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 /* required last entry */
111 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
113 #ifdef CONFIG_IXGBE_DCA
114 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
116 static struct notifier_block dca_notifier = {
117 .notifier_call = ixgbe_notify_dca,
123 #ifdef CONFIG_PCI_IOV
124 static unsigned int max_vfs;
125 module_param(max_vfs, uint, 0);
126 MODULE_PARM_DESC(max_vfs,
127 "Maximum number of virtual functions to allocate per physical function");
128 #endif /* CONFIG_PCI_IOV */
130 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
131 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
132 MODULE_LICENSE("GPL");
133 MODULE_VERSION(DRV_VERSION);
135 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
137 static inline void ixgbe_disable_sriov(struct ixgbe_adapter *adapter)
139 struct ixgbe_hw *hw = &adapter->hw;
144 #ifdef CONFIG_PCI_IOV
145 /* disable iov and allow time for transactions to clear */
146 pci_disable_sriov(adapter->pdev);
149 /* turn off device IOV mode */
150 gcr = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
151 gcr &= ~(IXGBE_GCR_EXT_SRIOV);
152 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr);
153 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
154 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
155 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
157 /* set default pool back to 0 */
158 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
159 vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
160 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
161 IXGBE_WRITE_FLUSH(hw);
163 /* take a breather then clean up driver data */
166 kfree(adapter->vfinfo);
167 adapter->vfinfo = NULL;
169 adapter->num_vfs = 0;
170 adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
173 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
175 if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
176 !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
177 schedule_work(&adapter->service_task);
180 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
182 BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
184 /* flush memory to make sure state is correct before next watchog */
185 smp_mb__before_clear_bit();
186 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
189 struct ixgbe_reg_info {
194 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
196 /* General Registers */
197 {IXGBE_CTRL, "CTRL"},
198 {IXGBE_STATUS, "STATUS"},
199 {IXGBE_CTRL_EXT, "CTRL_EXT"},
201 /* Interrupt Registers */
202 {IXGBE_EICR, "EICR"},
205 {IXGBE_SRRCTL(0), "SRRCTL"},
206 {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
207 {IXGBE_RDLEN(0), "RDLEN"},
208 {IXGBE_RDH(0), "RDH"},
209 {IXGBE_RDT(0), "RDT"},
210 {IXGBE_RXDCTL(0), "RXDCTL"},
211 {IXGBE_RDBAL(0), "RDBAL"},
212 {IXGBE_RDBAH(0), "RDBAH"},
215 {IXGBE_TDBAL(0), "TDBAL"},
216 {IXGBE_TDBAH(0), "TDBAH"},
217 {IXGBE_TDLEN(0), "TDLEN"},
218 {IXGBE_TDH(0), "TDH"},
219 {IXGBE_TDT(0), "TDT"},
220 {IXGBE_TXDCTL(0), "TXDCTL"},
222 /* List Terminator */
228 * ixgbe_regdump - register printout routine
230 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
236 switch (reginfo->ofs) {
237 case IXGBE_SRRCTL(0):
238 for (i = 0; i < 64; i++)
239 regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
241 case IXGBE_DCA_RXCTRL(0):
242 for (i = 0; i < 64; i++)
243 regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
246 for (i = 0; i < 64; i++)
247 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
250 for (i = 0; i < 64; i++)
251 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
254 for (i = 0; i < 64; i++)
255 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
257 case IXGBE_RXDCTL(0):
258 for (i = 0; i < 64; i++)
259 regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
262 for (i = 0; i < 64; i++)
263 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
266 for (i = 0; i < 64; i++)
267 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
270 for (i = 0; i < 64; i++)
271 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
274 for (i = 0; i < 64; i++)
275 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
278 for (i = 0; i < 64; i++)
279 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
282 for (i = 0; i < 64; i++)
283 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
286 for (i = 0; i < 64; i++)
287 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
289 case IXGBE_TXDCTL(0):
290 for (i = 0; i < 64; i++)
291 regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
294 pr_info("%-15s %08x\n", reginfo->name,
295 IXGBE_READ_REG(hw, reginfo->ofs));
299 for (i = 0; i < 8; i++) {
300 snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i*8, i*8+7);
301 pr_err("%-15s", rname);
302 for (j = 0; j < 8; j++)
303 pr_cont(" %08x", regs[i*8+j]);
310 * ixgbe_dump - Print registers, tx-rings and rx-rings
312 static void ixgbe_dump(struct ixgbe_adapter *adapter)
314 struct net_device *netdev = adapter->netdev;
315 struct ixgbe_hw *hw = &adapter->hw;
316 struct ixgbe_reg_info *reginfo;
318 struct ixgbe_ring *tx_ring;
319 struct ixgbe_tx_buffer *tx_buffer_info;
320 union ixgbe_adv_tx_desc *tx_desc;
321 struct my_u0 { u64 a; u64 b; } *u0;
322 struct ixgbe_ring *rx_ring;
323 union ixgbe_adv_rx_desc *rx_desc;
324 struct ixgbe_rx_buffer *rx_buffer_info;
328 if (!netif_msg_hw(adapter))
331 /* Print netdevice Info */
333 dev_info(&adapter->pdev->dev, "Net device Info\n");
334 pr_info("Device Name state "
335 "trans_start last_rx\n");
336 pr_info("%-15s %016lX %016lX %016lX\n",
343 /* Print Registers */
344 dev_info(&adapter->pdev->dev, "Register Dump\n");
345 pr_info(" Register Name Value\n");
346 for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
347 reginfo->name; reginfo++) {
348 ixgbe_regdump(hw, reginfo);
351 /* Print TX Ring Summary */
352 if (!netdev || !netif_running(netdev))
355 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
356 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
357 for (n = 0; n < adapter->num_tx_queues; n++) {
358 tx_ring = adapter->tx_ring[n];
360 &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
361 pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
362 n, tx_ring->next_to_use, tx_ring->next_to_clean,
363 (u64)tx_buffer_info->dma,
364 tx_buffer_info->length,
365 tx_buffer_info->next_to_watch,
366 (u64)tx_buffer_info->time_stamp);
370 if (!netif_msg_tx_done(adapter))
371 goto rx_ring_summary;
373 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
375 /* Transmit Descriptor Formats
377 * Advanced Transmit Descriptor
378 * +--------------------------------------------------------------+
379 * 0 | Buffer Address [63:0] |
380 * +--------------------------------------------------------------+
381 * 8 | PAYLEN | PORTS | IDX | STA | DCMD |DTYP | RSV | DTALEN |
382 * +--------------------------------------------------------------+
383 * 63 46 45 40 39 36 35 32 31 24 23 20 19 0
386 for (n = 0; n < adapter->num_tx_queues; n++) {
387 tx_ring = adapter->tx_ring[n];
388 pr_info("------------------------------------\n");
389 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
390 pr_info("------------------------------------\n");
391 pr_info("T [desc] [address 63:0 ] "
392 "[PlPOIdStDDt Ln] [bi->dma ] "
393 "leng ntw timestamp bi->skb\n");
395 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
396 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
397 tx_buffer_info = &tx_ring->tx_buffer_info[i];
398 u0 = (struct my_u0 *)tx_desc;
399 pr_info("T [0x%03X] %016llX %016llX %016llX"
400 " %04X %p %016llX %p", i,
403 (u64)tx_buffer_info->dma,
404 tx_buffer_info->length,
405 tx_buffer_info->next_to_watch,
406 (u64)tx_buffer_info->time_stamp,
407 tx_buffer_info->skb);
408 if (i == tx_ring->next_to_use &&
409 i == tx_ring->next_to_clean)
411 else if (i == tx_ring->next_to_use)
413 else if (i == tx_ring->next_to_clean)
418 if (netif_msg_pktdata(adapter) &&
419 tx_buffer_info->dma != 0)
420 print_hex_dump(KERN_INFO, "",
421 DUMP_PREFIX_ADDRESS, 16, 1,
422 phys_to_virt(tx_buffer_info->dma),
423 tx_buffer_info->length, true);
427 /* Print RX Rings Summary */
429 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
430 pr_info("Queue [NTU] [NTC]\n");
431 for (n = 0; n < adapter->num_rx_queues; n++) {
432 rx_ring = adapter->rx_ring[n];
433 pr_info("%5d %5X %5X\n",
434 n, rx_ring->next_to_use, rx_ring->next_to_clean);
438 if (!netif_msg_rx_status(adapter))
441 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
443 /* Advanced Receive Descriptor (Read) Format
445 * +-----------------------------------------------------+
446 * 0 | Packet Buffer Address [63:1] |A0/NSE|
447 * +----------------------------------------------+------+
448 * 8 | Header Buffer Address [63:1] | DD |
449 * +-----------------------------------------------------+
452 * Advanced Receive Descriptor (Write-Back) Format
454 * 63 48 47 32 31 30 21 20 16 15 4 3 0
455 * +------------------------------------------------------+
456 * 0 | Packet IP |SPH| HDR_LEN | RSV|Packet| RSS |
457 * | Checksum Ident | | | | Type | Type |
458 * +------------------------------------------------------+
459 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
460 * +------------------------------------------------------+
461 * 63 48 47 32 31 20 19 0
463 for (n = 0; n < adapter->num_rx_queues; n++) {
464 rx_ring = adapter->rx_ring[n];
465 pr_info("------------------------------------\n");
466 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
467 pr_info("------------------------------------\n");
468 pr_info("R [desc] [ PktBuf A0] "
469 "[ HeadBuf DD] [bi->dma ] [bi->skb] "
470 "<-- Adv Rx Read format\n");
471 pr_info("RWB[desc] [PcsmIpSHl PtRs] "
472 "[vl er S cks ln] ---------------- [bi->skb] "
473 "<-- Adv Rx Write-Back format\n");
475 for (i = 0; i < rx_ring->count; i++) {
476 rx_buffer_info = &rx_ring->rx_buffer_info[i];
477 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
478 u0 = (struct my_u0 *)rx_desc;
479 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
480 if (staterr & IXGBE_RXD_STAT_DD) {
481 /* Descriptor Done */
482 pr_info("RWB[0x%03X] %016llX "
483 "%016llX ---------------- %p", i,
486 rx_buffer_info->skb);
488 pr_info("R [0x%03X] %016llX "
489 "%016llX %016llX %p", i,
492 (u64)rx_buffer_info->dma,
493 rx_buffer_info->skb);
495 if (netif_msg_pktdata(adapter)) {
496 print_hex_dump(KERN_INFO, "",
497 DUMP_PREFIX_ADDRESS, 16, 1,
498 phys_to_virt(rx_buffer_info->dma),
499 rx_ring->rx_buf_len, true);
501 if (rx_ring->rx_buf_len
502 < IXGBE_RXBUFFER_2048)
503 print_hex_dump(KERN_INFO, "",
504 DUMP_PREFIX_ADDRESS, 16, 1,
506 rx_buffer_info->page_dma +
507 rx_buffer_info->page_offset
513 if (i == rx_ring->next_to_use)
515 else if (i == rx_ring->next_to_clean)
527 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
531 /* Let firmware take over control of h/w */
532 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
533 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
534 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
537 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
541 /* Let firmware know the driver has taken over */
542 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
543 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
544 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
548 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
549 * @adapter: pointer to adapter struct
550 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
551 * @queue: queue to map the corresponding interrupt to
552 * @msix_vector: the vector to map to the corresponding queue
555 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
556 u8 queue, u8 msix_vector)
559 struct ixgbe_hw *hw = &adapter->hw;
560 switch (hw->mac.type) {
561 case ixgbe_mac_82598EB:
562 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
565 index = (((direction * 64) + queue) >> 2) & 0x1F;
566 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
567 ivar &= ~(0xFF << (8 * (queue & 0x3)));
568 ivar |= (msix_vector << (8 * (queue & 0x3)));
569 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
571 case ixgbe_mac_82599EB:
573 if (direction == -1) {
575 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
576 index = ((queue & 1) * 8);
577 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
578 ivar &= ~(0xFF << index);
579 ivar |= (msix_vector << index);
580 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
583 /* tx or rx causes */
584 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
585 index = ((16 * (queue & 1)) + (8 * direction));
586 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
587 ivar &= ~(0xFF << index);
588 ivar |= (msix_vector << index);
589 IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
597 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
602 switch (adapter->hw.mac.type) {
603 case ixgbe_mac_82598EB:
604 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
605 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
607 case ixgbe_mac_82599EB:
609 mask = (qmask & 0xFFFFFFFF);
610 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
611 mask = (qmask >> 32);
612 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
619 static inline void ixgbe_unmap_tx_resource(struct ixgbe_ring *ring,
620 struct ixgbe_tx_buffer *tx_buffer)
622 if (tx_buffer->dma) {
623 if (tx_buffer->tx_flags & IXGBE_TX_FLAGS_MAPPED_AS_PAGE)
624 dma_unmap_page(ring->dev,
629 dma_unmap_single(ring->dev,
637 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_ring *tx_ring,
638 struct ixgbe_tx_buffer *tx_buffer_info)
640 ixgbe_unmap_tx_resource(tx_ring, tx_buffer_info);
641 if (tx_buffer_info->skb)
642 dev_kfree_skb_any(tx_buffer_info->skb);
643 tx_buffer_info->skb = NULL;
644 /* tx_buffer_info must be completely set up in the transmit path */
647 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
649 struct ixgbe_hw *hw = &adapter->hw;
650 struct ixgbe_hw_stats *hwstats = &adapter->stats;
655 if ((hw->fc.current_mode == ixgbe_fc_full) ||
656 (hw->fc.current_mode == ixgbe_fc_rx_pause)) {
657 switch (hw->mac.type) {
658 case ixgbe_mac_82598EB:
659 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
662 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
664 hwstats->lxoffrxc += data;
666 /* refill credits (no tx hang) if we received xoff */
670 for (i = 0; i < adapter->num_tx_queues; i++)
671 clear_bit(__IXGBE_HANG_CHECK_ARMED,
672 &adapter->tx_ring[i]->state);
674 } else if (!(adapter->dcb_cfg.pfc_mode_enable))
677 /* update stats for each tc, only valid with PFC enabled */
678 for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
679 switch (hw->mac.type) {
680 case ixgbe_mac_82598EB:
681 xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
684 xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
686 hwstats->pxoffrxc[i] += xoff[i];
689 /* disarm tx queues that have received xoff frames */
690 for (i = 0; i < adapter->num_tx_queues; i++) {
691 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
692 u8 tc = tx_ring->dcb_tc;
695 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
699 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
701 return ring->tx_stats.completed;
704 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
706 struct ixgbe_adapter *adapter = netdev_priv(ring->netdev);
707 struct ixgbe_hw *hw = &adapter->hw;
709 u32 head = IXGBE_READ_REG(hw, IXGBE_TDH(ring->reg_idx));
710 u32 tail = IXGBE_READ_REG(hw, IXGBE_TDT(ring->reg_idx));
713 return (head < tail) ?
714 tail - head : (tail + ring->count - head);
719 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
721 u32 tx_done = ixgbe_get_tx_completed(tx_ring);
722 u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
723 u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
726 clear_check_for_tx_hang(tx_ring);
729 * Check for a hung queue, but be thorough. This verifies
730 * that a transmit has been completed since the previous
731 * check AND there is at least one packet pending. The
732 * ARMED bit is set to indicate a potential hang. The
733 * bit is cleared if a pause frame is received to remove
734 * false hang detection due to PFC or 802.3x frames. By
735 * requiring this to fail twice we avoid races with
736 * pfc clearing the ARMED bit and conditions where we
737 * run the check_tx_hang logic with a transmit completion
738 * pending but without time to complete it yet.
740 if ((tx_done_old == tx_done) && tx_pending) {
741 /* make sure it is true for two checks in a row */
742 ret = test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
745 /* update completed stats and continue */
746 tx_ring->tx_stats.tx_done_old = tx_done;
747 /* reset the countdown */
748 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
755 * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
756 * @adapter: driver private struct
758 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
761 /* Do the reset outside of interrupt context */
762 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
763 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
764 ixgbe_service_event_schedule(adapter);
769 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
770 * @q_vector: structure containing interrupt and ring information
771 * @tx_ring: tx ring to clean
773 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
774 struct ixgbe_ring *tx_ring)
776 struct ixgbe_adapter *adapter = q_vector->adapter;
777 struct ixgbe_tx_buffer *tx_buffer;
778 union ixgbe_adv_tx_desc *tx_desc;
779 unsigned int total_bytes = 0, total_packets = 0;
780 unsigned int budget = q_vector->tx.work_limit;
781 u16 i = tx_ring->next_to_clean;
783 tx_buffer = &tx_ring->tx_buffer_info[i];
784 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
786 for (; budget; budget--) {
787 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
789 /* if next_to_watch is not set then there is no work pending */
793 /* if DD is not set pending work has not been completed */
794 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
797 /* count the packet as being completed */
798 tx_ring->tx_stats.completed++;
800 /* clear next_to_watch to prevent false hangs */
801 tx_buffer->next_to_watch = NULL;
803 /* prevent any other reads prior to eop_desc being verified */
807 ixgbe_unmap_tx_resource(tx_ring, tx_buffer);
808 tx_desc->wb.status = 0;
809 if (likely(tx_desc == eop_desc)) {
811 dev_kfree_skb_any(tx_buffer->skb);
812 tx_buffer->skb = NULL;
814 total_bytes += tx_buffer->bytecount;
815 total_packets += tx_buffer->gso_segs;
821 if (unlikely(i == tx_ring->count)) {
824 tx_buffer = tx_ring->tx_buffer_info;
825 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, 0);
831 tx_ring->next_to_clean = i;
832 u64_stats_update_begin(&tx_ring->syncp);
833 tx_ring->stats.bytes += total_bytes;
834 tx_ring->stats.packets += total_packets;
835 u64_stats_update_end(&tx_ring->syncp);
836 q_vector->tx.total_bytes += total_bytes;
837 q_vector->tx.total_packets += total_packets;
839 if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
840 /* schedule immediate reset if we believe we hung */
841 struct ixgbe_hw *hw = &adapter->hw;
842 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
843 e_err(drv, "Detected Tx Unit Hang\n"
845 " TDH, TDT <%x>, <%x>\n"
846 " next_to_use <%x>\n"
847 " next_to_clean <%x>\n"
848 "tx_buffer_info[next_to_clean]\n"
849 " time_stamp <%lx>\n"
851 tx_ring->queue_index,
852 IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
853 IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
854 tx_ring->next_to_use, i,
855 tx_ring->tx_buffer_info[i].time_stamp, jiffies);
857 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
860 "tx hang %d detected on queue %d, resetting adapter\n",
861 adapter->tx_timeout_count + 1, tx_ring->queue_index);
863 /* schedule immediate reset if we believe we hung */
864 ixgbe_tx_timeout_reset(adapter);
866 /* the adapter is about to reset, no point in enabling stuff */
870 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
871 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
872 (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
873 /* Make sure that anybody stopping the queue after this
874 * sees the new next_to_clean.
877 if (__netif_subqueue_stopped(tx_ring->netdev, tx_ring->queue_index) &&
878 !test_bit(__IXGBE_DOWN, &adapter->state)) {
879 netif_wake_subqueue(tx_ring->netdev, tx_ring->queue_index);
880 ++tx_ring->tx_stats.restart_queue;
887 #ifdef CONFIG_IXGBE_DCA
888 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
889 struct ixgbe_ring *rx_ring,
892 struct ixgbe_hw *hw = &adapter->hw;
894 u8 reg_idx = rx_ring->reg_idx;
896 rxctrl = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(reg_idx));
897 switch (hw->mac.type) {
898 case ixgbe_mac_82598EB:
899 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
900 rxctrl |= dca3_get_tag(rx_ring->dev, cpu);
902 case ixgbe_mac_82599EB:
904 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK_82599;
905 rxctrl |= (dca3_get_tag(rx_ring->dev, cpu) <<
906 IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599);
911 rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
912 rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
913 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_RRO_EN);
914 IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
917 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
918 struct ixgbe_ring *tx_ring,
921 struct ixgbe_hw *hw = &adapter->hw;
923 u8 reg_idx = tx_ring->reg_idx;
925 switch (hw->mac.type) {
926 case ixgbe_mac_82598EB:
927 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(reg_idx));
928 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
929 txctrl |= dca3_get_tag(tx_ring->dev, cpu);
930 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
931 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(reg_idx), txctrl);
933 case ixgbe_mac_82599EB:
935 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(reg_idx));
936 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK_82599;
937 txctrl |= (dca3_get_tag(tx_ring->dev, cpu) <<
938 IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599);
939 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
940 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(reg_idx), txctrl);
947 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
949 struct ixgbe_adapter *adapter = q_vector->adapter;
950 struct ixgbe_ring *ring;
953 if (q_vector->cpu == cpu)
956 for (ring = q_vector->tx.ring; ring != NULL; ring = ring->next)
957 ixgbe_update_tx_dca(adapter, ring, cpu);
959 for (ring = q_vector->rx.ring; ring != NULL; ring = ring->next)
960 ixgbe_update_rx_dca(adapter, ring, cpu);
967 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
972 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
975 /* always use CB2 mode, difference is masked in the CB driver */
976 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
978 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
979 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
983 for (i = 0; i < num_q_vectors; i++) {
984 adapter->q_vector[i]->cpu = -1;
985 ixgbe_update_dca(adapter->q_vector[i]);
989 static int __ixgbe_notify_dca(struct device *dev, void *data)
991 struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
992 unsigned long event = *(unsigned long *)data;
994 if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
998 case DCA_PROVIDER_ADD:
999 /* if we're already enabled, don't do it again */
1000 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1002 if (dca_add_requester(dev) == 0) {
1003 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1004 ixgbe_setup_dca(adapter);
1007 /* Fall Through since DCA is disabled. */
1008 case DCA_PROVIDER_REMOVE:
1009 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1010 dca_remove_requester(dev);
1011 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1012 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
1019 #endif /* CONFIG_IXGBE_DCA */
1021 static inline void ixgbe_rx_hash(union ixgbe_adv_rx_desc *rx_desc,
1022 struct sk_buff *skb)
1024 skb->rxhash = le32_to_cpu(rx_desc->wb.lower.hi_dword.rss);
1028 * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1029 * @adapter: address of board private structure
1030 * @rx_desc: advanced rx descriptor
1032 * Returns : true if it is FCoE pkt
1034 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_adapter *adapter,
1035 union ixgbe_adv_rx_desc *rx_desc)
1037 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1039 return (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
1040 ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1041 (cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1042 IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1046 * ixgbe_receive_skb - Send a completed packet up the stack
1047 * @adapter: board private structure
1048 * @skb: packet to send up
1049 * @status: hardware indication of status of receive
1050 * @rx_ring: rx descriptor ring (for a specific queue) to setup
1051 * @rx_desc: rx descriptor
1053 static void ixgbe_receive_skb(struct ixgbe_q_vector *q_vector,
1054 struct sk_buff *skb, u8 status,
1055 struct ixgbe_ring *ring,
1056 union ixgbe_adv_rx_desc *rx_desc)
1058 struct ixgbe_adapter *adapter = q_vector->adapter;
1059 struct napi_struct *napi = &q_vector->napi;
1060 bool is_vlan = (status & IXGBE_RXD_STAT_VP);
1061 u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
1063 if (is_vlan && (tag & VLAN_VID_MASK))
1064 __vlan_hwaccel_put_tag(skb, tag);
1066 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
1067 napi_gro_receive(napi, skb);
1073 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1074 * @adapter: address of board private structure
1075 * @status_err: hardware indication of status of receive
1076 * @skb: skb currently being received and modified
1077 * @status_err: status error value of last descriptor in packet
1079 static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
1080 union ixgbe_adv_rx_desc *rx_desc,
1081 struct sk_buff *skb,
1084 skb->ip_summed = CHECKSUM_NONE;
1086 /* Rx csum disabled */
1087 if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
1090 /* if IP and error */
1091 if ((status_err & IXGBE_RXD_STAT_IPCS) &&
1092 (status_err & IXGBE_RXDADV_ERR_IPE)) {
1093 adapter->hw_csum_rx_error++;
1097 if (!(status_err & IXGBE_RXD_STAT_L4CS))
1100 if (status_err & IXGBE_RXDADV_ERR_TCPE) {
1101 u16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1104 * 82599 errata, UDP frames with a 0 checksum can be marked as
1107 if ((pkt_info & IXGBE_RXDADV_PKTTYPE_UDP) &&
1108 (adapter->hw.mac.type == ixgbe_mac_82599EB))
1111 adapter->hw_csum_rx_error++;
1115 /* It must be a TCP or UDP packet with a valid checksum */
1116 skb->ip_summed = CHECKSUM_UNNECESSARY;
1119 static inline void ixgbe_release_rx_desc(struct ixgbe_ring *rx_ring, u32 val)
1122 * Force memory writes to complete before letting h/w
1123 * know there are new descriptors to fetch. (Only
1124 * applicable for weak-ordered memory model archs,
1128 writel(val, rx_ring->tail);
1132 * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
1133 * @rx_ring: ring to place buffers on
1134 * @cleaned_count: number of buffers to replace
1136 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1138 union ixgbe_adv_rx_desc *rx_desc;
1139 struct ixgbe_rx_buffer *bi;
1140 struct sk_buff *skb;
1141 u16 i = rx_ring->next_to_use;
1143 /* do nothing if no valid netdev defined */
1144 if (!rx_ring->netdev)
1147 while (cleaned_count--) {
1148 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1149 bi = &rx_ring->rx_buffer_info[i];
1153 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1154 rx_ring->rx_buf_len);
1156 rx_ring->rx_stats.alloc_rx_buff_failed++;
1159 /* initialize queue mapping */
1160 skb_record_rx_queue(skb, rx_ring->queue_index);
1165 bi->dma = dma_map_single(rx_ring->dev,
1167 rx_ring->rx_buf_len,
1169 if (dma_mapping_error(rx_ring->dev, bi->dma)) {
1170 rx_ring->rx_stats.alloc_rx_buff_failed++;
1176 if (ring_is_ps_enabled(rx_ring)) {
1178 bi->page = netdev_alloc_page(rx_ring->netdev);
1180 rx_ring->rx_stats.alloc_rx_page_failed++;
1185 if (!bi->page_dma) {
1186 /* use a half page if we're re-using */
1187 bi->page_offset ^= PAGE_SIZE / 2;
1188 bi->page_dma = dma_map_page(rx_ring->dev,
1193 if (dma_mapping_error(rx_ring->dev,
1195 rx_ring->rx_stats.alloc_rx_page_failed++;
1201 /* Refresh the desc even if buffer_addrs didn't change
1202 * because each write-back erases this info. */
1203 rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
1204 rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
1206 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
1207 rx_desc->read.hdr_addr = 0;
1211 if (i == rx_ring->count)
1216 if (rx_ring->next_to_use != i) {
1217 rx_ring->next_to_use = i;
1218 ixgbe_release_rx_desc(rx_ring, i);
1222 static inline u16 ixgbe_get_hlen(union ixgbe_adv_rx_desc *rx_desc)
1224 /* HW will not DMA in data larger than the given buffer, even if it
1225 * parses the (NFS, of course) header to be larger. In that case, it
1226 * fills the header buffer and spills the rest into the page.
1228 u16 hdr_info = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info);
1229 u16 hlen = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
1230 IXGBE_RXDADV_HDRBUFLEN_SHIFT;
1231 if (hlen > IXGBE_RX_HDR_SIZE)
1232 hlen = IXGBE_RX_HDR_SIZE;
1237 * ixgbe_transform_rsc_queue - change rsc queue into a full packet
1238 * @skb: pointer to the last skb in the rsc queue
1240 * This function changes a queue full of hw rsc buffers into a completed
1241 * packet. It uses the ->prev pointers to find the first packet and then
1242 * turns it into the frag list owner.
1244 static inline struct sk_buff *ixgbe_transform_rsc_queue(struct sk_buff *skb)
1246 unsigned int frag_list_size = 0;
1247 unsigned int skb_cnt = 1;
1250 struct sk_buff *prev = skb->prev;
1251 frag_list_size += skb->len;
1257 skb_shinfo(skb)->frag_list = skb->next;
1259 skb->len += frag_list_size;
1260 skb->data_len += frag_list_size;
1261 skb->truesize += frag_list_size;
1262 IXGBE_RSC_CB(skb)->skb_cnt = skb_cnt;
1267 static inline bool ixgbe_get_rsc_state(union ixgbe_adv_rx_desc *rx_desc)
1269 return !!(le32_to_cpu(rx_desc->wb.lower.lo_dword.data) &
1270 IXGBE_RXDADV_RSCCNT_MASK);
1273 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1274 struct ixgbe_ring *rx_ring,
1277 struct ixgbe_adapter *adapter = q_vector->adapter;
1278 union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
1279 struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
1280 struct sk_buff *skb;
1281 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1282 const int current_node = numa_node_id();
1285 #endif /* IXGBE_FCOE */
1288 u16 cleaned_count = 0;
1289 bool pkt_is_rsc = false;
1291 i = rx_ring->next_to_clean;
1292 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1293 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1295 while (staterr & IXGBE_RXD_STAT_DD) {
1298 rmb(); /* read descriptor and rx_buffer_info after status DD */
1300 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1302 skb = rx_buffer_info->skb;
1303 rx_buffer_info->skb = NULL;
1304 prefetch(skb->data);
1306 if (ring_is_rsc_enabled(rx_ring))
1307 pkt_is_rsc = ixgbe_get_rsc_state(rx_desc);
1309 /* if this is a skb from previous receive DMA will be 0 */
1310 if (rx_buffer_info->dma) {
1313 !(staterr & IXGBE_RXD_STAT_EOP) &&
1316 * When HWRSC is enabled, delay unmapping
1317 * of the first packet. It carries the
1318 * header information, HW may still
1319 * access the header after the writeback.
1320 * Only unmap it when EOP is reached
1322 IXGBE_RSC_CB(skb)->delay_unmap = true;
1323 IXGBE_RSC_CB(skb)->dma = rx_buffer_info->dma;
1325 dma_unmap_single(rx_ring->dev,
1326 rx_buffer_info->dma,
1327 rx_ring->rx_buf_len,
1330 rx_buffer_info->dma = 0;
1332 if (ring_is_ps_enabled(rx_ring)) {
1333 hlen = ixgbe_get_hlen(rx_desc);
1334 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1336 hlen = le16_to_cpu(rx_desc->wb.upper.length);
1341 /* assume packet split since header is unmapped */
1342 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1346 dma_unmap_page(rx_ring->dev,
1347 rx_buffer_info->page_dma,
1350 rx_buffer_info->page_dma = 0;
1351 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1352 rx_buffer_info->page,
1353 rx_buffer_info->page_offset,
1356 if ((page_count(rx_buffer_info->page) == 1) &&
1357 (page_to_nid(rx_buffer_info->page) == current_node))
1358 get_page(rx_buffer_info->page);
1360 rx_buffer_info->page = NULL;
1362 skb->len += upper_len;
1363 skb->data_len += upper_len;
1364 skb->truesize += upper_len;
1368 if (i == rx_ring->count)
1371 next_rxd = IXGBE_RX_DESC_ADV(rx_ring, i);
1376 u32 nextp = (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
1377 IXGBE_RXDADV_NEXTP_SHIFT;
1378 next_buffer = &rx_ring->rx_buffer_info[nextp];
1380 next_buffer = &rx_ring->rx_buffer_info[i];
1383 if (!(staterr & IXGBE_RXD_STAT_EOP)) {
1384 if (ring_is_ps_enabled(rx_ring)) {
1385 rx_buffer_info->skb = next_buffer->skb;
1386 rx_buffer_info->dma = next_buffer->dma;
1387 next_buffer->skb = skb;
1388 next_buffer->dma = 0;
1390 skb->next = next_buffer->skb;
1391 skb->next->prev = skb;
1393 rx_ring->rx_stats.non_eop_descs++;
1398 skb = ixgbe_transform_rsc_queue(skb);
1399 /* if we got here without RSC the packet is invalid */
1401 __pskb_trim(skb, 0);
1402 rx_buffer_info->skb = skb;
1407 if (ring_is_rsc_enabled(rx_ring)) {
1408 if (IXGBE_RSC_CB(skb)->delay_unmap) {
1409 dma_unmap_single(rx_ring->dev,
1410 IXGBE_RSC_CB(skb)->dma,
1411 rx_ring->rx_buf_len,
1413 IXGBE_RSC_CB(skb)->dma = 0;
1414 IXGBE_RSC_CB(skb)->delay_unmap = false;
1418 if (ring_is_ps_enabled(rx_ring))
1419 rx_ring->rx_stats.rsc_count +=
1420 skb_shinfo(skb)->nr_frags;
1422 rx_ring->rx_stats.rsc_count +=
1423 IXGBE_RSC_CB(skb)->skb_cnt;
1424 rx_ring->rx_stats.rsc_flush++;
1427 /* ERR_MASK will only have valid bits if EOP set */
1428 if (unlikely(staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK)) {
1429 dev_kfree_skb_any(skb);
1433 ixgbe_rx_checksum(adapter, rx_desc, skb, staterr);
1434 if (adapter->netdev->features & NETIF_F_RXHASH)
1435 ixgbe_rx_hash(rx_desc, skb);
1437 /* probably a little skewed due to removing CRC */
1438 total_rx_bytes += skb->len;
1441 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1443 /* if ddp, not passing to ULD unless for FCP_RSP or error */
1444 if (ixgbe_rx_is_fcoe(adapter, rx_desc)) {
1445 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb,
1448 dev_kfree_skb_any(skb);
1452 #endif /* IXGBE_FCOE */
1453 ixgbe_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
1457 rx_desc->wb.upper.status_error = 0;
1462 /* return some buffers to hardware, one at a time is too slow */
1463 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
1464 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1468 /* use prefetched values */
1470 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1473 rx_ring->next_to_clean = i;
1474 cleaned_count = ixgbe_desc_unused(rx_ring);
1477 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1480 /* include DDPed FCoE data */
1481 if (ddp_bytes > 0) {
1484 mss = rx_ring->netdev->mtu - sizeof(struct fcoe_hdr) -
1485 sizeof(struct fc_frame_header) -
1486 sizeof(struct fcoe_crc_eof);
1489 total_rx_bytes += ddp_bytes;
1490 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
1492 #endif /* IXGBE_FCOE */
1494 u64_stats_update_begin(&rx_ring->syncp);
1495 rx_ring->stats.packets += total_rx_packets;
1496 rx_ring->stats.bytes += total_rx_bytes;
1497 u64_stats_update_end(&rx_ring->syncp);
1498 q_vector->rx.total_packets += total_rx_packets;
1499 q_vector->rx.total_bytes += total_rx_bytes;
1505 * ixgbe_configure_msix - Configure MSI-X hardware
1506 * @adapter: board private structure
1508 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1511 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1513 struct ixgbe_q_vector *q_vector;
1514 int q_vectors, v_idx;
1517 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1519 /* Populate MSIX to EITR Select */
1520 if (adapter->num_vfs > 32) {
1521 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
1522 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
1526 * Populate the IVAR table and set the ITR values to the
1527 * corresponding register.
1529 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1530 struct ixgbe_ring *ring;
1531 q_vector = adapter->q_vector[v_idx];
1533 for (ring = q_vector->rx.ring; ring != NULL; ring = ring->next)
1534 ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1536 for (ring = q_vector->tx.ring; ring != NULL; ring = ring->next)
1537 ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1539 if (q_vector->tx.ring && !q_vector->rx.ring)
1541 q_vector->eitr = adapter->tx_eitr_param;
1542 else if (q_vector->rx.ring)
1544 q_vector->eitr = adapter->rx_eitr_param;
1546 ixgbe_write_eitr(q_vector);
1549 switch (adapter->hw.mac.type) {
1550 case ixgbe_mac_82598EB:
1551 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1554 case ixgbe_mac_82599EB:
1555 case ixgbe_mac_X540:
1556 ixgbe_set_ivar(adapter, -1, 1, v_idx);
1562 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1564 /* set up to autoclear timer, and the vectors */
1565 mask = IXGBE_EIMS_ENABLE_MASK;
1566 if (adapter->num_vfs)
1567 mask &= ~(IXGBE_EIMS_OTHER |
1568 IXGBE_EIMS_MAILBOX |
1571 mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
1572 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1575 enum latency_range {
1579 latency_invalid = 255
1583 * ixgbe_update_itr - update the dynamic ITR value based on statistics
1584 * @q_vector: structure containing interrupt and ring information
1585 * @ring_container: structure containing ring performance data
1587 * Stores a new ITR value based on packets and byte
1588 * counts during the last interrupt. The advantage of per interrupt
1589 * computation is faster updates and more accurate ITR for the current
1590 * traffic pattern. Constants in this function were computed
1591 * based on theoretical maximum wire speed and thresholds were set based
1592 * on testing data as well as attempting to minimize response time
1593 * while increasing bulk throughput.
1594 * this functionality is controlled by the InterruptThrottleRate module
1595 * parameter (see ixgbe_param.c)
1597 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
1598 struct ixgbe_ring_container *ring_container)
1601 struct ixgbe_adapter *adapter = q_vector->adapter;
1602 int bytes = ring_container->total_bytes;
1603 int packets = ring_container->total_packets;
1605 u8 itr_setting = ring_container->itr;
1610 /* simple throttlerate management
1611 * 0-20MB/s lowest (100000 ints/s)
1612 * 20-100MB/s low (20000 ints/s)
1613 * 100-1249MB/s bulk (8000 ints/s)
1615 /* what was last interrupt timeslice? */
1616 timepassed_us = 1000000/q_vector->eitr;
1617 bytes_perint = bytes / timepassed_us; /* bytes/usec */
1619 switch (itr_setting) {
1620 case lowest_latency:
1621 if (bytes_perint > adapter->eitr_low)
1622 itr_setting = low_latency;
1625 if (bytes_perint > adapter->eitr_high)
1626 itr_setting = bulk_latency;
1627 else if (bytes_perint <= adapter->eitr_low)
1628 itr_setting = lowest_latency;
1631 if (bytes_perint <= adapter->eitr_high)
1632 itr_setting = low_latency;
1636 /* clear work counters since we have the values we need */
1637 ring_container->total_bytes = 0;
1638 ring_container->total_packets = 0;
1640 /* write updated itr to ring container */
1641 ring_container->itr = itr_setting;
1645 * ixgbe_write_eitr - write EITR register in hardware specific way
1646 * @q_vector: structure containing interrupt and ring information
1648 * This function is made to be called by ethtool and by the driver
1649 * when it needs to update EITR registers at runtime. Hardware
1650 * specific quirks/differences are taken care of here.
1652 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1654 struct ixgbe_adapter *adapter = q_vector->adapter;
1655 struct ixgbe_hw *hw = &adapter->hw;
1656 int v_idx = q_vector->v_idx;
1657 u32 itr_reg = EITR_INTS_PER_SEC_TO_REG(q_vector->eitr);
1659 switch (adapter->hw.mac.type) {
1660 case ixgbe_mac_82598EB:
1661 /* must write high and low 16 bits to reset counter */
1662 itr_reg |= (itr_reg << 16);
1664 case ixgbe_mac_82599EB:
1665 case ixgbe_mac_X540:
1667 * 82599 and X540 can support a value of zero, so allow it for
1668 * max interrupt rate, but there is an errata where it can
1669 * not be zero with RSC
1672 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
1676 * set the WDIS bit to not clear the timer bits and cause an
1677 * immediate assertion of the interrupt
1679 itr_reg |= IXGBE_EITR_CNT_WDIS;
1684 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1687 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
1689 u32 new_itr = q_vector->eitr;
1692 ixgbe_update_itr(q_vector, &q_vector->tx);
1693 ixgbe_update_itr(q_vector, &q_vector->rx);
1695 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
1697 switch (current_itr) {
1698 /* counts and packets in update_itr are dependent on these numbers */
1699 case lowest_latency:
1703 new_itr = 20000; /* aka hwitr = ~200 */
1712 if (new_itr != q_vector->eitr) {
1713 /* do an exponential smoothing */
1714 new_itr = ((q_vector->eitr * 9) + new_itr)/10;
1716 /* save the algorithm value here */
1717 q_vector->eitr = new_itr;
1719 ixgbe_write_eitr(q_vector);
1724 * ixgbe_check_overtemp_subtask - check for over tempurature
1725 * @adapter: pointer to adapter
1727 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
1729 struct ixgbe_hw *hw = &adapter->hw;
1730 u32 eicr = adapter->interrupt_event;
1732 if (test_bit(__IXGBE_DOWN, &adapter->state))
1735 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
1736 !(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
1739 adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
1741 switch (hw->device_id) {
1742 case IXGBE_DEV_ID_82599_T3_LOM:
1744 * Since the warning interrupt is for both ports
1745 * we don't have to check if:
1746 * - This interrupt wasn't for our port.
1747 * - We may have missed the interrupt so always have to
1748 * check if we got a LSC
1750 if (!(eicr & IXGBE_EICR_GPI_SDP0) &&
1751 !(eicr & IXGBE_EICR_LSC))
1754 if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
1756 bool link_up = false;
1758 hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
1764 /* Check if this is not due to overtemp */
1765 if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
1770 if (!(eicr & IXGBE_EICR_GPI_SDP0))
1775 "Network adapter has been stopped because it has over heated. "
1776 "Restart the computer. If the problem persists, "
1777 "power off the system and replace the adapter\n");
1779 adapter->interrupt_event = 0;
1782 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
1784 struct ixgbe_hw *hw = &adapter->hw;
1786 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
1787 (eicr & IXGBE_EICR_GPI_SDP1)) {
1788 e_crit(probe, "Fan has stopped, replace the adapter\n");
1789 /* write to clear the interrupt */
1790 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1794 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
1796 struct ixgbe_hw *hw = &adapter->hw;
1798 if (eicr & IXGBE_EICR_GPI_SDP2) {
1799 /* Clear the interrupt */
1800 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
1801 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1802 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
1803 ixgbe_service_event_schedule(adapter);
1807 if (eicr & IXGBE_EICR_GPI_SDP1) {
1808 /* Clear the interrupt */
1809 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1810 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1811 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
1812 ixgbe_service_event_schedule(adapter);
1817 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1819 struct ixgbe_hw *hw = &adapter->hw;
1822 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1823 adapter->link_check_timeout = jiffies;
1824 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1825 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1826 IXGBE_WRITE_FLUSH(hw);
1827 ixgbe_service_event_schedule(adapter);
1831 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
1835 struct ixgbe_hw *hw = &adapter->hw;
1837 switch (hw->mac.type) {
1838 case ixgbe_mac_82598EB:
1839 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1840 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
1842 case ixgbe_mac_82599EB:
1843 case ixgbe_mac_X540:
1844 mask = (qmask & 0xFFFFFFFF);
1846 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
1847 mask = (qmask >> 32);
1849 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
1854 /* skip the flush */
1857 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
1861 struct ixgbe_hw *hw = &adapter->hw;
1863 switch (hw->mac.type) {
1864 case ixgbe_mac_82598EB:
1865 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1866 IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
1868 case ixgbe_mac_82599EB:
1869 case ixgbe_mac_X540:
1870 mask = (qmask & 0xFFFFFFFF);
1872 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
1873 mask = (qmask >> 32);
1875 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
1880 /* skip the flush */
1884 * ixgbe_irq_enable - Enable default interrupt generation settings
1885 * @adapter: board private structure
1887 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
1890 u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
1892 /* don't reenable LSC while waiting for link */
1893 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
1894 mask &= ~IXGBE_EIMS_LSC;
1896 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
1897 mask |= IXGBE_EIMS_GPI_SDP0;
1898 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
1899 mask |= IXGBE_EIMS_GPI_SDP1;
1900 switch (adapter->hw.mac.type) {
1901 case ixgbe_mac_82599EB:
1902 case ixgbe_mac_X540:
1903 mask |= IXGBE_EIMS_ECC;
1904 mask |= IXGBE_EIMS_GPI_SDP1;
1905 mask |= IXGBE_EIMS_GPI_SDP2;
1906 mask |= IXGBE_EIMS_MAILBOX;
1911 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
1912 !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
1913 mask |= IXGBE_EIMS_FLOW_DIR;
1915 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1917 ixgbe_irq_enable_queues(adapter, ~0);
1919 IXGBE_WRITE_FLUSH(&adapter->hw);
1922 static irqreturn_t ixgbe_msix_other(int irq, void *data)
1924 struct ixgbe_adapter *adapter = data;
1925 struct ixgbe_hw *hw = &adapter->hw;
1929 * Workaround for Silicon errata. Use clear-by-write instead
1930 * of clear-by-read. Reading with EICS will return the
1931 * interrupt causes without clearing, which later be done
1932 * with the write to EICR.
1934 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1935 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1937 if (eicr & IXGBE_EICR_LSC)
1938 ixgbe_check_lsc(adapter);
1940 if (eicr & IXGBE_EICR_MAILBOX)
1941 ixgbe_msg_task(adapter);
1943 switch (hw->mac.type) {
1944 case ixgbe_mac_82599EB:
1945 case ixgbe_mac_X540:
1946 if (eicr & IXGBE_EICR_ECC)
1947 e_info(link, "Received unrecoverable ECC Err, please "
1949 /* Handle Flow Director Full threshold interrupt */
1950 if (eicr & IXGBE_EICR_FLOW_DIR) {
1951 int reinit_count = 0;
1953 for (i = 0; i < adapter->num_tx_queues; i++) {
1954 struct ixgbe_ring *ring = adapter->tx_ring[i];
1955 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
1960 /* no more flow director interrupts until after init */
1961 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
1962 adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
1963 ixgbe_service_event_schedule(adapter);
1966 ixgbe_check_sfp_event(adapter, eicr);
1967 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
1968 ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC))) {
1969 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1970 adapter->interrupt_event = eicr;
1971 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
1972 ixgbe_service_event_schedule(adapter);
1980 ixgbe_check_fan_failure(adapter, eicr);
1982 /* re-enable the original interrupt state, no lsc, no queues */
1983 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1984 ixgbe_irq_enable(adapter, false, false);
1989 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
1991 struct ixgbe_q_vector *q_vector = data;
1993 /* EIAM disabled interrupts (on this vector) for us */
1995 if (q_vector->rx.ring || q_vector->tx.ring)
1996 napi_schedule(&q_vector->napi);
2001 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
2004 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2005 struct ixgbe_ring *rx_ring = a->rx_ring[r_idx];
2007 rx_ring->q_vector = q_vector;
2008 rx_ring->next = q_vector->rx.ring;
2009 q_vector->rx.ring = rx_ring;
2010 q_vector->rx.count++;
2013 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
2016 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2017 struct ixgbe_ring *tx_ring = a->tx_ring[t_idx];
2019 tx_ring->q_vector = q_vector;
2020 tx_ring->next = q_vector->tx.ring;
2021 q_vector->tx.ring = tx_ring;
2022 q_vector->tx.count++;
2023 q_vector->tx.work_limit = a->tx_work_limit;
2027 * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
2028 * @adapter: board private structure to initialize
2030 * This function maps descriptor rings to the queue-specific vectors
2031 * we were allotted through the MSI-X enabling code. Ideally, we'd have
2032 * one vector per ring/queue, but on a constrained vector budget, we
2033 * group the rings as "efficiently" as possible. You would add new
2034 * mapping configurations in here.
2036 static void ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter)
2038 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2039 int rxr_remaining = adapter->num_rx_queues, rxr_idx = 0;
2040 int txr_remaining = adapter->num_tx_queues, txr_idx = 0;
2043 /* only one q_vector if MSI-X is disabled. */
2044 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2048 * If we don't have enough vectors for a 1-to-1 mapping, we'll have to
2049 * group them so there are multiple queues per vector.
2051 * Re-adjusting *qpv takes care of the remainder.
2053 for (; v_start < q_vectors && rxr_remaining; v_start++) {
2054 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_start);
2055 for (; rqpv; rqpv--, rxr_idx++, rxr_remaining--)
2056 map_vector_to_rxq(adapter, v_start, rxr_idx);
2060 * If there are not enough q_vectors for each ring to have it's own
2061 * vector then we must pair up Rx/Tx on a each vector
2063 if ((v_start + txr_remaining) > q_vectors)
2066 for (; v_start < q_vectors && txr_remaining; v_start++) {
2067 int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_start);
2068 for (; tqpv; tqpv--, txr_idx++, txr_remaining--)
2069 map_vector_to_txq(adapter, v_start, txr_idx);
2074 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2075 * @adapter: board private structure
2077 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2078 * interrupts from the kernel.
2080 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2082 struct net_device *netdev = adapter->netdev;
2083 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2087 for (vector = 0; vector < q_vectors; vector++) {
2088 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2089 struct msix_entry *entry = &adapter->msix_entries[vector];
2091 if (q_vector->tx.ring && q_vector->rx.ring) {
2092 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2093 "%s-%s-%d", netdev->name, "TxRx", ri++);
2095 } else if (q_vector->rx.ring) {
2096 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2097 "%s-%s-%d", netdev->name, "rx", ri++);
2098 } else if (q_vector->tx.ring) {
2099 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2100 "%s-%s-%d", netdev->name, "tx", ti++);
2102 /* skip this unused q_vector */
2105 err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
2106 q_vector->name, q_vector);
2108 e_err(probe, "request_irq failed for MSIX interrupt "
2109 "Error: %d\n", err);
2110 goto free_queue_irqs;
2112 /* If Flow Director is enabled, set interrupt affinity */
2113 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2114 /* assign the mask for this irq */
2115 irq_set_affinity_hint(entry->vector,
2116 q_vector->affinity_mask);
2120 err = request_irq(adapter->msix_entries[vector].vector,
2121 ixgbe_msix_other, 0, netdev->name, adapter);
2123 e_err(probe, "request_irq for msix_lsc failed: %d\n", err);
2124 goto free_queue_irqs;
2132 irq_set_affinity_hint(adapter->msix_entries[vector].vector,
2134 free_irq(adapter->msix_entries[vector].vector,
2135 adapter->q_vector[vector]);
2137 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2138 pci_disable_msix(adapter->pdev);
2139 kfree(adapter->msix_entries);
2140 adapter->msix_entries = NULL;
2145 * ixgbe_intr - legacy mode Interrupt Handler
2146 * @irq: interrupt number
2147 * @data: pointer to a network interface device structure
2149 static irqreturn_t ixgbe_intr(int irq, void *data)
2151 struct ixgbe_adapter *adapter = data;
2152 struct ixgbe_hw *hw = &adapter->hw;
2153 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2157 * Workaround for silicon errata on 82598. Mask the interrupts
2158 * before the read of EICR.
2160 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2162 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2163 * therefore no explict interrupt disable is necessary */
2164 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2167 * shared interrupt alert!
2168 * make sure interrupts are enabled because the read will
2169 * have disabled interrupts due to EIAM
2170 * finish the workaround of silicon errata on 82598. Unmask
2171 * the interrupt that we masked before the EICR read.
2173 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2174 ixgbe_irq_enable(adapter, true, true);
2175 return IRQ_NONE; /* Not our interrupt */
2178 if (eicr & IXGBE_EICR_LSC)
2179 ixgbe_check_lsc(adapter);
2181 switch (hw->mac.type) {
2182 case ixgbe_mac_82599EB:
2183 ixgbe_check_sfp_event(adapter, eicr);
2184 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2185 ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC))) {
2186 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2187 adapter->interrupt_event = eicr;
2188 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2189 ixgbe_service_event_schedule(adapter);
2197 ixgbe_check_fan_failure(adapter, eicr);
2199 if (napi_schedule_prep(&(q_vector->napi))) {
2200 /* would disable interrupts here but EIAM disabled it */
2201 __napi_schedule(&(q_vector->napi));
2205 * re-enable link(maybe) and non-queue interrupts, no flush.
2206 * ixgbe_poll will re-enable the queue interrupts
2209 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2210 ixgbe_irq_enable(adapter, false, false);
2215 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
2217 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2220 /* legacy and MSI only use one vector */
2221 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2224 for (i = 0; i < adapter->num_rx_queues; i++) {
2225 adapter->rx_ring[i]->q_vector = NULL;
2226 adapter->rx_ring[i]->next = NULL;
2228 for (i = 0; i < adapter->num_tx_queues; i++) {
2229 adapter->tx_ring[i]->q_vector = NULL;
2230 adapter->tx_ring[i]->next = NULL;
2233 for (i = 0; i < q_vectors; i++) {
2234 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
2235 memset(&q_vector->rx, 0, sizeof(struct ixgbe_ring_container));
2236 memset(&q_vector->tx, 0, sizeof(struct ixgbe_ring_container));
2241 * ixgbe_request_irq - initialize interrupts
2242 * @adapter: board private structure
2244 * Attempts to configure interrupts using the best available
2245 * capabilities of the hardware and kernel.
2247 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2249 struct net_device *netdev = adapter->netdev;
2252 /* map all of the rings to the q_vectors */
2253 ixgbe_map_rings_to_vectors(adapter);
2255 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2256 err = ixgbe_request_msix_irqs(adapter);
2257 else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
2258 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2259 netdev->name, adapter);
2261 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2262 netdev->name, adapter);
2265 e_err(probe, "request_irq failed, Error %d\n", err);
2267 /* place q_vectors and rings back into a known good state */
2268 ixgbe_reset_q_vectors(adapter);
2274 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2276 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2279 q_vectors = adapter->num_msix_vectors;
2281 free_irq(adapter->msix_entries[i].vector, adapter);
2284 for (; i >= 0; i--) {
2285 /* free only the irqs that were actually requested */
2286 if (!adapter->q_vector[i]->rx.ring &&
2287 !adapter->q_vector[i]->tx.ring)
2290 /* clear the affinity_mask in the IRQ descriptor */
2291 irq_set_affinity_hint(adapter->msix_entries[i].vector,
2294 free_irq(adapter->msix_entries[i].vector,
2295 adapter->q_vector[i]);
2298 free_irq(adapter->pdev->irq, adapter);
2301 /* clear q_vector state information */
2302 ixgbe_reset_q_vectors(adapter);
2306 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2307 * @adapter: board private structure
2309 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2311 switch (adapter->hw.mac.type) {
2312 case ixgbe_mac_82598EB:
2313 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2315 case ixgbe_mac_82599EB:
2316 case ixgbe_mac_X540:
2317 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2318 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2319 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2324 IXGBE_WRITE_FLUSH(&adapter->hw);
2325 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2327 for (i = 0; i < adapter->num_msix_vectors; i++)
2328 synchronize_irq(adapter->msix_entries[i].vector);
2330 synchronize_irq(adapter->pdev->irq);
2335 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2338 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2340 struct ixgbe_hw *hw = &adapter->hw;
2342 IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
2343 EITR_INTS_PER_SEC_TO_REG(adapter->rx_eitr_param));
2345 ixgbe_set_ivar(adapter, 0, 0, 0);
2346 ixgbe_set_ivar(adapter, 1, 0, 0);
2348 e_info(hw, "Legacy interrupt IVAR setup done\n");
2352 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2353 * @adapter: board private structure
2354 * @ring: structure containing ring specific data
2356 * Configure the Tx descriptor ring after a reset.
2358 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2359 struct ixgbe_ring *ring)
2361 struct ixgbe_hw *hw = &adapter->hw;
2362 u64 tdba = ring->dma;
2364 u32 txdctl = IXGBE_TXDCTL_ENABLE;
2365 u8 reg_idx = ring->reg_idx;
2367 /* disable queue to avoid issues while updating state */
2368 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
2369 IXGBE_WRITE_FLUSH(hw);
2371 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2372 (tdba & DMA_BIT_MASK(32)));
2373 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2374 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2375 ring->count * sizeof(union ixgbe_adv_tx_desc));
2376 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2377 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2378 ring->tail = hw->hw_addr + IXGBE_TDT(reg_idx);
2381 * set WTHRESH to encourage burst writeback, it should not be set
2382 * higher than 1 when ITR is 0 as it could cause false TX hangs
2384 * In order to avoid issues WTHRESH + PTHRESH should always be equal
2385 * to or less than the number of on chip descriptors, which is
2388 if (!adapter->tx_itr_setting || !adapter->rx_itr_setting)
2389 txdctl |= (1 << 16); /* WTHRESH = 1 */
2391 txdctl |= (8 << 16); /* WTHRESH = 8 */
2393 /* PTHRESH=32 is needed to avoid a Tx hang with DFP enabled. */
2394 txdctl |= (1 << 8) | /* HTHRESH = 1 */
2395 32; /* PTHRESH = 32 */
2397 /* reinitialize flowdirector state */
2398 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2399 adapter->atr_sample_rate) {
2400 ring->atr_sample_rate = adapter->atr_sample_rate;
2401 ring->atr_count = 0;
2402 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
2404 ring->atr_sample_rate = 0;
2407 clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
2410 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
2412 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2413 if (hw->mac.type == ixgbe_mac_82598EB &&
2414 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2417 /* poll to verify queue is enabled */
2419 usleep_range(1000, 2000);
2420 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2421 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
2423 e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
2426 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
2428 struct ixgbe_hw *hw = &adapter->hw;
2431 u8 tcs = netdev_get_num_tc(adapter->netdev);
2433 if (hw->mac.type == ixgbe_mac_82598EB)
2436 /* disable the arbiter while setting MTQC */
2437 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2438 rttdcs |= IXGBE_RTTDCS_ARBDIS;
2439 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2441 /* set transmit pool layout */
2442 switch (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
2443 case (IXGBE_FLAG_SRIOV_ENABLED):
2444 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2445 (IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF));
2449 reg = IXGBE_MTQC_64Q_1PB;
2451 reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
2453 reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
2455 IXGBE_WRITE_REG(hw, IXGBE_MTQC, reg);
2457 /* Enable Security TX Buffer IFG for multiple pb */
2459 reg = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
2460 reg |= IXGBE_SECTX_DCB;
2461 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, reg);
2466 /* re-enable the arbiter */
2467 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2468 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2472 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2473 * @adapter: board private structure
2475 * Configure the Tx unit of the MAC after a reset.
2477 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2479 struct ixgbe_hw *hw = &adapter->hw;
2483 ixgbe_setup_mtqc(adapter);
2485 if (hw->mac.type != ixgbe_mac_82598EB) {
2486 /* DMATXCTL.EN must be before Tx queues are enabled */
2487 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2488 dmatxctl |= IXGBE_DMATXCTL_TE;
2489 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2492 /* Setup the HW Tx Head and Tail descriptor pointers */
2493 for (i = 0; i < adapter->num_tx_queues; i++)
2494 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
2497 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2499 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2500 struct ixgbe_ring *rx_ring)
2503 u8 reg_idx = rx_ring->reg_idx;
2505 switch (adapter->hw.mac.type) {
2506 case ixgbe_mac_82598EB: {
2507 struct ixgbe_ring_feature *feature = adapter->ring_feature;
2508 const int mask = feature[RING_F_RSS].mask;
2509 reg_idx = reg_idx & mask;
2512 case ixgbe_mac_82599EB:
2513 case ixgbe_mac_X540:
2518 srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx));
2520 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
2521 srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
2522 if (adapter->num_vfs)
2523 srrctl |= IXGBE_SRRCTL_DROP_EN;
2525 srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
2526 IXGBE_SRRCTL_BSIZEHDR_MASK;
2528 if (ring_is_ps_enabled(rx_ring)) {
2529 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
2530 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2532 srrctl |= (PAGE_SIZE / 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2534 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2536 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
2537 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2538 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2541 IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx), srrctl);
2544 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2546 struct ixgbe_hw *hw = &adapter->hw;
2547 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2548 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2549 0x6A3E67EA, 0x14364D17, 0x3BED200D};
2550 u32 mrqc = 0, reta = 0;
2553 u8 tcs = netdev_get_num_tc(adapter->netdev);
2554 int maxq = adapter->ring_feature[RING_F_RSS].indices;
2557 maxq = min(maxq, adapter->num_tx_queues / tcs);
2559 /* Fill out hash function seeds */
2560 for (i = 0; i < 10; i++)
2561 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2563 /* Fill out redirection table */
2564 for (i = 0, j = 0; i < 128; i++, j++) {
2567 /* reta = 4-byte sliding window of
2568 * 0x00..(indices-1)(indices-1)00..etc. */
2569 reta = (reta << 8) | (j * 0x11);
2571 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2574 /* Disable indicating checksum in descriptor, enables RSS hash */
2575 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2576 rxcsum |= IXGBE_RXCSUM_PCSD;
2577 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2579 if (adapter->hw.mac.type == ixgbe_mac_82598EB &&
2580 (adapter->flags & IXGBE_FLAG_RSS_ENABLED)) {
2581 mrqc = IXGBE_MRQC_RSSEN;
2583 int mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2584 | IXGBE_FLAG_SRIOV_ENABLED);
2587 case (IXGBE_FLAG_RSS_ENABLED):
2589 mrqc = IXGBE_MRQC_RSSEN;
2591 mrqc = IXGBE_MRQC_RTRSS4TCEN;
2593 mrqc = IXGBE_MRQC_RTRSS8TCEN;
2595 case (IXGBE_FLAG_SRIOV_ENABLED):
2596 mrqc = IXGBE_MRQC_VMDQEN;
2603 /* Perform hash on these packet types */
2604 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2605 | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2606 | IXGBE_MRQC_RSS_FIELD_IPV6
2607 | IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
2609 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2613 * ixgbe_configure_rscctl - enable RSC for the indicated ring
2614 * @adapter: address of board private structure
2615 * @index: index of ring to set
2617 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
2618 struct ixgbe_ring *ring)
2620 struct ixgbe_hw *hw = &adapter->hw;
2623 u8 reg_idx = ring->reg_idx;
2625 if (!ring_is_rsc_enabled(ring))
2628 rx_buf_len = ring->rx_buf_len;
2629 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
2630 rscctrl |= IXGBE_RSCCTL_RSCEN;
2632 * we must limit the number of descriptors so that the
2633 * total size of max desc * buf_len is not greater
2636 if (ring_is_ps_enabled(ring)) {
2637 #if (MAX_SKB_FRAGS > 16)
2638 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2639 #elif (MAX_SKB_FRAGS > 8)
2640 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2641 #elif (MAX_SKB_FRAGS > 4)
2642 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2644 rscctrl |= IXGBE_RSCCTL_MAXDESC_1;
2647 if (rx_buf_len < IXGBE_RXBUFFER_4096)
2648 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2649 else if (rx_buf_len < IXGBE_RXBUFFER_8192)
2650 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2652 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2654 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
2658 * ixgbe_set_uta - Set unicast filter table address
2659 * @adapter: board private structure
2661 * The unicast table address is a register array of 32-bit registers.
2662 * The table is meant to be used in a way similar to how the MTA is used
2663 * however due to certain limitations in the hardware it is necessary to
2664 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
2665 * enable bit to allow vlan tag stripping when promiscuous mode is enabled
2667 static void ixgbe_set_uta(struct ixgbe_adapter *adapter)
2669 struct ixgbe_hw *hw = &adapter->hw;
2672 /* The UTA table only exists on 82599 hardware and newer */
2673 if (hw->mac.type < ixgbe_mac_82599EB)
2676 /* we only need to do this if VMDq is enabled */
2677 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
2680 for (i = 0; i < 128; i++)
2681 IXGBE_WRITE_REG(hw, IXGBE_UTA(i), ~0);
2684 #define IXGBE_MAX_RX_DESC_POLL 10
2685 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
2686 struct ixgbe_ring *ring)
2688 struct ixgbe_hw *hw = &adapter->hw;
2689 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
2691 u8 reg_idx = ring->reg_idx;
2693 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2694 if (hw->mac.type == ixgbe_mac_82598EB &&
2695 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2699 usleep_range(1000, 2000);
2700 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2701 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
2704 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
2705 "the polling period\n", reg_idx);
2709 void ixgbe_disable_rx_queue(struct ixgbe_adapter *adapter,
2710 struct ixgbe_ring *ring)
2712 struct ixgbe_hw *hw = &adapter->hw;
2713 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
2715 u8 reg_idx = ring->reg_idx;
2717 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2718 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
2720 /* write value back with RXDCTL.ENABLE bit cleared */
2721 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
2723 if (hw->mac.type == ixgbe_mac_82598EB &&
2724 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2727 /* the hardware may take up to 100us to really disable the rx queue */
2730 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2731 } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
2734 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not cleared within "
2735 "the polling period\n", reg_idx);
2739 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
2740 struct ixgbe_ring *ring)
2742 struct ixgbe_hw *hw = &adapter->hw;
2743 u64 rdba = ring->dma;
2745 u8 reg_idx = ring->reg_idx;
2747 /* disable queue to avoid issues while updating state */
2748 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2749 ixgbe_disable_rx_queue(adapter, ring);
2751 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
2752 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
2753 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
2754 ring->count * sizeof(union ixgbe_adv_rx_desc));
2755 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
2756 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
2757 ring->tail = hw->hw_addr + IXGBE_RDT(reg_idx);
2759 ixgbe_configure_srrctl(adapter, ring);
2760 ixgbe_configure_rscctl(adapter, ring);
2762 /* If operating in IOV mode set RLPML for X540 */
2763 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
2764 hw->mac.type == ixgbe_mac_X540) {
2765 rxdctl &= ~IXGBE_RXDCTL_RLPMLMASK;
2766 rxdctl |= ((ring->netdev->mtu + ETH_HLEN +
2767 ETH_FCS_LEN + VLAN_HLEN) | IXGBE_RXDCTL_RLPML_EN);
2770 if (hw->mac.type == ixgbe_mac_82598EB) {
2772 * enable cache line friendly hardware writes:
2773 * PTHRESH=32 descriptors (half the internal cache),
2774 * this also removes ugly rx_no_buffer_count increment
2775 * HTHRESH=4 descriptors (to minimize latency on fetch)
2776 * WTHRESH=8 burst writeback up to two cache lines
2778 rxdctl &= ~0x3FFFFF;
2782 /* enable receive descriptor ring */
2783 rxdctl |= IXGBE_RXDCTL_ENABLE;
2784 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
2786 ixgbe_rx_desc_queue_enable(adapter, ring);
2787 ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
2790 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
2792 struct ixgbe_hw *hw = &adapter->hw;
2795 /* PSRTYPE must be initialized in non 82598 adapters */
2796 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
2797 IXGBE_PSRTYPE_UDPHDR |
2798 IXGBE_PSRTYPE_IPV4HDR |
2799 IXGBE_PSRTYPE_L2HDR |
2800 IXGBE_PSRTYPE_IPV6HDR;
2802 if (hw->mac.type == ixgbe_mac_82598EB)
2805 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED)
2806 psrtype |= (adapter->num_rx_queues_per_pool << 29);
2808 for (p = 0; p < adapter->num_rx_pools; p++)
2809 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(adapter->num_vfs + p),
2813 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
2815 struct ixgbe_hw *hw = &adapter->hw;
2818 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->antispoofing_enabled =
2856 (adapter->num_vfs != 0)),
2860 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
2862 struct ixgbe_hw *hw = &adapter->hw;
2863 struct net_device *netdev = adapter->netdev;
2864 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2866 struct ixgbe_ring *rx_ring;
2870 /* Decide whether to use packet split mode or not */
2872 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
2874 /* Do not use packet split if we're in SR-IOV Mode */
2875 if (adapter->num_vfs)
2876 adapter->flags &= ~IXGBE_FLAG_RX_PS_ENABLED;
2878 /* Disable packet split due to 82599 erratum #45 */
2879 if (hw->mac.type == ixgbe_mac_82599EB)
2880 adapter->flags &= ~IXGBE_FLAG_RX_PS_ENABLED;
2882 /* Set the RX buffer length according to the mode */
2883 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
2884 rx_buf_len = IXGBE_RX_HDR_SIZE;
2886 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
2887 (netdev->mtu <= ETH_DATA_LEN))
2888 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
2890 rx_buf_len = ALIGN(max_frame + VLAN_HLEN, 1024);
2894 /* adjust max frame to be able to do baby jumbo for FCoE */
2895 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
2896 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
2897 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
2899 #endif /* IXGBE_FCOE */
2900 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
2901 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
2902 mhadd &= ~IXGBE_MHADD_MFS_MASK;
2903 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
2905 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
2908 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2909 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
2910 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2911 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
2914 * Setup the HW Rx Head and Tail Descriptor Pointers and
2915 * the Base and Length of the Rx Descriptor Ring
2917 for (i = 0; i < adapter->num_rx_queues; i++) {
2918 rx_ring = adapter->rx_ring[i];
2919 rx_ring->rx_buf_len = rx_buf_len;
2921 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)
2922 set_ring_ps_enabled(rx_ring);
2924 clear_ring_ps_enabled(rx_ring);
2926 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
2927 set_ring_rsc_enabled(rx_ring);
2929 clear_ring_rsc_enabled(rx_ring);
2932 if (netdev->features & NETIF_F_FCOE_MTU) {
2933 struct ixgbe_ring_feature *f;
2934 f = &adapter->ring_feature[RING_F_FCOE];
2935 if ((i >= f->mask) && (i < f->mask + f->indices)) {
2936 clear_ring_ps_enabled(rx_ring);
2937 if (rx_buf_len < IXGBE_FCOE_JUMBO_FRAME_SIZE)
2938 rx_ring->rx_buf_len =
2939 IXGBE_FCOE_JUMBO_FRAME_SIZE;
2940 } else if (!ring_is_rsc_enabled(rx_ring) &&
2941 !ring_is_ps_enabled(rx_ring)) {
2942 rx_ring->rx_buf_len =
2943 IXGBE_FCOE_JUMBO_FRAME_SIZE;
2946 #endif /* IXGBE_FCOE */
2950 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
2952 struct ixgbe_hw *hw = &adapter->hw;
2953 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2955 switch (hw->mac.type) {
2956 case ixgbe_mac_82598EB:
2958 * For VMDq support of different descriptor types or
2959 * buffer sizes through the use of multiple SRRCTL
2960 * registers, RDRXCTL.MVMEN must be set to 1
2962 * also, the manual doesn't mention it clearly but DCA hints
2963 * will only use queue 0's tags unless this bit is set. Side
2964 * effects of setting this bit are only that SRRCTL must be
2965 * fully programmed [0..15]
2967 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
2969 case ixgbe_mac_82599EB:
2970 case ixgbe_mac_X540:
2971 /* Disable RSC for ACK packets */
2972 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
2973 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
2974 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
2975 /* hardware requires some bits to be set by default */
2976 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
2977 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
2980 /* We should do nothing since we don't know this hardware */
2984 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
2988 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
2989 * @adapter: board private structure
2991 * Configure the Rx unit of the MAC after a reset.
2993 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
2995 struct ixgbe_hw *hw = &adapter->hw;
2999 /* disable receives while setting up the descriptors */
3000 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3001 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3003 ixgbe_setup_psrtype(adapter);
3004 ixgbe_setup_rdrxctl(adapter);
3006 /* Program registers for the distribution of queues */
3007 ixgbe_setup_mrqc(adapter);
3009 ixgbe_set_uta(adapter);
3011 /* set_rx_buffer_len must be called before ring initialization */
3012 ixgbe_set_rx_buffer_len(adapter);
3015 * Setup the HW Rx Head and Tail Descriptor Pointers and
3016 * the Base and Length of the Rx Descriptor Ring
3018 for (i = 0; i < adapter->num_rx_queues; i++)
3019 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3021 /* disable drop enable for 82598 parts */
3022 if (hw->mac.type == ixgbe_mac_82598EB)
3023 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3025 /* enable all receives */
3026 rxctrl |= IXGBE_RXCTRL_RXEN;
3027 hw->mac.ops.enable_rx_dma(hw, rxctrl);
3030 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3032 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3033 struct ixgbe_hw *hw = &adapter->hw;
3034 int pool_ndx = adapter->num_vfs;
3036 /* add VID to filter table */
3037 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, true);
3038 set_bit(vid, adapter->active_vlans);
3041 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3043 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3044 struct ixgbe_hw *hw = &adapter->hw;
3045 int pool_ndx = adapter->num_vfs;
3047 /* remove VID from filter table */
3048 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, false);
3049 clear_bit(vid, adapter->active_vlans);
3053 * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3054 * @adapter: driver data
3056 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
3058 struct ixgbe_hw *hw = &adapter->hw;
3061 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3062 vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
3063 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3067 * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3068 * @adapter: driver data
3070 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
3072 struct ixgbe_hw *hw = &adapter->hw;
3075 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3076 vlnctrl |= IXGBE_VLNCTRL_VFE;
3077 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3078 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3082 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3083 * @adapter: driver data
3085 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3087 struct ixgbe_hw *hw = &adapter->hw;
3091 switch (hw->mac.type) {
3092 case ixgbe_mac_82598EB:
3093 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3094 vlnctrl &= ~IXGBE_VLNCTRL_VME;
3095 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3097 case ixgbe_mac_82599EB:
3098 case ixgbe_mac_X540:
3099 for (i = 0; i < adapter->num_rx_queues; i++) {
3100 j = adapter->rx_ring[i]->reg_idx;
3101 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3102 vlnctrl &= ~IXGBE_RXDCTL_VME;
3103 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3112 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3113 * @adapter: driver data
3115 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3117 struct ixgbe_hw *hw = &adapter->hw;
3121 switch (hw->mac.type) {
3122 case ixgbe_mac_82598EB:
3123 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3124 vlnctrl |= IXGBE_VLNCTRL_VME;
3125 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3127 case ixgbe_mac_82599EB:
3128 case ixgbe_mac_X540:
3129 for (i = 0; i < adapter->num_rx_queues; i++) {
3130 j = adapter->rx_ring[i]->reg_idx;
3131 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3132 vlnctrl |= IXGBE_RXDCTL_VME;
3133 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3141 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3145 ixgbe_vlan_rx_add_vid(adapter->netdev, 0);
3147 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3148 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
3152 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3153 * @netdev: network interface device structure
3155 * Writes unicast address list to the RAR table.
3156 * Returns: -ENOMEM on failure/insufficient address space
3157 * 0 on no addresses written
3158 * X on writing X addresses to the RAR table
3160 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3162 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3163 struct ixgbe_hw *hw = &adapter->hw;
3164 unsigned int vfn = adapter->num_vfs;
3165 unsigned int rar_entries = IXGBE_MAX_PF_MACVLANS;
3168 /* return ENOMEM indicating insufficient memory for addresses */
3169 if (netdev_uc_count(netdev) > rar_entries)
3172 if (!netdev_uc_empty(netdev) && rar_entries) {
3173 struct netdev_hw_addr *ha;
3174 /* return error if we do not support writing to RAR table */
3175 if (!hw->mac.ops.set_rar)
3178 netdev_for_each_uc_addr(ha, netdev) {
3181 hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3186 /* write the addresses in reverse order to avoid write combining */
3187 for (; rar_entries > 0 ; rar_entries--)
3188 hw->mac.ops.clear_rar(hw, rar_entries);
3194 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3195 * @netdev: network interface device structure
3197 * The set_rx_method entry point is called whenever the unicast/multicast
3198 * address list or the network interface flags are updated. This routine is
3199 * responsible for configuring the hardware for proper unicast, multicast and
3202 void ixgbe_set_rx_mode(struct net_device *netdev)
3204 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3205 struct ixgbe_hw *hw = &adapter->hw;
3206 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3209 /* Check for Promiscuous and All Multicast modes */
3211 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3213 /* set all bits that we expect to always be set */
3214 fctrl |= IXGBE_FCTRL_BAM;
3215 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3216 fctrl |= IXGBE_FCTRL_PMCF;
3218 /* clear the bits we are changing the status of */
3219 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3221 if (netdev->flags & IFF_PROMISC) {
3222 hw->addr_ctrl.user_set_promisc = true;
3223 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3224 vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3225 /* don't hardware filter vlans in promisc mode */
3226 ixgbe_vlan_filter_disable(adapter);
3228 if (netdev->flags & IFF_ALLMULTI) {
3229 fctrl |= IXGBE_FCTRL_MPE;
3230 vmolr |= IXGBE_VMOLR_MPE;
3233 * Write addresses to the MTA, if the attempt fails
3234 * then we should just turn on promiscuous mode so
3235 * that we can at least receive multicast traffic
3237 hw->mac.ops.update_mc_addr_list(hw, netdev);
3238 vmolr |= IXGBE_VMOLR_ROMPE;
3240 ixgbe_vlan_filter_enable(adapter);
3241 hw->addr_ctrl.user_set_promisc = false;
3243 * Write addresses to available RAR registers, if there is not
3244 * sufficient space to store all the addresses then enable
3245 * unicast promiscuous mode
3247 count = ixgbe_write_uc_addr_list(netdev);
3249 fctrl |= IXGBE_FCTRL_UPE;
3250 vmolr |= IXGBE_VMOLR_ROPE;
3254 if (adapter->num_vfs) {
3255 ixgbe_restore_vf_multicasts(adapter);
3256 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(adapter->num_vfs)) &
3257 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3259 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(adapter->num_vfs), vmolr);
3262 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3264 if (netdev->features & NETIF_F_HW_VLAN_RX)
3265 ixgbe_vlan_strip_enable(adapter);
3267 ixgbe_vlan_strip_disable(adapter);
3270 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3273 struct ixgbe_q_vector *q_vector;
3274 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3276 /* legacy and MSI only use one vector */
3277 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3280 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3281 q_vector = adapter->q_vector[q_idx];
3282 napi_enable(&q_vector->napi);
3286 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3289 struct ixgbe_q_vector *q_vector;
3290 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3292 /* legacy and MSI only use one vector */
3293 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3296 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3297 q_vector = adapter->q_vector[q_idx];
3298 napi_disable(&q_vector->napi);
3302 #ifdef CONFIG_IXGBE_DCB
3304 * ixgbe_configure_dcb - Configure DCB hardware
3305 * @adapter: ixgbe adapter struct
3307 * This is called by the driver on open to configure the DCB hardware.
3308 * This is also called by the gennetlink interface when reconfiguring
3311 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3313 struct ixgbe_hw *hw = &adapter->hw;
3314 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3316 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3317 if (hw->mac.type == ixgbe_mac_82598EB)
3318 netif_set_gso_max_size(adapter->netdev, 65536);
3322 if (hw->mac.type == ixgbe_mac_82598EB)
3323 netif_set_gso_max_size(adapter->netdev, 32768);
3326 /* Enable VLAN tag insert/strip */
3327 adapter->netdev->features |= NETIF_F_HW_VLAN_RX;
3329 hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
3331 /* reconfigure the hardware */
3332 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
3334 if (adapter->netdev->features & NETIF_F_FCOE_MTU)
3335 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
3337 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3339 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3341 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
3343 struct net_device *dev = adapter->netdev;
3345 if (adapter->ixgbe_ieee_ets)
3346 dev->dcbnl_ops->ieee_setets(dev,
3347 adapter->ixgbe_ieee_ets);
3348 if (adapter->ixgbe_ieee_pfc)
3349 dev->dcbnl_ops->ieee_setpfc(dev,
3350 adapter->ixgbe_ieee_pfc);
3353 /* Enable RSS Hash per TC */
3354 if (hw->mac.type != ixgbe_mac_82598EB) {
3358 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
3360 u8 cnt = adapter->netdev->tc_to_txq[i].count;
3365 reg |= msb << IXGBE_RQTC_SHIFT_TC(i);
3367 IXGBE_WRITE_REG(hw, IXGBE_RQTC, reg);
3373 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
3376 int num_tc = netdev_get_num_tc(adapter->netdev);
3377 struct ixgbe_hw *hw = &adapter->hw;
3379 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3380 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3381 hdrm = 64 << adapter->fdir_pballoc;
3383 hw->mac.ops.set_rxpba(&adapter->hw, num_tc, hdrm, PBA_STRATEGY_EQUAL);
3386 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
3388 struct ixgbe_hw *hw = &adapter->hw;
3389 struct hlist_node *node, *node2;
3390 struct ixgbe_fdir_filter *filter;
3392 spin_lock(&adapter->fdir_perfect_lock);
3394 if (!hlist_empty(&adapter->fdir_filter_list))
3395 ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
3397 hlist_for_each_entry_safe(filter, node, node2,
3398 &adapter->fdir_filter_list, fdir_node) {
3399 ixgbe_fdir_write_perfect_filter_82599(hw,
3402 (filter->action == IXGBE_FDIR_DROP_QUEUE) ?
3403 IXGBE_FDIR_DROP_QUEUE :
3404 adapter->rx_ring[filter->action]->reg_idx);
3407 spin_unlock(&adapter->fdir_perfect_lock);
3410 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3412 struct net_device *netdev = adapter->netdev;
3413 struct ixgbe_hw *hw = &adapter->hw;
3416 ixgbe_configure_pb(adapter);
3417 #ifdef CONFIG_IXGBE_DCB
3418 ixgbe_configure_dcb(adapter);
3421 ixgbe_set_rx_mode(netdev);
3422 ixgbe_restore_vlan(adapter);
3425 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
3426 ixgbe_configure_fcoe(adapter);
3428 #endif /* IXGBE_FCOE */
3429 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3430 for (i = 0; i < adapter->num_tx_queues; i++)
3431 adapter->tx_ring[i]->atr_sample_rate =
3432 adapter->atr_sample_rate;
3433 ixgbe_init_fdir_signature_82599(hw, adapter->fdir_pballoc);
3434 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3435 ixgbe_init_fdir_perfect_82599(&adapter->hw,
3436 adapter->fdir_pballoc);
3437 ixgbe_fdir_filter_restore(adapter);
3439 ixgbe_configure_virtualization(adapter);
3441 ixgbe_configure_tx(adapter);
3442 ixgbe_configure_rx(adapter);
3445 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3447 switch (hw->phy.type) {
3448 case ixgbe_phy_sfp_avago:
3449 case ixgbe_phy_sfp_ftl:
3450 case ixgbe_phy_sfp_intel:
3451 case ixgbe_phy_sfp_unknown:
3452 case ixgbe_phy_sfp_passive_tyco:
3453 case ixgbe_phy_sfp_passive_unknown:
3454 case ixgbe_phy_sfp_active_unknown:
3455 case ixgbe_phy_sfp_ftl_active:
3463 * ixgbe_sfp_link_config - set up SFP+ link
3464 * @adapter: pointer to private adapter struct
3466 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3469 * We are assuming the worst case scenerio here, and that
3470 * is that an SFP was inserted/removed after the reset
3471 * but before SFP detection was enabled. As such the best
3472 * solution is to just start searching as soon as we start
3474 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3475 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
3477 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
3481 * ixgbe_non_sfp_link_config - set up non-SFP+ link
3482 * @hw: pointer to private hardware struct
3484 * Returns 0 on success, negative on failure
3486 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3489 bool negotiation, link_up = false;
3490 u32 ret = IXGBE_ERR_LINK_SETUP;
3492 if (hw->mac.ops.check_link)
3493 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3498 autoneg = hw->phy.autoneg_advertised;
3499 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
3500 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
3505 if (hw->mac.ops.setup_link)
3506 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
3511 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
3513 struct ixgbe_hw *hw = &adapter->hw;
3516 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3517 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
3519 gpie |= IXGBE_GPIE_EIAME;
3521 * use EIAM to auto-mask when MSI-X interrupt is asserted
3522 * this saves a register write for every interrupt
3524 switch (hw->mac.type) {
3525 case ixgbe_mac_82598EB:
3526 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3528 case ixgbe_mac_82599EB:
3529 case ixgbe_mac_X540:
3531 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
3532 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
3536 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
3537 * specifically only auto mask tx and rx interrupts */
3538 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3541 /* XXX: to interrupt immediately for EICS writes, enable this */
3542 /* gpie |= IXGBE_GPIE_EIMEN; */
3544 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3545 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
3546 gpie |= IXGBE_GPIE_VTMODE_64;
3549 /* Enable fan failure interrupt */
3550 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3551 gpie |= IXGBE_SDP1_GPIEN;
3553 if (hw->mac.type == ixgbe_mac_82599EB) {
3554 gpie |= IXGBE_SDP1_GPIEN;
3555 gpie |= IXGBE_SDP2_GPIEN;
3558 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3561 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
3563 struct ixgbe_hw *hw = &adapter->hw;
3567 ixgbe_get_hw_control(adapter);
3568 ixgbe_setup_gpie(adapter);
3570 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3571 ixgbe_configure_msix(adapter);
3573 ixgbe_configure_msi_and_legacy(adapter);
3575 /* enable the optics for both mult-speed fiber and 82599 SFP+ fiber */
3576 if (hw->mac.ops.enable_tx_laser &&
3577 ((hw->phy.multispeed_fiber) ||
3578 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
3579 (hw->mac.type == ixgbe_mac_82599EB))))
3580 hw->mac.ops.enable_tx_laser(hw);
3582 clear_bit(__IXGBE_DOWN, &adapter->state);
3583 ixgbe_napi_enable_all(adapter);
3585 if (ixgbe_is_sfp(hw)) {
3586 ixgbe_sfp_link_config(adapter);
3588 err = ixgbe_non_sfp_link_config(hw);
3590 e_err(probe, "link_config FAILED %d\n", err);
3593 /* clear any pending interrupts, may auto mask */
3594 IXGBE_READ_REG(hw, IXGBE_EICR);
3595 ixgbe_irq_enable(adapter, true, true);
3598 * If this adapter has a fan, check to see if we had a failure
3599 * before we enabled the interrupt.
3601 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3602 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
3603 if (esdp & IXGBE_ESDP_SDP1)
3604 e_crit(drv, "Fan has stopped, replace the adapter\n");
3607 /* enable transmits */
3608 netif_tx_start_all_queues(adapter->netdev);
3610 /* bring the link up in the watchdog, this could race with our first
3611 * link up interrupt but shouldn't be a problem */
3612 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3613 adapter->link_check_timeout = jiffies;
3614 mod_timer(&adapter->service_timer, jiffies);
3616 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
3617 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
3618 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
3619 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
3624 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
3626 WARN_ON(in_interrupt());
3627 /* put off any impending NetWatchDogTimeout */
3628 adapter->netdev->trans_start = jiffies;
3630 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
3631 usleep_range(1000, 2000);
3632 ixgbe_down(adapter);
3634 * If SR-IOV enabled then wait a bit before bringing the adapter
3635 * back up to give the VFs time to respond to the reset. The
3636 * two second wait is based upon the watchdog timer cycle in
3639 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3642 clear_bit(__IXGBE_RESETTING, &adapter->state);
3645 int ixgbe_up(struct ixgbe_adapter *adapter)
3647 /* hardware has been reset, we need to reload some things */
3648 ixgbe_configure(adapter);
3650 return ixgbe_up_complete(adapter);
3653 void ixgbe_reset(struct ixgbe_adapter *adapter)
3655 struct ixgbe_hw *hw = &adapter->hw;
3658 /* lock SFP init bit to prevent race conditions with the watchdog */
3659 while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
3660 usleep_range(1000, 2000);
3662 /* clear all SFP and link config related flags while holding SFP_INIT */
3663 adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
3664 IXGBE_FLAG2_SFP_NEEDS_RESET);
3665 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
3667 err = hw->mac.ops.init_hw(hw);
3670 case IXGBE_ERR_SFP_NOT_PRESENT:
3671 case IXGBE_ERR_SFP_NOT_SUPPORTED:
3673 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
3674 e_dev_err("master disable timed out\n");
3676 case IXGBE_ERR_EEPROM_VERSION:
3677 /* We are running on a pre-production device, log a warning */
3678 e_dev_warn("This device is a pre-production adapter/LOM. "
3679 "Please be aware there may be issuesassociated with "
3680 "your hardware. If you are experiencing problems "
3681 "please contact your Intel or hardware "
3682 "representative who provided you with this "
3686 e_dev_err("Hardware Error: %d\n", err);
3689 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
3691 /* reprogram the RAR[0] in case user changed it. */
3692 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
3697 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
3698 * @rx_ring: ring to free buffers from
3700 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
3702 struct device *dev = rx_ring->dev;
3706 /* ring already cleared, nothing to do */
3707 if (!rx_ring->rx_buffer_info)
3710 /* Free all the Rx ring sk_buffs */
3711 for (i = 0; i < rx_ring->count; i++) {
3712 struct ixgbe_rx_buffer *rx_buffer_info;
3714 rx_buffer_info = &rx_ring->rx_buffer_info[i];
3715 if (rx_buffer_info->dma) {
3716 dma_unmap_single(rx_ring->dev, rx_buffer_info->dma,
3717 rx_ring->rx_buf_len,
3719 rx_buffer_info->dma = 0;
3721 if (rx_buffer_info->skb) {
3722 struct sk_buff *skb = rx_buffer_info->skb;
3723 rx_buffer_info->skb = NULL;
3725 struct sk_buff *this = skb;
3726 if (IXGBE_RSC_CB(this)->delay_unmap) {
3727 dma_unmap_single(dev,
3728 IXGBE_RSC_CB(this)->dma,
3729 rx_ring->rx_buf_len,
3731 IXGBE_RSC_CB(this)->dma = 0;
3732 IXGBE_RSC_CB(skb)->delay_unmap = false;
3735 dev_kfree_skb(this);
3738 if (!rx_buffer_info->page)
3740 if (rx_buffer_info->page_dma) {
3741 dma_unmap_page(dev, rx_buffer_info->page_dma,
3742 PAGE_SIZE / 2, DMA_FROM_DEVICE);
3743 rx_buffer_info->page_dma = 0;
3745 put_page(rx_buffer_info->page);
3746 rx_buffer_info->page = NULL;
3747 rx_buffer_info->page_offset = 0;
3750 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
3751 memset(rx_ring->rx_buffer_info, 0, size);
3753 /* Zero out the descriptor ring */
3754 memset(rx_ring->desc, 0, rx_ring->size);
3756 rx_ring->next_to_clean = 0;
3757 rx_ring->next_to_use = 0;
3761 * ixgbe_clean_tx_ring - Free Tx Buffers
3762 * @tx_ring: ring to be cleaned
3764 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
3766 struct ixgbe_tx_buffer *tx_buffer_info;
3770 /* ring already cleared, nothing to do */
3771 if (!tx_ring->tx_buffer_info)
3774 /* Free all the Tx ring sk_buffs */
3775 for (i = 0; i < tx_ring->count; i++) {
3776 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3777 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
3780 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
3781 memset(tx_ring->tx_buffer_info, 0, size);
3783 /* Zero out the descriptor ring */
3784 memset(tx_ring->desc, 0, tx_ring->size);
3786 tx_ring->next_to_use = 0;
3787 tx_ring->next_to_clean = 0;
3791 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
3792 * @adapter: board private structure
3794 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
3798 for (i = 0; i < adapter->num_rx_queues; i++)
3799 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
3803 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
3804 * @adapter: board private structure
3806 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
3810 for (i = 0; i < adapter->num_tx_queues; i++)
3811 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
3814 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
3816 struct hlist_node *node, *node2;
3817 struct ixgbe_fdir_filter *filter;
3819 spin_lock(&adapter->fdir_perfect_lock);
3821 hlist_for_each_entry_safe(filter, node, node2,
3822 &adapter->fdir_filter_list, fdir_node) {
3823 hlist_del(&filter->fdir_node);
3826 adapter->fdir_filter_count = 0;
3828 spin_unlock(&adapter->fdir_perfect_lock);
3831 void ixgbe_down(struct ixgbe_adapter *adapter)
3833 struct net_device *netdev = adapter->netdev;
3834 struct ixgbe_hw *hw = &adapter->hw;
3838 /* signal that we are down to the interrupt handler */
3839 set_bit(__IXGBE_DOWN, &adapter->state);
3841 /* disable receives */
3842 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3843 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3845 /* disable all enabled rx queues */
3846 for (i = 0; i < adapter->num_rx_queues; i++)
3847 /* this call also flushes the previous write */
3848 ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]);
3850 usleep_range(10000, 20000);
3852 netif_tx_stop_all_queues(netdev);
3854 /* call carrier off first to avoid false dev_watchdog timeouts */
3855 netif_carrier_off(netdev);
3856 netif_tx_disable(netdev);
3858 ixgbe_irq_disable(adapter);
3860 ixgbe_napi_disable_all(adapter);
3862 adapter->flags2 &= ~(IXGBE_FLAG2_FDIR_REQUIRES_REINIT |
3863 IXGBE_FLAG2_RESET_REQUESTED);
3864 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
3866 del_timer_sync(&adapter->service_timer);
3868 if (adapter->num_vfs) {
3869 /* Clear EITR Select mapping */
3870 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
3872 /* Mark all the VFs as inactive */
3873 for (i = 0 ; i < adapter->num_vfs; i++)
3874 adapter->vfinfo[i].clear_to_send = 0;
3876 /* ping all the active vfs to let them know we are going down */
3877 ixgbe_ping_all_vfs(adapter);
3879 /* Disable all VFTE/VFRE TX/RX */
3880 ixgbe_disable_tx_rx(adapter);
3883 /* disable transmits in the hardware now that interrupts are off */
3884 for (i = 0; i < adapter->num_tx_queues; i++) {
3885 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
3886 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
3889 /* Disable the Tx DMA engine on 82599 and X540 */
3890 switch (hw->mac.type) {
3891 case ixgbe_mac_82599EB:
3892 case ixgbe_mac_X540:
3893 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
3894 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
3895 ~IXGBE_DMATXCTL_TE));
3901 if (!pci_channel_offline(adapter->pdev))
3902 ixgbe_reset(adapter);
3904 /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
3905 if (hw->mac.ops.disable_tx_laser &&
3906 ((hw->phy.multispeed_fiber) ||
3907 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
3908 (hw->mac.type == ixgbe_mac_82599EB))))
3909 hw->mac.ops.disable_tx_laser(hw);
3911 ixgbe_clean_all_tx_rings(adapter);
3912 ixgbe_clean_all_rx_rings(adapter);
3914 #ifdef CONFIG_IXGBE_DCA
3915 /* since we reset the hardware DCA settings were cleared */
3916 ixgbe_setup_dca(adapter);
3921 * ixgbe_poll - NAPI Rx polling callback
3922 * @napi: structure for representing this polling device
3923 * @budget: how many packets driver is allowed to clean
3925 * This function is used for legacy and MSI, NAPI mode
3927 static int ixgbe_poll(struct napi_struct *napi, int budget)
3929 struct ixgbe_q_vector *q_vector =
3930 container_of(napi, struct ixgbe_q_vector, napi);
3931 struct ixgbe_adapter *adapter = q_vector->adapter;
3932 struct ixgbe_ring *ring;
3933 int per_ring_budget;
3934 bool clean_complete = true;
3936 #ifdef CONFIG_IXGBE_DCA
3937 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
3938 ixgbe_update_dca(q_vector);
3941 for (ring = q_vector->tx.ring; ring != NULL; ring = ring->next)
3942 clean_complete &= !!ixgbe_clean_tx_irq(q_vector, ring);
3944 /* attempt to distribute budget to each queue fairly, but don't allow
3945 * the budget to go below 1 because we'll exit polling */
3946 if (q_vector->rx.count > 1)
3947 per_ring_budget = max(budget/q_vector->rx.count, 1);
3949 per_ring_budget = budget;
3951 for (ring = q_vector->rx.ring; ring != NULL; ring = ring->next)
3952 clean_complete &= ixgbe_clean_rx_irq(q_vector, ring,
3955 /* If all work not completed, return budget and keep polling */
3956 if (!clean_complete)
3959 /* all work done, exit the polling mode */
3960 napi_complete(napi);
3961 if (adapter->rx_itr_setting & 1)
3962 ixgbe_set_itr(q_vector);
3963 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3964 ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
3970 * ixgbe_tx_timeout - Respond to a Tx Hang
3971 * @netdev: network interface device structure
3973 static void ixgbe_tx_timeout(struct net_device *netdev)
3975 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3977 /* Do the reset outside of interrupt context */
3978 ixgbe_tx_timeout_reset(adapter);
3982 * ixgbe_set_rss_queues: Allocate queues for RSS
3983 * @adapter: board private structure to initialize
3985 * This is our "base" multiqueue mode. RSS (Receive Side Scaling) will try
3986 * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
3989 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
3992 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
3994 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3996 adapter->num_rx_queues = f->indices;
3997 adapter->num_tx_queues = f->indices;
4007 * ixgbe_set_fdir_queues: Allocate queues for Flow Director
4008 * @adapter: board private structure to initialize
4010 * Flow Director is an advanced Rx filter, attempting to get Rx flows back
4011 * to the original CPU that initiated the Tx session. This runs in addition
4012 * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
4013 * Rx load across CPUs using RSS.
4016 static inline bool ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
4019 struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
4021 f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
4024 /* Flow Director must have RSS enabled */
4025 if ((adapter->flags & IXGBE_FLAG_RSS_ENABLED) &&
4026 (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)) {
4027 adapter->num_tx_queues = f_fdir->indices;
4028 adapter->num_rx_queues = f_fdir->indices;
4031 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4038 * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
4039 * @adapter: board private structure to initialize
4041 * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
4042 * The ring feature mask is not used as a mask for FCoE, as it can take any 8
4043 * rx queues out of the max number of rx queues, instead, it is used as the
4044 * index of the first rx queue used by FCoE.
4047 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
4049 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4051 if (!(adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
4054 f->indices = min((int)num_online_cpus(), f->indices);
4056 adapter->num_rx_queues = 1;
4057 adapter->num_tx_queues = 1;
4059 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4060 e_info(probe, "FCoE enabled with RSS\n");
4061 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)
4062 ixgbe_set_fdir_queues(adapter);
4064 ixgbe_set_rss_queues(adapter);
4067 /* adding FCoE rx rings to the end */
4068 f->mask = adapter->num_rx_queues;
4069 adapter->num_rx_queues += f->indices;
4070 adapter->num_tx_queues += f->indices;
4074 #endif /* IXGBE_FCOE */
4076 /* Artificial max queue cap per traffic class in DCB mode */
4077 #define DCB_QUEUE_CAP 8
4079 #ifdef CONFIG_IXGBE_DCB
4080 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
4082 int per_tc_q, q, i, offset = 0;
4083 struct net_device *dev = adapter->netdev;
4084 int tcs = netdev_get_num_tc(dev);
4089 /* Map queue offset and counts onto allocated tx queues */
4090 per_tc_q = min(dev->num_tx_queues / tcs, (unsigned int)DCB_QUEUE_CAP);
4091 q = min((int)num_online_cpus(), per_tc_q);
4093 for (i = 0; i < tcs; i++) {
4094 netdev_set_prio_tc_map(dev, i, i);
4095 netdev_set_tc_queue(dev, i, q, offset);
4099 adapter->num_tx_queues = q * tcs;
4100 adapter->num_rx_queues = q * tcs;
4103 /* FCoE enabled queues require special configuration indexed
4104 * by feature specific indices and mask. Here we map FCoE
4105 * indices onto the DCB queue pairs allowing FCoE to own
4106 * configuration later.
4108 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
4110 struct ixgbe_ring_feature *f =
4111 &adapter->ring_feature[RING_F_FCOE];
4113 tc = netdev_get_prio_tc_map(dev, adapter->fcoe.up);
4114 f->indices = dev->tc_to_txq[tc].count;
4115 f->mask = dev->tc_to_txq[tc].offset;
4124 * ixgbe_set_sriov_queues: Allocate queues for IOV use
4125 * @adapter: board private structure to initialize
4127 * IOV doesn't actually use anything, so just NAK the
4128 * request for now and let the other queue routines
4129 * figure out what to do.
4131 static inline bool ixgbe_set_sriov_queues(struct ixgbe_adapter *adapter)
4137 * ixgbe_set_num_queues: Allocate queues for device, feature dependent
4138 * @adapter: board private structure to initialize
4140 * This is the top level queue allocation routine. The order here is very
4141 * important, starting with the "most" number of features turned on at once,
4142 * and ending with the smallest set of features. This way large combinations
4143 * can be allocated if they're turned on, and smaller combinations are the
4144 * fallthrough conditions.
4147 static int ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
4149 /* Start with base case */
4150 adapter->num_rx_queues = 1;
4151 adapter->num_tx_queues = 1;
4152 adapter->num_rx_pools = adapter->num_rx_queues;
4153 adapter->num_rx_queues_per_pool = 1;
4155 if (ixgbe_set_sriov_queues(adapter))
4158 #ifdef CONFIG_IXGBE_DCB
4159 if (ixgbe_set_dcb_queues(adapter))
4164 if (ixgbe_set_fcoe_queues(adapter))
4167 #endif /* IXGBE_FCOE */
4168 if (ixgbe_set_fdir_queues(adapter))
4171 if (ixgbe_set_rss_queues(adapter))
4174 /* fallback to base case */
4175 adapter->num_rx_queues = 1;
4176 adapter->num_tx_queues = 1;
4179 /* Notify the stack of the (possibly) reduced queue counts. */
4180 netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
4181 return netif_set_real_num_rx_queues(adapter->netdev,
4182 adapter->num_rx_queues);
4185 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
4188 int err, vector_threshold;
4190 /* We'll want at least 3 (vector_threshold):
4193 * 3) Other (Link Status Change, etc.)
4194 * 4) TCP Timer (optional)
4196 vector_threshold = MIN_MSIX_COUNT;
4198 /* The more we get, the more we will assign to Tx/Rx Cleanup
4199 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
4200 * Right now, we simply care about how many we'll get; we'll
4201 * set them up later while requesting irq's.
4203 while (vectors >= vector_threshold) {
4204 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
4206 if (!err) /* Success in acquiring all requested vectors. */
4209 vectors = 0; /* Nasty failure, quit now */
4210 else /* err == number of vectors we should try again with */
4214 if (vectors < vector_threshold) {
4215 /* Can't allocate enough MSI-X interrupts? Oh well.
4216 * This just means we'll go with either a single MSI
4217 * vector or fall back to legacy interrupts.
4219 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4220 "Unable to allocate MSI-X interrupts\n");
4221 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4222 kfree(adapter->msix_entries);
4223 adapter->msix_entries = NULL;
4225 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
4227 * Adjust for only the vectors we'll use, which is minimum
4228 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
4229 * vectors we were allocated.
4231 adapter->num_msix_vectors = min(vectors,
4232 adapter->max_msix_q_vectors + NON_Q_VECTORS);
4237 * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
4238 * @adapter: board private structure to initialize
4240 * Cache the descriptor ring offsets for RSS to the assigned rings.
4243 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
4247 if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED))
4250 for (i = 0; i < adapter->num_rx_queues; i++)
4251 adapter->rx_ring[i]->reg_idx = i;
4252 for (i = 0; i < adapter->num_tx_queues; i++)
4253 adapter->tx_ring[i]->reg_idx = i;
4258 #ifdef CONFIG_IXGBE_DCB
4260 /* ixgbe_get_first_reg_idx - Return first register index associated with ring */
4261 static void ixgbe_get_first_reg_idx(struct ixgbe_adapter *adapter, u8 tc,
4262 unsigned int *tx, unsigned int *rx)
4264 struct net_device *dev = adapter->netdev;
4265 struct ixgbe_hw *hw = &adapter->hw;
4266 u8 num_tcs = netdev_get_num_tc(dev);
4271 switch (hw->mac.type) {
4272 case ixgbe_mac_82598EB:
4276 case ixgbe_mac_82599EB:
4277 case ixgbe_mac_X540:
4282 } else if (tc < 5) {
4283 *tx = ((tc + 2) << 4);
4285 } else if (tc < num_tcs) {
4286 *tx = ((tc + 8) << 3);
4315 * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
4316 * @adapter: board private structure to initialize
4318 * Cache the descriptor ring offsets for DCB to the assigned rings.
4321 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
4323 struct net_device *dev = adapter->netdev;
4325 u8 num_tcs = netdev_get_num_tc(dev);
4330 for (i = 0, k = 0; i < num_tcs; i++) {
4331 unsigned int tx_s, rx_s;
4332 u16 count = dev->tc_to_txq[i].count;
4334 ixgbe_get_first_reg_idx(adapter, i, &tx_s, &rx_s);
4335 for (j = 0; j < count; j++, k++) {
4336 adapter->tx_ring[k]->reg_idx = tx_s + j;
4337 adapter->rx_ring[k]->reg_idx = rx_s + j;
4338 adapter->tx_ring[k]->dcb_tc = i;
4339 adapter->rx_ring[k]->dcb_tc = i;
4348 * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
4349 * @adapter: board private structure to initialize
4351 * Cache the descriptor ring offsets for Flow Director to the assigned rings.
4354 static inline bool ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
4359 if ((adapter->flags & IXGBE_FLAG_RSS_ENABLED) &&
4360 (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)) {
4361 for (i = 0; i < adapter->num_rx_queues; i++)
4362 adapter->rx_ring[i]->reg_idx = i;
4363 for (i = 0; i < adapter->num_tx_queues; i++)
4364 adapter->tx_ring[i]->reg_idx = i;
4373 * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
4374 * @adapter: board private structure to initialize
4376 * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
4379 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
4381 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4383 u8 fcoe_rx_i = 0, fcoe_tx_i = 0;
4385 if (!(adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
4388 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4389 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)
4390 ixgbe_cache_ring_fdir(adapter);
4392 ixgbe_cache_ring_rss(adapter);
4394 fcoe_rx_i = f->mask;
4395 fcoe_tx_i = f->mask;
4397 for (i = 0; i < f->indices; i++, fcoe_rx_i++, fcoe_tx_i++) {
4398 adapter->rx_ring[f->mask + i]->reg_idx = fcoe_rx_i;
4399 adapter->tx_ring[f->mask + i]->reg_idx = fcoe_tx_i;
4404 #endif /* IXGBE_FCOE */
4406 * ixgbe_cache_ring_sriov - Descriptor ring to register mapping for sriov
4407 * @adapter: board private structure to initialize
4409 * SR-IOV doesn't use any descriptor rings but changes the default if
4410 * no other mapping is used.
4413 static inline bool ixgbe_cache_ring_sriov(struct ixgbe_adapter *adapter)
4415 adapter->rx_ring[0]->reg_idx = adapter->num_vfs * 2;
4416 adapter->tx_ring[0]->reg_idx = adapter->num_vfs * 2;
4417 if (adapter->num_vfs)
4424 * ixgbe_cache_ring_register - Descriptor ring to register mapping
4425 * @adapter: board private structure to initialize
4427 * Once we know the feature-set enabled for the device, we'll cache
4428 * the register offset the descriptor ring is assigned to.
4430 * Note, the order the various feature calls is important. It must start with
4431 * the "most" features enabled at the same time, then trickle down to the
4432 * least amount of features turned on at once.
4434 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
4436 /* start with default case */
4437 adapter->rx_ring[0]->reg_idx = 0;
4438 adapter->tx_ring[0]->reg_idx = 0;
4440 if (ixgbe_cache_ring_sriov(adapter))
4443 #ifdef CONFIG_IXGBE_DCB
4444 if (ixgbe_cache_ring_dcb(adapter))
4449 if (ixgbe_cache_ring_fcoe(adapter))
4451 #endif /* IXGBE_FCOE */
4453 if (ixgbe_cache_ring_fdir(adapter))
4456 if (ixgbe_cache_ring_rss(adapter))
4461 * ixgbe_alloc_queues - Allocate memory for all rings
4462 * @adapter: board private structure to initialize
4464 * We allocate one ring per queue at run-time since we don't know the
4465 * number of queues at compile-time. The polling_netdev array is
4466 * intended for Multiqueue, but should work fine with a single queue.
4468 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
4470 int rx = 0, tx = 0, nid = adapter->node;
4472 if (nid < 0 || !node_online(nid))
4473 nid = first_online_node;
4475 for (; tx < adapter->num_tx_queues; tx++) {
4476 struct ixgbe_ring *ring;
4478 ring = kzalloc_node(sizeof(*ring), GFP_KERNEL, nid);
4480 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
4482 goto err_allocation;
4483 ring->count = adapter->tx_ring_count;
4484 ring->queue_index = tx;
4485 ring->numa_node = nid;
4486 ring->dev = &adapter->pdev->dev;
4487 ring->netdev = adapter->netdev;
4489 adapter->tx_ring[tx] = ring;
4492 for (; rx < adapter->num_rx_queues; rx++) {
4493 struct ixgbe_ring *ring;
4495 ring = kzalloc_node(sizeof(*ring), GFP_KERNEL, nid);
4497 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
4499 goto err_allocation;
4500 ring->count = adapter->rx_ring_count;
4501 ring->queue_index = rx;
4502 ring->numa_node = nid;
4503 ring->dev = &adapter->pdev->dev;
4504 ring->netdev = adapter->netdev;
4506 adapter->rx_ring[rx] = ring;
4509 ixgbe_cache_ring_register(adapter);
4515 kfree(adapter->tx_ring[--tx]);
4518 kfree(adapter->rx_ring[--rx]);
4523 * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
4524 * @adapter: board private structure to initialize
4526 * Attempt to configure the interrupts using the best available
4527 * capabilities of the hardware and the kernel.
4529 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
4531 struct ixgbe_hw *hw = &adapter->hw;
4533 int vector, v_budget;
4536 * It's easy to be greedy for MSI-X vectors, but it really
4537 * doesn't do us much good if we have a lot more vectors
4538 * than CPU's. So let's be conservative and only ask for
4539 * (roughly) the same number of vectors as there are CPU's.
4541 v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
4542 (int)num_online_cpus()) + NON_Q_VECTORS;
4545 * At the same time, hardware can only support a maximum of
4546 * hw.mac->max_msix_vectors vectors. With features
4547 * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
4548 * descriptor queues supported by our device. Thus, we cap it off in
4549 * those rare cases where the cpu count also exceeds our vector limit.
4551 v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
4553 /* A failure in MSI-X entry allocation isn't fatal, but it does
4554 * mean we disable MSI-X capabilities of the adapter. */
4555 adapter->msix_entries = kcalloc(v_budget,
4556 sizeof(struct msix_entry), GFP_KERNEL);
4557 if (adapter->msix_entries) {
4558 for (vector = 0; vector < v_budget; vector++)
4559 adapter->msix_entries[vector].entry = vector;
4561 ixgbe_acquire_msix_vectors(adapter, v_budget);
4563 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4567 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
4568 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
4569 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
4571 "ATR is not supported while multiple "
4572 "queues are disabled. Disabling Flow Director\n");
4574 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4575 adapter->atr_sample_rate = 0;
4576 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4577 ixgbe_disable_sriov(adapter);
4579 err = ixgbe_set_num_queues(adapter);
4583 err = pci_enable_msi(adapter->pdev);
4585 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
4587 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4588 "Unable to allocate MSI interrupt, "
4589 "falling back to legacy. Error: %d\n", err);
4599 * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
4600 * @adapter: board private structure to initialize
4602 * We allocate one q_vector per queue interrupt. If allocation fails we
4605 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
4607 int v_idx, num_q_vectors;
4608 struct ixgbe_q_vector *q_vector;
4610 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4611 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4615 for (v_idx = 0; v_idx < num_q_vectors; v_idx++) {
4616 q_vector = kzalloc_node(sizeof(struct ixgbe_q_vector),
4617 GFP_KERNEL, adapter->node);
4619 q_vector = kzalloc(sizeof(struct ixgbe_q_vector),
4624 q_vector->adapter = adapter;
4625 q_vector->v_idx = v_idx;
4627 /* Allocate the affinity_hint cpumask, configure the mask */
4628 if (!alloc_cpumask_var(&q_vector->affinity_mask, GFP_KERNEL))
4630 cpumask_set_cpu(v_idx, q_vector->affinity_mask);
4632 if (q_vector->tx.count && !q_vector->rx.count)
4633 q_vector->eitr = adapter->tx_eitr_param;
4635 q_vector->eitr = adapter->rx_eitr_param;
4637 netif_napi_add(adapter->netdev, &q_vector->napi,
4639 adapter->q_vector[v_idx] = q_vector;
4647 q_vector = adapter->q_vector[v_idx];
4648 netif_napi_del(&q_vector->napi);
4649 free_cpumask_var(q_vector->affinity_mask);
4651 adapter->q_vector[v_idx] = NULL;
4657 * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
4658 * @adapter: board private structure to initialize
4660 * This function frees the memory allocated to the q_vectors. In addition if
4661 * NAPI is enabled it will delete any references to the NAPI struct prior
4662 * to freeing the q_vector.
4664 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
4666 int v_idx, num_q_vectors;
4668 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4669 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4673 for (v_idx = 0; v_idx < num_q_vectors; v_idx++) {
4674 struct ixgbe_q_vector *q_vector = adapter->q_vector[v_idx];
4675 adapter->q_vector[v_idx] = NULL;
4676 netif_napi_del(&q_vector->napi);
4677 free_cpumask_var(q_vector->affinity_mask);
4682 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
4684 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4685 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4686 pci_disable_msix(adapter->pdev);
4687 kfree(adapter->msix_entries);
4688 adapter->msix_entries = NULL;
4689 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
4690 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
4691 pci_disable_msi(adapter->pdev);
4696 * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
4697 * @adapter: board private structure to initialize
4699 * We determine which interrupt scheme to use based on...
4700 * - Kernel support (MSI, MSI-X)
4701 * - which can be user-defined (via MODULE_PARAM)
4702 * - Hardware queue count (num_*_queues)
4703 * - defined by miscellaneous hardware support/features (RSS, etc.)
4705 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
4709 /* Number of supported queues */
4710 err = ixgbe_set_num_queues(adapter);
4714 err = ixgbe_set_interrupt_capability(adapter);
4716 e_dev_err("Unable to setup interrupt capabilities\n");
4717 goto err_set_interrupt;
4720 err = ixgbe_alloc_q_vectors(adapter);
4722 e_dev_err("Unable to allocate memory for queue vectors\n");
4723 goto err_alloc_q_vectors;
4726 err = ixgbe_alloc_queues(adapter);
4728 e_dev_err("Unable to allocate memory for queues\n");
4729 goto err_alloc_queues;
4732 e_dev_info("Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
4733 (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
4734 adapter->num_rx_queues, adapter->num_tx_queues);
4736 set_bit(__IXGBE_DOWN, &adapter->state);
4741 ixgbe_free_q_vectors(adapter);
4742 err_alloc_q_vectors:
4743 ixgbe_reset_interrupt_capability(adapter);
4749 * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
4750 * @adapter: board private structure to clear interrupt scheme on
4752 * We go through and clear interrupt specific resources and reset the structure
4753 * to pre-load conditions
4755 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
4759 for (i = 0; i < adapter->num_tx_queues; i++) {
4760 kfree(adapter->tx_ring[i]);
4761 adapter->tx_ring[i] = NULL;
4763 for (i = 0; i < adapter->num_rx_queues; i++) {
4764 struct ixgbe_ring *ring = adapter->rx_ring[i];
4766 /* ixgbe_get_stats64() might access this ring, we must wait
4767 * a grace period before freeing it.
4769 kfree_rcu(ring, rcu);
4770 adapter->rx_ring[i] = NULL;
4773 adapter->num_tx_queues = 0;
4774 adapter->num_rx_queues = 0;
4776 ixgbe_free_q_vectors(adapter);
4777 ixgbe_reset_interrupt_capability(adapter);
4781 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4782 * @adapter: board private structure to initialize
4784 * ixgbe_sw_init initializes the Adapter private data structure.
4785 * Fields are initialized based on PCI device information and
4786 * OS network device settings (MTU size).
4788 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
4790 struct ixgbe_hw *hw = &adapter->hw;
4791 struct pci_dev *pdev = adapter->pdev;
4792 struct net_device *dev = adapter->netdev;
4794 #ifdef CONFIG_IXGBE_DCB
4796 struct tc_configuration *tc;
4798 int max_frame = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
4800 /* PCI config space info */
4802 hw->vendor_id = pdev->vendor;
4803 hw->device_id = pdev->device;
4804 hw->revision_id = pdev->revision;
4805 hw->subsystem_vendor_id = pdev->subsystem_vendor;
4806 hw->subsystem_device_id = pdev->subsystem_device;
4808 /* Set capability flags */
4809 rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
4810 adapter->ring_feature[RING_F_RSS].indices = rss;
4811 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
4812 switch (hw->mac.type) {
4813 case ixgbe_mac_82598EB:
4814 if (hw->device_id == IXGBE_DEV_ID_82598AT)
4815 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4816 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
4818 case ixgbe_mac_82599EB:
4819 case ixgbe_mac_X540:
4820 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
4821 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4822 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4823 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
4824 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4825 /* Flow Director hash filters enabled */
4826 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
4827 adapter->atr_sample_rate = 20;
4828 adapter->ring_feature[RING_F_FDIR].indices =
4829 IXGBE_MAX_FDIR_INDICES;
4830 adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
4832 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4833 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4834 adapter->ring_feature[RING_F_FCOE].indices = 0;
4835 #ifdef CONFIG_IXGBE_DCB
4836 /* Default traffic class to use for FCoE */
4837 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
4839 #endif /* IXGBE_FCOE */
4845 /* n-tuple support exists, always init our spinlock */
4846 spin_lock_init(&adapter->fdir_perfect_lock);
4848 #ifdef CONFIG_IXGBE_DCB
4849 /* Configure DCB traffic classes */
4850 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
4851 tc = &adapter->dcb_cfg.tc_config[j];
4852 tc->path[DCB_TX_CONFIG].bwg_id = 0;
4853 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
4854 tc->path[DCB_RX_CONFIG].bwg_id = 0;
4855 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
4856 tc->dcb_pfc = pfc_disabled;
4858 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
4859 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
4860 adapter->dcb_cfg.pfc_mode_enable = false;
4861 adapter->dcb_set_bitmap = 0x00;
4862 adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
4863 ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
4868 /* default flow control settings */
4869 hw->fc.requested_mode = ixgbe_fc_full;
4870 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
4872 adapter->last_lfc_mode = hw->fc.current_mode;
4874 hw->fc.high_water = FC_HIGH_WATER(max_frame);
4875 hw->fc.low_water = FC_LOW_WATER(max_frame);
4876 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
4877 hw->fc.send_xon = true;
4878 hw->fc.disable_fc_autoneg = false;
4880 /* enable itr by default in dynamic mode */
4881 adapter->rx_itr_setting = 1;
4882 adapter->rx_eitr_param = 20000;
4883 adapter->tx_itr_setting = 1;
4884 adapter->tx_eitr_param = 10000;
4886 /* set defaults for eitr in MegaBytes */
4887 adapter->eitr_low = 10;
4888 adapter->eitr_high = 20;
4890 /* set default ring sizes */
4891 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
4892 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
4894 /* set default work limits */
4895 adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
4897 /* initialize eeprom parameters */
4898 if (ixgbe_init_eeprom_params_generic(hw)) {
4899 e_dev_err("EEPROM initialization failed\n");
4903 /* enable rx csum by default */
4904 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
4906 /* get assigned NUMA node */
4907 adapter->node = dev_to_node(&pdev->dev);
4909 set_bit(__IXGBE_DOWN, &adapter->state);
4915 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
4916 * @tx_ring: tx descriptor ring (for a specific queue) to setup
4918 * Return 0 on success, negative on failure
4920 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
4922 struct device *dev = tx_ring->dev;
4925 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4926 tx_ring->tx_buffer_info = vzalloc_node(size, tx_ring->numa_node);
4927 if (!tx_ring->tx_buffer_info)
4928 tx_ring->tx_buffer_info = vzalloc(size);
4929 if (!tx_ring->tx_buffer_info)
4932 /* round up to nearest 4K */
4933 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
4934 tx_ring->size = ALIGN(tx_ring->size, 4096);
4936 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
4937 &tx_ring->dma, GFP_KERNEL);
4941 tx_ring->next_to_use = 0;
4942 tx_ring->next_to_clean = 0;
4946 vfree(tx_ring->tx_buffer_info);
4947 tx_ring->tx_buffer_info = NULL;
4948 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
4953 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
4954 * @adapter: board private structure
4956 * If this function returns with an error, then it's possible one or
4957 * more of the rings is populated (while the rest are not). It is the
4958 * callers duty to clean those orphaned rings.
4960 * Return 0 on success, negative on failure
4962 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
4966 for (i = 0; i < adapter->num_tx_queues; i++) {
4967 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
4970 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
4978 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
4979 * @rx_ring: rx descriptor ring (for a specific queue) to setup
4981 * Returns 0 on success, negative on failure
4983 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
4985 struct device *dev = rx_ring->dev;
4988 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4989 rx_ring->rx_buffer_info = vzalloc_node(size, rx_ring->numa_node);
4990 if (!rx_ring->rx_buffer_info)
4991 rx_ring->rx_buffer_info = vzalloc(size);
4992 if (!rx_ring->rx_buffer_info)
4995 /* Round up to nearest 4K */
4996 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
4997 rx_ring->size = ALIGN(rx_ring->size, 4096);
4999 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
5000 &rx_ring->dma, GFP_KERNEL);
5005 rx_ring->next_to_clean = 0;
5006 rx_ring->next_to_use = 0;
5010 vfree(rx_ring->rx_buffer_info);
5011 rx_ring->rx_buffer_info = NULL;
5012 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
5017 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
5018 * @adapter: board private structure
5020 * If this function returns with an error, then it's possible one or
5021 * more of the rings is populated (while the rest are not). It is the
5022 * callers duty to clean those orphaned rings.
5024 * Return 0 on success, negative on failure
5026 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
5030 for (i = 0; i < adapter->num_rx_queues; i++) {
5031 err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
5034 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
5042 * ixgbe_free_tx_resources - Free Tx Resources per Queue
5043 * @tx_ring: Tx descriptor ring for a specific queue
5045 * Free all transmit software resources
5047 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
5049 ixgbe_clean_tx_ring(tx_ring);
5051 vfree(tx_ring->tx_buffer_info);
5052 tx_ring->tx_buffer_info = NULL;
5054 /* if not set, then don't free */
5058 dma_free_coherent(tx_ring->dev, tx_ring->size,
5059 tx_ring->desc, tx_ring->dma);
5061 tx_ring->desc = NULL;
5065 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5066 * @adapter: board private structure
5068 * Free all transmit software resources
5070 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
5074 for (i = 0; i < adapter->num_tx_queues; i++)
5075 if (adapter->tx_ring[i]->desc)
5076 ixgbe_free_tx_resources(adapter->tx_ring[i]);
5080 * ixgbe_free_rx_resources - Free Rx Resources
5081 * @rx_ring: ring to clean the resources from
5083 * Free all receive software resources
5085 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
5087 ixgbe_clean_rx_ring(rx_ring);
5089 vfree(rx_ring->rx_buffer_info);
5090 rx_ring->rx_buffer_info = NULL;
5092 /* if not set, then don't free */
5096 dma_free_coherent(rx_ring->dev, rx_ring->size,
5097 rx_ring->desc, rx_ring->dma);
5099 rx_ring->desc = NULL;
5103 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5104 * @adapter: board private structure
5106 * Free all receive software resources
5108 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5112 for (i = 0; i < adapter->num_rx_queues; i++)
5113 if (adapter->rx_ring[i]->desc)
5114 ixgbe_free_rx_resources(adapter->rx_ring[i]);
5118 * ixgbe_change_mtu - Change the Maximum Transfer Unit
5119 * @netdev: network interface device structure
5120 * @new_mtu: new value for maximum frame size
5122 * Returns 0 on success, negative on failure
5124 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5126 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5127 struct ixgbe_hw *hw = &adapter->hw;
5128 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5130 /* MTU < 68 is an error and causes problems on some kernels */
5131 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED &&
5132 hw->mac.type != ixgbe_mac_X540) {
5133 if ((new_mtu < 68) || (max_frame > MAXIMUM_ETHERNET_VLAN_SIZE))
5136 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5140 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5141 /* must set new MTU before calling down or up */
5142 netdev->mtu = new_mtu;
5144 hw->fc.high_water = FC_HIGH_WATER(max_frame);
5145 hw->fc.low_water = FC_LOW_WATER(max_frame);
5147 if (netif_running(netdev))
5148 ixgbe_reinit_locked(adapter);
5154 * ixgbe_open - Called when a network interface is made active
5155 * @netdev: network interface device structure
5157 * Returns 0 on success, negative value on failure
5159 * The open entry point is called when a network interface is made
5160 * active by the system (IFF_UP). At this point all resources needed
5161 * for transmit and receive operations are allocated, the interrupt
5162 * handler is registered with the OS, the watchdog timer is started,
5163 * and the stack is notified that the interface is ready.
5165 static int ixgbe_open(struct net_device *netdev)
5167 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5170 /* disallow open during test */
5171 if (test_bit(__IXGBE_TESTING, &adapter->state))
5174 netif_carrier_off(netdev);
5176 /* allocate transmit descriptors */
5177 err = ixgbe_setup_all_tx_resources(adapter);
5181 /* allocate receive descriptors */
5182 err = ixgbe_setup_all_rx_resources(adapter);
5186 ixgbe_configure(adapter);
5188 err = ixgbe_request_irq(adapter);
5192 err = ixgbe_up_complete(adapter);
5196 netif_tx_start_all_queues(netdev);
5201 ixgbe_release_hw_control(adapter);
5202 ixgbe_free_irq(adapter);
5205 ixgbe_free_all_rx_resources(adapter);
5207 ixgbe_free_all_tx_resources(adapter);
5208 ixgbe_reset(adapter);
5214 * ixgbe_close - Disables a network interface
5215 * @netdev: network interface device structure
5217 * Returns 0, this is not allowed to fail
5219 * The close entry point is called when an interface is de-activated
5220 * by the OS. The hardware is still under the drivers control, but
5221 * needs to be disabled. A global MAC reset is issued to stop the
5222 * hardware, and all transmit and receive resources are freed.
5224 static int ixgbe_close(struct net_device *netdev)
5226 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5228 ixgbe_down(adapter);
5229 ixgbe_free_irq(adapter);
5231 ixgbe_fdir_filter_exit(adapter);
5233 ixgbe_free_all_tx_resources(adapter);
5234 ixgbe_free_all_rx_resources(adapter);
5236 ixgbe_release_hw_control(adapter);
5242 static int ixgbe_resume(struct pci_dev *pdev)
5244 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5245 struct net_device *netdev = adapter->netdev;
5248 pci_set_power_state(pdev, PCI_D0);
5249 pci_restore_state(pdev);
5251 * pci_restore_state clears dev->state_saved so call
5252 * pci_save_state to restore it.
5254 pci_save_state(pdev);
5256 err = pci_enable_device_mem(pdev);
5258 e_dev_err("Cannot enable PCI device from suspend\n");
5261 pci_set_master(pdev);
5263 pci_wake_from_d3(pdev, false);
5265 err = ixgbe_init_interrupt_scheme(adapter);
5267 e_dev_err("Cannot initialize interrupts for device\n");
5271 ixgbe_reset(adapter);
5273 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5275 if (netif_running(netdev)) {
5276 err = ixgbe_open(netdev);
5281 netif_device_attach(netdev);
5285 #endif /* CONFIG_PM */
5287 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5289 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5290 struct net_device *netdev = adapter->netdev;
5291 struct ixgbe_hw *hw = &adapter->hw;
5293 u32 wufc = adapter->wol;
5298 netif_device_detach(netdev);
5300 if (netif_running(netdev)) {
5301 ixgbe_down(adapter);
5302 ixgbe_free_irq(adapter);
5303 ixgbe_free_all_tx_resources(adapter);
5304 ixgbe_free_all_rx_resources(adapter);
5307 ixgbe_clear_interrupt_scheme(adapter);
5309 kfree(adapter->ixgbe_ieee_pfc);
5310 kfree(adapter->ixgbe_ieee_ets);
5314 retval = pci_save_state(pdev);
5320 ixgbe_set_rx_mode(netdev);
5322 /* turn on all-multi mode if wake on multicast is enabled */
5323 if (wufc & IXGBE_WUFC_MC) {
5324 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5325 fctrl |= IXGBE_FCTRL_MPE;
5326 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5329 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5330 ctrl |= IXGBE_CTRL_GIO_DIS;
5331 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5333 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5335 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5336 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5339 switch (hw->mac.type) {
5340 case ixgbe_mac_82598EB:
5341 pci_wake_from_d3(pdev, false);
5343 case ixgbe_mac_82599EB:
5344 case ixgbe_mac_X540:
5345 pci_wake_from_d3(pdev, !!wufc);
5351 *enable_wake = !!wufc;
5353 ixgbe_release_hw_control(adapter);
5355 pci_disable_device(pdev);
5361 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5366 retval = __ixgbe_shutdown(pdev, &wake);
5371 pci_prepare_to_sleep(pdev);
5373 pci_wake_from_d3(pdev, false);
5374 pci_set_power_state(pdev, PCI_D3hot);
5379 #endif /* CONFIG_PM */
5381 static void ixgbe_shutdown(struct pci_dev *pdev)
5385 __ixgbe_shutdown(pdev, &wake);
5387 if (system_state == SYSTEM_POWER_OFF) {
5388 pci_wake_from_d3(pdev, wake);
5389 pci_set_power_state(pdev, PCI_D3hot);
5394 * ixgbe_update_stats - Update the board statistics counters.
5395 * @adapter: board private structure
5397 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5399 struct net_device *netdev = adapter->netdev;
5400 struct ixgbe_hw *hw = &adapter->hw;
5401 struct ixgbe_hw_stats *hwstats = &adapter->stats;
5403 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5404 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
5405 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
5406 u64 bytes = 0, packets = 0;
5408 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5409 test_bit(__IXGBE_RESETTING, &adapter->state))
5412 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5415 for (i = 0; i < 16; i++)
5416 adapter->hw_rx_no_dma_resources +=
5417 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5418 for (i = 0; i < adapter->num_rx_queues; i++) {
5419 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
5420 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
5422 adapter->rsc_total_count = rsc_count;
5423 adapter->rsc_total_flush = rsc_flush;
5426 for (i = 0; i < adapter->num_rx_queues; i++) {
5427 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
5428 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
5429 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
5430 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
5431 bytes += rx_ring->stats.bytes;
5432 packets += rx_ring->stats.packets;
5434 adapter->non_eop_descs = non_eop_descs;
5435 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5436 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5437 netdev->stats.rx_bytes = bytes;
5438 netdev->stats.rx_packets = packets;
5442 /* gather some stats to the adapter struct that are per queue */
5443 for (i = 0; i < adapter->num_tx_queues; i++) {
5444 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5445 restart_queue += tx_ring->tx_stats.restart_queue;
5446 tx_busy += tx_ring->tx_stats.tx_busy;
5447 bytes += tx_ring->stats.bytes;
5448 packets += tx_ring->stats.packets;
5450 adapter->restart_queue = restart_queue;
5451 adapter->tx_busy = tx_busy;
5452 netdev->stats.tx_bytes = bytes;
5453 netdev->stats.tx_packets = packets;
5455 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5456 for (i = 0; i < 8; i++) {
5457 /* for packet buffers not used, the register should read 0 */
5458 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5460 hwstats->mpc[i] += mpc;
5461 total_mpc += hwstats->mpc[i];
5462 if (hw->mac.type == ixgbe_mac_82598EB)
5463 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5464 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5465 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5466 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5467 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5468 switch (hw->mac.type) {
5469 case ixgbe_mac_82598EB:
5470 hwstats->pxonrxc[i] +=
5471 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5473 case ixgbe_mac_82599EB:
5474 case ixgbe_mac_X540:
5475 hwstats->pxonrxc[i] +=
5476 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5481 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5482 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5484 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5485 /* work around hardware counting issue */
5486 hwstats->gprc -= missed_rx;
5488 ixgbe_update_xoff_received(adapter);
5490 /* 82598 hardware only has a 32 bit counter in the high register */
5491 switch (hw->mac.type) {
5492 case ixgbe_mac_82598EB:
5493 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5494 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5495 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5496 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5498 case ixgbe_mac_X540:
5499 /* OS2BMC stats are X540 only*/
5500 hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
5501 hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
5502 hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
5503 hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
5504 case ixgbe_mac_82599EB:
5505 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5506 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
5507 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5508 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
5509 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5510 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5511 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5512 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5513 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5515 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5516 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5517 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5518 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5519 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5520 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5521 #endif /* IXGBE_FCOE */
5526 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5527 hwstats->bprc += bprc;
5528 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5529 if (hw->mac.type == ixgbe_mac_82598EB)
5530 hwstats->mprc -= bprc;
5531 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5532 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5533 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5534 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5535 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5536 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5537 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5538 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5539 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5540 hwstats->lxontxc += lxon;
5541 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5542 hwstats->lxofftxc += lxoff;
5543 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5544 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5545 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5547 * 82598 errata - tx of flow control packets is included in tx counters
5549 xon_off_tot = lxon + lxoff;
5550 hwstats->gptc -= xon_off_tot;
5551 hwstats->mptc -= xon_off_tot;
5552 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5553 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5554 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5555 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5556 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5557 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5558 hwstats->ptc64 -= xon_off_tot;
5559 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5560 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5561 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5562 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5563 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5564 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5566 /* Fill out the OS statistics structure */
5567 netdev->stats.multicast = hwstats->mprc;
5570 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
5571 netdev->stats.rx_dropped = 0;
5572 netdev->stats.rx_length_errors = hwstats->rlec;
5573 netdev->stats.rx_crc_errors = hwstats->crcerrs;
5574 netdev->stats.rx_missed_errors = total_mpc;
5578 * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
5579 * @adapter - pointer to the device adapter structure
5581 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
5583 struct ixgbe_hw *hw = &adapter->hw;
5586 if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
5589 adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
5591 /* if interface is down do nothing */
5592 if (test_bit(__IXGBE_DOWN, &adapter->state))
5595 /* do nothing if we are not using signature filters */
5596 if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
5599 adapter->fdir_overflow++;
5601 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5602 for (i = 0; i < adapter->num_tx_queues; i++)
5603 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
5604 &(adapter->tx_ring[i]->state));
5605 /* re-enable flow director interrupts */
5606 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
5608 e_err(probe, "failed to finish FDIR re-initialization, "
5609 "ignored adding FDIR ATR filters\n");
5614 * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
5615 * @adapter - pointer to the device adapter structure
5617 * This function serves two purposes. First it strobes the interrupt lines
5618 * in order to make certain interrupts are occuring. Secondly it sets the
5619 * bits needed to check for TX hangs. As a result we should immediately
5620 * determine if a hang has occured.
5622 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
5624 struct ixgbe_hw *hw = &adapter->hw;
5628 /* If we're down or resetting, just bail */
5629 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5630 test_bit(__IXGBE_RESETTING, &adapter->state))
5633 /* Force detection of hung controller */
5634 if (netif_carrier_ok(adapter->netdev)) {
5635 for (i = 0; i < adapter->num_tx_queues; i++)
5636 set_check_for_tx_hang(adapter->tx_ring[i]);
5639 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5641 * for legacy and MSI interrupts don't set any bits
5642 * that are enabled for EIAM, because this operation
5643 * would set *both* EIMS and EICS for any bit in EIAM
5645 IXGBE_WRITE_REG(hw, IXGBE_EICS,
5646 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5648 /* get one bit for every active tx/rx interrupt vector */
5649 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
5650 struct ixgbe_q_vector *qv = adapter->q_vector[i];
5651 if (qv->rx.ring || qv->tx.ring)
5652 eics |= ((u64)1 << i);
5656 /* Cause software interrupt to ensure rings are cleaned */
5657 ixgbe_irq_rearm_queues(adapter, eics);
5662 * ixgbe_watchdog_update_link - update the link status
5663 * @adapter - pointer to the device adapter structure
5664 * @link_speed - pointer to a u32 to store the link_speed
5666 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
5668 struct ixgbe_hw *hw = &adapter->hw;
5669 u32 link_speed = adapter->link_speed;
5670 bool link_up = adapter->link_up;
5673 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
5676 if (hw->mac.ops.check_link) {
5677 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5679 /* always assume link is up, if no check link function */
5680 link_speed = IXGBE_LINK_SPEED_10GB_FULL;
5684 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5685 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
5686 hw->mac.ops.fc_enable(hw, i);
5688 hw->mac.ops.fc_enable(hw, 0);
5693 time_after(jiffies, (adapter->link_check_timeout +
5694 IXGBE_TRY_LINK_TIMEOUT))) {
5695 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5696 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5697 IXGBE_WRITE_FLUSH(hw);
5700 adapter->link_up = link_up;
5701 adapter->link_speed = link_speed;
5705 * ixgbe_watchdog_link_is_up - update netif_carrier status and
5706 * print link up message
5707 * @adapter - pointer to the device adapter structure
5709 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
5711 struct net_device *netdev = adapter->netdev;
5712 struct ixgbe_hw *hw = &adapter->hw;
5713 u32 link_speed = adapter->link_speed;
5714 bool flow_rx, flow_tx;
5716 /* only continue if link was previously down */
5717 if (netif_carrier_ok(netdev))
5720 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
5722 switch (hw->mac.type) {
5723 case ixgbe_mac_82598EB: {
5724 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5725 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5726 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5727 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5730 case ixgbe_mac_X540:
5731 case ixgbe_mac_82599EB: {
5732 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5733 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5734 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5735 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5743 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
5744 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5746 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5748 (link_speed == IXGBE_LINK_SPEED_100_FULL ?
5751 ((flow_rx && flow_tx) ? "RX/TX" :
5753 (flow_tx ? "TX" : "None"))));
5755 netif_carrier_on(netdev);
5756 ixgbe_check_vf_rate_limit(adapter);
5760 * ixgbe_watchdog_link_is_down - update netif_carrier status and
5761 * print link down message
5762 * @adapter - pointer to the adapter structure
5764 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter* adapter)
5766 struct net_device *netdev = adapter->netdev;
5767 struct ixgbe_hw *hw = &adapter->hw;
5769 adapter->link_up = false;
5770 adapter->link_speed = 0;
5772 /* only continue if link was up previously */
5773 if (!netif_carrier_ok(netdev))
5776 /* poll for SFP+ cable when link is down */
5777 if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
5778 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5780 e_info(drv, "NIC Link is Down\n");
5781 netif_carrier_off(netdev);
5785 * ixgbe_watchdog_flush_tx - flush queues on link down
5786 * @adapter - pointer to the device adapter structure
5788 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
5791 int some_tx_pending = 0;
5793 if (!netif_carrier_ok(adapter->netdev)) {
5794 for (i = 0; i < adapter->num_tx_queues; i++) {
5795 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5796 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5797 some_tx_pending = 1;
5802 if (some_tx_pending) {
5803 /* We've lost link, so the controller stops DMA,
5804 * but we've got queued Tx work that's never going
5805 * to get done, so reset controller to flush Tx.
5806 * (Do the reset outside of interrupt context).
5808 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
5813 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
5817 /* Do not perform spoof check for 82598 */
5818 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
5821 ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
5824 * ssvpc register is cleared on read, if zero then no
5825 * spoofed packets in the last interval.
5830 e_warn(drv, "%d Spoofed packets detected\n", ssvpc);
5834 * ixgbe_watchdog_subtask - check and bring link up
5835 * @adapter - pointer to the device adapter structure
5837 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
5839 /* if interface is down do nothing */
5840 if (test_bit(__IXGBE_DOWN, &adapter->state))
5843 ixgbe_watchdog_update_link(adapter);
5845 if (adapter->link_up)
5846 ixgbe_watchdog_link_is_up(adapter);
5848 ixgbe_watchdog_link_is_down(adapter);
5850 ixgbe_spoof_check(adapter);
5851 ixgbe_update_stats(adapter);
5853 ixgbe_watchdog_flush_tx(adapter);
5857 * ixgbe_sfp_detection_subtask - poll for SFP+ cable
5858 * @adapter - the ixgbe adapter structure
5860 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
5862 struct ixgbe_hw *hw = &adapter->hw;
5865 /* not searching for SFP so there is nothing to do here */
5866 if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
5867 !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
5870 /* someone else is in init, wait until next service event */
5871 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5874 err = hw->phy.ops.identify_sfp(hw);
5875 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
5878 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
5879 /* If no cable is present, then we need to reset
5880 * the next time we find a good cable. */
5881 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
5888 /* exit if reset not needed */
5889 if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
5892 adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
5895 * A module may be identified correctly, but the EEPROM may not have
5896 * support for that module. setup_sfp() will fail in that case, so
5897 * we should not allow that module to load.
5899 if (hw->mac.type == ixgbe_mac_82598EB)
5900 err = hw->phy.ops.reset(hw);
5902 err = hw->mac.ops.setup_sfp(hw);
5904 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
5907 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
5908 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
5911 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5913 if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
5914 (adapter->netdev->reg_state == NETREG_REGISTERED)) {
5915 e_dev_err("failed to initialize because an unsupported "
5916 "SFP+ module type was detected.\n");
5917 e_dev_err("Reload the driver after installing a "
5918 "supported module.\n");
5919 unregister_netdev(adapter->netdev);
5924 * ixgbe_sfp_link_config_subtask - set up link SFP after module install
5925 * @adapter - the ixgbe adapter structure
5927 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
5929 struct ixgbe_hw *hw = &adapter->hw;
5933 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
5936 /* someone else is in init, wait until next service event */
5937 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5940 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
5942 autoneg = hw->phy.autoneg_advertised;
5943 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
5944 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
5945 hw->mac.autotry_restart = false;
5946 if (hw->mac.ops.setup_link)
5947 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
5949 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5950 adapter->link_check_timeout = jiffies;
5951 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5955 * ixgbe_service_timer - Timer Call-back
5956 * @data: pointer to adapter cast into an unsigned long
5958 static void ixgbe_service_timer(unsigned long data)
5960 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5961 unsigned long next_event_offset;
5963 /* poll faster when waiting for link */
5964 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
5965 next_event_offset = HZ / 10;
5967 next_event_offset = HZ * 2;
5969 /* Reset the timer */
5970 mod_timer(&adapter->service_timer, next_event_offset + jiffies);
5972 ixgbe_service_event_schedule(adapter);
5975 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
5977 if (!(adapter->flags2 & IXGBE_FLAG2_RESET_REQUESTED))
5980 adapter->flags2 &= ~IXGBE_FLAG2_RESET_REQUESTED;
5982 /* If we're already down or resetting, just bail */
5983 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5984 test_bit(__IXGBE_RESETTING, &adapter->state))
5987 ixgbe_dump(adapter);
5988 netdev_err(adapter->netdev, "Reset adapter\n");
5989 adapter->tx_timeout_count++;
5991 ixgbe_reinit_locked(adapter);
5995 * ixgbe_service_task - manages and runs subtasks
5996 * @work: pointer to work_struct containing our data
5998 static void ixgbe_service_task(struct work_struct *work)
6000 struct ixgbe_adapter *adapter = container_of(work,
6001 struct ixgbe_adapter,
6004 ixgbe_reset_subtask(adapter);
6005 ixgbe_sfp_detection_subtask(adapter);
6006 ixgbe_sfp_link_config_subtask(adapter);
6007 ixgbe_check_overtemp_subtask(adapter);
6008 ixgbe_watchdog_subtask(adapter);
6009 ixgbe_fdir_reinit_subtask(adapter);
6010 ixgbe_check_hang_subtask(adapter);
6012 ixgbe_service_event_complete(adapter);
6015 void ixgbe_tx_ctxtdesc(struct ixgbe_ring *tx_ring, u32 vlan_macip_lens,
6016 u32 fcoe_sof_eof, u32 type_tucmd, u32 mss_l4len_idx)
6018 struct ixgbe_adv_tx_context_desc *context_desc;
6019 u16 i = tx_ring->next_to_use;
6021 context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
6024 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
6026 /* set bits to identify this as an advanced context descriptor */
6027 type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
6029 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
6030 context_desc->seqnum_seed = cpu_to_le32(fcoe_sof_eof);
6031 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd);
6032 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
6035 static int ixgbe_tso(struct ixgbe_ring *tx_ring, struct sk_buff *skb,
6036 u32 tx_flags, __be16 protocol, u8 *hdr_len)
6039 u32 vlan_macip_lens, type_tucmd;
6040 u32 mss_l4len_idx, l4len;
6042 if (!skb_is_gso(skb))
6045 if (skb_header_cloned(skb)) {
6046 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
6051 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
6052 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
6054 if (protocol == __constant_htons(ETH_P_IP)) {
6055 struct iphdr *iph = ip_hdr(skb);
6058 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
6062 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
6063 } else if (skb_is_gso_v6(skb)) {
6064 ipv6_hdr(skb)->payload_len = 0;
6065 tcp_hdr(skb)->check =
6066 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
6067 &ipv6_hdr(skb)->daddr,
6071 l4len = tcp_hdrlen(skb);
6072 *hdr_len = skb_transport_offset(skb) + l4len;
6074 /* mss_l4len_id: use 1 as index for TSO */
6075 mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
6076 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
6077 mss_l4len_idx |= 1 << IXGBE_ADVTXD_IDX_SHIFT;
6079 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
6080 vlan_macip_lens = skb_network_header_len(skb);
6081 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
6082 vlan_macip_lens |= tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
6084 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0, type_tucmd,
6090 static bool ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
6091 struct sk_buff *skb, u32 tx_flags,
6094 u32 vlan_macip_lens = 0;
6095 u32 mss_l4len_idx = 0;
6098 if (skb->ip_summed != CHECKSUM_PARTIAL) {
6099 if (!(tx_flags & IXGBE_TX_FLAGS_HW_VLAN) &&
6100 !(tx_flags & IXGBE_TX_FLAGS_TXSW))
6105 case __constant_htons(ETH_P_IP):
6106 vlan_macip_lens |= skb_network_header_len(skb);
6107 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
6108 l4_hdr = ip_hdr(skb)->protocol;
6110 case __constant_htons(ETH_P_IPV6):
6111 vlan_macip_lens |= skb_network_header_len(skb);
6112 l4_hdr = ipv6_hdr(skb)->nexthdr;
6115 if (unlikely(net_ratelimit())) {
6116 dev_warn(tx_ring->dev,
6117 "partial checksum but proto=%x!\n",
6125 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6126 mss_l4len_idx = tcp_hdrlen(skb) <<
6127 IXGBE_ADVTXD_L4LEN_SHIFT;
6130 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
6131 mss_l4len_idx = sizeof(struct sctphdr) <<
6132 IXGBE_ADVTXD_L4LEN_SHIFT;
6135 mss_l4len_idx = sizeof(struct udphdr) <<
6136 IXGBE_ADVTXD_L4LEN_SHIFT;
6139 if (unlikely(net_ratelimit())) {
6140 dev_warn(tx_ring->dev,
6141 "partial checksum but l4 proto=%x!\n",
6148 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
6149 vlan_macip_lens |= tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
6151 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0,
6152 type_tucmd, mss_l4len_idx);
6154 return (skb->ip_summed == CHECKSUM_PARTIAL);
6157 static __le32 ixgbe_tx_cmd_type(u32 tx_flags)
6159 /* set type for advanced descriptor with frame checksum insertion */
6160 __le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
6161 IXGBE_ADVTXD_DCMD_IFCS |
6162 IXGBE_ADVTXD_DCMD_DEXT);
6164 /* set HW vlan bit if vlan is present */
6165 if (tx_flags & IXGBE_TX_FLAGS_HW_VLAN)
6166 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
6168 /* set segmentation enable bits for TSO/FSO */
6170 if ((tx_flags & IXGBE_TX_FLAGS_TSO) || (tx_flags & IXGBE_TX_FLAGS_FSO))
6172 if (tx_flags & IXGBE_TX_FLAGS_TSO)
6174 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
6179 static __le32 ixgbe_tx_olinfo_status(u32 tx_flags, unsigned int paylen)
6181 __le32 olinfo_status =
6182 cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
6184 if (tx_flags & IXGBE_TX_FLAGS_TSO) {
6185 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM |
6186 (1 << IXGBE_ADVTXD_IDX_SHIFT));
6187 /* enble IPv4 checksum for TSO */
6188 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
6189 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
6192 /* enable L4 checksum for TSO and TX checksum offload */
6193 if (tx_flags & IXGBE_TX_FLAGS_CSUM)
6194 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
6197 /* use index 1 context for FCOE/FSO */
6198 if (tx_flags & IXGBE_TX_FLAGS_FCOE)
6199 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC |
6200 (1 << IXGBE_ADVTXD_IDX_SHIFT));
6204 * Check Context must be set if Tx switch is enabled, which it
6205 * always is for case where virtual functions are running
6207 if (tx_flags & IXGBE_TX_FLAGS_TXSW)
6208 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
6210 return olinfo_status;
6213 #define IXGBE_TXD_CMD (IXGBE_TXD_CMD_EOP | \
6216 static void ixgbe_tx_map(struct ixgbe_ring *tx_ring,
6217 struct sk_buff *skb,
6218 struct ixgbe_tx_buffer *first,
6222 struct device *dev = tx_ring->dev;
6223 struct ixgbe_tx_buffer *tx_buffer_info;
6224 union ixgbe_adv_tx_desc *tx_desc;
6226 __le32 cmd_type, olinfo_status;
6227 struct skb_frag_struct *frag;
6229 unsigned int data_len = skb->data_len;
6230 unsigned int size = skb_headlen(skb);
6232 u32 paylen = skb->len - hdr_len;
6233 u16 i = tx_ring->next_to_use;
6237 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6238 if (data_len >= sizeof(struct fcoe_crc_eof)) {
6239 data_len -= sizeof(struct fcoe_crc_eof);
6241 size -= sizeof(struct fcoe_crc_eof) - data_len;
6247 dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE);
6248 if (dma_mapping_error(dev, dma))
6251 cmd_type = ixgbe_tx_cmd_type(tx_flags);
6252 olinfo_status = ixgbe_tx_olinfo_status(tx_flags, paylen);
6254 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
6257 while (size > IXGBE_MAX_DATA_PER_TXD) {
6258 tx_desc->read.buffer_addr = cpu_to_le64(dma + offset);
6259 tx_desc->read.cmd_type_len =
6260 cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
6261 tx_desc->read.olinfo_status = olinfo_status;
6263 offset += IXGBE_MAX_DATA_PER_TXD;
6264 size -= IXGBE_MAX_DATA_PER_TXD;
6268 if (i == tx_ring->count) {
6269 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, 0);
6274 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6275 tx_buffer_info->length = offset + size;
6276 tx_buffer_info->tx_flags = tx_flags;
6277 tx_buffer_info->dma = dma;
6279 tx_desc->read.buffer_addr = cpu_to_le64(dma + offset);
6280 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
6281 tx_desc->read.olinfo_status = olinfo_status;
6286 frag = &skb_shinfo(skb)->frags[f];
6288 size = min_t(unsigned int, data_len, frag->size);
6296 tx_flags |= IXGBE_TX_FLAGS_MAPPED_AS_PAGE;
6298 dma = dma_map_page(dev, frag->page, frag->page_offset,
6299 size, DMA_TO_DEVICE);
6300 if (dma_mapping_error(dev, dma))
6305 if (i == tx_ring->count) {
6306 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, 0);
6311 tx_desc->read.cmd_type_len |= cpu_to_le32(IXGBE_TXD_CMD);
6314 if (i == tx_ring->count)
6317 tx_ring->next_to_use = i;
6319 if (tx_flags & IXGBE_TX_FLAGS_TSO)
6320 gso_segs = skb_shinfo(skb)->gso_segs;
6322 /* adjust for FCoE Sequence Offload */
6323 else if (tx_flags & IXGBE_TX_FLAGS_FSO)
6324 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
6325 skb_shinfo(skb)->gso_size);
6326 #endif /* IXGBE_FCOE */
6330 /* multiply data chunks by size of headers */
6331 tx_buffer_info->bytecount = paylen + (gso_segs * hdr_len);
6332 tx_buffer_info->gso_segs = gso_segs;
6333 tx_buffer_info->skb = skb;
6335 /* set the timestamp */
6336 first->time_stamp = jiffies;
6339 * Force memory writes to complete before letting h/w
6340 * know there are new descriptors to fetch. (Only
6341 * applicable for weak-ordered memory model archs,
6346 /* set next_to_watch value indicating a packet is present */
6347 first->next_to_watch = tx_desc;
6349 /* notify HW of packet */
6350 writel(i, tx_ring->tail);
6354 dev_err(dev, "TX DMA map failed\n");
6356 /* clear dma mappings for failed tx_buffer_info map */
6358 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6359 ixgbe_unmap_tx_resource(tx_ring, tx_buffer_info);
6360 if (tx_buffer_info == first)
6367 dev_kfree_skb_any(skb);
6369 tx_ring->next_to_use = i;
6372 static void ixgbe_atr(struct ixgbe_ring *ring, struct sk_buff *skb,
6373 u32 tx_flags, __be16 protocol)
6375 struct ixgbe_q_vector *q_vector = ring->q_vector;
6376 union ixgbe_atr_hash_dword input = { .dword = 0 };
6377 union ixgbe_atr_hash_dword common = { .dword = 0 };
6379 unsigned char *network;
6381 struct ipv6hdr *ipv6;
6386 /* if ring doesn't have a interrupt vector, cannot perform ATR */
6390 /* do nothing if sampling is disabled */
6391 if (!ring->atr_sample_rate)
6396 /* snag network header to get L4 type and address */
6397 hdr.network = skb_network_header(skb);
6399 /* Currently only IPv4/IPv6 with TCP is supported */
6400 if ((protocol != __constant_htons(ETH_P_IPV6) ||
6401 hdr.ipv6->nexthdr != IPPROTO_TCP) &&
6402 (protocol != __constant_htons(ETH_P_IP) ||
6403 hdr.ipv4->protocol != IPPROTO_TCP))
6408 /* skip this packet since it is invalid or the socket is closing */
6412 /* sample on all syn packets or once every atr sample count */
6413 if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
6416 /* reset sample count */
6417 ring->atr_count = 0;
6419 vlan_id = htons(tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
6422 * src and dst are inverted, think how the receiver sees them
6424 * The input is broken into two sections, a non-compressed section
6425 * containing vm_pool, vlan_id, and flow_type. The rest of the data
6426 * is XORed together and stored in the compressed dword.
6428 input.formatted.vlan_id = vlan_id;
6431 * since src port and flex bytes occupy the same word XOR them together
6432 * and write the value to source port portion of compressed dword
6434 if (tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
6435 common.port.src ^= th->dest ^ __constant_htons(ETH_P_8021Q);
6437 common.port.src ^= th->dest ^ protocol;
6438 common.port.dst ^= th->source;
6440 if (protocol == __constant_htons(ETH_P_IP)) {
6441 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
6442 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
6444 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
6445 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
6446 hdr.ipv6->saddr.s6_addr32[1] ^
6447 hdr.ipv6->saddr.s6_addr32[2] ^
6448 hdr.ipv6->saddr.s6_addr32[3] ^
6449 hdr.ipv6->daddr.s6_addr32[0] ^
6450 hdr.ipv6->daddr.s6_addr32[1] ^
6451 hdr.ipv6->daddr.s6_addr32[2] ^
6452 hdr.ipv6->daddr.s6_addr32[3];
6455 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
6456 ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
6457 input, common, ring->queue_index);
6460 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6462 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
6463 /* Herbert's original patch had:
6464 * smp_mb__after_netif_stop_queue();
6465 * but since that doesn't exist yet, just open code it. */
6468 /* We need to check again in a case another CPU has just
6469 * made room available. */
6470 if (likely(ixgbe_desc_unused(tx_ring) < size))
6473 /* A reprieve! - use start_queue because it doesn't call schedule */
6474 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
6475 ++tx_ring->tx_stats.restart_queue;
6479 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6481 if (likely(ixgbe_desc_unused(tx_ring) >= size))
6483 return __ixgbe_maybe_stop_tx(tx_ring, size);
6486 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6488 struct ixgbe_adapter *adapter = netdev_priv(dev);
6489 int txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
6492 __be16 protocol = vlan_get_protocol(skb);
6494 if (((protocol == htons(ETH_P_FCOE)) ||
6495 (protocol == htons(ETH_P_FIP))) &&
6496 (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)) {
6497 txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
6498 txq += adapter->ring_feature[RING_F_FCOE].mask;
6503 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6504 while (unlikely(txq >= dev->real_num_tx_queues))
6505 txq -= dev->real_num_tx_queues;
6509 return skb_tx_hash(dev, skb);
6512 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
6513 struct ixgbe_adapter *adapter,
6514 struct ixgbe_ring *tx_ring)
6516 struct ixgbe_tx_buffer *first;
6519 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6522 u16 count = TXD_USE_COUNT(skb_headlen(skb));
6523 __be16 protocol = skb->protocol;
6527 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
6528 * + 1 desc for skb_head_len/IXGBE_MAX_DATA_PER_TXD,
6529 * + 2 desc gap to keep tail from touching head,
6530 * + 1 desc for context descriptor,
6531 * otherwise try next time
6533 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6534 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6535 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6537 count += skb_shinfo(skb)->nr_frags;
6539 if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
6540 tx_ring->tx_stats.tx_busy++;
6541 return NETDEV_TX_BUSY;
6544 #ifdef CONFIG_PCI_IOV
6545 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6546 tx_flags |= IXGBE_TX_FLAGS_TXSW;
6549 /* if we have a HW VLAN tag being added default to the HW one */
6550 if (vlan_tx_tag_present(skb)) {
6551 tx_flags |= vlan_tx_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
6552 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6553 /* else if it is a SW VLAN check the next protocol and store the tag */
6554 } else if (protocol == __constant_htons(ETH_P_8021Q)) {
6555 struct vlan_hdr *vhdr, _vhdr;
6556 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
6560 protocol = vhdr->h_vlan_encapsulated_proto;
6561 tx_flags |= ntohs(vhdr->h_vlan_TCI) << IXGBE_TX_FLAGS_VLAN_SHIFT;
6562 tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
6565 if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
6566 ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
6567 (skb->priority != TC_PRIO_CONTROL))) {
6568 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6569 tx_flags |= tx_ring->dcb_tc <<
6570 IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
6571 if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
6572 struct vlan_ethhdr *vhdr;
6573 if (skb_header_cloned(skb) &&
6574 pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
6576 vhdr = (struct vlan_ethhdr *)skb->data;
6577 vhdr->h_vlan_TCI = htons(tx_flags >>
6578 IXGBE_TX_FLAGS_VLAN_SHIFT);
6580 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6584 /* record the location of the first descriptor for this packet */
6585 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
6588 /* setup tx offload for FCoE */
6589 if ((protocol == __constant_htons(ETH_P_FCOE)) &&
6590 (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)) {
6591 tso = ixgbe_fso(tx_ring, skb, tx_flags, &hdr_len);
6595 tx_flags |= IXGBE_TX_FLAGS_FSO |
6596 IXGBE_TX_FLAGS_FCOE;
6598 tx_flags |= IXGBE_TX_FLAGS_FCOE;
6603 #endif /* IXGBE_FCOE */
6604 /* setup IPv4/IPv6 offloads */
6605 if (protocol == __constant_htons(ETH_P_IP))
6606 tx_flags |= IXGBE_TX_FLAGS_IPV4;
6608 tso = ixgbe_tso(tx_ring, skb, tx_flags, protocol, &hdr_len);
6612 tx_flags |= IXGBE_TX_FLAGS_TSO;
6613 else if (ixgbe_tx_csum(tx_ring, skb, tx_flags, protocol))
6614 tx_flags |= IXGBE_TX_FLAGS_CSUM;
6616 /* add the ATR filter if ATR is on */
6617 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
6618 ixgbe_atr(tx_ring, skb, tx_flags, protocol);
6622 #endif /* IXGBE_FCOE */
6623 ixgbe_tx_map(tx_ring, skb, first, tx_flags, hdr_len);
6625 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
6627 return NETDEV_TX_OK;
6630 dev_kfree_skb_any(skb);
6631 return NETDEV_TX_OK;
6634 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
6636 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6637 struct ixgbe_ring *tx_ring;
6639 tx_ring = adapter->tx_ring[skb->queue_mapping];
6640 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
6644 * ixgbe_set_mac - Change the Ethernet Address of the NIC
6645 * @netdev: network interface device structure
6646 * @p: pointer to an address structure
6648 * Returns 0 on success, negative on failure
6650 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6652 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6653 struct ixgbe_hw *hw = &adapter->hw;
6654 struct sockaddr *addr = p;
6656 if (!is_valid_ether_addr(addr->sa_data))
6657 return -EADDRNOTAVAIL;
6659 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6660 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6662 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
6669 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6671 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6672 struct ixgbe_hw *hw = &adapter->hw;
6676 if (prtad != hw->phy.mdio.prtad)
6678 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6684 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6685 u16 addr, u16 value)
6687 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6688 struct ixgbe_hw *hw = &adapter->hw;
6690 if (prtad != hw->phy.mdio.prtad)
6692 return hw->phy.ops.write_reg(hw, addr, devad, value);
6695 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6697 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6699 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6703 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6705 * @netdev: network interface device structure
6707 * Returns non-zero on failure
6709 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6712 struct ixgbe_adapter *adapter = netdev_priv(dev);
6713 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6715 if (is_valid_ether_addr(mac->san_addr)) {
6717 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6724 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6726 * @netdev: network interface device structure
6728 * Returns non-zero on failure
6730 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6733 struct ixgbe_adapter *adapter = netdev_priv(dev);
6734 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6736 if (is_valid_ether_addr(mac->san_addr)) {
6738 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6744 #ifdef CONFIG_NET_POLL_CONTROLLER
6746 * Polling 'interrupt' - used by things like netconsole to send skbs
6747 * without having to re-enable interrupts. It's not called while
6748 * the interrupt routine is executing.
6750 static void ixgbe_netpoll(struct net_device *netdev)
6752 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6755 /* if interface is down do nothing */
6756 if (test_bit(__IXGBE_DOWN, &adapter->state))
6759 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
6760 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
6761 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
6762 for (i = 0; i < num_q_vectors; i++) {
6763 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
6764 ixgbe_msix_clean_rings(0, q_vector);
6767 ixgbe_intr(adapter->pdev->irq, netdev);
6769 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
6773 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
6774 struct rtnl_link_stats64 *stats)
6776 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6780 for (i = 0; i < adapter->num_rx_queues; i++) {
6781 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
6787 start = u64_stats_fetch_begin_bh(&ring->syncp);
6788 packets = ring->stats.packets;
6789 bytes = ring->stats.bytes;
6790 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6791 stats->rx_packets += packets;
6792 stats->rx_bytes += bytes;
6796 for (i = 0; i < adapter->num_tx_queues; i++) {
6797 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->tx_ring[i]);
6803 start = u64_stats_fetch_begin_bh(&ring->syncp);
6804 packets = ring->stats.packets;
6805 bytes = ring->stats.bytes;
6806 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6807 stats->tx_packets += packets;
6808 stats->tx_bytes += bytes;
6812 /* following stats updated by ixgbe_watchdog_task() */
6813 stats->multicast = netdev->stats.multicast;
6814 stats->rx_errors = netdev->stats.rx_errors;
6815 stats->rx_length_errors = netdev->stats.rx_length_errors;
6816 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
6817 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
6821 /* ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
6822 * #adapter: pointer to ixgbe_adapter
6823 * @tc: number of traffic classes currently enabled
6825 * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
6826 * 802.1Q priority maps to a packet buffer that exists.
6828 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
6830 struct ixgbe_hw *hw = &adapter->hw;
6834 /* 82598 have a static priority to TC mapping that can not
6835 * be changed so no validation is needed.
6837 if (hw->mac.type == ixgbe_mac_82598EB)
6840 reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
6843 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
6844 u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
6846 /* If up2tc is out of bounds default to zero */
6848 reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
6852 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
6858 /* ixgbe_setup_tc - routine to configure net_device for multiple traffic
6861 * @netdev: net device to configure
6862 * @tc: number of traffic classes to enable
6864 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
6866 struct ixgbe_adapter *adapter = netdev_priv(dev);
6867 struct ixgbe_hw *hw = &adapter->hw;
6869 /* Multiple traffic classes requires multiple queues */
6870 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
6871 e_err(drv, "Enable failed, needs MSI-X\n");
6875 /* Hardware supports up to 8 traffic classes */
6876 if (tc > MAX_TRAFFIC_CLASS ||
6877 (hw->mac.type == ixgbe_mac_82598EB && tc < MAX_TRAFFIC_CLASS))
6880 /* Hardware has to reinitialize queues and interrupts to
6881 * match packet buffer alignment. Unfortunantly, the
6882 * hardware is not flexible enough to do this dynamically.
6884 if (netif_running(dev))
6886 ixgbe_clear_interrupt_scheme(adapter);
6889 netdev_set_num_tc(dev, tc);
6890 adapter->last_lfc_mode = adapter->hw.fc.current_mode;
6892 adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
6893 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
6895 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
6896 adapter->hw.fc.requested_mode = ixgbe_fc_none;
6898 netdev_reset_tc(dev);
6900 adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
6902 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
6903 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
6905 adapter->temp_dcb_cfg.pfc_mode_enable = false;
6906 adapter->dcb_cfg.pfc_mode_enable = false;
6909 ixgbe_init_interrupt_scheme(adapter);
6910 ixgbe_validate_rtr(adapter, tc);
6911 if (netif_running(dev))
6917 void ixgbe_do_reset(struct net_device *netdev)
6919 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6921 if (netif_running(netdev))
6922 ixgbe_reinit_locked(adapter);
6924 ixgbe_reset(adapter);
6927 static u32 ixgbe_fix_features(struct net_device *netdev, u32 data)
6929 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6932 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
6933 data &= ~NETIF_F_HW_VLAN_RX;
6936 /* return error if RXHASH is being enabled when RSS is not supported */
6937 if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED))
6938 data &= ~NETIF_F_RXHASH;
6940 /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
6941 if (!(data & NETIF_F_RXCSUM))
6942 data &= ~NETIF_F_LRO;
6944 /* Turn off LRO if not RSC capable or invalid ITR settings */
6945 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)) {
6946 data &= ~NETIF_F_LRO;
6947 } else if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
6948 (adapter->rx_itr_setting != 1 &&
6949 adapter->rx_itr_setting > IXGBE_MAX_RSC_INT_RATE)) {
6950 data &= ~NETIF_F_LRO;
6951 e_info(probe, "rx-usecs set too low, not enabling RSC\n");
6957 static int ixgbe_set_features(struct net_device *netdev, u32 data)
6959 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6960 bool need_reset = false;
6962 /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
6963 if (!(data & NETIF_F_RXCSUM))
6964 adapter->flags &= ~IXGBE_FLAG_RX_CSUM_ENABLED;
6966 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
6968 /* Make sure RSC matches LRO, reset if change */
6969 if (!!(data & NETIF_F_LRO) !=
6970 !!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
6971 adapter->flags2 ^= IXGBE_FLAG2_RSC_ENABLED;
6972 switch (adapter->hw.mac.type) {
6973 case ixgbe_mac_X540:
6974 case ixgbe_mac_82599EB:
6983 * Check if Flow Director n-tuple support was enabled or disabled. If
6984 * the state changed, we need to reset.
6986 if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)) {
6987 /* turn off ATR, enable perfect filters and reset */
6988 if (data & NETIF_F_NTUPLE) {
6989 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
6990 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
6993 } else if (!(data & NETIF_F_NTUPLE)) {
6994 /* turn off Flow Director, set ATR and reset */
6995 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
6996 if ((adapter->flags & IXGBE_FLAG_RSS_ENABLED) &&
6997 !(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
6998 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
7003 ixgbe_do_reset(netdev);
7009 static const struct net_device_ops ixgbe_netdev_ops = {
7010 .ndo_open = ixgbe_open,
7011 .ndo_stop = ixgbe_close,
7012 .ndo_start_xmit = ixgbe_xmit_frame,
7013 .ndo_select_queue = ixgbe_select_queue,
7014 .ndo_set_rx_mode = ixgbe_set_rx_mode,
7015 .ndo_validate_addr = eth_validate_addr,
7016 .ndo_set_mac_address = ixgbe_set_mac,
7017 .ndo_change_mtu = ixgbe_change_mtu,
7018 .ndo_tx_timeout = ixgbe_tx_timeout,
7019 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
7020 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
7021 .ndo_do_ioctl = ixgbe_ioctl,
7022 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
7023 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
7024 .ndo_set_vf_tx_rate = ixgbe_ndo_set_vf_bw,
7025 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
7026 .ndo_get_stats64 = ixgbe_get_stats64,
7027 .ndo_setup_tc = ixgbe_setup_tc,
7028 #ifdef CONFIG_NET_POLL_CONTROLLER
7029 .ndo_poll_controller = ixgbe_netpoll,
7032 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
7033 .ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
7034 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
7035 .ndo_fcoe_enable = ixgbe_fcoe_enable,
7036 .ndo_fcoe_disable = ixgbe_fcoe_disable,
7037 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
7038 #endif /* IXGBE_FCOE */
7039 .ndo_set_features = ixgbe_set_features,
7040 .ndo_fix_features = ixgbe_fix_features,
7043 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
7044 const struct ixgbe_info *ii)
7046 #ifdef CONFIG_PCI_IOV
7047 struct ixgbe_hw *hw = &adapter->hw;
7049 int num_vf_macvlans, i;
7050 struct vf_macvlans *mv_list;
7052 if (hw->mac.type == ixgbe_mac_82598EB || !max_vfs)
7055 /* The 82599 supports up to 64 VFs per physical function
7056 * but this implementation limits allocation to 63 so that
7057 * basic networking resources are still available to the
7060 adapter->num_vfs = (max_vfs > 63) ? 63 : max_vfs;
7061 adapter->flags |= IXGBE_FLAG_SRIOV_ENABLED;
7062 err = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
7064 e_err(probe, "Failed to enable PCI sriov: %d\n", err);
7068 num_vf_macvlans = hw->mac.num_rar_entries -
7069 (IXGBE_MAX_PF_MACVLANS + 1 + adapter->num_vfs);
7071 adapter->mv_list = mv_list = kcalloc(num_vf_macvlans,
7072 sizeof(struct vf_macvlans),
7075 /* Initialize list of VF macvlans */
7076 INIT_LIST_HEAD(&adapter->vf_mvs.l);
7077 for (i = 0; i < num_vf_macvlans; i++) {
7079 mv_list->free = true;
7080 mv_list->rar_entry = hw->mac.num_rar_entries -
7081 (i + adapter->num_vfs + 1);
7082 list_add(&mv_list->l, &adapter->vf_mvs.l);
7087 /* If call to enable VFs succeeded then allocate memory
7088 * for per VF control structures.
7091 kcalloc(adapter->num_vfs,
7092 sizeof(struct vf_data_storage), GFP_KERNEL);
7093 if (adapter->vfinfo) {
7094 /* Now that we're sure SR-IOV is enabled
7095 * and memory allocated set up the mailbox parameters
7097 ixgbe_init_mbx_params_pf(hw);
7098 memcpy(&hw->mbx.ops, ii->mbx_ops,
7099 sizeof(hw->mbx.ops));
7101 /* Disable RSC when in SR-IOV mode */
7102 adapter->flags2 &= ~(IXGBE_FLAG2_RSC_CAPABLE |
7103 IXGBE_FLAG2_RSC_ENABLED);
7108 e_err(probe, "Unable to allocate memory for VF Data Storage - "
7109 "SRIOV disabled\n");
7110 pci_disable_sriov(adapter->pdev);
7113 adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
7114 adapter->num_vfs = 0;
7115 #endif /* CONFIG_PCI_IOV */
7119 * ixgbe_probe - Device Initialization Routine
7120 * @pdev: PCI device information struct
7121 * @ent: entry in ixgbe_pci_tbl
7123 * Returns 0 on success, negative on failure
7125 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
7126 * The OS initialization, configuring of the adapter private structure,
7127 * and a hardware reset occur.
7129 static int __devinit ixgbe_probe(struct pci_dev *pdev,
7130 const struct pci_device_id *ent)
7132 struct net_device *netdev;
7133 struct ixgbe_adapter *adapter = NULL;
7134 struct ixgbe_hw *hw;
7135 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
7136 static int cards_found;
7137 int i, err, pci_using_dac;
7138 u8 part_str[IXGBE_PBANUM_LENGTH];
7139 unsigned int indices = num_possible_cpus();
7145 /* Catch broken hardware that put the wrong VF device ID in
7146 * the PCIe SR-IOV capability.
7148 if (pdev->is_virtfn) {
7149 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
7150 pci_name(pdev), pdev->vendor, pdev->device);
7154 err = pci_enable_device_mem(pdev);
7158 if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
7159 !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
7162 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
7164 err = dma_set_coherent_mask(&pdev->dev,
7168 "No usable DMA configuration, aborting\n");
7175 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
7176 IORESOURCE_MEM), ixgbe_driver_name);
7179 "pci_request_selected_regions failed 0x%x\n", err);
7183 pci_enable_pcie_error_reporting(pdev);
7185 pci_set_master(pdev);
7186 pci_save_state(pdev);
7188 #ifdef CONFIG_IXGBE_DCB
7189 indices *= MAX_TRAFFIC_CLASS;
7192 if (ii->mac == ixgbe_mac_82598EB)
7193 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
7195 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
7198 indices += min_t(unsigned int, num_possible_cpus(),
7199 IXGBE_MAX_FCOE_INDICES);
7201 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
7204 goto err_alloc_etherdev;
7207 SET_NETDEV_DEV(netdev, &pdev->dev);
7209 adapter = netdev_priv(netdev);
7210 pci_set_drvdata(pdev, adapter);
7212 adapter->netdev = netdev;
7213 adapter->pdev = pdev;
7216 adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
7218 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
7219 pci_resource_len(pdev, 0));
7225 for (i = 1; i <= 5; i++) {
7226 if (pci_resource_len(pdev, i) == 0)
7230 netdev->netdev_ops = &ixgbe_netdev_ops;
7231 ixgbe_set_ethtool_ops(netdev);
7232 netdev->watchdog_timeo = 5 * HZ;
7233 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
7235 adapter->bd_number = cards_found;
7238 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
7239 hw->mac.type = ii->mac;
7242 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
7243 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
7244 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
7245 if (!(eec & (1 << 8)))
7246 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
7249 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
7250 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
7251 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
7252 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
7253 hw->phy.mdio.mmds = 0;
7254 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
7255 hw->phy.mdio.dev = netdev;
7256 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
7257 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
7259 ii->get_invariants(hw);
7261 /* setup the private structure */
7262 err = ixgbe_sw_init(adapter);
7266 /* Make it possible the adapter to be woken up via WOL */
7267 switch (adapter->hw.mac.type) {
7268 case ixgbe_mac_82599EB:
7269 case ixgbe_mac_X540:
7270 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7277 * If there is a fan on this device and it has failed log the
7280 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
7281 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
7282 if (esdp & IXGBE_ESDP_SDP1)
7283 e_crit(probe, "Fan has stopped, replace the adapter\n");
7286 /* reset_hw fills in the perm_addr as well */
7287 hw->phy.reset_if_overtemp = true;
7288 err = hw->mac.ops.reset_hw(hw);
7289 hw->phy.reset_if_overtemp = false;
7290 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
7291 hw->mac.type == ixgbe_mac_82598EB) {
7293 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
7294 e_dev_err("failed to load because an unsupported SFP+ "
7295 "module type was detected.\n");
7296 e_dev_err("Reload the driver after installing a supported "
7300 e_dev_err("HW Init failed: %d\n", err);
7304 ixgbe_probe_vf(adapter, ii);
7306 netdev->features = NETIF_F_SG |
7309 NETIF_F_HW_VLAN_TX |
7310 NETIF_F_HW_VLAN_RX |
7311 NETIF_F_HW_VLAN_FILTER |
7317 netdev->hw_features = netdev->features;
7319 switch (adapter->hw.mac.type) {
7320 case ixgbe_mac_82599EB:
7321 case ixgbe_mac_X540:
7322 netdev->features |= NETIF_F_SCTP_CSUM;
7323 netdev->hw_features |= NETIF_F_SCTP_CSUM |
7330 netdev->vlan_features |= NETIF_F_TSO;
7331 netdev->vlan_features |= NETIF_F_TSO6;
7332 netdev->vlan_features |= NETIF_F_IP_CSUM;
7333 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
7334 netdev->vlan_features |= NETIF_F_SG;
7336 netdev->priv_flags |= IFF_UNICAST_FLT;
7338 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7339 adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
7340 IXGBE_FLAG_DCB_ENABLED);
7342 #ifdef CONFIG_IXGBE_DCB
7343 netdev->dcbnl_ops = &dcbnl_ops;
7347 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7348 if (hw->mac.ops.get_device_caps) {
7349 hw->mac.ops.get_device_caps(hw, &device_caps);
7350 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
7351 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
7354 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7355 netdev->vlan_features |= NETIF_F_FCOE_CRC;
7356 netdev->vlan_features |= NETIF_F_FSO;
7357 netdev->vlan_features |= NETIF_F_FCOE_MTU;
7359 #endif /* IXGBE_FCOE */
7360 if (pci_using_dac) {
7361 netdev->features |= NETIF_F_HIGHDMA;
7362 netdev->vlan_features |= NETIF_F_HIGHDMA;
7365 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
7366 netdev->hw_features |= NETIF_F_LRO;
7367 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
7368 netdev->features |= NETIF_F_LRO;
7370 /* make sure the EEPROM is good */
7371 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
7372 e_dev_err("The EEPROM Checksum Is Not Valid\n");
7377 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
7378 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
7380 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
7381 e_dev_err("invalid MAC address\n");
7386 /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
7387 if (hw->mac.ops.disable_tx_laser &&
7388 ((hw->phy.multispeed_fiber) ||
7389 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
7390 (hw->mac.type == ixgbe_mac_82599EB))))
7391 hw->mac.ops.disable_tx_laser(hw);
7393 setup_timer(&adapter->service_timer, &ixgbe_service_timer,
7394 (unsigned long) adapter);
7396 INIT_WORK(&adapter->service_task, ixgbe_service_task);
7397 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
7399 err = ixgbe_init_interrupt_scheme(adapter);
7403 if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED)) {
7404 netdev->hw_features &= ~NETIF_F_RXHASH;
7405 netdev->features &= ~NETIF_F_RXHASH;
7408 switch (pdev->device) {
7409 case IXGBE_DEV_ID_82599_SFP:
7410 /* Only this subdevice supports WOL */
7411 if (pdev->subsystem_device == IXGBE_SUBDEV_ID_82599_SFP)
7412 adapter->wol = IXGBE_WUFC_MAG;
7414 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
7415 /* All except this subdevice support WOL */
7416 if (pdev->subsystem_device != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
7417 adapter->wol = IXGBE_WUFC_MAG;
7419 case IXGBE_DEV_ID_82599_KX4:
7420 adapter->wol = IXGBE_WUFC_MAG;
7426 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
7428 /* pick up the PCI bus settings for reporting later */
7429 hw->mac.ops.get_bus_info(hw);
7431 /* print bus type/speed/width info */
7432 e_dev_info("(PCI Express:%s:%s) %pM\n",
7433 (hw->bus.speed == ixgbe_bus_speed_5000 ? "5.0GT/s" :
7434 hw->bus.speed == ixgbe_bus_speed_2500 ? "2.5GT/s" :
7436 (hw->bus.width == ixgbe_bus_width_pcie_x8 ? "Width x8" :
7437 hw->bus.width == ixgbe_bus_width_pcie_x4 ? "Width x4" :
7438 hw->bus.width == ixgbe_bus_width_pcie_x1 ? "Width x1" :
7442 err = ixgbe_read_pba_string_generic(hw, part_str, IXGBE_PBANUM_LENGTH);
7444 strncpy(part_str, "Unknown", IXGBE_PBANUM_LENGTH);
7445 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
7446 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
7447 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
7450 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
7451 hw->mac.type, hw->phy.type, part_str);
7453 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
7454 e_dev_warn("PCI-Express bandwidth available for this card is "
7455 "not sufficient for optimal performance.\n");
7456 e_dev_warn("For optimal performance a x8 PCI-Express slot "
7460 /* save off EEPROM version number */
7461 hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
7463 /* reset the hardware with the new settings */
7464 err = hw->mac.ops.start_hw(hw);
7466 if (err == IXGBE_ERR_EEPROM_VERSION) {
7467 /* We are running on a pre-production device, log a warning */
7468 e_dev_warn("This device is a pre-production adapter/LOM. "
7469 "Please be aware there may be issues associated "
7470 "with your hardware. If you are experiencing "
7471 "problems please contact your Intel or hardware "
7472 "representative who provided you with this "
7475 strcpy(netdev->name, "eth%d");
7476 err = register_netdev(netdev);
7480 /* carrier off reporting is important to ethtool even BEFORE open */
7481 netif_carrier_off(netdev);
7483 #ifdef CONFIG_IXGBE_DCA
7484 if (dca_add_requester(&pdev->dev) == 0) {
7485 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
7486 ixgbe_setup_dca(adapter);
7489 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7490 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
7491 for (i = 0; i < adapter->num_vfs; i++)
7492 ixgbe_vf_configuration(pdev, (i | 0x10000000));
7495 /* Inform firmware of driver version */
7496 if (hw->mac.ops.set_fw_drv_ver)
7497 hw->mac.ops.set_fw_drv_ver(hw, MAJ, MIN, BUILD,
7500 /* add san mac addr to netdev */
7501 ixgbe_add_sanmac_netdev(netdev);
7503 e_dev_info("Intel(R) 10 Gigabit Network Connection\n");
7508 ixgbe_release_hw_control(adapter);
7509 ixgbe_clear_interrupt_scheme(adapter);
7512 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7513 ixgbe_disable_sriov(adapter);
7514 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
7515 iounmap(hw->hw_addr);
7517 free_netdev(netdev);
7519 pci_release_selected_regions(pdev,
7520 pci_select_bars(pdev, IORESOURCE_MEM));
7523 pci_disable_device(pdev);
7528 * ixgbe_remove - Device Removal Routine
7529 * @pdev: PCI device information struct
7531 * ixgbe_remove is called by the PCI subsystem to alert the driver
7532 * that it should release a PCI device. The could be caused by a
7533 * Hot-Plug event, or because the driver is going to be removed from
7536 static void __devexit ixgbe_remove(struct pci_dev *pdev)
7538 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7539 struct net_device *netdev = adapter->netdev;
7541 set_bit(__IXGBE_DOWN, &adapter->state);
7542 cancel_work_sync(&adapter->service_task);
7544 #ifdef CONFIG_IXGBE_DCA
7545 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
7546 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
7547 dca_remove_requester(&pdev->dev);
7548 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
7553 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
7554 ixgbe_cleanup_fcoe(adapter);
7556 #endif /* IXGBE_FCOE */
7558 /* remove the added san mac */
7559 ixgbe_del_sanmac_netdev(netdev);
7561 if (netdev->reg_state == NETREG_REGISTERED)
7562 unregister_netdev(netdev);
7564 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7565 ixgbe_disable_sriov(adapter);
7567 ixgbe_clear_interrupt_scheme(adapter);
7569 ixgbe_release_hw_control(adapter);
7571 iounmap(adapter->hw.hw_addr);
7572 pci_release_selected_regions(pdev, pci_select_bars(pdev,
7575 e_dev_info("complete\n");
7577 free_netdev(netdev);
7579 pci_disable_pcie_error_reporting(pdev);
7581 pci_disable_device(pdev);
7585 * ixgbe_io_error_detected - called when PCI error is detected
7586 * @pdev: Pointer to PCI device
7587 * @state: The current pci connection state
7589 * This function is called after a PCI bus error affecting
7590 * this device has been detected.
7592 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
7593 pci_channel_state_t state)
7595 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7596 struct net_device *netdev = adapter->netdev;
7598 netif_device_detach(netdev);
7600 if (state == pci_channel_io_perm_failure)
7601 return PCI_ERS_RESULT_DISCONNECT;
7603 if (netif_running(netdev))
7604 ixgbe_down(adapter);
7605 pci_disable_device(pdev);
7607 /* Request a slot reset. */
7608 return PCI_ERS_RESULT_NEED_RESET;
7612 * ixgbe_io_slot_reset - called after the pci bus has been reset.
7613 * @pdev: Pointer to PCI device
7615 * Restart the card from scratch, as if from a cold-boot.
7617 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7619 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7620 pci_ers_result_t result;
7623 if (pci_enable_device_mem(pdev)) {
7624 e_err(probe, "Cannot re-enable PCI device after reset.\n");
7625 result = PCI_ERS_RESULT_DISCONNECT;
7627 pci_set_master(pdev);
7628 pci_restore_state(pdev);
7629 pci_save_state(pdev);
7631 pci_wake_from_d3(pdev, false);
7633 ixgbe_reset(adapter);
7634 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7635 result = PCI_ERS_RESULT_RECOVERED;
7638 err = pci_cleanup_aer_uncorrect_error_status(pdev);
7640 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7641 "failed 0x%0x\n", err);
7642 /* non-fatal, continue */
7649 * ixgbe_io_resume - called when traffic can start flowing again.
7650 * @pdev: Pointer to PCI device
7652 * This callback is called when the error recovery driver tells us that
7653 * its OK to resume normal operation.
7655 static void ixgbe_io_resume(struct pci_dev *pdev)
7657 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7658 struct net_device *netdev = adapter->netdev;
7660 if (netif_running(netdev)) {
7661 if (ixgbe_up(adapter)) {
7662 e_info(probe, "ixgbe_up failed after reset\n");
7667 netif_device_attach(netdev);
7670 static struct pci_error_handlers ixgbe_err_handler = {
7671 .error_detected = ixgbe_io_error_detected,
7672 .slot_reset = ixgbe_io_slot_reset,
7673 .resume = ixgbe_io_resume,
7676 static struct pci_driver ixgbe_driver = {
7677 .name = ixgbe_driver_name,
7678 .id_table = ixgbe_pci_tbl,
7679 .probe = ixgbe_probe,
7680 .remove = __devexit_p(ixgbe_remove),
7682 .suspend = ixgbe_suspend,
7683 .resume = ixgbe_resume,
7685 .shutdown = ixgbe_shutdown,
7686 .err_handler = &ixgbe_err_handler
7690 * ixgbe_init_module - Driver Registration Routine
7692 * ixgbe_init_module is the first routine called when the driver is
7693 * loaded. All it does is register with the PCI subsystem.
7695 static int __init ixgbe_init_module(void)
7698 pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
7699 pr_info("%s\n", ixgbe_copyright);
7701 #ifdef CONFIG_IXGBE_DCA
7702 dca_register_notify(&dca_notifier);
7705 ret = pci_register_driver(&ixgbe_driver);
7709 module_init(ixgbe_init_module);
7712 * ixgbe_exit_module - Driver Exit Cleanup Routine
7714 * ixgbe_exit_module is called just before the driver is removed
7717 static void __exit ixgbe_exit_module(void)
7719 #ifdef CONFIG_IXGBE_DCA
7720 dca_unregister_notify(&dca_notifier);
7722 pci_unregister_driver(&ixgbe_driver);
7723 rcu_barrier(); /* Wait for completion of call_rcu()'s */
7726 #ifdef CONFIG_IXGBE_DCA
7727 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
7732 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
7733 __ixgbe_notify_dca);
7735 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7738 #endif /* CONFIG_IXGBE_DCA */
7740 module_exit(ixgbe_exit_module);