aboutsummaryrefslogtreecommitdiffstats
path: root/drivers/net/ethernet/intel/igb/igb_main.c
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/net/ethernet/intel/igb/igb_main.c')
-rw-r--r--drivers/net/ethernet/intel/igb/igb_main.c7119
1 files changed, 7119 insertions, 0 deletions
diff --git a/drivers/net/ethernet/intel/igb/igb_main.c b/drivers/net/ethernet/intel/igb/igb_main.c
new file mode 100644
index 000000000000..ced544499f1b
--- /dev/null
+++ b/drivers/net/ethernet/intel/igb/igb_main.c
@@ -0,0 +1,7119 @@
1/*******************************************************************************
2
3 Intel(R) Gigabit Ethernet Linux driver
4 Copyright(c) 2007-2011 Intel Corporation.
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/bitops.h>
32#include <linux/vmalloc.h>
33#include <linux/pagemap.h>
34#include <linux/netdevice.h>
35#include <linux/ipv6.h>
36#include <linux/slab.h>
37#include <net/checksum.h>
38#include <net/ip6_checksum.h>
39#include <linux/net_tstamp.h>
40#include <linux/mii.h>
41#include <linux/ethtool.h>
42#include <linux/if.h>
43#include <linux/if_vlan.h>
44#include <linux/pci.h>
45#include <linux/pci-aspm.h>
46#include <linux/delay.h>
47#include <linux/interrupt.h>
48#include <linux/ip.h>
49#include <linux/tcp.h>
50#include <linux/sctp.h>
51#include <linux/if_ether.h>
52#include <linux/aer.h>
53#include <linux/prefetch.h>
54#ifdef CONFIG_IGB_DCA
55#include <linux/dca.h>
56#endif
57#include "igb.h"
58
59#define MAJ 3
60#define MIN 2
61#define BUILD 10
62#define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) "." \
63__stringify(BUILD) "-k"
64char igb_driver_name[] = "igb";
65char igb_driver_version[] = DRV_VERSION;
66static const char igb_driver_string[] =
67 "Intel(R) Gigabit Ethernet Network Driver";
68static const char igb_copyright[] = "Copyright (c) 2007-2011 Intel Corporation.";
69
70static const struct e1000_info *igb_info_tbl[] = {
71 [board_82575] = &e1000_82575_info,
72};
73
74static DEFINE_PCI_DEVICE_TABLE(igb_pci_tbl) = {
75 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_COPPER), board_82575 },
76 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_FIBER), board_82575 },
77 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SERDES), board_82575 },
78 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SGMII), board_82575 },
79 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER), board_82575 },
80 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_FIBER), board_82575 },
81 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_QUAD_FIBER), board_82575 },
82 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SERDES), board_82575 },
83 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SGMII), board_82575 },
84 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER_DUAL), board_82575 },
85 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SGMII), board_82575 },
86 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SERDES), board_82575 },
87 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_BACKPLANE), board_82575 },
88 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SFP), board_82575 },
89 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576), board_82575 },
90 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS), board_82575 },
91 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS_SERDES), board_82575 },
92 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_FIBER), board_82575 },
93 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES), board_82575 },
94 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES_QUAD), board_82575 },
95 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER_ET2), board_82575 },
96 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER), board_82575 },
97 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
98 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
99 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
100 /* required last entry */
101 {0, }
102};
103
104MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
105
106void igb_reset(struct igb_adapter *);
107static int igb_setup_all_tx_resources(struct igb_adapter *);
108static int igb_setup_all_rx_resources(struct igb_adapter *);
109static void igb_free_all_tx_resources(struct igb_adapter *);
110static void igb_free_all_rx_resources(struct igb_adapter *);
111static void igb_setup_mrqc(struct igb_adapter *);
112static int igb_probe(struct pci_dev *, const struct pci_device_id *);
113static void __devexit igb_remove(struct pci_dev *pdev);
114static void igb_init_hw_timer(struct igb_adapter *adapter);
115static int igb_sw_init(struct igb_adapter *);
116static int igb_open(struct net_device *);
117static int igb_close(struct net_device *);
118static void igb_configure_tx(struct igb_adapter *);
119static void igb_configure_rx(struct igb_adapter *);
120static void igb_clean_all_tx_rings(struct igb_adapter *);
121static void igb_clean_all_rx_rings(struct igb_adapter *);
122static void igb_clean_tx_ring(struct igb_ring *);
123static void igb_clean_rx_ring(struct igb_ring *);
124static void igb_set_rx_mode(struct net_device *);
125static void igb_update_phy_info(unsigned long);
126static void igb_watchdog(unsigned long);
127static void igb_watchdog_task(struct work_struct *);
128static netdev_tx_t igb_xmit_frame(struct sk_buff *skb, struct net_device *);
129static struct rtnl_link_stats64 *igb_get_stats64(struct net_device *dev,
130 struct rtnl_link_stats64 *stats);
131static int igb_change_mtu(struct net_device *, int);
132static int igb_set_mac(struct net_device *, void *);
133static void igb_set_uta(struct igb_adapter *adapter);
134static irqreturn_t igb_intr(int irq, void *);
135static irqreturn_t igb_intr_msi(int irq, void *);
136static irqreturn_t igb_msix_other(int irq, void *);
137static irqreturn_t igb_msix_ring(int irq, void *);
138#ifdef CONFIG_IGB_DCA
139static void igb_update_dca(struct igb_q_vector *);
140static void igb_setup_dca(struct igb_adapter *);
141#endif /* CONFIG_IGB_DCA */
142static int igb_poll(struct napi_struct *, int);
143static bool igb_clean_tx_irq(struct igb_q_vector *);
144static bool igb_clean_rx_irq(struct igb_q_vector *, int);
145static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
146static void igb_tx_timeout(struct net_device *);
147static void igb_reset_task(struct work_struct *);
148static void igb_vlan_mode(struct net_device *netdev, u32 features);
149static void igb_vlan_rx_add_vid(struct net_device *, u16);
150static void igb_vlan_rx_kill_vid(struct net_device *, u16);
151static void igb_restore_vlan(struct igb_adapter *);
152static void igb_rar_set_qsel(struct igb_adapter *, u8 *, u32 , u8);
153static void igb_ping_all_vfs(struct igb_adapter *);
154static void igb_msg_task(struct igb_adapter *);
155static void igb_vmm_control(struct igb_adapter *);
156static int igb_set_vf_mac(struct igb_adapter *, int, unsigned char *);
157static void igb_restore_vf_multicasts(struct igb_adapter *adapter);
158static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac);
159static int igb_ndo_set_vf_vlan(struct net_device *netdev,
160 int vf, u16 vlan, u8 qos);
161static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf, int tx_rate);
162static int igb_ndo_get_vf_config(struct net_device *netdev, int vf,
163 struct ifla_vf_info *ivi);
164static void igb_check_vf_rate_limit(struct igb_adapter *);
165
166#ifdef CONFIG_PCI_IOV
167static int igb_vf_configure(struct igb_adapter *adapter, int vf);
168static int igb_find_enabled_vfs(struct igb_adapter *adapter);
169static int igb_check_vf_assignment(struct igb_adapter *adapter);
170#endif
171
172#ifdef CONFIG_PM
173static int igb_suspend(struct pci_dev *, pm_message_t);
174static int igb_resume(struct pci_dev *);
175#endif
176static void igb_shutdown(struct pci_dev *);
177#ifdef CONFIG_IGB_DCA
178static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
179static struct notifier_block dca_notifier = {
180 .notifier_call = igb_notify_dca,
181 .next = NULL,
182 .priority = 0
183};
184#endif
185#ifdef CONFIG_NET_POLL_CONTROLLER
186/* for netdump / net console */
187static void igb_netpoll(struct net_device *);
188#endif
189#ifdef CONFIG_PCI_IOV
190static unsigned int max_vfs = 0;
191module_param(max_vfs, uint, 0);
192MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate "
193 "per physical function");
194#endif /* CONFIG_PCI_IOV */
195
196static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
197 pci_channel_state_t);
198static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
199static void igb_io_resume(struct pci_dev *);
200
201static struct pci_error_handlers igb_err_handler = {
202 .error_detected = igb_io_error_detected,
203 .slot_reset = igb_io_slot_reset,
204 .resume = igb_io_resume,
205};
206
207static void igb_init_dmac(struct igb_adapter *adapter, u32 pba);
208
209static struct pci_driver igb_driver = {
210 .name = igb_driver_name,
211 .id_table = igb_pci_tbl,
212 .probe = igb_probe,
213 .remove = __devexit_p(igb_remove),
214#ifdef CONFIG_PM
215 /* Power Management Hooks */
216 .suspend = igb_suspend,
217 .resume = igb_resume,
218#endif
219 .shutdown = igb_shutdown,
220 .err_handler = &igb_err_handler
221};
222
223MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
224MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
225MODULE_LICENSE("GPL");
226MODULE_VERSION(DRV_VERSION);
227
228struct igb_reg_info {
229 u32 ofs;
230 char *name;
231};
232
233static const struct igb_reg_info igb_reg_info_tbl[] = {
234
235 /* General Registers */
236 {E1000_CTRL, "CTRL"},
237 {E1000_STATUS, "STATUS"},
238 {E1000_CTRL_EXT, "CTRL_EXT"},
239
240 /* Interrupt Registers */
241 {E1000_ICR, "ICR"},
242
243 /* RX Registers */
244 {E1000_RCTL, "RCTL"},
245 {E1000_RDLEN(0), "RDLEN"},
246 {E1000_RDH(0), "RDH"},
247 {E1000_RDT(0), "RDT"},
248 {E1000_RXDCTL(0), "RXDCTL"},
249 {E1000_RDBAL(0), "RDBAL"},
250 {E1000_RDBAH(0), "RDBAH"},
251
252 /* TX Registers */
253 {E1000_TCTL, "TCTL"},
254 {E1000_TDBAL(0), "TDBAL"},
255 {E1000_TDBAH(0), "TDBAH"},
256 {E1000_TDLEN(0), "TDLEN"},
257 {E1000_TDH(0), "TDH"},
258 {E1000_TDT(0), "TDT"},
259 {E1000_TXDCTL(0), "TXDCTL"},
260 {E1000_TDFH, "TDFH"},
261 {E1000_TDFT, "TDFT"},
262 {E1000_TDFHS, "TDFHS"},
263 {E1000_TDFPC, "TDFPC"},
264
265 /* List Terminator */
266 {}
267};
268
269/*
270 * igb_regdump - register printout routine
271 */
272static void igb_regdump(struct e1000_hw *hw, struct igb_reg_info *reginfo)
273{
274 int n = 0;
275 char rname[16];
276 u32 regs[8];
277
278 switch (reginfo->ofs) {
279 case E1000_RDLEN(0):
280 for (n = 0; n < 4; n++)
281 regs[n] = rd32(E1000_RDLEN(n));
282 break;
283 case E1000_RDH(0):
284 for (n = 0; n < 4; n++)
285 regs[n] = rd32(E1000_RDH(n));
286 break;
287 case E1000_RDT(0):
288 for (n = 0; n < 4; n++)
289 regs[n] = rd32(E1000_RDT(n));
290 break;
291 case E1000_RXDCTL(0):
292 for (n = 0; n < 4; n++)
293 regs[n] = rd32(E1000_RXDCTL(n));
294 break;
295 case E1000_RDBAL(0):
296 for (n = 0; n < 4; n++)
297 regs[n] = rd32(E1000_RDBAL(n));
298 break;
299 case E1000_RDBAH(0):
300 for (n = 0; n < 4; n++)
301 regs[n] = rd32(E1000_RDBAH(n));
302 break;
303 case E1000_TDBAL(0):
304 for (n = 0; n < 4; n++)
305 regs[n] = rd32(E1000_RDBAL(n));
306 break;
307 case E1000_TDBAH(0):
308 for (n = 0; n < 4; n++)
309 regs[n] = rd32(E1000_TDBAH(n));
310 break;
311 case E1000_TDLEN(0):
312 for (n = 0; n < 4; n++)
313 regs[n] = rd32(E1000_TDLEN(n));
314 break;
315 case E1000_TDH(0):
316 for (n = 0; n < 4; n++)
317 regs[n] = rd32(E1000_TDH(n));
318 break;
319 case E1000_TDT(0):
320 for (n = 0; n < 4; n++)
321 regs[n] = rd32(E1000_TDT(n));
322 break;
323 case E1000_TXDCTL(0):
324 for (n = 0; n < 4; n++)
325 regs[n] = rd32(E1000_TXDCTL(n));
326 break;
327 default:
328 printk(KERN_INFO "%-15s %08x\n",
329 reginfo->name, rd32(reginfo->ofs));
330 return;
331 }
332
333 snprintf(rname, 16, "%s%s", reginfo->name, "[0-3]");
334 printk(KERN_INFO "%-15s ", rname);
335 for (n = 0; n < 4; n++)
336 printk(KERN_CONT "%08x ", regs[n]);
337 printk(KERN_CONT "\n");
338}
339
340/*
341 * igb_dump - Print registers, tx-rings and rx-rings
342 */
343static void igb_dump(struct igb_adapter *adapter)
344{
345 struct net_device *netdev = adapter->netdev;
346 struct e1000_hw *hw = &adapter->hw;
347 struct igb_reg_info *reginfo;
348 struct igb_ring *tx_ring;
349 union e1000_adv_tx_desc *tx_desc;
350 struct my_u0 { u64 a; u64 b; } *u0;
351 struct igb_ring *rx_ring;
352 union e1000_adv_rx_desc *rx_desc;
353 u32 staterr;
354 u16 i, n;
355
356 if (!netif_msg_hw(adapter))
357 return;
358
359 /* Print netdevice Info */
360 if (netdev) {
361 dev_info(&adapter->pdev->dev, "Net device Info\n");
362 printk(KERN_INFO "Device Name state "
363 "trans_start last_rx\n");
364 printk(KERN_INFO "%-15s %016lX %016lX %016lX\n",
365 netdev->name,
366 netdev->state,
367 netdev->trans_start,
368 netdev->last_rx);
369 }
370
371 /* Print Registers */
372 dev_info(&adapter->pdev->dev, "Register Dump\n");
373 printk(KERN_INFO " Register Name Value\n");
374 for (reginfo = (struct igb_reg_info *)igb_reg_info_tbl;
375 reginfo->name; reginfo++) {
376 igb_regdump(hw, reginfo);
377 }
378
379 /* Print TX Ring Summary */
380 if (!netdev || !netif_running(netdev))
381 goto exit;
382
383 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
384 printk(KERN_INFO "Queue [NTU] [NTC] [bi(ntc)->dma ]"
385 " leng ntw timestamp\n");
386 for (n = 0; n < adapter->num_tx_queues; n++) {
387 struct igb_tx_buffer *buffer_info;
388 tx_ring = adapter->tx_ring[n];
389 buffer_info = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
390 printk(KERN_INFO " %5d %5X %5X %016llX %04X %p %016llX\n",
391 n, tx_ring->next_to_use, tx_ring->next_to_clean,
392 (u64)buffer_info->dma,
393 buffer_info->length,
394 buffer_info->next_to_watch,
395 (u64)buffer_info->time_stamp);
396 }
397
398 /* Print TX Rings */
399 if (!netif_msg_tx_done(adapter))
400 goto rx_ring_summary;
401
402 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
403
404 /* Transmit Descriptor Formats
405 *
406 * Advanced Transmit Descriptor
407 * +--------------------------------------------------------------+
408 * 0 | Buffer Address [63:0] |
409 * +--------------------------------------------------------------+
410 * 8 | PAYLEN | PORTS |CC|IDX | STA | DCMD |DTYP|MAC|RSV| DTALEN |
411 * +--------------------------------------------------------------+
412 * 63 46 45 40 39 38 36 35 32 31 24 15 0
413 */
414
415 for (n = 0; n < adapter->num_tx_queues; n++) {
416 tx_ring = adapter->tx_ring[n];
417 printk(KERN_INFO "------------------------------------\n");
418 printk(KERN_INFO "TX QUEUE INDEX = %d\n", tx_ring->queue_index);
419 printk(KERN_INFO "------------------------------------\n");
420 printk(KERN_INFO "T [desc] [address 63:0 ] "
421 "[PlPOCIStDDM Ln] [bi->dma ] "
422 "leng ntw timestamp bi->skb\n");
423
424 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
425 struct igb_tx_buffer *buffer_info;
426 tx_desc = IGB_TX_DESC(tx_ring, i);
427 buffer_info = &tx_ring->tx_buffer_info[i];
428 u0 = (struct my_u0 *)tx_desc;
429 printk(KERN_INFO "T [0x%03X] %016llX %016llX %016llX"
430 " %04X %p %016llX %p", i,
431 le64_to_cpu(u0->a),
432 le64_to_cpu(u0->b),
433 (u64)buffer_info->dma,
434 buffer_info->length,
435 buffer_info->next_to_watch,
436 (u64)buffer_info->time_stamp,
437 buffer_info->skb);
438 if (i == tx_ring->next_to_use &&
439 i == tx_ring->next_to_clean)
440 printk(KERN_CONT " NTC/U\n");
441 else if (i == tx_ring->next_to_use)
442 printk(KERN_CONT " NTU\n");
443 else if (i == tx_ring->next_to_clean)
444 printk(KERN_CONT " NTC\n");
445 else
446 printk(KERN_CONT "\n");
447
448 if (netif_msg_pktdata(adapter) && buffer_info->dma != 0)
449 print_hex_dump(KERN_INFO, "",
450 DUMP_PREFIX_ADDRESS,
451 16, 1, phys_to_virt(buffer_info->dma),
452 buffer_info->length, true);
453 }
454 }
455
456 /* Print RX Rings Summary */
457rx_ring_summary:
458 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
459 printk(KERN_INFO "Queue [NTU] [NTC]\n");
460 for (n = 0; n < adapter->num_rx_queues; n++) {
461 rx_ring = adapter->rx_ring[n];
462 printk(KERN_INFO " %5d %5X %5X\n", n,
463 rx_ring->next_to_use, rx_ring->next_to_clean);
464 }
465
466 /* Print RX Rings */
467 if (!netif_msg_rx_status(adapter))
468 goto exit;
469
470 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
471
472 /* Advanced Receive Descriptor (Read) Format
473 * 63 1 0
474 * +-----------------------------------------------------+
475 * 0 | Packet Buffer Address [63:1] |A0/NSE|
476 * +----------------------------------------------+------+
477 * 8 | Header Buffer Address [63:1] | DD |
478 * +-----------------------------------------------------+
479 *
480 *
481 * Advanced Receive Descriptor (Write-Back) Format
482 *
483 * 63 48 47 32 31 30 21 20 17 16 4 3 0
484 * +------------------------------------------------------+
485 * 0 | Packet IP |SPH| HDR_LEN | RSV|Packet| RSS |
486 * | Checksum Ident | | | | Type | Type |
487 * +------------------------------------------------------+
488 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
489 * +------------------------------------------------------+
490 * 63 48 47 32 31 20 19 0
491 */
492
493 for (n = 0; n < adapter->num_rx_queues; n++) {
494 rx_ring = adapter->rx_ring[n];
495 printk(KERN_INFO "------------------------------------\n");
496 printk(KERN_INFO "RX QUEUE INDEX = %d\n", rx_ring->queue_index);
497 printk(KERN_INFO "------------------------------------\n");
498 printk(KERN_INFO "R [desc] [ PktBuf A0] "
499 "[ HeadBuf DD] [bi->dma ] [bi->skb] "
500 "<-- Adv Rx Read format\n");
501 printk(KERN_INFO "RWB[desc] [PcsmIpSHl PtRs] "
502 "[vl er S cks ln] ---------------- [bi->skb] "
503 "<-- Adv Rx Write-Back format\n");
504
505 for (i = 0; i < rx_ring->count; i++) {
506 struct igb_rx_buffer *buffer_info;
507 buffer_info = &rx_ring->rx_buffer_info[i];
508 rx_desc = IGB_RX_DESC(rx_ring, i);
509 u0 = (struct my_u0 *)rx_desc;
510 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
511 if (staterr & E1000_RXD_STAT_DD) {
512 /* Descriptor Done */
513 printk(KERN_INFO "RWB[0x%03X] %016llX "
514 "%016llX ---------------- %p", i,
515 le64_to_cpu(u0->a),
516 le64_to_cpu(u0->b),
517 buffer_info->skb);
518 } else {
519 printk(KERN_INFO "R [0x%03X] %016llX "
520 "%016llX %016llX %p", i,
521 le64_to_cpu(u0->a),
522 le64_to_cpu(u0->b),
523 (u64)buffer_info->dma,
524 buffer_info->skb);
525
526 if (netif_msg_pktdata(adapter)) {
527 print_hex_dump(KERN_INFO, "",
528 DUMP_PREFIX_ADDRESS,
529 16, 1,
530 phys_to_virt(buffer_info->dma),
531 IGB_RX_HDR_LEN, true);
532 print_hex_dump(KERN_INFO, "",
533 DUMP_PREFIX_ADDRESS,
534 16, 1,
535 phys_to_virt(
536 buffer_info->page_dma +
537 buffer_info->page_offset),
538 PAGE_SIZE/2, true);
539 }
540 }
541
542 if (i == rx_ring->next_to_use)
543 printk(KERN_CONT " NTU\n");
544 else if (i == rx_ring->next_to_clean)
545 printk(KERN_CONT " NTC\n");
546 else
547 printk(KERN_CONT "\n");
548
549 }
550 }
551
552exit:
553 return;
554}
555
556
557/**
558 * igb_read_clock - read raw cycle counter (to be used by time counter)
559 */
560static cycle_t igb_read_clock(const struct cyclecounter *tc)
561{
562 struct igb_adapter *adapter =
563 container_of(tc, struct igb_adapter, cycles);
564 struct e1000_hw *hw = &adapter->hw;
565 u64 stamp = 0;
566 int shift = 0;
567
568 /*
569 * The timestamp latches on lowest register read. For the 82580
570 * the lowest register is SYSTIMR instead of SYSTIML. However we never
571 * adjusted TIMINCA so SYSTIMR will just read as all 0s so ignore it.
572 */
573 if (hw->mac.type >= e1000_82580) {
574 stamp = rd32(E1000_SYSTIMR) >> 8;
575 shift = IGB_82580_TSYNC_SHIFT;
576 }
577
578 stamp |= (u64)rd32(E1000_SYSTIML) << shift;
579 stamp |= (u64)rd32(E1000_SYSTIMH) << (shift + 32);
580 return stamp;
581}
582
583/**
584 * igb_get_hw_dev - return device
585 * used by hardware layer to print debugging information
586 **/
587struct net_device *igb_get_hw_dev(struct e1000_hw *hw)
588{
589 struct igb_adapter *adapter = hw->back;
590 return adapter->netdev;
591}
592
593/**
594 * igb_init_module - Driver Registration Routine
595 *
596 * igb_init_module is the first routine called when the driver is
597 * loaded. All it does is register with the PCI subsystem.
598 **/
599static int __init igb_init_module(void)
600{
601 int ret;
602 printk(KERN_INFO "%s - version %s\n",
603 igb_driver_string, igb_driver_version);
604
605 printk(KERN_INFO "%s\n", igb_copyright);
606
607#ifdef CONFIG_IGB_DCA
608 dca_register_notify(&dca_notifier);
609#endif
610 ret = pci_register_driver(&igb_driver);
611 return ret;
612}
613
614module_init(igb_init_module);
615
616/**
617 * igb_exit_module - Driver Exit Cleanup Routine
618 *
619 * igb_exit_module is called just before the driver is removed
620 * from memory.
621 **/
622static void __exit igb_exit_module(void)
623{
624#ifdef CONFIG_IGB_DCA
625 dca_unregister_notify(&dca_notifier);
626#endif
627 pci_unregister_driver(&igb_driver);
628}
629
630module_exit(igb_exit_module);
631
632#define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
633/**
634 * igb_cache_ring_register - Descriptor ring to register mapping
635 * @adapter: board private structure to initialize
636 *
637 * Once we know the feature-set enabled for the device, we'll cache
638 * the register offset the descriptor ring is assigned to.
639 **/
640static void igb_cache_ring_register(struct igb_adapter *adapter)
641{
642 int i = 0, j = 0;
643 u32 rbase_offset = adapter->vfs_allocated_count;
644
645 switch (adapter->hw.mac.type) {
646 case e1000_82576:
647 /* The queues are allocated for virtualization such that VF 0
648 * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
649 * In order to avoid collision we start at the first free queue
650 * and continue consuming queues in the same sequence
651 */
652 if (adapter->vfs_allocated_count) {
653 for (; i < adapter->rss_queues; i++)
654 adapter->rx_ring[i]->reg_idx = rbase_offset +
655 Q_IDX_82576(i);
656 }
657 case e1000_82575:
658 case e1000_82580:
659 case e1000_i350:
660 default:
661 for (; i < adapter->num_rx_queues; i++)
662 adapter->rx_ring[i]->reg_idx = rbase_offset + i;
663 for (; j < adapter->num_tx_queues; j++)
664 adapter->tx_ring[j]->reg_idx = rbase_offset + j;
665 break;
666 }
667}
668
669static void igb_free_queues(struct igb_adapter *adapter)
670{
671 int i;
672
673 for (i = 0; i < adapter->num_tx_queues; i++) {
674 kfree(adapter->tx_ring[i]);
675 adapter->tx_ring[i] = NULL;
676 }
677 for (i = 0; i < adapter->num_rx_queues; i++) {
678 kfree(adapter->rx_ring[i]);
679 adapter->rx_ring[i] = NULL;
680 }
681 adapter->num_rx_queues = 0;
682 adapter->num_tx_queues = 0;
683}
684
685/**
686 * igb_alloc_queues - Allocate memory for all rings
687 * @adapter: board private structure to initialize
688 *
689 * We allocate one ring per queue at run-time since we don't know the
690 * number of queues at compile-time.
691 **/
692static int igb_alloc_queues(struct igb_adapter *adapter)
693{
694 struct igb_ring *ring;
695 int i;
696 int orig_node = adapter->node;
697
698 for (i = 0; i < adapter->num_tx_queues; i++) {
699 if (orig_node == -1) {
700 int cur_node = next_online_node(adapter->node);
701 if (cur_node == MAX_NUMNODES)
702 cur_node = first_online_node;
703 adapter->node = cur_node;
704 }
705 ring = kzalloc_node(sizeof(struct igb_ring), GFP_KERNEL,
706 adapter->node);
707 if (!ring)
708 ring = kzalloc(sizeof(struct igb_ring), GFP_KERNEL);
709 if (!ring)
710 goto err;
711 ring->count = adapter->tx_ring_count;
712 ring->queue_index = i;
713 ring->dev = &adapter->pdev->dev;
714 ring->netdev = adapter->netdev;
715 ring->numa_node = adapter->node;
716 /* For 82575, context index must be unique per ring. */
717 if (adapter->hw.mac.type == e1000_82575)
718 set_bit(IGB_RING_FLAG_TX_CTX_IDX, &ring->flags);
719 adapter->tx_ring[i] = ring;
720 }
721 /* Restore the adapter's original node */
722 adapter->node = orig_node;
723
724 for (i = 0; i < adapter->num_rx_queues; i++) {
725 if (orig_node == -1) {
726 int cur_node = next_online_node(adapter->node);
727 if (cur_node == MAX_NUMNODES)
728 cur_node = first_online_node;
729 adapter->node = cur_node;
730 }
731 ring = kzalloc_node(sizeof(struct igb_ring), GFP_KERNEL,
732 adapter->node);
733 if (!ring)
734 ring = kzalloc(sizeof(struct igb_ring), GFP_KERNEL);
735 if (!ring)
736 goto err;
737 ring->count = adapter->rx_ring_count;
738 ring->queue_index = i;
739 ring->dev = &adapter->pdev->dev;
740 ring->netdev = adapter->netdev;
741 ring->numa_node = adapter->node;
742 /* set flag indicating ring supports SCTP checksum offload */
743 if (adapter->hw.mac.type >= e1000_82576)
744 set_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags);
745
746 /* On i350, loopback VLAN packets have the tag byte-swapped. */
747 if (adapter->hw.mac.type == e1000_i350)
748 set_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &ring->flags);
749
750 adapter->rx_ring[i] = ring;
751 }
752 /* Restore the adapter's original node */
753 adapter->node = orig_node;
754
755 igb_cache_ring_register(adapter);
756
757 return 0;
758
759err:
760 /* Restore the adapter's original node */
761 adapter->node = orig_node;
762 igb_free_queues(adapter);
763
764 return -ENOMEM;
765}
766
767/**
768 * igb_write_ivar - configure ivar for given MSI-X vector
769 * @hw: pointer to the HW structure
770 * @msix_vector: vector number we are allocating to a given ring
771 * @index: row index of IVAR register to write within IVAR table
772 * @offset: column offset of in IVAR, should be multiple of 8
773 *
774 * This function is intended to handle the writing of the IVAR register
775 * for adapters 82576 and newer. The IVAR table consists of 2 columns,
776 * each containing an cause allocation for an Rx and Tx ring, and a
777 * variable number of rows depending on the number of queues supported.
778 **/
779static void igb_write_ivar(struct e1000_hw *hw, int msix_vector,
780 int index, int offset)
781{
782 u32 ivar = array_rd32(E1000_IVAR0, index);
783
784 /* clear any bits that are currently set */
785 ivar &= ~((u32)0xFF << offset);
786
787 /* write vector and valid bit */
788 ivar |= (msix_vector | E1000_IVAR_VALID) << offset;
789
790 array_wr32(E1000_IVAR0, index, ivar);
791}
792
793#define IGB_N0_QUEUE -1
794static void igb_assign_vector(struct igb_q_vector *q_vector, int msix_vector)
795{
796 struct igb_adapter *adapter = q_vector->adapter;
797 struct e1000_hw *hw = &adapter->hw;
798 int rx_queue = IGB_N0_QUEUE;
799 int tx_queue = IGB_N0_QUEUE;
800 u32 msixbm = 0;
801
802 if (q_vector->rx.ring)
803 rx_queue = q_vector->rx.ring->reg_idx;
804 if (q_vector->tx.ring)
805 tx_queue = q_vector->tx.ring->reg_idx;
806
807 switch (hw->mac.type) {
808 case e1000_82575:
809 /* The 82575 assigns vectors using a bitmask, which matches the
810 bitmask for the EICR/EIMS/EIMC registers. To assign one
811 or more queues to a vector, we write the appropriate bits
812 into the MSIXBM register for that vector. */
813 if (rx_queue > IGB_N0_QUEUE)
814 msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
815 if (tx_queue > IGB_N0_QUEUE)
816 msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
817 if (!adapter->msix_entries && msix_vector == 0)
818 msixbm |= E1000_EIMS_OTHER;
819 array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
820 q_vector->eims_value = msixbm;
821 break;
822 case e1000_82576:
823 /*
824 * 82576 uses a table that essentially consists of 2 columns
825 * with 8 rows. The ordering is column-major so we use the
826 * lower 3 bits as the row index, and the 4th bit as the
827 * column offset.
828 */
829 if (rx_queue > IGB_N0_QUEUE)
830 igb_write_ivar(hw, msix_vector,
831 rx_queue & 0x7,
832 (rx_queue & 0x8) << 1);
833 if (tx_queue > IGB_N0_QUEUE)
834 igb_write_ivar(hw, msix_vector,
835 tx_queue & 0x7,
836 ((tx_queue & 0x8) << 1) + 8);
837 q_vector->eims_value = 1 << msix_vector;
838 break;
839 case e1000_82580:
840 case e1000_i350:
841 /*
842 * On 82580 and newer adapters the scheme is similar to 82576
843 * however instead of ordering column-major we have things
844 * ordered row-major. So we traverse the table by using
845 * bit 0 as the column offset, and the remaining bits as the
846 * row index.
847 */
848 if (rx_queue > IGB_N0_QUEUE)
849 igb_write_ivar(hw, msix_vector,
850 rx_queue >> 1,
851 (rx_queue & 0x1) << 4);
852 if (tx_queue > IGB_N0_QUEUE)
853 igb_write_ivar(hw, msix_vector,
854 tx_queue >> 1,
855 ((tx_queue & 0x1) << 4) + 8);
856 q_vector->eims_value = 1 << msix_vector;
857 break;
858 default:
859 BUG();
860 break;
861 }
862
863 /* add q_vector eims value to global eims_enable_mask */
864 adapter->eims_enable_mask |= q_vector->eims_value;
865
866 /* configure q_vector to set itr on first interrupt */
867 q_vector->set_itr = 1;
868}
869
870/**
871 * igb_configure_msix - Configure MSI-X hardware
872 *
873 * igb_configure_msix sets up the hardware to properly
874 * generate MSI-X interrupts.
875 **/
876static void igb_configure_msix(struct igb_adapter *adapter)
877{
878 u32 tmp;
879 int i, vector = 0;
880 struct e1000_hw *hw = &adapter->hw;
881
882 adapter->eims_enable_mask = 0;
883
884 /* set vector for other causes, i.e. link changes */
885 switch (hw->mac.type) {
886 case e1000_82575:
887 tmp = rd32(E1000_CTRL_EXT);
888 /* enable MSI-X PBA support*/
889 tmp |= E1000_CTRL_EXT_PBA_CLR;
890
891 /* Auto-Mask interrupts upon ICR read. */
892 tmp |= E1000_CTRL_EXT_EIAME;
893 tmp |= E1000_CTRL_EXT_IRCA;
894
895 wr32(E1000_CTRL_EXT, tmp);
896
897 /* enable msix_other interrupt */
898 array_wr32(E1000_MSIXBM(0), vector++,
899 E1000_EIMS_OTHER);
900 adapter->eims_other = E1000_EIMS_OTHER;
901
902 break;
903
904 case e1000_82576:
905 case e1000_82580:
906 case e1000_i350:
907 /* Turn on MSI-X capability first, or our settings
908 * won't stick. And it will take days to debug. */
909 wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
910 E1000_GPIE_PBA | E1000_GPIE_EIAME |
911 E1000_GPIE_NSICR);
912
913 /* enable msix_other interrupt */
914 adapter->eims_other = 1 << vector;
915 tmp = (vector++ | E1000_IVAR_VALID) << 8;
916
917 wr32(E1000_IVAR_MISC, tmp);
918 break;
919 default:
920 /* do nothing, since nothing else supports MSI-X */
921 break;
922 } /* switch (hw->mac.type) */
923
924 adapter->eims_enable_mask |= adapter->eims_other;
925
926 for (i = 0; i < adapter->num_q_vectors; i++)
927 igb_assign_vector(adapter->q_vector[i], vector++);
928
929 wrfl();
930}
931
932/**
933 * igb_request_msix - Initialize MSI-X interrupts
934 *
935 * igb_request_msix allocates MSI-X vectors and requests interrupts from the
936 * kernel.
937 **/
938static int igb_request_msix(struct igb_adapter *adapter)
939{
940 struct net_device *netdev = adapter->netdev;
941 struct e1000_hw *hw = &adapter->hw;
942 int i, err = 0, vector = 0;
943
944 err = request_irq(adapter->msix_entries[vector].vector,
945 igb_msix_other, 0, netdev->name, adapter);
946 if (err)
947 goto out;
948 vector++;
949
950 for (i = 0; i < adapter->num_q_vectors; i++) {
951 struct igb_q_vector *q_vector = adapter->q_vector[i];
952
953 q_vector->itr_register = hw->hw_addr + E1000_EITR(vector);
954
955 if (q_vector->rx.ring && q_vector->tx.ring)
956 sprintf(q_vector->name, "%s-TxRx-%u", netdev->name,
957 q_vector->rx.ring->queue_index);
958 else if (q_vector->tx.ring)
959 sprintf(q_vector->name, "%s-tx-%u", netdev->name,
960 q_vector->tx.ring->queue_index);
961 else if (q_vector->rx.ring)
962 sprintf(q_vector->name, "%s-rx-%u", netdev->name,
963 q_vector->rx.ring->queue_index);
964 else
965 sprintf(q_vector->name, "%s-unused", netdev->name);
966
967 err = request_irq(adapter->msix_entries[vector].vector,
968 igb_msix_ring, 0, q_vector->name,
969 q_vector);
970 if (err)
971 goto out;
972 vector++;
973 }
974
975 igb_configure_msix(adapter);
976 return 0;
977out:
978 return err;
979}
980
981static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
982{
983 if (adapter->msix_entries) {
984 pci_disable_msix(adapter->pdev);
985 kfree(adapter->msix_entries);
986 adapter->msix_entries = NULL;
987 } else if (adapter->flags & IGB_FLAG_HAS_MSI) {
988 pci_disable_msi(adapter->pdev);
989 }
990}
991
992/**
993 * igb_free_q_vectors - Free memory allocated for interrupt vectors
994 * @adapter: board private structure to initialize
995 *
996 * This function frees the memory allocated to the q_vectors. In addition if
997 * NAPI is enabled it will delete any references to the NAPI struct prior
998 * to freeing the q_vector.
999 **/
1000static void igb_free_q_vectors(struct igb_adapter *adapter)
1001{
1002 int v_idx;
1003
1004 for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
1005 struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
1006 adapter->q_vector[v_idx] = NULL;
1007 if (!q_vector)
1008 continue;
1009 netif_napi_del(&q_vector->napi);
1010 kfree(q_vector);
1011 }
1012 adapter->num_q_vectors = 0;
1013}
1014
1015/**
1016 * igb_clear_interrupt_scheme - reset the device to a state of no interrupts
1017 *
1018 * This function resets the device so that it has 0 rx queues, tx queues, and
1019 * MSI-X interrupts allocated.
1020 */
1021static void igb_clear_interrupt_scheme(struct igb_adapter *adapter)
1022{
1023 igb_free_queues(adapter);
1024 igb_free_q_vectors(adapter);
1025 igb_reset_interrupt_capability(adapter);
1026}
1027
1028/**
1029 * igb_set_interrupt_capability - set MSI or MSI-X if supported
1030 *
1031 * Attempt to configure interrupts using the best available
1032 * capabilities of the hardware and kernel.
1033 **/
1034static int igb_set_interrupt_capability(struct igb_adapter *adapter)
1035{
1036 int err;
1037 int numvecs, i;
1038
1039 /* Number of supported queues. */
1040 adapter->num_rx_queues = adapter->rss_queues;
1041 if (adapter->vfs_allocated_count)
1042 adapter->num_tx_queues = 1;
1043 else
1044 adapter->num_tx_queues = adapter->rss_queues;
1045
1046 /* start with one vector for every rx queue */
1047 numvecs = adapter->num_rx_queues;
1048
1049 /* if tx handler is separate add 1 for every tx queue */
1050 if (!(adapter->flags & IGB_FLAG_QUEUE_PAIRS))
1051 numvecs += adapter->num_tx_queues;
1052
1053 /* store the number of vectors reserved for queues */
1054 adapter->num_q_vectors = numvecs;
1055
1056 /* add 1 vector for link status interrupts */
1057 numvecs++;
1058 adapter->msix_entries = kcalloc(numvecs, sizeof(struct msix_entry),
1059 GFP_KERNEL);
1060 if (!adapter->msix_entries)
1061 goto msi_only;
1062
1063 for (i = 0; i < numvecs; i++)
1064 adapter->msix_entries[i].entry = i;
1065
1066 err = pci_enable_msix(adapter->pdev,
1067 adapter->msix_entries,
1068 numvecs);
1069 if (err == 0)
1070 goto out;
1071
1072 igb_reset_interrupt_capability(adapter);
1073
1074 /* If we can't do MSI-X, try MSI */
1075msi_only:
1076#ifdef CONFIG_PCI_IOV
1077 /* disable SR-IOV for non MSI-X configurations */
1078 if (adapter->vf_data) {
1079 struct e1000_hw *hw = &adapter->hw;
1080 /* disable iov and allow time for transactions to clear */
1081 pci_disable_sriov(adapter->pdev);
1082 msleep(500);
1083
1084 kfree(adapter->vf_data);
1085 adapter->vf_data = NULL;
1086 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
1087 wrfl();
1088 msleep(100);
1089 dev_info(&adapter->pdev->dev, "IOV Disabled\n");
1090 }
1091#endif
1092 adapter->vfs_allocated_count = 0;
1093 adapter->rss_queues = 1;
1094 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
1095 adapter->num_rx_queues = 1;
1096 adapter->num_tx_queues = 1;
1097 adapter->num_q_vectors = 1;
1098 if (!pci_enable_msi(adapter->pdev))
1099 adapter->flags |= IGB_FLAG_HAS_MSI;
1100out:
1101 /* Notify the stack of the (possibly) reduced queue counts. */
1102 netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
1103 return netif_set_real_num_rx_queues(adapter->netdev,
1104 adapter->num_rx_queues);
1105}
1106
1107/**
1108 * igb_alloc_q_vectors - Allocate memory for interrupt vectors
1109 * @adapter: board private structure to initialize
1110 *
1111 * We allocate one q_vector per queue interrupt. If allocation fails we
1112 * return -ENOMEM.
1113 **/
1114static int igb_alloc_q_vectors(struct igb_adapter *adapter)
1115{
1116 struct igb_q_vector *q_vector;
1117 struct e1000_hw *hw = &adapter->hw;
1118 int v_idx;
1119 int orig_node = adapter->node;
1120
1121 for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
1122 if ((adapter->num_q_vectors == (adapter->num_rx_queues +
1123 adapter->num_tx_queues)) &&
1124 (adapter->num_rx_queues == v_idx))
1125 adapter->node = orig_node;
1126 if (orig_node == -1) {
1127 int cur_node = next_online_node(adapter->node);
1128 if (cur_node == MAX_NUMNODES)
1129 cur_node = first_online_node;
1130 adapter->node = cur_node;
1131 }
1132 q_vector = kzalloc_node(sizeof(struct igb_q_vector), GFP_KERNEL,
1133 adapter->node);
1134 if (!q_vector)
1135 q_vector = kzalloc(sizeof(struct igb_q_vector),
1136 GFP_KERNEL);
1137 if (!q_vector)
1138 goto err_out;
1139 q_vector->adapter = adapter;
1140 q_vector->itr_register = hw->hw_addr + E1000_EITR(0);
1141 q_vector->itr_val = IGB_START_ITR;
1142 netif_napi_add(adapter->netdev, &q_vector->napi, igb_poll, 64);
1143 adapter->q_vector[v_idx] = q_vector;
1144 }
1145 /* Restore the adapter's original node */
1146 adapter->node = orig_node;
1147
1148 return 0;
1149
1150err_out:
1151 /* Restore the adapter's original node */
1152 adapter->node = orig_node;
1153 igb_free_q_vectors(adapter);
1154 return -ENOMEM;
1155}
1156
1157static void igb_map_rx_ring_to_vector(struct igb_adapter *adapter,
1158 int ring_idx, int v_idx)
1159{
1160 struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
1161
1162 q_vector->rx.ring = adapter->rx_ring[ring_idx];
1163 q_vector->rx.ring->q_vector = q_vector;
1164 q_vector->rx.count++;
1165 q_vector->itr_val = adapter->rx_itr_setting;
1166 if (q_vector->itr_val && q_vector->itr_val <= 3)
1167 q_vector->itr_val = IGB_START_ITR;
1168}
1169
1170static void igb_map_tx_ring_to_vector(struct igb_adapter *adapter,
1171 int ring_idx, int v_idx)
1172{
1173 struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
1174
1175 q_vector->tx.ring = adapter->tx_ring[ring_idx];
1176 q_vector->tx.ring->q_vector = q_vector;
1177 q_vector->tx.count++;
1178 q_vector->itr_val = adapter->tx_itr_setting;
1179 q_vector->tx.work_limit = adapter->tx_work_limit;
1180 if (q_vector->itr_val && q_vector->itr_val <= 3)
1181 q_vector->itr_val = IGB_START_ITR;
1182}
1183
1184/**
1185 * igb_map_ring_to_vector - maps allocated queues to vectors
1186 *
1187 * This function maps the recently allocated queues to vectors.
1188 **/
1189static int igb_map_ring_to_vector(struct igb_adapter *adapter)
1190{
1191 int i;
1192 int v_idx = 0;
1193
1194 if ((adapter->num_q_vectors < adapter->num_rx_queues) ||
1195 (adapter->num_q_vectors < adapter->num_tx_queues))
1196 return -ENOMEM;
1197
1198 if (adapter->num_q_vectors >=
1199 (adapter->num_rx_queues + adapter->num_tx_queues)) {
1200 for (i = 0; i < adapter->num_rx_queues; i++)
1201 igb_map_rx_ring_to_vector(adapter, i, v_idx++);
1202 for (i = 0; i < adapter->num_tx_queues; i++)
1203 igb_map_tx_ring_to_vector(adapter, i, v_idx++);
1204 } else {
1205 for (i = 0; i < adapter->num_rx_queues; i++) {
1206 if (i < adapter->num_tx_queues)
1207 igb_map_tx_ring_to_vector(adapter, i, v_idx);
1208 igb_map_rx_ring_to_vector(adapter, i, v_idx++);
1209 }
1210 for (; i < adapter->num_tx_queues; i++)
1211 igb_map_tx_ring_to_vector(adapter, i, v_idx++);
1212 }
1213 return 0;
1214}
1215
1216/**
1217 * igb_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
1218 *
1219 * This function initializes the interrupts and allocates all of the queues.
1220 **/
1221static int igb_init_interrupt_scheme(struct igb_adapter *adapter)
1222{
1223 struct pci_dev *pdev = adapter->pdev;
1224 int err;
1225
1226 err = igb_set_interrupt_capability(adapter);
1227 if (err)
1228 return err;
1229
1230 err = igb_alloc_q_vectors(adapter);
1231 if (err) {
1232 dev_err(&pdev->dev, "Unable to allocate memory for vectors\n");
1233 goto err_alloc_q_vectors;
1234 }
1235
1236 err = igb_alloc_queues(adapter);
1237 if (err) {
1238 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
1239 goto err_alloc_queues;
1240 }
1241
1242 err = igb_map_ring_to_vector(adapter);
1243 if (err) {
1244 dev_err(&pdev->dev, "Invalid q_vector to ring mapping\n");
1245 goto err_map_queues;
1246 }
1247
1248
1249 return 0;
1250err_map_queues:
1251 igb_free_queues(adapter);
1252err_alloc_queues:
1253 igb_free_q_vectors(adapter);
1254err_alloc_q_vectors:
1255 igb_reset_interrupt_capability(adapter);
1256 return err;
1257}
1258
1259/**
1260 * igb_request_irq - initialize interrupts
1261 *
1262 * Attempts to configure interrupts using the best available
1263 * capabilities of the hardware and kernel.
1264 **/
1265static int igb_request_irq(struct igb_adapter *adapter)
1266{
1267 struct net_device *netdev = adapter->netdev;
1268 struct pci_dev *pdev = adapter->pdev;
1269 int err = 0;
1270
1271 if (adapter->msix_entries) {
1272 err = igb_request_msix(adapter);
1273 if (!err)
1274 goto request_done;
1275 /* fall back to MSI */
1276 igb_clear_interrupt_scheme(adapter);
1277 if (!pci_enable_msi(pdev))
1278 adapter->flags |= IGB_FLAG_HAS_MSI;
1279 igb_free_all_tx_resources(adapter);
1280 igb_free_all_rx_resources(adapter);
1281 adapter->num_tx_queues = 1;
1282 adapter->num_rx_queues = 1;
1283 adapter->num_q_vectors = 1;
1284 err = igb_alloc_q_vectors(adapter);
1285 if (err) {
1286 dev_err(&pdev->dev,
1287 "Unable to allocate memory for vectors\n");
1288 goto request_done;
1289 }
1290 err = igb_alloc_queues(adapter);
1291 if (err) {
1292 dev_err(&pdev->dev,
1293 "Unable to allocate memory for queues\n");
1294 igb_free_q_vectors(adapter);
1295 goto request_done;
1296 }
1297 igb_setup_all_tx_resources(adapter);
1298 igb_setup_all_rx_resources(adapter);
1299 }
1300
1301 igb_assign_vector(adapter->q_vector[0], 0);
1302
1303 if (adapter->flags & IGB_FLAG_HAS_MSI) {
1304 err = request_irq(pdev->irq, igb_intr_msi, 0,
1305 netdev->name, adapter);
1306 if (!err)
1307 goto request_done;
1308
1309 /* fall back to legacy interrupts */
1310 igb_reset_interrupt_capability(adapter);
1311 adapter->flags &= ~IGB_FLAG_HAS_MSI;
1312 }
1313
1314 err = request_irq(pdev->irq, igb_intr, IRQF_SHARED,
1315 netdev->name, adapter);
1316
1317 if (err)
1318 dev_err(&pdev->dev, "Error %d getting interrupt\n",
1319 err);
1320
1321request_done:
1322 return err;
1323}
1324
1325static void igb_free_irq(struct igb_adapter *adapter)
1326{
1327 if (adapter->msix_entries) {
1328 int vector = 0, i;
1329
1330 free_irq(adapter->msix_entries[vector++].vector, adapter);
1331
1332 for (i = 0; i < adapter->num_q_vectors; i++)
1333 free_irq(adapter->msix_entries[vector++].vector,
1334 adapter->q_vector[i]);
1335 } else {
1336 free_irq(adapter->pdev->irq, adapter);
1337 }
1338}
1339
1340/**
1341 * igb_irq_disable - Mask off interrupt generation on the NIC
1342 * @adapter: board private structure
1343 **/
1344static void igb_irq_disable(struct igb_adapter *adapter)
1345{
1346 struct e1000_hw *hw = &adapter->hw;
1347
1348 /*
1349 * we need to be careful when disabling interrupts. The VFs are also
1350 * mapped into these registers and so clearing the bits can cause
1351 * issues on the VF drivers so we only need to clear what we set
1352 */
1353 if (adapter->msix_entries) {
1354 u32 regval = rd32(E1000_EIAM);
1355 wr32(E1000_EIAM, regval & ~adapter->eims_enable_mask);
1356 wr32(E1000_EIMC, adapter->eims_enable_mask);
1357 regval = rd32(E1000_EIAC);
1358 wr32(E1000_EIAC, regval & ~adapter->eims_enable_mask);
1359 }
1360
1361 wr32(E1000_IAM, 0);
1362 wr32(E1000_IMC, ~0);
1363 wrfl();
1364 if (adapter->msix_entries) {
1365 int i;
1366 for (i = 0; i < adapter->num_q_vectors; i++)
1367 synchronize_irq(adapter->msix_entries[i].vector);
1368 } else {
1369 synchronize_irq(adapter->pdev->irq);
1370 }
1371}
1372
1373/**
1374 * igb_irq_enable - Enable default interrupt generation settings
1375 * @adapter: board private structure
1376 **/
1377static void igb_irq_enable(struct igb_adapter *adapter)
1378{
1379 struct e1000_hw *hw = &adapter->hw;
1380
1381 if (adapter->msix_entries) {
1382 u32 ims = E1000_IMS_LSC | E1000_IMS_DOUTSYNC | E1000_IMS_DRSTA;
1383 u32 regval = rd32(E1000_EIAC);
1384 wr32(E1000_EIAC, regval | adapter->eims_enable_mask);
1385 regval = rd32(E1000_EIAM);
1386 wr32(E1000_EIAM, regval | adapter->eims_enable_mask);
1387 wr32(E1000_EIMS, adapter->eims_enable_mask);
1388 if (adapter->vfs_allocated_count) {
1389 wr32(E1000_MBVFIMR, 0xFF);
1390 ims |= E1000_IMS_VMMB;
1391 }
1392 wr32(E1000_IMS, ims);
1393 } else {
1394 wr32(E1000_IMS, IMS_ENABLE_MASK |
1395 E1000_IMS_DRSTA);
1396 wr32(E1000_IAM, IMS_ENABLE_MASK |
1397 E1000_IMS_DRSTA);
1398 }
1399}
1400
1401static void igb_update_mng_vlan(struct igb_adapter *adapter)
1402{
1403 struct e1000_hw *hw = &adapter->hw;
1404 u16 vid = adapter->hw.mng_cookie.vlan_id;
1405 u16 old_vid = adapter->mng_vlan_id;
1406
1407 if (hw->mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1408 /* add VID to filter table */
1409 igb_vfta_set(hw, vid, true);
1410 adapter->mng_vlan_id = vid;
1411 } else {
1412 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1413 }
1414
1415 if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
1416 (vid != old_vid) &&
1417 !test_bit(old_vid, adapter->active_vlans)) {
1418 /* remove VID from filter table */
1419 igb_vfta_set(hw, old_vid, false);
1420 }
1421}
1422
1423/**
1424 * igb_release_hw_control - release control of the h/w to f/w
1425 * @adapter: address of board private structure
1426 *
1427 * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
1428 * For ASF and Pass Through versions of f/w this means that the
1429 * driver is no longer loaded.
1430 *
1431 **/
1432static void igb_release_hw_control(struct igb_adapter *adapter)
1433{
1434 struct e1000_hw *hw = &adapter->hw;
1435 u32 ctrl_ext;
1436
1437 /* Let firmware take over control of h/w */
1438 ctrl_ext = rd32(E1000_CTRL_EXT);
1439 wr32(E1000_CTRL_EXT,
1440 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1441}
1442
1443/**
1444 * igb_get_hw_control - get control of the h/w from f/w
1445 * @adapter: address of board private structure
1446 *
1447 * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
1448 * For ASF and Pass Through versions of f/w this means that
1449 * the driver is loaded.
1450 *
1451 **/
1452static void igb_get_hw_control(struct igb_adapter *adapter)
1453{
1454 struct e1000_hw *hw = &adapter->hw;
1455 u32 ctrl_ext;
1456
1457 /* Let firmware know the driver has taken over */
1458 ctrl_ext = rd32(E1000_CTRL_EXT);
1459 wr32(E1000_CTRL_EXT,
1460 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1461}
1462
1463/**
1464 * igb_configure - configure the hardware for RX and TX
1465 * @adapter: private board structure
1466 **/
1467static void igb_configure(struct igb_adapter *adapter)
1468{
1469 struct net_device *netdev = adapter->netdev;
1470 int i;
1471
1472 igb_get_hw_control(adapter);
1473 igb_set_rx_mode(netdev);
1474
1475 igb_restore_vlan(adapter);
1476
1477 igb_setup_tctl(adapter);
1478 igb_setup_mrqc(adapter);
1479 igb_setup_rctl(adapter);
1480
1481 igb_configure_tx(adapter);
1482 igb_configure_rx(adapter);
1483
1484 igb_rx_fifo_flush_82575(&adapter->hw);
1485
1486 /* call igb_desc_unused which always leaves
1487 * at least 1 descriptor unused to make sure
1488 * next_to_use != next_to_clean */
1489 for (i = 0; i < adapter->num_rx_queues; i++) {
1490 struct igb_ring *ring = adapter->rx_ring[i];
1491 igb_alloc_rx_buffers(ring, igb_desc_unused(ring));
1492 }
1493}
1494
1495/**
1496 * igb_power_up_link - Power up the phy/serdes link
1497 * @adapter: address of board private structure
1498 **/
1499void igb_power_up_link(struct igb_adapter *adapter)
1500{
1501 if (adapter->hw.phy.media_type == e1000_media_type_copper)
1502 igb_power_up_phy_copper(&adapter->hw);
1503 else
1504 igb_power_up_serdes_link_82575(&adapter->hw);
1505}
1506
1507/**
1508 * igb_power_down_link - Power down the phy/serdes link
1509 * @adapter: address of board private structure
1510 */
1511static void igb_power_down_link(struct igb_adapter *adapter)
1512{
1513 if (adapter->hw.phy.media_type == e1000_media_type_copper)
1514 igb_power_down_phy_copper_82575(&adapter->hw);
1515 else
1516 igb_shutdown_serdes_link_82575(&adapter->hw);
1517}
1518
1519/**
1520 * igb_up - Open the interface and prepare it to handle traffic
1521 * @adapter: board private structure
1522 **/
1523int igb_up(struct igb_adapter *adapter)
1524{
1525 struct e1000_hw *hw = &adapter->hw;
1526 int i;
1527
1528 /* hardware has been reset, we need to reload some things */
1529 igb_configure(adapter);
1530
1531 clear_bit(__IGB_DOWN, &adapter->state);
1532
1533 for (i = 0; i < adapter->num_q_vectors; i++)
1534 napi_enable(&(adapter->q_vector[i]->napi));
1535
1536 if (adapter->msix_entries)
1537 igb_configure_msix(adapter);
1538 else
1539 igb_assign_vector(adapter->q_vector[0], 0);
1540
1541 /* Clear any pending interrupts. */
1542 rd32(E1000_ICR);
1543 igb_irq_enable(adapter);
1544
1545 /* notify VFs that reset has been completed */
1546 if (adapter->vfs_allocated_count) {
1547 u32 reg_data = rd32(E1000_CTRL_EXT);
1548 reg_data |= E1000_CTRL_EXT_PFRSTD;
1549 wr32(E1000_CTRL_EXT, reg_data);
1550 }
1551
1552 netif_tx_start_all_queues(adapter->netdev);
1553
1554 /* start the watchdog. */
1555 hw->mac.get_link_status = 1;
1556 schedule_work(&adapter->watchdog_task);
1557
1558 return 0;
1559}
1560
1561void igb_down(struct igb_adapter *adapter)
1562{
1563 struct net_device *netdev = adapter->netdev;
1564 struct e1000_hw *hw = &adapter->hw;
1565 u32 tctl, rctl;
1566 int i;
1567
1568 /* signal that we're down so the interrupt handler does not
1569 * reschedule our watchdog timer */
1570 set_bit(__IGB_DOWN, &adapter->state);
1571
1572 /* disable receives in the hardware */
1573 rctl = rd32(E1000_RCTL);
1574 wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
1575 /* flush and sleep below */
1576
1577 netif_tx_stop_all_queues(netdev);
1578
1579 /* disable transmits in the hardware */
1580 tctl = rd32(E1000_TCTL);
1581 tctl &= ~E1000_TCTL_EN;
1582 wr32(E1000_TCTL, tctl);
1583 /* flush both disables and wait for them to finish */
1584 wrfl();
1585 msleep(10);
1586
1587 for (i = 0; i < adapter->num_q_vectors; i++)
1588 napi_disable(&(adapter->q_vector[i]->napi));
1589
1590 igb_irq_disable(adapter);
1591
1592 del_timer_sync(&adapter->watchdog_timer);
1593 del_timer_sync(&adapter->phy_info_timer);
1594
1595 netif_carrier_off(netdev);
1596
1597 /* record the stats before reset*/
1598 spin_lock(&adapter->stats64_lock);
1599 igb_update_stats(adapter, &adapter->stats64);
1600 spin_unlock(&adapter->stats64_lock);
1601
1602 adapter->link_speed = 0;
1603 adapter->link_duplex = 0;
1604
1605 if (!pci_channel_offline(adapter->pdev))
1606 igb_reset(adapter);
1607 igb_clean_all_tx_rings(adapter);
1608 igb_clean_all_rx_rings(adapter);
1609#ifdef CONFIG_IGB_DCA
1610
1611 /* since we reset the hardware DCA settings were cleared */
1612 igb_setup_dca(adapter);
1613#endif
1614}
1615
1616void igb_reinit_locked(struct igb_adapter *adapter)
1617{
1618 WARN_ON(in_interrupt());
1619 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
1620 msleep(1);
1621 igb_down(adapter);
1622 igb_up(adapter);
1623 clear_bit(__IGB_RESETTING, &adapter->state);
1624}
1625
1626void igb_reset(struct igb_adapter *adapter)
1627{
1628 struct pci_dev *pdev = adapter->pdev;
1629 struct e1000_hw *hw = &adapter->hw;
1630 struct e1000_mac_info *mac = &hw->mac;
1631 struct e1000_fc_info *fc = &hw->fc;
1632 u32 pba = 0, tx_space, min_tx_space, min_rx_space;
1633 u16 hwm;
1634
1635 /* Repartition Pba for greater than 9k mtu
1636 * To take effect CTRL.RST is required.
1637 */
1638 switch (mac->type) {
1639 case e1000_i350:
1640 case e1000_82580:
1641 pba = rd32(E1000_RXPBS);
1642 pba = igb_rxpbs_adjust_82580(pba);
1643 break;
1644 case e1000_82576:
1645 pba = rd32(E1000_RXPBS);
1646 pba &= E1000_RXPBS_SIZE_MASK_82576;
1647 break;
1648 case e1000_82575:
1649 default:
1650 pba = E1000_PBA_34K;
1651 break;
1652 }
1653
1654 if ((adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) &&
1655 (mac->type < e1000_82576)) {
1656 /* adjust PBA for jumbo frames */
1657 wr32(E1000_PBA, pba);
1658
1659 /* To maintain wire speed transmits, the Tx FIFO should be
1660 * large enough to accommodate two full transmit packets,
1661 * rounded up to the next 1KB and expressed in KB. Likewise,
1662 * the Rx FIFO should be large enough to accommodate at least
1663 * one full receive packet and is similarly rounded up and
1664 * expressed in KB. */
1665 pba = rd32(E1000_PBA);
1666 /* upper 16 bits has Tx packet buffer allocation size in KB */
1667 tx_space = pba >> 16;
1668 /* lower 16 bits has Rx packet buffer allocation size in KB */
1669 pba &= 0xffff;
1670 /* the tx fifo also stores 16 bytes of information about the tx
1671 * but don't include ethernet FCS because hardware appends it */
1672 min_tx_space = (adapter->max_frame_size +
1673 sizeof(union e1000_adv_tx_desc) -
1674 ETH_FCS_LEN) * 2;
1675 min_tx_space = ALIGN(min_tx_space, 1024);
1676 min_tx_space >>= 10;
1677 /* software strips receive CRC, so leave room for it */
1678 min_rx_space = adapter->max_frame_size;
1679 min_rx_space = ALIGN(min_rx_space, 1024);
1680 min_rx_space >>= 10;
1681
1682 /* If current Tx allocation is less than the min Tx FIFO size,
1683 * and the min Tx FIFO size is less than the current Rx FIFO
1684 * allocation, take space away from current Rx allocation */
1685 if (tx_space < min_tx_space &&
1686 ((min_tx_space - tx_space) < pba)) {
1687 pba = pba - (min_tx_space - tx_space);
1688
1689 /* if short on rx space, rx wins and must trump tx
1690 * adjustment */
1691 if (pba < min_rx_space)
1692 pba = min_rx_space;
1693 }
1694 wr32(E1000_PBA, pba);
1695 }
1696
1697 /* flow control settings */
1698 /* The high water mark must be low enough to fit one full frame
1699 * (or the size used for early receive) above it in the Rx FIFO.
1700 * Set it to the lower of:
1701 * - 90% of the Rx FIFO size, or
1702 * - the full Rx FIFO size minus one full frame */
1703 hwm = min(((pba << 10) * 9 / 10),
1704 ((pba << 10) - 2 * adapter->max_frame_size));
1705
1706 fc->high_water = hwm & 0xFFF0; /* 16-byte granularity */
1707 fc->low_water = fc->high_water - 16;
1708 fc->pause_time = 0xFFFF;
1709 fc->send_xon = 1;
1710 fc->current_mode = fc->requested_mode;
1711
1712 /* disable receive for all VFs and wait one second */
1713 if (adapter->vfs_allocated_count) {
1714 int i;
1715 for (i = 0 ; i < adapter->vfs_allocated_count; i++)
1716 adapter->vf_data[i].flags &= IGB_VF_FLAG_PF_SET_MAC;
1717
1718 /* ping all the active vfs to let them know we are going down */
1719 igb_ping_all_vfs(adapter);
1720
1721 /* disable transmits and receives */
1722 wr32(E1000_VFRE, 0);
1723 wr32(E1000_VFTE, 0);
1724 }
1725
1726 /* Allow time for pending master requests to run */
1727 hw->mac.ops.reset_hw(hw);
1728 wr32(E1000_WUC, 0);
1729
1730 if (hw->mac.ops.init_hw(hw))
1731 dev_err(&pdev->dev, "Hardware Error\n");
1732
1733 igb_init_dmac(adapter, pba);
1734 if (!netif_running(adapter->netdev))
1735 igb_power_down_link(adapter);
1736
1737 igb_update_mng_vlan(adapter);
1738
1739 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
1740 wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
1741
1742 igb_get_phy_info(hw);
1743}
1744
1745static u32 igb_fix_features(struct net_device *netdev, u32 features)
1746{
1747 /*
1748 * Since there is no support for separate rx/tx vlan accel
1749 * enable/disable make sure tx flag is always in same state as rx.
1750 */
1751 if (features & NETIF_F_HW_VLAN_RX)
1752 features |= NETIF_F_HW_VLAN_TX;
1753 else
1754 features &= ~NETIF_F_HW_VLAN_TX;
1755
1756 return features;
1757}
1758
1759static int igb_set_features(struct net_device *netdev, u32 features)
1760{
1761 u32 changed = netdev->features ^ features;
1762
1763 if (changed & NETIF_F_HW_VLAN_RX)
1764 igb_vlan_mode(netdev, features);
1765
1766 return 0;
1767}
1768
1769static const struct net_device_ops igb_netdev_ops = {
1770 .ndo_open = igb_open,
1771 .ndo_stop = igb_close,
1772 .ndo_start_xmit = igb_xmit_frame,
1773 .ndo_get_stats64 = igb_get_stats64,
1774 .ndo_set_rx_mode = igb_set_rx_mode,
1775 .ndo_set_mac_address = igb_set_mac,
1776 .ndo_change_mtu = igb_change_mtu,
1777 .ndo_do_ioctl = igb_ioctl,
1778 .ndo_tx_timeout = igb_tx_timeout,
1779 .ndo_validate_addr = eth_validate_addr,
1780 .ndo_vlan_rx_add_vid = igb_vlan_rx_add_vid,
1781 .ndo_vlan_rx_kill_vid = igb_vlan_rx_kill_vid,
1782 .ndo_set_vf_mac = igb_ndo_set_vf_mac,
1783 .ndo_set_vf_vlan = igb_ndo_set_vf_vlan,
1784 .ndo_set_vf_tx_rate = igb_ndo_set_vf_bw,
1785 .ndo_get_vf_config = igb_ndo_get_vf_config,
1786#ifdef CONFIG_NET_POLL_CONTROLLER
1787 .ndo_poll_controller = igb_netpoll,
1788#endif
1789 .ndo_fix_features = igb_fix_features,
1790 .ndo_set_features = igb_set_features,
1791};
1792
1793/**
1794 * igb_probe - Device Initialization Routine
1795 * @pdev: PCI device information struct
1796 * @ent: entry in igb_pci_tbl
1797 *
1798 * Returns 0 on success, negative on failure
1799 *
1800 * igb_probe initializes an adapter identified by a pci_dev structure.
1801 * The OS initialization, configuring of the adapter private structure,
1802 * and a hardware reset occur.
1803 **/
1804static int __devinit igb_probe(struct pci_dev *pdev,
1805 const struct pci_device_id *ent)
1806{
1807 struct net_device *netdev;
1808 struct igb_adapter *adapter;
1809 struct e1000_hw *hw;
1810 u16 eeprom_data = 0;
1811 s32 ret_val;
1812 static int global_quad_port_a; /* global quad port a indication */
1813 const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
1814 unsigned long mmio_start, mmio_len;
1815 int err, pci_using_dac;
1816 u16 eeprom_apme_mask = IGB_EEPROM_APME;
1817 u8 part_str[E1000_PBANUM_LENGTH];
1818
1819 /* Catch broken hardware that put the wrong VF device ID in
1820 * the PCIe SR-IOV capability.
1821 */
1822 if (pdev->is_virtfn) {
1823 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
1824 pci_name(pdev), pdev->vendor, pdev->device);
1825 return -EINVAL;
1826 }
1827
1828 err = pci_enable_device_mem(pdev);
1829 if (err)
1830 return err;
1831
1832 pci_using_dac = 0;
1833 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
1834 if (!err) {
1835 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
1836 if (!err)
1837 pci_using_dac = 1;
1838 } else {
1839 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
1840 if (err) {
1841 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(32));
1842 if (err) {
1843 dev_err(&pdev->dev, "No usable DMA "
1844 "configuration, aborting\n");
1845 goto err_dma;
1846 }
1847 }
1848 }
1849
1850 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
1851 IORESOURCE_MEM),
1852 igb_driver_name);
1853 if (err)
1854 goto err_pci_reg;
1855
1856 pci_enable_pcie_error_reporting(pdev);
1857
1858 pci_set_master(pdev);
1859 pci_save_state(pdev);
1860
1861 err = -ENOMEM;
1862 netdev = alloc_etherdev_mq(sizeof(struct igb_adapter),
1863 IGB_MAX_TX_QUEUES);
1864 if (!netdev)
1865 goto err_alloc_etherdev;
1866
1867 SET_NETDEV_DEV(netdev, &pdev->dev);
1868
1869 pci_set_drvdata(pdev, netdev);
1870 adapter = netdev_priv(netdev);
1871 adapter->netdev = netdev;
1872 adapter->pdev = pdev;
1873 hw = &adapter->hw;
1874 hw->back = adapter;
1875 adapter->msg_enable = NETIF_MSG_DRV | NETIF_MSG_PROBE;
1876
1877 mmio_start = pci_resource_start(pdev, 0);
1878 mmio_len = pci_resource_len(pdev, 0);
1879
1880 err = -EIO;
1881 hw->hw_addr = ioremap(mmio_start, mmio_len);
1882 if (!hw->hw_addr)
1883 goto err_ioremap;
1884
1885 netdev->netdev_ops = &igb_netdev_ops;
1886 igb_set_ethtool_ops(netdev);
1887 netdev->watchdog_timeo = 5 * HZ;
1888
1889 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
1890
1891 netdev->mem_start = mmio_start;
1892 netdev->mem_end = mmio_start + mmio_len;
1893
1894 /* PCI config space info */
1895 hw->vendor_id = pdev->vendor;
1896 hw->device_id = pdev->device;
1897 hw->revision_id = pdev->revision;
1898 hw->subsystem_vendor_id = pdev->subsystem_vendor;
1899 hw->subsystem_device_id = pdev->subsystem_device;
1900
1901 /* Copy the default MAC, PHY and NVM function pointers */
1902 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
1903 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
1904 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
1905 /* Initialize skew-specific constants */
1906 err = ei->get_invariants(hw);
1907 if (err)
1908 goto err_sw_init;
1909
1910 /* setup the private structure */
1911 err = igb_sw_init(adapter);
1912 if (err)
1913 goto err_sw_init;
1914
1915 igb_get_bus_info_pcie(hw);
1916
1917 hw->phy.autoneg_wait_to_complete = false;
1918
1919 /* Copper options */
1920 if (hw->phy.media_type == e1000_media_type_copper) {
1921 hw->phy.mdix = AUTO_ALL_MODES;
1922 hw->phy.disable_polarity_correction = false;
1923 hw->phy.ms_type = e1000_ms_hw_default;
1924 }
1925
1926 if (igb_check_reset_block(hw))
1927 dev_info(&pdev->dev,
1928 "PHY reset is blocked due to SOL/IDER session.\n");
1929
1930 /*
1931 * features is initialized to 0 in allocation, it might have bits
1932 * set by igb_sw_init so we should use an or instead of an
1933 * assignment.
1934 */
1935 netdev->features |= NETIF_F_SG |
1936 NETIF_F_IP_CSUM |
1937 NETIF_F_IPV6_CSUM |
1938 NETIF_F_TSO |
1939 NETIF_F_TSO6 |
1940 NETIF_F_RXHASH |
1941 NETIF_F_RXCSUM |
1942 NETIF_F_HW_VLAN_RX |
1943 NETIF_F_HW_VLAN_TX;
1944
1945 /* copy netdev features into list of user selectable features */
1946 netdev->hw_features |= netdev->features;
1947
1948 /* set this bit last since it cannot be part of hw_features */
1949 netdev->features |= NETIF_F_HW_VLAN_FILTER;
1950
1951 netdev->vlan_features |= NETIF_F_TSO |
1952 NETIF_F_TSO6 |
1953 NETIF_F_IP_CSUM |
1954 NETIF_F_IPV6_CSUM |
1955 NETIF_F_SG;
1956
1957 if (pci_using_dac) {
1958 netdev->features |= NETIF_F_HIGHDMA;
1959 netdev->vlan_features |= NETIF_F_HIGHDMA;
1960 }
1961
1962 if (hw->mac.type >= e1000_82576) {
1963 netdev->hw_features |= NETIF_F_SCTP_CSUM;
1964 netdev->features |= NETIF_F_SCTP_CSUM;
1965 }
1966
1967 netdev->priv_flags |= IFF_UNICAST_FLT;
1968
1969 adapter->en_mng_pt = igb_enable_mng_pass_thru(hw);
1970
1971 /* before reading the NVM, reset the controller to put the device in a
1972 * known good starting state */
1973 hw->mac.ops.reset_hw(hw);
1974
1975 /* make sure the NVM is good */
1976 if (hw->nvm.ops.validate(hw) < 0) {
1977 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
1978 err = -EIO;
1979 goto err_eeprom;
1980 }
1981
1982 /* copy the MAC address out of the NVM */
1983 if (hw->mac.ops.read_mac_addr(hw))
1984 dev_err(&pdev->dev, "NVM Read Error\n");
1985
1986 memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
1987 memcpy(netdev->perm_addr, hw->mac.addr, netdev->addr_len);
1988
1989 if (!is_valid_ether_addr(netdev->perm_addr)) {
1990 dev_err(&pdev->dev, "Invalid MAC Address\n");
1991 err = -EIO;
1992 goto err_eeprom;
1993 }
1994
1995 setup_timer(&adapter->watchdog_timer, igb_watchdog,
1996 (unsigned long) adapter);
1997 setup_timer(&adapter->phy_info_timer, igb_update_phy_info,
1998 (unsigned long) adapter);
1999
2000 INIT_WORK(&adapter->reset_task, igb_reset_task);
2001 INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
2002
2003 /* Initialize link properties that are user-changeable */
2004 adapter->fc_autoneg = true;
2005 hw->mac.autoneg = true;
2006 hw->phy.autoneg_advertised = 0x2f;
2007
2008 hw->fc.requested_mode = e1000_fc_default;
2009 hw->fc.current_mode = e1000_fc_default;
2010
2011 igb_validate_mdi_setting(hw);
2012
2013 /* Initial Wake on LAN setting If APM wake is enabled in the EEPROM,
2014 * enable the ACPI Magic Packet filter
2015 */
2016
2017 if (hw->bus.func == 0)
2018 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
2019 else if (hw->mac.type >= e1000_82580)
2020 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A +
2021 NVM_82580_LAN_FUNC_OFFSET(hw->bus.func), 1,
2022 &eeprom_data);
2023 else if (hw->bus.func == 1)
2024 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
2025
2026 if (eeprom_data & eeprom_apme_mask)
2027 adapter->eeprom_wol |= E1000_WUFC_MAG;
2028
2029 /* now that we have the eeprom settings, apply the special cases where
2030 * the eeprom may be wrong or the board simply won't support wake on
2031 * lan on a particular port */
2032 switch (pdev->device) {
2033 case E1000_DEV_ID_82575GB_QUAD_COPPER:
2034 adapter->eeprom_wol = 0;
2035 break;
2036 case E1000_DEV_ID_82575EB_FIBER_SERDES:
2037 case E1000_DEV_ID_82576_FIBER:
2038 case E1000_DEV_ID_82576_SERDES:
2039 /* Wake events only supported on port A for dual fiber
2040 * regardless of eeprom setting */
2041 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
2042 adapter->eeprom_wol = 0;
2043 break;
2044 case E1000_DEV_ID_82576_QUAD_COPPER:
2045 case E1000_DEV_ID_82576_QUAD_COPPER_ET2:
2046 /* if quad port adapter, disable WoL on all but port A */
2047 if (global_quad_port_a != 0)
2048 adapter->eeprom_wol = 0;
2049 else
2050 adapter->flags |= IGB_FLAG_QUAD_PORT_A;
2051 /* Reset for multiple quad port adapters */
2052 if (++global_quad_port_a == 4)
2053 global_quad_port_a = 0;
2054 break;
2055 }
2056
2057 /* initialize the wol settings based on the eeprom settings */
2058 adapter->wol = adapter->eeprom_wol;
2059 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
2060
2061 /* reset the hardware with the new settings */
2062 igb_reset(adapter);
2063
2064 /* let the f/w know that the h/w is now under the control of the
2065 * driver. */
2066 igb_get_hw_control(adapter);
2067
2068 strcpy(netdev->name, "eth%d");
2069 err = register_netdev(netdev);
2070 if (err)
2071 goto err_register;
2072
2073 /* carrier off reporting is important to ethtool even BEFORE open */
2074 netif_carrier_off(netdev);
2075
2076#ifdef CONFIG_IGB_DCA
2077 if (dca_add_requester(&pdev->dev) == 0) {
2078 adapter->flags |= IGB_FLAG_DCA_ENABLED;
2079 dev_info(&pdev->dev, "DCA enabled\n");
2080 igb_setup_dca(adapter);
2081 }
2082
2083#endif
2084 /* do hw tstamp init after resetting */
2085 igb_init_hw_timer(adapter);
2086
2087 dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
2088 /* print bus type/speed/width info */
2089 dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
2090 netdev->name,
2091 ((hw->bus.speed == e1000_bus_speed_2500) ? "2.5Gb/s" :
2092 (hw->bus.speed == e1000_bus_speed_5000) ? "5.0Gb/s" :
2093 "unknown"),
2094 ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
2095 (hw->bus.width == e1000_bus_width_pcie_x2) ? "Width x2" :
2096 (hw->bus.width == e1000_bus_width_pcie_x1) ? "Width x1" :
2097 "unknown"),
2098 netdev->dev_addr);
2099
2100 ret_val = igb_read_part_string(hw, part_str, E1000_PBANUM_LENGTH);
2101 if (ret_val)
2102 strcpy(part_str, "Unknown");
2103 dev_info(&pdev->dev, "%s: PBA No: %s\n", netdev->name, part_str);
2104 dev_info(&pdev->dev,
2105 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
2106 adapter->msix_entries ? "MSI-X" :
2107 (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
2108 adapter->num_rx_queues, adapter->num_tx_queues);
2109 switch (hw->mac.type) {
2110 case e1000_i350:
2111 igb_set_eee_i350(hw);
2112 break;
2113 default:
2114 break;
2115 }
2116 return 0;
2117
2118err_register:
2119 igb_release_hw_control(adapter);
2120err_eeprom:
2121 if (!igb_check_reset_block(hw))
2122 igb_reset_phy(hw);
2123
2124 if (hw->flash_address)
2125 iounmap(hw->flash_address);
2126err_sw_init:
2127 igb_clear_interrupt_scheme(adapter);
2128 iounmap(hw->hw_addr);
2129err_ioremap:
2130 free_netdev(netdev);
2131err_alloc_etherdev:
2132 pci_release_selected_regions(pdev,
2133 pci_select_bars(pdev, IORESOURCE_MEM));
2134err_pci_reg:
2135err_dma:
2136 pci_disable_device(pdev);
2137 return err;
2138}
2139
2140/**
2141 * igb_remove - Device Removal Routine
2142 * @pdev: PCI device information struct
2143 *
2144 * igb_remove is called by the PCI subsystem to alert the driver
2145 * that it should release a PCI device. The could be caused by a
2146 * Hot-Plug event, or because the driver is going to be removed from
2147 * memory.
2148 **/
2149static void __devexit igb_remove(struct pci_dev *pdev)
2150{
2151 struct net_device *netdev = pci_get_drvdata(pdev);
2152 struct igb_adapter *adapter = netdev_priv(netdev);
2153 struct e1000_hw *hw = &adapter->hw;
2154
2155 /*
2156 * The watchdog timer may be rescheduled, so explicitly
2157 * disable watchdog from being rescheduled.
2158 */
2159 set_bit(__IGB_DOWN, &adapter->state);
2160 del_timer_sync(&adapter->watchdog_timer);
2161 del_timer_sync(&adapter->phy_info_timer);
2162
2163 cancel_work_sync(&adapter->reset_task);
2164 cancel_work_sync(&adapter->watchdog_task);
2165
2166#ifdef CONFIG_IGB_DCA
2167 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
2168 dev_info(&pdev->dev, "DCA disabled\n");
2169 dca_remove_requester(&pdev->dev);
2170 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
2171 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
2172 }
2173#endif
2174
2175 /* Release control of h/w to f/w. If f/w is AMT enabled, this
2176 * would have already happened in close and is redundant. */
2177 igb_release_hw_control(adapter);
2178
2179 unregister_netdev(netdev);
2180
2181 igb_clear_interrupt_scheme(adapter);
2182
2183#ifdef CONFIG_PCI_IOV
2184 /* reclaim resources allocated to VFs */
2185 if (adapter->vf_data) {
2186 /* disable iov and allow time for transactions to clear */
2187 if (!igb_check_vf_assignment(adapter)) {
2188 pci_disable_sriov(pdev);
2189 msleep(500);
2190 } else {
2191 dev_info(&pdev->dev, "VF(s) assigned to guests!\n");
2192 }
2193
2194 kfree(adapter->vf_data);
2195 adapter->vf_data = NULL;
2196 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
2197 wrfl();
2198 msleep(100);
2199 dev_info(&pdev->dev, "IOV Disabled\n");
2200 }
2201#endif
2202
2203 iounmap(hw->hw_addr);
2204 if (hw->flash_address)
2205 iounmap(hw->flash_address);
2206 pci_release_selected_regions(pdev,
2207 pci_select_bars(pdev, IORESOURCE_MEM));
2208
2209 kfree(adapter->shadow_vfta);
2210 free_netdev(netdev);
2211
2212 pci_disable_pcie_error_reporting(pdev);
2213
2214 pci_disable_device(pdev);
2215}
2216
2217/**
2218 * igb_probe_vfs - Initialize vf data storage and add VFs to pci config space
2219 * @adapter: board private structure to initialize
2220 *
2221 * This function initializes the vf specific data storage and then attempts to
2222 * allocate the VFs. The reason for ordering it this way is because it is much
2223 * mor expensive time wise to disable SR-IOV than it is to allocate and free
2224 * the memory for the VFs.
2225 **/
2226static void __devinit igb_probe_vfs(struct igb_adapter * adapter)
2227{
2228#ifdef CONFIG_PCI_IOV
2229 struct pci_dev *pdev = adapter->pdev;
2230 int old_vfs = igb_find_enabled_vfs(adapter);
2231 int i;
2232
2233 if (old_vfs) {
2234 dev_info(&pdev->dev, "%d pre-allocated VFs found - override "
2235 "max_vfs setting of %d\n", old_vfs, max_vfs);
2236 adapter->vfs_allocated_count = old_vfs;
2237 }
2238
2239 if (!adapter->vfs_allocated_count)
2240 return;
2241
2242 adapter->vf_data = kcalloc(adapter->vfs_allocated_count,
2243 sizeof(struct vf_data_storage), GFP_KERNEL);
2244 /* if allocation failed then we do not support SR-IOV */
2245 if (!adapter->vf_data) {
2246 adapter->vfs_allocated_count = 0;
2247 dev_err(&pdev->dev, "Unable to allocate memory for VF "
2248 "Data Storage\n");
2249 goto out;
2250 }
2251
2252 if (!old_vfs) {
2253 if (pci_enable_sriov(pdev, adapter->vfs_allocated_count))
2254 goto err_out;
2255 }
2256 dev_info(&pdev->dev, "%d VFs allocated\n",
2257 adapter->vfs_allocated_count);
2258 for (i = 0; i < adapter->vfs_allocated_count; i++)
2259 igb_vf_configure(adapter, i);
2260
2261 /* DMA Coalescing is not supported in IOV mode. */
2262 adapter->flags &= ~IGB_FLAG_DMAC;
2263 goto out;
2264err_out:
2265 kfree(adapter->vf_data);
2266 adapter->vf_data = NULL;
2267 adapter->vfs_allocated_count = 0;
2268out:
2269 return;
2270#endif /* CONFIG_PCI_IOV */
2271}
2272
2273/**
2274 * igb_init_hw_timer - Initialize hardware timer used with IEEE 1588 timestamp
2275 * @adapter: board private structure to initialize
2276 *
2277 * igb_init_hw_timer initializes the function pointer and values for the hw
2278 * timer found in hardware.
2279 **/
2280static void igb_init_hw_timer(struct igb_adapter *adapter)
2281{
2282 struct e1000_hw *hw = &adapter->hw;
2283
2284 switch (hw->mac.type) {
2285 case e1000_i350:
2286 case e1000_82580:
2287 memset(&adapter->cycles, 0, sizeof(adapter->cycles));
2288 adapter->cycles.read = igb_read_clock;
2289 adapter->cycles.mask = CLOCKSOURCE_MASK(64);
2290 adapter->cycles.mult = 1;
2291 /*
2292 * The 82580 timesync updates the system timer every 8ns by 8ns
2293 * and the value cannot be shifted. Instead we need to shift
2294 * the registers to generate a 64bit timer value. As a result
2295 * SYSTIMR/L/H, TXSTMPL/H, RXSTMPL/H all have to be shifted by
2296 * 24 in order to generate a larger value for synchronization.
2297 */
2298 adapter->cycles.shift = IGB_82580_TSYNC_SHIFT;
2299 /* disable system timer temporarily by setting bit 31 */
2300 wr32(E1000_TSAUXC, 0x80000000);
2301 wrfl();
2302
2303 /* Set registers so that rollover occurs soon to test this. */
2304 wr32(E1000_SYSTIMR, 0x00000000);
2305 wr32(E1000_SYSTIML, 0x80000000);
2306 wr32(E1000_SYSTIMH, 0x000000FF);
2307 wrfl();
2308
2309 /* enable system timer by clearing bit 31 */
2310 wr32(E1000_TSAUXC, 0x0);
2311 wrfl();
2312
2313 timecounter_init(&adapter->clock,
2314 &adapter->cycles,
2315 ktime_to_ns(ktime_get_real()));
2316 /*
2317 * Synchronize our NIC clock against system wall clock. NIC
2318 * time stamp reading requires ~3us per sample, each sample
2319 * was pretty stable even under load => only require 10
2320 * samples for each offset comparison.
2321 */
2322 memset(&adapter->compare, 0, sizeof(adapter->compare));
2323 adapter->compare.source = &adapter->clock;
2324 adapter->compare.target = ktime_get_real;
2325 adapter->compare.num_samples = 10;
2326 timecompare_update(&adapter->compare, 0);
2327 break;
2328 case e1000_82576:
2329 /*
2330 * Initialize hardware timer: we keep it running just in case
2331 * that some program needs it later on.
2332 */
2333 memset(&adapter->cycles, 0, sizeof(adapter->cycles));
2334 adapter->cycles.read = igb_read_clock;
2335 adapter->cycles.mask = CLOCKSOURCE_MASK(64);
2336 adapter->cycles.mult = 1;
2337 /**
2338 * Scale the NIC clock cycle by a large factor so that
2339 * relatively small clock corrections can be added or
2340 * subtracted at each clock tick. The drawbacks of a large
2341 * factor are a) that the clock register overflows more quickly
2342 * (not such a big deal) and b) that the increment per tick has
2343 * to fit into 24 bits. As a result we need to use a shift of
2344 * 19 so we can fit a value of 16 into the TIMINCA register.
2345 */
2346 adapter->cycles.shift = IGB_82576_TSYNC_SHIFT;
2347 wr32(E1000_TIMINCA,
2348 (1 << E1000_TIMINCA_16NS_SHIFT) |
2349 (16 << IGB_82576_TSYNC_SHIFT));
2350
2351 /* Set registers so that rollover occurs soon to test this. */
2352 wr32(E1000_SYSTIML, 0x00000000);
2353 wr32(E1000_SYSTIMH, 0xFF800000);
2354 wrfl();
2355
2356 timecounter_init(&adapter->clock,
2357 &adapter->cycles,
2358 ktime_to_ns(ktime_get_real()));
2359 /*
2360 * Synchronize our NIC clock against system wall clock. NIC
2361 * time stamp reading requires ~3us per sample, each sample
2362 * was pretty stable even under load => only require 10
2363 * samples for each offset comparison.
2364 */
2365 memset(&adapter->compare, 0, sizeof(adapter->compare));
2366 adapter->compare.source = &adapter->clock;
2367 adapter->compare.target = ktime_get_real;
2368 adapter->compare.num_samples = 10;
2369 timecompare_update(&adapter->compare, 0);
2370 break;
2371 case e1000_82575:
2372 /* 82575 does not support timesync */
2373 default:
2374 break;
2375 }
2376
2377}
2378
2379/**
2380 * igb_sw_init - Initialize general software structures (struct igb_adapter)
2381 * @adapter: board private structure to initialize
2382 *
2383 * igb_sw_init initializes the Adapter private data structure.
2384 * Fields are initialized based on PCI device information and
2385 * OS network device settings (MTU size).
2386 **/
2387static int __devinit igb_sw_init(struct igb_adapter *adapter)
2388{
2389 struct e1000_hw *hw = &adapter->hw;
2390 struct net_device *netdev = adapter->netdev;
2391 struct pci_dev *pdev = adapter->pdev;
2392
2393 pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
2394
2395 /* set default ring sizes */
2396 adapter->tx_ring_count = IGB_DEFAULT_TXD;
2397 adapter->rx_ring_count = IGB_DEFAULT_RXD;
2398
2399 /* set default ITR values */
2400 adapter->rx_itr_setting = IGB_DEFAULT_ITR;
2401 adapter->tx_itr_setting = IGB_DEFAULT_ITR;
2402
2403 /* set default work limits */
2404 adapter->tx_work_limit = IGB_DEFAULT_TX_WORK;
2405
2406 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN +
2407 VLAN_HLEN;
2408 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2409
2410 adapter->node = -1;
2411
2412 spin_lock_init(&adapter->stats64_lock);
2413#ifdef CONFIG_PCI_IOV
2414 switch (hw->mac.type) {
2415 case e1000_82576:
2416 case e1000_i350:
2417 if (max_vfs > 7) {
2418 dev_warn(&pdev->dev,
2419 "Maximum of 7 VFs per PF, using max\n");
2420 adapter->vfs_allocated_count = 7;
2421 } else
2422 adapter->vfs_allocated_count = max_vfs;
2423 break;
2424 default:
2425 break;
2426 }
2427#endif /* CONFIG_PCI_IOV */
2428 adapter->rss_queues = min_t(u32, IGB_MAX_RX_QUEUES, num_online_cpus());
2429 /* i350 cannot do RSS and SR-IOV at the same time */
2430 if (hw->mac.type == e1000_i350 && adapter->vfs_allocated_count)
2431 adapter->rss_queues = 1;
2432
2433 /*
2434 * if rss_queues > 4 or vfs are going to be allocated with rss_queues
2435 * then we should combine the queues into a queue pair in order to
2436 * conserve interrupts due to limited supply
2437 */
2438 if ((adapter->rss_queues > 4) ||
2439 ((adapter->rss_queues > 1) && (adapter->vfs_allocated_count > 6)))
2440 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
2441
2442 /* Setup and initialize a copy of the hw vlan table array */
2443 adapter->shadow_vfta = kzalloc(sizeof(u32) *
2444 E1000_VLAN_FILTER_TBL_SIZE,
2445 GFP_ATOMIC);
2446
2447 /* This call may decrease the number of queues */
2448 if (igb_init_interrupt_scheme(adapter)) {
2449 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
2450 return -ENOMEM;
2451 }
2452
2453 igb_probe_vfs(adapter);
2454
2455 /* Explicitly disable IRQ since the NIC can be in any state. */
2456 igb_irq_disable(adapter);
2457
2458 if (hw->mac.type == e1000_i350)
2459 adapter->flags &= ~IGB_FLAG_DMAC;
2460
2461 set_bit(__IGB_DOWN, &adapter->state);
2462 return 0;
2463}
2464
2465/**
2466 * igb_open - Called when a network interface is made active
2467 * @netdev: network interface device structure
2468 *
2469 * Returns 0 on success, negative value on failure
2470 *
2471 * The open entry point is called when a network interface is made
2472 * active by the system (IFF_UP). At this point all resources needed
2473 * for transmit and receive operations are allocated, the interrupt
2474 * handler is registered with the OS, the watchdog timer is started,
2475 * and the stack is notified that the interface is ready.
2476 **/
2477static int igb_open(struct net_device *netdev)
2478{
2479 struct igb_adapter *adapter = netdev_priv(netdev);
2480 struct e1000_hw *hw = &adapter->hw;
2481 int err;
2482 int i;
2483
2484 /* disallow open during test */
2485 if (test_bit(__IGB_TESTING, &adapter->state))
2486 return -EBUSY;
2487
2488 netif_carrier_off(netdev);
2489
2490 /* allocate transmit descriptors */
2491 err = igb_setup_all_tx_resources(adapter);
2492 if (err)
2493 goto err_setup_tx;
2494
2495 /* allocate receive descriptors */
2496 err = igb_setup_all_rx_resources(adapter);
2497 if (err)
2498 goto err_setup_rx;
2499
2500 igb_power_up_link(adapter);
2501
2502 /* before we allocate an interrupt, we must be ready to handle it.
2503 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
2504 * as soon as we call pci_request_irq, so we have to setup our
2505 * clean_rx handler before we do so. */
2506 igb_configure(adapter);
2507
2508 err = igb_request_irq(adapter);
2509 if (err)
2510 goto err_req_irq;
2511
2512 /* From here on the code is the same as igb_up() */
2513 clear_bit(__IGB_DOWN, &adapter->state);
2514
2515 for (i = 0; i < adapter->num_q_vectors; i++)
2516 napi_enable(&(adapter->q_vector[i]->napi));
2517
2518 /* Clear any pending interrupts. */
2519 rd32(E1000_ICR);
2520
2521 igb_irq_enable(adapter);
2522
2523 /* notify VFs that reset has been completed */
2524 if (adapter->vfs_allocated_count) {
2525 u32 reg_data = rd32(E1000_CTRL_EXT);
2526 reg_data |= E1000_CTRL_EXT_PFRSTD;
2527 wr32(E1000_CTRL_EXT, reg_data);
2528 }
2529
2530 netif_tx_start_all_queues(netdev);
2531
2532 /* start the watchdog. */
2533 hw->mac.get_link_status = 1;
2534 schedule_work(&adapter->watchdog_task);
2535
2536 return 0;
2537
2538err_req_irq:
2539 igb_release_hw_control(adapter);
2540 igb_power_down_link(adapter);
2541 igb_free_all_rx_resources(adapter);
2542err_setup_rx:
2543 igb_free_all_tx_resources(adapter);
2544err_setup_tx:
2545 igb_reset(adapter);
2546
2547 return err;
2548}
2549
2550/**
2551 * igb_close - Disables a network interface
2552 * @netdev: network interface device structure
2553 *
2554 * Returns 0, this is not allowed to fail
2555 *
2556 * The close entry point is called when an interface is de-activated
2557 * by the OS. The hardware is still under the driver's control, but
2558 * needs to be disabled. A global MAC reset is issued to stop the
2559 * hardware, and all transmit and receive resources are freed.
2560 **/
2561static int igb_close(struct net_device *netdev)
2562{
2563 struct igb_adapter *adapter = netdev_priv(netdev);
2564
2565 WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
2566 igb_down(adapter);
2567
2568 igb_free_irq(adapter);
2569
2570 igb_free_all_tx_resources(adapter);
2571 igb_free_all_rx_resources(adapter);
2572
2573 return 0;
2574}
2575
2576/**
2577 * igb_setup_tx_resources - allocate Tx resources (Descriptors)
2578 * @tx_ring: tx descriptor ring (for a specific queue) to setup
2579 *
2580 * Return 0 on success, negative on failure
2581 **/
2582int igb_setup_tx_resources(struct igb_ring *tx_ring)
2583{
2584 struct device *dev = tx_ring->dev;
2585 int orig_node = dev_to_node(dev);
2586 int size;
2587
2588 size = sizeof(struct igb_tx_buffer) * tx_ring->count;
2589 tx_ring->tx_buffer_info = vzalloc_node(size, tx_ring->numa_node);
2590 if (!tx_ring->tx_buffer_info)
2591 tx_ring->tx_buffer_info = vzalloc(size);
2592 if (!tx_ring->tx_buffer_info)
2593 goto err;
2594
2595 /* round up to nearest 4K */
2596 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
2597 tx_ring->size = ALIGN(tx_ring->size, 4096);
2598
2599 set_dev_node(dev, tx_ring->numa_node);
2600 tx_ring->desc = dma_alloc_coherent(dev,
2601 tx_ring->size,
2602 &tx_ring->dma,
2603 GFP_KERNEL);
2604 set_dev_node(dev, orig_node);
2605 if (!tx_ring->desc)
2606 tx_ring->desc = dma_alloc_coherent(dev,
2607 tx_ring->size,
2608 &tx_ring->dma,
2609 GFP_KERNEL);
2610
2611 if (!tx_ring->desc)
2612 goto err;
2613
2614 tx_ring->next_to_use = 0;
2615 tx_ring->next_to_clean = 0;
2616
2617 return 0;
2618
2619err:
2620 vfree(tx_ring->tx_buffer_info);
2621 dev_err(dev,
2622 "Unable to allocate memory for the transmit descriptor ring\n");
2623 return -ENOMEM;
2624}
2625
2626/**
2627 * igb_setup_all_tx_resources - wrapper to allocate Tx resources
2628 * (Descriptors) for all queues
2629 * @adapter: board private structure
2630 *
2631 * Return 0 on success, negative on failure
2632 **/
2633static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
2634{
2635 struct pci_dev *pdev = adapter->pdev;
2636 int i, err = 0;
2637
2638 for (i = 0; i < adapter->num_tx_queues; i++) {
2639 err = igb_setup_tx_resources(adapter->tx_ring[i]);
2640 if (err) {
2641 dev_err(&pdev->dev,
2642 "Allocation for Tx Queue %u failed\n", i);
2643 for (i--; i >= 0; i--)
2644 igb_free_tx_resources(adapter->tx_ring[i]);
2645 break;
2646 }
2647 }
2648
2649 return err;
2650}
2651
2652/**
2653 * igb_setup_tctl - configure the transmit control registers
2654 * @adapter: Board private structure
2655 **/
2656void igb_setup_tctl(struct igb_adapter *adapter)
2657{
2658 struct e1000_hw *hw = &adapter->hw;
2659 u32 tctl;
2660
2661 /* disable queue 0 which is enabled by default on 82575 and 82576 */
2662 wr32(E1000_TXDCTL(0), 0);
2663
2664 /* Program the Transmit Control Register */
2665 tctl = rd32(E1000_TCTL);
2666 tctl &= ~E1000_TCTL_CT;
2667 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2668 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2669
2670 igb_config_collision_dist(hw);
2671
2672 /* Enable transmits */
2673 tctl |= E1000_TCTL_EN;
2674
2675 wr32(E1000_TCTL, tctl);
2676}
2677
2678/**
2679 * igb_configure_tx_ring - Configure transmit ring after Reset
2680 * @adapter: board private structure
2681 * @ring: tx ring to configure
2682 *
2683 * Configure a transmit ring after a reset.
2684 **/
2685void igb_configure_tx_ring(struct igb_adapter *adapter,
2686 struct igb_ring *ring)
2687{
2688 struct e1000_hw *hw = &adapter->hw;
2689 u32 txdctl = 0;
2690 u64 tdba = ring->dma;
2691 int reg_idx = ring->reg_idx;
2692
2693 /* disable the queue */
2694 wr32(E1000_TXDCTL(reg_idx), 0);
2695 wrfl();
2696 mdelay(10);
2697
2698 wr32(E1000_TDLEN(reg_idx),
2699 ring->count * sizeof(union e1000_adv_tx_desc));
2700 wr32(E1000_TDBAL(reg_idx),
2701 tdba & 0x00000000ffffffffULL);
2702 wr32(E1000_TDBAH(reg_idx), tdba >> 32);
2703
2704 ring->tail = hw->hw_addr + E1000_TDT(reg_idx);
2705 wr32(E1000_TDH(reg_idx), 0);
2706 writel(0, ring->tail);
2707
2708 txdctl |= IGB_TX_PTHRESH;
2709 txdctl |= IGB_TX_HTHRESH << 8;
2710 txdctl |= IGB_TX_WTHRESH << 16;
2711
2712 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
2713 wr32(E1000_TXDCTL(reg_idx), txdctl);
2714}
2715
2716/**
2717 * igb_configure_tx - Configure transmit Unit after Reset
2718 * @adapter: board private structure
2719 *
2720 * Configure the Tx unit of the MAC after a reset.
2721 **/
2722static void igb_configure_tx(struct igb_adapter *adapter)
2723{
2724 int i;
2725
2726 for (i = 0; i < adapter->num_tx_queues; i++)
2727 igb_configure_tx_ring(adapter, adapter->tx_ring[i]);
2728}
2729
2730/**
2731 * igb_setup_rx_resources - allocate Rx resources (Descriptors)
2732 * @rx_ring: rx descriptor ring (for a specific queue) to setup
2733 *
2734 * Returns 0 on success, negative on failure
2735 **/
2736int igb_setup_rx_resources(struct igb_ring *rx_ring)
2737{
2738 struct device *dev = rx_ring->dev;
2739 int orig_node = dev_to_node(dev);
2740 int size, desc_len;
2741
2742 size = sizeof(struct igb_rx_buffer) * rx_ring->count;
2743 rx_ring->rx_buffer_info = vzalloc_node(size, rx_ring->numa_node);
2744 if (!rx_ring->rx_buffer_info)
2745 rx_ring->rx_buffer_info = vzalloc(size);
2746 if (!rx_ring->rx_buffer_info)
2747 goto err;
2748
2749 desc_len = sizeof(union e1000_adv_rx_desc);
2750
2751 /* Round up to nearest 4K */
2752 rx_ring->size = rx_ring->count * desc_len;
2753 rx_ring->size = ALIGN(rx_ring->size, 4096);
2754
2755 set_dev_node(dev, rx_ring->numa_node);
2756 rx_ring->desc = dma_alloc_coherent(dev,
2757 rx_ring->size,
2758 &rx_ring->dma,
2759 GFP_KERNEL);
2760 set_dev_node(dev, orig_node);
2761 if (!rx_ring->desc)
2762 rx_ring->desc = dma_alloc_coherent(dev,
2763 rx_ring->size,
2764 &rx_ring->dma,
2765 GFP_KERNEL);
2766
2767 if (!rx_ring->desc)
2768 goto err;
2769
2770 rx_ring->next_to_clean = 0;
2771 rx_ring->next_to_use = 0;
2772
2773 return 0;
2774
2775err:
2776 vfree(rx_ring->rx_buffer_info);
2777 rx_ring->rx_buffer_info = NULL;
2778 dev_err(dev, "Unable to allocate memory for the receive descriptor"
2779 " ring\n");
2780 return -ENOMEM;
2781}
2782
2783/**
2784 * igb_setup_all_rx_resources - wrapper to allocate Rx resources
2785 * (Descriptors) for all queues
2786 * @adapter: board private structure
2787 *
2788 * Return 0 on success, negative on failure
2789 **/
2790static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
2791{
2792 struct pci_dev *pdev = adapter->pdev;
2793 int i, err = 0;
2794
2795 for (i = 0; i < adapter->num_rx_queues; i++) {
2796 err = igb_setup_rx_resources(adapter->rx_ring[i]);
2797 if (err) {
2798 dev_err(&pdev->dev,
2799 "Allocation for Rx Queue %u failed\n", i);
2800 for (i--; i >= 0; i--)
2801 igb_free_rx_resources(adapter->rx_ring[i]);
2802 break;
2803 }
2804 }
2805
2806 return err;
2807}
2808
2809/**
2810 * igb_setup_mrqc - configure the multiple receive queue control registers
2811 * @adapter: Board private structure
2812 **/
2813static void igb_setup_mrqc(struct igb_adapter *adapter)
2814{
2815 struct e1000_hw *hw = &adapter->hw;
2816 u32 mrqc, rxcsum;
2817 u32 j, num_rx_queues, shift = 0, shift2 = 0;
2818 union e1000_reta {
2819 u32 dword;
2820 u8 bytes[4];
2821 } reta;
2822 static const u8 rsshash[40] = {
2823 0x6d, 0x5a, 0x56, 0xda, 0x25, 0x5b, 0x0e, 0xc2, 0x41, 0x67,
2824 0x25, 0x3d, 0x43, 0xa3, 0x8f, 0xb0, 0xd0, 0xca, 0x2b, 0xcb,
2825 0xae, 0x7b, 0x30, 0xb4, 0x77, 0xcb, 0x2d, 0xa3, 0x80, 0x30,
2826 0xf2, 0x0c, 0x6a, 0x42, 0xb7, 0x3b, 0xbe, 0xac, 0x01, 0xfa };
2827
2828 /* Fill out hash function seeds */
2829 for (j = 0; j < 10; j++) {
2830 u32 rsskey = rsshash[(j * 4)];
2831 rsskey |= rsshash[(j * 4) + 1] << 8;
2832 rsskey |= rsshash[(j * 4) + 2] << 16;
2833 rsskey |= rsshash[(j * 4) + 3] << 24;
2834 array_wr32(E1000_RSSRK(0), j, rsskey);
2835 }
2836
2837 num_rx_queues = adapter->rss_queues;
2838
2839 if (adapter->vfs_allocated_count) {
2840 /* 82575 and 82576 supports 2 RSS queues for VMDq */
2841 switch (hw->mac.type) {
2842 case e1000_i350:
2843 case e1000_82580:
2844 num_rx_queues = 1;
2845 shift = 0;
2846 break;
2847 case e1000_82576:
2848 shift = 3;
2849 num_rx_queues = 2;
2850 break;
2851 case e1000_82575:
2852 shift = 2;
2853 shift2 = 6;
2854 default:
2855 break;
2856 }
2857 } else {
2858 if (hw->mac.type == e1000_82575)
2859 shift = 6;
2860 }
2861
2862 for (j = 0; j < (32 * 4); j++) {
2863 reta.bytes[j & 3] = (j % num_rx_queues) << shift;
2864 if (shift2)
2865 reta.bytes[j & 3] |= num_rx_queues << shift2;
2866 if ((j & 3) == 3)
2867 wr32(E1000_RETA(j >> 2), reta.dword);
2868 }
2869
2870 /*
2871 * Disable raw packet checksumming so that RSS hash is placed in
2872 * descriptor on writeback. No need to enable TCP/UDP/IP checksum
2873 * offloads as they are enabled by default
2874 */
2875 rxcsum = rd32(E1000_RXCSUM);
2876 rxcsum |= E1000_RXCSUM_PCSD;
2877
2878 if (adapter->hw.mac.type >= e1000_82576)
2879 /* Enable Receive Checksum Offload for SCTP */
2880 rxcsum |= E1000_RXCSUM_CRCOFL;
2881
2882 /* Don't need to set TUOFL or IPOFL, they default to 1 */
2883 wr32(E1000_RXCSUM, rxcsum);
2884
2885 /* If VMDq is enabled then we set the appropriate mode for that, else
2886 * we default to RSS so that an RSS hash is calculated per packet even
2887 * if we are only using one queue */
2888 if (adapter->vfs_allocated_count) {
2889 if (hw->mac.type > e1000_82575) {
2890 /* Set the default pool for the PF's first queue */
2891 u32 vtctl = rd32(E1000_VT_CTL);
2892 vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK |
2893 E1000_VT_CTL_DISABLE_DEF_POOL);
2894 vtctl |= adapter->vfs_allocated_count <<
2895 E1000_VT_CTL_DEFAULT_POOL_SHIFT;
2896 wr32(E1000_VT_CTL, vtctl);
2897 }
2898 if (adapter->rss_queues > 1)
2899 mrqc = E1000_MRQC_ENABLE_VMDQ_RSS_2Q;
2900 else
2901 mrqc = E1000_MRQC_ENABLE_VMDQ;
2902 } else {
2903 mrqc = E1000_MRQC_ENABLE_RSS_4Q;
2904 }
2905 igb_vmm_control(adapter);
2906
2907 /*
2908 * Generate RSS hash based on TCP port numbers and/or
2909 * IPv4/v6 src and dst addresses since UDP cannot be
2910 * hashed reliably due to IP fragmentation
2911 */
2912 mrqc |= E1000_MRQC_RSS_FIELD_IPV4 |
2913 E1000_MRQC_RSS_FIELD_IPV4_TCP |
2914 E1000_MRQC_RSS_FIELD_IPV6 |
2915 E1000_MRQC_RSS_FIELD_IPV6_TCP |
2916 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX;
2917
2918 wr32(E1000_MRQC, mrqc);
2919}
2920
2921/**
2922 * igb_setup_rctl - configure the receive control registers
2923 * @adapter: Board private structure
2924 **/
2925void igb_setup_rctl(struct igb_adapter *adapter)
2926{
2927 struct e1000_hw *hw = &adapter->hw;
2928 u32 rctl;
2929
2930 rctl = rd32(E1000_RCTL);
2931
2932 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2933 rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
2934
2935 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
2936 (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2937
2938 /*
2939 * enable stripping of CRC. It's unlikely this will break BMC
2940 * redirection as it did with e1000. Newer features require
2941 * that the HW strips the CRC.
2942 */
2943 rctl |= E1000_RCTL_SECRC;
2944
2945 /* disable store bad packets and clear size bits. */
2946 rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
2947
2948 /* enable LPE to prevent packets larger than max_frame_size */
2949 rctl |= E1000_RCTL_LPE;
2950
2951 /* disable queue 0 to prevent tail write w/o re-config */
2952 wr32(E1000_RXDCTL(0), 0);
2953
2954 /* Attention!!! For SR-IOV PF driver operations you must enable
2955 * queue drop for all VF and PF queues to prevent head of line blocking
2956 * if an un-trusted VF does not provide descriptors to hardware.
2957 */
2958 if (adapter->vfs_allocated_count) {
2959 /* set all queue drop enable bits */
2960 wr32(E1000_QDE, ALL_QUEUES);
2961 }
2962
2963 wr32(E1000_RCTL, rctl);
2964}
2965
2966static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
2967 int vfn)
2968{
2969 struct e1000_hw *hw = &adapter->hw;
2970 u32 vmolr;
2971
2972 /* if it isn't the PF check to see if VFs are enabled and
2973 * increase the size to support vlan tags */
2974 if (vfn < adapter->vfs_allocated_count &&
2975 adapter->vf_data[vfn].vlans_enabled)
2976 size += VLAN_TAG_SIZE;
2977
2978 vmolr = rd32(E1000_VMOLR(vfn));
2979 vmolr &= ~E1000_VMOLR_RLPML_MASK;
2980 vmolr |= size | E1000_VMOLR_LPE;
2981 wr32(E1000_VMOLR(vfn), vmolr);
2982
2983 return 0;
2984}
2985
2986/**
2987 * igb_rlpml_set - set maximum receive packet size
2988 * @adapter: board private structure
2989 *
2990 * Configure maximum receivable packet size.
2991 **/
2992static void igb_rlpml_set(struct igb_adapter *adapter)
2993{
2994 u32 max_frame_size = adapter->max_frame_size;
2995 struct e1000_hw *hw = &adapter->hw;
2996 u16 pf_id = adapter->vfs_allocated_count;
2997
2998 if (pf_id) {
2999 igb_set_vf_rlpml(adapter, max_frame_size, pf_id);
3000 /*
3001 * If we're in VMDQ or SR-IOV mode, then set global RLPML
3002 * to our max jumbo frame size, in case we need to enable
3003 * jumbo frames on one of the rings later.
3004 * This will not pass over-length frames into the default
3005 * queue because it's gated by the VMOLR.RLPML.
3006 */
3007 max_frame_size = MAX_JUMBO_FRAME_SIZE;
3008 }
3009
3010 wr32(E1000_RLPML, max_frame_size);
3011}
3012
3013static inline void igb_set_vmolr(struct igb_adapter *adapter,
3014 int vfn, bool aupe)
3015{
3016 struct e1000_hw *hw = &adapter->hw;
3017 u32 vmolr;
3018
3019 /*
3020 * This register exists only on 82576 and newer so if we are older then
3021 * we should exit and do nothing
3022 */
3023 if (hw->mac.type < e1000_82576)
3024 return;
3025
3026 vmolr = rd32(E1000_VMOLR(vfn));
3027 vmolr |= E1000_VMOLR_STRVLAN; /* Strip vlan tags */
3028 if (aupe)
3029 vmolr |= E1000_VMOLR_AUPE; /* Accept untagged packets */
3030 else
3031 vmolr &= ~(E1000_VMOLR_AUPE); /* Tagged packets ONLY */
3032
3033 /* clear all bits that might not be set */
3034 vmolr &= ~(E1000_VMOLR_BAM | E1000_VMOLR_RSSE);
3035
3036 if (adapter->rss_queues > 1 && vfn == adapter->vfs_allocated_count)
3037 vmolr |= E1000_VMOLR_RSSE; /* enable RSS */
3038 /*
3039 * for VMDq only allow the VFs and pool 0 to accept broadcast and
3040 * multicast packets
3041 */
3042 if (vfn <= adapter->vfs_allocated_count)
3043 vmolr |= E1000_VMOLR_BAM; /* Accept broadcast */
3044
3045 wr32(E1000_VMOLR(vfn), vmolr);
3046}
3047
3048/**
3049 * igb_configure_rx_ring - Configure a receive ring after Reset
3050 * @adapter: board private structure
3051 * @ring: receive ring to be configured
3052 *
3053 * Configure the Rx unit of the MAC after a reset.
3054 **/
3055void igb_configure_rx_ring(struct igb_adapter *adapter,
3056 struct igb_ring *ring)
3057{
3058 struct e1000_hw *hw = &adapter->hw;
3059 u64 rdba = ring->dma;
3060 int reg_idx = ring->reg_idx;
3061 u32 srrctl = 0, rxdctl = 0;
3062
3063 /* disable the queue */
3064 wr32(E1000_RXDCTL(reg_idx), 0);
3065
3066 /* Set DMA base address registers */
3067 wr32(E1000_RDBAL(reg_idx),
3068 rdba & 0x00000000ffffffffULL);
3069 wr32(E1000_RDBAH(reg_idx), rdba >> 32);
3070 wr32(E1000_RDLEN(reg_idx),
3071 ring->count * sizeof(union e1000_adv_rx_desc));
3072
3073 /* initialize head and tail */
3074 ring->tail = hw->hw_addr + E1000_RDT(reg_idx);
3075 wr32(E1000_RDH(reg_idx), 0);
3076 writel(0, ring->tail);
3077
3078 /* set descriptor configuration */
3079 srrctl = IGB_RX_HDR_LEN << E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
3080#if (PAGE_SIZE / 2) > IGB_RXBUFFER_16384
3081 srrctl |= IGB_RXBUFFER_16384 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
3082#else
3083 srrctl |= (PAGE_SIZE / 2) >> E1000_SRRCTL_BSIZEPKT_SHIFT;
3084#endif
3085 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
3086 if (hw->mac.type >= e1000_82580)
3087 srrctl |= E1000_SRRCTL_TIMESTAMP;
3088 /* Only set Drop Enable if we are supporting multiple queues */
3089 if (adapter->vfs_allocated_count || adapter->num_rx_queues > 1)
3090 srrctl |= E1000_SRRCTL_DROP_EN;
3091
3092 wr32(E1000_SRRCTL(reg_idx), srrctl);
3093
3094 /* set filtering for VMDQ pools */
3095 igb_set_vmolr(adapter, reg_idx & 0x7, true);
3096
3097 rxdctl |= IGB_RX_PTHRESH;
3098 rxdctl |= IGB_RX_HTHRESH << 8;
3099 rxdctl |= IGB_RX_WTHRESH << 16;
3100
3101 /* enable receive descriptor fetching */
3102 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
3103 wr32(E1000_RXDCTL(reg_idx), rxdctl);
3104}
3105
3106/**
3107 * igb_configure_rx - Configure receive Unit after Reset
3108 * @adapter: board private structure
3109 *
3110 * Configure the Rx unit of the MAC after a reset.
3111 **/
3112static void igb_configure_rx(struct igb_adapter *adapter)
3113{
3114 int i;
3115
3116 /* set UTA to appropriate mode */
3117 igb_set_uta(adapter);
3118
3119 /* set the correct pool for the PF default MAC address in entry 0 */
3120 igb_rar_set_qsel(adapter, adapter->hw.mac.addr, 0,
3121 adapter->vfs_allocated_count);
3122
3123 /* Setup the HW Rx Head and Tail Descriptor Pointers and
3124 * the Base and Length of the Rx Descriptor Ring */
3125 for (i = 0; i < adapter->num_rx_queues; i++)
3126 igb_configure_rx_ring(adapter, adapter->rx_ring[i]);
3127}
3128
3129/**
3130 * igb_free_tx_resources - Free Tx Resources per Queue
3131 * @tx_ring: Tx descriptor ring for a specific queue
3132 *
3133 * Free all transmit software resources
3134 **/
3135void igb_free_tx_resources(struct igb_ring *tx_ring)
3136{
3137 igb_clean_tx_ring(tx_ring);
3138
3139 vfree(tx_ring->tx_buffer_info);
3140 tx_ring->tx_buffer_info = NULL;
3141
3142 /* if not set, then don't free */
3143 if (!tx_ring->desc)
3144 return;
3145
3146 dma_free_coherent(tx_ring->dev, tx_ring->size,
3147 tx_ring->desc, tx_ring->dma);
3148
3149 tx_ring->desc = NULL;
3150}
3151
3152/**
3153 * igb_free_all_tx_resources - Free Tx Resources for All Queues
3154 * @adapter: board private structure
3155 *
3156 * Free all transmit software resources
3157 **/
3158static void igb_free_all_tx_resources(struct igb_adapter *adapter)
3159{
3160 int i;
3161
3162 for (i = 0; i < adapter->num_tx_queues; i++)
3163 igb_free_tx_resources(adapter->tx_ring[i]);
3164}
3165
3166void igb_unmap_and_free_tx_resource(struct igb_ring *ring,
3167 struct igb_tx_buffer *tx_buffer)
3168{
3169 if (tx_buffer->skb) {
3170 dev_kfree_skb_any(tx_buffer->skb);
3171 if (tx_buffer->dma)
3172 dma_unmap_single(ring->dev,
3173 tx_buffer->dma,
3174 tx_buffer->length,
3175 DMA_TO_DEVICE);
3176 } else if (tx_buffer->dma) {
3177 dma_unmap_page(ring->dev,
3178 tx_buffer->dma,
3179 tx_buffer->length,
3180 DMA_TO_DEVICE);
3181 }
3182 tx_buffer->next_to_watch = NULL;
3183 tx_buffer->skb = NULL;
3184 tx_buffer->dma = 0;
3185 /* buffer_info must be completely set up in the transmit path */
3186}
3187
3188/**
3189 * igb_clean_tx_ring - Free Tx Buffers
3190 * @tx_ring: ring to be cleaned
3191 **/
3192static void igb_clean_tx_ring(struct igb_ring *tx_ring)
3193{
3194 struct igb_tx_buffer *buffer_info;
3195 unsigned long size;
3196 u16 i;
3197
3198 if (!tx_ring->tx_buffer_info)
3199 return;
3200 /* Free all the Tx ring sk_buffs */
3201
3202 for (i = 0; i < tx_ring->count; i++) {
3203 buffer_info = &tx_ring->tx_buffer_info[i];
3204 igb_unmap_and_free_tx_resource(tx_ring, buffer_info);
3205 }
3206
3207 size = sizeof(struct igb_tx_buffer) * tx_ring->count;
3208 memset(tx_ring->tx_buffer_info, 0, size);
3209
3210 /* Zero out the descriptor ring */
3211 memset(tx_ring->desc, 0, tx_ring->size);
3212
3213 tx_ring->next_to_use = 0;
3214 tx_ring->next_to_clean = 0;
3215}
3216
3217/**
3218 * igb_clean_all_tx_rings - Free Tx Buffers for all queues
3219 * @adapter: board private structure
3220 **/
3221static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
3222{
3223 int i;
3224
3225 for (i = 0; i < adapter->num_tx_queues; i++)
3226 igb_clean_tx_ring(adapter->tx_ring[i]);
3227}
3228
3229/**
3230 * igb_free_rx_resources - Free Rx Resources
3231 * @rx_ring: ring to clean the resources from
3232 *
3233 * Free all receive software resources
3234 **/
3235void igb_free_rx_resources(struct igb_ring *rx_ring)
3236{
3237 igb_clean_rx_ring(rx_ring);
3238
3239 vfree(rx_ring->rx_buffer_info);
3240 rx_ring->rx_buffer_info = NULL;
3241
3242 /* if not set, then don't free */
3243 if (!rx_ring->desc)
3244 return;
3245
3246 dma_free_coherent(rx_ring->dev, rx_ring->size,
3247 rx_ring->desc, rx_ring->dma);
3248
3249 rx_ring->desc = NULL;
3250}
3251
3252/**
3253 * igb_free_all_rx_resources - Free Rx Resources for All Queues
3254 * @adapter: board private structure
3255 *
3256 * Free all receive software resources
3257 **/
3258static void igb_free_all_rx_resources(struct igb_adapter *adapter)
3259{
3260 int i;
3261
3262 for (i = 0; i < adapter->num_rx_queues; i++)
3263 igb_free_rx_resources(adapter->rx_ring[i]);
3264}
3265
3266/**
3267 * igb_clean_rx_ring - Free Rx Buffers per Queue
3268 * @rx_ring: ring to free buffers from
3269 **/
3270static void igb_clean_rx_ring(struct igb_ring *rx_ring)
3271{
3272 unsigned long size;
3273 u16 i;
3274
3275 if (!rx_ring->rx_buffer_info)
3276 return;
3277
3278 /* Free all the Rx ring sk_buffs */
3279 for (i = 0; i < rx_ring->count; i++) {
3280 struct igb_rx_buffer *buffer_info = &rx_ring->rx_buffer_info[i];
3281 if (buffer_info->dma) {
3282 dma_unmap_single(rx_ring->dev,
3283 buffer_info->dma,
3284 IGB_RX_HDR_LEN,
3285 DMA_FROM_DEVICE);
3286 buffer_info->dma = 0;
3287 }
3288
3289 if (buffer_info->skb) {
3290 dev_kfree_skb(buffer_info->skb);
3291 buffer_info->skb = NULL;
3292 }
3293 if (buffer_info->page_dma) {
3294 dma_unmap_page(rx_ring->dev,
3295 buffer_info->page_dma,
3296 PAGE_SIZE / 2,
3297 DMA_FROM_DEVICE);
3298 buffer_info->page_dma = 0;
3299 }
3300 if (buffer_info->page) {
3301 put_page(buffer_info->page);
3302 buffer_info->page = NULL;
3303 buffer_info->page_offset = 0;
3304 }
3305 }
3306
3307 size = sizeof(struct igb_rx_buffer) * rx_ring->count;
3308 memset(rx_ring->rx_buffer_info, 0, size);
3309
3310 /* Zero out the descriptor ring */
3311 memset(rx_ring->desc, 0, rx_ring->size);
3312
3313 rx_ring->next_to_clean = 0;
3314 rx_ring->next_to_use = 0;
3315}
3316
3317/**
3318 * igb_clean_all_rx_rings - Free Rx Buffers for all queues
3319 * @adapter: board private structure
3320 **/
3321static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
3322{
3323 int i;
3324
3325 for (i = 0; i < adapter->num_rx_queues; i++)
3326 igb_clean_rx_ring(adapter->rx_ring[i]);
3327}
3328
3329/**
3330 * igb_set_mac - Change the Ethernet Address of the NIC
3331 * @netdev: network interface device structure
3332 * @p: pointer to an address structure
3333 *
3334 * Returns 0 on success, negative on failure
3335 **/
3336static int igb_set_mac(struct net_device *netdev, void *p)
3337{
3338 struct igb_adapter *adapter = netdev_priv(netdev);
3339 struct e1000_hw *hw = &adapter->hw;
3340 struct sockaddr *addr = p;
3341
3342 if (!is_valid_ether_addr(addr->sa_data))
3343 return -EADDRNOTAVAIL;
3344
3345 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3346 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
3347
3348 /* set the correct pool for the new PF MAC address in entry 0 */
3349 igb_rar_set_qsel(adapter, hw->mac.addr, 0,
3350 adapter->vfs_allocated_count);
3351
3352 return 0;
3353}
3354
3355/**
3356 * igb_write_mc_addr_list - write multicast addresses to MTA
3357 * @netdev: network interface device structure
3358 *
3359 * Writes multicast address list to the MTA hash table.
3360 * Returns: -ENOMEM on failure
3361 * 0 on no addresses written
3362 * X on writing X addresses to MTA
3363 **/
3364static int igb_write_mc_addr_list(struct net_device *netdev)
3365{
3366 struct igb_adapter *adapter = netdev_priv(netdev);
3367 struct e1000_hw *hw = &adapter->hw;
3368 struct netdev_hw_addr *ha;
3369 u8 *mta_list;
3370 int i;
3371
3372 if (netdev_mc_empty(netdev)) {
3373 /* nothing to program, so clear mc list */
3374 igb_update_mc_addr_list(hw, NULL, 0);
3375 igb_restore_vf_multicasts(adapter);
3376 return 0;
3377 }
3378
3379 mta_list = kzalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
3380 if (!mta_list)
3381 return -ENOMEM;
3382
3383 /* The shared function expects a packed array of only addresses. */
3384 i = 0;
3385 netdev_for_each_mc_addr(ha, netdev)
3386 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3387
3388 igb_update_mc_addr_list(hw, mta_list, i);
3389 kfree(mta_list);
3390
3391 return netdev_mc_count(netdev);
3392}
3393
3394/**
3395 * igb_write_uc_addr_list - write unicast addresses to RAR table
3396 * @netdev: network interface device structure
3397 *
3398 * Writes unicast address list to the RAR table.
3399 * Returns: -ENOMEM on failure/insufficient address space
3400 * 0 on no addresses written
3401 * X on writing X addresses to the RAR table
3402 **/
3403static int igb_write_uc_addr_list(struct net_device *netdev)
3404{
3405 struct igb_adapter *adapter = netdev_priv(netdev);
3406 struct e1000_hw *hw = &adapter->hw;
3407 unsigned int vfn = adapter->vfs_allocated_count;
3408 unsigned int rar_entries = hw->mac.rar_entry_count - (vfn + 1);
3409 int count = 0;
3410
3411 /* return ENOMEM indicating insufficient memory for addresses */
3412 if (netdev_uc_count(netdev) > rar_entries)
3413 return -ENOMEM;
3414
3415 if (!netdev_uc_empty(netdev) && rar_entries) {
3416 struct netdev_hw_addr *ha;
3417
3418 netdev_for_each_uc_addr(ha, netdev) {
3419 if (!rar_entries)
3420 break;
3421 igb_rar_set_qsel(adapter, ha->addr,
3422 rar_entries--,
3423 vfn);
3424 count++;
3425 }
3426 }
3427 /* write the addresses in reverse order to avoid write combining */
3428 for (; rar_entries > 0 ; rar_entries--) {
3429 wr32(E1000_RAH(rar_entries), 0);
3430 wr32(E1000_RAL(rar_entries), 0);
3431 }
3432 wrfl();
3433
3434 return count;
3435}
3436
3437/**
3438 * igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
3439 * @netdev: network interface device structure
3440 *
3441 * The set_rx_mode entry point is called whenever the unicast or multicast
3442 * address lists or the network interface flags are updated. This routine is
3443 * responsible for configuring the hardware for proper unicast, multicast,
3444 * promiscuous mode, and all-multi behavior.
3445 **/
3446static void igb_set_rx_mode(struct net_device *netdev)
3447{
3448 struct igb_adapter *adapter = netdev_priv(netdev);
3449 struct e1000_hw *hw = &adapter->hw;
3450 unsigned int vfn = adapter->vfs_allocated_count;
3451 u32 rctl, vmolr = 0;
3452 int count;
3453
3454 /* Check for Promiscuous and All Multicast modes */
3455 rctl = rd32(E1000_RCTL);
3456
3457 /* clear the effected bits */
3458 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE | E1000_RCTL_VFE);
3459
3460 if (netdev->flags & IFF_PROMISC) {
3461 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3462 vmolr |= (E1000_VMOLR_ROPE | E1000_VMOLR_MPME);
3463 } else {
3464 if (netdev->flags & IFF_ALLMULTI) {
3465 rctl |= E1000_RCTL_MPE;
3466 vmolr |= E1000_VMOLR_MPME;
3467 } else {
3468 /*
3469 * Write addresses to the MTA, if the attempt fails
3470 * then we should just turn on promiscuous mode so
3471 * that we can at least receive multicast traffic
3472 */
3473 count = igb_write_mc_addr_list(netdev);
3474 if (count < 0) {
3475 rctl |= E1000_RCTL_MPE;
3476 vmolr |= E1000_VMOLR_MPME;
3477 } else if (count) {
3478 vmolr |= E1000_VMOLR_ROMPE;
3479 }
3480 }
3481 /*
3482 * Write addresses to available RAR registers, if there is not
3483 * sufficient space to store all the addresses then enable
3484 * unicast promiscuous mode
3485 */
3486 count = igb_write_uc_addr_list(netdev);
3487 if (count < 0) {
3488 rctl |= E1000_RCTL_UPE;
3489 vmolr |= E1000_VMOLR_ROPE;
3490 }
3491 rctl |= E1000_RCTL_VFE;
3492 }
3493 wr32(E1000_RCTL, rctl);
3494
3495 /*
3496 * In order to support SR-IOV and eventually VMDq it is necessary to set
3497 * the VMOLR to enable the appropriate modes. Without this workaround
3498 * we will have issues with VLAN tag stripping not being done for frames
3499 * that are only arriving because we are the default pool
3500 */
3501 if (hw->mac.type < e1000_82576)
3502 return;
3503
3504 vmolr |= rd32(E1000_VMOLR(vfn)) &
3505 ~(E1000_VMOLR_ROPE | E1000_VMOLR_MPME | E1000_VMOLR_ROMPE);
3506 wr32(E1000_VMOLR(vfn), vmolr);
3507 igb_restore_vf_multicasts(adapter);
3508}
3509
3510static void igb_check_wvbr(struct igb_adapter *adapter)
3511{
3512 struct e1000_hw *hw = &adapter->hw;
3513 u32 wvbr = 0;
3514
3515 switch (hw->mac.type) {
3516 case e1000_82576:
3517 case e1000_i350:
3518 if (!(wvbr = rd32(E1000_WVBR)))
3519 return;
3520 break;
3521 default:
3522 break;
3523 }
3524
3525 adapter->wvbr |= wvbr;
3526}
3527
3528#define IGB_STAGGERED_QUEUE_OFFSET 8
3529
3530static void igb_spoof_check(struct igb_adapter *adapter)
3531{
3532 int j;
3533
3534 if (!adapter->wvbr)
3535 return;
3536
3537 for(j = 0; j < adapter->vfs_allocated_count; j++) {
3538 if (adapter->wvbr & (1 << j) ||
3539 adapter->wvbr & (1 << (j + IGB_STAGGERED_QUEUE_OFFSET))) {
3540 dev_warn(&adapter->pdev->dev,
3541 "Spoof event(s) detected on VF %d\n", j);
3542 adapter->wvbr &=
3543 ~((1 << j) |
3544 (1 << (j + IGB_STAGGERED_QUEUE_OFFSET)));
3545 }
3546 }
3547}
3548
3549/* Need to wait a few seconds after link up to get diagnostic information from
3550 * the phy */
3551static void igb_update_phy_info(unsigned long data)
3552{
3553 struct igb_adapter *adapter = (struct igb_adapter *) data;
3554 igb_get_phy_info(&adapter->hw);
3555}
3556
3557/**
3558 * igb_has_link - check shared code for link and determine up/down
3559 * @adapter: pointer to driver private info
3560 **/
3561bool igb_has_link(struct igb_adapter *adapter)
3562{
3563 struct e1000_hw *hw = &adapter->hw;
3564 bool link_active = false;
3565 s32 ret_val = 0;
3566
3567 /* get_link_status is set on LSC (link status) interrupt or
3568 * rx sequence error interrupt. get_link_status will stay
3569 * false until the e1000_check_for_link establishes link
3570 * for copper adapters ONLY
3571 */
3572 switch (hw->phy.media_type) {
3573 case e1000_media_type_copper:
3574 if (hw->mac.get_link_status) {
3575 ret_val = hw->mac.ops.check_for_link(hw);
3576 link_active = !hw->mac.get_link_status;
3577 } else {
3578 link_active = true;
3579 }
3580 break;
3581 case e1000_media_type_internal_serdes:
3582 ret_val = hw->mac.ops.check_for_link(hw);
3583 link_active = hw->mac.serdes_has_link;
3584 break;
3585 default:
3586 case e1000_media_type_unknown:
3587 break;
3588 }
3589
3590 return link_active;
3591}
3592
3593static bool igb_thermal_sensor_event(struct e1000_hw *hw, u32 event)
3594{
3595 bool ret = false;
3596 u32 ctrl_ext, thstat;
3597
3598 /* check for thermal sensor event on i350, copper only */
3599 if (hw->mac.type == e1000_i350) {
3600 thstat = rd32(E1000_THSTAT);
3601 ctrl_ext = rd32(E1000_CTRL_EXT);
3602
3603 if ((hw->phy.media_type == e1000_media_type_copper) &&
3604 !(ctrl_ext & E1000_CTRL_EXT_LINK_MODE_SGMII)) {
3605 ret = !!(thstat & event);
3606 }
3607 }
3608
3609 return ret;
3610}
3611
3612/**
3613 * igb_watchdog - Timer Call-back
3614 * @data: pointer to adapter cast into an unsigned long
3615 **/
3616static void igb_watchdog(unsigned long data)
3617{
3618 struct igb_adapter *adapter = (struct igb_adapter *)data;
3619 /* Do the rest outside of interrupt context */
3620 schedule_work(&adapter->watchdog_task);
3621}
3622
3623static void igb_watchdog_task(struct work_struct *work)
3624{
3625 struct igb_adapter *adapter = container_of(work,
3626 struct igb_adapter,
3627 watchdog_task);
3628 struct e1000_hw *hw = &adapter->hw;
3629 struct net_device *netdev = adapter->netdev;
3630 u32 link;
3631 int i;
3632
3633 link = igb_has_link(adapter);
3634 if (link) {
3635 if (!netif_carrier_ok(netdev)) {
3636 u32 ctrl;
3637 hw->mac.ops.get_speed_and_duplex(hw,
3638 &adapter->link_speed,
3639 &adapter->link_duplex);
3640
3641 ctrl = rd32(E1000_CTRL);
3642 /* Links status message must follow this format */
3643 printk(KERN_INFO "igb: %s NIC Link is Up %d Mbps %s, "
3644 "Flow Control: %s\n",
3645 netdev->name,
3646 adapter->link_speed,
3647 adapter->link_duplex == FULL_DUPLEX ?
3648 "Full Duplex" : "Half Duplex",
3649 ((ctrl & E1000_CTRL_TFCE) &&
3650 (ctrl & E1000_CTRL_RFCE)) ? "RX/TX" :
3651 ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3652 ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None")));
3653
3654 /* check for thermal sensor event */
3655 if (igb_thermal_sensor_event(hw, E1000_THSTAT_LINK_THROTTLE)) {
3656 printk(KERN_INFO "igb: %s The network adapter "
3657 "link speed was downshifted "
3658 "because it overheated.\n",
3659 netdev->name);
3660 }
3661
3662 /* adjust timeout factor according to speed/duplex */
3663 adapter->tx_timeout_factor = 1;
3664 switch (adapter->link_speed) {
3665 case SPEED_10:
3666 adapter->tx_timeout_factor = 14;
3667 break;
3668 case SPEED_100:
3669 /* maybe add some timeout factor ? */
3670 break;
3671 }
3672
3673 netif_carrier_on(netdev);
3674
3675 igb_ping_all_vfs(adapter);
3676 igb_check_vf_rate_limit(adapter);
3677
3678 /* link state has changed, schedule phy info update */
3679 if (!test_bit(__IGB_DOWN, &adapter->state))
3680 mod_timer(&adapter->phy_info_timer,
3681 round_jiffies(jiffies + 2 * HZ));
3682 }
3683 } else {
3684 if (netif_carrier_ok(netdev)) {
3685 adapter->link_speed = 0;
3686 adapter->link_duplex = 0;
3687
3688 /* check for thermal sensor event */
3689 if (igb_thermal_sensor_event(hw, E1000_THSTAT_PWR_DOWN)) {
3690 printk(KERN_ERR "igb: %s The network adapter "
3691 "was stopped because it "
3692 "overheated.\n",
3693 netdev->name);
3694 }
3695
3696 /* Links status message must follow this format */
3697 printk(KERN_INFO "igb: %s NIC Link is Down\n",
3698 netdev->name);
3699 netif_carrier_off(netdev);
3700
3701 igb_ping_all_vfs(adapter);
3702
3703 /* link state has changed, schedule phy info update */
3704 if (!test_bit(__IGB_DOWN, &adapter->state))
3705 mod_timer(&adapter->phy_info_timer,
3706 round_jiffies(jiffies + 2 * HZ));
3707 }
3708 }
3709
3710 spin_lock(&adapter->stats64_lock);
3711 igb_update_stats(adapter, &adapter->stats64);
3712 spin_unlock(&adapter->stats64_lock);
3713
3714 for (i = 0; i < adapter->num_tx_queues; i++) {
3715 struct igb_ring *tx_ring = adapter->tx_ring[i];
3716 if (!netif_carrier_ok(netdev)) {
3717 /* We've lost link, so the controller stops DMA,
3718 * but we've got queued Tx work that's never going
3719 * to get done, so reset controller to flush Tx.
3720 * (Do the reset outside of interrupt context). */
3721 if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) {
3722 adapter->tx_timeout_count++;
3723 schedule_work(&adapter->reset_task);
3724 /* return immediately since reset is imminent */
3725 return;
3726 }
3727 }
3728
3729 /* Force detection of hung controller every watchdog period */
3730 set_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
3731 }
3732
3733 /* Cause software interrupt to ensure rx ring is cleaned */
3734 if (adapter->msix_entries) {
3735 u32 eics = 0;
3736 for (i = 0; i < adapter->num_q_vectors; i++)
3737 eics |= adapter->q_vector[i]->eims_value;
3738 wr32(E1000_EICS, eics);
3739 } else {
3740 wr32(E1000_ICS, E1000_ICS_RXDMT0);
3741 }
3742
3743 igb_spoof_check(adapter);
3744
3745 /* Reset the timer */
3746 if (!test_bit(__IGB_DOWN, &adapter->state))
3747 mod_timer(&adapter->watchdog_timer,
3748 round_jiffies(jiffies + 2 * HZ));
3749}
3750
3751enum latency_range {
3752 lowest_latency = 0,
3753 low_latency = 1,
3754 bulk_latency = 2,
3755 latency_invalid = 255
3756};
3757
3758/**
3759 * igb_update_ring_itr - update the dynamic ITR value based on packet size
3760 *
3761 * Stores a new ITR value based on strictly on packet size. This
3762 * algorithm is less sophisticated than that used in igb_update_itr,
3763 * due to the difficulty of synchronizing statistics across multiple
3764 * receive rings. The divisors and thresholds used by this function
3765 * were determined based on theoretical maximum wire speed and testing
3766 * data, in order to minimize response time while increasing bulk
3767 * throughput.
3768 * This functionality is controlled by the InterruptThrottleRate module
3769 * parameter (see igb_param.c)
3770 * NOTE: This function is called only when operating in a multiqueue
3771 * receive environment.
3772 * @q_vector: pointer to q_vector
3773 **/
3774static void igb_update_ring_itr(struct igb_q_vector *q_vector)
3775{
3776 int new_val = q_vector->itr_val;
3777 int avg_wire_size = 0;
3778 struct igb_adapter *adapter = q_vector->adapter;
3779 unsigned int packets;
3780
3781 /* For non-gigabit speeds, just fix the interrupt rate at 4000
3782 * ints/sec - ITR timer value of 120 ticks.
3783 */
3784 if (adapter->link_speed != SPEED_1000) {
3785 new_val = IGB_4K_ITR;
3786 goto set_itr_val;
3787 }
3788
3789 packets = q_vector->rx.total_packets;
3790 if (packets)
3791 avg_wire_size = q_vector->rx.total_bytes / packets;
3792
3793 packets = q_vector->tx.total_packets;
3794 if (packets)
3795 avg_wire_size = max_t(u32, avg_wire_size,
3796 q_vector->tx.total_bytes / packets);
3797
3798 /* if avg_wire_size isn't set no work was done */
3799 if (!avg_wire_size)
3800 goto clear_counts;
3801
3802 /* Add 24 bytes to size to account for CRC, preamble, and gap */
3803 avg_wire_size += 24;
3804
3805 /* Don't starve jumbo frames */
3806 avg_wire_size = min(avg_wire_size, 3000);
3807
3808 /* Give a little boost to mid-size frames */
3809 if ((avg_wire_size > 300) && (avg_wire_size < 1200))
3810 new_val = avg_wire_size / 3;
3811 else
3812 new_val = avg_wire_size / 2;
3813
3814 /* conservative mode (itr 3) eliminates the lowest_latency setting */
3815 if (new_val < IGB_20K_ITR &&
3816 ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
3817 (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
3818 new_val = IGB_20K_ITR;
3819
3820set_itr_val:
3821 if (new_val != q_vector->itr_val) {
3822 q_vector->itr_val = new_val;
3823 q_vector->set_itr = 1;
3824 }
3825clear_counts:
3826 q_vector->rx.total_bytes = 0;
3827 q_vector->rx.total_packets = 0;
3828 q_vector->tx.total_bytes = 0;
3829 q_vector->tx.total_packets = 0;
3830}
3831
3832/**
3833 * igb_update_itr - update the dynamic ITR value based on statistics
3834 * Stores a new ITR value based on packets and byte
3835 * counts during the last interrupt. The advantage of per interrupt
3836 * computation is faster updates and more accurate ITR for the current
3837 * traffic pattern. Constants in this function were computed
3838 * based on theoretical maximum wire speed and thresholds were set based
3839 * on testing data as well as attempting to minimize response time
3840 * while increasing bulk throughput.
3841 * this functionality is controlled by the InterruptThrottleRate module
3842 * parameter (see igb_param.c)
3843 * NOTE: These calculations are only valid when operating in a single-
3844 * queue environment.
3845 * @q_vector: pointer to q_vector
3846 * @ring_container: ring info to update the itr for
3847 **/
3848static void igb_update_itr(struct igb_q_vector *q_vector,
3849 struct igb_ring_container *ring_container)
3850{
3851 unsigned int packets = ring_container->total_packets;
3852 unsigned int bytes = ring_container->total_bytes;
3853 u8 itrval = ring_container->itr;
3854
3855 /* no packets, exit with status unchanged */
3856 if (packets == 0)
3857 return;
3858
3859 switch (itrval) {
3860 case lowest_latency:
3861 /* handle TSO and jumbo frames */
3862 if (bytes/packets > 8000)
3863 itrval = bulk_latency;
3864 else if ((packets < 5) && (bytes > 512))
3865 itrval = low_latency;
3866 break;
3867 case low_latency: /* 50 usec aka 20000 ints/s */
3868 if (bytes > 10000) {
3869 /* this if handles the TSO accounting */
3870 if (bytes/packets > 8000) {
3871 itrval = bulk_latency;
3872 } else if ((packets < 10) || ((bytes/packets) > 1200)) {
3873 itrval = bulk_latency;
3874 } else if ((packets > 35)) {
3875 itrval = lowest_latency;
3876 }
3877 } else if (bytes/packets > 2000) {
3878 itrval = bulk_latency;
3879 } else if (packets <= 2 && bytes < 512) {
3880 itrval = lowest_latency;
3881 }
3882 break;
3883 case bulk_latency: /* 250 usec aka 4000 ints/s */
3884 if (bytes > 25000) {
3885 if (packets > 35)
3886 itrval = low_latency;
3887 } else if (bytes < 1500) {
3888 itrval = low_latency;
3889 }
3890 break;
3891 }
3892
3893 /* clear work counters since we have the values we need */
3894 ring_container->total_bytes = 0;
3895 ring_container->total_packets = 0;
3896
3897 /* write updated itr to ring container */
3898 ring_container->itr = itrval;
3899}
3900
3901static void igb_set_itr(struct igb_q_vector *q_vector)
3902{
3903 struct igb_adapter *adapter = q_vector->adapter;
3904 u32 new_itr = q_vector->itr_val;
3905 u8 current_itr = 0;
3906
3907 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
3908 if (adapter->link_speed != SPEED_1000) {
3909 current_itr = 0;
3910 new_itr = IGB_4K_ITR;
3911 goto set_itr_now;
3912 }
3913
3914 igb_update_itr(q_vector, &q_vector->tx);
3915 igb_update_itr(q_vector, &q_vector->rx);
3916
3917 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
3918
3919 /* conservative mode (itr 3) eliminates the lowest_latency setting */
3920 if (current_itr == lowest_latency &&
3921 ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
3922 (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
3923 current_itr = low_latency;
3924
3925 switch (current_itr) {
3926 /* counts and packets in update_itr are dependent on these numbers */
3927 case lowest_latency:
3928 new_itr = IGB_70K_ITR; /* 70,000 ints/sec */
3929 break;
3930 case low_latency:
3931 new_itr = IGB_20K_ITR; /* 20,000 ints/sec */
3932 break;
3933 case bulk_latency:
3934 new_itr = IGB_4K_ITR; /* 4,000 ints/sec */
3935 break;
3936 default:
3937 break;
3938 }
3939
3940set_itr_now:
3941 if (new_itr != q_vector->itr_val) {
3942 /* this attempts to bias the interrupt rate towards Bulk
3943 * by adding intermediate steps when interrupt rate is
3944 * increasing */
3945 new_itr = new_itr > q_vector->itr_val ?
3946 max((new_itr * q_vector->itr_val) /
3947 (new_itr + (q_vector->itr_val >> 2)),
3948 new_itr) :
3949 new_itr;
3950 /* Don't write the value here; it resets the adapter's
3951 * internal timer, and causes us to delay far longer than
3952 * we should between interrupts. Instead, we write the ITR
3953 * value at the beginning of the next interrupt so the timing
3954 * ends up being correct.
3955 */
3956 q_vector->itr_val = new_itr;
3957 q_vector->set_itr = 1;
3958 }
3959}
3960
3961void igb_tx_ctxtdesc(struct igb_ring *tx_ring, u32 vlan_macip_lens,
3962 u32 type_tucmd, u32 mss_l4len_idx)
3963{
3964 struct e1000_adv_tx_context_desc *context_desc;
3965 u16 i = tx_ring->next_to_use;
3966
3967 context_desc = IGB_TX_CTXTDESC(tx_ring, i);
3968
3969 i++;
3970 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
3971
3972 /* set bits to identify this as an advanced context descriptor */
3973 type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
3974
3975 /* For 82575, context index must be unique per ring. */
3976 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
3977 mss_l4len_idx |= tx_ring->reg_idx << 4;
3978
3979 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
3980 context_desc->seqnum_seed = 0;
3981 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd);
3982 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
3983}
3984
3985static int igb_tso(struct igb_ring *tx_ring,
3986 struct igb_tx_buffer *first,
3987 u8 *hdr_len)
3988{
3989 struct sk_buff *skb = first->skb;
3990 u32 vlan_macip_lens, type_tucmd;
3991 u32 mss_l4len_idx, l4len;
3992
3993 if (!skb_is_gso(skb))
3994 return 0;
3995
3996 if (skb_header_cloned(skb)) {
3997 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3998 if (err)
3999 return err;
4000 }
4001
4002 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
4003 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
4004
4005 if (first->protocol == __constant_htons(ETH_P_IP)) {
4006 struct iphdr *iph = ip_hdr(skb);
4007 iph->tot_len = 0;
4008 iph->check = 0;
4009 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
4010 iph->daddr, 0,
4011 IPPROTO_TCP,
4012 0);
4013 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
4014 first->tx_flags |= IGB_TX_FLAGS_TSO |
4015 IGB_TX_FLAGS_CSUM |
4016 IGB_TX_FLAGS_IPV4;
4017 } else if (skb_is_gso_v6(skb)) {
4018 ipv6_hdr(skb)->payload_len = 0;
4019 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4020 &ipv6_hdr(skb)->daddr,
4021 0, IPPROTO_TCP, 0);
4022 first->tx_flags |= IGB_TX_FLAGS_TSO |
4023 IGB_TX_FLAGS_CSUM;
4024 }
4025
4026 /* compute header lengths */
4027 l4len = tcp_hdrlen(skb);
4028 *hdr_len = skb_transport_offset(skb) + l4len;
4029
4030 /* update gso size and bytecount with header size */
4031 first->gso_segs = skb_shinfo(skb)->gso_segs;
4032 first->bytecount += (first->gso_segs - 1) * *hdr_len;
4033
4034 /* MSS L4LEN IDX */
4035 mss_l4len_idx = l4len << E1000_ADVTXD_L4LEN_SHIFT;
4036 mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT;
4037
4038 /* VLAN MACLEN IPLEN */
4039 vlan_macip_lens = skb_network_header_len(skb);
4040 vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
4041 vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
4042
4043 igb_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
4044
4045 return 1;
4046}
4047
4048static void igb_tx_csum(struct igb_ring *tx_ring, struct igb_tx_buffer *first)
4049{
4050 struct sk_buff *skb = first->skb;
4051 u32 vlan_macip_lens = 0;
4052 u32 mss_l4len_idx = 0;
4053 u32 type_tucmd = 0;
4054
4055 if (skb->ip_summed != CHECKSUM_PARTIAL) {
4056 if (!(first->tx_flags & IGB_TX_FLAGS_VLAN))
4057 return;
4058 } else {
4059 u8 l4_hdr = 0;
4060 switch (first->protocol) {
4061 case __constant_htons(ETH_P_IP):
4062 vlan_macip_lens |= skb_network_header_len(skb);
4063 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
4064 l4_hdr = ip_hdr(skb)->protocol;
4065 break;
4066 case __constant_htons(ETH_P_IPV6):
4067 vlan_macip_lens |= skb_network_header_len(skb);
4068 l4_hdr = ipv6_hdr(skb)->nexthdr;
4069 break;
4070 default:
4071 if (unlikely(net_ratelimit())) {
4072 dev_warn(tx_ring->dev,
4073 "partial checksum but proto=%x!\n",
4074 first->protocol);
4075 }
4076 break;
4077 }
4078
4079 switch (l4_hdr) {
4080 case IPPROTO_TCP:
4081 type_tucmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
4082 mss_l4len_idx = tcp_hdrlen(skb) <<
4083 E1000_ADVTXD_L4LEN_SHIFT;
4084 break;
4085 case IPPROTO_SCTP:
4086 type_tucmd |= E1000_ADVTXD_TUCMD_L4T_SCTP;
4087 mss_l4len_idx = sizeof(struct sctphdr) <<
4088 E1000_ADVTXD_L4LEN_SHIFT;
4089 break;
4090 case IPPROTO_UDP:
4091 mss_l4len_idx = sizeof(struct udphdr) <<
4092 E1000_ADVTXD_L4LEN_SHIFT;
4093 break;
4094 default:
4095 if (unlikely(net_ratelimit())) {
4096 dev_warn(tx_ring->dev,
4097 "partial checksum but l4 proto=%x!\n",
4098 l4_hdr);
4099 }
4100 break;
4101 }
4102
4103 /* update TX checksum flag */
4104 first->tx_flags |= IGB_TX_FLAGS_CSUM;
4105 }
4106
4107 vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
4108 vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
4109
4110 igb_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
4111}
4112
4113static __le32 igb_tx_cmd_type(u32 tx_flags)
4114{
4115 /* set type for advanced descriptor with frame checksum insertion */
4116 __le32 cmd_type = cpu_to_le32(E1000_ADVTXD_DTYP_DATA |
4117 E1000_ADVTXD_DCMD_IFCS |
4118 E1000_ADVTXD_DCMD_DEXT);
4119
4120 /* set HW vlan bit if vlan is present */
4121 if (tx_flags & IGB_TX_FLAGS_VLAN)
4122 cmd_type |= cpu_to_le32(E1000_ADVTXD_DCMD_VLE);
4123
4124 /* set timestamp bit if present */
4125 if (tx_flags & IGB_TX_FLAGS_TSTAMP)
4126 cmd_type |= cpu_to_le32(E1000_ADVTXD_MAC_TSTAMP);
4127
4128 /* set segmentation bits for TSO */
4129 if (tx_flags & IGB_TX_FLAGS_TSO)
4130 cmd_type |= cpu_to_le32(E1000_ADVTXD_DCMD_TSE);
4131
4132 return cmd_type;
4133}
4134
4135static void igb_tx_olinfo_status(struct igb_ring *tx_ring,
4136 union e1000_adv_tx_desc *tx_desc,
4137 u32 tx_flags, unsigned int paylen)
4138{
4139 u32 olinfo_status = paylen << E1000_ADVTXD_PAYLEN_SHIFT;
4140
4141 /* 82575 requires a unique index per ring if any offload is enabled */
4142 if ((tx_flags & (IGB_TX_FLAGS_CSUM | IGB_TX_FLAGS_VLAN)) &&
4143 test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
4144 olinfo_status |= tx_ring->reg_idx << 4;
4145
4146 /* insert L4 checksum */
4147 if (tx_flags & IGB_TX_FLAGS_CSUM) {
4148 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
4149
4150 /* insert IPv4 checksum */
4151 if (tx_flags & IGB_TX_FLAGS_IPV4)
4152 olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
4153 }
4154
4155 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
4156}
4157
4158/*
4159 * The largest size we can write to the descriptor is 65535. In order to
4160 * maintain a power of two alignment we have to limit ourselves to 32K.
4161 */
4162#define IGB_MAX_TXD_PWR 15
4163#define IGB_MAX_DATA_PER_TXD (1<<IGB_MAX_TXD_PWR)
4164
4165static void igb_tx_map(struct igb_ring *tx_ring,
4166 struct igb_tx_buffer *first,
4167 const u8 hdr_len)
4168{
4169 struct sk_buff *skb = first->skb;
4170 struct igb_tx_buffer *tx_buffer_info;
4171 union e1000_adv_tx_desc *tx_desc;
4172 dma_addr_t dma;
4173 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
4174 unsigned int data_len = skb->data_len;
4175 unsigned int size = skb_headlen(skb);
4176 unsigned int paylen = skb->len - hdr_len;
4177 __le32 cmd_type;
4178 u32 tx_flags = first->tx_flags;
4179 u16 i = tx_ring->next_to_use;
4180
4181 tx_desc = IGB_TX_DESC(tx_ring, i);
4182
4183 igb_tx_olinfo_status(tx_ring, tx_desc, tx_flags, paylen);
4184 cmd_type = igb_tx_cmd_type(tx_flags);
4185
4186 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
4187 if (dma_mapping_error(tx_ring->dev, dma))
4188 goto dma_error;
4189
4190 /* record length, and DMA address */
4191 first->length = size;
4192 first->dma = dma;
4193 tx_desc->read.buffer_addr = cpu_to_le64(dma);
4194
4195 for (;;) {
4196 while (unlikely(size > IGB_MAX_DATA_PER_TXD)) {
4197 tx_desc->read.cmd_type_len =
4198 cmd_type | cpu_to_le32(IGB_MAX_DATA_PER_TXD);
4199
4200 i++;
4201 tx_desc++;
4202 if (i == tx_ring->count) {
4203 tx_desc = IGB_TX_DESC(tx_ring, 0);
4204 i = 0;
4205 }
4206
4207 dma += IGB_MAX_DATA_PER_TXD;
4208 size -= IGB_MAX_DATA_PER_TXD;
4209
4210 tx_desc->read.olinfo_status = 0;
4211 tx_desc->read.buffer_addr = cpu_to_le64(dma);
4212 }
4213
4214 if (likely(!data_len))
4215 break;
4216
4217 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
4218
4219 i++;
4220 tx_desc++;
4221 if (i == tx_ring->count) {
4222 tx_desc = IGB_TX_DESC(tx_ring, 0);
4223 i = 0;
4224 }
4225
4226 size = skb_frag_size(frag);
4227 data_len -= size;
4228
4229 dma = skb_frag_dma_map(tx_ring->dev, frag, 0,
4230 size, DMA_TO_DEVICE);
4231 if (dma_mapping_error(tx_ring->dev, dma))
4232 goto dma_error;
4233
4234 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4235 tx_buffer_info->length = size;
4236 tx_buffer_info->dma = dma;
4237
4238 tx_desc->read.olinfo_status = 0;
4239 tx_desc->read.buffer_addr = cpu_to_le64(dma);
4240
4241 frag++;
4242 }
4243
4244 /* write last descriptor with RS and EOP bits */
4245 cmd_type |= cpu_to_le32(size) | cpu_to_le32(IGB_TXD_DCMD);
4246 tx_desc->read.cmd_type_len = cmd_type;
4247
4248 /* set the timestamp */
4249 first->time_stamp = jiffies;
4250
4251 /*
4252 * Force memory writes to complete before letting h/w know there
4253 * are new descriptors to fetch. (Only applicable for weak-ordered
4254 * memory model archs, such as IA-64).
4255 *
4256 * We also need this memory barrier to make certain all of the
4257 * status bits have been updated before next_to_watch is written.
4258 */
4259 wmb();
4260
4261 /* set next_to_watch value indicating a packet is present */
4262 first->next_to_watch = tx_desc;
4263
4264 i++;
4265 if (i == tx_ring->count)
4266 i = 0;
4267
4268 tx_ring->next_to_use = i;
4269
4270 writel(i, tx_ring->tail);
4271
4272 /* we need this if more than one processor can write to our tail
4273 * at a time, it syncronizes IO on IA64/Altix systems */
4274 mmiowb();
4275
4276 return;
4277
4278dma_error:
4279 dev_err(tx_ring->dev, "TX DMA map failed\n");
4280
4281 /* clear dma mappings for failed tx_buffer_info map */
4282 for (;;) {
4283 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4284 igb_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
4285 if (tx_buffer_info == first)
4286 break;
4287 if (i == 0)
4288 i = tx_ring->count;
4289 i--;
4290 }
4291
4292 tx_ring->next_to_use = i;
4293}
4294
4295static int __igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
4296{
4297 struct net_device *netdev = tx_ring->netdev;
4298
4299 netif_stop_subqueue(netdev, tx_ring->queue_index);
4300
4301 /* Herbert's original patch had:
4302 * smp_mb__after_netif_stop_queue();
4303 * but since that doesn't exist yet, just open code it. */
4304 smp_mb();
4305
4306 /* We need to check again in a case another CPU has just
4307 * made room available. */
4308 if (igb_desc_unused(tx_ring) < size)
4309 return -EBUSY;
4310
4311 /* A reprieve! */
4312 netif_wake_subqueue(netdev, tx_ring->queue_index);
4313
4314 u64_stats_update_begin(&tx_ring->tx_syncp2);
4315 tx_ring->tx_stats.restart_queue2++;
4316 u64_stats_update_end(&tx_ring->tx_syncp2);
4317
4318 return 0;
4319}
4320
4321static inline int igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
4322{
4323 if (igb_desc_unused(tx_ring) >= size)
4324 return 0;
4325 return __igb_maybe_stop_tx(tx_ring, size);
4326}
4327
4328netdev_tx_t igb_xmit_frame_ring(struct sk_buff *skb,
4329 struct igb_ring *tx_ring)
4330{
4331 struct igb_tx_buffer *first;
4332 int tso;
4333 u32 tx_flags = 0;
4334 __be16 protocol = vlan_get_protocol(skb);
4335 u8 hdr_len = 0;
4336
4337 /* need: 1 descriptor per page,
4338 * + 2 desc gap to keep tail from touching head,
4339 * + 1 desc for skb->data,
4340 * + 1 desc for context descriptor,
4341 * otherwise try next time */
4342 if (igb_maybe_stop_tx(tx_ring, skb_shinfo(skb)->nr_frags + 4)) {
4343 /* this is a hard error */
4344 return NETDEV_TX_BUSY;
4345 }
4346
4347 /* record the location of the first descriptor for this packet */
4348 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
4349 first->skb = skb;
4350 first->bytecount = skb->len;
4351 first->gso_segs = 1;
4352
4353 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
4354 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
4355 tx_flags |= IGB_TX_FLAGS_TSTAMP;
4356 }
4357
4358 if (vlan_tx_tag_present(skb)) {
4359 tx_flags |= IGB_TX_FLAGS_VLAN;
4360 tx_flags |= (vlan_tx_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
4361 }
4362
4363 /* record initial flags and protocol */
4364 first->tx_flags = tx_flags;
4365 first->protocol = protocol;
4366
4367 tso = igb_tso(tx_ring, first, &hdr_len);
4368 if (tso < 0)
4369 goto out_drop;
4370 else if (!tso)
4371 igb_tx_csum(tx_ring, first);
4372
4373 igb_tx_map(tx_ring, first, hdr_len);
4374
4375 /* Make sure there is space in the ring for the next send. */
4376 igb_maybe_stop_tx(tx_ring, MAX_SKB_FRAGS + 4);
4377
4378 return NETDEV_TX_OK;
4379
4380out_drop:
4381 igb_unmap_and_free_tx_resource(tx_ring, first);
4382
4383 return NETDEV_TX_OK;
4384}
4385
4386static inline struct igb_ring *igb_tx_queue_mapping(struct igb_adapter *adapter,
4387 struct sk_buff *skb)
4388{
4389 unsigned int r_idx = skb->queue_mapping;
4390
4391 if (r_idx >= adapter->num_tx_queues)
4392 r_idx = r_idx % adapter->num_tx_queues;
4393
4394 return adapter->tx_ring[r_idx];
4395}
4396
4397static netdev_tx_t igb_xmit_frame(struct sk_buff *skb,
4398 struct net_device *netdev)
4399{
4400 struct igb_adapter *adapter = netdev_priv(netdev);
4401
4402 if (test_bit(__IGB_DOWN, &adapter->state)) {
4403 dev_kfree_skb_any(skb);
4404 return NETDEV_TX_OK;
4405 }
4406
4407 if (skb->len <= 0) {
4408 dev_kfree_skb_any(skb);
4409 return NETDEV_TX_OK;
4410 }
4411
4412 /*
4413 * The minimum packet size with TCTL.PSP set is 17 so pad the skb
4414 * in order to meet this minimum size requirement.
4415 */
4416 if (skb->len < 17) {
4417 if (skb_padto(skb, 17))
4418 return NETDEV_TX_OK;
4419 skb->len = 17;
4420 }
4421
4422 return igb_xmit_frame_ring(skb, igb_tx_queue_mapping(adapter, skb));
4423}
4424
4425/**
4426 * igb_tx_timeout - Respond to a Tx Hang
4427 * @netdev: network interface device structure
4428 **/
4429static void igb_tx_timeout(struct net_device *netdev)
4430{
4431 struct igb_adapter *adapter = netdev_priv(netdev);
4432 struct e1000_hw *hw = &adapter->hw;
4433
4434 /* Do the reset outside of interrupt context */
4435 adapter->tx_timeout_count++;
4436
4437 if (hw->mac.type >= e1000_82580)
4438 hw->dev_spec._82575.global_device_reset = true;
4439
4440 schedule_work(&adapter->reset_task);
4441 wr32(E1000_EICS,
4442 (adapter->eims_enable_mask & ~adapter->eims_other));
4443}
4444
4445static void igb_reset_task(struct work_struct *work)
4446{
4447 struct igb_adapter *adapter;
4448 adapter = container_of(work, struct igb_adapter, reset_task);
4449
4450 igb_dump(adapter);
4451 netdev_err(adapter->netdev, "Reset adapter\n");
4452 igb_reinit_locked(adapter);
4453}
4454
4455/**
4456 * igb_get_stats64 - Get System Network Statistics
4457 * @netdev: network interface device structure
4458 * @stats: rtnl_link_stats64 pointer
4459 *
4460 **/
4461static struct rtnl_link_stats64 *igb_get_stats64(struct net_device *netdev,
4462 struct rtnl_link_stats64 *stats)
4463{
4464 struct igb_adapter *adapter = netdev_priv(netdev);
4465
4466 spin_lock(&adapter->stats64_lock);
4467 igb_update_stats(adapter, &adapter->stats64);
4468 memcpy(stats, &adapter->stats64, sizeof(*stats));
4469 spin_unlock(&adapter->stats64_lock);
4470
4471 return stats;
4472}
4473
4474/**
4475 * igb_change_mtu - Change the Maximum Transfer Unit
4476 * @netdev: network interface device structure
4477 * @new_mtu: new value for maximum frame size
4478 *
4479 * Returns 0 on success, negative on failure
4480 **/
4481static int igb_change_mtu(struct net_device *netdev, int new_mtu)
4482{
4483 struct igb_adapter *adapter = netdev_priv(netdev);
4484 struct pci_dev *pdev = adapter->pdev;
4485 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
4486
4487 if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
4488 dev_err(&pdev->dev, "Invalid MTU setting\n");
4489 return -EINVAL;
4490 }
4491
4492#define MAX_STD_JUMBO_FRAME_SIZE 9238
4493 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
4494 dev_err(&pdev->dev, "MTU > 9216 not supported.\n");
4495 return -EINVAL;
4496 }
4497
4498 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
4499 msleep(1);
4500
4501 /* igb_down has a dependency on max_frame_size */
4502 adapter->max_frame_size = max_frame;
4503
4504 if (netif_running(netdev))
4505 igb_down(adapter);
4506
4507 dev_info(&pdev->dev, "changing MTU from %d to %d\n",
4508 netdev->mtu, new_mtu);
4509 netdev->mtu = new_mtu;
4510
4511 if (netif_running(netdev))
4512 igb_up(adapter);
4513 else
4514 igb_reset(adapter);
4515
4516 clear_bit(__IGB_RESETTING, &adapter->state);
4517
4518 return 0;
4519}
4520
4521/**
4522 * igb_update_stats - Update the board statistics counters
4523 * @adapter: board private structure
4524 **/
4525
4526void igb_update_stats(struct igb_adapter *adapter,
4527 struct rtnl_link_stats64 *net_stats)
4528{
4529 struct e1000_hw *hw = &adapter->hw;
4530 struct pci_dev *pdev = adapter->pdev;
4531 u32 reg, mpc;
4532 u16 phy_tmp;
4533 int i;
4534 u64 bytes, packets;
4535 unsigned int start;
4536 u64 _bytes, _packets;
4537
4538#define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
4539
4540 /*
4541 * Prevent stats update while adapter is being reset, or if the pci
4542 * connection is down.
4543 */
4544 if (adapter->link_speed == 0)
4545 return;
4546 if (pci_channel_offline(pdev))
4547 return;
4548
4549 bytes = 0;
4550 packets = 0;
4551 for (i = 0; i < adapter->num_rx_queues; i++) {
4552 u32 rqdpc_tmp = rd32(E1000_RQDPC(i)) & 0x0FFF;
4553 struct igb_ring *ring = adapter->rx_ring[i];
4554
4555 ring->rx_stats.drops += rqdpc_tmp;
4556 net_stats->rx_fifo_errors += rqdpc_tmp;
4557
4558 do {
4559 start = u64_stats_fetch_begin_bh(&ring->rx_syncp);
4560 _bytes = ring->rx_stats.bytes;
4561 _packets = ring->rx_stats.packets;
4562 } while (u64_stats_fetch_retry_bh(&ring->rx_syncp, start));
4563 bytes += _bytes;
4564 packets += _packets;
4565 }
4566
4567 net_stats->rx_bytes = bytes;
4568 net_stats->rx_packets = packets;
4569
4570 bytes = 0;
4571 packets = 0;
4572 for (i = 0; i < adapter->num_tx_queues; i++) {
4573 struct igb_ring *ring = adapter->tx_ring[i];
4574 do {
4575 start = u64_stats_fetch_begin_bh(&ring->tx_syncp);
4576 _bytes = ring->tx_stats.bytes;
4577 _packets = ring->tx_stats.packets;
4578 } while (u64_stats_fetch_retry_bh(&ring->tx_syncp, start));
4579 bytes += _bytes;
4580 packets += _packets;
4581 }
4582 net_stats->tx_bytes = bytes;
4583 net_stats->tx_packets = packets;
4584
4585 /* read stats registers */
4586 adapter->stats.crcerrs += rd32(E1000_CRCERRS);
4587 adapter->stats.gprc += rd32(E1000_GPRC);
4588 adapter->stats.gorc += rd32(E1000_GORCL);
4589 rd32(E1000_GORCH); /* clear GORCL */
4590 adapter->stats.bprc += rd32(E1000_BPRC);
4591 adapter->stats.mprc += rd32(E1000_MPRC);
4592 adapter->stats.roc += rd32(E1000_ROC);
4593
4594 adapter->stats.prc64 += rd32(E1000_PRC64);
4595 adapter->stats.prc127 += rd32(E1000_PRC127);
4596 adapter->stats.prc255 += rd32(E1000_PRC255);
4597 adapter->stats.prc511 += rd32(E1000_PRC511);
4598 adapter->stats.prc1023 += rd32(E1000_PRC1023);
4599 adapter->stats.prc1522 += rd32(E1000_PRC1522);
4600 adapter->stats.symerrs += rd32(E1000_SYMERRS);
4601 adapter->stats.sec += rd32(E1000_SEC);
4602
4603 mpc = rd32(E1000_MPC);
4604 adapter->stats.mpc += mpc;
4605 net_stats->rx_fifo_errors += mpc;
4606 adapter->stats.scc += rd32(E1000_SCC);
4607 adapter->stats.ecol += rd32(E1000_ECOL);
4608 adapter->stats.mcc += rd32(E1000_MCC);
4609 adapter->stats.latecol += rd32(E1000_LATECOL);
4610 adapter->stats.dc += rd32(E1000_DC);
4611 adapter->stats.rlec += rd32(E1000_RLEC);
4612 adapter->stats.xonrxc += rd32(E1000_XONRXC);
4613 adapter->stats.xontxc += rd32(E1000_XONTXC);
4614 adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
4615 adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
4616 adapter->stats.fcruc += rd32(E1000_FCRUC);
4617 adapter->stats.gptc += rd32(E1000_GPTC);
4618 adapter->stats.gotc += rd32(E1000_GOTCL);
4619 rd32(E1000_GOTCH); /* clear GOTCL */
4620 adapter->stats.rnbc += rd32(E1000_RNBC);
4621 adapter->stats.ruc += rd32(E1000_RUC);
4622 adapter->stats.rfc += rd32(E1000_RFC);
4623 adapter->stats.rjc += rd32(E1000_RJC);
4624 adapter->stats.tor += rd32(E1000_TORH);
4625 adapter->stats.tot += rd32(E1000_TOTH);
4626 adapter->stats.tpr += rd32(E1000_TPR);
4627
4628 adapter->stats.ptc64 += rd32(E1000_PTC64);
4629 adapter->stats.ptc127 += rd32(E1000_PTC127);
4630 adapter->stats.ptc255 += rd32(E1000_PTC255);
4631 adapter->stats.ptc511 += rd32(E1000_PTC511);
4632 adapter->stats.ptc1023 += rd32(E1000_PTC1023);
4633 adapter->stats.ptc1522 += rd32(E1000_PTC1522);
4634
4635 adapter->stats.mptc += rd32(E1000_MPTC);
4636 adapter->stats.bptc += rd32(E1000_BPTC);
4637
4638 adapter->stats.tpt += rd32(E1000_TPT);
4639 adapter->stats.colc += rd32(E1000_COLC);
4640
4641 adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
4642 /* read internal phy specific stats */
4643 reg = rd32(E1000_CTRL_EXT);
4644 if (!(reg & E1000_CTRL_EXT_LINK_MODE_MASK)) {
4645 adapter->stats.rxerrc += rd32(E1000_RXERRC);
4646 adapter->stats.tncrs += rd32(E1000_TNCRS);
4647 }
4648
4649 adapter->stats.tsctc += rd32(E1000_TSCTC);
4650 adapter->stats.tsctfc += rd32(E1000_TSCTFC);
4651
4652 adapter->stats.iac += rd32(E1000_IAC);
4653 adapter->stats.icrxoc += rd32(E1000_ICRXOC);
4654 adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
4655 adapter->stats.icrxatc += rd32(E1000_ICRXATC);
4656 adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
4657 adapter->stats.ictxatc += rd32(E1000_ICTXATC);
4658 adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
4659 adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
4660 adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
4661
4662 /* Fill out the OS statistics structure */
4663 net_stats->multicast = adapter->stats.mprc;
4664 net_stats->collisions = adapter->stats.colc;
4665
4666 /* Rx Errors */
4667
4668 /* RLEC on some newer hardware can be incorrect so build
4669 * our own version based on RUC and ROC */
4670 net_stats->rx_errors = adapter->stats.rxerrc +
4671 adapter->stats.crcerrs + adapter->stats.algnerrc +
4672 adapter->stats.ruc + adapter->stats.roc +
4673 adapter->stats.cexterr;
4674 net_stats->rx_length_errors = adapter->stats.ruc +
4675 adapter->stats.roc;
4676 net_stats->rx_crc_errors = adapter->stats.crcerrs;
4677 net_stats->rx_frame_errors = adapter->stats.algnerrc;
4678 net_stats->rx_missed_errors = adapter->stats.mpc;
4679
4680 /* Tx Errors */
4681 net_stats->tx_errors = adapter->stats.ecol +
4682 adapter->stats.latecol;
4683 net_stats->tx_aborted_errors = adapter->stats.ecol;
4684 net_stats->tx_window_errors = adapter->stats.latecol;
4685 net_stats->tx_carrier_errors = adapter->stats.tncrs;
4686
4687 /* Tx Dropped needs to be maintained elsewhere */
4688
4689 /* Phy Stats */
4690 if (hw->phy.media_type == e1000_media_type_copper) {
4691 if ((adapter->link_speed == SPEED_1000) &&
4692 (!igb_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
4693 phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
4694 adapter->phy_stats.idle_errors += phy_tmp;
4695 }
4696 }
4697
4698 /* Management Stats */
4699 adapter->stats.mgptc += rd32(E1000_MGTPTC);
4700 adapter->stats.mgprc += rd32(E1000_MGTPRC);
4701 adapter->stats.mgpdc += rd32(E1000_MGTPDC);
4702
4703 /* OS2BMC Stats */
4704 reg = rd32(E1000_MANC);
4705 if (reg & E1000_MANC_EN_BMC2OS) {
4706 adapter->stats.o2bgptc += rd32(E1000_O2BGPTC);
4707 adapter->stats.o2bspc += rd32(E1000_O2BSPC);
4708 adapter->stats.b2ospc += rd32(E1000_B2OSPC);
4709 adapter->stats.b2ogprc += rd32(E1000_B2OGPRC);
4710 }
4711}
4712
4713static irqreturn_t igb_msix_other(int irq, void *data)
4714{
4715 struct igb_adapter *adapter = data;
4716 struct e1000_hw *hw = &adapter->hw;
4717 u32 icr = rd32(E1000_ICR);
4718 /* reading ICR causes bit 31 of EICR to be cleared */
4719
4720 if (icr & E1000_ICR_DRSTA)
4721 schedule_work(&adapter->reset_task);
4722
4723 if (icr & E1000_ICR_DOUTSYNC) {
4724 /* HW is reporting DMA is out of sync */
4725 adapter->stats.doosync++;
4726 /* The DMA Out of Sync is also indication of a spoof event
4727 * in IOV mode. Check the Wrong VM Behavior register to
4728 * see if it is really a spoof event. */
4729 igb_check_wvbr(adapter);
4730 }
4731
4732 /* Check for a mailbox event */
4733 if (icr & E1000_ICR_VMMB)
4734 igb_msg_task(adapter);
4735
4736 if (icr & E1000_ICR_LSC) {
4737 hw->mac.get_link_status = 1;
4738 /* guard against interrupt when we're going down */
4739 if (!test_bit(__IGB_DOWN, &adapter->state))
4740 mod_timer(&adapter->watchdog_timer, jiffies + 1);
4741 }
4742
4743 wr32(E1000_EIMS, adapter->eims_other);
4744
4745 return IRQ_HANDLED;
4746}
4747
4748static void igb_write_itr(struct igb_q_vector *q_vector)
4749{
4750 struct igb_adapter *adapter = q_vector->adapter;
4751 u32 itr_val = q_vector->itr_val & 0x7FFC;
4752
4753 if (!q_vector->set_itr)
4754 return;
4755
4756 if (!itr_val)
4757 itr_val = 0x4;
4758
4759 if (adapter->hw.mac.type == e1000_82575)
4760 itr_val |= itr_val << 16;
4761 else
4762 itr_val |= E1000_EITR_CNT_IGNR;
4763
4764 writel(itr_val, q_vector->itr_register);
4765 q_vector->set_itr = 0;
4766}
4767
4768static irqreturn_t igb_msix_ring(int irq, void *data)
4769{
4770 struct igb_q_vector *q_vector = data;
4771
4772 /* Write the ITR value calculated from the previous interrupt. */
4773 igb_write_itr(q_vector);
4774
4775 napi_schedule(&q_vector->napi);
4776
4777 return IRQ_HANDLED;
4778}
4779
4780#ifdef CONFIG_IGB_DCA
4781static void igb_update_dca(struct igb_q_vector *q_vector)
4782{
4783 struct igb_adapter *adapter = q_vector->adapter;
4784 struct e1000_hw *hw = &adapter->hw;
4785 int cpu = get_cpu();
4786
4787 if (q_vector->cpu == cpu)
4788 goto out_no_update;
4789
4790 if (q_vector->tx.ring) {
4791 int q = q_vector->tx.ring->reg_idx;
4792 u32 dca_txctrl = rd32(E1000_DCA_TXCTRL(q));
4793 if (hw->mac.type == e1000_82575) {
4794 dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK;
4795 dca_txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
4796 } else {
4797 dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK_82576;
4798 dca_txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu) <<
4799 E1000_DCA_TXCTRL_CPUID_SHIFT;
4800 }
4801 dca_txctrl |= E1000_DCA_TXCTRL_DESC_DCA_EN;
4802 wr32(E1000_DCA_TXCTRL(q), dca_txctrl);
4803 }
4804 if (q_vector->rx.ring) {
4805 int q = q_vector->rx.ring->reg_idx;
4806 u32 dca_rxctrl = rd32(E1000_DCA_RXCTRL(q));
4807 if (hw->mac.type == e1000_82575) {
4808 dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK;
4809 dca_rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
4810 } else {
4811 dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK_82576;
4812 dca_rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu) <<
4813 E1000_DCA_RXCTRL_CPUID_SHIFT;
4814 }
4815 dca_rxctrl |= E1000_DCA_RXCTRL_DESC_DCA_EN;
4816 dca_rxctrl |= E1000_DCA_RXCTRL_HEAD_DCA_EN;
4817 dca_rxctrl |= E1000_DCA_RXCTRL_DATA_DCA_EN;
4818 wr32(E1000_DCA_RXCTRL(q), dca_rxctrl);
4819 }
4820 q_vector->cpu = cpu;
4821out_no_update:
4822 put_cpu();
4823}
4824
4825static void igb_setup_dca(struct igb_adapter *adapter)
4826{
4827 struct e1000_hw *hw = &adapter->hw;
4828 int i;
4829
4830 if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
4831 return;
4832
4833 /* Always use CB2 mode, difference is masked in the CB driver. */
4834 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
4835
4836 for (i = 0; i < adapter->num_q_vectors; i++) {
4837 adapter->q_vector[i]->cpu = -1;
4838 igb_update_dca(adapter->q_vector[i]);
4839 }
4840}
4841
4842static int __igb_notify_dca(struct device *dev, void *data)
4843{
4844 struct net_device *netdev = dev_get_drvdata(dev);
4845 struct igb_adapter *adapter = netdev_priv(netdev);
4846 struct pci_dev *pdev = adapter->pdev;
4847 struct e1000_hw *hw = &adapter->hw;
4848 unsigned long event = *(unsigned long *)data;
4849
4850 switch (event) {
4851 case DCA_PROVIDER_ADD:
4852 /* if already enabled, don't do it again */
4853 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
4854 break;
4855 if (dca_add_requester(dev) == 0) {
4856 adapter->flags |= IGB_FLAG_DCA_ENABLED;
4857 dev_info(&pdev->dev, "DCA enabled\n");
4858 igb_setup_dca(adapter);
4859 break;
4860 }
4861 /* Fall Through since DCA is disabled. */
4862 case DCA_PROVIDER_REMOVE:
4863 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
4864 /* without this a class_device is left
4865 * hanging around in the sysfs model */
4866 dca_remove_requester(dev);
4867 dev_info(&pdev->dev, "DCA disabled\n");
4868 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
4869 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
4870 }
4871 break;
4872 }
4873
4874 return 0;
4875}
4876
4877static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
4878 void *p)
4879{
4880 int ret_val;
4881
4882 ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
4883 __igb_notify_dca);
4884
4885 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
4886}
4887#endif /* CONFIG_IGB_DCA */
4888
4889#ifdef CONFIG_PCI_IOV
4890static int igb_vf_configure(struct igb_adapter *adapter, int vf)
4891{
4892 unsigned char mac_addr[ETH_ALEN];
4893 struct pci_dev *pdev = adapter->pdev;
4894 struct e1000_hw *hw = &adapter->hw;
4895 struct pci_dev *pvfdev;
4896 unsigned int device_id;
4897 u16 thisvf_devfn;
4898
4899 random_ether_addr(mac_addr);
4900 igb_set_vf_mac(adapter, vf, mac_addr);
4901
4902 switch (adapter->hw.mac.type) {
4903 case e1000_82576:
4904 device_id = IGB_82576_VF_DEV_ID;
4905 /* VF Stride for 82576 is 2 */
4906 thisvf_devfn = (pdev->devfn + 0x80 + (vf << 1)) |
4907 (pdev->devfn & 1);
4908 break;
4909 case e1000_i350:
4910 device_id = IGB_I350_VF_DEV_ID;
4911 /* VF Stride for I350 is 4 */
4912 thisvf_devfn = (pdev->devfn + 0x80 + (vf << 2)) |
4913 (pdev->devfn & 3);
4914 break;
4915 default:
4916 device_id = 0;
4917 thisvf_devfn = 0;
4918 break;
4919 }
4920
4921 pvfdev = pci_get_device(hw->vendor_id, device_id, NULL);
4922 while (pvfdev) {
4923 if (pvfdev->devfn == thisvf_devfn)
4924 break;
4925 pvfdev = pci_get_device(hw->vendor_id,
4926 device_id, pvfdev);
4927 }
4928
4929 if (pvfdev)
4930 adapter->vf_data[vf].vfdev = pvfdev;
4931 else
4932 dev_err(&pdev->dev,
4933 "Couldn't find pci dev ptr for VF %4.4x\n",
4934 thisvf_devfn);
4935 return pvfdev != NULL;
4936}
4937
4938static int igb_find_enabled_vfs(struct igb_adapter *adapter)
4939{
4940 struct e1000_hw *hw = &adapter->hw;
4941 struct pci_dev *pdev = adapter->pdev;
4942 struct pci_dev *pvfdev;
4943 u16 vf_devfn = 0;
4944 u16 vf_stride;
4945 unsigned int device_id;
4946 int vfs_found = 0;
4947
4948 switch (adapter->hw.mac.type) {
4949 case e1000_82576:
4950 device_id = IGB_82576_VF_DEV_ID;
4951 /* VF Stride for 82576 is 2 */
4952 vf_stride = 2;
4953 break;
4954 case e1000_i350:
4955 device_id = IGB_I350_VF_DEV_ID;
4956 /* VF Stride for I350 is 4 */
4957 vf_stride = 4;
4958 break;
4959 default:
4960 device_id = 0;
4961 vf_stride = 0;
4962 break;
4963 }
4964
4965 vf_devfn = pdev->devfn + 0x80;
4966 pvfdev = pci_get_device(hw->vendor_id, device_id, NULL);
4967 while (pvfdev) {
4968 if (pvfdev->devfn == vf_devfn)
4969 vfs_found++;
4970 vf_devfn += vf_stride;
4971 pvfdev = pci_get_device(hw->vendor_id,
4972 device_id, pvfdev);
4973 }
4974
4975 return vfs_found;
4976}
4977
4978static int igb_check_vf_assignment(struct igb_adapter *adapter)
4979{
4980 int i;
4981 for (i = 0; i < adapter->vfs_allocated_count; i++) {
4982 if (adapter->vf_data[i].vfdev) {
4983 if (adapter->vf_data[i].vfdev->dev_flags &
4984 PCI_DEV_FLAGS_ASSIGNED)
4985 return true;
4986 }
4987 }
4988 return false;
4989}
4990
4991#endif
4992static void igb_ping_all_vfs(struct igb_adapter *adapter)
4993{
4994 struct e1000_hw *hw = &adapter->hw;
4995 u32 ping;
4996 int i;
4997
4998 for (i = 0 ; i < adapter->vfs_allocated_count; i++) {
4999 ping = E1000_PF_CONTROL_MSG;
5000 if (adapter->vf_data[i].flags & IGB_VF_FLAG_CTS)
5001 ping |= E1000_VT_MSGTYPE_CTS;
5002 igb_write_mbx(hw, &ping, 1, i);
5003 }
5004}
5005
5006static int igb_set_vf_promisc(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
5007{
5008 struct e1000_hw *hw = &adapter->hw;
5009 u32 vmolr = rd32(E1000_VMOLR(vf));
5010 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5011
5012 vf_data->flags &= ~(IGB_VF_FLAG_UNI_PROMISC |
5013 IGB_VF_FLAG_MULTI_PROMISC);
5014 vmolr &= ~(E1000_VMOLR_ROPE | E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
5015
5016 if (*msgbuf & E1000_VF_SET_PROMISC_MULTICAST) {
5017 vmolr |= E1000_VMOLR_MPME;
5018 vf_data->flags |= IGB_VF_FLAG_MULTI_PROMISC;
5019 *msgbuf &= ~E1000_VF_SET_PROMISC_MULTICAST;
5020 } else {
5021 /*
5022 * if we have hashes and we are clearing a multicast promisc
5023 * flag we need to write the hashes to the MTA as this step
5024 * was previously skipped
5025 */
5026 if (vf_data->num_vf_mc_hashes > 30) {
5027 vmolr |= E1000_VMOLR_MPME;
5028 } else if (vf_data->num_vf_mc_hashes) {
5029 int j;
5030 vmolr |= E1000_VMOLR_ROMPE;
5031 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
5032 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
5033 }
5034 }
5035
5036 wr32(E1000_VMOLR(vf), vmolr);
5037
5038 /* there are flags left unprocessed, likely not supported */
5039 if (*msgbuf & E1000_VT_MSGINFO_MASK)
5040 return -EINVAL;
5041
5042 return 0;
5043
5044}
5045
5046static int igb_set_vf_multicasts(struct igb_adapter *adapter,
5047 u32 *msgbuf, u32 vf)
5048{
5049 int n = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
5050 u16 *hash_list = (u16 *)&msgbuf[1];
5051 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5052 int i;
5053
5054 /* salt away the number of multicast addresses assigned
5055 * to this VF for later use to restore when the PF multi cast
5056 * list changes
5057 */
5058 vf_data->num_vf_mc_hashes = n;
5059
5060 /* only up to 30 hash values supported */
5061 if (n > 30)
5062 n = 30;
5063
5064 /* store the hashes for later use */
5065 for (i = 0; i < n; i++)
5066 vf_data->vf_mc_hashes[i] = hash_list[i];
5067
5068 /* Flush and reset the mta with the new values */
5069 igb_set_rx_mode(adapter->netdev);
5070
5071 return 0;
5072}
5073
5074static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
5075{
5076 struct e1000_hw *hw = &adapter->hw;
5077 struct vf_data_storage *vf_data;
5078 int i, j;
5079
5080 for (i = 0; i < adapter->vfs_allocated_count; i++) {
5081 u32 vmolr = rd32(E1000_VMOLR(i));
5082 vmolr &= ~(E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
5083
5084 vf_data = &adapter->vf_data[i];
5085
5086 if ((vf_data->num_vf_mc_hashes > 30) ||
5087 (vf_data->flags & IGB_VF_FLAG_MULTI_PROMISC)) {
5088 vmolr |= E1000_VMOLR_MPME;
5089 } else if (vf_data->num_vf_mc_hashes) {
5090 vmolr |= E1000_VMOLR_ROMPE;
5091 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
5092 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
5093 }
5094 wr32(E1000_VMOLR(i), vmolr);
5095 }
5096}
5097
5098static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
5099{
5100 struct e1000_hw *hw = &adapter->hw;
5101 u32 pool_mask, reg, vid;
5102 int i;
5103
5104 pool_mask = 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
5105
5106 /* Find the vlan filter for this id */
5107 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5108 reg = rd32(E1000_VLVF(i));
5109
5110 /* remove the vf from the pool */
5111 reg &= ~pool_mask;
5112
5113 /* if pool is empty then remove entry from vfta */
5114 if (!(reg & E1000_VLVF_POOLSEL_MASK) &&
5115 (reg & E1000_VLVF_VLANID_ENABLE)) {
5116 reg = 0;
5117 vid = reg & E1000_VLVF_VLANID_MASK;
5118 igb_vfta_set(hw, vid, false);
5119 }
5120
5121 wr32(E1000_VLVF(i), reg);
5122 }
5123
5124 adapter->vf_data[vf].vlans_enabled = 0;
5125}
5126
5127static s32 igb_vlvf_set(struct igb_adapter *adapter, u32 vid, bool add, u32 vf)
5128{
5129 struct e1000_hw *hw = &adapter->hw;
5130 u32 reg, i;
5131
5132 /* The vlvf table only exists on 82576 hardware and newer */
5133 if (hw->mac.type < e1000_82576)
5134 return -1;
5135
5136 /* we only need to do this if VMDq is enabled */
5137 if (!adapter->vfs_allocated_count)
5138 return -1;
5139
5140 /* Find the vlan filter for this id */
5141 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5142 reg = rd32(E1000_VLVF(i));
5143 if ((reg & E1000_VLVF_VLANID_ENABLE) &&
5144 vid == (reg & E1000_VLVF_VLANID_MASK))
5145 break;
5146 }
5147
5148 if (add) {
5149 if (i == E1000_VLVF_ARRAY_SIZE) {
5150 /* Did not find a matching VLAN ID entry that was
5151 * enabled. Search for a free filter entry, i.e.
5152 * one without the enable bit set
5153 */
5154 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5155 reg = rd32(E1000_VLVF(i));
5156 if (!(reg & E1000_VLVF_VLANID_ENABLE))
5157 break;
5158 }
5159 }
5160 if (i < E1000_VLVF_ARRAY_SIZE) {
5161 /* Found an enabled/available entry */
5162 reg |= 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
5163
5164 /* if !enabled we need to set this up in vfta */
5165 if (!(reg & E1000_VLVF_VLANID_ENABLE)) {
5166 /* add VID to filter table */
5167 igb_vfta_set(hw, vid, true);
5168 reg |= E1000_VLVF_VLANID_ENABLE;
5169 }
5170 reg &= ~E1000_VLVF_VLANID_MASK;
5171 reg |= vid;
5172 wr32(E1000_VLVF(i), reg);
5173
5174 /* do not modify RLPML for PF devices */
5175 if (vf >= adapter->vfs_allocated_count)
5176 return 0;
5177
5178 if (!adapter->vf_data[vf].vlans_enabled) {
5179 u32 size;
5180 reg = rd32(E1000_VMOLR(vf));
5181 size = reg & E1000_VMOLR_RLPML_MASK;
5182 size += 4;
5183 reg &= ~E1000_VMOLR_RLPML_MASK;
5184 reg |= size;
5185 wr32(E1000_VMOLR(vf), reg);
5186 }
5187
5188 adapter->vf_data[vf].vlans_enabled++;
5189 }
5190 } else {
5191 if (i < E1000_VLVF_ARRAY_SIZE) {
5192 /* remove vf from the pool */
5193 reg &= ~(1 << (E1000_VLVF_POOLSEL_SHIFT + vf));
5194 /* if pool is empty then remove entry from vfta */
5195 if (!(reg & E1000_VLVF_POOLSEL_MASK)) {
5196 reg = 0;
5197 igb_vfta_set(hw, vid, false);
5198 }
5199 wr32(E1000_VLVF(i), reg);
5200
5201 /* do not modify RLPML for PF devices */
5202 if (vf >= adapter->vfs_allocated_count)
5203 return 0;
5204
5205 adapter->vf_data[vf].vlans_enabled--;
5206 if (!adapter->vf_data[vf].vlans_enabled) {
5207 u32 size;
5208 reg = rd32(E1000_VMOLR(vf));
5209 size = reg & E1000_VMOLR_RLPML_MASK;
5210 size -= 4;
5211 reg &= ~E1000_VMOLR_RLPML_MASK;
5212 reg |= size;
5213 wr32(E1000_VMOLR(vf), reg);
5214 }
5215 }
5216 }
5217 return 0;
5218}
5219
5220static void igb_set_vmvir(struct igb_adapter *adapter, u32 vid, u32 vf)
5221{
5222 struct e1000_hw *hw = &adapter->hw;
5223
5224 if (vid)
5225 wr32(E1000_VMVIR(vf), (vid | E1000_VMVIR_VLANA_DEFAULT));
5226 else
5227 wr32(E1000_VMVIR(vf), 0);
5228}
5229
5230static int igb_ndo_set_vf_vlan(struct net_device *netdev,
5231 int vf, u16 vlan, u8 qos)
5232{
5233 int err = 0;
5234 struct igb_adapter *adapter = netdev_priv(netdev);
5235
5236 if ((vf >= adapter->vfs_allocated_count) || (vlan > 4095) || (qos > 7))
5237 return -EINVAL;
5238 if (vlan || qos) {
5239 err = igb_vlvf_set(adapter, vlan, !!vlan, vf);
5240 if (err)
5241 goto out;
5242 igb_set_vmvir(adapter, vlan | (qos << VLAN_PRIO_SHIFT), vf);
5243 igb_set_vmolr(adapter, vf, !vlan);
5244 adapter->vf_data[vf].pf_vlan = vlan;
5245 adapter->vf_data[vf].pf_qos = qos;
5246 dev_info(&adapter->pdev->dev,
5247 "Setting VLAN %d, QOS 0x%x on VF %d\n", vlan, qos, vf);
5248 if (test_bit(__IGB_DOWN, &adapter->state)) {
5249 dev_warn(&adapter->pdev->dev,
5250 "The VF VLAN has been set,"
5251 " but the PF device is not up.\n");
5252 dev_warn(&adapter->pdev->dev,
5253 "Bring the PF device up before"
5254 " attempting to use the VF device.\n");
5255 }
5256 } else {
5257 igb_vlvf_set(adapter, adapter->vf_data[vf].pf_vlan,
5258 false, vf);
5259 igb_set_vmvir(adapter, vlan, vf);
5260 igb_set_vmolr(adapter, vf, true);
5261 adapter->vf_data[vf].pf_vlan = 0;
5262 adapter->vf_data[vf].pf_qos = 0;
5263 }
5264out:
5265 return err;
5266}
5267
5268static int igb_set_vf_vlan(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
5269{
5270 int add = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
5271 int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
5272
5273 return igb_vlvf_set(adapter, vid, add, vf);
5274}
5275
5276static inline void igb_vf_reset(struct igb_adapter *adapter, u32 vf)
5277{
5278 /* clear flags - except flag that indicates PF has set the MAC */
5279 adapter->vf_data[vf].flags &= IGB_VF_FLAG_PF_SET_MAC;
5280 adapter->vf_data[vf].last_nack = jiffies;
5281
5282 /* reset offloads to defaults */
5283 igb_set_vmolr(adapter, vf, true);
5284
5285 /* reset vlans for device */
5286 igb_clear_vf_vfta(adapter, vf);
5287 if (adapter->vf_data[vf].pf_vlan)
5288 igb_ndo_set_vf_vlan(adapter->netdev, vf,
5289 adapter->vf_data[vf].pf_vlan,
5290 adapter->vf_data[vf].pf_qos);
5291 else
5292 igb_clear_vf_vfta(adapter, vf);
5293
5294 /* reset multicast table array for vf */
5295 adapter->vf_data[vf].num_vf_mc_hashes = 0;
5296
5297 /* Flush and reset the mta with the new values */
5298 igb_set_rx_mode(adapter->netdev);
5299}
5300
5301static void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
5302{
5303 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
5304
5305 /* generate a new mac address as we were hotplug removed/added */
5306 if (!(adapter->vf_data[vf].flags & IGB_VF_FLAG_PF_SET_MAC))
5307 random_ether_addr(vf_mac);
5308
5309 /* process remaining reset events */
5310 igb_vf_reset(adapter, vf);
5311}
5312
5313static void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
5314{
5315 struct e1000_hw *hw = &adapter->hw;
5316 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
5317 int rar_entry = hw->mac.rar_entry_count - (vf + 1);
5318 u32 reg, msgbuf[3];
5319 u8 *addr = (u8 *)(&msgbuf[1]);
5320
5321 /* process all the same items cleared in a function level reset */
5322 igb_vf_reset(adapter, vf);
5323
5324 /* set vf mac address */
5325 igb_rar_set_qsel(adapter, vf_mac, rar_entry, vf);
5326
5327 /* enable transmit and receive for vf */
5328 reg = rd32(E1000_VFTE);
5329 wr32(E1000_VFTE, reg | (1 << vf));
5330 reg = rd32(E1000_VFRE);
5331 wr32(E1000_VFRE, reg | (1 << vf));
5332
5333 adapter->vf_data[vf].flags |= IGB_VF_FLAG_CTS;
5334
5335 /* reply to reset with ack and vf mac address */
5336 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
5337 memcpy(addr, vf_mac, 6);
5338 igb_write_mbx(hw, msgbuf, 3, vf);
5339}
5340
5341static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf)
5342{
5343 /*
5344 * The VF MAC Address is stored in a packed array of bytes
5345 * starting at the second 32 bit word of the msg array
5346 */
5347 unsigned char *addr = (char *)&msg[1];
5348 int err = -1;
5349
5350 if (is_valid_ether_addr(addr))
5351 err = igb_set_vf_mac(adapter, vf, addr);
5352
5353 return err;
5354}
5355
5356static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf)
5357{
5358 struct e1000_hw *hw = &adapter->hw;
5359 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5360 u32 msg = E1000_VT_MSGTYPE_NACK;
5361
5362 /* if device isn't clear to send it shouldn't be reading either */
5363 if (!(vf_data->flags & IGB_VF_FLAG_CTS) &&
5364 time_after(jiffies, vf_data->last_nack + (2 * HZ))) {
5365 igb_write_mbx(hw, &msg, 1, vf);
5366 vf_data->last_nack = jiffies;
5367 }
5368}
5369
5370static void igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
5371{
5372 struct pci_dev *pdev = adapter->pdev;
5373 u32 msgbuf[E1000_VFMAILBOX_SIZE];
5374 struct e1000_hw *hw = &adapter->hw;
5375 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5376 s32 retval;
5377
5378 retval = igb_read_mbx(hw, msgbuf, E1000_VFMAILBOX_SIZE, vf);
5379
5380 if (retval) {
5381 /* if receive failed revoke VF CTS stats and restart init */
5382 dev_err(&pdev->dev, "Error receiving message from VF\n");
5383 vf_data->flags &= ~IGB_VF_FLAG_CTS;
5384 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
5385 return;
5386 goto out;
5387 }
5388
5389 /* this is a message we already processed, do nothing */
5390 if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
5391 return;
5392
5393 /*
5394 * until the vf completes a reset it should not be
5395 * allowed to start any configuration.
5396 */
5397
5398 if (msgbuf[0] == E1000_VF_RESET) {
5399 igb_vf_reset_msg(adapter, vf);
5400 return;
5401 }
5402
5403 if (!(vf_data->flags & IGB_VF_FLAG_CTS)) {
5404 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
5405 return;
5406 retval = -1;
5407 goto out;
5408 }
5409
5410 switch ((msgbuf[0] & 0xFFFF)) {
5411 case E1000_VF_SET_MAC_ADDR:
5412 retval = -EINVAL;
5413 if (!(vf_data->flags & IGB_VF_FLAG_PF_SET_MAC))
5414 retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
5415 else
5416 dev_warn(&pdev->dev,
5417 "VF %d attempted to override administratively "
5418 "set MAC address\nReload the VF driver to "
5419 "resume operations\n", vf);
5420 break;
5421 case E1000_VF_SET_PROMISC:
5422 retval = igb_set_vf_promisc(adapter, msgbuf, vf);
5423 break;
5424 case E1000_VF_SET_MULTICAST:
5425 retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
5426 break;
5427 case E1000_VF_SET_LPE:
5428 retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf);
5429 break;
5430 case E1000_VF_SET_VLAN:
5431 retval = -1;
5432 if (vf_data->pf_vlan)
5433 dev_warn(&pdev->dev,
5434 "VF %d attempted to override administratively "
5435 "set VLAN tag\nReload the VF driver to "
5436 "resume operations\n", vf);
5437 else
5438 retval = igb_set_vf_vlan(adapter, msgbuf, vf);
5439 break;
5440 default:
5441 dev_err(&pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
5442 retval = -1;
5443 break;
5444 }
5445
5446 msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
5447out:
5448 /* notify the VF of the results of what it sent us */
5449 if (retval)
5450 msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
5451 else
5452 msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
5453
5454 igb_write_mbx(hw, msgbuf, 1, vf);
5455}
5456
5457static void igb_msg_task(struct igb_adapter *adapter)
5458{
5459 struct e1000_hw *hw = &adapter->hw;
5460 u32 vf;
5461
5462 for (vf = 0; vf < adapter->vfs_allocated_count; vf++) {
5463 /* process any reset requests */
5464 if (!igb_check_for_rst(hw, vf))
5465 igb_vf_reset_event(adapter, vf);
5466
5467 /* process any messages pending */
5468 if (!igb_check_for_msg(hw, vf))
5469 igb_rcv_msg_from_vf(adapter, vf);
5470
5471 /* process any acks */
5472 if (!igb_check_for_ack(hw, vf))
5473 igb_rcv_ack_from_vf(adapter, vf);
5474 }
5475}
5476
5477/**
5478 * igb_set_uta - Set unicast filter table address
5479 * @adapter: board private structure
5480 *
5481 * The unicast table address is a register array of 32-bit registers.
5482 * The table is meant to be used in a way similar to how the MTA is used
5483 * however due to certain limitations in the hardware it is necessary to
5484 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
5485 * enable bit to allow vlan tag stripping when promiscuous mode is enabled
5486 **/
5487static void igb_set_uta(struct igb_adapter *adapter)
5488{
5489 struct e1000_hw *hw = &adapter->hw;
5490 int i;
5491
5492 /* The UTA table only exists on 82576 hardware and newer */
5493 if (hw->mac.type < e1000_82576)
5494 return;
5495
5496 /* we only need to do this if VMDq is enabled */
5497 if (!adapter->vfs_allocated_count)
5498 return;
5499
5500 for (i = 0; i < hw->mac.uta_reg_count; i++)
5501 array_wr32(E1000_UTA, i, ~0);
5502}
5503
5504/**
5505 * igb_intr_msi - Interrupt Handler
5506 * @irq: interrupt number
5507 * @data: pointer to a network interface device structure
5508 **/
5509static irqreturn_t igb_intr_msi(int irq, void *data)
5510{
5511 struct igb_adapter *adapter = data;
5512 struct igb_q_vector *q_vector = adapter->q_vector[0];
5513 struct e1000_hw *hw = &adapter->hw;
5514 /* read ICR disables interrupts using IAM */
5515 u32 icr = rd32(E1000_ICR);
5516
5517 igb_write_itr(q_vector);
5518
5519 if (icr & E1000_ICR_DRSTA)
5520 schedule_work(&adapter->reset_task);
5521
5522 if (icr & E1000_ICR_DOUTSYNC) {
5523 /* HW is reporting DMA is out of sync */
5524 adapter->stats.doosync++;
5525 }
5526
5527 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
5528 hw->mac.get_link_status = 1;
5529 if (!test_bit(__IGB_DOWN, &adapter->state))
5530 mod_timer(&adapter->watchdog_timer, jiffies + 1);
5531 }
5532
5533 napi_schedule(&q_vector->napi);
5534
5535 return IRQ_HANDLED;
5536}
5537
5538/**
5539 * igb_intr - Legacy Interrupt Handler
5540 * @irq: interrupt number
5541 * @data: pointer to a network interface device structure
5542 **/
5543static irqreturn_t igb_intr(int irq, void *data)
5544{
5545 struct igb_adapter *adapter = data;
5546 struct igb_q_vector *q_vector = adapter->q_vector[0];
5547 struct e1000_hw *hw = &adapter->hw;
5548 /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked. No
5549 * need for the IMC write */
5550 u32 icr = rd32(E1000_ICR);
5551
5552 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
5553 * not set, then the adapter didn't send an interrupt */
5554 if (!(icr & E1000_ICR_INT_ASSERTED))
5555 return IRQ_NONE;
5556
5557 igb_write_itr(q_vector);
5558
5559 if (icr & E1000_ICR_DRSTA)
5560 schedule_work(&adapter->reset_task);
5561
5562 if (icr & E1000_ICR_DOUTSYNC) {
5563 /* HW is reporting DMA is out of sync */
5564 adapter->stats.doosync++;
5565 }
5566
5567 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
5568 hw->mac.get_link_status = 1;
5569 /* guard against interrupt when we're going down */
5570 if (!test_bit(__IGB_DOWN, &adapter->state))
5571 mod_timer(&adapter->watchdog_timer, jiffies + 1);
5572 }
5573
5574 napi_schedule(&q_vector->napi);
5575
5576 return IRQ_HANDLED;
5577}
5578
5579void igb_ring_irq_enable(struct igb_q_vector *q_vector)
5580{
5581 struct igb_adapter *adapter = q_vector->adapter;
5582 struct e1000_hw *hw = &adapter->hw;
5583
5584 if ((q_vector->rx.ring && (adapter->rx_itr_setting & 3)) ||
5585 (!q_vector->rx.ring && (adapter->tx_itr_setting & 3))) {
5586 if ((adapter->num_q_vectors == 1) && !adapter->vf_data)
5587 igb_set_itr(q_vector);
5588 else
5589 igb_update_ring_itr(q_vector);
5590 }
5591
5592 if (!test_bit(__IGB_DOWN, &adapter->state)) {
5593 if (adapter->msix_entries)
5594 wr32(E1000_EIMS, q_vector->eims_value);
5595 else
5596 igb_irq_enable(adapter);
5597 }
5598}
5599
5600/**
5601 * igb_poll - NAPI Rx polling callback
5602 * @napi: napi polling structure
5603 * @budget: count of how many packets we should handle
5604 **/
5605static int igb_poll(struct napi_struct *napi, int budget)
5606{
5607 struct igb_q_vector *q_vector = container_of(napi,
5608 struct igb_q_vector,
5609 napi);
5610 bool clean_complete = true;
5611
5612#ifdef CONFIG_IGB_DCA
5613 if (q_vector->adapter->flags & IGB_FLAG_DCA_ENABLED)
5614 igb_update_dca(q_vector);
5615#endif
5616 if (q_vector->tx.ring)
5617 clean_complete = igb_clean_tx_irq(q_vector);
5618
5619 if (q_vector->rx.ring)
5620 clean_complete &= igb_clean_rx_irq(q_vector, budget);
5621
5622 /* If all work not completed, return budget and keep polling */
5623 if (!clean_complete)
5624 return budget;
5625
5626 /* If not enough Rx work done, exit the polling mode */
5627 napi_complete(napi);
5628 igb_ring_irq_enable(q_vector);
5629
5630 return 0;
5631}
5632
5633/**
5634 * igb_systim_to_hwtstamp - convert system time value to hw timestamp
5635 * @adapter: board private structure
5636 * @shhwtstamps: timestamp structure to update
5637 * @regval: unsigned 64bit system time value.
5638 *
5639 * We need to convert the system time value stored in the RX/TXSTMP registers
5640 * into a hwtstamp which can be used by the upper level timestamping functions
5641 */
5642static void igb_systim_to_hwtstamp(struct igb_adapter *adapter,
5643 struct skb_shared_hwtstamps *shhwtstamps,
5644 u64 regval)
5645{
5646 u64 ns;
5647
5648 /*
5649 * The 82580 starts with 1ns at bit 0 in RX/TXSTMPL, shift this up to
5650 * 24 to match clock shift we setup earlier.
5651 */
5652 if (adapter->hw.mac.type >= e1000_82580)
5653 regval <<= IGB_82580_TSYNC_SHIFT;
5654
5655 ns = timecounter_cyc2time(&adapter->clock, regval);
5656 timecompare_update(&adapter->compare, ns);
5657 memset(shhwtstamps, 0, sizeof(struct skb_shared_hwtstamps));
5658 shhwtstamps->hwtstamp = ns_to_ktime(ns);
5659 shhwtstamps->syststamp = timecompare_transform(&adapter->compare, ns);
5660}
5661
5662/**
5663 * igb_tx_hwtstamp - utility function which checks for TX time stamp
5664 * @q_vector: pointer to q_vector containing needed info
5665 * @buffer: pointer to igb_tx_buffer structure
5666 *
5667 * If we were asked to do hardware stamping and such a time stamp is
5668 * available, then it must have been for this skb here because we only
5669 * allow only one such packet into the queue.
5670 */
5671static void igb_tx_hwtstamp(struct igb_q_vector *q_vector,
5672 struct igb_tx_buffer *buffer_info)
5673{
5674 struct igb_adapter *adapter = q_vector->adapter;
5675 struct e1000_hw *hw = &adapter->hw;
5676 struct skb_shared_hwtstamps shhwtstamps;
5677 u64 regval;
5678
5679 /* if skb does not support hw timestamp or TX stamp not valid exit */
5680 if (likely(!(buffer_info->tx_flags & IGB_TX_FLAGS_TSTAMP)) ||
5681 !(rd32(E1000_TSYNCTXCTL) & E1000_TSYNCTXCTL_VALID))
5682 return;
5683
5684 regval = rd32(E1000_TXSTMPL);
5685 regval |= (u64)rd32(E1000_TXSTMPH) << 32;
5686
5687 igb_systim_to_hwtstamp(adapter, &shhwtstamps, regval);
5688 skb_tstamp_tx(buffer_info->skb, &shhwtstamps);
5689}
5690
5691/**
5692 * igb_clean_tx_irq - Reclaim resources after transmit completes
5693 * @q_vector: pointer to q_vector containing needed info
5694 * returns true if ring is completely cleaned
5695 **/
5696static bool igb_clean_tx_irq(struct igb_q_vector *q_vector)
5697{
5698 struct igb_adapter *adapter = q_vector->adapter;
5699 struct igb_ring *tx_ring = q_vector->tx.ring;
5700 struct igb_tx_buffer *tx_buffer;
5701 union e1000_adv_tx_desc *tx_desc, *eop_desc;
5702 unsigned int total_bytes = 0, total_packets = 0;
5703 unsigned int budget = q_vector->tx.work_limit;
5704 unsigned int i = tx_ring->next_to_clean;
5705
5706 if (test_bit(__IGB_DOWN, &adapter->state))
5707 return true;
5708
5709 tx_buffer = &tx_ring->tx_buffer_info[i];
5710 tx_desc = IGB_TX_DESC(tx_ring, i);
5711 i -= tx_ring->count;
5712
5713 for (; budget; budget--) {
5714 eop_desc = tx_buffer->next_to_watch;
5715
5716 /* prevent any other reads prior to eop_desc */
5717 rmb();
5718
5719 /* if next_to_watch is not set then there is no work pending */
5720 if (!eop_desc)
5721 break;
5722
5723 /* if DD is not set pending work has not been completed */
5724 if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)))
5725 break;
5726
5727 /* clear next_to_watch to prevent false hangs */
5728 tx_buffer->next_to_watch = NULL;
5729
5730 /* update the statistics for this packet */
5731 total_bytes += tx_buffer->bytecount;
5732 total_packets += tx_buffer->gso_segs;
5733
5734 /* retrieve hardware timestamp */
5735 igb_tx_hwtstamp(q_vector, tx_buffer);
5736
5737 /* free the skb */
5738 dev_kfree_skb_any(tx_buffer->skb);
5739 tx_buffer->skb = NULL;
5740
5741 /* unmap skb header data */
5742 dma_unmap_single(tx_ring->dev,
5743 tx_buffer->dma,
5744 tx_buffer->length,
5745 DMA_TO_DEVICE);
5746
5747 /* clear last DMA location and unmap remaining buffers */
5748 while (tx_desc != eop_desc) {
5749 tx_buffer->dma = 0;
5750
5751 tx_buffer++;
5752 tx_desc++;
5753 i++;
5754 if (unlikely(!i)) {
5755 i -= tx_ring->count;
5756 tx_buffer = tx_ring->tx_buffer_info;
5757 tx_desc = IGB_TX_DESC(tx_ring, 0);
5758 }
5759
5760 /* unmap any remaining paged data */
5761 if (tx_buffer->dma) {
5762 dma_unmap_page(tx_ring->dev,
5763 tx_buffer->dma,
5764 tx_buffer->length,
5765 DMA_TO_DEVICE);
5766 }
5767 }
5768
5769 /* clear last DMA location */
5770 tx_buffer->dma = 0;
5771
5772 /* move us one more past the eop_desc for start of next pkt */
5773 tx_buffer++;
5774 tx_desc++;
5775 i++;
5776 if (unlikely(!i)) {
5777 i -= tx_ring->count;
5778 tx_buffer = tx_ring->tx_buffer_info;
5779 tx_desc = IGB_TX_DESC(tx_ring, 0);
5780 }
5781 }
5782
5783 i += tx_ring->count;
5784 tx_ring->next_to_clean = i;
5785 u64_stats_update_begin(&tx_ring->tx_syncp);
5786 tx_ring->tx_stats.bytes += total_bytes;
5787 tx_ring->tx_stats.packets += total_packets;
5788 u64_stats_update_end(&tx_ring->tx_syncp);
5789 q_vector->tx.total_bytes += total_bytes;
5790 q_vector->tx.total_packets += total_packets;
5791
5792 if (test_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags)) {
5793 struct e1000_hw *hw = &adapter->hw;
5794
5795 eop_desc = tx_buffer->next_to_watch;
5796
5797 /* Detect a transmit hang in hardware, this serializes the
5798 * check with the clearing of time_stamp and movement of i */
5799 clear_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
5800 if (eop_desc &&
5801 time_after(jiffies, tx_buffer->time_stamp +
5802 (adapter->tx_timeout_factor * HZ)) &&
5803 !(rd32(E1000_STATUS) & E1000_STATUS_TXOFF)) {
5804
5805 /* detected Tx unit hang */
5806 dev_err(tx_ring->dev,
5807 "Detected Tx Unit Hang\n"
5808 " Tx Queue <%d>\n"
5809 " TDH <%x>\n"
5810 " TDT <%x>\n"
5811 " next_to_use <%x>\n"
5812 " next_to_clean <%x>\n"
5813 "buffer_info[next_to_clean]\n"
5814 " time_stamp <%lx>\n"
5815 " next_to_watch <%p>\n"
5816 " jiffies <%lx>\n"
5817 " desc.status <%x>\n",
5818 tx_ring->queue_index,
5819 rd32(E1000_TDH(tx_ring->reg_idx)),
5820 readl(tx_ring->tail),
5821 tx_ring->next_to_use,
5822 tx_ring->next_to_clean,
5823 tx_buffer->time_stamp,
5824 eop_desc,
5825 jiffies,
5826 eop_desc->wb.status);
5827 netif_stop_subqueue(tx_ring->netdev,
5828 tx_ring->queue_index);
5829
5830 /* we are about to reset, no point in enabling stuff */
5831 return true;
5832 }
5833 }
5834
5835 if (unlikely(total_packets &&
5836 netif_carrier_ok(tx_ring->netdev) &&
5837 igb_desc_unused(tx_ring) >= IGB_TX_QUEUE_WAKE)) {
5838 /* Make sure that anybody stopping the queue after this
5839 * sees the new next_to_clean.
5840 */
5841 smp_mb();
5842 if (__netif_subqueue_stopped(tx_ring->netdev,
5843 tx_ring->queue_index) &&
5844 !(test_bit(__IGB_DOWN, &adapter->state))) {
5845 netif_wake_subqueue(tx_ring->netdev,
5846 tx_ring->queue_index);
5847
5848 u64_stats_update_begin(&tx_ring->tx_syncp);
5849 tx_ring->tx_stats.restart_queue++;
5850 u64_stats_update_end(&tx_ring->tx_syncp);
5851 }
5852 }
5853
5854 return !!budget;
5855}
5856
5857static inline void igb_rx_checksum(struct igb_ring *ring,
5858 union e1000_adv_rx_desc *rx_desc,
5859 struct sk_buff *skb)
5860{
5861 skb_checksum_none_assert(skb);
5862
5863 /* Ignore Checksum bit is set */
5864 if (igb_test_staterr(rx_desc, E1000_RXD_STAT_IXSM))
5865 return;
5866
5867 /* Rx checksum disabled via ethtool */
5868 if (!(ring->netdev->features & NETIF_F_RXCSUM))
5869 return;
5870
5871 /* TCP/UDP checksum error bit is set */
5872 if (igb_test_staterr(rx_desc,
5873 E1000_RXDEXT_STATERR_TCPE |
5874 E1000_RXDEXT_STATERR_IPE)) {
5875 /*
5876 * work around errata with sctp packets where the TCPE aka
5877 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
5878 * packets, (aka let the stack check the crc32c)
5879 */
5880 if (!((skb->len == 60) &&
5881 test_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags))) {
5882 u64_stats_update_begin(&ring->rx_syncp);
5883 ring->rx_stats.csum_err++;
5884 u64_stats_update_end(&ring->rx_syncp);
5885 }
5886 /* let the stack verify checksum errors */
5887 return;
5888 }
5889 /* It must be a TCP or UDP packet with a valid checksum */
5890 if (igb_test_staterr(rx_desc, E1000_RXD_STAT_TCPCS |
5891 E1000_RXD_STAT_UDPCS))
5892 skb->ip_summed = CHECKSUM_UNNECESSARY;
5893
5894 dev_dbg(ring->dev, "cksum success: bits %08X\n",
5895 le32_to_cpu(rx_desc->wb.upper.status_error));
5896}
5897
5898static inline void igb_rx_hash(struct igb_ring *ring,
5899 union e1000_adv_rx_desc *rx_desc,
5900 struct sk_buff *skb)
5901{
5902 if (ring->netdev->features & NETIF_F_RXHASH)
5903 skb->rxhash = le32_to_cpu(rx_desc->wb.lower.hi_dword.rss);
5904}
5905
5906static void igb_rx_hwtstamp(struct igb_q_vector *q_vector,
5907 union e1000_adv_rx_desc *rx_desc,
5908 struct sk_buff *skb)
5909{
5910 struct igb_adapter *adapter = q_vector->adapter;
5911 struct e1000_hw *hw = &adapter->hw;
5912 u64 regval;
5913
5914 if (!igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP |
5915 E1000_RXDADV_STAT_TS))
5916 return;
5917
5918 /*
5919 * If this bit is set, then the RX registers contain the time stamp. No
5920 * other packet will be time stamped until we read these registers, so
5921 * read the registers to make them available again. Because only one
5922 * packet can be time stamped at a time, we know that the register
5923 * values must belong to this one here and therefore we don't need to
5924 * compare any of the additional attributes stored for it.
5925 *
5926 * If nothing went wrong, then it should have a shared tx_flags that we
5927 * can turn into a skb_shared_hwtstamps.
5928 */
5929 if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
5930 u32 *stamp = (u32 *)skb->data;
5931 regval = le32_to_cpu(*(stamp + 2));
5932 regval |= (u64)le32_to_cpu(*(stamp + 3)) << 32;
5933 skb_pull(skb, IGB_TS_HDR_LEN);
5934 } else {
5935 if(!(rd32(E1000_TSYNCRXCTL) & E1000_TSYNCRXCTL_VALID))
5936 return;
5937
5938 regval = rd32(E1000_RXSTMPL);
5939 regval |= (u64)rd32(E1000_RXSTMPH) << 32;
5940 }
5941
5942 igb_systim_to_hwtstamp(adapter, skb_hwtstamps(skb), regval);
5943}
5944
5945static void igb_rx_vlan(struct igb_ring *ring,
5946 union e1000_adv_rx_desc *rx_desc,
5947 struct sk_buff *skb)
5948{
5949 if (igb_test_staterr(rx_desc, E1000_RXD_STAT_VP)) {
5950 u16 vid;
5951 if (igb_test_staterr(rx_desc, E1000_RXDEXT_STATERR_LB) &&
5952 test_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &ring->flags))
5953 vid = be16_to_cpu(rx_desc->wb.upper.vlan);
5954 else
5955 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
5956
5957 __vlan_hwaccel_put_tag(skb, vid);
5958 }
5959}
5960
5961static inline u16 igb_get_hlen(union e1000_adv_rx_desc *rx_desc)
5962{
5963 /* HW will not DMA in data larger than the given buffer, even if it
5964 * parses the (NFS, of course) header to be larger. In that case, it
5965 * fills the header buffer and spills the rest into the page.
5966 */
5967 u16 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hdr_info) &
5968 E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
5969 if (hlen > IGB_RX_HDR_LEN)
5970 hlen = IGB_RX_HDR_LEN;
5971 return hlen;
5972}
5973
5974static bool igb_clean_rx_irq(struct igb_q_vector *q_vector, int budget)
5975{
5976 struct igb_ring *rx_ring = q_vector->rx.ring;
5977 union e1000_adv_rx_desc *rx_desc;
5978 const int current_node = numa_node_id();
5979 unsigned int total_bytes = 0, total_packets = 0;
5980 u16 cleaned_count = igb_desc_unused(rx_ring);
5981 u16 i = rx_ring->next_to_clean;
5982
5983 rx_desc = IGB_RX_DESC(rx_ring, i);
5984
5985 while (igb_test_staterr(rx_desc, E1000_RXD_STAT_DD)) {
5986 struct igb_rx_buffer *buffer_info = &rx_ring->rx_buffer_info[i];
5987 struct sk_buff *skb = buffer_info->skb;
5988 union e1000_adv_rx_desc *next_rxd;
5989
5990 buffer_info->skb = NULL;
5991 prefetch(skb->data);
5992
5993 i++;
5994 if (i == rx_ring->count)
5995 i = 0;
5996
5997 next_rxd = IGB_RX_DESC(rx_ring, i);
5998 prefetch(next_rxd);
5999
6000 /*
6001 * This memory barrier is needed to keep us from reading
6002 * any other fields out of the rx_desc until we know the
6003 * RXD_STAT_DD bit is set
6004 */
6005 rmb();
6006
6007 if (!skb_is_nonlinear(skb)) {
6008 __skb_put(skb, igb_get_hlen(rx_desc));
6009 dma_unmap_single(rx_ring->dev, buffer_info->dma,
6010 IGB_RX_HDR_LEN,
6011 DMA_FROM_DEVICE);
6012 buffer_info->dma = 0;
6013 }
6014
6015 if (rx_desc->wb.upper.length) {
6016 u16 length = le16_to_cpu(rx_desc->wb.upper.length);
6017
6018 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
6019 buffer_info->page,
6020 buffer_info->page_offset,
6021 length);
6022
6023 skb->len += length;
6024 skb->data_len += length;
6025 skb->truesize += PAGE_SIZE / 2;
6026
6027 if ((page_count(buffer_info->page) != 1) ||
6028 (page_to_nid(buffer_info->page) != current_node))
6029 buffer_info->page = NULL;
6030 else
6031 get_page(buffer_info->page);
6032
6033 dma_unmap_page(rx_ring->dev, buffer_info->page_dma,
6034 PAGE_SIZE / 2, DMA_FROM_DEVICE);
6035 buffer_info->page_dma = 0;
6036 }
6037
6038 if (!igb_test_staterr(rx_desc, E1000_RXD_STAT_EOP)) {
6039 struct igb_rx_buffer *next_buffer;
6040 next_buffer = &rx_ring->rx_buffer_info[i];
6041 buffer_info->skb = next_buffer->skb;
6042 buffer_info->dma = next_buffer->dma;
6043 next_buffer->skb = skb;
6044 next_buffer->dma = 0;
6045 goto next_desc;
6046 }
6047
6048 if (igb_test_staterr(rx_desc,
6049 E1000_RXDEXT_ERR_FRAME_ERR_MASK)) {
6050 dev_kfree_skb_any(skb);
6051 goto next_desc;
6052 }
6053
6054 igb_rx_hwtstamp(q_vector, rx_desc, skb);
6055 igb_rx_hash(rx_ring, rx_desc, skb);
6056 igb_rx_checksum(rx_ring, rx_desc, skb);
6057 igb_rx_vlan(rx_ring, rx_desc, skb);
6058
6059 total_bytes += skb->len;
6060 total_packets++;
6061
6062 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
6063
6064 napi_gro_receive(&q_vector->napi, skb);
6065
6066 budget--;
6067next_desc:
6068 if (!budget)
6069 break;
6070
6071 cleaned_count++;
6072 /* return some buffers to hardware, one at a time is too slow */
6073 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
6074 igb_alloc_rx_buffers(rx_ring, cleaned_count);
6075 cleaned_count = 0;
6076 }
6077
6078 /* use prefetched values */
6079 rx_desc = next_rxd;
6080 }
6081
6082 rx_ring->next_to_clean = i;
6083 u64_stats_update_begin(&rx_ring->rx_syncp);
6084 rx_ring->rx_stats.packets += total_packets;
6085 rx_ring->rx_stats.bytes += total_bytes;
6086 u64_stats_update_end(&rx_ring->rx_syncp);
6087 q_vector->rx.total_packets += total_packets;
6088 q_vector->rx.total_bytes += total_bytes;
6089
6090 if (cleaned_count)
6091 igb_alloc_rx_buffers(rx_ring, cleaned_count);
6092
6093 return !!budget;
6094}
6095
6096static bool igb_alloc_mapped_skb(struct igb_ring *rx_ring,
6097 struct igb_rx_buffer *bi)
6098{
6099 struct sk_buff *skb = bi->skb;
6100 dma_addr_t dma = bi->dma;
6101
6102 if (dma)
6103 return true;
6104
6105 if (likely(!skb)) {
6106 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
6107 IGB_RX_HDR_LEN);
6108 bi->skb = skb;
6109 if (!skb) {
6110 rx_ring->rx_stats.alloc_failed++;
6111 return false;
6112 }
6113
6114 /* initialize skb for ring */
6115 skb_record_rx_queue(skb, rx_ring->queue_index);
6116 }
6117
6118 dma = dma_map_single(rx_ring->dev, skb->data,
6119 IGB_RX_HDR_LEN, DMA_FROM_DEVICE);
6120
6121 if (dma_mapping_error(rx_ring->dev, dma)) {
6122 rx_ring->rx_stats.alloc_failed++;
6123 return false;
6124 }
6125
6126 bi->dma = dma;
6127 return true;
6128}
6129
6130static bool igb_alloc_mapped_page(struct igb_ring *rx_ring,
6131 struct igb_rx_buffer *bi)
6132{
6133 struct page *page = bi->page;
6134 dma_addr_t page_dma = bi->page_dma;
6135 unsigned int page_offset = bi->page_offset ^ (PAGE_SIZE / 2);
6136
6137 if (page_dma)
6138 return true;
6139
6140 if (!page) {
6141 page = netdev_alloc_page(rx_ring->netdev);
6142 bi->page = page;
6143 if (unlikely(!page)) {
6144 rx_ring->rx_stats.alloc_failed++;
6145 return false;
6146 }
6147 }
6148
6149 page_dma = dma_map_page(rx_ring->dev, page,
6150 page_offset, PAGE_SIZE / 2,
6151 DMA_FROM_DEVICE);
6152
6153 if (dma_mapping_error(rx_ring->dev, page_dma)) {
6154 rx_ring->rx_stats.alloc_failed++;
6155 return false;
6156 }
6157
6158 bi->page_dma = page_dma;
6159 bi->page_offset = page_offset;
6160 return true;
6161}
6162
6163/**
6164 * igb_alloc_rx_buffers - Replace used receive buffers; packet split
6165 * @adapter: address of board private structure
6166 **/
6167void igb_alloc_rx_buffers(struct igb_ring *rx_ring, u16 cleaned_count)
6168{
6169 union e1000_adv_rx_desc *rx_desc;
6170 struct igb_rx_buffer *bi;
6171 u16 i = rx_ring->next_to_use;
6172
6173 rx_desc = IGB_RX_DESC(rx_ring, i);
6174 bi = &rx_ring->rx_buffer_info[i];
6175 i -= rx_ring->count;
6176
6177 while (cleaned_count--) {
6178 if (!igb_alloc_mapped_skb(rx_ring, bi))
6179 break;
6180
6181 /* Refresh the desc even if buffer_addrs didn't change
6182 * because each write-back erases this info. */
6183 rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
6184
6185 if (!igb_alloc_mapped_page(rx_ring, bi))
6186 break;
6187
6188 rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
6189
6190 rx_desc++;
6191 bi++;
6192 i++;
6193 if (unlikely(!i)) {
6194 rx_desc = IGB_RX_DESC(rx_ring, 0);
6195 bi = rx_ring->rx_buffer_info;
6196 i -= rx_ring->count;
6197 }
6198
6199 /* clear the hdr_addr for the next_to_use descriptor */
6200 rx_desc->read.hdr_addr = 0;
6201 }
6202
6203 i += rx_ring->count;
6204
6205 if (rx_ring->next_to_use != i) {
6206 rx_ring->next_to_use = i;
6207
6208 /* Force memory writes to complete before letting h/w
6209 * know there are new descriptors to fetch. (Only
6210 * applicable for weak-ordered memory model archs,
6211 * such as IA-64). */
6212 wmb();
6213 writel(i, rx_ring->tail);
6214 }
6215}
6216
6217/**
6218 * igb_mii_ioctl -
6219 * @netdev:
6220 * @ifreq:
6221 * @cmd:
6222 **/
6223static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
6224{
6225 struct igb_adapter *adapter = netdev_priv(netdev);
6226 struct mii_ioctl_data *data = if_mii(ifr);
6227
6228 if (adapter->hw.phy.media_type != e1000_media_type_copper)
6229 return -EOPNOTSUPP;
6230
6231 switch (cmd) {
6232 case SIOCGMIIPHY:
6233 data->phy_id = adapter->hw.phy.addr;
6234 break;
6235 case SIOCGMIIREG:
6236 if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
6237 &data->val_out))
6238 return -EIO;
6239 break;
6240 case SIOCSMIIREG:
6241 default:
6242 return -EOPNOTSUPP;
6243 }
6244 return 0;
6245}
6246
6247/**
6248 * igb_hwtstamp_ioctl - control hardware time stamping
6249 * @netdev:
6250 * @ifreq:
6251 * @cmd:
6252 *
6253 * Outgoing time stamping can be enabled and disabled. Play nice and
6254 * disable it when requested, although it shouldn't case any overhead
6255 * when no packet needs it. At most one packet in the queue may be
6256 * marked for time stamping, otherwise it would be impossible to tell
6257 * for sure to which packet the hardware time stamp belongs.
6258 *
6259 * Incoming time stamping has to be configured via the hardware
6260 * filters. Not all combinations are supported, in particular event
6261 * type has to be specified. Matching the kind of event packet is
6262 * not supported, with the exception of "all V2 events regardless of
6263 * level 2 or 4".
6264 *
6265 **/
6266static int igb_hwtstamp_ioctl(struct net_device *netdev,
6267 struct ifreq *ifr, int cmd)
6268{
6269 struct igb_adapter *adapter = netdev_priv(netdev);
6270 struct e1000_hw *hw = &adapter->hw;
6271 struct hwtstamp_config config;
6272 u32 tsync_tx_ctl = E1000_TSYNCTXCTL_ENABLED;
6273 u32 tsync_rx_ctl = E1000_TSYNCRXCTL_ENABLED;
6274 u32 tsync_rx_cfg = 0;
6275 bool is_l4 = false;
6276 bool is_l2 = false;
6277 u32 regval;
6278
6279 if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
6280 return -EFAULT;
6281
6282 /* reserved for future extensions */
6283 if (config.flags)
6284 return -EINVAL;
6285
6286 switch (config.tx_type) {
6287 case HWTSTAMP_TX_OFF:
6288 tsync_tx_ctl = 0;
6289 case HWTSTAMP_TX_ON:
6290 break;
6291 default:
6292 return -ERANGE;
6293 }
6294
6295 switch (config.rx_filter) {
6296 case HWTSTAMP_FILTER_NONE:
6297 tsync_rx_ctl = 0;
6298 break;
6299 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
6300 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
6301 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
6302 case HWTSTAMP_FILTER_ALL:
6303 /*
6304 * register TSYNCRXCFG must be set, therefore it is not
6305 * possible to time stamp both Sync and Delay_Req messages
6306 * => fall back to time stamping all packets
6307 */
6308 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_ALL;
6309 config.rx_filter = HWTSTAMP_FILTER_ALL;
6310 break;
6311 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
6312 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L4_V1;
6313 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V1_SYNC_MESSAGE;
6314 is_l4 = true;
6315 break;
6316 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
6317 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L4_V1;
6318 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V1_DELAY_REQ_MESSAGE;
6319 is_l4 = true;
6320 break;
6321 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
6322 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
6323 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
6324 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V2_SYNC_MESSAGE;
6325 is_l2 = true;
6326 is_l4 = true;
6327 config.rx_filter = HWTSTAMP_FILTER_SOME;
6328 break;
6329 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
6330 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
6331 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
6332 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V2_DELAY_REQ_MESSAGE;
6333 is_l2 = true;
6334 is_l4 = true;
6335 config.rx_filter = HWTSTAMP_FILTER_SOME;
6336 break;
6337 case HWTSTAMP_FILTER_PTP_V2_EVENT:
6338 case HWTSTAMP_FILTER_PTP_V2_SYNC:
6339 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
6340 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_EVENT_V2;
6341 config.rx_filter = HWTSTAMP_FILTER_PTP_V2_EVENT;
6342 is_l2 = true;
6343 is_l4 = true;
6344 break;
6345 default:
6346 return -ERANGE;
6347 }
6348
6349 if (hw->mac.type == e1000_82575) {
6350 if (tsync_rx_ctl | tsync_tx_ctl)
6351 return -EINVAL;
6352 return 0;
6353 }
6354
6355 /*
6356 * Per-packet timestamping only works if all packets are
6357 * timestamped, so enable timestamping in all packets as
6358 * long as one rx filter was configured.
6359 */
6360 if ((hw->mac.type >= e1000_82580) && tsync_rx_ctl) {
6361 tsync_rx_ctl = E1000_TSYNCRXCTL_ENABLED;
6362 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_ALL;
6363 }
6364
6365 /* enable/disable TX */
6366 regval = rd32(E1000_TSYNCTXCTL);
6367 regval &= ~E1000_TSYNCTXCTL_ENABLED;
6368 regval |= tsync_tx_ctl;
6369 wr32(E1000_TSYNCTXCTL, regval);
6370
6371 /* enable/disable RX */
6372 regval = rd32(E1000_TSYNCRXCTL);
6373 regval &= ~(E1000_TSYNCRXCTL_ENABLED | E1000_TSYNCRXCTL_TYPE_MASK);
6374 regval |= tsync_rx_ctl;
6375 wr32(E1000_TSYNCRXCTL, regval);
6376
6377 /* define which PTP packets are time stamped */
6378 wr32(E1000_TSYNCRXCFG, tsync_rx_cfg);
6379
6380 /* define ethertype filter for timestamped packets */
6381 if (is_l2)
6382 wr32(E1000_ETQF(3),
6383 (E1000_ETQF_FILTER_ENABLE | /* enable filter */
6384 E1000_ETQF_1588 | /* enable timestamping */
6385 ETH_P_1588)); /* 1588 eth protocol type */
6386 else
6387 wr32(E1000_ETQF(3), 0);
6388
6389#define PTP_PORT 319
6390 /* L4 Queue Filter[3]: filter by destination port and protocol */
6391 if (is_l4) {
6392 u32 ftqf = (IPPROTO_UDP /* UDP */
6393 | E1000_FTQF_VF_BP /* VF not compared */
6394 | E1000_FTQF_1588_TIME_STAMP /* Enable Timestamping */
6395 | E1000_FTQF_MASK); /* mask all inputs */
6396 ftqf &= ~E1000_FTQF_MASK_PROTO_BP; /* enable protocol check */
6397
6398 wr32(E1000_IMIR(3), htons(PTP_PORT));
6399 wr32(E1000_IMIREXT(3),
6400 (E1000_IMIREXT_SIZE_BP | E1000_IMIREXT_CTRL_BP));
6401 if (hw->mac.type == e1000_82576) {
6402 /* enable source port check */
6403 wr32(E1000_SPQF(3), htons(PTP_PORT));
6404 ftqf &= ~E1000_FTQF_MASK_SOURCE_PORT_BP;
6405 }
6406 wr32(E1000_FTQF(3), ftqf);
6407 } else {
6408 wr32(E1000_FTQF(3), E1000_FTQF_MASK);
6409 }
6410 wrfl();
6411
6412 adapter->hwtstamp_config = config;
6413
6414 /* clear TX/RX time stamp registers, just to be sure */
6415 regval = rd32(E1000_TXSTMPH);
6416 regval = rd32(E1000_RXSTMPH);
6417
6418 return copy_to_user(ifr->ifr_data, &config, sizeof(config)) ?
6419 -EFAULT : 0;
6420}
6421
6422/**
6423 * igb_ioctl -
6424 * @netdev:
6425 * @ifreq:
6426 * @cmd:
6427 **/
6428static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
6429{
6430 switch (cmd) {
6431 case SIOCGMIIPHY:
6432 case SIOCGMIIREG:
6433 case SIOCSMIIREG:
6434 return igb_mii_ioctl(netdev, ifr, cmd);
6435 case SIOCSHWTSTAMP:
6436 return igb_hwtstamp_ioctl(netdev, ifr, cmd);
6437 default:
6438 return -EOPNOTSUPP;
6439 }
6440}
6441
6442s32 igb_read_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
6443{
6444 struct igb_adapter *adapter = hw->back;
6445 u16 cap_offset;
6446
6447 cap_offset = adapter->pdev->pcie_cap;
6448 if (!cap_offset)
6449 return -E1000_ERR_CONFIG;
6450
6451 pci_read_config_word(adapter->pdev, cap_offset + reg, value);
6452
6453 return 0;
6454}
6455
6456s32 igb_write_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
6457{
6458 struct igb_adapter *adapter = hw->back;
6459 u16 cap_offset;
6460
6461 cap_offset = adapter->pdev->pcie_cap;
6462 if (!cap_offset)
6463 return -E1000_ERR_CONFIG;
6464
6465 pci_write_config_word(adapter->pdev, cap_offset + reg, *value);
6466
6467 return 0;
6468}
6469
6470static void igb_vlan_mode(struct net_device *netdev, u32 features)
6471{
6472 struct igb_adapter *adapter = netdev_priv(netdev);
6473 struct e1000_hw *hw = &adapter->hw;
6474 u32 ctrl, rctl;
6475 bool enable = !!(features & NETIF_F_HW_VLAN_RX);
6476
6477 if (enable) {
6478 /* enable VLAN tag insert/strip */
6479 ctrl = rd32(E1000_CTRL);
6480 ctrl |= E1000_CTRL_VME;
6481 wr32(E1000_CTRL, ctrl);
6482
6483 /* Disable CFI check */
6484 rctl = rd32(E1000_RCTL);
6485 rctl &= ~E1000_RCTL_CFIEN;
6486 wr32(E1000_RCTL, rctl);
6487 } else {
6488 /* disable VLAN tag insert/strip */
6489 ctrl = rd32(E1000_CTRL);
6490 ctrl &= ~E1000_CTRL_VME;
6491 wr32(E1000_CTRL, ctrl);
6492 }
6493
6494 igb_rlpml_set(adapter);
6495}
6496
6497static void igb_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
6498{
6499 struct igb_adapter *adapter = netdev_priv(netdev);
6500 struct e1000_hw *hw = &adapter->hw;
6501 int pf_id = adapter->vfs_allocated_count;
6502
6503 /* attempt to add filter to vlvf array */
6504 igb_vlvf_set(adapter, vid, true, pf_id);
6505
6506 /* add the filter since PF can receive vlans w/o entry in vlvf */
6507 igb_vfta_set(hw, vid, true);
6508
6509 set_bit(vid, adapter->active_vlans);
6510}
6511
6512static void igb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
6513{
6514 struct igb_adapter *adapter = netdev_priv(netdev);
6515 struct e1000_hw *hw = &adapter->hw;
6516 int pf_id = adapter->vfs_allocated_count;
6517 s32 err;
6518
6519 /* remove vlan from VLVF table array */
6520 err = igb_vlvf_set(adapter, vid, false, pf_id);
6521
6522 /* if vid was not present in VLVF just remove it from table */
6523 if (err)
6524 igb_vfta_set(hw, vid, false);
6525
6526 clear_bit(vid, adapter->active_vlans);
6527}
6528
6529static void igb_restore_vlan(struct igb_adapter *adapter)
6530{
6531 u16 vid;
6532
6533 igb_vlan_mode(adapter->netdev, adapter->netdev->features);
6534
6535 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
6536 igb_vlan_rx_add_vid(adapter->netdev, vid);
6537}
6538
6539int igb_set_spd_dplx(struct igb_adapter *adapter, u32 spd, u8 dplx)
6540{
6541 struct pci_dev *pdev = adapter->pdev;
6542 struct e1000_mac_info *mac = &adapter->hw.mac;
6543
6544 mac->autoneg = 0;
6545
6546 /* Make sure dplx is at most 1 bit and lsb of speed is not set
6547 * for the switch() below to work */
6548 if ((spd & 1) || (dplx & ~1))
6549 goto err_inval;
6550
6551 /* Fiber NIC's only allow 1000 Gbps Full duplex */
6552 if ((adapter->hw.phy.media_type == e1000_media_type_internal_serdes) &&
6553 spd != SPEED_1000 &&
6554 dplx != DUPLEX_FULL)
6555 goto err_inval;
6556
6557 switch (spd + dplx) {
6558 case SPEED_10 + DUPLEX_HALF:
6559 mac->forced_speed_duplex = ADVERTISE_10_HALF;
6560 break;
6561 case SPEED_10 + DUPLEX_FULL:
6562 mac->forced_speed_duplex = ADVERTISE_10_FULL;
6563 break;
6564 case SPEED_100 + DUPLEX_HALF:
6565 mac->forced_speed_duplex = ADVERTISE_100_HALF;
6566 break;
6567 case SPEED_100 + DUPLEX_FULL:
6568 mac->forced_speed_duplex = ADVERTISE_100_FULL;
6569 break;
6570 case SPEED_1000 + DUPLEX_FULL:
6571 mac->autoneg = 1;
6572 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
6573 break;
6574 case SPEED_1000 + DUPLEX_HALF: /* not supported */
6575 default:
6576 goto err_inval;
6577 }
6578 return 0;
6579
6580err_inval:
6581 dev_err(&pdev->dev, "Unsupported Speed/Duplex configuration\n");
6582 return -EINVAL;
6583}
6584
6585static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake)
6586{
6587 struct net_device *netdev = pci_get_drvdata(pdev);
6588 struct igb_adapter *adapter = netdev_priv(netdev);
6589 struct e1000_hw *hw = &adapter->hw;
6590 u32 ctrl, rctl, status;
6591 u32 wufc = adapter->wol;
6592#ifdef CONFIG_PM
6593 int retval = 0;
6594#endif
6595
6596 netif_device_detach(netdev);
6597
6598 if (netif_running(netdev))
6599 igb_close(netdev);
6600
6601 igb_clear_interrupt_scheme(adapter);
6602
6603#ifdef CONFIG_PM
6604 retval = pci_save_state(pdev);
6605 if (retval)
6606 return retval;
6607#endif
6608
6609 status = rd32(E1000_STATUS);
6610 if (status & E1000_STATUS_LU)
6611 wufc &= ~E1000_WUFC_LNKC;
6612
6613 if (wufc) {
6614 igb_setup_rctl(adapter);
6615 igb_set_rx_mode(netdev);
6616
6617 /* turn on all-multi mode if wake on multicast is enabled */
6618 if (wufc & E1000_WUFC_MC) {
6619 rctl = rd32(E1000_RCTL);
6620 rctl |= E1000_RCTL_MPE;
6621 wr32(E1000_RCTL, rctl);
6622 }
6623
6624 ctrl = rd32(E1000_CTRL);
6625 /* advertise wake from D3Cold */
6626 #define E1000_CTRL_ADVD3WUC 0x00100000
6627 /* phy power management enable */
6628 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
6629 ctrl |= E1000_CTRL_ADVD3WUC;
6630 wr32(E1000_CTRL, ctrl);
6631
6632 /* Allow time for pending master requests to run */
6633 igb_disable_pcie_master(hw);
6634
6635 wr32(E1000_WUC, E1000_WUC_PME_EN);
6636 wr32(E1000_WUFC, wufc);
6637 } else {
6638 wr32(E1000_WUC, 0);
6639 wr32(E1000_WUFC, 0);
6640 }
6641
6642 *enable_wake = wufc || adapter->en_mng_pt;
6643 if (!*enable_wake)
6644 igb_power_down_link(adapter);
6645 else
6646 igb_power_up_link(adapter);
6647
6648 /* Release control of h/w to f/w. If f/w is AMT enabled, this
6649 * would have already happened in close and is redundant. */
6650 igb_release_hw_control(adapter);
6651
6652 pci_disable_device(pdev);
6653
6654 return 0;
6655}
6656
6657#ifdef CONFIG_PM
6658static int igb_suspend(struct pci_dev *pdev, pm_message_t state)
6659{
6660 int retval;
6661 bool wake;
6662
6663 retval = __igb_shutdown(pdev, &wake);
6664 if (retval)
6665 return retval;
6666
6667 if (wake) {
6668 pci_prepare_to_sleep(pdev);
6669 } else {
6670 pci_wake_from_d3(pdev, false);
6671 pci_set_power_state(pdev, PCI_D3hot);
6672 }
6673
6674 return 0;
6675}
6676
6677static int igb_resume(struct pci_dev *pdev)
6678{
6679 struct net_device *netdev = pci_get_drvdata(pdev);
6680 struct igb_adapter *adapter = netdev_priv(netdev);
6681 struct e1000_hw *hw = &adapter->hw;
6682 u32 err;
6683
6684 pci_set_power_state(pdev, PCI_D0);
6685 pci_restore_state(pdev);
6686 pci_save_state(pdev);
6687
6688 err = pci_enable_device_mem(pdev);
6689 if (err) {
6690 dev_err(&pdev->dev,
6691 "igb: Cannot enable PCI device from suspend\n");
6692 return err;
6693 }
6694 pci_set_master(pdev);
6695
6696 pci_enable_wake(pdev, PCI_D3hot, 0);
6697 pci_enable_wake(pdev, PCI_D3cold, 0);
6698
6699 if (igb_init_interrupt_scheme(adapter)) {
6700 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
6701 return -ENOMEM;
6702 }
6703
6704 igb_reset(adapter);
6705
6706 /* let the f/w know that the h/w is now under the control of the
6707 * driver. */
6708 igb_get_hw_control(adapter);
6709
6710 wr32(E1000_WUS, ~0);
6711
6712 if (netif_running(netdev)) {
6713 err = igb_open(netdev);
6714 if (err)
6715 return err;
6716 }
6717
6718 netif_device_attach(netdev);
6719
6720 return 0;
6721}
6722#endif
6723
6724static void igb_shutdown(struct pci_dev *pdev)
6725{
6726 bool wake;
6727
6728 __igb_shutdown(pdev, &wake);
6729
6730 if (system_state == SYSTEM_POWER_OFF) {
6731 pci_wake_from_d3(pdev, wake);
6732 pci_set_power_state(pdev, PCI_D3hot);
6733 }
6734}
6735
6736#ifdef CONFIG_NET_POLL_CONTROLLER
6737/*
6738 * Polling 'interrupt' - used by things like netconsole to send skbs
6739 * without having to re-enable interrupts. It's not called while
6740 * the interrupt routine is executing.
6741 */
6742static void igb_netpoll(struct net_device *netdev)
6743{
6744 struct igb_adapter *adapter = netdev_priv(netdev);
6745 struct e1000_hw *hw = &adapter->hw;
6746 struct igb_q_vector *q_vector;
6747 int i;
6748
6749 for (i = 0; i < adapter->num_q_vectors; i++) {
6750 q_vector = adapter->q_vector[i];
6751 if (adapter->msix_entries)
6752 wr32(E1000_EIMC, q_vector->eims_value);
6753 else
6754 igb_irq_disable(adapter);
6755 napi_schedule(&q_vector->napi);
6756 }
6757}
6758#endif /* CONFIG_NET_POLL_CONTROLLER */
6759
6760/**
6761 * igb_io_error_detected - called when PCI error is detected
6762 * @pdev: Pointer to PCI device
6763 * @state: The current pci connection state
6764 *
6765 * This function is called after a PCI bus error affecting
6766 * this device has been detected.
6767 */
6768static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
6769 pci_channel_state_t state)
6770{
6771 struct net_device *netdev = pci_get_drvdata(pdev);
6772 struct igb_adapter *adapter = netdev_priv(netdev);
6773
6774 netif_device_detach(netdev);
6775
6776 if (state == pci_channel_io_perm_failure)
6777 return PCI_ERS_RESULT_DISCONNECT;
6778
6779 if (netif_running(netdev))
6780 igb_down(adapter);
6781 pci_disable_device(pdev);
6782
6783 /* Request a slot slot reset. */
6784 return PCI_ERS_RESULT_NEED_RESET;
6785}
6786
6787/**
6788 * igb_io_slot_reset - called after the pci bus has been reset.
6789 * @pdev: Pointer to PCI device
6790 *
6791 * Restart the card from scratch, as if from a cold-boot. Implementation
6792 * resembles the first-half of the igb_resume routine.
6793 */
6794static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
6795{
6796 struct net_device *netdev = pci_get_drvdata(pdev);
6797 struct igb_adapter *adapter = netdev_priv(netdev);
6798 struct e1000_hw *hw = &adapter->hw;
6799 pci_ers_result_t result;
6800 int err;
6801
6802 if (pci_enable_device_mem(pdev)) {
6803 dev_err(&pdev->dev,
6804 "Cannot re-enable PCI device after reset.\n");
6805 result = PCI_ERS_RESULT_DISCONNECT;
6806 } else {
6807 pci_set_master(pdev);
6808 pci_restore_state(pdev);
6809 pci_save_state(pdev);
6810
6811 pci_enable_wake(pdev, PCI_D3hot, 0);
6812 pci_enable_wake(pdev, PCI_D3cold, 0);
6813
6814 igb_reset(adapter);
6815 wr32(E1000_WUS, ~0);
6816 result = PCI_ERS_RESULT_RECOVERED;
6817 }
6818
6819 err = pci_cleanup_aer_uncorrect_error_status(pdev);
6820 if (err) {
6821 dev_err(&pdev->dev, "pci_cleanup_aer_uncorrect_error_status "
6822 "failed 0x%0x\n", err);
6823 /* non-fatal, continue */
6824 }
6825
6826 return result;
6827}
6828
6829/**
6830 * igb_io_resume - called when traffic can start flowing again.
6831 * @pdev: Pointer to PCI device
6832 *
6833 * This callback is called when the error recovery driver tells us that
6834 * its OK to resume normal operation. Implementation resembles the
6835 * second-half of the igb_resume routine.
6836 */
6837static void igb_io_resume(struct pci_dev *pdev)
6838{
6839 struct net_device *netdev = pci_get_drvdata(pdev);
6840 struct igb_adapter *adapter = netdev_priv(netdev);
6841
6842 if (netif_running(netdev)) {
6843 if (igb_up(adapter)) {
6844 dev_err(&pdev->dev, "igb_up failed after reset\n");
6845 return;
6846 }
6847 }
6848
6849 netif_device_attach(netdev);
6850
6851 /* let the f/w know that the h/w is now under the control of the
6852 * driver. */
6853 igb_get_hw_control(adapter);
6854}
6855
6856static void igb_rar_set_qsel(struct igb_adapter *adapter, u8 *addr, u32 index,
6857 u8 qsel)
6858{
6859 u32 rar_low, rar_high;
6860 struct e1000_hw *hw = &adapter->hw;
6861
6862 /* HW expects these in little endian so we reverse the byte order
6863 * from network order (big endian) to little endian
6864 */
6865 rar_low = ((u32) addr[0] | ((u32) addr[1] << 8) |
6866 ((u32) addr[2] << 16) | ((u32) addr[3] << 24));
6867 rar_high = ((u32) addr[4] | ((u32) addr[5] << 8));
6868
6869 /* Indicate to hardware the Address is Valid. */
6870 rar_high |= E1000_RAH_AV;
6871
6872 if (hw->mac.type == e1000_82575)
6873 rar_high |= E1000_RAH_POOL_1 * qsel;
6874 else
6875 rar_high |= E1000_RAH_POOL_1 << qsel;
6876
6877 wr32(E1000_RAL(index), rar_low);
6878 wrfl();
6879 wr32(E1000_RAH(index), rar_high);
6880 wrfl();
6881}
6882
6883static int igb_set_vf_mac(struct igb_adapter *adapter,
6884 int vf, unsigned char *mac_addr)
6885{
6886 struct e1000_hw *hw = &adapter->hw;
6887 /* VF MAC addresses start at end of receive addresses and moves
6888 * torwards the first, as a result a collision should not be possible */
6889 int rar_entry = hw->mac.rar_entry_count - (vf + 1);
6890
6891 memcpy(adapter->vf_data[vf].vf_mac_addresses, mac_addr, ETH_ALEN);
6892
6893 igb_rar_set_qsel(adapter, mac_addr, rar_entry, vf);
6894
6895 return 0;
6896}
6897
6898static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
6899{
6900 struct igb_adapter *adapter = netdev_priv(netdev);
6901 if (!is_valid_ether_addr(mac) || (vf >= adapter->vfs_allocated_count))
6902 return -EINVAL;
6903 adapter->vf_data[vf].flags |= IGB_VF_FLAG_PF_SET_MAC;
6904 dev_info(&adapter->pdev->dev, "setting MAC %pM on VF %d\n", mac, vf);
6905 dev_info(&adapter->pdev->dev, "Reload the VF driver to make this"
6906 " change effective.");
6907 if (test_bit(__IGB_DOWN, &adapter->state)) {
6908 dev_warn(&adapter->pdev->dev, "The VF MAC address has been set,"
6909 " but the PF device is not up.\n");
6910 dev_warn(&adapter->pdev->dev, "Bring the PF device up before"
6911 " attempting to use the VF device.\n");
6912 }
6913 return igb_set_vf_mac(adapter, vf, mac);
6914}
6915
6916static int igb_link_mbps(int internal_link_speed)
6917{
6918 switch (internal_link_speed) {
6919 case SPEED_100:
6920 return 100;
6921 case SPEED_1000:
6922 return 1000;
6923 default:
6924 return 0;
6925 }
6926}
6927
6928static void igb_set_vf_rate_limit(struct e1000_hw *hw, int vf, int tx_rate,
6929 int link_speed)
6930{
6931 int rf_dec, rf_int;
6932 u32 bcnrc_val;
6933
6934 if (tx_rate != 0) {
6935 /* Calculate the rate factor values to set */
6936 rf_int = link_speed / tx_rate;
6937 rf_dec = (link_speed - (rf_int * tx_rate));
6938 rf_dec = (rf_dec * (1<<E1000_RTTBCNRC_RF_INT_SHIFT)) / tx_rate;
6939
6940 bcnrc_val = E1000_RTTBCNRC_RS_ENA;
6941 bcnrc_val |= ((rf_int<<E1000_RTTBCNRC_RF_INT_SHIFT) &
6942 E1000_RTTBCNRC_RF_INT_MASK);
6943 bcnrc_val |= (rf_dec & E1000_RTTBCNRC_RF_DEC_MASK);
6944 } else {
6945 bcnrc_val = 0;
6946 }
6947
6948 wr32(E1000_RTTDQSEL, vf); /* vf X uses queue X */
6949 wr32(E1000_RTTBCNRC, bcnrc_val);
6950}
6951
6952static void igb_check_vf_rate_limit(struct igb_adapter *adapter)
6953{
6954 int actual_link_speed, i;
6955 bool reset_rate = false;
6956
6957 /* VF TX rate limit was not set or not supported */
6958 if ((adapter->vf_rate_link_speed == 0) ||
6959 (adapter->hw.mac.type != e1000_82576))
6960 return;
6961
6962 actual_link_speed = igb_link_mbps(adapter->link_speed);
6963 if (actual_link_speed != adapter->vf_rate_link_speed) {
6964 reset_rate = true;
6965 adapter->vf_rate_link_speed = 0;
6966 dev_info(&adapter->pdev->dev,
6967 "Link speed has been changed. VF Transmit "
6968 "rate is disabled\n");
6969 }
6970
6971 for (i = 0; i < adapter->vfs_allocated_count; i++) {
6972 if (reset_rate)
6973 adapter->vf_data[i].tx_rate = 0;
6974
6975 igb_set_vf_rate_limit(&adapter->hw, i,
6976 adapter->vf_data[i].tx_rate,
6977 actual_link_speed);
6978 }
6979}
6980
6981static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf, int tx_rate)
6982{
6983 struct igb_adapter *adapter = netdev_priv(netdev);
6984 struct e1000_hw *hw = &adapter->hw;
6985 int actual_link_speed;
6986
6987 if (hw->mac.type != e1000_82576)
6988 return -EOPNOTSUPP;
6989
6990 actual_link_speed = igb_link_mbps(adapter->link_speed);
6991 if ((vf >= adapter->vfs_allocated_count) ||
6992 (!(rd32(E1000_STATUS) & E1000_STATUS_LU)) ||
6993 (tx_rate < 0) || (tx_rate > actual_link_speed))
6994 return -EINVAL;
6995
6996 adapter->vf_rate_link_speed = actual_link_speed;
6997 adapter->vf_data[vf].tx_rate = (u16)tx_rate;
6998 igb_set_vf_rate_limit(hw, vf, tx_rate, actual_link_speed);
6999
7000 return 0;
7001}
7002
7003static int igb_ndo_get_vf_config(struct net_device *netdev,
7004 int vf, struct ifla_vf_info *ivi)
7005{
7006 struct igb_adapter *adapter = netdev_priv(netdev);
7007 if (vf >= adapter->vfs_allocated_count)
7008 return -EINVAL;
7009 ivi->vf = vf;
7010 memcpy(&ivi->mac, adapter->vf_data[vf].vf_mac_addresses, ETH_ALEN);
7011 ivi->tx_rate = adapter->vf_data[vf].tx_rate;
7012 ivi->vlan = adapter->vf_data[vf].pf_vlan;
7013 ivi->qos = adapter->vf_data[vf].pf_qos;
7014 return 0;
7015}
7016
7017static void igb_vmm_control(struct igb_adapter *adapter)
7018{
7019 struct e1000_hw *hw = &adapter->hw;
7020 u32 reg;
7021
7022 switch (hw->mac.type) {
7023 case e1000_82575:
7024 default:
7025 /* replication is not supported for 82575 */
7026 return;
7027 case e1000_82576:
7028 /* notify HW that the MAC is adding vlan tags */
7029 reg = rd32(E1000_DTXCTL);
7030 reg |= E1000_DTXCTL_VLAN_ADDED;
7031 wr32(E1000_DTXCTL, reg);
7032 case e1000_82580:
7033 /* enable replication vlan tag stripping */
7034 reg = rd32(E1000_RPLOLR);
7035 reg |= E1000_RPLOLR_STRVLAN;
7036 wr32(E1000_RPLOLR, reg);
7037 case e1000_i350:
7038 /* none of the above registers are supported by i350 */
7039 break;
7040 }
7041
7042 if (adapter->vfs_allocated_count) {
7043 igb_vmdq_set_loopback_pf(hw, true);
7044 igb_vmdq_set_replication_pf(hw, true);
7045 igb_vmdq_set_anti_spoofing_pf(hw, true,
7046 adapter->vfs_allocated_count);
7047 } else {
7048 igb_vmdq_set_loopback_pf(hw, false);
7049 igb_vmdq_set_replication_pf(hw, false);
7050 }
7051}
7052
7053static void igb_init_dmac(struct igb_adapter *adapter, u32 pba)
7054{
7055 struct e1000_hw *hw = &adapter->hw;
7056 u32 dmac_thr;
7057 u16 hwm;
7058
7059 if (hw->mac.type > e1000_82580) {
7060 if (adapter->flags & IGB_FLAG_DMAC) {
7061 u32 reg;
7062
7063 /* force threshold to 0. */
7064 wr32(E1000_DMCTXTH, 0);
7065
7066 /*
7067 * DMA Coalescing high water mark needs to be higher
7068 * than the RX threshold. set hwm to PBA - 2 * max
7069 * frame size
7070 */
7071 hwm = pba - (2 * adapter->max_frame_size);
7072 reg = rd32(E1000_DMACR);
7073 reg &= ~E1000_DMACR_DMACTHR_MASK;
7074 dmac_thr = pba - 4;
7075
7076 reg |= ((dmac_thr << E1000_DMACR_DMACTHR_SHIFT)
7077 & E1000_DMACR_DMACTHR_MASK);
7078
7079 /* transition to L0x or L1 if available..*/
7080 reg |= (E1000_DMACR_DMAC_EN | E1000_DMACR_DMAC_LX_MASK);
7081
7082 /* watchdog timer= +-1000 usec in 32usec intervals */
7083 reg |= (1000 >> 5);
7084 wr32(E1000_DMACR, reg);
7085
7086 /*
7087 * no lower threshold to disable
7088 * coalescing(smart fifb)-UTRESH=0
7089 */
7090 wr32(E1000_DMCRTRH, 0);
7091 wr32(E1000_FCRTC, hwm);
7092
7093 reg = (IGB_DMCTLX_DCFLUSH_DIS | 0x4);
7094
7095 wr32(E1000_DMCTLX, reg);
7096
7097 /*
7098 * free space in tx packet buffer to wake from
7099 * DMA coal
7100 */
7101 wr32(E1000_DMCTXTH, (IGB_MIN_TXPBSIZE -
7102 (IGB_TX_BUF_4096 + adapter->max_frame_size)) >> 6);
7103
7104 /*
7105 * make low power state decision controlled
7106 * by DMA coal
7107 */
7108 reg = rd32(E1000_PCIEMISC);
7109 reg &= ~E1000_PCIEMISC_LX_DECISION;
7110 wr32(E1000_PCIEMISC, reg);
7111 } /* endif adapter->dmac is not disabled */
7112 } else if (hw->mac.type == e1000_82580) {
7113 u32 reg = rd32(E1000_PCIEMISC);
7114 wr32(E1000_PCIEMISC, reg & ~E1000_PCIEMISC_LX_DECISION);
7115 wr32(E1000_DMACR, 0);
7116 }
7117}
7118
7119/* igb_main.c */
generated by cgit v1.2.2 (git 2.25.0) at 2025-10-03 22:31:24 -0400