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 <linux/slab.h>
36 #include <net/checksum.h>
37 #include <net/ip6_checksum.h>
38 #include <linux/net_tstamp.h>
39 #include <linux/mii.h>
40 #include <linux/ethtool.h>
41 #include <linux/if_vlan.h>
42 #include <linux/pci.h>
43 #include <linux/pci-aspm.h>
44 #include <linux/delay.h>
45 #include <linux/interrupt.h>
46 #include <linux/if_ether.h>
47 #include <linux/aer.h>
49 #include <linux/dca.h>
53 #define DRV_VERSION "2.1.0-k2"
54 char igb_driver_name[] = "igb";
55 char igb_driver_version[] = DRV_VERSION;
56 static const char igb_driver_string[] =
57 "Intel(R) Gigabit Ethernet Network Driver";
58 static const char igb_copyright[] = "Copyright (c) 2007-2009 Intel Corporation.";
60 static const struct e1000_info *igb_info_tbl[] = {
61 [board_82575] = &e1000_82575_info,
64 static DEFINE_PCI_DEVICE_TABLE(igb_pci_tbl) = {
65 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_COPPER), board_82575 },
66 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_FIBER), board_82575 },
67 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SERDES), board_82575 },
68 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SGMII), board_82575 },
69 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER), board_82575 },
70 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_FIBER), board_82575 },
71 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SERDES), board_82575 },
72 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SGMII), board_82575 },
73 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER_DUAL), board_82575 },
74 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576), board_82575 },
75 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS), board_82575 },
76 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS_SERDES), board_82575 },
77 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_FIBER), board_82575 },
78 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES), board_82575 },
79 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES_QUAD), board_82575 },
80 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER_ET2), board_82575 },
81 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER), board_82575 },
82 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
83 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
84 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
85 /* required last entry */
89 MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
91 void igb_reset(struct igb_adapter *);
92 static int igb_setup_all_tx_resources(struct igb_adapter *);
93 static int igb_setup_all_rx_resources(struct igb_adapter *);
94 static void igb_free_all_tx_resources(struct igb_adapter *);
95 static void igb_free_all_rx_resources(struct igb_adapter *);
96 static void igb_setup_mrqc(struct igb_adapter *);
97 void igb_update_stats(struct igb_adapter *);
98 static int igb_probe(struct pci_dev *, const struct pci_device_id *);
99 static void __devexit igb_remove(struct pci_dev *pdev);
100 static int igb_sw_init(struct igb_adapter *);
101 static int igb_open(struct net_device *);
102 static int igb_close(struct net_device *);
103 static void igb_configure_tx(struct igb_adapter *);
104 static void igb_configure_rx(struct igb_adapter *);
105 static void igb_clean_all_tx_rings(struct igb_adapter *);
106 static void igb_clean_all_rx_rings(struct igb_adapter *);
107 static void igb_clean_tx_ring(struct igb_ring *);
108 static void igb_clean_rx_ring(struct igb_ring *);
109 static void igb_set_rx_mode(struct net_device *);
110 static void igb_update_phy_info(unsigned long);
111 static void igb_watchdog(unsigned long);
112 static void igb_watchdog_task(struct work_struct *);
113 static netdev_tx_t igb_xmit_frame_adv(struct sk_buff *skb, struct net_device *);
114 static struct net_device_stats *igb_get_stats(struct net_device *);
115 static int igb_change_mtu(struct net_device *, int);
116 static int igb_set_mac(struct net_device *, void *);
117 static void igb_set_uta(struct igb_adapter *adapter);
118 static irqreturn_t igb_intr(int irq, void *);
119 static irqreturn_t igb_intr_msi(int irq, void *);
120 static irqreturn_t igb_msix_other(int irq, void *);
121 static irqreturn_t igb_msix_ring(int irq, void *);
122 #ifdef CONFIG_IGB_DCA
123 static void igb_update_dca(struct igb_q_vector *);
124 static void igb_setup_dca(struct igb_adapter *);
125 #endif /* CONFIG_IGB_DCA */
126 static bool igb_clean_tx_irq(struct igb_q_vector *);
127 static int igb_poll(struct napi_struct *, int);
128 static bool igb_clean_rx_irq_adv(struct igb_q_vector *, int *, int);
129 static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
130 static void igb_tx_timeout(struct net_device *);
131 static void igb_reset_task(struct work_struct *);
132 static void igb_vlan_rx_register(struct net_device *, struct vlan_group *);
133 static void igb_vlan_rx_add_vid(struct net_device *, u16);
134 static void igb_vlan_rx_kill_vid(struct net_device *, u16);
135 static void igb_restore_vlan(struct igb_adapter *);
136 static void igb_rar_set_qsel(struct igb_adapter *, u8 *, u32 , u8);
137 static void igb_ping_all_vfs(struct igb_adapter *);
138 static void igb_msg_task(struct igb_adapter *);
139 static void igb_vmm_control(struct igb_adapter *);
140 static int igb_set_vf_mac(struct igb_adapter *, int, unsigned char *);
141 static void igb_restore_vf_multicasts(struct igb_adapter *adapter);
142 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac);
143 static int igb_ndo_set_vf_vlan(struct net_device *netdev,
144 int vf, u16 vlan, u8 qos);
145 static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf, int tx_rate);
146 static int igb_ndo_get_vf_config(struct net_device *netdev, int vf,
147 struct ifla_vf_info *ivi);
150 static int igb_suspend(struct pci_dev *, pm_message_t);
151 static int igb_resume(struct pci_dev *);
153 static void igb_shutdown(struct pci_dev *);
154 #ifdef CONFIG_IGB_DCA
155 static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
156 static struct notifier_block dca_notifier = {
157 .notifier_call = igb_notify_dca,
162 #ifdef CONFIG_NET_POLL_CONTROLLER
163 /* for netdump / net console */
164 static void igb_netpoll(struct net_device *);
166 #ifdef CONFIG_PCI_IOV
167 static unsigned int max_vfs = 0;
168 module_param(max_vfs, uint, 0);
169 MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate "
170 "per physical function");
171 #endif /* CONFIG_PCI_IOV */
173 static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
174 pci_channel_state_t);
175 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
176 static void igb_io_resume(struct pci_dev *);
178 static struct pci_error_handlers igb_err_handler = {
179 .error_detected = igb_io_error_detected,
180 .slot_reset = igb_io_slot_reset,
181 .resume = igb_io_resume,
185 static struct pci_driver igb_driver = {
186 .name = igb_driver_name,
187 .id_table = igb_pci_tbl,
189 .remove = __devexit_p(igb_remove),
191 /* Power Managment Hooks */
192 .suspend = igb_suspend,
193 .resume = igb_resume,
195 .shutdown = igb_shutdown,
196 .err_handler = &igb_err_handler
199 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
200 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
201 MODULE_LICENSE("GPL");
202 MODULE_VERSION(DRV_VERSION);
205 * igb_read_clock - read raw cycle counter (to be used by time counter)
207 static cycle_t igb_read_clock(const struct cyclecounter *tc)
209 struct igb_adapter *adapter =
210 container_of(tc, struct igb_adapter, cycles);
211 struct e1000_hw *hw = &adapter->hw;
216 * The timestamp latches on lowest register read. For the 82580
217 * the lowest register is SYSTIMR instead of SYSTIML. However we never
218 * adjusted TIMINCA so SYSTIMR will just read as all 0s so ignore it.
220 if (hw->mac.type == e1000_82580) {
221 stamp = rd32(E1000_SYSTIMR) >> 8;
222 shift = IGB_82580_TSYNC_SHIFT;
225 stamp |= (u64)rd32(E1000_SYSTIML) << shift;
226 stamp |= (u64)rd32(E1000_SYSTIMH) << (shift + 32);
231 * igb_get_hw_dev - return device
232 * used by hardware layer to print debugging information
234 struct net_device *igb_get_hw_dev(struct e1000_hw *hw)
236 struct igb_adapter *adapter = hw->back;
237 return adapter->netdev;
241 * igb_init_module - Driver Registration Routine
243 * igb_init_module is the first routine called when the driver is
244 * loaded. All it does is register with the PCI subsystem.
246 static int __init igb_init_module(void)
249 printk(KERN_INFO "%s - version %s\n",
250 igb_driver_string, igb_driver_version);
252 printk(KERN_INFO "%s\n", igb_copyright);
254 #ifdef CONFIG_IGB_DCA
255 dca_register_notify(&dca_notifier);
257 ret = pci_register_driver(&igb_driver);
261 module_init(igb_init_module);
264 * igb_exit_module - Driver Exit Cleanup Routine
266 * igb_exit_module is called just before the driver is removed
269 static void __exit igb_exit_module(void)
271 #ifdef CONFIG_IGB_DCA
272 dca_unregister_notify(&dca_notifier);
274 pci_unregister_driver(&igb_driver);
277 module_exit(igb_exit_module);
279 #define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
281 * igb_cache_ring_register - Descriptor ring to register mapping
282 * @adapter: board private structure to initialize
284 * Once we know the feature-set enabled for the device, we'll cache
285 * the register offset the descriptor ring is assigned to.
287 static void igb_cache_ring_register(struct igb_adapter *adapter)
290 u32 rbase_offset = adapter->vfs_allocated_count;
292 switch (adapter->hw.mac.type) {
294 /* The queues are allocated for virtualization such that VF 0
295 * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
296 * In order to avoid collision we start at the first free queue
297 * and continue consuming queues in the same sequence
299 if (adapter->vfs_allocated_count) {
300 for (; i < adapter->rss_queues; i++)
301 adapter->rx_ring[i]->reg_idx = rbase_offset +
303 for (; j < adapter->rss_queues; j++)
304 adapter->tx_ring[j]->reg_idx = rbase_offset +
311 for (; i < adapter->num_rx_queues; i++)
312 adapter->rx_ring[i]->reg_idx = rbase_offset + i;
313 for (; j < adapter->num_tx_queues; j++)
314 adapter->tx_ring[j]->reg_idx = rbase_offset + j;
319 static void igb_free_queues(struct igb_adapter *adapter)
323 for (i = 0; i < adapter->num_tx_queues; i++) {
324 kfree(adapter->tx_ring[i]);
325 adapter->tx_ring[i] = NULL;
327 for (i = 0; i < adapter->num_rx_queues; i++) {
328 kfree(adapter->rx_ring[i]);
329 adapter->rx_ring[i] = NULL;
331 adapter->num_rx_queues = 0;
332 adapter->num_tx_queues = 0;
336 * igb_alloc_queues - Allocate memory for all rings
337 * @adapter: board private structure to initialize
339 * We allocate one ring per queue at run-time since we don't know the
340 * number of queues at compile-time.
342 static int igb_alloc_queues(struct igb_adapter *adapter)
344 struct igb_ring *ring;
347 for (i = 0; i < adapter->num_tx_queues; i++) {
348 ring = kzalloc(sizeof(struct igb_ring), GFP_KERNEL);
351 ring->count = adapter->tx_ring_count;
352 ring->queue_index = i;
353 ring->dev = &adapter->pdev->dev;
354 ring->netdev = adapter->netdev;
355 /* For 82575, context index must be unique per ring. */
356 if (adapter->hw.mac.type == e1000_82575)
357 ring->flags = IGB_RING_FLAG_TX_CTX_IDX;
358 adapter->tx_ring[i] = ring;
361 for (i = 0; i < adapter->num_rx_queues; i++) {
362 ring = kzalloc(sizeof(struct igb_ring), GFP_KERNEL);
365 ring->count = adapter->rx_ring_count;
366 ring->queue_index = i;
367 ring->dev = &adapter->pdev->dev;
368 ring->netdev = adapter->netdev;
369 ring->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
370 ring->flags = IGB_RING_FLAG_RX_CSUM; /* enable rx checksum */
371 /* set flag indicating ring supports SCTP checksum offload */
372 if (adapter->hw.mac.type >= e1000_82576)
373 ring->flags |= IGB_RING_FLAG_RX_SCTP_CSUM;
374 adapter->rx_ring[i] = ring;
377 igb_cache_ring_register(adapter);
382 igb_free_queues(adapter);
387 #define IGB_N0_QUEUE -1
388 static void igb_assign_vector(struct igb_q_vector *q_vector, int msix_vector)
391 struct igb_adapter *adapter = q_vector->adapter;
392 struct e1000_hw *hw = &adapter->hw;
394 int rx_queue = IGB_N0_QUEUE;
395 int tx_queue = IGB_N0_QUEUE;
397 if (q_vector->rx_ring)
398 rx_queue = q_vector->rx_ring->reg_idx;
399 if (q_vector->tx_ring)
400 tx_queue = q_vector->tx_ring->reg_idx;
402 switch (hw->mac.type) {
404 /* The 82575 assigns vectors using a bitmask, which matches the
405 bitmask for the EICR/EIMS/EIMC registers. To assign one
406 or more queues to a vector, we write the appropriate bits
407 into the MSIXBM register for that vector. */
408 if (rx_queue > IGB_N0_QUEUE)
409 msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
410 if (tx_queue > IGB_N0_QUEUE)
411 msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
412 if (!adapter->msix_entries && msix_vector == 0)
413 msixbm |= E1000_EIMS_OTHER;
414 array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
415 q_vector->eims_value = msixbm;
418 /* 82576 uses a table-based method for assigning vectors.
419 Each queue has a single entry in the table to which we write
420 a vector number along with a "valid" bit. Sadly, the layout
421 of the table is somewhat counterintuitive. */
422 if (rx_queue > IGB_N0_QUEUE) {
423 index = (rx_queue & 0x7);
424 ivar = array_rd32(E1000_IVAR0, index);
426 /* vector goes into low byte of register */
427 ivar = ivar & 0xFFFFFF00;
428 ivar |= msix_vector | E1000_IVAR_VALID;
430 /* vector goes into third byte of register */
431 ivar = ivar & 0xFF00FFFF;
432 ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
434 array_wr32(E1000_IVAR0, index, ivar);
436 if (tx_queue > IGB_N0_QUEUE) {
437 index = (tx_queue & 0x7);
438 ivar = array_rd32(E1000_IVAR0, index);
440 /* vector goes into second byte of register */
441 ivar = ivar & 0xFFFF00FF;
442 ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
444 /* vector goes into high byte of register */
445 ivar = ivar & 0x00FFFFFF;
446 ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
448 array_wr32(E1000_IVAR0, index, ivar);
450 q_vector->eims_value = 1 << msix_vector;
454 /* 82580 uses the same table-based approach as 82576 but has fewer
455 entries as a result we carry over for queues greater than 4. */
456 if (rx_queue > IGB_N0_QUEUE) {
457 index = (rx_queue >> 1);
458 ivar = array_rd32(E1000_IVAR0, index);
459 if (rx_queue & 0x1) {
460 /* vector goes into third byte of register */
461 ivar = ivar & 0xFF00FFFF;
462 ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
464 /* vector goes into low byte of register */
465 ivar = ivar & 0xFFFFFF00;
466 ivar |= msix_vector | E1000_IVAR_VALID;
468 array_wr32(E1000_IVAR0, index, ivar);
470 if (tx_queue > IGB_N0_QUEUE) {
471 index = (tx_queue >> 1);
472 ivar = array_rd32(E1000_IVAR0, index);
473 if (tx_queue & 0x1) {
474 /* vector goes into high byte of register */
475 ivar = ivar & 0x00FFFFFF;
476 ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
478 /* vector goes into second byte of register */
479 ivar = ivar & 0xFFFF00FF;
480 ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
482 array_wr32(E1000_IVAR0, index, ivar);
484 q_vector->eims_value = 1 << msix_vector;
491 /* add q_vector eims value to global eims_enable_mask */
492 adapter->eims_enable_mask |= q_vector->eims_value;
494 /* configure q_vector to set itr on first interrupt */
495 q_vector->set_itr = 1;
499 * igb_configure_msix - Configure MSI-X hardware
501 * igb_configure_msix sets up the hardware to properly
502 * generate MSI-X interrupts.
504 static void igb_configure_msix(struct igb_adapter *adapter)
508 struct e1000_hw *hw = &adapter->hw;
510 adapter->eims_enable_mask = 0;
512 /* set vector for other causes, i.e. link changes */
513 switch (hw->mac.type) {
515 tmp = rd32(E1000_CTRL_EXT);
516 /* enable MSI-X PBA support*/
517 tmp |= E1000_CTRL_EXT_PBA_CLR;
519 /* Auto-Mask interrupts upon ICR read. */
520 tmp |= E1000_CTRL_EXT_EIAME;
521 tmp |= E1000_CTRL_EXT_IRCA;
523 wr32(E1000_CTRL_EXT, tmp);
525 /* enable msix_other interrupt */
526 array_wr32(E1000_MSIXBM(0), vector++,
528 adapter->eims_other = E1000_EIMS_OTHER;
535 /* Turn on MSI-X capability first, or our settings
536 * won't stick. And it will take days to debug. */
537 wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
538 E1000_GPIE_PBA | E1000_GPIE_EIAME |
541 /* enable msix_other interrupt */
542 adapter->eims_other = 1 << vector;
543 tmp = (vector++ | E1000_IVAR_VALID) << 8;
545 wr32(E1000_IVAR_MISC, tmp);
548 /* do nothing, since nothing else supports MSI-X */
550 } /* switch (hw->mac.type) */
552 adapter->eims_enable_mask |= adapter->eims_other;
554 for (i = 0; i < adapter->num_q_vectors; i++)
555 igb_assign_vector(adapter->q_vector[i], vector++);
561 * igb_request_msix - Initialize MSI-X interrupts
563 * igb_request_msix allocates MSI-X vectors and requests interrupts from the
566 static int igb_request_msix(struct igb_adapter *adapter)
568 struct net_device *netdev = adapter->netdev;
569 struct e1000_hw *hw = &adapter->hw;
570 int i, err = 0, vector = 0;
572 err = request_irq(adapter->msix_entries[vector].vector,
573 igb_msix_other, 0, netdev->name, adapter);
578 for (i = 0; i < adapter->num_q_vectors; i++) {
579 struct igb_q_vector *q_vector = adapter->q_vector[i];
581 q_vector->itr_register = hw->hw_addr + E1000_EITR(vector);
583 if (q_vector->rx_ring && q_vector->tx_ring)
584 sprintf(q_vector->name, "%s-TxRx-%u", netdev->name,
585 q_vector->rx_ring->queue_index);
586 else if (q_vector->tx_ring)
587 sprintf(q_vector->name, "%s-tx-%u", netdev->name,
588 q_vector->tx_ring->queue_index);
589 else if (q_vector->rx_ring)
590 sprintf(q_vector->name, "%s-rx-%u", netdev->name,
591 q_vector->rx_ring->queue_index);
593 sprintf(q_vector->name, "%s-unused", netdev->name);
595 err = request_irq(adapter->msix_entries[vector].vector,
596 igb_msix_ring, 0, q_vector->name,
603 igb_configure_msix(adapter);
609 static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
611 if (adapter->msix_entries) {
612 pci_disable_msix(adapter->pdev);
613 kfree(adapter->msix_entries);
614 adapter->msix_entries = NULL;
615 } else if (adapter->flags & IGB_FLAG_HAS_MSI) {
616 pci_disable_msi(adapter->pdev);
621 * igb_free_q_vectors - Free memory allocated for interrupt vectors
622 * @adapter: board private structure to initialize
624 * This function frees the memory allocated to the q_vectors. In addition if
625 * NAPI is enabled it will delete any references to the NAPI struct prior
626 * to freeing the q_vector.
628 static void igb_free_q_vectors(struct igb_adapter *adapter)
632 for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
633 struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
634 adapter->q_vector[v_idx] = NULL;
637 netif_napi_del(&q_vector->napi);
640 adapter->num_q_vectors = 0;
644 * igb_clear_interrupt_scheme - reset the device to a state of no interrupts
646 * This function resets the device so that it has 0 rx queues, tx queues, and
647 * MSI-X interrupts allocated.
649 static void igb_clear_interrupt_scheme(struct igb_adapter *adapter)
651 igb_free_queues(adapter);
652 igb_free_q_vectors(adapter);
653 igb_reset_interrupt_capability(adapter);
657 * igb_set_interrupt_capability - set MSI or MSI-X if supported
659 * Attempt to configure interrupts using the best available
660 * capabilities of the hardware and kernel.
662 static void igb_set_interrupt_capability(struct igb_adapter *adapter)
667 /* Number of supported queues. */
668 adapter->num_rx_queues = adapter->rss_queues;
669 adapter->num_tx_queues = adapter->rss_queues;
671 /* start with one vector for every rx queue */
672 numvecs = adapter->num_rx_queues;
674 /* if tx handler is separate add 1 for every tx queue */
675 if (!(adapter->flags & IGB_FLAG_QUEUE_PAIRS))
676 numvecs += adapter->num_tx_queues;
678 /* store the number of vectors reserved for queues */
679 adapter->num_q_vectors = numvecs;
681 /* add 1 vector for link status interrupts */
683 adapter->msix_entries = kcalloc(numvecs, sizeof(struct msix_entry),
685 if (!adapter->msix_entries)
688 for (i = 0; i < numvecs; i++)
689 adapter->msix_entries[i].entry = i;
691 err = pci_enable_msix(adapter->pdev,
692 adapter->msix_entries,
697 igb_reset_interrupt_capability(adapter);
699 /* If we can't do MSI-X, try MSI */
701 #ifdef CONFIG_PCI_IOV
702 /* disable SR-IOV for non MSI-X configurations */
703 if (adapter->vf_data) {
704 struct e1000_hw *hw = &adapter->hw;
705 /* disable iov and allow time for transactions to clear */
706 pci_disable_sriov(adapter->pdev);
709 kfree(adapter->vf_data);
710 adapter->vf_data = NULL;
711 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
713 dev_info(&adapter->pdev->dev, "IOV Disabled\n");
716 adapter->vfs_allocated_count = 0;
717 adapter->rss_queues = 1;
718 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
719 adapter->num_rx_queues = 1;
720 adapter->num_tx_queues = 1;
721 adapter->num_q_vectors = 1;
722 if (!pci_enable_msi(adapter->pdev))
723 adapter->flags |= IGB_FLAG_HAS_MSI;
725 /* Notify the stack of the (possibly) reduced Tx Queue count. */
726 adapter->netdev->real_num_tx_queues = adapter->num_tx_queues;
731 * igb_alloc_q_vectors - Allocate memory for interrupt vectors
732 * @adapter: board private structure to initialize
734 * We allocate one q_vector per queue interrupt. If allocation fails we
737 static int igb_alloc_q_vectors(struct igb_adapter *adapter)
739 struct igb_q_vector *q_vector;
740 struct e1000_hw *hw = &adapter->hw;
743 for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
744 q_vector = kzalloc(sizeof(struct igb_q_vector), GFP_KERNEL);
747 q_vector->adapter = adapter;
748 q_vector->itr_register = hw->hw_addr + E1000_EITR(0);
749 q_vector->itr_val = IGB_START_ITR;
750 netif_napi_add(adapter->netdev, &q_vector->napi, igb_poll, 64);
751 adapter->q_vector[v_idx] = q_vector;
756 igb_free_q_vectors(adapter);
760 static void igb_map_rx_ring_to_vector(struct igb_adapter *adapter,
761 int ring_idx, int v_idx)
763 struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
765 q_vector->rx_ring = adapter->rx_ring[ring_idx];
766 q_vector->rx_ring->q_vector = q_vector;
767 q_vector->itr_val = adapter->rx_itr_setting;
768 if (q_vector->itr_val && q_vector->itr_val <= 3)
769 q_vector->itr_val = IGB_START_ITR;
772 static void igb_map_tx_ring_to_vector(struct igb_adapter *adapter,
773 int ring_idx, int v_idx)
775 struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
777 q_vector->tx_ring = adapter->tx_ring[ring_idx];
778 q_vector->tx_ring->q_vector = q_vector;
779 q_vector->itr_val = adapter->tx_itr_setting;
780 if (q_vector->itr_val && q_vector->itr_val <= 3)
781 q_vector->itr_val = IGB_START_ITR;
785 * igb_map_ring_to_vector - maps allocated queues to vectors
787 * This function maps the recently allocated queues to vectors.
789 static int igb_map_ring_to_vector(struct igb_adapter *adapter)
794 if ((adapter->num_q_vectors < adapter->num_rx_queues) ||
795 (adapter->num_q_vectors < adapter->num_tx_queues))
798 if (adapter->num_q_vectors >=
799 (adapter->num_rx_queues + adapter->num_tx_queues)) {
800 for (i = 0; i < adapter->num_rx_queues; i++)
801 igb_map_rx_ring_to_vector(adapter, i, v_idx++);
802 for (i = 0; i < adapter->num_tx_queues; i++)
803 igb_map_tx_ring_to_vector(adapter, i, v_idx++);
805 for (i = 0; i < adapter->num_rx_queues; i++) {
806 if (i < adapter->num_tx_queues)
807 igb_map_tx_ring_to_vector(adapter, i, v_idx);
808 igb_map_rx_ring_to_vector(adapter, i, v_idx++);
810 for (; i < adapter->num_tx_queues; i++)
811 igb_map_tx_ring_to_vector(adapter, i, v_idx++);
817 * igb_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
819 * This function initializes the interrupts and allocates all of the queues.
821 static int igb_init_interrupt_scheme(struct igb_adapter *adapter)
823 struct pci_dev *pdev = adapter->pdev;
826 igb_set_interrupt_capability(adapter);
828 err = igb_alloc_q_vectors(adapter);
830 dev_err(&pdev->dev, "Unable to allocate memory for vectors\n");
831 goto err_alloc_q_vectors;
834 err = igb_alloc_queues(adapter);
836 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
837 goto err_alloc_queues;
840 err = igb_map_ring_to_vector(adapter);
842 dev_err(&pdev->dev, "Invalid q_vector to ring mapping\n");
849 igb_free_queues(adapter);
851 igb_free_q_vectors(adapter);
853 igb_reset_interrupt_capability(adapter);
858 * igb_request_irq - initialize interrupts
860 * Attempts to configure interrupts using the best available
861 * capabilities of the hardware and kernel.
863 static int igb_request_irq(struct igb_adapter *adapter)
865 struct net_device *netdev = adapter->netdev;
866 struct pci_dev *pdev = adapter->pdev;
869 if (adapter->msix_entries) {
870 err = igb_request_msix(adapter);
873 /* fall back to MSI */
874 igb_clear_interrupt_scheme(adapter);
875 if (!pci_enable_msi(adapter->pdev))
876 adapter->flags |= IGB_FLAG_HAS_MSI;
877 igb_free_all_tx_resources(adapter);
878 igb_free_all_rx_resources(adapter);
879 adapter->num_tx_queues = 1;
880 adapter->num_rx_queues = 1;
881 adapter->num_q_vectors = 1;
882 err = igb_alloc_q_vectors(adapter);
885 "Unable to allocate memory for vectors\n");
888 err = igb_alloc_queues(adapter);
891 "Unable to allocate memory for queues\n");
892 igb_free_q_vectors(adapter);
895 igb_setup_all_tx_resources(adapter);
896 igb_setup_all_rx_resources(adapter);
898 igb_assign_vector(adapter->q_vector[0], 0);
901 if (adapter->flags & IGB_FLAG_HAS_MSI) {
902 err = request_irq(adapter->pdev->irq, igb_intr_msi, 0,
903 netdev->name, adapter);
907 /* fall back to legacy interrupts */
908 igb_reset_interrupt_capability(adapter);
909 adapter->flags &= ~IGB_FLAG_HAS_MSI;
912 err = request_irq(adapter->pdev->irq, igb_intr, IRQF_SHARED,
913 netdev->name, adapter);
916 dev_err(&adapter->pdev->dev, "Error %d getting interrupt\n",
923 static void igb_free_irq(struct igb_adapter *adapter)
925 if (adapter->msix_entries) {
928 free_irq(adapter->msix_entries[vector++].vector, adapter);
930 for (i = 0; i < adapter->num_q_vectors; i++) {
931 struct igb_q_vector *q_vector = adapter->q_vector[i];
932 free_irq(adapter->msix_entries[vector++].vector,
936 free_irq(adapter->pdev->irq, adapter);
941 * igb_irq_disable - Mask off interrupt generation on the NIC
942 * @adapter: board private structure
944 static void igb_irq_disable(struct igb_adapter *adapter)
946 struct e1000_hw *hw = &adapter->hw;
949 * we need to be careful when disabling interrupts. The VFs are also
950 * mapped into these registers and so clearing the bits can cause
951 * issues on the VF drivers so we only need to clear what we set
953 if (adapter->msix_entries) {
954 u32 regval = rd32(E1000_EIAM);
955 wr32(E1000_EIAM, regval & ~adapter->eims_enable_mask);
956 wr32(E1000_EIMC, adapter->eims_enable_mask);
957 regval = rd32(E1000_EIAC);
958 wr32(E1000_EIAC, regval & ~adapter->eims_enable_mask);
964 synchronize_irq(adapter->pdev->irq);
968 * igb_irq_enable - Enable default interrupt generation settings
969 * @adapter: board private structure
971 static void igb_irq_enable(struct igb_adapter *adapter)
973 struct e1000_hw *hw = &adapter->hw;
975 if (adapter->msix_entries) {
976 u32 ims = E1000_IMS_LSC | E1000_IMS_DOUTSYNC;
977 u32 regval = rd32(E1000_EIAC);
978 wr32(E1000_EIAC, regval | adapter->eims_enable_mask);
979 regval = rd32(E1000_EIAM);
980 wr32(E1000_EIAM, regval | adapter->eims_enable_mask);
981 wr32(E1000_EIMS, adapter->eims_enable_mask);
982 if (adapter->vfs_allocated_count) {
983 wr32(E1000_MBVFIMR, 0xFF);
984 ims |= E1000_IMS_VMMB;
986 if (adapter->hw.mac.type == e1000_82580)
987 ims |= E1000_IMS_DRSTA;
989 wr32(E1000_IMS, ims);
991 wr32(E1000_IMS, IMS_ENABLE_MASK |
993 wr32(E1000_IAM, IMS_ENABLE_MASK |
998 static void igb_update_mng_vlan(struct igb_adapter *adapter)
1000 struct e1000_hw *hw = &adapter->hw;
1001 u16 vid = adapter->hw.mng_cookie.vlan_id;
1002 u16 old_vid = adapter->mng_vlan_id;
1004 if (hw->mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1005 /* add VID to filter table */
1006 igb_vfta_set(hw, vid, true);
1007 adapter->mng_vlan_id = vid;
1009 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1012 if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
1014 !vlan_group_get_device(adapter->vlgrp, old_vid)) {
1015 /* remove VID from filter table */
1016 igb_vfta_set(hw, old_vid, false);
1021 * igb_release_hw_control - release control of the h/w to f/w
1022 * @adapter: address of board private structure
1024 * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
1025 * For ASF and Pass Through versions of f/w this means that the
1026 * driver is no longer loaded.
1029 static void igb_release_hw_control(struct igb_adapter *adapter)
1031 struct e1000_hw *hw = &adapter->hw;
1034 /* Let firmware take over control of h/w */
1035 ctrl_ext = rd32(E1000_CTRL_EXT);
1036 wr32(E1000_CTRL_EXT,
1037 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1041 * igb_get_hw_control - get control of the h/w from f/w
1042 * @adapter: address of board private structure
1044 * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
1045 * For ASF and Pass Through versions of f/w this means that
1046 * the driver is loaded.
1049 static void igb_get_hw_control(struct igb_adapter *adapter)
1051 struct e1000_hw *hw = &adapter->hw;
1054 /* Let firmware know the driver has taken over */
1055 ctrl_ext = rd32(E1000_CTRL_EXT);
1056 wr32(E1000_CTRL_EXT,
1057 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1061 * igb_configure - configure the hardware for RX and TX
1062 * @adapter: private board structure
1064 static void igb_configure(struct igb_adapter *adapter)
1066 struct net_device *netdev = adapter->netdev;
1069 igb_get_hw_control(adapter);
1070 igb_set_rx_mode(netdev);
1072 igb_restore_vlan(adapter);
1074 igb_setup_tctl(adapter);
1075 igb_setup_mrqc(adapter);
1076 igb_setup_rctl(adapter);
1078 igb_configure_tx(adapter);
1079 igb_configure_rx(adapter);
1081 igb_rx_fifo_flush_82575(&adapter->hw);
1083 /* call igb_desc_unused which always leaves
1084 * at least 1 descriptor unused to make sure
1085 * next_to_use != next_to_clean */
1086 for (i = 0; i < adapter->num_rx_queues; i++) {
1087 struct igb_ring *ring = adapter->rx_ring[i];
1088 igb_alloc_rx_buffers_adv(ring, igb_desc_unused(ring));
1093 * igb_power_up_link - Power up the phy/serdes link
1094 * @adapter: address of board private structure
1096 void igb_power_up_link(struct igb_adapter *adapter)
1098 if (adapter->hw.phy.media_type == e1000_media_type_copper)
1099 igb_power_up_phy_copper(&adapter->hw);
1101 igb_power_up_serdes_link_82575(&adapter->hw);
1105 * igb_power_down_link - Power down the phy/serdes link
1106 * @adapter: address of board private structure
1108 static void igb_power_down_link(struct igb_adapter *adapter)
1110 if (adapter->hw.phy.media_type == e1000_media_type_copper)
1111 igb_power_down_phy_copper_82575(&adapter->hw);
1113 igb_shutdown_serdes_link_82575(&adapter->hw);
1117 * igb_up - Open the interface and prepare it to handle traffic
1118 * @adapter: board private structure
1120 int igb_up(struct igb_adapter *adapter)
1122 struct e1000_hw *hw = &adapter->hw;
1125 /* hardware has been reset, we need to reload some things */
1126 igb_configure(adapter);
1128 clear_bit(__IGB_DOWN, &adapter->state);
1130 for (i = 0; i < adapter->num_q_vectors; i++) {
1131 struct igb_q_vector *q_vector = adapter->q_vector[i];
1132 napi_enable(&q_vector->napi);
1134 if (adapter->msix_entries)
1135 igb_configure_msix(adapter);
1137 igb_assign_vector(adapter->q_vector[0], 0);
1139 /* Clear any pending interrupts. */
1141 igb_irq_enable(adapter);
1143 /* notify VFs that reset has been completed */
1144 if (adapter->vfs_allocated_count) {
1145 u32 reg_data = rd32(E1000_CTRL_EXT);
1146 reg_data |= E1000_CTRL_EXT_PFRSTD;
1147 wr32(E1000_CTRL_EXT, reg_data);
1150 netif_tx_start_all_queues(adapter->netdev);
1152 /* start the watchdog. */
1153 hw->mac.get_link_status = 1;
1154 schedule_work(&adapter->watchdog_task);
1159 void igb_down(struct igb_adapter *adapter)
1161 struct net_device *netdev = adapter->netdev;
1162 struct e1000_hw *hw = &adapter->hw;
1166 /* signal that we're down so the interrupt handler does not
1167 * reschedule our watchdog timer */
1168 set_bit(__IGB_DOWN, &adapter->state);
1170 /* disable receives in the hardware */
1171 rctl = rd32(E1000_RCTL);
1172 wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
1173 /* flush and sleep below */
1175 netif_tx_stop_all_queues(netdev);
1177 /* disable transmits in the hardware */
1178 tctl = rd32(E1000_TCTL);
1179 tctl &= ~E1000_TCTL_EN;
1180 wr32(E1000_TCTL, tctl);
1181 /* flush both disables and wait for them to finish */
1185 for (i = 0; i < adapter->num_q_vectors; i++) {
1186 struct igb_q_vector *q_vector = adapter->q_vector[i];
1187 napi_disable(&q_vector->napi);
1190 igb_irq_disable(adapter);
1192 del_timer_sync(&adapter->watchdog_timer);
1193 del_timer_sync(&adapter->phy_info_timer);
1195 netif_carrier_off(netdev);
1197 /* record the stats before reset*/
1198 igb_update_stats(adapter);
1200 adapter->link_speed = 0;
1201 adapter->link_duplex = 0;
1203 if (!pci_channel_offline(adapter->pdev))
1205 igb_clean_all_tx_rings(adapter);
1206 igb_clean_all_rx_rings(adapter);
1207 #ifdef CONFIG_IGB_DCA
1209 /* since we reset the hardware DCA settings were cleared */
1210 igb_setup_dca(adapter);
1214 void igb_reinit_locked(struct igb_adapter *adapter)
1216 WARN_ON(in_interrupt());
1217 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
1221 clear_bit(__IGB_RESETTING, &adapter->state);
1224 void igb_reset(struct igb_adapter *adapter)
1226 struct pci_dev *pdev = adapter->pdev;
1227 struct e1000_hw *hw = &adapter->hw;
1228 struct e1000_mac_info *mac = &hw->mac;
1229 struct e1000_fc_info *fc = &hw->fc;
1230 u32 pba = 0, tx_space, min_tx_space, min_rx_space;
1233 /* Repartition Pba for greater than 9k mtu
1234 * To take effect CTRL.RST is required.
1236 switch (mac->type) {
1239 pba = rd32(E1000_RXPBS);
1240 pba = igb_rxpbs_adjust_82580(pba);
1243 pba = rd32(E1000_RXPBS);
1244 pba &= E1000_RXPBS_SIZE_MASK_82576;
1248 pba = E1000_PBA_34K;
1252 if ((adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) &&
1253 (mac->type < e1000_82576)) {
1254 /* adjust PBA for jumbo frames */
1255 wr32(E1000_PBA, pba);
1257 /* To maintain wire speed transmits, the Tx FIFO should be
1258 * large enough to accommodate two full transmit packets,
1259 * rounded up to the next 1KB and expressed in KB. Likewise,
1260 * the Rx FIFO should be large enough to accommodate at least
1261 * one full receive packet and is similarly rounded up and
1262 * expressed in KB. */
1263 pba = rd32(E1000_PBA);
1264 /* upper 16 bits has Tx packet buffer allocation size in KB */
1265 tx_space = pba >> 16;
1266 /* lower 16 bits has Rx packet buffer allocation size in KB */
1268 /* the tx fifo also stores 16 bytes of information about the tx
1269 * but don't include ethernet FCS because hardware appends it */
1270 min_tx_space = (adapter->max_frame_size +
1271 sizeof(union e1000_adv_tx_desc) -
1273 min_tx_space = ALIGN(min_tx_space, 1024);
1274 min_tx_space >>= 10;
1275 /* software strips receive CRC, so leave room for it */
1276 min_rx_space = adapter->max_frame_size;
1277 min_rx_space = ALIGN(min_rx_space, 1024);
1278 min_rx_space >>= 10;
1280 /* If current Tx allocation is less than the min Tx FIFO size,
1281 * and the min Tx FIFO size is less than the current Rx FIFO
1282 * allocation, take space away from current Rx allocation */
1283 if (tx_space < min_tx_space &&
1284 ((min_tx_space - tx_space) < pba)) {
1285 pba = pba - (min_tx_space - tx_space);
1287 /* if short on rx space, rx wins and must trump tx
1289 if (pba < min_rx_space)
1292 wr32(E1000_PBA, pba);
1295 /* flow control settings */
1296 /* The high water mark must be low enough to fit one full frame
1297 * (or the size used for early receive) above it in the Rx FIFO.
1298 * Set it to the lower of:
1299 * - 90% of the Rx FIFO size, or
1300 * - the full Rx FIFO size minus one full frame */
1301 hwm = min(((pba << 10) * 9 / 10),
1302 ((pba << 10) - 2 * adapter->max_frame_size));
1304 fc->high_water = hwm & 0xFFF0; /* 16-byte granularity */
1305 fc->low_water = fc->high_water - 16;
1306 fc->pause_time = 0xFFFF;
1308 fc->current_mode = fc->requested_mode;
1310 /* disable receive for all VFs and wait one second */
1311 if (adapter->vfs_allocated_count) {
1313 for (i = 0 ; i < adapter->vfs_allocated_count; i++)
1314 adapter->vf_data[i].flags = 0;
1316 /* ping all the active vfs to let them know we are going down */
1317 igb_ping_all_vfs(adapter);
1319 /* disable transmits and receives */
1320 wr32(E1000_VFRE, 0);
1321 wr32(E1000_VFTE, 0);
1324 /* Allow time for pending master requests to run */
1325 hw->mac.ops.reset_hw(hw);
1328 if (hw->mac.ops.init_hw(hw))
1329 dev_err(&pdev->dev, "Hardware Error\n");
1331 if (hw->mac.type == e1000_82580) {
1332 u32 reg = rd32(E1000_PCIEMISC);
1333 wr32(E1000_PCIEMISC,
1334 reg & ~E1000_PCIEMISC_LX_DECISION);
1336 if (!netif_running(adapter->netdev))
1337 igb_power_down_link(adapter);
1339 igb_update_mng_vlan(adapter);
1341 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
1342 wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
1344 igb_get_phy_info(hw);
1347 static const struct net_device_ops igb_netdev_ops = {
1348 .ndo_open = igb_open,
1349 .ndo_stop = igb_close,
1350 .ndo_start_xmit = igb_xmit_frame_adv,
1351 .ndo_get_stats = igb_get_stats,
1352 .ndo_set_rx_mode = igb_set_rx_mode,
1353 .ndo_set_multicast_list = igb_set_rx_mode,
1354 .ndo_set_mac_address = igb_set_mac,
1355 .ndo_change_mtu = igb_change_mtu,
1356 .ndo_do_ioctl = igb_ioctl,
1357 .ndo_tx_timeout = igb_tx_timeout,
1358 .ndo_validate_addr = eth_validate_addr,
1359 .ndo_vlan_rx_register = igb_vlan_rx_register,
1360 .ndo_vlan_rx_add_vid = igb_vlan_rx_add_vid,
1361 .ndo_vlan_rx_kill_vid = igb_vlan_rx_kill_vid,
1362 .ndo_set_vf_mac = igb_ndo_set_vf_mac,
1363 .ndo_set_vf_vlan = igb_ndo_set_vf_vlan,
1364 .ndo_set_vf_tx_rate = igb_ndo_set_vf_bw,
1365 .ndo_get_vf_config = igb_ndo_get_vf_config,
1366 #ifdef CONFIG_NET_POLL_CONTROLLER
1367 .ndo_poll_controller = igb_netpoll,
1372 * igb_probe - Device Initialization Routine
1373 * @pdev: PCI device information struct
1374 * @ent: entry in igb_pci_tbl
1376 * Returns 0 on success, negative on failure
1378 * igb_probe initializes an adapter identified by a pci_dev structure.
1379 * The OS initialization, configuring of the adapter private structure,
1380 * and a hardware reset occur.
1382 static int __devinit igb_probe(struct pci_dev *pdev,
1383 const struct pci_device_id *ent)
1385 struct net_device *netdev;
1386 struct igb_adapter *adapter;
1387 struct e1000_hw *hw;
1388 u16 eeprom_data = 0;
1389 static int global_quad_port_a; /* global quad port a indication */
1390 const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
1391 unsigned long mmio_start, mmio_len;
1392 int err, pci_using_dac;
1393 u16 eeprom_apme_mask = IGB_EEPROM_APME;
1396 err = pci_enable_device_mem(pdev);
1401 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
1403 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
1407 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
1409 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(32));
1411 dev_err(&pdev->dev, "No usable DMA "
1412 "configuration, aborting\n");
1418 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
1424 pci_enable_pcie_error_reporting(pdev);
1426 pci_set_master(pdev);
1427 pci_save_state(pdev);
1430 netdev = alloc_etherdev_mq(sizeof(struct igb_adapter),
1431 IGB_ABS_MAX_TX_QUEUES);
1433 goto err_alloc_etherdev;
1435 SET_NETDEV_DEV(netdev, &pdev->dev);
1437 pci_set_drvdata(pdev, netdev);
1438 adapter = netdev_priv(netdev);
1439 adapter->netdev = netdev;
1440 adapter->pdev = pdev;
1443 adapter->msg_enable = NETIF_MSG_DRV | NETIF_MSG_PROBE;
1445 mmio_start = pci_resource_start(pdev, 0);
1446 mmio_len = pci_resource_len(pdev, 0);
1449 hw->hw_addr = ioremap(mmio_start, mmio_len);
1453 netdev->netdev_ops = &igb_netdev_ops;
1454 igb_set_ethtool_ops(netdev);
1455 netdev->watchdog_timeo = 5 * HZ;
1457 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
1459 netdev->mem_start = mmio_start;
1460 netdev->mem_end = mmio_start + mmio_len;
1462 /* PCI config space info */
1463 hw->vendor_id = pdev->vendor;
1464 hw->device_id = pdev->device;
1465 hw->revision_id = pdev->revision;
1466 hw->subsystem_vendor_id = pdev->subsystem_vendor;
1467 hw->subsystem_device_id = pdev->subsystem_device;
1469 /* Copy the default MAC, PHY and NVM function pointers */
1470 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
1471 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
1472 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
1473 /* Initialize skew-specific constants */
1474 err = ei->get_invariants(hw);
1478 /* setup the private structure */
1479 err = igb_sw_init(adapter);
1483 igb_get_bus_info_pcie(hw);
1485 hw->phy.autoneg_wait_to_complete = false;
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 case E1000_DEV_ID_82576_QUAD_COPPER_ET2:
1598 /* if quad port adapter, disable WoL on all but port A */
1599 if (global_quad_port_a != 0)
1600 adapter->eeprom_wol = 0;
1602 adapter->flags |= IGB_FLAG_QUAD_PORT_A;
1603 /* Reset for multiple quad port adapters */
1604 if (++global_quad_port_a == 4)
1605 global_quad_port_a = 0;
1609 /* initialize the wol settings based on the eeprom settings */
1610 adapter->wol = adapter->eeprom_wol;
1611 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
1613 /* reset the hardware with the new settings */
1616 /* let the f/w know that the h/w is now under the control of the
1618 igb_get_hw_control(adapter);
1620 strcpy(netdev->name, "eth%d");
1621 err = register_netdev(netdev);
1625 /* carrier off reporting is important to ethtool even BEFORE open */
1626 netif_carrier_off(netdev);
1628 #ifdef CONFIG_IGB_DCA
1629 if (dca_add_requester(&pdev->dev) == 0) {
1630 adapter->flags |= IGB_FLAG_DCA_ENABLED;
1631 dev_info(&pdev->dev, "DCA enabled\n");
1632 igb_setup_dca(adapter);
1636 dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
1637 /* print bus type/speed/width info */
1638 dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
1640 ((hw->bus.speed == e1000_bus_speed_2500) ? "2.5Gb/s" :
1641 (hw->bus.speed == e1000_bus_speed_5000) ? "5.0Gb/s" :
1643 ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
1644 (hw->bus.width == e1000_bus_width_pcie_x2) ? "Width x2" :
1645 (hw->bus.width == e1000_bus_width_pcie_x1) ? "Width x1" :
1649 igb_read_part_num(hw, &part_num);
1650 dev_info(&pdev->dev, "%s: PBA No: %06x-%03x\n", netdev->name,
1651 (part_num >> 8), (part_num & 0xff));
1653 dev_info(&pdev->dev,
1654 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
1655 adapter->msix_entries ? "MSI-X" :
1656 (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
1657 adapter->num_rx_queues, adapter->num_tx_queues);
1662 igb_release_hw_control(adapter);
1664 if (!igb_check_reset_block(hw))
1667 if (hw->flash_address)
1668 iounmap(hw->flash_address);
1670 igb_clear_interrupt_scheme(adapter);
1671 iounmap(hw->hw_addr);
1673 free_netdev(netdev);
1675 pci_release_selected_regions(pdev,
1676 pci_select_bars(pdev, IORESOURCE_MEM));
1679 pci_disable_device(pdev);
1684 * igb_remove - Device Removal Routine
1685 * @pdev: PCI device information struct
1687 * igb_remove is called by the PCI subsystem to alert the driver
1688 * that it should release a PCI device. The could be caused by a
1689 * Hot-Plug event, or because the driver is going to be removed from
1692 static void __devexit igb_remove(struct pci_dev *pdev)
1694 struct net_device *netdev = pci_get_drvdata(pdev);
1695 struct igb_adapter *adapter = netdev_priv(netdev);
1696 struct e1000_hw *hw = &adapter->hw;
1698 /* flush_scheduled work may reschedule our watchdog task, so
1699 * explicitly disable watchdog tasks from being rescheduled */
1700 set_bit(__IGB_DOWN, &adapter->state);
1701 del_timer_sync(&adapter->watchdog_timer);
1702 del_timer_sync(&adapter->phy_info_timer);
1704 flush_scheduled_work();
1706 #ifdef CONFIG_IGB_DCA
1707 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
1708 dev_info(&pdev->dev, "DCA disabled\n");
1709 dca_remove_requester(&pdev->dev);
1710 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
1711 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
1715 /* Release control of h/w to f/w. If f/w is AMT enabled, this
1716 * would have already happened in close and is redundant. */
1717 igb_release_hw_control(adapter);
1719 unregister_netdev(netdev);
1721 igb_clear_interrupt_scheme(adapter);
1723 #ifdef CONFIG_PCI_IOV
1724 /* reclaim resources allocated to VFs */
1725 if (adapter->vf_data) {
1726 /* disable iov and allow time for transactions to clear */
1727 pci_disable_sriov(pdev);
1730 kfree(adapter->vf_data);
1731 adapter->vf_data = NULL;
1732 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
1734 dev_info(&pdev->dev, "IOV Disabled\n");
1738 iounmap(hw->hw_addr);
1739 if (hw->flash_address)
1740 iounmap(hw->flash_address);
1741 pci_release_selected_regions(pdev,
1742 pci_select_bars(pdev, IORESOURCE_MEM));
1744 free_netdev(netdev);
1746 pci_disable_pcie_error_reporting(pdev);
1748 pci_disable_device(pdev);
1752 * igb_probe_vfs - Initialize vf data storage and add VFs to pci config space
1753 * @adapter: board private structure to initialize
1755 * This function initializes the vf specific data storage and then attempts to
1756 * allocate the VFs. The reason for ordering it this way is because it is much
1757 * mor expensive time wise to disable SR-IOV than it is to allocate and free
1758 * the memory for the VFs.
1760 static void __devinit igb_probe_vfs(struct igb_adapter * adapter)
1762 #ifdef CONFIG_PCI_IOV
1763 struct pci_dev *pdev = adapter->pdev;
1765 if (adapter->vfs_allocated_count > 7)
1766 adapter->vfs_allocated_count = 7;
1768 if (adapter->vfs_allocated_count) {
1769 adapter->vf_data = kcalloc(adapter->vfs_allocated_count,
1770 sizeof(struct vf_data_storage),
1772 /* if allocation failed then we do not support SR-IOV */
1773 if (!adapter->vf_data) {
1774 adapter->vfs_allocated_count = 0;
1775 dev_err(&pdev->dev, "Unable to allocate memory for VF "
1780 if (pci_enable_sriov(pdev, adapter->vfs_allocated_count)) {
1781 kfree(adapter->vf_data);
1782 adapter->vf_data = NULL;
1783 #endif /* CONFIG_PCI_IOV */
1784 adapter->vfs_allocated_count = 0;
1785 #ifdef CONFIG_PCI_IOV
1787 unsigned char mac_addr[ETH_ALEN];
1789 dev_info(&pdev->dev, "%d vfs allocated\n",
1790 adapter->vfs_allocated_count);
1791 for (i = 0; i < adapter->vfs_allocated_count; i++) {
1792 random_ether_addr(mac_addr);
1793 igb_set_vf_mac(adapter, i, mac_addr);
1796 #endif /* CONFIG_PCI_IOV */
1801 * igb_init_hw_timer - Initialize hardware timer used with IEEE 1588 timestamp
1802 * @adapter: board private structure to initialize
1804 * igb_init_hw_timer initializes the function pointer and values for the hw
1805 * timer found in hardware.
1807 static void igb_init_hw_timer(struct igb_adapter *adapter)
1809 struct e1000_hw *hw = &adapter->hw;
1811 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 igb_power_up_link(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 igb_power_down_link(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 device *dev = tx_ring->dev;
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 = dma_alloc_coherent(dev,
2104 tx_ring->next_to_use = 0;
2105 tx_ring->next_to_clean = 0;
2109 vfree(tx_ring->buffer_info);
2111 "Unable to allocate memory for the transmit descriptor ring\n");
2116 * igb_setup_all_tx_resources - wrapper to allocate Tx resources
2117 * (Descriptors) for all queues
2118 * @adapter: board private structure
2120 * Return 0 on success, negative on failure
2122 static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
2124 struct pci_dev *pdev = adapter->pdev;
2127 for (i = 0; i < adapter->num_tx_queues; i++) {
2128 err = igb_setup_tx_resources(adapter->tx_ring[i]);
2131 "Allocation for Tx Queue %u failed\n", i);
2132 for (i--; i >= 0; i--)
2133 igb_free_tx_resources(adapter->tx_ring[i]);
2138 for (i = 0; i < IGB_ABS_MAX_TX_QUEUES; i++) {
2139 int r_idx = i % adapter->num_tx_queues;
2140 adapter->multi_tx_table[i] = adapter->tx_ring[r_idx];
2146 * igb_setup_tctl - configure the transmit control registers
2147 * @adapter: Board private structure
2149 void igb_setup_tctl(struct igb_adapter *adapter)
2151 struct e1000_hw *hw = &adapter->hw;
2154 /* disable queue 0 which is enabled by default on 82575 and 82576 */
2155 wr32(E1000_TXDCTL(0), 0);
2157 /* Program the Transmit Control Register */
2158 tctl = rd32(E1000_TCTL);
2159 tctl &= ~E1000_TCTL_CT;
2160 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2161 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2163 igb_config_collision_dist(hw);
2165 /* Enable transmits */
2166 tctl |= E1000_TCTL_EN;
2168 wr32(E1000_TCTL, tctl);
2172 * igb_configure_tx_ring - Configure transmit ring after Reset
2173 * @adapter: board private structure
2174 * @ring: tx ring to configure
2176 * Configure a transmit ring after a reset.
2178 void igb_configure_tx_ring(struct igb_adapter *adapter,
2179 struct igb_ring *ring)
2181 struct e1000_hw *hw = &adapter->hw;
2183 u64 tdba = ring->dma;
2184 int reg_idx = ring->reg_idx;
2186 /* disable the queue */
2187 txdctl = rd32(E1000_TXDCTL(reg_idx));
2188 wr32(E1000_TXDCTL(reg_idx),
2189 txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
2193 wr32(E1000_TDLEN(reg_idx),
2194 ring->count * sizeof(union e1000_adv_tx_desc));
2195 wr32(E1000_TDBAL(reg_idx),
2196 tdba & 0x00000000ffffffffULL);
2197 wr32(E1000_TDBAH(reg_idx), tdba >> 32);
2199 ring->head = hw->hw_addr + E1000_TDH(reg_idx);
2200 ring->tail = hw->hw_addr + E1000_TDT(reg_idx);
2201 writel(0, ring->head);
2202 writel(0, ring->tail);
2204 txdctl |= IGB_TX_PTHRESH;
2205 txdctl |= IGB_TX_HTHRESH << 8;
2206 txdctl |= IGB_TX_WTHRESH << 16;
2208 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
2209 wr32(E1000_TXDCTL(reg_idx), txdctl);
2213 * igb_configure_tx - Configure transmit Unit after Reset
2214 * @adapter: board private structure
2216 * Configure the Tx unit of the MAC after a reset.
2218 static void igb_configure_tx(struct igb_adapter *adapter)
2222 for (i = 0; i < adapter->num_tx_queues; i++)
2223 igb_configure_tx_ring(adapter, adapter->tx_ring[i]);
2227 * igb_setup_rx_resources - allocate Rx resources (Descriptors)
2228 * @rx_ring: rx descriptor ring (for a specific queue) to setup
2230 * Returns 0 on success, negative on failure
2232 int igb_setup_rx_resources(struct igb_ring *rx_ring)
2234 struct device *dev = rx_ring->dev;
2237 size = sizeof(struct igb_buffer) * rx_ring->count;
2238 rx_ring->buffer_info = vmalloc(size);
2239 if (!rx_ring->buffer_info)
2241 memset(rx_ring->buffer_info, 0, size);
2243 desc_len = sizeof(union e1000_adv_rx_desc);
2245 /* Round up to nearest 4K */
2246 rx_ring->size = rx_ring->count * desc_len;
2247 rx_ring->size = ALIGN(rx_ring->size, 4096);
2249 rx_ring->desc = dma_alloc_coherent(dev,
2257 rx_ring->next_to_clean = 0;
2258 rx_ring->next_to_use = 0;
2263 vfree(rx_ring->buffer_info);
2264 rx_ring->buffer_info = NULL;
2265 dev_err(dev, "Unable to allocate memory for the receive descriptor"
2271 * igb_setup_all_rx_resources - wrapper to allocate Rx resources
2272 * (Descriptors) for all queues
2273 * @adapter: board private structure
2275 * Return 0 on success, negative on failure
2277 static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
2279 struct pci_dev *pdev = adapter->pdev;
2282 for (i = 0; i < adapter->num_rx_queues; i++) {
2283 err = igb_setup_rx_resources(adapter->rx_ring[i]);
2286 "Allocation for Rx Queue %u failed\n", i);
2287 for (i--; i >= 0; i--)
2288 igb_free_rx_resources(adapter->rx_ring[i]);
2297 * igb_setup_mrqc - configure the multiple receive queue control registers
2298 * @adapter: Board private structure
2300 static void igb_setup_mrqc(struct igb_adapter *adapter)
2302 struct e1000_hw *hw = &adapter->hw;
2304 u32 j, num_rx_queues, shift = 0, shift2 = 0;
2309 static const u8 rsshash[40] = {
2310 0x6d, 0x5a, 0x56, 0xda, 0x25, 0x5b, 0x0e, 0xc2, 0x41, 0x67,
2311 0x25, 0x3d, 0x43, 0xa3, 0x8f, 0xb0, 0xd0, 0xca, 0x2b, 0xcb,
2312 0xae, 0x7b, 0x30, 0xb4, 0x77, 0xcb, 0x2d, 0xa3, 0x80, 0x30,
2313 0xf2, 0x0c, 0x6a, 0x42, 0xb7, 0x3b, 0xbe, 0xac, 0x01, 0xfa };
2315 /* Fill out hash function seeds */
2316 for (j = 0; j < 10; j++) {
2317 u32 rsskey = rsshash[(j * 4)];
2318 rsskey |= rsshash[(j * 4) + 1] << 8;
2319 rsskey |= rsshash[(j * 4) + 2] << 16;
2320 rsskey |= rsshash[(j * 4) + 3] << 24;
2321 array_wr32(E1000_RSSRK(0), j, rsskey);
2324 num_rx_queues = adapter->rss_queues;
2326 if (adapter->vfs_allocated_count) {
2327 /* 82575 and 82576 supports 2 RSS queues for VMDq */
2328 switch (hw->mac.type) {
2345 if (hw->mac.type == e1000_82575)
2349 for (j = 0; j < (32 * 4); j++) {
2350 reta.bytes[j & 3] = (j % num_rx_queues) << shift;
2352 reta.bytes[j & 3] |= num_rx_queues << shift2;
2354 wr32(E1000_RETA(j >> 2), reta.dword);
2358 * Disable raw packet checksumming so that RSS hash is placed in
2359 * descriptor on writeback. No need to enable TCP/UDP/IP checksum
2360 * offloads as they are enabled by default
2362 rxcsum = rd32(E1000_RXCSUM);
2363 rxcsum |= E1000_RXCSUM_PCSD;
2365 if (adapter->hw.mac.type >= e1000_82576)
2366 /* Enable Receive Checksum Offload for SCTP */
2367 rxcsum |= E1000_RXCSUM_CRCOFL;
2369 /* Don't need to set TUOFL or IPOFL, they default to 1 */
2370 wr32(E1000_RXCSUM, rxcsum);
2372 /* If VMDq is enabled then we set the appropriate mode for that, else
2373 * we default to RSS so that an RSS hash is calculated per packet even
2374 * if we are only using one queue */
2375 if (adapter->vfs_allocated_count) {
2376 if (hw->mac.type > e1000_82575) {
2377 /* Set the default pool for the PF's first queue */
2378 u32 vtctl = rd32(E1000_VT_CTL);
2379 vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK |
2380 E1000_VT_CTL_DISABLE_DEF_POOL);
2381 vtctl |= adapter->vfs_allocated_count <<
2382 E1000_VT_CTL_DEFAULT_POOL_SHIFT;
2383 wr32(E1000_VT_CTL, vtctl);
2385 if (adapter->rss_queues > 1)
2386 mrqc = E1000_MRQC_ENABLE_VMDQ_RSS_2Q;
2388 mrqc = E1000_MRQC_ENABLE_VMDQ;
2390 mrqc = E1000_MRQC_ENABLE_RSS_4Q;
2392 igb_vmm_control(adapter);
2394 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
2395 E1000_MRQC_RSS_FIELD_IPV4_TCP);
2396 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6 |
2397 E1000_MRQC_RSS_FIELD_IPV6_TCP);
2398 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4_UDP |
2399 E1000_MRQC_RSS_FIELD_IPV6_UDP);
2400 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
2401 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
2403 wr32(E1000_MRQC, mrqc);
2407 * igb_setup_rctl - configure the receive control registers
2408 * @adapter: Board private structure
2410 void igb_setup_rctl(struct igb_adapter *adapter)
2412 struct e1000_hw *hw = &adapter->hw;
2415 rctl = rd32(E1000_RCTL);
2417 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2418 rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
2420 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
2421 (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2424 * enable stripping of CRC. It's unlikely this will break BMC
2425 * redirection as it did with e1000. Newer features require
2426 * that the HW strips the CRC.
2428 rctl |= E1000_RCTL_SECRC;
2430 /* disable store bad packets and clear size bits. */
2431 rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
2433 /* enable LPE to prevent packets larger than max_frame_size */
2434 rctl |= E1000_RCTL_LPE;
2436 /* disable queue 0 to prevent tail write w/o re-config */
2437 wr32(E1000_RXDCTL(0), 0);
2439 /* Attention!!! For SR-IOV PF driver operations you must enable
2440 * queue drop for all VF and PF queues to prevent head of line blocking
2441 * if an un-trusted VF does not provide descriptors to hardware.
2443 if (adapter->vfs_allocated_count) {
2444 /* set all queue drop enable bits */
2445 wr32(E1000_QDE, ALL_QUEUES);
2448 wr32(E1000_RCTL, rctl);
2451 static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
2454 struct e1000_hw *hw = &adapter->hw;
2457 /* if it isn't the PF check to see if VFs are enabled and
2458 * increase the size to support vlan tags */
2459 if (vfn < adapter->vfs_allocated_count &&
2460 adapter->vf_data[vfn].vlans_enabled)
2461 size += VLAN_TAG_SIZE;
2463 vmolr = rd32(E1000_VMOLR(vfn));
2464 vmolr &= ~E1000_VMOLR_RLPML_MASK;
2465 vmolr |= size | E1000_VMOLR_LPE;
2466 wr32(E1000_VMOLR(vfn), vmolr);
2472 * igb_rlpml_set - set maximum receive packet size
2473 * @adapter: board private structure
2475 * Configure maximum receivable packet size.
2477 static void igb_rlpml_set(struct igb_adapter *adapter)
2479 u32 max_frame_size = adapter->max_frame_size;
2480 struct e1000_hw *hw = &adapter->hw;
2481 u16 pf_id = adapter->vfs_allocated_count;
2484 max_frame_size += VLAN_TAG_SIZE;
2486 /* if vfs are enabled we set RLPML to the largest possible request
2487 * size and set the VMOLR RLPML to the size we need */
2489 igb_set_vf_rlpml(adapter, max_frame_size, pf_id);
2490 max_frame_size = MAX_JUMBO_FRAME_SIZE;
2493 wr32(E1000_RLPML, max_frame_size);
2496 static inline void igb_set_vmolr(struct igb_adapter *adapter,
2499 struct e1000_hw *hw = &adapter->hw;
2503 * This register exists only on 82576 and newer so if we are older then
2504 * we should exit and do nothing
2506 if (hw->mac.type < e1000_82576)
2509 vmolr = rd32(E1000_VMOLR(vfn));
2510 vmolr |= E1000_VMOLR_STRVLAN; /* Strip vlan tags */
2512 vmolr |= E1000_VMOLR_AUPE; /* Accept untagged packets */
2514 vmolr &= ~(E1000_VMOLR_AUPE); /* Tagged packets ONLY */
2516 /* clear all bits that might not be set */
2517 vmolr &= ~(E1000_VMOLR_BAM | E1000_VMOLR_RSSE);
2519 if (adapter->rss_queues > 1 && vfn == adapter->vfs_allocated_count)
2520 vmolr |= E1000_VMOLR_RSSE; /* enable RSS */
2522 * for VMDq only allow the VFs and pool 0 to accept broadcast and
2525 if (vfn <= adapter->vfs_allocated_count)
2526 vmolr |= E1000_VMOLR_BAM; /* Accept broadcast */
2528 wr32(E1000_VMOLR(vfn), vmolr);
2532 * igb_configure_rx_ring - Configure a receive ring after Reset
2533 * @adapter: board private structure
2534 * @ring: receive ring to be configured
2536 * Configure the Rx unit of the MAC after a reset.
2538 void igb_configure_rx_ring(struct igb_adapter *adapter,
2539 struct igb_ring *ring)
2541 struct e1000_hw *hw = &adapter->hw;
2542 u64 rdba = ring->dma;
2543 int reg_idx = ring->reg_idx;
2546 /* disable the queue */
2547 rxdctl = rd32(E1000_RXDCTL(reg_idx));
2548 wr32(E1000_RXDCTL(reg_idx),
2549 rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
2551 /* Set DMA base address registers */
2552 wr32(E1000_RDBAL(reg_idx),
2553 rdba & 0x00000000ffffffffULL);
2554 wr32(E1000_RDBAH(reg_idx), rdba >> 32);
2555 wr32(E1000_RDLEN(reg_idx),
2556 ring->count * sizeof(union e1000_adv_rx_desc));
2558 /* initialize head and tail */
2559 ring->head = hw->hw_addr + E1000_RDH(reg_idx);
2560 ring->tail = hw->hw_addr + E1000_RDT(reg_idx);
2561 writel(0, ring->head);
2562 writel(0, ring->tail);
2564 /* set descriptor configuration */
2565 if (ring->rx_buffer_len < IGB_RXBUFFER_1024) {
2566 srrctl = ALIGN(ring->rx_buffer_len, 64) <<
2567 E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
2568 #if (PAGE_SIZE / 2) > IGB_RXBUFFER_16384
2569 srrctl |= IGB_RXBUFFER_16384 >>
2570 E1000_SRRCTL_BSIZEPKT_SHIFT;
2572 srrctl |= (PAGE_SIZE / 2) >>
2573 E1000_SRRCTL_BSIZEPKT_SHIFT;
2575 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2577 srrctl = ALIGN(ring->rx_buffer_len, 1024) >>
2578 E1000_SRRCTL_BSIZEPKT_SHIFT;
2579 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
2581 if (hw->mac.type == e1000_82580)
2582 srrctl |= E1000_SRRCTL_TIMESTAMP;
2583 /* Only set Drop Enable if we are supporting multiple queues */
2584 if (adapter->vfs_allocated_count || adapter->num_rx_queues > 1)
2585 srrctl |= E1000_SRRCTL_DROP_EN;
2587 wr32(E1000_SRRCTL(reg_idx), srrctl);
2589 /* set filtering for VMDQ pools */
2590 igb_set_vmolr(adapter, reg_idx & 0x7, true);
2592 /* enable receive descriptor fetching */
2593 rxdctl = rd32(E1000_RXDCTL(reg_idx));
2594 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
2595 rxdctl &= 0xFFF00000;
2596 rxdctl |= IGB_RX_PTHRESH;
2597 rxdctl |= IGB_RX_HTHRESH << 8;
2598 rxdctl |= IGB_RX_WTHRESH << 16;
2599 wr32(E1000_RXDCTL(reg_idx), rxdctl);
2603 * igb_configure_rx - Configure receive Unit after Reset
2604 * @adapter: board private structure
2606 * Configure the Rx unit of the MAC after a reset.
2608 static void igb_configure_rx(struct igb_adapter *adapter)
2612 /* set UTA to appropriate mode */
2613 igb_set_uta(adapter);
2615 /* set the correct pool for the PF default MAC address in entry 0 */
2616 igb_rar_set_qsel(adapter, adapter->hw.mac.addr, 0,
2617 adapter->vfs_allocated_count);
2619 /* Setup the HW Rx Head and Tail Descriptor Pointers and
2620 * the Base and Length of the Rx Descriptor Ring */
2621 for (i = 0; i < adapter->num_rx_queues; i++)
2622 igb_configure_rx_ring(adapter, adapter->rx_ring[i]);
2626 * igb_free_tx_resources - Free Tx Resources per Queue
2627 * @tx_ring: Tx descriptor ring for a specific queue
2629 * Free all transmit software resources
2631 void igb_free_tx_resources(struct igb_ring *tx_ring)
2633 igb_clean_tx_ring(tx_ring);
2635 vfree(tx_ring->buffer_info);
2636 tx_ring->buffer_info = NULL;
2638 /* if not set, then don't free */
2642 dma_free_coherent(tx_ring->dev, tx_ring->size,
2643 tx_ring->desc, tx_ring->dma);
2645 tx_ring->desc = NULL;
2649 * igb_free_all_tx_resources - Free Tx Resources for All Queues
2650 * @adapter: board private structure
2652 * Free all transmit software resources
2654 static void igb_free_all_tx_resources(struct igb_adapter *adapter)
2658 for (i = 0; i < adapter->num_tx_queues; i++)
2659 igb_free_tx_resources(adapter->tx_ring[i]);
2662 void igb_unmap_and_free_tx_resource(struct igb_ring *tx_ring,
2663 struct igb_buffer *buffer_info)
2665 if (buffer_info->dma) {
2666 if (buffer_info->mapped_as_page)
2667 dma_unmap_page(tx_ring->dev,
2669 buffer_info->length,
2672 dma_unmap_single(tx_ring->dev,
2674 buffer_info->length,
2676 buffer_info->dma = 0;
2678 if (buffer_info->skb) {
2679 dev_kfree_skb_any(buffer_info->skb);
2680 buffer_info->skb = NULL;
2682 buffer_info->time_stamp = 0;
2683 buffer_info->length = 0;
2684 buffer_info->next_to_watch = 0;
2685 buffer_info->mapped_as_page = false;
2689 * igb_clean_tx_ring - Free Tx Buffers
2690 * @tx_ring: ring to be cleaned
2692 static void igb_clean_tx_ring(struct igb_ring *tx_ring)
2694 struct igb_buffer *buffer_info;
2698 if (!tx_ring->buffer_info)
2700 /* Free all the Tx ring sk_buffs */
2702 for (i = 0; i < tx_ring->count; i++) {
2703 buffer_info = &tx_ring->buffer_info[i];
2704 igb_unmap_and_free_tx_resource(tx_ring, buffer_info);
2707 size = sizeof(struct igb_buffer) * tx_ring->count;
2708 memset(tx_ring->buffer_info, 0, size);
2710 /* Zero out the descriptor ring */
2711 memset(tx_ring->desc, 0, tx_ring->size);
2713 tx_ring->next_to_use = 0;
2714 tx_ring->next_to_clean = 0;
2718 * igb_clean_all_tx_rings - Free Tx Buffers for all queues
2719 * @adapter: board private structure
2721 static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
2725 for (i = 0; i < adapter->num_tx_queues; i++)
2726 igb_clean_tx_ring(adapter->tx_ring[i]);
2730 * igb_free_rx_resources - Free Rx Resources
2731 * @rx_ring: ring to clean the resources from
2733 * Free all receive software resources
2735 void igb_free_rx_resources(struct igb_ring *rx_ring)
2737 igb_clean_rx_ring(rx_ring);
2739 vfree(rx_ring->buffer_info);
2740 rx_ring->buffer_info = NULL;
2742 /* if not set, then don't free */
2746 dma_free_coherent(rx_ring->dev, rx_ring->size,
2747 rx_ring->desc, rx_ring->dma);
2749 rx_ring->desc = NULL;
2753 * igb_free_all_rx_resources - Free Rx Resources for All Queues
2754 * @adapter: board private structure
2756 * Free all receive software resources
2758 static void igb_free_all_rx_resources(struct igb_adapter *adapter)
2762 for (i = 0; i < adapter->num_rx_queues; i++)
2763 igb_free_rx_resources(adapter->rx_ring[i]);
2767 * igb_clean_rx_ring - Free Rx Buffers per Queue
2768 * @rx_ring: ring to free buffers from
2770 static void igb_clean_rx_ring(struct igb_ring *rx_ring)
2772 struct igb_buffer *buffer_info;
2776 if (!rx_ring->buffer_info)
2779 /* Free all the Rx ring sk_buffs */
2780 for (i = 0; i < rx_ring->count; i++) {
2781 buffer_info = &rx_ring->buffer_info[i];
2782 if (buffer_info->dma) {
2783 dma_unmap_single(rx_ring->dev,
2785 rx_ring->rx_buffer_len,
2787 buffer_info->dma = 0;
2790 if (buffer_info->skb) {
2791 dev_kfree_skb(buffer_info->skb);
2792 buffer_info->skb = NULL;
2794 if (buffer_info->page_dma) {
2795 dma_unmap_page(rx_ring->dev,
2796 buffer_info->page_dma,
2799 buffer_info->page_dma = 0;
2801 if (buffer_info->page) {
2802 put_page(buffer_info->page);
2803 buffer_info->page = NULL;
2804 buffer_info->page_offset = 0;
2808 size = sizeof(struct igb_buffer) * rx_ring->count;
2809 memset(rx_ring->buffer_info, 0, size);
2811 /* Zero out the descriptor ring */
2812 memset(rx_ring->desc, 0, rx_ring->size);
2814 rx_ring->next_to_clean = 0;
2815 rx_ring->next_to_use = 0;
2819 * igb_clean_all_rx_rings - Free Rx Buffers for all queues
2820 * @adapter: board private structure
2822 static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
2826 for (i = 0; i < adapter->num_rx_queues; i++)
2827 igb_clean_rx_ring(adapter->rx_ring[i]);
2831 * igb_set_mac - Change the Ethernet Address of the NIC
2832 * @netdev: network interface device structure
2833 * @p: pointer to an address structure
2835 * Returns 0 on success, negative on failure
2837 static int igb_set_mac(struct net_device *netdev, void *p)
2839 struct igb_adapter *adapter = netdev_priv(netdev);
2840 struct e1000_hw *hw = &adapter->hw;
2841 struct sockaddr *addr = p;
2843 if (!is_valid_ether_addr(addr->sa_data))
2844 return -EADDRNOTAVAIL;
2846 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2847 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
2849 /* set the correct pool for the new PF MAC address in entry 0 */
2850 igb_rar_set_qsel(adapter, hw->mac.addr, 0,
2851 adapter->vfs_allocated_count);
2857 * igb_write_mc_addr_list - write multicast addresses to MTA
2858 * @netdev: network interface device structure
2860 * Writes multicast address list to the MTA hash table.
2861 * Returns: -ENOMEM on failure
2862 * 0 on no addresses written
2863 * X on writing X addresses to MTA
2865 static int igb_write_mc_addr_list(struct net_device *netdev)
2867 struct igb_adapter *adapter = netdev_priv(netdev);
2868 struct e1000_hw *hw = &adapter->hw;
2869 struct netdev_hw_addr *ha;
2873 if (netdev_mc_empty(netdev)) {
2874 /* nothing to program, so clear mc list */
2875 igb_update_mc_addr_list(hw, NULL, 0);
2876 igb_restore_vf_multicasts(adapter);
2880 mta_list = kzalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
2884 /* The shared function expects a packed array of only addresses. */
2886 netdev_for_each_mc_addr(ha, netdev)
2887 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
2889 igb_update_mc_addr_list(hw, mta_list, i);
2892 return netdev_mc_count(netdev);
2896 * igb_write_uc_addr_list - write unicast addresses to RAR table
2897 * @netdev: network interface device structure
2899 * Writes unicast address list to the RAR table.
2900 * Returns: -ENOMEM on failure/insufficient address space
2901 * 0 on no addresses written
2902 * X on writing X addresses to the RAR table
2904 static int igb_write_uc_addr_list(struct net_device *netdev)
2906 struct igb_adapter *adapter = netdev_priv(netdev);
2907 struct e1000_hw *hw = &adapter->hw;
2908 unsigned int vfn = adapter->vfs_allocated_count;
2909 unsigned int rar_entries = hw->mac.rar_entry_count - (vfn + 1);
2912 /* return ENOMEM indicating insufficient memory for addresses */
2913 if (netdev_uc_count(netdev) > rar_entries)
2916 if (!netdev_uc_empty(netdev) && rar_entries) {
2917 struct netdev_hw_addr *ha;
2919 netdev_for_each_uc_addr(ha, netdev) {
2922 igb_rar_set_qsel(adapter, ha->addr,
2928 /* write the addresses in reverse order to avoid write combining */
2929 for (; rar_entries > 0 ; rar_entries--) {
2930 wr32(E1000_RAH(rar_entries), 0);
2931 wr32(E1000_RAL(rar_entries), 0);
2939 * igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
2940 * @netdev: network interface device structure
2942 * The set_rx_mode entry point is called whenever the unicast or multicast
2943 * address lists or the network interface flags are updated. This routine is
2944 * responsible for configuring the hardware for proper unicast, multicast,
2945 * promiscuous mode, and all-multi behavior.
2947 static void igb_set_rx_mode(struct net_device *netdev)
2949 struct igb_adapter *adapter = netdev_priv(netdev);
2950 struct e1000_hw *hw = &adapter->hw;
2951 unsigned int vfn = adapter->vfs_allocated_count;
2952 u32 rctl, vmolr = 0;
2955 /* Check for Promiscuous and All Multicast modes */
2956 rctl = rd32(E1000_RCTL);
2958 /* clear the effected bits */
2959 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE | E1000_RCTL_VFE);
2961 if (netdev->flags & IFF_PROMISC) {
2962 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2963 vmolr |= (E1000_VMOLR_ROPE | E1000_VMOLR_MPME);
2965 if (netdev->flags & IFF_ALLMULTI) {
2966 rctl |= E1000_RCTL_MPE;
2967 vmolr |= E1000_VMOLR_MPME;
2970 * Write addresses to the MTA, if the attempt fails
2971 * then we should just turn on promiscous mode so
2972 * that we can at least receive multicast traffic
2974 count = igb_write_mc_addr_list(netdev);
2976 rctl |= E1000_RCTL_MPE;
2977 vmolr |= E1000_VMOLR_MPME;
2979 vmolr |= E1000_VMOLR_ROMPE;
2983 * Write addresses to available RAR registers, if there is not
2984 * sufficient space to store all the addresses then enable
2985 * unicast promiscous mode
2987 count = igb_write_uc_addr_list(netdev);
2989 rctl |= E1000_RCTL_UPE;
2990 vmolr |= E1000_VMOLR_ROPE;
2992 rctl |= E1000_RCTL_VFE;
2994 wr32(E1000_RCTL, rctl);
2997 * In order to support SR-IOV and eventually VMDq it is necessary to set
2998 * the VMOLR to enable the appropriate modes. Without this workaround
2999 * we will have issues with VLAN tag stripping not being done for frames
3000 * that are only arriving because we are the default pool
3002 if (hw->mac.type < e1000_82576)
3005 vmolr |= rd32(E1000_VMOLR(vfn)) &
3006 ~(E1000_VMOLR_ROPE | E1000_VMOLR_MPME | E1000_VMOLR_ROMPE);
3007 wr32(E1000_VMOLR(vfn), vmolr);
3008 igb_restore_vf_multicasts(adapter);
3011 /* Need to wait a few seconds after link up to get diagnostic information from
3013 static void igb_update_phy_info(unsigned long data)
3015 struct igb_adapter *adapter = (struct igb_adapter *) data;
3016 igb_get_phy_info(&adapter->hw);
3020 * igb_has_link - check shared code for link and determine up/down
3021 * @adapter: pointer to driver private info
3023 bool igb_has_link(struct igb_adapter *adapter)
3025 struct e1000_hw *hw = &adapter->hw;
3026 bool link_active = false;
3029 /* get_link_status is set on LSC (link status) interrupt or
3030 * rx sequence error interrupt. get_link_status will stay
3031 * false until the e1000_check_for_link establishes link
3032 * for copper adapters ONLY
3034 switch (hw->phy.media_type) {
3035 case e1000_media_type_copper:
3036 if (hw->mac.get_link_status) {
3037 ret_val = hw->mac.ops.check_for_link(hw);
3038 link_active = !hw->mac.get_link_status;
3043 case e1000_media_type_internal_serdes:
3044 ret_val = hw->mac.ops.check_for_link(hw);
3045 link_active = hw->mac.serdes_has_link;
3048 case e1000_media_type_unknown:
3056 * igb_watchdog - Timer Call-back
3057 * @data: pointer to adapter cast into an unsigned long
3059 static void igb_watchdog(unsigned long data)
3061 struct igb_adapter *adapter = (struct igb_adapter *)data;
3062 /* Do the rest outside of interrupt context */
3063 schedule_work(&adapter->watchdog_task);
3066 static void igb_watchdog_task(struct work_struct *work)
3068 struct igb_adapter *adapter = container_of(work,
3071 struct e1000_hw *hw = &adapter->hw;
3072 struct net_device *netdev = adapter->netdev;
3076 link = igb_has_link(adapter);
3078 if (!netif_carrier_ok(netdev)) {
3080 hw->mac.ops.get_speed_and_duplex(hw,
3081 &adapter->link_speed,
3082 &adapter->link_duplex);
3084 ctrl = rd32(E1000_CTRL);
3085 /* Links status message must follow this format */
3086 printk(KERN_INFO "igb: %s NIC Link is Up %d Mbps %s, "
3087 "Flow Control: %s\n",
3089 adapter->link_speed,
3090 adapter->link_duplex == FULL_DUPLEX ?
3091 "Full Duplex" : "Half Duplex",
3092 ((ctrl & E1000_CTRL_TFCE) &&
3093 (ctrl & E1000_CTRL_RFCE)) ? "RX/TX" :
3094 ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3095 ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None")));
3097 /* adjust timeout factor according to speed/duplex */
3098 adapter->tx_timeout_factor = 1;
3099 switch (adapter->link_speed) {
3101 adapter->tx_timeout_factor = 14;
3104 /* maybe add some timeout factor ? */
3108 netif_carrier_on(netdev);
3110 igb_ping_all_vfs(adapter);
3112 /* link state has changed, schedule phy info update */
3113 if (!test_bit(__IGB_DOWN, &adapter->state))
3114 mod_timer(&adapter->phy_info_timer,
3115 round_jiffies(jiffies + 2 * HZ));
3118 if (netif_carrier_ok(netdev)) {
3119 adapter->link_speed = 0;
3120 adapter->link_duplex = 0;
3121 /* Links status message must follow this format */
3122 printk(KERN_INFO "igb: %s NIC Link is Down\n",
3124 netif_carrier_off(netdev);
3126 igb_ping_all_vfs(adapter);
3128 /* link state has changed, schedule phy info update */
3129 if (!test_bit(__IGB_DOWN, &adapter->state))
3130 mod_timer(&adapter->phy_info_timer,
3131 round_jiffies(jiffies + 2 * HZ));
3135 igb_update_stats(adapter);
3137 for (i = 0; i < adapter->num_tx_queues; i++) {
3138 struct igb_ring *tx_ring = adapter->tx_ring[i];
3139 if (!netif_carrier_ok(netdev)) {
3140 /* We've lost link, so the controller stops DMA,
3141 * but we've got queued Tx work that's never going
3142 * to get done, so reset controller to flush Tx.
3143 * (Do the reset outside of interrupt context). */
3144 if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) {
3145 adapter->tx_timeout_count++;
3146 schedule_work(&adapter->reset_task);
3147 /* return immediately since reset is imminent */
3152 /* Force detection of hung controller every watchdog period */
3153 tx_ring->detect_tx_hung = true;
3156 /* Cause software interrupt to ensure rx ring is cleaned */
3157 if (adapter->msix_entries) {
3159 for (i = 0; i < adapter->num_q_vectors; i++) {
3160 struct igb_q_vector *q_vector = adapter->q_vector[i];
3161 eics |= q_vector->eims_value;
3163 wr32(E1000_EICS, eics);
3165 wr32(E1000_ICS, E1000_ICS_RXDMT0);
3168 /* Reset the timer */
3169 if (!test_bit(__IGB_DOWN, &adapter->state))
3170 mod_timer(&adapter->watchdog_timer,
3171 round_jiffies(jiffies + 2 * HZ));
3174 enum latency_range {
3178 latency_invalid = 255
3182 * igb_update_ring_itr - update the dynamic ITR value based on packet size
3184 * Stores a new ITR value based on strictly on packet size. This
3185 * algorithm is less sophisticated than that used in igb_update_itr,
3186 * due to the difficulty of synchronizing statistics across multiple
3187 * receive rings. The divisors and thresholds used by this fuction
3188 * were determined based on theoretical maximum wire speed and testing
3189 * data, in order to minimize response time while increasing bulk
3191 * This functionality is controlled by the InterruptThrottleRate module
3192 * parameter (see igb_param.c)
3193 * NOTE: This function is called only when operating in a multiqueue
3194 * receive environment.
3195 * @q_vector: pointer to q_vector
3197 static void igb_update_ring_itr(struct igb_q_vector *q_vector)
3199 int new_val = q_vector->itr_val;
3200 int avg_wire_size = 0;
3201 struct igb_adapter *adapter = q_vector->adapter;
3203 /* For non-gigabit speeds, just fix the interrupt rate at 4000
3204 * ints/sec - ITR timer value of 120 ticks.
3206 if (adapter->link_speed != SPEED_1000) {
3211 if (q_vector->rx_ring && q_vector->rx_ring->total_packets) {
3212 struct igb_ring *ring = q_vector->rx_ring;
3213 avg_wire_size = ring->total_bytes / ring->total_packets;
3216 if (q_vector->tx_ring && q_vector->tx_ring->total_packets) {
3217 struct igb_ring *ring = q_vector->tx_ring;
3218 avg_wire_size = max_t(u32, avg_wire_size,
3219 (ring->total_bytes /
3220 ring->total_packets));
3223 /* if avg_wire_size isn't set no work was done */
3227 /* Add 24 bytes to size to account for CRC, preamble, and gap */
3228 avg_wire_size += 24;
3230 /* Don't starve jumbo frames */
3231 avg_wire_size = min(avg_wire_size, 3000);
3233 /* Give a little boost to mid-size frames */
3234 if ((avg_wire_size > 300) && (avg_wire_size < 1200))
3235 new_val = avg_wire_size / 3;
3237 new_val = avg_wire_size / 2;
3239 /* when in itr mode 3 do not exceed 20K ints/sec */
3240 if (adapter->rx_itr_setting == 3 && new_val < 196)
3244 if (new_val != q_vector->itr_val) {
3245 q_vector->itr_val = new_val;
3246 q_vector->set_itr = 1;
3249 if (q_vector->rx_ring) {
3250 q_vector->rx_ring->total_bytes = 0;
3251 q_vector->rx_ring->total_packets = 0;
3253 if (q_vector->tx_ring) {
3254 q_vector->tx_ring->total_bytes = 0;
3255 q_vector->tx_ring->total_packets = 0;
3260 * igb_update_itr - update the dynamic ITR value based on statistics
3261 * Stores a new ITR value based on packets and byte
3262 * counts during the last interrupt. The advantage of per interrupt
3263 * computation is faster updates and more accurate ITR for the current
3264 * traffic pattern. Constants in this function were computed
3265 * based on theoretical maximum wire speed and thresholds were set based
3266 * on testing data as well as attempting to minimize response time
3267 * while increasing bulk throughput.
3268 * this functionality is controlled by the InterruptThrottleRate module
3269 * parameter (see igb_param.c)
3270 * NOTE: These calculations are only valid when operating in a single-
3271 * queue environment.
3272 * @adapter: pointer to adapter
3273 * @itr_setting: current q_vector->itr_val
3274 * @packets: the number of packets during this measurement interval
3275 * @bytes: the number of bytes during this measurement interval
3277 static unsigned int igb_update_itr(struct igb_adapter *adapter, u16 itr_setting,
3278 int packets, int bytes)
3280 unsigned int retval = itr_setting;
3283 goto update_itr_done;
3285 switch (itr_setting) {
3286 case lowest_latency:
3287 /* handle TSO and jumbo frames */
3288 if (bytes/packets > 8000)
3289 retval = bulk_latency;
3290 else if ((packets < 5) && (bytes > 512))
3291 retval = low_latency;
3293 case low_latency: /* 50 usec aka 20000 ints/s */
3294 if (bytes > 10000) {
3295 /* this if handles the TSO accounting */
3296 if (bytes/packets > 8000) {
3297 retval = bulk_latency;
3298 } else if ((packets < 10) || ((bytes/packets) > 1200)) {
3299 retval = bulk_latency;
3300 } else if ((packets > 35)) {
3301 retval = lowest_latency;
3303 } else if (bytes/packets > 2000) {
3304 retval = bulk_latency;
3305 } else if (packets <= 2 && bytes < 512) {
3306 retval = lowest_latency;
3309 case bulk_latency: /* 250 usec aka 4000 ints/s */
3310 if (bytes > 25000) {
3312 retval = low_latency;
3313 } else if (bytes < 1500) {
3314 retval = low_latency;
3323 static void igb_set_itr(struct igb_adapter *adapter)
3325 struct igb_q_vector *q_vector = adapter->q_vector[0];
3327 u32 new_itr = q_vector->itr_val;
3329 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
3330 if (adapter->link_speed != SPEED_1000) {
3336 adapter->rx_itr = igb_update_itr(adapter,
3338 q_vector->rx_ring->total_packets,
3339 q_vector->rx_ring->total_bytes);
3341 adapter->tx_itr = igb_update_itr(adapter,
3343 q_vector->tx_ring->total_packets,
3344 q_vector->tx_ring->total_bytes);
3345 current_itr = max(adapter->rx_itr, adapter->tx_itr);
3347 /* conservative mode (itr 3) eliminates the lowest_latency setting */
3348 if (adapter->rx_itr_setting == 3 && current_itr == lowest_latency)
3349 current_itr = low_latency;
3351 switch (current_itr) {
3352 /* counts and packets in update_itr are dependent on these numbers */
3353 case lowest_latency:
3354 new_itr = 56; /* aka 70,000 ints/sec */
3357 new_itr = 196; /* aka 20,000 ints/sec */
3360 new_itr = 980; /* aka 4,000 ints/sec */
3367 q_vector->rx_ring->total_bytes = 0;
3368 q_vector->rx_ring->total_packets = 0;
3369 q_vector->tx_ring->total_bytes = 0;
3370 q_vector->tx_ring->total_packets = 0;
3372 if (new_itr != q_vector->itr_val) {
3373 /* this attempts to bias the interrupt rate towards Bulk
3374 * by adding intermediate steps when interrupt rate is
3376 new_itr = new_itr > q_vector->itr_val ?
3377 max((new_itr * q_vector->itr_val) /
3378 (new_itr + (q_vector->itr_val >> 2)),
3381 /* Don't write the value here; it resets the adapter's
3382 * internal timer, and causes us to delay far longer than
3383 * we should between interrupts. Instead, we write the ITR
3384 * value at the beginning of the next interrupt so the timing
3385 * ends up being correct.
3387 q_vector->itr_val = new_itr;
3388 q_vector->set_itr = 1;
3394 #define IGB_TX_FLAGS_CSUM 0x00000001
3395 #define IGB_TX_FLAGS_VLAN 0x00000002
3396 #define IGB_TX_FLAGS_TSO 0x00000004
3397 #define IGB_TX_FLAGS_IPV4 0x00000008
3398 #define IGB_TX_FLAGS_TSTAMP 0x00000010
3399 #define IGB_TX_FLAGS_VLAN_MASK 0xffff0000
3400 #define IGB_TX_FLAGS_VLAN_SHIFT 16
3402 static inline int igb_tso_adv(struct igb_ring *tx_ring,
3403 struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
3405 struct e1000_adv_tx_context_desc *context_desc;
3408 struct igb_buffer *buffer_info;
3409 u32 info = 0, tu_cmd = 0;
3413 if (skb_header_cloned(skb)) {
3414 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3419 l4len = tcp_hdrlen(skb);
3422 if (skb->protocol == htons(ETH_P_IP)) {
3423 struct iphdr *iph = ip_hdr(skb);
3426 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3430 } else if (skb_is_gso_v6(skb)) {
3431 ipv6_hdr(skb)->payload_len = 0;
3432 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3433 &ipv6_hdr(skb)->daddr,
3437 i = tx_ring->next_to_use;
3439 buffer_info = &tx_ring->buffer_info[i];
3440 context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
3441 /* VLAN MACLEN IPLEN */
3442 if (tx_flags & IGB_TX_FLAGS_VLAN)
3443 info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
3444 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
3445 *hdr_len += skb_network_offset(skb);
3446 info |= skb_network_header_len(skb);
3447 *hdr_len += skb_network_header_len(skb);
3448 context_desc->vlan_macip_lens = cpu_to_le32(info);
3450 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3451 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
3453 if (skb->protocol == htons(ETH_P_IP))
3454 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
3455 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
3457 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
3460 mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
3461 mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
3463 /* For 82575, context index must be unique per ring. */
3464 if (tx_ring->flags & IGB_RING_FLAG_TX_CTX_IDX)
3465 mss_l4len_idx |= tx_ring->reg_idx << 4;
3467 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
3468 context_desc->seqnum_seed = 0;
3470 buffer_info->time_stamp = jiffies;
3471 buffer_info->next_to_watch = i;
3472 buffer_info->dma = 0;
3474 if (i == tx_ring->count)
3477 tx_ring->next_to_use = i;
3482 static inline bool igb_tx_csum_adv(struct igb_ring *tx_ring,
3483 struct sk_buff *skb, u32 tx_flags)
3485 struct e1000_adv_tx_context_desc *context_desc;
3486 struct device *dev = tx_ring->dev;
3487 struct igb_buffer *buffer_info;
3488 u32 info = 0, tu_cmd = 0;
3491 if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
3492 (tx_flags & IGB_TX_FLAGS_VLAN)) {
3493 i = tx_ring->next_to_use;
3494 buffer_info = &tx_ring->buffer_info[i];
3495 context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
3497 if (tx_flags & IGB_TX_FLAGS_VLAN)
3498 info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
3500 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
3501 if (skb->ip_summed == CHECKSUM_PARTIAL)
3502 info |= skb_network_header_len(skb);
3504 context_desc->vlan_macip_lens = cpu_to_le32(info);
3506 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
3508 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3511 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3512 const struct vlan_ethhdr *vhdr =
3513 (const struct vlan_ethhdr*)skb->data;
3515 protocol = vhdr->h_vlan_encapsulated_proto;
3517 protocol = skb->protocol;
3521 case cpu_to_be16(ETH_P_IP):
3522 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
3523 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
3524 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
3525 else if (ip_hdr(skb)->protocol == IPPROTO_SCTP)
3526 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_SCTP;
3528 case cpu_to_be16(ETH_P_IPV6):
3529 /* XXX what about other V6 headers?? */
3530 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3531 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
3532 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_SCTP)
3533 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_SCTP;
3536 if (unlikely(net_ratelimit()))
3538 "partial checksum but proto=%x!\n",
3544 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
3545 context_desc->seqnum_seed = 0;
3546 if (tx_ring->flags & IGB_RING_FLAG_TX_CTX_IDX)
3547 context_desc->mss_l4len_idx =
3548 cpu_to_le32(tx_ring->reg_idx << 4);
3550 buffer_info->time_stamp = jiffies;
3551 buffer_info->next_to_watch = i;
3552 buffer_info->dma = 0;
3555 if (i == tx_ring->count)
3557 tx_ring->next_to_use = i;
3564 #define IGB_MAX_TXD_PWR 16
3565 #define IGB_MAX_DATA_PER_TXD (1<<IGB_MAX_TXD_PWR)
3567 static inline int igb_tx_map_adv(struct igb_ring *tx_ring, struct sk_buff *skb,
3570 struct igb_buffer *buffer_info;
3571 struct device *dev = tx_ring->dev;
3572 unsigned int len = skb_headlen(skb);
3573 unsigned int count = 0, i;
3576 i = tx_ring->next_to_use;
3578 buffer_info = &tx_ring->buffer_info[i];
3579 BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
3580 buffer_info->length = len;
3581 /* set time_stamp *before* dma to help avoid a possible race */
3582 buffer_info->time_stamp = jiffies;
3583 buffer_info->next_to_watch = i;
3584 buffer_info->dma = dma_map_single(dev, skb->data, len,
3586 if (dma_mapping_error(dev, buffer_info->dma))
3589 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
3590 struct skb_frag_struct *frag;
3594 if (i == tx_ring->count)
3597 frag = &skb_shinfo(skb)->frags[f];
3600 buffer_info = &tx_ring->buffer_info[i];
3601 BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
3602 buffer_info->length = len;
3603 buffer_info->time_stamp = jiffies;
3604 buffer_info->next_to_watch = i;
3605 buffer_info->mapped_as_page = true;
3606 buffer_info->dma = dma_map_page(dev,
3611 if (dma_mapping_error(dev, buffer_info->dma))
3616 tx_ring->buffer_info[i].skb = skb;
3617 tx_ring->buffer_info[i].gso_segs = skb_shinfo(skb)->gso_segs ?: 1;
3618 tx_ring->buffer_info[first].next_to_watch = i;
3623 dev_err(dev, "TX DMA map failed\n");
3625 /* clear timestamp and dma mappings for failed buffer_info mapping */
3626 buffer_info->dma = 0;
3627 buffer_info->time_stamp = 0;
3628 buffer_info->length = 0;
3629 buffer_info->next_to_watch = 0;
3630 buffer_info->mapped_as_page = false;
3632 /* clear timestamp and dma mappings for remaining portion of packet */
3637 buffer_info = &tx_ring->buffer_info[i];
3638 igb_unmap_and_free_tx_resource(tx_ring, buffer_info);
3644 static inline void igb_tx_queue_adv(struct igb_ring *tx_ring,
3645 u32 tx_flags, int count, u32 paylen,
3648 union e1000_adv_tx_desc *tx_desc;
3649 struct igb_buffer *buffer_info;
3650 u32 olinfo_status = 0, cmd_type_len;
3651 unsigned int i = tx_ring->next_to_use;
3653 cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
3654 E1000_ADVTXD_DCMD_DEXT);
3656 if (tx_flags & IGB_TX_FLAGS_VLAN)
3657 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
3659 if (tx_flags & IGB_TX_FLAGS_TSTAMP)
3660 cmd_type_len |= E1000_ADVTXD_MAC_TSTAMP;
3662 if (tx_flags & IGB_TX_FLAGS_TSO) {
3663 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
3665 /* insert tcp checksum */
3666 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
3668 /* insert ip checksum */
3669 if (tx_flags & IGB_TX_FLAGS_IPV4)
3670 olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
3672 } else if (tx_flags & IGB_TX_FLAGS_CSUM) {
3673 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
3676 if ((tx_ring->flags & IGB_RING_FLAG_TX_CTX_IDX) &&
3677 (tx_flags & (IGB_TX_FLAGS_CSUM |
3679 IGB_TX_FLAGS_VLAN)))
3680 olinfo_status |= tx_ring->reg_idx << 4;
3682 olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
3685 buffer_info = &tx_ring->buffer_info[i];
3686 tx_desc = E1000_TX_DESC_ADV(*tx_ring, i);
3687 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
3688 tx_desc->read.cmd_type_len =
3689 cpu_to_le32(cmd_type_len | buffer_info->length);
3690 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
3693 if (i == tx_ring->count)
3695 } while (count > 0);
3697 tx_desc->read.cmd_type_len |= cpu_to_le32(IGB_ADVTXD_DCMD);
3698 /* Force memory writes to complete before letting h/w
3699 * know there are new descriptors to fetch. (Only
3700 * applicable for weak-ordered memory model archs,
3701 * such as IA-64). */
3704 tx_ring->next_to_use = i;
3705 writel(i, tx_ring->tail);
3706 /* we need this if more than one processor can write to our tail
3707 * at a time, it syncronizes IO on IA64/Altix systems */
3711 static int __igb_maybe_stop_tx(struct igb_ring *tx_ring, int size)
3713 struct net_device *netdev = tx_ring->netdev;
3715 netif_stop_subqueue(netdev, tx_ring->queue_index);
3717 /* Herbert's original patch had:
3718 * smp_mb__after_netif_stop_queue();
3719 * but since that doesn't exist yet, just open code it. */
3722 /* We need to check again in a case another CPU has just
3723 * made room available. */
3724 if (igb_desc_unused(tx_ring) < size)
3728 netif_wake_subqueue(netdev, tx_ring->queue_index);
3729 tx_ring->tx_stats.restart_queue++;
3733 static inline int igb_maybe_stop_tx(struct igb_ring *tx_ring, int size)
3735 if (igb_desc_unused(tx_ring) >= size)
3737 return __igb_maybe_stop_tx(tx_ring, size);
3740 netdev_tx_t igb_xmit_frame_ring_adv(struct sk_buff *skb,
3741 struct igb_ring *tx_ring)
3743 struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
3748 union skb_shared_tx *shtx = skb_tx(skb);
3750 /* need: 1 descriptor per page,
3751 * + 2 desc gap to keep tail from touching head,
3752 * + 1 desc for skb->data,
3753 * + 1 desc for context descriptor,
3754 * otherwise try next time */
3755 if (igb_maybe_stop_tx(tx_ring, skb_shinfo(skb)->nr_frags + 4)) {
3756 /* this is a hard error */
3757 return NETDEV_TX_BUSY;
3760 if (unlikely(shtx->hardware)) {
3761 shtx->in_progress = 1;
3762 tx_flags |= IGB_TX_FLAGS_TSTAMP;
3765 if (vlan_tx_tag_present(skb) && adapter->vlgrp) {
3766 tx_flags |= IGB_TX_FLAGS_VLAN;
3767 tx_flags |= (vlan_tx_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
3770 if (skb->protocol == htons(ETH_P_IP))
3771 tx_flags |= IGB_TX_FLAGS_IPV4;
3773 first = tx_ring->next_to_use;
3774 if (skb_is_gso(skb)) {
3775 tso = igb_tso_adv(tx_ring, skb, tx_flags, &hdr_len);
3778 dev_kfree_skb_any(skb);
3779 return NETDEV_TX_OK;
3784 tx_flags |= IGB_TX_FLAGS_TSO;
3785 else if (igb_tx_csum_adv(tx_ring, skb, tx_flags) &&
3786 (skb->ip_summed == CHECKSUM_PARTIAL))
3787 tx_flags |= IGB_TX_FLAGS_CSUM;
3790 * count reflects descriptors mapped, if 0 or less then mapping error
3791 * has occured and we need to rewind the descriptor queue
3793 count = igb_tx_map_adv(tx_ring, skb, first);
3795 dev_kfree_skb_any(skb);
3796 tx_ring->buffer_info[first].time_stamp = 0;
3797 tx_ring->next_to_use = first;
3798 return NETDEV_TX_OK;
3801 igb_tx_queue_adv(tx_ring, tx_flags, count, skb->len, hdr_len);
3803 /* Make sure there is space in the ring for the next send. */
3804 igb_maybe_stop_tx(tx_ring, MAX_SKB_FRAGS + 4);
3806 return NETDEV_TX_OK;
3809 static netdev_tx_t igb_xmit_frame_adv(struct sk_buff *skb,
3810 struct net_device *netdev)
3812 struct igb_adapter *adapter = netdev_priv(netdev);
3813 struct igb_ring *tx_ring;
3816 if (test_bit(__IGB_DOWN, &adapter->state)) {
3817 dev_kfree_skb_any(skb);
3818 return NETDEV_TX_OK;
3821 if (skb->len <= 0) {
3822 dev_kfree_skb_any(skb);
3823 return NETDEV_TX_OK;
3826 r_idx = skb->queue_mapping & (IGB_ABS_MAX_TX_QUEUES - 1);
3827 tx_ring = adapter->multi_tx_table[r_idx];
3829 /* This goes back to the question of how to logically map a tx queue
3830 * to a flow. Right now, performance is impacted slightly negatively
3831 * if using multiple tx queues. If the stack breaks away from a
3832 * single qdisc implementation, we can look at this again. */
3833 return igb_xmit_frame_ring_adv(skb, tx_ring);
3837 * igb_tx_timeout - Respond to a Tx Hang
3838 * @netdev: network interface device structure
3840 static void igb_tx_timeout(struct net_device *netdev)
3842 struct igb_adapter *adapter = netdev_priv(netdev);
3843 struct e1000_hw *hw = &adapter->hw;
3845 /* Do the reset outside of interrupt context */
3846 adapter->tx_timeout_count++;
3848 if (hw->mac.type == e1000_82580)
3849 hw->dev_spec._82575.global_device_reset = true;
3851 schedule_work(&adapter->reset_task);
3853 (adapter->eims_enable_mask & ~adapter->eims_other));
3856 static void igb_reset_task(struct work_struct *work)
3858 struct igb_adapter *adapter;
3859 adapter = container_of(work, struct igb_adapter, reset_task);
3861 igb_reinit_locked(adapter);
3865 * igb_get_stats - Get System Network Statistics
3866 * @netdev: network interface device structure
3868 * Returns the address of the device statistics structure.
3869 * The statistics are actually updated from the timer callback.
3871 static struct net_device_stats *igb_get_stats(struct net_device *netdev)
3873 /* only return the current stats */
3874 return &netdev->stats;
3878 * igb_change_mtu - Change the Maximum Transfer Unit
3879 * @netdev: network interface device structure
3880 * @new_mtu: new value for maximum frame size
3882 * Returns 0 on success, negative on failure
3884 static int igb_change_mtu(struct net_device *netdev, int new_mtu)
3886 struct igb_adapter *adapter = netdev_priv(netdev);
3887 struct pci_dev *pdev = adapter->pdev;
3888 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3889 u32 rx_buffer_len, i;
3891 if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3892 dev_err(&pdev->dev, "Invalid MTU setting\n");
3896 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
3897 dev_err(&pdev->dev, "MTU > 9216 not supported.\n");
3901 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
3904 /* igb_down has a dependency on max_frame_size */
3905 adapter->max_frame_size = max_frame;
3907 /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3908 * means we reserve 2 more, this pushes us to allocate from the next
3910 * i.e. RXBUFFER_2048 --> size-4096 slab
3913 if (adapter->hw.mac.type == e1000_82580)
3914 max_frame += IGB_TS_HDR_LEN;
3916 if (max_frame <= IGB_RXBUFFER_1024)
3917 rx_buffer_len = IGB_RXBUFFER_1024;
3918 else if (max_frame <= MAXIMUM_ETHERNET_VLAN_SIZE)
3919 rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3921 rx_buffer_len = IGB_RXBUFFER_128;
3923 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN + IGB_TS_HDR_LEN) ||
3924 (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE + IGB_TS_HDR_LEN))
3925 rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE + IGB_TS_HDR_LEN;
3927 if ((adapter->hw.mac.type == e1000_82580) &&
3928 (rx_buffer_len == IGB_RXBUFFER_128))
3929 rx_buffer_len += IGB_RXBUFFER_64;
3931 if (netif_running(netdev))
3934 dev_info(&pdev->dev, "changing MTU from %d to %d\n",
3935 netdev->mtu, new_mtu);
3936 netdev->mtu = new_mtu;
3938 for (i = 0; i < adapter->num_rx_queues; i++)
3939 adapter->rx_ring[i]->rx_buffer_len = rx_buffer_len;
3941 if (netif_running(netdev))
3946 clear_bit(__IGB_RESETTING, &adapter->state);
3952 * igb_update_stats - Update the board statistics counters
3953 * @adapter: board private structure
3956 void igb_update_stats(struct igb_adapter *adapter)
3958 struct net_device_stats *net_stats = igb_get_stats(adapter->netdev);
3959 struct e1000_hw *hw = &adapter->hw;
3960 struct pci_dev *pdev = adapter->pdev;
3966 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3969 * Prevent stats update while adapter is being reset, or if the pci
3970 * connection is down.
3972 if (adapter->link_speed == 0)
3974 if (pci_channel_offline(pdev))
3979 for (i = 0; i < adapter->num_rx_queues; i++) {
3980 u32 rqdpc_tmp = rd32(E1000_RQDPC(i)) & 0x0FFF;
3981 struct igb_ring *ring = adapter->rx_ring[i];
3982 ring->rx_stats.drops += rqdpc_tmp;
3983 net_stats->rx_fifo_errors += rqdpc_tmp;
3984 bytes += ring->rx_stats.bytes;
3985 packets += ring->rx_stats.packets;
3988 net_stats->rx_bytes = bytes;
3989 net_stats->rx_packets = packets;
3993 for (i = 0; i < adapter->num_tx_queues; i++) {
3994 struct igb_ring *ring = adapter->tx_ring[i];
3995 bytes += ring->tx_stats.bytes;
3996 packets += ring->tx_stats.packets;
3998 net_stats->tx_bytes = bytes;
3999 net_stats->tx_packets = packets;
4001 /* read stats registers */
4002 adapter->stats.crcerrs += rd32(E1000_CRCERRS);
4003 adapter->stats.gprc += rd32(E1000_GPRC);
4004 adapter->stats.gorc += rd32(E1000_GORCL);
4005 rd32(E1000_GORCH); /* clear GORCL */
4006 adapter->stats.bprc += rd32(E1000_BPRC);
4007 adapter->stats.mprc += rd32(E1000_MPRC);
4008 adapter->stats.roc += rd32(E1000_ROC);
4010 adapter->stats.prc64 += rd32(E1000_PRC64);
4011 adapter->stats.prc127 += rd32(E1000_PRC127);
4012 adapter->stats.prc255 += rd32(E1000_PRC255);
4013 adapter->stats.prc511 += rd32(E1000_PRC511);
4014 adapter->stats.prc1023 += rd32(E1000_PRC1023);
4015 adapter->stats.prc1522 += rd32(E1000_PRC1522);
4016 adapter->stats.symerrs += rd32(E1000_SYMERRS);
4017 adapter->stats.sec += rd32(E1000_SEC);
4019 mpc = rd32(E1000_MPC);
4020 adapter->stats.mpc += mpc;
4021 net_stats->rx_fifo_errors += mpc;
4022 adapter->stats.scc += rd32(E1000_SCC);
4023 adapter->stats.ecol += rd32(E1000_ECOL);
4024 adapter->stats.mcc += rd32(E1000_MCC);
4025 adapter->stats.latecol += rd32(E1000_LATECOL);
4026 adapter->stats.dc += rd32(E1000_DC);
4027 adapter->stats.rlec += rd32(E1000_RLEC);
4028 adapter->stats.xonrxc += rd32(E1000_XONRXC);
4029 adapter->stats.xontxc += rd32(E1000_XONTXC);
4030 adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
4031 adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
4032 adapter->stats.fcruc += rd32(E1000_FCRUC);
4033 adapter->stats.gptc += rd32(E1000_GPTC);
4034 adapter->stats.gotc += rd32(E1000_GOTCL);
4035 rd32(E1000_GOTCH); /* clear GOTCL */
4036 adapter->stats.rnbc += rd32(E1000_RNBC);
4037 adapter->stats.ruc += rd32(E1000_RUC);
4038 adapter->stats.rfc += rd32(E1000_RFC);
4039 adapter->stats.rjc += rd32(E1000_RJC);
4040 adapter->stats.tor += rd32(E1000_TORH);
4041 adapter->stats.tot += rd32(E1000_TOTH);
4042 adapter->stats.tpr += rd32(E1000_TPR);
4044 adapter->stats.ptc64 += rd32(E1000_PTC64);
4045 adapter->stats.ptc127 += rd32(E1000_PTC127);
4046 adapter->stats.ptc255 += rd32(E1000_PTC255);
4047 adapter->stats.ptc511 += rd32(E1000_PTC511);
4048 adapter->stats.ptc1023 += rd32(E1000_PTC1023);
4049 adapter->stats.ptc1522 += rd32(E1000_PTC1522);
4051 adapter->stats.mptc += rd32(E1000_MPTC);
4052 adapter->stats.bptc += rd32(E1000_BPTC);
4054 adapter->stats.tpt += rd32(E1000_TPT);
4055 adapter->stats.colc += rd32(E1000_COLC);
4057 adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
4058 /* read internal phy specific stats */
4059 reg = rd32(E1000_CTRL_EXT);
4060 if (!(reg & E1000_CTRL_EXT_LINK_MODE_MASK)) {
4061 adapter->stats.rxerrc += rd32(E1000_RXERRC);
4062 adapter->stats.tncrs += rd32(E1000_TNCRS);
4065 adapter->stats.tsctc += rd32(E1000_TSCTC);
4066 adapter->stats.tsctfc += rd32(E1000_TSCTFC);
4068 adapter->stats.iac += rd32(E1000_IAC);
4069 adapter->stats.icrxoc += rd32(E1000_ICRXOC);
4070 adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
4071 adapter->stats.icrxatc += rd32(E1000_ICRXATC);
4072 adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
4073 adapter->stats.ictxatc += rd32(E1000_ICTXATC);
4074 adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
4075 adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
4076 adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
4078 /* Fill out the OS statistics structure */
4079 net_stats->multicast = adapter->stats.mprc;
4080 net_stats->collisions = adapter->stats.colc;
4084 /* RLEC on some newer hardware can be incorrect so build
4085 * our own version based on RUC and ROC */
4086 net_stats->rx_errors = adapter->stats.rxerrc +
4087 adapter->stats.crcerrs + adapter->stats.algnerrc +
4088 adapter->stats.ruc + adapter->stats.roc +
4089 adapter->stats.cexterr;
4090 net_stats->rx_length_errors = adapter->stats.ruc +
4092 net_stats->rx_crc_errors = adapter->stats.crcerrs;
4093 net_stats->rx_frame_errors = adapter->stats.algnerrc;
4094 net_stats->rx_missed_errors = adapter->stats.mpc;
4097 net_stats->tx_errors = adapter->stats.ecol +
4098 adapter->stats.latecol;
4099 net_stats->tx_aborted_errors = adapter->stats.ecol;
4100 net_stats->tx_window_errors = adapter->stats.latecol;
4101 net_stats->tx_carrier_errors = adapter->stats.tncrs;
4103 /* Tx Dropped needs to be maintained elsewhere */
4106 if (hw->phy.media_type == e1000_media_type_copper) {
4107 if ((adapter->link_speed == SPEED_1000) &&
4108 (!igb_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
4109 phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
4110 adapter->phy_stats.idle_errors += phy_tmp;
4114 /* Management Stats */
4115 adapter->stats.mgptc += rd32(E1000_MGTPTC);
4116 adapter->stats.mgprc += rd32(E1000_MGTPRC);
4117 adapter->stats.mgpdc += rd32(E1000_MGTPDC);
4120 static irqreturn_t igb_msix_other(int irq, void *data)
4122 struct igb_adapter *adapter = data;
4123 struct e1000_hw *hw = &adapter->hw;
4124 u32 icr = rd32(E1000_ICR);
4125 /* reading ICR causes bit 31 of EICR to be cleared */
4127 if (icr & E1000_ICR_DRSTA)
4128 schedule_work(&adapter->reset_task);
4130 if (icr & E1000_ICR_DOUTSYNC) {
4131 /* HW is reporting DMA is out of sync */
4132 adapter->stats.doosync++;
4135 /* Check for a mailbox event */
4136 if (icr & E1000_ICR_VMMB)
4137 igb_msg_task(adapter);
4139 if (icr & E1000_ICR_LSC) {
4140 hw->mac.get_link_status = 1;
4141 /* guard against interrupt when we're going down */
4142 if (!test_bit(__IGB_DOWN, &adapter->state))
4143 mod_timer(&adapter->watchdog_timer, jiffies + 1);
4146 if (adapter->vfs_allocated_count)
4147 wr32(E1000_IMS, E1000_IMS_LSC |
4149 E1000_IMS_DOUTSYNC);
4151 wr32(E1000_IMS, E1000_IMS_LSC | E1000_IMS_DOUTSYNC);
4152 wr32(E1000_EIMS, adapter->eims_other);
4157 static void igb_write_itr(struct igb_q_vector *q_vector)
4159 struct igb_adapter *adapter = q_vector->adapter;
4160 u32 itr_val = q_vector->itr_val & 0x7FFC;
4162 if (!q_vector->set_itr)
4168 if (adapter->hw.mac.type == e1000_82575)
4169 itr_val |= itr_val << 16;
4171 itr_val |= 0x8000000;
4173 writel(itr_val, q_vector->itr_register);
4174 q_vector->set_itr = 0;
4177 static irqreturn_t igb_msix_ring(int irq, void *data)
4179 struct igb_q_vector *q_vector = data;
4181 /* Write the ITR value calculated from the previous interrupt. */
4182 igb_write_itr(q_vector);
4184 napi_schedule(&q_vector->napi);
4189 #ifdef CONFIG_IGB_DCA
4190 static void igb_update_dca(struct igb_q_vector *q_vector)
4192 struct igb_adapter *adapter = q_vector->adapter;
4193 struct e1000_hw *hw = &adapter->hw;
4194 int cpu = get_cpu();
4196 if (q_vector->cpu == cpu)
4199 if (q_vector->tx_ring) {
4200 int q = q_vector->tx_ring->reg_idx;
4201 u32 dca_txctrl = rd32(E1000_DCA_TXCTRL(q));
4202 if (hw->mac.type == e1000_82575) {
4203 dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK;
4204 dca_txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
4206 dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK_82576;
4207 dca_txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu) <<
4208 E1000_DCA_TXCTRL_CPUID_SHIFT;
4210 dca_txctrl |= E1000_DCA_TXCTRL_DESC_DCA_EN;
4211 wr32(E1000_DCA_TXCTRL(q), dca_txctrl);
4213 if (q_vector->rx_ring) {
4214 int q = q_vector->rx_ring->reg_idx;
4215 u32 dca_rxctrl = rd32(E1000_DCA_RXCTRL(q));
4216 if (hw->mac.type == e1000_82575) {
4217 dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK;
4218 dca_rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
4220 dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK_82576;
4221 dca_rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu) <<
4222 E1000_DCA_RXCTRL_CPUID_SHIFT;
4224 dca_rxctrl |= E1000_DCA_RXCTRL_DESC_DCA_EN;
4225 dca_rxctrl |= E1000_DCA_RXCTRL_HEAD_DCA_EN;
4226 dca_rxctrl |= E1000_DCA_RXCTRL_DATA_DCA_EN;
4227 wr32(E1000_DCA_RXCTRL(q), dca_rxctrl);
4229 q_vector->cpu = cpu;
4234 static void igb_setup_dca(struct igb_adapter *adapter)
4236 struct e1000_hw *hw = &adapter->hw;
4239 if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
4242 /* Always use CB2 mode, difference is masked in the CB driver. */
4243 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
4245 for (i = 0; i < adapter->num_q_vectors; i++) {
4246 adapter->q_vector[i]->cpu = -1;
4247 igb_update_dca(adapter->q_vector[i]);
4251 static int __igb_notify_dca(struct device *dev, void *data)
4253 struct net_device *netdev = dev_get_drvdata(dev);
4254 struct igb_adapter *adapter = netdev_priv(netdev);
4255 struct pci_dev *pdev = adapter->pdev;
4256 struct e1000_hw *hw = &adapter->hw;
4257 unsigned long event = *(unsigned long *)data;
4260 case DCA_PROVIDER_ADD:
4261 /* if already enabled, don't do it again */
4262 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
4264 if (dca_add_requester(dev) == 0) {
4265 adapter->flags |= IGB_FLAG_DCA_ENABLED;
4266 dev_info(&pdev->dev, "DCA enabled\n");
4267 igb_setup_dca(adapter);
4270 /* Fall Through since DCA is disabled. */
4271 case DCA_PROVIDER_REMOVE:
4272 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
4273 /* without this a class_device is left
4274 * hanging around in the sysfs model */
4275 dca_remove_requester(dev);
4276 dev_info(&pdev->dev, "DCA disabled\n");
4277 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
4278 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
4286 static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
4291 ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
4294 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
4296 #endif /* CONFIG_IGB_DCA */
4298 static void igb_ping_all_vfs(struct igb_adapter *adapter)
4300 struct e1000_hw *hw = &adapter->hw;
4304 for (i = 0 ; i < adapter->vfs_allocated_count; i++) {
4305 ping = E1000_PF_CONTROL_MSG;
4306 if (adapter->vf_data[i].flags & IGB_VF_FLAG_CTS)
4307 ping |= E1000_VT_MSGTYPE_CTS;
4308 igb_write_mbx(hw, &ping, 1, i);
4312 static int igb_set_vf_promisc(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
4314 struct e1000_hw *hw = &adapter->hw;
4315 u32 vmolr = rd32(E1000_VMOLR(vf));
4316 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
4318 vf_data->flags |= ~(IGB_VF_FLAG_UNI_PROMISC |
4319 IGB_VF_FLAG_MULTI_PROMISC);
4320 vmolr &= ~(E1000_VMOLR_ROPE | E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
4322 if (*msgbuf & E1000_VF_SET_PROMISC_MULTICAST) {
4323 vmolr |= E1000_VMOLR_MPME;
4324 *msgbuf &= ~E1000_VF_SET_PROMISC_MULTICAST;
4327 * if we have hashes and we are clearing a multicast promisc
4328 * flag we need to write the hashes to the MTA as this step
4329 * was previously skipped
4331 if (vf_data->num_vf_mc_hashes > 30) {
4332 vmolr |= E1000_VMOLR_MPME;
4333 } else if (vf_data->num_vf_mc_hashes) {
4335 vmolr |= E1000_VMOLR_ROMPE;
4336 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
4337 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
4341 wr32(E1000_VMOLR(vf), vmolr);
4343 /* there are flags left unprocessed, likely not supported */
4344 if (*msgbuf & E1000_VT_MSGINFO_MASK)
4351 static int igb_set_vf_multicasts(struct igb_adapter *adapter,
4352 u32 *msgbuf, u32 vf)
4354 int n = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
4355 u16 *hash_list = (u16 *)&msgbuf[1];
4356 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
4359 /* salt away the number of multicast addresses assigned
4360 * to this VF for later use to restore when the PF multi cast
4363 vf_data->num_vf_mc_hashes = n;
4365 /* only up to 30 hash values supported */
4369 /* store the hashes for later use */
4370 for (i = 0; i < n; i++)
4371 vf_data->vf_mc_hashes[i] = hash_list[i];
4373 /* Flush and reset the mta with the new values */
4374 igb_set_rx_mode(adapter->netdev);
4379 static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
4381 struct e1000_hw *hw = &adapter->hw;
4382 struct vf_data_storage *vf_data;
4385 for (i = 0; i < adapter->vfs_allocated_count; i++) {
4386 u32 vmolr = rd32(E1000_VMOLR(i));
4387 vmolr &= ~(E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
4389 vf_data = &adapter->vf_data[i];
4391 if ((vf_data->num_vf_mc_hashes > 30) ||
4392 (vf_data->flags & IGB_VF_FLAG_MULTI_PROMISC)) {
4393 vmolr |= E1000_VMOLR_MPME;
4394 } else if (vf_data->num_vf_mc_hashes) {
4395 vmolr |= E1000_VMOLR_ROMPE;
4396 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
4397 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
4399 wr32(E1000_VMOLR(i), vmolr);
4403 static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
4405 struct e1000_hw *hw = &adapter->hw;
4406 u32 pool_mask, reg, vid;
4409 pool_mask = 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
4411 /* Find the vlan filter for this id */
4412 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
4413 reg = rd32(E1000_VLVF(i));
4415 /* remove the vf from the pool */
4418 /* if pool is empty then remove entry from vfta */
4419 if (!(reg & E1000_VLVF_POOLSEL_MASK) &&
4420 (reg & E1000_VLVF_VLANID_ENABLE)) {
4422 vid = reg & E1000_VLVF_VLANID_MASK;
4423 igb_vfta_set(hw, vid, false);
4426 wr32(E1000_VLVF(i), reg);
4429 adapter->vf_data[vf].vlans_enabled = 0;
4432 static s32 igb_vlvf_set(struct igb_adapter *adapter, u32 vid, bool add, u32 vf)
4434 struct e1000_hw *hw = &adapter->hw;
4437 /* The vlvf table only exists on 82576 hardware and newer */
4438 if (hw->mac.type < e1000_82576)
4441 /* we only need to do this if VMDq is enabled */
4442 if (!adapter->vfs_allocated_count)
4445 /* Find the vlan filter for this id */
4446 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
4447 reg = rd32(E1000_VLVF(i));
4448 if ((reg & E1000_VLVF_VLANID_ENABLE) &&
4449 vid == (reg & E1000_VLVF_VLANID_MASK))
4454 if (i == E1000_VLVF_ARRAY_SIZE) {
4455 /* Did not find a matching VLAN ID entry that was
4456 * enabled. Search for a free filter entry, i.e.
4457 * one without the enable bit set
4459 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
4460 reg = rd32(E1000_VLVF(i));
4461 if (!(reg & E1000_VLVF_VLANID_ENABLE))
4465 if (i < E1000_VLVF_ARRAY_SIZE) {
4466 /* Found an enabled/available entry */
4467 reg |= 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
4469 /* if !enabled we need to set this up in vfta */
4470 if (!(reg & E1000_VLVF_VLANID_ENABLE)) {
4471 /* add VID to filter table */
4472 igb_vfta_set(hw, vid, true);
4473 reg |= E1000_VLVF_VLANID_ENABLE;
4475 reg &= ~E1000_VLVF_VLANID_MASK;
4477 wr32(E1000_VLVF(i), reg);
4479 /* do not modify RLPML for PF devices */
4480 if (vf >= adapter->vfs_allocated_count)
4483 if (!adapter->vf_data[vf].vlans_enabled) {
4485 reg = rd32(E1000_VMOLR(vf));
4486 size = reg & E1000_VMOLR_RLPML_MASK;
4488 reg &= ~E1000_VMOLR_RLPML_MASK;
4490 wr32(E1000_VMOLR(vf), reg);
4493 adapter->vf_data[vf].vlans_enabled++;
4497 if (i < E1000_VLVF_ARRAY_SIZE) {
4498 /* remove vf from the pool */
4499 reg &= ~(1 << (E1000_VLVF_POOLSEL_SHIFT + vf));
4500 /* if pool is empty then remove entry from vfta */
4501 if (!(reg & E1000_VLVF_POOLSEL_MASK)) {
4503 igb_vfta_set(hw, vid, false);
4505 wr32(E1000_VLVF(i), reg);
4507 /* do not modify RLPML for PF devices */
4508 if (vf >= adapter->vfs_allocated_count)
4511 adapter->vf_data[vf].vlans_enabled--;
4512 if (!adapter->vf_data[vf].vlans_enabled) {
4514 reg = rd32(E1000_VMOLR(vf));
4515 size = reg & E1000_VMOLR_RLPML_MASK;
4517 reg &= ~E1000_VMOLR_RLPML_MASK;
4519 wr32(E1000_VMOLR(vf), reg);
4526 static void igb_set_vmvir(struct igb_adapter *adapter, u32 vid, u32 vf)
4528 struct e1000_hw *hw = &adapter->hw;
4531 wr32(E1000_VMVIR(vf), (vid | E1000_VMVIR_VLANA_DEFAULT));
4533 wr32(E1000_VMVIR(vf), 0);
4536 static int igb_ndo_set_vf_vlan(struct net_device *netdev,
4537 int vf, u16 vlan, u8 qos)
4540 struct igb_adapter *adapter = netdev_priv(netdev);
4542 if ((vf >= adapter->vfs_allocated_count) || (vlan > 4095) || (qos > 7))
4545 err = igb_vlvf_set(adapter, vlan, !!vlan, vf);
4548 igb_set_vmvir(adapter, vlan | (qos << VLAN_PRIO_SHIFT), vf);
4549 igb_set_vmolr(adapter, vf, !vlan);
4550 adapter->vf_data[vf].pf_vlan = vlan;
4551 adapter->vf_data[vf].pf_qos = qos;
4552 dev_info(&adapter->pdev->dev,
4553 "Setting VLAN %d, QOS 0x%x on VF %d\n", vlan, qos, vf);
4554 if (test_bit(__IGB_DOWN, &adapter->state)) {
4555 dev_warn(&adapter->pdev->dev,
4556 "The VF VLAN has been set,"
4557 " but the PF device is not up.\n");
4558 dev_warn(&adapter->pdev->dev,
4559 "Bring the PF device up before"
4560 " attempting to use the VF device.\n");
4563 igb_vlvf_set(adapter, adapter->vf_data[vf].pf_vlan,
4565 igb_set_vmvir(adapter, vlan, vf);
4566 igb_set_vmolr(adapter, vf, true);
4567 adapter->vf_data[vf].pf_vlan = 0;
4568 adapter->vf_data[vf].pf_qos = 0;
4574 static int igb_set_vf_vlan(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
4576 int add = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
4577 int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
4579 return igb_vlvf_set(adapter, vid, add, vf);
4582 static inline void igb_vf_reset(struct igb_adapter *adapter, u32 vf)
4585 adapter->vf_data[vf].flags &= ~(IGB_VF_FLAG_PF_SET_MAC);
4586 adapter->vf_data[vf].last_nack = jiffies;
4588 /* reset offloads to defaults */
4589 igb_set_vmolr(adapter, vf, true);
4591 /* reset vlans for device */
4592 igb_clear_vf_vfta(adapter, vf);
4593 if (adapter->vf_data[vf].pf_vlan)
4594 igb_ndo_set_vf_vlan(adapter->netdev, vf,
4595 adapter->vf_data[vf].pf_vlan,
4596 adapter->vf_data[vf].pf_qos);
4598 igb_clear_vf_vfta(adapter, vf);
4600 /* reset multicast table array for vf */
4601 adapter->vf_data[vf].num_vf_mc_hashes = 0;
4603 /* Flush and reset the mta with the new values */
4604 igb_set_rx_mode(adapter->netdev);
4607 static void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
4609 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
4611 /* generate a new mac address as we were hotplug removed/added */
4612 if (!(adapter->vf_data[vf].flags & IGB_VF_FLAG_PF_SET_MAC))
4613 random_ether_addr(vf_mac);
4615 /* process remaining reset events */
4616 igb_vf_reset(adapter, vf);
4619 static void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
4621 struct e1000_hw *hw = &adapter->hw;
4622 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
4623 int rar_entry = hw->mac.rar_entry_count - (vf + 1);
4625 u8 *addr = (u8 *)(&msgbuf[1]);
4627 /* process all the same items cleared in a function level reset */
4628 igb_vf_reset(adapter, vf);
4630 /* set vf mac address */
4631 igb_rar_set_qsel(adapter, vf_mac, rar_entry, vf);
4633 /* enable transmit and receive for vf */
4634 reg = rd32(E1000_VFTE);
4635 wr32(E1000_VFTE, reg | (1 << vf));
4636 reg = rd32(E1000_VFRE);
4637 wr32(E1000_VFRE, reg | (1 << vf));
4639 adapter->vf_data[vf].flags = IGB_VF_FLAG_CTS;
4641 /* reply to reset with ack and vf mac address */
4642 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
4643 memcpy(addr, vf_mac, 6);
4644 igb_write_mbx(hw, msgbuf, 3, vf);
4647 static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf)
4649 unsigned char *addr = (char *)&msg[1];
4652 if (is_valid_ether_addr(addr))
4653 err = igb_set_vf_mac(adapter, vf, addr);
4658 static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf)
4660 struct e1000_hw *hw = &adapter->hw;
4661 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
4662 u32 msg = E1000_VT_MSGTYPE_NACK;
4664 /* if device isn't clear to send it shouldn't be reading either */
4665 if (!(vf_data->flags & IGB_VF_FLAG_CTS) &&
4666 time_after(jiffies, vf_data->last_nack + (2 * HZ))) {
4667 igb_write_mbx(hw, &msg, 1, vf);
4668 vf_data->last_nack = jiffies;
4672 static void igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
4674 struct pci_dev *pdev = adapter->pdev;
4675 u32 msgbuf[E1000_VFMAILBOX_SIZE];
4676 struct e1000_hw *hw = &adapter->hw;
4677 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
4680 retval = igb_read_mbx(hw, msgbuf, E1000_VFMAILBOX_SIZE, vf);
4683 /* if receive failed revoke VF CTS stats and restart init */
4684 dev_err(&pdev->dev, "Error receiving message from VF\n");
4685 vf_data->flags &= ~IGB_VF_FLAG_CTS;
4686 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
4691 /* this is a message we already processed, do nothing */
4692 if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
4696 * until the vf completes a reset it should not be
4697 * allowed to start any configuration.
4700 if (msgbuf[0] == E1000_VF_RESET) {
4701 igb_vf_reset_msg(adapter, vf);
4705 if (!(vf_data->flags & IGB_VF_FLAG_CTS)) {
4706 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
4712 switch ((msgbuf[0] & 0xFFFF)) {
4713 case E1000_VF_SET_MAC_ADDR:
4714 retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
4716 case E1000_VF_SET_PROMISC:
4717 retval = igb_set_vf_promisc(adapter, msgbuf, vf);
4719 case E1000_VF_SET_MULTICAST:
4720 retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
4722 case E1000_VF_SET_LPE:
4723 retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf);
4725 case E1000_VF_SET_VLAN:
4726 if (adapter->vf_data[vf].pf_vlan)
4729 retval = igb_set_vf_vlan(adapter, msgbuf, vf);
4732 dev_err(&pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
4737 msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
4739 /* notify the VF of the results of what it sent us */
4741 msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
4743 msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
4745 igb_write_mbx(hw, msgbuf, 1, vf);
4748 static void igb_msg_task(struct igb_adapter *adapter)
4750 struct e1000_hw *hw = &adapter->hw;
4753 for (vf = 0; vf < adapter->vfs_allocated_count; vf++) {
4754 /* process any reset requests */
4755 if (!igb_check_for_rst(hw, vf))
4756 igb_vf_reset_event(adapter, vf);
4758 /* process any messages pending */
4759 if (!igb_check_for_msg(hw, vf))
4760 igb_rcv_msg_from_vf(adapter, vf);
4762 /* process any acks */
4763 if (!igb_check_for_ack(hw, vf))
4764 igb_rcv_ack_from_vf(adapter, vf);
4769 * igb_set_uta - Set unicast filter table address
4770 * @adapter: board private structure
4772 * The unicast table address is a register array of 32-bit registers.
4773 * The table is meant to be used in a way similar to how the MTA is used
4774 * however due to certain limitations in the hardware it is necessary to
4775 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscous
4776 * enable bit to allow vlan tag stripping when promiscous mode is enabled
4778 static void igb_set_uta(struct igb_adapter *adapter)
4780 struct e1000_hw *hw = &adapter->hw;
4783 /* The UTA table only exists on 82576 hardware and newer */
4784 if (hw->mac.type < e1000_82576)
4787 /* we only need to do this if VMDq is enabled */
4788 if (!adapter->vfs_allocated_count)
4791 for (i = 0; i < hw->mac.uta_reg_count; i++)
4792 array_wr32(E1000_UTA, i, ~0);
4796 * igb_intr_msi - Interrupt Handler
4797 * @irq: interrupt number
4798 * @data: pointer to a network interface device structure
4800 static irqreturn_t igb_intr_msi(int irq, void *data)
4802 struct igb_adapter *adapter = data;
4803 struct igb_q_vector *q_vector = adapter->q_vector[0];
4804 struct e1000_hw *hw = &adapter->hw;
4805 /* read ICR disables interrupts using IAM */
4806 u32 icr = rd32(E1000_ICR);
4808 igb_write_itr(q_vector);
4810 if (icr & E1000_ICR_DRSTA)
4811 schedule_work(&adapter->reset_task);
4813 if (icr & E1000_ICR_DOUTSYNC) {
4814 /* HW is reporting DMA is out of sync */
4815 adapter->stats.doosync++;
4818 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
4819 hw->mac.get_link_status = 1;
4820 if (!test_bit(__IGB_DOWN, &adapter->state))
4821 mod_timer(&adapter->watchdog_timer, jiffies + 1);
4824 napi_schedule(&q_vector->napi);
4830 * igb_intr - Legacy Interrupt Handler
4831 * @irq: interrupt number
4832 * @data: pointer to a network interface device structure
4834 static irqreturn_t igb_intr(int irq, void *data)
4836 struct igb_adapter *adapter = data;
4837 struct igb_q_vector *q_vector = adapter->q_vector[0];
4838 struct e1000_hw *hw = &adapter->hw;
4839 /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked. No
4840 * need for the IMC write */
4841 u32 icr = rd32(E1000_ICR);
4843 return IRQ_NONE; /* Not our interrupt */
4845 igb_write_itr(q_vector);
4847 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
4848 * not set, then the adapter didn't send an interrupt */
4849 if (!(icr & E1000_ICR_INT_ASSERTED))
4852 if (icr & E1000_ICR_DRSTA)
4853 schedule_work(&adapter->reset_task);
4855 if (icr & E1000_ICR_DOUTSYNC) {
4856 /* HW is reporting DMA is out of sync */
4857 adapter->stats.doosync++;
4860 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
4861 hw->mac.get_link_status = 1;
4862 /* guard against interrupt when we're going down */
4863 if (!test_bit(__IGB_DOWN, &adapter->state))
4864 mod_timer(&adapter->watchdog_timer, jiffies + 1);
4867 napi_schedule(&q_vector->napi);
4872 static inline void igb_ring_irq_enable(struct igb_q_vector *q_vector)
4874 struct igb_adapter *adapter = q_vector->adapter;
4875 struct e1000_hw *hw = &adapter->hw;
4877 if ((q_vector->rx_ring && (adapter->rx_itr_setting & 3)) ||
4878 (!q_vector->rx_ring && (adapter->tx_itr_setting & 3))) {
4879 if (!adapter->msix_entries)
4880 igb_set_itr(adapter);
4882 igb_update_ring_itr(q_vector);
4885 if (!test_bit(__IGB_DOWN, &adapter->state)) {
4886 if (adapter->msix_entries)
4887 wr32(E1000_EIMS, q_vector->eims_value);
4889 igb_irq_enable(adapter);
4894 * igb_poll - NAPI Rx polling callback
4895 * @napi: napi polling structure
4896 * @budget: count of how many packets we should handle
4898 static int igb_poll(struct napi_struct *napi, int budget)
4900 struct igb_q_vector *q_vector = container_of(napi,
4901 struct igb_q_vector,
4903 int tx_clean_complete = 1, work_done = 0;
4905 #ifdef CONFIG_IGB_DCA
4906 if (q_vector->adapter->flags & IGB_FLAG_DCA_ENABLED)
4907 igb_update_dca(q_vector);
4909 if (q_vector->tx_ring)
4910 tx_clean_complete = igb_clean_tx_irq(q_vector);
4912 if (q_vector->rx_ring)
4913 igb_clean_rx_irq_adv(q_vector, &work_done, budget);
4915 if (!tx_clean_complete)
4918 /* If not enough Rx work done, exit the polling mode */
4919 if (work_done < budget) {
4920 napi_complete(napi);
4921 igb_ring_irq_enable(q_vector);
4928 * igb_systim_to_hwtstamp - convert system time value to hw timestamp
4929 * @adapter: board private structure
4930 * @shhwtstamps: timestamp structure to update
4931 * @regval: unsigned 64bit system time value.
4933 * We need to convert the system time value stored in the RX/TXSTMP registers
4934 * into a hwtstamp which can be used by the upper level timestamping functions
4936 static void igb_systim_to_hwtstamp(struct igb_adapter *adapter,
4937 struct skb_shared_hwtstamps *shhwtstamps,
4943 * The 82580 starts with 1ns at bit 0 in RX/TXSTMPL, shift this up to
4944 * 24 to match clock shift we setup earlier.
4946 if (adapter->hw.mac.type == e1000_82580)
4947 regval <<= IGB_82580_TSYNC_SHIFT;
4949 ns = timecounter_cyc2time(&adapter->clock, regval);
4950 timecompare_update(&adapter->compare, ns);
4951 memset(shhwtstamps, 0, sizeof(struct skb_shared_hwtstamps));
4952 shhwtstamps->hwtstamp = ns_to_ktime(ns);
4953 shhwtstamps->syststamp = timecompare_transform(&adapter->compare, ns);
4957 * igb_tx_hwtstamp - utility function which checks for TX time stamp
4958 * @q_vector: pointer to q_vector containing needed info
4959 * @skb: packet that was just sent
4961 * If we were asked to do hardware stamping and such a time stamp is
4962 * available, then it must have been for this skb here because we only
4963 * allow only one such packet into the queue.
4965 static void igb_tx_hwtstamp(struct igb_q_vector *q_vector, struct sk_buff *skb)
4967 struct igb_adapter *adapter = q_vector->adapter;
4968 union skb_shared_tx *shtx = skb_tx(skb);
4969 struct e1000_hw *hw = &adapter->hw;
4970 struct skb_shared_hwtstamps shhwtstamps;
4973 /* if skb does not support hw timestamp or TX stamp not valid exit */
4974 if (likely(!shtx->hardware) ||
4975 !(rd32(E1000_TSYNCTXCTL) & E1000_TSYNCTXCTL_VALID))
4978 regval = rd32(E1000_TXSTMPL);
4979 regval |= (u64)rd32(E1000_TXSTMPH) << 32;
4981 igb_systim_to_hwtstamp(adapter, &shhwtstamps, regval);
4982 skb_tstamp_tx(skb, &shhwtstamps);
4986 * igb_clean_tx_irq - Reclaim resources after transmit completes
4987 * @q_vector: pointer to q_vector containing needed info
4988 * returns true if ring is completely cleaned
4990 static bool igb_clean_tx_irq(struct igb_q_vector *q_vector)
4992 struct igb_adapter *adapter = q_vector->adapter;
4993 struct igb_ring *tx_ring = q_vector->tx_ring;
4994 struct net_device *netdev = tx_ring->netdev;
4995 struct e1000_hw *hw = &adapter->hw;
4996 struct igb_buffer *buffer_info;
4997 struct sk_buff *skb;
4998 union e1000_adv_tx_desc *tx_desc, *eop_desc;
4999 unsigned int total_bytes = 0, total_packets = 0;
5000 unsigned int i, eop, count = 0;
5001 bool cleaned = false;
5003 i = tx_ring->next_to_clean;
5004 eop = tx_ring->buffer_info[i].next_to_watch;
5005 eop_desc = E1000_TX_DESC_ADV(*tx_ring, eop);
5007 while ((eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)) &&
5008 (count < tx_ring->count)) {
5009 for (cleaned = false; !cleaned; count++) {
5010 tx_desc = E1000_TX_DESC_ADV(*tx_ring, i);
5011 buffer_info = &tx_ring->buffer_info[i];
5012 cleaned = (i == eop);
5013 skb = buffer_info->skb;
5016 unsigned int segs, bytecount;
5017 /* gso_segs is currently only valid for tcp */
5018 segs = buffer_info->gso_segs;
5019 /* multiply data chunks by size of headers */
5020 bytecount = ((segs - 1) * skb_headlen(skb)) +
5022 total_packets += segs;
5023 total_bytes += bytecount;
5025 igb_tx_hwtstamp(q_vector, skb);
5028 igb_unmap_and_free_tx_resource(tx_ring, buffer_info);
5029 tx_desc->wb.status = 0;
5032 if (i == tx_ring->count)
5035 eop = tx_ring->buffer_info[i].next_to_watch;
5036 eop_desc = E1000_TX_DESC_ADV(*tx_ring, eop);
5039 tx_ring->next_to_clean = i;
5041 if (unlikely(count &&
5042 netif_carrier_ok(netdev) &&
5043 igb_desc_unused(tx_ring) >= IGB_TX_QUEUE_WAKE)) {
5044 /* Make sure that anybody stopping the queue after this
5045 * sees the new next_to_clean.
5048 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
5049 !(test_bit(__IGB_DOWN, &adapter->state))) {
5050 netif_wake_subqueue(netdev, tx_ring->queue_index);
5051 tx_ring->tx_stats.restart_queue++;
5055 if (tx_ring->detect_tx_hung) {
5056 /* Detect a transmit hang in hardware, this serializes the
5057 * check with the clearing of time_stamp and movement of i */
5058 tx_ring->detect_tx_hung = false;
5059 if (tx_ring->buffer_info[i].time_stamp &&
5060 time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
5061 (adapter->tx_timeout_factor * HZ)) &&
5062 !(rd32(E1000_STATUS) & E1000_STATUS_TXOFF)) {
5064 /* detected Tx unit hang */
5065 dev_err(tx_ring->dev,
5066 "Detected Tx Unit Hang\n"
5070 " next_to_use <%x>\n"
5071 " next_to_clean <%x>\n"
5072 "buffer_info[next_to_clean]\n"
5073 " time_stamp <%lx>\n"
5074 " next_to_watch <%x>\n"
5076 " desc.status <%x>\n",
5077 tx_ring->queue_index,
5078 readl(tx_ring->head),
5079 readl(tx_ring->tail),
5080 tx_ring->next_to_use,
5081 tx_ring->next_to_clean,
5082 tx_ring->buffer_info[eop].time_stamp,
5085 eop_desc->wb.status);
5086 netif_stop_subqueue(netdev, tx_ring->queue_index);
5089 tx_ring->total_bytes += total_bytes;
5090 tx_ring->total_packets += total_packets;
5091 tx_ring->tx_stats.bytes += total_bytes;
5092 tx_ring->tx_stats.packets += total_packets;
5093 return (count < tx_ring->count);
5097 * igb_receive_skb - helper function to handle rx indications
5098 * @q_vector: structure containing interrupt and ring information
5099 * @skb: packet to send up
5100 * @vlan_tag: vlan tag for packet
5102 static void igb_receive_skb(struct igb_q_vector *q_vector,
5103 struct sk_buff *skb,
5106 struct igb_adapter *adapter = q_vector->adapter;
5108 if (vlan_tag && adapter->vlgrp)
5109 vlan_gro_receive(&q_vector->napi, adapter->vlgrp,
5112 napi_gro_receive(&q_vector->napi, skb);
5115 static inline void igb_rx_checksum_adv(struct igb_ring *ring,
5116 u32 status_err, struct sk_buff *skb)
5118 skb->ip_summed = CHECKSUM_NONE;
5120 /* Ignore Checksum bit is set or checksum is disabled through ethtool */
5121 if (!(ring->flags & IGB_RING_FLAG_RX_CSUM) ||
5122 (status_err & E1000_RXD_STAT_IXSM))
5125 /* TCP/UDP checksum error bit is set */
5127 (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
5129 * work around errata with sctp packets where the TCPE aka
5130 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
5131 * packets, (aka let the stack check the crc32c)
5133 if ((skb->len == 60) &&
5134 (ring->flags & IGB_RING_FLAG_RX_SCTP_CSUM))
5135 ring->rx_stats.csum_err++;
5137 /* let the stack verify checksum errors */
5140 /* It must be a TCP or UDP packet with a valid checksum */
5141 if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
5142 skb->ip_summed = CHECKSUM_UNNECESSARY;
5144 dev_dbg(ring->dev, "cksum success: bits %08X\n", status_err);
5147 static void igb_rx_hwtstamp(struct igb_q_vector *q_vector, u32 staterr,
5148 struct sk_buff *skb)
5150 struct igb_adapter *adapter = q_vector->adapter;
5151 struct e1000_hw *hw = &adapter->hw;
5155 * If this bit is set, then the RX registers contain the time stamp. No
5156 * other packet will be time stamped until we read these registers, so
5157 * read the registers to make them available again. Because only one
5158 * packet can be time stamped at a time, we know that the register
5159 * values must belong to this one here and therefore we don't need to
5160 * compare any of the additional attributes stored for it.
5162 * If nothing went wrong, then it should have a skb_shared_tx that we
5163 * can turn into a skb_shared_hwtstamps.
5165 if (staterr & E1000_RXDADV_STAT_TSIP) {
5166 u32 *stamp = (u32 *)skb->data;
5167 regval = le32_to_cpu(*(stamp + 2));
5168 regval |= (u64)le32_to_cpu(*(stamp + 3)) << 32;
5169 skb_pull(skb, IGB_TS_HDR_LEN);
5171 if(!(rd32(E1000_TSYNCRXCTL) & E1000_TSYNCRXCTL_VALID))
5174 regval = rd32(E1000_RXSTMPL);
5175 regval |= (u64)rd32(E1000_RXSTMPH) << 32;
5178 igb_systim_to_hwtstamp(adapter, skb_hwtstamps(skb), regval);
5180 static inline u16 igb_get_hlen(struct igb_ring *rx_ring,
5181 union e1000_adv_rx_desc *rx_desc)
5183 /* HW will not DMA in data larger than the given buffer, even if it
5184 * parses the (NFS, of course) header to be larger. In that case, it
5185 * fills the header buffer and spills the rest into the page.
5187 u16 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hdr_info) &
5188 E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
5189 if (hlen > rx_ring->rx_buffer_len)
5190 hlen = rx_ring->rx_buffer_len;
5194 static bool igb_clean_rx_irq_adv(struct igb_q_vector *q_vector,
5195 int *work_done, int budget)
5197 struct igb_ring *rx_ring = q_vector->rx_ring;
5198 struct net_device *netdev = rx_ring->netdev;
5199 struct device *dev = rx_ring->dev;
5200 union e1000_adv_rx_desc *rx_desc , *next_rxd;
5201 struct igb_buffer *buffer_info , *next_buffer;
5202 struct sk_buff *skb;
5203 bool cleaned = false;
5204 int cleaned_count = 0;
5205 int current_node = numa_node_id();
5206 unsigned int total_bytes = 0, total_packets = 0;
5212 i = rx_ring->next_to_clean;
5213 buffer_info = &rx_ring->buffer_info[i];
5214 rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
5215 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
5217 while (staterr & E1000_RXD_STAT_DD) {
5218 if (*work_done >= budget)
5222 skb = buffer_info->skb;
5223 prefetch(skb->data - NET_IP_ALIGN);
5224 buffer_info->skb = NULL;
5227 if (i == rx_ring->count)
5230 next_rxd = E1000_RX_DESC_ADV(*rx_ring, i);
5232 next_buffer = &rx_ring->buffer_info[i];
5234 length = le16_to_cpu(rx_desc->wb.upper.length);
5238 if (buffer_info->dma) {
5239 dma_unmap_single(dev, buffer_info->dma,
5240 rx_ring->rx_buffer_len,
5242 buffer_info->dma = 0;
5243 if (rx_ring->rx_buffer_len >= IGB_RXBUFFER_1024) {
5244 skb_put(skb, length);
5247 skb_put(skb, igb_get_hlen(rx_ring, rx_desc));
5251 dma_unmap_page(dev, buffer_info->page_dma,
5252 PAGE_SIZE / 2, DMA_FROM_DEVICE);
5253 buffer_info->page_dma = 0;
5255 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
5257 buffer_info->page_offset,
5260 if ((page_count(buffer_info->page) != 1) ||
5261 (page_to_nid(buffer_info->page) != current_node))
5262 buffer_info->page = NULL;
5264 get_page(buffer_info->page);
5267 skb->data_len += length;
5268 skb->truesize += length;
5271 if (!(staterr & E1000_RXD_STAT_EOP)) {
5272 buffer_info->skb = next_buffer->skb;
5273 buffer_info->dma = next_buffer->dma;
5274 next_buffer->skb = skb;
5275 next_buffer->dma = 0;
5279 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
5280 dev_kfree_skb_irq(skb);
5284 if (staterr & (E1000_RXDADV_STAT_TSIP | E1000_RXDADV_STAT_TS))
5285 igb_rx_hwtstamp(q_vector, staterr, skb);
5286 total_bytes += skb->len;
5289 igb_rx_checksum_adv(rx_ring, staterr, skb);
5291 skb->protocol = eth_type_trans(skb, netdev);
5292 skb_record_rx_queue(skb, rx_ring->queue_index);
5294 vlan_tag = ((staterr & E1000_RXD_STAT_VP) ?
5295 le16_to_cpu(rx_desc->wb.upper.vlan) : 0);
5297 igb_receive_skb(q_vector, skb, vlan_tag);
5300 rx_desc->wb.upper.status_error = 0;
5302 /* return some buffers to hardware, one at a time is too slow */
5303 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
5304 igb_alloc_rx_buffers_adv(rx_ring, cleaned_count);
5308 /* use prefetched values */
5310 buffer_info = next_buffer;
5311 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
5314 rx_ring->next_to_clean = i;
5315 cleaned_count = igb_desc_unused(rx_ring);
5318 igb_alloc_rx_buffers_adv(rx_ring, cleaned_count);
5320 rx_ring->total_packets += total_packets;
5321 rx_ring->total_bytes += total_bytes;
5322 rx_ring->rx_stats.packets += total_packets;
5323 rx_ring->rx_stats.bytes += total_bytes;
5328 * igb_alloc_rx_buffers_adv - Replace used receive buffers; packet split
5329 * @adapter: address of board private structure
5331 void igb_alloc_rx_buffers_adv(struct igb_ring *rx_ring, int cleaned_count)
5333 struct net_device *netdev = rx_ring->netdev;
5334 union e1000_adv_rx_desc *rx_desc;
5335 struct igb_buffer *buffer_info;
5336 struct sk_buff *skb;
5340 i = rx_ring->next_to_use;
5341 buffer_info = &rx_ring->buffer_info[i];
5343 bufsz = rx_ring->rx_buffer_len;
5345 while (cleaned_count--) {
5346 rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
5348 if ((bufsz < IGB_RXBUFFER_1024) && !buffer_info->page_dma) {
5349 if (!buffer_info->page) {
5350 buffer_info->page = netdev_alloc_page(netdev);
5351 if (!buffer_info->page) {
5352 rx_ring->rx_stats.alloc_failed++;
5355 buffer_info->page_offset = 0;
5357 buffer_info->page_offset ^= PAGE_SIZE / 2;
5359 buffer_info->page_dma =
5360 dma_map_page(rx_ring->dev, buffer_info->page,
5361 buffer_info->page_offset,
5364 if (dma_mapping_error(rx_ring->dev,
5365 buffer_info->page_dma)) {
5366 buffer_info->page_dma = 0;
5367 rx_ring->rx_stats.alloc_failed++;
5372 skb = buffer_info->skb;
5374 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
5376 rx_ring->rx_stats.alloc_failed++;
5380 buffer_info->skb = skb;
5382 if (!buffer_info->dma) {
5383 buffer_info->dma = dma_map_single(rx_ring->dev,
5387 if (dma_mapping_error(rx_ring->dev,
5388 buffer_info->dma)) {
5389 buffer_info->dma = 0;
5390 rx_ring->rx_stats.alloc_failed++;
5394 /* Refresh the desc even if buffer_addrs didn't change because
5395 * each write-back erases this info. */
5396 if (bufsz < IGB_RXBUFFER_1024) {
5397 rx_desc->read.pkt_addr =
5398 cpu_to_le64(buffer_info->page_dma);
5399 rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
5401 rx_desc->read.pkt_addr = cpu_to_le64(buffer_info->dma);
5402 rx_desc->read.hdr_addr = 0;
5406 if (i == rx_ring->count)
5408 buffer_info = &rx_ring->buffer_info[i];
5412 if (rx_ring->next_to_use != i) {
5413 rx_ring->next_to_use = i;
5415 i = (rx_ring->count - 1);
5419 /* Force memory writes to complete before letting h/w
5420 * know there are new descriptors to fetch. (Only
5421 * applicable for weak-ordered memory model archs,
5422 * such as IA-64). */
5424 writel(i, rx_ring->tail);
5434 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5436 struct igb_adapter *adapter = netdev_priv(netdev);
5437 struct mii_ioctl_data *data = if_mii(ifr);
5439 if (adapter->hw.phy.media_type != e1000_media_type_copper)
5444 data->phy_id = adapter->hw.phy.addr;
5447 if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
5459 * igb_hwtstamp_ioctl - control hardware time stamping
5464 * Outgoing time stamping can be enabled and disabled. Play nice and
5465 * disable it when requested, although it shouldn't case any overhead
5466 * when no packet needs it. At most one packet in the queue may be
5467 * marked for time stamping, otherwise it would be impossible to tell
5468 * for sure to which packet the hardware time stamp belongs.
5470 * Incoming time stamping has to be configured via the hardware
5471 * filters. Not all combinations are supported, in particular event
5472 * type has to be specified. Matching the kind of event packet is
5473 * not supported, with the exception of "all V2 events regardless of
5477 static int igb_hwtstamp_ioctl(struct net_device *netdev,
5478 struct ifreq *ifr, int cmd)
5480 struct igb_adapter *adapter = netdev_priv(netdev);
5481 struct e1000_hw *hw = &adapter->hw;
5482 struct hwtstamp_config config;
5483 u32 tsync_tx_ctl = E1000_TSYNCTXCTL_ENABLED;
5484 u32 tsync_rx_ctl = E1000_TSYNCRXCTL_ENABLED;
5485 u32 tsync_rx_cfg = 0;
5490 if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
5493 /* reserved for future extensions */
5497 switch (config.tx_type) {
5498 case HWTSTAMP_TX_OFF:
5500 case HWTSTAMP_TX_ON:
5506 switch (config.rx_filter) {
5507 case HWTSTAMP_FILTER_NONE:
5510 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
5511 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
5512 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
5513 case HWTSTAMP_FILTER_ALL:
5515 * register TSYNCRXCFG must be set, therefore it is not
5516 * possible to time stamp both Sync and Delay_Req messages
5517 * => fall back to time stamping all packets
5519 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_ALL;
5520 config.rx_filter = HWTSTAMP_FILTER_ALL;
5522 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
5523 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L4_V1;
5524 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V1_SYNC_MESSAGE;
5527 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
5528 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L4_V1;
5529 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V1_DELAY_REQ_MESSAGE;
5532 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
5533 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
5534 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
5535 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V2_SYNC_MESSAGE;
5538 config.rx_filter = HWTSTAMP_FILTER_SOME;
5540 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
5541 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
5542 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
5543 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V2_DELAY_REQ_MESSAGE;
5546 config.rx_filter = HWTSTAMP_FILTER_SOME;
5548 case HWTSTAMP_FILTER_PTP_V2_EVENT:
5549 case HWTSTAMP_FILTER_PTP_V2_SYNC:
5550 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
5551 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_EVENT_V2;
5552 config.rx_filter = HWTSTAMP_FILTER_PTP_V2_EVENT;
5559 if (hw->mac.type == e1000_82575) {
5560 if (tsync_rx_ctl | tsync_tx_ctl)
5566 * Per-packet timestamping only works if all packets are
5567 * timestamped, so enable timestamping in all packets as
5568 * long as one rx filter was configured.
5570 if ((hw->mac.type == e1000_82580) && tsync_rx_ctl) {
5571 tsync_rx_ctl = E1000_TSYNCRXCTL_ENABLED;
5572 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_ALL;
5575 /* enable/disable TX */
5576 regval = rd32(E1000_TSYNCTXCTL);
5577 regval &= ~E1000_TSYNCTXCTL_ENABLED;
5578 regval |= tsync_tx_ctl;
5579 wr32(E1000_TSYNCTXCTL, regval);
5581 /* enable/disable RX */
5582 regval = rd32(E1000_TSYNCRXCTL);
5583 regval &= ~(E1000_TSYNCRXCTL_ENABLED | E1000_TSYNCRXCTL_TYPE_MASK);
5584 regval |= tsync_rx_ctl;
5585 wr32(E1000_TSYNCRXCTL, regval);
5587 /* define which PTP packets are time stamped */
5588 wr32(E1000_TSYNCRXCFG, tsync_rx_cfg);
5590 /* define ethertype filter for timestamped packets */
5593 (E1000_ETQF_FILTER_ENABLE | /* enable filter */
5594 E1000_ETQF_1588 | /* enable timestamping */
5595 ETH_P_1588)); /* 1588 eth protocol type */
5597 wr32(E1000_ETQF(3), 0);
5599 #define PTP_PORT 319
5600 /* L4 Queue Filter[3]: filter by destination port and protocol */
5602 u32 ftqf = (IPPROTO_UDP /* UDP */
5603 | E1000_FTQF_VF_BP /* VF not compared */
5604 | E1000_FTQF_1588_TIME_STAMP /* Enable Timestamping */
5605 | E1000_FTQF_MASK); /* mask all inputs */
5606 ftqf &= ~E1000_FTQF_MASK_PROTO_BP; /* enable protocol check */
5608 wr32(E1000_IMIR(3), htons(PTP_PORT));
5609 wr32(E1000_IMIREXT(3),
5610 (E1000_IMIREXT_SIZE_BP | E1000_IMIREXT_CTRL_BP));
5611 if (hw->mac.type == e1000_82576) {
5612 /* enable source port check */
5613 wr32(E1000_SPQF(3), htons(PTP_PORT));
5614 ftqf &= ~E1000_FTQF_MASK_SOURCE_PORT_BP;
5616 wr32(E1000_FTQF(3), ftqf);
5618 wr32(E1000_FTQF(3), E1000_FTQF_MASK);
5622 adapter->hwtstamp_config = config;
5624 /* clear TX/RX time stamp registers, just to be sure */
5625 regval = rd32(E1000_TXSTMPH);
5626 regval = rd32(E1000_RXSTMPH);
5628 return copy_to_user(ifr->ifr_data, &config, sizeof(config)) ?
5638 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5644 return igb_mii_ioctl(netdev, ifr, cmd);
5646 return igb_hwtstamp_ioctl(netdev, ifr, cmd);
5652 s32 igb_read_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
5654 struct igb_adapter *adapter = hw->back;
5657 cap_offset = pci_find_capability(adapter->pdev, PCI_CAP_ID_EXP);
5659 return -E1000_ERR_CONFIG;
5661 pci_read_config_word(adapter->pdev, cap_offset + reg, value);
5666 s32 igb_write_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
5668 struct igb_adapter *adapter = hw->back;
5671 cap_offset = pci_find_capability(adapter->pdev, PCI_CAP_ID_EXP);
5673 return -E1000_ERR_CONFIG;
5675 pci_write_config_word(adapter->pdev, cap_offset + reg, *value);
5680 static void igb_vlan_rx_register(struct net_device *netdev,
5681 struct vlan_group *grp)
5683 struct igb_adapter *adapter = netdev_priv(netdev);
5684 struct e1000_hw *hw = &adapter->hw;
5687 igb_irq_disable(adapter);
5688 adapter->vlgrp = grp;
5691 /* enable VLAN tag insert/strip */
5692 ctrl = rd32(E1000_CTRL);
5693 ctrl |= E1000_CTRL_VME;
5694 wr32(E1000_CTRL, ctrl);
5696 /* Disable CFI check */
5697 rctl = rd32(E1000_RCTL);
5698 rctl &= ~E1000_RCTL_CFIEN;
5699 wr32(E1000_RCTL, rctl);
5701 /* disable VLAN tag insert/strip */
5702 ctrl = rd32(E1000_CTRL);
5703 ctrl &= ~E1000_CTRL_VME;
5704 wr32(E1000_CTRL, ctrl);
5707 igb_rlpml_set(adapter);
5709 if (!test_bit(__IGB_DOWN, &adapter->state))
5710 igb_irq_enable(adapter);
5713 static void igb_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
5715 struct igb_adapter *adapter = netdev_priv(netdev);
5716 struct e1000_hw *hw = &adapter->hw;
5717 int pf_id = adapter->vfs_allocated_count;
5719 /* attempt to add filter to vlvf array */
5720 igb_vlvf_set(adapter, vid, true, pf_id);
5722 /* add the filter since PF can receive vlans w/o entry in vlvf */
5723 igb_vfta_set(hw, vid, true);
5726 static void igb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
5728 struct igb_adapter *adapter = netdev_priv(netdev);
5729 struct e1000_hw *hw = &adapter->hw;
5730 int pf_id = adapter->vfs_allocated_count;
5733 igb_irq_disable(adapter);
5734 vlan_group_set_device(adapter->vlgrp, vid, NULL);
5736 if (!test_bit(__IGB_DOWN, &adapter->state))
5737 igb_irq_enable(adapter);
5739 /* remove vlan from VLVF table array */
5740 err = igb_vlvf_set(adapter, vid, false, pf_id);
5742 /* if vid was not present in VLVF just remove it from table */
5744 igb_vfta_set(hw, vid, false);
5747 static void igb_restore_vlan(struct igb_adapter *adapter)
5749 igb_vlan_rx_register(adapter->netdev, adapter->vlgrp);
5751 if (adapter->vlgrp) {
5753 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
5754 if (!vlan_group_get_device(adapter->vlgrp, vid))
5756 igb_vlan_rx_add_vid(adapter->netdev, vid);
5761 int igb_set_spd_dplx(struct igb_adapter *adapter, u16 spddplx)
5763 struct pci_dev *pdev = adapter->pdev;
5764 struct e1000_mac_info *mac = &adapter->hw.mac;
5769 case SPEED_10 + DUPLEX_HALF:
5770 mac->forced_speed_duplex = ADVERTISE_10_HALF;
5772 case SPEED_10 + DUPLEX_FULL:
5773 mac->forced_speed_duplex = ADVERTISE_10_FULL;
5775 case SPEED_100 + DUPLEX_HALF:
5776 mac->forced_speed_duplex = ADVERTISE_100_HALF;
5778 case SPEED_100 + DUPLEX_FULL:
5779 mac->forced_speed_duplex = ADVERTISE_100_FULL;
5781 case SPEED_1000 + DUPLEX_FULL:
5783 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
5785 case SPEED_1000 + DUPLEX_HALF: /* not supported */
5787 dev_err(&pdev->dev, "Unsupported Speed/Duplex configuration\n");
5793 static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake)
5795 struct net_device *netdev = pci_get_drvdata(pdev);
5796 struct igb_adapter *adapter = netdev_priv(netdev);
5797 struct e1000_hw *hw = &adapter->hw;
5798 u32 ctrl, rctl, status;
5799 u32 wufc = adapter->wol;
5804 netif_device_detach(netdev);
5806 if (netif_running(netdev))
5809 igb_clear_interrupt_scheme(adapter);
5812 retval = pci_save_state(pdev);
5817 status = rd32(E1000_STATUS);
5818 if (status & E1000_STATUS_LU)
5819 wufc &= ~E1000_WUFC_LNKC;
5822 igb_setup_rctl(adapter);
5823 igb_set_rx_mode(netdev);
5825 /* turn on all-multi mode if wake on multicast is enabled */
5826 if (wufc & E1000_WUFC_MC) {
5827 rctl = rd32(E1000_RCTL);
5828 rctl |= E1000_RCTL_MPE;
5829 wr32(E1000_RCTL, rctl);
5832 ctrl = rd32(E1000_CTRL);
5833 /* advertise wake from D3Cold */
5834 #define E1000_CTRL_ADVD3WUC 0x00100000
5835 /* phy power management enable */
5836 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5837 ctrl |= E1000_CTRL_ADVD3WUC;
5838 wr32(E1000_CTRL, ctrl);
5840 /* Allow time for pending master requests to run */
5841 igb_disable_pcie_master(hw);
5843 wr32(E1000_WUC, E1000_WUC_PME_EN);
5844 wr32(E1000_WUFC, wufc);
5847 wr32(E1000_WUFC, 0);
5850 *enable_wake = wufc || adapter->en_mng_pt;
5852 igb_power_down_link(adapter);
5854 igb_power_up_link(adapter);
5856 /* Release control of h/w to f/w. If f/w is AMT enabled, this
5857 * would have already happened in close and is redundant. */
5858 igb_release_hw_control(adapter);
5860 pci_disable_device(pdev);
5866 static int igb_suspend(struct pci_dev *pdev, pm_message_t state)
5871 retval = __igb_shutdown(pdev, &wake);
5876 pci_prepare_to_sleep(pdev);
5878 pci_wake_from_d3(pdev, false);
5879 pci_set_power_state(pdev, PCI_D3hot);
5885 static int igb_resume(struct pci_dev *pdev)
5887 struct net_device *netdev = pci_get_drvdata(pdev);
5888 struct igb_adapter *adapter = netdev_priv(netdev);
5889 struct e1000_hw *hw = &adapter->hw;
5892 pci_set_power_state(pdev, PCI_D0);
5893 pci_restore_state(pdev);
5894 pci_save_state(pdev);
5896 err = pci_enable_device_mem(pdev);
5899 "igb: Cannot enable PCI device from suspend\n");
5902 pci_set_master(pdev);
5904 pci_enable_wake(pdev, PCI_D3hot, 0);
5905 pci_enable_wake(pdev, PCI_D3cold, 0);
5907 if (igb_init_interrupt_scheme(adapter)) {
5908 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
5914 /* let the f/w know that the h/w is now under the control of the
5916 igb_get_hw_control(adapter);
5918 wr32(E1000_WUS, ~0);
5920 if (netif_running(netdev)) {
5921 err = igb_open(netdev);
5926 netif_device_attach(netdev);
5932 static void igb_shutdown(struct pci_dev *pdev)
5936 __igb_shutdown(pdev, &wake);
5938 if (system_state == SYSTEM_POWER_OFF) {
5939 pci_wake_from_d3(pdev, wake);
5940 pci_set_power_state(pdev, PCI_D3hot);
5944 #ifdef CONFIG_NET_POLL_CONTROLLER
5946 * Polling 'interrupt' - used by things like netconsole to send skbs
5947 * without having to re-enable interrupts. It's not called while
5948 * the interrupt routine is executing.
5950 static void igb_netpoll(struct net_device *netdev)
5952 struct igb_adapter *adapter = netdev_priv(netdev);
5953 struct e1000_hw *hw = &adapter->hw;
5956 if (!adapter->msix_entries) {
5957 struct igb_q_vector *q_vector = adapter->q_vector[0];
5958 igb_irq_disable(adapter);
5959 napi_schedule(&q_vector->napi);
5963 for (i = 0; i < adapter->num_q_vectors; i++) {
5964 struct igb_q_vector *q_vector = adapter->q_vector[i];
5965 wr32(E1000_EIMC, q_vector->eims_value);
5966 napi_schedule(&q_vector->napi);
5969 #endif /* CONFIG_NET_POLL_CONTROLLER */
5972 * igb_io_error_detected - called when PCI error is detected
5973 * @pdev: Pointer to PCI device
5974 * @state: The current pci connection state
5976 * This function is called after a PCI bus error affecting
5977 * this device has been detected.
5979 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
5980 pci_channel_state_t state)
5982 struct net_device *netdev = pci_get_drvdata(pdev);
5983 struct igb_adapter *adapter = netdev_priv(netdev);
5985 netif_device_detach(netdev);
5987 if (state == pci_channel_io_perm_failure)
5988 return PCI_ERS_RESULT_DISCONNECT;
5990 if (netif_running(netdev))
5992 pci_disable_device(pdev);
5994 /* Request a slot slot reset. */
5995 return PCI_ERS_RESULT_NEED_RESET;
5999 * igb_io_slot_reset - called after the pci bus has been reset.
6000 * @pdev: Pointer to PCI device
6002 * Restart the card from scratch, as if from a cold-boot. Implementation
6003 * resembles the first-half of the igb_resume routine.
6005 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
6007 struct net_device *netdev = pci_get_drvdata(pdev);
6008 struct igb_adapter *adapter = netdev_priv(netdev);
6009 struct e1000_hw *hw = &adapter->hw;
6010 pci_ers_result_t result;
6013 if (pci_enable_device_mem(pdev)) {
6015 "Cannot re-enable PCI device after reset.\n");
6016 result = PCI_ERS_RESULT_DISCONNECT;
6018 pci_set_master(pdev);
6019 pci_restore_state(pdev);
6020 pci_save_state(pdev);
6022 pci_enable_wake(pdev, PCI_D3hot, 0);
6023 pci_enable_wake(pdev, PCI_D3cold, 0);
6026 wr32(E1000_WUS, ~0);
6027 result = PCI_ERS_RESULT_RECOVERED;
6030 err = pci_cleanup_aer_uncorrect_error_status(pdev);
6032 dev_err(&pdev->dev, "pci_cleanup_aer_uncorrect_error_status "
6033 "failed 0x%0x\n", err);
6034 /* non-fatal, continue */
6041 * igb_io_resume - called when traffic can start flowing again.
6042 * @pdev: Pointer to PCI device
6044 * This callback is called when the error recovery driver tells us that
6045 * its OK to resume normal operation. Implementation resembles the
6046 * second-half of the igb_resume routine.
6048 static void igb_io_resume(struct pci_dev *pdev)
6050 struct net_device *netdev = pci_get_drvdata(pdev);
6051 struct igb_adapter *adapter = netdev_priv(netdev);
6053 if (netif_running(netdev)) {
6054 if (igb_up(adapter)) {
6055 dev_err(&pdev->dev, "igb_up failed after reset\n");
6060 netif_device_attach(netdev);
6062 /* let the f/w know that the h/w is now under the control of the
6064 igb_get_hw_control(adapter);
6067 static void igb_rar_set_qsel(struct igb_adapter *adapter, u8 *addr, u32 index,
6070 u32 rar_low, rar_high;
6071 struct e1000_hw *hw = &adapter->hw;
6073 /* HW expects these in little endian so we reverse the byte order
6074 * from network order (big endian) to little endian
6076 rar_low = ((u32) addr[0] | ((u32) addr[1] << 8) |
6077 ((u32) addr[2] << 16) | ((u32) addr[3] << 24));
6078 rar_high = ((u32) addr[4] | ((u32) addr[5] << 8));
6080 /* Indicate to hardware the Address is Valid. */
6081 rar_high |= E1000_RAH_AV;
6083 if (hw->mac.type == e1000_82575)
6084 rar_high |= E1000_RAH_POOL_1 * qsel;
6086 rar_high |= E1000_RAH_POOL_1 << qsel;
6088 wr32(E1000_RAL(index), rar_low);
6090 wr32(E1000_RAH(index), rar_high);
6094 static int igb_set_vf_mac(struct igb_adapter *adapter,
6095 int vf, unsigned char *mac_addr)
6097 struct e1000_hw *hw = &adapter->hw;
6098 /* VF MAC addresses start at end of receive addresses and moves
6099 * torwards the first, as a result a collision should not be possible */
6100 int rar_entry = hw->mac.rar_entry_count - (vf + 1);
6102 memcpy(adapter->vf_data[vf].vf_mac_addresses, mac_addr, ETH_ALEN);
6104 igb_rar_set_qsel(adapter, mac_addr, rar_entry, vf);
6109 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
6111 struct igb_adapter *adapter = netdev_priv(netdev);
6112 if (!is_valid_ether_addr(mac) || (vf >= adapter->vfs_allocated_count))
6114 adapter->vf_data[vf].flags |= IGB_VF_FLAG_PF_SET_MAC;
6115 dev_info(&adapter->pdev->dev, "setting MAC %pM on VF %d\n", mac, vf);
6116 dev_info(&adapter->pdev->dev, "Reload the VF driver to make this"
6117 " change effective.");
6118 if (test_bit(__IGB_DOWN, &adapter->state)) {
6119 dev_warn(&adapter->pdev->dev, "The VF MAC address has been set,"
6120 " but the PF device is not up.\n");
6121 dev_warn(&adapter->pdev->dev, "Bring the PF device up before"
6122 " attempting to use the VF device.\n");
6124 return igb_set_vf_mac(adapter, vf, mac);
6127 static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf, int tx_rate)
6132 static int igb_ndo_get_vf_config(struct net_device *netdev,
6133 int vf, struct ifla_vf_info *ivi)
6135 struct igb_adapter *adapter = netdev_priv(netdev);
6136 if (vf >= adapter->vfs_allocated_count)
6139 memcpy(&ivi->mac, adapter->vf_data[vf].vf_mac_addresses, ETH_ALEN);
6141 ivi->vlan = adapter->vf_data[vf].pf_vlan;
6142 ivi->qos = adapter->vf_data[vf].pf_qos;
6146 static void igb_vmm_control(struct igb_adapter *adapter)
6148 struct e1000_hw *hw = &adapter->hw;
6151 switch (hw->mac.type) {
6154 /* replication is not supported for 82575 */
6157 /* notify HW that the MAC is adding vlan tags */
6158 reg = rd32(E1000_DTXCTL);
6159 reg |= E1000_DTXCTL_VLAN_ADDED;
6160 wr32(E1000_DTXCTL, reg);
6162 /* enable replication vlan tag stripping */
6163 reg = rd32(E1000_RPLOLR);
6164 reg |= E1000_RPLOLR_STRVLAN;
6165 wr32(E1000_RPLOLR, reg);
6167 /* none of the above registers are supported by i350 */
6171 if (adapter->vfs_allocated_count) {
6172 igb_vmdq_set_loopback_pf(hw, true);
6173 igb_vmdq_set_replication_pf(hw, true);
6175 igb_vmdq_set_loopback_pf(hw, false);
6176 igb_vmdq_set_replication_pf(hw, false);