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 if (adapter->hw.mac.type == e1000_82576)
1349 netdev->features |= NETIF_F_SCTP_CSUM;
1351 adapter->en_mng_pt = igb_enable_mng_pass_thru(&adapter->hw);
1353 /* before reading the NVM, reset the controller to put the device in a
1354 * known good starting state */
1355 hw->mac.ops.reset_hw(hw);
1357 /* make sure the NVM is good */
1358 if (igb_validate_nvm_checksum(hw) < 0) {
1359 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
1364 /* copy the MAC address out of the NVM */
1365 if (hw->mac.ops.read_mac_addr(hw))
1366 dev_err(&pdev->dev, "NVM Read Error\n");
1368 memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
1369 memcpy(netdev->perm_addr, hw->mac.addr, netdev->addr_len);
1371 if (!is_valid_ether_addr(netdev->perm_addr)) {
1372 dev_err(&pdev->dev, "Invalid MAC Address\n");
1377 setup_timer(&adapter->watchdog_timer, &igb_watchdog,
1378 (unsigned long) adapter);
1379 setup_timer(&adapter->phy_info_timer, &igb_update_phy_info,
1380 (unsigned long) adapter);
1382 INIT_WORK(&adapter->reset_task, igb_reset_task);
1383 INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
1385 /* Initialize link properties that are user-changeable */
1386 adapter->fc_autoneg = true;
1387 hw->mac.autoneg = true;
1388 hw->phy.autoneg_advertised = 0x2f;
1390 hw->fc.original_type = e1000_fc_default;
1391 hw->fc.type = e1000_fc_default;
1393 adapter->itr_setting = IGB_DEFAULT_ITR;
1394 adapter->itr = IGB_START_ITR;
1396 igb_validate_mdi_setting(hw);
1398 /* Initial Wake on LAN setting If APM wake is enabled in the EEPROM,
1399 * enable the ACPI Magic Packet filter
1402 if (hw->bus.func == 0)
1403 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
1404 else if (hw->bus.func == 1)
1405 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
1407 if (eeprom_data & eeprom_apme_mask)
1408 adapter->eeprom_wol |= E1000_WUFC_MAG;
1410 /* now that we have the eeprom settings, apply the special cases where
1411 * the eeprom may be wrong or the board simply won't support wake on
1412 * lan on a particular port */
1413 switch (pdev->device) {
1414 case E1000_DEV_ID_82575GB_QUAD_COPPER:
1415 adapter->eeprom_wol = 0;
1417 case E1000_DEV_ID_82575EB_FIBER_SERDES:
1418 case E1000_DEV_ID_82576_FIBER:
1419 case E1000_DEV_ID_82576_SERDES:
1420 /* Wake events only supported on port A for dual fiber
1421 * regardless of eeprom setting */
1422 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
1423 adapter->eeprom_wol = 0;
1425 case E1000_DEV_ID_82576_QUAD_COPPER:
1426 /* if quad port adapter, disable WoL on all but port A */
1427 if (global_quad_port_a != 0)
1428 adapter->eeprom_wol = 0;
1430 adapter->flags |= IGB_FLAG_QUAD_PORT_A;
1431 /* Reset for multiple quad port adapters */
1432 if (++global_quad_port_a == 4)
1433 global_quad_port_a = 0;
1437 /* initialize the wol settings based on the eeprom settings */
1438 adapter->wol = adapter->eeprom_wol;
1439 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
1441 /* reset the hardware with the new settings */
1444 /* let the f/w know that the h/w is now under the control of the
1446 igb_get_hw_control(adapter);
1448 strcpy(netdev->name, "eth%d");
1449 err = register_netdev(netdev);
1453 /* carrier off reporting is important to ethtool even BEFORE open */
1454 netif_carrier_off(netdev);
1456 #ifdef CONFIG_IGB_DCA
1457 if (dca_add_requester(&pdev->dev) == 0) {
1458 adapter->flags |= IGB_FLAG_DCA_ENABLED;
1459 dev_info(&pdev->dev, "DCA enabled\n");
1460 /* Always use CB2 mode, difference is masked
1461 * in the CB driver. */
1462 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
1463 igb_setup_dca(adapter);
1468 * Initialize hardware timer: we keep it running just in case
1469 * that some program needs it later on.
1471 memset(&adapter->cycles, 0, sizeof(adapter->cycles));
1472 adapter->cycles.read = igb_read_clock;
1473 adapter->cycles.mask = CLOCKSOURCE_MASK(64);
1474 adapter->cycles.mult = 1;
1475 adapter->cycles.shift = IGB_TSYNC_SHIFT;
1478 IGB_TSYNC_CYCLE_TIME_IN_NANOSECONDS * IGB_TSYNC_SCALE);
1481 * Avoid rollover while we initialize by resetting the time counter.
1483 wr32(E1000_SYSTIML, 0x00000000);
1484 wr32(E1000_SYSTIMH, 0x00000000);
1487 * Set registers so that rollover occurs soon to test this.
1489 wr32(E1000_SYSTIML, 0x00000000);
1490 wr32(E1000_SYSTIMH, 0xFF800000);
1493 timecounter_init(&adapter->clock,
1495 ktime_to_ns(ktime_get_real()));
1498 * Synchronize our NIC clock against system wall clock. NIC
1499 * time stamp reading requires ~3us per sample, each sample
1500 * was pretty stable even under load => only require 10
1501 * samples for each offset comparison.
1503 memset(&adapter->compare, 0, sizeof(adapter->compare));
1504 adapter->compare.source = &adapter->clock;
1505 adapter->compare.target = ktime_get_real;
1506 adapter->compare.num_samples = 10;
1507 timecompare_update(&adapter->compare, 0);
1513 "igb: %s: hw %p initialized timer\n",
1514 igb_get_time_str(adapter, buffer),
1519 dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
1520 /* print bus type/speed/width info */
1521 dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
1523 ((hw->bus.speed == e1000_bus_speed_2500)
1524 ? "2.5Gb/s" : "unknown"),
1525 ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
1526 (hw->bus.width == e1000_bus_width_pcie_x2) ? "Width x2" :
1527 (hw->bus.width == e1000_bus_width_pcie_x1) ? "Width x1" :
1531 igb_read_part_num(hw, &part_num);
1532 dev_info(&pdev->dev, "%s: PBA No: %06x-%03x\n", netdev->name,
1533 (part_num >> 8), (part_num & 0xff));
1535 dev_info(&pdev->dev,
1536 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
1537 adapter->msix_entries ? "MSI-X" :
1538 (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
1539 adapter->num_rx_queues, adapter->num_tx_queues);
1544 igb_release_hw_control(adapter);
1546 if (!igb_check_reset_block(hw))
1549 if (hw->flash_address)
1550 iounmap(hw->flash_address);
1552 igb_free_queues(adapter);
1554 iounmap(hw->hw_addr);
1556 free_netdev(netdev);
1558 pci_release_selected_regions(pdev, pci_select_bars(pdev,
1562 pci_disable_device(pdev);
1567 * igb_remove - Device Removal Routine
1568 * @pdev: PCI device information struct
1570 * igb_remove is called by the PCI subsystem to alert the driver
1571 * that it should release a PCI device. The could be caused by a
1572 * Hot-Plug event, or because the driver is going to be removed from
1575 static void __devexit igb_remove(struct pci_dev *pdev)
1577 struct net_device *netdev = pci_get_drvdata(pdev);
1578 struct igb_adapter *adapter = netdev_priv(netdev);
1579 struct e1000_hw *hw = &adapter->hw;
1582 /* flush_scheduled work may reschedule our watchdog task, so
1583 * explicitly disable watchdog tasks from being rescheduled */
1584 set_bit(__IGB_DOWN, &adapter->state);
1585 del_timer_sync(&adapter->watchdog_timer);
1586 del_timer_sync(&adapter->phy_info_timer);
1588 flush_scheduled_work();
1590 #ifdef CONFIG_IGB_DCA
1591 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
1592 dev_info(&pdev->dev, "DCA disabled\n");
1593 dca_remove_requester(&pdev->dev);
1594 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
1595 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
1599 /* Release control of h/w to f/w. If f/w is AMT enabled, this
1600 * would have already happened in close and is redundant. */
1601 igb_release_hw_control(adapter);
1603 unregister_netdev(netdev);
1605 if (!igb_check_reset_block(&adapter->hw))
1606 igb_reset_phy(&adapter->hw);
1608 igb_reset_interrupt_capability(adapter);
1610 igb_free_queues(adapter);
1612 #ifdef CONFIG_PCI_IOV
1613 /* reclaim resources allocated to VFs */
1614 if (adapter->vf_data) {
1615 /* disable iov and allow time for transactions to clear */
1616 pci_disable_sriov(pdev);
1619 kfree(adapter->vf_data);
1620 adapter->vf_data = NULL;
1621 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
1623 dev_info(&pdev->dev, "IOV Disabled\n");
1626 iounmap(hw->hw_addr);
1627 if (hw->flash_address)
1628 iounmap(hw->flash_address);
1629 pci_release_selected_regions(pdev, pci_select_bars(pdev,
1632 free_netdev(netdev);
1634 err = pci_disable_pcie_error_reporting(pdev);
1637 "pci_disable_pcie_error_reporting failed 0x%x\n", err);
1639 pci_disable_device(pdev);
1643 * igb_sw_init - Initialize general software structures (struct igb_adapter)
1644 * @adapter: board private structure to initialize
1646 * igb_sw_init initializes the Adapter private data structure.
1647 * Fields are initialized based on PCI device information and
1648 * OS network device settings (MTU size).
1650 static int __devinit igb_sw_init(struct igb_adapter *adapter)
1652 struct e1000_hw *hw = &adapter->hw;
1653 struct net_device *netdev = adapter->netdev;
1654 struct pci_dev *pdev = adapter->pdev;
1656 pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
1658 adapter->tx_ring_count = IGB_DEFAULT_TXD;
1659 adapter->rx_ring_count = IGB_DEFAULT_RXD;
1660 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1661 adapter->rx_ps_hdr_size = 0; /* disable packet split */
1662 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1663 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1665 /* This call may decrease the number of queues depending on
1666 * interrupt mode. */
1667 igb_set_interrupt_capability(adapter);
1669 if (igb_alloc_queues(adapter)) {
1670 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
1674 /* Explicitly disable IRQ since the NIC can be in any state. */
1675 igb_irq_disable(adapter);
1677 set_bit(__IGB_DOWN, &adapter->state);
1682 * igb_open - Called when a network interface is made active
1683 * @netdev: network interface device structure
1685 * Returns 0 on success, negative value on failure
1687 * The open entry point is called when a network interface is made
1688 * active by the system (IFF_UP). At this point all resources needed
1689 * for transmit and receive operations are allocated, the interrupt
1690 * handler is registered with the OS, the watchdog timer is started,
1691 * and the stack is notified that the interface is ready.
1693 static int igb_open(struct net_device *netdev)
1695 struct igb_adapter *adapter = netdev_priv(netdev);
1696 struct e1000_hw *hw = &adapter->hw;
1700 /* disallow open during test */
1701 if (test_bit(__IGB_TESTING, &adapter->state))
1704 netif_carrier_off(netdev);
1706 /* allocate transmit descriptors */
1707 err = igb_setup_all_tx_resources(adapter);
1711 /* allocate receive descriptors */
1712 err = igb_setup_all_rx_resources(adapter);
1716 /* e1000_power_up_phy(adapter); */
1718 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1719 if ((adapter->hw.mng_cookie.status &
1720 E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
1721 igb_update_mng_vlan(adapter);
1723 /* before we allocate an interrupt, we must be ready to handle it.
1724 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1725 * as soon as we call pci_request_irq, so we have to setup our
1726 * clean_rx handler before we do so. */
1727 igb_configure(adapter);
1729 igb_vmm_control(adapter);
1730 igb_set_rah_pool(hw, adapter->vfs_allocated_count, 0);
1731 igb_set_vmolr(hw, adapter->vfs_allocated_count);
1733 err = igb_request_irq(adapter);
1737 /* From here on the code is the same as igb_up() */
1738 clear_bit(__IGB_DOWN, &adapter->state);
1740 for (i = 0; i < adapter->num_rx_queues; i++)
1741 napi_enable(&adapter->rx_ring[i].napi);
1743 /* Clear any pending interrupts. */
1746 igb_irq_enable(adapter);
1748 netif_tx_start_all_queues(netdev);
1750 /* Fire a link status change interrupt to start the watchdog. */
1751 wr32(E1000_ICS, E1000_ICS_LSC);
1756 igb_release_hw_control(adapter);
1757 /* e1000_power_down_phy(adapter); */
1758 igb_free_all_rx_resources(adapter);
1760 igb_free_all_tx_resources(adapter);
1768 * igb_close - Disables a network interface
1769 * @netdev: network interface device structure
1771 * Returns 0, this is not allowed to fail
1773 * The close entry point is called when an interface is de-activated
1774 * by the OS. The hardware is still under the driver's control, but
1775 * needs to be disabled. A global MAC reset is issued to stop the
1776 * hardware, and all transmit and receive resources are freed.
1778 static int igb_close(struct net_device *netdev)
1780 struct igb_adapter *adapter = netdev_priv(netdev);
1782 WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
1785 igb_free_irq(adapter);
1787 igb_free_all_tx_resources(adapter);
1788 igb_free_all_rx_resources(adapter);
1790 /* kill manageability vlan ID if supported, but not if a vlan with
1791 * the same ID is registered on the host OS (let 8021q kill it) */
1792 if ((adapter->hw.mng_cookie.status &
1793 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1795 vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
1796 igb_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1802 * igb_setup_tx_resources - allocate Tx resources (Descriptors)
1803 * @adapter: board private structure
1804 * @tx_ring: tx descriptor ring (for a specific queue) to setup
1806 * Return 0 on success, negative on failure
1808 int igb_setup_tx_resources(struct igb_adapter *adapter,
1809 struct igb_ring *tx_ring)
1811 struct pci_dev *pdev = adapter->pdev;
1814 size = sizeof(struct igb_buffer) * tx_ring->count;
1815 tx_ring->buffer_info = vmalloc(size);
1816 if (!tx_ring->buffer_info)
1818 memset(tx_ring->buffer_info, 0, size);
1820 /* round up to nearest 4K */
1821 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
1822 tx_ring->size = ALIGN(tx_ring->size, 4096);
1824 tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
1830 tx_ring->adapter = adapter;
1831 tx_ring->next_to_use = 0;
1832 tx_ring->next_to_clean = 0;
1836 vfree(tx_ring->buffer_info);
1837 dev_err(&adapter->pdev->dev,
1838 "Unable to allocate memory for the transmit descriptor ring\n");
1843 * igb_setup_all_tx_resources - wrapper to allocate Tx resources
1844 * (Descriptors) for all queues
1845 * @adapter: board private structure
1847 * Return 0 on success, negative on failure
1849 static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
1854 for (i = 0; i < adapter->num_tx_queues; i++) {
1855 err = igb_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1857 dev_err(&adapter->pdev->dev,
1858 "Allocation for Tx Queue %u failed\n", i);
1859 for (i--; i >= 0; i--)
1860 igb_free_tx_resources(&adapter->tx_ring[i]);
1865 for (i = 0; i < IGB_MAX_TX_QUEUES; i++) {
1866 r_idx = i % adapter->num_tx_queues;
1867 adapter->multi_tx_table[i] = &adapter->tx_ring[r_idx];
1873 * igb_configure_tx - Configure transmit Unit after Reset
1874 * @adapter: board private structure
1876 * Configure the Tx unit of the MAC after a reset.
1878 static void igb_configure_tx(struct igb_adapter *adapter)
1881 struct e1000_hw *hw = &adapter->hw;
1886 for (i = 0; i < adapter->num_tx_queues; i++) {
1887 struct igb_ring *ring = &adapter->tx_ring[i];
1889 wr32(E1000_TDLEN(j),
1890 ring->count * sizeof(union e1000_adv_tx_desc));
1892 wr32(E1000_TDBAL(j),
1893 tdba & 0x00000000ffffffffULL);
1894 wr32(E1000_TDBAH(j), tdba >> 32);
1896 ring->head = E1000_TDH(j);
1897 ring->tail = E1000_TDT(j);
1898 writel(0, hw->hw_addr + ring->tail);
1899 writel(0, hw->hw_addr + ring->head);
1900 txdctl = rd32(E1000_TXDCTL(j));
1901 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1902 wr32(E1000_TXDCTL(j), txdctl);
1904 /* Turn off Relaxed Ordering on head write-backs. The
1905 * writebacks MUST be delivered in order or it will
1906 * completely screw up our bookeeping.
1908 txctrl = rd32(E1000_DCA_TXCTRL(j));
1909 txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1910 wr32(E1000_DCA_TXCTRL(j), txctrl);
1913 /* disable queue 0 to prevent tail bump w/o re-configuration */
1914 if (adapter->vfs_allocated_count)
1915 wr32(E1000_TXDCTL(0), 0);
1917 /* Program the Transmit Control Register */
1918 tctl = rd32(E1000_TCTL);
1919 tctl &= ~E1000_TCTL_CT;
1920 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1921 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1923 igb_config_collision_dist(hw);
1925 /* Setup Transmit Descriptor Settings for eop descriptor */
1926 adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_RS;
1928 /* Enable transmits */
1929 tctl |= E1000_TCTL_EN;
1931 wr32(E1000_TCTL, tctl);
1935 * igb_setup_rx_resources - allocate Rx resources (Descriptors)
1936 * @adapter: board private structure
1937 * @rx_ring: rx descriptor ring (for a specific queue) to setup
1939 * Returns 0 on success, negative on failure
1941 int igb_setup_rx_resources(struct igb_adapter *adapter,
1942 struct igb_ring *rx_ring)
1944 struct pci_dev *pdev = adapter->pdev;
1947 size = sizeof(struct igb_buffer) * rx_ring->count;
1948 rx_ring->buffer_info = vmalloc(size);
1949 if (!rx_ring->buffer_info)
1951 memset(rx_ring->buffer_info, 0, size);
1953 desc_len = sizeof(union e1000_adv_rx_desc);
1955 /* Round up to nearest 4K */
1956 rx_ring->size = rx_ring->count * desc_len;
1957 rx_ring->size = ALIGN(rx_ring->size, 4096);
1959 rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size,
1965 rx_ring->next_to_clean = 0;
1966 rx_ring->next_to_use = 0;
1968 rx_ring->adapter = adapter;
1973 vfree(rx_ring->buffer_info);
1974 dev_err(&adapter->pdev->dev, "Unable to allocate memory for "
1975 "the receive descriptor ring\n");
1980 * igb_setup_all_rx_resources - wrapper to allocate Rx resources
1981 * (Descriptors) for all queues
1982 * @adapter: board private structure
1984 * Return 0 on success, negative on failure
1986 static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
1990 for (i = 0; i < adapter->num_rx_queues; i++) {
1991 err = igb_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1993 dev_err(&adapter->pdev->dev,
1994 "Allocation for Rx Queue %u failed\n", i);
1995 for (i--; i >= 0; i--)
1996 igb_free_rx_resources(&adapter->rx_ring[i]);
2005 * igb_setup_rctl - configure the receive control registers
2006 * @adapter: Board private structure
2008 static void igb_setup_rctl(struct igb_adapter *adapter)
2010 struct e1000_hw *hw = &adapter->hw;
2015 rctl = rd32(E1000_RCTL);
2017 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2018 rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
2020 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
2021 (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2024 * enable stripping of CRC. It's unlikely this will break BMC
2025 * redirection as it did with e1000. Newer features require
2026 * that the HW strips the CRC.
2028 rctl |= E1000_RCTL_SECRC;
2031 * disable store bad packets and clear size bits.
2033 rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
2035 /* enable LPE when to prevent packets larger than max_frame_size */
2036 rctl |= E1000_RCTL_LPE;
2038 /* Setup buffer sizes */
2039 switch (adapter->rx_buffer_len) {
2040 case IGB_RXBUFFER_256:
2041 rctl |= E1000_RCTL_SZ_256;
2043 case IGB_RXBUFFER_512:
2044 rctl |= E1000_RCTL_SZ_512;
2047 srrctl = ALIGN(adapter->rx_buffer_len, 1024)
2048 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
2052 /* 82575 and greater support packet-split where the protocol
2053 * header is placed in skb->data and the packet data is
2054 * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2055 * In the case of a non-split, skb->data is linearly filled,
2056 * followed by the page buffers. Therefore, skb->data is
2057 * sized to hold the largest protocol header.
2059 /* allocations using alloc_page take too long for regular MTU
2060 * so only enable packet split for jumbo frames */
2061 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2062 adapter->rx_ps_hdr_size = IGB_RXBUFFER_128;
2063 srrctl |= adapter->rx_ps_hdr_size <<
2064 E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
2065 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2067 adapter->rx_ps_hdr_size = 0;
2068 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
2071 /* Attention!!! For SR-IOV PF driver operations you must enable
2072 * queue drop for all VF and PF queues to prevent head of line blocking
2073 * if an un-trusted VF does not provide descriptors to hardware.
2075 if (adapter->vfs_allocated_count) {
2078 /* set all queue drop enable bits */
2079 wr32(E1000_QDE, ALL_QUEUES);
2080 srrctl |= E1000_SRRCTL_DROP_EN;
2082 /* disable queue 0 to prevent tail write w/o re-config */
2083 wr32(E1000_RXDCTL(0), 0);
2085 vmolr = rd32(E1000_VMOLR(adapter->vfs_allocated_count));
2086 if (rctl & E1000_RCTL_LPE)
2087 vmolr |= E1000_VMOLR_LPE;
2088 if (adapter->num_rx_queues > 1)
2089 vmolr |= E1000_VMOLR_RSSE;
2090 wr32(E1000_VMOLR(adapter->vfs_allocated_count), vmolr);
2093 for (i = 0; i < adapter->num_rx_queues; i++) {
2094 int j = adapter->rx_ring[i].reg_idx;
2095 wr32(E1000_SRRCTL(j), srrctl);
2098 wr32(E1000_RCTL, rctl);
2102 * igb_rlpml_set - set maximum receive packet size
2103 * @adapter: board private structure
2105 * Configure maximum receivable packet size.
2107 static void igb_rlpml_set(struct igb_adapter *adapter)
2109 u32 max_frame_size = adapter->max_frame_size;
2110 struct e1000_hw *hw = &adapter->hw;
2111 u16 pf_id = adapter->vfs_allocated_count;
2114 max_frame_size += VLAN_TAG_SIZE;
2116 /* if vfs are enabled we set RLPML to the largest possible request
2117 * size and set the VMOLR RLPML to the size we need */
2119 igb_set_vf_rlpml(adapter, max_frame_size, pf_id);
2120 max_frame_size = MAX_STD_JUMBO_FRAME_SIZE + VLAN_TAG_SIZE;
2123 wr32(E1000_RLPML, max_frame_size);
2127 * igb_configure_vt_default_pool - Configure VT default pool
2128 * @adapter: board private structure
2130 * Configure the default pool
2132 static void igb_configure_vt_default_pool(struct igb_adapter *adapter)
2134 struct e1000_hw *hw = &adapter->hw;
2135 u16 pf_id = adapter->vfs_allocated_count;
2138 /* not in sr-iov mode - do nothing */
2142 vtctl = rd32(E1000_VT_CTL);
2143 vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK |
2144 E1000_VT_CTL_DISABLE_DEF_POOL);
2145 vtctl |= pf_id << E1000_VT_CTL_DEFAULT_POOL_SHIFT;
2146 wr32(E1000_VT_CTL, vtctl);
2150 * igb_configure_rx - Configure receive Unit after Reset
2151 * @adapter: board private structure
2153 * Configure the Rx unit of the MAC after a reset.
2155 static void igb_configure_rx(struct igb_adapter *adapter)
2158 struct e1000_hw *hw = &adapter->hw;
2163 /* disable receives while setting up the descriptors */
2164 rctl = rd32(E1000_RCTL);
2165 wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
2169 if (adapter->itr_setting > 3)
2170 wr32(E1000_ITR, adapter->itr);
2172 /* Setup the HW Rx Head and Tail Descriptor Pointers and
2173 * the Base and Length of the Rx Descriptor Ring */
2174 for (i = 0; i < adapter->num_rx_queues; i++) {
2175 struct igb_ring *ring = &adapter->rx_ring[i];
2176 int j = ring->reg_idx;
2178 wr32(E1000_RDBAL(j),
2179 rdba & 0x00000000ffffffffULL);
2180 wr32(E1000_RDBAH(j), rdba >> 32);
2181 wr32(E1000_RDLEN(j),
2182 ring->count * sizeof(union e1000_adv_rx_desc));
2184 ring->head = E1000_RDH(j);
2185 ring->tail = E1000_RDT(j);
2186 writel(0, hw->hw_addr + ring->tail);
2187 writel(0, hw->hw_addr + ring->head);
2189 rxdctl = rd32(E1000_RXDCTL(j));
2190 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
2191 rxdctl &= 0xFFF00000;
2192 rxdctl |= IGB_RX_PTHRESH;
2193 rxdctl |= IGB_RX_HTHRESH << 8;
2194 rxdctl |= IGB_RX_WTHRESH << 16;
2195 wr32(E1000_RXDCTL(j), rxdctl);
2198 if (adapter->num_rx_queues > 1) {
2207 get_random_bytes(&random[0], 40);
2209 if (hw->mac.type >= e1000_82576)
2213 for (j = 0; j < (32 * 4); j++) {
2215 adapter->rx_ring[(j % adapter->num_rx_queues)].reg_idx << shift;
2218 hw->hw_addr + E1000_RETA(0) + (j & ~3));
2220 if (adapter->vfs_allocated_count)
2221 mrqc = E1000_MRQC_ENABLE_VMDQ_RSS_2Q;
2223 mrqc = E1000_MRQC_ENABLE_RSS_4Q;
2225 /* Fill out hash function seeds */
2226 for (j = 0; j < 10; j++)
2227 array_wr32(E1000_RSSRK(0), j, random[j]);
2229 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
2230 E1000_MRQC_RSS_FIELD_IPV4_TCP);
2231 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6 |
2232 E1000_MRQC_RSS_FIELD_IPV6_TCP);
2233 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4_UDP |
2234 E1000_MRQC_RSS_FIELD_IPV6_UDP);
2235 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
2236 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
2238 wr32(E1000_MRQC, mrqc);
2239 } else if (adapter->vfs_allocated_count) {
2240 /* Enable multi-queue for sr-iov */
2241 wr32(E1000_MRQC, E1000_MRQC_ENABLE_VMDQ);
2244 /* Enable Receive Checksum Offload for TCP and UDP */
2245 rxcsum = rd32(E1000_RXCSUM);
2246 /* Disable raw packet checksumming */
2247 rxcsum |= E1000_RXCSUM_PCSD;
2249 if (adapter->hw.mac.type == e1000_82576)
2250 /* Enable Receive Checksum Offload for SCTP */
2251 rxcsum |= E1000_RXCSUM_CRCOFL;
2253 /* Don't need to set TUOFL or IPOFL, they default to 1 */
2254 wr32(E1000_RXCSUM, rxcsum);
2256 /* Set the default pool for the PF's first queue */
2257 igb_configure_vt_default_pool(adapter);
2259 igb_rlpml_set(adapter);
2261 /* Enable Receives */
2262 wr32(E1000_RCTL, rctl);
2266 * igb_free_tx_resources - Free Tx Resources per Queue
2267 * @tx_ring: Tx descriptor ring for a specific queue
2269 * Free all transmit software resources
2271 void igb_free_tx_resources(struct igb_ring *tx_ring)
2273 struct pci_dev *pdev = tx_ring->adapter->pdev;
2275 igb_clean_tx_ring(tx_ring);
2277 vfree(tx_ring->buffer_info);
2278 tx_ring->buffer_info = NULL;
2280 pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
2282 tx_ring->desc = NULL;
2286 * igb_free_all_tx_resources - Free Tx Resources for All Queues
2287 * @adapter: board private structure
2289 * Free all transmit software resources
2291 static void igb_free_all_tx_resources(struct igb_adapter *adapter)
2295 for (i = 0; i < adapter->num_tx_queues; i++)
2296 igb_free_tx_resources(&adapter->tx_ring[i]);
2299 static void igb_unmap_and_free_tx_resource(struct igb_adapter *adapter,
2300 struct igb_buffer *buffer_info)
2302 buffer_info->dma = 0;
2303 if (buffer_info->skb) {
2304 skb_dma_unmap(&adapter->pdev->dev, buffer_info->skb,
2306 dev_kfree_skb_any(buffer_info->skb);
2307 buffer_info->skb = NULL;
2309 buffer_info->time_stamp = 0;
2310 /* buffer_info must be completely set up in the transmit path */
2314 * igb_clean_tx_ring - Free Tx Buffers
2315 * @tx_ring: ring to be cleaned
2317 static void igb_clean_tx_ring(struct igb_ring *tx_ring)
2319 struct igb_adapter *adapter = tx_ring->adapter;
2320 struct igb_buffer *buffer_info;
2324 if (!tx_ring->buffer_info)
2326 /* Free all the Tx ring sk_buffs */
2328 for (i = 0; i < tx_ring->count; i++) {
2329 buffer_info = &tx_ring->buffer_info[i];
2330 igb_unmap_and_free_tx_resource(adapter, buffer_info);
2333 size = sizeof(struct igb_buffer) * tx_ring->count;
2334 memset(tx_ring->buffer_info, 0, size);
2336 /* Zero out the descriptor ring */
2338 memset(tx_ring->desc, 0, tx_ring->size);
2340 tx_ring->next_to_use = 0;
2341 tx_ring->next_to_clean = 0;
2343 writel(0, adapter->hw.hw_addr + tx_ring->head);
2344 writel(0, adapter->hw.hw_addr + tx_ring->tail);
2348 * igb_clean_all_tx_rings - Free Tx Buffers for all queues
2349 * @adapter: board private structure
2351 static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
2355 for (i = 0; i < adapter->num_tx_queues; i++)
2356 igb_clean_tx_ring(&adapter->tx_ring[i]);
2360 * igb_free_rx_resources - Free Rx Resources
2361 * @rx_ring: ring to clean the resources from
2363 * Free all receive software resources
2365 void igb_free_rx_resources(struct igb_ring *rx_ring)
2367 struct pci_dev *pdev = rx_ring->adapter->pdev;
2369 igb_clean_rx_ring(rx_ring);
2371 vfree(rx_ring->buffer_info);
2372 rx_ring->buffer_info = NULL;
2374 pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
2376 rx_ring->desc = NULL;
2380 * igb_free_all_rx_resources - Free Rx Resources for All Queues
2381 * @adapter: board private structure
2383 * Free all receive software resources
2385 static void igb_free_all_rx_resources(struct igb_adapter *adapter)
2389 for (i = 0; i < adapter->num_rx_queues; i++)
2390 igb_free_rx_resources(&adapter->rx_ring[i]);
2394 * igb_clean_rx_ring - Free Rx Buffers per Queue
2395 * @rx_ring: ring to free buffers from
2397 static void igb_clean_rx_ring(struct igb_ring *rx_ring)
2399 struct igb_adapter *adapter = rx_ring->adapter;
2400 struct igb_buffer *buffer_info;
2401 struct pci_dev *pdev = adapter->pdev;
2405 if (!rx_ring->buffer_info)
2407 /* Free all the Rx ring sk_buffs */
2408 for (i = 0; i < rx_ring->count; i++) {
2409 buffer_info = &rx_ring->buffer_info[i];
2410 if (buffer_info->dma) {
2411 if (adapter->rx_ps_hdr_size)
2412 pci_unmap_single(pdev, buffer_info->dma,
2413 adapter->rx_ps_hdr_size,
2414 PCI_DMA_FROMDEVICE);
2416 pci_unmap_single(pdev, buffer_info->dma,
2417 adapter->rx_buffer_len,
2418 PCI_DMA_FROMDEVICE);
2419 buffer_info->dma = 0;
2422 if (buffer_info->skb) {
2423 dev_kfree_skb(buffer_info->skb);
2424 buffer_info->skb = NULL;
2426 if (buffer_info->page) {
2427 if (buffer_info->page_dma)
2428 pci_unmap_page(pdev, buffer_info->page_dma,
2430 PCI_DMA_FROMDEVICE);
2431 put_page(buffer_info->page);
2432 buffer_info->page = NULL;
2433 buffer_info->page_dma = 0;
2434 buffer_info->page_offset = 0;
2438 size = sizeof(struct igb_buffer) * rx_ring->count;
2439 memset(rx_ring->buffer_info, 0, size);
2441 /* Zero out the descriptor ring */
2442 memset(rx_ring->desc, 0, rx_ring->size);
2444 rx_ring->next_to_clean = 0;
2445 rx_ring->next_to_use = 0;
2447 writel(0, adapter->hw.hw_addr + rx_ring->head);
2448 writel(0, adapter->hw.hw_addr + rx_ring->tail);
2452 * igb_clean_all_rx_rings - Free Rx Buffers for all queues
2453 * @adapter: board private structure
2455 static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
2459 for (i = 0; i < adapter->num_rx_queues; i++)
2460 igb_clean_rx_ring(&adapter->rx_ring[i]);
2464 * igb_set_mac - Change the Ethernet Address of the NIC
2465 * @netdev: network interface device structure
2466 * @p: pointer to an address structure
2468 * Returns 0 on success, negative on failure
2470 static int igb_set_mac(struct net_device *netdev, void *p)
2472 struct igb_adapter *adapter = netdev_priv(netdev);
2473 struct e1000_hw *hw = &adapter->hw;
2474 struct sockaddr *addr = p;
2476 if (!is_valid_ether_addr(addr->sa_data))
2477 return -EADDRNOTAVAIL;
2479 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2480 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
2482 hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
2484 igb_set_rah_pool(hw, adapter->vfs_allocated_count, 0);
2490 * igb_set_multi - Multicast and Promiscuous mode set
2491 * @netdev: network interface device structure
2493 * The set_multi entry point is called whenever the multicast address
2494 * list or the network interface flags are updated. This routine is
2495 * responsible for configuring the hardware for proper multicast,
2496 * promiscuous mode, and all-multi behavior.
2498 static void igb_set_multi(struct net_device *netdev)
2500 struct igb_adapter *adapter = netdev_priv(netdev);
2501 struct e1000_hw *hw = &adapter->hw;
2502 struct e1000_mac_info *mac = &hw->mac;
2503 struct dev_mc_list *mc_ptr;
2504 u8 *mta_list = NULL;
2508 /* Check for Promiscuous and All Multicast modes */
2510 rctl = rd32(E1000_RCTL);
2512 if (netdev->flags & IFF_PROMISC) {
2513 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2514 rctl &= ~E1000_RCTL_VFE;
2516 if (netdev->flags & IFF_ALLMULTI) {
2517 rctl |= E1000_RCTL_MPE;
2518 rctl &= ~E1000_RCTL_UPE;
2520 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2521 rctl |= E1000_RCTL_VFE;
2523 wr32(E1000_RCTL, rctl);
2525 if (netdev->mc_count) {
2526 mta_list = kzalloc(netdev->mc_count * 6, GFP_ATOMIC);
2528 dev_err(&adapter->pdev->dev,
2529 "failed to allocate multicast filter list\n");
2534 /* The shared function expects a packed array of only addresses. */
2535 mc_ptr = netdev->mc_list;
2537 for (i = 0; i < netdev->mc_count; i++) {
2540 memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr, ETH_ALEN);
2541 mc_ptr = mc_ptr->next;
2543 igb_update_mc_addr_list(hw, mta_list, i,
2544 adapter->vfs_allocated_count + 1,
2545 mac->rar_entry_count);
2547 igb_set_mc_list_pools(adapter, i, mac->rar_entry_count);
2548 igb_restore_vf_multicasts(adapter);
2553 /* Need to wait a few seconds after link up to get diagnostic information from
2555 static void igb_update_phy_info(unsigned long data)
2557 struct igb_adapter *adapter = (struct igb_adapter *) data;
2558 igb_get_phy_info(&adapter->hw);
2562 * igb_has_link - check shared code for link and determine up/down
2563 * @adapter: pointer to driver private info
2565 static bool igb_has_link(struct igb_adapter *adapter)
2567 struct e1000_hw *hw = &adapter->hw;
2568 bool link_active = false;
2571 /* get_link_status is set on LSC (link status) interrupt or
2572 * rx sequence error interrupt. get_link_status will stay
2573 * false until the e1000_check_for_link establishes link
2574 * for copper adapters ONLY
2576 switch (hw->phy.media_type) {
2577 case e1000_media_type_copper:
2578 if (hw->mac.get_link_status) {
2579 ret_val = hw->mac.ops.check_for_link(hw);
2580 link_active = !hw->mac.get_link_status;
2585 case e1000_media_type_fiber:
2586 ret_val = hw->mac.ops.check_for_link(hw);
2587 link_active = !!(rd32(E1000_STATUS) & E1000_STATUS_LU);
2589 case e1000_media_type_internal_serdes:
2590 ret_val = hw->mac.ops.check_for_link(hw);
2591 link_active = hw->mac.serdes_has_link;
2594 case e1000_media_type_unknown:
2602 * igb_watchdog - Timer Call-back
2603 * @data: pointer to adapter cast into an unsigned long
2605 static void igb_watchdog(unsigned long data)
2607 struct igb_adapter *adapter = (struct igb_adapter *)data;
2608 /* Do the rest outside of interrupt context */
2609 schedule_work(&adapter->watchdog_task);
2612 static void igb_watchdog_task(struct work_struct *work)
2614 struct igb_adapter *adapter = container_of(work,
2615 struct igb_adapter, watchdog_task);
2616 struct e1000_hw *hw = &adapter->hw;
2617 struct net_device *netdev = adapter->netdev;
2618 struct igb_ring *tx_ring = adapter->tx_ring;
2623 link = igb_has_link(adapter);
2624 if ((netif_carrier_ok(netdev)) && link)
2628 if (!netif_carrier_ok(netdev)) {
2630 hw->mac.ops.get_speed_and_duplex(&adapter->hw,
2631 &adapter->link_speed,
2632 &adapter->link_duplex);
2634 ctrl = rd32(E1000_CTRL);
2635 /* Links status message must follow this format */
2636 printk(KERN_INFO "igb: %s NIC Link is Up %d Mbps %s, "
2637 "Flow Control: %s\n",
2639 adapter->link_speed,
2640 adapter->link_duplex == FULL_DUPLEX ?
2641 "Full Duplex" : "Half Duplex",
2642 ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2643 E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2644 E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2645 E1000_CTRL_TFCE) ? "TX" : "None")));
2647 /* tweak tx_queue_len according to speed/duplex and
2648 * adjust the timeout factor */
2649 netdev->tx_queue_len = adapter->tx_queue_len;
2650 adapter->tx_timeout_factor = 1;
2651 switch (adapter->link_speed) {
2653 netdev->tx_queue_len = 10;
2654 adapter->tx_timeout_factor = 14;
2657 netdev->tx_queue_len = 100;
2658 /* maybe add some timeout factor ? */
2662 netif_carrier_on(netdev);
2664 igb_ping_all_vfs(adapter);
2666 /* link state has changed, schedule phy info update */
2667 if (!test_bit(__IGB_DOWN, &adapter->state))
2668 mod_timer(&adapter->phy_info_timer,
2669 round_jiffies(jiffies + 2 * HZ));
2672 if (netif_carrier_ok(netdev)) {
2673 adapter->link_speed = 0;
2674 adapter->link_duplex = 0;
2675 /* Links status message must follow this format */
2676 printk(KERN_INFO "igb: %s NIC Link is Down\n",
2678 netif_carrier_off(netdev);
2680 igb_ping_all_vfs(adapter);
2682 /* link state has changed, schedule phy info update */
2683 if (!test_bit(__IGB_DOWN, &adapter->state))
2684 mod_timer(&adapter->phy_info_timer,
2685 round_jiffies(jiffies + 2 * HZ));
2690 igb_update_stats(adapter);
2692 hw->mac.tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2693 adapter->tpt_old = adapter->stats.tpt;
2694 hw->mac.collision_delta = adapter->stats.colc - adapter->colc_old;
2695 adapter->colc_old = adapter->stats.colc;
2697 adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
2698 adapter->gorc_old = adapter->stats.gorc;
2699 adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
2700 adapter->gotc_old = adapter->stats.gotc;
2702 igb_update_adaptive(&adapter->hw);
2704 if (!netif_carrier_ok(netdev)) {
2705 if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) {
2706 /* We've lost link, so the controller stops DMA,
2707 * but we've got queued Tx work that's never going
2708 * to get done, so reset controller to flush Tx.
2709 * (Do the reset outside of interrupt context). */
2710 adapter->tx_timeout_count++;
2711 schedule_work(&adapter->reset_task);
2712 /* return immediately since reset is imminent */
2717 /* Cause software interrupt to ensure rx ring is cleaned */
2718 if (adapter->msix_entries) {
2719 for (i = 0; i < adapter->num_rx_queues; i++)
2720 eics |= adapter->rx_ring[i].eims_value;
2721 wr32(E1000_EICS, eics);
2723 wr32(E1000_ICS, E1000_ICS_RXDMT0);
2726 /* Force detection of hung controller every watchdog period */
2727 tx_ring->detect_tx_hung = true;
2729 /* Reset the timer */
2730 if (!test_bit(__IGB_DOWN, &adapter->state))
2731 mod_timer(&adapter->watchdog_timer,
2732 round_jiffies(jiffies + 2 * HZ));
2735 enum latency_range {
2739 latency_invalid = 255
2744 * igb_update_ring_itr - update the dynamic ITR value based on packet size
2746 * Stores a new ITR value based on strictly on packet size. This
2747 * algorithm is less sophisticated than that used in igb_update_itr,
2748 * due to the difficulty of synchronizing statistics across multiple
2749 * receive rings. The divisors and thresholds used by this fuction
2750 * were determined based on theoretical maximum wire speed and testing
2751 * data, in order to minimize response time while increasing bulk
2753 * This functionality is controlled by the InterruptThrottleRate module
2754 * parameter (see igb_param.c)
2755 * NOTE: This function is called only when operating in a multiqueue
2756 * receive environment.
2757 * @rx_ring: pointer to ring
2759 static void igb_update_ring_itr(struct igb_ring *rx_ring)
2761 int new_val = rx_ring->itr_val;
2762 int avg_wire_size = 0;
2763 struct igb_adapter *adapter = rx_ring->adapter;
2765 if (!rx_ring->total_packets)
2766 goto clear_counts; /* no packets, so don't do anything */
2768 /* For non-gigabit speeds, just fix the interrupt rate at 4000
2769 * ints/sec - ITR timer value of 120 ticks.
2771 if (adapter->link_speed != SPEED_1000) {
2775 avg_wire_size = rx_ring->total_bytes / rx_ring->total_packets;
2777 /* Add 24 bytes to size to account for CRC, preamble, and gap */
2778 avg_wire_size += 24;
2780 /* Don't starve jumbo frames */
2781 avg_wire_size = min(avg_wire_size, 3000);
2783 /* Give a little boost to mid-size frames */
2784 if ((avg_wire_size > 300) && (avg_wire_size < 1200))
2785 new_val = avg_wire_size / 3;
2787 new_val = avg_wire_size / 2;
2790 if (new_val != rx_ring->itr_val) {
2791 rx_ring->itr_val = new_val;
2792 rx_ring->set_itr = 1;
2795 rx_ring->total_bytes = 0;
2796 rx_ring->total_packets = 0;
2800 * igb_update_itr - update the dynamic ITR value based on statistics
2801 * Stores a new ITR value based on packets and byte
2802 * counts during the last interrupt. The advantage of per interrupt
2803 * computation is faster updates and more accurate ITR for the current
2804 * traffic pattern. Constants in this function were computed
2805 * based on theoretical maximum wire speed and thresholds were set based
2806 * on testing data as well as attempting to minimize response time
2807 * while increasing bulk throughput.
2808 * this functionality is controlled by the InterruptThrottleRate module
2809 * parameter (see igb_param.c)
2810 * NOTE: These calculations are only valid when operating in a single-
2811 * queue environment.
2812 * @adapter: pointer to adapter
2813 * @itr_setting: current adapter->itr
2814 * @packets: the number of packets during this measurement interval
2815 * @bytes: the number of bytes during this measurement interval
2817 static unsigned int igb_update_itr(struct igb_adapter *adapter, u16 itr_setting,
2818 int packets, int bytes)
2820 unsigned int retval = itr_setting;
2823 goto update_itr_done;
2825 switch (itr_setting) {
2826 case lowest_latency:
2827 /* handle TSO and jumbo frames */
2828 if (bytes/packets > 8000)
2829 retval = bulk_latency;
2830 else if ((packets < 5) && (bytes > 512))
2831 retval = low_latency;
2833 case low_latency: /* 50 usec aka 20000 ints/s */
2834 if (bytes > 10000) {
2835 /* this if handles the TSO accounting */
2836 if (bytes/packets > 8000) {
2837 retval = bulk_latency;
2838 } else if ((packets < 10) || ((bytes/packets) > 1200)) {
2839 retval = bulk_latency;
2840 } else if ((packets > 35)) {
2841 retval = lowest_latency;
2843 } else if (bytes/packets > 2000) {
2844 retval = bulk_latency;
2845 } else if (packets <= 2 && bytes < 512) {
2846 retval = lowest_latency;
2849 case bulk_latency: /* 250 usec aka 4000 ints/s */
2850 if (bytes > 25000) {
2852 retval = low_latency;
2853 } else if (bytes < 1500) {
2854 retval = low_latency;
2863 static void igb_set_itr(struct igb_adapter *adapter)
2866 u32 new_itr = adapter->itr;
2868 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2869 if (adapter->link_speed != SPEED_1000) {
2875 adapter->rx_itr = igb_update_itr(adapter,
2877 adapter->rx_ring->total_packets,
2878 adapter->rx_ring->total_bytes);
2880 if (adapter->rx_ring->buddy) {
2881 adapter->tx_itr = igb_update_itr(adapter,
2883 adapter->tx_ring->total_packets,
2884 adapter->tx_ring->total_bytes);
2885 current_itr = max(adapter->rx_itr, adapter->tx_itr);
2887 current_itr = adapter->rx_itr;
2890 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2891 if (adapter->itr_setting == 3 && current_itr == lowest_latency)
2892 current_itr = low_latency;
2894 switch (current_itr) {
2895 /* counts and packets in update_itr are dependent on these numbers */
2896 case lowest_latency:
2897 new_itr = 56; /* aka 70,000 ints/sec */
2900 new_itr = 196; /* aka 20,000 ints/sec */
2903 new_itr = 980; /* aka 4,000 ints/sec */
2910 adapter->rx_ring->total_bytes = 0;
2911 adapter->rx_ring->total_packets = 0;
2912 if (adapter->rx_ring->buddy) {
2913 adapter->rx_ring->buddy->total_bytes = 0;
2914 adapter->rx_ring->buddy->total_packets = 0;
2917 if (new_itr != adapter->itr) {
2918 /* this attempts to bias the interrupt rate towards Bulk
2919 * by adding intermediate steps when interrupt rate is
2921 new_itr = new_itr > adapter->itr ?
2922 max((new_itr * adapter->itr) /
2923 (new_itr + (adapter->itr >> 2)), new_itr) :
2925 /* Don't write the value here; it resets the adapter's
2926 * internal timer, and causes us to delay far longer than
2927 * we should between interrupts. Instead, we write the ITR
2928 * value at the beginning of the next interrupt so the timing
2929 * ends up being correct.
2931 adapter->itr = new_itr;
2932 adapter->rx_ring->itr_val = new_itr;
2933 adapter->rx_ring->set_itr = 1;
2940 #define IGB_TX_FLAGS_CSUM 0x00000001
2941 #define IGB_TX_FLAGS_VLAN 0x00000002
2942 #define IGB_TX_FLAGS_TSO 0x00000004
2943 #define IGB_TX_FLAGS_IPV4 0x00000008
2944 #define IGB_TX_FLAGS_TSTAMP 0x00000010
2945 #define IGB_TX_FLAGS_VLAN_MASK 0xffff0000
2946 #define IGB_TX_FLAGS_VLAN_SHIFT 16
2948 static inline int igb_tso_adv(struct igb_adapter *adapter,
2949 struct igb_ring *tx_ring,
2950 struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
2952 struct e1000_adv_tx_context_desc *context_desc;
2955 struct igb_buffer *buffer_info;
2956 u32 info = 0, tu_cmd = 0;
2957 u32 mss_l4len_idx, l4len;
2960 if (skb_header_cloned(skb)) {
2961 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2966 l4len = tcp_hdrlen(skb);
2969 if (skb->protocol == htons(ETH_P_IP)) {
2970 struct iphdr *iph = ip_hdr(skb);
2973 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2977 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
2978 ipv6_hdr(skb)->payload_len = 0;
2979 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2980 &ipv6_hdr(skb)->daddr,
2984 i = tx_ring->next_to_use;
2986 buffer_info = &tx_ring->buffer_info[i];
2987 context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
2988 /* VLAN MACLEN IPLEN */
2989 if (tx_flags & IGB_TX_FLAGS_VLAN)
2990 info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
2991 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2992 *hdr_len += skb_network_offset(skb);
2993 info |= skb_network_header_len(skb);
2994 *hdr_len += skb_network_header_len(skb);
2995 context_desc->vlan_macip_lens = cpu_to_le32(info);
2997 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
2998 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
3000 if (skb->protocol == htons(ETH_P_IP))
3001 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
3002 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
3004 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
3007 mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
3008 mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
3010 /* For 82575, context index must be unique per ring. */
3011 if (adapter->flags & IGB_FLAG_NEED_CTX_IDX)
3012 mss_l4len_idx |= tx_ring->queue_index << 4;
3014 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
3015 context_desc->seqnum_seed = 0;
3017 buffer_info->time_stamp = jiffies;
3018 buffer_info->next_to_watch = i;
3019 buffer_info->dma = 0;
3021 if (i == tx_ring->count)
3024 tx_ring->next_to_use = i;
3029 static inline bool igb_tx_csum_adv(struct igb_adapter *adapter,
3030 struct igb_ring *tx_ring,
3031 struct sk_buff *skb, u32 tx_flags)
3033 struct e1000_adv_tx_context_desc *context_desc;
3035 struct igb_buffer *buffer_info;
3036 u32 info = 0, tu_cmd = 0;
3038 if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
3039 (tx_flags & IGB_TX_FLAGS_VLAN)) {
3040 i = tx_ring->next_to_use;
3041 buffer_info = &tx_ring->buffer_info[i];
3042 context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
3044 if (tx_flags & IGB_TX_FLAGS_VLAN)
3045 info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
3046 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
3047 if (skb->ip_summed == CHECKSUM_PARTIAL)
3048 info |= skb_network_header_len(skb);
3050 context_desc->vlan_macip_lens = cpu_to_le32(info);
3052 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
3054 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3057 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3058 const struct vlan_ethhdr *vhdr =
3059 (const struct vlan_ethhdr*)skb->data;
3061 protocol = vhdr->h_vlan_encapsulated_proto;
3063 protocol = skb->protocol;
3067 case cpu_to_be16(ETH_P_IP):
3068 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
3069 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
3070 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
3071 else if (ip_hdr(skb)->protocol == IPPROTO_SCTP)
3072 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_SCTP;
3074 case cpu_to_be16(ETH_P_IPV6):
3075 /* XXX what about other V6 headers?? */
3076 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3077 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
3078 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_SCTP)
3079 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_SCTP;
3082 if (unlikely(net_ratelimit()))
3083 dev_warn(&adapter->pdev->dev,
3084 "partial checksum but proto=%x!\n",
3090 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
3091 context_desc->seqnum_seed = 0;
3092 if (adapter->flags & IGB_FLAG_NEED_CTX_IDX)
3093 context_desc->mss_l4len_idx =
3094 cpu_to_le32(tx_ring->queue_index << 4);
3096 context_desc->mss_l4len_idx = 0;
3098 buffer_info->time_stamp = jiffies;
3099 buffer_info->next_to_watch = i;
3100 buffer_info->dma = 0;
3103 if (i == tx_ring->count)
3105 tx_ring->next_to_use = i;
3112 #define IGB_MAX_TXD_PWR 16
3113 #define IGB_MAX_DATA_PER_TXD (1<<IGB_MAX_TXD_PWR)
3115 static inline int igb_tx_map_adv(struct igb_adapter *adapter,
3116 struct igb_ring *tx_ring, struct sk_buff *skb,
3119 struct igb_buffer *buffer_info;
3120 unsigned int len = skb_headlen(skb);
3121 unsigned int count = 0, i;
3125 i = tx_ring->next_to_use;
3127 if (skb_dma_map(&adapter->pdev->dev, skb, DMA_TO_DEVICE)) {
3128 dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
3132 map = skb_shinfo(skb)->dma_maps;
3134 buffer_info = &tx_ring->buffer_info[i];
3135 BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
3136 buffer_info->length = len;
3137 /* set time_stamp *before* dma to help avoid a possible race */
3138 buffer_info->time_stamp = jiffies;
3139 buffer_info->next_to_watch = i;
3140 buffer_info->dma = map[count];
3143 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
3144 struct skb_frag_struct *frag;
3147 if (i == tx_ring->count)
3150 frag = &skb_shinfo(skb)->frags[f];
3153 buffer_info = &tx_ring->buffer_info[i];
3154 BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
3155 buffer_info->length = len;
3156 buffer_info->time_stamp = jiffies;
3157 buffer_info->next_to_watch = i;
3158 buffer_info->dma = map[count];
3162 tx_ring->buffer_info[i].skb = skb;
3163 tx_ring->buffer_info[first].next_to_watch = i;
3168 static inline void igb_tx_queue_adv(struct igb_adapter *adapter,
3169 struct igb_ring *tx_ring,
3170 int tx_flags, int count, u32 paylen,
3173 union e1000_adv_tx_desc *tx_desc = NULL;
3174 struct igb_buffer *buffer_info;
3175 u32 olinfo_status = 0, cmd_type_len;
3178 cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
3179 E1000_ADVTXD_DCMD_DEXT);
3181 if (tx_flags & IGB_TX_FLAGS_VLAN)
3182 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
3184 if (tx_flags & IGB_TX_FLAGS_TSTAMP)
3185 cmd_type_len |= E1000_ADVTXD_MAC_TSTAMP;
3187 if (tx_flags & IGB_TX_FLAGS_TSO) {
3188 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
3190 /* insert tcp checksum */
3191 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
3193 /* insert ip checksum */
3194 if (tx_flags & IGB_TX_FLAGS_IPV4)
3195 olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
3197 } else if (tx_flags & IGB_TX_FLAGS_CSUM) {
3198 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
3201 if ((adapter->flags & IGB_FLAG_NEED_CTX_IDX) &&
3202 (tx_flags & (IGB_TX_FLAGS_CSUM | IGB_TX_FLAGS_TSO |
3203 IGB_TX_FLAGS_VLAN)))
3204 olinfo_status |= tx_ring->queue_index << 4;
3206 olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
3208 i = tx_ring->next_to_use;
3210 buffer_info = &tx_ring->buffer_info[i];
3211 tx_desc = E1000_TX_DESC_ADV(*tx_ring, i);
3212 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
3213 tx_desc->read.cmd_type_len =
3214 cpu_to_le32(cmd_type_len | buffer_info->length);
3215 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
3217 if (i == tx_ring->count)
3221 tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
3222 /* Force memory writes to complete before letting h/w
3223 * know there are new descriptors to fetch. (Only
3224 * applicable for weak-ordered memory model archs,
3225 * such as IA-64). */
3228 tx_ring->next_to_use = i;
3229 writel(i, adapter->hw.hw_addr + tx_ring->tail);
3230 /* we need this if more than one processor can write to our tail
3231 * at a time, it syncronizes IO on IA64/Altix systems */
3235 static int __igb_maybe_stop_tx(struct net_device *netdev,
3236 struct igb_ring *tx_ring, int size)
3238 struct igb_adapter *adapter = netdev_priv(netdev);
3240 netif_stop_subqueue(netdev, tx_ring->queue_index);
3242 /* Herbert's original patch had:
3243 * smp_mb__after_netif_stop_queue();
3244 * but since that doesn't exist yet, just open code it. */
3247 /* We need to check again in a case another CPU has just
3248 * made room available. */
3249 if (igb_desc_unused(tx_ring) < size)
3253 netif_wake_subqueue(netdev, tx_ring->queue_index);
3254 ++adapter->restart_queue;
3258 static int igb_maybe_stop_tx(struct net_device *netdev,
3259 struct igb_ring *tx_ring, int size)
3261 if (igb_desc_unused(tx_ring) >= size)
3263 return __igb_maybe_stop_tx(netdev, tx_ring, size);
3266 static int igb_xmit_frame_ring_adv(struct sk_buff *skb,
3267 struct net_device *netdev,
3268 struct igb_ring *tx_ring)
3270 struct igb_adapter *adapter = netdev_priv(netdev);
3272 unsigned int tx_flags = 0;
3276 union skb_shared_tx *shtx;
3278 if (test_bit(__IGB_DOWN, &adapter->state)) {
3279 dev_kfree_skb_any(skb);
3280 return NETDEV_TX_OK;
3283 if (skb->len <= 0) {
3284 dev_kfree_skb_any(skb);
3285 return NETDEV_TX_OK;
3288 /* need: 1 descriptor per page,
3289 * + 2 desc gap to keep tail from touching head,
3290 * + 1 desc for skb->data,
3291 * + 1 desc for context descriptor,
3292 * otherwise try next time */
3293 if (igb_maybe_stop_tx(netdev, tx_ring, skb_shinfo(skb)->nr_frags + 4)) {
3294 /* this is a hard error */
3295 return NETDEV_TX_BUSY;
3299 * TODO: check that there currently is no other packet with
3300 * time stamping in the queue
3302 * When doing time stamping, keep the connection to the socket
3303 * a while longer: it is still needed by skb_hwtstamp_tx(),
3304 * called either in igb_tx_hwtstamp() or by our caller when
3305 * doing software time stamping.
3308 if (unlikely(shtx->hardware)) {
3309 shtx->in_progress = 1;
3310 tx_flags |= IGB_TX_FLAGS_TSTAMP;
3313 if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
3314 tx_flags |= IGB_TX_FLAGS_VLAN;
3315 tx_flags |= (vlan_tx_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
3318 if (skb->protocol == htons(ETH_P_IP))
3319 tx_flags |= IGB_TX_FLAGS_IPV4;
3321 first = tx_ring->next_to_use;
3322 tso = skb_is_gso(skb) ? igb_tso_adv(adapter, tx_ring, skb, tx_flags,
3326 dev_kfree_skb_any(skb);
3327 return NETDEV_TX_OK;
3331 tx_flags |= IGB_TX_FLAGS_TSO;
3332 else if (igb_tx_csum_adv(adapter, tx_ring, skb, tx_flags) &&
3333 (skb->ip_summed == CHECKSUM_PARTIAL))
3334 tx_flags |= IGB_TX_FLAGS_CSUM;
3337 * count reflects descriptors mapped, if 0 then mapping error
3338 * has occured and we need to rewind the descriptor queue
3340 count = igb_tx_map_adv(adapter, tx_ring, skb, first);
3343 igb_tx_queue_adv(adapter, tx_ring, tx_flags, count,
3345 netdev->trans_start = jiffies;
3346 /* Make sure there is space in the ring for the next send. */
3347 igb_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 4);
3349 dev_kfree_skb_any(skb);
3350 tx_ring->buffer_info[first].time_stamp = 0;
3351 tx_ring->next_to_use = first;
3354 return NETDEV_TX_OK;
3357 static int igb_xmit_frame_adv(struct sk_buff *skb, struct net_device *netdev)
3359 struct igb_adapter *adapter = netdev_priv(netdev);
3360 struct igb_ring *tx_ring;
3363 r_idx = skb->queue_mapping & (IGB_ABS_MAX_TX_QUEUES - 1);
3364 tx_ring = adapter->multi_tx_table[r_idx];
3366 /* This goes back to the question of how to logically map a tx queue
3367 * to a flow. Right now, performance is impacted slightly negatively
3368 * if using multiple tx queues. If the stack breaks away from a
3369 * single qdisc implementation, we can look at this again. */
3370 return (igb_xmit_frame_ring_adv(skb, netdev, tx_ring));
3374 * igb_tx_timeout - Respond to a Tx Hang
3375 * @netdev: network interface device structure
3377 static void igb_tx_timeout(struct net_device *netdev)
3379 struct igb_adapter *adapter = netdev_priv(netdev);
3380 struct e1000_hw *hw = &adapter->hw;
3382 /* Do the reset outside of interrupt context */
3383 adapter->tx_timeout_count++;
3384 schedule_work(&adapter->reset_task);
3386 (adapter->eims_enable_mask & ~adapter->eims_other));
3389 static void igb_reset_task(struct work_struct *work)
3391 struct igb_adapter *adapter;
3392 adapter = container_of(work, struct igb_adapter, reset_task);
3394 igb_reinit_locked(adapter);
3398 * igb_get_stats - Get System Network Statistics
3399 * @netdev: network interface device structure
3401 * Returns the address of the device statistics structure.
3402 * The statistics are actually updated from the timer callback.
3404 static struct net_device_stats *igb_get_stats(struct net_device *netdev)
3406 struct igb_adapter *adapter = netdev_priv(netdev);
3408 /* only return the current stats */
3409 return &adapter->net_stats;
3413 * igb_change_mtu - Change the Maximum Transfer Unit
3414 * @netdev: network interface device structure
3415 * @new_mtu: new value for maximum frame size
3417 * Returns 0 on success, negative on failure
3419 static int igb_change_mtu(struct net_device *netdev, int new_mtu)
3421 struct igb_adapter *adapter = netdev_priv(netdev);
3422 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3424 if ((max_frame < ETH_ZLEN + ETH_FCS_LEN) ||
3425 (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3426 dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
3430 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
3431 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
3435 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
3438 /* igb_down has a dependency on max_frame_size */
3439 adapter->max_frame_size = max_frame;
3440 if (netif_running(netdev))
3443 /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3444 * means we reserve 2 more, this pushes us to allocate from the next
3446 * i.e. RXBUFFER_2048 --> size-4096 slab
3449 if (max_frame <= IGB_RXBUFFER_256)
3450 adapter->rx_buffer_len = IGB_RXBUFFER_256;
3451 else if (max_frame <= IGB_RXBUFFER_512)
3452 adapter->rx_buffer_len = IGB_RXBUFFER_512;
3453 else if (max_frame <= IGB_RXBUFFER_1024)
3454 adapter->rx_buffer_len = IGB_RXBUFFER_1024;
3455 else if (max_frame <= IGB_RXBUFFER_2048)
3456 adapter->rx_buffer_len = IGB_RXBUFFER_2048;
3458 #if (PAGE_SIZE / 2) > IGB_RXBUFFER_16384
3459 adapter->rx_buffer_len = IGB_RXBUFFER_16384;
3461 adapter->rx_buffer_len = PAGE_SIZE / 2;
3464 /* if sr-iov is enabled we need to force buffer size to 1K or larger */
3465 if (adapter->vfs_allocated_count &&
3466 (adapter->rx_buffer_len < IGB_RXBUFFER_1024))
3467 adapter->rx_buffer_len = IGB_RXBUFFER_1024;
3469 /* adjust allocation if LPE protects us, and we aren't using SBP */
3470 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
3471 (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE))
3472 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3474 dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
3475 netdev->mtu, new_mtu);
3476 netdev->mtu = new_mtu;
3478 if (netif_running(netdev))
3483 clear_bit(__IGB_RESETTING, &adapter->state);
3489 * igb_update_stats - Update the board statistics counters
3490 * @adapter: board private structure
3493 void igb_update_stats(struct igb_adapter *adapter)
3495 struct e1000_hw *hw = &adapter->hw;
3496 struct pci_dev *pdev = adapter->pdev;
3499 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3502 * Prevent stats update while adapter is being reset, or if the pci
3503 * connection is down.
3505 if (adapter->link_speed == 0)
3507 if (pci_channel_offline(pdev))
3510 adapter->stats.crcerrs += rd32(E1000_CRCERRS);
3511 adapter->stats.gprc += rd32(E1000_GPRC);
3512 adapter->stats.gorc += rd32(E1000_GORCL);
3513 rd32(E1000_GORCH); /* clear GORCL */
3514 adapter->stats.bprc += rd32(E1000_BPRC);
3515 adapter->stats.mprc += rd32(E1000_MPRC);
3516 adapter->stats.roc += rd32(E1000_ROC);
3518 adapter->stats.prc64 += rd32(E1000_PRC64);
3519 adapter->stats.prc127 += rd32(E1000_PRC127);
3520 adapter->stats.prc255 += rd32(E1000_PRC255);
3521 adapter->stats.prc511 += rd32(E1000_PRC511);
3522 adapter->stats.prc1023 += rd32(E1000_PRC1023);
3523 adapter->stats.prc1522 += rd32(E1000_PRC1522);
3524 adapter->stats.symerrs += rd32(E1000_SYMERRS);
3525 adapter->stats.sec += rd32(E1000_SEC);
3527 adapter->stats.mpc += rd32(E1000_MPC);
3528 adapter->stats.scc += rd32(E1000_SCC);
3529 adapter->stats.ecol += rd32(E1000_ECOL);
3530 adapter->stats.mcc += rd32(E1000_MCC);
3531 adapter->stats.latecol += rd32(E1000_LATECOL);
3532 adapter->stats.dc += rd32(E1000_DC);
3533 adapter->stats.rlec += rd32(E1000_RLEC);
3534 adapter->stats.xonrxc += rd32(E1000_XONRXC);
3535 adapter->stats.xontxc += rd32(E1000_XONTXC);
3536 adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
3537 adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
3538 adapter->stats.fcruc += rd32(E1000_FCRUC);
3539 adapter->stats.gptc += rd32(E1000_GPTC);
3540 adapter->stats.gotc += rd32(E1000_GOTCL);
3541 rd32(E1000_GOTCH); /* clear GOTCL */
3542 adapter->stats.rnbc += rd32(E1000_RNBC);
3543 adapter->stats.ruc += rd32(E1000_RUC);
3544 adapter->stats.rfc += rd32(E1000_RFC);
3545 adapter->stats.rjc += rd32(E1000_RJC);
3546 adapter->stats.tor += rd32(E1000_TORH);
3547 adapter->stats.tot += rd32(E1000_TOTH);
3548 adapter->stats.tpr += rd32(E1000_TPR);
3550 adapter->stats.ptc64 += rd32(E1000_PTC64);
3551 adapter->stats.ptc127 += rd32(E1000_PTC127);
3552 adapter->stats.ptc255 += rd32(E1000_PTC255);
3553 adapter->stats.ptc511 += rd32(E1000_PTC511);
3554 adapter->stats.ptc1023 += rd32(E1000_PTC1023);
3555 adapter->stats.ptc1522 += rd32(E1000_PTC1522);
3557 adapter->stats.mptc += rd32(E1000_MPTC);
3558 adapter->stats.bptc += rd32(E1000_BPTC);
3560 /* used for adaptive IFS */
3562 hw->mac.tx_packet_delta = rd32(E1000_TPT);
3563 adapter->stats.tpt += hw->mac.tx_packet_delta;
3564 hw->mac.collision_delta = rd32(E1000_COLC);
3565 adapter->stats.colc += hw->mac.collision_delta;
3567 adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
3568 adapter->stats.rxerrc += rd32(E1000_RXERRC);
3569 adapter->stats.tncrs += rd32(E1000_TNCRS);
3570 adapter->stats.tsctc += rd32(E1000_TSCTC);
3571 adapter->stats.tsctfc += rd32(E1000_TSCTFC);
3573 adapter->stats.iac += rd32(E1000_IAC);
3574 adapter->stats.icrxoc += rd32(E1000_ICRXOC);
3575 adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
3576 adapter->stats.icrxatc += rd32(E1000_ICRXATC);
3577 adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
3578 adapter->stats.ictxatc += rd32(E1000_ICTXATC);
3579 adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
3580 adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
3581 adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
3583 /* Fill out the OS statistics structure */
3584 adapter->net_stats.multicast = adapter->stats.mprc;
3585 adapter->net_stats.collisions = adapter->stats.colc;
3589 /* RLEC on some newer hardware can be incorrect so build
3590 * our own version based on RUC and ROC */
3591 adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3592 adapter->stats.crcerrs + adapter->stats.algnerrc +
3593 adapter->stats.ruc + adapter->stats.roc +
3594 adapter->stats.cexterr;
3595 adapter->net_stats.rx_length_errors = adapter->stats.ruc +
3597 adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3598 adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3599 adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3602 adapter->net_stats.tx_errors = adapter->stats.ecol +
3603 adapter->stats.latecol;
3604 adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3605 adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3606 adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3608 /* Tx Dropped needs to be maintained elsewhere */
3611 if (hw->phy.media_type == e1000_media_type_copper) {
3612 if ((adapter->link_speed == SPEED_1000) &&
3613 (!igb_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
3614 phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3615 adapter->phy_stats.idle_errors += phy_tmp;
3619 /* Management Stats */
3620 adapter->stats.mgptc += rd32(E1000_MGTPTC);
3621 adapter->stats.mgprc += rd32(E1000_MGTPRC);
3622 adapter->stats.mgpdc += rd32(E1000_MGTPDC);
3625 static irqreturn_t igb_msix_other(int irq, void *data)
3627 struct net_device *netdev = data;
3628 struct igb_adapter *adapter = netdev_priv(netdev);
3629 struct e1000_hw *hw = &adapter->hw;
3630 u32 icr = rd32(E1000_ICR);
3632 /* reading ICR causes bit 31 of EICR to be cleared */
3634 if(icr & E1000_ICR_DOUTSYNC) {
3635 /* HW is reporting DMA is out of sync */
3636 adapter->stats.doosync++;
3639 /* Check for a mailbox event */
3640 if (icr & E1000_ICR_VMMB)
3641 igb_msg_task(adapter);
3643 if (icr & E1000_ICR_LSC) {
3644 hw->mac.get_link_status = 1;
3645 /* guard against interrupt when we're going down */
3646 if (!test_bit(__IGB_DOWN, &adapter->state))
3647 mod_timer(&adapter->watchdog_timer, jiffies + 1);
3650 wr32(E1000_IMS, E1000_IMS_LSC | E1000_IMS_DOUTSYNC | E1000_IMS_VMMB);
3651 wr32(E1000_EIMS, adapter->eims_other);
3656 static irqreturn_t igb_msix_tx(int irq, void *data)
3658 struct igb_ring *tx_ring = data;
3659 struct igb_adapter *adapter = tx_ring->adapter;
3660 struct e1000_hw *hw = &adapter->hw;
3662 #ifdef CONFIG_IGB_DCA
3663 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
3664 igb_update_tx_dca(tx_ring);
3667 tx_ring->total_bytes = 0;
3668 tx_ring->total_packets = 0;
3670 /* auto mask will automatically reenable the interrupt when we write
3672 if (!igb_clean_tx_irq(tx_ring))
3673 /* Ring was not completely cleaned, so fire another interrupt */
3674 wr32(E1000_EICS, tx_ring->eims_value);
3676 wr32(E1000_EIMS, tx_ring->eims_value);
3681 static void igb_write_itr(struct igb_ring *ring)
3683 struct e1000_hw *hw = &ring->adapter->hw;
3684 if ((ring->adapter->itr_setting & 3) && ring->set_itr) {
3685 switch (hw->mac.type) {
3687 wr32(ring->itr_register, ring->itr_val |
3691 wr32(ring->itr_register, ring->itr_val |
3692 (ring->itr_val << 16));
3699 static irqreturn_t igb_msix_rx(int irq, void *data)
3701 struct igb_ring *rx_ring = data;
3703 /* Write the ITR value calculated at the end of the
3704 * previous interrupt.
3707 igb_write_itr(rx_ring);
3709 if (napi_schedule_prep(&rx_ring->napi))
3710 __napi_schedule(&rx_ring->napi);
3712 #ifdef CONFIG_IGB_DCA
3713 if (rx_ring->adapter->flags & IGB_FLAG_DCA_ENABLED)
3714 igb_update_rx_dca(rx_ring);
3719 #ifdef CONFIG_IGB_DCA
3720 static void igb_update_rx_dca(struct igb_ring *rx_ring)
3723 struct igb_adapter *adapter = rx_ring->adapter;
3724 struct e1000_hw *hw = &adapter->hw;
3725 int cpu = get_cpu();
3726 int q = rx_ring->reg_idx;
3728 if (rx_ring->cpu != cpu) {
3729 dca_rxctrl = rd32(E1000_DCA_RXCTRL(q));
3730 if (hw->mac.type == e1000_82576) {
3731 dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK_82576;
3732 dca_rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu) <<
3733 E1000_DCA_RXCTRL_CPUID_SHIFT;
3735 dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK;
3736 dca_rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
3738 dca_rxctrl |= E1000_DCA_RXCTRL_DESC_DCA_EN;
3739 dca_rxctrl |= E1000_DCA_RXCTRL_HEAD_DCA_EN;
3740 dca_rxctrl |= E1000_DCA_RXCTRL_DATA_DCA_EN;
3741 wr32(E1000_DCA_RXCTRL(q), dca_rxctrl);
3747 static void igb_update_tx_dca(struct igb_ring *tx_ring)
3750 struct igb_adapter *adapter = tx_ring->adapter;
3751 struct e1000_hw *hw = &adapter->hw;
3752 int cpu = get_cpu();
3753 int q = tx_ring->reg_idx;
3755 if (tx_ring->cpu != cpu) {
3756 dca_txctrl = rd32(E1000_DCA_TXCTRL(q));
3757 if (hw->mac.type == e1000_82576) {
3758 dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK_82576;
3759 dca_txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu) <<
3760 E1000_DCA_TXCTRL_CPUID_SHIFT;
3762 dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK;
3763 dca_txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
3765 dca_txctrl |= E1000_DCA_TXCTRL_DESC_DCA_EN;
3766 wr32(E1000_DCA_TXCTRL(q), dca_txctrl);
3772 static void igb_setup_dca(struct igb_adapter *adapter)
3776 if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
3779 for (i = 0; i < adapter->num_tx_queues; i++) {
3780 adapter->tx_ring[i].cpu = -1;
3781 igb_update_tx_dca(&adapter->tx_ring[i]);
3783 for (i = 0; i < adapter->num_rx_queues; i++) {
3784 adapter->rx_ring[i].cpu = -1;
3785 igb_update_rx_dca(&adapter->rx_ring[i]);
3789 static int __igb_notify_dca(struct device *dev, void *data)
3791 struct net_device *netdev = dev_get_drvdata(dev);
3792 struct igb_adapter *adapter = netdev_priv(netdev);
3793 struct e1000_hw *hw = &adapter->hw;
3794 unsigned long event = *(unsigned long *)data;
3797 case DCA_PROVIDER_ADD:
3798 /* if already enabled, don't do it again */
3799 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
3801 /* Always use CB2 mode, difference is masked
3802 * in the CB driver. */
3803 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
3804 if (dca_add_requester(dev) == 0) {
3805 adapter->flags |= IGB_FLAG_DCA_ENABLED;
3806 dev_info(&adapter->pdev->dev, "DCA enabled\n");
3807 igb_setup_dca(adapter);
3810 /* Fall Through since DCA is disabled. */
3811 case DCA_PROVIDER_REMOVE:
3812 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
3813 /* without this a class_device is left
3814 * hanging around in the sysfs model */
3815 dca_remove_requester(dev);
3816 dev_info(&adapter->pdev->dev, "DCA disabled\n");
3817 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
3818 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
3826 static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
3831 ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
3834 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
3836 #endif /* CONFIG_IGB_DCA */
3838 static void igb_ping_all_vfs(struct igb_adapter *adapter)
3840 struct e1000_hw *hw = &adapter->hw;
3844 for (i = 0 ; i < adapter->vfs_allocated_count; i++) {
3845 ping = E1000_PF_CONTROL_MSG;
3846 if (adapter->vf_data[i].clear_to_send)
3847 ping |= E1000_VT_MSGTYPE_CTS;
3848 igb_write_mbx(hw, &ping, 1, i);
3852 static int igb_set_vf_multicasts(struct igb_adapter *adapter,
3853 u32 *msgbuf, u32 vf)
3855 int n = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
3856 u16 *hash_list = (u16 *)&msgbuf[1];
3857 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
3860 /* only up to 30 hash values supported */
3864 /* salt away the number of multi cast addresses assigned
3865 * to this VF for later use to restore when the PF multi cast
3868 vf_data->num_vf_mc_hashes = n;
3870 /* VFs are limited to using the MTA hash table for their multicast
3872 for (i = 0; i < n; i++)
3873 vf_data->vf_mc_hashes[i] = hash_list[i];;
3875 /* Flush and reset the mta with the new values */
3876 igb_set_multi(adapter->netdev);
3881 static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
3883 struct e1000_hw *hw = &adapter->hw;
3884 struct vf_data_storage *vf_data;
3887 for (i = 0; i < adapter->vfs_allocated_count; i++) {
3888 vf_data = &adapter->vf_data[i];
3889 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
3890 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
3894 static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
3896 struct e1000_hw *hw = &adapter->hw;
3897 u32 pool_mask, reg, vid;
3900 pool_mask = 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
3902 /* Find the vlan filter for this id */
3903 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
3904 reg = rd32(E1000_VLVF(i));
3906 /* remove the vf from the pool */
3909 /* if pool is empty then remove entry from vfta */
3910 if (!(reg & E1000_VLVF_POOLSEL_MASK) &&
3911 (reg & E1000_VLVF_VLANID_ENABLE)) {
3913 vid = reg & E1000_VLVF_VLANID_MASK;
3914 igb_vfta_set(hw, vid, false);
3917 wr32(E1000_VLVF(i), reg);
3921 static s32 igb_vlvf_set(struct igb_adapter *adapter, u32 vid, bool add, u32 vf)
3923 struct e1000_hw *hw = &adapter->hw;
3926 /* It is an error to call this function when VFs are not enabled */
3927 if (!adapter->vfs_allocated_count)
3930 /* Find the vlan filter for this id */
3931 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
3932 reg = rd32(E1000_VLVF(i));
3933 if ((reg & E1000_VLVF_VLANID_ENABLE) &&
3934 vid == (reg & E1000_VLVF_VLANID_MASK))
3939 if (i == E1000_VLVF_ARRAY_SIZE) {
3940 /* Did not find a matching VLAN ID entry that was
3941 * enabled. Search for a free filter entry, i.e.
3942 * one without the enable bit set
3944 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
3945 reg = rd32(E1000_VLVF(i));
3946 if (!(reg & E1000_VLVF_VLANID_ENABLE))
3950 if (i < E1000_VLVF_ARRAY_SIZE) {
3951 /* Found an enabled/available entry */
3952 reg |= 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
3954 /* if !enabled we need to set this up in vfta */
3955 if (!(reg & E1000_VLVF_VLANID_ENABLE)) {
3956 /* add VID to filter table, if bit already set
3957 * PF must have added it outside of table */
3958 if (igb_vfta_set(hw, vid, true))
3959 reg |= 1 << (E1000_VLVF_POOLSEL_SHIFT +
3960 adapter->vfs_allocated_count);
3961 reg |= E1000_VLVF_VLANID_ENABLE;
3963 reg &= ~E1000_VLVF_VLANID_MASK;
3966 wr32(E1000_VLVF(i), reg);
3970 if (i < E1000_VLVF_ARRAY_SIZE) {
3971 /* remove vf from the pool */
3972 reg &= ~(1 << (E1000_VLVF_POOLSEL_SHIFT + vf));
3973 /* if pool is empty then remove entry from vfta */
3974 if (!(reg & E1000_VLVF_POOLSEL_MASK)) {
3976 igb_vfta_set(hw, vid, false);
3978 wr32(E1000_VLVF(i), reg);
3985 static int igb_set_vf_vlan(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
3987 int add = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
3988 int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
3990 return igb_vlvf_set(adapter, vid, add, vf);
3993 static inline void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
3995 struct e1000_hw *hw = &adapter->hw;
3997 /* disable mailbox functionality for vf */
3998 adapter->vf_data[vf].clear_to_send = false;
4000 /* reset offloads to defaults */
4001 igb_set_vmolr(hw, vf);
4003 /* reset vlans for device */
4004 igb_clear_vf_vfta(adapter, vf);
4006 /* reset multicast table array for vf */
4007 adapter->vf_data[vf].num_vf_mc_hashes = 0;
4009 /* Flush and reset the mta with the new values */
4010 igb_set_multi(adapter->netdev);
4013 static inline void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
4015 struct e1000_hw *hw = &adapter->hw;
4016 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
4018 u8 *addr = (u8 *)(&msgbuf[1]);
4020 /* process all the same items cleared in a function level reset */
4021 igb_vf_reset_event(adapter, vf);
4023 /* set vf mac address */
4024 igb_rar_set(hw, vf_mac, vf + 1);
4025 igb_set_rah_pool(hw, vf, vf + 1);
4027 /* enable transmit and receive for vf */
4028 reg = rd32(E1000_VFTE);
4029 wr32(E1000_VFTE, reg | (1 << vf));
4030 reg = rd32(E1000_VFRE);
4031 wr32(E1000_VFRE, reg | (1 << vf));
4033 /* enable mailbox functionality for vf */
4034 adapter->vf_data[vf].clear_to_send = true;
4036 /* reply to reset with ack and vf mac address */
4037 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
4038 memcpy(addr, vf_mac, 6);
4039 igb_write_mbx(hw, msgbuf, 3, vf);
4042 static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf)
4044 unsigned char *addr = (char *)&msg[1];
4047 if (is_valid_ether_addr(addr))
4048 err = igb_set_vf_mac(adapter, vf, addr);
4054 static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf)
4056 struct e1000_hw *hw = &adapter->hw;
4057 u32 msg = E1000_VT_MSGTYPE_NACK;
4059 /* if device isn't clear to send it shouldn't be reading either */
4060 if (!adapter->vf_data[vf].clear_to_send)
4061 igb_write_mbx(hw, &msg, 1, vf);
4065 static void igb_msg_task(struct igb_adapter *adapter)
4067 struct e1000_hw *hw = &adapter->hw;
4070 for (vf = 0; vf < adapter->vfs_allocated_count; vf++) {
4071 /* process any reset requests */
4072 if (!igb_check_for_rst(hw, vf)) {
4073 adapter->vf_data[vf].clear_to_send = false;
4074 igb_vf_reset_event(adapter, vf);
4077 /* process any messages pending */
4078 if (!igb_check_for_msg(hw, vf))
4079 igb_rcv_msg_from_vf(adapter, vf);
4081 /* process any acks */
4082 if (!igb_check_for_ack(hw, vf))
4083 igb_rcv_ack_from_vf(adapter, vf);
4088 static int igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
4090 u32 mbx_size = E1000_VFMAILBOX_SIZE;
4091 u32 msgbuf[mbx_size];
4092 struct e1000_hw *hw = &adapter->hw;
4095 retval = igb_read_mbx(hw, msgbuf, mbx_size, vf);
4098 dev_err(&adapter->pdev->dev,
4099 "Error receiving message from VF\n");
4101 /* this is a message we already processed, do nothing */
4102 if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
4106 * until the vf completes a reset it should not be
4107 * allowed to start any configuration.
4110 if (msgbuf[0] == E1000_VF_RESET) {
4111 igb_vf_reset_msg(adapter, vf);
4116 if (!adapter->vf_data[vf].clear_to_send) {
4117 msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
4118 igb_write_mbx(hw, msgbuf, 1, vf);
4122 switch ((msgbuf[0] & 0xFFFF)) {
4123 case E1000_VF_SET_MAC_ADDR:
4124 retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
4126 case E1000_VF_SET_MULTICAST:
4127 retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
4129 case E1000_VF_SET_LPE:
4130 retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf);
4132 case E1000_VF_SET_VLAN:
4133 retval = igb_set_vf_vlan(adapter, msgbuf, vf);
4136 dev_err(&adapter->pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
4141 /* notify the VF of the results of what it sent us */
4143 msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
4145 msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
4147 msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
4149 igb_write_mbx(hw, msgbuf, 1, vf);
4155 * igb_intr_msi - Interrupt Handler
4156 * @irq: interrupt number
4157 * @data: pointer to a network interface device structure
4159 static irqreturn_t igb_intr_msi(int irq, void *data)
4161 struct net_device *netdev = data;
4162 struct igb_adapter *adapter = netdev_priv(netdev);
4163 struct e1000_hw *hw = &adapter->hw;
4164 /* read ICR disables interrupts using IAM */
4165 u32 icr = rd32(E1000_ICR);
4167 igb_write_itr(adapter->rx_ring);
4169 if(icr & E1000_ICR_DOUTSYNC) {
4170 /* HW is reporting DMA is out of sync */
4171 adapter->stats.doosync++;
4174 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
4175 hw->mac.get_link_status = 1;
4176 if (!test_bit(__IGB_DOWN, &adapter->state))
4177 mod_timer(&adapter->watchdog_timer, jiffies + 1);
4180 napi_schedule(&adapter->rx_ring[0].napi);
4186 * igb_intr - Legacy Interrupt Handler
4187 * @irq: interrupt number
4188 * @data: pointer to a network interface device structure
4190 static irqreturn_t igb_intr(int irq, void *data)
4192 struct net_device *netdev = data;
4193 struct igb_adapter *adapter = netdev_priv(netdev);
4194 struct e1000_hw *hw = &adapter->hw;
4195 /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked. No
4196 * need for the IMC write */
4197 u32 icr = rd32(E1000_ICR);
4199 return IRQ_NONE; /* Not our interrupt */
4201 igb_write_itr(adapter->rx_ring);
4203 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
4204 * not set, then the adapter didn't send an interrupt */
4205 if (!(icr & E1000_ICR_INT_ASSERTED))
4208 if(icr & E1000_ICR_DOUTSYNC) {
4209 /* HW is reporting DMA is out of sync */
4210 adapter->stats.doosync++;
4213 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
4214 hw->mac.get_link_status = 1;
4215 /* guard against interrupt when we're going down */
4216 if (!test_bit(__IGB_DOWN, &adapter->state))
4217 mod_timer(&adapter->watchdog_timer, jiffies + 1);
4220 napi_schedule(&adapter->rx_ring[0].napi);
4225 static inline void igb_rx_irq_enable(struct igb_ring *rx_ring)
4227 struct igb_adapter *adapter = rx_ring->adapter;
4228 struct e1000_hw *hw = &adapter->hw;
4230 if (adapter->itr_setting & 3) {
4231 if (adapter->num_rx_queues == 1)
4232 igb_set_itr(adapter);
4234 igb_update_ring_itr(rx_ring);
4237 if (!test_bit(__IGB_DOWN, &adapter->state)) {
4238 if (adapter->msix_entries)
4239 wr32(E1000_EIMS, rx_ring->eims_value);
4241 igb_irq_enable(adapter);
4246 * igb_poll - NAPI Rx polling callback
4247 * @napi: napi polling structure
4248 * @budget: count of how many packets we should handle
4250 static int igb_poll(struct napi_struct *napi, int budget)
4252 struct igb_ring *rx_ring = container_of(napi, struct igb_ring, napi);
4255 #ifdef CONFIG_IGB_DCA
4256 if (rx_ring->adapter->flags & IGB_FLAG_DCA_ENABLED)
4257 igb_update_rx_dca(rx_ring);
4259 igb_clean_rx_irq_adv(rx_ring, &work_done, budget);
4261 if (rx_ring->buddy) {
4262 #ifdef CONFIG_IGB_DCA
4263 if (rx_ring->adapter->flags & IGB_FLAG_DCA_ENABLED)
4264 igb_update_tx_dca(rx_ring->buddy);
4266 if (!igb_clean_tx_irq(rx_ring->buddy))
4270 /* If not enough Rx work done, exit the polling mode */
4271 if (work_done < budget) {
4272 napi_complete(napi);
4273 igb_rx_irq_enable(rx_ring);
4280 * igb_hwtstamp - utility function which checks for TX time stamp
4281 * @adapter: board private structure
4282 * @skb: packet that was just sent
4284 * If we were asked to do hardware stamping and such a time stamp is
4285 * available, then it must have been for this skb here because we only
4286 * allow only one such packet into the queue.
4288 static void igb_tx_hwtstamp(struct igb_adapter *adapter, struct sk_buff *skb)
4290 union skb_shared_tx *shtx = skb_tx(skb);
4291 struct e1000_hw *hw = &adapter->hw;
4293 if (unlikely(shtx->hardware)) {
4294 u32 valid = rd32(E1000_TSYNCTXCTL) & E1000_TSYNCTXCTL_VALID;
4296 u64 regval = rd32(E1000_TXSTMPL);
4298 struct skb_shared_hwtstamps shhwtstamps;
4300 memset(&shhwtstamps, 0, sizeof(shhwtstamps));
4301 regval |= (u64)rd32(E1000_TXSTMPH) << 32;
4302 ns = timecounter_cyc2time(&adapter->clock,
4304 timecompare_update(&adapter->compare, ns);
4305 shhwtstamps.hwtstamp = ns_to_ktime(ns);
4306 shhwtstamps.syststamp =
4307 timecompare_transform(&adapter->compare, ns);
4308 skb_tstamp_tx(skb, &shhwtstamps);
4314 * igb_clean_tx_irq - Reclaim resources after transmit completes
4315 * @adapter: board private structure
4316 * returns true if ring is completely cleaned
4318 static bool igb_clean_tx_irq(struct igb_ring *tx_ring)
4320 struct igb_adapter *adapter = tx_ring->adapter;
4321 struct net_device *netdev = adapter->netdev;
4322 struct e1000_hw *hw = &adapter->hw;
4323 struct igb_buffer *buffer_info;
4324 struct sk_buff *skb;
4325 union e1000_adv_tx_desc *tx_desc, *eop_desc;
4326 unsigned int total_bytes = 0, total_packets = 0;
4327 unsigned int i, eop, count = 0;
4328 bool cleaned = false;
4330 i = tx_ring->next_to_clean;
4331 eop = tx_ring->buffer_info[i].next_to_watch;
4332 eop_desc = E1000_TX_DESC_ADV(*tx_ring, eop);
4334 while ((eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)) &&
4335 (count < tx_ring->count)) {
4336 for (cleaned = false; !cleaned; count++) {
4337 tx_desc = E1000_TX_DESC_ADV(*tx_ring, i);
4338 buffer_info = &tx_ring->buffer_info[i];
4339 cleaned = (i == eop);
4340 skb = buffer_info->skb;
4343 unsigned int segs, bytecount;
4344 /* gso_segs is currently only valid for tcp */
4345 segs = skb_shinfo(skb)->gso_segs ?: 1;
4346 /* multiply data chunks by size of headers */
4347 bytecount = ((segs - 1) * skb_headlen(skb)) +
4349 total_packets += segs;
4350 total_bytes += bytecount;
4352 igb_tx_hwtstamp(adapter, skb);
4355 igb_unmap_and_free_tx_resource(adapter, buffer_info);
4356 tx_desc->wb.status = 0;
4359 if (i == tx_ring->count)
4362 eop = tx_ring->buffer_info[i].next_to_watch;
4363 eop_desc = E1000_TX_DESC_ADV(*tx_ring, eop);
4366 tx_ring->next_to_clean = i;
4368 if (unlikely(count &&
4369 netif_carrier_ok(netdev) &&
4370 igb_desc_unused(tx_ring) >= IGB_TX_QUEUE_WAKE)) {
4371 /* Make sure that anybody stopping the queue after this
4372 * sees the new next_to_clean.
4375 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
4376 !(test_bit(__IGB_DOWN, &adapter->state))) {
4377 netif_wake_subqueue(netdev, tx_ring->queue_index);
4378 ++adapter->restart_queue;
4382 if (tx_ring->detect_tx_hung) {
4383 /* Detect a transmit hang in hardware, this serializes the
4384 * check with the clearing of time_stamp and movement of i */
4385 tx_ring->detect_tx_hung = false;
4386 if (tx_ring->buffer_info[i].time_stamp &&
4387 time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
4388 (adapter->tx_timeout_factor * HZ))
4389 && !(rd32(E1000_STATUS) &
4390 E1000_STATUS_TXOFF)) {
4392 /* detected Tx unit hang */
4393 dev_err(&adapter->pdev->dev,
4394 "Detected Tx Unit Hang\n"
4398 " next_to_use <%x>\n"
4399 " next_to_clean <%x>\n"
4400 "buffer_info[next_to_clean]\n"
4401 " time_stamp <%lx>\n"
4402 " next_to_watch <%x>\n"
4404 " desc.status <%x>\n",
4405 tx_ring->queue_index,
4406 readl(adapter->hw.hw_addr + tx_ring->head),
4407 readl(adapter->hw.hw_addr + tx_ring->tail),
4408 tx_ring->next_to_use,
4409 tx_ring->next_to_clean,
4410 tx_ring->buffer_info[i].time_stamp,
4413 eop_desc->wb.status);
4414 netif_stop_subqueue(netdev, tx_ring->queue_index);
4417 tx_ring->total_bytes += total_bytes;
4418 tx_ring->total_packets += total_packets;
4419 tx_ring->tx_stats.bytes += total_bytes;
4420 tx_ring->tx_stats.packets += total_packets;
4421 adapter->net_stats.tx_bytes += total_bytes;
4422 adapter->net_stats.tx_packets += total_packets;
4423 return (count < tx_ring->count);
4427 * igb_receive_skb - helper function to handle rx indications
4428 * @ring: pointer to receive ring receving this packet
4429 * @status: descriptor status field as written by hardware
4430 * @rx_desc: receive descriptor containing vlan and type information.
4431 * @skb: pointer to sk_buff to be indicated to stack
4433 static void igb_receive_skb(struct igb_ring *ring, u8 status,
4434 union e1000_adv_rx_desc * rx_desc,
4435 struct sk_buff *skb)
4437 struct igb_adapter * adapter = ring->adapter;
4438 bool vlan_extracted = (adapter->vlgrp && (status & E1000_RXD_STAT_VP));
4440 skb_record_rx_queue(skb, ring->queue_index);
4442 vlan_gro_receive(&ring->napi, adapter->vlgrp,
4443 le16_to_cpu(rx_desc->wb.upper.vlan),
4446 napi_gro_receive(&ring->napi, skb);
4449 static inline void igb_rx_checksum_adv(struct igb_adapter *adapter,
4450 u32 status_err, struct sk_buff *skb)
4452 skb->ip_summed = CHECKSUM_NONE;
4454 /* Ignore Checksum bit is set or checksum is disabled through ethtool */
4455 if ((status_err & E1000_RXD_STAT_IXSM) ||
4456 (adapter->flags & IGB_FLAG_RX_CSUM_DISABLED))
4458 /* TCP/UDP checksum error bit is set */
4460 (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
4462 * work around errata with sctp packets where the TCPE aka
4463 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
4464 * packets, (aka let the stack check the crc32c)
4466 if (!((adapter->hw.mac.type == e1000_82576) &&
4468 adapter->hw_csum_err++;
4469 /* let the stack verify checksum errors */
4472 /* It must be a TCP or UDP packet with a valid checksum */
4473 if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
4474 skb->ip_summed = CHECKSUM_UNNECESSARY;
4476 dev_dbg(&adapter->pdev->dev, "cksum success: bits %08X\n", status_err);
4477 adapter->hw_csum_good++;
4480 static bool igb_clean_rx_irq_adv(struct igb_ring *rx_ring,
4481 int *work_done, int budget)
4483 struct igb_adapter *adapter = rx_ring->adapter;
4484 struct net_device *netdev = adapter->netdev;
4485 struct e1000_hw *hw = &adapter->hw;
4486 struct pci_dev *pdev = adapter->pdev;
4487 union e1000_adv_rx_desc *rx_desc , *next_rxd;
4488 struct igb_buffer *buffer_info , *next_buffer;
4489 struct sk_buff *skb;
4490 bool cleaned = false;
4491 int cleaned_count = 0;
4492 unsigned int total_bytes = 0, total_packets = 0;
4494 u32 length, hlen, staterr;
4496 i = rx_ring->next_to_clean;
4497 buffer_info = &rx_ring->buffer_info[i];
4498 rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
4499 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
4501 while (staterr & E1000_RXD_STAT_DD) {
4502 if (*work_done >= budget)
4506 skb = buffer_info->skb;
4507 prefetch(skb->data - NET_IP_ALIGN);
4508 buffer_info->skb = NULL;
4511 if (i == rx_ring->count)
4513 next_rxd = E1000_RX_DESC_ADV(*rx_ring, i);
4515 next_buffer = &rx_ring->buffer_info[i];
4517 length = le16_to_cpu(rx_desc->wb.upper.length);
4521 if (!adapter->rx_ps_hdr_size) {
4522 pci_unmap_single(pdev, buffer_info->dma,
4523 adapter->rx_buffer_len +
4525 PCI_DMA_FROMDEVICE);
4526 skb_put(skb, length);
4530 /* HW will not DMA in data larger than the given buffer, even
4531 * if it parses the (NFS, of course) header to be larger. In
4532 * that case, it fills the header buffer and spills the rest
4535 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hdr_info) &
4536 E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
4537 if (hlen > adapter->rx_ps_hdr_size)
4538 hlen = adapter->rx_ps_hdr_size;
4540 if (!skb_shinfo(skb)->nr_frags) {
4541 pci_unmap_single(pdev, buffer_info->dma,
4542 adapter->rx_ps_hdr_size + NET_IP_ALIGN,
4543 PCI_DMA_FROMDEVICE);
4548 pci_unmap_page(pdev, buffer_info->page_dma,
4549 PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
4550 buffer_info->page_dma = 0;
4552 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags++,
4554 buffer_info->page_offset,
4557 if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
4558 (page_count(buffer_info->page) != 1))
4559 buffer_info->page = NULL;
4561 get_page(buffer_info->page);
4564 skb->data_len += length;
4566 skb->truesize += length;
4569 if (!(staterr & E1000_RXD_STAT_EOP)) {
4570 buffer_info->skb = next_buffer->skb;
4571 buffer_info->dma = next_buffer->dma;
4572 next_buffer->skb = skb;
4573 next_buffer->dma = 0;
4578 * If this bit is set, then the RX registers contain
4579 * the time stamp. No other packet will be time
4580 * stamped until we read these registers, so read the
4581 * registers to make them available again. Because
4582 * only one packet can be time stamped at a time, we
4583 * know that the register values must belong to this
4584 * one here and therefore we don't need to compare
4585 * any of the additional attributes stored for it.
4587 * If nothing went wrong, then it should have a
4588 * skb_shared_tx that we can turn into a
4589 * skb_shared_hwtstamps.
4591 * TODO: can time stamping be triggered (thus locking
4592 * the registers) without the packet reaching this point
4593 * here? In that case RX time stamping would get stuck.
4595 * TODO: in "time stamp all packets" mode this bit is
4596 * not set. Need a global flag for this mode and then
4597 * always read the registers. Cannot be done without
4600 if (unlikely(staterr & E1000_RXD_STAT_TS)) {
4603 struct skb_shared_hwtstamps *shhwtstamps =
4606 WARN(!(rd32(E1000_TSYNCRXCTL) & E1000_TSYNCRXCTL_VALID),
4607 "igb: no RX time stamp available for time stamped packet");
4608 regval = rd32(E1000_RXSTMPL);
4609 regval |= (u64)rd32(E1000_RXSTMPH) << 32;
4610 ns = timecounter_cyc2time(&adapter->clock, regval);
4611 timecompare_update(&adapter->compare, ns);
4612 memset(shhwtstamps, 0, sizeof(*shhwtstamps));
4613 shhwtstamps->hwtstamp = ns_to_ktime(ns);
4614 shhwtstamps->syststamp =
4615 timecompare_transform(&adapter->compare, ns);
4618 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
4619 dev_kfree_skb_irq(skb);
4623 total_bytes += skb->len;
4626 igb_rx_checksum_adv(adapter, staterr, skb);
4628 skb->protocol = eth_type_trans(skb, netdev);
4630 igb_receive_skb(rx_ring, staterr, rx_desc, skb);
4633 rx_desc->wb.upper.status_error = 0;
4635 /* return some buffers to hardware, one at a time is too slow */
4636 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
4637 igb_alloc_rx_buffers_adv(rx_ring, cleaned_count);
4641 /* use prefetched values */
4643 buffer_info = next_buffer;
4644 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
4647 rx_ring->next_to_clean = i;
4648 cleaned_count = igb_desc_unused(rx_ring);
4651 igb_alloc_rx_buffers_adv(rx_ring, cleaned_count);
4653 rx_ring->total_packets += total_packets;
4654 rx_ring->total_bytes += total_bytes;
4655 rx_ring->rx_stats.packets += total_packets;
4656 rx_ring->rx_stats.bytes += total_bytes;
4657 adapter->net_stats.rx_bytes += total_bytes;
4658 adapter->net_stats.rx_packets += total_packets;
4663 * igb_alloc_rx_buffers_adv - Replace used receive buffers; packet split
4664 * @adapter: address of board private structure
4666 static void igb_alloc_rx_buffers_adv(struct igb_ring *rx_ring,
4669 struct igb_adapter *adapter = rx_ring->adapter;
4670 struct net_device *netdev = adapter->netdev;
4671 struct pci_dev *pdev = adapter->pdev;
4672 union e1000_adv_rx_desc *rx_desc;
4673 struct igb_buffer *buffer_info;
4674 struct sk_buff *skb;
4678 i = rx_ring->next_to_use;
4679 buffer_info = &rx_ring->buffer_info[i];
4681 if (adapter->rx_ps_hdr_size)
4682 bufsz = adapter->rx_ps_hdr_size;
4684 bufsz = adapter->rx_buffer_len;
4685 bufsz += NET_IP_ALIGN;
4687 while (cleaned_count--) {
4688 rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
4690 if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) {
4691 if (!buffer_info->page) {
4692 buffer_info->page = alloc_page(GFP_ATOMIC);
4693 if (!buffer_info->page) {
4694 adapter->alloc_rx_buff_failed++;
4697 buffer_info->page_offset = 0;
4699 buffer_info->page_offset ^= PAGE_SIZE / 2;
4701 buffer_info->page_dma =
4702 pci_map_page(pdev, buffer_info->page,
4703 buffer_info->page_offset,
4705 PCI_DMA_FROMDEVICE);
4708 if (!buffer_info->skb) {
4709 skb = netdev_alloc_skb(netdev, bufsz);
4711 adapter->alloc_rx_buff_failed++;
4715 /* Make buffer alignment 2 beyond a 16 byte boundary
4716 * this will result in a 16 byte aligned IP header after
4717 * the 14 byte MAC header is removed
4719 skb_reserve(skb, NET_IP_ALIGN);
4721 buffer_info->skb = skb;
4722 buffer_info->dma = pci_map_single(pdev, skb->data,
4724 PCI_DMA_FROMDEVICE);
4726 /* Refresh the desc even if buffer_addrs didn't change because
4727 * each write-back erases this info. */
4728 if (adapter->rx_ps_hdr_size) {
4729 rx_desc->read.pkt_addr =
4730 cpu_to_le64(buffer_info->page_dma);
4731 rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
4733 rx_desc->read.pkt_addr =
4734 cpu_to_le64(buffer_info->dma);
4735 rx_desc->read.hdr_addr = 0;
4739 if (i == rx_ring->count)
4741 buffer_info = &rx_ring->buffer_info[i];
4745 if (rx_ring->next_to_use != i) {
4746 rx_ring->next_to_use = i;
4748 i = (rx_ring->count - 1);
4752 /* Force memory writes to complete before letting h/w
4753 * know there are new descriptors to fetch. (Only
4754 * applicable for weak-ordered memory model archs,
4755 * such as IA-64). */
4757 writel(i, adapter->hw.hw_addr + rx_ring->tail);
4767 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4769 struct igb_adapter *adapter = netdev_priv(netdev);
4770 struct mii_ioctl_data *data = if_mii(ifr);
4772 if (adapter->hw.phy.media_type != e1000_media_type_copper)
4777 data->phy_id = adapter->hw.phy.addr;
4780 if (!capable(CAP_NET_ADMIN))
4782 if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
4794 * igb_hwtstamp_ioctl - control hardware time stamping
4799 * Outgoing time stamping can be enabled and disabled. Play nice and
4800 * disable it when requested, although it shouldn't case any overhead
4801 * when no packet needs it. At most one packet in the queue may be
4802 * marked for time stamping, otherwise it would be impossible to tell
4803 * for sure to which packet the hardware time stamp belongs.
4805 * Incoming time stamping has to be configured via the hardware
4806 * filters. Not all combinations are supported, in particular event
4807 * type has to be specified. Matching the kind of event packet is
4808 * not supported, with the exception of "all V2 events regardless of
4812 static int igb_hwtstamp_ioctl(struct net_device *netdev,
4813 struct ifreq *ifr, int cmd)
4815 struct igb_adapter *adapter = netdev_priv(netdev);
4816 struct e1000_hw *hw = &adapter->hw;
4817 struct hwtstamp_config config;
4818 u32 tsync_tx_ctl_bit = E1000_TSYNCTXCTL_ENABLED;
4819 u32 tsync_rx_ctl_bit = E1000_TSYNCRXCTL_ENABLED;
4820 u32 tsync_rx_ctl_type = 0;
4821 u32 tsync_rx_cfg = 0;
4824 short port = 319; /* PTP */
4827 if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
4830 /* reserved for future extensions */
4834 switch (config.tx_type) {
4835 case HWTSTAMP_TX_OFF:
4836 tsync_tx_ctl_bit = 0;
4838 case HWTSTAMP_TX_ON:
4839 tsync_tx_ctl_bit = E1000_TSYNCTXCTL_ENABLED;
4845 switch (config.rx_filter) {
4846 case HWTSTAMP_FILTER_NONE:
4847 tsync_rx_ctl_bit = 0;
4849 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
4850 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
4851 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
4852 case HWTSTAMP_FILTER_ALL:
4854 * register TSYNCRXCFG must be set, therefore it is not
4855 * possible to time stamp both Sync and Delay_Req messages
4856 * => fall back to time stamping all packets
4858 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_ALL;
4859 config.rx_filter = HWTSTAMP_FILTER_ALL;
4861 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
4862 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_L4_V1;
4863 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V1_SYNC_MESSAGE;
4866 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
4867 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_L4_V1;
4868 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V1_DELAY_REQ_MESSAGE;
4871 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
4872 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
4873 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
4874 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V2_SYNC_MESSAGE;
4877 config.rx_filter = HWTSTAMP_FILTER_SOME;
4879 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
4880 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
4881 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
4882 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V2_DELAY_REQ_MESSAGE;
4885 config.rx_filter = HWTSTAMP_FILTER_SOME;
4887 case HWTSTAMP_FILTER_PTP_V2_EVENT:
4888 case HWTSTAMP_FILTER_PTP_V2_SYNC:
4889 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
4890 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_EVENT_V2;
4891 config.rx_filter = HWTSTAMP_FILTER_PTP_V2_EVENT;
4898 /* enable/disable TX */
4899 regval = rd32(E1000_TSYNCTXCTL);
4900 regval = (regval & ~E1000_TSYNCTXCTL_ENABLED) | tsync_tx_ctl_bit;
4901 wr32(E1000_TSYNCTXCTL, regval);
4903 /* enable/disable RX, define which PTP packets are time stamped */
4904 regval = rd32(E1000_TSYNCRXCTL);
4905 regval = (regval & ~E1000_TSYNCRXCTL_ENABLED) | tsync_rx_ctl_bit;
4906 regval = (regval & ~0xE) | tsync_rx_ctl_type;
4907 wr32(E1000_TSYNCRXCTL, regval);
4908 wr32(E1000_TSYNCRXCFG, tsync_rx_cfg);
4911 * Ethertype Filter Queue Filter[0][15:0] = 0x88F7
4912 * (Ethertype to filter on)
4913 * Ethertype Filter Queue Filter[0][26] = 0x1 (Enable filter)
4914 * Ethertype Filter Queue Filter[0][30] = 0x1 (Enable Timestamping)
4916 wr32(E1000_ETQF0, is_l2 ? 0x440088f7 : 0);
4918 /* L4 Queue Filter[0]: only filter by source and destination port */
4919 wr32(E1000_SPQF0, htons(port));
4920 wr32(E1000_IMIREXT(0), is_l4 ?
4921 ((1<<12) | (1<<19) /* bypass size and control flags */) : 0);
4922 wr32(E1000_IMIR(0), is_l4 ?
4924 | (0<<16) /* immediate interrupt disabled */
4925 | 0 /* (1<<17) bit cleared: do not bypass
4926 destination port check */)
4928 wr32(E1000_FTQF0, is_l4 ?
4930 | (1<<15) /* VF not compared */
4931 | (1<<27) /* Enable Timestamping */
4932 | (7<<28) /* only source port filter enabled,
4933 source/target address and protocol
4935 : ((1<<15) | (15<<28) /* all mask bits set = filter not
4940 adapter->hwtstamp_config = config;
4942 /* clear TX/RX time stamp registers, just to be sure */
4943 regval = rd32(E1000_TXSTMPH);
4944 regval = rd32(E1000_RXSTMPH);
4946 return copy_to_user(ifr->ifr_data, &config, sizeof(config)) ?
4956 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4962 return igb_mii_ioctl(netdev, ifr, cmd);
4964 return igb_hwtstamp_ioctl(netdev, ifr, cmd);
4970 static void igb_vlan_rx_register(struct net_device *netdev,
4971 struct vlan_group *grp)
4973 struct igb_adapter *adapter = netdev_priv(netdev);
4974 struct e1000_hw *hw = &adapter->hw;
4977 igb_irq_disable(adapter);
4978 adapter->vlgrp = grp;
4981 /* enable VLAN tag insert/strip */
4982 ctrl = rd32(E1000_CTRL);
4983 ctrl |= E1000_CTRL_VME;
4984 wr32(E1000_CTRL, ctrl);
4986 /* enable VLAN receive filtering */
4987 rctl = rd32(E1000_RCTL);
4988 rctl &= ~E1000_RCTL_CFIEN;
4989 wr32(E1000_RCTL, rctl);
4990 igb_update_mng_vlan(adapter);
4992 /* disable VLAN tag insert/strip */
4993 ctrl = rd32(E1000_CTRL);
4994 ctrl &= ~E1000_CTRL_VME;
4995 wr32(E1000_CTRL, ctrl);
4997 if (adapter->mng_vlan_id != (u16)IGB_MNG_VLAN_NONE) {
4998 igb_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
4999 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
5003 igb_rlpml_set(adapter);
5005 if (!test_bit(__IGB_DOWN, &adapter->state))
5006 igb_irq_enable(adapter);
5009 static void igb_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
5011 struct igb_adapter *adapter = netdev_priv(netdev);
5012 struct e1000_hw *hw = &adapter->hw;
5013 int pf_id = adapter->vfs_allocated_count;
5015 if ((hw->mng_cookie.status &
5016 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
5017 (vid == adapter->mng_vlan_id))
5020 /* add vid to vlvf if sr-iov is enabled,
5021 * if that fails add directly to filter table */
5022 if (igb_vlvf_set(adapter, vid, true, pf_id))
5023 igb_vfta_set(hw, vid, true);
5027 static void igb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
5029 struct igb_adapter *adapter = netdev_priv(netdev);
5030 struct e1000_hw *hw = &adapter->hw;
5031 int pf_id = adapter->vfs_allocated_count;
5033 igb_irq_disable(adapter);
5034 vlan_group_set_device(adapter->vlgrp, vid, NULL);
5036 if (!test_bit(__IGB_DOWN, &adapter->state))
5037 igb_irq_enable(adapter);
5039 if ((adapter->hw.mng_cookie.status &
5040 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
5041 (vid == adapter->mng_vlan_id)) {
5042 /* release control to f/w */
5043 igb_release_hw_control(adapter);
5047 /* remove vid from vlvf if sr-iov is enabled,
5048 * if not in vlvf remove from vfta */
5049 if (igb_vlvf_set(adapter, vid, false, pf_id))
5050 igb_vfta_set(hw, vid, false);
5053 static void igb_restore_vlan(struct igb_adapter *adapter)
5055 igb_vlan_rx_register(adapter->netdev, adapter->vlgrp);
5057 if (adapter->vlgrp) {
5059 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
5060 if (!vlan_group_get_device(adapter->vlgrp, vid))
5062 igb_vlan_rx_add_vid(adapter->netdev, vid);
5067 int igb_set_spd_dplx(struct igb_adapter *adapter, u16 spddplx)
5069 struct e1000_mac_info *mac = &adapter->hw.mac;
5073 /* Fiber NICs only allow 1000 gbps Full duplex */
5074 if ((adapter->hw.phy.media_type == e1000_media_type_fiber) &&
5075 spddplx != (SPEED_1000 + DUPLEX_FULL)) {
5076 dev_err(&adapter->pdev->dev,
5077 "Unsupported Speed/Duplex configuration\n");
5082 case SPEED_10 + DUPLEX_HALF:
5083 mac->forced_speed_duplex = ADVERTISE_10_HALF;
5085 case SPEED_10 + DUPLEX_FULL:
5086 mac->forced_speed_duplex = ADVERTISE_10_FULL;
5088 case SPEED_100 + DUPLEX_HALF:
5089 mac->forced_speed_duplex = ADVERTISE_100_HALF;
5091 case SPEED_100 + DUPLEX_FULL:
5092 mac->forced_speed_duplex = ADVERTISE_100_FULL;
5094 case SPEED_1000 + DUPLEX_FULL:
5096 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
5098 case SPEED_1000 + DUPLEX_HALF: /* not supported */
5100 dev_err(&adapter->pdev->dev,
5101 "Unsupported Speed/Duplex configuration\n");
5107 static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake)
5109 struct net_device *netdev = pci_get_drvdata(pdev);
5110 struct igb_adapter *adapter = netdev_priv(netdev);
5111 struct e1000_hw *hw = &adapter->hw;
5112 u32 ctrl, rctl, status;
5113 u32 wufc = adapter->wol;
5118 netif_device_detach(netdev);
5120 if (netif_running(netdev))
5123 igb_reset_interrupt_capability(adapter);
5125 igb_free_queues(adapter);
5128 retval = pci_save_state(pdev);
5133 status = rd32(E1000_STATUS);
5134 if (status & E1000_STATUS_LU)
5135 wufc &= ~E1000_WUFC_LNKC;
5138 igb_setup_rctl(adapter);
5139 igb_set_multi(netdev);
5141 /* turn on all-multi mode if wake on multicast is enabled */
5142 if (wufc & E1000_WUFC_MC) {
5143 rctl = rd32(E1000_RCTL);
5144 rctl |= E1000_RCTL_MPE;
5145 wr32(E1000_RCTL, rctl);
5148 ctrl = rd32(E1000_CTRL);
5149 /* advertise wake from D3Cold */
5150 #define E1000_CTRL_ADVD3WUC 0x00100000
5151 /* phy power management enable */
5152 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5153 ctrl |= E1000_CTRL_ADVD3WUC;
5154 wr32(E1000_CTRL, ctrl);
5156 /* Allow time for pending master requests to run */
5157 igb_disable_pcie_master(&adapter->hw);
5159 wr32(E1000_WUC, E1000_WUC_PME_EN);
5160 wr32(E1000_WUFC, wufc);
5163 wr32(E1000_WUFC, 0);
5166 *enable_wake = wufc || adapter->en_mng_pt;
5168 igb_shutdown_fiber_serdes_link_82575(hw);
5170 /* Release control of h/w to f/w. If f/w is AMT enabled, this
5171 * would have already happened in close and is redundant. */
5172 igb_release_hw_control(adapter);
5174 pci_disable_device(pdev);
5180 static int igb_suspend(struct pci_dev *pdev, pm_message_t state)
5185 retval = __igb_shutdown(pdev, &wake);
5190 pci_prepare_to_sleep(pdev);
5192 pci_wake_from_d3(pdev, false);
5193 pci_set_power_state(pdev, PCI_D3hot);
5199 static int igb_resume(struct pci_dev *pdev)
5201 struct net_device *netdev = pci_get_drvdata(pdev);
5202 struct igb_adapter *adapter = netdev_priv(netdev);
5203 struct e1000_hw *hw = &adapter->hw;
5206 pci_set_power_state(pdev, PCI_D0);
5207 pci_restore_state(pdev);
5209 err = pci_enable_device_mem(pdev);
5212 "igb: Cannot enable PCI device from suspend\n");
5215 pci_set_master(pdev);
5217 pci_enable_wake(pdev, PCI_D3hot, 0);
5218 pci_enable_wake(pdev, PCI_D3cold, 0);
5220 igb_set_interrupt_capability(adapter);
5222 if (igb_alloc_queues(adapter)) {
5223 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
5227 /* e1000_power_up_phy(adapter); */
5231 /* let the f/w know that the h/w is now under the control of the
5233 igb_get_hw_control(adapter);
5235 wr32(E1000_WUS, ~0);
5237 if (netif_running(netdev)) {
5238 err = igb_open(netdev);
5243 netif_device_attach(netdev);
5249 static void igb_shutdown(struct pci_dev *pdev)
5253 __igb_shutdown(pdev, &wake);
5255 if (system_state == SYSTEM_POWER_OFF) {
5256 pci_wake_from_d3(pdev, wake);
5257 pci_set_power_state(pdev, PCI_D3hot);
5261 #ifdef CONFIG_NET_POLL_CONTROLLER
5263 * Polling 'interrupt' - used by things like netconsole to send skbs
5264 * without having to re-enable interrupts. It's not called while
5265 * the interrupt routine is executing.
5267 static void igb_netpoll(struct net_device *netdev)
5269 struct igb_adapter *adapter = netdev_priv(netdev);
5270 struct e1000_hw *hw = &adapter->hw;
5273 if (!adapter->msix_entries) {
5274 igb_irq_disable(adapter);
5275 napi_schedule(&adapter->rx_ring[0].napi);
5279 for (i = 0; i < adapter->num_tx_queues; i++) {
5280 struct igb_ring *tx_ring = &adapter->tx_ring[i];
5281 wr32(E1000_EIMC, tx_ring->eims_value);
5282 igb_clean_tx_irq(tx_ring);
5283 wr32(E1000_EIMS, tx_ring->eims_value);
5286 for (i = 0; i < adapter->num_rx_queues; i++) {
5287 struct igb_ring *rx_ring = &adapter->rx_ring[i];
5288 wr32(E1000_EIMC, rx_ring->eims_value);
5289 napi_schedule(&rx_ring->napi);
5292 #endif /* CONFIG_NET_POLL_CONTROLLER */
5295 * igb_io_error_detected - called when PCI error is detected
5296 * @pdev: Pointer to PCI device
5297 * @state: The current pci connection state
5299 * This function is called after a PCI bus error affecting
5300 * this device has been detected.
5302 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
5303 pci_channel_state_t state)
5305 struct net_device *netdev = pci_get_drvdata(pdev);
5306 struct igb_adapter *adapter = netdev_priv(netdev);
5308 netif_device_detach(netdev);
5310 if (netif_running(netdev))
5312 pci_disable_device(pdev);
5314 /* Request a slot slot reset. */
5315 return PCI_ERS_RESULT_NEED_RESET;
5319 * igb_io_slot_reset - called after the pci bus has been reset.
5320 * @pdev: Pointer to PCI device
5322 * Restart the card from scratch, as if from a cold-boot. Implementation
5323 * resembles the first-half of the igb_resume routine.
5325 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
5327 struct net_device *netdev = pci_get_drvdata(pdev);
5328 struct igb_adapter *adapter = netdev_priv(netdev);
5329 struct e1000_hw *hw = &adapter->hw;
5330 pci_ers_result_t result;
5333 if (pci_enable_device_mem(pdev)) {
5335 "Cannot re-enable PCI device after reset.\n");
5336 result = PCI_ERS_RESULT_DISCONNECT;
5338 pci_set_master(pdev);
5339 pci_restore_state(pdev);
5341 pci_enable_wake(pdev, PCI_D3hot, 0);
5342 pci_enable_wake(pdev, PCI_D3cold, 0);
5345 wr32(E1000_WUS, ~0);
5346 result = PCI_ERS_RESULT_RECOVERED;
5349 err = pci_cleanup_aer_uncorrect_error_status(pdev);
5351 dev_err(&pdev->dev, "pci_cleanup_aer_uncorrect_error_status "
5352 "failed 0x%0x\n", err);
5353 /* non-fatal, continue */
5360 * igb_io_resume - called when traffic can start flowing again.
5361 * @pdev: Pointer to PCI device
5363 * This callback is called when the error recovery driver tells us that
5364 * its OK to resume normal operation. Implementation resembles the
5365 * second-half of the igb_resume routine.
5367 static void igb_io_resume(struct pci_dev *pdev)
5369 struct net_device *netdev = pci_get_drvdata(pdev);
5370 struct igb_adapter *adapter = netdev_priv(netdev);
5372 if (netif_running(netdev)) {
5373 if (igb_up(adapter)) {
5374 dev_err(&pdev->dev, "igb_up failed after reset\n");
5379 netif_device_attach(netdev);
5381 /* let the f/w know that the h/w is now under the control of the
5383 igb_get_hw_control(adapter);
5386 static inline void igb_set_vmolr(struct e1000_hw *hw, int vfn)
5390 reg_data = rd32(E1000_VMOLR(vfn));
5391 reg_data |= E1000_VMOLR_BAM | /* Accept broadcast */
5392 E1000_VMOLR_ROPE | /* Accept packets matched in UTA */
5393 E1000_VMOLR_ROMPE | /* Accept packets matched in MTA */
5394 E1000_VMOLR_AUPE | /* Accept untagged packets */
5395 E1000_VMOLR_STRVLAN; /* Strip vlan tags */
5396 wr32(E1000_VMOLR(vfn), reg_data);
5399 static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
5402 struct e1000_hw *hw = &adapter->hw;
5405 vmolr = rd32(E1000_VMOLR(vfn));
5406 vmolr &= ~E1000_VMOLR_RLPML_MASK;
5407 vmolr |= size | E1000_VMOLR_LPE;
5408 wr32(E1000_VMOLR(vfn), vmolr);
5413 static inline void igb_set_rah_pool(struct e1000_hw *hw, int pool, int entry)
5417 reg_data = rd32(E1000_RAH(entry));
5418 reg_data &= ~E1000_RAH_POOL_MASK;
5419 reg_data |= E1000_RAH_POOL_1 << pool;;
5420 wr32(E1000_RAH(entry), reg_data);
5423 static void igb_set_mc_list_pools(struct igb_adapter *adapter,
5424 int entry_count, u16 total_rar_filters)
5426 struct e1000_hw *hw = &adapter->hw;
5427 int i = adapter->vfs_allocated_count + 1;
5429 if ((i + entry_count) < total_rar_filters)
5430 total_rar_filters = i + entry_count;
5432 for (; i < total_rar_filters; i++)
5433 igb_set_rah_pool(hw, adapter->vfs_allocated_count, i);
5436 static int igb_set_vf_mac(struct igb_adapter *adapter,
5437 int vf, unsigned char *mac_addr)
5439 struct e1000_hw *hw = &adapter->hw;
5440 int rar_entry = vf + 1; /* VF MAC addresses start at entry 1 */
5442 igb_rar_set(hw, mac_addr, rar_entry);
5444 memcpy(adapter->vf_data[vf].vf_mac_addresses, mac_addr, ETH_ALEN);
5446 igb_set_rah_pool(hw, vf, rar_entry);
5451 static void igb_vmm_control(struct igb_adapter *adapter)
5453 struct e1000_hw *hw = &adapter->hw;
5456 if (!adapter->vfs_allocated_count)
5459 /* VF's need PF reset indication before they
5460 * can send/receive mail */
5461 reg_data = rd32(E1000_CTRL_EXT);
5462 reg_data |= E1000_CTRL_EXT_PFRSTD;
5463 wr32(E1000_CTRL_EXT, reg_data);
5465 igb_vmdq_set_loopback_pf(hw, true);
5466 igb_vmdq_set_replication_pf(hw, true);