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_QUAD_COPPER), board_82575 },
69 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
70 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
71 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
72 /* required last entry */
76 MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
78 void igb_reset(struct igb_adapter *);
79 static int igb_setup_all_tx_resources(struct igb_adapter *);
80 static int igb_setup_all_rx_resources(struct igb_adapter *);
81 static void igb_free_all_tx_resources(struct igb_adapter *);
82 static void igb_free_all_rx_resources(struct igb_adapter *);
83 void igb_update_stats(struct igb_adapter *);
84 static int igb_probe(struct pci_dev *, const struct pci_device_id *);
85 static void __devexit igb_remove(struct pci_dev *pdev);
86 static int igb_sw_init(struct igb_adapter *);
87 static int igb_open(struct net_device *);
88 static int igb_close(struct net_device *);
89 static void igb_configure_tx(struct igb_adapter *);
90 static void igb_configure_rx(struct igb_adapter *);
91 static void igb_setup_rctl(struct igb_adapter *);
92 static void igb_clean_all_tx_rings(struct igb_adapter *);
93 static void igb_clean_all_rx_rings(struct igb_adapter *);
94 static void igb_clean_tx_ring(struct igb_ring *);
95 static void igb_clean_rx_ring(struct igb_ring *);
96 static void igb_set_multi(struct net_device *);
97 static void igb_update_phy_info(unsigned long);
98 static void igb_watchdog(unsigned long);
99 static void igb_watchdog_task(struct work_struct *);
100 static int igb_xmit_frame_ring_adv(struct sk_buff *, struct net_device *,
102 static int igb_xmit_frame_adv(struct sk_buff *skb, struct net_device *);
103 static struct net_device_stats *igb_get_stats(struct net_device *);
104 static int igb_change_mtu(struct net_device *, int);
105 static int igb_set_mac(struct net_device *, void *);
106 static irqreturn_t igb_intr(int irq, void *);
107 static irqreturn_t igb_intr_msi(int irq, void *);
108 static irqreturn_t igb_msix_other(int irq, void *);
109 static irqreturn_t igb_msix_rx(int irq, void *);
110 static irqreturn_t igb_msix_tx(int irq, void *);
111 #ifdef CONFIG_IGB_DCA
112 static void igb_update_rx_dca(struct igb_ring *);
113 static void igb_update_tx_dca(struct igb_ring *);
114 static void igb_setup_dca(struct igb_adapter *);
115 #endif /* CONFIG_IGB_DCA */
116 static bool igb_clean_tx_irq(struct igb_ring *);
117 static int igb_poll(struct napi_struct *, int);
118 static bool igb_clean_rx_irq_adv(struct igb_ring *, int *, int);
119 static void igb_alloc_rx_buffers_adv(struct igb_ring *, int);
120 static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
121 static void igb_tx_timeout(struct net_device *);
122 static void igb_reset_task(struct work_struct *);
123 static void igb_vlan_rx_register(struct net_device *, struct vlan_group *);
124 static void igb_vlan_rx_add_vid(struct net_device *, u16);
125 static void igb_vlan_rx_kill_vid(struct net_device *, u16);
126 static void igb_restore_vlan(struct igb_adapter *);
127 static void igb_ping_all_vfs(struct igb_adapter *);
128 static void igb_msg_task(struct igb_adapter *);
129 static int igb_rcv_msg_from_vf(struct igb_adapter *, u32);
130 static inline void igb_set_rah_pool(struct e1000_hw *, int , int);
131 static void igb_set_mc_list_pools(struct igb_adapter *, int, u16);
132 static void igb_vmm_control(struct igb_adapter *);
133 static inline void igb_set_vmolr(struct e1000_hw *, int);
134 static inline int igb_set_vf_rlpml(struct igb_adapter *, int, int);
135 static int igb_set_vf_mac(struct igb_adapter *adapter, int, unsigned char *);
136 static void igb_restore_vf_multicasts(struct igb_adapter *adapter);
139 static int igb_suspend(struct pci_dev *, pm_message_t);
140 static int igb_resume(struct pci_dev *);
142 static void igb_shutdown(struct pci_dev *);
143 #ifdef CONFIG_IGB_DCA
144 static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
145 static struct notifier_block dca_notifier = {
146 .notifier_call = igb_notify_dca,
151 #ifdef CONFIG_NET_POLL_CONTROLLER
152 /* for netdump / net console */
153 static void igb_netpoll(struct net_device *);
155 #ifdef CONFIG_PCI_IOV
156 static unsigned int max_vfs = 0;
157 module_param(max_vfs, uint, 0);
158 MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate "
159 "per physical function");
160 #endif /* CONFIG_PCI_IOV */
162 static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
163 pci_channel_state_t);
164 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
165 static void igb_io_resume(struct pci_dev *);
167 static struct pci_error_handlers igb_err_handler = {
168 .error_detected = igb_io_error_detected,
169 .slot_reset = igb_io_slot_reset,
170 .resume = igb_io_resume,
174 static struct pci_driver igb_driver = {
175 .name = igb_driver_name,
176 .id_table = igb_pci_tbl,
178 .remove = __devexit_p(igb_remove),
180 /* Power Managment Hooks */
181 .suspend = igb_suspend,
182 .resume = igb_resume,
184 .shutdown = igb_shutdown,
185 .err_handler = &igb_err_handler
188 static int global_quad_port_a; /* global quad port a indication */
190 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
191 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
192 MODULE_LICENSE("GPL");
193 MODULE_VERSION(DRV_VERSION);
196 * Scale the NIC clock cycle by a large factor so that
197 * relatively small clock corrections can be added or
198 * substracted at each clock tick. The drawbacks of a
199 * large factor are a) that the clock register overflows
200 * more quickly (not such a big deal) and b) that the
201 * increment per tick has to fit into 24 bits.
204 * TIMINCA = IGB_TSYNC_CYCLE_TIME_IN_NANOSECONDS *
206 * TIMINCA += TIMINCA * adjustment [ppm] / 1e9
208 * The base scale factor is intentionally a power of two
209 * so that the division in %struct timecounter can be done with
212 #define IGB_TSYNC_SHIFT (19)
213 #define IGB_TSYNC_SCALE (1<<IGB_TSYNC_SHIFT)
216 * The duration of one clock cycle of the NIC.
218 * @todo This hard-coded value is part of the specification and might change
219 * in future hardware revisions. Add revision check.
221 #define IGB_TSYNC_CYCLE_TIME_IN_NANOSECONDS 16
223 #if (IGB_TSYNC_SCALE * IGB_TSYNC_CYCLE_TIME_IN_NANOSECONDS) >= (1<<24)
224 # error IGB_TSYNC_SCALE and/or IGB_TSYNC_CYCLE_TIME_IN_NANOSECONDS are too large to fit into TIMINCA
228 * igb_read_clock - read raw cycle counter (to be used by time counter)
230 static cycle_t igb_read_clock(const struct cyclecounter *tc)
232 struct igb_adapter *adapter =
233 container_of(tc, struct igb_adapter, cycles);
234 struct e1000_hw *hw = &adapter->hw;
237 stamp = rd32(E1000_SYSTIML);
238 stamp |= (u64)rd32(E1000_SYSTIMH) << 32ULL;
245 * igb_get_hw_dev_name - return device name string
246 * used by hardware layer to print debugging information
248 char *igb_get_hw_dev_name(struct e1000_hw *hw)
250 struct igb_adapter *adapter = hw->back;
251 return adapter->netdev->name;
255 * igb_get_time_str - format current NIC and system time as string
257 static char *igb_get_time_str(struct igb_adapter *adapter,
260 cycle_t hw = adapter->cycles.read(&adapter->cycles);
261 struct timespec nic = ns_to_timespec(timecounter_read(&adapter->clock));
263 struct timespec delta;
264 getnstimeofday(&sys);
266 delta = timespec_sub(nic, sys);
269 "HW %llu, NIC %ld.%09lus, SYS %ld.%09lus, NIC-SYS %lds + %09luns",
271 (long)nic.tv_sec, nic.tv_nsec,
272 (long)sys.tv_sec, sys.tv_nsec,
273 (long)delta.tv_sec, delta.tv_nsec);
280 * igb_desc_unused - calculate if we have unused descriptors
282 static int igb_desc_unused(struct igb_ring *ring)
284 if (ring->next_to_clean > ring->next_to_use)
285 return ring->next_to_clean - ring->next_to_use - 1;
287 return ring->count + ring->next_to_clean - ring->next_to_use - 1;
291 * igb_init_module - Driver Registration Routine
293 * igb_init_module is the first routine called when the driver is
294 * loaded. All it does is register with the PCI subsystem.
296 static int __init igb_init_module(void)
299 printk(KERN_INFO "%s - version %s\n",
300 igb_driver_string, igb_driver_version);
302 printk(KERN_INFO "%s\n", igb_copyright);
304 global_quad_port_a = 0;
306 #ifdef CONFIG_IGB_DCA
307 dca_register_notify(&dca_notifier);
310 ret = pci_register_driver(&igb_driver);
314 module_init(igb_init_module);
317 * igb_exit_module - Driver Exit Cleanup Routine
319 * igb_exit_module is called just before the driver is removed
322 static void __exit igb_exit_module(void)
324 #ifdef CONFIG_IGB_DCA
325 dca_unregister_notify(&dca_notifier);
327 pci_unregister_driver(&igb_driver);
330 module_exit(igb_exit_module);
332 #define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
334 * igb_cache_ring_register - Descriptor ring to register mapping
335 * @adapter: board private structure to initialize
337 * Once we know the feature-set enabled for the device, we'll cache
338 * the register offset the descriptor ring is assigned to.
340 static void igb_cache_ring_register(struct igb_adapter *adapter)
343 unsigned int rbase_offset = adapter->vfs_allocated_count;
345 switch (adapter->hw.mac.type) {
347 /* The queues are allocated for virtualization such that VF 0
348 * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
349 * In order to avoid collision we start at the first free queue
350 * and continue consuming queues in the same sequence
352 for (i = 0; i < adapter->num_rx_queues; i++)
353 adapter->rx_ring[i].reg_idx = rbase_offset +
355 for (i = 0; i < adapter->num_tx_queues; i++)
356 adapter->tx_ring[i].reg_idx = rbase_offset +
361 for (i = 0; i < adapter->num_rx_queues; i++)
362 adapter->rx_ring[i].reg_idx = i;
363 for (i = 0; i < adapter->num_tx_queues; i++)
364 adapter->tx_ring[i].reg_idx = i;
370 * igb_alloc_queues - Allocate memory for all rings
371 * @adapter: board private structure to initialize
373 * We allocate one ring per queue at run-time since we don't know the
374 * number of queues at compile-time.
376 static int igb_alloc_queues(struct igb_adapter *adapter)
380 adapter->tx_ring = kcalloc(adapter->num_tx_queues,
381 sizeof(struct igb_ring), GFP_KERNEL);
382 if (!adapter->tx_ring)
385 adapter->rx_ring = kcalloc(adapter->num_rx_queues,
386 sizeof(struct igb_ring), GFP_KERNEL);
387 if (!adapter->rx_ring) {
388 kfree(adapter->tx_ring);
392 adapter->rx_ring->buddy = adapter->tx_ring;
394 for (i = 0; i < adapter->num_tx_queues; i++) {
395 struct igb_ring *ring = &(adapter->tx_ring[i]);
396 ring->count = adapter->tx_ring_count;
397 ring->adapter = adapter;
398 ring->queue_index = i;
400 for (i = 0; i < adapter->num_rx_queues; i++) {
401 struct igb_ring *ring = &(adapter->rx_ring[i]);
402 ring->count = adapter->rx_ring_count;
403 ring->adapter = adapter;
404 ring->queue_index = i;
405 ring->itr_register = E1000_ITR;
407 /* set a default napi handler for each rx_ring */
408 netif_napi_add(adapter->netdev, &ring->napi, igb_poll, 64);
411 igb_cache_ring_register(adapter);
415 static void igb_free_queues(struct igb_adapter *adapter)
419 for (i = 0; i < adapter->num_rx_queues; i++)
420 netif_napi_del(&adapter->rx_ring[i].napi);
422 adapter->num_rx_queues = 0;
423 adapter->num_tx_queues = 0;
425 kfree(adapter->tx_ring);
426 kfree(adapter->rx_ring);
429 #define IGB_N0_QUEUE -1
430 static void igb_assign_vector(struct igb_adapter *adapter, int rx_queue,
431 int tx_queue, int msix_vector)
434 struct e1000_hw *hw = &adapter->hw;
437 switch (hw->mac.type) {
439 /* The 82575 assigns vectors using a bitmask, which matches the
440 bitmask for the EICR/EIMS/EIMC registers. To assign one
441 or more queues to a vector, we write the appropriate bits
442 into the MSIXBM register for that vector. */
443 if (rx_queue > IGB_N0_QUEUE) {
444 msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
445 adapter->rx_ring[rx_queue].eims_value = msixbm;
447 if (tx_queue > IGB_N0_QUEUE) {
448 msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
449 adapter->tx_ring[tx_queue].eims_value =
450 E1000_EICR_TX_QUEUE0 << tx_queue;
452 array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
455 /* 82576 uses a table-based method for assigning vectors.
456 Each queue has a single entry in the table to which we write
457 a vector number along with a "valid" bit. Sadly, the layout
458 of the table is somewhat counterintuitive. */
459 if (rx_queue > IGB_N0_QUEUE) {
460 index = (rx_queue >> 1) + adapter->vfs_allocated_count;
461 ivar = array_rd32(E1000_IVAR0, index);
462 if (rx_queue & 0x1) {
463 /* vector goes into third byte of register */
464 ivar = ivar & 0xFF00FFFF;
465 ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
467 /* vector goes into low byte of register */
468 ivar = ivar & 0xFFFFFF00;
469 ivar |= msix_vector | E1000_IVAR_VALID;
471 adapter->rx_ring[rx_queue].eims_value= 1 << msix_vector;
472 array_wr32(E1000_IVAR0, index, ivar);
474 if (tx_queue > IGB_N0_QUEUE) {
475 index = (tx_queue >> 1) + adapter->vfs_allocated_count;
476 ivar = array_rd32(E1000_IVAR0, index);
477 if (tx_queue & 0x1) {
478 /* vector goes into high byte of register */
479 ivar = ivar & 0x00FFFFFF;
480 ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
482 /* vector goes into second byte of register */
483 ivar = ivar & 0xFFFF00FF;
484 ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
486 adapter->tx_ring[tx_queue].eims_value= 1 << msix_vector;
487 array_wr32(E1000_IVAR0, index, ivar);
497 * igb_configure_msix - Configure MSI-X hardware
499 * igb_configure_msix sets up the hardware to properly
500 * generate MSI-X interrupts.
502 static void igb_configure_msix(struct igb_adapter *adapter)
506 struct e1000_hw *hw = &adapter->hw;
508 adapter->eims_enable_mask = 0;
509 if (hw->mac.type == e1000_82576)
510 /* Turn on MSI-X capability first, or our settings
511 * won't stick. And it will take days to debug. */
512 wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
513 E1000_GPIE_PBA | E1000_GPIE_EIAME |
516 for (i = 0; i < adapter->num_tx_queues; i++) {
517 struct igb_ring *tx_ring = &adapter->tx_ring[i];
518 igb_assign_vector(adapter, IGB_N0_QUEUE, i, vector++);
519 adapter->eims_enable_mask |= tx_ring->eims_value;
520 if (tx_ring->itr_val)
521 writel(tx_ring->itr_val,
522 hw->hw_addr + tx_ring->itr_register);
524 writel(1, hw->hw_addr + tx_ring->itr_register);
527 for (i = 0; i < adapter->num_rx_queues; i++) {
528 struct igb_ring *rx_ring = &adapter->rx_ring[i];
529 rx_ring->buddy = NULL;
530 igb_assign_vector(adapter, i, IGB_N0_QUEUE, vector++);
531 adapter->eims_enable_mask |= rx_ring->eims_value;
532 if (rx_ring->itr_val)
533 writel(rx_ring->itr_val,
534 hw->hw_addr + rx_ring->itr_register);
536 writel(1, hw->hw_addr + rx_ring->itr_register);
540 /* set vector for other causes, i.e. link changes */
541 switch (hw->mac.type) {
543 array_wr32(E1000_MSIXBM(0), vector++,
546 tmp = rd32(E1000_CTRL_EXT);
547 /* enable MSI-X PBA support*/
548 tmp |= E1000_CTRL_EXT_PBA_CLR;
550 /* Auto-Mask interrupts upon ICR read. */
551 tmp |= E1000_CTRL_EXT_EIAME;
552 tmp |= E1000_CTRL_EXT_IRCA;
554 wr32(E1000_CTRL_EXT, tmp);
555 adapter->eims_enable_mask |= E1000_EIMS_OTHER;
556 adapter->eims_other = E1000_EIMS_OTHER;
561 tmp = (vector++ | E1000_IVAR_VALID) << 8;
562 wr32(E1000_IVAR_MISC, tmp);
564 adapter->eims_enable_mask = (1 << (vector)) - 1;
565 adapter->eims_other = 1 << (vector - 1);
568 /* do nothing, since nothing else supports MSI-X */
570 } /* switch (hw->mac.type) */
575 * igb_request_msix - Initialize MSI-X interrupts
577 * igb_request_msix allocates MSI-X vectors and requests interrupts from the
580 static int igb_request_msix(struct igb_adapter *adapter)
582 struct net_device *netdev = adapter->netdev;
583 int i, err = 0, vector = 0;
587 for (i = 0; i < adapter->num_tx_queues; i++) {
588 struct igb_ring *ring = &(adapter->tx_ring[i]);
589 sprintf(ring->name, "%s-tx-%d", netdev->name, i);
590 err = request_irq(adapter->msix_entries[vector].vector,
591 &igb_msix_tx, 0, ring->name,
592 &(adapter->tx_ring[i]));
595 ring->itr_register = E1000_EITR(0) + (vector << 2);
596 ring->itr_val = 976; /* ~4000 ints/sec */
599 for (i = 0; i < adapter->num_rx_queues; i++) {
600 struct igb_ring *ring = &(adapter->rx_ring[i]);
601 if (strlen(netdev->name) < (IFNAMSIZ - 5))
602 sprintf(ring->name, "%s-rx-%d", netdev->name, i);
604 memcpy(ring->name, netdev->name, IFNAMSIZ);
605 err = request_irq(adapter->msix_entries[vector].vector,
606 &igb_msix_rx, 0, ring->name,
607 &(adapter->rx_ring[i]));
610 ring->itr_register = E1000_EITR(0) + (vector << 2);
611 ring->itr_val = adapter->itr;
615 err = request_irq(adapter->msix_entries[vector].vector,
616 &igb_msix_other, 0, netdev->name, netdev);
620 igb_configure_msix(adapter);
626 static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
628 if (adapter->msix_entries) {
629 pci_disable_msix(adapter->pdev);
630 kfree(adapter->msix_entries);
631 adapter->msix_entries = NULL;
632 } else if (adapter->flags & IGB_FLAG_HAS_MSI)
633 pci_disable_msi(adapter->pdev);
639 * igb_set_interrupt_capability - set MSI or MSI-X if supported
641 * Attempt to configure interrupts using the best available
642 * capabilities of the hardware and kernel.
644 static void igb_set_interrupt_capability(struct igb_adapter *adapter)
649 /* Number of supported queues. */
650 /* Having more queues than CPUs doesn't make sense. */
651 adapter->num_rx_queues = min_t(u32, IGB_MAX_RX_QUEUES, num_online_cpus());
652 adapter->num_tx_queues = min_t(u32, IGB_MAX_TX_QUEUES, num_online_cpus());
654 numvecs = adapter->num_tx_queues + adapter->num_rx_queues + 1;
655 adapter->msix_entries = kcalloc(numvecs, sizeof(struct msix_entry),
657 if (!adapter->msix_entries)
660 for (i = 0; i < numvecs; i++)
661 adapter->msix_entries[i].entry = i;
663 err = pci_enable_msix(adapter->pdev,
664 adapter->msix_entries,
669 igb_reset_interrupt_capability(adapter);
671 /* If we can't do MSI-X, try MSI */
673 #ifdef CONFIG_PCI_IOV
674 /* disable SR-IOV for non MSI-X configurations */
675 if (adapter->vf_data) {
676 struct e1000_hw *hw = &adapter->hw;
677 /* disable iov and allow time for transactions to clear */
678 pci_disable_sriov(adapter->pdev);
681 kfree(adapter->vf_data);
682 adapter->vf_data = NULL;
683 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
685 dev_info(&adapter->pdev->dev, "IOV Disabled\n");
688 adapter->num_rx_queues = 1;
689 adapter->num_tx_queues = 1;
690 if (!pci_enable_msi(adapter->pdev))
691 adapter->flags |= IGB_FLAG_HAS_MSI;
693 /* Notify the stack of the (possibly) reduced Tx Queue count. */
694 adapter->netdev->real_num_tx_queues = adapter->num_tx_queues;
699 * igb_request_irq - initialize interrupts
701 * Attempts to configure interrupts using the best available
702 * capabilities of the hardware and kernel.
704 static int igb_request_irq(struct igb_adapter *adapter)
706 struct net_device *netdev = adapter->netdev;
707 struct e1000_hw *hw = &adapter->hw;
710 if (adapter->msix_entries) {
711 err = igb_request_msix(adapter);
714 /* fall back to MSI */
715 igb_reset_interrupt_capability(adapter);
716 if (!pci_enable_msi(adapter->pdev))
717 adapter->flags |= IGB_FLAG_HAS_MSI;
718 igb_free_all_tx_resources(adapter);
719 igb_free_all_rx_resources(adapter);
720 adapter->num_rx_queues = 1;
721 igb_alloc_queues(adapter);
723 switch (hw->mac.type) {
725 wr32(E1000_MSIXBM(0),
726 (E1000_EICR_RX_QUEUE0 | E1000_EIMS_OTHER));
729 wr32(E1000_IVAR0, E1000_IVAR_VALID);
736 if (adapter->flags & IGB_FLAG_HAS_MSI) {
737 err = request_irq(adapter->pdev->irq, &igb_intr_msi, 0,
738 netdev->name, netdev);
741 /* fall back to legacy interrupts */
742 igb_reset_interrupt_capability(adapter);
743 adapter->flags &= ~IGB_FLAG_HAS_MSI;
746 err = request_irq(adapter->pdev->irq, &igb_intr, IRQF_SHARED,
747 netdev->name, netdev);
750 dev_err(&adapter->pdev->dev, "Error %d getting interrupt\n",
757 static void igb_free_irq(struct igb_adapter *adapter)
759 struct net_device *netdev = adapter->netdev;
761 if (adapter->msix_entries) {
764 for (i = 0; i < adapter->num_tx_queues; i++)
765 free_irq(adapter->msix_entries[vector++].vector,
766 &(adapter->tx_ring[i]));
767 for (i = 0; i < adapter->num_rx_queues; i++)
768 free_irq(adapter->msix_entries[vector++].vector,
769 &(adapter->rx_ring[i]));
771 free_irq(adapter->msix_entries[vector++].vector, netdev);
775 free_irq(adapter->pdev->irq, netdev);
779 * igb_irq_disable - Mask off interrupt generation on the NIC
780 * @adapter: board private structure
782 static void igb_irq_disable(struct igb_adapter *adapter)
784 struct e1000_hw *hw = &adapter->hw;
786 if (adapter->msix_entries) {
788 wr32(E1000_EIMC, ~0);
795 synchronize_irq(adapter->pdev->irq);
799 * igb_irq_enable - Enable default interrupt generation settings
800 * @adapter: board private structure
802 static void igb_irq_enable(struct igb_adapter *adapter)
804 struct e1000_hw *hw = &adapter->hw;
806 if (adapter->msix_entries) {
807 wr32(E1000_EIAC, adapter->eims_enable_mask);
808 wr32(E1000_EIAM, adapter->eims_enable_mask);
809 wr32(E1000_EIMS, adapter->eims_enable_mask);
810 if (adapter->vfs_allocated_count)
811 wr32(E1000_MBVFIMR, 0xFF);
812 wr32(E1000_IMS, (E1000_IMS_LSC | E1000_IMS_VMMB |
813 E1000_IMS_DOUTSYNC));
815 wr32(E1000_IMS, IMS_ENABLE_MASK);
816 wr32(E1000_IAM, IMS_ENABLE_MASK);
820 static void igb_update_mng_vlan(struct igb_adapter *adapter)
822 struct net_device *netdev = adapter->netdev;
823 u16 vid = adapter->hw.mng_cookie.vlan_id;
824 u16 old_vid = adapter->mng_vlan_id;
825 if (adapter->vlgrp) {
826 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
827 if (adapter->hw.mng_cookie.status &
828 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
829 igb_vlan_rx_add_vid(netdev, vid);
830 adapter->mng_vlan_id = vid;
832 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
834 if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
836 !vlan_group_get_device(adapter->vlgrp, old_vid))
837 igb_vlan_rx_kill_vid(netdev, old_vid);
839 adapter->mng_vlan_id = vid;
844 * igb_release_hw_control - release control of the h/w to f/w
845 * @adapter: address of board private structure
847 * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
848 * For ASF and Pass Through versions of f/w this means that the
849 * driver is no longer loaded.
852 static void igb_release_hw_control(struct igb_adapter *adapter)
854 struct e1000_hw *hw = &adapter->hw;
857 /* Let firmware take over control of h/w */
858 ctrl_ext = rd32(E1000_CTRL_EXT);
860 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
865 * igb_get_hw_control - get control of the h/w from f/w
866 * @adapter: address of board private structure
868 * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
869 * For ASF and Pass Through versions of f/w this means that
870 * the driver is loaded.
873 static void igb_get_hw_control(struct igb_adapter *adapter)
875 struct e1000_hw *hw = &adapter->hw;
878 /* Let firmware know the driver has taken over */
879 ctrl_ext = rd32(E1000_CTRL_EXT);
881 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
885 * igb_configure - configure the hardware for RX and TX
886 * @adapter: private board structure
888 static void igb_configure(struct igb_adapter *adapter)
890 struct net_device *netdev = adapter->netdev;
893 igb_get_hw_control(adapter);
894 igb_set_multi(netdev);
896 igb_restore_vlan(adapter);
898 igb_configure_tx(adapter);
899 igb_setup_rctl(adapter);
900 igb_configure_rx(adapter);
902 igb_rx_fifo_flush_82575(&adapter->hw);
904 /* call igb_desc_unused which always leaves
905 * at least 1 descriptor unused to make sure
906 * next_to_use != next_to_clean */
907 for (i = 0; i < adapter->num_rx_queues; i++) {
908 struct igb_ring *ring = &adapter->rx_ring[i];
909 igb_alloc_rx_buffers_adv(ring, igb_desc_unused(ring));
913 adapter->tx_queue_len = netdev->tx_queue_len;
918 * igb_up - Open the interface and prepare it to handle traffic
919 * @adapter: board private structure
922 int igb_up(struct igb_adapter *adapter)
924 struct e1000_hw *hw = &adapter->hw;
927 /* hardware has been reset, we need to reload some things */
928 igb_configure(adapter);
930 clear_bit(__IGB_DOWN, &adapter->state);
932 for (i = 0; i < adapter->num_rx_queues; i++)
933 napi_enable(&adapter->rx_ring[i].napi);
934 if (adapter->msix_entries)
935 igb_configure_msix(adapter);
937 igb_vmm_control(adapter);
938 igb_set_rah_pool(hw, adapter->vfs_allocated_count, 0);
939 igb_set_vmolr(hw, adapter->vfs_allocated_count);
941 /* Clear any pending interrupts. */
943 igb_irq_enable(adapter);
945 netif_tx_start_all_queues(adapter->netdev);
947 /* Fire a link change interrupt to start the watchdog. */
948 wr32(E1000_ICS, E1000_ICS_LSC);
952 void igb_down(struct igb_adapter *adapter)
954 struct e1000_hw *hw = &adapter->hw;
955 struct net_device *netdev = adapter->netdev;
959 /* signal that we're down so the interrupt handler does not
960 * reschedule our watchdog timer */
961 set_bit(__IGB_DOWN, &adapter->state);
963 /* disable receives in the hardware */
964 rctl = rd32(E1000_RCTL);
965 wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
966 /* flush and sleep below */
968 netif_tx_stop_all_queues(netdev);
970 /* disable transmits in the hardware */
971 tctl = rd32(E1000_TCTL);
972 tctl &= ~E1000_TCTL_EN;
973 wr32(E1000_TCTL, tctl);
974 /* flush both disables and wait for them to finish */
978 for (i = 0; i < adapter->num_rx_queues; i++)
979 napi_disable(&adapter->rx_ring[i].napi);
981 igb_irq_disable(adapter);
983 del_timer_sync(&adapter->watchdog_timer);
984 del_timer_sync(&adapter->phy_info_timer);
986 netdev->tx_queue_len = adapter->tx_queue_len;
987 netif_carrier_off(netdev);
989 /* record the stats before reset*/
990 igb_update_stats(adapter);
992 adapter->link_speed = 0;
993 adapter->link_duplex = 0;
995 if (!pci_channel_offline(adapter->pdev))
997 igb_clean_all_tx_rings(adapter);
998 igb_clean_all_rx_rings(adapter);
1001 void igb_reinit_locked(struct igb_adapter *adapter)
1003 WARN_ON(in_interrupt());
1004 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
1008 clear_bit(__IGB_RESETTING, &adapter->state);
1011 void igb_reset(struct igb_adapter *adapter)
1013 struct e1000_hw *hw = &adapter->hw;
1014 struct e1000_mac_info *mac = &hw->mac;
1015 struct e1000_fc_info *fc = &hw->fc;
1016 u32 pba = 0, tx_space, min_tx_space, min_rx_space;
1019 /* Repartition Pba for greater than 9k mtu
1020 * To take effect CTRL.RST is required.
1022 switch (mac->type) {
1024 pba = E1000_PBA_64K;
1028 pba = E1000_PBA_34K;
1032 if ((adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) &&
1033 (mac->type < e1000_82576)) {
1034 /* adjust PBA for jumbo frames */
1035 wr32(E1000_PBA, pba);
1037 /* To maintain wire speed transmits, the Tx FIFO should be
1038 * large enough to accommodate two full transmit packets,
1039 * rounded up to the next 1KB and expressed in KB. Likewise,
1040 * the Rx FIFO should be large enough to accommodate at least
1041 * one full receive packet and is similarly rounded up and
1042 * expressed in KB. */
1043 pba = rd32(E1000_PBA);
1044 /* upper 16 bits has Tx packet buffer allocation size in KB */
1045 tx_space = pba >> 16;
1046 /* lower 16 bits has Rx packet buffer allocation size in KB */
1048 /* the tx fifo also stores 16 bytes of information about the tx
1049 * but don't include ethernet FCS because hardware appends it */
1050 min_tx_space = (adapter->max_frame_size +
1051 sizeof(union e1000_adv_tx_desc) -
1053 min_tx_space = ALIGN(min_tx_space, 1024);
1054 min_tx_space >>= 10;
1055 /* software strips receive CRC, so leave room for it */
1056 min_rx_space = adapter->max_frame_size;
1057 min_rx_space = ALIGN(min_rx_space, 1024);
1058 min_rx_space >>= 10;
1060 /* If current Tx allocation is less than the min Tx FIFO size,
1061 * and the min Tx FIFO size is less than the current Rx FIFO
1062 * allocation, take space away from current Rx allocation */
1063 if (tx_space < min_tx_space &&
1064 ((min_tx_space - tx_space) < pba)) {
1065 pba = pba - (min_tx_space - tx_space);
1067 /* if short on rx space, rx wins and must trump tx
1069 if (pba < min_rx_space)
1072 wr32(E1000_PBA, pba);
1075 /* flow control settings */
1076 /* The high water mark must be low enough to fit one full frame
1077 * (or the size used for early receive) above it in the Rx FIFO.
1078 * Set it to the lower of:
1079 * - 90% of the Rx FIFO size, or
1080 * - the full Rx FIFO size minus one full frame */
1081 hwm = min(((pba << 10) * 9 / 10),
1082 ((pba << 10) - 2 * adapter->max_frame_size));
1084 if (mac->type < e1000_82576) {
1085 fc->high_water = hwm & 0xFFF8; /* 8-byte granularity */
1086 fc->low_water = fc->high_water - 8;
1088 fc->high_water = hwm & 0xFFF0; /* 16-byte granularity */
1089 fc->low_water = fc->high_water - 16;
1091 fc->pause_time = 0xFFFF;
1093 fc->type = fc->original_type;
1095 /* disable receive for all VFs and wait one second */
1096 if (adapter->vfs_allocated_count) {
1098 for (i = 0 ; i < adapter->vfs_allocated_count; i++)
1099 adapter->vf_data[i].clear_to_send = false;
1101 /* ping all the active vfs to let them know we are going down */
1102 igb_ping_all_vfs(adapter);
1104 /* disable transmits and receives */
1105 wr32(E1000_VFRE, 0);
1106 wr32(E1000_VFTE, 0);
1109 /* Allow time for pending master requests to run */
1110 adapter->hw.mac.ops.reset_hw(&adapter->hw);
1113 if (adapter->hw.mac.ops.init_hw(&adapter->hw))
1114 dev_err(&adapter->pdev->dev, "Hardware Error\n");
1116 igb_update_mng_vlan(adapter);
1118 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
1119 wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
1121 igb_reset_adaptive(&adapter->hw);
1122 igb_get_phy_info(&adapter->hw);
1125 static const struct net_device_ops igb_netdev_ops = {
1126 .ndo_open = igb_open,
1127 .ndo_stop = igb_close,
1128 .ndo_start_xmit = igb_xmit_frame_adv,
1129 .ndo_get_stats = igb_get_stats,
1130 .ndo_set_multicast_list = igb_set_multi,
1131 .ndo_set_mac_address = igb_set_mac,
1132 .ndo_change_mtu = igb_change_mtu,
1133 .ndo_do_ioctl = igb_ioctl,
1134 .ndo_tx_timeout = igb_tx_timeout,
1135 .ndo_validate_addr = eth_validate_addr,
1136 .ndo_vlan_rx_register = igb_vlan_rx_register,
1137 .ndo_vlan_rx_add_vid = igb_vlan_rx_add_vid,
1138 .ndo_vlan_rx_kill_vid = igb_vlan_rx_kill_vid,
1139 #ifdef CONFIG_NET_POLL_CONTROLLER
1140 .ndo_poll_controller = igb_netpoll,
1145 * igb_probe - Device Initialization Routine
1146 * @pdev: PCI device information struct
1147 * @ent: entry in igb_pci_tbl
1149 * Returns 0 on success, negative on failure
1151 * igb_probe initializes an adapter identified by a pci_dev structure.
1152 * The OS initialization, configuring of the adapter private structure,
1153 * and a hardware reset occur.
1155 static int __devinit igb_probe(struct pci_dev *pdev,
1156 const struct pci_device_id *ent)
1158 struct net_device *netdev;
1159 struct igb_adapter *adapter;
1160 struct e1000_hw *hw;
1161 const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
1162 unsigned long mmio_start, mmio_len;
1163 int err, pci_using_dac;
1164 u16 eeprom_data = 0;
1165 u16 eeprom_apme_mask = IGB_EEPROM_APME;
1168 err = pci_enable_device_mem(pdev);
1173 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
1175 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
1179 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
1181 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
1183 dev_err(&pdev->dev, "No usable DMA "
1184 "configuration, aborting\n");
1190 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
1196 err = pci_enable_pcie_error_reporting(pdev);
1198 dev_err(&pdev->dev, "pci_enable_pcie_error_reporting failed "
1200 /* non-fatal, continue */
1203 pci_set_master(pdev);
1204 pci_save_state(pdev);
1207 netdev = alloc_etherdev_mq(sizeof(struct igb_adapter),
1208 IGB_ABS_MAX_TX_QUEUES);
1210 goto err_alloc_etherdev;
1212 SET_NETDEV_DEV(netdev, &pdev->dev);
1214 pci_set_drvdata(pdev, netdev);
1215 adapter = netdev_priv(netdev);
1216 adapter->netdev = netdev;
1217 adapter->pdev = pdev;
1220 adapter->msg_enable = NETIF_MSG_DRV | NETIF_MSG_PROBE;
1222 mmio_start = pci_resource_start(pdev, 0);
1223 mmio_len = pci_resource_len(pdev, 0);
1226 hw->hw_addr = ioremap(mmio_start, mmio_len);
1230 netdev->netdev_ops = &igb_netdev_ops;
1231 igb_set_ethtool_ops(netdev);
1232 netdev->watchdog_timeo = 5 * HZ;
1234 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
1236 netdev->mem_start = mmio_start;
1237 netdev->mem_end = mmio_start + mmio_len;
1239 /* PCI config space info */
1240 hw->vendor_id = pdev->vendor;
1241 hw->device_id = pdev->device;
1242 hw->revision_id = pdev->revision;
1243 hw->subsystem_vendor_id = pdev->subsystem_vendor;
1244 hw->subsystem_device_id = pdev->subsystem_device;
1246 /* setup the private structure */
1248 /* Copy the default MAC, PHY and NVM function pointers */
1249 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
1250 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
1251 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
1252 /* Initialize skew-specific constants */
1253 err = ei->get_invariants(hw);
1257 #ifdef CONFIG_PCI_IOV
1258 /* since iov functionality isn't critical to base device function we
1259 * can accept failure. If it fails we don't allow iov to be enabled */
1260 if (hw->mac.type == e1000_82576) {
1261 /* 82576 supports a maximum of 7 VFs in addition to the PF */
1262 unsigned int num_vfs = (max_vfs > 7) ? 7 : max_vfs;
1264 unsigned char mac_addr[ETH_ALEN];
1267 adapter->vf_data = kcalloc(num_vfs,
1268 sizeof(struct vf_data_storage),
1270 if (!adapter->vf_data) {
1272 "Could not allocate VF private data - "
1273 "IOV enable failed\n");
1275 err = pci_enable_sriov(pdev, num_vfs);
1277 adapter->vfs_allocated_count = num_vfs;
1278 dev_info(&pdev->dev,
1279 "%d vfs allocated\n",
1282 i < adapter->vfs_allocated_count;
1284 random_ether_addr(mac_addr);
1285 igb_set_vf_mac(adapter, i,
1289 kfree(adapter->vf_data);
1290 adapter->vf_data = NULL;
1297 /* setup the private structure */
1298 err = igb_sw_init(adapter);
1302 igb_get_bus_info_pcie(hw);
1305 switch (hw->mac.type) {
1307 adapter->flags |= IGB_FLAG_NEED_CTX_IDX;
1314 hw->phy.autoneg_wait_to_complete = false;
1315 hw->mac.adaptive_ifs = true;
1317 /* Copper options */
1318 if (hw->phy.media_type == e1000_media_type_copper) {
1319 hw->phy.mdix = AUTO_ALL_MODES;
1320 hw->phy.disable_polarity_correction = false;
1321 hw->phy.ms_type = e1000_ms_hw_default;
1324 if (igb_check_reset_block(hw))
1325 dev_info(&pdev->dev,
1326 "PHY reset is blocked due to SOL/IDER session.\n");
1328 netdev->features = NETIF_F_SG |
1330 NETIF_F_HW_VLAN_TX |
1331 NETIF_F_HW_VLAN_RX |
1332 NETIF_F_HW_VLAN_FILTER;
1334 netdev->features |= NETIF_F_IPV6_CSUM;
1335 netdev->features |= NETIF_F_TSO;
1336 netdev->features |= NETIF_F_TSO6;
1338 netdev->features |= NETIF_F_GRO;
1340 netdev->vlan_features |= NETIF_F_TSO;
1341 netdev->vlan_features |= NETIF_F_TSO6;
1342 netdev->vlan_features |= NETIF_F_IP_CSUM;
1343 netdev->vlan_features |= NETIF_F_SG;
1346 netdev->features |= NETIF_F_HIGHDMA;
1348 adapter->en_mng_pt = igb_enable_mng_pass_thru(&adapter->hw);
1350 /* before reading the NVM, reset the controller to put the device in a
1351 * known good starting state */
1352 hw->mac.ops.reset_hw(hw);
1354 /* make sure the NVM is good */
1355 if (igb_validate_nvm_checksum(hw) < 0) {
1356 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
1361 /* copy the MAC address out of the NVM */
1362 if (hw->mac.ops.read_mac_addr(hw))
1363 dev_err(&pdev->dev, "NVM Read Error\n");
1365 memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
1366 memcpy(netdev->perm_addr, hw->mac.addr, netdev->addr_len);
1368 if (!is_valid_ether_addr(netdev->perm_addr)) {
1369 dev_err(&pdev->dev, "Invalid MAC Address\n");
1374 setup_timer(&adapter->watchdog_timer, &igb_watchdog,
1375 (unsigned long) adapter);
1376 setup_timer(&adapter->phy_info_timer, &igb_update_phy_info,
1377 (unsigned long) adapter);
1379 INIT_WORK(&adapter->reset_task, igb_reset_task);
1380 INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
1382 /* Initialize link properties that are user-changeable */
1383 adapter->fc_autoneg = true;
1384 hw->mac.autoneg = true;
1385 hw->phy.autoneg_advertised = 0x2f;
1387 hw->fc.original_type = e1000_fc_default;
1388 hw->fc.type = e1000_fc_default;
1390 adapter->itr_setting = IGB_DEFAULT_ITR;
1391 adapter->itr = IGB_START_ITR;
1393 igb_validate_mdi_setting(hw);
1395 adapter->rx_csum = 1;
1397 /* Initial Wake on LAN setting If APM wake is enabled in the EEPROM,
1398 * enable the ACPI Magic Packet filter
1401 if (hw->bus.func == 0)
1402 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
1403 else if (hw->bus.func == 1)
1404 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
1406 if (eeprom_data & eeprom_apme_mask)
1407 adapter->eeprom_wol |= E1000_WUFC_MAG;
1409 /* now that we have the eeprom settings, apply the special cases where
1410 * the eeprom may be wrong or the board simply won't support wake on
1411 * lan on a particular port */
1412 switch (pdev->device) {
1413 case E1000_DEV_ID_82575GB_QUAD_COPPER:
1414 adapter->eeprom_wol = 0;
1416 case E1000_DEV_ID_82575EB_FIBER_SERDES:
1417 case E1000_DEV_ID_82576_FIBER:
1418 case E1000_DEV_ID_82576_SERDES:
1419 /* Wake events only supported on port A for dual fiber
1420 * regardless of eeprom setting */
1421 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
1422 adapter->eeprom_wol = 0;
1424 case E1000_DEV_ID_82576_QUAD_COPPER:
1425 /* if quad port adapter, disable WoL on all but port A */
1426 if (global_quad_port_a != 0)
1427 adapter->eeprom_wol = 0;
1429 adapter->flags |= IGB_FLAG_QUAD_PORT_A;
1430 /* Reset for multiple quad port adapters */
1431 if (++global_quad_port_a == 4)
1432 global_quad_port_a = 0;
1436 /* initialize the wol settings based on the eeprom settings */
1437 adapter->wol = adapter->eeprom_wol;
1438 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
1440 /* reset the hardware with the new settings */
1443 /* let the f/w know that the h/w is now under the control of the
1445 igb_get_hw_control(adapter);
1447 strcpy(netdev->name, "eth%d");
1448 err = register_netdev(netdev);
1452 /* carrier off reporting is important to ethtool even BEFORE open */
1453 netif_carrier_off(netdev);
1455 #ifdef CONFIG_IGB_DCA
1456 if (dca_add_requester(&pdev->dev) == 0) {
1457 adapter->flags |= IGB_FLAG_DCA_ENABLED;
1458 dev_info(&pdev->dev, "DCA enabled\n");
1459 /* Always use CB2 mode, difference is masked
1460 * in the CB driver. */
1461 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
1462 igb_setup_dca(adapter);
1467 * Initialize hardware timer: we keep it running just in case
1468 * that some program needs it later on.
1470 memset(&adapter->cycles, 0, sizeof(adapter->cycles));
1471 adapter->cycles.read = igb_read_clock;
1472 adapter->cycles.mask = CLOCKSOURCE_MASK(64);
1473 adapter->cycles.mult = 1;
1474 adapter->cycles.shift = IGB_TSYNC_SHIFT;
1477 IGB_TSYNC_CYCLE_TIME_IN_NANOSECONDS * IGB_TSYNC_SCALE);
1480 * Avoid rollover while we initialize by resetting the time counter.
1482 wr32(E1000_SYSTIML, 0x00000000);
1483 wr32(E1000_SYSTIMH, 0x00000000);
1486 * Set registers so that rollover occurs soon to test this.
1488 wr32(E1000_SYSTIML, 0x00000000);
1489 wr32(E1000_SYSTIMH, 0xFF800000);
1492 timecounter_init(&adapter->clock,
1494 ktime_to_ns(ktime_get_real()));
1497 * Synchronize our NIC clock against system wall clock. NIC
1498 * time stamp reading requires ~3us per sample, each sample
1499 * was pretty stable even under load => only require 10
1500 * samples for each offset comparison.
1502 memset(&adapter->compare, 0, sizeof(adapter->compare));
1503 adapter->compare.source = &adapter->clock;
1504 adapter->compare.target = ktime_get_real;
1505 adapter->compare.num_samples = 10;
1506 timecompare_update(&adapter->compare, 0);
1512 "igb: %s: hw %p initialized timer\n",
1513 igb_get_time_str(adapter, buffer),
1518 dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
1519 /* print bus type/speed/width info */
1520 dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
1522 ((hw->bus.speed == e1000_bus_speed_2500)
1523 ? "2.5Gb/s" : "unknown"),
1524 ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
1525 (hw->bus.width == e1000_bus_width_pcie_x2) ? "Width x2" :
1526 (hw->bus.width == e1000_bus_width_pcie_x1) ? "Width x1" :
1530 igb_read_part_num(hw, &part_num);
1531 dev_info(&pdev->dev, "%s: PBA No: %06x-%03x\n", netdev->name,
1532 (part_num >> 8), (part_num & 0xff));
1534 dev_info(&pdev->dev,
1535 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
1536 adapter->msix_entries ? "MSI-X" :
1537 (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
1538 adapter->num_rx_queues, adapter->num_tx_queues);
1543 igb_release_hw_control(adapter);
1545 if (!igb_check_reset_block(hw))
1548 if (hw->flash_address)
1549 iounmap(hw->flash_address);
1551 igb_free_queues(adapter);
1553 iounmap(hw->hw_addr);
1555 free_netdev(netdev);
1557 pci_release_selected_regions(pdev, pci_select_bars(pdev,
1561 pci_disable_device(pdev);
1566 * igb_remove - Device Removal Routine
1567 * @pdev: PCI device information struct
1569 * igb_remove is called by the PCI subsystem to alert the driver
1570 * that it should release a PCI device. The could be caused by a
1571 * Hot-Plug event, or because the driver is going to be removed from
1574 static void __devexit igb_remove(struct pci_dev *pdev)
1576 struct net_device *netdev = pci_get_drvdata(pdev);
1577 struct igb_adapter *adapter = netdev_priv(netdev);
1578 struct e1000_hw *hw = &adapter->hw;
1581 /* flush_scheduled work may reschedule our watchdog task, so
1582 * explicitly disable watchdog tasks from being rescheduled */
1583 set_bit(__IGB_DOWN, &adapter->state);
1584 del_timer_sync(&adapter->watchdog_timer);
1585 del_timer_sync(&adapter->phy_info_timer);
1587 flush_scheduled_work();
1589 #ifdef CONFIG_IGB_DCA
1590 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
1591 dev_info(&pdev->dev, "DCA disabled\n");
1592 dca_remove_requester(&pdev->dev);
1593 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
1594 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
1598 /* Release control of h/w to f/w. If f/w is AMT enabled, this
1599 * would have already happened in close and is redundant. */
1600 igb_release_hw_control(adapter);
1602 unregister_netdev(netdev);
1604 if (!igb_check_reset_block(&adapter->hw))
1605 igb_reset_phy(&adapter->hw);
1607 igb_reset_interrupt_capability(adapter);
1609 igb_free_queues(adapter);
1611 #ifdef CONFIG_PCI_IOV
1612 /* reclaim resources allocated to VFs */
1613 if (adapter->vf_data) {
1614 /* disable iov and allow time for transactions to clear */
1615 pci_disable_sriov(pdev);
1618 kfree(adapter->vf_data);
1619 adapter->vf_data = NULL;
1620 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
1622 dev_info(&pdev->dev, "IOV Disabled\n");
1625 iounmap(hw->hw_addr);
1626 if (hw->flash_address)
1627 iounmap(hw->flash_address);
1628 pci_release_selected_regions(pdev, pci_select_bars(pdev,
1631 free_netdev(netdev);
1633 err = pci_disable_pcie_error_reporting(pdev);
1636 "pci_disable_pcie_error_reporting failed 0x%x\n", err);
1638 pci_disable_device(pdev);
1642 * igb_sw_init - Initialize general software structures (struct igb_adapter)
1643 * @adapter: board private structure to initialize
1645 * igb_sw_init initializes the Adapter private data structure.
1646 * Fields are initialized based on PCI device information and
1647 * OS network device settings (MTU size).
1649 static int __devinit igb_sw_init(struct igb_adapter *adapter)
1651 struct e1000_hw *hw = &adapter->hw;
1652 struct net_device *netdev = adapter->netdev;
1653 struct pci_dev *pdev = adapter->pdev;
1655 pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
1657 adapter->tx_ring_count = IGB_DEFAULT_TXD;
1658 adapter->rx_ring_count = IGB_DEFAULT_RXD;
1659 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1660 adapter->rx_ps_hdr_size = 0; /* disable packet split */
1661 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1662 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1664 /* This call may decrease the number of queues depending on
1665 * interrupt mode. */
1666 igb_set_interrupt_capability(adapter);
1668 if (igb_alloc_queues(adapter)) {
1669 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
1673 /* Explicitly disable IRQ since the NIC can be in any state. */
1674 igb_irq_disable(adapter);
1676 set_bit(__IGB_DOWN, &adapter->state);
1681 * igb_open - Called when a network interface is made active
1682 * @netdev: network interface device structure
1684 * Returns 0 on success, negative value on failure
1686 * The open entry point is called when a network interface is made
1687 * active by the system (IFF_UP). At this point all resources needed
1688 * for transmit and receive operations are allocated, the interrupt
1689 * handler is registered with the OS, the watchdog timer is started,
1690 * and the stack is notified that the interface is ready.
1692 static int igb_open(struct net_device *netdev)
1694 struct igb_adapter *adapter = netdev_priv(netdev);
1695 struct e1000_hw *hw = &adapter->hw;
1699 /* disallow open during test */
1700 if (test_bit(__IGB_TESTING, &adapter->state))
1703 netif_carrier_off(netdev);
1705 /* allocate transmit descriptors */
1706 err = igb_setup_all_tx_resources(adapter);
1710 /* allocate receive descriptors */
1711 err = igb_setup_all_rx_resources(adapter);
1715 /* e1000_power_up_phy(adapter); */
1717 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1718 if ((adapter->hw.mng_cookie.status &
1719 E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
1720 igb_update_mng_vlan(adapter);
1722 /* before we allocate an interrupt, we must be ready to handle it.
1723 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1724 * as soon as we call pci_request_irq, so we have to setup our
1725 * clean_rx handler before we do so. */
1726 igb_configure(adapter);
1728 igb_vmm_control(adapter);
1729 igb_set_rah_pool(hw, adapter->vfs_allocated_count, 0);
1730 igb_set_vmolr(hw, adapter->vfs_allocated_count);
1732 err = igb_request_irq(adapter);
1736 /* From here on the code is the same as igb_up() */
1737 clear_bit(__IGB_DOWN, &adapter->state);
1739 for (i = 0; i < adapter->num_rx_queues; i++)
1740 napi_enable(&adapter->rx_ring[i].napi);
1742 /* Clear any pending interrupts. */
1745 igb_irq_enable(adapter);
1747 netif_tx_start_all_queues(netdev);
1749 /* Fire a link status change interrupt to start the watchdog. */
1750 wr32(E1000_ICS, E1000_ICS_LSC);
1755 igb_release_hw_control(adapter);
1756 /* e1000_power_down_phy(adapter); */
1757 igb_free_all_rx_resources(adapter);
1759 igb_free_all_tx_resources(adapter);
1767 * igb_close - Disables a network interface
1768 * @netdev: network interface device structure
1770 * Returns 0, this is not allowed to fail
1772 * The close entry point is called when an interface is de-activated
1773 * by the OS. The hardware is still under the driver's control, but
1774 * needs to be disabled. A global MAC reset is issued to stop the
1775 * hardware, and all transmit and receive resources are freed.
1777 static int igb_close(struct net_device *netdev)
1779 struct igb_adapter *adapter = netdev_priv(netdev);
1781 WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
1784 igb_free_irq(adapter);
1786 igb_free_all_tx_resources(adapter);
1787 igb_free_all_rx_resources(adapter);
1789 /* kill manageability vlan ID if supported, but not if a vlan with
1790 * the same ID is registered on the host OS (let 8021q kill it) */
1791 if ((adapter->hw.mng_cookie.status &
1792 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1794 vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
1795 igb_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1801 * igb_setup_tx_resources - allocate Tx resources (Descriptors)
1802 * @adapter: board private structure
1803 * @tx_ring: tx descriptor ring (for a specific queue) to setup
1805 * Return 0 on success, negative on failure
1807 int igb_setup_tx_resources(struct igb_adapter *adapter,
1808 struct igb_ring *tx_ring)
1810 struct pci_dev *pdev = adapter->pdev;
1813 size = sizeof(struct igb_buffer) * tx_ring->count;
1814 tx_ring->buffer_info = vmalloc(size);
1815 if (!tx_ring->buffer_info)
1817 memset(tx_ring->buffer_info, 0, size);
1819 /* round up to nearest 4K */
1820 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
1821 tx_ring->size = ALIGN(tx_ring->size, 4096);
1823 tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
1829 tx_ring->adapter = adapter;
1830 tx_ring->next_to_use = 0;
1831 tx_ring->next_to_clean = 0;
1835 vfree(tx_ring->buffer_info);
1836 dev_err(&adapter->pdev->dev,
1837 "Unable to allocate memory for the transmit descriptor ring\n");
1842 * igb_setup_all_tx_resources - wrapper to allocate Tx resources
1843 * (Descriptors) for all queues
1844 * @adapter: board private structure
1846 * Return 0 on success, negative on failure
1848 static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
1853 for (i = 0; i < adapter->num_tx_queues; i++) {
1854 err = igb_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1856 dev_err(&adapter->pdev->dev,
1857 "Allocation for Tx Queue %u failed\n", i);
1858 for (i--; i >= 0; i--)
1859 igb_free_tx_resources(&adapter->tx_ring[i]);
1864 for (i = 0; i < IGB_MAX_TX_QUEUES; i++) {
1865 r_idx = i % adapter->num_tx_queues;
1866 adapter->multi_tx_table[i] = &adapter->tx_ring[r_idx];
1872 * igb_configure_tx - Configure transmit Unit after Reset
1873 * @adapter: board private structure
1875 * Configure the Tx unit of the MAC after a reset.
1877 static void igb_configure_tx(struct igb_adapter *adapter)
1880 struct e1000_hw *hw = &adapter->hw;
1885 for (i = 0; i < adapter->num_tx_queues; i++) {
1886 struct igb_ring *ring = &adapter->tx_ring[i];
1888 wr32(E1000_TDLEN(j),
1889 ring->count * sizeof(union e1000_adv_tx_desc));
1891 wr32(E1000_TDBAL(j),
1892 tdba & 0x00000000ffffffffULL);
1893 wr32(E1000_TDBAH(j), tdba >> 32);
1895 ring->head = E1000_TDH(j);
1896 ring->tail = E1000_TDT(j);
1897 writel(0, hw->hw_addr + ring->tail);
1898 writel(0, hw->hw_addr + ring->head);
1899 txdctl = rd32(E1000_TXDCTL(j));
1900 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1901 wr32(E1000_TXDCTL(j), txdctl);
1903 /* Turn off Relaxed Ordering on head write-backs. The
1904 * writebacks MUST be delivered in order or it will
1905 * completely screw up our bookeeping.
1907 txctrl = rd32(E1000_DCA_TXCTRL(j));
1908 txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1909 wr32(E1000_DCA_TXCTRL(j), txctrl);
1912 /* disable queue 0 to prevent tail bump w/o re-configuration */
1913 if (adapter->vfs_allocated_count)
1914 wr32(E1000_TXDCTL(0), 0);
1916 /* Program the Transmit Control Register */
1917 tctl = rd32(E1000_TCTL);
1918 tctl &= ~E1000_TCTL_CT;
1919 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1920 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1922 igb_config_collision_dist(hw);
1924 /* Setup Transmit Descriptor Settings for eop descriptor */
1925 adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_RS;
1927 /* Enable transmits */
1928 tctl |= E1000_TCTL_EN;
1930 wr32(E1000_TCTL, tctl);
1934 * igb_setup_rx_resources - allocate Rx resources (Descriptors)
1935 * @adapter: board private structure
1936 * @rx_ring: rx descriptor ring (for a specific queue) to setup
1938 * Returns 0 on success, negative on failure
1940 int igb_setup_rx_resources(struct igb_adapter *adapter,
1941 struct igb_ring *rx_ring)
1943 struct pci_dev *pdev = adapter->pdev;
1946 size = sizeof(struct igb_buffer) * rx_ring->count;
1947 rx_ring->buffer_info = vmalloc(size);
1948 if (!rx_ring->buffer_info)
1950 memset(rx_ring->buffer_info, 0, size);
1952 desc_len = sizeof(union e1000_adv_rx_desc);
1954 /* Round up to nearest 4K */
1955 rx_ring->size = rx_ring->count * desc_len;
1956 rx_ring->size = ALIGN(rx_ring->size, 4096);
1958 rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size,
1964 rx_ring->next_to_clean = 0;
1965 rx_ring->next_to_use = 0;
1967 rx_ring->adapter = adapter;
1972 vfree(rx_ring->buffer_info);
1973 dev_err(&adapter->pdev->dev, "Unable to allocate memory for "
1974 "the receive descriptor ring\n");
1979 * igb_setup_all_rx_resources - wrapper to allocate Rx resources
1980 * (Descriptors) for all queues
1981 * @adapter: board private structure
1983 * Return 0 on success, negative on failure
1985 static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
1989 for (i = 0; i < adapter->num_rx_queues; i++) {
1990 err = igb_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1992 dev_err(&adapter->pdev->dev,
1993 "Allocation for Rx Queue %u failed\n", i);
1994 for (i--; i >= 0; i--)
1995 igb_free_rx_resources(&adapter->rx_ring[i]);
2004 * igb_setup_rctl - configure the receive control registers
2005 * @adapter: Board private structure
2007 static void igb_setup_rctl(struct igb_adapter *adapter)
2009 struct e1000_hw *hw = &adapter->hw;
2014 rctl = rd32(E1000_RCTL);
2016 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2017 rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
2019 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
2020 (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2023 * enable stripping of CRC. It's unlikely this will break BMC
2024 * redirection as it did with e1000. Newer features require
2025 * that the HW strips the CRC.
2027 rctl |= E1000_RCTL_SECRC;
2030 * disable store bad packets and clear size bits.
2032 rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
2034 /* enable LPE when to prevent packets larger than max_frame_size */
2035 rctl |= E1000_RCTL_LPE;
2037 /* Setup buffer sizes */
2038 switch (adapter->rx_buffer_len) {
2039 case IGB_RXBUFFER_256:
2040 rctl |= E1000_RCTL_SZ_256;
2042 case IGB_RXBUFFER_512:
2043 rctl |= E1000_RCTL_SZ_512;
2046 srrctl = ALIGN(adapter->rx_buffer_len, 1024)
2047 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
2051 /* 82575 and greater support packet-split where the protocol
2052 * header is placed in skb->data and the packet data is
2053 * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2054 * In the case of a non-split, skb->data is linearly filled,
2055 * followed by the page buffers. Therefore, skb->data is
2056 * sized to hold the largest protocol header.
2058 /* allocations using alloc_page take too long for regular MTU
2059 * so only enable packet split for jumbo frames */
2060 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2061 adapter->rx_ps_hdr_size = IGB_RXBUFFER_128;
2062 srrctl |= adapter->rx_ps_hdr_size <<
2063 E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
2064 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2066 adapter->rx_ps_hdr_size = 0;
2067 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
2070 /* Attention!!! For SR-IOV PF driver operations you must enable
2071 * queue drop for all VF and PF queues to prevent head of line blocking
2072 * if an un-trusted VF does not provide descriptors to hardware.
2074 if (adapter->vfs_allocated_count) {
2077 j = adapter->rx_ring[0].reg_idx;
2079 /* set all queue drop enable bits */
2080 wr32(E1000_QDE, ALL_QUEUES);
2081 srrctl |= E1000_SRRCTL_DROP_EN;
2083 /* disable queue 0 to prevent tail write w/o re-config */
2084 wr32(E1000_RXDCTL(0), 0);
2086 vmolr = rd32(E1000_VMOLR(j));
2087 if (rctl & E1000_RCTL_LPE)
2088 vmolr |= E1000_VMOLR_LPE;
2089 if (adapter->num_rx_queues > 0)
2090 vmolr |= E1000_VMOLR_RSSE;
2091 wr32(E1000_VMOLR(j), vmolr);
2094 for (i = 0; i < adapter->num_rx_queues; i++) {
2095 j = adapter->rx_ring[i].reg_idx;
2096 wr32(E1000_SRRCTL(j), srrctl);
2099 wr32(E1000_RCTL, rctl);
2103 * igb_rlpml_set - set maximum receive packet size
2104 * @adapter: board private structure
2106 * Configure maximum receivable packet size.
2108 static void igb_rlpml_set(struct igb_adapter *adapter)
2110 u32 max_frame_size = adapter->max_frame_size;
2111 struct e1000_hw *hw = &adapter->hw;
2112 u16 pf_id = adapter->vfs_allocated_count;
2115 max_frame_size += VLAN_TAG_SIZE;
2117 /* if vfs are enabled we set RLPML to the largest possible request
2118 * size and set the VMOLR RLPML to the size we need */
2120 igb_set_vf_rlpml(adapter, max_frame_size, pf_id);
2121 max_frame_size = MAX_STD_JUMBO_FRAME_SIZE + VLAN_TAG_SIZE;
2124 wr32(E1000_RLPML, max_frame_size);
2128 * igb_configure_vt_default_pool - Configure VT default pool
2129 * @adapter: board private structure
2131 * Configure the default pool
2133 static void igb_configure_vt_default_pool(struct igb_adapter *adapter)
2135 struct e1000_hw *hw = &adapter->hw;
2136 u16 pf_id = adapter->vfs_allocated_count;
2139 /* not in sr-iov mode - do nothing */
2143 vtctl = rd32(E1000_VT_CTL);
2144 vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK |
2145 E1000_VT_CTL_DISABLE_DEF_POOL);
2146 vtctl |= pf_id << E1000_VT_CTL_DEFAULT_POOL_SHIFT;
2147 wr32(E1000_VT_CTL, vtctl);
2151 * igb_configure_rx - Configure receive Unit after Reset
2152 * @adapter: board private structure
2154 * Configure the Rx unit of the MAC after a reset.
2156 static void igb_configure_rx(struct igb_adapter *adapter)
2159 struct e1000_hw *hw = &adapter->hw;
2164 /* disable receives while setting up the descriptors */
2165 rctl = rd32(E1000_RCTL);
2166 wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
2170 if (adapter->itr_setting > 3)
2171 wr32(E1000_ITR, adapter->itr);
2173 /* Setup the HW Rx Head and Tail Descriptor Pointers and
2174 * the Base and Length of the Rx Descriptor Ring */
2175 for (i = 0; i < adapter->num_rx_queues; i++) {
2176 struct igb_ring *ring = &adapter->rx_ring[i];
2177 int j = ring->reg_idx;
2179 wr32(E1000_RDBAL(j),
2180 rdba & 0x00000000ffffffffULL);
2181 wr32(E1000_RDBAH(j), rdba >> 32);
2182 wr32(E1000_RDLEN(j),
2183 ring->count * sizeof(union e1000_adv_rx_desc));
2185 ring->head = E1000_RDH(j);
2186 ring->tail = E1000_RDT(j);
2187 writel(0, hw->hw_addr + ring->tail);
2188 writel(0, hw->hw_addr + ring->head);
2190 rxdctl = rd32(E1000_RXDCTL(j));
2191 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
2192 rxdctl &= 0xFFF00000;
2193 rxdctl |= IGB_RX_PTHRESH;
2194 rxdctl |= IGB_RX_HTHRESH << 8;
2195 rxdctl |= IGB_RX_WTHRESH << 16;
2196 wr32(E1000_RXDCTL(j), rxdctl);
2199 if (adapter->num_rx_queues > 1) {
2208 get_random_bytes(&random[0], 40);
2210 if (hw->mac.type >= e1000_82576)
2214 for (j = 0; j < (32 * 4); j++) {
2216 adapter->rx_ring[(j % adapter->num_rx_queues)].reg_idx << shift;
2219 hw->hw_addr + E1000_RETA(0) + (j & ~3));
2221 if (adapter->vfs_allocated_count)
2222 mrqc = E1000_MRQC_ENABLE_VMDQ_RSS_2Q;
2224 mrqc = E1000_MRQC_ENABLE_RSS_4Q;
2226 /* Fill out hash function seeds */
2227 for (j = 0; j < 10; j++)
2228 array_wr32(E1000_RSSRK(0), j, random[j]);
2230 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
2231 E1000_MRQC_RSS_FIELD_IPV4_TCP);
2232 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6 |
2233 E1000_MRQC_RSS_FIELD_IPV6_TCP);
2234 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4_UDP |
2235 E1000_MRQC_RSS_FIELD_IPV6_UDP);
2236 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
2237 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
2239 wr32(E1000_MRQC, mrqc);
2240 } else if (adapter->vfs_allocated_count) {
2241 /* Enable multi-queue for sr-iov */
2242 wr32(E1000_MRQC, E1000_MRQC_ENABLE_VMDQ);
2245 /* Enable Receive Checksum Offload for TCP and UDP */
2246 rxcsum = rd32(E1000_RXCSUM);
2247 /* Disable raw packet checksumming */
2248 rxcsum |= E1000_RXCSUM_PCSD;
2249 /* Don't need to set TUOFL or IPOFL, they default to 1 */
2250 if (!adapter->rx_csum)
2251 rxcsum &= ~(E1000_RXCSUM_TUOFL | E1000_RXCSUM_IPOFL);
2252 wr32(E1000_RXCSUM, rxcsum);
2254 /* Set the default pool for the PF's first queue */
2255 igb_configure_vt_default_pool(adapter);
2257 igb_rlpml_set(adapter);
2259 /* Enable Receives */
2260 wr32(E1000_RCTL, rctl);
2264 * igb_free_tx_resources - Free Tx Resources per Queue
2265 * @tx_ring: Tx descriptor ring for a specific queue
2267 * Free all transmit software resources
2269 void igb_free_tx_resources(struct igb_ring *tx_ring)
2271 struct pci_dev *pdev = tx_ring->adapter->pdev;
2273 igb_clean_tx_ring(tx_ring);
2275 vfree(tx_ring->buffer_info);
2276 tx_ring->buffer_info = NULL;
2278 pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
2280 tx_ring->desc = NULL;
2284 * igb_free_all_tx_resources - Free Tx Resources for All Queues
2285 * @adapter: board private structure
2287 * Free all transmit software resources
2289 static void igb_free_all_tx_resources(struct igb_adapter *adapter)
2293 for (i = 0; i < adapter->num_tx_queues; i++)
2294 igb_free_tx_resources(&adapter->tx_ring[i]);
2297 static void igb_unmap_and_free_tx_resource(struct igb_adapter *adapter,
2298 struct igb_buffer *buffer_info)
2300 buffer_info->dma = 0;
2301 if (buffer_info->skb) {
2302 skb_dma_unmap(&adapter->pdev->dev, buffer_info->skb,
2304 dev_kfree_skb_any(buffer_info->skb);
2305 buffer_info->skb = NULL;
2307 buffer_info->time_stamp = 0;
2308 /* buffer_info must be completely set up in the transmit path */
2312 * igb_clean_tx_ring - Free Tx Buffers
2313 * @tx_ring: ring to be cleaned
2315 static void igb_clean_tx_ring(struct igb_ring *tx_ring)
2317 struct igb_adapter *adapter = tx_ring->adapter;
2318 struct igb_buffer *buffer_info;
2322 if (!tx_ring->buffer_info)
2324 /* Free all the Tx ring sk_buffs */
2326 for (i = 0; i < tx_ring->count; i++) {
2327 buffer_info = &tx_ring->buffer_info[i];
2328 igb_unmap_and_free_tx_resource(adapter, buffer_info);
2331 size = sizeof(struct igb_buffer) * tx_ring->count;
2332 memset(tx_ring->buffer_info, 0, size);
2334 /* Zero out the descriptor ring */
2336 memset(tx_ring->desc, 0, tx_ring->size);
2338 tx_ring->next_to_use = 0;
2339 tx_ring->next_to_clean = 0;
2341 writel(0, adapter->hw.hw_addr + tx_ring->head);
2342 writel(0, adapter->hw.hw_addr + tx_ring->tail);
2346 * igb_clean_all_tx_rings - Free Tx Buffers for all queues
2347 * @adapter: board private structure
2349 static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
2353 for (i = 0; i < adapter->num_tx_queues; i++)
2354 igb_clean_tx_ring(&adapter->tx_ring[i]);
2358 * igb_free_rx_resources - Free Rx Resources
2359 * @rx_ring: ring to clean the resources from
2361 * Free all receive software resources
2363 void igb_free_rx_resources(struct igb_ring *rx_ring)
2365 struct pci_dev *pdev = rx_ring->adapter->pdev;
2367 igb_clean_rx_ring(rx_ring);
2369 vfree(rx_ring->buffer_info);
2370 rx_ring->buffer_info = NULL;
2372 pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
2374 rx_ring->desc = NULL;
2378 * igb_free_all_rx_resources - Free Rx Resources for All Queues
2379 * @adapter: board private structure
2381 * Free all receive software resources
2383 static void igb_free_all_rx_resources(struct igb_adapter *adapter)
2387 for (i = 0; i < adapter->num_rx_queues; i++)
2388 igb_free_rx_resources(&adapter->rx_ring[i]);
2392 * igb_clean_rx_ring - Free Rx Buffers per Queue
2393 * @rx_ring: ring to free buffers from
2395 static void igb_clean_rx_ring(struct igb_ring *rx_ring)
2397 struct igb_adapter *adapter = rx_ring->adapter;
2398 struct igb_buffer *buffer_info;
2399 struct pci_dev *pdev = adapter->pdev;
2403 if (!rx_ring->buffer_info)
2405 /* Free all the Rx ring sk_buffs */
2406 for (i = 0; i < rx_ring->count; i++) {
2407 buffer_info = &rx_ring->buffer_info[i];
2408 if (buffer_info->dma) {
2409 if (adapter->rx_ps_hdr_size)
2410 pci_unmap_single(pdev, buffer_info->dma,
2411 adapter->rx_ps_hdr_size,
2412 PCI_DMA_FROMDEVICE);
2414 pci_unmap_single(pdev, buffer_info->dma,
2415 adapter->rx_buffer_len,
2416 PCI_DMA_FROMDEVICE);
2417 buffer_info->dma = 0;
2420 if (buffer_info->skb) {
2421 dev_kfree_skb(buffer_info->skb);
2422 buffer_info->skb = NULL;
2424 if (buffer_info->page) {
2425 if (buffer_info->page_dma)
2426 pci_unmap_page(pdev, buffer_info->page_dma,
2428 PCI_DMA_FROMDEVICE);
2429 put_page(buffer_info->page);
2430 buffer_info->page = NULL;
2431 buffer_info->page_dma = 0;
2432 buffer_info->page_offset = 0;
2436 size = sizeof(struct igb_buffer) * rx_ring->count;
2437 memset(rx_ring->buffer_info, 0, size);
2439 /* Zero out the descriptor ring */
2440 memset(rx_ring->desc, 0, rx_ring->size);
2442 rx_ring->next_to_clean = 0;
2443 rx_ring->next_to_use = 0;
2445 writel(0, adapter->hw.hw_addr + rx_ring->head);
2446 writel(0, adapter->hw.hw_addr + rx_ring->tail);
2450 * igb_clean_all_rx_rings - Free Rx Buffers for all queues
2451 * @adapter: board private structure
2453 static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
2457 for (i = 0; i < adapter->num_rx_queues; i++)
2458 igb_clean_rx_ring(&adapter->rx_ring[i]);
2462 * igb_set_mac - Change the Ethernet Address of the NIC
2463 * @netdev: network interface device structure
2464 * @p: pointer to an address structure
2466 * Returns 0 on success, negative on failure
2468 static int igb_set_mac(struct net_device *netdev, void *p)
2470 struct igb_adapter *adapter = netdev_priv(netdev);
2471 struct e1000_hw *hw = &adapter->hw;
2472 struct sockaddr *addr = p;
2474 if (!is_valid_ether_addr(addr->sa_data))
2475 return -EADDRNOTAVAIL;
2477 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2478 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
2480 hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
2482 igb_set_rah_pool(hw, adapter->vfs_allocated_count, 0);
2488 * igb_set_multi - Multicast and Promiscuous mode set
2489 * @netdev: network interface device structure
2491 * The set_multi entry point is called whenever the multicast address
2492 * list or the network interface flags are updated. This routine is
2493 * responsible for configuring the hardware for proper multicast,
2494 * promiscuous mode, and all-multi behavior.
2496 static void igb_set_multi(struct net_device *netdev)
2498 struct igb_adapter *adapter = netdev_priv(netdev);
2499 struct e1000_hw *hw = &adapter->hw;
2500 struct e1000_mac_info *mac = &hw->mac;
2501 struct dev_mc_list *mc_ptr;
2502 u8 *mta_list = NULL;
2506 /* Check for Promiscuous and All Multicast modes */
2508 rctl = rd32(E1000_RCTL);
2510 if (netdev->flags & IFF_PROMISC) {
2511 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2512 rctl &= ~E1000_RCTL_VFE;
2514 if (netdev->flags & IFF_ALLMULTI) {
2515 rctl |= E1000_RCTL_MPE;
2516 rctl &= ~E1000_RCTL_UPE;
2518 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2519 rctl |= E1000_RCTL_VFE;
2521 wr32(E1000_RCTL, rctl);
2523 if (netdev->mc_count) {
2524 mta_list = kzalloc(netdev->mc_count * 6, GFP_ATOMIC);
2526 dev_err(&adapter->pdev->dev,
2527 "failed to allocate multicast filter list\n");
2532 /* The shared function expects a packed array of only addresses. */
2533 mc_ptr = netdev->mc_list;
2535 for (i = 0; i < netdev->mc_count; i++) {
2538 memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr, ETH_ALEN);
2539 mc_ptr = mc_ptr->next;
2541 igb_update_mc_addr_list(hw, mta_list, i,
2542 adapter->vfs_allocated_count + 1,
2543 mac->rar_entry_count);
2545 igb_set_mc_list_pools(adapter, i, mac->rar_entry_count);
2546 igb_restore_vf_multicasts(adapter);
2551 /* Need to wait a few seconds after link up to get diagnostic information from
2553 static void igb_update_phy_info(unsigned long data)
2555 struct igb_adapter *adapter = (struct igb_adapter *) data;
2556 igb_get_phy_info(&adapter->hw);
2560 * igb_has_link - check shared code for link and determine up/down
2561 * @adapter: pointer to driver private info
2563 static bool igb_has_link(struct igb_adapter *adapter)
2565 struct e1000_hw *hw = &adapter->hw;
2566 bool link_active = false;
2569 /* get_link_status is set on LSC (link status) interrupt or
2570 * rx sequence error interrupt. get_link_status will stay
2571 * false until the e1000_check_for_link establishes link
2572 * for copper adapters ONLY
2574 switch (hw->phy.media_type) {
2575 case e1000_media_type_copper:
2576 if (hw->mac.get_link_status) {
2577 ret_val = hw->mac.ops.check_for_link(hw);
2578 link_active = !hw->mac.get_link_status;
2583 case e1000_media_type_fiber:
2584 ret_val = hw->mac.ops.check_for_link(hw);
2585 link_active = !!(rd32(E1000_STATUS) & E1000_STATUS_LU);
2587 case e1000_media_type_internal_serdes:
2588 ret_val = hw->mac.ops.check_for_link(hw);
2589 link_active = hw->mac.serdes_has_link;
2592 case e1000_media_type_unknown:
2600 * igb_watchdog - Timer Call-back
2601 * @data: pointer to adapter cast into an unsigned long
2603 static void igb_watchdog(unsigned long data)
2605 struct igb_adapter *adapter = (struct igb_adapter *)data;
2606 /* Do the rest outside of interrupt context */
2607 schedule_work(&adapter->watchdog_task);
2610 static void igb_watchdog_task(struct work_struct *work)
2612 struct igb_adapter *adapter = container_of(work,
2613 struct igb_adapter, watchdog_task);
2614 struct e1000_hw *hw = &adapter->hw;
2615 struct net_device *netdev = adapter->netdev;
2616 struct igb_ring *tx_ring = adapter->tx_ring;
2621 link = igb_has_link(adapter);
2622 if ((netif_carrier_ok(netdev)) && link)
2626 if (!netif_carrier_ok(netdev)) {
2628 hw->mac.ops.get_speed_and_duplex(&adapter->hw,
2629 &adapter->link_speed,
2630 &adapter->link_duplex);
2632 ctrl = rd32(E1000_CTRL);
2633 /* Links status message must follow this format */
2634 printk(KERN_INFO "igb: %s NIC Link is Up %d Mbps %s, "
2635 "Flow Control: %s\n",
2637 adapter->link_speed,
2638 adapter->link_duplex == FULL_DUPLEX ?
2639 "Full Duplex" : "Half Duplex",
2640 ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2641 E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2642 E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2643 E1000_CTRL_TFCE) ? "TX" : "None")));
2645 /* tweak tx_queue_len according to speed/duplex and
2646 * adjust the timeout factor */
2647 netdev->tx_queue_len = adapter->tx_queue_len;
2648 adapter->tx_timeout_factor = 1;
2649 switch (adapter->link_speed) {
2651 netdev->tx_queue_len = 10;
2652 adapter->tx_timeout_factor = 14;
2655 netdev->tx_queue_len = 100;
2656 /* maybe add some timeout factor ? */
2660 netif_carrier_on(netdev);
2662 igb_ping_all_vfs(adapter);
2664 /* link state has changed, schedule phy info update */
2665 if (!test_bit(__IGB_DOWN, &adapter->state))
2666 mod_timer(&adapter->phy_info_timer,
2667 round_jiffies(jiffies + 2 * HZ));
2670 if (netif_carrier_ok(netdev)) {
2671 adapter->link_speed = 0;
2672 adapter->link_duplex = 0;
2673 /* Links status message must follow this format */
2674 printk(KERN_INFO "igb: %s NIC Link is Down\n",
2676 netif_carrier_off(netdev);
2678 igb_ping_all_vfs(adapter);
2680 /* link state has changed, schedule phy info update */
2681 if (!test_bit(__IGB_DOWN, &adapter->state))
2682 mod_timer(&adapter->phy_info_timer,
2683 round_jiffies(jiffies + 2 * HZ));
2688 igb_update_stats(adapter);
2690 hw->mac.tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2691 adapter->tpt_old = adapter->stats.tpt;
2692 hw->mac.collision_delta = adapter->stats.colc - adapter->colc_old;
2693 adapter->colc_old = adapter->stats.colc;
2695 adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
2696 adapter->gorc_old = adapter->stats.gorc;
2697 adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
2698 adapter->gotc_old = adapter->stats.gotc;
2700 igb_update_adaptive(&adapter->hw);
2702 if (!netif_carrier_ok(netdev)) {
2703 if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) {
2704 /* We've lost link, so the controller stops DMA,
2705 * but we've got queued Tx work that's never going
2706 * to get done, so reset controller to flush Tx.
2707 * (Do the reset outside of interrupt context). */
2708 adapter->tx_timeout_count++;
2709 schedule_work(&adapter->reset_task);
2713 /* Cause software interrupt to ensure rx ring is cleaned */
2714 if (adapter->msix_entries) {
2715 for (i = 0; i < adapter->num_rx_queues; i++)
2716 eics |= adapter->rx_ring[i].eims_value;
2717 wr32(E1000_EICS, eics);
2719 wr32(E1000_ICS, E1000_ICS_RXDMT0);
2722 /* Force detection of hung controller every watchdog period */
2723 tx_ring->detect_tx_hung = true;
2725 /* Reset the timer */
2726 if (!test_bit(__IGB_DOWN, &adapter->state))
2727 mod_timer(&adapter->watchdog_timer,
2728 round_jiffies(jiffies + 2 * HZ));
2731 enum latency_range {
2735 latency_invalid = 255
2740 * igb_update_ring_itr - update the dynamic ITR value based on packet size
2742 * Stores a new ITR value based on strictly on packet size. This
2743 * algorithm is less sophisticated than that used in igb_update_itr,
2744 * due to the difficulty of synchronizing statistics across multiple
2745 * receive rings. The divisors and thresholds used by this fuction
2746 * were determined based on theoretical maximum wire speed and testing
2747 * data, in order to minimize response time while increasing bulk
2749 * This functionality is controlled by the InterruptThrottleRate module
2750 * parameter (see igb_param.c)
2751 * NOTE: This function is called only when operating in a multiqueue
2752 * receive environment.
2753 * @rx_ring: pointer to ring
2755 static void igb_update_ring_itr(struct igb_ring *rx_ring)
2757 int new_val = rx_ring->itr_val;
2758 int avg_wire_size = 0;
2759 struct igb_adapter *adapter = rx_ring->adapter;
2761 if (!rx_ring->total_packets)
2762 goto clear_counts; /* no packets, so don't do anything */
2764 /* For non-gigabit speeds, just fix the interrupt rate at 4000
2765 * ints/sec - ITR timer value of 120 ticks.
2767 if (adapter->link_speed != SPEED_1000) {
2771 avg_wire_size = rx_ring->total_bytes / rx_ring->total_packets;
2773 /* Add 24 bytes to size to account for CRC, preamble, and gap */
2774 avg_wire_size += 24;
2776 /* Don't starve jumbo frames */
2777 avg_wire_size = min(avg_wire_size, 3000);
2779 /* Give a little boost to mid-size frames */
2780 if ((avg_wire_size > 300) && (avg_wire_size < 1200))
2781 new_val = avg_wire_size / 3;
2783 new_val = avg_wire_size / 2;
2786 if (new_val != rx_ring->itr_val) {
2787 rx_ring->itr_val = new_val;
2788 rx_ring->set_itr = 1;
2791 rx_ring->total_bytes = 0;
2792 rx_ring->total_packets = 0;
2796 * igb_update_itr - update the dynamic ITR value based on statistics
2797 * Stores a new ITR value based on packets and byte
2798 * counts during the last interrupt. The advantage of per interrupt
2799 * computation is faster updates and more accurate ITR for the current
2800 * traffic pattern. Constants in this function were computed
2801 * based on theoretical maximum wire speed and thresholds were set based
2802 * on testing data as well as attempting to minimize response time
2803 * while increasing bulk throughput.
2804 * this functionality is controlled by the InterruptThrottleRate module
2805 * parameter (see igb_param.c)
2806 * NOTE: These calculations are only valid when operating in a single-
2807 * queue environment.
2808 * @adapter: pointer to adapter
2809 * @itr_setting: current adapter->itr
2810 * @packets: the number of packets during this measurement interval
2811 * @bytes: the number of bytes during this measurement interval
2813 static unsigned int igb_update_itr(struct igb_adapter *adapter, u16 itr_setting,
2814 int packets, int bytes)
2816 unsigned int retval = itr_setting;
2819 goto update_itr_done;
2821 switch (itr_setting) {
2822 case lowest_latency:
2823 /* handle TSO and jumbo frames */
2824 if (bytes/packets > 8000)
2825 retval = bulk_latency;
2826 else if ((packets < 5) && (bytes > 512))
2827 retval = low_latency;
2829 case low_latency: /* 50 usec aka 20000 ints/s */
2830 if (bytes > 10000) {
2831 /* this if handles the TSO accounting */
2832 if (bytes/packets > 8000) {
2833 retval = bulk_latency;
2834 } else if ((packets < 10) || ((bytes/packets) > 1200)) {
2835 retval = bulk_latency;
2836 } else if ((packets > 35)) {
2837 retval = lowest_latency;
2839 } else if (bytes/packets > 2000) {
2840 retval = bulk_latency;
2841 } else if (packets <= 2 && bytes < 512) {
2842 retval = lowest_latency;
2845 case bulk_latency: /* 250 usec aka 4000 ints/s */
2846 if (bytes > 25000) {
2848 retval = low_latency;
2849 } else if (bytes < 1500) {
2850 retval = low_latency;
2859 static void igb_set_itr(struct igb_adapter *adapter)
2862 u32 new_itr = adapter->itr;
2864 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2865 if (adapter->link_speed != SPEED_1000) {
2871 adapter->rx_itr = igb_update_itr(adapter,
2873 adapter->rx_ring->total_packets,
2874 adapter->rx_ring->total_bytes);
2876 if (adapter->rx_ring->buddy) {
2877 adapter->tx_itr = igb_update_itr(adapter,
2879 adapter->tx_ring->total_packets,
2880 adapter->tx_ring->total_bytes);
2881 current_itr = max(adapter->rx_itr, adapter->tx_itr);
2883 current_itr = adapter->rx_itr;
2886 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2887 if (adapter->itr_setting == 3 && current_itr == lowest_latency)
2888 current_itr = low_latency;
2890 switch (current_itr) {
2891 /* counts and packets in update_itr are dependent on these numbers */
2892 case lowest_latency:
2893 new_itr = 56; /* aka 70,000 ints/sec */
2896 new_itr = 196; /* aka 20,000 ints/sec */
2899 new_itr = 980; /* aka 4,000 ints/sec */
2906 adapter->rx_ring->total_bytes = 0;
2907 adapter->rx_ring->total_packets = 0;
2908 if (adapter->rx_ring->buddy) {
2909 adapter->rx_ring->buddy->total_bytes = 0;
2910 adapter->rx_ring->buddy->total_packets = 0;
2913 if (new_itr != adapter->itr) {
2914 /* this attempts to bias the interrupt rate towards Bulk
2915 * by adding intermediate steps when interrupt rate is
2917 new_itr = new_itr > adapter->itr ?
2918 max((new_itr * adapter->itr) /
2919 (new_itr + (adapter->itr >> 2)), new_itr) :
2921 /* Don't write the value here; it resets the adapter's
2922 * internal timer, and causes us to delay far longer than
2923 * we should between interrupts. Instead, we write the ITR
2924 * value at the beginning of the next interrupt so the timing
2925 * ends up being correct.
2927 adapter->itr = new_itr;
2928 adapter->rx_ring->itr_val = new_itr;
2929 adapter->rx_ring->set_itr = 1;
2936 #define IGB_TX_FLAGS_CSUM 0x00000001
2937 #define IGB_TX_FLAGS_VLAN 0x00000002
2938 #define IGB_TX_FLAGS_TSO 0x00000004
2939 #define IGB_TX_FLAGS_IPV4 0x00000008
2940 #define IGB_TX_FLAGS_TSTAMP 0x00000010
2941 #define IGB_TX_FLAGS_VLAN_MASK 0xffff0000
2942 #define IGB_TX_FLAGS_VLAN_SHIFT 16
2944 static inline int igb_tso_adv(struct igb_adapter *adapter,
2945 struct igb_ring *tx_ring,
2946 struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
2948 struct e1000_adv_tx_context_desc *context_desc;
2951 struct igb_buffer *buffer_info;
2952 u32 info = 0, tu_cmd = 0;
2953 u32 mss_l4len_idx, l4len;
2956 if (skb_header_cloned(skb)) {
2957 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2962 l4len = tcp_hdrlen(skb);
2965 if (skb->protocol == htons(ETH_P_IP)) {
2966 struct iphdr *iph = ip_hdr(skb);
2969 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2973 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
2974 ipv6_hdr(skb)->payload_len = 0;
2975 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2976 &ipv6_hdr(skb)->daddr,
2980 i = tx_ring->next_to_use;
2982 buffer_info = &tx_ring->buffer_info[i];
2983 context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
2984 /* VLAN MACLEN IPLEN */
2985 if (tx_flags & IGB_TX_FLAGS_VLAN)
2986 info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
2987 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2988 *hdr_len += skb_network_offset(skb);
2989 info |= skb_network_header_len(skb);
2990 *hdr_len += skb_network_header_len(skb);
2991 context_desc->vlan_macip_lens = cpu_to_le32(info);
2993 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
2994 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2996 if (skb->protocol == htons(ETH_P_IP))
2997 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2998 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
3000 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
3003 mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
3004 mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
3006 /* For 82575, context index must be unique per ring. */
3007 if (adapter->flags & IGB_FLAG_NEED_CTX_IDX)
3008 mss_l4len_idx |= tx_ring->queue_index << 4;
3010 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
3011 context_desc->seqnum_seed = 0;
3013 buffer_info->time_stamp = jiffies;
3014 buffer_info->next_to_watch = i;
3015 buffer_info->dma = 0;
3017 if (i == tx_ring->count)
3020 tx_ring->next_to_use = i;
3025 static inline bool igb_tx_csum_adv(struct igb_adapter *adapter,
3026 struct igb_ring *tx_ring,
3027 struct sk_buff *skb, u32 tx_flags)
3029 struct e1000_adv_tx_context_desc *context_desc;
3031 struct igb_buffer *buffer_info;
3032 u32 info = 0, tu_cmd = 0;
3034 if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
3035 (tx_flags & IGB_TX_FLAGS_VLAN)) {
3036 i = tx_ring->next_to_use;
3037 buffer_info = &tx_ring->buffer_info[i];
3038 context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
3040 if (tx_flags & IGB_TX_FLAGS_VLAN)
3041 info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
3042 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
3043 if (skb->ip_summed == CHECKSUM_PARTIAL)
3044 info |= skb_network_header_len(skb);
3046 context_desc->vlan_macip_lens = cpu_to_le32(info);
3048 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
3050 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3053 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3054 const struct vlan_ethhdr *vhdr =
3055 (const struct vlan_ethhdr*)skb->data;
3057 protocol = vhdr->h_vlan_encapsulated_proto;
3059 protocol = skb->protocol;
3063 case cpu_to_be16(ETH_P_IP):
3064 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
3065 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
3066 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
3068 case cpu_to_be16(ETH_P_IPV6):
3069 /* XXX what about other V6 headers?? */
3070 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3071 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
3074 if (unlikely(net_ratelimit()))
3075 dev_warn(&adapter->pdev->dev,
3076 "partial checksum but proto=%x!\n",
3082 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
3083 context_desc->seqnum_seed = 0;
3084 if (adapter->flags & IGB_FLAG_NEED_CTX_IDX)
3085 context_desc->mss_l4len_idx =
3086 cpu_to_le32(tx_ring->queue_index << 4);
3088 context_desc->mss_l4len_idx = 0;
3090 buffer_info->time_stamp = jiffies;
3091 buffer_info->next_to_watch = i;
3092 buffer_info->dma = 0;
3095 if (i == tx_ring->count)
3097 tx_ring->next_to_use = i;
3104 #define IGB_MAX_TXD_PWR 16
3105 #define IGB_MAX_DATA_PER_TXD (1<<IGB_MAX_TXD_PWR)
3107 static inline int igb_tx_map_adv(struct igb_adapter *adapter,
3108 struct igb_ring *tx_ring, struct sk_buff *skb,
3111 struct igb_buffer *buffer_info;
3112 unsigned int len = skb_headlen(skb);
3113 unsigned int count = 0, i;
3117 i = tx_ring->next_to_use;
3119 if (skb_dma_map(&adapter->pdev->dev, skb, DMA_TO_DEVICE)) {
3120 dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
3124 map = skb_shinfo(skb)->dma_maps;
3126 buffer_info = &tx_ring->buffer_info[i];
3127 BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
3128 buffer_info->length = len;
3129 /* set time_stamp *before* dma to help avoid a possible race */
3130 buffer_info->time_stamp = jiffies;
3131 buffer_info->next_to_watch = i;
3132 buffer_info->dma = map[count];
3135 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
3136 struct skb_frag_struct *frag;
3139 if (i == tx_ring->count)
3142 frag = &skb_shinfo(skb)->frags[f];
3145 buffer_info = &tx_ring->buffer_info[i];
3146 BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
3147 buffer_info->length = len;
3148 buffer_info->time_stamp = jiffies;
3149 buffer_info->next_to_watch = i;
3150 buffer_info->dma = map[count];
3154 tx_ring->buffer_info[i].skb = skb;
3155 tx_ring->buffer_info[first].next_to_watch = i;
3160 static inline void igb_tx_queue_adv(struct igb_adapter *adapter,
3161 struct igb_ring *tx_ring,
3162 int tx_flags, int count, u32 paylen,
3165 union e1000_adv_tx_desc *tx_desc = NULL;
3166 struct igb_buffer *buffer_info;
3167 u32 olinfo_status = 0, cmd_type_len;
3170 cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
3171 E1000_ADVTXD_DCMD_DEXT);
3173 if (tx_flags & IGB_TX_FLAGS_VLAN)
3174 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
3176 if (tx_flags & IGB_TX_FLAGS_TSTAMP)
3177 cmd_type_len |= E1000_ADVTXD_MAC_TSTAMP;
3179 if (tx_flags & IGB_TX_FLAGS_TSO) {
3180 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
3182 /* insert tcp checksum */
3183 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
3185 /* insert ip checksum */
3186 if (tx_flags & IGB_TX_FLAGS_IPV4)
3187 olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
3189 } else if (tx_flags & IGB_TX_FLAGS_CSUM) {
3190 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
3193 if ((adapter->flags & IGB_FLAG_NEED_CTX_IDX) &&
3194 (tx_flags & (IGB_TX_FLAGS_CSUM | IGB_TX_FLAGS_TSO |
3195 IGB_TX_FLAGS_VLAN)))
3196 olinfo_status |= tx_ring->queue_index << 4;
3198 olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
3200 i = tx_ring->next_to_use;
3202 buffer_info = &tx_ring->buffer_info[i];
3203 tx_desc = E1000_TX_DESC_ADV(*tx_ring, i);
3204 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
3205 tx_desc->read.cmd_type_len =
3206 cpu_to_le32(cmd_type_len | buffer_info->length);
3207 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
3209 if (i == tx_ring->count)
3213 tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
3214 /* Force memory writes to complete before letting h/w
3215 * know there are new descriptors to fetch. (Only
3216 * applicable for weak-ordered memory model archs,
3217 * such as IA-64). */
3220 tx_ring->next_to_use = i;
3221 writel(i, adapter->hw.hw_addr + tx_ring->tail);
3222 /* we need this if more than one processor can write to our tail
3223 * at a time, it syncronizes IO on IA64/Altix systems */
3227 static int __igb_maybe_stop_tx(struct net_device *netdev,
3228 struct igb_ring *tx_ring, int size)
3230 struct igb_adapter *adapter = netdev_priv(netdev);
3232 netif_stop_subqueue(netdev, tx_ring->queue_index);
3234 /* Herbert's original patch had:
3235 * smp_mb__after_netif_stop_queue();
3236 * but since that doesn't exist yet, just open code it. */
3239 /* We need to check again in a case another CPU has just
3240 * made room available. */
3241 if (igb_desc_unused(tx_ring) < size)
3245 netif_wake_subqueue(netdev, tx_ring->queue_index);
3246 ++adapter->restart_queue;
3250 static int igb_maybe_stop_tx(struct net_device *netdev,
3251 struct igb_ring *tx_ring, int size)
3253 if (igb_desc_unused(tx_ring) >= size)
3255 return __igb_maybe_stop_tx(netdev, tx_ring, size);
3258 static int igb_xmit_frame_ring_adv(struct sk_buff *skb,
3259 struct net_device *netdev,
3260 struct igb_ring *tx_ring)
3262 struct igb_adapter *adapter = netdev_priv(netdev);
3264 unsigned int tx_flags = 0;
3268 union skb_shared_tx *shtx;
3270 if (test_bit(__IGB_DOWN, &adapter->state)) {
3271 dev_kfree_skb_any(skb);
3272 return NETDEV_TX_OK;
3275 if (skb->len <= 0) {
3276 dev_kfree_skb_any(skb);
3277 return NETDEV_TX_OK;
3280 /* need: 1 descriptor per page,
3281 * + 2 desc gap to keep tail from touching head,
3282 * + 1 desc for skb->data,
3283 * + 1 desc for context descriptor,
3284 * otherwise try next time */
3285 if (igb_maybe_stop_tx(netdev, tx_ring, skb_shinfo(skb)->nr_frags + 4)) {
3286 /* this is a hard error */
3287 return NETDEV_TX_BUSY;
3291 * TODO: check that there currently is no other packet with
3292 * time stamping in the queue
3294 * When doing time stamping, keep the connection to the socket
3295 * a while longer: it is still needed by skb_hwtstamp_tx(),
3296 * called either in igb_tx_hwtstamp() or by our caller when
3297 * doing software time stamping.
3300 if (unlikely(shtx->hardware)) {
3301 shtx->in_progress = 1;
3302 tx_flags |= IGB_TX_FLAGS_TSTAMP;
3305 if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
3306 tx_flags |= IGB_TX_FLAGS_VLAN;
3307 tx_flags |= (vlan_tx_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
3310 if (skb->protocol == htons(ETH_P_IP))
3311 tx_flags |= IGB_TX_FLAGS_IPV4;
3313 first = tx_ring->next_to_use;
3314 tso = skb_is_gso(skb) ? igb_tso_adv(adapter, tx_ring, skb, tx_flags,
3318 dev_kfree_skb_any(skb);
3319 return NETDEV_TX_OK;
3323 tx_flags |= IGB_TX_FLAGS_TSO;
3324 else if (igb_tx_csum_adv(adapter, tx_ring, skb, tx_flags) &&
3325 (skb->ip_summed == CHECKSUM_PARTIAL))
3326 tx_flags |= IGB_TX_FLAGS_CSUM;
3329 * count reflects descriptors mapped, if 0 then mapping error
3330 * has occured and we need to rewind the descriptor queue
3332 count = igb_tx_map_adv(adapter, tx_ring, skb, first);
3335 igb_tx_queue_adv(adapter, tx_ring, tx_flags, count,
3337 netdev->trans_start = jiffies;
3338 /* Make sure there is space in the ring for the next send. */
3339 igb_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 4);
3341 dev_kfree_skb_any(skb);
3342 tx_ring->buffer_info[first].time_stamp = 0;
3343 tx_ring->next_to_use = first;
3346 return NETDEV_TX_OK;
3349 static int igb_xmit_frame_adv(struct sk_buff *skb, struct net_device *netdev)
3351 struct igb_adapter *adapter = netdev_priv(netdev);
3352 struct igb_ring *tx_ring;
3355 r_idx = skb->queue_mapping & (IGB_ABS_MAX_TX_QUEUES - 1);
3356 tx_ring = adapter->multi_tx_table[r_idx];
3358 /* This goes back to the question of how to logically map a tx queue
3359 * to a flow. Right now, performance is impacted slightly negatively
3360 * if using multiple tx queues. If the stack breaks away from a
3361 * single qdisc implementation, we can look at this again. */
3362 return (igb_xmit_frame_ring_adv(skb, netdev, tx_ring));
3366 * igb_tx_timeout - Respond to a Tx Hang
3367 * @netdev: network interface device structure
3369 static void igb_tx_timeout(struct net_device *netdev)
3371 struct igb_adapter *adapter = netdev_priv(netdev);
3372 struct e1000_hw *hw = &adapter->hw;
3374 /* Do the reset outside of interrupt context */
3375 adapter->tx_timeout_count++;
3376 schedule_work(&adapter->reset_task);
3378 (adapter->eims_enable_mask & ~adapter->eims_other));
3381 static void igb_reset_task(struct work_struct *work)
3383 struct igb_adapter *adapter;
3384 adapter = container_of(work, struct igb_adapter, reset_task);
3386 igb_reinit_locked(adapter);
3390 * igb_get_stats - Get System Network Statistics
3391 * @netdev: network interface device structure
3393 * Returns the address of the device statistics structure.
3394 * The statistics are actually updated from the timer callback.
3396 static struct net_device_stats *igb_get_stats(struct net_device *netdev)
3398 struct igb_adapter *adapter = netdev_priv(netdev);
3400 /* only return the current stats */
3401 return &adapter->net_stats;
3405 * igb_change_mtu - Change the Maximum Transfer Unit
3406 * @netdev: network interface device structure
3407 * @new_mtu: new value for maximum frame size
3409 * Returns 0 on success, negative on failure
3411 static int igb_change_mtu(struct net_device *netdev, int new_mtu)
3413 struct igb_adapter *adapter = netdev_priv(netdev);
3414 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3416 if ((max_frame < ETH_ZLEN + ETH_FCS_LEN) ||
3417 (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3418 dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
3422 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
3423 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
3427 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
3430 /* igb_down has a dependency on max_frame_size */
3431 adapter->max_frame_size = max_frame;
3432 if (netif_running(netdev))
3435 /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3436 * means we reserve 2 more, this pushes us to allocate from the next
3438 * i.e. RXBUFFER_2048 --> size-4096 slab
3441 if (max_frame <= IGB_RXBUFFER_256)
3442 adapter->rx_buffer_len = IGB_RXBUFFER_256;
3443 else if (max_frame <= IGB_RXBUFFER_512)
3444 adapter->rx_buffer_len = IGB_RXBUFFER_512;
3445 else if (max_frame <= IGB_RXBUFFER_1024)
3446 adapter->rx_buffer_len = IGB_RXBUFFER_1024;
3447 else if (max_frame <= IGB_RXBUFFER_2048)
3448 adapter->rx_buffer_len = IGB_RXBUFFER_2048;
3450 #if (PAGE_SIZE / 2) > IGB_RXBUFFER_16384
3451 adapter->rx_buffer_len = IGB_RXBUFFER_16384;
3453 adapter->rx_buffer_len = PAGE_SIZE / 2;
3456 /* if sr-iov is enabled we need to force buffer size to 1K or larger */
3457 if (adapter->vfs_allocated_count &&
3458 (adapter->rx_buffer_len < IGB_RXBUFFER_1024))
3459 adapter->rx_buffer_len = IGB_RXBUFFER_1024;
3461 /* adjust allocation if LPE protects us, and we aren't using SBP */
3462 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
3463 (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE))
3464 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3466 dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
3467 netdev->mtu, new_mtu);
3468 netdev->mtu = new_mtu;
3470 if (netif_running(netdev))
3475 clear_bit(__IGB_RESETTING, &adapter->state);
3481 * igb_update_stats - Update the board statistics counters
3482 * @adapter: board private structure
3485 void igb_update_stats(struct igb_adapter *adapter)
3487 struct e1000_hw *hw = &adapter->hw;
3488 struct pci_dev *pdev = adapter->pdev;
3491 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3494 * Prevent stats update while adapter is being reset, or if the pci
3495 * connection is down.
3497 if (adapter->link_speed == 0)
3499 if (pci_channel_offline(pdev))
3502 adapter->stats.crcerrs += rd32(E1000_CRCERRS);
3503 adapter->stats.gprc += rd32(E1000_GPRC);
3504 adapter->stats.gorc += rd32(E1000_GORCL);
3505 rd32(E1000_GORCH); /* clear GORCL */
3506 adapter->stats.bprc += rd32(E1000_BPRC);
3507 adapter->stats.mprc += rd32(E1000_MPRC);
3508 adapter->stats.roc += rd32(E1000_ROC);
3510 adapter->stats.prc64 += rd32(E1000_PRC64);
3511 adapter->stats.prc127 += rd32(E1000_PRC127);
3512 adapter->stats.prc255 += rd32(E1000_PRC255);
3513 adapter->stats.prc511 += rd32(E1000_PRC511);
3514 adapter->stats.prc1023 += rd32(E1000_PRC1023);
3515 adapter->stats.prc1522 += rd32(E1000_PRC1522);
3516 adapter->stats.symerrs += rd32(E1000_SYMERRS);
3517 adapter->stats.sec += rd32(E1000_SEC);
3519 adapter->stats.mpc += rd32(E1000_MPC);
3520 adapter->stats.scc += rd32(E1000_SCC);
3521 adapter->stats.ecol += rd32(E1000_ECOL);
3522 adapter->stats.mcc += rd32(E1000_MCC);
3523 adapter->stats.latecol += rd32(E1000_LATECOL);
3524 adapter->stats.dc += rd32(E1000_DC);
3525 adapter->stats.rlec += rd32(E1000_RLEC);
3526 adapter->stats.xonrxc += rd32(E1000_XONRXC);
3527 adapter->stats.xontxc += rd32(E1000_XONTXC);
3528 adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
3529 adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
3530 adapter->stats.fcruc += rd32(E1000_FCRUC);
3531 adapter->stats.gptc += rd32(E1000_GPTC);
3532 adapter->stats.gotc += rd32(E1000_GOTCL);
3533 rd32(E1000_GOTCH); /* clear GOTCL */
3534 adapter->stats.rnbc += rd32(E1000_RNBC);
3535 adapter->stats.ruc += rd32(E1000_RUC);
3536 adapter->stats.rfc += rd32(E1000_RFC);
3537 adapter->stats.rjc += rd32(E1000_RJC);
3538 adapter->stats.tor += rd32(E1000_TORH);
3539 adapter->stats.tot += rd32(E1000_TOTH);
3540 adapter->stats.tpr += rd32(E1000_TPR);
3542 adapter->stats.ptc64 += rd32(E1000_PTC64);
3543 adapter->stats.ptc127 += rd32(E1000_PTC127);
3544 adapter->stats.ptc255 += rd32(E1000_PTC255);
3545 adapter->stats.ptc511 += rd32(E1000_PTC511);
3546 adapter->stats.ptc1023 += rd32(E1000_PTC1023);
3547 adapter->stats.ptc1522 += rd32(E1000_PTC1522);
3549 adapter->stats.mptc += rd32(E1000_MPTC);
3550 adapter->stats.bptc += rd32(E1000_BPTC);
3552 /* used for adaptive IFS */
3554 hw->mac.tx_packet_delta = rd32(E1000_TPT);
3555 adapter->stats.tpt += hw->mac.tx_packet_delta;
3556 hw->mac.collision_delta = rd32(E1000_COLC);
3557 adapter->stats.colc += hw->mac.collision_delta;
3559 adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
3560 adapter->stats.rxerrc += rd32(E1000_RXERRC);
3561 adapter->stats.tncrs += rd32(E1000_TNCRS);
3562 adapter->stats.tsctc += rd32(E1000_TSCTC);
3563 adapter->stats.tsctfc += rd32(E1000_TSCTFC);
3565 adapter->stats.iac += rd32(E1000_IAC);
3566 adapter->stats.icrxoc += rd32(E1000_ICRXOC);
3567 adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
3568 adapter->stats.icrxatc += rd32(E1000_ICRXATC);
3569 adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
3570 adapter->stats.ictxatc += rd32(E1000_ICTXATC);
3571 adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
3572 adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
3573 adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
3575 /* Fill out the OS statistics structure */
3576 adapter->net_stats.multicast = adapter->stats.mprc;
3577 adapter->net_stats.collisions = adapter->stats.colc;
3581 /* RLEC on some newer hardware can be incorrect so build
3582 * our own version based on RUC and ROC */
3583 adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3584 adapter->stats.crcerrs + adapter->stats.algnerrc +
3585 adapter->stats.ruc + adapter->stats.roc +
3586 adapter->stats.cexterr;
3587 adapter->net_stats.rx_length_errors = adapter->stats.ruc +
3589 adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3590 adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3591 adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3594 adapter->net_stats.tx_errors = adapter->stats.ecol +
3595 adapter->stats.latecol;
3596 adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3597 adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3598 adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3600 /* Tx Dropped needs to be maintained elsewhere */
3603 if (hw->phy.media_type == e1000_media_type_copper) {
3604 if ((adapter->link_speed == SPEED_1000) &&
3605 (!igb_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
3606 phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3607 adapter->phy_stats.idle_errors += phy_tmp;
3611 /* Management Stats */
3612 adapter->stats.mgptc += rd32(E1000_MGTPTC);
3613 adapter->stats.mgprc += rd32(E1000_MGTPRC);
3614 adapter->stats.mgpdc += rd32(E1000_MGTPDC);
3617 static irqreturn_t igb_msix_other(int irq, void *data)
3619 struct net_device *netdev = data;
3620 struct igb_adapter *adapter = netdev_priv(netdev);
3621 struct e1000_hw *hw = &adapter->hw;
3622 u32 icr = rd32(E1000_ICR);
3624 /* reading ICR causes bit 31 of EICR to be cleared */
3626 if(icr & E1000_ICR_DOUTSYNC) {
3627 /* HW is reporting DMA is out of sync */
3628 adapter->stats.doosync++;
3631 /* Check for a mailbox event */
3632 if (icr & E1000_ICR_VMMB)
3633 igb_msg_task(adapter);
3635 if (icr & E1000_ICR_LSC) {
3636 hw->mac.get_link_status = 1;
3637 /* guard against interrupt when we're going down */
3638 if (!test_bit(__IGB_DOWN, &adapter->state))
3639 mod_timer(&adapter->watchdog_timer, jiffies + 1);
3642 wr32(E1000_IMS, E1000_IMS_LSC | E1000_IMS_DOUTSYNC | E1000_IMS_VMMB);
3643 wr32(E1000_EIMS, adapter->eims_other);
3648 static irqreturn_t igb_msix_tx(int irq, void *data)
3650 struct igb_ring *tx_ring = data;
3651 struct igb_adapter *adapter = tx_ring->adapter;
3652 struct e1000_hw *hw = &adapter->hw;
3654 #ifdef CONFIG_IGB_DCA
3655 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
3656 igb_update_tx_dca(tx_ring);
3659 tx_ring->total_bytes = 0;
3660 tx_ring->total_packets = 0;
3662 /* auto mask will automatically reenable the interrupt when we write
3664 if (!igb_clean_tx_irq(tx_ring))
3665 /* Ring was not completely cleaned, so fire another interrupt */
3666 wr32(E1000_EICS, tx_ring->eims_value);
3668 wr32(E1000_EIMS, tx_ring->eims_value);
3673 static void igb_write_itr(struct igb_ring *ring)
3675 struct e1000_hw *hw = &ring->adapter->hw;
3676 if ((ring->adapter->itr_setting & 3) && ring->set_itr) {
3677 switch (hw->mac.type) {
3679 wr32(ring->itr_register, ring->itr_val |
3683 wr32(ring->itr_register, ring->itr_val |
3684 (ring->itr_val << 16));
3691 static irqreturn_t igb_msix_rx(int irq, void *data)
3693 struct igb_ring *rx_ring = data;
3695 /* Write the ITR value calculated at the end of the
3696 * previous interrupt.
3699 igb_write_itr(rx_ring);
3701 if (napi_schedule_prep(&rx_ring->napi))
3702 __napi_schedule(&rx_ring->napi);
3704 #ifdef CONFIG_IGB_DCA
3705 if (rx_ring->adapter->flags & IGB_FLAG_DCA_ENABLED)
3706 igb_update_rx_dca(rx_ring);
3711 #ifdef CONFIG_IGB_DCA
3712 static void igb_update_rx_dca(struct igb_ring *rx_ring)
3715 struct igb_adapter *adapter = rx_ring->adapter;
3716 struct e1000_hw *hw = &adapter->hw;
3717 int cpu = get_cpu();
3718 int q = rx_ring->reg_idx;
3720 if (rx_ring->cpu != cpu) {
3721 dca_rxctrl = rd32(E1000_DCA_RXCTRL(q));
3722 if (hw->mac.type == e1000_82576) {
3723 dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK_82576;
3724 dca_rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu) <<
3725 E1000_DCA_RXCTRL_CPUID_SHIFT;
3727 dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK;
3728 dca_rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
3730 dca_rxctrl |= E1000_DCA_RXCTRL_DESC_DCA_EN;
3731 dca_rxctrl |= E1000_DCA_RXCTRL_HEAD_DCA_EN;
3732 dca_rxctrl |= E1000_DCA_RXCTRL_DATA_DCA_EN;
3733 wr32(E1000_DCA_RXCTRL(q), dca_rxctrl);
3739 static void igb_update_tx_dca(struct igb_ring *tx_ring)
3742 struct igb_adapter *adapter = tx_ring->adapter;
3743 struct e1000_hw *hw = &adapter->hw;
3744 int cpu = get_cpu();
3745 int q = tx_ring->reg_idx;
3747 if (tx_ring->cpu != cpu) {
3748 dca_txctrl = rd32(E1000_DCA_TXCTRL(q));
3749 if (hw->mac.type == e1000_82576) {
3750 dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK_82576;
3751 dca_txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu) <<
3752 E1000_DCA_TXCTRL_CPUID_SHIFT;
3754 dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK;
3755 dca_txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
3757 dca_txctrl |= E1000_DCA_TXCTRL_DESC_DCA_EN;
3758 wr32(E1000_DCA_TXCTRL(q), dca_txctrl);
3764 static void igb_setup_dca(struct igb_adapter *adapter)
3768 if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
3771 for (i = 0; i < adapter->num_tx_queues; i++) {
3772 adapter->tx_ring[i].cpu = -1;
3773 igb_update_tx_dca(&adapter->tx_ring[i]);
3775 for (i = 0; i < adapter->num_rx_queues; i++) {
3776 adapter->rx_ring[i].cpu = -1;
3777 igb_update_rx_dca(&adapter->rx_ring[i]);
3781 static int __igb_notify_dca(struct device *dev, void *data)
3783 struct net_device *netdev = dev_get_drvdata(dev);
3784 struct igb_adapter *adapter = netdev_priv(netdev);
3785 struct e1000_hw *hw = &adapter->hw;
3786 unsigned long event = *(unsigned long *)data;
3789 case DCA_PROVIDER_ADD:
3790 /* if already enabled, don't do it again */
3791 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
3793 /* Always use CB2 mode, difference is masked
3794 * in the CB driver. */
3795 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
3796 if (dca_add_requester(dev) == 0) {
3797 adapter->flags |= IGB_FLAG_DCA_ENABLED;
3798 dev_info(&adapter->pdev->dev, "DCA enabled\n");
3799 igb_setup_dca(adapter);
3802 /* Fall Through since DCA is disabled. */
3803 case DCA_PROVIDER_REMOVE:
3804 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
3805 /* without this a class_device is left
3806 * hanging around in the sysfs model */
3807 dca_remove_requester(dev);
3808 dev_info(&adapter->pdev->dev, "DCA disabled\n");
3809 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
3810 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
3818 static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
3823 ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
3826 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
3828 #endif /* CONFIG_IGB_DCA */
3830 static void igb_ping_all_vfs(struct igb_adapter *adapter)
3832 struct e1000_hw *hw = &adapter->hw;
3836 for (i = 0 ; i < adapter->vfs_allocated_count; i++) {
3837 ping = E1000_PF_CONTROL_MSG;
3838 if (adapter->vf_data[i].clear_to_send)
3839 ping |= E1000_VT_MSGTYPE_CTS;
3840 igb_write_mbx(hw, &ping, 1, i);
3844 static int igb_set_vf_multicasts(struct igb_adapter *adapter,
3845 u32 *msgbuf, u32 vf)
3847 int n = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
3848 u16 *hash_list = (u16 *)&msgbuf[1];
3849 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
3852 /* only up to 30 hash values supported */
3856 /* salt away the number of multi cast addresses assigned
3857 * to this VF for later use to restore when the PF multi cast
3860 vf_data->num_vf_mc_hashes = n;
3862 /* VFs are limited to using the MTA hash table for their multicast
3864 for (i = 0; i < n; i++)
3865 vf_data->vf_mc_hashes[i] = hash_list[i];;
3867 /* Flush and reset the mta with the new values */
3868 igb_set_multi(adapter->netdev);
3873 static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
3875 struct e1000_hw *hw = &adapter->hw;
3876 struct vf_data_storage *vf_data;
3879 for (i = 0; i < adapter->vfs_allocated_count; i++) {
3880 vf_data = &adapter->vf_data[i];
3881 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
3882 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
3886 static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
3888 struct e1000_hw *hw = &adapter->hw;
3889 u32 pool_mask, reg, vid;
3892 pool_mask = 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
3894 /* Find the vlan filter for this id */
3895 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
3896 reg = rd32(E1000_VLVF(i));
3898 /* remove the vf from the pool */
3901 /* if pool is empty then remove entry from vfta */
3902 if (!(reg & E1000_VLVF_POOLSEL_MASK) &&
3903 (reg & E1000_VLVF_VLANID_ENABLE)) {
3905 vid = reg & E1000_VLVF_VLANID_MASK;
3906 igb_vfta_set(hw, vid, false);
3909 wr32(E1000_VLVF(i), reg);
3913 static s32 igb_vlvf_set(struct igb_adapter *adapter, u32 vid, bool add, u32 vf)
3915 struct e1000_hw *hw = &adapter->hw;
3918 /* It is an error to call this function when VFs are not enabled */
3919 if (!adapter->vfs_allocated_count)
3922 /* Find the vlan filter for this id */
3923 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
3924 reg = rd32(E1000_VLVF(i));
3925 if ((reg & E1000_VLVF_VLANID_ENABLE) &&
3926 vid == (reg & E1000_VLVF_VLANID_MASK))
3931 if (i == E1000_VLVF_ARRAY_SIZE) {
3932 /* Did not find a matching VLAN ID entry that was
3933 * enabled. Search for a free filter entry, i.e.
3934 * one without the enable bit set
3936 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
3937 reg = rd32(E1000_VLVF(i));
3938 if (!(reg & E1000_VLVF_VLANID_ENABLE))
3942 if (i < E1000_VLVF_ARRAY_SIZE) {
3943 /* Found an enabled/available entry */
3944 reg |= 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
3946 /* if !enabled we need to set this up in vfta */
3947 if (!(reg & E1000_VLVF_VLANID_ENABLE)) {
3948 /* add VID to filter table, if bit already set
3949 * PF must have added it outside of table */
3950 if (igb_vfta_set(hw, vid, true))
3951 reg |= 1 << (E1000_VLVF_POOLSEL_SHIFT +
3952 adapter->vfs_allocated_count);
3953 reg |= E1000_VLVF_VLANID_ENABLE;
3955 reg &= ~E1000_VLVF_VLANID_MASK;
3958 wr32(E1000_VLVF(i), reg);
3962 if (i < E1000_VLVF_ARRAY_SIZE) {
3963 /* remove vf from the pool */
3964 reg &= ~(1 << (E1000_VLVF_POOLSEL_SHIFT + vf));
3965 /* if pool is empty then remove entry from vfta */
3966 if (!(reg & E1000_VLVF_POOLSEL_MASK)) {
3968 igb_vfta_set(hw, vid, false);
3970 wr32(E1000_VLVF(i), reg);
3977 static int igb_set_vf_vlan(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
3979 int add = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
3980 int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
3982 return igb_vlvf_set(adapter, vid, add, vf);
3985 static inline void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
3987 struct e1000_hw *hw = &adapter->hw;
3989 /* disable mailbox functionality for vf */
3990 adapter->vf_data[vf].clear_to_send = false;
3992 /* reset offloads to defaults */
3993 igb_set_vmolr(hw, vf);
3995 /* reset vlans for device */
3996 igb_clear_vf_vfta(adapter, vf);
3998 /* reset multicast table array for vf */
3999 adapter->vf_data[vf].num_vf_mc_hashes = 0;
4001 /* Flush and reset the mta with the new values */
4002 igb_set_multi(adapter->netdev);
4005 static inline void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
4007 struct e1000_hw *hw = &adapter->hw;
4008 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
4010 u8 *addr = (u8 *)(&msgbuf[1]);
4012 /* process all the same items cleared in a function level reset */
4013 igb_vf_reset_event(adapter, vf);
4015 /* set vf mac address */
4016 igb_rar_set(hw, vf_mac, vf + 1);
4017 igb_set_rah_pool(hw, vf, vf + 1);
4019 /* enable transmit and receive for vf */
4020 reg = rd32(E1000_VFTE);
4021 wr32(E1000_VFTE, reg | (1 << vf));
4022 reg = rd32(E1000_VFRE);
4023 wr32(E1000_VFRE, reg | (1 << vf));
4025 /* enable mailbox functionality for vf */
4026 adapter->vf_data[vf].clear_to_send = true;
4028 /* reply to reset with ack and vf mac address */
4029 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
4030 memcpy(addr, vf_mac, 6);
4031 igb_write_mbx(hw, msgbuf, 3, vf);
4034 static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf)
4036 unsigned char *addr = (char *)&msg[1];
4039 if (is_valid_ether_addr(addr))
4040 err = igb_set_vf_mac(adapter, vf, addr);
4046 static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf)
4048 struct e1000_hw *hw = &adapter->hw;
4049 u32 msg = E1000_VT_MSGTYPE_NACK;
4051 /* if device isn't clear to send it shouldn't be reading either */
4052 if (!adapter->vf_data[vf].clear_to_send)
4053 igb_write_mbx(hw, &msg, 1, vf);
4057 static void igb_msg_task(struct igb_adapter *adapter)
4059 struct e1000_hw *hw = &adapter->hw;
4062 for (vf = 0; vf < adapter->vfs_allocated_count; vf++) {
4063 /* process any reset requests */
4064 if (!igb_check_for_rst(hw, vf)) {
4065 adapter->vf_data[vf].clear_to_send = false;
4066 igb_vf_reset_event(adapter, vf);
4069 /* process any messages pending */
4070 if (!igb_check_for_msg(hw, vf))
4071 igb_rcv_msg_from_vf(adapter, vf);
4073 /* process any acks */
4074 if (!igb_check_for_ack(hw, vf))
4075 igb_rcv_ack_from_vf(adapter, vf);
4080 static int igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
4082 u32 mbx_size = E1000_VFMAILBOX_SIZE;
4083 u32 msgbuf[mbx_size];
4084 struct e1000_hw *hw = &adapter->hw;
4087 retval = igb_read_mbx(hw, msgbuf, mbx_size, vf);
4090 dev_err(&adapter->pdev->dev,
4091 "Error receiving message from VF\n");
4093 /* this is a message we already processed, do nothing */
4094 if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
4098 * until the vf completes a reset it should not be
4099 * allowed to start any configuration.
4102 if (msgbuf[0] == E1000_VF_RESET) {
4103 igb_vf_reset_msg(adapter, vf);
4108 if (!adapter->vf_data[vf].clear_to_send) {
4109 msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
4110 igb_write_mbx(hw, msgbuf, 1, vf);
4114 switch ((msgbuf[0] & 0xFFFF)) {
4115 case E1000_VF_SET_MAC_ADDR:
4116 retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
4118 case E1000_VF_SET_MULTICAST:
4119 retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
4121 case E1000_VF_SET_LPE:
4122 retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf);
4124 case E1000_VF_SET_VLAN:
4125 retval = igb_set_vf_vlan(adapter, msgbuf, vf);
4128 dev_err(&adapter->pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
4133 /* notify the VF of the results of what it sent us */
4135 msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
4137 msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
4139 msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
4141 igb_write_mbx(hw, msgbuf, 1, vf);
4147 * igb_intr_msi - Interrupt Handler
4148 * @irq: interrupt number
4149 * @data: pointer to a network interface device structure
4151 static irqreturn_t igb_intr_msi(int irq, void *data)
4153 struct net_device *netdev = data;
4154 struct igb_adapter *adapter = netdev_priv(netdev);
4155 struct e1000_hw *hw = &adapter->hw;
4156 /* read ICR disables interrupts using IAM */
4157 u32 icr = rd32(E1000_ICR);
4159 igb_write_itr(adapter->rx_ring);
4161 if(icr & E1000_ICR_DOUTSYNC) {
4162 /* HW is reporting DMA is out of sync */
4163 adapter->stats.doosync++;
4166 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
4167 hw->mac.get_link_status = 1;
4168 if (!test_bit(__IGB_DOWN, &adapter->state))
4169 mod_timer(&adapter->watchdog_timer, jiffies + 1);
4172 napi_schedule(&adapter->rx_ring[0].napi);
4178 * igb_intr - Legacy Interrupt Handler
4179 * @irq: interrupt number
4180 * @data: pointer to a network interface device structure
4182 static irqreturn_t igb_intr(int irq, void *data)
4184 struct net_device *netdev = data;
4185 struct igb_adapter *adapter = netdev_priv(netdev);
4186 struct e1000_hw *hw = &adapter->hw;
4187 /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked. No
4188 * need for the IMC write */
4189 u32 icr = rd32(E1000_ICR);
4191 return IRQ_NONE; /* Not our interrupt */
4193 igb_write_itr(adapter->rx_ring);
4195 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
4196 * not set, then the adapter didn't send an interrupt */
4197 if (!(icr & E1000_ICR_INT_ASSERTED))
4200 if(icr & E1000_ICR_DOUTSYNC) {
4201 /* HW is reporting DMA is out of sync */
4202 adapter->stats.doosync++;
4205 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
4206 hw->mac.get_link_status = 1;
4207 /* guard against interrupt when we're going down */
4208 if (!test_bit(__IGB_DOWN, &adapter->state))
4209 mod_timer(&adapter->watchdog_timer, jiffies + 1);
4212 napi_schedule(&adapter->rx_ring[0].napi);
4217 static inline void igb_rx_irq_enable(struct igb_ring *rx_ring)
4219 struct igb_adapter *adapter = rx_ring->adapter;
4220 struct e1000_hw *hw = &adapter->hw;
4222 if (adapter->itr_setting & 3) {
4223 if (adapter->num_rx_queues == 1)
4224 igb_set_itr(adapter);
4226 igb_update_ring_itr(rx_ring);
4229 if (!test_bit(__IGB_DOWN, &adapter->state)) {
4230 if (adapter->msix_entries)
4231 wr32(E1000_EIMS, rx_ring->eims_value);
4233 igb_irq_enable(adapter);
4238 * igb_poll - NAPI Rx polling callback
4239 * @napi: napi polling structure
4240 * @budget: count of how many packets we should handle
4242 static int igb_poll(struct napi_struct *napi, int budget)
4244 struct igb_ring *rx_ring = container_of(napi, struct igb_ring, napi);
4247 #ifdef CONFIG_IGB_DCA
4248 if (rx_ring->adapter->flags & IGB_FLAG_DCA_ENABLED)
4249 igb_update_rx_dca(rx_ring);
4251 igb_clean_rx_irq_adv(rx_ring, &work_done, budget);
4253 if (rx_ring->buddy) {
4254 #ifdef CONFIG_IGB_DCA
4255 if (rx_ring->adapter->flags & IGB_FLAG_DCA_ENABLED)
4256 igb_update_tx_dca(rx_ring->buddy);
4258 if (!igb_clean_tx_irq(rx_ring->buddy))
4262 /* If not enough Rx work done, exit the polling mode */
4263 if (work_done < budget) {
4264 napi_complete(napi);
4265 igb_rx_irq_enable(rx_ring);
4272 * igb_hwtstamp - utility function which checks for TX time stamp
4273 * @adapter: board private structure
4274 * @skb: packet that was just sent
4276 * If we were asked to do hardware stamping and such a time stamp is
4277 * available, then it must have been for this skb here because we only
4278 * allow only one such packet into the queue.
4280 static void igb_tx_hwtstamp(struct igb_adapter *adapter, struct sk_buff *skb)
4282 union skb_shared_tx *shtx = skb_tx(skb);
4283 struct e1000_hw *hw = &adapter->hw;
4285 if (unlikely(shtx->hardware)) {
4286 u32 valid = rd32(E1000_TSYNCTXCTL) & E1000_TSYNCTXCTL_VALID;
4288 u64 regval = rd32(E1000_TXSTMPL);
4290 struct skb_shared_hwtstamps shhwtstamps;
4292 memset(&shhwtstamps, 0, sizeof(shhwtstamps));
4293 regval |= (u64)rd32(E1000_TXSTMPH) << 32;
4294 ns = timecounter_cyc2time(&adapter->clock,
4296 timecompare_update(&adapter->compare, ns);
4297 shhwtstamps.hwtstamp = ns_to_ktime(ns);
4298 shhwtstamps.syststamp =
4299 timecompare_transform(&adapter->compare, ns);
4300 skb_tstamp_tx(skb, &shhwtstamps);
4306 * igb_clean_tx_irq - Reclaim resources after transmit completes
4307 * @adapter: board private structure
4308 * returns true if ring is completely cleaned
4310 static bool igb_clean_tx_irq(struct igb_ring *tx_ring)
4312 struct igb_adapter *adapter = tx_ring->adapter;
4313 struct net_device *netdev = adapter->netdev;
4314 struct e1000_hw *hw = &adapter->hw;
4315 struct igb_buffer *buffer_info;
4316 struct sk_buff *skb;
4317 union e1000_adv_tx_desc *tx_desc, *eop_desc;
4318 unsigned int total_bytes = 0, total_packets = 0;
4319 unsigned int i, eop, count = 0;
4320 bool cleaned = false;
4322 i = tx_ring->next_to_clean;
4323 eop = tx_ring->buffer_info[i].next_to_watch;
4324 eop_desc = E1000_TX_DESC_ADV(*tx_ring, eop);
4326 while ((eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)) &&
4327 (count < tx_ring->count)) {
4328 for (cleaned = false; !cleaned; count++) {
4329 tx_desc = E1000_TX_DESC_ADV(*tx_ring, i);
4330 buffer_info = &tx_ring->buffer_info[i];
4331 cleaned = (i == eop);
4332 skb = buffer_info->skb;
4335 unsigned int segs, bytecount;
4336 /* gso_segs is currently only valid for tcp */
4337 segs = skb_shinfo(skb)->gso_segs ?: 1;
4338 /* multiply data chunks by size of headers */
4339 bytecount = ((segs - 1) * skb_headlen(skb)) +
4341 total_packets += segs;
4342 total_bytes += bytecount;
4344 igb_tx_hwtstamp(adapter, skb);
4347 igb_unmap_and_free_tx_resource(adapter, buffer_info);
4348 tx_desc->wb.status = 0;
4351 if (i == tx_ring->count)
4354 eop = tx_ring->buffer_info[i].next_to_watch;
4355 eop_desc = E1000_TX_DESC_ADV(*tx_ring, eop);
4358 tx_ring->next_to_clean = i;
4360 if (unlikely(count &&
4361 netif_carrier_ok(netdev) &&
4362 igb_desc_unused(tx_ring) >= IGB_TX_QUEUE_WAKE)) {
4363 /* Make sure that anybody stopping the queue after this
4364 * sees the new next_to_clean.
4367 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
4368 !(test_bit(__IGB_DOWN, &adapter->state))) {
4369 netif_wake_subqueue(netdev, tx_ring->queue_index);
4370 ++adapter->restart_queue;
4374 if (tx_ring->detect_tx_hung) {
4375 /* Detect a transmit hang in hardware, this serializes the
4376 * check with the clearing of time_stamp and movement of i */
4377 tx_ring->detect_tx_hung = false;
4378 if (tx_ring->buffer_info[i].time_stamp &&
4379 time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
4380 (adapter->tx_timeout_factor * HZ))
4381 && !(rd32(E1000_STATUS) &
4382 E1000_STATUS_TXOFF)) {
4384 /* detected Tx unit hang */
4385 dev_err(&adapter->pdev->dev,
4386 "Detected Tx Unit Hang\n"
4390 " next_to_use <%x>\n"
4391 " next_to_clean <%x>\n"
4392 "buffer_info[next_to_clean]\n"
4393 " time_stamp <%lx>\n"
4394 " next_to_watch <%x>\n"
4396 " desc.status <%x>\n",
4397 tx_ring->queue_index,
4398 readl(adapter->hw.hw_addr + tx_ring->head),
4399 readl(adapter->hw.hw_addr + tx_ring->tail),
4400 tx_ring->next_to_use,
4401 tx_ring->next_to_clean,
4402 tx_ring->buffer_info[i].time_stamp,
4405 eop_desc->wb.status);
4406 netif_stop_subqueue(netdev, tx_ring->queue_index);
4409 tx_ring->total_bytes += total_bytes;
4410 tx_ring->total_packets += total_packets;
4411 tx_ring->tx_stats.bytes += total_bytes;
4412 tx_ring->tx_stats.packets += total_packets;
4413 adapter->net_stats.tx_bytes += total_bytes;
4414 adapter->net_stats.tx_packets += total_packets;
4415 return (count < tx_ring->count);
4419 * igb_receive_skb - helper function to handle rx indications
4420 * @ring: pointer to receive ring receving this packet
4421 * @status: descriptor status field as written by hardware
4422 * @rx_desc: receive descriptor containing vlan and type information.
4423 * @skb: pointer to sk_buff to be indicated to stack
4425 static void igb_receive_skb(struct igb_ring *ring, u8 status,
4426 union e1000_adv_rx_desc * rx_desc,
4427 struct sk_buff *skb)
4429 struct igb_adapter * adapter = ring->adapter;
4430 bool vlan_extracted = (adapter->vlgrp && (status & E1000_RXD_STAT_VP));
4432 skb_record_rx_queue(skb, ring->queue_index);
4434 vlan_gro_receive(&ring->napi, adapter->vlgrp,
4435 le16_to_cpu(rx_desc->wb.upper.vlan),
4438 napi_gro_receive(&ring->napi, skb);
4441 static inline void igb_rx_checksum_adv(struct igb_adapter *adapter,
4442 u32 status_err, struct sk_buff *skb)
4444 skb->ip_summed = CHECKSUM_NONE;
4446 /* Ignore Checksum bit is set or checksum is disabled through ethtool */
4447 if ((status_err & E1000_RXD_STAT_IXSM) || !adapter->rx_csum)
4449 /* TCP/UDP checksum error bit is set */
4451 (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
4452 /* let the stack verify checksum errors */
4453 adapter->hw_csum_err++;
4456 /* It must be a TCP or UDP packet with a valid checksum */
4457 if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
4458 skb->ip_summed = CHECKSUM_UNNECESSARY;
4460 adapter->hw_csum_good++;
4463 static bool igb_clean_rx_irq_adv(struct igb_ring *rx_ring,
4464 int *work_done, int budget)
4466 struct igb_adapter *adapter = rx_ring->adapter;
4467 struct net_device *netdev = adapter->netdev;
4468 struct e1000_hw *hw = &adapter->hw;
4469 struct pci_dev *pdev = adapter->pdev;
4470 union e1000_adv_rx_desc *rx_desc , *next_rxd;
4471 struct igb_buffer *buffer_info , *next_buffer;
4472 struct sk_buff *skb;
4473 bool cleaned = false;
4474 int cleaned_count = 0;
4475 unsigned int total_bytes = 0, total_packets = 0;
4477 u32 length, hlen, staterr;
4479 i = rx_ring->next_to_clean;
4480 buffer_info = &rx_ring->buffer_info[i];
4481 rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
4482 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
4484 while (staterr & E1000_RXD_STAT_DD) {
4485 if (*work_done >= budget)
4489 skb = buffer_info->skb;
4490 prefetch(skb->data - NET_IP_ALIGN);
4491 buffer_info->skb = NULL;
4494 if (i == rx_ring->count)
4496 next_rxd = E1000_RX_DESC_ADV(*rx_ring, i);
4498 next_buffer = &rx_ring->buffer_info[i];
4500 length = le16_to_cpu(rx_desc->wb.upper.length);
4504 if (!adapter->rx_ps_hdr_size) {
4505 pci_unmap_single(pdev, buffer_info->dma,
4506 adapter->rx_buffer_len +
4508 PCI_DMA_FROMDEVICE);
4509 skb_put(skb, length);
4513 /* HW will not DMA in data larger than the given buffer, even
4514 * if it parses the (NFS, of course) header to be larger. In
4515 * that case, it fills the header buffer and spills the rest
4518 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hdr_info) &
4519 E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
4520 if (hlen > adapter->rx_ps_hdr_size)
4521 hlen = adapter->rx_ps_hdr_size;
4523 if (!skb_shinfo(skb)->nr_frags) {
4524 pci_unmap_single(pdev, buffer_info->dma,
4525 adapter->rx_ps_hdr_size + NET_IP_ALIGN,
4526 PCI_DMA_FROMDEVICE);
4531 pci_unmap_page(pdev, buffer_info->page_dma,
4532 PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
4533 buffer_info->page_dma = 0;
4535 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags++,
4537 buffer_info->page_offset,
4540 if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
4541 (page_count(buffer_info->page) != 1))
4542 buffer_info->page = NULL;
4544 get_page(buffer_info->page);
4547 skb->data_len += length;
4549 skb->truesize += length;
4552 if (!(staterr & E1000_RXD_STAT_EOP)) {
4553 buffer_info->skb = next_buffer->skb;
4554 buffer_info->dma = next_buffer->dma;
4555 next_buffer->skb = skb;
4556 next_buffer->dma = 0;
4561 * If this bit is set, then the RX registers contain
4562 * the time stamp. No other packet will be time
4563 * stamped until we read these registers, so read the
4564 * registers to make them available again. Because
4565 * only one packet can be time stamped at a time, we
4566 * know that the register values must belong to this
4567 * one here and therefore we don't need to compare
4568 * any of the additional attributes stored for it.
4570 * If nothing went wrong, then it should have a
4571 * skb_shared_tx that we can turn into a
4572 * skb_shared_hwtstamps.
4574 * TODO: can time stamping be triggered (thus locking
4575 * the registers) without the packet reaching this point
4576 * here? In that case RX time stamping would get stuck.
4578 * TODO: in "time stamp all packets" mode this bit is
4579 * not set. Need a global flag for this mode and then
4580 * always read the registers. Cannot be done without
4583 if (unlikely(staterr & E1000_RXD_STAT_TS)) {
4586 struct skb_shared_hwtstamps *shhwtstamps =
4589 WARN(!(rd32(E1000_TSYNCRXCTL) & E1000_TSYNCRXCTL_VALID),
4590 "igb: no RX time stamp available for time stamped packet");
4591 regval = rd32(E1000_RXSTMPL);
4592 regval |= (u64)rd32(E1000_RXSTMPH) << 32;
4593 ns = timecounter_cyc2time(&adapter->clock, regval);
4594 timecompare_update(&adapter->compare, ns);
4595 memset(shhwtstamps, 0, sizeof(*shhwtstamps));
4596 shhwtstamps->hwtstamp = ns_to_ktime(ns);
4597 shhwtstamps->syststamp =
4598 timecompare_transform(&adapter->compare, ns);
4601 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
4602 dev_kfree_skb_irq(skb);
4606 total_bytes += skb->len;
4609 igb_rx_checksum_adv(adapter, staterr, skb);
4611 skb->protocol = eth_type_trans(skb, netdev);
4613 igb_receive_skb(rx_ring, staterr, rx_desc, skb);
4616 rx_desc->wb.upper.status_error = 0;
4618 /* return some buffers to hardware, one at a time is too slow */
4619 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
4620 igb_alloc_rx_buffers_adv(rx_ring, cleaned_count);
4624 /* use prefetched values */
4626 buffer_info = next_buffer;
4627 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
4630 rx_ring->next_to_clean = i;
4631 cleaned_count = igb_desc_unused(rx_ring);
4634 igb_alloc_rx_buffers_adv(rx_ring, cleaned_count);
4636 rx_ring->total_packets += total_packets;
4637 rx_ring->total_bytes += total_bytes;
4638 rx_ring->rx_stats.packets += total_packets;
4639 rx_ring->rx_stats.bytes += total_bytes;
4640 adapter->net_stats.rx_bytes += total_bytes;
4641 adapter->net_stats.rx_packets += total_packets;
4646 * igb_alloc_rx_buffers_adv - Replace used receive buffers; packet split
4647 * @adapter: address of board private structure
4649 static void igb_alloc_rx_buffers_adv(struct igb_ring *rx_ring,
4652 struct igb_adapter *adapter = rx_ring->adapter;
4653 struct net_device *netdev = adapter->netdev;
4654 struct pci_dev *pdev = adapter->pdev;
4655 union e1000_adv_rx_desc *rx_desc;
4656 struct igb_buffer *buffer_info;
4657 struct sk_buff *skb;
4661 i = rx_ring->next_to_use;
4662 buffer_info = &rx_ring->buffer_info[i];
4664 if (adapter->rx_ps_hdr_size)
4665 bufsz = adapter->rx_ps_hdr_size;
4667 bufsz = adapter->rx_buffer_len;
4668 bufsz += NET_IP_ALIGN;
4670 while (cleaned_count--) {
4671 rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
4673 if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) {
4674 if (!buffer_info->page) {
4675 buffer_info->page = alloc_page(GFP_ATOMIC);
4676 if (!buffer_info->page) {
4677 adapter->alloc_rx_buff_failed++;
4680 buffer_info->page_offset = 0;
4682 buffer_info->page_offset ^= PAGE_SIZE / 2;
4684 buffer_info->page_dma =
4685 pci_map_page(pdev, buffer_info->page,
4686 buffer_info->page_offset,
4688 PCI_DMA_FROMDEVICE);
4691 if (!buffer_info->skb) {
4692 skb = netdev_alloc_skb(netdev, bufsz);
4694 adapter->alloc_rx_buff_failed++;
4698 /* Make buffer alignment 2 beyond a 16 byte boundary
4699 * this will result in a 16 byte aligned IP header after
4700 * the 14 byte MAC header is removed
4702 skb_reserve(skb, NET_IP_ALIGN);
4704 buffer_info->skb = skb;
4705 buffer_info->dma = pci_map_single(pdev, skb->data,
4707 PCI_DMA_FROMDEVICE);
4709 /* Refresh the desc even if buffer_addrs didn't change because
4710 * each write-back erases this info. */
4711 if (adapter->rx_ps_hdr_size) {
4712 rx_desc->read.pkt_addr =
4713 cpu_to_le64(buffer_info->page_dma);
4714 rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
4716 rx_desc->read.pkt_addr =
4717 cpu_to_le64(buffer_info->dma);
4718 rx_desc->read.hdr_addr = 0;
4722 if (i == rx_ring->count)
4724 buffer_info = &rx_ring->buffer_info[i];
4728 if (rx_ring->next_to_use != i) {
4729 rx_ring->next_to_use = i;
4731 i = (rx_ring->count - 1);
4735 /* Force memory writes to complete before letting h/w
4736 * know there are new descriptors to fetch. (Only
4737 * applicable for weak-ordered memory model archs,
4738 * such as IA-64). */
4740 writel(i, adapter->hw.hw_addr + rx_ring->tail);
4750 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4752 struct igb_adapter *adapter = netdev_priv(netdev);
4753 struct mii_ioctl_data *data = if_mii(ifr);
4755 if (adapter->hw.phy.media_type != e1000_media_type_copper)
4760 data->phy_id = adapter->hw.phy.addr;
4763 if (!capable(CAP_NET_ADMIN))
4765 if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
4777 * igb_hwtstamp_ioctl - control hardware time stamping
4782 * Outgoing time stamping can be enabled and disabled. Play nice and
4783 * disable it when requested, although it shouldn't case any overhead
4784 * when no packet needs it. At most one packet in the queue may be
4785 * marked for time stamping, otherwise it would be impossible to tell
4786 * for sure to which packet the hardware time stamp belongs.
4788 * Incoming time stamping has to be configured via the hardware
4789 * filters. Not all combinations are supported, in particular event
4790 * type has to be specified. Matching the kind of event packet is
4791 * not supported, with the exception of "all V2 events regardless of
4795 static int igb_hwtstamp_ioctl(struct net_device *netdev,
4796 struct ifreq *ifr, int cmd)
4798 struct igb_adapter *adapter = netdev_priv(netdev);
4799 struct e1000_hw *hw = &adapter->hw;
4800 struct hwtstamp_config config;
4801 u32 tsync_tx_ctl_bit = E1000_TSYNCTXCTL_ENABLED;
4802 u32 tsync_rx_ctl_bit = E1000_TSYNCRXCTL_ENABLED;
4803 u32 tsync_rx_ctl_type = 0;
4804 u32 tsync_rx_cfg = 0;
4807 short port = 319; /* PTP */
4810 if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
4813 /* reserved for future extensions */
4817 switch (config.tx_type) {
4818 case HWTSTAMP_TX_OFF:
4819 tsync_tx_ctl_bit = 0;
4821 case HWTSTAMP_TX_ON:
4822 tsync_tx_ctl_bit = E1000_TSYNCTXCTL_ENABLED;
4828 switch (config.rx_filter) {
4829 case HWTSTAMP_FILTER_NONE:
4830 tsync_rx_ctl_bit = 0;
4832 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
4833 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
4834 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
4835 case HWTSTAMP_FILTER_ALL:
4837 * register TSYNCRXCFG must be set, therefore it is not
4838 * possible to time stamp both Sync and Delay_Req messages
4839 * => fall back to time stamping all packets
4841 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_ALL;
4842 config.rx_filter = HWTSTAMP_FILTER_ALL;
4844 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
4845 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_L4_V1;
4846 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V1_SYNC_MESSAGE;
4849 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
4850 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_L4_V1;
4851 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V1_DELAY_REQ_MESSAGE;
4854 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
4855 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
4856 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
4857 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V2_SYNC_MESSAGE;
4860 config.rx_filter = HWTSTAMP_FILTER_SOME;
4862 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
4863 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
4864 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
4865 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V2_DELAY_REQ_MESSAGE;
4868 config.rx_filter = HWTSTAMP_FILTER_SOME;
4870 case HWTSTAMP_FILTER_PTP_V2_EVENT:
4871 case HWTSTAMP_FILTER_PTP_V2_SYNC:
4872 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
4873 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_EVENT_V2;
4874 config.rx_filter = HWTSTAMP_FILTER_PTP_V2_EVENT;
4881 /* enable/disable TX */
4882 regval = rd32(E1000_TSYNCTXCTL);
4883 regval = (regval & ~E1000_TSYNCTXCTL_ENABLED) | tsync_tx_ctl_bit;
4884 wr32(E1000_TSYNCTXCTL, regval);
4886 /* enable/disable RX, define which PTP packets are time stamped */
4887 regval = rd32(E1000_TSYNCRXCTL);
4888 regval = (regval & ~E1000_TSYNCRXCTL_ENABLED) | tsync_rx_ctl_bit;
4889 regval = (regval & ~0xE) | tsync_rx_ctl_type;
4890 wr32(E1000_TSYNCRXCTL, regval);
4891 wr32(E1000_TSYNCRXCFG, tsync_rx_cfg);
4894 * Ethertype Filter Queue Filter[0][15:0] = 0x88F7
4895 * (Ethertype to filter on)
4896 * Ethertype Filter Queue Filter[0][26] = 0x1 (Enable filter)
4897 * Ethertype Filter Queue Filter[0][30] = 0x1 (Enable Timestamping)
4899 wr32(E1000_ETQF0, is_l2 ? 0x440088f7 : 0);
4901 /* L4 Queue Filter[0]: only filter by source and destination port */
4902 wr32(E1000_SPQF0, htons(port));
4903 wr32(E1000_IMIREXT(0), is_l4 ?
4904 ((1<<12) | (1<<19) /* bypass size and control flags */) : 0);
4905 wr32(E1000_IMIR(0), is_l4 ?
4907 | (0<<16) /* immediate interrupt disabled */
4908 | 0 /* (1<<17) bit cleared: do not bypass
4909 destination port check */)
4911 wr32(E1000_FTQF0, is_l4 ?
4913 | (1<<15) /* VF not compared */
4914 | (1<<27) /* Enable Timestamping */
4915 | (7<<28) /* only source port filter enabled,
4916 source/target address and protocol
4918 : ((1<<15) | (15<<28) /* all mask bits set = filter not
4923 adapter->hwtstamp_config = config;
4925 /* clear TX/RX time stamp registers, just to be sure */
4926 regval = rd32(E1000_TXSTMPH);
4927 regval = rd32(E1000_RXSTMPH);
4929 return copy_to_user(ifr->ifr_data, &config, sizeof(config)) ?
4939 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4945 return igb_mii_ioctl(netdev, ifr, cmd);
4947 return igb_hwtstamp_ioctl(netdev, ifr, cmd);
4953 static void igb_vlan_rx_register(struct net_device *netdev,
4954 struct vlan_group *grp)
4956 struct igb_adapter *adapter = netdev_priv(netdev);
4957 struct e1000_hw *hw = &adapter->hw;
4960 igb_irq_disable(adapter);
4961 adapter->vlgrp = grp;
4964 /* enable VLAN tag insert/strip */
4965 ctrl = rd32(E1000_CTRL);
4966 ctrl |= E1000_CTRL_VME;
4967 wr32(E1000_CTRL, ctrl);
4969 /* enable VLAN receive filtering */
4970 rctl = rd32(E1000_RCTL);
4971 rctl &= ~E1000_RCTL_CFIEN;
4972 wr32(E1000_RCTL, rctl);
4973 igb_update_mng_vlan(adapter);
4975 /* disable VLAN tag insert/strip */
4976 ctrl = rd32(E1000_CTRL);
4977 ctrl &= ~E1000_CTRL_VME;
4978 wr32(E1000_CTRL, ctrl);
4980 if (adapter->mng_vlan_id != (u16)IGB_MNG_VLAN_NONE) {
4981 igb_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
4982 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
4986 igb_rlpml_set(adapter);
4988 if (!test_bit(__IGB_DOWN, &adapter->state))
4989 igb_irq_enable(adapter);
4992 static void igb_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
4994 struct igb_adapter *adapter = netdev_priv(netdev);
4995 struct e1000_hw *hw = &adapter->hw;
4996 int pf_id = adapter->vfs_allocated_count;
4998 if ((hw->mng_cookie.status &
4999 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
5000 (vid == adapter->mng_vlan_id))
5003 /* add vid to vlvf if sr-iov is enabled,
5004 * if that fails add directly to filter table */
5005 if (igb_vlvf_set(adapter, vid, true, pf_id))
5006 igb_vfta_set(hw, vid, true);
5010 static void igb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
5012 struct igb_adapter *adapter = netdev_priv(netdev);
5013 struct e1000_hw *hw = &adapter->hw;
5014 int pf_id = adapter->vfs_allocated_count;
5016 igb_irq_disable(adapter);
5017 vlan_group_set_device(adapter->vlgrp, vid, NULL);
5019 if (!test_bit(__IGB_DOWN, &adapter->state))
5020 igb_irq_enable(adapter);
5022 if ((adapter->hw.mng_cookie.status &
5023 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
5024 (vid == adapter->mng_vlan_id)) {
5025 /* release control to f/w */
5026 igb_release_hw_control(adapter);
5030 /* remove vid from vlvf if sr-iov is enabled,
5031 * if not in vlvf remove from vfta */
5032 if (igb_vlvf_set(adapter, vid, false, pf_id))
5033 igb_vfta_set(hw, vid, false);
5036 static void igb_restore_vlan(struct igb_adapter *adapter)
5038 igb_vlan_rx_register(adapter->netdev, adapter->vlgrp);
5040 if (adapter->vlgrp) {
5042 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
5043 if (!vlan_group_get_device(adapter->vlgrp, vid))
5045 igb_vlan_rx_add_vid(adapter->netdev, vid);
5050 int igb_set_spd_dplx(struct igb_adapter *adapter, u16 spddplx)
5052 struct e1000_mac_info *mac = &adapter->hw.mac;
5056 /* Fiber NICs only allow 1000 gbps Full duplex */
5057 if ((adapter->hw.phy.media_type == e1000_media_type_fiber) &&
5058 spddplx != (SPEED_1000 + DUPLEX_FULL)) {
5059 dev_err(&adapter->pdev->dev,
5060 "Unsupported Speed/Duplex configuration\n");
5065 case SPEED_10 + DUPLEX_HALF:
5066 mac->forced_speed_duplex = ADVERTISE_10_HALF;
5068 case SPEED_10 + DUPLEX_FULL:
5069 mac->forced_speed_duplex = ADVERTISE_10_FULL;
5071 case SPEED_100 + DUPLEX_HALF:
5072 mac->forced_speed_duplex = ADVERTISE_100_HALF;
5074 case SPEED_100 + DUPLEX_FULL:
5075 mac->forced_speed_duplex = ADVERTISE_100_FULL;
5077 case SPEED_1000 + DUPLEX_FULL:
5079 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
5081 case SPEED_1000 + DUPLEX_HALF: /* not supported */
5083 dev_err(&adapter->pdev->dev,
5084 "Unsupported Speed/Duplex configuration\n");
5090 static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake)
5092 struct net_device *netdev = pci_get_drvdata(pdev);
5093 struct igb_adapter *adapter = netdev_priv(netdev);
5094 struct e1000_hw *hw = &adapter->hw;
5095 u32 ctrl, rctl, status;
5096 u32 wufc = adapter->wol;
5101 netif_device_detach(netdev);
5103 if (netif_running(netdev))
5106 igb_reset_interrupt_capability(adapter);
5108 igb_free_queues(adapter);
5111 retval = pci_save_state(pdev);
5116 status = rd32(E1000_STATUS);
5117 if (status & E1000_STATUS_LU)
5118 wufc &= ~E1000_WUFC_LNKC;
5121 igb_setup_rctl(adapter);
5122 igb_set_multi(netdev);
5124 /* turn on all-multi mode if wake on multicast is enabled */
5125 if (wufc & E1000_WUFC_MC) {
5126 rctl = rd32(E1000_RCTL);
5127 rctl |= E1000_RCTL_MPE;
5128 wr32(E1000_RCTL, rctl);
5131 ctrl = rd32(E1000_CTRL);
5132 /* advertise wake from D3Cold */
5133 #define E1000_CTRL_ADVD3WUC 0x00100000
5134 /* phy power management enable */
5135 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5136 ctrl |= E1000_CTRL_ADVD3WUC;
5137 wr32(E1000_CTRL, ctrl);
5139 /* Allow time for pending master requests to run */
5140 igb_disable_pcie_master(&adapter->hw);
5142 wr32(E1000_WUC, E1000_WUC_PME_EN);
5143 wr32(E1000_WUFC, wufc);
5146 wr32(E1000_WUFC, 0);
5149 *enable_wake = wufc || adapter->en_mng_pt;
5151 igb_shutdown_fiber_serdes_link_82575(hw);
5153 /* Release control of h/w to f/w. If f/w is AMT enabled, this
5154 * would have already happened in close and is redundant. */
5155 igb_release_hw_control(adapter);
5157 pci_disable_device(pdev);
5163 static int igb_suspend(struct pci_dev *pdev, pm_message_t state)
5168 retval = __igb_shutdown(pdev, &wake);
5173 pci_prepare_to_sleep(pdev);
5175 pci_wake_from_d3(pdev, false);
5176 pci_set_power_state(pdev, PCI_D3hot);
5182 static int igb_resume(struct pci_dev *pdev)
5184 struct net_device *netdev = pci_get_drvdata(pdev);
5185 struct igb_adapter *adapter = netdev_priv(netdev);
5186 struct e1000_hw *hw = &adapter->hw;
5189 pci_set_power_state(pdev, PCI_D0);
5190 pci_restore_state(pdev);
5192 err = pci_enable_device_mem(pdev);
5195 "igb: Cannot enable PCI device from suspend\n");
5198 pci_set_master(pdev);
5200 pci_enable_wake(pdev, PCI_D3hot, 0);
5201 pci_enable_wake(pdev, PCI_D3cold, 0);
5203 igb_set_interrupt_capability(adapter);
5205 if (igb_alloc_queues(adapter)) {
5206 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
5210 /* e1000_power_up_phy(adapter); */
5214 /* let the f/w know that the h/w is now under the control of the
5216 igb_get_hw_control(adapter);
5218 wr32(E1000_WUS, ~0);
5220 if (netif_running(netdev)) {
5221 err = igb_open(netdev);
5226 netif_device_attach(netdev);
5232 static void igb_shutdown(struct pci_dev *pdev)
5236 __igb_shutdown(pdev, &wake);
5238 if (system_state == SYSTEM_POWER_OFF) {
5239 pci_wake_from_d3(pdev, wake);
5240 pci_set_power_state(pdev, PCI_D3hot);
5244 #ifdef CONFIG_NET_POLL_CONTROLLER
5246 * Polling 'interrupt' - used by things like netconsole to send skbs
5247 * without having to re-enable interrupts. It's not called while
5248 * the interrupt routine is executing.
5250 static void igb_netpoll(struct net_device *netdev)
5252 struct igb_adapter *adapter = netdev_priv(netdev);
5253 struct e1000_hw *hw = &adapter->hw;
5256 if (!adapter->msix_entries) {
5257 igb_irq_disable(adapter);
5258 napi_schedule(&adapter->rx_ring[0].napi);
5262 for (i = 0; i < adapter->num_tx_queues; i++) {
5263 struct igb_ring *tx_ring = &adapter->tx_ring[i];
5264 wr32(E1000_EIMC, tx_ring->eims_value);
5265 igb_clean_tx_irq(tx_ring);
5266 wr32(E1000_EIMS, tx_ring->eims_value);
5269 for (i = 0; i < adapter->num_rx_queues; i++) {
5270 struct igb_ring *rx_ring = &adapter->rx_ring[i];
5271 wr32(E1000_EIMC, rx_ring->eims_value);
5272 napi_schedule(&rx_ring->napi);
5275 #endif /* CONFIG_NET_POLL_CONTROLLER */
5278 * igb_io_error_detected - called when PCI error is detected
5279 * @pdev: Pointer to PCI device
5280 * @state: The current pci connection state
5282 * This function is called after a PCI bus error affecting
5283 * this device has been detected.
5285 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
5286 pci_channel_state_t state)
5288 struct net_device *netdev = pci_get_drvdata(pdev);
5289 struct igb_adapter *adapter = netdev_priv(netdev);
5291 netif_device_detach(netdev);
5293 if (netif_running(netdev))
5295 pci_disable_device(pdev);
5297 /* Request a slot slot reset. */
5298 return PCI_ERS_RESULT_NEED_RESET;
5302 * igb_io_slot_reset - called after the pci bus has been reset.
5303 * @pdev: Pointer to PCI device
5305 * Restart the card from scratch, as if from a cold-boot. Implementation
5306 * resembles the first-half of the igb_resume routine.
5308 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
5310 struct net_device *netdev = pci_get_drvdata(pdev);
5311 struct igb_adapter *adapter = netdev_priv(netdev);
5312 struct e1000_hw *hw = &adapter->hw;
5313 pci_ers_result_t result;
5316 if (pci_enable_device_mem(pdev)) {
5318 "Cannot re-enable PCI device after reset.\n");
5319 result = PCI_ERS_RESULT_DISCONNECT;
5321 pci_set_master(pdev);
5322 pci_restore_state(pdev);
5324 pci_enable_wake(pdev, PCI_D3hot, 0);
5325 pci_enable_wake(pdev, PCI_D3cold, 0);
5328 wr32(E1000_WUS, ~0);
5329 result = PCI_ERS_RESULT_RECOVERED;
5332 err = pci_cleanup_aer_uncorrect_error_status(pdev);
5334 dev_err(&pdev->dev, "pci_cleanup_aer_uncorrect_error_status "
5335 "failed 0x%0x\n", err);
5336 /* non-fatal, continue */
5343 * igb_io_resume - called when traffic can start flowing again.
5344 * @pdev: Pointer to PCI device
5346 * This callback is called when the error recovery driver tells us that
5347 * its OK to resume normal operation. Implementation resembles the
5348 * second-half of the igb_resume routine.
5350 static void igb_io_resume(struct pci_dev *pdev)
5352 struct net_device *netdev = pci_get_drvdata(pdev);
5353 struct igb_adapter *adapter = netdev_priv(netdev);
5355 if (netif_running(netdev)) {
5356 if (igb_up(adapter)) {
5357 dev_err(&pdev->dev, "igb_up failed after reset\n");
5362 netif_device_attach(netdev);
5364 /* let the f/w know that the h/w is now under the control of the
5366 igb_get_hw_control(adapter);
5369 static inline void igb_set_vmolr(struct e1000_hw *hw, int vfn)
5373 reg_data = rd32(E1000_VMOLR(vfn));
5374 reg_data |= E1000_VMOLR_BAM | /* Accept broadcast */
5375 E1000_VMOLR_ROPE | /* Accept packets matched in UTA */
5376 E1000_VMOLR_ROMPE | /* Accept packets matched in MTA */
5377 E1000_VMOLR_AUPE | /* Accept untagged packets */
5378 E1000_VMOLR_STRVLAN; /* Strip vlan tags */
5379 wr32(E1000_VMOLR(vfn), reg_data);
5382 static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
5385 struct e1000_hw *hw = &adapter->hw;
5388 vmolr = rd32(E1000_VMOLR(vfn));
5389 vmolr &= ~E1000_VMOLR_RLPML_MASK;
5390 vmolr |= size | E1000_VMOLR_LPE;
5391 wr32(E1000_VMOLR(vfn), vmolr);
5396 static inline void igb_set_rah_pool(struct e1000_hw *hw, int pool, int entry)
5400 reg_data = rd32(E1000_RAH(entry));
5401 reg_data &= ~E1000_RAH_POOL_MASK;
5402 reg_data |= E1000_RAH_POOL_1 << pool;;
5403 wr32(E1000_RAH(entry), reg_data);
5406 static void igb_set_mc_list_pools(struct igb_adapter *adapter,
5407 int entry_count, u16 total_rar_filters)
5409 struct e1000_hw *hw = &adapter->hw;
5410 int i = adapter->vfs_allocated_count + 1;
5412 if ((i + entry_count) < total_rar_filters)
5413 total_rar_filters = i + entry_count;
5415 for (; i < total_rar_filters; i++)
5416 igb_set_rah_pool(hw, adapter->vfs_allocated_count, i);
5419 static int igb_set_vf_mac(struct igb_adapter *adapter,
5420 int vf, unsigned char *mac_addr)
5422 struct e1000_hw *hw = &adapter->hw;
5423 int rar_entry = vf + 1; /* VF MAC addresses start at entry 1 */
5425 igb_rar_set(hw, mac_addr, rar_entry);
5427 memcpy(adapter->vf_data[vf].vf_mac_addresses, mac_addr, ETH_ALEN);
5429 igb_set_rah_pool(hw, vf, rar_entry);
5434 static void igb_vmm_control(struct igb_adapter *adapter)
5436 struct e1000_hw *hw = &adapter->hw;
5439 if (!adapter->vfs_allocated_count)
5442 /* VF's need PF reset indication before they
5443 * can send/receive mail */
5444 reg_data = rd32(E1000_CTRL_EXT);
5445 reg_data |= E1000_CTRL_EXT_PFRSTD;
5446 wr32(E1000_CTRL_EXT, reg_data);
5448 igb_vmdq_set_loopback_pf(hw, true);
5449 igb_vmdq_set_replication_pf(hw, true);