1 /*******************************************************************************
3 Intel 10 Gigabit PCI Express Linux driver
4 Copyright(c) 1999 - 2011 Intel Corporation.
6 This program is free software; you can redistribute it and/or modify it
7 under the terms and conditions of the GNU General Public License,
8 version 2, as published by the Free Software Foundation.
10 This program is distributed in the hope it will be useful, but WITHOUT
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
23 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26 *******************************************************************************/
28 #include <linux/types.h>
29 #include <linux/module.h>
30 #include <linux/pci.h>
31 #include <linux/netdevice.h>
32 #include <linux/vmalloc.h>
33 #include <linux/string.h>
36 #include <linux/tcp.h>
37 #include <linux/pkt_sched.h>
38 #include <linux/ipv6.h>
39 #include <linux/slab.h>
40 #include <net/checksum.h>
41 #include <net/ip6_checksum.h>
42 #include <linux/ethtool.h>
43 #include <linux/if_vlan.h>
44 #include <scsi/fc/fc_fcoe.h>
47 #include "ixgbe_common.h"
48 #include "ixgbe_dcb_82599.h"
49 #include "ixgbe_sriov.h"
51 char ixgbe_driver_name[] = "ixgbe";
52 static const char ixgbe_driver_string[] =
53 "Intel(R) 10 Gigabit PCI Express Network Driver";
58 #define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) "." \
59 __stringify(BUILD) "-k" __stringify(KFIX)
60 const char ixgbe_driver_version[] = DRV_VERSION;
61 static const char ixgbe_copyright[] =
62 "Copyright (c) 1999-2011 Intel Corporation.";
64 static const struct ixgbe_info *ixgbe_info_tbl[] = {
65 [board_82598] = &ixgbe_82598_info,
66 [board_82599] = &ixgbe_82599_info,
67 [board_X540] = &ixgbe_X540_info,
70 /* ixgbe_pci_tbl - PCI Device ID Table
72 * Wildcard entries (PCI_ANY_ID) should come last
73 * Last entry must be all 0s
75 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
76 * Class, Class Mask, private data (not used) }
78 static DEFINE_PCI_DEVICE_TABLE(ixgbe_pci_tbl) = {
79 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598),
81 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT),
83 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT),
85 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT),
87 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2),
89 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4),
91 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT),
93 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT),
95 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM),
97 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR),
99 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM),
101 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX),
103 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4),
105 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM),
107 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR),
109 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP),
111 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM),
113 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ),
115 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4),
117 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE),
119 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE),
121 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM),
123 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE),
125 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T),
127 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF2),
130 /* required last entry */
133 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
135 #ifdef CONFIG_IXGBE_DCA
136 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
138 static struct notifier_block dca_notifier = {
139 .notifier_call = ixgbe_notify_dca,
145 #ifdef CONFIG_PCI_IOV
146 static unsigned int max_vfs;
147 module_param(max_vfs, uint, 0);
148 MODULE_PARM_DESC(max_vfs,
149 "Maximum number of virtual functions to allocate per physical function");
150 #endif /* CONFIG_PCI_IOV */
152 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
153 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
154 MODULE_LICENSE("GPL");
155 MODULE_VERSION(DRV_VERSION);
157 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
159 static inline void ixgbe_disable_sriov(struct ixgbe_adapter *adapter)
161 struct ixgbe_hw *hw = &adapter->hw;
166 #ifdef CONFIG_PCI_IOV
167 /* disable iov and allow time for transactions to clear */
168 pci_disable_sriov(adapter->pdev);
171 /* turn off device IOV mode */
172 gcr = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
173 gcr &= ~(IXGBE_GCR_EXT_SRIOV);
174 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr);
175 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
176 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
177 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
179 /* set default pool back to 0 */
180 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
181 vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
182 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
184 /* take a breather then clean up driver data */
187 kfree(adapter->vfinfo);
188 adapter->vfinfo = NULL;
190 adapter->num_vfs = 0;
191 adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
194 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
196 if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
197 !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
198 schedule_work(&adapter->service_task);
201 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
203 BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
205 /* flush memory to make sure state is correct before next watchog */
206 smp_mb__before_clear_bit();
207 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
210 struct ixgbe_reg_info {
215 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
217 /* General Registers */
218 {IXGBE_CTRL, "CTRL"},
219 {IXGBE_STATUS, "STATUS"},
220 {IXGBE_CTRL_EXT, "CTRL_EXT"},
222 /* Interrupt Registers */
223 {IXGBE_EICR, "EICR"},
226 {IXGBE_SRRCTL(0), "SRRCTL"},
227 {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
228 {IXGBE_RDLEN(0), "RDLEN"},
229 {IXGBE_RDH(0), "RDH"},
230 {IXGBE_RDT(0), "RDT"},
231 {IXGBE_RXDCTL(0), "RXDCTL"},
232 {IXGBE_RDBAL(0), "RDBAL"},
233 {IXGBE_RDBAH(0), "RDBAH"},
236 {IXGBE_TDBAL(0), "TDBAL"},
237 {IXGBE_TDBAH(0), "TDBAH"},
238 {IXGBE_TDLEN(0), "TDLEN"},
239 {IXGBE_TDH(0), "TDH"},
240 {IXGBE_TDT(0), "TDT"},
241 {IXGBE_TXDCTL(0), "TXDCTL"},
243 /* List Terminator */
249 * ixgbe_regdump - register printout routine
251 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
257 switch (reginfo->ofs) {
258 case IXGBE_SRRCTL(0):
259 for (i = 0; i < 64; i++)
260 regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
262 case IXGBE_DCA_RXCTRL(0):
263 for (i = 0; i < 64; i++)
264 regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
267 for (i = 0; i < 64; i++)
268 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
271 for (i = 0; i < 64; i++)
272 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
275 for (i = 0; i < 64; i++)
276 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
278 case IXGBE_RXDCTL(0):
279 for (i = 0; i < 64; i++)
280 regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
283 for (i = 0; i < 64; i++)
284 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
287 for (i = 0; i < 64; i++)
288 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
291 for (i = 0; i < 64; i++)
292 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
295 for (i = 0; i < 64; i++)
296 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
299 for (i = 0; i < 64; i++)
300 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
303 for (i = 0; i < 64; i++)
304 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
307 for (i = 0; i < 64; i++)
308 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
310 case IXGBE_TXDCTL(0):
311 for (i = 0; i < 64; i++)
312 regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
315 pr_info("%-15s %08x\n", reginfo->name,
316 IXGBE_READ_REG(hw, reginfo->ofs));
320 for (i = 0; i < 8; i++) {
321 snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i*8, i*8+7);
322 pr_err("%-15s", rname);
323 for (j = 0; j < 8; j++)
324 pr_cont(" %08x", regs[i*8+j]);
331 * ixgbe_dump - Print registers, tx-rings and rx-rings
333 static void ixgbe_dump(struct ixgbe_adapter *adapter)
335 struct net_device *netdev = adapter->netdev;
336 struct ixgbe_hw *hw = &adapter->hw;
337 struct ixgbe_reg_info *reginfo;
339 struct ixgbe_ring *tx_ring;
340 struct ixgbe_tx_buffer *tx_buffer_info;
341 union ixgbe_adv_tx_desc *tx_desc;
342 struct my_u0 { u64 a; u64 b; } *u0;
343 struct ixgbe_ring *rx_ring;
344 union ixgbe_adv_rx_desc *rx_desc;
345 struct ixgbe_rx_buffer *rx_buffer_info;
349 if (!netif_msg_hw(adapter))
352 /* Print netdevice Info */
354 dev_info(&adapter->pdev->dev, "Net device Info\n");
355 pr_info("Device Name state "
356 "trans_start last_rx\n");
357 pr_info("%-15s %016lX %016lX %016lX\n",
364 /* Print Registers */
365 dev_info(&adapter->pdev->dev, "Register Dump\n");
366 pr_info(" Register Name Value\n");
367 for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
368 reginfo->name; reginfo++) {
369 ixgbe_regdump(hw, reginfo);
372 /* Print TX Ring Summary */
373 if (!netdev || !netif_running(netdev))
376 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
377 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
378 for (n = 0; n < adapter->num_tx_queues; n++) {
379 tx_ring = adapter->tx_ring[n];
381 &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
382 pr_info(" %5d %5X %5X %016llX %04X %3X %016llX\n",
383 n, tx_ring->next_to_use, tx_ring->next_to_clean,
384 (u64)tx_buffer_info->dma,
385 tx_buffer_info->length,
386 tx_buffer_info->next_to_watch,
387 (u64)tx_buffer_info->time_stamp);
391 if (!netif_msg_tx_done(adapter))
392 goto rx_ring_summary;
394 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
396 /* Transmit Descriptor Formats
398 * Advanced Transmit Descriptor
399 * +--------------------------------------------------------------+
400 * 0 | Buffer Address [63:0] |
401 * +--------------------------------------------------------------+
402 * 8 | PAYLEN | PORTS | IDX | STA | DCMD |DTYP | RSV | DTALEN |
403 * +--------------------------------------------------------------+
404 * 63 46 45 40 39 36 35 32 31 24 23 20 19 0
407 for (n = 0; n < adapter->num_tx_queues; n++) {
408 tx_ring = adapter->tx_ring[n];
409 pr_info("------------------------------------\n");
410 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
411 pr_info("------------------------------------\n");
412 pr_info("T [desc] [address 63:0 ] "
413 "[PlPOIdStDDt Ln] [bi->dma ] "
414 "leng ntw timestamp bi->skb\n");
416 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
417 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
418 tx_buffer_info = &tx_ring->tx_buffer_info[i];
419 u0 = (struct my_u0 *)tx_desc;
420 pr_info("T [0x%03X] %016llX %016llX %016llX"
421 " %04X %3X %016llX %p", i,
424 (u64)tx_buffer_info->dma,
425 tx_buffer_info->length,
426 tx_buffer_info->next_to_watch,
427 (u64)tx_buffer_info->time_stamp,
428 tx_buffer_info->skb);
429 if (i == tx_ring->next_to_use &&
430 i == tx_ring->next_to_clean)
432 else if (i == tx_ring->next_to_use)
434 else if (i == tx_ring->next_to_clean)
439 if (netif_msg_pktdata(adapter) &&
440 tx_buffer_info->dma != 0)
441 print_hex_dump(KERN_INFO, "",
442 DUMP_PREFIX_ADDRESS, 16, 1,
443 phys_to_virt(tx_buffer_info->dma),
444 tx_buffer_info->length, true);
448 /* Print RX Rings Summary */
450 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
451 pr_info("Queue [NTU] [NTC]\n");
452 for (n = 0; n < adapter->num_rx_queues; n++) {
453 rx_ring = adapter->rx_ring[n];
454 pr_info("%5d %5X %5X\n",
455 n, rx_ring->next_to_use, rx_ring->next_to_clean);
459 if (!netif_msg_rx_status(adapter))
462 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
464 /* Advanced Receive Descriptor (Read) Format
466 * +-----------------------------------------------------+
467 * 0 | Packet Buffer Address [63:1] |A0/NSE|
468 * +----------------------------------------------+------+
469 * 8 | Header Buffer Address [63:1] | DD |
470 * +-----------------------------------------------------+
473 * Advanced Receive Descriptor (Write-Back) Format
475 * 63 48 47 32 31 30 21 20 16 15 4 3 0
476 * +------------------------------------------------------+
477 * 0 | Packet IP |SPH| HDR_LEN | RSV|Packet| RSS |
478 * | Checksum Ident | | | | Type | Type |
479 * +------------------------------------------------------+
480 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
481 * +------------------------------------------------------+
482 * 63 48 47 32 31 20 19 0
484 for (n = 0; n < adapter->num_rx_queues; n++) {
485 rx_ring = adapter->rx_ring[n];
486 pr_info("------------------------------------\n");
487 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
488 pr_info("------------------------------------\n");
489 pr_info("R [desc] [ PktBuf A0] "
490 "[ HeadBuf DD] [bi->dma ] [bi->skb] "
491 "<-- Adv Rx Read format\n");
492 pr_info("RWB[desc] [PcsmIpSHl PtRs] "
493 "[vl er S cks ln] ---------------- [bi->skb] "
494 "<-- Adv Rx Write-Back format\n");
496 for (i = 0; i < rx_ring->count; i++) {
497 rx_buffer_info = &rx_ring->rx_buffer_info[i];
498 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
499 u0 = (struct my_u0 *)rx_desc;
500 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
501 if (staterr & IXGBE_RXD_STAT_DD) {
502 /* Descriptor Done */
503 pr_info("RWB[0x%03X] %016llX "
504 "%016llX ---------------- %p", i,
507 rx_buffer_info->skb);
509 pr_info("R [0x%03X] %016llX "
510 "%016llX %016llX %p", i,
513 (u64)rx_buffer_info->dma,
514 rx_buffer_info->skb);
516 if (netif_msg_pktdata(adapter)) {
517 print_hex_dump(KERN_INFO, "",
518 DUMP_PREFIX_ADDRESS, 16, 1,
519 phys_to_virt(rx_buffer_info->dma),
520 rx_ring->rx_buf_len, true);
522 if (rx_ring->rx_buf_len
523 < IXGBE_RXBUFFER_2048)
524 print_hex_dump(KERN_INFO, "",
525 DUMP_PREFIX_ADDRESS, 16, 1,
527 rx_buffer_info->page_dma +
528 rx_buffer_info->page_offset
534 if (i == rx_ring->next_to_use)
536 else if (i == rx_ring->next_to_clean)
548 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
552 /* Let firmware take over control of h/w */
553 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
554 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
555 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
558 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
562 /* Let firmware know the driver has taken over */
563 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
564 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
565 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
569 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
570 * @adapter: pointer to adapter struct
571 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
572 * @queue: queue to map the corresponding interrupt to
573 * @msix_vector: the vector to map to the corresponding queue
576 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
577 u8 queue, u8 msix_vector)
580 struct ixgbe_hw *hw = &adapter->hw;
581 switch (hw->mac.type) {
582 case ixgbe_mac_82598EB:
583 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
586 index = (((direction * 64) + queue) >> 2) & 0x1F;
587 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
588 ivar &= ~(0xFF << (8 * (queue & 0x3)));
589 ivar |= (msix_vector << (8 * (queue & 0x3)));
590 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
592 case ixgbe_mac_82599EB:
594 if (direction == -1) {
596 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
597 index = ((queue & 1) * 8);
598 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
599 ivar &= ~(0xFF << index);
600 ivar |= (msix_vector << index);
601 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
604 /* tx or rx causes */
605 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
606 index = ((16 * (queue & 1)) + (8 * direction));
607 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
608 ivar &= ~(0xFF << index);
609 ivar |= (msix_vector << index);
610 IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
618 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
623 switch (adapter->hw.mac.type) {
624 case ixgbe_mac_82598EB:
625 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
626 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
628 case ixgbe_mac_82599EB:
630 mask = (qmask & 0xFFFFFFFF);
631 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
632 mask = (qmask >> 32);
633 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
640 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_ring *tx_ring,
641 struct ixgbe_tx_buffer *tx_buffer_info)
643 if (tx_buffer_info->dma) {
644 if (tx_buffer_info->mapped_as_page)
645 dma_unmap_page(tx_ring->dev,
647 tx_buffer_info->length,
650 dma_unmap_single(tx_ring->dev,
652 tx_buffer_info->length,
654 tx_buffer_info->dma = 0;
656 if (tx_buffer_info->skb) {
657 dev_kfree_skb_any(tx_buffer_info->skb);
658 tx_buffer_info->skb = NULL;
660 tx_buffer_info->time_stamp = 0;
661 /* tx_buffer_info must be completely set up in the transmit path */
665 * ixgbe_dcb_txq_to_tc - convert a reg index to a traffic class
666 * @adapter: driver private struct
667 * @index: reg idx of queue to query (0-127)
669 * Helper function to determine the traffic index for a particular
672 * Returns : a tc index for use in range 0-7, or 0-3
674 static u8 ixgbe_dcb_txq_to_tc(struct ixgbe_adapter *adapter, u8 reg_idx)
677 int dcb_i = netdev_get_num_tc(adapter->netdev);
679 /* if DCB is not enabled the queues have no TC */
680 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
683 /* check valid range */
684 if (reg_idx >= adapter->hw.mac.max_tx_queues)
687 switch (adapter->hw.mac.type) {
688 case ixgbe_mac_82598EB:
692 if (dcb_i != 4 && dcb_i != 8)
695 /* if VMDq is enabled the lowest order bits determine TC */
696 if (adapter->flags & (IXGBE_FLAG_SRIOV_ENABLED |
697 IXGBE_FLAG_VMDQ_ENABLED)) {
698 tc = reg_idx & (dcb_i - 1);
703 * Convert the reg_idx into the correct TC. This bitmask
704 * targets the last full 32 ring traffic class and assigns
705 * it a value of 1. From there the rest of the rings are
706 * based on shifting the mask further up to include the
707 * reg_idx / 16 and then reg_idx / 8. It assumes dcB_i
708 * will only ever be 8 or 4 and that reg_idx will never
709 * be greater then 128. The code without the power of 2
710 * optimizations would be:
711 * (((reg_idx % 32) + 32) * dcb_i) >> (9 - reg_idx / 32)
713 tc = ((reg_idx & 0X1F) + 0x20) * dcb_i;
714 tc >>= 9 - (reg_idx >> 5);
720 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
722 struct ixgbe_hw *hw = &adapter->hw;
723 struct ixgbe_hw_stats *hwstats = &adapter->stats;
728 if ((hw->fc.current_mode == ixgbe_fc_full) ||
729 (hw->fc.current_mode == ixgbe_fc_rx_pause)) {
730 switch (hw->mac.type) {
731 case ixgbe_mac_82598EB:
732 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
735 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
737 hwstats->lxoffrxc += data;
739 /* refill credits (no tx hang) if we received xoff */
743 for (i = 0; i < adapter->num_tx_queues; i++)
744 clear_bit(__IXGBE_HANG_CHECK_ARMED,
745 &adapter->tx_ring[i]->state);
747 } else if (!(adapter->dcb_cfg.pfc_mode_enable))
750 /* update stats for each tc, only valid with PFC enabled */
751 for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
752 switch (hw->mac.type) {
753 case ixgbe_mac_82598EB:
754 xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
757 xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
759 hwstats->pxoffrxc[i] += xoff[i];
762 /* disarm tx queues that have received xoff frames */
763 for (i = 0; i < adapter->num_tx_queues; i++) {
764 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
765 u32 tc = ixgbe_dcb_txq_to_tc(adapter, tx_ring->reg_idx);
768 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
772 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
774 return ring->tx_stats.completed;
777 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
779 struct ixgbe_adapter *adapter = netdev_priv(ring->netdev);
780 struct ixgbe_hw *hw = &adapter->hw;
782 u32 head = IXGBE_READ_REG(hw, IXGBE_TDH(ring->reg_idx));
783 u32 tail = IXGBE_READ_REG(hw, IXGBE_TDT(ring->reg_idx));
786 return (head < tail) ?
787 tail - head : (tail + ring->count - head);
792 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
794 u32 tx_done = ixgbe_get_tx_completed(tx_ring);
795 u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
796 u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
799 clear_check_for_tx_hang(tx_ring);
802 * Check for a hung queue, but be thorough. This verifies
803 * that a transmit has been completed since the previous
804 * check AND there is at least one packet pending. The
805 * ARMED bit is set to indicate a potential hang. The
806 * bit is cleared if a pause frame is received to remove
807 * false hang detection due to PFC or 802.3x frames. By
808 * requiring this to fail twice we avoid races with
809 * pfc clearing the ARMED bit and conditions where we
810 * run the check_tx_hang logic with a transmit completion
811 * pending but without time to complete it yet.
813 if ((tx_done_old == tx_done) && tx_pending) {
814 /* make sure it is true for two checks in a row */
815 ret = test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
818 /* update completed stats and continue */
819 tx_ring->tx_stats.tx_done_old = tx_done;
820 /* reset the countdown */
821 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
827 #define IXGBE_MAX_TXD_PWR 14
828 #define IXGBE_MAX_DATA_PER_TXD (1 << IXGBE_MAX_TXD_PWR)
830 /* Tx Descriptors needed, worst case */
831 #define TXD_USE_COUNT(S) (((S) >> IXGBE_MAX_TXD_PWR) + \
832 (((S) & (IXGBE_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
833 #define DESC_NEEDED (TXD_USE_COUNT(IXGBE_MAX_DATA_PER_TXD) /* skb->data */ + \
834 MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1) /* for context */
837 * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
838 * @adapter: driver private struct
840 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
843 /* Do the reset outside of interrupt context */
844 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
845 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
846 ixgbe_service_event_schedule(adapter);
851 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
852 * @q_vector: structure containing interrupt and ring information
853 * @tx_ring: tx ring to clean
855 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
856 struct ixgbe_ring *tx_ring)
858 struct ixgbe_adapter *adapter = q_vector->adapter;
859 union ixgbe_adv_tx_desc *tx_desc, *eop_desc;
860 struct ixgbe_tx_buffer *tx_buffer_info;
861 unsigned int total_bytes = 0, total_packets = 0;
862 u16 i, eop, count = 0;
864 i = tx_ring->next_to_clean;
865 eop = tx_ring->tx_buffer_info[i].next_to_watch;
866 eop_desc = IXGBE_TX_DESC_ADV(tx_ring, eop);
868 while ((eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)) &&
869 (count < tx_ring->work_limit)) {
870 bool cleaned = false;
871 rmb(); /* read buffer_info after eop_desc */
872 for ( ; !cleaned; count++) {
873 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
874 tx_buffer_info = &tx_ring->tx_buffer_info[i];
876 tx_desc->wb.status = 0;
877 cleaned = (i == eop);
880 if (i == tx_ring->count)
883 if (cleaned && tx_buffer_info->skb) {
884 total_bytes += tx_buffer_info->bytecount;
885 total_packets += tx_buffer_info->gso_segs;
888 ixgbe_unmap_and_free_tx_resource(tx_ring,
892 tx_ring->tx_stats.completed++;
893 eop = tx_ring->tx_buffer_info[i].next_to_watch;
894 eop_desc = IXGBE_TX_DESC_ADV(tx_ring, eop);
897 tx_ring->next_to_clean = i;
898 tx_ring->total_bytes += total_bytes;
899 tx_ring->total_packets += total_packets;
900 u64_stats_update_begin(&tx_ring->syncp);
901 tx_ring->stats.packets += total_packets;
902 tx_ring->stats.bytes += total_bytes;
903 u64_stats_update_end(&tx_ring->syncp);
905 if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
906 /* schedule immediate reset if we believe we hung */
907 struct ixgbe_hw *hw = &adapter->hw;
908 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, eop);
909 e_err(drv, "Detected Tx Unit Hang\n"
911 " TDH, TDT <%x>, <%x>\n"
912 " next_to_use <%x>\n"
913 " next_to_clean <%x>\n"
914 "tx_buffer_info[next_to_clean]\n"
915 " time_stamp <%lx>\n"
917 tx_ring->queue_index,
918 IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
919 IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
920 tx_ring->next_to_use, eop,
921 tx_ring->tx_buffer_info[eop].time_stamp, jiffies);
923 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
926 "tx hang %d detected on queue %d, resetting adapter\n",
927 adapter->tx_timeout_count + 1, tx_ring->queue_index);
929 /* schedule immediate reset if we believe we hung */
930 ixgbe_tx_timeout_reset(adapter);
932 /* the adapter is about to reset, no point in enabling stuff */
936 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
937 if (unlikely(count && netif_carrier_ok(tx_ring->netdev) &&
938 (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
939 /* Make sure that anybody stopping the queue after this
940 * sees the new next_to_clean.
943 if (__netif_subqueue_stopped(tx_ring->netdev, tx_ring->queue_index) &&
944 !test_bit(__IXGBE_DOWN, &adapter->state)) {
945 netif_wake_subqueue(tx_ring->netdev, tx_ring->queue_index);
946 ++tx_ring->tx_stats.restart_queue;
950 return count < tx_ring->work_limit;
953 #ifdef CONFIG_IXGBE_DCA
954 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
955 struct ixgbe_ring *rx_ring,
958 struct ixgbe_hw *hw = &adapter->hw;
960 u8 reg_idx = rx_ring->reg_idx;
962 rxctrl = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(reg_idx));
963 switch (hw->mac.type) {
964 case ixgbe_mac_82598EB:
965 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
966 rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
968 case ixgbe_mac_82599EB:
970 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK_82599;
971 rxctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
972 IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599);
977 rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
978 rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
979 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_RRO_EN);
980 IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
983 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
984 struct ixgbe_ring *tx_ring,
987 struct ixgbe_hw *hw = &adapter->hw;
989 u8 reg_idx = tx_ring->reg_idx;
991 switch (hw->mac.type) {
992 case ixgbe_mac_82598EB:
993 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(reg_idx));
994 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
995 txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
996 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
997 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(reg_idx), txctrl);
999 case ixgbe_mac_82599EB:
1000 case ixgbe_mac_X540:
1001 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(reg_idx));
1002 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK_82599;
1003 txctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
1004 IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599);
1005 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
1006 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(reg_idx), txctrl);
1013 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
1015 struct ixgbe_adapter *adapter = q_vector->adapter;
1016 int cpu = get_cpu();
1020 if (q_vector->cpu == cpu)
1023 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1024 for (i = 0; i < q_vector->txr_count; i++) {
1025 ixgbe_update_tx_dca(adapter, adapter->tx_ring[r_idx], cpu);
1026 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1030 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1031 for (i = 0; i < q_vector->rxr_count; i++) {
1032 ixgbe_update_rx_dca(adapter, adapter->rx_ring[r_idx], cpu);
1033 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1037 q_vector->cpu = cpu;
1042 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
1047 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
1050 /* always use CB2 mode, difference is masked in the CB driver */
1051 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
1053 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
1054 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1058 for (i = 0; i < num_q_vectors; i++) {
1059 adapter->q_vector[i]->cpu = -1;
1060 ixgbe_update_dca(adapter->q_vector[i]);
1064 static int __ixgbe_notify_dca(struct device *dev, void *data)
1066 struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1067 unsigned long event = *(unsigned long *)data;
1069 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
1073 case DCA_PROVIDER_ADD:
1074 /* if we're already enabled, don't do it again */
1075 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1077 if (dca_add_requester(dev) == 0) {
1078 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1079 ixgbe_setup_dca(adapter);
1082 /* Fall Through since DCA is disabled. */
1083 case DCA_PROVIDER_REMOVE:
1084 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1085 dca_remove_requester(dev);
1086 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1087 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
1094 #endif /* CONFIG_IXGBE_DCA */
1096 static inline void ixgbe_rx_hash(union ixgbe_adv_rx_desc *rx_desc,
1097 struct sk_buff *skb)
1099 skb->rxhash = le32_to_cpu(rx_desc->wb.lower.hi_dword.rss);
1103 * ixgbe_receive_skb - Send a completed packet up the stack
1104 * @adapter: board private structure
1105 * @skb: packet to send up
1106 * @status: hardware indication of status of receive
1107 * @rx_ring: rx descriptor ring (for a specific queue) to setup
1108 * @rx_desc: rx descriptor
1110 static void ixgbe_receive_skb(struct ixgbe_q_vector *q_vector,
1111 struct sk_buff *skb, u8 status,
1112 struct ixgbe_ring *ring,
1113 union ixgbe_adv_rx_desc *rx_desc)
1115 struct ixgbe_adapter *adapter = q_vector->adapter;
1116 struct napi_struct *napi = &q_vector->napi;
1117 bool is_vlan = (status & IXGBE_RXD_STAT_VP);
1118 u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
1120 if (is_vlan && (tag & VLAN_VID_MASK))
1121 __vlan_hwaccel_put_tag(skb, tag);
1123 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
1124 napi_gro_receive(napi, skb);
1130 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1131 * @adapter: address of board private structure
1132 * @status_err: hardware indication of status of receive
1133 * @skb: skb currently being received and modified
1135 static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
1136 union ixgbe_adv_rx_desc *rx_desc,
1137 struct sk_buff *skb)
1139 u32 status_err = le32_to_cpu(rx_desc->wb.upper.status_error);
1141 skb_checksum_none_assert(skb);
1143 /* Rx csum disabled */
1144 if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
1147 /* if IP and error */
1148 if ((status_err & IXGBE_RXD_STAT_IPCS) &&
1149 (status_err & IXGBE_RXDADV_ERR_IPE)) {
1150 adapter->hw_csum_rx_error++;
1154 if (!(status_err & IXGBE_RXD_STAT_L4CS))
1157 if (status_err & IXGBE_RXDADV_ERR_TCPE) {
1158 u16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1161 * 82599 errata, UDP frames with a 0 checksum can be marked as
1164 if ((pkt_info & IXGBE_RXDADV_PKTTYPE_UDP) &&
1165 (adapter->hw.mac.type == ixgbe_mac_82599EB))
1168 adapter->hw_csum_rx_error++;
1172 /* It must be a TCP or UDP packet with a valid checksum */
1173 skb->ip_summed = CHECKSUM_UNNECESSARY;
1176 static inline void ixgbe_release_rx_desc(struct ixgbe_ring *rx_ring, u32 val)
1179 * Force memory writes to complete before letting h/w
1180 * know there are new descriptors to fetch. (Only
1181 * applicable for weak-ordered memory model archs,
1185 writel(val, rx_ring->tail);
1189 * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
1190 * @rx_ring: ring to place buffers on
1191 * @cleaned_count: number of buffers to replace
1193 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1195 union ixgbe_adv_rx_desc *rx_desc;
1196 struct ixgbe_rx_buffer *bi;
1197 struct sk_buff *skb;
1198 u16 i = rx_ring->next_to_use;
1200 /* do nothing if no valid netdev defined */
1201 if (!rx_ring->netdev)
1204 while (cleaned_count--) {
1205 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1206 bi = &rx_ring->rx_buffer_info[i];
1210 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1211 rx_ring->rx_buf_len);
1213 rx_ring->rx_stats.alloc_rx_buff_failed++;
1216 /* initialize queue mapping */
1217 skb_record_rx_queue(skb, rx_ring->queue_index);
1222 bi->dma = dma_map_single(rx_ring->dev,
1224 rx_ring->rx_buf_len,
1226 if (dma_mapping_error(rx_ring->dev, bi->dma)) {
1227 rx_ring->rx_stats.alloc_rx_buff_failed++;
1233 if (ring_is_ps_enabled(rx_ring)) {
1235 bi->page = netdev_alloc_page(rx_ring->netdev);
1237 rx_ring->rx_stats.alloc_rx_page_failed++;
1242 if (!bi->page_dma) {
1243 /* use a half page if we're re-using */
1244 bi->page_offset ^= PAGE_SIZE / 2;
1245 bi->page_dma = dma_map_page(rx_ring->dev,
1250 if (dma_mapping_error(rx_ring->dev,
1252 rx_ring->rx_stats.alloc_rx_page_failed++;
1258 /* Refresh the desc even if buffer_addrs didn't change
1259 * because each write-back erases this info. */
1260 rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
1261 rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
1263 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
1264 rx_desc->read.hdr_addr = 0;
1268 if (i == rx_ring->count)
1273 if (rx_ring->next_to_use != i) {
1274 rx_ring->next_to_use = i;
1275 ixgbe_release_rx_desc(rx_ring, i);
1279 static inline u16 ixgbe_get_hlen(union ixgbe_adv_rx_desc *rx_desc)
1281 /* HW will not DMA in data larger than the given buffer, even if it
1282 * parses the (NFS, of course) header to be larger. In that case, it
1283 * fills the header buffer and spills the rest into the page.
1285 u16 hdr_info = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info);
1286 u16 hlen = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
1287 IXGBE_RXDADV_HDRBUFLEN_SHIFT;
1288 if (hlen > IXGBE_RX_HDR_SIZE)
1289 hlen = IXGBE_RX_HDR_SIZE;
1294 * ixgbe_transform_rsc_queue - change rsc queue into a full packet
1295 * @skb: pointer to the last skb in the rsc queue
1297 * This function changes a queue full of hw rsc buffers into a completed
1298 * packet. It uses the ->prev pointers to find the first packet and then
1299 * turns it into the frag list owner.
1301 static inline struct sk_buff *ixgbe_transform_rsc_queue(struct sk_buff *skb)
1303 unsigned int frag_list_size = 0;
1304 unsigned int skb_cnt = 1;
1307 struct sk_buff *prev = skb->prev;
1308 frag_list_size += skb->len;
1314 skb_shinfo(skb)->frag_list = skb->next;
1316 skb->len += frag_list_size;
1317 skb->data_len += frag_list_size;
1318 skb->truesize += frag_list_size;
1319 IXGBE_RSC_CB(skb)->skb_cnt = skb_cnt;
1324 static inline bool ixgbe_get_rsc_state(union ixgbe_adv_rx_desc *rx_desc)
1326 return !!(le32_to_cpu(rx_desc->wb.lower.lo_dword.data) &
1327 IXGBE_RXDADV_RSCCNT_MASK);
1330 static void ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1331 struct ixgbe_ring *rx_ring,
1332 int *work_done, int work_to_do)
1334 struct ixgbe_adapter *adapter = q_vector->adapter;
1335 union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
1336 struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
1337 struct sk_buff *skb;
1338 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1339 const int current_node = numa_node_id();
1342 #endif /* IXGBE_FCOE */
1345 u16 cleaned_count = 0;
1346 bool pkt_is_rsc = false;
1348 i = rx_ring->next_to_clean;
1349 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1350 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1352 while (staterr & IXGBE_RXD_STAT_DD) {
1355 rmb(); /* read descriptor and rx_buffer_info after status DD */
1357 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1359 skb = rx_buffer_info->skb;
1360 rx_buffer_info->skb = NULL;
1361 prefetch(skb->data);
1363 if (ring_is_rsc_enabled(rx_ring))
1364 pkt_is_rsc = ixgbe_get_rsc_state(rx_desc);
1366 /* if this is a skb from previous receive DMA will be 0 */
1367 if (rx_buffer_info->dma) {
1370 !(staterr & IXGBE_RXD_STAT_EOP) &&
1373 * When HWRSC is enabled, delay unmapping
1374 * of the first packet. It carries the
1375 * header information, HW may still
1376 * access the header after the writeback.
1377 * Only unmap it when EOP is reached
1379 IXGBE_RSC_CB(skb)->delay_unmap = true;
1380 IXGBE_RSC_CB(skb)->dma = rx_buffer_info->dma;
1382 dma_unmap_single(rx_ring->dev,
1383 rx_buffer_info->dma,
1384 rx_ring->rx_buf_len,
1387 rx_buffer_info->dma = 0;
1389 if (ring_is_ps_enabled(rx_ring)) {
1390 hlen = ixgbe_get_hlen(rx_desc);
1391 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1393 hlen = le16_to_cpu(rx_desc->wb.upper.length);
1398 /* assume packet split since header is unmapped */
1399 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1403 dma_unmap_page(rx_ring->dev,
1404 rx_buffer_info->page_dma,
1407 rx_buffer_info->page_dma = 0;
1408 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1409 rx_buffer_info->page,
1410 rx_buffer_info->page_offset,
1413 if ((page_count(rx_buffer_info->page) == 1) &&
1414 (page_to_nid(rx_buffer_info->page) == current_node))
1415 get_page(rx_buffer_info->page);
1417 rx_buffer_info->page = NULL;
1419 skb->len += upper_len;
1420 skb->data_len += upper_len;
1421 skb->truesize += upper_len;
1425 if (i == rx_ring->count)
1428 next_rxd = IXGBE_RX_DESC_ADV(rx_ring, i);
1433 u32 nextp = (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
1434 IXGBE_RXDADV_NEXTP_SHIFT;
1435 next_buffer = &rx_ring->rx_buffer_info[nextp];
1437 next_buffer = &rx_ring->rx_buffer_info[i];
1440 if (!(staterr & IXGBE_RXD_STAT_EOP)) {
1441 if (ring_is_ps_enabled(rx_ring)) {
1442 rx_buffer_info->skb = next_buffer->skb;
1443 rx_buffer_info->dma = next_buffer->dma;
1444 next_buffer->skb = skb;
1445 next_buffer->dma = 0;
1447 skb->next = next_buffer->skb;
1448 skb->next->prev = skb;
1450 rx_ring->rx_stats.non_eop_descs++;
1455 skb = ixgbe_transform_rsc_queue(skb);
1456 /* if we got here without RSC the packet is invalid */
1458 __pskb_trim(skb, 0);
1459 rx_buffer_info->skb = skb;
1464 if (ring_is_rsc_enabled(rx_ring)) {
1465 if (IXGBE_RSC_CB(skb)->delay_unmap) {
1466 dma_unmap_single(rx_ring->dev,
1467 IXGBE_RSC_CB(skb)->dma,
1468 rx_ring->rx_buf_len,
1470 IXGBE_RSC_CB(skb)->dma = 0;
1471 IXGBE_RSC_CB(skb)->delay_unmap = false;
1475 if (ring_is_ps_enabled(rx_ring))
1476 rx_ring->rx_stats.rsc_count +=
1477 skb_shinfo(skb)->nr_frags;
1479 rx_ring->rx_stats.rsc_count +=
1480 IXGBE_RSC_CB(skb)->skb_cnt;
1481 rx_ring->rx_stats.rsc_flush++;
1484 /* ERR_MASK will only have valid bits if EOP set */
1485 if (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) {
1486 /* trim packet back to size 0 and recycle it */
1487 __pskb_trim(skb, 0);
1488 rx_buffer_info->skb = skb;
1492 ixgbe_rx_checksum(adapter, rx_desc, skb);
1493 if (adapter->netdev->features & NETIF_F_RXHASH)
1494 ixgbe_rx_hash(rx_desc, skb);
1496 /* probably a little skewed due to removing CRC */
1497 total_rx_bytes += skb->len;
1500 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1502 /* if ddp, not passing to ULD unless for FCP_RSP or error */
1503 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
1504 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
1508 #endif /* IXGBE_FCOE */
1509 ixgbe_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
1512 rx_desc->wb.upper.status_error = 0;
1515 if (*work_done >= work_to_do)
1518 /* return some buffers to hardware, one at a time is too slow */
1519 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
1520 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1524 /* use prefetched values */
1526 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1529 rx_ring->next_to_clean = i;
1530 cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
1533 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1536 /* include DDPed FCoE data */
1537 if (ddp_bytes > 0) {
1540 mss = rx_ring->netdev->mtu - sizeof(struct fcoe_hdr) -
1541 sizeof(struct fc_frame_header) -
1542 sizeof(struct fcoe_crc_eof);
1545 total_rx_bytes += ddp_bytes;
1546 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
1548 #endif /* IXGBE_FCOE */
1550 rx_ring->total_packets += total_rx_packets;
1551 rx_ring->total_bytes += total_rx_bytes;
1552 u64_stats_update_begin(&rx_ring->syncp);
1553 rx_ring->stats.packets += total_rx_packets;
1554 rx_ring->stats.bytes += total_rx_bytes;
1555 u64_stats_update_end(&rx_ring->syncp);
1558 static int ixgbe_clean_rxonly(struct napi_struct *, int);
1560 * ixgbe_configure_msix - Configure MSI-X hardware
1561 * @adapter: board private structure
1563 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1566 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1568 struct ixgbe_q_vector *q_vector;
1569 int i, q_vectors, v_idx, r_idx;
1572 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1575 * Populate the IVAR table and set the ITR values to the
1576 * corresponding register.
1578 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1579 q_vector = adapter->q_vector[v_idx];
1580 /* XXX for_each_set_bit(...) */
1581 r_idx = find_first_bit(q_vector->rxr_idx,
1582 adapter->num_rx_queues);
1584 for (i = 0; i < q_vector->rxr_count; i++) {
1585 u8 reg_idx = adapter->rx_ring[r_idx]->reg_idx;
1586 ixgbe_set_ivar(adapter, 0, reg_idx, v_idx);
1587 r_idx = find_next_bit(q_vector->rxr_idx,
1588 adapter->num_rx_queues,
1591 r_idx = find_first_bit(q_vector->txr_idx,
1592 adapter->num_tx_queues);
1594 for (i = 0; i < q_vector->txr_count; i++) {
1595 u8 reg_idx = adapter->tx_ring[r_idx]->reg_idx;
1596 ixgbe_set_ivar(adapter, 1, reg_idx, v_idx);
1597 r_idx = find_next_bit(q_vector->txr_idx,
1598 adapter->num_tx_queues,
1602 if (q_vector->txr_count && !q_vector->rxr_count)
1604 q_vector->eitr = adapter->tx_eitr_param;
1605 else if (q_vector->rxr_count)
1607 q_vector->eitr = adapter->rx_eitr_param;
1609 ixgbe_write_eitr(q_vector);
1610 /* If Flow Director is enabled, set interrupt affinity */
1611 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
1612 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)) {
1614 * Allocate the affinity_hint cpumask, assign the mask
1615 * for this vector, and set our affinity_hint for
1618 if (!alloc_cpumask_var(&q_vector->affinity_mask,
1621 cpumask_set_cpu(v_idx, q_vector->affinity_mask);
1622 irq_set_affinity_hint(adapter->msix_entries[v_idx].vector,
1623 q_vector->affinity_mask);
1627 switch (adapter->hw.mac.type) {
1628 case ixgbe_mac_82598EB:
1629 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1632 case ixgbe_mac_82599EB:
1633 case ixgbe_mac_X540:
1634 ixgbe_set_ivar(adapter, -1, 1, v_idx);
1640 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1642 /* set up to autoclear timer, and the vectors */
1643 mask = IXGBE_EIMS_ENABLE_MASK;
1644 if (adapter->num_vfs)
1645 mask &= ~(IXGBE_EIMS_OTHER |
1646 IXGBE_EIMS_MAILBOX |
1649 mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
1650 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1653 enum latency_range {
1657 latency_invalid = 255
1661 * ixgbe_update_itr - update the dynamic ITR value based on statistics
1662 * @adapter: pointer to adapter
1663 * @eitr: eitr setting (ints per sec) to give last timeslice
1664 * @itr_setting: current throttle rate in ints/second
1665 * @packets: the number of packets during this measurement interval
1666 * @bytes: the number of bytes during this measurement interval
1668 * Stores a new ITR value based on packets and byte
1669 * counts during the last interrupt. The advantage of per interrupt
1670 * computation is faster updates and more accurate ITR for the current
1671 * traffic pattern. Constants in this function were computed
1672 * based on theoretical maximum wire speed and thresholds were set based
1673 * on testing data as well as attempting to minimize response time
1674 * while increasing bulk throughput.
1675 * this functionality is controlled by the InterruptThrottleRate module
1676 * parameter (see ixgbe_param.c)
1678 static u8 ixgbe_update_itr(struct ixgbe_adapter *adapter,
1679 u32 eitr, u8 itr_setting,
1680 int packets, int bytes)
1682 unsigned int retval = itr_setting;
1687 goto update_itr_done;
1690 /* simple throttlerate management
1691 * 0-20MB/s lowest (100000 ints/s)
1692 * 20-100MB/s low (20000 ints/s)
1693 * 100-1249MB/s bulk (8000 ints/s)
1695 /* what was last interrupt timeslice? */
1696 timepassed_us = 1000000/eitr;
1697 bytes_perint = bytes / timepassed_us; /* bytes/usec */
1699 switch (itr_setting) {
1700 case lowest_latency:
1701 if (bytes_perint > adapter->eitr_low)
1702 retval = low_latency;
1705 if (bytes_perint > adapter->eitr_high)
1706 retval = bulk_latency;
1707 else if (bytes_perint <= adapter->eitr_low)
1708 retval = lowest_latency;
1711 if (bytes_perint <= adapter->eitr_high)
1712 retval = low_latency;
1721 * ixgbe_write_eitr - write EITR register in hardware specific way
1722 * @q_vector: structure containing interrupt and ring information
1724 * This function is made to be called by ethtool and by the driver
1725 * when it needs to update EITR registers at runtime. Hardware
1726 * specific quirks/differences are taken care of here.
1728 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1730 struct ixgbe_adapter *adapter = q_vector->adapter;
1731 struct ixgbe_hw *hw = &adapter->hw;
1732 int v_idx = q_vector->v_idx;
1733 u32 itr_reg = EITR_INTS_PER_SEC_TO_REG(q_vector->eitr);
1735 switch (adapter->hw.mac.type) {
1736 case ixgbe_mac_82598EB:
1737 /* must write high and low 16 bits to reset counter */
1738 itr_reg |= (itr_reg << 16);
1740 case ixgbe_mac_82599EB:
1741 case ixgbe_mac_X540:
1743 * 82599 and X540 can support a value of zero, so allow it for
1744 * max interrupt rate, but there is an errata where it can
1745 * not be zero with RSC
1748 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
1752 * set the WDIS bit to not clear the timer bits and cause an
1753 * immediate assertion of the interrupt
1755 itr_reg |= IXGBE_EITR_CNT_WDIS;
1760 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1763 static void ixgbe_set_itr_msix(struct ixgbe_q_vector *q_vector)
1765 struct ixgbe_adapter *adapter = q_vector->adapter;
1768 u8 current_itr, ret_itr;
1770 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1771 for (i = 0; i < q_vector->txr_count; i++) {
1772 struct ixgbe_ring *tx_ring = adapter->tx_ring[r_idx];
1773 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1775 tx_ring->total_packets,
1776 tx_ring->total_bytes);
1777 /* if the result for this queue would decrease interrupt
1778 * rate for this vector then use that result */
1779 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
1780 q_vector->tx_itr - 1 : ret_itr);
1781 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1785 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1786 for (i = 0; i < q_vector->rxr_count; i++) {
1787 struct ixgbe_ring *rx_ring = adapter->rx_ring[r_idx];
1788 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1790 rx_ring->total_packets,
1791 rx_ring->total_bytes);
1792 /* if the result for this queue would decrease interrupt
1793 * rate for this vector then use that result */
1794 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
1795 q_vector->rx_itr - 1 : ret_itr);
1796 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1800 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1802 switch (current_itr) {
1803 /* counts and packets in update_itr are dependent on these numbers */
1804 case lowest_latency:
1808 new_itr = 20000; /* aka hwitr = ~200 */
1816 if (new_itr != q_vector->eitr) {
1817 /* do an exponential smoothing */
1818 new_itr = ((q_vector->eitr * 9) + new_itr)/10;
1820 /* save the algorithm value here, not the smoothed one */
1821 q_vector->eitr = new_itr;
1823 ixgbe_write_eitr(q_vector);
1828 * ixgbe_check_overtemp_task - worker thread to check over tempurature
1829 * @work: pointer to work_struct containing our data
1831 static void ixgbe_check_overtemp_task(struct work_struct *work)
1833 struct ixgbe_adapter *adapter = container_of(work,
1834 struct ixgbe_adapter,
1835 check_overtemp_task);
1836 struct ixgbe_hw *hw = &adapter->hw;
1837 u32 eicr = adapter->interrupt_event;
1839 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
1842 switch (hw->device_id) {
1843 case IXGBE_DEV_ID_82599_T3_LOM: {
1845 bool link_up = false;
1847 if (hw->mac.ops.check_link)
1848 hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
1850 if (((eicr & IXGBE_EICR_GPI_SDP0) && (!link_up)) ||
1851 (eicr & IXGBE_EICR_LSC))
1852 /* Check if this is due to overtemp */
1853 if (hw->phy.ops.check_overtemp(hw) == IXGBE_ERR_OVERTEMP)
1858 if (!(eicr & IXGBE_EICR_GPI_SDP0))
1863 "Network adapter has been stopped because it has over heated. "
1864 "Restart the computer. If the problem persists, "
1865 "power off the system and replace the adapter\n");
1866 /* write to clear the interrupt */
1867 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP0);
1870 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
1872 struct ixgbe_hw *hw = &adapter->hw;
1874 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
1875 (eicr & IXGBE_EICR_GPI_SDP1)) {
1876 e_crit(probe, "Fan has stopped, replace the adapter\n");
1877 /* write to clear the interrupt */
1878 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1882 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
1884 struct ixgbe_hw *hw = &adapter->hw;
1886 if (eicr & IXGBE_EICR_GPI_SDP2) {
1887 /* Clear the interrupt */
1888 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
1889 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1890 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
1891 ixgbe_service_event_schedule(adapter);
1895 if (eicr & IXGBE_EICR_GPI_SDP1) {
1896 /* Clear the interrupt */
1897 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1898 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1899 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
1900 ixgbe_service_event_schedule(adapter);
1905 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1907 struct ixgbe_hw *hw = &adapter->hw;
1910 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1911 adapter->link_check_timeout = jiffies;
1912 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1913 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1914 IXGBE_WRITE_FLUSH(hw);
1915 ixgbe_service_event_schedule(adapter);
1919 static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
1921 struct net_device *netdev = data;
1922 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1923 struct ixgbe_hw *hw = &adapter->hw;
1927 * Workaround for Silicon errata. Use clear-by-write instead
1928 * of clear-by-read. Reading with EICS will return the
1929 * interrupt causes without clearing, which later be done
1930 * with the write to EICR.
1932 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1933 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1935 if (eicr & IXGBE_EICR_LSC)
1936 ixgbe_check_lsc(adapter);
1938 if (eicr & IXGBE_EICR_MAILBOX)
1939 ixgbe_msg_task(adapter);
1941 switch (hw->mac.type) {
1942 case ixgbe_mac_82599EB:
1943 ixgbe_check_sfp_event(adapter, eicr);
1944 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
1945 ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC))) {
1946 adapter->interrupt_event = eicr;
1947 schedule_work(&adapter->check_overtemp_task);
1949 /* now fallthrough to handle Flow Director */
1950 case ixgbe_mac_X540:
1951 /* Handle Flow Director Full threshold interrupt */
1952 if (eicr & IXGBE_EICR_FLOW_DIR) {
1954 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_FLOW_DIR);
1955 /* Disable transmits before FDIR Re-initialization */
1956 netif_tx_stop_all_queues(netdev);
1957 for (i = 0; i < adapter->num_tx_queues; i++) {
1958 struct ixgbe_ring *tx_ring =
1959 adapter->tx_ring[i];
1960 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
1962 schedule_work(&adapter->fdir_reinit_task);
1970 ixgbe_check_fan_failure(adapter, eicr);
1972 /* re-enable the original interrupt state, no lsc, no queues */
1973 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1974 IXGBE_WRITE_REG(hw, IXGBE_EIMS, eicr &
1975 ~(IXGBE_EIMS_LSC | IXGBE_EIMS_RTX_QUEUE));
1980 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
1984 struct ixgbe_hw *hw = &adapter->hw;
1986 switch (hw->mac.type) {
1987 case ixgbe_mac_82598EB:
1988 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1989 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
1991 case ixgbe_mac_82599EB:
1992 case ixgbe_mac_X540:
1993 mask = (qmask & 0xFFFFFFFF);
1995 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
1996 mask = (qmask >> 32);
1998 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2003 /* skip the flush */
2006 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
2010 struct ixgbe_hw *hw = &adapter->hw;
2012 switch (hw->mac.type) {
2013 case ixgbe_mac_82598EB:
2014 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2015 IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
2017 case ixgbe_mac_82599EB:
2018 case ixgbe_mac_X540:
2019 mask = (qmask & 0xFFFFFFFF);
2021 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
2022 mask = (qmask >> 32);
2024 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
2029 /* skip the flush */
2032 static irqreturn_t ixgbe_msix_clean_tx(int irq, void *data)
2034 struct ixgbe_q_vector *q_vector = data;
2035 struct ixgbe_adapter *adapter = q_vector->adapter;
2036 struct ixgbe_ring *tx_ring;
2039 if (!q_vector->txr_count)
2042 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
2043 for (i = 0; i < q_vector->txr_count; i++) {
2044 tx_ring = adapter->tx_ring[r_idx];
2045 tx_ring->total_bytes = 0;
2046 tx_ring->total_packets = 0;
2047 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
2051 /* EIAM disabled interrupts (on this vector) for us */
2052 napi_schedule(&q_vector->napi);
2058 * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
2060 * @data: pointer to our q_vector struct for this interrupt vector
2062 static irqreturn_t ixgbe_msix_clean_rx(int irq, void *data)
2064 struct ixgbe_q_vector *q_vector = data;
2065 struct ixgbe_adapter *adapter = q_vector->adapter;
2066 struct ixgbe_ring *rx_ring;
2070 #ifdef CONFIG_IXGBE_DCA
2071 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2072 ixgbe_update_dca(q_vector);
2075 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2076 for (i = 0; i < q_vector->rxr_count; i++) {
2077 rx_ring = adapter->rx_ring[r_idx];
2078 rx_ring->total_bytes = 0;
2079 rx_ring->total_packets = 0;
2080 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
2084 if (!q_vector->rxr_count)
2087 /* EIAM disabled interrupts (on this vector) for us */
2088 napi_schedule(&q_vector->napi);
2093 static irqreturn_t ixgbe_msix_clean_many(int irq, void *data)
2095 struct ixgbe_q_vector *q_vector = data;
2096 struct ixgbe_adapter *adapter = q_vector->adapter;
2097 struct ixgbe_ring *ring;
2101 if (!q_vector->txr_count && !q_vector->rxr_count)
2104 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
2105 for (i = 0; i < q_vector->txr_count; i++) {
2106 ring = adapter->tx_ring[r_idx];
2107 ring->total_bytes = 0;
2108 ring->total_packets = 0;
2109 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
2113 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2114 for (i = 0; i < q_vector->rxr_count; i++) {
2115 ring = adapter->rx_ring[r_idx];
2116 ring->total_bytes = 0;
2117 ring->total_packets = 0;
2118 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
2122 /* EIAM disabled interrupts (on this vector) for us */
2123 napi_schedule(&q_vector->napi);
2129 * ixgbe_clean_rxonly - 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 is optimized for cleaning one queue only on a single
2136 static int ixgbe_clean_rxonly(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 *rx_ring = NULL;
2145 #ifdef CONFIG_IXGBE_DCA
2146 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2147 ixgbe_update_dca(q_vector);
2150 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2151 rx_ring = adapter->rx_ring[r_idx];
2153 ixgbe_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
2155 /* If all Rx work done, exit the polling mode */
2156 if (work_done < budget) {
2157 napi_complete(napi);
2158 if (adapter->rx_itr_setting & 1)
2159 ixgbe_set_itr_msix(q_vector);
2160 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2161 ixgbe_irq_enable_queues(adapter,
2162 ((u64)1 << q_vector->v_idx));
2169 * ixgbe_clean_rxtx_many - msix (aka one shot) rx clean routine
2170 * @napi: napi struct with our devices info in it
2171 * @budget: amount of work driver is allowed to do this pass, in packets
2173 * This function will clean more than one rx queue associated with a
2176 static int ixgbe_clean_rxtx_many(struct napi_struct *napi, int budget)
2178 struct ixgbe_q_vector *q_vector =
2179 container_of(napi, struct ixgbe_q_vector, napi);
2180 struct ixgbe_adapter *adapter = q_vector->adapter;
2181 struct ixgbe_ring *ring = NULL;
2182 int work_done = 0, i;
2184 bool tx_clean_complete = true;
2186 #ifdef CONFIG_IXGBE_DCA
2187 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2188 ixgbe_update_dca(q_vector);
2191 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
2192 for (i = 0; i < q_vector->txr_count; i++) {
2193 ring = adapter->tx_ring[r_idx];
2194 tx_clean_complete &= ixgbe_clean_tx_irq(q_vector, ring);
2195 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
2199 /* attempt to distribute budget to each queue fairly, but don't allow
2200 * the budget to go below 1 because we'll exit polling */
2201 budget /= (q_vector->rxr_count ?: 1);
2202 budget = max(budget, 1);
2203 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2204 for (i = 0; i < q_vector->rxr_count; i++) {
2205 ring = adapter->rx_ring[r_idx];
2206 ixgbe_clean_rx_irq(q_vector, ring, &work_done, budget);
2207 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
2211 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2212 ring = adapter->rx_ring[r_idx];
2213 /* If all Rx work done, exit the polling mode */
2214 if (work_done < budget) {
2215 napi_complete(napi);
2216 if (adapter->rx_itr_setting & 1)
2217 ixgbe_set_itr_msix(q_vector);
2218 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2219 ixgbe_irq_enable_queues(adapter,
2220 ((u64)1 << q_vector->v_idx));
2228 * ixgbe_clean_txonly - msix (aka one shot) tx clean routine
2229 * @napi: napi struct with our devices info in it
2230 * @budget: amount of work driver is allowed to do this pass, in packets
2232 * This function is optimized for cleaning one queue only on a single
2235 static int ixgbe_clean_txonly(struct napi_struct *napi, int budget)
2237 struct ixgbe_q_vector *q_vector =
2238 container_of(napi, struct ixgbe_q_vector, napi);
2239 struct ixgbe_adapter *adapter = q_vector->adapter;
2240 struct ixgbe_ring *tx_ring = NULL;
2244 #ifdef CONFIG_IXGBE_DCA
2245 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2246 ixgbe_update_dca(q_vector);
2249 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
2250 tx_ring = adapter->tx_ring[r_idx];
2252 if (!ixgbe_clean_tx_irq(q_vector, tx_ring))
2255 /* If all Tx work done, exit the polling mode */
2256 if (work_done < budget) {
2257 napi_complete(napi);
2258 if (adapter->tx_itr_setting & 1)
2259 ixgbe_set_itr_msix(q_vector);
2260 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2261 ixgbe_irq_enable_queues(adapter,
2262 ((u64)1 << q_vector->v_idx));
2268 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
2271 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2272 struct ixgbe_ring *rx_ring = a->rx_ring[r_idx];
2274 set_bit(r_idx, q_vector->rxr_idx);
2275 q_vector->rxr_count++;
2276 rx_ring->q_vector = q_vector;
2279 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
2282 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2283 struct ixgbe_ring *tx_ring = a->tx_ring[t_idx];
2285 set_bit(t_idx, q_vector->txr_idx);
2286 q_vector->txr_count++;
2287 tx_ring->q_vector = q_vector;
2291 * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
2292 * @adapter: board private structure to initialize
2294 * This function maps descriptor rings to the queue-specific vectors
2295 * we were allotted through the MSI-X enabling code. Ideally, we'd have
2296 * one vector per ring/queue, but on a constrained vector budget, we
2297 * group the rings as "efficiently" as possible. You would add new
2298 * mapping configurations in here.
2300 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter)
2304 int rxr_idx = 0, txr_idx = 0;
2305 int rxr_remaining = adapter->num_rx_queues;
2306 int txr_remaining = adapter->num_tx_queues;
2311 /* No mapping required if MSI-X is disabled. */
2312 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2315 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2318 * The ideal configuration...
2319 * We have enough vectors to map one per queue.
2321 if (q_vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
2322 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
2323 map_vector_to_rxq(adapter, v_start, rxr_idx);
2325 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
2326 map_vector_to_txq(adapter, v_start, txr_idx);
2332 * If we don't have enough vectors for a 1-to-1
2333 * mapping, we'll have to group them so there are
2334 * multiple queues per vector.
2336 /* Re-adjusting *qpv takes care of the remainder. */
2337 for (i = v_start; i < q_vectors; i++) {
2338 rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - i);
2339 for (j = 0; j < rqpv; j++) {
2340 map_vector_to_rxq(adapter, i, rxr_idx);
2344 tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - i);
2345 for (j = 0; j < tqpv; j++) {
2346 map_vector_to_txq(adapter, i, txr_idx);
2356 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2357 * @adapter: board private structure
2359 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2360 * interrupts from the kernel.
2362 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2364 struct net_device *netdev = adapter->netdev;
2365 irqreturn_t (*handler)(int, void *);
2366 int i, vector, q_vectors, err;
2369 /* Decrement for Other and TCP Timer vectors */
2370 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2372 err = ixgbe_map_rings_to_vectors(adapter);
2376 #define SET_HANDLER(_v) (((_v)->rxr_count && (_v)->txr_count) \
2377 ? &ixgbe_msix_clean_many : \
2378 (_v)->rxr_count ? &ixgbe_msix_clean_rx : \
2379 (_v)->txr_count ? &ixgbe_msix_clean_tx : \
2381 for (vector = 0; vector < q_vectors; vector++) {
2382 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2383 handler = SET_HANDLER(q_vector);
2385 if (handler == &ixgbe_msix_clean_rx) {
2386 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2387 "%s-%s-%d", netdev->name, "rx", ri++);
2388 } else if (handler == &ixgbe_msix_clean_tx) {
2389 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2390 "%s-%s-%d", netdev->name, "tx", ti++);
2391 } else if (handler == &ixgbe_msix_clean_many) {
2392 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2393 "%s-%s-%d", netdev->name, "TxRx", ri++);
2396 /* skip this unused q_vector */
2399 err = request_irq(adapter->msix_entries[vector].vector,
2400 handler, 0, q_vector->name,
2403 e_err(probe, "request_irq failed for MSIX interrupt "
2404 "Error: %d\n", err);
2405 goto free_queue_irqs;
2409 sprintf(adapter->lsc_int_name, "%s:lsc", netdev->name);
2410 err = request_irq(adapter->msix_entries[vector].vector,
2411 ixgbe_msix_lsc, 0, adapter->lsc_int_name, netdev);
2413 e_err(probe, "request_irq for msix_lsc failed: %d\n", err);
2414 goto free_queue_irqs;
2420 for (i = vector - 1; i >= 0; i--)
2421 free_irq(adapter->msix_entries[--vector].vector,
2422 adapter->q_vector[i]);
2423 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2424 pci_disable_msix(adapter->pdev);
2425 kfree(adapter->msix_entries);
2426 adapter->msix_entries = NULL;
2430 static void ixgbe_set_itr(struct ixgbe_adapter *adapter)
2432 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2433 struct ixgbe_ring *rx_ring = adapter->rx_ring[0];
2434 struct ixgbe_ring *tx_ring = adapter->tx_ring[0];
2435 u32 new_itr = q_vector->eitr;
2438 q_vector->tx_itr = ixgbe_update_itr(adapter, new_itr,
2440 tx_ring->total_packets,
2441 tx_ring->total_bytes);
2442 q_vector->rx_itr = ixgbe_update_itr(adapter, new_itr,
2444 rx_ring->total_packets,
2445 rx_ring->total_bytes);
2447 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
2449 switch (current_itr) {
2450 /* counts and packets in update_itr are dependent on these numbers */
2451 case lowest_latency:
2455 new_itr = 20000; /* aka hwitr = ~200 */
2464 if (new_itr != q_vector->eitr) {
2465 /* do an exponential smoothing */
2466 new_itr = ((q_vector->eitr * 9) + new_itr)/10;
2468 /* save the algorithm value here */
2469 q_vector->eitr = new_itr;
2471 ixgbe_write_eitr(q_vector);
2476 * ixgbe_irq_enable - Enable default interrupt generation settings
2477 * @adapter: board private structure
2479 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2484 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2485 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2486 mask |= IXGBE_EIMS_GPI_SDP0;
2487 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2488 mask |= IXGBE_EIMS_GPI_SDP1;
2489 switch (adapter->hw.mac.type) {
2490 case ixgbe_mac_82599EB:
2491 case ixgbe_mac_X540:
2492 mask |= IXGBE_EIMS_ECC;
2493 mask |= IXGBE_EIMS_GPI_SDP1;
2494 mask |= IXGBE_EIMS_GPI_SDP2;
2495 if (adapter->num_vfs)
2496 mask |= IXGBE_EIMS_MAILBOX;
2501 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
2502 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
2503 mask |= IXGBE_EIMS_FLOW_DIR;
2505 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2507 ixgbe_irq_enable_queues(adapter, ~0);
2509 IXGBE_WRITE_FLUSH(&adapter->hw);
2511 if (adapter->num_vfs > 32) {
2512 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
2513 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2518 * ixgbe_intr - legacy mode Interrupt Handler
2519 * @irq: interrupt number
2520 * @data: pointer to a network interface device structure
2522 static irqreturn_t ixgbe_intr(int irq, void *data)
2524 struct net_device *netdev = data;
2525 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2526 struct ixgbe_hw *hw = &adapter->hw;
2527 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2531 * Workaround for silicon errata on 82598. Mask the interrupts
2532 * before the read of EICR.
2534 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2536 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2537 * therefore no explict interrupt disable is necessary */
2538 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2541 * shared interrupt alert!
2542 * make sure interrupts are enabled because the read will
2543 * have disabled interrupts due to EIAM
2544 * finish the workaround of silicon errata on 82598. Unmask
2545 * the interrupt that we masked before the EICR read.
2547 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2548 ixgbe_irq_enable(adapter, true, true);
2549 return IRQ_NONE; /* Not our interrupt */
2552 if (eicr & IXGBE_EICR_LSC)
2553 ixgbe_check_lsc(adapter);
2555 switch (hw->mac.type) {
2556 case ixgbe_mac_82599EB:
2557 ixgbe_check_sfp_event(adapter, eicr);
2558 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2559 ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC))) {
2560 adapter->interrupt_event = eicr;
2561 schedule_work(&adapter->check_overtemp_task);
2568 ixgbe_check_fan_failure(adapter, eicr);
2570 if (napi_schedule_prep(&(q_vector->napi))) {
2571 adapter->tx_ring[0]->total_packets = 0;
2572 adapter->tx_ring[0]->total_bytes = 0;
2573 adapter->rx_ring[0]->total_packets = 0;
2574 adapter->rx_ring[0]->total_bytes = 0;
2575 /* would disable interrupts here but EIAM disabled it */
2576 __napi_schedule(&(q_vector->napi));
2580 * re-enable link(maybe) and non-queue interrupts, no flush.
2581 * ixgbe_poll will re-enable the queue interrupts
2584 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2585 ixgbe_irq_enable(adapter, false, false);
2590 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
2592 int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2594 for (i = 0; i < q_vectors; i++) {
2595 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
2596 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
2597 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
2598 q_vector->rxr_count = 0;
2599 q_vector->txr_count = 0;
2604 * ixgbe_request_irq - initialize interrupts
2605 * @adapter: board private structure
2607 * Attempts to configure interrupts using the best available
2608 * capabilities of the hardware and kernel.
2610 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2612 struct net_device *netdev = adapter->netdev;
2615 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2616 err = ixgbe_request_msix_irqs(adapter);
2617 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
2618 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2619 netdev->name, netdev);
2621 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2622 netdev->name, netdev);
2626 e_err(probe, "request_irq failed, Error %d\n", err);
2631 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2633 struct net_device *netdev = adapter->netdev;
2635 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2638 q_vectors = adapter->num_msix_vectors;
2641 free_irq(adapter->msix_entries[i].vector, netdev);
2644 for (; i >= 0; i--) {
2645 /* free only the irqs that were actually requested */
2646 if (!adapter->q_vector[i]->rxr_count &&
2647 !adapter->q_vector[i]->txr_count)
2650 free_irq(adapter->msix_entries[i].vector,
2651 adapter->q_vector[i]);
2654 ixgbe_reset_q_vectors(adapter);
2656 free_irq(adapter->pdev->irq, netdev);
2661 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2662 * @adapter: board private structure
2664 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2666 switch (adapter->hw.mac.type) {
2667 case ixgbe_mac_82598EB:
2668 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2670 case ixgbe_mac_82599EB:
2671 case ixgbe_mac_X540:
2672 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2673 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2674 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2675 if (adapter->num_vfs > 32)
2676 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
2681 IXGBE_WRITE_FLUSH(&adapter->hw);
2682 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2684 for (i = 0; i < adapter->num_msix_vectors; i++)
2685 synchronize_irq(adapter->msix_entries[i].vector);
2687 synchronize_irq(adapter->pdev->irq);
2692 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2695 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2697 struct ixgbe_hw *hw = &adapter->hw;
2699 IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
2700 EITR_INTS_PER_SEC_TO_REG(adapter->rx_eitr_param));
2702 ixgbe_set_ivar(adapter, 0, 0, 0);
2703 ixgbe_set_ivar(adapter, 1, 0, 0);
2705 map_vector_to_rxq(adapter, 0, 0);
2706 map_vector_to_txq(adapter, 0, 0);
2708 e_info(hw, "Legacy interrupt IVAR setup done\n");
2712 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2713 * @adapter: board private structure
2714 * @ring: structure containing ring specific data
2716 * Configure the Tx descriptor ring after a reset.
2718 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2719 struct ixgbe_ring *ring)
2721 struct ixgbe_hw *hw = &adapter->hw;
2722 u64 tdba = ring->dma;
2725 u8 reg_idx = ring->reg_idx;
2727 /* disable queue to avoid issues while updating state */
2728 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2729 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx),
2730 txdctl & ~IXGBE_TXDCTL_ENABLE);
2731 IXGBE_WRITE_FLUSH(hw);
2733 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2734 (tdba & DMA_BIT_MASK(32)));
2735 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2736 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2737 ring->count * sizeof(union ixgbe_adv_tx_desc));
2738 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2739 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2740 ring->tail = hw->hw_addr + IXGBE_TDT(reg_idx);
2742 /* configure fetching thresholds */
2743 if (adapter->rx_itr_setting == 0) {
2744 /* cannot set wthresh when itr==0 */
2745 txdctl &= ~0x007F0000;
2747 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
2748 txdctl |= (8 << 16);
2750 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2751 /* PThresh workaround for Tx hang with DFP enabled. */
2755 /* reinitialize flowdirector state */
2756 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2757 adapter->atr_sample_rate) {
2758 ring->atr_sample_rate = adapter->atr_sample_rate;
2759 ring->atr_count = 0;
2760 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
2762 ring->atr_sample_rate = 0;
2765 clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
2768 txdctl |= IXGBE_TXDCTL_ENABLE;
2769 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
2771 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2772 if (hw->mac.type == ixgbe_mac_82598EB &&
2773 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2776 /* poll to verify queue is enabled */
2778 usleep_range(1000, 2000);
2779 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2780 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
2782 e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
2785 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
2787 struct ixgbe_hw *hw = &adapter->hw;
2791 if (hw->mac.type == ixgbe_mac_82598EB)
2794 /* disable the arbiter while setting MTQC */
2795 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2796 rttdcs |= IXGBE_RTTDCS_ARBDIS;
2797 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2799 /* set transmit pool layout */
2800 mask = (IXGBE_FLAG_SRIOV_ENABLED | IXGBE_FLAG_DCB_ENABLED);
2801 switch (adapter->flags & mask) {
2803 case (IXGBE_FLAG_SRIOV_ENABLED):
2804 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2805 (IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF));
2808 case (IXGBE_FLAG_DCB_ENABLED):
2809 /* We enable 8 traffic classes, DCB only */
2810 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2811 (IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ));
2815 IXGBE_WRITE_REG(hw, IXGBE_MTQC, IXGBE_MTQC_64Q_1PB);
2819 /* re-enable the arbiter */
2820 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2821 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2825 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2826 * @adapter: board private structure
2828 * Configure the Tx unit of the MAC after a reset.
2830 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2832 struct ixgbe_hw *hw = &adapter->hw;
2836 ixgbe_setup_mtqc(adapter);
2838 if (hw->mac.type != ixgbe_mac_82598EB) {
2839 /* DMATXCTL.EN must be before Tx queues are enabled */
2840 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2841 dmatxctl |= IXGBE_DMATXCTL_TE;
2842 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2845 /* Setup the HW Tx Head and Tail descriptor pointers */
2846 for (i = 0; i < adapter->num_tx_queues; i++)
2847 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
2850 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2852 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2853 struct ixgbe_ring *rx_ring)
2856 u8 reg_idx = rx_ring->reg_idx;
2858 switch (adapter->hw.mac.type) {
2859 case ixgbe_mac_82598EB: {
2860 struct ixgbe_ring_feature *feature = adapter->ring_feature;
2861 const int mask = feature[RING_F_RSS].mask;
2862 reg_idx = reg_idx & mask;
2865 case ixgbe_mac_82599EB:
2866 case ixgbe_mac_X540:
2871 srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx));
2873 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
2874 srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
2875 if (adapter->num_vfs)
2876 srrctl |= IXGBE_SRRCTL_DROP_EN;
2878 srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
2879 IXGBE_SRRCTL_BSIZEHDR_MASK;
2881 if (ring_is_ps_enabled(rx_ring)) {
2882 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
2883 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2885 srrctl |= (PAGE_SIZE / 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2887 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2889 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
2890 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2891 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2894 IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx), srrctl);
2897 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2899 struct ixgbe_hw *hw = &adapter->hw;
2900 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2901 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2902 0x6A3E67EA, 0x14364D17, 0x3BED200D};
2903 u32 mrqc = 0, reta = 0;
2908 /* Fill out hash function seeds */
2909 for (i = 0; i < 10; i++)
2910 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2912 /* Fill out redirection table */
2913 for (i = 0, j = 0; i < 128; i++, j++) {
2914 if (j == adapter->ring_feature[RING_F_RSS].indices)
2916 /* reta = 4-byte sliding window of
2917 * 0x00..(indices-1)(indices-1)00..etc. */
2918 reta = (reta << 8) | (j * 0x11);
2920 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2923 /* Disable indicating checksum in descriptor, enables RSS hash */
2924 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2925 rxcsum |= IXGBE_RXCSUM_PCSD;
2926 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2928 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
2929 mask = adapter->flags & IXGBE_FLAG_RSS_ENABLED;
2931 mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2932 #ifdef CONFIG_IXGBE_DCB
2933 | IXGBE_FLAG_DCB_ENABLED
2935 | IXGBE_FLAG_SRIOV_ENABLED
2939 #ifdef CONFIG_IXGBE_DCB
2940 case (IXGBE_FLAG_DCB_ENABLED | IXGBE_FLAG_RSS_ENABLED):
2941 mrqc = IXGBE_MRQC_RTRSS8TCEN;
2943 case (IXGBE_FLAG_DCB_ENABLED):
2944 mrqc = IXGBE_MRQC_RT8TCEN;
2946 #endif /* CONFIG_IXGBE_DCB */
2947 case (IXGBE_FLAG_RSS_ENABLED):
2948 mrqc = IXGBE_MRQC_RSSEN;
2950 case (IXGBE_FLAG_SRIOV_ENABLED):
2951 mrqc = IXGBE_MRQC_VMDQEN;
2957 /* Perform hash on these packet types */
2958 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2959 | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2960 | IXGBE_MRQC_RSS_FIELD_IPV6
2961 | IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
2963 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2967 * ixgbe_clear_rscctl - disable RSC for the indicated ring
2968 * @adapter: address of board private structure
2969 * @ring: structure containing ring specific data
2971 void ixgbe_clear_rscctl(struct ixgbe_adapter *adapter,
2972 struct ixgbe_ring *ring)
2974 struct ixgbe_hw *hw = &adapter->hw;
2976 u8 reg_idx = ring->reg_idx;
2978 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
2979 rscctrl &= ~IXGBE_RSCCTL_RSCEN;
2980 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
2984 * ixgbe_configure_rscctl - enable RSC for the indicated ring
2985 * @adapter: address of board private structure
2986 * @index: index of ring to set
2988 void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
2989 struct ixgbe_ring *ring)
2991 struct ixgbe_hw *hw = &adapter->hw;
2994 u8 reg_idx = ring->reg_idx;
2996 if (!ring_is_rsc_enabled(ring))
2999 rx_buf_len = ring->rx_buf_len;
3000 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
3001 rscctrl |= IXGBE_RSCCTL_RSCEN;
3003 * we must limit the number of descriptors so that the
3004 * total size of max desc * buf_len is not greater
3007 if (ring_is_ps_enabled(ring)) {
3008 #if (MAX_SKB_FRAGS > 16)
3009 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
3010 #elif (MAX_SKB_FRAGS > 8)
3011 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
3012 #elif (MAX_SKB_FRAGS > 4)
3013 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
3015 rscctrl |= IXGBE_RSCCTL_MAXDESC_1;
3018 if (rx_buf_len < IXGBE_RXBUFFER_4096)
3019 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
3020 else if (rx_buf_len < IXGBE_RXBUFFER_8192)
3021 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
3023 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
3025 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
3029 * ixgbe_set_uta - Set unicast filter table address
3030 * @adapter: board private structure
3032 * The unicast table address is a register array of 32-bit registers.
3033 * The table is meant to be used in a way similar to how the MTA is used
3034 * however due to certain limitations in the hardware it is necessary to
3035 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
3036 * enable bit to allow vlan tag stripping when promiscuous mode is enabled
3038 static void ixgbe_set_uta(struct ixgbe_adapter *adapter)
3040 struct ixgbe_hw *hw = &adapter->hw;
3043 /* The UTA table only exists on 82599 hardware and newer */
3044 if (hw->mac.type < ixgbe_mac_82599EB)
3047 /* we only need to do this if VMDq is enabled */
3048 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
3051 for (i = 0; i < 128; i++)
3052 IXGBE_WRITE_REG(hw, IXGBE_UTA(i), ~0);
3055 #define IXGBE_MAX_RX_DESC_POLL 10
3056 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
3057 struct ixgbe_ring *ring)
3059 struct ixgbe_hw *hw = &adapter->hw;
3060 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3062 u8 reg_idx = ring->reg_idx;
3064 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3065 if (hw->mac.type == ixgbe_mac_82598EB &&
3066 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3070 usleep_range(1000, 2000);
3071 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3072 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
3075 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
3076 "the polling period\n", reg_idx);
3080 void ixgbe_disable_rx_queue(struct ixgbe_adapter *adapter,
3081 struct ixgbe_ring *ring)
3083 struct ixgbe_hw *hw = &adapter->hw;
3084 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3086 u8 reg_idx = ring->reg_idx;
3088 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3089 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
3091 /* write value back with RXDCTL.ENABLE bit cleared */
3092 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3094 if (hw->mac.type == ixgbe_mac_82598EB &&
3095 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3098 /* the hardware may take up to 100us to really disable the rx queue */
3101 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3102 } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
3105 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not cleared within "
3106 "the polling period\n", reg_idx);
3110 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
3111 struct ixgbe_ring *ring)
3113 struct ixgbe_hw *hw = &adapter->hw;
3114 u64 rdba = ring->dma;
3116 u8 reg_idx = ring->reg_idx;
3118 /* disable queue to avoid issues while updating state */
3119 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3120 ixgbe_disable_rx_queue(adapter, ring);
3122 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
3123 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
3124 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
3125 ring->count * sizeof(union ixgbe_adv_rx_desc));
3126 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
3127 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
3128 ring->tail = hw->hw_addr + IXGBE_RDT(reg_idx);
3130 ixgbe_configure_srrctl(adapter, ring);
3131 ixgbe_configure_rscctl(adapter, ring);
3133 /* If operating in IOV mode set RLPML for X540 */
3134 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
3135 hw->mac.type == ixgbe_mac_X540) {
3136 rxdctl &= ~IXGBE_RXDCTL_RLPMLMASK;
3137 rxdctl |= ((ring->netdev->mtu + ETH_HLEN +
3138 ETH_FCS_LEN + VLAN_HLEN) | IXGBE_RXDCTL_RLPML_EN);
3141 if (hw->mac.type == ixgbe_mac_82598EB) {
3143 * enable cache line friendly hardware writes:
3144 * PTHRESH=32 descriptors (half the internal cache),
3145 * this also removes ugly rx_no_buffer_count increment
3146 * HTHRESH=4 descriptors (to minimize latency on fetch)
3147 * WTHRESH=8 burst writeback up to two cache lines
3149 rxdctl &= ~0x3FFFFF;
3153 /* enable receive descriptor ring */
3154 rxdctl |= IXGBE_RXDCTL_ENABLE;
3155 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3157 ixgbe_rx_desc_queue_enable(adapter, ring);
3158 ixgbe_alloc_rx_buffers(ring, IXGBE_DESC_UNUSED(ring));
3161 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
3163 struct ixgbe_hw *hw = &adapter->hw;
3166 /* PSRTYPE must be initialized in non 82598 adapters */
3167 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
3168 IXGBE_PSRTYPE_UDPHDR |
3169 IXGBE_PSRTYPE_IPV4HDR |
3170 IXGBE_PSRTYPE_L2HDR |
3171 IXGBE_PSRTYPE_IPV6HDR;
3173 if (hw->mac.type == ixgbe_mac_82598EB)
3176 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED)
3177 psrtype |= (adapter->num_rx_queues_per_pool << 29);
3179 for (p = 0; p < adapter->num_rx_pools; p++)
3180 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(adapter->num_vfs + p),
3184 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
3186 struct ixgbe_hw *hw = &adapter->hw;
3189 u32 reg_offset, vf_shift;
3192 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
3195 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
3196 vt_reg_bits = IXGBE_VMD_CTL_VMDQ_EN | IXGBE_VT_CTL_REPLEN;
3197 vt_reg_bits |= (adapter->num_vfs << IXGBE_VT_CTL_POOL_SHIFT);
3198 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl | vt_reg_bits);
3200 vf_shift = adapter->num_vfs % 32;
3201 reg_offset = (adapter->num_vfs > 32) ? 1 : 0;
3203 /* Enable only the PF's pool for Tx/Rx */
3204 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (1 << vf_shift));
3205 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), 0);
3206 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (1 << vf_shift));
3207 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), 0);
3208 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3210 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
3211 hw->mac.ops.set_vmdq(hw, 0, adapter->num_vfs);
3214 * Set up VF register offsets for selected VT Mode,
3215 * i.e. 32 or 64 VFs for SR-IOV
3217 gcr_ext = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
3218 gcr_ext |= IXGBE_GCR_EXT_MSIX_EN;
3219 gcr_ext |= IXGBE_GCR_EXT_VT_MODE_64;
3220 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
3222 /* enable Tx loopback for VF/PF communication */
3223 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3224 /* Enable MAC Anti-Spoofing */
3225 hw->mac.ops.set_mac_anti_spoofing(hw,
3226 (adapter->antispoofing_enabled =
3227 (adapter->num_vfs != 0)),
3231 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
3233 struct ixgbe_hw *hw = &adapter->hw;
3234 struct net_device *netdev = adapter->netdev;
3235 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3237 struct ixgbe_ring *rx_ring;
3241 /* Decide whether to use packet split mode or not */
3243 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
3245 /* Do not use packet split if we're in SR-IOV Mode */
3246 if (adapter->num_vfs)
3247 adapter->flags &= ~IXGBE_FLAG_RX_PS_ENABLED;
3249 /* Disable packet split due to 82599 erratum #45 */
3250 if (hw->mac.type == ixgbe_mac_82599EB)
3251 adapter->flags &= ~IXGBE_FLAG_RX_PS_ENABLED;
3253 /* Set the RX buffer length according to the mode */
3254 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
3255 rx_buf_len = IXGBE_RX_HDR_SIZE;
3257 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
3258 (netdev->mtu <= ETH_DATA_LEN))
3259 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3261 rx_buf_len = ALIGN(max_frame + VLAN_HLEN, 1024);
3265 /* adjust max frame to be able to do baby jumbo for FCoE */
3266 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
3267 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
3268 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3270 #endif /* IXGBE_FCOE */
3271 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
3272 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
3273 mhadd &= ~IXGBE_MHADD_MFS_MASK;
3274 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
3276 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
3279 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
3280 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
3281 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
3282 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
3285 * Setup the HW Rx Head and Tail Descriptor Pointers and
3286 * the Base and Length of the Rx Descriptor Ring
3288 for (i = 0; i < adapter->num_rx_queues; i++) {
3289 rx_ring = adapter->rx_ring[i];
3290 rx_ring->rx_buf_len = rx_buf_len;
3292 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)
3293 set_ring_ps_enabled(rx_ring);
3295 clear_ring_ps_enabled(rx_ring);
3297 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
3298 set_ring_rsc_enabled(rx_ring);
3300 clear_ring_rsc_enabled(rx_ring);
3303 if (netdev->features & NETIF_F_FCOE_MTU) {
3304 struct ixgbe_ring_feature *f;
3305 f = &adapter->ring_feature[RING_F_FCOE];
3306 if ((i >= f->mask) && (i < f->mask + f->indices)) {
3307 clear_ring_ps_enabled(rx_ring);
3308 if (rx_buf_len < IXGBE_FCOE_JUMBO_FRAME_SIZE)
3309 rx_ring->rx_buf_len =
3310 IXGBE_FCOE_JUMBO_FRAME_SIZE;
3311 } else if (!ring_is_rsc_enabled(rx_ring) &&
3312 !ring_is_ps_enabled(rx_ring)) {
3313 rx_ring->rx_buf_len =
3314 IXGBE_FCOE_JUMBO_FRAME_SIZE;
3317 #endif /* IXGBE_FCOE */
3321 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
3323 struct ixgbe_hw *hw = &adapter->hw;
3324 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
3326 switch (hw->mac.type) {
3327 case ixgbe_mac_82598EB:
3329 * For VMDq support of different descriptor types or
3330 * buffer sizes through the use of multiple SRRCTL
3331 * registers, RDRXCTL.MVMEN must be set to 1
3333 * also, the manual doesn't mention it clearly but DCA hints
3334 * will only use queue 0's tags unless this bit is set. Side
3335 * effects of setting this bit are only that SRRCTL must be
3336 * fully programmed [0..15]
3338 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
3340 case ixgbe_mac_82599EB:
3341 case ixgbe_mac_X540:
3342 /* Disable RSC for ACK packets */
3343 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
3344 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
3345 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3346 /* hardware requires some bits to be set by default */
3347 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3348 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3351 /* We should do nothing since we don't know this hardware */
3355 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3359 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3360 * @adapter: board private structure
3362 * Configure the Rx unit of the MAC after a reset.
3364 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3366 struct ixgbe_hw *hw = &adapter->hw;
3370 /* disable receives while setting up the descriptors */
3371 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3372 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3374 ixgbe_setup_psrtype(adapter);
3375 ixgbe_setup_rdrxctl(adapter);
3377 /* Program registers for the distribution of queues */
3378 ixgbe_setup_mrqc(adapter);
3380 ixgbe_set_uta(adapter);
3382 /* set_rx_buffer_len must be called before ring initialization */
3383 ixgbe_set_rx_buffer_len(adapter);
3386 * Setup the HW Rx Head and Tail Descriptor Pointers and
3387 * the Base and Length of the Rx Descriptor Ring
3389 for (i = 0; i < adapter->num_rx_queues; i++)
3390 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3392 /* disable drop enable for 82598 parts */
3393 if (hw->mac.type == ixgbe_mac_82598EB)
3394 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3396 /* enable all receives */
3397 rxctrl |= IXGBE_RXCTRL_RXEN;
3398 hw->mac.ops.enable_rx_dma(hw, rxctrl);
3401 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3403 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3404 struct ixgbe_hw *hw = &adapter->hw;
3405 int pool_ndx = adapter->num_vfs;
3407 /* add VID to filter table */
3408 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, true);
3409 set_bit(vid, adapter->active_vlans);
3412 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3414 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3415 struct ixgbe_hw *hw = &adapter->hw;
3416 int pool_ndx = adapter->num_vfs;
3418 /* remove VID from filter table */
3419 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, false);
3420 clear_bit(vid, adapter->active_vlans);
3424 * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3425 * @adapter: driver data
3427 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
3429 struct ixgbe_hw *hw = &adapter->hw;
3432 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3433 vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
3434 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3438 * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3439 * @adapter: driver data
3441 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
3443 struct ixgbe_hw *hw = &adapter->hw;
3446 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3447 vlnctrl |= IXGBE_VLNCTRL_VFE;
3448 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3449 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3453 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3454 * @adapter: driver data
3456 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3458 struct ixgbe_hw *hw = &adapter->hw;
3462 switch (hw->mac.type) {
3463 case ixgbe_mac_82598EB:
3464 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3465 vlnctrl &= ~IXGBE_VLNCTRL_VME;
3466 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3468 case ixgbe_mac_82599EB:
3469 case ixgbe_mac_X540:
3470 for (i = 0; i < adapter->num_rx_queues; i++) {
3471 j = adapter->rx_ring[i]->reg_idx;
3472 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3473 vlnctrl &= ~IXGBE_RXDCTL_VME;
3474 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3483 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3484 * @adapter: driver data
3486 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3488 struct ixgbe_hw *hw = &adapter->hw;
3492 switch (hw->mac.type) {
3493 case ixgbe_mac_82598EB:
3494 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3495 vlnctrl |= IXGBE_VLNCTRL_VME;
3496 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3498 case ixgbe_mac_82599EB:
3499 case ixgbe_mac_X540:
3500 for (i = 0; i < adapter->num_rx_queues; i++) {
3501 j = adapter->rx_ring[i]->reg_idx;
3502 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3503 vlnctrl |= IXGBE_RXDCTL_VME;
3504 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3512 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3516 ixgbe_vlan_rx_add_vid(adapter->netdev, 0);
3518 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3519 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
3523 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3524 * @netdev: network interface device structure
3526 * Writes unicast address list to the RAR table.
3527 * Returns: -ENOMEM on failure/insufficient address space
3528 * 0 on no addresses written
3529 * X on writing X addresses to the RAR table
3531 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3533 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3534 struct ixgbe_hw *hw = &adapter->hw;
3535 unsigned int vfn = adapter->num_vfs;
3536 unsigned int rar_entries = IXGBE_MAX_PF_MACVLANS;
3539 /* return ENOMEM indicating insufficient memory for addresses */
3540 if (netdev_uc_count(netdev) > rar_entries)
3543 if (!netdev_uc_empty(netdev) && rar_entries) {
3544 struct netdev_hw_addr *ha;
3545 /* return error if we do not support writing to RAR table */
3546 if (!hw->mac.ops.set_rar)
3549 netdev_for_each_uc_addr(ha, netdev) {
3552 hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3557 /* write the addresses in reverse order to avoid write combining */
3558 for (; rar_entries > 0 ; rar_entries--)
3559 hw->mac.ops.clear_rar(hw, rar_entries);
3565 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3566 * @netdev: network interface device structure
3568 * The set_rx_method entry point is called whenever the unicast/multicast
3569 * address list or the network interface flags are updated. This routine is
3570 * responsible for configuring the hardware for proper unicast, multicast and
3573 void ixgbe_set_rx_mode(struct net_device *netdev)
3575 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3576 struct ixgbe_hw *hw = &adapter->hw;
3577 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3580 /* Check for Promiscuous and All Multicast modes */
3582 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3584 /* set all bits that we expect to always be set */
3585 fctrl |= IXGBE_FCTRL_BAM;
3586 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3587 fctrl |= IXGBE_FCTRL_PMCF;
3589 /* clear the bits we are changing the status of */
3590 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3592 if (netdev->flags & IFF_PROMISC) {
3593 hw->addr_ctrl.user_set_promisc = true;
3594 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3595 vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3596 /* don't hardware filter vlans in promisc mode */
3597 ixgbe_vlan_filter_disable(adapter);
3599 if (netdev->flags & IFF_ALLMULTI) {
3600 fctrl |= IXGBE_FCTRL_MPE;
3601 vmolr |= IXGBE_VMOLR_MPE;
3604 * Write addresses to the MTA, if the attempt fails
3605 * then we should just turn on promiscuous mode so
3606 * that we can at least receive multicast traffic
3608 hw->mac.ops.update_mc_addr_list(hw, netdev);
3609 vmolr |= IXGBE_VMOLR_ROMPE;
3611 ixgbe_vlan_filter_enable(adapter);
3612 hw->addr_ctrl.user_set_promisc = false;
3614 * Write addresses to available RAR registers, if there is not
3615 * sufficient space to store all the addresses then enable
3616 * unicast promiscuous mode
3618 count = ixgbe_write_uc_addr_list(netdev);
3620 fctrl |= IXGBE_FCTRL_UPE;
3621 vmolr |= IXGBE_VMOLR_ROPE;
3625 if (adapter->num_vfs) {
3626 ixgbe_restore_vf_multicasts(adapter);
3627 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(adapter->num_vfs)) &
3628 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3630 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(adapter->num_vfs), vmolr);
3633 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3635 if (netdev->features & NETIF_F_HW_VLAN_RX)
3636 ixgbe_vlan_strip_enable(adapter);
3638 ixgbe_vlan_strip_disable(adapter);
3641 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3644 struct ixgbe_q_vector *q_vector;
3645 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3647 /* legacy and MSI only use one vector */
3648 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3651 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3652 struct napi_struct *napi;
3653 q_vector = adapter->q_vector[q_idx];
3654 napi = &q_vector->napi;
3655 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3656 if (!q_vector->rxr_count || !q_vector->txr_count) {
3657 if (q_vector->txr_count == 1)
3658 napi->poll = &ixgbe_clean_txonly;
3659 else if (q_vector->rxr_count == 1)
3660 napi->poll = &ixgbe_clean_rxonly;
3668 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3671 struct ixgbe_q_vector *q_vector;
3672 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3674 /* legacy and MSI only use one vector */
3675 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3678 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3679 q_vector = adapter->q_vector[q_idx];
3680 napi_disable(&q_vector->napi);
3684 #ifdef CONFIG_IXGBE_DCB
3686 * ixgbe_configure_dcb - Configure DCB hardware
3687 * @adapter: ixgbe adapter struct
3689 * This is called by the driver on open to configure the DCB hardware.
3690 * This is also called by the gennetlink interface when reconfiguring
3693 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3695 struct ixgbe_hw *hw = &adapter->hw;
3696 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3698 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3699 if (hw->mac.type == ixgbe_mac_82598EB)
3700 netif_set_gso_max_size(adapter->netdev, 65536);
3704 if (hw->mac.type == ixgbe_mac_82598EB)
3705 netif_set_gso_max_size(adapter->netdev, 32768);
3708 /* Enable VLAN tag insert/strip */
3709 adapter->netdev->features |= NETIF_F_HW_VLAN_RX;
3711 hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
3713 /* reconfigure the hardware */
3714 if (adapter->dcbx_cap & (DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE)) {
3716 if (adapter->netdev->features & NETIF_F_FCOE_MTU)
3717 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
3719 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3721 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3723 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
3725 struct net_device *dev = adapter->netdev;
3727 if (adapter->ixgbe_ieee_ets)
3728 dev->dcbnl_ops->ieee_setets(dev,
3729 adapter->ixgbe_ieee_ets);
3730 if (adapter->ixgbe_ieee_pfc)
3731 dev->dcbnl_ops->ieee_setpfc(dev,
3732 adapter->ixgbe_ieee_pfc);
3735 /* Enable RSS Hash per TC */
3736 if (hw->mac.type != ixgbe_mac_82598EB) {
3740 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
3742 u8 cnt = adapter->netdev->tc_to_txq[i].count;
3747 reg |= msb << IXGBE_RQTC_SHIFT_TC(i);
3749 IXGBE_WRITE_REG(hw, IXGBE_RQTC, reg);
3754 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3756 struct net_device *netdev = adapter->netdev;
3757 struct ixgbe_hw *hw = &adapter->hw;
3760 #ifdef CONFIG_IXGBE_DCB
3761 ixgbe_configure_dcb(adapter);
3764 ixgbe_set_rx_mode(netdev);
3765 ixgbe_restore_vlan(adapter);
3768 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
3769 ixgbe_configure_fcoe(adapter);
3771 #endif /* IXGBE_FCOE */
3772 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3773 for (i = 0; i < adapter->num_tx_queues; i++)
3774 adapter->tx_ring[i]->atr_sample_rate =
3775 adapter->atr_sample_rate;
3776 ixgbe_init_fdir_signature_82599(hw, adapter->fdir_pballoc);
3777 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3778 ixgbe_init_fdir_perfect_82599(hw, adapter->fdir_pballoc);
3780 ixgbe_configure_virtualization(adapter);
3782 ixgbe_configure_tx(adapter);
3783 ixgbe_configure_rx(adapter);
3786 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3788 switch (hw->phy.type) {
3789 case ixgbe_phy_sfp_avago:
3790 case ixgbe_phy_sfp_ftl:
3791 case ixgbe_phy_sfp_intel:
3792 case ixgbe_phy_sfp_unknown:
3793 case ixgbe_phy_sfp_passive_tyco:
3794 case ixgbe_phy_sfp_passive_unknown:
3795 case ixgbe_phy_sfp_active_unknown:
3796 case ixgbe_phy_sfp_ftl_active:
3804 * ixgbe_sfp_link_config - set up SFP+ link
3805 * @adapter: pointer to private adapter struct
3807 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3810 * We are assuming the worst case scenerio here, and that
3811 * is that an SFP was inserted/removed after the reset
3812 * but before SFP detection was enabled. As such the best
3813 * solution is to just start searching as soon as we start
3815 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3816 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
3818 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
3822 * ixgbe_non_sfp_link_config - set up non-SFP+ link
3823 * @hw: pointer to private hardware struct
3825 * Returns 0 on success, negative on failure
3827 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3830 bool negotiation, link_up = false;
3831 u32 ret = IXGBE_ERR_LINK_SETUP;
3833 if (hw->mac.ops.check_link)
3834 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3839 autoneg = hw->phy.autoneg_advertised;
3840 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
3841 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
3846 if (hw->mac.ops.setup_link)
3847 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
3852 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
3854 struct ixgbe_hw *hw = &adapter->hw;
3857 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3858 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
3860 gpie |= IXGBE_GPIE_EIAME;
3862 * use EIAM to auto-mask when MSI-X interrupt is asserted
3863 * this saves a register write for every interrupt
3865 switch (hw->mac.type) {
3866 case ixgbe_mac_82598EB:
3867 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3869 case ixgbe_mac_82599EB:
3870 case ixgbe_mac_X540:
3872 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
3873 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
3877 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
3878 * specifically only auto mask tx and rx interrupts */
3879 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3882 /* XXX: to interrupt immediately for EICS writes, enable this */
3883 /* gpie |= IXGBE_GPIE_EIMEN; */
3885 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3886 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
3887 gpie |= IXGBE_GPIE_VTMODE_64;
3890 /* Enable fan failure interrupt */
3891 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3892 gpie |= IXGBE_SDP1_GPIEN;
3894 if (hw->mac.type == ixgbe_mac_82599EB) {
3895 gpie |= IXGBE_SDP1_GPIEN;
3896 gpie |= IXGBE_SDP2_GPIEN;
3899 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3902 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
3904 struct ixgbe_hw *hw = &adapter->hw;
3908 ixgbe_get_hw_control(adapter);
3909 ixgbe_setup_gpie(adapter);
3911 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3912 ixgbe_configure_msix(adapter);
3914 ixgbe_configure_msi_and_legacy(adapter);
3916 /* enable the optics for both mult-speed fiber and 82599 SFP+ fiber */
3917 if (hw->mac.ops.enable_tx_laser &&
3918 ((hw->phy.multispeed_fiber) ||
3919 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
3920 (hw->mac.type == ixgbe_mac_82599EB))))
3921 hw->mac.ops.enable_tx_laser(hw);
3923 clear_bit(__IXGBE_DOWN, &adapter->state);
3924 ixgbe_napi_enable_all(adapter);
3926 if (ixgbe_is_sfp(hw)) {
3927 ixgbe_sfp_link_config(adapter);
3929 err = ixgbe_non_sfp_link_config(hw);
3931 e_err(probe, "link_config FAILED %d\n", err);
3934 /* clear any pending interrupts, may auto mask */
3935 IXGBE_READ_REG(hw, IXGBE_EICR);
3936 ixgbe_irq_enable(adapter, true, true);
3939 * If this adapter has a fan, check to see if we had a failure
3940 * before we enabled the interrupt.
3942 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3943 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
3944 if (esdp & IXGBE_ESDP_SDP1)
3945 e_crit(drv, "Fan has stopped, replace the adapter\n");
3948 /* enable transmits */
3949 netif_tx_start_all_queues(adapter->netdev);
3951 /* bring the link up in the watchdog, this could race with our first
3952 * link up interrupt but shouldn't be a problem */
3953 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3954 adapter->link_check_timeout = jiffies;
3955 mod_timer(&adapter->service_timer, jiffies);
3957 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
3958 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
3959 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
3960 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
3965 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
3967 WARN_ON(in_interrupt());
3968 /* put off any impending NetWatchDogTimeout */
3969 adapter->netdev->trans_start = jiffies;
3971 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
3972 usleep_range(1000, 2000);
3973 ixgbe_down(adapter);
3975 * If SR-IOV enabled then wait a bit before bringing the adapter
3976 * back up to give the VFs time to respond to the reset. The
3977 * two second wait is based upon the watchdog timer cycle in
3980 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3983 clear_bit(__IXGBE_RESETTING, &adapter->state);
3986 int ixgbe_up(struct ixgbe_adapter *adapter)
3988 /* hardware has been reset, we need to reload some things */
3989 ixgbe_configure(adapter);
3991 return ixgbe_up_complete(adapter);
3994 void ixgbe_reset(struct ixgbe_adapter *adapter)
3996 struct ixgbe_hw *hw = &adapter->hw;
3999 /* lock SFP init bit to prevent race conditions with the watchdog */
4000 while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
4001 usleep_range(1000, 2000);
4003 /* clear all SFP and link config related flags while holding SFP_INIT */
4004 adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
4005 IXGBE_FLAG2_SFP_NEEDS_RESET);
4006 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
4008 err = hw->mac.ops.init_hw(hw);
4011 case IXGBE_ERR_SFP_NOT_PRESENT:
4012 case IXGBE_ERR_SFP_NOT_SUPPORTED:
4014 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
4015 e_dev_err("master disable timed out\n");
4017 case IXGBE_ERR_EEPROM_VERSION:
4018 /* We are running on a pre-production device, log a warning */
4019 e_dev_warn("This device is a pre-production adapter/LOM. "
4020 "Please be aware there may be issuesassociated with "
4021 "your hardware. If you are experiencing problems "
4022 "please contact your Intel or hardware "
4023 "representative who provided you with this "
4027 e_dev_err("Hardware Error: %d\n", err);
4030 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
4032 /* reprogram the RAR[0] in case user changed it. */
4033 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
4038 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
4039 * @rx_ring: ring to free buffers from
4041 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
4043 struct device *dev = rx_ring->dev;
4047 /* ring already cleared, nothing to do */
4048 if (!rx_ring->rx_buffer_info)
4051 /* Free all the Rx ring sk_buffs */
4052 for (i = 0; i < rx_ring->count; i++) {
4053 struct ixgbe_rx_buffer *rx_buffer_info;
4055 rx_buffer_info = &rx_ring->rx_buffer_info[i];
4056 if (rx_buffer_info->dma) {
4057 dma_unmap_single(rx_ring->dev, rx_buffer_info->dma,
4058 rx_ring->rx_buf_len,
4060 rx_buffer_info->dma = 0;
4062 if (rx_buffer_info->skb) {
4063 struct sk_buff *skb = rx_buffer_info->skb;
4064 rx_buffer_info->skb = NULL;
4066 struct sk_buff *this = skb;
4067 if (IXGBE_RSC_CB(this)->delay_unmap) {
4068 dma_unmap_single(dev,
4069 IXGBE_RSC_CB(this)->dma,
4070 rx_ring->rx_buf_len,
4072 IXGBE_RSC_CB(this)->dma = 0;
4073 IXGBE_RSC_CB(skb)->delay_unmap = false;
4076 dev_kfree_skb(this);
4079 if (!rx_buffer_info->page)
4081 if (rx_buffer_info->page_dma) {
4082 dma_unmap_page(dev, rx_buffer_info->page_dma,
4083 PAGE_SIZE / 2, DMA_FROM_DEVICE);
4084 rx_buffer_info->page_dma = 0;
4086 put_page(rx_buffer_info->page);
4087 rx_buffer_info->page = NULL;
4088 rx_buffer_info->page_offset = 0;
4091 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4092 memset(rx_ring->rx_buffer_info, 0, size);
4094 /* Zero out the descriptor ring */
4095 memset(rx_ring->desc, 0, rx_ring->size);
4097 rx_ring->next_to_clean = 0;
4098 rx_ring->next_to_use = 0;
4102 * ixgbe_clean_tx_ring - Free Tx Buffers
4103 * @tx_ring: ring to be cleaned
4105 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
4107 struct ixgbe_tx_buffer *tx_buffer_info;
4111 /* ring already cleared, nothing to do */
4112 if (!tx_ring->tx_buffer_info)
4115 /* Free all the Tx ring sk_buffs */
4116 for (i = 0; i < tx_ring->count; i++) {
4117 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4118 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
4121 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4122 memset(tx_ring->tx_buffer_info, 0, size);
4124 /* Zero out the descriptor ring */
4125 memset(tx_ring->desc, 0, tx_ring->size);
4127 tx_ring->next_to_use = 0;
4128 tx_ring->next_to_clean = 0;
4132 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
4133 * @adapter: board private structure
4135 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
4139 for (i = 0; i < adapter->num_rx_queues; i++)
4140 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
4144 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
4145 * @adapter: board private structure
4147 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
4151 for (i = 0; i < adapter->num_tx_queues; i++)
4152 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
4155 void ixgbe_down(struct ixgbe_adapter *adapter)
4157 struct net_device *netdev = adapter->netdev;
4158 struct ixgbe_hw *hw = &adapter->hw;
4162 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4164 /* signal that we are down to the interrupt handler */
4165 set_bit(__IXGBE_DOWN, &adapter->state);
4167 /* disable receive for all VFs and wait one second */
4168 if (adapter->num_vfs) {
4169 /* ping all the active vfs to let them know we are going down */
4170 ixgbe_ping_all_vfs(adapter);
4172 /* Disable all VFTE/VFRE TX/RX */
4173 ixgbe_disable_tx_rx(adapter);
4175 /* Mark all the VFs as inactive */
4176 for (i = 0 ; i < adapter->num_vfs; i++)
4177 adapter->vfinfo[i].clear_to_send = 0;
4180 /* disable receives */
4181 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4182 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
4184 /* disable all enabled rx queues */
4185 for (i = 0; i < adapter->num_rx_queues; i++)
4186 /* this call also flushes the previous write */
4187 ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]);
4189 usleep_range(10000, 20000);
4191 netif_tx_stop_all_queues(netdev);
4193 /* call carrier off first to avoid false dev_watchdog timeouts */
4194 netif_carrier_off(netdev);
4195 netif_tx_disable(netdev);
4197 ixgbe_irq_disable(adapter);
4199 ixgbe_napi_disable_all(adapter);
4201 adapter->flags2 &= ~IXGBE_FLAG2_RESET_REQUESTED;
4202 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
4204 del_timer_sync(&adapter->service_timer);
4206 /* Cleanup the affinity_hint CPU mask memory and callback */
4207 for (i = 0; i < num_q_vectors; i++) {
4208 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
4209 /* clear the affinity_mask in the IRQ descriptor */
4210 irq_set_affinity_hint(adapter->msix_entries[i]. vector, NULL);
4211 /* release the CPU mask memory */
4212 free_cpumask_var(q_vector->affinity_mask);
4215 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
4216 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
4217 cancel_work_sync(&adapter->fdir_reinit_task);
4219 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
4220 cancel_work_sync(&adapter->check_overtemp_task);
4222 /* disable transmits in the hardware now that interrupts are off */
4223 for (i = 0; i < adapter->num_tx_queues; i++) {
4224 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
4225 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
4226 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx),
4227 (txdctl & ~IXGBE_TXDCTL_ENABLE));
4229 /* Disable the Tx DMA engine on 82599 */
4230 switch (hw->mac.type) {
4231 case ixgbe_mac_82599EB:
4232 case ixgbe_mac_X540:
4233 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
4234 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
4235 ~IXGBE_DMATXCTL_TE));
4241 if (!pci_channel_offline(adapter->pdev))
4242 ixgbe_reset(adapter);
4244 /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
4245 if (hw->mac.ops.disable_tx_laser &&
4246 ((hw->phy.multispeed_fiber) ||
4247 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
4248 (hw->mac.type == ixgbe_mac_82599EB))))
4249 hw->mac.ops.disable_tx_laser(hw);
4251 ixgbe_clean_all_tx_rings(adapter);
4252 ixgbe_clean_all_rx_rings(adapter);
4254 #ifdef CONFIG_IXGBE_DCA
4255 /* since we reset the hardware DCA settings were cleared */
4256 ixgbe_setup_dca(adapter);
4261 * ixgbe_poll - NAPI Rx polling callback
4262 * @napi: structure for representing this polling device
4263 * @budget: how many packets driver is allowed to clean
4265 * This function is used for legacy and MSI, NAPI mode
4267 static int ixgbe_poll(struct napi_struct *napi, int budget)
4269 struct ixgbe_q_vector *q_vector =
4270 container_of(napi, struct ixgbe_q_vector, napi);
4271 struct ixgbe_adapter *adapter = q_vector->adapter;
4272 int tx_clean_complete, work_done = 0;
4274 #ifdef CONFIG_IXGBE_DCA
4275 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
4276 ixgbe_update_dca(q_vector);
4279 tx_clean_complete = ixgbe_clean_tx_irq(q_vector, adapter->tx_ring[0]);
4280 ixgbe_clean_rx_irq(q_vector, adapter->rx_ring[0], &work_done, budget);
4282 if (!tx_clean_complete)
4285 /* If budget not fully consumed, exit the polling mode */
4286 if (work_done < budget) {
4287 napi_complete(napi);
4288 if (adapter->rx_itr_setting & 1)
4289 ixgbe_set_itr(adapter);
4290 if (!test_bit(__IXGBE_DOWN, &adapter->state))
4291 ixgbe_irq_enable_queues(adapter, IXGBE_EIMS_RTX_QUEUE);
4297 * ixgbe_tx_timeout - Respond to a Tx Hang
4298 * @netdev: network interface device structure
4300 static void ixgbe_tx_timeout(struct net_device *netdev)
4302 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4304 /* Do the reset outside of interrupt context */
4305 ixgbe_tx_timeout_reset(adapter);
4309 * ixgbe_set_rss_queues: Allocate queues for RSS
4310 * @adapter: board private structure to initialize
4312 * This is our "base" multiqueue mode. RSS (Receive Side Scaling) will try
4313 * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
4316 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
4319 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
4321 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4323 adapter->num_rx_queues = f->indices;
4324 adapter->num_tx_queues = f->indices;
4334 * ixgbe_set_fdir_queues: Allocate queues for Flow Director
4335 * @adapter: board private structure to initialize
4337 * Flow Director is an advanced Rx filter, attempting to get Rx flows back
4338 * to the original CPU that initiated the Tx session. This runs in addition
4339 * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
4340 * Rx load across CPUs using RSS.
4343 static inline bool ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
4346 struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
4348 f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
4351 /* Flow Director must have RSS enabled */
4352 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
4353 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
4354 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)))) {
4355 adapter->num_tx_queues = f_fdir->indices;
4356 adapter->num_rx_queues = f_fdir->indices;
4359 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4360 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4367 * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
4368 * @adapter: board private structure to initialize
4370 * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
4371 * The ring feature mask is not used as a mask for FCoE, as it can take any 8
4372 * rx queues out of the max number of rx queues, instead, it is used as the
4373 * index of the first rx queue used by FCoE.
4376 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
4378 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4380 if (!(adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
4383 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4384 #ifdef CONFIG_IXGBE_DCB
4386 struct net_device *dev = adapter->netdev;
4388 tc = netdev_get_prio_tc_map(dev, adapter->fcoe.up);
4389 f->indices = dev->tc_to_txq[tc].count;
4390 f->mask = dev->tc_to_txq[tc].offset;
4393 f->indices = min((int)num_online_cpus(), f->indices);
4395 adapter->num_rx_queues = 1;
4396 adapter->num_tx_queues = 1;
4398 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4399 e_info(probe, "FCoE enabled with RSS\n");
4400 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4401 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
4402 ixgbe_set_fdir_queues(adapter);
4404 ixgbe_set_rss_queues(adapter);
4406 /* adding FCoE rx rings to the end */
4407 f->mask = adapter->num_rx_queues;
4408 adapter->num_rx_queues += f->indices;
4409 adapter->num_tx_queues += f->indices;
4414 #endif /* IXGBE_FCOE */
4416 #ifdef CONFIG_IXGBE_DCB
4417 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
4420 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_DCB];
4423 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
4427 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
4428 q = min((int)num_online_cpus(), MAX_TRAFFIC_CLASS);
4433 adapter->num_rx_queues = f->indices;
4434 adapter->num_tx_queues = f->indices;
4438 /* FCoE enabled queues require special configuration done through
4439 * configure_fcoe() and others. Here we map FCoE indices onto the
4440 * DCB queue pairs allowing FCoE to own configuration later.
4442 ixgbe_set_fcoe_queues(adapter);
4450 * ixgbe_set_sriov_queues: Allocate queues for IOV use
4451 * @adapter: board private structure to initialize
4453 * IOV doesn't actually use anything, so just NAK the
4454 * request for now and let the other queue routines
4455 * figure out what to do.
4457 static inline bool ixgbe_set_sriov_queues(struct ixgbe_adapter *adapter)
4463 * ixgbe_set_num_queues: Allocate queues for device, feature dependent
4464 * @adapter: board private structure to initialize
4466 * This is the top level queue allocation routine. The order here is very
4467 * important, starting with the "most" number of features turned on at once,
4468 * and ending with the smallest set of features. This way large combinations
4469 * can be allocated if they're turned on, and smaller combinations are the
4470 * fallthrough conditions.
4473 static int ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
4475 /* Start with base case */
4476 adapter->num_rx_queues = 1;
4477 adapter->num_tx_queues = 1;
4478 adapter->num_rx_pools = adapter->num_rx_queues;
4479 adapter->num_rx_queues_per_pool = 1;
4481 if (ixgbe_set_sriov_queues(adapter))
4484 #ifdef CONFIG_IXGBE_DCB
4485 if (ixgbe_set_dcb_queues(adapter))
4490 if (ixgbe_set_fcoe_queues(adapter))
4493 #endif /* IXGBE_FCOE */
4494 if (ixgbe_set_fdir_queues(adapter))
4497 if (ixgbe_set_rss_queues(adapter))
4500 /* fallback to base case */
4501 adapter->num_rx_queues = 1;
4502 adapter->num_tx_queues = 1;
4505 /* Notify the stack of the (possibly) reduced queue counts. */
4506 netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
4507 return netif_set_real_num_rx_queues(adapter->netdev,
4508 adapter->num_rx_queues);
4511 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
4514 int err, vector_threshold;
4516 /* We'll want at least 3 (vector_threshold):
4519 * 3) Other (Link Status Change, etc.)
4520 * 4) TCP Timer (optional)
4522 vector_threshold = MIN_MSIX_COUNT;
4524 /* The more we get, the more we will assign to Tx/Rx Cleanup
4525 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
4526 * Right now, we simply care about how many we'll get; we'll
4527 * set them up later while requesting irq's.
4529 while (vectors >= vector_threshold) {
4530 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
4532 if (!err) /* Success in acquiring all requested vectors. */
4535 vectors = 0; /* Nasty failure, quit now */
4536 else /* err == number of vectors we should try again with */
4540 if (vectors < vector_threshold) {
4541 /* Can't allocate enough MSI-X interrupts? Oh well.
4542 * This just means we'll go with either a single MSI
4543 * vector or fall back to legacy interrupts.
4545 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4546 "Unable to allocate MSI-X interrupts\n");
4547 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4548 kfree(adapter->msix_entries);
4549 adapter->msix_entries = NULL;
4551 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
4553 * Adjust for only the vectors we'll use, which is minimum
4554 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
4555 * vectors we were allocated.
4557 adapter->num_msix_vectors = min(vectors,
4558 adapter->max_msix_q_vectors + NON_Q_VECTORS);
4563 * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
4564 * @adapter: board private structure to initialize
4566 * Cache the descriptor ring offsets for RSS to the assigned rings.
4569 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
4573 if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED))
4576 for (i = 0; i < adapter->num_rx_queues; i++)
4577 adapter->rx_ring[i]->reg_idx = i;
4578 for (i = 0; i < adapter->num_tx_queues; i++)
4579 adapter->tx_ring[i]->reg_idx = i;
4584 #ifdef CONFIG_IXGBE_DCB
4586 /* ixgbe_get_first_reg_idx - Return first register index associated with ring */
4587 static void ixgbe_get_first_reg_idx(struct ixgbe_adapter *adapter, u8 tc,
4588 unsigned int *tx, unsigned int *rx)
4590 struct net_device *dev = adapter->netdev;
4591 struct ixgbe_hw *hw = &adapter->hw;
4592 u8 num_tcs = netdev_get_num_tc(dev);
4597 switch (hw->mac.type) {
4598 case ixgbe_mac_82598EB:
4602 case ixgbe_mac_82599EB:
4603 case ixgbe_mac_X540:
4608 } else if (tc < 5) {
4609 *tx = ((tc + 2) << 4);
4611 } else if (tc < num_tcs) {
4612 *tx = ((tc + 8) << 3);
4615 } else if (num_tcs == 4) {
4640 #define IXGBE_MAX_Q_PER_TC (IXGBE_MAX_DCB_INDICES / MAX_TRAFFIC_CLASS)
4642 /* ixgbe_setup_tc - routine to configure net_device for multiple traffic
4645 * @netdev: net device to configure
4646 * @tc: number of traffic classes to enable
4648 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
4651 unsigned int q, offset = 0;
4654 netdev_reset_tc(dev);
4656 struct ixgbe_adapter *adapter = netdev_priv(dev);
4658 /* Hardware supports up to 8 traffic classes */
4659 if (tc > MAX_TRAFFIC_CLASS || netdev_set_num_tc(dev, tc))
4662 /* Partition Tx queues evenly amongst traffic classes */
4663 for (i = 0; i < tc; i++) {
4664 q = min((int)num_online_cpus(), IXGBE_MAX_Q_PER_TC);
4665 netdev_set_prio_tc_map(dev, i, i);
4666 netdev_set_tc_queue(dev, i, q, offset);
4670 /* This enables multiple traffic class support in the hardware
4671 * which defaults to strict priority transmission by default.
4672 * If traffic classes are already enabled perhaps through DCB
4673 * code path then existing configuration will be used.
4675 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
4676 dev->dcbnl_ops && dev->dcbnl_ops->setdcbx) {
4677 struct ieee_ets ets = {
4678 .prio_tc = {0, 1, 2, 3, 4, 5, 6, 7},
4680 u8 mode = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_IEEE;
4682 dev->dcbnl_ops->setdcbx(dev, mode);
4683 dev->dcbnl_ops->ieee_setets(dev, &ets);
4690 * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
4691 * @adapter: board private structure to initialize
4693 * Cache the descriptor ring offsets for DCB to the assigned rings.
4696 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
4698 struct net_device *dev = adapter->netdev;
4700 u8 num_tcs = netdev_get_num_tc(dev);
4702 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
4705 for (i = 0, k = 0; i < num_tcs; i++) {
4706 unsigned int tx_s, rx_s;
4707 u16 count = dev->tc_to_txq[i].count;
4709 ixgbe_get_first_reg_idx(adapter, i, &tx_s, &rx_s);
4710 for (j = 0; j < count; j++, k++) {
4711 adapter->tx_ring[k]->reg_idx = tx_s + j;
4712 adapter->rx_ring[k]->reg_idx = rx_s + j;
4713 adapter->tx_ring[k]->dcb_tc = i;
4714 adapter->rx_ring[k]->dcb_tc = i;
4723 * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
4724 * @adapter: board private structure to initialize
4726 * Cache the descriptor ring offsets for Flow Director to the assigned rings.
4729 static inline bool ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
4734 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
4735 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4736 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))) {
4737 for (i = 0; i < adapter->num_rx_queues; i++)
4738 adapter->rx_ring[i]->reg_idx = i;
4739 for (i = 0; i < adapter->num_tx_queues; i++)
4740 adapter->tx_ring[i]->reg_idx = i;
4749 * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
4750 * @adapter: board private structure to initialize
4752 * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
4755 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
4757 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4759 u8 fcoe_rx_i = 0, fcoe_tx_i = 0;
4761 if (!(adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
4764 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4765 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4766 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
4767 ixgbe_cache_ring_fdir(adapter);
4769 ixgbe_cache_ring_rss(adapter);
4771 fcoe_rx_i = f->mask;
4772 fcoe_tx_i = f->mask;
4774 for (i = 0; i < f->indices; i++, fcoe_rx_i++, fcoe_tx_i++) {
4775 adapter->rx_ring[f->mask + i]->reg_idx = fcoe_rx_i;
4776 adapter->tx_ring[f->mask + i]->reg_idx = fcoe_tx_i;
4781 #endif /* IXGBE_FCOE */
4783 * ixgbe_cache_ring_sriov - Descriptor ring to register mapping for sriov
4784 * @adapter: board private structure to initialize
4786 * SR-IOV doesn't use any descriptor rings but changes the default if
4787 * no other mapping is used.
4790 static inline bool ixgbe_cache_ring_sriov(struct ixgbe_adapter *adapter)
4792 adapter->rx_ring[0]->reg_idx = adapter->num_vfs * 2;
4793 adapter->tx_ring[0]->reg_idx = adapter->num_vfs * 2;
4794 if (adapter->num_vfs)
4801 * ixgbe_cache_ring_register - Descriptor ring to register mapping
4802 * @adapter: board private structure to initialize
4804 * Once we know the feature-set enabled for the device, we'll cache
4805 * the register offset the descriptor ring is assigned to.
4807 * Note, the order the various feature calls is important. It must start with
4808 * the "most" features enabled at the same time, then trickle down to the
4809 * least amount of features turned on at once.
4811 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
4813 /* start with default case */
4814 adapter->rx_ring[0]->reg_idx = 0;
4815 adapter->tx_ring[0]->reg_idx = 0;
4817 if (ixgbe_cache_ring_sriov(adapter))
4820 #ifdef CONFIG_IXGBE_DCB
4821 if (ixgbe_cache_ring_dcb(adapter))
4826 if (ixgbe_cache_ring_fcoe(adapter))
4828 #endif /* IXGBE_FCOE */
4830 if (ixgbe_cache_ring_fdir(adapter))
4833 if (ixgbe_cache_ring_rss(adapter))
4838 * ixgbe_alloc_queues - Allocate memory for all rings
4839 * @adapter: board private structure to initialize
4841 * We allocate one ring per queue at run-time since we don't know the
4842 * number of queues at compile-time. The polling_netdev array is
4843 * intended for Multiqueue, but should work fine with a single queue.
4845 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
4847 int rx = 0, tx = 0, nid = adapter->node;
4849 if (nid < 0 || !node_online(nid))
4850 nid = first_online_node;
4852 for (; tx < adapter->num_tx_queues; tx++) {
4853 struct ixgbe_ring *ring;
4855 ring = kzalloc_node(sizeof(*ring), GFP_KERNEL, nid);
4857 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
4859 goto err_allocation;
4860 ring->count = adapter->tx_ring_count;
4861 ring->queue_index = tx;
4862 ring->numa_node = nid;
4863 ring->dev = &adapter->pdev->dev;
4864 ring->netdev = adapter->netdev;
4866 adapter->tx_ring[tx] = ring;
4869 for (; rx < adapter->num_rx_queues; rx++) {
4870 struct ixgbe_ring *ring;
4872 ring = kzalloc_node(sizeof(*ring), GFP_KERNEL, nid);
4874 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
4876 goto err_allocation;
4877 ring->count = adapter->rx_ring_count;
4878 ring->queue_index = rx;
4879 ring->numa_node = nid;
4880 ring->dev = &adapter->pdev->dev;
4881 ring->netdev = adapter->netdev;
4883 adapter->rx_ring[rx] = ring;
4886 ixgbe_cache_ring_register(adapter);
4892 kfree(adapter->tx_ring[--tx]);
4895 kfree(adapter->rx_ring[--rx]);
4900 * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
4901 * @adapter: board private structure to initialize
4903 * Attempt to configure the interrupts using the best available
4904 * capabilities of the hardware and the kernel.
4906 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
4908 struct ixgbe_hw *hw = &adapter->hw;
4910 int vector, v_budget;
4913 * It's easy to be greedy for MSI-X vectors, but it really
4914 * doesn't do us much good if we have a lot more vectors
4915 * than CPU's. So let's be conservative and only ask for
4916 * (roughly) the same number of vectors as there are CPU's.
4918 v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
4919 (int)num_online_cpus()) + NON_Q_VECTORS;
4922 * At the same time, hardware can only support a maximum of
4923 * hw.mac->max_msix_vectors vectors. With features
4924 * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
4925 * descriptor queues supported by our device. Thus, we cap it off in
4926 * those rare cases where the cpu count also exceeds our vector limit.
4928 v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
4930 /* A failure in MSI-X entry allocation isn't fatal, but it does
4931 * mean we disable MSI-X capabilities of the adapter. */
4932 adapter->msix_entries = kcalloc(v_budget,
4933 sizeof(struct msix_entry), GFP_KERNEL);
4934 if (adapter->msix_entries) {
4935 for (vector = 0; vector < v_budget; vector++)
4936 adapter->msix_entries[vector].entry = vector;
4938 ixgbe_acquire_msix_vectors(adapter, v_budget);
4940 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4944 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
4945 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
4946 if (adapter->flags & (IXGBE_FLAG_FDIR_HASH_CAPABLE |
4947 IXGBE_FLAG_FDIR_PERFECT_CAPABLE)) {
4949 "Flow Director is not supported while multiple "
4950 "queues are disabled. Disabling Flow Director\n");
4952 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4953 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4954 adapter->atr_sample_rate = 0;
4955 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4956 ixgbe_disable_sriov(adapter);
4958 err = ixgbe_set_num_queues(adapter);
4962 err = pci_enable_msi(adapter->pdev);
4964 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
4966 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4967 "Unable to allocate MSI interrupt, "
4968 "falling back to legacy. Error: %d\n", err);
4978 * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
4979 * @adapter: board private structure to initialize
4981 * We allocate one q_vector per queue interrupt. If allocation fails we
4984 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
4986 int q_idx, num_q_vectors;
4987 struct ixgbe_q_vector *q_vector;
4988 int (*poll)(struct napi_struct *, int);
4990 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4991 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4992 poll = &ixgbe_clean_rxtx_many;
4998 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4999 q_vector = kzalloc_node(sizeof(struct ixgbe_q_vector),
5000 GFP_KERNEL, adapter->node);
5002 q_vector = kzalloc(sizeof(struct ixgbe_q_vector),
5006 q_vector->adapter = adapter;
5007 if (q_vector->txr_count && !q_vector->rxr_count)
5008 q_vector->eitr = adapter->tx_eitr_param;
5010 q_vector->eitr = adapter->rx_eitr_param;
5011 q_vector->v_idx = q_idx;
5012 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
5013 adapter->q_vector[q_idx] = q_vector;
5021 q_vector = adapter->q_vector[q_idx];
5022 netif_napi_del(&q_vector->napi);
5024 adapter->q_vector[q_idx] = NULL;
5030 * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
5031 * @adapter: board private structure to initialize
5033 * This function frees the memory allocated to the q_vectors. In addition if
5034 * NAPI is enabled it will delete any references to the NAPI struct prior
5035 * to freeing the q_vector.
5037 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
5039 int q_idx, num_q_vectors;
5041 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
5042 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
5046 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
5047 struct ixgbe_q_vector *q_vector = adapter->q_vector[q_idx];
5048 adapter->q_vector[q_idx] = NULL;
5049 netif_napi_del(&q_vector->napi);
5054 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
5056 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
5057 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
5058 pci_disable_msix(adapter->pdev);
5059 kfree(adapter->msix_entries);
5060 adapter->msix_entries = NULL;
5061 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
5062 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
5063 pci_disable_msi(adapter->pdev);
5068 * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
5069 * @adapter: board private structure to initialize
5071 * We determine which interrupt scheme to use based on...
5072 * - Kernel support (MSI, MSI-X)
5073 * - which can be user-defined (via MODULE_PARAM)
5074 * - Hardware queue count (num_*_queues)
5075 * - defined by miscellaneous hardware support/features (RSS, etc.)
5077 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
5081 /* Number of supported queues */
5082 err = ixgbe_set_num_queues(adapter);
5086 err = ixgbe_set_interrupt_capability(adapter);
5088 e_dev_err("Unable to setup interrupt capabilities\n");
5089 goto err_set_interrupt;
5092 err = ixgbe_alloc_q_vectors(adapter);
5094 e_dev_err("Unable to allocate memory for queue vectors\n");
5095 goto err_alloc_q_vectors;
5098 err = ixgbe_alloc_queues(adapter);
5100 e_dev_err("Unable to allocate memory for queues\n");
5101 goto err_alloc_queues;
5104 e_dev_info("Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
5105 (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
5106 adapter->num_rx_queues, adapter->num_tx_queues);
5108 set_bit(__IXGBE_DOWN, &adapter->state);
5113 ixgbe_free_q_vectors(adapter);
5114 err_alloc_q_vectors:
5115 ixgbe_reset_interrupt_capability(adapter);
5120 static void ring_free_rcu(struct rcu_head *head)
5122 kfree(container_of(head, struct ixgbe_ring, rcu));
5126 * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
5127 * @adapter: board private structure to clear interrupt scheme on
5129 * We go through and clear interrupt specific resources and reset the structure
5130 * to pre-load conditions
5132 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
5136 for (i = 0; i < adapter->num_tx_queues; i++) {
5137 kfree(adapter->tx_ring[i]);
5138 adapter->tx_ring[i] = NULL;
5140 for (i = 0; i < adapter->num_rx_queues; i++) {
5141 struct ixgbe_ring *ring = adapter->rx_ring[i];
5143 /* ixgbe_get_stats64() might access this ring, we must wait
5144 * a grace period before freeing it.
5146 call_rcu(&ring->rcu, ring_free_rcu);
5147 adapter->rx_ring[i] = NULL;
5150 adapter->num_tx_queues = 0;
5151 adapter->num_rx_queues = 0;
5153 ixgbe_free_q_vectors(adapter);
5154 ixgbe_reset_interrupt_capability(adapter);
5158 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
5159 * @adapter: board private structure to initialize
5161 * ixgbe_sw_init initializes the Adapter private data structure.
5162 * Fields are initialized based on PCI device information and
5163 * OS network device settings (MTU size).
5165 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
5167 struct ixgbe_hw *hw = &adapter->hw;
5168 struct pci_dev *pdev = adapter->pdev;
5169 struct net_device *dev = adapter->netdev;
5171 #ifdef CONFIG_IXGBE_DCB
5173 struct tc_configuration *tc;
5175 int max_frame = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
5177 /* PCI config space info */
5179 hw->vendor_id = pdev->vendor;
5180 hw->device_id = pdev->device;
5181 hw->revision_id = pdev->revision;
5182 hw->subsystem_vendor_id = pdev->subsystem_vendor;
5183 hw->subsystem_device_id = pdev->subsystem_device;
5185 /* Set capability flags */
5186 rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
5187 adapter->ring_feature[RING_F_RSS].indices = rss;
5188 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
5189 adapter->ring_feature[RING_F_DCB].indices = IXGBE_MAX_DCB_INDICES;
5190 switch (hw->mac.type) {
5191 case ixgbe_mac_82598EB:
5192 if (hw->device_id == IXGBE_DEV_ID_82598AT)
5193 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
5194 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
5196 case ixgbe_mac_82599EB:
5197 case ixgbe_mac_X540:
5198 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
5199 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
5200 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
5201 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
5202 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
5203 /* n-tuple support exists, always init our spinlock */
5204 spin_lock_init(&adapter->fdir_perfect_lock);
5205 /* Flow Director hash filters enabled */
5206 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
5207 adapter->atr_sample_rate = 20;
5208 adapter->ring_feature[RING_F_FDIR].indices =
5209 IXGBE_MAX_FDIR_INDICES;
5210 adapter->fdir_pballoc = 0;
5212 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
5213 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
5214 adapter->ring_feature[RING_F_FCOE].indices = 0;
5215 #ifdef CONFIG_IXGBE_DCB
5216 /* Default traffic class to use for FCoE */
5217 adapter->fcoe.tc = IXGBE_FCOE_DEFTC;
5218 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
5220 #endif /* IXGBE_FCOE */
5226 #ifdef CONFIG_IXGBE_DCB
5227 /* Configure DCB traffic classes */
5228 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
5229 tc = &adapter->dcb_cfg.tc_config[j];
5230 tc->path[DCB_TX_CONFIG].bwg_id = 0;
5231 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
5232 tc->path[DCB_RX_CONFIG].bwg_id = 0;
5233 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
5234 tc->dcb_pfc = pfc_disabled;
5236 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
5237 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
5238 adapter->dcb_cfg.rx_pba_cfg = pba_equal;
5239 adapter->dcb_cfg.pfc_mode_enable = false;
5240 adapter->dcb_set_bitmap = 0x00;
5241 adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
5242 ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
5247 /* default flow control settings */
5248 hw->fc.requested_mode = ixgbe_fc_full;
5249 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
5251 adapter->last_lfc_mode = hw->fc.current_mode;
5253 hw->fc.high_water = FC_HIGH_WATER(max_frame);
5254 hw->fc.low_water = FC_LOW_WATER(max_frame);
5255 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
5256 hw->fc.send_xon = true;
5257 hw->fc.disable_fc_autoneg = false;
5259 /* enable itr by default in dynamic mode */
5260 adapter->rx_itr_setting = 1;
5261 adapter->rx_eitr_param = 20000;
5262 adapter->tx_itr_setting = 1;
5263 adapter->tx_eitr_param = 10000;
5265 /* set defaults for eitr in MegaBytes */
5266 adapter->eitr_low = 10;
5267 adapter->eitr_high = 20;
5269 /* set default ring sizes */
5270 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
5271 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
5273 /* initialize eeprom parameters */
5274 if (ixgbe_init_eeprom_params_generic(hw)) {
5275 e_dev_err("EEPROM initialization failed\n");
5279 /* enable rx csum by default */
5280 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
5282 /* get assigned NUMA node */
5283 adapter->node = dev_to_node(&pdev->dev);
5285 set_bit(__IXGBE_DOWN, &adapter->state);
5291 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
5292 * @tx_ring: tx descriptor ring (for a specific queue) to setup
5294 * Return 0 on success, negative on failure
5296 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
5298 struct device *dev = tx_ring->dev;
5301 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
5302 tx_ring->tx_buffer_info = vzalloc_node(size, tx_ring->numa_node);
5303 if (!tx_ring->tx_buffer_info)
5304 tx_ring->tx_buffer_info = vzalloc(size);
5305 if (!tx_ring->tx_buffer_info)
5308 /* round up to nearest 4K */
5309 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
5310 tx_ring->size = ALIGN(tx_ring->size, 4096);
5312 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
5313 &tx_ring->dma, GFP_KERNEL);
5317 tx_ring->next_to_use = 0;
5318 tx_ring->next_to_clean = 0;
5319 tx_ring->work_limit = tx_ring->count;
5323 vfree(tx_ring->tx_buffer_info);
5324 tx_ring->tx_buffer_info = NULL;
5325 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
5330 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
5331 * @adapter: board private structure
5333 * If this function returns with an error, then it's possible one or
5334 * more of the rings is populated (while the rest are not). It is the
5335 * callers duty to clean those orphaned rings.
5337 * Return 0 on success, negative on failure
5339 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
5343 for (i = 0; i < adapter->num_tx_queues; i++) {
5344 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
5347 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
5355 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
5356 * @rx_ring: rx descriptor ring (for a specific queue) to setup
5358 * Returns 0 on success, negative on failure
5360 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
5362 struct device *dev = rx_ring->dev;
5365 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
5366 rx_ring->rx_buffer_info = vzalloc_node(size, rx_ring->numa_node);
5367 if (!rx_ring->rx_buffer_info)
5368 rx_ring->rx_buffer_info = vzalloc(size);
5369 if (!rx_ring->rx_buffer_info)
5372 /* Round up to nearest 4K */
5373 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
5374 rx_ring->size = ALIGN(rx_ring->size, 4096);
5376 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
5377 &rx_ring->dma, GFP_KERNEL);
5382 rx_ring->next_to_clean = 0;
5383 rx_ring->next_to_use = 0;
5387 vfree(rx_ring->rx_buffer_info);
5388 rx_ring->rx_buffer_info = NULL;
5389 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
5394 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
5395 * @adapter: board private structure
5397 * If this function returns with an error, then it's possible one or
5398 * more of the rings is populated (while the rest are not). It is the
5399 * callers duty to clean those orphaned rings.
5401 * Return 0 on success, negative on failure
5403 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
5407 for (i = 0; i < adapter->num_rx_queues; i++) {
5408 err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
5411 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
5419 * ixgbe_free_tx_resources - Free Tx Resources per Queue
5420 * @tx_ring: Tx descriptor ring for a specific queue
5422 * Free all transmit software resources
5424 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
5426 ixgbe_clean_tx_ring(tx_ring);
5428 vfree(tx_ring->tx_buffer_info);
5429 tx_ring->tx_buffer_info = NULL;
5431 /* if not set, then don't free */
5435 dma_free_coherent(tx_ring->dev, tx_ring->size,
5436 tx_ring->desc, tx_ring->dma);
5438 tx_ring->desc = NULL;
5442 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5443 * @adapter: board private structure
5445 * Free all transmit software resources
5447 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
5451 for (i = 0; i < adapter->num_tx_queues; i++)
5452 if (adapter->tx_ring[i]->desc)
5453 ixgbe_free_tx_resources(adapter->tx_ring[i]);
5457 * ixgbe_free_rx_resources - Free Rx Resources
5458 * @rx_ring: ring to clean the resources from
5460 * Free all receive software resources
5462 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
5464 ixgbe_clean_rx_ring(rx_ring);
5466 vfree(rx_ring->rx_buffer_info);
5467 rx_ring->rx_buffer_info = NULL;
5469 /* if not set, then don't free */
5473 dma_free_coherent(rx_ring->dev, rx_ring->size,
5474 rx_ring->desc, rx_ring->dma);
5476 rx_ring->desc = NULL;
5480 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5481 * @adapter: board private structure
5483 * Free all receive software resources
5485 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5489 for (i = 0; i < adapter->num_rx_queues; i++)
5490 if (adapter->rx_ring[i]->desc)
5491 ixgbe_free_rx_resources(adapter->rx_ring[i]);
5495 * ixgbe_change_mtu - Change the Maximum Transfer Unit
5496 * @netdev: network interface device structure
5497 * @new_mtu: new value for maximum frame size
5499 * Returns 0 on success, negative on failure
5501 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5503 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5504 struct ixgbe_hw *hw = &adapter->hw;
5505 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5507 /* MTU < 68 is an error and causes problems on some kernels */
5508 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED &&
5509 hw->mac.type != ixgbe_mac_X540) {
5510 if ((new_mtu < 68) || (max_frame > MAXIMUM_ETHERNET_VLAN_SIZE))
5513 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5517 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5518 /* must set new MTU before calling down or up */
5519 netdev->mtu = new_mtu;
5521 hw->fc.high_water = FC_HIGH_WATER(max_frame);
5522 hw->fc.low_water = FC_LOW_WATER(max_frame);
5524 if (netif_running(netdev))
5525 ixgbe_reinit_locked(adapter);
5531 * ixgbe_open - Called when a network interface is made active
5532 * @netdev: network interface device structure
5534 * Returns 0 on success, negative value on failure
5536 * The open entry point is called when a network interface is made
5537 * active by the system (IFF_UP). At this point all resources needed
5538 * for transmit and receive operations are allocated, the interrupt
5539 * handler is registered with the OS, the watchdog timer is started,
5540 * and the stack is notified that the interface is ready.
5542 static int ixgbe_open(struct net_device *netdev)
5544 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5547 /* disallow open during test */
5548 if (test_bit(__IXGBE_TESTING, &adapter->state))
5551 netif_carrier_off(netdev);
5553 /* allocate transmit descriptors */
5554 err = ixgbe_setup_all_tx_resources(adapter);
5558 /* allocate receive descriptors */
5559 err = ixgbe_setup_all_rx_resources(adapter);
5563 ixgbe_configure(adapter);
5565 err = ixgbe_request_irq(adapter);
5569 err = ixgbe_up_complete(adapter);
5573 netif_tx_start_all_queues(netdev);
5578 ixgbe_release_hw_control(adapter);
5579 ixgbe_free_irq(adapter);
5582 ixgbe_free_all_rx_resources(adapter);
5584 ixgbe_free_all_tx_resources(adapter);
5585 ixgbe_reset(adapter);
5591 * ixgbe_close - Disables a network interface
5592 * @netdev: network interface device structure
5594 * Returns 0, this is not allowed to fail
5596 * The close entry point is called when an interface is de-activated
5597 * by the OS. The hardware is still under the drivers control, but
5598 * needs to be disabled. A global MAC reset is issued to stop the
5599 * hardware, and all transmit and receive resources are freed.
5601 static int ixgbe_close(struct net_device *netdev)
5603 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5605 ixgbe_down(adapter);
5606 ixgbe_free_irq(adapter);
5608 ixgbe_free_all_tx_resources(adapter);
5609 ixgbe_free_all_rx_resources(adapter);
5611 ixgbe_release_hw_control(adapter);
5617 static int ixgbe_resume(struct pci_dev *pdev)
5619 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5620 struct net_device *netdev = adapter->netdev;
5623 pci_set_power_state(pdev, PCI_D0);
5624 pci_restore_state(pdev);
5626 * pci_restore_state clears dev->state_saved so call
5627 * pci_save_state to restore it.
5629 pci_save_state(pdev);
5631 err = pci_enable_device_mem(pdev);
5633 e_dev_err("Cannot enable PCI device from suspend\n");
5636 pci_set_master(pdev);
5638 pci_wake_from_d3(pdev, false);
5640 err = ixgbe_init_interrupt_scheme(adapter);
5642 e_dev_err("Cannot initialize interrupts for device\n");
5646 ixgbe_reset(adapter);
5648 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5650 if (netif_running(netdev)) {
5651 err = ixgbe_open(netdev);
5656 netif_device_attach(netdev);
5660 #endif /* CONFIG_PM */
5662 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5664 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5665 struct net_device *netdev = adapter->netdev;
5666 struct ixgbe_hw *hw = &adapter->hw;
5668 u32 wufc = adapter->wol;
5673 netif_device_detach(netdev);
5675 if (netif_running(netdev)) {
5676 ixgbe_down(adapter);
5677 ixgbe_free_irq(adapter);
5678 ixgbe_free_all_tx_resources(adapter);
5679 ixgbe_free_all_rx_resources(adapter);
5682 ixgbe_clear_interrupt_scheme(adapter);
5684 kfree(adapter->ixgbe_ieee_pfc);
5685 kfree(adapter->ixgbe_ieee_ets);
5689 retval = pci_save_state(pdev);
5695 ixgbe_set_rx_mode(netdev);
5697 /* turn on all-multi mode if wake on multicast is enabled */
5698 if (wufc & IXGBE_WUFC_MC) {
5699 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5700 fctrl |= IXGBE_FCTRL_MPE;
5701 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5704 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5705 ctrl |= IXGBE_CTRL_GIO_DIS;
5706 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5708 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5710 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5711 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5714 switch (hw->mac.type) {
5715 case ixgbe_mac_82598EB:
5716 pci_wake_from_d3(pdev, false);
5718 case ixgbe_mac_82599EB:
5719 case ixgbe_mac_X540:
5720 pci_wake_from_d3(pdev, !!wufc);
5726 *enable_wake = !!wufc;
5728 ixgbe_release_hw_control(adapter);
5730 pci_disable_device(pdev);
5736 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5741 retval = __ixgbe_shutdown(pdev, &wake);
5746 pci_prepare_to_sleep(pdev);
5748 pci_wake_from_d3(pdev, false);
5749 pci_set_power_state(pdev, PCI_D3hot);
5754 #endif /* CONFIG_PM */
5756 static void ixgbe_shutdown(struct pci_dev *pdev)
5760 __ixgbe_shutdown(pdev, &wake);
5762 if (system_state == SYSTEM_POWER_OFF) {
5763 pci_wake_from_d3(pdev, wake);
5764 pci_set_power_state(pdev, PCI_D3hot);
5769 * ixgbe_update_stats - Update the board statistics counters.
5770 * @adapter: board private structure
5772 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5774 struct net_device *netdev = adapter->netdev;
5775 struct ixgbe_hw *hw = &adapter->hw;
5776 struct ixgbe_hw_stats *hwstats = &adapter->stats;
5778 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5779 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
5780 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
5781 u64 bytes = 0, packets = 0;
5783 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5784 test_bit(__IXGBE_RESETTING, &adapter->state))
5787 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5790 for (i = 0; i < 16; i++)
5791 adapter->hw_rx_no_dma_resources +=
5792 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5793 for (i = 0; i < adapter->num_rx_queues; i++) {
5794 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
5795 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
5797 adapter->rsc_total_count = rsc_count;
5798 adapter->rsc_total_flush = rsc_flush;
5801 for (i = 0; i < adapter->num_rx_queues; i++) {
5802 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
5803 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
5804 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
5805 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
5806 bytes += rx_ring->stats.bytes;
5807 packets += rx_ring->stats.packets;
5809 adapter->non_eop_descs = non_eop_descs;
5810 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5811 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5812 netdev->stats.rx_bytes = bytes;
5813 netdev->stats.rx_packets = packets;
5817 /* gather some stats to the adapter struct that are per queue */
5818 for (i = 0; i < adapter->num_tx_queues; i++) {
5819 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5820 restart_queue += tx_ring->tx_stats.restart_queue;
5821 tx_busy += tx_ring->tx_stats.tx_busy;
5822 bytes += tx_ring->stats.bytes;
5823 packets += tx_ring->stats.packets;
5825 adapter->restart_queue = restart_queue;
5826 adapter->tx_busy = tx_busy;
5827 netdev->stats.tx_bytes = bytes;
5828 netdev->stats.tx_packets = packets;
5830 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5831 for (i = 0; i < 8; i++) {
5832 /* for packet buffers not used, the register should read 0 */
5833 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5835 hwstats->mpc[i] += mpc;
5836 total_mpc += hwstats->mpc[i];
5837 if (hw->mac.type == ixgbe_mac_82598EB)
5838 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5839 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5840 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5841 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5842 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5843 switch (hw->mac.type) {
5844 case ixgbe_mac_82598EB:
5845 hwstats->pxonrxc[i] +=
5846 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5848 case ixgbe_mac_82599EB:
5849 case ixgbe_mac_X540:
5850 hwstats->pxonrxc[i] +=
5851 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5856 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5857 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5859 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5860 /* work around hardware counting issue */
5861 hwstats->gprc -= missed_rx;
5863 ixgbe_update_xoff_received(adapter);
5865 /* 82598 hardware only has a 32 bit counter in the high register */
5866 switch (hw->mac.type) {
5867 case ixgbe_mac_82598EB:
5868 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5869 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5870 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5871 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5873 case ixgbe_mac_X540:
5874 /* OS2BMC stats are X540 only*/
5875 hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
5876 hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
5877 hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
5878 hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
5879 case ixgbe_mac_82599EB:
5880 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5881 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
5882 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5883 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
5884 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5885 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5886 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5887 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5888 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5890 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5891 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5892 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5893 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5894 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5895 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5896 #endif /* IXGBE_FCOE */
5901 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5902 hwstats->bprc += bprc;
5903 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5904 if (hw->mac.type == ixgbe_mac_82598EB)
5905 hwstats->mprc -= bprc;
5906 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5907 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5908 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5909 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5910 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5911 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5912 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5913 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5914 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5915 hwstats->lxontxc += lxon;
5916 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5917 hwstats->lxofftxc += lxoff;
5918 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5919 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5920 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5922 * 82598 errata - tx of flow control packets is included in tx counters
5924 xon_off_tot = lxon + lxoff;
5925 hwstats->gptc -= xon_off_tot;
5926 hwstats->mptc -= xon_off_tot;
5927 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5928 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5929 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5930 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5931 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5932 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5933 hwstats->ptc64 -= xon_off_tot;
5934 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5935 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5936 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5937 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5938 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5939 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5941 /* Fill out the OS statistics structure */
5942 netdev->stats.multicast = hwstats->mprc;
5945 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
5946 netdev->stats.rx_dropped = 0;
5947 netdev->stats.rx_length_errors = hwstats->rlec;
5948 netdev->stats.rx_crc_errors = hwstats->crcerrs;
5949 netdev->stats.rx_missed_errors = total_mpc;
5953 * ixgbe_fdir_reinit_task - worker thread to reinit FDIR filter table
5954 * @work: pointer to work_struct containing our data
5956 static void ixgbe_fdir_reinit_task(struct work_struct *work)
5958 struct ixgbe_adapter *adapter = container_of(work,
5959 struct ixgbe_adapter,
5961 struct ixgbe_hw *hw = &adapter->hw;
5964 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5965 for (i = 0; i < adapter->num_tx_queues; i++)
5966 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
5967 &(adapter->tx_ring[i]->state));
5969 e_err(probe, "failed to finish FDIR re-initialization, "
5970 "ignored adding FDIR ATR filters\n");
5972 /* Done FDIR Re-initialization, enable transmits */
5973 netif_tx_start_all_queues(adapter->netdev);
5977 * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
5978 * @adapter - pointer to the device adapter structure
5980 * This function serves two purposes. First it strobes the interrupt lines
5981 * in order to make certain interrupts are occuring. Secondly it sets the
5982 * bits needed to check for TX hangs. As a result we should immediately
5983 * determine if a hang has occured.
5985 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
5987 struct ixgbe_hw *hw = &adapter->hw;
5991 /* If we're down or resetting, just bail */
5992 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5993 test_bit(__IXGBE_RESETTING, &adapter->state))
5996 /* Force detection of hung controller */
5997 if (netif_carrier_ok(adapter->netdev)) {
5998 for (i = 0; i < adapter->num_tx_queues; i++)
5999 set_check_for_tx_hang(adapter->tx_ring[i]);
6002 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
6004 * for legacy and MSI interrupts don't set any bits
6005 * that are enabled for EIAM, because this operation
6006 * would set *both* EIMS and EICS for any bit in EIAM
6008 IXGBE_WRITE_REG(hw, IXGBE_EICS,
6009 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
6011 /* get one bit for every active tx/rx interrupt vector */
6012 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
6013 struct ixgbe_q_vector *qv = adapter->q_vector[i];
6014 if (qv->rxr_count || qv->txr_count)
6015 eics |= ((u64)1 << i);
6019 /* Cause software interrupt to ensure rings are cleaned */
6020 ixgbe_irq_rearm_queues(adapter, eics);
6025 * ixgbe_watchdog_update_link - update the link status
6026 * @adapter - pointer to the device adapter structure
6027 * @link_speed - pointer to a u32 to store the link_speed
6029 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
6031 struct ixgbe_hw *hw = &adapter->hw;
6032 u32 link_speed = adapter->link_speed;
6033 bool link_up = adapter->link_up;
6036 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
6039 if (hw->mac.ops.check_link) {
6040 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
6042 /* always assume link is up, if no check link function */
6043 link_speed = IXGBE_LINK_SPEED_10GB_FULL;
6047 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6048 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
6049 hw->mac.ops.fc_enable(hw, i);
6051 hw->mac.ops.fc_enable(hw, 0);
6056 time_after(jiffies, (adapter->link_check_timeout +
6057 IXGBE_TRY_LINK_TIMEOUT))) {
6058 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
6059 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
6060 IXGBE_WRITE_FLUSH(hw);
6063 adapter->link_up = link_up;
6064 adapter->link_speed = link_speed;
6068 * ixgbe_watchdog_link_is_up - update netif_carrier status and
6069 * print link up message
6070 * @adapter - pointer to the device adapter structure
6072 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
6074 struct net_device *netdev = adapter->netdev;
6075 struct ixgbe_hw *hw = &adapter->hw;
6076 u32 link_speed = adapter->link_speed;
6077 bool flow_rx, flow_tx;
6079 /* only continue if link was previously down */
6080 if (netif_carrier_ok(netdev))
6083 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
6085 switch (hw->mac.type) {
6086 case ixgbe_mac_82598EB: {
6087 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
6088 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
6089 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
6090 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
6093 case ixgbe_mac_X540:
6094 case ixgbe_mac_82599EB: {
6095 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
6096 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
6097 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
6098 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
6106 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
6107 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
6109 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
6111 (link_speed == IXGBE_LINK_SPEED_100_FULL ?
6114 ((flow_rx && flow_tx) ? "RX/TX" :
6116 (flow_tx ? "TX" : "None"))));
6118 netif_carrier_on(netdev);
6119 #ifdef HAVE_IPLINK_VF_CONFIG
6120 ixgbe_check_vf_rate_limit(adapter);
6121 #endif /* HAVE_IPLINK_VF_CONFIG */
6125 * ixgbe_watchdog_link_is_down - update netif_carrier status and
6126 * print link down message
6127 * @adapter - pointer to the adapter structure
6129 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter* adapter)
6131 struct net_device *netdev = adapter->netdev;
6132 struct ixgbe_hw *hw = &adapter->hw;
6134 adapter->link_up = false;
6135 adapter->link_speed = 0;
6137 /* only continue if link was up previously */
6138 if (!netif_carrier_ok(netdev))
6141 /* poll for SFP+ cable when link is down */
6142 if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
6143 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
6145 e_info(drv, "NIC Link is Down\n");
6146 netif_carrier_off(netdev);
6150 * ixgbe_watchdog_flush_tx - flush queues on link down
6151 * @adapter - pointer to the device adapter structure
6153 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
6156 int some_tx_pending = 0;
6158 if (!netif_carrier_ok(adapter->netdev)) {
6159 for (i = 0; i < adapter->num_tx_queues; i++) {
6160 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
6161 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
6162 some_tx_pending = 1;
6167 if (some_tx_pending) {
6168 /* We've lost link, so the controller stops DMA,
6169 * but we've got queued Tx work that's never going
6170 * to get done, so reset controller to flush Tx.
6171 * (Do the reset outside of interrupt context).
6173 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
6178 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
6182 /* Do not perform spoof check for 82598 */
6183 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
6186 ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
6189 * ssvpc register is cleared on read, if zero then no
6190 * spoofed packets in the last interval.
6195 e_warn(drv, "%d Spoofed packets detected\n", ssvpc);
6199 * ixgbe_watchdog_subtask - check and bring link up
6200 * @adapter - pointer to the device adapter structure
6202 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
6204 /* if interface is down do nothing */
6205 if (test_bit(__IXGBE_DOWN, &adapter->state))
6208 ixgbe_watchdog_update_link(adapter);
6210 if (adapter->link_up)
6211 ixgbe_watchdog_link_is_up(adapter);
6213 ixgbe_watchdog_link_is_down(adapter);
6215 ixgbe_spoof_check(adapter);
6216 ixgbe_update_stats(adapter);
6218 ixgbe_watchdog_flush_tx(adapter);
6222 * ixgbe_sfp_detection_subtask - poll for SFP+ cable
6223 * @adapter - the ixgbe adapter structure
6225 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
6227 struct ixgbe_hw *hw = &adapter->hw;
6230 /* not searching for SFP so there is nothing to do here */
6231 if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
6232 !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
6235 /* someone else is in init, wait until next service event */
6236 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
6239 err = hw->phy.ops.identify_sfp(hw);
6240 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
6243 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
6244 /* If no cable is present, then we need to reset
6245 * the next time we find a good cable. */
6246 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
6253 /* exit if reset not needed */
6254 if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
6257 adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
6260 * A module may be identified correctly, but the EEPROM may not have
6261 * support for that module. setup_sfp() will fail in that case, so
6262 * we should not allow that module to load.
6264 if (hw->mac.type == ixgbe_mac_82598EB)
6265 err = hw->phy.ops.reset(hw);
6267 err = hw->mac.ops.setup_sfp(hw);
6269 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
6272 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
6273 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
6276 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
6278 if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
6279 (adapter->netdev->reg_state == NETREG_REGISTERED)) {
6280 e_dev_err("failed to initialize because an unsupported "
6281 "SFP+ module type was detected.\n");
6282 e_dev_err("Reload the driver after installing a "
6283 "supported module.\n");
6284 unregister_netdev(adapter->netdev);
6289 * ixgbe_sfp_link_config_subtask - set up link SFP after module install
6290 * @adapter - the ixgbe adapter structure
6292 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
6294 struct ixgbe_hw *hw = &adapter->hw;
6298 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
6301 /* someone else is in init, wait until next service event */
6302 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
6305 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
6307 autoneg = hw->phy.autoneg_advertised;
6308 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
6309 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
6310 hw->mac.autotry_restart = false;
6311 if (hw->mac.ops.setup_link)
6312 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
6314 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
6315 adapter->link_check_timeout = jiffies;
6316 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
6320 * ixgbe_service_timer - Timer Call-back
6321 * @data: pointer to adapter cast into an unsigned long
6323 static void ixgbe_service_timer(unsigned long data)
6325 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
6326 unsigned long next_event_offset;
6328 /* poll faster when waiting for link */
6329 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
6330 next_event_offset = HZ / 10;
6332 next_event_offset = HZ * 2;
6334 /* Reset the timer */
6335 mod_timer(&adapter->service_timer, next_event_offset + jiffies);
6337 ixgbe_service_event_schedule(adapter);
6340 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
6342 if (!(adapter->flags2 & IXGBE_FLAG2_RESET_REQUESTED))
6345 adapter->flags2 &= ~IXGBE_FLAG2_RESET_REQUESTED;
6347 /* If we're already down or resetting, just bail */
6348 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6349 test_bit(__IXGBE_RESETTING, &adapter->state))
6352 ixgbe_dump(adapter);
6353 netdev_err(adapter->netdev, "Reset adapter\n");
6354 adapter->tx_timeout_count++;
6356 ixgbe_reinit_locked(adapter);
6360 * ixgbe_service_task - manages and runs subtasks
6361 * @work: pointer to work_struct containing our data
6363 static void ixgbe_service_task(struct work_struct *work)
6365 struct ixgbe_adapter *adapter = container_of(work,
6366 struct ixgbe_adapter,
6369 ixgbe_reset_subtask(adapter);
6370 ixgbe_sfp_detection_subtask(adapter);
6371 ixgbe_sfp_link_config_subtask(adapter);
6372 ixgbe_watchdog_subtask(adapter);
6373 ixgbe_check_hang_subtask(adapter);
6375 ixgbe_service_event_complete(adapter);
6378 static int ixgbe_tso(struct ixgbe_adapter *adapter,
6379 struct ixgbe_ring *tx_ring, struct sk_buff *skb,
6380 u32 tx_flags, u8 *hdr_len, __be16 protocol)
6382 struct ixgbe_adv_tx_context_desc *context_desc;
6385 struct ixgbe_tx_buffer *tx_buffer_info;
6386 u32 vlan_macip_lens = 0, type_tucmd_mlhl;
6387 u32 mss_l4len_idx, l4len;
6389 if (skb_is_gso(skb)) {
6390 if (skb_header_cloned(skb)) {
6391 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
6395 l4len = tcp_hdrlen(skb);
6398 if (protocol == htons(ETH_P_IP)) {
6399 struct iphdr *iph = ip_hdr(skb);
6402 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
6406 } else if (skb_is_gso_v6(skb)) {
6407 ipv6_hdr(skb)->payload_len = 0;
6408 tcp_hdr(skb)->check =
6409 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
6410 &ipv6_hdr(skb)->daddr,
6414 i = tx_ring->next_to_use;
6416 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6417 context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
6419 /* VLAN MACLEN IPLEN */
6420 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
6422 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
6423 vlan_macip_lens |= ((skb_network_offset(skb)) <<
6424 IXGBE_ADVTXD_MACLEN_SHIFT);
6425 *hdr_len += skb_network_offset(skb);
6427 (skb_transport_header(skb) - skb_network_header(skb));
6429 (skb_transport_header(skb) - skb_network_header(skb));
6430 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
6431 context_desc->seqnum_seed = 0;
6433 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
6434 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
6435 IXGBE_ADVTXD_DTYP_CTXT);
6437 if (protocol == htons(ETH_P_IP))
6438 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
6439 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6440 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
6444 (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
6445 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
6446 /* use index 1 for TSO */
6447 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6448 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
6450 tx_buffer_info->time_stamp = jiffies;
6451 tx_buffer_info->next_to_watch = i;
6454 if (i == tx_ring->count)
6456 tx_ring->next_to_use = i;
6463 static u32 ixgbe_psum(struct ixgbe_adapter *adapter, struct sk_buff *skb,
6469 case cpu_to_be16(ETH_P_IP):
6470 rtn |= IXGBE_ADVTXD_TUCMD_IPV4;
6471 switch (ip_hdr(skb)->protocol) {
6473 rtn |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6476 rtn |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
6480 case cpu_to_be16(ETH_P_IPV6):
6481 /* XXX what about other V6 headers?? */
6482 switch (ipv6_hdr(skb)->nexthdr) {
6484 rtn |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6487 rtn |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
6492 if (unlikely(net_ratelimit()))
6493 e_warn(probe, "partial checksum but proto=%x!\n",
6501 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
6502 struct ixgbe_ring *tx_ring,
6503 struct sk_buff *skb, u32 tx_flags,
6506 struct ixgbe_adv_tx_context_desc *context_desc;
6508 struct ixgbe_tx_buffer *tx_buffer_info;
6509 u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
6511 if (skb->ip_summed == CHECKSUM_PARTIAL ||
6512 (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
6513 i = tx_ring->next_to_use;
6514 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6515 context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
6517 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
6519 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
6520 vlan_macip_lens |= (skb_network_offset(skb) <<
6521 IXGBE_ADVTXD_MACLEN_SHIFT);
6522 if (skb->ip_summed == CHECKSUM_PARTIAL)
6523 vlan_macip_lens |= (skb_transport_header(skb) -
6524 skb_network_header(skb));
6526 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
6527 context_desc->seqnum_seed = 0;
6529 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
6530 IXGBE_ADVTXD_DTYP_CTXT);
6532 if (skb->ip_summed == CHECKSUM_PARTIAL)
6533 type_tucmd_mlhl |= ixgbe_psum(adapter, skb, protocol);
6535 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
6536 /* use index zero for tx checksum offload */
6537 context_desc->mss_l4len_idx = 0;
6539 tx_buffer_info->time_stamp = jiffies;
6540 tx_buffer_info->next_to_watch = i;
6543 if (i == tx_ring->count)
6545 tx_ring->next_to_use = i;
6553 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
6554 struct ixgbe_ring *tx_ring,
6555 struct sk_buff *skb, u32 tx_flags,
6556 unsigned int first, const u8 hdr_len)
6558 struct device *dev = tx_ring->dev;
6559 struct ixgbe_tx_buffer *tx_buffer_info;
6561 unsigned int total = skb->len;
6562 unsigned int offset = 0, size, count = 0, i;
6563 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
6565 unsigned int bytecount = skb->len;
6568 i = tx_ring->next_to_use;
6570 if (tx_flags & IXGBE_TX_FLAGS_FCOE)
6571 /* excluding fcoe_crc_eof for FCoE */
6572 total -= sizeof(struct fcoe_crc_eof);
6574 len = min(skb_headlen(skb), total);
6576 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6577 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
6579 tx_buffer_info->length = size;
6580 tx_buffer_info->mapped_as_page = false;
6581 tx_buffer_info->dma = dma_map_single(dev,
6583 size, DMA_TO_DEVICE);
6584 if (dma_mapping_error(dev, tx_buffer_info->dma))
6586 tx_buffer_info->time_stamp = jiffies;
6587 tx_buffer_info->next_to_watch = i;
6596 if (i == tx_ring->count)
6601 for (f = 0; f < nr_frags; f++) {
6602 struct skb_frag_struct *frag;
6604 frag = &skb_shinfo(skb)->frags[f];
6605 len = min((unsigned int)frag->size, total);
6606 offset = frag->page_offset;
6610 if (i == tx_ring->count)
6613 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6614 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
6616 tx_buffer_info->length = size;
6617 tx_buffer_info->dma = dma_map_page(dev,
6621 tx_buffer_info->mapped_as_page = true;
6622 if (dma_mapping_error(dev, tx_buffer_info->dma))
6624 tx_buffer_info->time_stamp = jiffies;
6625 tx_buffer_info->next_to_watch = i;
6636 if (tx_flags & IXGBE_TX_FLAGS_TSO)
6637 gso_segs = skb_shinfo(skb)->gso_segs;
6639 /* adjust for FCoE Sequence Offload */
6640 else if (tx_flags & IXGBE_TX_FLAGS_FSO)
6641 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
6642 skb_shinfo(skb)->gso_size);
6643 #endif /* IXGBE_FCOE */
6644 bytecount += (gso_segs - 1) * hdr_len;
6646 /* multiply data chunks by size of headers */
6647 tx_ring->tx_buffer_info[i].bytecount = bytecount;
6648 tx_ring->tx_buffer_info[i].gso_segs = gso_segs;
6649 tx_ring->tx_buffer_info[i].skb = skb;
6650 tx_ring->tx_buffer_info[first].next_to_watch = i;
6655 e_dev_err("TX DMA map failed\n");
6657 /* clear timestamp and dma mappings for failed tx_buffer_info map */
6658 tx_buffer_info->dma = 0;
6659 tx_buffer_info->time_stamp = 0;
6660 tx_buffer_info->next_to_watch = 0;
6664 /* clear timestamp and dma mappings for remaining portion of packet */
6667 i += tx_ring->count;
6669 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6670 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
6676 static void ixgbe_tx_queue(struct ixgbe_ring *tx_ring,
6677 int tx_flags, int count, u32 paylen, u8 hdr_len)
6679 union ixgbe_adv_tx_desc *tx_desc = NULL;
6680 struct ixgbe_tx_buffer *tx_buffer_info;
6681 u32 olinfo_status = 0, cmd_type_len = 0;
6683 u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
6685 cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
6687 cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
6689 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
6690 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
6692 if (tx_flags & IXGBE_TX_FLAGS_TSO) {
6693 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
6695 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
6696 IXGBE_ADVTXD_POPTS_SHIFT;
6698 /* use index 1 context for tso */
6699 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6700 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
6701 olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
6702 IXGBE_ADVTXD_POPTS_SHIFT;
6704 } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
6705 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
6706 IXGBE_ADVTXD_POPTS_SHIFT;
6708 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6709 olinfo_status |= IXGBE_ADVTXD_CC;
6710 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6711 if (tx_flags & IXGBE_TX_FLAGS_FSO)
6712 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
6715 olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
6717 i = tx_ring->next_to_use;
6719 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6720 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
6721 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
6722 tx_desc->read.cmd_type_len =
6723 cpu_to_le32(cmd_type_len | tx_buffer_info->length);
6724 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
6726 if (i == tx_ring->count)
6730 tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
6733 * Force memory writes to complete before letting h/w
6734 * know there are new descriptors to fetch. (Only
6735 * applicable for weak-ordered memory model archs,
6740 tx_ring->next_to_use = i;
6741 writel(i, tx_ring->tail);
6744 static void ixgbe_atr(struct ixgbe_ring *ring, struct sk_buff *skb,
6745 u32 tx_flags, __be16 protocol)
6747 struct ixgbe_q_vector *q_vector = ring->q_vector;
6748 union ixgbe_atr_hash_dword input = { .dword = 0 };
6749 union ixgbe_atr_hash_dword common = { .dword = 0 };
6751 unsigned char *network;
6753 struct ipv6hdr *ipv6;
6758 /* if ring doesn't have a interrupt vector, cannot perform ATR */
6762 /* do nothing if sampling is disabled */
6763 if (!ring->atr_sample_rate)
6768 /* snag network header to get L4 type and address */
6769 hdr.network = skb_network_header(skb);
6771 /* Currently only IPv4/IPv6 with TCP is supported */
6772 if ((protocol != __constant_htons(ETH_P_IPV6) ||
6773 hdr.ipv6->nexthdr != IPPROTO_TCP) &&
6774 (protocol != __constant_htons(ETH_P_IP) ||
6775 hdr.ipv4->protocol != IPPROTO_TCP))
6780 /* skip this packet since the socket is closing */
6784 /* sample on all syn packets or once every atr sample count */
6785 if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
6788 /* reset sample count */
6789 ring->atr_count = 0;
6791 vlan_id = htons(tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
6794 * src and dst are inverted, think how the receiver sees them
6796 * The input is broken into two sections, a non-compressed section
6797 * containing vm_pool, vlan_id, and flow_type. The rest of the data
6798 * is XORed together and stored in the compressed dword.
6800 input.formatted.vlan_id = vlan_id;
6803 * since src port and flex bytes occupy the same word XOR them together
6804 * and write the value to source port portion of compressed dword
6807 common.port.src ^= th->dest ^ __constant_htons(ETH_P_8021Q);
6809 common.port.src ^= th->dest ^ protocol;
6810 common.port.dst ^= th->source;
6812 if (protocol == __constant_htons(ETH_P_IP)) {
6813 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
6814 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
6816 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
6817 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
6818 hdr.ipv6->saddr.s6_addr32[1] ^
6819 hdr.ipv6->saddr.s6_addr32[2] ^
6820 hdr.ipv6->saddr.s6_addr32[3] ^
6821 hdr.ipv6->daddr.s6_addr32[0] ^
6822 hdr.ipv6->daddr.s6_addr32[1] ^
6823 hdr.ipv6->daddr.s6_addr32[2] ^
6824 hdr.ipv6->daddr.s6_addr32[3];
6827 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
6828 ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
6829 input, common, ring->queue_index);
6832 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, int size)
6834 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
6835 /* Herbert's original patch had:
6836 * smp_mb__after_netif_stop_queue();
6837 * but since that doesn't exist yet, just open code it. */
6840 /* We need to check again in a case another CPU has just
6841 * made room available. */
6842 if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
6845 /* A reprieve! - use start_queue because it doesn't call schedule */
6846 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
6847 ++tx_ring->tx_stats.restart_queue;
6851 static int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, int size)
6853 if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
6855 return __ixgbe_maybe_stop_tx(tx_ring, size);
6858 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6860 struct ixgbe_adapter *adapter = netdev_priv(dev);
6861 int txq = smp_processor_id();
6865 protocol = vlan_get_protocol(skb);
6867 if (((protocol == htons(ETH_P_FCOE)) ||
6868 (protocol == htons(ETH_P_FIP))) &&
6869 (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)) {
6870 txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
6871 txq += adapter->ring_feature[RING_F_FCOE].mask;
6876 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6877 while (unlikely(txq >= dev->real_num_tx_queues))
6878 txq -= dev->real_num_tx_queues;
6882 return skb_tx_hash(dev, skb);
6885 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
6886 struct ixgbe_adapter *adapter,
6887 struct ixgbe_ring *tx_ring)
6890 unsigned int tx_flags = 0;
6897 protocol = vlan_get_protocol(skb);
6899 if (vlan_tx_tag_present(skb)) {
6900 tx_flags |= vlan_tx_tag_get(skb);
6901 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6902 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6903 tx_flags |= tx_ring->dcb_tc << 13;
6905 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6906 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6907 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED &&
6908 skb->priority != TC_PRIO_CONTROL) {
6909 tx_flags |= tx_ring->dcb_tc << 13;
6910 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6911 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6915 /* for FCoE with DCB, we force the priority to what
6916 * was specified by the switch */
6917 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED &&
6918 (protocol == htons(ETH_P_FCOE)))
6919 tx_flags |= IXGBE_TX_FLAGS_FCOE;
6922 /* four things can cause us to need a context descriptor */
6923 if (skb_is_gso(skb) ||
6924 (skb->ip_summed == CHECKSUM_PARTIAL) ||
6925 (tx_flags & IXGBE_TX_FLAGS_VLAN) ||
6926 (tx_flags & IXGBE_TX_FLAGS_FCOE))
6929 count += TXD_USE_COUNT(skb_headlen(skb));
6930 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6931 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6933 if (ixgbe_maybe_stop_tx(tx_ring, count)) {
6934 tx_ring->tx_stats.tx_busy++;
6935 return NETDEV_TX_BUSY;
6938 first = tx_ring->next_to_use;
6939 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6941 /* setup tx offload for FCoE */
6942 tso = ixgbe_fso(adapter, tx_ring, skb, tx_flags, &hdr_len);
6944 dev_kfree_skb_any(skb);
6945 return NETDEV_TX_OK;
6948 tx_flags |= IXGBE_TX_FLAGS_FSO;
6949 #endif /* IXGBE_FCOE */
6951 if (protocol == htons(ETH_P_IP))
6952 tx_flags |= IXGBE_TX_FLAGS_IPV4;
6953 tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len,
6956 dev_kfree_skb_any(skb);
6957 return NETDEV_TX_OK;
6961 tx_flags |= IXGBE_TX_FLAGS_TSO;
6962 else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags,
6964 (skb->ip_summed == CHECKSUM_PARTIAL))
6965 tx_flags |= IXGBE_TX_FLAGS_CSUM;
6968 count = ixgbe_tx_map(adapter, tx_ring, skb, tx_flags, first, hdr_len);
6970 /* add the ATR filter if ATR is on */
6971 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
6972 ixgbe_atr(tx_ring, skb, tx_flags, protocol);
6973 ixgbe_tx_queue(tx_ring, tx_flags, count, skb->len, hdr_len);
6974 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
6977 dev_kfree_skb_any(skb);
6978 tx_ring->tx_buffer_info[first].time_stamp = 0;
6979 tx_ring->next_to_use = first;
6982 return NETDEV_TX_OK;
6985 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
6987 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6988 struct ixgbe_ring *tx_ring;
6990 tx_ring = adapter->tx_ring[skb->queue_mapping];
6991 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
6995 * ixgbe_set_mac - Change the Ethernet Address of the NIC
6996 * @netdev: network interface device structure
6997 * @p: pointer to an address structure
6999 * Returns 0 on success, negative on failure
7001 static int ixgbe_set_mac(struct net_device *netdev, void *p)
7003 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7004 struct ixgbe_hw *hw = &adapter->hw;
7005 struct sockaddr *addr = p;
7007 if (!is_valid_ether_addr(addr->sa_data))
7008 return -EADDRNOTAVAIL;
7010 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
7011 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
7013 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
7020 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
7022 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7023 struct ixgbe_hw *hw = &adapter->hw;
7027 if (prtad != hw->phy.mdio.prtad)
7029 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
7035 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
7036 u16 addr, u16 value)
7038 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7039 struct ixgbe_hw *hw = &adapter->hw;
7041 if (prtad != hw->phy.mdio.prtad)
7043 return hw->phy.ops.write_reg(hw, addr, devad, value);
7046 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
7048 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7050 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
7054 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
7056 * @netdev: network interface device structure
7058 * Returns non-zero on failure
7060 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
7063 struct ixgbe_adapter *adapter = netdev_priv(dev);
7064 struct ixgbe_mac_info *mac = &adapter->hw.mac;
7066 if (is_valid_ether_addr(mac->san_addr)) {
7068 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
7075 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
7077 * @netdev: network interface device structure
7079 * Returns non-zero on failure
7081 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
7084 struct ixgbe_adapter *adapter = netdev_priv(dev);
7085 struct ixgbe_mac_info *mac = &adapter->hw.mac;
7087 if (is_valid_ether_addr(mac->san_addr)) {
7089 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
7095 #ifdef CONFIG_NET_POLL_CONTROLLER
7097 * Polling 'interrupt' - used by things like netconsole to send skbs
7098 * without having to re-enable interrupts. It's not called while
7099 * the interrupt routine is executing.
7101 static void ixgbe_netpoll(struct net_device *netdev)
7103 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7106 /* if interface is down do nothing */
7107 if (test_bit(__IXGBE_DOWN, &adapter->state))
7110 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
7111 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
7112 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
7113 for (i = 0; i < num_q_vectors; i++) {
7114 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
7115 ixgbe_msix_clean_many(0, q_vector);
7118 ixgbe_intr(adapter->pdev->irq, netdev);
7120 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
7124 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
7125 struct rtnl_link_stats64 *stats)
7127 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7131 for (i = 0; i < adapter->num_rx_queues; i++) {
7132 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
7138 start = u64_stats_fetch_begin_bh(&ring->syncp);
7139 packets = ring->stats.packets;
7140 bytes = ring->stats.bytes;
7141 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
7142 stats->rx_packets += packets;
7143 stats->rx_bytes += bytes;
7147 for (i = 0; i < adapter->num_tx_queues; i++) {
7148 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->tx_ring[i]);
7154 start = u64_stats_fetch_begin_bh(&ring->syncp);
7155 packets = ring->stats.packets;
7156 bytes = ring->stats.bytes;
7157 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
7158 stats->tx_packets += packets;
7159 stats->tx_bytes += bytes;
7163 /* following stats updated by ixgbe_watchdog_task() */
7164 stats->multicast = netdev->stats.multicast;
7165 stats->rx_errors = netdev->stats.rx_errors;
7166 stats->rx_length_errors = netdev->stats.rx_length_errors;
7167 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
7168 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
7173 static const struct net_device_ops ixgbe_netdev_ops = {
7174 .ndo_open = ixgbe_open,
7175 .ndo_stop = ixgbe_close,
7176 .ndo_start_xmit = ixgbe_xmit_frame,
7177 .ndo_select_queue = ixgbe_select_queue,
7178 .ndo_set_rx_mode = ixgbe_set_rx_mode,
7179 .ndo_set_multicast_list = ixgbe_set_rx_mode,
7180 .ndo_validate_addr = eth_validate_addr,
7181 .ndo_set_mac_address = ixgbe_set_mac,
7182 .ndo_change_mtu = ixgbe_change_mtu,
7183 .ndo_tx_timeout = ixgbe_tx_timeout,
7184 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
7185 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
7186 .ndo_do_ioctl = ixgbe_ioctl,
7187 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
7188 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
7189 .ndo_set_vf_tx_rate = ixgbe_ndo_set_vf_bw,
7190 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
7191 .ndo_get_stats64 = ixgbe_get_stats64,
7192 #ifdef CONFIG_IXGBE_DCB
7193 .ndo_setup_tc = ixgbe_setup_tc,
7195 #ifdef CONFIG_NET_POLL_CONTROLLER
7196 .ndo_poll_controller = ixgbe_netpoll,
7199 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
7200 .ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
7201 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
7202 .ndo_fcoe_enable = ixgbe_fcoe_enable,
7203 .ndo_fcoe_disable = ixgbe_fcoe_disable,
7204 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
7205 #endif /* IXGBE_FCOE */
7208 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
7209 const struct ixgbe_info *ii)
7211 #ifdef CONFIG_PCI_IOV
7212 struct ixgbe_hw *hw = &adapter->hw;
7214 int num_vf_macvlans, i;
7215 struct vf_macvlans *mv_list;
7217 if (hw->mac.type == ixgbe_mac_82598EB || !max_vfs)
7220 /* The 82599 supports up to 64 VFs per physical function
7221 * but this implementation limits allocation to 63 so that
7222 * basic networking resources are still available to the
7225 adapter->num_vfs = (max_vfs > 63) ? 63 : max_vfs;
7226 adapter->flags |= IXGBE_FLAG_SRIOV_ENABLED;
7227 err = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
7229 e_err(probe, "Failed to enable PCI sriov: %d\n", err);
7233 num_vf_macvlans = hw->mac.num_rar_entries -
7234 (IXGBE_MAX_PF_MACVLANS + 1 + adapter->num_vfs);
7236 adapter->mv_list = mv_list = kcalloc(num_vf_macvlans,
7237 sizeof(struct vf_macvlans),
7240 /* Initialize list of VF macvlans */
7241 INIT_LIST_HEAD(&adapter->vf_mvs.l);
7242 for (i = 0; i < num_vf_macvlans; i++) {
7244 mv_list->free = true;
7245 mv_list->rar_entry = hw->mac.num_rar_entries -
7246 (i + adapter->num_vfs + 1);
7247 list_add(&mv_list->l, &adapter->vf_mvs.l);
7252 /* If call to enable VFs succeeded then allocate memory
7253 * for per VF control structures.
7256 kcalloc(adapter->num_vfs,
7257 sizeof(struct vf_data_storage), GFP_KERNEL);
7258 if (adapter->vfinfo) {
7259 /* Now that we're sure SR-IOV is enabled
7260 * and memory allocated set up the mailbox parameters
7262 ixgbe_init_mbx_params_pf(hw);
7263 memcpy(&hw->mbx.ops, ii->mbx_ops,
7264 sizeof(hw->mbx.ops));
7266 /* Disable RSC when in SR-IOV mode */
7267 adapter->flags2 &= ~(IXGBE_FLAG2_RSC_CAPABLE |
7268 IXGBE_FLAG2_RSC_ENABLED);
7273 e_err(probe, "Unable to allocate memory for VF Data Storage - "
7274 "SRIOV disabled\n");
7275 pci_disable_sriov(adapter->pdev);
7278 adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
7279 adapter->num_vfs = 0;
7280 #endif /* CONFIG_PCI_IOV */
7284 * ixgbe_probe - Device Initialization Routine
7285 * @pdev: PCI device information struct
7286 * @ent: entry in ixgbe_pci_tbl
7288 * Returns 0 on success, negative on failure
7290 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
7291 * The OS initialization, configuring of the adapter private structure,
7292 * and a hardware reset occur.
7294 static int __devinit ixgbe_probe(struct pci_dev *pdev,
7295 const struct pci_device_id *ent)
7297 struct net_device *netdev;
7298 struct ixgbe_adapter *adapter = NULL;
7299 struct ixgbe_hw *hw;
7300 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
7301 static int cards_found;
7302 int i, err, pci_using_dac;
7303 u8 part_str[IXGBE_PBANUM_LENGTH];
7304 unsigned int indices = num_possible_cpus();
7310 /* Catch broken hardware that put the wrong VF device ID in
7311 * the PCIe SR-IOV capability.
7313 if (pdev->is_virtfn) {
7314 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
7315 pci_name(pdev), pdev->vendor, pdev->device);
7319 err = pci_enable_device_mem(pdev);
7323 if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
7324 !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
7327 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
7329 err = dma_set_coherent_mask(&pdev->dev,
7333 "No usable DMA configuration, aborting\n");
7340 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
7341 IORESOURCE_MEM), ixgbe_driver_name);
7344 "pci_request_selected_regions failed 0x%x\n", err);
7348 pci_enable_pcie_error_reporting(pdev);
7350 pci_set_master(pdev);
7351 pci_save_state(pdev);
7353 if (ii->mac == ixgbe_mac_82598EB)
7354 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
7356 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
7358 #if defined(CONFIG_DCB)
7359 indices = max_t(unsigned int, indices, IXGBE_MAX_DCB_INDICES);
7360 #elif defined(IXGBE_FCOE)
7361 indices += min_t(unsigned int, num_possible_cpus(),
7362 IXGBE_MAX_FCOE_INDICES);
7364 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
7367 goto err_alloc_etherdev;
7370 SET_NETDEV_DEV(netdev, &pdev->dev);
7372 adapter = netdev_priv(netdev);
7373 pci_set_drvdata(pdev, adapter);
7375 adapter->netdev = netdev;
7376 adapter->pdev = pdev;
7379 adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
7381 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
7382 pci_resource_len(pdev, 0));
7388 for (i = 1; i <= 5; i++) {
7389 if (pci_resource_len(pdev, i) == 0)
7393 netdev->netdev_ops = &ixgbe_netdev_ops;
7394 ixgbe_set_ethtool_ops(netdev);
7395 netdev->watchdog_timeo = 5 * HZ;
7396 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
7398 adapter->bd_number = cards_found;
7401 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
7402 hw->mac.type = ii->mac;
7405 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
7406 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
7407 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
7408 if (!(eec & (1 << 8)))
7409 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
7412 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
7413 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
7414 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
7415 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
7416 hw->phy.mdio.mmds = 0;
7417 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
7418 hw->phy.mdio.dev = netdev;
7419 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
7420 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
7422 ii->get_invariants(hw);
7424 /* setup the private structure */
7425 err = ixgbe_sw_init(adapter);
7429 /* Make it possible the adapter to be woken up via WOL */
7430 switch (adapter->hw.mac.type) {
7431 case ixgbe_mac_82599EB:
7432 case ixgbe_mac_X540:
7433 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7440 * If there is a fan on this device and it has failed log the
7443 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
7444 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
7445 if (esdp & IXGBE_ESDP_SDP1)
7446 e_crit(probe, "Fan has stopped, replace the adapter\n");
7449 /* reset_hw fills in the perm_addr as well */
7450 hw->phy.reset_if_overtemp = true;
7451 err = hw->mac.ops.reset_hw(hw);
7452 hw->phy.reset_if_overtemp = false;
7453 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
7454 hw->mac.type == ixgbe_mac_82598EB) {
7456 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
7457 e_dev_err("failed to load because an unsupported SFP+ "
7458 "module type was detected.\n");
7459 e_dev_err("Reload the driver after installing a supported "
7463 e_dev_err("HW Init failed: %d\n", err);
7467 ixgbe_probe_vf(adapter, ii);
7469 netdev->features = NETIF_F_SG |
7471 NETIF_F_HW_VLAN_TX |
7472 NETIF_F_HW_VLAN_RX |
7473 NETIF_F_HW_VLAN_FILTER;
7475 netdev->features |= NETIF_F_IPV6_CSUM;
7476 netdev->features |= NETIF_F_TSO;
7477 netdev->features |= NETIF_F_TSO6;
7478 netdev->features |= NETIF_F_GRO;
7479 netdev->features |= NETIF_F_RXHASH;
7481 switch (adapter->hw.mac.type) {
7482 case ixgbe_mac_82599EB:
7483 case ixgbe_mac_X540:
7484 netdev->features |= NETIF_F_SCTP_CSUM;
7490 netdev->vlan_features |= NETIF_F_TSO;
7491 netdev->vlan_features |= NETIF_F_TSO6;
7492 netdev->vlan_features |= NETIF_F_IP_CSUM;
7493 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
7494 netdev->vlan_features |= NETIF_F_SG;
7496 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7497 adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
7498 IXGBE_FLAG_DCB_ENABLED);
7500 #ifdef CONFIG_IXGBE_DCB
7501 netdev->dcbnl_ops = &dcbnl_ops;
7505 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7506 if (hw->mac.ops.get_device_caps) {
7507 hw->mac.ops.get_device_caps(hw, &device_caps);
7508 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
7509 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
7512 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7513 netdev->vlan_features |= NETIF_F_FCOE_CRC;
7514 netdev->vlan_features |= NETIF_F_FSO;
7515 netdev->vlan_features |= NETIF_F_FCOE_MTU;
7517 #endif /* IXGBE_FCOE */
7518 if (pci_using_dac) {
7519 netdev->features |= NETIF_F_HIGHDMA;
7520 netdev->vlan_features |= NETIF_F_HIGHDMA;
7523 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
7524 netdev->features |= NETIF_F_LRO;
7526 /* make sure the EEPROM is good */
7527 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
7528 e_dev_err("The EEPROM Checksum Is Not Valid\n");
7533 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
7534 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
7536 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
7537 e_dev_err("invalid MAC address\n");
7542 /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
7543 if (hw->mac.ops.disable_tx_laser &&
7544 ((hw->phy.multispeed_fiber) ||
7545 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
7546 (hw->mac.type == ixgbe_mac_82599EB))))
7547 hw->mac.ops.disable_tx_laser(hw);
7549 setup_timer(&adapter->service_timer, &ixgbe_service_timer,
7550 (unsigned long) adapter);
7552 INIT_WORK(&adapter->service_task, ixgbe_service_task);
7553 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
7555 err = ixgbe_init_interrupt_scheme(adapter);
7559 if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED))
7560 netdev->features &= ~NETIF_F_RXHASH;
7562 switch (pdev->device) {
7563 case IXGBE_DEV_ID_82599_SFP:
7564 /* Only this subdevice supports WOL */
7565 if (pdev->subsystem_device == IXGBE_SUBDEV_ID_82599_SFP)
7566 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
7567 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
7569 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
7570 /* All except this subdevice support WOL */
7571 if (pdev->subsystem_device != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
7572 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
7573 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
7575 case IXGBE_DEV_ID_82599_KX4:
7576 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
7577 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
7583 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
7585 /* pick up the PCI bus settings for reporting later */
7586 hw->mac.ops.get_bus_info(hw);
7588 /* print bus type/speed/width info */
7589 e_dev_info("(PCI Express:%s:%s) %pM\n",
7590 (hw->bus.speed == ixgbe_bus_speed_5000 ? "5.0GT/s" :
7591 hw->bus.speed == ixgbe_bus_speed_2500 ? "2.5GT/s" :
7593 (hw->bus.width == ixgbe_bus_width_pcie_x8 ? "Width x8" :
7594 hw->bus.width == ixgbe_bus_width_pcie_x4 ? "Width x4" :
7595 hw->bus.width == ixgbe_bus_width_pcie_x1 ? "Width x1" :
7599 err = ixgbe_read_pba_string_generic(hw, part_str, IXGBE_PBANUM_LENGTH);
7601 strncpy(part_str, "Unknown", IXGBE_PBANUM_LENGTH);
7602 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
7603 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
7604 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
7607 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
7608 hw->mac.type, hw->phy.type, part_str);
7610 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
7611 e_dev_warn("PCI-Express bandwidth available for this card is "
7612 "not sufficient for optimal performance.\n");
7613 e_dev_warn("For optimal performance a x8 PCI-Express slot "
7617 /* save off EEPROM version number */
7618 hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
7620 /* reset the hardware with the new settings */
7621 err = hw->mac.ops.start_hw(hw);
7623 if (err == IXGBE_ERR_EEPROM_VERSION) {
7624 /* We are running on a pre-production device, log a warning */
7625 e_dev_warn("This device is a pre-production adapter/LOM. "
7626 "Please be aware there may be issues associated "
7627 "with your hardware. If you are experiencing "
7628 "problems please contact your Intel or hardware "
7629 "representative who provided you with this "
7632 strcpy(netdev->name, "eth%d");
7633 err = register_netdev(netdev);
7637 /* carrier off reporting is important to ethtool even BEFORE open */
7638 netif_carrier_off(netdev);
7640 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
7641 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
7642 INIT_WORK(&adapter->fdir_reinit_task, ixgbe_fdir_reinit_task);
7644 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
7645 INIT_WORK(&adapter->check_overtemp_task,
7646 ixgbe_check_overtemp_task);
7647 #ifdef CONFIG_IXGBE_DCA
7648 if (dca_add_requester(&pdev->dev) == 0) {
7649 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
7650 ixgbe_setup_dca(adapter);
7653 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7654 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
7655 for (i = 0; i < adapter->num_vfs; i++)
7656 ixgbe_vf_configuration(pdev, (i | 0x10000000));
7659 /* add san mac addr to netdev */
7660 ixgbe_add_sanmac_netdev(netdev);
7662 e_dev_info("Intel(R) 10 Gigabit Network Connection\n");
7667 ixgbe_release_hw_control(adapter);
7668 ixgbe_clear_interrupt_scheme(adapter);
7671 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7672 ixgbe_disable_sriov(adapter);
7673 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
7674 iounmap(hw->hw_addr);
7676 free_netdev(netdev);
7678 pci_release_selected_regions(pdev,
7679 pci_select_bars(pdev, IORESOURCE_MEM));
7682 pci_disable_device(pdev);
7687 * ixgbe_remove - Device Removal Routine
7688 * @pdev: PCI device information struct
7690 * ixgbe_remove is called by the PCI subsystem to alert the driver
7691 * that it should release a PCI device. The could be caused by a
7692 * Hot-Plug event, or because the driver is going to be removed from
7695 static void __devexit ixgbe_remove(struct pci_dev *pdev)
7697 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7698 struct net_device *netdev = adapter->netdev;
7700 set_bit(__IXGBE_DOWN, &adapter->state);
7701 cancel_work_sync(&adapter->service_task);
7703 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
7704 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
7705 cancel_work_sync(&adapter->fdir_reinit_task);
7706 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
7707 cancel_work_sync(&adapter->check_overtemp_task);
7709 #ifdef CONFIG_IXGBE_DCA
7710 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
7711 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
7712 dca_remove_requester(&pdev->dev);
7713 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
7718 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
7719 ixgbe_cleanup_fcoe(adapter);
7721 #endif /* IXGBE_FCOE */
7723 /* remove the added san mac */
7724 ixgbe_del_sanmac_netdev(netdev);
7726 if (netdev->reg_state == NETREG_REGISTERED)
7727 unregister_netdev(netdev);
7729 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7730 ixgbe_disable_sriov(adapter);
7732 ixgbe_clear_interrupt_scheme(adapter);
7734 ixgbe_release_hw_control(adapter);
7736 iounmap(adapter->hw.hw_addr);
7737 pci_release_selected_regions(pdev, pci_select_bars(pdev,
7740 e_dev_info("complete\n");
7742 free_netdev(netdev);
7744 pci_disable_pcie_error_reporting(pdev);
7746 pci_disable_device(pdev);
7750 * ixgbe_io_error_detected - called when PCI error is detected
7751 * @pdev: Pointer to PCI device
7752 * @state: The current pci connection state
7754 * This function is called after a PCI bus error affecting
7755 * this device has been detected.
7757 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
7758 pci_channel_state_t state)
7760 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7761 struct net_device *netdev = adapter->netdev;
7763 netif_device_detach(netdev);
7765 if (state == pci_channel_io_perm_failure)
7766 return PCI_ERS_RESULT_DISCONNECT;
7768 if (netif_running(netdev))
7769 ixgbe_down(adapter);
7770 pci_disable_device(pdev);
7772 /* Request a slot reset. */
7773 return PCI_ERS_RESULT_NEED_RESET;
7777 * ixgbe_io_slot_reset - called after the pci bus has been reset.
7778 * @pdev: Pointer to PCI device
7780 * Restart the card from scratch, as if from a cold-boot.
7782 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7784 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7785 pci_ers_result_t result;
7788 if (pci_enable_device_mem(pdev)) {
7789 e_err(probe, "Cannot re-enable PCI device after reset.\n");
7790 result = PCI_ERS_RESULT_DISCONNECT;
7792 pci_set_master(pdev);
7793 pci_restore_state(pdev);
7794 pci_save_state(pdev);
7796 pci_wake_from_d3(pdev, false);
7798 ixgbe_reset(adapter);
7799 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7800 result = PCI_ERS_RESULT_RECOVERED;
7803 err = pci_cleanup_aer_uncorrect_error_status(pdev);
7805 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7806 "failed 0x%0x\n", err);
7807 /* non-fatal, continue */
7814 * ixgbe_io_resume - called when traffic can start flowing again.
7815 * @pdev: Pointer to PCI device
7817 * This callback is called when the error recovery driver tells us that
7818 * its OK to resume normal operation.
7820 static void ixgbe_io_resume(struct pci_dev *pdev)
7822 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7823 struct net_device *netdev = adapter->netdev;
7825 if (netif_running(netdev)) {
7826 if (ixgbe_up(adapter)) {
7827 e_info(probe, "ixgbe_up failed after reset\n");
7832 netif_device_attach(netdev);
7835 static struct pci_error_handlers ixgbe_err_handler = {
7836 .error_detected = ixgbe_io_error_detected,
7837 .slot_reset = ixgbe_io_slot_reset,
7838 .resume = ixgbe_io_resume,
7841 static struct pci_driver ixgbe_driver = {
7842 .name = ixgbe_driver_name,
7843 .id_table = ixgbe_pci_tbl,
7844 .probe = ixgbe_probe,
7845 .remove = __devexit_p(ixgbe_remove),
7847 .suspend = ixgbe_suspend,
7848 .resume = ixgbe_resume,
7850 .shutdown = ixgbe_shutdown,
7851 .err_handler = &ixgbe_err_handler
7855 * ixgbe_init_module - Driver Registration Routine
7857 * ixgbe_init_module is the first routine called when the driver is
7858 * loaded. All it does is register with the PCI subsystem.
7860 static int __init ixgbe_init_module(void)
7863 pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
7864 pr_info("%s\n", ixgbe_copyright);
7866 #ifdef CONFIG_IXGBE_DCA
7867 dca_register_notify(&dca_notifier);
7870 ret = pci_register_driver(&ixgbe_driver);
7874 module_init(ixgbe_init_module);
7877 * ixgbe_exit_module - Driver Exit Cleanup Routine
7879 * ixgbe_exit_module is called just before the driver is removed
7882 static void __exit ixgbe_exit_module(void)
7884 #ifdef CONFIG_IXGBE_DCA
7885 dca_unregister_notify(&dca_notifier);
7887 pci_unregister_driver(&ixgbe_driver);
7888 rcu_barrier(); /* Wait for completion of call_rcu()'s */
7891 #ifdef CONFIG_IXGBE_DCA
7892 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
7897 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
7898 __ixgbe_notify_dca);
7900 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7903 #endif /* CONFIG_IXGBE_DCA */
7905 module_exit(ixgbe_exit_module);