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),
128 /* required last entry */
131 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
133 #ifdef CONFIG_IXGBE_DCA
134 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
136 static struct notifier_block dca_notifier = {
137 .notifier_call = ixgbe_notify_dca,
143 #ifdef CONFIG_PCI_IOV
144 static unsigned int max_vfs;
145 module_param(max_vfs, uint, 0);
146 MODULE_PARM_DESC(max_vfs,
147 "Maximum number of virtual functions to allocate per physical function");
148 #endif /* CONFIG_PCI_IOV */
150 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
151 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
152 MODULE_LICENSE("GPL");
153 MODULE_VERSION(DRV_VERSION);
155 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
157 static inline void ixgbe_disable_sriov(struct ixgbe_adapter *adapter)
159 struct ixgbe_hw *hw = &adapter->hw;
164 #ifdef CONFIG_PCI_IOV
165 /* disable iov and allow time for transactions to clear */
166 pci_disable_sriov(adapter->pdev);
169 /* turn off device IOV mode */
170 gcr = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
171 gcr &= ~(IXGBE_GCR_EXT_SRIOV);
172 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr);
173 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
174 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
175 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
177 /* set default pool back to 0 */
178 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
179 vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
180 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
182 /* take a breather then clean up driver data */
185 kfree(adapter->vfinfo);
186 adapter->vfinfo = NULL;
188 adapter->num_vfs = 0;
189 adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
192 struct ixgbe_reg_info {
197 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
199 /* General Registers */
200 {IXGBE_CTRL, "CTRL"},
201 {IXGBE_STATUS, "STATUS"},
202 {IXGBE_CTRL_EXT, "CTRL_EXT"},
204 /* Interrupt Registers */
205 {IXGBE_EICR, "EICR"},
208 {IXGBE_SRRCTL(0), "SRRCTL"},
209 {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
210 {IXGBE_RDLEN(0), "RDLEN"},
211 {IXGBE_RDH(0), "RDH"},
212 {IXGBE_RDT(0), "RDT"},
213 {IXGBE_RXDCTL(0), "RXDCTL"},
214 {IXGBE_RDBAL(0), "RDBAL"},
215 {IXGBE_RDBAH(0), "RDBAH"},
218 {IXGBE_TDBAL(0), "TDBAL"},
219 {IXGBE_TDBAH(0), "TDBAH"},
220 {IXGBE_TDLEN(0), "TDLEN"},
221 {IXGBE_TDH(0), "TDH"},
222 {IXGBE_TDT(0), "TDT"},
223 {IXGBE_TXDCTL(0), "TXDCTL"},
225 /* List Terminator */
231 * ixgbe_regdump - register printout routine
233 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
239 switch (reginfo->ofs) {
240 case IXGBE_SRRCTL(0):
241 for (i = 0; i < 64; i++)
242 regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
244 case IXGBE_DCA_RXCTRL(0):
245 for (i = 0; i < 64; i++)
246 regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
249 for (i = 0; i < 64; i++)
250 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
253 for (i = 0; i < 64; i++)
254 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
257 for (i = 0; i < 64; i++)
258 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
260 case IXGBE_RXDCTL(0):
261 for (i = 0; i < 64; i++)
262 regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
265 for (i = 0; i < 64; i++)
266 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
269 for (i = 0; i < 64; i++)
270 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
273 for (i = 0; i < 64; i++)
274 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
277 for (i = 0; i < 64; i++)
278 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
281 for (i = 0; i < 64; i++)
282 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
285 for (i = 0; i < 64; i++)
286 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
289 for (i = 0; i < 64; i++)
290 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
292 case IXGBE_TXDCTL(0):
293 for (i = 0; i < 64; i++)
294 regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
297 pr_info("%-15s %08x\n", reginfo->name,
298 IXGBE_READ_REG(hw, reginfo->ofs));
302 for (i = 0; i < 8; i++) {
303 snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i*8, i*8+7);
304 pr_err("%-15s", rname);
305 for (j = 0; j < 8; j++)
306 pr_cont(" %08x", regs[i*8+j]);
313 * ixgbe_dump - Print registers, tx-rings and rx-rings
315 static void ixgbe_dump(struct ixgbe_adapter *adapter)
317 struct net_device *netdev = adapter->netdev;
318 struct ixgbe_hw *hw = &adapter->hw;
319 struct ixgbe_reg_info *reginfo;
321 struct ixgbe_ring *tx_ring;
322 struct ixgbe_tx_buffer *tx_buffer_info;
323 union ixgbe_adv_tx_desc *tx_desc;
324 struct my_u0 { u64 a; u64 b; } *u0;
325 struct ixgbe_ring *rx_ring;
326 union ixgbe_adv_rx_desc *rx_desc;
327 struct ixgbe_rx_buffer *rx_buffer_info;
331 if (!netif_msg_hw(adapter))
334 /* Print netdevice Info */
336 dev_info(&adapter->pdev->dev, "Net device Info\n");
337 pr_info("Device Name state "
338 "trans_start last_rx\n");
339 pr_info("%-15s %016lX %016lX %016lX\n",
346 /* Print Registers */
347 dev_info(&adapter->pdev->dev, "Register Dump\n");
348 pr_info(" Register Name Value\n");
349 for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
350 reginfo->name; reginfo++) {
351 ixgbe_regdump(hw, reginfo);
354 /* Print TX Ring Summary */
355 if (!netdev || !netif_running(netdev))
358 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
359 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
360 for (n = 0; n < adapter->num_tx_queues; n++) {
361 tx_ring = adapter->tx_ring[n];
363 &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
364 pr_info(" %5d %5X %5X %016llX %04X %3X %016llX\n",
365 n, tx_ring->next_to_use, tx_ring->next_to_clean,
366 (u64)tx_buffer_info->dma,
367 tx_buffer_info->length,
368 tx_buffer_info->next_to_watch,
369 (u64)tx_buffer_info->time_stamp);
373 if (!netif_msg_tx_done(adapter))
374 goto rx_ring_summary;
376 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
378 /* Transmit Descriptor Formats
380 * Advanced Transmit Descriptor
381 * +--------------------------------------------------------------+
382 * 0 | Buffer Address [63:0] |
383 * +--------------------------------------------------------------+
384 * 8 | PAYLEN | PORTS | IDX | STA | DCMD |DTYP | RSV | DTALEN |
385 * +--------------------------------------------------------------+
386 * 63 46 45 40 39 36 35 32 31 24 23 20 19 0
389 for (n = 0; n < adapter->num_tx_queues; n++) {
390 tx_ring = adapter->tx_ring[n];
391 pr_info("------------------------------------\n");
392 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
393 pr_info("------------------------------------\n");
394 pr_info("T [desc] [address 63:0 ] "
395 "[PlPOIdStDDt Ln] [bi->dma ] "
396 "leng ntw timestamp bi->skb\n");
398 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
399 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
400 tx_buffer_info = &tx_ring->tx_buffer_info[i];
401 u0 = (struct my_u0 *)tx_desc;
402 pr_info("T [0x%03X] %016llX %016llX %016llX"
403 " %04X %3X %016llX %p", i,
406 (u64)tx_buffer_info->dma,
407 tx_buffer_info->length,
408 tx_buffer_info->next_to_watch,
409 (u64)tx_buffer_info->time_stamp,
410 tx_buffer_info->skb);
411 if (i == tx_ring->next_to_use &&
412 i == tx_ring->next_to_clean)
414 else if (i == tx_ring->next_to_use)
416 else if (i == tx_ring->next_to_clean)
421 if (netif_msg_pktdata(adapter) &&
422 tx_buffer_info->dma != 0)
423 print_hex_dump(KERN_INFO, "",
424 DUMP_PREFIX_ADDRESS, 16, 1,
425 phys_to_virt(tx_buffer_info->dma),
426 tx_buffer_info->length, true);
430 /* Print RX Rings Summary */
432 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
433 pr_info("Queue [NTU] [NTC]\n");
434 for (n = 0; n < adapter->num_rx_queues; n++) {
435 rx_ring = adapter->rx_ring[n];
436 pr_info("%5d %5X %5X\n",
437 n, rx_ring->next_to_use, rx_ring->next_to_clean);
441 if (!netif_msg_rx_status(adapter))
444 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
446 /* Advanced Receive Descriptor (Read) Format
448 * +-----------------------------------------------------+
449 * 0 | Packet Buffer Address [63:1] |A0/NSE|
450 * +----------------------------------------------+------+
451 * 8 | Header Buffer Address [63:1] | DD |
452 * +-----------------------------------------------------+
455 * Advanced Receive Descriptor (Write-Back) Format
457 * 63 48 47 32 31 30 21 20 16 15 4 3 0
458 * +------------------------------------------------------+
459 * 0 | Packet IP |SPH| HDR_LEN | RSV|Packet| RSS |
460 * | Checksum Ident | | | | Type | Type |
461 * +------------------------------------------------------+
462 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
463 * +------------------------------------------------------+
464 * 63 48 47 32 31 20 19 0
466 for (n = 0; n < adapter->num_rx_queues; n++) {
467 rx_ring = adapter->rx_ring[n];
468 pr_info("------------------------------------\n");
469 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
470 pr_info("------------------------------------\n");
471 pr_info("R [desc] [ PktBuf A0] "
472 "[ HeadBuf DD] [bi->dma ] [bi->skb] "
473 "<-- Adv Rx Read format\n");
474 pr_info("RWB[desc] [PcsmIpSHl PtRs] "
475 "[vl er S cks ln] ---------------- [bi->skb] "
476 "<-- Adv Rx Write-Back format\n");
478 for (i = 0; i < rx_ring->count; i++) {
479 rx_buffer_info = &rx_ring->rx_buffer_info[i];
480 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
481 u0 = (struct my_u0 *)rx_desc;
482 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
483 if (staterr & IXGBE_RXD_STAT_DD) {
484 /* Descriptor Done */
485 pr_info("RWB[0x%03X] %016llX "
486 "%016llX ---------------- %p", i,
489 rx_buffer_info->skb);
491 pr_info("R [0x%03X] %016llX "
492 "%016llX %016llX %p", i,
495 (u64)rx_buffer_info->dma,
496 rx_buffer_info->skb);
498 if (netif_msg_pktdata(adapter)) {
499 print_hex_dump(KERN_INFO, "",
500 DUMP_PREFIX_ADDRESS, 16, 1,
501 phys_to_virt(rx_buffer_info->dma),
502 rx_ring->rx_buf_len, true);
504 if (rx_ring->rx_buf_len
505 < IXGBE_RXBUFFER_2048)
506 print_hex_dump(KERN_INFO, "",
507 DUMP_PREFIX_ADDRESS, 16, 1,
509 rx_buffer_info->page_dma +
510 rx_buffer_info->page_offset
516 if (i == rx_ring->next_to_use)
518 else if (i == rx_ring->next_to_clean)
530 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
534 /* Let firmware take over control of h/w */
535 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
536 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
537 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
540 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
544 /* Let firmware know the driver has taken over */
545 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
546 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
547 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
551 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
552 * @adapter: pointer to adapter struct
553 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
554 * @queue: queue to map the corresponding interrupt to
555 * @msix_vector: the vector to map to the corresponding queue
558 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
559 u8 queue, u8 msix_vector)
562 struct ixgbe_hw *hw = &adapter->hw;
563 switch (hw->mac.type) {
564 case ixgbe_mac_82598EB:
565 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
568 index = (((direction * 64) + queue) >> 2) & 0x1F;
569 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
570 ivar &= ~(0xFF << (8 * (queue & 0x3)));
571 ivar |= (msix_vector << (8 * (queue & 0x3)));
572 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
574 case ixgbe_mac_82599EB:
576 if (direction == -1) {
578 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
579 index = ((queue & 1) * 8);
580 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
581 ivar &= ~(0xFF << index);
582 ivar |= (msix_vector << index);
583 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
586 /* tx or rx causes */
587 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
588 index = ((16 * (queue & 1)) + (8 * direction));
589 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
590 ivar &= ~(0xFF << index);
591 ivar |= (msix_vector << index);
592 IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
600 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
605 switch (adapter->hw.mac.type) {
606 case ixgbe_mac_82598EB:
607 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
608 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
610 case ixgbe_mac_82599EB:
612 mask = (qmask & 0xFFFFFFFF);
613 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
614 mask = (qmask >> 32);
615 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
622 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_ring *tx_ring,
623 struct ixgbe_tx_buffer *tx_buffer_info)
625 if (tx_buffer_info->dma) {
626 if (tx_buffer_info->mapped_as_page)
627 dma_unmap_page(tx_ring->dev,
629 tx_buffer_info->length,
632 dma_unmap_single(tx_ring->dev,
634 tx_buffer_info->length,
636 tx_buffer_info->dma = 0;
638 if (tx_buffer_info->skb) {
639 dev_kfree_skb_any(tx_buffer_info->skb);
640 tx_buffer_info->skb = NULL;
642 tx_buffer_info->time_stamp = 0;
643 /* tx_buffer_info must be completely set up in the transmit path */
647 * ixgbe_dcb_txq_to_tc - convert a reg index to a traffic class
648 * @adapter: driver private struct
649 * @index: reg idx of queue to query (0-127)
651 * Helper function to determine the traffic index for a particular
654 * Returns : a tc index for use in range 0-7, or 0-3
656 static u8 ixgbe_dcb_txq_to_tc(struct ixgbe_adapter *adapter, u8 reg_idx)
659 int dcb_i = netdev_get_num_tc(adapter->netdev);
661 /* if DCB is not enabled the queues have no TC */
662 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
665 /* check valid range */
666 if (reg_idx >= adapter->hw.mac.max_tx_queues)
669 switch (adapter->hw.mac.type) {
670 case ixgbe_mac_82598EB:
674 if (dcb_i != 4 && dcb_i != 8)
677 /* if VMDq is enabled the lowest order bits determine TC */
678 if (adapter->flags & (IXGBE_FLAG_SRIOV_ENABLED |
679 IXGBE_FLAG_VMDQ_ENABLED)) {
680 tc = reg_idx & (dcb_i - 1);
685 * Convert the reg_idx into the correct TC. This bitmask
686 * targets the last full 32 ring traffic class and assigns
687 * it a value of 1. From there the rest of the rings are
688 * based on shifting the mask further up to include the
689 * reg_idx / 16 and then reg_idx / 8. It assumes dcB_i
690 * will only ever be 8 or 4 and that reg_idx will never
691 * be greater then 128. The code without the power of 2
692 * optimizations would be:
693 * (((reg_idx % 32) + 32) * dcb_i) >> (9 - reg_idx / 32)
695 tc = ((reg_idx & 0X1F) + 0x20) * dcb_i;
696 tc >>= 9 - (reg_idx >> 5);
702 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
704 struct ixgbe_hw *hw = &adapter->hw;
705 struct ixgbe_hw_stats *hwstats = &adapter->stats;
710 if ((hw->fc.current_mode == ixgbe_fc_full) ||
711 (hw->fc.current_mode == ixgbe_fc_rx_pause)) {
712 switch (hw->mac.type) {
713 case ixgbe_mac_82598EB:
714 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
717 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
719 hwstats->lxoffrxc += data;
721 /* refill credits (no tx hang) if we received xoff */
725 for (i = 0; i < adapter->num_tx_queues; i++)
726 clear_bit(__IXGBE_HANG_CHECK_ARMED,
727 &adapter->tx_ring[i]->state);
729 } else if (!(adapter->dcb_cfg.pfc_mode_enable))
732 /* update stats for each tc, only valid with PFC enabled */
733 for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
734 switch (hw->mac.type) {
735 case ixgbe_mac_82598EB:
736 xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
739 xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
741 hwstats->pxoffrxc[i] += xoff[i];
744 /* disarm tx queues that have received xoff frames */
745 for (i = 0; i < adapter->num_tx_queues; i++) {
746 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
747 u32 tc = ixgbe_dcb_txq_to_tc(adapter, tx_ring->reg_idx);
750 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
754 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
756 return ring->tx_stats.completed;
759 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
761 struct ixgbe_adapter *adapter = netdev_priv(ring->netdev);
762 struct ixgbe_hw *hw = &adapter->hw;
764 u32 head = IXGBE_READ_REG(hw, IXGBE_TDH(ring->reg_idx));
765 u32 tail = IXGBE_READ_REG(hw, IXGBE_TDT(ring->reg_idx));
768 return (head < tail) ?
769 tail - head : (tail + ring->count - head);
774 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
776 u32 tx_done = ixgbe_get_tx_completed(tx_ring);
777 u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
778 u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
781 clear_check_for_tx_hang(tx_ring);
784 * Check for a hung queue, but be thorough. This verifies
785 * that a transmit has been completed since the previous
786 * check AND there is at least one packet pending. The
787 * ARMED bit is set to indicate a potential hang. The
788 * bit is cleared if a pause frame is received to remove
789 * false hang detection due to PFC or 802.3x frames. By
790 * requiring this to fail twice we avoid races with
791 * pfc clearing the ARMED bit and conditions where we
792 * run the check_tx_hang logic with a transmit completion
793 * pending but without time to complete it yet.
795 if ((tx_done_old == tx_done) && tx_pending) {
796 /* make sure it is true for two checks in a row */
797 ret = test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
800 /* update completed stats and continue */
801 tx_ring->tx_stats.tx_done_old = tx_done;
802 /* reset the countdown */
803 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
809 #define IXGBE_MAX_TXD_PWR 14
810 #define IXGBE_MAX_DATA_PER_TXD (1 << IXGBE_MAX_TXD_PWR)
812 /* Tx Descriptors needed, worst case */
813 #define TXD_USE_COUNT(S) (((S) >> IXGBE_MAX_TXD_PWR) + \
814 (((S) & (IXGBE_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
815 #define DESC_NEEDED (TXD_USE_COUNT(IXGBE_MAX_DATA_PER_TXD) /* skb->data */ + \
816 MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1) /* for context */
818 static void ixgbe_tx_timeout(struct net_device *netdev);
821 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
822 * @q_vector: structure containing interrupt and ring information
823 * @tx_ring: tx ring to clean
825 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
826 struct ixgbe_ring *tx_ring)
828 struct ixgbe_adapter *adapter = q_vector->adapter;
829 union ixgbe_adv_tx_desc *tx_desc, *eop_desc;
830 struct ixgbe_tx_buffer *tx_buffer_info;
831 unsigned int total_bytes = 0, total_packets = 0;
832 u16 i, eop, count = 0;
834 i = tx_ring->next_to_clean;
835 eop = tx_ring->tx_buffer_info[i].next_to_watch;
836 eop_desc = IXGBE_TX_DESC_ADV(tx_ring, eop);
838 while ((eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)) &&
839 (count < tx_ring->work_limit)) {
840 bool cleaned = false;
841 rmb(); /* read buffer_info after eop_desc */
842 for ( ; !cleaned; count++) {
843 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
844 tx_buffer_info = &tx_ring->tx_buffer_info[i];
846 tx_desc->wb.status = 0;
847 cleaned = (i == eop);
850 if (i == tx_ring->count)
853 if (cleaned && tx_buffer_info->skb) {
854 total_bytes += tx_buffer_info->bytecount;
855 total_packets += tx_buffer_info->gso_segs;
858 ixgbe_unmap_and_free_tx_resource(tx_ring,
862 tx_ring->tx_stats.completed++;
863 eop = tx_ring->tx_buffer_info[i].next_to_watch;
864 eop_desc = IXGBE_TX_DESC_ADV(tx_ring, eop);
867 tx_ring->next_to_clean = i;
868 tx_ring->total_bytes += total_bytes;
869 tx_ring->total_packets += total_packets;
870 u64_stats_update_begin(&tx_ring->syncp);
871 tx_ring->stats.packets += total_packets;
872 tx_ring->stats.bytes += total_bytes;
873 u64_stats_update_end(&tx_ring->syncp);
875 if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
876 /* schedule immediate reset if we believe we hung */
877 struct ixgbe_hw *hw = &adapter->hw;
878 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, eop);
879 e_err(drv, "Detected Tx Unit Hang\n"
881 " TDH, TDT <%x>, <%x>\n"
882 " next_to_use <%x>\n"
883 " next_to_clean <%x>\n"
884 "tx_buffer_info[next_to_clean]\n"
885 " time_stamp <%lx>\n"
887 tx_ring->queue_index,
888 IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
889 IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
890 tx_ring->next_to_use, eop,
891 tx_ring->tx_buffer_info[eop].time_stamp, jiffies);
893 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
896 "tx hang %d detected on queue %d, resetting adapter\n",
897 adapter->tx_timeout_count + 1, tx_ring->queue_index);
899 /* schedule immediate reset if we believe we hung */
900 ixgbe_tx_timeout(adapter->netdev);
902 /* the adapter is about to reset, no point in enabling stuff */
906 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
907 if (unlikely(count && netif_carrier_ok(tx_ring->netdev) &&
908 (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
909 /* Make sure that anybody stopping the queue after this
910 * sees the new next_to_clean.
913 if (__netif_subqueue_stopped(tx_ring->netdev, tx_ring->queue_index) &&
914 !test_bit(__IXGBE_DOWN, &adapter->state)) {
915 netif_wake_subqueue(tx_ring->netdev, tx_ring->queue_index);
916 ++tx_ring->tx_stats.restart_queue;
920 return count < tx_ring->work_limit;
923 #ifdef CONFIG_IXGBE_DCA
924 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
925 struct ixgbe_ring *rx_ring,
928 struct ixgbe_hw *hw = &adapter->hw;
930 u8 reg_idx = rx_ring->reg_idx;
932 rxctrl = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(reg_idx));
933 switch (hw->mac.type) {
934 case ixgbe_mac_82598EB:
935 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
936 rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
938 case ixgbe_mac_82599EB:
940 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK_82599;
941 rxctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
942 IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599);
947 rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
948 rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
949 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_RRO_EN);
950 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_WRO_EN |
951 IXGBE_DCA_RXCTRL_DESC_HSRO_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 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
970 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(reg_idx), txctrl);
972 case ixgbe_mac_82599EB:
974 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(reg_idx));
975 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK_82599;
976 txctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
977 IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599);
978 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
979 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
980 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(reg_idx), txctrl);
987 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
989 struct ixgbe_adapter *adapter = q_vector->adapter;
994 if (q_vector->cpu == cpu)
997 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
998 for (i = 0; i < q_vector->txr_count; i++) {
999 ixgbe_update_tx_dca(adapter, adapter->tx_ring[r_idx], cpu);
1000 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1004 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1005 for (i = 0; i < q_vector->rxr_count; i++) {
1006 ixgbe_update_rx_dca(adapter, adapter->rx_ring[r_idx], cpu);
1007 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1011 q_vector->cpu = cpu;
1016 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
1021 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
1024 /* always use CB2 mode, difference is masked in the CB driver */
1025 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
1027 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
1028 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1032 for (i = 0; i < num_q_vectors; i++) {
1033 adapter->q_vector[i]->cpu = -1;
1034 ixgbe_update_dca(adapter->q_vector[i]);
1038 static int __ixgbe_notify_dca(struct device *dev, void *data)
1040 struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1041 unsigned long event = *(unsigned long *)data;
1043 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
1047 case DCA_PROVIDER_ADD:
1048 /* if we're already enabled, don't do it again */
1049 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1051 if (dca_add_requester(dev) == 0) {
1052 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1053 ixgbe_setup_dca(adapter);
1056 /* Fall Through since DCA is disabled. */
1057 case DCA_PROVIDER_REMOVE:
1058 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1059 dca_remove_requester(dev);
1060 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1061 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
1069 #endif /* CONFIG_IXGBE_DCA */
1071 * ixgbe_receive_skb - Send a completed packet up the stack
1072 * @adapter: board private structure
1073 * @skb: packet to send up
1074 * @status: hardware indication of status of receive
1075 * @rx_ring: rx descriptor ring (for a specific queue) to setup
1076 * @rx_desc: rx descriptor
1078 static void ixgbe_receive_skb(struct ixgbe_q_vector *q_vector,
1079 struct sk_buff *skb, u8 status,
1080 struct ixgbe_ring *ring,
1081 union ixgbe_adv_rx_desc *rx_desc)
1083 struct ixgbe_adapter *adapter = q_vector->adapter;
1084 struct napi_struct *napi = &q_vector->napi;
1085 bool is_vlan = (status & IXGBE_RXD_STAT_VP);
1086 u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
1088 if (is_vlan && (tag & VLAN_VID_MASK))
1089 __vlan_hwaccel_put_tag(skb, tag);
1091 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
1092 napi_gro_receive(napi, skb);
1098 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1099 * @adapter: address of board private structure
1100 * @status_err: hardware indication of status of receive
1101 * @skb: skb currently being received and modified
1103 static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
1104 union ixgbe_adv_rx_desc *rx_desc,
1105 struct sk_buff *skb)
1107 u32 status_err = le32_to_cpu(rx_desc->wb.upper.status_error);
1109 skb_checksum_none_assert(skb);
1111 /* Rx csum disabled */
1112 if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
1115 /* if IP and error */
1116 if ((status_err & IXGBE_RXD_STAT_IPCS) &&
1117 (status_err & IXGBE_RXDADV_ERR_IPE)) {
1118 adapter->hw_csum_rx_error++;
1122 if (!(status_err & IXGBE_RXD_STAT_L4CS))
1125 if (status_err & IXGBE_RXDADV_ERR_TCPE) {
1126 u16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1129 * 82599 errata, UDP frames with a 0 checksum can be marked as
1132 if ((pkt_info & IXGBE_RXDADV_PKTTYPE_UDP) &&
1133 (adapter->hw.mac.type == ixgbe_mac_82599EB))
1136 adapter->hw_csum_rx_error++;
1140 /* It must be a TCP or UDP packet with a valid checksum */
1141 skb->ip_summed = CHECKSUM_UNNECESSARY;
1144 static inline void ixgbe_release_rx_desc(struct ixgbe_ring *rx_ring, u32 val)
1147 * Force memory writes to complete before letting h/w
1148 * know there are new descriptors to fetch. (Only
1149 * applicable for weak-ordered memory model archs,
1153 writel(val, rx_ring->tail);
1157 * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
1158 * @rx_ring: ring to place buffers on
1159 * @cleaned_count: number of buffers to replace
1161 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1163 union ixgbe_adv_rx_desc *rx_desc;
1164 struct ixgbe_rx_buffer *bi;
1165 struct sk_buff *skb;
1166 u16 i = rx_ring->next_to_use;
1168 /* do nothing if no valid netdev defined */
1169 if (!rx_ring->netdev)
1172 while (cleaned_count--) {
1173 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1174 bi = &rx_ring->rx_buffer_info[i];
1178 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1179 rx_ring->rx_buf_len);
1181 rx_ring->rx_stats.alloc_rx_buff_failed++;
1184 /* initialize queue mapping */
1185 skb_record_rx_queue(skb, rx_ring->queue_index);
1190 bi->dma = dma_map_single(rx_ring->dev,
1192 rx_ring->rx_buf_len,
1194 if (dma_mapping_error(rx_ring->dev, bi->dma)) {
1195 rx_ring->rx_stats.alloc_rx_buff_failed++;
1201 if (ring_is_ps_enabled(rx_ring)) {
1203 bi->page = netdev_alloc_page(rx_ring->netdev);
1205 rx_ring->rx_stats.alloc_rx_page_failed++;
1210 if (!bi->page_dma) {
1211 /* use a half page if we're re-using */
1212 bi->page_offset ^= PAGE_SIZE / 2;
1213 bi->page_dma = dma_map_page(rx_ring->dev,
1218 if (dma_mapping_error(rx_ring->dev,
1220 rx_ring->rx_stats.alloc_rx_page_failed++;
1226 /* Refresh the desc even if buffer_addrs didn't change
1227 * because each write-back erases this info. */
1228 rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
1229 rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
1231 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
1232 rx_desc->read.hdr_addr = 0;
1236 if (i == rx_ring->count)
1241 if (rx_ring->next_to_use != i) {
1242 rx_ring->next_to_use = i;
1243 ixgbe_release_rx_desc(rx_ring, i);
1247 static inline u16 ixgbe_get_hlen(union ixgbe_adv_rx_desc *rx_desc)
1249 /* HW will not DMA in data larger than the given buffer, even if it
1250 * parses the (NFS, of course) header to be larger. In that case, it
1251 * fills the header buffer and spills the rest into the page.
1253 u16 hdr_info = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info);
1254 u16 hlen = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
1255 IXGBE_RXDADV_HDRBUFLEN_SHIFT;
1256 if (hlen > IXGBE_RX_HDR_SIZE)
1257 hlen = IXGBE_RX_HDR_SIZE;
1262 * ixgbe_transform_rsc_queue - change rsc queue into a full packet
1263 * @skb: pointer to the last skb in the rsc queue
1265 * This function changes a queue full of hw rsc buffers into a completed
1266 * packet. It uses the ->prev pointers to find the first packet and then
1267 * turns it into the frag list owner.
1269 static inline struct sk_buff *ixgbe_transform_rsc_queue(struct sk_buff *skb)
1271 unsigned int frag_list_size = 0;
1272 unsigned int skb_cnt = 1;
1275 struct sk_buff *prev = skb->prev;
1276 frag_list_size += skb->len;
1282 skb_shinfo(skb)->frag_list = skb->next;
1284 skb->len += frag_list_size;
1285 skb->data_len += frag_list_size;
1286 skb->truesize += frag_list_size;
1287 IXGBE_RSC_CB(skb)->skb_cnt = skb_cnt;
1292 static inline bool ixgbe_get_rsc_state(union ixgbe_adv_rx_desc *rx_desc)
1294 return !!(le32_to_cpu(rx_desc->wb.lower.lo_dword.data) &
1295 IXGBE_RXDADV_RSCCNT_MASK);
1298 static void ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1299 struct ixgbe_ring *rx_ring,
1300 int *work_done, int work_to_do)
1302 struct ixgbe_adapter *adapter = q_vector->adapter;
1303 union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
1304 struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
1305 struct sk_buff *skb;
1306 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1307 const int current_node = numa_node_id();
1310 #endif /* IXGBE_FCOE */
1313 u16 cleaned_count = 0;
1314 bool pkt_is_rsc = false;
1316 i = rx_ring->next_to_clean;
1317 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1318 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1320 while (staterr & IXGBE_RXD_STAT_DD) {
1323 rmb(); /* read descriptor and rx_buffer_info after status DD */
1325 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1327 skb = rx_buffer_info->skb;
1328 rx_buffer_info->skb = NULL;
1329 prefetch(skb->data);
1331 if (ring_is_rsc_enabled(rx_ring))
1332 pkt_is_rsc = ixgbe_get_rsc_state(rx_desc);
1334 /* if this is a skb from previous receive DMA will be 0 */
1335 if (rx_buffer_info->dma) {
1338 !(staterr & IXGBE_RXD_STAT_EOP) &&
1341 * When HWRSC is enabled, delay unmapping
1342 * of the first packet. It carries the
1343 * header information, HW may still
1344 * access the header after the writeback.
1345 * Only unmap it when EOP is reached
1347 IXGBE_RSC_CB(skb)->delay_unmap = true;
1348 IXGBE_RSC_CB(skb)->dma = rx_buffer_info->dma;
1350 dma_unmap_single(rx_ring->dev,
1351 rx_buffer_info->dma,
1352 rx_ring->rx_buf_len,
1355 rx_buffer_info->dma = 0;
1357 if (ring_is_ps_enabled(rx_ring)) {
1358 hlen = ixgbe_get_hlen(rx_desc);
1359 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1361 hlen = le16_to_cpu(rx_desc->wb.upper.length);
1366 /* assume packet split since header is unmapped */
1367 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1371 dma_unmap_page(rx_ring->dev,
1372 rx_buffer_info->page_dma,
1375 rx_buffer_info->page_dma = 0;
1376 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1377 rx_buffer_info->page,
1378 rx_buffer_info->page_offset,
1381 if ((page_count(rx_buffer_info->page) == 1) &&
1382 (page_to_nid(rx_buffer_info->page) == current_node))
1383 get_page(rx_buffer_info->page);
1385 rx_buffer_info->page = NULL;
1387 skb->len += upper_len;
1388 skb->data_len += upper_len;
1389 skb->truesize += upper_len;
1393 if (i == rx_ring->count)
1396 next_rxd = IXGBE_RX_DESC_ADV(rx_ring, i);
1401 u32 nextp = (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
1402 IXGBE_RXDADV_NEXTP_SHIFT;
1403 next_buffer = &rx_ring->rx_buffer_info[nextp];
1405 next_buffer = &rx_ring->rx_buffer_info[i];
1408 if (!(staterr & IXGBE_RXD_STAT_EOP)) {
1409 if (ring_is_ps_enabled(rx_ring)) {
1410 rx_buffer_info->skb = next_buffer->skb;
1411 rx_buffer_info->dma = next_buffer->dma;
1412 next_buffer->skb = skb;
1413 next_buffer->dma = 0;
1415 skb->next = next_buffer->skb;
1416 skb->next->prev = skb;
1418 rx_ring->rx_stats.non_eop_descs++;
1423 skb = ixgbe_transform_rsc_queue(skb);
1424 /* if we got here without RSC the packet is invalid */
1426 __pskb_trim(skb, 0);
1427 rx_buffer_info->skb = skb;
1432 if (ring_is_rsc_enabled(rx_ring)) {
1433 if (IXGBE_RSC_CB(skb)->delay_unmap) {
1434 dma_unmap_single(rx_ring->dev,
1435 IXGBE_RSC_CB(skb)->dma,
1436 rx_ring->rx_buf_len,
1438 IXGBE_RSC_CB(skb)->dma = 0;
1439 IXGBE_RSC_CB(skb)->delay_unmap = false;
1443 if (ring_is_ps_enabled(rx_ring))
1444 rx_ring->rx_stats.rsc_count +=
1445 skb_shinfo(skb)->nr_frags;
1447 rx_ring->rx_stats.rsc_count +=
1448 IXGBE_RSC_CB(skb)->skb_cnt;
1449 rx_ring->rx_stats.rsc_flush++;
1452 /* ERR_MASK will only have valid bits if EOP set */
1453 if (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) {
1454 /* trim packet back to size 0 and recycle it */
1455 __pskb_trim(skb, 0);
1456 rx_buffer_info->skb = skb;
1460 ixgbe_rx_checksum(adapter, rx_desc, skb);
1462 /* probably a little skewed due to removing CRC */
1463 total_rx_bytes += skb->len;
1466 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1468 /* if ddp, not passing to ULD unless for FCP_RSP or error */
1469 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
1470 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
1474 #endif /* IXGBE_FCOE */
1475 ixgbe_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
1478 rx_desc->wb.upper.status_error = 0;
1481 if (*work_done >= work_to_do)
1484 /* return some buffers to hardware, one at a time is too slow */
1485 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
1486 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1490 /* use prefetched values */
1492 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1495 rx_ring->next_to_clean = i;
1496 cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
1499 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1502 /* include DDPed FCoE data */
1503 if (ddp_bytes > 0) {
1506 mss = rx_ring->netdev->mtu - sizeof(struct fcoe_hdr) -
1507 sizeof(struct fc_frame_header) -
1508 sizeof(struct fcoe_crc_eof);
1511 total_rx_bytes += ddp_bytes;
1512 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
1514 #endif /* IXGBE_FCOE */
1516 rx_ring->total_packets += total_rx_packets;
1517 rx_ring->total_bytes += total_rx_bytes;
1518 u64_stats_update_begin(&rx_ring->syncp);
1519 rx_ring->stats.packets += total_rx_packets;
1520 rx_ring->stats.bytes += total_rx_bytes;
1521 u64_stats_update_end(&rx_ring->syncp);
1524 static int ixgbe_clean_rxonly(struct napi_struct *, int);
1526 * ixgbe_configure_msix - Configure MSI-X hardware
1527 * @adapter: board private structure
1529 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1532 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1534 struct ixgbe_q_vector *q_vector;
1535 int i, q_vectors, v_idx, r_idx;
1538 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1541 * Populate the IVAR table and set the ITR values to the
1542 * corresponding register.
1544 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1545 q_vector = adapter->q_vector[v_idx];
1546 /* XXX for_each_set_bit(...) */
1547 r_idx = find_first_bit(q_vector->rxr_idx,
1548 adapter->num_rx_queues);
1550 for (i = 0; i < q_vector->rxr_count; i++) {
1551 u8 reg_idx = adapter->rx_ring[r_idx]->reg_idx;
1552 ixgbe_set_ivar(adapter, 0, reg_idx, v_idx);
1553 r_idx = find_next_bit(q_vector->rxr_idx,
1554 adapter->num_rx_queues,
1557 r_idx = find_first_bit(q_vector->txr_idx,
1558 adapter->num_tx_queues);
1560 for (i = 0; i < q_vector->txr_count; i++) {
1561 u8 reg_idx = adapter->tx_ring[r_idx]->reg_idx;
1562 ixgbe_set_ivar(adapter, 1, reg_idx, v_idx);
1563 r_idx = find_next_bit(q_vector->txr_idx,
1564 adapter->num_tx_queues,
1568 if (q_vector->txr_count && !q_vector->rxr_count)
1570 q_vector->eitr = adapter->tx_eitr_param;
1571 else if (q_vector->rxr_count)
1573 q_vector->eitr = adapter->rx_eitr_param;
1575 ixgbe_write_eitr(q_vector);
1576 /* If Flow Director is enabled, set interrupt affinity */
1577 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
1578 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)) {
1580 * Allocate the affinity_hint cpumask, assign the mask
1581 * for this vector, and set our affinity_hint for
1584 if (!alloc_cpumask_var(&q_vector->affinity_mask,
1587 cpumask_set_cpu(v_idx, q_vector->affinity_mask);
1588 irq_set_affinity_hint(adapter->msix_entries[v_idx].vector,
1589 q_vector->affinity_mask);
1593 switch (adapter->hw.mac.type) {
1594 case ixgbe_mac_82598EB:
1595 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1598 case ixgbe_mac_82599EB:
1599 case ixgbe_mac_X540:
1600 ixgbe_set_ivar(adapter, -1, 1, v_idx);
1606 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1608 /* set up to autoclear timer, and the vectors */
1609 mask = IXGBE_EIMS_ENABLE_MASK;
1610 if (adapter->num_vfs)
1611 mask &= ~(IXGBE_EIMS_OTHER |
1612 IXGBE_EIMS_MAILBOX |
1615 mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
1616 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1619 enum latency_range {
1623 latency_invalid = 255
1627 * ixgbe_update_itr - update the dynamic ITR value based on statistics
1628 * @adapter: pointer to adapter
1629 * @eitr: eitr setting (ints per sec) to give last timeslice
1630 * @itr_setting: current throttle rate in ints/second
1631 * @packets: the number of packets during this measurement interval
1632 * @bytes: the number of bytes during this measurement interval
1634 * Stores a new ITR value based on packets and byte
1635 * counts during the last interrupt. The advantage of per interrupt
1636 * computation is faster updates and more accurate ITR for the current
1637 * traffic pattern. Constants in this function were computed
1638 * based on theoretical maximum wire speed and thresholds were set based
1639 * on testing data as well as attempting to minimize response time
1640 * while increasing bulk throughput.
1641 * this functionality is controlled by the InterruptThrottleRate module
1642 * parameter (see ixgbe_param.c)
1644 static u8 ixgbe_update_itr(struct ixgbe_adapter *adapter,
1645 u32 eitr, u8 itr_setting,
1646 int packets, int bytes)
1648 unsigned int retval = itr_setting;
1653 goto update_itr_done;
1656 /* simple throttlerate management
1657 * 0-20MB/s lowest (100000 ints/s)
1658 * 20-100MB/s low (20000 ints/s)
1659 * 100-1249MB/s bulk (8000 ints/s)
1661 /* what was last interrupt timeslice? */
1662 timepassed_us = 1000000/eitr;
1663 bytes_perint = bytes / timepassed_us; /* bytes/usec */
1665 switch (itr_setting) {
1666 case lowest_latency:
1667 if (bytes_perint > adapter->eitr_low)
1668 retval = low_latency;
1671 if (bytes_perint > adapter->eitr_high)
1672 retval = bulk_latency;
1673 else if (bytes_perint <= adapter->eitr_low)
1674 retval = lowest_latency;
1677 if (bytes_perint <= adapter->eitr_high)
1678 retval = low_latency;
1687 * ixgbe_write_eitr - write EITR register in hardware specific way
1688 * @q_vector: structure containing interrupt and ring information
1690 * This function is made to be called by ethtool and by the driver
1691 * when it needs to update EITR registers at runtime. Hardware
1692 * specific quirks/differences are taken care of here.
1694 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1696 struct ixgbe_adapter *adapter = q_vector->adapter;
1697 struct ixgbe_hw *hw = &adapter->hw;
1698 int v_idx = q_vector->v_idx;
1699 u32 itr_reg = EITR_INTS_PER_SEC_TO_REG(q_vector->eitr);
1701 switch (adapter->hw.mac.type) {
1702 case ixgbe_mac_82598EB:
1703 /* must write high and low 16 bits to reset counter */
1704 itr_reg |= (itr_reg << 16);
1706 case ixgbe_mac_82599EB:
1707 case ixgbe_mac_X540:
1709 * 82599 and X540 can support a value of zero, so allow it for
1710 * max interrupt rate, but there is an errata where it can
1711 * not be zero with RSC
1714 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
1718 * set the WDIS bit to not clear the timer bits and cause an
1719 * immediate assertion of the interrupt
1721 itr_reg |= IXGBE_EITR_CNT_WDIS;
1726 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1729 static void ixgbe_set_itr_msix(struct ixgbe_q_vector *q_vector)
1731 struct ixgbe_adapter *adapter = q_vector->adapter;
1734 u8 current_itr, ret_itr;
1736 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1737 for (i = 0; i < q_vector->txr_count; i++) {
1738 struct ixgbe_ring *tx_ring = adapter->tx_ring[r_idx];
1739 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1741 tx_ring->total_packets,
1742 tx_ring->total_bytes);
1743 /* if the result for this queue would decrease interrupt
1744 * rate for this vector then use that result */
1745 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
1746 q_vector->tx_itr - 1 : ret_itr);
1747 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1751 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1752 for (i = 0; i < q_vector->rxr_count; i++) {
1753 struct ixgbe_ring *rx_ring = adapter->rx_ring[r_idx];
1754 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1756 rx_ring->total_packets,
1757 rx_ring->total_bytes);
1758 /* if the result for this queue would decrease interrupt
1759 * rate for this vector then use that result */
1760 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
1761 q_vector->rx_itr - 1 : ret_itr);
1762 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1766 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1768 switch (current_itr) {
1769 /* counts and packets in update_itr are dependent on these numbers */
1770 case lowest_latency:
1774 new_itr = 20000; /* aka hwitr = ~200 */
1782 if (new_itr != q_vector->eitr) {
1783 /* do an exponential smoothing */
1784 new_itr = ((q_vector->eitr * 9) + new_itr)/10;
1786 /* save the algorithm value here, not the smoothed one */
1787 q_vector->eitr = new_itr;
1789 ixgbe_write_eitr(q_vector);
1794 * ixgbe_check_overtemp_task - worker thread to check over tempurature
1795 * @work: pointer to work_struct containing our data
1797 static void ixgbe_check_overtemp_task(struct work_struct *work)
1799 struct ixgbe_adapter *adapter = container_of(work,
1800 struct ixgbe_adapter,
1801 check_overtemp_task);
1802 struct ixgbe_hw *hw = &adapter->hw;
1803 u32 eicr = adapter->interrupt_event;
1805 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
1808 switch (hw->device_id) {
1809 case IXGBE_DEV_ID_82599_T3_LOM: {
1811 bool link_up = false;
1813 if (hw->mac.ops.check_link)
1814 hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
1816 if (((eicr & IXGBE_EICR_GPI_SDP0) && (!link_up)) ||
1817 (eicr & IXGBE_EICR_LSC))
1818 /* Check if this is due to overtemp */
1819 if (hw->phy.ops.check_overtemp(hw) == IXGBE_ERR_OVERTEMP)
1824 if (!(eicr & IXGBE_EICR_GPI_SDP0))
1829 "Network adapter has been stopped because it has over heated. "
1830 "Restart the computer. If the problem persists, "
1831 "power off the system and replace the adapter\n");
1832 /* write to clear the interrupt */
1833 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP0);
1836 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
1838 struct ixgbe_hw *hw = &adapter->hw;
1840 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
1841 (eicr & IXGBE_EICR_GPI_SDP1)) {
1842 e_crit(probe, "Fan has stopped, replace the adapter\n");
1843 /* write to clear the interrupt */
1844 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1848 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
1850 struct ixgbe_hw *hw = &adapter->hw;
1852 if (eicr & IXGBE_EICR_GPI_SDP2) {
1853 /* Clear the interrupt */
1854 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
1855 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1856 schedule_work(&adapter->sfp_config_module_task);
1859 if (eicr & IXGBE_EICR_GPI_SDP1) {
1860 /* Clear the interrupt */
1861 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1862 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1863 schedule_work(&adapter->multispeed_fiber_task);
1867 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1869 struct ixgbe_hw *hw = &adapter->hw;
1872 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1873 adapter->link_check_timeout = jiffies;
1874 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1875 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1876 IXGBE_WRITE_FLUSH(hw);
1877 schedule_work(&adapter->watchdog_task);
1881 static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
1883 struct net_device *netdev = data;
1884 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1885 struct ixgbe_hw *hw = &adapter->hw;
1889 * Workaround for Silicon errata. Use clear-by-write instead
1890 * of clear-by-read. Reading with EICS will return the
1891 * interrupt causes without clearing, which later be done
1892 * with the write to EICR.
1894 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1895 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1897 if (eicr & IXGBE_EICR_LSC)
1898 ixgbe_check_lsc(adapter);
1900 if (eicr & IXGBE_EICR_MAILBOX)
1901 ixgbe_msg_task(adapter);
1903 switch (hw->mac.type) {
1904 case ixgbe_mac_82599EB:
1905 ixgbe_check_sfp_event(adapter, eicr);
1906 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
1907 ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC))) {
1908 adapter->interrupt_event = eicr;
1909 schedule_work(&adapter->check_overtemp_task);
1911 /* now fallthrough to handle Flow Director */
1912 case ixgbe_mac_X540:
1913 /* Handle Flow Director Full threshold interrupt */
1914 if (eicr & IXGBE_EICR_FLOW_DIR) {
1916 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_FLOW_DIR);
1917 /* Disable transmits before FDIR Re-initialization */
1918 netif_tx_stop_all_queues(netdev);
1919 for (i = 0; i < adapter->num_tx_queues; i++) {
1920 struct ixgbe_ring *tx_ring =
1921 adapter->tx_ring[i];
1922 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
1924 schedule_work(&adapter->fdir_reinit_task);
1932 ixgbe_check_fan_failure(adapter, eicr);
1934 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1935 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_OTHER);
1940 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
1944 struct ixgbe_hw *hw = &adapter->hw;
1946 switch (hw->mac.type) {
1947 case ixgbe_mac_82598EB:
1948 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1949 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
1951 case ixgbe_mac_82599EB:
1952 case ixgbe_mac_X540:
1953 mask = (qmask & 0xFFFFFFFF);
1955 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
1956 mask = (qmask >> 32);
1958 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
1963 /* skip the flush */
1966 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
1970 struct ixgbe_hw *hw = &adapter->hw;
1972 switch (hw->mac.type) {
1973 case ixgbe_mac_82598EB:
1974 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1975 IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
1977 case ixgbe_mac_82599EB:
1978 case ixgbe_mac_X540:
1979 mask = (qmask & 0xFFFFFFFF);
1981 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
1982 mask = (qmask >> 32);
1984 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
1989 /* skip the flush */
1992 static irqreturn_t ixgbe_msix_clean_tx(int irq, void *data)
1994 struct ixgbe_q_vector *q_vector = data;
1995 struct ixgbe_adapter *adapter = q_vector->adapter;
1996 struct ixgbe_ring *tx_ring;
1999 if (!q_vector->txr_count)
2002 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
2003 for (i = 0; i < q_vector->txr_count; i++) {
2004 tx_ring = adapter->tx_ring[r_idx];
2005 tx_ring->total_bytes = 0;
2006 tx_ring->total_packets = 0;
2007 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
2011 /* EIAM disabled interrupts (on this vector) for us */
2012 napi_schedule(&q_vector->napi);
2018 * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
2020 * @data: pointer to our q_vector struct for this interrupt vector
2022 static irqreturn_t ixgbe_msix_clean_rx(int irq, void *data)
2024 struct ixgbe_q_vector *q_vector = data;
2025 struct ixgbe_adapter *adapter = q_vector->adapter;
2026 struct ixgbe_ring *rx_ring;
2030 #ifdef CONFIG_IXGBE_DCA
2031 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2032 ixgbe_update_dca(q_vector);
2035 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2036 for (i = 0; i < q_vector->rxr_count; i++) {
2037 rx_ring = adapter->rx_ring[r_idx];
2038 rx_ring->total_bytes = 0;
2039 rx_ring->total_packets = 0;
2040 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
2044 if (!q_vector->rxr_count)
2047 /* EIAM disabled interrupts (on this vector) for us */
2048 napi_schedule(&q_vector->napi);
2053 static irqreturn_t ixgbe_msix_clean_many(int irq, void *data)
2055 struct ixgbe_q_vector *q_vector = data;
2056 struct ixgbe_adapter *adapter = q_vector->adapter;
2057 struct ixgbe_ring *ring;
2061 if (!q_vector->txr_count && !q_vector->rxr_count)
2064 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
2065 for (i = 0; i < q_vector->txr_count; i++) {
2066 ring = adapter->tx_ring[r_idx];
2067 ring->total_bytes = 0;
2068 ring->total_packets = 0;
2069 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
2073 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2074 for (i = 0; i < q_vector->rxr_count; i++) {
2075 ring = adapter->rx_ring[r_idx];
2076 ring->total_bytes = 0;
2077 ring->total_packets = 0;
2078 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
2082 /* EIAM disabled interrupts (on this vector) for us */
2083 napi_schedule(&q_vector->napi);
2089 * ixgbe_clean_rxonly - msix (aka one shot) rx clean routine
2090 * @napi: napi struct with our devices info in it
2091 * @budget: amount of work driver is allowed to do this pass, in packets
2093 * This function is optimized for cleaning one queue only on a single
2096 static int ixgbe_clean_rxonly(struct napi_struct *napi, int budget)
2098 struct ixgbe_q_vector *q_vector =
2099 container_of(napi, struct ixgbe_q_vector, napi);
2100 struct ixgbe_adapter *adapter = q_vector->adapter;
2101 struct ixgbe_ring *rx_ring = NULL;
2105 #ifdef CONFIG_IXGBE_DCA
2106 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2107 ixgbe_update_dca(q_vector);
2110 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2111 rx_ring = adapter->rx_ring[r_idx];
2113 ixgbe_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
2115 /* If all Rx work done, exit the polling mode */
2116 if (work_done < budget) {
2117 napi_complete(napi);
2118 if (adapter->rx_itr_setting & 1)
2119 ixgbe_set_itr_msix(q_vector);
2120 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2121 ixgbe_irq_enable_queues(adapter,
2122 ((u64)1 << q_vector->v_idx));
2129 * ixgbe_clean_rxtx_many - msix (aka one shot) rx clean routine
2130 * @napi: napi struct with our devices info in it
2131 * @budget: amount of work driver is allowed to do this pass, in packets
2133 * This function will clean more than one rx queue associated with a
2136 static int ixgbe_clean_rxtx_many(struct napi_struct *napi, int budget)
2138 struct ixgbe_q_vector *q_vector =
2139 container_of(napi, struct ixgbe_q_vector, napi);
2140 struct ixgbe_adapter *adapter = q_vector->adapter;
2141 struct ixgbe_ring *ring = NULL;
2142 int work_done = 0, i;
2144 bool tx_clean_complete = true;
2146 #ifdef CONFIG_IXGBE_DCA
2147 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2148 ixgbe_update_dca(q_vector);
2151 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
2152 for (i = 0; i < q_vector->txr_count; i++) {
2153 ring = adapter->tx_ring[r_idx];
2154 tx_clean_complete &= ixgbe_clean_tx_irq(q_vector, ring);
2155 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
2159 /* attempt to distribute budget to each queue fairly, but don't allow
2160 * the budget to go below 1 because we'll exit polling */
2161 budget /= (q_vector->rxr_count ?: 1);
2162 budget = max(budget, 1);
2163 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2164 for (i = 0; i < q_vector->rxr_count; i++) {
2165 ring = adapter->rx_ring[r_idx];
2166 ixgbe_clean_rx_irq(q_vector, ring, &work_done, budget);
2167 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
2171 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2172 ring = adapter->rx_ring[r_idx];
2173 /* If all Rx work done, exit the polling mode */
2174 if (work_done < budget) {
2175 napi_complete(napi);
2176 if (adapter->rx_itr_setting & 1)
2177 ixgbe_set_itr_msix(q_vector);
2178 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2179 ixgbe_irq_enable_queues(adapter,
2180 ((u64)1 << q_vector->v_idx));
2188 * ixgbe_clean_txonly - msix (aka one shot) tx clean routine
2189 * @napi: napi struct with our devices info in it
2190 * @budget: amount of work driver is allowed to do this pass, in packets
2192 * This function is optimized for cleaning one queue only on a single
2195 static int ixgbe_clean_txonly(struct napi_struct *napi, int budget)
2197 struct ixgbe_q_vector *q_vector =
2198 container_of(napi, struct ixgbe_q_vector, napi);
2199 struct ixgbe_adapter *adapter = q_vector->adapter;
2200 struct ixgbe_ring *tx_ring = NULL;
2204 #ifdef CONFIG_IXGBE_DCA
2205 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2206 ixgbe_update_dca(q_vector);
2209 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
2210 tx_ring = adapter->tx_ring[r_idx];
2212 if (!ixgbe_clean_tx_irq(q_vector, tx_ring))
2215 /* If all Tx work done, exit the polling mode */
2216 if (work_done < budget) {
2217 napi_complete(napi);
2218 if (adapter->tx_itr_setting & 1)
2219 ixgbe_set_itr_msix(q_vector);
2220 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2221 ixgbe_irq_enable_queues(adapter,
2222 ((u64)1 << q_vector->v_idx));
2228 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
2231 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2232 struct ixgbe_ring *rx_ring = a->rx_ring[r_idx];
2234 set_bit(r_idx, q_vector->rxr_idx);
2235 q_vector->rxr_count++;
2236 rx_ring->q_vector = q_vector;
2239 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
2242 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2243 struct ixgbe_ring *tx_ring = a->tx_ring[t_idx];
2245 set_bit(t_idx, q_vector->txr_idx);
2246 q_vector->txr_count++;
2247 tx_ring->q_vector = q_vector;
2251 * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
2252 * @adapter: board private structure to initialize
2254 * This function maps descriptor rings to the queue-specific vectors
2255 * we were allotted through the MSI-X enabling code. Ideally, we'd have
2256 * one vector per ring/queue, but on a constrained vector budget, we
2257 * group the rings as "efficiently" as possible. You would add new
2258 * mapping configurations in here.
2260 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter)
2264 int rxr_idx = 0, txr_idx = 0;
2265 int rxr_remaining = adapter->num_rx_queues;
2266 int txr_remaining = adapter->num_tx_queues;
2271 /* No mapping required if MSI-X is disabled. */
2272 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2275 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2278 * The ideal configuration...
2279 * We have enough vectors to map one per queue.
2281 if (q_vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
2282 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
2283 map_vector_to_rxq(adapter, v_start, rxr_idx);
2285 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
2286 map_vector_to_txq(adapter, v_start, txr_idx);
2292 * If we don't have enough vectors for a 1-to-1
2293 * mapping, we'll have to group them so there are
2294 * multiple queues per vector.
2296 /* Re-adjusting *qpv takes care of the remainder. */
2297 for (i = v_start; i < q_vectors; i++) {
2298 rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - i);
2299 for (j = 0; j < rqpv; j++) {
2300 map_vector_to_rxq(adapter, i, rxr_idx);
2304 tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - i);
2305 for (j = 0; j < tqpv; j++) {
2306 map_vector_to_txq(adapter, i, txr_idx);
2316 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2317 * @adapter: board private structure
2319 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2320 * interrupts from the kernel.
2322 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2324 struct net_device *netdev = adapter->netdev;
2325 irqreturn_t (*handler)(int, void *);
2326 int i, vector, q_vectors, err;
2329 /* Decrement for Other and TCP Timer vectors */
2330 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2332 err = ixgbe_map_rings_to_vectors(adapter);
2336 #define SET_HANDLER(_v) (((_v)->rxr_count && (_v)->txr_count) \
2337 ? &ixgbe_msix_clean_many : \
2338 (_v)->rxr_count ? &ixgbe_msix_clean_rx : \
2339 (_v)->txr_count ? &ixgbe_msix_clean_tx : \
2341 for (vector = 0; vector < q_vectors; vector++) {
2342 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2343 handler = SET_HANDLER(q_vector);
2345 if (handler == &ixgbe_msix_clean_rx) {
2346 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2347 "%s-%s-%d", netdev->name, "rx", ri++);
2348 } else if (handler == &ixgbe_msix_clean_tx) {
2349 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2350 "%s-%s-%d", netdev->name, "tx", ti++);
2351 } else if (handler == &ixgbe_msix_clean_many) {
2352 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2353 "%s-%s-%d", netdev->name, "TxRx", ri++);
2356 /* skip this unused q_vector */
2359 err = request_irq(adapter->msix_entries[vector].vector,
2360 handler, 0, q_vector->name,
2363 e_err(probe, "request_irq failed for MSIX interrupt "
2364 "Error: %d\n", err);
2365 goto free_queue_irqs;
2369 sprintf(adapter->lsc_int_name, "%s:lsc", netdev->name);
2370 err = request_irq(adapter->msix_entries[vector].vector,
2371 ixgbe_msix_lsc, 0, adapter->lsc_int_name, netdev);
2373 e_err(probe, "request_irq for msix_lsc failed: %d\n", err);
2374 goto free_queue_irqs;
2380 for (i = vector - 1; i >= 0; i--)
2381 free_irq(adapter->msix_entries[--vector].vector,
2382 adapter->q_vector[i]);
2383 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2384 pci_disable_msix(adapter->pdev);
2385 kfree(adapter->msix_entries);
2386 adapter->msix_entries = NULL;
2390 static void ixgbe_set_itr(struct ixgbe_adapter *adapter)
2392 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2393 struct ixgbe_ring *rx_ring = adapter->rx_ring[0];
2394 struct ixgbe_ring *tx_ring = adapter->tx_ring[0];
2395 u32 new_itr = q_vector->eitr;
2398 q_vector->tx_itr = ixgbe_update_itr(adapter, new_itr,
2400 tx_ring->total_packets,
2401 tx_ring->total_bytes);
2402 q_vector->rx_itr = ixgbe_update_itr(adapter, new_itr,
2404 rx_ring->total_packets,
2405 rx_ring->total_bytes);
2407 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
2409 switch (current_itr) {
2410 /* counts and packets in update_itr are dependent on these numbers */
2411 case lowest_latency:
2415 new_itr = 20000; /* aka hwitr = ~200 */
2424 if (new_itr != q_vector->eitr) {
2425 /* do an exponential smoothing */
2426 new_itr = ((q_vector->eitr * 9) + new_itr)/10;
2428 /* save the algorithm value here */
2429 q_vector->eitr = new_itr;
2431 ixgbe_write_eitr(q_vector);
2436 * ixgbe_irq_enable - Enable default interrupt generation settings
2437 * @adapter: board private structure
2439 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2444 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2445 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2446 mask |= IXGBE_EIMS_GPI_SDP0;
2447 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2448 mask |= IXGBE_EIMS_GPI_SDP1;
2449 switch (adapter->hw.mac.type) {
2450 case ixgbe_mac_82599EB:
2451 case ixgbe_mac_X540:
2452 mask |= IXGBE_EIMS_ECC;
2453 mask |= IXGBE_EIMS_GPI_SDP1;
2454 mask |= IXGBE_EIMS_GPI_SDP2;
2455 if (adapter->num_vfs)
2456 mask |= IXGBE_EIMS_MAILBOX;
2461 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
2462 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
2463 mask |= IXGBE_EIMS_FLOW_DIR;
2465 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2467 ixgbe_irq_enable_queues(adapter, ~0);
2469 IXGBE_WRITE_FLUSH(&adapter->hw);
2471 if (adapter->num_vfs > 32) {
2472 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
2473 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2478 * ixgbe_intr - legacy mode Interrupt Handler
2479 * @irq: interrupt number
2480 * @data: pointer to a network interface device structure
2482 static irqreturn_t ixgbe_intr(int irq, void *data)
2484 struct net_device *netdev = data;
2485 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2486 struct ixgbe_hw *hw = &adapter->hw;
2487 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2491 * Workaround for silicon errata on 82598. Mask the interrupts
2492 * before the read of EICR.
2494 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2496 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2497 * therefore no explict interrupt disable is necessary */
2498 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2501 * shared interrupt alert!
2502 * make sure interrupts are enabled because the read will
2503 * have disabled interrupts due to EIAM
2504 * finish the workaround of silicon errata on 82598. Unmask
2505 * the interrupt that we masked before the EICR read.
2507 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2508 ixgbe_irq_enable(adapter, true, true);
2509 return IRQ_NONE; /* Not our interrupt */
2512 if (eicr & IXGBE_EICR_LSC)
2513 ixgbe_check_lsc(adapter);
2515 switch (hw->mac.type) {
2516 case ixgbe_mac_82599EB:
2517 ixgbe_check_sfp_event(adapter, eicr);
2518 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2519 ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC))) {
2520 adapter->interrupt_event = eicr;
2521 schedule_work(&adapter->check_overtemp_task);
2528 ixgbe_check_fan_failure(adapter, eicr);
2530 if (napi_schedule_prep(&(q_vector->napi))) {
2531 adapter->tx_ring[0]->total_packets = 0;
2532 adapter->tx_ring[0]->total_bytes = 0;
2533 adapter->rx_ring[0]->total_packets = 0;
2534 adapter->rx_ring[0]->total_bytes = 0;
2535 /* would disable interrupts here but EIAM disabled it */
2536 __napi_schedule(&(q_vector->napi));
2540 * re-enable link(maybe) and non-queue interrupts, no flush.
2541 * ixgbe_poll will re-enable the queue interrupts
2544 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2545 ixgbe_irq_enable(adapter, false, false);
2550 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
2552 int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2554 for (i = 0; i < q_vectors; i++) {
2555 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
2556 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
2557 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
2558 q_vector->rxr_count = 0;
2559 q_vector->txr_count = 0;
2564 * ixgbe_request_irq - initialize interrupts
2565 * @adapter: board private structure
2567 * Attempts to configure interrupts using the best available
2568 * capabilities of the hardware and kernel.
2570 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2572 struct net_device *netdev = adapter->netdev;
2575 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2576 err = ixgbe_request_msix_irqs(adapter);
2577 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
2578 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2579 netdev->name, netdev);
2581 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2582 netdev->name, netdev);
2586 e_err(probe, "request_irq failed, Error %d\n", err);
2591 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2593 struct net_device *netdev = adapter->netdev;
2595 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2598 q_vectors = adapter->num_msix_vectors;
2601 free_irq(adapter->msix_entries[i].vector, netdev);
2604 for (; i >= 0; i--) {
2605 /* free only the irqs that were actually requested */
2606 if (!adapter->q_vector[i]->rxr_count &&
2607 !adapter->q_vector[i]->txr_count)
2610 free_irq(adapter->msix_entries[i].vector,
2611 adapter->q_vector[i]);
2614 ixgbe_reset_q_vectors(adapter);
2616 free_irq(adapter->pdev->irq, netdev);
2621 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2622 * @adapter: board private structure
2624 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2626 switch (adapter->hw.mac.type) {
2627 case ixgbe_mac_82598EB:
2628 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2630 case ixgbe_mac_82599EB:
2631 case ixgbe_mac_X540:
2632 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2633 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2634 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2635 if (adapter->num_vfs > 32)
2636 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
2641 IXGBE_WRITE_FLUSH(&adapter->hw);
2642 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2644 for (i = 0; i < adapter->num_msix_vectors; i++)
2645 synchronize_irq(adapter->msix_entries[i].vector);
2647 synchronize_irq(adapter->pdev->irq);
2652 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2655 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2657 struct ixgbe_hw *hw = &adapter->hw;
2659 IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
2660 EITR_INTS_PER_SEC_TO_REG(adapter->rx_eitr_param));
2662 ixgbe_set_ivar(adapter, 0, 0, 0);
2663 ixgbe_set_ivar(adapter, 1, 0, 0);
2665 map_vector_to_rxq(adapter, 0, 0);
2666 map_vector_to_txq(adapter, 0, 0);
2668 e_info(hw, "Legacy interrupt IVAR setup done\n");
2672 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2673 * @adapter: board private structure
2674 * @ring: structure containing ring specific data
2676 * Configure the Tx descriptor ring after a reset.
2678 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2679 struct ixgbe_ring *ring)
2681 struct ixgbe_hw *hw = &adapter->hw;
2682 u64 tdba = ring->dma;
2685 u8 reg_idx = ring->reg_idx;
2687 /* disable queue to avoid issues while updating state */
2688 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2689 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx),
2690 txdctl & ~IXGBE_TXDCTL_ENABLE);
2691 IXGBE_WRITE_FLUSH(hw);
2693 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2694 (tdba & DMA_BIT_MASK(32)));
2695 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2696 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2697 ring->count * sizeof(union ixgbe_adv_tx_desc));
2698 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2699 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2700 ring->tail = hw->hw_addr + IXGBE_TDT(reg_idx);
2702 /* configure fetching thresholds */
2703 if (adapter->rx_itr_setting == 0) {
2704 /* cannot set wthresh when itr==0 */
2705 txdctl &= ~0x007F0000;
2707 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
2708 txdctl |= (8 << 16);
2710 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2711 /* PThresh workaround for Tx hang with DFP enabled. */
2715 /* reinitialize flowdirector state */
2716 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2717 adapter->atr_sample_rate) {
2718 ring->atr_sample_rate = adapter->atr_sample_rate;
2719 ring->atr_count = 0;
2720 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
2722 ring->atr_sample_rate = 0;
2725 clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
2728 txdctl |= IXGBE_TXDCTL_ENABLE;
2729 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
2731 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2732 if (hw->mac.type == ixgbe_mac_82598EB &&
2733 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2736 /* poll to verify queue is enabled */
2739 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2740 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
2742 e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
2745 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
2747 struct ixgbe_hw *hw = &adapter->hw;
2751 if (hw->mac.type == ixgbe_mac_82598EB)
2754 /* disable the arbiter while setting MTQC */
2755 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2756 rttdcs |= IXGBE_RTTDCS_ARBDIS;
2757 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2759 /* set transmit pool layout */
2760 mask = (IXGBE_FLAG_SRIOV_ENABLED | IXGBE_FLAG_DCB_ENABLED);
2761 switch (adapter->flags & mask) {
2763 case (IXGBE_FLAG_SRIOV_ENABLED):
2764 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2765 (IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF));
2768 case (IXGBE_FLAG_DCB_ENABLED):
2769 /* We enable 8 traffic classes, DCB only */
2770 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2771 (IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ));
2775 IXGBE_WRITE_REG(hw, IXGBE_MTQC, IXGBE_MTQC_64Q_1PB);
2779 /* re-enable the arbiter */
2780 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2781 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2785 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2786 * @adapter: board private structure
2788 * Configure the Tx unit of the MAC after a reset.
2790 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2792 struct ixgbe_hw *hw = &adapter->hw;
2796 ixgbe_setup_mtqc(adapter);
2798 if (hw->mac.type != ixgbe_mac_82598EB) {
2799 /* DMATXCTL.EN must be before Tx queues are enabled */
2800 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2801 dmatxctl |= IXGBE_DMATXCTL_TE;
2802 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2805 /* Setup the HW Tx Head and Tail descriptor pointers */
2806 for (i = 0; i < adapter->num_tx_queues; i++)
2807 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
2810 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2812 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2813 struct ixgbe_ring *rx_ring)
2816 u8 reg_idx = rx_ring->reg_idx;
2818 switch (adapter->hw.mac.type) {
2819 case ixgbe_mac_82598EB: {
2820 struct ixgbe_ring_feature *feature = adapter->ring_feature;
2821 const int mask = feature[RING_F_RSS].mask;
2822 reg_idx = reg_idx & mask;
2825 case ixgbe_mac_82599EB:
2826 case ixgbe_mac_X540:
2831 srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx));
2833 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
2834 srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
2835 if (adapter->num_vfs)
2836 srrctl |= IXGBE_SRRCTL_DROP_EN;
2838 srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
2839 IXGBE_SRRCTL_BSIZEHDR_MASK;
2841 if (ring_is_ps_enabled(rx_ring)) {
2842 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
2843 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2845 srrctl |= (PAGE_SIZE / 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2847 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2849 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
2850 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2851 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2854 IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx), srrctl);
2857 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2859 struct ixgbe_hw *hw = &adapter->hw;
2860 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2861 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2862 0x6A3E67EA, 0x14364D17, 0x3BED200D};
2863 u32 mrqc = 0, reta = 0;
2868 /* Fill out hash function seeds */
2869 for (i = 0; i < 10; i++)
2870 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2872 /* Fill out redirection table */
2873 for (i = 0, j = 0; i < 128; i++, j++) {
2874 if (j == adapter->ring_feature[RING_F_RSS].indices)
2876 /* reta = 4-byte sliding window of
2877 * 0x00..(indices-1)(indices-1)00..etc. */
2878 reta = (reta << 8) | (j * 0x11);
2880 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2883 /* Disable indicating checksum in descriptor, enables RSS hash */
2884 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2885 rxcsum |= IXGBE_RXCSUM_PCSD;
2886 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2888 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
2889 mask = adapter->flags & IXGBE_FLAG_RSS_ENABLED;
2891 mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2892 #ifdef CONFIG_IXGBE_DCB
2893 | IXGBE_FLAG_DCB_ENABLED
2895 | IXGBE_FLAG_SRIOV_ENABLED
2899 #ifdef CONFIG_IXGBE_DCB
2900 case (IXGBE_FLAG_DCB_ENABLED | IXGBE_FLAG_RSS_ENABLED):
2901 mrqc = IXGBE_MRQC_RTRSS8TCEN;
2903 case (IXGBE_FLAG_DCB_ENABLED):
2904 mrqc = IXGBE_MRQC_RT8TCEN;
2906 #endif /* CONFIG_IXGBE_DCB */
2907 case (IXGBE_FLAG_RSS_ENABLED):
2908 mrqc = IXGBE_MRQC_RSSEN;
2910 case (IXGBE_FLAG_SRIOV_ENABLED):
2911 mrqc = IXGBE_MRQC_VMDQEN;
2917 /* Perform hash on these packet types */
2918 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2919 | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2920 | IXGBE_MRQC_RSS_FIELD_IPV6
2921 | IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
2923 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2927 * ixgbe_clear_rscctl - disable RSC for the indicated ring
2928 * @adapter: address of board private structure
2929 * @ring: structure containing ring specific data
2931 void ixgbe_clear_rscctl(struct ixgbe_adapter *adapter,
2932 struct ixgbe_ring *ring)
2934 struct ixgbe_hw *hw = &adapter->hw;
2936 u8 reg_idx = ring->reg_idx;
2938 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
2939 rscctrl &= ~IXGBE_RSCCTL_RSCEN;
2940 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
2944 * ixgbe_configure_rscctl - enable RSC for the indicated ring
2945 * @adapter: address of board private structure
2946 * @index: index of ring to set
2948 void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
2949 struct ixgbe_ring *ring)
2951 struct ixgbe_hw *hw = &adapter->hw;
2954 u8 reg_idx = ring->reg_idx;
2956 if (!ring_is_rsc_enabled(ring))
2959 rx_buf_len = ring->rx_buf_len;
2960 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
2961 rscctrl |= IXGBE_RSCCTL_RSCEN;
2963 * we must limit the number of descriptors so that the
2964 * total size of max desc * buf_len is not greater
2967 if (ring_is_ps_enabled(ring)) {
2968 #if (MAX_SKB_FRAGS > 16)
2969 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2970 #elif (MAX_SKB_FRAGS > 8)
2971 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2972 #elif (MAX_SKB_FRAGS > 4)
2973 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2975 rscctrl |= IXGBE_RSCCTL_MAXDESC_1;
2978 if (rx_buf_len < IXGBE_RXBUFFER_4096)
2979 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2980 else if (rx_buf_len < IXGBE_RXBUFFER_8192)
2981 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2983 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2985 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
2989 * ixgbe_set_uta - Set unicast filter table address
2990 * @adapter: board private structure
2992 * The unicast table address is a register array of 32-bit registers.
2993 * The table is meant to be used in a way similar to how the MTA is used
2994 * however due to certain limitations in the hardware it is necessary to
2995 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
2996 * enable bit to allow vlan tag stripping when promiscuous mode is enabled
2998 static void ixgbe_set_uta(struct ixgbe_adapter *adapter)
3000 struct ixgbe_hw *hw = &adapter->hw;
3003 /* The UTA table only exists on 82599 hardware and newer */
3004 if (hw->mac.type < ixgbe_mac_82599EB)
3007 /* we only need to do this if VMDq is enabled */
3008 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
3011 for (i = 0; i < 128; i++)
3012 IXGBE_WRITE_REG(hw, IXGBE_UTA(i), ~0);
3015 #define IXGBE_MAX_RX_DESC_POLL 10
3016 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
3017 struct ixgbe_ring *ring)
3019 struct ixgbe_hw *hw = &adapter->hw;
3020 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3022 u8 reg_idx = ring->reg_idx;
3024 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3025 if (hw->mac.type == ixgbe_mac_82598EB &&
3026 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3031 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3032 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
3035 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
3036 "the polling period\n", reg_idx);
3040 void ixgbe_disable_rx_queue(struct ixgbe_adapter *adapter,
3041 struct ixgbe_ring *ring)
3043 struct ixgbe_hw *hw = &adapter->hw;
3044 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3046 u8 reg_idx = ring->reg_idx;
3048 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3049 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
3051 /* write value back with RXDCTL.ENABLE bit cleared */
3052 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3054 if (hw->mac.type == ixgbe_mac_82598EB &&
3055 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3058 /* the hardware may take up to 100us to really disable the rx queue */
3061 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3062 } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
3065 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not cleared within "
3066 "the polling period\n", reg_idx);
3070 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
3071 struct ixgbe_ring *ring)
3073 struct ixgbe_hw *hw = &adapter->hw;
3074 u64 rdba = ring->dma;
3076 u8 reg_idx = ring->reg_idx;
3078 /* disable queue to avoid issues while updating state */
3079 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3080 ixgbe_disable_rx_queue(adapter, ring);
3082 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
3083 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
3084 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
3085 ring->count * sizeof(union ixgbe_adv_rx_desc));
3086 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
3087 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
3088 ring->tail = hw->hw_addr + IXGBE_RDT(reg_idx);
3090 ixgbe_configure_srrctl(adapter, ring);
3091 ixgbe_configure_rscctl(adapter, ring);
3093 /* If operating in IOV mode set RLPML for X540 */
3094 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
3095 hw->mac.type == ixgbe_mac_X540) {
3096 rxdctl &= ~IXGBE_RXDCTL_RLPMLMASK;
3097 rxdctl |= ((ring->netdev->mtu + ETH_HLEN +
3098 ETH_FCS_LEN + VLAN_HLEN) | IXGBE_RXDCTL_RLPML_EN);
3101 if (hw->mac.type == ixgbe_mac_82598EB) {
3103 * enable cache line friendly hardware writes:
3104 * PTHRESH=32 descriptors (half the internal cache),
3105 * this also removes ugly rx_no_buffer_count increment
3106 * HTHRESH=4 descriptors (to minimize latency on fetch)
3107 * WTHRESH=8 burst writeback up to two cache lines
3109 rxdctl &= ~0x3FFFFF;
3113 /* enable receive descriptor ring */
3114 rxdctl |= IXGBE_RXDCTL_ENABLE;
3115 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3117 ixgbe_rx_desc_queue_enable(adapter, ring);
3118 ixgbe_alloc_rx_buffers(ring, IXGBE_DESC_UNUSED(ring));
3121 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
3123 struct ixgbe_hw *hw = &adapter->hw;
3126 /* PSRTYPE must be initialized in non 82598 adapters */
3127 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
3128 IXGBE_PSRTYPE_UDPHDR |
3129 IXGBE_PSRTYPE_IPV4HDR |
3130 IXGBE_PSRTYPE_L2HDR |
3131 IXGBE_PSRTYPE_IPV6HDR;
3133 if (hw->mac.type == ixgbe_mac_82598EB)
3136 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED)
3137 psrtype |= (adapter->num_rx_queues_per_pool << 29);
3139 for (p = 0; p < adapter->num_rx_pools; p++)
3140 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(adapter->num_vfs + p),
3144 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
3146 struct ixgbe_hw *hw = &adapter->hw;
3149 u32 reg_offset, vf_shift;
3152 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
3155 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
3156 vt_reg_bits = IXGBE_VMD_CTL_VMDQ_EN | IXGBE_VT_CTL_REPLEN;
3157 vt_reg_bits |= (adapter->num_vfs << IXGBE_VT_CTL_POOL_SHIFT);
3158 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl | vt_reg_bits);
3160 vf_shift = adapter->num_vfs % 32;
3161 reg_offset = (adapter->num_vfs > 32) ? 1 : 0;
3163 /* Enable only the PF's pool for Tx/Rx */
3164 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (1 << vf_shift));
3165 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), 0);
3166 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (1 << vf_shift));
3167 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), 0);
3168 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3170 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
3171 hw->mac.ops.set_vmdq(hw, 0, adapter->num_vfs);
3174 * Set up VF register offsets for selected VT Mode,
3175 * i.e. 32 or 64 VFs for SR-IOV
3177 gcr_ext = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
3178 gcr_ext |= IXGBE_GCR_EXT_MSIX_EN;
3179 gcr_ext |= IXGBE_GCR_EXT_VT_MODE_64;
3180 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
3182 /* enable Tx loopback for VF/PF communication */
3183 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3184 /* Enable MAC Anti-Spoofing */
3185 hw->mac.ops.set_mac_anti_spoofing(hw, (adapter->num_vfs != 0),
3189 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
3191 struct ixgbe_hw *hw = &adapter->hw;
3192 struct net_device *netdev = adapter->netdev;
3193 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3195 struct ixgbe_ring *rx_ring;
3199 /* Decide whether to use packet split mode or not */
3201 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
3203 /* Do not use packet split if we're in SR-IOV Mode */
3204 if (adapter->num_vfs)
3205 adapter->flags &= ~IXGBE_FLAG_RX_PS_ENABLED;
3207 /* Disable packet split due to 82599 erratum #45 */
3208 if (hw->mac.type == ixgbe_mac_82599EB)
3209 adapter->flags &= ~IXGBE_FLAG_RX_PS_ENABLED;
3211 /* Set the RX buffer length according to the mode */
3212 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
3213 rx_buf_len = IXGBE_RX_HDR_SIZE;
3215 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
3216 (netdev->mtu <= ETH_DATA_LEN))
3217 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3219 rx_buf_len = ALIGN(max_frame + VLAN_HLEN, 1024);
3223 /* adjust max frame to be able to do baby jumbo for FCoE */
3224 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
3225 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
3226 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3228 #endif /* IXGBE_FCOE */
3229 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
3230 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
3231 mhadd &= ~IXGBE_MHADD_MFS_MASK;
3232 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
3234 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
3237 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
3238 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
3239 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
3240 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
3243 * Setup the HW Rx Head and Tail Descriptor Pointers and
3244 * the Base and Length of the Rx Descriptor Ring
3246 for (i = 0; i < adapter->num_rx_queues; i++) {
3247 rx_ring = adapter->rx_ring[i];
3248 rx_ring->rx_buf_len = rx_buf_len;
3250 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)
3251 set_ring_ps_enabled(rx_ring);
3253 clear_ring_ps_enabled(rx_ring);
3255 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
3256 set_ring_rsc_enabled(rx_ring);
3258 clear_ring_rsc_enabled(rx_ring);
3261 if (netdev->features & NETIF_F_FCOE_MTU) {
3262 struct ixgbe_ring_feature *f;
3263 f = &adapter->ring_feature[RING_F_FCOE];
3264 if ((i >= f->mask) && (i < f->mask + f->indices)) {
3265 clear_ring_ps_enabled(rx_ring);
3266 if (rx_buf_len < IXGBE_FCOE_JUMBO_FRAME_SIZE)
3267 rx_ring->rx_buf_len =
3268 IXGBE_FCOE_JUMBO_FRAME_SIZE;
3269 } else if (!ring_is_rsc_enabled(rx_ring) &&
3270 !ring_is_ps_enabled(rx_ring)) {
3271 rx_ring->rx_buf_len =
3272 IXGBE_FCOE_JUMBO_FRAME_SIZE;
3275 #endif /* IXGBE_FCOE */
3279 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
3281 struct ixgbe_hw *hw = &adapter->hw;
3282 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
3284 switch (hw->mac.type) {
3285 case ixgbe_mac_82598EB:
3287 * For VMDq support of different descriptor types or
3288 * buffer sizes through the use of multiple SRRCTL
3289 * registers, RDRXCTL.MVMEN must be set to 1
3291 * also, the manual doesn't mention it clearly but DCA hints
3292 * will only use queue 0's tags unless this bit is set. Side
3293 * effects of setting this bit are only that SRRCTL must be
3294 * fully programmed [0..15]
3296 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
3298 case ixgbe_mac_82599EB:
3299 case ixgbe_mac_X540:
3300 /* Disable RSC for ACK packets */
3301 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
3302 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
3303 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3304 /* hardware requires some bits to be set by default */
3305 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3306 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3309 /* We should do nothing since we don't know this hardware */
3313 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3317 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3318 * @adapter: board private structure
3320 * Configure the Rx unit of the MAC after a reset.
3322 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3324 struct ixgbe_hw *hw = &adapter->hw;
3328 /* disable receives while setting up the descriptors */
3329 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3330 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3332 ixgbe_setup_psrtype(adapter);
3333 ixgbe_setup_rdrxctl(adapter);
3335 /* Program registers for the distribution of queues */
3336 ixgbe_setup_mrqc(adapter);
3338 ixgbe_set_uta(adapter);
3340 /* set_rx_buffer_len must be called before ring initialization */
3341 ixgbe_set_rx_buffer_len(adapter);
3344 * Setup the HW Rx Head and Tail Descriptor Pointers and
3345 * the Base and Length of the Rx Descriptor Ring
3347 for (i = 0; i < adapter->num_rx_queues; i++)
3348 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3350 /* disable drop enable for 82598 parts */
3351 if (hw->mac.type == ixgbe_mac_82598EB)
3352 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3354 /* enable all receives */
3355 rxctrl |= IXGBE_RXCTRL_RXEN;
3356 hw->mac.ops.enable_rx_dma(hw, rxctrl);
3359 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3361 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3362 struct ixgbe_hw *hw = &adapter->hw;
3363 int pool_ndx = adapter->num_vfs;
3365 /* add VID to filter table */
3366 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, true);
3367 set_bit(vid, adapter->active_vlans);
3370 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3372 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3373 struct ixgbe_hw *hw = &adapter->hw;
3374 int pool_ndx = adapter->num_vfs;
3376 /* remove VID from filter table */
3377 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, false);
3378 clear_bit(vid, adapter->active_vlans);
3382 * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3383 * @adapter: driver data
3385 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
3387 struct ixgbe_hw *hw = &adapter->hw;
3390 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3391 vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
3392 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3396 * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3397 * @adapter: driver data
3399 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
3401 struct ixgbe_hw *hw = &adapter->hw;
3404 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3405 vlnctrl |= IXGBE_VLNCTRL_VFE;
3406 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3407 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3411 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3412 * @adapter: driver data
3414 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3416 struct ixgbe_hw *hw = &adapter->hw;
3420 switch (hw->mac.type) {
3421 case ixgbe_mac_82598EB:
3422 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3423 vlnctrl &= ~IXGBE_VLNCTRL_VME;
3424 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3426 case ixgbe_mac_82599EB:
3427 case ixgbe_mac_X540:
3428 for (i = 0; i < adapter->num_rx_queues; i++) {
3429 j = adapter->rx_ring[i]->reg_idx;
3430 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3431 vlnctrl &= ~IXGBE_RXDCTL_VME;
3432 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3441 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3442 * @adapter: driver data
3444 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3446 struct ixgbe_hw *hw = &adapter->hw;
3450 switch (hw->mac.type) {
3451 case ixgbe_mac_82598EB:
3452 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3453 vlnctrl |= IXGBE_VLNCTRL_VME;
3454 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3456 case ixgbe_mac_82599EB:
3457 case ixgbe_mac_X540:
3458 for (i = 0; i < adapter->num_rx_queues; i++) {
3459 j = adapter->rx_ring[i]->reg_idx;
3460 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3461 vlnctrl |= IXGBE_RXDCTL_VME;
3462 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3470 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3474 ixgbe_vlan_rx_add_vid(adapter->netdev, 0);
3476 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3477 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
3481 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3482 * @netdev: network interface device structure
3484 * Writes unicast address list to the RAR table.
3485 * Returns: -ENOMEM on failure/insufficient address space
3486 * 0 on no addresses written
3487 * X on writing X addresses to the RAR table
3489 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3491 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3492 struct ixgbe_hw *hw = &adapter->hw;
3493 unsigned int vfn = adapter->num_vfs;
3494 unsigned int rar_entries = hw->mac.num_rar_entries - (vfn + 1);
3497 /* return ENOMEM indicating insufficient memory for addresses */
3498 if (netdev_uc_count(netdev) > rar_entries)
3501 if (!netdev_uc_empty(netdev) && rar_entries) {
3502 struct netdev_hw_addr *ha;
3503 /* return error if we do not support writing to RAR table */
3504 if (!hw->mac.ops.set_rar)
3507 netdev_for_each_uc_addr(ha, netdev) {
3510 hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3515 /* write the addresses in reverse order to avoid write combining */
3516 for (; rar_entries > 0 ; rar_entries--)
3517 hw->mac.ops.clear_rar(hw, rar_entries);
3523 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3524 * @netdev: network interface device structure
3526 * The set_rx_method entry point is called whenever the unicast/multicast
3527 * address list or the network interface flags are updated. This routine is
3528 * responsible for configuring the hardware for proper unicast, multicast and
3531 void ixgbe_set_rx_mode(struct net_device *netdev)
3533 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3534 struct ixgbe_hw *hw = &adapter->hw;
3535 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3538 /* Check for Promiscuous and All Multicast modes */
3540 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3542 /* set all bits that we expect to always be set */
3543 fctrl |= IXGBE_FCTRL_BAM;
3544 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3545 fctrl |= IXGBE_FCTRL_PMCF;
3547 /* clear the bits we are changing the status of */
3548 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3550 if (netdev->flags & IFF_PROMISC) {
3551 hw->addr_ctrl.user_set_promisc = true;
3552 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3553 vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3554 /* don't hardware filter vlans in promisc mode */
3555 ixgbe_vlan_filter_disable(adapter);
3557 if (netdev->flags & IFF_ALLMULTI) {
3558 fctrl |= IXGBE_FCTRL_MPE;
3559 vmolr |= IXGBE_VMOLR_MPE;
3562 * Write addresses to the MTA, if the attempt fails
3563 * then we should just turn on promiscuous mode so
3564 * that we can at least receive multicast traffic
3566 hw->mac.ops.update_mc_addr_list(hw, netdev);
3567 vmolr |= IXGBE_VMOLR_ROMPE;
3569 ixgbe_vlan_filter_enable(adapter);
3570 hw->addr_ctrl.user_set_promisc = false;
3572 * Write addresses to available RAR registers, if there is not
3573 * sufficient space to store all the addresses then enable
3574 * unicast promiscuous mode
3576 count = ixgbe_write_uc_addr_list(netdev);
3578 fctrl |= IXGBE_FCTRL_UPE;
3579 vmolr |= IXGBE_VMOLR_ROPE;
3583 if (adapter->num_vfs) {
3584 ixgbe_restore_vf_multicasts(adapter);
3585 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(adapter->num_vfs)) &
3586 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3588 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(adapter->num_vfs), vmolr);
3591 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3593 if (netdev->features & NETIF_F_HW_VLAN_RX)
3594 ixgbe_vlan_strip_enable(adapter);
3596 ixgbe_vlan_strip_disable(adapter);
3599 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3602 struct ixgbe_q_vector *q_vector;
3603 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3605 /* legacy and MSI only use one vector */
3606 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3609 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3610 struct napi_struct *napi;
3611 q_vector = adapter->q_vector[q_idx];
3612 napi = &q_vector->napi;
3613 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3614 if (!q_vector->rxr_count || !q_vector->txr_count) {
3615 if (q_vector->txr_count == 1)
3616 napi->poll = &ixgbe_clean_txonly;
3617 else if (q_vector->rxr_count == 1)
3618 napi->poll = &ixgbe_clean_rxonly;
3626 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3629 struct ixgbe_q_vector *q_vector;
3630 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3632 /* legacy and MSI only use one vector */
3633 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3636 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3637 q_vector = adapter->q_vector[q_idx];
3638 napi_disable(&q_vector->napi);
3642 #ifdef CONFIG_IXGBE_DCB
3644 * ixgbe_configure_dcb - Configure DCB hardware
3645 * @adapter: ixgbe adapter struct
3647 * This is called by the driver on open to configure the DCB hardware.
3648 * This is also called by the gennetlink interface when reconfiguring
3651 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3653 struct ixgbe_hw *hw = &adapter->hw;
3654 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3656 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3657 if (hw->mac.type == ixgbe_mac_82598EB)
3658 netif_set_gso_max_size(adapter->netdev, 65536);
3662 if (hw->mac.type == ixgbe_mac_82598EB)
3663 netif_set_gso_max_size(adapter->netdev, 32768);
3666 /* Enable VLAN tag insert/strip */
3667 adapter->netdev->features |= NETIF_F_HW_VLAN_RX;
3669 hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
3671 /* reconfigure the hardware */
3672 if (adapter->dcbx_cap & (DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE)) {
3674 if (adapter->netdev->features & NETIF_F_FCOE_MTU)
3675 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
3677 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3679 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3681 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
3683 struct net_device *dev = adapter->netdev;
3685 if (adapter->ixgbe_ieee_ets)
3686 dev->dcbnl_ops->ieee_setets(dev,
3687 adapter->ixgbe_ieee_ets);
3688 if (adapter->ixgbe_ieee_pfc)
3689 dev->dcbnl_ops->ieee_setpfc(dev,
3690 adapter->ixgbe_ieee_pfc);
3693 /* Enable RSS Hash per TC */
3694 if (hw->mac.type != ixgbe_mac_82598EB) {
3698 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
3700 u8 cnt = adapter->netdev->tc_to_txq[i].count;
3705 reg |= msb << IXGBE_RQTC_SHIFT_TC(i);
3707 IXGBE_WRITE_REG(hw, IXGBE_RQTC, reg);
3712 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3714 struct net_device *netdev = adapter->netdev;
3715 struct ixgbe_hw *hw = &adapter->hw;
3718 #ifdef CONFIG_IXGBE_DCB
3719 ixgbe_configure_dcb(adapter);
3722 ixgbe_set_rx_mode(netdev);
3723 ixgbe_restore_vlan(adapter);
3726 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
3727 ixgbe_configure_fcoe(adapter);
3729 #endif /* IXGBE_FCOE */
3730 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3731 for (i = 0; i < adapter->num_tx_queues; i++)
3732 adapter->tx_ring[i]->atr_sample_rate =
3733 adapter->atr_sample_rate;
3734 ixgbe_init_fdir_signature_82599(hw, adapter->fdir_pballoc);
3735 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3736 ixgbe_init_fdir_perfect_82599(hw, adapter->fdir_pballoc);
3738 ixgbe_configure_virtualization(adapter);
3740 ixgbe_configure_tx(adapter);
3741 ixgbe_configure_rx(adapter);
3744 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3746 switch (hw->phy.type) {
3747 case ixgbe_phy_sfp_avago:
3748 case ixgbe_phy_sfp_ftl:
3749 case ixgbe_phy_sfp_intel:
3750 case ixgbe_phy_sfp_unknown:
3751 case ixgbe_phy_sfp_passive_tyco:
3752 case ixgbe_phy_sfp_passive_unknown:
3753 case ixgbe_phy_sfp_active_unknown:
3754 case ixgbe_phy_sfp_ftl_active:
3762 * ixgbe_sfp_link_config - set up SFP+ link
3763 * @adapter: pointer to private adapter struct
3765 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3767 struct ixgbe_hw *hw = &adapter->hw;
3769 if (hw->phy.multispeed_fiber) {
3771 * In multispeed fiber setups, the device may not have
3772 * had a physical connection when the driver loaded.
3773 * If that's the case, the initial link configuration
3774 * couldn't get the MAC into 10G or 1G mode, so we'll
3775 * never have a link status change interrupt fire.
3776 * We need to try and force an autonegotiation
3777 * session, then bring up link.
3779 if (hw->mac.ops.setup_sfp)
3780 hw->mac.ops.setup_sfp(hw);
3781 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
3782 schedule_work(&adapter->multispeed_fiber_task);
3785 * Direct Attach Cu and non-multispeed fiber modules
3786 * still need to be configured properly prior to
3789 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_MOD_TASK))
3790 schedule_work(&adapter->sfp_config_module_task);
3795 * ixgbe_non_sfp_link_config - set up non-SFP+ link
3796 * @hw: pointer to private hardware struct
3798 * Returns 0 on success, negative on failure
3800 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3803 bool negotiation, link_up = false;
3804 u32 ret = IXGBE_ERR_LINK_SETUP;
3806 if (hw->mac.ops.check_link)
3807 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3812 autoneg = hw->phy.autoneg_advertised;
3813 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
3814 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
3819 if (hw->mac.ops.setup_link)
3820 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
3825 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
3827 struct ixgbe_hw *hw = &adapter->hw;
3830 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3831 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
3833 gpie |= IXGBE_GPIE_EIAME;
3835 * use EIAM to auto-mask when MSI-X interrupt is asserted
3836 * this saves a register write for every interrupt
3838 switch (hw->mac.type) {
3839 case ixgbe_mac_82598EB:
3840 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3842 case ixgbe_mac_82599EB:
3843 case ixgbe_mac_X540:
3845 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
3846 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
3850 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
3851 * specifically only auto mask tx and rx interrupts */
3852 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3855 /* XXX: to interrupt immediately for EICS writes, enable this */
3856 /* gpie |= IXGBE_GPIE_EIMEN; */
3858 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3859 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
3860 gpie |= IXGBE_GPIE_VTMODE_64;
3863 /* Enable fan failure interrupt */
3864 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3865 gpie |= IXGBE_SDP1_GPIEN;
3867 if (hw->mac.type == ixgbe_mac_82599EB)
3868 gpie |= IXGBE_SDP1_GPIEN;
3869 gpie |= IXGBE_SDP2_GPIEN;
3871 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3874 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
3876 struct ixgbe_hw *hw = &adapter->hw;
3880 ixgbe_get_hw_control(adapter);
3881 ixgbe_setup_gpie(adapter);
3883 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3884 ixgbe_configure_msix(adapter);
3886 ixgbe_configure_msi_and_legacy(adapter);
3888 /* enable the optics for both mult-speed fiber and 82599 SFP+ fiber */
3889 if (hw->mac.ops.enable_tx_laser &&
3890 ((hw->phy.multispeed_fiber) ||
3891 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
3892 (hw->mac.type == ixgbe_mac_82599EB))))
3893 hw->mac.ops.enable_tx_laser(hw);
3895 clear_bit(__IXGBE_DOWN, &adapter->state);
3896 ixgbe_napi_enable_all(adapter);
3898 if (ixgbe_is_sfp(hw)) {
3899 ixgbe_sfp_link_config(adapter);
3901 err = ixgbe_non_sfp_link_config(hw);
3903 e_err(probe, "link_config FAILED %d\n", err);
3906 /* clear any pending interrupts, may auto mask */
3907 IXGBE_READ_REG(hw, IXGBE_EICR);
3908 ixgbe_irq_enable(adapter, true, true);
3911 * If this adapter has a fan, check to see if we had a failure
3912 * before we enabled the interrupt.
3914 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3915 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
3916 if (esdp & IXGBE_ESDP_SDP1)
3917 e_crit(drv, "Fan has stopped, replace the adapter\n");
3921 * For hot-pluggable SFP+ devices, a new SFP+ module may have
3922 * arrived before interrupts were enabled but after probe. Such
3923 * devices wouldn't have their type identified yet. We need to
3924 * kick off the SFP+ module setup first, then try to bring up link.
3925 * If we're not hot-pluggable SFP+, we just need to configure link
3928 if (hw->phy.type == ixgbe_phy_none)
3929 schedule_work(&adapter->sfp_config_module_task);
3931 /* enable transmits */
3932 netif_tx_start_all_queues(adapter->netdev);
3934 /* bring the link up in the watchdog, this could race with our first
3935 * link up interrupt but shouldn't be a problem */
3936 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3937 adapter->link_check_timeout = jiffies;
3938 mod_timer(&adapter->watchdog_timer, jiffies);
3940 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
3941 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
3942 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
3943 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
3948 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
3950 WARN_ON(in_interrupt());
3951 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
3953 ixgbe_down(adapter);
3955 * If SR-IOV enabled then wait a bit before bringing the adapter
3956 * back up to give the VFs time to respond to the reset. The
3957 * two second wait is based upon the watchdog timer cycle in
3960 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3963 clear_bit(__IXGBE_RESETTING, &adapter->state);
3966 int ixgbe_up(struct ixgbe_adapter *adapter)
3968 /* hardware has been reset, we need to reload some things */
3969 ixgbe_configure(adapter);
3971 return ixgbe_up_complete(adapter);
3974 void ixgbe_reset(struct ixgbe_adapter *adapter)
3976 struct ixgbe_hw *hw = &adapter->hw;
3979 err = hw->mac.ops.init_hw(hw);
3982 case IXGBE_ERR_SFP_NOT_PRESENT:
3984 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
3985 e_dev_err("master disable timed out\n");
3987 case IXGBE_ERR_EEPROM_VERSION:
3988 /* We are running on a pre-production device, log a warning */
3989 e_dev_warn("This device is a pre-production adapter/LOM. "
3990 "Please be aware there may be issuesassociated with "
3991 "your hardware. If you are experiencing problems "
3992 "please contact your Intel or hardware "
3993 "representative who provided you with this "
3997 e_dev_err("Hardware Error: %d\n", err);
4000 /* reprogram the RAR[0] in case user changed it. */
4001 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
4006 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
4007 * @rx_ring: ring to free buffers from
4009 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
4011 struct device *dev = rx_ring->dev;
4015 /* ring already cleared, nothing to do */
4016 if (!rx_ring->rx_buffer_info)
4019 /* Free all the Rx ring sk_buffs */
4020 for (i = 0; i < rx_ring->count; i++) {
4021 struct ixgbe_rx_buffer *rx_buffer_info;
4023 rx_buffer_info = &rx_ring->rx_buffer_info[i];
4024 if (rx_buffer_info->dma) {
4025 dma_unmap_single(rx_ring->dev, rx_buffer_info->dma,
4026 rx_ring->rx_buf_len,
4028 rx_buffer_info->dma = 0;
4030 if (rx_buffer_info->skb) {
4031 struct sk_buff *skb = rx_buffer_info->skb;
4032 rx_buffer_info->skb = NULL;
4034 struct sk_buff *this = skb;
4035 if (IXGBE_RSC_CB(this)->delay_unmap) {
4036 dma_unmap_single(dev,
4037 IXGBE_RSC_CB(this)->dma,
4038 rx_ring->rx_buf_len,
4040 IXGBE_RSC_CB(this)->dma = 0;
4041 IXGBE_RSC_CB(skb)->delay_unmap = false;
4044 dev_kfree_skb(this);
4047 if (!rx_buffer_info->page)
4049 if (rx_buffer_info->page_dma) {
4050 dma_unmap_page(dev, rx_buffer_info->page_dma,
4051 PAGE_SIZE / 2, DMA_FROM_DEVICE);
4052 rx_buffer_info->page_dma = 0;
4054 put_page(rx_buffer_info->page);
4055 rx_buffer_info->page = NULL;
4056 rx_buffer_info->page_offset = 0;
4059 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4060 memset(rx_ring->rx_buffer_info, 0, size);
4062 /* Zero out the descriptor ring */
4063 memset(rx_ring->desc, 0, rx_ring->size);
4065 rx_ring->next_to_clean = 0;
4066 rx_ring->next_to_use = 0;
4070 * ixgbe_clean_tx_ring - Free Tx Buffers
4071 * @tx_ring: ring to be cleaned
4073 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
4075 struct ixgbe_tx_buffer *tx_buffer_info;
4079 /* ring already cleared, nothing to do */
4080 if (!tx_ring->tx_buffer_info)
4083 /* Free all the Tx ring sk_buffs */
4084 for (i = 0; i < tx_ring->count; i++) {
4085 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4086 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
4089 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4090 memset(tx_ring->tx_buffer_info, 0, size);
4092 /* Zero out the descriptor ring */
4093 memset(tx_ring->desc, 0, tx_ring->size);
4095 tx_ring->next_to_use = 0;
4096 tx_ring->next_to_clean = 0;
4100 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
4101 * @adapter: board private structure
4103 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
4107 for (i = 0; i < adapter->num_rx_queues; i++)
4108 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
4112 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
4113 * @adapter: board private structure
4115 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
4119 for (i = 0; i < adapter->num_tx_queues; i++)
4120 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
4123 void ixgbe_down(struct ixgbe_adapter *adapter)
4125 struct net_device *netdev = adapter->netdev;
4126 struct ixgbe_hw *hw = &adapter->hw;
4130 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4132 /* signal that we are down to the interrupt handler */
4133 set_bit(__IXGBE_DOWN, &adapter->state);
4135 /* disable receive for all VFs and wait one second */
4136 if (adapter->num_vfs) {
4137 /* ping all the active vfs to let them know we are going down */
4138 ixgbe_ping_all_vfs(adapter);
4140 /* Disable all VFTE/VFRE TX/RX */
4141 ixgbe_disable_tx_rx(adapter);
4143 /* Mark all the VFs as inactive */
4144 for (i = 0 ; i < adapter->num_vfs; i++)
4145 adapter->vfinfo[i].clear_to_send = 0;
4148 /* disable receives */
4149 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4150 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
4152 /* disable all enabled rx queues */
4153 for (i = 0; i < adapter->num_rx_queues; i++)
4154 /* this call also flushes the previous write */
4155 ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]);
4159 netif_tx_stop_all_queues(netdev);
4161 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
4162 del_timer_sync(&adapter->sfp_timer);
4163 del_timer_sync(&adapter->watchdog_timer);
4164 cancel_work_sync(&adapter->watchdog_task);
4166 netif_carrier_off(netdev);
4167 netif_tx_disable(netdev);
4169 ixgbe_irq_disable(adapter);
4171 ixgbe_napi_disable_all(adapter);
4173 /* Cleanup the affinity_hint CPU mask memory and callback */
4174 for (i = 0; i < num_q_vectors; i++) {
4175 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
4176 /* clear the affinity_mask in the IRQ descriptor */
4177 irq_set_affinity_hint(adapter->msix_entries[i]. vector, NULL);
4178 /* release the CPU mask memory */
4179 free_cpumask_var(q_vector->affinity_mask);
4182 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
4183 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
4184 cancel_work_sync(&adapter->fdir_reinit_task);
4186 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
4187 cancel_work_sync(&adapter->check_overtemp_task);
4189 /* disable transmits in the hardware now that interrupts are off */
4190 for (i = 0; i < adapter->num_tx_queues; i++) {
4191 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
4192 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
4193 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx),
4194 (txdctl & ~IXGBE_TXDCTL_ENABLE));
4196 /* Disable the Tx DMA engine on 82599 */
4197 switch (hw->mac.type) {
4198 case ixgbe_mac_82599EB:
4199 case ixgbe_mac_X540:
4200 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
4201 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
4202 ~IXGBE_DMATXCTL_TE));
4208 /* clear n-tuple filters that are cached */
4209 ethtool_ntuple_flush(netdev);
4211 if (!pci_channel_offline(adapter->pdev))
4212 ixgbe_reset(adapter);
4214 /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
4215 if (hw->mac.ops.disable_tx_laser &&
4216 ((hw->phy.multispeed_fiber) ||
4217 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
4218 (hw->mac.type == ixgbe_mac_82599EB))))
4219 hw->mac.ops.disable_tx_laser(hw);
4221 ixgbe_clean_all_tx_rings(adapter);
4222 ixgbe_clean_all_rx_rings(adapter);
4224 #ifdef CONFIG_IXGBE_DCA
4225 /* since we reset the hardware DCA settings were cleared */
4226 ixgbe_setup_dca(adapter);
4231 * ixgbe_poll - NAPI Rx polling callback
4232 * @napi: structure for representing this polling device
4233 * @budget: how many packets driver is allowed to clean
4235 * This function is used for legacy and MSI, NAPI mode
4237 static int ixgbe_poll(struct napi_struct *napi, int budget)
4239 struct ixgbe_q_vector *q_vector =
4240 container_of(napi, struct ixgbe_q_vector, napi);
4241 struct ixgbe_adapter *adapter = q_vector->adapter;
4242 int tx_clean_complete, work_done = 0;
4244 #ifdef CONFIG_IXGBE_DCA
4245 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
4246 ixgbe_update_dca(q_vector);
4249 tx_clean_complete = ixgbe_clean_tx_irq(q_vector, adapter->tx_ring[0]);
4250 ixgbe_clean_rx_irq(q_vector, adapter->rx_ring[0], &work_done, budget);
4252 if (!tx_clean_complete)
4255 /* If budget not fully consumed, exit the polling mode */
4256 if (work_done < budget) {
4257 napi_complete(napi);
4258 if (adapter->rx_itr_setting & 1)
4259 ixgbe_set_itr(adapter);
4260 if (!test_bit(__IXGBE_DOWN, &adapter->state))
4261 ixgbe_irq_enable_queues(adapter, IXGBE_EIMS_RTX_QUEUE);
4267 * ixgbe_tx_timeout - Respond to a Tx Hang
4268 * @netdev: network interface device structure
4270 static void ixgbe_tx_timeout(struct net_device *netdev)
4272 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4274 adapter->tx_timeout_count++;
4276 /* Do the reset outside of interrupt context */
4277 schedule_work(&adapter->reset_task);
4280 static void ixgbe_reset_task(struct work_struct *work)
4282 struct ixgbe_adapter *adapter;
4283 adapter = container_of(work, struct ixgbe_adapter, reset_task);
4285 /* If we're already down or resetting, just bail */
4286 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
4287 test_bit(__IXGBE_RESETTING, &adapter->state))
4290 ixgbe_dump(adapter);
4291 netdev_err(adapter->netdev, "Reset adapter\n");
4292 ixgbe_reinit_locked(adapter);
4296 * ixgbe_set_rss_queues: Allocate queues for RSS
4297 * @adapter: board private structure to initialize
4299 * This is our "base" multiqueue mode. RSS (Receive Side Scaling) will try
4300 * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
4303 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
4306 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
4308 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4310 adapter->num_rx_queues = f->indices;
4311 adapter->num_tx_queues = f->indices;
4321 * ixgbe_set_fdir_queues: Allocate queues for Flow Director
4322 * @adapter: board private structure to initialize
4324 * Flow Director is an advanced Rx filter, attempting to get Rx flows back
4325 * to the original CPU that initiated the Tx session. This runs in addition
4326 * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
4327 * Rx load across CPUs using RSS.
4330 static inline bool ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
4333 struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
4335 f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
4338 /* Flow Director must have RSS enabled */
4339 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
4340 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
4341 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)))) {
4342 adapter->num_tx_queues = f_fdir->indices;
4343 adapter->num_rx_queues = f_fdir->indices;
4346 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4347 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4354 * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
4355 * @adapter: board private structure to initialize
4357 * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
4358 * The ring feature mask is not used as a mask for FCoE, as it can take any 8
4359 * rx queues out of the max number of rx queues, instead, it is used as the
4360 * index of the first rx queue used by FCoE.
4363 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
4365 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4367 if (!(adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
4370 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4371 #ifdef CONFIG_IXGBE_DCB
4373 struct net_device *dev = adapter->netdev;
4375 tc = netdev_get_prio_tc_map(dev, adapter->fcoe.up);
4376 f->indices = dev->tc_to_txq[tc].count;
4377 f->mask = dev->tc_to_txq[tc].offset;
4380 f->indices = min((int)num_online_cpus(), f->indices);
4382 adapter->num_rx_queues = 1;
4383 adapter->num_tx_queues = 1;
4385 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4386 e_info(probe, "FCoE enabled with RSS\n");
4387 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4388 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
4389 ixgbe_set_fdir_queues(adapter);
4391 ixgbe_set_rss_queues(adapter);
4393 /* adding FCoE rx rings to the end */
4394 f->mask = adapter->num_rx_queues;
4395 adapter->num_rx_queues += f->indices;
4396 adapter->num_tx_queues += f->indices;
4401 #endif /* IXGBE_FCOE */
4403 #ifdef CONFIG_IXGBE_DCB
4404 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
4407 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_DCB];
4410 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
4414 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
4415 q = min((int)num_online_cpus(), MAX_TRAFFIC_CLASS);
4420 adapter->num_rx_queues = f->indices;
4421 adapter->num_tx_queues = f->indices;
4425 /* FCoE enabled queues require special configuration done through
4426 * configure_fcoe() and others. Here we map FCoE indices onto the
4427 * DCB queue pairs allowing FCoE to own configuration later.
4429 ixgbe_set_fcoe_queues(adapter);
4437 * ixgbe_set_sriov_queues: Allocate queues for IOV use
4438 * @adapter: board private structure to initialize
4440 * IOV doesn't actually use anything, so just NAK the
4441 * request for now and let the other queue routines
4442 * figure out what to do.
4444 static inline bool ixgbe_set_sriov_queues(struct ixgbe_adapter *adapter)
4450 * ixgbe_set_num_queues: Allocate queues for device, feature dependent
4451 * @adapter: board private structure to initialize
4453 * This is the top level queue allocation routine. The order here is very
4454 * important, starting with the "most" number of features turned on at once,
4455 * and ending with the smallest set of features. This way large combinations
4456 * can be allocated if they're turned on, and smaller combinations are the
4457 * fallthrough conditions.
4460 static int ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
4462 /* Start with base case */
4463 adapter->num_rx_queues = 1;
4464 adapter->num_tx_queues = 1;
4465 adapter->num_rx_pools = adapter->num_rx_queues;
4466 adapter->num_rx_queues_per_pool = 1;
4468 if (ixgbe_set_sriov_queues(adapter))
4471 #ifdef CONFIG_IXGBE_DCB
4472 if (ixgbe_set_dcb_queues(adapter))
4477 if (ixgbe_set_fcoe_queues(adapter))
4480 #endif /* IXGBE_FCOE */
4481 if (ixgbe_set_fdir_queues(adapter))
4484 if (ixgbe_set_rss_queues(adapter))
4487 /* fallback to base case */
4488 adapter->num_rx_queues = 1;
4489 adapter->num_tx_queues = 1;
4492 /* Notify the stack of the (possibly) reduced queue counts. */
4493 netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
4494 return netif_set_real_num_rx_queues(adapter->netdev,
4495 adapter->num_rx_queues);
4498 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
4501 int err, vector_threshold;
4503 /* We'll want at least 3 (vector_threshold):
4506 * 3) Other (Link Status Change, etc.)
4507 * 4) TCP Timer (optional)
4509 vector_threshold = MIN_MSIX_COUNT;
4511 /* The more we get, the more we will assign to Tx/Rx Cleanup
4512 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
4513 * Right now, we simply care about how many we'll get; we'll
4514 * set them up later while requesting irq's.
4516 while (vectors >= vector_threshold) {
4517 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
4519 if (!err) /* Success in acquiring all requested vectors. */
4522 vectors = 0; /* Nasty failure, quit now */
4523 else /* err == number of vectors we should try again with */
4527 if (vectors < vector_threshold) {
4528 /* Can't allocate enough MSI-X interrupts? Oh well.
4529 * This just means we'll go with either a single MSI
4530 * vector or fall back to legacy interrupts.
4532 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4533 "Unable to allocate MSI-X interrupts\n");
4534 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4535 kfree(adapter->msix_entries);
4536 adapter->msix_entries = NULL;
4538 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
4540 * Adjust for only the vectors we'll use, which is minimum
4541 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
4542 * vectors we were allocated.
4544 adapter->num_msix_vectors = min(vectors,
4545 adapter->max_msix_q_vectors + NON_Q_VECTORS);
4550 * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
4551 * @adapter: board private structure to initialize
4553 * Cache the descriptor ring offsets for RSS to the assigned rings.
4556 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
4560 if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED))
4563 for (i = 0; i < adapter->num_rx_queues; i++)
4564 adapter->rx_ring[i]->reg_idx = i;
4565 for (i = 0; i < adapter->num_tx_queues; i++)
4566 adapter->tx_ring[i]->reg_idx = i;
4571 #ifdef CONFIG_IXGBE_DCB
4573 /* ixgbe_get_first_reg_idx - Return first register index associated with ring */
4574 void ixgbe_get_first_reg_idx(struct ixgbe_adapter *adapter, u8 tc,
4575 unsigned int *tx, unsigned int *rx)
4577 struct net_device *dev = adapter->netdev;
4578 struct ixgbe_hw *hw = &adapter->hw;
4579 u8 num_tcs = netdev_get_num_tc(dev);
4584 switch (hw->mac.type) {
4585 case ixgbe_mac_82598EB:
4589 case ixgbe_mac_82599EB:
4590 case ixgbe_mac_X540:
4595 } else if (tc < 5) {
4596 *tx = ((tc + 2) << 4);
4598 } else if (tc < num_tcs) {
4599 *tx = ((tc + 8) << 3);
4602 } else if (num_tcs == 4) {
4627 #define IXGBE_MAX_Q_PER_TC (IXGBE_MAX_DCB_INDICES / MAX_TRAFFIC_CLASS)
4629 /* ixgbe_setup_tc - routine to configure net_device for multiple traffic
4632 * @netdev: net device to configure
4633 * @tc: number of traffic classes to enable
4635 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
4638 unsigned int q, offset = 0;
4641 netdev_reset_tc(dev);
4643 struct ixgbe_adapter *adapter = netdev_priv(dev);
4645 /* Hardware supports up to 8 traffic classes */
4646 if (tc > MAX_TRAFFIC_CLASS || netdev_set_num_tc(dev, tc))
4649 /* Partition Tx queues evenly amongst traffic classes */
4650 for (i = 0; i < tc; i++) {
4651 q = min((int)num_online_cpus(), IXGBE_MAX_Q_PER_TC);
4652 netdev_set_prio_tc_map(dev, i, i);
4653 netdev_set_tc_queue(dev, i, q, offset);
4657 /* This enables multiple traffic class support in the hardware
4658 * which defaults to strict priority transmission by default.
4659 * If traffic classes are already enabled perhaps through DCB
4660 * code path then existing configuration will be used.
4662 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
4663 dev->dcbnl_ops && dev->dcbnl_ops->setdcbx) {
4664 struct ieee_ets ets = {
4665 .prio_tc = {0, 1, 2, 3, 4, 5, 6, 7},
4667 u8 mode = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_IEEE;
4669 dev->dcbnl_ops->setdcbx(dev, mode);
4670 dev->dcbnl_ops->ieee_setets(dev, &ets);
4677 * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
4678 * @adapter: board private structure to initialize
4680 * Cache the descriptor ring offsets for DCB to the assigned rings.
4683 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
4685 struct net_device *dev = adapter->netdev;
4687 u8 num_tcs = netdev_get_num_tc(dev);
4689 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
4692 for (i = 0, k = 0; i < num_tcs; i++) {
4693 unsigned int tx_s, rx_s;
4694 u16 count = dev->tc_to_txq[i].count;
4696 ixgbe_get_first_reg_idx(adapter, i, &tx_s, &rx_s);
4697 for (j = 0; j < count; j++, k++) {
4698 adapter->tx_ring[k]->reg_idx = tx_s + j;
4699 adapter->rx_ring[k]->reg_idx = rx_s + j;
4700 adapter->tx_ring[k]->dcb_tc = i;
4701 adapter->rx_ring[k]->dcb_tc = i;
4710 * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
4711 * @adapter: board private structure to initialize
4713 * Cache the descriptor ring offsets for Flow Director to the assigned rings.
4716 static inline bool ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
4721 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
4722 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4723 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))) {
4724 for (i = 0; i < adapter->num_rx_queues; i++)
4725 adapter->rx_ring[i]->reg_idx = i;
4726 for (i = 0; i < adapter->num_tx_queues; i++)
4727 adapter->tx_ring[i]->reg_idx = i;
4736 * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
4737 * @adapter: board private structure to initialize
4739 * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
4742 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
4744 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4746 u8 fcoe_rx_i = 0, fcoe_tx_i = 0;
4748 if (!(adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
4751 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4752 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4753 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
4754 ixgbe_cache_ring_fdir(adapter);
4756 ixgbe_cache_ring_rss(adapter);
4758 fcoe_rx_i = f->mask;
4759 fcoe_tx_i = f->mask;
4761 for (i = 0; i < f->indices; i++, fcoe_rx_i++, fcoe_tx_i++) {
4762 adapter->rx_ring[f->mask + i]->reg_idx = fcoe_rx_i;
4763 adapter->tx_ring[f->mask + i]->reg_idx = fcoe_tx_i;
4768 #endif /* IXGBE_FCOE */
4770 * ixgbe_cache_ring_sriov - Descriptor ring to register mapping for sriov
4771 * @adapter: board private structure to initialize
4773 * SR-IOV doesn't use any descriptor rings but changes the default if
4774 * no other mapping is used.
4777 static inline bool ixgbe_cache_ring_sriov(struct ixgbe_adapter *adapter)
4779 adapter->rx_ring[0]->reg_idx = adapter->num_vfs * 2;
4780 adapter->tx_ring[0]->reg_idx = adapter->num_vfs * 2;
4781 if (adapter->num_vfs)
4788 * ixgbe_cache_ring_register - Descriptor ring to register mapping
4789 * @adapter: board private structure to initialize
4791 * Once we know the feature-set enabled for the device, we'll cache
4792 * the register offset the descriptor ring is assigned to.
4794 * Note, the order the various feature calls is important. It must start with
4795 * the "most" features enabled at the same time, then trickle down to the
4796 * least amount of features turned on at once.
4798 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
4800 /* start with default case */
4801 adapter->rx_ring[0]->reg_idx = 0;
4802 adapter->tx_ring[0]->reg_idx = 0;
4804 if (ixgbe_cache_ring_sriov(adapter))
4807 #ifdef CONFIG_IXGBE_DCB
4808 if (ixgbe_cache_ring_dcb(adapter))
4813 if (ixgbe_cache_ring_fcoe(adapter))
4815 #endif /* IXGBE_FCOE */
4817 if (ixgbe_cache_ring_fdir(adapter))
4820 if (ixgbe_cache_ring_rss(adapter))
4825 * ixgbe_alloc_queues - Allocate memory for all rings
4826 * @adapter: board private structure to initialize
4828 * We allocate one ring per queue at run-time since we don't know the
4829 * number of queues at compile-time. The polling_netdev array is
4830 * intended for Multiqueue, but should work fine with a single queue.
4832 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
4834 int rx = 0, tx = 0, nid = adapter->node;
4836 if (nid < 0 || !node_online(nid))
4837 nid = first_online_node;
4839 for (; tx < adapter->num_tx_queues; tx++) {
4840 struct ixgbe_ring *ring;
4842 ring = kzalloc_node(sizeof(*ring), GFP_KERNEL, nid);
4844 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
4846 goto err_allocation;
4847 ring->count = adapter->tx_ring_count;
4848 ring->queue_index = tx;
4849 ring->numa_node = nid;
4850 ring->dev = &adapter->pdev->dev;
4851 ring->netdev = adapter->netdev;
4853 adapter->tx_ring[tx] = ring;
4856 for (; rx < adapter->num_rx_queues; rx++) {
4857 struct ixgbe_ring *ring;
4859 ring = kzalloc_node(sizeof(*ring), GFP_KERNEL, nid);
4861 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
4863 goto err_allocation;
4864 ring->count = adapter->rx_ring_count;
4865 ring->queue_index = rx;
4866 ring->numa_node = nid;
4867 ring->dev = &adapter->pdev->dev;
4868 ring->netdev = adapter->netdev;
4870 adapter->rx_ring[rx] = ring;
4873 ixgbe_cache_ring_register(adapter);
4879 kfree(adapter->tx_ring[--tx]);
4882 kfree(adapter->rx_ring[--rx]);
4887 * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
4888 * @adapter: board private structure to initialize
4890 * Attempt to configure the interrupts using the best available
4891 * capabilities of the hardware and the kernel.
4893 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
4895 struct ixgbe_hw *hw = &adapter->hw;
4897 int vector, v_budget;
4900 * It's easy to be greedy for MSI-X vectors, but it really
4901 * doesn't do us much good if we have a lot more vectors
4902 * than CPU's. So let's be conservative and only ask for
4903 * (roughly) the same number of vectors as there are CPU's.
4905 v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
4906 (int)num_online_cpus()) + NON_Q_VECTORS;
4909 * At the same time, hardware can only support a maximum of
4910 * hw.mac->max_msix_vectors vectors. With features
4911 * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
4912 * descriptor queues supported by our device. Thus, we cap it off in
4913 * those rare cases where the cpu count also exceeds our vector limit.
4915 v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
4917 /* A failure in MSI-X entry allocation isn't fatal, but it does
4918 * mean we disable MSI-X capabilities of the adapter. */
4919 adapter->msix_entries = kcalloc(v_budget,
4920 sizeof(struct msix_entry), GFP_KERNEL);
4921 if (adapter->msix_entries) {
4922 for (vector = 0; vector < v_budget; vector++)
4923 adapter->msix_entries[vector].entry = vector;
4925 ixgbe_acquire_msix_vectors(adapter, v_budget);
4927 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4931 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
4932 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
4933 if (adapter->flags & (IXGBE_FLAG_FDIR_HASH_CAPABLE |
4934 IXGBE_FLAG_FDIR_PERFECT_CAPABLE)) {
4936 "Flow Director is not supported while multiple "
4937 "queues are disabled. Disabling Flow Director\n");
4939 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4940 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4941 adapter->atr_sample_rate = 0;
4942 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4943 ixgbe_disable_sriov(adapter);
4945 err = ixgbe_set_num_queues(adapter);
4949 err = pci_enable_msi(adapter->pdev);
4951 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
4953 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4954 "Unable to allocate MSI interrupt, "
4955 "falling back to legacy. Error: %d\n", err);
4965 * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
4966 * @adapter: board private structure to initialize
4968 * We allocate one q_vector per queue interrupt. If allocation fails we
4971 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
4973 int q_idx, num_q_vectors;
4974 struct ixgbe_q_vector *q_vector;
4975 int (*poll)(struct napi_struct *, int);
4977 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4978 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4979 poll = &ixgbe_clean_rxtx_many;
4985 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4986 q_vector = kzalloc_node(sizeof(struct ixgbe_q_vector),
4987 GFP_KERNEL, adapter->node);
4989 q_vector = kzalloc(sizeof(struct ixgbe_q_vector),
4993 q_vector->adapter = adapter;
4994 if (q_vector->txr_count && !q_vector->rxr_count)
4995 q_vector->eitr = adapter->tx_eitr_param;
4997 q_vector->eitr = adapter->rx_eitr_param;
4998 q_vector->v_idx = q_idx;
4999 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
5000 adapter->q_vector[q_idx] = q_vector;
5008 q_vector = adapter->q_vector[q_idx];
5009 netif_napi_del(&q_vector->napi);
5011 adapter->q_vector[q_idx] = NULL;
5017 * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
5018 * @adapter: board private structure to initialize
5020 * This function frees the memory allocated to the q_vectors. In addition if
5021 * NAPI is enabled it will delete any references to the NAPI struct prior
5022 * to freeing the q_vector.
5024 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
5026 int q_idx, num_q_vectors;
5028 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
5029 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
5033 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
5034 struct ixgbe_q_vector *q_vector = adapter->q_vector[q_idx];
5035 adapter->q_vector[q_idx] = NULL;
5036 netif_napi_del(&q_vector->napi);
5041 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
5043 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
5044 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
5045 pci_disable_msix(adapter->pdev);
5046 kfree(adapter->msix_entries);
5047 adapter->msix_entries = NULL;
5048 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
5049 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
5050 pci_disable_msi(adapter->pdev);
5055 * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
5056 * @adapter: board private structure to initialize
5058 * We determine which interrupt scheme to use based on...
5059 * - Kernel support (MSI, MSI-X)
5060 * - which can be user-defined (via MODULE_PARAM)
5061 * - Hardware queue count (num_*_queues)
5062 * - defined by miscellaneous hardware support/features (RSS, etc.)
5064 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
5068 /* Number of supported queues */
5069 err = ixgbe_set_num_queues(adapter);
5073 err = ixgbe_set_interrupt_capability(adapter);
5075 e_dev_err("Unable to setup interrupt capabilities\n");
5076 goto err_set_interrupt;
5079 err = ixgbe_alloc_q_vectors(adapter);
5081 e_dev_err("Unable to allocate memory for queue vectors\n");
5082 goto err_alloc_q_vectors;
5085 err = ixgbe_alloc_queues(adapter);
5087 e_dev_err("Unable to allocate memory for queues\n");
5088 goto err_alloc_queues;
5091 e_dev_info("Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
5092 (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
5093 adapter->num_rx_queues, adapter->num_tx_queues);
5095 set_bit(__IXGBE_DOWN, &adapter->state);
5100 ixgbe_free_q_vectors(adapter);
5101 err_alloc_q_vectors:
5102 ixgbe_reset_interrupt_capability(adapter);
5107 static void ring_free_rcu(struct rcu_head *head)
5109 kfree(container_of(head, struct ixgbe_ring, rcu));
5113 * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
5114 * @adapter: board private structure to clear interrupt scheme on
5116 * We go through and clear interrupt specific resources and reset the structure
5117 * to pre-load conditions
5119 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
5123 for (i = 0; i < adapter->num_tx_queues; i++) {
5124 kfree(adapter->tx_ring[i]);
5125 adapter->tx_ring[i] = NULL;
5127 for (i = 0; i < adapter->num_rx_queues; i++) {
5128 struct ixgbe_ring *ring = adapter->rx_ring[i];
5130 /* ixgbe_get_stats64() might access this ring, we must wait
5131 * a grace period before freeing it.
5133 call_rcu(&ring->rcu, ring_free_rcu);
5134 adapter->rx_ring[i] = NULL;
5137 adapter->num_tx_queues = 0;
5138 adapter->num_rx_queues = 0;
5140 ixgbe_free_q_vectors(adapter);
5141 ixgbe_reset_interrupt_capability(adapter);
5145 * ixgbe_sfp_timer - worker thread to find a missing module
5146 * @data: pointer to our adapter struct
5148 static void ixgbe_sfp_timer(unsigned long data)
5150 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5153 * Do the sfp_timer outside of interrupt context due to the
5154 * delays that sfp+ detection requires
5156 schedule_work(&adapter->sfp_task);
5160 * ixgbe_sfp_task - worker thread to find a missing module
5161 * @work: pointer to work_struct containing our data
5163 static void ixgbe_sfp_task(struct work_struct *work)
5165 struct ixgbe_adapter *adapter = container_of(work,
5166 struct ixgbe_adapter,
5168 struct ixgbe_hw *hw = &adapter->hw;
5170 if ((hw->phy.type == ixgbe_phy_nl) &&
5171 (hw->phy.sfp_type == ixgbe_sfp_type_not_present)) {
5172 s32 ret = hw->phy.ops.identify_sfp(hw);
5173 if (ret == IXGBE_ERR_SFP_NOT_PRESENT)
5175 ret = hw->phy.ops.reset(hw);
5176 if (ret == IXGBE_ERR_SFP_NOT_SUPPORTED) {
5177 e_dev_err("failed to initialize because an unsupported "
5178 "SFP+ module type was detected.\n");
5179 e_dev_err("Reload the driver after installing a "
5180 "supported module.\n");
5181 unregister_netdev(adapter->netdev);
5183 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
5185 /* don't need this routine any more */
5186 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
5190 if (test_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state))
5191 mod_timer(&adapter->sfp_timer,
5192 round_jiffies(jiffies + (2 * HZ)));
5196 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
5197 * @adapter: board private structure to initialize
5199 * ixgbe_sw_init initializes the Adapter private data structure.
5200 * Fields are initialized based on PCI device information and
5201 * OS network device settings (MTU size).
5203 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
5205 struct ixgbe_hw *hw = &adapter->hw;
5206 struct pci_dev *pdev = adapter->pdev;
5207 struct net_device *dev = adapter->netdev;
5209 #ifdef CONFIG_IXGBE_DCB
5211 struct tc_configuration *tc;
5213 int max_frame = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
5215 /* PCI config space info */
5217 hw->vendor_id = pdev->vendor;
5218 hw->device_id = pdev->device;
5219 hw->revision_id = pdev->revision;
5220 hw->subsystem_vendor_id = pdev->subsystem_vendor;
5221 hw->subsystem_device_id = pdev->subsystem_device;
5223 /* Set capability flags */
5224 rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
5225 adapter->ring_feature[RING_F_RSS].indices = rss;
5226 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
5227 adapter->ring_feature[RING_F_DCB].indices = IXGBE_MAX_DCB_INDICES;
5228 switch (hw->mac.type) {
5229 case ixgbe_mac_82598EB:
5230 if (hw->device_id == IXGBE_DEV_ID_82598AT)
5231 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
5232 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
5234 case ixgbe_mac_82599EB:
5235 case ixgbe_mac_X540:
5236 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
5237 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
5238 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
5239 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
5240 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
5241 /* n-tuple support exists, always init our spinlock */
5242 spin_lock_init(&adapter->fdir_perfect_lock);
5243 /* Flow Director hash filters enabled */
5244 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
5245 adapter->atr_sample_rate = 20;
5246 adapter->ring_feature[RING_F_FDIR].indices =
5247 IXGBE_MAX_FDIR_INDICES;
5248 adapter->fdir_pballoc = 0;
5250 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
5251 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
5252 adapter->ring_feature[RING_F_FCOE].indices = 0;
5253 #ifdef CONFIG_IXGBE_DCB
5254 /* Default traffic class to use for FCoE */
5255 adapter->fcoe.tc = IXGBE_FCOE_DEFTC;
5256 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
5258 #endif /* IXGBE_FCOE */
5264 #ifdef CONFIG_IXGBE_DCB
5265 /* Configure DCB traffic classes */
5266 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
5267 tc = &adapter->dcb_cfg.tc_config[j];
5268 tc->path[DCB_TX_CONFIG].bwg_id = 0;
5269 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
5270 tc->path[DCB_RX_CONFIG].bwg_id = 0;
5271 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
5272 tc->dcb_pfc = pfc_disabled;
5274 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
5275 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
5276 adapter->dcb_cfg.rx_pba_cfg = pba_equal;
5277 adapter->dcb_cfg.pfc_mode_enable = false;
5278 adapter->dcb_set_bitmap = 0x00;
5279 adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
5280 ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
5285 /* default flow control settings */
5286 hw->fc.requested_mode = ixgbe_fc_full;
5287 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
5289 adapter->last_lfc_mode = hw->fc.current_mode;
5291 hw->fc.high_water = FC_HIGH_WATER(max_frame);
5292 hw->fc.low_water = FC_LOW_WATER(max_frame);
5293 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
5294 hw->fc.send_xon = true;
5295 hw->fc.disable_fc_autoneg = false;
5297 /* enable itr by default in dynamic mode */
5298 adapter->rx_itr_setting = 1;
5299 adapter->rx_eitr_param = 20000;
5300 adapter->tx_itr_setting = 1;
5301 adapter->tx_eitr_param = 10000;
5303 /* set defaults for eitr in MegaBytes */
5304 adapter->eitr_low = 10;
5305 adapter->eitr_high = 20;
5307 /* set default ring sizes */
5308 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
5309 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
5311 /* initialize eeprom parameters */
5312 if (ixgbe_init_eeprom_params_generic(hw)) {
5313 e_dev_err("EEPROM initialization failed\n");
5317 /* enable rx csum by default */
5318 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
5320 /* get assigned NUMA node */
5321 adapter->node = dev_to_node(&pdev->dev);
5323 set_bit(__IXGBE_DOWN, &adapter->state);
5329 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
5330 * @tx_ring: tx descriptor ring (for a specific queue) to setup
5332 * Return 0 on success, negative on failure
5334 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
5336 struct device *dev = tx_ring->dev;
5339 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
5340 tx_ring->tx_buffer_info = vzalloc_node(size, tx_ring->numa_node);
5341 if (!tx_ring->tx_buffer_info)
5342 tx_ring->tx_buffer_info = vzalloc(size);
5343 if (!tx_ring->tx_buffer_info)
5346 /* round up to nearest 4K */
5347 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
5348 tx_ring->size = ALIGN(tx_ring->size, 4096);
5350 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
5351 &tx_ring->dma, GFP_KERNEL);
5355 tx_ring->next_to_use = 0;
5356 tx_ring->next_to_clean = 0;
5357 tx_ring->work_limit = tx_ring->count;
5361 vfree(tx_ring->tx_buffer_info);
5362 tx_ring->tx_buffer_info = NULL;
5363 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
5368 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
5369 * @adapter: board private structure
5371 * If this function returns with an error, then it's possible one or
5372 * more of the rings is populated (while the rest are not). It is the
5373 * callers duty to clean those orphaned rings.
5375 * Return 0 on success, negative on failure
5377 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
5381 for (i = 0; i < adapter->num_tx_queues; i++) {
5382 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
5385 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
5393 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
5394 * @rx_ring: rx descriptor ring (for a specific queue) to setup
5396 * Returns 0 on success, negative on failure
5398 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
5400 struct device *dev = rx_ring->dev;
5403 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
5404 rx_ring->rx_buffer_info = vzalloc_node(size, rx_ring->numa_node);
5405 if (!rx_ring->rx_buffer_info)
5406 rx_ring->rx_buffer_info = vzalloc(size);
5407 if (!rx_ring->rx_buffer_info)
5410 /* Round up to nearest 4K */
5411 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
5412 rx_ring->size = ALIGN(rx_ring->size, 4096);
5414 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
5415 &rx_ring->dma, GFP_KERNEL);
5420 rx_ring->next_to_clean = 0;
5421 rx_ring->next_to_use = 0;
5425 vfree(rx_ring->rx_buffer_info);
5426 rx_ring->rx_buffer_info = NULL;
5427 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
5432 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
5433 * @adapter: board private structure
5435 * If this function returns with an error, then it's possible one or
5436 * more of the rings is populated (while the rest are not). It is the
5437 * callers duty to clean those orphaned rings.
5439 * Return 0 on success, negative on failure
5441 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
5445 for (i = 0; i < adapter->num_rx_queues; i++) {
5446 err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
5449 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
5457 * ixgbe_free_tx_resources - Free Tx Resources per Queue
5458 * @tx_ring: Tx descriptor ring for a specific queue
5460 * Free all transmit software resources
5462 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
5464 ixgbe_clean_tx_ring(tx_ring);
5466 vfree(tx_ring->tx_buffer_info);
5467 tx_ring->tx_buffer_info = NULL;
5469 /* if not set, then don't free */
5473 dma_free_coherent(tx_ring->dev, tx_ring->size,
5474 tx_ring->desc, tx_ring->dma);
5476 tx_ring->desc = NULL;
5480 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5481 * @adapter: board private structure
5483 * Free all transmit software resources
5485 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
5489 for (i = 0; i < adapter->num_tx_queues; i++)
5490 if (adapter->tx_ring[i]->desc)
5491 ixgbe_free_tx_resources(adapter->tx_ring[i]);
5495 * ixgbe_free_rx_resources - Free Rx Resources
5496 * @rx_ring: ring to clean the resources from
5498 * Free all receive software resources
5500 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
5502 ixgbe_clean_rx_ring(rx_ring);
5504 vfree(rx_ring->rx_buffer_info);
5505 rx_ring->rx_buffer_info = NULL;
5507 /* if not set, then don't free */
5511 dma_free_coherent(rx_ring->dev, rx_ring->size,
5512 rx_ring->desc, rx_ring->dma);
5514 rx_ring->desc = NULL;
5518 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5519 * @adapter: board private structure
5521 * Free all receive software resources
5523 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5527 for (i = 0; i < adapter->num_rx_queues; i++)
5528 if (adapter->rx_ring[i]->desc)
5529 ixgbe_free_rx_resources(adapter->rx_ring[i]);
5533 * ixgbe_change_mtu - Change the Maximum Transfer Unit
5534 * @netdev: network interface device structure
5535 * @new_mtu: new value for maximum frame size
5537 * Returns 0 on success, negative on failure
5539 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5541 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5542 struct ixgbe_hw *hw = &adapter->hw;
5543 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5545 /* MTU < 68 is an error and causes problems on some kernels */
5546 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED &&
5547 hw->mac.type != ixgbe_mac_X540) {
5548 if ((new_mtu < 68) || (max_frame > MAXIMUM_ETHERNET_VLAN_SIZE))
5551 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5555 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5556 /* must set new MTU before calling down or up */
5557 netdev->mtu = new_mtu;
5559 hw->fc.high_water = FC_HIGH_WATER(max_frame);
5560 hw->fc.low_water = FC_LOW_WATER(max_frame);
5562 if (netif_running(netdev))
5563 ixgbe_reinit_locked(adapter);
5569 * ixgbe_open - Called when a network interface is made active
5570 * @netdev: network interface device structure
5572 * Returns 0 on success, negative value on failure
5574 * The open entry point is called when a network interface is made
5575 * active by the system (IFF_UP). At this point all resources needed
5576 * for transmit and receive operations are allocated, the interrupt
5577 * handler is registered with the OS, the watchdog timer is started,
5578 * and the stack is notified that the interface is ready.
5580 static int ixgbe_open(struct net_device *netdev)
5582 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5585 /* disallow open during test */
5586 if (test_bit(__IXGBE_TESTING, &adapter->state))
5589 netif_carrier_off(netdev);
5591 /* allocate transmit descriptors */
5592 err = ixgbe_setup_all_tx_resources(adapter);
5596 /* allocate receive descriptors */
5597 err = ixgbe_setup_all_rx_resources(adapter);
5601 ixgbe_configure(adapter);
5603 err = ixgbe_request_irq(adapter);
5607 err = ixgbe_up_complete(adapter);
5611 netif_tx_start_all_queues(netdev);
5616 ixgbe_release_hw_control(adapter);
5617 ixgbe_free_irq(adapter);
5620 ixgbe_free_all_rx_resources(adapter);
5622 ixgbe_free_all_tx_resources(adapter);
5623 ixgbe_reset(adapter);
5629 * ixgbe_close - Disables a network interface
5630 * @netdev: network interface device structure
5632 * Returns 0, this is not allowed to fail
5634 * The close entry point is called when an interface is de-activated
5635 * by the OS. The hardware is still under the drivers control, but
5636 * needs to be disabled. A global MAC reset is issued to stop the
5637 * hardware, and all transmit and receive resources are freed.
5639 static int ixgbe_close(struct net_device *netdev)
5641 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5643 ixgbe_down(adapter);
5644 ixgbe_free_irq(adapter);
5646 ixgbe_free_all_tx_resources(adapter);
5647 ixgbe_free_all_rx_resources(adapter);
5649 ixgbe_release_hw_control(adapter);
5655 static int ixgbe_resume(struct pci_dev *pdev)
5657 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5658 struct net_device *netdev = adapter->netdev;
5661 pci_set_power_state(pdev, PCI_D0);
5662 pci_restore_state(pdev);
5664 * pci_restore_state clears dev->state_saved so call
5665 * pci_save_state to restore it.
5667 pci_save_state(pdev);
5669 err = pci_enable_device_mem(pdev);
5671 e_dev_err("Cannot enable PCI device from suspend\n");
5674 pci_set_master(pdev);
5676 pci_wake_from_d3(pdev, false);
5678 err = ixgbe_init_interrupt_scheme(adapter);
5680 e_dev_err("Cannot initialize interrupts for device\n");
5684 ixgbe_reset(adapter);
5686 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5688 if (netif_running(netdev)) {
5689 err = ixgbe_open(netdev);
5694 netif_device_attach(netdev);
5698 #endif /* CONFIG_PM */
5700 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5702 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5703 struct net_device *netdev = adapter->netdev;
5704 struct ixgbe_hw *hw = &adapter->hw;
5706 u32 wufc = adapter->wol;
5711 netif_device_detach(netdev);
5713 if (netif_running(netdev)) {
5714 ixgbe_down(adapter);
5715 ixgbe_free_irq(adapter);
5716 ixgbe_free_all_tx_resources(adapter);
5717 ixgbe_free_all_rx_resources(adapter);
5720 ixgbe_clear_interrupt_scheme(adapter);
5722 kfree(adapter->ixgbe_ieee_pfc);
5723 kfree(adapter->ixgbe_ieee_ets);
5727 retval = pci_save_state(pdev);
5733 ixgbe_set_rx_mode(netdev);
5735 /* turn on all-multi mode if wake on multicast is enabled */
5736 if (wufc & IXGBE_WUFC_MC) {
5737 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5738 fctrl |= IXGBE_FCTRL_MPE;
5739 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5742 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5743 ctrl |= IXGBE_CTRL_GIO_DIS;
5744 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5746 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5748 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5749 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5752 switch (hw->mac.type) {
5753 case ixgbe_mac_82598EB:
5754 pci_wake_from_d3(pdev, false);
5756 case ixgbe_mac_82599EB:
5757 case ixgbe_mac_X540:
5758 pci_wake_from_d3(pdev, !!wufc);
5764 *enable_wake = !!wufc;
5766 ixgbe_release_hw_control(adapter);
5768 pci_disable_device(pdev);
5774 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5779 retval = __ixgbe_shutdown(pdev, &wake);
5784 pci_prepare_to_sleep(pdev);
5786 pci_wake_from_d3(pdev, false);
5787 pci_set_power_state(pdev, PCI_D3hot);
5792 #endif /* CONFIG_PM */
5794 static void ixgbe_shutdown(struct pci_dev *pdev)
5798 __ixgbe_shutdown(pdev, &wake);
5800 if (system_state == SYSTEM_POWER_OFF) {
5801 pci_wake_from_d3(pdev, wake);
5802 pci_set_power_state(pdev, PCI_D3hot);
5807 * ixgbe_update_stats - Update the board statistics counters.
5808 * @adapter: board private structure
5810 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5812 struct net_device *netdev = adapter->netdev;
5813 struct ixgbe_hw *hw = &adapter->hw;
5814 struct ixgbe_hw_stats *hwstats = &adapter->stats;
5816 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5817 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
5818 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
5819 u64 bytes = 0, packets = 0;
5821 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5822 test_bit(__IXGBE_RESETTING, &adapter->state))
5825 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5828 for (i = 0; i < 16; i++)
5829 adapter->hw_rx_no_dma_resources +=
5830 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5831 for (i = 0; i < adapter->num_rx_queues; i++) {
5832 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
5833 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
5835 adapter->rsc_total_count = rsc_count;
5836 adapter->rsc_total_flush = rsc_flush;
5839 for (i = 0; i < adapter->num_rx_queues; i++) {
5840 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
5841 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
5842 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
5843 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
5844 bytes += rx_ring->stats.bytes;
5845 packets += rx_ring->stats.packets;
5847 adapter->non_eop_descs = non_eop_descs;
5848 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5849 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5850 netdev->stats.rx_bytes = bytes;
5851 netdev->stats.rx_packets = packets;
5855 /* gather some stats to the adapter struct that are per queue */
5856 for (i = 0; i < adapter->num_tx_queues; i++) {
5857 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5858 restart_queue += tx_ring->tx_stats.restart_queue;
5859 tx_busy += tx_ring->tx_stats.tx_busy;
5860 bytes += tx_ring->stats.bytes;
5861 packets += tx_ring->stats.packets;
5863 adapter->restart_queue = restart_queue;
5864 adapter->tx_busy = tx_busy;
5865 netdev->stats.tx_bytes = bytes;
5866 netdev->stats.tx_packets = packets;
5868 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5869 for (i = 0; i < 8; i++) {
5870 /* for packet buffers not used, the register should read 0 */
5871 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5873 hwstats->mpc[i] += mpc;
5874 total_mpc += hwstats->mpc[i];
5875 if (hw->mac.type == ixgbe_mac_82598EB)
5876 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5877 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5878 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5879 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5880 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5881 switch (hw->mac.type) {
5882 case ixgbe_mac_82598EB:
5883 hwstats->pxonrxc[i] +=
5884 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5886 case ixgbe_mac_82599EB:
5887 case ixgbe_mac_X540:
5888 hwstats->pxonrxc[i] +=
5889 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5894 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5895 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5897 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5898 /* work around hardware counting issue */
5899 hwstats->gprc -= missed_rx;
5901 ixgbe_update_xoff_received(adapter);
5903 /* 82598 hardware only has a 32 bit counter in the high register */
5904 switch (hw->mac.type) {
5905 case ixgbe_mac_82598EB:
5906 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5907 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5908 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5909 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5911 case ixgbe_mac_82599EB:
5912 case ixgbe_mac_X540:
5913 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5914 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
5915 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5916 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
5917 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5918 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5919 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5920 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5921 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5923 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5924 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5925 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5926 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5927 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5928 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5929 #endif /* IXGBE_FCOE */
5934 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5935 hwstats->bprc += bprc;
5936 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5937 if (hw->mac.type == ixgbe_mac_82598EB)
5938 hwstats->mprc -= bprc;
5939 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5940 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5941 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5942 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5943 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5944 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5945 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5946 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5947 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5948 hwstats->lxontxc += lxon;
5949 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5950 hwstats->lxofftxc += lxoff;
5951 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5952 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5953 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5955 * 82598 errata - tx of flow control packets is included in tx counters
5957 xon_off_tot = lxon + lxoff;
5958 hwstats->gptc -= xon_off_tot;
5959 hwstats->mptc -= xon_off_tot;
5960 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5961 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5962 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5963 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5964 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5965 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5966 hwstats->ptc64 -= xon_off_tot;
5967 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5968 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5969 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5970 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5971 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5972 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5974 /* Fill out the OS statistics structure */
5975 netdev->stats.multicast = hwstats->mprc;
5978 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
5979 netdev->stats.rx_dropped = 0;
5980 netdev->stats.rx_length_errors = hwstats->rlec;
5981 netdev->stats.rx_crc_errors = hwstats->crcerrs;
5982 netdev->stats.rx_missed_errors = total_mpc;
5986 * ixgbe_watchdog - Timer Call-back
5987 * @data: pointer to adapter cast into an unsigned long
5989 static void ixgbe_watchdog(unsigned long data)
5991 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5992 struct ixgbe_hw *hw = &adapter->hw;
5997 * Do the watchdog outside of interrupt context due to the lovely
5998 * delays that some of the newer hardware requires
6001 if (test_bit(__IXGBE_DOWN, &adapter->state))
6002 goto watchdog_short_circuit;
6004 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
6006 * for legacy and MSI interrupts don't set any bits
6007 * that are enabled for EIAM, because this operation
6008 * would set *both* EIMS and EICS for any bit in EIAM
6010 IXGBE_WRITE_REG(hw, IXGBE_EICS,
6011 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
6012 goto watchdog_reschedule;
6015 /* get one bit for every active tx/rx interrupt vector */
6016 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
6017 struct ixgbe_q_vector *qv = adapter->q_vector[i];
6018 if (qv->rxr_count || qv->txr_count)
6019 eics |= ((u64)1 << i);
6022 /* Cause software interrupt to ensure rx rings are cleaned */
6023 ixgbe_irq_rearm_queues(adapter, eics);
6025 watchdog_reschedule:
6026 /* Reset the timer */
6027 mod_timer(&adapter->watchdog_timer, round_jiffies(jiffies + 2 * HZ));
6029 watchdog_short_circuit:
6030 schedule_work(&adapter->watchdog_task);
6034 * ixgbe_multispeed_fiber_task - worker thread to configure multispeed fiber
6035 * @work: pointer to work_struct containing our data
6037 static void ixgbe_multispeed_fiber_task(struct work_struct *work)
6039 struct ixgbe_adapter *adapter = container_of(work,
6040 struct ixgbe_adapter,
6041 multispeed_fiber_task);
6042 struct ixgbe_hw *hw = &adapter->hw;
6046 adapter->flags |= IXGBE_FLAG_IN_SFP_LINK_TASK;
6047 autoneg = hw->phy.autoneg_advertised;
6048 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
6049 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
6050 hw->mac.autotry_restart = false;
6051 if (hw->mac.ops.setup_link)
6052 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
6053 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
6054 adapter->flags &= ~IXGBE_FLAG_IN_SFP_LINK_TASK;
6058 * ixgbe_sfp_config_module_task - worker thread to configure a new SFP+ module
6059 * @work: pointer to work_struct containing our data
6061 static void ixgbe_sfp_config_module_task(struct work_struct *work)
6063 struct ixgbe_adapter *adapter = container_of(work,
6064 struct ixgbe_adapter,
6065 sfp_config_module_task);
6066 struct ixgbe_hw *hw = &adapter->hw;
6069 adapter->flags |= IXGBE_FLAG_IN_SFP_MOD_TASK;
6071 /* Time for electrical oscillations to settle down */
6073 err = hw->phy.ops.identify_sfp(hw);
6075 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
6076 e_dev_err("failed to initialize because an unsupported SFP+ "
6077 "module type was detected.\n");
6078 e_dev_err("Reload the driver after installing a supported "
6080 unregister_netdev(adapter->netdev);
6083 if (hw->mac.ops.setup_sfp)
6084 hw->mac.ops.setup_sfp(hw);
6086 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
6087 /* This will also work for DA Twinax connections */
6088 schedule_work(&adapter->multispeed_fiber_task);
6089 adapter->flags &= ~IXGBE_FLAG_IN_SFP_MOD_TASK;
6093 * ixgbe_fdir_reinit_task - worker thread to reinit FDIR filter table
6094 * @work: pointer to work_struct containing our data
6096 static void ixgbe_fdir_reinit_task(struct work_struct *work)
6098 struct ixgbe_adapter *adapter = container_of(work,
6099 struct ixgbe_adapter,
6101 struct ixgbe_hw *hw = &adapter->hw;
6104 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
6105 for (i = 0; i < adapter->num_tx_queues; i++)
6106 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
6107 &(adapter->tx_ring[i]->state));
6109 e_err(probe, "failed to finish FDIR re-initialization, "
6110 "ignored adding FDIR ATR filters\n");
6112 /* Done FDIR Re-initialization, enable transmits */
6113 netif_tx_start_all_queues(adapter->netdev);
6116 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
6120 /* Do not perform spoof check for 82598 */
6121 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
6124 ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
6127 * ssvpc register is cleared on read, if zero then no
6128 * spoofed packets in the last interval.
6133 e_warn(drv, "%d Spoofed packets detected\n", ssvpc);
6136 static DEFINE_MUTEX(ixgbe_watchdog_lock);
6139 * ixgbe_watchdog_task - worker thread to bring link up
6140 * @work: pointer to work_struct containing our data
6142 static void ixgbe_watchdog_task(struct work_struct *work)
6144 struct ixgbe_adapter *adapter = container_of(work,
6145 struct ixgbe_adapter,
6147 struct net_device *netdev = adapter->netdev;
6148 struct ixgbe_hw *hw = &adapter->hw;
6152 struct ixgbe_ring *tx_ring;
6153 int some_tx_pending = 0;
6155 mutex_lock(&ixgbe_watchdog_lock);
6157 link_up = adapter->link_up;
6158 link_speed = adapter->link_speed;
6160 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
6161 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
6164 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6165 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
6166 hw->mac.ops.fc_enable(hw, i);
6168 hw->mac.ops.fc_enable(hw, 0);
6171 hw->mac.ops.fc_enable(hw, 0);
6176 time_after(jiffies, (adapter->link_check_timeout +
6177 IXGBE_TRY_LINK_TIMEOUT))) {
6178 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
6179 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
6181 adapter->link_up = link_up;
6182 adapter->link_speed = link_speed;
6186 if (!netif_carrier_ok(netdev)) {
6187 bool flow_rx, flow_tx;
6189 switch (hw->mac.type) {
6190 case ixgbe_mac_82598EB: {
6191 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
6192 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
6193 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
6194 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
6197 case ixgbe_mac_82599EB:
6198 case ixgbe_mac_X540: {
6199 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
6200 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
6201 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
6202 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
6211 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
6212 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
6214 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
6216 (link_speed == IXGBE_LINK_SPEED_100_FULL ?
6219 ((flow_rx && flow_tx) ? "RX/TX" :
6221 (flow_tx ? "TX" : "None"))));
6223 netif_carrier_on(netdev);
6224 ixgbe_check_vf_rate_limit(adapter);
6226 /* Force detection of hung controller */
6227 for (i = 0; i < adapter->num_tx_queues; i++) {
6228 tx_ring = adapter->tx_ring[i];
6229 set_check_for_tx_hang(tx_ring);
6233 adapter->link_up = false;
6234 adapter->link_speed = 0;
6235 if (netif_carrier_ok(netdev)) {
6236 e_info(drv, "NIC Link is Down\n");
6237 netif_carrier_off(netdev);
6241 if (!netif_carrier_ok(netdev)) {
6242 for (i = 0; i < adapter->num_tx_queues; i++) {
6243 tx_ring = adapter->tx_ring[i];
6244 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
6245 some_tx_pending = 1;
6250 if (some_tx_pending) {
6251 /* We've lost link, so the controller stops DMA,
6252 * but we've got queued Tx work that's never going
6253 * to get done, so reset controller to flush Tx.
6254 * (Do the reset outside of interrupt context).
6256 schedule_work(&adapter->reset_task);
6260 ixgbe_spoof_check(adapter);
6261 ixgbe_update_stats(adapter);
6262 mutex_unlock(&ixgbe_watchdog_lock);
6265 static int ixgbe_tso(struct ixgbe_adapter *adapter,
6266 struct ixgbe_ring *tx_ring, struct sk_buff *skb,
6267 u32 tx_flags, u8 *hdr_len, __be16 protocol)
6269 struct ixgbe_adv_tx_context_desc *context_desc;
6272 struct ixgbe_tx_buffer *tx_buffer_info;
6273 u32 vlan_macip_lens = 0, type_tucmd_mlhl;
6274 u32 mss_l4len_idx, l4len;
6276 if (skb_is_gso(skb)) {
6277 if (skb_header_cloned(skb)) {
6278 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
6282 l4len = tcp_hdrlen(skb);
6285 if (protocol == htons(ETH_P_IP)) {
6286 struct iphdr *iph = ip_hdr(skb);
6289 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
6293 } else if (skb_is_gso_v6(skb)) {
6294 ipv6_hdr(skb)->payload_len = 0;
6295 tcp_hdr(skb)->check =
6296 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
6297 &ipv6_hdr(skb)->daddr,
6301 i = tx_ring->next_to_use;
6303 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6304 context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
6306 /* VLAN MACLEN IPLEN */
6307 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
6309 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
6310 vlan_macip_lens |= ((skb_network_offset(skb)) <<
6311 IXGBE_ADVTXD_MACLEN_SHIFT);
6312 *hdr_len += skb_network_offset(skb);
6314 (skb_transport_header(skb) - skb_network_header(skb));
6316 (skb_transport_header(skb) - skb_network_header(skb));
6317 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
6318 context_desc->seqnum_seed = 0;
6320 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
6321 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
6322 IXGBE_ADVTXD_DTYP_CTXT);
6324 if (protocol == htons(ETH_P_IP))
6325 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
6326 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6327 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
6331 (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
6332 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
6333 /* use index 1 for TSO */
6334 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6335 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
6337 tx_buffer_info->time_stamp = jiffies;
6338 tx_buffer_info->next_to_watch = i;
6341 if (i == tx_ring->count)
6343 tx_ring->next_to_use = i;
6350 static u32 ixgbe_psum(struct ixgbe_adapter *adapter, struct sk_buff *skb,
6356 case cpu_to_be16(ETH_P_IP):
6357 rtn |= IXGBE_ADVTXD_TUCMD_IPV4;
6358 switch (ip_hdr(skb)->protocol) {
6360 rtn |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6363 rtn |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
6367 case cpu_to_be16(ETH_P_IPV6):
6368 /* XXX what about other V6 headers?? */
6369 switch (ipv6_hdr(skb)->nexthdr) {
6371 rtn |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6374 rtn |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
6379 if (unlikely(net_ratelimit()))
6380 e_warn(probe, "partial checksum but proto=%x!\n",
6388 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
6389 struct ixgbe_ring *tx_ring,
6390 struct sk_buff *skb, u32 tx_flags,
6393 struct ixgbe_adv_tx_context_desc *context_desc;
6395 struct ixgbe_tx_buffer *tx_buffer_info;
6396 u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
6398 if (skb->ip_summed == CHECKSUM_PARTIAL ||
6399 (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
6400 i = tx_ring->next_to_use;
6401 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6402 context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
6404 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
6406 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
6407 vlan_macip_lens |= (skb_network_offset(skb) <<
6408 IXGBE_ADVTXD_MACLEN_SHIFT);
6409 if (skb->ip_summed == CHECKSUM_PARTIAL)
6410 vlan_macip_lens |= (skb_transport_header(skb) -
6411 skb_network_header(skb));
6413 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
6414 context_desc->seqnum_seed = 0;
6416 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
6417 IXGBE_ADVTXD_DTYP_CTXT);
6419 if (skb->ip_summed == CHECKSUM_PARTIAL)
6420 type_tucmd_mlhl |= ixgbe_psum(adapter, skb, protocol);
6422 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
6423 /* use index zero for tx checksum offload */
6424 context_desc->mss_l4len_idx = 0;
6426 tx_buffer_info->time_stamp = jiffies;
6427 tx_buffer_info->next_to_watch = i;
6430 if (i == tx_ring->count)
6432 tx_ring->next_to_use = i;
6440 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
6441 struct ixgbe_ring *tx_ring,
6442 struct sk_buff *skb, u32 tx_flags,
6443 unsigned int first, const u8 hdr_len)
6445 struct device *dev = tx_ring->dev;
6446 struct ixgbe_tx_buffer *tx_buffer_info;
6448 unsigned int total = skb->len;
6449 unsigned int offset = 0, size, count = 0, i;
6450 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
6452 unsigned int bytecount = skb->len;
6455 i = tx_ring->next_to_use;
6457 if (tx_flags & IXGBE_TX_FLAGS_FCOE)
6458 /* excluding fcoe_crc_eof for FCoE */
6459 total -= sizeof(struct fcoe_crc_eof);
6461 len = min(skb_headlen(skb), total);
6463 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6464 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
6466 tx_buffer_info->length = size;
6467 tx_buffer_info->mapped_as_page = false;
6468 tx_buffer_info->dma = dma_map_single(dev,
6470 size, DMA_TO_DEVICE);
6471 if (dma_mapping_error(dev, tx_buffer_info->dma))
6473 tx_buffer_info->time_stamp = jiffies;
6474 tx_buffer_info->next_to_watch = i;
6483 if (i == tx_ring->count)
6488 for (f = 0; f < nr_frags; f++) {
6489 struct skb_frag_struct *frag;
6491 frag = &skb_shinfo(skb)->frags[f];
6492 len = min((unsigned int)frag->size, total);
6493 offset = frag->page_offset;
6497 if (i == tx_ring->count)
6500 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6501 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
6503 tx_buffer_info->length = size;
6504 tx_buffer_info->dma = dma_map_page(dev,
6508 tx_buffer_info->mapped_as_page = true;
6509 if (dma_mapping_error(dev, tx_buffer_info->dma))
6511 tx_buffer_info->time_stamp = jiffies;
6512 tx_buffer_info->next_to_watch = i;
6523 if (tx_flags & IXGBE_TX_FLAGS_TSO)
6524 gso_segs = skb_shinfo(skb)->gso_segs;
6526 /* adjust for FCoE Sequence Offload */
6527 else if (tx_flags & IXGBE_TX_FLAGS_FSO)
6528 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
6529 skb_shinfo(skb)->gso_size);
6530 #endif /* IXGBE_FCOE */
6531 bytecount += (gso_segs - 1) * hdr_len;
6533 /* multiply data chunks by size of headers */
6534 tx_ring->tx_buffer_info[i].bytecount = bytecount;
6535 tx_ring->tx_buffer_info[i].gso_segs = gso_segs;
6536 tx_ring->tx_buffer_info[i].skb = skb;
6537 tx_ring->tx_buffer_info[first].next_to_watch = i;
6542 e_dev_err("TX DMA map failed\n");
6544 /* clear timestamp and dma mappings for failed tx_buffer_info map */
6545 tx_buffer_info->dma = 0;
6546 tx_buffer_info->time_stamp = 0;
6547 tx_buffer_info->next_to_watch = 0;
6551 /* clear timestamp and dma mappings for remaining portion of packet */
6554 i += tx_ring->count;
6556 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6557 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
6563 static void ixgbe_tx_queue(struct ixgbe_ring *tx_ring,
6564 int tx_flags, int count, u32 paylen, u8 hdr_len)
6566 union ixgbe_adv_tx_desc *tx_desc = NULL;
6567 struct ixgbe_tx_buffer *tx_buffer_info;
6568 u32 olinfo_status = 0, cmd_type_len = 0;
6570 u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
6572 cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
6574 cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
6576 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
6577 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
6579 if (tx_flags & IXGBE_TX_FLAGS_TSO) {
6580 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
6582 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
6583 IXGBE_ADVTXD_POPTS_SHIFT;
6585 /* use index 1 context for tso */
6586 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6587 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
6588 olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
6589 IXGBE_ADVTXD_POPTS_SHIFT;
6591 } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
6592 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
6593 IXGBE_ADVTXD_POPTS_SHIFT;
6595 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6596 olinfo_status |= IXGBE_ADVTXD_CC;
6597 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6598 if (tx_flags & IXGBE_TX_FLAGS_FSO)
6599 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
6602 olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
6604 i = tx_ring->next_to_use;
6606 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6607 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
6608 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
6609 tx_desc->read.cmd_type_len =
6610 cpu_to_le32(cmd_type_len | tx_buffer_info->length);
6611 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
6613 if (i == tx_ring->count)
6617 tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
6620 * Force memory writes to complete before letting h/w
6621 * know there are new descriptors to fetch. (Only
6622 * applicable for weak-ordered memory model archs,
6627 tx_ring->next_to_use = i;
6628 writel(i, tx_ring->tail);
6631 static void ixgbe_atr(struct ixgbe_ring *ring, struct sk_buff *skb,
6632 u32 tx_flags, __be16 protocol)
6634 struct ixgbe_q_vector *q_vector = ring->q_vector;
6635 union ixgbe_atr_hash_dword input = { .dword = 0 };
6636 union ixgbe_atr_hash_dword common = { .dword = 0 };
6638 unsigned char *network;
6640 struct ipv6hdr *ipv6;
6645 /* if ring doesn't have a interrupt vector, cannot perform ATR */
6649 /* do nothing if sampling is disabled */
6650 if (!ring->atr_sample_rate)
6655 /* snag network header to get L4 type and address */
6656 hdr.network = skb_network_header(skb);
6658 /* Currently only IPv4/IPv6 with TCP is supported */
6659 if ((protocol != __constant_htons(ETH_P_IPV6) ||
6660 hdr.ipv6->nexthdr != IPPROTO_TCP) &&
6661 (protocol != __constant_htons(ETH_P_IP) ||
6662 hdr.ipv4->protocol != IPPROTO_TCP))
6667 /* skip this packet since the socket is closing */
6671 /* sample on all syn packets or once every atr sample count */
6672 if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
6675 /* reset sample count */
6676 ring->atr_count = 0;
6678 vlan_id = htons(tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
6681 * src and dst are inverted, think how the receiver sees them
6683 * The input is broken into two sections, a non-compressed section
6684 * containing vm_pool, vlan_id, and flow_type. The rest of the data
6685 * is XORed together and stored in the compressed dword.
6687 input.formatted.vlan_id = vlan_id;
6690 * since src port and flex bytes occupy the same word XOR them together
6691 * and write the value to source port portion of compressed dword
6694 common.port.src ^= th->dest ^ __constant_htons(ETH_P_8021Q);
6696 common.port.src ^= th->dest ^ protocol;
6697 common.port.dst ^= th->source;
6699 if (protocol == __constant_htons(ETH_P_IP)) {
6700 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
6701 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
6703 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
6704 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
6705 hdr.ipv6->saddr.s6_addr32[1] ^
6706 hdr.ipv6->saddr.s6_addr32[2] ^
6707 hdr.ipv6->saddr.s6_addr32[3] ^
6708 hdr.ipv6->daddr.s6_addr32[0] ^
6709 hdr.ipv6->daddr.s6_addr32[1] ^
6710 hdr.ipv6->daddr.s6_addr32[2] ^
6711 hdr.ipv6->daddr.s6_addr32[3];
6714 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
6715 ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
6716 input, common, ring->queue_index);
6719 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, int size)
6721 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
6722 /* Herbert's original patch had:
6723 * smp_mb__after_netif_stop_queue();
6724 * but since that doesn't exist yet, just open code it. */
6727 /* We need to check again in a case another CPU has just
6728 * made room available. */
6729 if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
6732 /* A reprieve! - use start_queue because it doesn't call schedule */
6733 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
6734 ++tx_ring->tx_stats.restart_queue;
6738 static int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, int size)
6740 if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
6742 return __ixgbe_maybe_stop_tx(tx_ring, size);
6745 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6747 struct ixgbe_adapter *adapter = netdev_priv(dev);
6748 int txq = smp_processor_id();
6752 protocol = vlan_get_protocol(skb);
6754 if (((protocol == htons(ETH_P_FCOE)) ||
6755 (protocol == htons(ETH_P_FIP))) &&
6756 (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)) {
6757 txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
6758 txq += adapter->ring_feature[RING_F_FCOE].mask;
6763 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6764 while (unlikely(txq >= dev->real_num_tx_queues))
6765 txq -= dev->real_num_tx_queues;
6769 return skb_tx_hash(dev, skb);
6772 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
6773 struct ixgbe_adapter *adapter,
6774 struct ixgbe_ring *tx_ring)
6777 unsigned int tx_flags = 0;
6784 protocol = vlan_get_protocol(skb);
6786 if (vlan_tx_tag_present(skb)) {
6787 tx_flags |= vlan_tx_tag_get(skb);
6788 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6789 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6790 tx_flags |= tx_ring->dcb_tc << 13;
6792 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6793 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6794 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED &&
6795 skb->priority != TC_PRIO_CONTROL) {
6796 tx_flags |= tx_ring->dcb_tc << 13;
6797 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6798 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6802 /* for FCoE with DCB, we force the priority to what
6803 * was specified by the switch */
6804 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED &&
6805 (protocol == htons(ETH_P_FCOE)))
6806 tx_flags |= IXGBE_TX_FLAGS_FCOE;
6809 /* four things can cause us to need a context descriptor */
6810 if (skb_is_gso(skb) ||
6811 (skb->ip_summed == CHECKSUM_PARTIAL) ||
6812 (tx_flags & IXGBE_TX_FLAGS_VLAN) ||
6813 (tx_flags & IXGBE_TX_FLAGS_FCOE))
6816 count += TXD_USE_COUNT(skb_headlen(skb));
6817 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6818 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6820 if (ixgbe_maybe_stop_tx(tx_ring, count)) {
6821 tx_ring->tx_stats.tx_busy++;
6822 return NETDEV_TX_BUSY;
6825 first = tx_ring->next_to_use;
6826 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6828 /* setup tx offload for FCoE */
6829 tso = ixgbe_fso(adapter, tx_ring, skb, tx_flags, &hdr_len);
6831 dev_kfree_skb_any(skb);
6832 return NETDEV_TX_OK;
6835 tx_flags |= IXGBE_TX_FLAGS_FSO;
6836 #endif /* IXGBE_FCOE */
6838 if (protocol == htons(ETH_P_IP))
6839 tx_flags |= IXGBE_TX_FLAGS_IPV4;
6840 tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len,
6843 dev_kfree_skb_any(skb);
6844 return NETDEV_TX_OK;
6848 tx_flags |= IXGBE_TX_FLAGS_TSO;
6849 else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags,
6851 (skb->ip_summed == CHECKSUM_PARTIAL))
6852 tx_flags |= IXGBE_TX_FLAGS_CSUM;
6855 count = ixgbe_tx_map(adapter, tx_ring, skb, tx_flags, first, hdr_len);
6857 /* add the ATR filter if ATR is on */
6858 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
6859 ixgbe_atr(tx_ring, skb, tx_flags, protocol);
6860 ixgbe_tx_queue(tx_ring, tx_flags, count, skb->len, hdr_len);
6861 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
6864 dev_kfree_skb_any(skb);
6865 tx_ring->tx_buffer_info[first].time_stamp = 0;
6866 tx_ring->next_to_use = first;
6869 return NETDEV_TX_OK;
6872 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
6874 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6875 struct ixgbe_ring *tx_ring;
6877 tx_ring = adapter->tx_ring[skb->queue_mapping];
6878 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
6882 * ixgbe_set_mac - Change the Ethernet Address of the NIC
6883 * @netdev: network interface device structure
6884 * @p: pointer to an address structure
6886 * Returns 0 on success, negative on failure
6888 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6890 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6891 struct ixgbe_hw *hw = &adapter->hw;
6892 struct sockaddr *addr = p;
6894 if (!is_valid_ether_addr(addr->sa_data))
6895 return -EADDRNOTAVAIL;
6897 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6898 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6900 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
6907 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6909 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6910 struct ixgbe_hw *hw = &adapter->hw;
6914 if (prtad != hw->phy.mdio.prtad)
6916 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6922 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6923 u16 addr, u16 value)
6925 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6926 struct ixgbe_hw *hw = &adapter->hw;
6928 if (prtad != hw->phy.mdio.prtad)
6930 return hw->phy.ops.write_reg(hw, addr, devad, value);
6933 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6935 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6937 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6941 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6943 * @netdev: network interface device structure
6945 * Returns non-zero on failure
6947 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6950 struct ixgbe_adapter *adapter = netdev_priv(dev);
6951 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6953 if (is_valid_ether_addr(mac->san_addr)) {
6955 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6962 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6964 * @netdev: network interface device structure
6966 * Returns non-zero on failure
6968 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6971 struct ixgbe_adapter *adapter = netdev_priv(dev);
6972 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6974 if (is_valid_ether_addr(mac->san_addr)) {
6976 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6982 #ifdef CONFIG_NET_POLL_CONTROLLER
6984 * Polling 'interrupt' - used by things like netconsole to send skbs
6985 * without having to re-enable interrupts. It's not called while
6986 * the interrupt routine is executing.
6988 static void ixgbe_netpoll(struct net_device *netdev)
6990 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6993 /* if interface is down do nothing */
6994 if (test_bit(__IXGBE_DOWN, &adapter->state))
6997 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
6998 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
6999 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
7000 for (i = 0; i < num_q_vectors; i++) {
7001 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
7002 ixgbe_msix_clean_many(0, q_vector);
7005 ixgbe_intr(adapter->pdev->irq, netdev);
7007 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
7011 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
7012 struct rtnl_link_stats64 *stats)
7014 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7018 for (i = 0; i < adapter->num_rx_queues; i++) {
7019 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
7025 start = u64_stats_fetch_begin_bh(&ring->syncp);
7026 packets = ring->stats.packets;
7027 bytes = ring->stats.bytes;
7028 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
7029 stats->rx_packets += packets;
7030 stats->rx_bytes += bytes;
7034 for (i = 0; i < adapter->num_tx_queues; i++) {
7035 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->tx_ring[i]);
7041 start = u64_stats_fetch_begin_bh(&ring->syncp);
7042 packets = ring->stats.packets;
7043 bytes = ring->stats.bytes;
7044 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
7045 stats->tx_packets += packets;
7046 stats->tx_bytes += bytes;
7050 /* following stats updated by ixgbe_watchdog_task() */
7051 stats->multicast = netdev->stats.multicast;
7052 stats->rx_errors = netdev->stats.rx_errors;
7053 stats->rx_length_errors = netdev->stats.rx_length_errors;
7054 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
7055 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
7060 static const struct net_device_ops ixgbe_netdev_ops = {
7061 .ndo_open = ixgbe_open,
7062 .ndo_stop = ixgbe_close,
7063 .ndo_start_xmit = ixgbe_xmit_frame,
7064 .ndo_select_queue = ixgbe_select_queue,
7065 .ndo_set_rx_mode = ixgbe_set_rx_mode,
7066 .ndo_set_multicast_list = ixgbe_set_rx_mode,
7067 .ndo_validate_addr = eth_validate_addr,
7068 .ndo_set_mac_address = ixgbe_set_mac,
7069 .ndo_change_mtu = ixgbe_change_mtu,
7070 .ndo_tx_timeout = ixgbe_tx_timeout,
7071 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
7072 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
7073 .ndo_do_ioctl = ixgbe_ioctl,
7074 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
7075 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
7076 .ndo_set_vf_tx_rate = ixgbe_ndo_set_vf_bw,
7077 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
7078 .ndo_get_stats64 = ixgbe_get_stats64,
7079 #ifdef CONFIG_IXGBE_DCB
7080 .ndo_setup_tc = ixgbe_setup_tc,
7082 #ifdef CONFIG_NET_POLL_CONTROLLER
7083 .ndo_poll_controller = ixgbe_netpoll,
7086 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
7087 .ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
7088 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
7089 .ndo_fcoe_enable = ixgbe_fcoe_enable,
7090 .ndo_fcoe_disable = ixgbe_fcoe_disable,
7091 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
7092 #endif /* IXGBE_FCOE */
7095 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
7096 const struct ixgbe_info *ii)
7098 #ifdef CONFIG_PCI_IOV
7099 struct ixgbe_hw *hw = &adapter->hw;
7102 if (hw->mac.type == ixgbe_mac_82598EB || !max_vfs)
7105 /* The 82599 supports up to 64 VFs per physical function
7106 * but this implementation limits allocation to 63 so that
7107 * basic networking resources are still available to the
7110 adapter->num_vfs = (max_vfs > 63) ? 63 : max_vfs;
7111 adapter->flags |= IXGBE_FLAG_SRIOV_ENABLED;
7112 err = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
7114 e_err(probe, "Failed to enable PCI sriov: %d\n", err);
7117 /* If call to enable VFs succeeded then allocate memory
7118 * for per VF control structures.
7121 kcalloc(adapter->num_vfs,
7122 sizeof(struct vf_data_storage), GFP_KERNEL);
7123 if (adapter->vfinfo) {
7124 /* Now that we're sure SR-IOV is enabled
7125 * and memory allocated set up the mailbox parameters
7127 ixgbe_init_mbx_params_pf(hw);
7128 memcpy(&hw->mbx.ops, ii->mbx_ops,
7129 sizeof(hw->mbx.ops));
7131 /* Disable RSC when in SR-IOV mode */
7132 adapter->flags2 &= ~(IXGBE_FLAG2_RSC_CAPABLE |
7133 IXGBE_FLAG2_RSC_ENABLED);
7138 e_err(probe, "Unable to allocate memory for VF Data Storage - "
7139 "SRIOV disabled\n");
7140 pci_disable_sriov(adapter->pdev);
7143 adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
7144 adapter->num_vfs = 0;
7145 #endif /* CONFIG_PCI_IOV */
7149 * ixgbe_probe - Device Initialization Routine
7150 * @pdev: PCI device information struct
7151 * @ent: entry in ixgbe_pci_tbl
7153 * Returns 0 on success, negative on failure
7155 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
7156 * The OS initialization, configuring of the adapter private structure,
7157 * and a hardware reset occur.
7159 static int __devinit ixgbe_probe(struct pci_dev *pdev,
7160 const struct pci_device_id *ent)
7162 struct net_device *netdev;
7163 struct ixgbe_adapter *adapter = NULL;
7164 struct ixgbe_hw *hw;
7165 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
7166 static int cards_found;
7167 int i, err, pci_using_dac;
7168 u8 part_str[IXGBE_PBANUM_LENGTH];
7169 unsigned int indices = num_possible_cpus();
7175 /* Catch broken hardware that put the wrong VF device ID in
7176 * the PCIe SR-IOV capability.
7178 if (pdev->is_virtfn) {
7179 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
7180 pci_name(pdev), pdev->vendor, pdev->device);
7184 err = pci_enable_device_mem(pdev);
7188 if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
7189 !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
7192 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
7194 err = dma_set_coherent_mask(&pdev->dev,
7198 "No usable DMA configuration, aborting\n");
7205 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
7206 IORESOURCE_MEM), ixgbe_driver_name);
7209 "pci_request_selected_regions failed 0x%x\n", err);
7213 pci_enable_pcie_error_reporting(pdev);
7215 pci_set_master(pdev);
7216 pci_save_state(pdev);
7218 if (ii->mac == ixgbe_mac_82598EB)
7219 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
7221 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
7223 #if defined(CONFIG_DCB)
7224 indices = max_t(unsigned int, indices, IXGBE_MAX_DCB_INDICES);
7225 #elif defined(IXGBE_FCOE)
7226 indices += min_t(unsigned int, num_possible_cpus(),
7227 IXGBE_MAX_FCOE_INDICES);
7229 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
7232 goto err_alloc_etherdev;
7235 SET_NETDEV_DEV(netdev, &pdev->dev);
7237 adapter = netdev_priv(netdev);
7238 pci_set_drvdata(pdev, adapter);
7240 adapter->netdev = netdev;
7241 adapter->pdev = pdev;
7244 adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
7246 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
7247 pci_resource_len(pdev, 0));
7253 for (i = 1; i <= 5; i++) {
7254 if (pci_resource_len(pdev, i) == 0)
7258 netdev->netdev_ops = &ixgbe_netdev_ops;
7259 ixgbe_set_ethtool_ops(netdev);
7260 netdev->watchdog_timeo = 5 * HZ;
7261 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
7263 adapter->bd_number = cards_found;
7266 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
7267 hw->mac.type = ii->mac;
7270 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
7271 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
7272 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
7273 if (!(eec & (1 << 8)))
7274 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
7277 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
7278 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
7279 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
7280 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
7281 hw->phy.mdio.mmds = 0;
7282 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
7283 hw->phy.mdio.dev = netdev;
7284 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
7285 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
7287 /* set up this timer and work struct before calling get_invariants
7288 * which might start the timer
7290 init_timer(&adapter->sfp_timer);
7291 adapter->sfp_timer.function = ixgbe_sfp_timer;
7292 adapter->sfp_timer.data = (unsigned long) adapter;
7294 INIT_WORK(&adapter->sfp_task, ixgbe_sfp_task);
7296 /* multispeed fiber has its own tasklet, called from GPI SDP1 context */
7297 INIT_WORK(&adapter->multispeed_fiber_task, ixgbe_multispeed_fiber_task);
7299 /* a new SFP+ module arrival, called from GPI SDP2 context */
7300 INIT_WORK(&adapter->sfp_config_module_task,
7301 ixgbe_sfp_config_module_task);
7303 ii->get_invariants(hw);
7305 /* setup the private structure */
7306 err = ixgbe_sw_init(adapter);
7310 /* Make it possible the adapter to be woken up via WOL */
7311 switch (adapter->hw.mac.type) {
7312 case ixgbe_mac_82599EB:
7313 case ixgbe_mac_X540:
7314 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7321 * If there is a fan on this device and it has failed log the
7324 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
7325 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
7326 if (esdp & IXGBE_ESDP_SDP1)
7327 e_crit(probe, "Fan has stopped, replace the adapter\n");
7330 /* reset_hw fills in the perm_addr as well */
7331 hw->phy.reset_if_overtemp = true;
7332 err = hw->mac.ops.reset_hw(hw);
7333 hw->phy.reset_if_overtemp = false;
7334 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
7335 hw->mac.type == ixgbe_mac_82598EB) {
7337 * Start a kernel thread to watch for a module to arrive.
7338 * Only do this for 82598, since 82599 will generate
7339 * interrupts on module arrival.
7341 set_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
7342 mod_timer(&adapter->sfp_timer,
7343 round_jiffies(jiffies + (2 * HZ)));
7345 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
7346 e_dev_err("failed to initialize because an unsupported SFP+ "
7347 "module type was detected.\n");
7348 e_dev_err("Reload the driver after installing a supported "
7352 e_dev_err("HW Init failed: %d\n", err);
7356 ixgbe_probe_vf(adapter, ii);
7358 netdev->features = NETIF_F_SG |
7360 NETIF_F_HW_VLAN_TX |
7361 NETIF_F_HW_VLAN_RX |
7362 NETIF_F_HW_VLAN_FILTER;
7364 netdev->features |= NETIF_F_IPV6_CSUM;
7365 netdev->features |= NETIF_F_TSO;
7366 netdev->features |= NETIF_F_TSO6;
7367 netdev->features |= NETIF_F_GRO;
7369 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
7370 netdev->features |= NETIF_F_SCTP_CSUM;
7372 netdev->vlan_features |= NETIF_F_TSO;
7373 netdev->vlan_features |= NETIF_F_TSO6;
7374 netdev->vlan_features |= NETIF_F_IP_CSUM;
7375 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
7376 netdev->vlan_features |= NETIF_F_SG;
7378 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7379 adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
7380 IXGBE_FLAG_DCB_ENABLED);
7382 #ifdef CONFIG_IXGBE_DCB
7383 netdev->dcbnl_ops = &dcbnl_ops;
7387 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7388 if (hw->mac.ops.get_device_caps) {
7389 hw->mac.ops.get_device_caps(hw, &device_caps);
7390 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
7391 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
7394 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7395 netdev->vlan_features |= NETIF_F_FCOE_CRC;
7396 netdev->vlan_features |= NETIF_F_FSO;
7397 netdev->vlan_features |= NETIF_F_FCOE_MTU;
7399 #endif /* IXGBE_FCOE */
7400 if (pci_using_dac) {
7401 netdev->features |= NETIF_F_HIGHDMA;
7402 netdev->vlan_features |= NETIF_F_HIGHDMA;
7405 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
7406 netdev->features |= NETIF_F_LRO;
7408 /* make sure the EEPROM is good */
7409 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
7410 e_dev_err("The EEPROM Checksum Is Not Valid\n");
7415 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
7416 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
7418 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
7419 e_dev_err("invalid MAC address\n");
7424 /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
7425 if (hw->mac.ops.disable_tx_laser &&
7426 ((hw->phy.multispeed_fiber) ||
7427 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
7428 (hw->mac.type == ixgbe_mac_82599EB))))
7429 hw->mac.ops.disable_tx_laser(hw);
7431 init_timer(&adapter->watchdog_timer);
7432 adapter->watchdog_timer.function = ixgbe_watchdog;
7433 adapter->watchdog_timer.data = (unsigned long)adapter;
7435 INIT_WORK(&adapter->reset_task, ixgbe_reset_task);
7436 INIT_WORK(&adapter->watchdog_task, ixgbe_watchdog_task);
7438 err = ixgbe_init_interrupt_scheme(adapter);
7442 switch (pdev->device) {
7443 case IXGBE_DEV_ID_82599_SFP:
7444 /* Only this subdevice supports WOL */
7445 if (pdev->subsystem_device == IXGBE_SUBDEV_ID_82599_SFP)
7446 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
7447 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
7449 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
7450 /* All except this subdevice support WOL */
7451 if (pdev->subsystem_device != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
7452 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
7453 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
7455 case IXGBE_DEV_ID_82599_KX4:
7456 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
7457 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
7463 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
7465 /* pick up the PCI bus settings for reporting later */
7466 hw->mac.ops.get_bus_info(hw);
7468 /* print bus type/speed/width info */
7469 e_dev_info("(PCI Express:%s:%s) %pM\n",
7470 (hw->bus.speed == ixgbe_bus_speed_5000 ? "5.0Gb/s" :
7471 hw->bus.speed == ixgbe_bus_speed_2500 ? "2.5Gb/s" :
7473 (hw->bus.width == ixgbe_bus_width_pcie_x8 ? "Width x8" :
7474 hw->bus.width == ixgbe_bus_width_pcie_x4 ? "Width x4" :
7475 hw->bus.width == ixgbe_bus_width_pcie_x1 ? "Width x1" :
7479 err = ixgbe_read_pba_string_generic(hw, part_str, IXGBE_PBANUM_LENGTH);
7481 strncpy(part_str, "Unknown", IXGBE_PBANUM_LENGTH);
7482 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
7483 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
7484 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
7487 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
7488 hw->mac.type, hw->phy.type, part_str);
7490 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
7491 e_dev_warn("PCI-Express bandwidth available for this card is "
7492 "not sufficient for optimal performance.\n");
7493 e_dev_warn("For optimal performance a x8 PCI-Express slot "
7497 /* save off EEPROM version number */
7498 hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
7500 /* reset the hardware with the new settings */
7501 err = hw->mac.ops.start_hw(hw);
7503 if (err == IXGBE_ERR_EEPROM_VERSION) {
7504 /* We are running on a pre-production device, log a warning */
7505 e_dev_warn("This device is a pre-production adapter/LOM. "
7506 "Please be aware there may be issues associated "
7507 "with your hardware. If you are experiencing "
7508 "problems please contact your Intel or hardware "
7509 "representative who provided you with this "
7512 strcpy(netdev->name, "eth%d");
7513 err = register_netdev(netdev);
7517 /* carrier off reporting is important to ethtool even BEFORE open */
7518 netif_carrier_off(netdev);
7520 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
7521 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
7522 INIT_WORK(&adapter->fdir_reinit_task, ixgbe_fdir_reinit_task);
7524 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
7525 INIT_WORK(&adapter->check_overtemp_task,
7526 ixgbe_check_overtemp_task);
7527 #ifdef CONFIG_IXGBE_DCA
7528 if (dca_add_requester(&pdev->dev) == 0) {
7529 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
7530 ixgbe_setup_dca(adapter);
7533 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7534 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
7535 for (i = 0; i < adapter->num_vfs; i++)
7536 ixgbe_vf_configuration(pdev, (i | 0x10000000));
7539 /* add san mac addr to netdev */
7540 ixgbe_add_sanmac_netdev(netdev);
7542 e_dev_info("Intel(R) 10 Gigabit Network Connection\n");
7547 ixgbe_release_hw_control(adapter);
7548 ixgbe_clear_interrupt_scheme(adapter);
7551 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7552 ixgbe_disable_sriov(adapter);
7553 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
7554 del_timer_sync(&adapter->sfp_timer);
7555 cancel_work_sync(&adapter->sfp_task);
7556 cancel_work_sync(&adapter->multispeed_fiber_task);
7557 cancel_work_sync(&adapter->sfp_config_module_task);
7558 iounmap(hw->hw_addr);
7560 free_netdev(netdev);
7562 pci_release_selected_regions(pdev,
7563 pci_select_bars(pdev, IORESOURCE_MEM));
7566 pci_disable_device(pdev);
7571 * ixgbe_remove - Device Removal Routine
7572 * @pdev: PCI device information struct
7574 * ixgbe_remove is called by the PCI subsystem to alert the driver
7575 * that it should release a PCI device. The could be caused by a
7576 * Hot-Plug event, or because the driver is going to be removed from
7579 static void __devexit ixgbe_remove(struct pci_dev *pdev)
7581 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7582 struct net_device *netdev = adapter->netdev;
7584 set_bit(__IXGBE_DOWN, &adapter->state);
7587 * The timers may be rescheduled, so explicitly disable them
7588 * from being rescheduled.
7590 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
7591 del_timer_sync(&adapter->watchdog_timer);
7592 del_timer_sync(&adapter->sfp_timer);
7594 cancel_work_sync(&adapter->watchdog_task);
7595 cancel_work_sync(&adapter->sfp_task);
7596 cancel_work_sync(&adapter->multispeed_fiber_task);
7597 cancel_work_sync(&adapter->sfp_config_module_task);
7598 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
7599 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
7600 cancel_work_sync(&adapter->fdir_reinit_task);
7601 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
7602 cancel_work_sync(&adapter->check_overtemp_task);
7604 #ifdef CONFIG_IXGBE_DCA
7605 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
7606 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
7607 dca_remove_requester(&pdev->dev);
7608 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
7613 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
7614 ixgbe_cleanup_fcoe(adapter);
7616 #endif /* IXGBE_FCOE */
7618 /* remove the added san mac */
7619 ixgbe_del_sanmac_netdev(netdev);
7621 if (netdev->reg_state == NETREG_REGISTERED)
7622 unregister_netdev(netdev);
7624 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7625 ixgbe_disable_sriov(adapter);
7627 ixgbe_clear_interrupt_scheme(adapter);
7629 ixgbe_release_hw_control(adapter);
7631 iounmap(adapter->hw.hw_addr);
7632 pci_release_selected_regions(pdev, pci_select_bars(pdev,
7635 e_dev_info("complete\n");
7637 free_netdev(netdev);
7639 pci_disable_pcie_error_reporting(pdev);
7641 pci_disable_device(pdev);
7645 * ixgbe_io_error_detected - called when PCI error is detected
7646 * @pdev: Pointer to PCI device
7647 * @state: The current pci connection state
7649 * This function is called after a PCI bus error affecting
7650 * this device has been detected.
7652 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
7653 pci_channel_state_t state)
7655 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7656 struct net_device *netdev = adapter->netdev;
7658 netif_device_detach(netdev);
7660 if (state == pci_channel_io_perm_failure)
7661 return PCI_ERS_RESULT_DISCONNECT;
7663 if (netif_running(netdev))
7664 ixgbe_down(adapter);
7665 pci_disable_device(pdev);
7667 /* Request a slot reset. */
7668 return PCI_ERS_RESULT_NEED_RESET;
7672 * ixgbe_io_slot_reset - called after the pci bus has been reset.
7673 * @pdev: Pointer to PCI device
7675 * Restart the card from scratch, as if from a cold-boot.
7677 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7679 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7680 pci_ers_result_t result;
7683 if (pci_enable_device_mem(pdev)) {
7684 e_err(probe, "Cannot re-enable PCI device after reset.\n");
7685 result = PCI_ERS_RESULT_DISCONNECT;
7687 pci_set_master(pdev);
7688 pci_restore_state(pdev);
7689 pci_save_state(pdev);
7691 pci_wake_from_d3(pdev, false);
7693 ixgbe_reset(adapter);
7694 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7695 result = PCI_ERS_RESULT_RECOVERED;
7698 err = pci_cleanup_aer_uncorrect_error_status(pdev);
7700 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7701 "failed 0x%0x\n", err);
7702 /* non-fatal, continue */
7709 * ixgbe_io_resume - called when traffic can start flowing again.
7710 * @pdev: Pointer to PCI device
7712 * This callback is called when the error recovery driver tells us that
7713 * its OK to resume normal operation.
7715 static void ixgbe_io_resume(struct pci_dev *pdev)
7717 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7718 struct net_device *netdev = adapter->netdev;
7720 if (netif_running(netdev)) {
7721 if (ixgbe_up(adapter)) {
7722 e_info(probe, "ixgbe_up failed after reset\n");
7727 netif_device_attach(netdev);
7730 static struct pci_error_handlers ixgbe_err_handler = {
7731 .error_detected = ixgbe_io_error_detected,
7732 .slot_reset = ixgbe_io_slot_reset,
7733 .resume = ixgbe_io_resume,
7736 static struct pci_driver ixgbe_driver = {
7737 .name = ixgbe_driver_name,
7738 .id_table = ixgbe_pci_tbl,
7739 .probe = ixgbe_probe,
7740 .remove = __devexit_p(ixgbe_remove),
7742 .suspend = ixgbe_suspend,
7743 .resume = ixgbe_resume,
7745 .shutdown = ixgbe_shutdown,
7746 .err_handler = &ixgbe_err_handler
7750 * ixgbe_init_module - Driver Registration Routine
7752 * ixgbe_init_module is the first routine called when the driver is
7753 * loaded. All it does is register with the PCI subsystem.
7755 static int __init ixgbe_init_module(void)
7758 pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
7759 pr_info("%s\n", ixgbe_copyright);
7761 #ifdef CONFIG_IXGBE_DCA
7762 dca_register_notify(&dca_notifier);
7765 ret = pci_register_driver(&ixgbe_driver);
7769 module_init(ixgbe_init_module);
7772 * ixgbe_exit_module - Driver Exit Cleanup Routine
7774 * ixgbe_exit_module is called just before the driver is removed
7777 static void __exit ixgbe_exit_module(void)
7779 #ifdef CONFIG_IXGBE_DCA
7780 dca_unregister_notify(&dca_notifier);
7782 pci_unregister_driver(&ixgbe_driver);
7783 rcu_barrier(); /* Wait for completion of call_rcu()'s */
7786 #ifdef CONFIG_IXGBE_DCA
7787 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
7792 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
7793 __ixgbe_notify_dca);
7795 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7798 #endif /* CONFIG_IXGBE_DCA */
7800 module_exit(ixgbe_exit_module);