1 /*******************************************************************************
3 Intel(R) Gigabit Ethernet Linux driver
4 Copyright(c) 2007-2009 Intel Corporation.
6 This program is free software; you can redistribute it and/or modify it
7 under the terms and conditions of the GNU General Public License,
8 version 2, as published by the Free Software Foundation.
10 This program is distributed in the hope it will be useful, but WITHOUT
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
23 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26 *******************************************************************************/
28 #include <linux/module.h>
29 #include <linux/types.h>
30 #include <linux/init.h>
31 #include <linux/vmalloc.h>
32 #include <linux/pagemap.h>
33 #include <linux/netdevice.h>
34 #include <linux/ipv6.h>
35 #include <net/checksum.h>
36 #include <net/ip6_checksum.h>
37 #include <linux/net_tstamp.h>
38 #include <linux/mii.h>
39 #include <linux/ethtool.h>
40 #include <linux/if_vlan.h>
41 #include <linux/pci.h>
42 #include <linux/pci-aspm.h>
43 #include <linux/delay.h>
44 #include <linux/interrupt.h>
45 #include <linux/if_ether.h>
46 #include <linux/aer.h>
48 #include <linux/dca.h>
52 #define DRV_VERSION "2.1.0-k2"
53 char igb_driver_name[] = "igb";
54 char igb_driver_version[] = DRV_VERSION;
55 static const char igb_driver_string[] =
56 "Intel(R) Gigabit Ethernet Network Driver";
57 static const char igb_copyright[] = "Copyright (c) 2007-2009 Intel Corporation.";
59 static const struct e1000_info *igb_info_tbl[] = {
60 [board_82575] = &e1000_82575_info,
63 static DEFINE_PCI_DEVICE_TABLE(igb_pci_tbl) = {
64 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER), board_82575 },
65 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_FIBER), board_82575 },
66 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SERDES), board_82575 },
67 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SGMII), board_82575 },
68 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER_DUAL), board_82575 },
69 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576), board_82575 },
70 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS), board_82575 },
71 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS_SERDES), board_82575 },
72 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_FIBER), board_82575 },
73 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES), board_82575 },
74 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES_QUAD), board_82575 },
75 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER), board_82575 },
76 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
77 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
78 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
79 /* required last entry */
83 MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
85 void igb_reset(struct igb_adapter *);
86 static int igb_setup_all_tx_resources(struct igb_adapter *);
87 static int igb_setup_all_rx_resources(struct igb_adapter *);
88 static void igb_free_all_tx_resources(struct igb_adapter *);
89 static void igb_free_all_rx_resources(struct igb_adapter *);
90 static void igb_setup_mrqc(struct igb_adapter *);
91 void igb_update_stats(struct igb_adapter *);
92 static int igb_probe(struct pci_dev *, const struct pci_device_id *);
93 static void __devexit igb_remove(struct pci_dev *pdev);
94 static int igb_sw_init(struct igb_adapter *);
95 static int igb_open(struct net_device *);
96 static int igb_close(struct net_device *);
97 static void igb_configure_tx(struct igb_adapter *);
98 static void igb_configure_rx(struct igb_adapter *);
99 static void igb_clean_all_tx_rings(struct igb_adapter *);
100 static void igb_clean_all_rx_rings(struct igb_adapter *);
101 static void igb_clean_tx_ring(struct igb_ring *);
102 static void igb_clean_rx_ring(struct igb_ring *);
103 static void igb_set_rx_mode(struct net_device *);
104 static void igb_update_phy_info(unsigned long);
105 static void igb_watchdog(unsigned long);
106 static void igb_watchdog_task(struct work_struct *);
107 static netdev_tx_t igb_xmit_frame_adv(struct sk_buff *skb, struct net_device *);
108 static struct net_device_stats *igb_get_stats(struct net_device *);
109 static int igb_change_mtu(struct net_device *, int);
110 static int igb_set_mac(struct net_device *, void *);
111 static void igb_set_uta(struct igb_adapter *adapter);
112 static irqreturn_t igb_intr(int irq, void *);
113 static irqreturn_t igb_intr_msi(int irq, void *);
114 static irqreturn_t igb_msix_other(int irq, void *);
115 static irqreturn_t igb_msix_ring(int irq, void *);
116 #ifdef CONFIG_IGB_DCA
117 static void igb_update_dca(struct igb_q_vector *);
118 static void igb_setup_dca(struct igb_adapter *);
119 #endif /* CONFIG_IGB_DCA */
120 static bool igb_clean_tx_irq(struct igb_q_vector *);
121 static int igb_poll(struct napi_struct *, int);
122 static bool igb_clean_rx_irq_adv(struct igb_q_vector *, int *, int);
123 static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
124 static void igb_tx_timeout(struct net_device *);
125 static void igb_reset_task(struct work_struct *);
126 static void igb_vlan_rx_register(struct net_device *, struct vlan_group *);
127 static void igb_vlan_rx_add_vid(struct net_device *, u16);
128 static void igb_vlan_rx_kill_vid(struct net_device *, u16);
129 static void igb_restore_vlan(struct igb_adapter *);
130 static void igb_rar_set_qsel(struct igb_adapter *, u8 *, u32 , u8);
131 static void igb_ping_all_vfs(struct igb_adapter *);
132 static void igb_msg_task(struct igb_adapter *);
133 static void igb_vmm_control(struct igb_adapter *);
134 static int igb_set_vf_mac(struct igb_adapter *, int, unsigned char *);
135 static void igb_restore_vf_multicasts(struct igb_adapter *adapter);
136 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac);
137 static int igb_ndo_set_vf_vlan(struct net_device *netdev,
138 int vf, u16 vlan, u8 qos);
139 static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf, int tx_rate);
140 static int igb_ndo_get_vf_config(struct net_device *netdev, int vf,
141 struct ifla_vf_info *ivi);
144 static int igb_suspend(struct pci_dev *, pm_message_t);
145 static int igb_resume(struct pci_dev *);
147 static void igb_shutdown(struct pci_dev *);
148 #ifdef CONFIG_IGB_DCA
149 static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
150 static struct notifier_block dca_notifier = {
151 .notifier_call = igb_notify_dca,
156 #ifdef CONFIG_NET_POLL_CONTROLLER
157 /* for netdump / net console */
158 static void igb_netpoll(struct net_device *);
160 #ifdef CONFIG_PCI_IOV
161 static unsigned int max_vfs = 0;
162 module_param(max_vfs, uint, 0);
163 MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate "
164 "per physical function");
165 #endif /* CONFIG_PCI_IOV */
167 static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
168 pci_channel_state_t);
169 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
170 static void igb_io_resume(struct pci_dev *);
172 static struct pci_error_handlers igb_err_handler = {
173 .error_detected = igb_io_error_detected,
174 .slot_reset = igb_io_slot_reset,
175 .resume = igb_io_resume,
179 static struct pci_driver igb_driver = {
180 .name = igb_driver_name,
181 .id_table = igb_pci_tbl,
183 .remove = __devexit_p(igb_remove),
185 /* Power Managment Hooks */
186 .suspend = igb_suspend,
187 .resume = igb_resume,
189 .shutdown = igb_shutdown,
190 .err_handler = &igb_err_handler
193 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
194 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
195 MODULE_LICENSE("GPL");
196 MODULE_VERSION(DRV_VERSION);
199 * igb_read_clock - read raw cycle counter (to be used by time counter)
201 static cycle_t igb_read_clock(const struct cyclecounter *tc)
203 struct igb_adapter *adapter =
204 container_of(tc, struct igb_adapter, cycles);
205 struct e1000_hw *hw = &adapter->hw;
210 * The timestamp latches on lowest register read. For the 82580
211 * the lowest register is SYSTIMR instead of SYSTIML. However we never
212 * adjusted TIMINCA so SYSTIMR will just read as all 0s so ignore it.
214 if (hw->mac.type == e1000_82580) {
215 stamp = rd32(E1000_SYSTIMR) >> 8;
216 shift = IGB_82580_TSYNC_SHIFT;
219 stamp |= (u64)rd32(E1000_SYSTIML) << shift;
220 stamp |= (u64)rd32(E1000_SYSTIMH) << (shift + 32);
226 * igb_get_hw_dev_name - return device name string
227 * used by hardware layer to print debugging information
229 char *igb_get_hw_dev_name(struct e1000_hw *hw)
231 struct igb_adapter *adapter = hw->back;
232 return adapter->netdev->name;
236 * igb_get_time_str - format current NIC and system time as string
238 static char *igb_get_time_str(struct igb_adapter *adapter,
241 cycle_t hw = adapter->cycles.read(&adapter->cycles);
242 struct timespec nic = ns_to_timespec(timecounter_read(&adapter->clock));
244 struct timespec delta;
245 getnstimeofday(&sys);
247 delta = timespec_sub(nic, sys);
250 "HW %llu, NIC %ld.%09lus, SYS %ld.%09lus, NIC-SYS %lds + %09luns",
252 (long)nic.tv_sec, nic.tv_nsec,
253 (long)sys.tv_sec, sys.tv_nsec,
254 (long)delta.tv_sec, delta.tv_nsec);
261 * igb_init_module - Driver Registration Routine
263 * igb_init_module is the first routine called when the driver is
264 * loaded. All it does is register with the PCI subsystem.
266 static int __init igb_init_module(void)
269 printk(KERN_INFO "%s - version %s\n",
270 igb_driver_string, igb_driver_version);
272 printk(KERN_INFO "%s\n", igb_copyright);
274 #ifdef CONFIG_IGB_DCA
275 dca_register_notify(&dca_notifier);
277 ret = pci_register_driver(&igb_driver);
281 module_init(igb_init_module);
284 * igb_exit_module - Driver Exit Cleanup Routine
286 * igb_exit_module is called just before the driver is removed
289 static void __exit igb_exit_module(void)
291 #ifdef CONFIG_IGB_DCA
292 dca_unregister_notify(&dca_notifier);
294 pci_unregister_driver(&igb_driver);
297 module_exit(igb_exit_module);
299 #define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
301 * igb_cache_ring_register - Descriptor ring to register mapping
302 * @adapter: board private structure to initialize
304 * Once we know the feature-set enabled for the device, we'll cache
305 * the register offset the descriptor ring is assigned to.
307 static void igb_cache_ring_register(struct igb_adapter *adapter)
310 u32 rbase_offset = adapter->vfs_allocated_count;
312 switch (adapter->hw.mac.type) {
314 /* The queues are allocated for virtualization such that VF 0
315 * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
316 * In order to avoid collision we start at the first free queue
317 * and continue consuming queues in the same sequence
319 if (adapter->vfs_allocated_count) {
320 for (; i < adapter->rss_queues; i++)
321 adapter->rx_ring[i]->reg_idx = rbase_offset +
323 for (; j < adapter->rss_queues; j++)
324 adapter->tx_ring[j]->reg_idx = rbase_offset +
330 for (; i < adapter->num_rx_queues; i++)
331 adapter->rx_ring[i]->reg_idx = rbase_offset + i;
332 for (; j < adapter->num_tx_queues; j++)
333 adapter->tx_ring[j]->reg_idx = rbase_offset + j;
338 static void igb_free_queues(struct igb_adapter *adapter)
342 for (i = 0; i < adapter->num_tx_queues; i++) {
343 kfree(adapter->tx_ring[i]);
344 adapter->tx_ring[i] = NULL;
346 for (i = 0; i < adapter->num_rx_queues; i++) {
347 kfree(adapter->rx_ring[i]);
348 adapter->rx_ring[i] = NULL;
350 adapter->num_rx_queues = 0;
351 adapter->num_tx_queues = 0;
355 * igb_alloc_queues - Allocate memory for all rings
356 * @adapter: board private structure to initialize
358 * We allocate one ring per queue at run-time since we don't know the
359 * number of queues at compile-time.
361 static int igb_alloc_queues(struct igb_adapter *adapter)
363 struct igb_ring *ring;
366 for (i = 0; i < adapter->num_tx_queues; i++) {
367 ring = kzalloc(sizeof(struct igb_ring), GFP_KERNEL);
370 ring->count = adapter->tx_ring_count;
371 ring->queue_index = i;
372 ring->pdev = adapter->pdev;
373 ring->netdev = adapter->netdev;
374 /* For 82575, context index must be unique per ring. */
375 if (adapter->hw.mac.type == e1000_82575)
376 ring->flags = IGB_RING_FLAG_TX_CTX_IDX;
377 adapter->tx_ring[i] = ring;
380 for (i = 0; i < adapter->num_rx_queues; i++) {
381 ring = kzalloc(sizeof(struct igb_ring), GFP_KERNEL);
384 ring->count = adapter->rx_ring_count;
385 ring->queue_index = i;
386 ring->pdev = adapter->pdev;
387 ring->netdev = adapter->netdev;
388 ring->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
389 ring->flags = IGB_RING_FLAG_RX_CSUM; /* enable rx checksum */
390 /* set flag indicating ring supports SCTP checksum offload */
391 if (adapter->hw.mac.type >= e1000_82576)
392 ring->flags |= IGB_RING_FLAG_RX_SCTP_CSUM;
393 adapter->rx_ring[i] = ring;
396 igb_cache_ring_register(adapter);
401 igb_free_queues(adapter);
406 #define IGB_N0_QUEUE -1
407 static void igb_assign_vector(struct igb_q_vector *q_vector, int msix_vector)
410 struct igb_adapter *adapter = q_vector->adapter;
411 struct e1000_hw *hw = &adapter->hw;
413 int rx_queue = IGB_N0_QUEUE;
414 int tx_queue = IGB_N0_QUEUE;
416 if (q_vector->rx_ring)
417 rx_queue = q_vector->rx_ring->reg_idx;
418 if (q_vector->tx_ring)
419 tx_queue = q_vector->tx_ring->reg_idx;
421 switch (hw->mac.type) {
423 /* The 82575 assigns vectors using a bitmask, which matches the
424 bitmask for the EICR/EIMS/EIMC registers. To assign one
425 or more queues to a vector, we write the appropriate bits
426 into the MSIXBM register for that vector. */
427 if (rx_queue > IGB_N0_QUEUE)
428 msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
429 if (tx_queue > IGB_N0_QUEUE)
430 msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
431 if (!adapter->msix_entries && msix_vector == 0)
432 msixbm |= E1000_EIMS_OTHER;
433 array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
434 q_vector->eims_value = msixbm;
437 /* 82576 uses a table-based method for assigning vectors.
438 Each queue has a single entry in the table to which we write
439 a vector number along with a "valid" bit. Sadly, the layout
440 of the table is somewhat counterintuitive. */
441 if (rx_queue > IGB_N0_QUEUE) {
442 index = (rx_queue & 0x7);
443 ivar = array_rd32(E1000_IVAR0, index);
445 /* vector goes into low byte of register */
446 ivar = ivar & 0xFFFFFF00;
447 ivar |= msix_vector | E1000_IVAR_VALID;
449 /* vector goes into third byte of register */
450 ivar = ivar & 0xFF00FFFF;
451 ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
453 array_wr32(E1000_IVAR0, index, ivar);
455 if (tx_queue > IGB_N0_QUEUE) {
456 index = (tx_queue & 0x7);
457 ivar = array_rd32(E1000_IVAR0, index);
459 /* vector goes into second byte of register */
460 ivar = ivar & 0xFFFF00FF;
461 ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
463 /* vector goes into high byte of register */
464 ivar = ivar & 0x00FFFFFF;
465 ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
467 array_wr32(E1000_IVAR0, index, ivar);
469 q_vector->eims_value = 1 << msix_vector;
472 /* 82580 uses the same table-based approach as 82576 but has fewer
473 entries as a result we carry over for queues greater than 4. */
474 if (rx_queue > IGB_N0_QUEUE) {
475 index = (rx_queue >> 1);
476 ivar = array_rd32(E1000_IVAR0, index);
477 if (rx_queue & 0x1) {
478 /* vector goes into third byte of register */
479 ivar = ivar & 0xFF00FFFF;
480 ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
482 /* vector goes into low byte of register */
483 ivar = ivar & 0xFFFFFF00;
484 ivar |= msix_vector | E1000_IVAR_VALID;
486 array_wr32(E1000_IVAR0, index, ivar);
488 if (tx_queue > IGB_N0_QUEUE) {
489 index = (tx_queue >> 1);
490 ivar = array_rd32(E1000_IVAR0, index);
491 if (tx_queue & 0x1) {
492 /* vector goes into high byte of register */
493 ivar = ivar & 0x00FFFFFF;
494 ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
496 /* vector goes into second byte of register */
497 ivar = ivar & 0xFFFF00FF;
498 ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
500 array_wr32(E1000_IVAR0, index, ivar);
502 q_vector->eims_value = 1 << msix_vector;
509 /* add q_vector eims value to global eims_enable_mask */
510 adapter->eims_enable_mask |= q_vector->eims_value;
512 /* configure q_vector to set itr on first interrupt */
513 q_vector->set_itr = 1;
517 * igb_configure_msix - Configure MSI-X hardware
519 * igb_configure_msix sets up the hardware to properly
520 * generate MSI-X interrupts.
522 static void igb_configure_msix(struct igb_adapter *adapter)
526 struct e1000_hw *hw = &adapter->hw;
528 adapter->eims_enable_mask = 0;
530 /* set vector for other causes, i.e. link changes */
531 switch (hw->mac.type) {
533 tmp = rd32(E1000_CTRL_EXT);
534 /* enable MSI-X PBA support*/
535 tmp |= E1000_CTRL_EXT_PBA_CLR;
537 /* Auto-Mask interrupts upon ICR read. */
538 tmp |= E1000_CTRL_EXT_EIAME;
539 tmp |= E1000_CTRL_EXT_IRCA;
541 wr32(E1000_CTRL_EXT, tmp);
543 /* enable msix_other interrupt */
544 array_wr32(E1000_MSIXBM(0), vector++,
546 adapter->eims_other = E1000_EIMS_OTHER;
552 /* Turn on MSI-X capability first, or our settings
553 * won't stick. And it will take days to debug. */
554 wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
555 E1000_GPIE_PBA | E1000_GPIE_EIAME |
558 /* enable msix_other interrupt */
559 adapter->eims_other = 1 << vector;
560 tmp = (vector++ | E1000_IVAR_VALID) << 8;
562 wr32(E1000_IVAR_MISC, tmp);
565 /* do nothing, since nothing else supports MSI-X */
567 } /* switch (hw->mac.type) */
569 adapter->eims_enable_mask |= adapter->eims_other;
571 for (i = 0; i < adapter->num_q_vectors; i++)
572 igb_assign_vector(adapter->q_vector[i], vector++);
578 * igb_request_msix - Initialize MSI-X interrupts
580 * igb_request_msix allocates MSI-X vectors and requests interrupts from the
583 static int igb_request_msix(struct igb_adapter *adapter)
585 struct net_device *netdev = adapter->netdev;
586 struct e1000_hw *hw = &adapter->hw;
587 int i, err = 0, vector = 0;
589 err = request_irq(adapter->msix_entries[vector].vector,
590 igb_msix_other, 0, netdev->name, adapter);
595 for (i = 0; i < adapter->num_q_vectors; i++) {
596 struct igb_q_vector *q_vector = adapter->q_vector[i];
598 q_vector->itr_register = hw->hw_addr + E1000_EITR(vector);
600 if (q_vector->rx_ring && q_vector->tx_ring)
601 sprintf(q_vector->name, "%s-TxRx-%u", netdev->name,
602 q_vector->rx_ring->queue_index);
603 else if (q_vector->tx_ring)
604 sprintf(q_vector->name, "%s-tx-%u", netdev->name,
605 q_vector->tx_ring->queue_index);
606 else if (q_vector->rx_ring)
607 sprintf(q_vector->name, "%s-rx-%u", netdev->name,
608 q_vector->rx_ring->queue_index);
610 sprintf(q_vector->name, "%s-unused", netdev->name);
612 err = request_irq(adapter->msix_entries[vector].vector,
613 igb_msix_ring, 0, q_vector->name,
620 igb_configure_msix(adapter);
626 static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
628 if (adapter->msix_entries) {
629 pci_disable_msix(adapter->pdev);
630 kfree(adapter->msix_entries);
631 adapter->msix_entries = NULL;
632 } else if (adapter->flags & IGB_FLAG_HAS_MSI) {
633 pci_disable_msi(adapter->pdev);
638 * igb_free_q_vectors - Free memory allocated for interrupt vectors
639 * @adapter: board private structure to initialize
641 * This function frees the memory allocated to the q_vectors. In addition if
642 * NAPI is enabled it will delete any references to the NAPI struct prior
643 * to freeing the q_vector.
645 static void igb_free_q_vectors(struct igb_adapter *adapter)
649 for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
650 struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
651 adapter->q_vector[v_idx] = NULL;
654 netif_napi_del(&q_vector->napi);
657 adapter->num_q_vectors = 0;
661 * igb_clear_interrupt_scheme - reset the device to a state of no interrupts
663 * This function resets the device so that it has 0 rx queues, tx queues, and
664 * MSI-X interrupts allocated.
666 static void igb_clear_interrupt_scheme(struct igb_adapter *adapter)
668 igb_free_queues(adapter);
669 igb_free_q_vectors(adapter);
670 igb_reset_interrupt_capability(adapter);
674 * igb_set_interrupt_capability - set MSI or MSI-X if supported
676 * Attempt to configure interrupts using the best available
677 * capabilities of the hardware and kernel.
679 static void igb_set_interrupt_capability(struct igb_adapter *adapter)
684 /* Number of supported queues. */
685 adapter->num_rx_queues = adapter->rss_queues;
686 adapter->num_tx_queues = adapter->rss_queues;
688 /* start with one vector for every rx queue */
689 numvecs = adapter->num_rx_queues;
691 /* if tx handler is separate add 1 for every tx queue */
692 if (!(adapter->flags & IGB_FLAG_QUEUE_PAIRS))
693 numvecs += adapter->num_tx_queues;
695 /* store the number of vectors reserved for queues */
696 adapter->num_q_vectors = numvecs;
698 /* add 1 vector for link status interrupts */
700 adapter->msix_entries = kcalloc(numvecs, sizeof(struct msix_entry),
702 if (!adapter->msix_entries)
705 for (i = 0; i < numvecs; i++)
706 adapter->msix_entries[i].entry = i;
708 err = pci_enable_msix(adapter->pdev,
709 adapter->msix_entries,
714 igb_reset_interrupt_capability(adapter);
716 /* If we can't do MSI-X, try MSI */
718 #ifdef CONFIG_PCI_IOV
719 /* disable SR-IOV for non MSI-X configurations */
720 if (adapter->vf_data) {
721 struct e1000_hw *hw = &adapter->hw;
722 /* disable iov and allow time for transactions to clear */
723 pci_disable_sriov(adapter->pdev);
726 kfree(adapter->vf_data);
727 adapter->vf_data = NULL;
728 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
730 dev_info(&adapter->pdev->dev, "IOV Disabled\n");
733 adapter->vfs_allocated_count = 0;
734 adapter->rss_queues = 1;
735 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
736 adapter->num_rx_queues = 1;
737 adapter->num_tx_queues = 1;
738 adapter->num_q_vectors = 1;
739 if (!pci_enable_msi(adapter->pdev))
740 adapter->flags |= IGB_FLAG_HAS_MSI;
742 /* Notify the stack of the (possibly) reduced Tx Queue count. */
743 adapter->netdev->real_num_tx_queues = adapter->num_tx_queues;
748 * igb_alloc_q_vectors - Allocate memory for interrupt vectors
749 * @adapter: board private structure to initialize
751 * We allocate one q_vector per queue interrupt. If allocation fails we
754 static int igb_alloc_q_vectors(struct igb_adapter *adapter)
756 struct igb_q_vector *q_vector;
757 struct e1000_hw *hw = &adapter->hw;
760 for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
761 q_vector = kzalloc(sizeof(struct igb_q_vector), GFP_KERNEL);
764 q_vector->adapter = adapter;
765 q_vector->itr_register = hw->hw_addr + E1000_EITR(0);
766 q_vector->itr_val = IGB_START_ITR;
767 netif_napi_add(adapter->netdev, &q_vector->napi, igb_poll, 64);
768 adapter->q_vector[v_idx] = q_vector;
773 igb_free_q_vectors(adapter);
777 static void igb_map_rx_ring_to_vector(struct igb_adapter *adapter,
778 int ring_idx, int v_idx)
780 struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
782 q_vector->rx_ring = adapter->rx_ring[ring_idx];
783 q_vector->rx_ring->q_vector = q_vector;
784 q_vector->itr_val = adapter->rx_itr_setting;
785 if (q_vector->itr_val && q_vector->itr_val <= 3)
786 q_vector->itr_val = IGB_START_ITR;
789 static void igb_map_tx_ring_to_vector(struct igb_adapter *adapter,
790 int ring_idx, int v_idx)
792 struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
794 q_vector->tx_ring = adapter->tx_ring[ring_idx];
795 q_vector->tx_ring->q_vector = q_vector;
796 q_vector->itr_val = adapter->tx_itr_setting;
797 if (q_vector->itr_val && q_vector->itr_val <= 3)
798 q_vector->itr_val = IGB_START_ITR;
802 * igb_map_ring_to_vector - maps allocated queues to vectors
804 * This function maps the recently allocated queues to vectors.
806 static int igb_map_ring_to_vector(struct igb_adapter *adapter)
811 if ((adapter->num_q_vectors < adapter->num_rx_queues) ||
812 (adapter->num_q_vectors < adapter->num_tx_queues))
815 if (adapter->num_q_vectors >=
816 (adapter->num_rx_queues + adapter->num_tx_queues)) {
817 for (i = 0; i < adapter->num_rx_queues; i++)
818 igb_map_rx_ring_to_vector(adapter, i, v_idx++);
819 for (i = 0; i < adapter->num_tx_queues; i++)
820 igb_map_tx_ring_to_vector(adapter, i, v_idx++);
822 for (i = 0; i < adapter->num_rx_queues; i++) {
823 if (i < adapter->num_tx_queues)
824 igb_map_tx_ring_to_vector(adapter, i, v_idx);
825 igb_map_rx_ring_to_vector(adapter, i, v_idx++);
827 for (; i < adapter->num_tx_queues; i++)
828 igb_map_tx_ring_to_vector(adapter, i, v_idx++);
834 * igb_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
836 * This function initializes the interrupts and allocates all of the queues.
838 static int igb_init_interrupt_scheme(struct igb_adapter *adapter)
840 struct pci_dev *pdev = adapter->pdev;
843 igb_set_interrupt_capability(adapter);
845 err = igb_alloc_q_vectors(adapter);
847 dev_err(&pdev->dev, "Unable to allocate memory for vectors\n");
848 goto err_alloc_q_vectors;
851 err = igb_alloc_queues(adapter);
853 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
854 goto err_alloc_queues;
857 err = igb_map_ring_to_vector(adapter);
859 dev_err(&pdev->dev, "Invalid q_vector to ring mapping\n");
866 igb_free_queues(adapter);
868 igb_free_q_vectors(adapter);
870 igb_reset_interrupt_capability(adapter);
875 * igb_request_irq - initialize interrupts
877 * Attempts to configure interrupts using the best available
878 * capabilities of the hardware and kernel.
880 static int igb_request_irq(struct igb_adapter *adapter)
882 struct net_device *netdev = adapter->netdev;
883 struct pci_dev *pdev = adapter->pdev;
886 if (adapter->msix_entries) {
887 err = igb_request_msix(adapter);
890 /* fall back to MSI */
891 igb_clear_interrupt_scheme(adapter);
892 if (!pci_enable_msi(adapter->pdev))
893 adapter->flags |= IGB_FLAG_HAS_MSI;
894 igb_free_all_tx_resources(adapter);
895 igb_free_all_rx_resources(adapter);
896 adapter->num_tx_queues = 1;
897 adapter->num_rx_queues = 1;
898 adapter->num_q_vectors = 1;
899 err = igb_alloc_q_vectors(adapter);
902 "Unable to allocate memory for vectors\n");
905 err = igb_alloc_queues(adapter);
908 "Unable to allocate memory for queues\n");
909 igb_free_q_vectors(adapter);
912 igb_setup_all_tx_resources(adapter);
913 igb_setup_all_rx_resources(adapter);
915 igb_assign_vector(adapter->q_vector[0], 0);
918 if (adapter->flags & IGB_FLAG_HAS_MSI) {
919 err = request_irq(adapter->pdev->irq, igb_intr_msi, 0,
920 netdev->name, adapter);
924 /* fall back to legacy interrupts */
925 igb_reset_interrupt_capability(adapter);
926 adapter->flags &= ~IGB_FLAG_HAS_MSI;
929 err = request_irq(adapter->pdev->irq, igb_intr, IRQF_SHARED,
930 netdev->name, adapter);
933 dev_err(&adapter->pdev->dev, "Error %d getting interrupt\n",
940 static void igb_free_irq(struct igb_adapter *adapter)
942 if (adapter->msix_entries) {
945 free_irq(adapter->msix_entries[vector++].vector, adapter);
947 for (i = 0; i < adapter->num_q_vectors; i++) {
948 struct igb_q_vector *q_vector = adapter->q_vector[i];
949 free_irq(adapter->msix_entries[vector++].vector,
953 free_irq(adapter->pdev->irq, adapter);
958 * igb_irq_disable - Mask off interrupt generation on the NIC
959 * @adapter: board private structure
961 static void igb_irq_disable(struct igb_adapter *adapter)
963 struct e1000_hw *hw = &adapter->hw;
966 * we need to be careful when disabling interrupts. The VFs are also
967 * mapped into these registers and so clearing the bits can cause
968 * issues on the VF drivers so we only need to clear what we set
970 if (adapter->msix_entries) {
971 u32 regval = rd32(E1000_EIAM);
972 wr32(E1000_EIAM, regval & ~adapter->eims_enable_mask);
973 wr32(E1000_EIMC, adapter->eims_enable_mask);
974 regval = rd32(E1000_EIAC);
975 wr32(E1000_EIAC, regval & ~adapter->eims_enable_mask);
981 synchronize_irq(adapter->pdev->irq);
985 * igb_irq_enable - Enable default interrupt generation settings
986 * @adapter: board private structure
988 static void igb_irq_enable(struct igb_adapter *adapter)
990 struct e1000_hw *hw = &adapter->hw;
992 if (adapter->msix_entries) {
993 u32 ims = E1000_IMS_LSC | E1000_IMS_DOUTSYNC;
994 u32 regval = rd32(E1000_EIAC);
995 wr32(E1000_EIAC, regval | adapter->eims_enable_mask);
996 regval = rd32(E1000_EIAM);
997 wr32(E1000_EIAM, regval | adapter->eims_enable_mask);
998 wr32(E1000_EIMS, adapter->eims_enable_mask);
999 if (adapter->vfs_allocated_count) {
1000 wr32(E1000_MBVFIMR, 0xFF);
1001 ims |= E1000_IMS_VMMB;
1003 if (adapter->hw.mac.type == e1000_82580)
1004 ims |= E1000_IMS_DRSTA;
1006 wr32(E1000_IMS, ims);
1008 wr32(E1000_IMS, IMS_ENABLE_MASK |
1010 wr32(E1000_IAM, IMS_ENABLE_MASK |
1015 static void igb_update_mng_vlan(struct igb_adapter *adapter)
1017 struct e1000_hw *hw = &adapter->hw;
1018 u16 vid = adapter->hw.mng_cookie.vlan_id;
1019 u16 old_vid = adapter->mng_vlan_id;
1021 if (hw->mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1022 /* add VID to filter table */
1023 igb_vfta_set(hw, vid, true);
1024 adapter->mng_vlan_id = vid;
1026 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1029 if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
1031 !vlan_group_get_device(adapter->vlgrp, old_vid)) {
1032 /* remove VID from filter table */
1033 igb_vfta_set(hw, old_vid, false);
1038 * igb_release_hw_control - release control of the h/w to f/w
1039 * @adapter: address of board private structure
1041 * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
1042 * For ASF and Pass Through versions of f/w this means that the
1043 * driver is no longer loaded.
1046 static void igb_release_hw_control(struct igb_adapter *adapter)
1048 struct e1000_hw *hw = &adapter->hw;
1051 /* Let firmware take over control of h/w */
1052 ctrl_ext = rd32(E1000_CTRL_EXT);
1053 wr32(E1000_CTRL_EXT,
1054 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1058 * igb_get_hw_control - get control of the h/w from f/w
1059 * @adapter: address of board private structure
1061 * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
1062 * For ASF and Pass Through versions of f/w this means that
1063 * the driver is loaded.
1066 static void igb_get_hw_control(struct igb_adapter *adapter)
1068 struct e1000_hw *hw = &adapter->hw;
1071 /* Let firmware know the driver has taken over */
1072 ctrl_ext = rd32(E1000_CTRL_EXT);
1073 wr32(E1000_CTRL_EXT,
1074 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1078 * igb_configure - configure the hardware for RX and TX
1079 * @adapter: private board structure
1081 static void igb_configure(struct igb_adapter *adapter)
1083 struct net_device *netdev = adapter->netdev;
1086 igb_get_hw_control(adapter);
1087 igb_set_rx_mode(netdev);
1089 igb_restore_vlan(adapter);
1091 igb_setup_tctl(adapter);
1092 igb_setup_mrqc(adapter);
1093 igb_setup_rctl(adapter);
1095 igb_configure_tx(adapter);
1096 igb_configure_rx(adapter);
1098 igb_rx_fifo_flush_82575(&adapter->hw);
1100 /* call igb_desc_unused which always leaves
1101 * at least 1 descriptor unused to make sure
1102 * next_to_use != next_to_clean */
1103 for (i = 0; i < adapter->num_rx_queues; i++) {
1104 struct igb_ring *ring = adapter->rx_ring[i];
1105 igb_alloc_rx_buffers_adv(ring, igb_desc_unused(ring));
1109 adapter->tx_queue_len = netdev->tx_queue_len;
1113 * igb_power_up_link - Power up the phy/serdes link
1114 * @adapter: address of board private structure
1116 void igb_power_up_link(struct igb_adapter *adapter)
1118 if (adapter->hw.phy.media_type == e1000_media_type_copper)
1119 igb_power_up_phy_copper(&adapter->hw);
1121 igb_power_up_serdes_link_82575(&adapter->hw);
1125 * igb_power_down_link - Power down the phy/serdes link
1126 * @adapter: address of board private structure
1128 static void igb_power_down_link(struct igb_adapter *adapter)
1130 if (adapter->hw.phy.media_type == e1000_media_type_copper)
1131 igb_power_down_phy_copper_82575(&adapter->hw);
1133 igb_shutdown_serdes_link_82575(&adapter->hw);
1137 * igb_up - Open the interface and prepare it to handle traffic
1138 * @adapter: board private structure
1140 int igb_up(struct igb_adapter *adapter)
1142 struct e1000_hw *hw = &adapter->hw;
1145 /* hardware has been reset, we need to reload some things */
1146 igb_configure(adapter);
1148 clear_bit(__IGB_DOWN, &adapter->state);
1150 for (i = 0; i < adapter->num_q_vectors; i++) {
1151 struct igb_q_vector *q_vector = adapter->q_vector[i];
1152 napi_enable(&q_vector->napi);
1154 if (adapter->msix_entries)
1155 igb_configure_msix(adapter);
1157 igb_assign_vector(adapter->q_vector[0], 0);
1159 /* Clear any pending interrupts. */
1161 igb_irq_enable(adapter);
1163 /* notify VFs that reset has been completed */
1164 if (adapter->vfs_allocated_count) {
1165 u32 reg_data = rd32(E1000_CTRL_EXT);
1166 reg_data |= E1000_CTRL_EXT_PFRSTD;
1167 wr32(E1000_CTRL_EXT, reg_data);
1170 netif_tx_start_all_queues(adapter->netdev);
1172 /* start the watchdog. */
1173 hw->mac.get_link_status = 1;
1174 schedule_work(&adapter->watchdog_task);
1179 void igb_down(struct igb_adapter *adapter)
1181 struct net_device *netdev = adapter->netdev;
1182 struct e1000_hw *hw = &adapter->hw;
1186 /* signal that we're down so the interrupt handler does not
1187 * reschedule our watchdog timer */
1188 set_bit(__IGB_DOWN, &adapter->state);
1190 /* disable receives in the hardware */
1191 rctl = rd32(E1000_RCTL);
1192 wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
1193 /* flush and sleep below */
1195 netif_tx_stop_all_queues(netdev);
1197 /* disable transmits in the hardware */
1198 tctl = rd32(E1000_TCTL);
1199 tctl &= ~E1000_TCTL_EN;
1200 wr32(E1000_TCTL, tctl);
1201 /* flush both disables and wait for them to finish */
1205 for (i = 0; i < adapter->num_q_vectors; i++) {
1206 struct igb_q_vector *q_vector = adapter->q_vector[i];
1207 napi_disable(&q_vector->napi);
1210 igb_irq_disable(adapter);
1212 del_timer_sync(&adapter->watchdog_timer);
1213 del_timer_sync(&adapter->phy_info_timer);
1215 netdev->tx_queue_len = adapter->tx_queue_len;
1216 netif_carrier_off(netdev);
1218 /* record the stats before reset*/
1219 igb_update_stats(adapter);
1221 adapter->link_speed = 0;
1222 adapter->link_duplex = 0;
1224 if (!pci_channel_offline(adapter->pdev))
1226 igb_clean_all_tx_rings(adapter);
1227 igb_clean_all_rx_rings(adapter);
1228 #ifdef CONFIG_IGB_DCA
1230 /* since we reset the hardware DCA settings were cleared */
1231 igb_setup_dca(adapter);
1235 void igb_reinit_locked(struct igb_adapter *adapter)
1237 WARN_ON(in_interrupt());
1238 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
1242 clear_bit(__IGB_RESETTING, &adapter->state);
1245 void igb_reset(struct igb_adapter *adapter)
1247 struct pci_dev *pdev = adapter->pdev;
1248 struct e1000_hw *hw = &adapter->hw;
1249 struct e1000_mac_info *mac = &hw->mac;
1250 struct e1000_fc_info *fc = &hw->fc;
1251 u32 pba = 0, tx_space, min_tx_space, min_rx_space;
1254 /* Repartition Pba for greater than 9k mtu
1255 * To take effect CTRL.RST is required.
1257 switch (mac->type) {
1259 pba = rd32(E1000_RXPBS);
1260 pba = igb_rxpbs_adjust_82580(pba);
1263 pba = rd32(E1000_RXPBS);
1264 pba &= E1000_RXPBS_SIZE_MASK_82576;
1268 pba = E1000_PBA_34K;
1272 if ((adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) &&
1273 (mac->type < e1000_82576)) {
1274 /* adjust PBA for jumbo frames */
1275 wr32(E1000_PBA, pba);
1277 /* To maintain wire speed transmits, the Tx FIFO should be
1278 * large enough to accommodate two full transmit packets,
1279 * rounded up to the next 1KB and expressed in KB. Likewise,
1280 * the Rx FIFO should be large enough to accommodate at least
1281 * one full receive packet and is similarly rounded up and
1282 * expressed in KB. */
1283 pba = rd32(E1000_PBA);
1284 /* upper 16 bits has Tx packet buffer allocation size in KB */
1285 tx_space = pba >> 16;
1286 /* lower 16 bits has Rx packet buffer allocation size in KB */
1288 /* the tx fifo also stores 16 bytes of information about the tx
1289 * but don't include ethernet FCS because hardware appends it */
1290 min_tx_space = (adapter->max_frame_size +
1291 sizeof(union e1000_adv_tx_desc) -
1293 min_tx_space = ALIGN(min_tx_space, 1024);
1294 min_tx_space >>= 10;
1295 /* software strips receive CRC, so leave room for it */
1296 min_rx_space = adapter->max_frame_size;
1297 min_rx_space = ALIGN(min_rx_space, 1024);
1298 min_rx_space >>= 10;
1300 /* If current Tx allocation is less than the min Tx FIFO size,
1301 * and the min Tx FIFO size is less than the current Rx FIFO
1302 * allocation, take space away from current Rx allocation */
1303 if (tx_space < min_tx_space &&
1304 ((min_tx_space - tx_space) < pba)) {
1305 pba = pba - (min_tx_space - tx_space);
1307 /* if short on rx space, rx wins and must trump tx
1309 if (pba < min_rx_space)
1312 wr32(E1000_PBA, pba);
1315 /* flow control settings */
1316 /* The high water mark must be low enough to fit one full frame
1317 * (or the size used for early receive) above it in the Rx FIFO.
1318 * Set it to the lower of:
1319 * - 90% of the Rx FIFO size, or
1320 * - the full Rx FIFO size minus one full frame */
1321 hwm = min(((pba << 10) * 9 / 10),
1322 ((pba << 10) - 2 * adapter->max_frame_size));
1324 fc->high_water = hwm & 0xFFF0; /* 16-byte granularity */
1325 fc->low_water = fc->high_water - 16;
1326 fc->pause_time = 0xFFFF;
1328 fc->current_mode = fc->requested_mode;
1330 /* disable receive for all VFs and wait one second */
1331 if (adapter->vfs_allocated_count) {
1333 for (i = 0 ; i < adapter->vfs_allocated_count; i++)
1334 adapter->vf_data[i].flags = 0;
1336 /* ping all the active vfs to let them know we are going down */
1337 igb_ping_all_vfs(adapter);
1339 /* disable transmits and receives */
1340 wr32(E1000_VFRE, 0);
1341 wr32(E1000_VFTE, 0);
1344 /* Allow time for pending master requests to run */
1345 hw->mac.ops.reset_hw(hw);
1348 if (hw->mac.ops.init_hw(hw))
1349 dev_err(&pdev->dev, "Hardware Error\n");
1351 if (hw->mac.type == e1000_82580) {
1352 u32 reg = rd32(E1000_PCIEMISC);
1353 wr32(E1000_PCIEMISC,
1354 reg & ~E1000_PCIEMISC_LX_DECISION);
1356 if (!netif_running(adapter->netdev))
1357 igb_power_down_link(adapter);
1359 igb_update_mng_vlan(adapter);
1361 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
1362 wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
1364 igb_get_phy_info(hw);
1367 static const struct net_device_ops igb_netdev_ops = {
1368 .ndo_open = igb_open,
1369 .ndo_stop = igb_close,
1370 .ndo_start_xmit = igb_xmit_frame_adv,
1371 .ndo_get_stats = igb_get_stats,
1372 .ndo_set_rx_mode = igb_set_rx_mode,
1373 .ndo_set_multicast_list = igb_set_rx_mode,
1374 .ndo_set_mac_address = igb_set_mac,
1375 .ndo_change_mtu = igb_change_mtu,
1376 .ndo_do_ioctl = igb_ioctl,
1377 .ndo_tx_timeout = igb_tx_timeout,
1378 .ndo_validate_addr = eth_validate_addr,
1379 .ndo_vlan_rx_register = igb_vlan_rx_register,
1380 .ndo_vlan_rx_add_vid = igb_vlan_rx_add_vid,
1381 .ndo_vlan_rx_kill_vid = igb_vlan_rx_kill_vid,
1382 .ndo_set_vf_mac = igb_ndo_set_vf_mac,
1383 .ndo_set_vf_vlan = igb_ndo_set_vf_vlan,
1384 .ndo_set_vf_tx_rate = igb_ndo_set_vf_bw,
1385 .ndo_get_vf_config = igb_ndo_get_vf_config,
1386 #ifdef CONFIG_NET_POLL_CONTROLLER
1387 .ndo_poll_controller = igb_netpoll,
1392 * igb_probe - Device Initialization Routine
1393 * @pdev: PCI device information struct
1394 * @ent: entry in igb_pci_tbl
1396 * Returns 0 on success, negative on failure
1398 * igb_probe initializes an adapter identified by a pci_dev structure.
1399 * The OS initialization, configuring of the adapter private structure,
1400 * and a hardware reset occur.
1402 static int __devinit igb_probe(struct pci_dev *pdev,
1403 const struct pci_device_id *ent)
1405 struct net_device *netdev;
1406 struct igb_adapter *adapter;
1407 struct e1000_hw *hw;
1408 u16 eeprom_data = 0;
1409 static int global_quad_port_a; /* global quad port a indication */
1410 const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
1411 unsigned long mmio_start, mmio_len;
1412 int err, pci_using_dac;
1413 u16 eeprom_apme_mask = IGB_EEPROM_APME;
1416 err = pci_enable_device_mem(pdev);
1421 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
1423 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
1427 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
1429 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
1431 dev_err(&pdev->dev, "No usable DMA "
1432 "configuration, aborting\n");
1438 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
1444 pci_enable_pcie_error_reporting(pdev);
1446 pci_set_master(pdev);
1447 pci_save_state(pdev);
1450 netdev = alloc_etherdev_mq(sizeof(struct igb_adapter),
1451 IGB_ABS_MAX_TX_QUEUES);
1453 goto err_alloc_etherdev;
1455 SET_NETDEV_DEV(netdev, &pdev->dev);
1457 pci_set_drvdata(pdev, netdev);
1458 adapter = netdev_priv(netdev);
1459 adapter->netdev = netdev;
1460 adapter->pdev = pdev;
1463 adapter->msg_enable = NETIF_MSG_DRV | NETIF_MSG_PROBE;
1465 mmio_start = pci_resource_start(pdev, 0);
1466 mmio_len = pci_resource_len(pdev, 0);
1469 hw->hw_addr = ioremap(mmio_start, mmio_len);
1473 netdev->netdev_ops = &igb_netdev_ops;
1474 igb_set_ethtool_ops(netdev);
1475 netdev->watchdog_timeo = 5 * HZ;
1477 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
1479 netdev->mem_start = mmio_start;
1480 netdev->mem_end = mmio_start + mmio_len;
1482 /* PCI config space info */
1483 hw->vendor_id = pdev->vendor;
1484 hw->device_id = pdev->device;
1485 hw->revision_id = pdev->revision;
1486 hw->subsystem_vendor_id = pdev->subsystem_vendor;
1487 hw->subsystem_device_id = pdev->subsystem_device;
1489 /* Copy the default MAC, PHY and NVM function pointers */
1490 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
1491 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
1492 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
1493 /* Initialize skew-specific constants */
1494 err = ei->get_invariants(hw);
1498 /* setup the private structure */
1499 err = igb_sw_init(adapter);
1503 igb_get_bus_info_pcie(hw);
1505 hw->phy.autoneg_wait_to_complete = false;
1507 /* Copper options */
1508 if (hw->phy.media_type == e1000_media_type_copper) {
1509 hw->phy.mdix = AUTO_ALL_MODES;
1510 hw->phy.disable_polarity_correction = false;
1511 hw->phy.ms_type = e1000_ms_hw_default;
1514 if (igb_check_reset_block(hw))
1515 dev_info(&pdev->dev,
1516 "PHY reset is blocked due to SOL/IDER session.\n");
1518 netdev->features = NETIF_F_SG |
1520 NETIF_F_HW_VLAN_TX |
1521 NETIF_F_HW_VLAN_RX |
1522 NETIF_F_HW_VLAN_FILTER;
1524 netdev->features |= NETIF_F_IPV6_CSUM;
1525 netdev->features |= NETIF_F_TSO;
1526 netdev->features |= NETIF_F_TSO6;
1527 netdev->features |= NETIF_F_GRO;
1529 netdev->vlan_features |= NETIF_F_TSO;
1530 netdev->vlan_features |= NETIF_F_TSO6;
1531 netdev->vlan_features |= NETIF_F_IP_CSUM;
1532 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
1533 netdev->vlan_features |= NETIF_F_SG;
1536 netdev->features |= NETIF_F_HIGHDMA;
1538 if (hw->mac.type >= e1000_82576)
1539 netdev->features |= NETIF_F_SCTP_CSUM;
1541 adapter->en_mng_pt = igb_enable_mng_pass_thru(hw);
1543 /* before reading the NVM, reset the controller to put the device in a
1544 * known good starting state */
1545 hw->mac.ops.reset_hw(hw);
1547 /* make sure the NVM is good */
1548 if (igb_validate_nvm_checksum(hw) < 0) {
1549 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
1554 /* copy the MAC address out of the NVM */
1555 if (hw->mac.ops.read_mac_addr(hw))
1556 dev_err(&pdev->dev, "NVM Read Error\n");
1558 memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
1559 memcpy(netdev->perm_addr, hw->mac.addr, netdev->addr_len);
1561 if (!is_valid_ether_addr(netdev->perm_addr)) {
1562 dev_err(&pdev->dev, "Invalid MAC Address\n");
1567 setup_timer(&adapter->watchdog_timer, &igb_watchdog,
1568 (unsigned long) adapter);
1569 setup_timer(&adapter->phy_info_timer, &igb_update_phy_info,
1570 (unsigned long) adapter);
1572 INIT_WORK(&adapter->reset_task, igb_reset_task);
1573 INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
1575 /* Initialize link properties that are user-changeable */
1576 adapter->fc_autoneg = true;
1577 hw->mac.autoneg = true;
1578 hw->phy.autoneg_advertised = 0x2f;
1580 hw->fc.requested_mode = e1000_fc_default;
1581 hw->fc.current_mode = e1000_fc_default;
1583 igb_validate_mdi_setting(hw);
1585 /* Initial Wake on LAN setting If APM wake is enabled in the EEPROM,
1586 * enable the ACPI Magic Packet filter
1589 if (hw->bus.func == 0)
1590 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
1591 else if (hw->mac.type == e1000_82580)
1592 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A +
1593 NVM_82580_LAN_FUNC_OFFSET(hw->bus.func), 1,
1595 else if (hw->bus.func == 1)
1596 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
1598 if (eeprom_data & eeprom_apme_mask)
1599 adapter->eeprom_wol |= E1000_WUFC_MAG;
1601 /* now that we have the eeprom settings, apply the special cases where
1602 * the eeprom may be wrong or the board simply won't support wake on
1603 * lan on a particular port */
1604 switch (pdev->device) {
1605 case E1000_DEV_ID_82575GB_QUAD_COPPER:
1606 adapter->eeprom_wol = 0;
1608 case E1000_DEV_ID_82575EB_FIBER_SERDES:
1609 case E1000_DEV_ID_82576_FIBER:
1610 case E1000_DEV_ID_82576_SERDES:
1611 /* Wake events only supported on port A for dual fiber
1612 * regardless of eeprom setting */
1613 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
1614 adapter->eeprom_wol = 0;
1616 case E1000_DEV_ID_82576_QUAD_COPPER:
1617 /* if quad port adapter, disable WoL on all but port A */
1618 if (global_quad_port_a != 0)
1619 adapter->eeprom_wol = 0;
1621 adapter->flags |= IGB_FLAG_QUAD_PORT_A;
1622 /* Reset for multiple quad port adapters */
1623 if (++global_quad_port_a == 4)
1624 global_quad_port_a = 0;
1628 /* initialize the wol settings based on the eeprom settings */
1629 adapter->wol = adapter->eeprom_wol;
1630 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
1632 /* reset the hardware with the new settings */
1635 /* let the f/w know that the h/w is now under the control of the
1637 igb_get_hw_control(adapter);
1639 strcpy(netdev->name, "eth%d");
1640 err = register_netdev(netdev);
1644 /* carrier off reporting is important to ethtool even BEFORE open */
1645 netif_carrier_off(netdev);
1647 #ifdef CONFIG_IGB_DCA
1648 if (dca_add_requester(&pdev->dev) == 0) {
1649 adapter->flags |= IGB_FLAG_DCA_ENABLED;
1650 dev_info(&pdev->dev, "DCA enabled\n");
1651 igb_setup_dca(adapter);
1655 dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
1656 /* print bus type/speed/width info */
1657 dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
1659 ((hw->bus.speed == e1000_bus_speed_2500) ? "2.5Gb/s" :
1661 ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
1662 (hw->bus.width == e1000_bus_width_pcie_x2) ? "Width x2" :
1663 (hw->bus.width == e1000_bus_width_pcie_x1) ? "Width x1" :
1667 igb_read_part_num(hw, &part_num);
1668 dev_info(&pdev->dev, "%s: PBA No: %06x-%03x\n", netdev->name,
1669 (part_num >> 8), (part_num & 0xff));
1671 dev_info(&pdev->dev,
1672 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
1673 adapter->msix_entries ? "MSI-X" :
1674 (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
1675 adapter->num_rx_queues, adapter->num_tx_queues);
1680 igb_release_hw_control(adapter);
1682 if (!igb_check_reset_block(hw))
1685 if (hw->flash_address)
1686 iounmap(hw->flash_address);
1688 igb_clear_interrupt_scheme(adapter);
1689 iounmap(hw->hw_addr);
1691 free_netdev(netdev);
1693 pci_release_selected_regions(pdev,
1694 pci_select_bars(pdev, IORESOURCE_MEM));
1697 pci_disable_device(pdev);
1702 * igb_remove - Device Removal Routine
1703 * @pdev: PCI device information struct
1705 * igb_remove is called by the PCI subsystem to alert the driver
1706 * that it should release a PCI device. The could be caused by a
1707 * Hot-Plug event, or because the driver is going to be removed from
1710 static void __devexit igb_remove(struct pci_dev *pdev)
1712 struct net_device *netdev = pci_get_drvdata(pdev);
1713 struct igb_adapter *adapter = netdev_priv(netdev);
1714 struct e1000_hw *hw = &adapter->hw;
1716 /* flush_scheduled work may reschedule our watchdog task, so
1717 * explicitly disable watchdog tasks from being rescheduled */
1718 set_bit(__IGB_DOWN, &adapter->state);
1719 del_timer_sync(&adapter->watchdog_timer);
1720 del_timer_sync(&adapter->phy_info_timer);
1722 flush_scheduled_work();
1724 #ifdef CONFIG_IGB_DCA
1725 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
1726 dev_info(&pdev->dev, "DCA disabled\n");
1727 dca_remove_requester(&pdev->dev);
1728 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
1729 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
1733 /* Release control of h/w to f/w. If f/w is AMT enabled, this
1734 * would have already happened in close and is redundant. */
1735 igb_release_hw_control(adapter);
1737 unregister_netdev(netdev);
1739 igb_clear_interrupt_scheme(adapter);
1741 #ifdef CONFIG_PCI_IOV
1742 /* reclaim resources allocated to VFs */
1743 if (adapter->vf_data) {
1744 /* disable iov and allow time for transactions to clear */
1745 pci_disable_sriov(pdev);
1748 kfree(adapter->vf_data);
1749 adapter->vf_data = NULL;
1750 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
1752 dev_info(&pdev->dev, "IOV Disabled\n");
1756 iounmap(hw->hw_addr);
1757 if (hw->flash_address)
1758 iounmap(hw->flash_address);
1759 pci_release_selected_regions(pdev,
1760 pci_select_bars(pdev, IORESOURCE_MEM));
1762 free_netdev(netdev);
1764 pci_disable_pcie_error_reporting(pdev);
1766 pci_disable_device(pdev);
1770 * igb_probe_vfs - Initialize vf data storage and add VFs to pci config space
1771 * @adapter: board private structure to initialize
1773 * This function initializes the vf specific data storage and then attempts to
1774 * allocate the VFs. The reason for ordering it this way is because it is much
1775 * mor expensive time wise to disable SR-IOV than it is to allocate and free
1776 * the memory for the VFs.
1778 static void __devinit igb_probe_vfs(struct igb_adapter * adapter)
1780 #ifdef CONFIG_PCI_IOV
1781 struct pci_dev *pdev = adapter->pdev;
1783 if (adapter->vfs_allocated_count > 7)
1784 adapter->vfs_allocated_count = 7;
1786 if (adapter->vfs_allocated_count) {
1787 adapter->vf_data = kcalloc(adapter->vfs_allocated_count,
1788 sizeof(struct vf_data_storage),
1790 /* if allocation failed then we do not support SR-IOV */
1791 if (!adapter->vf_data) {
1792 adapter->vfs_allocated_count = 0;
1793 dev_err(&pdev->dev, "Unable to allocate memory for VF "
1798 if (pci_enable_sriov(pdev, adapter->vfs_allocated_count)) {
1799 kfree(adapter->vf_data);
1800 adapter->vf_data = NULL;
1801 #endif /* CONFIG_PCI_IOV */
1802 adapter->vfs_allocated_count = 0;
1803 #ifdef CONFIG_PCI_IOV
1805 unsigned char mac_addr[ETH_ALEN];
1807 dev_info(&pdev->dev, "%d vfs allocated\n",
1808 adapter->vfs_allocated_count);
1809 for (i = 0; i < adapter->vfs_allocated_count; i++) {
1810 random_ether_addr(mac_addr);
1811 igb_set_vf_mac(adapter, i, mac_addr);
1814 #endif /* CONFIG_PCI_IOV */
1819 * igb_init_hw_timer - Initialize hardware timer used with IEEE 1588 timestamp
1820 * @adapter: board private structure to initialize
1822 * igb_init_hw_timer initializes the function pointer and values for the hw
1823 * timer found in hardware.
1825 static void igb_init_hw_timer(struct igb_adapter *adapter)
1827 struct e1000_hw *hw = &adapter->hw;
1829 switch (hw->mac.type) {
1831 memset(&adapter->cycles, 0, sizeof(adapter->cycles));
1832 adapter->cycles.read = igb_read_clock;
1833 adapter->cycles.mask = CLOCKSOURCE_MASK(64);
1834 adapter->cycles.mult = 1;
1836 * The 82580 timesync updates the system timer every 8ns by 8ns
1837 * and the value cannot be shifted. Instead we need to shift
1838 * the registers to generate a 64bit timer value. As a result
1839 * SYSTIMR/L/H, TXSTMPL/H, RXSTMPL/H all have to be shifted by
1840 * 24 in order to generate a larger value for synchronization.
1842 adapter->cycles.shift = IGB_82580_TSYNC_SHIFT;
1843 /* disable system timer temporarily by setting bit 31 */
1844 wr32(E1000_TSAUXC, 0x80000000);
1847 /* Set registers so that rollover occurs soon to test this. */
1848 wr32(E1000_SYSTIMR, 0x00000000);
1849 wr32(E1000_SYSTIML, 0x80000000);
1850 wr32(E1000_SYSTIMH, 0x000000FF);
1853 /* enable system timer by clearing bit 31 */
1854 wr32(E1000_TSAUXC, 0x0);
1857 timecounter_init(&adapter->clock,
1859 ktime_to_ns(ktime_get_real()));
1861 * Synchronize our NIC clock against system wall clock. NIC
1862 * time stamp reading requires ~3us per sample, each sample
1863 * was pretty stable even under load => only require 10
1864 * samples for each offset comparison.
1866 memset(&adapter->compare, 0, sizeof(adapter->compare));
1867 adapter->compare.source = &adapter->clock;
1868 adapter->compare.target = ktime_get_real;
1869 adapter->compare.num_samples = 10;
1870 timecompare_update(&adapter->compare, 0);
1874 * Initialize hardware timer: we keep it running just in case
1875 * that some program needs it later on.
1877 memset(&adapter->cycles, 0, sizeof(adapter->cycles));
1878 adapter->cycles.read = igb_read_clock;
1879 adapter->cycles.mask = CLOCKSOURCE_MASK(64);
1880 adapter->cycles.mult = 1;
1882 * Scale the NIC clock cycle by a large factor so that
1883 * relatively small clock corrections can be added or
1884 * substracted at each clock tick. The drawbacks of a large
1885 * factor are a) that the clock register overflows more quickly
1886 * (not such a big deal) and b) that the increment per tick has
1887 * to fit into 24 bits. As a result we need to use a shift of
1888 * 19 so we can fit a value of 16 into the TIMINCA register.
1890 adapter->cycles.shift = IGB_82576_TSYNC_SHIFT;
1892 (1 << E1000_TIMINCA_16NS_SHIFT) |
1893 (16 << IGB_82576_TSYNC_SHIFT));
1895 /* Set registers so that rollover occurs soon to test this. */
1896 wr32(E1000_SYSTIML, 0x00000000);
1897 wr32(E1000_SYSTIMH, 0xFF800000);
1900 timecounter_init(&adapter->clock,
1902 ktime_to_ns(ktime_get_real()));
1904 * Synchronize our NIC clock against system wall clock. NIC
1905 * time stamp reading requires ~3us per sample, each sample
1906 * was pretty stable even under load => only require 10
1907 * samples for each offset comparison.
1909 memset(&adapter->compare, 0, sizeof(adapter->compare));
1910 adapter->compare.source = &adapter->clock;
1911 adapter->compare.target = ktime_get_real;
1912 adapter->compare.num_samples = 10;
1913 timecompare_update(&adapter->compare, 0);
1916 /* 82575 does not support timesync */
1924 * igb_sw_init - Initialize general software structures (struct igb_adapter)
1925 * @adapter: board private structure to initialize
1927 * igb_sw_init initializes the Adapter private data structure.
1928 * Fields are initialized based on PCI device information and
1929 * OS network device settings (MTU size).
1931 static int __devinit igb_sw_init(struct igb_adapter *adapter)
1933 struct e1000_hw *hw = &adapter->hw;
1934 struct net_device *netdev = adapter->netdev;
1935 struct pci_dev *pdev = adapter->pdev;
1937 pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
1939 adapter->tx_ring_count = IGB_DEFAULT_TXD;
1940 adapter->rx_ring_count = IGB_DEFAULT_RXD;
1941 adapter->rx_itr_setting = IGB_DEFAULT_ITR;
1942 adapter->tx_itr_setting = IGB_DEFAULT_ITR;
1944 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1945 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1947 #ifdef CONFIG_PCI_IOV
1948 if (hw->mac.type == e1000_82576)
1949 adapter->vfs_allocated_count = max_vfs;
1951 #endif /* CONFIG_PCI_IOV */
1952 adapter->rss_queues = min_t(u32, IGB_MAX_RX_QUEUES, num_online_cpus());
1955 * if rss_queues > 4 or vfs are going to be allocated with rss_queues
1956 * then we should combine the queues into a queue pair in order to
1957 * conserve interrupts due to limited supply
1959 if ((adapter->rss_queues > 4) ||
1960 ((adapter->rss_queues > 1) && (adapter->vfs_allocated_count > 6)))
1961 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
1963 /* This call may decrease the number of queues */
1964 if (igb_init_interrupt_scheme(adapter)) {
1965 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
1969 igb_init_hw_timer(adapter);
1970 igb_probe_vfs(adapter);
1972 /* Explicitly disable IRQ since the NIC can be in any state. */
1973 igb_irq_disable(adapter);
1975 set_bit(__IGB_DOWN, &adapter->state);
1980 * igb_open - Called when a network interface is made active
1981 * @netdev: network interface device structure
1983 * Returns 0 on success, negative value on failure
1985 * The open entry point is called when a network interface is made
1986 * active by the system (IFF_UP). At this point all resources needed
1987 * for transmit and receive operations are allocated, the interrupt
1988 * handler is registered with the OS, the watchdog timer is started,
1989 * and the stack is notified that the interface is ready.
1991 static int igb_open(struct net_device *netdev)
1993 struct igb_adapter *adapter = netdev_priv(netdev);
1994 struct e1000_hw *hw = &adapter->hw;
1998 /* disallow open during test */
1999 if (test_bit(__IGB_TESTING, &adapter->state))
2002 netif_carrier_off(netdev);
2004 /* allocate transmit descriptors */
2005 err = igb_setup_all_tx_resources(adapter);
2009 /* allocate receive descriptors */
2010 err = igb_setup_all_rx_resources(adapter);
2014 igb_power_up_link(adapter);
2016 /* before we allocate an interrupt, we must be ready to handle it.
2017 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
2018 * as soon as we call pci_request_irq, so we have to setup our
2019 * clean_rx handler before we do so. */
2020 igb_configure(adapter);
2022 err = igb_request_irq(adapter);
2026 /* From here on the code is the same as igb_up() */
2027 clear_bit(__IGB_DOWN, &adapter->state);
2029 for (i = 0; i < adapter->num_q_vectors; i++) {
2030 struct igb_q_vector *q_vector = adapter->q_vector[i];
2031 napi_enable(&q_vector->napi);
2034 /* Clear any pending interrupts. */
2037 igb_irq_enable(adapter);
2039 /* notify VFs that reset has been completed */
2040 if (adapter->vfs_allocated_count) {
2041 u32 reg_data = rd32(E1000_CTRL_EXT);
2042 reg_data |= E1000_CTRL_EXT_PFRSTD;
2043 wr32(E1000_CTRL_EXT, reg_data);
2046 netif_tx_start_all_queues(netdev);
2048 /* start the watchdog. */
2049 hw->mac.get_link_status = 1;
2050 schedule_work(&adapter->watchdog_task);
2055 igb_release_hw_control(adapter);
2056 igb_power_down_link(adapter);
2057 igb_free_all_rx_resources(adapter);
2059 igb_free_all_tx_resources(adapter);
2067 * igb_close - Disables a network interface
2068 * @netdev: network interface device structure
2070 * Returns 0, this is not allowed to fail
2072 * The close entry point is called when an interface is de-activated
2073 * by the OS. The hardware is still under the driver's control, but
2074 * needs to be disabled. A global MAC reset is issued to stop the
2075 * hardware, and all transmit and receive resources are freed.
2077 static int igb_close(struct net_device *netdev)
2079 struct igb_adapter *adapter = netdev_priv(netdev);
2081 WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
2084 igb_free_irq(adapter);
2086 igb_free_all_tx_resources(adapter);
2087 igb_free_all_rx_resources(adapter);
2093 * igb_setup_tx_resources - allocate Tx resources (Descriptors)
2094 * @tx_ring: tx descriptor ring (for a specific queue) to setup
2096 * Return 0 on success, negative on failure
2098 int igb_setup_tx_resources(struct igb_ring *tx_ring)
2100 struct pci_dev *pdev = tx_ring->pdev;
2103 size = sizeof(struct igb_buffer) * tx_ring->count;
2104 tx_ring->buffer_info = vmalloc(size);
2105 if (!tx_ring->buffer_info)
2107 memset(tx_ring->buffer_info, 0, size);
2109 /* round up to nearest 4K */
2110 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
2111 tx_ring->size = ALIGN(tx_ring->size, 4096);
2113 tx_ring->desc = pci_alloc_consistent(pdev,
2120 tx_ring->next_to_use = 0;
2121 tx_ring->next_to_clean = 0;
2125 vfree(tx_ring->buffer_info);
2127 "Unable to allocate memory for the transmit descriptor ring\n");
2132 * igb_setup_all_tx_resources - wrapper to allocate Tx resources
2133 * (Descriptors) for all queues
2134 * @adapter: board private structure
2136 * Return 0 on success, negative on failure
2138 static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
2140 struct pci_dev *pdev = adapter->pdev;
2143 for (i = 0; i < adapter->num_tx_queues; i++) {
2144 err = igb_setup_tx_resources(adapter->tx_ring[i]);
2147 "Allocation for Tx Queue %u failed\n", i);
2148 for (i--; i >= 0; i--)
2149 igb_free_tx_resources(adapter->tx_ring[i]);
2154 for (i = 0; i < IGB_ABS_MAX_TX_QUEUES; i++) {
2155 int r_idx = i % adapter->num_tx_queues;
2156 adapter->multi_tx_table[i] = adapter->tx_ring[r_idx];
2162 * igb_setup_tctl - configure the transmit control registers
2163 * @adapter: Board private structure
2165 void igb_setup_tctl(struct igb_adapter *adapter)
2167 struct e1000_hw *hw = &adapter->hw;
2170 /* disable queue 0 which is enabled by default on 82575 and 82576 */
2171 wr32(E1000_TXDCTL(0), 0);
2173 /* Program the Transmit Control Register */
2174 tctl = rd32(E1000_TCTL);
2175 tctl &= ~E1000_TCTL_CT;
2176 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2177 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2179 igb_config_collision_dist(hw);
2181 /* Enable transmits */
2182 tctl |= E1000_TCTL_EN;
2184 wr32(E1000_TCTL, tctl);
2188 * igb_configure_tx_ring - Configure transmit ring after Reset
2189 * @adapter: board private structure
2190 * @ring: tx ring to configure
2192 * Configure a transmit ring after a reset.
2194 void igb_configure_tx_ring(struct igb_adapter *adapter,
2195 struct igb_ring *ring)
2197 struct e1000_hw *hw = &adapter->hw;
2199 u64 tdba = ring->dma;
2200 int reg_idx = ring->reg_idx;
2202 /* disable the queue */
2203 txdctl = rd32(E1000_TXDCTL(reg_idx));
2204 wr32(E1000_TXDCTL(reg_idx),
2205 txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
2209 wr32(E1000_TDLEN(reg_idx),
2210 ring->count * sizeof(union e1000_adv_tx_desc));
2211 wr32(E1000_TDBAL(reg_idx),
2212 tdba & 0x00000000ffffffffULL);
2213 wr32(E1000_TDBAH(reg_idx), tdba >> 32);
2215 ring->head = hw->hw_addr + E1000_TDH(reg_idx);
2216 ring->tail = hw->hw_addr + E1000_TDT(reg_idx);
2217 writel(0, ring->head);
2218 writel(0, ring->tail);
2220 txdctl |= IGB_TX_PTHRESH;
2221 txdctl |= IGB_TX_HTHRESH << 8;
2222 txdctl |= IGB_TX_WTHRESH << 16;
2224 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
2225 wr32(E1000_TXDCTL(reg_idx), txdctl);
2229 * igb_configure_tx - Configure transmit Unit after Reset
2230 * @adapter: board private structure
2232 * Configure the Tx unit of the MAC after a reset.
2234 static void igb_configure_tx(struct igb_adapter *adapter)
2238 for (i = 0; i < adapter->num_tx_queues; i++)
2239 igb_configure_tx_ring(adapter, adapter->tx_ring[i]);
2243 * igb_setup_rx_resources - allocate Rx resources (Descriptors)
2244 * @rx_ring: rx descriptor ring (for a specific queue) to setup
2246 * Returns 0 on success, negative on failure
2248 int igb_setup_rx_resources(struct igb_ring *rx_ring)
2250 struct pci_dev *pdev = rx_ring->pdev;
2253 size = sizeof(struct igb_buffer) * rx_ring->count;
2254 rx_ring->buffer_info = vmalloc(size);
2255 if (!rx_ring->buffer_info)
2257 memset(rx_ring->buffer_info, 0, size);
2259 desc_len = sizeof(union e1000_adv_rx_desc);
2261 /* Round up to nearest 4K */
2262 rx_ring->size = rx_ring->count * desc_len;
2263 rx_ring->size = ALIGN(rx_ring->size, 4096);
2265 rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size,
2271 rx_ring->next_to_clean = 0;
2272 rx_ring->next_to_use = 0;
2277 vfree(rx_ring->buffer_info);
2278 rx_ring->buffer_info = NULL;
2279 dev_err(&pdev->dev, "Unable to allocate memory for "
2280 "the receive descriptor ring\n");
2285 * igb_setup_all_rx_resources - wrapper to allocate Rx resources
2286 * (Descriptors) for all queues
2287 * @adapter: board private structure
2289 * Return 0 on success, negative on failure
2291 static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
2293 struct pci_dev *pdev = adapter->pdev;
2296 for (i = 0; i < adapter->num_rx_queues; i++) {
2297 err = igb_setup_rx_resources(adapter->rx_ring[i]);
2300 "Allocation for Rx Queue %u failed\n", i);
2301 for (i--; i >= 0; i--)
2302 igb_free_rx_resources(adapter->rx_ring[i]);
2311 * igb_setup_mrqc - configure the multiple receive queue control registers
2312 * @adapter: Board private structure
2314 static void igb_setup_mrqc(struct igb_adapter *adapter)
2316 struct e1000_hw *hw = &adapter->hw;
2318 u32 j, num_rx_queues, shift = 0, shift2 = 0;
2323 static const u8 rsshash[40] = {
2324 0x6d, 0x5a, 0x56, 0xda, 0x25, 0x5b, 0x0e, 0xc2, 0x41, 0x67,
2325 0x25, 0x3d, 0x43, 0xa3, 0x8f, 0xb0, 0xd0, 0xca, 0x2b, 0xcb,
2326 0xae, 0x7b, 0x30, 0xb4, 0x77, 0xcb, 0x2d, 0xa3, 0x80, 0x30,
2327 0xf2, 0x0c, 0x6a, 0x42, 0xb7, 0x3b, 0xbe, 0xac, 0x01, 0xfa };
2329 /* Fill out hash function seeds */
2330 for (j = 0; j < 10; j++) {
2331 u32 rsskey = rsshash[(j * 4)];
2332 rsskey |= rsshash[(j * 4) + 1] << 8;
2333 rsskey |= rsshash[(j * 4) + 2] << 16;
2334 rsskey |= rsshash[(j * 4) + 3] << 24;
2335 array_wr32(E1000_RSSRK(0), j, rsskey);
2338 num_rx_queues = adapter->rss_queues;
2340 if (adapter->vfs_allocated_count) {
2341 /* 82575 and 82576 supports 2 RSS queues for VMDq */
2342 switch (hw->mac.type) {
2358 if (hw->mac.type == e1000_82575)
2362 for (j = 0; j < (32 * 4); j++) {
2363 reta.bytes[j & 3] = (j % num_rx_queues) << shift;
2365 reta.bytes[j & 3] |= num_rx_queues << shift2;
2367 wr32(E1000_RETA(j >> 2), reta.dword);
2371 * Disable raw packet checksumming so that RSS hash is placed in
2372 * descriptor on writeback. No need to enable TCP/UDP/IP checksum
2373 * offloads as they are enabled by default
2375 rxcsum = rd32(E1000_RXCSUM);
2376 rxcsum |= E1000_RXCSUM_PCSD;
2378 if (adapter->hw.mac.type >= e1000_82576)
2379 /* Enable Receive Checksum Offload for SCTP */
2380 rxcsum |= E1000_RXCSUM_CRCOFL;
2382 /* Don't need to set TUOFL or IPOFL, they default to 1 */
2383 wr32(E1000_RXCSUM, rxcsum);
2385 /* If VMDq is enabled then we set the appropriate mode for that, else
2386 * we default to RSS so that an RSS hash is calculated per packet even
2387 * if we are only using one queue */
2388 if (adapter->vfs_allocated_count) {
2389 if (hw->mac.type > e1000_82575) {
2390 /* Set the default pool for the PF's first queue */
2391 u32 vtctl = rd32(E1000_VT_CTL);
2392 vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK |
2393 E1000_VT_CTL_DISABLE_DEF_POOL);
2394 vtctl |= adapter->vfs_allocated_count <<
2395 E1000_VT_CTL_DEFAULT_POOL_SHIFT;
2396 wr32(E1000_VT_CTL, vtctl);
2398 if (adapter->rss_queues > 1)
2399 mrqc = E1000_MRQC_ENABLE_VMDQ_RSS_2Q;
2401 mrqc = E1000_MRQC_ENABLE_VMDQ;
2403 mrqc = E1000_MRQC_ENABLE_RSS_4Q;
2405 igb_vmm_control(adapter);
2407 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
2408 E1000_MRQC_RSS_FIELD_IPV4_TCP);
2409 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6 |
2410 E1000_MRQC_RSS_FIELD_IPV6_TCP);
2411 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4_UDP |
2412 E1000_MRQC_RSS_FIELD_IPV6_UDP);
2413 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
2414 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
2416 wr32(E1000_MRQC, mrqc);
2420 * igb_setup_rctl - configure the receive control registers
2421 * @adapter: Board private structure
2423 void igb_setup_rctl(struct igb_adapter *adapter)
2425 struct e1000_hw *hw = &adapter->hw;
2428 rctl = rd32(E1000_RCTL);
2430 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2431 rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
2433 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
2434 (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2437 * enable stripping of CRC. It's unlikely this will break BMC
2438 * redirection as it did with e1000. Newer features require
2439 * that the HW strips the CRC.
2441 rctl |= E1000_RCTL_SECRC;
2443 /* disable store bad packets and clear size bits. */
2444 rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
2446 /* enable LPE to prevent packets larger than max_frame_size */
2447 rctl |= E1000_RCTL_LPE;
2449 /* disable queue 0 to prevent tail write w/o re-config */
2450 wr32(E1000_RXDCTL(0), 0);
2452 /* Attention!!! For SR-IOV PF driver operations you must enable
2453 * queue drop for all VF and PF queues to prevent head of line blocking
2454 * if an un-trusted VF does not provide descriptors to hardware.
2456 if (adapter->vfs_allocated_count) {
2457 /* set all queue drop enable bits */
2458 wr32(E1000_QDE, ALL_QUEUES);
2461 wr32(E1000_RCTL, rctl);
2464 static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
2467 struct e1000_hw *hw = &adapter->hw;
2470 /* if it isn't the PF check to see if VFs are enabled and
2471 * increase the size to support vlan tags */
2472 if (vfn < adapter->vfs_allocated_count &&
2473 adapter->vf_data[vfn].vlans_enabled)
2474 size += VLAN_TAG_SIZE;
2476 vmolr = rd32(E1000_VMOLR(vfn));
2477 vmolr &= ~E1000_VMOLR_RLPML_MASK;
2478 vmolr |= size | E1000_VMOLR_LPE;
2479 wr32(E1000_VMOLR(vfn), vmolr);
2485 * igb_rlpml_set - set maximum receive packet size
2486 * @adapter: board private structure
2488 * Configure maximum receivable packet size.
2490 static void igb_rlpml_set(struct igb_adapter *adapter)
2492 u32 max_frame_size = adapter->max_frame_size;
2493 struct e1000_hw *hw = &adapter->hw;
2494 u16 pf_id = adapter->vfs_allocated_count;
2497 max_frame_size += VLAN_TAG_SIZE;
2499 /* if vfs are enabled we set RLPML to the largest possible request
2500 * size and set the VMOLR RLPML to the size we need */
2502 igb_set_vf_rlpml(adapter, max_frame_size, pf_id);
2503 max_frame_size = MAX_JUMBO_FRAME_SIZE;
2506 wr32(E1000_RLPML, max_frame_size);
2509 static inline void igb_set_vmolr(struct igb_adapter *adapter,
2512 struct e1000_hw *hw = &adapter->hw;
2516 * This register exists only on 82576 and newer so if we are older then
2517 * we should exit and do nothing
2519 if (hw->mac.type < e1000_82576)
2522 vmolr = rd32(E1000_VMOLR(vfn));
2523 vmolr |= E1000_VMOLR_STRVLAN; /* Strip vlan tags */
2525 vmolr |= E1000_VMOLR_AUPE; /* Accept untagged packets */
2527 vmolr &= ~(E1000_VMOLR_AUPE); /* Tagged packets ONLY */
2529 /* clear all bits that might not be set */
2530 vmolr &= ~(E1000_VMOLR_BAM | E1000_VMOLR_RSSE);
2532 if (adapter->rss_queues > 1 && vfn == adapter->vfs_allocated_count)
2533 vmolr |= E1000_VMOLR_RSSE; /* enable RSS */
2535 * for VMDq only allow the VFs and pool 0 to accept broadcast and
2538 if (vfn <= adapter->vfs_allocated_count)
2539 vmolr |= E1000_VMOLR_BAM; /* Accept broadcast */
2541 wr32(E1000_VMOLR(vfn), vmolr);
2545 * igb_configure_rx_ring - Configure a receive ring after Reset
2546 * @adapter: board private structure
2547 * @ring: receive ring to be configured
2549 * Configure the Rx unit of the MAC after a reset.
2551 void igb_configure_rx_ring(struct igb_adapter *adapter,
2552 struct igb_ring *ring)
2554 struct e1000_hw *hw = &adapter->hw;
2555 u64 rdba = ring->dma;
2556 int reg_idx = ring->reg_idx;
2559 /* disable the queue */
2560 rxdctl = rd32(E1000_RXDCTL(reg_idx));
2561 wr32(E1000_RXDCTL(reg_idx),
2562 rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
2564 /* Set DMA base address registers */
2565 wr32(E1000_RDBAL(reg_idx),
2566 rdba & 0x00000000ffffffffULL);
2567 wr32(E1000_RDBAH(reg_idx), rdba >> 32);
2568 wr32(E1000_RDLEN(reg_idx),
2569 ring->count * sizeof(union e1000_adv_rx_desc));
2571 /* initialize head and tail */
2572 ring->head = hw->hw_addr + E1000_RDH(reg_idx);
2573 ring->tail = hw->hw_addr + E1000_RDT(reg_idx);
2574 writel(0, ring->head);
2575 writel(0, ring->tail);
2577 /* set descriptor configuration */
2578 if (ring->rx_buffer_len < IGB_RXBUFFER_1024) {
2579 srrctl = ALIGN(ring->rx_buffer_len, 64) <<
2580 E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
2581 #if (PAGE_SIZE / 2) > IGB_RXBUFFER_16384
2582 srrctl |= IGB_RXBUFFER_16384 >>
2583 E1000_SRRCTL_BSIZEPKT_SHIFT;
2585 srrctl |= (PAGE_SIZE / 2) >>
2586 E1000_SRRCTL_BSIZEPKT_SHIFT;
2588 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2590 srrctl = ALIGN(ring->rx_buffer_len, 1024) >>
2591 E1000_SRRCTL_BSIZEPKT_SHIFT;
2592 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
2594 /* Only set Drop Enable if we are supporting multiple queues */
2595 if (adapter->vfs_allocated_count || adapter->num_rx_queues > 1)
2596 srrctl |= E1000_SRRCTL_DROP_EN;
2598 wr32(E1000_SRRCTL(reg_idx), srrctl);
2600 /* set filtering for VMDQ pools */
2601 igb_set_vmolr(adapter, reg_idx & 0x7, true);
2603 /* enable receive descriptor fetching */
2604 rxdctl = rd32(E1000_RXDCTL(reg_idx));
2605 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
2606 rxdctl &= 0xFFF00000;
2607 rxdctl |= IGB_RX_PTHRESH;
2608 rxdctl |= IGB_RX_HTHRESH << 8;
2609 rxdctl |= IGB_RX_WTHRESH << 16;
2610 wr32(E1000_RXDCTL(reg_idx), rxdctl);
2614 * igb_configure_rx - Configure receive Unit after Reset
2615 * @adapter: board private structure
2617 * Configure the Rx unit of the MAC after a reset.
2619 static void igb_configure_rx(struct igb_adapter *adapter)
2623 /* set UTA to appropriate mode */
2624 igb_set_uta(adapter);
2626 /* set the correct pool for the PF default MAC address in entry 0 */
2627 igb_rar_set_qsel(adapter, adapter->hw.mac.addr, 0,
2628 adapter->vfs_allocated_count);
2630 /* Setup the HW Rx Head and Tail Descriptor Pointers and
2631 * the Base and Length of the Rx Descriptor Ring */
2632 for (i = 0; i < adapter->num_rx_queues; i++)
2633 igb_configure_rx_ring(adapter, adapter->rx_ring[i]);
2637 * igb_free_tx_resources - Free Tx Resources per Queue
2638 * @tx_ring: Tx descriptor ring for a specific queue
2640 * Free all transmit software resources
2642 void igb_free_tx_resources(struct igb_ring *tx_ring)
2644 igb_clean_tx_ring(tx_ring);
2646 vfree(tx_ring->buffer_info);
2647 tx_ring->buffer_info = NULL;
2649 /* if not set, then don't free */
2653 pci_free_consistent(tx_ring->pdev, tx_ring->size,
2654 tx_ring->desc, tx_ring->dma);
2656 tx_ring->desc = NULL;
2660 * igb_free_all_tx_resources - Free Tx Resources for All Queues
2661 * @adapter: board private structure
2663 * Free all transmit software resources
2665 static void igb_free_all_tx_resources(struct igb_adapter *adapter)
2669 for (i = 0; i < adapter->num_tx_queues; i++)
2670 igb_free_tx_resources(adapter->tx_ring[i]);
2673 void igb_unmap_and_free_tx_resource(struct igb_ring *tx_ring,
2674 struct igb_buffer *buffer_info)
2676 if (buffer_info->dma) {
2677 if (buffer_info->mapped_as_page)
2678 pci_unmap_page(tx_ring->pdev,
2680 buffer_info->length,
2683 pci_unmap_single(tx_ring->pdev,
2685 buffer_info->length,
2687 buffer_info->dma = 0;
2689 if (buffer_info->skb) {
2690 dev_kfree_skb_any(buffer_info->skb);
2691 buffer_info->skb = NULL;
2693 buffer_info->time_stamp = 0;
2694 buffer_info->length = 0;
2695 buffer_info->next_to_watch = 0;
2696 buffer_info->mapped_as_page = false;
2700 * igb_clean_tx_ring - Free Tx Buffers
2701 * @tx_ring: ring to be cleaned
2703 static void igb_clean_tx_ring(struct igb_ring *tx_ring)
2705 struct igb_buffer *buffer_info;
2709 if (!tx_ring->buffer_info)
2711 /* Free all the Tx ring sk_buffs */
2713 for (i = 0; i < tx_ring->count; i++) {
2714 buffer_info = &tx_ring->buffer_info[i];
2715 igb_unmap_and_free_tx_resource(tx_ring, buffer_info);
2718 size = sizeof(struct igb_buffer) * tx_ring->count;
2719 memset(tx_ring->buffer_info, 0, size);
2721 /* Zero out the descriptor ring */
2722 memset(tx_ring->desc, 0, tx_ring->size);
2724 tx_ring->next_to_use = 0;
2725 tx_ring->next_to_clean = 0;
2729 * igb_clean_all_tx_rings - Free Tx Buffers for all queues
2730 * @adapter: board private structure
2732 static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
2736 for (i = 0; i < adapter->num_tx_queues; i++)
2737 igb_clean_tx_ring(adapter->tx_ring[i]);
2741 * igb_free_rx_resources - Free Rx Resources
2742 * @rx_ring: ring to clean the resources from
2744 * Free all receive software resources
2746 void igb_free_rx_resources(struct igb_ring *rx_ring)
2748 igb_clean_rx_ring(rx_ring);
2750 vfree(rx_ring->buffer_info);
2751 rx_ring->buffer_info = NULL;
2753 /* if not set, then don't free */
2757 pci_free_consistent(rx_ring->pdev, rx_ring->size,
2758 rx_ring->desc, rx_ring->dma);
2760 rx_ring->desc = NULL;
2764 * igb_free_all_rx_resources - Free Rx Resources for All Queues
2765 * @adapter: board private structure
2767 * Free all receive software resources
2769 static void igb_free_all_rx_resources(struct igb_adapter *adapter)
2773 for (i = 0; i < adapter->num_rx_queues; i++)
2774 igb_free_rx_resources(adapter->rx_ring[i]);
2778 * igb_clean_rx_ring - Free Rx Buffers per Queue
2779 * @rx_ring: ring to free buffers from
2781 static void igb_clean_rx_ring(struct igb_ring *rx_ring)
2783 struct igb_buffer *buffer_info;
2787 if (!rx_ring->buffer_info)
2790 /* Free all the Rx ring sk_buffs */
2791 for (i = 0; i < rx_ring->count; i++) {
2792 buffer_info = &rx_ring->buffer_info[i];
2793 if (buffer_info->dma) {
2794 pci_unmap_single(rx_ring->pdev,
2796 rx_ring->rx_buffer_len,
2797 PCI_DMA_FROMDEVICE);
2798 buffer_info->dma = 0;
2801 if (buffer_info->skb) {
2802 dev_kfree_skb(buffer_info->skb);
2803 buffer_info->skb = NULL;
2805 if (buffer_info->page_dma) {
2806 pci_unmap_page(rx_ring->pdev,
2807 buffer_info->page_dma,
2809 PCI_DMA_FROMDEVICE);
2810 buffer_info->page_dma = 0;
2812 if (buffer_info->page) {
2813 put_page(buffer_info->page);
2814 buffer_info->page = NULL;
2815 buffer_info->page_offset = 0;
2819 size = sizeof(struct igb_buffer) * rx_ring->count;
2820 memset(rx_ring->buffer_info, 0, size);
2822 /* Zero out the descriptor ring */
2823 memset(rx_ring->desc, 0, rx_ring->size);
2825 rx_ring->next_to_clean = 0;
2826 rx_ring->next_to_use = 0;
2830 * igb_clean_all_rx_rings - Free Rx Buffers for all queues
2831 * @adapter: board private structure
2833 static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
2837 for (i = 0; i < adapter->num_rx_queues; i++)
2838 igb_clean_rx_ring(adapter->rx_ring[i]);
2842 * igb_set_mac - Change the Ethernet Address of the NIC
2843 * @netdev: network interface device structure
2844 * @p: pointer to an address structure
2846 * Returns 0 on success, negative on failure
2848 static int igb_set_mac(struct net_device *netdev, void *p)
2850 struct igb_adapter *adapter = netdev_priv(netdev);
2851 struct e1000_hw *hw = &adapter->hw;
2852 struct sockaddr *addr = p;
2854 if (!is_valid_ether_addr(addr->sa_data))
2855 return -EADDRNOTAVAIL;
2857 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2858 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
2860 /* set the correct pool for the new PF MAC address in entry 0 */
2861 igb_rar_set_qsel(adapter, hw->mac.addr, 0,
2862 adapter->vfs_allocated_count);
2868 * igb_write_mc_addr_list - write multicast addresses to MTA
2869 * @netdev: network interface device structure
2871 * Writes multicast address list to the MTA hash table.
2872 * Returns: -ENOMEM on failure
2873 * 0 on no addresses written
2874 * X on writing X addresses to MTA
2876 static int igb_write_mc_addr_list(struct net_device *netdev)
2878 struct igb_adapter *adapter = netdev_priv(netdev);
2879 struct e1000_hw *hw = &adapter->hw;
2880 struct dev_mc_list *mc_ptr;
2884 if (netdev_mc_empty(netdev)) {
2885 /* nothing to program, so clear mc list */
2886 igb_update_mc_addr_list(hw, NULL, 0);
2887 igb_restore_vf_multicasts(adapter);
2891 mta_list = kzalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
2895 /* The shared function expects a packed array of only addresses. */
2897 netdev_for_each_mc_addr(mc_ptr, netdev)
2898 memcpy(mta_list + (i++ * ETH_ALEN), mc_ptr->dmi_addr, ETH_ALEN);
2900 igb_update_mc_addr_list(hw, mta_list, i);
2903 return netdev_mc_count(netdev);
2907 * igb_write_uc_addr_list - write unicast addresses to RAR table
2908 * @netdev: network interface device structure
2910 * Writes unicast address list to the RAR table.
2911 * Returns: -ENOMEM on failure/insufficient address space
2912 * 0 on no addresses written
2913 * X on writing X addresses to the RAR table
2915 static int igb_write_uc_addr_list(struct net_device *netdev)
2917 struct igb_adapter *adapter = netdev_priv(netdev);
2918 struct e1000_hw *hw = &adapter->hw;
2919 unsigned int vfn = adapter->vfs_allocated_count;
2920 unsigned int rar_entries = hw->mac.rar_entry_count - (vfn + 1);
2923 /* return ENOMEM indicating insufficient memory for addresses */
2924 if (netdev_uc_count(netdev) > rar_entries)
2927 if (!netdev_uc_empty(netdev) && rar_entries) {
2928 struct netdev_hw_addr *ha;
2930 netdev_for_each_uc_addr(ha, netdev) {
2933 igb_rar_set_qsel(adapter, ha->addr,
2939 /* write the addresses in reverse order to avoid write combining */
2940 for (; rar_entries > 0 ; rar_entries--) {
2941 wr32(E1000_RAH(rar_entries), 0);
2942 wr32(E1000_RAL(rar_entries), 0);
2950 * igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
2951 * @netdev: network interface device structure
2953 * The set_rx_mode entry point is called whenever the unicast or multicast
2954 * address lists or the network interface flags are updated. This routine is
2955 * responsible for configuring the hardware for proper unicast, multicast,
2956 * promiscuous mode, and all-multi behavior.
2958 static void igb_set_rx_mode(struct net_device *netdev)
2960 struct igb_adapter *adapter = netdev_priv(netdev);
2961 struct e1000_hw *hw = &adapter->hw;
2962 unsigned int vfn = adapter->vfs_allocated_count;
2963 u32 rctl, vmolr = 0;
2966 /* Check for Promiscuous and All Multicast modes */
2967 rctl = rd32(E1000_RCTL);
2969 /* clear the effected bits */
2970 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE | E1000_RCTL_VFE);
2972 if (netdev->flags & IFF_PROMISC) {
2973 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2974 vmolr |= (E1000_VMOLR_ROPE | E1000_VMOLR_MPME);
2976 if (netdev->flags & IFF_ALLMULTI) {
2977 rctl |= E1000_RCTL_MPE;
2978 vmolr |= E1000_VMOLR_MPME;
2981 * Write addresses to the MTA, if the attempt fails
2982 * then we should just turn on promiscous mode so
2983 * that we can at least receive multicast traffic
2985 count = igb_write_mc_addr_list(netdev);
2987 rctl |= E1000_RCTL_MPE;
2988 vmolr |= E1000_VMOLR_MPME;
2990 vmolr |= E1000_VMOLR_ROMPE;
2994 * Write addresses to available RAR registers, if there is not
2995 * sufficient space to store all the addresses then enable
2996 * unicast promiscous mode
2998 count = igb_write_uc_addr_list(netdev);
3000 rctl |= E1000_RCTL_UPE;
3001 vmolr |= E1000_VMOLR_ROPE;
3003 rctl |= E1000_RCTL_VFE;
3005 wr32(E1000_RCTL, rctl);
3008 * In order to support SR-IOV and eventually VMDq it is necessary to set
3009 * the VMOLR to enable the appropriate modes. Without this workaround
3010 * we will have issues with VLAN tag stripping not being done for frames
3011 * that are only arriving because we are the default pool
3013 if (hw->mac.type < e1000_82576)
3016 vmolr |= rd32(E1000_VMOLR(vfn)) &
3017 ~(E1000_VMOLR_ROPE | E1000_VMOLR_MPME | E1000_VMOLR_ROMPE);
3018 wr32(E1000_VMOLR(vfn), vmolr);
3019 igb_restore_vf_multicasts(adapter);
3022 /* Need to wait a few seconds after link up to get diagnostic information from
3024 static void igb_update_phy_info(unsigned long data)
3026 struct igb_adapter *adapter = (struct igb_adapter *) data;
3027 igb_get_phy_info(&adapter->hw);
3031 * igb_has_link - check shared code for link and determine up/down
3032 * @adapter: pointer to driver private info
3034 bool igb_has_link(struct igb_adapter *adapter)
3036 struct e1000_hw *hw = &adapter->hw;
3037 bool link_active = false;
3040 /* get_link_status is set on LSC (link status) interrupt or
3041 * rx sequence error interrupt. get_link_status will stay
3042 * false until the e1000_check_for_link establishes link
3043 * for copper adapters ONLY
3045 switch (hw->phy.media_type) {
3046 case e1000_media_type_copper:
3047 if (hw->mac.get_link_status) {
3048 ret_val = hw->mac.ops.check_for_link(hw);
3049 link_active = !hw->mac.get_link_status;
3054 case e1000_media_type_internal_serdes:
3055 ret_val = hw->mac.ops.check_for_link(hw);
3056 link_active = hw->mac.serdes_has_link;
3059 case e1000_media_type_unknown:
3067 * igb_watchdog - Timer Call-back
3068 * @data: pointer to adapter cast into an unsigned long
3070 static void igb_watchdog(unsigned long data)
3072 struct igb_adapter *adapter = (struct igb_adapter *)data;
3073 /* Do the rest outside of interrupt context */
3074 schedule_work(&adapter->watchdog_task);
3077 static void igb_watchdog_task(struct work_struct *work)
3079 struct igb_adapter *adapter = container_of(work,
3082 struct e1000_hw *hw = &adapter->hw;
3083 struct net_device *netdev = adapter->netdev;
3087 link = igb_has_link(adapter);
3089 if (!netif_carrier_ok(netdev)) {
3091 hw->mac.ops.get_speed_and_duplex(hw,
3092 &adapter->link_speed,
3093 &adapter->link_duplex);
3095 ctrl = rd32(E1000_CTRL);
3096 /* Links status message must follow this format */
3097 printk(KERN_INFO "igb: %s NIC Link is Up %d Mbps %s, "
3098 "Flow Control: %s\n",
3100 adapter->link_speed,
3101 adapter->link_duplex == FULL_DUPLEX ?
3102 "Full Duplex" : "Half Duplex",
3103 ((ctrl & E1000_CTRL_TFCE) &&
3104 (ctrl & E1000_CTRL_RFCE)) ? "RX/TX" :
3105 ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3106 ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None")));
3108 /* tweak tx_queue_len according to speed/duplex and
3109 * adjust the timeout factor */
3110 netdev->tx_queue_len = adapter->tx_queue_len;
3111 adapter->tx_timeout_factor = 1;
3112 switch (adapter->link_speed) {
3114 netdev->tx_queue_len = 10;
3115 adapter->tx_timeout_factor = 14;
3118 netdev->tx_queue_len = 100;
3119 /* maybe add some timeout factor ? */
3123 netif_carrier_on(netdev);
3125 igb_ping_all_vfs(adapter);
3127 /* link state has changed, schedule phy info update */
3128 if (!test_bit(__IGB_DOWN, &adapter->state))
3129 mod_timer(&adapter->phy_info_timer,
3130 round_jiffies(jiffies + 2 * HZ));
3133 if (netif_carrier_ok(netdev)) {
3134 adapter->link_speed = 0;
3135 adapter->link_duplex = 0;
3136 /* Links status message must follow this format */
3137 printk(KERN_INFO "igb: %s NIC Link is Down\n",
3139 netif_carrier_off(netdev);
3141 igb_ping_all_vfs(adapter);
3143 /* link state has changed, schedule phy info update */
3144 if (!test_bit(__IGB_DOWN, &adapter->state))
3145 mod_timer(&adapter->phy_info_timer,
3146 round_jiffies(jiffies + 2 * HZ));
3150 igb_update_stats(adapter);
3152 for (i = 0; i < adapter->num_tx_queues; i++) {
3153 struct igb_ring *tx_ring = adapter->tx_ring[i];
3154 if (!netif_carrier_ok(netdev)) {
3155 /* We've lost link, so the controller stops DMA,
3156 * but we've got queued Tx work that's never going
3157 * to get done, so reset controller to flush Tx.
3158 * (Do the reset outside of interrupt context). */
3159 if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) {
3160 adapter->tx_timeout_count++;
3161 schedule_work(&adapter->reset_task);
3162 /* return immediately since reset is imminent */
3167 /* Force detection of hung controller every watchdog period */
3168 tx_ring->detect_tx_hung = true;
3171 /* Cause software interrupt to ensure rx ring is cleaned */
3172 if (adapter->msix_entries) {
3174 for (i = 0; i < adapter->num_q_vectors; i++) {
3175 struct igb_q_vector *q_vector = adapter->q_vector[i];
3176 eics |= q_vector->eims_value;
3178 wr32(E1000_EICS, eics);
3180 wr32(E1000_ICS, E1000_ICS_RXDMT0);
3183 /* Reset the timer */
3184 if (!test_bit(__IGB_DOWN, &adapter->state))
3185 mod_timer(&adapter->watchdog_timer,
3186 round_jiffies(jiffies + 2 * HZ));
3189 enum latency_range {
3193 latency_invalid = 255
3197 * igb_update_ring_itr - update the dynamic ITR value based on packet size
3199 * Stores a new ITR value based on strictly on packet size. This
3200 * algorithm is less sophisticated than that used in igb_update_itr,
3201 * due to the difficulty of synchronizing statistics across multiple
3202 * receive rings. The divisors and thresholds used by this fuction
3203 * were determined based on theoretical maximum wire speed and testing
3204 * data, in order to minimize response time while increasing bulk
3206 * This functionality is controlled by the InterruptThrottleRate module
3207 * parameter (see igb_param.c)
3208 * NOTE: This function is called only when operating in a multiqueue
3209 * receive environment.
3210 * @q_vector: pointer to q_vector
3212 static void igb_update_ring_itr(struct igb_q_vector *q_vector)
3214 int new_val = q_vector->itr_val;
3215 int avg_wire_size = 0;
3216 struct igb_adapter *adapter = q_vector->adapter;
3218 /* For non-gigabit speeds, just fix the interrupt rate at 4000
3219 * ints/sec - ITR timer value of 120 ticks.
3221 if (adapter->link_speed != SPEED_1000) {
3226 if (q_vector->rx_ring && q_vector->rx_ring->total_packets) {
3227 struct igb_ring *ring = q_vector->rx_ring;
3228 avg_wire_size = ring->total_bytes / ring->total_packets;
3231 if (q_vector->tx_ring && q_vector->tx_ring->total_packets) {
3232 struct igb_ring *ring = q_vector->tx_ring;
3233 avg_wire_size = max_t(u32, avg_wire_size,
3234 (ring->total_bytes /
3235 ring->total_packets));
3238 /* if avg_wire_size isn't set no work was done */
3242 /* Add 24 bytes to size to account for CRC, preamble, and gap */
3243 avg_wire_size += 24;
3245 /* Don't starve jumbo frames */
3246 avg_wire_size = min(avg_wire_size, 3000);
3248 /* Give a little boost to mid-size frames */
3249 if ((avg_wire_size > 300) && (avg_wire_size < 1200))
3250 new_val = avg_wire_size / 3;
3252 new_val = avg_wire_size / 2;
3254 /* when in itr mode 3 do not exceed 20K ints/sec */
3255 if (adapter->rx_itr_setting == 3 && new_val < 196)
3259 if (new_val != q_vector->itr_val) {
3260 q_vector->itr_val = new_val;
3261 q_vector->set_itr = 1;
3264 if (q_vector->rx_ring) {
3265 q_vector->rx_ring->total_bytes = 0;
3266 q_vector->rx_ring->total_packets = 0;
3268 if (q_vector->tx_ring) {
3269 q_vector->tx_ring->total_bytes = 0;
3270 q_vector->tx_ring->total_packets = 0;
3275 * igb_update_itr - update the dynamic ITR value based on statistics
3276 * Stores a new ITR value based on packets and byte
3277 * counts during the last interrupt. The advantage of per interrupt
3278 * computation is faster updates and more accurate ITR for the current
3279 * traffic pattern. Constants in this function were computed
3280 * based on theoretical maximum wire speed and thresholds were set based
3281 * on testing data as well as attempting to minimize response time
3282 * while increasing bulk throughput.
3283 * this functionality is controlled by the InterruptThrottleRate module
3284 * parameter (see igb_param.c)
3285 * NOTE: These calculations are only valid when operating in a single-
3286 * queue environment.
3287 * @adapter: pointer to adapter
3288 * @itr_setting: current q_vector->itr_val
3289 * @packets: the number of packets during this measurement interval
3290 * @bytes: the number of bytes during this measurement interval
3292 static unsigned int igb_update_itr(struct igb_adapter *adapter, u16 itr_setting,
3293 int packets, int bytes)
3295 unsigned int retval = itr_setting;
3298 goto update_itr_done;
3300 switch (itr_setting) {
3301 case lowest_latency:
3302 /* handle TSO and jumbo frames */
3303 if (bytes/packets > 8000)
3304 retval = bulk_latency;
3305 else if ((packets < 5) && (bytes > 512))
3306 retval = low_latency;
3308 case low_latency: /* 50 usec aka 20000 ints/s */
3309 if (bytes > 10000) {
3310 /* this if handles the TSO accounting */
3311 if (bytes/packets > 8000) {
3312 retval = bulk_latency;
3313 } else if ((packets < 10) || ((bytes/packets) > 1200)) {
3314 retval = bulk_latency;
3315 } else if ((packets > 35)) {
3316 retval = lowest_latency;
3318 } else if (bytes/packets > 2000) {
3319 retval = bulk_latency;
3320 } else if (packets <= 2 && bytes < 512) {
3321 retval = lowest_latency;
3324 case bulk_latency: /* 250 usec aka 4000 ints/s */
3325 if (bytes > 25000) {
3327 retval = low_latency;
3328 } else if (bytes < 1500) {
3329 retval = low_latency;
3338 static void igb_set_itr(struct igb_adapter *adapter)
3340 struct igb_q_vector *q_vector = adapter->q_vector[0];
3342 u32 new_itr = q_vector->itr_val;
3344 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
3345 if (adapter->link_speed != SPEED_1000) {
3351 adapter->rx_itr = igb_update_itr(adapter,
3353 q_vector->rx_ring->total_packets,
3354 q_vector->rx_ring->total_bytes);
3356 adapter->tx_itr = igb_update_itr(adapter,
3358 q_vector->tx_ring->total_packets,
3359 q_vector->tx_ring->total_bytes);
3360 current_itr = max(adapter->rx_itr, adapter->tx_itr);
3362 /* conservative mode (itr 3) eliminates the lowest_latency setting */
3363 if (adapter->rx_itr_setting == 3 && current_itr == lowest_latency)
3364 current_itr = low_latency;
3366 switch (current_itr) {
3367 /* counts and packets in update_itr are dependent on these numbers */
3368 case lowest_latency:
3369 new_itr = 56; /* aka 70,000 ints/sec */
3372 new_itr = 196; /* aka 20,000 ints/sec */
3375 new_itr = 980; /* aka 4,000 ints/sec */
3382 q_vector->rx_ring->total_bytes = 0;
3383 q_vector->rx_ring->total_packets = 0;
3384 q_vector->tx_ring->total_bytes = 0;
3385 q_vector->tx_ring->total_packets = 0;
3387 if (new_itr != q_vector->itr_val) {
3388 /* this attempts to bias the interrupt rate towards Bulk
3389 * by adding intermediate steps when interrupt rate is
3391 new_itr = new_itr > q_vector->itr_val ?
3392 max((new_itr * q_vector->itr_val) /
3393 (new_itr + (q_vector->itr_val >> 2)),
3396 /* Don't write the value here; it resets the adapter's
3397 * internal timer, and causes us to delay far longer than
3398 * we should between interrupts. Instead, we write the ITR
3399 * value at the beginning of the next interrupt so the timing
3400 * ends up being correct.
3402 q_vector->itr_val = new_itr;
3403 q_vector->set_itr = 1;
3409 #define IGB_TX_FLAGS_CSUM 0x00000001
3410 #define IGB_TX_FLAGS_VLAN 0x00000002
3411 #define IGB_TX_FLAGS_TSO 0x00000004
3412 #define IGB_TX_FLAGS_IPV4 0x00000008
3413 #define IGB_TX_FLAGS_TSTAMP 0x00000010
3414 #define IGB_TX_FLAGS_VLAN_MASK 0xffff0000
3415 #define IGB_TX_FLAGS_VLAN_SHIFT 16
3417 static inline int igb_tso_adv(struct igb_ring *tx_ring,
3418 struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
3420 struct e1000_adv_tx_context_desc *context_desc;
3423 struct igb_buffer *buffer_info;
3424 u32 info = 0, tu_cmd = 0;
3428 if (skb_header_cloned(skb)) {
3429 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3434 l4len = tcp_hdrlen(skb);
3437 if (skb->protocol == htons(ETH_P_IP)) {
3438 struct iphdr *iph = ip_hdr(skb);
3441 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3445 } else if (skb_is_gso_v6(skb)) {
3446 ipv6_hdr(skb)->payload_len = 0;
3447 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3448 &ipv6_hdr(skb)->daddr,
3452 i = tx_ring->next_to_use;
3454 buffer_info = &tx_ring->buffer_info[i];
3455 context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
3456 /* VLAN MACLEN IPLEN */
3457 if (tx_flags & IGB_TX_FLAGS_VLAN)
3458 info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
3459 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
3460 *hdr_len += skb_network_offset(skb);
3461 info |= skb_network_header_len(skb);
3462 *hdr_len += skb_network_header_len(skb);
3463 context_desc->vlan_macip_lens = cpu_to_le32(info);
3465 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3466 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
3468 if (skb->protocol == htons(ETH_P_IP))
3469 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
3470 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
3472 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
3475 mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
3476 mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
3478 /* For 82575, context index must be unique per ring. */
3479 if (tx_ring->flags & IGB_RING_FLAG_TX_CTX_IDX)
3480 mss_l4len_idx |= tx_ring->reg_idx << 4;
3482 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
3483 context_desc->seqnum_seed = 0;
3485 buffer_info->time_stamp = jiffies;
3486 buffer_info->next_to_watch = i;
3487 buffer_info->dma = 0;
3489 if (i == tx_ring->count)
3492 tx_ring->next_to_use = i;
3497 static inline bool igb_tx_csum_adv(struct igb_ring *tx_ring,
3498 struct sk_buff *skb, u32 tx_flags)
3500 struct e1000_adv_tx_context_desc *context_desc;
3501 struct pci_dev *pdev = tx_ring->pdev;
3502 struct igb_buffer *buffer_info;
3503 u32 info = 0, tu_cmd = 0;
3506 if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
3507 (tx_flags & IGB_TX_FLAGS_VLAN)) {
3508 i = tx_ring->next_to_use;
3509 buffer_info = &tx_ring->buffer_info[i];
3510 context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
3512 if (tx_flags & IGB_TX_FLAGS_VLAN)
3513 info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
3515 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
3516 if (skb->ip_summed == CHECKSUM_PARTIAL)
3517 info |= skb_network_header_len(skb);
3519 context_desc->vlan_macip_lens = cpu_to_le32(info);
3521 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
3523 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3526 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3527 const struct vlan_ethhdr *vhdr =
3528 (const struct vlan_ethhdr*)skb->data;
3530 protocol = vhdr->h_vlan_encapsulated_proto;
3532 protocol = skb->protocol;
3536 case cpu_to_be16(ETH_P_IP):
3537 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
3538 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
3539 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
3540 else if (ip_hdr(skb)->protocol == IPPROTO_SCTP)
3541 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_SCTP;
3543 case cpu_to_be16(ETH_P_IPV6):
3544 /* XXX what about other V6 headers?? */
3545 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3546 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
3547 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_SCTP)
3548 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_SCTP;
3551 if (unlikely(net_ratelimit()))
3552 dev_warn(&pdev->dev,
3553 "partial checksum but proto=%x!\n",
3559 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
3560 context_desc->seqnum_seed = 0;
3561 if (tx_ring->flags & IGB_RING_FLAG_TX_CTX_IDX)
3562 context_desc->mss_l4len_idx =
3563 cpu_to_le32(tx_ring->reg_idx << 4);
3565 buffer_info->time_stamp = jiffies;
3566 buffer_info->next_to_watch = i;
3567 buffer_info->dma = 0;
3570 if (i == tx_ring->count)
3572 tx_ring->next_to_use = i;
3579 #define IGB_MAX_TXD_PWR 16
3580 #define IGB_MAX_DATA_PER_TXD (1<<IGB_MAX_TXD_PWR)
3582 static inline int igb_tx_map_adv(struct igb_ring *tx_ring, struct sk_buff *skb,
3585 struct igb_buffer *buffer_info;
3586 struct pci_dev *pdev = tx_ring->pdev;
3587 unsigned int len = skb_headlen(skb);
3588 unsigned int count = 0, i;
3591 i = tx_ring->next_to_use;
3593 buffer_info = &tx_ring->buffer_info[i];
3594 BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
3595 buffer_info->length = len;
3596 /* set time_stamp *before* dma to help avoid a possible race */
3597 buffer_info->time_stamp = jiffies;
3598 buffer_info->next_to_watch = i;
3599 buffer_info->dma = pci_map_single(pdev, skb->data, len,
3601 if (pci_dma_mapping_error(pdev, buffer_info->dma))
3604 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
3605 struct skb_frag_struct *frag;
3609 if (i == tx_ring->count)
3612 frag = &skb_shinfo(skb)->frags[f];
3615 buffer_info = &tx_ring->buffer_info[i];
3616 BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
3617 buffer_info->length = len;
3618 buffer_info->time_stamp = jiffies;
3619 buffer_info->next_to_watch = i;
3620 buffer_info->mapped_as_page = true;
3621 buffer_info->dma = pci_map_page(pdev,
3626 if (pci_dma_mapping_error(pdev, buffer_info->dma))
3631 tx_ring->buffer_info[i].skb = skb;
3632 tx_ring->buffer_info[i].gso_segs = skb_shinfo(skb)->gso_segs ?: 1;
3633 tx_ring->buffer_info[first].next_to_watch = i;
3638 dev_err(&pdev->dev, "TX DMA map failed\n");
3640 /* clear timestamp and dma mappings for failed buffer_info mapping */
3641 buffer_info->dma = 0;
3642 buffer_info->time_stamp = 0;
3643 buffer_info->length = 0;
3644 buffer_info->next_to_watch = 0;
3645 buffer_info->mapped_as_page = false;
3647 /* clear timestamp and dma mappings for remaining portion of packet */
3652 buffer_info = &tx_ring->buffer_info[i];
3653 igb_unmap_and_free_tx_resource(tx_ring, buffer_info);
3659 static inline void igb_tx_queue_adv(struct igb_ring *tx_ring,
3660 u32 tx_flags, int count, u32 paylen,
3663 union e1000_adv_tx_desc *tx_desc;
3664 struct igb_buffer *buffer_info;
3665 u32 olinfo_status = 0, cmd_type_len;
3666 unsigned int i = tx_ring->next_to_use;
3668 cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
3669 E1000_ADVTXD_DCMD_DEXT);
3671 if (tx_flags & IGB_TX_FLAGS_VLAN)
3672 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
3674 if (tx_flags & IGB_TX_FLAGS_TSTAMP)
3675 cmd_type_len |= E1000_ADVTXD_MAC_TSTAMP;
3677 if (tx_flags & IGB_TX_FLAGS_TSO) {
3678 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
3680 /* insert tcp checksum */
3681 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
3683 /* insert ip checksum */
3684 if (tx_flags & IGB_TX_FLAGS_IPV4)
3685 olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
3687 } else if (tx_flags & IGB_TX_FLAGS_CSUM) {
3688 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
3691 if ((tx_ring->flags & IGB_RING_FLAG_TX_CTX_IDX) &&
3692 (tx_flags & (IGB_TX_FLAGS_CSUM |
3694 IGB_TX_FLAGS_VLAN)))
3695 olinfo_status |= tx_ring->reg_idx << 4;
3697 olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
3700 buffer_info = &tx_ring->buffer_info[i];
3701 tx_desc = E1000_TX_DESC_ADV(*tx_ring, i);
3702 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
3703 tx_desc->read.cmd_type_len =
3704 cpu_to_le32(cmd_type_len | buffer_info->length);
3705 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
3708 if (i == tx_ring->count)
3710 } while (count > 0);
3712 tx_desc->read.cmd_type_len |= cpu_to_le32(IGB_ADVTXD_DCMD);
3713 /* Force memory writes to complete before letting h/w
3714 * know there are new descriptors to fetch. (Only
3715 * applicable for weak-ordered memory model archs,
3716 * such as IA-64). */
3719 tx_ring->next_to_use = i;
3720 writel(i, tx_ring->tail);
3721 /* we need this if more than one processor can write to our tail
3722 * at a time, it syncronizes IO on IA64/Altix systems */
3726 static int __igb_maybe_stop_tx(struct igb_ring *tx_ring, int size)
3728 struct net_device *netdev = tx_ring->netdev;
3730 netif_stop_subqueue(netdev, tx_ring->queue_index);
3732 /* Herbert's original patch had:
3733 * smp_mb__after_netif_stop_queue();
3734 * but since that doesn't exist yet, just open code it. */
3737 /* We need to check again in a case another CPU has just
3738 * made room available. */
3739 if (igb_desc_unused(tx_ring) < size)
3743 netif_wake_subqueue(netdev, tx_ring->queue_index);
3744 tx_ring->tx_stats.restart_queue++;
3748 static inline int igb_maybe_stop_tx(struct igb_ring *tx_ring, int size)
3750 if (igb_desc_unused(tx_ring) >= size)
3752 return __igb_maybe_stop_tx(tx_ring, size);
3755 netdev_tx_t igb_xmit_frame_ring_adv(struct sk_buff *skb,
3756 struct igb_ring *tx_ring)
3758 struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
3763 union skb_shared_tx *shtx = skb_tx(skb);
3765 /* need: 1 descriptor per page,
3766 * + 2 desc gap to keep tail from touching head,
3767 * + 1 desc for skb->data,
3768 * + 1 desc for context descriptor,
3769 * otherwise try next time */
3770 if (igb_maybe_stop_tx(tx_ring, skb_shinfo(skb)->nr_frags + 4)) {
3771 /* this is a hard error */
3772 return NETDEV_TX_BUSY;
3775 if (unlikely(shtx->hardware)) {
3776 shtx->in_progress = 1;
3777 tx_flags |= IGB_TX_FLAGS_TSTAMP;
3780 if (vlan_tx_tag_present(skb) && adapter->vlgrp) {
3781 tx_flags |= IGB_TX_FLAGS_VLAN;
3782 tx_flags |= (vlan_tx_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
3785 if (skb->protocol == htons(ETH_P_IP))
3786 tx_flags |= IGB_TX_FLAGS_IPV4;
3788 first = tx_ring->next_to_use;
3789 if (skb_is_gso(skb)) {
3790 tso = igb_tso_adv(tx_ring, skb, tx_flags, &hdr_len);
3793 dev_kfree_skb_any(skb);
3794 return NETDEV_TX_OK;
3799 tx_flags |= IGB_TX_FLAGS_TSO;
3800 else if (igb_tx_csum_adv(tx_ring, skb, tx_flags) &&
3801 (skb->ip_summed == CHECKSUM_PARTIAL))
3802 tx_flags |= IGB_TX_FLAGS_CSUM;
3805 * count reflects descriptors mapped, if 0 or less then mapping error
3806 * has occured and we need to rewind the descriptor queue
3808 count = igb_tx_map_adv(tx_ring, skb, first);
3810 dev_kfree_skb_any(skb);
3811 tx_ring->buffer_info[first].time_stamp = 0;
3812 tx_ring->next_to_use = first;
3813 return NETDEV_TX_OK;
3816 igb_tx_queue_adv(tx_ring, tx_flags, count, skb->len, hdr_len);
3818 /* Make sure there is space in the ring for the next send. */
3819 igb_maybe_stop_tx(tx_ring, MAX_SKB_FRAGS + 4);
3821 return NETDEV_TX_OK;
3824 static netdev_tx_t igb_xmit_frame_adv(struct sk_buff *skb,
3825 struct net_device *netdev)
3827 struct igb_adapter *adapter = netdev_priv(netdev);
3828 struct igb_ring *tx_ring;
3831 if (test_bit(__IGB_DOWN, &adapter->state)) {
3832 dev_kfree_skb_any(skb);
3833 return NETDEV_TX_OK;
3836 if (skb->len <= 0) {
3837 dev_kfree_skb_any(skb);
3838 return NETDEV_TX_OK;
3841 r_idx = skb->queue_mapping & (IGB_ABS_MAX_TX_QUEUES - 1);
3842 tx_ring = adapter->multi_tx_table[r_idx];
3844 /* This goes back to the question of how to logically map a tx queue
3845 * to a flow. Right now, performance is impacted slightly negatively
3846 * if using multiple tx queues. If the stack breaks away from a
3847 * single qdisc implementation, we can look at this again. */
3848 return igb_xmit_frame_ring_adv(skb, tx_ring);
3852 * igb_tx_timeout - Respond to a Tx Hang
3853 * @netdev: network interface device structure
3855 static void igb_tx_timeout(struct net_device *netdev)
3857 struct igb_adapter *adapter = netdev_priv(netdev);
3858 struct e1000_hw *hw = &adapter->hw;
3860 /* Do the reset outside of interrupt context */
3861 adapter->tx_timeout_count++;
3863 if (hw->mac.type == e1000_82580)
3864 hw->dev_spec._82575.global_device_reset = true;
3866 schedule_work(&adapter->reset_task);
3868 (adapter->eims_enable_mask & ~adapter->eims_other));
3871 static void igb_reset_task(struct work_struct *work)
3873 struct igb_adapter *adapter;
3874 adapter = container_of(work, struct igb_adapter, reset_task);
3876 igb_reinit_locked(adapter);
3880 * igb_get_stats - Get System Network Statistics
3881 * @netdev: network interface device structure
3883 * Returns the address of the device statistics structure.
3884 * The statistics are actually updated from the timer callback.
3886 static struct net_device_stats *igb_get_stats(struct net_device *netdev)
3888 /* only return the current stats */
3889 return &netdev->stats;
3893 * igb_change_mtu - Change the Maximum Transfer Unit
3894 * @netdev: network interface device structure
3895 * @new_mtu: new value for maximum frame size
3897 * Returns 0 on success, negative on failure
3899 static int igb_change_mtu(struct net_device *netdev, int new_mtu)
3901 struct igb_adapter *adapter = netdev_priv(netdev);
3902 struct pci_dev *pdev = adapter->pdev;
3903 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3904 u32 rx_buffer_len, i;
3906 if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3907 dev_err(&pdev->dev, "Invalid MTU setting\n");
3911 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
3912 dev_err(&pdev->dev, "MTU > 9216 not supported.\n");
3916 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
3919 /* igb_down has a dependency on max_frame_size */
3920 adapter->max_frame_size = max_frame;
3922 /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3923 * means we reserve 2 more, this pushes us to allocate from the next
3925 * i.e. RXBUFFER_2048 --> size-4096 slab
3928 if (max_frame <= IGB_RXBUFFER_1024)
3929 rx_buffer_len = IGB_RXBUFFER_1024;
3930 else if (max_frame <= MAXIMUM_ETHERNET_VLAN_SIZE)
3931 rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3933 rx_buffer_len = IGB_RXBUFFER_128;
3935 if (netif_running(netdev))
3938 dev_info(&pdev->dev, "changing MTU from %d to %d\n",
3939 netdev->mtu, new_mtu);
3940 netdev->mtu = new_mtu;
3942 for (i = 0; i < adapter->num_rx_queues; i++)
3943 adapter->rx_ring[i]->rx_buffer_len = rx_buffer_len;
3945 if (netif_running(netdev))
3950 clear_bit(__IGB_RESETTING, &adapter->state);
3956 * igb_update_stats - Update the board statistics counters
3957 * @adapter: board private structure
3960 void igb_update_stats(struct igb_adapter *adapter)
3962 struct net_device_stats *net_stats = igb_get_stats(adapter->netdev);
3963 struct e1000_hw *hw = &adapter->hw;
3964 struct pci_dev *pdev = adapter->pdev;
3970 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3973 * Prevent stats update while adapter is being reset, or if the pci
3974 * connection is down.
3976 if (adapter->link_speed == 0)
3978 if (pci_channel_offline(pdev))
3983 for (i = 0; i < adapter->num_rx_queues; i++) {
3984 u32 rqdpc_tmp = rd32(E1000_RQDPC(i)) & 0x0FFF;
3985 struct igb_ring *ring = adapter->rx_ring[i];
3986 ring->rx_stats.drops += rqdpc_tmp;
3987 net_stats->rx_fifo_errors += rqdpc_tmp;
3988 bytes += ring->rx_stats.bytes;
3989 packets += ring->rx_stats.packets;
3992 net_stats->rx_bytes = bytes;
3993 net_stats->rx_packets = packets;
3997 for (i = 0; i < adapter->num_tx_queues; i++) {
3998 struct igb_ring *ring = adapter->tx_ring[i];
3999 bytes += ring->tx_stats.bytes;
4000 packets += ring->tx_stats.packets;
4002 net_stats->tx_bytes = bytes;
4003 net_stats->tx_packets = packets;
4005 /* read stats registers */
4006 adapter->stats.crcerrs += rd32(E1000_CRCERRS);
4007 adapter->stats.gprc += rd32(E1000_GPRC);
4008 adapter->stats.gorc += rd32(E1000_GORCL);
4009 rd32(E1000_GORCH); /* clear GORCL */
4010 adapter->stats.bprc += rd32(E1000_BPRC);
4011 adapter->stats.mprc += rd32(E1000_MPRC);
4012 adapter->stats.roc += rd32(E1000_ROC);
4014 adapter->stats.prc64 += rd32(E1000_PRC64);
4015 adapter->stats.prc127 += rd32(E1000_PRC127);
4016 adapter->stats.prc255 += rd32(E1000_PRC255);
4017 adapter->stats.prc511 += rd32(E1000_PRC511);
4018 adapter->stats.prc1023 += rd32(E1000_PRC1023);
4019 adapter->stats.prc1522 += rd32(E1000_PRC1522);
4020 adapter->stats.symerrs += rd32(E1000_SYMERRS);
4021 adapter->stats.sec += rd32(E1000_SEC);
4023 adapter->stats.mpc += rd32(E1000_MPC);
4024 adapter->stats.scc += rd32(E1000_SCC);
4025 adapter->stats.ecol += rd32(E1000_ECOL);
4026 adapter->stats.mcc += rd32(E1000_MCC);
4027 adapter->stats.latecol += rd32(E1000_LATECOL);
4028 adapter->stats.dc += rd32(E1000_DC);
4029 adapter->stats.rlec += rd32(E1000_RLEC);
4030 adapter->stats.xonrxc += rd32(E1000_XONRXC);
4031 adapter->stats.xontxc += rd32(E1000_XONTXC);
4032 adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
4033 adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
4034 adapter->stats.fcruc += rd32(E1000_FCRUC);
4035 adapter->stats.gptc += rd32(E1000_GPTC);
4036 adapter->stats.gotc += rd32(E1000_GOTCL);
4037 rd32(E1000_GOTCH); /* clear GOTCL */
4038 rnbc = rd32(E1000_RNBC);
4039 adapter->stats.rnbc += rnbc;
4040 net_stats->rx_fifo_errors += rnbc;
4041 adapter->stats.ruc += rd32(E1000_RUC);
4042 adapter->stats.rfc += rd32(E1000_RFC);
4043 adapter->stats.rjc += rd32(E1000_RJC);
4044 adapter->stats.tor += rd32(E1000_TORH);
4045 adapter->stats.tot += rd32(E1000_TOTH);
4046 adapter->stats.tpr += rd32(E1000_TPR);
4048 adapter->stats.ptc64 += rd32(E1000_PTC64);
4049 adapter->stats.ptc127 += rd32(E1000_PTC127);
4050 adapter->stats.ptc255 += rd32(E1000_PTC255);
4051 adapter->stats.ptc511 += rd32(E1000_PTC511);
4052 adapter->stats.ptc1023 += rd32(E1000_PTC1023);
4053 adapter->stats.ptc1522 += rd32(E1000_PTC1522);
4055 adapter->stats.mptc += rd32(E1000_MPTC);
4056 adapter->stats.bptc += rd32(E1000_BPTC);
4058 adapter->stats.tpt += rd32(E1000_TPT);
4059 adapter->stats.colc += rd32(E1000_COLC);
4061 adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
4062 /* read internal phy specific stats */
4063 reg = rd32(E1000_CTRL_EXT);
4064 if (!(reg & E1000_CTRL_EXT_LINK_MODE_MASK)) {
4065 adapter->stats.rxerrc += rd32(E1000_RXERRC);
4066 adapter->stats.tncrs += rd32(E1000_TNCRS);
4069 adapter->stats.tsctc += rd32(E1000_TSCTC);
4070 adapter->stats.tsctfc += rd32(E1000_TSCTFC);
4072 adapter->stats.iac += rd32(E1000_IAC);
4073 adapter->stats.icrxoc += rd32(E1000_ICRXOC);
4074 adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
4075 adapter->stats.icrxatc += rd32(E1000_ICRXATC);
4076 adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
4077 adapter->stats.ictxatc += rd32(E1000_ICTXATC);
4078 adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
4079 adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
4080 adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
4082 /* Fill out the OS statistics structure */
4083 net_stats->multicast = adapter->stats.mprc;
4084 net_stats->collisions = adapter->stats.colc;
4088 /* RLEC on some newer hardware can be incorrect so build
4089 * our own version based on RUC and ROC */
4090 net_stats->rx_errors = adapter->stats.rxerrc +
4091 adapter->stats.crcerrs + adapter->stats.algnerrc +
4092 adapter->stats.ruc + adapter->stats.roc +
4093 adapter->stats.cexterr;
4094 net_stats->rx_length_errors = adapter->stats.ruc +
4096 net_stats->rx_crc_errors = adapter->stats.crcerrs;
4097 net_stats->rx_frame_errors = adapter->stats.algnerrc;
4098 net_stats->rx_missed_errors = adapter->stats.mpc;
4101 net_stats->tx_errors = adapter->stats.ecol +
4102 adapter->stats.latecol;
4103 net_stats->tx_aborted_errors = adapter->stats.ecol;
4104 net_stats->tx_window_errors = adapter->stats.latecol;
4105 net_stats->tx_carrier_errors = adapter->stats.tncrs;
4107 /* Tx Dropped needs to be maintained elsewhere */
4110 if (hw->phy.media_type == e1000_media_type_copper) {
4111 if ((adapter->link_speed == SPEED_1000) &&
4112 (!igb_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
4113 phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
4114 adapter->phy_stats.idle_errors += phy_tmp;
4118 /* Management Stats */
4119 adapter->stats.mgptc += rd32(E1000_MGTPTC);
4120 adapter->stats.mgprc += rd32(E1000_MGTPRC);
4121 adapter->stats.mgpdc += rd32(E1000_MGTPDC);
4124 static irqreturn_t igb_msix_other(int irq, void *data)
4126 struct igb_adapter *adapter = data;
4127 struct e1000_hw *hw = &adapter->hw;
4128 u32 icr = rd32(E1000_ICR);
4129 /* reading ICR causes bit 31 of EICR to be cleared */
4131 if (icr & E1000_ICR_DRSTA)
4132 schedule_work(&adapter->reset_task);
4134 if (icr & E1000_ICR_DOUTSYNC) {
4135 /* HW is reporting DMA is out of sync */
4136 adapter->stats.doosync++;
4139 /* Check for a mailbox event */
4140 if (icr & E1000_ICR_VMMB)
4141 igb_msg_task(adapter);
4143 if (icr & E1000_ICR_LSC) {
4144 hw->mac.get_link_status = 1;
4145 /* guard against interrupt when we're going down */
4146 if (!test_bit(__IGB_DOWN, &adapter->state))
4147 mod_timer(&adapter->watchdog_timer, jiffies + 1);
4150 if (adapter->vfs_allocated_count)
4151 wr32(E1000_IMS, E1000_IMS_LSC |
4153 E1000_IMS_DOUTSYNC);
4155 wr32(E1000_IMS, E1000_IMS_LSC | E1000_IMS_DOUTSYNC);
4156 wr32(E1000_EIMS, adapter->eims_other);
4161 static void igb_write_itr(struct igb_q_vector *q_vector)
4163 struct igb_adapter *adapter = q_vector->adapter;
4164 u32 itr_val = q_vector->itr_val & 0x7FFC;
4166 if (!q_vector->set_itr)
4172 if (adapter->hw.mac.type == e1000_82575)
4173 itr_val |= itr_val << 16;
4175 itr_val |= 0x8000000;
4177 writel(itr_val, q_vector->itr_register);
4178 q_vector->set_itr = 0;
4181 static irqreturn_t igb_msix_ring(int irq, void *data)
4183 struct igb_q_vector *q_vector = data;
4185 /* Write the ITR value calculated from the previous interrupt. */
4186 igb_write_itr(q_vector);
4188 napi_schedule(&q_vector->napi);
4193 #ifdef CONFIG_IGB_DCA
4194 static void igb_update_dca(struct igb_q_vector *q_vector)
4196 struct igb_adapter *adapter = q_vector->adapter;
4197 struct e1000_hw *hw = &adapter->hw;
4198 int cpu = get_cpu();
4200 if (q_vector->cpu == cpu)
4203 if (q_vector->tx_ring) {
4204 int q = q_vector->tx_ring->reg_idx;
4205 u32 dca_txctrl = rd32(E1000_DCA_TXCTRL(q));
4206 if (hw->mac.type == e1000_82575) {
4207 dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK;
4208 dca_txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
4210 dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK_82576;
4211 dca_txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu) <<
4212 E1000_DCA_TXCTRL_CPUID_SHIFT;
4214 dca_txctrl |= E1000_DCA_TXCTRL_DESC_DCA_EN;
4215 wr32(E1000_DCA_TXCTRL(q), dca_txctrl);
4217 if (q_vector->rx_ring) {
4218 int q = q_vector->rx_ring->reg_idx;
4219 u32 dca_rxctrl = rd32(E1000_DCA_RXCTRL(q));
4220 if (hw->mac.type == e1000_82575) {
4221 dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK;
4222 dca_rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
4224 dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK_82576;
4225 dca_rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu) <<
4226 E1000_DCA_RXCTRL_CPUID_SHIFT;
4228 dca_rxctrl |= E1000_DCA_RXCTRL_DESC_DCA_EN;
4229 dca_rxctrl |= E1000_DCA_RXCTRL_HEAD_DCA_EN;
4230 dca_rxctrl |= E1000_DCA_RXCTRL_DATA_DCA_EN;
4231 wr32(E1000_DCA_RXCTRL(q), dca_rxctrl);
4233 q_vector->cpu = cpu;
4238 static void igb_setup_dca(struct igb_adapter *adapter)
4240 struct e1000_hw *hw = &adapter->hw;
4243 if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
4246 /* Always use CB2 mode, difference is masked in the CB driver. */
4247 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
4249 for (i = 0; i < adapter->num_q_vectors; i++) {
4250 adapter->q_vector[i]->cpu = -1;
4251 igb_update_dca(adapter->q_vector[i]);
4255 static int __igb_notify_dca(struct device *dev, void *data)
4257 struct net_device *netdev = dev_get_drvdata(dev);
4258 struct igb_adapter *adapter = netdev_priv(netdev);
4259 struct pci_dev *pdev = adapter->pdev;
4260 struct e1000_hw *hw = &adapter->hw;
4261 unsigned long event = *(unsigned long *)data;
4264 case DCA_PROVIDER_ADD:
4265 /* if already enabled, don't do it again */
4266 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
4268 if (dca_add_requester(dev) == 0) {
4269 adapter->flags |= IGB_FLAG_DCA_ENABLED;
4270 dev_info(&pdev->dev, "DCA enabled\n");
4271 igb_setup_dca(adapter);
4274 /* Fall Through since DCA is disabled. */
4275 case DCA_PROVIDER_REMOVE:
4276 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
4277 /* without this a class_device is left
4278 * hanging around in the sysfs model */
4279 dca_remove_requester(dev);
4280 dev_info(&pdev->dev, "DCA disabled\n");
4281 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
4282 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
4290 static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
4295 ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
4298 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
4300 #endif /* CONFIG_IGB_DCA */
4302 static void igb_ping_all_vfs(struct igb_adapter *adapter)
4304 struct e1000_hw *hw = &adapter->hw;
4308 for (i = 0 ; i < adapter->vfs_allocated_count; i++) {
4309 ping = E1000_PF_CONTROL_MSG;
4310 if (adapter->vf_data[i].flags & IGB_VF_FLAG_CTS)
4311 ping |= E1000_VT_MSGTYPE_CTS;
4312 igb_write_mbx(hw, &ping, 1, i);
4316 static int igb_set_vf_promisc(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
4318 struct e1000_hw *hw = &adapter->hw;
4319 u32 vmolr = rd32(E1000_VMOLR(vf));
4320 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
4322 vf_data->flags |= ~(IGB_VF_FLAG_UNI_PROMISC |
4323 IGB_VF_FLAG_MULTI_PROMISC);
4324 vmolr &= ~(E1000_VMOLR_ROPE | E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
4326 if (*msgbuf & E1000_VF_SET_PROMISC_MULTICAST) {
4327 vmolr |= E1000_VMOLR_MPME;
4328 *msgbuf &= ~E1000_VF_SET_PROMISC_MULTICAST;
4331 * if we have hashes and we are clearing a multicast promisc
4332 * flag we need to write the hashes to the MTA as this step
4333 * was previously skipped
4335 if (vf_data->num_vf_mc_hashes > 30) {
4336 vmolr |= E1000_VMOLR_MPME;
4337 } else if (vf_data->num_vf_mc_hashes) {
4339 vmolr |= E1000_VMOLR_ROMPE;
4340 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
4341 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
4345 wr32(E1000_VMOLR(vf), vmolr);
4347 /* there are flags left unprocessed, likely not supported */
4348 if (*msgbuf & E1000_VT_MSGINFO_MASK)
4355 static int igb_set_vf_multicasts(struct igb_adapter *adapter,
4356 u32 *msgbuf, u32 vf)
4358 int n = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
4359 u16 *hash_list = (u16 *)&msgbuf[1];
4360 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
4363 /* salt away the number of multicast addresses assigned
4364 * to this VF for later use to restore when the PF multi cast
4367 vf_data->num_vf_mc_hashes = n;
4369 /* only up to 30 hash values supported */
4373 /* store the hashes for later use */
4374 for (i = 0; i < n; i++)
4375 vf_data->vf_mc_hashes[i] = hash_list[i];
4377 /* Flush and reset the mta with the new values */
4378 igb_set_rx_mode(adapter->netdev);
4383 static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
4385 struct e1000_hw *hw = &adapter->hw;
4386 struct vf_data_storage *vf_data;
4389 for (i = 0; i < adapter->vfs_allocated_count; i++) {
4390 u32 vmolr = rd32(E1000_VMOLR(i));
4391 vmolr &= ~(E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
4393 vf_data = &adapter->vf_data[i];
4395 if ((vf_data->num_vf_mc_hashes > 30) ||
4396 (vf_data->flags & IGB_VF_FLAG_MULTI_PROMISC)) {
4397 vmolr |= E1000_VMOLR_MPME;
4398 } else if (vf_data->num_vf_mc_hashes) {
4399 vmolr |= E1000_VMOLR_ROMPE;
4400 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
4401 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
4403 wr32(E1000_VMOLR(i), vmolr);
4407 static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
4409 struct e1000_hw *hw = &adapter->hw;
4410 u32 pool_mask, reg, vid;
4413 pool_mask = 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
4415 /* Find the vlan filter for this id */
4416 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
4417 reg = rd32(E1000_VLVF(i));
4419 /* remove the vf from the pool */
4422 /* if pool is empty then remove entry from vfta */
4423 if (!(reg & E1000_VLVF_POOLSEL_MASK) &&
4424 (reg & E1000_VLVF_VLANID_ENABLE)) {
4426 vid = reg & E1000_VLVF_VLANID_MASK;
4427 igb_vfta_set(hw, vid, false);
4430 wr32(E1000_VLVF(i), reg);
4433 adapter->vf_data[vf].vlans_enabled = 0;
4436 static s32 igb_vlvf_set(struct igb_adapter *adapter, u32 vid, bool add, u32 vf)
4438 struct e1000_hw *hw = &adapter->hw;
4441 /* The vlvf table only exists on 82576 hardware and newer */
4442 if (hw->mac.type < e1000_82576)
4445 /* we only need to do this if VMDq is enabled */
4446 if (!adapter->vfs_allocated_count)
4449 /* Find the vlan filter for this id */
4450 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
4451 reg = rd32(E1000_VLVF(i));
4452 if ((reg & E1000_VLVF_VLANID_ENABLE) &&
4453 vid == (reg & E1000_VLVF_VLANID_MASK))
4458 if (i == E1000_VLVF_ARRAY_SIZE) {
4459 /* Did not find a matching VLAN ID entry that was
4460 * enabled. Search for a free filter entry, i.e.
4461 * one without the enable bit set
4463 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
4464 reg = rd32(E1000_VLVF(i));
4465 if (!(reg & E1000_VLVF_VLANID_ENABLE))
4469 if (i < E1000_VLVF_ARRAY_SIZE) {
4470 /* Found an enabled/available entry */
4471 reg |= 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
4473 /* if !enabled we need to set this up in vfta */
4474 if (!(reg & E1000_VLVF_VLANID_ENABLE)) {
4475 /* add VID to filter table */
4476 igb_vfta_set(hw, vid, true);
4477 reg |= E1000_VLVF_VLANID_ENABLE;
4479 reg &= ~E1000_VLVF_VLANID_MASK;
4481 wr32(E1000_VLVF(i), reg);
4483 /* do not modify RLPML for PF devices */
4484 if (vf >= adapter->vfs_allocated_count)
4487 if (!adapter->vf_data[vf].vlans_enabled) {
4489 reg = rd32(E1000_VMOLR(vf));
4490 size = reg & E1000_VMOLR_RLPML_MASK;
4492 reg &= ~E1000_VMOLR_RLPML_MASK;
4494 wr32(E1000_VMOLR(vf), reg);
4497 adapter->vf_data[vf].vlans_enabled++;
4501 if (i < E1000_VLVF_ARRAY_SIZE) {
4502 /* remove vf from the pool */
4503 reg &= ~(1 << (E1000_VLVF_POOLSEL_SHIFT + vf));
4504 /* if pool is empty then remove entry from vfta */
4505 if (!(reg & E1000_VLVF_POOLSEL_MASK)) {
4507 igb_vfta_set(hw, vid, false);
4509 wr32(E1000_VLVF(i), reg);
4511 /* do not modify RLPML for PF devices */
4512 if (vf >= adapter->vfs_allocated_count)
4515 adapter->vf_data[vf].vlans_enabled--;
4516 if (!adapter->vf_data[vf].vlans_enabled) {
4518 reg = rd32(E1000_VMOLR(vf));
4519 size = reg & E1000_VMOLR_RLPML_MASK;
4521 reg &= ~E1000_VMOLR_RLPML_MASK;
4523 wr32(E1000_VMOLR(vf), reg);
4530 static void igb_set_vmvir(struct igb_adapter *adapter, u32 vid, u32 vf)
4532 struct e1000_hw *hw = &adapter->hw;
4535 wr32(E1000_VMVIR(vf), (vid | E1000_VMVIR_VLANA_DEFAULT));
4537 wr32(E1000_VMVIR(vf), 0);
4540 static int igb_ndo_set_vf_vlan(struct net_device *netdev,
4541 int vf, u16 vlan, u8 qos)
4544 struct igb_adapter *adapter = netdev_priv(netdev);
4546 if ((vf >= adapter->vfs_allocated_count) || (vlan > 4095) || (qos > 7))
4549 err = igb_vlvf_set(adapter, vlan, !!vlan, vf);
4552 igb_set_vmvir(adapter, vlan | (qos << VLAN_PRIO_SHIFT), vf);
4553 igb_set_vmolr(adapter, vf, !vlan);
4554 adapter->vf_data[vf].pf_vlan = vlan;
4555 adapter->vf_data[vf].pf_qos = qos;
4556 dev_info(&adapter->pdev->dev,
4557 "Setting VLAN %d, QOS 0x%x on VF %d\n", vlan, qos, vf);
4558 if (test_bit(__IGB_DOWN, &adapter->state)) {
4559 dev_warn(&adapter->pdev->dev,
4560 "The VF VLAN has been set,"
4561 " but the PF device is not up.\n");
4562 dev_warn(&adapter->pdev->dev,
4563 "Bring the PF device up before"
4564 " attempting to use the VF device.\n");
4567 igb_vlvf_set(adapter, adapter->vf_data[vf].pf_vlan,
4569 igb_set_vmvir(adapter, vlan, vf);
4570 igb_set_vmolr(adapter, vf, true);
4571 adapter->vf_data[vf].pf_vlan = 0;
4572 adapter->vf_data[vf].pf_qos = 0;
4578 static int igb_set_vf_vlan(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
4580 int add = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
4581 int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
4583 return igb_vlvf_set(adapter, vid, add, vf);
4586 static inline void igb_vf_reset(struct igb_adapter *adapter, u32 vf)
4589 adapter->vf_data[vf].flags &= ~(IGB_VF_FLAG_PF_SET_MAC);
4590 adapter->vf_data[vf].last_nack = jiffies;
4592 /* reset offloads to defaults */
4593 igb_set_vmolr(adapter, vf, true);
4595 /* reset vlans for device */
4596 igb_clear_vf_vfta(adapter, vf);
4597 if (adapter->vf_data[vf].pf_vlan)
4598 igb_ndo_set_vf_vlan(adapter->netdev, vf,
4599 adapter->vf_data[vf].pf_vlan,
4600 adapter->vf_data[vf].pf_qos);
4602 igb_clear_vf_vfta(adapter, vf);
4604 /* reset multicast table array for vf */
4605 adapter->vf_data[vf].num_vf_mc_hashes = 0;
4607 /* Flush and reset the mta with the new values */
4608 igb_set_rx_mode(adapter->netdev);
4611 static void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
4613 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
4615 /* generate a new mac address as we were hotplug removed/added */
4616 if (!(adapter->vf_data[vf].flags & IGB_VF_FLAG_PF_SET_MAC))
4617 random_ether_addr(vf_mac);
4619 /* process remaining reset events */
4620 igb_vf_reset(adapter, vf);
4623 static void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
4625 struct e1000_hw *hw = &adapter->hw;
4626 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
4627 int rar_entry = hw->mac.rar_entry_count - (vf + 1);
4629 u8 *addr = (u8 *)(&msgbuf[1]);
4631 /* process all the same items cleared in a function level reset */
4632 igb_vf_reset(adapter, vf);
4634 /* set vf mac address */
4635 igb_rar_set_qsel(adapter, vf_mac, rar_entry, vf);
4637 /* enable transmit and receive for vf */
4638 reg = rd32(E1000_VFTE);
4639 wr32(E1000_VFTE, reg | (1 << vf));
4640 reg = rd32(E1000_VFRE);
4641 wr32(E1000_VFRE, reg | (1 << vf));
4643 adapter->vf_data[vf].flags = IGB_VF_FLAG_CTS;
4645 /* reply to reset with ack and vf mac address */
4646 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
4647 memcpy(addr, vf_mac, 6);
4648 igb_write_mbx(hw, msgbuf, 3, vf);
4651 static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf)
4653 unsigned char *addr = (char *)&msg[1];
4656 if (is_valid_ether_addr(addr))
4657 err = igb_set_vf_mac(adapter, vf, addr);
4662 static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf)
4664 struct e1000_hw *hw = &adapter->hw;
4665 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
4666 u32 msg = E1000_VT_MSGTYPE_NACK;
4668 /* if device isn't clear to send it shouldn't be reading either */
4669 if (!(vf_data->flags & IGB_VF_FLAG_CTS) &&
4670 time_after(jiffies, vf_data->last_nack + (2 * HZ))) {
4671 igb_write_mbx(hw, &msg, 1, vf);
4672 vf_data->last_nack = jiffies;
4676 static void igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
4678 struct pci_dev *pdev = adapter->pdev;
4679 u32 msgbuf[E1000_VFMAILBOX_SIZE];
4680 struct e1000_hw *hw = &adapter->hw;
4681 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
4684 retval = igb_read_mbx(hw, msgbuf, E1000_VFMAILBOX_SIZE, vf);
4687 /* if receive failed revoke VF CTS stats and restart init */
4688 dev_err(&pdev->dev, "Error receiving message from VF\n");
4689 vf_data->flags &= ~IGB_VF_FLAG_CTS;
4690 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
4695 /* this is a message we already processed, do nothing */
4696 if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
4700 * until the vf completes a reset it should not be
4701 * allowed to start any configuration.
4704 if (msgbuf[0] == E1000_VF_RESET) {
4705 igb_vf_reset_msg(adapter, vf);
4709 if (!(vf_data->flags & IGB_VF_FLAG_CTS)) {
4710 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
4716 switch ((msgbuf[0] & 0xFFFF)) {
4717 case E1000_VF_SET_MAC_ADDR:
4718 retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
4720 case E1000_VF_SET_PROMISC:
4721 retval = igb_set_vf_promisc(adapter, msgbuf, vf);
4723 case E1000_VF_SET_MULTICAST:
4724 retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
4726 case E1000_VF_SET_LPE:
4727 retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf);
4729 case E1000_VF_SET_VLAN:
4730 if (adapter->vf_data[vf].pf_vlan)
4733 retval = igb_set_vf_vlan(adapter, msgbuf, vf);
4736 dev_err(&pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
4741 msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
4743 /* notify the VF of the results of what it sent us */
4745 msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
4747 msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
4749 igb_write_mbx(hw, msgbuf, 1, vf);
4752 static void igb_msg_task(struct igb_adapter *adapter)
4754 struct e1000_hw *hw = &adapter->hw;
4757 for (vf = 0; vf < adapter->vfs_allocated_count; vf++) {
4758 /* process any reset requests */
4759 if (!igb_check_for_rst(hw, vf))
4760 igb_vf_reset_event(adapter, vf);
4762 /* process any messages pending */
4763 if (!igb_check_for_msg(hw, vf))
4764 igb_rcv_msg_from_vf(adapter, vf);
4766 /* process any acks */
4767 if (!igb_check_for_ack(hw, vf))
4768 igb_rcv_ack_from_vf(adapter, vf);
4773 * igb_set_uta - Set unicast filter table address
4774 * @adapter: board private structure
4776 * The unicast table address is a register array of 32-bit registers.
4777 * The table is meant to be used in a way similar to how the MTA is used
4778 * however due to certain limitations in the hardware it is necessary to
4779 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscous
4780 * enable bit to allow vlan tag stripping when promiscous mode is enabled
4782 static void igb_set_uta(struct igb_adapter *adapter)
4784 struct e1000_hw *hw = &adapter->hw;
4787 /* The UTA table only exists on 82576 hardware and newer */
4788 if (hw->mac.type < e1000_82576)
4791 /* we only need to do this if VMDq is enabled */
4792 if (!adapter->vfs_allocated_count)
4795 for (i = 0; i < hw->mac.uta_reg_count; i++)
4796 array_wr32(E1000_UTA, i, ~0);
4800 * igb_intr_msi - Interrupt Handler
4801 * @irq: interrupt number
4802 * @data: pointer to a network interface device structure
4804 static irqreturn_t igb_intr_msi(int irq, void *data)
4806 struct igb_adapter *adapter = data;
4807 struct igb_q_vector *q_vector = adapter->q_vector[0];
4808 struct e1000_hw *hw = &adapter->hw;
4809 /* read ICR disables interrupts using IAM */
4810 u32 icr = rd32(E1000_ICR);
4812 igb_write_itr(q_vector);
4814 if (icr & E1000_ICR_DRSTA)
4815 schedule_work(&adapter->reset_task);
4817 if (icr & E1000_ICR_DOUTSYNC) {
4818 /* HW is reporting DMA is out of sync */
4819 adapter->stats.doosync++;
4822 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
4823 hw->mac.get_link_status = 1;
4824 if (!test_bit(__IGB_DOWN, &adapter->state))
4825 mod_timer(&adapter->watchdog_timer, jiffies + 1);
4828 napi_schedule(&q_vector->napi);
4834 * igb_intr - Legacy Interrupt Handler
4835 * @irq: interrupt number
4836 * @data: pointer to a network interface device structure
4838 static irqreturn_t igb_intr(int irq, void *data)
4840 struct igb_adapter *adapter = data;
4841 struct igb_q_vector *q_vector = adapter->q_vector[0];
4842 struct e1000_hw *hw = &adapter->hw;
4843 /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked. No
4844 * need for the IMC write */
4845 u32 icr = rd32(E1000_ICR);
4847 return IRQ_NONE; /* Not our interrupt */
4849 igb_write_itr(q_vector);
4851 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
4852 * not set, then the adapter didn't send an interrupt */
4853 if (!(icr & E1000_ICR_INT_ASSERTED))
4856 if (icr & E1000_ICR_DRSTA)
4857 schedule_work(&adapter->reset_task);
4859 if (icr & E1000_ICR_DOUTSYNC) {
4860 /* HW is reporting DMA is out of sync */
4861 adapter->stats.doosync++;
4864 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
4865 hw->mac.get_link_status = 1;
4866 /* guard against interrupt when we're going down */
4867 if (!test_bit(__IGB_DOWN, &adapter->state))
4868 mod_timer(&adapter->watchdog_timer, jiffies + 1);
4871 napi_schedule(&q_vector->napi);
4876 static inline void igb_ring_irq_enable(struct igb_q_vector *q_vector)
4878 struct igb_adapter *adapter = q_vector->adapter;
4879 struct e1000_hw *hw = &adapter->hw;
4881 if ((q_vector->rx_ring && (adapter->rx_itr_setting & 3)) ||
4882 (!q_vector->rx_ring && (adapter->tx_itr_setting & 3))) {
4883 if (!adapter->msix_entries)
4884 igb_set_itr(adapter);
4886 igb_update_ring_itr(q_vector);
4889 if (!test_bit(__IGB_DOWN, &adapter->state)) {
4890 if (adapter->msix_entries)
4891 wr32(E1000_EIMS, q_vector->eims_value);
4893 igb_irq_enable(adapter);
4898 * igb_poll - NAPI Rx polling callback
4899 * @napi: napi polling structure
4900 * @budget: count of how many packets we should handle
4902 static int igb_poll(struct napi_struct *napi, int budget)
4904 struct igb_q_vector *q_vector = container_of(napi,
4905 struct igb_q_vector,
4907 int tx_clean_complete = 1, work_done = 0;
4909 #ifdef CONFIG_IGB_DCA
4910 if (q_vector->adapter->flags & IGB_FLAG_DCA_ENABLED)
4911 igb_update_dca(q_vector);
4913 if (q_vector->tx_ring)
4914 tx_clean_complete = igb_clean_tx_irq(q_vector);
4916 if (q_vector->rx_ring)
4917 igb_clean_rx_irq_adv(q_vector, &work_done, budget);
4919 if (!tx_clean_complete)
4922 /* If not enough Rx work done, exit the polling mode */
4923 if (work_done < budget) {
4924 napi_complete(napi);
4925 igb_ring_irq_enable(q_vector);
4932 * igb_systim_to_hwtstamp - convert system time value to hw timestamp
4933 * @adapter: board private structure
4934 * @shhwtstamps: timestamp structure to update
4935 * @regval: unsigned 64bit system time value.
4937 * We need to convert the system time value stored in the RX/TXSTMP registers
4938 * into a hwtstamp which can be used by the upper level timestamping functions
4940 static void igb_systim_to_hwtstamp(struct igb_adapter *adapter,
4941 struct skb_shared_hwtstamps *shhwtstamps,
4947 * The 82580 starts with 1ns at bit 0 in RX/TXSTMPL, shift this up to
4948 * 24 to match clock shift we setup earlier.
4950 if (adapter->hw.mac.type == e1000_82580)
4951 regval <<= IGB_82580_TSYNC_SHIFT;
4953 ns = timecounter_cyc2time(&adapter->clock, regval);
4954 timecompare_update(&adapter->compare, ns);
4955 memset(shhwtstamps, 0, sizeof(struct skb_shared_hwtstamps));
4956 shhwtstamps->hwtstamp = ns_to_ktime(ns);
4957 shhwtstamps->syststamp = timecompare_transform(&adapter->compare, ns);
4961 * igb_tx_hwtstamp - utility function which checks for TX time stamp
4962 * @q_vector: pointer to q_vector containing needed info
4963 * @skb: packet that was just sent
4965 * If we were asked to do hardware stamping and such a time stamp is
4966 * available, then it must have been for this skb here because we only
4967 * allow only one such packet into the queue.
4969 static void igb_tx_hwtstamp(struct igb_q_vector *q_vector, struct sk_buff *skb)
4971 struct igb_adapter *adapter = q_vector->adapter;
4972 union skb_shared_tx *shtx = skb_tx(skb);
4973 struct e1000_hw *hw = &adapter->hw;
4974 struct skb_shared_hwtstamps shhwtstamps;
4977 /* if skb does not support hw timestamp or TX stamp not valid exit */
4978 if (likely(!shtx->hardware) ||
4979 !(rd32(E1000_TSYNCTXCTL) & E1000_TSYNCTXCTL_VALID))
4982 regval = rd32(E1000_TXSTMPL);
4983 regval |= (u64)rd32(E1000_TXSTMPH) << 32;
4985 igb_systim_to_hwtstamp(adapter, &shhwtstamps, regval);
4986 skb_tstamp_tx(skb, &shhwtstamps);
4990 * igb_clean_tx_irq - Reclaim resources after transmit completes
4991 * @q_vector: pointer to q_vector containing needed info
4992 * returns true if ring is completely cleaned
4994 static bool igb_clean_tx_irq(struct igb_q_vector *q_vector)
4996 struct igb_adapter *adapter = q_vector->adapter;
4997 struct igb_ring *tx_ring = q_vector->tx_ring;
4998 struct net_device *netdev = tx_ring->netdev;
4999 struct e1000_hw *hw = &adapter->hw;
5000 struct igb_buffer *buffer_info;
5001 struct sk_buff *skb;
5002 union e1000_adv_tx_desc *tx_desc, *eop_desc;
5003 unsigned int total_bytes = 0, total_packets = 0;
5004 unsigned int i, eop, count = 0;
5005 bool cleaned = false;
5007 i = tx_ring->next_to_clean;
5008 eop = tx_ring->buffer_info[i].next_to_watch;
5009 eop_desc = E1000_TX_DESC_ADV(*tx_ring, eop);
5011 while ((eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)) &&
5012 (count < tx_ring->count)) {
5013 for (cleaned = false; !cleaned; count++) {
5014 tx_desc = E1000_TX_DESC_ADV(*tx_ring, i);
5015 buffer_info = &tx_ring->buffer_info[i];
5016 cleaned = (i == eop);
5017 skb = buffer_info->skb;
5020 unsigned int segs, bytecount;
5021 /* gso_segs is currently only valid for tcp */
5022 segs = buffer_info->gso_segs;
5023 /* multiply data chunks by size of headers */
5024 bytecount = ((segs - 1) * skb_headlen(skb)) +
5026 total_packets += segs;
5027 total_bytes += bytecount;
5029 igb_tx_hwtstamp(q_vector, skb);
5032 igb_unmap_and_free_tx_resource(tx_ring, buffer_info);
5033 tx_desc->wb.status = 0;
5036 if (i == tx_ring->count)
5039 eop = tx_ring->buffer_info[i].next_to_watch;
5040 eop_desc = E1000_TX_DESC_ADV(*tx_ring, eop);
5043 tx_ring->next_to_clean = i;
5045 if (unlikely(count &&
5046 netif_carrier_ok(netdev) &&
5047 igb_desc_unused(tx_ring) >= IGB_TX_QUEUE_WAKE)) {
5048 /* Make sure that anybody stopping the queue after this
5049 * sees the new next_to_clean.
5052 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
5053 !(test_bit(__IGB_DOWN, &adapter->state))) {
5054 netif_wake_subqueue(netdev, tx_ring->queue_index);
5055 tx_ring->tx_stats.restart_queue++;
5059 if (tx_ring->detect_tx_hung) {
5060 /* Detect a transmit hang in hardware, this serializes the
5061 * check with the clearing of time_stamp and movement of i */
5062 tx_ring->detect_tx_hung = false;
5063 if (tx_ring->buffer_info[i].time_stamp &&
5064 time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
5065 (adapter->tx_timeout_factor * HZ)) &&
5066 !(rd32(E1000_STATUS) & E1000_STATUS_TXOFF)) {
5068 /* detected Tx unit hang */
5069 dev_err(&tx_ring->pdev->dev,
5070 "Detected Tx Unit Hang\n"
5074 " next_to_use <%x>\n"
5075 " next_to_clean <%x>\n"
5076 "buffer_info[next_to_clean]\n"
5077 " time_stamp <%lx>\n"
5078 " next_to_watch <%x>\n"
5080 " desc.status <%x>\n",
5081 tx_ring->queue_index,
5082 readl(tx_ring->head),
5083 readl(tx_ring->tail),
5084 tx_ring->next_to_use,
5085 tx_ring->next_to_clean,
5086 tx_ring->buffer_info[eop].time_stamp,
5089 eop_desc->wb.status);
5090 netif_stop_subqueue(netdev, tx_ring->queue_index);
5093 tx_ring->total_bytes += total_bytes;
5094 tx_ring->total_packets += total_packets;
5095 tx_ring->tx_stats.bytes += total_bytes;
5096 tx_ring->tx_stats.packets += total_packets;
5097 return (count < tx_ring->count);
5101 * igb_receive_skb - helper function to handle rx indications
5102 * @q_vector: structure containing interrupt and ring information
5103 * @skb: packet to send up
5104 * @vlan_tag: vlan tag for packet
5106 static void igb_receive_skb(struct igb_q_vector *q_vector,
5107 struct sk_buff *skb,
5110 struct igb_adapter *adapter = q_vector->adapter;
5113 vlan_gro_receive(&q_vector->napi, adapter->vlgrp,
5116 napi_gro_receive(&q_vector->napi, skb);
5119 static inline void igb_rx_checksum_adv(struct igb_ring *ring,
5120 u32 status_err, struct sk_buff *skb)
5122 skb->ip_summed = CHECKSUM_NONE;
5124 /* Ignore Checksum bit is set or checksum is disabled through ethtool */
5125 if (!(ring->flags & IGB_RING_FLAG_RX_CSUM) ||
5126 (status_err & E1000_RXD_STAT_IXSM))
5129 /* TCP/UDP checksum error bit is set */
5131 (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
5133 * work around errata with sctp packets where the TCPE aka
5134 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
5135 * packets, (aka let the stack check the crc32c)
5137 if ((skb->len == 60) &&
5138 (ring->flags & IGB_RING_FLAG_RX_SCTP_CSUM))
5139 ring->rx_stats.csum_err++;
5141 /* let the stack verify checksum errors */
5144 /* It must be a TCP or UDP packet with a valid checksum */
5145 if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
5146 skb->ip_summed = CHECKSUM_UNNECESSARY;
5148 dev_dbg(&ring->pdev->dev, "cksum success: bits %08X\n", status_err);
5151 static inline void igb_rx_hwtstamp(struct igb_q_vector *q_vector, u32 staterr,
5152 struct sk_buff *skb)
5154 struct igb_adapter *adapter = q_vector->adapter;
5155 struct e1000_hw *hw = &adapter->hw;
5159 * If this bit is set, then the RX registers contain the time stamp. No
5160 * other packet will be time stamped until we read these registers, so
5161 * read the registers to make them available again. Because only one
5162 * packet can be time stamped at a time, we know that the register
5163 * values must belong to this one here and therefore we don't need to
5164 * compare any of the additional attributes stored for it.
5166 * If nothing went wrong, then it should have a skb_shared_tx that we
5167 * can turn into a skb_shared_hwtstamps.
5169 if (likely(!(staterr & E1000_RXDADV_STAT_TS)))
5171 if (!(rd32(E1000_TSYNCRXCTL) & E1000_TSYNCRXCTL_VALID))
5174 regval = rd32(E1000_RXSTMPL);
5175 regval |= (u64)rd32(E1000_RXSTMPH) << 32;
5177 igb_systim_to_hwtstamp(adapter, skb_hwtstamps(skb), regval);
5179 static inline u16 igb_get_hlen(struct igb_ring *rx_ring,
5180 union e1000_adv_rx_desc *rx_desc)
5182 /* HW will not DMA in data larger than the given buffer, even if it
5183 * parses the (NFS, of course) header to be larger. In that case, it
5184 * fills the header buffer and spills the rest into the page.
5186 u16 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hdr_info) &
5187 E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
5188 if (hlen > rx_ring->rx_buffer_len)
5189 hlen = rx_ring->rx_buffer_len;
5193 static bool igb_clean_rx_irq_adv(struct igb_q_vector *q_vector,
5194 int *work_done, int budget)
5196 struct igb_ring *rx_ring = q_vector->rx_ring;
5197 struct net_device *netdev = rx_ring->netdev;
5198 struct pci_dev *pdev = rx_ring->pdev;
5199 union e1000_adv_rx_desc *rx_desc , *next_rxd;
5200 struct igb_buffer *buffer_info , *next_buffer;
5201 struct sk_buff *skb;
5202 bool cleaned = false;
5203 int cleaned_count = 0;
5204 int current_node = numa_node_id();
5205 unsigned int total_bytes = 0, total_packets = 0;
5211 i = rx_ring->next_to_clean;
5212 buffer_info = &rx_ring->buffer_info[i];
5213 rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
5214 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
5216 while (staterr & E1000_RXD_STAT_DD) {
5217 if (*work_done >= budget)
5221 skb = buffer_info->skb;
5222 prefetch(skb->data - NET_IP_ALIGN);
5223 buffer_info->skb = NULL;
5226 if (i == rx_ring->count)
5229 next_rxd = E1000_RX_DESC_ADV(*rx_ring, i);
5231 next_buffer = &rx_ring->buffer_info[i];
5233 length = le16_to_cpu(rx_desc->wb.upper.length);
5237 if (buffer_info->dma) {
5238 pci_unmap_single(pdev, buffer_info->dma,
5239 rx_ring->rx_buffer_len,
5240 PCI_DMA_FROMDEVICE);
5241 buffer_info->dma = 0;
5242 if (rx_ring->rx_buffer_len >= IGB_RXBUFFER_1024) {
5243 skb_put(skb, length);
5246 skb_put(skb, igb_get_hlen(rx_ring, rx_desc));
5250 pci_unmap_page(pdev, buffer_info->page_dma,
5251 PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
5252 buffer_info->page_dma = 0;
5254 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags++,
5256 buffer_info->page_offset,
5259 if ((page_count(buffer_info->page) != 1) ||
5260 (page_to_nid(buffer_info->page) != current_node))
5261 buffer_info->page = NULL;
5263 get_page(buffer_info->page);
5266 skb->data_len += length;
5267 skb->truesize += length;
5270 if (!(staterr & E1000_RXD_STAT_EOP)) {
5271 buffer_info->skb = next_buffer->skb;
5272 buffer_info->dma = next_buffer->dma;
5273 next_buffer->skb = skb;
5274 next_buffer->dma = 0;
5278 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
5279 dev_kfree_skb_irq(skb);
5283 igb_rx_hwtstamp(q_vector, staterr, skb);
5284 total_bytes += skb->len;
5287 igb_rx_checksum_adv(rx_ring, staterr, skb);
5289 skb->protocol = eth_type_trans(skb, netdev);
5290 skb_record_rx_queue(skb, rx_ring->queue_index);
5292 vlan_tag = ((staterr & E1000_RXD_STAT_VP) ?
5293 le16_to_cpu(rx_desc->wb.upper.vlan) : 0);
5295 igb_receive_skb(q_vector, skb, vlan_tag);
5298 rx_desc->wb.upper.status_error = 0;
5300 /* return some buffers to hardware, one at a time is too slow */
5301 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
5302 igb_alloc_rx_buffers_adv(rx_ring, cleaned_count);
5306 /* use prefetched values */
5308 buffer_info = next_buffer;
5309 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
5312 rx_ring->next_to_clean = i;
5313 cleaned_count = igb_desc_unused(rx_ring);
5316 igb_alloc_rx_buffers_adv(rx_ring, cleaned_count);
5318 rx_ring->total_packets += total_packets;
5319 rx_ring->total_bytes += total_bytes;
5320 rx_ring->rx_stats.packets += total_packets;
5321 rx_ring->rx_stats.bytes += total_bytes;
5326 * igb_alloc_rx_buffers_adv - Replace used receive buffers; packet split
5327 * @adapter: address of board private structure
5329 void igb_alloc_rx_buffers_adv(struct igb_ring *rx_ring, int cleaned_count)
5331 struct net_device *netdev = rx_ring->netdev;
5332 union e1000_adv_rx_desc *rx_desc;
5333 struct igb_buffer *buffer_info;
5334 struct sk_buff *skb;
5338 i = rx_ring->next_to_use;
5339 buffer_info = &rx_ring->buffer_info[i];
5341 bufsz = rx_ring->rx_buffer_len;
5343 while (cleaned_count--) {
5344 rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
5346 if ((bufsz < IGB_RXBUFFER_1024) && !buffer_info->page_dma) {
5347 if (!buffer_info->page) {
5348 buffer_info->page = netdev_alloc_page(netdev);
5349 if (!buffer_info->page) {
5350 rx_ring->rx_stats.alloc_failed++;
5353 buffer_info->page_offset = 0;
5355 buffer_info->page_offset ^= PAGE_SIZE / 2;
5357 buffer_info->page_dma =
5358 pci_map_page(rx_ring->pdev, buffer_info->page,
5359 buffer_info->page_offset,
5361 PCI_DMA_FROMDEVICE);
5362 if (pci_dma_mapping_error(rx_ring->pdev,
5363 buffer_info->page_dma)) {
5364 buffer_info->page_dma = 0;
5365 rx_ring->rx_stats.alloc_failed++;
5370 skb = buffer_info->skb;
5372 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
5374 rx_ring->rx_stats.alloc_failed++;
5378 buffer_info->skb = skb;
5380 if (!buffer_info->dma) {
5381 buffer_info->dma = pci_map_single(rx_ring->pdev,
5384 PCI_DMA_FROMDEVICE);
5385 if (pci_dma_mapping_error(rx_ring->pdev,
5386 buffer_info->dma)) {
5387 buffer_info->dma = 0;
5388 rx_ring->rx_stats.alloc_failed++;
5392 /* Refresh the desc even if buffer_addrs didn't change because
5393 * each write-back erases this info. */
5394 if (bufsz < IGB_RXBUFFER_1024) {
5395 rx_desc->read.pkt_addr =
5396 cpu_to_le64(buffer_info->page_dma);
5397 rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
5399 rx_desc->read.pkt_addr = cpu_to_le64(buffer_info->dma);
5400 rx_desc->read.hdr_addr = 0;
5404 if (i == rx_ring->count)
5406 buffer_info = &rx_ring->buffer_info[i];
5410 if (rx_ring->next_to_use != i) {
5411 rx_ring->next_to_use = i;
5413 i = (rx_ring->count - 1);
5417 /* Force memory writes to complete before letting h/w
5418 * know there are new descriptors to fetch. (Only
5419 * applicable for weak-ordered memory model archs,
5420 * such as IA-64). */
5422 writel(i, rx_ring->tail);
5432 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5434 struct igb_adapter *adapter = netdev_priv(netdev);
5435 struct mii_ioctl_data *data = if_mii(ifr);
5437 if (adapter->hw.phy.media_type != e1000_media_type_copper)
5442 data->phy_id = adapter->hw.phy.addr;
5445 if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
5457 * igb_hwtstamp_ioctl - control hardware time stamping
5462 * Outgoing time stamping can be enabled and disabled. Play nice and
5463 * disable it when requested, although it shouldn't case any overhead
5464 * when no packet needs it. At most one packet in the queue may be
5465 * marked for time stamping, otherwise it would be impossible to tell
5466 * for sure to which packet the hardware time stamp belongs.
5468 * Incoming time stamping has to be configured via the hardware
5469 * filters. Not all combinations are supported, in particular event
5470 * type has to be specified. Matching the kind of event packet is
5471 * not supported, with the exception of "all V2 events regardless of
5475 static int igb_hwtstamp_ioctl(struct net_device *netdev,
5476 struct ifreq *ifr, int cmd)
5478 struct igb_adapter *adapter = netdev_priv(netdev);
5479 struct e1000_hw *hw = &adapter->hw;
5480 struct hwtstamp_config config;
5481 u32 tsync_tx_ctl = E1000_TSYNCTXCTL_ENABLED;
5482 u32 tsync_rx_ctl = E1000_TSYNCRXCTL_ENABLED;
5483 u32 tsync_rx_cfg = 0;
5488 if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
5491 /* reserved for future extensions */
5495 switch (config.tx_type) {
5496 case HWTSTAMP_TX_OFF:
5498 case HWTSTAMP_TX_ON:
5504 switch (config.rx_filter) {
5505 case HWTSTAMP_FILTER_NONE:
5508 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
5509 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
5510 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
5511 case HWTSTAMP_FILTER_ALL:
5513 * register TSYNCRXCFG must be set, therefore it is not
5514 * possible to time stamp both Sync and Delay_Req messages
5515 * => fall back to time stamping all packets
5517 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_ALL;
5518 config.rx_filter = HWTSTAMP_FILTER_ALL;
5520 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
5521 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L4_V1;
5522 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V1_SYNC_MESSAGE;
5525 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
5526 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L4_V1;
5527 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V1_DELAY_REQ_MESSAGE;
5530 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
5531 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
5532 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
5533 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V2_SYNC_MESSAGE;
5536 config.rx_filter = HWTSTAMP_FILTER_SOME;
5538 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
5539 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
5540 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
5541 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V2_DELAY_REQ_MESSAGE;
5544 config.rx_filter = HWTSTAMP_FILTER_SOME;
5546 case HWTSTAMP_FILTER_PTP_V2_EVENT:
5547 case HWTSTAMP_FILTER_PTP_V2_SYNC:
5548 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
5549 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_EVENT_V2;
5550 config.rx_filter = HWTSTAMP_FILTER_PTP_V2_EVENT;
5557 if (hw->mac.type == e1000_82575) {
5558 if (tsync_rx_ctl | tsync_tx_ctl)
5563 /* enable/disable TX */
5564 regval = rd32(E1000_TSYNCTXCTL);
5565 regval &= ~E1000_TSYNCTXCTL_ENABLED;
5566 regval |= tsync_tx_ctl;
5567 wr32(E1000_TSYNCTXCTL, regval);
5569 /* enable/disable RX */
5570 regval = rd32(E1000_TSYNCRXCTL);
5571 regval &= ~(E1000_TSYNCRXCTL_ENABLED | E1000_TSYNCRXCTL_TYPE_MASK);
5572 regval |= tsync_rx_ctl;
5573 wr32(E1000_TSYNCRXCTL, regval);
5575 /* define which PTP packets are time stamped */
5576 wr32(E1000_TSYNCRXCFG, tsync_rx_cfg);
5578 /* define ethertype filter for timestamped packets */
5581 (E1000_ETQF_FILTER_ENABLE | /* enable filter */
5582 E1000_ETQF_1588 | /* enable timestamping */
5583 ETH_P_1588)); /* 1588 eth protocol type */
5585 wr32(E1000_ETQF(3), 0);
5587 #define PTP_PORT 319
5588 /* L4 Queue Filter[3]: filter by destination port and protocol */
5590 u32 ftqf = (IPPROTO_UDP /* UDP */
5591 | E1000_FTQF_VF_BP /* VF not compared */
5592 | E1000_FTQF_1588_TIME_STAMP /* Enable Timestamping */
5593 | E1000_FTQF_MASK); /* mask all inputs */
5594 ftqf &= ~E1000_FTQF_MASK_PROTO_BP; /* enable protocol check */
5596 wr32(E1000_IMIR(3), htons(PTP_PORT));
5597 wr32(E1000_IMIREXT(3),
5598 (E1000_IMIREXT_SIZE_BP | E1000_IMIREXT_CTRL_BP));
5599 if (hw->mac.type == e1000_82576) {
5600 /* enable source port check */
5601 wr32(E1000_SPQF(3), htons(PTP_PORT));
5602 ftqf &= ~E1000_FTQF_MASK_SOURCE_PORT_BP;
5604 wr32(E1000_FTQF(3), ftqf);
5606 wr32(E1000_FTQF(3), E1000_FTQF_MASK);
5610 adapter->hwtstamp_config = config;
5612 /* clear TX/RX time stamp registers, just to be sure */
5613 regval = rd32(E1000_TXSTMPH);
5614 regval = rd32(E1000_RXSTMPH);
5616 return copy_to_user(ifr->ifr_data, &config, sizeof(config)) ?
5626 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5632 return igb_mii_ioctl(netdev, ifr, cmd);
5634 return igb_hwtstamp_ioctl(netdev, ifr, cmd);
5640 s32 igb_read_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
5642 struct igb_adapter *adapter = hw->back;
5645 cap_offset = pci_find_capability(adapter->pdev, PCI_CAP_ID_EXP);
5647 return -E1000_ERR_CONFIG;
5649 pci_read_config_word(adapter->pdev, cap_offset + reg, value);
5654 s32 igb_write_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
5656 struct igb_adapter *adapter = hw->back;
5659 cap_offset = pci_find_capability(adapter->pdev, PCI_CAP_ID_EXP);
5661 return -E1000_ERR_CONFIG;
5663 pci_write_config_word(adapter->pdev, cap_offset + reg, *value);
5668 static void igb_vlan_rx_register(struct net_device *netdev,
5669 struct vlan_group *grp)
5671 struct igb_adapter *adapter = netdev_priv(netdev);
5672 struct e1000_hw *hw = &adapter->hw;
5675 igb_irq_disable(adapter);
5676 adapter->vlgrp = grp;
5679 /* enable VLAN tag insert/strip */
5680 ctrl = rd32(E1000_CTRL);
5681 ctrl |= E1000_CTRL_VME;
5682 wr32(E1000_CTRL, ctrl);
5684 /* Disable CFI check */
5685 rctl = rd32(E1000_RCTL);
5686 rctl &= ~E1000_RCTL_CFIEN;
5687 wr32(E1000_RCTL, rctl);
5689 /* disable VLAN tag insert/strip */
5690 ctrl = rd32(E1000_CTRL);
5691 ctrl &= ~E1000_CTRL_VME;
5692 wr32(E1000_CTRL, ctrl);
5695 igb_rlpml_set(adapter);
5697 if (!test_bit(__IGB_DOWN, &adapter->state))
5698 igb_irq_enable(adapter);
5701 static void igb_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
5703 struct igb_adapter *adapter = netdev_priv(netdev);
5704 struct e1000_hw *hw = &adapter->hw;
5705 int pf_id = adapter->vfs_allocated_count;
5707 /* attempt to add filter to vlvf array */
5708 igb_vlvf_set(adapter, vid, true, pf_id);
5710 /* add the filter since PF can receive vlans w/o entry in vlvf */
5711 igb_vfta_set(hw, vid, true);
5714 static void igb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
5716 struct igb_adapter *adapter = netdev_priv(netdev);
5717 struct e1000_hw *hw = &adapter->hw;
5718 int pf_id = adapter->vfs_allocated_count;
5721 igb_irq_disable(adapter);
5722 vlan_group_set_device(adapter->vlgrp, vid, NULL);
5724 if (!test_bit(__IGB_DOWN, &adapter->state))
5725 igb_irq_enable(adapter);
5727 /* remove vlan from VLVF table array */
5728 err = igb_vlvf_set(adapter, vid, false, pf_id);
5730 /* if vid was not present in VLVF just remove it from table */
5732 igb_vfta_set(hw, vid, false);
5735 static void igb_restore_vlan(struct igb_adapter *adapter)
5737 igb_vlan_rx_register(adapter->netdev, adapter->vlgrp);
5739 if (adapter->vlgrp) {
5741 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
5742 if (!vlan_group_get_device(adapter->vlgrp, vid))
5744 igb_vlan_rx_add_vid(adapter->netdev, vid);
5749 int igb_set_spd_dplx(struct igb_adapter *adapter, u16 spddplx)
5751 struct pci_dev *pdev = adapter->pdev;
5752 struct e1000_mac_info *mac = &adapter->hw.mac;
5757 case SPEED_10 + DUPLEX_HALF:
5758 mac->forced_speed_duplex = ADVERTISE_10_HALF;
5760 case SPEED_10 + DUPLEX_FULL:
5761 mac->forced_speed_duplex = ADVERTISE_10_FULL;
5763 case SPEED_100 + DUPLEX_HALF:
5764 mac->forced_speed_duplex = ADVERTISE_100_HALF;
5766 case SPEED_100 + DUPLEX_FULL:
5767 mac->forced_speed_duplex = ADVERTISE_100_FULL;
5769 case SPEED_1000 + DUPLEX_FULL:
5771 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
5773 case SPEED_1000 + DUPLEX_HALF: /* not supported */
5775 dev_err(&pdev->dev, "Unsupported Speed/Duplex configuration\n");
5781 static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake)
5783 struct net_device *netdev = pci_get_drvdata(pdev);
5784 struct igb_adapter *adapter = netdev_priv(netdev);
5785 struct e1000_hw *hw = &adapter->hw;
5786 u32 ctrl, rctl, status;
5787 u32 wufc = adapter->wol;
5792 netif_device_detach(netdev);
5794 if (netif_running(netdev))
5797 igb_clear_interrupt_scheme(adapter);
5800 retval = pci_save_state(pdev);
5805 status = rd32(E1000_STATUS);
5806 if (status & E1000_STATUS_LU)
5807 wufc &= ~E1000_WUFC_LNKC;
5810 igb_setup_rctl(adapter);
5811 igb_set_rx_mode(netdev);
5813 /* turn on all-multi mode if wake on multicast is enabled */
5814 if (wufc & E1000_WUFC_MC) {
5815 rctl = rd32(E1000_RCTL);
5816 rctl |= E1000_RCTL_MPE;
5817 wr32(E1000_RCTL, rctl);
5820 ctrl = rd32(E1000_CTRL);
5821 /* advertise wake from D3Cold */
5822 #define E1000_CTRL_ADVD3WUC 0x00100000
5823 /* phy power management enable */
5824 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5825 ctrl |= E1000_CTRL_ADVD3WUC;
5826 wr32(E1000_CTRL, ctrl);
5828 /* Allow time for pending master requests to run */
5829 igb_disable_pcie_master(hw);
5831 wr32(E1000_WUC, E1000_WUC_PME_EN);
5832 wr32(E1000_WUFC, wufc);
5835 wr32(E1000_WUFC, 0);
5838 *enable_wake = wufc || adapter->en_mng_pt;
5840 igb_power_down_link(adapter);
5842 igb_power_up_link(adapter);
5844 /* Release control of h/w to f/w. If f/w is AMT enabled, this
5845 * would have already happened in close and is redundant. */
5846 igb_release_hw_control(adapter);
5848 pci_disable_device(pdev);
5854 static int igb_suspend(struct pci_dev *pdev, pm_message_t state)
5859 retval = __igb_shutdown(pdev, &wake);
5864 pci_prepare_to_sleep(pdev);
5866 pci_wake_from_d3(pdev, false);
5867 pci_set_power_state(pdev, PCI_D3hot);
5873 static int igb_resume(struct pci_dev *pdev)
5875 struct net_device *netdev = pci_get_drvdata(pdev);
5876 struct igb_adapter *adapter = netdev_priv(netdev);
5877 struct e1000_hw *hw = &adapter->hw;
5880 pci_set_power_state(pdev, PCI_D0);
5881 pci_restore_state(pdev);
5882 pci_save_state(pdev);
5884 err = pci_enable_device_mem(pdev);
5887 "igb: Cannot enable PCI device from suspend\n");
5890 pci_set_master(pdev);
5892 pci_enable_wake(pdev, PCI_D3hot, 0);
5893 pci_enable_wake(pdev, PCI_D3cold, 0);
5895 if (igb_init_interrupt_scheme(adapter)) {
5896 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
5902 /* let the f/w know that the h/w is now under the control of the
5904 igb_get_hw_control(adapter);
5906 wr32(E1000_WUS, ~0);
5908 if (netif_running(netdev)) {
5909 err = igb_open(netdev);
5914 netif_device_attach(netdev);
5920 static void igb_shutdown(struct pci_dev *pdev)
5924 __igb_shutdown(pdev, &wake);
5926 if (system_state == SYSTEM_POWER_OFF) {
5927 pci_wake_from_d3(pdev, wake);
5928 pci_set_power_state(pdev, PCI_D3hot);
5932 #ifdef CONFIG_NET_POLL_CONTROLLER
5934 * Polling 'interrupt' - used by things like netconsole to send skbs
5935 * without having to re-enable interrupts. It's not called while
5936 * the interrupt routine is executing.
5938 static void igb_netpoll(struct net_device *netdev)
5940 struct igb_adapter *adapter = netdev_priv(netdev);
5941 struct e1000_hw *hw = &adapter->hw;
5944 if (!adapter->msix_entries) {
5945 struct igb_q_vector *q_vector = adapter->q_vector[0];
5946 igb_irq_disable(adapter);
5947 napi_schedule(&q_vector->napi);
5951 for (i = 0; i < adapter->num_q_vectors; i++) {
5952 struct igb_q_vector *q_vector = adapter->q_vector[i];
5953 wr32(E1000_EIMC, q_vector->eims_value);
5954 napi_schedule(&q_vector->napi);
5957 #endif /* CONFIG_NET_POLL_CONTROLLER */
5960 * igb_io_error_detected - called when PCI error is detected
5961 * @pdev: Pointer to PCI device
5962 * @state: The current pci connection state
5964 * This function is called after a PCI bus error affecting
5965 * this device has been detected.
5967 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
5968 pci_channel_state_t state)
5970 struct net_device *netdev = pci_get_drvdata(pdev);
5971 struct igb_adapter *adapter = netdev_priv(netdev);
5973 netif_device_detach(netdev);
5975 if (state == pci_channel_io_perm_failure)
5976 return PCI_ERS_RESULT_DISCONNECT;
5978 if (netif_running(netdev))
5980 pci_disable_device(pdev);
5982 /* Request a slot slot reset. */
5983 return PCI_ERS_RESULT_NEED_RESET;
5987 * igb_io_slot_reset - called after the pci bus has been reset.
5988 * @pdev: Pointer to PCI device
5990 * Restart the card from scratch, as if from a cold-boot. Implementation
5991 * resembles the first-half of the igb_resume routine.
5993 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
5995 struct net_device *netdev = pci_get_drvdata(pdev);
5996 struct igb_adapter *adapter = netdev_priv(netdev);
5997 struct e1000_hw *hw = &adapter->hw;
5998 pci_ers_result_t result;
6001 if (pci_enable_device_mem(pdev)) {
6003 "Cannot re-enable PCI device after reset.\n");
6004 result = PCI_ERS_RESULT_DISCONNECT;
6006 pci_set_master(pdev);
6007 pci_restore_state(pdev);
6008 pci_save_state(pdev);
6010 pci_enable_wake(pdev, PCI_D3hot, 0);
6011 pci_enable_wake(pdev, PCI_D3cold, 0);
6014 wr32(E1000_WUS, ~0);
6015 result = PCI_ERS_RESULT_RECOVERED;
6018 err = pci_cleanup_aer_uncorrect_error_status(pdev);
6020 dev_err(&pdev->dev, "pci_cleanup_aer_uncorrect_error_status "
6021 "failed 0x%0x\n", err);
6022 /* non-fatal, continue */
6029 * igb_io_resume - called when traffic can start flowing again.
6030 * @pdev: Pointer to PCI device
6032 * This callback is called when the error recovery driver tells us that
6033 * its OK to resume normal operation. Implementation resembles the
6034 * second-half of the igb_resume routine.
6036 static void igb_io_resume(struct pci_dev *pdev)
6038 struct net_device *netdev = pci_get_drvdata(pdev);
6039 struct igb_adapter *adapter = netdev_priv(netdev);
6041 if (netif_running(netdev)) {
6042 if (igb_up(adapter)) {
6043 dev_err(&pdev->dev, "igb_up failed after reset\n");
6048 netif_device_attach(netdev);
6050 /* let the f/w know that the h/w is now under the control of the
6052 igb_get_hw_control(adapter);
6055 static void igb_rar_set_qsel(struct igb_adapter *adapter, u8 *addr, u32 index,
6058 u32 rar_low, rar_high;
6059 struct e1000_hw *hw = &adapter->hw;
6061 /* HW expects these in little endian so we reverse the byte order
6062 * from network order (big endian) to little endian
6064 rar_low = ((u32) addr[0] | ((u32) addr[1] << 8) |
6065 ((u32) addr[2] << 16) | ((u32) addr[3] << 24));
6066 rar_high = ((u32) addr[4] | ((u32) addr[5] << 8));
6068 /* Indicate to hardware the Address is Valid. */
6069 rar_high |= E1000_RAH_AV;
6071 if (hw->mac.type == e1000_82575)
6072 rar_high |= E1000_RAH_POOL_1 * qsel;
6074 rar_high |= E1000_RAH_POOL_1 << qsel;
6076 wr32(E1000_RAL(index), rar_low);
6078 wr32(E1000_RAH(index), rar_high);
6082 static int igb_set_vf_mac(struct igb_adapter *adapter,
6083 int vf, unsigned char *mac_addr)
6085 struct e1000_hw *hw = &adapter->hw;
6086 /* VF MAC addresses start at end of receive addresses and moves
6087 * torwards the first, as a result a collision should not be possible */
6088 int rar_entry = hw->mac.rar_entry_count - (vf + 1);
6090 memcpy(adapter->vf_data[vf].vf_mac_addresses, mac_addr, ETH_ALEN);
6092 igb_rar_set_qsel(adapter, mac_addr, rar_entry, vf);
6097 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
6099 struct igb_adapter *adapter = netdev_priv(netdev);
6100 if (!is_valid_ether_addr(mac) || (vf >= adapter->vfs_allocated_count))
6102 adapter->vf_data[vf].flags |= IGB_VF_FLAG_PF_SET_MAC;
6103 dev_info(&adapter->pdev->dev, "setting MAC %pM on VF %d\n", mac, vf);
6104 dev_info(&adapter->pdev->dev, "Reload the VF driver to make this"
6105 " change effective.");
6106 if (test_bit(__IGB_DOWN, &adapter->state)) {
6107 dev_warn(&adapter->pdev->dev, "The VF MAC address has been set,"
6108 " but the PF device is not up.\n");
6109 dev_warn(&adapter->pdev->dev, "Bring the PF device up before"
6110 " attempting to use the VF device.\n");
6112 return igb_set_vf_mac(adapter, vf, mac);
6115 static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf, int tx_rate)
6120 static int igb_ndo_get_vf_config(struct net_device *netdev,
6121 int vf, struct ifla_vf_info *ivi)
6123 struct igb_adapter *adapter = netdev_priv(netdev);
6124 if (vf >= adapter->vfs_allocated_count)
6127 memcpy(&ivi->mac, adapter->vf_data[vf].vf_mac_addresses, ETH_ALEN);
6129 ivi->vlan = adapter->vf_data[vf].pf_vlan;
6130 ivi->qos = adapter->vf_data[vf].pf_qos;
6134 static void igb_vmm_control(struct igb_adapter *adapter)
6136 struct e1000_hw *hw = &adapter->hw;
6139 /* replication is not supported for 82575 */
6140 if (hw->mac.type == e1000_82575)
6143 /* enable replication vlan tag stripping */
6144 reg = rd32(E1000_RPLOLR);
6145 reg |= E1000_RPLOLR_STRVLAN;
6146 wr32(E1000_RPLOLR, reg);
6148 /* notify HW that the MAC is adding vlan tags */
6149 reg = rd32(E1000_DTXCTL);
6150 reg |= E1000_DTXCTL_VLAN_ADDED;
6151 wr32(E1000_DTXCTL, reg);
6153 if (adapter->vfs_allocated_count) {
6154 igb_vmdq_set_loopback_pf(hw, true);
6155 igb_vmdq_set_replication_pf(hw, true);
6157 igb_vmdq_set_loopback_pf(hw, false);
6158 igb_vmdq_set_replication_pf(hw, false);