1 /*******************************************************************************
3 Intel(R) Gigabit Ethernet Linux driver
4 Copyright(c) 2007-2009 Intel Corporation.
6 This program is free software; you can redistribute it and/or modify it
7 under the terms and conditions of the GNU General Public License,
8 version 2, as published by the Free Software Foundation.
10 This program is distributed in the hope it will be useful, but WITHOUT
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
23 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26 *******************************************************************************/
28 #include <linux/module.h>
29 #include <linux/types.h>
30 #include <linux/init.h>
31 #include <linux/vmalloc.h>
32 #include <linux/pagemap.h>
33 #include <linux/netdevice.h>
34 #include <linux/ipv6.h>
35 #include <net/checksum.h>
36 #include <net/ip6_checksum.h>
37 #include <linux/net_tstamp.h>
38 #include <linux/mii.h>
39 #include <linux/ethtool.h>
40 #include <linux/if_vlan.h>
41 #include <linux/pci.h>
42 #include <linux/pci-aspm.h>
43 #include <linux/delay.h>
44 #include <linux/interrupt.h>
45 #include <linux/if_ether.h>
46 #include <linux/aer.h>
48 #include <linux/dca.h>
52 #define DRV_VERSION "1.3.16-k2"
53 char igb_driver_name[] = "igb";
54 char igb_driver_version[] = DRV_VERSION;
55 static const char igb_driver_string[] =
56 "Intel(R) Gigabit Ethernet Network Driver";
57 static const char igb_copyright[] = "Copyright (c) 2007-2009 Intel Corporation.";
59 static const struct e1000_info *igb_info_tbl[] = {
60 [board_82575] = &e1000_82575_info,
63 static struct pci_device_id igb_pci_tbl[] = {
64 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576), board_82575 },
65 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS), board_82575 },
66 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_FIBER), board_82575 },
67 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES), board_82575 },
68 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER), board_82575 },
69 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
70 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
71 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
72 /* required last entry */
76 MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
78 void igb_reset(struct igb_adapter *);
79 static int igb_setup_all_tx_resources(struct igb_adapter *);
80 static int igb_setup_all_rx_resources(struct igb_adapter *);
81 static void igb_free_all_tx_resources(struct igb_adapter *);
82 static void igb_free_all_rx_resources(struct igb_adapter *);
83 void igb_update_stats(struct igb_adapter *);
84 static int igb_probe(struct pci_dev *, const struct pci_device_id *);
85 static void __devexit igb_remove(struct pci_dev *pdev);
86 static int igb_sw_init(struct igb_adapter *);
87 static int igb_open(struct net_device *);
88 static int igb_close(struct net_device *);
89 static void igb_configure_tx(struct igb_adapter *);
90 static void igb_configure_rx(struct igb_adapter *);
91 static void igb_setup_rctl(struct igb_adapter *);
92 static void igb_clean_all_tx_rings(struct igb_adapter *);
93 static void igb_clean_all_rx_rings(struct igb_adapter *);
94 static void igb_clean_tx_ring(struct igb_ring *);
95 static void igb_clean_rx_ring(struct igb_ring *);
96 static void igb_set_multi(struct net_device *);
97 static void igb_update_phy_info(unsigned long);
98 static void igb_watchdog(unsigned long);
99 static void igb_watchdog_task(struct work_struct *);
100 static int igb_xmit_frame_ring_adv(struct sk_buff *, struct net_device *,
102 static int igb_xmit_frame_adv(struct sk_buff *skb, struct net_device *);
103 static struct net_device_stats *igb_get_stats(struct net_device *);
104 static int igb_change_mtu(struct net_device *, int);
105 static int igb_set_mac(struct net_device *, void *);
106 static irqreturn_t igb_intr(int irq, void *);
107 static irqreturn_t igb_intr_msi(int irq, void *);
108 static irqreturn_t igb_msix_other(int irq, void *);
109 static irqreturn_t igb_msix_rx(int irq, void *);
110 static irqreturn_t igb_msix_tx(int irq, void *);
111 #ifdef CONFIG_IGB_DCA
112 static void igb_update_rx_dca(struct igb_ring *);
113 static void igb_update_tx_dca(struct igb_ring *);
114 static void igb_setup_dca(struct igb_adapter *);
115 #endif /* CONFIG_IGB_DCA */
116 static bool igb_clean_tx_irq(struct igb_ring *);
117 static int igb_poll(struct napi_struct *, int);
118 static bool igb_clean_rx_irq_adv(struct igb_ring *, int *, int);
119 static void igb_alloc_rx_buffers_adv(struct igb_ring *, int);
120 static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
121 static void igb_tx_timeout(struct net_device *);
122 static void igb_reset_task(struct work_struct *);
123 static void igb_vlan_rx_register(struct net_device *, struct vlan_group *);
124 static void igb_vlan_rx_add_vid(struct net_device *, u16);
125 static void igb_vlan_rx_kill_vid(struct net_device *, u16);
126 static void igb_restore_vlan(struct igb_adapter *);
127 static void igb_ping_all_vfs(struct igb_adapter *);
128 static void igb_msg_task(struct igb_adapter *);
129 static int igb_rcv_msg_from_vf(struct igb_adapter *, u32);
130 static inline void igb_set_rah_pool(struct e1000_hw *, int , int);
131 static void igb_set_mc_list_pools(struct igb_adapter *, int, u16);
132 static void igb_vmm_control(struct igb_adapter *);
133 static inline void igb_set_vmolr(struct e1000_hw *, int);
134 static inline int igb_set_vf_rlpml(struct igb_adapter *, int, int);
135 static int igb_set_vf_mac(struct igb_adapter *adapter, int, unsigned char *);
136 static void igb_restore_vf_multicasts(struct igb_adapter *adapter);
139 static int igb_suspend(struct pci_dev *, pm_message_t);
140 static int igb_resume(struct pci_dev *);
142 static void igb_shutdown(struct pci_dev *);
143 #ifdef CONFIG_IGB_DCA
144 static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
145 static struct notifier_block dca_notifier = {
146 .notifier_call = igb_notify_dca,
151 #ifdef CONFIG_NET_POLL_CONTROLLER
152 /* for netdump / net console */
153 static void igb_netpoll(struct net_device *);
155 #ifdef CONFIG_PCI_IOV
156 static unsigned int max_vfs = 0;
157 module_param(max_vfs, uint, 0);
158 MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate "
159 "per physical function");
160 #endif /* CONFIG_PCI_IOV */
162 static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
163 pci_channel_state_t);
164 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
165 static void igb_io_resume(struct pci_dev *);
167 static struct pci_error_handlers igb_err_handler = {
168 .error_detected = igb_io_error_detected,
169 .slot_reset = igb_io_slot_reset,
170 .resume = igb_io_resume,
174 static struct pci_driver igb_driver = {
175 .name = igb_driver_name,
176 .id_table = igb_pci_tbl,
178 .remove = __devexit_p(igb_remove),
180 /* Power Managment Hooks */
181 .suspend = igb_suspend,
182 .resume = igb_resume,
184 .shutdown = igb_shutdown,
185 .err_handler = &igb_err_handler
188 static int global_quad_port_a; /* global quad port a indication */
190 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
191 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
192 MODULE_LICENSE("GPL");
193 MODULE_VERSION(DRV_VERSION);
196 * Scale the NIC clock cycle by a large factor so that
197 * relatively small clock corrections can be added or
198 * substracted at each clock tick. The drawbacks of a
199 * large factor are a) that the clock register overflows
200 * more quickly (not such a big deal) and b) that the
201 * increment per tick has to fit into 24 bits.
204 * TIMINCA = IGB_TSYNC_CYCLE_TIME_IN_NANOSECONDS *
206 * TIMINCA += TIMINCA * adjustment [ppm] / 1e9
208 * The base scale factor is intentionally a power of two
209 * so that the division in %struct timecounter can be done with
212 #define IGB_TSYNC_SHIFT (19)
213 #define IGB_TSYNC_SCALE (1<<IGB_TSYNC_SHIFT)
216 * The duration of one clock cycle of the NIC.
218 * @todo This hard-coded value is part of the specification and might change
219 * in future hardware revisions. Add revision check.
221 #define IGB_TSYNC_CYCLE_TIME_IN_NANOSECONDS 16
223 #if (IGB_TSYNC_SCALE * IGB_TSYNC_CYCLE_TIME_IN_NANOSECONDS) >= (1<<24)
224 # error IGB_TSYNC_SCALE and/or IGB_TSYNC_CYCLE_TIME_IN_NANOSECONDS are too large to fit into TIMINCA
228 * igb_read_clock - read raw cycle counter (to be used by time counter)
230 static cycle_t igb_read_clock(const struct cyclecounter *tc)
232 struct igb_adapter *adapter =
233 container_of(tc, struct igb_adapter, cycles);
234 struct e1000_hw *hw = &adapter->hw;
237 stamp = rd32(E1000_SYSTIML);
238 stamp |= (u64)rd32(E1000_SYSTIMH) << 32ULL;
245 * igb_get_hw_dev_name - return device name string
246 * used by hardware layer to print debugging information
248 char *igb_get_hw_dev_name(struct e1000_hw *hw)
250 struct igb_adapter *adapter = hw->back;
251 return adapter->netdev->name;
255 * igb_get_time_str - format current NIC and system time as string
257 static char *igb_get_time_str(struct igb_adapter *adapter,
260 cycle_t hw = adapter->cycles.read(&adapter->cycles);
261 struct timespec nic = ns_to_timespec(timecounter_read(&adapter->clock));
263 struct timespec delta;
264 getnstimeofday(&sys);
266 delta = timespec_sub(nic, sys);
269 "HW %llu, NIC %ld.%09lus, SYS %ld.%09lus, NIC-SYS %lds + %09luns",
271 (long)nic.tv_sec, nic.tv_nsec,
272 (long)sys.tv_sec, sys.tv_nsec,
273 (long)delta.tv_sec, delta.tv_nsec);
280 * igb_desc_unused - calculate if we have unused descriptors
282 static int igb_desc_unused(struct igb_ring *ring)
284 if (ring->next_to_clean > ring->next_to_use)
285 return ring->next_to_clean - ring->next_to_use - 1;
287 return ring->count + ring->next_to_clean - ring->next_to_use - 1;
291 * igb_init_module - Driver Registration Routine
293 * igb_init_module is the first routine called when the driver is
294 * loaded. All it does is register with the PCI subsystem.
296 static int __init igb_init_module(void)
299 printk(KERN_INFO "%s - version %s\n",
300 igb_driver_string, igb_driver_version);
302 printk(KERN_INFO "%s\n", igb_copyright);
304 global_quad_port_a = 0;
306 #ifdef CONFIG_IGB_DCA
307 dca_register_notify(&dca_notifier);
310 ret = pci_register_driver(&igb_driver);
314 module_init(igb_init_module);
317 * igb_exit_module - Driver Exit Cleanup Routine
319 * igb_exit_module is called just before the driver is removed
322 static void __exit igb_exit_module(void)
324 #ifdef CONFIG_IGB_DCA
325 dca_unregister_notify(&dca_notifier);
327 pci_unregister_driver(&igb_driver);
330 module_exit(igb_exit_module);
332 #define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
334 * igb_cache_ring_register - Descriptor ring to register mapping
335 * @adapter: board private structure to initialize
337 * Once we know the feature-set enabled for the device, we'll cache
338 * the register offset the descriptor ring is assigned to.
340 static void igb_cache_ring_register(struct igb_adapter *adapter)
343 unsigned int rbase_offset = adapter->vfs_allocated_count;
345 switch (adapter->hw.mac.type) {
347 /* The queues are allocated for virtualization such that VF 0
348 * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
349 * In order to avoid collision we start at the first free queue
350 * and continue consuming queues in the same sequence
352 for (i = 0; i < adapter->num_rx_queues; i++)
353 adapter->rx_ring[i].reg_idx = rbase_offset +
355 for (i = 0; i < adapter->num_tx_queues; i++)
356 adapter->tx_ring[i].reg_idx = rbase_offset +
361 for (i = 0; i < adapter->num_rx_queues; i++)
362 adapter->rx_ring[i].reg_idx = i;
363 for (i = 0; i < adapter->num_tx_queues; i++)
364 adapter->tx_ring[i].reg_idx = i;
370 * igb_alloc_queues - Allocate memory for all rings
371 * @adapter: board private structure to initialize
373 * We allocate one ring per queue at run-time since we don't know the
374 * number of queues at compile-time.
376 static int igb_alloc_queues(struct igb_adapter *adapter)
380 adapter->tx_ring = kcalloc(adapter->num_tx_queues,
381 sizeof(struct igb_ring), GFP_KERNEL);
382 if (!adapter->tx_ring)
385 adapter->rx_ring = kcalloc(adapter->num_rx_queues,
386 sizeof(struct igb_ring), GFP_KERNEL);
387 if (!adapter->rx_ring) {
388 kfree(adapter->tx_ring);
392 adapter->rx_ring->buddy = adapter->tx_ring;
394 for (i = 0; i < adapter->num_tx_queues; i++) {
395 struct igb_ring *ring = &(adapter->tx_ring[i]);
396 ring->count = adapter->tx_ring_count;
397 ring->adapter = adapter;
398 ring->queue_index = i;
400 for (i = 0; i < adapter->num_rx_queues; i++) {
401 struct igb_ring *ring = &(adapter->rx_ring[i]);
402 ring->count = adapter->rx_ring_count;
403 ring->adapter = adapter;
404 ring->queue_index = i;
405 ring->itr_register = E1000_ITR;
407 /* set a default napi handler for each rx_ring */
408 netif_napi_add(adapter->netdev, &ring->napi, igb_poll, 64);
411 igb_cache_ring_register(adapter);
415 static void igb_free_queues(struct igb_adapter *adapter)
419 for (i = 0; i < adapter->num_rx_queues; i++)
420 netif_napi_del(&adapter->rx_ring[i].napi);
422 adapter->num_rx_queues = 0;
423 adapter->num_tx_queues = 0;
425 kfree(adapter->tx_ring);
426 kfree(adapter->rx_ring);
429 #define IGB_N0_QUEUE -1
430 static void igb_assign_vector(struct igb_adapter *adapter, int rx_queue,
431 int tx_queue, int msix_vector)
434 struct e1000_hw *hw = &adapter->hw;
437 switch (hw->mac.type) {
439 /* The 82575 assigns vectors using a bitmask, which matches the
440 bitmask for the EICR/EIMS/EIMC registers. To assign one
441 or more queues to a vector, we write the appropriate bits
442 into the MSIXBM register for that vector. */
443 if (rx_queue > IGB_N0_QUEUE) {
444 msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
445 adapter->rx_ring[rx_queue].eims_value = msixbm;
447 if (tx_queue > IGB_N0_QUEUE) {
448 msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
449 adapter->tx_ring[tx_queue].eims_value =
450 E1000_EICR_TX_QUEUE0 << tx_queue;
452 array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
455 /* 82576 uses a table-based method for assigning vectors.
456 Each queue has a single entry in the table to which we write
457 a vector number along with a "valid" bit. Sadly, the layout
458 of the table is somewhat counterintuitive. */
459 if (rx_queue > IGB_N0_QUEUE) {
460 index = (rx_queue >> 1) + adapter->vfs_allocated_count;
461 ivar = array_rd32(E1000_IVAR0, index);
462 if (rx_queue & 0x1) {
463 /* vector goes into third byte of register */
464 ivar = ivar & 0xFF00FFFF;
465 ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
467 /* vector goes into low byte of register */
468 ivar = ivar & 0xFFFFFF00;
469 ivar |= msix_vector | E1000_IVAR_VALID;
471 adapter->rx_ring[rx_queue].eims_value= 1 << msix_vector;
472 array_wr32(E1000_IVAR0, index, ivar);
474 if (tx_queue > IGB_N0_QUEUE) {
475 index = (tx_queue >> 1) + adapter->vfs_allocated_count;
476 ivar = array_rd32(E1000_IVAR0, index);
477 if (tx_queue & 0x1) {
478 /* vector goes into high byte of register */
479 ivar = ivar & 0x00FFFFFF;
480 ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
482 /* vector goes into second byte of register */
483 ivar = ivar & 0xFFFF00FF;
484 ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
486 adapter->tx_ring[tx_queue].eims_value= 1 << msix_vector;
487 array_wr32(E1000_IVAR0, index, ivar);
497 * igb_configure_msix - Configure MSI-X hardware
499 * igb_configure_msix sets up the hardware to properly
500 * generate MSI-X interrupts.
502 static void igb_configure_msix(struct igb_adapter *adapter)
506 struct e1000_hw *hw = &adapter->hw;
508 adapter->eims_enable_mask = 0;
509 if (hw->mac.type == e1000_82576)
510 /* Turn on MSI-X capability first, or our settings
511 * won't stick. And it will take days to debug. */
512 wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
513 E1000_GPIE_PBA | E1000_GPIE_EIAME |
516 for (i = 0; i < adapter->num_tx_queues; i++) {
517 struct igb_ring *tx_ring = &adapter->tx_ring[i];
518 igb_assign_vector(adapter, IGB_N0_QUEUE, i, vector++);
519 adapter->eims_enable_mask |= tx_ring->eims_value;
520 if (tx_ring->itr_val)
521 writel(tx_ring->itr_val,
522 hw->hw_addr + tx_ring->itr_register);
524 writel(1, hw->hw_addr + tx_ring->itr_register);
527 for (i = 0; i < adapter->num_rx_queues; i++) {
528 struct igb_ring *rx_ring = &adapter->rx_ring[i];
529 rx_ring->buddy = NULL;
530 igb_assign_vector(adapter, i, IGB_N0_QUEUE, vector++);
531 adapter->eims_enable_mask |= rx_ring->eims_value;
532 if (rx_ring->itr_val)
533 writel(rx_ring->itr_val,
534 hw->hw_addr + rx_ring->itr_register);
536 writel(1, hw->hw_addr + rx_ring->itr_register);
540 /* set vector for other causes, i.e. link changes */
541 switch (hw->mac.type) {
543 array_wr32(E1000_MSIXBM(0), vector++,
546 tmp = rd32(E1000_CTRL_EXT);
547 /* enable MSI-X PBA support*/
548 tmp |= E1000_CTRL_EXT_PBA_CLR;
550 /* Auto-Mask interrupts upon ICR read. */
551 tmp |= E1000_CTRL_EXT_EIAME;
552 tmp |= E1000_CTRL_EXT_IRCA;
554 wr32(E1000_CTRL_EXT, tmp);
555 adapter->eims_enable_mask |= E1000_EIMS_OTHER;
556 adapter->eims_other = E1000_EIMS_OTHER;
561 tmp = (vector++ | E1000_IVAR_VALID) << 8;
562 wr32(E1000_IVAR_MISC, tmp);
564 adapter->eims_enable_mask = (1 << (vector)) - 1;
565 adapter->eims_other = 1 << (vector - 1);
568 /* do nothing, since nothing else supports MSI-X */
570 } /* switch (hw->mac.type) */
575 * igb_request_msix - Initialize MSI-X interrupts
577 * igb_request_msix allocates MSI-X vectors and requests interrupts from the
580 static int igb_request_msix(struct igb_adapter *adapter)
582 struct net_device *netdev = adapter->netdev;
583 int i, err = 0, vector = 0;
587 for (i = 0; i < adapter->num_tx_queues; i++) {
588 struct igb_ring *ring = &(adapter->tx_ring[i]);
589 sprintf(ring->name, "%s-tx-%d", netdev->name, i);
590 err = request_irq(adapter->msix_entries[vector].vector,
591 &igb_msix_tx, 0, ring->name,
592 &(adapter->tx_ring[i]));
595 ring->itr_register = E1000_EITR(0) + (vector << 2);
596 ring->itr_val = 976; /* ~4000 ints/sec */
599 for (i = 0; i < adapter->num_rx_queues; i++) {
600 struct igb_ring *ring = &(adapter->rx_ring[i]);
601 if (strlen(netdev->name) < (IFNAMSIZ - 5))
602 sprintf(ring->name, "%s-rx-%d", netdev->name, i);
604 memcpy(ring->name, netdev->name, IFNAMSIZ);
605 err = request_irq(adapter->msix_entries[vector].vector,
606 &igb_msix_rx, 0, ring->name,
607 &(adapter->rx_ring[i]));
610 ring->itr_register = E1000_EITR(0) + (vector << 2);
611 ring->itr_val = adapter->itr;
615 err = request_irq(adapter->msix_entries[vector].vector,
616 &igb_msix_other, 0, netdev->name, netdev);
620 igb_configure_msix(adapter);
626 static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
628 if (adapter->msix_entries) {
629 pci_disable_msix(adapter->pdev);
630 kfree(adapter->msix_entries);
631 adapter->msix_entries = NULL;
632 } else if (adapter->flags & IGB_FLAG_HAS_MSI)
633 pci_disable_msi(adapter->pdev);
639 * igb_set_interrupt_capability - set MSI or MSI-X if supported
641 * Attempt to configure interrupts using the best available
642 * capabilities of the hardware and kernel.
644 static void igb_set_interrupt_capability(struct igb_adapter *adapter)
649 /* Number of supported queues. */
650 /* Having more queues than CPUs doesn't make sense. */
651 adapter->num_rx_queues = min_t(u32, IGB_MAX_RX_QUEUES, num_online_cpus());
652 adapter->num_tx_queues = min_t(u32, IGB_MAX_TX_QUEUES, num_online_cpus());
654 numvecs = adapter->num_tx_queues + adapter->num_rx_queues + 1;
655 adapter->msix_entries = kcalloc(numvecs, sizeof(struct msix_entry),
657 if (!adapter->msix_entries)
660 for (i = 0; i < numvecs; i++)
661 adapter->msix_entries[i].entry = i;
663 err = pci_enable_msix(adapter->pdev,
664 adapter->msix_entries,
669 igb_reset_interrupt_capability(adapter);
671 /* If we can't do MSI-X, try MSI */
673 #ifdef CONFIG_PCI_IOV
674 /* disable SR-IOV for non MSI-X configurations */
675 if (adapter->vf_data) {
676 struct e1000_hw *hw = &adapter->hw;
677 /* disable iov and allow time for transactions to clear */
678 pci_disable_sriov(adapter->pdev);
681 kfree(adapter->vf_data);
682 adapter->vf_data = NULL;
683 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
685 dev_info(&adapter->pdev->dev, "IOV Disabled\n");
688 adapter->num_rx_queues = 1;
689 adapter->num_tx_queues = 1;
690 if (!pci_enable_msi(adapter->pdev))
691 adapter->flags |= IGB_FLAG_HAS_MSI;
693 /* Notify the stack of the (possibly) reduced Tx Queue count. */
694 adapter->netdev->real_num_tx_queues = adapter->num_tx_queues;
699 * igb_request_irq - initialize interrupts
701 * Attempts to configure interrupts using the best available
702 * capabilities of the hardware and kernel.
704 static int igb_request_irq(struct igb_adapter *adapter)
706 struct net_device *netdev = adapter->netdev;
707 struct e1000_hw *hw = &adapter->hw;
710 if (adapter->msix_entries) {
711 err = igb_request_msix(adapter);
714 /* fall back to MSI */
715 igb_reset_interrupt_capability(adapter);
716 if (!pci_enable_msi(adapter->pdev))
717 adapter->flags |= IGB_FLAG_HAS_MSI;
718 igb_free_all_tx_resources(adapter);
719 igb_free_all_rx_resources(adapter);
720 adapter->num_rx_queues = 1;
721 igb_alloc_queues(adapter);
723 switch (hw->mac.type) {
725 wr32(E1000_MSIXBM(0),
726 (E1000_EICR_RX_QUEUE0 | E1000_EIMS_OTHER));
729 wr32(E1000_IVAR0, E1000_IVAR_VALID);
736 if (adapter->flags & IGB_FLAG_HAS_MSI) {
737 err = request_irq(adapter->pdev->irq, &igb_intr_msi, 0,
738 netdev->name, netdev);
741 /* fall back to legacy interrupts */
742 igb_reset_interrupt_capability(adapter);
743 adapter->flags &= ~IGB_FLAG_HAS_MSI;
746 err = request_irq(adapter->pdev->irq, &igb_intr, IRQF_SHARED,
747 netdev->name, netdev);
750 dev_err(&adapter->pdev->dev, "Error %d getting interrupt\n",
757 static void igb_free_irq(struct igb_adapter *adapter)
759 struct net_device *netdev = adapter->netdev;
761 if (adapter->msix_entries) {
764 for (i = 0; i < adapter->num_tx_queues; i++)
765 free_irq(adapter->msix_entries[vector++].vector,
766 &(adapter->tx_ring[i]));
767 for (i = 0; i < adapter->num_rx_queues; i++)
768 free_irq(adapter->msix_entries[vector++].vector,
769 &(adapter->rx_ring[i]));
771 free_irq(adapter->msix_entries[vector++].vector, netdev);
775 free_irq(adapter->pdev->irq, netdev);
779 * igb_irq_disable - Mask off interrupt generation on the NIC
780 * @adapter: board private structure
782 static void igb_irq_disable(struct igb_adapter *adapter)
784 struct e1000_hw *hw = &adapter->hw;
786 if (adapter->msix_entries) {
788 wr32(E1000_EIMC, ~0);
795 synchronize_irq(adapter->pdev->irq);
799 * igb_irq_enable - Enable default interrupt generation settings
800 * @adapter: board private structure
802 static void igb_irq_enable(struct igb_adapter *adapter)
804 struct e1000_hw *hw = &adapter->hw;
806 if (adapter->msix_entries) {
807 wr32(E1000_EIAC, adapter->eims_enable_mask);
808 wr32(E1000_EIAM, adapter->eims_enable_mask);
809 wr32(E1000_EIMS, adapter->eims_enable_mask);
810 if (adapter->vfs_allocated_count)
811 wr32(E1000_MBVFIMR, 0xFF);
812 wr32(E1000_IMS, (E1000_IMS_LSC | E1000_IMS_VMMB |
813 E1000_IMS_DOUTSYNC));
815 wr32(E1000_IMS, IMS_ENABLE_MASK);
816 wr32(E1000_IAM, IMS_ENABLE_MASK);
820 static void igb_update_mng_vlan(struct igb_adapter *adapter)
822 struct net_device *netdev = adapter->netdev;
823 u16 vid = adapter->hw.mng_cookie.vlan_id;
824 u16 old_vid = adapter->mng_vlan_id;
825 if (adapter->vlgrp) {
826 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
827 if (adapter->hw.mng_cookie.status &
828 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
829 igb_vlan_rx_add_vid(netdev, vid);
830 adapter->mng_vlan_id = vid;
832 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
834 if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
836 !vlan_group_get_device(adapter->vlgrp, old_vid))
837 igb_vlan_rx_kill_vid(netdev, old_vid);
839 adapter->mng_vlan_id = vid;
844 * igb_release_hw_control - release control of the h/w to f/w
845 * @adapter: address of board private structure
847 * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
848 * For ASF and Pass Through versions of f/w this means that the
849 * driver is no longer loaded.
852 static void igb_release_hw_control(struct igb_adapter *adapter)
854 struct e1000_hw *hw = &adapter->hw;
857 /* Let firmware take over control of h/w */
858 ctrl_ext = rd32(E1000_CTRL_EXT);
860 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
865 * igb_get_hw_control - get control of the h/w from f/w
866 * @adapter: address of board private structure
868 * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
869 * For ASF and Pass Through versions of f/w this means that
870 * the driver is loaded.
873 static void igb_get_hw_control(struct igb_adapter *adapter)
875 struct e1000_hw *hw = &adapter->hw;
878 /* Let firmware know the driver has taken over */
879 ctrl_ext = rd32(E1000_CTRL_EXT);
881 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
885 * igb_configure - configure the hardware for RX and TX
886 * @adapter: private board structure
888 static void igb_configure(struct igb_adapter *adapter)
890 struct net_device *netdev = adapter->netdev;
893 igb_get_hw_control(adapter);
894 igb_set_multi(netdev);
896 igb_restore_vlan(adapter);
898 igb_configure_tx(adapter);
899 igb_setup_rctl(adapter);
900 igb_configure_rx(adapter);
902 igb_rx_fifo_flush_82575(&adapter->hw);
904 /* call igb_desc_unused which always leaves
905 * at least 1 descriptor unused to make sure
906 * next_to_use != next_to_clean */
907 for (i = 0; i < adapter->num_rx_queues; i++) {
908 struct igb_ring *ring = &adapter->rx_ring[i];
909 igb_alloc_rx_buffers_adv(ring, igb_desc_unused(ring));
913 adapter->tx_queue_len = netdev->tx_queue_len;
918 * igb_up - Open the interface and prepare it to handle traffic
919 * @adapter: board private structure
922 int igb_up(struct igb_adapter *adapter)
924 struct e1000_hw *hw = &adapter->hw;
927 /* hardware has been reset, we need to reload some things */
928 igb_configure(adapter);
930 clear_bit(__IGB_DOWN, &adapter->state);
932 for (i = 0; i < adapter->num_rx_queues; i++)
933 napi_enable(&adapter->rx_ring[i].napi);
934 if (adapter->msix_entries)
935 igb_configure_msix(adapter);
937 igb_vmm_control(adapter);
938 igb_set_rah_pool(hw, adapter->vfs_allocated_count, 0);
939 igb_set_vmolr(hw, adapter->vfs_allocated_count);
941 /* Clear any pending interrupts. */
943 igb_irq_enable(adapter);
945 netif_tx_start_all_queues(adapter->netdev);
947 /* Fire a link change interrupt to start the watchdog. */
948 wr32(E1000_ICS, E1000_ICS_LSC);
952 void igb_down(struct igb_adapter *adapter)
954 struct e1000_hw *hw = &adapter->hw;
955 struct net_device *netdev = adapter->netdev;
959 /* signal that we're down so the interrupt handler does not
960 * reschedule our watchdog timer */
961 set_bit(__IGB_DOWN, &adapter->state);
963 /* disable receives in the hardware */
964 rctl = rd32(E1000_RCTL);
965 wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
966 /* flush and sleep below */
968 netif_tx_stop_all_queues(netdev);
970 /* disable transmits in the hardware */
971 tctl = rd32(E1000_TCTL);
972 tctl &= ~E1000_TCTL_EN;
973 wr32(E1000_TCTL, tctl);
974 /* flush both disables and wait for them to finish */
978 for (i = 0; i < adapter->num_rx_queues; i++)
979 napi_disable(&adapter->rx_ring[i].napi);
981 igb_irq_disable(adapter);
983 del_timer_sync(&adapter->watchdog_timer);
984 del_timer_sync(&adapter->phy_info_timer);
986 netdev->tx_queue_len = adapter->tx_queue_len;
987 netif_carrier_off(netdev);
989 /* record the stats before reset*/
990 igb_update_stats(adapter);
992 adapter->link_speed = 0;
993 adapter->link_duplex = 0;
995 if (!pci_channel_offline(adapter->pdev))
997 igb_clean_all_tx_rings(adapter);
998 igb_clean_all_rx_rings(adapter);
1001 void igb_reinit_locked(struct igb_adapter *adapter)
1003 WARN_ON(in_interrupt());
1004 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
1008 clear_bit(__IGB_RESETTING, &adapter->state);
1011 void igb_reset(struct igb_adapter *adapter)
1013 struct e1000_hw *hw = &adapter->hw;
1014 struct e1000_mac_info *mac = &hw->mac;
1015 struct e1000_fc_info *fc = &hw->fc;
1016 u32 pba = 0, tx_space, min_tx_space, min_rx_space;
1019 /* Repartition Pba for greater than 9k mtu
1020 * To take effect CTRL.RST is required.
1022 switch (mac->type) {
1024 pba = E1000_PBA_64K;
1028 pba = E1000_PBA_34K;
1032 if ((adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) &&
1033 (mac->type < e1000_82576)) {
1034 /* adjust PBA for jumbo frames */
1035 wr32(E1000_PBA, pba);
1037 /* To maintain wire speed transmits, the Tx FIFO should be
1038 * large enough to accommodate two full transmit packets,
1039 * rounded up to the next 1KB and expressed in KB. Likewise,
1040 * the Rx FIFO should be large enough to accommodate at least
1041 * one full receive packet and is similarly rounded up and
1042 * expressed in KB. */
1043 pba = rd32(E1000_PBA);
1044 /* upper 16 bits has Tx packet buffer allocation size in KB */
1045 tx_space = pba >> 16;
1046 /* lower 16 bits has Rx packet buffer allocation size in KB */
1048 /* the tx fifo also stores 16 bytes of information about the tx
1049 * but don't include ethernet FCS because hardware appends it */
1050 min_tx_space = (adapter->max_frame_size +
1051 sizeof(union e1000_adv_tx_desc) -
1053 min_tx_space = ALIGN(min_tx_space, 1024);
1054 min_tx_space >>= 10;
1055 /* software strips receive CRC, so leave room for it */
1056 min_rx_space = adapter->max_frame_size;
1057 min_rx_space = ALIGN(min_rx_space, 1024);
1058 min_rx_space >>= 10;
1060 /* If current Tx allocation is less than the min Tx FIFO size,
1061 * and the min Tx FIFO size is less than the current Rx FIFO
1062 * allocation, take space away from current Rx allocation */
1063 if (tx_space < min_tx_space &&
1064 ((min_tx_space - tx_space) < pba)) {
1065 pba = pba - (min_tx_space - tx_space);
1067 /* if short on rx space, rx wins and must trump tx
1069 if (pba < min_rx_space)
1072 wr32(E1000_PBA, pba);
1075 /* flow control settings */
1076 /* The high water mark must be low enough to fit one full frame
1077 * (or the size used for early receive) above it in the Rx FIFO.
1078 * Set it to the lower of:
1079 * - 90% of the Rx FIFO size, or
1080 * - the full Rx FIFO size minus one full frame */
1081 hwm = min(((pba << 10) * 9 / 10),
1082 ((pba << 10) - 2 * adapter->max_frame_size));
1084 if (mac->type < e1000_82576) {
1085 fc->high_water = hwm & 0xFFF8; /* 8-byte granularity */
1086 fc->low_water = fc->high_water - 8;
1088 fc->high_water = hwm & 0xFFF0; /* 16-byte granularity */
1089 fc->low_water = fc->high_water - 16;
1091 fc->pause_time = 0xFFFF;
1093 fc->type = fc->original_type;
1095 /* disable receive for all VFs and wait one second */
1096 if (adapter->vfs_allocated_count) {
1098 for (i = 0 ; i < adapter->vfs_allocated_count; i++)
1099 adapter->vf_data[i].clear_to_send = false;
1101 /* ping all the active vfs to let them know we are going down */
1102 igb_ping_all_vfs(adapter);
1104 /* disable transmits and receives */
1105 wr32(E1000_VFRE, 0);
1106 wr32(E1000_VFTE, 0);
1109 /* Allow time for pending master requests to run */
1110 adapter->hw.mac.ops.reset_hw(&adapter->hw);
1113 if (adapter->hw.mac.ops.init_hw(&adapter->hw))
1114 dev_err(&adapter->pdev->dev, "Hardware Error\n");
1116 igb_update_mng_vlan(adapter);
1118 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
1119 wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
1121 igb_reset_adaptive(&adapter->hw);
1122 igb_get_phy_info(&adapter->hw);
1125 static const struct net_device_ops igb_netdev_ops = {
1126 .ndo_open = igb_open,
1127 .ndo_stop = igb_close,
1128 .ndo_start_xmit = igb_xmit_frame_adv,
1129 .ndo_get_stats = igb_get_stats,
1130 .ndo_set_multicast_list = igb_set_multi,
1131 .ndo_set_mac_address = igb_set_mac,
1132 .ndo_change_mtu = igb_change_mtu,
1133 .ndo_do_ioctl = igb_ioctl,
1134 .ndo_tx_timeout = igb_tx_timeout,
1135 .ndo_validate_addr = eth_validate_addr,
1136 .ndo_vlan_rx_register = igb_vlan_rx_register,
1137 .ndo_vlan_rx_add_vid = igb_vlan_rx_add_vid,
1138 .ndo_vlan_rx_kill_vid = igb_vlan_rx_kill_vid,
1139 #ifdef CONFIG_NET_POLL_CONTROLLER
1140 .ndo_poll_controller = igb_netpoll,
1145 * igb_probe - Device Initialization Routine
1146 * @pdev: PCI device information struct
1147 * @ent: entry in igb_pci_tbl
1149 * Returns 0 on success, negative on failure
1151 * igb_probe initializes an adapter identified by a pci_dev structure.
1152 * The OS initialization, configuring of the adapter private structure,
1153 * and a hardware reset occur.
1155 static int __devinit igb_probe(struct pci_dev *pdev,
1156 const struct pci_device_id *ent)
1158 struct net_device *netdev;
1159 struct igb_adapter *adapter;
1160 struct e1000_hw *hw;
1161 const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
1162 unsigned long mmio_start, mmio_len;
1163 int err, pci_using_dac;
1164 u16 eeprom_data = 0;
1165 u16 eeprom_apme_mask = IGB_EEPROM_APME;
1168 err = pci_enable_device_mem(pdev);
1173 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
1175 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
1179 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
1181 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
1183 dev_err(&pdev->dev, "No usable DMA "
1184 "configuration, aborting\n");
1190 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
1196 err = pci_enable_pcie_error_reporting(pdev);
1198 dev_err(&pdev->dev, "pci_enable_pcie_error_reporting failed "
1200 /* non-fatal, continue */
1203 pci_set_master(pdev);
1204 pci_save_state(pdev);
1207 netdev = alloc_etherdev_mq(sizeof(struct igb_adapter),
1208 IGB_ABS_MAX_TX_QUEUES);
1210 goto err_alloc_etherdev;
1212 SET_NETDEV_DEV(netdev, &pdev->dev);
1214 pci_set_drvdata(pdev, netdev);
1215 adapter = netdev_priv(netdev);
1216 adapter->netdev = netdev;
1217 adapter->pdev = pdev;
1220 adapter->msg_enable = NETIF_MSG_DRV | NETIF_MSG_PROBE;
1222 mmio_start = pci_resource_start(pdev, 0);
1223 mmio_len = pci_resource_len(pdev, 0);
1226 hw->hw_addr = ioremap(mmio_start, mmio_len);
1230 netdev->netdev_ops = &igb_netdev_ops;
1231 igb_set_ethtool_ops(netdev);
1232 netdev->watchdog_timeo = 5 * HZ;
1234 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
1236 netdev->mem_start = mmio_start;
1237 netdev->mem_end = mmio_start + mmio_len;
1239 /* PCI config space info */
1240 hw->vendor_id = pdev->vendor;
1241 hw->device_id = pdev->device;
1242 hw->revision_id = pdev->revision;
1243 hw->subsystem_vendor_id = pdev->subsystem_vendor;
1244 hw->subsystem_device_id = pdev->subsystem_device;
1246 /* setup the private structure */
1248 /* Copy the default MAC, PHY and NVM function pointers */
1249 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
1250 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
1251 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
1252 /* Initialize skew-specific constants */
1253 err = ei->get_invariants(hw);
1257 #ifdef CONFIG_PCI_IOV
1258 /* since iov functionality isn't critical to base device function we
1259 * can accept failure. If it fails we don't allow iov to be enabled */
1260 if (hw->mac.type == e1000_82576) {
1261 /* 82576 supports a maximum of 7 VFs in addition to the PF */
1262 unsigned int num_vfs = (max_vfs > 7) ? 7 : max_vfs;
1264 unsigned char mac_addr[ETH_ALEN];
1267 adapter->vf_data = kcalloc(num_vfs,
1268 sizeof(struct vf_data_storage),
1270 if (!adapter->vf_data) {
1272 "Could not allocate VF private data - "
1273 "IOV enable failed\n");
1275 err = pci_enable_sriov(pdev, num_vfs);
1277 adapter->vfs_allocated_count = num_vfs;
1278 dev_info(&pdev->dev,
1279 "%d vfs allocated\n",
1282 i < adapter->vfs_allocated_count;
1284 random_ether_addr(mac_addr);
1285 igb_set_vf_mac(adapter, i,
1289 kfree(adapter->vf_data);
1290 adapter->vf_data = NULL;
1297 /* setup the private structure */
1298 err = igb_sw_init(adapter);
1302 igb_get_bus_info_pcie(hw);
1305 switch (hw->mac.type) {
1307 adapter->flags |= IGB_FLAG_NEED_CTX_IDX;
1314 hw->phy.autoneg_wait_to_complete = false;
1315 hw->mac.adaptive_ifs = true;
1317 /* Copper options */
1318 if (hw->phy.media_type == e1000_media_type_copper) {
1319 hw->phy.mdix = AUTO_ALL_MODES;
1320 hw->phy.disable_polarity_correction = false;
1321 hw->phy.ms_type = e1000_ms_hw_default;
1324 if (igb_check_reset_block(hw))
1325 dev_info(&pdev->dev,
1326 "PHY reset is blocked due to SOL/IDER session.\n");
1328 netdev->features = NETIF_F_SG |
1330 NETIF_F_HW_VLAN_TX |
1331 NETIF_F_HW_VLAN_RX |
1332 NETIF_F_HW_VLAN_FILTER;
1334 netdev->features |= NETIF_F_IPV6_CSUM;
1335 netdev->features |= NETIF_F_TSO;
1336 netdev->features |= NETIF_F_TSO6;
1338 netdev->features |= NETIF_F_GRO;
1340 netdev->vlan_features |= NETIF_F_TSO;
1341 netdev->vlan_features |= NETIF_F_TSO6;
1342 netdev->vlan_features |= NETIF_F_IP_CSUM;
1343 netdev->vlan_features |= NETIF_F_SG;
1346 netdev->features |= NETIF_F_HIGHDMA;
1348 adapter->en_mng_pt = igb_enable_mng_pass_thru(&adapter->hw);
1350 /* before reading the NVM, reset the controller to put the device in a
1351 * known good starting state */
1352 hw->mac.ops.reset_hw(hw);
1354 /* make sure the NVM is good */
1355 if (igb_validate_nvm_checksum(hw) < 0) {
1356 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
1361 /* copy the MAC address out of the NVM */
1362 if (hw->mac.ops.read_mac_addr(hw))
1363 dev_err(&pdev->dev, "NVM Read Error\n");
1365 memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
1366 memcpy(netdev->perm_addr, hw->mac.addr, netdev->addr_len);
1368 if (!is_valid_ether_addr(netdev->perm_addr)) {
1369 dev_err(&pdev->dev, "Invalid MAC Address\n");
1374 setup_timer(&adapter->watchdog_timer, &igb_watchdog,
1375 (unsigned long) adapter);
1376 setup_timer(&adapter->phy_info_timer, &igb_update_phy_info,
1377 (unsigned long) adapter);
1379 INIT_WORK(&adapter->reset_task, igb_reset_task);
1380 INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
1382 /* Initialize link properties that are user-changeable */
1383 adapter->fc_autoneg = true;
1384 hw->mac.autoneg = true;
1385 hw->phy.autoneg_advertised = 0x2f;
1387 hw->fc.original_type = e1000_fc_default;
1388 hw->fc.type = e1000_fc_default;
1390 adapter->itr_setting = IGB_DEFAULT_ITR;
1391 adapter->itr = IGB_START_ITR;
1393 igb_validate_mdi_setting(hw);
1395 adapter->rx_csum = 1;
1397 /* Initial Wake on LAN setting If APM wake is enabled in the EEPROM,
1398 * enable the ACPI Magic Packet filter
1401 if (hw->bus.func == 0)
1402 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
1403 else if (hw->bus.func == 1)
1404 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
1406 if (eeprom_data & eeprom_apme_mask)
1407 adapter->eeprom_wol |= E1000_WUFC_MAG;
1409 /* now that we have the eeprom settings, apply the special cases where
1410 * the eeprom may be wrong or the board simply won't support wake on
1411 * lan on a particular port */
1412 switch (pdev->device) {
1413 case E1000_DEV_ID_82575GB_QUAD_COPPER:
1414 adapter->eeprom_wol = 0;
1416 case E1000_DEV_ID_82575EB_FIBER_SERDES:
1417 case E1000_DEV_ID_82576_FIBER:
1418 case E1000_DEV_ID_82576_SERDES:
1419 /* Wake events only supported on port A for dual fiber
1420 * regardless of eeprom setting */
1421 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
1422 adapter->eeprom_wol = 0;
1424 case E1000_DEV_ID_82576_QUAD_COPPER:
1425 /* if quad port adapter, disable WoL on all but port A */
1426 if (global_quad_port_a != 0)
1427 adapter->eeprom_wol = 0;
1429 adapter->flags |= IGB_FLAG_QUAD_PORT_A;
1430 /* Reset for multiple quad port adapters */
1431 if (++global_quad_port_a == 4)
1432 global_quad_port_a = 0;
1436 /* initialize the wol settings based on the eeprom settings */
1437 adapter->wol = adapter->eeprom_wol;
1438 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
1440 /* reset the hardware with the new settings */
1443 /* let the f/w know that the h/w is now under the control of the
1445 igb_get_hw_control(adapter);
1447 strcpy(netdev->name, "eth%d");
1448 err = register_netdev(netdev);
1452 /* carrier off reporting is important to ethtool even BEFORE open */
1453 netif_carrier_off(netdev);
1455 #ifdef CONFIG_IGB_DCA
1456 if (dca_add_requester(&pdev->dev) == 0) {
1457 adapter->flags |= IGB_FLAG_DCA_ENABLED;
1458 dev_info(&pdev->dev, "DCA enabled\n");
1459 /* Always use CB2 mode, difference is masked
1460 * in the CB driver. */
1461 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
1462 igb_setup_dca(adapter);
1467 * Initialize hardware timer: we keep it running just in case
1468 * that some program needs it later on.
1470 memset(&adapter->cycles, 0, sizeof(adapter->cycles));
1471 adapter->cycles.read = igb_read_clock;
1472 adapter->cycles.mask = CLOCKSOURCE_MASK(64);
1473 adapter->cycles.mult = 1;
1474 adapter->cycles.shift = IGB_TSYNC_SHIFT;
1477 IGB_TSYNC_CYCLE_TIME_IN_NANOSECONDS * IGB_TSYNC_SCALE);
1480 * Avoid rollover while we initialize by resetting the time counter.
1482 wr32(E1000_SYSTIML, 0x00000000);
1483 wr32(E1000_SYSTIMH, 0x00000000);
1486 * Set registers so that rollover occurs soon to test this.
1488 wr32(E1000_SYSTIML, 0x00000000);
1489 wr32(E1000_SYSTIMH, 0xFF800000);
1492 timecounter_init(&adapter->clock,
1494 ktime_to_ns(ktime_get_real()));
1497 * Synchronize our NIC clock against system wall clock. NIC
1498 * time stamp reading requires ~3us per sample, each sample
1499 * was pretty stable even under load => only require 10
1500 * samples for each offset comparison.
1502 memset(&adapter->compare, 0, sizeof(adapter->compare));
1503 adapter->compare.source = &adapter->clock;
1504 adapter->compare.target = ktime_get_real;
1505 adapter->compare.num_samples = 10;
1506 timecompare_update(&adapter->compare, 0);
1512 "igb: %s: hw %p initialized timer\n",
1513 igb_get_time_str(adapter, buffer),
1518 dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
1519 /* print bus type/speed/width info */
1520 dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
1522 ((hw->bus.speed == e1000_bus_speed_2500)
1523 ? "2.5Gb/s" : "unknown"),
1524 ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
1525 (hw->bus.width == e1000_bus_width_pcie_x2) ? "Width x2" :
1526 (hw->bus.width == e1000_bus_width_pcie_x1) ? "Width x1" :
1530 igb_read_part_num(hw, &part_num);
1531 dev_info(&pdev->dev, "%s: PBA No: %06x-%03x\n", netdev->name,
1532 (part_num >> 8), (part_num & 0xff));
1534 dev_info(&pdev->dev,
1535 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
1536 adapter->msix_entries ? "MSI-X" :
1537 (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
1538 adapter->num_rx_queues, adapter->num_tx_queues);
1543 igb_release_hw_control(adapter);
1545 if (!igb_check_reset_block(hw))
1548 if (hw->flash_address)
1549 iounmap(hw->flash_address);
1551 igb_free_queues(adapter);
1553 iounmap(hw->hw_addr);
1555 free_netdev(netdev);
1557 pci_release_selected_regions(pdev, pci_select_bars(pdev,
1561 pci_disable_device(pdev);
1566 * igb_remove - Device Removal Routine
1567 * @pdev: PCI device information struct
1569 * igb_remove is called by the PCI subsystem to alert the driver
1570 * that it should release a PCI device. The could be caused by a
1571 * Hot-Plug event, or because the driver is going to be removed from
1574 static void __devexit igb_remove(struct pci_dev *pdev)
1576 struct net_device *netdev = pci_get_drvdata(pdev);
1577 struct igb_adapter *adapter = netdev_priv(netdev);
1578 struct e1000_hw *hw = &adapter->hw;
1581 /* flush_scheduled work may reschedule our watchdog task, so
1582 * explicitly disable watchdog tasks from being rescheduled */
1583 set_bit(__IGB_DOWN, &adapter->state);
1584 del_timer_sync(&adapter->watchdog_timer);
1585 del_timer_sync(&adapter->phy_info_timer);
1587 flush_scheduled_work();
1589 #ifdef CONFIG_IGB_DCA
1590 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
1591 dev_info(&pdev->dev, "DCA disabled\n");
1592 dca_remove_requester(&pdev->dev);
1593 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
1594 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
1598 /* Release control of h/w to f/w. If f/w is AMT enabled, this
1599 * would have already happened in close and is redundant. */
1600 igb_release_hw_control(adapter);
1602 unregister_netdev(netdev);
1604 if (!igb_check_reset_block(&adapter->hw))
1605 igb_reset_phy(&adapter->hw);
1607 igb_reset_interrupt_capability(adapter);
1609 igb_free_queues(adapter);
1611 #ifdef CONFIG_PCI_IOV
1612 /* reclaim resources allocated to VFs */
1613 if (adapter->vf_data) {
1614 /* disable iov and allow time for transactions to clear */
1615 pci_disable_sriov(pdev);
1618 kfree(adapter->vf_data);
1619 adapter->vf_data = NULL;
1620 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
1622 dev_info(&pdev->dev, "IOV Disabled\n");
1625 iounmap(hw->hw_addr);
1626 if (hw->flash_address)
1627 iounmap(hw->flash_address);
1628 pci_release_selected_regions(pdev, pci_select_bars(pdev,
1631 free_netdev(netdev);
1633 err = pci_disable_pcie_error_reporting(pdev);
1636 "pci_disable_pcie_error_reporting failed 0x%x\n", err);
1638 pci_disable_device(pdev);
1642 * igb_sw_init - Initialize general software structures (struct igb_adapter)
1643 * @adapter: board private structure to initialize
1645 * igb_sw_init initializes the Adapter private data structure.
1646 * Fields are initialized based on PCI device information and
1647 * OS network device settings (MTU size).
1649 static int __devinit igb_sw_init(struct igb_adapter *adapter)
1651 struct e1000_hw *hw = &adapter->hw;
1652 struct net_device *netdev = adapter->netdev;
1653 struct pci_dev *pdev = adapter->pdev;
1655 pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
1657 adapter->tx_ring_count = IGB_DEFAULT_TXD;
1658 adapter->rx_ring_count = IGB_DEFAULT_RXD;
1659 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1660 adapter->rx_ps_hdr_size = 0; /* disable packet split */
1661 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1662 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1664 /* This call may decrease the number of queues depending on
1665 * interrupt mode. */
1666 igb_set_interrupt_capability(adapter);
1668 if (igb_alloc_queues(adapter)) {
1669 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
1673 /* Explicitly disable IRQ since the NIC can be in any state. */
1674 igb_irq_disable(adapter);
1676 set_bit(__IGB_DOWN, &adapter->state);
1681 * igb_open - Called when a network interface is made active
1682 * @netdev: network interface device structure
1684 * Returns 0 on success, negative value on failure
1686 * The open entry point is called when a network interface is made
1687 * active by the system (IFF_UP). At this point all resources needed
1688 * for transmit and receive operations are allocated, the interrupt
1689 * handler is registered with the OS, the watchdog timer is started,
1690 * and the stack is notified that the interface is ready.
1692 static int igb_open(struct net_device *netdev)
1694 struct igb_adapter *adapter = netdev_priv(netdev);
1695 struct e1000_hw *hw = &adapter->hw;
1699 /* disallow open during test */
1700 if (test_bit(__IGB_TESTING, &adapter->state))
1703 netif_carrier_off(netdev);
1705 /* allocate transmit descriptors */
1706 err = igb_setup_all_tx_resources(adapter);
1710 /* allocate receive descriptors */
1711 err = igb_setup_all_rx_resources(adapter);
1715 /* e1000_power_up_phy(adapter); */
1717 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1718 if ((adapter->hw.mng_cookie.status &
1719 E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
1720 igb_update_mng_vlan(adapter);
1722 /* before we allocate an interrupt, we must be ready to handle it.
1723 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1724 * as soon as we call pci_request_irq, so we have to setup our
1725 * clean_rx handler before we do so. */
1726 igb_configure(adapter);
1728 igb_vmm_control(adapter);
1729 igb_set_rah_pool(hw, adapter->vfs_allocated_count, 0);
1730 igb_set_vmolr(hw, adapter->vfs_allocated_count);
1732 err = igb_request_irq(adapter);
1736 /* From here on the code is the same as igb_up() */
1737 clear_bit(__IGB_DOWN, &adapter->state);
1739 for (i = 0; i < adapter->num_rx_queues; i++)
1740 napi_enable(&adapter->rx_ring[i].napi);
1742 /* Clear any pending interrupts. */
1745 igb_irq_enable(adapter);
1747 netif_tx_start_all_queues(netdev);
1749 /* Fire a link status change interrupt to start the watchdog. */
1750 wr32(E1000_ICS, E1000_ICS_LSC);
1755 igb_release_hw_control(adapter);
1756 /* e1000_power_down_phy(adapter); */
1757 igb_free_all_rx_resources(adapter);
1759 igb_free_all_tx_resources(adapter);
1767 * igb_close - Disables a network interface
1768 * @netdev: network interface device structure
1770 * Returns 0, this is not allowed to fail
1772 * The close entry point is called when an interface is de-activated
1773 * by the OS. The hardware is still under the driver's control, but
1774 * needs to be disabled. A global MAC reset is issued to stop the
1775 * hardware, and all transmit and receive resources are freed.
1777 static int igb_close(struct net_device *netdev)
1779 struct igb_adapter *adapter = netdev_priv(netdev);
1781 WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
1784 igb_free_irq(adapter);
1786 igb_free_all_tx_resources(adapter);
1787 igb_free_all_rx_resources(adapter);
1789 /* kill manageability vlan ID if supported, but not if a vlan with
1790 * the same ID is registered on the host OS (let 8021q kill it) */
1791 if ((adapter->hw.mng_cookie.status &
1792 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1794 vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
1795 igb_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1801 * igb_setup_tx_resources - allocate Tx resources (Descriptors)
1802 * @adapter: board private structure
1803 * @tx_ring: tx descriptor ring (for a specific queue) to setup
1805 * Return 0 on success, negative on failure
1807 int igb_setup_tx_resources(struct igb_adapter *adapter,
1808 struct igb_ring *tx_ring)
1810 struct pci_dev *pdev = adapter->pdev;
1813 size = sizeof(struct igb_buffer) * tx_ring->count;
1814 tx_ring->buffer_info = vmalloc(size);
1815 if (!tx_ring->buffer_info)
1817 memset(tx_ring->buffer_info, 0, size);
1819 /* round up to nearest 4K */
1820 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
1821 tx_ring->size = ALIGN(tx_ring->size, 4096);
1823 tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
1829 tx_ring->adapter = adapter;
1830 tx_ring->next_to_use = 0;
1831 tx_ring->next_to_clean = 0;
1835 vfree(tx_ring->buffer_info);
1836 dev_err(&adapter->pdev->dev,
1837 "Unable to allocate memory for the transmit descriptor ring\n");
1842 * igb_setup_all_tx_resources - wrapper to allocate Tx resources
1843 * (Descriptors) for all queues
1844 * @adapter: board private structure
1846 * Return 0 on success, negative on failure
1848 static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
1853 for (i = 0; i < adapter->num_tx_queues; i++) {
1854 err = igb_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1856 dev_err(&adapter->pdev->dev,
1857 "Allocation for Tx Queue %u failed\n", i);
1858 for (i--; i >= 0; i--)
1859 igb_free_tx_resources(&adapter->tx_ring[i]);
1864 for (i = 0; i < IGB_MAX_TX_QUEUES; i++) {
1865 r_idx = i % adapter->num_tx_queues;
1866 adapter->multi_tx_table[i] = &adapter->tx_ring[r_idx];
1872 * igb_configure_tx - Configure transmit Unit after Reset
1873 * @adapter: board private structure
1875 * Configure the Tx unit of the MAC after a reset.
1877 static void igb_configure_tx(struct igb_adapter *adapter)
1880 struct e1000_hw *hw = &adapter->hw;
1885 for (i = 0; i < adapter->num_tx_queues; i++) {
1886 struct igb_ring *ring = &adapter->tx_ring[i];
1888 wr32(E1000_TDLEN(j),
1889 ring->count * sizeof(union e1000_adv_tx_desc));
1891 wr32(E1000_TDBAL(j),
1892 tdba & 0x00000000ffffffffULL);
1893 wr32(E1000_TDBAH(j), tdba >> 32);
1895 ring->head = E1000_TDH(j);
1896 ring->tail = E1000_TDT(j);
1897 writel(0, hw->hw_addr + ring->tail);
1898 writel(0, hw->hw_addr + ring->head);
1899 txdctl = rd32(E1000_TXDCTL(j));
1900 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1901 wr32(E1000_TXDCTL(j), txdctl);
1903 /* Turn off Relaxed Ordering on head write-backs. The
1904 * writebacks MUST be delivered in order or it will
1905 * completely screw up our bookeeping.
1907 txctrl = rd32(E1000_DCA_TXCTRL(j));
1908 txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1909 wr32(E1000_DCA_TXCTRL(j), txctrl);
1912 /* disable queue 0 to prevent tail bump w/o re-configuration */
1913 if (adapter->vfs_allocated_count)
1914 wr32(E1000_TXDCTL(0), 0);
1916 /* Program the Transmit Control Register */
1917 tctl = rd32(E1000_TCTL);
1918 tctl &= ~E1000_TCTL_CT;
1919 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1920 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1922 igb_config_collision_dist(hw);
1924 /* Setup Transmit Descriptor Settings for eop descriptor */
1925 adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_RS;
1927 /* Enable transmits */
1928 tctl |= E1000_TCTL_EN;
1930 wr32(E1000_TCTL, tctl);
1934 * igb_setup_rx_resources - allocate Rx resources (Descriptors)
1935 * @adapter: board private structure
1936 * @rx_ring: rx descriptor ring (for a specific queue) to setup
1938 * Returns 0 on success, negative on failure
1940 int igb_setup_rx_resources(struct igb_adapter *adapter,
1941 struct igb_ring *rx_ring)
1943 struct pci_dev *pdev = adapter->pdev;
1946 size = sizeof(struct igb_buffer) * rx_ring->count;
1947 rx_ring->buffer_info = vmalloc(size);
1948 if (!rx_ring->buffer_info)
1950 memset(rx_ring->buffer_info, 0, size);
1952 desc_len = sizeof(union e1000_adv_rx_desc);
1954 /* Round up to nearest 4K */
1955 rx_ring->size = rx_ring->count * desc_len;
1956 rx_ring->size = ALIGN(rx_ring->size, 4096);
1958 rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size,
1964 rx_ring->next_to_clean = 0;
1965 rx_ring->next_to_use = 0;
1967 rx_ring->adapter = adapter;
1972 vfree(rx_ring->buffer_info);
1973 dev_err(&adapter->pdev->dev, "Unable to allocate memory for "
1974 "the receive descriptor ring\n");
1979 * igb_setup_all_rx_resources - wrapper to allocate Rx resources
1980 * (Descriptors) for all queues
1981 * @adapter: board private structure
1983 * Return 0 on success, negative on failure
1985 static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
1989 for (i = 0; i < adapter->num_rx_queues; i++) {
1990 err = igb_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1992 dev_err(&adapter->pdev->dev,
1993 "Allocation for Rx Queue %u failed\n", i);
1994 for (i--; i >= 0; i--)
1995 igb_free_rx_resources(&adapter->rx_ring[i]);
2004 * igb_setup_rctl - configure the receive control registers
2005 * @adapter: Board private structure
2007 static void igb_setup_rctl(struct igb_adapter *adapter)
2009 struct e1000_hw *hw = &adapter->hw;
2014 rctl = rd32(E1000_RCTL);
2016 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2017 rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
2019 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
2020 (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2023 * enable stripping of CRC. It's unlikely this will break BMC
2024 * redirection as it did with e1000. Newer features require
2025 * that the HW strips the CRC.
2027 rctl |= E1000_RCTL_SECRC;
2030 * disable store bad packets and clear size bits.
2032 rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
2034 /* enable LPE when to prevent packets larger than max_frame_size */
2035 rctl |= E1000_RCTL_LPE;
2037 /* Setup buffer sizes */
2038 switch (adapter->rx_buffer_len) {
2039 case IGB_RXBUFFER_256:
2040 rctl |= E1000_RCTL_SZ_256;
2042 case IGB_RXBUFFER_512:
2043 rctl |= E1000_RCTL_SZ_512;
2046 srrctl = ALIGN(adapter->rx_buffer_len, 1024)
2047 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
2051 /* 82575 and greater support packet-split where the protocol
2052 * header is placed in skb->data and the packet data is
2053 * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2054 * In the case of a non-split, skb->data is linearly filled,
2055 * followed by the page buffers. Therefore, skb->data is
2056 * sized to hold the largest protocol header.
2058 /* allocations using alloc_page take too long for regular MTU
2059 * so only enable packet split for jumbo frames */
2060 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2061 adapter->rx_ps_hdr_size = IGB_RXBUFFER_128;
2062 srrctl |= adapter->rx_ps_hdr_size <<
2063 E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
2064 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2066 adapter->rx_ps_hdr_size = 0;
2067 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
2070 /* Attention!!! For SR-IOV PF driver operations you must enable
2071 * queue drop for all VF and PF queues to prevent head of line blocking
2072 * if an un-trusted VF does not provide descriptors to hardware.
2074 if (adapter->vfs_allocated_count) {
2077 j = adapter->rx_ring[0].reg_idx;
2079 /* set all queue drop enable bits */
2080 wr32(E1000_QDE, ALL_QUEUES);
2081 srrctl |= E1000_SRRCTL_DROP_EN;
2083 /* disable queue 0 to prevent tail write w/o re-config */
2084 wr32(E1000_RXDCTL(0), 0);
2086 vmolr = rd32(E1000_VMOLR(j));
2087 if (rctl & E1000_RCTL_LPE)
2088 vmolr |= E1000_VMOLR_LPE;
2089 if (adapter->num_rx_queues > 0)
2090 vmolr |= E1000_VMOLR_RSSE;
2091 wr32(E1000_VMOLR(j), vmolr);
2094 for (i = 0; i < adapter->num_rx_queues; i++) {
2095 j = adapter->rx_ring[i].reg_idx;
2096 wr32(E1000_SRRCTL(j), srrctl);
2099 wr32(E1000_RCTL, rctl);
2103 * igb_rlpml_set - set maximum receive packet size
2104 * @adapter: board private structure
2106 * Configure maximum receivable packet size.
2108 static void igb_rlpml_set(struct igb_adapter *adapter)
2110 u32 max_frame_size = adapter->max_frame_size;
2111 struct e1000_hw *hw = &adapter->hw;
2112 u16 pf_id = adapter->vfs_allocated_count;
2115 max_frame_size += VLAN_TAG_SIZE;
2117 /* if vfs are enabled we set RLPML to the largest possible request
2118 * size and set the VMOLR RLPML to the size we need */
2120 igb_set_vf_rlpml(adapter, max_frame_size, pf_id);
2121 max_frame_size = MAX_STD_JUMBO_FRAME_SIZE + VLAN_TAG_SIZE;
2124 wr32(E1000_RLPML, max_frame_size);
2128 * igb_configure_vt_default_pool - Configure VT default pool
2129 * @adapter: board private structure
2131 * Configure the default pool
2133 static void igb_configure_vt_default_pool(struct igb_adapter *adapter)
2135 struct e1000_hw *hw = &adapter->hw;
2136 u16 pf_id = adapter->vfs_allocated_count;
2139 /* not in sr-iov mode - do nothing */
2143 vtctl = rd32(E1000_VT_CTL);
2144 vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK |
2145 E1000_VT_CTL_DISABLE_DEF_POOL);
2146 vtctl |= pf_id << E1000_VT_CTL_DEFAULT_POOL_SHIFT;
2147 wr32(E1000_VT_CTL, vtctl);
2151 * igb_configure_rx - Configure receive Unit after Reset
2152 * @adapter: board private structure
2154 * Configure the Rx unit of the MAC after a reset.
2156 static void igb_configure_rx(struct igb_adapter *adapter)
2159 struct e1000_hw *hw = &adapter->hw;
2164 /* disable receives while setting up the descriptors */
2165 rctl = rd32(E1000_RCTL);
2166 wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
2170 if (adapter->itr_setting > 3)
2171 wr32(E1000_ITR, adapter->itr);
2173 /* Setup the HW Rx Head and Tail Descriptor Pointers and
2174 * the Base and Length of the Rx Descriptor Ring */
2175 for (i = 0; i < adapter->num_rx_queues; i++) {
2176 struct igb_ring *ring = &adapter->rx_ring[i];
2177 int j = ring->reg_idx;
2179 wr32(E1000_RDBAL(j),
2180 rdba & 0x00000000ffffffffULL);
2181 wr32(E1000_RDBAH(j), rdba >> 32);
2182 wr32(E1000_RDLEN(j),
2183 ring->count * sizeof(union e1000_adv_rx_desc));
2185 ring->head = E1000_RDH(j);
2186 ring->tail = E1000_RDT(j);
2187 writel(0, hw->hw_addr + ring->tail);
2188 writel(0, hw->hw_addr + ring->head);
2190 rxdctl = rd32(E1000_RXDCTL(j));
2191 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
2192 rxdctl &= 0xFFF00000;
2193 rxdctl |= IGB_RX_PTHRESH;
2194 rxdctl |= IGB_RX_HTHRESH << 8;
2195 rxdctl |= IGB_RX_WTHRESH << 16;
2196 wr32(E1000_RXDCTL(j), rxdctl);
2199 if (adapter->num_rx_queues > 1) {
2208 get_random_bytes(&random[0], 40);
2210 if (hw->mac.type >= e1000_82576)
2214 for (j = 0; j < (32 * 4); j++) {
2216 adapter->rx_ring[(j % adapter->num_rx_queues)].reg_idx << shift;
2219 hw->hw_addr + E1000_RETA(0) + (j & ~3));
2221 if (adapter->vfs_allocated_count)
2222 mrqc = E1000_MRQC_ENABLE_VMDQ_RSS_2Q;
2224 mrqc = E1000_MRQC_ENABLE_RSS_4Q;
2226 /* Fill out hash function seeds */
2227 for (j = 0; j < 10; j++)
2228 array_wr32(E1000_RSSRK(0), j, random[j]);
2230 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
2231 E1000_MRQC_RSS_FIELD_IPV4_TCP);
2232 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6 |
2233 E1000_MRQC_RSS_FIELD_IPV6_TCP);
2234 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4_UDP |
2235 E1000_MRQC_RSS_FIELD_IPV6_UDP);
2236 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
2237 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
2240 wr32(E1000_MRQC, mrqc);
2242 /* Multiqueue and raw packet checksumming are mutually
2243 * exclusive. Note that this not the same as TCP/IP
2244 * checksumming, which works fine. */
2245 rxcsum = rd32(E1000_RXCSUM);
2246 rxcsum |= E1000_RXCSUM_PCSD;
2247 wr32(E1000_RXCSUM, rxcsum);
2249 /* Enable multi-queue for sr-iov */
2250 if (adapter->vfs_allocated_count)
2251 wr32(E1000_MRQC, E1000_MRQC_ENABLE_VMDQ);
2252 /* Enable Receive Checksum Offload for TCP and UDP */
2253 rxcsum = rd32(E1000_RXCSUM);
2254 if (adapter->rx_csum)
2255 rxcsum |= E1000_RXCSUM_TUOFL | E1000_RXCSUM_IPPCSE;
2257 rxcsum &= ~(E1000_RXCSUM_TUOFL | E1000_RXCSUM_IPPCSE);
2259 wr32(E1000_RXCSUM, rxcsum);
2262 /* Set the default pool for the PF's first queue */
2263 igb_configure_vt_default_pool(adapter);
2265 igb_rlpml_set(adapter);
2267 /* Enable Receives */
2268 wr32(E1000_RCTL, rctl);
2272 * igb_free_tx_resources - Free Tx Resources per Queue
2273 * @tx_ring: Tx descriptor ring for a specific queue
2275 * Free all transmit software resources
2277 void igb_free_tx_resources(struct igb_ring *tx_ring)
2279 struct pci_dev *pdev = tx_ring->adapter->pdev;
2281 igb_clean_tx_ring(tx_ring);
2283 vfree(tx_ring->buffer_info);
2284 tx_ring->buffer_info = NULL;
2286 pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
2288 tx_ring->desc = NULL;
2292 * igb_free_all_tx_resources - Free Tx Resources for All Queues
2293 * @adapter: board private structure
2295 * Free all transmit software resources
2297 static void igb_free_all_tx_resources(struct igb_adapter *adapter)
2301 for (i = 0; i < adapter->num_tx_queues; i++)
2302 igb_free_tx_resources(&adapter->tx_ring[i]);
2305 static void igb_unmap_and_free_tx_resource(struct igb_adapter *adapter,
2306 struct igb_buffer *buffer_info)
2308 buffer_info->dma = 0;
2309 if (buffer_info->skb) {
2310 skb_dma_unmap(&adapter->pdev->dev, buffer_info->skb,
2312 dev_kfree_skb_any(buffer_info->skb);
2313 buffer_info->skb = NULL;
2315 buffer_info->time_stamp = 0;
2316 /* buffer_info must be completely set up in the transmit path */
2320 * igb_clean_tx_ring - Free Tx Buffers
2321 * @tx_ring: ring to be cleaned
2323 static void igb_clean_tx_ring(struct igb_ring *tx_ring)
2325 struct igb_adapter *adapter = tx_ring->adapter;
2326 struct igb_buffer *buffer_info;
2330 if (!tx_ring->buffer_info)
2332 /* Free all the Tx ring sk_buffs */
2334 for (i = 0; i < tx_ring->count; i++) {
2335 buffer_info = &tx_ring->buffer_info[i];
2336 igb_unmap_and_free_tx_resource(adapter, buffer_info);
2339 size = sizeof(struct igb_buffer) * tx_ring->count;
2340 memset(tx_ring->buffer_info, 0, size);
2342 /* Zero out the descriptor ring */
2344 memset(tx_ring->desc, 0, tx_ring->size);
2346 tx_ring->next_to_use = 0;
2347 tx_ring->next_to_clean = 0;
2349 writel(0, adapter->hw.hw_addr + tx_ring->head);
2350 writel(0, adapter->hw.hw_addr + tx_ring->tail);
2354 * igb_clean_all_tx_rings - Free Tx Buffers for all queues
2355 * @adapter: board private structure
2357 static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
2361 for (i = 0; i < adapter->num_tx_queues; i++)
2362 igb_clean_tx_ring(&adapter->tx_ring[i]);
2366 * igb_free_rx_resources - Free Rx Resources
2367 * @rx_ring: ring to clean the resources from
2369 * Free all receive software resources
2371 void igb_free_rx_resources(struct igb_ring *rx_ring)
2373 struct pci_dev *pdev = rx_ring->adapter->pdev;
2375 igb_clean_rx_ring(rx_ring);
2377 vfree(rx_ring->buffer_info);
2378 rx_ring->buffer_info = NULL;
2380 pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
2382 rx_ring->desc = NULL;
2386 * igb_free_all_rx_resources - Free Rx Resources for All Queues
2387 * @adapter: board private structure
2389 * Free all receive software resources
2391 static void igb_free_all_rx_resources(struct igb_adapter *adapter)
2395 for (i = 0; i < adapter->num_rx_queues; i++)
2396 igb_free_rx_resources(&adapter->rx_ring[i]);
2400 * igb_clean_rx_ring - Free Rx Buffers per Queue
2401 * @rx_ring: ring to free buffers from
2403 static void igb_clean_rx_ring(struct igb_ring *rx_ring)
2405 struct igb_adapter *adapter = rx_ring->adapter;
2406 struct igb_buffer *buffer_info;
2407 struct pci_dev *pdev = adapter->pdev;
2411 if (!rx_ring->buffer_info)
2413 /* Free all the Rx ring sk_buffs */
2414 for (i = 0; i < rx_ring->count; i++) {
2415 buffer_info = &rx_ring->buffer_info[i];
2416 if (buffer_info->dma) {
2417 if (adapter->rx_ps_hdr_size)
2418 pci_unmap_single(pdev, buffer_info->dma,
2419 adapter->rx_ps_hdr_size,
2420 PCI_DMA_FROMDEVICE);
2422 pci_unmap_single(pdev, buffer_info->dma,
2423 adapter->rx_buffer_len,
2424 PCI_DMA_FROMDEVICE);
2425 buffer_info->dma = 0;
2428 if (buffer_info->skb) {
2429 dev_kfree_skb(buffer_info->skb);
2430 buffer_info->skb = NULL;
2432 if (buffer_info->page) {
2433 if (buffer_info->page_dma)
2434 pci_unmap_page(pdev, buffer_info->page_dma,
2436 PCI_DMA_FROMDEVICE);
2437 put_page(buffer_info->page);
2438 buffer_info->page = NULL;
2439 buffer_info->page_dma = 0;
2440 buffer_info->page_offset = 0;
2444 size = sizeof(struct igb_buffer) * rx_ring->count;
2445 memset(rx_ring->buffer_info, 0, size);
2447 /* Zero out the descriptor ring */
2448 memset(rx_ring->desc, 0, rx_ring->size);
2450 rx_ring->next_to_clean = 0;
2451 rx_ring->next_to_use = 0;
2453 writel(0, adapter->hw.hw_addr + rx_ring->head);
2454 writel(0, adapter->hw.hw_addr + rx_ring->tail);
2458 * igb_clean_all_rx_rings - Free Rx Buffers for all queues
2459 * @adapter: board private structure
2461 static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
2465 for (i = 0; i < adapter->num_rx_queues; i++)
2466 igb_clean_rx_ring(&adapter->rx_ring[i]);
2470 * igb_set_mac - Change the Ethernet Address of the NIC
2471 * @netdev: network interface device structure
2472 * @p: pointer to an address structure
2474 * Returns 0 on success, negative on failure
2476 static int igb_set_mac(struct net_device *netdev, void *p)
2478 struct igb_adapter *adapter = netdev_priv(netdev);
2479 struct e1000_hw *hw = &adapter->hw;
2480 struct sockaddr *addr = p;
2482 if (!is_valid_ether_addr(addr->sa_data))
2483 return -EADDRNOTAVAIL;
2485 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2486 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
2488 hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
2490 igb_set_rah_pool(hw, adapter->vfs_allocated_count, 0);
2496 * igb_set_multi - Multicast and Promiscuous mode set
2497 * @netdev: network interface device structure
2499 * The set_multi entry point is called whenever the multicast address
2500 * list or the network interface flags are updated. This routine is
2501 * responsible for configuring the hardware for proper multicast,
2502 * promiscuous mode, and all-multi behavior.
2504 static void igb_set_multi(struct net_device *netdev)
2506 struct igb_adapter *adapter = netdev_priv(netdev);
2507 struct e1000_hw *hw = &adapter->hw;
2508 struct e1000_mac_info *mac = &hw->mac;
2509 struct dev_mc_list *mc_ptr;
2510 u8 *mta_list = NULL;
2514 /* Check for Promiscuous and All Multicast modes */
2516 rctl = rd32(E1000_RCTL);
2518 if (netdev->flags & IFF_PROMISC) {
2519 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2520 rctl &= ~E1000_RCTL_VFE;
2522 if (netdev->flags & IFF_ALLMULTI) {
2523 rctl |= E1000_RCTL_MPE;
2524 rctl &= ~E1000_RCTL_UPE;
2526 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2527 rctl |= E1000_RCTL_VFE;
2529 wr32(E1000_RCTL, rctl);
2531 if (netdev->mc_count) {
2532 mta_list = kzalloc(netdev->mc_count * 6, GFP_ATOMIC);
2534 dev_err(&adapter->pdev->dev,
2535 "failed to allocate multicast filter list\n");
2540 /* The shared function expects a packed array of only addresses. */
2541 mc_ptr = netdev->mc_list;
2543 for (i = 0; i < netdev->mc_count; i++) {
2546 memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr, ETH_ALEN);
2547 mc_ptr = mc_ptr->next;
2549 igb_update_mc_addr_list(hw, mta_list, i,
2550 adapter->vfs_allocated_count + 1,
2551 mac->rar_entry_count);
2553 igb_set_mc_list_pools(adapter, i, mac->rar_entry_count);
2554 igb_restore_vf_multicasts(adapter);
2559 /* Need to wait a few seconds after link up to get diagnostic information from
2561 static void igb_update_phy_info(unsigned long data)
2563 struct igb_adapter *adapter = (struct igb_adapter *) data;
2564 igb_get_phy_info(&adapter->hw);
2568 * igb_has_link - check shared code for link and determine up/down
2569 * @adapter: pointer to driver private info
2571 static bool igb_has_link(struct igb_adapter *adapter)
2573 struct e1000_hw *hw = &adapter->hw;
2574 bool link_active = false;
2577 /* get_link_status is set on LSC (link status) interrupt or
2578 * rx sequence error interrupt. get_link_status will stay
2579 * false until the e1000_check_for_link establishes link
2580 * for copper adapters ONLY
2582 switch (hw->phy.media_type) {
2583 case e1000_media_type_copper:
2584 if (hw->mac.get_link_status) {
2585 ret_val = hw->mac.ops.check_for_link(hw);
2586 link_active = !hw->mac.get_link_status;
2591 case e1000_media_type_fiber:
2592 ret_val = hw->mac.ops.check_for_link(hw);
2593 link_active = !!(rd32(E1000_STATUS) & E1000_STATUS_LU);
2595 case e1000_media_type_internal_serdes:
2596 ret_val = hw->mac.ops.check_for_link(hw);
2597 link_active = hw->mac.serdes_has_link;
2600 case e1000_media_type_unknown:
2608 * igb_watchdog - Timer Call-back
2609 * @data: pointer to adapter cast into an unsigned long
2611 static void igb_watchdog(unsigned long data)
2613 struct igb_adapter *adapter = (struct igb_adapter *)data;
2614 /* Do the rest outside of interrupt context */
2615 schedule_work(&adapter->watchdog_task);
2618 static void igb_watchdog_task(struct work_struct *work)
2620 struct igb_adapter *adapter = container_of(work,
2621 struct igb_adapter, watchdog_task);
2622 struct e1000_hw *hw = &adapter->hw;
2623 struct net_device *netdev = adapter->netdev;
2624 struct igb_ring *tx_ring = adapter->tx_ring;
2629 link = igb_has_link(adapter);
2630 if ((netif_carrier_ok(netdev)) && link)
2634 if (!netif_carrier_ok(netdev)) {
2636 hw->mac.ops.get_speed_and_duplex(&adapter->hw,
2637 &adapter->link_speed,
2638 &adapter->link_duplex);
2640 ctrl = rd32(E1000_CTRL);
2641 /* Links status message must follow this format */
2642 printk(KERN_INFO "igb: %s NIC Link is Up %d Mbps %s, "
2643 "Flow Control: %s\n",
2645 adapter->link_speed,
2646 adapter->link_duplex == FULL_DUPLEX ?
2647 "Full Duplex" : "Half Duplex",
2648 ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2649 E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2650 E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2651 E1000_CTRL_TFCE) ? "TX" : "None")));
2653 /* tweak tx_queue_len according to speed/duplex and
2654 * adjust the timeout factor */
2655 netdev->tx_queue_len = adapter->tx_queue_len;
2656 adapter->tx_timeout_factor = 1;
2657 switch (adapter->link_speed) {
2659 netdev->tx_queue_len = 10;
2660 adapter->tx_timeout_factor = 14;
2663 netdev->tx_queue_len = 100;
2664 /* maybe add some timeout factor ? */
2668 netif_carrier_on(netdev);
2670 igb_ping_all_vfs(adapter);
2672 /* link state has changed, schedule phy info update */
2673 if (!test_bit(__IGB_DOWN, &adapter->state))
2674 mod_timer(&adapter->phy_info_timer,
2675 round_jiffies(jiffies + 2 * HZ));
2678 if (netif_carrier_ok(netdev)) {
2679 adapter->link_speed = 0;
2680 adapter->link_duplex = 0;
2681 /* Links status message must follow this format */
2682 printk(KERN_INFO "igb: %s NIC Link is Down\n",
2684 netif_carrier_off(netdev);
2686 igb_ping_all_vfs(adapter);
2688 /* link state has changed, schedule phy info update */
2689 if (!test_bit(__IGB_DOWN, &adapter->state))
2690 mod_timer(&adapter->phy_info_timer,
2691 round_jiffies(jiffies + 2 * HZ));
2696 igb_update_stats(adapter);
2698 hw->mac.tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2699 adapter->tpt_old = adapter->stats.tpt;
2700 hw->mac.collision_delta = adapter->stats.colc - adapter->colc_old;
2701 adapter->colc_old = adapter->stats.colc;
2703 adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
2704 adapter->gorc_old = adapter->stats.gorc;
2705 adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
2706 adapter->gotc_old = adapter->stats.gotc;
2708 igb_update_adaptive(&adapter->hw);
2710 if (!netif_carrier_ok(netdev)) {
2711 if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) {
2712 /* We've lost link, so the controller stops DMA,
2713 * but we've got queued Tx work that's never going
2714 * to get done, so reset controller to flush Tx.
2715 * (Do the reset outside of interrupt context). */
2716 adapter->tx_timeout_count++;
2717 schedule_work(&adapter->reset_task);
2721 /* Cause software interrupt to ensure rx ring is cleaned */
2722 if (adapter->msix_entries) {
2723 for (i = 0; i < adapter->num_rx_queues; i++)
2724 eics |= adapter->rx_ring[i].eims_value;
2725 wr32(E1000_EICS, eics);
2727 wr32(E1000_ICS, E1000_ICS_RXDMT0);
2730 /* Force detection of hung controller every watchdog period */
2731 tx_ring->detect_tx_hung = true;
2733 /* Reset the timer */
2734 if (!test_bit(__IGB_DOWN, &adapter->state))
2735 mod_timer(&adapter->watchdog_timer,
2736 round_jiffies(jiffies + 2 * HZ));
2739 enum latency_range {
2743 latency_invalid = 255
2748 * igb_update_ring_itr - update the dynamic ITR value based on packet size
2750 * Stores a new ITR value based on strictly on packet size. This
2751 * algorithm is less sophisticated than that used in igb_update_itr,
2752 * due to the difficulty of synchronizing statistics across multiple
2753 * receive rings. The divisors and thresholds used by this fuction
2754 * were determined based on theoretical maximum wire speed and testing
2755 * data, in order to minimize response time while increasing bulk
2757 * This functionality is controlled by the InterruptThrottleRate module
2758 * parameter (see igb_param.c)
2759 * NOTE: This function is called only when operating in a multiqueue
2760 * receive environment.
2761 * @rx_ring: pointer to ring
2763 static void igb_update_ring_itr(struct igb_ring *rx_ring)
2765 int new_val = rx_ring->itr_val;
2766 int avg_wire_size = 0;
2767 struct igb_adapter *adapter = rx_ring->adapter;
2769 if (!rx_ring->total_packets)
2770 goto clear_counts; /* no packets, so don't do anything */
2772 /* For non-gigabit speeds, just fix the interrupt rate at 4000
2773 * ints/sec - ITR timer value of 120 ticks.
2775 if (adapter->link_speed != SPEED_1000) {
2779 avg_wire_size = rx_ring->total_bytes / rx_ring->total_packets;
2781 /* Add 24 bytes to size to account for CRC, preamble, and gap */
2782 avg_wire_size += 24;
2784 /* Don't starve jumbo frames */
2785 avg_wire_size = min(avg_wire_size, 3000);
2787 /* Give a little boost to mid-size frames */
2788 if ((avg_wire_size > 300) && (avg_wire_size < 1200))
2789 new_val = avg_wire_size / 3;
2791 new_val = avg_wire_size / 2;
2794 if (new_val != rx_ring->itr_val) {
2795 rx_ring->itr_val = new_val;
2796 rx_ring->set_itr = 1;
2799 rx_ring->total_bytes = 0;
2800 rx_ring->total_packets = 0;
2804 * igb_update_itr - update the dynamic ITR value based on statistics
2805 * Stores a new ITR value based on packets and byte
2806 * counts during the last interrupt. The advantage of per interrupt
2807 * computation is faster updates and more accurate ITR for the current
2808 * traffic pattern. Constants in this function were computed
2809 * based on theoretical maximum wire speed and thresholds were set based
2810 * on testing data as well as attempting to minimize response time
2811 * while increasing bulk throughput.
2812 * this functionality is controlled by the InterruptThrottleRate module
2813 * parameter (see igb_param.c)
2814 * NOTE: These calculations are only valid when operating in a single-
2815 * queue environment.
2816 * @adapter: pointer to adapter
2817 * @itr_setting: current adapter->itr
2818 * @packets: the number of packets during this measurement interval
2819 * @bytes: the number of bytes during this measurement interval
2821 static unsigned int igb_update_itr(struct igb_adapter *adapter, u16 itr_setting,
2822 int packets, int bytes)
2824 unsigned int retval = itr_setting;
2827 goto update_itr_done;
2829 switch (itr_setting) {
2830 case lowest_latency:
2831 /* handle TSO and jumbo frames */
2832 if (bytes/packets > 8000)
2833 retval = bulk_latency;
2834 else if ((packets < 5) && (bytes > 512))
2835 retval = low_latency;
2837 case low_latency: /* 50 usec aka 20000 ints/s */
2838 if (bytes > 10000) {
2839 /* this if handles the TSO accounting */
2840 if (bytes/packets > 8000) {
2841 retval = bulk_latency;
2842 } else if ((packets < 10) || ((bytes/packets) > 1200)) {
2843 retval = bulk_latency;
2844 } else if ((packets > 35)) {
2845 retval = lowest_latency;
2847 } else if (bytes/packets > 2000) {
2848 retval = bulk_latency;
2849 } else if (packets <= 2 && bytes < 512) {
2850 retval = lowest_latency;
2853 case bulk_latency: /* 250 usec aka 4000 ints/s */
2854 if (bytes > 25000) {
2856 retval = low_latency;
2857 } else if (bytes < 1500) {
2858 retval = low_latency;
2867 static void igb_set_itr(struct igb_adapter *adapter)
2870 u32 new_itr = adapter->itr;
2872 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2873 if (adapter->link_speed != SPEED_1000) {
2879 adapter->rx_itr = igb_update_itr(adapter,
2881 adapter->rx_ring->total_packets,
2882 adapter->rx_ring->total_bytes);
2884 if (adapter->rx_ring->buddy) {
2885 adapter->tx_itr = igb_update_itr(adapter,
2887 adapter->tx_ring->total_packets,
2888 adapter->tx_ring->total_bytes);
2889 current_itr = max(adapter->rx_itr, adapter->tx_itr);
2891 current_itr = adapter->rx_itr;
2894 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2895 if (adapter->itr_setting == 3 && current_itr == lowest_latency)
2896 current_itr = low_latency;
2898 switch (current_itr) {
2899 /* counts and packets in update_itr are dependent on these numbers */
2900 case lowest_latency:
2904 new_itr = 20000; /* aka hwitr = ~200 */
2914 adapter->rx_ring->total_bytes = 0;
2915 adapter->rx_ring->total_packets = 0;
2916 if (adapter->rx_ring->buddy) {
2917 adapter->rx_ring->buddy->total_bytes = 0;
2918 adapter->rx_ring->buddy->total_packets = 0;
2921 if (new_itr != adapter->itr) {
2922 /* this attempts to bias the interrupt rate towards Bulk
2923 * by adding intermediate steps when interrupt rate is
2925 new_itr = new_itr > adapter->itr ?
2926 min(adapter->itr + (new_itr >> 2), new_itr) :
2928 /* Don't write the value here; it resets the adapter's
2929 * internal timer, and causes us to delay far longer than
2930 * we should between interrupts. Instead, we write the ITR
2931 * value at the beginning of the next interrupt so the timing
2932 * ends up being correct.
2934 adapter->itr = new_itr;
2935 adapter->rx_ring->itr_val = 1000000000 / (new_itr * 256);
2936 adapter->rx_ring->set_itr = 1;
2943 #define IGB_TX_FLAGS_CSUM 0x00000001
2944 #define IGB_TX_FLAGS_VLAN 0x00000002
2945 #define IGB_TX_FLAGS_TSO 0x00000004
2946 #define IGB_TX_FLAGS_IPV4 0x00000008
2947 #define IGB_TX_FLAGS_TSTAMP 0x00000010
2948 #define IGB_TX_FLAGS_VLAN_MASK 0xffff0000
2949 #define IGB_TX_FLAGS_VLAN_SHIFT 16
2951 static inline int igb_tso_adv(struct igb_adapter *adapter,
2952 struct igb_ring *tx_ring,
2953 struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
2955 struct e1000_adv_tx_context_desc *context_desc;
2958 struct igb_buffer *buffer_info;
2959 u32 info = 0, tu_cmd = 0;
2960 u32 mss_l4len_idx, l4len;
2963 if (skb_header_cloned(skb)) {
2964 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2969 l4len = tcp_hdrlen(skb);
2972 if (skb->protocol == htons(ETH_P_IP)) {
2973 struct iphdr *iph = ip_hdr(skb);
2976 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2980 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
2981 ipv6_hdr(skb)->payload_len = 0;
2982 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2983 &ipv6_hdr(skb)->daddr,
2987 i = tx_ring->next_to_use;
2989 buffer_info = &tx_ring->buffer_info[i];
2990 context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
2991 /* VLAN MACLEN IPLEN */
2992 if (tx_flags & IGB_TX_FLAGS_VLAN)
2993 info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
2994 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2995 *hdr_len += skb_network_offset(skb);
2996 info |= skb_network_header_len(skb);
2997 *hdr_len += skb_network_header_len(skb);
2998 context_desc->vlan_macip_lens = cpu_to_le32(info);
3000 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3001 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
3003 if (skb->protocol == htons(ETH_P_IP))
3004 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
3005 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
3007 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
3010 mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
3011 mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
3013 /* For 82575, context index must be unique per ring. */
3014 if (adapter->flags & IGB_FLAG_NEED_CTX_IDX)
3015 mss_l4len_idx |= tx_ring->queue_index << 4;
3017 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
3018 context_desc->seqnum_seed = 0;
3020 buffer_info->time_stamp = jiffies;
3021 buffer_info->next_to_watch = i;
3022 buffer_info->dma = 0;
3024 if (i == tx_ring->count)
3027 tx_ring->next_to_use = i;
3032 static inline bool igb_tx_csum_adv(struct igb_adapter *adapter,
3033 struct igb_ring *tx_ring,
3034 struct sk_buff *skb, u32 tx_flags)
3036 struct e1000_adv_tx_context_desc *context_desc;
3038 struct igb_buffer *buffer_info;
3039 u32 info = 0, tu_cmd = 0;
3041 if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
3042 (tx_flags & IGB_TX_FLAGS_VLAN)) {
3043 i = tx_ring->next_to_use;
3044 buffer_info = &tx_ring->buffer_info[i];
3045 context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
3047 if (tx_flags & IGB_TX_FLAGS_VLAN)
3048 info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
3049 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
3050 if (skb->ip_summed == CHECKSUM_PARTIAL)
3051 info |= skb_network_header_len(skb);
3053 context_desc->vlan_macip_lens = cpu_to_le32(info);
3055 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
3057 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3060 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3061 const struct vlan_ethhdr *vhdr =
3062 (const struct vlan_ethhdr*)skb->data;
3064 protocol = vhdr->h_vlan_encapsulated_proto;
3066 protocol = skb->protocol;
3070 case cpu_to_be16(ETH_P_IP):
3071 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
3072 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
3073 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
3075 case cpu_to_be16(ETH_P_IPV6):
3076 /* XXX what about other V6 headers?? */
3077 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3078 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
3081 if (unlikely(net_ratelimit()))
3082 dev_warn(&adapter->pdev->dev,
3083 "partial checksum but proto=%x!\n",
3089 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
3090 context_desc->seqnum_seed = 0;
3091 if (adapter->flags & IGB_FLAG_NEED_CTX_IDX)
3092 context_desc->mss_l4len_idx =
3093 cpu_to_le32(tx_ring->queue_index << 4);
3095 context_desc->mss_l4len_idx = 0;
3097 buffer_info->time_stamp = jiffies;
3098 buffer_info->next_to_watch = i;
3099 buffer_info->dma = 0;
3102 if (i == tx_ring->count)
3104 tx_ring->next_to_use = i;
3111 #define IGB_MAX_TXD_PWR 16
3112 #define IGB_MAX_DATA_PER_TXD (1<<IGB_MAX_TXD_PWR)
3114 static inline int igb_tx_map_adv(struct igb_adapter *adapter,
3115 struct igb_ring *tx_ring, struct sk_buff *skb,
3118 struct igb_buffer *buffer_info;
3119 unsigned int len = skb_headlen(skb);
3120 unsigned int count = 0, i;
3124 i = tx_ring->next_to_use;
3126 if (skb_dma_map(&adapter->pdev->dev, skb, DMA_TO_DEVICE)) {
3127 dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
3131 map = skb_shinfo(skb)->dma_maps;
3133 buffer_info = &tx_ring->buffer_info[i];
3134 BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
3135 buffer_info->length = len;
3136 /* set time_stamp *before* dma to help avoid a possible race */
3137 buffer_info->time_stamp = jiffies;
3138 buffer_info->next_to_watch = i;
3139 buffer_info->dma = map[count];
3142 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
3143 struct skb_frag_struct *frag;
3146 if (i == tx_ring->count)
3149 frag = &skb_shinfo(skb)->frags[f];
3152 buffer_info = &tx_ring->buffer_info[i];
3153 BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
3154 buffer_info->length = len;
3155 buffer_info->time_stamp = jiffies;
3156 buffer_info->next_to_watch = i;
3157 buffer_info->dma = map[count];
3161 tx_ring->buffer_info[i].skb = skb;
3162 tx_ring->buffer_info[first].next_to_watch = i;
3167 static inline void igb_tx_queue_adv(struct igb_adapter *adapter,
3168 struct igb_ring *tx_ring,
3169 int tx_flags, int count, u32 paylen,
3172 union e1000_adv_tx_desc *tx_desc = NULL;
3173 struct igb_buffer *buffer_info;
3174 u32 olinfo_status = 0, cmd_type_len;
3177 cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
3178 E1000_ADVTXD_DCMD_DEXT);
3180 if (tx_flags & IGB_TX_FLAGS_VLAN)
3181 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
3183 if (tx_flags & IGB_TX_FLAGS_TSTAMP)
3184 cmd_type_len |= E1000_ADVTXD_MAC_TSTAMP;
3186 if (tx_flags & IGB_TX_FLAGS_TSO) {
3187 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
3189 /* insert tcp checksum */
3190 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
3192 /* insert ip checksum */
3193 if (tx_flags & IGB_TX_FLAGS_IPV4)
3194 olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
3196 } else if (tx_flags & IGB_TX_FLAGS_CSUM) {
3197 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
3200 if ((adapter->flags & IGB_FLAG_NEED_CTX_IDX) &&
3201 (tx_flags & (IGB_TX_FLAGS_CSUM | IGB_TX_FLAGS_TSO |
3202 IGB_TX_FLAGS_VLAN)))
3203 olinfo_status |= tx_ring->queue_index << 4;
3205 olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
3207 i = tx_ring->next_to_use;
3209 buffer_info = &tx_ring->buffer_info[i];
3210 tx_desc = E1000_TX_DESC_ADV(*tx_ring, i);
3211 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
3212 tx_desc->read.cmd_type_len =
3213 cpu_to_le32(cmd_type_len | buffer_info->length);
3214 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
3216 if (i == tx_ring->count)
3220 tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
3221 /* Force memory writes to complete before letting h/w
3222 * know there are new descriptors to fetch. (Only
3223 * applicable for weak-ordered memory model archs,
3224 * such as IA-64). */
3227 tx_ring->next_to_use = i;
3228 writel(i, adapter->hw.hw_addr + tx_ring->tail);
3229 /* we need this if more than one processor can write to our tail
3230 * at a time, it syncronizes IO on IA64/Altix systems */
3234 static int __igb_maybe_stop_tx(struct net_device *netdev,
3235 struct igb_ring *tx_ring, int size)
3237 struct igb_adapter *adapter = netdev_priv(netdev);
3239 netif_stop_subqueue(netdev, tx_ring->queue_index);
3241 /* Herbert's original patch had:
3242 * smp_mb__after_netif_stop_queue();
3243 * but since that doesn't exist yet, just open code it. */
3246 /* We need to check again in a case another CPU has just
3247 * made room available. */
3248 if (igb_desc_unused(tx_ring) < size)
3252 netif_wake_subqueue(netdev, tx_ring->queue_index);
3253 ++adapter->restart_queue;
3257 static int igb_maybe_stop_tx(struct net_device *netdev,
3258 struct igb_ring *tx_ring, int size)
3260 if (igb_desc_unused(tx_ring) >= size)
3262 return __igb_maybe_stop_tx(netdev, tx_ring, size);
3265 static int igb_xmit_frame_ring_adv(struct sk_buff *skb,
3266 struct net_device *netdev,
3267 struct igb_ring *tx_ring)
3269 struct igb_adapter *adapter = netdev_priv(netdev);
3271 unsigned int tx_flags = 0;
3275 union skb_shared_tx *shtx;
3277 if (test_bit(__IGB_DOWN, &adapter->state)) {
3278 dev_kfree_skb_any(skb);
3279 return NETDEV_TX_OK;
3282 if (skb->len <= 0) {
3283 dev_kfree_skb_any(skb);
3284 return NETDEV_TX_OK;
3287 /* need: 1 descriptor per page,
3288 * + 2 desc gap to keep tail from touching head,
3289 * + 1 desc for skb->data,
3290 * + 1 desc for context descriptor,
3291 * otherwise try next time */
3292 if (igb_maybe_stop_tx(netdev, tx_ring, skb_shinfo(skb)->nr_frags + 4)) {
3293 /* this is a hard error */
3294 return NETDEV_TX_BUSY;
3298 * TODO: check that there currently is no other packet with
3299 * time stamping in the queue
3301 * When doing time stamping, keep the connection to the socket
3302 * a while longer: it is still needed by skb_hwtstamp_tx(),
3303 * called either in igb_tx_hwtstamp() or by our caller when
3304 * doing software time stamping.
3307 if (unlikely(shtx->hardware)) {
3308 shtx->in_progress = 1;
3309 tx_flags |= IGB_TX_FLAGS_TSTAMP;
3312 if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
3313 tx_flags |= IGB_TX_FLAGS_VLAN;
3314 tx_flags |= (vlan_tx_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
3317 if (skb->protocol == htons(ETH_P_IP))
3318 tx_flags |= IGB_TX_FLAGS_IPV4;
3320 first = tx_ring->next_to_use;
3321 tso = skb_is_gso(skb) ? igb_tso_adv(adapter, tx_ring, skb, tx_flags,
3325 dev_kfree_skb_any(skb);
3326 return NETDEV_TX_OK;
3330 tx_flags |= IGB_TX_FLAGS_TSO;
3331 else if (igb_tx_csum_adv(adapter, tx_ring, skb, tx_flags) &&
3332 (skb->ip_summed == CHECKSUM_PARTIAL))
3333 tx_flags |= IGB_TX_FLAGS_CSUM;
3336 * count reflects descriptors mapped, if 0 then mapping error
3337 * has occured and we need to rewind the descriptor queue
3339 count = igb_tx_map_adv(adapter, tx_ring, skb, first);
3342 igb_tx_queue_adv(adapter, tx_ring, tx_flags, count,
3344 netdev->trans_start = jiffies;
3345 /* Make sure there is space in the ring for the next send. */
3346 igb_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 4);
3348 dev_kfree_skb_any(skb);
3349 tx_ring->buffer_info[first].time_stamp = 0;
3350 tx_ring->next_to_use = first;
3353 return NETDEV_TX_OK;
3356 static int igb_xmit_frame_adv(struct sk_buff *skb, struct net_device *netdev)
3358 struct igb_adapter *adapter = netdev_priv(netdev);
3359 struct igb_ring *tx_ring;
3362 r_idx = skb->queue_mapping & (IGB_ABS_MAX_TX_QUEUES - 1);
3363 tx_ring = adapter->multi_tx_table[r_idx];
3365 /* This goes back to the question of how to logically map a tx queue
3366 * to a flow. Right now, performance is impacted slightly negatively
3367 * if using multiple tx queues. If the stack breaks away from a
3368 * single qdisc implementation, we can look at this again. */
3369 return (igb_xmit_frame_ring_adv(skb, netdev, tx_ring));
3373 * igb_tx_timeout - Respond to a Tx Hang
3374 * @netdev: network interface device structure
3376 static void igb_tx_timeout(struct net_device *netdev)
3378 struct igb_adapter *adapter = netdev_priv(netdev);
3379 struct e1000_hw *hw = &adapter->hw;
3381 /* Do the reset outside of interrupt context */
3382 adapter->tx_timeout_count++;
3383 schedule_work(&adapter->reset_task);
3385 (adapter->eims_enable_mask & ~adapter->eims_other));
3388 static void igb_reset_task(struct work_struct *work)
3390 struct igb_adapter *adapter;
3391 adapter = container_of(work, struct igb_adapter, reset_task);
3393 igb_reinit_locked(adapter);
3397 * igb_get_stats - Get System Network Statistics
3398 * @netdev: network interface device structure
3400 * Returns the address of the device statistics structure.
3401 * The statistics are actually updated from the timer callback.
3403 static struct net_device_stats *igb_get_stats(struct net_device *netdev)
3405 struct igb_adapter *adapter = netdev_priv(netdev);
3407 /* only return the current stats */
3408 return &adapter->net_stats;
3412 * igb_change_mtu - Change the Maximum Transfer Unit
3413 * @netdev: network interface device structure
3414 * @new_mtu: new value for maximum frame size
3416 * Returns 0 on success, negative on failure
3418 static int igb_change_mtu(struct net_device *netdev, int new_mtu)
3420 struct igb_adapter *adapter = netdev_priv(netdev);
3421 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3423 if ((max_frame < ETH_ZLEN + ETH_FCS_LEN) ||
3424 (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3425 dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
3429 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
3430 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
3434 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
3437 /* igb_down has a dependency on max_frame_size */
3438 adapter->max_frame_size = max_frame;
3439 if (netif_running(netdev))
3442 /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3443 * means we reserve 2 more, this pushes us to allocate from the next
3445 * i.e. RXBUFFER_2048 --> size-4096 slab
3448 if (max_frame <= IGB_RXBUFFER_256)
3449 adapter->rx_buffer_len = IGB_RXBUFFER_256;
3450 else if (max_frame <= IGB_RXBUFFER_512)
3451 adapter->rx_buffer_len = IGB_RXBUFFER_512;
3452 else if (max_frame <= IGB_RXBUFFER_1024)
3453 adapter->rx_buffer_len = IGB_RXBUFFER_1024;
3454 else if (max_frame <= IGB_RXBUFFER_2048)
3455 adapter->rx_buffer_len = IGB_RXBUFFER_2048;
3457 #if (PAGE_SIZE / 2) > IGB_RXBUFFER_16384
3458 adapter->rx_buffer_len = IGB_RXBUFFER_16384;
3460 adapter->rx_buffer_len = PAGE_SIZE / 2;
3463 /* if sr-iov is enabled we need to force buffer size to 1K or larger */
3464 if (adapter->vfs_allocated_count &&
3465 (adapter->rx_buffer_len < IGB_RXBUFFER_1024))
3466 adapter->rx_buffer_len = IGB_RXBUFFER_1024;
3468 /* adjust allocation if LPE protects us, and we aren't using SBP */
3469 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
3470 (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE))
3471 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3473 dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
3474 netdev->mtu, new_mtu);
3475 netdev->mtu = new_mtu;
3477 if (netif_running(netdev))
3482 clear_bit(__IGB_RESETTING, &adapter->state);
3488 * igb_update_stats - Update the board statistics counters
3489 * @adapter: board private structure
3492 void igb_update_stats(struct igb_adapter *adapter)
3494 struct e1000_hw *hw = &adapter->hw;
3495 struct pci_dev *pdev = adapter->pdev;
3498 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3501 * Prevent stats update while adapter is being reset, or if the pci
3502 * connection is down.
3504 if (adapter->link_speed == 0)
3506 if (pci_channel_offline(pdev))
3509 adapter->stats.crcerrs += rd32(E1000_CRCERRS);
3510 adapter->stats.gprc += rd32(E1000_GPRC);
3511 adapter->stats.gorc += rd32(E1000_GORCL);
3512 rd32(E1000_GORCH); /* clear GORCL */
3513 adapter->stats.bprc += rd32(E1000_BPRC);
3514 adapter->stats.mprc += rd32(E1000_MPRC);
3515 adapter->stats.roc += rd32(E1000_ROC);
3517 adapter->stats.prc64 += rd32(E1000_PRC64);
3518 adapter->stats.prc127 += rd32(E1000_PRC127);
3519 adapter->stats.prc255 += rd32(E1000_PRC255);
3520 adapter->stats.prc511 += rd32(E1000_PRC511);
3521 adapter->stats.prc1023 += rd32(E1000_PRC1023);
3522 adapter->stats.prc1522 += rd32(E1000_PRC1522);
3523 adapter->stats.symerrs += rd32(E1000_SYMERRS);
3524 adapter->stats.sec += rd32(E1000_SEC);
3526 adapter->stats.mpc += rd32(E1000_MPC);
3527 adapter->stats.scc += rd32(E1000_SCC);
3528 adapter->stats.ecol += rd32(E1000_ECOL);
3529 adapter->stats.mcc += rd32(E1000_MCC);
3530 adapter->stats.latecol += rd32(E1000_LATECOL);
3531 adapter->stats.dc += rd32(E1000_DC);
3532 adapter->stats.rlec += rd32(E1000_RLEC);
3533 adapter->stats.xonrxc += rd32(E1000_XONRXC);
3534 adapter->stats.xontxc += rd32(E1000_XONTXC);
3535 adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
3536 adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
3537 adapter->stats.fcruc += rd32(E1000_FCRUC);
3538 adapter->stats.gptc += rd32(E1000_GPTC);
3539 adapter->stats.gotc += rd32(E1000_GOTCL);
3540 rd32(E1000_GOTCH); /* clear GOTCL */
3541 adapter->stats.rnbc += rd32(E1000_RNBC);
3542 adapter->stats.ruc += rd32(E1000_RUC);
3543 adapter->stats.rfc += rd32(E1000_RFC);
3544 adapter->stats.rjc += rd32(E1000_RJC);
3545 adapter->stats.tor += rd32(E1000_TORH);
3546 adapter->stats.tot += rd32(E1000_TOTH);
3547 adapter->stats.tpr += rd32(E1000_TPR);
3549 adapter->stats.ptc64 += rd32(E1000_PTC64);
3550 adapter->stats.ptc127 += rd32(E1000_PTC127);
3551 adapter->stats.ptc255 += rd32(E1000_PTC255);
3552 adapter->stats.ptc511 += rd32(E1000_PTC511);
3553 adapter->stats.ptc1023 += rd32(E1000_PTC1023);
3554 adapter->stats.ptc1522 += rd32(E1000_PTC1522);
3556 adapter->stats.mptc += rd32(E1000_MPTC);
3557 adapter->stats.bptc += rd32(E1000_BPTC);
3559 /* used for adaptive IFS */
3561 hw->mac.tx_packet_delta = rd32(E1000_TPT);
3562 adapter->stats.tpt += hw->mac.tx_packet_delta;
3563 hw->mac.collision_delta = rd32(E1000_COLC);
3564 adapter->stats.colc += hw->mac.collision_delta;
3566 adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
3567 adapter->stats.rxerrc += rd32(E1000_RXERRC);
3568 adapter->stats.tncrs += rd32(E1000_TNCRS);
3569 adapter->stats.tsctc += rd32(E1000_TSCTC);
3570 adapter->stats.tsctfc += rd32(E1000_TSCTFC);
3572 adapter->stats.iac += rd32(E1000_IAC);
3573 adapter->stats.icrxoc += rd32(E1000_ICRXOC);
3574 adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
3575 adapter->stats.icrxatc += rd32(E1000_ICRXATC);
3576 adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
3577 adapter->stats.ictxatc += rd32(E1000_ICTXATC);
3578 adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
3579 adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
3580 adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
3582 /* Fill out the OS statistics structure */
3583 adapter->net_stats.multicast = adapter->stats.mprc;
3584 adapter->net_stats.collisions = adapter->stats.colc;
3588 /* RLEC on some newer hardware can be incorrect so build
3589 * our own version based on RUC and ROC */
3590 adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3591 adapter->stats.crcerrs + adapter->stats.algnerrc +
3592 adapter->stats.ruc + adapter->stats.roc +
3593 adapter->stats.cexterr;
3594 adapter->net_stats.rx_length_errors = adapter->stats.ruc +
3596 adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3597 adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3598 adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3601 adapter->net_stats.tx_errors = adapter->stats.ecol +
3602 adapter->stats.latecol;
3603 adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3604 adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3605 adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3607 /* Tx Dropped needs to be maintained elsewhere */
3610 if (hw->phy.media_type == e1000_media_type_copper) {
3611 if ((adapter->link_speed == SPEED_1000) &&
3612 (!igb_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
3613 phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3614 adapter->phy_stats.idle_errors += phy_tmp;
3618 /* Management Stats */
3619 adapter->stats.mgptc += rd32(E1000_MGTPTC);
3620 adapter->stats.mgprc += rd32(E1000_MGTPRC);
3621 adapter->stats.mgpdc += rd32(E1000_MGTPDC);
3624 static irqreturn_t igb_msix_other(int irq, void *data)
3626 struct net_device *netdev = data;
3627 struct igb_adapter *adapter = netdev_priv(netdev);
3628 struct e1000_hw *hw = &adapter->hw;
3629 u32 icr = rd32(E1000_ICR);
3631 /* reading ICR causes bit 31 of EICR to be cleared */
3633 if(icr & E1000_ICR_DOUTSYNC) {
3634 /* HW is reporting DMA is out of sync */
3635 adapter->stats.doosync++;
3638 /* Check for a mailbox event */
3639 if (icr & E1000_ICR_VMMB)
3640 igb_msg_task(adapter);
3642 if (icr & E1000_ICR_LSC) {
3643 hw->mac.get_link_status = 1;
3644 /* guard against interrupt when we're going down */
3645 if (!test_bit(__IGB_DOWN, &adapter->state))
3646 mod_timer(&adapter->watchdog_timer, jiffies + 1);
3649 wr32(E1000_IMS, E1000_IMS_LSC | E1000_IMS_DOUTSYNC | E1000_IMS_VMMB);
3650 wr32(E1000_EIMS, adapter->eims_other);
3655 static irqreturn_t igb_msix_tx(int irq, void *data)
3657 struct igb_ring *tx_ring = data;
3658 struct igb_adapter *adapter = tx_ring->adapter;
3659 struct e1000_hw *hw = &adapter->hw;
3661 #ifdef CONFIG_IGB_DCA
3662 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
3663 igb_update_tx_dca(tx_ring);
3666 tx_ring->total_bytes = 0;
3667 tx_ring->total_packets = 0;
3669 /* auto mask will automatically reenable the interrupt when we write
3671 if (!igb_clean_tx_irq(tx_ring))
3672 /* Ring was not completely cleaned, so fire another interrupt */
3673 wr32(E1000_EICS, tx_ring->eims_value);
3675 wr32(E1000_EIMS, tx_ring->eims_value);
3680 static void igb_write_itr(struct igb_ring *ring)
3682 struct e1000_hw *hw = &ring->adapter->hw;
3683 if ((ring->adapter->itr_setting & 3) && ring->set_itr) {
3684 switch (hw->mac.type) {
3686 wr32(ring->itr_register, ring->itr_val |
3690 wr32(ring->itr_register, ring->itr_val |
3691 (ring->itr_val << 16));
3698 static irqreturn_t igb_msix_rx(int irq, void *data)
3700 struct igb_ring *rx_ring = data;
3702 /* Write the ITR value calculated at the end of the
3703 * previous interrupt.
3706 igb_write_itr(rx_ring);
3708 if (napi_schedule_prep(&rx_ring->napi))
3709 __napi_schedule(&rx_ring->napi);
3711 #ifdef CONFIG_IGB_DCA
3712 if (rx_ring->adapter->flags & IGB_FLAG_DCA_ENABLED)
3713 igb_update_rx_dca(rx_ring);
3718 #ifdef CONFIG_IGB_DCA
3719 static void igb_update_rx_dca(struct igb_ring *rx_ring)
3722 struct igb_adapter *adapter = rx_ring->adapter;
3723 struct e1000_hw *hw = &adapter->hw;
3724 int cpu = get_cpu();
3725 int q = rx_ring->reg_idx;
3727 if (rx_ring->cpu != cpu) {
3728 dca_rxctrl = rd32(E1000_DCA_RXCTRL(q));
3729 if (hw->mac.type == e1000_82576) {
3730 dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK_82576;
3731 dca_rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu) <<
3732 E1000_DCA_RXCTRL_CPUID_SHIFT;
3734 dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK;
3735 dca_rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
3737 dca_rxctrl |= E1000_DCA_RXCTRL_DESC_DCA_EN;
3738 dca_rxctrl |= E1000_DCA_RXCTRL_HEAD_DCA_EN;
3739 dca_rxctrl |= E1000_DCA_RXCTRL_DATA_DCA_EN;
3740 wr32(E1000_DCA_RXCTRL(q), dca_rxctrl);
3746 static void igb_update_tx_dca(struct igb_ring *tx_ring)
3749 struct igb_adapter *adapter = tx_ring->adapter;
3750 struct e1000_hw *hw = &adapter->hw;
3751 int cpu = get_cpu();
3752 int q = tx_ring->reg_idx;
3754 if (tx_ring->cpu != cpu) {
3755 dca_txctrl = rd32(E1000_DCA_TXCTRL(q));
3756 if (hw->mac.type == e1000_82576) {
3757 dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK_82576;
3758 dca_txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu) <<
3759 E1000_DCA_TXCTRL_CPUID_SHIFT;
3761 dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK;
3762 dca_txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
3764 dca_txctrl |= E1000_DCA_TXCTRL_DESC_DCA_EN;
3765 wr32(E1000_DCA_TXCTRL(q), dca_txctrl);
3771 static void igb_setup_dca(struct igb_adapter *adapter)
3775 if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
3778 for (i = 0; i < adapter->num_tx_queues; i++) {
3779 adapter->tx_ring[i].cpu = -1;
3780 igb_update_tx_dca(&adapter->tx_ring[i]);
3782 for (i = 0; i < adapter->num_rx_queues; i++) {
3783 adapter->rx_ring[i].cpu = -1;
3784 igb_update_rx_dca(&adapter->rx_ring[i]);
3788 static int __igb_notify_dca(struct device *dev, void *data)
3790 struct net_device *netdev = dev_get_drvdata(dev);
3791 struct igb_adapter *adapter = netdev_priv(netdev);
3792 struct e1000_hw *hw = &adapter->hw;
3793 unsigned long event = *(unsigned long *)data;
3796 case DCA_PROVIDER_ADD:
3797 /* if already enabled, don't do it again */
3798 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
3800 /* Always use CB2 mode, difference is masked
3801 * in the CB driver. */
3802 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
3803 if (dca_add_requester(dev) == 0) {
3804 adapter->flags |= IGB_FLAG_DCA_ENABLED;
3805 dev_info(&adapter->pdev->dev, "DCA enabled\n");
3806 igb_setup_dca(adapter);
3809 /* Fall Through since DCA is disabled. */
3810 case DCA_PROVIDER_REMOVE:
3811 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
3812 /* without this a class_device is left
3813 * hanging around in the sysfs model */
3814 dca_remove_requester(dev);
3815 dev_info(&adapter->pdev->dev, "DCA disabled\n");
3816 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
3817 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
3825 static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
3830 ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
3833 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
3835 #endif /* CONFIG_IGB_DCA */
3837 static void igb_ping_all_vfs(struct igb_adapter *adapter)
3839 struct e1000_hw *hw = &adapter->hw;
3843 for (i = 0 ; i < adapter->vfs_allocated_count; i++) {
3844 ping = E1000_PF_CONTROL_MSG;
3845 if (adapter->vf_data[i].clear_to_send)
3846 ping |= E1000_VT_MSGTYPE_CTS;
3847 igb_write_mbx(hw, &ping, 1, i);
3851 static int igb_set_vf_multicasts(struct igb_adapter *adapter,
3852 u32 *msgbuf, u32 vf)
3854 int n = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
3855 u16 *hash_list = (u16 *)&msgbuf[1];
3856 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
3859 /* only up to 30 hash values supported */
3863 /* salt away the number of multi cast addresses assigned
3864 * to this VF for later use to restore when the PF multi cast
3867 vf_data->num_vf_mc_hashes = n;
3869 /* VFs are limited to using the MTA hash table for their multicast
3871 for (i = 0; i < n; i++)
3872 vf_data->vf_mc_hashes[i] = hash_list[i];;
3874 /* Flush and reset the mta with the new values */
3875 igb_set_multi(adapter->netdev);
3880 static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
3882 struct e1000_hw *hw = &adapter->hw;
3883 struct vf_data_storage *vf_data;
3886 for (i = 0; i < adapter->vfs_allocated_count; i++) {
3887 vf_data = &adapter->vf_data[i];
3888 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
3889 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
3893 static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
3895 struct e1000_hw *hw = &adapter->hw;
3896 u32 pool_mask, reg, vid;
3899 pool_mask = 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
3901 /* Find the vlan filter for this id */
3902 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
3903 reg = rd32(E1000_VLVF(i));
3905 /* remove the vf from the pool */
3908 /* if pool is empty then remove entry from vfta */
3909 if (!(reg & E1000_VLVF_POOLSEL_MASK) &&
3910 (reg & E1000_VLVF_VLANID_ENABLE)) {
3912 vid = reg & E1000_VLVF_VLANID_MASK;
3913 igb_vfta_set(hw, vid, false);
3916 wr32(E1000_VLVF(i), reg);
3920 static s32 igb_vlvf_set(struct igb_adapter *adapter, u32 vid, bool add, u32 vf)
3922 struct e1000_hw *hw = &adapter->hw;
3925 /* It is an error to call this function when VFs are not enabled */
3926 if (!adapter->vfs_allocated_count)
3929 /* Find the vlan filter for this id */
3930 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
3931 reg = rd32(E1000_VLVF(i));
3932 if ((reg & E1000_VLVF_VLANID_ENABLE) &&
3933 vid == (reg & E1000_VLVF_VLANID_MASK))
3938 if (i == E1000_VLVF_ARRAY_SIZE) {
3939 /* Did not find a matching VLAN ID entry that was
3940 * enabled. Search for a free filter entry, i.e.
3941 * one without the enable bit set
3943 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
3944 reg = rd32(E1000_VLVF(i));
3945 if (!(reg & E1000_VLVF_VLANID_ENABLE))
3949 if (i < E1000_VLVF_ARRAY_SIZE) {
3950 /* Found an enabled/available entry */
3951 reg |= 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
3953 /* if !enabled we need to set this up in vfta */
3954 if (!(reg & E1000_VLVF_VLANID_ENABLE)) {
3955 /* add VID to filter table, if bit already set
3956 * PF must have added it outside of table */
3957 if (igb_vfta_set(hw, vid, true))
3958 reg |= 1 << (E1000_VLVF_POOLSEL_SHIFT +
3959 adapter->vfs_allocated_count);
3960 reg |= E1000_VLVF_VLANID_ENABLE;
3962 reg &= ~E1000_VLVF_VLANID_MASK;
3965 wr32(E1000_VLVF(i), reg);
3969 if (i < E1000_VLVF_ARRAY_SIZE) {
3970 /* remove vf from the pool */
3971 reg &= ~(1 << (E1000_VLVF_POOLSEL_SHIFT + vf));
3972 /* if pool is empty then remove entry from vfta */
3973 if (!(reg & E1000_VLVF_POOLSEL_MASK)) {
3975 igb_vfta_set(hw, vid, false);
3977 wr32(E1000_VLVF(i), reg);
3984 static int igb_set_vf_vlan(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
3986 int add = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
3987 int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
3989 return igb_vlvf_set(adapter, vid, add, vf);
3992 static inline void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
3994 struct e1000_hw *hw = &adapter->hw;
3996 /* disable mailbox functionality for vf */
3997 adapter->vf_data[vf].clear_to_send = false;
3999 /* reset offloads to defaults */
4000 igb_set_vmolr(hw, vf);
4002 /* reset vlans for device */
4003 igb_clear_vf_vfta(adapter, vf);
4005 /* reset multicast table array for vf */
4006 adapter->vf_data[vf].num_vf_mc_hashes = 0;
4008 /* Flush and reset the mta with the new values */
4009 igb_set_multi(adapter->netdev);
4012 static inline void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
4014 struct e1000_hw *hw = &adapter->hw;
4015 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
4017 u8 *addr = (u8 *)(&msgbuf[1]);
4019 /* process all the same items cleared in a function level reset */
4020 igb_vf_reset_event(adapter, vf);
4022 /* set vf mac address */
4023 igb_rar_set(hw, vf_mac, vf + 1);
4024 igb_set_rah_pool(hw, vf, vf + 1);
4026 /* enable transmit and receive for vf */
4027 reg = rd32(E1000_VFTE);
4028 wr32(E1000_VFTE, reg | (1 << vf));
4029 reg = rd32(E1000_VFRE);
4030 wr32(E1000_VFRE, reg | (1 << vf));
4032 /* enable mailbox functionality for vf */
4033 adapter->vf_data[vf].clear_to_send = true;
4035 /* reply to reset with ack and vf mac address */
4036 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
4037 memcpy(addr, vf_mac, 6);
4038 igb_write_mbx(hw, msgbuf, 3, vf);
4041 static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf)
4043 unsigned char *addr = (char *)&msg[1];
4046 if (is_valid_ether_addr(addr))
4047 err = igb_set_vf_mac(adapter, vf, addr);
4053 static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf)
4055 struct e1000_hw *hw = &adapter->hw;
4056 u32 msg = E1000_VT_MSGTYPE_NACK;
4058 /* if device isn't clear to send it shouldn't be reading either */
4059 if (!adapter->vf_data[vf].clear_to_send)
4060 igb_write_mbx(hw, &msg, 1, vf);
4064 static void igb_msg_task(struct igb_adapter *adapter)
4066 struct e1000_hw *hw = &adapter->hw;
4069 for (vf = 0; vf < adapter->vfs_allocated_count; vf++) {
4070 /* process any reset requests */
4071 if (!igb_check_for_rst(hw, vf)) {
4072 adapter->vf_data[vf].clear_to_send = false;
4073 igb_vf_reset_event(adapter, vf);
4076 /* process any messages pending */
4077 if (!igb_check_for_msg(hw, vf))
4078 igb_rcv_msg_from_vf(adapter, vf);
4080 /* process any acks */
4081 if (!igb_check_for_ack(hw, vf))
4082 igb_rcv_ack_from_vf(adapter, vf);
4087 static int igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
4089 u32 mbx_size = E1000_VFMAILBOX_SIZE;
4090 u32 msgbuf[mbx_size];
4091 struct e1000_hw *hw = &adapter->hw;
4094 retval = igb_read_mbx(hw, msgbuf, mbx_size, vf);
4097 dev_err(&adapter->pdev->dev,
4098 "Error receiving message from VF\n");
4100 /* this is a message we already processed, do nothing */
4101 if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
4105 * until the vf completes a reset it should not be
4106 * allowed to start any configuration.
4109 if (msgbuf[0] == E1000_VF_RESET) {
4110 igb_vf_reset_msg(adapter, vf);
4115 if (!adapter->vf_data[vf].clear_to_send) {
4116 msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
4117 igb_write_mbx(hw, msgbuf, 1, vf);
4121 switch ((msgbuf[0] & 0xFFFF)) {
4122 case E1000_VF_SET_MAC_ADDR:
4123 retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
4125 case E1000_VF_SET_MULTICAST:
4126 retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
4128 case E1000_VF_SET_LPE:
4129 retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf);
4131 case E1000_VF_SET_VLAN:
4132 retval = igb_set_vf_vlan(adapter, msgbuf, vf);
4135 dev_err(&adapter->pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
4140 /* notify the VF of the results of what it sent us */
4142 msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
4144 msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
4146 msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
4148 igb_write_mbx(hw, msgbuf, 1, vf);
4154 * igb_intr_msi - Interrupt Handler
4155 * @irq: interrupt number
4156 * @data: pointer to a network interface device structure
4158 static irqreturn_t igb_intr_msi(int irq, void *data)
4160 struct net_device *netdev = data;
4161 struct igb_adapter *adapter = netdev_priv(netdev);
4162 struct e1000_hw *hw = &adapter->hw;
4163 /* read ICR disables interrupts using IAM */
4164 u32 icr = rd32(E1000_ICR);
4166 igb_write_itr(adapter->rx_ring);
4168 if(icr & E1000_ICR_DOUTSYNC) {
4169 /* HW is reporting DMA is out of sync */
4170 adapter->stats.doosync++;
4173 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
4174 hw->mac.get_link_status = 1;
4175 if (!test_bit(__IGB_DOWN, &adapter->state))
4176 mod_timer(&adapter->watchdog_timer, jiffies + 1);
4179 napi_schedule(&adapter->rx_ring[0].napi);
4185 * igb_intr - Legacy Interrupt Handler
4186 * @irq: interrupt number
4187 * @data: pointer to a network interface device structure
4189 static irqreturn_t igb_intr(int irq, void *data)
4191 struct net_device *netdev = data;
4192 struct igb_adapter *adapter = netdev_priv(netdev);
4193 struct e1000_hw *hw = &adapter->hw;
4194 /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked. No
4195 * need for the IMC write */
4196 u32 icr = rd32(E1000_ICR);
4198 return IRQ_NONE; /* Not our interrupt */
4200 igb_write_itr(adapter->rx_ring);
4202 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
4203 * not set, then the adapter didn't send an interrupt */
4204 if (!(icr & E1000_ICR_INT_ASSERTED))
4207 if(icr & E1000_ICR_DOUTSYNC) {
4208 /* HW is reporting DMA is out of sync */
4209 adapter->stats.doosync++;
4212 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
4213 hw->mac.get_link_status = 1;
4214 /* guard against interrupt when we're going down */
4215 if (!test_bit(__IGB_DOWN, &adapter->state))
4216 mod_timer(&adapter->watchdog_timer, jiffies + 1);
4219 napi_schedule(&adapter->rx_ring[0].napi);
4224 static inline void igb_rx_irq_enable(struct igb_ring *rx_ring)
4226 struct igb_adapter *adapter = rx_ring->adapter;
4227 struct e1000_hw *hw = &adapter->hw;
4229 if (adapter->itr_setting & 3) {
4230 if (adapter->num_rx_queues == 1)
4231 igb_set_itr(adapter);
4233 igb_update_ring_itr(rx_ring);
4236 if (!test_bit(__IGB_DOWN, &adapter->state)) {
4237 if (adapter->msix_entries)
4238 wr32(E1000_EIMS, rx_ring->eims_value);
4240 igb_irq_enable(adapter);
4245 * igb_poll - NAPI Rx polling callback
4246 * @napi: napi polling structure
4247 * @budget: count of how many packets we should handle
4249 static int igb_poll(struct napi_struct *napi, int budget)
4251 struct igb_ring *rx_ring = container_of(napi, struct igb_ring, napi);
4254 #ifdef CONFIG_IGB_DCA
4255 if (rx_ring->adapter->flags & IGB_FLAG_DCA_ENABLED)
4256 igb_update_rx_dca(rx_ring);
4258 igb_clean_rx_irq_adv(rx_ring, &work_done, budget);
4260 if (rx_ring->buddy) {
4261 #ifdef CONFIG_IGB_DCA
4262 if (rx_ring->adapter->flags & IGB_FLAG_DCA_ENABLED)
4263 igb_update_tx_dca(rx_ring->buddy);
4265 if (!igb_clean_tx_irq(rx_ring->buddy))
4269 /* If not enough Rx work done, exit the polling mode */
4270 if (work_done < budget) {
4271 napi_complete(napi);
4272 igb_rx_irq_enable(rx_ring);
4279 * igb_hwtstamp - utility function which checks for TX time stamp
4280 * @adapter: board private structure
4281 * @skb: packet that was just sent
4283 * If we were asked to do hardware stamping and such a time stamp is
4284 * available, then it must have been for this skb here because we only
4285 * allow only one such packet into the queue.
4287 static void igb_tx_hwtstamp(struct igb_adapter *adapter, struct sk_buff *skb)
4289 union skb_shared_tx *shtx = skb_tx(skb);
4290 struct e1000_hw *hw = &adapter->hw;
4292 if (unlikely(shtx->hardware)) {
4293 u32 valid = rd32(E1000_TSYNCTXCTL) & E1000_TSYNCTXCTL_VALID;
4295 u64 regval = rd32(E1000_TXSTMPL);
4297 struct skb_shared_hwtstamps shhwtstamps;
4299 memset(&shhwtstamps, 0, sizeof(shhwtstamps));
4300 regval |= (u64)rd32(E1000_TXSTMPH) << 32;
4301 ns = timecounter_cyc2time(&adapter->clock,
4303 timecompare_update(&adapter->compare, ns);
4304 shhwtstamps.hwtstamp = ns_to_ktime(ns);
4305 shhwtstamps.syststamp =
4306 timecompare_transform(&adapter->compare, ns);
4307 skb_tstamp_tx(skb, &shhwtstamps);
4313 * igb_clean_tx_irq - Reclaim resources after transmit completes
4314 * @adapter: board private structure
4315 * returns true if ring is completely cleaned
4317 static bool igb_clean_tx_irq(struct igb_ring *tx_ring)
4319 struct igb_adapter *adapter = tx_ring->adapter;
4320 struct net_device *netdev = adapter->netdev;
4321 struct e1000_hw *hw = &adapter->hw;
4322 struct igb_buffer *buffer_info;
4323 struct sk_buff *skb;
4324 union e1000_adv_tx_desc *tx_desc, *eop_desc;
4325 unsigned int total_bytes = 0, total_packets = 0;
4326 unsigned int i, eop, count = 0;
4327 bool cleaned = false;
4329 i = tx_ring->next_to_clean;
4330 eop = tx_ring->buffer_info[i].next_to_watch;
4331 eop_desc = E1000_TX_DESC_ADV(*tx_ring, eop);
4333 while ((eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)) &&
4334 (count < tx_ring->count)) {
4335 for (cleaned = false; !cleaned; count++) {
4336 tx_desc = E1000_TX_DESC_ADV(*tx_ring, i);
4337 buffer_info = &tx_ring->buffer_info[i];
4338 cleaned = (i == eop);
4339 skb = buffer_info->skb;
4342 unsigned int segs, bytecount;
4343 /* gso_segs is currently only valid for tcp */
4344 segs = skb_shinfo(skb)->gso_segs ?: 1;
4345 /* multiply data chunks by size of headers */
4346 bytecount = ((segs - 1) * skb_headlen(skb)) +
4348 total_packets += segs;
4349 total_bytes += bytecount;
4351 igb_tx_hwtstamp(adapter, skb);
4354 igb_unmap_and_free_tx_resource(adapter, buffer_info);
4355 tx_desc->wb.status = 0;
4358 if (i == tx_ring->count)
4361 eop = tx_ring->buffer_info[i].next_to_watch;
4362 eop_desc = E1000_TX_DESC_ADV(*tx_ring, eop);
4365 tx_ring->next_to_clean = i;
4367 if (unlikely(count &&
4368 netif_carrier_ok(netdev) &&
4369 igb_desc_unused(tx_ring) >= IGB_TX_QUEUE_WAKE)) {
4370 /* Make sure that anybody stopping the queue after this
4371 * sees the new next_to_clean.
4374 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
4375 !(test_bit(__IGB_DOWN, &adapter->state))) {
4376 netif_wake_subqueue(netdev, tx_ring->queue_index);
4377 ++adapter->restart_queue;
4381 if (tx_ring->detect_tx_hung) {
4382 /* Detect a transmit hang in hardware, this serializes the
4383 * check with the clearing of time_stamp and movement of i */
4384 tx_ring->detect_tx_hung = false;
4385 if (tx_ring->buffer_info[i].time_stamp &&
4386 time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
4387 (adapter->tx_timeout_factor * HZ))
4388 && !(rd32(E1000_STATUS) &
4389 E1000_STATUS_TXOFF)) {
4391 /* detected Tx unit hang */
4392 dev_err(&adapter->pdev->dev,
4393 "Detected Tx Unit Hang\n"
4397 " next_to_use <%x>\n"
4398 " next_to_clean <%x>\n"
4399 "buffer_info[next_to_clean]\n"
4400 " time_stamp <%lx>\n"
4401 " next_to_watch <%x>\n"
4403 " desc.status <%x>\n",
4404 tx_ring->queue_index,
4405 readl(adapter->hw.hw_addr + tx_ring->head),
4406 readl(adapter->hw.hw_addr + tx_ring->tail),
4407 tx_ring->next_to_use,
4408 tx_ring->next_to_clean,
4409 tx_ring->buffer_info[i].time_stamp,
4412 eop_desc->wb.status);
4413 netif_stop_subqueue(netdev, tx_ring->queue_index);
4416 tx_ring->total_bytes += total_bytes;
4417 tx_ring->total_packets += total_packets;
4418 tx_ring->tx_stats.bytes += total_bytes;
4419 tx_ring->tx_stats.packets += total_packets;
4420 adapter->net_stats.tx_bytes += total_bytes;
4421 adapter->net_stats.tx_packets += total_packets;
4422 return (count < tx_ring->count);
4426 * igb_receive_skb - helper function to handle rx indications
4427 * @ring: pointer to receive ring receving this packet
4428 * @status: descriptor status field as written by hardware
4429 * @rx_desc: receive descriptor containing vlan and type information.
4430 * @skb: pointer to sk_buff to be indicated to stack
4432 static void igb_receive_skb(struct igb_ring *ring, u8 status,
4433 union e1000_adv_rx_desc * rx_desc,
4434 struct sk_buff *skb)
4436 struct igb_adapter * adapter = ring->adapter;
4437 bool vlan_extracted = (adapter->vlgrp && (status & E1000_RXD_STAT_VP));
4439 skb_record_rx_queue(skb, ring->queue_index);
4440 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
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 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
4450 le16_to_cpu(rx_desc->wb.upper.vlan));
4452 netif_receive_skb(skb);
4456 static inline void igb_rx_checksum_adv(struct igb_adapter *adapter,
4457 u32 status_err, struct sk_buff *skb)
4459 skb->ip_summed = CHECKSUM_NONE;
4461 /* Ignore Checksum bit is set or checksum is disabled through ethtool */
4462 if ((status_err & E1000_RXD_STAT_IXSM) || !adapter->rx_csum)
4464 /* TCP/UDP checksum error bit is set */
4466 (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
4467 /* let the stack verify checksum errors */
4468 adapter->hw_csum_err++;
4471 /* It must be a TCP or UDP packet with a valid checksum */
4472 if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
4473 skb->ip_summed = CHECKSUM_UNNECESSARY;
4475 adapter->hw_csum_good++;
4478 static bool igb_clean_rx_irq_adv(struct igb_ring *rx_ring,
4479 int *work_done, int budget)
4481 struct igb_adapter *adapter = rx_ring->adapter;
4482 struct net_device *netdev = adapter->netdev;
4483 struct e1000_hw *hw = &adapter->hw;
4484 struct pci_dev *pdev = adapter->pdev;
4485 union e1000_adv_rx_desc *rx_desc , *next_rxd;
4486 struct igb_buffer *buffer_info , *next_buffer;
4487 struct sk_buff *skb;
4488 bool cleaned = false;
4489 int cleaned_count = 0;
4490 unsigned int total_bytes = 0, total_packets = 0;
4492 u32 length, hlen, staterr;
4494 i = rx_ring->next_to_clean;
4495 buffer_info = &rx_ring->buffer_info[i];
4496 rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
4497 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
4499 while (staterr & E1000_RXD_STAT_DD) {
4500 if (*work_done >= budget)
4504 skb = buffer_info->skb;
4505 prefetch(skb->data - NET_IP_ALIGN);
4506 buffer_info->skb = NULL;
4509 if (i == rx_ring->count)
4511 next_rxd = E1000_RX_DESC_ADV(*rx_ring, i);
4513 next_buffer = &rx_ring->buffer_info[i];
4515 length = le16_to_cpu(rx_desc->wb.upper.length);
4519 if (!adapter->rx_ps_hdr_size) {
4520 pci_unmap_single(pdev, buffer_info->dma,
4521 adapter->rx_buffer_len +
4523 PCI_DMA_FROMDEVICE);
4524 skb_put(skb, length);
4528 /* HW will not DMA in data larger than the given buffer, even
4529 * if it parses the (NFS, of course) header to be larger. In
4530 * that case, it fills the header buffer and spills the rest
4533 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hdr_info) &
4534 E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
4535 if (hlen > adapter->rx_ps_hdr_size)
4536 hlen = adapter->rx_ps_hdr_size;
4538 if (!skb_shinfo(skb)->nr_frags) {
4539 pci_unmap_single(pdev, buffer_info->dma,
4540 adapter->rx_ps_hdr_size + NET_IP_ALIGN,
4541 PCI_DMA_FROMDEVICE);
4546 pci_unmap_page(pdev, buffer_info->page_dma,
4547 PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
4548 buffer_info->page_dma = 0;
4550 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags++,
4552 buffer_info->page_offset,
4555 if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
4556 (page_count(buffer_info->page) != 1))
4557 buffer_info->page = NULL;
4559 get_page(buffer_info->page);
4562 skb->data_len += length;
4564 skb->truesize += length;
4567 if (!(staterr & E1000_RXD_STAT_EOP)) {
4568 buffer_info->skb = next_buffer->skb;
4569 buffer_info->dma = next_buffer->dma;
4570 next_buffer->skb = skb;
4571 next_buffer->dma = 0;
4576 * If this bit is set, then the RX registers contain
4577 * the time stamp. No other packet will be time
4578 * stamped until we read these registers, so read the
4579 * registers to make them available again. Because
4580 * only one packet can be time stamped at a time, we
4581 * know that the register values must belong to this
4582 * one here and therefore we don't need to compare
4583 * any of the additional attributes stored for it.
4585 * If nothing went wrong, then it should have a
4586 * skb_shared_tx that we can turn into a
4587 * skb_shared_hwtstamps.
4589 * TODO: can time stamping be triggered (thus locking
4590 * the registers) without the packet reaching this point
4591 * here? In that case RX time stamping would get stuck.
4593 * TODO: in "time stamp all packets" mode this bit is
4594 * not set. Need a global flag for this mode and then
4595 * always read the registers. Cannot be done without
4598 if (unlikely(staterr & E1000_RXD_STAT_TS)) {
4601 struct skb_shared_hwtstamps *shhwtstamps =
4604 WARN(!(rd32(E1000_TSYNCRXCTL) & E1000_TSYNCRXCTL_VALID),
4605 "igb: no RX time stamp available for time stamped packet");
4606 regval = rd32(E1000_RXSTMPL);
4607 regval |= (u64)rd32(E1000_RXSTMPH) << 32;
4608 ns = timecounter_cyc2time(&adapter->clock, regval);
4609 timecompare_update(&adapter->compare, ns);
4610 memset(shhwtstamps, 0, sizeof(*shhwtstamps));
4611 shhwtstamps->hwtstamp = ns_to_ktime(ns);
4612 shhwtstamps->syststamp =
4613 timecompare_transform(&adapter->compare, ns);
4616 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
4617 dev_kfree_skb_irq(skb);
4621 total_bytes += skb->len;
4624 igb_rx_checksum_adv(adapter, staterr, skb);
4626 skb->protocol = eth_type_trans(skb, netdev);
4628 igb_receive_skb(rx_ring, staterr, rx_desc, skb);
4631 rx_desc->wb.upper.status_error = 0;
4633 /* return some buffers to hardware, one at a time is too slow */
4634 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
4635 igb_alloc_rx_buffers_adv(rx_ring, cleaned_count);
4639 /* use prefetched values */
4641 buffer_info = next_buffer;
4642 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
4645 rx_ring->next_to_clean = i;
4646 cleaned_count = igb_desc_unused(rx_ring);
4649 igb_alloc_rx_buffers_adv(rx_ring, cleaned_count);
4651 rx_ring->total_packets += total_packets;
4652 rx_ring->total_bytes += total_bytes;
4653 rx_ring->rx_stats.packets += total_packets;
4654 rx_ring->rx_stats.bytes += total_bytes;
4655 adapter->net_stats.rx_bytes += total_bytes;
4656 adapter->net_stats.rx_packets += total_packets;
4661 * igb_alloc_rx_buffers_adv - Replace used receive buffers; packet split
4662 * @adapter: address of board private structure
4664 static void igb_alloc_rx_buffers_adv(struct igb_ring *rx_ring,
4667 struct igb_adapter *adapter = rx_ring->adapter;
4668 struct net_device *netdev = adapter->netdev;
4669 struct pci_dev *pdev = adapter->pdev;
4670 union e1000_adv_rx_desc *rx_desc;
4671 struct igb_buffer *buffer_info;
4672 struct sk_buff *skb;
4676 i = rx_ring->next_to_use;
4677 buffer_info = &rx_ring->buffer_info[i];
4679 if (adapter->rx_ps_hdr_size)
4680 bufsz = adapter->rx_ps_hdr_size;
4682 bufsz = adapter->rx_buffer_len;
4683 bufsz += NET_IP_ALIGN;
4685 while (cleaned_count--) {
4686 rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
4688 if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) {
4689 if (!buffer_info->page) {
4690 buffer_info->page = alloc_page(GFP_ATOMIC);
4691 if (!buffer_info->page) {
4692 adapter->alloc_rx_buff_failed++;
4695 buffer_info->page_offset = 0;
4697 buffer_info->page_offset ^= PAGE_SIZE / 2;
4699 buffer_info->page_dma =
4700 pci_map_page(pdev, buffer_info->page,
4701 buffer_info->page_offset,
4703 PCI_DMA_FROMDEVICE);
4706 if (!buffer_info->skb) {
4707 skb = netdev_alloc_skb(netdev, bufsz);
4709 adapter->alloc_rx_buff_failed++;
4713 /* Make buffer alignment 2 beyond a 16 byte boundary
4714 * this will result in a 16 byte aligned IP header after
4715 * the 14 byte MAC header is removed
4717 skb_reserve(skb, NET_IP_ALIGN);
4719 buffer_info->skb = skb;
4720 buffer_info->dma = pci_map_single(pdev, skb->data,
4722 PCI_DMA_FROMDEVICE);
4724 /* Refresh the desc even if buffer_addrs didn't change because
4725 * each write-back erases this info. */
4726 if (adapter->rx_ps_hdr_size) {
4727 rx_desc->read.pkt_addr =
4728 cpu_to_le64(buffer_info->page_dma);
4729 rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
4731 rx_desc->read.pkt_addr =
4732 cpu_to_le64(buffer_info->dma);
4733 rx_desc->read.hdr_addr = 0;
4737 if (i == rx_ring->count)
4739 buffer_info = &rx_ring->buffer_info[i];
4743 if (rx_ring->next_to_use != i) {
4744 rx_ring->next_to_use = i;
4746 i = (rx_ring->count - 1);
4750 /* Force memory writes to complete before letting h/w
4751 * know there are new descriptors to fetch. (Only
4752 * applicable for weak-ordered memory model archs,
4753 * such as IA-64). */
4755 writel(i, adapter->hw.hw_addr + rx_ring->tail);
4765 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4767 struct igb_adapter *adapter = netdev_priv(netdev);
4768 struct mii_ioctl_data *data = if_mii(ifr);
4770 if (adapter->hw.phy.media_type != e1000_media_type_copper)
4775 data->phy_id = adapter->hw.phy.addr;
4778 if (!capable(CAP_NET_ADMIN))
4780 if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
4792 * igb_hwtstamp_ioctl - control hardware time stamping
4797 * Outgoing time stamping can be enabled and disabled. Play nice and
4798 * disable it when requested, although it shouldn't case any overhead
4799 * when no packet needs it. At most one packet in the queue may be
4800 * marked for time stamping, otherwise it would be impossible to tell
4801 * for sure to which packet the hardware time stamp belongs.
4803 * Incoming time stamping has to be configured via the hardware
4804 * filters. Not all combinations are supported, in particular event
4805 * type has to be specified. Matching the kind of event packet is
4806 * not supported, with the exception of "all V2 events regardless of
4810 static int igb_hwtstamp_ioctl(struct net_device *netdev,
4811 struct ifreq *ifr, int cmd)
4813 struct igb_adapter *adapter = netdev_priv(netdev);
4814 struct e1000_hw *hw = &adapter->hw;
4815 struct hwtstamp_config config;
4816 u32 tsync_tx_ctl_bit = E1000_TSYNCTXCTL_ENABLED;
4817 u32 tsync_rx_ctl_bit = E1000_TSYNCRXCTL_ENABLED;
4818 u32 tsync_rx_ctl_type = 0;
4819 u32 tsync_rx_cfg = 0;
4822 short port = 319; /* PTP */
4825 if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
4828 /* reserved for future extensions */
4832 switch (config.tx_type) {
4833 case HWTSTAMP_TX_OFF:
4834 tsync_tx_ctl_bit = 0;
4836 case HWTSTAMP_TX_ON:
4837 tsync_tx_ctl_bit = E1000_TSYNCTXCTL_ENABLED;
4843 switch (config.rx_filter) {
4844 case HWTSTAMP_FILTER_NONE:
4845 tsync_rx_ctl_bit = 0;
4847 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
4848 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
4849 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
4850 case HWTSTAMP_FILTER_ALL:
4852 * register TSYNCRXCFG must be set, therefore it is not
4853 * possible to time stamp both Sync and Delay_Req messages
4854 * => fall back to time stamping all packets
4856 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_ALL;
4857 config.rx_filter = HWTSTAMP_FILTER_ALL;
4859 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
4860 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_L4_V1;
4861 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V1_SYNC_MESSAGE;
4864 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
4865 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_L4_V1;
4866 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V1_DELAY_REQ_MESSAGE;
4869 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
4870 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
4871 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
4872 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V2_SYNC_MESSAGE;
4875 config.rx_filter = HWTSTAMP_FILTER_SOME;
4877 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
4878 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
4879 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
4880 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V2_DELAY_REQ_MESSAGE;
4883 config.rx_filter = HWTSTAMP_FILTER_SOME;
4885 case HWTSTAMP_FILTER_PTP_V2_EVENT:
4886 case HWTSTAMP_FILTER_PTP_V2_SYNC:
4887 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
4888 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_EVENT_V2;
4889 config.rx_filter = HWTSTAMP_FILTER_PTP_V2_EVENT;
4896 /* enable/disable TX */
4897 regval = rd32(E1000_TSYNCTXCTL);
4898 regval = (regval & ~E1000_TSYNCTXCTL_ENABLED) | tsync_tx_ctl_bit;
4899 wr32(E1000_TSYNCTXCTL, regval);
4901 /* enable/disable RX, define which PTP packets are time stamped */
4902 regval = rd32(E1000_TSYNCRXCTL);
4903 regval = (regval & ~E1000_TSYNCRXCTL_ENABLED) | tsync_rx_ctl_bit;
4904 regval = (regval & ~0xE) | tsync_rx_ctl_type;
4905 wr32(E1000_TSYNCRXCTL, regval);
4906 wr32(E1000_TSYNCRXCFG, tsync_rx_cfg);
4909 * Ethertype Filter Queue Filter[0][15:0] = 0x88F7
4910 * (Ethertype to filter on)
4911 * Ethertype Filter Queue Filter[0][26] = 0x1 (Enable filter)
4912 * Ethertype Filter Queue Filter[0][30] = 0x1 (Enable Timestamping)
4914 wr32(E1000_ETQF0, is_l2 ? 0x440088f7 : 0);
4916 /* L4 Queue Filter[0]: only filter by source and destination port */
4917 wr32(E1000_SPQF0, htons(port));
4918 wr32(E1000_IMIREXT(0), is_l4 ?
4919 ((1<<12) | (1<<19) /* bypass size and control flags */) : 0);
4920 wr32(E1000_IMIR(0), is_l4 ?
4922 | (0<<16) /* immediate interrupt disabled */
4923 | 0 /* (1<<17) bit cleared: do not bypass
4924 destination port check */)
4926 wr32(E1000_FTQF0, is_l4 ?
4928 | (1<<15) /* VF not compared */
4929 | (1<<27) /* Enable Timestamping */
4930 | (7<<28) /* only source port filter enabled,
4931 source/target address and protocol
4933 : ((1<<15) | (15<<28) /* all mask bits set = filter not
4938 adapter->hwtstamp_config = config;
4940 /* clear TX/RX time stamp registers, just to be sure */
4941 regval = rd32(E1000_TXSTMPH);
4942 regval = rd32(E1000_RXSTMPH);
4944 return copy_to_user(ifr->ifr_data, &config, sizeof(config)) ?
4954 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4960 return igb_mii_ioctl(netdev, ifr, cmd);
4962 return igb_hwtstamp_ioctl(netdev, ifr, cmd);
4968 static void igb_vlan_rx_register(struct net_device *netdev,
4969 struct vlan_group *grp)
4971 struct igb_adapter *adapter = netdev_priv(netdev);
4972 struct e1000_hw *hw = &adapter->hw;
4975 igb_irq_disable(adapter);
4976 adapter->vlgrp = grp;
4979 /* enable VLAN tag insert/strip */
4980 ctrl = rd32(E1000_CTRL);
4981 ctrl |= E1000_CTRL_VME;
4982 wr32(E1000_CTRL, ctrl);
4984 /* enable VLAN receive filtering */
4985 rctl = rd32(E1000_RCTL);
4986 rctl &= ~E1000_RCTL_CFIEN;
4987 wr32(E1000_RCTL, rctl);
4988 igb_update_mng_vlan(adapter);
4990 /* disable VLAN tag insert/strip */
4991 ctrl = rd32(E1000_CTRL);
4992 ctrl &= ~E1000_CTRL_VME;
4993 wr32(E1000_CTRL, ctrl);
4995 if (adapter->mng_vlan_id != (u16)IGB_MNG_VLAN_NONE) {
4996 igb_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
4997 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
5001 igb_rlpml_set(adapter);
5003 if (!test_bit(__IGB_DOWN, &adapter->state))
5004 igb_irq_enable(adapter);
5007 static void igb_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
5009 struct igb_adapter *adapter = netdev_priv(netdev);
5010 struct e1000_hw *hw = &adapter->hw;
5011 int pf_id = adapter->vfs_allocated_count;
5013 if ((hw->mng_cookie.status &
5014 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
5015 (vid == adapter->mng_vlan_id))
5018 /* add vid to vlvf if sr-iov is enabled,
5019 * if that fails add directly to filter table */
5020 if (igb_vlvf_set(adapter, vid, true, pf_id))
5021 igb_vfta_set(hw, vid, true);
5025 static void igb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
5027 struct igb_adapter *adapter = netdev_priv(netdev);
5028 struct e1000_hw *hw = &adapter->hw;
5029 int pf_id = adapter->vfs_allocated_count;
5031 igb_irq_disable(adapter);
5032 vlan_group_set_device(adapter->vlgrp, vid, NULL);
5034 if (!test_bit(__IGB_DOWN, &adapter->state))
5035 igb_irq_enable(adapter);
5037 if ((adapter->hw.mng_cookie.status &
5038 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
5039 (vid == adapter->mng_vlan_id)) {
5040 /* release control to f/w */
5041 igb_release_hw_control(adapter);
5045 /* remove vid from vlvf if sr-iov is enabled,
5046 * if not in vlvf remove from vfta */
5047 if (igb_vlvf_set(adapter, vid, false, pf_id))
5048 igb_vfta_set(hw, vid, false);
5051 static void igb_restore_vlan(struct igb_adapter *adapter)
5053 igb_vlan_rx_register(adapter->netdev, adapter->vlgrp);
5055 if (adapter->vlgrp) {
5057 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
5058 if (!vlan_group_get_device(adapter->vlgrp, vid))
5060 igb_vlan_rx_add_vid(adapter->netdev, vid);
5065 int igb_set_spd_dplx(struct igb_adapter *adapter, u16 spddplx)
5067 struct e1000_mac_info *mac = &adapter->hw.mac;
5071 /* Fiber NICs only allow 1000 gbps Full duplex */
5072 if ((adapter->hw.phy.media_type == e1000_media_type_fiber) &&
5073 spddplx != (SPEED_1000 + DUPLEX_FULL)) {
5074 dev_err(&adapter->pdev->dev,
5075 "Unsupported Speed/Duplex configuration\n");
5080 case SPEED_10 + DUPLEX_HALF:
5081 mac->forced_speed_duplex = ADVERTISE_10_HALF;
5083 case SPEED_10 + DUPLEX_FULL:
5084 mac->forced_speed_duplex = ADVERTISE_10_FULL;
5086 case SPEED_100 + DUPLEX_HALF:
5087 mac->forced_speed_duplex = ADVERTISE_100_HALF;
5089 case SPEED_100 + DUPLEX_FULL:
5090 mac->forced_speed_duplex = ADVERTISE_100_FULL;
5092 case SPEED_1000 + DUPLEX_FULL:
5094 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
5096 case SPEED_1000 + DUPLEX_HALF: /* not supported */
5098 dev_err(&adapter->pdev->dev,
5099 "Unsupported Speed/Duplex configuration\n");
5105 static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake)
5107 struct net_device *netdev = pci_get_drvdata(pdev);
5108 struct igb_adapter *adapter = netdev_priv(netdev);
5109 struct e1000_hw *hw = &adapter->hw;
5110 u32 ctrl, rctl, status;
5111 u32 wufc = adapter->wol;
5116 netif_device_detach(netdev);
5118 if (netif_running(netdev))
5121 igb_reset_interrupt_capability(adapter);
5123 igb_free_queues(adapter);
5126 retval = pci_save_state(pdev);
5131 status = rd32(E1000_STATUS);
5132 if (status & E1000_STATUS_LU)
5133 wufc &= ~E1000_WUFC_LNKC;
5136 igb_setup_rctl(adapter);
5137 igb_set_multi(netdev);
5139 /* turn on all-multi mode if wake on multicast is enabled */
5140 if (wufc & E1000_WUFC_MC) {
5141 rctl = rd32(E1000_RCTL);
5142 rctl |= E1000_RCTL_MPE;
5143 wr32(E1000_RCTL, rctl);
5146 ctrl = rd32(E1000_CTRL);
5147 /* advertise wake from D3Cold */
5148 #define E1000_CTRL_ADVD3WUC 0x00100000
5149 /* phy power management enable */
5150 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5151 ctrl |= E1000_CTRL_ADVD3WUC;
5152 wr32(E1000_CTRL, ctrl);
5154 /* Allow time for pending master requests to run */
5155 igb_disable_pcie_master(&adapter->hw);
5157 wr32(E1000_WUC, E1000_WUC_PME_EN);
5158 wr32(E1000_WUFC, wufc);
5161 wr32(E1000_WUFC, 0);
5164 *enable_wake = wufc || adapter->en_mng_pt;
5166 igb_shutdown_fiber_serdes_link_82575(hw);
5168 /* Release control of h/w to f/w. If f/w is AMT enabled, this
5169 * would have already happened in close and is redundant. */
5170 igb_release_hw_control(adapter);
5172 pci_disable_device(pdev);
5178 static int igb_suspend(struct pci_dev *pdev, pm_message_t state)
5183 retval = __igb_shutdown(pdev, &wake);
5188 pci_prepare_to_sleep(pdev);
5190 pci_wake_from_d3(pdev, false);
5191 pci_set_power_state(pdev, PCI_D3hot);
5197 static int igb_resume(struct pci_dev *pdev)
5199 struct net_device *netdev = pci_get_drvdata(pdev);
5200 struct igb_adapter *adapter = netdev_priv(netdev);
5201 struct e1000_hw *hw = &adapter->hw;
5204 pci_set_power_state(pdev, PCI_D0);
5205 pci_restore_state(pdev);
5207 err = pci_enable_device_mem(pdev);
5210 "igb: Cannot enable PCI device from suspend\n");
5213 pci_set_master(pdev);
5215 pci_enable_wake(pdev, PCI_D3hot, 0);
5216 pci_enable_wake(pdev, PCI_D3cold, 0);
5218 igb_set_interrupt_capability(adapter);
5220 if (igb_alloc_queues(adapter)) {
5221 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
5225 /* e1000_power_up_phy(adapter); */
5229 /* let the f/w know that the h/w is now under the control of the
5231 igb_get_hw_control(adapter);
5233 wr32(E1000_WUS, ~0);
5235 if (netif_running(netdev)) {
5236 err = igb_open(netdev);
5241 netif_device_attach(netdev);
5247 static void igb_shutdown(struct pci_dev *pdev)
5251 __igb_shutdown(pdev, &wake);
5253 if (system_state == SYSTEM_POWER_OFF) {
5254 pci_wake_from_d3(pdev, wake);
5255 pci_set_power_state(pdev, PCI_D3hot);
5259 #ifdef CONFIG_NET_POLL_CONTROLLER
5261 * Polling 'interrupt' - used by things like netconsole to send skbs
5262 * without having to re-enable interrupts. It's not called while
5263 * the interrupt routine is executing.
5265 static void igb_netpoll(struct net_device *netdev)
5267 struct igb_adapter *adapter = netdev_priv(netdev);
5268 struct e1000_hw *hw = &adapter->hw;
5271 if (!adapter->msix_entries) {
5272 igb_irq_disable(adapter);
5273 napi_schedule(&adapter->rx_ring[0].napi);
5277 for (i = 0; i < adapter->num_tx_queues; i++) {
5278 struct igb_ring *tx_ring = &adapter->tx_ring[i];
5279 wr32(E1000_EIMC, tx_ring->eims_value);
5280 igb_clean_tx_irq(tx_ring);
5281 wr32(E1000_EIMS, tx_ring->eims_value);
5284 for (i = 0; i < adapter->num_rx_queues; i++) {
5285 struct igb_ring *rx_ring = &adapter->rx_ring[i];
5286 wr32(E1000_EIMC, rx_ring->eims_value);
5287 napi_schedule(&rx_ring->napi);
5290 #endif /* CONFIG_NET_POLL_CONTROLLER */
5293 * igb_io_error_detected - called when PCI error is detected
5294 * @pdev: Pointer to PCI device
5295 * @state: The current pci connection state
5297 * This function is called after a PCI bus error affecting
5298 * this device has been detected.
5300 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
5301 pci_channel_state_t state)
5303 struct net_device *netdev = pci_get_drvdata(pdev);
5304 struct igb_adapter *adapter = netdev_priv(netdev);
5306 netif_device_detach(netdev);
5308 if (netif_running(netdev))
5310 pci_disable_device(pdev);
5312 /* Request a slot slot reset. */
5313 return PCI_ERS_RESULT_NEED_RESET;
5317 * igb_io_slot_reset - called after the pci bus has been reset.
5318 * @pdev: Pointer to PCI device
5320 * Restart the card from scratch, as if from a cold-boot. Implementation
5321 * resembles the first-half of the igb_resume routine.
5323 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
5325 struct net_device *netdev = pci_get_drvdata(pdev);
5326 struct igb_adapter *adapter = netdev_priv(netdev);
5327 struct e1000_hw *hw = &adapter->hw;
5328 pci_ers_result_t result;
5331 if (pci_enable_device_mem(pdev)) {
5333 "Cannot re-enable PCI device after reset.\n");
5334 result = PCI_ERS_RESULT_DISCONNECT;
5336 pci_set_master(pdev);
5337 pci_restore_state(pdev);
5339 pci_enable_wake(pdev, PCI_D3hot, 0);
5340 pci_enable_wake(pdev, PCI_D3cold, 0);
5343 wr32(E1000_WUS, ~0);
5344 result = PCI_ERS_RESULT_RECOVERED;
5347 err = pci_cleanup_aer_uncorrect_error_status(pdev);
5349 dev_err(&pdev->dev, "pci_cleanup_aer_uncorrect_error_status "
5350 "failed 0x%0x\n", err);
5351 /* non-fatal, continue */
5358 * igb_io_resume - called when traffic can start flowing again.
5359 * @pdev: Pointer to PCI device
5361 * This callback is called when the error recovery driver tells us that
5362 * its OK to resume normal operation. Implementation resembles the
5363 * second-half of the igb_resume routine.
5365 static void igb_io_resume(struct pci_dev *pdev)
5367 struct net_device *netdev = pci_get_drvdata(pdev);
5368 struct igb_adapter *adapter = netdev_priv(netdev);
5370 if (netif_running(netdev)) {
5371 if (igb_up(adapter)) {
5372 dev_err(&pdev->dev, "igb_up failed after reset\n");
5377 netif_device_attach(netdev);
5379 /* let the f/w know that the h/w is now under the control of the
5381 igb_get_hw_control(adapter);
5384 static inline void igb_set_vmolr(struct e1000_hw *hw, int vfn)
5388 reg_data = rd32(E1000_VMOLR(vfn));
5389 reg_data |= E1000_VMOLR_BAM | /* Accept broadcast */
5390 E1000_VMOLR_ROPE | /* Accept packets matched in UTA */
5391 E1000_VMOLR_ROMPE | /* Accept packets matched in MTA */
5392 E1000_VMOLR_AUPE | /* Accept untagged packets */
5393 E1000_VMOLR_STRVLAN; /* Strip vlan tags */
5394 wr32(E1000_VMOLR(vfn), reg_data);
5397 static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
5400 struct e1000_hw *hw = &adapter->hw;
5403 vmolr = rd32(E1000_VMOLR(vfn));
5404 vmolr &= ~E1000_VMOLR_RLPML_MASK;
5405 vmolr |= size | E1000_VMOLR_LPE;
5406 wr32(E1000_VMOLR(vfn), vmolr);
5411 static inline void igb_set_rah_pool(struct e1000_hw *hw, int pool, int entry)
5415 reg_data = rd32(E1000_RAH(entry));
5416 reg_data &= ~E1000_RAH_POOL_MASK;
5417 reg_data |= E1000_RAH_POOL_1 << pool;;
5418 wr32(E1000_RAH(entry), reg_data);
5421 static void igb_set_mc_list_pools(struct igb_adapter *adapter,
5422 int entry_count, u16 total_rar_filters)
5424 struct e1000_hw *hw = &adapter->hw;
5425 int i = adapter->vfs_allocated_count + 1;
5427 if ((i + entry_count) < total_rar_filters)
5428 total_rar_filters = i + entry_count;
5430 for (; i < total_rar_filters; i++)
5431 igb_set_rah_pool(hw, adapter->vfs_allocated_count, i);
5434 static int igb_set_vf_mac(struct igb_adapter *adapter,
5435 int vf, unsigned char *mac_addr)
5437 struct e1000_hw *hw = &adapter->hw;
5438 int rar_entry = vf + 1; /* VF MAC addresses start at entry 1 */
5440 igb_rar_set(hw, mac_addr, rar_entry);
5442 memcpy(adapter->vf_data[vf].vf_mac_addresses, mac_addr, ETH_ALEN);
5444 igb_set_rah_pool(hw, vf, rar_entry);
5449 static void igb_vmm_control(struct igb_adapter *adapter)
5451 struct e1000_hw *hw = &adapter->hw;
5454 if (!adapter->vfs_allocated_count)
5457 /* VF's need PF reset indication before they
5458 * can send/receive mail */
5459 reg_data = rd32(E1000_CTRL_EXT);
5460 reg_data |= E1000_CTRL_EXT_PFRSTD;
5461 wr32(E1000_CTRL_EXT, reg_data);
5463 igb_vmdq_set_loopback_pf(hw, true);
5464 igb_vmdq_set_replication_pf(hw, true);