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 static 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 only the irqs that were actually requested */
2601 if (!adapter->q_vector[i]->rxr_count &&
2602 !adapter->q_vector[i]->txr_count)
2605 free_irq(adapter->msix_entries[i].vector,
2606 adapter->q_vector[i]);
2609 ixgbe_reset_q_vectors(adapter);
2611 free_irq(adapter->pdev->irq, netdev);
2616 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2617 * @adapter: board private structure
2619 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2621 switch (adapter->hw.mac.type) {
2622 case ixgbe_mac_82598EB:
2623 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2625 case ixgbe_mac_82599EB:
2626 case ixgbe_mac_X540:
2627 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2628 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2629 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2630 if (adapter->num_vfs > 32)
2631 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
2636 IXGBE_WRITE_FLUSH(&adapter->hw);
2637 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2639 for (i = 0; i < adapter->num_msix_vectors; i++)
2640 synchronize_irq(adapter->msix_entries[i].vector);
2642 synchronize_irq(adapter->pdev->irq);
2647 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2650 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2652 struct ixgbe_hw *hw = &adapter->hw;
2654 IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
2655 EITR_INTS_PER_SEC_TO_REG(adapter->rx_eitr_param));
2657 ixgbe_set_ivar(adapter, 0, 0, 0);
2658 ixgbe_set_ivar(adapter, 1, 0, 0);
2660 map_vector_to_rxq(adapter, 0, 0);
2661 map_vector_to_txq(adapter, 0, 0);
2663 e_info(hw, "Legacy interrupt IVAR setup done\n");
2667 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2668 * @adapter: board private structure
2669 * @ring: structure containing ring specific data
2671 * Configure the Tx descriptor ring after a reset.
2673 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2674 struct ixgbe_ring *ring)
2676 struct ixgbe_hw *hw = &adapter->hw;
2677 u64 tdba = ring->dma;
2680 u8 reg_idx = ring->reg_idx;
2682 /* disable queue to avoid issues while updating state */
2683 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2684 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx),
2685 txdctl & ~IXGBE_TXDCTL_ENABLE);
2686 IXGBE_WRITE_FLUSH(hw);
2688 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2689 (tdba & DMA_BIT_MASK(32)));
2690 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2691 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2692 ring->count * sizeof(union ixgbe_adv_tx_desc));
2693 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2694 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2695 ring->tail = hw->hw_addr + IXGBE_TDT(reg_idx);
2697 /* configure fetching thresholds */
2698 if (adapter->rx_itr_setting == 0) {
2699 /* cannot set wthresh when itr==0 */
2700 txdctl &= ~0x007F0000;
2702 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
2703 txdctl |= (8 << 16);
2705 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2706 /* PThresh workaround for Tx hang with DFP enabled. */
2710 /* reinitialize flowdirector state */
2711 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2712 adapter->atr_sample_rate) {
2713 ring->atr_sample_rate = adapter->atr_sample_rate;
2714 ring->atr_count = 0;
2715 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
2717 ring->atr_sample_rate = 0;
2720 clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
2723 txdctl |= IXGBE_TXDCTL_ENABLE;
2724 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
2726 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2727 if (hw->mac.type == ixgbe_mac_82598EB &&
2728 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2731 /* poll to verify queue is enabled */
2734 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2735 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
2737 e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
2740 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
2742 struct ixgbe_hw *hw = &adapter->hw;
2746 if (hw->mac.type == ixgbe_mac_82598EB)
2749 /* disable the arbiter while setting MTQC */
2750 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2751 rttdcs |= IXGBE_RTTDCS_ARBDIS;
2752 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2754 /* set transmit pool layout */
2755 mask = (IXGBE_FLAG_SRIOV_ENABLED | IXGBE_FLAG_DCB_ENABLED);
2756 switch (adapter->flags & mask) {
2758 case (IXGBE_FLAG_SRIOV_ENABLED):
2759 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2760 (IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF));
2763 case (IXGBE_FLAG_DCB_ENABLED):
2764 /* We enable 8 traffic classes, DCB only */
2765 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2766 (IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ));
2770 IXGBE_WRITE_REG(hw, IXGBE_MTQC, IXGBE_MTQC_64Q_1PB);
2774 /* re-enable the arbiter */
2775 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2776 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2780 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2781 * @adapter: board private structure
2783 * Configure the Tx unit of the MAC after a reset.
2785 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2787 struct ixgbe_hw *hw = &adapter->hw;
2791 ixgbe_setup_mtqc(adapter);
2793 if (hw->mac.type != ixgbe_mac_82598EB) {
2794 /* DMATXCTL.EN must be before Tx queues are enabled */
2795 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2796 dmatxctl |= IXGBE_DMATXCTL_TE;
2797 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2800 /* Setup the HW Tx Head and Tail descriptor pointers */
2801 for (i = 0; i < adapter->num_tx_queues; i++)
2802 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
2805 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2807 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2808 struct ixgbe_ring *rx_ring)
2811 u8 reg_idx = rx_ring->reg_idx;
2813 switch (adapter->hw.mac.type) {
2814 case ixgbe_mac_82598EB: {
2815 struct ixgbe_ring_feature *feature = adapter->ring_feature;
2816 const int mask = feature[RING_F_RSS].mask;
2817 reg_idx = reg_idx & mask;
2820 case ixgbe_mac_82599EB:
2821 case ixgbe_mac_X540:
2826 srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx));
2828 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
2829 srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
2830 if (adapter->num_vfs)
2831 srrctl |= IXGBE_SRRCTL_DROP_EN;
2833 srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
2834 IXGBE_SRRCTL_BSIZEHDR_MASK;
2836 if (ring_is_ps_enabled(rx_ring)) {
2837 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
2838 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2840 srrctl |= (PAGE_SIZE / 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2842 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2844 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
2845 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2846 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2849 IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx), srrctl);
2852 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2854 struct ixgbe_hw *hw = &adapter->hw;
2855 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2856 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2857 0x6A3E67EA, 0x14364D17, 0x3BED200D};
2858 u32 mrqc = 0, reta = 0;
2863 /* Fill out hash function seeds */
2864 for (i = 0; i < 10; i++)
2865 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2867 /* Fill out redirection table */
2868 for (i = 0, j = 0; i < 128; i++, j++) {
2869 if (j == adapter->ring_feature[RING_F_RSS].indices)
2871 /* reta = 4-byte sliding window of
2872 * 0x00..(indices-1)(indices-1)00..etc. */
2873 reta = (reta << 8) | (j * 0x11);
2875 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2878 /* Disable indicating checksum in descriptor, enables RSS hash */
2879 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2880 rxcsum |= IXGBE_RXCSUM_PCSD;
2881 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2883 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
2884 mask = adapter->flags & IXGBE_FLAG_RSS_ENABLED;
2886 mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2887 #ifdef CONFIG_IXGBE_DCB
2888 | IXGBE_FLAG_DCB_ENABLED
2890 | IXGBE_FLAG_SRIOV_ENABLED
2894 case (IXGBE_FLAG_RSS_ENABLED):
2895 mrqc = IXGBE_MRQC_RSSEN;
2897 case (IXGBE_FLAG_SRIOV_ENABLED):
2898 mrqc = IXGBE_MRQC_VMDQEN;
2900 #ifdef CONFIG_IXGBE_DCB
2901 case (IXGBE_FLAG_DCB_ENABLED):
2902 mrqc = IXGBE_MRQC_RT8TCEN;
2904 #endif /* CONFIG_IXGBE_DCB */
2909 /* Perform hash on these packet types */
2910 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2911 | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2912 | IXGBE_MRQC_RSS_FIELD_IPV6
2913 | IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
2915 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2919 * ixgbe_clear_rscctl - disable RSC for the indicated ring
2920 * @adapter: address of board private structure
2921 * @ring: structure containing ring specific data
2923 void ixgbe_clear_rscctl(struct ixgbe_adapter *adapter,
2924 struct ixgbe_ring *ring)
2926 struct ixgbe_hw *hw = &adapter->hw;
2928 u8 reg_idx = ring->reg_idx;
2930 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
2931 rscctrl &= ~IXGBE_RSCCTL_RSCEN;
2932 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
2936 * ixgbe_configure_rscctl - enable RSC for the indicated ring
2937 * @adapter: address of board private structure
2938 * @index: index of ring to set
2940 void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
2941 struct ixgbe_ring *ring)
2943 struct ixgbe_hw *hw = &adapter->hw;
2946 u8 reg_idx = ring->reg_idx;
2948 if (!ring_is_rsc_enabled(ring))
2951 rx_buf_len = ring->rx_buf_len;
2952 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
2953 rscctrl |= IXGBE_RSCCTL_RSCEN;
2955 * we must limit the number of descriptors so that the
2956 * total size of max desc * buf_len is not greater
2959 if (ring_is_ps_enabled(ring)) {
2960 #if (MAX_SKB_FRAGS > 16)
2961 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2962 #elif (MAX_SKB_FRAGS > 8)
2963 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2964 #elif (MAX_SKB_FRAGS > 4)
2965 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2967 rscctrl |= IXGBE_RSCCTL_MAXDESC_1;
2970 if (rx_buf_len < IXGBE_RXBUFFER_4096)
2971 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2972 else if (rx_buf_len < IXGBE_RXBUFFER_8192)
2973 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2975 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2977 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
2981 * ixgbe_set_uta - Set unicast filter table address
2982 * @adapter: board private structure
2984 * The unicast table address is a register array of 32-bit registers.
2985 * The table is meant to be used in a way similar to how the MTA is used
2986 * however due to certain limitations in the hardware it is necessary to
2987 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
2988 * enable bit to allow vlan tag stripping when promiscuous mode is enabled
2990 static void ixgbe_set_uta(struct ixgbe_adapter *adapter)
2992 struct ixgbe_hw *hw = &adapter->hw;
2995 /* The UTA table only exists on 82599 hardware and newer */
2996 if (hw->mac.type < ixgbe_mac_82599EB)
2999 /* we only need to do this if VMDq is enabled */
3000 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
3003 for (i = 0; i < 128; i++)
3004 IXGBE_WRITE_REG(hw, IXGBE_UTA(i), ~0);
3007 #define IXGBE_MAX_RX_DESC_POLL 10
3008 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
3009 struct ixgbe_ring *ring)
3011 struct ixgbe_hw *hw = &adapter->hw;
3012 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3014 u8 reg_idx = ring->reg_idx;
3016 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3017 if (hw->mac.type == ixgbe_mac_82598EB &&
3018 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3023 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3024 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
3027 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
3028 "the polling period\n", reg_idx);
3032 void ixgbe_disable_rx_queue(struct ixgbe_adapter *adapter,
3033 struct ixgbe_ring *ring)
3035 struct ixgbe_hw *hw = &adapter->hw;
3036 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3038 u8 reg_idx = ring->reg_idx;
3040 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3041 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
3043 /* write value back with RXDCTL.ENABLE bit cleared */
3044 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3046 if (hw->mac.type == ixgbe_mac_82598EB &&
3047 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3050 /* the hardware may take up to 100us to really disable the rx queue */
3053 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3054 } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
3057 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not cleared within "
3058 "the polling period\n", reg_idx);
3062 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
3063 struct ixgbe_ring *ring)
3065 struct ixgbe_hw *hw = &adapter->hw;
3066 u64 rdba = ring->dma;
3068 u8 reg_idx = ring->reg_idx;
3070 /* disable queue to avoid issues while updating state */
3071 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3072 ixgbe_disable_rx_queue(adapter, ring);
3074 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
3075 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
3076 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
3077 ring->count * sizeof(union ixgbe_adv_rx_desc));
3078 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
3079 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
3080 ring->tail = hw->hw_addr + IXGBE_RDT(reg_idx);
3082 ixgbe_configure_srrctl(adapter, ring);
3083 ixgbe_configure_rscctl(adapter, ring);
3085 /* If operating in IOV mode set RLPML for X540 */
3086 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
3087 hw->mac.type == ixgbe_mac_X540) {
3088 rxdctl &= ~IXGBE_RXDCTL_RLPMLMASK;
3089 rxdctl |= ((ring->netdev->mtu + ETH_HLEN +
3090 ETH_FCS_LEN + VLAN_HLEN) | IXGBE_RXDCTL_RLPML_EN);
3093 if (hw->mac.type == ixgbe_mac_82598EB) {
3095 * enable cache line friendly hardware writes:
3096 * PTHRESH=32 descriptors (half the internal cache),
3097 * this also removes ugly rx_no_buffer_count increment
3098 * HTHRESH=4 descriptors (to minimize latency on fetch)
3099 * WTHRESH=8 burst writeback up to two cache lines
3101 rxdctl &= ~0x3FFFFF;
3105 /* enable receive descriptor ring */
3106 rxdctl |= IXGBE_RXDCTL_ENABLE;
3107 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3109 ixgbe_rx_desc_queue_enable(adapter, ring);
3110 ixgbe_alloc_rx_buffers(ring, IXGBE_DESC_UNUSED(ring));
3113 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
3115 struct ixgbe_hw *hw = &adapter->hw;
3118 /* PSRTYPE must be initialized in non 82598 adapters */
3119 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
3120 IXGBE_PSRTYPE_UDPHDR |
3121 IXGBE_PSRTYPE_IPV4HDR |
3122 IXGBE_PSRTYPE_L2HDR |
3123 IXGBE_PSRTYPE_IPV6HDR;
3125 if (hw->mac.type == ixgbe_mac_82598EB)
3128 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED)
3129 psrtype |= (adapter->num_rx_queues_per_pool << 29);
3131 for (p = 0; p < adapter->num_rx_pools; p++)
3132 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(adapter->num_vfs + p),
3136 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
3138 struct ixgbe_hw *hw = &adapter->hw;
3141 u32 reg_offset, vf_shift;
3144 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
3147 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
3148 vt_reg_bits = IXGBE_VMD_CTL_VMDQ_EN | IXGBE_VT_CTL_REPLEN;
3149 vt_reg_bits |= (adapter->num_vfs << IXGBE_VT_CTL_POOL_SHIFT);
3150 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl | vt_reg_bits);
3152 vf_shift = adapter->num_vfs % 32;
3153 reg_offset = (adapter->num_vfs > 32) ? 1 : 0;
3155 /* Enable only the PF's pool for Tx/Rx */
3156 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (1 << vf_shift));
3157 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), 0);
3158 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (1 << vf_shift));
3159 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), 0);
3160 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3162 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
3163 hw->mac.ops.set_vmdq(hw, 0, adapter->num_vfs);
3166 * Set up VF register offsets for selected VT Mode,
3167 * i.e. 32 or 64 VFs for SR-IOV
3169 gcr_ext = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
3170 gcr_ext |= IXGBE_GCR_EXT_MSIX_EN;
3171 gcr_ext |= IXGBE_GCR_EXT_VT_MODE_64;
3172 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
3174 /* enable Tx loopback for VF/PF communication */
3175 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3176 /* Enable MAC Anti-Spoofing */
3177 hw->mac.ops.set_mac_anti_spoofing(hw, (adapter->num_vfs != 0),
3181 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
3183 struct ixgbe_hw *hw = &adapter->hw;
3184 struct net_device *netdev = adapter->netdev;
3185 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3187 struct ixgbe_ring *rx_ring;
3191 /* Decide whether to use packet split mode or not */
3193 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
3195 /* Do not use packet split if we're in SR-IOV Mode */
3196 if (adapter->num_vfs)
3197 adapter->flags &= ~IXGBE_FLAG_RX_PS_ENABLED;
3199 /* Disable packet split due to 82599 erratum #45 */
3200 if (hw->mac.type == ixgbe_mac_82599EB)
3201 adapter->flags &= ~IXGBE_FLAG_RX_PS_ENABLED;
3203 /* Set the RX buffer length according to the mode */
3204 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
3205 rx_buf_len = IXGBE_RX_HDR_SIZE;
3207 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
3208 (netdev->mtu <= ETH_DATA_LEN))
3209 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3211 rx_buf_len = ALIGN(max_frame + VLAN_HLEN, 1024);
3215 /* adjust max frame to be able to do baby jumbo for FCoE */
3216 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
3217 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
3218 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3220 #endif /* IXGBE_FCOE */
3221 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
3222 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
3223 mhadd &= ~IXGBE_MHADD_MFS_MASK;
3224 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
3226 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
3229 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
3230 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
3231 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
3232 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
3235 * Setup the HW Rx Head and Tail Descriptor Pointers and
3236 * the Base and Length of the Rx Descriptor Ring
3238 for (i = 0; i < adapter->num_rx_queues; i++) {
3239 rx_ring = adapter->rx_ring[i];
3240 rx_ring->rx_buf_len = rx_buf_len;
3242 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)
3243 set_ring_ps_enabled(rx_ring);
3245 clear_ring_ps_enabled(rx_ring);
3247 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
3248 set_ring_rsc_enabled(rx_ring);
3250 clear_ring_rsc_enabled(rx_ring);
3253 if (netdev->features & NETIF_F_FCOE_MTU) {
3254 struct ixgbe_ring_feature *f;
3255 f = &adapter->ring_feature[RING_F_FCOE];
3256 if ((i >= f->mask) && (i < f->mask + f->indices)) {
3257 clear_ring_ps_enabled(rx_ring);
3258 if (rx_buf_len < IXGBE_FCOE_JUMBO_FRAME_SIZE)
3259 rx_ring->rx_buf_len =
3260 IXGBE_FCOE_JUMBO_FRAME_SIZE;
3261 } else if (!ring_is_rsc_enabled(rx_ring) &&
3262 !ring_is_ps_enabled(rx_ring)) {
3263 rx_ring->rx_buf_len =
3264 IXGBE_FCOE_JUMBO_FRAME_SIZE;
3267 #endif /* IXGBE_FCOE */
3271 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
3273 struct ixgbe_hw *hw = &adapter->hw;
3274 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
3276 switch (hw->mac.type) {
3277 case ixgbe_mac_82598EB:
3279 * For VMDq support of different descriptor types or
3280 * buffer sizes through the use of multiple SRRCTL
3281 * registers, RDRXCTL.MVMEN must be set to 1
3283 * also, the manual doesn't mention it clearly but DCA hints
3284 * will only use queue 0's tags unless this bit is set. Side
3285 * effects of setting this bit are only that SRRCTL must be
3286 * fully programmed [0..15]
3288 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
3290 case ixgbe_mac_82599EB:
3291 case ixgbe_mac_X540:
3292 /* Disable RSC for ACK packets */
3293 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
3294 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
3295 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3296 /* hardware requires some bits to be set by default */
3297 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3298 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3301 /* We should do nothing since we don't know this hardware */
3305 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3309 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3310 * @adapter: board private structure
3312 * Configure the Rx unit of the MAC after a reset.
3314 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3316 struct ixgbe_hw *hw = &adapter->hw;
3320 /* disable receives while setting up the descriptors */
3321 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3322 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3324 ixgbe_setup_psrtype(adapter);
3325 ixgbe_setup_rdrxctl(adapter);
3327 /* Program registers for the distribution of queues */
3328 ixgbe_setup_mrqc(adapter);
3330 ixgbe_set_uta(adapter);
3332 /* set_rx_buffer_len must be called before ring initialization */
3333 ixgbe_set_rx_buffer_len(adapter);
3336 * Setup the HW Rx Head and Tail Descriptor Pointers and
3337 * the Base and Length of the Rx Descriptor Ring
3339 for (i = 0; i < adapter->num_rx_queues; i++)
3340 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3342 /* disable drop enable for 82598 parts */
3343 if (hw->mac.type == ixgbe_mac_82598EB)
3344 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3346 /* enable all receives */
3347 rxctrl |= IXGBE_RXCTRL_RXEN;
3348 hw->mac.ops.enable_rx_dma(hw, rxctrl);
3351 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3353 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3354 struct ixgbe_hw *hw = &adapter->hw;
3355 int pool_ndx = adapter->num_vfs;
3357 /* add VID to filter table */
3358 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, true);
3359 set_bit(vid, adapter->active_vlans);
3362 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3364 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3365 struct ixgbe_hw *hw = &adapter->hw;
3366 int pool_ndx = adapter->num_vfs;
3368 /* remove VID from filter table */
3369 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, false);
3370 clear_bit(vid, adapter->active_vlans);
3374 * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3375 * @adapter: driver data
3377 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
3379 struct ixgbe_hw *hw = &adapter->hw;
3382 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3383 vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
3384 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3388 * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3389 * @adapter: driver data
3391 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
3393 struct ixgbe_hw *hw = &adapter->hw;
3396 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3397 vlnctrl |= IXGBE_VLNCTRL_VFE;
3398 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3399 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3403 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3404 * @adapter: driver data
3406 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3408 struct ixgbe_hw *hw = &adapter->hw;
3412 switch (hw->mac.type) {
3413 case ixgbe_mac_82598EB:
3414 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3415 vlnctrl &= ~IXGBE_VLNCTRL_VME;
3416 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3418 case ixgbe_mac_82599EB:
3419 case ixgbe_mac_X540:
3420 for (i = 0; i < adapter->num_rx_queues; i++) {
3421 j = adapter->rx_ring[i]->reg_idx;
3422 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3423 vlnctrl &= ~IXGBE_RXDCTL_VME;
3424 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3433 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3434 * @adapter: driver data
3436 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3438 struct ixgbe_hw *hw = &adapter->hw;
3442 switch (hw->mac.type) {
3443 case ixgbe_mac_82598EB:
3444 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3445 vlnctrl |= IXGBE_VLNCTRL_VME;
3446 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3448 case ixgbe_mac_82599EB:
3449 case ixgbe_mac_X540:
3450 for (i = 0; i < adapter->num_rx_queues; i++) {
3451 j = adapter->rx_ring[i]->reg_idx;
3452 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3453 vlnctrl |= IXGBE_RXDCTL_VME;
3454 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3462 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3466 ixgbe_vlan_rx_add_vid(adapter->netdev, 0);
3468 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3469 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
3473 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3474 * @netdev: network interface device structure
3476 * Writes unicast address list to the RAR table.
3477 * Returns: -ENOMEM on failure/insufficient address space
3478 * 0 on no addresses written
3479 * X on writing X addresses to the RAR table
3481 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3483 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3484 struct ixgbe_hw *hw = &adapter->hw;
3485 unsigned int vfn = adapter->num_vfs;
3486 unsigned int rar_entries = hw->mac.num_rar_entries - (vfn + 1);
3489 /* return ENOMEM indicating insufficient memory for addresses */
3490 if (netdev_uc_count(netdev) > rar_entries)
3493 if (!netdev_uc_empty(netdev) && rar_entries) {
3494 struct netdev_hw_addr *ha;
3495 /* return error if we do not support writing to RAR table */
3496 if (!hw->mac.ops.set_rar)
3499 netdev_for_each_uc_addr(ha, netdev) {
3502 hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3507 /* write the addresses in reverse order to avoid write combining */
3508 for (; rar_entries > 0 ; rar_entries--)
3509 hw->mac.ops.clear_rar(hw, rar_entries);
3515 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3516 * @netdev: network interface device structure
3518 * The set_rx_method entry point is called whenever the unicast/multicast
3519 * address list or the network interface flags are updated. This routine is
3520 * responsible for configuring the hardware for proper unicast, multicast and
3523 void ixgbe_set_rx_mode(struct net_device *netdev)
3525 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3526 struct ixgbe_hw *hw = &adapter->hw;
3527 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3530 /* Check for Promiscuous and All Multicast modes */
3532 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3534 /* set all bits that we expect to always be set */
3535 fctrl |= IXGBE_FCTRL_BAM;
3536 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3537 fctrl |= IXGBE_FCTRL_PMCF;
3539 /* clear the bits we are changing the status of */
3540 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3542 if (netdev->flags & IFF_PROMISC) {
3543 hw->addr_ctrl.user_set_promisc = true;
3544 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3545 vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3546 /* don't hardware filter vlans in promisc mode */
3547 ixgbe_vlan_filter_disable(adapter);
3549 if (netdev->flags & IFF_ALLMULTI) {
3550 fctrl |= IXGBE_FCTRL_MPE;
3551 vmolr |= IXGBE_VMOLR_MPE;
3554 * Write addresses to the MTA, if the attempt fails
3555 * then we should just turn on promiscous mode so
3556 * that we can at least receive multicast traffic
3558 hw->mac.ops.update_mc_addr_list(hw, netdev);
3559 vmolr |= IXGBE_VMOLR_ROMPE;
3561 ixgbe_vlan_filter_enable(adapter);
3562 hw->addr_ctrl.user_set_promisc = false;
3564 * Write addresses to available RAR registers, if there is not
3565 * sufficient space to store all the addresses then enable
3566 * unicast promiscous mode
3568 count = ixgbe_write_uc_addr_list(netdev);
3570 fctrl |= IXGBE_FCTRL_UPE;
3571 vmolr |= IXGBE_VMOLR_ROPE;
3575 if (adapter->num_vfs) {
3576 ixgbe_restore_vf_multicasts(adapter);
3577 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(adapter->num_vfs)) &
3578 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3580 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(adapter->num_vfs), vmolr);
3583 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3585 if (netdev->features & NETIF_F_HW_VLAN_RX)
3586 ixgbe_vlan_strip_enable(adapter);
3588 ixgbe_vlan_strip_disable(adapter);
3591 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3594 struct ixgbe_q_vector *q_vector;
3595 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3597 /* legacy and MSI only use one vector */
3598 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3601 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3602 struct napi_struct *napi;
3603 q_vector = adapter->q_vector[q_idx];
3604 napi = &q_vector->napi;
3605 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3606 if (!q_vector->rxr_count || !q_vector->txr_count) {
3607 if (q_vector->txr_count == 1)
3608 napi->poll = &ixgbe_clean_txonly;
3609 else if (q_vector->rxr_count == 1)
3610 napi->poll = &ixgbe_clean_rxonly;
3618 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3621 struct ixgbe_q_vector *q_vector;
3622 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3624 /* legacy and MSI only use one vector */
3625 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3628 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3629 q_vector = adapter->q_vector[q_idx];
3630 napi_disable(&q_vector->napi);
3634 #ifdef CONFIG_IXGBE_DCB
3636 * ixgbe_configure_dcb - Configure DCB hardware
3637 * @adapter: ixgbe adapter struct
3639 * This is called by the driver on open to configure the DCB hardware.
3640 * This is also called by the gennetlink interface when reconfiguring
3643 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3645 struct ixgbe_hw *hw = &adapter->hw;
3646 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3648 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3649 if (hw->mac.type == ixgbe_mac_82598EB)
3650 netif_set_gso_max_size(adapter->netdev, 65536);
3654 if (hw->mac.type == ixgbe_mac_82598EB)
3655 netif_set_gso_max_size(adapter->netdev, 32768);
3658 if (adapter->netdev->features & NETIF_F_FCOE_MTU)
3659 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
3662 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3664 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3667 /* Enable VLAN tag insert/strip */
3668 adapter->netdev->features |= NETIF_F_HW_VLAN_RX;
3670 hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
3672 /* reconfigure the hardware */
3673 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
3677 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3679 struct net_device *netdev = adapter->netdev;
3680 struct ixgbe_hw *hw = &adapter->hw;
3683 #ifdef CONFIG_IXGBE_DCB
3684 ixgbe_configure_dcb(adapter);
3687 ixgbe_set_rx_mode(netdev);
3688 ixgbe_restore_vlan(adapter);
3691 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
3692 ixgbe_configure_fcoe(adapter);
3694 #endif /* IXGBE_FCOE */
3695 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3696 for (i = 0; i < adapter->num_tx_queues; i++)
3697 adapter->tx_ring[i]->atr_sample_rate =
3698 adapter->atr_sample_rate;
3699 ixgbe_init_fdir_signature_82599(hw, adapter->fdir_pballoc);
3700 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3701 ixgbe_init_fdir_perfect_82599(hw, adapter->fdir_pballoc);
3703 ixgbe_configure_virtualization(adapter);
3705 ixgbe_configure_tx(adapter);
3706 ixgbe_configure_rx(adapter);
3709 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3711 switch (hw->phy.type) {
3712 case ixgbe_phy_sfp_avago:
3713 case ixgbe_phy_sfp_ftl:
3714 case ixgbe_phy_sfp_intel:
3715 case ixgbe_phy_sfp_unknown:
3716 case ixgbe_phy_sfp_passive_tyco:
3717 case ixgbe_phy_sfp_passive_unknown:
3718 case ixgbe_phy_sfp_active_unknown:
3719 case ixgbe_phy_sfp_ftl_active:
3727 * ixgbe_sfp_link_config - set up SFP+ link
3728 * @adapter: pointer to private adapter struct
3730 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3732 struct ixgbe_hw *hw = &adapter->hw;
3734 if (hw->phy.multispeed_fiber) {
3736 * In multispeed fiber setups, the device may not have
3737 * had a physical connection when the driver loaded.
3738 * If that's the case, the initial link configuration
3739 * couldn't get the MAC into 10G or 1G mode, so we'll
3740 * never have a link status change interrupt fire.
3741 * We need to try and force an autonegotiation
3742 * session, then bring up link.
3744 if (hw->mac.ops.setup_sfp)
3745 hw->mac.ops.setup_sfp(hw);
3746 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
3747 schedule_work(&adapter->multispeed_fiber_task);
3750 * Direct Attach Cu and non-multispeed fiber modules
3751 * still need to be configured properly prior to
3754 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_MOD_TASK))
3755 schedule_work(&adapter->sfp_config_module_task);
3760 * ixgbe_non_sfp_link_config - set up non-SFP+ link
3761 * @hw: pointer to private hardware struct
3763 * Returns 0 on success, negative on failure
3765 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3768 bool negotiation, link_up = false;
3769 u32 ret = IXGBE_ERR_LINK_SETUP;
3771 if (hw->mac.ops.check_link)
3772 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3777 if (hw->mac.ops.get_link_capabilities)
3778 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
3783 if (hw->mac.ops.setup_link)
3784 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
3789 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
3791 struct ixgbe_hw *hw = &adapter->hw;
3794 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3795 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
3797 gpie |= IXGBE_GPIE_EIAME;
3799 * use EIAM to auto-mask when MSI-X interrupt is asserted
3800 * this saves a register write for every interrupt
3802 switch (hw->mac.type) {
3803 case ixgbe_mac_82598EB:
3804 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3806 case ixgbe_mac_82599EB:
3807 case ixgbe_mac_X540:
3809 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
3810 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
3814 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
3815 * specifically only auto mask tx and rx interrupts */
3816 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3819 /* XXX: to interrupt immediately for EICS writes, enable this */
3820 /* gpie |= IXGBE_GPIE_EIMEN; */
3822 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3823 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
3824 gpie |= IXGBE_GPIE_VTMODE_64;
3827 /* Enable fan failure interrupt */
3828 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3829 gpie |= IXGBE_SDP1_GPIEN;
3831 if (hw->mac.type == ixgbe_mac_82599EB)
3832 gpie |= IXGBE_SDP1_GPIEN;
3833 gpie |= IXGBE_SDP2_GPIEN;
3835 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3838 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
3840 struct ixgbe_hw *hw = &adapter->hw;
3844 ixgbe_get_hw_control(adapter);
3845 ixgbe_setup_gpie(adapter);
3847 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3848 ixgbe_configure_msix(adapter);
3850 ixgbe_configure_msi_and_legacy(adapter);
3852 /* enable the optics for both mult-speed fiber and 82599 SFP+ fiber */
3853 if (hw->mac.ops.enable_tx_laser &&
3854 ((hw->phy.multispeed_fiber) ||
3855 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
3856 (hw->mac.type == ixgbe_mac_82599EB))))
3857 hw->mac.ops.enable_tx_laser(hw);
3859 clear_bit(__IXGBE_DOWN, &adapter->state);
3860 ixgbe_napi_enable_all(adapter);
3862 if (ixgbe_is_sfp(hw)) {
3863 ixgbe_sfp_link_config(adapter);
3865 err = ixgbe_non_sfp_link_config(hw);
3867 e_err(probe, "link_config FAILED %d\n", err);
3870 /* clear any pending interrupts, may auto mask */
3871 IXGBE_READ_REG(hw, IXGBE_EICR);
3872 ixgbe_irq_enable(adapter, true, true);
3875 * If this adapter has a fan, check to see if we had a failure
3876 * before we enabled the interrupt.
3878 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3879 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
3880 if (esdp & IXGBE_ESDP_SDP1)
3881 e_crit(drv, "Fan has stopped, replace the adapter\n");
3885 * For hot-pluggable SFP+ devices, a new SFP+ module may have
3886 * arrived before interrupts were enabled but after probe. Such
3887 * devices wouldn't have their type identified yet. We need to
3888 * kick off the SFP+ module setup first, then try to bring up link.
3889 * If we're not hot-pluggable SFP+, we just need to configure link
3892 if (hw->phy.type == ixgbe_phy_unknown)
3893 schedule_work(&adapter->sfp_config_module_task);
3895 /* enable transmits */
3896 netif_tx_start_all_queues(adapter->netdev);
3898 /* bring the link up in the watchdog, this could race with our first
3899 * link up interrupt but shouldn't be a problem */
3900 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3901 adapter->link_check_timeout = jiffies;
3902 mod_timer(&adapter->watchdog_timer, jiffies);
3904 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
3905 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
3906 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
3907 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
3912 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
3914 WARN_ON(in_interrupt());
3915 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
3917 ixgbe_down(adapter);
3919 * If SR-IOV enabled then wait a bit before bringing the adapter
3920 * back up to give the VFs time to respond to the reset. The
3921 * two second wait is based upon the watchdog timer cycle in
3924 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3927 clear_bit(__IXGBE_RESETTING, &adapter->state);
3930 int ixgbe_up(struct ixgbe_adapter *adapter)
3932 /* hardware has been reset, we need to reload some things */
3933 ixgbe_configure(adapter);
3935 return ixgbe_up_complete(adapter);
3938 void ixgbe_reset(struct ixgbe_adapter *adapter)
3940 struct ixgbe_hw *hw = &adapter->hw;
3943 err = hw->mac.ops.init_hw(hw);
3946 case IXGBE_ERR_SFP_NOT_PRESENT:
3948 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
3949 e_dev_err("master disable timed out\n");
3951 case IXGBE_ERR_EEPROM_VERSION:
3952 /* We are running on a pre-production device, log a warning */
3953 e_dev_warn("This device is a pre-production adapter/LOM. "
3954 "Please be aware there may be issuesassociated with "
3955 "your hardware. If you are experiencing problems "
3956 "please contact your Intel or hardware "
3957 "representative who provided you with this "
3961 e_dev_err("Hardware Error: %d\n", err);
3964 /* reprogram the RAR[0] in case user changed it. */
3965 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
3970 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
3971 * @rx_ring: ring to free buffers from
3973 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
3975 struct device *dev = rx_ring->dev;
3979 /* ring already cleared, nothing to do */
3980 if (!rx_ring->rx_buffer_info)
3983 /* Free all the Rx ring sk_buffs */
3984 for (i = 0; i < rx_ring->count; i++) {
3985 struct ixgbe_rx_buffer *rx_buffer_info;
3987 rx_buffer_info = &rx_ring->rx_buffer_info[i];
3988 if (rx_buffer_info->dma) {
3989 dma_unmap_single(rx_ring->dev, rx_buffer_info->dma,
3990 rx_ring->rx_buf_len,
3992 rx_buffer_info->dma = 0;
3994 if (rx_buffer_info->skb) {
3995 struct sk_buff *skb = rx_buffer_info->skb;
3996 rx_buffer_info->skb = NULL;
3998 struct sk_buff *this = skb;
3999 if (IXGBE_RSC_CB(this)->delay_unmap) {
4000 dma_unmap_single(dev,
4001 IXGBE_RSC_CB(this)->dma,
4002 rx_ring->rx_buf_len,
4004 IXGBE_RSC_CB(this)->dma = 0;
4005 IXGBE_RSC_CB(skb)->delay_unmap = false;
4008 dev_kfree_skb(this);
4011 if (!rx_buffer_info->page)
4013 if (rx_buffer_info->page_dma) {
4014 dma_unmap_page(dev, rx_buffer_info->page_dma,
4015 PAGE_SIZE / 2, DMA_FROM_DEVICE);
4016 rx_buffer_info->page_dma = 0;
4018 put_page(rx_buffer_info->page);
4019 rx_buffer_info->page = NULL;
4020 rx_buffer_info->page_offset = 0;
4023 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4024 memset(rx_ring->rx_buffer_info, 0, size);
4026 /* Zero out the descriptor ring */
4027 memset(rx_ring->desc, 0, rx_ring->size);
4029 rx_ring->next_to_clean = 0;
4030 rx_ring->next_to_use = 0;
4034 * ixgbe_clean_tx_ring - Free Tx Buffers
4035 * @tx_ring: ring to be cleaned
4037 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
4039 struct ixgbe_tx_buffer *tx_buffer_info;
4043 /* ring already cleared, nothing to do */
4044 if (!tx_ring->tx_buffer_info)
4047 /* Free all the Tx ring sk_buffs */
4048 for (i = 0; i < tx_ring->count; i++) {
4049 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4050 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
4053 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4054 memset(tx_ring->tx_buffer_info, 0, size);
4056 /* Zero out the descriptor ring */
4057 memset(tx_ring->desc, 0, tx_ring->size);
4059 tx_ring->next_to_use = 0;
4060 tx_ring->next_to_clean = 0;
4064 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
4065 * @adapter: board private structure
4067 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
4071 for (i = 0; i < adapter->num_rx_queues; i++)
4072 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
4076 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
4077 * @adapter: board private structure
4079 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
4083 for (i = 0; i < adapter->num_tx_queues; i++)
4084 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
4087 void ixgbe_down(struct ixgbe_adapter *adapter)
4089 struct net_device *netdev = adapter->netdev;
4090 struct ixgbe_hw *hw = &adapter->hw;
4094 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4096 /* signal that we are down to the interrupt handler */
4097 set_bit(__IXGBE_DOWN, &adapter->state);
4099 /* disable receive for all VFs and wait one second */
4100 if (adapter->num_vfs) {
4101 /* ping all the active vfs to let them know we are going down */
4102 ixgbe_ping_all_vfs(adapter);
4104 /* Disable all VFTE/VFRE TX/RX */
4105 ixgbe_disable_tx_rx(adapter);
4107 /* Mark all the VFs as inactive */
4108 for (i = 0 ; i < adapter->num_vfs; i++)
4109 adapter->vfinfo[i].clear_to_send = 0;
4112 /* disable receives */
4113 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4114 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
4116 /* disable all enabled rx queues */
4117 for (i = 0; i < adapter->num_rx_queues; i++)
4118 /* this call also flushes the previous write */
4119 ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]);
4123 netif_tx_stop_all_queues(netdev);
4125 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
4126 del_timer_sync(&adapter->sfp_timer);
4127 del_timer_sync(&adapter->watchdog_timer);
4128 cancel_work_sync(&adapter->watchdog_task);
4130 netif_carrier_off(netdev);
4131 netif_tx_disable(netdev);
4133 ixgbe_irq_disable(adapter);
4135 ixgbe_napi_disable_all(adapter);
4137 /* Cleanup the affinity_hint CPU mask memory and callback */
4138 for (i = 0; i < num_q_vectors; i++) {
4139 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
4140 /* clear the affinity_mask in the IRQ descriptor */
4141 irq_set_affinity_hint(adapter->msix_entries[i]. vector, NULL);
4142 /* release the CPU mask memory */
4143 free_cpumask_var(q_vector->affinity_mask);
4146 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
4147 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
4148 cancel_work_sync(&adapter->fdir_reinit_task);
4150 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
4151 cancel_work_sync(&adapter->check_overtemp_task);
4153 /* disable transmits in the hardware now that interrupts are off */
4154 for (i = 0; i < adapter->num_tx_queues; i++) {
4155 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
4156 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
4157 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx),
4158 (txdctl & ~IXGBE_TXDCTL_ENABLE));
4160 /* Disable the Tx DMA engine on 82599 */
4161 switch (hw->mac.type) {
4162 case ixgbe_mac_82599EB:
4163 case ixgbe_mac_X540:
4164 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
4165 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
4166 ~IXGBE_DMATXCTL_TE));
4172 /* clear n-tuple filters that are cached */
4173 ethtool_ntuple_flush(netdev);
4175 if (!pci_channel_offline(adapter->pdev))
4176 ixgbe_reset(adapter);
4178 /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
4179 if (hw->mac.ops.disable_tx_laser &&
4180 ((hw->phy.multispeed_fiber) ||
4181 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
4182 (hw->mac.type == ixgbe_mac_82599EB))))
4183 hw->mac.ops.disable_tx_laser(hw);
4185 ixgbe_clean_all_tx_rings(adapter);
4186 ixgbe_clean_all_rx_rings(adapter);
4188 #ifdef CONFIG_IXGBE_DCA
4189 /* since we reset the hardware DCA settings were cleared */
4190 ixgbe_setup_dca(adapter);
4195 * ixgbe_poll - NAPI Rx polling callback
4196 * @napi: structure for representing this polling device
4197 * @budget: how many packets driver is allowed to clean
4199 * This function is used for legacy and MSI, NAPI mode
4201 static int ixgbe_poll(struct napi_struct *napi, int budget)
4203 struct ixgbe_q_vector *q_vector =
4204 container_of(napi, struct ixgbe_q_vector, napi);
4205 struct ixgbe_adapter *adapter = q_vector->adapter;
4206 int tx_clean_complete, work_done = 0;
4208 #ifdef CONFIG_IXGBE_DCA
4209 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
4210 ixgbe_update_dca(q_vector);
4213 tx_clean_complete = ixgbe_clean_tx_irq(q_vector, adapter->tx_ring[0]);
4214 ixgbe_clean_rx_irq(q_vector, adapter->rx_ring[0], &work_done, budget);
4216 if (!tx_clean_complete)
4219 /* If budget not fully consumed, exit the polling mode */
4220 if (work_done < budget) {
4221 napi_complete(napi);
4222 if (adapter->rx_itr_setting & 1)
4223 ixgbe_set_itr(adapter);
4224 if (!test_bit(__IXGBE_DOWN, &adapter->state))
4225 ixgbe_irq_enable_queues(adapter, IXGBE_EIMS_RTX_QUEUE);
4231 * ixgbe_tx_timeout - Respond to a Tx Hang
4232 * @netdev: network interface device structure
4234 static void ixgbe_tx_timeout(struct net_device *netdev)
4236 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4238 adapter->tx_timeout_count++;
4240 /* Do the reset outside of interrupt context */
4241 schedule_work(&adapter->reset_task);
4244 static void ixgbe_reset_task(struct work_struct *work)
4246 struct ixgbe_adapter *adapter;
4247 adapter = container_of(work, struct ixgbe_adapter, reset_task);
4249 /* If we're already down or resetting, just bail */
4250 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
4251 test_bit(__IXGBE_RESETTING, &adapter->state))
4254 ixgbe_dump(adapter);
4255 netdev_err(adapter->netdev, "Reset adapter\n");
4256 ixgbe_reinit_locked(adapter);
4259 #ifdef CONFIG_IXGBE_DCB
4260 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
4263 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_DCB];
4265 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
4269 adapter->num_rx_queues = f->indices;
4270 adapter->num_tx_queues = f->indices;
4278 * ixgbe_set_rss_queues: Allocate queues for RSS
4279 * @adapter: board private structure to initialize
4281 * This is our "base" multiqueue mode. RSS (Receive Side Scaling) will try
4282 * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
4285 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
4288 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
4290 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4292 adapter->num_rx_queues = f->indices;
4293 adapter->num_tx_queues = f->indices;
4303 * ixgbe_set_fdir_queues: Allocate queues for Flow Director
4304 * @adapter: board private structure to initialize
4306 * Flow Director is an advanced Rx filter, attempting to get Rx flows back
4307 * to the original CPU that initiated the Tx session. This runs in addition
4308 * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
4309 * Rx load across CPUs using RSS.
4312 static inline bool ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
4315 struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
4317 f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
4320 /* Flow Director must have RSS enabled */
4321 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
4322 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
4323 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)))) {
4324 adapter->num_tx_queues = f_fdir->indices;
4325 adapter->num_rx_queues = f_fdir->indices;
4328 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4329 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4336 * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
4337 * @adapter: board private structure to initialize
4339 * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
4340 * The ring feature mask is not used as a mask for FCoE, as it can take any 8
4341 * rx queues out of the max number of rx queues, instead, it is used as the
4342 * index of the first rx queue used by FCoE.
4345 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
4348 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4350 f->indices = min((int)num_online_cpus(), f->indices);
4351 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
4352 adapter->num_rx_queues = 1;
4353 adapter->num_tx_queues = 1;
4354 #ifdef CONFIG_IXGBE_DCB
4355 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4356 e_info(probe, "FCoE enabled with DCB\n");
4357 ixgbe_set_dcb_queues(adapter);
4360 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4361 e_info(probe, "FCoE enabled with RSS\n");
4362 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4363 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
4364 ixgbe_set_fdir_queues(adapter);
4366 ixgbe_set_rss_queues(adapter);
4368 /* adding FCoE rx rings to the end */
4369 f->mask = adapter->num_rx_queues;
4370 adapter->num_rx_queues += f->indices;
4371 adapter->num_tx_queues += f->indices;
4379 #endif /* IXGBE_FCOE */
4381 * ixgbe_set_sriov_queues: Allocate queues for IOV use
4382 * @adapter: board private structure to initialize
4384 * IOV doesn't actually use anything, so just NAK the
4385 * request for now and let the other queue routines
4386 * figure out what to do.
4388 static inline bool ixgbe_set_sriov_queues(struct ixgbe_adapter *adapter)
4394 * ixgbe_set_num_queues: Allocate queues for device, feature dependant
4395 * @adapter: board private structure to initialize
4397 * This is the top level queue allocation routine. The order here is very
4398 * important, starting with the "most" number of features turned on at once,
4399 * and ending with the smallest set of features. This way large combinations
4400 * can be allocated if they're turned on, and smaller combinations are the
4401 * fallthrough conditions.
4404 static int ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
4406 /* Start with base case */
4407 adapter->num_rx_queues = 1;
4408 adapter->num_tx_queues = 1;
4409 adapter->num_rx_pools = adapter->num_rx_queues;
4410 adapter->num_rx_queues_per_pool = 1;
4412 if (ixgbe_set_sriov_queues(adapter))
4416 if (ixgbe_set_fcoe_queues(adapter))
4419 #endif /* IXGBE_FCOE */
4420 #ifdef CONFIG_IXGBE_DCB
4421 if (ixgbe_set_dcb_queues(adapter))
4425 if (ixgbe_set_fdir_queues(adapter))
4428 if (ixgbe_set_rss_queues(adapter))
4431 /* fallback to base case */
4432 adapter->num_rx_queues = 1;
4433 adapter->num_tx_queues = 1;
4436 /* Notify the stack of the (possibly) reduced queue counts. */
4437 netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
4438 return netif_set_real_num_rx_queues(adapter->netdev,
4439 adapter->num_rx_queues);
4442 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
4445 int err, vector_threshold;
4447 /* We'll want at least 3 (vector_threshold):
4450 * 3) Other (Link Status Change, etc.)
4451 * 4) TCP Timer (optional)
4453 vector_threshold = MIN_MSIX_COUNT;
4455 /* The more we get, the more we will assign to Tx/Rx Cleanup
4456 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
4457 * Right now, we simply care about how many we'll get; we'll
4458 * set them up later while requesting irq's.
4460 while (vectors >= vector_threshold) {
4461 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
4463 if (!err) /* Success in acquiring all requested vectors. */
4466 vectors = 0; /* Nasty failure, quit now */
4467 else /* err == number of vectors we should try again with */
4471 if (vectors < vector_threshold) {
4472 /* Can't allocate enough MSI-X interrupts? Oh well.
4473 * This just means we'll go with either a single MSI
4474 * vector or fall back to legacy interrupts.
4476 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4477 "Unable to allocate MSI-X interrupts\n");
4478 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4479 kfree(adapter->msix_entries);
4480 adapter->msix_entries = NULL;
4482 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
4484 * Adjust for only the vectors we'll use, which is minimum
4485 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
4486 * vectors we were allocated.
4488 adapter->num_msix_vectors = min(vectors,
4489 adapter->max_msix_q_vectors + NON_Q_VECTORS);
4494 * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
4495 * @adapter: board private structure to initialize
4497 * Cache the descriptor ring offsets for RSS to the assigned rings.
4500 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
4504 if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED))
4507 for (i = 0; i < adapter->num_rx_queues; i++)
4508 adapter->rx_ring[i]->reg_idx = i;
4509 for (i = 0; i < adapter->num_tx_queues; i++)
4510 adapter->tx_ring[i]->reg_idx = i;
4515 #ifdef CONFIG_IXGBE_DCB
4517 * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
4518 * @adapter: board private structure to initialize
4520 * Cache the descriptor ring offsets for DCB to the assigned rings.
4523 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
4527 int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
4529 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
4532 /* the number of queues is assumed to be symmetric */
4533 switch (adapter->hw.mac.type) {
4534 case ixgbe_mac_82598EB:
4535 for (i = 0; i < dcb_i; i++) {
4536 adapter->rx_ring[i]->reg_idx = i << 3;
4537 adapter->tx_ring[i]->reg_idx = i << 2;
4541 case ixgbe_mac_82599EB:
4542 case ixgbe_mac_X540:
4545 * Tx TC0 starts at: descriptor queue 0
4546 * Tx TC1 starts at: descriptor queue 32
4547 * Tx TC2 starts at: descriptor queue 64
4548 * Tx TC3 starts at: descriptor queue 80
4549 * Tx TC4 starts at: descriptor queue 96
4550 * Tx TC5 starts at: descriptor queue 104
4551 * Tx TC6 starts at: descriptor queue 112
4552 * Tx TC7 starts at: descriptor queue 120
4554 * Rx TC0-TC7 are offset by 16 queues each
4556 for (i = 0; i < 3; i++) {
4557 adapter->tx_ring[i]->reg_idx = i << 5;
4558 adapter->rx_ring[i]->reg_idx = i << 4;
4560 for ( ; i < 5; i++) {
4561 adapter->tx_ring[i]->reg_idx = ((i + 2) << 4);
4562 adapter->rx_ring[i]->reg_idx = i << 4;
4564 for ( ; i < dcb_i; i++) {
4565 adapter->tx_ring[i]->reg_idx = ((i + 8) << 3);
4566 adapter->rx_ring[i]->reg_idx = i << 4;
4569 } else if (dcb_i == 4) {
4571 * Tx TC0 starts at: descriptor queue 0
4572 * Tx TC1 starts at: descriptor queue 64
4573 * Tx TC2 starts at: descriptor queue 96
4574 * Tx TC3 starts at: descriptor queue 112
4576 * Rx TC0-TC3 are offset by 32 queues each
4578 adapter->tx_ring[0]->reg_idx = 0;
4579 adapter->tx_ring[1]->reg_idx = 64;
4580 adapter->tx_ring[2]->reg_idx = 96;
4581 adapter->tx_ring[3]->reg_idx = 112;
4582 for (i = 0 ; i < dcb_i; i++)
4583 adapter->rx_ring[i]->reg_idx = i << 5;
4595 * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
4596 * @adapter: board private structure to initialize
4598 * Cache the descriptor ring offsets for Flow Director to the assigned rings.
4601 static inline bool ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
4606 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
4607 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4608 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))) {
4609 for (i = 0; i < adapter->num_rx_queues; i++)
4610 adapter->rx_ring[i]->reg_idx = i;
4611 for (i = 0; i < adapter->num_tx_queues; i++)
4612 adapter->tx_ring[i]->reg_idx = i;
4621 * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
4622 * @adapter: board private structure to initialize
4624 * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
4627 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
4629 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4631 u8 fcoe_rx_i = 0, fcoe_tx_i = 0;
4633 if (!(adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
4636 #ifdef CONFIG_IXGBE_DCB
4637 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4638 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
4640 ixgbe_cache_ring_dcb(adapter);
4641 /* find out queues in TC for FCoE */
4642 fcoe_rx_i = adapter->rx_ring[fcoe->tc]->reg_idx + 1;
4643 fcoe_tx_i = adapter->tx_ring[fcoe->tc]->reg_idx + 1;
4645 * In 82599, the number of Tx queues for each traffic
4646 * class for both 8-TC and 4-TC modes are:
4647 * TCs : TC0 TC1 TC2 TC3 TC4 TC5 TC6 TC7
4648 * 8 TCs: 32 32 16 16 8 8 8 8
4649 * 4 TCs: 64 64 32 32
4650 * We have max 8 queues for FCoE, where 8 the is
4651 * FCoE redirection table size. If TC for FCoE is
4652 * less than or equal to TC3, we have enough queues
4653 * to add max of 8 queues for FCoE, so we start FCoE
4654 * Tx queue from the next one, i.e., reg_idx + 1.
4655 * If TC for FCoE is above TC3, implying 8 TC mode,
4656 * and we need 8 for FCoE, we have to take all queues
4657 * in that traffic class for FCoE.
4659 if ((f->indices == IXGBE_FCRETA_SIZE) && (fcoe->tc > 3))
4662 #endif /* CONFIG_IXGBE_DCB */
4663 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4664 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4665 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
4666 ixgbe_cache_ring_fdir(adapter);
4668 ixgbe_cache_ring_rss(adapter);
4670 fcoe_rx_i = f->mask;
4671 fcoe_tx_i = f->mask;
4673 for (i = 0; i < f->indices; i++, fcoe_rx_i++, fcoe_tx_i++) {
4674 adapter->rx_ring[f->mask + i]->reg_idx = fcoe_rx_i;
4675 adapter->tx_ring[f->mask + i]->reg_idx = fcoe_tx_i;
4680 #endif /* IXGBE_FCOE */
4682 * ixgbe_cache_ring_sriov - Descriptor ring to register mapping for sriov
4683 * @adapter: board private structure to initialize
4685 * SR-IOV doesn't use any descriptor rings but changes the default if
4686 * no other mapping is used.
4689 static inline bool ixgbe_cache_ring_sriov(struct ixgbe_adapter *adapter)
4691 adapter->rx_ring[0]->reg_idx = adapter->num_vfs * 2;
4692 adapter->tx_ring[0]->reg_idx = adapter->num_vfs * 2;
4693 if (adapter->num_vfs)
4700 * ixgbe_cache_ring_register - Descriptor ring to register mapping
4701 * @adapter: board private structure to initialize
4703 * Once we know the feature-set enabled for the device, we'll cache
4704 * the register offset the descriptor ring is assigned to.
4706 * Note, the order the various feature calls is important. It must start with
4707 * the "most" features enabled at the same time, then trickle down to the
4708 * least amount of features turned on at once.
4710 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
4712 /* start with default case */
4713 adapter->rx_ring[0]->reg_idx = 0;
4714 adapter->tx_ring[0]->reg_idx = 0;
4716 if (ixgbe_cache_ring_sriov(adapter))
4720 if (ixgbe_cache_ring_fcoe(adapter))
4723 #endif /* IXGBE_FCOE */
4724 #ifdef CONFIG_IXGBE_DCB
4725 if (ixgbe_cache_ring_dcb(adapter))
4729 if (ixgbe_cache_ring_fdir(adapter))
4732 if (ixgbe_cache_ring_rss(adapter))
4737 * ixgbe_alloc_queues - Allocate memory for all rings
4738 * @adapter: board private structure to initialize
4740 * We allocate one ring per queue at run-time since we don't know the
4741 * number of queues at compile-time. The polling_netdev array is
4742 * intended for Multiqueue, but should work fine with a single queue.
4744 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
4746 int rx = 0, tx = 0, nid = adapter->node;
4748 if (nid < 0 || !node_online(nid))
4749 nid = first_online_node;
4751 for (; tx < adapter->num_tx_queues; tx++) {
4752 struct ixgbe_ring *ring;
4754 ring = kzalloc_node(sizeof(*ring), GFP_KERNEL, nid);
4756 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
4758 goto err_allocation;
4759 ring->count = adapter->tx_ring_count;
4760 ring->queue_index = tx;
4761 ring->numa_node = nid;
4762 ring->dev = &adapter->pdev->dev;
4763 ring->netdev = adapter->netdev;
4765 adapter->tx_ring[tx] = ring;
4768 for (; rx < adapter->num_rx_queues; rx++) {
4769 struct ixgbe_ring *ring;
4771 ring = kzalloc_node(sizeof(*ring), GFP_KERNEL, nid);
4773 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
4775 goto err_allocation;
4776 ring->count = adapter->rx_ring_count;
4777 ring->queue_index = rx;
4778 ring->numa_node = nid;
4779 ring->dev = &adapter->pdev->dev;
4780 ring->netdev = adapter->netdev;
4782 adapter->rx_ring[rx] = ring;
4785 ixgbe_cache_ring_register(adapter);
4791 kfree(adapter->tx_ring[--tx]);
4794 kfree(adapter->rx_ring[--rx]);
4799 * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
4800 * @adapter: board private structure to initialize
4802 * Attempt to configure the interrupts using the best available
4803 * capabilities of the hardware and the kernel.
4805 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
4807 struct ixgbe_hw *hw = &adapter->hw;
4809 int vector, v_budget;
4812 * It's easy to be greedy for MSI-X vectors, but it really
4813 * doesn't do us much good if we have a lot more vectors
4814 * than CPU's. So let's be conservative and only ask for
4815 * (roughly) the same number of vectors as there are CPU's.
4817 v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
4818 (int)num_online_cpus()) + NON_Q_VECTORS;
4821 * At the same time, hardware can only support a maximum of
4822 * hw.mac->max_msix_vectors vectors. With features
4823 * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
4824 * descriptor queues supported by our device. Thus, we cap it off in
4825 * those rare cases where the cpu count also exceeds our vector limit.
4827 v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
4829 /* A failure in MSI-X entry allocation isn't fatal, but it does
4830 * mean we disable MSI-X capabilities of the adapter. */
4831 adapter->msix_entries = kcalloc(v_budget,
4832 sizeof(struct msix_entry), GFP_KERNEL);
4833 if (adapter->msix_entries) {
4834 for (vector = 0; vector < v_budget; vector++)
4835 adapter->msix_entries[vector].entry = vector;
4837 ixgbe_acquire_msix_vectors(adapter, v_budget);
4839 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4843 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
4844 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
4845 if (adapter->flags & (IXGBE_FLAG_FDIR_HASH_CAPABLE |
4846 IXGBE_FLAG_FDIR_PERFECT_CAPABLE)) {
4848 "Flow Director is not supported while multiple "
4849 "queues are disabled. Disabling Flow Director\n");
4851 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4852 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4853 adapter->atr_sample_rate = 0;
4854 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4855 ixgbe_disable_sriov(adapter);
4857 err = ixgbe_set_num_queues(adapter);
4861 err = pci_enable_msi(adapter->pdev);
4863 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
4865 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4866 "Unable to allocate MSI interrupt, "
4867 "falling back to legacy. Error: %d\n", err);
4877 * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
4878 * @adapter: board private structure to initialize
4880 * We allocate one q_vector per queue interrupt. If allocation fails we
4883 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
4885 int q_idx, num_q_vectors;
4886 struct ixgbe_q_vector *q_vector;
4887 int (*poll)(struct napi_struct *, int);
4889 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4890 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4891 poll = &ixgbe_clean_rxtx_many;
4897 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4898 q_vector = kzalloc_node(sizeof(struct ixgbe_q_vector),
4899 GFP_KERNEL, adapter->node);
4901 q_vector = kzalloc(sizeof(struct ixgbe_q_vector),
4905 q_vector->adapter = adapter;
4906 if (q_vector->txr_count && !q_vector->rxr_count)
4907 q_vector->eitr = adapter->tx_eitr_param;
4909 q_vector->eitr = adapter->rx_eitr_param;
4910 q_vector->v_idx = q_idx;
4911 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
4912 adapter->q_vector[q_idx] = q_vector;
4920 q_vector = adapter->q_vector[q_idx];
4921 netif_napi_del(&q_vector->napi);
4923 adapter->q_vector[q_idx] = NULL;
4929 * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
4930 * @adapter: board private structure to initialize
4932 * This function frees the memory allocated to the q_vectors. In addition if
4933 * NAPI is enabled it will delete any references to the NAPI struct prior
4934 * to freeing the q_vector.
4936 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
4938 int q_idx, num_q_vectors;
4940 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4941 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4945 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4946 struct ixgbe_q_vector *q_vector = adapter->q_vector[q_idx];
4947 adapter->q_vector[q_idx] = NULL;
4948 netif_napi_del(&q_vector->napi);
4953 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
4955 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4956 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4957 pci_disable_msix(adapter->pdev);
4958 kfree(adapter->msix_entries);
4959 adapter->msix_entries = NULL;
4960 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
4961 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
4962 pci_disable_msi(adapter->pdev);
4967 * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
4968 * @adapter: board private structure to initialize
4970 * We determine which interrupt scheme to use based on...
4971 * - Kernel support (MSI, MSI-X)
4972 * - which can be user-defined (via MODULE_PARAM)
4973 * - Hardware queue count (num_*_queues)
4974 * - defined by miscellaneous hardware support/features (RSS, etc.)
4976 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
4980 /* Number of supported queues */
4981 err = ixgbe_set_num_queues(adapter);
4985 err = ixgbe_set_interrupt_capability(adapter);
4987 e_dev_err("Unable to setup interrupt capabilities\n");
4988 goto err_set_interrupt;
4991 err = ixgbe_alloc_q_vectors(adapter);
4993 e_dev_err("Unable to allocate memory for queue vectors\n");
4994 goto err_alloc_q_vectors;
4997 err = ixgbe_alloc_queues(adapter);
4999 e_dev_err("Unable to allocate memory for queues\n");
5000 goto err_alloc_queues;
5003 e_dev_info("Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
5004 (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
5005 adapter->num_rx_queues, adapter->num_tx_queues);
5007 set_bit(__IXGBE_DOWN, &adapter->state);
5012 ixgbe_free_q_vectors(adapter);
5013 err_alloc_q_vectors:
5014 ixgbe_reset_interrupt_capability(adapter);
5019 static void ring_free_rcu(struct rcu_head *head)
5021 kfree(container_of(head, struct ixgbe_ring, rcu));
5025 * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
5026 * @adapter: board private structure to clear interrupt scheme on
5028 * We go through and clear interrupt specific resources and reset the structure
5029 * to pre-load conditions
5031 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
5035 for (i = 0; i < adapter->num_tx_queues; i++) {
5036 kfree(adapter->tx_ring[i]);
5037 adapter->tx_ring[i] = NULL;
5039 for (i = 0; i < adapter->num_rx_queues; i++) {
5040 struct ixgbe_ring *ring = adapter->rx_ring[i];
5042 /* ixgbe_get_stats64() might access this ring, we must wait
5043 * a grace period before freeing it.
5045 call_rcu(&ring->rcu, ring_free_rcu);
5046 adapter->rx_ring[i] = NULL;
5049 adapter->num_tx_queues = 0;
5050 adapter->num_rx_queues = 0;
5052 ixgbe_free_q_vectors(adapter);
5053 ixgbe_reset_interrupt_capability(adapter);
5057 * ixgbe_sfp_timer - worker thread to find a missing module
5058 * @data: pointer to our adapter struct
5060 static void ixgbe_sfp_timer(unsigned long data)
5062 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5065 * Do the sfp_timer outside of interrupt context due to the
5066 * delays that sfp+ detection requires
5068 schedule_work(&adapter->sfp_task);
5072 * ixgbe_sfp_task - worker thread to find a missing module
5073 * @work: pointer to work_struct containing our data
5075 static void ixgbe_sfp_task(struct work_struct *work)
5077 struct ixgbe_adapter *adapter = container_of(work,
5078 struct ixgbe_adapter,
5080 struct ixgbe_hw *hw = &adapter->hw;
5082 if ((hw->phy.type == ixgbe_phy_nl) &&
5083 (hw->phy.sfp_type == ixgbe_sfp_type_not_present)) {
5084 s32 ret = hw->phy.ops.identify_sfp(hw);
5085 if (ret == IXGBE_ERR_SFP_NOT_PRESENT)
5087 ret = hw->phy.ops.reset(hw);
5088 if (ret == IXGBE_ERR_SFP_NOT_SUPPORTED) {
5089 e_dev_err("failed to initialize because an unsupported "
5090 "SFP+ module type was detected.\n");
5091 e_dev_err("Reload the driver after installing a "
5092 "supported module.\n");
5093 unregister_netdev(adapter->netdev);
5095 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
5097 /* don't need this routine any more */
5098 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
5102 if (test_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state))
5103 mod_timer(&adapter->sfp_timer,
5104 round_jiffies(jiffies + (2 * HZ)));
5108 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
5109 * @adapter: board private structure to initialize
5111 * ixgbe_sw_init initializes the Adapter private data structure.
5112 * Fields are initialized based on PCI device information and
5113 * OS network device settings (MTU size).
5115 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
5117 struct ixgbe_hw *hw = &adapter->hw;
5118 struct pci_dev *pdev = adapter->pdev;
5119 struct net_device *dev = adapter->netdev;
5121 #ifdef CONFIG_IXGBE_DCB
5123 struct tc_configuration *tc;
5125 int max_frame = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
5127 /* PCI config space info */
5129 hw->vendor_id = pdev->vendor;
5130 hw->device_id = pdev->device;
5131 hw->revision_id = pdev->revision;
5132 hw->subsystem_vendor_id = pdev->subsystem_vendor;
5133 hw->subsystem_device_id = pdev->subsystem_device;
5135 /* Set capability flags */
5136 rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
5137 adapter->ring_feature[RING_F_RSS].indices = rss;
5138 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
5139 adapter->ring_feature[RING_F_DCB].indices = IXGBE_MAX_DCB_INDICES;
5140 switch (hw->mac.type) {
5141 case ixgbe_mac_82598EB:
5142 if (hw->device_id == IXGBE_DEV_ID_82598AT)
5143 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
5144 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
5146 case ixgbe_mac_82599EB:
5147 case ixgbe_mac_X540:
5148 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
5149 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
5150 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
5151 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
5152 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
5153 /* n-tuple support exists, always init our spinlock */
5154 spin_lock_init(&adapter->fdir_perfect_lock);
5155 /* Flow Director hash filters enabled */
5156 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
5157 adapter->atr_sample_rate = 20;
5158 adapter->ring_feature[RING_F_FDIR].indices =
5159 IXGBE_MAX_FDIR_INDICES;
5160 adapter->fdir_pballoc = 0;
5162 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
5163 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
5164 adapter->ring_feature[RING_F_FCOE].indices = 0;
5165 #ifdef CONFIG_IXGBE_DCB
5166 /* Default traffic class to use for FCoE */
5167 adapter->fcoe.tc = IXGBE_FCOE_DEFTC;
5168 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
5170 #endif /* IXGBE_FCOE */
5176 #ifdef CONFIG_IXGBE_DCB
5177 /* Configure DCB traffic classes */
5178 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
5179 tc = &adapter->dcb_cfg.tc_config[j];
5180 tc->path[DCB_TX_CONFIG].bwg_id = 0;
5181 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
5182 tc->path[DCB_RX_CONFIG].bwg_id = 0;
5183 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
5184 tc->dcb_pfc = pfc_disabled;
5186 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
5187 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
5188 adapter->dcb_cfg.rx_pba_cfg = pba_equal;
5189 adapter->dcb_cfg.pfc_mode_enable = false;
5190 adapter->dcb_set_bitmap = 0x00;
5191 ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
5192 adapter->ring_feature[RING_F_DCB].indices);
5196 /* default flow control settings */
5197 hw->fc.requested_mode = ixgbe_fc_full;
5198 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
5200 adapter->last_lfc_mode = hw->fc.current_mode;
5202 hw->fc.high_water = FC_HIGH_WATER(max_frame);
5203 hw->fc.low_water = FC_LOW_WATER(max_frame);
5204 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
5205 hw->fc.send_xon = true;
5206 hw->fc.disable_fc_autoneg = false;
5208 /* enable itr by default in dynamic mode */
5209 adapter->rx_itr_setting = 1;
5210 adapter->rx_eitr_param = 20000;
5211 adapter->tx_itr_setting = 1;
5212 adapter->tx_eitr_param = 10000;
5214 /* set defaults for eitr in MegaBytes */
5215 adapter->eitr_low = 10;
5216 adapter->eitr_high = 20;
5218 /* set default ring sizes */
5219 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
5220 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
5222 /* initialize eeprom parameters */
5223 if (ixgbe_init_eeprom_params_generic(hw)) {
5224 e_dev_err("EEPROM initialization failed\n");
5228 /* enable rx csum by default */
5229 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
5231 /* get assigned NUMA node */
5232 adapter->node = dev_to_node(&pdev->dev);
5234 set_bit(__IXGBE_DOWN, &adapter->state);
5240 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
5241 * @tx_ring: tx descriptor ring (for a specific queue) to setup
5243 * Return 0 on success, negative on failure
5245 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
5247 struct device *dev = tx_ring->dev;
5250 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
5251 tx_ring->tx_buffer_info = vzalloc_node(size, tx_ring->numa_node);
5252 if (!tx_ring->tx_buffer_info)
5253 tx_ring->tx_buffer_info = vzalloc(size);
5254 if (!tx_ring->tx_buffer_info)
5257 /* round up to nearest 4K */
5258 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
5259 tx_ring->size = ALIGN(tx_ring->size, 4096);
5261 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
5262 &tx_ring->dma, GFP_KERNEL);
5266 tx_ring->next_to_use = 0;
5267 tx_ring->next_to_clean = 0;
5268 tx_ring->work_limit = tx_ring->count;
5272 vfree(tx_ring->tx_buffer_info);
5273 tx_ring->tx_buffer_info = NULL;
5274 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
5279 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
5280 * @adapter: board private structure
5282 * If this function returns with an error, then it's possible one or
5283 * more of the rings is populated (while the rest are not). It is the
5284 * callers duty to clean those orphaned rings.
5286 * Return 0 on success, negative on failure
5288 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
5292 for (i = 0; i < adapter->num_tx_queues; i++) {
5293 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
5296 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
5304 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
5305 * @rx_ring: rx descriptor ring (for a specific queue) to setup
5307 * Returns 0 on success, negative on failure
5309 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
5311 struct device *dev = rx_ring->dev;
5314 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
5315 rx_ring->rx_buffer_info = vzalloc_node(size, rx_ring->numa_node);
5316 if (!rx_ring->rx_buffer_info)
5317 rx_ring->rx_buffer_info = vzalloc(size);
5318 if (!rx_ring->rx_buffer_info)
5321 /* Round up to nearest 4K */
5322 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
5323 rx_ring->size = ALIGN(rx_ring->size, 4096);
5325 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
5326 &rx_ring->dma, GFP_KERNEL);
5331 rx_ring->next_to_clean = 0;
5332 rx_ring->next_to_use = 0;
5336 vfree(rx_ring->rx_buffer_info);
5337 rx_ring->rx_buffer_info = NULL;
5338 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
5343 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
5344 * @adapter: board private structure
5346 * If this function returns with an error, then it's possible one or
5347 * more of the rings is populated (while the rest are not). It is the
5348 * callers duty to clean those orphaned rings.
5350 * Return 0 on success, negative on failure
5352 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
5356 for (i = 0; i < adapter->num_rx_queues; i++) {
5357 err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
5360 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
5368 * ixgbe_free_tx_resources - Free Tx Resources per Queue
5369 * @tx_ring: Tx descriptor ring for a specific queue
5371 * Free all transmit software resources
5373 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
5375 ixgbe_clean_tx_ring(tx_ring);
5377 vfree(tx_ring->tx_buffer_info);
5378 tx_ring->tx_buffer_info = NULL;
5380 /* if not set, then don't free */
5384 dma_free_coherent(tx_ring->dev, tx_ring->size,
5385 tx_ring->desc, tx_ring->dma);
5387 tx_ring->desc = NULL;
5391 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5392 * @adapter: board private structure
5394 * Free all transmit software resources
5396 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
5400 for (i = 0; i < adapter->num_tx_queues; i++)
5401 if (adapter->tx_ring[i]->desc)
5402 ixgbe_free_tx_resources(adapter->tx_ring[i]);
5406 * ixgbe_free_rx_resources - Free Rx Resources
5407 * @rx_ring: ring to clean the resources from
5409 * Free all receive software resources
5411 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
5413 ixgbe_clean_rx_ring(rx_ring);
5415 vfree(rx_ring->rx_buffer_info);
5416 rx_ring->rx_buffer_info = NULL;
5418 /* if not set, then don't free */
5422 dma_free_coherent(rx_ring->dev, rx_ring->size,
5423 rx_ring->desc, rx_ring->dma);
5425 rx_ring->desc = NULL;
5429 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5430 * @adapter: board private structure
5432 * Free all receive software resources
5434 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5438 for (i = 0; i < adapter->num_rx_queues; i++)
5439 if (adapter->rx_ring[i]->desc)
5440 ixgbe_free_rx_resources(adapter->rx_ring[i]);
5444 * ixgbe_change_mtu - Change the Maximum Transfer Unit
5445 * @netdev: network interface device structure
5446 * @new_mtu: new value for maximum frame size
5448 * Returns 0 on success, negative on failure
5450 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5452 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5453 struct ixgbe_hw *hw = &adapter->hw;
5454 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5456 /* MTU < 68 is an error and causes problems on some kernels */
5457 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED &&
5458 hw->mac.type != ixgbe_mac_X540) {
5459 if ((new_mtu < 68) || (max_frame > MAXIMUM_ETHERNET_VLAN_SIZE))
5462 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5466 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5467 /* must set new MTU before calling down or up */
5468 netdev->mtu = new_mtu;
5470 hw->fc.high_water = FC_HIGH_WATER(max_frame);
5471 hw->fc.low_water = FC_LOW_WATER(max_frame);
5473 if (netif_running(netdev))
5474 ixgbe_reinit_locked(adapter);
5480 * ixgbe_open - Called when a network interface is made active
5481 * @netdev: network interface device structure
5483 * Returns 0 on success, negative value on failure
5485 * The open entry point is called when a network interface is made
5486 * active by the system (IFF_UP). At this point all resources needed
5487 * for transmit and receive operations are allocated, the interrupt
5488 * handler is registered with the OS, the watchdog timer is started,
5489 * and the stack is notified that the interface is ready.
5491 static int ixgbe_open(struct net_device *netdev)
5493 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5496 /* disallow open during test */
5497 if (test_bit(__IXGBE_TESTING, &adapter->state))
5500 netif_carrier_off(netdev);
5502 /* allocate transmit descriptors */
5503 err = ixgbe_setup_all_tx_resources(adapter);
5507 /* allocate receive descriptors */
5508 err = ixgbe_setup_all_rx_resources(adapter);
5512 ixgbe_configure(adapter);
5514 err = ixgbe_request_irq(adapter);
5518 err = ixgbe_up_complete(adapter);
5522 netif_tx_start_all_queues(netdev);
5527 ixgbe_release_hw_control(adapter);
5528 ixgbe_free_irq(adapter);
5531 ixgbe_free_all_rx_resources(adapter);
5533 ixgbe_free_all_tx_resources(adapter);
5534 ixgbe_reset(adapter);
5540 * ixgbe_close - Disables a network interface
5541 * @netdev: network interface device structure
5543 * Returns 0, this is not allowed to fail
5545 * The close entry point is called when an interface is de-activated
5546 * by the OS. The hardware is still under the drivers control, but
5547 * needs to be disabled. A global MAC reset is issued to stop the
5548 * hardware, and all transmit and receive resources are freed.
5550 static int ixgbe_close(struct net_device *netdev)
5552 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5554 ixgbe_down(adapter);
5555 ixgbe_free_irq(adapter);
5557 ixgbe_free_all_tx_resources(adapter);
5558 ixgbe_free_all_rx_resources(adapter);
5560 ixgbe_release_hw_control(adapter);
5566 static int ixgbe_resume(struct pci_dev *pdev)
5568 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5569 struct net_device *netdev = adapter->netdev;
5572 pci_set_power_state(pdev, PCI_D0);
5573 pci_restore_state(pdev);
5575 * pci_restore_state clears dev->state_saved so call
5576 * pci_save_state to restore it.
5578 pci_save_state(pdev);
5580 err = pci_enable_device_mem(pdev);
5582 e_dev_err("Cannot enable PCI device from suspend\n");
5585 pci_set_master(pdev);
5587 pci_wake_from_d3(pdev, false);
5589 err = ixgbe_init_interrupt_scheme(adapter);
5591 e_dev_err("Cannot initialize interrupts for device\n");
5595 ixgbe_reset(adapter);
5597 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5599 if (netif_running(netdev)) {
5600 err = ixgbe_open(netdev);
5605 netif_device_attach(netdev);
5609 #endif /* CONFIG_PM */
5611 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5613 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5614 struct net_device *netdev = adapter->netdev;
5615 struct ixgbe_hw *hw = &adapter->hw;
5617 u32 wufc = adapter->wol;
5622 netif_device_detach(netdev);
5624 if (netif_running(netdev)) {
5625 ixgbe_down(adapter);
5626 ixgbe_free_irq(adapter);
5627 ixgbe_free_all_tx_resources(adapter);
5628 ixgbe_free_all_rx_resources(adapter);
5631 ixgbe_clear_interrupt_scheme(adapter);
5633 kfree(adapter->ixgbe_ieee_pfc);
5634 kfree(adapter->ixgbe_ieee_ets);
5638 retval = pci_save_state(pdev);
5644 ixgbe_set_rx_mode(netdev);
5646 /* turn on all-multi mode if wake on multicast is enabled */
5647 if (wufc & IXGBE_WUFC_MC) {
5648 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5649 fctrl |= IXGBE_FCTRL_MPE;
5650 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5653 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5654 ctrl |= IXGBE_CTRL_GIO_DIS;
5655 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5657 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5659 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5660 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5663 switch (hw->mac.type) {
5664 case ixgbe_mac_82598EB:
5665 pci_wake_from_d3(pdev, false);
5667 case ixgbe_mac_82599EB:
5668 case ixgbe_mac_X540:
5669 pci_wake_from_d3(pdev, !!wufc);
5675 *enable_wake = !!wufc;
5677 ixgbe_release_hw_control(adapter);
5679 pci_disable_device(pdev);
5685 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5690 retval = __ixgbe_shutdown(pdev, &wake);
5695 pci_prepare_to_sleep(pdev);
5697 pci_wake_from_d3(pdev, false);
5698 pci_set_power_state(pdev, PCI_D3hot);
5703 #endif /* CONFIG_PM */
5705 static void ixgbe_shutdown(struct pci_dev *pdev)
5709 __ixgbe_shutdown(pdev, &wake);
5711 if (system_state == SYSTEM_POWER_OFF) {
5712 pci_wake_from_d3(pdev, wake);
5713 pci_set_power_state(pdev, PCI_D3hot);
5718 * ixgbe_update_stats - Update the board statistics counters.
5719 * @adapter: board private structure
5721 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5723 struct net_device *netdev = adapter->netdev;
5724 struct ixgbe_hw *hw = &adapter->hw;
5725 struct ixgbe_hw_stats *hwstats = &adapter->stats;
5727 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5728 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
5729 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
5730 u64 bytes = 0, packets = 0;
5732 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5733 test_bit(__IXGBE_RESETTING, &adapter->state))
5736 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5739 for (i = 0; i < 16; i++)
5740 adapter->hw_rx_no_dma_resources +=
5741 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5742 for (i = 0; i < adapter->num_rx_queues; i++) {
5743 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
5744 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
5746 adapter->rsc_total_count = rsc_count;
5747 adapter->rsc_total_flush = rsc_flush;
5750 for (i = 0; i < adapter->num_rx_queues; i++) {
5751 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
5752 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
5753 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
5754 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
5755 bytes += rx_ring->stats.bytes;
5756 packets += rx_ring->stats.packets;
5758 adapter->non_eop_descs = non_eop_descs;
5759 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5760 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5761 netdev->stats.rx_bytes = bytes;
5762 netdev->stats.rx_packets = packets;
5766 /* gather some stats to the adapter struct that are per queue */
5767 for (i = 0; i < adapter->num_tx_queues; i++) {
5768 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5769 restart_queue += tx_ring->tx_stats.restart_queue;
5770 tx_busy += tx_ring->tx_stats.tx_busy;
5771 bytes += tx_ring->stats.bytes;
5772 packets += tx_ring->stats.packets;
5774 adapter->restart_queue = restart_queue;
5775 adapter->tx_busy = tx_busy;
5776 netdev->stats.tx_bytes = bytes;
5777 netdev->stats.tx_packets = packets;
5779 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5780 for (i = 0; i < 8; i++) {
5781 /* for packet buffers not used, the register should read 0 */
5782 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5784 hwstats->mpc[i] += mpc;
5785 total_mpc += hwstats->mpc[i];
5786 if (hw->mac.type == ixgbe_mac_82598EB)
5787 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5788 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5789 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5790 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5791 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5792 switch (hw->mac.type) {
5793 case ixgbe_mac_82598EB:
5794 hwstats->pxonrxc[i] +=
5795 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5797 case ixgbe_mac_82599EB:
5798 case ixgbe_mac_X540:
5799 hwstats->pxonrxc[i] +=
5800 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5805 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5806 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5808 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5809 /* work around hardware counting issue */
5810 hwstats->gprc -= missed_rx;
5812 ixgbe_update_xoff_received(adapter);
5814 /* 82598 hardware only has a 32 bit counter in the high register */
5815 switch (hw->mac.type) {
5816 case ixgbe_mac_82598EB:
5817 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5818 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5819 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5820 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5822 case ixgbe_mac_82599EB:
5823 case ixgbe_mac_X540:
5824 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5825 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
5826 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5827 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
5828 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5829 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5830 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5831 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5832 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5834 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5835 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5836 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5837 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5838 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5839 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5840 #endif /* IXGBE_FCOE */
5845 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5846 hwstats->bprc += bprc;
5847 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5848 if (hw->mac.type == ixgbe_mac_82598EB)
5849 hwstats->mprc -= bprc;
5850 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5851 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5852 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5853 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5854 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5855 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5856 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5857 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5858 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5859 hwstats->lxontxc += lxon;
5860 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5861 hwstats->lxofftxc += lxoff;
5862 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5863 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5864 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5866 * 82598 errata - tx of flow control packets is included in tx counters
5868 xon_off_tot = lxon + lxoff;
5869 hwstats->gptc -= xon_off_tot;
5870 hwstats->mptc -= xon_off_tot;
5871 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5872 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5873 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5874 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5875 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5876 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5877 hwstats->ptc64 -= xon_off_tot;
5878 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5879 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5880 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5881 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5882 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5883 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5885 /* Fill out the OS statistics structure */
5886 netdev->stats.multicast = hwstats->mprc;
5889 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
5890 netdev->stats.rx_dropped = 0;
5891 netdev->stats.rx_length_errors = hwstats->rlec;
5892 netdev->stats.rx_crc_errors = hwstats->crcerrs;
5893 netdev->stats.rx_missed_errors = total_mpc;
5897 * ixgbe_watchdog - Timer Call-back
5898 * @data: pointer to adapter cast into an unsigned long
5900 static void ixgbe_watchdog(unsigned long data)
5902 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5903 struct ixgbe_hw *hw = &adapter->hw;
5908 * Do the watchdog outside of interrupt context due to the lovely
5909 * delays that some of the newer hardware requires
5912 if (test_bit(__IXGBE_DOWN, &adapter->state))
5913 goto watchdog_short_circuit;
5915 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5917 * for legacy and MSI interrupts don't set any bits
5918 * that are enabled for EIAM, because this operation
5919 * would set *both* EIMS and EICS for any bit in EIAM
5921 IXGBE_WRITE_REG(hw, IXGBE_EICS,
5922 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5923 goto watchdog_reschedule;
5926 /* get one bit for every active tx/rx interrupt vector */
5927 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
5928 struct ixgbe_q_vector *qv = adapter->q_vector[i];
5929 if (qv->rxr_count || qv->txr_count)
5930 eics |= ((u64)1 << i);
5933 /* Cause software interrupt to ensure rx rings are cleaned */
5934 ixgbe_irq_rearm_queues(adapter, eics);
5936 watchdog_reschedule:
5937 /* Reset the timer */
5938 mod_timer(&adapter->watchdog_timer, round_jiffies(jiffies + 2 * HZ));
5940 watchdog_short_circuit:
5941 schedule_work(&adapter->watchdog_task);
5945 * ixgbe_multispeed_fiber_task - worker thread to configure multispeed fiber
5946 * @work: pointer to work_struct containing our data
5948 static void ixgbe_multispeed_fiber_task(struct work_struct *work)
5950 struct ixgbe_adapter *adapter = container_of(work,
5951 struct ixgbe_adapter,
5952 multispeed_fiber_task);
5953 struct ixgbe_hw *hw = &adapter->hw;
5957 adapter->flags |= IXGBE_FLAG_IN_SFP_LINK_TASK;
5958 autoneg = hw->phy.autoneg_advertised;
5959 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
5960 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
5961 hw->mac.autotry_restart = false;
5962 if (hw->mac.ops.setup_link)
5963 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
5964 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5965 adapter->flags &= ~IXGBE_FLAG_IN_SFP_LINK_TASK;
5969 * ixgbe_sfp_config_module_task - worker thread to configure a new SFP+ module
5970 * @work: pointer to work_struct containing our data
5972 static void ixgbe_sfp_config_module_task(struct work_struct *work)
5974 struct ixgbe_adapter *adapter = container_of(work,
5975 struct ixgbe_adapter,
5976 sfp_config_module_task);
5977 struct ixgbe_hw *hw = &adapter->hw;
5980 adapter->flags |= IXGBE_FLAG_IN_SFP_MOD_TASK;
5982 /* Time for electrical oscillations to settle down */
5984 err = hw->phy.ops.identify_sfp(hw);
5986 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
5987 e_dev_err("failed to initialize because an unsupported SFP+ "
5988 "module type was detected.\n");
5989 e_dev_err("Reload the driver after installing a supported "
5991 unregister_netdev(adapter->netdev);
5994 if (hw->mac.ops.setup_sfp)
5995 hw->mac.ops.setup_sfp(hw);
5997 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
5998 /* This will also work for DA Twinax connections */
5999 schedule_work(&adapter->multispeed_fiber_task);
6000 adapter->flags &= ~IXGBE_FLAG_IN_SFP_MOD_TASK;
6004 * ixgbe_fdir_reinit_task - worker thread to reinit FDIR filter table
6005 * @work: pointer to work_struct containing our data
6007 static void ixgbe_fdir_reinit_task(struct work_struct *work)
6009 struct ixgbe_adapter *adapter = container_of(work,
6010 struct ixgbe_adapter,
6012 struct ixgbe_hw *hw = &adapter->hw;
6015 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
6016 for (i = 0; i < adapter->num_tx_queues; i++)
6017 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
6018 &(adapter->tx_ring[i]->state));
6020 e_err(probe, "failed to finish FDIR re-initialization, "
6021 "ignored adding FDIR ATR filters\n");
6023 /* Done FDIR Re-initialization, enable transmits */
6024 netif_tx_start_all_queues(adapter->netdev);
6027 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
6031 /* Do not perform spoof check for 82598 */
6032 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
6035 ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
6038 * ssvpc register is cleared on read, if zero then no
6039 * spoofed packets in the last interval.
6044 e_warn(drv, "%d Spoofed packets detected\n", ssvpc);
6047 static DEFINE_MUTEX(ixgbe_watchdog_lock);
6050 * ixgbe_watchdog_task - worker thread to bring link up
6051 * @work: pointer to work_struct containing our data
6053 static void ixgbe_watchdog_task(struct work_struct *work)
6055 struct ixgbe_adapter *adapter = container_of(work,
6056 struct ixgbe_adapter,
6058 struct net_device *netdev = adapter->netdev;
6059 struct ixgbe_hw *hw = &adapter->hw;
6063 struct ixgbe_ring *tx_ring;
6064 int some_tx_pending = 0;
6066 mutex_lock(&ixgbe_watchdog_lock);
6068 link_up = adapter->link_up;
6069 link_speed = adapter->link_speed;
6071 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
6072 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
6075 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6076 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
6077 hw->mac.ops.fc_enable(hw, i);
6079 hw->mac.ops.fc_enable(hw, 0);
6082 hw->mac.ops.fc_enable(hw, 0);
6087 time_after(jiffies, (adapter->link_check_timeout +
6088 IXGBE_TRY_LINK_TIMEOUT))) {
6089 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
6090 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
6092 adapter->link_up = link_up;
6093 adapter->link_speed = link_speed;
6097 if (!netif_carrier_ok(netdev)) {
6098 bool flow_rx, flow_tx;
6100 switch (hw->mac.type) {
6101 case ixgbe_mac_82598EB: {
6102 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
6103 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
6104 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
6105 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
6108 case ixgbe_mac_82599EB:
6109 case ixgbe_mac_X540: {
6110 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
6111 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
6112 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
6113 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
6122 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
6123 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
6125 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
6127 (link_speed == IXGBE_LINK_SPEED_100_FULL ?
6130 ((flow_rx && flow_tx) ? "RX/TX" :
6132 (flow_tx ? "TX" : "None"))));
6134 netif_carrier_on(netdev);
6136 /* Force detection of hung controller */
6137 for (i = 0; i < adapter->num_tx_queues; i++) {
6138 tx_ring = adapter->tx_ring[i];
6139 set_check_for_tx_hang(tx_ring);
6143 adapter->link_up = false;
6144 adapter->link_speed = 0;
6145 if (netif_carrier_ok(netdev)) {
6146 e_info(drv, "NIC Link is Down\n");
6147 netif_carrier_off(netdev);
6151 if (!netif_carrier_ok(netdev)) {
6152 for (i = 0; i < adapter->num_tx_queues; i++) {
6153 tx_ring = adapter->tx_ring[i];
6154 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
6155 some_tx_pending = 1;
6160 if (some_tx_pending) {
6161 /* We've lost link, so the controller stops DMA,
6162 * but we've got queued Tx work that's never going
6163 * to get done, so reset controller to flush Tx.
6164 * (Do the reset outside of interrupt context).
6166 schedule_work(&adapter->reset_task);
6170 ixgbe_spoof_check(adapter);
6171 ixgbe_update_stats(adapter);
6172 mutex_unlock(&ixgbe_watchdog_lock);
6175 static int ixgbe_tso(struct ixgbe_adapter *adapter,
6176 struct ixgbe_ring *tx_ring, struct sk_buff *skb,
6177 u32 tx_flags, u8 *hdr_len, __be16 protocol)
6179 struct ixgbe_adv_tx_context_desc *context_desc;
6182 struct ixgbe_tx_buffer *tx_buffer_info;
6183 u32 vlan_macip_lens = 0, type_tucmd_mlhl;
6184 u32 mss_l4len_idx, l4len;
6186 if (skb_is_gso(skb)) {
6187 if (skb_header_cloned(skb)) {
6188 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
6192 l4len = tcp_hdrlen(skb);
6195 if (protocol == htons(ETH_P_IP)) {
6196 struct iphdr *iph = ip_hdr(skb);
6199 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
6203 } else if (skb_is_gso_v6(skb)) {
6204 ipv6_hdr(skb)->payload_len = 0;
6205 tcp_hdr(skb)->check =
6206 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
6207 &ipv6_hdr(skb)->daddr,
6211 i = tx_ring->next_to_use;
6213 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6214 context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
6216 /* VLAN MACLEN IPLEN */
6217 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
6219 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
6220 vlan_macip_lens |= ((skb_network_offset(skb)) <<
6221 IXGBE_ADVTXD_MACLEN_SHIFT);
6222 *hdr_len += skb_network_offset(skb);
6224 (skb_transport_header(skb) - skb_network_header(skb));
6226 (skb_transport_header(skb) - skb_network_header(skb));
6227 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
6228 context_desc->seqnum_seed = 0;
6230 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
6231 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
6232 IXGBE_ADVTXD_DTYP_CTXT);
6234 if (protocol == htons(ETH_P_IP))
6235 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
6236 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6237 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
6241 (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
6242 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
6243 /* use index 1 for TSO */
6244 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6245 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
6247 tx_buffer_info->time_stamp = jiffies;
6248 tx_buffer_info->next_to_watch = i;
6251 if (i == tx_ring->count)
6253 tx_ring->next_to_use = i;
6260 static u32 ixgbe_psum(struct ixgbe_adapter *adapter, struct sk_buff *skb,
6266 case cpu_to_be16(ETH_P_IP):
6267 rtn |= IXGBE_ADVTXD_TUCMD_IPV4;
6268 switch (ip_hdr(skb)->protocol) {
6270 rtn |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6273 rtn |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
6277 case cpu_to_be16(ETH_P_IPV6):
6278 /* XXX what about other V6 headers?? */
6279 switch (ipv6_hdr(skb)->nexthdr) {
6281 rtn |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6284 rtn |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
6289 if (unlikely(net_ratelimit()))
6290 e_warn(probe, "partial checksum but proto=%x!\n",
6298 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
6299 struct ixgbe_ring *tx_ring,
6300 struct sk_buff *skb, u32 tx_flags,
6303 struct ixgbe_adv_tx_context_desc *context_desc;
6305 struct ixgbe_tx_buffer *tx_buffer_info;
6306 u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
6308 if (skb->ip_summed == CHECKSUM_PARTIAL ||
6309 (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
6310 i = tx_ring->next_to_use;
6311 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6312 context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
6314 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
6316 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
6317 vlan_macip_lens |= (skb_network_offset(skb) <<
6318 IXGBE_ADVTXD_MACLEN_SHIFT);
6319 if (skb->ip_summed == CHECKSUM_PARTIAL)
6320 vlan_macip_lens |= (skb_transport_header(skb) -
6321 skb_network_header(skb));
6323 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
6324 context_desc->seqnum_seed = 0;
6326 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
6327 IXGBE_ADVTXD_DTYP_CTXT);
6329 if (skb->ip_summed == CHECKSUM_PARTIAL)
6330 type_tucmd_mlhl |= ixgbe_psum(adapter, skb, protocol);
6332 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
6333 /* use index zero for tx checksum offload */
6334 context_desc->mss_l4len_idx = 0;
6336 tx_buffer_info->time_stamp = jiffies;
6337 tx_buffer_info->next_to_watch = i;
6340 if (i == tx_ring->count)
6342 tx_ring->next_to_use = i;
6350 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
6351 struct ixgbe_ring *tx_ring,
6352 struct sk_buff *skb, u32 tx_flags,
6353 unsigned int first, const u8 hdr_len)
6355 struct device *dev = tx_ring->dev;
6356 struct ixgbe_tx_buffer *tx_buffer_info;
6358 unsigned int total = skb->len;
6359 unsigned int offset = 0, size, count = 0, i;
6360 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
6362 unsigned int bytecount = skb->len;
6365 i = tx_ring->next_to_use;
6367 if (tx_flags & IXGBE_TX_FLAGS_FCOE)
6368 /* excluding fcoe_crc_eof for FCoE */
6369 total -= sizeof(struct fcoe_crc_eof);
6371 len = min(skb_headlen(skb), total);
6373 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6374 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
6376 tx_buffer_info->length = size;
6377 tx_buffer_info->mapped_as_page = false;
6378 tx_buffer_info->dma = dma_map_single(dev,
6380 size, DMA_TO_DEVICE);
6381 if (dma_mapping_error(dev, tx_buffer_info->dma))
6383 tx_buffer_info->time_stamp = jiffies;
6384 tx_buffer_info->next_to_watch = i;
6393 if (i == tx_ring->count)
6398 for (f = 0; f < nr_frags; f++) {
6399 struct skb_frag_struct *frag;
6401 frag = &skb_shinfo(skb)->frags[f];
6402 len = min((unsigned int)frag->size, total);
6403 offset = frag->page_offset;
6407 if (i == tx_ring->count)
6410 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6411 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
6413 tx_buffer_info->length = size;
6414 tx_buffer_info->dma = dma_map_page(dev,
6418 tx_buffer_info->mapped_as_page = true;
6419 if (dma_mapping_error(dev, tx_buffer_info->dma))
6421 tx_buffer_info->time_stamp = jiffies;
6422 tx_buffer_info->next_to_watch = i;
6433 if (tx_flags & IXGBE_TX_FLAGS_TSO)
6434 gso_segs = skb_shinfo(skb)->gso_segs;
6436 /* adjust for FCoE Sequence Offload */
6437 else if (tx_flags & IXGBE_TX_FLAGS_FSO)
6438 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
6439 skb_shinfo(skb)->gso_size);
6440 #endif /* IXGBE_FCOE */
6441 bytecount += (gso_segs - 1) * hdr_len;
6443 /* multiply data chunks by size of headers */
6444 tx_ring->tx_buffer_info[i].bytecount = bytecount;
6445 tx_ring->tx_buffer_info[i].gso_segs = gso_segs;
6446 tx_ring->tx_buffer_info[i].skb = skb;
6447 tx_ring->tx_buffer_info[first].next_to_watch = i;
6452 e_dev_err("TX DMA map failed\n");
6454 /* clear timestamp and dma mappings for failed tx_buffer_info map */
6455 tx_buffer_info->dma = 0;
6456 tx_buffer_info->time_stamp = 0;
6457 tx_buffer_info->next_to_watch = 0;
6461 /* clear timestamp and dma mappings for remaining portion of packet */
6464 i += tx_ring->count;
6466 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6467 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
6473 static void ixgbe_tx_queue(struct ixgbe_ring *tx_ring,
6474 int tx_flags, int count, u32 paylen, u8 hdr_len)
6476 union ixgbe_adv_tx_desc *tx_desc = NULL;
6477 struct ixgbe_tx_buffer *tx_buffer_info;
6478 u32 olinfo_status = 0, cmd_type_len = 0;
6480 u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
6482 cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
6484 cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
6486 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
6487 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
6489 if (tx_flags & IXGBE_TX_FLAGS_TSO) {
6490 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
6492 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
6493 IXGBE_ADVTXD_POPTS_SHIFT;
6495 /* use index 1 context for tso */
6496 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6497 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
6498 olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
6499 IXGBE_ADVTXD_POPTS_SHIFT;
6501 } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
6502 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
6503 IXGBE_ADVTXD_POPTS_SHIFT;
6505 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6506 olinfo_status |= IXGBE_ADVTXD_CC;
6507 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6508 if (tx_flags & IXGBE_TX_FLAGS_FSO)
6509 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
6512 olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
6514 i = tx_ring->next_to_use;
6516 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6517 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
6518 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
6519 tx_desc->read.cmd_type_len =
6520 cpu_to_le32(cmd_type_len | tx_buffer_info->length);
6521 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
6523 if (i == tx_ring->count)
6527 tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
6530 * Force memory writes to complete before letting h/w
6531 * know there are new descriptors to fetch. (Only
6532 * applicable for weak-ordered memory model archs,
6537 tx_ring->next_to_use = i;
6538 writel(i, tx_ring->tail);
6541 static void ixgbe_atr(struct ixgbe_ring *ring, struct sk_buff *skb,
6542 u32 tx_flags, __be16 protocol)
6544 struct ixgbe_q_vector *q_vector = ring->q_vector;
6545 union ixgbe_atr_hash_dword input = { .dword = 0 };
6546 union ixgbe_atr_hash_dword common = { .dword = 0 };
6548 unsigned char *network;
6550 struct ipv6hdr *ipv6;
6555 /* if ring doesn't have a interrupt vector, cannot perform ATR */
6559 /* do nothing if sampling is disabled */
6560 if (!ring->atr_sample_rate)
6565 /* snag network header to get L4 type and address */
6566 hdr.network = skb_network_header(skb);
6568 /* Currently only IPv4/IPv6 with TCP is supported */
6569 if ((protocol != __constant_htons(ETH_P_IPV6) ||
6570 hdr.ipv6->nexthdr != IPPROTO_TCP) &&
6571 (protocol != __constant_htons(ETH_P_IP) ||
6572 hdr.ipv4->protocol != IPPROTO_TCP))
6577 /* skip this packet since the socket is closing */
6581 /* sample on all syn packets or once every atr sample count */
6582 if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
6585 /* reset sample count */
6586 ring->atr_count = 0;
6588 vlan_id = htons(tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
6591 * src and dst are inverted, think how the receiver sees them
6593 * The input is broken into two sections, a non-compressed section
6594 * containing vm_pool, vlan_id, and flow_type. The rest of the data
6595 * is XORed together and stored in the compressed dword.
6597 input.formatted.vlan_id = vlan_id;
6600 * since src port and flex bytes occupy the same word XOR them together
6601 * and write the value to source port portion of compressed dword
6604 common.port.src ^= th->dest ^ __constant_htons(ETH_P_8021Q);
6606 common.port.src ^= th->dest ^ protocol;
6607 common.port.dst ^= th->source;
6609 if (protocol == __constant_htons(ETH_P_IP)) {
6610 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
6611 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
6613 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
6614 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
6615 hdr.ipv6->saddr.s6_addr32[1] ^
6616 hdr.ipv6->saddr.s6_addr32[2] ^
6617 hdr.ipv6->saddr.s6_addr32[3] ^
6618 hdr.ipv6->daddr.s6_addr32[0] ^
6619 hdr.ipv6->daddr.s6_addr32[1] ^
6620 hdr.ipv6->daddr.s6_addr32[2] ^
6621 hdr.ipv6->daddr.s6_addr32[3];
6624 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
6625 ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
6626 input, common, ring->queue_index);
6629 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, int size)
6631 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
6632 /* Herbert's original patch had:
6633 * smp_mb__after_netif_stop_queue();
6634 * but since that doesn't exist yet, just open code it. */
6637 /* We need to check again in a case another CPU has just
6638 * made room available. */
6639 if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
6642 /* A reprieve! - use start_queue because it doesn't call schedule */
6643 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
6644 ++tx_ring->tx_stats.restart_queue;
6648 static int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, int size)
6650 if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
6652 return __ixgbe_maybe_stop_tx(tx_ring, size);
6655 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6657 struct ixgbe_adapter *adapter = netdev_priv(dev);
6658 int txq = smp_processor_id();
6662 protocol = vlan_get_protocol(skb);
6664 if ((protocol == htons(ETH_P_FCOE)) ||
6665 (protocol == htons(ETH_P_FIP))) {
6666 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
6667 txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
6668 txq += adapter->ring_feature[RING_F_FCOE].mask;
6670 #ifdef CONFIG_IXGBE_DCB
6671 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6672 txq = adapter->fcoe.up;
6679 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6680 while (unlikely(txq >= dev->real_num_tx_queues))
6681 txq -= dev->real_num_tx_queues;
6685 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6686 if (skb->priority == TC_PRIO_CONTROL)
6687 txq = adapter->ring_feature[RING_F_DCB].indices-1;
6689 txq = (skb->vlan_tci & IXGBE_TX_FLAGS_VLAN_PRIO_MASK)
6694 return skb_tx_hash(dev, skb);
6697 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
6698 struct ixgbe_adapter *adapter,
6699 struct ixgbe_ring *tx_ring)
6702 unsigned int tx_flags = 0;
6709 protocol = vlan_get_protocol(skb);
6711 if (vlan_tx_tag_present(skb)) {
6712 tx_flags |= vlan_tx_tag_get(skb);
6713 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6714 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6715 tx_flags |= ((skb->queue_mapping & 0x7) << 13);
6717 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6718 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6719 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED &&
6720 skb->priority != TC_PRIO_CONTROL) {
6721 tx_flags |= ((skb->queue_mapping & 0x7) << 13);
6722 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6723 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6727 /* for FCoE with DCB, we force the priority to what
6728 * was specified by the switch */
6729 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED &&
6730 (protocol == htons(ETH_P_FCOE) ||
6731 protocol == htons(ETH_P_FIP))) {
6732 #ifdef CONFIG_IXGBE_DCB
6733 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6734 tx_flags &= ~(IXGBE_TX_FLAGS_VLAN_PRIO_MASK
6735 << IXGBE_TX_FLAGS_VLAN_SHIFT);
6736 tx_flags |= ((adapter->fcoe.up << 13)
6737 << IXGBE_TX_FLAGS_VLAN_SHIFT);
6740 /* flag for FCoE offloads */
6741 if (protocol == htons(ETH_P_FCOE))
6742 tx_flags |= IXGBE_TX_FLAGS_FCOE;
6746 /* four things can cause us to need a context descriptor */
6747 if (skb_is_gso(skb) ||
6748 (skb->ip_summed == CHECKSUM_PARTIAL) ||
6749 (tx_flags & IXGBE_TX_FLAGS_VLAN) ||
6750 (tx_flags & IXGBE_TX_FLAGS_FCOE))
6753 count += TXD_USE_COUNT(skb_headlen(skb));
6754 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6755 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6757 if (ixgbe_maybe_stop_tx(tx_ring, count)) {
6758 tx_ring->tx_stats.tx_busy++;
6759 return NETDEV_TX_BUSY;
6762 first = tx_ring->next_to_use;
6763 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6765 /* setup tx offload for FCoE */
6766 tso = ixgbe_fso(adapter, tx_ring, skb, tx_flags, &hdr_len);
6768 dev_kfree_skb_any(skb);
6769 return NETDEV_TX_OK;
6772 tx_flags |= IXGBE_TX_FLAGS_FSO;
6773 #endif /* IXGBE_FCOE */
6775 if (protocol == htons(ETH_P_IP))
6776 tx_flags |= IXGBE_TX_FLAGS_IPV4;
6777 tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len,
6780 dev_kfree_skb_any(skb);
6781 return NETDEV_TX_OK;
6785 tx_flags |= IXGBE_TX_FLAGS_TSO;
6786 else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags,
6788 (skb->ip_summed == CHECKSUM_PARTIAL))
6789 tx_flags |= IXGBE_TX_FLAGS_CSUM;
6792 count = ixgbe_tx_map(adapter, tx_ring, skb, tx_flags, first, hdr_len);
6794 /* add the ATR filter if ATR is on */
6795 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
6796 ixgbe_atr(tx_ring, skb, tx_flags, protocol);
6797 ixgbe_tx_queue(tx_ring, tx_flags, count, skb->len, hdr_len);
6798 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
6801 dev_kfree_skb_any(skb);
6802 tx_ring->tx_buffer_info[first].time_stamp = 0;
6803 tx_ring->next_to_use = first;
6806 return NETDEV_TX_OK;
6809 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
6811 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6812 struct ixgbe_ring *tx_ring;
6814 tx_ring = adapter->tx_ring[skb->queue_mapping];
6815 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
6819 * ixgbe_set_mac - Change the Ethernet Address of the NIC
6820 * @netdev: network interface device structure
6821 * @p: pointer to an address structure
6823 * Returns 0 on success, negative on failure
6825 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6827 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6828 struct ixgbe_hw *hw = &adapter->hw;
6829 struct sockaddr *addr = p;
6831 if (!is_valid_ether_addr(addr->sa_data))
6832 return -EADDRNOTAVAIL;
6834 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6835 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6837 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
6844 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6846 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6847 struct ixgbe_hw *hw = &adapter->hw;
6851 if (prtad != hw->phy.mdio.prtad)
6853 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6859 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6860 u16 addr, u16 value)
6862 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6863 struct ixgbe_hw *hw = &adapter->hw;
6865 if (prtad != hw->phy.mdio.prtad)
6867 return hw->phy.ops.write_reg(hw, addr, devad, value);
6870 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6872 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6874 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6878 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6880 * @netdev: network interface device structure
6882 * Returns non-zero on failure
6884 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6887 struct ixgbe_adapter *adapter = netdev_priv(dev);
6888 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6890 if (is_valid_ether_addr(mac->san_addr)) {
6892 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6899 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6901 * @netdev: network interface device structure
6903 * Returns non-zero on failure
6905 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6908 struct ixgbe_adapter *adapter = netdev_priv(dev);
6909 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6911 if (is_valid_ether_addr(mac->san_addr)) {
6913 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6919 #ifdef CONFIG_NET_POLL_CONTROLLER
6921 * Polling 'interrupt' - used by things like netconsole to send skbs
6922 * without having to re-enable interrupts. It's not called while
6923 * the interrupt routine is executing.
6925 static void ixgbe_netpoll(struct net_device *netdev)
6927 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6930 /* if interface is down do nothing */
6931 if (test_bit(__IXGBE_DOWN, &adapter->state))
6934 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
6935 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
6936 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
6937 for (i = 0; i < num_q_vectors; i++) {
6938 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
6939 ixgbe_msix_clean_many(0, q_vector);
6942 ixgbe_intr(adapter->pdev->irq, netdev);
6944 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
6948 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
6949 struct rtnl_link_stats64 *stats)
6951 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6955 for (i = 0; i < adapter->num_rx_queues; i++) {
6956 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
6962 start = u64_stats_fetch_begin_bh(&ring->syncp);
6963 packets = ring->stats.packets;
6964 bytes = ring->stats.bytes;
6965 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6966 stats->rx_packets += packets;
6967 stats->rx_bytes += bytes;
6971 for (i = 0; i < adapter->num_tx_queues; i++) {
6972 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->tx_ring[i]);
6978 start = u64_stats_fetch_begin_bh(&ring->syncp);
6979 packets = ring->stats.packets;
6980 bytes = ring->stats.bytes;
6981 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6982 stats->tx_packets += packets;
6983 stats->tx_bytes += bytes;
6987 /* following stats updated by ixgbe_watchdog_task() */
6988 stats->multicast = netdev->stats.multicast;
6989 stats->rx_errors = netdev->stats.rx_errors;
6990 stats->rx_length_errors = netdev->stats.rx_length_errors;
6991 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
6992 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
6997 static const struct net_device_ops ixgbe_netdev_ops = {
6998 .ndo_open = ixgbe_open,
6999 .ndo_stop = ixgbe_close,
7000 .ndo_start_xmit = ixgbe_xmit_frame,
7001 .ndo_select_queue = ixgbe_select_queue,
7002 .ndo_set_rx_mode = ixgbe_set_rx_mode,
7003 .ndo_set_multicast_list = ixgbe_set_rx_mode,
7004 .ndo_validate_addr = eth_validate_addr,
7005 .ndo_set_mac_address = ixgbe_set_mac,
7006 .ndo_change_mtu = ixgbe_change_mtu,
7007 .ndo_tx_timeout = ixgbe_tx_timeout,
7008 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
7009 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
7010 .ndo_do_ioctl = ixgbe_ioctl,
7011 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
7012 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
7013 .ndo_set_vf_tx_rate = ixgbe_ndo_set_vf_bw,
7014 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
7015 .ndo_get_stats64 = ixgbe_get_stats64,
7016 #ifdef CONFIG_NET_POLL_CONTROLLER
7017 .ndo_poll_controller = ixgbe_netpoll,
7020 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
7021 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
7022 .ndo_fcoe_enable = ixgbe_fcoe_enable,
7023 .ndo_fcoe_disable = ixgbe_fcoe_disable,
7024 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
7025 #endif /* IXGBE_FCOE */
7028 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
7029 const struct ixgbe_info *ii)
7031 #ifdef CONFIG_PCI_IOV
7032 struct ixgbe_hw *hw = &adapter->hw;
7035 if (hw->mac.type == ixgbe_mac_82598EB || !max_vfs)
7038 /* The 82599 supports up to 64 VFs per physical function
7039 * but this implementation limits allocation to 63 so that
7040 * basic networking resources are still available to the
7043 adapter->num_vfs = (max_vfs > 63) ? 63 : max_vfs;
7044 adapter->flags |= IXGBE_FLAG_SRIOV_ENABLED;
7045 err = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
7047 e_err(probe, "Failed to enable PCI sriov: %d\n", err);
7050 /* If call to enable VFs succeeded then allocate memory
7051 * for per VF control structures.
7054 kcalloc(adapter->num_vfs,
7055 sizeof(struct vf_data_storage), GFP_KERNEL);
7056 if (adapter->vfinfo) {
7057 /* Now that we're sure SR-IOV is enabled
7058 * and memory allocated set up the mailbox parameters
7060 ixgbe_init_mbx_params_pf(hw);
7061 memcpy(&hw->mbx.ops, ii->mbx_ops,
7062 sizeof(hw->mbx.ops));
7064 /* Disable RSC when in SR-IOV mode */
7065 adapter->flags2 &= ~(IXGBE_FLAG2_RSC_CAPABLE |
7066 IXGBE_FLAG2_RSC_ENABLED);
7071 e_err(probe, "Unable to allocate memory for VF Data Storage - "
7072 "SRIOV disabled\n");
7073 pci_disable_sriov(adapter->pdev);
7076 adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
7077 adapter->num_vfs = 0;
7078 #endif /* CONFIG_PCI_IOV */
7082 * ixgbe_probe - Device Initialization Routine
7083 * @pdev: PCI device information struct
7084 * @ent: entry in ixgbe_pci_tbl
7086 * Returns 0 on success, negative on failure
7088 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
7089 * The OS initialization, configuring of the adapter private structure,
7090 * and a hardware reset occur.
7092 static int __devinit ixgbe_probe(struct pci_dev *pdev,
7093 const struct pci_device_id *ent)
7095 struct net_device *netdev;
7096 struct ixgbe_adapter *adapter = NULL;
7097 struct ixgbe_hw *hw;
7098 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
7099 static int cards_found;
7100 int i, err, pci_using_dac;
7101 u8 part_str[IXGBE_PBANUM_LENGTH];
7102 unsigned int indices = num_possible_cpus();
7108 /* Catch broken hardware that put the wrong VF device ID in
7109 * the PCIe SR-IOV capability.
7111 if (pdev->is_virtfn) {
7112 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
7113 pci_name(pdev), pdev->vendor, pdev->device);
7117 err = pci_enable_device_mem(pdev);
7121 if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
7122 !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
7125 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
7127 err = dma_set_coherent_mask(&pdev->dev,
7131 "No usable DMA configuration, aborting\n");
7138 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
7139 IORESOURCE_MEM), ixgbe_driver_name);
7142 "pci_request_selected_regions failed 0x%x\n", err);
7146 pci_enable_pcie_error_reporting(pdev);
7148 pci_set_master(pdev);
7149 pci_save_state(pdev);
7151 if (ii->mac == ixgbe_mac_82598EB)
7152 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
7154 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
7156 indices = max_t(unsigned int, indices, IXGBE_MAX_DCB_INDICES);
7158 indices += min_t(unsigned int, num_possible_cpus(),
7159 IXGBE_MAX_FCOE_INDICES);
7161 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
7164 goto err_alloc_etherdev;
7167 SET_NETDEV_DEV(netdev, &pdev->dev);
7169 adapter = netdev_priv(netdev);
7170 pci_set_drvdata(pdev, adapter);
7172 adapter->netdev = netdev;
7173 adapter->pdev = pdev;
7176 adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
7178 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
7179 pci_resource_len(pdev, 0));
7185 for (i = 1; i <= 5; i++) {
7186 if (pci_resource_len(pdev, i) == 0)
7190 netdev->netdev_ops = &ixgbe_netdev_ops;
7191 ixgbe_set_ethtool_ops(netdev);
7192 netdev->watchdog_timeo = 5 * HZ;
7193 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
7195 adapter->bd_number = cards_found;
7198 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
7199 hw->mac.type = ii->mac;
7202 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
7203 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
7204 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
7205 if (!(eec & (1 << 8)))
7206 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
7209 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
7210 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
7211 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
7212 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
7213 hw->phy.mdio.mmds = 0;
7214 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
7215 hw->phy.mdio.dev = netdev;
7216 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
7217 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
7219 /* set up this timer and work struct before calling get_invariants
7220 * which might start the timer
7222 init_timer(&adapter->sfp_timer);
7223 adapter->sfp_timer.function = ixgbe_sfp_timer;
7224 adapter->sfp_timer.data = (unsigned long) adapter;
7226 INIT_WORK(&adapter->sfp_task, ixgbe_sfp_task);
7228 /* multispeed fiber has its own tasklet, called from GPI SDP1 context */
7229 INIT_WORK(&adapter->multispeed_fiber_task, ixgbe_multispeed_fiber_task);
7231 /* a new SFP+ module arrival, called from GPI SDP2 context */
7232 INIT_WORK(&adapter->sfp_config_module_task,
7233 ixgbe_sfp_config_module_task);
7235 ii->get_invariants(hw);
7237 /* setup the private structure */
7238 err = ixgbe_sw_init(adapter);
7242 /* Make it possible the adapter to be woken up via WOL */
7243 switch (adapter->hw.mac.type) {
7244 case ixgbe_mac_82599EB:
7245 case ixgbe_mac_X540:
7246 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7253 * If there is a fan on this device and it has failed log the
7256 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
7257 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
7258 if (esdp & IXGBE_ESDP_SDP1)
7259 e_crit(probe, "Fan has stopped, replace the adapter\n");
7262 /* reset_hw fills in the perm_addr as well */
7263 hw->phy.reset_if_overtemp = true;
7264 err = hw->mac.ops.reset_hw(hw);
7265 hw->phy.reset_if_overtemp = false;
7266 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
7267 hw->mac.type == ixgbe_mac_82598EB) {
7269 * Start a kernel thread to watch for a module to arrive.
7270 * Only do this for 82598, since 82599 will generate
7271 * interrupts on module arrival.
7273 set_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
7274 mod_timer(&adapter->sfp_timer,
7275 round_jiffies(jiffies + (2 * HZ)));
7277 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
7278 e_dev_err("failed to initialize because an unsupported SFP+ "
7279 "module type was detected.\n");
7280 e_dev_err("Reload the driver after installing a supported "
7284 e_dev_err("HW Init failed: %d\n", err);
7288 ixgbe_probe_vf(adapter, ii);
7290 netdev->features = NETIF_F_SG |
7292 NETIF_F_HW_VLAN_TX |
7293 NETIF_F_HW_VLAN_RX |
7294 NETIF_F_HW_VLAN_FILTER;
7296 netdev->features |= NETIF_F_IPV6_CSUM;
7297 netdev->features |= NETIF_F_TSO;
7298 netdev->features |= NETIF_F_TSO6;
7299 netdev->features |= NETIF_F_GRO;
7301 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
7302 netdev->features |= NETIF_F_SCTP_CSUM;
7304 netdev->vlan_features |= NETIF_F_TSO;
7305 netdev->vlan_features |= NETIF_F_TSO6;
7306 netdev->vlan_features |= NETIF_F_IP_CSUM;
7307 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
7308 netdev->vlan_features |= NETIF_F_SG;
7310 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7311 adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
7312 IXGBE_FLAG_DCB_ENABLED);
7313 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
7314 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
7316 #ifdef CONFIG_IXGBE_DCB
7317 netdev->dcbnl_ops = &dcbnl_ops;
7321 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7322 if (hw->mac.ops.get_device_caps) {
7323 hw->mac.ops.get_device_caps(hw, &device_caps);
7324 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
7325 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
7328 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7329 netdev->vlan_features |= NETIF_F_FCOE_CRC;
7330 netdev->vlan_features |= NETIF_F_FSO;
7331 netdev->vlan_features |= NETIF_F_FCOE_MTU;
7333 #endif /* IXGBE_FCOE */
7334 if (pci_using_dac) {
7335 netdev->features |= NETIF_F_HIGHDMA;
7336 netdev->vlan_features |= NETIF_F_HIGHDMA;
7339 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
7340 netdev->features |= NETIF_F_LRO;
7342 /* make sure the EEPROM is good */
7343 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
7344 e_dev_err("The EEPROM Checksum Is Not Valid\n");
7349 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
7350 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
7352 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
7353 e_dev_err("invalid MAC address\n");
7358 /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
7359 if (hw->mac.ops.disable_tx_laser &&
7360 ((hw->phy.multispeed_fiber) ||
7361 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
7362 (hw->mac.type == ixgbe_mac_82599EB))))
7363 hw->mac.ops.disable_tx_laser(hw);
7365 init_timer(&adapter->watchdog_timer);
7366 adapter->watchdog_timer.function = ixgbe_watchdog;
7367 adapter->watchdog_timer.data = (unsigned long)adapter;
7369 INIT_WORK(&adapter->reset_task, ixgbe_reset_task);
7370 INIT_WORK(&adapter->watchdog_task, ixgbe_watchdog_task);
7372 err = ixgbe_init_interrupt_scheme(adapter);
7376 switch (pdev->device) {
7377 case IXGBE_DEV_ID_82599_SFP:
7378 /* Only this subdevice supports WOL */
7379 if (pdev->subsystem_device == IXGBE_SUBDEV_ID_82599_SFP)
7380 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
7381 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
7383 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
7384 /* All except this subdevice support WOL */
7385 if (pdev->subsystem_device != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
7386 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
7387 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
7389 case IXGBE_DEV_ID_82599_KX4:
7390 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
7391 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
7397 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
7399 /* pick up the PCI bus settings for reporting later */
7400 hw->mac.ops.get_bus_info(hw);
7402 /* print bus type/speed/width info */
7403 e_dev_info("(PCI Express:%s:%s) %pM\n",
7404 (hw->bus.speed == ixgbe_bus_speed_5000 ? "5.0Gb/s" :
7405 hw->bus.speed == ixgbe_bus_speed_2500 ? "2.5Gb/s" :
7407 (hw->bus.width == ixgbe_bus_width_pcie_x8 ? "Width x8" :
7408 hw->bus.width == ixgbe_bus_width_pcie_x4 ? "Width x4" :
7409 hw->bus.width == ixgbe_bus_width_pcie_x1 ? "Width x1" :
7413 err = ixgbe_read_pba_string_generic(hw, part_str, IXGBE_PBANUM_LENGTH);
7415 strncpy(part_str, "Unknown", IXGBE_PBANUM_LENGTH);
7416 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
7417 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
7418 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
7421 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
7422 hw->mac.type, hw->phy.type, part_str);
7424 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
7425 e_dev_warn("PCI-Express bandwidth available for this card is "
7426 "not sufficient for optimal performance.\n");
7427 e_dev_warn("For optimal performance a x8 PCI-Express slot "
7431 /* save off EEPROM version number */
7432 hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
7434 /* reset the hardware with the new settings */
7435 err = hw->mac.ops.start_hw(hw);
7437 if (err == IXGBE_ERR_EEPROM_VERSION) {
7438 /* We are running on a pre-production device, log a warning */
7439 e_dev_warn("This device is a pre-production adapter/LOM. "
7440 "Please be aware there may be issues associated "
7441 "with your hardware. If you are experiencing "
7442 "problems please contact your Intel or hardware "
7443 "representative who provided you with this "
7446 strcpy(netdev->name, "eth%d");
7447 err = register_netdev(netdev);
7451 /* carrier off reporting is important to ethtool even BEFORE open */
7452 netif_carrier_off(netdev);
7454 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
7455 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
7456 INIT_WORK(&adapter->fdir_reinit_task, ixgbe_fdir_reinit_task);
7458 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
7459 INIT_WORK(&adapter->check_overtemp_task,
7460 ixgbe_check_overtemp_task);
7461 #ifdef CONFIG_IXGBE_DCA
7462 if (dca_add_requester(&pdev->dev) == 0) {
7463 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
7464 ixgbe_setup_dca(adapter);
7467 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7468 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
7469 for (i = 0; i < adapter->num_vfs; i++)
7470 ixgbe_vf_configuration(pdev, (i | 0x10000000));
7473 /* add san mac addr to netdev */
7474 ixgbe_add_sanmac_netdev(netdev);
7476 e_dev_info("Intel(R) 10 Gigabit Network Connection\n");
7481 ixgbe_release_hw_control(adapter);
7482 ixgbe_clear_interrupt_scheme(adapter);
7485 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7486 ixgbe_disable_sriov(adapter);
7487 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
7488 del_timer_sync(&adapter->sfp_timer);
7489 cancel_work_sync(&adapter->sfp_task);
7490 cancel_work_sync(&adapter->multispeed_fiber_task);
7491 cancel_work_sync(&adapter->sfp_config_module_task);
7492 iounmap(hw->hw_addr);
7494 free_netdev(netdev);
7496 pci_release_selected_regions(pdev,
7497 pci_select_bars(pdev, IORESOURCE_MEM));
7500 pci_disable_device(pdev);
7505 * ixgbe_remove - Device Removal Routine
7506 * @pdev: PCI device information struct
7508 * ixgbe_remove is called by the PCI subsystem to alert the driver
7509 * that it should release a PCI device. The could be caused by a
7510 * Hot-Plug event, or because the driver is going to be removed from
7513 static void __devexit ixgbe_remove(struct pci_dev *pdev)
7515 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7516 struct net_device *netdev = adapter->netdev;
7518 set_bit(__IXGBE_DOWN, &adapter->state);
7521 * The timers may be rescheduled, so explicitly disable them
7522 * from being rescheduled.
7524 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
7525 del_timer_sync(&adapter->watchdog_timer);
7526 del_timer_sync(&adapter->sfp_timer);
7528 cancel_work_sync(&adapter->watchdog_task);
7529 cancel_work_sync(&adapter->sfp_task);
7530 cancel_work_sync(&adapter->multispeed_fiber_task);
7531 cancel_work_sync(&adapter->sfp_config_module_task);
7532 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
7533 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
7534 cancel_work_sync(&adapter->fdir_reinit_task);
7535 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
7536 cancel_work_sync(&adapter->check_overtemp_task);
7538 #ifdef CONFIG_IXGBE_DCA
7539 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
7540 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
7541 dca_remove_requester(&pdev->dev);
7542 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
7547 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
7548 ixgbe_cleanup_fcoe(adapter);
7550 #endif /* IXGBE_FCOE */
7552 /* remove the added san mac */
7553 ixgbe_del_sanmac_netdev(netdev);
7555 if (netdev->reg_state == NETREG_REGISTERED)
7556 unregister_netdev(netdev);
7558 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7559 ixgbe_disable_sriov(adapter);
7561 ixgbe_clear_interrupt_scheme(adapter);
7563 ixgbe_release_hw_control(adapter);
7565 iounmap(adapter->hw.hw_addr);
7566 pci_release_selected_regions(pdev, pci_select_bars(pdev,
7569 e_dev_info("complete\n");
7571 free_netdev(netdev);
7573 pci_disable_pcie_error_reporting(pdev);
7575 pci_disable_device(pdev);
7579 * ixgbe_io_error_detected - called when PCI error is detected
7580 * @pdev: Pointer to PCI device
7581 * @state: The current pci connection state
7583 * This function is called after a PCI bus error affecting
7584 * this device has been detected.
7586 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
7587 pci_channel_state_t state)
7589 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7590 struct net_device *netdev = adapter->netdev;
7592 netif_device_detach(netdev);
7594 if (state == pci_channel_io_perm_failure)
7595 return PCI_ERS_RESULT_DISCONNECT;
7597 if (netif_running(netdev))
7598 ixgbe_down(adapter);
7599 pci_disable_device(pdev);
7601 /* Request a slot reset. */
7602 return PCI_ERS_RESULT_NEED_RESET;
7606 * ixgbe_io_slot_reset - called after the pci bus has been reset.
7607 * @pdev: Pointer to PCI device
7609 * Restart the card from scratch, as if from a cold-boot.
7611 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7613 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7614 pci_ers_result_t result;
7617 if (pci_enable_device_mem(pdev)) {
7618 e_err(probe, "Cannot re-enable PCI device after reset.\n");
7619 result = PCI_ERS_RESULT_DISCONNECT;
7621 pci_set_master(pdev);
7622 pci_restore_state(pdev);
7623 pci_save_state(pdev);
7625 pci_wake_from_d3(pdev, false);
7627 ixgbe_reset(adapter);
7628 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7629 result = PCI_ERS_RESULT_RECOVERED;
7632 err = pci_cleanup_aer_uncorrect_error_status(pdev);
7634 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7635 "failed 0x%0x\n", err);
7636 /* non-fatal, continue */
7643 * ixgbe_io_resume - called when traffic can start flowing again.
7644 * @pdev: Pointer to PCI device
7646 * This callback is called when the error recovery driver tells us that
7647 * its OK to resume normal operation.
7649 static void ixgbe_io_resume(struct pci_dev *pdev)
7651 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7652 struct net_device *netdev = adapter->netdev;
7654 if (netif_running(netdev)) {
7655 if (ixgbe_up(adapter)) {
7656 e_info(probe, "ixgbe_up failed after reset\n");
7661 netif_device_attach(netdev);
7664 static struct pci_error_handlers ixgbe_err_handler = {
7665 .error_detected = ixgbe_io_error_detected,
7666 .slot_reset = ixgbe_io_slot_reset,
7667 .resume = ixgbe_io_resume,
7670 static struct pci_driver ixgbe_driver = {
7671 .name = ixgbe_driver_name,
7672 .id_table = ixgbe_pci_tbl,
7673 .probe = ixgbe_probe,
7674 .remove = __devexit_p(ixgbe_remove),
7676 .suspend = ixgbe_suspend,
7677 .resume = ixgbe_resume,
7679 .shutdown = ixgbe_shutdown,
7680 .err_handler = &ixgbe_err_handler
7684 * ixgbe_init_module - Driver Registration Routine
7686 * ixgbe_init_module is the first routine called when the driver is
7687 * loaded. All it does is register with the PCI subsystem.
7689 static int __init ixgbe_init_module(void)
7692 pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
7693 pr_info("%s\n", ixgbe_copyright);
7695 #ifdef CONFIG_IXGBE_DCA
7696 dca_register_notify(&dca_notifier);
7699 ret = pci_register_driver(&ixgbe_driver);
7703 module_init(ixgbe_init_module);
7706 * ixgbe_exit_module - Driver Exit Cleanup Routine
7708 * ixgbe_exit_module is called just before the driver is removed
7711 static void __exit ixgbe_exit_module(void)
7713 #ifdef CONFIG_IXGBE_DCA
7714 dca_unregister_notify(&dca_notifier);
7716 pci_unregister_driver(&ixgbe_driver);
7717 rcu_barrier(); /* Wait for completion of call_rcu()'s */
7720 #ifdef CONFIG_IXGBE_DCA
7721 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
7726 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
7727 __ixgbe_notify_dca);
7729 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7732 #endif /* CONFIG_IXGBE_DCA */
7734 module_exit(ixgbe_exit_module);