1 /*******************************************************************************
3 Intel 10 Gigabit PCI Express Linux driver
4 Copyright(c) 1999 - 2011 Intel Corporation.
6 This program is free software; you can redistribute it and/or modify it
7 under the terms and conditions of the GNU General Public License,
8 version 2, as published by the Free Software Foundation.
10 This program is distributed in the hope it will be useful, but WITHOUT
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
23 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26 *******************************************************************************/
28 #include <linux/types.h>
29 #include <linux/module.h>
30 #include <linux/pci.h>
31 #include <linux/netdevice.h>
32 #include <linux/vmalloc.h>
33 #include <linux/string.h>
35 #include <linux/interrupt.h>
37 #include <linux/tcp.h>
38 #include <linux/sctp.h>
39 #include <linux/pkt_sched.h>
40 #include <linux/ipv6.h>
41 #include <linux/slab.h>
42 #include <net/checksum.h>
43 #include <net/ip6_checksum.h>
44 #include <linux/ethtool.h>
46 #include <linux/if_vlan.h>
47 #include <linux/prefetch.h>
48 #include <scsi/fc/fc_fcoe.h>
51 #include "ixgbe_common.h"
52 #include "ixgbe_dcb_82599.h"
53 #include "ixgbe_sriov.h"
55 char ixgbe_driver_name[] = "ixgbe";
56 static const char ixgbe_driver_string[] =
57 "Intel(R) 10 Gigabit PCI Express Network Driver";
61 #define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) "." \
62 __stringify(BUILD) "-k"
63 const char ixgbe_driver_version[] = DRV_VERSION;
64 static const char ixgbe_copyright[] =
65 "Copyright (c) 1999-2011 Intel Corporation.";
67 static const struct ixgbe_info *ixgbe_info_tbl[] = {
68 [board_82598] = &ixgbe_82598_info,
69 [board_82599] = &ixgbe_82599_info,
70 [board_X540] = &ixgbe_X540_info,
73 /* ixgbe_pci_tbl - PCI Device ID Table
75 * Wildcard entries (PCI_ANY_ID) should come last
76 * Last entry must be all 0s
78 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
79 * Class, Class Mask, private data (not used) }
81 static DEFINE_PCI_DEVICE_TABLE(ixgbe_pci_tbl) = {
82 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598), board_82598 },
83 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT), board_82598 },
84 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT), board_82598 },
85 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT), board_82598 },
86 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2), board_82598 },
87 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4), board_82598 },
88 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT), board_82598 },
89 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT), board_82598 },
90 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM), board_82598 },
91 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR), board_82598 },
92 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM), board_82598 },
93 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX), board_82598 },
94 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4), board_82599 },
95 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM), board_82599 },
96 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR), board_82599 },
97 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP), board_82599 },
98 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM), board_82599 },
99 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ), board_82599 },
100 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4), board_82599 },
101 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE), board_82599 },
102 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE), board_82599 },
103 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM), board_82599 },
104 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE), board_82599 },
105 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T), board_X540 },
106 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF2), board_82599 },
107 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_LS), board_82599 },
108 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599EN_SFP), board_82599 },
109 /* required last entry */
112 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
114 #ifdef CONFIG_IXGBE_DCA
115 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
117 static struct notifier_block dca_notifier = {
118 .notifier_call = ixgbe_notify_dca,
124 #ifdef CONFIG_PCI_IOV
125 static unsigned int max_vfs;
126 module_param(max_vfs, uint, 0);
127 MODULE_PARM_DESC(max_vfs,
128 "Maximum number of virtual functions to allocate per physical function");
129 #endif /* CONFIG_PCI_IOV */
131 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
132 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
133 MODULE_LICENSE("GPL");
134 MODULE_VERSION(DRV_VERSION);
136 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
138 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
140 if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
141 !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
142 schedule_work(&adapter->service_task);
145 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
147 BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
149 /* flush memory to make sure state is correct before next watchdog */
150 smp_mb__before_clear_bit();
151 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
154 struct ixgbe_reg_info {
159 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
161 /* General Registers */
162 {IXGBE_CTRL, "CTRL"},
163 {IXGBE_STATUS, "STATUS"},
164 {IXGBE_CTRL_EXT, "CTRL_EXT"},
166 /* Interrupt Registers */
167 {IXGBE_EICR, "EICR"},
170 {IXGBE_SRRCTL(0), "SRRCTL"},
171 {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
172 {IXGBE_RDLEN(0), "RDLEN"},
173 {IXGBE_RDH(0), "RDH"},
174 {IXGBE_RDT(0), "RDT"},
175 {IXGBE_RXDCTL(0), "RXDCTL"},
176 {IXGBE_RDBAL(0), "RDBAL"},
177 {IXGBE_RDBAH(0), "RDBAH"},
180 {IXGBE_TDBAL(0), "TDBAL"},
181 {IXGBE_TDBAH(0), "TDBAH"},
182 {IXGBE_TDLEN(0), "TDLEN"},
183 {IXGBE_TDH(0), "TDH"},
184 {IXGBE_TDT(0), "TDT"},
185 {IXGBE_TXDCTL(0), "TXDCTL"},
187 /* List Terminator */
193 * ixgbe_regdump - register printout routine
195 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
201 switch (reginfo->ofs) {
202 case IXGBE_SRRCTL(0):
203 for (i = 0; i < 64; i++)
204 regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
206 case IXGBE_DCA_RXCTRL(0):
207 for (i = 0; i < 64; i++)
208 regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
211 for (i = 0; i < 64; i++)
212 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
215 for (i = 0; i < 64; i++)
216 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
219 for (i = 0; i < 64; i++)
220 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
222 case IXGBE_RXDCTL(0):
223 for (i = 0; i < 64; i++)
224 regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
227 for (i = 0; i < 64; i++)
228 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
231 for (i = 0; i < 64; i++)
232 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
235 for (i = 0; i < 64; i++)
236 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
239 for (i = 0; i < 64; i++)
240 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
243 for (i = 0; i < 64; i++)
244 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
247 for (i = 0; i < 64; i++)
248 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
251 for (i = 0; i < 64; i++)
252 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
254 case IXGBE_TXDCTL(0):
255 for (i = 0; i < 64; i++)
256 regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
259 pr_info("%-15s %08x\n", reginfo->name,
260 IXGBE_READ_REG(hw, reginfo->ofs));
264 for (i = 0; i < 8; i++) {
265 snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i*8, i*8+7);
266 pr_err("%-15s", rname);
267 for (j = 0; j < 8; j++)
268 pr_cont(" %08x", regs[i*8+j]);
275 * ixgbe_dump - Print registers, tx-rings and rx-rings
277 static void ixgbe_dump(struct ixgbe_adapter *adapter)
279 struct net_device *netdev = adapter->netdev;
280 struct ixgbe_hw *hw = &adapter->hw;
281 struct ixgbe_reg_info *reginfo;
283 struct ixgbe_ring *tx_ring;
284 struct ixgbe_tx_buffer *tx_buffer_info;
285 union ixgbe_adv_tx_desc *tx_desc;
286 struct my_u0 { u64 a; u64 b; } *u0;
287 struct ixgbe_ring *rx_ring;
288 union ixgbe_adv_rx_desc *rx_desc;
289 struct ixgbe_rx_buffer *rx_buffer_info;
293 if (!netif_msg_hw(adapter))
296 /* Print netdevice Info */
298 dev_info(&adapter->pdev->dev, "Net device Info\n");
299 pr_info("Device Name state "
300 "trans_start last_rx\n");
301 pr_info("%-15s %016lX %016lX %016lX\n",
308 /* Print Registers */
309 dev_info(&adapter->pdev->dev, "Register Dump\n");
310 pr_info(" Register Name Value\n");
311 for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
312 reginfo->name; reginfo++) {
313 ixgbe_regdump(hw, reginfo);
316 /* Print TX Ring Summary */
317 if (!netdev || !netif_running(netdev))
320 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
321 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
322 for (n = 0; n < adapter->num_tx_queues; n++) {
323 tx_ring = adapter->tx_ring[n];
325 &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
326 pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
327 n, tx_ring->next_to_use, tx_ring->next_to_clean,
328 (u64)tx_buffer_info->dma,
329 tx_buffer_info->length,
330 tx_buffer_info->next_to_watch,
331 (u64)tx_buffer_info->time_stamp);
335 if (!netif_msg_tx_done(adapter))
336 goto rx_ring_summary;
338 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
340 /* Transmit Descriptor Formats
342 * Advanced Transmit Descriptor
343 * +--------------------------------------------------------------+
344 * 0 | Buffer Address [63:0] |
345 * +--------------------------------------------------------------+
346 * 8 | PAYLEN | PORTS | IDX | STA | DCMD |DTYP | RSV | DTALEN |
347 * +--------------------------------------------------------------+
348 * 63 46 45 40 39 36 35 32 31 24 23 20 19 0
351 for (n = 0; n < adapter->num_tx_queues; n++) {
352 tx_ring = adapter->tx_ring[n];
353 pr_info("------------------------------------\n");
354 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
355 pr_info("------------------------------------\n");
356 pr_info("T [desc] [address 63:0 ] "
357 "[PlPOIdStDDt Ln] [bi->dma ] "
358 "leng ntw timestamp bi->skb\n");
360 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
361 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
362 tx_buffer_info = &tx_ring->tx_buffer_info[i];
363 u0 = (struct my_u0 *)tx_desc;
364 pr_info("T [0x%03X] %016llX %016llX %016llX"
365 " %04X %p %016llX %p", i,
368 (u64)tx_buffer_info->dma,
369 tx_buffer_info->length,
370 tx_buffer_info->next_to_watch,
371 (u64)tx_buffer_info->time_stamp,
372 tx_buffer_info->skb);
373 if (i == tx_ring->next_to_use &&
374 i == tx_ring->next_to_clean)
376 else if (i == tx_ring->next_to_use)
378 else if (i == tx_ring->next_to_clean)
383 if (netif_msg_pktdata(adapter) &&
384 tx_buffer_info->dma != 0)
385 print_hex_dump(KERN_INFO, "",
386 DUMP_PREFIX_ADDRESS, 16, 1,
387 phys_to_virt(tx_buffer_info->dma),
388 tx_buffer_info->length, true);
392 /* Print RX Rings Summary */
394 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
395 pr_info("Queue [NTU] [NTC]\n");
396 for (n = 0; n < adapter->num_rx_queues; n++) {
397 rx_ring = adapter->rx_ring[n];
398 pr_info("%5d %5X %5X\n",
399 n, rx_ring->next_to_use, rx_ring->next_to_clean);
403 if (!netif_msg_rx_status(adapter))
406 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
408 /* Advanced Receive Descriptor (Read) Format
410 * +-----------------------------------------------------+
411 * 0 | Packet Buffer Address [63:1] |A0/NSE|
412 * +----------------------------------------------+------+
413 * 8 | Header Buffer Address [63:1] | DD |
414 * +-----------------------------------------------------+
417 * Advanced Receive Descriptor (Write-Back) Format
419 * 63 48 47 32 31 30 21 20 16 15 4 3 0
420 * +------------------------------------------------------+
421 * 0 | Packet IP |SPH| HDR_LEN | RSV|Packet| RSS |
422 * | Checksum Ident | | | | Type | Type |
423 * +------------------------------------------------------+
424 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
425 * +------------------------------------------------------+
426 * 63 48 47 32 31 20 19 0
428 for (n = 0; n < adapter->num_rx_queues; n++) {
429 rx_ring = adapter->rx_ring[n];
430 pr_info("------------------------------------\n");
431 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
432 pr_info("------------------------------------\n");
433 pr_info("R [desc] [ PktBuf A0] "
434 "[ HeadBuf DD] [bi->dma ] [bi->skb] "
435 "<-- Adv Rx Read format\n");
436 pr_info("RWB[desc] [PcsmIpSHl PtRs] "
437 "[vl er S cks ln] ---------------- [bi->skb] "
438 "<-- Adv Rx Write-Back format\n");
440 for (i = 0; i < rx_ring->count; i++) {
441 rx_buffer_info = &rx_ring->rx_buffer_info[i];
442 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
443 u0 = (struct my_u0 *)rx_desc;
444 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
445 if (staterr & IXGBE_RXD_STAT_DD) {
446 /* Descriptor Done */
447 pr_info("RWB[0x%03X] %016llX "
448 "%016llX ---------------- %p", i,
451 rx_buffer_info->skb);
453 pr_info("R [0x%03X] %016llX "
454 "%016llX %016llX %p", i,
457 (u64)rx_buffer_info->dma,
458 rx_buffer_info->skb);
460 if (netif_msg_pktdata(adapter)) {
461 print_hex_dump(KERN_INFO, "",
462 DUMP_PREFIX_ADDRESS, 16, 1,
463 phys_to_virt(rx_buffer_info->dma),
464 rx_ring->rx_buf_len, true);
466 if (rx_ring->rx_buf_len
468 print_hex_dump(KERN_INFO, "",
469 DUMP_PREFIX_ADDRESS, 16, 1,
471 rx_buffer_info->page_dma +
472 rx_buffer_info->page_offset
478 if (i == rx_ring->next_to_use)
480 else if (i == rx_ring->next_to_clean)
492 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
496 /* Let firmware take over control of h/w */
497 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
498 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
499 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
502 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
506 /* Let firmware know the driver has taken over */
507 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
508 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
509 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
513 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
514 * @adapter: pointer to adapter struct
515 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
516 * @queue: queue to map the corresponding interrupt to
517 * @msix_vector: the vector to map to the corresponding queue
520 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
521 u8 queue, u8 msix_vector)
524 struct ixgbe_hw *hw = &adapter->hw;
525 switch (hw->mac.type) {
526 case ixgbe_mac_82598EB:
527 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
530 index = (((direction * 64) + queue) >> 2) & 0x1F;
531 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
532 ivar &= ~(0xFF << (8 * (queue & 0x3)));
533 ivar |= (msix_vector << (8 * (queue & 0x3)));
534 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
536 case ixgbe_mac_82599EB:
538 if (direction == -1) {
540 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
541 index = ((queue & 1) * 8);
542 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
543 ivar &= ~(0xFF << index);
544 ivar |= (msix_vector << index);
545 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
548 /* tx or rx causes */
549 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
550 index = ((16 * (queue & 1)) + (8 * direction));
551 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
552 ivar &= ~(0xFF << index);
553 ivar |= (msix_vector << index);
554 IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
562 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
567 switch (adapter->hw.mac.type) {
568 case ixgbe_mac_82598EB:
569 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
570 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
572 case ixgbe_mac_82599EB:
574 mask = (qmask & 0xFFFFFFFF);
575 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
576 mask = (qmask >> 32);
577 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
584 static inline void ixgbe_unmap_tx_resource(struct ixgbe_ring *ring,
585 struct ixgbe_tx_buffer *tx_buffer)
587 if (tx_buffer->dma) {
588 if (tx_buffer->tx_flags & IXGBE_TX_FLAGS_MAPPED_AS_PAGE)
589 dma_unmap_page(ring->dev,
594 dma_unmap_single(ring->dev,
602 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_ring *tx_ring,
603 struct ixgbe_tx_buffer *tx_buffer_info)
605 ixgbe_unmap_tx_resource(tx_ring, tx_buffer_info);
606 if (tx_buffer_info->skb)
607 dev_kfree_skb_any(tx_buffer_info->skb);
608 tx_buffer_info->skb = NULL;
609 /* tx_buffer_info must be completely set up in the transmit path */
612 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
614 struct ixgbe_hw *hw = &adapter->hw;
615 struct ixgbe_hw_stats *hwstats = &adapter->stats;
620 if ((hw->fc.current_mode == ixgbe_fc_full) ||
621 (hw->fc.current_mode == ixgbe_fc_rx_pause)) {
622 switch (hw->mac.type) {
623 case ixgbe_mac_82598EB:
624 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
627 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
629 hwstats->lxoffrxc += data;
631 /* refill credits (no tx hang) if we received xoff */
635 for (i = 0; i < adapter->num_tx_queues; i++)
636 clear_bit(__IXGBE_HANG_CHECK_ARMED,
637 &adapter->tx_ring[i]->state);
639 } else if (!(adapter->dcb_cfg.pfc_mode_enable))
642 /* update stats for each tc, only valid with PFC enabled */
643 for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
644 switch (hw->mac.type) {
645 case ixgbe_mac_82598EB:
646 xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
649 xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
651 hwstats->pxoffrxc[i] += xoff[i];
654 /* disarm tx queues that have received xoff frames */
655 for (i = 0; i < adapter->num_tx_queues; i++) {
656 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
657 u8 tc = tx_ring->dcb_tc;
660 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
664 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
666 return ring->tx_stats.completed;
669 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
671 struct ixgbe_adapter *adapter = netdev_priv(ring->netdev);
672 struct ixgbe_hw *hw = &adapter->hw;
674 u32 head = IXGBE_READ_REG(hw, IXGBE_TDH(ring->reg_idx));
675 u32 tail = IXGBE_READ_REG(hw, IXGBE_TDT(ring->reg_idx));
678 return (head < tail) ?
679 tail - head : (tail + ring->count - head);
684 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
686 u32 tx_done = ixgbe_get_tx_completed(tx_ring);
687 u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
688 u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
691 clear_check_for_tx_hang(tx_ring);
694 * Check for a hung queue, but be thorough. This verifies
695 * that a transmit has been completed since the previous
696 * check AND there is at least one packet pending. The
697 * ARMED bit is set to indicate a potential hang. The
698 * bit is cleared if a pause frame is received to remove
699 * false hang detection due to PFC or 802.3x frames. By
700 * requiring this to fail twice we avoid races with
701 * pfc clearing the ARMED bit and conditions where we
702 * run the check_tx_hang logic with a transmit completion
703 * pending but without time to complete it yet.
705 if ((tx_done_old == tx_done) && tx_pending) {
706 /* make sure it is true for two checks in a row */
707 ret = test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
710 /* update completed stats and continue */
711 tx_ring->tx_stats.tx_done_old = tx_done;
712 /* reset the countdown */
713 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
720 * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
721 * @adapter: driver private struct
723 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
726 /* Do the reset outside of interrupt context */
727 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
728 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
729 ixgbe_service_event_schedule(adapter);
734 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
735 * @q_vector: structure containing interrupt and ring information
736 * @tx_ring: tx ring to clean
738 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
739 struct ixgbe_ring *tx_ring)
741 struct ixgbe_adapter *adapter = q_vector->adapter;
742 struct ixgbe_tx_buffer *tx_buffer;
743 union ixgbe_adv_tx_desc *tx_desc;
744 unsigned int total_bytes = 0, total_packets = 0;
745 unsigned int budget = q_vector->tx.work_limit;
746 u16 i = tx_ring->next_to_clean;
748 tx_buffer = &tx_ring->tx_buffer_info[i];
749 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
751 for (; budget; budget--) {
752 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
754 /* if next_to_watch is not set then there is no work pending */
758 /* if DD is not set pending work has not been completed */
759 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
762 /* count the packet as being completed */
763 tx_ring->tx_stats.completed++;
765 /* clear next_to_watch to prevent false hangs */
766 tx_buffer->next_to_watch = NULL;
768 /* prevent any other reads prior to eop_desc being verified */
772 ixgbe_unmap_tx_resource(tx_ring, tx_buffer);
773 tx_desc->wb.status = 0;
774 if (likely(tx_desc == eop_desc)) {
776 dev_kfree_skb_any(tx_buffer->skb);
777 tx_buffer->skb = NULL;
779 total_bytes += tx_buffer->bytecount;
780 total_packets += tx_buffer->gso_segs;
786 if (unlikely(i == tx_ring->count)) {
789 tx_buffer = tx_ring->tx_buffer_info;
790 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, 0);
796 tx_ring->next_to_clean = i;
797 u64_stats_update_begin(&tx_ring->syncp);
798 tx_ring->stats.bytes += total_bytes;
799 tx_ring->stats.packets += total_packets;
800 u64_stats_update_end(&tx_ring->syncp);
801 q_vector->tx.total_bytes += total_bytes;
802 q_vector->tx.total_packets += total_packets;
804 if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
805 /* schedule immediate reset if we believe we hung */
806 struct ixgbe_hw *hw = &adapter->hw;
807 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
808 e_err(drv, "Detected Tx Unit Hang\n"
810 " TDH, TDT <%x>, <%x>\n"
811 " next_to_use <%x>\n"
812 " next_to_clean <%x>\n"
813 "tx_buffer_info[next_to_clean]\n"
814 " time_stamp <%lx>\n"
816 tx_ring->queue_index,
817 IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
818 IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
819 tx_ring->next_to_use, i,
820 tx_ring->tx_buffer_info[i].time_stamp, jiffies);
822 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
825 "tx hang %d detected on queue %d, resetting adapter\n",
826 adapter->tx_timeout_count + 1, tx_ring->queue_index);
828 /* schedule immediate reset if we believe we hung */
829 ixgbe_tx_timeout_reset(adapter);
831 /* the adapter is about to reset, no point in enabling stuff */
835 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
836 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
837 (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
838 /* Make sure that anybody stopping the queue after this
839 * sees the new next_to_clean.
842 if (__netif_subqueue_stopped(tx_ring->netdev, tx_ring->queue_index) &&
843 !test_bit(__IXGBE_DOWN, &adapter->state)) {
844 netif_wake_subqueue(tx_ring->netdev, tx_ring->queue_index);
845 ++tx_ring->tx_stats.restart_queue;
852 #ifdef CONFIG_IXGBE_DCA
853 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
854 struct ixgbe_ring *rx_ring,
857 struct ixgbe_hw *hw = &adapter->hw;
859 u8 reg_idx = rx_ring->reg_idx;
861 rxctrl = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(reg_idx));
862 switch (hw->mac.type) {
863 case ixgbe_mac_82598EB:
864 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
865 rxctrl |= dca3_get_tag(rx_ring->dev, cpu);
867 case ixgbe_mac_82599EB:
869 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK_82599;
870 rxctrl |= (dca3_get_tag(rx_ring->dev, cpu) <<
871 IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599);
876 rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
877 rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
878 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_RRO_EN);
879 IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
882 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
883 struct ixgbe_ring *tx_ring,
886 struct ixgbe_hw *hw = &adapter->hw;
888 u8 reg_idx = tx_ring->reg_idx;
890 switch (hw->mac.type) {
891 case ixgbe_mac_82598EB:
892 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(reg_idx));
893 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
894 txctrl |= dca3_get_tag(tx_ring->dev, cpu);
895 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
896 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(reg_idx), txctrl);
898 case ixgbe_mac_82599EB:
900 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(reg_idx));
901 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK_82599;
902 txctrl |= (dca3_get_tag(tx_ring->dev, cpu) <<
903 IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599);
904 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
905 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(reg_idx), txctrl);
912 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
914 struct ixgbe_adapter *adapter = q_vector->adapter;
915 struct ixgbe_ring *ring;
918 if (q_vector->cpu == cpu)
921 for (ring = q_vector->tx.ring; ring != NULL; ring = ring->next)
922 ixgbe_update_tx_dca(adapter, ring, cpu);
924 for (ring = q_vector->rx.ring; ring != NULL; ring = ring->next)
925 ixgbe_update_rx_dca(adapter, ring, cpu);
932 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
937 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
940 /* always use CB2 mode, difference is masked in the CB driver */
941 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
943 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
944 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
948 for (i = 0; i < num_q_vectors; i++) {
949 adapter->q_vector[i]->cpu = -1;
950 ixgbe_update_dca(adapter->q_vector[i]);
954 static int __ixgbe_notify_dca(struct device *dev, void *data)
956 struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
957 unsigned long event = *(unsigned long *)data;
959 if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
963 case DCA_PROVIDER_ADD:
964 /* if we're already enabled, don't do it again */
965 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
967 if (dca_add_requester(dev) == 0) {
968 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
969 ixgbe_setup_dca(adapter);
972 /* Fall Through since DCA is disabled. */
973 case DCA_PROVIDER_REMOVE:
974 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
975 dca_remove_requester(dev);
976 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
977 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
984 #endif /* CONFIG_IXGBE_DCA */
986 static inline void ixgbe_rx_hash(union ixgbe_adv_rx_desc *rx_desc,
989 skb->rxhash = le32_to_cpu(rx_desc->wb.lower.hi_dword.rss);
993 * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
994 * @adapter: address of board private structure
995 * @rx_desc: advanced rx descriptor
997 * Returns : true if it is FCoE pkt
999 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_adapter *adapter,
1000 union ixgbe_adv_rx_desc *rx_desc)
1002 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1004 return (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
1005 ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1006 (cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1007 IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1011 * ixgbe_receive_skb - Send a completed packet up the stack
1012 * @adapter: board private structure
1013 * @skb: packet to send up
1014 * @status: hardware indication of status of receive
1015 * @rx_ring: rx descriptor ring (for a specific queue) to setup
1016 * @rx_desc: rx descriptor
1018 static void ixgbe_receive_skb(struct ixgbe_q_vector *q_vector,
1019 struct sk_buff *skb, u8 status,
1020 struct ixgbe_ring *ring,
1021 union ixgbe_adv_rx_desc *rx_desc)
1023 struct ixgbe_adapter *adapter = q_vector->adapter;
1024 struct napi_struct *napi = &q_vector->napi;
1025 bool is_vlan = (status & IXGBE_RXD_STAT_VP);
1026 u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
1028 if (is_vlan && (tag & VLAN_VID_MASK))
1029 __vlan_hwaccel_put_tag(skb, tag);
1031 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
1032 napi_gro_receive(napi, skb);
1038 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1039 * @adapter: address of board private structure
1040 * @status_err: hardware indication of status of receive
1041 * @skb: skb currently being received and modified
1042 * @status_err: status error value of last descriptor in packet
1044 static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
1045 union ixgbe_adv_rx_desc *rx_desc,
1046 struct sk_buff *skb,
1049 skb->ip_summed = CHECKSUM_NONE;
1051 /* Rx csum disabled */
1052 if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
1055 /* if IP and error */
1056 if ((status_err & IXGBE_RXD_STAT_IPCS) &&
1057 (status_err & IXGBE_RXDADV_ERR_IPE)) {
1058 adapter->hw_csum_rx_error++;
1062 if (!(status_err & IXGBE_RXD_STAT_L4CS))
1065 if (status_err & IXGBE_RXDADV_ERR_TCPE) {
1066 u16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1069 * 82599 errata, UDP frames with a 0 checksum can be marked as
1072 if ((pkt_info & IXGBE_RXDADV_PKTTYPE_UDP) &&
1073 (adapter->hw.mac.type == ixgbe_mac_82599EB))
1076 adapter->hw_csum_rx_error++;
1080 /* It must be a TCP or UDP packet with a valid checksum */
1081 skb->ip_summed = CHECKSUM_UNNECESSARY;
1084 static inline void ixgbe_release_rx_desc(struct ixgbe_ring *rx_ring, u32 val)
1087 * Force memory writes to complete before letting h/w
1088 * know there are new descriptors to fetch. (Only
1089 * applicable for weak-ordered memory model archs,
1093 writel(val, rx_ring->tail);
1097 * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
1098 * @rx_ring: ring to place buffers on
1099 * @cleaned_count: number of buffers to replace
1101 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1103 union ixgbe_adv_rx_desc *rx_desc;
1104 struct ixgbe_rx_buffer *bi;
1105 struct sk_buff *skb;
1106 u16 i = rx_ring->next_to_use;
1108 /* do nothing if no valid netdev defined */
1109 if (!rx_ring->netdev)
1112 while (cleaned_count--) {
1113 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1114 bi = &rx_ring->rx_buffer_info[i];
1118 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1119 rx_ring->rx_buf_len);
1121 rx_ring->rx_stats.alloc_rx_buff_failed++;
1124 /* initialize queue mapping */
1125 skb_record_rx_queue(skb, rx_ring->queue_index);
1130 bi->dma = dma_map_single(rx_ring->dev,
1132 rx_ring->rx_buf_len,
1134 if (dma_mapping_error(rx_ring->dev, bi->dma)) {
1135 rx_ring->rx_stats.alloc_rx_buff_failed++;
1141 if (ring_is_ps_enabled(rx_ring)) {
1143 bi->page = alloc_page(GFP_ATOMIC | __GFP_COLD);
1145 rx_ring->rx_stats.alloc_rx_page_failed++;
1150 if (!bi->page_dma) {
1151 /* use a half page if we're re-using */
1152 bi->page_offset ^= PAGE_SIZE / 2;
1153 bi->page_dma = dma_map_page(rx_ring->dev,
1158 if (dma_mapping_error(rx_ring->dev,
1160 rx_ring->rx_stats.alloc_rx_page_failed++;
1166 /* Refresh the desc even if buffer_addrs didn't change
1167 * because each write-back erases this info. */
1168 rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
1169 rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
1171 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
1172 rx_desc->read.hdr_addr = 0;
1176 if (i == rx_ring->count)
1181 if (rx_ring->next_to_use != i) {
1182 rx_ring->next_to_use = i;
1183 ixgbe_release_rx_desc(rx_ring, i);
1187 static inline u16 ixgbe_get_hlen(union ixgbe_adv_rx_desc *rx_desc)
1189 /* HW will not DMA in data larger than the given buffer, even if it
1190 * parses the (NFS, of course) header to be larger. In that case, it
1191 * fills the header buffer and spills the rest into the page.
1193 u16 hdr_info = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info);
1194 u16 hlen = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
1195 IXGBE_RXDADV_HDRBUFLEN_SHIFT;
1196 if (hlen > IXGBE_RX_HDR_SIZE)
1197 hlen = IXGBE_RX_HDR_SIZE;
1202 * ixgbe_transform_rsc_queue - change rsc queue into a full packet
1203 * @skb: pointer to the last skb in the rsc queue
1205 * This function changes a queue full of hw rsc buffers into a completed
1206 * packet. It uses the ->prev pointers to find the first packet and then
1207 * turns it into the frag list owner.
1209 static inline struct sk_buff *ixgbe_transform_rsc_queue(struct sk_buff *skb)
1211 unsigned int frag_list_size = 0;
1212 unsigned int skb_cnt = 1;
1215 struct sk_buff *prev = skb->prev;
1216 frag_list_size += skb->len;
1222 skb_shinfo(skb)->frag_list = skb->next;
1224 skb->len += frag_list_size;
1225 skb->data_len += frag_list_size;
1226 skb->truesize += frag_list_size;
1227 IXGBE_RSC_CB(skb)->skb_cnt = skb_cnt;
1232 static inline bool ixgbe_get_rsc_state(union ixgbe_adv_rx_desc *rx_desc)
1234 return !!(le32_to_cpu(rx_desc->wb.lower.lo_dword.data) &
1235 IXGBE_RXDADV_RSCCNT_MASK);
1238 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1239 struct ixgbe_ring *rx_ring,
1242 struct ixgbe_adapter *adapter = q_vector->adapter;
1243 union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
1244 struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
1245 struct sk_buff *skb;
1246 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1247 const int current_node = numa_node_id();
1250 #endif /* IXGBE_FCOE */
1253 u16 cleaned_count = 0;
1254 bool pkt_is_rsc = false;
1256 i = rx_ring->next_to_clean;
1257 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1258 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1260 while (staterr & IXGBE_RXD_STAT_DD) {
1263 rmb(); /* read descriptor and rx_buffer_info after status DD */
1265 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1267 skb = rx_buffer_info->skb;
1268 rx_buffer_info->skb = NULL;
1269 prefetch(skb->data);
1271 if (ring_is_rsc_enabled(rx_ring))
1272 pkt_is_rsc = ixgbe_get_rsc_state(rx_desc);
1274 /* linear means we are building an skb from multiple pages */
1275 if (!skb_is_nonlinear(skb)) {
1278 !(staterr & IXGBE_RXD_STAT_EOP) &&
1281 * When HWRSC is enabled, delay unmapping
1282 * of the first packet. It carries the
1283 * header information, HW may still
1284 * access the header after the writeback.
1285 * Only unmap it when EOP is reached
1287 IXGBE_RSC_CB(skb)->delay_unmap = true;
1288 IXGBE_RSC_CB(skb)->dma = rx_buffer_info->dma;
1290 dma_unmap_single(rx_ring->dev,
1291 rx_buffer_info->dma,
1292 rx_ring->rx_buf_len,
1295 rx_buffer_info->dma = 0;
1297 if (ring_is_ps_enabled(rx_ring)) {
1298 hlen = ixgbe_get_hlen(rx_desc);
1299 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1301 hlen = le16_to_cpu(rx_desc->wb.upper.length);
1306 /* assume packet split since header is unmapped */
1307 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1311 dma_unmap_page(rx_ring->dev,
1312 rx_buffer_info->page_dma,
1315 rx_buffer_info->page_dma = 0;
1316 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1317 rx_buffer_info->page,
1318 rx_buffer_info->page_offset,
1321 if ((page_count(rx_buffer_info->page) == 1) &&
1322 (page_to_nid(rx_buffer_info->page) == current_node))
1323 get_page(rx_buffer_info->page);
1325 rx_buffer_info->page = NULL;
1327 skb->len += upper_len;
1328 skb->data_len += upper_len;
1329 skb->truesize += PAGE_SIZE / 2;
1333 if (i == rx_ring->count)
1336 next_rxd = IXGBE_RX_DESC_ADV(rx_ring, i);
1341 u32 nextp = (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
1342 IXGBE_RXDADV_NEXTP_SHIFT;
1343 next_buffer = &rx_ring->rx_buffer_info[nextp];
1345 next_buffer = &rx_ring->rx_buffer_info[i];
1348 if (!(staterr & IXGBE_RXD_STAT_EOP)) {
1349 if (ring_is_ps_enabled(rx_ring)) {
1350 rx_buffer_info->skb = next_buffer->skb;
1351 rx_buffer_info->dma = next_buffer->dma;
1352 next_buffer->skb = skb;
1353 next_buffer->dma = 0;
1355 skb->next = next_buffer->skb;
1356 skb->next->prev = skb;
1358 rx_ring->rx_stats.non_eop_descs++;
1363 skb = ixgbe_transform_rsc_queue(skb);
1364 /* if we got here without RSC the packet is invalid */
1366 __pskb_trim(skb, 0);
1367 rx_buffer_info->skb = skb;
1372 if (ring_is_rsc_enabled(rx_ring)) {
1373 if (IXGBE_RSC_CB(skb)->delay_unmap) {
1374 dma_unmap_single(rx_ring->dev,
1375 IXGBE_RSC_CB(skb)->dma,
1376 rx_ring->rx_buf_len,
1378 IXGBE_RSC_CB(skb)->dma = 0;
1379 IXGBE_RSC_CB(skb)->delay_unmap = false;
1383 if (ring_is_ps_enabled(rx_ring))
1384 rx_ring->rx_stats.rsc_count +=
1385 skb_shinfo(skb)->nr_frags;
1387 rx_ring->rx_stats.rsc_count +=
1388 IXGBE_RSC_CB(skb)->skb_cnt;
1389 rx_ring->rx_stats.rsc_flush++;
1392 /* ERR_MASK will only have valid bits if EOP set */
1393 if (unlikely(staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK)) {
1394 dev_kfree_skb_any(skb);
1398 ixgbe_rx_checksum(adapter, rx_desc, skb, staterr);
1399 if (adapter->netdev->features & NETIF_F_RXHASH)
1400 ixgbe_rx_hash(rx_desc, skb);
1402 /* probably a little skewed due to removing CRC */
1403 total_rx_bytes += skb->len;
1406 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1408 /* if ddp, not passing to ULD unless for FCP_RSP or error */
1409 if (ixgbe_rx_is_fcoe(adapter, rx_desc)) {
1410 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb,
1413 dev_kfree_skb_any(skb);
1417 #endif /* IXGBE_FCOE */
1418 ixgbe_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
1422 rx_desc->wb.upper.status_error = 0;
1427 /* return some buffers to hardware, one at a time is too slow */
1428 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
1429 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1433 /* use prefetched values */
1435 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1438 rx_ring->next_to_clean = i;
1439 cleaned_count = ixgbe_desc_unused(rx_ring);
1442 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1445 /* include DDPed FCoE data */
1446 if (ddp_bytes > 0) {
1449 mss = rx_ring->netdev->mtu - sizeof(struct fcoe_hdr) -
1450 sizeof(struct fc_frame_header) -
1451 sizeof(struct fcoe_crc_eof);
1454 total_rx_bytes += ddp_bytes;
1455 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
1457 #endif /* IXGBE_FCOE */
1459 u64_stats_update_begin(&rx_ring->syncp);
1460 rx_ring->stats.packets += total_rx_packets;
1461 rx_ring->stats.bytes += total_rx_bytes;
1462 u64_stats_update_end(&rx_ring->syncp);
1463 q_vector->rx.total_packets += total_rx_packets;
1464 q_vector->rx.total_bytes += total_rx_bytes;
1470 * ixgbe_configure_msix - Configure MSI-X hardware
1471 * @adapter: board private structure
1473 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1476 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1478 struct ixgbe_q_vector *q_vector;
1479 int q_vectors, v_idx;
1482 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1484 /* Populate MSIX to EITR Select */
1485 if (adapter->num_vfs > 32) {
1486 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
1487 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
1491 * Populate the IVAR table and set the ITR values to the
1492 * corresponding register.
1494 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1495 struct ixgbe_ring *ring;
1496 q_vector = adapter->q_vector[v_idx];
1498 for (ring = q_vector->rx.ring; ring != NULL; ring = ring->next)
1499 ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1501 for (ring = q_vector->tx.ring; ring != NULL; ring = ring->next)
1502 ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1504 if (q_vector->tx.ring && !q_vector->rx.ring) {
1505 /* tx only vector */
1506 if (adapter->tx_itr_setting == 1)
1507 q_vector->itr = IXGBE_10K_ITR;
1509 q_vector->itr = adapter->tx_itr_setting;
1511 /* rx or rx/tx vector */
1512 if (adapter->rx_itr_setting == 1)
1513 q_vector->itr = IXGBE_20K_ITR;
1515 q_vector->itr = adapter->rx_itr_setting;
1518 ixgbe_write_eitr(q_vector);
1521 switch (adapter->hw.mac.type) {
1522 case ixgbe_mac_82598EB:
1523 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1526 case ixgbe_mac_82599EB:
1527 case ixgbe_mac_X540:
1528 ixgbe_set_ivar(adapter, -1, 1, v_idx);
1533 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1535 /* set up to autoclear timer, and the vectors */
1536 mask = IXGBE_EIMS_ENABLE_MASK;
1537 mask &= ~(IXGBE_EIMS_OTHER |
1538 IXGBE_EIMS_MAILBOX |
1541 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1544 enum latency_range {
1548 latency_invalid = 255
1552 * ixgbe_update_itr - update the dynamic ITR value based on statistics
1553 * @q_vector: structure containing interrupt and ring information
1554 * @ring_container: structure containing ring performance data
1556 * Stores a new ITR value based on packets and byte
1557 * counts during the last interrupt. The advantage of per interrupt
1558 * computation is faster updates and more accurate ITR for the current
1559 * traffic pattern. Constants in this function were computed
1560 * based on theoretical maximum wire speed and thresholds were set based
1561 * on testing data as well as attempting to minimize response time
1562 * while increasing bulk throughput.
1563 * this functionality is controlled by the InterruptThrottleRate module
1564 * parameter (see ixgbe_param.c)
1566 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
1567 struct ixgbe_ring_container *ring_container)
1570 struct ixgbe_adapter *adapter = q_vector->adapter;
1571 int bytes = ring_container->total_bytes;
1572 int packets = ring_container->total_packets;
1574 u8 itr_setting = ring_container->itr;
1579 /* simple throttlerate management
1580 * 0-20MB/s lowest (100000 ints/s)
1581 * 20-100MB/s low (20000 ints/s)
1582 * 100-1249MB/s bulk (8000 ints/s)
1584 /* what was last interrupt timeslice? */
1585 timepassed_us = q_vector->itr >> 2;
1586 bytes_perint = bytes / timepassed_us; /* bytes/usec */
1588 switch (itr_setting) {
1589 case lowest_latency:
1590 if (bytes_perint > adapter->eitr_low)
1591 itr_setting = low_latency;
1594 if (bytes_perint > adapter->eitr_high)
1595 itr_setting = bulk_latency;
1596 else if (bytes_perint <= adapter->eitr_low)
1597 itr_setting = lowest_latency;
1600 if (bytes_perint <= adapter->eitr_high)
1601 itr_setting = low_latency;
1605 /* clear work counters since we have the values we need */
1606 ring_container->total_bytes = 0;
1607 ring_container->total_packets = 0;
1609 /* write updated itr to ring container */
1610 ring_container->itr = itr_setting;
1614 * ixgbe_write_eitr - write EITR register in hardware specific way
1615 * @q_vector: structure containing interrupt and ring information
1617 * This function is made to be called by ethtool and by the driver
1618 * when it needs to update EITR registers at runtime. Hardware
1619 * specific quirks/differences are taken care of here.
1621 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1623 struct ixgbe_adapter *adapter = q_vector->adapter;
1624 struct ixgbe_hw *hw = &adapter->hw;
1625 int v_idx = q_vector->v_idx;
1626 u32 itr_reg = q_vector->itr;
1628 switch (adapter->hw.mac.type) {
1629 case ixgbe_mac_82598EB:
1630 /* must write high and low 16 bits to reset counter */
1631 itr_reg |= (itr_reg << 16);
1633 case ixgbe_mac_82599EB:
1634 case ixgbe_mac_X540:
1636 * set the WDIS bit to not clear the timer bits and cause an
1637 * immediate assertion of the interrupt
1639 itr_reg |= IXGBE_EITR_CNT_WDIS;
1644 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1647 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
1649 u32 new_itr = q_vector->itr;
1652 ixgbe_update_itr(q_vector, &q_vector->tx);
1653 ixgbe_update_itr(q_vector, &q_vector->rx);
1655 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
1657 switch (current_itr) {
1658 /* counts and packets in update_itr are dependent on these numbers */
1659 case lowest_latency:
1660 new_itr = IXGBE_100K_ITR;
1663 new_itr = IXGBE_20K_ITR;
1666 new_itr = IXGBE_8K_ITR;
1672 if (new_itr != q_vector->itr) {
1673 /* do an exponential smoothing */
1674 new_itr = (10 * new_itr * q_vector->itr) /
1675 ((9 * new_itr) + q_vector->itr);
1677 /* save the algorithm value here */
1678 q_vector->itr = new_itr & IXGBE_MAX_EITR;
1680 ixgbe_write_eitr(q_vector);
1685 * ixgbe_check_overtemp_subtask - check for over tempurature
1686 * @adapter: pointer to adapter
1688 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
1690 struct ixgbe_hw *hw = &adapter->hw;
1691 u32 eicr = adapter->interrupt_event;
1693 if (test_bit(__IXGBE_DOWN, &adapter->state))
1696 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
1697 !(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
1700 adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
1702 switch (hw->device_id) {
1703 case IXGBE_DEV_ID_82599_T3_LOM:
1705 * Since the warning interrupt is for both ports
1706 * we don't have to check if:
1707 * - This interrupt wasn't for our port.
1708 * - We may have missed the interrupt so always have to
1709 * check if we got a LSC
1711 if (!(eicr & IXGBE_EICR_GPI_SDP0) &&
1712 !(eicr & IXGBE_EICR_LSC))
1715 if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
1717 bool link_up = false;
1719 hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
1725 /* Check if this is not due to overtemp */
1726 if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
1731 if (!(eicr & IXGBE_EICR_GPI_SDP0))
1736 "Network adapter has been stopped because it has over heated. "
1737 "Restart the computer. If the problem persists, "
1738 "power off the system and replace the adapter\n");
1740 adapter->interrupt_event = 0;
1743 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
1745 struct ixgbe_hw *hw = &adapter->hw;
1747 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
1748 (eicr & IXGBE_EICR_GPI_SDP1)) {
1749 e_crit(probe, "Fan has stopped, replace the adapter\n");
1750 /* write to clear the interrupt */
1751 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1755 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
1757 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
1760 switch (adapter->hw.mac.type) {
1761 case ixgbe_mac_82599EB:
1763 * Need to check link state so complete overtemp check
1766 if (((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)) &&
1767 (!test_bit(__IXGBE_DOWN, &adapter->state))) {
1768 adapter->interrupt_event = eicr;
1769 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
1770 ixgbe_service_event_schedule(adapter);
1774 case ixgbe_mac_X540:
1775 if (!(eicr & IXGBE_EICR_TS))
1783 "Network adapter has been stopped because it has over heated. "
1784 "Restart the computer. If the problem persists, "
1785 "power off the system and replace the adapter\n");
1788 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
1790 struct ixgbe_hw *hw = &adapter->hw;
1792 if (eicr & IXGBE_EICR_GPI_SDP2) {
1793 /* Clear the interrupt */
1794 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
1795 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1796 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
1797 ixgbe_service_event_schedule(adapter);
1801 if (eicr & IXGBE_EICR_GPI_SDP1) {
1802 /* Clear the interrupt */
1803 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1804 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1805 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
1806 ixgbe_service_event_schedule(adapter);
1811 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1813 struct ixgbe_hw *hw = &adapter->hw;
1816 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1817 adapter->link_check_timeout = jiffies;
1818 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1819 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1820 IXGBE_WRITE_FLUSH(hw);
1821 ixgbe_service_event_schedule(adapter);
1825 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
1829 struct ixgbe_hw *hw = &adapter->hw;
1831 switch (hw->mac.type) {
1832 case ixgbe_mac_82598EB:
1833 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1834 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
1836 case ixgbe_mac_82599EB:
1837 case ixgbe_mac_X540:
1838 mask = (qmask & 0xFFFFFFFF);
1840 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
1841 mask = (qmask >> 32);
1843 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
1848 /* skip the flush */
1851 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
1855 struct ixgbe_hw *hw = &adapter->hw;
1857 switch (hw->mac.type) {
1858 case ixgbe_mac_82598EB:
1859 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1860 IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
1862 case ixgbe_mac_82599EB:
1863 case ixgbe_mac_X540:
1864 mask = (qmask & 0xFFFFFFFF);
1866 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
1867 mask = (qmask >> 32);
1869 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
1874 /* skip the flush */
1878 * ixgbe_irq_enable - Enable default interrupt generation settings
1879 * @adapter: board private structure
1881 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
1884 u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
1886 /* don't reenable LSC while waiting for link */
1887 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
1888 mask &= ~IXGBE_EIMS_LSC;
1890 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
1891 switch (adapter->hw.mac.type) {
1892 case ixgbe_mac_82599EB:
1893 mask |= IXGBE_EIMS_GPI_SDP0;
1895 case ixgbe_mac_X540:
1896 mask |= IXGBE_EIMS_TS;
1901 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
1902 mask |= IXGBE_EIMS_GPI_SDP1;
1903 switch (adapter->hw.mac.type) {
1904 case ixgbe_mac_82599EB:
1905 mask |= IXGBE_EIMS_GPI_SDP1;
1906 mask |= IXGBE_EIMS_GPI_SDP2;
1907 case ixgbe_mac_X540:
1908 mask |= IXGBE_EIMS_ECC;
1909 mask |= IXGBE_EIMS_MAILBOX;
1914 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
1915 !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
1916 mask |= IXGBE_EIMS_FLOW_DIR;
1918 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1920 ixgbe_irq_enable_queues(adapter, ~0);
1922 IXGBE_WRITE_FLUSH(&adapter->hw);
1925 static irqreturn_t ixgbe_msix_other(int irq, void *data)
1927 struct ixgbe_adapter *adapter = data;
1928 struct ixgbe_hw *hw = &adapter->hw;
1932 * Workaround for Silicon errata. Use clear-by-write instead
1933 * of clear-by-read. Reading with EICS will return the
1934 * interrupt causes without clearing, which later be done
1935 * with the write to EICR.
1937 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1938 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1940 if (eicr & IXGBE_EICR_LSC)
1941 ixgbe_check_lsc(adapter);
1943 if (eicr & IXGBE_EICR_MAILBOX)
1944 ixgbe_msg_task(adapter);
1946 switch (hw->mac.type) {
1947 case ixgbe_mac_82599EB:
1948 case ixgbe_mac_X540:
1949 if (eicr & IXGBE_EICR_ECC)
1950 e_info(link, "Received unrecoverable ECC Err, please "
1952 /* Handle Flow Director Full threshold interrupt */
1953 if (eicr & IXGBE_EICR_FLOW_DIR) {
1954 int reinit_count = 0;
1956 for (i = 0; i < adapter->num_tx_queues; i++) {
1957 struct ixgbe_ring *ring = adapter->tx_ring[i];
1958 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
1963 /* no more flow director interrupts until after init */
1964 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
1965 adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
1966 ixgbe_service_event_schedule(adapter);
1969 ixgbe_check_sfp_event(adapter, eicr);
1970 ixgbe_check_overtemp_event(adapter, eicr);
1976 ixgbe_check_fan_failure(adapter, eicr);
1978 /* re-enable the original interrupt state, no lsc, no queues */
1979 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1980 ixgbe_irq_enable(adapter, false, false);
1985 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
1987 struct ixgbe_q_vector *q_vector = data;
1989 /* EIAM disabled interrupts (on this vector) for us */
1991 if (q_vector->rx.ring || q_vector->tx.ring)
1992 napi_schedule(&q_vector->napi);
1997 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
2000 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2001 struct ixgbe_ring *rx_ring = a->rx_ring[r_idx];
2003 rx_ring->q_vector = q_vector;
2004 rx_ring->next = q_vector->rx.ring;
2005 q_vector->rx.ring = rx_ring;
2006 q_vector->rx.count++;
2009 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
2012 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2013 struct ixgbe_ring *tx_ring = a->tx_ring[t_idx];
2015 tx_ring->q_vector = q_vector;
2016 tx_ring->next = q_vector->tx.ring;
2017 q_vector->tx.ring = tx_ring;
2018 q_vector->tx.count++;
2019 q_vector->tx.work_limit = a->tx_work_limit;
2023 * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
2024 * @adapter: board private structure to initialize
2026 * This function maps descriptor rings to the queue-specific vectors
2027 * we were allotted through the MSI-X enabling code. Ideally, we'd have
2028 * one vector per ring/queue, but on a constrained vector budget, we
2029 * group the rings as "efficiently" as possible. You would add new
2030 * mapping configurations in here.
2032 static void ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter)
2034 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2035 int rxr_remaining = adapter->num_rx_queues, rxr_idx = 0;
2036 int txr_remaining = adapter->num_tx_queues, txr_idx = 0;
2039 /* only one q_vector if MSI-X is disabled. */
2040 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2044 * If we don't have enough vectors for a 1-to-1 mapping, we'll have to
2045 * group them so there are multiple queues per vector.
2047 * Re-adjusting *qpv takes care of the remainder.
2049 for (; v_start < q_vectors && rxr_remaining; v_start++) {
2050 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_start);
2051 for (; rqpv; rqpv--, rxr_idx++, rxr_remaining--)
2052 map_vector_to_rxq(adapter, v_start, rxr_idx);
2056 * If there are not enough q_vectors for each ring to have it's own
2057 * vector then we must pair up Rx/Tx on a each vector
2059 if ((v_start + txr_remaining) > q_vectors)
2062 for (; v_start < q_vectors && txr_remaining; v_start++) {
2063 int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_start);
2064 for (; tqpv; tqpv--, txr_idx++, txr_remaining--)
2065 map_vector_to_txq(adapter, v_start, txr_idx);
2070 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2071 * @adapter: board private structure
2073 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2074 * interrupts from the kernel.
2076 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2078 struct net_device *netdev = adapter->netdev;
2079 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2083 for (vector = 0; vector < q_vectors; vector++) {
2084 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2085 struct msix_entry *entry = &adapter->msix_entries[vector];
2087 if (q_vector->tx.ring && q_vector->rx.ring) {
2088 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2089 "%s-%s-%d", netdev->name, "TxRx", ri++);
2091 } else if (q_vector->rx.ring) {
2092 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2093 "%s-%s-%d", netdev->name, "rx", ri++);
2094 } else if (q_vector->tx.ring) {
2095 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2096 "%s-%s-%d", netdev->name, "tx", ti++);
2098 /* skip this unused q_vector */
2101 err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
2102 q_vector->name, q_vector);
2104 e_err(probe, "request_irq failed for MSIX interrupt "
2105 "Error: %d\n", err);
2106 goto free_queue_irqs;
2108 /* If Flow Director is enabled, set interrupt affinity */
2109 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2110 /* assign the mask for this irq */
2111 irq_set_affinity_hint(entry->vector,
2112 q_vector->affinity_mask);
2116 err = request_irq(adapter->msix_entries[vector].vector,
2117 ixgbe_msix_other, 0, netdev->name, adapter);
2119 e_err(probe, "request_irq for msix_lsc failed: %d\n", err);
2120 goto free_queue_irqs;
2128 irq_set_affinity_hint(adapter->msix_entries[vector].vector,
2130 free_irq(adapter->msix_entries[vector].vector,
2131 adapter->q_vector[vector]);
2133 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2134 pci_disable_msix(adapter->pdev);
2135 kfree(adapter->msix_entries);
2136 adapter->msix_entries = NULL;
2141 * ixgbe_intr - legacy mode Interrupt Handler
2142 * @irq: interrupt number
2143 * @data: pointer to a network interface device structure
2145 static irqreturn_t ixgbe_intr(int irq, void *data)
2147 struct ixgbe_adapter *adapter = data;
2148 struct ixgbe_hw *hw = &adapter->hw;
2149 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2153 * Workaround for silicon errata on 82598. Mask the interrupts
2154 * before the read of EICR.
2156 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2158 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2159 * therefore no explicit interrupt disable is necessary */
2160 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2163 * shared interrupt alert!
2164 * make sure interrupts are enabled because the read will
2165 * have disabled interrupts due to EIAM
2166 * finish the workaround of silicon errata on 82598. Unmask
2167 * the interrupt that we masked before the EICR read.
2169 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2170 ixgbe_irq_enable(adapter, true, true);
2171 return IRQ_NONE; /* Not our interrupt */
2174 if (eicr & IXGBE_EICR_LSC)
2175 ixgbe_check_lsc(adapter);
2177 switch (hw->mac.type) {
2178 case ixgbe_mac_82599EB:
2179 ixgbe_check_sfp_event(adapter, eicr);
2181 case ixgbe_mac_X540:
2182 if (eicr & IXGBE_EICR_ECC)
2183 e_info(link, "Received unrecoverable ECC err, please "
2185 ixgbe_check_overtemp_event(adapter, eicr);
2191 ixgbe_check_fan_failure(adapter, eicr);
2193 if (napi_schedule_prep(&(q_vector->napi))) {
2194 /* would disable interrupts here but EIAM disabled it */
2195 __napi_schedule(&(q_vector->napi));
2199 * re-enable link(maybe) and non-queue interrupts, no flush.
2200 * ixgbe_poll will re-enable the queue interrupts
2203 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2204 ixgbe_irq_enable(adapter, false, false);
2209 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
2211 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2214 /* legacy and MSI only use one vector */
2215 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2218 for (i = 0; i < adapter->num_rx_queues; i++) {
2219 adapter->rx_ring[i]->q_vector = NULL;
2220 adapter->rx_ring[i]->next = NULL;
2222 for (i = 0; i < adapter->num_tx_queues; i++) {
2223 adapter->tx_ring[i]->q_vector = NULL;
2224 adapter->tx_ring[i]->next = NULL;
2227 for (i = 0; i < q_vectors; i++) {
2228 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
2229 memset(&q_vector->rx, 0, sizeof(struct ixgbe_ring_container));
2230 memset(&q_vector->tx, 0, sizeof(struct ixgbe_ring_container));
2235 * ixgbe_request_irq - initialize interrupts
2236 * @adapter: board private structure
2238 * Attempts to configure interrupts using the best available
2239 * capabilities of the hardware and kernel.
2241 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2243 struct net_device *netdev = adapter->netdev;
2246 /* map all of the rings to the q_vectors */
2247 ixgbe_map_rings_to_vectors(adapter);
2249 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2250 err = ixgbe_request_msix_irqs(adapter);
2251 else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
2252 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2253 netdev->name, adapter);
2255 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2256 netdev->name, adapter);
2259 e_err(probe, "request_irq failed, Error %d\n", err);
2261 /* place q_vectors and rings back into a known good state */
2262 ixgbe_reset_q_vectors(adapter);
2268 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2270 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2273 q_vectors = adapter->num_msix_vectors;
2275 free_irq(adapter->msix_entries[i].vector, adapter);
2278 for (; i >= 0; i--) {
2279 /* free only the irqs that were actually requested */
2280 if (!adapter->q_vector[i]->rx.ring &&
2281 !adapter->q_vector[i]->tx.ring)
2284 /* clear the affinity_mask in the IRQ descriptor */
2285 irq_set_affinity_hint(adapter->msix_entries[i].vector,
2288 free_irq(adapter->msix_entries[i].vector,
2289 adapter->q_vector[i]);
2292 free_irq(adapter->pdev->irq, adapter);
2295 /* clear q_vector state information */
2296 ixgbe_reset_q_vectors(adapter);
2300 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2301 * @adapter: board private structure
2303 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2305 switch (adapter->hw.mac.type) {
2306 case ixgbe_mac_82598EB:
2307 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2309 case ixgbe_mac_82599EB:
2310 case ixgbe_mac_X540:
2311 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2312 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2313 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2318 IXGBE_WRITE_FLUSH(&adapter->hw);
2319 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2321 for (i = 0; i < adapter->num_msix_vectors; i++)
2322 synchronize_irq(adapter->msix_entries[i].vector);
2324 synchronize_irq(adapter->pdev->irq);
2329 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2332 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2334 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2337 if (adapter->rx_itr_setting == 1)
2338 q_vector->itr = IXGBE_20K_ITR;
2340 q_vector->itr = adapter->rx_itr_setting;
2342 ixgbe_write_eitr(q_vector);
2344 ixgbe_set_ivar(adapter, 0, 0, 0);
2345 ixgbe_set_ivar(adapter, 1, 0, 0);
2347 e_info(hw, "Legacy interrupt IVAR setup done\n");
2351 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2352 * @adapter: board private structure
2353 * @ring: structure containing ring specific data
2355 * Configure the Tx descriptor ring after a reset.
2357 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2358 struct ixgbe_ring *ring)
2360 struct ixgbe_hw *hw = &adapter->hw;
2361 u64 tdba = ring->dma;
2363 u32 txdctl = IXGBE_TXDCTL_ENABLE;
2364 u8 reg_idx = ring->reg_idx;
2366 /* disable queue to avoid issues while updating state */
2367 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
2368 IXGBE_WRITE_FLUSH(hw);
2370 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2371 (tdba & DMA_BIT_MASK(32)));
2372 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2373 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2374 ring->count * sizeof(union ixgbe_adv_tx_desc));
2375 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2376 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2377 ring->tail = hw->hw_addr + IXGBE_TDT(reg_idx);
2380 * set WTHRESH to encourage burst writeback, it should not be set
2381 * higher than 1 when ITR is 0 as it could cause false TX hangs
2383 * In order to avoid issues WTHRESH + PTHRESH should always be equal
2384 * to or less than the number of on chip descriptors, which is
2387 if (!adapter->tx_itr_setting || !adapter->rx_itr_setting)
2388 txdctl |= (1 << 16); /* WTHRESH = 1 */
2390 txdctl |= (8 << 16); /* WTHRESH = 8 */
2392 /* PTHRESH=32 is needed to avoid a Tx hang with DFP enabled. */
2393 txdctl |= (1 << 8) | /* HTHRESH = 1 */
2394 32; /* PTHRESH = 32 */
2396 /* reinitialize flowdirector state */
2397 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2398 adapter->atr_sample_rate) {
2399 ring->atr_sample_rate = adapter->atr_sample_rate;
2400 ring->atr_count = 0;
2401 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
2403 ring->atr_sample_rate = 0;
2406 clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
2409 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
2411 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2412 if (hw->mac.type == ixgbe_mac_82598EB &&
2413 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2416 /* poll to verify queue is enabled */
2418 usleep_range(1000, 2000);
2419 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2420 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
2422 e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
2425 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
2427 struct ixgbe_hw *hw = &adapter->hw;
2430 u8 tcs = netdev_get_num_tc(adapter->netdev);
2432 if (hw->mac.type == ixgbe_mac_82598EB)
2435 /* disable the arbiter while setting MTQC */
2436 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2437 rttdcs |= IXGBE_RTTDCS_ARBDIS;
2438 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2440 /* set transmit pool layout */
2441 switch (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
2442 case (IXGBE_FLAG_SRIOV_ENABLED):
2443 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2444 (IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF));
2448 reg = IXGBE_MTQC_64Q_1PB;
2450 reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
2452 reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
2454 IXGBE_WRITE_REG(hw, IXGBE_MTQC, reg);
2456 /* Enable Security TX Buffer IFG for multiple pb */
2458 reg = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
2459 reg |= IXGBE_SECTX_DCB;
2460 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, reg);
2465 /* re-enable the arbiter */
2466 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2467 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2471 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2472 * @adapter: board private structure
2474 * Configure the Tx unit of the MAC after a reset.
2476 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2478 struct ixgbe_hw *hw = &adapter->hw;
2482 ixgbe_setup_mtqc(adapter);
2484 if (hw->mac.type != ixgbe_mac_82598EB) {
2485 /* DMATXCTL.EN must be before Tx queues are enabled */
2486 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2487 dmatxctl |= IXGBE_DMATXCTL_TE;
2488 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2491 /* Setup the HW Tx Head and Tail descriptor pointers */
2492 for (i = 0; i < adapter->num_tx_queues; i++)
2493 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
2496 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2498 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2499 struct ixgbe_ring *rx_ring)
2502 u8 reg_idx = rx_ring->reg_idx;
2504 switch (adapter->hw.mac.type) {
2505 case ixgbe_mac_82598EB: {
2506 struct ixgbe_ring_feature *feature = adapter->ring_feature;
2507 const int mask = feature[RING_F_RSS].mask;
2508 reg_idx = reg_idx & mask;
2511 case ixgbe_mac_82599EB:
2512 case ixgbe_mac_X540:
2517 srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx));
2519 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
2520 srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
2521 if (adapter->num_vfs)
2522 srrctl |= IXGBE_SRRCTL_DROP_EN;
2524 srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
2525 IXGBE_SRRCTL_BSIZEHDR_MASK;
2527 if (ring_is_ps_enabled(rx_ring)) {
2528 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
2529 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2531 srrctl |= (PAGE_SIZE / 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2533 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2535 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
2536 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2537 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2540 IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx), srrctl);
2543 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2545 struct ixgbe_hw *hw = &adapter->hw;
2546 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2547 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2548 0x6A3E67EA, 0x14364D17, 0x3BED200D};
2549 u32 mrqc = 0, reta = 0;
2552 u8 tcs = netdev_get_num_tc(adapter->netdev);
2553 int maxq = adapter->ring_feature[RING_F_RSS].indices;
2556 maxq = min(maxq, adapter->num_tx_queues / tcs);
2558 /* Fill out hash function seeds */
2559 for (i = 0; i < 10; i++)
2560 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2562 /* Fill out redirection table */
2563 for (i = 0, j = 0; i < 128; i++, j++) {
2566 /* reta = 4-byte sliding window of
2567 * 0x00..(indices-1)(indices-1)00..etc. */
2568 reta = (reta << 8) | (j * 0x11);
2570 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2573 /* Disable indicating checksum in descriptor, enables RSS hash */
2574 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2575 rxcsum |= IXGBE_RXCSUM_PCSD;
2576 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2578 if (adapter->hw.mac.type == ixgbe_mac_82598EB &&
2579 (adapter->flags & IXGBE_FLAG_RSS_ENABLED)) {
2580 mrqc = IXGBE_MRQC_RSSEN;
2582 int mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2583 | IXGBE_FLAG_SRIOV_ENABLED);
2586 case (IXGBE_FLAG_RSS_ENABLED):
2588 mrqc = IXGBE_MRQC_RSSEN;
2590 mrqc = IXGBE_MRQC_RTRSS4TCEN;
2592 mrqc = IXGBE_MRQC_RTRSS8TCEN;
2594 case (IXGBE_FLAG_SRIOV_ENABLED):
2595 mrqc = IXGBE_MRQC_VMDQEN;
2602 /* Perform hash on these packet types */
2603 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2604 | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2605 | IXGBE_MRQC_RSS_FIELD_IPV6
2606 | IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
2608 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2612 * ixgbe_configure_rscctl - enable RSC for the indicated ring
2613 * @adapter: address of board private structure
2614 * @index: index of ring to set
2616 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
2617 struct ixgbe_ring *ring)
2619 struct ixgbe_hw *hw = &adapter->hw;
2622 u8 reg_idx = ring->reg_idx;
2624 if (!ring_is_rsc_enabled(ring))
2627 rx_buf_len = ring->rx_buf_len;
2628 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
2629 rscctrl |= IXGBE_RSCCTL_RSCEN;
2631 * we must limit the number of descriptors so that the
2632 * total size of max desc * buf_len is not greater
2635 if (ring_is_ps_enabled(ring)) {
2636 #if (MAX_SKB_FRAGS > 16)
2637 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2638 #elif (MAX_SKB_FRAGS > 8)
2639 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2640 #elif (MAX_SKB_FRAGS > 4)
2641 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2643 rscctrl |= IXGBE_RSCCTL_MAXDESC_1;
2646 if (rx_buf_len < IXGBE_RXBUFFER_4K)
2647 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2648 else if (rx_buf_len < IXGBE_RXBUFFER_8K)
2649 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2651 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2653 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
2657 * ixgbe_set_uta - Set unicast filter table address
2658 * @adapter: board private structure
2660 * The unicast table address is a register array of 32-bit registers.
2661 * The table is meant to be used in a way similar to how the MTA is used
2662 * however due to certain limitations in the hardware it is necessary to
2663 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
2664 * enable bit to allow vlan tag stripping when promiscuous mode is enabled
2666 static void ixgbe_set_uta(struct ixgbe_adapter *adapter)
2668 struct ixgbe_hw *hw = &adapter->hw;
2671 /* The UTA table only exists on 82599 hardware and newer */
2672 if (hw->mac.type < ixgbe_mac_82599EB)
2675 /* we only need to do this if VMDq is enabled */
2676 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
2679 for (i = 0; i < 128; i++)
2680 IXGBE_WRITE_REG(hw, IXGBE_UTA(i), ~0);
2683 #define IXGBE_MAX_RX_DESC_POLL 10
2684 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
2685 struct ixgbe_ring *ring)
2687 struct ixgbe_hw *hw = &adapter->hw;
2688 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
2690 u8 reg_idx = ring->reg_idx;
2692 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2693 if (hw->mac.type == ixgbe_mac_82598EB &&
2694 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2698 usleep_range(1000, 2000);
2699 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2700 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
2703 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
2704 "the polling period\n", reg_idx);
2708 void ixgbe_disable_rx_queue(struct ixgbe_adapter *adapter,
2709 struct ixgbe_ring *ring)
2711 struct ixgbe_hw *hw = &adapter->hw;
2712 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
2714 u8 reg_idx = ring->reg_idx;
2716 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2717 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
2719 /* write value back with RXDCTL.ENABLE bit cleared */
2720 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
2722 if (hw->mac.type == ixgbe_mac_82598EB &&
2723 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2726 /* the hardware may take up to 100us to really disable the rx queue */
2729 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2730 } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
2733 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not cleared within "
2734 "the polling period\n", reg_idx);
2738 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
2739 struct ixgbe_ring *ring)
2741 struct ixgbe_hw *hw = &adapter->hw;
2742 u64 rdba = ring->dma;
2744 u8 reg_idx = ring->reg_idx;
2746 /* disable queue to avoid issues while updating state */
2747 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2748 ixgbe_disable_rx_queue(adapter, ring);
2750 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
2751 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
2752 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
2753 ring->count * sizeof(union ixgbe_adv_rx_desc));
2754 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
2755 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
2756 ring->tail = hw->hw_addr + IXGBE_RDT(reg_idx);
2758 ixgbe_configure_srrctl(adapter, ring);
2759 ixgbe_configure_rscctl(adapter, ring);
2761 /* If operating in IOV mode set RLPML for X540 */
2762 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
2763 hw->mac.type == ixgbe_mac_X540) {
2764 rxdctl &= ~IXGBE_RXDCTL_RLPMLMASK;
2765 rxdctl |= ((ring->netdev->mtu + ETH_HLEN +
2766 ETH_FCS_LEN + VLAN_HLEN) | IXGBE_RXDCTL_RLPML_EN);
2769 if (hw->mac.type == ixgbe_mac_82598EB) {
2771 * enable cache line friendly hardware writes:
2772 * PTHRESH=32 descriptors (half the internal cache),
2773 * this also removes ugly rx_no_buffer_count increment
2774 * HTHRESH=4 descriptors (to minimize latency on fetch)
2775 * WTHRESH=8 burst writeback up to two cache lines
2777 rxdctl &= ~0x3FFFFF;
2781 /* enable receive descriptor ring */
2782 rxdctl |= IXGBE_RXDCTL_ENABLE;
2783 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
2785 ixgbe_rx_desc_queue_enable(adapter, ring);
2786 ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
2789 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
2791 struct ixgbe_hw *hw = &adapter->hw;
2794 /* PSRTYPE must be initialized in non 82598 adapters */
2795 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
2796 IXGBE_PSRTYPE_UDPHDR |
2797 IXGBE_PSRTYPE_IPV4HDR |
2798 IXGBE_PSRTYPE_L2HDR |
2799 IXGBE_PSRTYPE_IPV6HDR;
2801 if (hw->mac.type == ixgbe_mac_82598EB)
2804 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED)
2805 psrtype |= (adapter->num_rx_queues_per_pool << 29);
2807 for (p = 0; p < adapter->num_rx_pools; p++)
2808 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(adapter->num_vfs + p),
2812 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
2814 struct ixgbe_hw *hw = &adapter->hw;
2817 u32 reg_offset, vf_shift;
2821 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
2824 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
2825 vt_reg_bits = IXGBE_VMD_CTL_VMDQ_EN | IXGBE_VT_CTL_REPLEN;
2826 vt_reg_bits |= (adapter->num_vfs << IXGBE_VT_CTL_POOL_SHIFT);
2827 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl | vt_reg_bits);
2829 vf_shift = adapter->num_vfs % 32;
2830 reg_offset = (adapter->num_vfs > 32) ? 1 : 0;
2832 /* Enable only the PF's pool for Tx/Rx */
2833 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (1 << vf_shift));
2834 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), 0);
2835 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (1 << vf_shift));
2836 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), 0);
2837 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
2839 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
2840 hw->mac.ops.set_vmdq(hw, 0, adapter->num_vfs);
2843 * Set up VF register offsets for selected VT Mode,
2844 * i.e. 32 or 64 VFs for SR-IOV
2846 gcr_ext = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
2847 gcr_ext |= IXGBE_GCR_EXT_MSIX_EN;
2848 gcr_ext |= IXGBE_GCR_EXT_VT_MODE_64;
2849 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
2851 /* enable Tx loopback for VF/PF communication */
2852 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
2853 /* Enable MAC Anti-Spoofing */
2854 hw->mac.ops.set_mac_anti_spoofing(hw,
2855 (adapter->num_vfs != 0),
2857 /* For VFs that have spoof checking turned off */
2858 for (i = 0; i < adapter->num_vfs; i++) {
2859 if (!adapter->vfinfo[i].spoofchk_enabled)
2860 ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i, false);
2864 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
2866 struct ixgbe_hw *hw = &adapter->hw;
2867 struct net_device *netdev = adapter->netdev;
2868 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2870 struct ixgbe_ring *rx_ring;
2874 /* Decide whether to use packet split mode or not */
2876 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
2878 /* Do not use packet split if we're in SR-IOV Mode */
2879 if (adapter->num_vfs)
2880 adapter->flags &= ~IXGBE_FLAG_RX_PS_ENABLED;
2882 /* Disable packet split due to 82599 erratum #45 */
2883 if (hw->mac.type == ixgbe_mac_82599EB)
2884 adapter->flags &= ~IXGBE_FLAG_RX_PS_ENABLED;
2887 /* adjust max frame to be able to do baby jumbo for FCoE */
2888 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
2889 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
2890 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
2892 #endif /* IXGBE_FCOE */
2893 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
2894 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
2895 mhadd &= ~IXGBE_MHADD_MFS_MASK;
2896 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
2898 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
2901 /* MHADD will allow an extra 4 bytes past for vlan tagged frames */
2902 max_frame += VLAN_HLEN;
2904 /* Set the RX buffer length according to the mode */
2905 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
2906 rx_buf_len = IXGBE_RX_HDR_SIZE;
2908 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
2909 (netdev->mtu <= ETH_DATA_LEN))
2910 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
2912 * Make best use of allocation by using all but 1K of a
2913 * power of 2 allocation that will be used for skb->head.
2915 else if (max_frame <= IXGBE_RXBUFFER_3K)
2916 rx_buf_len = IXGBE_RXBUFFER_3K;
2917 else if (max_frame <= IXGBE_RXBUFFER_7K)
2918 rx_buf_len = IXGBE_RXBUFFER_7K;
2919 else if (max_frame <= IXGBE_RXBUFFER_15K)
2920 rx_buf_len = IXGBE_RXBUFFER_15K;
2922 rx_buf_len = IXGBE_MAX_RXBUFFER;
2925 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2926 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
2927 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2928 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
2931 * Setup the HW Rx Head and Tail Descriptor Pointers and
2932 * the Base and Length of the Rx Descriptor Ring
2934 for (i = 0; i < adapter->num_rx_queues; i++) {
2935 rx_ring = adapter->rx_ring[i];
2936 rx_ring->rx_buf_len = rx_buf_len;
2938 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)
2939 set_ring_ps_enabled(rx_ring);
2941 clear_ring_ps_enabled(rx_ring);
2943 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
2944 set_ring_rsc_enabled(rx_ring);
2946 clear_ring_rsc_enabled(rx_ring);
2949 if (netdev->features & NETIF_F_FCOE_MTU) {
2950 struct ixgbe_ring_feature *f;
2951 f = &adapter->ring_feature[RING_F_FCOE];
2952 if ((i >= f->mask) && (i < f->mask + f->indices)) {
2953 clear_ring_ps_enabled(rx_ring);
2954 if (rx_buf_len < IXGBE_FCOE_JUMBO_FRAME_SIZE)
2955 rx_ring->rx_buf_len =
2956 IXGBE_FCOE_JUMBO_FRAME_SIZE;
2957 } else if (!ring_is_rsc_enabled(rx_ring) &&
2958 !ring_is_ps_enabled(rx_ring)) {
2959 rx_ring->rx_buf_len =
2960 IXGBE_FCOE_JUMBO_FRAME_SIZE;
2963 #endif /* IXGBE_FCOE */
2967 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
2969 struct ixgbe_hw *hw = &adapter->hw;
2970 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2972 switch (hw->mac.type) {
2973 case ixgbe_mac_82598EB:
2975 * For VMDq support of different descriptor types or
2976 * buffer sizes through the use of multiple SRRCTL
2977 * registers, RDRXCTL.MVMEN must be set to 1
2979 * also, the manual doesn't mention it clearly but DCA hints
2980 * will only use queue 0's tags unless this bit is set. Side
2981 * effects of setting this bit are only that SRRCTL must be
2982 * fully programmed [0..15]
2984 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
2986 case ixgbe_mac_82599EB:
2987 case ixgbe_mac_X540:
2988 /* Disable RSC for ACK packets */
2989 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
2990 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
2991 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
2992 /* hardware requires some bits to be set by default */
2993 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
2994 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
2997 /* We should do nothing since we don't know this hardware */
3001 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3005 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3006 * @adapter: board private structure
3008 * Configure the Rx unit of the MAC after a reset.
3010 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3012 struct ixgbe_hw *hw = &adapter->hw;
3016 /* disable receives while setting up the descriptors */
3017 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3018 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3020 ixgbe_setup_psrtype(adapter);
3021 ixgbe_setup_rdrxctl(adapter);
3023 /* Program registers for the distribution of queues */
3024 ixgbe_setup_mrqc(adapter);
3026 ixgbe_set_uta(adapter);
3028 /* set_rx_buffer_len must be called before ring initialization */
3029 ixgbe_set_rx_buffer_len(adapter);
3032 * Setup the HW Rx Head and Tail Descriptor Pointers and
3033 * the Base and Length of the Rx Descriptor Ring
3035 for (i = 0; i < adapter->num_rx_queues; i++)
3036 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3038 /* disable drop enable for 82598 parts */
3039 if (hw->mac.type == ixgbe_mac_82598EB)
3040 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3042 /* enable all receives */
3043 rxctrl |= IXGBE_RXCTRL_RXEN;
3044 hw->mac.ops.enable_rx_dma(hw, rxctrl);
3047 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3049 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3050 struct ixgbe_hw *hw = &adapter->hw;
3051 int pool_ndx = adapter->num_vfs;
3053 /* add VID to filter table */
3054 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, true);
3055 set_bit(vid, adapter->active_vlans);
3060 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3062 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3063 struct ixgbe_hw *hw = &adapter->hw;
3064 int pool_ndx = adapter->num_vfs;
3066 /* remove VID from filter table */
3067 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, false);
3068 clear_bit(vid, adapter->active_vlans);
3074 * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3075 * @adapter: driver data
3077 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
3079 struct ixgbe_hw *hw = &adapter->hw;
3082 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3083 vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
3084 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3088 * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3089 * @adapter: driver data
3091 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
3093 struct ixgbe_hw *hw = &adapter->hw;
3096 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3097 vlnctrl |= IXGBE_VLNCTRL_VFE;
3098 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3099 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3103 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3104 * @adapter: driver data
3106 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3108 struct ixgbe_hw *hw = &adapter->hw;
3112 switch (hw->mac.type) {
3113 case ixgbe_mac_82598EB:
3114 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3115 vlnctrl &= ~IXGBE_VLNCTRL_VME;
3116 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3118 case ixgbe_mac_82599EB:
3119 case ixgbe_mac_X540:
3120 for (i = 0; i < adapter->num_rx_queues; i++) {
3121 j = adapter->rx_ring[i]->reg_idx;
3122 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3123 vlnctrl &= ~IXGBE_RXDCTL_VME;
3124 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3133 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3134 * @adapter: driver data
3136 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3138 struct ixgbe_hw *hw = &adapter->hw;
3142 switch (hw->mac.type) {
3143 case ixgbe_mac_82598EB:
3144 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3145 vlnctrl |= IXGBE_VLNCTRL_VME;
3146 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3148 case ixgbe_mac_82599EB:
3149 case ixgbe_mac_X540:
3150 for (i = 0; i < adapter->num_rx_queues; i++) {
3151 j = adapter->rx_ring[i]->reg_idx;
3152 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3153 vlnctrl |= IXGBE_RXDCTL_VME;
3154 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3162 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3166 ixgbe_vlan_rx_add_vid(adapter->netdev, 0);
3168 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3169 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
3173 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3174 * @netdev: network interface device structure
3176 * Writes unicast address list to the RAR table.
3177 * Returns: -ENOMEM on failure/insufficient address space
3178 * 0 on no addresses written
3179 * X on writing X addresses to the RAR table
3181 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3183 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3184 struct ixgbe_hw *hw = &adapter->hw;
3185 unsigned int vfn = adapter->num_vfs;
3186 unsigned int rar_entries = IXGBE_MAX_PF_MACVLANS;
3189 /* return ENOMEM indicating insufficient memory for addresses */
3190 if (netdev_uc_count(netdev) > rar_entries)
3193 if (!netdev_uc_empty(netdev) && rar_entries) {
3194 struct netdev_hw_addr *ha;
3195 /* return error if we do not support writing to RAR table */
3196 if (!hw->mac.ops.set_rar)
3199 netdev_for_each_uc_addr(ha, netdev) {
3202 hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3207 /* write the addresses in reverse order to avoid write combining */
3208 for (; rar_entries > 0 ; rar_entries--)
3209 hw->mac.ops.clear_rar(hw, rar_entries);
3215 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3216 * @netdev: network interface device structure
3218 * The set_rx_method entry point is called whenever the unicast/multicast
3219 * address list or the network interface flags are updated. This routine is
3220 * responsible for configuring the hardware for proper unicast, multicast and
3223 void ixgbe_set_rx_mode(struct net_device *netdev)
3225 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3226 struct ixgbe_hw *hw = &adapter->hw;
3227 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3230 /* Check for Promiscuous and All Multicast modes */
3232 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3234 /* set all bits that we expect to always be set */
3235 fctrl |= IXGBE_FCTRL_BAM;
3236 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3237 fctrl |= IXGBE_FCTRL_PMCF;
3239 /* clear the bits we are changing the status of */
3240 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3242 if (netdev->flags & IFF_PROMISC) {
3243 hw->addr_ctrl.user_set_promisc = true;
3244 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3245 vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3246 /* don't hardware filter vlans in promisc mode */
3247 ixgbe_vlan_filter_disable(adapter);
3249 if (netdev->flags & IFF_ALLMULTI) {
3250 fctrl |= IXGBE_FCTRL_MPE;
3251 vmolr |= IXGBE_VMOLR_MPE;
3254 * Write addresses to the MTA, if the attempt fails
3255 * then we should just turn on promiscuous mode so
3256 * that we can at least receive multicast traffic
3258 hw->mac.ops.update_mc_addr_list(hw, netdev);
3259 vmolr |= IXGBE_VMOLR_ROMPE;
3261 ixgbe_vlan_filter_enable(adapter);
3262 hw->addr_ctrl.user_set_promisc = false;
3264 * Write addresses to available RAR registers, if there is not
3265 * sufficient space to store all the addresses then enable
3266 * unicast promiscuous mode
3268 count = ixgbe_write_uc_addr_list(netdev);
3270 fctrl |= IXGBE_FCTRL_UPE;
3271 vmolr |= IXGBE_VMOLR_ROPE;
3275 if (adapter->num_vfs) {
3276 ixgbe_restore_vf_multicasts(adapter);
3277 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(adapter->num_vfs)) &
3278 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3280 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(adapter->num_vfs), vmolr);
3283 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3285 if (netdev->features & NETIF_F_HW_VLAN_RX)
3286 ixgbe_vlan_strip_enable(adapter);
3288 ixgbe_vlan_strip_disable(adapter);
3291 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3294 struct ixgbe_q_vector *q_vector;
3295 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3297 /* legacy and MSI only use one vector */
3298 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3301 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3302 q_vector = adapter->q_vector[q_idx];
3303 napi_enable(&q_vector->napi);
3307 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3310 struct ixgbe_q_vector *q_vector;
3311 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3313 /* legacy and MSI only use one vector */
3314 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3317 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3318 q_vector = adapter->q_vector[q_idx];
3319 napi_disable(&q_vector->napi);
3323 #ifdef CONFIG_IXGBE_DCB
3325 * ixgbe_configure_dcb - Configure DCB hardware
3326 * @adapter: ixgbe adapter struct
3328 * This is called by the driver on open to configure the DCB hardware.
3329 * This is also called by the gennetlink interface when reconfiguring
3332 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3334 struct ixgbe_hw *hw = &adapter->hw;
3335 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3337 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3338 if (hw->mac.type == ixgbe_mac_82598EB)
3339 netif_set_gso_max_size(adapter->netdev, 65536);
3343 if (hw->mac.type == ixgbe_mac_82598EB)
3344 netif_set_gso_max_size(adapter->netdev, 32768);
3347 /* Enable VLAN tag insert/strip */
3348 adapter->netdev->features |= NETIF_F_HW_VLAN_RX;
3350 hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
3353 if (adapter->netdev->features & NETIF_F_FCOE_MTU)
3354 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
3357 /* reconfigure the hardware */
3358 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
3359 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3361 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3363 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
3364 } else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
3365 ixgbe_dcb_hw_ets(&adapter->hw,
3366 adapter->ixgbe_ieee_ets,
3368 ixgbe_dcb_hw_pfc_config(&adapter->hw,
3369 adapter->ixgbe_ieee_pfc->pfc_en,
3370 adapter->ixgbe_ieee_ets->prio_tc);
3373 /* Enable RSS Hash per TC */
3374 if (hw->mac.type != ixgbe_mac_82598EB) {
3378 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
3380 u8 cnt = adapter->netdev->tc_to_txq[i].count;
3385 reg |= msb << IXGBE_RQTC_SHIFT_TC(i);
3387 IXGBE_WRITE_REG(hw, IXGBE_RQTC, reg);
3392 /* Additional bittime to account for IXGBE framing */
3393 #define IXGBE_ETH_FRAMING 20
3396 * ixgbe_hpbthresh - calculate high water mark for flow control
3398 * @adapter: board private structure to calculate for
3399 * @pb - packet buffer to calculate
3401 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
3403 struct ixgbe_hw *hw = &adapter->hw;
3404 struct net_device *dev = adapter->netdev;
3405 int link, tc, kb, marker;
3408 /* Calculate max LAN frame size */
3409 tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
3412 /* FCoE traffic class uses FCOE jumbo frames */
3413 if (dev->features & NETIF_F_FCOE_MTU) {
3416 #ifdef CONFIG_IXGBE_DCB
3417 fcoe_pb = netdev_get_prio_tc_map(dev, adapter->fcoe.up);
3420 if (fcoe_pb == pb && tc < IXGBE_FCOE_JUMBO_FRAME_SIZE)
3421 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3425 /* Calculate delay value for device */
3426 switch (hw->mac.type) {
3427 case ixgbe_mac_X540:
3428 dv_id = IXGBE_DV_X540(link, tc);
3431 dv_id = IXGBE_DV(link, tc);
3435 /* Loopback switch introduces additional latency */
3436 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3437 dv_id += IXGBE_B2BT(tc);
3439 /* Delay value is calculated in bit times convert to KB */
3440 kb = IXGBE_BT2KB(dv_id);
3441 rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
3443 marker = rx_pba - kb;
3445 /* It is possible that the packet buffer is not large enough
3446 * to provide required headroom. In this case throw an error
3447 * to user and a do the best we can.
3450 e_warn(drv, "Packet Buffer(%i) can not provide enough"
3451 "headroom to support flow control."
3452 "Decrease MTU or number of traffic classes\n", pb);
3460 * ixgbe_lpbthresh - calculate low water mark for for flow control
3462 * @adapter: board private structure to calculate for
3463 * @pb - packet buffer to calculate
3465 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter)
3467 struct ixgbe_hw *hw = &adapter->hw;
3468 struct net_device *dev = adapter->netdev;
3472 /* Calculate max LAN frame size */
3473 tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
3475 /* Calculate delay value for device */
3476 switch (hw->mac.type) {
3477 case ixgbe_mac_X540:
3478 dv_id = IXGBE_LOW_DV_X540(tc);
3481 dv_id = IXGBE_LOW_DV(tc);
3485 /* Delay value is calculated in bit times convert to KB */
3486 return IXGBE_BT2KB(dv_id);
3490 * ixgbe_pbthresh_setup - calculate and setup high low water marks
3492 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
3494 struct ixgbe_hw *hw = &adapter->hw;
3495 int num_tc = netdev_get_num_tc(adapter->netdev);
3501 hw->fc.low_water = ixgbe_lpbthresh(adapter);
3503 for (i = 0; i < num_tc; i++) {
3504 hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
3506 /* Low water marks must not be larger than high water marks */
3507 if (hw->fc.low_water > hw->fc.high_water[i])
3508 hw->fc.low_water = 0;
3512 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
3514 struct ixgbe_hw *hw = &adapter->hw;
3516 u8 tc = netdev_get_num_tc(adapter->netdev);
3518 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3519 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3520 hdrm = 32 << adapter->fdir_pballoc;
3524 hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
3525 ixgbe_pbthresh_setup(adapter);
3528 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
3530 struct ixgbe_hw *hw = &adapter->hw;
3531 struct hlist_node *node, *node2;
3532 struct ixgbe_fdir_filter *filter;
3534 spin_lock(&adapter->fdir_perfect_lock);
3536 if (!hlist_empty(&adapter->fdir_filter_list))
3537 ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
3539 hlist_for_each_entry_safe(filter, node, node2,
3540 &adapter->fdir_filter_list, fdir_node) {
3541 ixgbe_fdir_write_perfect_filter_82599(hw,
3544 (filter->action == IXGBE_FDIR_DROP_QUEUE) ?
3545 IXGBE_FDIR_DROP_QUEUE :
3546 adapter->rx_ring[filter->action]->reg_idx);
3549 spin_unlock(&adapter->fdir_perfect_lock);
3552 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3554 ixgbe_configure_pb(adapter);
3555 #ifdef CONFIG_IXGBE_DCB
3556 ixgbe_configure_dcb(adapter);
3559 ixgbe_set_rx_mode(adapter->netdev);
3560 ixgbe_restore_vlan(adapter);
3563 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
3564 ixgbe_configure_fcoe(adapter);
3566 #endif /* IXGBE_FCOE */
3567 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3568 ixgbe_init_fdir_signature_82599(&adapter->hw,
3569 adapter->fdir_pballoc);
3570 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3571 ixgbe_init_fdir_perfect_82599(&adapter->hw,
3572 adapter->fdir_pballoc);
3573 ixgbe_fdir_filter_restore(adapter);
3576 ixgbe_configure_virtualization(adapter);
3578 ixgbe_configure_tx(adapter);
3579 ixgbe_configure_rx(adapter);
3582 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3584 switch (hw->phy.type) {
3585 case ixgbe_phy_sfp_avago:
3586 case ixgbe_phy_sfp_ftl:
3587 case ixgbe_phy_sfp_intel:
3588 case ixgbe_phy_sfp_unknown:
3589 case ixgbe_phy_sfp_passive_tyco:
3590 case ixgbe_phy_sfp_passive_unknown:
3591 case ixgbe_phy_sfp_active_unknown:
3592 case ixgbe_phy_sfp_ftl_active:
3595 if (hw->mac.type == ixgbe_mac_82598EB)
3603 * ixgbe_sfp_link_config - set up SFP+ link
3604 * @adapter: pointer to private adapter struct
3606 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3609 * We are assuming the worst case scenario here, and that
3610 * is that an SFP was inserted/removed after the reset
3611 * but before SFP detection was enabled. As such the best
3612 * solution is to just start searching as soon as we start
3614 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3615 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
3617 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
3621 * ixgbe_non_sfp_link_config - set up non-SFP+ link
3622 * @hw: pointer to private hardware struct
3624 * Returns 0 on success, negative on failure
3626 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3629 bool negotiation, link_up = false;
3630 u32 ret = IXGBE_ERR_LINK_SETUP;
3632 if (hw->mac.ops.check_link)
3633 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3638 autoneg = hw->phy.autoneg_advertised;
3639 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
3640 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
3645 if (hw->mac.ops.setup_link)
3646 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
3651 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
3653 struct ixgbe_hw *hw = &adapter->hw;
3656 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3657 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
3659 gpie |= IXGBE_GPIE_EIAME;
3661 * use EIAM to auto-mask when MSI-X interrupt is asserted
3662 * this saves a register write for every interrupt
3664 switch (hw->mac.type) {
3665 case ixgbe_mac_82598EB:
3666 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3668 case ixgbe_mac_82599EB:
3669 case ixgbe_mac_X540:
3671 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
3672 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
3676 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
3677 * specifically only auto mask tx and rx interrupts */
3678 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3681 /* XXX: to interrupt immediately for EICS writes, enable this */
3682 /* gpie |= IXGBE_GPIE_EIMEN; */
3684 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3685 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
3686 gpie |= IXGBE_GPIE_VTMODE_64;
3689 /* Enable Thermal over heat sensor interrupt */
3690 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
3691 switch (adapter->hw.mac.type) {
3692 case ixgbe_mac_82599EB:
3693 gpie |= IXGBE_SDP0_GPIEN;
3695 case ixgbe_mac_X540:
3696 gpie |= IXGBE_EIMS_TS;
3703 /* Enable fan failure interrupt */
3704 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3705 gpie |= IXGBE_SDP1_GPIEN;
3707 if (hw->mac.type == ixgbe_mac_82599EB) {
3708 gpie |= IXGBE_SDP1_GPIEN;
3709 gpie |= IXGBE_SDP2_GPIEN;
3712 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3715 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
3717 struct ixgbe_hw *hw = &adapter->hw;
3721 ixgbe_get_hw_control(adapter);
3722 ixgbe_setup_gpie(adapter);
3724 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3725 ixgbe_configure_msix(adapter);
3727 ixgbe_configure_msi_and_legacy(adapter);
3729 /* enable the optics for both mult-speed fiber and 82599 SFP+ fiber */
3730 if (hw->mac.ops.enable_tx_laser &&
3731 ((hw->phy.multispeed_fiber) ||
3732 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
3733 (hw->mac.type == ixgbe_mac_82599EB))))
3734 hw->mac.ops.enable_tx_laser(hw);
3736 clear_bit(__IXGBE_DOWN, &adapter->state);
3737 ixgbe_napi_enable_all(adapter);
3739 if (ixgbe_is_sfp(hw)) {
3740 ixgbe_sfp_link_config(adapter);
3742 err = ixgbe_non_sfp_link_config(hw);
3744 e_err(probe, "link_config FAILED %d\n", err);
3747 /* clear any pending interrupts, may auto mask */
3748 IXGBE_READ_REG(hw, IXGBE_EICR);
3749 ixgbe_irq_enable(adapter, true, true);
3752 * If this adapter has a fan, check to see if we had a failure
3753 * before we enabled the interrupt.
3755 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3756 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
3757 if (esdp & IXGBE_ESDP_SDP1)
3758 e_crit(drv, "Fan has stopped, replace the adapter\n");
3761 /* enable transmits */
3762 netif_tx_start_all_queues(adapter->netdev);
3764 /* bring the link up in the watchdog, this could race with our first
3765 * link up interrupt but shouldn't be a problem */
3766 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3767 adapter->link_check_timeout = jiffies;
3768 mod_timer(&adapter->service_timer, jiffies);
3770 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
3771 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
3772 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
3773 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
3776 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
3778 WARN_ON(in_interrupt());
3779 /* put off any impending NetWatchDogTimeout */
3780 adapter->netdev->trans_start = jiffies;
3782 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
3783 usleep_range(1000, 2000);
3784 ixgbe_down(adapter);
3786 * If SR-IOV enabled then wait a bit before bringing the adapter
3787 * back up to give the VFs time to respond to the reset. The
3788 * two second wait is based upon the watchdog timer cycle in
3791 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3794 clear_bit(__IXGBE_RESETTING, &adapter->state);
3797 void ixgbe_up(struct ixgbe_adapter *adapter)
3799 /* hardware has been reset, we need to reload some things */
3800 ixgbe_configure(adapter);
3802 ixgbe_up_complete(adapter);
3805 void ixgbe_reset(struct ixgbe_adapter *adapter)
3807 struct ixgbe_hw *hw = &adapter->hw;
3810 /* lock SFP init bit to prevent race conditions with the watchdog */
3811 while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
3812 usleep_range(1000, 2000);
3814 /* clear all SFP and link config related flags while holding SFP_INIT */
3815 adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
3816 IXGBE_FLAG2_SFP_NEEDS_RESET);
3817 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
3819 err = hw->mac.ops.init_hw(hw);
3822 case IXGBE_ERR_SFP_NOT_PRESENT:
3823 case IXGBE_ERR_SFP_NOT_SUPPORTED:
3825 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
3826 e_dev_err("master disable timed out\n");
3828 case IXGBE_ERR_EEPROM_VERSION:
3829 /* We are running on a pre-production device, log a warning */
3830 e_dev_warn("This device is a pre-production adapter/LOM. "
3831 "Please be aware there may be issues associated with "
3832 "your hardware. If you are experiencing problems "
3833 "please contact your Intel or hardware "
3834 "representative who provided you with this "
3838 e_dev_err("Hardware Error: %d\n", err);
3841 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
3843 /* reprogram the RAR[0] in case user changed it. */
3844 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
3849 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
3850 * @rx_ring: ring to free buffers from
3852 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
3854 struct device *dev = rx_ring->dev;
3858 /* ring already cleared, nothing to do */
3859 if (!rx_ring->rx_buffer_info)
3862 /* Free all the Rx ring sk_buffs */
3863 for (i = 0; i < rx_ring->count; i++) {
3864 struct ixgbe_rx_buffer *rx_buffer_info;
3866 rx_buffer_info = &rx_ring->rx_buffer_info[i];
3867 if (rx_buffer_info->dma) {
3868 dma_unmap_single(rx_ring->dev, rx_buffer_info->dma,
3869 rx_ring->rx_buf_len,
3871 rx_buffer_info->dma = 0;
3873 if (rx_buffer_info->skb) {
3874 struct sk_buff *skb = rx_buffer_info->skb;
3875 rx_buffer_info->skb = NULL;
3877 struct sk_buff *this = skb;
3878 if (IXGBE_RSC_CB(this)->delay_unmap) {
3879 dma_unmap_single(dev,
3880 IXGBE_RSC_CB(this)->dma,
3881 rx_ring->rx_buf_len,
3883 IXGBE_RSC_CB(this)->dma = 0;
3884 IXGBE_RSC_CB(skb)->delay_unmap = false;
3887 dev_kfree_skb(this);
3890 if (!rx_buffer_info->page)
3892 if (rx_buffer_info->page_dma) {
3893 dma_unmap_page(dev, rx_buffer_info->page_dma,
3894 PAGE_SIZE / 2, DMA_FROM_DEVICE);
3895 rx_buffer_info->page_dma = 0;
3897 put_page(rx_buffer_info->page);
3898 rx_buffer_info->page = NULL;
3899 rx_buffer_info->page_offset = 0;
3902 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
3903 memset(rx_ring->rx_buffer_info, 0, size);
3905 /* Zero out the descriptor ring */
3906 memset(rx_ring->desc, 0, rx_ring->size);
3908 rx_ring->next_to_clean = 0;
3909 rx_ring->next_to_use = 0;
3913 * ixgbe_clean_tx_ring - Free Tx Buffers
3914 * @tx_ring: ring to be cleaned
3916 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
3918 struct ixgbe_tx_buffer *tx_buffer_info;
3922 /* ring already cleared, nothing to do */
3923 if (!tx_ring->tx_buffer_info)
3926 /* Free all the Tx ring sk_buffs */
3927 for (i = 0; i < tx_ring->count; i++) {
3928 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3929 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
3932 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
3933 memset(tx_ring->tx_buffer_info, 0, size);
3935 /* Zero out the descriptor ring */
3936 memset(tx_ring->desc, 0, tx_ring->size);
3938 tx_ring->next_to_use = 0;
3939 tx_ring->next_to_clean = 0;
3943 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
3944 * @adapter: board private structure
3946 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
3950 for (i = 0; i < adapter->num_rx_queues; i++)
3951 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
3955 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
3956 * @adapter: board private structure
3958 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
3962 for (i = 0; i < adapter->num_tx_queues; i++)
3963 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
3966 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
3968 struct hlist_node *node, *node2;
3969 struct ixgbe_fdir_filter *filter;
3971 spin_lock(&adapter->fdir_perfect_lock);
3973 hlist_for_each_entry_safe(filter, node, node2,
3974 &adapter->fdir_filter_list, fdir_node) {
3975 hlist_del(&filter->fdir_node);
3978 adapter->fdir_filter_count = 0;
3980 spin_unlock(&adapter->fdir_perfect_lock);
3983 void ixgbe_down(struct ixgbe_adapter *adapter)
3985 struct net_device *netdev = adapter->netdev;
3986 struct ixgbe_hw *hw = &adapter->hw;
3990 /* signal that we are down to the interrupt handler */
3991 set_bit(__IXGBE_DOWN, &adapter->state);
3993 /* disable receives */
3994 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3995 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3997 /* disable all enabled rx queues */
3998 for (i = 0; i < adapter->num_rx_queues; i++)
3999 /* this call also flushes the previous write */
4000 ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]);
4002 usleep_range(10000, 20000);
4004 netif_tx_stop_all_queues(netdev);
4006 /* call carrier off first to avoid false dev_watchdog timeouts */
4007 netif_carrier_off(netdev);
4008 netif_tx_disable(netdev);
4010 ixgbe_irq_disable(adapter);
4012 ixgbe_napi_disable_all(adapter);
4014 adapter->flags2 &= ~(IXGBE_FLAG2_FDIR_REQUIRES_REINIT |
4015 IXGBE_FLAG2_RESET_REQUESTED);
4016 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
4018 del_timer_sync(&adapter->service_timer);
4020 if (adapter->num_vfs) {
4021 /* Clear EITR Select mapping */
4022 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
4024 /* Mark all the VFs as inactive */
4025 for (i = 0 ; i < adapter->num_vfs; i++)
4026 adapter->vfinfo[i].clear_to_send = false;
4028 /* ping all the active vfs to let them know we are going down */
4029 ixgbe_ping_all_vfs(adapter);
4031 /* Disable all VFTE/VFRE TX/RX */
4032 ixgbe_disable_tx_rx(adapter);
4035 /* disable transmits in the hardware now that interrupts are off */
4036 for (i = 0; i < adapter->num_tx_queues; i++) {
4037 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
4038 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
4041 /* Disable the Tx DMA engine on 82599 and X540 */
4042 switch (hw->mac.type) {
4043 case ixgbe_mac_82599EB:
4044 case ixgbe_mac_X540:
4045 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
4046 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
4047 ~IXGBE_DMATXCTL_TE));
4053 if (!pci_channel_offline(adapter->pdev))
4054 ixgbe_reset(adapter);
4056 /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
4057 if (hw->mac.ops.disable_tx_laser &&
4058 ((hw->phy.multispeed_fiber) ||
4059 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
4060 (hw->mac.type == ixgbe_mac_82599EB))))
4061 hw->mac.ops.disable_tx_laser(hw);
4063 ixgbe_clean_all_tx_rings(adapter);
4064 ixgbe_clean_all_rx_rings(adapter);
4066 #ifdef CONFIG_IXGBE_DCA
4067 /* since we reset the hardware DCA settings were cleared */
4068 ixgbe_setup_dca(adapter);
4073 * ixgbe_poll - NAPI Rx polling callback
4074 * @napi: structure for representing this polling device
4075 * @budget: how many packets driver is allowed to clean
4077 * This function is used for legacy and MSI, NAPI mode
4079 static int ixgbe_poll(struct napi_struct *napi, int budget)
4081 struct ixgbe_q_vector *q_vector =
4082 container_of(napi, struct ixgbe_q_vector, napi);
4083 struct ixgbe_adapter *adapter = q_vector->adapter;
4084 struct ixgbe_ring *ring;
4085 int per_ring_budget;
4086 bool clean_complete = true;
4088 #ifdef CONFIG_IXGBE_DCA
4089 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
4090 ixgbe_update_dca(q_vector);
4093 for (ring = q_vector->tx.ring; ring != NULL; ring = ring->next)
4094 clean_complete &= !!ixgbe_clean_tx_irq(q_vector, ring);
4096 /* attempt to distribute budget to each queue fairly, but don't allow
4097 * the budget to go below 1 because we'll exit polling */
4098 if (q_vector->rx.count > 1)
4099 per_ring_budget = max(budget/q_vector->rx.count, 1);
4101 per_ring_budget = budget;
4103 for (ring = q_vector->rx.ring; ring != NULL; ring = ring->next)
4104 clean_complete &= ixgbe_clean_rx_irq(q_vector, ring,
4107 /* If all work not completed, return budget and keep polling */
4108 if (!clean_complete)
4111 /* all work done, exit the polling mode */
4112 napi_complete(napi);
4113 if (adapter->rx_itr_setting & 1)
4114 ixgbe_set_itr(q_vector);
4115 if (!test_bit(__IXGBE_DOWN, &adapter->state))
4116 ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
4122 * ixgbe_tx_timeout - Respond to a Tx Hang
4123 * @netdev: network interface device structure
4125 static void ixgbe_tx_timeout(struct net_device *netdev)
4127 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4129 /* Do the reset outside of interrupt context */
4130 ixgbe_tx_timeout_reset(adapter);
4134 * ixgbe_set_rss_queues: Allocate queues for RSS
4135 * @adapter: board private structure to initialize
4137 * This is our "base" multiqueue mode. RSS (Receive Side Scaling) will try
4138 * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
4141 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
4144 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
4146 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4148 adapter->num_rx_queues = f->indices;
4149 adapter->num_tx_queues = f->indices;
4159 * ixgbe_set_fdir_queues: Allocate queues for Flow Director
4160 * @adapter: board private structure to initialize
4162 * Flow Director is an advanced Rx filter, attempting to get Rx flows back
4163 * to the original CPU that initiated the Tx session. This runs in addition
4164 * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
4165 * Rx load across CPUs using RSS.
4168 static inline bool ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
4171 struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
4173 f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
4176 /* Flow Director must have RSS enabled */
4177 if ((adapter->flags & IXGBE_FLAG_RSS_ENABLED) &&
4178 (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)) {
4179 adapter->num_tx_queues = f_fdir->indices;
4180 adapter->num_rx_queues = f_fdir->indices;
4183 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4190 * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
4191 * @adapter: board private structure to initialize
4193 * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
4194 * The ring feature mask is not used as a mask for FCoE, as it can take any 8
4195 * rx queues out of the max number of rx queues, instead, it is used as the
4196 * index of the first rx queue used by FCoE.
4199 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
4201 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4203 if (!(adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
4206 f->indices = min((int)num_online_cpus(), f->indices);
4208 adapter->num_rx_queues = 1;
4209 adapter->num_tx_queues = 1;
4211 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4212 e_info(probe, "FCoE enabled with RSS\n");
4213 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)
4214 ixgbe_set_fdir_queues(adapter);
4216 ixgbe_set_rss_queues(adapter);
4219 /* adding FCoE rx rings to the end */
4220 f->mask = adapter->num_rx_queues;
4221 adapter->num_rx_queues += f->indices;
4222 adapter->num_tx_queues += f->indices;
4226 #endif /* IXGBE_FCOE */
4228 /* Artificial max queue cap per traffic class in DCB mode */
4229 #define DCB_QUEUE_CAP 8
4231 #ifdef CONFIG_IXGBE_DCB
4232 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
4234 int per_tc_q, q, i, offset = 0;
4235 struct net_device *dev = adapter->netdev;
4236 int tcs = netdev_get_num_tc(dev);
4241 /* Map queue offset and counts onto allocated tx queues */
4242 per_tc_q = min(dev->num_tx_queues / tcs, (unsigned int)DCB_QUEUE_CAP);
4243 q = min((int)num_online_cpus(), per_tc_q);
4245 for (i = 0; i < tcs; i++) {
4246 netdev_set_tc_queue(dev, i, q, offset);
4250 adapter->num_tx_queues = q * tcs;
4251 adapter->num_rx_queues = q * tcs;
4254 /* FCoE enabled queues require special configuration indexed
4255 * by feature specific indices and mask. Here we map FCoE
4256 * indices onto the DCB queue pairs allowing FCoE to own
4257 * configuration later.
4259 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
4261 struct ixgbe_ring_feature *f =
4262 &adapter->ring_feature[RING_F_FCOE];
4264 tc = netdev_get_prio_tc_map(dev, adapter->fcoe.up);
4265 f->indices = dev->tc_to_txq[tc].count;
4266 f->mask = dev->tc_to_txq[tc].offset;
4275 * ixgbe_set_sriov_queues: Allocate queues for IOV use
4276 * @adapter: board private structure to initialize
4278 * IOV doesn't actually use anything, so just NAK the
4279 * request for now and let the other queue routines
4280 * figure out what to do.
4282 static inline bool ixgbe_set_sriov_queues(struct ixgbe_adapter *adapter)
4288 * ixgbe_set_num_queues: Allocate queues for device, feature dependent
4289 * @adapter: board private structure to initialize
4291 * This is the top level queue allocation routine. The order here is very
4292 * important, starting with the "most" number of features turned on at once,
4293 * and ending with the smallest set of features. This way large combinations
4294 * can be allocated if they're turned on, and smaller combinations are the
4295 * fallthrough conditions.
4298 static int ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
4300 /* Start with base case */
4301 adapter->num_rx_queues = 1;
4302 adapter->num_tx_queues = 1;
4303 adapter->num_rx_pools = adapter->num_rx_queues;
4304 adapter->num_rx_queues_per_pool = 1;
4306 if (ixgbe_set_sriov_queues(adapter))
4309 #ifdef CONFIG_IXGBE_DCB
4310 if (ixgbe_set_dcb_queues(adapter))
4315 if (ixgbe_set_fcoe_queues(adapter))
4318 #endif /* IXGBE_FCOE */
4319 if (ixgbe_set_fdir_queues(adapter))
4322 if (ixgbe_set_rss_queues(adapter))
4325 /* fallback to base case */
4326 adapter->num_rx_queues = 1;
4327 adapter->num_tx_queues = 1;
4330 /* Notify the stack of the (possibly) reduced queue counts. */
4331 netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
4332 return netif_set_real_num_rx_queues(adapter->netdev,
4333 adapter->num_rx_queues);
4336 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
4339 int err, vector_threshold;
4341 /* We'll want at least 3 (vector_threshold):
4344 * 3) Other (Link Status Change, etc.)
4345 * 4) TCP Timer (optional)
4347 vector_threshold = MIN_MSIX_COUNT;
4349 /* The more we get, the more we will assign to Tx/Rx Cleanup
4350 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
4351 * Right now, we simply care about how many we'll get; we'll
4352 * set them up later while requesting irq's.
4354 while (vectors >= vector_threshold) {
4355 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
4357 if (!err) /* Success in acquiring all requested vectors. */
4360 vectors = 0; /* Nasty failure, quit now */
4361 else /* err == number of vectors we should try again with */
4365 if (vectors < vector_threshold) {
4366 /* Can't allocate enough MSI-X interrupts? Oh well.
4367 * This just means we'll go with either a single MSI
4368 * vector or fall back to legacy interrupts.
4370 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4371 "Unable to allocate MSI-X interrupts\n");
4372 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4373 kfree(adapter->msix_entries);
4374 adapter->msix_entries = NULL;
4376 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
4378 * Adjust for only the vectors we'll use, which is minimum
4379 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
4380 * vectors we were allocated.
4382 adapter->num_msix_vectors = min(vectors,
4383 adapter->max_msix_q_vectors + NON_Q_VECTORS);
4388 * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
4389 * @adapter: board private structure to initialize
4391 * Cache the descriptor ring offsets for RSS to the assigned rings.
4394 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
4398 if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED))
4401 for (i = 0; i < adapter->num_rx_queues; i++)
4402 adapter->rx_ring[i]->reg_idx = i;
4403 for (i = 0; i < adapter->num_tx_queues; i++)
4404 adapter->tx_ring[i]->reg_idx = i;
4409 #ifdef CONFIG_IXGBE_DCB
4411 /* ixgbe_get_first_reg_idx - Return first register index associated with ring */
4412 static void ixgbe_get_first_reg_idx(struct ixgbe_adapter *adapter, u8 tc,
4413 unsigned int *tx, unsigned int *rx)
4415 struct net_device *dev = adapter->netdev;
4416 struct ixgbe_hw *hw = &adapter->hw;
4417 u8 num_tcs = netdev_get_num_tc(dev);
4422 switch (hw->mac.type) {
4423 case ixgbe_mac_82598EB:
4427 case ixgbe_mac_82599EB:
4428 case ixgbe_mac_X540:
4433 } else if (tc < 5) {
4434 *tx = ((tc + 2) << 4);
4436 } else if (tc < num_tcs) {
4437 *tx = ((tc + 8) << 3);
4466 * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
4467 * @adapter: board private structure to initialize
4469 * Cache the descriptor ring offsets for DCB to the assigned rings.
4472 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
4474 struct net_device *dev = adapter->netdev;
4476 u8 num_tcs = netdev_get_num_tc(dev);
4481 for (i = 0, k = 0; i < num_tcs; i++) {
4482 unsigned int tx_s, rx_s;
4483 u16 count = dev->tc_to_txq[i].count;
4485 ixgbe_get_first_reg_idx(adapter, i, &tx_s, &rx_s);
4486 for (j = 0; j < count; j++, k++) {
4487 adapter->tx_ring[k]->reg_idx = tx_s + j;
4488 adapter->rx_ring[k]->reg_idx = rx_s + j;
4489 adapter->tx_ring[k]->dcb_tc = i;
4490 adapter->rx_ring[k]->dcb_tc = i;
4499 * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
4500 * @adapter: board private structure to initialize
4502 * Cache the descriptor ring offsets for Flow Director to the assigned rings.
4505 static inline bool ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
4510 if ((adapter->flags & IXGBE_FLAG_RSS_ENABLED) &&
4511 (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)) {
4512 for (i = 0; i < adapter->num_rx_queues; i++)
4513 adapter->rx_ring[i]->reg_idx = i;
4514 for (i = 0; i < adapter->num_tx_queues; i++)
4515 adapter->tx_ring[i]->reg_idx = i;
4524 * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
4525 * @adapter: board private structure to initialize
4527 * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
4530 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
4532 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4534 u8 fcoe_rx_i = 0, fcoe_tx_i = 0;
4536 if (!(adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
4539 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4540 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)
4541 ixgbe_cache_ring_fdir(adapter);
4543 ixgbe_cache_ring_rss(adapter);
4545 fcoe_rx_i = f->mask;
4546 fcoe_tx_i = f->mask;
4548 for (i = 0; i < f->indices; i++, fcoe_rx_i++, fcoe_tx_i++) {
4549 adapter->rx_ring[f->mask + i]->reg_idx = fcoe_rx_i;
4550 adapter->tx_ring[f->mask + i]->reg_idx = fcoe_tx_i;
4555 #endif /* IXGBE_FCOE */
4557 * ixgbe_cache_ring_sriov - Descriptor ring to register mapping for sriov
4558 * @adapter: board private structure to initialize
4560 * SR-IOV doesn't use any descriptor rings but changes the default if
4561 * no other mapping is used.
4564 static inline bool ixgbe_cache_ring_sriov(struct ixgbe_adapter *adapter)
4566 adapter->rx_ring[0]->reg_idx = adapter->num_vfs * 2;
4567 adapter->tx_ring[0]->reg_idx = adapter->num_vfs * 2;
4568 if (adapter->num_vfs)
4575 * ixgbe_cache_ring_register - Descriptor ring to register mapping
4576 * @adapter: board private structure to initialize
4578 * Once we know the feature-set enabled for the device, we'll cache
4579 * the register offset the descriptor ring is assigned to.
4581 * Note, the order the various feature calls is important. It must start with
4582 * the "most" features enabled at the same time, then trickle down to the
4583 * least amount of features turned on at once.
4585 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
4587 /* start with default case */
4588 adapter->rx_ring[0]->reg_idx = 0;
4589 adapter->tx_ring[0]->reg_idx = 0;
4591 if (ixgbe_cache_ring_sriov(adapter))
4594 #ifdef CONFIG_IXGBE_DCB
4595 if (ixgbe_cache_ring_dcb(adapter))
4600 if (ixgbe_cache_ring_fcoe(adapter))
4602 #endif /* IXGBE_FCOE */
4604 if (ixgbe_cache_ring_fdir(adapter))
4607 if (ixgbe_cache_ring_rss(adapter))
4612 * ixgbe_alloc_queues - Allocate memory for all rings
4613 * @adapter: board private structure to initialize
4615 * We allocate one ring per queue at run-time since we don't know the
4616 * number of queues at compile-time. The polling_netdev array is
4617 * intended for Multiqueue, but should work fine with a single queue.
4619 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
4621 int rx = 0, tx = 0, nid = adapter->node;
4623 if (nid < 0 || !node_online(nid))
4624 nid = first_online_node;
4626 for (; tx < adapter->num_tx_queues; tx++) {
4627 struct ixgbe_ring *ring;
4629 ring = kzalloc_node(sizeof(*ring), GFP_KERNEL, nid);
4631 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
4633 goto err_allocation;
4634 ring->count = adapter->tx_ring_count;
4635 ring->queue_index = tx;
4636 ring->numa_node = nid;
4637 ring->dev = &adapter->pdev->dev;
4638 ring->netdev = adapter->netdev;
4640 adapter->tx_ring[tx] = ring;
4643 for (; rx < adapter->num_rx_queues; rx++) {
4644 struct ixgbe_ring *ring;
4646 ring = kzalloc_node(sizeof(*ring), GFP_KERNEL, nid);
4648 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
4650 goto err_allocation;
4651 ring->count = adapter->rx_ring_count;
4652 ring->queue_index = rx;
4653 ring->numa_node = nid;
4654 ring->dev = &adapter->pdev->dev;
4655 ring->netdev = adapter->netdev;
4657 adapter->rx_ring[rx] = ring;
4660 ixgbe_cache_ring_register(adapter);
4666 kfree(adapter->tx_ring[--tx]);
4669 kfree(adapter->rx_ring[--rx]);
4674 * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
4675 * @adapter: board private structure to initialize
4677 * Attempt to configure the interrupts using the best available
4678 * capabilities of the hardware and the kernel.
4680 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
4682 struct ixgbe_hw *hw = &adapter->hw;
4684 int vector, v_budget;
4687 * It's easy to be greedy for MSI-X vectors, but it really
4688 * doesn't do us much good if we have a lot more vectors
4689 * than CPU's. So let's be conservative and only ask for
4690 * (roughly) the same number of vectors as there are CPU's.
4692 v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
4693 (int)num_online_cpus()) + NON_Q_VECTORS;
4696 * At the same time, hardware can only support a maximum of
4697 * hw.mac->max_msix_vectors vectors. With features
4698 * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
4699 * descriptor queues supported by our device. Thus, we cap it off in
4700 * those rare cases where the cpu count also exceeds our vector limit.
4702 v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
4704 /* A failure in MSI-X entry allocation isn't fatal, but it does
4705 * mean we disable MSI-X capabilities of the adapter. */
4706 adapter->msix_entries = kcalloc(v_budget,
4707 sizeof(struct msix_entry), GFP_KERNEL);
4708 if (adapter->msix_entries) {
4709 for (vector = 0; vector < v_budget; vector++)
4710 adapter->msix_entries[vector].entry = vector;
4712 ixgbe_acquire_msix_vectors(adapter, v_budget);
4714 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4718 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
4719 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
4720 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
4722 "ATR is not supported while multiple "
4723 "queues are disabled. Disabling Flow Director\n");
4725 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4726 adapter->atr_sample_rate = 0;
4727 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4728 ixgbe_disable_sriov(adapter);
4730 err = ixgbe_set_num_queues(adapter);
4734 err = pci_enable_msi(adapter->pdev);
4736 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
4738 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4739 "Unable to allocate MSI interrupt, "
4740 "falling back to legacy. Error: %d\n", err);
4750 * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
4751 * @adapter: board private structure to initialize
4753 * We allocate one q_vector per queue interrupt. If allocation fails we
4756 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
4758 int v_idx, num_q_vectors;
4759 struct ixgbe_q_vector *q_vector;
4761 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4762 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4766 for (v_idx = 0; v_idx < num_q_vectors; v_idx++) {
4767 q_vector = kzalloc_node(sizeof(struct ixgbe_q_vector),
4768 GFP_KERNEL, adapter->node);
4770 q_vector = kzalloc(sizeof(struct ixgbe_q_vector),
4775 q_vector->adapter = adapter;
4776 q_vector->v_idx = v_idx;
4778 /* Allocate the affinity_hint cpumask, configure the mask */
4779 if (!alloc_cpumask_var(&q_vector->affinity_mask, GFP_KERNEL))
4781 cpumask_set_cpu(v_idx, q_vector->affinity_mask);
4782 netif_napi_add(adapter->netdev, &q_vector->napi,
4784 adapter->q_vector[v_idx] = q_vector;
4792 q_vector = adapter->q_vector[v_idx];
4793 netif_napi_del(&q_vector->napi);
4794 free_cpumask_var(q_vector->affinity_mask);
4796 adapter->q_vector[v_idx] = NULL;
4802 * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
4803 * @adapter: board private structure to initialize
4805 * This function frees the memory allocated to the q_vectors. In addition if
4806 * NAPI is enabled it will delete any references to the NAPI struct prior
4807 * to freeing the q_vector.
4809 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
4811 int v_idx, num_q_vectors;
4813 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4814 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4818 for (v_idx = 0; v_idx < num_q_vectors; v_idx++) {
4819 struct ixgbe_q_vector *q_vector = adapter->q_vector[v_idx];
4820 adapter->q_vector[v_idx] = NULL;
4821 netif_napi_del(&q_vector->napi);
4822 free_cpumask_var(q_vector->affinity_mask);
4827 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
4829 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4830 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4831 pci_disable_msix(adapter->pdev);
4832 kfree(adapter->msix_entries);
4833 adapter->msix_entries = NULL;
4834 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
4835 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
4836 pci_disable_msi(adapter->pdev);
4841 * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
4842 * @adapter: board private structure to initialize
4844 * We determine which interrupt scheme to use based on...
4845 * - Kernel support (MSI, MSI-X)
4846 * - which can be user-defined (via MODULE_PARAM)
4847 * - Hardware queue count (num_*_queues)
4848 * - defined by miscellaneous hardware support/features (RSS, etc.)
4850 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
4854 /* Number of supported queues */
4855 err = ixgbe_set_num_queues(adapter);
4859 err = ixgbe_set_interrupt_capability(adapter);
4861 e_dev_err("Unable to setup interrupt capabilities\n");
4862 goto err_set_interrupt;
4865 err = ixgbe_alloc_q_vectors(adapter);
4867 e_dev_err("Unable to allocate memory for queue vectors\n");
4868 goto err_alloc_q_vectors;
4871 err = ixgbe_alloc_queues(adapter);
4873 e_dev_err("Unable to allocate memory for queues\n");
4874 goto err_alloc_queues;
4877 e_dev_info("Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
4878 (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
4879 adapter->num_rx_queues, adapter->num_tx_queues);
4881 set_bit(__IXGBE_DOWN, &adapter->state);
4886 ixgbe_free_q_vectors(adapter);
4887 err_alloc_q_vectors:
4888 ixgbe_reset_interrupt_capability(adapter);
4894 * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
4895 * @adapter: board private structure to clear interrupt scheme on
4897 * We go through and clear interrupt specific resources and reset the structure
4898 * to pre-load conditions
4900 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
4904 for (i = 0; i < adapter->num_tx_queues; i++) {
4905 kfree(adapter->tx_ring[i]);
4906 adapter->tx_ring[i] = NULL;
4908 for (i = 0; i < adapter->num_rx_queues; i++) {
4909 struct ixgbe_ring *ring = adapter->rx_ring[i];
4911 /* ixgbe_get_stats64() might access this ring, we must wait
4912 * a grace period before freeing it.
4914 kfree_rcu(ring, rcu);
4915 adapter->rx_ring[i] = NULL;
4918 adapter->num_tx_queues = 0;
4919 adapter->num_rx_queues = 0;
4921 ixgbe_free_q_vectors(adapter);
4922 ixgbe_reset_interrupt_capability(adapter);
4926 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4927 * @adapter: board private structure to initialize
4929 * ixgbe_sw_init initializes the Adapter private data structure.
4930 * Fields are initialized based on PCI device information and
4931 * OS network device settings (MTU size).
4933 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
4935 struct ixgbe_hw *hw = &adapter->hw;
4936 struct pci_dev *pdev = adapter->pdev;
4938 #ifdef CONFIG_IXGBE_DCB
4940 struct tc_configuration *tc;
4943 /* PCI config space info */
4945 hw->vendor_id = pdev->vendor;
4946 hw->device_id = pdev->device;
4947 hw->revision_id = pdev->revision;
4948 hw->subsystem_vendor_id = pdev->subsystem_vendor;
4949 hw->subsystem_device_id = pdev->subsystem_device;
4951 /* Set capability flags */
4952 rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
4953 adapter->ring_feature[RING_F_RSS].indices = rss;
4954 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
4955 switch (hw->mac.type) {
4956 case ixgbe_mac_82598EB:
4957 if (hw->device_id == IXGBE_DEV_ID_82598AT)
4958 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4959 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
4961 case ixgbe_mac_X540:
4962 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4963 case ixgbe_mac_82599EB:
4964 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
4965 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4966 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4967 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
4968 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4969 /* Flow Director hash filters enabled */
4970 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
4971 adapter->atr_sample_rate = 20;
4972 adapter->ring_feature[RING_F_FDIR].indices =
4973 IXGBE_MAX_FDIR_INDICES;
4974 adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
4976 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4977 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4978 adapter->ring_feature[RING_F_FCOE].indices = 0;
4979 #ifdef CONFIG_IXGBE_DCB
4980 /* Default traffic class to use for FCoE */
4981 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
4983 #endif /* IXGBE_FCOE */
4989 /* n-tuple support exists, always init our spinlock */
4990 spin_lock_init(&adapter->fdir_perfect_lock);
4992 #ifdef CONFIG_IXGBE_DCB
4993 switch (hw->mac.type) {
4994 case ixgbe_mac_X540:
4995 adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
4996 adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
4999 adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
5000 adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
5004 /* Configure DCB traffic classes */
5005 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
5006 tc = &adapter->dcb_cfg.tc_config[j];
5007 tc->path[DCB_TX_CONFIG].bwg_id = 0;
5008 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
5009 tc->path[DCB_RX_CONFIG].bwg_id = 0;
5010 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
5011 tc->dcb_pfc = pfc_disabled;
5014 /* Initialize default user to priority mapping, UPx->TC0 */
5015 tc = &adapter->dcb_cfg.tc_config[0];
5016 tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
5017 tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
5019 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
5020 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
5021 adapter->dcb_cfg.pfc_mode_enable = false;
5022 adapter->dcb_set_bitmap = 0x00;
5023 adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
5024 ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
5029 /* default flow control settings */
5030 hw->fc.requested_mode = ixgbe_fc_full;
5031 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
5033 adapter->last_lfc_mode = hw->fc.current_mode;
5035 ixgbe_pbthresh_setup(adapter);
5036 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
5037 hw->fc.send_xon = true;
5038 hw->fc.disable_fc_autoneg = false;
5040 /* enable itr by default in dynamic mode */
5041 adapter->rx_itr_setting = 1;
5042 adapter->tx_itr_setting = 1;
5044 /* set defaults for eitr in MegaBytes */
5045 adapter->eitr_low = 10;
5046 adapter->eitr_high = 20;
5048 /* set default ring sizes */
5049 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
5050 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
5052 /* set default work limits */
5053 adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
5055 /* initialize eeprom parameters */
5056 if (ixgbe_init_eeprom_params_generic(hw)) {
5057 e_dev_err("EEPROM initialization failed\n");
5061 /* enable rx csum by default */
5062 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
5064 /* get assigned NUMA node */
5065 adapter->node = dev_to_node(&pdev->dev);
5067 set_bit(__IXGBE_DOWN, &adapter->state);
5073 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
5074 * @tx_ring: tx descriptor ring (for a specific queue) to setup
5076 * Return 0 on success, negative on failure
5078 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
5080 struct device *dev = tx_ring->dev;
5083 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
5084 tx_ring->tx_buffer_info = vzalloc_node(size, tx_ring->numa_node);
5085 if (!tx_ring->tx_buffer_info)
5086 tx_ring->tx_buffer_info = vzalloc(size);
5087 if (!tx_ring->tx_buffer_info)
5090 /* round up to nearest 4K */
5091 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
5092 tx_ring->size = ALIGN(tx_ring->size, 4096);
5094 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
5095 &tx_ring->dma, GFP_KERNEL);
5099 tx_ring->next_to_use = 0;
5100 tx_ring->next_to_clean = 0;
5104 vfree(tx_ring->tx_buffer_info);
5105 tx_ring->tx_buffer_info = NULL;
5106 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
5111 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
5112 * @adapter: board private structure
5114 * If this function returns with an error, then it's possible one or
5115 * more of the rings is populated (while the rest are not). It is the
5116 * callers duty to clean those orphaned rings.
5118 * Return 0 on success, negative on failure
5120 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
5124 for (i = 0; i < adapter->num_tx_queues; i++) {
5125 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
5128 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
5136 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
5137 * @rx_ring: rx descriptor ring (for a specific queue) to setup
5139 * Returns 0 on success, negative on failure
5141 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
5143 struct device *dev = rx_ring->dev;
5146 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
5147 rx_ring->rx_buffer_info = vzalloc_node(size, rx_ring->numa_node);
5148 if (!rx_ring->rx_buffer_info)
5149 rx_ring->rx_buffer_info = vzalloc(size);
5150 if (!rx_ring->rx_buffer_info)
5153 /* Round up to nearest 4K */
5154 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
5155 rx_ring->size = ALIGN(rx_ring->size, 4096);
5157 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
5158 &rx_ring->dma, GFP_KERNEL);
5163 rx_ring->next_to_clean = 0;
5164 rx_ring->next_to_use = 0;
5168 vfree(rx_ring->rx_buffer_info);
5169 rx_ring->rx_buffer_info = NULL;
5170 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
5175 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
5176 * @adapter: board private structure
5178 * If this function returns with an error, then it's possible one or
5179 * more of the rings is populated (while the rest are not). It is the
5180 * callers duty to clean those orphaned rings.
5182 * Return 0 on success, negative on failure
5184 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
5188 for (i = 0; i < adapter->num_rx_queues; i++) {
5189 err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
5192 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
5200 * ixgbe_free_tx_resources - Free Tx Resources per Queue
5201 * @tx_ring: Tx descriptor ring for a specific queue
5203 * Free all transmit software resources
5205 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
5207 ixgbe_clean_tx_ring(tx_ring);
5209 vfree(tx_ring->tx_buffer_info);
5210 tx_ring->tx_buffer_info = NULL;
5212 /* if not set, then don't free */
5216 dma_free_coherent(tx_ring->dev, tx_ring->size,
5217 tx_ring->desc, tx_ring->dma);
5219 tx_ring->desc = NULL;
5223 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5224 * @adapter: board private structure
5226 * Free all transmit software resources
5228 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
5232 for (i = 0; i < adapter->num_tx_queues; i++)
5233 if (adapter->tx_ring[i]->desc)
5234 ixgbe_free_tx_resources(adapter->tx_ring[i]);
5238 * ixgbe_free_rx_resources - Free Rx Resources
5239 * @rx_ring: ring to clean the resources from
5241 * Free all receive software resources
5243 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
5245 ixgbe_clean_rx_ring(rx_ring);
5247 vfree(rx_ring->rx_buffer_info);
5248 rx_ring->rx_buffer_info = NULL;
5250 /* if not set, then don't free */
5254 dma_free_coherent(rx_ring->dev, rx_ring->size,
5255 rx_ring->desc, rx_ring->dma);
5257 rx_ring->desc = NULL;
5261 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5262 * @adapter: board private structure
5264 * Free all receive software resources
5266 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5270 for (i = 0; i < adapter->num_rx_queues; i++)
5271 if (adapter->rx_ring[i]->desc)
5272 ixgbe_free_rx_resources(adapter->rx_ring[i]);
5276 * ixgbe_change_mtu - Change the Maximum Transfer Unit
5277 * @netdev: network interface device structure
5278 * @new_mtu: new value for maximum frame size
5280 * Returns 0 on success, negative on failure
5282 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5284 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5285 struct ixgbe_hw *hw = &adapter->hw;
5286 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5288 /* MTU < 68 is an error and causes problems on some kernels */
5289 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED &&
5290 hw->mac.type != ixgbe_mac_X540) {
5291 if ((new_mtu < 68) || (max_frame > MAXIMUM_ETHERNET_VLAN_SIZE))
5294 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5298 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5299 /* must set new MTU before calling down or up */
5300 netdev->mtu = new_mtu;
5302 if (netif_running(netdev))
5303 ixgbe_reinit_locked(adapter);
5309 * ixgbe_open - Called when a network interface is made active
5310 * @netdev: network interface device structure
5312 * Returns 0 on success, negative value on failure
5314 * The open entry point is called when a network interface is made
5315 * active by the system (IFF_UP). At this point all resources needed
5316 * for transmit and receive operations are allocated, the interrupt
5317 * handler is registered with the OS, the watchdog timer is started,
5318 * and the stack is notified that the interface is ready.
5320 static int ixgbe_open(struct net_device *netdev)
5322 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5325 /* disallow open during test */
5326 if (test_bit(__IXGBE_TESTING, &adapter->state))
5329 netif_carrier_off(netdev);
5331 /* allocate transmit descriptors */
5332 err = ixgbe_setup_all_tx_resources(adapter);
5336 /* allocate receive descriptors */
5337 err = ixgbe_setup_all_rx_resources(adapter);
5341 ixgbe_configure(adapter);
5343 err = ixgbe_request_irq(adapter);
5347 ixgbe_up_complete(adapter);
5353 ixgbe_free_all_rx_resources(adapter);
5355 ixgbe_free_all_tx_resources(adapter);
5356 ixgbe_reset(adapter);
5362 * ixgbe_close - Disables a network interface
5363 * @netdev: network interface device structure
5365 * Returns 0, this is not allowed to fail
5367 * The close entry point is called when an interface is de-activated
5368 * by the OS. The hardware is still under the drivers control, but
5369 * needs to be disabled. A global MAC reset is issued to stop the
5370 * hardware, and all transmit and receive resources are freed.
5372 static int ixgbe_close(struct net_device *netdev)
5374 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5376 ixgbe_down(adapter);
5377 ixgbe_free_irq(adapter);
5379 ixgbe_fdir_filter_exit(adapter);
5381 ixgbe_free_all_tx_resources(adapter);
5382 ixgbe_free_all_rx_resources(adapter);
5384 ixgbe_release_hw_control(adapter);
5390 static int ixgbe_resume(struct pci_dev *pdev)
5392 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5393 struct net_device *netdev = adapter->netdev;
5396 pci_set_power_state(pdev, PCI_D0);
5397 pci_restore_state(pdev);
5399 * pci_restore_state clears dev->state_saved so call
5400 * pci_save_state to restore it.
5402 pci_save_state(pdev);
5404 err = pci_enable_device_mem(pdev);
5406 e_dev_err("Cannot enable PCI device from suspend\n");
5409 pci_set_master(pdev);
5411 pci_wake_from_d3(pdev, false);
5413 err = ixgbe_init_interrupt_scheme(adapter);
5415 e_dev_err("Cannot initialize interrupts for device\n");
5419 ixgbe_reset(adapter);
5421 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5423 if (netif_running(netdev)) {
5424 err = ixgbe_open(netdev);
5429 netif_device_attach(netdev);
5433 #endif /* CONFIG_PM */
5435 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5437 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5438 struct net_device *netdev = adapter->netdev;
5439 struct ixgbe_hw *hw = &adapter->hw;
5441 u32 wufc = adapter->wol;
5446 netif_device_detach(netdev);
5448 if (netif_running(netdev)) {
5449 ixgbe_down(adapter);
5450 ixgbe_free_irq(adapter);
5451 ixgbe_free_all_tx_resources(adapter);
5452 ixgbe_free_all_rx_resources(adapter);
5455 ixgbe_clear_interrupt_scheme(adapter);
5457 kfree(adapter->ixgbe_ieee_pfc);
5458 kfree(adapter->ixgbe_ieee_ets);
5462 retval = pci_save_state(pdev);
5468 ixgbe_set_rx_mode(netdev);
5470 /* turn on all-multi mode if wake on multicast is enabled */
5471 if (wufc & IXGBE_WUFC_MC) {
5472 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5473 fctrl |= IXGBE_FCTRL_MPE;
5474 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5477 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5478 ctrl |= IXGBE_CTRL_GIO_DIS;
5479 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5481 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5483 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5484 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5487 switch (hw->mac.type) {
5488 case ixgbe_mac_82598EB:
5489 pci_wake_from_d3(pdev, false);
5491 case ixgbe_mac_82599EB:
5492 case ixgbe_mac_X540:
5493 pci_wake_from_d3(pdev, !!wufc);
5499 *enable_wake = !!wufc;
5501 ixgbe_release_hw_control(adapter);
5503 pci_disable_device(pdev);
5509 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5514 retval = __ixgbe_shutdown(pdev, &wake);
5519 pci_prepare_to_sleep(pdev);
5521 pci_wake_from_d3(pdev, false);
5522 pci_set_power_state(pdev, PCI_D3hot);
5527 #endif /* CONFIG_PM */
5529 static void ixgbe_shutdown(struct pci_dev *pdev)
5533 __ixgbe_shutdown(pdev, &wake);
5535 if (system_state == SYSTEM_POWER_OFF) {
5536 pci_wake_from_d3(pdev, wake);
5537 pci_set_power_state(pdev, PCI_D3hot);
5542 * ixgbe_update_stats - Update the board statistics counters.
5543 * @adapter: board private structure
5545 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5547 struct net_device *netdev = adapter->netdev;
5548 struct ixgbe_hw *hw = &adapter->hw;
5549 struct ixgbe_hw_stats *hwstats = &adapter->stats;
5551 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5552 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
5553 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
5554 u64 bytes = 0, packets = 0;
5556 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
5558 u64 fcoe_noddp_counts_sum = 0, fcoe_noddp_ext_buff_counts_sum = 0;
5559 #endif /* IXGBE_FCOE */
5561 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5562 test_bit(__IXGBE_RESETTING, &adapter->state))
5565 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5568 for (i = 0; i < 16; i++)
5569 adapter->hw_rx_no_dma_resources +=
5570 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5571 for (i = 0; i < adapter->num_rx_queues; i++) {
5572 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
5573 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
5575 adapter->rsc_total_count = rsc_count;
5576 adapter->rsc_total_flush = rsc_flush;
5579 for (i = 0; i < adapter->num_rx_queues; i++) {
5580 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
5581 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
5582 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
5583 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
5584 bytes += rx_ring->stats.bytes;
5585 packets += rx_ring->stats.packets;
5587 adapter->non_eop_descs = non_eop_descs;
5588 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5589 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5590 netdev->stats.rx_bytes = bytes;
5591 netdev->stats.rx_packets = packets;
5595 /* gather some stats to the adapter struct that are per queue */
5596 for (i = 0; i < adapter->num_tx_queues; i++) {
5597 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5598 restart_queue += tx_ring->tx_stats.restart_queue;
5599 tx_busy += tx_ring->tx_stats.tx_busy;
5600 bytes += tx_ring->stats.bytes;
5601 packets += tx_ring->stats.packets;
5603 adapter->restart_queue = restart_queue;
5604 adapter->tx_busy = tx_busy;
5605 netdev->stats.tx_bytes = bytes;
5606 netdev->stats.tx_packets = packets;
5608 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5610 /* 8 register reads */
5611 for (i = 0; i < 8; i++) {
5612 /* for packet buffers not used, the register should read 0 */
5613 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5615 hwstats->mpc[i] += mpc;
5616 total_mpc += hwstats->mpc[i];
5617 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5618 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5619 switch (hw->mac.type) {
5620 case ixgbe_mac_82598EB:
5621 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5622 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5623 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5624 hwstats->pxonrxc[i] +=
5625 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5627 case ixgbe_mac_82599EB:
5628 case ixgbe_mac_X540:
5629 hwstats->pxonrxc[i] +=
5630 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5637 /*16 register reads */
5638 for (i = 0; i < 16; i++) {
5639 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5640 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5641 if ((hw->mac.type == ixgbe_mac_82599EB) ||
5642 (hw->mac.type == ixgbe_mac_X540)) {
5643 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
5644 IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
5645 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
5646 IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
5650 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5651 /* work around hardware counting issue */
5652 hwstats->gprc -= missed_rx;
5654 ixgbe_update_xoff_received(adapter);
5656 /* 82598 hardware only has a 32 bit counter in the high register */
5657 switch (hw->mac.type) {
5658 case ixgbe_mac_82598EB:
5659 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5660 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5661 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5662 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5664 case ixgbe_mac_X540:
5665 /* OS2BMC stats are X540 only*/
5666 hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
5667 hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
5668 hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
5669 hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
5670 case ixgbe_mac_82599EB:
5671 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5672 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
5673 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5674 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
5675 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5676 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5677 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5678 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5679 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5681 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5682 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5683 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5684 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5685 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5686 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5687 /* Add up per cpu counters for total ddp aloc fail */
5688 if (fcoe->pcpu_noddp && fcoe->pcpu_noddp_ext_buff) {
5689 for_each_possible_cpu(cpu) {
5690 fcoe_noddp_counts_sum +=
5691 *per_cpu_ptr(fcoe->pcpu_noddp, cpu);
5692 fcoe_noddp_ext_buff_counts_sum +=
5694 pcpu_noddp_ext_buff, cpu);
5697 hwstats->fcoe_noddp = fcoe_noddp_counts_sum;
5698 hwstats->fcoe_noddp_ext_buff = fcoe_noddp_ext_buff_counts_sum;
5699 #endif /* IXGBE_FCOE */
5704 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5705 hwstats->bprc += bprc;
5706 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5707 if (hw->mac.type == ixgbe_mac_82598EB)
5708 hwstats->mprc -= bprc;
5709 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5710 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5711 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5712 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5713 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5714 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5715 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5716 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5717 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5718 hwstats->lxontxc += lxon;
5719 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5720 hwstats->lxofftxc += lxoff;
5721 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5722 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5724 * 82598 errata - tx of flow control packets is included in tx counters
5726 xon_off_tot = lxon + lxoff;
5727 hwstats->gptc -= xon_off_tot;
5728 hwstats->mptc -= xon_off_tot;
5729 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5730 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5731 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5732 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5733 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5734 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5735 hwstats->ptc64 -= xon_off_tot;
5736 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5737 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5738 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5739 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5740 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5741 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5743 /* Fill out the OS statistics structure */
5744 netdev->stats.multicast = hwstats->mprc;
5747 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
5748 netdev->stats.rx_dropped = 0;
5749 netdev->stats.rx_length_errors = hwstats->rlec;
5750 netdev->stats.rx_crc_errors = hwstats->crcerrs;
5751 netdev->stats.rx_missed_errors = total_mpc;
5755 * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
5756 * @adapter - pointer to the device adapter structure
5758 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
5760 struct ixgbe_hw *hw = &adapter->hw;
5763 if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
5766 adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
5768 /* if interface is down do nothing */
5769 if (test_bit(__IXGBE_DOWN, &adapter->state))
5772 /* do nothing if we are not using signature filters */
5773 if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
5776 adapter->fdir_overflow++;
5778 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5779 for (i = 0; i < adapter->num_tx_queues; i++)
5780 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
5781 &(adapter->tx_ring[i]->state));
5782 /* re-enable flow director interrupts */
5783 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
5785 e_err(probe, "failed to finish FDIR re-initialization, "
5786 "ignored adding FDIR ATR filters\n");
5791 * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
5792 * @adapter - pointer to the device adapter structure
5794 * This function serves two purposes. First it strobes the interrupt lines
5795 * in order to make certain interrupts are occurring. Secondly it sets the
5796 * bits needed to check for TX hangs. As a result we should immediately
5797 * determine if a hang has occurred.
5799 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
5801 struct ixgbe_hw *hw = &adapter->hw;
5805 /* If we're down or resetting, just bail */
5806 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5807 test_bit(__IXGBE_RESETTING, &adapter->state))
5810 /* Force detection of hung controller */
5811 if (netif_carrier_ok(adapter->netdev)) {
5812 for (i = 0; i < adapter->num_tx_queues; i++)
5813 set_check_for_tx_hang(adapter->tx_ring[i]);
5816 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5818 * for legacy and MSI interrupts don't set any bits
5819 * that are enabled for EIAM, because this operation
5820 * would set *both* EIMS and EICS for any bit in EIAM
5822 IXGBE_WRITE_REG(hw, IXGBE_EICS,
5823 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5825 /* get one bit for every active tx/rx interrupt vector */
5826 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
5827 struct ixgbe_q_vector *qv = adapter->q_vector[i];
5828 if (qv->rx.ring || qv->tx.ring)
5829 eics |= ((u64)1 << i);
5833 /* Cause software interrupt to ensure rings are cleaned */
5834 ixgbe_irq_rearm_queues(adapter, eics);
5839 * ixgbe_watchdog_update_link - update the link status
5840 * @adapter - pointer to the device adapter structure
5841 * @link_speed - pointer to a u32 to store the link_speed
5843 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
5845 struct ixgbe_hw *hw = &adapter->hw;
5846 u32 link_speed = adapter->link_speed;
5847 bool link_up = adapter->link_up;
5850 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
5853 if (hw->mac.ops.check_link) {
5854 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5856 /* always assume link is up, if no check link function */
5857 link_speed = IXGBE_LINK_SPEED_10GB_FULL;
5861 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5862 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
5863 hw->mac.ops.fc_enable(hw, i);
5865 hw->mac.ops.fc_enable(hw, 0);
5870 time_after(jiffies, (adapter->link_check_timeout +
5871 IXGBE_TRY_LINK_TIMEOUT))) {
5872 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5873 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5874 IXGBE_WRITE_FLUSH(hw);
5877 adapter->link_up = link_up;
5878 adapter->link_speed = link_speed;
5882 * ixgbe_watchdog_link_is_up - update netif_carrier status and
5883 * print link up message
5884 * @adapter - pointer to the device adapter structure
5886 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
5888 struct net_device *netdev = adapter->netdev;
5889 struct ixgbe_hw *hw = &adapter->hw;
5890 u32 link_speed = adapter->link_speed;
5891 bool flow_rx, flow_tx;
5893 /* only continue if link was previously down */
5894 if (netif_carrier_ok(netdev))
5897 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
5899 switch (hw->mac.type) {
5900 case ixgbe_mac_82598EB: {
5901 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5902 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5903 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5904 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5907 case ixgbe_mac_X540:
5908 case ixgbe_mac_82599EB: {
5909 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5910 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5911 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5912 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5920 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
5921 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5923 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5925 (link_speed == IXGBE_LINK_SPEED_100_FULL ?
5928 ((flow_rx && flow_tx) ? "RX/TX" :
5930 (flow_tx ? "TX" : "None"))));
5932 netif_carrier_on(netdev);
5933 ixgbe_check_vf_rate_limit(adapter);
5937 * ixgbe_watchdog_link_is_down - update netif_carrier status and
5938 * print link down message
5939 * @adapter - pointer to the adapter structure
5941 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter* adapter)
5943 struct net_device *netdev = adapter->netdev;
5944 struct ixgbe_hw *hw = &adapter->hw;
5946 adapter->link_up = false;
5947 adapter->link_speed = 0;
5949 /* only continue if link was up previously */
5950 if (!netif_carrier_ok(netdev))
5953 /* poll for SFP+ cable when link is down */
5954 if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
5955 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5957 e_info(drv, "NIC Link is Down\n");
5958 netif_carrier_off(netdev);
5962 * ixgbe_watchdog_flush_tx - flush queues on link down
5963 * @adapter - pointer to the device adapter structure
5965 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
5968 int some_tx_pending = 0;
5970 if (!netif_carrier_ok(adapter->netdev)) {
5971 for (i = 0; i < adapter->num_tx_queues; i++) {
5972 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5973 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5974 some_tx_pending = 1;
5979 if (some_tx_pending) {
5980 /* We've lost link, so the controller stops DMA,
5981 * but we've got queued Tx work that's never going
5982 * to get done, so reset controller to flush Tx.
5983 * (Do the reset outside of interrupt context).
5985 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
5990 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
5994 /* Do not perform spoof check for 82598 */
5995 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
5998 ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
6001 * ssvpc register is cleared on read, if zero then no
6002 * spoofed packets in the last interval.
6007 e_warn(drv, "%d Spoofed packets detected\n", ssvpc);
6011 * ixgbe_watchdog_subtask - check and bring link up
6012 * @adapter - pointer to the device adapter structure
6014 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
6016 /* if interface is down do nothing */
6017 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6018 test_bit(__IXGBE_RESETTING, &adapter->state))
6021 ixgbe_watchdog_update_link(adapter);
6023 if (adapter->link_up)
6024 ixgbe_watchdog_link_is_up(adapter);
6026 ixgbe_watchdog_link_is_down(adapter);
6028 ixgbe_spoof_check(adapter);
6029 ixgbe_update_stats(adapter);
6031 ixgbe_watchdog_flush_tx(adapter);
6035 * ixgbe_sfp_detection_subtask - poll for SFP+ cable
6036 * @adapter - the ixgbe adapter structure
6038 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
6040 struct ixgbe_hw *hw = &adapter->hw;
6043 /* not searching for SFP so there is nothing to do here */
6044 if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
6045 !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
6048 /* someone else is in init, wait until next service event */
6049 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
6052 err = hw->phy.ops.identify_sfp(hw);
6053 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
6056 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
6057 /* If no cable is present, then we need to reset
6058 * the next time we find a good cable. */
6059 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
6066 /* exit if reset not needed */
6067 if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
6070 adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
6073 * A module may be identified correctly, but the EEPROM may not have
6074 * support for that module. setup_sfp() will fail in that case, so
6075 * we should not allow that module to load.
6077 if (hw->mac.type == ixgbe_mac_82598EB)
6078 err = hw->phy.ops.reset(hw);
6080 err = hw->mac.ops.setup_sfp(hw);
6082 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
6085 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
6086 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
6089 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
6091 if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
6092 (adapter->netdev->reg_state == NETREG_REGISTERED)) {
6093 e_dev_err("failed to initialize because an unsupported "
6094 "SFP+ module type was detected.\n");
6095 e_dev_err("Reload the driver after installing a "
6096 "supported module.\n");
6097 unregister_netdev(adapter->netdev);
6102 * ixgbe_sfp_link_config_subtask - set up link SFP after module install
6103 * @adapter - the ixgbe adapter structure
6105 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
6107 struct ixgbe_hw *hw = &adapter->hw;
6111 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
6114 /* someone else is in init, wait until next service event */
6115 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
6118 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
6120 autoneg = hw->phy.autoneg_advertised;
6121 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
6122 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
6123 if (hw->mac.ops.setup_link)
6124 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
6126 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
6127 adapter->link_check_timeout = jiffies;
6128 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
6131 #ifdef CONFIG_PCI_IOV
6132 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
6135 struct ixgbe_hw *hw = &adapter->hw;
6136 struct net_device *netdev = adapter->netdev;
6140 gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
6141 if (gpc) /* If incrementing then no need for the check below */
6144 * Check to see if a bad DMA write target from an errant or
6145 * malicious VF has caused a PCIe error. If so then we can
6146 * issue a VFLR to the offending VF(s) and then resume without
6147 * requesting a full slot reset.
6150 for (vf = 0; vf < adapter->num_vfs; vf++) {
6151 ciaa = (vf << 16) | 0x80000000;
6152 /* 32 bit read so align, we really want status at offset 6 */
6153 ciaa |= PCI_COMMAND;
6154 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
6155 ciad = IXGBE_READ_REG(hw, IXGBE_CIAD_82599);
6157 /* disable debug mode asap after reading data */
6158 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
6159 /* Get the upper 16 bits which will be the PCI status reg */
6161 if (ciad & PCI_STATUS_REC_MASTER_ABORT) {
6162 netdev_err(netdev, "VF %d Hung DMA\n", vf);
6164 ciaa = (vf << 16) | 0x80000000;
6166 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
6167 ciad = 0x00008000; /* VFLR */
6168 IXGBE_WRITE_REG(hw, IXGBE_CIAD_82599, ciad);
6170 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
6177 * ixgbe_service_timer - Timer Call-back
6178 * @data: pointer to adapter cast into an unsigned long
6180 static void ixgbe_service_timer(unsigned long data)
6182 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
6183 unsigned long next_event_offset;
6186 #ifdef CONFIG_PCI_IOV
6190 * don't bother with SR-IOV VF DMA hang check if there are
6191 * no VFs or the link is down
6193 if (!adapter->num_vfs ||
6194 (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)) {
6196 goto normal_timer_service;
6199 /* If we have VFs allocated then we must check for DMA hangs */
6200 ixgbe_check_for_bad_vf(adapter);
6201 next_event_offset = HZ / 50;
6202 adapter->timer_event_accumulator++;
6204 if (adapter->timer_event_accumulator >= 100) {
6206 adapter->timer_event_accumulator = 0;
6209 goto schedule_event;
6211 normal_timer_service:
6213 /* poll faster when waiting for link */
6214 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
6215 next_event_offset = HZ / 10;
6217 next_event_offset = HZ * 2;
6219 #ifdef CONFIG_PCI_IOV
6222 /* Reset the timer */
6223 mod_timer(&adapter->service_timer, next_event_offset + jiffies);
6226 ixgbe_service_event_schedule(adapter);
6229 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
6231 if (!(adapter->flags2 & IXGBE_FLAG2_RESET_REQUESTED))
6234 adapter->flags2 &= ~IXGBE_FLAG2_RESET_REQUESTED;
6236 /* If we're already down or resetting, just bail */
6237 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6238 test_bit(__IXGBE_RESETTING, &adapter->state))
6241 ixgbe_dump(adapter);
6242 netdev_err(adapter->netdev, "Reset adapter\n");
6243 adapter->tx_timeout_count++;
6245 ixgbe_reinit_locked(adapter);
6249 * ixgbe_service_task - manages and runs subtasks
6250 * @work: pointer to work_struct containing our data
6252 static void ixgbe_service_task(struct work_struct *work)
6254 struct ixgbe_adapter *adapter = container_of(work,
6255 struct ixgbe_adapter,
6258 ixgbe_reset_subtask(adapter);
6259 ixgbe_sfp_detection_subtask(adapter);
6260 ixgbe_sfp_link_config_subtask(adapter);
6261 ixgbe_check_overtemp_subtask(adapter);
6262 ixgbe_watchdog_subtask(adapter);
6263 ixgbe_fdir_reinit_subtask(adapter);
6264 ixgbe_check_hang_subtask(adapter);
6266 ixgbe_service_event_complete(adapter);
6269 void ixgbe_tx_ctxtdesc(struct ixgbe_ring *tx_ring, u32 vlan_macip_lens,
6270 u32 fcoe_sof_eof, u32 type_tucmd, u32 mss_l4len_idx)
6272 struct ixgbe_adv_tx_context_desc *context_desc;
6273 u16 i = tx_ring->next_to_use;
6275 context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
6278 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
6280 /* set bits to identify this as an advanced context descriptor */
6281 type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
6283 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
6284 context_desc->seqnum_seed = cpu_to_le32(fcoe_sof_eof);
6285 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd);
6286 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
6289 static int ixgbe_tso(struct ixgbe_ring *tx_ring, struct sk_buff *skb,
6290 u32 tx_flags, __be16 protocol, u8 *hdr_len)
6293 u32 vlan_macip_lens, type_tucmd;
6294 u32 mss_l4len_idx, l4len;
6296 if (!skb_is_gso(skb))
6299 if (skb_header_cloned(skb)) {
6300 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
6305 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
6306 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
6308 if (protocol == __constant_htons(ETH_P_IP)) {
6309 struct iphdr *iph = ip_hdr(skb);
6312 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
6316 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
6317 } else if (skb_is_gso_v6(skb)) {
6318 ipv6_hdr(skb)->payload_len = 0;
6319 tcp_hdr(skb)->check =
6320 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
6321 &ipv6_hdr(skb)->daddr,
6325 l4len = tcp_hdrlen(skb);
6326 *hdr_len = skb_transport_offset(skb) + l4len;
6328 /* mss_l4len_id: use 1 as index for TSO */
6329 mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
6330 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
6331 mss_l4len_idx |= 1 << IXGBE_ADVTXD_IDX_SHIFT;
6333 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
6334 vlan_macip_lens = skb_network_header_len(skb);
6335 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
6336 vlan_macip_lens |= tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
6338 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0, type_tucmd,
6344 static bool ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
6345 struct sk_buff *skb, u32 tx_flags,
6348 u32 vlan_macip_lens = 0;
6349 u32 mss_l4len_idx = 0;
6352 if (skb->ip_summed != CHECKSUM_PARTIAL) {
6353 if (!(tx_flags & IXGBE_TX_FLAGS_HW_VLAN) &&
6354 !(tx_flags & IXGBE_TX_FLAGS_TXSW))
6359 case __constant_htons(ETH_P_IP):
6360 vlan_macip_lens |= skb_network_header_len(skb);
6361 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
6362 l4_hdr = ip_hdr(skb)->protocol;
6364 case __constant_htons(ETH_P_IPV6):
6365 vlan_macip_lens |= skb_network_header_len(skb);
6366 l4_hdr = ipv6_hdr(skb)->nexthdr;
6369 if (unlikely(net_ratelimit())) {
6370 dev_warn(tx_ring->dev,
6371 "partial checksum but proto=%x!\n",
6379 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6380 mss_l4len_idx = tcp_hdrlen(skb) <<
6381 IXGBE_ADVTXD_L4LEN_SHIFT;
6384 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
6385 mss_l4len_idx = sizeof(struct sctphdr) <<
6386 IXGBE_ADVTXD_L4LEN_SHIFT;
6389 mss_l4len_idx = sizeof(struct udphdr) <<
6390 IXGBE_ADVTXD_L4LEN_SHIFT;
6393 if (unlikely(net_ratelimit())) {
6394 dev_warn(tx_ring->dev,
6395 "partial checksum but l4 proto=%x!\n",
6402 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
6403 vlan_macip_lens |= tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
6405 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0,
6406 type_tucmd, mss_l4len_idx);
6408 return (skb->ip_summed == CHECKSUM_PARTIAL);
6411 static __le32 ixgbe_tx_cmd_type(u32 tx_flags)
6413 /* set type for advanced descriptor with frame checksum insertion */
6414 __le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
6415 IXGBE_ADVTXD_DCMD_IFCS |
6416 IXGBE_ADVTXD_DCMD_DEXT);
6418 /* set HW vlan bit if vlan is present */
6419 if (tx_flags & IXGBE_TX_FLAGS_HW_VLAN)
6420 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
6422 /* set segmentation enable bits for TSO/FSO */
6424 if ((tx_flags & IXGBE_TX_FLAGS_TSO) || (tx_flags & IXGBE_TX_FLAGS_FSO))
6426 if (tx_flags & IXGBE_TX_FLAGS_TSO)
6428 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
6433 static __le32 ixgbe_tx_olinfo_status(u32 tx_flags, unsigned int paylen)
6435 __le32 olinfo_status =
6436 cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
6438 if (tx_flags & IXGBE_TX_FLAGS_TSO) {
6439 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM |
6440 (1 << IXGBE_ADVTXD_IDX_SHIFT));
6441 /* enble IPv4 checksum for TSO */
6442 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
6443 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
6446 /* enable L4 checksum for TSO and TX checksum offload */
6447 if (tx_flags & IXGBE_TX_FLAGS_CSUM)
6448 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
6451 /* use index 1 context for FCOE/FSO */
6452 if (tx_flags & IXGBE_TX_FLAGS_FCOE)
6453 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC |
6454 (1 << IXGBE_ADVTXD_IDX_SHIFT));
6458 * Check Context must be set if Tx switch is enabled, which it
6459 * always is for case where virtual functions are running
6461 if (tx_flags & IXGBE_TX_FLAGS_TXSW)
6462 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
6464 return olinfo_status;
6467 #define IXGBE_TXD_CMD (IXGBE_TXD_CMD_EOP | \
6470 static void ixgbe_tx_map(struct ixgbe_ring *tx_ring,
6471 struct sk_buff *skb,
6472 struct ixgbe_tx_buffer *first,
6476 struct device *dev = tx_ring->dev;
6477 struct ixgbe_tx_buffer *tx_buffer_info;
6478 union ixgbe_adv_tx_desc *tx_desc;
6480 __le32 cmd_type, olinfo_status;
6481 struct skb_frag_struct *frag;
6483 unsigned int data_len = skb->data_len;
6484 unsigned int size = skb_headlen(skb);
6486 u32 paylen = skb->len - hdr_len;
6487 u16 i = tx_ring->next_to_use;
6491 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6492 if (data_len >= sizeof(struct fcoe_crc_eof)) {
6493 data_len -= sizeof(struct fcoe_crc_eof);
6495 size -= sizeof(struct fcoe_crc_eof) - data_len;
6501 dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE);
6502 if (dma_mapping_error(dev, dma))
6505 cmd_type = ixgbe_tx_cmd_type(tx_flags);
6506 olinfo_status = ixgbe_tx_olinfo_status(tx_flags, paylen);
6508 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
6511 while (size > IXGBE_MAX_DATA_PER_TXD) {
6512 tx_desc->read.buffer_addr = cpu_to_le64(dma + offset);
6513 tx_desc->read.cmd_type_len =
6514 cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
6515 tx_desc->read.olinfo_status = olinfo_status;
6517 offset += IXGBE_MAX_DATA_PER_TXD;
6518 size -= IXGBE_MAX_DATA_PER_TXD;
6522 if (i == tx_ring->count) {
6523 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, 0);
6528 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6529 tx_buffer_info->length = offset + size;
6530 tx_buffer_info->tx_flags = tx_flags;
6531 tx_buffer_info->dma = dma;
6533 tx_desc->read.buffer_addr = cpu_to_le64(dma + offset);
6534 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
6535 tx_desc->read.olinfo_status = olinfo_status;
6540 frag = &skb_shinfo(skb)->frags[f];
6542 size = min_t(unsigned int, data_len, skb_frag_size(frag));
6544 size = skb_frag_size(frag);
6550 tx_flags |= IXGBE_TX_FLAGS_MAPPED_AS_PAGE;
6552 dma = skb_frag_dma_map(dev, frag, 0, size, DMA_TO_DEVICE);
6553 if (dma_mapping_error(dev, dma))
6558 if (i == tx_ring->count) {
6559 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, 0);
6564 tx_desc->read.cmd_type_len |= cpu_to_le32(IXGBE_TXD_CMD);
6567 if (i == tx_ring->count)
6570 tx_ring->next_to_use = i;
6572 if (tx_flags & IXGBE_TX_FLAGS_TSO)
6573 gso_segs = skb_shinfo(skb)->gso_segs;
6575 /* adjust for FCoE Sequence Offload */
6576 else if (tx_flags & IXGBE_TX_FLAGS_FSO)
6577 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
6578 skb_shinfo(skb)->gso_size);
6579 #endif /* IXGBE_FCOE */
6583 /* multiply data chunks by size of headers */
6584 tx_buffer_info->bytecount = paylen + (gso_segs * hdr_len);
6585 tx_buffer_info->gso_segs = gso_segs;
6586 tx_buffer_info->skb = skb;
6588 /* set the timestamp */
6589 first->time_stamp = jiffies;
6592 * Force memory writes to complete before letting h/w
6593 * know there are new descriptors to fetch. (Only
6594 * applicable for weak-ordered memory model archs,
6599 /* set next_to_watch value indicating a packet is present */
6600 first->next_to_watch = tx_desc;
6602 /* notify HW of packet */
6603 writel(i, tx_ring->tail);
6607 dev_err(dev, "TX DMA map failed\n");
6609 /* clear dma mappings for failed tx_buffer_info map */
6611 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6612 ixgbe_unmap_tx_resource(tx_ring, tx_buffer_info);
6613 if (tx_buffer_info == first)
6620 dev_kfree_skb_any(skb);
6622 tx_ring->next_to_use = i;
6625 static void ixgbe_atr(struct ixgbe_ring *ring, struct sk_buff *skb,
6626 u32 tx_flags, __be16 protocol)
6628 struct ixgbe_q_vector *q_vector = ring->q_vector;
6629 union ixgbe_atr_hash_dword input = { .dword = 0 };
6630 union ixgbe_atr_hash_dword common = { .dword = 0 };
6632 unsigned char *network;
6634 struct ipv6hdr *ipv6;
6639 /* if ring doesn't have a interrupt vector, cannot perform ATR */
6643 /* do nothing if sampling is disabled */
6644 if (!ring->atr_sample_rate)
6649 /* snag network header to get L4 type and address */
6650 hdr.network = skb_network_header(skb);
6652 /* Currently only IPv4/IPv6 with TCP is supported */
6653 if ((protocol != __constant_htons(ETH_P_IPV6) ||
6654 hdr.ipv6->nexthdr != IPPROTO_TCP) &&
6655 (protocol != __constant_htons(ETH_P_IP) ||
6656 hdr.ipv4->protocol != IPPROTO_TCP))
6661 /* skip this packet since it is invalid or the socket is closing */
6665 /* sample on all syn packets or once every atr sample count */
6666 if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
6669 /* reset sample count */
6670 ring->atr_count = 0;
6672 vlan_id = htons(tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
6675 * src and dst are inverted, think how the receiver sees them
6677 * The input is broken into two sections, a non-compressed section
6678 * containing vm_pool, vlan_id, and flow_type. The rest of the data
6679 * is XORed together and stored in the compressed dword.
6681 input.formatted.vlan_id = vlan_id;
6684 * since src port and flex bytes occupy the same word XOR them together
6685 * and write the value to source port portion of compressed dword
6687 if (tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
6688 common.port.src ^= th->dest ^ __constant_htons(ETH_P_8021Q);
6690 common.port.src ^= th->dest ^ protocol;
6691 common.port.dst ^= th->source;
6693 if (protocol == __constant_htons(ETH_P_IP)) {
6694 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
6695 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
6697 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
6698 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
6699 hdr.ipv6->saddr.s6_addr32[1] ^
6700 hdr.ipv6->saddr.s6_addr32[2] ^
6701 hdr.ipv6->saddr.s6_addr32[3] ^
6702 hdr.ipv6->daddr.s6_addr32[0] ^
6703 hdr.ipv6->daddr.s6_addr32[1] ^
6704 hdr.ipv6->daddr.s6_addr32[2] ^
6705 hdr.ipv6->daddr.s6_addr32[3];
6708 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
6709 ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
6710 input, common, ring->queue_index);
6713 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6715 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
6716 /* Herbert's original patch had:
6717 * smp_mb__after_netif_stop_queue();
6718 * but since that doesn't exist yet, just open code it. */
6721 /* We need to check again in a case another CPU has just
6722 * made room available. */
6723 if (likely(ixgbe_desc_unused(tx_ring) < size))
6726 /* A reprieve! - use start_queue because it doesn't call schedule */
6727 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
6728 ++tx_ring->tx_stats.restart_queue;
6732 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6734 if (likely(ixgbe_desc_unused(tx_ring) >= size))
6736 return __ixgbe_maybe_stop_tx(tx_ring, size);
6739 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6741 struct ixgbe_adapter *adapter = netdev_priv(dev);
6742 int txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
6745 __be16 protocol = vlan_get_protocol(skb);
6747 if (((protocol == htons(ETH_P_FCOE)) ||
6748 (protocol == htons(ETH_P_FIP))) &&
6749 (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)) {
6750 txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
6751 txq += adapter->ring_feature[RING_F_FCOE].mask;
6756 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6757 while (unlikely(txq >= dev->real_num_tx_queues))
6758 txq -= dev->real_num_tx_queues;
6762 return skb_tx_hash(dev, skb);
6765 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
6766 struct ixgbe_adapter *adapter,
6767 struct ixgbe_ring *tx_ring)
6769 struct ixgbe_tx_buffer *first;
6772 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6775 u16 count = TXD_USE_COUNT(skb_headlen(skb));
6776 __be16 protocol = skb->protocol;
6780 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
6781 * + 1 desc for skb_head_len/IXGBE_MAX_DATA_PER_TXD,
6782 * + 2 desc gap to keep tail from touching head,
6783 * + 1 desc for context descriptor,
6784 * otherwise try next time
6786 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6787 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6788 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6790 count += skb_shinfo(skb)->nr_frags;
6792 if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
6793 tx_ring->tx_stats.tx_busy++;
6794 return NETDEV_TX_BUSY;
6797 #ifdef CONFIG_PCI_IOV
6798 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6799 tx_flags |= IXGBE_TX_FLAGS_TXSW;
6802 /* if we have a HW VLAN tag being added default to the HW one */
6803 if (vlan_tx_tag_present(skb)) {
6804 tx_flags |= vlan_tx_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
6805 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6806 /* else if it is a SW VLAN check the next protocol and store the tag */
6807 } else if (protocol == __constant_htons(ETH_P_8021Q)) {
6808 struct vlan_hdr *vhdr, _vhdr;
6809 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
6813 protocol = vhdr->h_vlan_encapsulated_proto;
6814 tx_flags |= ntohs(vhdr->h_vlan_TCI) << IXGBE_TX_FLAGS_VLAN_SHIFT;
6815 tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
6818 /* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
6819 if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
6820 ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
6821 (skb->priority != TC_PRIO_CONTROL))) {
6822 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6823 tx_flags |= (skb->priority & 0x7) <<
6824 IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
6825 if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
6826 struct vlan_ethhdr *vhdr;
6827 if (skb_header_cloned(skb) &&
6828 pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
6830 vhdr = (struct vlan_ethhdr *)skb->data;
6831 vhdr->h_vlan_TCI = htons(tx_flags >>
6832 IXGBE_TX_FLAGS_VLAN_SHIFT);
6834 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6838 /* record the location of the first descriptor for this packet */
6839 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
6842 /* setup tx offload for FCoE */
6843 if ((protocol == __constant_htons(ETH_P_FCOE)) &&
6844 (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)) {
6845 tso = ixgbe_fso(tx_ring, skb, tx_flags, &hdr_len);
6849 tx_flags |= IXGBE_TX_FLAGS_FSO |
6850 IXGBE_TX_FLAGS_FCOE;
6852 tx_flags |= IXGBE_TX_FLAGS_FCOE;
6857 #endif /* IXGBE_FCOE */
6858 /* setup IPv4/IPv6 offloads */
6859 if (protocol == __constant_htons(ETH_P_IP))
6860 tx_flags |= IXGBE_TX_FLAGS_IPV4;
6862 tso = ixgbe_tso(tx_ring, skb, tx_flags, protocol, &hdr_len);
6866 tx_flags |= IXGBE_TX_FLAGS_TSO;
6867 else if (ixgbe_tx_csum(tx_ring, skb, tx_flags, protocol))
6868 tx_flags |= IXGBE_TX_FLAGS_CSUM;
6870 /* add the ATR filter if ATR is on */
6871 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
6872 ixgbe_atr(tx_ring, skb, tx_flags, protocol);
6876 #endif /* IXGBE_FCOE */
6877 ixgbe_tx_map(tx_ring, skb, first, tx_flags, hdr_len);
6879 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
6881 return NETDEV_TX_OK;
6884 dev_kfree_skb_any(skb);
6885 return NETDEV_TX_OK;
6888 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
6890 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6891 struct ixgbe_ring *tx_ring;
6893 tx_ring = adapter->tx_ring[skb->queue_mapping];
6894 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
6898 * ixgbe_set_mac - Change the Ethernet Address of the NIC
6899 * @netdev: network interface device structure
6900 * @p: pointer to an address structure
6902 * Returns 0 on success, negative on failure
6904 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6906 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6907 struct ixgbe_hw *hw = &adapter->hw;
6908 struct sockaddr *addr = p;
6910 if (!is_valid_ether_addr(addr->sa_data))
6911 return -EADDRNOTAVAIL;
6913 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6914 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6916 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
6923 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6925 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6926 struct ixgbe_hw *hw = &adapter->hw;
6930 if (prtad != hw->phy.mdio.prtad)
6932 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6938 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6939 u16 addr, u16 value)
6941 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6942 struct ixgbe_hw *hw = &adapter->hw;
6944 if (prtad != hw->phy.mdio.prtad)
6946 return hw->phy.ops.write_reg(hw, addr, devad, value);
6949 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6951 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6953 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6957 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6959 * @netdev: network interface device structure
6961 * Returns non-zero on failure
6963 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6966 struct ixgbe_adapter *adapter = netdev_priv(dev);
6967 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6969 if (is_valid_ether_addr(mac->san_addr)) {
6971 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6978 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6980 * @netdev: network interface device structure
6982 * Returns non-zero on failure
6984 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6987 struct ixgbe_adapter *adapter = netdev_priv(dev);
6988 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6990 if (is_valid_ether_addr(mac->san_addr)) {
6992 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6998 #ifdef CONFIG_NET_POLL_CONTROLLER
7000 * Polling 'interrupt' - used by things like netconsole to send skbs
7001 * without having to re-enable interrupts. It's not called while
7002 * the interrupt routine is executing.
7004 static void ixgbe_netpoll(struct net_device *netdev)
7006 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7009 /* if interface is down do nothing */
7010 if (test_bit(__IXGBE_DOWN, &adapter->state))
7013 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
7014 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
7015 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
7016 for (i = 0; i < num_q_vectors; i++) {
7017 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
7018 ixgbe_msix_clean_rings(0, q_vector);
7021 ixgbe_intr(adapter->pdev->irq, netdev);
7023 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
7027 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
7028 struct rtnl_link_stats64 *stats)
7030 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7034 for (i = 0; i < adapter->num_rx_queues; i++) {
7035 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
7041 start = u64_stats_fetch_begin_bh(&ring->syncp);
7042 packets = ring->stats.packets;
7043 bytes = ring->stats.bytes;
7044 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
7045 stats->rx_packets += packets;
7046 stats->rx_bytes += bytes;
7050 for (i = 0; i < adapter->num_tx_queues; i++) {
7051 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->tx_ring[i]);
7057 start = u64_stats_fetch_begin_bh(&ring->syncp);
7058 packets = ring->stats.packets;
7059 bytes = ring->stats.bytes;
7060 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
7061 stats->tx_packets += packets;
7062 stats->tx_bytes += bytes;
7066 /* following stats updated by ixgbe_watchdog_task() */
7067 stats->multicast = netdev->stats.multicast;
7068 stats->rx_errors = netdev->stats.rx_errors;
7069 stats->rx_length_errors = netdev->stats.rx_length_errors;
7070 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
7071 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
7075 /* ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
7076 * #adapter: pointer to ixgbe_adapter
7077 * @tc: number of traffic classes currently enabled
7079 * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
7080 * 802.1Q priority maps to a packet buffer that exists.
7082 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
7084 struct ixgbe_hw *hw = &adapter->hw;
7088 /* 82598 have a static priority to TC mapping that can not
7089 * be changed so no validation is needed.
7091 if (hw->mac.type == ixgbe_mac_82598EB)
7094 reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
7097 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
7098 u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
7100 /* If up2tc is out of bounds default to zero */
7102 reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
7106 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
7112 /* ixgbe_setup_tc - routine to configure net_device for multiple traffic
7115 * @netdev: net device to configure
7116 * @tc: number of traffic classes to enable
7118 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
7120 struct ixgbe_adapter *adapter = netdev_priv(dev);
7121 struct ixgbe_hw *hw = &adapter->hw;
7123 /* Multiple traffic classes requires multiple queues */
7124 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
7125 e_err(drv, "Enable failed, needs MSI-X\n");
7129 /* Hardware supports up to 8 traffic classes */
7130 if (tc > adapter->dcb_cfg.num_tcs.pg_tcs ||
7131 (hw->mac.type == ixgbe_mac_82598EB && tc < MAX_TRAFFIC_CLASS))
7134 /* Hardware has to reinitialize queues and interrupts to
7135 * match packet buffer alignment. Unfortunately, the
7136 * hardware is not flexible enough to do this dynamically.
7138 if (netif_running(dev))
7140 ixgbe_clear_interrupt_scheme(adapter);
7143 netdev_set_num_tc(dev, tc);
7144 adapter->last_lfc_mode = adapter->hw.fc.current_mode;
7146 adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
7147 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
7149 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
7150 adapter->hw.fc.requested_mode = ixgbe_fc_none;
7152 netdev_reset_tc(dev);
7154 adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
7156 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
7157 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
7159 adapter->temp_dcb_cfg.pfc_mode_enable = false;
7160 adapter->dcb_cfg.pfc_mode_enable = false;
7163 ixgbe_init_interrupt_scheme(adapter);
7164 ixgbe_validate_rtr(adapter, tc);
7165 if (netif_running(dev))
7171 void ixgbe_do_reset(struct net_device *netdev)
7173 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7175 if (netif_running(netdev))
7176 ixgbe_reinit_locked(adapter);
7178 ixgbe_reset(adapter);
7181 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
7182 netdev_features_t data)
7184 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7187 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
7188 data &= ~NETIF_F_HW_VLAN_RX;
7191 /* return error if RXHASH is being enabled when RSS is not supported */
7192 if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED))
7193 data &= ~NETIF_F_RXHASH;
7195 /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
7196 if (!(data & NETIF_F_RXCSUM))
7197 data &= ~NETIF_F_LRO;
7199 /* Turn off LRO if not RSC capable or invalid ITR settings */
7200 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)) {
7201 data &= ~NETIF_F_LRO;
7202 } else if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
7203 (adapter->rx_itr_setting != 1 &&
7204 adapter->rx_itr_setting > IXGBE_MAX_RSC_INT_RATE)) {
7205 data &= ~NETIF_F_LRO;
7206 e_info(probe, "rx-usecs set too low, not enabling RSC\n");
7212 static int ixgbe_set_features(struct net_device *netdev,
7213 netdev_features_t data)
7215 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7216 bool need_reset = false;
7218 /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
7219 if (!(data & NETIF_F_RXCSUM))
7220 adapter->flags &= ~IXGBE_FLAG_RX_CSUM_ENABLED;
7222 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
7224 /* Make sure RSC matches LRO, reset if change */
7225 if (!!(data & NETIF_F_LRO) !=
7226 !!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
7227 adapter->flags2 ^= IXGBE_FLAG2_RSC_ENABLED;
7228 switch (adapter->hw.mac.type) {
7229 case ixgbe_mac_X540:
7230 case ixgbe_mac_82599EB:
7239 * Check if Flow Director n-tuple support was enabled or disabled. If
7240 * the state changed, we need to reset.
7242 if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)) {
7243 /* turn off ATR, enable perfect filters and reset */
7244 if (data & NETIF_F_NTUPLE) {
7245 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
7246 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
7249 } else if (!(data & NETIF_F_NTUPLE)) {
7250 /* turn off Flow Director, set ATR and reset */
7251 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
7252 if ((adapter->flags & IXGBE_FLAG_RSS_ENABLED) &&
7253 !(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
7254 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
7259 ixgbe_do_reset(netdev);
7265 static const struct net_device_ops ixgbe_netdev_ops = {
7266 .ndo_open = ixgbe_open,
7267 .ndo_stop = ixgbe_close,
7268 .ndo_start_xmit = ixgbe_xmit_frame,
7269 .ndo_select_queue = ixgbe_select_queue,
7270 .ndo_set_rx_mode = ixgbe_set_rx_mode,
7271 .ndo_validate_addr = eth_validate_addr,
7272 .ndo_set_mac_address = ixgbe_set_mac,
7273 .ndo_change_mtu = ixgbe_change_mtu,
7274 .ndo_tx_timeout = ixgbe_tx_timeout,
7275 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
7276 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
7277 .ndo_do_ioctl = ixgbe_ioctl,
7278 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
7279 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
7280 .ndo_set_vf_tx_rate = ixgbe_ndo_set_vf_bw,
7281 .ndo_set_vf_spoofchk = ixgbe_ndo_set_vf_spoofchk,
7282 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
7283 .ndo_get_stats64 = ixgbe_get_stats64,
7284 .ndo_setup_tc = ixgbe_setup_tc,
7285 #ifdef CONFIG_NET_POLL_CONTROLLER
7286 .ndo_poll_controller = ixgbe_netpoll,
7289 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
7290 .ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
7291 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
7292 .ndo_fcoe_enable = ixgbe_fcoe_enable,
7293 .ndo_fcoe_disable = ixgbe_fcoe_disable,
7294 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
7295 #endif /* IXGBE_FCOE */
7296 .ndo_set_features = ixgbe_set_features,
7297 .ndo_fix_features = ixgbe_fix_features,
7300 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
7301 const struct ixgbe_info *ii)
7303 #ifdef CONFIG_PCI_IOV
7304 struct ixgbe_hw *hw = &adapter->hw;
7306 if (hw->mac.type == ixgbe_mac_82598EB)
7309 /* The 82599 supports up to 64 VFs per physical function
7310 * but this implementation limits allocation to 63 so that
7311 * basic networking resources are still available to the
7314 adapter->num_vfs = (max_vfs > 63) ? 63 : max_vfs;
7315 ixgbe_enable_sriov(adapter, ii);
7316 #endif /* CONFIG_PCI_IOV */
7320 * ixgbe_probe - Device Initialization Routine
7321 * @pdev: PCI device information struct
7322 * @ent: entry in ixgbe_pci_tbl
7324 * Returns 0 on success, negative on failure
7326 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
7327 * The OS initialization, configuring of the adapter private structure,
7328 * and a hardware reset occur.
7330 static int __devinit ixgbe_probe(struct pci_dev *pdev,
7331 const struct pci_device_id *ent)
7333 struct net_device *netdev;
7334 struct ixgbe_adapter *adapter = NULL;
7335 struct ixgbe_hw *hw;
7336 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
7337 static int cards_found;
7338 int i, err, pci_using_dac;
7339 u8 part_str[IXGBE_PBANUM_LENGTH];
7340 unsigned int indices = num_possible_cpus();
7347 /* Catch broken hardware that put the wrong VF device ID in
7348 * the PCIe SR-IOV capability.
7350 if (pdev->is_virtfn) {
7351 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
7352 pci_name(pdev), pdev->vendor, pdev->device);
7356 err = pci_enable_device_mem(pdev);
7360 if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
7361 !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
7364 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
7366 err = dma_set_coherent_mask(&pdev->dev,
7370 "No usable DMA configuration, aborting\n");
7377 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
7378 IORESOURCE_MEM), ixgbe_driver_name);
7381 "pci_request_selected_regions failed 0x%x\n", err);
7385 pci_enable_pcie_error_reporting(pdev);
7387 pci_set_master(pdev);
7388 pci_save_state(pdev);
7390 #ifdef CONFIG_IXGBE_DCB
7391 indices *= MAX_TRAFFIC_CLASS;
7394 if (ii->mac == ixgbe_mac_82598EB)
7395 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
7397 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
7400 indices += min_t(unsigned int, num_possible_cpus(),
7401 IXGBE_MAX_FCOE_INDICES);
7403 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
7406 goto err_alloc_etherdev;
7409 SET_NETDEV_DEV(netdev, &pdev->dev);
7411 adapter = netdev_priv(netdev);
7412 pci_set_drvdata(pdev, adapter);
7414 adapter->netdev = netdev;
7415 adapter->pdev = pdev;
7418 adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
7420 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
7421 pci_resource_len(pdev, 0));
7427 for (i = 1; i <= 5; i++) {
7428 if (pci_resource_len(pdev, i) == 0)
7432 netdev->netdev_ops = &ixgbe_netdev_ops;
7433 ixgbe_set_ethtool_ops(netdev);
7434 netdev->watchdog_timeo = 5 * HZ;
7435 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
7437 adapter->bd_number = cards_found;
7440 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
7441 hw->mac.type = ii->mac;
7444 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
7445 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
7446 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
7447 if (!(eec & (1 << 8)))
7448 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
7451 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
7452 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
7453 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
7454 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
7455 hw->phy.mdio.mmds = 0;
7456 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
7457 hw->phy.mdio.dev = netdev;
7458 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
7459 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
7461 ii->get_invariants(hw);
7463 /* setup the private structure */
7464 err = ixgbe_sw_init(adapter);
7468 /* Make it possible the adapter to be woken up via WOL */
7469 switch (adapter->hw.mac.type) {
7470 case ixgbe_mac_82599EB:
7471 case ixgbe_mac_X540:
7472 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7479 * If there is a fan on this device and it has failed log the
7482 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
7483 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
7484 if (esdp & IXGBE_ESDP_SDP1)
7485 e_crit(probe, "Fan has stopped, replace the adapter\n");
7488 /* reset_hw fills in the perm_addr as well */
7489 hw->phy.reset_if_overtemp = true;
7490 err = hw->mac.ops.reset_hw(hw);
7491 hw->phy.reset_if_overtemp = false;
7492 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
7493 hw->mac.type == ixgbe_mac_82598EB) {
7495 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
7496 e_dev_err("failed to load because an unsupported SFP+ "
7497 "module type was detected.\n");
7498 e_dev_err("Reload the driver after installing a supported "
7502 e_dev_err("HW Init failed: %d\n", err);
7506 ixgbe_probe_vf(adapter, ii);
7508 netdev->features = NETIF_F_SG |
7511 NETIF_F_HW_VLAN_TX |
7512 NETIF_F_HW_VLAN_RX |
7513 NETIF_F_HW_VLAN_FILTER |
7519 netdev->hw_features = netdev->features;
7521 switch (adapter->hw.mac.type) {
7522 case ixgbe_mac_82599EB:
7523 case ixgbe_mac_X540:
7524 netdev->features |= NETIF_F_SCTP_CSUM;
7525 netdev->hw_features |= NETIF_F_SCTP_CSUM |
7532 netdev->vlan_features |= NETIF_F_TSO;
7533 netdev->vlan_features |= NETIF_F_TSO6;
7534 netdev->vlan_features |= NETIF_F_IP_CSUM;
7535 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
7536 netdev->vlan_features |= NETIF_F_SG;
7538 netdev->priv_flags |= IFF_UNICAST_FLT;
7540 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7541 adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
7542 IXGBE_FLAG_DCB_ENABLED);
7544 #ifdef CONFIG_IXGBE_DCB
7545 netdev->dcbnl_ops = &dcbnl_ops;
7549 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7550 if (hw->mac.ops.get_device_caps) {
7551 hw->mac.ops.get_device_caps(hw, &device_caps);
7552 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
7553 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
7556 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7557 netdev->vlan_features |= NETIF_F_FCOE_CRC;
7558 netdev->vlan_features |= NETIF_F_FSO;
7559 netdev->vlan_features |= NETIF_F_FCOE_MTU;
7561 #endif /* IXGBE_FCOE */
7562 if (pci_using_dac) {
7563 netdev->features |= NETIF_F_HIGHDMA;
7564 netdev->vlan_features |= NETIF_F_HIGHDMA;
7567 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
7568 netdev->hw_features |= NETIF_F_LRO;
7569 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
7570 netdev->features |= NETIF_F_LRO;
7572 /* make sure the EEPROM is good */
7573 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
7574 e_dev_err("The EEPROM Checksum Is Not Valid\n");
7579 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
7580 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
7582 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
7583 e_dev_err("invalid MAC address\n");
7588 setup_timer(&adapter->service_timer, &ixgbe_service_timer,
7589 (unsigned long) adapter);
7591 INIT_WORK(&adapter->service_task, ixgbe_service_task);
7592 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
7594 err = ixgbe_init_interrupt_scheme(adapter);
7598 if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED)) {
7599 netdev->hw_features &= ~NETIF_F_RXHASH;
7600 netdev->features &= ~NETIF_F_RXHASH;
7603 /* WOL not supported for all but the following */
7605 switch (pdev->device) {
7606 case IXGBE_DEV_ID_82599_SFP:
7607 /* Only this subdevice supports WOL */
7608 if (pdev->subsystem_device == IXGBE_SUBDEV_ID_82599_SFP)
7609 adapter->wol = IXGBE_WUFC_MAG;
7611 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
7612 /* All except this subdevice support WOL */
7613 if (pdev->subsystem_device != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
7614 adapter->wol = IXGBE_WUFC_MAG;
7616 case IXGBE_DEV_ID_82599_KX4:
7617 adapter->wol = IXGBE_WUFC_MAG;
7619 case IXGBE_DEV_ID_X540T:
7620 /* Check eeprom to see if it is enabled */
7621 hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
7622 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
7624 if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
7625 ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
7626 (hw->bus.func == 0)))
7627 adapter->wol = IXGBE_WUFC_MAG;
7630 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
7632 /* save off EEPROM version number */
7633 hw->eeprom.ops.read(hw, 0x2e, &adapter->eeprom_verh);
7634 hw->eeprom.ops.read(hw, 0x2d, &adapter->eeprom_verl);
7636 /* pick up the PCI bus settings for reporting later */
7637 hw->mac.ops.get_bus_info(hw);
7639 /* print bus type/speed/width info */
7640 e_dev_info("(PCI Express:%s:%s) %pM\n",
7641 (hw->bus.speed == ixgbe_bus_speed_5000 ? "5.0GT/s" :
7642 hw->bus.speed == ixgbe_bus_speed_2500 ? "2.5GT/s" :
7644 (hw->bus.width == ixgbe_bus_width_pcie_x8 ? "Width x8" :
7645 hw->bus.width == ixgbe_bus_width_pcie_x4 ? "Width x4" :
7646 hw->bus.width == ixgbe_bus_width_pcie_x1 ? "Width x1" :
7650 err = ixgbe_read_pba_string_generic(hw, part_str, IXGBE_PBANUM_LENGTH);
7652 strncpy(part_str, "Unknown", IXGBE_PBANUM_LENGTH);
7653 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
7654 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
7655 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
7658 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
7659 hw->mac.type, hw->phy.type, part_str);
7661 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
7662 e_dev_warn("PCI-Express bandwidth available for this card is "
7663 "not sufficient for optimal performance.\n");
7664 e_dev_warn("For optimal performance a x8 PCI-Express slot "
7668 /* reset the hardware with the new settings */
7669 err = hw->mac.ops.start_hw(hw);
7671 if (err == IXGBE_ERR_EEPROM_VERSION) {
7672 /* We are running on a pre-production device, log a warning */
7673 e_dev_warn("This device is a pre-production adapter/LOM. "
7674 "Please be aware there may be issues associated "
7675 "with your hardware. If you are experiencing "
7676 "problems please contact your Intel or hardware "
7677 "representative who provided you with this "
7680 strcpy(netdev->name, "eth%d");
7681 err = register_netdev(netdev);
7685 /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
7686 if (hw->mac.ops.disable_tx_laser &&
7687 ((hw->phy.multispeed_fiber) ||
7688 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
7689 (hw->mac.type == ixgbe_mac_82599EB))))
7690 hw->mac.ops.disable_tx_laser(hw);
7692 /* carrier off reporting is important to ethtool even BEFORE open */
7693 netif_carrier_off(netdev);
7695 #ifdef CONFIG_IXGBE_DCA
7696 if (dca_add_requester(&pdev->dev) == 0) {
7697 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
7698 ixgbe_setup_dca(adapter);
7701 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7702 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
7703 for (i = 0; i < adapter->num_vfs; i++)
7704 ixgbe_vf_configuration(pdev, (i | 0x10000000));
7707 /* firmware requires driver version to be 0xFFFFFFFF
7708 * since os does not support feature
7710 if (hw->mac.ops.set_fw_drv_ver)
7711 hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF,
7714 /* add san mac addr to netdev */
7715 ixgbe_add_sanmac_netdev(netdev);
7717 e_dev_info("Intel(R) 10 Gigabit Network Connection\n");
7722 ixgbe_release_hw_control(adapter);
7723 ixgbe_clear_interrupt_scheme(adapter);
7726 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7727 ixgbe_disable_sriov(adapter);
7728 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
7729 iounmap(hw->hw_addr);
7731 free_netdev(netdev);
7733 pci_release_selected_regions(pdev,
7734 pci_select_bars(pdev, IORESOURCE_MEM));
7737 pci_disable_device(pdev);
7742 * ixgbe_remove - Device Removal Routine
7743 * @pdev: PCI device information struct
7745 * ixgbe_remove is called by the PCI subsystem to alert the driver
7746 * that it should release a PCI device. The could be caused by a
7747 * Hot-Plug event, or because the driver is going to be removed from
7750 static void __devexit ixgbe_remove(struct pci_dev *pdev)
7752 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7753 struct net_device *netdev = adapter->netdev;
7755 set_bit(__IXGBE_DOWN, &adapter->state);
7756 cancel_work_sync(&adapter->service_task);
7758 #ifdef CONFIG_IXGBE_DCA
7759 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
7760 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
7761 dca_remove_requester(&pdev->dev);
7762 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
7767 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
7768 ixgbe_cleanup_fcoe(adapter);
7770 #endif /* IXGBE_FCOE */
7772 /* remove the added san mac */
7773 ixgbe_del_sanmac_netdev(netdev);
7775 if (netdev->reg_state == NETREG_REGISTERED)
7776 unregister_netdev(netdev);
7778 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7779 if (!(ixgbe_check_vf_assignment(adapter)))
7780 ixgbe_disable_sriov(adapter);
7782 e_dev_warn("Unloading driver while VFs are assigned "
7783 "- VFs will not be deallocated\n");
7786 ixgbe_clear_interrupt_scheme(adapter);
7788 ixgbe_release_hw_control(adapter);
7790 iounmap(adapter->hw.hw_addr);
7791 pci_release_selected_regions(pdev, pci_select_bars(pdev,
7794 e_dev_info("complete\n");
7796 free_netdev(netdev);
7798 pci_disable_pcie_error_reporting(pdev);
7800 pci_disable_device(pdev);
7804 * ixgbe_io_error_detected - called when PCI error is detected
7805 * @pdev: Pointer to PCI device
7806 * @state: The current pci connection state
7808 * This function is called after a PCI bus error affecting
7809 * this device has been detected.
7811 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
7812 pci_channel_state_t state)
7814 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7815 struct net_device *netdev = adapter->netdev;
7817 #ifdef CONFIG_PCI_IOV
7818 struct pci_dev *bdev, *vfdev;
7819 u32 dw0, dw1, dw2, dw3;
7821 u16 req_id, pf_func;
7823 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
7824 adapter->num_vfs == 0)
7825 goto skip_bad_vf_detection;
7827 bdev = pdev->bus->self;
7828 while (bdev && (bdev->pcie_type != PCI_EXP_TYPE_ROOT_PORT))
7829 bdev = bdev->bus->self;
7832 goto skip_bad_vf_detection;
7834 pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
7836 goto skip_bad_vf_detection;
7838 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG, &dw0);
7839 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 4, &dw1);
7840 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 8, &dw2);
7841 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 12, &dw3);
7844 /* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
7845 if (!(req_id & 0x0080))
7846 goto skip_bad_vf_detection;
7848 pf_func = req_id & 0x01;
7849 if ((pf_func & 1) == (pdev->devfn & 1)) {
7850 unsigned int device_id;
7852 vf = (req_id & 0x7F) >> 1;
7853 e_dev_err("VF %d has caused a PCIe error\n", vf);
7854 e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
7855 "%8.8x\tdw3: %8.8x\n",
7856 dw0, dw1, dw2, dw3);
7857 switch (adapter->hw.mac.type) {
7858 case ixgbe_mac_82599EB:
7859 device_id = IXGBE_82599_VF_DEVICE_ID;
7861 case ixgbe_mac_X540:
7862 device_id = IXGBE_X540_VF_DEVICE_ID;
7869 /* Find the pci device of the offending VF */
7870 vfdev = pci_get_device(IXGBE_INTEL_VENDOR_ID, device_id, NULL);
7872 if (vfdev->devfn == (req_id & 0xFF))
7874 vfdev = pci_get_device(IXGBE_INTEL_VENDOR_ID,
7878 * There's a slim chance the VF could have been hot plugged,
7879 * so if it is no longer present we don't need to issue the
7880 * VFLR. Just clean up the AER in that case.
7883 e_dev_err("Issuing VFLR to VF %d\n", vf);
7884 pci_write_config_dword(vfdev, 0xA8, 0x00008000);
7887 pci_cleanup_aer_uncorrect_error_status(pdev);
7891 * Even though the error may have occurred on the other port
7892 * we still need to increment the vf error reference count for
7893 * both ports because the I/O resume function will be called
7896 adapter->vferr_refcount++;
7898 return PCI_ERS_RESULT_RECOVERED;
7900 skip_bad_vf_detection:
7901 #endif /* CONFIG_PCI_IOV */
7902 netif_device_detach(netdev);
7904 if (state == pci_channel_io_perm_failure)
7905 return PCI_ERS_RESULT_DISCONNECT;
7907 if (netif_running(netdev))
7908 ixgbe_down(adapter);
7909 pci_disable_device(pdev);
7911 /* Request a slot reset. */
7912 return PCI_ERS_RESULT_NEED_RESET;
7916 * ixgbe_io_slot_reset - called after the pci bus has been reset.
7917 * @pdev: Pointer to PCI device
7919 * Restart the card from scratch, as if from a cold-boot.
7921 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7923 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7924 pci_ers_result_t result;
7927 if (pci_enable_device_mem(pdev)) {
7928 e_err(probe, "Cannot re-enable PCI device after reset.\n");
7929 result = PCI_ERS_RESULT_DISCONNECT;
7931 pci_set_master(pdev);
7932 pci_restore_state(pdev);
7933 pci_save_state(pdev);
7935 pci_wake_from_d3(pdev, false);
7937 ixgbe_reset(adapter);
7938 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7939 result = PCI_ERS_RESULT_RECOVERED;
7942 err = pci_cleanup_aer_uncorrect_error_status(pdev);
7944 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7945 "failed 0x%0x\n", err);
7946 /* non-fatal, continue */
7953 * ixgbe_io_resume - called when traffic can start flowing again.
7954 * @pdev: Pointer to PCI device
7956 * This callback is called when the error recovery driver tells us that
7957 * its OK to resume normal operation.
7959 static void ixgbe_io_resume(struct pci_dev *pdev)
7961 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7962 struct net_device *netdev = adapter->netdev;
7964 #ifdef CONFIG_PCI_IOV
7965 if (adapter->vferr_refcount) {
7966 e_info(drv, "Resuming after VF err\n");
7967 adapter->vferr_refcount--;
7972 if (netif_running(netdev))
7975 netif_device_attach(netdev);
7978 static struct pci_error_handlers ixgbe_err_handler = {
7979 .error_detected = ixgbe_io_error_detected,
7980 .slot_reset = ixgbe_io_slot_reset,
7981 .resume = ixgbe_io_resume,
7984 static struct pci_driver ixgbe_driver = {
7985 .name = ixgbe_driver_name,
7986 .id_table = ixgbe_pci_tbl,
7987 .probe = ixgbe_probe,
7988 .remove = __devexit_p(ixgbe_remove),
7990 .suspend = ixgbe_suspend,
7991 .resume = ixgbe_resume,
7993 .shutdown = ixgbe_shutdown,
7994 .err_handler = &ixgbe_err_handler
7998 * ixgbe_init_module - Driver Registration Routine
8000 * ixgbe_init_module is the first routine called when the driver is
8001 * loaded. All it does is register with the PCI subsystem.
8003 static int __init ixgbe_init_module(void)
8006 pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
8007 pr_info("%s\n", ixgbe_copyright);
8009 #ifdef CONFIG_IXGBE_DCA
8010 dca_register_notify(&dca_notifier);
8013 ret = pci_register_driver(&ixgbe_driver);
8017 module_init(ixgbe_init_module);
8020 * ixgbe_exit_module - Driver Exit Cleanup Routine
8022 * ixgbe_exit_module is called just before the driver is removed
8025 static void __exit ixgbe_exit_module(void)
8027 #ifdef CONFIG_IXGBE_DCA
8028 dca_unregister_notify(&dca_notifier);
8030 pci_unregister_driver(&ixgbe_driver);
8031 rcu_barrier(); /* Wait for completion of call_rcu()'s */
8034 #ifdef CONFIG_IXGBE_DCA
8035 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
8040 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
8041 __ixgbe_notify_dca);
8043 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
8046 #endif /* CONFIG_IXGBE_DCA */
8048 module_exit(ixgbe_exit_module);