]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blame - drivers/net/igb/igb_main.c
projects / mirror_ubuntu-zesty-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
igb: rework igb_set_multi so that vfs are properly updated
[mirror_ubuntu-zesty-kernel.git] / drivers / net / igb / igb_main.c
CommitLineData
9d5c8243
AK		 1/*******************************************************************************
2
3 Intel(R) Gigabit Ethernet Linux driver
86d5d38f 4 Copyright(c) 2007-2009 Intel Corporation.
9d5c8243
AK		 5
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.
9
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
13 more details.
14
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.
18
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
21
22 Contact Information:
23 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
25
26*******************************************************************************/
27
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>
9d5c8243
AK		 34#include <linux/ipv6.h>
35#include <net/checksum.h>
36#include <net/ip6_checksum.h>
c6cb090b 37#include <linux/net_tstamp.h>
9d5c8243
AK		 38#include <linux/mii.h>
39#include <linux/ethtool.h>
40#include <linux/if_vlan.h>
41#include <linux/pci.h>
c54106bb 42#include <linux/pci-aspm.h>
9d5c8243
AK		 43#include <linux/delay.h>
44#include <linux/interrupt.h>
45#include <linux/if_ether.h>
40a914fa 46#include <linux/aer.h>
421e02f0 47#ifdef CONFIG_IGB_DCA
fe4506b6
JC		 48#include <linux/dca.h>
49#endif
9d5c8243
AK		 50#include "igb.h"
51
86d5d38f 52#define DRV_VERSION "1.3.16-k2"
9d5c8243
AK		 53char igb_driver_name[] = "igb";
54char igb_driver_version[] = DRV_VERSION;
55static const char igb_driver_string[] =
56 "Intel(R) Gigabit Ethernet Network Driver";
86d5d38f 57static const char igb_copyright[] = "Copyright (c) 2007-2009 Intel Corporation.";
9d5c8243 58
9d5c8243
AK		 59static const struct e1000_info *igb_info_tbl[] = {
60 [board_82575] = &e1000_82575_info,
61};
62
63static struct pci_device_id igb_pci_tbl[] = {
2d064c06 64 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576), board_82575 },
9eb2341d 65 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS), board_82575 },
2d064c06
AD		 66 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_FIBER), board_82575 },
67 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES), board_82575 },
c8ea5ea9 68 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER), board_82575 },
9d5c8243
AK		 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 */
73 {0, }
74};
75
76MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
77
78void igb_reset(struct igb_adapter *);
79static int igb_setup_all_tx_resources(struct igb_adapter *);
80static int igb_setup_all_rx_resources(struct igb_adapter *);
81static void igb_free_all_tx_resources(struct igb_adapter *);
82static void igb_free_all_rx_resources(struct igb_adapter *);
9d5c8243
AK		 83void igb_update_stats(struct igb_adapter *);
84static int igb_probe(struct pci_dev *, const struct pci_device_id *);
85static void __devexit igb_remove(struct pci_dev *pdev);
86static int igb_sw_init(struct igb_adapter *);
87static int igb_open(struct net_device *);
88static int igb_close(struct net_device *);
89static void igb_configure_tx(struct igb_adapter *);
90static void igb_configure_rx(struct igb_adapter *);
91static void igb_setup_rctl(struct igb_adapter *);
92static void igb_clean_all_tx_rings(struct igb_adapter *);
93static void igb_clean_all_rx_rings(struct igb_adapter *);
3b644cf6
MW		 94static void igb_clean_tx_ring(struct igb_ring *);
95static void igb_clean_rx_ring(struct igb_ring *);
9d5c8243
AK		 96static void igb_set_multi(struct net_device *);
97static void igb_update_phy_info(unsigned long);
98static void igb_watchdog(unsigned long);
99static void igb_watchdog_task(struct work_struct *);
100static int igb_xmit_frame_ring_adv(struct sk_buff *, struct net_device *,
101 struct igb_ring *);
102static int igb_xmit_frame_adv(struct sk_buff *skb, struct net_device *);
103static struct net_device_stats *igb_get_stats(struct net_device *);
104static int igb_change_mtu(struct net_device *, int);
105static int igb_set_mac(struct net_device *, void *);
106static irqreturn_t igb_intr(int irq, void *);
107static irqreturn_t igb_intr_msi(int irq, void *);
108static irqreturn_t igb_msix_other(int irq, void *);
109static irqreturn_t igb_msix_rx(int irq, void *);
110static irqreturn_t igb_msix_tx(int irq, void *);
421e02f0 111#ifdef CONFIG_IGB_DCA
fe4506b6
JC		 112static void igb_update_rx_dca(struct igb_ring *);
113static void igb_update_tx_dca(struct igb_ring *);
114static void igb_setup_dca(struct igb_adapter *);
421e02f0 115#endif /* CONFIG_IGB_DCA */
3b644cf6 116static bool igb_clean_tx_irq(struct igb_ring *);
661086df 117static int igb_poll(struct napi_struct *, int);
3b644cf6
MW		 118static bool igb_clean_rx_irq_adv(struct igb_ring *, int *, int);
119static void igb_alloc_rx_buffers_adv(struct igb_ring *, int);
9d5c8243
AK		 120static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
121static void igb_tx_timeout(struct net_device *);
122static void igb_reset_task(struct work_struct *);
123static void igb_vlan_rx_register(struct net_device *, struct vlan_group *);
124static void igb_vlan_rx_add_vid(struct net_device *, u16);
125static void igb_vlan_rx_kill_vid(struct net_device *, u16);
126static void igb_restore_vlan(struct igb_adapter *);
4ae196df
AD		 127static void igb_ping_all_vfs(struct igb_adapter *);
128static void igb_msg_task(struct igb_adapter *);
129static int igb_rcv_msg_from_vf(struct igb_adapter *, u32);
e1739522
AD		 130static inline void igb_set_rah_pool(struct e1000_hw *, int , int);
131static void igb_set_mc_list_pools(struct igb_adapter *, int, u16);
4ae196df 132static void igb_vmm_control(struct igb_adapter *);
e1739522 133static inline void igb_set_vmolr(struct e1000_hw *, int);
4ae196df
AD		 134static inline int igb_set_vf_rlpml(struct igb_adapter *, int, int);
135static int igb_set_vf_mac(struct igb_adapter *adapter, int, unsigned char *);
136static void igb_restore_vf_multicasts(struct igb_adapter *adapter);
9d5c8243
AK		 137
138static int igb_suspend(struct pci_dev *, pm_message_t);
139#ifdef CONFIG_PM
140static int igb_resume(struct pci_dev *);
141#endif
142static void igb_shutdown(struct pci_dev *);
421e02f0 143#ifdef CONFIG_IGB_DCA
fe4506b6
JC		 144static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
145static struct notifier_block dca_notifier = {
146 .notifier_call = igb_notify_dca,
147 .next = NULL,
148 .priority = 0
149};
150#endif
9d5c8243
AK		 151#ifdef CONFIG_NET_POLL_CONTROLLER
152/* for netdump / net console */
153static void igb_netpoll(struct net_device *);
154#endif
155
37680117
AD		 156#ifdef CONFIG_PCI_IOV
157static ssize_t igb_set_num_vfs(struct device *, struct device_attribute *,
158 const char *, size_t);
159static ssize_t igb_show_num_vfs(struct device *, struct device_attribute *,
160 char *);
161DEVICE_ATTR(num_vfs, S_IRUGO | S_IWUSR, igb_show_num_vfs, igb_set_num_vfs);
162#endif
9d5c8243
AK		 163static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
164 pci_channel_state_t);
165static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
166static void igb_io_resume(struct pci_dev *);
167
168static struct pci_error_handlers igb_err_handler = {
169 .error_detected = igb_io_error_detected,
170 .slot_reset = igb_io_slot_reset,
171 .resume = igb_io_resume,
172};
173
174
175static struct pci_driver igb_driver = {
176 .name = igb_driver_name,
177 .id_table = igb_pci_tbl,
178 .probe = igb_probe,
179 .remove = __devexit_p(igb_remove),
180#ifdef CONFIG_PM
181 /* Power Managment Hooks */
182 .suspend = igb_suspend,
183 .resume = igb_resume,
184#endif
185 .shutdown = igb_shutdown,
186 .err_handler = &igb_err_handler
187};
188
7dfc16fa
AD		 189static int global_quad_port_a; /* global quad port a indication */
190
9d5c8243
AK		 191MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
192MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
193MODULE_LICENSE("GPL");
194MODULE_VERSION(DRV_VERSION);
195
38c845c7
PO		 196/**
197 * Scale the NIC clock cycle by a large factor so that
198 * relatively small clock corrections can be added or
199 * substracted at each clock tick. The drawbacks of a
200 * large factor are a) that the clock register overflows
201 * more quickly (not such a big deal) and b) that the
202 * increment per tick has to fit into 24 bits.
203 *
204 * Note that
205 * TIMINCA = IGB_TSYNC_CYCLE_TIME_IN_NANOSECONDS *
206 * IGB_TSYNC_SCALE
207 * TIMINCA += TIMINCA * adjustment [ppm] / 1e9
208 *
209 * The base scale factor is intentionally a power of two
210 * so that the division in %struct timecounter can be done with
211 * a shift.
212 */
213#define IGB_TSYNC_SHIFT (19)
214#define IGB_TSYNC_SCALE (1<<IGB_TSYNC_SHIFT)
215
216/**
217 * The duration of one clock cycle of the NIC.
218 *
219 * @todo This hard-coded value is part of the specification and might change
220 * in future hardware revisions. Add revision check.
221 */
222#define IGB_TSYNC_CYCLE_TIME_IN_NANOSECONDS 16
223
224#if (IGB_TSYNC_SCALE * IGB_TSYNC_CYCLE_TIME_IN_NANOSECONDS) >= (1<<24)
225# error IGB_TSYNC_SCALE and/or IGB_TSYNC_CYCLE_TIME_IN_NANOSECONDS are too large to fit into TIMINCA
226#endif
227
228/**
229 * igb_read_clock - read raw cycle counter (to be used by time counter)
230 */
231static cycle_t igb_read_clock(const struct cyclecounter *tc)
232{
233 struct igb_adapter *adapter =
234 container_of(tc, struct igb_adapter, cycles);
235 struct e1000_hw *hw = &adapter->hw;
236 u64 stamp;
237
238 stamp = rd32(E1000_SYSTIML);
239 stamp |= (u64)rd32(E1000_SYSTIMH) << 32ULL;
240
241 return stamp;
242}
243
9d5c8243
AK		 244#ifdef DEBUG
245/**
246 * igb_get_hw_dev_name - return device name string
247 * used by hardware layer to print debugging information
248 **/
249char *igb_get_hw_dev_name(struct e1000_hw *hw)
250{
251 struct igb_adapter *adapter = hw->back;
252 return adapter->netdev->name;
253}
38c845c7
PO		 254
255/**
256 * igb_get_time_str - format current NIC and system time as string
257 */
258static char *igb_get_time_str(struct igb_adapter *adapter,
259 char buffer[160])
260{
261 cycle_t hw = adapter->cycles.read(&adapter->cycles);
262 struct timespec nic = ns_to_timespec(timecounter_read(&adapter->clock));
263 struct timespec sys;
264 struct timespec delta;
265 getnstimeofday(&sys);
266
267 delta = timespec_sub(nic, sys);
268
269 sprintf(buffer,
33af6bcc
PO		 270 "HW %llu, NIC %ld.%09lus, SYS %ld.%09lus, NIC-SYS %lds + %09luns",
271 hw,
38c845c7
PO		 272 (long)nic.tv_sec, nic.tv_nsec,
273 (long)sys.tv_sec, sys.tv_nsec,
274 (long)delta.tv_sec, delta.tv_nsec);
275
276 return buffer;
277}
9d5c8243
AK		 278#endif
279
c493ea45
AD		 280/**
281 * igb_desc_unused - calculate if we have unused descriptors
282 **/
283static int igb_desc_unused(struct igb_ring *ring)
284{
285 if (ring->next_to_clean > ring->next_to_use)
286 return ring->next_to_clean - ring->next_to_use - 1;
287
288 return ring->count + ring->next_to_clean - ring->next_to_use - 1;
289}
290
9d5c8243
AK		 291/**
292 * igb_init_module - Driver Registration Routine
293 *
294 * igb_init_module is the first routine called when the driver is
295 * loaded. All it does is register with the PCI subsystem.
296 **/
297static int __init igb_init_module(void)
298{
299 int ret;
300 printk(KERN_INFO "%s - version %s\n",
301 igb_driver_string, igb_driver_version);
302
303 printk(KERN_INFO "%s\n", igb_copyright);
304
7dfc16fa
AD		 305 global_quad_port_a = 0;
306
421e02f0 307#ifdef CONFIG_IGB_DCA
fe4506b6
JC		 308 dca_register_notify(&dca_notifier);
309#endif
bbd98fe4
AD		 310
311 ret = pci_register_driver(&igb_driver);
9d5c8243
AK		 312 return ret;
313}
314
315module_init(igb_init_module);
316
317/**
318 * igb_exit_module - Driver Exit Cleanup Routine
319 *
320 * igb_exit_module is called just before the driver is removed
321 * from memory.
322 **/
323static void __exit igb_exit_module(void)
324{
421e02f0 325#ifdef CONFIG_IGB_DCA
fe4506b6
JC		 326 dca_unregister_notify(&dca_notifier);
327#endif
9d5c8243
AK		 328 pci_unregister_driver(&igb_driver);
329}
330
331module_exit(igb_exit_module);
332
26bc19ec
AD		 333#define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
334/**
335 * igb_cache_ring_register - Descriptor ring to register mapping
336 * @adapter: board private structure to initialize
337 *
338 * Once we know the feature-set enabled for the device, we'll cache
339 * the register offset the descriptor ring is assigned to.
340 **/
341static void igb_cache_ring_register(struct igb_adapter *adapter)
342{
343 int i;
1bfaf07b 344 unsigned int rbase_offset = adapter->vfs_allocated_count;
26bc19ec
AD		 345
346 switch (adapter->hw.mac.type) {
347 case e1000_82576:
348 /* The queues are allocated for virtualization such that VF 0
349 * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
350 * In order to avoid collision we start at the first free queue
351 * and continue consuming queues in the same sequence
352 */
353 for (i = 0; i < adapter->num_rx_queues; i++)
1bfaf07b
AD		 354 adapter->rx_ring[i].reg_idx = rbase_offset +
355 Q_IDX_82576(i);
26bc19ec 356 for (i = 0; i < adapter->num_tx_queues; i++)
1bfaf07b
AD		 357 adapter->tx_ring[i].reg_idx = rbase_offset +
358 Q_IDX_82576(i);
26bc19ec
AD		 359 break;
360 case e1000_82575:
361 default:
362 for (i = 0; i < adapter->num_rx_queues; i++)
363 adapter->rx_ring[i].reg_idx = i;
364 for (i = 0; i < adapter->num_tx_queues; i++)
365 adapter->tx_ring[i].reg_idx = i;
366 break;
367 }
368}
369
9d5c8243
AK		 370/**
371 * igb_alloc_queues - Allocate memory for all rings
372 * @adapter: board private structure to initialize
373 *
374 * We allocate one ring per queue at run-time since we don't know the
375 * number of queues at compile-time.
376 **/
377static int igb_alloc_queues(struct igb_adapter *adapter)
378{
379 int i;
380
381 adapter->tx_ring = kcalloc(adapter->num_tx_queues,
382 sizeof(struct igb_ring), GFP_KERNEL);
383 if (!adapter->tx_ring)
384 return -ENOMEM;
385
386 adapter->rx_ring = kcalloc(adapter->num_rx_queues,
387 sizeof(struct igb_ring), GFP_KERNEL);
388 if (!adapter->rx_ring) {
389 kfree(adapter->tx_ring);
390 return -ENOMEM;
391 }
392
6eb5a7f1
AD		 393 adapter->rx_ring->buddy = adapter->tx_ring;
394
661086df
PWJ		 395 for (i = 0; i < adapter->num_tx_queues; i++) {
396 struct igb_ring *ring = &(adapter->tx_ring[i]);
68fd9910 397 ring->count = adapter->tx_ring_count;
661086df
PWJ		 398 ring->adapter = adapter;
399 ring->queue_index = i;
400 }
9d5c8243
AK		 401 for (i = 0; i < adapter->num_rx_queues; i++) {
402 struct igb_ring *ring = &(adapter->rx_ring[i]);
68fd9910 403 ring->count = adapter->rx_ring_count;
9d5c8243 404 ring->adapter = adapter;
844290e5 405 ring->queue_index = i;
9d5c8243
AK		 406 ring->itr_register = E1000_ITR;
407
844290e5 408 /* set a default napi handler for each rx_ring */
661086df 409 netif_napi_add(adapter->netdev, &ring->napi, igb_poll, 64);
9d5c8243 410 }
26bc19ec
AD		 411
412 igb_cache_ring_register(adapter);
9d5c8243
AK		 413 return 0;
414}
415
a88f10ec
AD		 416static void igb_free_queues(struct igb_adapter *adapter)
417{
418 int i;
419
420 for (i = 0; i < adapter->num_rx_queues; i++)
421 netif_napi_del(&adapter->rx_ring[i].napi);
422
423 kfree(adapter->tx_ring);
424 kfree(adapter->rx_ring);
425}
426
9d5c8243
AK		 427#define IGB_N0_QUEUE -1
428static void igb_assign_vector(struct igb_adapter *adapter, int rx_queue,
429 int tx_queue, int msix_vector)
430{
431 u32 msixbm = 0;
432 struct e1000_hw *hw = &adapter->hw;
2d064c06
AD		 433 u32 ivar, index;
434
435 switch (hw->mac.type) {
436 case e1000_82575:
9d5c8243
AK		 437 /* The 82575 assigns vectors using a bitmask, which matches the
438 bitmask for the EICR/EIMS/EIMC registers. To assign one
439 or more queues to a vector, we write the appropriate bits
440 into the MSIXBM register for that vector. */
441 if (rx_queue > IGB_N0_QUEUE) {
442 msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
443 adapter->rx_ring[rx_queue].eims_value = msixbm;
444 }
445 if (tx_queue > IGB_N0_QUEUE) {
446 msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
447 adapter->tx_ring[tx_queue].eims_value =
448 E1000_EICR_TX_QUEUE0 << tx_queue;
449 }
450 array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
2d064c06
AD		 451 break;
452 case e1000_82576:
26bc19ec 453 /* 82576 uses a table-based method for assigning vectors.
2d064c06
AD		 454 Each queue has a single entry in the table to which we write
455 a vector number along with a "valid" bit. Sadly, the layout
456 of the table is somewhat counterintuitive. */
457 if (rx_queue > IGB_N0_QUEUE) {
1bfaf07b 458 index = (rx_queue >> 1) + adapter->vfs_allocated_count;
2d064c06 459 ivar = array_rd32(E1000_IVAR0, index);
26bc19ec 460 if (rx_queue & 0x1) {
2d064c06
AD		 461 /* vector goes into third byte of register */
462 ivar = ivar & 0xFF00FFFF;
463 ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
26bc19ec
AD		 464 } else {
465 /* vector goes into low byte of register */
466 ivar = ivar & 0xFFFFFF00;
467 ivar |= msix_vector | E1000_IVAR_VALID;
2d064c06
AD		 468 }
469 adapter->rx_ring[rx_queue].eims_value= 1 << msix_vector;
470 array_wr32(E1000_IVAR0, index, ivar);
471 }
472 if (tx_queue > IGB_N0_QUEUE) {
1bfaf07b 473 index = (tx_queue >> 1) + adapter->vfs_allocated_count;
2d064c06 474 ivar = array_rd32(E1000_IVAR0, index);
26bc19ec 475 if (tx_queue & 0x1) {
2d064c06
AD		 476 /* vector goes into high byte of register */
477 ivar = ivar & 0x00FFFFFF;
478 ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
26bc19ec
AD		 479 } else {
480 /* vector goes into second byte of register */
481 ivar = ivar & 0xFFFF00FF;
482 ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
2d064c06
AD		 483 }
484 adapter->tx_ring[tx_queue].eims_value= 1 << msix_vector;
485 array_wr32(E1000_IVAR0, index, ivar);
486 }
487 break;
488 default:
489 BUG();
490 break;
491 }
9d5c8243
AK		 492}
493
494/**
495 * igb_configure_msix - Configure MSI-X hardware
496 *
497 * igb_configure_msix sets up the hardware to properly
498 * generate MSI-X interrupts.
499 **/
500static void igb_configure_msix(struct igb_adapter *adapter)
501{
502 u32 tmp;
503 int i, vector = 0;
504 struct e1000_hw *hw = &adapter->hw;
505
506 adapter->eims_enable_mask = 0;
2d064c06
AD		 507 if (hw->mac.type == e1000_82576)
508 /* Turn on MSI-X capability first, or our settings
509 * won't stick. And it will take days to debug. */
510 wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
eebbbdba 511 E1000_GPIE_PBA | E1000_GPIE_EIAME |
2d064c06 512 E1000_GPIE_NSICR);
9d5c8243
AK		 513
514 for (i = 0; i < adapter->num_tx_queues; i++) {
515 struct igb_ring *tx_ring = &adapter->tx_ring[i];
516 igb_assign_vector(adapter, IGB_N0_QUEUE, i, vector++);
517 adapter->eims_enable_mask |= tx_ring->eims_value;
518 if (tx_ring->itr_val)
6eb5a7f1 519 writel(tx_ring->itr_val,
9d5c8243
AK		 520 hw->hw_addr + tx_ring->itr_register);
521 else
522 writel(1, hw->hw_addr + tx_ring->itr_register);
523 }
524
525 for (i = 0; i < adapter->num_rx_queues; i++) {
526 struct igb_ring *rx_ring = &adapter->rx_ring[i];
25ac3c24 527 rx_ring->buddy = NULL;
9d5c8243
AK		 528 igb_assign_vector(adapter, i, IGB_N0_QUEUE, vector++);
529 adapter->eims_enable_mask |= rx_ring->eims_value;
530 if (rx_ring->itr_val)
6eb5a7f1 531 writel(rx_ring->itr_val,
9d5c8243
AK		 532 hw->hw_addr + rx_ring->itr_register);
533 else
534 writel(1, hw->hw_addr + rx_ring->itr_register);
535 }
536
537
538 /* set vector for other causes, i.e. link changes */
2d064c06
AD		 539 switch (hw->mac.type) {
540 case e1000_82575:
9d5c8243
AK		 541 array_wr32(E1000_MSIXBM(0), vector++,
542 E1000_EIMS_OTHER);
543
9d5c8243
AK		 544 tmp = rd32(E1000_CTRL_EXT);
545 /* enable MSI-X PBA support*/
546 tmp |= E1000_CTRL_EXT_PBA_CLR;
547
548 /* Auto-Mask interrupts upon ICR read. */
549 tmp |= E1000_CTRL_EXT_EIAME;
550 tmp |= E1000_CTRL_EXT_IRCA;
551
552 wr32(E1000_CTRL_EXT, tmp);
553 adapter->eims_enable_mask |= E1000_EIMS_OTHER;
844290e5 554 adapter->eims_other = E1000_EIMS_OTHER;
9d5c8243 555
2d064c06
AD		 556 break;
557
558 case e1000_82576:
559 tmp = (vector++ | E1000_IVAR_VALID) << 8;
560 wr32(E1000_IVAR_MISC, tmp);
561
562 adapter->eims_enable_mask = (1 << (vector)) - 1;
563 adapter->eims_other = 1 << (vector - 1);
564 break;
565 default:
566 /* do nothing, since nothing else supports MSI-X */
567 break;
568 } /* switch (hw->mac.type) */
9d5c8243
AK		 569 wrfl();
570}
571
572/**
573 * igb_request_msix - Initialize MSI-X interrupts
574 *
575 * igb_request_msix allocates MSI-X vectors and requests interrupts from the
576 * kernel.
577 **/
578static int igb_request_msix(struct igb_adapter *adapter)
579{
580 struct net_device *netdev = adapter->netdev;
581 int i, err = 0, vector = 0;
582
583 vector = 0;
584
585 for (i = 0; i < adapter->num_tx_queues; i++) {
586 struct igb_ring *ring = &(adapter->tx_ring[i]);
cb7b48f6 587 sprintf(ring->name, "%s-tx-%d", netdev->name, i);
9d5c8243
AK		 588 err = request_irq(adapter->msix_entries[vector].vector,
589 &igb_msix_tx, 0, ring->name,
590 &(adapter->tx_ring[i]));
591 if (err)
592 goto out;
593 ring->itr_register = E1000_EITR(0) + (vector << 2);
6eb5a7f1 594 ring->itr_val = 976; /* ~4000 ints/sec */
9d5c8243
AK		 595 vector++;
596 }
597 for (i = 0; i < adapter->num_rx_queues; i++) {
598 struct igb_ring *ring = &(adapter->rx_ring[i]);
599 if (strlen(netdev->name) < (IFNAMSIZ - 5))
cb7b48f6 600 sprintf(ring->name, "%s-rx-%d", netdev->name, i);
9d5c8243
AK		 601 else
602 memcpy(ring->name, netdev->name, IFNAMSIZ);
603 err = request_irq(adapter->msix_entries[vector].vector,
604 &igb_msix_rx, 0, ring->name,
605 &(adapter->rx_ring[i]));
606 if (err)
607 goto out;
608 ring->itr_register = E1000_EITR(0) + (vector << 2);
609 ring->itr_val = adapter->itr;
610 vector++;
611 }
612
613 err = request_irq(adapter->msix_entries[vector].vector,
614 &igb_msix_other, 0, netdev->name, netdev);
615 if (err)
616 goto out;
617
9d5c8243
AK		 618 igb_configure_msix(adapter);
619 return 0;
620out:
621 return err;
622}
623
624static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
625{
626 if (adapter->msix_entries) {
627 pci_disable_msix(adapter->pdev);
628 kfree(adapter->msix_entries);
629 adapter->msix_entries = NULL;
7dfc16fa 630 } else if (adapter->flags & IGB_FLAG_HAS_MSI)
9d5c8243
AK		 631 pci_disable_msi(adapter->pdev);
632 return;
633}
634
635
636/**
637 * igb_set_interrupt_capability - set MSI or MSI-X if supported
638 *
639 * Attempt to configure interrupts using the best available
640 * capabilities of the hardware and kernel.
641 **/
642static void igb_set_interrupt_capability(struct igb_adapter *adapter)
643{
644 int err;
645 int numvecs, i;
646
83b7180d
AD		 647 /* Number of supported queues. */
648 /* Having more queues than CPUs doesn't make sense. */
649 adapter->num_rx_queues = min_t(u32, IGB_MAX_RX_QUEUES, num_online_cpus());
650 adapter->num_tx_queues = min_t(u32, IGB_MAX_TX_QUEUES, num_online_cpus());
651
9d5c8243
AK		 652 numvecs = adapter->num_tx_queues + adapter->num_rx_queues + 1;
653 adapter->msix_entries = kcalloc(numvecs, sizeof(struct msix_entry),
654 GFP_KERNEL);
655 if (!adapter->msix_entries)
656 goto msi_only;
657
658 for (i = 0; i < numvecs; i++)
659 adapter->msix_entries[i].entry = i;
660
661 err = pci_enable_msix(adapter->pdev,
662 adapter->msix_entries,
663 numvecs);
664 if (err == 0)
34a20e89 665 goto out;
9d5c8243
AK		 666
667 igb_reset_interrupt_capability(adapter);
668
669 /* If we can't do MSI-X, try MSI */
670msi_only:
671 adapter->num_rx_queues = 1;
661086df 672 adapter->num_tx_queues = 1;
9d5c8243 673 if (!pci_enable_msi(adapter->pdev))
7dfc16fa 674 adapter->flags |= IGB_FLAG_HAS_MSI;
34a20e89 675out:
661086df 676 /* Notify the stack of the (possibly) reduced Tx Queue count. */
fd2ea0a7 677 adapter->netdev->real_num_tx_queues = adapter->num_tx_queues;
9d5c8243
AK		 678 return;
679}
680
681/**
682 * igb_request_irq - initialize interrupts
683 *
684 * Attempts to configure interrupts using the best available
685 * capabilities of the hardware and kernel.
686 **/
687static int igb_request_irq(struct igb_adapter *adapter)
688{
689 struct net_device *netdev = adapter->netdev;
690 struct e1000_hw *hw = &adapter->hw;
691 int err = 0;
692
693 if (adapter->msix_entries) {
694 err = igb_request_msix(adapter);
844290e5 695 if (!err)
9d5c8243 696 goto request_done;
9d5c8243
AK		 697 /* fall back to MSI */
698 igb_reset_interrupt_capability(adapter);
699 if (!pci_enable_msi(adapter->pdev))
7dfc16fa 700 adapter->flags |= IGB_FLAG_HAS_MSI;
9d5c8243
AK		 701 igb_free_all_tx_resources(adapter);
702 igb_free_all_rx_resources(adapter);
703 adapter->num_rx_queues = 1;
704 igb_alloc_queues(adapter);
844290e5 705 } else {
2d064c06
AD		 706 switch (hw->mac.type) {
707 case e1000_82575:
708 wr32(E1000_MSIXBM(0),
709 (E1000_EICR_RX_QUEUE0 | E1000_EIMS_OTHER));
710 break;
711 case e1000_82576:
712 wr32(E1000_IVAR0, E1000_IVAR_VALID);
713 break;
714 default:
715 break;
716 }
9d5c8243 717 }
844290e5 718
7dfc16fa 719 if (adapter->flags & IGB_FLAG_HAS_MSI) {
9d5c8243
AK		 720 err = request_irq(adapter->pdev->irq, &igb_intr_msi, 0,
721 netdev->name, netdev);
722 if (!err)
723 goto request_done;
724 /* fall back to legacy interrupts */
725 igb_reset_interrupt_capability(adapter);
7dfc16fa 726 adapter->flags &= ~IGB_FLAG_HAS_MSI;
9d5c8243
AK		 727 }
728
729 err = request_irq(adapter->pdev->irq, &igb_intr, IRQF_SHARED,
730 netdev->name, netdev);
731
6cb5e577 732 if (err)
9d5c8243
AK		 733 dev_err(&adapter->pdev->dev, "Error %d getting interrupt\n",
734 err);
9d5c8243
AK		 735
736request_done:
737 return err;
738}
739
740static void igb_free_irq(struct igb_adapter *adapter)
741{
742 struct net_device *netdev = adapter->netdev;
743
744 if (adapter->msix_entries) {
745 int vector = 0, i;
746
747 for (i = 0; i < adapter->num_tx_queues; i++)
748 free_irq(adapter->msix_entries[vector++].vector,
749 &(adapter->tx_ring[i]));
750 for (i = 0; i < adapter->num_rx_queues; i++)
751 free_irq(adapter->msix_entries[vector++].vector,
752 &(adapter->rx_ring[i]));
753
754 free_irq(adapter->msix_entries[vector++].vector, netdev);
755 return;
756 }
757
758 free_irq(adapter->pdev->irq, netdev);
759}
760
761/**
762 * igb_irq_disable - Mask off interrupt generation on the NIC
763 * @adapter: board private structure
764 **/
765static void igb_irq_disable(struct igb_adapter *adapter)
766{
767 struct e1000_hw *hw = &adapter->hw;
768
769 if (adapter->msix_entries) {
844290e5 770 wr32(E1000_EIAM, 0);
9d5c8243
AK		 771 wr32(E1000_EIMC, ~0);
772 wr32(E1000_EIAC, 0);
773 }
844290e5
PW		 774
775 wr32(E1000_IAM, 0);
9d5c8243
AK		 776 wr32(E1000_IMC, ~0);
777 wrfl();
778 synchronize_irq(adapter->pdev->irq);
779}
780
781/**
782 * igb_irq_enable - Enable default interrupt generation settings
783 * @adapter: board private structure
784 **/
785static void igb_irq_enable(struct igb_adapter *adapter)
786{
787 struct e1000_hw *hw = &adapter->hw;
788
789 if (adapter->msix_entries) {
844290e5
PW		 790 wr32(E1000_EIAC, adapter->eims_enable_mask);
791 wr32(E1000_EIAM, adapter->eims_enable_mask);
792 wr32(E1000_EIMS, adapter->eims_enable_mask);
4ae196df
AD		 793 if (adapter->vfs_allocated_count)
794 wr32(E1000_MBVFIMR, 0xFF);
795 wr32(E1000_IMS, (E1000_IMS_LSC | E1000_IMS_VMMB |
796 E1000_IMS_DOUTSYNC));
844290e5
PW		 797 } else {
798 wr32(E1000_IMS, IMS_ENABLE_MASK);
799 wr32(E1000_IAM, IMS_ENABLE_MASK);
800 }
9d5c8243
AK		 801}
802
803static void igb_update_mng_vlan(struct igb_adapter *adapter)
804{
805 struct net_device *netdev = adapter->netdev;
806 u16 vid = adapter->hw.mng_cookie.vlan_id;
807 u16 old_vid = adapter->mng_vlan_id;
808 if (adapter->vlgrp) {
809 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
810 if (adapter->hw.mng_cookie.status &
811 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
812 igb_vlan_rx_add_vid(netdev, vid);
813 adapter->mng_vlan_id = vid;
814 } else
815 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
816
817 if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
818 (vid != old_vid) &&
819 !vlan_group_get_device(adapter->vlgrp, old_vid))
820 igb_vlan_rx_kill_vid(netdev, old_vid);
821 } else
822 adapter->mng_vlan_id = vid;
823 }
824}
825
826/**
827 * igb_release_hw_control - release control of the h/w to f/w
828 * @adapter: address of board private structure
829 *
830 * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
831 * For ASF and Pass Through versions of f/w this means that the
832 * driver is no longer loaded.
833 *
834 **/
835static void igb_release_hw_control(struct igb_adapter *adapter)
836{
837 struct e1000_hw *hw = &adapter->hw;
838 u32 ctrl_ext;
839
840 /* Let firmware take over control of h/w */
841 ctrl_ext = rd32(E1000_CTRL_EXT);
842 wr32(E1000_CTRL_EXT,
843 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
844}
845
846
847/**
848 * igb_get_hw_control - get control of the h/w from f/w
849 * @adapter: address of board private structure
850 *
851 * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
852 * For ASF and Pass Through versions of f/w this means that
853 * the driver is loaded.
854 *
855 **/
856static void igb_get_hw_control(struct igb_adapter *adapter)
857{
858 struct e1000_hw *hw = &adapter->hw;
859 u32 ctrl_ext;
860
861 /* Let firmware know the driver has taken over */
862 ctrl_ext = rd32(E1000_CTRL_EXT);
863 wr32(E1000_CTRL_EXT,
864 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
865}
866
9d5c8243
AK		 867/**
868 * igb_configure - configure the hardware for RX and TX
869 * @adapter: private board structure
870 **/
871static void igb_configure(struct igb_adapter *adapter)
872{
873 struct net_device *netdev = adapter->netdev;
874 int i;
875
876 igb_get_hw_control(adapter);
877 igb_set_multi(netdev);
878
879 igb_restore_vlan(adapter);
9d5c8243
AK		 880
881 igb_configure_tx(adapter);
882 igb_setup_rctl(adapter);
883 igb_configure_rx(adapter);
662d7205
AD		 884
885 igb_rx_fifo_flush_82575(&adapter->hw);
886
c493ea45 887 /* call igb_desc_unused which always leaves
9d5c8243
AK		 888 * at least 1 descriptor unused to make sure
889 * next_to_use != next_to_clean */
890 for (i = 0; i < adapter->num_rx_queues; i++) {
891 struct igb_ring *ring = &adapter->rx_ring[i];
c493ea45 892 igb_alloc_rx_buffers_adv(ring, igb_desc_unused(ring));
9d5c8243
AK		 893 }
894
895
896 adapter->tx_queue_len = netdev->tx_queue_len;
897}
898
899
900/**
901 * igb_up - Open the interface and prepare it to handle traffic
902 * @adapter: board private structure
903 **/
904
905int igb_up(struct igb_adapter *adapter)
906{
907 struct e1000_hw *hw = &adapter->hw;
908 int i;
909
910 /* hardware has been reset, we need to reload some things */
911 igb_configure(adapter);
912
913 clear_bit(__IGB_DOWN, &adapter->state);
914
844290e5
PW		 915 for (i = 0; i < adapter->num_rx_queues; i++)
916 napi_enable(&adapter->rx_ring[i].napi);
917 if (adapter->msix_entries)
9d5c8243 918 igb_configure_msix(adapter);
9d5c8243 919
4ae196df 920 igb_vmm_control(adapter);
e1739522
AD		 921 igb_set_rah_pool(hw, adapter->vfs_allocated_count, 0);
922 igb_set_vmolr(hw, adapter->vfs_allocated_count);
923
9d5c8243
AK		 924 /* Clear any pending interrupts. */
925 rd32(E1000_ICR);
926 igb_irq_enable(adapter);
927
928 /* Fire a link change interrupt to start the watchdog. */
929 wr32(E1000_ICS, E1000_ICS_LSC);
930 return 0;
931}
932
933void igb_down(struct igb_adapter *adapter)
934{
935 struct e1000_hw *hw = &adapter->hw;
936 struct net_device *netdev = adapter->netdev;
937 u32 tctl, rctl;
938 int i;
939
940 /* signal that we're down so the interrupt handler does not
941 * reschedule our watchdog timer */
942 set_bit(__IGB_DOWN, &adapter->state);
943
944 /* disable receives in the hardware */
945 rctl = rd32(E1000_RCTL);
946 wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
947 /* flush and sleep below */
948
fd2ea0a7 949 netif_tx_stop_all_queues(netdev);
9d5c8243
AK		 950
951 /* disable transmits in the hardware */
952 tctl = rd32(E1000_TCTL);
953 tctl &= ~E1000_TCTL_EN;
954 wr32(E1000_TCTL, tctl);
955 /* flush both disables and wait for them to finish */
956 wrfl();
957 msleep(10);
958
844290e5
PW		 959 for (i = 0; i < adapter->num_rx_queues; i++)
960 napi_disable(&adapter->rx_ring[i].napi);
9d5c8243 961
9d5c8243
AK		 962 igb_irq_disable(adapter);
963
964 del_timer_sync(&adapter->watchdog_timer);
965 del_timer_sync(&adapter->phy_info_timer);
966
967 netdev->tx_queue_len = adapter->tx_queue_len;
968 netif_carrier_off(netdev);
04fe6358
AD		 969
970 /* record the stats before reset*/
971 igb_update_stats(adapter);
972
9d5c8243
AK		 973 adapter->link_speed = 0;
974 adapter->link_duplex = 0;
975
3023682e
JK		 976 if (!pci_channel_offline(adapter->pdev))
977 igb_reset(adapter);
9d5c8243
AK		 978 igb_clean_all_tx_rings(adapter);
979 igb_clean_all_rx_rings(adapter);
980}
981
982void igb_reinit_locked(struct igb_adapter *adapter)
983{
984 WARN_ON(in_interrupt());
985 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
986 msleep(1);
987 igb_down(adapter);
988 igb_up(adapter);
989 clear_bit(__IGB_RESETTING, &adapter->state);
990}
991
992void igb_reset(struct igb_adapter *adapter)
993{
994 struct e1000_hw *hw = &adapter->hw;
2d064c06
AD		 995 struct e1000_mac_info *mac = &hw->mac;
996 struct e1000_fc_info *fc = &hw->fc;
9d5c8243
AK		 997 u32 pba = 0, tx_space, min_tx_space, min_rx_space;
998 u16 hwm;
999
1000 /* Repartition Pba for greater than 9k mtu
1001 * To take effect CTRL.RST is required.
1002 */
fa4dfae0
AD		 1003 switch (mac->type) {
1004 case e1000_82576:
2d064c06 1005 pba = E1000_PBA_64K;
fa4dfae0
AD		 1006 break;
1007 case e1000_82575:
1008 default:
1009 pba = E1000_PBA_34K;
1010 break;
2d064c06 1011 }
9d5c8243 1012
2d064c06
AD		 1013 if ((adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) &&
1014 (mac->type < e1000_82576)) {
9d5c8243
AK		 1015 /* adjust PBA for jumbo frames */
1016 wr32(E1000_PBA, pba);
1017
1018 /* To maintain wire speed transmits, the Tx FIFO should be
1019 * large enough to accommodate two full transmit packets,
1020 * rounded up to the next 1KB and expressed in KB. Likewise,
1021 * the Rx FIFO should be large enough to accommodate at least
1022 * one full receive packet and is similarly rounded up and
1023 * expressed in KB. */
1024 pba = rd32(E1000_PBA);
1025 /* upper 16 bits has Tx packet buffer allocation size in KB */
1026 tx_space = pba >> 16;
1027 /* lower 16 bits has Rx packet buffer allocation size in KB */
1028 pba &= 0xffff;
1029 /* the tx fifo also stores 16 bytes of information about the tx
1030 * but don't include ethernet FCS because hardware appends it */
1031 min_tx_space = (adapter->max_frame_size +
85e8d004 1032 sizeof(union e1000_adv_tx_desc) -
9d5c8243
AK		 1033 ETH_FCS_LEN) * 2;
1034 min_tx_space = ALIGN(min_tx_space, 1024);
1035 min_tx_space >>= 10;
1036 /* software strips receive CRC, so leave room for it */
1037 min_rx_space = adapter->max_frame_size;
1038 min_rx_space = ALIGN(min_rx_space, 1024);
1039 min_rx_space >>= 10;
1040
1041 /* If current Tx allocation is less than the min Tx FIFO size,
1042 * and the min Tx FIFO size is less than the current Rx FIFO
1043 * allocation, take space away from current Rx allocation */
1044 if (tx_space < min_tx_space &&
1045 ((min_tx_space - tx_space) < pba)) {
1046 pba = pba - (min_tx_space - tx_space);
1047
1048 /* if short on rx space, rx wins and must trump tx
1049 * adjustment */
1050 if (pba < min_rx_space)
1051 pba = min_rx_space;
1052 }
2d064c06 1053 wr32(E1000_PBA, pba);
9d5c8243 1054 }
9d5c8243
AK		 1055
1056 /* flow control settings */
1057 /* The high water mark must be low enough to fit one full frame
1058 * (or the size used for early receive) above it in the Rx FIFO.
1059 * Set it to the lower of:
1060 * - 90% of the Rx FIFO size, or
1061 * - the full Rx FIFO size minus one full frame */
1062 hwm = min(((pba << 10) * 9 / 10),
2d064c06 1063 ((pba << 10) - 2 * adapter->max_frame_size));
9d5c8243 1064
2d064c06
AD		 1065 if (mac->type < e1000_82576) {
1066 fc->high_water = hwm & 0xFFF8; /* 8-byte granularity */
1067 fc->low_water = fc->high_water - 8;
1068 } else {
1069 fc->high_water = hwm & 0xFFF0; /* 16-byte granularity */
1070 fc->low_water = fc->high_water - 16;
1071 }
9d5c8243
AK		 1072 fc->pause_time = 0xFFFF;
1073 fc->send_xon = 1;
1074 fc->type = fc->original_type;
1075
4ae196df
AD		 1076 /* disable receive for all VFs and wait one second */
1077 if (adapter->vfs_allocated_count) {
1078 int i;
1079 for (i = 0 ; i < adapter->vfs_allocated_count; i++)
1080 adapter->vf_data[i].clear_to_send = false;
1081
1082 /* ping all the active vfs to let them know we are going down */
1083 igb_ping_all_vfs(adapter);
1084
1085 /* disable transmits and receives */
1086 wr32(E1000_VFRE, 0);
1087 wr32(E1000_VFTE, 0);
1088 }
1089
9d5c8243
AK		 1090 /* Allow time for pending master requests to run */
1091 adapter->hw.mac.ops.reset_hw(&adapter->hw);
1092 wr32(E1000_WUC, 0);
1093
1094 if (adapter->hw.mac.ops.init_hw(&adapter->hw))
1095 dev_err(&adapter->pdev->dev, "Hardware Error\n");
1096
1097 igb_update_mng_vlan(adapter);
1098
1099 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
1100 wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
1101
1102 igb_reset_adaptive(&adapter->hw);
f5f4cf08 1103 igb_get_phy_info(&adapter->hw);
9d5c8243
AK		 1104}
1105
2e5c6922
SH		 1106static const struct net_device_ops igb_netdev_ops = {
1107 .ndo_open = igb_open,
1108 .ndo_stop = igb_close,
00829823 1109 .ndo_start_xmit = igb_xmit_frame_adv,
2e5c6922
SH		 1110 .ndo_get_stats = igb_get_stats,
1111 .ndo_set_multicast_list = igb_set_multi,
1112 .ndo_set_mac_address = igb_set_mac,
1113 .ndo_change_mtu = igb_change_mtu,
1114 .ndo_do_ioctl = igb_ioctl,
1115 .ndo_tx_timeout = igb_tx_timeout,
1116 .ndo_validate_addr = eth_validate_addr,
1117 .ndo_vlan_rx_register = igb_vlan_rx_register,
1118 .ndo_vlan_rx_add_vid = igb_vlan_rx_add_vid,
1119 .ndo_vlan_rx_kill_vid = igb_vlan_rx_kill_vid,
1120#ifdef CONFIG_NET_POLL_CONTROLLER
1121 .ndo_poll_controller = igb_netpoll,
1122#endif
1123};
1124
9d5c8243
AK		 1125/**
1126 * igb_probe - Device Initialization Routine
1127 * @pdev: PCI device information struct
1128 * @ent: entry in igb_pci_tbl
1129 *
1130 * Returns 0 on success, negative on failure
1131 *
1132 * igb_probe initializes an adapter identified by a pci_dev structure.
1133 * The OS initialization, configuring of the adapter private structure,
1134 * and a hardware reset occur.
1135 **/
1136static int __devinit igb_probe(struct pci_dev *pdev,
1137 const struct pci_device_id *ent)
1138{
1139 struct net_device *netdev;
1140 struct igb_adapter *adapter;
1141 struct e1000_hw *hw;
1142 const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
1143 unsigned long mmio_start, mmio_len;
2d6a5e95 1144 int err, pci_using_dac;
682337fe 1145 u16 eeprom_data = 0;
9d5c8243
AK		 1146 u16 eeprom_apme_mask = IGB_EEPROM_APME;
1147 u32 part_num;
1148
aed5dec3 1149 err = pci_enable_device_mem(pdev);
9d5c8243
AK		 1150 if (err)
1151 return err;
1152
1153 pci_using_dac = 0;
1154 err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
1155 if (!err) {
1156 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
1157 if (!err)
1158 pci_using_dac = 1;
1159 } else {
1160 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
1161 if (err) {
1162 err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
1163 if (err) {
1164 dev_err(&pdev->dev, "No usable DMA "
1165 "configuration, aborting\n");
1166 goto err_dma;
1167 }
1168 }
1169 }
1170
aed5dec3
AD		 1171 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
1172 IORESOURCE_MEM),
1173 igb_driver_name);
9d5c8243
AK		 1174 if (err)
1175 goto err_pci_reg;
1176
ea943d41
JK		 1177 err = pci_enable_pcie_error_reporting(pdev);
1178 if (err) {
1179 dev_err(&pdev->dev, "pci_enable_pcie_error_reporting failed "
1180 "0x%x\n", err);
1181 /* non-fatal, continue */
1182 }
40a914fa 1183
9d5c8243 1184 pci_set_master(pdev);
c682fc23 1185 pci_save_state(pdev);
9d5c8243
AK		 1186
1187 err = -ENOMEM;
1bfaf07b
AD		 1188 netdev = alloc_etherdev_mq(sizeof(struct igb_adapter),
1189 IGB_ABS_MAX_TX_QUEUES);
9d5c8243
AK		 1190 if (!netdev)
1191 goto err_alloc_etherdev;
1192
1193 SET_NETDEV_DEV(netdev, &pdev->dev);
1194
1195 pci_set_drvdata(pdev, netdev);
1196 adapter = netdev_priv(netdev);
1197 adapter->netdev = netdev;
1198 adapter->pdev = pdev;
1199 hw = &adapter->hw;
1200 hw->back = adapter;
1201 adapter->msg_enable = NETIF_MSG_DRV | NETIF_MSG_PROBE;
1202
1203 mmio_start = pci_resource_start(pdev, 0);
1204 mmio_len = pci_resource_len(pdev, 0);
1205
1206 err = -EIO;
28b0759c
AD		 1207 hw->hw_addr = ioremap(mmio_start, mmio_len);
1208 if (!hw->hw_addr)
9d5c8243
AK		 1209 goto err_ioremap;
1210
2e5c6922 1211 netdev->netdev_ops = &igb_netdev_ops;
9d5c8243 1212 igb_set_ethtool_ops(netdev);
9d5c8243 1213 netdev->watchdog_timeo = 5 * HZ;
9d5c8243
AK		 1214
1215 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
1216
1217 netdev->mem_start = mmio_start;
1218 netdev->mem_end = mmio_start + mmio_len;
1219
9d5c8243
AK		 1220 /* PCI config space info */
1221 hw->vendor_id = pdev->vendor;
1222 hw->device_id = pdev->device;
1223 hw->revision_id = pdev->revision;
1224 hw->subsystem_vendor_id = pdev->subsystem_vendor;
1225 hw->subsystem_device_id = pdev->subsystem_device;
1226
1227 /* setup the private structure */
1228 hw->back = adapter;
1229 /* Copy the default MAC, PHY and NVM function pointers */
1230 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
1231 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
1232 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
1233 /* Initialize skew-specific constants */
1234 err = ei->get_invariants(hw);
1235 if (err)
450c87c8 1236 goto err_sw_init;
9d5c8243 1237
450c87c8 1238 /* setup the private structure */
9d5c8243
AK		 1239 err = igb_sw_init(adapter);
1240 if (err)
1241 goto err_sw_init;
1242
1243 igb_get_bus_info_pcie(hw);
1244
7dfc16fa
AD		 1245 /* set flags */
1246 switch (hw->mac.type) {
7dfc16fa 1247 case e1000_82575:
7dfc16fa
AD		 1248 adapter->flags |= IGB_FLAG_NEED_CTX_IDX;
1249 break;
bbd98fe4 1250 case e1000_82576:
7dfc16fa
AD		 1251 default:
1252 break;
1253 }
1254
9d5c8243
AK		 1255 hw->phy.autoneg_wait_to_complete = false;
1256 hw->mac.adaptive_ifs = true;
1257
1258 /* Copper options */
1259 if (hw->phy.media_type == e1000_media_type_copper) {
1260 hw->phy.mdix = AUTO_ALL_MODES;
1261 hw->phy.disable_polarity_correction = false;
1262 hw->phy.ms_type = e1000_ms_hw_default;
1263 }
1264
1265 if (igb_check_reset_block(hw))
1266 dev_info(&pdev->dev,
1267 "PHY reset is blocked due to SOL/IDER session.\n");
1268
1269 netdev->features = NETIF_F_SG |
7d8eb29e 1270 NETIF_F_IP_CSUM |
9d5c8243
AK		 1271 NETIF_F_HW_VLAN_TX |
1272 NETIF_F_HW_VLAN_RX |
1273 NETIF_F_HW_VLAN_FILTER;
1274
7d8eb29e 1275 netdev->features |= NETIF_F_IPV6_CSUM;
9d5c8243 1276 netdev->features |= NETIF_F_TSO;
9d5c8243 1277 netdev->features |= NETIF_F_TSO6;
48f29ffc 1278
5c0999b7 1279 netdev->features |= NETIF_F_GRO;
d3352520 1280
48f29ffc
JK		 1281 netdev->vlan_features |= NETIF_F_TSO;
1282 netdev->vlan_features |= NETIF_F_TSO6;
7d8eb29e 1283 netdev->vlan_features |= NETIF_F_IP_CSUM;
48f29ffc
JK		 1284 netdev->vlan_features |= NETIF_F_SG;
1285
9d5c8243
AK		 1286 if (pci_using_dac)
1287 netdev->features |= NETIF_F_HIGHDMA;
1288
9d5c8243
AK		 1289 adapter->en_mng_pt = igb_enable_mng_pass_thru(&adapter->hw);
1290
1291 /* before reading the NVM, reset the controller to put the device in a
1292 * known good starting state */
1293 hw->mac.ops.reset_hw(hw);
1294
1295 /* make sure the NVM is good */
1296 if (igb_validate_nvm_checksum(hw) < 0) {
1297 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
1298 err = -EIO;
1299 goto err_eeprom;
1300 }
1301
1302 /* copy the MAC address out of the NVM */
1303 if (hw->mac.ops.read_mac_addr(hw))
1304 dev_err(&pdev->dev, "NVM Read Error\n");
1305
1306 memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
1307 memcpy(netdev->perm_addr, hw->mac.addr, netdev->addr_len);
1308
1309 if (!is_valid_ether_addr(netdev->perm_addr)) {
1310 dev_err(&pdev->dev, "Invalid MAC Address\n");
1311 err = -EIO;
1312 goto err_eeprom;
1313 }
1314
0e340485
AD		 1315 setup_timer(&adapter->watchdog_timer, &igb_watchdog,
1316 (unsigned long) adapter);
1317 setup_timer(&adapter->phy_info_timer, &igb_update_phy_info,
1318 (unsigned long) adapter);
9d5c8243
AK		 1319
1320 INIT_WORK(&adapter->reset_task, igb_reset_task);
1321 INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
1322
450c87c8 1323 /* Initialize link properties that are user-changeable */
9d5c8243
AK		 1324 adapter->fc_autoneg = true;
1325 hw->mac.autoneg = true;
1326 hw->phy.autoneg_advertised = 0x2f;
1327
1328 hw->fc.original_type = e1000_fc_default;
1329 hw->fc.type = e1000_fc_default;
1330
cbd347ad 1331 adapter->itr_setting = IGB_DEFAULT_ITR;
9d5c8243
AK		 1332 adapter->itr = IGB_START_ITR;
1333
1334 igb_validate_mdi_setting(hw);
1335
1336 adapter->rx_csum = 1;
1337
1338 /* Initial Wake on LAN setting If APM wake is enabled in the EEPROM,
1339 * enable the ACPI Magic Packet filter
1340 */
1341
a2cf8b6c 1342 if (hw->bus.func == 0)
312c75ae 1343 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
a2cf8b6c
AD		 1344 else if (hw->bus.func == 1)
1345 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
9d5c8243
AK		 1346
1347 if (eeprom_data & eeprom_apme_mask)
1348 adapter->eeprom_wol |= E1000_WUFC_MAG;
1349
1350 /* now that we have the eeprom settings, apply the special cases where
1351 * the eeprom may be wrong or the board simply won't support wake on
1352 * lan on a particular port */
1353 switch (pdev->device) {
1354 case E1000_DEV_ID_82575GB_QUAD_COPPER:
1355 adapter->eeprom_wol = 0;
1356 break;
1357 case E1000_DEV_ID_82575EB_FIBER_SERDES:
2d064c06
AD		 1358 case E1000_DEV_ID_82576_FIBER:
1359 case E1000_DEV_ID_82576_SERDES:
9d5c8243
AK		 1360 /* Wake events only supported on port A for dual fiber
1361 * regardless of eeprom setting */
1362 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
1363 adapter->eeprom_wol = 0;
1364 break;
c8ea5ea9
AD		 1365 case E1000_DEV_ID_82576_QUAD_COPPER:
1366 /* if quad port adapter, disable WoL on all but port A */
1367 if (global_quad_port_a != 0)
1368 adapter->eeprom_wol = 0;
1369 else
1370 adapter->flags |= IGB_FLAG_QUAD_PORT_A;
1371 /* Reset for multiple quad port adapters */
1372 if (++global_quad_port_a == 4)
1373 global_quad_port_a = 0;
1374 break;
9d5c8243
AK		 1375 }
1376
1377 /* initialize the wol settings based on the eeprom settings */
1378 adapter->wol = adapter->eeprom_wol;
e1b86d84 1379 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
9d5c8243
AK		 1380
1381 /* reset the hardware with the new settings */
1382 igb_reset(adapter);
1383
1384 /* let the f/w know that the h/w is now under the control of the
1385 * driver. */
1386 igb_get_hw_control(adapter);
1387
1388 /* tell the stack to leave us alone until igb_open() is called */
1389 netif_carrier_off(netdev);
fd2ea0a7 1390 netif_tx_stop_all_queues(netdev);
9d5c8243
AK		 1391
1392 strcpy(netdev->name, "eth%d");
1393 err = register_netdev(netdev);
1394 if (err)
1395 goto err_register;
1396
37680117
AD		 1397#ifdef CONFIG_PCI_IOV
1398 /* since iov functionality isn't critical to base device function we
1399 * can accept failure. If it fails we don't allow iov to be enabled */
1400 if (hw->mac.type == e1000_82576) {
1401 err = pci_enable_sriov(pdev, 0);
1402 if (!err)
1403 err = device_create_file(&netdev->dev,
1404 &dev_attr_num_vfs);
1405 if (err)
1406 dev_err(&pdev->dev, "Failed to initialize IOV\n");
1407 }
1408
1409#endif
421e02f0 1410#ifdef CONFIG_IGB_DCA
bbd98fe4 1411 if (dca_add_requester(&pdev->dev) == 0) {
7dfc16fa 1412 adapter->flags |= IGB_FLAG_DCA_ENABLED;
fe4506b6
JC		 1413 dev_info(&pdev->dev, "DCA enabled\n");
1414 /* Always use CB2 mode, difference is masked
1415 * in the CB driver. */
cbd347ad 1416 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
fe4506b6
JC		 1417 igb_setup_dca(adapter);
1418 }
1419#endif
1420
38c845c7
PO		 1421 /*
1422 * Initialize hardware timer: we keep it running just in case
1423 * that some program needs it later on.
1424 */
1425 memset(&adapter->cycles, 0, sizeof(adapter->cycles));
1426 adapter->cycles.read = igb_read_clock;
1427 adapter->cycles.mask = CLOCKSOURCE_MASK(64);
1428 adapter->cycles.mult = 1;
1429 adapter->cycles.shift = IGB_TSYNC_SHIFT;
1430 wr32(E1000_TIMINCA,
1431 (1<<24) |
1432 IGB_TSYNC_CYCLE_TIME_IN_NANOSECONDS * IGB_TSYNC_SCALE);
1433#if 0
1434 /*
1435 * Avoid rollover while we initialize by resetting the time counter.
1436 */
1437 wr32(E1000_SYSTIML, 0x00000000);
1438 wr32(E1000_SYSTIMH, 0x00000000);
1439#else
1440 /*
1441 * Set registers so that rollover occurs soon to test this.
1442 */
1443 wr32(E1000_SYSTIML, 0x00000000);
1444 wr32(E1000_SYSTIMH, 0xFF800000);
1445#endif
1446 wrfl();
1447 timecounter_init(&adapter->clock,
1448 &adapter->cycles,
1449 ktime_to_ns(ktime_get_real()));
1450
33af6bcc
PO		 1451 /*
1452 * Synchronize our NIC clock against system wall clock. NIC
1453 * time stamp reading requires ~3us per sample, each sample
1454 * was pretty stable even under load => only require 10
1455 * samples for each offset comparison.
1456 */
1457 memset(&adapter->compare, 0, sizeof(adapter->compare));
1458 adapter->compare.source = &adapter->clock;
1459 adapter->compare.target = ktime_get_real;
1460 adapter->compare.num_samples = 10;
1461 timecompare_update(&adapter->compare, 0);
1462
38c845c7
PO		 1463#ifdef DEBUG
1464 {
1465 char buffer[160];
1466 printk(KERN_DEBUG
1467 "igb: %s: hw %p initialized timer\n",
1468 igb_get_time_str(adapter, buffer),
1469 &adapter->hw);
1470 }
1471#endif
1472
9d5c8243
AK		 1473 dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
1474 /* print bus type/speed/width info */
7c510e4b 1475 dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
9d5c8243
AK		 1476 netdev->name,
1477 ((hw->bus.speed == e1000_bus_speed_2500)
1478 ? "2.5Gb/s" : "unknown"),
1479 ((hw->bus.width == e1000_bus_width_pcie_x4)
1480 ? "Width x4" : (hw->bus.width == e1000_bus_width_pcie_x1)
1481 ? "Width x1" : "unknown"),
7c510e4b 1482 netdev->dev_addr);
9d5c8243
AK		 1483
1484 igb_read_part_num(hw, &part_num);
1485 dev_info(&pdev->dev, "%s: PBA No: %06x-%03x\n", netdev->name,
1486 (part_num >> 8), (part_num & 0xff));
1487
1488 dev_info(&pdev->dev,
1489 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
1490 adapter->msix_entries ? "MSI-X" :
7dfc16fa 1491 (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
9d5c8243
AK		 1492 adapter->num_rx_queues, adapter->num_tx_queues);
1493
9d5c8243
AK		 1494 return 0;
1495
1496err_register:
1497 igb_release_hw_control(adapter);
1498err_eeprom:
1499 if (!igb_check_reset_block(hw))
f5f4cf08 1500 igb_reset_phy(hw);
9d5c8243
AK		 1501
1502 if (hw->flash_address)
1503 iounmap(hw->flash_address);
1504
a88f10ec 1505 igb_free_queues(adapter);
9d5c8243 1506err_sw_init:
9d5c8243
AK		 1507 iounmap(hw->hw_addr);
1508err_ioremap:
1509 free_netdev(netdev);
1510err_alloc_etherdev:
aed5dec3
AD		 1511 pci_release_selected_regions(pdev, pci_select_bars(pdev,
1512 IORESOURCE_MEM));
9d5c8243
AK		 1513err_pci_reg:
1514err_dma:
1515 pci_disable_device(pdev);
1516 return err;
1517}
1518
1519/**
1520 * igb_remove - Device Removal Routine
1521 * @pdev: PCI device information struct
1522 *
1523 * igb_remove is called by the PCI subsystem to alert the driver
1524 * that it should release a PCI device. The could be caused by a
1525 * Hot-Plug event, or because the driver is going to be removed from
1526 * memory.
1527 **/
1528static void __devexit igb_remove(struct pci_dev *pdev)
1529{
1530 struct net_device *netdev = pci_get_drvdata(pdev);
1531 struct igb_adapter *adapter = netdev_priv(netdev);
fe4506b6 1532 struct e1000_hw *hw = &adapter->hw;
ea943d41 1533 int err;
9d5c8243
AK		 1534
1535 /* flush_scheduled work may reschedule our watchdog task, so
1536 * explicitly disable watchdog tasks from being rescheduled */
1537 set_bit(__IGB_DOWN, &adapter->state);
1538 del_timer_sync(&adapter->watchdog_timer);
1539 del_timer_sync(&adapter->phy_info_timer);
1540
1541 flush_scheduled_work();
1542
421e02f0 1543#ifdef CONFIG_IGB_DCA
7dfc16fa 1544 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
fe4506b6
JC		 1545 dev_info(&pdev->dev, "DCA disabled\n");
1546 dca_remove_requester(&pdev->dev);
7dfc16fa 1547 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
cbd347ad 1548 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
fe4506b6
JC		 1549 }
1550#endif
1551
9d5c8243
AK		 1552 /* Release control of h/w to f/w. If f/w is AMT enabled, this
1553 * would have already happened in close and is redundant. */
1554 igb_release_hw_control(adapter);
1555
1556 unregister_netdev(netdev);
1557
f5f4cf08
AD		 1558 if (!igb_check_reset_block(&adapter->hw))
1559 igb_reset_phy(&adapter->hw);
9d5c8243 1560
9d5c8243
AK		 1561 igb_reset_interrupt_capability(adapter);
1562
a88f10ec 1563 igb_free_queues(adapter);
9d5c8243 1564
37680117
AD		 1565#ifdef CONFIG_PCI_IOV
1566 /* reclaim resources allocated to VFs */
1567 if (adapter->vf_data) {
1568 /* disable iov and allow time for transactions to clear */
1569 pci_disable_sriov(pdev);
1570 msleep(500);
1571
1572 kfree(adapter->vf_data);
1573 adapter->vf_data = NULL;
1574 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
1575 msleep(100);
1576 dev_info(&pdev->dev, "IOV Disabled\n");
1577 }
1578#endif
28b0759c
AD		 1579 iounmap(hw->hw_addr);
1580 if (hw->flash_address)
1581 iounmap(hw->flash_address);
aed5dec3
AD		 1582 pci_release_selected_regions(pdev, pci_select_bars(pdev,
1583 IORESOURCE_MEM));
9d5c8243
AK		 1584
1585 free_netdev(netdev);
1586
ea943d41
JK		 1587 err = pci_disable_pcie_error_reporting(pdev);
1588 if (err)
1589 dev_err(&pdev->dev,
1590 "pci_disable_pcie_error_reporting failed 0x%x\n", err);
40a914fa 1591
9d5c8243
AK		 1592 pci_disable_device(pdev);
1593}
1594
1595/**
1596 * igb_sw_init - Initialize general software structures (struct igb_adapter)
1597 * @adapter: board private structure to initialize
1598 *
1599 * igb_sw_init initializes the Adapter private data structure.
1600 * Fields are initialized based on PCI device information and
1601 * OS network device settings (MTU size).
1602 **/
1603static int __devinit igb_sw_init(struct igb_adapter *adapter)
1604{
1605 struct e1000_hw *hw = &adapter->hw;
1606 struct net_device *netdev = adapter->netdev;
1607 struct pci_dev *pdev = adapter->pdev;
1608
1609 pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
1610
68fd9910
AD		 1611 adapter->tx_ring_count = IGB_DEFAULT_TXD;
1612 adapter->rx_ring_count = IGB_DEFAULT_RXD;
9d5c8243
AK		 1613 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1614 adapter->rx_ps_hdr_size = 0; /* disable packet split */
1615 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1616 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1617
661086df
PWJ		 1618 /* This call may decrease the number of queues depending on
1619 * interrupt mode. */
9d5c8243
AK		 1620 igb_set_interrupt_capability(adapter);
1621
1622 if (igb_alloc_queues(adapter)) {
1623 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
1624 return -ENOMEM;
1625 }
1626
1627 /* Explicitly disable IRQ since the NIC can be in any state. */
1628 igb_irq_disable(adapter);
1629
1630 set_bit(__IGB_DOWN, &adapter->state);
1631 return 0;
1632}
1633
1634/**
1635 * igb_open - Called when a network interface is made active
1636 * @netdev: network interface device structure
1637 *
1638 * Returns 0 on success, negative value on failure
1639 *
1640 * The open entry point is called when a network interface is made
1641 * active by the system (IFF_UP). At this point all resources needed
1642 * for transmit and receive operations are allocated, the interrupt
1643 * handler is registered with the OS, the watchdog timer is started,
1644 * and the stack is notified that the interface is ready.
1645 **/
1646static int igb_open(struct net_device *netdev)
1647{
1648 struct igb_adapter *adapter = netdev_priv(netdev);
1649 struct e1000_hw *hw = &adapter->hw;
1650 int err;
1651 int i;
1652
1653 /* disallow open during test */
1654 if (test_bit(__IGB_TESTING, &adapter->state))
1655 return -EBUSY;
1656
1657 /* allocate transmit descriptors */
1658 err = igb_setup_all_tx_resources(adapter);
1659 if (err)
1660 goto err_setup_tx;
1661
1662 /* allocate receive descriptors */
1663 err = igb_setup_all_rx_resources(adapter);
1664 if (err)
1665 goto err_setup_rx;
1666
1667 /* e1000_power_up_phy(adapter); */
1668
1669 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1670 if ((adapter->hw.mng_cookie.status &
1671 E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
1672 igb_update_mng_vlan(adapter);
1673
1674 /* before we allocate an interrupt, we must be ready to handle it.
1675 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1676 * as soon as we call pci_request_irq, so we have to setup our
1677 * clean_rx handler before we do so. */
1678 igb_configure(adapter);
1679
4ae196df 1680 igb_vmm_control(adapter);
e1739522
AD		 1681 igb_set_rah_pool(hw, adapter->vfs_allocated_count, 0);
1682 igb_set_vmolr(hw, adapter->vfs_allocated_count);
1683
9d5c8243
AK		 1684 err = igb_request_irq(adapter);
1685 if (err)
1686 goto err_req_irq;
1687
1688 /* From here on the code is the same as igb_up() */
1689 clear_bit(__IGB_DOWN, &adapter->state);
1690
844290e5
PW		 1691 for (i = 0; i < adapter->num_rx_queues; i++)
1692 napi_enable(&adapter->rx_ring[i].napi);
9d5c8243
AK		 1693
1694 /* Clear any pending interrupts. */
1695 rd32(E1000_ICR);
844290e5
PW		 1696
1697 igb_irq_enable(adapter);
1698
d55b53ff
JK		 1699 netif_tx_start_all_queues(netdev);
1700
9d5c8243
AK		 1701 /* Fire a link status change interrupt to start the watchdog. */
1702 wr32(E1000_ICS, E1000_ICS_LSC);
1703
1704 return 0;
1705
1706err_req_irq:
1707 igb_release_hw_control(adapter);
1708 /* e1000_power_down_phy(adapter); */
1709 igb_free_all_rx_resources(adapter);
1710err_setup_rx:
1711 igb_free_all_tx_resources(adapter);
1712err_setup_tx:
1713 igb_reset(adapter);
1714
1715 return err;
1716}
1717
1718/**
1719 * igb_close - Disables a network interface
1720 * @netdev: network interface device structure
1721 *
1722 * Returns 0, this is not allowed to fail
1723 *
1724 * The close entry point is called when an interface is de-activated
1725 * by the OS. The hardware is still under the driver's control, but
1726 * needs to be disabled. A global MAC reset is issued to stop the
1727 * hardware, and all transmit and receive resources are freed.
1728 **/
1729static int igb_close(struct net_device *netdev)
1730{
1731 struct igb_adapter *adapter = netdev_priv(netdev);
1732
1733 WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
1734 igb_down(adapter);
1735
1736 igb_free_irq(adapter);
1737
1738 igb_free_all_tx_resources(adapter);
1739 igb_free_all_rx_resources(adapter);
1740
1741 /* kill manageability vlan ID if supported, but not if a vlan with
1742 * the same ID is registered on the host OS (let 8021q kill it) */
1743 if ((adapter->hw.mng_cookie.status &
1744 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1745 !(adapter->vlgrp &&
1746 vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
1747 igb_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1748
1749 return 0;
1750}
1751
1752/**
1753 * igb_setup_tx_resources - allocate Tx resources (Descriptors)
1754 * @adapter: board private structure
1755 * @tx_ring: tx descriptor ring (for a specific queue) to setup
1756 *
1757 * Return 0 on success, negative on failure
1758 **/
9d5c8243
AK		 1759int igb_setup_tx_resources(struct igb_adapter *adapter,
1760 struct igb_ring *tx_ring)
1761{
1762 struct pci_dev *pdev = adapter->pdev;
1763 int size;
1764
1765 size = sizeof(struct igb_buffer) * tx_ring->count;
1766 tx_ring->buffer_info = vmalloc(size);
1767 if (!tx_ring->buffer_info)
1768 goto err;
1769 memset(tx_ring->buffer_info, 0, size);
1770
1771 /* round up to nearest 4K */
85e8d004 1772 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
9d5c8243
AK		 1773 tx_ring->size = ALIGN(tx_ring->size, 4096);
1774
1775 tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
1776 &tx_ring->dma);
1777
1778 if (!tx_ring->desc)
1779 goto err;
1780
1781 tx_ring->adapter = adapter;
1782 tx_ring->next_to_use = 0;
1783 tx_ring->next_to_clean = 0;
9d5c8243
AK		 1784 return 0;
1785
1786err:
1787 vfree(tx_ring->buffer_info);
1788 dev_err(&adapter->pdev->dev,
1789 "Unable to allocate memory for the transmit descriptor ring\n");
1790 return -ENOMEM;
1791}
1792
1793/**
1794 * igb_setup_all_tx_resources - wrapper to allocate Tx resources
1795 * (Descriptors) for all queues
1796 * @adapter: board private structure
1797 *
1798 * Return 0 on success, negative on failure
1799 **/
1800static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
1801{
1802 int i, err = 0;
661086df 1803 int r_idx;
9d5c8243
AK		 1804
1805 for (i = 0; i < adapter->num_tx_queues; i++) {
1806 err = igb_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1807 if (err) {
1808 dev_err(&adapter->pdev->dev,
1809 "Allocation for Tx Queue %u failed\n", i);
1810 for (i--; i >= 0; i--)
3b644cf6 1811 igb_free_tx_resources(&adapter->tx_ring[i]);
9d5c8243
AK		 1812 break;
1813 }
1814 }
1815
661086df
PWJ		 1816 for (i = 0; i < IGB_MAX_TX_QUEUES; i++) {
1817 r_idx = i % adapter->num_tx_queues;
1818 adapter->multi_tx_table[i] = &adapter->tx_ring[r_idx];
eebbbdba 1819 }
9d5c8243
AK		 1820 return err;
1821}
1822
1823/**
1824 * igb_configure_tx - Configure transmit Unit after Reset
1825 * @adapter: board private structure
1826 *
1827 * Configure the Tx unit of the MAC after a reset.
1828 **/
1829static void igb_configure_tx(struct igb_adapter *adapter)
1830{
0e014cb1 1831 u64 tdba;
9d5c8243
AK		 1832 struct e1000_hw *hw = &adapter->hw;
1833 u32 tctl;
1834 u32 txdctl, txctrl;
26bc19ec 1835 int i, j;
9d5c8243
AK		 1836
1837 for (i = 0; i < adapter->num_tx_queues; i++) {
73cd78f1 1838 struct igb_ring *ring = &adapter->tx_ring[i];
26bc19ec
AD		 1839 j = ring->reg_idx;
1840 wr32(E1000_TDLEN(j),
85e8d004 1841 ring->count * sizeof(union e1000_adv_tx_desc));
9d5c8243 1842 tdba = ring->dma;
26bc19ec 1843 wr32(E1000_TDBAL(j),
73cd78f1 1844 tdba & 0x00000000ffffffffULL);
26bc19ec 1845 wr32(E1000_TDBAH(j), tdba >> 32);
9d5c8243 1846
26bc19ec
AD		 1847 ring->head = E1000_TDH(j);
1848 ring->tail = E1000_TDT(j);
9d5c8243
AK		 1849 writel(0, hw->hw_addr + ring->tail);
1850 writel(0, hw->hw_addr + ring->head);
26bc19ec 1851 txdctl = rd32(E1000_TXDCTL(j));
9d5c8243 1852 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
26bc19ec 1853 wr32(E1000_TXDCTL(j), txdctl);
9d5c8243
AK		 1854
1855 /* Turn off Relaxed Ordering on head write-backs. The
1856 * writebacks MUST be delivered in order or it will
1857 * completely screw up our bookeeping.
1858 */
26bc19ec 1859 txctrl = rd32(E1000_DCA_TXCTRL(j));
9d5c8243 1860 txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
26bc19ec 1861 wr32(E1000_DCA_TXCTRL(j), txctrl);
9d5c8243
AK		 1862 }
1863
e1739522
AD		 1864 /* disable queue 0 to prevent tail bump w/o re-configuration */
1865 if (adapter->vfs_allocated_count)
1866 wr32(E1000_TXDCTL(0), 0);
9d5c8243
AK		 1867
1868 /* Program the Transmit Control Register */
9d5c8243
AK		 1869 tctl = rd32(E1000_TCTL);
1870 tctl &= ~E1000_TCTL_CT;
1871 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1872 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1873
1874 igb_config_collision_dist(hw);
1875
1876 /* Setup Transmit Descriptor Settings for eop descriptor */
1877 adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_RS;
1878
1879 /* Enable transmits */
1880 tctl |= E1000_TCTL_EN;
1881
1882 wr32(E1000_TCTL, tctl);
1883}
1884
1885/**
1886 * igb_setup_rx_resources - allocate Rx resources (Descriptors)
1887 * @adapter: board private structure
1888 * @rx_ring: rx descriptor ring (for a specific queue) to setup
1889 *
1890 * Returns 0 on success, negative on failure
1891 **/
9d5c8243
AK		 1892int igb_setup_rx_resources(struct igb_adapter *adapter,
1893 struct igb_ring *rx_ring)
1894{
1895 struct pci_dev *pdev = adapter->pdev;
1896 int size, desc_len;
1897
1898 size = sizeof(struct igb_buffer) * rx_ring->count;
1899 rx_ring->buffer_info = vmalloc(size);
1900 if (!rx_ring->buffer_info)
1901 goto err;
1902 memset(rx_ring->buffer_info, 0, size);
1903
1904 desc_len = sizeof(union e1000_adv_rx_desc);
1905
1906 /* Round up to nearest 4K */
1907 rx_ring->size = rx_ring->count * desc_len;
1908 rx_ring->size = ALIGN(rx_ring->size, 4096);
1909
1910 rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size,
1911 &rx_ring->dma);
1912
1913 if (!rx_ring->desc)
1914 goto err;
1915
1916 rx_ring->next_to_clean = 0;
1917 rx_ring->next_to_use = 0;
9d5c8243
AK		 1918
1919 rx_ring->adapter = adapter;
9d5c8243
AK		 1920
1921 return 0;
1922
1923err:
1924 vfree(rx_ring->buffer_info);
1925 dev_err(&adapter->pdev->dev, "Unable to allocate memory for "
1926 "the receive descriptor ring\n");
1927 return -ENOMEM;
1928}
1929
1930/**
1931 * igb_setup_all_rx_resources - wrapper to allocate Rx resources
1932 * (Descriptors) for all queues
1933 * @adapter: board private structure
1934 *
1935 * Return 0 on success, negative on failure
1936 **/
1937static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
1938{
1939 int i, err = 0;
1940
1941 for (i = 0; i < adapter->num_rx_queues; i++) {
1942 err = igb_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1943 if (err) {
1944 dev_err(&adapter->pdev->dev,
1945 "Allocation for Rx Queue %u failed\n", i);
1946 for (i--; i >= 0; i--)
3b644cf6 1947 igb_free_rx_resources(&adapter->rx_ring[i]);
9d5c8243
AK		 1948 break;
1949 }
1950 }
1951
1952 return err;
1953}
1954
1955/**
1956 * igb_setup_rctl - configure the receive control registers
1957 * @adapter: Board private structure
1958 **/
1959static void igb_setup_rctl(struct igb_adapter *adapter)
1960{
1961 struct e1000_hw *hw = &adapter->hw;
1962 u32 rctl;
1963 u32 srrctl = 0;
26bc19ec 1964 int i, j;
9d5c8243
AK		 1965
1966 rctl = rd32(E1000_RCTL);
1967
1968 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
69d728ba 1969 rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
9d5c8243 1970
69d728ba 1971 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
28b0759c 1972 (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
9d5c8243 1973
87cb7e8c
AK		 1974 /*
1975 * enable stripping of CRC. It's unlikely this will break BMC
1976 * redirection as it did with e1000. Newer features require
1977 * that the HW strips the CRC.
73cd78f1 1978 */
87cb7e8c 1979 rctl |= E1000_RCTL_SECRC;
9d5c8243 1980
9b07f3d3 1981 /*
ec54d7d6 1982 * disable store bad packets and clear size bits.
9b07f3d3 1983 */
ec54d7d6 1984 rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
9d5c8243 1985
ec54d7d6 1986 /* enable LPE when to prevent packets larger than max_frame_size */
9b07f3d3 1987 rctl |= E1000_RCTL_LPE;
b4557be2
AD		 1988
1989 /* Setup buffer sizes */
1990 switch (adapter->rx_buffer_len) {
1991 case IGB_RXBUFFER_256:
1992 rctl |= E1000_RCTL_SZ_256;
1993 break;
1994 case IGB_RXBUFFER_512:
1995 rctl |= E1000_RCTL_SZ_512;
1996 break;
1997 default:
1998 srrctl = ALIGN(adapter->rx_buffer_len, 1024)
1999 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
2000 break;
9d5c8243
AK		 2001 }
2002
2003 /* 82575 and greater support packet-split where the protocol
2004 * header is placed in skb->data and the packet data is
2005 * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2006 * In the case of a non-split, skb->data is linearly filled,
2007 * followed by the page buffers. Therefore, skb->data is
2008 * sized to hold the largest protocol header.
2009 */
2010 /* allocations using alloc_page take too long for regular MTU
2011 * so only enable packet split for jumbo frames */
ec54d7d6 2012 if (adapter->netdev->mtu > ETH_DATA_LEN) {
9d5c8243 2013 adapter->rx_ps_hdr_size = IGB_RXBUFFER_128;
bf36c1a0 2014 srrctl |= adapter->rx_ps_hdr_size <<
9d5c8243 2015 E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
9d5c8243
AK		 2016 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2017 } else {
2018 adapter->rx_ps_hdr_size = 0;
2019 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
2020 }
2021
e1739522
AD		 2022 /* Attention!!! For SR-IOV PF driver operations you must enable
2023 * queue drop for all VF and PF queues to prevent head of line blocking
2024 * if an un-trusted VF does not provide descriptors to hardware.
2025 */
2026 if (adapter->vfs_allocated_count) {
2027 u32 vmolr;
2028
2029 j = adapter->rx_ring[0].reg_idx;
2030
2031 /* set all queue drop enable bits */
2032 wr32(E1000_QDE, ALL_QUEUES);
2033 srrctl |= E1000_SRRCTL_DROP_EN;
2034
2035 /* disable queue 0 to prevent tail write w/o re-config */
2036 wr32(E1000_RXDCTL(0), 0);
2037
2038 vmolr = rd32(E1000_VMOLR(j));
2039 if (rctl & E1000_RCTL_LPE)
2040 vmolr |= E1000_VMOLR_LPE;
2041 if (adapter->num_rx_queues > 0)
2042 vmolr |= E1000_VMOLR_RSSE;
2043 wr32(E1000_VMOLR(j), vmolr);
2044 }
2045
26bc19ec
AD		 2046 for (i = 0; i < adapter->num_rx_queues; i++) {
2047 j = adapter->rx_ring[i].reg_idx;
2048 wr32(E1000_SRRCTL(j), srrctl);
2049 }
9d5c8243
AK		 2050
2051 wr32(E1000_RCTL, rctl);
2052}
2053
e1739522
AD		 2054/**
2055 * igb_rlpml_set - set maximum receive packet size
2056 * @adapter: board private structure
2057 *
2058 * Configure maximum receivable packet size.
2059 **/
2060static void igb_rlpml_set(struct igb_adapter *adapter)
2061{
2062 u32 max_frame_size = adapter->max_frame_size;
2063 struct e1000_hw *hw = &adapter->hw;
2064 u16 pf_id = adapter->vfs_allocated_count;
2065
2066 if (adapter->vlgrp)
2067 max_frame_size += VLAN_TAG_SIZE;
2068
2069 /* if vfs are enabled we set RLPML to the largest possible request
2070 * size and set the VMOLR RLPML to the size we need */
2071 if (pf_id) {
2072 igb_set_vf_rlpml(adapter, max_frame_size, pf_id);
2073 max_frame_size = MAX_STD_JUMBO_FRAME_SIZE + VLAN_TAG_SIZE;
2074 }
2075
2076 wr32(E1000_RLPML, max_frame_size);
2077}
2078
2079/**
2080 * igb_configure_vt_default_pool - Configure VT default pool
2081 * @adapter: board private structure
2082 *
2083 * Configure the default pool
2084 **/
2085static void igb_configure_vt_default_pool(struct igb_adapter *adapter)
2086{
2087 struct e1000_hw *hw = &adapter->hw;
2088 u16 pf_id = adapter->vfs_allocated_count;
2089 u32 vtctl;
2090
2091 /* not in sr-iov mode - do nothing */
2092 if (!pf_id)
2093 return;
2094
2095 vtctl = rd32(E1000_VT_CTL);
2096 vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK |
2097 E1000_VT_CTL_DISABLE_DEF_POOL);
2098 vtctl |= pf_id << E1000_VT_CTL_DEFAULT_POOL_SHIFT;
2099 wr32(E1000_VT_CTL, vtctl);
2100}
2101
9d5c8243
AK		 2102/**
2103 * igb_configure_rx - Configure receive Unit after Reset
2104 * @adapter: board private structure
2105 *
2106 * Configure the Rx unit of the MAC after a reset.
2107 **/
2108static void igb_configure_rx(struct igb_adapter *adapter)
2109{
2110 u64 rdba;
2111 struct e1000_hw *hw = &adapter->hw;
2112 u32 rctl, rxcsum;
2113 u32 rxdctl;
9107584e 2114 int i;
9d5c8243
AK		 2115
2116 /* disable receives while setting up the descriptors */
2117 rctl = rd32(E1000_RCTL);
2118 wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
2119 wrfl();
2120 mdelay(10);
2121
2122 if (adapter->itr_setting > 3)
6eb5a7f1 2123 wr32(E1000_ITR, adapter->itr);
9d5c8243
AK		 2124
2125 /* Setup the HW Rx Head and Tail Descriptor Pointers and
2126 * the Base and Length of the Rx Descriptor Ring */
2127 for (i = 0; i < adapter->num_rx_queues; i++) {
73cd78f1 2128 struct igb_ring *ring = &adapter->rx_ring[i];
9107584e 2129 int j = ring->reg_idx;
9d5c8243 2130 rdba = ring->dma;
26bc19ec 2131 wr32(E1000_RDBAL(j),
73cd78f1 2132 rdba & 0x00000000ffffffffULL);
26bc19ec
AD		 2133 wr32(E1000_RDBAH(j), rdba >> 32);
2134 wr32(E1000_RDLEN(j),
73cd78f1 2135 ring->count * sizeof(union e1000_adv_rx_desc));
9d5c8243 2136
26bc19ec
AD		 2137 ring->head = E1000_RDH(j);
2138 ring->tail = E1000_RDT(j);
9d5c8243
AK		 2139 writel(0, hw->hw_addr + ring->tail);
2140 writel(0, hw->hw_addr + ring->head);
2141
26bc19ec 2142 rxdctl = rd32(E1000_RXDCTL(j));
9d5c8243
AK		 2143 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
2144 rxdctl &= 0xFFF00000;
2145 rxdctl |= IGB_RX_PTHRESH;
2146 rxdctl |= IGB_RX_HTHRESH << 8;
2147 rxdctl |= IGB_RX_WTHRESH << 16;
26bc19ec 2148 wr32(E1000_RXDCTL(j), rxdctl);
9d5c8243
AK		 2149 }
2150
2151 if (adapter->num_rx_queues > 1) {
2152 u32 random[10];
2153 u32 mrqc;
2154 u32 j, shift;
2155 union e1000_reta {
2156 u32 dword;
2157 u8 bytes[4];
2158 } reta;
2159
2160 get_random_bytes(&random[0], 40);
2161
2d064c06
AD		 2162 if (hw->mac.type >= e1000_82576)
2163 shift = 0;
2164 else
2165 shift = 6;
9d5c8243
AK		 2166 for (j = 0; j < (32 * 4); j++) {
2167 reta.bytes[j & 3] =
26bc19ec 2168 adapter->rx_ring[(j % adapter->num_rx_queues)].reg_idx << shift;
9d5c8243
AK		 2169 if ((j & 3) == 3)
2170 writel(reta.dword,
2171 hw->hw_addr + E1000_RETA(0) + (j & ~3));
2172 }
e1739522
AD		 2173 if (adapter->vfs_allocated_count)
2174 mrqc = E1000_MRQC_ENABLE_VMDQ_RSS_2Q;
2175 else
2176 mrqc = E1000_MRQC_ENABLE_RSS_4Q;
9d5c8243
AK		 2177
2178 /* Fill out hash function seeds */
2179 for (j = 0; j < 10; j++)
2180 array_wr32(E1000_RSSRK(0), j, random[j]);
2181
2182 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
2183 E1000_MRQC_RSS_FIELD_IPV4_TCP);
2184 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6 |
2185 E1000_MRQC_RSS_FIELD_IPV6_TCP);
2186 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4_UDP |
2187 E1000_MRQC_RSS_FIELD_IPV6_UDP);
2188 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
2189 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
2190
2191
2192 wr32(E1000_MRQC, mrqc);
2193
2194 /* Multiqueue and raw packet checksumming are mutually
2195 * exclusive. Note that this not the same as TCP/IP
2196 * checksumming, which works fine. */
2197 rxcsum = rd32(E1000_RXCSUM);
2198 rxcsum |= E1000_RXCSUM_PCSD;
2199 wr32(E1000_RXCSUM, rxcsum);
2200 } else {
e1739522
AD		 2201 /* Enable multi-queue for sr-iov */
2202 if (adapter->vfs_allocated_count)
2203 wr32(E1000_MRQC, E1000_MRQC_ENABLE_VMDQ);
9d5c8243
AK		 2204 /* Enable Receive Checksum Offload for TCP and UDP */
2205 rxcsum = rd32(E1000_RXCSUM);
56fbbb4e
AD		 2206 if (adapter->rx_csum)
2207 rxcsum |= E1000_RXCSUM_TUOFL | E1000_RXCSUM_IPPCSE;
2208 else
2209 rxcsum &= ~(E1000_RXCSUM_TUOFL | E1000_RXCSUM_IPPCSE);
2210
9d5c8243
AK		 2211 wr32(E1000_RXCSUM, rxcsum);
2212 }
2213
e1739522
AD		 2214 /* Set the default pool for the PF's first queue */
2215 igb_configure_vt_default_pool(adapter);
2216
2217 igb_rlpml_set(adapter);
9d5c8243
AK		 2218
2219 /* Enable Receives */
2220 wr32(E1000_RCTL, rctl);
2221}
2222
2223/**
2224 * igb_free_tx_resources - Free Tx Resources per Queue
9d5c8243
AK		 2225 * @tx_ring: Tx descriptor ring for a specific queue
2226 *
2227 * Free all transmit software resources
2228 **/
68fd9910 2229void igb_free_tx_resources(struct igb_ring *tx_ring)
9d5c8243 2230{
3b644cf6 2231 struct pci_dev *pdev = tx_ring->adapter->pdev;
9d5c8243 2232
3b644cf6 2233 igb_clean_tx_ring(tx_ring);
9d5c8243
AK		 2234
2235 vfree(tx_ring->buffer_info);
2236 tx_ring->buffer_info = NULL;
2237
2238 pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
2239
2240 tx_ring->desc = NULL;
2241}
2242
2243/**
2244 * igb_free_all_tx_resources - Free Tx Resources for All Queues
2245 * @adapter: board private structure
2246 *
2247 * Free all transmit software resources
2248 **/
2249static void igb_free_all_tx_resources(struct igb_adapter *adapter)
2250{
2251 int i;
2252
2253 for (i = 0; i < adapter->num_tx_queues; i++)
3b644cf6 2254 igb_free_tx_resources(&adapter->tx_ring[i]);
9d5c8243
AK		 2255}
2256
2257static void igb_unmap_and_free_tx_resource(struct igb_adapter *adapter,
2258 struct igb_buffer *buffer_info)
2259{
2260 if (buffer_info->dma) {
2261 pci_unmap_page(adapter->pdev,
2262 buffer_info->dma,
2263 buffer_info->length,
2264 PCI_DMA_TODEVICE);
2265 buffer_info->dma = 0;
2266 }
2267 if (buffer_info->skb) {
2268 dev_kfree_skb_any(buffer_info->skb);
2269 buffer_info->skb = NULL;
2270 }
2271 buffer_info->time_stamp = 0;
73cd78f1 2272 buffer_info->next_to_watch = 0;
9d5c8243
AK		 2273 /* buffer_info must be completely set up in the transmit path */
2274}
2275
2276/**
2277 * igb_clean_tx_ring - Free Tx Buffers
9d5c8243
AK		 2278 * @tx_ring: ring to be cleaned
2279 **/
3b644cf6 2280static void igb_clean_tx_ring(struct igb_ring *tx_ring)
9d5c8243 2281{
3b644cf6 2282 struct igb_adapter *adapter = tx_ring->adapter;
9d5c8243
AK		 2283 struct igb_buffer *buffer_info;
2284 unsigned long size;
2285 unsigned int i;
2286
2287 if (!tx_ring->buffer_info)
2288 return;
2289 /* Free all the Tx ring sk_buffs */
2290
2291 for (i = 0; i < tx_ring->count; i++) {
2292 buffer_info = &tx_ring->buffer_info[i];
2293 igb_unmap_and_free_tx_resource(adapter, buffer_info);
2294 }
2295
2296 size = sizeof(struct igb_buffer) * tx_ring->count;
2297 memset(tx_ring->buffer_info, 0, size);
2298
2299 /* Zero out the descriptor ring */
2300
2301 memset(tx_ring->desc, 0, tx_ring->size);
2302
2303 tx_ring->next_to_use = 0;
2304 tx_ring->next_to_clean = 0;
2305
2306 writel(0, adapter->hw.hw_addr + tx_ring->head);
2307 writel(0, adapter->hw.hw_addr + tx_ring->tail);
2308}
2309
2310/**
2311 * igb_clean_all_tx_rings - Free Tx Buffers for all queues
2312 * @adapter: board private structure
2313 **/
2314static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
2315{
2316 int i;
2317
2318 for (i = 0; i < adapter->num_tx_queues; i++)
3b644cf6 2319 igb_clean_tx_ring(&adapter->tx_ring[i]);
9d5c8243
AK		 2320}
2321
2322/**
2323 * igb_free_rx_resources - Free Rx Resources
9d5c8243
AK		 2324 * @rx_ring: ring to clean the resources from
2325 *
2326 * Free all receive software resources
2327 **/
68fd9910 2328void igb_free_rx_resources(struct igb_ring *rx_ring)
9d5c8243 2329{
3b644cf6 2330 struct pci_dev *pdev = rx_ring->adapter->pdev;
9d5c8243 2331
3b644cf6 2332 igb_clean_rx_ring(rx_ring);
9d5c8243
AK		 2333
2334 vfree(rx_ring->buffer_info);
2335 rx_ring->buffer_info = NULL;
2336
2337 pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
2338
2339 rx_ring->desc = NULL;
2340}
2341
2342/**
2343 * igb_free_all_rx_resources - Free Rx Resources for All Queues
2344 * @adapter: board private structure
2345 *
2346 * Free all receive software resources
2347 **/
2348static void igb_free_all_rx_resources(struct igb_adapter *adapter)
2349{
2350 int i;
2351
2352 for (i = 0; i < adapter->num_rx_queues; i++)
3b644cf6 2353 igb_free_rx_resources(&adapter->rx_ring[i]);
9d5c8243
AK		 2354}
2355
2356/**
2357 * igb_clean_rx_ring - Free Rx Buffers per Queue
9d5c8243
AK		 2358 * @rx_ring: ring to free buffers from
2359 **/
3b644cf6 2360static void igb_clean_rx_ring(struct igb_ring *rx_ring)
9d5c8243 2361{
3b644cf6 2362 struct igb_adapter *adapter = rx_ring->adapter;
9d5c8243
AK		 2363 struct igb_buffer *buffer_info;
2364 struct pci_dev *pdev = adapter->pdev;
2365 unsigned long size;
2366 unsigned int i;
2367
2368 if (!rx_ring->buffer_info)
2369 return;
2370 /* Free all the Rx ring sk_buffs */
2371 for (i = 0; i < rx_ring->count; i++) {
2372 buffer_info = &rx_ring->buffer_info[i];
2373 if (buffer_info->dma) {
2374 if (adapter->rx_ps_hdr_size)
2375 pci_unmap_single(pdev, buffer_info->dma,
2376 adapter->rx_ps_hdr_size,
2377 PCI_DMA_FROMDEVICE);
2378 else
2379 pci_unmap_single(pdev, buffer_info->dma,
2380 adapter->rx_buffer_len,
2381 PCI_DMA_FROMDEVICE);
2382 buffer_info->dma = 0;
2383 }
2384
2385 if (buffer_info->skb) {
2386 dev_kfree_skb(buffer_info->skb);
2387 buffer_info->skb = NULL;
2388 }
2389 if (buffer_info->page) {
bf36c1a0
AD		 2390 if (buffer_info->page_dma)
2391 pci_unmap_page(pdev, buffer_info->page_dma,
2392 PAGE_SIZE / 2,
2393 PCI_DMA_FROMDEVICE);
9d5c8243
AK		 2394 put_page(buffer_info->page);
2395 buffer_info->page = NULL;
2396 buffer_info->page_dma = 0;
bf36c1a0 2397 buffer_info->page_offset = 0;
9d5c8243
AK		 2398 }
2399 }
2400
9d5c8243
AK		 2401 size = sizeof(struct igb_buffer) * rx_ring->count;
2402 memset(rx_ring->buffer_info, 0, size);
2403
2404 /* Zero out the descriptor ring */
2405 memset(rx_ring->desc, 0, rx_ring->size);
2406
2407 rx_ring->next_to_clean = 0;
2408 rx_ring->next_to_use = 0;
2409
2410 writel(0, adapter->hw.hw_addr + rx_ring->head);
2411 writel(0, adapter->hw.hw_addr + rx_ring->tail);
2412}
2413
2414/**
2415 * igb_clean_all_rx_rings - Free Rx Buffers for all queues
2416 * @adapter: board private structure
2417 **/
2418static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
2419{
2420 int i;
2421
2422 for (i = 0; i < adapter->num_rx_queues; i++)
3b644cf6 2423 igb_clean_rx_ring(&adapter->rx_ring[i]);
9d5c8243
AK		 2424}
2425
2426/**
2427 * igb_set_mac - Change the Ethernet Address of the NIC
2428 * @netdev: network interface device structure
2429 * @p: pointer to an address structure
2430 *
2431 * Returns 0 on success, negative on failure
2432 **/
2433static int igb_set_mac(struct net_device *netdev, void *p)
2434{
2435 struct igb_adapter *adapter = netdev_priv(netdev);
28b0759c 2436 struct e1000_hw *hw = &adapter->hw;
9d5c8243
AK		 2437 struct sockaddr *addr = p;
2438
2439 if (!is_valid_ether_addr(addr->sa_data))
2440 return -EADDRNOTAVAIL;
2441
2442 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
28b0759c 2443 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
9d5c8243 2444
28b0759c 2445 hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
9d5c8243 2446
e1739522
AD		 2447 igb_set_rah_pool(hw, adapter->vfs_allocated_count, 0);
2448
9d5c8243
AK		 2449 return 0;
2450}
2451
2452/**
2453 * igb_set_multi - Multicast and Promiscuous mode set
2454 * @netdev: network interface device structure
2455 *
2456 * The set_multi entry point is called whenever the multicast address
2457 * list or the network interface flags are updated. This routine is
2458 * responsible for configuring the hardware for proper multicast,
2459 * promiscuous mode, and all-multi behavior.
2460 **/
2461static void igb_set_multi(struct net_device *netdev)
2462{
2463 struct igb_adapter *adapter = netdev_priv(netdev);
2464 struct e1000_hw *hw = &adapter->hw;
2465 struct e1000_mac_info *mac = &hw->mac;
2466 struct dev_mc_list *mc_ptr;
c5cd11e3 2467 u8 *mta_list = NULL;
9d5c8243
AK		 2468 u32 rctl;
2469 int i;
2470
2471 /* Check for Promiscuous and All Multicast modes */
2472
2473 rctl = rd32(E1000_RCTL);
2474
746b9f02 2475 if (netdev->flags & IFF_PROMISC) {
9d5c8243 2476 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
746b9f02
PM		 2477 rctl &= ~E1000_RCTL_VFE;
2478 } else {
2479 if (netdev->flags & IFF_ALLMULTI) {
2480 rctl |= E1000_RCTL_MPE;
2481 rctl &= ~E1000_RCTL_UPE;
2482 } else
2483 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
78ed11a5 2484 rctl |= E1000_RCTL_VFE;
746b9f02 2485 }
9d5c8243
AK		 2486 wr32(E1000_RCTL, rctl);
2487
c5cd11e3
AD		 2488 if (netdev->mc_count) {
2489 mta_list = kzalloc(netdev->mc_count * 6, GFP_ATOMIC);
2490 if (!mta_list) {
2491 dev_err(&adapter->pdev->dev,
2492 "failed to allocate multicast filter list\n");
2493 return;
2494 }
9d5c8243
AK		 2495 }
2496
9d5c8243
AK		 2497 /* The shared function expects a packed array of only addresses. */
2498 mc_ptr = netdev->mc_list;
2499
2500 for (i = 0; i < netdev->mc_count; i++) {
2501 if (!mc_ptr)
2502 break;
2503 memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr, ETH_ALEN);
2504 mc_ptr = mc_ptr->next;
2505 }
e1739522
AD		 2506 igb_update_mc_addr_list(hw, mta_list, i,
2507 adapter->vfs_allocated_count + 1,
2508 mac->rar_entry_count);
2509
2510 igb_set_mc_list_pools(adapter, i, mac->rar_entry_count);
4ae196df
AD		 2511 igb_restore_vf_multicasts(adapter);
2512
9d5c8243
AK		 2513 kfree(mta_list);
2514}
2515
2516/* Need to wait a few seconds after link up to get diagnostic information from
2517 * the phy */
2518static void igb_update_phy_info(unsigned long data)
2519{
2520 struct igb_adapter *adapter = (struct igb_adapter *) data;
f5f4cf08 2521 igb_get_phy_info(&adapter->hw);
9d5c8243
AK		 2522}
2523
4d6b725e
AD		 2524/**
2525 * igb_has_link - check shared code for link and determine up/down
2526 * @adapter: pointer to driver private info
2527 **/
2528static bool igb_has_link(struct igb_adapter *adapter)
2529{
2530 struct e1000_hw *hw = &adapter->hw;
2531 bool link_active = false;
2532 s32 ret_val = 0;
2533
2534 /* get_link_status is set on LSC (link status) interrupt or
2535 * rx sequence error interrupt. get_link_status will stay
2536 * false until the e1000_check_for_link establishes link
2537 * for copper adapters ONLY
2538 */
2539 switch (hw->phy.media_type) {
2540 case e1000_media_type_copper:
2541 if (hw->mac.get_link_status) {
2542 ret_val = hw->mac.ops.check_for_link(hw);
2543 link_active = !hw->mac.get_link_status;
2544 } else {
2545 link_active = true;
2546 }
2547 break;
2548 case e1000_media_type_fiber:
2549 ret_val = hw->mac.ops.check_for_link(hw);
2550 link_active = !!(rd32(E1000_STATUS) & E1000_STATUS_LU);
2551 break;
2552 case e1000_media_type_internal_serdes:
2553 ret_val = hw->mac.ops.check_for_link(hw);
2554 link_active = hw->mac.serdes_has_link;
2555 break;
2556 default:
2557 case e1000_media_type_unknown:
2558 break;
2559 }
2560
2561 return link_active;
2562}
2563
9d5c8243
AK		 2564/**
2565 * igb_watchdog - Timer Call-back
2566 * @data: pointer to adapter cast into an unsigned long
2567 **/
2568static void igb_watchdog(unsigned long data)
2569{
2570 struct igb_adapter *adapter = (struct igb_adapter *)data;
2571 /* Do the rest outside of interrupt context */
2572 schedule_work(&adapter->watchdog_task);
2573}
2574
2575static void igb_watchdog_task(struct work_struct *work)
2576{
2577 struct igb_adapter *adapter = container_of(work,
2578 struct igb_adapter, watchdog_task);
2579 struct e1000_hw *hw = &adapter->hw;
9d5c8243
AK		 2580 struct net_device *netdev = adapter->netdev;
2581 struct igb_ring *tx_ring = adapter->tx_ring;
9d5c8243 2582 u32 link;
7a6ea550 2583 u32 eics = 0;
7a6ea550 2584 int i;
9d5c8243 2585
4d6b725e
AD		 2586 link = igb_has_link(adapter);
2587 if ((netif_carrier_ok(netdev)) && link)
9d5c8243
AK		 2588 goto link_up;
2589
9d5c8243
AK		 2590 if (link) {
2591 if (!netif_carrier_ok(netdev)) {
2592 u32 ctrl;
2593 hw->mac.ops.get_speed_and_duplex(&adapter->hw,
2594 &adapter->link_speed,
2595 &adapter->link_duplex);
2596
2597 ctrl = rd32(E1000_CTRL);
527d47c1
AD		 2598 /* Links status message must follow this format */
2599 printk(KERN_INFO "igb: %s NIC Link is Up %d Mbps %s, "
9d5c8243 2600 "Flow Control: %s\n",
527d47c1 2601 netdev->name,
9d5c8243
AK		 2602 adapter->link_speed,
2603 adapter->link_duplex == FULL_DUPLEX ?
2604 "Full Duplex" : "Half Duplex",
2605 ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2606 E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2607 E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2608 E1000_CTRL_TFCE) ? "TX" : "None")));
2609
2610 /* tweak tx_queue_len according to speed/duplex and
2611 * adjust the timeout factor */
2612 netdev->tx_queue_len = adapter->tx_queue_len;
2613 adapter->tx_timeout_factor = 1;
2614 switch (adapter->link_speed) {
2615 case SPEED_10:
2616 netdev->tx_queue_len = 10;
2617 adapter->tx_timeout_factor = 14;
2618 break;
2619 case SPEED_100:
2620 netdev->tx_queue_len = 100;
2621 /* maybe add some timeout factor ? */
2622 break;
2623 }
2624
2625 netif_carrier_on(netdev);
fd2ea0a7 2626 netif_tx_wake_all_queues(netdev);
9d5c8243 2627
4ae196df
AD		 2628 igb_ping_all_vfs(adapter);
2629
4b1a9877 2630 /* link state has changed, schedule phy info update */
9d5c8243
AK		 2631 if (!test_bit(__IGB_DOWN, &adapter->state))
2632 mod_timer(&adapter->phy_info_timer,
2633 round_jiffies(jiffies + 2 * HZ));
2634 }
2635 } else {
2636 if (netif_carrier_ok(netdev)) {
2637 adapter->link_speed = 0;
2638 adapter->link_duplex = 0;
527d47c1
AD		 2639 /* Links status message must follow this format */
2640 printk(KERN_INFO "igb: %s NIC Link is Down\n",
2641 netdev->name);
9d5c8243 2642 netif_carrier_off(netdev);
fd2ea0a7 2643 netif_tx_stop_all_queues(netdev);
4b1a9877 2644
4ae196df
AD		 2645 igb_ping_all_vfs(adapter);
2646
4b1a9877 2647 /* link state has changed, schedule phy info update */
9d5c8243
AK		 2648 if (!test_bit(__IGB_DOWN, &adapter->state))
2649 mod_timer(&adapter->phy_info_timer,
2650 round_jiffies(jiffies + 2 * HZ));
2651 }
2652 }
2653
2654link_up:
2655 igb_update_stats(adapter);
2656
4b1a9877 2657 hw->mac.tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
9d5c8243 2658 adapter->tpt_old = adapter->stats.tpt;
4b1a9877 2659 hw->mac.collision_delta = adapter->stats.colc - adapter->colc_old;
9d5c8243
AK		 2660 adapter->colc_old = adapter->stats.colc;
2661
2662 adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
2663 adapter->gorc_old = adapter->stats.gorc;
2664 adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
2665 adapter->gotc_old = adapter->stats.gotc;
2666
2667 igb_update_adaptive(&adapter->hw);
2668
2669 if (!netif_carrier_ok(netdev)) {
c493ea45 2670 if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) {
9d5c8243
AK		 2671 /* We've lost link, so the controller stops DMA,
2672 * but we've got queued Tx work that's never going
2673 * to get done, so reset controller to flush Tx.
2674 * (Do the reset outside of interrupt context). */
2675 adapter->tx_timeout_count++;
2676 schedule_work(&adapter->reset_task);
2677 }
2678 }
2679
2680 /* Cause software interrupt to ensure rx ring is cleaned */
7a6ea550
AD		 2681 if (adapter->msix_entries) {
2682 for (i = 0; i < adapter->num_rx_queues; i++)
2683 eics |= adapter->rx_ring[i].eims_value;
2684 wr32(E1000_EICS, eics);
2685 } else {
2686 wr32(E1000_ICS, E1000_ICS_RXDMT0);
2687 }
9d5c8243
AK		 2688
2689 /* Force detection of hung controller every watchdog period */
2690 tx_ring->detect_tx_hung = true;
2691
2692 /* Reset the timer */
2693 if (!test_bit(__IGB_DOWN, &adapter->state))
2694 mod_timer(&adapter->watchdog_timer,
2695 round_jiffies(jiffies + 2 * HZ));
2696}
2697
2698enum latency_range {
2699 lowest_latency = 0,
2700 low_latency = 1,
2701 bulk_latency = 2,
2702 latency_invalid = 255
2703};
2704
2705
6eb5a7f1
AD		 2706/**
2707 * igb_update_ring_itr - update the dynamic ITR value based on packet size
2708 *
2709 * Stores a new ITR value based on strictly on packet size. This
2710 * algorithm is less sophisticated than that used in igb_update_itr,
2711 * due to the difficulty of synchronizing statistics across multiple
2712 * receive rings. The divisors and thresholds used by this fuction
2713 * were determined based on theoretical maximum wire speed and testing
2714 * data, in order to minimize response time while increasing bulk
2715 * throughput.
2716 * This functionality is controlled by the InterruptThrottleRate module
2717 * parameter (see igb_param.c)
2718 * NOTE: This function is called only when operating in a multiqueue
2719 * receive environment.
2720 * @rx_ring: pointer to ring
2721 **/
2722static void igb_update_ring_itr(struct igb_ring *rx_ring)
9d5c8243 2723{
6eb5a7f1
AD		 2724 int new_val = rx_ring->itr_val;
2725 int avg_wire_size = 0;
2726 struct igb_adapter *adapter = rx_ring->adapter;
9d5c8243 2727
6eb5a7f1
AD		 2728 if (!rx_ring->total_packets)
2729 goto clear_counts; /* no packets, so don't do anything */
9d5c8243 2730
6eb5a7f1
AD		 2731 /* For non-gigabit speeds, just fix the interrupt rate at 4000
2732 * ints/sec - ITR timer value of 120 ticks.
2733 */
2734 if (adapter->link_speed != SPEED_1000) {
2735 new_val = 120;
2736 goto set_itr_val;
9d5c8243 2737 }
6eb5a7f1 2738 avg_wire_size = rx_ring->total_bytes / rx_ring->total_packets;
9d5c8243 2739
6eb5a7f1
AD		 2740 /* Add 24 bytes to size to account for CRC, preamble, and gap */
2741 avg_wire_size += 24;
2742
2743 /* Don't starve jumbo frames */
2744 avg_wire_size = min(avg_wire_size, 3000);
9d5c8243 2745
6eb5a7f1
AD		 2746 /* Give a little boost to mid-size frames */
2747 if ((avg_wire_size > 300) && (avg_wire_size < 1200))
2748 new_val = avg_wire_size / 3;
2749 else
2750 new_val = avg_wire_size / 2;
9d5c8243 2751
6eb5a7f1 2752set_itr_val:
9d5c8243
AK		 2753 if (new_val != rx_ring->itr_val) {
2754 rx_ring->itr_val = new_val;
6eb5a7f1 2755 rx_ring->set_itr = 1;
9d5c8243 2756 }
6eb5a7f1
AD		 2757clear_counts:
2758 rx_ring->total_bytes = 0;
2759 rx_ring->total_packets = 0;
9d5c8243
AK		 2760}
2761
2762/**
2763 * igb_update_itr - update the dynamic ITR value based on statistics
2764 * Stores a new ITR value based on packets and byte
2765 * counts during the last interrupt. The advantage of per interrupt
2766 * computation is faster updates and more accurate ITR for the current
2767 * traffic pattern. Constants in this function were computed
2768 * based on theoretical maximum wire speed and thresholds were set based
2769 * on testing data as well as attempting to minimize response time
2770 * while increasing bulk throughput.
2771 * this functionality is controlled by the InterruptThrottleRate module
2772 * parameter (see igb_param.c)
2773 * NOTE: These calculations are only valid when operating in a single-
2774 * queue environment.
2775 * @adapter: pointer to adapter
2776 * @itr_setting: current adapter->itr
2777 * @packets: the number of packets during this measurement interval
2778 * @bytes: the number of bytes during this measurement interval
2779 **/
2780static unsigned int igb_update_itr(struct igb_adapter *adapter, u16 itr_setting,
2781 int packets, int bytes)
2782{
2783 unsigned int retval = itr_setting;
2784
2785 if (packets == 0)
2786 goto update_itr_done;
2787
2788 switch (itr_setting) {
2789 case lowest_latency:
2790 /* handle TSO and jumbo frames */
2791 if (bytes/packets > 8000)
2792 retval = bulk_latency;
2793 else if ((packets < 5) && (bytes > 512))
2794 retval = low_latency;
2795 break;
2796 case low_latency: /* 50 usec aka 20000 ints/s */
2797 if (bytes > 10000) {
2798 /* this if handles the TSO accounting */
2799 if (bytes/packets > 8000) {
2800 retval = bulk_latency;
2801 } else if ((packets < 10) || ((bytes/packets) > 1200)) {
2802 retval = bulk_latency;
2803 } else if ((packets > 35)) {
2804 retval = lowest_latency;
2805 }
2806 } else if (bytes/packets > 2000) {
2807 retval = bulk_latency;
2808 } else if (packets <= 2 && bytes < 512) {
2809 retval = lowest_latency;
2810 }
2811 break;
2812 case bulk_latency: /* 250 usec aka 4000 ints/s */
2813 if (bytes > 25000) {
2814 if (packets > 35)
2815 retval = low_latency;
1e5c3d21 2816 } else if (bytes < 1500) {
9d5c8243
AK		 2817 retval = low_latency;
2818 }
2819 break;
2820 }
2821
2822update_itr_done:
2823 return retval;
2824}
2825
6eb5a7f1 2826static void igb_set_itr(struct igb_adapter *adapter)
9d5c8243
AK		 2827{
2828 u16 current_itr;
2829 u32 new_itr = adapter->itr;
2830
2831 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2832 if (adapter->link_speed != SPEED_1000) {
2833 current_itr = 0;
2834 new_itr = 4000;
2835 goto set_itr_now;
2836 }
2837
2838 adapter->rx_itr = igb_update_itr(adapter,
2839 adapter->rx_itr,
2840 adapter->rx_ring->total_packets,
2841 adapter->rx_ring->total_bytes);
9d5c8243 2842
6eb5a7f1 2843 if (adapter->rx_ring->buddy) {
9d5c8243
AK		 2844 adapter->tx_itr = igb_update_itr(adapter,
2845 adapter->tx_itr,
2846 adapter->tx_ring->total_packets,
2847 adapter->tx_ring->total_bytes);
9d5c8243
AK		 2848 current_itr = max(adapter->rx_itr, adapter->tx_itr);
2849 } else {
2850 current_itr = adapter->rx_itr;
2851 }
2852
6eb5a7f1 2853 /* conservative mode (itr 3) eliminates the lowest_latency setting */
73cd78f1 2854 if (adapter->itr_setting == 3 && current_itr == lowest_latency)
6eb5a7f1
AD		 2855 current_itr = low_latency;
2856
9d5c8243
AK		 2857 switch (current_itr) {
2858 /* counts and packets in update_itr are dependent on these numbers */
2859 case lowest_latency:
2860 new_itr = 70000;
2861 break;
2862 case low_latency:
2863 new_itr = 20000; /* aka hwitr = ~200 */
2864 break;
2865 case bulk_latency:
2866 new_itr = 4000;
2867 break;
2868 default:
2869 break;
2870 }
2871
2872set_itr_now:
6eb5a7f1
AD		 2873 adapter->rx_ring->total_bytes = 0;
2874 adapter->rx_ring->total_packets = 0;
2875 if (adapter->rx_ring->buddy) {
2876 adapter->rx_ring->buddy->total_bytes = 0;
2877 adapter->rx_ring->buddy->total_packets = 0;
2878 }
2879
9d5c8243
AK		 2880 if (new_itr != adapter->itr) {
2881 /* this attempts to bias the interrupt rate towards Bulk
2882 * by adding intermediate steps when interrupt rate is
2883 * increasing */
2884 new_itr = new_itr > adapter->itr ?
2885 min(adapter->itr + (new_itr >> 2), new_itr) :
2886 new_itr;
2887 /* Don't write the value here; it resets the adapter's
2888 * internal timer, and causes us to delay far longer than
2889 * we should between interrupts. Instead, we write the ITR
2890 * value at the beginning of the next interrupt so the timing
2891 * ends up being correct.
2892 */
2893 adapter->itr = new_itr;
6eb5a7f1
AD		 2894 adapter->rx_ring->itr_val = 1000000000 / (new_itr * 256);
2895 adapter->rx_ring->set_itr = 1;
9d5c8243
AK		 2896 }
2897
2898 return;
2899}
2900
2901
2902#define IGB_TX_FLAGS_CSUM 0x00000001
2903#define IGB_TX_FLAGS_VLAN 0x00000002
2904#define IGB_TX_FLAGS_TSO 0x00000004
2905#define IGB_TX_FLAGS_IPV4 0x00000008
33af6bcc 2906#define IGB_TX_FLAGS_TSTAMP 0x00000010
9d5c8243
AK		 2907#define IGB_TX_FLAGS_VLAN_MASK 0xffff0000
2908#define IGB_TX_FLAGS_VLAN_SHIFT 16
2909
2910static inline int igb_tso_adv(struct igb_adapter *adapter,
2911 struct igb_ring *tx_ring,
2912 struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
2913{
2914 struct e1000_adv_tx_context_desc *context_desc;
2915 unsigned int i;
2916 int err;
2917 struct igb_buffer *buffer_info;
2918 u32 info = 0, tu_cmd = 0;
2919 u32 mss_l4len_idx, l4len;
2920 *hdr_len = 0;
2921
2922 if (skb_header_cloned(skb)) {
2923 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2924 if (err)
2925 return err;
2926 }
2927
2928 l4len = tcp_hdrlen(skb);
2929 *hdr_len += l4len;
2930
2931 if (skb->protocol == htons(ETH_P_IP)) {
2932 struct iphdr *iph = ip_hdr(skb);
2933 iph->tot_len = 0;
2934 iph->check = 0;
2935 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2936 iph->daddr, 0,
2937 IPPROTO_TCP,
2938 0);
2939 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
2940 ipv6_hdr(skb)->payload_len = 0;
2941 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2942 &ipv6_hdr(skb)->daddr,
2943 0, IPPROTO_TCP, 0);
2944 }
2945
2946 i = tx_ring->next_to_use;
2947
2948 buffer_info = &tx_ring->buffer_info[i];
2949 context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
2950 /* VLAN MACLEN IPLEN */
2951 if (tx_flags & IGB_TX_FLAGS_VLAN)
2952 info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
2953 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2954 *hdr_len += skb_network_offset(skb);
2955 info |= skb_network_header_len(skb);
2956 *hdr_len += skb_network_header_len(skb);
2957 context_desc->vlan_macip_lens = cpu_to_le32(info);
2958
2959 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
2960 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2961
2962 if (skb->protocol == htons(ETH_P_IP))
2963 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2964 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2965
2966 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2967
2968 /* MSS L4LEN IDX */
2969 mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
2970 mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
2971
73cd78f1 2972 /* For 82575, context index must be unique per ring. */
7dfc16fa
AD		 2973 if (adapter->flags & IGB_FLAG_NEED_CTX_IDX)
2974 mss_l4len_idx |= tx_ring->queue_index << 4;
9d5c8243
AK		 2975
2976 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
2977 context_desc->seqnum_seed = 0;
2978
2979 buffer_info->time_stamp = jiffies;
0e014cb1 2980 buffer_info->next_to_watch = i;
9d5c8243
AK		 2981 buffer_info->dma = 0;
2982 i++;
2983 if (i == tx_ring->count)
2984 i = 0;
2985
2986 tx_ring->next_to_use = i;
2987
2988 return true;
2989}
2990
2991static inline bool igb_tx_csum_adv(struct igb_adapter *adapter,
2992 struct igb_ring *tx_ring,
2993 struct sk_buff *skb, u32 tx_flags)
2994{
2995 struct e1000_adv_tx_context_desc *context_desc;
2996 unsigned int i;
2997 struct igb_buffer *buffer_info;
2998 u32 info = 0, tu_cmd = 0;
2999
3000 if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
3001 (tx_flags & IGB_TX_FLAGS_VLAN)) {
3002 i = tx_ring->next_to_use;
3003 buffer_info = &tx_ring->buffer_info[i];
3004 context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
3005
3006 if (tx_flags & IGB_TX_FLAGS_VLAN)
3007 info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
3008 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
3009 if (skb->ip_summed == CHECKSUM_PARTIAL)
3010 info |= skb_network_header_len(skb);
3011
3012 context_desc->vlan_macip_lens = cpu_to_le32(info);
3013
3014 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
3015
3016 if (skb->ip_summed == CHECKSUM_PARTIAL) {
fa4a7ef3
AJ		 3017 __be16 protocol;
3018
3019 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3020 const struct vlan_ethhdr *vhdr =
3021 (const struct vlan_ethhdr*)skb->data;
3022
3023 protocol = vhdr->h_vlan_encapsulated_proto;
3024 } else {
3025 protocol = skb->protocol;
3026 }
3027
3028 switch (protocol) {
09640e63 3029 case cpu_to_be16(ETH_P_IP):
9d5c8243 3030 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
44b0cda3
MW		 3031 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
3032 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
3033 break;
09640e63 3034 case cpu_to_be16(ETH_P_IPV6):
44b0cda3
MW		 3035 /* XXX what about other V6 headers?? */
3036 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3037 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
3038 break;
3039 default:
3040 if (unlikely(net_ratelimit()))
3041 dev_warn(&adapter->pdev->dev,
3042 "partial checksum but proto=%x!\n",
3043 skb->protocol);
3044 break;
3045 }
9d5c8243
AK		 3046 }
3047
3048 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
3049 context_desc->seqnum_seed = 0;
7dfc16fa
AD		 3050 if (adapter->flags & IGB_FLAG_NEED_CTX_IDX)
3051 context_desc->mss_l4len_idx =
3052 cpu_to_le32(tx_ring->queue_index << 4);
265de409
AD		 3053 else
3054 context_desc->mss_l4len_idx = 0;
9d5c8243
AK		 3055
3056 buffer_info->time_stamp = jiffies;
0e014cb1 3057 buffer_info->next_to_watch = i;
9d5c8243
AK		 3058 buffer_info->dma = 0;
3059
3060 i++;
3061 if (i == tx_ring->count)
3062 i = 0;
3063 tx_ring->next_to_use = i;
3064
3065 return true;
3066 }
9d5c8243
AK		 3067 return false;
3068}
3069
3070#define IGB_MAX_TXD_PWR 16
3071#define IGB_MAX_DATA_PER_TXD (1<<IGB_MAX_TXD_PWR)
3072
3073static inline int igb_tx_map_adv(struct igb_adapter *adapter,
0e014cb1
AD		 3074 struct igb_ring *tx_ring, struct sk_buff *skb,
3075 unsigned int first)
9d5c8243
AK		 3076{
3077 struct igb_buffer *buffer_info;
3078 unsigned int len = skb_headlen(skb);
3079 unsigned int count = 0, i;
3080 unsigned int f;
3081
3082 i = tx_ring->next_to_use;
3083
3084 buffer_info = &tx_ring->buffer_info[i];
3085 BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
3086 buffer_info->length = len;
3087 /* set time_stamp *before* dma to help avoid a possible race */
3088 buffer_info->time_stamp = jiffies;
0e014cb1 3089 buffer_info->next_to_watch = i;
9d5c8243
AK		 3090 buffer_info->dma = pci_map_single(adapter->pdev, skb->data, len,
3091 PCI_DMA_TODEVICE);
3092 count++;
3093 i++;
3094 if (i == tx_ring->count)
3095 i = 0;
3096
3097 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
3098 struct skb_frag_struct *frag;
3099
3100 frag = &skb_shinfo(skb)->frags[f];
3101 len = frag->size;
3102
3103 buffer_info = &tx_ring->buffer_info[i];
3104 BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
3105 buffer_info->length = len;
3106 buffer_info->time_stamp = jiffies;
0e014cb1 3107 buffer_info->next_to_watch = i;
9d5c8243
AK		 3108 buffer_info->dma = pci_map_page(adapter->pdev,
3109 frag->page,
3110 frag->page_offset,
3111 len,
3112 PCI_DMA_TODEVICE);
3113
3114 count++;
3115 i++;
3116 if (i == tx_ring->count)
3117 i = 0;
3118 }
3119
0e014cb1 3120 i = ((i == 0) ? tx_ring->count - 1 : i - 1);
9d5c8243 3121 tx_ring->buffer_info[i].skb = skb;
0e014cb1 3122 tx_ring->buffer_info[first].next_to_watch = i;
9d5c8243
AK		 3123
3124 return count;
3125}
3126
3127static inline void igb_tx_queue_adv(struct igb_adapter *adapter,
3128 struct igb_ring *tx_ring,
3129 int tx_flags, int count, u32 paylen,
3130 u8 hdr_len)
3131{
3132 union e1000_adv_tx_desc *tx_desc = NULL;
3133 struct igb_buffer *buffer_info;
3134 u32 olinfo_status = 0, cmd_type_len;
3135 unsigned int i;
3136
3137 cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
3138 E1000_ADVTXD_DCMD_DEXT);
3139
3140 if (tx_flags & IGB_TX_FLAGS_VLAN)
3141 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
3142
33af6bcc
PO		 3143 if (tx_flags & IGB_TX_FLAGS_TSTAMP)
3144 cmd_type_len |= E1000_ADVTXD_MAC_TSTAMP;
3145
9d5c8243
AK		 3146 if (tx_flags & IGB_TX_FLAGS_TSO) {
3147 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
3148
3149 /* insert tcp checksum */
3150 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
3151
3152 /* insert ip checksum */
3153 if (tx_flags & IGB_TX_FLAGS_IPV4)
3154 olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
3155
3156 } else if (tx_flags & IGB_TX_FLAGS_CSUM) {
3157 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
3158 }
3159
7dfc16fa
AD		 3160 if ((adapter->flags & IGB_FLAG_NEED_CTX_IDX) &&
3161 (tx_flags & (IGB_TX_FLAGS_CSUM | IGB_TX_FLAGS_TSO |
3162 IGB_TX_FLAGS_VLAN)))
661086df 3163 olinfo_status |= tx_ring->queue_index << 4;
9d5c8243
AK		 3164
3165 olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
3166
3167 i = tx_ring->next_to_use;
3168 while (count--) {
3169 buffer_info = &tx_ring->buffer_info[i];
3170 tx_desc = E1000_TX_DESC_ADV(*tx_ring, i);
3171 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
3172 tx_desc->read.cmd_type_len =
3173 cpu_to_le32(cmd_type_len | buffer_info->length);
3174 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
3175 i++;
3176 if (i == tx_ring->count)
3177 i = 0;
3178 }
3179
3180 tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
3181 /* Force memory writes to complete before letting h/w
3182 * know there are new descriptors to fetch. (Only
3183 * applicable for weak-ordered memory model archs,
3184 * such as IA-64). */
3185 wmb();
3186
3187 tx_ring->next_to_use = i;
3188 writel(i, adapter->hw.hw_addr + tx_ring->tail);
3189 /* we need this if more than one processor can write to our tail
3190 * at a time, it syncronizes IO on IA64/Altix systems */
3191 mmiowb();
3192}
3193
3194static int __igb_maybe_stop_tx(struct net_device *netdev,
3195 struct igb_ring *tx_ring, int size)
3196{
3197 struct igb_adapter *adapter = netdev_priv(netdev);
3198
661086df 3199 netif_stop_subqueue(netdev, tx_ring->queue_index);
661086df 3200
9d5c8243
AK		 3201 /* Herbert's original patch had:
3202 * smp_mb__after_netif_stop_queue();
3203 * but since that doesn't exist yet, just open code it. */
3204 smp_mb();
3205
3206 /* We need to check again in a case another CPU has just
3207 * made room available. */
c493ea45 3208 if (igb_desc_unused(tx_ring) < size)
9d5c8243
AK		 3209 return -EBUSY;
3210
3211 /* A reprieve! */
661086df 3212 netif_wake_subqueue(netdev, tx_ring->queue_index);
9d5c8243
AK		 3213 ++adapter->restart_queue;
3214 return 0;
3215}
3216
3217static int igb_maybe_stop_tx(struct net_device *netdev,
3218 struct igb_ring *tx_ring, int size)
3219{
c493ea45 3220 if (igb_desc_unused(tx_ring) >= size)
9d5c8243
AK		 3221 return 0;
3222 return __igb_maybe_stop_tx(netdev, tx_ring, size);
3223}
3224
9d5c8243
AK		 3225static int igb_xmit_frame_ring_adv(struct sk_buff *skb,
3226 struct net_device *netdev,
3227 struct igb_ring *tx_ring)
3228{
3229 struct igb_adapter *adapter = netdev_priv(netdev);
0e014cb1 3230 unsigned int first;
9d5c8243 3231 unsigned int tx_flags = 0;
9d5c8243
AK		 3232 u8 hdr_len = 0;
3233 int tso = 0;
33af6bcc 3234 union skb_shared_tx *shtx;
9d5c8243 3235
9d5c8243
AK		 3236 if (test_bit(__IGB_DOWN, &adapter->state)) {
3237 dev_kfree_skb_any(skb);
3238 return NETDEV_TX_OK;
3239 }
3240
3241 if (skb->len <= 0) {
3242 dev_kfree_skb_any(skb);
3243 return NETDEV_TX_OK;
3244 }
3245
9d5c8243
AK		 3246 /* need: 1 descriptor per page,
3247 * + 2 desc gap to keep tail from touching head,
3248 * + 1 desc for skb->data,
3249 * + 1 desc for context descriptor,
3250 * otherwise try next time */
3251 if (igb_maybe_stop_tx(netdev, tx_ring, skb_shinfo(skb)->nr_frags + 4)) {
3252 /* this is a hard error */
9d5c8243
AK		 3253 return NETDEV_TX_BUSY;
3254 }
33af6bcc
PO		 3255
3256 /*
3257 * TODO: check that there currently is no other packet with
3258 * time stamping in the queue
3259 *
3260 * When doing time stamping, keep the connection to the socket
3261 * a while longer: it is still needed by skb_hwtstamp_tx(),
3262 * called either in igb_tx_hwtstamp() or by our caller when
3263 * doing software time stamping.
3264 */
3265 shtx = skb_tx(skb);
3266 if (unlikely(shtx->hardware)) {
3267 shtx->in_progress = 1;
3268 tx_flags |= IGB_TX_FLAGS_TSTAMP;
33af6bcc 3269 }
9d5c8243
AK		 3270
3271 if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
3272 tx_flags |= IGB_TX_FLAGS_VLAN;
3273 tx_flags |= (vlan_tx_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
3274 }
3275
661086df
PWJ		 3276 if (skb->protocol == htons(ETH_P_IP))
3277 tx_flags |= IGB_TX_FLAGS_IPV4;
3278
0e014cb1 3279 first = tx_ring->next_to_use;
9d5c8243
AK		 3280 tso = skb_is_gso(skb) ? igb_tso_adv(adapter, tx_ring, skb, tx_flags,
3281 &hdr_len) : 0;
3282
3283 if (tso < 0) {
3284 dev_kfree_skb_any(skb);
9d5c8243
AK		 3285 return NETDEV_TX_OK;
3286 }
3287
3288 if (tso)
3289 tx_flags |= IGB_TX_FLAGS_TSO;
bc1cbd34
AD		 3290 else if (igb_tx_csum_adv(adapter, tx_ring, skb, tx_flags) &&
3291 (skb->ip_summed == CHECKSUM_PARTIAL))
3292 tx_flags |= IGB_TX_FLAGS_CSUM;
9d5c8243 3293
9d5c8243 3294 igb_tx_queue_adv(adapter, tx_ring, tx_flags,
0e014cb1 3295 igb_tx_map_adv(adapter, tx_ring, skb, first),
9d5c8243
AK		 3296 skb->len, hdr_len);
3297
3298 netdev->trans_start = jiffies;
3299
3300 /* Make sure there is space in the ring for the next send. */
3301 igb_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 4);
3302
9d5c8243
AK		 3303 return NETDEV_TX_OK;
3304}
3305
3306static int igb_xmit_frame_adv(struct sk_buff *skb, struct net_device *netdev)
3307{
3308 struct igb_adapter *adapter = netdev_priv(netdev);
661086df
PWJ		 3309 struct igb_ring *tx_ring;
3310
661086df 3311 int r_idx = 0;
1bfaf07b 3312 r_idx = skb->queue_mapping & (IGB_ABS_MAX_TX_QUEUES - 1);
661086df 3313 tx_ring = adapter->multi_tx_table[r_idx];
9d5c8243
AK		 3314
3315 /* This goes back to the question of how to logically map a tx queue
3316 * to a flow. Right now, performance is impacted slightly negatively
3317 * if using multiple tx queues. If the stack breaks away from a
3318 * single qdisc implementation, we can look at this again. */
3319 return (igb_xmit_frame_ring_adv(skb, netdev, tx_ring));
3320}
3321
3322/**
3323 * igb_tx_timeout - Respond to a Tx Hang
3324 * @netdev: network interface device structure
3325 **/
3326static void igb_tx_timeout(struct net_device *netdev)
3327{
3328 struct igb_adapter *adapter = netdev_priv(netdev);
3329 struct e1000_hw *hw = &adapter->hw;
3330
3331 /* Do the reset outside of interrupt context */
3332 adapter->tx_timeout_count++;
3333 schedule_work(&adapter->reset_task);
265de409
AD		 3334 wr32(E1000_EICS,
3335 (adapter->eims_enable_mask & ~adapter->eims_other));
9d5c8243
AK		 3336}
3337
3338static void igb_reset_task(struct work_struct *work)
3339{
3340 struct igb_adapter *adapter;
3341 adapter = container_of(work, struct igb_adapter, reset_task);
3342
3343 igb_reinit_locked(adapter);
3344}
3345
3346/**
3347 * igb_get_stats - Get System Network Statistics
3348 * @netdev: network interface device structure
3349 *
3350 * Returns the address of the device statistics structure.
3351 * The statistics are actually updated from the timer callback.
3352 **/
73cd78f1 3353static struct net_device_stats *igb_get_stats(struct net_device *netdev)
9d5c8243
AK		 3354{
3355 struct igb_adapter *adapter = netdev_priv(netdev);
3356
3357 /* only return the current stats */
3358 return &adapter->net_stats;
3359}
3360
3361/**
3362 * igb_change_mtu - Change the Maximum Transfer Unit
3363 * @netdev: network interface device structure
3364 * @new_mtu: new value for maximum frame size
3365 *
3366 * Returns 0 on success, negative on failure
3367 **/
3368static int igb_change_mtu(struct net_device *netdev, int new_mtu)
3369{
3370 struct igb_adapter *adapter = netdev_priv(netdev);
3371 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3372
3373 if ((max_frame < ETH_ZLEN + ETH_FCS_LEN) ||
3374 (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3375 dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
3376 return -EINVAL;
3377 }
3378
9d5c8243
AK		 3379 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
3380 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
3381 return -EINVAL;
3382 }
3383
3384 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
3385 msleep(1);
73cd78f1 3386
9d5c8243
AK		 3387 /* igb_down has a dependency on max_frame_size */
3388 adapter->max_frame_size = max_frame;
3389 if (netif_running(netdev))
3390 igb_down(adapter);
3391
3392 /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3393 * means we reserve 2 more, this pushes us to allocate from the next
3394 * larger slab size.
3395 * i.e. RXBUFFER_2048 --> size-4096 slab
3396 */
3397
3398 if (max_frame <= IGB_RXBUFFER_256)
3399 adapter->rx_buffer_len = IGB_RXBUFFER_256;
3400 else if (max_frame <= IGB_RXBUFFER_512)
3401 adapter->rx_buffer_len = IGB_RXBUFFER_512;
3402 else if (max_frame <= IGB_RXBUFFER_1024)
3403 adapter->rx_buffer_len = IGB_RXBUFFER_1024;
3404 else if (max_frame <= IGB_RXBUFFER_2048)
3405 adapter->rx_buffer_len = IGB_RXBUFFER_2048;
3406 else
bf36c1a0
AD		 3407#if (PAGE_SIZE / 2) > IGB_RXBUFFER_16384
3408 adapter->rx_buffer_len = IGB_RXBUFFER_16384;
3409#else
3410 adapter->rx_buffer_len = PAGE_SIZE / 2;
3411#endif
e1739522
AD		 3412
3413 /* if sr-iov is enabled we need to force buffer size to 1K or larger */
3414 if (adapter->vfs_allocated_count &&
3415 (adapter->rx_buffer_len < IGB_RXBUFFER_1024))
3416 adapter->rx_buffer_len = IGB_RXBUFFER_1024;
3417
9d5c8243
AK		 3418 /* adjust allocation if LPE protects us, and we aren't using SBP */
3419 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
3420 (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE))
3421 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3422
3423 dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
3424 netdev->mtu, new_mtu);
3425 netdev->mtu = new_mtu;
3426
3427 if (netif_running(netdev))
3428 igb_up(adapter);
3429 else
3430 igb_reset(adapter);
3431
3432 clear_bit(__IGB_RESETTING, &adapter->state);
3433
3434 return 0;
3435}
3436
3437/**
3438 * igb_update_stats - Update the board statistics counters
3439 * @adapter: board private structure
3440 **/
3441
3442void igb_update_stats(struct igb_adapter *adapter)
3443{
3444 struct e1000_hw *hw = &adapter->hw;
3445 struct pci_dev *pdev = adapter->pdev;
3446 u16 phy_tmp;
3447
3448#define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3449
3450 /*
3451 * Prevent stats update while adapter is being reset, or if the pci
3452 * connection is down.
3453 */
3454 if (adapter->link_speed == 0)
3455 return;
3456 if (pci_channel_offline(pdev))
3457 return;
3458
3459 adapter->stats.crcerrs += rd32(E1000_CRCERRS);
3460 adapter->stats.gprc += rd32(E1000_GPRC);
3461 adapter->stats.gorc += rd32(E1000_GORCL);
3462 rd32(E1000_GORCH); /* clear GORCL */
3463 adapter->stats.bprc += rd32(E1000_BPRC);
3464 adapter->stats.mprc += rd32(E1000_MPRC);
3465 adapter->stats.roc += rd32(E1000_ROC);
3466
3467 adapter->stats.prc64 += rd32(E1000_PRC64);
3468 adapter->stats.prc127 += rd32(E1000_PRC127);
3469 adapter->stats.prc255 += rd32(E1000_PRC255);
3470 adapter->stats.prc511 += rd32(E1000_PRC511);
3471 adapter->stats.prc1023 += rd32(E1000_PRC1023);
3472 adapter->stats.prc1522 += rd32(E1000_PRC1522);
3473 adapter->stats.symerrs += rd32(E1000_SYMERRS);
3474 adapter->stats.sec += rd32(E1000_SEC);
3475
3476 adapter->stats.mpc += rd32(E1000_MPC);
3477 adapter->stats.scc += rd32(E1000_SCC);
3478 adapter->stats.ecol += rd32(E1000_ECOL);
3479 adapter->stats.mcc += rd32(E1000_MCC);
3480 adapter->stats.latecol += rd32(E1000_LATECOL);
3481 adapter->stats.dc += rd32(E1000_DC);
3482 adapter->stats.rlec += rd32(E1000_RLEC);
3483 adapter->stats.xonrxc += rd32(E1000_XONRXC);
3484 adapter->stats.xontxc += rd32(E1000_XONTXC);
3485 adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
3486 adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
3487 adapter->stats.fcruc += rd32(E1000_FCRUC);
3488 adapter->stats.gptc += rd32(E1000_GPTC);
3489 adapter->stats.gotc += rd32(E1000_GOTCL);
3490 rd32(E1000_GOTCH); /* clear GOTCL */
3491 adapter->stats.rnbc += rd32(E1000_RNBC);
3492 adapter->stats.ruc += rd32(E1000_RUC);
3493 adapter->stats.rfc += rd32(E1000_RFC);
3494 adapter->stats.rjc += rd32(E1000_RJC);
3495 adapter->stats.tor += rd32(E1000_TORH);
3496 adapter->stats.tot += rd32(E1000_TOTH);
3497 adapter->stats.tpr += rd32(E1000_TPR);
3498
3499 adapter->stats.ptc64 += rd32(E1000_PTC64);
3500 adapter->stats.ptc127 += rd32(E1000_PTC127);
3501 adapter->stats.ptc255 += rd32(E1000_PTC255);
3502 adapter->stats.ptc511 += rd32(E1000_PTC511);
3503 adapter->stats.ptc1023 += rd32(E1000_PTC1023);
3504 adapter->stats.ptc1522 += rd32(E1000_PTC1522);
3505
3506 adapter->stats.mptc += rd32(E1000_MPTC);
3507 adapter->stats.bptc += rd32(E1000_BPTC);
3508
3509 /* used for adaptive IFS */
3510
3511 hw->mac.tx_packet_delta = rd32(E1000_TPT);
3512 adapter->stats.tpt += hw->mac.tx_packet_delta;
3513 hw->mac.collision_delta = rd32(E1000_COLC);
3514 adapter->stats.colc += hw->mac.collision_delta;
3515
3516 adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
3517 adapter->stats.rxerrc += rd32(E1000_RXERRC);
3518 adapter->stats.tncrs += rd32(E1000_TNCRS);
3519 adapter->stats.tsctc += rd32(E1000_TSCTC);
3520 adapter->stats.tsctfc += rd32(E1000_TSCTFC);
3521
3522 adapter->stats.iac += rd32(E1000_IAC);
3523 adapter->stats.icrxoc += rd32(E1000_ICRXOC);
3524 adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
3525 adapter->stats.icrxatc += rd32(E1000_ICRXATC);
3526 adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
3527 adapter->stats.ictxatc += rd32(E1000_ICTXATC);
3528 adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
3529 adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
3530 adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
3531
3532 /* Fill out the OS statistics structure */
3533 adapter->net_stats.multicast = adapter->stats.mprc;
3534 adapter->net_stats.collisions = adapter->stats.colc;
3535
3536 /* Rx Errors */
3537
3538 /* RLEC on some newer hardware can be incorrect so build
3539 * our own version based on RUC and ROC */
3540 adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3541 adapter->stats.crcerrs + adapter->stats.algnerrc +
3542 adapter->stats.ruc + adapter->stats.roc +
3543 adapter->stats.cexterr;
3544 adapter->net_stats.rx_length_errors = adapter->stats.ruc +
3545 adapter->stats.roc;
3546 adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3547 adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3548 adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3549
3550 /* Tx Errors */
3551 adapter->net_stats.tx_errors = adapter->stats.ecol +
3552 adapter->stats.latecol;
3553 adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3554 adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3555 adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3556
3557 /* Tx Dropped needs to be maintained elsewhere */
3558
3559 /* Phy Stats */
3560 if (hw->phy.media_type == e1000_media_type_copper) {
3561 if ((adapter->link_speed == SPEED_1000) &&
73cd78f1 3562 (!igb_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
9d5c8243
AK		 3563 phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3564 adapter->phy_stats.idle_errors += phy_tmp;
3565 }
3566 }
3567
3568 /* Management Stats */
3569 adapter->stats.mgptc += rd32(E1000_MGTPTC);
3570 adapter->stats.mgprc += rd32(E1000_MGTPRC);
3571 adapter->stats.mgpdc += rd32(E1000_MGTPDC);
3572}
3573
9d5c8243
AK		 3574static irqreturn_t igb_msix_other(int irq, void *data)
3575{
3576 struct net_device *netdev = data;
3577 struct igb_adapter *adapter = netdev_priv(netdev);
3578 struct e1000_hw *hw = &adapter->hw;
844290e5 3579 u32 icr = rd32(E1000_ICR);
9d5c8243 3580
844290e5 3581 /* reading ICR causes bit 31 of EICR to be cleared */
dda0e083
AD		 3582
3583 if(icr & E1000_ICR_DOUTSYNC) {
3584 /* HW is reporting DMA is out of sync */
3585 adapter->stats.doosync++;
3586 }
eebbbdba 3587
4ae196df
AD		 3588 /* Check for a mailbox event */
3589 if (icr & E1000_ICR_VMMB)
3590 igb_msg_task(adapter);
3591
3592 if (icr & E1000_ICR_LSC) {
3593 hw->mac.get_link_status = 1;
3594 /* guard against interrupt when we're going down */
3595 if (!test_bit(__IGB_DOWN, &adapter->state))
3596 mod_timer(&adapter->watchdog_timer, jiffies + 1);
3597 }
3598
3599 wr32(E1000_IMS, E1000_IMS_LSC | E1000_IMS_DOUTSYNC | E1000_IMS_VMMB);
844290e5 3600 wr32(E1000_EIMS, adapter->eims_other);
9d5c8243
AK		 3601
3602 return IRQ_HANDLED;
3603}
3604
3605static irqreturn_t igb_msix_tx(int irq, void *data)
3606{
3607 struct igb_ring *tx_ring = data;
3608 struct igb_adapter *adapter = tx_ring->adapter;
3609 struct e1000_hw *hw = &adapter->hw;
3610
421e02f0 3611#ifdef CONFIG_IGB_DCA
7dfc16fa 3612 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
fe4506b6
JC		 3613 igb_update_tx_dca(tx_ring);
3614#endif
73cd78f1 3615
9d5c8243
AK		 3616 tx_ring->total_bytes = 0;
3617 tx_ring->total_packets = 0;
661086df
PWJ		 3618
3619 /* auto mask will automatically reenable the interrupt when we write
3620 * EICS */
3b644cf6 3621 if (!igb_clean_tx_irq(tx_ring))
9d5c8243
AK		 3622 /* Ring was not completely cleaned, so fire another interrupt */
3623 wr32(E1000_EICS, tx_ring->eims_value);
661086df 3624 else
9d5c8243 3625 wr32(E1000_EIMS, tx_ring->eims_value);
661086df 3626
9d5c8243
AK		 3627 return IRQ_HANDLED;
3628}
3629
6eb5a7f1
AD		 3630static void igb_write_itr(struct igb_ring *ring)
3631{
3632 struct e1000_hw *hw = &ring->adapter->hw;
3633 if ((ring->adapter->itr_setting & 3) && ring->set_itr) {
3634 switch (hw->mac.type) {
3635 case e1000_82576:
73cd78f1 3636 wr32(ring->itr_register, ring->itr_val |
6eb5a7f1
AD		 3637 0x80000000);
3638 break;
3639 default:
73cd78f1 3640 wr32(ring->itr_register, ring->itr_val |
6eb5a7f1
AD		 3641 (ring->itr_val << 16));
3642 break;
3643 }
3644 ring->set_itr = 0;
3645 }
3646}
3647
9d5c8243
AK		 3648static irqreturn_t igb_msix_rx(int irq, void *data)
3649{
3650 struct igb_ring *rx_ring = data;
9d5c8243 3651
844290e5
PW		 3652 /* Write the ITR value calculated at the end of the
3653 * previous interrupt.
3654 */
9d5c8243 3655
6eb5a7f1 3656 igb_write_itr(rx_ring);
9d5c8243 3657
288379f0
BH		 3658 if (napi_schedule_prep(&rx_ring->napi))
3659 __napi_schedule(&rx_ring->napi);
844290e5 3660
421e02f0 3661#ifdef CONFIG_IGB_DCA
8d253320 3662 if (rx_ring->adapter->flags & IGB_FLAG_DCA_ENABLED)
fe4506b6
JC		 3663 igb_update_rx_dca(rx_ring);
3664#endif
3665 return IRQ_HANDLED;
3666}
3667
421e02f0 3668#ifdef CONFIG_IGB_DCA
fe4506b6
JC		 3669static void igb_update_rx_dca(struct igb_ring *rx_ring)
3670{
3671 u32 dca_rxctrl;
3672 struct igb_adapter *adapter = rx_ring->adapter;
3673 struct e1000_hw *hw = &adapter->hw;
3674 int cpu = get_cpu();
26bc19ec 3675 int q = rx_ring->reg_idx;
fe4506b6
JC		 3676
3677 if (rx_ring->cpu != cpu) {
3678 dca_rxctrl = rd32(E1000_DCA_RXCTRL(q));
2d064c06
AD		 3679 if (hw->mac.type == e1000_82576) {
3680 dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK_82576;
92be7917 3681 dca_rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu) <<
2d064c06
AD		 3682 E1000_DCA_RXCTRL_CPUID_SHIFT;
3683 } else {
3684 dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK;
92be7917 3685 dca_rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
2d064c06 3686 }
fe4506b6
JC		 3687 dca_rxctrl |= E1000_DCA_RXCTRL_DESC_DCA_EN;
3688 dca_rxctrl |= E1000_DCA_RXCTRL_HEAD_DCA_EN;
3689 dca_rxctrl |= E1000_DCA_RXCTRL_DATA_DCA_EN;
3690 wr32(E1000_DCA_RXCTRL(q), dca_rxctrl);
3691 rx_ring->cpu = cpu;
3692 }
3693 put_cpu();
3694}
3695
3696static void igb_update_tx_dca(struct igb_ring *tx_ring)
3697{
3698 u32 dca_txctrl;
3699 struct igb_adapter *adapter = tx_ring->adapter;
3700 struct e1000_hw *hw = &adapter->hw;
3701 int cpu = get_cpu();
26bc19ec 3702 int q = tx_ring->reg_idx;
fe4506b6
JC		 3703
3704 if (tx_ring->cpu != cpu) {
3705 dca_txctrl = rd32(E1000_DCA_TXCTRL(q));
2d064c06
AD		 3706 if (hw->mac.type == e1000_82576) {
3707 dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK_82576;
92be7917 3708 dca_txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu) <<
2d064c06
AD		 3709 E1000_DCA_TXCTRL_CPUID_SHIFT;
3710 } else {
3711 dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK;
92be7917 3712 dca_txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
2d064c06 3713 }
fe4506b6
JC		 3714 dca_txctrl |= E1000_DCA_TXCTRL_DESC_DCA_EN;
3715 wr32(E1000_DCA_TXCTRL(q), dca_txctrl);
3716 tx_ring->cpu = cpu;
3717 }
3718 put_cpu();
3719}
3720
3721static void igb_setup_dca(struct igb_adapter *adapter)
3722{
3723 int i;
3724
7dfc16fa 3725 if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
fe4506b6
JC		 3726 return;
3727
3728 for (i = 0; i < adapter->num_tx_queues; i++) {
3729 adapter->tx_ring[i].cpu = -1;
3730 igb_update_tx_dca(&adapter->tx_ring[i]);
3731 }
3732 for (i = 0; i < adapter->num_rx_queues; i++) {
3733 adapter->rx_ring[i].cpu = -1;
3734 igb_update_rx_dca(&adapter->rx_ring[i]);
3735 }
3736}
3737
3738static int __igb_notify_dca(struct device *dev, void *data)
3739{
3740 struct net_device *netdev = dev_get_drvdata(dev);
3741 struct igb_adapter *adapter = netdev_priv(netdev);
3742 struct e1000_hw *hw = &adapter->hw;
3743 unsigned long event = *(unsigned long *)data;
3744
3745 switch (event) {
3746 case DCA_PROVIDER_ADD:
3747 /* if already enabled, don't do it again */
7dfc16fa 3748 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
fe4506b6 3749 break;
fe4506b6
JC		 3750 /* Always use CB2 mode, difference is masked
3751 * in the CB driver. */
cbd347ad 3752 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
fe4506b6 3753 if (dca_add_requester(dev) == 0) {
bbd98fe4 3754 adapter->flags |= IGB_FLAG_DCA_ENABLED;
fe4506b6
JC		 3755 dev_info(&adapter->pdev->dev, "DCA enabled\n");
3756 igb_setup_dca(adapter);
3757 break;
3758 }
3759 /* Fall Through since DCA is disabled. */
3760 case DCA_PROVIDER_REMOVE:
7dfc16fa 3761 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
fe4506b6
JC		 3762 /* without this a class_device is left
3763 * hanging around in the sysfs model */
3764 dca_remove_requester(dev);
3765 dev_info(&adapter->pdev->dev, "DCA disabled\n");
7dfc16fa 3766 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
cbd347ad 3767 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
fe4506b6
JC		 3768 }
3769 break;
3770 }
bbd98fe4 3771
fe4506b6 3772 return 0;
9d5c8243
AK		 3773}
3774
fe4506b6
JC		 3775static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
3776 void *p)
3777{
3778 int ret_val;
3779
3780 ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
3781 __igb_notify_dca);
3782
3783 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
3784}
421e02f0 3785#endif /* CONFIG_IGB_DCA */
9d5c8243 3786
4ae196df
AD		 3787static void igb_ping_all_vfs(struct igb_adapter *adapter)
3788{
3789 struct e1000_hw *hw = &adapter->hw;
3790 u32 ping;
3791 int i;
3792
3793 for (i = 0 ; i < adapter->vfs_allocated_count; i++) {
3794 ping = E1000_PF_CONTROL_MSG;
3795 if (adapter->vf_data[i].clear_to_send)
3796 ping |= E1000_VT_MSGTYPE_CTS;
3797 igb_write_mbx(hw, &ping, 1, i);
3798 }
3799}
3800
3801static int igb_set_vf_multicasts(struct igb_adapter *adapter,
3802 u32 *msgbuf, u32 vf)
3803{
3804 int n = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
3805 u16 *hash_list = (u16 *)&msgbuf[1];
3806 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
3807 int i;
3808
3809 /* only up to 30 hash values supported */
3810 if (n > 30)
3811 n = 30;
3812
3813 /* salt away the number of multi cast addresses assigned
3814 * to this VF for later use to restore when the PF multi cast
3815 * list changes
3816 */
3817 vf_data->num_vf_mc_hashes = n;
3818
3819 /* VFs are limited to using the MTA hash table for their multicast
3820 * addresses */
3821 for (i = 0; i < n; i++)
3822 vf_data->vf_mc_hashes[i] = hash_list[i];;
3823
3824 /* Flush and reset the mta with the new values */
3825 igb_set_multi(adapter->netdev);
3826
3827 return 0;
3828}
3829
3830static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
3831{
3832 struct e1000_hw *hw = &adapter->hw;
3833 struct vf_data_storage *vf_data;
3834 int i, j;
3835
3836 for (i = 0; i < adapter->vfs_allocated_count; i++) {
3837 vf_data = &adapter->vf_data[i];
75f4f382 3838 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
4ae196df
AD		 3839 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
3840 }
3841}
3842
3843static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
3844{
3845 struct e1000_hw *hw = &adapter->hw;
3846 u32 pool_mask, reg, vid;
3847 int i;
3848
3849 pool_mask = 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
3850
3851 /* Find the vlan filter for this id */
3852 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
3853 reg = rd32(E1000_VLVF(i));
3854
3855 /* remove the vf from the pool */
3856 reg &= ~pool_mask;
3857
3858 /* if pool is empty then remove entry from vfta */
3859 if (!(reg & E1000_VLVF_POOLSEL_MASK) &&
3860 (reg & E1000_VLVF_VLANID_ENABLE)) {
3861 reg = 0;
3862 vid = reg & E1000_VLVF_VLANID_MASK;
3863 igb_vfta_set(hw, vid, false);
3864 }
3865
3866 wr32(E1000_VLVF(i), reg);
3867 }
3868}
3869
3870static s32 igb_vlvf_set(struct igb_adapter *adapter, u32 vid, bool add, u32 vf)
3871{
3872 struct e1000_hw *hw = &adapter->hw;
3873 u32 reg, i;
3874
3875 /* It is an error to call this function when VFs are not enabled */
3876 if (!adapter->vfs_allocated_count)
3877 return -1;
3878
3879 /* Find the vlan filter for this id */
3880 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
3881 reg = rd32(E1000_VLVF(i));
3882 if ((reg & E1000_VLVF_VLANID_ENABLE) &&
3883 vid == (reg & E1000_VLVF_VLANID_MASK))
3884 break;
3885 }
3886
3887 if (add) {
3888 if (i == E1000_VLVF_ARRAY_SIZE) {
3889 /* Did not find a matching VLAN ID entry that was
3890 * enabled. Search for a free filter entry, i.e.
3891 * one without the enable bit set
3892 */
3893 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
3894 reg = rd32(E1000_VLVF(i));
3895 if (!(reg & E1000_VLVF_VLANID_ENABLE))
3896 break;
3897 }
3898 }
3899 if (i < E1000_VLVF_ARRAY_SIZE) {
3900 /* Found an enabled/available entry */
3901 reg |= 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
3902
3903 /* if !enabled we need to set this up in vfta */
3904 if (!(reg & E1000_VLVF_VLANID_ENABLE)) {
cad6d05f
AD		 3905 /* add VID to filter table, if bit already set
3906 * PF must have added it outside of table */
3907 if (igb_vfta_set(hw, vid, true))
3908 reg |= 1 << (E1000_VLVF_POOLSEL_SHIFT +
3909 adapter->vfs_allocated_count);
4ae196df
AD		 3910 reg |= E1000_VLVF_VLANID_ENABLE;
3911 }
cad6d05f
AD		 3912 reg &= ~E1000_VLVF_VLANID_MASK;
3913 reg |= vid;
4ae196df
AD		 3914
3915 wr32(E1000_VLVF(i), reg);
3916 return 0;
3917 }
3918 } else {
3919 if (i < E1000_VLVF_ARRAY_SIZE) {
3920 /* remove vf from the pool */
3921 reg &= ~(1 << (E1000_VLVF_POOLSEL_SHIFT + vf));
3922 /* if pool is empty then remove entry from vfta */
3923 if (!(reg & E1000_VLVF_POOLSEL_MASK)) {
3924 reg = 0;
3925 igb_vfta_set(hw, vid, false);
3926 }
3927 wr32(E1000_VLVF(i), reg);
3928 return 0;
3929 }
3930 }
3931 return -1;
3932}
3933
3934static int igb_set_vf_vlan(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
3935{
3936 int add = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
3937 int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
3938
3939 return igb_vlvf_set(adapter, vid, add, vf);
3940}
3941
3942static inline void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
3943{
3944 struct e1000_hw *hw = &adapter->hw;
3945
3946 /* disable mailbox functionality for vf */
3947 adapter->vf_data[vf].clear_to_send = false;
3948
3949 /* reset offloads to defaults */
3950 igb_set_vmolr(hw, vf);
3951
3952 /* reset vlans for device */
3953 igb_clear_vf_vfta(adapter, vf);
3954
3955 /* reset multicast table array for vf */
3956 adapter->vf_data[vf].num_vf_mc_hashes = 0;
3957
3958 /* Flush and reset the mta with the new values */
3959 igb_set_multi(adapter->netdev);
3960}
3961
3962static inline void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
3963{
3964 struct e1000_hw *hw = &adapter->hw;
3965 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
3966 u32 reg, msgbuf[3];
3967 u8 *addr = (u8 *)(&msgbuf[1]);
3968
3969 /* process all the same items cleared in a function level reset */
3970 igb_vf_reset_event(adapter, vf);
3971
3972 /* set vf mac address */
3973 igb_rar_set(hw, vf_mac, vf + 1);
3974 igb_set_rah_pool(hw, vf, vf + 1);
3975
3976 /* enable transmit and receive for vf */
3977 reg = rd32(E1000_VFTE);
3978 wr32(E1000_VFTE, reg | (1 << vf));
3979 reg = rd32(E1000_VFRE);
3980 wr32(E1000_VFRE, reg | (1 << vf));
3981
3982 /* enable mailbox functionality for vf */
3983 adapter->vf_data[vf].clear_to_send = true;
3984
3985 /* reply to reset with ack and vf mac address */
3986 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
3987 memcpy(addr, vf_mac, 6);
3988 igb_write_mbx(hw, msgbuf, 3, vf);
3989}
3990
3991static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf)
3992{
3993 unsigned char *addr = (char *)&msg[1];
3994 int err = -1;
3995
3996 if (is_valid_ether_addr(addr))
3997 err = igb_set_vf_mac(adapter, vf, addr);
3998
3999 return err;
4000
4001}
4002
4003static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf)
4004{
4005 struct e1000_hw *hw = &adapter->hw;
4006 u32 msg = E1000_VT_MSGTYPE_NACK;
4007
4008 /* if device isn't clear to send it shouldn't be reading either */
4009 if (!adapter->vf_data[vf].clear_to_send)
4010 igb_write_mbx(hw, &msg, 1, vf);
4011}
4012
4013
4014static void igb_msg_task(struct igb_adapter *adapter)
4015{
4016 struct e1000_hw *hw = &adapter->hw;
4017 u32 vf;
4018
4019 for (vf = 0; vf < adapter->vfs_allocated_count; vf++) {
4020 /* process any reset requests */
4021 if (!igb_check_for_rst(hw, vf)) {
4022 adapter->vf_data[vf].clear_to_send = false;
4023 igb_vf_reset_event(adapter, vf);
4024 }
4025
4026 /* process any messages pending */
4027 if (!igb_check_for_msg(hw, vf))
4028 igb_rcv_msg_from_vf(adapter, vf);
4029
4030 /* process any acks */
4031 if (!igb_check_for_ack(hw, vf))
4032 igb_rcv_ack_from_vf(adapter, vf);
4033
4034 }
4035}
4036
4037static int igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
4038{
4039 u32 mbx_size = E1000_VFMAILBOX_SIZE;
4040 u32 msgbuf[mbx_size];
4041 struct e1000_hw *hw = &adapter->hw;
4042 s32 retval;
4043
4044 retval = igb_read_mbx(hw, msgbuf, mbx_size, vf);
4045
4046 if (retval)
4047 dev_err(&adapter->pdev->dev,
4048 "Error receiving message from VF\n");
4049
4050 /* this is a message we already processed, do nothing */
4051 if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
4052 return retval;
4053
4054 /*
4055 * until the vf completes a reset it should not be
4056 * allowed to start any configuration.
4057 */
4058
4059 if (msgbuf[0] == E1000_VF_RESET) {
4060 igb_vf_reset_msg(adapter, vf);
4061
4062 return retval;
4063 }
4064
4065 if (!adapter->vf_data[vf].clear_to_send) {
4066 msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
4067 igb_write_mbx(hw, msgbuf, 1, vf);
4068 return retval;
4069 }
4070
4071 switch ((msgbuf[0] & 0xFFFF)) {
4072 case E1000_VF_SET_MAC_ADDR:
4073 retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
4074 break;
4075 case E1000_VF_SET_MULTICAST:
4076 retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
4077 break;
4078 case E1000_VF_SET_LPE:
4079 retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf);
4080 break;
4081 case E1000_VF_SET_VLAN:
4082 retval = igb_set_vf_vlan(adapter, msgbuf, vf);
4083 break;
4084 default:
4085 dev_err(&adapter->pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
4086 retval = -1;
4087 break;
4088 }
4089
4090 /* notify the VF of the results of what it sent us */
4091 if (retval)
4092 msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
4093 else
4094 msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
4095
4096 msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
4097
4098 igb_write_mbx(hw, msgbuf, 1, vf);
4099
4100 return retval;
4101}
4102
9d5c8243
AK		 4103/**
4104 * igb_intr_msi - Interrupt Handler
4105 * @irq: interrupt number
4106 * @data: pointer to a network interface device structure
4107 **/
4108static irqreturn_t igb_intr_msi(int irq, void *data)
4109{
4110 struct net_device *netdev = data;
4111 struct igb_adapter *adapter = netdev_priv(netdev);
9d5c8243
AK		 4112 struct e1000_hw *hw = &adapter->hw;
4113 /* read ICR disables interrupts using IAM */
4114 u32 icr = rd32(E1000_ICR);
4115
6eb5a7f1 4116 igb_write_itr(adapter->rx_ring);
9d5c8243 4117
dda0e083
AD		 4118 if(icr & E1000_ICR_DOUTSYNC) {
4119 /* HW is reporting DMA is out of sync */
4120 adapter->stats.doosync++;
4121 }
4122
9d5c8243
AK		 4123 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
4124 hw->mac.get_link_status = 1;
4125 if (!test_bit(__IGB_DOWN, &adapter->state))
4126 mod_timer(&adapter->watchdog_timer, jiffies + 1);
4127 }
4128
288379f0 4129 napi_schedule(&adapter->rx_ring[0].napi);
9d5c8243
AK		 4130
4131 return IRQ_HANDLED;
4132}
4133
4134/**
4a3c6433 4135 * igb_intr - Legacy Interrupt Handler
9d5c8243
AK		 4136 * @irq: interrupt number
4137 * @data: pointer to a network interface device structure
4138 **/
4139static irqreturn_t igb_intr(int irq, void *data)
4140{
4141 struct net_device *netdev = data;
4142 struct igb_adapter *adapter = netdev_priv(netdev);
9d5c8243
AK		 4143 struct e1000_hw *hw = &adapter->hw;
4144 /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked. No
4145 * need for the IMC write */
4146 u32 icr = rd32(E1000_ICR);
9d5c8243
AK		 4147 if (!icr)
4148 return IRQ_NONE; /* Not our interrupt */
4149
6eb5a7f1 4150 igb_write_itr(adapter->rx_ring);
9d5c8243
AK		 4151
4152 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
4153 * not set, then the adapter didn't send an interrupt */
4154 if (!(icr & E1000_ICR_INT_ASSERTED))
4155 return IRQ_NONE;
4156
dda0e083
AD		 4157 if(icr & E1000_ICR_DOUTSYNC) {
4158 /* HW is reporting DMA is out of sync */
4159 adapter->stats.doosync++;
4160 }
4161
9d5c8243
AK		 4162 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
4163 hw->mac.get_link_status = 1;
4164 /* guard against interrupt when we're going down */
4165 if (!test_bit(__IGB_DOWN, &adapter->state))
4166 mod_timer(&adapter->watchdog_timer, jiffies + 1);
4167 }
4168
288379f0 4169 napi_schedule(&adapter->rx_ring[0].napi);
9d5c8243
AK		 4170
4171 return IRQ_HANDLED;
4172}
4173
46544258 4174static inline void igb_rx_irq_enable(struct igb_ring *rx_ring)
9d5c8243 4175{
661086df 4176 struct igb_adapter *adapter = rx_ring->adapter;
46544258 4177 struct e1000_hw *hw = &adapter->hw;
9d5c8243 4178
46544258
AD		 4179 if (adapter->itr_setting & 3) {
4180 if (adapter->num_rx_queues == 1)
6eb5a7f1 4181 igb_set_itr(adapter);
46544258
AD		 4182 else
4183 igb_update_ring_itr(rx_ring);
9d5c8243
AK		 4184 }
4185
46544258
AD		 4186 if (!test_bit(__IGB_DOWN, &adapter->state)) {
4187 if (adapter->msix_entries)
4188 wr32(E1000_EIMS, rx_ring->eims_value);
4189 else
4190 igb_irq_enable(adapter);
4191 }
9d5c8243
AK		 4192}
4193
46544258
AD		 4194/**
4195 * igb_poll - NAPI Rx polling callback
4196 * @napi: napi polling structure
4197 * @budget: count of how many packets we should handle
4198 **/
4199static int igb_poll(struct napi_struct *napi, int budget)
9d5c8243
AK		 4200{
4201 struct igb_ring *rx_ring = container_of(napi, struct igb_ring, napi);
9d5c8243
AK		 4202 int work_done = 0;
4203
421e02f0 4204#ifdef CONFIG_IGB_DCA
bd38e5d1 4205 if (rx_ring->adapter->flags & IGB_FLAG_DCA_ENABLED)
fe4506b6
JC		 4206 igb_update_rx_dca(rx_ring);
4207#endif
3b644cf6 4208 igb_clean_rx_irq_adv(rx_ring, &work_done, budget);
9d5c8243 4209
46544258
AD		 4210 if (rx_ring->buddy) {
4211#ifdef CONFIG_IGB_DCA
bd38e5d1 4212 if (rx_ring->adapter->flags & IGB_FLAG_DCA_ENABLED)
46544258
AD		 4213 igb_update_tx_dca(rx_ring->buddy);
4214#endif
4215 if (!igb_clean_tx_irq(rx_ring->buddy))
4216 work_done = budget;
4217 }
4218
9d5c8243 4219 /* If not enough Rx work done, exit the polling mode */
5e6d5b17 4220 if (work_done < budget) {
288379f0 4221 napi_complete(napi);
46544258 4222 igb_rx_irq_enable(rx_ring);
9d5c8243
AK		 4223 }
4224
46544258 4225 return work_done;
9d5c8243 4226}
6d8126f9 4227
33af6bcc
PO		 4228/**
4229 * igb_hwtstamp - utility function which checks for TX time stamp
4230 * @adapter: board private structure
4231 * @skb: packet that was just sent
4232 *
4233 * If we were asked to do hardware stamping and such a time stamp is
4234 * available, then it must have been for this skb here because we only
4235 * allow only one such packet into the queue.
4236 */
4237static void igb_tx_hwtstamp(struct igb_adapter *adapter, struct sk_buff *skb)
4238{
4239 union skb_shared_tx *shtx = skb_tx(skb);
4240 struct e1000_hw *hw = &adapter->hw;
4241
4242 if (unlikely(shtx->hardware)) {
4243 u32 valid = rd32(E1000_TSYNCTXCTL) & E1000_TSYNCTXCTL_VALID;
4244 if (valid) {
4245 u64 regval = rd32(E1000_TXSTMPL);
4246 u64 ns;
4247 struct skb_shared_hwtstamps shhwtstamps;
4248
4249 memset(&shhwtstamps, 0, sizeof(shhwtstamps));
4250 regval |= (u64)rd32(E1000_TXSTMPH) << 32;
4251 ns = timecounter_cyc2time(&adapter->clock,
4252 regval);
4253 timecompare_update(&adapter->compare, ns);
4254 shhwtstamps.hwtstamp = ns_to_ktime(ns);
4255 shhwtstamps.syststamp =
4256 timecompare_transform(&adapter->compare, ns);
4257 skb_tstamp_tx(skb, &shhwtstamps);
4258 }
33af6bcc
PO		 4259 }
4260}
4261
9d5c8243
AK		 4262/**
4263 * igb_clean_tx_irq - Reclaim resources after transmit completes
4264 * @adapter: board private structure
4265 * returns true if ring is completely cleaned
4266 **/
3b644cf6 4267static bool igb_clean_tx_irq(struct igb_ring *tx_ring)
9d5c8243 4268{
3b644cf6 4269 struct igb_adapter *adapter = tx_ring->adapter;
3b644cf6 4270 struct net_device *netdev = adapter->netdev;
0e014cb1 4271 struct e1000_hw *hw = &adapter->hw;
9d5c8243
AK		 4272 struct igb_buffer *buffer_info;
4273 struct sk_buff *skb;
0e014cb1 4274 union e1000_adv_tx_desc *tx_desc, *eop_desc;
9d5c8243 4275 unsigned int total_bytes = 0, total_packets = 0;
0e014cb1
AD		 4276 unsigned int i, eop, count = 0;
4277 bool cleaned = false;
9d5c8243 4278
9d5c8243 4279 i = tx_ring->next_to_clean;
0e014cb1
AD		 4280 eop = tx_ring->buffer_info[i].next_to_watch;
4281 eop_desc = E1000_TX_DESC_ADV(*tx_ring, eop);
4282
4283 while ((eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)) &&
4284 (count < tx_ring->count)) {
4285 for (cleaned = false; !cleaned; count++) {
4286 tx_desc = E1000_TX_DESC_ADV(*tx_ring, i);
9d5c8243 4287 buffer_info = &tx_ring->buffer_info[i];
0e014cb1 4288 cleaned = (i == eop);
9d5c8243
AK		 4289 skb = buffer_info->skb;
4290
4291 if (skb) {
4292 unsigned int segs, bytecount;
4293 /* gso_segs is currently only valid for tcp */
4294 segs = skb_shinfo(skb)->gso_segs ?: 1;
4295 /* multiply data chunks by size of headers */
4296 bytecount = ((segs - 1) * skb_headlen(skb)) +
4297 skb->len;
4298 total_packets += segs;
4299 total_bytes += bytecount;
33af6bcc
PO		 4300
4301 igb_tx_hwtstamp(adapter, skb);
9d5c8243
AK		 4302 }
4303
4304 igb_unmap_and_free_tx_resource(adapter, buffer_info);
0e014cb1 4305 tx_desc->wb.status = 0;
9d5c8243
AK		 4306
4307 i++;
4308 if (i == tx_ring->count)
4309 i = 0;
9d5c8243 4310 }
0e014cb1
AD		 4311 eop = tx_ring->buffer_info[i].next_to_watch;
4312 eop_desc = E1000_TX_DESC_ADV(*tx_ring, eop);
4313 }
4314
9d5c8243
AK		 4315 tx_ring->next_to_clean = i;
4316
fc7d345d 4317 if (unlikely(count &&
9d5c8243 4318 netif_carrier_ok(netdev) &&
c493ea45 4319 igb_desc_unused(tx_ring) >= IGB_TX_QUEUE_WAKE)) {
9d5c8243
AK		 4320 /* Make sure that anybody stopping the queue after this
4321 * sees the new next_to_clean.
4322 */
4323 smp_mb();
661086df
PWJ		 4324 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
4325 !(test_bit(__IGB_DOWN, &adapter->state))) {
4326 netif_wake_subqueue(netdev, tx_ring->queue_index);
4327 ++adapter->restart_queue;
4328 }
9d5c8243
AK		 4329 }
4330
4331 if (tx_ring->detect_tx_hung) {
4332 /* Detect a transmit hang in hardware, this serializes the
4333 * check with the clearing of time_stamp and movement of i */
4334 tx_ring->detect_tx_hung = false;
4335 if (tx_ring->buffer_info[i].time_stamp &&
4336 time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
4337 (adapter->tx_timeout_factor * HZ))
4338 && !(rd32(E1000_STATUS) &
4339 E1000_STATUS_TXOFF)) {
4340
9d5c8243
AK		 4341 /* detected Tx unit hang */
4342 dev_err(&adapter->pdev->dev,
4343 "Detected Tx Unit Hang\n"
2d064c06 4344 " Tx Queue <%d>\n"
9d5c8243
AK		 4345 " TDH <%x>\n"
4346 " TDT <%x>\n"
4347 " next_to_use <%x>\n"
4348 " next_to_clean <%x>\n"
9d5c8243
AK		 4349 "buffer_info[next_to_clean]\n"
4350 " time_stamp <%lx>\n"
0e014cb1 4351 " next_to_watch <%x>\n"
9d5c8243
AK		 4352 " jiffies <%lx>\n"
4353 " desc.status <%x>\n",
2d064c06 4354 tx_ring->queue_index,
9d5c8243
AK		 4355 readl(adapter->hw.hw_addr + tx_ring->head),
4356 readl(adapter->hw.hw_addr + tx_ring->tail),
4357 tx_ring->next_to_use,
4358 tx_ring->next_to_clean,
9d5c8243 4359 tx_ring->buffer_info[i].time_stamp,
0e014cb1 4360 eop,
9d5c8243 4361 jiffies,
0e014cb1 4362 eop_desc->wb.status);
661086df 4363 netif_stop_subqueue(netdev, tx_ring->queue_index);
9d5c8243
AK		 4364 }
4365 }
4366 tx_ring->total_bytes += total_bytes;
4367 tx_ring->total_packets += total_packets;
e21ed353
AD		 4368 tx_ring->tx_stats.bytes += total_bytes;
4369 tx_ring->tx_stats.packets += total_packets;
9d5c8243
AK		 4370 adapter->net_stats.tx_bytes += total_bytes;
4371 adapter->net_stats.tx_packets += total_packets;
0e014cb1 4372 return (count < tx_ring->count);
9d5c8243
AK		 4373}
4374
9d5c8243
AK		 4375/**
4376 * igb_receive_skb - helper function to handle rx indications
eebbbdba 4377 * @ring: pointer to receive ring receving this packet
9d5c8243 4378 * @status: descriptor status field as written by hardware
73cd78f1 4379 * @rx_desc: receive descriptor containing vlan and type information.
9d5c8243
AK		 4380 * @skb: pointer to sk_buff to be indicated to stack
4381 **/
d3352520
AD		 4382static void igb_receive_skb(struct igb_ring *ring, u8 status,
4383 union e1000_adv_rx_desc * rx_desc,
4384 struct sk_buff *skb)
4385{
4386 struct igb_adapter * adapter = ring->adapter;
4387 bool vlan_extracted = (adapter->vlgrp && (status & E1000_RXD_STAT_VP));
4388
0c8dfc83 4389 skb_record_rx_queue(skb, ring->queue_index);
5c0999b7 4390 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
d3352520 4391 if (vlan_extracted)
5c0999b7
HX		 4392 vlan_gro_receive(&ring->napi, adapter->vlgrp,
4393 le16_to_cpu(rx_desc->wb.upper.vlan),
4394 skb);
d3352520 4395 else
5c0999b7 4396 napi_gro_receive(&ring->napi, skb);
d3352520 4397 } else {
d3352520
AD		 4398 if (vlan_extracted)
4399 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
4400 le16_to_cpu(rx_desc->wb.upper.vlan));
4401 else
d3352520 4402 netif_receive_skb(skb);
d3352520 4403 }
9d5c8243
AK		 4404}
4405
9d5c8243
AK		 4406static inline void igb_rx_checksum_adv(struct igb_adapter *adapter,
4407 u32 status_err, struct sk_buff *skb)
4408{
4409 skb->ip_summed = CHECKSUM_NONE;
4410
4411 /* Ignore Checksum bit is set or checksum is disabled through ethtool */
4412 if ((status_err & E1000_RXD_STAT_IXSM) || !adapter->rx_csum)
4413 return;
4414 /* TCP/UDP checksum error bit is set */
4415 if (status_err &
4416 (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
4417 /* let the stack verify checksum errors */
4418 adapter->hw_csum_err++;
4419 return;
4420 }
4421 /* It must be a TCP or UDP packet with a valid checksum */
4422 if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
4423 skb->ip_summed = CHECKSUM_UNNECESSARY;
4424
4425 adapter->hw_csum_good++;
4426}
4427
3b644cf6
MW		 4428static bool igb_clean_rx_irq_adv(struct igb_ring *rx_ring,
4429 int *work_done, int budget)
9d5c8243 4430{
3b644cf6 4431 struct igb_adapter *adapter = rx_ring->adapter;
9d5c8243 4432 struct net_device *netdev = adapter->netdev;
33af6bcc 4433 struct e1000_hw *hw = &adapter->hw;
9d5c8243
AK		 4434 struct pci_dev *pdev = adapter->pdev;
4435 union e1000_adv_rx_desc *rx_desc , *next_rxd;
4436 struct igb_buffer *buffer_info , *next_buffer;
4437 struct sk_buff *skb;
9d5c8243
AK		 4438 bool cleaned = false;
4439 int cleaned_count = 0;
4440 unsigned int total_bytes = 0, total_packets = 0;
73cd78f1
AD		 4441 unsigned int i;
4442 u32 length, hlen, staterr;
9d5c8243
AK		 4443
4444 i = rx_ring->next_to_clean;
69d3ca53 4445 buffer_info = &rx_ring->buffer_info[i];
9d5c8243
AK		 4446 rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
4447 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
4448
4449 while (staterr & E1000_RXD_STAT_DD) {
4450 if (*work_done >= budget)
4451 break;
4452 (*work_done)++;
9d5c8243 4453
69d3ca53
AD		 4454 skb = buffer_info->skb;
4455 prefetch(skb->data - NET_IP_ALIGN);
4456 buffer_info->skb = NULL;
4457
4458 i++;
4459 if (i == rx_ring->count)
4460 i = 0;
4461 next_rxd = E1000_RX_DESC_ADV(*rx_ring, i);
4462 prefetch(next_rxd);
4463 next_buffer = &rx_ring->buffer_info[i];
9d5c8243
AK		 4464
4465 length = le16_to_cpu(rx_desc->wb.upper.length);
4466 cleaned = true;
4467 cleaned_count++;
4468
bf36c1a0
AD		 4469 if (!adapter->rx_ps_hdr_size) {
4470 pci_unmap_single(pdev, buffer_info->dma,
4471 adapter->rx_buffer_len +
4472 NET_IP_ALIGN,
4473 PCI_DMA_FROMDEVICE);
4474 skb_put(skb, length);
4475 goto send_up;
9d5c8243
AK		 4476 }
4477
69d3ca53
AD		 4478 /* HW will not DMA in data larger than the given buffer, even
4479 * if it parses the (NFS, of course) header to be larger. In
4480 * that case, it fills the header buffer and spills the rest
4481 * into the page.
4482 */
4483 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hdr_info) &
4484 E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
4485 if (hlen > adapter->rx_ps_hdr_size)
4486 hlen = adapter->rx_ps_hdr_size;
4487
bf36c1a0
AD		 4488 if (!skb_shinfo(skb)->nr_frags) {
4489 pci_unmap_single(pdev, buffer_info->dma,
73cd78f1 4490 adapter->rx_ps_hdr_size + NET_IP_ALIGN,
bf36c1a0
AD		 4491 PCI_DMA_FROMDEVICE);
4492 skb_put(skb, hlen);
4493 }
4494
4495 if (length) {
9d5c8243 4496 pci_unmap_page(pdev, buffer_info->page_dma,
bf36c1a0 4497 PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
9d5c8243 4498 buffer_info->page_dma = 0;
bf36c1a0
AD		 4499
4500 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags++,
4501 buffer_info->page,
4502 buffer_info->page_offset,
4503 length);
4504
4505 if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
4506 (page_count(buffer_info->page) != 1))
4507 buffer_info->page = NULL;
4508 else
4509 get_page(buffer_info->page);
9d5c8243
AK		 4510
4511 skb->len += length;
4512 skb->data_len += length;
9d5c8243 4513
bf36c1a0 4514 skb->truesize += length;
9d5c8243 4515 }
9d5c8243 4516
bf36c1a0 4517 if (!(staterr & E1000_RXD_STAT_EOP)) {
b2d56536
AD		 4518 buffer_info->skb = next_buffer->skb;
4519 buffer_info->dma = next_buffer->dma;
4520 next_buffer->skb = skb;
4521 next_buffer->dma = 0;
bf36c1a0
AD		 4522 goto next_desc;
4523 }
69d3ca53 4524send_up:
33af6bcc
PO		 4525 /*
4526 * If this bit is set, then the RX registers contain
4527 * the time stamp. No other packet will be time
4528 * stamped until we read these registers, so read the
4529 * registers to make them available again. Because
4530 * only one packet can be time stamped at a time, we
4531 * know that the register values must belong to this
4532 * one here and therefore we don't need to compare
4533 * any of the additional attributes stored for it.
4534 *
4535 * If nothing went wrong, then it should have a
4536 * skb_shared_tx that we can turn into a
4537 * skb_shared_hwtstamps.
4538 *
4539 * TODO: can time stamping be triggered (thus locking
4540 * the registers) without the packet reaching this point
4541 * here? In that case RX time stamping would get stuck.
4542 *
4543 * TODO: in "time stamp all packets" mode this bit is
4544 * not set. Need a global flag for this mode and then
4545 * always read the registers. Cannot be done without
4546 * a race condition.
4547 */
4548 if (unlikely(staterr & E1000_RXD_STAT_TS)) {
4549 u64 regval;
4550 u64 ns;
4551 struct skb_shared_hwtstamps *shhwtstamps =
4552 skb_hwtstamps(skb);
4553
4554 WARN(!(rd32(E1000_TSYNCRXCTL) & E1000_TSYNCRXCTL_VALID),
4555 "igb: no RX time stamp available for time stamped packet");
4556 regval = rd32(E1000_RXSTMPL);
4557 regval |= (u64)rd32(E1000_RXSTMPH) << 32;
4558 ns = timecounter_cyc2time(&adapter->clock, regval);
4559 timecompare_update(&adapter->compare, ns);
4560 memset(shhwtstamps, 0, sizeof(*shhwtstamps));
4561 shhwtstamps->hwtstamp = ns_to_ktime(ns);
4562 shhwtstamps->syststamp =
4563 timecompare_transform(&adapter->compare, ns);
4564 }
4565
9d5c8243
AK		 4566 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
4567 dev_kfree_skb_irq(skb);
4568 goto next_desc;
4569 }
9d5c8243
AK		 4570
4571 total_bytes += skb->len;
4572 total_packets++;
4573
4574 igb_rx_checksum_adv(adapter, staterr, skb);
4575
4576 skb->protocol = eth_type_trans(skb, netdev);
4577
d3352520 4578 igb_receive_skb(rx_ring, staterr, rx_desc, skb);
9d5c8243 4579
9d5c8243
AK		 4580next_desc:
4581 rx_desc->wb.upper.status_error = 0;
4582
4583 /* return some buffers to hardware, one at a time is too slow */
4584 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
3b644cf6 4585 igb_alloc_rx_buffers_adv(rx_ring, cleaned_count);
9d5c8243
AK		 4586 cleaned_count = 0;
4587 }
4588
4589 /* use prefetched values */
4590 rx_desc = next_rxd;
4591 buffer_info = next_buffer;
9d5c8243
AK		 4592 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
4593 }
bf36c1a0 4594
9d5c8243 4595 rx_ring->next_to_clean = i;
c493ea45 4596 cleaned_count = igb_desc_unused(rx_ring);
9d5c8243
AK		 4597
4598 if (cleaned_count)
3b644cf6 4599 igb_alloc_rx_buffers_adv(rx_ring, cleaned_count);
9d5c8243
AK		 4600
4601 rx_ring->total_packets += total_packets;
4602 rx_ring->total_bytes += total_bytes;
4603 rx_ring->rx_stats.packets += total_packets;
4604 rx_ring->rx_stats.bytes += total_bytes;
4605 adapter->net_stats.rx_bytes += total_bytes;
4606 adapter->net_stats.rx_packets += total_packets;
4607 return cleaned;
4608}
4609
9d5c8243
AK		 4610/**
4611 * igb_alloc_rx_buffers_adv - Replace used receive buffers; packet split
4612 * @adapter: address of board private structure
4613 **/
3b644cf6 4614static void igb_alloc_rx_buffers_adv(struct igb_ring *rx_ring,
9d5c8243
AK		 4615 int cleaned_count)
4616{
3b644cf6 4617 struct igb_adapter *adapter = rx_ring->adapter;
9d5c8243
AK		 4618 struct net_device *netdev = adapter->netdev;
4619 struct pci_dev *pdev = adapter->pdev;
4620 union e1000_adv_rx_desc *rx_desc;
4621 struct igb_buffer *buffer_info;
4622 struct sk_buff *skb;
4623 unsigned int i;
db761762 4624 int bufsz;
9d5c8243
AK		 4625
4626 i = rx_ring->next_to_use;
4627 buffer_info = &rx_ring->buffer_info[i];
4628
db761762
AD		 4629 if (adapter->rx_ps_hdr_size)
4630 bufsz = adapter->rx_ps_hdr_size;
4631 else
4632 bufsz = adapter->rx_buffer_len;
4633 bufsz += NET_IP_ALIGN;
4634
9d5c8243
AK		 4635 while (cleaned_count--) {
4636 rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
4637
bf36c1a0 4638 if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) {
9d5c8243 4639 if (!buffer_info->page) {
bf36c1a0
AD		 4640 buffer_info->page = alloc_page(GFP_ATOMIC);
4641 if (!buffer_info->page) {
4642 adapter->alloc_rx_buff_failed++;
4643 goto no_buffers;
4644 }
4645 buffer_info->page_offset = 0;
4646 } else {
4647 buffer_info->page_offset ^= PAGE_SIZE / 2;
9d5c8243
AK		 4648 }
4649 buffer_info->page_dma =
db761762 4650 pci_map_page(pdev, buffer_info->page,
bf36c1a0
AD		 4651 buffer_info->page_offset,
4652 PAGE_SIZE / 2,
9d5c8243
AK		 4653 PCI_DMA_FROMDEVICE);
4654 }
4655
4656 if (!buffer_info->skb) {
9d5c8243 4657 skb = netdev_alloc_skb(netdev, bufsz);
9d5c8243
AK		 4658 if (!skb) {
4659 adapter->alloc_rx_buff_failed++;
4660 goto no_buffers;
4661 }
4662
4663 /* Make buffer alignment 2 beyond a 16 byte boundary
4664 * this will result in a 16 byte aligned IP header after
4665 * the 14 byte MAC header is removed
4666 */
4667 skb_reserve(skb, NET_IP_ALIGN);
4668
4669 buffer_info->skb = skb;
4670 buffer_info->dma = pci_map_single(pdev, skb->data,
4671 bufsz,
4672 PCI_DMA_FROMDEVICE);
9d5c8243
AK		 4673 }
4674 /* Refresh the desc even if buffer_addrs didn't change because
4675 * each write-back erases this info. */
4676 if (adapter->rx_ps_hdr_size) {
4677 rx_desc->read.pkt_addr =
4678 cpu_to_le64(buffer_info->page_dma);
4679 rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
4680 } else {
4681 rx_desc->read.pkt_addr =
4682 cpu_to_le64(buffer_info->dma);
4683 rx_desc->read.hdr_addr = 0;
4684 }
4685
4686 i++;
4687 if (i == rx_ring->count)
4688 i = 0;
4689 buffer_info = &rx_ring->buffer_info[i];
4690 }
4691
4692no_buffers:
4693 if (rx_ring->next_to_use != i) {
4694 rx_ring->next_to_use = i;
4695 if (i == 0)
4696 i = (rx_ring->count - 1);
4697 else
4698 i--;
4699
4700 /* Force memory writes to complete before letting h/w
4701 * know there are new descriptors to fetch. (Only
4702 * applicable for weak-ordered memory model archs,
4703 * such as IA-64). */
4704 wmb();
4705 writel(i, adapter->hw.hw_addr + rx_ring->tail);
4706 }
4707}
4708
4709/**
4710 * igb_mii_ioctl -
4711 * @netdev:
4712 * @ifreq:
4713 * @cmd:
4714 **/
4715static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4716{
4717 struct igb_adapter *adapter = netdev_priv(netdev);
4718 struct mii_ioctl_data *data = if_mii(ifr);
4719
4720 if (adapter->hw.phy.media_type != e1000_media_type_copper)
4721 return -EOPNOTSUPP;
4722
4723 switch (cmd) {
4724 case SIOCGMIIPHY:
4725 data->phy_id = adapter->hw.phy.addr;
4726 break;
4727 case SIOCGMIIREG:
4728 if (!capable(CAP_NET_ADMIN))
4729 return -EPERM;
f5f4cf08
AD		 4730 if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
4731 &data->val_out))
9d5c8243
AK		 4732 return -EIO;
4733 break;
4734 case SIOCSMIIREG:
4735 default:
4736 return -EOPNOTSUPP;
4737 }
4738 return 0;
4739}
4740
c6cb090b
PO		 4741/**
4742 * igb_hwtstamp_ioctl - control hardware time stamping
4743 * @netdev:
4744 * @ifreq:
4745 * @cmd:
4746 *
33af6bcc
PO		 4747 * Outgoing time stamping can be enabled and disabled. Play nice and
4748 * disable it when requested, although it shouldn't case any overhead
4749 * when no packet needs it. At most one packet in the queue may be
4750 * marked for time stamping, otherwise it would be impossible to tell
4751 * for sure to which packet the hardware time stamp belongs.
4752 *
4753 * Incoming time stamping has to be configured via the hardware
4754 * filters. Not all combinations are supported, in particular event
4755 * type has to be specified. Matching the kind of event packet is
4756 * not supported, with the exception of "all V2 events regardless of
4757 * level 2 or 4".
4758 *
c6cb090b
PO		 4759 **/
4760static int igb_hwtstamp_ioctl(struct net_device *netdev,
4761 struct ifreq *ifr, int cmd)
4762{
33af6bcc
PO		 4763 struct igb_adapter *adapter = netdev_priv(netdev);
4764 struct e1000_hw *hw = &adapter->hw;
c6cb090b 4765 struct hwtstamp_config config;
33af6bcc
PO		 4766 u32 tsync_tx_ctl_bit = E1000_TSYNCTXCTL_ENABLED;
4767 u32 tsync_rx_ctl_bit = E1000_TSYNCRXCTL_ENABLED;
4768 u32 tsync_rx_ctl_type = 0;
4769 u32 tsync_rx_cfg = 0;
4770 int is_l4 = 0;
4771 int is_l2 = 0;
4772 short port = 319; /* PTP */
4773 u32 regval;
c6cb090b
PO		 4774
4775 if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
4776 return -EFAULT;
4777
4778 /* reserved for future extensions */
4779 if (config.flags)
4780 return -EINVAL;
4781
33af6bcc
PO		 4782 switch (config.tx_type) {
4783 case HWTSTAMP_TX_OFF:
4784 tsync_tx_ctl_bit = 0;
4785 break;
4786 case HWTSTAMP_TX_ON:
4787 tsync_tx_ctl_bit = E1000_TSYNCTXCTL_ENABLED;
4788 break;
4789 default:
4790 return -ERANGE;
4791 }
4792
4793 switch (config.rx_filter) {
4794 case HWTSTAMP_FILTER_NONE:
4795 tsync_rx_ctl_bit = 0;
4796 break;
4797 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
4798 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
4799 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
4800 case HWTSTAMP_FILTER_ALL:
4801 /*
4802 * register TSYNCRXCFG must be set, therefore it is not
4803 * possible to time stamp both Sync and Delay_Req messages
4804 * => fall back to time stamping all packets
4805 */
4806 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_ALL;
4807 config.rx_filter = HWTSTAMP_FILTER_ALL;
4808 break;
4809 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
4810 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_L4_V1;
4811 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V1_SYNC_MESSAGE;
4812 is_l4 = 1;
4813 break;
4814 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
4815 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_L4_V1;
4816 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V1_DELAY_REQ_MESSAGE;
4817 is_l4 = 1;
4818 break;
4819 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
4820 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
4821 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
4822 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V2_SYNC_MESSAGE;
4823 is_l2 = 1;
4824 is_l4 = 1;
4825 config.rx_filter = HWTSTAMP_FILTER_SOME;
4826 break;
4827 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
4828 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
4829 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
4830 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V2_DELAY_REQ_MESSAGE;
4831 is_l2 = 1;
4832 is_l4 = 1;
4833 config.rx_filter = HWTSTAMP_FILTER_SOME;
4834 break;
4835 case HWTSTAMP_FILTER_PTP_V2_EVENT:
4836 case HWTSTAMP_FILTER_PTP_V2_SYNC:
4837 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
4838 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_EVENT_V2;
4839 config.rx_filter = HWTSTAMP_FILTER_PTP_V2_EVENT;
4840 is_l2 = 1;
4841 break;
4842 default:
4843 return -ERANGE;
4844 }
4845
4846 /* enable/disable TX */
4847 regval = rd32(E1000_TSYNCTXCTL);
4848 regval = (regval & ~E1000_TSYNCTXCTL_ENABLED) | tsync_tx_ctl_bit;
4849 wr32(E1000_TSYNCTXCTL, regval);
4850
4851 /* enable/disable RX, define which PTP packets are time stamped */
4852 regval = rd32(E1000_TSYNCRXCTL);
4853 regval = (regval & ~E1000_TSYNCRXCTL_ENABLED) | tsync_rx_ctl_bit;
4854 regval = (regval & ~0xE) | tsync_rx_ctl_type;
4855 wr32(E1000_TSYNCRXCTL, regval);
4856 wr32(E1000_TSYNCRXCFG, tsync_rx_cfg);
4857
4858 /*
4859 * Ethertype Filter Queue Filter[0][15:0] = 0x88F7
4860 * (Ethertype to filter on)
4861 * Ethertype Filter Queue Filter[0][26] = 0x1 (Enable filter)
4862 * Ethertype Filter Queue Filter[0][30] = 0x1 (Enable Timestamping)
4863 */
4864 wr32(E1000_ETQF0, is_l2 ? 0x440088f7 : 0);
4865
4866 /* L4 Queue Filter[0]: only filter by source and destination port */
4867 wr32(E1000_SPQF0, htons(port));
4868 wr32(E1000_IMIREXT(0), is_l4 ?
4869 ((1<<12) | (1<<19) /* bypass size and control flags */) : 0);
4870 wr32(E1000_IMIR(0), is_l4 ?
4871 (htons(port)
4872 | (0<<16) /* immediate interrupt disabled */
4873 | 0 /* (1<<17) bit cleared: do not bypass
4874 destination port check */)
4875 : 0);
4876 wr32(E1000_FTQF0, is_l4 ?
4877 (0x11 /* UDP */
4878 | (1<<15) /* VF not compared */
4879 | (1<<27) /* Enable Timestamping */
4880 | (7<<28) /* only source port filter enabled,
4881 source/target address and protocol
4882 masked */)
4883 : ((1<<15) | (15<<28) /* all mask bits set = filter not
4884 enabled */));
4885
4886 wrfl();
4887
4888 adapter->hwtstamp_config = config;
4889
4890 /* clear TX/RX time stamp registers, just to be sure */
4891 regval = rd32(E1000_TXSTMPH);
4892 regval = rd32(E1000_RXSTMPH);
c6cb090b 4893
33af6bcc
PO		 4894 return copy_to_user(ifr->ifr_data, &config, sizeof(config)) ?
4895 -EFAULT : 0;
c6cb090b
PO		 4896}
4897
9d5c8243
AK		 4898/**
4899 * igb_ioctl -
4900 * @netdev:
4901 * @ifreq:
4902 * @cmd:
4903 **/
4904static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4905{
4906 switch (cmd) {
4907 case SIOCGMIIPHY:
4908 case SIOCGMIIREG:
4909 case SIOCSMIIREG:
4910 return igb_mii_ioctl(netdev, ifr, cmd);
c6cb090b
PO		 4911 case SIOCSHWTSTAMP:
4912 return igb_hwtstamp_ioctl(netdev, ifr, cmd);
9d5c8243
AK		 4913 default:
4914 return -EOPNOTSUPP;
4915 }
4916}
4917
4918static void igb_vlan_rx_register(struct net_device *netdev,
4919 struct vlan_group *grp)
4920{
4921 struct igb_adapter *adapter = netdev_priv(netdev);
4922 struct e1000_hw *hw = &adapter->hw;
4923 u32 ctrl, rctl;
4924
4925 igb_irq_disable(adapter);
4926 adapter->vlgrp = grp;
4927
4928 if (grp) {
4929 /* enable VLAN tag insert/strip */
4930 ctrl = rd32(E1000_CTRL);
4931 ctrl |= E1000_CTRL_VME;
4932 wr32(E1000_CTRL, ctrl);
4933
4934 /* enable VLAN receive filtering */
4935 rctl = rd32(E1000_RCTL);
9d5c8243
AK		 4936 rctl &= ~E1000_RCTL_CFIEN;
4937 wr32(E1000_RCTL, rctl);
4938 igb_update_mng_vlan(adapter);
9d5c8243
AK		 4939 } else {
4940 /* disable VLAN tag insert/strip */
4941 ctrl = rd32(E1000_CTRL);
4942 ctrl &= ~E1000_CTRL_VME;
4943 wr32(E1000_CTRL, ctrl);
4944
9d5c8243
AK		 4945 if (adapter->mng_vlan_id != (u16)IGB_MNG_VLAN_NONE) {
4946 igb_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
4947 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
4948 }
9d5c8243
AK		 4949 }
4950
e1739522
AD		 4951 igb_rlpml_set(adapter);
4952
9d5c8243
AK		 4953 if (!test_bit(__IGB_DOWN, &adapter->state))
4954 igb_irq_enable(adapter);
4955}
4956
4957static void igb_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
4958{
4959 struct igb_adapter *adapter = netdev_priv(netdev);
4960 struct e1000_hw *hw = &adapter->hw;
4ae196df 4961 int pf_id = adapter->vfs_allocated_count;
9d5c8243 4962
28b0759c 4963 if ((hw->mng_cookie.status &
9d5c8243
AK		 4964 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
4965 (vid == adapter->mng_vlan_id))
4966 return;
4ae196df
AD		 4967
4968 /* add vid to vlvf if sr-iov is enabled,
4969 * if that fails add directly to filter table */
4970 if (igb_vlvf_set(adapter, vid, true, pf_id))
4971 igb_vfta_set(hw, vid, true);
4972
9d5c8243
AK		 4973}
4974
4975static void igb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
4976{
4977 struct igb_adapter *adapter = netdev_priv(netdev);
4978 struct e1000_hw *hw = &adapter->hw;
4ae196df 4979 int pf_id = adapter->vfs_allocated_count;
9d5c8243
AK		 4980
4981 igb_irq_disable(adapter);
4982 vlan_group_set_device(adapter->vlgrp, vid, NULL);
4983
4984 if (!test_bit(__IGB_DOWN, &adapter->state))
4985 igb_irq_enable(adapter);
4986
4987 if ((adapter->hw.mng_cookie.status &
4988 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
4989 (vid == adapter->mng_vlan_id)) {
4990 /* release control to f/w */
4991 igb_release_hw_control(adapter);
4992 return;
4993 }
4994
4ae196df
AD		 4995 /* remove vid from vlvf if sr-iov is enabled,
4996 * if not in vlvf remove from vfta */
4997 if (igb_vlvf_set(adapter, vid, false, pf_id))
4998 igb_vfta_set(hw, vid, false);
9d5c8243
AK		 4999}
5000
5001static void igb_restore_vlan(struct igb_adapter *adapter)
5002{
5003 igb_vlan_rx_register(adapter->netdev, adapter->vlgrp);
5004
5005 if (adapter->vlgrp) {
5006 u16 vid;
5007 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
5008 if (!vlan_group_get_device(adapter->vlgrp, vid))
5009 continue;
5010 igb_vlan_rx_add_vid(adapter->netdev, vid);
5011 }
5012 }
5013}
5014
5015int igb_set_spd_dplx(struct igb_adapter *adapter, u16 spddplx)
5016{
5017 struct e1000_mac_info *mac = &adapter->hw.mac;
5018
5019 mac->autoneg = 0;
5020
5021 /* Fiber NICs only allow 1000 gbps Full duplex */
5022 if ((adapter->hw.phy.media_type == e1000_media_type_fiber) &&
5023 spddplx != (SPEED_1000 + DUPLEX_FULL)) {
5024 dev_err(&adapter->pdev->dev,
5025 "Unsupported Speed/Duplex configuration\n");
5026 return -EINVAL;
5027 }
5028
5029 switch (spddplx) {
5030 case SPEED_10 + DUPLEX_HALF:
5031 mac->forced_speed_duplex = ADVERTISE_10_HALF;
5032 break;
5033 case SPEED_10 + DUPLEX_FULL:
5034 mac->forced_speed_duplex = ADVERTISE_10_FULL;
5035 break;
5036 case SPEED_100 + DUPLEX_HALF:
5037 mac->forced_speed_duplex = ADVERTISE_100_HALF;
5038 break;
5039 case SPEED_100 + DUPLEX_FULL:
5040 mac->forced_speed_duplex = ADVERTISE_100_FULL;
5041 break;
5042 case SPEED_1000 + DUPLEX_FULL:
5043 mac->autoneg = 1;
5044 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
5045 break;
5046 case SPEED_1000 + DUPLEX_HALF: /* not supported */
5047 default:
5048 dev_err(&adapter->pdev->dev,
5049 "Unsupported Speed/Duplex configuration\n");
5050 return -EINVAL;
5051 }
5052 return 0;
5053}
5054
9d5c8243
AK		 5055static int igb_suspend(struct pci_dev *pdev, pm_message_t state)
5056{
5057 struct net_device *netdev = pci_get_drvdata(pdev);
5058 struct igb_adapter *adapter = netdev_priv(netdev);
5059 struct e1000_hw *hw = &adapter->hw;
2d064c06 5060 u32 ctrl, rctl, status;
9d5c8243
AK		 5061 u32 wufc = adapter->wol;
5062#ifdef CONFIG_PM
5063 int retval = 0;
5064#endif
5065
5066 netif_device_detach(netdev);
5067
a88f10ec
AD		 5068 if (netif_running(netdev))
5069 igb_close(netdev);
5070
5071 igb_reset_interrupt_capability(adapter);
5072
5073 igb_free_queues(adapter);
9d5c8243
AK		 5074
5075#ifdef CONFIG_PM
5076 retval = pci_save_state(pdev);
5077 if (retval)
5078 return retval;
5079#endif
5080
5081 status = rd32(E1000_STATUS);
5082 if (status & E1000_STATUS_LU)
5083 wufc &= ~E1000_WUFC_LNKC;
5084
5085 if (wufc) {
5086 igb_setup_rctl(adapter);
5087 igb_set_multi(netdev);
5088
5089 /* turn on all-multi mode if wake on multicast is enabled */
5090 if (wufc & E1000_WUFC_MC) {
5091 rctl = rd32(E1000_RCTL);
5092 rctl |= E1000_RCTL_MPE;
5093 wr32(E1000_RCTL, rctl);
5094 }
5095
5096 ctrl = rd32(E1000_CTRL);
5097 /* advertise wake from D3Cold */
5098 #define E1000_CTRL_ADVD3WUC 0x00100000
5099 /* phy power management enable */
5100 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5101 ctrl |= E1000_CTRL_ADVD3WUC;
5102 wr32(E1000_CTRL, ctrl);
5103
9d5c8243
AK		 5104 /* Allow time for pending master requests to run */
5105 igb_disable_pcie_master(&adapter->hw);
5106
5107 wr32(E1000_WUC, E1000_WUC_PME_EN);
5108 wr32(E1000_WUFC, wufc);
9d5c8243
AK		 5109 } else {
5110 wr32(E1000_WUC, 0);
5111 wr32(E1000_WUFC, 0);
9d5c8243
AK		 5112 }
5113
2d064c06
AD		 5114 /* make sure adapter isn't asleep if manageability/wol is enabled */
5115 if (wufc || adapter->en_mng_pt) {
9d5c8243
AK		 5116 pci_enable_wake(pdev, PCI_D3hot, 1);
5117 pci_enable_wake(pdev, PCI_D3cold, 1);
2d064c06
AD		 5118 } else {
5119 igb_shutdown_fiber_serdes_link_82575(hw);
5120 pci_enable_wake(pdev, PCI_D3hot, 0);
5121 pci_enable_wake(pdev, PCI_D3cold, 0);
9d5c8243
AK		 5122 }
5123
5124 /* Release control of h/w to f/w. If f/w is AMT enabled, this
5125 * would have already happened in close and is redundant. */
5126 igb_release_hw_control(adapter);
5127
5128 pci_disable_device(pdev);
5129
5130 pci_set_power_state(pdev, pci_choose_state(pdev, state));
5131
5132 return 0;
5133}
5134
5135#ifdef CONFIG_PM
5136static int igb_resume(struct pci_dev *pdev)
5137{
5138 struct net_device *netdev = pci_get_drvdata(pdev);
5139 struct igb_adapter *adapter = netdev_priv(netdev);
5140 struct e1000_hw *hw = &adapter->hw;
5141 u32 err;
5142
5143 pci_set_power_state(pdev, PCI_D0);
5144 pci_restore_state(pdev);
42bfd33a 5145
aed5dec3 5146 err = pci_enable_device_mem(pdev);
9d5c8243
AK		 5147 if (err) {
5148 dev_err(&pdev->dev,
5149 "igb: Cannot enable PCI device from suspend\n");
5150 return err;
5151 }
5152 pci_set_master(pdev);
5153
5154 pci_enable_wake(pdev, PCI_D3hot, 0);
5155 pci_enable_wake(pdev, PCI_D3cold, 0);
5156
a88f10ec
AD		 5157 igb_set_interrupt_capability(adapter);
5158
5159 if (igb_alloc_queues(adapter)) {
5160 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
5161 return -ENOMEM;
9d5c8243
AK		 5162 }
5163
5164 /* e1000_power_up_phy(adapter); */
5165
5166 igb_reset(adapter);
a8564f03
AD		 5167
5168 /* let the f/w know that the h/w is now under the control of the
5169 * driver. */
5170 igb_get_hw_control(adapter);
5171
9d5c8243
AK		 5172 wr32(E1000_WUS, ~0);
5173
a88f10ec
AD		 5174 if (netif_running(netdev)) {
5175 err = igb_open(netdev);
5176 if (err)
5177 return err;
5178 }
9d5c8243
AK		 5179
5180 netif_device_attach(netdev);
5181
9d5c8243
AK		 5182 return 0;
5183}
5184#endif
5185
5186static void igb_shutdown(struct pci_dev *pdev)
5187{
5188 igb_suspend(pdev, PMSG_SUSPEND);
5189}
5190
5191#ifdef CONFIG_NET_POLL_CONTROLLER
5192/*
5193 * Polling 'interrupt' - used by things like netconsole to send skbs
5194 * without having to re-enable interrupts. It's not called while
5195 * the interrupt routine is executing.
5196 */
5197static void igb_netpoll(struct net_device *netdev)
5198{
5199 struct igb_adapter *adapter = netdev_priv(netdev);
eebbbdba 5200 struct e1000_hw *hw = &adapter->hw;
9d5c8243 5201 int i;
9d5c8243 5202
eebbbdba
AD		 5203 if (!adapter->msix_entries) {
5204 igb_irq_disable(adapter);
5205 napi_schedule(&adapter->rx_ring[0].napi);
5206 return;
5207 }
9d5c8243 5208
eebbbdba
AD		 5209 for (i = 0; i < adapter->num_tx_queues; i++) {
5210 struct igb_ring *tx_ring = &adapter->tx_ring[i];
5211 wr32(E1000_EIMC, tx_ring->eims_value);
5212 igb_clean_tx_irq(tx_ring);
5213 wr32(E1000_EIMS, tx_ring->eims_value);
5214 }
9d5c8243 5215
eebbbdba
AD		 5216 for (i = 0; i < adapter->num_rx_queues; i++) {
5217 struct igb_ring *rx_ring = &adapter->rx_ring[i];
5218 wr32(E1000_EIMC, rx_ring->eims_value);
5219 napi_schedule(&rx_ring->napi);
5220 }
9d5c8243
AK		 5221}
5222#endif /* CONFIG_NET_POLL_CONTROLLER */
5223
5224/**
5225 * igb_io_error_detected - called when PCI error is detected
5226 * @pdev: Pointer to PCI device
5227 * @state: The current pci connection state
5228 *
5229 * This function is called after a PCI bus error affecting
5230 * this device has been detected.
5231 */
5232static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
5233 pci_channel_state_t state)
5234{
5235 struct net_device *netdev = pci_get_drvdata(pdev);
5236 struct igb_adapter *adapter = netdev_priv(netdev);
5237
5238 netif_device_detach(netdev);
5239
5240 if (netif_running(netdev))
5241 igb_down(adapter);
5242 pci_disable_device(pdev);
5243
5244 /* Request a slot slot reset. */
5245 return PCI_ERS_RESULT_NEED_RESET;
5246}
5247
5248/**
5249 * igb_io_slot_reset - called after the pci bus has been reset.
5250 * @pdev: Pointer to PCI device
5251 *
5252 * Restart the card from scratch, as if from a cold-boot. Implementation
5253 * resembles the first-half of the igb_resume routine.
5254 */
5255static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
5256{
5257 struct net_device *netdev = pci_get_drvdata(pdev);
5258 struct igb_adapter *adapter = netdev_priv(netdev);
5259 struct e1000_hw *hw = &adapter->hw;
40a914fa 5260 pci_ers_result_t result;
42bfd33a 5261 int err;
9d5c8243 5262
aed5dec3 5263 if (pci_enable_device_mem(pdev)) {
9d5c8243
AK		 5264 dev_err(&pdev->dev,
5265 "Cannot re-enable PCI device after reset.\n");
40a914fa
AD		 5266 result = PCI_ERS_RESULT_DISCONNECT;
5267 } else {
5268 pci_set_master(pdev);
5269 pci_restore_state(pdev);
9d5c8243 5270
40a914fa
AD		 5271 pci_enable_wake(pdev, PCI_D3hot, 0);
5272 pci_enable_wake(pdev, PCI_D3cold, 0);
9d5c8243 5273
40a914fa
AD		 5274 igb_reset(adapter);
5275 wr32(E1000_WUS, ~0);
5276 result = PCI_ERS_RESULT_RECOVERED;
5277 }
9d5c8243 5278
ea943d41
JK		 5279 err = pci_cleanup_aer_uncorrect_error_status(pdev);
5280 if (err) {
5281 dev_err(&pdev->dev, "pci_cleanup_aer_uncorrect_error_status "
5282 "failed 0x%0x\n", err);
5283 /* non-fatal, continue */
5284 }
40a914fa
AD		 5285
5286 return result;
9d5c8243
AK		 5287}
5288
5289/**
5290 * igb_io_resume - called when traffic can start flowing again.
5291 * @pdev: Pointer to PCI device
5292 *
5293 * This callback is called when the error recovery driver tells us that
5294 * its OK to resume normal operation. Implementation resembles the
5295 * second-half of the igb_resume routine.
5296 */
5297static void igb_io_resume(struct pci_dev *pdev)
5298{
5299 struct net_device *netdev = pci_get_drvdata(pdev);
5300 struct igb_adapter *adapter = netdev_priv(netdev);
5301
9d5c8243
AK		 5302 if (netif_running(netdev)) {
5303 if (igb_up(adapter)) {
5304 dev_err(&pdev->dev, "igb_up failed after reset\n");
5305 return;
5306 }
5307 }
5308
5309 netif_device_attach(netdev);
5310
5311 /* let the f/w know that the h/w is now under the control of the
5312 * driver. */
5313 igb_get_hw_control(adapter);
9d5c8243
AK		 5314}
5315
e1739522
AD		 5316static inline void igb_set_vmolr(struct e1000_hw *hw, int vfn)
5317{
5318 u32 reg_data;
5319
5320 reg_data = rd32(E1000_VMOLR(vfn));
5321 reg_data |= E1000_VMOLR_BAM | /* Accept broadcast */
5322 E1000_VMOLR_ROPE | /* Accept packets matched in UTA */
5323 E1000_VMOLR_ROMPE | /* Accept packets matched in MTA */
5324 E1000_VMOLR_AUPE | /* Accept untagged packets */
5325 E1000_VMOLR_STRVLAN; /* Strip vlan tags */
5326 wr32(E1000_VMOLR(vfn), reg_data);
5327}
5328
4ae196df
AD		 5329static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
5330 int vfn)
e1739522
AD		 5331{
5332 struct e1000_hw *hw = &adapter->hw;
5333 u32 vmolr;
5334
5335 vmolr = rd32(E1000_VMOLR(vfn));
5336 vmolr &= ~E1000_VMOLR_RLPML_MASK;
5337 vmolr |= size | E1000_VMOLR_LPE;
5338 wr32(E1000_VMOLR(vfn), vmolr);
4ae196df
AD		 5339
5340 return 0;
e1739522
AD		 5341}
5342
5343static inline void igb_set_rah_pool(struct e1000_hw *hw, int pool, int entry)
5344{
5345 u32 reg_data;
5346
5347 reg_data = rd32(E1000_RAH(entry));
5348 reg_data &= ~E1000_RAH_POOL_MASK;
5349 reg_data |= E1000_RAH_POOL_1 << pool;;
5350 wr32(E1000_RAH(entry), reg_data);
5351}
5352
5353static void igb_set_mc_list_pools(struct igb_adapter *adapter,
5354 int entry_count, u16 total_rar_filters)
5355{
5356 struct e1000_hw *hw = &adapter->hw;
5357 int i = adapter->vfs_allocated_count + 1;
5358
5359 if ((i + entry_count) < total_rar_filters)
5360 total_rar_filters = i + entry_count;
5361
5362 for (; i < total_rar_filters; i++)
5363 igb_set_rah_pool(hw, adapter->vfs_allocated_count, i);
5364}
5365
4ae196df
AD		 5366static int igb_set_vf_mac(struct igb_adapter *adapter,
5367 int vf, unsigned char *mac_addr)
5368{
5369 struct e1000_hw *hw = &adapter->hw;
5370 int rar_entry = vf + 1; /* VF MAC addresses start at entry 1 */
5371
5372 igb_rar_set(hw, mac_addr, rar_entry);
5373
37680117 5374 memcpy(adapter->vf_data[vf].vf_mac_addresses, mac_addr, ETH_ALEN);
4ae196df
AD		 5375
5376 igb_set_rah_pool(hw, vf, rar_entry);
5377
5378 return 0;
5379}
5380
5381static void igb_vmm_control(struct igb_adapter *adapter)
5382{
5383 struct e1000_hw *hw = &adapter->hw;
5384 u32 reg_data;
5385
5386 if (!adapter->vfs_allocated_count)
5387 return;
5388
5389 /* VF's need PF reset indication before they
5390 * can send/receive mail */
5391 reg_data = rd32(E1000_CTRL_EXT);
5392 reg_data |= E1000_CTRL_EXT_PFRSTD;
5393 wr32(E1000_CTRL_EXT, reg_data);
5394
5395 igb_vmdq_set_loopback_pf(hw, true);
5396 igb_vmdq_set_replication_pf(hw, true);
5397}
5398
37680117
AD		 5399#ifdef CONFIG_PCI_IOV
5400static ssize_t igb_show_num_vfs(struct device *dev,
5401 struct device_attribute *attr, char *buf)
5402{
5403 struct igb_adapter *adapter = netdev_priv(to_net_dev(dev));
5404
5405 return sprintf(buf, "%d\n", adapter->vfs_allocated_count);
5406}
5407
5408static ssize_t igb_set_num_vfs(struct device *dev,
5409 struct device_attribute *attr,
5410 const char *buf, size_t count)
5411{
5412 struct net_device *netdev = to_net_dev(dev);
5413 struct igb_adapter *adapter = netdev_priv(netdev);
5414 struct e1000_hw *hw = &adapter->hw;
5415 struct pci_dev *pdev = adapter->pdev;
5416 unsigned int num_vfs, i;
5417 unsigned char mac_addr[ETH_ALEN];
5418 int err;
5419
5420 sscanf(buf, "%u", &num_vfs);
5421
5422 if (num_vfs > 7)
5423 num_vfs = 7;
5424
5425 /* value unchanged do nothing */
5426 if (num_vfs == adapter->vfs_allocated_count)
5427 return count;
5428
5429 if (netdev->flags & IFF_UP)
5430 igb_close(netdev);
5431
5432 igb_reset_interrupt_capability(adapter);
5433 igb_free_queues(adapter);
5434 adapter->tx_ring = NULL;
5435 adapter->rx_ring = NULL;
5436 adapter->vfs_allocated_count = 0;
5437
5438 /* reclaim resources allocated to VFs since we are changing count */
5439 if (adapter->vf_data) {
5440 /* disable iov and allow time for transactions to clear */
5441 pci_disable_sriov(pdev);
5442 msleep(500);
5443
5444 kfree(adapter->vf_data);
5445 adapter->vf_data = NULL;
5446 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
5447 msleep(100);
5448 dev_info(&pdev->dev, "IOV Disabled\n");
5449 }
5450
5451 if (num_vfs) {
5452 adapter->vf_data = kcalloc(num_vfs,
5453 sizeof(struct vf_data_storage),
5454 GFP_KERNEL);
5455 if (!adapter->vf_data) {
5456 dev_err(&pdev->dev, "Could not allocate VF private "
5457 "data - IOV enable failed\n");
5458 } else {
5459 err = pci_enable_sriov(pdev, num_vfs);
5460 if (!err) {
5461 adapter->vfs_allocated_count = num_vfs;
5462 dev_info(&pdev->dev, "%d vfs allocated\n", num_vfs);
5463 for (i = 0; i < adapter->vfs_allocated_count; i++) {
5464 random_ether_addr(mac_addr);
5465 igb_set_vf_mac(adapter, i, mac_addr);
5466 }
5467 } else {
5468 kfree(adapter->vf_data);
5469 adapter->vf_data = NULL;
5470 }
5471 }
5472 }
5473
5474 igb_set_interrupt_capability(adapter);
5475 igb_alloc_queues(adapter);
5476 igb_reset(adapter);
5477
5478 if (netdev->flags & IFF_UP)
5479 igb_open(netdev);
5480
5481 return count;
5482}
5483#endif /* CONFIG_PCI_IOV */
9d5c8243 5484/* igb_main.c */
ubuntu-zesty-kernel mirror