1 /*******************************************************************************
3 Intel 10 Gigabit PCI Express Linux driver
4 Copyright(c) 1999 - 2010 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";
55 #define DRV_VERSION "3.2.9-k2"
56 const char ixgbe_driver_version[] = DRV_VERSION;
57 static char ixgbe_copyright[] = "Copyright (c) 1999-2010 Intel Corporation.";
59 static const struct ixgbe_info *ixgbe_info_tbl[] = {
60 [board_82598] = &ixgbe_82598_info,
61 [board_82599] = &ixgbe_82599_info,
62 [board_X540] = &ixgbe_X540_info,
65 /* ixgbe_pci_tbl - PCI Device ID Table
67 * Wildcard entries (PCI_ANY_ID) should come last
68 * Last entry must be all 0s
70 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
71 * Class, Class Mask, private data (not used) }
73 static DEFINE_PCI_DEVICE_TABLE(ixgbe_pci_tbl) = {
74 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598),
76 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT),
78 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT),
80 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT),
82 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2),
84 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4),
86 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT),
88 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT),
90 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM),
92 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR),
94 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM),
96 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX),
98 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4),
100 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM),
102 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR),
104 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP),
106 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM),
108 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ),
110 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4),
112 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE),
114 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE),
116 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM),
118 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE),
120 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T),
123 /* required last entry */
126 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
128 #ifdef CONFIG_IXGBE_DCA
129 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
131 static struct notifier_block dca_notifier = {
132 .notifier_call = ixgbe_notify_dca,
138 #ifdef CONFIG_PCI_IOV
139 static unsigned int max_vfs;
140 module_param(max_vfs, uint, 0);
141 MODULE_PARM_DESC(max_vfs,
142 "Maximum number of virtual functions to allocate per physical function");
143 #endif /* CONFIG_PCI_IOV */
145 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
146 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
147 MODULE_LICENSE("GPL");
148 MODULE_VERSION(DRV_VERSION);
150 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
152 static inline void ixgbe_disable_sriov(struct ixgbe_adapter *adapter)
154 struct ixgbe_hw *hw = &adapter->hw;
159 #ifdef CONFIG_PCI_IOV
160 /* disable iov and allow time for transactions to clear */
161 pci_disable_sriov(adapter->pdev);
164 /* turn off device IOV mode */
165 gcr = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
166 gcr &= ~(IXGBE_GCR_EXT_SRIOV);
167 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr);
168 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
169 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
170 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
172 /* set default pool back to 0 */
173 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
174 vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
175 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
177 /* take a breather then clean up driver data */
180 kfree(adapter->vfinfo);
181 adapter->vfinfo = NULL;
183 adapter->num_vfs = 0;
184 adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
187 struct ixgbe_reg_info {
192 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
194 /* General Registers */
195 {IXGBE_CTRL, "CTRL"},
196 {IXGBE_STATUS, "STATUS"},
197 {IXGBE_CTRL_EXT, "CTRL_EXT"},
199 /* Interrupt Registers */
200 {IXGBE_EICR, "EICR"},
203 {IXGBE_SRRCTL(0), "SRRCTL"},
204 {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
205 {IXGBE_RDLEN(0), "RDLEN"},
206 {IXGBE_RDH(0), "RDH"},
207 {IXGBE_RDT(0), "RDT"},
208 {IXGBE_RXDCTL(0), "RXDCTL"},
209 {IXGBE_RDBAL(0), "RDBAL"},
210 {IXGBE_RDBAH(0), "RDBAH"},
213 {IXGBE_TDBAL(0), "TDBAL"},
214 {IXGBE_TDBAH(0), "TDBAH"},
215 {IXGBE_TDLEN(0), "TDLEN"},
216 {IXGBE_TDH(0), "TDH"},
217 {IXGBE_TDT(0), "TDT"},
218 {IXGBE_TXDCTL(0), "TXDCTL"},
220 /* List Terminator */
226 * ixgbe_regdump - register printout routine
228 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
234 switch (reginfo->ofs) {
235 case IXGBE_SRRCTL(0):
236 for (i = 0; i < 64; i++)
237 regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
239 case IXGBE_DCA_RXCTRL(0):
240 for (i = 0; i < 64; i++)
241 regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
244 for (i = 0; i < 64; i++)
245 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
248 for (i = 0; i < 64; i++)
249 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
252 for (i = 0; i < 64; i++)
253 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
255 case IXGBE_RXDCTL(0):
256 for (i = 0; i < 64; i++)
257 regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
260 for (i = 0; i < 64; i++)
261 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
264 for (i = 0; i < 64; i++)
265 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
268 for (i = 0; i < 64; i++)
269 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
272 for (i = 0; i < 64; i++)
273 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
276 for (i = 0; i < 64; i++)
277 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
280 for (i = 0; i < 64; i++)
281 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
284 for (i = 0; i < 64; i++)
285 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
287 case IXGBE_TXDCTL(0):
288 for (i = 0; i < 64; i++)
289 regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
292 pr_info("%-15s %08x\n", reginfo->name,
293 IXGBE_READ_REG(hw, reginfo->ofs));
297 for (i = 0; i < 8; i++) {
298 snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i*8, i*8+7);
299 pr_err("%-15s", rname);
300 for (j = 0; j < 8; j++)
301 pr_cont(" %08x", regs[i*8+j]);
308 * ixgbe_dump - Print registers, tx-rings and rx-rings
310 static void ixgbe_dump(struct ixgbe_adapter *adapter)
312 struct net_device *netdev = adapter->netdev;
313 struct ixgbe_hw *hw = &adapter->hw;
314 struct ixgbe_reg_info *reginfo;
316 struct ixgbe_ring *tx_ring;
317 struct ixgbe_tx_buffer *tx_buffer_info;
318 union ixgbe_adv_tx_desc *tx_desc;
319 struct my_u0 { u64 a; u64 b; } *u0;
320 struct ixgbe_ring *rx_ring;
321 union ixgbe_adv_rx_desc *rx_desc;
322 struct ixgbe_rx_buffer *rx_buffer_info;
326 if (!netif_msg_hw(adapter))
329 /* Print netdevice Info */
331 dev_info(&adapter->pdev->dev, "Net device Info\n");
332 pr_info("Device Name state "
333 "trans_start last_rx\n");
334 pr_info("%-15s %016lX %016lX %016lX\n",
341 /* Print Registers */
342 dev_info(&adapter->pdev->dev, "Register Dump\n");
343 pr_info(" Register Name Value\n");
344 for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
345 reginfo->name; reginfo++) {
346 ixgbe_regdump(hw, reginfo);
349 /* Print TX Ring Summary */
350 if (!netdev || !netif_running(netdev))
353 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
354 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
355 for (n = 0; n < adapter->num_tx_queues; n++) {
356 tx_ring = adapter->tx_ring[n];
358 &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
359 pr_info(" %5d %5X %5X %016llX %04X %3X %016llX\n",
360 n, tx_ring->next_to_use, tx_ring->next_to_clean,
361 (u64)tx_buffer_info->dma,
362 tx_buffer_info->length,
363 tx_buffer_info->next_to_watch,
364 (u64)tx_buffer_info->time_stamp);
368 if (!netif_msg_tx_done(adapter))
369 goto rx_ring_summary;
371 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
373 /* Transmit Descriptor Formats
375 * Advanced Transmit Descriptor
376 * +--------------------------------------------------------------+
377 * 0 | Buffer Address [63:0] |
378 * +--------------------------------------------------------------+
379 * 8 | PAYLEN | PORTS | IDX | STA | DCMD |DTYP | RSV | DTALEN |
380 * +--------------------------------------------------------------+
381 * 63 46 45 40 39 36 35 32 31 24 23 20 19 0
384 for (n = 0; n < adapter->num_tx_queues; n++) {
385 tx_ring = adapter->tx_ring[n];
386 pr_info("------------------------------------\n");
387 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
388 pr_info("------------------------------------\n");
389 pr_info("T [desc] [address 63:0 ] "
390 "[PlPOIdStDDt Ln] [bi->dma ] "
391 "leng ntw timestamp bi->skb\n");
393 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
394 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
395 tx_buffer_info = &tx_ring->tx_buffer_info[i];
396 u0 = (struct my_u0 *)tx_desc;
397 pr_info("T [0x%03X] %016llX %016llX %016llX"
398 " %04X %3X %016llX %p", i,
401 (u64)tx_buffer_info->dma,
402 tx_buffer_info->length,
403 tx_buffer_info->next_to_watch,
404 (u64)tx_buffer_info->time_stamp,
405 tx_buffer_info->skb);
406 if (i == tx_ring->next_to_use &&
407 i == tx_ring->next_to_clean)
409 else if (i == tx_ring->next_to_use)
411 else if (i == tx_ring->next_to_clean)
416 if (netif_msg_pktdata(adapter) &&
417 tx_buffer_info->dma != 0)
418 print_hex_dump(KERN_INFO, "",
419 DUMP_PREFIX_ADDRESS, 16, 1,
420 phys_to_virt(tx_buffer_info->dma),
421 tx_buffer_info->length, true);
425 /* Print RX Rings Summary */
427 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
428 pr_info("Queue [NTU] [NTC]\n");
429 for (n = 0; n < adapter->num_rx_queues; n++) {
430 rx_ring = adapter->rx_ring[n];
431 pr_info("%5d %5X %5X\n",
432 n, rx_ring->next_to_use, rx_ring->next_to_clean);
436 if (!netif_msg_rx_status(adapter))
439 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
441 /* Advanced Receive Descriptor (Read) Format
443 * +-----------------------------------------------------+
444 * 0 | Packet Buffer Address [63:1] |A0/NSE|
445 * +----------------------------------------------+------+
446 * 8 | Header Buffer Address [63:1] | DD |
447 * +-----------------------------------------------------+
450 * Advanced Receive Descriptor (Write-Back) Format
452 * 63 48 47 32 31 30 21 20 16 15 4 3 0
453 * +------------------------------------------------------+
454 * 0 | Packet IP |SPH| HDR_LEN | RSV|Packet| RSS |
455 * | Checksum Ident | | | | Type | Type |
456 * +------------------------------------------------------+
457 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
458 * +------------------------------------------------------+
459 * 63 48 47 32 31 20 19 0
461 for (n = 0; n < adapter->num_rx_queues; n++) {
462 rx_ring = adapter->rx_ring[n];
463 pr_info("------------------------------------\n");
464 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
465 pr_info("------------------------------------\n");
466 pr_info("R [desc] [ PktBuf A0] "
467 "[ HeadBuf DD] [bi->dma ] [bi->skb] "
468 "<-- Adv Rx Read format\n");
469 pr_info("RWB[desc] [PcsmIpSHl PtRs] "
470 "[vl er S cks ln] ---------------- [bi->skb] "
471 "<-- Adv Rx Write-Back format\n");
473 for (i = 0; i < rx_ring->count; i++) {
474 rx_buffer_info = &rx_ring->rx_buffer_info[i];
475 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
476 u0 = (struct my_u0 *)rx_desc;
477 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
478 if (staterr & IXGBE_RXD_STAT_DD) {
479 /* Descriptor Done */
480 pr_info("RWB[0x%03X] %016llX "
481 "%016llX ---------------- %p", i,
484 rx_buffer_info->skb);
486 pr_info("R [0x%03X] %016llX "
487 "%016llX %016llX %p", i,
490 (u64)rx_buffer_info->dma,
491 rx_buffer_info->skb);
493 if (netif_msg_pktdata(adapter)) {
494 print_hex_dump(KERN_INFO, "",
495 DUMP_PREFIX_ADDRESS, 16, 1,
496 phys_to_virt(rx_buffer_info->dma),
497 rx_ring->rx_buf_len, true);
499 if (rx_ring->rx_buf_len
500 < IXGBE_RXBUFFER_2048)
501 print_hex_dump(KERN_INFO, "",
502 DUMP_PREFIX_ADDRESS, 16, 1,
504 rx_buffer_info->page_dma +
505 rx_buffer_info->page_offset
511 if (i == rx_ring->next_to_use)
513 else if (i == rx_ring->next_to_clean)
525 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
529 /* Let firmware take over control of h/w */
530 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
531 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
532 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
535 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
539 /* Let firmware know the driver has taken over */
540 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
541 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
542 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
546 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
547 * @adapter: pointer to adapter struct
548 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
549 * @queue: queue to map the corresponding interrupt to
550 * @msix_vector: the vector to map to the corresponding queue
553 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
554 u8 queue, u8 msix_vector)
557 struct ixgbe_hw *hw = &adapter->hw;
558 switch (hw->mac.type) {
559 case ixgbe_mac_82598EB:
560 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
563 index = (((direction * 64) + queue) >> 2) & 0x1F;
564 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
565 ivar &= ~(0xFF << (8 * (queue & 0x3)));
566 ivar |= (msix_vector << (8 * (queue & 0x3)));
567 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
569 case ixgbe_mac_82599EB:
571 if (direction == -1) {
573 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
574 index = ((queue & 1) * 8);
575 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
576 ivar &= ~(0xFF << index);
577 ivar |= (msix_vector << index);
578 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
581 /* tx or rx causes */
582 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
583 index = ((16 * (queue & 1)) + (8 * direction));
584 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
585 ivar &= ~(0xFF << index);
586 ivar |= (msix_vector << index);
587 IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
595 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
600 switch (adapter->hw.mac.type) {
601 case ixgbe_mac_82598EB:
602 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
603 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
605 case ixgbe_mac_82599EB:
607 mask = (qmask & 0xFFFFFFFF);
608 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
609 mask = (qmask >> 32);
610 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
617 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_ring *tx_ring,
618 struct ixgbe_tx_buffer *tx_buffer_info)
620 if (tx_buffer_info->dma) {
621 if (tx_buffer_info->mapped_as_page)
622 dma_unmap_page(tx_ring->dev,
624 tx_buffer_info->length,
627 dma_unmap_single(tx_ring->dev,
629 tx_buffer_info->length,
631 tx_buffer_info->dma = 0;
633 if (tx_buffer_info->skb) {
634 dev_kfree_skb_any(tx_buffer_info->skb);
635 tx_buffer_info->skb = NULL;
637 tx_buffer_info->time_stamp = 0;
638 /* tx_buffer_info must be completely set up in the transmit path */
642 * ixgbe_dcb_txq_to_tc - convert a reg index to a traffic class
643 * @adapter: driver private struct
644 * @index: reg idx of queue to query (0-127)
646 * Helper function to determine the traffic index for a paticular
649 * Returns : a tc index for use in range 0-7, or 0-3
651 u8 ixgbe_dcb_txq_to_tc(struct ixgbe_adapter *adapter, u8 reg_idx)
654 int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
656 /* if DCB is not enabled the queues have no TC */
657 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
660 /* check valid range */
661 if (reg_idx >= adapter->hw.mac.max_tx_queues)
664 switch (adapter->hw.mac.type) {
665 case ixgbe_mac_82598EB:
669 if (dcb_i != 4 && dcb_i != 8)
672 /* if VMDq is enabled the lowest order bits determine TC */
673 if (adapter->flags & (IXGBE_FLAG_SRIOV_ENABLED |
674 IXGBE_FLAG_VMDQ_ENABLED)) {
675 tc = reg_idx & (dcb_i - 1);
680 * Convert the reg_idx into the correct TC. This bitmask
681 * targets the last full 32 ring traffic class and assigns
682 * it a value of 1. From there the rest of the rings are
683 * based on shifting the mask further up to include the
684 * reg_idx / 16 and then reg_idx / 8. It assumes dcB_i
685 * will only ever be 8 or 4 and that reg_idx will never
686 * be greater then 128. The code without the power of 2
687 * optimizations would be:
688 * (((reg_idx % 32) + 32) * dcb_i) >> (9 - reg_idx / 32)
690 tc = ((reg_idx & 0X1F) + 0x20) * dcb_i;
691 tc >>= 9 - (reg_idx >> 5);
697 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
699 struct ixgbe_hw *hw = &adapter->hw;
700 struct ixgbe_hw_stats *hwstats = &adapter->stats;
705 if ((hw->fc.current_mode == ixgbe_fc_full) ||
706 (hw->fc.current_mode == ixgbe_fc_rx_pause)) {
707 switch (hw->mac.type) {
708 case ixgbe_mac_82598EB:
709 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
712 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
714 hwstats->lxoffrxc += data;
716 /* refill credits (no tx hang) if we received xoff */
720 for (i = 0; i < adapter->num_tx_queues; i++)
721 clear_bit(__IXGBE_HANG_CHECK_ARMED,
722 &adapter->tx_ring[i]->state);
724 } else if (!(adapter->dcb_cfg.pfc_mode_enable))
727 /* update stats for each tc, only valid with PFC enabled */
728 for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
729 switch (hw->mac.type) {
730 case ixgbe_mac_82598EB:
731 xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
734 xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
736 hwstats->pxoffrxc[i] += xoff[i];
739 /* disarm tx queues that have received xoff frames */
740 for (i = 0; i < adapter->num_tx_queues; i++) {
741 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
742 u32 tc = ixgbe_dcb_txq_to_tc(adapter, tx_ring->reg_idx);
745 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
749 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
751 return ring->tx_stats.completed;
754 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
756 struct ixgbe_adapter *adapter = netdev_priv(ring->netdev);
757 struct ixgbe_hw *hw = &adapter->hw;
759 u32 head = IXGBE_READ_REG(hw, IXGBE_TDH(ring->reg_idx));
760 u32 tail = IXGBE_READ_REG(hw, IXGBE_TDT(ring->reg_idx));
763 return (head < tail) ?
764 tail - head : (tail + ring->count - head);
769 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
771 u32 tx_done = ixgbe_get_tx_completed(tx_ring);
772 u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
773 u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
776 clear_check_for_tx_hang(tx_ring);
779 * Check for a hung queue, but be thorough. This verifies
780 * that a transmit has been completed since the previous
781 * check AND there is at least one packet pending. The
782 * ARMED bit is set to indicate a potential hang. The
783 * bit is cleared if a pause frame is received to remove
784 * false hang detection due to PFC or 802.3x frames. By
785 * requiring this to fail twice we avoid races with
786 * pfc clearing the ARMED bit and conditions where we
787 * run the check_tx_hang logic with a transmit completion
788 * pending but without time to complete it yet.
790 if ((tx_done_old == tx_done) && tx_pending) {
791 /* make sure it is true for two checks in a row */
792 ret = test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
795 /* update completed stats and continue */
796 tx_ring->tx_stats.tx_done_old = tx_done;
797 /* reset the countdown */
798 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
804 #define IXGBE_MAX_TXD_PWR 14
805 #define IXGBE_MAX_DATA_PER_TXD (1 << IXGBE_MAX_TXD_PWR)
807 /* Tx Descriptors needed, worst case */
808 #define TXD_USE_COUNT(S) (((S) >> IXGBE_MAX_TXD_PWR) + \
809 (((S) & (IXGBE_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
810 #define DESC_NEEDED (TXD_USE_COUNT(IXGBE_MAX_DATA_PER_TXD) /* skb->data */ + \
811 MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1) /* for context */
813 static void ixgbe_tx_timeout(struct net_device *netdev);
816 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
817 * @q_vector: structure containing interrupt and ring information
818 * @tx_ring: tx ring to clean
820 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
821 struct ixgbe_ring *tx_ring)
823 struct ixgbe_adapter *adapter = q_vector->adapter;
824 union ixgbe_adv_tx_desc *tx_desc, *eop_desc;
825 struct ixgbe_tx_buffer *tx_buffer_info;
826 unsigned int total_bytes = 0, total_packets = 0;
827 u16 i, eop, count = 0;
829 i = tx_ring->next_to_clean;
830 eop = tx_ring->tx_buffer_info[i].next_to_watch;
831 eop_desc = IXGBE_TX_DESC_ADV(tx_ring, eop);
833 while ((eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)) &&
834 (count < tx_ring->work_limit)) {
835 bool cleaned = false;
836 rmb(); /* read buffer_info after eop_desc */
837 for ( ; !cleaned; count++) {
838 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
839 tx_buffer_info = &tx_ring->tx_buffer_info[i];
841 tx_desc->wb.status = 0;
842 cleaned = (i == eop);
845 if (i == tx_ring->count)
848 if (cleaned && tx_buffer_info->skb) {
849 total_bytes += tx_buffer_info->bytecount;
850 total_packets += tx_buffer_info->gso_segs;
853 ixgbe_unmap_and_free_tx_resource(tx_ring,
857 tx_ring->tx_stats.completed++;
858 eop = tx_ring->tx_buffer_info[i].next_to_watch;
859 eop_desc = IXGBE_TX_DESC_ADV(tx_ring, eop);
862 tx_ring->next_to_clean = i;
863 tx_ring->total_bytes += total_bytes;
864 tx_ring->total_packets += total_packets;
865 u64_stats_update_begin(&tx_ring->syncp);
866 tx_ring->stats.packets += total_packets;
867 tx_ring->stats.bytes += total_bytes;
868 u64_stats_update_end(&tx_ring->syncp);
870 if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
871 /* schedule immediate reset if we believe we hung */
872 struct ixgbe_hw *hw = &adapter->hw;
873 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, eop);
874 e_err(drv, "Detected Tx Unit Hang\n"
876 " TDH, TDT <%x>, <%x>\n"
877 " next_to_use <%x>\n"
878 " next_to_clean <%x>\n"
879 "tx_buffer_info[next_to_clean]\n"
880 " time_stamp <%lx>\n"
882 tx_ring->queue_index,
883 IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
884 IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
885 tx_ring->next_to_use, eop,
886 tx_ring->tx_buffer_info[eop].time_stamp, jiffies);
888 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
891 "tx hang %d detected on queue %d, resetting adapter\n",
892 adapter->tx_timeout_count + 1, tx_ring->queue_index);
894 /* schedule immediate reset if we believe we hung */
895 ixgbe_tx_timeout(adapter->netdev);
897 /* the adapter is about to reset, no point in enabling stuff */
901 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
902 if (unlikely(count && netif_carrier_ok(tx_ring->netdev) &&
903 (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
904 /* Make sure that anybody stopping the queue after this
905 * sees the new next_to_clean.
908 if (__netif_subqueue_stopped(tx_ring->netdev, tx_ring->queue_index) &&
909 !test_bit(__IXGBE_DOWN, &adapter->state)) {
910 netif_wake_subqueue(tx_ring->netdev, tx_ring->queue_index);
911 ++tx_ring->tx_stats.restart_queue;
915 return count < tx_ring->work_limit;
918 #ifdef CONFIG_IXGBE_DCA
919 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
920 struct ixgbe_ring *rx_ring,
923 struct ixgbe_hw *hw = &adapter->hw;
925 u8 reg_idx = rx_ring->reg_idx;
927 rxctrl = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(reg_idx));
928 switch (hw->mac.type) {
929 case ixgbe_mac_82598EB:
930 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
931 rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
933 case ixgbe_mac_82599EB:
935 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK_82599;
936 rxctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
937 IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599);
942 rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
943 rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
944 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_RRO_EN);
945 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_WRO_EN |
946 IXGBE_DCA_RXCTRL_DESC_HSRO_EN);
947 IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
950 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
951 struct ixgbe_ring *tx_ring,
954 struct ixgbe_hw *hw = &adapter->hw;
956 u8 reg_idx = tx_ring->reg_idx;
958 switch (hw->mac.type) {
959 case ixgbe_mac_82598EB:
960 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(reg_idx));
961 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
962 txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
963 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
964 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
965 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(reg_idx), txctrl);
967 case ixgbe_mac_82599EB:
969 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(reg_idx));
970 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK_82599;
971 txctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
972 IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599);
973 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
974 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
975 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(reg_idx), txctrl);
982 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
984 struct ixgbe_adapter *adapter = q_vector->adapter;
989 if (q_vector->cpu == cpu)
992 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
993 for (i = 0; i < q_vector->txr_count; i++) {
994 ixgbe_update_tx_dca(adapter, adapter->tx_ring[r_idx], cpu);
995 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
999 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1000 for (i = 0; i < q_vector->rxr_count; i++) {
1001 ixgbe_update_rx_dca(adapter, adapter->rx_ring[r_idx], cpu);
1002 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1006 q_vector->cpu = cpu;
1011 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
1016 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
1019 /* always use CB2 mode, difference is masked in the CB driver */
1020 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
1022 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
1023 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1027 for (i = 0; i < num_q_vectors; i++) {
1028 adapter->q_vector[i]->cpu = -1;
1029 ixgbe_update_dca(adapter->q_vector[i]);
1033 static int __ixgbe_notify_dca(struct device *dev, void *data)
1035 struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1036 unsigned long event = *(unsigned long *)data;
1038 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
1042 case DCA_PROVIDER_ADD:
1043 /* if we're already enabled, don't do it again */
1044 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1046 if (dca_add_requester(dev) == 0) {
1047 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1048 ixgbe_setup_dca(adapter);
1051 /* Fall Through since DCA is disabled. */
1052 case DCA_PROVIDER_REMOVE:
1053 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1054 dca_remove_requester(dev);
1055 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1056 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
1064 #endif /* CONFIG_IXGBE_DCA */
1066 * ixgbe_receive_skb - Send a completed packet up the stack
1067 * @adapter: board private structure
1068 * @skb: packet to send up
1069 * @status: hardware indication of status of receive
1070 * @rx_ring: rx descriptor ring (for a specific queue) to setup
1071 * @rx_desc: rx descriptor
1073 static void ixgbe_receive_skb(struct ixgbe_q_vector *q_vector,
1074 struct sk_buff *skb, u8 status,
1075 struct ixgbe_ring *ring,
1076 union ixgbe_adv_rx_desc *rx_desc)
1078 struct ixgbe_adapter *adapter = q_vector->adapter;
1079 struct napi_struct *napi = &q_vector->napi;
1080 bool is_vlan = (status & IXGBE_RXD_STAT_VP);
1081 u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
1083 if (is_vlan && (tag & VLAN_VID_MASK))
1084 __vlan_hwaccel_put_tag(skb, tag);
1086 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
1087 napi_gro_receive(napi, skb);
1093 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1094 * @adapter: address of board private structure
1095 * @status_err: hardware indication of status of receive
1096 * @skb: skb currently being received and modified
1098 static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
1099 union ixgbe_adv_rx_desc *rx_desc,
1100 struct sk_buff *skb)
1102 u32 status_err = le32_to_cpu(rx_desc->wb.upper.status_error);
1104 skb_checksum_none_assert(skb);
1106 /* Rx csum disabled */
1107 if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
1110 /* if IP and error */
1111 if ((status_err & IXGBE_RXD_STAT_IPCS) &&
1112 (status_err & IXGBE_RXDADV_ERR_IPE)) {
1113 adapter->hw_csum_rx_error++;
1117 if (!(status_err & IXGBE_RXD_STAT_L4CS))
1120 if (status_err & IXGBE_RXDADV_ERR_TCPE) {
1121 u16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1124 * 82599 errata, UDP frames with a 0 checksum can be marked as
1127 if ((pkt_info & IXGBE_RXDADV_PKTTYPE_UDP) &&
1128 (adapter->hw.mac.type == ixgbe_mac_82599EB))
1131 adapter->hw_csum_rx_error++;
1135 /* It must be a TCP or UDP packet with a valid checksum */
1136 skb->ip_summed = CHECKSUM_UNNECESSARY;
1139 static inline void ixgbe_release_rx_desc(struct ixgbe_ring *rx_ring, u32 val)
1142 * Force memory writes to complete before letting h/w
1143 * know there are new descriptors to fetch. (Only
1144 * applicable for weak-ordered memory model archs,
1148 writel(val, rx_ring->tail);
1152 * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
1153 * @rx_ring: ring to place buffers on
1154 * @cleaned_count: number of buffers to replace
1156 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1158 union ixgbe_adv_rx_desc *rx_desc;
1159 struct ixgbe_rx_buffer *bi;
1160 struct sk_buff *skb;
1161 u16 i = rx_ring->next_to_use;
1163 /* do nothing if no valid netdev defined */
1164 if (!rx_ring->netdev)
1167 while (cleaned_count--) {
1168 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1169 bi = &rx_ring->rx_buffer_info[i];
1173 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1174 rx_ring->rx_buf_len);
1176 rx_ring->rx_stats.alloc_rx_buff_failed++;
1179 /* initialize queue mapping */
1180 skb_record_rx_queue(skb, rx_ring->queue_index);
1185 bi->dma = dma_map_single(rx_ring->dev,
1187 rx_ring->rx_buf_len,
1189 if (dma_mapping_error(rx_ring->dev, bi->dma)) {
1190 rx_ring->rx_stats.alloc_rx_buff_failed++;
1196 if (ring_is_ps_enabled(rx_ring)) {
1198 bi->page = netdev_alloc_page(rx_ring->netdev);
1200 rx_ring->rx_stats.alloc_rx_page_failed++;
1205 if (!bi->page_dma) {
1206 /* use a half page if we're re-using */
1207 bi->page_offset ^= PAGE_SIZE / 2;
1208 bi->page_dma = dma_map_page(rx_ring->dev,
1213 if (dma_mapping_error(rx_ring->dev,
1215 rx_ring->rx_stats.alloc_rx_page_failed++;
1221 /* Refresh the desc even if buffer_addrs didn't change
1222 * because each write-back erases this info. */
1223 rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
1224 rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
1226 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
1227 rx_desc->read.hdr_addr = 0;
1231 if (i == rx_ring->count)
1236 if (rx_ring->next_to_use != i) {
1237 rx_ring->next_to_use = i;
1238 ixgbe_release_rx_desc(rx_ring, i);
1242 static inline u16 ixgbe_get_hlen(union ixgbe_adv_rx_desc *rx_desc)
1244 /* HW will not DMA in data larger than the given buffer, even if it
1245 * parses the (NFS, of course) header to be larger. In that case, it
1246 * fills the header buffer and spills the rest into the page.
1248 u16 hdr_info = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info);
1249 u16 hlen = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
1250 IXGBE_RXDADV_HDRBUFLEN_SHIFT;
1251 if (hlen > IXGBE_RX_HDR_SIZE)
1252 hlen = IXGBE_RX_HDR_SIZE;
1257 * ixgbe_transform_rsc_queue - change rsc queue into a full packet
1258 * @skb: pointer to the last skb in the rsc queue
1260 * This function changes a queue full of hw rsc buffers into a completed
1261 * packet. It uses the ->prev pointers to find the first packet and then
1262 * turns it into the frag list owner.
1264 static inline struct sk_buff *ixgbe_transform_rsc_queue(struct sk_buff *skb)
1266 unsigned int frag_list_size = 0;
1267 unsigned int skb_cnt = 1;
1270 struct sk_buff *prev = skb->prev;
1271 frag_list_size += skb->len;
1277 skb_shinfo(skb)->frag_list = skb->next;
1279 skb->len += frag_list_size;
1280 skb->data_len += frag_list_size;
1281 skb->truesize += frag_list_size;
1282 IXGBE_RSC_CB(skb)->skb_cnt = skb_cnt;
1287 static inline bool ixgbe_get_rsc_state(union ixgbe_adv_rx_desc *rx_desc)
1289 return !!(le32_to_cpu(rx_desc->wb.lower.lo_dword.data) &
1290 IXGBE_RXDADV_RSCCNT_MASK);
1293 static void ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1294 struct ixgbe_ring *rx_ring,
1295 int *work_done, int work_to_do)
1297 struct ixgbe_adapter *adapter = q_vector->adapter;
1298 union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
1299 struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
1300 struct sk_buff *skb;
1301 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1302 const int current_node = numa_node_id();
1305 #endif /* IXGBE_FCOE */
1308 u16 cleaned_count = 0;
1309 bool pkt_is_rsc = false;
1311 i = rx_ring->next_to_clean;
1312 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1313 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1315 while (staterr & IXGBE_RXD_STAT_DD) {
1318 rmb(); /* read descriptor and rx_buffer_info after status DD */
1320 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1322 skb = rx_buffer_info->skb;
1323 rx_buffer_info->skb = NULL;
1324 prefetch(skb->data);
1326 if (ring_is_rsc_enabled(rx_ring))
1327 pkt_is_rsc = ixgbe_get_rsc_state(rx_desc);
1329 /* if this is a skb from previous receive DMA will be 0 */
1330 if (rx_buffer_info->dma) {
1333 !(staterr & IXGBE_RXD_STAT_EOP) &&
1336 * When HWRSC is enabled, delay unmapping
1337 * of the first packet. It carries the
1338 * header information, HW may still
1339 * access the header after the writeback.
1340 * Only unmap it when EOP is reached
1342 IXGBE_RSC_CB(skb)->delay_unmap = true;
1343 IXGBE_RSC_CB(skb)->dma = rx_buffer_info->dma;
1345 dma_unmap_single(rx_ring->dev,
1346 rx_buffer_info->dma,
1347 rx_ring->rx_buf_len,
1350 rx_buffer_info->dma = 0;
1352 if (ring_is_ps_enabled(rx_ring)) {
1353 hlen = ixgbe_get_hlen(rx_desc);
1354 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1356 hlen = le16_to_cpu(rx_desc->wb.upper.length);
1361 /* assume packet split since header is unmapped */
1362 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1366 dma_unmap_page(rx_ring->dev,
1367 rx_buffer_info->page_dma,
1370 rx_buffer_info->page_dma = 0;
1371 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1372 rx_buffer_info->page,
1373 rx_buffer_info->page_offset,
1376 if ((page_count(rx_buffer_info->page) == 1) &&
1377 (page_to_nid(rx_buffer_info->page) == current_node))
1378 get_page(rx_buffer_info->page);
1380 rx_buffer_info->page = NULL;
1382 skb->len += upper_len;
1383 skb->data_len += upper_len;
1384 skb->truesize += upper_len;
1388 if (i == rx_ring->count)
1391 next_rxd = IXGBE_RX_DESC_ADV(rx_ring, i);
1396 u32 nextp = (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
1397 IXGBE_RXDADV_NEXTP_SHIFT;
1398 next_buffer = &rx_ring->rx_buffer_info[nextp];
1400 next_buffer = &rx_ring->rx_buffer_info[i];
1403 if (!(staterr & IXGBE_RXD_STAT_EOP)) {
1404 if (ring_is_ps_enabled(rx_ring)) {
1405 rx_buffer_info->skb = next_buffer->skb;
1406 rx_buffer_info->dma = next_buffer->dma;
1407 next_buffer->skb = skb;
1408 next_buffer->dma = 0;
1410 skb->next = next_buffer->skb;
1411 skb->next->prev = skb;
1413 rx_ring->rx_stats.non_eop_descs++;
1418 skb = ixgbe_transform_rsc_queue(skb);
1419 /* if we got here without RSC the packet is invalid */
1421 __pskb_trim(skb, 0);
1422 rx_buffer_info->skb = skb;
1427 if (ring_is_rsc_enabled(rx_ring)) {
1428 if (IXGBE_RSC_CB(skb)->delay_unmap) {
1429 dma_unmap_single(rx_ring->dev,
1430 IXGBE_RSC_CB(skb)->dma,
1431 rx_ring->rx_buf_len,
1433 IXGBE_RSC_CB(skb)->dma = 0;
1434 IXGBE_RSC_CB(skb)->delay_unmap = false;
1438 if (ring_is_ps_enabled(rx_ring))
1439 rx_ring->rx_stats.rsc_count +=
1440 skb_shinfo(skb)->nr_frags;
1442 rx_ring->rx_stats.rsc_count +=
1443 IXGBE_RSC_CB(skb)->skb_cnt;
1444 rx_ring->rx_stats.rsc_flush++;
1447 /* ERR_MASK will only have valid bits if EOP set */
1448 if (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) {
1449 /* trim packet back to size 0 and recycle it */
1450 __pskb_trim(skb, 0);
1451 rx_buffer_info->skb = skb;
1455 ixgbe_rx_checksum(adapter, rx_desc, skb);
1457 /* probably a little skewed due to removing CRC */
1458 total_rx_bytes += skb->len;
1461 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1463 /* if ddp, not passing to ULD unless for FCP_RSP or error */
1464 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
1465 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
1469 #endif /* IXGBE_FCOE */
1470 ixgbe_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
1473 rx_desc->wb.upper.status_error = 0;
1476 if (*work_done >= work_to_do)
1479 /* return some buffers to hardware, one at a time is too slow */
1480 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
1481 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1485 /* use prefetched values */
1487 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1490 rx_ring->next_to_clean = i;
1491 cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
1494 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1497 /* include DDPed FCoE data */
1498 if (ddp_bytes > 0) {
1501 mss = rx_ring->netdev->mtu - sizeof(struct fcoe_hdr) -
1502 sizeof(struct fc_frame_header) -
1503 sizeof(struct fcoe_crc_eof);
1506 total_rx_bytes += ddp_bytes;
1507 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
1509 #endif /* IXGBE_FCOE */
1511 rx_ring->total_packets += total_rx_packets;
1512 rx_ring->total_bytes += total_rx_bytes;
1513 u64_stats_update_begin(&rx_ring->syncp);
1514 rx_ring->stats.packets += total_rx_packets;
1515 rx_ring->stats.bytes += total_rx_bytes;
1516 u64_stats_update_end(&rx_ring->syncp);
1519 static int ixgbe_clean_rxonly(struct napi_struct *, int);
1521 * ixgbe_configure_msix - Configure MSI-X hardware
1522 * @adapter: board private structure
1524 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1527 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1529 struct ixgbe_q_vector *q_vector;
1530 int i, q_vectors, v_idx, r_idx;
1533 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1536 * Populate the IVAR table and set the ITR values to the
1537 * corresponding register.
1539 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1540 q_vector = adapter->q_vector[v_idx];
1541 /* XXX for_each_set_bit(...) */
1542 r_idx = find_first_bit(q_vector->rxr_idx,
1543 adapter->num_rx_queues);
1545 for (i = 0; i < q_vector->rxr_count; i++) {
1546 u8 reg_idx = adapter->rx_ring[r_idx]->reg_idx;
1547 ixgbe_set_ivar(adapter, 0, reg_idx, v_idx);
1548 r_idx = find_next_bit(q_vector->rxr_idx,
1549 adapter->num_rx_queues,
1552 r_idx = find_first_bit(q_vector->txr_idx,
1553 adapter->num_tx_queues);
1555 for (i = 0; i < q_vector->txr_count; i++) {
1556 u8 reg_idx = adapter->tx_ring[r_idx]->reg_idx;
1557 ixgbe_set_ivar(adapter, 1, reg_idx, v_idx);
1558 r_idx = find_next_bit(q_vector->txr_idx,
1559 adapter->num_tx_queues,
1563 if (q_vector->txr_count && !q_vector->rxr_count)
1565 q_vector->eitr = adapter->tx_eitr_param;
1566 else if (q_vector->rxr_count)
1568 q_vector->eitr = adapter->rx_eitr_param;
1570 ixgbe_write_eitr(q_vector);
1571 /* If Flow Director is enabled, set interrupt affinity */
1572 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
1573 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)) {
1575 * Allocate the affinity_hint cpumask, assign the mask
1576 * for this vector, and set our affinity_hint for
1579 if (!alloc_cpumask_var(&q_vector->affinity_mask,
1582 cpumask_set_cpu(v_idx, q_vector->affinity_mask);
1583 irq_set_affinity_hint(adapter->msix_entries[v_idx].vector,
1584 q_vector->affinity_mask);
1588 switch (adapter->hw.mac.type) {
1589 case ixgbe_mac_82598EB:
1590 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1593 case ixgbe_mac_82599EB:
1594 case ixgbe_mac_X540:
1595 ixgbe_set_ivar(adapter, -1, 1, v_idx);
1601 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1603 /* set up to autoclear timer, and the vectors */
1604 mask = IXGBE_EIMS_ENABLE_MASK;
1605 if (adapter->num_vfs)
1606 mask &= ~(IXGBE_EIMS_OTHER |
1607 IXGBE_EIMS_MAILBOX |
1610 mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
1611 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1614 enum latency_range {
1618 latency_invalid = 255
1622 * ixgbe_update_itr - update the dynamic ITR value based on statistics
1623 * @adapter: pointer to adapter
1624 * @eitr: eitr setting (ints per sec) to give last timeslice
1625 * @itr_setting: current throttle rate in ints/second
1626 * @packets: the number of packets during this measurement interval
1627 * @bytes: the number of bytes during this measurement interval
1629 * Stores a new ITR value based on packets and byte
1630 * counts during the last interrupt. The advantage of per interrupt
1631 * computation is faster updates and more accurate ITR for the current
1632 * traffic pattern. Constants in this function were computed
1633 * based on theoretical maximum wire speed and thresholds were set based
1634 * on testing data as well as attempting to minimize response time
1635 * while increasing bulk throughput.
1636 * this functionality is controlled by the InterruptThrottleRate module
1637 * parameter (see ixgbe_param.c)
1639 static u8 ixgbe_update_itr(struct ixgbe_adapter *adapter,
1640 u32 eitr, u8 itr_setting,
1641 int packets, int bytes)
1643 unsigned int retval = itr_setting;
1648 goto update_itr_done;
1651 /* simple throttlerate management
1652 * 0-20MB/s lowest (100000 ints/s)
1653 * 20-100MB/s low (20000 ints/s)
1654 * 100-1249MB/s bulk (8000 ints/s)
1656 /* what was last interrupt timeslice? */
1657 timepassed_us = 1000000/eitr;
1658 bytes_perint = bytes / timepassed_us; /* bytes/usec */
1660 switch (itr_setting) {
1661 case lowest_latency:
1662 if (bytes_perint > adapter->eitr_low)
1663 retval = low_latency;
1666 if (bytes_perint > adapter->eitr_high)
1667 retval = bulk_latency;
1668 else if (bytes_perint <= adapter->eitr_low)
1669 retval = lowest_latency;
1672 if (bytes_perint <= adapter->eitr_high)
1673 retval = low_latency;
1682 * ixgbe_write_eitr - write EITR register in hardware specific way
1683 * @q_vector: structure containing interrupt and ring information
1685 * This function is made to be called by ethtool and by the driver
1686 * when it needs to update EITR registers at runtime. Hardware
1687 * specific quirks/differences are taken care of here.
1689 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1691 struct ixgbe_adapter *adapter = q_vector->adapter;
1692 struct ixgbe_hw *hw = &adapter->hw;
1693 int v_idx = q_vector->v_idx;
1694 u32 itr_reg = EITR_INTS_PER_SEC_TO_REG(q_vector->eitr);
1696 switch (adapter->hw.mac.type) {
1697 case ixgbe_mac_82598EB:
1698 /* must write high and low 16 bits to reset counter */
1699 itr_reg |= (itr_reg << 16);
1701 case ixgbe_mac_82599EB:
1702 case ixgbe_mac_X540:
1704 * 82599 and X540 can support a value of zero, so allow it for
1705 * max interrupt rate, but there is an errata where it can
1706 * not be zero with RSC
1709 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
1713 * set the WDIS bit to not clear the timer bits and cause an
1714 * immediate assertion of the interrupt
1716 itr_reg |= IXGBE_EITR_CNT_WDIS;
1721 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1724 static void ixgbe_set_itr_msix(struct ixgbe_q_vector *q_vector)
1726 struct ixgbe_adapter *adapter = q_vector->adapter;
1729 u8 current_itr, ret_itr;
1731 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1732 for (i = 0; i < q_vector->txr_count; i++) {
1733 struct ixgbe_ring *tx_ring = adapter->tx_ring[r_idx];
1734 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1736 tx_ring->total_packets,
1737 tx_ring->total_bytes);
1738 /* if the result for this queue would decrease interrupt
1739 * rate for this vector then use that result */
1740 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
1741 q_vector->tx_itr - 1 : ret_itr);
1742 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1746 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1747 for (i = 0; i < q_vector->rxr_count; i++) {
1748 struct ixgbe_ring *rx_ring = adapter->rx_ring[r_idx];
1749 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1751 rx_ring->total_packets,
1752 rx_ring->total_bytes);
1753 /* if the result for this queue would decrease interrupt
1754 * rate for this vector then use that result */
1755 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
1756 q_vector->rx_itr - 1 : ret_itr);
1757 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1761 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1763 switch (current_itr) {
1764 /* counts and packets in update_itr are dependent on these numbers */
1765 case lowest_latency:
1769 new_itr = 20000; /* aka hwitr = ~200 */
1777 if (new_itr != q_vector->eitr) {
1778 /* do an exponential smoothing */
1779 new_itr = ((q_vector->eitr * 9) + new_itr)/10;
1781 /* save the algorithm value here, not the smoothed one */
1782 q_vector->eitr = new_itr;
1784 ixgbe_write_eitr(q_vector);
1789 * ixgbe_check_overtemp_task - worker thread to check over tempurature
1790 * @work: pointer to work_struct containing our data
1792 static void ixgbe_check_overtemp_task(struct work_struct *work)
1794 struct ixgbe_adapter *adapter = container_of(work,
1795 struct ixgbe_adapter,
1796 check_overtemp_task);
1797 struct ixgbe_hw *hw = &adapter->hw;
1798 u32 eicr = adapter->interrupt_event;
1800 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
1803 switch (hw->device_id) {
1804 case IXGBE_DEV_ID_82599_T3_LOM: {
1806 bool link_up = false;
1808 if (hw->mac.ops.check_link)
1809 hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
1811 if (((eicr & IXGBE_EICR_GPI_SDP0) && (!link_up)) ||
1812 (eicr & IXGBE_EICR_LSC))
1813 /* Check if this is due to overtemp */
1814 if (hw->phy.ops.check_overtemp(hw) == IXGBE_ERR_OVERTEMP)
1819 if (!(eicr & IXGBE_EICR_GPI_SDP0))
1824 "Network adapter has been stopped because it has over heated. "
1825 "Restart the computer. If the problem persists, "
1826 "power off the system and replace the adapter\n");
1827 /* write to clear the interrupt */
1828 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP0);
1831 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
1833 struct ixgbe_hw *hw = &adapter->hw;
1835 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
1836 (eicr & IXGBE_EICR_GPI_SDP1)) {
1837 e_crit(probe, "Fan has stopped, replace the adapter\n");
1838 /* write to clear the interrupt */
1839 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1843 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
1845 struct ixgbe_hw *hw = &adapter->hw;
1847 if (eicr & IXGBE_EICR_GPI_SDP2) {
1848 /* Clear the interrupt */
1849 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
1850 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1851 schedule_work(&adapter->sfp_config_module_task);
1854 if (eicr & IXGBE_EICR_GPI_SDP1) {
1855 /* Clear the interrupt */
1856 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1857 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1858 schedule_work(&adapter->multispeed_fiber_task);
1862 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1864 struct ixgbe_hw *hw = &adapter->hw;
1867 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1868 adapter->link_check_timeout = jiffies;
1869 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1870 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1871 IXGBE_WRITE_FLUSH(hw);
1872 schedule_work(&adapter->watchdog_task);
1876 static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
1878 struct net_device *netdev = data;
1879 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1880 struct ixgbe_hw *hw = &adapter->hw;
1884 * Workaround for Silicon errata. Use clear-by-write instead
1885 * of clear-by-read. Reading with EICS will return the
1886 * interrupt causes without clearing, which later be done
1887 * with the write to EICR.
1889 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1890 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1892 if (eicr & IXGBE_EICR_LSC)
1893 ixgbe_check_lsc(adapter);
1895 if (eicr & IXGBE_EICR_MAILBOX)
1896 ixgbe_msg_task(adapter);
1898 switch (hw->mac.type) {
1899 case ixgbe_mac_82599EB:
1900 ixgbe_check_sfp_event(adapter, eicr);
1901 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
1902 ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC))) {
1903 adapter->interrupt_event = eicr;
1904 schedule_work(&adapter->check_overtemp_task);
1906 /* now fallthrough to handle Flow Director */
1907 case ixgbe_mac_X540:
1908 /* Handle Flow Director Full threshold interrupt */
1909 if (eicr & IXGBE_EICR_FLOW_DIR) {
1911 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_FLOW_DIR);
1912 /* Disable transmits before FDIR Re-initialization */
1913 netif_tx_stop_all_queues(netdev);
1914 for (i = 0; i < adapter->num_tx_queues; i++) {
1915 struct ixgbe_ring *tx_ring =
1916 adapter->tx_ring[i];
1917 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
1919 schedule_work(&adapter->fdir_reinit_task);
1927 ixgbe_check_fan_failure(adapter, eicr);
1929 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1930 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_OTHER);
1935 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
1939 struct ixgbe_hw *hw = &adapter->hw;
1941 switch (hw->mac.type) {
1942 case ixgbe_mac_82598EB:
1943 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1944 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
1946 case ixgbe_mac_82599EB:
1947 case ixgbe_mac_X540:
1948 mask = (qmask & 0xFFFFFFFF);
1950 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
1951 mask = (qmask >> 32);
1953 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
1958 /* skip the flush */
1961 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
1965 struct ixgbe_hw *hw = &adapter->hw;
1967 switch (hw->mac.type) {
1968 case ixgbe_mac_82598EB:
1969 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1970 IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
1972 case ixgbe_mac_82599EB:
1973 case ixgbe_mac_X540:
1974 mask = (qmask & 0xFFFFFFFF);
1976 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
1977 mask = (qmask >> 32);
1979 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
1984 /* skip the flush */
1987 static irqreturn_t ixgbe_msix_clean_tx(int irq, void *data)
1989 struct ixgbe_q_vector *q_vector = data;
1990 struct ixgbe_adapter *adapter = q_vector->adapter;
1991 struct ixgbe_ring *tx_ring;
1994 if (!q_vector->txr_count)
1997 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1998 for (i = 0; i < q_vector->txr_count; i++) {
1999 tx_ring = adapter->tx_ring[r_idx];
2000 tx_ring->total_bytes = 0;
2001 tx_ring->total_packets = 0;
2002 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
2006 /* EIAM disabled interrupts (on this vector) for us */
2007 napi_schedule(&q_vector->napi);
2013 * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
2015 * @data: pointer to our q_vector struct for this interrupt vector
2017 static irqreturn_t ixgbe_msix_clean_rx(int irq, void *data)
2019 struct ixgbe_q_vector *q_vector = data;
2020 struct ixgbe_adapter *adapter = q_vector->adapter;
2021 struct ixgbe_ring *rx_ring;
2025 #ifdef CONFIG_IXGBE_DCA
2026 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2027 ixgbe_update_dca(q_vector);
2030 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2031 for (i = 0; i < q_vector->rxr_count; i++) {
2032 rx_ring = adapter->rx_ring[r_idx];
2033 rx_ring->total_bytes = 0;
2034 rx_ring->total_packets = 0;
2035 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
2039 if (!q_vector->rxr_count)
2042 /* EIAM disabled interrupts (on this vector) for us */
2043 napi_schedule(&q_vector->napi);
2048 static irqreturn_t ixgbe_msix_clean_many(int irq, void *data)
2050 struct ixgbe_q_vector *q_vector = data;
2051 struct ixgbe_adapter *adapter = q_vector->adapter;
2052 struct ixgbe_ring *ring;
2056 if (!q_vector->txr_count && !q_vector->rxr_count)
2059 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
2060 for (i = 0; i < q_vector->txr_count; i++) {
2061 ring = adapter->tx_ring[r_idx];
2062 ring->total_bytes = 0;
2063 ring->total_packets = 0;
2064 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
2068 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2069 for (i = 0; i < q_vector->rxr_count; i++) {
2070 ring = adapter->rx_ring[r_idx];
2071 ring->total_bytes = 0;
2072 ring->total_packets = 0;
2073 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
2077 /* EIAM disabled interrupts (on this vector) for us */
2078 napi_schedule(&q_vector->napi);
2084 * ixgbe_clean_rxonly - msix (aka one shot) rx clean routine
2085 * @napi: napi struct with our devices info in it
2086 * @budget: amount of work driver is allowed to do this pass, in packets
2088 * This function is optimized for cleaning one queue only on a single
2091 static int ixgbe_clean_rxonly(struct napi_struct *napi, int budget)
2093 struct ixgbe_q_vector *q_vector =
2094 container_of(napi, struct ixgbe_q_vector, napi);
2095 struct ixgbe_adapter *adapter = q_vector->adapter;
2096 struct ixgbe_ring *rx_ring = NULL;
2100 #ifdef CONFIG_IXGBE_DCA
2101 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2102 ixgbe_update_dca(q_vector);
2105 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2106 rx_ring = adapter->rx_ring[r_idx];
2108 ixgbe_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
2110 /* If all Rx work done, exit the polling mode */
2111 if (work_done < budget) {
2112 napi_complete(napi);
2113 if (adapter->rx_itr_setting & 1)
2114 ixgbe_set_itr_msix(q_vector);
2115 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2116 ixgbe_irq_enable_queues(adapter,
2117 ((u64)1 << q_vector->v_idx));
2124 * ixgbe_clean_rxtx_many - msix (aka one shot) rx clean routine
2125 * @napi: napi struct with our devices info in it
2126 * @budget: amount of work driver is allowed to do this pass, in packets
2128 * This function will clean more than one rx queue associated with a
2131 static int ixgbe_clean_rxtx_many(struct napi_struct *napi, int budget)
2133 struct ixgbe_q_vector *q_vector =
2134 container_of(napi, struct ixgbe_q_vector, napi);
2135 struct ixgbe_adapter *adapter = q_vector->adapter;
2136 struct ixgbe_ring *ring = NULL;
2137 int work_done = 0, i;
2139 bool tx_clean_complete = true;
2141 #ifdef CONFIG_IXGBE_DCA
2142 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2143 ixgbe_update_dca(q_vector);
2146 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
2147 for (i = 0; i < q_vector->txr_count; i++) {
2148 ring = adapter->tx_ring[r_idx];
2149 tx_clean_complete &= ixgbe_clean_tx_irq(q_vector, ring);
2150 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
2154 /* attempt to distribute budget to each queue fairly, but don't allow
2155 * the budget to go below 1 because we'll exit polling */
2156 budget /= (q_vector->rxr_count ?: 1);
2157 budget = max(budget, 1);
2158 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2159 for (i = 0; i < q_vector->rxr_count; i++) {
2160 ring = adapter->rx_ring[r_idx];
2161 ixgbe_clean_rx_irq(q_vector, ring, &work_done, budget);
2162 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
2166 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2167 ring = adapter->rx_ring[r_idx];
2168 /* If all Rx work done, exit the polling mode */
2169 if (work_done < budget) {
2170 napi_complete(napi);
2171 if (adapter->rx_itr_setting & 1)
2172 ixgbe_set_itr_msix(q_vector);
2173 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2174 ixgbe_irq_enable_queues(adapter,
2175 ((u64)1 << q_vector->v_idx));
2183 * ixgbe_clean_txonly - msix (aka one shot) tx clean routine
2184 * @napi: napi struct with our devices info in it
2185 * @budget: amount of work driver is allowed to do this pass, in packets
2187 * This function is optimized for cleaning one queue only on a single
2190 static int ixgbe_clean_txonly(struct napi_struct *napi, int budget)
2192 struct ixgbe_q_vector *q_vector =
2193 container_of(napi, struct ixgbe_q_vector, napi);
2194 struct ixgbe_adapter *adapter = q_vector->adapter;
2195 struct ixgbe_ring *tx_ring = NULL;
2199 #ifdef CONFIG_IXGBE_DCA
2200 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2201 ixgbe_update_dca(q_vector);
2204 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
2205 tx_ring = adapter->tx_ring[r_idx];
2207 if (!ixgbe_clean_tx_irq(q_vector, tx_ring))
2210 /* If all Tx work done, exit the polling mode */
2211 if (work_done < budget) {
2212 napi_complete(napi);
2213 if (adapter->tx_itr_setting & 1)
2214 ixgbe_set_itr_msix(q_vector);
2215 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2216 ixgbe_irq_enable_queues(adapter,
2217 ((u64)1 << q_vector->v_idx));
2223 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
2226 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2227 struct ixgbe_ring *rx_ring = a->rx_ring[r_idx];
2229 set_bit(r_idx, q_vector->rxr_idx);
2230 q_vector->rxr_count++;
2231 rx_ring->q_vector = q_vector;
2234 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
2237 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2238 struct ixgbe_ring *tx_ring = a->tx_ring[t_idx];
2240 set_bit(t_idx, q_vector->txr_idx);
2241 q_vector->txr_count++;
2242 tx_ring->q_vector = q_vector;
2246 * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
2247 * @adapter: board private structure to initialize
2249 * This function maps descriptor rings to the queue-specific vectors
2250 * we were allotted through the MSI-X enabling code. Ideally, we'd have
2251 * one vector per ring/queue, but on a constrained vector budget, we
2252 * group the rings as "efficiently" as possible. You would add new
2253 * mapping configurations in here.
2255 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter)
2259 int rxr_idx = 0, txr_idx = 0;
2260 int rxr_remaining = adapter->num_rx_queues;
2261 int txr_remaining = adapter->num_tx_queues;
2266 /* No mapping required if MSI-X is disabled. */
2267 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2270 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2273 * The ideal configuration...
2274 * We have enough vectors to map one per queue.
2276 if (q_vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
2277 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
2278 map_vector_to_rxq(adapter, v_start, rxr_idx);
2280 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
2281 map_vector_to_txq(adapter, v_start, txr_idx);
2287 * If we don't have enough vectors for a 1-to-1
2288 * mapping, we'll have to group them so there are
2289 * multiple queues per vector.
2291 /* Re-adjusting *qpv takes care of the remainder. */
2292 for (i = v_start; i < q_vectors; i++) {
2293 rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - i);
2294 for (j = 0; j < rqpv; j++) {
2295 map_vector_to_rxq(adapter, i, rxr_idx);
2299 tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - i);
2300 for (j = 0; j < tqpv; j++) {
2301 map_vector_to_txq(adapter, i, txr_idx);
2311 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2312 * @adapter: board private structure
2314 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2315 * interrupts from the kernel.
2317 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2319 struct net_device *netdev = adapter->netdev;
2320 irqreturn_t (*handler)(int, void *);
2321 int i, vector, q_vectors, err;
2324 /* Decrement for Other and TCP Timer vectors */
2325 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2327 err = ixgbe_map_rings_to_vectors(adapter);
2331 #define SET_HANDLER(_v) (((_v)->rxr_count && (_v)->txr_count) \
2332 ? &ixgbe_msix_clean_many : \
2333 (_v)->rxr_count ? &ixgbe_msix_clean_rx : \
2334 (_v)->txr_count ? &ixgbe_msix_clean_tx : \
2336 for (vector = 0; vector < q_vectors; vector++) {
2337 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2338 handler = SET_HANDLER(q_vector);
2340 if (handler == &ixgbe_msix_clean_rx) {
2341 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2342 "%s-%s-%d", netdev->name, "rx", ri++);
2343 } else if (handler == &ixgbe_msix_clean_tx) {
2344 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2345 "%s-%s-%d", netdev->name, "tx", ti++);
2346 } else if (handler == &ixgbe_msix_clean_many) {
2347 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2348 "%s-%s-%d", netdev->name, "TxRx", ri++);
2351 /* skip this unused q_vector */
2354 err = request_irq(adapter->msix_entries[vector].vector,
2355 handler, 0, q_vector->name,
2358 e_err(probe, "request_irq failed for MSIX interrupt "
2359 "Error: %d\n", err);
2360 goto free_queue_irqs;
2364 sprintf(adapter->lsc_int_name, "%s:lsc", netdev->name);
2365 err = request_irq(adapter->msix_entries[vector].vector,
2366 ixgbe_msix_lsc, 0, adapter->lsc_int_name, netdev);
2368 e_err(probe, "request_irq for msix_lsc failed: %d\n", err);
2369 goto free_queue_irqs;
2375 for (i = vector - 1; i >= 0; i--)
2376 free_irq(adapter->msix_entries[--vector].vector,
2377 adapter->q_vector[i]);
2378 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2379 pci_disable_msix(adapter->pdev);
2380 kfree(adapter->msix_entries);
2381 adapter->msix_entries = NULL;
2385 static void ixgbe_set_itr(struct ixgbe_adapter *adapter)
2387 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2388 struct ixgbe_ring *rx_ring = adapter->rx_ring[0];
2389 struct ixgbe_ring *tx_ring = adapter->tx_ring[0];
2390 u32 new_itr = q_vector->eitr;
2393 q_vector->tx_itr = ixgbe_update_itr(adapter, new_itr,
2395 tx_ring->total_packets,
2396 tx_ring->total_bytes);
2397 q_vector->rx_itr = ixgbe_update_itr(adapter, new_itr,
2399 rx_ring->total_packets,
2400 rx_ring->total_bytes);
2402 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
2404 switch (current_itr) {
2405 /* counts and packets in update_itr are dependent on these numbers */
2406 case lowest_latency:
2410 new_itr = 20000; /* aka hwitr = ~200 */
2419 if (new_itr != q_vector->eitr) {
2420 /* do an exponential smoothing */
2421 new_itr = ((q_vector->eitr * 9) + new_itr)/10;
2423 /* save the algorithm value here */
2424 q_vector->eitr = new_itr;
2426 ixgbe_write_eitr(q_vector);
2431 * ixgbe_irq_enable - Enable default interrupt generation settings
2432 * @adapter: board private structure
2434 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2439 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2440 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2441 mask |= IXGBE_EIMS_GPI_SDP0;
2442 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2443 mask |= IXGBE_EIMS_GPI_SDP1;
2444 switch (adapter->hw.mac.type) {
2445 case ixgbe_mac_82599EB:
2446 case ixgbe_mac_X540:
2447 mask |= IXGBE_EIMS_ECC;
2448 mask |= IXGBE_EIMS_GPI_SDP1;
2449 mask |= IXGBE_EIMS_GPI_SDP2;
2450 if (adapter->num_vfs)
2451 mask |= IXGBE_EIMS_MAILBOX;
2456 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
2457 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
2458 mask |= IXGBE_EIMS_FLOW_DIR;
2460 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2462 ixgbe_irq_enable_queues(adapter, ~0);
2464 IXGBE_WRITE_FLUSH(&adapter->hw);
2466 if (adapter->num_vfs > 32) {
2467 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
2468 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2473 * ixgbe_intr - legacy mode Interrupt Handler
2474 * @irq: interrupt number
2475 * @data: pointer to a network interface device structure
2477 static irqreturn_t ixgbe_intr(int irq, void *data)
2479 struct net_device *netdev = data;
2480 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2481 struct ixgbe_hw *hw = &adapter->hw;
2482 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2486 * Workaround for silicon errata on 82598. Mask the interrupts
2487 * before the read of EICR.
2489 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2491 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2492 * therefore no explict interrupt disable is necessary */
2493 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2496 * shared interrupt alert!
2497 * make sure interrupts are enabled because the read will
2498 * have disabled interrupts due to EIAM
2499 * finish the workaround of silicon errata on 82598. Unmask
2500 * the interrupt that we masked before the EICR read.
2502 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2503 ixgbe_irq_enable(adapter, true, true);
2504 return IRQ_NONE; /* Not our interrupt */
2507 if (eicr & IXGBE_EICR_LSC)
2508 ixgbe_check_lsc(adapter);
2510 switch (hw->mac.type) {
2511 case ixgbe_mac_82599EB:
2512 ixgbe_check_sfp_event(adapter, eicr);
2513 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2514 ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC))) {
2515 adapter->interrupt_event = eicr;
2516 schedule_work(&adapter->check_overtemp_task);
2523 ixgbe_check_fan_failure(adapter, eicr);
2525 if (napi_schedule_prep(&(q_vector->napi))) {
2526 adapter->tx_ring[0]->total_packets = 0;
2527 adapter->tx_ring[0]->total_bytes = 0;
2528 adapter->rx_ring[0]->total_packets = 0;
2529 adapter->rx_ring[0]->total_bytes = 0;
2530 /* would disable interrupts here but EIAM disabled it */
2531 __napi_schedule(&(q_vector->napi));
2535 * re-enable link(maybe) and non-queue interrupts, no flush.
2536 * ixgbe_poll will re-enable the queue interrupts
2539 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2540 ixgbe_irq_enable(adapter, false, false);
2545 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
2547 int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2549 for (i = 0; i < q_vectors; i++) {
2550 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
2551 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
2552 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
2553 q_vector->rxr_count = 0;
2554 q_vector->txr_count = 0;
2559 * ixgbe_request_irq - initialize interrupts
2560 * @adapter: board private structure
2562 * Attempts to configure interrupts using the best available
2563 * capabilities of the hardware and kernel.
2565 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2567 struct net_device *netdev = adapter->netdev;
2570 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2571 err = ixgbe_request_msix_irqs(adapter);
2572 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
2573 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2574 netdev->name, netdev);
2576 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2577 netdev->name, netdev);
2581 e_err(probe, "request_irq failed, Error %d\n", err);
2586 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2588 struct net_device *netdev = adapter->netdev;
2590 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2593 q_vectors = adapter->num_msix_vectors;
2596 free_irq(adapter->msix_entries[i].vector, netdev);
2599 for (; i >= 0; i--) {
2600 free_irq(adapter->msix_entries[i].vector,
2601 adapter->q_vector[i]);
2604 ixgbe_reset_q_vectors(adapter);
2606 free_irq(adapter->pdev->irq, netdev);
2611 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2612 * @adapter: board private structure
2614 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2616 switch (adapter->hw.mac.type) {
2617 case ixgbe_mac_82598EB:
2618 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2620 case ixgbe_mac_82599EB:
2621 case ixgbe_mac_X540:
2622 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2623 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2624 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2625 if (adapter->num_vfs > 32)
2626 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
2631 IXGBE_WRITE_FLUSH(&adapter->hw);
2632 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2634 for (i = 0; i < adapter->num_msix_vectors; i++)
2635 synchronize_irq(adapter->msix_entries[i].vector);
2637 synchronize_irq(adapter->pdev->irq);
2642 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2645 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2647 struct ixgbe_hw *hw = &adapter->hw;
2649 IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
2650 EITR_INTS_PER_SEC_TO_REG(adapter->rx_eitr_param));
2652 ixgbe_set_ivar(adapter, 0, 0, 0);
2653 ixgbe_set_ivar(adapter, 1, 0, 0);
2655 map_vector_to_rxq(adapter, 0, 0);
2656 map_vector_to_txq(adapter, 0, 0);
2658 e_info(hw, "Legacy interrupt IVAR setup done\n");
2662 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2663 * @adapter: board private structure
2664 * @ring: structure containing ring specific data
2666 * Configure the Tx descriptor ring after a reset.
2668 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2669 struct ixgbe_ring *ring)
2671 struct ixgbe_hw *hw = &adapter->hw;
2672 u64 tdba = ring->dma;
2675 u8 reg_idx = ring->reg_idx;
2677 /* disable queue to avoid issues while updating state */
2678 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2679 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx),
2680 txdctl & ~IXGBE_TXDCTL_ENABLE);
2681 IXGBE_WRITE_FLUSH(hw);
2683 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2684 (tdba & DMA_BIT_MASK(32)));
2685 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2686 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2687 ring->count * sizeof(union ixgbe_adv_tx_desc));
2688 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2689 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2690 ring->tail = hw->hw_addr + IXGBE_TDT(reg_idx);
2692 /* configure fetching thresholds */
2693 if (adapter->rx_itr_setting == 0) {
2694 /* cannot set wthresh when itr==0 */
2695 txdctl &= ~0x007F0000;
2697 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
2698 txdctl |= (8 << 16);
2700 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2701 /* PThresh workaround for Tx hang with DFP enabled. */
2705 /* reinitialize flowdirector state */
2706 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2707 adapter->atr_sample_rate) {
2708 ring->atr_sample_rate = adapter->atr_sample_rate;
2709 ring->atr_count = 0;
2710 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
2712 ring->atr_sample_rate = 0;
2715 clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
2718 txdctl |= IXGBE_TXDCTL_ENABLE;
2719 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
2721 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2722 if (hw->mac.type == ixgbe_mac_82598EB &&
2723 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2726 /* poll to verify queue is enabled */
2729 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2730 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
2732 e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
2735 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
2737 struct ixgbe_hw *hw = &adapter->hw;
2741 if (hw->mac.type == ixgbe_mac_82598EB)
2744 /* disable the arbiter while setting MTQC */
2745 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2746 rttdcs |= IXGBE_RTTDCS_ARBDIS;
2747 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2749 /* set transmit pool layout */
2750 mask = (IXGBE_FLAG_SRIOV_ENABLED | IXGBE_FLAG_DCB_ENABLED);
2751 switch (adapter->flags & mask) {
2753 case (IXGBE_FLAG_SRIOV_ENABLED):
2754 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2755 (IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF));
2758 case (IXGBE_FLAG_DCB_ENABLED):
2759 /* We enable 8 traffic classes, DCB only */
2760 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2761 (IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ));
2765 IXGBE_WRITE_REG(hw, IXGBE_MTQC, IXGBE_MTQC_64Q_1PB);
2769 /* re-enable the arbiter */
2770 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2771 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2775 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2776 * @adapter: board private structure
2778 * Configure the Tx unit of the MAC after a reset.
2780 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2782 struct ixgbe_hw *hw = &adapter->hw;
2786 ixgbe_setup_mtqc(adapter);
2788 if (hw->mac.type != ixgbe_mac_82598EB) {
2789 /* DMATXCTL.EN must be before Tx queues are enabled */
2790 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2791 dmatxctl |= IXGBE_DMATXCTL_TE;
2792 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2795 /* Setup the HW Tx Head and Tail descriptor pointers */
2796 for (i = 0; i < adapter->num_tx_queues; i++)
2797 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
2800 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2802 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2803 struct ixgbe_ring *rx_ring)
2806 u8 reg_idx = rx_ring->reg_idx;
2808 switch (adapter->hw.mac.type) {
2809 case ixgbe_mac_82598EB: {
2810 struct ixgbe_ring_feature *feature = adapter->ring_feature;
2811 const int mask = feature[RING_F_RSS].mask;
2812 reg_idx = reg_idx & mask;
2815 case ixgbe_mac_82599EB:
2816 case ixgbe_mac_X540:
2821 srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx));
2823 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
2824 srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
2825 if (adapter->num_vfs)
2826 srrctl |= IXGBE_SRRCTL_DROP_EN;
2828 srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
2829 IXGBE_SRRCTL_BSIZEHDR_MASK;
2831 if (ring_is_ps_enabled(rx_ring)) {
2832 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
2833 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2835 srrctl |= (PAGE_SIZE / 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2837 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2839 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
2840 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2841 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2844 IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx), srrctl);
2847 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2849 struct ixgbe_hw *hw = &adapter->hw;
2850 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2851 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2852 0x6A3E67EA, 0x14364D17, 0x3BED200D};
2853 u32 mrqc = 0, reta = 0;
2858 /* Fill out hash function seeds */
2859 for (i = 0; i < 10; i++)
2860 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2862 /* Fill out redirection table */
2863 for (i = 0, j = 0; i < 128; i++, j++) {
2864 if (j == adapter->ring_feature[RING_F_RSS].indices)
2866 /* reta = 4-byte sliding window of
2867 * 0x00..(indices-1)(indices-1)00..etc. */
2868 reta = (reta << 8) | (j * 0x11);
2870 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2873 /* Disable indicating checksum in descriptor, enables RSS hash */
2874 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2875 rxcsum |= IXGBE_RXCSUM_PCSD;
2876 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2878 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
2879 mask = adapter->flags & IXGBE_FLAG_RSS_ENABLED;
2881 mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2882 #ifdef CONFIG_IXGBE_DCB
2883 | IXGBE_FLAG_DCB_ENABLED
2885 | IXGBE_FLAG_SRIOV_ENABLED
2889 case (IXGBE_FLAG_RSS_ENABLED):
2890 mrqc = IXGBE_MRQC_RSSEN;
2892 case (IXGBE_FLAG_SRIOV_ENABLED):
2893 mrqc = IXGBE_MRQC_VMDQEN;
2895 #ifdef CONFIG_IXGBE_DCB
2896 case (IXGBE_FLAG_DCB_ENABLED):
2897 mrqc = IXGBE_MRQC_RT8TCEN;
2899 #endif /* CONFIG_IXGBE_DCB */
2904 /* Perform hash on these packet types */
2905 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2906 | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2907 | IXGBE_MRQC_RSS_FIELD_IPV6
2908 | IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
2910 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2914 * ixgbe_clear_rscctl - disable RSC for the indicated ring
2915 * @adapter: address of board private structure
2916 * @ring: structure containing ring specific data
2918 void ixgbe_clear_rscctl(struct ixgbe_adapter *adapter,
2919 struct ixgbe_ring *ring)
2921 struct ixgbe_hw *hw = &adapter->hw;
2923 u8 reg_idx = ring->reg_idx;
2925 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
2926 rscctrl &= ~IXGBE_RSCCTL_RSCEN;
2927 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
2931 * ixgbe_configure_rscctl - enable RSC for the indicated ring
2932 * @adapter: address of board private structure
2933 * @index: index of ring to set
2935 void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
2936 struct ixgbe_ring *ring)
2938 struct ixgbe_hw *hw = &adapter->hw;
2941 u8 reg_idx = ring->reg_idx;
2943 if (!ring_is_rsc_enabled(ring))
2946 rx_buf_len = ring->rx_buf_len;
2947 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
2948 rscctrl |= IXGBE_RSCCTL_RSCEN;
2950 * we must limit the number of descriptors so that the
2951 * total size of max desc * buf_len is not greater
2954 if (ring_is_ps_enabled(ring)) {
2955 #if (MAX_SKB_FRAGS > 16)
2956 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2957 #elif (MAX_SKB_FRAGS > 8)
2958 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2959 #elif (MAX_SKB_FRAGS > 4)
2960 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2962 rscctrl |= IXGBE_RSCCTL_MAXDESC_1;
2965 if (rx_buf_len < IXGBE_RXBUFFER_4096)
2966 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2967 else if (rx_buf_len < IXGBE_RXBUFFER_8192)
2968 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2970 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2972 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
2976 * ixgbe_set_uta - Set unicast filter table address
2977 * @adapter: board private structure
2979 * The unicast table address is a register array of 32-bit registers.
2980 * The table is meant to be used in a way similar to how the MTA is used
2981 * however due to certain limitations in the hardware it is necessary to
2982 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
2983 * enable bit to allow vlan tag stripping when promiscuous mode is enabled
2985 static void ixgbe_set_uta(struct ixgbe_adapter *adapter)
2987 struct ixgbe_hw *hw = &adapter->hw;
2990 /* The UTA table only exists on 82599 hardware and newer */
2991 if (hw->mac.type < ixgbe_mac_82599EB)
2994 /* we only need to do this if VMDq is enabled */
2995 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
2998 for (i = 0; i < 128; i++)
2999 IXGBE_WRITE_REG(hw, IXGBE_UTA(i), ~0);
3002 #define IXGBE_MAX_RX_DESC_POLL 10
3003 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
3004 struct ixgbe_ring *ring)
3006 struct ixgbe_hw *hw = &adapter->hw;
3007 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3009 u8 reg_idx = ring->reg_idx;
3011 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3012 if (hw->mac.type == ixgbe_mac_82598EB &&
3013 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3018 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3019 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
3022 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
3023 "the polling period\n", reg_idx);
3027 void ixgbe_disable_rx_queue(struct ixgbe_adapter *adapter,
3028 struct ixgbe_ring *ring)
3030 struct ixgbe_hw *hw = &adapter->hw;
3031 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3033 u8 reg_idx = ring->reg_idx;
3035 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3036 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
3038 /* write value back with RXDCTL.ENABLE bit cleared */
3039 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3041 if (hw->mac.type == ixgbe_mac_82598EB &&
3042 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3045 /* the hardware may take up to 100us to really disable the rx queue */
3048 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3049 } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
3052 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not cleared within "
3053 "the polling period\n", reg_idx);
3057 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
3058 struct ixgbe_ring *ring)
3060 struct ixgbe_hw *hw = &adapter->hw;
3061 u64 rdba = ring->dma;
3063 u8 reg_idx = ring->reg_idx;
3065 /* disable queue to avoid issues while updating state */
3066 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3067 ixgbe_disable_rx_queue(adapter, ring);
3069 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
3070 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
3071 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
3072 ring->count * sizeof(union ixgbe_adv_rx_desc));
3073 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
3074 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
3075 ring->tail = hw->hw_addr + IXGBE_RDT(reg_idx);
3077 ixgbe_configure_srrctl(adapter, ring);
3078 ixgbe_configure_rscctl(adapter, ring);
3080 if (hw->mac.type == ixgbe_mac_82598EB) {
3082 * enable cache line friendly hardware writes:
3083 * PTHRESH=32 descriptors (half the internal cache),
3084 * this also removes ugly rx_no_buffer_count increment
3085 * HTHRESH=4 descriptors (to minimize latency on fetch)
3086 * WTHRESH=8 burst writeback up to two cache lines
3088 rxdctl &= ~0x3FFFFF;
3092 /* enable receive descriptor ring */
3093 rxdctl |= IXGBE_RXDCTL_ENABLE;
3094 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3096 ixgbe_rx_desc_queue_enable(adapter, ring);
3097 ixgbe_alloc_rx_buffers(ring, IXGBE_DESC_UNUSED(ring));
3100 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
3102 struct ixgbe_hw *hw = &adapter->hw;
3105 /* PSRTYPE must be initialized in non 82598 adapters */
3106 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
3107 IXGBE_PSRTYPE_UDPHDR |
3108 IXGBE_PSRTYPE_IPV4HDR |
3109 IXGBE_PSRTYPE_L2HDR |
3110 IXGBE_PSRTYPE_IPV6HDR;
3112 if (hw->mac.type == ixgbe_mac_82598EB)
3115 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED)
3116 psrtype |= (adapter->num_rx_queues_per_pool << 29);
3118 for (p = 0; p < adapter->num_rx_pools; p++)
3119 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(adapter->num_vfs + p),
3123 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
3125 struct ixgbe_hw *hw = &adapter->hw;
3128 u32 reg_offset, vf_shift;
3131 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
3134 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
3135 vt_reg_bits = IXGBE_VMD_CTL_VMDQ_EN | IXGBE_VT_CTL_REPLEN;
3136 vt_reg_bits |= (adapter->num_vfs << IXGBE_VT_CTL_POOL_SHIFT);
3137 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl | vt_reg_bits);
3139 vf_shift = adapter->num_vfs % 32;
3140 reg_offset = (adapter->num_vfs > 32) ? 1 : 0;
3142 /* Enable only the PF's pool for Tx/Rx */
3143 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (1 << vf_shift));
3144 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), 0);
3145 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (1 << vf_shift));
3146 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), 0);
3147 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3149 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
3150 hw->mac.ops.set_vmdq(hw, 0, adapter->num_vfs);
3153 * Set up VF register offsets for selected VT Mode,
3154 * i.e. 32 or 64 VFs for SR-IOV
3156 gcr_ext = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
3157 gcr_ext |= IXGBE_GCR_EXT_MSIX_EN;
3158 gcr_ext |= IXGBE_GCR_EXT_VT_MODE_64;
3159 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
3161 /* enable Tx loopback for VF/PF communication */
3162 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3163 /* Enable MAC Anti-Spoofing */
3164 hw->mac.ops.set_mac_anti_spoofing(hw, (adapter->num_vfs != 0),
3168 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
3170 struct ixgbe_hw *hw = &adapter->hw;
3171 struct net_device *netdev = adapter->netdev;
3172 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3174 struct ixgbe_ring *rx_ring;
3178 /* Decide whether to use packet split mode or not */
3180 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
3182 /* Do not use packet split if we're in SR-IOV Mode */
3183 if (adapter->num_vfs)
3184 adapter->flags &= ~IXGBE_FLAG_RX_PS_ENABLED;
3186 /* Disable packet split due to 82599 erratum #45 */
3187 if (hw->mac.type == ixgbe_mac_82599EB)
3188 adapter->flags &= ~IXGBE_FLAG_RX_PS_ENABLED;
3190 /* Set the RX buffer length according to the mode */
3191 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
3192 rx_buf_len = IXGBE_RX_HDR_SIZE;
3194 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
3195 (netdev->mtu <= ETH_DATA_LEN))
3196 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3198 rx_buf_len = ALIGN(max_frame + VLAN_HLEN, 1024);
3202 /* adjust max frame to be able to do baby jumbo for FCoE */
3203 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
3204 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
3205 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3207 #endif /* IXGBE_FCOE */
3208 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
3209 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
3210 mhadd &= ~IXGBE_MHADD_MFS_MASK;
3211 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
3213 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
3216 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
3217 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
3218 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
3219 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
3222 * Setup the HW Rx Head and Tail Descriptor Pointers and
3223 * the Base and Length of the Rx Descriptor Ring
3225 for (i = 0; i < adapter->num_rx_queues; i++) {
3226 rx_ring = adapter->rx_ring[i];
3227 rx_ring->rx_buf_len = rx_buf_len;
3229 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)
3230 set_ring_ps_enabled(rx_ring);
3232 clear_ring_ps_enabled(rx_ring);
3234 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
3235 set_ring_rsc_enabled(rx_ring);
3237 clear_ring_rsc_enabled(rx_ring);
3240 if (netdev->features & NETIF_F_FCOE_MTU) {
3241 struct ixgbe_ring_feature *f;
3242 f = &adapter->ring_feature[RING_F_FCOE];
3243 if ((i >= f->mask) && (i < f->mask + f->indices)) {
3244 clear_ring_ps_enabled(rx_ring);
3245 if (rx_buf_len < IXGBE_FCOE_JUMBO_FRAME_SIZE)
3246 rx_ring->rx_buf_len =
3247 IXGBE_FCOE_JUMBO_FRAME_SIZE;
3248 } else if (!ring_is_rsc_enabled(rx_ring) &&
3249 !ring_is_ps_enabled(rx_ring)) {
3250 rx_ring->rx_buf_len =
3251 IXGBE_FCOE_JUMBO_FRAME_SIZE;
3254 #endif /* IXGBE_FCOE */
3258 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
3260 struct ixgbe_hw *hw = &adapter->hw;
3261 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
3263 switch (hw->mac.type) {
3264 case ixgbe_mac_82598EB:
3266 * For VMDq support of different descriptor types or
3267 * buffer sizes through the use of multiple SRRCTL
3268 * registers, RDRXCTL.MVMEN must be set to 1
3270 * also, the manual doesn't mention it clearly but DCA hints
3271 * will only use queue 0's tags unless this bit is set. Side
3272 * effects of setting this bit are only that SRRCTL must be
3273 * fully programmed [0..15]
3275 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
3277 case ixgbe_mac_82599EB:
3278 case ixgbe_mac_X540:
3279 /* Disable RSC for ACK packets */
3280 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
3281 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
3282 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3283 /* hardware requires some bits to be set by default */
3284 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3285 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3288 /* We should do nothing since we don't know this hardware */
3292 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3296 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3297 * @adapter: board private structure
3299 * Configure the Rx unit of the MAC after a reset.
3301 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3303 struct ixgbe_hw *hw = &adapter->hw;
3307 /* disable receives while setting up the descriptors */
3308 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3309 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3311 ixgbe_setup_psrtype(adapter);
3312 ixgbe_setup_rdrxctl(adapter);
3314 /* Program registers for the distribution of queues */
3315 ixgbe_setup_mrqc(adapter);
3317 ixgbe_set_uta(adapter);
3319 /* set_rx_buffer_len must be called before ring initialization */
3320 ixgbe_set_rx_buffer_len(adapter);
3323 * Setup the HW Rx Head and Tail Descriptor Pointers and
3324 * the Base and Length of the Rx Descriptor Ring
3326 for (i = 0; i < adapter->num_rx_queues; i++)
3327 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3329 /* disable drop enable for 82598 parts */
3330 if (hw->mac.type == ixgbe_mac_82598EB)
3331 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3333 /* enable all receives */
3334 rxctrl |= IXGBE_RXCTRL_RXEN;
3335 hw->mac.ops.enable_rx_dma(hw, rxctrl);
3338 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3340 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3341 struct ixgbe_hw *hw = &adapter->hw;
3342 int pool_ndx = adapter->num_vfs;
3344 /* add VID to filter table */
3345 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, true);
3346 set_bit(vid, adapter->active_vlans);
3349 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3351 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3352 struct ixgbe_hw *hw = &adapter->hw;
3353 int pool_ndx = adapter->num_vfs;
3355 /* remove VID from filter table */
3356 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, false);
3357 clear_bit(vid, adapter->active_vlans);
3361 * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3362 * @adapter: driver data
3364 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
3366 struct ixgbe_hw *hw = &adapter->hw;
3369 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3370 vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
3371 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3375 * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3376 * @adapter: driver data
3378 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
3380 struct ixgbe_hw *hw = &adapter->hw;
3383 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3384 vlnctrl |= IXGBE_VLNCTRL_VFE;
3385 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3386 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3390 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3391 * @adapter: driver data
3393 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3395 struct ixgbe_hw *hw = &adapter->hw;
3399 switch (hw->mac.type) {
3400 case ixgbe_mac_82598EB:
3401 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3402 vlnctrl &= ~IXGBE_VLNCTRL_VME;
3403 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3405 case ixgbe_mac_82599EB:
3406 case ixgbe_mac_X540:
3407 for (i = 0; i < adapter->num_rx_queues; i++) {
3408 j = adapter->rx_ring[i]->reg_idx;
3409 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3410 vlnctrl &= ~IXGBE_RXDCTL_VME;
3411 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3420 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3421 * @adapter: driver data
3423 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3425 struct ixgbe_hw *hw = &adapter->hw;
3429 switch (hw->mac.type) {
3430 case ixgbe_mac_82598EB:
3431 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3432 vlnctrl |= IXGBE_VLNCTRL_VME;
3433 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3435 case ixgbe_mac_82599EB:
3436 case ixgbe_mac_X540:
3437 for (i = 0; i < adapter->num_rx_queues; i++) {
3438 j = adapter->rx_ring[i]->reg_idx;
3439 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3440 vlnctrl |= IXGBE_RXDCTL_VME;
3441 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3449 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3453 ixgbe_vlan_rx_add_vid(adapter->netdev, 0);
3455 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3456 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
3460 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3461 * @netdev: network interface device structure
3463 * Writes unicast address list to the RAR table.
3464 * Returns: -ENOMEM on failure/insufficient address space
3465 * 0 on no addresses written
3466 * X on writing X addresses to the RAR table
3468 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3470 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3471 struct ixgbe_hw *hw = &adapter->hw;
3472 unsigned int vfn = adapter->num_vfs;
3473 unsigned int rar_entries = hw->mac.num_rar_entries - (vfn + 1);
3476 /* return ENOMEM indicating insufficient memory for addresses */
3477 if (netdev_uc_count(netdev) > rar_entries)
3480 if (!netdev_uc_empty(netdev) && rar_entries) {
3481 struct netdev_hw_addr *ha;
3482 /* return error if we do not support writing to RAR table */
3483 if (!hw->mac.ops.set_rar)
3486 netdev_for_each_uc_addr(ha, netdev) {
3489 hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3494 /* write the addresses in reverse order to avoid write combining */
3495 for (; rar_entries > 0 ; rar_entries--)
3496 hw->mac.ops.clear_rar(hw, rar_entries);
3502 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3503 * @netdev: network interface device structure
3505 * The set_rx_method entry point is called whenever the unicast/multicast
3506 * address list or the network interface flags are updated. This routine is
3507 * responsible for configuring the hardware for proper unicast, multicast and
3510 void ixgbe_set_rx_mode(struct net_device *netdev)
3512 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3513 struct ixgbe_hw *hw = &adapter->hw;
3514 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3517 /* Check for Promiscuous and All Multicast modes */
3519 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3521 /* set all bits that we expect to always be set */
3522 fctrl |= IXGBE_FCTRL_BAM;
3523 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3524 fctrl |= IXGBE_FCTRL_PMCF;
3526 /* clear the bits we are changing the status of */
3527 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3529 if (netdev->flags & IFF_PROMISC) {
3530 hw->addr_ctrl.user_set_promisc = true;
3531 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3532 vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3533 /* don't hardware filter vlans in promisc mode */
3534 ixgbe_vlan_filter_disable(adapter);
3536 if (netdev->flags & IFF_ALLMULTI) {
3537 fctrl |= IXGBE_FCTRL_MPE;
3538 vmolr |= IXGBE_VMOLR_MPE;
3541 * Write addresses to the MTA, if the attempt fails
3542 * then we should just turn on promiscous mode so
3543 * that we can at least receive multicast traffic
3545 hw->mac.ops.update_mc_addr_list(hw, netdev);
3546 vmolr |= IXGBE_VMOLR_ROMPE;
3548 ixgbe_vlan_filter_enable(adapter);
3549 hw->addr_ctrl.user_set_promisc = false;
3551 * Write addresses to available RAR registers, if there is not
3552 * sufficient space to store all the addresses then enable
3553 * unicast promiscous mode
3555 count = ixgbe_write_uc_addr_list(netdev);
3557 fctrl |= IXGBE_FCTRL_UPE;
3558 vmolr |= IXGBE_VMOLR_ROPE;
3562 if (adapter->num_vfs) {
3563 ixgbe_restore_vf_multicasts(adapter);
3564 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(adapter->num_vfs)) &
3565 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3567 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(adapter->num_vfs), vmolr);
3570 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3572 if (netdev->features & NETIF_F_HW_VLAN_RX)
3573 ixgbe_vlan_strip_enable(adapter);
3575 ixgbe_vlan_strip_disable(adapter);
3578 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3581 struct ixgbe_q_vector *q_vector;
3582 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3584 /* legacy and MSI only use one vector */
3585 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3588 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3589 struct napi_struct *napi;
3590 q_vector = adapter->q_vector[q_idx];
3591 napi = &q_vector->napi;
3592 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3593 if (!q_vector->rxr_count || !q_vector->txr_count) {
3594 if (q_vector->txr_count == 1)
3595 napi->poll = &ixgbe_clean_txonly;
3596 else if (q_vector->rxr_count == 1)
3597 napi->poll = &ixgbe_clean_rxonly;
3605 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3608 struct ixgbe_q_vector *q_vector;
3609 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3611 /* legacy and MSI only use one vector */
3612 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3615 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3616 q_vector = adapter->q_vector[q_idx];
3617 napi_disable(&q_vector->napi);
3621 #ifdef CONFIG_IXGBE_DCB
3623 * ixgbe_configure_dcb - Configure DCB hardware
3624 * @adapter: ixgbe adapter struct
3626 * This is called by the driver on open to configure the DCB hardware.
3627 * This is also called by the gennetlink interface when reconfiguring
3630 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3632 struct ixgbe_hw *hw = &adapter->hw;
3633 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3635 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3636 if (hw->mac.type == ixgbe_mac_82598EB)
3637 netif_set_gso_max_size(adapter->netdev, 65536);
3641 if (hw->mac.type == ixgbe_mac_82598EB)
3642 netif_set_gso_max_size(adapter->netdev, 32768);
3645 if (adapter->netdev->features & NETIF_F_FCOE_MTU)
3646 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
3649 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3651 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3654 /* Enable VLAN tag insert/strip */
3655 adapter->netdev->features |= NETIF_F_HW_VLAN_RX;
3657 hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
3659 /* reconfigure the hardware */
3660 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
3664 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3666 struct net_device *netdev = adapter->netdev;
3667 struct ixgbe_hw *hw = &adapter->hw;
3670 #ifdef CONFIG_IXGBE_DCB
3671 ixgbe_configure_dcb(adapter);
3674 ixgbe_set_rx_mode(netdev);
3675 ixgbe_restore_vlan(adapter);
3678 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
3679 ixgbe_configure_fcoe(adapter);
3681 #endif /* IXGBE_FCOE */
3682 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3683 for (i = 0; i < adapter->num_tx_queues; i++)
3684 adapter->tx_ring[i]->atr_sample_rate =
3685 adapter->atr_sample_rate;
3686 ixgbe_init_fdir_signature_82599(hw, adapter->fdir_pballoc);
3687 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3688 ixgbe_init_fdir_perfect_82599(hw, adapter->fdir_pballoc);
3690 ixgbe_configure_virtualization(adapter);
3692 ixgbe_configure_tx(adapter);
3693 ixgbe_configure_rx(adapter);
3696 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3698 switch (hw->phy.type) {
3699 case ixgbe_phy_sfp_avago:
3700 case ixgbe_phy_sfp_ftl:
3701 case ixgbe_phy_sfp_intel:
3702 case ixgbe_phy_sfp_unknown:
3703 case ixgbe_phy_sfp_passive_tyco:
3704 case ixgbe_phy_sfp_passive_unknown:
3705 case ixgbe_phy_sfp_active_unknown:
3706 case ixgbe_phy_sfp_ftl_active:
3714 * ixgbe_sfp_link_config - set up SFP+ link
3715 * @adapter: pointer to private adapter struct
3717 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3719 struct ixgbe_hw *hw = &adapter->hw;
3721 if (hw->phy.multispeed_fiber) {
3723 * In multispeed fiber setups, the device may not have
3724 * had a physical connection when the driver loaded.
3725 * If that's the case, the initial link configuration
3726 * couldn't get the MAC into 10G or 1G mode, so we'll
3727 * never have a link status change interrupt fire.
3728 * We need to try and force an autonegotiation
3729 * session, then bring up link.
3731 hw->mac.ops.setup_sfp(hw);
3732 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
3733 schedule_work(&adapter->multispeed_fiber_task);
3736 * Direct Attach Cu and non-multispeed fiber modules
3737 * still need to be configured properly prior to
3740 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_MOD_TASK))
3741 schedule_work(&adapter->sfp_config_module_task);
3746 * ixgbe_non_sfp_link_config - set up non-SFP+ link
3747 * @hw: pointer to private hardware struct
3749 * Returns 0 on success, negative on failure
3751 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3754 bool negotiation, link_up = false;
3755 u32 ret = IXGBE_ERR_LINK_SETUP;
3757 if (hw->mac.ops.check_link)
3758 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3763 if (hw->mac.ops.get_link_capabilities)
3764 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
3769 if (hw->mac.ops.setup_link)
3770 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
3775 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
3777 struct ixgbe_hw *hw = &adapter->hw;
3780 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3781 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
3783 gpie |= IXGBE_GPIE_EIAME;
3785 * use EIAM to auto-mask when MSI-X interrupt is asserted
3786 * this saves a register write for every interrupt
3788 switch (hw->mac.type) {
3789 case ixgbe_mac_82598EB:
3790 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3792 case ixgbe_mac_82599EB:
3793 case ixgbe_mac_X540:
3795 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
3796 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
3800 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
3801 * specifically only auto mask tx and rx interrupts */
3802 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3805 /* XXX: to interrupt immediately for EICS writes, enable this */
3806 /* gpie |= IXGBE_GPIE_EIMEN; */
3808 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3809 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
3810 gpie |= IXGBE_GPIE_VTMODE_64;
3813 /* Enable fan failure interrupt */
3814 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3815 gpie |= IXGBE_SDP1_GPIEN;
3817 if (hw->mac.type == ixgbe_mac_82599EB)
3818 gpie |= IXGBE_SDP1_GPIEN;
3819 gpie |= IXGBE_SDP2_GPIEN;
3821 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3824 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
3826 struct ixgbe_hw *hw = &adapter->hw;
3830 ixgbe_get_hw_control(adapter);
3831 ixgbe_setup_gpie(adapter);
3833 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3834 ixgbe_configure_msix(adapter);
3836 ixgbe_configure_msi_and_legacy(adapter);
3838 /* enable the optics for both mult-speed fiber and 82599 SFP+ fiber */
3839 if (hw->mac.ops.enable_tx_laser &&
3840 ((hw->phy.multispeed_fiber) ||
3841 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
3842 (hw->mac.type == ixgbe_mac_82599EB))))
3843 hw->mac.ops.enable_tx_laser(hw);
3845 clear_bit(__IXGBE_DOWN, &adapter->state);
3846 ixgbe_napi_enable_all(adapter);
3848 if (ixgbe_is_sfp(hw)) {
3849 ixgbe_sfp_link_config(adapter);
3851 err = ixgbe_non_sfp_link_config(hw);
3853 e_err(probe, "link_config FAILED %d\n", err);
3856 /* clear any pending interrupts, may auto mask */
3857 IXGBE_READ_REG(hw, IXGBE_EICR);
3858 ixgbe_irq_enable(adapter, true, true);
3861 * If this adapter has a fan, check to see if we had a failure
3862 * before we enabled the interrupt.
3864 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3865 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
3866 if (esdp & IXGBE_ESDP_SDP1)
3867 e_crit(drv, "Fan has stopped, replace the adapter\n");
3871 * For hot-pluggable SFP+ devices, a new SFP+ module may have
3872 * arrived before interrupts were enabled but after probe. Such
3873 * devices wouldn't have their type identified yet. We need to
3874 * kick off the SFP+ module setup first, then try to bring up link.
3875 * If we're not hot-pluggable SFP+, we just need to configure link
3878 if (hw->phy.type == ixgbe_phy_unknown)
3879 schedule_work(&adapter->sfp_config_module_task);
3881 /* enable transmits */
3882 netif_tx_start_all_queues(adapter->netdev);
3884 /* bring the link up in the watchdog, this could race with our first
3885 * link up interrupt but shouldn't be a problem */
3886 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3887 adapter->link_check_timeout = jiffies;
3888 mod_timer(&adapter->watchdog_timer, jiffies);
3890 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
3891 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
3892 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
3893 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
3898 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
3900 WARN_ON(in_interrupt());
3901 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
3903 ixgbe_down(adapter);
3905 * If SR-IOV enabled then wait a bit before bringing the adapter
3906 * back up to give the VFs time to respond to the reset. The
3907 * two second wait is based upon the watchdog timer cycle in
3910 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3913 clear_bit(__IXGBE_RESETTING, &adapter->state);
3916 int ixgbe_up(struct ixgbe_adapter *adapter)
3918 /* hardware has been reset, we need to reload some things */
3919 ixgbe_configure(adapter);
3921 return ixgbe_up_complete(adapter);
3924 void ixgbe_reset(struct ixgbe_adapter *adapter)
3926 struct ixgbe_hw *hw = &adapter->hw;
3929 err = hw->mac.ops.init_hw(hw);
3932 case IXGBE_ERR_SFP_NOT_PRESENT:
3934 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
3935 e_dev_err("master disable timed out\n");
3937 case IXGBE_ERR_EEPROM_VERSION:
3938 /* We are running on a pre-production device, log a warning */
3939 e_dev_warn("This device is a pre-production adapter/LOM. "
3940 "Please be aware there may be issuesassociated with "
3941 "your hardware. If you are experiencing problems "
3942 "please contact your Intel or hardware "
3943 "representative who provided you with this "
3947 e_dev_err("Hardware Error: %d\n", err);
3950 /* reprogram the RAR[0] in case user changed it. */
3951 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
3956 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
3957 * @rx_ring: ring to free buffers from
3959 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
3961 struct device *dev = rx_ring->dev;
3965 /* ring already cleared, nothing to do */
3966 if (!rx_ring->rx_buffer_info)
3969 /* Free all the Rx ring sk_buffs */
3970 for (i = 0; i < rx_ring->count; i++) {
3971 struct ixgbe_rx_buffer *rx_buffer_info;
3973 rx_buffer_info = &rx_ring->rx_buffer_info[i];
3974 if (rx_buffer_info->dma) {
3975 dma_unmap_single(rx_ring->dev, rx_buffer_info->dma,
3976 rx_ring->rx_buf_len,
3978 rx_buffer_info->dma = 0;
3980 if (rx_buffer_info->skb) {
3981 struct sk_buff *skb = rx_buffer_info->skb;
3982 rx_buffer_info->skb = NULL;
3984 struct sk_buff *this = skb;
3985 if (IXGBE_RSC_CB(this)->delay_unmap) {
3986 dma_unmap_single(dev,
3987 IXGBE_RSC_CB(this)->dma,
3988 rx_ring->rx_buf_len,
3990 IXGBE_RSC_CB(this)->dma = 0;
3991 IXGBE_RSC_CB(skb)->delay_unmap = false;
3994 dev_kfree_skb(this);
3997 if (!rx_buffer_info->page)
3999 if (rx_buffer_info->page_dma) {
4000 dma_unmap_page(dev, rx_buffer_info->page_dma,
4001 PAGE_SIZE / 2, DMA_FROM_DEVICE);
4002 rx_buffer_info->page_dma = 0;
4004 put_page(rx_buffer_info->page);
4005 rx_buffer_info->page = NULL;
4006 rx_buffer_info->page_offset = 0;
4009 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4010 memset(rx_ring->rx_buffer_info, 0, size);
4012 /* Zero out the descriptor ring */
4013 memset(rx_ring->desc, 0, rx_ring->size);
4015 rx_ring->next_to_clean = 0;
4016 rx_ring->next_to_use = 0;
4020 * ixgbe_clean_tx_ring - Free Tx Buffers
4021 * @tx_ring: ring to be cleaned
4023 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
4025 struct ixgbe_tx_buffer *tx_buffer_info;
4029 /* ring already cleared, nothing to do */
4030 if (!tx_ring->tx_buffer_info)
4033 /* Free all the Tx ring sk_buffs */
4034 for (i = 0; i < tx_ring->count; i++) {
4035 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4036 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
4039 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4040 memset(tx_ring->tx_buffer_info, 0, size);
4042 /* Zero out the descriptor ring */
4043 memset(tx_ring->desc, 0, tx_ring->size);
4045 tx_ring->next_to_use = 0;
4046 tx_ring->next_to_clean = 0;
4050 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
4051 * @adapter: board private structure
4053 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
4057 for (i = 0; i < adapter->num_rx_queues; i++)
4058 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
4062 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
4063 * @adapter: board private structure
4065 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
4069 for (i = 0; i < adapter->num_tx_queues; i++)
4070 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
4073 void ixgbe_down(struct ixgbe_adapter *adapter)
4075 struct net_device *netdev = adapter->netdev;
4076 struct ixgbe_hw *hw = &adapter->hw;
4080 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4082 /* signal that we are down to the interrupt handler */
4083 set_bit(__IXGBE_DOWN, &adapter->state);
4085 /* disable receive for all VFs and wait one second */
4086 if (adapter->num_vfs) {
4087 /* ping all the active vfs to let them know we are going down */
4088 ixgbe_ping_all_vfs(adapter);
4090 /* Disable all VFTE/VFRE TX/RX */
4091 ixgbe_disable_tx_rx(adapter);
4093 /* Mark all the VFs as inactive */
4094 for (i = 0 ; i < adapter->num_vfs; i++)
4095 adapter->vfinfo[i].clear_to_send = 0;
4098 /* disable receives */
4099 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4100 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
4102 /* disable all enabled rx queues */
4103 for (i = 0; i < adapter->num_rx_queues; i++)
4104 /* this call also flushes the previous write */
4105 ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]);
4109 netif_tx_stop_all_queues(netdev);
4111 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
4112 del_timer_sync(&adapter->sfp_timer);
4113 del_timer_sync(&adapter->watchdog_timer);
4114 cancel_work_sync(&adapter->watchdog_task);
4116 netif_carrier_off(netdev);
4117 netif_tx_disable(netdev);
4119 ixgbe_irq_disable(adapter);
4121 ixgbe_napi_disable_all(adapter);
4123 /* Cleanup the affinity_hint CPU mask memory and callback */
4124 for (i = 0; i < num_q_vectors; i++) {
4125 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
4126 /* clear the affinity_mask in the IRQ descriptor */
4127 irq_set_affinity_hint(adapter->msix_entries[i]. vector, NULL);
4128 /* release the CPU mask memory */
4129 free_cpumask_var(q_vector->affinity_mask);
4132 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
4133 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
4134 cancel_work_sync(&adapter->fdir_reinit_task);
4136 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
4137 cancel_work_sync(&adapter->check_overtemp_task);
4139 /* disable transmits in the hardware now that interrupts are off */
4140 for (i = 0; i < adapter->num_tx_queues; i++) {
4141 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
4142 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
4143 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx),
4144 (txdctl & ~IXGBE_TXDCTL_ENABLE));
4146 /* Disable the Tx DMA engine on 82599 */
4147 switch (hw->mac.type) {
4148 case ixgbe_mac_82599EB:
4149 case ixgbe_mac_X540:
4150 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
4151 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
4152 ~IXGBE_DMATXCTL_TE));
4158 /* clear n-tuple filters that are cached */
4159 ethtool_ntuple_flush(netdev);
4161 if (!pci_channel_offline(adapter->pdev))
4162 ixgbe_reset(adapter);
4164 /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
4165 if (hw->mac.ops.disable_tx_laser &&
4166 ((hw->phy.multispeed_fiber) ||
4167 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
4168 (hw->mac.type == ixgbe_mac_82599EB))))
4169 hw->mac.ops.disable_tx_laser(hw);
4171 ixgbe_clean_all_tx_rings(adapter);
4172 ixgbe_clean_all_rx_rings(adapter);
4174 #ifdef CONFIG_IXGBE_DCA
4175 /* since we reset the hardware DCA settings were cleared */
4176 ixgbe_setup_dca(adapter);
4181 * ixgbe_poll - NAPI Rx polling callback
4182 * @napi: structure for representing this polling device
4183 * @budget: how many packets driver is allowed to clean
4185 * This function is used for legacy and MSI, NAPI mode
4187 static int ixgbe_poll(struct napi_struct *napi, int budget)
4189 struct ixgbe_q_vector *q_vector =
4190 container_of(napi, struct ixgbe_q_vector, napi);
4191 struct ixgbe_adapter *adapter = q_vector->adapter;
4192 int tx_clean_complete, work_done = 0;
4194 #ifdef CONFIG_IXGBE_DCA
4195 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
4196 ixgbe_update_dca(q_vector);
4199 tx_clean_complete = ixgbe_clean_tx_irq(q_vector, adapter->tx_ring[0]);
4200 ixgbe_clean_rx_irq(q_vector, adapter->rx_ring[0], &work_done, budget);
4202 if (!tx_clean_complete)
4205 /* If budget not fully consumed, exit the polling mode */
4206 if (work_done < budget) {
4207 napi_complete(napi);
4208 if (adapter->rx_itr_setting & 1)
4209 ixgbe_set_itr(adapter);
4210 if (!test_bit(__IXGBE_DOWN, &adapter->state))
4211 ixgbe_irq_enable_queues(adapter, IXGBE_EIMS_RTX_QUEUE);
4217 * ixgbe_tx_timeout - Respond to a Tx Hang
4218 * @netdev: network interface device structure
4220 static void ixgbe_tx_timeout(struct net_device *netdev)
4222 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4224 adapter->tx_timeout_count++;
4226 /* Do the reset outside of interrupt context */
4227 schedule_work(&adapter->reset_task);
4230 static void ixgbe_reset_task(struct work_struct *work)
4232 struct ixgbe_adapter *adapter;
4233 adapter = container_of(work, struct ixgbe_adapter, reset_task);
4235 /* If we're already down or resetting, just bail */
4236 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
4237 test_bit(__IXGBE_RESETTING, &adapter->state))
4240 ixgbe_dump(adapter);
4241 netdev_err(adapter->netdev, "Reset adapter\n");
4242 ixgbe_reinit_locked(adapter);
4245 #ifdef CONFIG_IXGBE_DCB
4246 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
4249 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_DCB];
4251 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
4255 adapter->num_rx_queues = f->indices;
4256 adapter->num_tx_queues = f->indices;
4264 * ixgbe_set_rss_queues: Allocate queues for RSS
4265 * @adapter: board private structure to initialize
4267 * This is our "base" multiqueue mode. RSS (Receive Side Scaling) will try
4268 * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
4271 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
4274 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
4276 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4278 adapter->num_rx_queues = f->indices;
4279 adapter->num_tx_queues = f->indices;
4289 * ixgbe_set_fdir_queues: Allocate queues for Flow Director
4290 * @adapter: board private structure to initialize
4292 * Flow Director is an advanced Rx filter, attempting to get Rx flows back
4293 * to the original CPU that initiated the Tx session. This runs in addition
4294 * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
4295 * Rx load across CPUs using RSS.
4298 static inline bool ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
4301 struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
4303 f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
4306 /* Flow Director must have RSS enabled */
4307 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
4308 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
4309 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)))) {
4310 adapter->num_tx_queues = f_fdir->indices;
4311 adapter->num_rx_queues = f_fdir->indices;
4314 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4315 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4322 * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
4323 * @adapter: board private structure to initialize
4325 * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
4326 * The ring feature mask is not used as a mask for FCoE, as it can take any 8
4327 * rx queues out of the max number of rx queues, instead, it is used as the
4328 * index of the first rx queue used by FCoE.
4331 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
4334 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4336 f->indices = min((int)num_online_cpus(), f->indices);
4337 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
4338 adapter->num_rx_queues = 1;
4339 adapter->num_tx_queues = 1;
4340 #ifdef CONFIG_IXGBE_DCB
4341 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4342 e_info(probe, "FCoE enabled with DCB\n");
4343 ixgbe_set_dcb_queues(adapter);
4346 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4347 e_info(probe, "FCoE enabled with RSS\n");
4348 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4349 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
4350 ixgbe_set_fdir_queues(adapter);
4352 ixgbe_set_rss_queues(adapter);
4354 /* adding FCoE rx rings to the end */
4355 f->mask = adapter->num_rx_queues;
4356 adapter->num_rx_queues += f->indices;
4357 adapter->num_tx_queues += f->indices;
4365 #endif /* IXGBE_FCOE */
4367 * ixgbe_set_sriov_queues: Allocate queues for IOV use
4368 * @adapter: board private structure to initialize
4370 * IOV doesn't actually use anything, so just NAK the
4371 * request for now and let the other queue routines
4372 * figure out what to do.
4374 static inline bool ixgbe_set_sriov_queues(struct ixgbe_adapter *adapter)
4380 * ixgbe_set_num_queues: Allocate queues for device, feature dependant
4381 * @adapter: board private structure to initialize
4383 * This is the top level queue allocation routine. The order here is very
4384 * important, starting with the "most" number of features turned on at once,
4385 * and ending with the smallest set of features. This way large combinations
4386 * can be allocated if they're turned on, and smaller combinations are the
4387 * fallthrough conditions.
4390 static int ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
4392 /* Start with base case */
4393 adapter->num_rx_queues = 1;
4394 adapter->num_tx_queues = 1;
4395 adapter->num_rx_pools = adapter->num_rx_queues;
4396 adapter->num_rx_queues_per_pool = 1;
4398 if (ixgbe_set_sriov_queues(adapter))
4402 if (ixgbe_set_fcoe_queues(adapter))
4405 #endif /* IXGBE_FCOE */
4406 #ifdef CONFIG_IXGBE_DCB
4407 if (ixgbe_set_dcb_queues(adapter))
4411 if (ixgbe_set_fdir_queues(adapter))
4414 if (ixgbe_set_rss_queues(adapter))
4417 /* fallback to base case */
4418 adapter->num_rx_queues = 1;
4419 adapter->num_tx_queues = 1;
4422 /* Notify the stack of the (possibly) reduced queue counts. */
4423 netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
4424 return netif_set_real_num_rx_queues(adapter->netdev,
4425 adapter->num_rx_queues);
4428 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
4431 int err, vector_threshold;
4433 /* We'll want at least 3 (vector_threshold):
4436 * 3) Other (Link Status Change, etc.)
4437 * 4) TCP Timer (optional)
4439 vector_threshold = MIN_MSIX_COUNT;
4441 /* The more we get, the more we will assign to Tx/Rx Cleanup
4442 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
4443 * Right now, we simply care about how many we'll get; we'll
4444 * set them up later while requesting irq's.
4446 while (vectors >= vector_threshold) {
4447 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
4449 if (!err) /* Success in acquiring all requested vectors. */
4452 vectors = 0; /* Nasty failure, quit now */
4453 else /* err == number of vectors we should try again with */
4457 if (vectors < vector_threshold) {
4458 /* Can't allocate enough MSI-X interrupts? Oh well.
4459 * This just means we'll go with either a single MSI
4460 * vector or fall back to legacy interrupts.
4462 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4463 "Unable to allocate MSI-X interrupts\n");
4464 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4465 kfree(adapter->msix_entries);
4466 adapter->msix_entries = NULL;
4468 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
4470 * Adjust for only the vectors we'll use, which is minimum
4471 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
4472 * vectors we were allocated.
4474 adapter->num_msix_vectors = min(vectors,
4475 adapter->max_msix_q_vectors + NON_Q_VECTORS);
4480 * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
4481 * @adapter: board private structure to initialize
4483 * Cache the descriptor ring offsets for RSS to the assigned rings.
4486 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
4490 if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED))
4493 for (i = 0; i < adapter->num_rx_queues; i++)
4494 adapter->rx_ring[i]->reg_idx = i;
4495 for (i = 0; i < adapter->num_tx_queues; i++)
4496 adapter->tx_ring[i]->reg_idx = i;
4501 #ifdef CONFIG_IXGBE_DCB
4503 * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
4504 * @adapter: board private structure to initialize
4506 * Cache the descriptor ring offsets for DCB to the assigned rings.
4509 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
4513 int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
4515 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
4518 /* the number of queues is assumed to be symmetric */
4519 switch (adapter->hw.mac.type) {
4520 case ixgbe_mac_82598EB:
4521 for (i = 0; i < dcb_i; i++) {
4522 adapter->rx_ring[i]->reg_idx = i << 3;
4523 adapter->tx_ring[i]->reg_idx = i << 2;
4527 case ixgbe_mac_82599EB:
4528 case ixgbe_mac_X540:
4531 * Tx TC0 starts at: descriptor queue 0
4532 * Tx TC1 starts at: descriptor queue 32
4533 * Tx TC2 starts at: descriptor queue 64
4534 * Tx TC3 starts at: descriptor queue 80
4535 * Tx TC4 starts at: descriptor queue 96
4536 * Tx TC5 starts at: descriptor queue 104
4537 * Tx TC6 starts at: descriptor queue 112
4538 * Tx TC7 starts at: descriptor queue 120
4540 * Rx TC0-TC7 are offset by 16 queues each
4542 for (i = 0; i < 3; i++) {
4543 adapter->tx_ring[i]->reg_idx = i << 5;
4544 adapter->rx_ring[i]->reg_idx = i << 4;
4546 for ( ; i < 5; i++) {
4547 adapter->tx_ring[i]->reg_idx = ((i + 2) << 4);
4548 adapter->rx_ring[i]->reg_idx = i << 4;
4550 for ( ; i < dcb_i; i++) {
4551 adapter->tx_ring[i]->reg_idx = ((i + 8) << 3);
4552 adapter->rx_ring[i]->reg_idx = i << 4;
4555 } else if (dcb_i == 4) {
4557 * Tx TC0 starts at: descriptor queue 0
4558 * Tx TC1 starts at: descriptor queue 64
4559 * Tx TC2 starts at: descriptor queue 96
4560 * Tx TC3 starts at: descriptor queue 112
4562 * Rx TC0-TC3 are offset by 32 queues each
4564 adapter->tx_ring[0]->reg_idx = 0;
4565 adapter->tx_ring[1]->reg_idx = 64;
4566 adapter->tx_ring[2]->reg_idx = 96;
4567 adapter->tx_ring[3]->reg_idx = 112;
4568 for (i = 0 ; i < dcb_i; i++)
4569 adapter->rx_ring[i]->reg_idx = i << 5;
4581 * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
4582 * @adapter: board private structure to initialize
4584 * Cache the descriptor ring offsets for Flow Director to the assigned rings.
4587 static inline bool ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
4592 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
4593 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4594 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))) {
4595 for (i = 0; i < adapter->num_rx_queues; i++)
4596 adapter->rx_ring[i]->reg_idx = i;
4597 for (i = 0; i < adapter->num_tx_queues; i++)
4598 adapter->tx_ring[i]->reg_idx = i;
4607 * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
4608 * @adapter: board private structure to initialize
4610 * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
4613 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
4615 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4617 u8 fcoe_rx_i = 0, fcoe_tx_i = 0;
4619 if (!(adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
4622 #ifdef CONFIG_IXGBE_DCB
4623 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4624 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
4626 ixgbe_cache_ring_dcb(adapter);
4627 /* find out queues in TC for FCoE */
4628 fcoe_rx_i = adapter->rx_ring[fcoe->tc]->reg_idx + 1;
4629 fcoe_tx_i = adapter->tx_ring[fcoe->tc]->reg_idx + 1;
4631 * In 82599, the number of Tx queues for each traffic
4632 * class for both 8-TC and 4-TC modes are:
4633 * TCs : TC0 TC1 TC2 TC3 TC4 TC5 TC6 TC7
4634 * 8 TCs: 32 32 16 16 8 8 8 8
4635 * 4 TCs: 64 64 32 32
4636 * We have max 8 queues for FCoE, where 8 the is
4637 * FCoE redirection table size. If TC for FCoE is
4638 * less than or equal to TC3, we have enough queues
4639 * to add max of 8 queues for FCoE, so we start FCoE
4640 * Tx queue from the next one, i.e., reg_idx + 1.
4641 * If TC for FCoE is above TC3, implying 8 TC mode,
4642 * and we need 8 for FCoE, we have to take all queues
4643 * in that traffic class for FCoE.
4645 if ((f->indices == IXGBE_FCRETA_SIZE) && (fcoe->tc > 3))
4648 #endif /* CONFIG_IXGBE_DCB */
4649 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4650 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4651 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
4652 ixgbe_cache_ring_fdir(adapter);
4654 ixgbe_cache_ring_rss(adapter);
4656 fcoe_rx_i = f->mask;
4657 fcoe_tx_i = f->mask;
4659 for (i = 0; i < f->indices; i++, fcoe_rx_i++, fcoe_tx_i++) {
4660 adapter->rx_ring[f->mask + i]->reg_idx = fcoe_rx_i;
4661 adapter->tx_ring[f->mask + i]->reg_idx = fcoe_tx_i;
4666 #endif /* IXGBE_FCOE */
4668 * ixgbe_cache_ring_sriov - Descriptor ring to register mapping for sriov
4669 * @adapter: board private structure to initialize
4671 * SR-IOV doesn't use any descriptor rings but changes the default if
4672 * no other mapping is used.
4675 static inline bool ixgbe_cache_ring_sriov(struct ixgbe_adapter *adapter)
4677 adapter->rx_ring[0]->reg_idx = adapter->num_vfs * 2;
4678 adapter->tx_ring[0]->reg_idx = adapter->num_vfs * 2;
4679 if (adapter->num_vfs)
4686 * ixgbe_cache_ring_register - Descriptor ring to register mapping
4687 * @adapter: board private structure to initialize
4689 * Once we know the feature-set enabled for the device, we'll cache
4690 * the register offset the descriptor ring is assigned to.
4692 * Note, the order the various feature calls is important. It must start with
4693 * the "most" features enabled at the same time, then trickle down to the
4694 * least amount of features turned on at once.
4696 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
4698 /* start with default case */
4699 adapter->rx_ring[0]->reg_idx = 0;
4700 adapter->tx_ring[0]->reg_idx = 0;
4702 if (ixgbe_cache_ring_sriov(adapter))
4706 if (ixgbe_cache_ring_fcoe(adapter))
4709 #endif /* IXGBE_FCOE */
4710 #ifdef CONFIG_IXGBE_DCB
4711 if (ixgbe_cache_ring_dcb(adapter))
4715 if (ixgbe_cache_ring_fdir(adapter))
4718 if (ixgbe_cache_ring_rss(adapter))
4723 * ixgbe_alloc_queues - Allocate memory for all rings
4724 * @adapter: board private structure to initialize
4726 * We allocate one ring per queue at run-time since we don't know the
4727 * number of queues at compile-time. The polling_netdev array is
4728 * intended for Multiqueue, but should work fine with a single queue.
4730 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
4732 int rx = 0, tx = 0, nid = adapter->node;
4734 if (nid < 0 || !node_online(nid))
4735 nid = first_online_node;
4737 for (; tx < adapter->num_tx_queues; tx++) {
4738 struct ixgbe_ring *ring;
4740 ring = kzalloc_node(sizeof(*ring), GFP_KERNEL, nid);
4742 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
4744 goto err_allocation;
4745 ring->count = adapter->tx_ring_count;
4746 ring->queue_index = tx;
4747 ring->numa_node = nid;
4748 ring->dev = &adapter->pdev->dev;
4749 ring->netdev = adapter->netdev;
4751 adapter->tx_ring[tx] = ring;
4754 for (; rx < adapter->num_rx_queues; rx++) {
4755 struct ixgbe_ring *ring;
4757 ring = kzalloc_node(sizeof(*ring), GFP_KERNEL, nid);
4759 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
4761 goto err_allocation;
4762 ring->count = adapter->rx_ring_count;
4763 ring->queue_index = rx;
4764 ring->numa_node = nid;
4765 ring->dev = &adapter->pdev->dev;
4766 ring->netdev = adapter->netdev;
4768 adapter->rx_ring[rx] = ring;
4771 ixgbe_cache_ring_register(adapter);
4777 kfree(adapter->tx_ring[--tx]);
4780 kfree(adapter->rx_ring[--rx]);
4785 * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
4786 * @adapter: board private structure to initialize
4788 * Attempt to configure the interrupts using the best available
4789 * capabilities of the hardware and the kernel.
4791 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
4793 struct ixgbe_hw *hw = &adapter->hw;
4795 int vector, v_budget;
4798 * It's easy to be greedy for MSI-X vectors, but it really
4799 * doesn't do us much good if we have a lot more vectors
4800 * than CPU's. So let's be conservative and only ask for
4801 * (roughly) the same number of vectors as there are CPU's.
4803 v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
4804 (int)num_online_cpus()) + NON_Q_VECTORS;
4807 * At the same time, hardware can only support a maximum of
4808 * hw.mac->max_msix_vectors vectors. With features
4809 * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
4810 * descriptor queues supported by our device. Thus, we cap it off in
4811 * those rare cases where the cpu count also exceeds our vector limit.
4813 v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
4815 /* A failure in MSI-X entry allocation isn't fatal, but it does
4816 * mean we disable MSI-X capabilities of the adapter. */
4817 adapter->msix_entries = kcalloc(v_budget,
4818 sizeof(struct msix_entry), GFP_KERNEL);
4819 if (adapter->msix_entries) {
4820 for (vector = 0; vector < v_budget; vector++)
4821 adapter->msix_entries[vector].entry = vector;
4823 ixgbe_acquire_msix_vectors(adapter, v_budget);
4825 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4829 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
4830 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
4831 if (adapter->flags & (IXGBE_FLAG_FDIR_HASH_CAPABLE |
4832 IXGBE_FLAG_FDIR_PERFECT_CAPABLE)) {
4834 "Flow Director is not supported while multiple "
4835 "queues are disabled. Disabling Flow Director\n");
4837 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4838 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4839 adapter->atr_sample_rate = 0;
4840 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4841 ixgbe_disable_sriov(adapter);
4843 err = ixgbe_set_num_queues(adapter);
4847 err = pci_enable_msi(adapter->pdev);
4849 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
4851 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4852 "Unable to allocate MSI interrupt, "
4853 "falling back to legacy. Error: %d\n", err);
4863 * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
4864 * @adapter: board private structure to initialize
4866 * We allocate one q_vector per queue interrupt. If allocation fails we
4869 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
4871 int q_idx, num_q_vectors;
4872 struct ixgbe_q_vector *q_vector;
4873 int (*poll)(struct napi_struct *, int);
4875 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4876 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4877 poll = &ixgbe_clean_rxtx_many;
4883 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4884 q_vector = kzalloc_node(sizeof(struct ixgbe_q_vector),
4885 GFP_KERNEL, adapter->node);
4887 q_vector = kzalloc(sizeof(struct ixgbe_q_vector),
4891 q_vector->adapter = adapter;
4892 if (q_vector->txr_count && !q_vector->rxr_count)
4893 q_vector->eitr = adapter->tx_eitr_param;
4895 q_vector->eitr = adapter->rx_eitr_param;
4896 q_vector->v_idx = q_idx;
4897 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
4898 adapter->q_vector[q_idx] = q_vector;
4906 q_vector = adapter->q_vector[q_idx];
4907 netif_napi_del(&q_vector->napi);
4909 adapter->q_vector[q_idx] = NULL;
4915 * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
4916 * @adapter: board private structure to initialize
4918 * This function frees the memory allocated to the q_vectors. In addition if
4919 * NAPI is enabled it will delete any references to the NAPI struct prior
4920 * to freeing the q_vector.
4922 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
4924 int q_idx, num_q_vectors;
4926 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4927 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4931 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4932 struct ixgbe_q_vector *q_vector = adapter->q_vector[q_idx];
4933 adapter->q_vector[q_idx] = NULL;
4934 netif_napi_del(&q_vector->napi);
4939 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
4941 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4942 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4943 pci_disable_msix(adapter->pdev);
4944 kfree(adapter->msix_entries);
4945 adapter->msix_entries = NULL;
4946 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
4947 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
4948 pci_disable_msi(adapter->pdev);
4953 * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
4954 * @adapter: board private structure to initialize
4956 * We determine which interrupt scheme to use based on...
4957 * - Kernel support (MSI, MSI-X)
4958 * - which can be user-defined (via MODULE_PARAM)
4959 * - Hardware queue count (num_*_queues)
4960 * - defined by miscellaneous hardware support/features (RSS, etc.)
4962 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
4966 /* Number of supported queues */
4967 err = ixgbe_set_num_queues(adapter);
4971 err = ixgbe_set_interrupt_capability(adapter);
4973 e_dev_err("Unable to setup interrupt capabilities\n");
4974 goto err_set_interrupt;
4977 err = ixgbe_alloc_q_vectors(adapter);
4979 e_dev_err("Unable to allocate memory for queue vectors\n");
4980 goto err_alloc_q_vectors;
4983 err = ixgbe_alloc_queues(adapter);
4985 e_dev_err("Unable to allocate memory for queues\n");
4986 goto err_alloc_queues;
4989 e_dev_info("Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
4990 (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
4991 adapter->num_rx_queues, adapter->num_tx_queues);
4993 set_bit(__IXGBE_DOWN, &adapter->state);
4998 ixgbe_free_q_vectors(adapter);
4999 err_alloc_q_vectors:
5000 ixgbe_reset_interrupt_capability(adapter);
5005 static void ring_free_rcu(struct rcu_head *head)
5007 kfree(container_of(head, struct ixgbe_ring, rcu));
5011 * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
5012 * @adapter: board private structure to clear interrupt scheme on
5014 * We go through and clear interrupt specific resources and reset the structure
5015 * to pre-load conditions
5017 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
5021 for (i = 0; i < adapter->num_tx_queues; i++) {
5022 kfree(adapter->tx_ring[i]);
5023 adapter->tx_ring[i] = NULL;
5025 for (i = 0; i < adapter->num_rx_queues; i++) {
5026 struct ixgbe_ring *ring = adapter->rx_ring[i];
5028 /* ixgbe_get_stats64() might access this ring, we must wait
5029 * a grace period before freeing it.
5031 call_rcu(&ring->rcu, ring_free_rcu);
5032 adapter->rx_ring[i] = NULL;
5035 adapter->num_tx_queues = 0;
5036 adapter->num_rx_queues = 0;
5038 ixgbe_free_q_vectors(adapter);
5039 ixgbe_reset_interrupt_capability(adapter);
5043 * ixgbe_sfp_timer - worker thread to find a missing module
5044 * @data: pointer to our adapter struct
5046 static void ixgbe_sfp_timer(unsigned long data)
5048 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5051 * Do the sfp_timer outside of interrupt context due to the
5052 * delays that sfp+ detection requires
5054 schedule_work(&adapter->sfp_task);
5058 * ixgbe_sfp_task - worker thread to find a missing module
5059 * @work: pointer to work_struct containing our data
5061 static void ixgbe_sfp_task(struct work_struct *work)
5063 struct ixgbe_adapter *adapter = container_of(work,
5064 struct ixgbe_adapter,
5066 struct ixgbe_hw *hw = &adapter->hw;
5068 if ((hw->phy.type == ixgbe_phy_nl) &&
5069 (hw->phy.sfp_type == ixgbe_sfp_type_not_present)) {
5070 s32 ret = hw->phy.ops.identify_sfp(hw);
5071 if (ret == IXGBE_ERR_SFP_NOT_PRESENT)
5073 ret = hw->phy.ops.reset(hw);
5074 if (ret == IXGBE_ERR_SFP_NOT_SUPPORTED) {
5075 e_dev_err("failed to initialize because an unsupported "
5076 "SFP+ module type was detected.\n");
5077 e_dev_err("Reload the driver after installing a "
5078 "supported module.\n");
5079 unregister_netdev(adapter->netdev);
5081 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
5083 /* don't need this routine any more */
5084 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
5088 if (test_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state))
5089 mod_timer(&adapter->sfp_timer,
5090 round_jiffies(jiffies + (2 * HZ)));
5094 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
5095 * @adapter: board private structure to initialize
5097 * ixgbe_sw_init initializes the Adapter private data structure.
5098 * Fields are initialized based on PCI device information and
5099 * OS network device settings (MTU size).
5101 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
5103 struct ixgbe_hw *hw = &adapter->hw;
5104 struct pci_dev *pdev = adapter->pdev;
5105 struct net_device *dev = adapter->netdev;
5107 #ifdef CONFIG_IXGBE_DCB
5109 struct tc_configuration *tc;
5111 int max_frame = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
5113 /* PCI config space info */
5115 hw->vendor_id = pdev->vendor;
5116 hw->device_id = pdev->device;
5117 hw->revision_id = pdev->revision;
5118 hw->subsystem_vendor_id = pdev->subsystem_vendor;
5119 hw->subsystem_device_id = pdev->subsystem_device;
5121 /* Set capability flags */
5122 rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
5123 adapter->ring_feature[RING_F_RSS].indices = rss;
5124 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
5125 adapter->ring_feature[RING_F_DCB].indices = IXGBE_MAX_DCB_INDICES;
5126 switch (hw->mac.type) {
5127 case ixgbe_mac_82598EB:
5128 if (hw->device_id == IXGBE_DEV_ID_82598AT)
5129 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
5130 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
5132 case ixgbe_mac_82599EB:
5133 case ixgbe_mac_X540:
5134 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
5135 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
5136 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
5137 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
5138 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
5139 /* n-tuple support exists, always init our spinlock */
5140 spin_lock_init(&adapter->fdir_perfect_lock);
5141 /* Flow Director hash filters enabled */
5142 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
5143 adapter->atr_sample_rate = 20;
5144 adapter->ring_feature[RING_F_FDIR].indices =
5145 IXGBE_MAX_FDIR_INDICES;
5146 adapter->fdir_pballoc = 0;
5148 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
5149 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
5150 adapter->ring_feature[RING_F_FCOE].indices = 0;
5151 #ifdef CONFIG_IXGBE_DCB
5152 /* Default traffic class to use for FCoE */
5153 adapter->fcoe.tc = IXGBE_FCOE_DEFTC;
5154 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
5156 #endif /* IXGBE_FCOE */
5162 #ifdef CONFIG_IXGBE_DCB
5163 /* Configure DCB traffic classes */
5164 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
5165 tc = &adapter->dcb_cfg.tc_config[j];
5166 tc->path[DCB_TX_CONFIG].bwg_id = 0;
5167 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
5168 tc->path[DCB_RX_CONFIG].bwg_id = 0;
5169 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
5170 tc->dcb_pfc = pfc_disabled;
5172 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
5173 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
5174 adapter->dcb_cfg.rx_pba_cfg = pba_equal;
5175 adapter->dcb_cfg.pfc_mode_enable = false;
5176 adapter->dcb_cfg.round_robin_enable = false;
5177 adapter->dcb_set_bitmap = 0x00;
5178 ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
5179 adapter->ring_feature[RING_F_DCB].indices);
5183 /* default flow control settings */
5184 hw->fc.requested_mode = ixgbe_fc_full;
5185 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
5187 adapter->last_lfc_mode = hw->fc.current_mode;
5189 hw->fc.high_water = FC_HIGH_WATER(max_frame);
5190 hw->fc.low_water = FC_LOW_WATER(max_frame);
5191 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
5192 hw->fc.send_xon = true;
5193 hw->fc.disable_fc_autoneg = false;
5195 /* enable itr by default in dynamic mode */
5196 adapter->rx_itr_setting = 1;
5197 adapter->rx_eitr_param = 20000;
5198 adapter->tx_itr_setting = 1;
5199 adapter->tx_eitr_param = 10000;
5201 /* set defaults for eitr in MegaBytes */
5202 adapter->eitr_low = 10;
5203 adapter->eitr_high = 20;
5205 /* set default ring sizes */
5206 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
5207 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
5209 /* initialize eeprom parameters */
5210 if (ixgbe_init_eeprom_params_generic(hw)) {
5211 e_dev_err("EEPROM initialization failed\n");
5215 /* enable rx csum by default */
5216 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
5218 /* get assigned NUMA node */
5219 adapter->node = dev_to_node(&pdev->dev);
5221 set_bit(__IXGBE_DOWN, &adapter->state);
5227 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
5228 * @tx_ring: tx descriptor ring (for a specific queue) to setup
5230 * Return 0 on success, negative on failure
5232 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
5234 struct device *dev = tx_ring->dev;
5237 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
5238 tx_ring->tx_buffer_info = vzalloc_node(size, tx_ring->numa_node);
5239 if (!tx_ring->tx_buffer_info)
5240 tx_ring->tx_buffer_info = vzalloc(size);
5241 if (!tx_ring->tx_buffer_info)
5244 /* round up to nearest 4K */
5245 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
5246 tx_ring->size = ALIGN(tx_ring->size, 4096);
5248 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
5249 &tx_ring->dma, GFP_KERNEL);
5253 tx_ring->next_to_use = 0;
5254 tx_ring->next_to_clean = 0;
5255 tx_ring->work_limit = tx_ring->count;
5259 vfree(tx_ring->tx_buffer_info);
5260 tx_ring->tx_buffer_info = NULL;
5261 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
5266 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
5267 * @adapter: board private structure
5269 * If this function returns with an error, then it's possible one or
5270 * more of the rings is populated (while the rest are not). It is the
5271 * callers duty to clean those orphaned rings.
5273 * Return 0 on success, negative on failure
5275 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
5279 for (i = 0; i < adapter->num_tx_queues; i++) {
5280 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
5283 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
5291 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
5292 * @rx_ring: rx descriptor ring (for a specific queue) to setup
5294 * Returns 0 on success, negative on failure
5296 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
5298 struct device *dev = rx_ring->dev;
5301 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
5302 rx_ring->rx_buffer_info = vzalloc_node(size, rx_ring->numa_node);
5303 if (!rx_ring->rx_buffer_info)
5304 rx_ring->rx_buffer_info = vzalloc(size);
5305 if (!rx_ring->rx_buffer_info)
5308 /* Round up to nearest 4K */
5309 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
5310 rx_ring->size = ALIGN(rx_ring->size, 4096);
5312 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
5313 &rx_ring->dma, GFP_KERNEL);
5318 rx_ring->next_to_clean = 0;
5319 rx_ring->next_to_use = 0;
5323 vfree(rx_ring->rx_buffer_info);
5324 rx_ring->rx_buffer_info = NULL;
5325 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
5330 * ixgbe_setup_all_rx_resources - allocate all queues Rx 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_rx_resources(struct ixgbe_adapter *adapter)
5343 for (i = 0; i < adapter->num_rx_queues; i++) {
5344 err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
5347 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
5355 * ixgbe_free_tx_resources - Free Tx Resources per Queue
5356 * @tx_ring: Tx descriptor ring for a specific queue
5358 * Free all transmit software resources
5360 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
5362 ixgbe_clean_tx_ring(tx_ring);
5364 vfree(tx_ring->tx_buffer_info);
5365 tx_ring->tx_buffer_info = NULL;
5367 /* if not set, then don't free */
5371 dma_free_coherent(tx_ring->dev, tx_ring->size,
5372 tx_ring->desc, tx_ring->dma);
5374 tx_ring->desc = NULL;
5378 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5379 * @adapter: board private structure
5381 * Free all transmit software resources
5383 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
5387 for (i = 0; i < adapter->num_tx_queues; i++)
5388 if (adapter->tx_ring[i]->desc)
5389 ixgbe_free_tx_resources(adapter->tx_ring[i]);
5393 * ixgbe_free_rx_resources - Free Rx Resources
5394 * @rx_ring: ring to clean the resources from
5396 * Free all receive software resources
5398 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
5400 ixgbe_clean_rx_ring(rx_ring);
5402 vfree(rx_ring->rx_buffer_info);
5403 rx_ring->rx_buffer_info = NULL;
5405 /* if not set, then don't free */
5409 dma_free_coherent(rx_ring->dev, rx_ring->size,
5410 rx_ring->desc, rx_ring->dma);
5412 rx_ring->desc = NULL;
5416 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5417 * @adapter: board private structure
5419 * Free all receive software resources
5421 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5425 for (i = 0; i < adapter->num_rx_queues; i++)
5426 if (adapter->rx_ring[i]->desc)
5427 ixgbe_free_rx_resources(adapter->rx_ring[i]);
5431 * ixgbe_change_mtu - Change the Maximum Transfer Unit
5432 * @netdev: network interface device structure
5433 * @new_mtu: new value for maximum frame size
5435 * Returns 0 on success, negative on failure
5437 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5439 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5440 struct ixgbe_hw *hw = &adapter->hw;
5441 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5443 /* MTU < 68 is an error and causes problems on some kernels */
5444 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5447 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5448 /* must set new MTU before calling down or up */
5449 netdev->mtu = new_mtu;
5451 hw->fc.high_water = FC_HIGH_WATER(max_frame);
5452 hw->fc.low_water = FC_LOW_WATER(max_frame);
5454 if (netif_running(netdev))
5455 ixgbe_reinit_locked(adapter);
5461 * ixgbe_open - Called when a network interface is made active
5462 * @netdev: network interface device structure
5464 * Returns 0 on success, negative value on failure
5466 * The open entry point is called when a network interface is made
5467 * active by the system (IFF_UP). At this point all resources needed
5468 * for transmit and receive operations are allocated, the interrupt
5469 * handler is registered with the OS, the watchdog timer is started,
5470 * and the stack is notified that the interface is ready.
5472 static int ixgbe_open(struct net_device *netdev)
5474 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5477 /* disallow open during test */
5478 if (test_bit(__IXGBE_TESTING, &adapter->state))
5481 netif_carrier_off(netdev);
5483 /* allocate transmit descriptors */
5484 err = ixgbe_setup_all_tx_resources(adapter);
5488 /* allocate receive descriptors */
5489 err = ixgbe_setup_all_rx_resources(adapter);
5493 ixgbe_configure(adapter);
5495 err = ixgbe_request_irq(adapter);
5499 err = ixgbe_up_complete(adapter);
5503 netif_tx_start_all_queues(netdev);
5508 ixgbe_release_hw_control(adapter);
5509 ixgbe_free_irq(adapter);
5512 ixgbe_free_all_rx_resources(adapter);
5514 ixgbe_free_all_tx_resources(adapter);
5515 ixgbe_reset(adapter);
5521 * ixgbe_close - Disables a network interface
5522 * @netdev: network interface device structure
5524 * Returns 0, this is not allowed to fail
5526 * The close entry point is called when an interface is de-activated
5527 * by the OS. The hardware is still under the drivers control, but
5528 * needs to be disabled. A global MAC reset is issued to stop the
5529 * hardware, and all transmit and receive resources are freed.
5531 static int ixgbe_close(struct net_device *netdev)
5533 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5535 ixgbe_down(adapter);
5536 ixgbe_free_irq(adapter);
5538 ixgbe_free_all_tx_resources(adapter);
5539 ixgbe_free_all_rx_resources(adapter);
5541 ixgbe_release_hw_control(adapter);
5547 static int ixgbe_resume(struct pci_dev *pdev)
5549 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5550 struct net_device *netdev = adapter->netdev;
5553 pci_set_power_state(pdev, PCI_D0);
5554 pci_restore_state(pdev);
5556 * pci_restore_state clears dev->state_saved so call
5557 * pci_save_state to restore it.
5559 pci_save_state(pdev);
5561 err = pci_enable_device_mem(pdev);
5563 e_dev_err("Cannot enable PCI device from suspend\n");
5566 pci_set_master(pdev);
5568 pci_wake_from_d3(pdev, false);
5570 err = ixgbe_init_interrupt_scheme(adapter);
5572 e_dev_err("Cannot initialize interrupts for device\n");
5576 ixgbe_reset(adapter);
5578 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5580 if (netif_running(netdev)) {
5581 err = ixgbe_open(netdev);
5586 netif_device_attach(netdev);
5590 #endif /* CONFIG_PM */
5592 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5594 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5595 struct net_device *netdev = adapter->netdev;
5596 struct ixgbe_hw *hw = &adapter->hw;
5598 u32 wufc = adapter->wol;
5603 netif_device_detach(netdev);
5605 if (netif_running(netdev)) {
5606 ixgbe_down(adapter);
5607 ixgbe_free_irq(adapter);
5608 ixgbe_free_all_tx_resources(adapter);
5609 ixgbe_free_all_rx_resources(adapter);
5612 ixgbe_clear_interrupt_scheme(adapter);
5615 retval = pci_save_state(pdev);
5621 ixgbe_set_rx_mode(netdev);
5623 /* turn on all-multi mode if wake on multicast is enabled */
5624 if (wufc & IXGBE_WUFC_MC) {
5625 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5626 fctrl |= IXGBE_FCTRL_MPE;
5627 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5630 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5631 ctrl |= IXGBE_CTRL_GIO_DIS;
5632 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5634 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5636 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5637 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5640 switch (hw->mac.type) {
5641 case ixgbe_mac_82598EB:
5642 pci_wake_from_d3(pdev, false);
5644 case ixgbe_mac_82599EB:
5645 case ixgbe_mac_X540:
5646 pci_wake_from_d3(pdev, !!wufc);
5652 *enable_wake = !!wufc;
5654 ixgbe_release_hw_control(adapter);
5656 pci_disable_device(pdev);
5662 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5667 retval = __ixgbe_shutdown(pdev, &wake);
5672 pci_prepare_to_sleep(pdev);
5674 pci_wake_from_d3(pdev, false);
5675 pci_set_power_state(pdev, PCI_D3hot);
5680 #endif /* CONFIG_PM */
5682 static void ixgbe_shutdown(struct pci_dev *pdev)
5686 __ixgbe_shutdown(pdev, &wake);
5688 if (system_state == SYSTEM_POWER_OFF) {
5689 pci_wake_from_d3(pdev, wake);
5690 pci_set_power_state(pdev, PCI_D3hot);
5695 * ixgbe_update_stats - Update the board statistics counters.
5696 * @adapter: board private structure
5698 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5700 struct net_device *netdev = adapter->netdev;
5701 struct ixgbe_hw *hw = &adapter->hw;
5702 struct ixgbe_hw_stats *hwstats = &adapter->stats;
5704 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5705 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
5706 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
5707 u64 bytes = 0, packets = 0;
5709 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5710 test_bit(__IXGBE_RESETTING, &adapter->state))
5713 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5716 for (i = 0; i < 16; i++)
5717 adapter->hw_rx_no_dma_resources +=
5718 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5719 for (i = 0; i < adapter->num_rx_queues; i++) {
5720 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
5721 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
5723 adapter->rsc_total_count = rsc_count;
5724 adapter->rsc_total_flush = rsc_flush;
5727 for (i = 0; i < adapter->num_rx_queues; i++) {
5728 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
5729 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
5730 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
5731 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
5732 bytes += rx_ring->stats.bytes;
5733 packets += rx_ring->stats.packets;
5735 adapter->non_eop_descs = non_eop_descs;
5736 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5737 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5738 netdev->stats.rx_bytes = bytes;
5739 netdev->stats.rx_packets = packets;
5743 /* gather some stats to the adapter struct that are per queue */
5744 for (i = 0; i < adapter->num_tx_queues; i++) {
5745 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5746 restart_queue += tx_ring->tx_stats.restart_queue;
5747 tx_busy += tx_ring->tx_stats.tx_busy;
5748 bytes += tx_ring->stats.bytes;
5749 packets += tx_ring->stats.packets;
5751 adapter->restart_queue = restart_queue;
5752 adapter->tx_busy = tx_busy;
5753 netdev->stats.tx_bytes = bytes;
5754 netdev->stats.tx_packets = packets;
5756 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5757 for (i = 0; i < 8; i++) {
5758 /* for packet buffers not used, the register should read 0 */
5759 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5761 hwstats->mpc[i] += mpc;
5762 total_mpc += hwstats->mpc[i];
5763 if (hw->mac.type == ixgbe_mac_82598EB)
5764 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5765 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5766 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5767 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5768 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5769 switch (hw->mac.type) {
5770 case ixgbe_mac_82598EB:
5771 hwstats->pxonrxc[i] +=
5772 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5774 case ixgbe_mac_82599EB:
5775 case ixgbe_mac_X540:
5776 hwstats->pxonrxc[i] +=
5777 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5782 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5783 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5785 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5786 /* work around hardware counting issue */
5787 hwstats->gprc -= missed_rx;
5789 ixgbe_update_xoff_received(adapter);
5791 /* 82598 hardware only has a 32 bit counter in the high register */
5792 switch (hw->mac.type) {
5793 case ixgbe_mac_82598EB:
5794 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5795 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5796 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5797 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5799 case ixgbe_mac_82599EB:
5800 case ixgbe_mac_X540:
5801 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5802 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
5803 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5804 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
5805 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5806 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5807 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5808 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5809 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5811 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5812 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5813 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5814 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5815 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5816 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5817 #endif /* IXGBE_FCOE */
5822 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5823 hwstats->bprc += bprc;
5824 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5825 if (hw->mac.type == ixgbe_mac_82598EB)
5826 hwstats->mprc -= bprc;
5827 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5828 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5829 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5830 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5831 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5832 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5833 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5834 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5835 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5836 hwstats->lxontxc += lxon;
5837 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5838 hwstats->lxofftxc += lxoff;
5839 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5840 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5841 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5843 * 82598 errata - tx of flow control packets is included in tx counters
5845 xon_off_tot = lxon + lxoff;
5846 hwstats->gptc -= xon_off_tot;
5847 hwstats->mptc -= xon_off_tot;
5848 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5849 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5850 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5851 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5852 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5853 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5854 hwstats->ptc64 -= xon_off_tot;
5855 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5856 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5857 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5858 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5859 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5860 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5862 /* Fill out the OS statistics structure */
5863 netdev->stats.multicast = hwstats->mprc;
5866 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
5867 netdev->stats.rx_dropped = 0;
5868 netdev->stats.rx_length_errors = hwstats->rlec;
5869 netdev->stats.rx_crc_errors = hwstats->crcerrs;
5870 netdev->stats.rx_missed_errors = total_mpc;
5874 * ixgbe_watchdog - Timer Call-back
5875 * @data: pointer to adapter cast into an unsigned long
5877 static void ixgbe_watchdog(unsigned long data)
5879 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5880 struct ixgbe_hw *hw = &adapter->hw;
5885 * Do the watchdog outside of interrupt context due to the lovely
5886 * delays that some of the newer hardware requires
5889 if (test_bit(__IXGBE_DOWN, &adapter->state))
5890 goto watchdog_short_circuit;
5892 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5894 * for legacy and MSI interrupts don't set any bits
5895 * that are enabled for EIAM, because this operation
5896 * would set *both* EIMS and EICS for any bit in EIAM
5898 IXGBE_WRITE_REG(hw, IXGBE_EICS,
5899 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5900 goto watchdog_reschedule;
5903 /* get one bit for every active tx/rx interrupt vector */
5904 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
5905 struct ixgbe_q_vector *qv = adapter->q_vector[i];
5906 if (qv->rxr_count || qv->txr_count)
5907 eics |= ((u64)1 << i);
5910 /* Cause software interrupt to ensure rx rings are cleaned */
5911 ixgbe_irq_rearm_queues(adapter, eics);
5913 watchdog_reschedule:
5914 /* Reset the timer */
5915 mod_timer(&adapter->watchdog_timer, round_jiffies(jiffies + 2 * HZ));
5917 watchdog_short_circuit:
5918 schedule_work(&adapter->watchdog_task);
5922 * ixgbe_multispeed_fiber_task - worker thread to configure multispeed fiber
5923 * @work: pointer to work_struct containing our data
5925 static void ixgbe_multispeed_fiber_task(struct work_struct *work)
5927 struct ixgbe_adapter *adapter = container_of(work,
5928 struct ixgbe_adapter,
5929 multispeed_fiber_task);
5930 struct ixgbe_hw *hw = &adapter->hw;
5934 adapter->flags |= IXGBE_FLAG_IN_SFP_LINK_TASK;
5935 autoneg = hw->phy.autoneg_advertised;
5936 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
5937 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
5938 hw->mac.autotry_restart = false;
5939 if (hw->mac.ops.setup_link)
5940 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
5941 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5942 adapter->flags &= ~IXGBE_FLAG_IN_SFP_LINK_TASK;
5946 * ixgbe_sfp_config_module_task - worker thread to configure a new SFP+ module
5947 * @work: pointer to work_struct containing our data
5949 static void ixgbe_sfp_config_module_task(struct work_struct *work)
5951 struct ixgbe_adapter *adapter = container_of(work,
5952 struct ixgbe_adapter,
5953 sfp_config_module_task);
5954 struct ixgbe_hw *hw = &adapter->hw;
5957 adapter->flags |= IXGBE_FLAG_IN_SFP_MOD_TASK;
5959 /* Time for electrical oscillations to settle down */
5961 err = hw->phy.ops.identify_sfp(hw);
5963 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
5964 e_dev_err("failed to initialize because an unsupported SFP+ "
5965 "module type was detected.\n");
5966 e_dev_err("Reload the driver after installing a supported "
5968 unregister_netdev(adapter->netdev);
5971 hw->mac.ops.setup_sfp(hw);
5973 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
5974 /* This will also work for DA Twinax connections */
5975 schedule_work(&adapter->multispeed_fiber_task);
5976 adapter->flags &= ~IXGBE_FLAG_IN_SFP_MOD_TASK;
5980 * ixgbe_fdir_reinit_task - worker thread to reinit FDIR filter table
5981 * @work: pointer to work_struct containing our data
5983 static void ixgbe_fdir_reinit_task(struct work_struct *work)
5985 struct ixgbe_adapter *adapter = container_of(work,
5986 struct ixgbe_adapter,
5988 struct ixgbe_hw *hw = &adapter->hw;
5991 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5992 for (i = 0; i < adapter->num_tx_queues; i++)
5993 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
5994 &(adapter->tx_ring[i]->state));
5996 e_err(probe, "failed to finish FDIR re-initialization, "
5997 "ignored adding FDIR ATR filters\n");
5999 /* Done FDIR Re-initialization, enable transmits */
6000 netif_tx_start_all_queues(adapter->netdev);
6003 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
6007 /* Do not perform spoof check for 82598 */
6008 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
6011 ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
6014 * ssvpc register is cleared on read, if zero then no
6015 * spoofed packets in the last interval.
6020 e_warn(drv, "%d Spoofed packets detected\n", ssvpc);
6023 static DEFINE_MUTEX(ixgbe_watchdog_lock);
6026 * ixgbe_watchdog_task - worker thread to bring link up
6027 * @work: pointer to work_struct containing our data
6029 static void ixgbe_watchdog_task(struct work_struct *work)
6031 struct ixgbe_adapter *adapter = container_of(work,
6032 struct ixgbe_adapter,
6034 struct net_device *netdev = adapter->netdev;
6035 struct ixgbe_hw *hw = &adapter->hw;
6039 struct ixgbe_ring *tx_ring;
6040 int some_tx_pending = 0;
6042 mutex_lock(&ixgbe_watchdog_lock);
6044 link_up = adapter->link_up;
6045 link_speed = adapter->link_speed;
6047 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
6048 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
6051 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6052 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
6053 hw->mac.ops.fc_enable(hw, i);
6055 hw->mac.ops.fc_enable(hw, 0);
6058 hw->mac.ops.fc_enable(hw, 0);
6063 time_after(jiffies, (adapter->link_check_timeout +
6064 IXGBE_TRY_LINK_TIMEOUT))) {
6065 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
6066 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
6068 adapter->link_up = link_up;
6069 adapter->link_speed = link_speed;
6073 if (!netif_carrier_ok(netdev)) {
6074 bool flow_rx, flow_tx;
6076 switch (hw->mac.type) {
6077 case ixgbe_mac_82598EB: {
6078 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
6079 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
6080 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
6081 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
6084 case ixgbe_mac_82599EB:
6085 case ixgbe_mac_X540: {
6086 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
6087 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
6088 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
6089 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
6098 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
6099 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
6101 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
6102 "1 Gbps" : "unknown speed")),
6103 ((flow_rx && flow_tx) ? "RX/TX" :
6105 (flow_tx ? "TX" : "None"))));
6107 netif_carrier_on(netdev);
6109 /* Force detection of hung controller */
6110 for (i = 0; i < adapter->num_tx_queues; i++) {
6111 tx_ring = adapter->tx_ring[i];
6112 set_check_for_tx_hang(tx_ring);
6116 adapter->link_up = false;
6117 adapter->link_speed = 0;
6118 if (netif_carrier_ok(netdev)) {
6119 e_info(drv, "NIC Link is Down\n");
6120 netif_carrier_off(netdev);
6124 if (!netif_carrier_ok(netdev)) {
6125 for (i = 0; i < adapter->num_tx_queues; i++) {
6126 tx_ring = adapter->tx_ring[i];
6127 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
6128 some_tx_pending = 1;
6133 if (some_tx_pending) {
6134 /* We've lost link, so the controller stops DMA,
6135 * but we've got queued Tx work that's never going
6136 * to get done, so reset controller to flush Tx.
6137 * (Do the reset outside of interrupt context).
6139 schedule_work(&adapter->reset_task);
6143 ixgbe_spoof_check(adapter);
6144 ixgbe_update_stats(adapter);
6145 mutex_unlock(&ixgbe_watchdog_lock);
6148 static int ixgbe_tso(struct ixgbe_adapter *adapter,
6149 struct ixgbe_ring *tx_ring, struct sk_buff *skb,
6150 u32 tx_flags, u8 *hdr_len, __be16 protocol)
6152 struct ixgbe_adv_tx_context_desc *context_desc;
6155 struct ixgbe_tx_buffer *tx_buffer_info;
6156 u32 vlan_macip_lens = 0, type_tucmd_mlhl;
6157 u32 mss_l4len_idx, l4len;
6159 if (skb_is_gso(skb)) {
6160 if (skb_header_cloned(skb)) {
6161 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
6165 l4len = tcp_hdrlen(skb);
6168 if (protocol == htons(ETH_P_IP)) {
6169 struct iphdr *iph = ip_hdr(skb);
6172 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
6176 } else if (skb_is_gso_v6(skb)) {
6177 ipv6_hdr(skb)->payload_len = 0;
6178 tcp_hdr(skb)->check =
6179 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
6180 &ipv6_hdr(skb)->daddr,
6184 i = tx_ring->next_to_use;
6186 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6187 context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
6189 /* VLAN MACLEN IPLEN */
6190 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
6192 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
6193 vlan_macip_lens |= ((skb_network_offset(skb)) <<
6194 IXGBE_ADVTXD_MACLEN_SHIFT);
6195 *hdr_len += skb_network_offset(skb);
6197 (skb_transport_header(skb) - skb_network_header(skb));
6199 (skb_transport_header(skb) - skb_network_header(skb));
6200 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
6201 context_desc->seqnum_seed = 0;
6203 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
6204 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
6205 IXGBE_ADVTXD_DTYP_CTXT);
6207 if (protocol == htons(ETH_P_IP))
6208 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
6209 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6210 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
6214 (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
6215 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
6216 /* use index 1 for TSO */
6217 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6218 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
6220 tx_buffer_info->time_stamp = jiffies;
6221 tx_buffer_info->next_to_watch = i;
6224 if (i == tx_ring->count)
6226 tx_ring->next_to_use = i;
6233 static u32 ixgbe_psum(struct ixgbe_adapter *adapter, struct sk_buff *skb,
6239 case cpu_to_be16(ETH_P_IP):
6240 rtn |= IXGBE_ADVTXD_TUCMD_IPV4;
6241 switch (ip_hdr(skb)->protocol) {
6243 rtn |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6246 rtn |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
6250 case cpu_to_be16(ETH_P_IPV6):
6251 /* XXX what about other V6 headers?? */
6252 switch (ipv6_hdr(skb)->nexthdr) {
6254 rtn |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6257 rtn |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
6262 if (unlikely(net_ratelimit()))
6263 e_warn(probe, "partial checksum but proto=%x!\n",
6271 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
6272 struct ixgbe_ring *tx_ring,
6273 struct sk_buff *skb, u32 tx_flags,
6276 struct ixgbe_adv_tx_context_desc *context_desc;
6278 struct ixgbe_tx_buffer *tx_buffer_info;
6279 u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
6281 if (skb->ip_summed == CHECKSUM_PARTIAL ||
6282 (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
6283 i = tx_ring->next_to_use;
6284 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6285 context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
6287 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
6289 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
6290 vlan_macip_lens |= (skb_network_offset(skb) <<
6291 IXGBE_ADVTXD_MACLEN_SHIFT);
6292 if (skb->ip_summed == CHECKSUM_PARTIAL)
6293 vlan_macip_lens |= (skb_transport_header(skb) -
6294 skb_network_header(skb));
6296 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
6297 context_desc->seqnum_seed = 0;
6299 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
6300 IXGBE_ADVTXD_DTYP_CTXT);
6302 if (skb->ip_summed == CHECKSUM_PARTIAL)
6303 type_tucmd_mlhl |= ixgbe_psum(adapter, skb, protocol);
6305 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
6306 /* use index zero for tx checksum offload */
6307 context_desc->mss_l4len_idx = 0;
6309 tx_buffer_info->time_stamp = jiffies;
6310 tx_buffer_info->next_to_watch = i;
6313 if (i == tx_ring->count)
6315 tx_ring->next_to_use = i;
6323 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
6324 struct ixgbe_ring *tx_ring,
6325 struct sk_buff *skb, u32 tx_flags,
6326 unsigned int first, const u8 hdr_len)
6328 struct device *dev = tx_ring->dev;
6329 struct ixgbe_tx_buffer *tx_buffer_info;
6331 unsigned int total = skb->len;
6332 unsigned int offset = 0, size, count = 0, i;
6333 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
6335 unsigned int bytecount = skb->len;
6338 i = tx_ring->next_to_use;
6340 if (tx_flags & IXGBE_TX_FLAGS_FCOE)
6341 /* excluding fcoe_crc_eof for FCoE */
6342 total -= sizeof(struct fcoe_crc_eof);
6344 len = min(skb_headlen(skb), total);
6346 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6347 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
6349 tx_buffer_info->length = size;
6350 tx_buffer_info->mapped_as_page = false;
6351 tx_buffer_info->dma = dma_map_single(dev,
6353 size, DMA_TO_DEVICE);
6354 if (dma_mapping_error(dev, tx_buffer_info->dma))
6356 tx_buffer_info->time_stamp = jiffies;
6357 tx_buffer_info->next_to_watch = i;
6366 if (i == tx_ring->count)
6371 for (f = 0; f < nr_frags; f++) {
6372 struct skb_frag_struct *frag;
6374 frag = &skb_shinfo(skb)->frags[f];
6375 len = min((unsigned int)frag->size, total);
6376 offset = frag->page_offset;
6380 if (i == tx_ring->count)
6383 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6384 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
6386 tx_buffer_info->length = size;
6387 tx_buffer_info->dma = dma_map_page(dev,
6391 tx_buffer_info->mapped_as_page = true;
6392 if (dma_mapping_error(dev, tx_buffer_info->dma))
6394 tx_buffer_info->time_stamp = jiffies;
6395 tx_buffer_info->next_to_watch = i;
6406 if (tx_flags & IXGBE_TX_FLAGS_TSO)
6407 gso_segs = skb_shinfo(skb)->gso_segs;
6409 /* adjust for FCoE Sequence Offload */
6410 else if (tx_flags & IXGBE_TX_FLAGS_FSO)
6411 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
6412 skb_shinfo(skb)->gso_size);
6413 #endif /* IXGBE_FCOE */
6414 bytecount += (gso_segs - 1) * hdr_len;
6416 /* multiply data chunks by size of headers */
6417 tx_ring->tx_buffer_info[i].bytecount = bytecount;
6418 tx_ring->tx_buffer_info[i].gso_segs = gso_segs;
6419 tx_ring->tx_buffer_info[i].skb = skb;
6420 tx_ring->tx_buffer_info[first].next_to_watch = i;
6425 e_dev_err("TX DMA map failed\n");
6427 /* clear timestamp and dma mappings for failed tx_buffer_info map */
6428 tx_buffer_info->dma = 0;
6429 tx_buffer_info->time_stamp = 0;
6430 tx_buffer_info->next_to_watch = 0;
6434 /* clear timestamp and dma mappings for remaining portion of packet */
6437 i += tx_ring->count;
6439 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6440 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
6446 static void ixgbe_tx_queue(struct ixgbe_ring *tx_ring,
6447 int tx_flags, int count, u32 paylen, u8 hdr_len)
6449 union ixgbe_adv_tx_desc *tx_desc = NULL;
6450 struct ixgbe_tx_buffer *tx_buffer_info;
6451 u32 olinfo_status = 0, cmd_type_len = 0;
6453 u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
6455 cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
6457 cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
6459 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
6460 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
6462 if (tx_flags & IXGBE_TX_FLAGS_TSO) {
6463 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
6465 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
6466 IXGBE_ADVTXD_POPTS_SHIFT;
6468 /* use index 1 context for tso */
6469 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6470 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
6471 olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
6472 IXGBE_ADVTXD_POPTS_SHIFT;
6474 } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
6475 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
6476 IXGBE_ADVTXD_POPTS_SHIFT;
6478 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6479 olinfo_status |= IXGBE_ADVTXD_CC;
6480 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6481 if (tx_flags & IXGBE_TX_FLAGS_FSO)
6482 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
6485 olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
6487 i = tx_ring->next_to_use;
6489 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6490 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
6491 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
6492 tx_desc->read.cmd_type_len =
6493 cpu_to_le32(cmd_type_len | tx_buffer_info->length);
6494 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
6496 if (i == tx_ring->count)
6500 tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
6503 * Force memory writes to complete before letting h/w
6504 * know there are new descriptors to fetch. (Only
6505 * applicable for weak-ordered memory model archs,
6510 tx_ring->next_to_use = i;
6511 writel(i, tx_ring->tail);
6514 static void ixgbe_atr(struct ixgbe_ring *ring, struct sk_buff *skb,
6515 u32 tx_flags, __be16 protocol)
6517 struct ixgbe_q_vector *q_vector = ring->q_vector;
6518 union ixgbe_atr_hash_dword input = { .dword = 0 };
6519 union ixgbe_atr_hash_dword common = { .dword = 0 };
6521 unsigned char *network;
6523 struct ipv6hdr *ipv6;
6528 /* if ring doesn't have a interrupt vector, cannot perform ATR */
6532 /* do nothing if sampling is disabled */
6533 if (!ring->atr_sample_rate)
6538 /* snag network header to get L4 type and address */
6539 hdr.network = skb_network_header(skb);
6541 /* Currently only IPv4/IPv6 with TCP is supported */
6542 if ((protocol != __constant_htons(ETH_P_IPV6) ||
6543 hdr.ipv6->nexthdr != IPPROTO_TCP) &&
6544 (protocol != __constant_htons(ETH_P_IP) ||
6545 hdr.ipv4->protocol != IPPROTO_TCP))
6550 /* skip this packet since the socket is closing */
6554 /* sample on all syn packets or once every atr sample count */
6555 if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
6558 /* reset sample count */
6559 ring->atr_count = 0;
6561 vlan_id = htons(tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
6564 * src and dst are inverted, think how the receiver sees them
6566 * The input is broken into two sections, a non-compressed section
6567 * containing vm_pool, vlan_id, and flow_type. The rest of the data
6568 * is XORed together and stored in the compressed dword.
6570 input.formatted.vlan_id = vlan_id;
6573 * since src port and flex bytes occupy the same word XOR them together
6574 * and write the value to source port portion of compressed dword
6577 common.port.src ^= th->dest ^ __constant_htons(ETH_P_8021Q);
6579 common.port.src ^= th->dest ^ protocol;
6580 common.port.dst ^= th->source;
6582 if (protocol == __constant_htons(ETH_P_IP)) {
6583 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
6584 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
6586 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
6587 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
6588 hdr.ipv6->saddr.s6_addr32[1] ^
6589 hdr.ipv6->saddr.s6_addr32[2] ^
6590 hdr.ipv6->saddr.s6_addr32[3] ^
6591 hdr.ipv6->daddr.s6_addr32[0] ^
6592 hdr.ipv6->daddr.s6_addr32[1] ^
6593 hdr.ipv6->daddr.s6_addr32[2] ^
6594 hdr.ipv6->daddr.s6_addr32[3];
6597 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
6598 ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
6599 input, common, ring->queue_index);
6602 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, int size)
6604 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
6605 /* Herbert's original patch had:
6606 * smp_mb__after_netif_stop_queue();
6607 * but since that doesn't exist yet, just open code it. */
6610 /* We need to check again in a case another CPU has just
6611 * made room available. */
6612 if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
6615 /* A reprieve! - use start_queue because it doesn't call schedule */
6616 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
6617 ++tx_ring->tx_stats.restart_queue;
6621 static int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, int size)
6623 if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
6625 return __ixgbe_maybe_stop_tx(tx_ring, size);
6628 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6630 struct ixgbe_adapter *adapter = netdev_priv(dev);
6631 int txq = smp_processor_id();
6635 protocol = vlan_get_protocol(skb);
6637 if ((protocol == htons(ETH_P_FCOE)) ||
6638 (protocol == htons(ETH_P_FIP))) {
6639 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
6640 txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
6641 txq += adapter->ring_feature[RING_F_FCOE].mask;
6643 #ifdef CONFIG_IXGBE_DCB
6644 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6645 txq = adapter->fcoe.up;
6652 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6653 while (unlikely(txq >= dev->real_num_tx_queues))
6654 txq -= dev->real_num_tx_queues;
6658 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6659 if (skb->priority == TC_PRIO_CONTROL)
6660 txq = adapter->ring_feature[RING_F_DCB].indices-1;
6662 txq = (skb->vlan_tci & IXGBE_TX_FLAGS_VLAN_PRIO_MASK)
6667 return skb_tx_hash(dev, skb);
6670 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
6671 struct ixgbe_adapter *adapter,
6672 struct ixgbe_ring *tx_ring)
6675 unsigned int tx_flags = 0;
6682 protocol = vlan_get_protocol(skb);
6684 if (vlan_tx_tag_present(skb)) {
6685 tx_flags |= vlan_tx_tag_get(skb);
6686 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6687 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6688 tx_flags |= ((skb->queue_mapping & 0x7) << 13);
6690 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6691 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6692 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED &&
6693 skb->priority != TC_PRIO_CONTROL) {
6694 tx_flags |= ((skb->queue_mapping & 0x7) << 13);
6695 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6696 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6700 /* for FCoE with DCB, we force the priority to what
6701 * was specified by the switch */
6702 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED &&
6703 (protocol == htons(ETH_P_FCOE) ||
6704 protocol == htons(ETH_P_FIP))) {
6705 #ifdef CONFIG_IXGBE_DCB
6706 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6707 tx_flags &= ~(IXGBE_TX_FLAGS_VLAN_PRIO_MASK
6708 << IXGBE_TX_FLAGS_VLAN_SHIFT);
6709 tx_flags |= ((adapter->fcoe.up << 13)
6710 << IXGBE_TX_FLAGS_VLAN_SHIFT);
6713 /* flag for FCoE offloads */
6714 if (protocol == htons(ETH_P_FCOE))
6715 tx_flags |= IXGBE_TX_FLAGS_FCOE;
6719 /* four things can cause us to need a context descriptor */
6720 if (skb_is_gso(skb) ||
6721 (skb->ip_summed == CHECKSUM_PARTIAL) ||
6722 (tx_flags & IXGBE_TX_FLAGS_VLAN) ||
6723 (tx_flags & IXGBE_TX_FLAGS_FCOE))
6726 count += TXD_USE_COUNT(skb_headlen(skb));
6727 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6728 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6730 if (ixgbe_maybe_stop_tx(tx_ring, count)) {
6731 tx_ring->tx_stats.tx_busy++;
6732 return NETDEV_TX_BUSY;
6735 first = tx_ring->next_to_use;
6736 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6738 /* setup tx offload for FCoE */
6739 tso = ixgbe_fso(adapter, tx_ring, skb, tx_flags, &hdr_len);
6741 dev_kfree_skb_any(skb);
6742 return NETDEV_TX_OK;
6745 tx_flags |= IXGBE_TX_FLAGS_FSO;
6746 #endif /* IXGBE_FCOE */
6748 if (protocol == htons(ETH_P_IP))
6749 tx_flags |= IXGBE_TX_FLAGS_IPV4;
6750 tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len,
6753 dev_kfree_skb_any(skb);
6754 return NETDEV_TX_OK;
6758 tx_flags |= IXGBE_TX_FLAGS_TSO;
6759 else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags,
6761 (skb->ip_summed == CHECKSUM_PARTIAL))
6762 tx_flags |= IXGBE_TX_FLAGS_CSUM;
6765 count = ixgbe_tx_map(adapter, tx_ring, skb, tx_flags, first, hdr_len);
6767 /* add the ATR filter if ATR is on */
6768 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
6769 ixgbe_atr(tx_ring, skb, tx_flags, protocol);
6770 ixgbe_tx_queue(tx_ring, tx_flags, count, skb->len, hdr_len);
6771 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
6774 dev_kfree_skb_any(skb);
6775 tx_ring->tx_buffer_info[first].time_stamp = 0;
6776 tx_ring->next_to_use = first;
6779 return NETDEV_TX_OK;
6782 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
6784 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6785 struct ixgbe_ring *tx_ring;
6787 tx_ring = adapter->tx_ring[skb->queue_mapping];
6788 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
6792 * ixgbe_set_mac - Change the Ethernet Address of the NIC
6793 * @netdev: network interface device structure
6794 * @p: pointer to an address structure
6796 * Returns 0 on success, negative on failure
6798 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6800 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6801 struct ixgbe_hw *hw = &adapter->hw;
6802 struct sockaddr *addr = p;
6804 if (!is_valid_ether_addr(addr->sa_data))
6805 return -EADDRNOTAVAIL;
6807 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6808 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6810 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
6817 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6819 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6820 struct ixgbe_hw *hw = &adapter->hw;
6824 if (prtad != hw->phy.mdio.prtad)
6826 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6832 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6833 u16 addr, u16 value)
6835 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6836 struct ixgbe_hw *hw = &adapter->hw;
6838 if (prtad != hw->phy.mdio.prtad)
6840 return hw->phy.ops.write_reg(hw, addr, devad, value);
6843 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6845 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6847 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6851 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6853 * @netdev: network interface device structure
6855 * Returns non-zero on failure
6857 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6860 struct ixgbe_adapter *adapter = netdev_priv(dev);
6861 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6863 if (is_valid_ether_addr(mac->san_addr)) {
6865 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6872 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6874 * @netdev: network interface device structure
6876 * Returns non-zero on failure
6878 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6881 struct ixgbe_adapter *adapter = netdev_priv(dev);
6882 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6884 if (is_valid_ether_addr(mac->san_addr)) {
6886 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6892 #ifdef CONFIG_NET_POLL_CONTROLLER
6894 * Polling 'interrupt' - used by things like netconsole to send skbs
6895 * without having to re-enable interrupts. It's not called while
6896 * the interrupt routine is executing.
6898 static void ixgbe_netpoll(struct net_device *netdev)
6900 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6903 /* if interface is down do nothing */
6904 if (test_bit(__IXGBE_DOWN, &adapter->state))
6907 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
6908 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
6909 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
6910 for (i = 0; i < num_q_vectors; i++) {
6911 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
6912 ixgbe_msix_clean_many(0, q_vector);
6915 ixgbe_intr(adapter->pdev->irq, netdev);
6917 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
6921 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
6922 struct rtnl_link_stats64 *stats)
6924 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6928 for (i = 0; i < adapter->num_rx_queues; i++) {
6929 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
6935 start = u64_stats_fetch_begin_bh(&ring->syncp);
6936 packets = ring->stats.packets;
6937 bytes = ring->stats.bytes;
6938 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6939 stats->rx_packets += packets;
6940 stats->rx_bytes += bytes;
6944 for (i = 0; i < adapter->num_tx_queues; i++) {
6945 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->tx_ring[i]);
6951 start = u64_stats_fetch_begin_bh(&ring->syncp);
6952 packets = ring->stats.packets;
6953 bytes = ring->stats.bytes;
6954 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6955 stats->tx_packets += packets;
6956 stats->tx_bytes += bytes;
6960 /* following stats updated by ixgbe_watchdog_task() */
6961 stats->multicast = netdev->stats.multicast;
6962 stats->rx_errors = netdev->stats.rx_errors;
6963 stats->rx_length_errors = netdev->stats.rx_length_errors;
6964 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
6965 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
6970 static const struct net_device_ops ixgbe_netdev_ops = {
6971 .ndo_open = ixgbe_open,
6972 .ndo_stop = ixgbe_close,
6973 .ndo_start_xmit = ixgbe_xmit_frame,
6974 .ndo_select_queue = ixgbe_select_queue,
6975 .ndo_set_rx_mode = ixgbe_set_rx_mode,
6976 .ndo_set_multicast_list = ixgbe_set_rx_mode,
6977 .ndo_validate_addr = eth_validate_addr,
6978 .ndo_set_mac_address = ixgbe_set_mac,
6979 .ndo_change_mtu = ixgbe_change_mtu,
6980 .ndo_tx_timeout = ixgbe_tx_timeout,
6981 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
6982 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
6983 .ndo_do_ioctl = ixgbe_ioctl,
6984 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
6985 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
6986 .ndo_set_vf_tx_rate = ixgbe_ndo_set_vf_bw,
6987 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
6988 .ndo_get_stats64 = ixgbe_get_stats64,
6989 #ifdef CONFIG_NET_POLL_CONTROLLER
6990 .ndo_poll_controller = ixgbe_netpoll,
6993 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
6994 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
6995 .ndo_fcoe_enable = ixgbe_fcoe_enable,
6996 .ndo_fcoe_disable = ixgbe_fcoe_disable,
6997 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
6998 #endif /* IXGBE_FCOE */
7001 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
7002 const struct ixgbe_info *ii)
7004 #ifdef CONFIG_PCI_IOV
7005 struct ixgbe_hw *hw = &adapter->hw;
7008 if (hw->mac.type == ixgbe_mac_82598EB || !max_vfs)
7011 /* The 82599 supports up to 64 VFs per physical function
7012 * but this implementation limits allocation to 63 so that
7013 * basic networking resources are still available to the
7016 adapter->num_vfs = (max_vfs > 63) ? 63 : max_vfs;
7017 adapter->flags |= IXGBE_FLAG_SRIOV_ENABLED;
7018 err = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
7020 e_err(probe, "Failed to enable PCI sriov: %d\n", err);
7023 /* If call to enable VFs succeeded then allocate memory
7024 * for per VF control structures.
7027 kcalloc(adapter->num_vfs,
7028 sizeof(struct vf_data_storage), GFP_KERNEL);
7029 if (adapter->vfinfo) {
7030 /* Now that we're sure SR-IOV is enabled
7031 * and memory allocated set up the mailbox parameters
7033 ixgbe_init_mbx_params_pf(hw);
7034 memcpy(&hw->mbx.ops, ii->mbx_ops,
7035 sizeof(hw->mbx.ops));
7037 /* Disable RSC when in SR-IOV mode */
7038 adapter->flags2 &= ~(IXGBE_FLAG2_RSC_CAPABLE |
7039 IXGBE_FLAG2_RSC_ENABLED);
7044 e_err(probe, "Unable to allocate memory for VF Data Storage - "
7045 "SRIOV disabled\n");
7046 pci_disable_sriov(adapter->pdev);
7049 adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
7050 adapter->num_vfs = 0;
7051 #endif /* CONFIG_PCI_IOV */
7055 * ixgbe_probe - Device Initialization Routine
7056 * @pdev: PCI device information struct
7057 * @ent: entry in ixgbe_pci_tbl
7059 * Returns 0 on success, negative on failure
7061 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
7062 * The OS initialization, configuring of the adapter private structure,
7063 * and a hardware reset occur.
7065 static int __devinit ixgbe_probe(struct pci_dev *pdev,
7066 const struct pci_device_id *ent)
7068 struct net_device *netdev;
7069 struct ixgbe_adapter *adapter = NULL;
7070 struct ixgbe_hw *hw;
7071 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
7072 static int cards_found;
7073 int i, err, pci_using_dac;
7074 u8 part_str[IXGBE_PBANUM_LENGTH];
7075 unsigned int indices = num_possible_cpus();
7081 /* Catch broken hardware that put the wrong VF device ID in
7082 * the PCIe SR-IOV capability.
7084 if (pdev->is_virtfn) {
7085 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
7086 pci_name(pdev), pdev->vendor, pdev->device);
7090 err = pci_enable_device_mem(pdev);
7094 if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
7095 !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
7098 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
7100 err = dma_set_coherent_mask(&pdev->dev,
7104 "No usable DMA configuration, aborting\n");
7111 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
7112 IORESOURCE_MEM), ixgbe_driver_name);
7115 "pci_request_selected_regions failed 0x%x\n", err);
7119 pci_enable_pcie_error_reporting(pdev);
7121 pci_set_master(pdev);
7122 pci_save_state(pdev);
7124 if (ii->mac == ixgbe_mac_82598EB)
7125 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
7127 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
7129 indices = max_t(unsigned int, indices, IXGBE_MAX_DCB_INDICES);
7131 indices += min_t(unsigned int, num_possible_cpus(),
7132 IXGBE_MAX_FCOE_INDICES);
7134 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
7137 goto err_alloc_etherdev;
7140 SET_NETDEV_DEV(netdev, &pdev->dev);
7142 adapter = netdev_priv(netdev);
7143 pci_set_drvdata(pdev, adapter);
7145 adapter->netdev = netdev;
7146 adapter->pdev = pdev;
7149 adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
7151 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
7152 pci_resource_len(pdev, 0));
7158 for (i = 1; i <= 5; i++) {
7159 if (pci_resource_len(pdev, i) == 0)
7163 netdev->netdev_ops = &ixgbe_netdev_ops;
7164 ixgbe_set_ethtool_ops(netdev);
7165 netdev->watchdog_timeo = 5 * HZ;
7166 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
7168 adapter->bd_number = cards_found;
7171 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
7172 hw->mac.type = ii->mac;
7175 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
7176 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
7177 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
7178 if (!(eec & (1 << 8)))
7179 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
7182 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
7183 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
7184 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
7185 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
7186 hw->phy.mdio.mmds = 0;
7187 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
7188 hw->phy.mdio.dev = netdev;
7189 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
7190 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
7192 /* set up this timer and work struct before calling get_invariants
7193 * which might start the timer
7195 init_timer(&adapter->sfp_timer);
7196 adapter->sfp_timer.function = ixgbe_sfp_timer;
7197 adapter->sfp_timer.data = (unsigned long) adapter;
7199 INIT_WORK(&adapter->sfp_task, ixgbe_sfp_task);
7201 /* multispeed fiber has its own tasklet, called from GPI SDP1 context */
7202 INIT_WORK(&adapter->multispeed_fiber_task, ixgbe_multispeed_fiber_task);
7204 /* a new SFP+ module arrival, called from GPI SDP2 context */
7205 INIT_WORK(&adapter->sfp_config_module_task,
7206 ixgbe_sfp_config_module_task);
7208 ii->get_invariants(hw);
7210 /* setup the private structure */
7211 err = ixgbe_sw_init(adapter);
7215 /* Make it possible the adapter to be woken up via WOL */
7216 switch (adapter->hw.mac.type) {
7217 case ixgbe_mac_82599EB:
7218 case ixgbe_mac_X540:
7219 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7226 * If there is a fan on this device and it has failed log the
7229 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
7230 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
7231 if (esdp & IXGBE_ESDP_SDP1)
7232 e_crit(probe, "Fan has stopped, replace the adapter\n");
7235 /* reset_hw fills in the perm_addr as well */
7236 hw->phy.reset_if_overtemp = true;
7237 err = hw->mac.ops.reset_hw(hw);
7238 hw->phy.reset_if_overtemp = false;
7239 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
7240 hw->mac.type == ixgbe_mac_82598EB) {
7242 * Start a kernel thread to watch for a module to arrive.
7243 * Only do this for 82598, since 82599 will generate
7244 * interrupts on module arrival.
7246 set_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
7247 mod_timer(&adapter->sfp_timer,
7248 round_jiffies(jiffies + (2 * HZ)));
7250 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
7251 e_dev_err("failed to initialize because an unsupported SFP+ "
7252 "module type was detected.\n");
7253 e_dev_err("Reload the driver after installing a supported "
7257 e_dev_err("HW Init failed: %d\n", err);
7261 ixgbe_probe_vf(adapter, ii);
7263 netdev->features = NETIF_F_SG |
7265 NETIF_F_HW_VLAN_TX |
7266 NETIF_F_HW_VLAN_RX |
7267 NETIF_F_HW_VLAN_FILTER;
7269 netdev->features |= NETIF_F_IPV6_CSUM;
7270 netdev->features |= NETIF_F_TSO;
7271 netdev->features |= NETIF_F_TSO6;
7272 netdev->features |= NETIF_F_GRO;
7274 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
7275 netdev->features |= NETIF_F_SCTP_CSUM;
7277 netdev->vlan_features |= NETIF_F_TSO;
7278 netdev->vlan_features |= NETIF_F_TSO6;
7279 netdev->vlan_features |= NETIF_F_IP_CSUM;
7280 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
7281 netdev->vlan_features |= NETIF_F_SG;
7283 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7284 adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
7285 IXGBE_FLAG_DCB_ENABLED);
7286 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
7287 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
7289 #ifdef CONFIG_IXGBE_DCB
7290 netdev->dcbnl_ops = &dcbnl_ops;
7294 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7295 if (hw->mac.ops.get_device_caps) {
7296 hw->mac.ops.get_device_caps(hw, &device_caps);
7297 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
7298 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
7301 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7302 netdev->vlan_features |= NETIF_F_FCOE_CRC;
7303 netdev->vlan_features |= NETIF_F_FSO;
7304 netdev->vlan_features |= NETIF_F_FCOE_MTU;
7306 #endif /* IXGBE_FCOE */
7307 if (pci_using_dac) {
7308 netdev->features |= NETIF_F_HIGHDMA;
7309 netdev->vlan_features |= NETIF_F_HIGHDMA;
7312 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
7313 netdev->features |= NETIF_F_LRO;
7315 /* make sure the EEPROM is good */
7316 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
7317 e_dev_err("The EEPROM Checksum Is Not Valid\n");
7322 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
7323 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
7325 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
7326 e_dev_err("invalid MAC address\n");
7331 /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
7332 if (hw->mac.ops.disable_tx_laser &&
7333 ((hw->phy.multispeed_fiber) ||
7334 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
7335 (hw->mac.type == ixgbe_mac_82599EB))))
7336 hw->mac.ops.disable_tx_laser(hw);
7338 init_timer(&adapter->watchdog_timer);
7339 adapter->watchdog_timer.function = ixgbe_watchdog;
7340 adapter->watchdog_timer.data = (unsigned long)adapter;
7342 INIT_WORK(&adapter->reset_task, ixgbe_reset_task);
7343 INIT_WORK(&adapter->watchdog_task, ixgbe_watchdog_task);
7345 err = ixgbe_init_interrupt_scheme(adapter);
7349 switch (pdev->device) {
7350 case IXGBE_DEV_ID_82599_SFP:
7351 /* Only this subdevice supports WOL */
7352 if (pdev->subsystem_device == IXGBE_SUBDEV_ID_82599_SFP)
7353 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
7354 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
7356 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
7357 /* All except this subdevice support WOL */
7358 if (pdev->subsystem_device != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
7359 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
7360 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
7362 case IXGBE_DEV_ID_82599_KX4:
7363 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
7364 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
7370 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
7372 /* pick up the PCI bus settings for reporting later */
7373 hw->mac.ops.get_bus_info(hw);
7375 /* print bus type/speed/width info */
7376 e_dev_info("(PCI Express:%s:%s) %pM\n",
7377 (hw->bus.speed == ixgbe_bus_speed_5000 ? "5.0Gb/s" :
7378 hw->bus.speed == ixgbe_bus_speed_2500 ? "2.5Gb/s" :
7380 (hw->bus.width == ixgbe_bus_width_pcie_x8 ? "Width x8" :
7381 hw->bus.width == ixgbe_bus_width_pcie_x4 ? "Width x4" :
7382 hw->bus.width == ixgbe_bus_width_pcie_x1 ? "Width x1" :
7386 err = ixgbe_read_pba_string_generic(hw, part_str, IXGBE_PBANUM_LENGTH);
7388 strncpy(part_str, "Unknown", IXGBE_PBANUM_LENGTH);
7389 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
7390 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
7391 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
7394 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
7395 hw->mac.type, hw->phy.type, part_str);
7397 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
7398 e_dev_warn("PCI-Express bandwidth available for this card is "
7399 "not sufficient for optimal performance.\n");
7400 e_dev_warn("For optimal performance a x8 PCI-Express slot "
7404 /* save off EEPROM version number */
7405 hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
7407 /* reset the hardware with the new settings */
7408 err = hw->mac.ops.start_hw(hw);
7410 if (err == IXGBE_ERR_EEPROM_VERSION) {
7411 /* We are running on a pre-production device, log a warning */
7412 e_dev_warn("This device is a pre-production adapter/LOM. "
7413 "Please be aware there may be issues associated "
7414 "with your hardware. If you are experiencing "
7415 "problems please contact your Intel or hardware "
7416 "representative who provided you with this "
7419 strcpy(netdev->name, "eth%d");
7420 err = register_netdev(netdev);
7424 /* carrier off reporting is important to ethtool even BEFORE open */
7425 netif_carrier_off(netdev);
7427 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
7428 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
7429 INIT_WORK(&adapter->fdir_reinit_task, ixgbe_fdir_reinit_task);
7431 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
7432 INIT_WORK(&adapter->check_overtemp_task,
7433 ixgbe_check_overtemp_task);
7434 #ifdef CONFIG_IXGBE_DCA
7435 if (dca_add_requester(&pdev->dev) == 0) {
7436 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
7437 ixgbe_setup_dca(adapter);
7440 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7441 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
7442 for (i = 0; i < adapter->num_vfs; i++)
7443 ixgbe_vf_configuration(pdev, (i | 0x10000000));
7446 /* add san mac addr to netdev */
7447 ixgbe_add_sanmac_netdev(netdev);
7449 e_dev_info("Intel(R) 10 Gigabit Network Connection\n");
7454 ixgbe_release_hw_control(adapter);
7455 ixgbe_clear_interrupt_scheme(adapter);
7458 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7459 ixgbe_disable_sriov(adapter);
7460 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
7461 del_timer_sync(&adapter->sfp_timer);
7462 cancel_work_sync(&adapter->sfp_task);
7463 cancel_work_sync(&adapter->multispeed_fiber_task);
7464 cancel_work_sync(&adapter->sfp_config_module_task);
7465 iounmap(hw->hw_addr);
7467 free_netdev(netdev);
7469 pci_release_selected_regions(pdev,
7470 pci_select_bars(pdev, IORESOURCE_MEM));
7473 pci_disable_device(pdev);
7478 * ixgbe_remove - Device Removal Routine
7479 * @pdev: PCI device information struct
7481 * ixgbe_remove is called by the PCI subsystem to alert the driver
7482 * that it should release a PCI device. The could be caused by a
7483 * Hot-Plug event, or because the driver is going to be removed from
7486 static void __devexit ixgbe_remove(struct pci_dev *pdev)
7488 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7489 struct net_device *netdev = adapter->netdev;
7491 set_bit(__IXGBE_DOWN, &adapter->state);
7494 * The timers may be rescheduled, so explicitly disable them
7495 * from being rescheduled.
7497 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
7498 del_timer_sync(&adapter->watchdog_timer);
7499 del_timer_sync(&adapter->sfp_timer);
7501 cancel_work_sync(&adapter->watchdog_task);
7502 cancel_work_sync(&adapter->sfp_task);
7503 cancel_work_sync(&adapter->multispeed_fiber_task);
7504 cancel_work_sync(&adapter->sfp_config_module_task);
7505 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
7506 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
7507 cancel_work_sync(&adapter->fdir_reinit_task);
7508 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
7509 cancel_work_sync(&adapter->check_overtemp_task);
7511 #ifdef CONFIG_IXGBE_DCA
7512 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
7513 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
7514 dca_remove_requester(&pdev->dev);
7515 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
7520 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
7521 ixgbe_cleanup_fcoe(adapter);
7523 #endif /* IXGBE_FCOE */
7525 /* remove the added san mac */
7526 ixgbe_del_sanmac_netdev(netdev);
7528 if (netdev->reg_state == NETREG_REGISTERED)
7529 unregister_netdev(netdev);
7531 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7532 ixgbe_disable_sriov(adapter);
7534 ixgbe_clear_interrupt_scheme(adapter);
7536 ixgbe_release_hw_control(adapter);
7538 iounmap(adapter->hw.hw_addr);
7539 pci_release_selected_regions(pdev, pci_select_bars(pdev,
7542 e_dev_info("complete\n");
7544 free_netdev(netdev);
7546 pci_disable_pcie_error_reporting(pdev);
7548 pci_disable_device(pdev);
7552 * ixgbe_io_error_detected - called when PCI error is detected
7553 * @pdev: Pointer to PCI device
7554 * @state: The current pci connection state
7556 * This function is called after a PCI bus error affecting
7557 * this device has been detected.
7559 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
7560 pci_channel_state_t state)
7562 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7563 struct net_device *netdev = adapter->netdev;
7565 netif_device_detach(netdev);
7567 if (state == pci_channel_io_perm_failure)
7568 return PCI_ERS_RESULT_DISCONNECT;
7570 if (netif_running(netdev))
7571 ixgbe_down(adapter);
7572 pci_disable_device(pdev);
7574 /* Request a slot reset. */
7575 return PCI_ERS_RESULT_NEED_RESET;
7579 * ixgbe_io_slot_reset - called after the pci bus has been reset.
7580 * @pdev: Pointer to PCI device
7582 * Restart the card from scratch, as if from a cold-boot.
7584 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7586 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7587 pci_ers_result_t result;
7590 if (pci_enable_device_mem(pdev)) {
7591 e_err(probe, "Cannot re-enable PCI device after reset.\n");
7592 result = PCI_ERS_RESULT_DISCONNECT;
7594 pci_set_master(pdev);
7595 pci_restore_state(pdev);
7596 pci_save_state(pdev);
7598 pci_wake_from_d3(pdev, false);
7600 ixgbe_reset(adapter);
7601 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7602 result = PCI_ERS_RESULT_RECOVERED;
7605 err = pci_cleanup_aer_uncorrect_error_status(pdev);
7607 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7608 "failed 0x%0x\n", err);
7609 /* non-fatal, continue */
7616 * ixgbe_io_resume - called when traffic can start flowing again.
7617 * @pdev: Pointer to PCI device
7619 * This callback is called when the error recovery driver tells us that
7620 * its OK to resume normal operation.
7622 static void ixgbe_io_resume(struct pci_dev *pdev)
7624 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7625 struct net_device *netdev = adapter->netdev;
7627 if (netif_running(netdev)) {
7628 if (ixgbe_up(adapter)) {
7629 e_info(probe, "ixgbe_up failed after reset\n");
7634 netif_device_attach(netdev);
7637 static struct pci_error_handlers ixgbe_err_handler = {
7638 .error_detected = ixgbe_io_error_detected,
7639 .slot_reset = ixgbe_io_slot_reset,
7640 .resume = ixgbe_io_resume,
7643 static struct pci_driver ixgbe_driver = {
7644 .name = ixgbe_driver_name,
7645 .id_table = ixgbe_pci_tbl,
7646 .probe = ixgbe_probe,
7647 .remove = __devexit_p(ixgbe_remove),
7649 .suspend = ixgbe_suspend,
7650 .resume = ixgbe_resume,
7652 .shutdown = ixgbe_shutdown,
7653 .err_handler = &ixgbe_err_handler
7657 * ixgbe_init_module - Driver Registration Routine
7659 * ixgbe_init_module is the first routine called when the driver is
7660 * loaded. All it does is register with the PCI subsystem.
7662 static int __init ixgbe_init_module(void)
7665 pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
7666 pr_info("%s\n", ixgbe_copyright);
7668 #ifdef CONFIG_IXGBE_DCA
7669 dca_register_notify(&dca_notifier);
7672 ret = pci_register_driver(&ixgbe_driver);
7676 module_init(ixgbe_init_module);
7679 * ixgbe_exit_module - Driver Exit Cleanup Routine
7681 * ixgbe_exit_module is called just before the driver is removed
7684 static void __exit ixgbe_exit_module(void)
7686 #ifdef CONFIG_IXGBE_DCA
7687 dca_unregister_notify(&dca_notifier);
7689 pci_unregister_driver(&ixgbe_driver);
7690 rcu_barrier(); /* Wait for completion of call_rcu()'s */
7693 #ifdef CONFIG_IXGBE_DCA
7694 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
7699 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
7700 __ixgbe_notify_dca);
7702 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7705 #endif /* CONFIG_IXGBE_DCA */
7708 * ixgbe_get_hw_dev return device
7709 * used by hardware layer to print debugging information
7711 struct net_device *ixgbe_get_hw_dev(struct ixgbe_hw *hw)
7713 struct ixgbe_adapter *adapter = hw->back;
7714 return adapter->netdev;
7717 module_exit(ixgbe_exit_module);