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>
36 #include <linux/tcp.h>
37 #include <linux/pkt_sched.h>
38 #include <linux/ipv6.h>
39 #include <linux/slab.h>
40 #include <net/checksum.h>
41 #include <net/ip6_checksum.h>
42 #include <linux/ethtool.h>
43 #include <linux/if_vlan.h>
44 #include <scsi/fc/fc_fcoe.h>
47 #include "ixgbe_common.h"
48 #include "ixgbe_dcb_82599.h"
49 #include "ixgbe_sriov.h"
51 char ixgbe_driver_name[] = "ixgbe";
52 static const char ixgbe_driver_string[] =
53 "Intel(R) 10 Gigabit PCI Express Network Driver";
58 #define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) "." \
59 __stringify(BUILD) "-k" __stringify(KFIX)
60 const char ixgbe_driver_version[] = DRV_VERSION;
61 static const char ixgbe_copyright[] =
62 "Copyright (c) 1999-2011 Intel Corporation.";
64 static const struct ixgbe_info *ixgbe_info_tbl[] = {
65 [board_82598] = &ixgbe_82598_info,
66 [board_82599] = &ixgbe_82599_info,
67 [board_X540] = &ixgbe_X540_info,
70 /* ixgbe_pci_tbl - PCI Device ID Table
72 * Wildcard entries (PCI_ANY_ID) should come last
73 * Last entry must be all 0s
75 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
76 * Class, Class Mask, private data (not used) }
78 static DEFINE_PCI_DEVICE_TABLE(ixgbe_pci_tbl) = {
79 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598),
81 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT),
83 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT),
85 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT),
87 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2),
89 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4),
91 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT),
93 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT),
95 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM),
97 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR),
99 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM),
101 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX),
103 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4),
105 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM),
107 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR),
109 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP),
111 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM),
113 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ),
115 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4),
117 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE),
119 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE),
121 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM),
123 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE),
125 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T),
127 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF2),
130 /* required last entry */
133 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
135 #ifdef CONFIG_IXGBE_DCA
136 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
138 static struct notifier_block dca_notifier = {
139 .notifier_call = ixgbe_notify_dca,
145 #ifdef CONFIG_PCI_IOV
146 static unsigned int max_vfs;
147 module_param(max_vfs, uint, 0);
148 MODULE_PARM_DESC(max_vfs,
149 "Maximum number of virtual functions to allocate per physical function");
150 #endif /* CONFIG_PCI_IOV */
152 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
153 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
154 MODULE_LICENSE("GPL");
155 MODULE_VERSION(DRV_VERSION);
157 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
159 static inline void ixgbe_disable_sriov(struct ixgbe_adapter *adapter)
161 struct ixgbe_hw *hw = &adapter->hw;
166 #ifdef CONFIG_PCI_IOV
167 /* disable iov and allow time for transactions to clear */
168 pci_disable_sriov(adapter->pdev);
171 /* turn off device IOV mode */
172 gcr = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
173 gcr &= ~(IXGBE_GCR_EXT_SRIOV);
174 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr);
175 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
176 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
177 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
179 /* set default pool back to 0 */
180 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
181 vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
182 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
184 /* take a breather then clean up driver data */
187 kfree(adapter->vfinfo);
188 adapter->vfinfo = NULL;
190 adapter->num_vfs = 0;
191 adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
194 struct ixgbe_reg_info {
199 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
201 /* General Registers */
202 {IXGBE_CTRL, "CTRL"},
203 {IXGBE_STATUS, "STATUS"},
204 {IXGBE_CTRL_EXT, "CTRL_EXT"},
206 /* Interrupt Registers */
207 {IXGBE_EICR, "EICR"},
210 {IXGBE_SRRCTL(0), "SRRCTL"},
211 {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
212 {IXGBE_RDLEN(0), "RDLEN"},
213 {IXGBE_RDH(0), "RDH"},
214 {IXGBE_RDT(0), "RDT"},
215 {IXGBE_RXDCTL(0), "RXDCTL"},
216 {IXGBE_RDBAL(0), "RDBAL"},
217 {IXGBE_RDBAH(0), "RDBAH"},
220 {IXGBE_TDBAL(0), "TDBAL"},
221 {IXGBE_TDBAH(0), "TDBAH"},
222 {IXGBE_TDLEN(0), "TDLEN"},
223 {IXGBE_TDH(0), "TDH"},
224 {IXGBE_TDT(0), "TDT"},
225 {IXGBE_TXDCTL(0), "TXDCTL"},
227 /* List Terminator */
233 * ixgbe_regdump - register printout routine
235 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
241 switch (reginfo->ofs) {
242 case IXGBE_SRRCTL(0):
243 for (i = 0; i < 64; i++)
244 regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
246 case IXGBE_DCA_RXCTRL(0):
247 for (i = 0; i < 64; i++)
248 regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
251 for (i = 0; i < 64; i++)
252 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
255 for (i = 0; i < 64; i++)
256 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
259 for (i = 0; i < 64; i++)
260 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
262 case IXGBE_RXDCTL(0):
263 for (i = 0; i < 64; i++)
264 regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
267 for (i = 0; i < 64; i++)
268 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
271 for (i = 0; i < 64; i++)
272 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
275 for (i = 0; i < 64; i++)
276 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
279 for (i = 0; i < 64; i++)
280 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
283 for (i = 0; i < 64; i++)
284 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
287 for (i = 0; i < 64; i++)
288 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
291 for (i = 0; i < 64; i++)
292 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
294 case IXGBE_TXDCTL(0):
295 for (i = 0; i < 64; i++)
296 regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
299 pr_info("%-15s %08x\n", reginfo->name,
300 IXGBE_READ_REG(hw, reginfo->ofs));
304 for (i = 0; i < 8; i++) {
305 snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i*8, i*8+7);
306 pr_err("%-15s", rname);
307 for (j = 0; j < 8; j++)
308 pr_cont(" %08x", regs[i*8+j]);
315 * ixgbe_dump - Print registers, tx-rings and rx-rings
317 static void ixgbe_dump(struct ixgbe_adapter *adapter)
319 struct net_device *netdev = adapter->netdev;
320 struct ixgbe_hw *hw = &adapter->hw;
321 struct ixgbe_reg_info *reginfo;
323 struct ixgbe_ring *tx_ring;
324 struct ixgbe_tx_buffer *tx_buffer_info;
325 union ixgbe_adv_tx_desc *tx_desc;
326 struct my_u0 { u64 a; u64 b; } *u0;
327 struct ixgbe_ring *rx_ring;
328 union ixgbe_adv_rx_desc *rx_desc;
329 struct ixgbe_rx_buffer *rx_buffer_info;
333 if (!netif_msg_hw(adapter))
336 /* Print netdevice Info */
338 dev_info(&adapter->pdev->dev, "Net device Info\n");
339 pr_info("Device Name state "
340 "trans_start last_rx\n");
341 pr_info("%-15s %016lX %016lX %016lX\n",
348 /* Print Registers */
349 dev_info(&adapter->pdev->dev, "Register Dump\n");
350 pr_info(" Register Name Value\n");
351 for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
352 reginfo->name; reginfo++) {
353 ixgbe_regdump(hw, reginfo);
356 /* Print TX Ring Summary */
357 if (!netdev || !netif_running(netdev))
360 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
361 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
362 for (n = 0; n < adapter->num_tx_queues; n++) {
363 tx_ring = adapter->tx_ring[n];
365 &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
366 pr_info(" %5d %5X %5X %016llX %04X %3X %016llX\n",
367 n, tx_ring->next_to_use, tx_ring->next_to_clean,
368 (u64)tx_buffer_info->dma,
369 tx_buffer_info->length,
370 tx_buffer_info->next_to_watch,
371 (u64)tx_buffer_info->time_stamp);
375 if (!netif_msg_tx_done(adapter))
376 goto rx_ring_summary;
378 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
380 /* Transmit Descriptor Formats
382 * Advanced Transmit Descriptor
383 * +--------------------------------------------------------------+
384 * 0 | Buffer Address [63:0] |
385 * +--------------------------------------------------------------+
386 * 8 | PAYLEN | PORTS | IDX | STA | DCMD |DTYP | RSV | DTALEN |
387 * +--------------------------------------------------------------+
388 * 63 46 45 40 39 36 35 32 31 24 23 20 19 0
391 for (n = 0; n < adapter->num_tx_queues; n++) {
392 tx_ring = adapter->tx_ring[n];
393 pr_info("------------------------------------\n");
394 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
395 pr_info("------------------------------------\n");
396 pr_info("T [desc] [address 63:0 ] "
397 "[PlPOIdStDDt Ln] [bi->dma ] "
398 "leng ntw timestamp bi->skb\n");
400 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
401 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
402 tx_buffer_info = &tx_ring->tx_buffer_info[i];
403 u0 = (struct my_u0 *)tx_desc;
404 pr_info("T [0x%03X] %016llX %016llX %016llX"
405 " %04X %3X %016llX %p", i,
408 (u64)tx_buffer_info->dma,
409 tx_buffer_info->length,
410 tx_buffer_info->next_to_watch,
411 (u64)tx_buffer_info->time_stamp,
412 tx_buffer_info->skb);
413 if (i == tx_ring->next_to_use &&
414 i == tx_ring->next_to_clean)
416 else if (i == tx_ring->next_to_use)
418 else if (i == tx_ring->next_to_clean)
423 if (netif_msg_pktdata(adapter) &&
424 tx_buffer_info->dma != 0)
425 print_hex_dump(KERN_INFO, "",
426 DUMP_PREFIX_ADDRESS, 16, 1,
427 phys_to_virt(tx_buffer_info->dma),
428 tx_buffer_info->length, true);
432 /* Print RX Rings Summary */
434 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
435 pr_info("Queue [NTU] [NTC]\n");
436 for (n = 0; n < adapter->num_rx_queues; n++) {
437 rx_ring = adapter->rx_ring[n];
438 pr_info("%5d %5X %5X\n",
439 n, rx_ring->next_to_use, rx_ring->next_to_clean);
443 if (!netif_msg_rx_status(adapter))
446 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
448 /* Advanced Receive Descriptor (Read) Format
450 * +-----------------------------------------------------+
451 * 0 | Packet Buffer Address [63:1] |A0/NSE|
452 * +----------------------------------------------+------+
453 * 8 | Header Buffer Address [63:1] | DD |
454 * +-----------------------------------------------------+
457 * Advanced Receive Descriptor (Write-Back) Format
459 * 63 48 47 32 31 30 21 20 16 15 4 3 0
460 * +------------------------------------------------------+
461 * 0 | Packet IP |SPH| HDR_LEN | RSV|Packet| RSS |
462 * | Checksum Ident | | | | Type | Type |
463 * +------------------------------------------------------+
464 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
465 * +------------------------------------------------------+
466 * 63 48 47 32 31 20 19 0
468 for (n = 0; n < adapter->num_rx_queues; n++) {
469 rx_ring = adapter->rx_ring[n];
470 pr_info("------------------------------------\n");
471 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
472 pr_info("------------------------------------\n");
473 pr_info("R [desc] [ PktBuf A0] "
474 "[ HeadBuf DD] [bi->dma ] [bi->skb] "
475 "<-- Adv Rx Read format\n");
476 pr_info("RWB[desc] [PcsmIpSHl PtRs] "
477 "[vl er S cks ln] ---------------- [bi->skb] "
478 "<-- Adv Rx Write-Back format\n");
480 for (i = 0; i < rx_ring->count; i++) {
481 rx_buffer_info = &rx_ring->rx_buffer_info[i];
482 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
483 u0 = (struct my_u0 *)rx_desc;
484 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
485 if (staterr & IXGBE_RXD_STAT_DD) {
486 /* Descriptor Done */
487 pr_info("RWB[0x%03X] %016llX "
488 "%016llX ---------------- %p", i,
491 rx_buffer_info->skb);
493 pr_info("R [0x%03X] %016llX "
494 "%016llX %016llX %p", i,
497 (u64)rx_buffer_info->dma,
498 rx_buffer_info->skb);
500 if (netif_msg_pktdata(adapter)) {
501 print_hex_dump(KERN_INFO, "",
502 DUMP_PREFIX_ADDRESS, 16, 1,
503 phys_to_virt(rx_buffer_info->dma),
504 rx_ring->rx_buf_len, true);
506 if (rx_ring->rx_buf_len
507 < IXGBE_RXBUFFER_2048)
508 print_hex_dump(KERN_INFO, "",
509 DUMP_PREFIX_ADDRESS, 16, 1,
511 rx_buffer_info->page_dma +
512 rx_buffer_info->page_offset
518 if (i == rx_ring->next_to_use)
520 else if (i == rx_ring->next_to_clean)
532 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
536 /* Let firmware take over control of h/w */
537 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
538 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
539 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
542 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
546 /* Let firmware know the driver has taken over */
547 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
548 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
549 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
553 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
554 * @adapter: pointer to adapter struct
555 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
556 * @queue: queue to map the corresponding interrupt to
557 * @msix_vector: the vector to map to the corresponding queue
560 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
561 u8 queue, u8 msix_vector)
564 struct ixgbe_hw *hw = &adapter->hw;
565 switch (hw->mac.type) {
566 case ixgbe_mac_82598EB:
567 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
570 index = (((direction * 64) + queue) >> 2) & 0x1F;
571 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
572 ivar &= ~(0xFF << (8 * (queue & 0x3)));
573 ivar |= (msix_vector << (8 * (queue & 0x3)));
574 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
576 case ixgbe_mac_82599EB:
578 if (direction == -1) {
580 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
581 index = ((queue & 1) * 8);
582 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
583 ivar &= ~(0xFF << index);
584 ivar |= (msix_vector << index);
585 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
588 /* tx or rx causes */
589 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
590 index = ((16 * (queue & 1)) + (8 * direction));
591 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
592 ivar &= ~(0xFF << index);
593 ivar |= (msix_vector << index);
594 IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
602 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
607 switch (adapter->hw.mac.type) {
608 case ixgbe_mac_82598EB:
609 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
610 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
612 case ixgbe_mac_82599EB:
614 mask = (qmask & 0xFFFFFFFF);
615 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
616 mask = (qmask >> 32);
617 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
624 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_ring *tx_ring,
625 struct ixgbe_tx_buffer *tx_buffer_info)
627 if (tx_buffer_info->dma) {
628 if (tx_buffer_info->mapped_as_page)
629 dma_unmap_page(tx_ring->dev,
631 tx_buffer_info->length,
634 dma_unmap_single(tx_ring->dev,
636 tx_buffer_info->length,
638 tx_buffer_info->dma = 0;
640 if (tx_buffer_info->skb) {
641 dev_kfree_skb_any(tx_buffer_info->skb);
642 tx_buffer_info->skb = NULL;
644 tx_buffer_info->time_stamp = 0;
645 /* tx_buffer_info must be completely set up in the transmit path */
649 * ixgbe_dcb_txq_to_tc - convert a reg index to a traffic class
650 * @adapter: driver private struct
651 * @index: reg idx of queue to query (0-127)
653 * Helper function to determine the traffic index for a particular
656 * Returns : a tc index for use in range 0-7, or 0-3
658 static u8 ixgbe_dcb_txq_to_tc(struct ixgbe_adapter *adapter, u8 reg_idx)
661 int dcb_i = netdev_get_num_tc(adapter->netdev);
663 /* if DCB is not enabled the queues have no TC */
664 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
667 /* check valid range */
668 if (reg_idx >= adapter->hw.mac.max_tx_queues)
671 switch (adapter->hw.mac.type) {
672 case ixgbe_mac_82598EB:
676 if (dcb_i != 4 && dcb_i != 8)
679 /* if VMDq is enabled the lowest order bits determine TC */
680 if (adapter->flags & (IXGBE_FLAG_SRIOV_ENABLED |
681 IXGBE_FLAG_VMDQ_ENABLED)) {
682 tc = reg_idx & (dcb_i - 1);
687 * Convert the reg_idx into the correct TC. This bitmask
688 * targets the last full 32 ring traffic class and assigns
689 * it a value of 1. From there the rest of the rings are
690 * based on shifting the mask further up to include the
691 * reg_idx / 16 and then reg_idx / 8. It assumes dcB_i
692 * will only ever be 8 or 4 and that reg_idx will never
693 * be greater then 128. The code without the power of 2
694 * optimizations would be:
695 * (((reg_idx % 32) + 32) * dcb_i) >> (9 - reg_idx / 32)
697 tc = ((reg_idx & 0X1F) + 0x20) * dcb_i;
698 tc >>= 9 - (reg_idx >> 5);
704 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
706 struct ixgbe_hw *hw = &adapter->hw;
707 struct ixgbe_hw_stats *hwstats = &adapter->stats;
712 if ((hw->fc.current_mode == ixgbe_fc_full) ||
713 (hw->fc.current_mode == ixgbe_fc_rx_pause)) {
714 switch (hw->mac.type) {
715 case ixgbe_mac_82598EB:
716 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
719 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
721 hwstats->lxoffrxc += data;
723 /* refill credits (no tx hang) if we received xoff */
727 for (i = 0; i < adapter->num_tx_queues; i++)
728 clear_bit(__IXGBE_HANG_CHECK_ARMED,
729 &adapter->tx_ring[i]->state);
731 } else if (!(adapter->dcb_cfg.pfc_mode_enable))
734 /* update stats for each tc, only valid with PFC enabled */
735 for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
736 switch (hw->mac.type) {
737 case ixgbe_mac_82598EB:
738 xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
741 xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
743 hwstats->pxoffrxc[i] += xoff[i];
746 /* disarm tx queues that have received xoff frames */
747 for (i = 0; i < adapter->num_tx_queues; i++) {
748 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
749 u32 tc = ixgbe_dcb_txq_to_tc(adapter, tx_ring->reg_idx);
752 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
756 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
758 return ring->tx_stats.completed;
761 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
763 struct ixgbe_adapter *adapter = netdev_priv(ring->netdev);
764 struct ixgbe_hw *hw = &adapter->hw;
766 u32 head = IXGBE_READ_REG(hw, IXGBE_TDH(ring->reg_idx));
767 u32 tail = IXGBE_READ_REG(hw, IXGBE_TDT(ring->reg_idx));
770 return (head < tail) ?
771 tail - head : (tail + ring->count - head);
776 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
778 u32 tx_done = ixgbe_get_tx_completed(tx_ring);
779 u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
780 u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
783 clear_check_for_tx_hang(tx_ring);
786 * Check for a hung queue, but be thorough. This verifies
787 * that a transmit has been completed since the previous
788 * check AND there is at least one packet pending. The
789 * ARMED bit is set to indicate a potential hang. The
790 * bit is cleared if a pause frame is received to remove
791 * false hang detection due to PFC or 802.3x frames. By
792 * requiring this to fail twice we avoid races with
793 * pfc clearing the ARMED bit and conditions where we
794 * run the check_tx_hang logic with a transmit completion
795 * pending but without time to complete it yet.
797 if ((tx_done_old == tx_done) && tx_pending) {
798 /* make sure it is true for two checks in a row */
799 ret = test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
802 /* update completed stats and continue */
803 tx_ring->tx_stats.tx_done_old = tx_done;
804 /* reset the countdown */
805 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
811 #define IXGBE_MAX_TXD_PWR 14
812 #define IXGBE_MAX_DATA_PER_TXD (1 << IXGBE_MAX_TXD_PWR)
814 /* Tx Descriptors needed, worst case */
815 #define TXD_USE_COUNT(S) (((S) >> IXGBE_MAX_TXD_PWR) + \
816 (((S) & (IXGBE_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
817 #define DESC_NEEDED (TXD_USE_COUNT(IXGBE_MAX_DATA_PER_TXD) /* skb->data */ + \
818 MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1) /* for context */
820 static void ixgbe_tx_timeout(struct net_device *netdev);
823 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
824 * @q_vector: structure containing interrupt and ring information
825 * @tx_ring: tx ring to clean
827 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
828 struct ixgbe_ring *tx_ring)
830 struct ixgbe_adapter *adapter = q_vector->adapter;
831 union ixgbe_adv_tx_desc *tx_desc, *eop_desc;
832 struct ixgbe_tx_buffer *tx_buffer_info;
833 unsigned int total_bytes = 0, total_packets = 0;
834 u16 i, eop, count = 0;
836 i = tx_ring->next_to_clean;
837 eop = tx_ring->tx_buffer_info[i].next_to_watch;
838 eop_desc = IXGBE_TX_DESC_ADV(tx_ring, eop);
840 while ((eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)) &&
841 (count < tx_ring->work_limit)) {
842 bool cleaned = false;
843 rmb(); /* read buffer_info after eop_desc */
844 for ( ; !cleaned; count++) {
845 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
846 tx_buffer_info = &tx_ring->tx_buffer_info[i];
848 tx_desc->wb.status = 0;
849 cleaned = (i == eop);
852 if (i == tx_ring->count)
855 if (cleaned && tx_buffer_info->skb) {
856 total_bytes += tx_buffer_info->bytecount;
857 total_packets += tx_buffer_info->gso_segs;
860 ixgbe_unmap_and_free_tx_resource(tx_ring,
864 tx_ring->tx_stats.completed++;
865 eop = tx_ring->tx_buffer_info[i].next_to_watch;
866 eop_desc = IXGBE_TX_DESC_ADV(tx_ring, eop);
869 tx_ring->next_to_clean = i;
870 tx_ring->total_bytes += total_bytes;
871 tx_ring->total_packets += total_packets;
872 u64_stats_update_begin(&tx_ring->syncp);
873 tx_ring->stats.packets += total_packets;
874 tx_ring->stats.bytes += total_bytes;
875 u64_stats_update_end(&tx_ring->syncp);
877 if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
878 /* schedule immediate reset if we believe we hung */
879 struct ixgbe_hw *hw = &adapter->hw;
880 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, eop);
881 e_err(drv, "Detected Tx Unit Hang\n"
883 " TDH, TDT <%x>, <%x>\n"
884 " next_to_use <%x>\n"
885 " next_to_clean <%x>\n"
886 "tx_buffer_info[next_to_clean]\n"
887 " time_stamp <%lx>\n"
889 tx_ring->queue_index,
890 IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
891 IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
892 tx_ring->next_to_use, eop,
893 tx_ring->tx_buffer_info[eop].time_stamp, jiffies);
895 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
898 "tx hang %d detected on queue %d, resetting adapter\n",
899 adapter->tx_timeout_count + 1, tx_ring->queue_index);
901 /* schedule immediate reset if we believe we hung */
902 ixgbe_tx_timeout(adapter->netdev);
904 /* the adapter is about to reset, no point in enabling stuff */
908 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
909 if (unlikely(count && netif_carrier_ok(tx_ring->netdev) &&
910 (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
911 /* Make sure that anybody stopping the queue after this
912 * sees the new next_to_clean.
915 if (__netif_subqueue_stopped(tx_ring->netdev, tx_ring->queue_index) &&
916 !test_bit(__IXGBE_DOWN, &adapter->state)) {
917 netif_wake_subqueue(tx_ring->netdev, tx_ring->queue_index);
918 ++tx_ring->tx_stats.restart_queue;
922 return count < tx_ring->work_limit;
925 #ifdef CONFIG_IXGBE_DCA
926 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
927 struct ixgbe_ring *rx_ring,
930 struct ixgbe_hw *hw = &adapter->hw;
932 u8 reg_idx = rx_ring->reg_idx;
934 rxctrl = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(reg_idx));
935 switch (hw->mac.type) {
936 case ixgbe_mac_82598EB:
937 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
938 rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
940 case ixgbe_mac_82599EB:
942 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK_82599;
943 rxctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
944 IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599);
949 rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
950 rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
951 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_RRO_EN);
952 IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
955 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
956 struct ixgbe_ring *tx_ring,
959 struct ixgbe_hw *hw = &adapter->hw;
961 u8 reg_idx = tx_ring->reg_idx;
963 switch (hw->mac.type) {
964 case ixgbe_mac_82598EB:
965 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(reg_idx));
966 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
967 txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
968 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
969 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(reg_idx), txctrl);
971 case ixgbe_mac_82599EB:
973 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(reg_idx));
974 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK_82599;
975 txctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
976 IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599);
977 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
978 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(reg_idx), txctrl);
985 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
987 struct ixgbe_adapter *adapter = q_vector->adapter;
992 if (q_vector->cpu == cpu)
995 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
996 for (i = 0; i < q_vector->txr_count; i++) {
997 ixgbe_update_tx_dca(adapter, adapter->tx_ring[r_idx], cpu);
998 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1002 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1003 for (i = 0; i < q_vector->rxr_count; i++) {
1004 ixgbe_update_rx_dca(adapter, adapter->rx_ring[r_idx], cpu);
1005 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1009 q_vector->cpu = cpu;
1014 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
1019 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
1022 /* always use CB2 mode, difference is masked in the CB driver */
1023 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
1025 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
1026 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1030 for (i = 0; i < num_q_vectors; i++) {
1031 adapter->q_vector[i]->cpu = -1;
1032 ixgbe_update_dca(adapter->q_vector[i]);
1036 static int __ixgbe_notify_dca(struct device *dev, void *data)
1038 struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1039 unsigned long event = *(unsigned long *)data;
1041 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
1045 case DCA_PROVIDER_ADD:
1046 /* if we're already enabled, don't do it again */
1047 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1049 if (dca_add_requester(dev) == 0) {
1050 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1051 ixgbe_setup_dca(adapter);
1054 /* Fall Through since DCA is disabled. */
1055 case DCA_PROVIDER_REMOVE:
1056 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1057 dca_remove_requester(dev);
1058 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1059 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
1067 #endif /* CONFIG_IXGBE_DCA */
1069 * ixgbe_receive_skb - Send a completed packet up the stack
1070 * @adapter: board private structure
1071 * @skb: packet to send up
1072 * @status: hardware indication of status of receive
1073 * @rx_ring: rx descriptor ring (for a specific queue) to setup
1074 * @rx_desc: rx descriptor
1076 static void ixgbe_receive_skb(struct ixgbe_q_vector *q_vector,
1077 struct sk_buff *skb, u8 status,
1078 struct ixgbe_ring *ring,
1079 union ixgbe_adv_rx_desc *rx_desc)
1081 struct ixgbe_adapter *adapter = q_vector->adapter;
1082 struct napi_struct *napi = &q_vector->napi;
1083 bool is_vlan = (status & IXGBE_RXD_STAT_VP);
1084 u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
1086 if (is_vlan && (tag & VLAN_VID_MASK))
1087 __vlan_hwaccel_put_tag(skb, tag);
1089 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
1090 napi_gro_receive(napi, skb);
1096 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1097 * @adapter: address of board private structure
1098 * @status_err: hardware indication of status of receive
1099 * @skb: skb currently being received and modified
1101 static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
1102 union ixgbe_adv_rx_desc *rx_desc,
1103 struct sk_buff *skb)
1105 u32 status_err = le32_to_cpu(rx_desc->wb.upper.status_error);
1107 skb_checksum_none_assert(skb);
1109 /* Rx csum disabled */
1110 if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
1113 /* if IP and error */
1114 if ((status_err & IXGBE_RXD_STAT_IPCS) &&
1115 (status_err & IXGBE_RXDADV_ERR_IPE)) {
1116 adapter->hw_csum_rx_error++;
1120 if (!(status_err & IXGBE_RXD_STAT_L4CS))
1123 if (status_err & IXGBE_RXDADV_ERR_TCPE) {
1124 u16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1127 * 82599 errata, UDP frames with a 0 checksum can be marked as
1130 if ((pkt_info & IXGBE_RXDADV_PKTTYPE_UDP) &&
1131 (adapter->hw.mac.type == ixgbe_mac_82599EB))
1134 adapter->hw_csum_rx_error++;
1138 /* It must be a TCP or UDP packet with a valid checksum */
1139 skb->ip_summed = CHECKSUM_UNNECESSARY;
1142 static inline void ixgbe_release_rx_desc(struct ixgbe_ring *rx_ring, u32 val)
1145 * Force memory writes to complete before letting h/w
1146 * know there are new descriptors to fetch. (Only
1147 * applicable for weak-ordered memory model archs,
1151 writel(val, rx_ring->tail);
1155 * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
1156 * @rx_ring: ring to place buffers on
1157 * @cleaned_count: number of buffers to replace
1159 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1161 union ixgbe_adv_rx_desc *rx_desc;
1162 struct ixgbe_rx_buffer *bi;
1163 struct sk_buff *skb;
1164 u16 i = rx_ring->next_to_use;
1166 /* do nothing if no valid netdev defined */
1167 if (!rx_ring->netdev)
1170 while (cleaned_count--) {
1171 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1172 bi = &rx_ring->rx_buffer_info[i];
1176 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1177 rx_ring->rx_buf_len);
1179 rx_ring->rx_stats.alloc_rx_buff_failed++;
1182 /* initialize queue mapping */
1183 skb_record_rx_queue(skb, rx_ring->queue_index);
1188 bi->dma = dma_map_single(rx_ring->dev,
1190 rx_ring->rx_buf_len,
1192 if (dma_mapping_error(rx_ring->dev, bi->dma)) {
1193 rx_ring->rx_stats.alloc_rx_buff_failed++;
1199 if (ring_is_ps_enabled(rx_ring)) {
1201 bi->page = netdev_alloc_page(rx_ring->netdev);
1203 rx_ring->rx_stats.alloc_rx_page_failed++;
1208 if (!bi->page_dma) {
1209 /* use a half page if we're re-using */
1210 bi->page_offset ^= PAGE_SIZE / 2;
1211 bi->page_dma = dma_map_page(rx_ring->dev,
1216 if (dma_mapping_error(rx_ring->dev,
1218 rx_ring->rx_stats.alloc_rx_page_failed++;
1224 /* Refresh the desc even if buffer_addrs didn't change
1225 * because each write-back erases this info. */
1226 rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
1227 rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
1229 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
1230 rx_desc->read.hdr_addr = 0;
1234 if (i == rx_ring->count)
1239 if (rx_ring->next_to_use != i) {
1240 rx_ring->next_to_use = i;
1241 ixgbe_release_rx_desc(rx_ring, i);
1245 static inline u16 ixgbe_get_hlen(union ixgbe_adv_rx_desc *rx_desc)
1247 /* HW will not DMA in data larger than the given buffer, even if it
1248 * parses the (NFS, of course) header to be larger. In that case, it
1249 * fills the header buffer and spills the rest into the page.
1251 u16 hdr_info = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info);
1252 u16 hlen = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
1253 IXGBE_RXDADV_HDRBUFLEN_SHIFT;
1254 if (hlen > IXGBE_RX_HDR_SIZE)
1255 hlen = IXGBE_RX_HDR_SIZE;
1260 * ixgbe_transform_rsc_queue - change rsc queue into a full packet
1261 * @skb: pointer to the last skb in the rsc queue
1263 * This function changes a queue full of hw rsc buffers into a completed
1264 * packet. It uses the ->prev pointers to find the first packet and then
1265 * turns it into the frag list owner.
1267 static inline struct sk_buff *ixgbe_transform_rsc_queue(struct sk_buff *skb)
1269 unsigned int frag_list_size = 0;
1270 unsigned int skb_cnt = 1;
1273 struct sk_buff *prev = skb->prev;
1274 frag_list_size += skb->len;
1280 skb_shinfo(skb)->frag_list = skb->next;
1282 skb->len += frag_list_size;
1283 skb->data_len += frag_list_size;
1284 skb->truesize += frag_list_size;
1285 IXGBE_RSC_CB(skb)->skb_cnt = skb_cnt;
1290 static inline bool ixgbe_get_rsc_state(union ixgbe_adv_rx_desc *rx_desc)
1292 return !!(le32_to_cpu(rx_desc->wb.lower.lo_dword.data) &
1293 IXGBE_RXDADV_RSCCNT_MASK);
1296 static void ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1297 struct ixgbe_ring *rx_ring,
1298 int *work_done, int work_to_do)
1300 struct ixgbe_adapter *adapter = q_vector->adapter;
1301 union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
1302 struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
1303 struct sk_buff *skb;
1304 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1305 const int current_node = numa_node_id();
1308 #endif /* IXGBE_FCOE */
1311 u16 cleaned_count = 0;
1312 bool pkt_is_rsc = false;
1314 i = rx_ring->next_to_clean;
1315 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1316 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1318 while (staterr & IXGBE_RXD_STAT_DD) {
1321 rmb(); /* read descriptor and rx_buffer_info after status DD */
1323 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1325 skb = rx_buffer_info->skb;
1326 rx_buffer_info->skb = NULL;
1327 prefetch(skb->data);
1329 if (ring_is_rsc_enabled(rx_ring))
1330 pkt_is_rsc = ixgbe_get_rsc_state(rx_desc);
1332 /* if this is a skb from previous receive DMA will be 0 */
1333 if (rx_buffer_info->dma) {
1336 !(staterr & IXGBE_RXD_STAT_EOP) &&
1339 * When HWRSC is enabled, delay unmapping
1340 * of the first packet. It carries the
1341 * header information, HW may still
1342 * access the header after the writeback.
1343 * Only unmap it when EOP is reached
1345 IXGBE_RSC_CB(skb)->delay_unmap = true;
1346 IXGBE_RSC_CB(skb)->dma = rx_buffer_info->dma;
1348 dma_unmap_single(rx_ring->dev,
1349 rx_buffer_info->dma,
1350 rx_ring->rx_buf_len,
1353 rx_buffer_info->dma = 0;
1355 if (ring_is_ps_enabled(rx_ring)) {
1356 hlen = ixgbe_get_hlen(rx_desc);
1357 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1359 hlen = le16_to_cpu(rx_desc->wb.upper.length);
1364 /* assume packet split since header is unmapped */
1365 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1369 dma_unmap_page(rx_ring->dev,
1370 rx_buffer_info->page_dma,
1373 rx_buffer_info->page_dma = 0;
1374 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1375 rx_buffer_info->page,
1376 rx_buffer_info->page_offset,
1379 if ((page_count(rx_buffer_info->page) == 1) &&
1380 (page_to_nid(rx_buffer_info->page) == current_node))
1381 get_page(rx_buffer_info->page);
1383 rx_buffer_info->page = NULL;
1385 skb->len += upper_len;
1386 skb->data_len += upper_len;
1387 skb->truesize += upper_len;
1391 if (i == rx_ring->count)
1394 next_rxd = IXGBE_RX_DESC_ADV(rx_ring, i);
1399 u32 nextp = (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
1400 IXGBE_RXDADV_NEXTP_SHIFT;
1401 next_buffer = &rx_ring->rx_buffer_info[nextp];
1403 next_buffer = &rx_ring->rx_buffer_info[i];
1406 if (!(staterr & IXGBE_RXD_STAT_EOP)) {
1407 if (ring_is_ps_enabled(rx_ring)) {
1408 rx_buffer_info->skb = next_buffer->skb;
1409 rx_buffer_info->dma = next_buffer->dma;
1410 next_buffer->skb = skb;
1411 next_buffer->dma = 0;
1413 skb->next = next_buffer->skb;
1414 skb->next->prev = skb;
1416 rx_ring->rx_stats.non_eop_descs++;
1421 skb = ixgbe_transform_rsc_queue(skb);
1422 /* if we got here without RSC the packet is invalid */
1424 __pskb_trim(skb, 0);
1425 rx_buffer_info->skb = skb;
1430 if (ring_is_rsc_enabled(rx_ring)) {
1431 if (IXGBE_RSC_CB(skb)->delay_unmap) {
1432 dma_unmap_single(rx_ring->dev,
1433 IXGBE_RSC_CB(skb)->dma,
1434 rx_ring->rx_buf_len,
1436 IXGBE_RSC_CB(skb)->dma = 0;
1437 IXGBE_RSC_CB(skb)->delay_unmap = false;
1441 if (ring_is_ps_enabled(rx_ring))
1442 rx_ring->rx_stats.rsc_count +=
1443 skb_shinfo(skb)->nr_frags;
1445 rx_ring->rx_stats.rsc_count +=
1446 IXGBE_RSC_CB(skb)->skb_cnt;
1447 rx_ring->rx_stats.rsc_flush++;
1450 /* ERR_MASK will only have valid bits if EOP set */
1451 if (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) {
1452 /* trim packet back to size 0 and recycle it */
1453 __pskb_trim(skb, 0);
1454 rx_buffer_info->skb = skb;
1458 ixgbe_rx_checksum(adapter, rx_desc, skb);
1460 /* probably a little skewed due to removing CRC */
1461 total_rx_bytes += skb->len;
1464 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1466 /* if ddp, not passing to ULD unless for FCP_RSP or error */
1467 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
1468 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
1472 #endif /* IXGBE_FCOE */
1473 ixgbe_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
1476 rx_desc->wb.upper.status_error = 0;
1479 if (*work_done >= work_to_do)
1482 /* return some buffers to hardware, one at a time is too slow */
1483 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
1484 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1488 /* use prefetched values */
1490 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1493 rx_ring->next_to_clean = i;
1494 cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
1497 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1500 /* include DDPed FCoE data */
1501 if (ddp_bytes > 0) {
1504 mss = rx_ring->netdev->mtu - sizeof(struct fcoe_hdr) -
1505 sizeof(struct fc_frame_header) -
1506 sizeof(struct fcoe_crc_eof);
1509 total_rx_bytes += ddp_bytes;
1510 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
1512 #endif /* IXGBE_FCOE */
1514 rx_ring->total_packets += total_rx_packets;
1515 rx_ring->total_bytes += total_rx_bytes;
1516 u64_stats_update_begin(&rx_ring->syncp);
1517 rx_ring->stats.packets += total_rx_packets;
1518 rx_ring->stats.bytes += total_rx_bytes;
1519 u64_stats_update_end(&rx_ring->syncp);
1522 static int ixgbe_clean_rxonly(struct napi_struct *, int);
1524 * ixgbe_configure_msix - Configure MSI-X hardware
1525 * @adapter: board private structure
1527 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1530 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1532 struct ixgbe_q_vector *q_vector;
1533 int i, q_vectors, v_idx, r_idx;
1536 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1539 * Populate the IVAR table and set the ITR values to the
1540 * corresponding register.
1542 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1543 q_vector = adapter->q_vector[v_idx];
1544 /* XXX for_each_set_bit(...) */
1545 r_idx = find_first_bit(q_vector->rxr_idx,
1546 adapter->num_rx_queues);
1548 for (i = 0; i < q_vector->rxr_count; i++) {
1549 u8 reg_idx = adapter->rx_ring[r_idx]->reg_idx;
1550 ixgbe_set_ivar(adapter, 0, reg_idx, v_idx);
1551 r_idx = find_next_bit(q_vector->rxr_idx,
1552 adapter->num_rx_queues,
1555 r_idx = find_first_bit(q_vector->txr_idx,
1556 adapter->num_tx_queues);
1558 for (i = 0; i < q_vector->txr_count; i++) {
1559 u8 reg_idx = adapter->tx_ring[r_idx]->reg_idx;
1560 ixgbe_set_ivar(adapter, 1, reg_idx, v_idx);
1561 r_idx = find_next_bit(q_vector->txr_idx,
1562 adapter->num_tx_queues,
1566 if (q_vector->txr_count && !q_vector->rxr_count)
1568 q_vector->eitr = adapter->tx_eitr_param;
1569 else if (q_vector->rxr_count)
1571 q_vector->eitr = adapter->rx_eitr_param;
1573 ixgbe_write_eitr(q_vector);
1574 /* If Flow Director is enabled, set interrupt affinity */
1575 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
1576 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)) {
1578 * Allocate the affinity_hint cpumask, assign the mask
1579 * for this vector, and set our affinity_hint for
1582 if (!alloc_cpumask_var(&q_vector->affinity_mask,
1585 cpumask_set_cpu(v_idx, q_vector->affinity_mask);
1586 irq_set_affinity_hint(adapter->msix_entries[v_idx].vector,
1587 q_vector->affinity_mask);
1591 switch (adapter->hw.mac.type) {
1592 case ixgbe_mac_82598EB:
1593 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1596 case ixgbe_mac_82599EB:
1597 case ixgbe_mac_X540:
1598 ixgbe_set_ivar(adapter, -1, 1, v_idx);
1604 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1606 /* set up to autoclear timer, and the vectors */
1607 mask = IXGBE_EIMS_ENABLE_MASK;
1608 if (adapter->num_vfs)
1609 mask &= ~(IXGBE_EIMS_OTHER |
1610 IXGBE_EIMS_MAILBOX |
1613 mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
1614 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1617 enum latency_range {
1621 latency_invalid = 255
1625 * ixgbe_update_itr - update the dynamic ITR value based on statistics
1626 * @adapter: pointer to adapter
1627 * @eitr: eitr setting (ints per sec) to give last timeslice
1628 * @itr_setting: current throttle rate in ints/second
1629 * @packets: the number of packets during this measurement interval
1630 * @bytes: the number of bytes during this measurement interval
1632 * Stores a new ITR value based on packets and byte
1633 * counts during the last interrupt. The advantage of per interrupt
1634 * computation is faster updates and more accurate ITR for the current
1635 * traffic pattern. Constants in this function were computed
1636 * based on theoretical maximum wire speed and thresholds were set based
1637 * on testing data as well as attempting to minimize response time
1638 * while increasing bulk throughput.
1639 * this functionality is controlled by the InterruptThrottleRate module
1640 * parameter (see ixgbe_param.c)
1642 static u8 ixgbe_update_itr(struct ixgbe_adapter *adapter,
1643 u32 eitr, u8 itr_setting,
1644 int packets, int bytes)
1646 unsigned int retval = itr_setting;
1651 goto update_itr_done;
1654 /* simple throttlerate management
1655 * 0-20MB/s lowest (100000 ints/s)
1656 * 20-100MB/s low (20000 ints/s)
1657 * 100-1249MB/s bulk (8000 ints/s)
1659 /* what was last interrupt timeslice? */
1660 timepassed_us = 1000000/eitr;
1661 bytes_perint = bytes / timepassed_us; /* bytes/usec */
1663 switch (itr_setting) {
1664 case lowest_latency:
1665 if (bytes_perint > adapter->eitr_low)
1666 retval = low_latency;
1669 if (bytes_perint > adapter->eitr_high)
1670 retval = bulk_latency;
1671 else if (bytes_perint <= adapter->eitr_low)
1672 retval = lowest_latency;
1675 if (bytes_perint <= adapter->eitr_high)
1676 retval = low_latency;
1685 * ixgbe_write_eitr - write EITR register in hardware specific way
1686 * @q_vector: structure containing interrupt and ring information
1688 * This function is made to be called by ethtool and by the driver
1689 * when it needs to update EITR registers at runtime. Hardware
1690 * specific quirks/differences are taken care of here.
1692 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1694 struct ixgbe_adapter *adapter = q_vector->adapter;
1695 struct ixgbe_hw *hw = &adapter->hw;
1696 int v_idx = q_vector->v_idx;
1697 u32 itr_reg = EITR_INTS_PER_SEC_TO_REG(q_vector->eitr);
1699 switch (adapter->hw.mac.type) {
1700 case ixgbe_mac_82598EB:
1701 /* must write high and low 16 bits to reset counter */
1702 itr_reg |= (itr_reg << 16);
1704 case ixgbe_mac_82599EB:
1705 case ixgbe_mac_X540:
1707 * 82599 and X540 can support a value of zero, so allow it for
1708 * max interrupt rate, but there is an errata where it can
1709 * not be zero with RSC
1712 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
1716 * set the WDIS bit to not clear the timer bits and cause an
1717 * immediate assertion of the interrupt
1719 itr_reg |= IXGBE_EITR_CNT_WDIS;
1724 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1727 static void ixgbe_set_itr_msix(struct ixgbe_q_vector *q_vector)
1729 struct ixgbe_adapter *adapter = q_vector->adapter;
1732 u8 current_itr, ret_itr;
1734 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1735 for (i = 0; i < q_vector->txr_count; i++) {
1736 struct ixgbe_ring *tx_ring = adapter->tx_ring[r_idx];
1737 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1739 tx_ring->total_packets,
1740 tx_ring->total_bytes);
1741 /* if the result for this queue would decrease interrupt
1742 * rate for this vector then use that result */
1743 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
1744 q_vector->tx_itr - 1 : ret_itr);
1745 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1749 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1750 for (i = 0; i < q_vector->rxr_count; i++) {
1751 struct ixgbe_ring *rx_ring = adapter->rx_ring[r_idx];
1752 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1754 rx_ring->total_packets,
1755 rx_ring->total_bytes);
1756 /* if the result for this queue would decrease interrupt
1757 * rate for this vector then use that result */
1758 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
1759 q_vector->rx_itr - 1 : ret_itr);
1760 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1764 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1766 switch (current_itr) {
1767 /* counts and packets in update_itr are dependent on these numbers */
1768 case lowest_latency:
1772 new_itr = 20000; /* aka hwitr = ~200 */
1780 if (new_itr != q_vector->eitr) {
1781 /* do an exponential smoothing */
1782 new_itr = ((q_vector->eitr * 9) + new_itr)/10;
1784 /* save the algorithm value here, not the smoothed one */
1785 q_vector->eitr = new_itr;
1787 ixgbe_write_eitr(q_vector);
1792 * ixgbe_check_overtemp_task - worker thread to check over tempurature
1793 * @work: pointer to work_struct containing our data
1795 static void ixgbe_check_overtemp_task(struct work_struct *work)
1797 struct ixgbe_adapter *adapter = container_of(work,
1798 struct ixgbe_adapter,
1799 check_overtemp_task);
1800 struct ixgbe_hw *hw = &adapter->hw;
1801 u32 eicr = adapter->interrupt_event;
1803 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
1806 switch (hw->device_id) {
1807 case IXGBE_DEV_ID_82599_T3_LOM: {
1809 bool link_up = false;
1811 if (hw->mac.ops.check_link)
1812 hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
1814 if (((eicr & IXGBE_EICR_GPI_SDP0) && (!link_up)) ||
1815 (eicr & IXGBE_EICR_LSC))
1816 /* Check if this is due to overtemp */
1817 if (hw->phy.ops.check_overtemp(hw) == IXGBE_ERR_OVERTEMP)
1822 if (!(eicr & IXGBE_EICR_GPI_SDP0))
1827 "Network adapter has been stopped because it has over heated. "
1828 "Restart the computer. If the problem persists, "
1829 "power off the system and replace the adapter\n");
1830 /* write to clear the interrupt */
1831 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP0);
1834 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
1836 struct ixgbe_hw *hw = &adapter->hw;
1838 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
1839 (eicr & IXGBE_EICR_GPI_SDP1)) {
1840 e_crit(probe, "Fan has stopped, replace the adapter\n");
1841 /* write to clear the interrupt */
1842 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1846 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
1848 struct ixgbe_hw *hw = &adapter->hw;
1850 if (eicr & IXGBE_EICR_GPI_SDP2) {
1851 /* Clear the interrupt */
1852 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
1853 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1854 schedule_work(&adapter->sfp_config_module_task);
1857 if (eicr & IXGBE_EICR_GPI_SDP1) {
1858 /* Clear the interrupt */
1859 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1860 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1861 schedule_work(&adapter->multispeed_fiber_task);
1865 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1867 struct ixgbe_hw *hw = &adapter->hw;
1870 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1871 adapter->link_check_timeout = jiffies;
1872 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1873 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1874 IXGBE_WRITE_FLUSH(hw);
1875 schedule_work(&adapter->watchdog_task);
1879 static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
1881 struct net_device *netdev = data;
1882 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1883 struct ixgbe_hw *hw = &adapter->hw;
1887 * Workaround for Silicon errata. Use clear-by-write instead
1888 * of clear-by-read. Reading with EICS will return the
1889 * interrupt causes without clearing, which later be done
1890 * with the write to EICR.
1892 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1893 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1895 if (eicr & IXGBE_EICR_LSC)
1896 ixgbe_check_lsc(adapter);
1898 if (eicr & IXGBE_EICR_MAILBOX)
1899 ixgbe_msg_task(adapter);
1901 switch (hw->mac.type) {
1902 case ixgbe_mac_82599EB:
1903 ixgbe_check_sfp_event(adapter, eicr);
1904 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
1905 ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC))) {
1906 adapter->interrupt_event = eicr;
1907 schedule_work(&adapter->check_overtemp_task);
1909 /* now fallthrough to handle Flow Director */
1910 case ixgbe_mac_X540:
1911 /* Handle Flow Director Full threshold interrupt */
1912 if (eicr & IXGBE_EICR_FLOW_DIR) {
1914 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_FLOW_DIR);
1915 /* Disable transmits before FDIR Re-initialization */
1916 netif_tx_stop_all_queues(netdev);
1917 for (i = 0; i < adapter->num_tx_queues; i++) {
1918 struct ixgbe_ring *tx_ring =
1919 adapter->tx_ring[i];
1920 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
1922 schedule_work(&adapter->fdir_reinit_task);
1930 ixgbe_check_fan_failure(adapter, eicr);
1932 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1933 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_OTHER);
1938 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
1942 struct ixgbe_hw *hw = &adapter->hw;
1944 switch (hw->mac.type) {
1945 case ixgbe_mac_82598EB:
1946 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1947 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
1949 case ixgbe_mac_82599EB:
1950 case ixgbe_mac_X540:
1951 mask = (qmask & 0xFFFFFFFF);
1953 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
1954 mask = (qmask >> 32);
1956 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
1961 /* skip the flush */
1964 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
1968 struct ixgbe_hw *hw = &adapter->hw;
1970 switch (hw->mac.type) {
1971 case ixgbe_mac_82598EB:
1972 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1973 IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
1975 case ixgbe_mac_82599EB:
1976 case ixgbe_mac_X540:
1977 mask = (qmask & 0xFFFFFFFF);
1979 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
1980 mask = (qmask >> 32);
1982 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
1987 /* skip the flush */
1990 static irqreturn_t ixgbe_msix_clean_tx(int irq, void *data)
1992 struct ixgbe_q_vector *q_vector = data;
1993 struct ixgbe_adapter *adapter = q_vector->adapter;
1994 struct ixgbe_ring *tx_ring;
1997 if (!q_vector->txr_count)
2000 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
2001 for (i = 0; i < q_vector->txr_count; i++) {
2002 tx_ring = adapter->tx_ring[r_idx];
2003 tx_ring->total_bytes = 0;
2004 tx_ring->total_packets = 0;
2005 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
2009 /* EIAM disabled interrupts (on this vector) for us */
2010 napi_schedule(&q_vector->napi);
2016 * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
2018 * @data: pointer to our q_vector struct for this interrupt vector
2020 static irqreturn_t ixgbe_msix_clean_rx(int irq, void *data)
2022 struct ixgbe_q_vector *q_vector = data;
2023 struct ixgbe_adapter *adapter = q_vector->adapter;
2024 struct ixgbe_ring *rx_ring;
2028 #ifdef CONFIG_IXGBE_DCA
2029 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2030 ixgbe_update_dca(q_vector);
2033 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2034 for (i = 0; i < q_vector->rxr_count; i++) {
2035 rx_ring = adapter->rx_ring[r_idx];
2036 rx_ring->total_bytes = 0;
2037 rx_ring->total_packets = 0;
2038 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
2042 if (!q_vector->rxr_count)
2045 /* EIAM disabled interrupts (on this vector) for us */
2046 napi_schedule(&q_vector->napi);
2051 static irqreturn_t ixgbe_msix_clean_many(int irq, void *data)
2053 struct ixgbe_q_vector *q_vector = data;
2054 struct ixgbe_adapter *adapter = q_vector->adapter;
2055 struct ixgbe_ring *ring;
2059 if (!q_vector->txr_count && !q_vector->rxr_count)
2062 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
2063 for (i = 0; i < q_vector->txr_count; i++) {
2064 ring = adapter->tx_ring[r_idx];
2065 ring->total_bytes = 0;
2066 ring->total_packets = 0;
2067 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
2071 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2072 for (i = 0; i < q_vector->rxr_count; i++) {
2073 ring = adapter->rx_ring[r_idx];
2074 ring->total_bytes = 0;
2075 ring->total_packets = 0;
2076 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
2080 /* EIAM disabled interrupts (on this vector) for us */
2081 napi_schedule(&q_vector->napi);
2087 * ixgbe_clean_rxonly - msix (aka one shot) rx clean routine
2088 * @napi: napi struct with our devices info in it
2089 * @budget: amount of work driver is allowed to do this pass, in packets
2091 * This function is optimized for cleaning one queue only on a single
2094 static int ixgbe_clean_rxonly(struct napi_struct *napi, int budget)
2096 struct ixgbe_q_vector *q_vector =
2097 container_of(napi, struct ixgbe_q_vector, napi);
2098 struct ixgbe_adapter *adapter = q_vector->adapter;
2099 struct ixgbe_ring *rx_ring = NULL;
2103 #ifdef CONFIG_IXGBE_DCA
2104 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2105 ixgbe_update_dca(q_vector);
2108 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2109 rx_ring = adapter->rx_ring[r_idx];
2111 ixgbe_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
2113 /* If all Rx work done, exit the polling mode */
2114 if (work_done < budget) {
2115 napi_complete(napi);
2116 if (adapter->rx_itr_setting & 1)
2117 ixgbe_set_itr_msix(q_vector);
2118 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2119 ixgbe_irq_enable_queues(adapter,
2120 ((u64)1 << q_vector->v_idx));
2127 * ixgbe_clean_rxtx_many - msix (aka one shot) rx clean routine
2128 * @napi: napi struct with our devices info in it
2129 * @budget: amount of work driver is allowed to do this pass, in packets
2131 * This function will clean more than one rx queue associated with a
2134 static int ixgbe_clean_rxtx_many(struct napi_struct *napi, int budget)
2136 struct ixgbe_q_vector *q_vector =
2137 container_of(napi, struct ixgbe_q_vector, napi);
2138 struct ixgbe_adapter *adapter = q_vector->adapter;
2139 struct ixgbe_ring *ring = NULL;
2140 int work_done = 0, i;
2142 bool tx_clean_complete = true;
2144 #ifdef CONFIG_IXGBE_DCA
2145 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2146 ixgbe_update_dca(q_vector);
2149 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
2150 for (i = 0; i < q_vector->txr_count; i++) {
2151 ring = adapter->tx_ring[r_idx];
2152 tx_clean_complete &= ixgbe_clean_tx_irq(q_vector, ring);
2153 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
2157 /* attempt to distribute budget to each queue fairly, but don't allow
2158 * the budget to go below 1 because we'll exit polling */
2159 budget /= (q_vector->rxr_count ?: 1);
2160 budget = max(budget, 1);
2161 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2162 for (i = 0; i < q_vector->rxr_count; i++) {
2163 ring = adapter->rx_ring[r_idx];
2164 ixgbe_clean_rx_irq(q_vector, ring, &work_done, budget);
2165 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
2169 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2170 ring = adapter->rx_ring[r_idx];
2171 /* If all Rx work done, exit the polling mode */
2172 if (work_done < budget) {
2173 napi_complete(napi);
2174 if (adapter->rx_itr_setting & 1)
2175 ixgbe_set_itr_msix(q_vector);
2176 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2177 ixgbe_irq_enable_queues(adapter,
2178 ((u64)1 << q_vector->v_idx));
2186 * ixgbe_clean_txonly - msix (aka one shot) tx clean routine
2187 * @napi: napi struct with our devices info in it
2188 * @budget: amount of work driver is allowed to do this pass, in packets
2190 * This function is optimized for cleaning one queue only on a single
2193 static int ixgbe_clean_txonly(struct napi_struct *napi, int budget)
2195 struct ixgbe_q_vector *q_vector =
2196 container_of(napi, struct ixgbe_q_vector, napi);
2197 struct ixgbe_adapter *adapter = q_vector->adapter;
2198 struct ixgbe_ring *tx_ring = NULL;
2202 #ifdef CONFIG_IXGBE_DCA
2203 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2204 ixgbe_update_dca(q_vector);
2207 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
2208 tx_ring = adapter->tx_ring[r_idx];
2210 if (!ixgbe_clean_tx_irq(q_vector, tx_ring))
2213 /* If all Tx work done, exit the polling mode */
2214 if (work_done < budget) {
2215 napi_complete(napi);
2216 if (adapter->tx_itr_setting & 1)
2217 ixgbe_set_itr_msix(q_vector);
2218 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2219 ixgbe_irq_enable_queues(adapter,
2220 ((u64)1 << q_vector->v_idx));
2226 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
2229 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2230 struct ixgbe_ring *rx_ring = a->rx_ring[r_idx];
2232 set_bit(r_idx, q_vector->rxr_idx);
2233 q_vector->rxr_count++;
2234 rx_ring->q_vector = q_vector;
2237 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
2240 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2241 struct ixgbe_ring *tx_ring = a->tx_ring[t_idx];
2243 set_bit(t_idx, q_vector->txr_idx);
2244 q_vector->txr_count++;
2245 tx_ring->q_vector = q_vector;
2249 * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
2250 * @adapter: board private structure to initialize
2252 * This function maps descriptor rings to the queue-specific vectors
2253 * we were allotted through the MSI-X enabling code. Ideally, we'd have
2254 * one vector per ring/queue, but on a constrained vector budget, we
2255 * group the rings as "efficiently" as possible. You would add new
2256 * mapping configurations in here.
2258 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter)
2262 int rxr_idx = 0, txr_idx = 0;
2263 int rxr_remaining = adapter->num_rx_queues;
2264 int txr_remaining = adapter->num_tx_queues;
2269 /* No mapping required if MSI-X is disabled. */
2270 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2273 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2276 * The ideal configuration...
2277 * We have enough vectors to map one per queue.
2279 if (q_vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
2280 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
2281 map_vector_to_rxq(adapter, v_start, rxr_idx);
2283 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
2284 map_vector_to_txq(adapter, v_start, txr_idx);
2290 * If we don't have enough vectors for a 1-to-1
2291 * mapping, we'll have to group them so there are
2292 * multiple queues per vector.
2294 /* Re-adjusting *qpv takes care of the remainder. */
2295 for (i = v_start; i < q_vectors; i++) {
2296 rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - i);
2297 for (j = 0; j < rqpv; j++) {
2298 map_vector_to_rxq(adapter, i, rxr_idx);
2302 tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - i);
2303 for (j = 0; j < tqpv; j++) {
2304 map_vector_to_txq(adapter, i, txr_idx);
2314 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2315 * @adapter: board private structure
2317 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2318 * interrupts from the kernel.
2320 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2322 struct net_device *netdev = adapter->netdev;
2323 irqreturn_t (*handler)(int, void *);
2324 int i, vector, q_vectors, err;
2327 /* Decrement for Other and TCP Timer vectors */
2328 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2330 err = ixgbe_map_rings_to_vectors(adapter);
2334 #define SET_HANDLER(_v) (((_v)->rxr_count && (_v)->txr_count) \
2335 ? &ixgbe_msix_clean_many : \
2336 (_v)->rxr_count ? &ixgbe_msix_clean_rx : \
2337 (_v)->txr_count ? &ixgbe_msix_clean_tx : \
2339 for (vector = 0; vector < q_vectors; vector++) {
2340 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2341 handler = SET_HANDLER(q_vector);
2343 if (handler == &ixgbe_msix_clean_rx) {
2344 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2345 "%s-%s-%d", netdev->name, "rx", ri++);
2346 } else if (handler == &ixgbe_msix_clean_tx) {
2347 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2348 "%s-%s-%d", netdev->name, "tx", ti++);
2349 } else if (handler == &ixgbe_msix_clean_many) {
2350 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2351 "%s-%s-%d", netdev->name, "TxRx", ri++);
2354 /* skip this unused q_vector */
2357 err = request_irq(adapter->msix_entries[vector].vector,
2358 handler, 0, q_vector->name,
2361 e_err(probe, "request_irq failed for MSIX interrupt "
2362 "Error: %d\n", err);
2363 goto free_queue_irqs;
2367 sprintf(adapter->lsc_int_name, "%s:lsc", netdev->name);
2368 err = request_irq(adapter->msix_entries[vector].vector,
2369 ixgbe_msix_lsc, 0, adapter->lsc_int_name, netdev);
2371 e_err(probe, "request_irq for msix_lsc failed: %d\n", err);
2372 goto free_queue_irqs;
2378 for (i = vector - 1; i >= 0; i--)
2379 free_irq(adapter->msix_entries[--vector].vector,
2380 adapter->q_vector[i]);
2381 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2382 pci_disable_msix(adapter->pdev);
2383 kfree(adapter->msix_entries);
2384 adapter->msix_entries = NULL;
2388 static void ixgbe_set_itr(struct ixgbe_adapter *adapter)
2390 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2391 struct ixgbe_ring *rx_ring = adapter->rx_ring[0];
2392 struct ixgbe_ring *tx_ring = adapter->tx_ring[0];
2393 u32 new_itr = q_vector->eitr;
2396 q_vector->tx_itr = ixgbe_update_itr(adapter, new_itr,
2398 tx_ring->total_packets,
2399 tx_ring->total_bytes);
2400 q_vector->rx_itr = ixgbe_update_itr(adapter, new_itr,
2402 rx_ring->total_packets,
2403 rx_ring->total_bytes);
2405 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
2407 switch (current_itr) {
2408 /* counts and packets in update_itr are dependent on these numbers */
2409 case lowest_latency:
2413 new_itr = 20000; /* aka hwitr = ~200 */
2422 if (new_itr != q_vector->eitr) {
2423 /* do an exponential smoothing */
2424 new_itr = ((q_vector->eitr * 9) + new_itr)/10;
2426 /* save the algorithm value here */
2427 q_vector->eitr = new_itr;
2429 ixgbe_write_eitr(q_vector);
2434 * ixgbe_irq_enable - Enable default interrupt generation settings
2435 * @adapter: board private structure
2437 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2442 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2443 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2444 mask |= IXGBE_EIMS_GPI_SDP0;
2445 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2446 mask |= IXGBE_EIMS_GPI_SDP1;
2447 switch (adapter->hw.mac.type) {
2448 case ixgbe_mac_82599EB:
2449 case ixgbe_mac_X540:
2450 mask |= IXGBE_EIMS_ECC;
2451 mask |= IXGBE_EIMS_GPI_SDP1;
2452 mask |= IXGBE_EIMS_GPI_SDP2;
2453 if (adapter->num_vfs)
2454 mask |= IXGBE_EIMS_MAILBOX;
2459 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
2460 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
2461 mask |= IXGBE_EIMS_FLOW_DIR;
2463 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2465 ixgbe_irq_enable_queues(adapter, ~0);
2467 IXGBE_WRITE_FLUSH(&adapter->hw);
2469 if (adapter->num_vfs > 32) {
2470 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
2471 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2476 * ixgbe_intr - legacy mode Interrupt Handler
2477 * @irq: interrupt number
2478 * @data: pointer to a network interface device structure
2480 static irqreturn_t ixgbe_intr(int irq, void *data)
2482 struct net_device *netdev = data;
2483 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2484 struct ixgbe_hw *hw = &adapter->hw;
2485 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2489 * Workaround for silicon errata on 82598. Mask the interrupts
2490 * before the read of EICR.
2492 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2494 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2495 * therefore no explict interrupt disable is necessary */
2496 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2499 * shared interrupt alert!
2500 * make sure interrupts are enabled because the read will
2501 * have disabled interrupts due to EIAM
2502 * finish the workaround of silicon errata on 82598. Unmask
2503 * the interrupt that we masked before the EICR read.
2505 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2506 ixgbe_irq_enable(adapter, true, true);
2507 return IRQ_NONE; /* Not our interrupt */
2510 if (eicr & IXGBE_EICR_LSC)
2511 ixgbe_check_lsc(adapter);
2513 switch (hw->mac.type) {
2514 case ixgbe_mac_82599EB:
2515 ixgbe_check_sfp_event(adapter, eicr);
2516 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2517 ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC))) {
2518 adapter->interrupt_event = eicr;
2519 schedule_work(&adapter->check_overtemp_task);
2526 ixgbe_check_fan_failure(adapter, eicr);
2528 if (napi_schedule_prep(&(q_vector->napi))) {
2529 adapter->tx_ring[0]->total_packets = 0;
2530 adapter->tx_ring[0]->total_bytes = 0;
2531 adapter->rx_ring[0]->total_packets = 0;
2532 adapter->rx_ring[0]->total_bytes = 0;
2533 /* would disable interrupts here but EIAM disabled it */
2534 __napi_schedule(&(q_vector->napi));
2538 * re-enable link(maybe) and non-queue interrupts, no flush.
2539 * ixgbe_poll will re-enable the queue interrupts
2542 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2543 ixgbe_irq_enable(adapter, false, false);
2548 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
2550 int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2552 for (i = 0; i < q_vectors; i++) {
2553 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
2554 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
2555 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
2556 q_vector->rxr_count = 0;
2557 q_vector->txr_count = 0;
2562 * ixgbe_request_irq - initialize interrupts
2563 * @adapter: board private structure
2565 * Attempts to configure interrupts using the best available
2566 * capabilities of the hardware and kernel.
2568 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2570 struct net_device *netdev = adapter->netdev;
2573 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2574 err = ixgbe_request_msix_irqs(adapter);
2575 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
2576 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2577 netdev->name, netdev);
2579 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2580 netdev->name, netdev);
2584 e_err(probe, "request_irq failed, Error %d\n", err);
2589 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2591 struct net_device *netdev = adapter->netdev;
2593 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2596 q_vectors = adapter->num_msix_vectors;
2599 free_irq(adapter->msix_entries[i].vector, netdev);
2602 for (; i >= 0; i--) {
2603 /* free only the irqs that were actually requested */
2604 if (!adapter->q_vector[i]->rxr_count &&
2605 !adapter->q_vector[i]->txr_count)
2608 free_irq(adapter->msix_entries[i].vector,
2609 adapter->q_vector[i]);
2612 ixgbe_reset_q_vectors(adapter);
2614 free_irq(adapter->pdev->irq, netdev);
2619 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2620 * @adapter: board private structure
2622 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2624 switch (adapter->hw.mac.type) {
2625 case ixgbe_mac_82598EB:
2626 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2628 case ixgbe_mac_82599EB:
2629 case ixgbe_mac_X540:
2630 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2631 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2632 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2633 if (adapter->num_vfs > 32)
2634 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
2639 IXGBE_WRITE_FLUSH(&adapter->hw);
2640 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2642 for (i = 0; i < adapter->num_msix_vectors; i++)
2643 synchronize_irq(adapter->msix_entries[i].vector);
2645 synchronize_irq(adapter->pdev->irq);
2650 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2653 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2655 struct ixgbe_hw *hw = &adapter->hw;
2657 IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
2658 EITR_INTS_PER_SEC_TO_REG(adapter->rx_eitr_param));
2660 ixgbe_set_ivar(adapter, 0, 0, 0);
2661 ixgbe_set_ivar(adapter, 1, 0, 0);
2663 map_vector_to_rxq(adapter, 0, 0);
2664 map_vector_to_txq(adapter, 0, 0);
2666 e_info(hw, "Legacy interrupt IVAR setup done\n");
2670 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2671 * @adapter: board private structure
2672 * @ring: structure containing ring specific data
2674 * Configure the Tx descriptor ring after a reset.
2676 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2677 struct ixgbe_ring *ring)
2679 struct ixgbe_hw *hw = &adapter->hw;
2680 u64 tdba = ring->dma;
2683 u8 reg_idx = ring->reg_idx;
2685 /* disable queue to avoid issues while updating state */
2686 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2687 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx),
2688 txdctl & ~IXGBE_TXDCTL_ENABLE);
2689 IXGBE_WRITE_FLUSH(hw);
2691 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2692 (tdba & DMA_BIT_MASK(32)));
2693 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2694 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2695 ring->count * sizeof(union ixgbe_adv_tx_desc));
2696 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2697 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2698 ring->tail = hw->hw_addr + IXGBE_TDT(reg_idx);
2700 /* configure fetching thresholds */
2701 if (adapter->rx_itr_setting == 0) {
2702 /* cannot set wthresh when itr==0 */
2703 txdctl &= ~0x007F0000;
2705 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
2706 txdctl |= (8 << 16);
2708 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2709 /* PThresh workaround for Tx hang with DFP enabled. */
2713 /* reinitialize flowdirector state */
2714 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2715 adapter->atr_sample_rate) {
2716 ring->atr_sample_rate = adapter->atr_sample_rate;
2717 ring->atr_count = 0;
2718 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
2720 ring->atr_sample_rate = 0;
2723 clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
2726 txdctl |= IXGBE_TXDCTL_ENABLE;
2727 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
2729 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2730 if (hw->mac.type == ixgbe_mac_82598EB &&
2731 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2734 /* poll to verify queue is enabled */
2736 usleep_range(1000, 2000);
2737 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2738 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
2740 e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
2743 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
2745 struct ixgbe_hw *hw = &adapter->hw;
2749 if (hw->mac.type == ixgbe_mac_82598EB)
2752 /* disable the arbiter while setting MTQC */
2753 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2754 rttdcs |= IXGBE_RTTDCS_ARBDIS;
2755 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2757 /* set transmit pool layout */
2758 mask = (IXGBE_FLAG_SRIOV_ENABLED | IXGBE_FLAG_DCB_ENABLED);
2759 switch (adapter->flags & mask) {
2761 case (IXGBE_FLAG_SRIOV_ENABLED):
2762 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2763 (IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF));
2766 case (IXGBE_FLAG_DCB_ENABLED):
2767 /* We enable 8 traffic classes, DCB only */
2768 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2769 (IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ));
2773 IXGBE_WRITE_REG(hw, IXGBE_MTQC, IXGBE_MTQC_64Q_1PB);
2777 /* re-enable the arbiter */
2778 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2779 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2783 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2784 * @adapter: board private structure
2786 * Configure the Tx unit of the MAC after a reset.
2788 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2790 struct ixgbe_hw *hw = &adapter->hw;
2794 ixgbe_setup_mtqc(adapter);
2796 if (hw->mac.type != ixgbe_mac_82598EB) {
2797 /* DMATXCTL.EN must be before Tx queues are enabled */
2798 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2799 dmatxctl |= IXGBE_DMATXCTL_TE;
2800 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2803 /* Setup the HW Tx Head and Tail descriptor pointers */
2804 for (i = 0; i < adapter->num_tx_queues; i++)
2805 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
2808 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2810 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2811 struct ixgbe_ring *rx_ring)
2814 u8 reg_idx = rx_ring->reg_idx;
2816 switch (adapter->hw.mac.type) {
2817 case ixgbe_mac_82598EB: {
2818 struct ixgbe_ring_feature *feature = adapter->ring_feature;
2819 const int mask = feature[RING_F_RSS].mask;
2820 reg_idx = reg_idx & mask;
2823 case ixgbe_mac_82599EB:
2824 case ixgbe_mac_X540:
2829 srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx));
2831 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
2832 srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
2833 if (adapter->num_vfs)
2834 srrctl |= IXGBE_SRRCTL_DROP_EN;
2836 srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
2837 IXGBE_SRRCTL_BSIZEHDR_MASK;
2839 if (ring_is_ps_enabled(rx_ring)) {
2840 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
2841 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2843 srrctl |= (PAGE_SIZE / 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2845 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2847 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
2848 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2849 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2852 IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx), srrctl);
2855 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2857 struct ixgbe_hw *hw = &adapter->hw;
2858 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2859 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2860 0x6A3E67EA, 0x14364D17, 0x3BED200D};
2861 u32 mrqc = 0, reta = 0;
2866 /* Fill out hash function seeds */
2867 for (i = 0; i < 10; i++)
2868 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2870 /* Fill out redirection table */
2871 for (i = 0, j = 0; i < 128; i++, j++) {
2872 if (j == adapter->ring_feature[RING_F_RSS].indices)
2874 /* reta = 4-byte sliding window of
2875 * 0x00..(indices-1)(indices-1)00..etc. */
2876 reta = (reta << 8) | (j * 0x11);
2878 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2881 /* Disable indicating checksum in descriptor, enables RSS hash */
2882 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2883 rxcsum |= IXGBE_RXCSUM_PCSD;
2884 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2886 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
2887 mask = adapter->flags & IXGBE_FLAG_RSS_ENABLED;
2889 mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2890 #ifdef CONFIG_IXGBE_DCB
2891 | IXGBE_FLAG_DCB_ENABLED
2893 | IXGBE_FLAG_SRIOV_ENABLED
2897 #ifdef CONFIG_IXGBE_DCB
2898 case (IXGBE_FLAG_DCB_ENABLED | IXGBE_FLAG_RSS_ENABLED):
2899 mrqc = IXGBE_MRQC_RTRSS8TCEN;
2901 case (IXGBE_FLAG_DCB_ENABLED):
2902 mrqc = IXGBE_MRQC_RT8TCEN;
2904 #endif /* CONFIG_IXGBE_DCB */
2905 case (IXGBE_FLAG_RSS_ENABLED):
2906 mrqc = IXGBE_MRQC_RSSEN;
2908 case (IXGBE_FLAG_SRIOV_ENABLED):
2909 mrqc = IXGBE_MRQC_VMDQEN;
2915 /* Perform hash on these packet types */
2916 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2917 | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2918 | IXGBE_MRQC_RSS_FIELD_IPV6
2919 | IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
2921 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2925 * ixgbe_clear_rscctl - disable RSC for the indicated ring
2926 * @adapter: address of board private structure
2927 * @ring: structure containing ring specific data
2929 void ixgbe_clear_rscctl(struct ixgbe_adapter *adapter,
2930 struct ixgbe_ring *ring)
2932 struct ixgbe_hw *hw = &adapter->hw;
2934 u8 reg_idx = ring->reg_idx;
2936 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
2937 rscctrl &= ~IXGBE_RSCCTL_RSCEN;
2938 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
2942 * ixgbe_configure_rscctl - enable RSC for the indicated ring
2943 * @adapter: address of board private structure
2944 * @index: index of ring to set
2946 void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
2947 struct ixgbe_ring *ring)
2949 struct ixgbe_hw *hw = &adapter->hw;
2952 u8 reg_idx = ring->reg_idx;
2954 if (!ring_is_rsc_enabled(ring))
2957 rx_buf_len = ring->rx_buf_len;
2958 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
2959 rscctrl |= IXGBE_RSCCTL_RSCEN;
2961 * we must limit the number of descriptors so that the
2962 * total size of max desc * buf_len is not greater
2965 if (ring_is_ps_enabled(ring)) {
2966 #if (MAX_SKB_FRAGS > 16)
2967 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2968 #elif (MAX_SKB_FRAGS > 8)
2969 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2970 #elif (MAX_SKB_FRAGS > 4)
2971 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2973 rscctrl |= IXGBE_RSCCTL_MAXDESC_1;
2976 if (rx_buf_len < IXGBE_RXBUFFER_4096)
2977 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2978 else if (rx_buf_len < IXGBE_RXBUFFER_8192)
2979 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2981 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2983 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
2987 * ixgbe_set_uta - Set unicast filter table address
2988 * @adapter: board private structure
2990 * The unicast table address is a register array of 32-bit registers.
2991 * The table is meant to be used in a way similar to how the MTA is used
2992 * however due to certain limitations in the hardware it is necessary to
2993 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
2994 * enable bit to allow vlan tag stripping when promiscuous mode is enabled
2996 static void ixgbe_set_uta(struct ixgbe_adapter *adapter)
2998 struct ixgbe_hw *hw = &adapter->hw;
3001 /* The UTA table only exists on 82599 hardware and newer */
3002 if (hw->mac.type < ixgbe_mac_82599EB)
3005 /* we only need to do this if VMDq is enabled */
3006 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
3009 for (i = 0; i < 128; i++)
3010 IXGBE_WRITE_REG(hw, IXGBE_UTA(i), ~0);
3013 #define IXGBE_MAX_RX_DESC_POLL 10
3014 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
3015 struct ixgbe_ring *ring)
3017 struct ixgbe_hw *hw = &adapter->hw;
3018 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3020 u8 reg_idx = ring->reg_idx;
3022 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3023 if (hw->mac.type == ixgbe_mac_82598EB &&
3024 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3028 usleep_range(1000, 2000);
3029 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3030 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
3033 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
3034 "the polling period\n", reg_idx);
3038 void ixgbe_disable_rx_queue(struct ixgbe_adapter *adapter,
3039 struct ixgbe_ring *ring)
3041 struct ixgbe_hw *hw = &adapter->hw;
3042 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3044 u8 reg_idx = ring->reg_idx;
3046 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3047 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
3049 /* write value back with RXDCTL.ENABLE bit cleared */
3050 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3052 if (hw->mac.type == ixgbe_mac_82598EB &&
3053 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3056 /* the hardware may take up to 100us to really disable the rx queue */
3059 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3060 } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
3063 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not cleared within "
3064 "the polling period\n", reg_idx);
3068 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
3069 struct ixgbe_ring *ring)
3071 struct ixgbe_hw *hw = &adapter->hw;
3072 u64 rdba = ring->dma;
3074 u8 reg_idx = ring->reg_idx;
3076 /* disable queue to avoid issues while updating state */
3077 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3078 ixgbe_disable_rx_queue(adapter, ring);
3080 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
3081 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
3082 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
3083 ring->count * sizeof(union ixgbe_adv_rx_desc));
3084 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
3085 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
3086 ring->tail = hw->hw_addr + IXGBE_RDT(reg_idx);
3088 ixgbe_configure_srrctl(adapter, ring);
3089 ixgbe_configure_rscctl(adapter, ring);
3091 /* If operating in IOV mode set RLPML for X540 */
3092 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
3093 hw->mac.type == ixgbe_mac_X540) {
3094 rxdctl &= ~IXGBE_RXDCTL_RLPMLMASK;
3095 rxdctl |= ((ring->netdev->mtu + ETH_HLEN +
3096 ETH_FCS_LEN + VLAN_HLEN) | IXGBE_RXDCTL_RLPML_EN);
3099 if (hw->mac.type == ixgbe_mac_82598EB) {
3101 * enable cache line friendly hardware writes:
3102 * PTHRESH=32 descriptors (half the internal cache),
3103 * this also removes ugly rx_no_buffer_count increment
3104 * HTHRESH=4 descriptors (to minimize latency on fetch)
3105 * WTHRESH=8 burst writeback up to two cache lines
3107 rxdctl &= ~0x3FFFFF;
3111 /* enable receive descriptor ring */
3112 rxdctl |= IXGBE_RXDCTL_ENABLE;
3113 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3115 ixgbe_rx_desc_queue_enable(adapter, ring);
3116 ixgbe_alloc_rx_buffers(ring, IXGBE_DESC_UNUSED(ring));
3119 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
3121 struct ixgbe_hw *hw = &adapter->hw;
3124 /* PSRTYPE must be initialized in non 82598 adapters */
3125 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
3126 IXGBE_PSRTYPE_UDPHDR |
3127 IXGBE_PSRTYPE_IPV4HDR |
3128 IXGBE_PSRTYPE_L2HDR |
3129 IXGBE_PSRTYPE_IPV6HDR;
3131 if (hw->mac.type == ixgbe_mac_82598EB)
3134 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED)
3135 psrtype |= (adapter->num_rx_queues_per_pool << 29);
3137 for (p = 0; p < adapter->num_rx_pools; p++)
3138 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(adapter->num_vfs + p),
3142 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
3144 struct ixgbe_hw *hw = &adapter->hw;
3147 u32 reg_offset, vf_shift;
3150 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
3153 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
3154 vt_reg_bits = IXGBE_VMD_CTL_VMDQ_EN | IXGBE_VT_CTL_REPLEN;
3155 vt_reg_bits |= (adapter->num_vfs << IXGBE_VT_CTL_POOL_SHIFT);
3156 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl | vt_reg_bits);
3158 vf_shift = adapter->num_vfs % 32;
3159 reg_offset = (adapter->num_vfs > 32) ? 1 : 0;
3161 /* Enable only the PF's pool for Tx/Rx */
3162 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (1 << vf_shift));
3163 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), 0);
3164 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (1 << vf_shift));
3165 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), 0);
3166 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3168 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
3169 hw->mac.ops.set_vmdq(hw, 0, adapter->num_vfs);
3172 * Set up VF register offsets for selected VT Mode,
3173 * i.e. 32 or 64 VFs for SR-IOV
3175 gcr_ext = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
3176 gcr_ext |= IXGBE_GCR_EXT_MSIX_EN;
3177 gcr_ext |= IXGBE_GCR_EXT_VT_MODE_64;
3178 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
3180 /* enable Tx loopback for VF/PF communication */
3181 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3182 /* Enable MAC Anti-Spoofing */
3183 hw->mac.ops.set_mac_anti_spoofing(hw, (adapter->num_vfs != 0),
3187 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
3189 struct ixgbe_hw *hw = &adapter->hw;
3190 struct net_device *netdev = adapter->netdev;
3191 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3193 struct ixgbe_ring *rx_ring;
3197 /* Decide whether to use packet split mode or not */
3199 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
3201 /* Do not use packet split if we're in SR-IOV Mode */
3202 if (adapter->num_vfs)
3203 adapter->flags &= ~IXGBE_FLAG_RX_PS_ENABLED;
3205 /* Disable packet split due to 82599 erratum #45 */
3206 if (hw->mac.type == ixgbe_mac_82599EB)
3207 adapter->flags &= ~IXGBE_FLAG_RX_PS_ENABLED;
3209 /* Set the RX buffer length according to the mode */
3210 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
3211 rx_buf_len = IXGBE_RX_HDR_SIZE;
3213 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
3214 (netdev->mtu <= ETH_DATA_LEN))
3215 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3217 rx_buf_len = ALIGN(max_frame + VLAN_HLEN, 1024);
3221 /* adjust max frame to be able to do baby jumbo for FCoE */
3222 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
3223 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
3224 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3226 #endif /* IXGBE_FCOE */
3227 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
3228 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
3229 mhadd &= ~IXGBE_MHADD_MFS_MASK;
3230 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
3232 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
3235 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
3236 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
3237 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
3238 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
3241 * Setup the HW Rx Head and Tail Descriptor Pointers and
3242 * the Base and Length of the Rx Descriptor Ring
3244 for (i = 0; i < adapter->num_rx_queues; i++) {
3245 rx_ring = adapter->rx_ring[i];
3246 rx_ring->rx_buf_len = rx_buf_len;
3248 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)
3249 set_ring_ps_enabled(rx_ring);
3251 clear_ring_ps_enabled(rx_ring);
3253 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
3254 set_ring_rsc_enabled(rx_ring);
3256 clear_ring_rsc_enabled(rx_ring);
3259 if (netdev->features & NETIF_F_FCOE_MTU) {
3260 struct ixgbe_ring_feature *f;
3261 f = &adapter->ring_feature[RING_F_FCOE];
3262 if ((i >= f->mask) && (i < f->mask + f->indices)) {
3263 clear_ring_ps_enabled(rx_ring);
3264 if (rx_buf_len < IXGBE_FCOE_JUMBO_FRAME_SIZE)
3265 rx_ring->rx_buf_len =
3266 IXGBE_FCOE_JUMBO_FRAME_SIZE;
3267 } else if (!ring_is_rsc_enabled(rx_ring) &&
3268 !ring_is_ps_enabled(rx_ring)) {
3269 rx_ring->rx_buf_len =
3270 IXGBE_FCOE_JUMBO_FRAME_SIZE;
3273 #endif /* IXGBE_FCOE */
3277 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
3279 struct ixgbe_hw *hw = &adapter->hw;
3280 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
3282 switch (hw->mac.type) {
3283 case ixgbe_mac_82598EB:
3285 * For VMDq support of different descriptor types or
3286 * buffer sizes through the use of multiple SRRCTL
3287 * registers, RDRXCTL.MVMEN must be set to 1
3289 * also, the manual doesn't mention it clearly but DCA hints
3290 * will only use queue 0's tags unless this bit is set. Side
3291 * effects of setting this bit are only that SRRCTL must be
3292 * fully programmed [0..15]
3294 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
3296 case ixgbe_mac_82599EB:
3297 case ixgbe_mac_X540:
3298 /* Disable RSC for ACK packets */
3299 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
3300 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
3301 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3302 /* hardware requires some bits to be set by default */
3303 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3304 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3307 /* We should do nothing since we don't know this hardware */
3311 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3315 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3316 * @adapter: board private structure
3318 * Configure the Rx unit of the MAC after a reset.
3320 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3322 struct ixgbe_hw *hw = &adapter->hw;
3326 /* disable receives while setting up the descriptors */
3327 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3328 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3330 ixgbe_setup_psrtype(adapter);
3331 ixgbe_setup_rdrxctl(adapter);
3333 /* Program registers for the distribution of queues */
3334 ixgbe_setup_mrqc(adapter);
3336 ixgbe_set_uta(adapter);
3338 /* set_rx_buffer_len must be called before ring initialization */
3339 ixgbe_set_rx_buffer_len(adapter);
3342 * Setup the HW Rx Head and Tail Descriptor Pointers and
3343 * the Base and Length of the Rx Descriptor Ring
3345 for (i = 0; i < adapter->num_rx_queues; i++)
3346 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3348 /* disable drop enable for 82598 parts */
3349 if (hw->mac.type == ixgbe_mac_82598EB)
3350 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3352 /* enable all receives */
3353 rxctrl |= IXGBE_RXCTRL_RXEN;
3354 hw->mac.ops.enable_rx_dma(hw, rxctrl);
3357 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3359 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3360 struct ixgbe_hw *hw = &adapter->hw;
3361 int pool_ndx = adapter->num_vfs;
3363 /* add VID to filter table */
3364 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, true);
3365 set_bit(vid, adapter->active_vlans);
3368 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3370 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3371 struct ixgbe_hw *hw = &adapter->hw;
3372 int pool_ndx = adapter->num_vfs;
3374 /* remove VID from filter table */
3375 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, false);
3376 clear_bit(vid, adapter->active_vlans);
3380 * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3381 * @adapter: driver data
3383 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
3385 struct ixgbe_hw *hw = &adapter->hw;
3388 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3389 vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
3390 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3394 * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3395 * @adapter: driver data
3397 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
3399 struct ixgbe_hw *hw = &adapter->hw;
3402 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3403 vlnctrl |= IXGBE_VLNCTRL_VFE;
3404 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3405 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3409 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3410 * @adapter: driver data
3412 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3414 struct ixgbe_hw *hw = &adapter->hw;
3418 switch (hw->mac.type) {
3419 case ixgbe_mac_82598EB:
3420 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3421 vlnctrl &= ~IXGBE_VLNCTRL_VME;
3422 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3424 case ixgbe_mac_82599EB:
3425 case ixgbe_mac_X540:
3426 for (i = 0; i < adapter->num_rx_queues; i++) {
3427 j = adapter->rx_ring[i]->reg_idx;
3428 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3429 vlnctrl &= ~IXGBE_RXDCTL_VME;
3430 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3439 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3440 * @adapter: driver data
3442 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3444 struct ixgbe_hw *hw = &adapter->hw;
3448 switch (hw->mac.type) {
3449 case ixgbe_mac_82598EB:
3450 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3451 vlnctrl |= IXGBE_VLNCTRL_VME;
3452 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3454 case ixgbe_mac_82599EB:
3455 case ixgbe_mac_X540:
3456 for (i = 0; i < adapter->num_rx_queues; i++) {
3457 j = adapter->rx_ring[i]->reg_idx;
3458 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3459 vlnctrl |= IXGBE_RXDCTL_VME;
3460 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3468 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3472 ixgbe_vlan_rx_add_vid(adapter->netdev, 0);
3474 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3475 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
3479 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3480 * @netdev: network interface device structure
3482 * Writes unicast address list to the RAR table.
3483 * Returns: -ENOMEM on failure/insufficient address space
3484 * 0 on no addresses written
3485 * X on writing X addresses to the RAR table
3487 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3489 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3490 struct ixgbe_hw *hw = &adapter->hw;
3491 unsigned int vfn = adapter->num_vfs;
3492 unsigned int rar_entries = hw->mac.num_rar_entries - (vfn + 1);
3495 /* return ENOMEM indicating insufficient memory for addresses */
3496 if (netdev_uc_count(netdev) > rar_entries)
3499 if (!netdev_uc_empty(netdev) && rar_entries) {
3500 struct netdev_hw_addr *ha;
3501 /* return error if we do not support writing to RAR table */
3502 if (!hw->mac.ops.set_rar)
3505 netdev_for_each_uc_addr(ha, netdev) {
3508 hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3513 /* write the addresses in reverse order to avoid write combining */
3514 for (; rar_entries > 0 ; rar_entries--)
3515 hw->mac.ops.clear_rar(hw, rar_entries);
3521 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3522 * @netdev: network interface device structure
3524 * The set_rx_method entry point is called whenever the unicast/multicast
3525 * address list or the network interface flags are updated. This routine is
3526 * responsible for configuring the hardware for proper unicast, multicast and
3529 void ixgbe_set_rx_mode(struct net_device *netdev)
3531 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3532 struct ixgbe_hw *hw = &adapter->hw;
3533 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3536 /* Check for Promiscuous and All Multicast modes */
3538 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3540 /* set all bits that we expect to always be set */
3541 fctrl |= IXGBE_FCTRL_BAM;
3542 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3543 fctrl |= IXGBE_FCTRL_PMCF;
3545 /* clear the bits we are changing the status of */
3546 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3548 if (netdev->flags & IFF_PROMISC) {
3549 hw->addr_ctrl.user_set_promisc = true;
3550 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3551 vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3552 /* don't hardware filter vlans in promisc mode */
3553 ixgbe_vlan_filter_disable(adapter);
3555 if (netdev->flags & IFF_ALLMULTI) {
3556 fctrl |= IXGBE_FCTRL_MPE;
3557 vmolr |= IXGBE_VMOLR_MPE;
3560 * Write addresses to the MTA, if the attempt fails
3561 * then we should just turn on promiscuous mode so
3562 * that we can at least receive multicast traffic
3564 hw->mac.ops.update_mc_addr_list(hw, netdev);
3565 vmolr |= IXGBE_VMOLR_ROMPE;
3567 ixgbe_vlan_filter_enable(adapter);
3568 hw->addr_ctrl.user_set_promisc = false;
3570 * Write addresses to available RAR registers, if there is not
3571 * sufficient space to store all the addresses then enable
3572 * unicast promiscuous mode
3574 count = ixgbe_write_uc_addr_list(netdev);
3576 fctrl |= IXGBE_FCTRL_UPE;
3577 vmolr |= IXGBE_VMOLR_ROPE;
3581 if (adapter->num_vfs) {
3582 ixgbe_restore_vf_multicasts(adapter);
3583 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(adapter->num_vfs)) &
3584 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3586 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(adapter->num_vfs), vmolr);
3589 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3591 if (netdev->features & NETIF_F_HW_VLAN_RX)
3592 ixgbe_vlan_strip_enable(adapter);
3594 ixgbe_vlan_strip_disable(adapter);
3597 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3600 struct ixgbe_q_vector *q_vector;
3601 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3603 /* legacy and MSI only use one vector */
3604 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3607 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3608 struct napi_struct *napi;
3609 q_vector = adapter->q_vector[q_idx];
3610 napi = &q_vector->napi;
3611 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3612 if (!q_vector->rxr_count || !q_vector->txr_count) {
3613 if (q_vector->txr_count == 1)
3614 napi->poll = &ixgbe_clean_txonly;
3615 else if (q_vector->rxr_count == 1)
3616 napi->poll = &ixgbe_clean_rxonly;
3624 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3627 struct ixgbe_q_vector *q_vector;
3628 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3630 /* legacy and MSI only use one vector */
3631 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3634 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3635 q_vector = adapter->q_vector[q_idx];
3636 napi_disable(&q_vector->napi);
3640 #ifdef CONFIG_IXGBE_DCB
3642 * ixgbe_configure_dcb - Configure DCB hardware
3643 * @adapter: ixgbe adapter struct
3645 * This is called by the driver on open to configure the DCB hardware.
3646 * This is also called by the gennetlink interface when reconfiguring
3649 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3651 struct ixgbe_hw *hw = &adapter->hw;
3652 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3654 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3655 if (hw->mac.type == ixgbe_mac_82598EB)
3656 netif_set_gso_max_size(adapter->netdev, 65536);
3660 if (hw->mac.type == ixgbe_mac_82598EB)
3661 netif_set_gso_max_size(adapter->netdev, 32768);
3664 /* Enable VLAN tag insert/strip */
3665 adapter->netdev->features |= NETIF_F_HW_VLAN_RX;
3667 hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
3669 /* reconfigure the hardware */
3670 if (adapter->dcbx_cap & (DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE)) {
3672 if (adapter->netdev->features & NETIF_F_FCOE_MTU)
3673 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
3675 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3677 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3679 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
3681 struct net_device *dev = adapter->netdev;
3683 if (adapter->ixgbe_ieee_ets)
3684 dev->dcbnl_ops->ieee_setets(dev,
3685 adapter->ixgbe_ieee_ets);
3686 if (adapter->ixgbe_ieee_pfc)
3687 dev->dcbnl_ops->ieee_setpfc(dev,
3688 adapter->ixgbe_ieee_pfc);
3691 /* Enable RSS Hash per TC */
3692 if (hw->mac.type != ixgbe_mac_82598EB) {
3696 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
3698 u8 cnt = adapter->netdev->tc_to_txq[i].count;
3703 reg |= msb << IXGBE_RQTC_SHIFT_TC(i);
3705 IXGBE_WRITE_REG(hw, IXGBE_RQTC, reg);
3710 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3712 struct net_device *netdev = adapter->netdev;
3713 struct ixgbe_hw *hw = &adapter->hw;
3716 #ifdef CONFIG_IXGBE_DCB
3717 ixgbe_configure_dcb(adapter);
3720 ixgbe_set_rx_mode(netdev);
3721 ixgbe_restore_vlan(adapter);
3724 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
3725 ixgbe_configure_fcoe(adapter);
3727 #endif /* IXGBE_FCOE */
3728 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3729 for (i = 0; i < adapter->num_tx_queues; i++)
3730 adapter->tx_ring[i]->atr_sample_rate =
3731 adapter->atr_sample_rate;
3732 ixgbe_init_fdir_signature_82599(hw, adapter->fdir_pballoc);
3733 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3734 ixgbe_init_fdir_perfect_82599(hw, adapter->fdir_pballoc);
3736 ixgbe_configure_virtualization(adapter);
3738 ixgbe_configure_tx(adapter);
3739 ixgbe_configure_rx(adapter);
3742 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3744 switch (hw->phy.type) {
3745 case ixgbe_phy_sfp_avago:
3746 case ixgbe_phy_sfp_ftl:
3747 case ixgbe_phy_sfp_intel:
3748 case ixgbe_phy_sfp_unknown:
3749 case ixgbe_phy_sfp_passive_tyco:
3750 case ixgbe_phy_sfp_passive_unknown:
3751 case ixgbe_phy_sfp_active_unknown:
3752 case ixgbe_phy_sfp_ftl_active:
3760 * ixgbe_sfp_link_config - set up SFP+ link
3761 * @adapter: pointer to private adapter struct
3763 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3765 struct ixgbe_hw *hw = &adapter->hw;
3767 if (hw->phy.multispeed_fiber) {
3769 * In multispeed fiber setups, the device may not have
3770 * had a physical connection when the driver loaded.
3771 * If that's the case, the initial link configuration
3772 * couldn't get the MAC into 10G or 1G mode, so we'll
3773 * never have a link status change interrupt fire.
3774 * We need to try and force an autonegotiation
3775 * session, then bring up link.
3777 if (hw->mac.ops.setup_sfp)
3778 hw->mac.ops.setup_sfp(hw);
3779 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
3780 schedule_work(&adapter->multispeed_fiber_task);
3783 * Direct Attach Cu and non-multispeed fiber modules
3784 * still need to be configured properly prior to
3787 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_MOD_TASK))
3788 schedule_work(&adapter->sfp_config_module_task);
3793 * ixgbe_non_sfp_link_config - set up non-SFP+ link
3794 * @hw: pointer to private hardware struct
3796 * Returns 0 on success, negative on failure
3798 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3801 bool negotiation, link_up = false;
3802 u32 ret = IXGBE_ERR_LINK_SETUP;
3804 if (hw->mac.ops.check_link)
3805 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3810 autoneg = hw->phy.autoneg_advertised;
3811 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
3812 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
3817 if (hw->mac.ops.setup_link)
3818 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
3823 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
3825 struct ixgbe_hw *hw = &adapter->hw;
3828 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3829 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
3831 gpie |= IXGBE_GPIE_EIAME;
3833 * use EIAM to auto-mask when MSI-X interrupt is asserted
3834 * this saves a register write for every interrupt
3836 switch (hw->mac.type) {
3837 case ixgbe_mac_82598EB:
3838 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3840 case ixgbe_mac_82599EB:
3841 case ixgbe_mac_X540:
3843 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
3844 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
3848 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
3849 * specifically only auto mask tx and rx interrupts */
3850 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3853 /* XXX: to interrupt immediately for EICS writes, enable this */
3854 /* gpie |= IXGBE_GPIE_EIMEN; */
3856 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3857 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
3858 gpie |= IXGBE_GPIE_VTMODE_64;
3861 /* Enable fan failure interrupt */
3862 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3863 gpie |= IXGBE_SDP1_GPIEN;
3865 if (hw->mac.type == ixgbe_mac_82599EB)
3866 gpie |= IXGBE_SDP1_GPIEN;
3867 gpie |= IXGBE_SDP2_GPIEN;
3869 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3872 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
3874 struct ixgbe_hw *hw = &adapter->hw;
3878 ixgbe_get_hw_control(adapter);
3879 ixgbe_setup_gpie(adapter);
3881 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3882 ixgbe_configure_msix(adapter);
3884 ixgbe_configure_msi_and_legacy(adapter);
3886 /* enable the optics for both mult-speed fiber and 82599 SFP+ fiber */
3887 if (hw->mac.ops.enable_tx_laser &&
3888 ((hw->phy.multispeed_fiber) ||
3889 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
3890 (hw->mac.type == ixgbe_mac_82599EB))))
3891 hw->mac.ops.enable_tx_laser(hw);
3893 clear_bit(__IXGBE_DOWN, &adapter->state);
3894 ixgbe_napi_enable_all(adapter);
3896 if (ixgbe_is_sfp(hw)) {
3897 ixgbe_sfp_link_config(adapter);
3899 err = ixgbe_non_sfp_link_config(hw);
3901 e_err(probe, "link_config FAILED %d\n", err);
3904 /* clear any pending interrupts, may auto mask */
3905 IXGBE_READ_REG(hw, IXGBE_EICR);
3906 ixgbe_irq_enable(adapter, true, true);
3909 * If this adapter has a fan, check to see if we had a failure
3910 * before we enabled the interrupt.
3912 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3913 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
3914 if (esdp & IXGBE_ESDP_SDP1)
3915 e_crit(drv, "Fan has stopped, replace the adapter\n");
3919 * For hot-pluggable SFP+ devices, a new SFP+ module may have
3920 * arrived before interrupts were enabled but after probe. Such
3921 * devices wouldn't have their type identified yet. We need to
3922 * kick off the SFP+ module setup first, then try to bring up link.
3923 * If we're not hot-pluggable SFP+, we just need to configure link
3926 if (hw->phy.type == ixgbe_phy_none)
3927 schedule_work(&adapter->sfp_config_module_task);
3929 /* enable transmits */
3930 netif_tx_start_all_queues(adapter->netdev);
3932 /* bring the link up in the watchdog, this could race with our first
3933 * link up interrupt but shouldn't be a problem */
3934 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3935 adapter->link_check_timeout = jiffies;
3936 mod_timer(&adapter->watchdog_timer, jiffies);
3938 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
3939 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
3940 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
3941 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
3946 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
3948 WARN_ON(in_interrupt());
3949 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
3950 usleep_range(1000, 2000);
3951 ixgbe_down(adapter);
3953 * If SR-IOV enabled then wait a bit before bringing the adapter
3954 * back up to give the VFs time to respond to the reset. The
3955 * two second wait is based upon the watchdog timer cycle in
3958 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3961 clear_bit(__IXGBE_RESETTING, &adapter->state);
3964 int ixgbe_up(struct ixgbe_adapter *adapter)
3966 /* hardware has been reset, we need to reload some things */
3967 ixgbe_configure(adapter);
3969 return ixgbe_up_complete(adapter);
3972 void ixgbe_reset(struct ixgbe_adapter *adapter)
3974 struct ixgbe_hw *hw = &adapter->hw;
3977 err = hw->mac.ops.init_hw(hw);
3980 case IXGBE_ERR_SFP_NOT_PRESENT:
3982 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
3983 e_dev_err("master disable timed out\n");
3985 case IXGBE_ERR_EEPROM_VERSION:
3986 /* We are running on a pre-production device, log a warning */
3987 e_dev_warn("This device is a pre-production adapter/LOM. "
3988 "Please be aware there may be issuesassociated with "
3989 "your hardware. If you are experiencing problems "
3990 "please contact your Intel or hardware "
3991 "representative who provided you with this "
3995 e_dev_err("Hardware Error: %d\n", err);
3998 /* reprogram the RAR[0] in case user changed it. */
3999 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
4004 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
4005 * @rx_ring: ring to free buffers from
4007 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
4009 struct device *dev = rx_ring->dev;
4013 /* ring already cleared, nothing to do */
4014 if (!rx_ring->rx_buffer_info)
4017 /* Free all the Rx ring sk_buffs */
4018 for (i = 0; i < rx_ring->count; i++) {
4019 struct ixgbe_rx_buffer *rx_buffer_info;
4021 rx_buffer_info = &rx_ring->rx_buffer_info[i];
4022 if (rx_buffer_info->dma) {
4023 dma_unmap_single(rx_ring->dev, rx_buffer_info->dma,
4024 rx_ring->rx_buf_len,
4026 rx_buffer_info->dma = 0;
4028 if (rx_buffer_info->skb) {
4029 struct sk_buff *skb = rx_buffer_info->skb;
4030 rx_buffer_info->skb = NULL;
4032 struct sk_buff *this = skb;
4033 if (IXGBE_RSC_CB(this)->delay_unmap) {
4034 dma_unmap_single(dev,
4035 IXGBE_RSC_CB(this)->dma,
4036 rx_ring->rx_buf_len,
4038 IXGBE_RSC_CB(this)->dma = 0;
4039 IXGBE_RSC_CB(skb)->delay_unmap = false;
4042 dev_kfree_skb(this);
4045 if (!rx_buffer_info->page)
4047 if (rx_buffer_info->page_dma) {
4048 dma_unmap_page(dev, rx_buffer_info->page_dma,
4049 PAGE_SIZE / 2, DMA_FROM_DEVICE);
4050 rx_buffer_info->page_dma = 0;
4052 put_page(rx_buffer_info->page);
4053 rx_buffer_info->page = NULL;
4054 rx_buffer_info->page_offset = 0;
4057 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4058 memset(rx_ring->rx_buffer_info, 0, size);
4060 /* Zero out the descriptor ring */
4061 memset(rx_ring->desc, 0, rx_ring->size);
4063 rx_ring->next_to_clean = 0;
4064 rx_ring->next_to_use = 0;
4068 * ixgbe_clean_tx_ring - Free Tx Buffers
4069 * @tx_ring: ring to be cleaned
4071 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
4073 struct ixgbe_tx_buffer *tx_buffer_info;
4077 /* ring already cleared, nothing to do */
4078 if (!tx_ring->tx_buffer_info)
4081 /* Free all the Tx ring sk_buffs */
4082 for (i = 0; i < tx_ring->count; i++) {
4083 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4084 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
4087 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4088 memset(tx_ring->tx_buffer_info, 0, size);
4090 /* Zero out the descriptor ring */
4091 memset(tx_ring->desc, 0, tx_ring->size);
4093 tx_ring->next_to_use = 0;
4094 tx_ring->next_to_clean = 0;
4098 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
4099 * @adapter: board private structure
4101 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
4105 for (i = 0; i < adapter->num_rx_queues; i++)
4106 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
4110 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
4111 * @adapter: board private structure
4113 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
4117 for (i = 0; i < adapter->num_tx_queues; i++)
4118 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
4121 void ixgbe_down(struct ixgbe_adapter *adapter)
4123 struct net_device *netdev = adapter->netdev;
4124 struct ixgbe_hw *hw = &adapter->hw;
4128 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4130 /* signal that we are down to the interrupt handler */
4131 set_bit(__IXGBE_DOWN, &adapter->state);
4133 /* disable receive for all VFs and wait one second */
4134 if (adapter->num_vfs) {
4135 /* ping all the active vfs to let them know we are going down */
4136 ixgbe_ping_all_vfs(adapter);
4138 /* Disable all VFTE/VFRE TX/RX */
4139 ixgbe_disable_tx_rx(adapter);
4141 /* Mark all the VFs as inactive */
4142 for (i = 0 ; i < adapter->num_vfs; i++)
4143 adapter->vfinfo[i].clear_to_send = 0;
4146 /* disable receives */
4147 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4148 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
4150 /* disable all enabled rx queues */
4151 for (i = 0; i < adapter->num_rx_queues; i++)
4152 /* this call also flushes the previous write */
4153 ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]);
4155 usleep_range(10000, 20000);
4157 netif_tx_stop_all_queues(netdev);
4159 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
4160 del_timer_sync(&adapter->sfp_timer);
4161 del_timer_sync(&adapter->watchdog_timer);
4162 cancel_work_sync(&adapter->watchdog_task);
4164 netif_carrier_off(netdev);
4165 netif_tx_disable(netdev);
4167 ixgbe_irq_disable(adapter);
4169 ixgbe_napi_disable_all(adapter);
4171 /* Cleanup the affinity_hint CPU mask memory and callback */
4172 for (i = 0; i < num_q_vectors; i++) {
4173 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
4174 /* clear the affinity_mask in the IRQ descriptor */
4175 irq_set_affinity_hint(adapter->msix_entries[i]. vector, NULL);
4176 /* release the CPU mask memory */
4177 free_cpumask_var(q_vector->affinity_mask);
4180 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
4181 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
4182 cancel_work_sync(&adapter->fdir_reinit_task);
4184 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
4185 cancel_work_sync(&adapter->check_overtemp_task);
4187 /* disable transmits in the hardware now that interrupts are off */
4188 for (i = 0; i < adapter->num_tx_queues; i++) {
4189 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
4190 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
4191 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx),
4192 (txdctl & ~IXGBE_TXDCTL_ENABLE));
4194 /* Disable the Tx DMA engine on 82599 */
4195 switch (hw->mac.type) {
4196 case ixgbe_mac_82599EB:
4197 case ixgbe_mac_X540:
4198 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
4199 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
4200 ~IXGBE_DMATXCTL_TE));
4206 if (!pci_channel_offline(adapter->pdev))
4207 ixgbe_reset(adapter);
4209 /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
4210 if (hw->mac.ops.disable_tx_laser &&
4211 ((hw->phy.multispeed_fiber) ||
4212 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
4213 (hw->mac.type == ixgbe_mac_82599EB))))
4214 hw->mac.ops.disable_tx_laser(hw);
4216 ixgbe_clean_all_tx_rings(adapter);
4217 ixgbe_clean_all_rx_rings(adapter);
4219 #ifdef CONFIG_IXGBE_DCA
4220 /* since we reset the hardware DCA settings were cleared */
4221 ixgbe_setup_dca(adapter);
4226 * ixgbe_poll - NAPI Rx polling callback
4227 * @napi: structure for representing this polling device
4228 * @budget: how many packets driver is allowed to clean
4230 * This function is used for legacy and MSI, NAPI mode
4232 static int ixgbe_poll(struct napi_struct *napi, int budget)
4234 struct ixgbe_q_vector *q_vector =
4235 container_of(napi, struct ixgbe_q_vector, napi);
4236 struct ixgbe_adapter *adapter = q_vector->adapter;
4237 int tx_clean_complete, work_done = 0;
4239 #ifdef CONFIG_IXGBE_DCA
4240 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
4241 ixgbe_update_dca(q_vector);
4244 tx_clean_complete = ixgbe_clean_tx_irq(q_vector, adapter->tx_ring[0]);
4245 ixgbe_clean_rx_irq(q_vector, adapter->rx_ring[0], &work_done, budget);
4247 if (!tx_clean_complete)
4250 /* If budget not fully consumed, exit the polling mode */
4251 if (work_done < budget) {
4252 napi_complete(napi);
4253 if (adapter->rx_itr_setting & 1)
4254 ixgbe_set_itr(adapter);
4255 if (!test_bit(__IXGBE_DOWN, &adapter->state))
4256 ixgbe_irq_enable_queues(adapter, IXGBE_EIMS_RTX_QUEUE);
4262 * ixgbe_tx_timeout - Respond to a Tx Hang
4263 * @netdev: network interface device structure
4265 static void ixgbe_tx_timeout(struct net_device *netdev)
4267 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4269 adapter->tx_timeout_count++;
4271 /* Do the reset outside of interrupt context */
4272 schedule_work(&adapter->reset_task);
4275 static void ixgbe_reset_task(struct work_struct *work)
4277 struct ixgbe_adapter *adapter;
4278 adapter = container_of(work, struct ixgbe_adapter, reset_task);
4280 /* If we're already down or resetting, just bail */
4281 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
4282 test_bit(__IXGBE_RESETTING, &adapter->state))
4285 ixgbe_dump(adapter);
4286 netdev_err(adapter->netdev, "Reset adapter\n");
4287 ixgbe_reinit_locked(adapter);
4291 * ixgbe_set_rss_queues: Allocate queues for RSS
4292 * @adapter: board private structure to initialize
4294 * This is our "base" multiqueue mode. RSS (Receive Side Scaling) will try
4295 * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
4298 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
4301 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
4303 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4305 adapter->num_rx_queues = f->indices;
4306 adapter->num_tx_queues = f->indices;
4316 * ixgbe_set_fdir_queues: Allocate queues for Flow Director
4317 * @adapter: board private structure to initialize
4319 * Flow Director is an advanced Rx filter, attempting to get Rx flows back
4320 * to the original CPU that initiated the Tx session. This runs in addition
4321 * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
4322 * Rx load across CPUs using RSS.
4325 static inline bool ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
4328 struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
4330 f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
4333 /* Flow Director must have RSS enabled */
4334 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
4335 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
4336 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)))) {
4337 adapter->num_tx_queues = f_fdir->indices;
4338 adapter->num_rx_queues = f_fdir->indices;
4341 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4342 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4349 * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
4350 * @adapter: board private structure to initialize
4352 * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
4353 * The ring feature mask is not used as a mask for FCoE, as it can take any 8
4354 * rx queues out of the max number of rx queues, instead, it is used as the
4355 * index of the first rx queue used by FCoE.
4358 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
4360 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4362 if (!(adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
4365 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4366 #ifdef CONFIG_IXGBE_DCB
4368 struct net_device *dev = adapter->netdev;
4370 tc = netdev_get_prio_tc_map(dev, adapter->fcoe.up);
4371 f->indices = dev->tc_to_txq[tc].count;
4372 f->mask = dev->tc_to_txq[tc].offset;
4375 f->indices = min((int)num_online_cpus(), f->indices);
4377 adapter->num_rx_queues = 1;
4378 adapter->num_tx_queues = 1;
4380 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4381 e_info(probe, "FCoE enabled with RSS\n");
4382 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4383 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
4384 ixgbe_set_fdir_queues(adapter);
4386 ixgbe_set_rss_queues(adapter);
4388 /* adding FCoE rx rings to the end */
4389 f->mask = adapter->num_rx_queues;
4390 adapter->num_rx_queues += f->indices;
4391 adapter->num_tx_queues += f->indices;
4396 #endif /* IXGBE_FCOE */
4398 #ifdef CONFIG_IXGBE_DCB
4399 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
4402 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_DCB];
4405 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
4409 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
4410 q = min((int)num_online_cpus(), MAX_TRAFFIC_CLASS);
4415 adapter->num_rx_queues = f->indices;
4416 adapter->num_tx_queues = f->indices;
4420 /* FCoE enabled queues require special configuration done through
4421 * configure_fcoe() and others. Here we map FCoE indices onto the
4422 * DCB queue pairs allowing FCoE to own configuration later.
4424 ixgbe_set_fcoe_queues(adapter);
4432 * ixgbe_set_sriov_queues: Allocate queues for IOV use
4433 * @adapter: board private structure to initialize
4435 * IOV doesn't actually use anything, so just NAK the
4436 * request for now and let the other queue routines
4437 * figure out what to do.
4439 static inline bool ixgbe_set_sriov_queues(struct ixgbe_adapter *adapter)
4445 * ixgbe_set_num_queues: Allocate queues for device, feature dependent
4446 * @adapter: board private structure to initialize
4448 * This is the top level queue allocation routine. The order here is very
4449 * important, starting with the "most" number of features turned on at once,
4450 * and ending with the smallest set of features. This way large combinations
4451 * can be allocated if they're turned on, and smaller combinations are the
4452 * fallthrough conditions.
4455 static int ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
4457 /* Start with base case */
4458 adapter->num_rx_queues = 1;
4459 adapter->num_tx_queues = 1;
4460 adapter->num_rx_pools = adapter->num_rx_queues;
4461 adapter->num_rx_queues_per_pool = 1;
4463 if (ixgbe_set_sriov_queues(adapter))
4466 #ifdef CONFIG_IXGBE_DCB
4467 if (ixgbe_set_dcb_queues(adapter))
4472 if (ixgbe_set_fcoe_queues(adapter))
4475 #endif /* IXGBE_FCOE */
4476 if (ixgbe_set_fdir_queues(adapter))
4479 if (ixgbe_set_rss_queues(adapter))
4482 /* fallback to base case */
4483 adapter->num_rx_queues = 1;
4484 adapter->num_tx_queues = 1;
4487 /* Notify the stack of the (possibly) reduced queue counts. */
4488 netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
4489 return netif_set_real_num_rx_queues(adapter->netdev,
4490 adapter->num_rx_queues);
4493 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
4496 int err, vector_threshold;
4498 /* We'll want at least 3 (vector_threshold):
4501 * 3) Other (Link Status Change, etc.)
4502 * 4) TCP Timer (optional)
4504 vector_threshold = MIN_MSIX_COUNT;
4506 /* The more we get, the more we will assign to Tx/Rx Cleanup
4507 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
4508 * Right now, we simply care about how many we'll get; we'll
4509 * set them up later while requesting irq's.
4511 while (vectors >= vector_threshold) {
4512 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
4514 if (!err) /* Success in acquiring all requested vectors. */
4517 vectors = 0; /* Nasty failure, quit now */
4518 else /* err == number of vectors we should try again with */
4522 if (vectors < vector_threshold) {
4523 /* Can't allocate enough MSI-X interrupts? Oh well.
4524 * This just means we'll go with either a single MSI
4525 * vector or fall back to legacy interrupts.
4527 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4528 "Unable to allocate MSI-X interrupts\n");
4529 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4530 kfree(adapter->msix_entries);
4531 adapter->msix_entries = NULL;
4533 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
4535 * Adjust for only the vectors we'll use, which is minimum
4536 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
4537 * vectors we were allocated.
4539 adapter->num_msix_vectors = min(vectors,
4540 adapter->max_msix_q_vectors + NON_Q_VECTORS);
4545 * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
4546 * @adapter: board private structure to initialize
4548 * Cache the descriptor ring offsets for RSS to the assigned rings.
4551 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
4555 if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED))
4558 for (i = 0; i < adapter->num_rx_queues; i++)
4559 adapter->rx_ring[i]->reg_idx = i;
4560 for (i = 0; i < adapter->num_tx_queues; i++)
4561 adapter->tx_ring[i]->reg_idx = i;
4566 #ifdef CONFIG_IXGBE_DCB
4568 /* ixgbe_get_first_reg_idx - Return first register index associated with ring */
4569 static void ixgbe_get_first_reg_idx(struct ixgbe_adapter *adapter, u8 tc,
4570 unsigned int *tx, unsigned int *rx)
4572 struct net_device *dev = adapter->netdev;
4573 struct ixgbe_hw *hw = &adapter->hw;
4574 u8 num_tcs = netdev_get_num_tc(dev);
4579 switch (hw->mac.type) {
4580 case ixgbe_mac_82598EB:
4584 case ixgbe_mac_82599EB:
4585 case ixgbe_mac_X540:
4590 } else if (tc < 5) {
4591 *tx = ((tc + 2) << 4);
4593 } else if (tc < num_tcs) {
4594 *tx = ((tc + 8) << 3);
4597 } else if (num_tcs == 4) {
4622 #define IXGBE_MAX_Q_PER_TC (IXGBE_MAX_DCB_INDICES / MAX_TRAFFIC_CLASS)
4624 /* ixgbe_setup_tc - routine to configure net_device for multiple traffic
4627 * @netdev: net device to configure
4628 * @tc: number of traffic classes to enable
4630 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
4633 unsigned int q, offset = 0;
4636 netdev_reset_tc(dev);
4638 struct ixgbe_adapter *adapter = netdev_priv(dev);
4640 /* Hardware supports up to 8 traffic classes */
4641 if (tc > MAX_TRAFFIC_CLASS || netdev_set_num_tc(dev, tc))
4644 /* Partition Tx queues evenly amongst traffic classes */
4645 for (i = 0; i < tc; i++) {
4646 q = min((int)num_online_cpus(), IXGBE_MAX_Q_PER_TC);
4647 netdev_set_prio_tc_map(dev, i, i);
4648 netdev_set_tc_queue(dev, i, q, offset);
4652 /* This enables multiple traffic class support in the hardware
4653 * which defaults to strict priority transmission by default.
4654 * If traffic classes are already enabled perhaps through DCB
4655 * code path then existing configuration will be used.
4657 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
4658 dev->dcbnl_ops && dev->dcbnl_ops->setdcbx) {
4659 struct ieee_ets ets = {
4660 .prio_tc = {0, 1, 2, 3, 4, 5, 6, 7},
4662 u8 mode = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_IEEE;
4664 dev->dcbnl_ops->setdcbx(dev, mode);
4665 dev->dcbnl_ops->ieee_setets(dev, &ets);
4672 * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
4673 * @adapter: board private structure to initialize
4675 * Cache the descriptor ring offsets for DCB to the assigned rings.
4678 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
4680 struct net_device *dev = adapter->netdev;
4682 u8 num_tcs = netdev_get_num_tc(dev);
4684 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
4687 for (i = 0, k = 0; i < num_tcs; i++) {
4688 unsigned int tx_s, rx_s;
4689 u16 count = dev->tc_to_txq[i].count;
4691 ixgbe_get_first_reg_idx(adapter, i, &tx_s, &rx_s);
4692 for (j = 0; j < count; j++, k++) {
4693 adapter->tx_ring[k]->reg_idx = tx_s + j;
4694 adapter->rx_ring[k]->reg_idx = rx_s + j;
4695 adapter->tx_ring[k]->dcb_tc = i;
4696 adapter->rx_ring[k]->dcb_tc = i;
4705 * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
4706 * @adapter: board private structure to initialize
4708 * Cache the descriptor ring offsets for Flow Director to the assigned rings.
4711 static inline bool ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
4716 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
4717 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4718 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))) {
4719 for (i = 0; i < adapter->num_rx_queues; i++)
4720 adapter->rx_ring[i]->reg_idx = i;
4721 for (i = 0; i < adapter->num_tx_queues; i++)
4722 adapter->tx_ring[i]->reg_idx = i;
4731 * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
4732 * @adapter: board private structure to initialize
4734 * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
4737 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
4739 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4741 u8 fcoe_rx_i = 0, fcoe_tx_i = 0;
4743 if (!(adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
4746 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4747 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4748 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
4749 ixgbe_cache_ring_fdir(adapter);
4751 ixgbe_cache_ring_rss(adapter);
4753 fcoe_rx_i = f->mask;
4754 fcoe_tx_i = f->mask;
4756 for (i = 0; i < f->indices; i++, fcoe_rx_i++, fcoe_tx_i++) {
4757 adapter->rx_ring[f->mask + i]->reg_idx = fcoe_rx_i;
4758 adapter->tx_ring[f->mask + i]->reg_idx = fcoe_tx_i;
4763 #endif /* IXGBE_FCOE */
4765 * ixgbe_cache_ring_sriov - Descriptor ring to register mapping for sriov
4766 * @adapter: board private structure to initialize
4768 * SR-IOV doesn't use any descriptor rings but changes the default if
4769 * no other mapping is used.
4772 static inline bool ixgbe_cache_ring_sriov(struct ixgbe_adapter *adapter)
4774 adapter->rx_ring[0]->reg_idx = adapter->num_vfs * 2;
4775 adapter->tx_ring[0]->reg_idx = adapter->num_vfs * 2;
4776 if (adapter->num_vfs)
4783 * ixgbe_cache_ring_register - Descriptor ring to register mapping
4784 * @adapter: board private structure to initialize
4786 * Once we know the feature-set enabled for the device, we'll cache
4787 * the register offset the descriptor ring is assigned to.
4789 * Note, the order the various feature calls is important. It must start with
4790 * the "most" features enabled at the same time, then trickle down to the
4791 * least amount of features turned on at once.
4793 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
4795 /* start with default case */
4796 adapter->rx_ring[0]->reg_idx = 0;
4797 adapter->tx_ring[0]->reg_idx = 0;
4799 if (ixgbe_cache_ring_sriov(adapter))
4802 #ifdef CONFIG_IXGBE_DCB
4803 if (ixgbe_cache_ring_dcb(adapter))
4808 if (ixgbe_cache_ring_fcoe(adapter))
4810 #endif /* IXGBE_FCOE */
4812 if (ixgbe_cache_ring_fdir(adapter))
4815 if (ixgbe_cache_ring_rss(adapter))
4820 * ixgbe_alloc_queues - Allocate memory for all rings
4821 * @adapter: board private structure to initialize
4823 * We allocate one ring per queue at run-time since we don't know the
4824 * number of queues at compile-time. The polling_netdev array is
4825 * intended for Multiqueue, but should work fine with a single queue.
4827 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
4829 int rx = 0, tx = 0, nid = adapter->node;
4831 if (nid < 0 || !node_online(nid))
4832 nid = first_online_node;
4834 for (; tx < adapter->num_tx_queues; tx++) {
4835 struct ixgbe_ring *ring;
4837 ring = kzalloc_node(sizeof(*ring), GFP_KERNEL, nid);
4839 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
4841 goto err_allocation;
4842 ring->count = adapter->tx_ring_count;
4843 ring->queue_index = tx;
4844 ring->numa_node = nid;
4845 ring->dev = &adapter->pdev->dev;
4846 ring->netdev = adapter->netdev;
4848 adapter->tx_ring[tx] = ring;
4851 for (; rx < adapter->num_rx_queues; rx++) {
4852 struct ixgbe_ring *ring;
4854 ring = kzalloc_node(sizeof(*ring), GFP_KERNEL, nid);
4856 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
4858 goto err_allocation;
4859 ring->count = adapter->rx_ring_count;
4860 ring->queue_index = rx;
4861 ring->numa_node = nid;
4862 ring->dev = &adapter->pdev->dev;
4863 ring->netdev = adapter->netdev;
4865 adapter->rx_ring[rx] = ring;
4868 ixgbe_cache_ring_register(adapter);
4874 kfree(adapter->tx_ring[--tx]);
4877 kfree(adapter->rx_ring[--rx]);
4882 * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
4883 * @adapter: board private structure to initialize
4885 * Attempt to configure the interrupts using the best available
4886 * capabilities of the hardware and the kernel.
4888 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
4890 struct ixgbe_hw *hw = &adapter->hw;
4892 int vector, v_budget;
4895 * It's easy to be greedy for MSI-X vectors, but it really
4896 * doesn't do us much good if we have a lot more vectors
4897 * than CPU's. So let's be conservative and only ask for
4898 * (roughly) the same number of vectors as there are CPU's.
4900 v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
4901 (int)num_online_cpus()) + NON_Q_VECTORS;
4904 * At the same time, hardware can only support a maximum of
4905 * hw.mac->max_msix_vectors vectors. With features
4906 * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
4907 * descriptor queues supported by our device. Thus, we cap it off in
4908 * those rare cases where the cpu count also exceeds our vector limit.
4910 v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
4912 /* A failure in MSI-X entry allocation isn't fatal, but it does
4913 * mean we disable MSI-X capabilities of the adapter. */
4914 adapter->msix_entries = kcalloc(v_budget,
4915 sizeof(struct msix_entry), GFP_KERNEL);
4916 if (adapter->msix_entries) {
4917 for (vector = 0; vector < v_budget; vector++)
4918 adapter->msix_entries[vector].entry = vector;
4920 ixgbe_acquire_msix_vectors(adapter, v_budget);
4922 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4926 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
4927 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
4928 if (adapter->flags & (IXGBE_FLAG_FDIR_HASH_CAPABLE |
4929 IXGBE_FLAG_FDIR_PERFECT_CAPABLE)) {
4931 "Flow Director is not supported while multiple "
4932 "queues are disabled. Disabling Flow Director\n");
4934 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4935 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4936 adapter->atr_sample_rate = 0;
4937 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4938 ixgbe_disable_sriov(adapter);
4940 err = ixgbe_set_num_queues(adapter);
4944 err = pci_enable_msi(adapter->pdev);
4946 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
4948 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4949 "Unable to allocate MSI interrupt, "
4950 "falling back to legacy. Error: %d\n", err);
4960 * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
4961 * @adapter: board private structure to initialize
4963 * We allocate one q_vector per queue interrupt. If allocation fails we
4966 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
4968 int q_idx, num_q_vectors;
4969 struct ixgbe_q_vector *q_vector;
4970 int (*poll)(struct napi_struct *, int);
4972 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4973 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4974 poll = &ixgbe_clean_rxtx_many;
4980 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4981 q_vector = kzalloc_node(sizeof(struct ixgbe_q_vector),
4982 GFP_KERNEL, adapter->node);
4984 q_vector = kzalloc(sizeof(struct ixgbe_q_vector),
4988 q_vector->adapter = adapter;
4989 if (q_vector->txr_count && !q_vector->rxr_count)
4990 q_vector->eitr = adapter->tx_eitr_param;
4992 q_vector->eitr = adapter->rx_eitr_param;
4993 q_vector->v_idx = q_idx;
4994 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
4995 adapter->q_vector[q_idx] = q_vector;
5003 q_vector = adapter->q_vector[q_idx];
5004 netif_napi_del(&q_vector->napi);
5006 adapter->q_vector[q_idx] = NULL;
5012 * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
5013 * @adapter: board private structure to initialize
5015 * This function frees the memory allocated to the q_vectors. In addition if
5016 * NAPI is enabled it will delete any references to the NAPI struct prior
5017 * to freeing the q_vector.
5019 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
5021 int q_idx, num_q_vectors;
5023 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
5024 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
5028 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
5029 struct ixgbe_q_vector *q_vector = adapter->q_vector[q_idx];
5030 adapter->q_vector[q_idx] = NULL;
5031 netif_napi_del(&q_vector->napi);
5036 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
5038 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
5039 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
5040 pci_disable_msix(adapter->pdev);
5041 kfree(adapter->msix_entries);
5042 adapter->msix_entries = NULL;
5043 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
5044 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
5045 pci_disable_msi(adapter->pdev);
5050 * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
5051 * @adapter: board private structure to initialize
5053 * We determine which interrupt scheme to use based on...
5054 * - Kernel support (MSI, MSI-X)
5055 * - which can be user-defined (via MODULE_PARAM)
5056 * - Hardware queue count (num_*_queues)
5057 * - defined by miscellaneous hardware support/features (RSS, etc.)
5059 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
5063 /* Number of supported queues */
5064 err = ixgbe_set_num_queues(adapter);
5068 err = ixgbe_set_interrupt_capability(adapter);
5070 e_dev_err("Unable to setup interrupt capabilities\n");
5071 goto err_set_interrupt;
5074 err = ixgbe_alloc_q_vectors(adapter);
5076 e_dev_err("Unable to allocate memory for queue vectors\n");
5077 goto err_alloc_q_vectors;
5080 err = ixgbe_alloc_queues(adapter);
5082 e_dev_err("Unable to allocate memory for queues\n");
5083 goto err_alloc_queues;
5086 e_dev_info("Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
5087 (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
5088 adapter->num_rx_queues, adapter->num_tx_queues);
5090 set_bit(__IXGBE_DOWN, &adapter->state);
5095 ixgbe_free_q_vectors(adapter);
5096 err_alloc_q_vectors:
5097 ixgbe_reset_interrupt_capability(adapter);
5102 static void ring_free_rcu(struct rcu_head *head)
5104 kfree(container_of(head, struct ixgbe_ring, rcu));
5108 * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
5109 * @adapter: board private structure to clear interrupt scheme on
5111 * We go through and clear interrupt specific resources and reset the structure
5112 * to pre-load conditions
5114 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
5118 for (i = 0; i < adapter->num_tx_queues; i++) {
5119 kfree(adapter->tx_ring[i]);
5120 adapter->tx_ring[i] = NULL;
5122 for (i = 0; i < adapter->num_rx_queues; i++) {
5123 struct ixgbe_ring *ring = adapter->rx_ring[i];
5125 /* ixgbe_get_stats64() might access this ring, we must wait
5126 * a grace period before freeing it.
5128 call_rcu(&ring->rcu, ring_free_rcu);
5129 adapter->rx_ring[i] = NULL;
5132 adapter->num_tx_queues = 0;
5133 adapter->num_rx_queues = 0;
5135 ixgbe_free_q_vectors(adapter);
5136 ixgbe_reset_interrupt_capability(adapter);
5140 * ixgbe_sfp_timer - worker thread to find a missing module
5141 * @data: pointer to our adapter struct
5143 static void ixgbe_sfp_timer(unsigned long data)
5145 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5148 * Do the sfp_timer outside of interrupt context due to the
5149 * delays that sfp+ detection requires
5151 schedule_work(&adapter->sfp_task);
5155 * ixgbe_sfp_task - worker thread to find a missing module
5156 * @work: pointer to work_struct containing our data
5158 static void ixgbe_sfp_task(struct work_struct *work)
5160 struct ixgbe_adapter *adapter = container_of(work,
5161 struct ixgbe_adapter,
5163 struct ixgbe_hw *hw = &adapter->hw;
5165 if ((hw->phy.type == ixgbe_phy_nl) &&
5166 (hw->phy.sfp_type == ixgbe_sfp_type_not_present)) {
5167 s32 ret = hw->phy.ops.identify_sfp(hw);
5168 if (ret == IXGBE_ERR_SFP_NOT_PRESENT)
5170 ret = hw->phy.ops.reset(hw);
5171 if (ret == IXGBE_ERR_SFP_NOT_SUPPORTED) {
5172 e_dev_err("failed to initialize because an unsupported "
5173 "SFP+ module type was detected.\n");
5174 e_dev_err("Reload the driver after installing a "
5175 "supported module.\n");
5176 unregister_netdev(adapter->netdev);
5178 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
5180 /* don't need this routine any more */
5181 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
5185 if (test_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state))
5186 mod_timer(&adapter->sfp_timer,
5187 round_jiffies(jiffies + (2 * HZ)));
5191 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
5192 * @adapter: board private structure to initialize
5194 * ixgbe_sw_init initializes the Adapter private data structure.
5195 * Fields are initialized based on PCI device information and
5196 * OS network device settings (MTU size).
5198 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
5200 struct ixgbe_hw *hw = &adapter->hw;
5201 struct pci_dev *pdev = adapter->pdev;
5202 struct net_device *dev = adapter->netdev;
5204 #ifdef CONFIG_IXGBE_DCB
5206 struct tc_configuration *tc;
5208 int max_frame = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
5210 /* PCI config space info */
5212 hw->vendor_id = pdev->vendor;
5213 hw->device_id = pdev->device;
5214 hw->revision_id = pdev->revision;
5215 hw->subsystem_vendor_id = pdev->subsystem_vendor;
5216 hw->subsystem_device_id = pdev->subsystem_device;
5218 /* Set capability flags */
5219 rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
5220 adapter->ring_feature[RING_F_RSS].indices = rss;
5221 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
5222 adapter->ring_feature[RING_F_DCB].indices = IXGBE_MAX_DCB_INDICES;
5223 switch (hw->mac.type) {
5224 case ixgbe_mac_82598EB:
5225 if (hw->device_id == IXGBE_DEV_ID_82598AT)
5226 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
5227 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
5229 case ixgbe_mac_82599EB:
5230 case ixgbe_mac_X540:
5231 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
5232 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
5233 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
5234 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
5235 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
5236 /* n-tuple support exists, always init our spinlock */
5237 spin_lock_init(&adapter->fdir_perfect_lock);
5238 /* Flow Director hash filters enabled */
5239 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
5240 adapter->atr_sample_rate = 20;
5241 adapter->ring_feature[RING_F_FDIR].indices =
5242 IXGBE_MAX_FDIR_INDICES;
5243 adapter->fdir_pballoc = 0;
5245 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
5246 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
5247 adapter->ring_feature[RING_F_FCOE].indices = 0;
5248 #ifdef CONFIG_IXGBE_DCB
5249 /* Default traffic class to use for FCoE */
5250 adapter->fcoe.tc = IXGBE_FCOE_DEFTC;
5251 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
5253 #endif /* IXGBE_FCOE */
5259 #ifdef CONFIG_IXGBE_DCB
5260 /* Configure DCB traffic classes */
5261 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
5262 tc = &adapter->dcb_cfg.tc_config[j];
5263 tc->path[DCB_TX_CONFIG].bwg_id = 0;
5264 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
5265 tc->path[DCB_RX_CONFIG].bwg_id = 0;
5266 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
5267 tc->dcb_pfc = pfc_disabled;
5269 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
5270 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
5271 adapter->dcb_cfg.rx_pba_cfg = pba_equal;
5272 adapter->dcb_cfg.pfc_mode_enable = false;
5273 adapter->dcb_set_bitmap = 0x00;
5274 adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
5275 ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
5280 /* default flow control settings */
5281 hw->fc.requested_mode = ixgbe_fc_full;
5282 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
5284 adapter->last_lfc_mode = hw->fc.current_mode;
5286 hw->fc.high_water = FC_HIGH_WATER(max_frame);
5287 hw->fc.low_water = FC_LOW_WATER(max_frame);
5288 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
5289 hw->fc.send_xon = true;
5290 hw->fc.disable_fc_autoneg = false;
5292 /* enable itr by default in dynamic mode */
5293 adapter->rx_itr_setting = 1;
5294 adapter->rx_eitr_param = 20000;
5295 adapter->tx_itr_setting = 1;
5296 adapter->tx_eitr_param = 10000;
5298 /* set defaults for eitr in MegaBytes */
5299 adapter->eitr_low = 10;
5300 adapter->eitr_high = 20;
5302 /* set default ring sizes */
5303 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
5304 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
5306 /* initialize eeprom parameters */
5307 if (ixgbe_init_eeprom_params_generic(hw)) {
5308 e_dev_err("EEPROM initialization failed\n");
5312 /* enable rx csum by default */
5313 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
5315 /* get assigned NUMA node */
5316 adapter->node = dev_to_node(&pdev->dev);
5318 set_bit(__IXGBE_DOWN, &adapter->state);
5324 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
5325 * @tx_ring: tx descriptor ring (for a specific queue) to setup
5327 * Return 0 on success, negative on failure
5329 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
5331 struct device *dev = tx_ring->dev;
5334 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
5335 tx_ring->tx_buffer_info = vzalloc_node(size, tx_ring->numa_node);
5336 if (!tx_ring->tx_buffer_info)
5337 tx_ring->tx_buffer_info = vzalloc(size);
5338 if (!tx_ring->tx_buffer_info)
5341 /* round up to nearest 4K */
5342 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
5343 tx_ring->size = ALIGN(tx_ring->size, 4096);
5345 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
5346 &tx_ring->dma, GFP_KERNEL);
5350 tx_ring->next_to_use = 0;
5351 tx_ring->next_to_clean = 0;
5352 tx_ring->work_limit = tx_ring->count;
5356 vfree(tx_ring->tx_buffer_info);
5357 tx_ring->tx_buffer_info = NULL;
5358 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
5363 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
5364 * @adapter: board private structure
5366 * If this function returns with an error, then it's possible one or
5367 * more of the rings is populated (while the rest are not). It is the
5368 * callers duty to clean those orphaned rings.
5370 * Return 0 on success, negative on failure
5372 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
5376 for (i = 0; i < adapter->num_tx_queues; i++) {
5377 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
5380 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
5388 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
5389 * @rx_ring: rx descriptor ring (for a specific queue) to setup
5391 * Returns 0 on success, negative on failure
5393 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
5395 struct device *dev = rx_ring->dev;
5398 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
5399 rx_ring->rx_buffer_info = vzalloc_node(size, rx_ring->numa_node);
5400 if (!rx_ring->rx_buffer_info)
5401 rx_ring->rx_buffer_info = vzalloc(size);
5402 if (!rx_ring->rx_buffer_info)
5405 /* Round up to nearest 4K */
5406 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
5407 rx_ring->size = ALIGN(rx_ring->size, 4096);
5409 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
5410 &rx_ring->dma, GFP_KERNEL);
5415 rx_ring->next_to_clean = 0;
5416 rx_ring->next_to_use = 0;
5420 vfree(rx_ring->rx_buffer_info);
5421 rx_ring->rx_buffer_info = NULL;
5422 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
5427 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
5428 * @adapter: board private structure
5430 * If this function returns with an error, then it's possible one or
5431 * more of the rings is populated (while the rest are not). It is the
5432 * callers duty to clean those orphaned rings.
5434 * Return 0 on success, negative on failure
5436 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
5440 for (i = 0; i < adapter->num_rx_queues; i++) {
5441 err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
5444 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
5452 * ixgbe_free_tx_resources - Free Tx Resources per Queue
5453 * @tx_ring: Tx descriptor ring for a specific queue
5455 * Free all transmit software resources
5457 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
5459 ixgbe_clean_tx_ring(tx_ring);
5461 vfree(tx_ring->tx_buffer_info);
5462 tx_ring->tx_buffer_info = NULL;
5464 /* if not set, then don't free */
5468 dma_free_coherent(tx_ring->dev, tx_ring->size,
5469 tx_ring->desc, tx_ring->dma);
5471 tx_ring->desc = NULL;
5475 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5476 * @adapter: board private structure
5478 * Free all transmit software resources
5480 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
5484 for (i = 0; i < adapter->num_tx_queues; i++)
5485 if (adapter->tx_ring[i]->desc)
5486 ixgbe_free_tx_resources(adapter->tx_ring[i]);
5490 * ixgbe_free_rx_resources - Free Rx Resources
5491 * @rx_ring: ring to clean the resources from
5493 * Free all receive software resources
5495 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
5497 ixgbe_clean_rx_ring(rx_ring);
5499 vfree(rx_ring->rx_buffer_info);
5500 rx_ring->rx_buffer_info = NULL;
5502 /* if not set, then don't free */
5506 dma_free_coherent(rx_ring->dev, rx_ring->size,
5507 rx_ring->desc, rx_ring->dma);
5509 rx_ring->desc = NULL;
5513 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5514 * @adapter: board private structure
5516 * Free all receive software resources
5518 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5522 for (i = 0; i < adapter->num_rx_queues; i++)
5523 if (adapter->rx_ring[i]->desc)
5524 ixgbe_free_rx_resources(adapter->rx_ring[i]);
5528 * ixgbe_change_mtu - Change the Maximum Transfer Unit
5529 * @netdev: network interface device structure
5530 * @new_mtu: new value for maximum frame size
5532 * Returns 0 on success, negative on failure
5534 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5536 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5537 struct ixgbe_hw *hw = &adapter->hw;
5538 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5540 /* MTU < 68 is an error and causes problems on some kernels */
5541 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED &&
5542 hw->mac.type != ixgbe_mac_X540) {
5543 if ((new_mtu < 68) || (max_frame > MAXIMUM_ETHERNET_VLAN_SIZE))
5546 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5550 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5551 /* must set new MTU before calling down or up */
5552 netdev->mtu = new_mtu;
5554 hw->fc.high_water = FC_HIGH_WATER(max_frame);
5555 hw->fc.low_water = FC_LOW_WATER(max_frame);
5557 if (netif_running(netdev))
5558 ixgbe_reinit_locked(adapter);
5564 * ixgbe_open - Called when a network interface is made active
5565 * @netdev: network interface device structure
5567 * Returns 0 on success, negative value on failure
5569 * The open entry point is called when a network interface is made
5570 * active by the system (IFF_UP). At this point all resources needed
5571 * for transmit and receive operations are allocated, the interrupt
5572 * handler is registered with the OS, the watchdog timer is started,
5573 * and the stack is notified that the interface is ready.
5575 static int ixgbe_open(struct net_device *netdev)
5577 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5580 /* disallow open during test */
5581 if (test_bit(__IXGBE_TESTING, &adapter->state))
5584 netif_carrier_off(netdev);
5586 /* allocate transmit descriptors */
5587 err = ixgbe_setup_all_tx_resources(adapter);
5591 /* allocate receive descriptors */
5592 err = ixgbe_setup_all_rx_resources(adapter);
5596 ixgbe_configure(adapter);
5598 err = ixgbe_request_irq(adapter);
5602 err = ixgbe_up_complete(adapter);
5606 netif_tx_start_all_queues(netdev);
5611 ixgbe_release_hw_control(adapter);
5612 ixgbe_free_irq(adapter);
5615 ixgbe_free_all_rx_resources(adapter);
5617 ixgbe_free_all_tx_resources(adapter);
5618 ixgbe_reset(adapter);
5624 * ixgbe_close - Disables a network interface
5625 * @netdev: network interface device structure
5627 * Returns 0, this is not allowed to fail
5629 * The close entry point is called when an interface is de-activated
5630 * by the OS. The hardware is still under the drivers control, but
5631 * needs to be disabled. A global MAC reset is issued to stop the
5632 * hardware, and all transmit and receive resources are freed.
5634 static int ixgbe_close(struct net_device *netdev)
5636 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5638 ixgbe_down(adapter);
5639 ixgbe_free_irq(adapter);
5641 ixgbe_free_all_tx_resources(adapter);
5642 ixgbe_free_all_rx_resources(adapter);
5644 ixgbe_release_hw_control(adapter);
5650 static int ixgbe_resume(struct pci_dev *pdev)
5652 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5653 struct net_device *netdev = adapter->netdev;
5656 pci_set_power_state(pdev, PCI_D0);
5657 pci_restore_state(pdev);
5659 * pci_restore_state clears dev->state_saved so call
5660 * pci_save_state to restore it.
5662 pci_save_state(pdev);
5664 err = pci_enable_device_mem(pdev);
5666 e_dev_err("Cannot enable PCI device from suspend\n");
5669 pci_set_master(pdev);
5671 pci_wake_from_d3(pdev, false);
5673 err = ixgbe_init_interrupt_scheme(adapter);
5675 e_dev_err("Cannot initialize interrupts for device\n");
5679 ixgbe_reset(adapter);
5681 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5683 if (netif_running(netdev)) {
5684 err = ixgbe_open(netdev);
5689 netif_device_attach(netdev);
5693 #endif /* CONFIG_PM */
5695 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5697 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5698 struct net_device *netdev = adapter->netdev;
5699 struct ixgbe_hw *hw = &adapter->hw;
5701 u32 wufc = adapter->wol;
5706 netif_device_detach(netdev);
5708 if (netif_running(netdev)) {
5709 ixgbe_down(adapter);
5710 ixgbe_free_irq(adapter);
5711 ixgbe_free_all_tx_resources(adapter);
5712 ixgbe_free_all_rx_resources(adapter);
5715 ixgbe_clear_interrupt_scheme(adapter);
5717 kfree(adapter->ixgbe_ieee_pfc);
5718 kfree(adapter->ixgbe_ieee_ets);
5722 retval = pci_save_state(pdev);
5728 ixgbe_set_rx_mode(netdev);
5730 /* turn on all-multi mode if wake on multicast is enabled */
5731 if (wufc & IXGBE_WUFC_MC) {
5732 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5733 fctrl |= IXGBE_FCTRL_MPE;
5734 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5737 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5738 ctrl |= IXGBE_CTRL_GIO_DIS;
5739 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5741 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5743 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5744 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5747 switch (hw->mac.type) {
5748 case ixgbe_mac_82598EB:
5749 pci_wake_from_d3(pdev, false);
5751 case ixgbe_mac_82599EB:
5752 case ixgbe_mac_X540:
5753 pci_wake_from_d3(pdev, !!wufc);
5759 *enable_wake = !!wufc;
5761 ixgbe_release_hw_control(adapter);
5763 pci_disable_device(pdev);
5769 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5774 retval = __ixgbe_shutdown(pdev, &wake);
5779 pci_prepare_to_sleep(pdev);
5781 pci_wake_from_d3(pdev, false);
5782 pci_set_power_state(pdev, PCI_D3hot);
5787 #endif /* CONFIG_PM */
5789 static void ixgbe_shutdown(struct pci_dev *pdev)
5793 __ixgbe_shutdown(pdev, &wake);
5795 if (system_state == SYSTEM_POWER_OFF) {
5796 pci_wake_from_d3(pdev, wake);
5797 pci_set_power_state(pdev, PCI_D3hot);
5802 * ixgbe_update_stats - Update the board statistics counters.
5803 * @adapter: board private structure
5805 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5807 struct net_device *netdev = adapter->netdev;
5808 struct ixgbe_hw *hw = &adapter->hw;
5809 struct ixgbe_hw_stats *hwstats = &adapter->stats;
5811 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5812 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
5813 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
5814 u64 bytes = 0, packets = 0;
5816 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5817 test_bit(__IXGBE_RESETTING, &adapter->state))
5820 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5823 for (i = 0; i < 16; i++)
5824 adapter->hw_rx_no_dma_resources +=
5825 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5826 for (i = 0; i < adapter->num_rx_queues; i++) {
5827 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
5828 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
5830 adapter->rsc_total_count = rsc_count;
5831 adapter->rsc_total_flush = rsc_flush;
5834 for (i = 0; i < adapter->num_rx_queues; i++) {
5835 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
5836 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
5837 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
5838 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
5839 bytes += rx_ring->stats.bytes;
5840 packets += rx_ring->stats.packets;
5842 adapter->non_eop_descs = non_eop_descs;
5843 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5844 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5845 netdev->stats.rx_bytes = bytes;
5846 netdev->stats.rx_packets = packets;
5850 /* gather some stats to the adapter struct that are per queue */
5851 for (i = 0; i < adapter->num_tx_queues; i++) {
5852 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5853 restart_queue += tx_ring->tx_stats.restart_queue;
5854 tx_busy += tx_ring->tx_stats.tx_busy;
5855 bytes += tx_ring->stats.bytes;
5856 packets += tx_ring->stats.packets;
5858 adapter->restart_queue = restart_queue;
5859 adapter->tx_busy = tx_busy;
5860 netdev->stats.tx_bytes = bytes;
5861 netdev->stats.tx_packets = packets;
5863 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5864 for (i = 0; i < 8; i++) {
5865 /* for packet buffers not used, the register should read 0 */
5866 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5868 hwstats->mpc[i] += mpc;
5869 total_mpc += hwstats->mpc[i];
5870 if (hw->mac.type == ixgbe_mac_82598EB)
5871 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5872 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5873 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5874 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5875 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5876 switch (hw->mac.type) {
5877 case ixgbe_mac_82598EB:
5878 hwstats->pxonrxc[i] +=
5879 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5881 case ixgbe_mac_82599EB:
5882 case ixgbe_mac_X540:
5883 hwstats->pxonrxc[i] +=
5884 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5889 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5890 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5892 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5893 /* work around hardware counting issue */
5894 hwstats->gprc -= missed_rx;
5896 ixgbe_update_xoff_received(adapter);
5898 /* 82598 hardware only has a 32 bit counter in the high register */
5899 switch (hw->mac.type) {
5900 case ixgbe_mac_82598EB:
5901 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5902 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5903 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5904 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5906 case ixgbe_mac_82599EB:
5907 case ixgbe_mac_X540:
5908 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5909 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
5910 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5911 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
5912 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5913 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5914 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5915 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5916 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5918 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5919 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5920 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5921 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5922 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5923 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5924 #endif /* IXGBE_FCOE */
5929 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5930 hwstats->bprc += bprc;
5931 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5932 if (hw->mac.type == ixgbe_mac_82598EB)
5933 hwstats->mprc -= bprc;
5934 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5935 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5936 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5937 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5938 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5939 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5940 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5941 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5942 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5943 hwstats->lxontxc += lxon;
5944 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5945 hwstats->lxofftxc += lxoff;
5946 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5947 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5948 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5950 * 82598 errata - tx of flow control packets is included in tx counters
5952 xon_off_tot = lxon + lxoff;
5953 hwstats->gptc -= xon_off_tot;
5954 hwstats->mptc -= xon_off_tot;
5955 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5956 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5957 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5958 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5959 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5960 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5961 hwstats->ptc64 -= xon_off_tot;
5962 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5963 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5964 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5965 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5966 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5967 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5969 /* Fill out the OS statistics structure */
5970 netdev->stats.multicast = hwstats->mprc;
5973 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
5974 netdev->stats.rx_dropped = 0;
5975 netdev->stats.rx_length_errors = hwstats->rlec;
5976 netdev->stats.rx_crc_errors = hwstats->crcerrs;
5977 netdev->stats.rx_missed_errors = total_mpc;
5981 * ixgbe_watchdog - Timer Call-back
5982 * @data: pointer to adapter cast into an unsigned long
5984 static void ixgbe_watchdog(unsigned long data)
5986 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5987 struct ixgbe_hw *hw = &adapter->hw;
5992 * Do the watchdog outside of interrupt context due to the lovely
5993 * delays that some of the newer hardware requires
5996 if (test_bit(__IXGBE_DOWN, &adapter->state))
5997 goto watchdog_short_circuit;
5999 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
6001 * for legacy and MSI interrupts don't set any bits
6002 * that are enabled for EIAM, because this operation
6003 * would set *both* EIMS and EICS for any bit in EIAM
6005 IXGBE_WRITE_REG(hw, IXGBE_EICS,
6006 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
6007 goto watchdog_reschedule;
6010 /* get one bit for every active tx/rx interrupt vector */
6011 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
6012 struct ixgbe_q_vector *qv = adapter->q_vector[i];
6013 if (qv->rxr_count || qv->txr_count)
6014 eics |= ((u64)1 << i);
6017 /* Cause software interrupt to ensure rx rings are cleaned */
6018 ixgbe_irq_rearm_queues(adapter, eics);
6020 watchdog_reschedule:
6021 /* Reset the timer */
6022 mod_timer(&adapter->watchdog_timer, round_jiffies(jiffies + 2 * HZ));
6024 watchdog_short_circuit:
6025 schedule_work(&adapter->watchdog_task);
6029 * ixgbe_multispeed_fiber_task - worker thread to configure multispeed fiber
6030 * @work: pointer to work_struct containing our data
6032 static void ixgbe_multispeed_fiber_task(struct work_struct *work)
6034 struct ixgbe_adapter *adapter = container_of(work,
6035 struct ixgbe_adapter,
6036 multispeed_fiber_task);
6037 struct ixgbe_hw *hw = &adapter->hw;
6041 adapter->flags |= IXGBE_FLAG_IN_SFP_LINK_TASK;
6042 autoneg = hw->phy.autoneg_advertised;
6043 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
6044 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
6045 hw->mac.autotry_restart = false;
6046 if (hw->mac.ops.setup_link)
6047 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
6048 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
6049 adapter->flags &= ~IXGBE_FLAG_IN_SFP_LINK_TASK;
6053 * ixgbe_sfp_config_module_task - worker thread to configure a new SFP+ module
6054 * @work: pointer to work_struct containing our data
6056 static void ixgbe_sfp_config_module_task(struct work_struct *work)
6058 struct ixgbe_adapter *adapter = container_of(work,
6059 struct ixgbe_adapter,
6060 sfp_config_module_task);
6061 struct ixgbe_hw *hw = &adapter->hw;
6064 adapter->flags |= IXGBE_FLAG_IN_SFP_MOD_TASK;
6066 /* Time for electrical oscillations to settle down */
6068 err = hw->phy.ops.identify_sfp(hw);
6070 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
6071 e_dev_err("failed to initialize because an unsupported SFP+ "
6072 "module type was detected.\n");
6073 e_dev_err("Reload the driver after installing a supported "
6075 unregister_netdev(adapter->netdev);
6078 if (hw->mac.ops.setup_sfp)
6079 hw->mac.ops.setup_sfp(hw);
6081 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
6082 /* This will also work for DA Twinax connections */
6083 schedule_work(&adapter->multispeed_fiber_task);
6084 adapter->flags &= ~IXGBE_FLAG_IN_SFP_MOD_TASK;
6088 * ixgbe_fdir_reinit_task - worker thread to reinit FDIR filter table
6089 * @work: pointer to work_struct containing our data
6091 static void ixgbe_fdir_reinit_task(struct work_struct *work)
6093 struct ixgbe_adapter *adapter = container_of(work,
6094 struct ixgbe_adapter,
6096 struct ixgbe_hw *hw = &adapter->hw;
6099 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
6100 for (i = 0; i < adapter->num_tx_queues; i++)
6101 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
6102 &(adapter->tx_ring[i]->state));
6104 e_err(probe, "failed to finish FDIR re-initialization, "
6105 "ignored adding FDIR ATR filters\n");
6107 /* Done FDIR Re-initialization, enable transmits */
6108 netif_tx_start_all_queues(adapter->netdev);
6111 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
6115 /* Do not perform spoof check for 82598 */
6116 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
6119 ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
6122 * ssvpc register is cleared on read, if zero then no
6123 * spoofed packets in the last interval.
6128 e_warn(drv, "%d Spoofed packets detected\n", ssvpc);
6131 static DEFINE_MUTEX(ixgbe_watchdog_lock);
6134 * ixgbe_watchdog_task - worker thread to bring link up
6135 * @work: pointer to work_struct containing our data
6137 static void ixgbe_watchdog_task(struct work_struct *work)
6139 struct ixgbe_adapter *adapter = container_of(work,
6140 struct ixgbe_adapter,
6142 struct net_device *netdev = adapter->netdev;
6143 struct ixgbe_hw *hw = &adapter->hw;
6147 struct ixgbe_ring *tx_ring;
6148 int some_tx_pending = 0;
6150 mutex_lock(&ixgbe_watchdog_lock);
6152 link_up = adapter->link_up;
6153 link_speed = adapter->link_speed;
6155 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
6156 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
6159 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6160 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
6161 hw->mac.ops.fc_enable(hw, i);
6163 hw->mac.ops.fc_enable(hw, 0);
6166 hw->mac.ops.fc_enable(hw, 0);
6171 time_after(jiffies, (adapter->link_check_timeout +
6172 IXGBE_TRY_LINK_TIMEOUT))) {
6173 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
6174 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
6176 adapter->link_up = link_up;
6177 adapter->link_speed = link_speed;
6181 if (!netif_carrier_ok(netdev)) {
6182 bool flow_rx, flow_tx;
6184 switch (hw->mac.type) {
6185 case ixgbe_mac_82598EB: {
6186 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
6187 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
6188 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
6189 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
6192 case ixgbe_mac_82599EB:
6193 case ixgbe_mac_X540: {
6194 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
6195 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
6196 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
6197 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
6206 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
6207 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
6209 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
6211 (link_speed == IXGBE_LINK_SPEED_100_FULL ?
6214 ((flow_rx && flow_tx) ? "RX/TX" :
6216 (flow_tx ? "TX" : "None"))));
6218 netif_carrier_on(netdev);
6219 ixgbe_check_vf_rate_limit(adapter);
6221 /* Force detection of hung controller */
6222 for (i = 0; i < adapter->num_tx_queues; i++) {
6223 tx_ring = adapter->tx_ring[i];
6224 set_check_for_tx_hang(tx_ring);
6228 adapter->link_up = false;
6229 adapter->link_speed = 0;
6230 if (netif_carrier_ok(netdev)) {
6231 e_info(drv, "NIC Link is Down\n");
6232 netif_carrier_off(netdev);
6236 if (!netif_carrier_ok(netdev)) {
6237 for (i = 0; i < adapter->num_tx_queues; i++) {
6238 tx_ring = adapter->tx_ring[i];
6239 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
6240 some_tx_pending = 1;
6245 if (some_tx_pending) {
6246 /* We've lost link, so the controller stops DMA,
6247 * but we've got queued Tx work that's never going
6248 * to get done, so reset controller to flush Tx.
6249 * (Do the reset outside of interrupt context).
6251 schedule_work(&adapter->reset_task);
6255 ixgbe_spoof_check(adapter);
6256 ixgbe_update_stats(adapter);
6257 mutex_unlock(&ixgbe_watchdog_lock);
6260 static int ixgbe_tso(struct ixgbe_adapter *adapter,
6261 struct ixgbe_ring *tx_ring, struct sk_buff *skb,
6262 u32 tx_flags, u8 *hdr_len, __be16 protocol)
6264 struct ixgbe_adv_tx_context_desc *context_desc;
6267 struct ixgbe_tx_buffer *tx_buffer_info;
6268 u32 vlan_macip_lens = 0, type_tucmd_mlhl;
6269 u32 mss_l4len_idx, l4len;
6271 if (skb_is_gso(skb)) {
6272 if (skb_header_cloned(skb)) {
6273 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
6277 l4len = tcp_hdrlen(skb);
6280 if (protocol == htons(ETH_P_IP)) {
6281 struct iphdr *iph = ip_hdr(skb);
6284 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
6288 } else if (skb_is_gso_v6(skb)) {
6289 ipv6_hdr(skb)->payload_len = 0;
6290 tcp_hdr(skb)->check =
6291 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
6292 &ipv6_hdr(skb)->daddr,
6296 i = tx_ring->next_to_use;
6298 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6299 context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
6301 /* VLAN MACLEN IPLEN */
6302 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
6304 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
6305 vlan_macip_lens |= ((skb_network_offset(skb)) <<
6306 IXGBE_ADVTXD_MACLEN_SHIFT);
6307 *hdr_len += skb_network_offset(skb);
6309 (skb_transport_header(skb) - skb_network_header(skb));
6311 (skb_transport_header(skb) - skb_network_header(skb));
6312 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
6313 context_desc->seqnum_seed = 0;
6315 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
6316 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
6317 IXGBE_ADVTXD_DTYP_CTXT);
6319 if (protocol == htons(ETH_P_IP))
6320 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
6321 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6322 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
6326 (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
6327 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
6328 /* use index 1 for TSO */
6329 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6330 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
6332 tx_buffer_info->time_stamp = jiffies;
6333 tx_buffer_info->next_to_watch = i;
6336 if (i == tx_ring->count)
6338 tx_ring->next_to_use = i;
6345 static u32 ixgbe_psum(struct ixgbe_adapter *adapter, struct sk_buff *skb,
6351 case cpu_to_be16(ETH_P_IP):
6352 rtn |= IXGBE_ADVTXD_TUCMD_IPV4;
6353 switch (ip_hdr(skb)->protocol) {
6355 rtn |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6358 rtn |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
6362 case cpu_to_be16(ETH_P_IPV6):
6363 /* XXX what about other V6 headers?? */
6364 switch (ipv6_hdr(skb)->nexthdr) {
6366 rtn |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6369 rtn |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
6374 if (unlikely(net_ratelimit()))
6375 e_warn(probe, "partial checksum but proto=%x!\n",
6383 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
6384 struct ixgbe_ring *tx_ring,
6385 struct sk_buff *skb, u32 tx_flags,
6388 struct ixgbe_adv_tx_context_desc *context_desc;
6390 struct ixgbe_tx_buffer *tx_buffer_info;
6391 u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
6393 if (skb->ip_summed == CHECKSUM_PARTIAL ||
6394 (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
6395 i = tx_ring->next_to_use;
6396 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6397 context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
6399 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
6401 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
6402 vlan_macip_lens |= (skb_network_offset(skb) <<
6403 IXGBE_ADVTXD_MACLEN_SHIFT);
6404 if (skb->ip_summed == CHECKSUM_PARTIAL)
6405 vlan_macip_lens |= (skb_transport_header(skb) -
6406 skb_network_header(skb));
6408 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
6409 context_desc->seqnum_seed = 0;
6411 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
6412 IXGBE_ADVTXD_DTYP_CTXT);
6414 if (skb->ip_summed == CHECKSUM_PARTIAL)
6415 type_tucmd_mlhl |= ixgbe_psum(adapter, skb, protocol);
6417 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
6418 /* use index zero for tx checksum offload */
6419 context_desc->mss_l4len_idx = 0;
6421 tx_buffer_info->time_stamp = jiffies;
6422 tx_buffer_info->next_to_watch = i;
6425 if (i == tx_ring->count)
6427 tx_ring->next_to_use = i;
6435 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
6436 struct ixgbe_ring *tx_ring,
6437 struct sk_buff *skb, u32 tx_flags,
6438 unsigned int first, const u8 hdr_len)
6440 struct device *dev = tx_ring->dev;
6441 struct ixgbe_tx_buffer *tx_buffer_info;
6443 unsigned int total = skb->len;
6444 unsigned int offset = 0, size, count = 0, i;
6445 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
6447 unsigned int bytecount = skb->len;
6450 i = tx_ring->next_to_use;
6452 if (tx_flags & IXGBE_TX_FLAGS_FCOE)
6453 /* excluding fcoe_crc_eof for FCoE */
6454 total -= sizeof(struct fcoe_crc_eof);
6456 len = min(skb_headlen(skb), total);
6458 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6459 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
6461 tx_buffer_info->length = size;
6462 tx_buffer_info->mapped_as_page = false;
6463 tx_buffer_info->dma = dma_map_single(dev,
6465 size, DMA_TO_DEVICE);
6466 if (dma_mapping_error(dev, tx_buffer_info->dma))
6468 tx_buffer_info->time_stamp = jiffies;
6469 tx_buffer_info->next_to_watch = i;
6478 if (i == tx_ring->count)
6483 for (f = 0; f < nr_frags; f++) {
6484 struct skb_frag_struct *frag;
6486 frag = &skb_shinfo(skb)->frags[f];
6487 len = min((unsigned int)frag->size, total);
6488 offset = frag->page_offset;
6492 if (i == tx_ring->count)
6495 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6496 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
6498 tx_buffer_info->length = size;
6499 tx_buffer_info->dma = dma_map_page(dev,
6503 tx_buffer_info->mapped_as_page = true;
6504 if (dma_mapping_error(dev, tx_buffer_info->dma))
6506 tx_buffer_info->time_stamp = jiffies;
6507 tx_buffer_info->next_to_watch = i;
6518 if (tx_flags & IXGBE_TX_FLAGS_TSO)
6519 gso_segs = skb_shinfo(skb)->gso_segs;
6521 /* adjust for FCoE Sequence Offload */
6522 else if (tx_flags & IXGBE_TX_FLAGS_FSO)
6523 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
6524 skb_shinfo(skb)->gso_size);
6525 #endif /* IXGBE_FCOE */
6526 bytecount += (gso_segs - 1) * hdr_len;
6528 /* multiply data chunks by size of headers */
6529 tx_ring->tx_buffer_info[i].bytecount = bytecount;
6530 tx_ring->tx_buffer_info[i].gso_segs = gso_segs;
6531 tx_ring->tx_buffer_info[i].skb = skb;
6532 tx_ring->tx_buffer_info[first].next_to_watch = i;
6537 e_dev_err("TX DMA map failed\n");
6539 /* clear timestamp and dma mappings for failed tx_buffer_info map */
6540 tx_buffer_info->dma = 0;
6541 tx_buffer_info->time_stamp = 0;
6542 tx_buffer_info->next_to_watch = 0;
6546 /* clear timestamp and dma mappings for remaining portion of packet */
6549 i += tx_ring->count;
6551 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6552 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
6558 static void ixgbe_tx_queue(struct ixgbe_ring *tx_ring,
6559 int tx_flags, int count, u32 paylen, u8 hdr_len)
6561 union ixgbe_adv_tx_desc *tx_desc = NULL;
6562 struct ixgbe_tx_buffer *tx_buffer_info;
6563 u32 olinfo_status = 0, cmd_type_len = 0;
6565 u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
6567 cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
6569 cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
6571 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
6572 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
6574 if (tx_flags & IXGBE_TX_FLAGS_TSO) {
6575 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
6577 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
6578 IXGBE_ADVTXD_POPTS_SHIFT;
6580 /* use index 1 context for tso */
6581 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6582 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
6583 olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
6584 IXGBE_ADVTXD_POPTS_SHIFT;
6586 } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
6587 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
6588 IXGBE_ADVTXD_POPTS_SHIFT;
6590 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6591 olinfo_status |= IXGBE_ADVTXD_CC;
6592 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6593 if (tx_flags & IXGBE_TX_FLAGS_FSO)
6594 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
6597 olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
6599 i = tx_ring->next_to_use;
6601 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6602 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
6603 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
6604 tx_desc->read.cmd_type_len =
6605 cpu_to_le32(cmd_type_len | tx_buffer_info->length);
6606 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
6608 if (i == tx_ring->count)
6612 tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
6615 * Force memory writes to complete before letting h/w
6616 * know there are new descriptors to fetch. (Only
6617 * applicable for weak-ordered memory model archs,
6622 tx_ring->next_to_use = i;
6623 writel(i, tx_ring->tail);
6626 static void ixgbe_atr(struct ixgbe_ring *ring, struct sk_buff *skb,
6627 u32 tx_flags, __be16 protocol)
6629 struct ixgbe_q_vector *q_vector = ring->q_vector;
6630 union ixgbe_atr_hash_dword input = { .dword = 0 };
6631 union ixgbe_atr_hash_dword common = { .dword = 0 };
6633 unsigned char *network;
6635 struct ipv6hdr *ipv6;
6640 /* if ring doesn't have a interrupt vector, cannot perform ATR */
6644 /* do nothing if sampling is disabled */
6645 if (!ring->atr_sample_rate)
6650 /* snag network header to get L4 type and address */
6651 hdr.network = skb_network_header(skb);
6653 /* Currently only IPv4/IPv6 with TCP is supported */
6654 if ((protocol != __constant_htons(ETH_P_IPV6) ||
6655 hdr.ipv6->nexthdr != IPPROTO_TCP) &&
6656 (protocol != __constant_htons(ETH_P_IP) ||
6657 hdr.ipv4->protocol != IPPROTO_TCP))
6662 /* skip this packet since the socket is closing */
6666 /* sample on all syn packets or once every atr sample count */
6667 if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
6670 /* reset sample count */
6671 ring->atr_count = 0;
6673 vlan_id = htons(tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
6676 * src and dst are inverted, think how the receiver sees them
6678 * The input is broken into two sections, a non-compressed section
6679 * containing vm_pool, vlan_id, and flow_type. The rest of the data
6680 * is XORed together and stored in the compressed dword.
6682 input.formatted.vlan_id = vlan_id;
6685 * since src port and flex bytes occupy the same word XOR them together
6686 * and write the value to source port portion of compressed dword
6689 common.port.src ^= th->dest ^ __constant_htons(ETH_P_8021Q);
6691 common.port.src ^= th->dest ^ protocol;
6692 common.port.dst ^= th->source;
6694 if (protocol == __constant_htons(ETH_P_IP)) {
6695 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
6696 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
6698 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
6699 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
6700 hdr.ipv6->saddr.s6_addr32[1] ^
6701 hdr.ipv6->saddr.s6_addr32[2] ^
6702 hdr.ipv6->saddr.s6_addr32[3] ^
6703 hdr.ipv6->daddr.s6_addr32[0] ^
6704 hdr.ipv6->daddr.s6_addr32[1] ^
6705 hdr.ipv6->daddr.s6_addr32[2] ^
6706 hdr.ipv6->daddr.s6_addr32[3];
6709 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
6710 ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
6711 input, common, ring->queue_index);
6714 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, int size)
6716 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
6717 /* Herbert's original patch had:
6718 * smp_mb__after_netif_stop_queue();
6719 * but since that doesn't exist yet, just open code it. */
6722 /* We need to check again in a case another CPU has just
6723 * made room available. */
6724 if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
6727 /* A reprieve! - use start_queue because it doesn't call schedule */
6728 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
6729 ++tx_ring->tx_stats.restart_queue;
6733 static int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, int size)
6735 if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
6737 return __ixgbe_maybe_stop_tx(tx_ring, size);
6740 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6742 struct ixgbe_adapter *adapter = netdev_priv(dev);
6743 int txq = smp_processor_id();
6747 protocol = vlan_get_protocol(skb);
6749 if (((protocol == htons(ETH_P_FCOE)) ||
6750 (protocol == htons(ETH_P_FIP))) &&
6751 (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)) {
6752 txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
6753 txq += adapter->ring_feature[RING_F_FCOE].mask;
6758 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6759 while (unlikely(txq >= dev->real_num_tx_queues))
6760 txq -= dev->real_num_tx_queues;
6764 return skb_tx_hash(dev, skb);
6767 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
6768 struct ixgbe_adapter *adapter,
6769 struct ixgbe_ring *tx_ring)
6772 unsigned int tx_flags = 0;
6779 protocol = vlan_get_protocol(skb);
6781 if (vlan_tx_tag_present(skb)) {
6782 tx_flags |= vlan_tx_tag_get(skb);
6783 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6784 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6785 tx_flags |= tx_ring->dcb_tc << 13;
6787 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6788 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6789 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED &&
6790 skb->priority != TC_PRIO_CONTROL) {
6791 tx_flags |= tx_ring->dcb_tc << 13;
6792 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6793 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6797 /* for FCoE with DCB, we force the priority to what
6798 * was specified by the switch */
6799 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED &&
6800 (protocol == htons(ETH_P_FCOE)))
6801 tx_flags |= IXGBE_TX_FLAGS_FCOE;
6804 /* four things can cause us to need a context descriptor */
6805 if (skb_is_gso(skb) ||
6806 (skb->ip_summed == CHECKSUM_PARTIAL) ||
6807 (tx_flags & IXGBE_TX_FLAGS_VLAN) ||
6808 (tx_flags & IXGBE_TX_FLAGS_FCOE))
6811 count += TXD_USE_COUNT(skb_headlen(skb));
6812 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6813 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6815 if (ixgbe_maybe_stop_tx(tx_ring, count)) {
6816 tx_ring->tx_stats.tx_busy++;
6817 return NETDEV_TX_BUSY;
6820 first = tx_ring->next_to_use;
6821 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6823 /* setup tx offload for FCoE */
6824 tso = ixgbe_fso(adapter, tx_ring, skb, tx_flags, &hdr_len);
6826 dev_kfree_skb_any(skb);
6827 return NETDEV_TX_OK;
6830 tx_flags |= IXGBE_TX_FLAGS_FSO;
6831 #endif /* IXGBE_FCOE */
6833 if (protocol == htons(ETH_P_IP))
6834 tx_flags |= IXGBE_TX_FLAGS_IPV4;
6835 tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len,
6838 dev_kfree_skb_any(skb);
6839 return NETDEV_TX_OK;
6843 tx_flags |= IXGBE_TX_FLAGS_TSO;
6844 else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags,
6846 (skb->ip_summed == CHECKSUM_PARTIAL))
6847 tx_flags |= IXGBE_TX_FLAGS_CSUM;
6850 count = ixgbe_tx_map(adapter, tx_ring, skb, tx_flags, first, hdr_len);
6852 /* add the ATR filter if ATR is on */
6853 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
6854 ixgbe_atr(tx_ring, skb, tx_flags, protocol);
6855 ixgbe_tx_queue(tx_ring, tx_flags, count, skb->len, hdr_len);
6856 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
6859 dev_kfree_skb_any(skb);
6860 tx_ring->tx_buffer_info[first].time_stamp = 0;
6861 tx_ring->next_to_use = first;
6864 return NETDEV_TX_OK;
6867 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
6869 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6870 struct ixgbe_ring *tx_ring;
6872 tx_ring = adapter->tx_ring[skb->queue_mapping];
6873 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
6877 * ixgbe_set_mac - Change the Ethernet Address of the NIC
6878 * @netdev: network interface device structure
6879 * @p: pointer to an address structure
6881 * Returns 0 on success, negative on failure
6883 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6885 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6886 struct ixgbe_hw *hw = &adapter->hw;
6887 struct sockaddr *addr = p;
6889 if (!is_valid_ether_addr(addr->sa_data))
6890 return -EADDRNOTAVAIL;
6892 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6893 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6895 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
6902 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6904 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6905 struct ixgbe_hw *hw = &adapter->hw;
6909 if (prtad != hw->phy.mdio.prtad)
6911 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6917 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6918 u16 addr, u16 value)
6920 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6921 struct ixgbe_hw *hw = &adapter->hw;
6923 if (prtad != hw->phy.mdio.prtad)
6925 return hw->phy.ops.write_reg(hw, addr, devad, value);
6928 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6930 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6932 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6936 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6938 * @netdev: network interface device structure
6940 * Returns non-zero on failure
6942 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6945 struct ixgbe_adapter *adapter = netdev_priv(dev);
6946 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6948 if (is_valid_ether_addr(mac->san_addr)) {
6950 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6957 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6959 * @netdev: network interface device structure
6961 * Returns non-zero on failure
6963 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6966 struct ixgbe_adapter *adapter = netdev_priv(dev);
6967 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6969 if (is_valid_ether_addr(mac->san_addr)) {
6971 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6977 #ifdef CONFIG_NET_POLL_CONTROLLER
6979 * Polling 'interrupt' - used by things like netconsole to send skbs
6980 * without having to re-enable interrupts. It's not called while
6981 * the interrupt routine is executing.
6983 static void ixgbe_netpoll(struct net_device *netdev)
6985 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6988 /* if interface is down do nothing */
6989 if (test_bit(__IXGBE_DOWN, &adapter->state))
6992 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
6993 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
6994 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
6995 for (i = 0; i < num_q_vectors; i++) {
6996 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
6997 ixgbe_msix_clean_many(0, q_vector);
7000 ixgbe_intr(adapter->pdev->irq, netdev);
7002 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
7006 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
7007 struct rtnl_link_stats64 *stats)
7009 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7013 for (i = 0; i < adapter->num_rx_queues; i++) {
7014 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
7020 start = u64_stats_fetch_begin_bh(&ring->syncp);
7021 packets = ring->stats.packets;
7022 bytes = ring->stats.bytes;
7023 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
7024 stats->rx_packets += packets;
7025 stats->rx_bytes += bytes;
7029 for (i = 0; i < adapter->num_tx_queues; i++) {
7030 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->tx_ring[i]);
7036 start = u64_stats_fetch_begin_bh(&ring->syncp);
7037 packets = ring->stats.packets;
7038 bytes = ring->stats.bytes;
7039 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
7040 stats->tx_packets += packets;
7041 stats->tx_bytes += bytes;
7045 /* following stats updated by ixgbe_watchdog_task() */
7046 stats->multicast = netdev->stats.multicast;
7047 stats->rx_errors = netdev->stats.rx_errors;
7048 stats->rx_length_errors = netdev->stats.rx_length_errors;
7049 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
7050 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
7055 static const struct net_device_ops ixgbe_netdev_ops = {
7056 .ndo_open = ixgbe_open,
7057 .ndo_stop = ixgbe_close,
7058 .ndo_start_xmit = ixgbe_xmit_frame,
7059 .ndo_select_queue = ixgbe_select_queue,
7060 .ndo_set_rx_mode = ixgbe_set_rx_mode,
7061 .ndo_set_multicast_list = ixgbe_set_rx_mode,
7062 .ndo_validate_addr = eth_validate_addr,
7063 .ndo_set_mac_address = ixgbe_set_mac,
7064 .ndo_change_mtu = ixgbe_change_mtu,
7065 .ndo_tx_timeout = ixgbe_tx_timeout,
7066 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
7067 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
7068 .ndo_do_ioctl = ixgbe_ioctl,
7069 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
7070 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
7071 .ndo_set_vf_tx_rate = ixgbe_ndo_set_vf_bw,
7072 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
7073 .ndo_get_stats64 = ixgbe_get_stats64,
7074 #ifdef CONFIG_IXGBE_DCB
7075 .ndo_setup_tc = ixgbe_setup_tc,
7077 #ifdef CONFIG_NET_POLL_CONTROLLER
7078 .ndo_poll_controller = ixgbe_netpoll,
7081 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
7082 .ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
7083 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
7084 .ndo_fcoe_enable = ixgbe_fcoe_enable,
7085 .ndo_fcoe_disable = ixgbe_fcoe_disable,
7086 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
7087 #endif /* IXGBE_FCOE */
7090 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
7091 const struct ixgbe_info *ii)
7093 #ifdef CONFIG_PCI_IOV
7094 struct ixgbe_hw *hw = &adapter->hw;
7097 if (hw->mac.type == ixgbe_mac_82598EB || !max_vfs)
7100 /* The 82599 supports up to 64 VFs per physical function
7101 * but this implementation limits allocation to 63 so that
7102 * basic networking resources are still available to the
7105 adapter->num_vfs = (max_vfs > 63) ? 63 : max_vfs;
7106 adapter->flags |= IXGBE_FLAG_SRIOV_ENABLED;
7107 err = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
7109 e_err(probe, "Failed to enable PCI sriov: %d\n", err);
7112 /* If call to enable VFs succeeded then allocate memory
7113 * for per VF control structures.
7116 kcalloc(adapter->num_vfs,
7117 sizeof(struct vf_data_storage), GFP_KERNEL);
7118 if (adapter->vfinfo) {
7119 /* Now that we're sure SR-IOV is enabled
7120 * and memory allocated set up the mailbox parameters
7122 ixgbe_init_mbx_params_pf(hw);
7123 memcpy(&hw->mbx.ops, ii->mbx_ops,
7124 sizeof(hw->mbx.ops));
7126 /* Disable RSC when in SR-IOV mode */
7127 adapter->flags2 &= ~(IXGBE_FLAG2_RSC_CAPABLE |
7128 IXGBE_FLAG2_RSC_ENABLED);
7133 e_err(probe, "Unable to allocate memory for VF Data Storage - "
7134 "SRIOV disabled\n");
7135 pci_disable_sriov(adapter->pdev);
7138 adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
7139 adapter->num_vfs = 0;
7140 #endif /* CONFIG_PCI_IOV */
7144 * ixgbe_probe - Device Initialization Routine
7145 * @pdev: PCI device information struct
7146 * @ent: entry in ixgbe_pci_tbl
7148 * Returns 0 on success, negative on failure
7150 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
7151 * The OS initialization, configuring of the adapter private structure,
7152 * and a hardware reset occur.
7154 static int __devinit ixgbe_probe(struct pci_dev *pdev,
7155 const struct pci_device_id *ent)
7157 struct net_device *netdev;
7158 struct ixgbe_adapter *adapter = NULL;
7159 struct ixgbe_hw *hw;
7160 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
7161 static int cards_found;
7162 int i, err, pci_using_dac;
7163 u8 part_str[IXGBE_PBANUM_LENGTH];
7164 unsigned int indices = num_possible_cpus();
7170 /* Catch broken hardware that put the wrong VF device ID in
7171 * the PCIe SR-IOV capability.
7173 if (pdev->is_virtfn) {
7174 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
7175 pci_name(pdev), pdev->vendor, pdev->device);
7179 err = pci_enable_device_mem(pdev);
7183 if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
7184 !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
7187 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
7189 err = dma_set_coherent_mask(&pdev->dev,
7193 "No usable DMA configuration, aborting\n");
7200 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
7201 IORESOURCE_MEM), ixgbe_driver_name);
7204 "pci_request_selected_regions failed 0x%x\n", err);
7208 pci_enable_pcie_error_reporting(pdev);
7210 pci_set_master(pdev);
7211 pci_save_state(pdev);
7213 if (ii->mac == ixgbe_mac_82598EB)
7214 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
7216 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
7218 #if defined(CONFIG_DCB)
7219 indices = max_t(unsigned int, indices, IXGBE_MAX_DCB_INDICES);
7220 #elif defined(IXGBE_FCOE)
7221 indices += min_t(unsigned int, num_possible_cpus(),
7222 IXGBE_MAX_FCOE_INDICES);
7224 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
7227 goto err_alloc_etherdev;
7230 SET_NETDEV_DEV(netdev, &pdev->dev);
7232 adapter = netdev_priv(netdev);
7233 pci_set_drvdata(pdev, adapter);
7235 adapter->netdev = netdev;
7236 adapter->pdev = pdev;
7239 adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
7241 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
7242 pci_resource_len(pdev, 0));
7248 for (i = 1; i <= 5; i++) {
7249 if (pci_resource_len(pdev, i) == 0)
7253 netdev->netdev_ops = &ixgbe_netdev_ops;
7254 ixgbe_set_ethtool_ops(netdev);
7255 netdev->watchdog_timeo = 5 * HZ;
7256 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
7258 adapter->bd_number = cards_found;
7261 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
7262 hw->mac.type = ii->mac;
7265 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
7266 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
7267 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
7268 if (!(eec & (1 << 8)))
7269 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
7272 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
7273 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
7274 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
7275 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
7276 hw->phy.mdio.mmds = 0;
7277 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
7278 hw->phy.mdio.dev = netdev;
7279 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
7280 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
7282 /* set up this timer and work struct before calling get_invariants
7283 * which might start the timer
7285 init_timer(&adapter->sfp_timer);
7286 adapter->sfp_timer.function = ixgbe_sfp_timer;
7287 adapter->sfp_timer.data = (unsigned long) adapter;
7289 INIT_WORK(&adapter->sfp_task, ixgbe_sfp_task);
7291 /* multispeed fiber has its own tasklet, called from GPI SDP1 context */
7292 INIT_WORK(&adapter->multispeed_fiber_task, ixgbe_multispeed_fiber_task);
7294 /* a new SFP+ module arrival, called from GPI SDP2 context */
7295 INIT_WORK(&adapter->sfp_config_module_task,
7296 ixgbe_sfp_config_module_task);
7298 ii->get_invariants(hw);
7300 /* setup the private structure */
7301 err = ixgbe_sw_init(adapter);
7305 /* Make it possible the adapter to be woken up via WOL */
7306 switch (adapter->hw.mac.type) {
7307 case ixgbe_mac_82599EB:
7308 case ixgbe_mac_X540:
7309 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7316 * If there is a fan on this device and it has failed log the
7319 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
7320 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
7321 if (esdp & IXGBE_ESDP_SDP1)
7322 e_crit(probe, "Fan has stopped, replace the adapter\n");
7325 /* reset_hw fills in the perm_addr as well */
7326 hw->phy.reset_if_overtemp = true;
7327 err = hw->mac.ops.reset_hw(hw);
7328 hw->phy.reset_if_overtemp = false;
7329 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
7330 hw->mac.type == ixgbe_mac_82598EB) {
7332 * Start a kernel thread to watch for a module to arrive.
7333 * Only do this for 82598, since 82599 will generate
7334 * interrupts on module arrival.
7336 set_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
7337 mod_timer(&adapter->sfp_timer,
7338 round_jiffies(jiffies + (2 * HZ)));
7340 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
7341 e_dev_err("failed to initialize because an unsupported SFP+ "
7342 "module type was detected.\n");
7343 e_dev_err("Reload the driver after installing a supported "
7347 e_dev_err("HW Init failed: %d\n", err);
7351 ixgbe_probe_vf(adapter, ii);
7353 netdev->features = NETIF_F_SG |
7355 NETIF_F_HW_VLAN_TX |
7356 NETIF_F_HW_VLAN_RX |
7357 NETIF_F_HW_VLAN_FILTER;
7359 netdev->features |= NETIF_F_IPV6_CSUM;
7360 netdev->features |= NETIF_F_TSO;
7361 netdev->features |= NETIF_F_TSO6;
7362 netdev->features |= NETIF_F_GRO;
7364 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
7365 netdev->features |= NETIF_F_SCTP_CSUM;
7367 netdev->vlan_features |= NETIF_F_TSO;
7368 netdev->vlan_features |= NETIF_F_TSO6;
7369 netdev->vlan_features |= NETIF_F_IP_CSUM;
7370 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
7371 netdev->vlan_features |= NETIF_F_SG;
7373 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7374 adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
7375 IXGBE_FLAG_DCB_ENABLED);
7377 #ifdef CONFIG_IXGBE_DCB
7378 netdev->dcbnl_ops = &dcbnl_ops;
7382 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7383 if (hw->mac.ops.get_device_caps) {
7384 hw->mac.ops.get_device_caps(hw, &device_caps);
7385 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
7386 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
7389 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7390 netdev->vlan_features |= NETIF_F_FCOE_CRC;
7391 netdev->vlan_features |= NETIF_F_FSO;
7392 netdev->vlan_features |= NETIF_F_FCOE_MTU;
7394 #endif /* IXGBE_FCOE */
7395 if (pci_using_dac) {
7396 netdev->features |= NETIF_F_HIGHDMA;
7397 netdev->vlan_features |= NETIF_F_HIGHDMA;
7400 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
7401 netdev->features |= NETIF_F_LRO;
7403 /* make sure the EEPROM is good */
7404 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
7405 e_dev_err("The EEPROM Checksum Is Not Valid\n");
7410 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
7411 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
7413 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
7414 e_dev_err("invalid MAC address\n");
7419 /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
7420 if (hw->mac.ops.disable_tx_laser &&
7421 ((hw->phy.multispeed_fiber) ||
7422 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
7423 (hw->mac.type == ixgbe_mac_82599EB))))
7424 hw->mac.ops.disable_tx_laser(hw);
7426 init_timer(&adapter->watchdog_timer);
7427 adapter->watchdog_timer.function = ixgbe_watchdog;
7428 adapter->watchdog_timer.data = (unsigned long)adapter;
7430 INIT_WORK(&adapter->reset_task, ixgbe_reset_task);
7431 INIT_WORK(&adapter->watchdog_task, ixgbe_watchdog_task);
7433 err = ixgbe_init_interrupt_scheme(adapter);
7437 switch (pdev->device) {
7438 case IXGBE_DEV_ID_82599_SFP:
7439 /* Only this subdevice supports WOL */
7440 if (pdev->subsystem_device == IXGBE_SUBDEV_ID_82599_SFP)
7441 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
7442 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
7444 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
7445 /* All except this subdevice support WOL */
7446 if (pdev->subsystem_device != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
7447 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
7448 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
7450 case IXGBE_DEV_ID_82599_KX4:
7451 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
7452 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
7458 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
7460 /* pick up the PCI bus settings for reporting later */
7461 hw->mac.ops.get_bus_info(hw);
7463 /* print bus type/speed/width info */
7464 e_dev_info("(PCI Express:%s:%s) %pM\n",
7465 (hw->bus.speed == ixgbe_bus_speed_5000 ? "5.0Gb/s" :
7466 hw->bus.speed == ixgbe_bus_speed_2500 ? "2.5Gb/s" :
7468 (hw->bus.width == ixgbe_bus_width_pcie_x8 ? "Width x8" :
7469 hw->bus.width == ixgbe_bus_width_pcie_x4 ? "Width x4" :
7470 hw->bus.width == ixgbe_bus_width_pcie_x1 ? "Width x1" :
7474 err = ixgbe_read_pba_string_generic(hw, part_str, IXGBE_PBANUM_LENGTH);
7476 strncpy(part_str, "Unknown", IXGBE_PBANUM_LENGTH);
7477 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
7478 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
7479 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
7482 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
7483 hw->mac.type, hw->phy.type, part_str);
7485 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
7486 e_dev_warn("PCI-Express bandwidth available for this card is "
7487 "not sufficient for optimal performance.\n");
7488 e_dev_warn("For optimal performance a x8 PCI-Express slot "
7492 /* save off EEPROM version number */
7493 hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
7495 /* reset the hardware with the new settings */
7496 err = hw->mac.ops.start_hw(hw);
7498 if (err == IXGBE_ERR_EEPROM_VERSION) {
7499 /* We are running on a pre-production device, log a warning */
7500 e_dev_warn("This device is a pre-production adapter/LOM. "
7501 "Please be aware there may be issues associated "
7502 "with your hardware. If you are experiencing "
7503 "problems please contact your Intel or hardware "
7504 "representative who provided you with this "
7507 strcpy(netdev->name, "eth%d");
7508 err = register_netdev(netdev);
7512 /* carrier off reporting is important to ethtool even BEFORE open */
7513 netif_carrier_off(netdev);
7515 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
7516 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
7517 INIT_WORK(&adapter->fdir_reinit_task, ixgbe_fdir_reinit_task);
7519 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
7520 INIT_WORK(&adapter->check_overtemp_task,
7521 ixgbe_check_overtemp_task);
7522 #ifdef CONFIG_IXGBE_DCA
7523 if (dca_add_requester(&pdev->dev) == 0) {
7524 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
7525 ixgbe_setup_dca(adapter);
7528 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7529 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
7530 for (i = 0; i < adapter->num_vfs; i++)
7531 ixgbe_vf_configuration(pdev, (i | 0x10000000));
7534 /* add san mac addr to netdev */
7535 ixgbe_add_sanmac_netdev(netdev);
7537 e_dev_info("Intel(R) 10 Gigabit Network Connection\n");
7542 ixgbe_release_hw_control(adapter);
7543 ixgbe_clear_interrupt_scheme(adapter);
7546 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7547 ixgbe_disable_sriov(adapter);
7548 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
7549 del_timer_sync(&adapter->sfp_timer);
7550 cancel_work_sync(&adapter->sfp_task);
7551 cancel_work_sync(&adapter->multispeed_fiber_task);
7552 cancel_work_sync(&adapter->sfp_config_module_task);
7553 iounmap(hw->hw_addr);
7555 free_netdev(netdev);
7557 pci_release_selected_regions(pdev,
7558 pci_select_bars(pdev, IORESOURCE_MEM));
7561 pci_disable_device(pdev);
7566 * ixgbe_remove - Device Removal Routine
7567 * @pdev: PCI device information struct
7569 * ixgbe_remove is called by the PCI subsystem to alert the driver
7570 * that it should release a PCI device. The could be caused by a
7571 * Hot-Plug event, or because the driver is going to be removed from
7574 static void __devexit ixgbe_remove(struct pci_dev *pdev)
7576 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7577 struct net_device *netdev = adapter->netdev;
7579 set_bit(__IXGBE_DOWN, &adapter->state);
7582 * The timers may be rescheduled, so explicitly disable them
7583 * from being rescheduled.
7585 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
7586 del_timer_sync(&adapter->watchdog_timer);
7587 del_timer_sync(&adapter->sfp_timer);
7589 cancel_work_sync(&adapter->watchdog_task);
7590 cancel_work_sync(&adapter->sfp_task);
7591 cancel_work_sync(&adapter->multispeed_fiber_task);
7592 cancel_work_sync(&adapter->sfp_config_module_task);
7593 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
7594 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
7595 cancel_work_sync(&adapter->fdir_reinit_task);
7596 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
7597 cancel_work_sync(&adapter->check_overtemp_task);
7599 #ifdef CONFIG_IXGBE_DCA
7600 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
7601 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
7602 dca_remove_requester(&pdev->dev);
7603 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
7608 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
7609 ixgbe_cleanup_fcoe(adapter);
7611 #endif /* IXGBE_FCOE */
7613 /* remove the added san mac */
7614 ixgbe_del_sanmac_netdev(netdev);
7616 if (netdev->reg_state == NETREG_REGISTERED)
7617 unregister_netdev(netdev);
7619 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7620 ixgbe_disable_sriov(adapter);
7622 ixgbe_clear_interrupt_scheme(adapter);
7624 ixgbe_release_hw_control(adapter);
7626 iounmap(adapter->hw.hw_addr);
7627 pci_release_selected_regions(pdev, pci_select_bars(pdev,
7630 e_dev_info("complete\n");
7632 free_netdev(netdev);
7634 pci_disable_pcie_error_reporting(pdev);
7636 pci_disable_device(pdev);
7640 * ixgbe_io_error_detected - called when PCI error is detected
7641 * @pdev: Pointer to PCI device
7642 * @state: The current pci connection state
7644 * This function is called after a PCI bus error affecting
7645 * this device has been detected.
7647 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
7648 pci_channel_state_t state)
7650 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7651 struct net_device *netdev = adapter->netdev;
7653 netif_device_detach(netdev);
7655 if (state == pci_channel_io_perm_failure)
7656 return PCI_ERS_RESULT_DISCONNECT;
7658 if (netif_running(netdev))
7659 ixgbe_down(adapter);
7660 pci_disable_device(pdev);
7662 /* Request a slot reset. */
7663 return PCI_ERS_RESULT_NEED_RESET;
7667 * ixgbe_io_slot_reset - called after the pci bus has been reset.
7668 * @pdev: Pointer to PCI device
7670 * Restart the card from scratch, as if from a cold-boot.
7672 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7674 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7675 pci_ers_result_t result;
7678 if (pci_enable_device_mem(pdev)) {
7679 e_err(probe, "Cannot re-enable PCI device after reset.\n");
7680 result = PCI_ERS_RESULT_DISCONNECT;
7682 pci_set_master(pdev);
7683 pci_restore_state(pdev);
7684 pci_save_state(pdev);
7686 pci_wake_from_d3(pdev, false);
7688 ixgbe_reset(adapter);
7689 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7690 result = PCI_ERS_RESULT_RECOVERED;
7693 err = pci_cleanup_aer_uncorrect_error_status(pdev);
7695 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7696 "failed 0x%0x\n", err);
7697 /* non-fatal, continue */
7704 * ixgbe_io_resume - called when traffic can start flowing again.
7705 * @pdev: Pointer to PCI device
7707 * This callback is called when the error recovery driver tells us that
7708 * its OK to resume normal operation.
7710 static void ixgbe_io_resume(struct pci_dev *pdev)
7712 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7713 struct net_device *netdev = adapter->netdev;
7715 if (netif_running(netdev)) {
7716 if (ixgbe_up(adapter)) {
7717 e_info(probe, "ixgbe_up failed after reset\n");
7722 netif_device_attach(netdev);
7725 static struct pci_error_handlers ixgbe_err_handler = {
7726 .error_detected = ixgbe_io_error_detected,
7727 .slot_reset = ixgbe_io_slot_reset,
7728 .resume = ixgbe_io_resume,
7731 static struct pci_driver ixgbe_driver = {
7732 .name = ixgbe_driver_name,
7733 .id_table = ixgbe_pci_tbl,
7734 .probe = ixgbe_probe,
7735 .remove = __devexit_p(ixgbe_remove),
7737 .suspend = ixgbe_suspend,
7738 .resume = ixgbe_resume,
7740 .shutdown = ixgbe_shutdown,
7741 .err_handler = &ixgbe_err_handler
7745 * ixgbe_init_module - Driver Registration Routine
7747 * ixgbe_init_module is the first routine called when the driver is
7748 * loaded. All it does is register with the PCI subsystem.
7750 static int __init ixgbe_init_module(void)
7753 pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
7754 pr_info("%s\n", ixgbe_copyright);
7756 #ifdef CONFIG_IXGBE_DCA
7757 dca_register_notify(&dca_notifier);
7760 ret = pci_register_driver(&ixgbe_driver);
7764 module_init(ixgbe_init_module);
7767 * ixgbe_exit_module - Driver Exit Cleanup Routine
7769 * ixgbe_exit_module is called just before the driver is removed
7772 static void __exit ixgbe_exit_module(void)
7774 #ifdef CONFIG_IXGBE_DCA
7775 dca_unregister_notify(&dca_notifier);
7777 pci_unregister_driver(&ixgbe_driver);
7778 rcu_barrier(); /* Wait for completion of call_rcu()'s */
7781 #ifdef CONFIG_IXGBE_DCA
7782 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
7787 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
7788 __ixgbe_notify_dca);
7790 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7793 #endif /* CONFIG_IXGBE_DCA */
7795 module_exit(ixgbe_exit_module);