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 "1.3.16-k2"
53 char igb_driver_name[] = "igb";
54 char igb_driver_version[] = DRV_VERSION;
55 static const char igb_driver_string[] =
56 "Intel(R) Gigabit Ethernet Network Driver";
57 static const char igb_copyright[] = "Copyright (c) 2007-2009 Intel Corporation.";
59 static const struct e1000_info *igb_info_tbl[] = {
60 [board_82575] = &e1000_82575_info,
63 static struct pci_device_id igb_pci_tbl[] = {
64 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576), board_82575 },
65 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS), board_82575 },
66 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_FIBER), board_82575 },
67 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES), board_82575 },
68 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES_QUAD), board_82575 },
69 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER), board_82575 },
70 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
71 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
72 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
73 /* required last entry */
77 MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
79 void igb_reset(struct igb_adapter *);
80 static int igb_setup_all_tx_resources(struct igb_adapter *);
81 static int igb_setup_all_rx_resources(struct igb_adapter *);
82 static void igb_free_all_tx_resources(struct igb_adapter *);
83 static void igb_free_all_rx_resources(struct igb_adapter *);
84 void igb_update_stats(struct igb_adapter *);
85 static int igb_probe(struct pci_dev *, const struct pci_device_id *);
86 static void __devexit igb_remove(struct pci_dev *pdev);
87 static int igb_sw_init(struct igb_adapter *);
88 static int igb_open(struct net_device *);
89 static int igb_close(struct net_device *);
90 static void igb_configure_tx(struct igb_adapter *);
91 static void igb_configure_rx(struct igb_adapter *);
92 static void igb_setup_rctl(struct igb_adapter *);
93 static void igb_clean_all_tx_rings(struct igb_adapter *);
94 static void igb_clean_all_rx_rings(struct igb_adapter *);
95 static void igb_clean_tx_ring(struct igb_ring *);
96 static void igb_clean_rx_ring(struct igb_ring *);
97 static void igb_set_rx_mode(struct net_device *);
98 static void igb_update_phy_info(unsigned long);
99 static void igb_watchdog(unsigned long);
100 static void igb_watchdog_task(struct work_struct *);
101 static netdev_tx_t igb_xmit_frame_ring_adv(struct sk_buff *,
104 static netdev_tx_t igb_xmit_frame_adv(struct sk_buff *skb,
105 struct net_device *);
106 static struct net_device_stats *igb_get_stats(struct net_device *);
107 static int igb_change_mtu(struct net_device *, int);
108 static int igb_set_mac(struct net_device *, void *);
109 static irqreturn_t igb_intr(int irq, void *);
110 static irqreturn_t igb_intr_msi(int irq, void *);
111 static irqreturn_t igb_msix_other(int irq, void *);
112 static irqreturn_t igb_msix_rx(int irq, void *);
113 static irqreturn_t igb_msix_tx(int irq, void *);
114 #ifdef CONFIG_IGB_DCA
115 static void igb_update_rx_dca(struct igb_ring *);
116 static void igb_update_tx_dca(struct igb_ring *);
117 static void igb_setup_dca(struct igb_adapter *);
118 #endif /* CONFIG_IGB_DCA */
119 static bool igb_clean_tx_irq(struct igb_ring *);
120 static int igb_poll(struct napi_struct *, int);
121 static bool igb_clean_rx_irq_adv(struct igb_ring *, int *, int);
122 static void igb_alloc_rx_buffers_adv(struct igb_ring *, 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 int igb_rcv_msg_from_vf(struct igb_adapter *, u32);
134 static void igb_vmm_control(struct igb_adapter *);
135 static int igb_set_vf_mac(struct igb_adapter *adapter, int, unsigned char *);
136 static void igb_restore_vf_multicasts(struct igb_adapter *adapter);
138 static inline void igb_set_vmolr(struct e1000_hw *hw, int vfn)
142 reg_data = rd32(E1000_VMOLR(vfn));
143 reg_data |= E1000_VMOLR_BAM | /* Accept broadcast */
144 E1000_VMOLR_ROPE | /* Accept packets matched in UTA */
145 E1000_VMOLR_ROMPE | /* Accept packets matched in MTA */
146 E1000_VMOLR_AUPE | /* Accept untagged packets */
147 E1000_VMOLR_STRVLAN; /* Strip vlan tags */
148 wr32(E1000_VMOLR(vfn), reg_data);
151 static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
154 struct e1000_hw *hw = &adapter->hw;
157 /* if it isn't the PF check to see if VFs are enabled and
158 * increase the size to support vlan tags */
159 if (vfn < adapter->vfs_allocated_count &&
160 adapter->vf_data[vfn].vlans_enabled)
161 size += VLAN_TAG_SIZE;
163 vmolr = rd32(E1000_VMOLR(vfn));
164 vmolr &= ~E1000_VMOLR_RLPML_MASK;
165 vmolr |= size | E1000_VMOLR_LPE;
166 wr32(E1000_VMOLR(vfn), vmolr);
172 static int igb_suspend(struct pci_dev *, pm_message_t);
173 static int igb_resume(struct pci_dev *);
175 static void igb_shutdown(struct pci_dev *);
176 #ifdef CONFIG_IGB_DCA
177 static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
178 static struct notifier_block dca_notifier = {
179 .notifier_call = igb_notify_dca,
184 #ifdef CONFIG_NET_POLL_CONTROLLER
185 /* for netdump / net console */
186 static void igb_netpoll(struct net_device *);
188 #ifdef CONFIG_PCI_IOV
189 static unsigned int max_vfs = 0;
190 module_param(max_vfs, uint, 0);
191 MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate "
192 "per physical function");
193 #endif /* CONFIG_PCI_IOV */
195 static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
196 pci_channel_state_t);
197 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
198 static void igb_io_resume(struct pci_dev *);
200 static struct pci_error_handlers igb_err_handler = {
201 .error_detected = igb_io_error_detected,
202 .slot_reset = igb_io_slot_reset,
203 .resume = igb_io_resume,
207 static struct pci_driver igb_driver = {
208 .name = igb_driver_name,
209 .id_table = igb_pci_tbl,
211 .remove = __devexit_p(igb_remove),
213 /* Power Managment Hooks */
214 .suspend = igb_suspend,
215 .resume = igb_resume,
217 .shutdown = igb_shutdown,
218 .err_handler = &igb_err_handler
221 static int global_quad_port_a; /* global quad port a indication */
223 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
224 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
225 MODULE_LICENSE("GPL");
226 MODULE_VERSION(DRV_VERSION);
229 * Scale the NIC clock cycle by a large factor so that
230 * relatively small clock corrections can be added or
231 * substracted at each clock tick. The drawbacks of a
232 * large factor are a) that the clock register overflows
233 * more quickly (not such a big deal) and b) that the
234 * increment per tick has to fit into 24 bits.
237 * TIMINCA = IGB_TSYNC_CYCLE_TIME_IN_NANOSECONDS *
239 * TIMINCA += TIMINCA * adjustment [ppm] / 1e9
241 * The base scale factor is intentionally a power of two
242 * so that the division in %struct timecounter can be done with
245 #define IGB_TSYNC_SHIFT (19)
246 #define IGB_TSYNC_SCALE (1<<IGB_TSYNC_SHIFT)
249 * The duration of one clock cycle of the NIC.
251 * @todo This hard-coded value is part of the specification and might change
252 * in future hardware revisions. Add revision check.
254 #define IGB_TSYNC_CYCLE_TIME_IN_NANOSECONDS 16
256 #if (IGB_TSYNC_SCALE * IGB_TSYNC_CYCLE_TIME_IN_NANOSECONDS) >= (1<<24)
257 # error IGB_TSYNC_SCALE and/or IGB_TSYNC_CYCLE_TIME_IN_NANOSECONDS are too large to fit into TIMINCA
261 * igb_read_clock - read raw cycle counter (to be used by time counter)
263 static cycle_t igb_read_clock(const struct cyclecounter *tc)
265 struct igb_adapter *adapter =
266 container_of(tc, struct igb_adapter, cycles);
267 struct e1000_hw *hw = &adapter->hw;
270 stamp = rd32(E1000_SYSTIML);
271 stamp |= (u64)rd32(E1000_SYSTIMH) << 32ULL;
278 * igb_get_hw_dev_name - return device name string
279 * used by hardware layer to print debugging information
281 char *igb_get_hw_dev_name(struct e1000_hw *hw)
283 struct igb_adapter *adapter = hw->back;
284 return adapter->netdev->name;
288 * igb_get_time_str - format current NIC and system time as string
290 static char *igb_get_time_str(struct igb_adapter *adapter,
293 cycle_t hw = adapter->cycles.read(&adapter->cycles);
294 struct timespec nic = ns_to_timespec(timecounter_read(&adapter->clock));
296 struct timespec delta;
297 getnstimeofday(&sys);
299 delta = timespec_sub(nic, sys);
302 "HW %llu, NIC %ld.%09lus, SYS %ld.%09lus, NIC-SYS %lds + %09luns",
304 (long)nic.tv_sec, nic.tv_nsec,
305 (long)sys.tv_sec, sys.tv_nsec,
306 (long)delta.tv_sec, delta.tv_nsec);
313 * igb_desc_unused - calculate if we have unused descriptors
315 static int igb_desc_unused(struct igb_ring *ring)
317 if (ring->next_to_clean > ring->next_to_use)
318 return ring->next_to_clean - ring->next_to_use - 1;
320 return ring->count + ring->next_to_clean - ring->next_to_use - 1;
324 * igb_init_module - Driver Registration Routine
326 * igb_init_module is the first routine called when the driver is
327 * loaded. All it does is register with the PCI subsystem.
329 static int __init igb_init_module(void)
332 printk(KERN_INFO "%s - version %s\n",
333 igb_driver_string, igb_driver_version);
335 printk(KERN_INFO "%s\n", igb_copyright);
337 global_quad_port_a = 0;
339 #ifdef CONFIG_IGB_DCA
340 dca_register_notify(&dca_notifier);
343 ret = pci_register_driver(&igb_driver);
347 module_init(igb_init_module);
350 * igb_exit_module - Driver Exit Cleanup Routine
352 * igb_exit_module is called just before the driver is removed
355 static void __exit igb_exit_module(void)
357 #ifdef CONFIG_IGB_DCA
358 dca_unregister_notify(&dca_notifier);
360 pci_unregister_driver(&igb_driver);
363 module_exit(igb_exit_module);
365 #define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
367 * igb_cache_ring_register - Descriptor ring to register mapping
368 * @adapter: board private structure to initialize
370 * Once we know the feature-set enabled for the device, we'll cache
371 * the register offset the descriptor ring is assigned to.
373 static void igb_cache_ring_register(struct igb_adapter *adapter)
376 unsigned int rbase_offset = adapter->vfs_allocated_count;
378 switch (adapter->hw.mac.type) {
380 /* The queues are allocated for virtualization such that VF 0
381 * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
382 * In order to avoid collision we start at the first free queue
383 * and continue consuming queues in the same sequence
385 for (i = 0; i < adapter->num_rx_queues; i++)
386 adapter->rx_ring[i].reg_idx = rbase_offset +
388 for (i = 0; i < adapter->num_tx_queues; i++)
389 adapter->tx_ring[i].reg_idx = rbase_offset +
394 for (i = 0; i < adapter->num_rx_queues; i++)
395 adapter->rx_ring[i].reg_idx = i;
396 for (i = 0; i < adapter->num_tx_queues; i++)
397 adapter->tx_ring[i].reg_idx = i;
403 * igb_alloc_queues - Allocate memory for all rings
404 * @adapter: board private structure to initialize
406 * We allocate one ring per queue at run-time since we don't know the
407 * number of queues at compile-time.
409 static int igb_alloc_queues(struct igb_adapter *adapter)
413 adapter->tx_ring = kcalloc(adapter->num_tx_queues,
414 sizeof(struct igb_ring), GFP_KERNEL);
415 if (!adapter->tx_ring)
418 adapter->rx_ring = kcalloc(adapter->num_rx_queues,
419 sizeof(struct igb_ring), GFP_KERNEL);
420 if (!adapter->rx_ring) {
421 kfree(adapter->tx_ring);
425 adapter->rx_ring->buddy = adapter->tx_ring;
427 for (i = 0; i < adapter->num_tx_queues; i++) {
428 struct igb_ring *ring = &(adapter->tx_ring[i]);
429 ring->count = adapter->tx_ring_count;
430 ring->adapter = adapter;
431 ring->queue_index = i;
433 for (i = 0; i < adapter->num_rx_queues; i++) {
434 struct igb_ring *ring = &(adapter->rx_ring[i]);
435 ring->count = adapter->rx_ring_count;
436 ring->adapter = adapter;
437 ring->queue_index = i;
438 ring->itr_register = E1000_ITR;
440 /* set a default napi handler for each rx_ring */
441 netif_napi_add(adapter->netdev, &ring->napi, igb_poll, 64);
444 igb_cache_ring_register(adapter);
448 static void igb_free_queues(struct igb_adapter *adapter)
452 for (i = 0; i < adapter->num_rx_queues; i++)
453 netif_napi_del(&adapter->rx_ring[i].napi);
455 adapter->num_rx_queues = 0;
456 adapter->num_tx_queues = 0;
458 kfree(adapter->tx_ring);
459 kfree(adapter->rx_ring);
462 #define IGB_N0_QUEUE -1
463 static void igb_assign_vector(struct igb_adapter *adapter, int rx_queue,
464 int tx_queue, int msix_vector)
467 struct e1000_hw *hw = &adapter->hw;
470 switch (hw->mac.type) {
472 /* The 82575 assigns vectors using a bitmask, which matches the
473 bitmask for the EICR/EIMS/EIMC registers. To assign one
474 or more queues to a vector, we write the appropriate bits
475 into the MSIXBM register for that vector. */
476 if (rx_queue > IGB_N0_QUEUE) {
477 msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
478 adapter->rx_ring[rx_queue].eims_value = msixbm;
480 if (tx_queue > IGB_N0_QUEUE) {
481 msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
482 adapter->tx_ring[tx_queue].eims_value =
483 E1000_EICR_TX_QUEUE0 << tx_queue;
485 array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
488 /* 82576 uses a table-based method for assigning vectors.
489 Each queue has a single entry in the table to which we write
490 a vector number along with a "valid" bit. Sadly, the layout
491 of the table is somewhat counterintuitive. */
492 if (rx_queue > IGB_N0_QUEUE) {
493 index = (rx_queue >> 1) + adapter->vfs_allocated_count;
494 ivar = array_rd32(E1000_IVAR0, index);
495 if (rx_queue & 0x1) {
496 /* vector goes into third byte of register */
497 ivar = ivar & 0xFF00FFFF;
498 ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
500 /* vector goes into low byte of register */
501 ivar = ivar & 0xFFFFFF00;
502 ivar |= msix_vector | E1000_IVAR_VALID;
504 adapter->rx_ring[rx_queue].eims_value= 1 << msix_vector;
505 array_wr32(E1000_IVAR0, index, ivar);
507 if (tx_queue > IGB_N0_QUEUE) {
508 index = (tx_queue >> 1) + adapter->vfs_allocated_count;
509 ivar = array_rd32(E1000_IVAR0, index);
510 if (tx_queue & 0x1) {
511 /* vector goes into high byte of register */
512 ivar = ivar & 0x00FFFFFF;
513 ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
515 /* vector goes into second byte of register */
516 ivar = ivar & 0xFFFF00FF;
517 ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
519 adapter->tx_ring[tx_queue].eims_value= 1 << msix_vector;
520 array_wr32(E1000_IVAR0, index, ivar);
530 * igb_configure_msix - Configure MSI-X hardware
532 * igb_configure_msix sets up the hardware to properly
533 * generate MSI-X interrupts.
535 static void igb_configure_msix(struct igb_adapter *adapter)
539 struct e1000_hw *hw = &adapter->hw;
541 adapter->eims_enable_mask = 0;
542 if (hw->mac.type == e1000_82576)
543 /* Turn on MSI-X capability first, or our settings
544 * won't stick. And it will take days to debug. */
545 wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
546 E1000_GPIE_PBA | E1000_GPIE_EIAME |
549 for (i = 0; i < adapter->num_tx_queues; i++) {
550 struct igb_ring *tx_ring = &adapter->tx_ring[i];
551 igb_assign_vector(adapter, IGB_N0_QUEUE, i, vector++);
552 adapter->eims_enable_mask |= tx_ring->eims_value;
553 if (tx_ring->itr_val)
554 writel(tx_ring->itr_val,
555 hw->hw_addr + tx_ring->itr_register);
557 writel(1, hw->hw_addr + tx_ring->itr_register);
560 for (i = 0; i < adapter->num_rx_queues; i++) {
561 struct igb_ring *rx_ring = &adapter->rx_ring[i];
562 rx_ring->buddy = NULL;
563 igb_assign_vector(adapter, i, IGB_N0_QUEUE, vector++);
564 adapter->eims_enable_mask |= rx_ring->eims_value;
565 if (rx_ring->itr_val)
566 writel(rx_ring->itr_val,
567 hw->hw_addr + rx_ring->itr_register);
569 writel(1, hw->hw_addr + rx_ring->itr_register);
573 /* set vector for other causes, i.e. link changes */
574 switch (hw->mac.type) {
576 array_wr32(E1000_MSIXBM(0), vector++,
579 tmp = rd32(E1000_CTRL_EXT);
580 /* enable MSI-X PBA support*/
581 tmp |= E1000_CTRL_EXT_PBA_CLR;
583 /* Auto-Mask interrupts upon ICR read. */
584 tmp |= E1000_CTRL_EXT_EIAME;
585 tmp |= E1000_CTRL_EXT_IRCA;
587 wr32(E1000_CTRL_EXT, tmp);
588 adapter->eims_enable_mask |= E1000_EIMS_OTHER;
589 adapter->eims_other = E1000_EIMS_OTHER;
594 tmp = (vector++ | E1000_IVAR_VALID) << 8;
595 wr32(E1000_IVAR_MISC, tmp);
597 adapter->eims_enable_mask = (1 << (vector)) - 1;
598 adapter->eims_other = 1 << (vector - 1);
601 /* do nothing, since nothing else supports MSI-X */
603 } /* switch (hw->mac.type) */
608 * igb_request_msix - Initialize MSI-X interrupts
610 * igb_request_msix allocates MSI-X vectors and requests interrupts from the
613 static int igb_request_msix(struct igb_adapter *adapter)
615 struct net_device *netdev = adapter->netdev;
616 int i, err = 0, vector = 0;
620 for (i = 0; i < adapter->num_tx_queues; i++) {
621 struct igb_ring *ring = &(adapter->tx_ring[i]);
622 sprintf(ring->name, "%s-tx-%d", netdev->name, i);
623 err = request_irq(adapter->msix_entries[vector].vector,
624 &igb_msix_tx, 0, ring->name,
625 &(adapter->tx_ring[i]));
628 ring->itr_register = E1000_EITR(0) + (vector << 2);
629 ring->itr_val = 976; /* ~4000 ints/sec */
632 for (i = 0; i < adapter->num_rx_queues; i++) {
633 struct igb_ring *ring = &(adapter->rx_ring[i]);
634 if (strlen(netdev->name) < (IFNAMSIZ - 5))
635 sprintf(ring->name, "%s-rx-%d", netdev->name, i);
637 memcpy(ring->name, netdev->name, IFNAMSIZ);
638 err = request_irq(adapter->msix_entries[vector].vector,
639 &igb_msix_rx, 0, ring->name,
640 &(adapter->rx_ring[i]));
643 ring->itr_register = E1000_EITR(0) + (vector << 2);
644 ring->itr_val = adapter->itr;
648 err = request_irq(adapter->msix_entries[vector].vector,
649 &igb_msix_other, 0, netdev->name, netdev);
653 igb_configure_msix(adapter);
659 static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
661 if (adapter->msix_entries) {
662 pci_disable_msix(adapter->pdev);
663 kfree(adapter->msix_entries);
664 adapter->msix_entries = NULL;
665 } else if (adapter->flags & IGB_FLAG_HAS_MSI)
666 pci_disable_msi(adapter->pdev);
672 * igb_set_interrupt_capability - set MSI or MSI-X if supported
674 * Attempt to configure interrupts using the best available
675 * capabilities of the hardware and kernel.
677 static void igb_set_interrupt_capability(struct igb_adapter *adapter)
682 /* Number of supported queues. */
683 /* Having more queues than CPUs doesn't make sense. */
684 adapter->num_rx_queues = min_t(u32, IGB_MAX_RX_QUEUES, num_online_cpus());
685 adapter->num_tx_queues = min_t(u32, IGB_MAX_TX_QUEUES, num_online_cpus());
687 numvecs = adapter->num_tx_queues + adapter->num_rx_queues + 1;
688 adapter->msix_entries = kcalloc(numvecs, sizeof(struct msix_entry),
690 if (!adapter->msix_entries)
693 for (i = 0; i < numvecs; i++)
694 adapter->msix_entries[i].entry = i;
696 err = pci_enable_msix(adapter->pdev,
697 adapter->msix_entries,
702 igb_reset_interrupt_capability(adapter);
704 /* If we can't do MSI-X, try MSI */
706 #ifdef CONFIG_PCI_IOV
707 /* disable SR-IOV for non MSI-X configurations */
708 if (adapter->vf_data) {
709 struct e1000_hw *hw = &adapter->hw;
710 /* disable iov and allow time for transactions to clear */
711 pci_disable_sriov(adapter->pdev);
714 kfree(adapter->vf_data);
715 adapter->vf_data = NULL;
716 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
718 dev_info(&adapter->pdev->dev, "IOV Disabled\n");
721 adapter->num_rx_queues = 1;
722 adapter->num_tx_queues = 1;
723 if (!pci_enable_msi(adapter->pdev))
724 adapter->flags |= IGB_FLAG_HAS_MSI;
726 /* Notify the stack of the (possibly) reduced Tx Queue count. */
727 adapter->netdev->real_num_tx_queues = adapter->num_tx_queues;
732 * igb_request_irq - initialize interrupts
734 * Attempts to configure interrupts using the best available
735 * capabilities of the hardware and kernel.
737 static int igb_request_irq(struct igb_adapter *adapter)
739 struct net_device *netdev = adapter->netdev;
740 struct e1000_hw *hw = &adapter->hw;
743 if (adapter->msix_entries) {
744 err = igb_request_msix(adapter);
747 /* fall back to MSI */
748 igb_reset_interrupt_capability(adapter);
749 if (!pci_enable_msi(adapter->pdev))
750 adapter->flags |= IGB_FLAG_HAS_MSI;
751 igb_free_all_tx_resources(adapter);
752 igb_free_all_rx_resources(adapter);
753 adapter->num_rx_queues = 1;
754 igb_alloc_queues(adapter);
756 switch (hw->mac.type) {
758 wr32(E1000_MSIXBM(0),
759 (E1000_EICR_RX_QUEUE0 | E1000_EIMS_OTHER));
762 wr32(E1000_IVAR0, E1000_IVAR_VALID);
769 if (adapter->flags & IGB_FLAG_HAS_MSI) {
770 err = request_irq(adapter->pdev->irq, &igb_intr_msi, 0,
771 netdev->name, netdev);
774 /* fall back to legacy interrupts */
775 igb_reset_interrupt_capability(adapter);
776 adapter->flags &= ~IGB_FLAG_HAS_MSI;
779 err = request_irq(adapter->pdev->irq, &igb_intr, IRQF_SHARED,
780 netdev->name, netdev);
783 dev_err(&adapter->pdev->dev, "Error %d getting interrupt\n",
790 static void igb_free_irq(struct igb_adapter *adapter)
792 struct net_device *netdev = adapter->netdev;
794 if (adapter->msix_entries) {
797 for (i = 0; i < adapter->num_tx_queues; i++)
798 free_irq(adapter->msix_entries[vector++].vector,
799 &(adapter->tx_ring[i]));
800 for (i = 0; i < adapter->num_rx_queues; i++)
801 free_irq(adapter->msix_entries[vector++].vector,
802 &(adapter->rx_ring[i]));
804 free_irq(adapter->msix_entries[vector++].vector, netdev);
808 free_irq(adapter->pdev->irq, netdev);
812 * igb_irq_disable - Mask off interrupt generation on the NIC
813 * @adapter: board private structure
815 static void igb_irq_disable(struct igb_adapter *adapter)
817 struct e1000_hw *hw = &adapter->hw;
819 if (adapter->msix_entries) {
820 u32 regval = rd32(E1000_EIAM);
821 wr32(E1000_EIAM, regval & ~adapter->eims_enable_mask);
822 wr32(E1000_EIMC, adapter->eims_enable_mask);
823 regval = rd32(E1000_EIAC);
824 wr32(E1000_EIAC, regval & ~adapter->eims_enable_mask);
830 synchronize_irq(adapter->pdev->irq);
834 * igb_irq_enable - Enable default interrupt generation settings
835 * @adapter: board private structure
837 static void igb_irq_enable(struct igb_adapter *adapter)
839 struct e1000_hw *hw = &adapter->hw;
841 if (adapter->msix_entries) {
842 u32 regval = rd32(E1000_EIAC);
843 wr32(E1000_EIAC, regval | adapter->eims_enable_mask);
844 regval = rd32(E1000_EIAM);
845 wr32(E1000_EIAM, regval | adapter->eims_enable_mask);
846 wr32(E1000_EIMS, adapter->eims_enable_mask);
847 if (adapter->vfs_allocated_count)
848 wr32(E1000_MBVFIMR, 0xFF);
849 wr32(E1000_IMS, (E1000_IMS_LSC | E1000_IMS_VMMB |
850 E1000_IMS_DOUTSYNC));
852 wr32(E1000_IMS, IMS_ENABLE_MASK);
853 wr32(E1000_IAM, IMS_ENABLE_MASK);
857 static void igb_update_mng_vlan(struct igb_adapter *adapter)
859 struct net_device *netdev = adapter->netdev;
860 u16 vid = adapter->hw.mng_cookie.vlan_id;
861 u16 old_vid = adapter->mng_vlan_id;
862 if (adapter->vlgrp) {
863 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
864 if (adapter->hw.mng_cookie.status &
865 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
866 igb_vlan_rx_add_vid(netdev, vid);
867 adapter->mng_vlan_id = vid;
869 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
871 if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
873 !vlan_group_get_device(adapter->vlgrp, old_vid))
874 igb_vlan_rx_kill_vid(netdev, old_vid);
876 adapter->mng_vlan_id = vid;
881 * igb_release_hw_control - release control of the h/w to f/w
882 * @adapter: address of board private structure
884 * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
885 * For ASF and Pass Through versions of f/w this means that the
886 * driver is no longer loaded.
889 static void igb_release_hw_control(struct igb_adapter *adapter)
891 struct e1000_hw *hw = &adapter->hw;
894 /* Let firmware take over control of h/w */
895 ctrl_ext = rd32(E1000_CTRL_EXT);
897 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
902 * igb_get_hw_control - get control of the h/w from f/w
903 * @adapter: address of board private structure
905 * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
906 * For ASF and Pass Through versions of f/w this means that
907 * the driver is loaded.
910 static void igb_get_hw_control(struct igb_adapter *adapter)
912 struct e1000_hw *hw = &adapter->hw;
915 /* Let firmware know the driver has taken over */
916 ctrl_ext = rd32(E1000_CTRL_EXT);
918 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
922 * igb_configure - configure the hardware for RX and TX
923 * @adapter: private board structure
925 static void igb_configure(struct igb_adapter *adapter)
927 struct net_device *netdev = adapter->netdev;
930 igb_get_hw_control(adapter);
931 igb_set_rx_mode(netdev);
933 igb_restore_vlan(adapter);
935 igb_configure_tx(adapter);
936 igb_setup_rctl(adapter);
937 igb_configure_rx(adapter);
939 igb_rx_fifo_flush_82575(&adapter->hw);
941 /* call igb_desc_unused which always leaves
942 * at least 1 descriptor unused to make sure
943 * next_to_use != next_to_clean */
944 for (i = 0; i < adapter->num_rx_queues; i++) {
945 struct igb_ring *ring = &adapter->rx_ring[i];
946 igb_alloc_rx_buffers_adv(ring, igb_desc_unused(ring));
950 adapter->tx_queue_len = netdev->tx_queue_len;
955 * igb_up - Open the interface and prepare it to handle traffic
956 * @adapter: board private structure
959 int igb_up(struct igb_adapter *adapter)
961 struct e1000_hw *hw = &adapter->hw;
964 /* hardware has been reset, we need to reload some things */
965 igb_configure(adapter);
967 clear_bit(__IGB_DOWN, &adapter->state);
969 for (i = 0; i < adapter->num_rx_queues; i++)
970 napi_enable(&adapter->rx_ring[i].napi);
971 if (adapter->msix_entries)
972 igb_configure_msix(adapter);
974 igb_vmm_control(adapter);
975 igb_set_vmolr(hw, adapter->vfs_allocated_count);
977 /* Clear any pending interrupts. */
979 igb_irq_enable(adapter);
981 netif_tx_start_all_queues(adapter->netdev);
983 /* Fire a link change interrupt to start the watchdog. */
984 wr32(E1000_ICS, E1000_ICS_LSC);
988 void igb_down(struct igb_adapter *adapter)
990 struct e1000_hw *hw = &adapter->hw;
991 struct net_device *netdev = adapter->netdev;
995 /* signal that we're down so the interrupt handler does not
996 * reschedule our watchdog timer */
997 set_bit(__IGB_DOWN, &adapter->state);
999 /* disable receives in the hardware */
1000 rctl = rd32(E1000_RCTL);
1001 wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
1002 /* flush and sleep below */
1004 netif_tx_stop_all_queues(netdev);
1006 /* disable transmits in the hardware */
1007 tctl = rd32(E1000_TCTL);
1008 tctl &= ~E1000_TCTL_EN;
1009 wr32(E1000_TCTL, tctl);
1010 /* flush both disables and wait for them to finish */
1014 for (i = 0; i < adapter->num_rx_queues; i++)
1015 napi_disable(&adapter->rx_ring[i].napi);
1017 igb_irq_disable(adapter);
1019 del_timer_sync(&adapter->watchdog_timer);
1020 del_timer_sync(&adapter->phy_info_timer);
1022 netdev->tx_queue_len = adapter->tx_queue_len;
1023 netif_carrier_off(netdev);
1025 /* record the stats before reset*/
1026 igb_update_stats(adapter);
1028 adapter->link_speed = 0;
1029 adapter->link_duplex = 0;
1031 if (!pci_channel_offline(adapter->pdev))
1033 igb_clean_all_tx_rings(adapter);
1034 igb_clean_all_rx_rings(adapter);
1035 #ifdef CONFIG_IGB_DCA
1037 /* since we reset the hardware DCA settings were cleared */
1038 igb_setup_dca(adapter);
1042 void igb_reinit_locked(struct igb_adapter *adapter)
1044 WARN_ON(in_interrupt());
1045 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
1049 clear_bit(__IGB_RESETTING, &adapter->state);
1052 void igb_reset(struct igb_adapter *adapter)
1054 struct e1000_hw *hw = &adapter->hw;
1055 struct e1000_mac_info *mac = &hw->mac;
1056 struct e1000_fc_info *fc = &hw->fc;
1057 u32 pba = 0, tx_space, min_tx_space, min_rx_space;
1060 /* Repartition Pba for greater than 9k mtu
1061 * To take effect CTRL.RST is required.
1063 switch (mac->type) {
1065 pba = E1000_PBA_64K;
1069 pba = E1000_PBA_34K;
1073 if ((adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) &&
1074 (mac->type < e1000_82576)) {
1075 /* adjust PBA for jumbo frames */
1076 wr32(E1000_PBA, pba);
1078 /* To maintain wire speed transmits, the Tx FIFO should be
1079 * large enough to accommodate two full transmit packets,
1080 * rounded up to the next 1KB and expressed in KB. Likewise,
1081 * the Rx FIFO should be large enough to accommodate at least
1082 * one full receive packet and is similarly rounded up and
1083 * expressed in KB. */
1084 pba = rd32(E1000_PBA);
1085 /* upper 16 bits has Tx packet buffer allocation size in KB */
1086 tx_space = pba >> 16;
1087 /* lower 16 bits has Rx packet buffer allocation size in KB */
1089 /* the tx fifo also stores 16 bytes of information about the tx
1090 * but don't include ethernet FCS because hardware appends it */
1091 min_tx_space = (adapter->max_frame_size +
1092 sizeof(union e1000_adv_tx_desc) -
1094 min_tx_space = ALIGN(min_tx_space, 1024);
1095 min_tx_space >>= 10;
1096 /* software strips receive CRC, so leave room for it */
1097 min_rx_space = adapter->max_frame_size;
1098 min_rx_space = ALIGN(min_rx_space, 1024);
1099 min_rx_space >>= 10;
1101 /* If current Tx allocation is less than the min Tx FIFO size,
1102 * and the min Tx FIFO size is less than the current Rx FIFO
1103 * allocation, take space away from current Rx allocation */
1104 if (tx_space < min_tx_space &&
1105 ((min_tx_space - tx_space) < pba)) {
1106 pba = pba - (min_tx_space - tx_space);
1108 /* if short on rx space, rx wins and must trump tx
1110 if (pba < min_rx_space)
1113 wr32(E1000_PBA, pba);
1116 /* flow control settings */
1117 /* The high water mark must be low enough to fit one full frame
1118 * (or the size used for early receive) above it in the Rx FIFO.
1119 * Set it to the lower of:
1120 * - 90% of the Rx FIFO size, or
1121 * - the full Rx FIFO size minus one full frame */
1122 hwm = min(((pba << 10) * 9 / 10),
1123 ((pba << 10) - 2 * adapter->max_frame_size));
1125 if (mac->type < e1000_82576) {
1126 fc->high_water = hwm & 0xFFF8; /* 8-byte granularity */
1127 fc->low_water = fc->high_water - 8;
1129 fc->high_water = hwm & 0xFFF0; /* 16-byte granularity */
1130 fc->low_water = fc->high_water - 16;
1132 fc->pause_time = 0xFFFF;
1134 fc->current_mode = fc->requested_mode;
1136 /* disable receive for all VFs and wait one second */
1137 if (adapter->vfs_allocated_count) {
1139 for (i = 0 ; i < adapter->vfs_allocated_count; i++)
1140 adapter->vf_data[i].clear_to_send = false;
1142 /* ping all the active vfs to let them know we are going down */
1143 igb_ping_all_vfs(adapter);
1145 /* disable transmits and receives */
1146 wr32(E1000_VFRE, 0);
1147 wr32(E1000_VFTE, 0);
1150 /* Allow time for pending master requests to run */
1151 adapter->hw.mac.ops.reset_hw(&adapter->hw);
1154 if (adapter->hw.mac.ops.init_hw(&adapter->hw))
1155 dev_err(&adapter->pdev->dev, "Hardware Error\n");
1157 igb_update_mng_vlan(adapter);
1159 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
1160 wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
1162 igb_reset_adaptive(&adapter->hw);
1163 igb_get_phy_info(&adapter->hw);
1166 static const struct net_device_ops igb_netdev_ops = {
1167 .ndo_open = igb_open,
1168 .ndo_stop = igb_close,
1169 .ndo_start_xmit = igb_xmit_frame_adv,
1170 .ndo_get_stats = igb_get_stats,
1171 .ndo_set_rx_mode = igb_set_rx_mode,
1172 .ndo_set_multicast_list = igb_set_rx_mode,
1173 .ndo_set_mac_address = igb_set_mac,
1174 .ndo_change_mtu = igb_change_mtu,
1175 .ndo_do_ioctl = igb_ioctl,
1176 .ndo_tx_timeout = igb_tx_timeout,
1177 .ndo_validate_addr = eth_validate_addr,
1178 .ndo_vlan_rx_register = igb_vlan_rx_register,
1179 .ndo_vlan_rx_add_vid = igb_vlan_rx_add_vid,
1180 .ndo_vlan_rx_kill_vid = igb_vlan_rx_kill_vid,
1181 #ifdef CONFIG_NET_POLL_CONTROLLER
1182 .ndo_poll_controller = igb_netpoll,
1187 * igb_probe - Device Initialization Routine
1188 * @pdev: PCI device information struct
1189 * @ent: entry in igb_pci_tbl
1191 * Returns 0 on success, negative on failure
1193 * igb_probe initializes an adapter identified by a pci_dev structure.
1194 * The OS initialization, configuring of the adapter private structure,
1195 * and a hardware reset occur.
1197 static int __devinit igb_probe(struct pci_dev *pdev,
1198 const struct pci_device_id *ent)
1200 struct net_device *netdev;
1201 struct igb_adapter *adapter;
1202 struct e1000_hw *hw;
1203 const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
1204 unsigned long mmio_start, mmio_len;
1205 int err, pci_using_dac;
1206 u16 eeprom_data = 0;
1207 u16 eeprom_apme_mask = IGB_EEPROM_APME;
1210 err = pci_enable_device_mem(pdev);
1215 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
1217 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
1221 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
1223 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
1225 dev_err(&pdev->dev, "No usable DMA "
1226 "configuration, aborting\n");
1232 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
1238 pci_enable_pcie_error_reporting(pdev);
1240 pci_set_master(pdev);
1241 pci_save_state(pdev);
1244 netdev = alloc_etherdev_mq(sizeof(struct igb_adapter),
1245 IGB_ABS_MAX_TX_QUEUES);
1247 goto err_alloc_etherdev;
1249 SET_NETDEV_DEV(netdev, &pdev->dev);
1251 pci_set_drvdata(pdev, netdev);
1252 adapter = netdev_priv(netdev);
1253 adapter->netdev = netdev;
1254 adapter->pdev = pdev;
1257 adapter->msg_enable = NETIF_MSG_DRV | NETIF_MSG_PROBE;
1259 mmio_start = pci_resource_start(pdev, 0);
1260 mmio_len = pci_resource_len(pdev, 0);
1263 hw->hw_addr = ioremap(mmio_start, mmio_len);
1267 netdev->netdev_ops = &igb_netdev_ops;
1268 igb_set_ethtool_ops(netdev);
1269 netdev->watchdog_timeo = 5 * HZ;
1271 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
1273 netdev->mem_start = mmio_start;
1274 netdev->mem_end = mmio_start + mmio_len;
1276 /* PCI config space info */
1277 hw->vendor_id = pdev->vendor;
1278 hw->device_id = pdev->device;
1279 hw->revision_id = pdev->revision;
1280 hw->subsystem_vendor_id = pdev->subsystem_vendor;
1281 hw->subsystem_device_id = pdev->subsystem_device;
1283 /* setup the private structure */
1285 /* Copy the default MAC, PHY and NVM function pointers */
1286 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
1287 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
1288 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
1289 /* Initialize skew-specific constants */
1290 err = ei->get_invariants(hw);
1294 #ifdef CONFIG_PCI_IOV
1295 /* since iov functionality isn't critical to base device function we
1296 * can accept failure. If it fails we don't allow iov to be enabled */
1297 if (hw->mac.type == e1000_82576) {
1298 /* 82576 supports a maximum of 7 VFs in addition to the PF */
1299 unsigned int num_vfs = (max_vfs > 7) ? 7 : max_vfs;
1301 unsigned char mac_addr[ETH_ALEN];
1304 adapter->vf_data = kcalloc(num_vfs,
1305 sizeof(struct vf_data_storage),
1307 if (!adapter->vf_data) {
1309 "Could not allocate VF private data - "
1310 "IOV enable failed\n");
1312 err = pci_enable_sriov(pdev, num_vfs);
1314 adapter->vfs_allocated_count = num_vfs;
1315 dev_info(&pdev->dev,
1316 "%d vfs allocated\n",
1319 i < adapter->vfs_allocated_count;
1321 random_ether_addr(mac_addr);
1322 igb_set_vf_mac(adapter, i,
1326 kfree(adapter->vf_data);
1327 adapter->vf_data = NULL;
1334 /* setup the private structure */
1335 err = igb_sw_init(adapter);
1339 igb_get_bus_info_pcie(hw);
1342 switch (hw->mac.type) {
1344 adapter->flags |= IGB_FLAG_NEED_CTX_IDX;
1351 hw->phy.autoneg_wait_to_complete = false;
1352 hw->mac.adaptive_ifs = true;
1354 /* Copper options */
1355 if (hw->phy.media_type == e1000_media_type_copper) {
1356 hw->phy.mdix = AUTO_ALL_MODES;
1357 hw->phy.disable_polarity_correction = false;
1358 hw->phy.ms_type = e1000_ms_hw_default;
1361 if (igb_check_reset_block(hw))
1362 dev_info(&pdev->dev,
1363 "PHY reset is blocked due to SOL/IDER session.\n");
1365 netdev->features = NETIF_F_SG |
1367 NETIF_F_HW_VLAN_TX |
1368 NETIF_F_HW_VLAN_RX |
1369 NETIF_F_HW_VLAN_FILTER;
1371 netdev->features |= NETIF_F_IPV6_CSUM;
1372 netdev->features |= NETIF_F_TSO;
1373 netdev->features |= NETIF_F_TSO6;
1375 netdev->features |= NETIF_F_GRO;
1377 netdev->vlan_features |= NETIF_F_TSO;
1378 netdev->vlan_features |= NETIF_F_TSO6;
1379 netdev->vlan_features |= NETIF_F_IP_CSUM;
1380 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
1381 netdev->vlan_features |= NETIF_F_SG;
1384 netdev->features |= NETIF_F_HIGHDMA;
1386 if (adapter->hw.mac.type == e1000_82576)
1387 netdev->features |= NETIF_F_SCTP_CSUM;
1389 adapter->en_mng_pt = igb_enable_mng_pass_thru(&adapter->hw);
1391 /* before reading the NVM, reset the controller to put the device in a
1392 * known good starting state */
1393 hw->mac.ops.reset_hw(hw);
1395 /* make sure the NVM is good */
1396 if (igb_validate_nvm_checksum(hw) < 0) {
1397 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
1402 /* copy the MAC address out of the NVM */
1403 if (hw->mac.ops.read_mac_addr(hw))
1404 dev_err(&pdev->dev, "NVM Read Error\n");
1406 memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
1407 memcpy(netdev->perm_addr, hw->mac.addr, netdev->addr_len);
1409 if (!is_valid_ether_addr(netdev->perm_addr)) {
1410 dev_err(&pdev->dev, "Invalid MAC Address\n");
1415 setup_timer(&adapter->watchdog_timer, &igb_watchdog,
1416 (unsigned long) adapter);
1417 setup_timer(&adapter->phy_info_timer, &igb_update_phy_info,
1418 (unsigned long) adapter);
1420 INIT_WORK(&adapter->reset_task, igb_reset_task);
1421 INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
1423 /* Initialize link properties that are user-changeable */
1424 adapter->fc_autoneg = true;
1425 hw->mac.autoneg = true;
1426 hw->phy.autoneg_advertised = 0x2f;
1428 hw->fc.requested_mode = e1000_fc_default;
1429 hw->fc.current_mode = e1000_fc_default;
1431 adapter->itr_setting = IGB_DEFAULT_ITR;
1432 adapter->itr = IGB_START_ITR;
1434 igb_validate_mdi_setting(hw);
1436 /* Initial Wake on LAN setting If APM wake is enabled in the EEPROM,
1437 * enable the ACPI Magic Packet filter
1440 if (hw->bus.func == 0)
1441 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
1442 else if (hw->bus.func == 1)
1443 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
1445 if (eeprom_data & eeprom_apme_mask)
1446 adapter->eeprom_wol |= E1000_WUFC_MAG;
1448 /* now that we have the eeprom settings, apply the special cases where
1449 * the eeprom may be wrong or the board simply won't support wake on
1450 * lan on a particular port */
1451 switch (pdev->device) {
1452 case E1000_DEV_ID_82575GB_QUAD_COPPER:
1453 adapter->eeprom_wol = 0;
1455 case E1000_DEV_ID_82575EB_FIBER_SERDES:
1456 case E1000_DEV_ID_82576_FIBER:
1457 case E1000_DEV_ID_82576_SERDES:
1458 /* Wake events only supported on port A for dual fiber
1459 * regardless of eeprom setting */
1460 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
1461 adapter->eeprom_wol = 0;
1463 case E1000_DEV_ID_82576_QUAD_COPPER:
1464 /* if quad port adapter, disable WoL on all but port A */
1465 if (global_quad_port_a != 0)
1466 adapter->eeprom_wol = 0;
1468 adapter->flags |= IGB_FLAG_QUAD_PORT_A;
1469 /* Reset for multiple quad port adapters */
1470 if (++global_quad_port_a == 4)
1471 global_quad_port_a = 0;
1475 /* initialize the wol settings based on the eeprom settings */
1476 adapter->wol = adapter->eeprom_wol;
1477 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
1479 /* reset the hardware with the new settings */
1482 /* let the f/w know that the h/w is now under the control of the
1484 igb_get_hw_control(adapter);
1486 strcpy(netdev->name, "eth%d");
1487 err = register_netdev(netdev);
1491 /* carrier off reporting is important to ethtool even BEFORE open */
1492 netif_carrier_off(netdev);
1494 #ifdef CONFIG_IGB_DCA
1495 if (dca_add_requester(&pdev->dev) == 0) {
1496 adapter->flags |= IGB_FLAG_DCA_ENABLED;
1497 dev_info(&pdev->dev, "DCA enabled\n");
1498 igb_setup_dca(adapter);
1503 * Initialize hardware timer: we keep it running just in case
1504 * that some program needs it later on.
1506 memset(&adapter->cycles, 0, sizeof(adapter->cycles));
1507 adapter->cycles.read = igb_read_clock;
1508 adapter->cycles.mask = CLOCKSOURCE_MASK(64);
1509 adapter->cycles.mult = 1;
1510 adapter->cycles.shift = IGB_TSYNC_SHIFT;
1513 IGB_TSYNC_CYCLE_TIME_IN_NANOSECONDS * IGB_TSYNC_SCALE);
1516 * Avoid rollover while we initialize by resetting the time counter.
1518 wr32(E1000_SYSTIML, 0x00000000);
1519 wr32(E1000_SYSTIMH, 0x00000000);
1522 * Set registers so that rollover occurs soon to test this.
1524 wr32(E1000_SYSTIML, 0x00000000);
1525 wr32(E1000_SYSTIMH, 0xFF800000);
1528 timecounter_init(&adapter->clock,
1530 ktime_to_ns(ktime_get_real()));
1533 * Synchronize our NIC clock against system wall clock. NIC
1534 * time stamp reading requires ~3us per sample, each sample
1535 * was pretty stable even under load => only require 10
1536 * samples for each offset comparison.
1538 memset(&adapter->compare, 0, sizeof(adapter->compare));
1539 adapter->compare.source = &adapter->clock;
1540 adapter->compare.target = ktime_get_real;
1541 adapter->compare.num_samples = 10;
1542 timecompare_update(&adapter->compare, 0);
1548 "igb: %s: hw %p initialized timer\n",
1549 igb_get_time_str(adapter, buffer),
1554 dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
1555 /* print bus type/speed/width info */
1556 dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
1558 ((hw->bus.speed == e1000_bus_speed_2500)
1559 ? "2.5Gb/s" : "unknown"),
1560 ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
1561 (hw->bus.width == e1000_bus_width_pcie_x2) ? "Width x2" :
1562 (hw->bus.width == e1000_bus_width_pcie_x1) ? "Width x1" :
1566 igb_read_part_num(hw, &part_num);
1567 dev_info(&pdev->dev, "%s: PBA No: %06x-%03x\n", netdev->name,
1568 (part_num >> 8), (part_num & 0xff));
1570 dev_info(&pdev->dev,
1571 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
1572 adapter->msix_entries ? "MSI-X" :
1573 (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
1574 adapter->num_rx_queues, adapter->num_tx_queues);
1579 igb_release_hw_control(adapter);
1581 if (!igb_check_reset_block(hw))
1584 if (hw->flash_address)
1585 iounmap(hw->flash_address);
1587 igb_free_queues(adapter);
1589 iounmap(hw->hw_addr);
1591 free_netdev(netdev);
1593 pci_release_selected_regions(pdev, pci_select_bars(pdev,
1597 pci_disable_device(pdev);
1602 * igb_remove - Device Removal Routine
1603 * @pdev: PCI device information struct
1605 * igb_remove is called by the PCI subsystem to alert the driver
1606 * that it should release a PCI device. The could be caused by a
1607 * Hot-Plug event, or because the driver is going to be removed from
1610 static void __devexit igb_remove(struct pci_dev *pdev)
1612 struct net_device *netdev = pci_get_drvdata(pdev);
1613 struct igb_adapter *adapter = netdev_priv(netdev);
1614 struct e1000_hw *hw = &adapter->hw;
1616 /* flush_scheduled work may reschedule our watchdog task, so
1617 * explicitly disable watchdog tasks from being rescheduled */
1618 set_bit(__IGB_DOWN, &adapter->state);
1619 del_timer_sync(&adapter->watchdog_timer);
1620 del_timer_sync(&adapter->phy_info_timer);
1622 flush_scheduled_work();
1624 #ifdef CONFIG_IGB_DCA
1625 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
1626 dev_info(&pdev->dev, "DCA disabled\n");
1627 dca_remove_requester(&pdev->dev);
1628 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
1629 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
1633 /* Release control of h/w to f/w. If f/w is AMT enabled, this
1634 * would have already happened in close and is redundant. */
1635 igb_release_hw_control(adapter);
1637 unregister_netdev(netdev);
1639 if (!igb_check_reset_block(&adapter->hw))
1640 igb_reset_phy(&adapter->hw);
1642 igb_reset_interrupt_capability(adapter);
1644 igb_free_queues(adapter);
1646 #ifdef CONFIG_PCI_IOV
1647 /* reclaim resources allocated to VFs */
1648 if (adapter->vf_data) {
1649 /* disable iov and allow time for transactions to clear */
1650 pci_disable_sriov(pdev);
1653 kfree(adapter->vf_data);
1654 adapter->vf_data = NULL;
1655 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
1657 dev_info(&pdev->dev, "IOV Disabled\n");
1660 iounmap(hw->hw_addr);
1661 if (hw->flash_address)
1662 iounmap(hw->flash_address);
1663 pci_release_selected_regions(pdev, pci_select_bars(pdev,
1666 free_netdev(netdev);
1668 pci_disable_pcie_error_reporting(pdev);
1670 pci_disable_device(pdev);
1674 * igb_sw_init - Initialize general software structures (struct igb_adapter)
1675 * @adapter: board private structure to initialize
1677 * igb_sw_init initializes the Adapter private data structure.
1678 * Fields are initialized based on PCI device information and
1679 * OS network device settings (MTU size).
1681 static int __devinit igb_sw_init(struct igb_adapter *adapter)
1683 struct e1000_hw *hw = &adapter->hw;
1684 struct net_device *netdev = adapter->netdev;
1685 struct pci_dev *pdev = adapter->pdev;
1687 pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
1689 adapter->tx_ring_count = IGB_DEFAULT_TXD;
1690 adapter->rx_ring_count = IGB_DEFAULT_RXD;
1691 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1692 adapter->rx_ps_hdr_size = 0; /* disable packet split */
1693 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1694 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1696 /* This call may decrease the number of queues depending on
1697 * interrupt mode. */
1698 igb_set_interrupt_capability(adapter);
1700 if (igb_alloc_queues(adapter)) {
1701 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
1705 /* Explicitly disable IRQ since the NIC can be in any state. */
1706 igb_irq_disable(adapter);
1708 set_bit(__IGB_DOWN, &adapter->state);
1713 * igb_open - Called when a network interface is made active
1714 * @netdev: network interface device structure
1716 * Returns 0 on success, negative value on failure
1718 * The open entry point is called when a network interface is made
1719 * active by the system (IFF_UP). At this point all resources needed
1720 * for transmit and receive operations are allocated, the interrupt
1721 * handler is registered with the OS, the watchdog timer is started,
1722 * and the stack is notified that the interface is ready.
1724 static int igb_open(struct net_device *netdev)
1726 struct igb_adapter *adapter = netdev_priv(netdev);
1727 struct e1000_hw *hw = &adapter->hw;
1731 /* disallow open during test */
1732 if (test_bit(__IGB_TESTING, &adapter->state))
1735 netif_carrier_off(netdev);
1737 /* allocate transmit descriptors */
1738 err = igb_setup_all_tx_resources(adapter);
1742 /* allocate receive descriptors */
1743 err = igb_setup_all_rx_resources(adapter);
1747 /* e1000_power_up_phy(adapter); */
1749 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1750 if ((adapter->hw.mng_cookie.status &
1751 E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
1752 igb_update_mng_vlan(adapter);
1754 /* before we allocate an interrupt, we must be ready to handle it.
1755 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1756 * as soon as we call pci_request_irq, so we have to setup our
1757 * clean_rx handler before we do so. */
1758 igb_configure(adapter);
1760 igb_vmm_control(adapter);
1761 igb_set_vmolr(hw, adapter->vfs_allocated_count);
1763 err = igb_request_irq(adapter);
1767 /* From here on the code is the same as igb_up() */
1768 clear_bit(__IGB_DOWN, &adapter->state);
1770 for (i = 0; i < adapter->num_rx_queues; i++)
1771 napi_enable(&adapter->rx_ring[i].napi);
1773 /* Clear any pending interrupts. */
1776 igb_irq_enable(adapter);
1778 netif_tx_start_all_queues(netdev);
1780 /* Fire a link status change interrupt to start the watchdog. */
1781 wr32(E1000_ICS, E1000_ICS_LSC);
1786 igb_release_hw_control(adapter);
1787 /* e1000_power_down_phy(adapter); */
1788 igb_free_all_rx_resources(adapter);
1790 igb_free_all_tx_resources(adapter);
1798 * igb_close - Disables a network interface
1799 * @netdev: network interface device structure
1801 * Returns 0, this is not allowed to fail
1803 * The close entry point is called when an interface is de-activated
1804 * by the OS. The hardware is still under the driver's control, but
1805 * needs to be disabled. A global MAC reset is issued to stop the
1806 * hardware, and all transmit and receive resources are freed.
1808 static int igb_close(struct net_device *netdev)
1810 struct igb_adapter *adapter = netdev_priv(netdev);
1812 WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
1815 igb_free_irq(adapter);
1817 igb_free_all_tx_resources(adapter);
1818 igb_free_all_rx_resources(adapter);
1820 /* kill manageability vlan ID if supported, but not if a vlan with
1821 * the same ID is registered on the host OS (let 8021q kill it) */
1822 if ((adapter->hw.mng_cookie.status &
1823 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1825 vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
1826 igb_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1832 * igb_setup_tx_resources - allocate Tx resources (Descriptors)
1833 * @adapter: board private structure
1834 * @tx_ring: tx descriptor ring (for a specific queue) to setup
1836 * Return 0 on success, negative on failure
1838 int igb_setup_tx_resources(struct igb_adapter *adapter,
1839 struct igb_ring *tx_ring)
1841 struct pci_dev *pdev = adapter->pdev;
1844 size = sizeof(struct igb_buffer) * tx_ring->count;
1845 tx_ring->buffer_info = vmalloc(size);
1846 if (!tx_ring->buffer_info)
1848 memset(tx_ring->buffer_info, 0, size);
1850 /* round up to nearest 4K */
1851 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
1852 tx_ring->size = ALIGN(tx_ring->size, 4096);
1854 tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
1860 tx_ring->adapter = adapter;
1861 tx_ring->next_to_use = 0;
1862 tx_ring->next_to_clean = 0;
1866 vfree(tx_ring->buffer_info);
1867 dev_err(&adapter->pdev->dev,
1868 "Unable to allocate memory for the transmit descriptor ring\n");
1873 * igb_setup_all_tx_resources - wrapper to allocate Tx resources
1874 * (Descriptors) for all queues
1875 * @adapter: board private structure
1877 * Return 0 on success, negative on failure
1879 static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
1884 for (i = 0; i < adapter->num_tx_queues; i++) {
1885 err = igb_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1887 dev_err(&adapter->pdev->dev,
1888 "Allocation for Tx Queue %u failed\n", i);
1889 for (i--; i >= 0; i--)
1890 igb_free_tx_resources(&adapter->tx_ring[i]);
1895 for (i = 0; i < IGB_MAX_TX_QUEUES; i++) {
1896 r_idx = i % adapter->num_tx_queues;
1897 adapter->multi_tx_table[i] = &adapter->tx_ring[r_idx];
1903 * igb_configure_tx - Configure transmit Unit after Reset
1904 * @adapter: board private structure
1906 * Configure the Tx unit of the MAC after a reset.
1908 static void igb_configure_tx(struct igb_adapter *adapter)
1911 struct e1000_hw *hw = &adapter->hw;
1916 for (i = 0; i < adapter->num_tx_queues; i++) {
1917 struct igb_ring *ring = &adapter->tx_ring[i];
1919 wr32(E1000_TDLEN(j),
1920 ring->count * sizeof(union e1000_adv_tx_desc));
1922 wr32(E1000_TDBAL(j),
1923 tdba & 0x00000000ffffffffULL);
1924 wr32(E1000_TDBAH(j), tdba >> 32);
1926 ring->head = E1000_TDH(j);
1927 ring->tail = E1000_TDT(j);
1928 writel(0, hw->hw_addr + ring->tail);
1929 writel(0, hw->hw_addr + ring->head);
1930 txdctl = rd32(E1000_TXDCTL(j));
1931 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1932 wr32(E1000_TXDCTL(j), txdctl);
1934 /* Turn off Relaxed Ordering on head write-backs. The
1935 * writebacks MUST be delivered in order or it will
1936 * completely screw up our bookeeping.
1938 txctrl = rd32(E1000_DCA_TXCTRL(j));
1939 txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1940 wr32(E1000_DCA_TXCTRL(j), txctrl);
1943 /* disable queue 0 to prevent tail bump w/o re-configuration */
1944 if (adapter->vfs_allocated_count)
1945 wr32(E1000_TXDCTL(0), 0);
1947 /* Program the Transmit Control Register */
1948 tctl = rd32(E1000_TCTL);
1949 tctl &= ~E1000_TCTL_CT;
1950 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1951 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1953 igb_config_collision_dist(hw);
1955 /* Setup Transmit Descriptor Settings for eop descriptor */
1956 adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_RS;
1958 /* Enable transmits */
1959 tctl |= E1000_TCTL_EN;
1961 wr32(E1000_TCTL, tctl);
1965 * igb_setup_rx_resources - allocate Rx resources (Descriptors)
1966 * @adapter: board private structure
1967 * @rx_ring: rx descriptor ring (for a specific queue) to setup
1969 * Returns 0 on success, negative on failure
1971 int igb_setup_rx_resources(struct igb_adapter *adapter,
1972 struct igb_ring *rx_ring)
1974 struct pci_dev *pdev = adapter->pdev;
1977 size = sizeof(struct igb_buffer) * rx_ring->count;
1978 rx_ring->buffer_info = vmalloc(size);
1979 if (!rx_ring->buffer_info)
1981 memset(rx_ring->buffer_info, 0, size);
1983 desc_len = sizeof(union e1000_adv_rx_desc);
1985 /* Round up to nearest 4K */
1986 rx_ring->size = rx_ring->count * desc_len;
1987 rx_ring->size = ALIGN(rx_ring->size, 4096);
1989 rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size,
1995 rx_ring->next_to_clean = 0;
1996 rx_ring->next_to_use = 0;
1998 rx_ring->adapter = adapter;
2003 vfree(rx_ring->buffer_info);
2004 dev_err(&adapter->pdev->dev, "Unable to allocate memory for "
2005 "the receive descriptor ring\n");
2010 * igb_setup_all_rx_resources - wrapper to allocate Rx resources
2011 * (Descriptors) for all queues
2012 * @adapter: board private structure
2014 * Return 0 on success, negative on failure
2016 static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
2020 for (i = 0; i < adapter->num_rx_queues; i++) {
2021 err = igb_setup_rx_resources(adapter, &adapter->rx_ring[i]);
2023 dev_err(&adapter->pdev->dev,
2024 "Allocation for Rx Queue %u failed\n", i);
2025 for (i--; i >= 0; i--)
2026 igb_free_rx_resources(&adapter->rx_ring[i]);
2035 * igb_setup_rctl - configure the receive control registers
2036 * @adapter: Board private structure
2038 static void igb_setup_rctl(struct igb_adapter *adapter)
2040 struct e1000_hw *hw = &adapter->hw;
2045 rctl = rd32(E1000_RCTL);
2047 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2048 rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
2050 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
2051 (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2054 * enable stripping of CRC. It's unlikely this will break BMC
2055 * redirection as it did with e1000. Newer features require
2056 * that the HW strips the CRC.
2058 rctl |= E1000_RCTL_SECRC;
2061 * disable store bad packets and clear size bits.
2063 rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
2065 /* enable LPE when to prevent packets larger than max_frame_size */
2066 rctl |= E1000_RCTL_LPE;
2068 /* Setup buffer sizes */
2069 switch (adapter->rx_buffer_len) {
2070 case IGB_RXBUFFER_256:
2071 rctl |= E1000_RCTL_SZ_256;
2073 case IGB_RXBUFFER_512:
2074 rctl |= E1000_RCTL_SZ_512;
2077 srrctl = ALIGN(adapter->rx_buffer_len, 1024)
2078 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
2082 /* 82575 and greater support packet-split where the protocol
2083 * header is placed in skb->data and the packet data is
2084 * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2085 * In the case of a non-split, skb->data is linearly filled,
2086 * followed by the page buffers. Therefore, skb->data is
2087 * sized to hold the largest protocol header.
2089 /* allocations using alloc_page take too long for regular MTU
2090 * so only enable packet split for jumbo frames */
2091 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2092 adapter->rx_ps_hdr_size = IGB_RXBUFFER_128;
2093 srrctl |= adapter->rx_ps_hdr_size <<
2094 E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
2095 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2097 adapter->rx_ps_hdr_size = 0;
2098 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
2101 /* Attention!!! For SR-IOV PF driver operations you must enable
2102 * queue drop for all VF and PF queues to prevent head of line blocking
2103 * if an un-trusted VF does not provide descriptors to hardware.
2105 if (adapter->vfs_allocated_count) {
2108 /* set all queue drop enable bits */
2109 wr32(E1000_QDE, ALL_QUEUES);
2110 srrctl |= E1000_SRRCTL_DROP_EN;
2112 /* disable queue 0 to prevent tail write w/o re-config */
2113 wr32(E1000_RXDCTL(0), 0);
2115 vmolr = rd32(E1000_VMOLR(adapter->vfs_allocated_count));
2116 if (rctl & E1000_RCTL_LPE)
2117 vmolr |= E1000_VMOLR_LPE;
2118 if (adapter->num_rx_queues > 1)
2119 vmolr |= E1000_VMOLR_RSSE;
2120 wr32(E1000_VMOLR(adapter->vfs_allocated_count), vmolr);
2123 for (i = 0; i < adapter->num_rx_queues; i++) {
2124 int j = adapter->rx_ring[i].reg_idx;
2125 wr32(E1000_SRRCTL(j), srrctl);
2128 wr32(E1000_RCTL, rctl);
2132 * igb_rlpml_set - set maximum receive packet size
2133 * @adapter: board private structure
2135 * Configure maximum receivable packet size.
2137 static void igb_rlpml_set(struct igb_adapter *adapter)
2139 u32 max_frame_size = adapter->max_frame_size;
2140 struct e1000_hw *hw = &adapter->hw;
2141 u16 pf_id = adapter->vfs_allocated_count;
2144 max_frame_size += VLAN_TAG_SIZE;
2146 /* if vfs are enabled we set RLPML to the largest possible request
2147 * size and set the VMOLR RLPML to the size we need */
2149 igb_set_vf_rlpml(adapter, max_frame_size, pf_id);
2150 max_frame_size = MAX_STD_JUMBO_FRAME_SIZE + VLAN_TAG_SIZE;
2153 wr32(E1000_RLPML, max_frame_size);
2157 * igb_configure_vt_default_pool - Configure VT default pool
2158 * @adapter: board private structure
2160 * Configure the default pool
2162 static void igb_configure_vt_default_pool(struct igb_adapter *adapter)
2164 struct e1000_hw *hw = &adapter->hw;
2165 u16 pf_id = adapter->vfs_allocated_count;
2168 /* not in sr-iov mode - do nothing */
2172 vtctl = rd32(E1000_VT_CTL);
2173 vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK |
2174 E1000_VT_CTL_DISABLE_DEF_POOL);
2175 vtctl |= pf_id << E1000_VT_CTL_DEFAULT_POOL_SHIFT;
2176 wr32(E1000_VT_CTL, vtctl);
2180 * igb_configure_rx - Configure receive Unit after Reset
2181 * @adapter: board private structure
2183 * Configure the Rx unit of the MAC after a reset.
2185 static void igb_configure_rx(struct igb_adapter *adapter)
2188 struct e1000_hw *hw = &adapter->hw;
2193 /* disable receives while setting up the descriptors */
2194 rctl = rd32(E1000_RCTL);
2195 wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
2199 if (adapter->itr_setting > 3)
2200 wr32(E1000_ITR, adapter->itr);
2202 /* Setup the HW Rx Head and Tail Descriptor Pointers and
2203 * the Base and Length of the Rx Descriptor Ring */
2204 for (i = 0; i < adapter->num_rx_queues; i++) {
2205 struct igb_ring *ring = &adapter->rx_ring[i];
2206 int j = ring->reg_idx;
2208 wr32(E1000_RDBAL(j),
2209 rdba & 0x00000000ffffffffULL);
2210 wr32(E1000_RDBAH(j), rdba >> 32);
2211 wr32(E1000_RDLEN(j),
2212 ring->count * sizeof(union e1000_adv_rx_desc));
2214 ring->head = E1000_RDH(j);
2215 ring->tail = E1000_RDT(j);
2216 writel(0, hw->hw_addr + ring->tail);
2217 writel(0, hw->hw_addr + ring->head);
2219 rxdctl = rd32(E1000_RXDCTL(j));
2220 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
2221 rxdctl &= 0xFFF00000;
2222 rxdctl |= IGB_RX_PTHRESH;
2223 rxdctl |= IGB_RX_HTHRESH << 8;
2224 rxdctl |= IGB_RX_WTHRESH << 16;
2225 wr32(E1000_RXDCTL(j), rxdctl);
2228 if (adapter->num_rx_queues > 1) {
2237 get_random_bytes(&random[0], 40);
2239 if (hw->mac.type >= e1000_82576)
2243 for (j = 0; j < (32 * 4); j++) {
2245 adapter->rx_ring[(j % adapter->num_rx_queues)].reg_idx << shift;
2248 hw->hw_addr + E1000_RETA(0) + (j & ~3));
2250 if (adapter->vfs_allocated_count)
2251 mrqc = E1000_MRQC_ENABLE_VMDQ_RSS_2Q;
2253 mrqc = E1000_MRQC_ENABLE_RSS_4Q;
2255 /* Fill out hash function seeds */
2256 for (j = 0; j < 10; j++)
2257 array_wr32(E1000_RSSRK(0), j, random[j]);
2259 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
2260 E1000_MRQC_RSS_FIELD_IPV4_TCP);
2261 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6 |
2262 E1000_MRQC_RSS_FIELD_IPV6_TCP);
2263 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4_UDP |
2264 E1000_MRQC_RSS_FIELD_IPV6_UDP);
2265 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
2266 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
2268 wr32(E1000_MRQC, mrqc);
2269 } else if (adapter->vfs_allocated_count) {
2270 /* Enable multi-queue for sr-iov */
2271 wr32(E1000_MRQC, E1000_MRQC_ENABLE_VMDQ);
2274 /* Enable Receive Checksum Offload for TCP and UDP */
2275 rxcsum = rd32(E1000_RXCSUM);
2276 /* Disable raw packet checksumming */
2277 rxcsum |= E1000_RXCSUM_PCSD;
2279 if (adapter->hw.mac.type == e1000_82576)
2280 /* Enable Receive Checksum Offload for SCTP */
2281 rxcsum |= E1000_RXCSUM_CRCOFL;
2283 /* Don't need to set TUOFL or IPOFL, they default to 1 */
2284 wr32(E1000_RXCSUM, rxcsum);
2286 /* Set the default pool for the PF's first queue */
2287 igb_configure_vt_default_pool(adapter);
2289 /* set the correct pool for the PF default MAC address in entry 0 */
2290 igb_rar_set_qsel(adapter, adapter->hw.mac.addr, 0,
2291 adapter->vfs_allocated_count);
2293 igb_rlpml_set(adapter);
2295 /* Enable Receives */
2296 wr32(E1000_RCTL, rctl);
2300 * igb_free_tx_resources - Free Tx Resources per Queue
2301 * @tx_ring: Tx descriptor ring for a specific queue
2303 * Free all transmit software resources
2305 void igb_free_tx_resources(struct igb_ring *tx_ring)
2307 struct pci_dev *pdev = tx_ring->adapter->pdev;
2309 igb_clean_tx_ring(tx_ring);
2311 vfree(tx_ring->buffer_info);
2312 tx_ring->buffer_info = NULL;
2314 pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
2316 tx_ring->desc = NULL;
2320 * igb_free_all_tx_resources - Free Tx Resources for All Queues
2321 * @adapter: board private structure
2323 * Free all transmit software resources
2325 static void igb_free_all_tx_resources(struct igb_adapter *adapter)
2329 for (i = 0; i < adapter->num_tx_queues; i++)
2330 igb_free_tx_resources(&adapter->tx_ring[i]);
2333 static void igb_unmap_and_free_tx_resource(struct igb_adapter *adapter,
2334 struct igb_buffer *buffer_info)
2336 buffer_info->dma = 0;
2337 if (buffer_info->skb) {
2338 skb_dma_unmap(&adapter->pdev->dev, buffer_info->skb,
2340 dev_kfree_skb_any(buffer_info->skb);
2341 buffer_info->skb = NULL;
2343 buffer_info->time_stamp = 0;
2344 /* buffer_info must be completely set up in the transmit path */
2348 * igb_clean_tx_ring - Free Tx Buffers
2349 * @tx_ring: ring to be cleaned
2351 static void igb_clean_tx_ring(struct igb_ring *tx_ring)
2353 struct igb_adapter *adapter = tx_ring->adapter;
2354 struct igb_buffer *buffer_info;
2358 if (!tx_ring->buffer_info)
2360 /* Free all the Tx ring sk_buffs */
2362 for (i = 0; i < tx_ring->count; i++) {
2363 buffer_info = &tx_ring->buffer_info[i];
2364 igb_unmap_and_free_tx_resource(adapter, buffer_info);
2367 size = sizeof(struct igb_buffer) * tx_ring->count;
2368 memset(tx_ring->buffer_info, 0, size);
2370 /* Zero out the descriptor ring */
2372 memset(tx_ring->desc, 0, tx_ring->size);
2374 tx_ring->next_to_use = 0;
2375 tx_ring->next_to_clean = 0;
2377 writel(0, adapter->hw.hw_addr + tx_ring->head);
2378 writel(0, adapter->hw.hw_addr + tx_ring->tail);
2382 * igb_clean_all_tx_rings - Free Tx Buffers for all queues
2383 * @adapter: board private structure
2385 static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
2389 for (i = 0; i < adapter->num_tx_queues; i++)
2390 igb_clean_tx_ring(&adapter->tx_ring[i]);
2394 * igb_free_rx_resources - Free Rx Resources
2395 * @rx_ring: ring to clean the resources from
2397 * Free all receive software resources
2399 void igb_free_rx_resources(struct igb_ring *rx_ring)
2401 struct pci_dev *pdev = rx_ring->adapter->pdev;
2403 igb_clean_rx_ring(rx_ring);
2405 vfree(rx_ring->buffer_info);
2406 rx_ring->buffer_info = NULL;
2408 pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
2410 rx_ring->desc = NULL;
2414 * igb_free_all_rx_resources - Free Rx Resources for All Queues
2415 * @adapter: board private structure
2417 * Free all receive software resources
2419 static void igb_free_all_rx_resources(struct igb_adapter *adapter)
2423 for (i = 0; i < adapter->num_rx_queues; i++)
2424 igb_free_rx_resources(&adapter->rx_ring[i]);
2428 * igb_clean_rx_ring - Free Rx Buffers per Queue
2429 * @rx_ring: ring to free buffers from
2431 static void igb_clean_rx_ring(struct igb_ring *rx_ring)
2433 struct igb_adapter *adapter = rx_ring->adapter;
2434 struct igb_buffer *buffer_info;
2435 struct pci_dev *pdev = adapter->pdev;
2439 if (!rx_ring->buffer_info)
2441 /* Free all the Rx ring sk_buffs */
2442 for (i = 0; i < rx_ring->count; i++) {
2443 buffer_info = &rx_ring->buffer_info[i];
2444 if (buffer_info->dma) {
2445 if (adapter->rx_ps_hdr_size)
2446 pci_unmap_single(pdev, buffer_info->dma,
2447 adapter->rx_ps_hdr_size,
2448 PCI_DMA_FROMDEVICE);
2450 pci_unmap_single(pdev, buffer_info->dma,
2451 adapter->rx_buffer_len,
2452 PCI_DMA_FROMDEVICE);
2453 buffer_info->dma = 0;
2456 if (buffer_info->skb) {
2457 dev_kfree_skb(buffer_info->skb);
2458 buffer_info->skb = NULL;
2460 if (buffer_info->page) {
2461 if (buffer_info->page_dma)
2462 pci_unmap_page(pdev, buffer_info->page_dma,
2464 PCI_DMA_FROMDEVICE);
2465 put_page(buffer_info->page);
2466 buffer_info->page = NULL;
2467 buffer_info->page_dma = 0;
2468 buffer_info->page_offset = 0;
2472 size = sizeof(struct igb_buffer) * rx_ring->count;
2473 memset(rx_ring->buffer_info, 0, size);
2475 /* Zero out the descriptor ring */
2476 memset(rx_ring->desc, 0, rx_ring->size);
2478 rx_ring->next_to_clean = 0;
2479 rx_ring->next_to_use = 0;
2481 writel(0, adapter->hw.hw_addr + rx_ring->head);
2482 writel(0, adapter->hw.hw_addr + rx_ring->tail);
2486 * igb_clean_all_rx_rings - Free Rx Buffers for all queues
2487 * @adapter: board private structure
2489 static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
2493 for (i = 0; i < adapter->num_rx_queues; i++)
2494 igb_clean_rx_ring(&adapter->rx_ring[i]);
2498 * igb_set_mac - Change the Ethernet Address of the NIC
2499 * @netdev: network interface device structure
2500 * @p: pointer to an address structure
2502 * Returns 0 on success, negative on failure
2504 static int igb_set_mac(struct net_device *netdev, void *p)
2506 struct igb_adapter *adapter = netdev_priv(netdev);
2507 struct e1000_hw *hw = &adapter->hw;
2508 struct sockaddr *addr = p;
2510 if (!is_valid_ether_addr(addr->sa_data))
2511 return -EADDRNOTAVAIL;
2513 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2514 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
2516 /* set the correct pool for the new PF MAC address in entry 0 */
2517 igb_rar_set_qsel(adapter, hw->mac.addr, 0,
2518 adapter->vfs_allocated_count);
2524 * igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
2525 * @netdev: network interface device structure
2527 * The set_rx_mode entry point is called whenever the unicast or multicast
2528 * address lists or the network interface flags are updated. This routine is
2529 * responsible for configuring the hardware for proper unicast, multicast,
2530 * promiscuous mode, and all-multi behavior.
2532 static void igb_set_rx_mode(struct net_device *netdev)
2534 struct igb_adapter *adapter = netdev_priv(netdev);
2535 struct e1000_hw *hw = &adapter->hw;
2536 unsigned int rar_entries = hw->mac.rar_entry_count -
2537 (adapter->vfs_allocated_count + 1);
2538 struct dev_mc_list *mc_ptr = netdev->mc_list;
2539 u8 *mta_list = NULL;
2543 /* Check for Promiscuous and All Multicast modes */
2544 rctl = rd32(E1000_RCTL);
2546 if (netdev->flags & IFF_PROMISC) {
2547 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2548 rctl &= ~E1000_RCTL_VFE;
2550 if (netdev->flags & IFF_ALLMULTI)
2551 rctl |= E1000_RCTL_MPE;
2553 rctl &= ~E1000_RCTL_MPE;
2555 if (netdev->uc.count > rar_entries)
2556 rctl |= E1000_RCTL_UPE;
2558 rctl &= ~E1000_RCTL_UPE;
2559 rctl |= E1000_RCTL_VFE;
2561 wr32(E1000_RCTL, rctl);
2563 if (netdev->uc.count && rar_entries) {
2564 struct netdev_hw_addr *ha;
2565 list_for_each_entry(ha, &netdev->uc.list, list) {
2568 igb_rar_set_qsel(adapter, ha->addr,
2570 adapter->vfs_allocated_count);
2573 /* write the addresses in reverse order to avoid write combining */
2574 for (; rar_entries > 0 ; rar_entries--) {
2575 wr32(E1000_RAH(rar_entries), 0);
2576 wr32(E1000_RAL(rar_entries), 0);
2580 if (!netdev->mc_count) {
2581 /* nothing to program, so clear mc list */
2582 igb_update_mc_addr_list(hw, NULL, 0);
2583 igb_restore_vf_multicasts(adapter);
2587 mta_list = kzalloc(netdev->mc_count * 6, GFP_ATOMIC);
2589 dev_err(&adapter->pdev->dev,
2590 "failed to allocate multicast filter list\n");
2594 /* The shared function expects a packed array of only addresses. */
2595 for (i = 0; i < netdev->mc_count; i++) {
2598 memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr, ETH_ALEN);
2599 mc_ptr = mc_ptr->next;
2601 igb_update_mc_addr_list(hw, mta_list, i);
2603 igb_restore_vf_multicasts(adapter);
2606 /* Need to wait a few seconds after link up to get diagnostic information from
2608 static void igb_update_phy_info(unsigned long data)
2610 struct igb_adapter *adapter = (struct igb_adapter *) data;
2611 igb_get_phy_info(&adapter->hw);
2615 * igb_has_link - check shared code for link and determine up/down
2616 * @adapter: pointer to driver private info
2618 static bool igb_has_link(struct igb_adapter *adapter)
2620 struct e1000_hw *hw = &adapter->hw;
2621 bool link_active = false;
2624 /* get_link_status is set on LSC (link status) interrupt or
2625 * rx sequence error interrupt. get_link_status will stay
2626 * false until the e1000_check_for_link establishes link
2627 * for copper adapters ONLY
2629 switch (hw->phy.media_type) {
2630 case e1000_media_type_copper:
2631 if (hw->mac.get_link_status) {
2632 ret_val = hw->mac.ops.check_for_link(hw);
2633 link_active = !hw->mac.get_link_status;
2638 case e1000_media_type_internal_serdes:
2639 ret_val = hw->mac.ops.check_for_link(hw);
2640 link_active = hw->mac.serdes_has_link;
2643 case e1000_media_type_unknown:
2651 * igb_watchdog - Timer Call-back
2652 * @data: pointer to adapter cast into an unsigned long
2654 static void igb_watchdog(unsigned long data)
2656 struct igb_adapter *adapter = (struct igb_adapter *)data;
2657 /* Do the rest outside of interrupt context */
2658 schedule_work(&adapter->watchdog_task);
2661 static void igb_watchdog_task(struct work_struct *work)
2663 struct igb_adapter *adapter = container_of(work,
2664 struct igb_adapter, watchdog_task);
2665 struct e1000_hw *hw = &adapter->hw;
2666 struct net_device *netdev = adapter->netdev;
2667 struct igb_ring *tx_ring = adapter->tx_ring;
2672 link = igb_has_link(adapter);
2673 if ((netif_carrier_ok(netdev)) && link)
2677 if (!netif_carrier_ok(netdev)) {
2679 hw->mac.ops.get_speed_and_duplex(&adapter->hw,
2680 &adapter->link_speed,
2681 &adapter->link_duplex);
2683 ctrl = rd32(E1000_CTRL);
2684 /* Links status message must follow this format */
2685 printk(KERN_INFO "igb: %s NIC Link is Up %d Mbps %s, "
2686 "Flow Control: %s\n",
2688 adapter->link_speed,
2689 adapter->link_duplex == FULL_DUPLEX ?
2690 "Full Duplex" : "Half Duplex",
2691 ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2692 E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2693 E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2694 E1000_CTRL_TFCE) ? "TX" : "None")));
2696 /* tweak tx_queue_len according to speed/duplex and
2697 * adjust the timeout factor */
2698 netdev->tx_queue_len = adapter->tx_queue_len;
2699 adapter->tx_timeout_factor = 1;
2700 switch (adapter->link_speed) {
2702 netdev->tx_queue_len = 10;
2703 adapter->tx_timeout_factor = 14;
2706 netdev->tx_queue_len = 100;
2707 /* maybe add some timeout factor ? */
2711 netif_carrier_on(netdev);
2713 igb_ping_all_vfs(adapter);
2715 /* link state has changed, schedule phy info update */
2716 if (!test_bit(__IGB_DOWN, &adapter->state))
2717 mod_timer(&adapter->phy_info_timer,
2718 round_jiffies(jiffies + 2 * HZ));
2721 if (netif_carrier_ok(netdev)) {
2722 adapter->link_speed = 0;
2723 adapter->link_duplex = 0;
2724 /* Links status message must follow this format */
2725 printk(KERN_INFO "igb: %s NIC Link is Down\n",
2727 netif_carrier_off(netdev);
2729 igb_ping_all_vfs(adapter);
2731 /* link state has changed, schedule phy info update */
2732 if (!test_bit(__IGB_DOWN, &adapter->state))
2733 mod_timer(&adapter->phy_info_timer,
2734 round_jiffies(jiffies + 2 * HZ));
2739 igb_update_stats(adapter);
2741 hw->mac.tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2742 adapter->tpt_old = adapter->stats.tpt;
2743 hw->mac.collision_delta = adapter->stats.colc - adapter->colc_old;
2744 adapter->colc_old = adapter->stats.colc;
2746 adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
2747 adapter->gorc_old = adapter->stats.gorc;
2748 adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
2749 adapter->gotc_old = adapter->stats.gotc;
2751 igb_update_adaptive(&adapter->hw);
2753 if (!netif_carrier_ok(netdev)) {
2754 if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) {
2755 /* We've lost link, so the controller stops DMA,
2756 * but we've got queued Tx work that's never going
2757 * to get done, so reset controller to flush Tx.
2758 * (Do the reset outside of interrupt context). */
2759 adapter->tx_timeout_count++;
2760 schedule_work(&adapter->reset_task);
2761 /* return immediately since reset is imminent */
2766 /* Cause software interrupt to ensure rx ring is cleaned */
2767 if (adapter->msix_entries) {
2768 for (i = 0; i < adapter->num_rx_queues; i++)
2769 eics |= adapter->rx_ring[i].eims_value;
2770 wr32(E1000_EICS, eics);
2772 wr32(E1000_ICS, E1000_ICS_RXDMT0);
2775 /* Force detection of hung controller every watchdog period */
2776 tx_ring->detect_tx_hung = true;
2778 /* Reset the timer */
2779 if (!test_bit(__IGB_DOWN, &adapter->state))
2780 mod_timer(&adapter->watchdog_timer,
2781 round_jiffies(jiffies + 2 * HZ));
2784 enum latency_range {
2788 latency_invalid = 255
2793 * igb_update_ring_itr - update the dynamic ITR value based on packet size
2795 * Stores a new ITR value based on strictly on packet size. This
2796 * algorithm is less sophisticated than that used in igb_update_itr,
2797 * due to the difficulty of synchronizing statistics across multiple
2798 * receive rings. The divisors and thresholds used by this fuction
2799 * were determined based on theoretical maximum wire speed and testing
2800 * data, in order to minimize response time while increasing bulk
2802 * This functionality is controlled by the InterruptThrottleRate module
2803 * parameter (see igb_param.c)
2804 * NOTE: This function is called only when operating in a multiqueue
2805 * receive environment.
2806 * @rx_ring: pointer to ring
2808 static void igb_update_ring_itr(struct igb_ring *rx_ring)
2810 int new_val = rx_ring->itr_val;
2811 int avg_wire_size = 0;
2812 struct igb_adapter *adapter = rx_ring->adapter;
2814 if (!rx_ring->total_packets)
2815 goto clear_counts; /* no packets, so don't do anything */
2817 /* For non-gigabit speeds, just fix the interrupt rate at 4000
2818 * ints/sec - ITR timer value of 120 ticks.
2820 if (adapter->link_speed != SPEED_1000) {
2824 avg_wire_size = rx_ring->total_bytes / rx_ring->total_packets;
2826 /* Add 24 bytes to size to account for CRC, preamble, and gap */
2827 avg_wire_size += 24;
2829 /* Don't starve jumbo frames */
2830 avg_wire_size = min(avg_wire_size, 3000);
2832 /* Give a little boost to mid-size frames */
2833 if ((avg_wire_size > 300) && (avg_wire_size < 1200))
2834 new_val = avg_wire_size / 3;
2836 new_val = avg_wire_size / 2;
2839 if (new_val != rx_ring->itr_val) {
2840 rx_ring->itr_val = new_val;
2841 rx_ring->set_itr = 1;
2844 rx_ring->total_bytes = 0;
2845 rx_ring->total_packets = 0;
2849 * igb_update_itr - update the dynamic ITR value based on statistics
2850 * Stores a new ITR value based on packets and byte
2851 * counts during the last interrupt. The advantage of per interrupt
2852 * computation is faster updates and more accurate ITR for the current
2853 * traffic pattern. Constants in this function were computed
2854 * based on theoretical maximum wire speed and thresholds were set based
2855 * on testing data as well as attempting to minimize response time
2856 * while increasing bulk throughput.
2857 * this functionality is controlled by the InterruptThrottleRate module
2858 * parameter (see igb_param.c)
2859 * NOTE: These calculations are only valid when operating in a single-
2860 * queue environment.
2861 * @adapter: pointer to adapter
2862 * @itr_setting: current adapter->itr
2863 * @packets: the number of packets during this measurement interval
2864 * @bytes: the number of bytes during this measurement interval
2866 static unsigned int igb_update_itr(struct igb_adapter *adapter, u16 itr_setting,
2867 int packets, int bytes)
2869 unsigned int retval = itr_setting;
2872 goto update_itr_done;
2874 switch (itr_setting) {
2875 case lowest_latency:
2876 /* handle TSO and jumbo frames */
2877 if (bytes/packets > 8000)
2878 retval = bulk_latency;
2879 else if ((packets < 5) && (bytes > 512))
2880 retval = low_latency;
2882 case low_latency: /* 50 usec aka 20000 ints/s */
2883 if (bytes > 10000) {
2884 /* this if handles the TSO accounting */
2885 if (bytes/packets > 8000) {
2886 retval = bulk_latency;
2887 } else if ((packets < 10) || ((bytes/packets) > 1200)) {
2888 retval = bulk_latency;
2889 } else if ((packets > 35)) {
2890 retval = lowest_latency;
2892 } else if (bytes/packets > 2000) {
2893 retval = bulk_latency;
2894 } else if (packets <= 2 && bytes < 512) {
2895 retval = lowest_latency;
2898 case bulk_latency: /* 250 usec aka 4000 ints/s */
2899 if (bytes > 25000) {
2901 retval = low_latency;
2902 } else if (bytes < 1500) {
2903 retval = low_latency;
2912 static void igb_set_itr(struct igb_adapter *adapter)
2915 u32 new_itr = adapter->itr;
2917 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2918 if (adapter->link_speed != SPEED_1000) {
2924 adapter->rx_itr = igb_update_itr(adapter,
2926 adapter->rx_ring->total_packets,
2927 adapter->rx_ring->total_bytes);
2929 if (adapter->rx_ring->buddy) {
2930 adapter->tx_itr = igb_update_itr(adapter,
2932 adapter->tx_ring->total_packets,
2933 adapter->tx_ring->total_bytes);
2934 current_itr = max(adapter->rx_itr, adapter->tx_itr);
2936 current_itr = adapter->rx_itr;
2939 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2940 if (adapter->itr_setting == 3 && current_itr == lowest_latency)
2941 current_itr = low_latency;
2943 switch (current_itr) {
2944 /* counts and packets in update_itr are dependent on these numbers */
2945 case lowest_latency:
2946 new_itr = 56; /* aka 70,000 ints/sec */
2949 new_itr = 196; /* aka 20,000 ints/sec */
2952 new_itr = 980; /* aka 4,000 ints/sec */
2959 adapter->rx_ring->total_bytes = 0;
2960 adapter->rx_ring->total_packets = 0;
2961 if (adapter->rx_ring->buddy) {
2962 adapter->rx_ring->buddy->total_bytes = 0;
2963 adapter->rx_ring->buddy->total_packets = 0;
2966 if (new_itr != adapter->itr) {
2967 /* this attempts to bias the interrupt rate towards Bulk
2968 * by adding intermediate steps when interrupt rate is
2970 new_itr = new_itr > adapter->itr ?
2971 max((new_itr * adapter->itr) /
2972 (new_itr + (adapter->itr >> 2)), new_itr) :
2974 /* Don't write the value here; it resets the adapter's
2975 * internal timer, and causes us to delay far longer than
2976 * we should between interrupts. Instead, we write the ITR
2977 * value at the beginning of the next interrupt so the timing
2978 * ends up being correct.
2980 adapter->itr = new_itr;
2981 adapter->rx_ring->itr_val = new_itr;
2982 adapter->rx_ring->set_itr = 1;
2989 #define IGB_TX_FLAGS_CSUM 0x00000001
2990 #define IGB_TX_FLAGS_VLAN 0x00000002
2991 #define IGB_TX_FLAGS_TSO 0x00000004
2992 #define IGB_TX_FLAGS_IPV4 0x00000008
2993 #define IGB_TX_FLAGS_TSTAMP 0x00000010
2994 #define IGB_TX_FLAGS_VLAN_MASK 0xffff0000
2995 #define IGB_TX_FLAGS_VLAN_SHIFT 16
2997 static inline int igb_tso_adv(struct igb_adapter *adapter,
2998 struct igb_ring *tx_ring,
2999 struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
3001 struct e1000_adv_tx_context_desc *context_desc;
3004 struct igb_buffer *buffer_info;
3005 u32 info = 0, tu_cmd = 0;
3006 u32 mss_l4len_idx, l4len;
3009 if (skb_header_cloned(skb)) {
3010 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3015 l4len = tcp_hdrlen(skb);
3018 if (skb->protocol == htons(ETH_P_IP)) {
3019 struct iphdr *iph = ip_hdr(skb);
3022 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3026 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
3027 ipv6_hdr(skb)->payload_len = 0;
3028 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3029 &ipv6_hdr(skb)->daddr,
3033 i = tx_ring->next_to_use;
3035 buffer_info = &tx_ring->buffer_info[i];
3036 context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
3037 /* VLAN MACLEN IPLEN */
3038 if (tx_flags & IGB_TX_FLAGS_VLAN)
3039 info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
3040 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
3041 *hdr_len += skb_network_offset(skb);
3042 info |= skb_network_header_len(skb);
3043 *hdr_len += skb_network_header_len(skb);
3044 context_desc->vlan_macip_lens = cpu_to_le32(info);
3046 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3047 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
3049 if (skb->protocol == htons(ETH_P_IP))
3050 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
3051 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
3053 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
3056 mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
3057 mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
3059 /* For 82575, context index must be unique per ring. */
3060 if (adapter->flags & IGB_FLAG_NEED_CTX_IDX)
3061 mss_l4len_idx |= tx_ring->queue_index << 4;
3063 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
3064 context_desc->seqnum_seed = 0;
3066 buffer_info->time_stamp = jiffies;
3067 buffer_info->next_to_watch = i;
3068 buffer_info->dma = 0;
3070 if (i == tx_ring->count)
3073 tx_ring->next_to_use = i;
3078 static inline bool igb_tx_csum_adv(struct igb_adapter *adapter,
3079 struct igb_ring *tx_ring,
3080 struct sk_buff *skb, u32 tx_flags)
3082 struct e1000_adv_tx_context_desc *context_desc;
3084 struct igb_buffer *buffer_info;
3085 u32 info = 0, tu_cmd = 0;
3087 if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
3088 (tx_flags & IGB_TX_FLAGS_VLAN)) {
3089 i = tx_ring->next_to_use;
3090 buffer_info = &tx_ring->buffer_info[i];
3091 context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
3093 if (tx_flags & IGB_TX_FLAGS_VLAN)
3094 info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
3095 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
3096 if (skb->ip_summed == CHECKSUM_PARTIAL)
3097 info |= skb_network_header_len(skb);
3099 context_desc->vlan_macip_lens = cpu_to_le32(info);
3101 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
3103 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3106 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3107 const struct vlan_ethhdr *vhdr =
3108 (const struct vlan_ethhdr*)skb->data;
3110 protocol = vhdr->h_vlan_encapsulated_proto;
3112 protocol = skb->protocol;
3116 case cpu_to_be16(ETH_P_IP):
3117 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
3118 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
3119 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
3120 else if (ip_hdr(skb)->protocol == IPPROTO_SCTP)
3121 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_SCTP;
3123 case cpu_to_be16(ETH_P_IPV6):
3124 /* XXX what about other V6 headers?? */
3125 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3126 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
3127 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_SCTP)
3128 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_SCTP;
3131 if (unlikely(net_ratelimit()))
3132 dev_warn(&adapter->pdev->dev,
3133 "partial checksum but proto=%x!\n",
3139 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
3140 context_desc->seqnum_seed = 0;
3141 if (adapter->flags & IGB_FLAG_NEED_CTX_IDX)
3142 context_desc->mss_l4len_idx =
3143 cpu_to_le32(tx_ring->queue_index << 4);
3145 context_desc->mss_l4len_idx = 0;
3147 buffer_info->time_stamp = jiffies;
3148 buffer_info->next_to_watch = i;
3149 buffer_info->dma = 0;
3152 if (i == tx_ring->count)
3154 tx_ring->next_to_use = i;
3161 #define IGB_MAX_TXD_PWR 16
3162 #define IGB_MAX_DATA_PER_TXD (1<<IGB_MAX_TXD_PWR)
3164 static inline int igb_tx_map_adv(struct igb_adapter *adapter,
3165 struct igb_ring *tx_ring, struct sk_buff *skb,
3168 struct igb_buffer *buffer_info;
3169 unsigned int len = skb_headlen(skb);
3170 unsigned int count = 0, i;
3174 i = tx_ring->next_to_use;
3176 if (skb_dma_map(&adapter->pdev->dev, skb, DMA_TO_DEVICE)) {
3177 dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
3181 map = skb_shinfo(skb)->dma_maps;
3183 buffer_info = &tx_ring->buffer_info[i];
3184 BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
3185 buffer_info->length = len;
3186 /* set time_stamp *before* dma to help avoid a possible race */
3187 buffer_info->time_stamp = jiffies;
3188 buffer_info->next_to_watch = i;
3189 buffer_info->dma = skb_shinfo(skb)->dma_head;
3191 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
3192 struct skb_frag_struct *frag;
3195 if (i == tx_ring->count)
3198 frag = &skb_shinfo(skb)->frags[f];
3201 buffer_info = &tx_ring->buffer_info[i];
3202 BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
3203 buffer_info->length = len;
3204 buffer_info->time_stamp = jiffies;
3205 buffer_info->next_to_watch = i;
3206 buffer_info->dma = map[count];
3210 tx_ring->buffer_info[i].skb = skb;
3211 tx_ring->buffer_info[first].next_to_watch = i;
3216 static inline void igb_tx_queue_adv(struct igb_adapter *adapter,
3217 struct igb_ring *tx_ring,
3218 int tx_flags, int count, u32 paylen,
3221 union e1000_adv_tx_desc *tx_desc = NULL;
3222 struct igb_buffer *buffer_info;
3223 u32 olinfo_status = 0, cmd_type_len;
3226 cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
3227 E1000_ADVTXD_DCMD_DEXT);
3229 if (tx_flags & IGB_TX_FLAGS_VLAN)
3230 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
3232 if (tx_flags & IGB_TX_FLAGS_TSTAMP)
3233 cmd_type_len |= E1000_ADVTXD_MAC_TSTAMP;
3235 if (tx_flags & IGB_TX_FLAGS_TSO) {
3236 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
3238 /* insert tcp checksum */
3239 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
3241 /* insert ip checksum */
3242 if (tx_flags & IGB_TX_FLAGS_IPV4)
3243 olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
3245 } else if (tx_flags & IGB_TX_FLAGS_CSUM) {
3246 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
3249 if ((adapter->flags & IGB_FLAG_NEED_CTX_IDX) &&
3250 (tx_flags & (IGB_TX_FLAGS_CSUM | IGB_TX_FLAGS_TSO |
3251 IGB_TX_FLAGS_VLAN)))
3252 olinfo_status |= tx_ring->queue_index << 4;
3254 olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
3256 i = tx_ring->next_to_use;
3258 buffer_info = &tx_ring->buffer_info[i];
3259 tx_desc = E1000_TX_DESC_ADV(*tx_ring, i);
3260 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
3261 tx_desc->read.cmd_type_len =
3262 cpu_to_le32(cmd_type_len | buffer_info->length);
3263 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
3265 if (i == tx_ring->count)
3269 tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
3270 /* Force memory writes to complete before letting h/w
3271 * know there are new descriptors to fetch. (Only
3272 * applicable for weak-ordered memory model archs,
3273 * such as IA-64). */
3276 tx_ring->next_to_use = i;
3277 writel(i, adapter->hw.hw_addr + tx_ring->tail);
3278 /* we need this if more than one processor can write to our tail
3279 * at a time, it syncronizes IO on IA64/Altix systems */
3283 static int __igb_maybe_stop_tx(struct net_device *netdev,
3284 struct igb_ring *tx_ring, int size)
3286 struct igb_adapter *adapter = netdev_priv(netdev);
3288 netif_stop_subqueue(netdev, tx_ring->queue_index);
3290 /* Herbert's original patch had:
3291 * smp_mb__after_netif_stop_queue();
3292 * but since that doesn't exist yet, just open code it. */
3295 /* We need to check again in a case another CPU has just
3296 * made room available. */
3297 if (igb_desc_unused(tx_ring) < size)
3301 netif_wake_subqueue(netdev, tx_ring->queue_index);
3302 ++adapter->restart_queue;
3306 static int igb_maybe_stop_tx(struct net_device *netdev,
3307 struct igb_ring *tx_ring, int size)
3309 if (igb_desc_unused(tx_ring) >= size)
3311 return __igb_maybe_stop_tx(netdev, tx_ring, size);
3314 static netdev_tx_t igb_xmit_frame_ring_adv(struct sk_buff *skb,
3315 struct net_device *netdev,
3316 struct igb_ring *tx_ring)
3318 struct igb_adapter *adapter = netdev_priv(netdev);
3320 unsigned int tx_flags = 0;
3324 union skb_shared_tx *shtx;
3326 if (test_bit(__IGB_DOWN, &adapter->state)) {
3327 dev_kfree_skb_any(skb);
3328 return NETDEV_TX_OK;
3331 if (skb->len <= 0) {
3332 dev_kfree_skb_any(skb);
3333 return NETDEV_TX_OK;
3336 /* need: 1 descriptor per page,
3337 * + 2 desc gap to keep tail from touching head,
3338 * + 1 desc for skb->data,
3339 * + 1 desc for context descriptor,
3340 * otherwise try next time */
3341 if (igb_maybe_stop_tx(netdev, tx_ring, skb_shinfo(skb)->nr_frags + 4)) {
3342 /* this is a hard error */
3343 return NETDEV_TX_BUSY;
3347 * TODO: check that there currently is no other packet with
3348 * time stamping in the queue
3350 * When doing time stamping, keep the connection to the socket
3351 * a while longer: it is still needed by skb_hwtstamp_tx(),
3352 * called either in igb_tx_hwtstamp() or by our caller when
3353 * doing software time stamping.
3356 if (unlikely(shtx->hardware)) {
3357 shtx->in_progress = 1;
3358 tx_flags |= IGB_TX_FLAGS_TSTAMP;
3361 if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
3362 tx_flags |= IGB_TX_FLAGS_VLAN;
3363 tx_flags |= (vlan_tx_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
3366 if (skb->protocol == htons(ETH_P_IP))
3367 tx_flags |= IGB_TX_FLAGS_IPV4;
3369 first = tx_ring->next_to_use;
3370 tso = skb_is_gso(skb) ? igb_tso_adv(adapter, tx_ring, skb, tx_flags,
3374 dev_kfree_skb_any(skb);
3375 return NETDEV_TX_OK;
3379 tx_flags |= IGB_TX_FLAGS_TSO;
3380 else if (igb_tx_csum_adv(adapter, tx_ring, skb, tx_flags) &&
3381 (skb->ip_summed == CHECKSUM_PARTIAL))
3382 tx_flags |= IGB_TX_FLAGS_CSUM;
3385 * count reflects descriptors mapped, if 0 then mapping error
3386 * has occured and we need to rewind the descriptor queue
3388 count = igb_tx_map_adv(adapter, tx_ring, skb, first);
3391 igb_tx_queue_adv(adapter, tx_ring, tx_flags, count,
3393 /* Make sure there is space in the ring for the next send. */
3394 igb_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 4);
3396 dev_kfree_skb_any(skb);
3397 tx_ring->buffer_info[first].time_stamp = 0;
3398 tx_ring->next_to_use = first;
3401 return NETDEV_TX_OK;
3404 static netdev_tx_t igb_xmit_frame_adv(struct sk_buff *skb,
3405 struct net_device *netdev)
3407 struct igb_adapter *adapter = netdev_priv(netdev);
3408 struct igb_ring *tx_ring;
3411 r_idx = skb->queue_mapping & (IGB_ABS_MAX_TX_QUEUES - 1);
3412 tx_ring = adapter->multi_tx_table[r_idx];
3414 /* This goes back to the question of how to logically map a tx queue
3415 * to a flow. Right now, performance is impacted slightly negatively
3416 * if using multiple tx queues. If the stack breaks away from a
3417 * single qdisc implementation, we can look at this again. */
3418 return igb_xmit_frame_ring_adv(skb, netdev, tx_ring);
3422 * igb_tx_timeout - Respond to a Tx Hang
3423 * @netdev: network interface device structure
3425 static void igb_tx_timeout(struct net_device *netdev)
3427 struct igb_adapter *adapter = netdev_priv(netdev);
3428 struct e1000_hw *hw = &adapter->hw;
3430 /* Do the reset outside of interrupt context */
3431 adapter->tx_timeout_count++;
3432 schedule_work(&adapter->reset_task);
3434 (adapter->eims_enable_mask & ~adapter->eims_other));
3437 static void igb_reset_task(struct work_struct *work)
3439 struct igb_adapter *adapter;
3440 adapter = container_of(work, struct igb_adapter, reset_task);
3442 igb_reinit_locked(adapter);
3446 * igb_get_stats - Get System Network Statistics
3447 * @netdev: network interface device structure
3449 * Returns the address of the device statistics structure.
3450 * The statistics are actually updated from the timer callback.
3452 static struct net_device_stats *igb_get_stats(struct net_device *netdev)
3454 struct igb_adapter *adapter = netdev_priv(netdev);
3456 /* only return the current stats */
3457 return &adapter->net_stats;
3461 * igb_change_mtu - Change the Maximum Transfer Unit
3462 * @netdev: network interface device structure
3463 * @new_mtu: new value for maximum frame size
3465 * Returns 0 on success, negative on failure
3467 static int igb_change_mtu(struct net_device *netdev, int new_mtu)
3469 struct igb_adapter *adapter = netdev_priv(netdev);
3470 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3472 if ((max_frame < ETH_ZLEN + ETH_FCS_LEN) ||
3473 (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3474 dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
3478 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
3479 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
3483 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
3486 /* igb_down has a dependency on max_frame_size */
3487 adapter->max_frame_size = max_frame;
3488 if (netif_running(netdev))
3491 /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3492 * means we reserve 2 more, this pushes us to allocate from the next
3494 * i.e. RXBUFFER_2048 --> size-4096 slab
3497 if (max_frame <= IGB_RXBUFFER_256)
3498 adapter->rx_buffer_len = IGB_RXBUFFER_256;
3499 else if (max_frame <= IGB_RXBUFFER_512)
3500 adapter->rx_buffer_len = IGB_RXBUFFER_512;
3501 else if (max_frame <= IGB_RXBUFFER_1024)
3502 adapter->rx_buffer_len = IGB_RXBUFFER_1024;
3503 else if (max_frame <= IGB_RXBUFFER_2048)
3504 adapter->rx_buffer_len = IGB_RXBUFFER_2048;
3506 #if (PAGE_SIZE / 2) > IGB_RXBUFFER_16384
3507 adapter->rx_buffer_len = IGB_RXBUFFER_16384;
3509 adapter->rx_buffer_len = PAGE_SIZE / 2;
3512 /* if sr-iov is enabled we need to force buffer size to 1K or larger */
3513 if (adapter->vfs_allocated_count &&
3514 (adapter->rx_buffer_len < IGB_RXBUFFER_1024))
3515 adapter->rx_buffer_len = IGB_RXBUFFER_1024;
3517 /* adjust allocation if LPE protects us, and we aren't using SBP */
3518 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
3519 (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE))
3520 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3522 dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
3523 netdev->mtu, new_mtu);
3524 netdev->mtu = new_mtu;
3526 if (netif_running(netdev))
3531 clear_bit(__IGB_RESETTING, &adapter->state);
3537 * igb_update_stats - Update the board statistics counters
3538 * @adapter: board private structure
3541 void igb_update_stats(struct igb_adapter *adapter)
3543 struct e1000_hw *hw = &adapter->hw;
3544 struct pci_dev *pdev = adapter->pdev;
3547 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3550 * Prevent stats update while adapter is being reset, or if the pci
3551 * connection is down.
3553 if (adapter->link_speed == 0)
3555 if (pci_channel_offline(pdev))
3558 adapter->stats.crcerrs += rd32(E1000_CRCERRS);
3559 adapter->stats.gprc += rd32(E1000_GPRC);
3560 adapter->stats.gorc += rd32(E1000_GORCL);
3561 rd32(E1000_GORCH); /* clear GORCL */
3562 adapter->stats.bprc += rd32(E1000_BPRC);
3563 adapter->stats.mprc += rd32(E1000_MPRC);
3564 adapter->stats.roc += rd32(E1000_ROC);
3566 adapter->stats.prc64 += rd32(E1000_PRC64);
3567 adapter->stats.prc127 += rd32(E1000_PRC127);
3568 adapter->stats.prc255 += rd32(E1000_PRC255);
3569 adapter->stats.prc511 += rd32(E1000_PRC511);
3570 adapter->stats.prc1023 += rd32(E1000_PRC1023);
3571 adapter->stats.prc1522 += rd32(E1000_PRC1522);
3572 adapter->stats.symerrs += rd32(E1000_SYMERRS);
3573 adapter->stats.sec += rd32(E1000_SEC);
3575 adapter->stats.mpc += rd32(E1000_MPC);
3576 adapter->stats.scc += rd32(E1000_SCC);
3577 adapter->stats.ecol += rd32(E1000_ECOL);
3578 adapter->stats.mcc += rd32(E1000_MCC);
3579 adapter->stats.latecol += rd32(E1000_LATECOL);
3580 adapter->stats.dc += rd32(E1000_DC);
3581 adapter->stats.rlec += rd32(E1000_RLEC);
3582 adapter->stats.xonrxc += rd32(E1000_XONRXC);
3583 adapter->stats.xontxc += rd32(E1000_XONTXC);
3584 adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
3585 adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
3586 adapter->stats.fcruc += rd32(E1000_FCRUC);
3587 adapter->stats.gptc += rd32(E1000_GPTC);
3588 adapter->stats.gotc += rd32(E1000_GOTCL);
3589 rd32(E1000_GOTCH); /* clear GOTCL */
3590 adapter->stats.rnbc += rd32(E1000_RNBC);
3591 adapter->stats.ruc += rd32(E1000_RUC);
3592 adapter->stats.rfc += rd32(E1000_RFC);
3593 adapter->stats.rjc += rd32(E1000_RJC);
3594 adapter->stats.tor += rd32(E1000_TORH);
3595 adapter->stats.tot += rd32(E1000_TOTH);
3596 adapter->stats.tpr += rd32(E1000_TPR);
3598 adapter->stats.ptc64 += rd32(E1000_PTC64);
3599 adapter->stats.ptc127 += rd32(E1000_PTC127);
3600 adapter->stats.ptc255 += rd32(E1000_PTC255);
3601 adapter->stats.ptc511 += rd32(E1000_PTC511);
3602 adapter->stats.ptc1023 += rd32(E1000_PTC1023);
3603 adapter->stats.ptc1522 += rd32(E1000_PTC1522);
3605 adapter->stats.mptc += rd32(E1000_MPTC);
3606 adapter->stats.bptc += rd32(E1000_BPTC);
3608 /* used for adaptive IFS */
3610 hw->mac.tx_packet_delta = rd32(E1000_TPT);
3611 adapter->stats.tpt += hw->mac.tx_packet_delta;
3612 hw->mac.collision_delta = rd32(E1000_COLC);
3613 adapter->stats.colc += hw->mac.collision_delta;
3615 adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
3616 adapter->stats.rxerrc += rd32(E1000_RXERRC);
3617 adapter->stats.tncrs += rd32(E1000_TNCRS);
3618 adapter->stats.tsctc += rd32(E1000_TSCTC);
3619 adapter->stats.tsctfc += rd32(E1000_TSCTFC);
3621 adapter->stats.iac += rd32(E1000_IAC);
3622 adapter->stats.icrxoc += rd32(E1000_ICRXOC);
3623 adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
3624 adapter->stats.icrxatc += rd32(E1000_ICRXATC);
3625 adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
3626 adapter->stats.ictxatc += rd32(E1000_ICTXATC);
3627 adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
3628 adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
3629 adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
3631 /* Fill out the OS statistics structure */
3632 adapter->net_stats.multicast = adapter->stats.mprc;
3633 adapter->net_stats.collisions = adapter->stats.colc;
3637 if (hw->mac.type != e1000_82575) {
3639 u64 rqdpc_total = 0;
3641 /* Read out drops stats per RX queue. Notice RQDPC (Receive
3642 * Queue Drop Packet Count) stats only gets incremented, if
3643 * the DROP_EN but it set (in the SRRCTL register for that
3644 * queue). If DROP_EN bit is NOT set, then the some what
3645 * equivalent count is stored in RNBC (not per queue basis).
3646 * Also note the drop count is due to lack of available
3649 for (i = 0; i < adapter->num_rx_queues; i++) {
3650 rqdpc_tmp = rd32(E1000_RQDPC(i)) & 0xFFF;
3651 adapter->rx_ring[i].rx_stats.drops += rqdpc_tmp;
3652 rqdpc_total += adapter->rx_ring[i].rx_stats.drops;
3654 adapter->net_stats.rx_fifo_errors = rqdpc_total;
3657 /* Note RNBC (Receive No Buffers Count) is an not an exact
3658 * drop count as the hardware FIFO might save the day. Thats
3659 * one of the reason for saving it in rx_fifo_errors, as its
3660 * potentially not a true drop.
3662 adapter->net_stats.rx_fifo_errors += adapter->stats.rnbc;
3664 /* RLEC on some newer hardware can be incorrect so build
3665 * our own version based on RUC and ROC */
3666 adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3667 adapter->stats.crcerrs + adapter->stats.algnerrc +
3668 adapter->stats.ruc + adapter->stats.roc +
3669 adapter->stats.cexterr;
3670 adapter->net_stats.rx_length_errors = adapter->stats.ruc +
3672 adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3673 adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3674 adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3677 adapter->net_stats.tx_errors = adapter->stats.ecol +
3678 adapter->stats.latecol;
3679 adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3680 adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3681 adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3683 /* Tx Dropped needs to be maintained elsewhere */
3686 if (hw->phy.media_type == e1000_media_type_copper) {
3687 if ((adapter->link_speed == SPEED_1000) &&
3688 (!igb_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
3689 phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3690 adapter->phy_stats.idle_errors += phy_tmp;
3694 /* Management Stats */
3695 adapter->stats.mgptc += rd32(E1000_MGTPTC);
3696 adapter->stats.mgprc += rd32(E1000_MGTPRC);
3697 adapter->stats.mgpdc += rd32(E1000_MGTPDC);
3700 static irqreturn_t igb_msix_other(int irq, void *data)
3702 struct net_device *netdev = data;
3703 struct igb_adapter *adapter = netdev_priv(netdev);
3704 struct e1000_hw *hw = &adapter->hw;
3705 u32 icr = rd32(E1000_ICR);
3707 /* reading ICR causes bit 31 of EICR to be cleared */
3709 if(icr & E1000_ICR_DOUTSYNC) {
3710 /* HW is reporting DMA is out of sync */
3711 adapter->stats.doosync++;
3714 /* Check for a mailbox event */
3715 if (icr & E1000_ICR_VMMB)
3716 igb_msg_task(adapter);
3718 if (icr & E1000_ICR_LSC) {
3719 hw->mac.get_link_status = 1;
3720 /* guard against interrupt when we're going down */
3721 if (!test_bit(__IGB_DOWN, &adapter->state))
3722 mod_timer(&adapter->watchdog_timer, jiffies + 1);
3725 wr32(E1000_IMS, E1000_IMS_LSC | E1000_IMS_DOUTSYNC | E1000_IMS_VMMB);
3726 wr32(E1000_EIMS, adapter->eims_other);
3731 static irqreturn_t igb_msix_tx(int irq, void *data)
3733 struct igb_ring *tx_ring = data;
3734 struct igb_adapter *adapter = tx_ring->adapter;
3735 struct e1000_hw *hw = &adapter->hw;
3737 #ifdef CONFIG_IGB_DCA
3738 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
3739 igb_update_tx_dca(tx_ring);
3742 tx_ring->total_bytes = 0;
3743 tx_ring->total_packets = 0;
3745 /* auto mask will automatically reenable the interrupt when we write
3747 if (!igb_clean_tx_irq(tx_ring))
3748 /* Ring was not completely cleaned, so fire another interrupt */
3749 wr32(E1000_EICS, tx_ring->eims_value);
3751 wr32(E1000_EIMS, tx_ring->eims_value);
3756 static void igb_write_itr(struct igb_ring *ring)
3758 struct e1000_hw *hw = &ring->adapter->hw;
3759 if ((ring->adapter->itr_setting & 3) && ring->set_itr) {
3760 switch (hw->mac.type) {
3762 wr32(ring->itr_register, ring->itr_val |
3766 wr32(ring->itr_register, ring->itr_val |
3767 (ring->itr_val << 16));
3774 static irqreturn_t igb_msix_rx(int irq, void *data)
3776 struct igb_ring *rx_ring = data;
3778 /* Write the ITR value calculated at the end of the
3779 * previous interrupt.
3782 igb_write_itr(rx_ring);
3784 if (napi_schedule_prep(&rx_ring->napi))
3785 __napi_schedule(&rx_ring->napi);
3787 #ifdef CONFIG_IGB_DCA
3788 if (rx_ring->adapter->flags & IGB_FLAG_DCA_ENABLED)
3789 igb_update_rx_dca(rx_ring);
3794 #ifdef CONFIG_IGB_DCA
3795 static void igb_update_rx_dca(struct igb_ring *rx_ring)
3798 struct igb_adapter *adapter = rx_ring->adapter;
3799 struct e1000_hw *hw = &adapter->hw;
3800 int cpu = get_cpu();
3801 int q = rx_ring->reg_idx;
3803 if (rx_ring->cpu != cpu) {
3804 dca_rxctrl = rd32(E1000_DCA_RXCTRL(q));
3805 if (hw->mac.type == e1000_82576) {
3806 dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK_82576;
3807 dca_rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu) <<
3808 E1000_DCA_RXCTRL_CPUID_SHIFT;
3810 dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK;
3811 dca_rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
3813 dca_rxctrl |= E1000_DCA_RXCTRL_DESC_DCA_EN;
3814 dca_rxctrl |= E1000_DCA_RXCTRL_HEAD_DCA_EN;
3815 dca_rxctrl |= E1000_DCA_RXCTRL_DATA_DCA_EN;
3816 wr32(E1000_DCA_RXCTRL(q), dca_rxctrl);
3822 static void igb_update_tx_dca(struct igb_ring *tx_ring)
3825 struct igb_adapter *adapter = tx_ring->adapter;
3826 struct e1000_hw *hw = &adapter->hw;
3827 int cpu = get_cpu();
3828 int q = tx_ring->reg_idx;
3830 if (tx_ring->cpu != cpu) {
3831 dca_txctrl = rd32(E1000_DCA_TXCTRL(q));
3832 if (hw->mac.type == e1000_82576) {
3833 dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK_82576;
3834 dca_txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu) <<
3835 E1000_DCA_TXCTRL_CPUID_SHIFT;
3837 dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK;
3838 dca_txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
3840 dca_txctrl |= E1000_DCA_TXCTRL_DESC_DCA_EN;
3841 wr32(E1000_DCA_TXCTRL(q), dca_txctrl);
3847 static void igb_setup_dca(struct igb_adapter *adapter)
3849 struct e1000_hw *hw = &adapter->hw;
3852 if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
3855 /* Always use CB2 mode, difference is masked in the CB driver. */
3856 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
3858 for (i = 0; i < adapter->num_tx_queues; i++) {
3859 adapter->tx_ring[i].cpu = -1;
3860 igb_update_tx_dca(&adapter->tx_ring[i]);
3862 for (i = 0; i < adapter->num_rx_queues; i++) {
3863 adapter->rx_ring[i].cpu = -1;
3864 igb_update_rx_dca(&adapter->rx_ring[i]);
3868 static int __igb_notify_dca(struct device *dev, void *data)
3870 struct net_device *netdev = dev_get_drvdata(dev);
3871 struct igb_adapter *adapter = netdev_priv(netdev);
3872 struct e1000_hw *hw = &adapter->hw;
3873 unsigned long event = *(unsigned long *)data;
3876 case DCA_PROVIDER_ADD:
3877 /* if already enabled, don't do it again */
3878 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
3880 /* Always use CB2 mode, difference is masked
3881 * in the CB driver. */
3882 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
3883 if (dca_add_requester(dev) == 0) {
3884 adapter->flags |= IGB_FLAG_DCA_ENABLED;
3885 dev_info(&adapter->pdev->dev, "DCA enabled\n");
3886 igb_setup_dca(adapter);
3889 /* Fall Through since DCA is disabled. */
3890 case DCA_PROVIDER_REMOVE:
3891 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
3892 /* without this a class_device is left
3893 * hanging around in the sysfs model */
3894 dca_remove_requester(dev);
3895 dev_info(&adapter->pdev->dev, "DCA disabled\n");
3896 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
3897 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
3905 static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
3910 ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
3913 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
3915 #endif /* CONFIG_IGB_DCA */
3917 static void igb_ping_all_vfs(struct igb_adapter *adapter)
3919 struct e1000_hw *hw = &adapter->hw;
3923 for (i = 0 ; i < adapter->vfs_allocated_count; i++) {
3924 ping = E1000_PF_CONTROL_MSG;
3925 if (adapter->vf_data[i].clear_to_send)
3926 ping |= E1000_VT_MSGTYPE_CTS;
3927 igb_write_mbx(hw, &ping, 1, i);
3931 static int igb_set_vf_multicasts(struct igb_adapter *adapter,
3932 u32 *msgbuf, u32 vf)
3934 int n = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
3935 u16 *hash_list = (u16 *)&msgbuf[1];
3936 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
3939 /* only up to 30 hash values supported */
3943 /* salt away the number of multi cast addresses assigned
3944 * to this VF for later use to restore when the PF multi cast
3947 vf_data->num_vf_mc_hashes = n;
3949 /* VFs are limited to using the MTA hash table for their multicast
3951 for (i = 0; i < n; i++)
3952 vf_data->vf_mc_hashes[i] = hash_list[i];
3954 /* Flush and reset the mta with the new values */
3955 igb_set_rx_mode(adapter->netdev);
3960 static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
3962 struct e1000_hw *hw = &adapter->hw;
3963 struct vf_data_storage *vf_data;
3966 for (i = 0; i < adapter->vfs_allocated_count; i++) {
3967 vf_data = &adapter->vf_data[i];
3968 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
3969 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
3973 static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
3975 struct e1000_hw *hw = &adapter->hw;
3976 u32 pool_mask, reg, vid;
3979 pool_mask = 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
3981 /* Find the vlan filter for this id */
3982 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
3983 reg = rd32(E1000_VLVF(i));
3985 /* remove the vf from the pool */
3988 /* if pool is empty then remove entry from vfta */
3989 if (!(reg & E1000_VLVF_POOLSEL_MASK) &&
3990 (reg & E1000_VLVF_VLANID_ENABLE)) {
3992 vid = reg & E1000_VLVF_VLANID_MASK;
3993 igb_vfta_set(hw, vid, false);
3996 wr32(E1000_VLVF(i), reg);
3999 adapter->vf_data[vf].vlans_enabled = 0;
4002 static s32 igb_vlvf_set(struct igb_adapter *adapter, u32 vid, bool add, u32 vf)
4004 struct e1000_hw *hw = &adapter->hw;
4007 /* It is an error to call this function when VFs are not enabled */
4008 if (!adapter->vfs_allocated_count)
4011 /* Find the vlan filter for this id */
4012 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
4013 reg = rd32(E1000_VLVF(i));
4014 if ((reg & E1000_VLVF_VLANID_ENABLE) &&
4015 vid == (reg & E1000_VLVF_VLANID_MASK))
4020 if (i == E1000_VLVF_ARRAY_SIZE) {
4021 /* Did not find a matching VLAN ID entry that was
4022 * enabled. Search for a free filter entry, i.e.
4023 * one without the enable bit set
4025 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
4026 reg = rd32(E1000_VLVF(i));
4027 if (!(reg & E1000_VLVF_VLANID_ENABLE))
4031 if (i < E1000_VLVF_ARRAY_SIZE) {
4032 /* Found an enabled/available entry */
4033 reg |= 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
4035 /* if !enabled we need to set this up in vfta */
4036 if (!(reg & E1000_VLVF_VLANID_ENABLE)) {
4037 /* add VID to filter table, if bit already set
4038 * PF must have added it outside of table */
4039 if (igb_vfta_set(hw, vid, true))
4040 reg |= 1 << (E1000_VLVF_POOLSEL_SHIFT +
4041 adapter->vfs_allocated_count);
4042 reg |= E1000_VLVF_VLANID_ENABLE;
4044 reg &= ~E1000_VLVF_VLANID_MASK;
4047 wr32(E1000_VLVF(i), reg);
4049 /* do not modify RLPML for PF devices */
4050 if (vf >= adapter->vfs_allocated_count)
4053 if (!adapter->vf_data[vf].vlans_enabled) {
4055 reg = rd32(E1000_VMOLR(vf));
4056 size = reg & E1000_VMOLR_RLPML_MASK;
4058 reg &= ~E1000_VMOLR_RLPML_MASK;
4060 wr32(E1000_VMOLR(vf), reg);
4062 adapter->vf_data[vf].vlans_enabled++;
4067 if (i < E1000_VLVF_ARRAY_SIZE) {
4068 /* remove vf from the pool */
4069 reg &= ~(1 << (E1000_VLVF_POOLSEL_SHIFT + vf));
4070 /* if pool is empty then remove entry from vfta */
4071 if (!(reg & E1000_VLVF_POOLSEL_MASK)) {
4073 igb_vfta_set(hw, vid, false);
4075 wr32(E1000_VLVF(i), reg);
4077 /* do not modify RLPML for PF devices */
4078 if (vf >= adapter->vfs_allocated_count)
4081 adapter->vf_data[vf].vlans_enabled--;
4082 if (!adapter->vf_data[vf].vlans_enabled) {
4084 reg = rd32(E1000_VMOLR(vf));
4085 size = reg & E1000_VMOLR_RLPML_MASK;
4087 reg &= ~E1000_VMOLR_RLPML_MASK;
4089 wr32(E1000_VMOLR(vf), reg);
4097 static int igb_set_vf_vlan(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
4099 int add = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
4100 int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
4102 return igb_vlvf_set(adapter, vid, add, vf);
4105 static inline void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
4107 struct e1000_hw *hw = &adapter->hw;
4109 /* disable mailbox functionality for vf */
4110 adapter->vf_data[vf].clear_to_send = false;
4112 /* reset offloads to defaults */
4113 igb_set_vmolr(hw, vf);
4115 /* reset vlans for device */
4116 igb_clear_vf_vfta(adapter, vf);
4118 /* reset multicast table array for vf */
4119 adapter->vf_data[vf].num_vf_mc_hashes = 0;
4121 /* Flush and reset the mta with the new values */
4122 igb_set_rx_mode(adapter->netdev);
4125 static inline void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
4127 struct e1000_hw *hw = &adapter->hw;
4128 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
4129 int rar_entry = hw->mac.rar_entry_count - (vf + 1);
4131 u8 *addr = (u8 *)(&msgbuf[1]);
4133 /* process all the same items cleared in a function level reset */
4134 igb_vf_reset_event(adapter, vf);
4136 /* set vf mac address */
4137 igb_rar_set_qsel(adapter, vf_mac, rar_entry, vf);
4139 /* enable transmit and receive for vf */
4140 reg = rd32(E1000_VFTE);
4141 wr32(E1000_VFTE, reg | (1 << vf));
4142 reg = rd32(E1000_VFRE);
4143 wr32(E1000_VFRE, reg | (1 << vf));
4145 /* enable mailbox functionality for vf */
4146 adapter->vf_data[vf].clear_to_send = true;
4148 /* reply to reset with ack and vf mac address */
4149 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
4150 memcpy(addr, vf_mac, 6);
4151 igb_write_mbx(hw, msgbuf, 3, vf);
4154 static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf)
4156 unsigned char *addr = (char *)&msg[1];
4159 if (is_valid_ether_addr(addr))
4160 err = igb_set_vf_mac(adapter, vf, addr);
4166 static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf)
4168 struct e1000_hw *hw = &adapter->hw;
4169 u32 msg = E1000_VT_MSGTYPE_NACK;
4171 /* if device isn't clear to send it shouldn't be reading either */
4172 if (!adapter->vf_data[vf].clear_to_send)
4173 igb_write_mbx(hw, &msg, 1, vf);
4177 static void igb_msg_task(struct igb_adapter *adapter)
4179 struct e1000_hw *hw = &adapter->hw;
4182 for (vf = 0; vf < adapter->vfs_allocated_count; vf++) {
4183 /* process any reset requests */
4184 if (!igb_check_for_rst(hw, vf)) {
4185 adapter->vf_data[vf].clear_to_send = false;
4186 igb_vf_reset_event(adapter, vf);
4189 /* process any messages pending */
4190 if (!igb_check_for_msg(hw, vf))
4191 igb_rcv_msg_from_vf(adapter, vf);
4193 /* process any acks */
4194 if (!igb_check_for_ack(hw, vf))
4195 igb_rcv_ack_from_vf(adapter, vf);
4200 static int igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
4202 u32 mbx_size = E1000_VFMAILBOX_SIZE;
4203 u32 msgbuf[mbx_size];
4204 struct e1000_hw *hw = &adapter->hw;
4207 retval = igb_read_mbx(hw, msgbuf, mbx_size, vf);
4210 dev_err(&adapter->pdev->dev,
4211 "Error receiving message from VF\n");
4213 /* this is a message we already processed, do nothing */
4214 if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
4218 * until the vf completes a reset it should not be
4219 * allowed to start any configuration.
4222 if (msgbuf[0] == E1000_VF_RESET) {
4223 igb_vf_reset_msg(adapter, vf);
4228 if (!adapter->vf_data[vf].clear_to_send) {
4229 msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
4230 igb_write_mbx(hw, msgbuf, 1, vf);
4234 switch ((msgbuf[0] & 0xFFFF)) {
4235 case E1000_VF_SET_MAC_ADDR:
4236 retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
4238 case E1000_VF_SET_MULTICAST:
4239 retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
4241 case E1000_VF_SET_LPE:
4242 retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf);
4244 case E1000_VF_SET_VLAN:
4245 retval = igb_set_vf_vlan(adapter, msgbuf, vf);
4248 dev_err(&adapter->pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
4253 /* notify the VF of the results of what it sent us */
4255 msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
4257 msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
4259 msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
4261 igb_write_mbx(hw, msgbuf, 1, vf);
4267 * igb_intr_msi - Interrupt Handler
4268 * @irq: interrupt number
4269 * @data: pointer to a network interface device structure
4271 static irqreturn_t igb_intr_msi(int irq, void *data)
4273 struct net_device *netdev = data;
4274 struct igb_adapter *adapter = netdev_priv(netdev);
4275 struct e1000_hw *hw = &adapter->hw;
4276 /* read ICR disables interrupts using IAM */
4277 u32 icr = rd32(E1000_ICR);
4279 igb_write_itr(adapter->rx_ring);
4281 if(icr & E1000_ICR_DOUTSYNC) {
4282 /* HW is reporting DMA is out of sync */
4283 adapter->stats.doosync++;
4286 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
4287 hw->mac.get_link_status = 1;
4288 if (!test_bit(__IGB_DOWN, &adapter->state))
4289 mod_timer(&adapter->watchdog_timer, jiffies + 1);
4292 napi_schedule(&adapter->rx_ring[0].napi);
4298 * igb_intr - Legacy Interrupt Handler
4299 * @irq: interrupt number
4300 * @data: pointer to a network interface device structure
4302 static irqreturn_t igb_intr(int irq, void *data)
4304 struct net_device *netdev = data;
4305 struct igb_adapter *adapter = netdev_priv(netdev);
4306 struct e1000_hw *hw = &adapter->hw;
4307 /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked. No
4308 * need for the IMC write */
4309 u32 icr = rd32(E1000_ICR);
4311 return IRQ_NONE; /* Not our interrupt */
4313 igb_write_itr(adapter->rx_ring);
4315 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
4316 * not set, then the adapter didn't send an interrupt */
4317 if (!(icr & E1000_ICR_INT_ASSERTED))
4320 if(icr & E1000_ICR_DOUTSYNC) {
4321 /* HW is reporting DMA is out of sync */
4322 adapter->stats.doosync++;
4325 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
4326 hw->mac.get_link_status = 1;
4327 /* guard against interrupt when we're going down */
4328 if (!test_bit(__IGB_DOWN, &adapter->state))
4329 mod_timer(&adapter->watchdog_timer, jiffies + 1);
4332 napi_schedule(&adapter->rx_ring[0].napi);
4337 static inline void igb_rx_irq_enable(struct igb_ring *rx_ring)
4339 struct igb_adapter *adapter = rx_ring->adapter;
4340 struct e1000_hw *hw = &adapter->hw;
4342 if (adapter->itr_setting & 3) {
4343 if (adapter->num_rx_queues == 1)
4344 igb_set_itr(adapter);
4346 igb_update_ring_itr(rx_ring);
4349 if (!test_bit(__IGB_DOWN, &adapter->state)) {
4350 if (adapter->msix_entries)
4351 wr32(E1000_EIMS, rx_ring->eims_value);
4353 igb_irq_enable(adapter);
4358 * igb_poll - NAPI Rx polling callback
4359 * @napi: napi polling structure
4360 * @budget: count of how many packets we should handle
4362 static int igb_poll(struct napi_struct *napi, int budget)
4364 struct igb_ring *rx_ring = container_of(napi, struct igb_ring, napi);
4367 #ifdef CONFIG_IGB_DCA
4368 if (rx_ring->adapter->flags & IGB_FLAG_DCA_ENABLED)
4369 igb_update_rx_dca(rx_ring);
4371 igb_clean_rx_irq_adv(rx_ring, &work_done, budget);
4373 if (rx_ring->buddy) {
4374 #ifdef CONFIG_IGB_DCA
4375 if (rx_ring->adapter->flags & IGB_FLAG_DCA_ENABLED)
4376 igb_update_tx_dca(rx_ring->buddy);
4378 if (!igb_clean_tx_irq(rx_ring->buddy))
4382 /* If not enough Rx work done, exit the polling mode */
4383 if (work_done < budget) {
4384 napi_complete(napi);
4385 igb_rx_irq_enable(rx_ring);
4392 * igb_hwtstamp - utility function which checks for TX time stamp
4393 * @adapter: board private structure
4394 * @skb: packet that was just sent
4396 * If we were asked to do hardware stamping and such a time stamp is
4397 * available, then it must have been for this skb here because we only
4398 * allow only one such packet into the queue.
4400 static void igb_tx_hwtstamp(struct igb_adapter *adapter, struct sk_buff *skb)
4402 union skb_shared_tx *shtx = skb_tx(skb);
4403 struct e1000_hw *hw = &adapter->hw;
4405 if (unlikely(shtx->hardware)) {
4406 u32 valid = rd32(E1000_TSYNCTXCTL) & E1000_TSYNCTXCTL_VALID;
4408 u64 regval = rd32(E1000_TXSTMPL);
4410 struct skb_shared_hwtstamps shhwtstamps;
4412 memset(&shhwtstamps, 0, sizeof(shhwtstamps));
4413 regval |= (u64)rd32(E1000_TXSTMPH) << 32;
4414 ns = timecounter_cyc2time(&adapter->clock,
4416 timecompare_update(&adapter->compare, ns);
4417 shhwtstamps.hwtstamp = ns_to_ktime(ns);
4418 shhwtstamps.syststamp =
4419 timecompare_transform(&adapter->compare, ns);
4420 skb_tstamp_tx(skb, &shhwtstamps);
4426 * igb_clean_tx_irq - Reclaim resources after transmit completes
4427 * @adapter: board private structure
4428 * returns true if ring is completely cleaned
4430 static bool igb_clean_tx_irq(struct igb_ring *tx_ring)
4432 struct igb_adapter *adapter = tx_ring->adapter;
4433 struct net_device *netdev = adapter->netdev;
4434 struct e1000_hw *hw = &adapter->hw;
4435 struct igb_buffer *buffer_info;
4436 struct sk_buff *skb;
4437 union e1000_adv_tx_desc *tx_desc, *eop_desc;
4438 unsigned int total_bytes = 0, total_packets = 0;
4439 unsigned int i, eop, count = 0;
4440 bool cleaned = false;
4442 i = tx_ring->next_to_clean;
4443 eop = tx_ring->buffer_info[i].next_to_watch;
4444 eop_desc = E1000_TX_DESC_ADV(*tx_ring, eop);
4446 while ((eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)) &&
4447 (count < tx_ring->count)) {
4448 for (cleaned = false; !cleaned; count++) {
4449 tx_desc = E1000_TX_DESC_ADV(*tx_ring, i);
4450 buffer_info = &tx_ring->buffer_info[i];
4451 cleaned = (i == eop);
4452 skb = buffer_info->skb;
4455 unsigned int segs, bytecount;
4456 /* gso_segs is currently only valid for tcp */
4457 segs = skb_shinfo(skb)->gso_segs ?: 1;
4458 /* multiply data chunks by size of headers */
4459 bytecount = ((segs - 1) * skb_headlen(skb)) +
4461 total_packets += segs;
4462 total_bytes += bytecount;
4464 igb_tx_hwtstamp(adapter, skb);
4467 igb_unmap_and_free_tx_resource(adapter, buffer_info);
4468 tx_desc->wb.status = 0;
4471 if (i == tx_ring->count)
4474 eop = tx_ring->buffer_info[i].next_to_watch;
4475 eop_desc = E1000_TX_DESC_ADV(*tx_ring, eop);
4478 tx_ring->next_to_clean = i;
4480 if (unlikely(count &&
4481 netif_carrier_ok(netdev) &&
4482 igb_desc_unused(tx_ring) >= IGB_TX_QUEUE_WAKE)) {
4483 /* Make sure that anybody stopping the queue after this
4484 * sees the new next_to_clean.
4487 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
4488 !(test_bit(__IGB_DOWN, &adapter->state))) {
4489 netif_wake_subqueue(netdev, tx_ring->queue_index);
4490 ++adapter->restart_queue;
4494 if (tx_ring->detect_tx_hung) {
4495 /* Detect a transmit hang in hardware, this serializes the
4496 * check with the clearing of time_stamp and movement of i */
4497 tx_ring->detect_tx_hung = false;
4498 if (tx_ring->buffer_info[i].time_stamp &&
4499 time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
4500 (adapter->tx_timeout_factor * HZ))
4501 && !(rd32(E1000_STATUS) &
4502 E1000_STATUS_TXOFF)) {
4504 /* detected Tx unit hang */
4505 dev_err(&adapter->pdev->dev,
4506 "Detected Tx Unit Hang\n"
4510 " next_to_use <%x>\n"
4511 " next_to_clean <%x>\n"
4512 "buffer_info[next_to_clean]\n"
4513 " time_stamp <%lx>\n"
4514 " next_to_watch <%x>\n"
4516 " desc.status <%x>\n",
4517 tx_ring->queue_index,
4518 readl(adapter->hw.hw_addr + tx_ring->head),
4519 readl(adapter->hw.hw_addr + tx_ring->tail),
4520 tx_ring->next_to_use,
4521 tx_ring->next_to_clean,
4522 tx_ring->buffer_info[i].time_stamp,
4525 eop_desc->wb.status);
4526 netif_stop_subqueue(netdev, tx_ring->queue_index);
4529 tx_ring->total_bytes += total_bytes;
4530 tx_ring->total_packets += total_packets;
4531 tx_ring->tx_stats.bytes += total_bytes;
4532 tx_ring->tx_stats.packets += total_packets;
4533 adapter->net_stats.tx_bytes += total_bytes;
4534 adapter->net_stats.tx_packets += total_packets;
4535 return (count < tx_ring->count);
4539 * igb_receive_skb - helper function to handle rx indications
4540 * @ring: pointer to receive ring receving this packet
4541 * @status: descriptor status field as written by hardware
4542 * @rx_desc: receive descriptor containing vlan and type information.
4543 * @skb: pointer to sk_buff to be indicated to stack
4545 static void igb_receive_skb(struct igb_ring *ring, u8 status,
4546 union e1000_adv_rx_desc * rx_desc,
4547 struct sk_buff *skb)
4549 struct igb_adapter * adapter = ring->adapter;
4550 bool vlan_extracted = (adapter->vlgrp && (status & E1000_RXD_STAT_VP));
4552 skb_record_rx_queue(skb, ring->queue_index);
4554 vlan_gro_receive(&ring->napi, adapter->vlgrp,
4555 le16_to_cpu(rx_desc->wb.upper.vlan),
4558 napi_gro_receive(&ring->napi, skb);
4561 static inline void igb_rx_checksum_adv(struct igb_adapter *adapter,
4562 u32 status_err, struct sk_buff *skb)
4564 skb->ip_summed = CHECKSUM_NONE;
4566 /* Ignore Checksum bit is set or checksum is disabled through ethtool */
4567 if ((status_err & E1000_RXD_STAT_IXSM) ||
4568 (adapter->flags & IGB_FLAG_RX_CSUM_DISABLED))
4570 /* TCP/UDP checksum error bit is set */
4572 (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
4574 * work around errata with sctp packets where the TCPE aka
4575 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
4576 * packets, (aka let the stack check the crc32c)
4578 if (!((adapter->hw.mac.type == e1000_82576) &&
4580 adapter->hw_csum_err++;
4581 /* let the stack verify checksum errors */
4584 /* It must be a TCP or UDP packet with a valid checksum */
4585 if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
4586 skb->ip_summed = CHECKSUM_UNNECESSARY;
4588 dev_dbg(&adapter->pdev->dev, "cksum success: bits %08X\n", status_err);
4589 adapter->hw_csum_good++;
4592 static inline u16 igb_get_hlen(struct igb_adapter *adapter,
4593 union e1000_adv_rx_desc *rx_desc)
4595 /* HW will not DMA in data larger than the given buffer, even if it
4596 * parses the (NFS, of course) header to be larger. In that case, it
4597 * fills the header buffer and spills the rest into the page.
4599 u16 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hdr_info) &
4600 E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
4601 if (hlen > adapter->rx_ps_hdr_size)
4602 hlen = adapter->rx_ps_hdr_size;
4606 static bool igb_clean_rx_irq_adv(struct igb_ring *rx_ring,
4607 int *work_done, int budget)
4609 struct igb_adapter *adapter = rx_ring->adapter;
4610 struct net_device *netdev = adapter->netdev;
4611 struct e1000_hw *hw = &adapter->hw;
4612 struct pci_dev *pdev = adapter->pdev;
4613 union e1000_adv_rx_desc *rx_desc , *next_rxd;
4614 struct igb_buffer *buffer_info , *next_buffer;
4615 struct sk_buff *skb;
4616 bool cleaned = false;
4617 int cleaned_count = 0;
4618 unsigned int total_bytes = 0, total_packets = 0;
4623 i = rx_ring->next_to_clean;
4624 buffer_info = &rx_ring->buffer_info[i];
4625 rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
4626 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
4628 while (staterr & E1000_RXD_STAT_DD) {
4629 if (*work_done >= budget)
4633 skb = buffer_info->skb;
4634 prefetch(skb->data - NET_IP_ALIGN);
4635 buffer_info->skb = NULL;
4638 if (i == rx_ring->count)
4640 next_rxd = E1000_RX_DESC_ADV(*rx_ring, i);
4642 next_buffer = &rx_ring->buffer_info[i];
4644 length = le16_to_cpu(rx_desc->wb.upper.length);
4648 /* this is the fast path for the non-packet split case */
4649 if (!adapter->rx_ps_hdr_size) {
4650 pci_unmap_single(pdev, buffer_info->dma,
4651 adapter->rx_buffer_len,
4652 PCI_DMA_FROMDEVICE);
4653 buffer_info->dma = 0;
4654 skb_put(skb, length);
4658 if (buffer_info->dma) {
4659 u16 hlen = igb_get_hlen(adapter, rx_desc);
4660 pci_unmap_single(pdev, buffer_info->dma,
4661 adapter->rx_ps_hdr_size,
4662 PCI_DMA_FROMDEVICE);
4663 buffer_info->dma = 0;
4668 pci_unmap_page(pdev, buffer_info->page_dma,
4669 PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
4670 buffer_info->page_dma = 0;
4672 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags++,
4674 buffer_info->page_offset,
4677 if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
4678 (page_count(buffer_info->page) != 1))
4679 buffer_info->page = NULL;
4681 get_page(buffer_info->page);
4684 skb->data_len += length;
4686 skb->truesize += length;
4689 if (!(staterr & E1000_RXD_STAT_EOP)) {
4690 buffer_info->skb = next_buffer->skb;
4691 buffer_info->dma = next_buffer->dma;
4692 next_buffer->skb = skb;
4693 next_buffer->dma = 0;
4698 * If this bit is set, then the RX registers contain
4699 * the time stamp. No other packet will be time
4700 * stamped until we read these registers, so read the
4701 * registers to make them available again. Because
4702 * only one packet can be time stamped at a time, we
4703 * know that the register values must belong to this
4704 * one here and therefore we don't need to compare
4705 * any of the additional attributes stored for it.
4707 * If nothing went wrong, then it should have a
4708 * skb_shared_tx that we can turn into a
4709 * skb_shared_hwtstamps.
4711 * TODO: can time stamping be triggered (thus locking
4712 * the registers) without the packet reaching this point
4713 * here? In that case RX time stamping would get stuck.
4715 * TODO: in "time stamp all packets" mode this bit is
4716 * not set. Need a global flag for this mode and then
4717 * always read the registers. Cannot be done without
4720 if (unlikely(staterr & E1000_RXD_STAT_TS)) {
4723 struct skb_shared_hwtstamps *shhwtstamps =
4726 WARN(!(rd32(E1000_TSYNCRXCTL) & E1000_TSYNCRXCTL_VALID),
4727 "igb: no RX time stamp available for time stamped packet");
4728 regval = rd32(E1000_RXSTMPL);
4729 regval |= (u64)rd32(E1000_RXSTMPH) << 32;
4730 ns = timecounter_cyc2time(&adapter->clock, regval);
4731 timecompare_update(&adapter->compare, ns);
4732 memset(shhwtstamps, 0, sizeof(*shhwtstamps));
4733 shhwtstamps->hwtstamp = ns_to_ktime(ns);
4734 shhwtstamps->syststamp =
4735 timecompare_transform(&adapter->compare, ns);
4738 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
4739 dev_kfree_skb_irq(skb);
4743 total_bytes += skb->len;
4746 igb_rx_checksum_adv(adapter, staterr, skb);
4748 skb->protocol = eth_type_trans(skb, netdev);
4750 igb_receive_skb(rx_ring, staterr, rx_desc, skb);
4753 rx_desc->wb.upper.status_error = 0;
4755 /* return some buffers to hardware, one at a time is too slow */
4756 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
4757 igb_alloc_rx_buffers_adv(rx_ring, cleaned_count);
4761 /* use prefetched values */
4763 buffer_info = next_buffer;
4764 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
4767 rx_ring->next_to_clean = i;
4768 cleaned_count = igb_desc_unused(rx_ring);
4771 igb_alloc_rx_buffers_adv(rx_ring, cleaned_count);
4773 rx_ring->total_packets += total_packets;
4774 rx_ring->total_bytes += total_bytes;
4775 rx_ring->rx_stats.packets += total_packets;
4776 rx_ring->rx_stats.bytes += total_bytes;
4777 adapter->net_stats.rx_bytes += total_bytes;
4778 adapter->net_stats.rx_packets += total_packets;
4783 * igb_alloc_rx_buffers_adv - Replace used receive buffers; packet split
4784 * @adapter: address of board private structure
4786 static void igb_alloc_rx_buffers_adv(struct igb_ring *rx_ring,
4789 struct igb_adapter *adapter = rx_ring->adapter;
4790 struct net_device *netdev = adapter->netdev;
4791 struct pci_dev *pdev = adapter->pdev;
4792 union e1000_adv_rx_desc *rx_desc;
4793 struct igb_buffer *buffer_info;
4794 struct sk_buff *skb;
4798 i = rx_ring->next_to_use;
4799 buffer_info = &rx_ring->buffer_info[i];
4801 if (adapter->rx_ps_hdr_size)
4802 bufsz = adapter->rx_ps_hdr_size;
4804 bufsz = adapter->rx_buffer_len;
4806 while (cleaned_count--) {
4807 rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
4809 if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) {
4810 if (!buffer_info->page) {
4811 buffer_info->page = alloc_page(GFP_ATOMIC);
4812 if (!buffer_info->page) {
4813 adapter->alloc_rx_buff_failed++;
4816 buffer_info->page_offset = 0;
4818 buffer_info->page_offset ^= PAGE_SIZE / 2;
4820 buffer_info->page_dma =
4821 pci_map_page(pdev, buffer_info->page,
4822 buffer_info->page_offset,
4824 PCI_DMA_FROMDEVICE);
4827 if (!buffer_info->skb) {
4828 skb = netdev_alloc_skb(netdev, bufsz + NET_IP_ALIGN);
4830 adapter->alloc_rx_buff_failed++;
4834 /* Make buffer alignment 2 beyond a 16 byte boundary
4835 * this will result in a 16 byte aligned IP header after
4836 * the 14 byte MAC header is removed
4838 skb_reserve(skb, NET_IP_ALIGN);
4840 buffer_info->skb = skb;
4841 buffer_info->dma = pci_map_single(pdev, skb->data,
4843 PCI_DMA_FROMDEVICE);
4845 /* Refresh the desc even if buffer_addrs didn't change because
4846 * each write-back erases this info. */
4847 if (adapter->rx_ps_hdr_size) {
4848 rx_desc->read.pkt_addr =
4849 cpu_to_le64(buffer_info->page_dma);
4850 rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
4852 rx_desc->read.pkt_addr =
4853 cpu_to_le64(buffer_info->dma);
4854 rx_desc->read.hdr_addr = 0;
4858 if (i == rx_ring->count)
4860 buffer_info = &rx_ring->buffer_info[i];
4864 if (rx_ring->next_to_use != i) {
4865 rx_ring->next_to_use = i;
4867 i = (rx_ring->count - 1);
4871 /* Force memory writes to complete before letting h/w
4872 * know there are new descriptors to fetch. (Only
4873 * applicable for weak-ordered memory model archs,
4874 * such as IA-64). */
4876 writel(i, adapter->hw.hw_addr + rx_ring->tail);
4886 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4888 struct igb_adapter *adapter = netdev_priv(netdev);
4889 struct mii_ioctl_data *data = if_mii(ifr);
4891 if (adapter->hw.phy.media_type != e1000_media_type_copper)
4896 data->phy_id = adapter->hw.phy.addr;
4899 if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
4911 * igb_hwtstamp_ioctl - control hardware time stamping
4916 * Outgoing time stamping can be enabled and disabled. Play nice and
4917 * disable it when requested, although it shouldn't case any overhead
4918 * when no packet needs it. At most one packet in the queue may be
4919 * marked for time stamping, otherwise it would be impossible to tell
4920 * for sure to which packet the hardware time stamp belongs.
4922 * Incoming time stamping has to be configured via the hardware
4923 * filters. Not all combinations are supported, in particular event
4924 * type has to be specified. Matching the kind of event packet is
4925 * not supported, with the exception of "all V2 events regardless of
4929 static int igb_hwtstamp_ioctl(struct net_device *netdev,
4930 struct ifreq *ifr, int cmd)
4932 struct igb_adapter *adapter = netdev_priv(netdev);
4933 struct e1000_hw *hw = &adapter->hw;
4934 struct hwtstamp_config config;
4935 u32 tsync_tx_ctl_bit = E1000_TSYNCTXCTL_ENABLED;
4936 u32 tsync_rx_ctl_bit = E1000_TSYNCRXCTL_ENABLED;
4937 u32 tsync_rx_ctl_type = 0;
4938 u32 tsync_rx_cfg = 0;
4941 short port = 319; /* PTP */
4944 if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
4947 /* reserved for future extensions */
4951 switch (config.tx_type) {
4952 case HWTSTAMP_TX_OFF:
4953 tsync_tx_ctl_bit = 0;
4955 case HWTSTAMP_TX_ON:
4956 tsync_tx_ctl_bit = E1000_TSYNCTXCTL_ENABLED;
4962 switch (config.rx_filter) {
4963 case HWTSTAMP_FILTER_NONE:
4964 tsync_rx_ctl_bit = 0;
4966 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
4967 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
4968 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
4969 case HWTSTAMP_FILTER_ALL:
4971 * register TSYNCRXCFG must be set, therefore it is not
4972 * possible to time stamp both Sync and Delay_Req messages
4973 * => fall back to time stamping all packets
4975 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_ALL;
4976 config.rx_filter = HWTSTAMP_FILTER_ALL;
4978 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
4979 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_L4_V1;
4980 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V1_SYNC_MESSAGE;
4983 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
4984 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_L4_V1;
4985 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V1_DELAY_REQ_MESSAGE;
4988 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
4989 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
4990 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
4991 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V2_SYNC_MESSAGE;
4994 config.rx_filter = HWTSTAMP_FILTER_SOME;
4996 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
4997 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
4998 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
4999 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V2_DELAY_REQ_MESSAGE;
5002 config.rx_filter = HWTSTAMP_FILTER_SOME;
5004 case HWTSTAMP_FILTER_PTP_V2_EVENT:
5005 case HWTSTAMP_FILTER_PTP_V2_SYNC:
5006 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
5007 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_EVENT_V2;
5008 config.rx_filter = HWTSTAMP_FILTER_PTP_V2_EVENT;
5015 /* enable/disable TX */
5016 regval = rd32(E1000_TSYNCTXCTL);
5017 regval = (regval & ~E1000_TSYNCTXCTL_ENABLED) | tsync_tx_ctl_bit;
5018 wr32(E1000_TSYNCTXCTL, regval);
5020 /* enable/disable RX, define which PTP packets are time stamped */
5021 regval = rd32(E1000_TSYNCRXCTL);
5022 regval = (regval & ~E1000_TSYNCRXCTL_ENABLED) | tsync_rx_ctl_bit;
5023 regval = (regval & ~0xE) | tsync_rx_ctl_type;
5024 wr32(E1000_TSYNCRXCTL, regval);
5025 wr32(E1000_TSYNCRXCFG, tsync_rx_cfg);
5028 * Ethertype Filter Queue Filter[0][15:0] = 0x88F7
5029 * (Ethertype to filter on)
5030 * Ethertype Filter Queue Filter[0][26] = 0x1 (Enable filter)
5031 * Ethertype Filter Queue Filter[0][30] = 0x1 (Enable Timestamping)
5033 wr32(E1000_ETQF0, is_l2 ? 0x440088f7 : 0);
5035 /* L4 Queue Filter[0]: only filter by source and destination port */
5036 wr32(E1000_SPQF0, htons(port));
5037 wr32(E1000_IMIREXT(0), is_l4 ?
5038 ((1<<12) | (1<<19) /* bypass size and control flags */) : 0);
5039 wr32(E1000_IMIR(0), is_l4 ?
5041 | (0<<16) /* immediate interrupt disabled */
5042 | 0 /* (1<<17) bit cleared: do not bypass
5043 destination port check */)
5045 wr32(E1000_FTQF0, is_l4 ?
5047 | (1<<15) /* VF not compared */
5048 | (1<<27) /* Enable Timestamping */
5049 | (7<<28) /* only source port filter enabled,
5050 source/target address and protocol
5052 : ((1<<15) | (15<<28) /* all mask bits set = filter not
5057 adapter->hwtstamp_config = config;
5059 /* clear TX/RX time stamp registers, just to be sure */
5060 regval = rd32(E1000_TXSTMPH);
5061 regval = rd32(E1000_RXSTMPH);
5063 return copy_to_user(ifr->ifr_data, &config, sizeof(config)) ?
5073 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5079 return igb_mii_ioctl(netdev, ifr, cmd);
5081 return igb_hwtstamp_ioctl(netdev, ifr, cmd);
5087 s32 igb_read_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
5089 struct igb_adapter *adapter = hw->back;
5092 cap_offset = pci_find_capability(adapter->pdev, PCI_CAP_ID_EXP);
5094 return -E1000_ERR_CONFIG;
5096 pci_read_config_word(adapter->pdev, cap_offset + reg, value);
5101 s32 igb_write_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
5103 struct igb_adapter *adapter = hw->back;
5106 cap_offset = pci_find_capability(adapter->pdev, PCI_CAP_ID_EXP);
5108 return -E1000_ERR_CONFIG;
5110 pci_write_config_word(adapter->pdev, cap_offset + reg, *value);
5115 static void igb_vlan_rx_register(struct net_device *netdev,
5116 struct vlan_group *grp)
5118 struct igb_adapter *adapter = netdev_priv(netdev);
5119 struct e1000_hw *hw = &adapter->hw;
5122 igb_irq_disable(adapter);
5123 adapter->vlgrp = grp;
5126 /* enable VLAN tag insert/strip */
5127 ctrl = rd32(E1000_CTRL);
5128 ctrl |= E1000_CTRL_VME;
5129 wr32(E1000_CTRL, ctrl);
5131 /* enable VLAN receive filtering */
5132 rctl = rd32(E1000_RCTL);
5133 rctl &= ~E1000_RCTL_CFIEN;
5134 wr32(E1000_RCTL, rctl);
5135 igb_update_mng_vlan(adapter);
5137 /* disable VLAN tag insert/strip */
5138 ctrl = rd32(E1000_CTRL);
5139 ctrl &= ~E1000_CTRL_VME;
5140 wr32(E1000_CTRL, ctrl);
5142 if (adapter->mng_vlan_id != (u16)IGB_MNG_VLAN_NONE) {
5143 igb_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
5144 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
5148 igb_rlpml_set(adapter);
5150 if (!test_bit(__IGB_DOWN, &adapter->state))
5151 igb_irq_enable(adapter);
5154 static void igb_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
5156 struct igb_adapter *adapter = netdev_priv(netdev);
5157 struct e1000_hw *hw = &adapter->hw;
5158 int pf_id = adapter->vfs_allocated_count;
5160 if ((hw->mng_cookie.status &
5161 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
5162 (vid == adapter->mng_vlan_id))
5165 /* add vid to vlvf if sr-iov is enabled,
5166 * if that fails add directly to filter table */
5167 if (igb_vlvf_set(adapter, vid, true, pf_id))
5168 igb_vfta_set(hw, vid, true);
5172 static void igb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
5174 struct igb_adapter *adapter = netdev_priv(netdev);
5175 struct e1000_hw *hw = &adapter->hw;
5176 int pf_id = adapter->vfs_allocated_count;
5178 igb_irq_disable(adapter);
5179 vlan_group_set_device(adapter->vlgrp, vid, NULL);
5181 if (!test_bit(__IGB_DOWN, &adapter->state))
5182 igb_irq_enable(adapter);
5184 if ((adapter->hw.mng_cookie.status &
5185 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
5186 (vid == adapter->mng_vlan_id)) {
5187 /* release control to f/w */
5188 igb_release_hw_control(adapter);
5192 /* remove vid from vlvf if sr-iov is enabled,
5193 * if not in vlvf remove from vfta */
5194 if (igb_vlvf_set(adapter, vid, false, pf_id))
5195 igb_vfta_set(hw, vid, false);
5198 static void igb_restore_vlan(struct igb_adapter *adapter)
5200 igb_vlan_rx_register(adapter->netdev, adapter->vlgrp);
5202 if (adapter->vlgrp) {
5204 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
5205 if (!vlan_group_get_device(adapter->vlgrp, vid))
5207 igb_vlan_rx_add_vid(adapter->netdev, vid);
5212 int igb_set_spd_dplx(struct igb_adapter *adapter, u16 spddplx)
5214 struct e1000_mac_info *mac = &adapter->hw.mac;
5219 case SPEED_10 + DUPLEX_HALF:
5220 mac->forced_speed_duplex = ADVERTISE_10_HALF;
5222 case SPEED_10 + DUPLEX_FULL:
5223 mac->forced_speed_duplex = ADVERTISE_10_FULL;
5225 case SPEED_100 + DUPLEX_HALF:
5226 mac->forced_speed_duplex = ADVERTISE_100_HALF;
5228 case SPEED_100 + DUPLEX_FULL:
5229 mac->forced_speed_duplex = ADVERTISE_100_FULL;
5231 case SPEED_1000 + DUPLEX_FULL:
5233 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
5235 case SPEED_1000 + DUPLEX_HALF: /* not supported */
5237 dev_err(&adapter->pdev->dev,
5238 "Unsupported Speed/Duplex configuration\n");
5244 static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake)
5246 struct net_device *netdev = pci_get_drvdata(pdev);
5247 struct igb_adapter *adapter = netdev_priv(netdev);
5248 struct e1000_hw *hw = &adapter->hw;
5249 u32 ctrl, rctl, status;
5250 u32 wufc = adapter->wol;
5255 netif_device_detach(netdev);
5257 if (netif_running(netdev))
5260 igb_reset_interrupt_capability(adapter);
5262 igb_free_queues(adapter);
5265 retval = pci_save_state(pdev);
5270 status = rd32(E1000_STATUS);
5271 if (status & E1000_STATUS_LU)
5272 wufc &= ~E1000_WUFC_LNKC;
5275 igb_setup_rctl(adapter);
5276 igb_set_rx_mode(netdev);
5278 /* turn on all-multi mode if wake on multicast is enabled */
5279 if (wufc & E1000_WUFC_MC) {
5280 rctl = rd32(E1000_RCTL);
5281 rctl |= E1000_RCTL_MPE;
5282 wr32(E1000_RCTL, rctl);
5285 ctrl = rd32(E1000_CTRL);
5286 /* advertise wake from D3Cold */
5287 #define E1000_CTRL_ADVD3WUC 0x00100000
5288 /* phy power management enable */
5289 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5290 ctrl |= E1000_CTRL_ADVD3WUC;
5291 wr32(E1000_CTRL, ctrl);
5293 /* Allow time for pending master requests to run */
5294 igb_disable_pcie_master(&adapter->hw);
5296 wr32(E1000_WUC, E1000_WUC_PME_EN);
5297 wr32(E1000_WUFC, wufc);
5300 wr32(E1000_WUFC, 0);
5303 *enable_wake = wufc || adapter->en_mng_pt;
5305 igb_shutdown_serdes_link_82575(hw);
5307 /* Release control of h/w to f/w. If f/w is AMT enabled, this
5308 * would have already happened in close and is redundant. */
5309 igb_release_hw_control(adapter);
5311 pci_disable_device(pdev);
5317 static int igb_suspend(struct pci_dev *pdev, pm_message_t state)
5322 retval = __igb_shutdown(pdev, &wake);
5327 pci_prepare_to_sleep(pdev);
5329 pci_wake_from_d3(pdev, false);
5330 pci_set_power_state(pdev, PCI_D3hot);
5336 static int igb_resume(struct pci_dev *pdev)
5338 struct net_device *netdev = pci_get_drvdata(pdev);
5339 struct igb_adapter *adapter = netdev_priv(netdev);
5340 struct e1000_hw *hw = &adapter->hw;
5343 pci_set_power_state(pdev, PCI_D0);
5344 pci_restore_state(pdev);
5346 err = pci_enable_device_mem(pdev);
5349 "igb: Cannot enable PCI device from suspend\n");
5352 pci_set_master(pdev);
5354 pci_enable_wake(pdev, PCI_D3hot, 0);
5355 pci_enable_wake(pdev, PCI_D3cold, 0);
5357 igb_set_interrupt_capability(adapter);
5359 if (igb_alloc_queues(adapter)) {
5360 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
5364 /* e1000_power_up_phy(adapter); */
5368 /* let the f/w know that the h/w is now under the control of the
5370 igb_get_hw_control(adapter);
5372 wr32(E1000_WUS, ~0);
5374 if (netif_running(netdev)) {
5375 err = igb_open(netdev);
5380 netif_device_attach(netdev);
5386 static void igb_shutdown(struct pci_dev *pdev)
5390 __igb_shutdown(pdev, &wake);
5392 if (system_state == SYSTEM_POWER_OFF) {
5393 pci_wake_from_d3(pdev, wake);
5394 pci_set_power_state(pdev, PCI_D3hot);
5398 #ifdef CONFIG_NET_POLL_CONTROLLER
5400 * Polling 'interrupt' - used by things like netconsole to send skbs
5401 * without having to re-enable interrupts. It's not called while
5402 * the interrupt routine is executing.
5404 static void igb_netpoll(struct net_device *netdev)
5406 struct igb_adapter *adapter = netdev_priv(netdev);
5407 struct e1000_hw *hw = &adapter->hw;
5410 if (!adapter->msix_entries) {
5411 igb_irq_disable(adapter);
5412 napi_schedule(&adapter->rx_ring[0].napi);
5416 for (i = 0; i < adapter->num_tx_queues; i++) {
5417 struct igb_ring *tx_ring = &adapter->tx_ring[i];
5418 wr32(E1000_EIMC, tx_ring->eims_value);
5419 igb_clean_tx_irq(tx_ring);
5420 wr32(E1000_EIMS, tx_ring->eims_value);
5423 for (i = 0; i < adapter->num_rx_queues; i++) {
5424 struct igb_ring *rx_ring = &adapter->rx_ring[i];
5425 wr32(E1000_EIMC, rx_ring->eims_value);
5426 napi_schedule(&rx_ring->napi);
5429 #endif /* CONFIG_NET_POLL_CONTROLLER */
5432 * igb_io_error_detected - called when PCI error is detected
5433 * @pdev: Pointer to PCI device
5434 * @state: The current pci connection state
5436 * This function is called after a PCI bus error affecting
5437 * this device has been detected.
5439 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
5440 pci_channel_state_t state)
5442 struct net_device *netdev = pci_get_drvdata(pdev);
5443 struct igb_adapter *adapter = netdev_priv(netdev);
5445 netif_device_detach(netdev);
5447 if (state == pci_channel_io_perm_failure)
5448 return PCI_ERS_RESULT_DISCONNECT;
5450 if (netif_running(netdev))
5452 pci_disable_device(pdev);
5454 /* Request a slot slot reset. */
5455 return PCI_ERS_RESULT_NEED_RESET;
5459 * igb_io_slot_reset - called after the pci bus has been reset.
5460 * @pdev: Pointer to PCI device
5462 * Restart the card from scratch, as if from a cold-boot. Implementation
5463 * resembles the first-half of the igb_resume routine.
5465 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
5467 struct net_device *netdev = pci_get_drvdata(pdev);
5468 struct igb_adapter *adapter = netdev_priv(netdev);
5469 struct e1000_hw *hw = &adapter->hw;
5470 pci_ers_result_t result;
5473 if (pci_enable_device_mem(pdev)) {
5475 "Cannot re-enable PCI device after reset.\n");
5476 result = PCI_ERS_RESULT_DISCONNECT;
5478 pci_set_master(pdev);
5479 pci_restore_state(pdev);
5481 pci_enable_wake(pdev, PCI_D3hot, 0);
5482 pci_enable_wake(pdev, PCI_D3cold, 0);
5485 wr32(E1000_WUS, ~0);
5486 result = PCI_ERS_RESULT_RECOVERED;
5489 err = pci_cleanup_aer_uncorrect_error_status(pdev);
5491 dev_err(&pdev->dev, "pci_cleanup_aer_uncorrect_error_status "
5492 "failed 0x%0x\n", err);
5493 /* non-fatal, continue */
5500 * igb_io_resume - called when traffic can start flowing again.
5501 * @pdev: Pointer to PCI device
5503 * This callback is called when the error recovery driver tells us that
5504 * its OK to resume normal operation. Implementation resembles the
5505 * second-half of the igb_resume routine.
5507 static void igb_io_resume(struct pci_dev *pdev)
5509 struct net_device *netdev = pci_get_drvdata(pdev);
5510 struct igb_adapter *adapter = netdev_priv(netdev);
5512 if (netif_running(netdev)) {
5513 if (igb_up(adapter)) {
5514 dev_err(&pdev->dev, "igb_up failed after reset\n");
5519 netif_device_attach(netdev);
5521 /* let the f/w know that the h/w is now under the control of the
5523 igb_get_hw_control(adapter);
5526 static void igb_rar_set_qsel(struct igb_adapter *adapter, u8 *addr, u32 index,
5529 u32 rar_low, rar_high;
5530 struct e1000_hw *hw = &adapter->hw;
5532 /* HW expects these in little endian so we reverse the byte order
5533 * from network order (big endian) to little endian
5535 rar_low = ((u32) addr[0] | ((u32) addr[1] << 8) |
5536 ((u32) addr[2] << 16) | ((u32) addr[3] << 24));
5537 rar_high = ((u32) addr[4] | ((u32) addr[5] << 8));
5539 /* Indicate to hardware the Address is Valid. */
5540 rar_high |= E1000_RAH_AV;
5542 if (hw->mac.type == e1000_82575)
5543 rar_high |= E1000_RAH_POOL_1 * qsel;
5545 rar_high |= E1000_RAH_POOL_1 << qsel;
5547 wr32(E1000_RAL(index), rar_low);
5549 wr32(E1000_RAH(index), rar_high);
5553 static int igb_set_vf_mac(struct igb_adapter *adapter,
5554 int vf, unsigned char *mac_addr)
5556 struct e1000_hw *hw = &adapter->hw;
5557 /* VF MAC addresses start at end of receive addresses and moves
5558 * torwards the first, as a result a collision should not be possible */
5559 int rar_entry = hw->mac.rar_entry_count - (vf + 1);
5561 memcpy(adapter->vf_data[vf].vf_mac_addresses, mac_addr, ETH_ALEN);
5563 igb_rar_set_qsel(adapter, mac_addr, rar_entry, vf);
5568 static void igb_vmm_control(struct igb_adapter *adapter)
5570 struct e1000_hw *hw = &adapter->hw;
5573 if (!adapter->vfs_allocated_count)
5576 /* VF's need PF reset indication before they
5577 * can send/receive mail */
5578 reg_data = rd32(E1000_CTRL_EXT);
5579 reg_data |= E1000_CTRL_EXT_PFRSTD;
5580 wr32(E1000_CTRL_EXT, reg_data);
5582 igb_vmdq_set_loopback_pf(hw, true);
5583 igb_vmdq_set_replication_pf(hw, true);