1 /*******************************************************************************
3 Intel 10 Gigabit PCI Express Linux driver
4 Copyright(c) 1999 - 2011 Intel Corporation.
6 This program is free software; you can redistribute it and/or modify it
7 under the terms and conditions of the GNU General Public License,
8 version 2, as published by the Free Software Foundation.
10 This program is distributed in the hope it will be useful, but WITHOUT
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
23 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26 *******************************************************************************/
28 #include <linux/types.h>
29 #include <linux/module.h>
30 #include <linux/pci.h>
31 #include <linux/netdevice.h>
32 #include <linux/vmalloc.h>
33 #include <linux/string.h>
35 #include <linux/interrupt.h>
37 #include <linux/tcp.h>
38 #include <linux/sctp.h>
39 #include <linux/pkt_sched.h>
40 #include <linux/ipv6.h>
41 #include <linux/slab.h>
42 #include <net/checksum.h>
43 #include <net/ip6_checksum.h>
44 #include <linux/ethtool.h>
46 #include <linux/if_vlan.h>
47 #include <linux/prefetch.h>
48 #include <scsi/fc/fc_fcoe.h>
51 #include "ixgbe_common.h"
52 #include "ixgbe_dcb_82599.h"
53 #include "ixgbe_sriov.h"
55 char ixgbe_driver_name[] = "ixgbe";
56 static const char ixgbe_driver_string[] =
57 "Intel(R) 10 Gigabit PCI Express Network Driver";
61 #define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) "." \
62 __stringify(BUILD) "-k"
63 const char ixgbe_driver_version[] = DRV_VERSION;
64 static const char ixgbe_copyright[] =
65 "Copyright (c) 1999-2011 Intel Corporation.";
67 static const struct ixgbe_info *ixgbe_info_tbl[] = {
68 [board_82598] = &ixgbe_82598_info,
69 [board_82599] = &ixgbe_82599_info,
70 [board_X540] = &ixgbe_X540_info,
73 /* ixgbe_pci_tbl - PCI Device ID Table
75 * Wildcard entries (PCI_ANY_ID) should come last
76 * Last entry must be all 0s
78 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
79 * Class, Class Mask, private data (not used) }
81 static DEFINE_PCI_DEVICE_TABLE(ixgbe_pci_tbl) = {
82 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598),
84 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT),
86 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT),
88 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT),
90 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2),
92 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4),
94 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT),
96 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT),
98 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM),
100 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR),
102 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM),
104 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX),
106 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4),
108 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM),
110 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR),
112 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP),
114 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM),
116 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ),
118 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4),
120 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE),
122 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE),
124 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM),
126 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE),
128 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T),
130 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF2),
132 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_LS),
135 /* required last entry */
138 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
140 #ifdef CONFIG_IXGBE_DCA
141 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
143 static struct notifier_block dca_notifier = {
144 .notifier_call = ixgbe_notify_dca,
150 #ifdef CONFIG_PCI_IOV
151 static unsigned int max_vfs;
152 module_param(max_vfs, uint, 0);
153 MODULE_PARM_DESC(max_vfs,
154 "Maximum number of virtual functions to allocate per physical function");
155 #endif /* CONFIG_PCI_IOV */
157 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
158 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
159 MODULE_LICENSE("GPL");
160 MODULE_VERSION(DRV_VERSION);
162 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
164 static inline void ixgbe_disable_sriov(struct ixgbe_adapter *adapter)
166 struct ixgbe_hw *hw = &adapter->hw;
171 #ifdef CONFIG_PCI_IOV
172 /* disable iov and allow time for transactions to clear */
173 pci_disable_sriov(adapter->pdev);
176 /* turn off device IOV mode */
177 gcr = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
178 gcr &= ~(IXGBE_GCR_EXT_SRIOV);
179 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr);
180 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
181 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
182 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
184 /* set default pool back to 0 */
185 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
186 vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
187 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
188 IXGBE_WRITE_FLUSH(hw);
190 /* take a breather then clean up driver data */
193 kfree(adapter->vfinfo);
194 adapter->vfinfo = NULL;
196 adapter->num_vfs = 0;
197 adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
200 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
202 if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
203 !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
204 schedule_work(&adapter->service_task);
207 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
209 BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
211 /* flush memory to make sure state is correct before next watchog */
212 smp_mb__before_clear_bit();
213 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
216 struct ixgbe_reg_info {
221 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
223 /* General Registers */
224 {IXGBE_CTRL, "CTRL"},
225 {IXGBE_STATUS, "STATUS"},
226 {IXGBE_CTRL_EXT, "CTRL_EXT"},
228 /* Interrupt Registers */
229 {IXGBE_EICR, "EICR"},
232 {IXGBE_SRRCTL(0), "SRRCTL"},
233 {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
234 {IXGBE_RDLEN(0), "RDLEN"},
235 {IXGBE_RDH(0), "RDH"},
236 {IXGBE_RDT(0), "RDT"},
237 {IXGBE_RXDCTL(0), "RXDCTL"},
238 {IXGBE_RDBAL(0), "RDBAL"},
239 {IXGBE_RDBAH(0), "RDBAH"},
242 {IXGBE_TDBAL(0), "TDBAL"},
243 {IXGBE_TDBAH(0), "TDBAH"},
244 {IXGBE_TDLEN(0), "TDLEN"},
245 {IXGBE_TDH(0), "TDH"},
246 {IXGBE_TDT(0), "TDT"},
247 {IXGBE_TXDCTL(0), "TXDCTL"},
249 /* List Terminator */
255 * ixgbe_regdump - register printout routine
257 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
263 switch (reginfo->ofs) {
264 case IXGBE_SRRCTL(0):
265 for (i = 0; i < 64; i++)
266 regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
268 case IXGBE_DCA_RXCTRL(0):
269 for (i = 0; i < 64; i++)
270 regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
273 for (i = 0; i < 64; i++)
274 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
277 for (i = 0; i < 64; i++)
278 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
281 for (i = 0; i < 64; i++)
282 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
284 case IXGBE_RXDCTL(0):
285 for (i = 0; i < 64; i++)
286 regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
289 for (i = 0; i < 64; i++)
290 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
293 for (i = 0; i < 64; i++)
294 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
297 for (i = 0; i < 64; i++)
298 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
301 for (i = 0; i < 64; i++)
302 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
305 for (i = 0; i < 64; i++)
306 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
309 for (i = 0; i < 64; i++)
310 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
313 for (i = 0; i < 64; i++)
314 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
316 case IXGBE_TXDCTL(0):
317 for (i = 0; i < 64; i++)
318 regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
321 pr_info("%-15s %08x\n", reginfo->name,
322 IXGBE_READ_REG(hw, reginfo->ofs));
326 for (i = 0; i < 8; i++) {
327 snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i*8, i*8+7);
328 pr_err("%-15s", rname);
329 for (j = 0; j < 8; j++)
330 pr_cont(" %08x", regs[i*8+j]);
337 * ixgbe_dump - Print registers, tx-rings and rx-rings
339 static void ixgbe_dump(struct ixgbe_adapter *adapter)
341 struct net_device *netdev = adapter->netdev;
342 struct ixgbe_hw *hw = &adapter->hw;
343 struct ixgbe_reg_info *reginfo;
345 struct ixgbe_ring *tx_ring;
346 struct ixgbe_tx_buffer *tx_buffer_info;
347 union ixgbe_adv_tx_desc *tx_desc;
348 struct my_u0 { u64 a; u64 b; } *u0;
349 struct ixgbe_ring *rx_ring;
350 union ixgbe_adv_rx_desc *rx_desc;
351 struct ixgbe_rx_buffer *rx_buffer_info;
355 if (!netif_msg_hw(adapter))
358 /* Print netdevice Info */
360 dev_info(&adapter->pdev->dev, "Net device Info\n");
361 pr_info("Device Name state "
362 "trans_start last_rx\n");
363 pr_info("%-15s %016lX %016lX %016lX\n",
370 /* Print Registers */
371 dev_info(&adapter->pdev->dev, "Register Dump\n");
372 pr_info(" Register Name Value\n");
373 for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
374 reginfo->name; reginfo++) {
375 ixgbe_regdump(hw, reginfo);
378 /* Print TX Ring Summary */
379 if (!netdev || !netif_running(netdev))
382 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
383 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
384 for (n = 0; n < adapter->num_tx_queues; n++) {
385 tx_ring = adapter->tx_ring[n];
387 &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
388 pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
389 n, tx_ring->next_to_use, tx_ring->next_to_clean,
390 (u64)tx_buffer_info->dma,
391 tx_buffer_info->length,
392 tx_buffer_info->next_to_watch,
393 (u64)tx_buffer_info->time_stamp);
397 if (!netif_msg_tx_done(adapter))
398 goto rx_ring_summary;
400 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
402 /* Transmit Descriptor Formats
404 * Advanced Transmit Descriptor
405 * +--------------------------------------------------------------+
406 * 0 | Buffer Address [63:0] |
407 * +--------------------------------------------------------------+
408 * 8 | PAYLEN | PORTS | IDX | STA | DCMD |DTYP | RSV | DTALEN |
409 * +--------------------------------------------------------------+
410 * 63 46 45 40 39 36 35 32 31 24 23 20 19 0
413 for (n = 0; n < adapter->num_tx_queues; n++) {
414 tx_ring = adapter->tx_ring[n];
415 pr_info("------------------------------------\n");
416 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
417 pr_info("------------------------------------\n");
418 pr_info("T [desc] [address 63:0 ] "
419 "[PlPOIdStDDt Ln] [bi->dma ] "
420 "leng ntw timestamp bi->skb\n");
422 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
423 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
424 tx_buffer_info = &tx_ring->tx_buffer_info[i];
425 u0 = (struct my_u0 *)tx_desc;
426 pr_info("T [0x%03X] %016llX %016llX %016llX"
427 " %04X %p %016llX %p", i,
430 (u64)tx_buffer_info->dma,
431 tx_buffer_info->length,
432 tx_buffer_info->next_to_watch,
433 (u64)tx_buffer_info->time_stamp,
434 tx_buffer_info->skb);
435 if (i == tx_ring->next_to_use &&
436 i == tx_ring->next_to_clean)
438 else if (i == tx_ring->next_to_use)
440 else if (i == tx_ring->next_to_clean)
445 if (netif_msg_pktdata(adapter) &&
446 tx_buffer_info->dma != 0)
447 print_hex_dump(KERN_INFO, "",
448 DUMP_PREFIX_ADDRESS, 16, 1,
449 phys_to_virt(tx_buffer_info->dma),
450 tx_buffer_info->length, true);
454 /* Print RX Rings Summary */
456 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
457 pr_info("Queue [NTU] [NTC]\n");
458 for (n = 0; n < adapter->num_rx_queues; n++) {
459 rx_ring = adapter->rx_ring[n];
460 pr_info("%5d %5X %5X\n",
461 n, rx_ring->next_to_use, rx_ring->next_to_clean);
465 if (!netif_msg_rx_status(adapter))
468 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
470 /* Advanced Receive Descriptor (Read) Format
472 * +-----------------------------------------------------+
473 * 0 | Packet Buffer Address [63:1] |A0/NSE|
474 * +----------------------------------------------+------+
475 * 8 | Header Buffer Address [63:1] | DD |
476 * +-----------------------------------------------------+
479 * Advanced Receive Descriptor (Write-Back) Format
481 * 63 48 47 32 31 30 21 20 16 15 4 3 0
482 * +------------------------------------------------------+
483 * 0 | Packet IP |SPH| HDR_LEN | RSV|Packet| RSS |
484 * | Checksum Ident | | | | Type | Type |
485 * +------------------------------------------------------+
486 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
487 * +------------------------------------------------------+
488 * 63 48 47 32 31 20 19 0
490 for (n = 0; n < adapter->num_rx_queues; n++) {
491 rx_ring = adapter->rx_ring[n];
492 pr_info("------------------------------------\n");
493 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
494 pr_info("------------------------------------\n");
495 pr_info("R [desc] [ PktBuf A0] "
496 "[ HeadBuf DD] [bi->dma ] [bi->skb] "
497 "<-- Adv Rx Read format\n");
498 pr_info("RWB[desc] [PcsmIpSHl PtRs] "
499 "[vl er S cks ln] ---------------- [bi->skb] "
500 "<-- Adv Rx Write-Back format\n");
502 for (i = 0; i < rx_ring->count; i++) {
503 rx_buffer_info = &rx_ring->rx_buffer_info[i];
504 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
505 u0 = (struct my_u0 *)rx_desc;
506 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
507 if (staterr & IXGBE_RXD_STAT_DD) {
508 /* Descriptor Done */
509 pr_info("RWB[0x%03X] %016llX "
510 "%016llX ---------------- %p", i,
513 rx_buffer_info->skb);
515 pr_info("R [0x%03X] %016llX "
516 "%016llX %016llX %p", i,
519 (u64)rx_buffer_info->dma,
520 rx_buffer_info->skb);
522 if (netif_msg_pktdata(adapter)) {
523 print_hex_dump(KERN_INFO, "",
524 DUMP_PREFIX_ADDRESS, 16, 1,
525 phys_to_virt(rx_buffer_info->dma),
526 rx_ring->rx_buf_len, true);
528 if (rx_ring->rx_buf_len
529 < IXGBE_RXBUFFER_2048)
530 print_hex_dump(KERN_INFO, "",
531 DUMP_PREFIX_ADDRESS, 16, 1,
533 rx_buffer_info->page_dma +
534 rx_buffer_info->page_offset
540 if (i == rx_ring->next_to_use)
542 else if (i == rx_ring->next_to_clean)
554 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
558 /* Let firmware take over control of h/w */
559 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
560 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
561 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
564 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
568 /* Let firmware know the driver has taken over */
569 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
570 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
571 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
575 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
576 * @adapter: pointer to adapter struct
577 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
578 * @queue: queue to map the corresponding interrupt to
579 * @msix_vector: the vector to map to the corresponding queue
582 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
583 u8 queue, u8 msix_vector)
586 struct ixgbe_hw *hw = &adapter->hw;
587 switch (hw->mac.type) {
588 case ixgbe_mac_82598EB:
589 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
592 index = (((direction * 64) + queue) >> 2) & 0x1F;
593 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
594 ivar &= ~(0xFF << (8 * (queue & 0x3)));
595 ivar |= (msix_vector << (8 * (queue & 0x3)));
596 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
598 case ixgbe_mac_82599EB:
600 if (direction == -1) {
602 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
603 index = ((queue & 1) * 8);
604 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
605 ivar &= ~(0xFF << index);
606 ivar |= (msix_vector << index);
607 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
610 /* tx or rx causes */
611 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
612 index = ((16 * (queue & 1)) + (8 * direction));
613 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
614 ivar &= ~(0xFF << index);
615 ivar |= (msix_vector << index);
616 IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
624 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
629 switch (adapter->hw.mac.type) {
630 case ixgbe_mac_82598EB:
631 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
632 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
634 case ixgbe_mac_82599EB:
636 mask = (qmask & 0xFFFFFFFF);
637 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
638 mask = (qmask >> 32);
639 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
646 static inline void ixgbe_unmap_tx_resource(struct ixgbe_ring *ring,
647 struct ixgbe_tx_buffer *tx_buffer)
649 if (tx_buffer->dma) {
650 if (tx_buffer->tx_flags & IXGBE_TX_FLAGS_MAPPED_AS_PAGE)
651 dma_unmap_page(ring->dev,
656 dma_unmap_single(ring->dev,
664 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_ring *tx_ring,
665 struct ixgbe_tx_buffer *tx_buffer_info)
667 ixgbe_unmap_tx_resource(tx_ring, tx_buffer_info);
668 if (tx_buffer_info->skb)
669 dev_kfree_skb_any(tx_buffer_info->skb);
670 tx_buffer_info->skb = NULL;
671 /* tx_buffer_info must be completely set up in the transmit path */
674 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
676 struct ixgbe_hw *hw = &adapter->hw;
677 struct ixgbe_hw_stats *hwstats = &adapter->stats;
682 if ((hw->fc.current_mode == ixgbe_fc_full) ||
683 (hw->fc.current_mode == ixgbe_fc_rx_pause)) {
684 switch (hw->mac.type) {
685 case ixgbe_mac_82598EB:
686 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
689 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
691 hwstats->lxoffrxc += data;
693 /* refill credits (no tx hang) if we received xoff */
697 for (i = 0; i < adapter->num_tx_queues; i++)
698 clear_bit(__IXGBE_HANG_CHECK_ARMED,
699 &adapter->tx_ring[i]->state);
701 } else if (!(adapter->dcb_cfg.pfc_mode_enable))
704 /* update stats for each tc, only valid with PFC enabled */
705 for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
706 switch (hw->mac.type) {
707 case ixgbe_mac_82598EB:
708 xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
711 xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
713 hwstats->pxoffrxc[i] += xoff[i];
716 /* disarm tx queues that have received xoff frames */
717 for (i = 0; i < adapter->num_tx_queues; i++) {
718 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
719 u8 tc = tx_ring->dcb_tc;
722 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
726 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
728 return ring->tx_stats.completed;
731 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
733 struct ixgbe_adapter *adapter = netdev_priv(ring->netdev);
734 struct ixgbe_hw *hw = &adapter->hw;
736 u32 head = IXGBE_READ_REG(hw, IXGBE_TDH(ring->reg_idx));
737 u32 tail = IXGBE_READ_REG(hw, IXGBE_TDT(ring->reg_idx));
740 return (head < tail) ?
741 tail - head : (tail + ring->count - head);
746 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
748 u32 tx_done = ixgbe_get_tx_completed(tx_ring);
749 u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
750 u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
753 clear_check_for_tx_hang(tx_ring);
756 * Check for a hung queue, but be thorough. This verifies
757 * that a transmit has been completed since the previous
758 * check AND there is at least one packet pending. The
759 * ARMED bit is set to indicate a potential hang. The
760 * bit is cleared if a pause frame is received to remove
761 * false hang detection due to PFC or 802.3x frames. By
762 * requiring this to fail twice we avoid races with
763 * pfc clearing the ARMED bit and conditions where we
764 * run the check_tx_hang logic with a transmit completion
765 * pending but without time to complete it yet.
767 if ((tx_done_old == tx_done) && tx_pending) {
768 /* make sure it is true for two checks in a row */
769 ret = test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
772 /* update completed stats and continue */
773 tx_ring->tx_stats.tx_done_old = tx_done;
774 /* reset the countdown */
775 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
782 * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
783 * @adapter: driver private struct
785 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
788 /* Do the reset outside of interrupt context */
789 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
790 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
791 ixgbe_service_event_schedule(adapter);
796 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
797 * @q_vector: structure containing interrupt and ring information
798 * @tx_ring: tx ring to clean
800 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
801 struct ixgbe_ring *tx_ring)
803 struct ixgbe_adapter *adapter = q_vector->adapter;
804 struct ixgbe_tx_buffer *tx_buffer;
805 union ixgbe_adv_tx_desc *tx_desc;
806 unsigned int total_bytes = 0, total_packets = 0;
807 unsigned int budget = q_vector->tx.work_limit;
808 u16 i = tx_ring->next_to_clean;
810 tx_buffer = &tx_ring->tx_buffer_info[i];
811 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
813 for (; budget; budget--) {
814 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
816 /* if next_to_watch is not set then there is no work pending */
820 /* if DD is not set pending work has not been completed */
821 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
824 /* count the packet as being completed */
825 tx_ring->tx_stats.completed++;
827 /* clear next_to_watch to prevent false hangs */
828 tx_buffer->next_to_watch = NULL;
830 /* prevent any other reads prior to eop_desc being verified */
834 ixgbe_unmap_tx_resource(tx_ring, tx_buffer);
835 tx_desc->wb.status = 0;
836 if (likely(tx_desc == eop_desc)) {
838 dev_kfree_skb_any(tx_buffer->skb);
839 tx_buffer->skb = NULL;
841 total_bytes += tx_buffer->bytecount;
842 total_packets += tx_buffer->gso_segs;
848 if (unlikely(i == tx_ring->count)) {
851 tx_buffer = tx_ring->tx_buffer_info;
852 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, 0);
858 tx_ring->next_to_clean = i;
859 u64_stats_update_begin(&tx_ring->syncp);
860 tx_ring->stats.bytes += total_bytes;
861 tx_ring->stats.packets += total_packets;
862 u64_stats_update_end(&tx_ring->syncp);
863 q_vector->tx.total_bytes += total_bytes;
864 q_vector->tx.total_packets += total_packets;
866 if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
867 /* schedule immediate reset if we believe we hung */
868 struct ixgbe_hw *hw = &adapter->hw;
869 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
870 e_err(drv, "Detected Tx Unit Hang\n"
872 " TDH, TDT <%x>, <%x>\n"
873 " next_to_use <%x>\n"
874 " next_to_clean <%x>\n"
875 "tx_buffer_info[next_to_clean]\n"
876 " time_stamp <%lx>\n"
878 tx_ring->queue_index,
879 IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
880 IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
881 tx_ring->next_to_use, i,
882 tx_ring->tx_buffer_info[i].time_stamp, jiffies);
884 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
887 "tx hang %d detected on queue %d, resetting adapter\n",
888 adapter->tx_timeout_count + 1, tx_ring->queue_index);
890 /* schedule immediate reset if we believe we hung */
891 ixgbe_tx_timeout_reset(adapter);
893 /* the adapter is about to reset, no point in enabling stuff */
897 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
898 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
899 (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
900 /* Make sure that anybody stopping the queue after this
901 * sees the new next_to_clean.
904 if (__netif_subqueue_stopped(tx_ring->netdev, tx_ring->queue_index) &&
905 !test_bit(__IXGBE_DOWN, &adapter->state)) {
906 netif_wake_subqueue(tx_ring->netdev, tx_ring->queue_index);
907 ++tx_ring->tx_stats.restart_queue;
914 #ifdef CONFIG_IXGBE_DCA
915 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
916 struct ixgbe_ring *rx_ring,
919 struct ixgbe_hw *hw = &adapter->hw;
921 u8 reg_idx = rx_ring->reg_idx;
923 rxctrl = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(reg_idx));
924 switch (hw->mac.type) {
925 case ixgbe_mac_82598EB:
926 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
927 rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
929 case ixgbe_mac_82599EB:
931 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK_82599;
932 rxctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
933 IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599);
938 rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
939 rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
940 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_RRO_EN);
941 IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
944 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
945 struct ixgbe_ring *tx_ring,
948 struct ixgbe_hw *hw = &adapter->hw;
950 u8 reg_idx = tx_ring->reg_idx;
952 switch (hw->mac.type) {
953 case ixgbe_mac_82598EB:
954 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(reg_idx));
955 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
956 txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
957 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
958 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(reg_idx), txctrl);
960 case ixgbe_mac_82599EB:
962 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(reg_idx));
963 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK_82599;
964 txctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
965 IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599);
966 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
967 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(reg_idx), txctrl);
974 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
976 struct ixgbe_adapter *adapter = q_vector->adapter;
977 struct ixgbe_ring *ring;
980 if (q_vector->cpu == cpu)
983 for (ring = q_vector->tx.ring; ring != NULL; ring = ring->next)
984 ixgbe_update_tx_dca(adapter, ring, cpu);
986 for (ring = q_vector->rx.ring; ring != NULL; ring = ring->next)
987 ixgbe_update_rx_dca(adapter, ring, cpu);
994 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
999 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
1002 /* always use CB2 mode, difference is masked in the CB driver */
1003 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
1005 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
1006 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1010 for (i = 0; i < num_q_vectors; i++) {
1011 adapter->q_vector[i]->cpu = -1;
1012 ixgbe_update_dca(adapter->q_vector[i]);
1016 static int __ixgbe_notify_dca(struct device *dev, void *data)
1018 struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1019 unsigned long event = *(unsigned long *)data;
1021 if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
1025 case DCA_PROVIDER_ADD:
1026 /* if we're already enabled, don't do it again */
1027 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1029 if (dca_add_requester(dev) == 0) {
1030 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1031 ixgbe_setup_dca(adapter);
1034 /* Fall Through since DCA is disabled. */
1035 case DCA_PROVIDER_REMOVE:
1036 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1037 dca_remove_requester(dev);
1038 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1039 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
1046 #endif /* CONFIG_IXGBE_DCA */
1048 static inline void ixgbe_rx_hash(union ixgbe_adv_rx_desc *rx_desc,
1049 struct sk_buff *skb)
1051 skb->rxhash = le32_to_cpu(rx_desc->wb.lower.hi_dword.rss);
1055 * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1056 * @adapter: address of board private structure
1057 * @rx_desc: advanced rx descriptor
1059 * Returns : true if it is FCoE pkt
1061 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_adapter *adapter,
1062 union ixgbe_adv_rx_desc *rx_desc)
1064 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1066 return (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
1067 ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1068 (cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1069 IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1073 * ixgbe_receive_skb - Send a completed packet up the stack
1074 * @adapter: board private structure
1075 * @skb: packet to send up
1076 * @status: hardware indication of status of receive
1077 * @rx_ring: rx descriptor ring (for a specific queue) to setup
1078 * @rx_desc: rx descriptor
1080 static void ixgbe_receive_skb(struct ixgbe_q_vector *q_vector,
1081 struct sk_buff *skb, u8 status,
1082 struct ixgbe_ring *ring,
1083 union ixgbe_adv_rx_desc *rx_desc)
1085 struct ixgbe_adapter *adapter = q_vector->adapter;
1086 struct napi_struct *napi = &q_vector->napi;
1087 bool is_vlan = (status & IXGBE_RXD_STAT_VP);
1088 u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
1090 if (is_vlan && (tag & VLAN_VID_MASK))
1091 __vlan_hwaccel_put_tag(skb, tag);
1093 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
1094 napi_gro_receive(napi, skb);
1100 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1101 * @adapter: address of board private structure
1102 * @status_err: hardware indication of status of receive
1103 * @skb: skb currently being received and modified
1104 * @status_err: status error value of last descriptor in packet
1106 static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
1107 union ixgbe_adv_rx_desc *rx_desc,
1108 struct sk_buff *skb,
1111 skb->ip_summed = CHECKSUM_NONE;
1113 /* Rx csum disabled */
1114 if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
1117 /* if IP and error */
1118 if ((status_err & IXGBE_RXD_STAT_IPCS) &&
1119 (status_err & IXGBE_RXDADV_ERR_IPE)) {
1120 adapter->hw_csum_rx_error++;
1124 if (!(status_err & IXGBE_RXD_STAT_L4CS))
1127 if (status_err & IXGBE_RXDADV_ERR_TCPE) {
1128 u16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1131 * 82599 errata, UDP frames with a 0 checksum can be marked as
1134 if ((pkt_info & IXGBE_RXDADV_PKTTYPE_UDP) &&
1135 (adapter->hw.mac.type == ixgbe_mac_82599EB))
1138 adapter->hw_csum_rx_error++;
1142 /* It must be a TCP or UDP packet with a valid checksum */
1143 skb->ip_summed = CHECKSUM_UNNECESSARY;
1146 static inline void ixgbe_release_rx_desc(struct ixgbe_ring *rx_ring, u32 val)
1149 * Force memory writes to complete before letting h/w
1150 * know there are new descriptors to fetch. (Only
1151 * applicable for weak-ordered memory model archs,
1155 writel(val, rx_ring->tail);
1159 * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
1160 * @rx_ring: ring to place buffers on
1161 * @cleaned_count: number of buffers to replace
1163 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1165 union ixgbe_adv_rx_desc *rx_desc;
1166 struct ixgbe_rx_buffer *bi;
1167 struct sk_buff *skb;
1168 u16 i = rx_ring->next_to_use;
1170 /* do nothing if no valid netdev defined */
1171 if (!rx_ring->netdev)
1174 while (cleaned_count--) {
1175 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1176 bi = &rx_ring->rx_buffer_info[i];
1180 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1181 rx_ring->rx_buf_len);
1183 rx_ring->rx_stats.alloc_rx_buff_failed++;
1186 /* initialize queue mapping */
1187 skb_record_rx_queue(skb, rx_ring->queue_index);
1192 bi->dma = dma_map_single(rx_ring->dev,
1194 rx_ring->rx_buf_len,
1196 if (dma_mapping_error(rx_ring->dev, bi->dma)) {
1197 rx_ring->rx_stats.alloc_rx_buff_failed++;
1203 if (ring_is_ps_enabled(rx_ring)) {
1205 bi->page = netdev_alloc_page(rx_ring->netdev);
1207 rx_ring->rx_stats.alloc_rx_page_failed++;
1212 if (!bi->page_dma) {
1213 /* use a half page if we're re-using */
1214 bi->page_offset ^= PAGE_SIZE / 2;
1215 bi->page_dma = dma_map_page(rx_ring->dev,
1220 if (dma_mapping_error(rx_ring->dev,
1222 rx_ring->rx_stats.alloc_rx_page_failed++;
1228 /* Refresh the desc even if buffer_addrs didn't change
1229 * because each write-back erases this info. */
1230 rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
1231 rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
1233 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
1234 rx_desc->read.hdr_addr = 0;
1238 if (i == rx_ring->count)
1243 if (rx_ring->next_to_use != i) {
1244 rx_ring->next_to_use = i;
1245 ixgbe_release_rx_desc(rx_ring, i);
1249 static inline u16 ixgbe_get_hlen(union ixgbe_adv_rx_desc *rx_desc)
1251 /* HW will not DMA in data larger than the given buffer, even if it
1252 * parses the (NFS, of course) header to be larger. In that case, it
1253 * fills the header buffer and spills the rest into the page.
1255 u16 hdr_info = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info);
1256 u16 hlen = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
1257 IXGBE_RXDADV_HDRBUFLEN_SHIFT;
1258 if (hlen > IXGBE_RX_HDR_SIZE)
1259 hlen = IXGBE_RX_HDR_SIZE;
1264 * ixgbe_transform_rsc_queue - change rsc queue into a full packet
1265 * @skb: pointer to the last skb in the rsc queue
1267 * This function changes a queue full of hw rsc buffers into a completed
1268 * packet. It uses the ->prev pointers to find the first packet and then
1269 * turns it into the frag list owner.
1271 static inline struct sk_buff *ixgbe_transform_rsc_queue(struct sk_buff *skb)
1273 unsigned int frag_list_size = 0;
1274 unsigned int skb_cnt = 1;
1277 struct sk_buff *prev = skb->prev;
1278 frag_list_size += skb->len;
1284 skb_shinfo(skb)->frag_list = skb->next;
1286 skb->len += frag_list_size;
1287 skb->data_len += frag_list_size;
1288 skb->truesize += frag_list_size;
1289 IXGBE_RSC_CB(skb)->skb_cnt = skb_cnt;
1294 static inline bool ixgbe_get_rsc_state(union ixgbe_adv_rx_desc *rx_desc)
1296 return !!(le32_to_cpu(rx_desc->wb.lower.lo_dword.data) &
1297 IXGBE_RXDADV_RSCCNT_MASK);
1300 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1301 struct ixgbe_ring *rx_ring,
1304 struct ixgbe_adapter *adapter = q_vector->adapter;
1305 union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
1306 struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
1307 struct sk_buff *skb;
1308 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1309 const int current_node = numa_node_id();
1312 #endif /* IXGBE_FCOE */
1315 u16 cleaned_count = 0;
1316 bool pkt_is_rsc = false;
1318 i = rx_ring->next_to_clean;
1319 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1320 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1322 while (staterr & IXGBE_RXD_STAT_DD) {
1325 rmb(); /* read descriptor and rx_buffer_info after status DD */
1327 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1329 skb = rx_buffer_info->skb;
1330 rx_buffer_info->skb = NULL;
1331 prefetch(skb->data);
1333 if (ring_is_rsc_enabled(rx_ring))
1334 pkt_is_rsc = ixgbe_get_rsc_state(rx_desc);
1336 /* if this is a skb from previous receive DMA will be 0 */
1337 if (rx_buffer_info->dma) {
1340 !(staterr & IXGBE_RXD_STAT_EOP) &&
1343 * When HWRSC is enabled, delay unmapping
1344 * of the first packet. It carries the
1345 * header information, HW may still
1346 * access the header after the writeback.
1347 * Only unmap it when EOP is reached
1349 IXGBE_RSC_CB(skb)->delay_unmap = true;
1350 IXGBE_RSC_CB(skb)->dma = rx_buffer_info->dma;
1352 dma_unmap_single(rx_ring->dev,
1353 rx_buffer_info->dma,
1354 rx_ring->rx_buf_len,
1357 rx_buffer_info->dma = 0;
1359 if (ring_is_ps_enabled(rx_ring)) {
1360 hlen = ixgbe_get_hlen(rx_desc);
1361 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1363 hlen = le16_to_cpu(rx_desc->wb.upper.length);
1368 /* assume packet split since header is unmapped */
1369 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1373 dma_unmap_page(rx_ring->dev,
1374 rx_buffer_info->page_dma,
1377 rx_buffer_info->page_dma = 0;
1378 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1379 rx_buffer_info->page,
1380 rx_buffer_info->page_offset,
1383 if ((page_count(rx_buffer_info->page) == 1) &&
1384 (page_to_nid(rx_buffer_info->page) == current_node))
1385 get_page(rx_buffer_info->page);
1387 rx_buffer_info->page = NULL;
1389 skb->len += upper_len;
1390 skb->data_len += upper_len;
1391 skb->truesize += upper_len;
1395 if (i == rx_ring->count)
1398 next_rxd = IXGBE_RX_DESC_ADV(rx_ring, i);
1403 u32 nextp = (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
1404 IXGBE_RXDADV_NEXTP_SHIFT;
1405 next_buffer = &rx_ring->rx_buffer_info[nextp];
1407 next_buffer = &rx_ring->rx_buffer_info[i];
1410 if (!(staterr & IXGBE_RXD_STAT_EOP)) {
1411 if (ring_is_ps_enabled(rx_ring)) {
1412 rx_buffer_info->skb = next_buffer->skb;
1413 rx_buffer_info->dma = next_buffer->dma;
1414 next_buffer->skb = skb;
1415 next_buffer->dma = 0;
1417 skb->next = next_buffer->skb;
1418 skb->next->prev = skb;
1420 rx_ring->rx_stats.non_eop_descs++;
1425 skb = ixgbe_transform_rsc_queue(skb);
1426 /* if we got here without RSC the packet is invalid */
1428 __pskb_trim(skb, 0);
1429 rx_buffer_info->skb = skb;
1434 if (ring_is_rsc_enabled(rx_ring)) {
1435 if (IXGBE_RSC_CB(skb)->delay_unmap) {
1436 dma_unmap_single(rx_ring->dev,
1437 IXGBE_RSC_CB(skb)->dma,
1438 rx_ring->rx_buf_len,
1440 IXGBE_RSC_CB(skb)->dma = 0;
1441 IXGBE_RSC_CB(skb)->delay_unmap = false;
1445 if (ring_is_ps_enabled(rx_ring))
1446 rx_ring->rx_stats.rsc_count +=
1447 skb_shinfo(skb)->nr_frags;
1449 rx_ring->rx_stats.rsc_count +=
1450 IXGBE_RSC_CB(skb)->skb_cnt;
1451 rx_ring->rx_stats.rsc_flush++;
1454 /* ERR_MASK will only have valid bits if EOP set */
1455 if (unlikely(staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK)) {
1456 dev_kfree_skb_any(skb);
1460 ixgbe_rx_checksum(adapter, rx_desc, skb, staterr);
1461 if (adapter->netdev->features & NETIF_F_RXHASH)
1462 ixgbe_rx_hash(rx_desc, skb);
1464 /* probably a little skewed due to removing CRC */
1465 total_rx_bytes += skb->len;
1468 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1470 /* if ddp, not passing to ULD unless for FCP_RSP or error */
1471 if (ixgbe_rx_is_fcoe(adapter, rx_desc)) {
1472 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb,
1475 dev_kfree_skb_any(skb);
1479 #endif /* IXGBE_FCOE */
1480 ixgbe_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
1484 rx_desc->wb.upper.status_error = 0;
1489 /* return some buffers to hardware, one at a time is too slow */
1490 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
1491 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1495 /* use prefetched values */
1497 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1500 rx_ring->next_to_clean = i;
1501 cleaned_count = ixgbe_desc_unused(rx_ring);
1504 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1507 /* include DDPed FCoE data */
1508 if (ddp_bytes > 0) {
1511 mss = rx_ring->netdev->mtu - sizeof(struct fcoe_hdr) -
1512 sizeof(struct fc_frame_header) -
1513 sizeof(struct fcoe_crc_eof);
1516 total_rx_bytes += ddp_bytes;
1517 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
1519 #endif /* IXGBE_FCOE */
1521 u64_stats_update_begin(&rx_ring->syncp);
1522 rx_ring->stats.packets += total_rx_packets;
1523 rx_ring->stats.bytes += total_rx_bytes;
1524 u64_stats_update_end(&rx_ring->syncp);
1525 q_vector->rx.total_packets += total_rx_packets;
1526 q_vector->rx.total_bytes += total_rx_bytes;
1532 * ixgbe_configure_msix - Configure MSI-X hardware
1533 * @adapter: board private structure
1535 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1538 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1540 struct ixgbe_q_vector *q_vector;
1541 int q_vectors, v_idx;
1544 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1547 * Populate the IVAR table and set the ITR values to the
1548 * corresponding register.
1550 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1551 struct ixgbe_ring *ring;
1552 q_vector = adapter->q_vector[v_idx];
1554 for (ring = q_vector->rx.ring; ring != NULL; ring = ring->next)
1555 ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1557 for (ring = q_vector->tx.ring; ring != NULL; ring = ring->next)
1558 ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1560 if (q_vector->tx.ring && !q_vector->rx.ring)
1562 q_vector->eitr = adapter->tx_eitr_param;
1563 else if (q_vector->rx.ring)
1565 q_vector->eitr = adapter->rx_eitr_param;
1567 ixgbe_write_eitr(q_vector);
1570 switch (adapter->hw.mac.type) {
1571 case ixgbe_mac_82598EB:
1572 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1575 case ixgbe_mac_82599EB:
1576 case ixgbe_mac_X540:
1577 ixgbe_set_ivar(adapter, -1, 1, v_idx);
1583 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1585 /* set up to autoclear timer, and the vectors */
1586 mask = IXGBE_EIMS_ENABLE_MASK;
1587 if (adapter->num_vfs)
1588 mask &= ~(IXGBE_EIMS_OTHER |
1589 IXGBE_EIMS_MAILBOX |
1592 mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
1593 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1596 enum latency_range {
1600 latency_invalid = 255
1604 * ixgbe_update_itr - update the dynamic ITR value based on statistics
1605 * @q_vector: structure containing interrupt and ring information
1606 * @ring_container: structure containing ring performance data
1608 * Stores a new ITR value based on packets and byte
1609 * counts during the last interrupt. The advantage of per interrupt
1610 * computation is faster updates and more accurate ITR for the current
1611 * traffic pattern. Constants in this function were computed
1612 * based on theoretical maximum wire speed and thresholds were set based
1613 * on testing data as well as attempting to minimize response time
1614 * while increasing bulk throughput.
1615 * this functionality is controlled by the InterruptThrottleRate module
1616 * parameter (see ixgbe_param.c)
1618 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
1619 struct ixgbe_ring_container *ring_container)
1622 struct ixgbe_adapter *adapter = q_vector->adapter;
1623 int bytes = ring_container->total_bytes;
1624 int packets = ring_container->total_packets;
1626 u8 itr_setting = ring_container->itr;
1631 /* simple throttlerate management
1632 * 0-20MB/s lowest (100000 ints/s)
1633 * 20-100MB/s low (20000 ints/s)
1634 * 100-1249MB/s bulk (8000 ints/s)
1636 /* what was last interrupt timeslice? */
1637 timepassed_us = 1000000/q_vector->eitr;
1638 bytes_perint = bytes / timepassed_us; /* bytes/usec */
1640 switch (itr_setting) {
1641 case lowest_latency:
1642 if (bytes_perint > adapter->eitr_low)
1643 itr_setting = low_latency;
1646 if (bytes_perint > adapter->eitr_high)
1647 itr_setting = bulk_latency;
1648 else if (bytes_perint <= adapter->eitr_low)
1649 itr_setting = lowest_latency;
1652 if (bytes_perint <= adapter->eitr_high)
1653 itr_setting = low_latency;
1657 /* clear work counters since we have the values we need */
1658 ring_container->total_bytes = 0;
1659 ring_container->total_packets = 0;
1661 /* write updated itr to ring container */
1662 ring_container->itr = itr_setting;
1666 * ixgbe_write_eitr - write EITR register in hardware specific way
1667 * @q_vector: structure containing interrupt and ring information
1669 * This function is made to be called by ethtool and by the driver
1670 * when it needs to update EITR registers at runtime. Hardware
1671 * specific quirks/differences are taken care of here.
1673 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1675 struct ixgbe_adapter *adapter = q_vector->adapter;
1676 struct ixgbe_hw *hw = &adapter->hw;
1677 int v_idx = q_vector->v_idx;
1678 u32 itr_reg = EITR_INTS_PER_SEC_TO_REG(q_vector->eitr);
1680 switch (adapter->hw.mac.type) {
1681 case ixgbe_mac_82598EB:
1682 /* must write high and low 16 bits to reset counter */
1683 itr_reg |= (itr_reg << 16);
1685 case ixgbe_mac_82599EB:
1686 case ixgbe_mac_X540:
1688 * 82599 and X540 can support a value of zero, so allow it for
1689 * max interrupt rate, but there is an errata where it can
1690 * not be zero with RSC
1693 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
1697 * set the WDIS bit to not clear the timer bits and cause an
1698 * immediate assertion of the interrupt
1700 itr_reg |= IXGBE_EITR_CNT_WDIS;
1705 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1708 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
1710 u32 new_itr = q_vector->eitr;
1713 ixgbe_update_itr(q_vector, &q_vector->tx);
1714 ixgbe_update_itr(q_vector, &q_vector->rx);
1716 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
1718 switch (current_itr) {
1719 /* counts and packets in update_itr are dependent on these numbers */
1720 case lowest_latency:
1724 new_itr = 20000; /* aka hwitr = ~200 */
1733 if (new_itr != q_vector->eitr) {
1734 /* do an exponential smoothing */
1735 new_itr = ((q_vector->eitr * 9) + new_itr)/10;
1737 /* save the algorithm value here */
1738 q_vector->eitr = new_itr;
1740 ixgbe_write_eitr(q_vector);
1745 * ixgbe_check_overtemp_subtask - check for over tempurature
1746 * @adapter: pointer to adapter
1748 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
1750 struct ixgbe_hw *hw = &adapter->hw;
1751 u32 eicr = adapter->interrupt_event;
1753 if (test_bit(__IXGBE_DOWN, &adapter->state))
1756 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
1757 !(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
1760 adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
1762 switch (hw->device_id) {
1763 case IXGBE_DEV_ID_82599_T3_LOM:
1765 * Since the warning interrupt is for both ports
1766 * we don't have to check if:
1767 * - This interrupt wasn't for our port.
1768 * - We may have missed the interrupt so always have to
1769 * check if we got a LSC
1771 if (!(eicr & IXGBE_EICR_GPI_SDP0) &&
1772 !(eicr & IXGBE_EICR_LSC))
1775 if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
1777 bool link_up = false;
1779 hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
1785 /* Check if this is not due to overtemp */
1786 if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
1791 if (!(eicr & IXGBE_EICR_GPI_SDP0))
1796 "Network adapter has been stopped because it has over heated. "
1797 "Restart the computer. If the problem persists, "
1798 "power off the system and replace the adapter\n");
1800 adapter->interrupt_event = 0;
1803 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
1805 struct ixgbe_hw *hw = &adapter->hw;
1807 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
1808 (eicr & IXGBE_EICR_GPI_SDP1)) {
1809 e_crit(probe, "Fan has stopped, replace the adapter\n");
1810 /* write to clear the interrupt */
1811 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1815 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
1817 struct ixgbe_hw *hw = &adapter->hw;
1819 if (eicr & IXGBE_EICR_GPI_SDP2) {
1820 /* Clear the interrupt */
1821 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
1822 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1823 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
1824 ixgbe_service_event_schedule(adapter);
1828 if (eicr & IXGBE_EICR_GPI_SDP1) {
1829 /* Clear the interrupt */
1830 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1831 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1832 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
1833 ixgbe_service_event_schedule(adapter);
1838 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1840 struct ixgbe_hw *hw = &adapter->hw;
1843 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1844 adapter->link_check_timeout = jiffies;
1845 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1846 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1847 IXGBE_WRITE_FLUSH(hw);
1848 ixgbe_service_event_schedule(adapter);
1852 static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
1854 struct ixgbe_adapter *adapter = data;
1855 struct ixgbe_hw *hw = &adapter->hw;
1859 * Workaround for Silicon errata. Use clear-by-write instead
1860 * of clear-by-read. Reading with EICS will return the
1861 * interrupt causes without clearing, which later be done
1862 * with the write to EICR.
1864 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1865 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1867 if (eicr & IXGBE_EICR_LSC)
1868 ixgbe_check_lsc(adapter);
1870 if (eicr & IXGBE_EICR_MAILBOX)
1871 ixgbe_msg_task(adapter);
1873 switch (hw->mac.type) {
1874 case ixgbe_mac_82599EB:
1875 case ixgbe_mac_X540:
1876 /* Handle Flow Director Full threshold interrupt */
1877 if (eicr & IXGBE_EICR_FLOW_DIR) {
1878 int reinit_count = 0;
1880 for (i = 0; i < adapter->num_tx_queues; i++) {
1881 struct ixgbe_ring *ring = adapter->tx_ring[i];
1882 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
1887 /* no more flow director interrupts until after init */
1888 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
1889 eicr &= ~IXGBE_EICR_FLOW_DIR;
1890 adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
1891 ixgbe_service_event_schedule(adapter);
1894 ixgbe_check_sfp_event(adapter, eicr);
1895 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
1896 ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC))) {
1897 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1898 adapter->interrupt_event = eicr;
1899 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
1900 ixgbe_service_event_schedule(adapter);
1908 ixgbe_check_fan_failure(adapter, eicr);
1910 /* re-enable the original interrupt state, no lsc, no queues */
1911 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1912 IXGBE_WRITE_REG(hw, IXGBE_EIMS, eicr &
1913 ~(IXGBE_EIMS_LSC | IXGBE_EIMS_RTX_QUEUE));
1918 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
1922 struct ixgbe_hw *hw = &adapter->hw;
1924 switch (hw->mac.type) {
1925 case ixgbe_mac_82598EB:
1926 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1927 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
1929 case ixgbe_mac_82599EB:
1930 case ixgbe_mac_X540:
1931 mask = (qmask & 0xFFFFFFFF);
1933 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
1934 mask = (qmask >> 32);
1936 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
1941 /* skip the flush */
1944 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
1948 struct ixgbe_hw *hw = &adapter->hw;
1950 switch (hw->mac.type) {
1951 case ixgbe_mac_82598EB:
1952 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1953 IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
1955 case ixgbe_mac_82599EB:
1956 case ixgbe_mac_X540:
1957 mask = (qmask & 0xFFFFFFFF);
1959 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
1960 mask = (qmask >> 32);
1962 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
1967 /* skip the flush */
1970 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
1972 struct ixgbe_q_vector *q_vector = data;
1974 /* EIAM disabled interrupts (on this vector) for us */
1976 if (q_vector->rx.ring || q_vector->tx.ring)
1977 napi_schedule(&q_vector->napi);
1982 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
1985 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
1986 struct ixgbe_ring *rx_ring = a->rx_ring[r_idx];
1988 rx_ring->q_vector = q_vector;
1989 rx_ring->next = q_vector->rx.ring;
1990 q_vector->rx.ring = rx_ring;
1991 q_vector->rx.count++;
1994 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
1997 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
1998 struct ixgbe_ring *tx_ring = a->tx_ring[t_idx];
2000 tx_ring->q_vector = q_vector;
2001 tx_ring->next = q_vector->tx.ring;
2002 q_vector->tx.ring = tx_ring;
2003 q_vector->tx.count++;
2004 q_vector->tx.work_limit = a->tx_work_limit;
2008 * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
2009 * @adapter: board private structure to initialize
2011 * This function maps descriptor rings to the queue-specific vectors
2012 * we were allotted through the MSI-X enabling code. Ideally, we'd have
2013 * one vector per ring/queue, but on a constrained vector budget, we
2014 * group the rings as "efficiently" as possible. You would add new
2015 * mapping configurations in here.
2017 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter)
2021 int rxr_idx = 0, txr_idx = 0;
2022 int rxr_remaining = adapter->num_rx_queues;
2023 int txr_remaining = adapter->num_tx_queues;
2028 /* No mapping required if MSI-X is disabled. */
2029 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2032 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2035 * The ideal configuration...
2036 * We have enough vectors to map one per queue.
2038 if (q_vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
2039 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
2040 map_vector_to_rxq(adapter, v_start, rxr_idx);
2042 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
2043 map_vector_to_txq(adapter, v_start, txr_idx);
2049 * If we don't have enough vectors for a 1-to-1
2050 * mapping, we'll have to group them so there are
2051 * multiple queues per vector.
2053 /* Re-adjusting *qpv takes care of the remainder. */
2054 for (i = v_start; i < q_vectors; i++) {
2055 rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - i);
2056 for (j = 0; j < rqpv; j++) {
2057 map_vector_to_rxq(adapter, i, rxr_idx);
2061 tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - i);
2062 for (j = 0; j < tqpv; j++) {
2063 map_vector_to_txq(adapter, i, txr_idx);
2073 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2074 * @adapter: board private structure
2076 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2077 * interrupts from the kernel.
2079 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2081 struct net_device *netdev = adapter->netdev;
2082 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2086 err = ixgbe_map_rings_to_vectors(adapter);
2090 for (vector = 0; vector < q_vectors; vector++) {
2091 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2092 struct msix_entry *entry = &adapter->msix_entries[vector];
2094 if (q_vector->tx.ring && q_vector->rx.ring) {
2095 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2096 "%s-%s-%d", netdev->name, "TxRx", ri++);
2098 } else if (q_vector->rx.ring) {
2099 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2100 "%s-%s-%d", netdev->name, "rx", ri++);
2101 } else if (q_vector->tx.ring) {
2102 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2103 "%s-%s-%d", netdev->name, "tx", ti++);
2105 /* skip this unused q_vector */
2108 err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
2109 q_vector->name, q_vector);
2111 e_err(probe, "request_irq failed for MSIX interrupt "
2112 "Error: %d\n", err);
2113 goto free_queue_irqs;
2115 /* If Flow Director is enabled, set interrupt affinity */
2116 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2117 /* assign the mask for this irq */
2118 irq_set_affinity_hint(entry->vector,
2119 q_vector->affinity_mask);
2123 sprintf(adapter->lsc_int_name, "%s:lsc", netdev->name);
2124 err = request_irq(adapter->msix_entries[vector].vector,
2125 ixgbe_msix_lsc, 0, adapter->lsc_int_name, adapter);
2127 e_err(probe, "request_irq for msix_lsc failed: %d\n", err);
2128 goto free_queue_irqs;
2136 irq_set_affinity_hint(adapter->msix_entries[vector].vector,
2138 free_irq(adapter->msix_entries[vector].vector,
2139 adapter->q_vector[vector]);
2141 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2142 pci_disable_msix(adapter->pdev);
2143 kfree(adapter->msix_entries);
2144 adapter->msix_entries = NULL;
2149 * ixgbe_irq_enable - Enable default interrupt generation settings
2150 * @adapter: board private structure
2152 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2157 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2158 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2159 mask |= IXGBE_EIMS_GPI_SDP0;
2160 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2161 mask |= IXGBE_EIMS_GPI_SDP1;
2162 switch (adapter->hw.mac.type) {
2163 case ixgbe_mac_82599EB:
2164 case ixgbe_mac_X540:
2165 mask |= IXGBE_EIMS_ECC;
2166 mask |= IXGBE_EIMS_GPI_SDP1;
2167 mask |= IXGBE_EIMS_GPI_SDP2;
2168 if (adapter->num_vfs)
2169 mask |= IXGBE_EIMS_MAILBOX;
2174 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)
2175 mask |= IXGBE_EIMS_FLOW_DIR;
2177 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2179 ixgbe_irq_enable_queues(adapter, ~0);
2181 IXGBE_WRITE_FLUSH(&adapter->hw);
2183 if (adapter->num_vfs > 32) {
2184 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
2185 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2190 * ixgbe_intr - legacy mode Interrupt Handler
2191 * @irq: interrupt number
2192 * @data: pointer to a network interface device structure
2194 static irqreturn_t ixgbe_intr(int irq, void *data)
2196 struct ixgbe_adapter *adapter = data;
2197 struct ixgbe_hw *hw = &adapter->hw;
2198 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2202 * Workaround for silicon errata on 82598. Mask the interrupts
2203 * before the read of EICR.
2205 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2207 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2208 * therefore no explict interrupt disable is necessary */
2209 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2212 * shared interrupt alert!
2213 * make sure interrupts are enabled because the read will
2214 * have disabled interrupts due to EIAM
2215 * finish the workaround of silicon errata on 82598. Unmask
2216 * the interrupt that we masked before the EICR read.
2218 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2219 ixgbe_irq_enable(adapter, true, true);
2220 return IRQ_NONE; /* Not our interrupt */
2223 if (eicr & IXGBE_EICR_LSC)
2224 ixgbe_check_lsc(adapter);
2226 switch (hw->mac.type) {
2227 case ixgbe_mac_82599EB:
2228 ixgbe_check_sfp_event(adapter, eicr);
2229 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2230 ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC))) {
2231 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2232 adapter->interrupt_event = eicr;
2233 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2234 ixgbe_service_event_schedule(adapter);
2242 ixgbe_check_fan_failure(adapter, eicr);
2244 if (napi_schedule_prep(&(q_vector->napi))) {
2245 /* would disable interrupts here but EIAM disabled it */
2246 __napi_schedule(&(q_vector->napi));
2250 * re-enable link(maybe) and non-queue interrupts, no flush.
2251 * ixgbe_poll will re-enable the queue interrupts
2254 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2255 ixgbe_irq_enable(adapter, false, false);
2260 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
2262 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2265 /* legacy and MSI only use one vector */
2266 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2269 for (i = 0; i < adapter->num_rx_queues; i++) {
2270 adapter->rx_ring[i]->q_vector = NULL;
2271 adapter->rx_ring[i]->next = NULL;
2273 for (i = 0; i < adapter->num_tx_queues; i++) {
2274 adapter->tx_ring[i]->q_vector = NULL;
2275 adapter->tx_ring[i]->next = NULL;
2278 for (i = 0; i < q_vectors; i++) {
2279 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
2280 memset(&q_vector->rx, 0, sizeof(struct ixgbe_ring_container));
2281 memset(&q_vector->tx, 0, sizeof(struct ixgbe_ring_container));
2286 * ixgbe_request_irq - initialize interrupts
2287 * @adapter: board private structure
2289 * Attempts to configure interrupts using the best available
2290 * capabilities of the hardware and kernel.
2292 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2294 struct net_device *netdev = adapter->netdev;
2297 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2298 err = ixgbe_request_msix_irqs(adapter);
2299 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
2300 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2301 netdev->name, adapter);
2303 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2304 netdev->name, adapter);
2308 e_err(probe, "request_irq failed, Error %d\n", err);
2313 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2315 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2318 q_vectors = adapter->num_msix_vectors;
2321 free_irq(adapter->msix_entries[i].vector, adapter);
2324 for (; i >= 0; i--) {
2325 /* free only the irqs that were actually requested */
2326 if (!adapter->q_vector[i]->rx.ring &&
2327 !adapter->q_vector[i]->tx.ring)
2330 /* clear the affinity_mask in the IRQ descriptor */
2331 irq_set_affinity_hint(adapter->msix_entries[i].vector,
2334 free_irq(adapter->msix_entries[i].vector,
2335 adapter->q_vector[i]);
2338 free_irq(adapter->pdev->irq, adapter);
2341 /* clear q_vector state information */
2342 ixgbe_reset_q_vectors(adapter);
2346 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2347 * @adapter: board private structure
2349 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2351 switch (adapter->hw.mac.type) {
2352 case ixgbe_mac_82598EB:
2353 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2355 case ixgbe_mac_82599EB:
2356 case ixgbe_mac_X540:
2357 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2358 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2359 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2360 if (adapter->num_vfs > 32)
2361 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
2366 IXGBE_WRITE_FLUSH(&adapter->hw);
2367 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2369 for (i = 0; i < adapter->num_msix_vectors; i++)
2370 synchronize_irq(adapter->msix_entries[i].vector);
2372 synchronize_irq(adapter->pdev->irq);
2377 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2380 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2382 struct ixgbe_hw *hw = &adapter->hw;
2384 IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
2385 EITR_INTS_PER_SEC_TO_REG(adapter->rx_eitr_param));
2387 ixgbe_set_ivar(adapter, 0, 0, 0);
2388 ixgbe_set_ivar(adapter, 1, 0, 0);
2390 map_vector_to_rxq(adapter, 0, 0);
2391 map_vector_to_txq(adapter, 0, 0);
2393 e_info(hw, "Legacy interrupt IVAR setup done\n");
2397 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2398 * @adapter: board private structure
2399 * @ring: structure containing ring specific data
2401 * Configure the Tx descriptor ring after a reset.
2403 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2404 struct ixgbe_ring *ring)
2406 struct ixgbe_hw *hw = &adapter->hw;
2407 u64 tdba = ring->dma;
2410 u8 reg_idx = ring->reg_idx;
2412 /* disable queue to avoid issues while updating state */
2413 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2414 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx),
2415 txdctl & ~IXGBE_TXDCTL_ENABLE);
2416 IXGBE_WRITE_FLUSH(hw);
2418 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2419 (tdba & DMA_BIT_MASK(32)));
2420 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2421 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2422 ring->count * sizeof(union ixgbe_adv_tx_desc));
2423 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2424 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2425 ring->tail = hw->hw_addr + IXGBE_TDT(reg_idx);
2427 /* configure fetching thresholds */
2428 if (adapter->rx_itr_setting == 0) {
2429 /* cannot set wthresh when itr==0 */
2430 txdctl &= ~0x007F0000;
2432 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
2433 txdctl |= (8 << 16);
2435 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2436 /* PThresh workaround for Tx hang with DFP enabled. */
2440 /* reinitialize flowdirector state */
2441 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2442 adapter->atr_sample_rate) {
2443 ring->atr_sample_rate = adapter->atr_sample_rate;
2444 ring->atr_count = 0;
2445 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
2447 ring->atr_sample_rate = 0;
2450 clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
2453 txdctl |= IXGBE_TXDCTL_ENABLE;
2454 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
2456 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2457 if (hw->mac.type == ixgbe_mac_82598EB &&
2458 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2461 /* poll to verify queue is enabled */
2463 usleep_range(1000, 2000);
2464 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2465 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
2467 e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
2470 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
2472 struct ixgbe_hw *hw = &adapter->hw;
2475 u8 tcs = netdev_get_num_tc(adapter->netdev);
2477 if (hw->mac.type == ixgbe_mac_82598EB)
2480 /* disable the arbiter while setting MTQC */
2481 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2482 rttdcs |= IXGBE_RTTDCS_ARBDIS;
2483 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2485 /* set transmit pool layout */
2486 switch (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
2487 case (IXGBE_FLAG_SRIOV_ENABLED):
2488 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2489 (IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF));
2493 reg = IXGBE_MTQC_64Q_1PB;
2495 reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
2497 reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
2499 IXGBE_WRITE_REG(hw, IXGBE_MTQC, reg);
2501 /* Enable Security TX Buffer IFG for multiple pb */
2503 reg = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
2504 reg |= IXGBE_SECTX_DCB;
2505 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, reg);
2510 /* re-enable the arbiter */
2511 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2512 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2516 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2517 * @adapter: board private structure
2519 * Configure the Tx unit of the MAC after a reset.
2521 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2523 struct ixgbe_hw *hw = &adapter->hw;
2527 ixgbe_setup_mtqc(adapter);
2529 if (hw->mac.type != ixgbe_mac_82598EB) {
2530 /* DMATXCTL.EN must be before Tx queues are enabled */
2531 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2532 dmatxctl |= IXGBE_DMATXCTL_TE;
2533 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2536 /* Setup the HW Tx Head and Tail descriptor pointers */
2537 for (i = 0; i < adapter->num_tx_queues; i++)
2538 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
2541 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2543 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2544 struct ixgbe_ring *rx_ring)
2547 u8 reg_idx = rx_ring->reg_idx;
2549 switch (adapter->hw.mac.type) {
2550 case ixgbe_mac_82598EB: {
2551 struct ixgbe_ring_feature *feature = adapter->ring_feature;
2552 const int mask = feature[RING_F_RSS].mask;
2553 reg_idx = reg_idx & mask;
2556 case ixgbe_mac_82599EB:
2557 case ixgbe_mac_X540:
2562 srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx));
2564 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
2565 srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
2566 if (adapter->num_vfs)
2567 srrctl |= IXGBE_SRRCTL_DROP_EN;
2569 srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
2570 IXGBE_SRRCTL_BSIZEHDR_MASK;
2572 if (ring_is_ps_enabled(rx_ring)) {
2573 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
2574 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2576 srrctl |= (PAGE_SIZE / 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2578 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2580 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
2581 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2582 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2585 IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx), srrctl);
2588 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2590 struct ixgbe_hw *hw = &adapter->hw;
2591 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2592 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2593 0x6A3E67EA, 0x14364D17, 0x3BED200D};
2594 u32 mrqc = 0, reta = 0;
2597 u8 tcs = netdev_get_num_tc(adapter->netdev);
2598 int maxq = adapter->ring_feature[RING_F_RSS].indices;
2601 maxq = min(maxq, adapter->num_tx_queues / tcs);
2603 /* Fill out hash function seeds */
2604 for (i = 0; i < 10; i++)
2605 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2607 /* Fill out redirection table */
2608 for (i = 0, j = 0; i < 128; i++, j++) {
2611 /* reta = 4-byte sliding window of
2612 * 0x00..(indices-1)(indices-1)00..etc. */
2613 reta = (reta << 8) | (j * 0x11);
2615 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2618 /* Disable indicating checksum in descriptor, enables RSS hash */
2619 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2620 rxcsum |= IXGBE_RXCSUM_PCSD;
2621 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2623 if (adapter->hw.mac.type == ixgbe_mac_82598EB &&
2624 (adapter->flags & IXGBE_FLAG_RSS_ENABLED)) {
2625 mrqc = IXGBE_MRQC_RSSEN;
2627 int mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2628 | IXGBE_FLAG_SRIOV_ENABLED);
2631 case (IXGBE_FLAG_RSS_ENABLED):
2633 mrqc = IXGBE_MRQC_RSSEN;
2635 mrqc = IXGBE_MRQC_RTRSS4TCEN;
2637 mrqc = IXGBE_MRQC_RTRSS8TCEN;
2639 case (IXGBE_FLAG_SRIOV_ENABLED):
2640 mrqc = IXGBE_MRQC_VMDQEN;
2647 /* Perform hash on these packet types */
2648 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2649 | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2650 | IXGBE_MRQC_RSS_FIELD_IPV6
2651 | IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
2653 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2657 * ixgbe_configure_rscctl - enable RSC for the indicated ring
2658 * @adapter: address of board private structure
2659 * @index: index of ring to set
2661 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
2662 struct ixgbe_ring *ring)
2664 struct ixgbe_hw *hw = &adapter->hw;
2667 u8 reg_idx = ring->reg_idx;
2669 if (!ring_is_rsc_enabled(ring))
2672 rx_buf_len = ring->rx_buf_len;
2673 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
2674 rscctrl |= IXGBE_RSCCTL_RSCEN;
2676 * we must limit the number of descriptors so that the
2677 * total size of max desc * buf_len is not greater
2680 if (ring_is_ps_enabled(ring)) {
2681 #if (MAX_SKB_FRAGS > 16)
2682 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2683 #elif (MAX_SKB_FRAGS > 8)
2684 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2685 #elif (MAX_SKB_FRAGS > 4)
2686 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2688 rscctrl |= IXGBE_RSCCTL_MAXDESC_1;
2691 if (rx_buf_len < IXGBE_RXBUFFER_4096)
2692 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2693 else if (rx_buf_len < IXGBE_RXBUFFER_8192)
2694 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2696 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2698 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
2702 * ixgbe_set_uta - Set unicast filter table address
2703 * @adapter: board private structure
2705 * The unicast table address is a register array of 32-bit registers.
2706 * The table is meant to be used in a way similar to how the MTA is used
2707 * however due to certain limitations in the hardware it is necessary to
2708 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
2709 * enable bit to allow vlan tag stripping when promiscuous mode is enabled
2711 static void ixgbe_set_uta(struct ixgbe_adapter *adapter)
2713 struct ixgbe_hw *hw = &adapter->hw;
2716 /* The UTA table only exists on 82599 hardware and newer */
2717 if (hw->mac.type < ixgbe_mac_82599EB)
2720 /* we only need to do this if VMDq is enabled */
2721 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
2724 for (i = 0; i < 128; i++)
2725 IXGBE_WRITE_REG(hw, IXGBE_UTA(i), ~0);
2728 #define IXGBE_MAX_RX_DESC_POLL 10
2729 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
2730 struct ixgbe_ring *ring)
2732 struct ixgbe_hw *hw = &adapter->hw;
2733 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
2735 u8 reg_idx = ring->reg_idx;
2737 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2738 if (hw->mac.type == ixgbe_mac_82598EB &&
2739 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2743 usleep_range(1000, 2000);
2744 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2745 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
2748 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
2749 "the polling period\n", reg_idx);
2753 void ixgbe_disable_rx_queue(struct ixgbe_adapter *adapter,
2754 struct ixgbe_ring *ring)
2756 struct ixgbe_hw *hw = &adapter->hw;
2757 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
2759 u8 reg_idx = ring->reg_idx;
2761 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2762 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
2764 /* write value back with RXDCTL.ENABLE bit cleared */
2765 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
2767 if (hw->mac.type == ixgbe_mac_82598EB &&
2768 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2771 /* the hardware may take up to 100us to really disable the rx queue */
2774 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2775 } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
2778 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not cleared within "
2779 "the polling period\n", reg_idx);
2783 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
2784 struct ixgbe_ring *ring)
2786 struct ixgbe_hw *hw = &adapter->hw;
2787 u64 rdba = ring->dma;
2789 u8 reg_idx = ring->reg_idx;
2791 /* disable queue to avoid issues while updating state */
2792 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2793 ixgbe_disable_rx_queue(adapter, ring);
2795 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
2796 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
2797 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
2798 ring->count * sizeof(union ixgbe_adv_rx_desc));
2799 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
2800 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
2801 ring->tail = hw->hw_addr + IXGBE_RDT(reg_idx);
2803 ixgbe_configure_srrctl(adapter, ring);
2804 ixgbe_configure_rscctl(adapter, ring);
2806 /* If operating in IOV mode set RLPML for X540 */
2807 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
2808 hw->mac.type == ixgbe_mac_X540) {
2809 rxdctl &= ~IXGBE_RXDCTL_RLPMLMASK;
2810 rxdctl |= ((ring->netdev->mtu + ETH_HLEN +
2811 ETH_FCS_LEN + VLAN_HLEN) | IXGBE_RXDCTL_RLPML_EN);
2814 if (hw->mac.type == ixgbe_mac_82598EB) {
2816 * enable cache line friendly hardware writes:
2817 * PTHRESH=32 descriptors (half the internal cache),
2818 * this also removes ugly rx_no_buffer_count increment
2819 * HTHRESH=4 descriptors (to minimize latency on fetch)
2820 * WTHRESH=8 burst writeback up to two cache lines
2822 rxdctl &= ~0x3FFFFF;
2826 /* enable receive descriptor ring */
2827 rxdctl |= IXGBE_RXDCTL_ENABLE;
2828 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
2830 ixgbe_rx_desc_queue_enable(adapter, ring);
2831 ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
2834 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
2836 struct ixgbe_hw *hw = &adapter->hw;
2839 /* PSRTYPE must be initialized in non 82598 adapters */
2840 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
2841 IXGBE_PSRTYPE_UDPHDR |
2842 IXGBE_PSRTYPE_IPV4HDR |
2843 IXGBE_PSRTYPE_L2HDR |
2844 IXGBE_PSRTYPE_IPV6HDR;
2846 if (hw->mac.type == ixgbe_mac_82598EB)
2849 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED)
2850 psrtype |= (adapter->num_rx_queues_per_pool << 29);
2852 for (p = 0; p < adapter->num_rx_pools; p++)
2853 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(adapter->num_vfs + p),
2857 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
2859 struct ixgbe_hw *hw = &adapter->hw;
2862 u32 reg_offset, vf_shift;
2865 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
2868 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
2869 vt_reg_bits = IXGBE_VMD_CTL_VMDQ_EN | IXGBE_VT_CTL_REPLEN;
2870 vt_reg_bits |= (adapter->num_vfs << IXGBE_VT_CTL_POOL_SHIFT);
2871 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl | vt_reg_bits);
2873 vf_shift = adapter->num_vfs % 32;
2874 reg_offset = (adapter->num_vfs > 32) ? 1 : 0;
2876 /* Enable only the PF's pool for Tx/Rx */
2877 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (1 << vf_shift));
2878 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), 0);
2879 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (1 << vf_shift));
2880 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), 0);
2881 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
2883 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
2884 hw->mac.ops.set_vmdq(hw, 0, adapter->num_vfs);
2887 * Set up VF register offsets for selected VT Mode,
2888 * i.e. 32 or 64 VFs for SR-IOV
2890 gcr_ext = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
2891 gcr_ext |= IXGBE_GCR_EXT_MSIX_EN;
2892 gcr_ext |= IXGBE_GCR_EXT_VT_MODE_64;
2893 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
2895 /* enable Tx loopback for VF/PF communication */
2896 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
2897 /* Enable MAC Anti-Spoofing */
2898 hw->mac.ops.set_mac_anti_spoofing(hw,
2899 (adapter->antispoofing_enabled =
2900 (adapter->num_vfs != 0)),
2904 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
2906 struct ixgbe_hw *hw = &adapter->hw;
2907 struct net_device *netdev = adapter->netdev;
2908 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2910 struct ixgbe_ring *rx_ring;
2914 /* Decide whether to use packet split mode or not */
2916 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
2918 /* Do not use packet split if we're in SR-IOV Mode */
2919 if (adapter->num_vfs)
2920 adapter->flags &= ~IXGBE_FLAG_RX_PS_ENABLED;
2922 /* Disable packet split due to 82599 erratum #45 */
2923 if (hw->mac.type == ixgbe_mac_82599EB)
2924 adapter->flags &= ~IXGBE_FLAG_RX_PS_ENABLED;
2926 /* Set the RX buffer length according to the mode */
2927 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
2928 rx_buf_len = IXGBE_RX_HDR_SIZE;
2930 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
2931 (netdev->mtu <= ETH_DATA_LEN))
2932 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
2934 rx_buf_len = ALIGN(max_frame + VLAN_HLEN, 1024);
2938 /* adjust max frame to be able to do baby jumbo for FCoE */
2939 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
2940 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
2941 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
2943 #endif /* IXGBE_FCOE */
2944 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
2945 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
2946 mhadd &= ~IXGBE_MHADD_MFS_MASK;
2947 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
2949 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
2952 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2953 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
2954 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2955 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
2958 * Setup the HW Rx Head and Tail Descriptor Pointers and
2959 * the Base and Length of the Rx Descriptor Ring
2961 for (i = 0; i < adapter->num_rx_queues; i++) {
2962 rx_ring = adapter->rx_ring[i];
2963 rx_ring->rx_buf_len = rx_buf_len;
2965 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)
2966 set_ring_ps_enabled(rx_ring);
2968 clear_ring_ps_enabled(rx_ring);
2970 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
2971 set_ring_rsc_enabled(rx_ring);
2973 clear_ring_rsc_enabled(rx_ring);
2976 if (netdev->features & NETIF_F_FCOE_MTU) {
2977 struct ixgbe_ring_feature *f;
2978 f = &adapter->ring_feature[RING_F_FCOE];
2979 if ((i >= f->mask) && (i < f->mask + f->indices)) {
2980 clear_ring_ps_enabled(rx_ring);
2981 if (rx_buf_len < IXGBE_FCOE_JUMBO_FRAME_SIZE)
2982 rx_ring->rx_buf_len =
2983 IXGBE_FCOE_JUMBO_FRAME_SIZE;
2984 } else if (!ring_is_rsc_enabled(rx_ring) &&
2985 !ring_is_ps_enabled(rx_ring)) {
2986 rx_ring->rx_buf_len =
2987 IXGBE_FCOE_JUMBO_FRAME_SIZE;
2990 #endif /* IXGBE_FCOE */
2994 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
2996 struct ixgbe_hw *hw = &adapter->hw;
2997 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2999 switch (hw->mac.type) {
3000 case ixgbe_mac_82598EB:
3002 * For VMDq support of different descriptor types or
3003 * buffer sizes through the use of multiple SRRCTL
3004 * registers, RDRXCTL.MVMEN must be set to 1
3006 * also, the manual doesn't mention it clearly but DCA hints
3007 * will only use queue 0's tags unless this bit is set. Side
3008 * effects of setting this bit are only that SRRCTL must be
3009 * fully programmed [0..15]
3011 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
3013 case ixgbe_mac_82599EB:
3014 case ixgbe_mac_X540:
3015 /* Disable RSC for ACK packets */
3016 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
3017 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
3018 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3019 /* hardware requires some bits to be set by default */
3020 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3021 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3024 /* We should do nothing since we don't know this hardware */
3028 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3032 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3033 * @adapter: board private structure
3035 * Configure the Rx unit of the MAC after a reset.
3037 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3039 struct ixgbe_hw *hw = &adapter->hw;
3043 /* disable receives while setting up the descriptors */
3044 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3045 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3047 ixgbe_setup_psrtype(adapter);
3048 ixgbe_setup_rdrxctl(adapter);
3050 /* Program registers for the distribution of queues */
3051 ixgbe_setup_mrqc(adapter);
3053 ixgbe_set_uta(adapter);
3055 /* set_rx_buffer_len must be called before ring initialization */
3056 ixgbe_set_rx_buffer_len(adapter);
3059 * Setup the HW Rx Head and Tail Descriptor Pointers and
3060 * the Base and Length of the Rx Descriptor Ring
3062 for (i = 0; i < adapter->num_rx_queues; i++)
3063 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3065 /* disable drop enable for 82598 parts */
3066 if (hw->mac.type == ixgbe_mac_82598EB)
3067 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3069 /* enable all receives */
3070 rxctrl |= IXGBE_RXCTRL_RXEN;
3071 hw->mac.ops.enable_rx_dma(hw, rxctrl);
3074 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3076 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3077 struct ixgbe_hw *hw = &adapter->hw;
3078 int pool_ndx = adapter->num_vfs;
3080 /* add VID to filter table */
3081 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, true);
3082 set_bit(vid, adapter->active_vlans);
3085 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3087 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3088 struct ixgbe_hw *hw = &adapter->hw;
3089 int pool_ndx = adapter->num_vfs;
3091 /* remove VID from filter table */
3092 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, false);
3093 clear_bit(vid, adapter->active_vlans);
3097 * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3098 * @adapter: driver data
3100 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
3102 struct ixgbe_hw *hw = &adapter->hw;
3105 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3106 vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
3107 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3111 * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3112 * @adapter: driver data
3114 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
3116 struct ixgbe_hw *hw = &adapter->hw;
3119 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3120 vlnctrl |= IXGBE_VLNCTRL_VFE;
3121 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3122 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3126 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3127 * @adapter: driver data
3129 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3131 struct ixgbe_hw *hw = &adapter->hw;
3135 switch (hw->mac.type) {
3136 case ixgbe_mac_82598EB:
3137 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3138 vlnctrl &= ~IXGBE_VLNCTRL_VME;
3139 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3141 case ixgbe_mac_82599EB:
3142 case ixgbe_mac_X540:
3143 for (i = 0; i < adapter->num_rx_queues; i++) {
3144 j = adapter->rx_ring[i]->reg_idx;
3145 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3146 vlnctrl &= ~IXGBE_RXDCTL_VME;
3147 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3156 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3157 * @adapter: driver data
3159 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3161 struct ixgbe_hw *hw = &adapter->hw;
3165 switch (hw->mac.type) {
3166 case ixgbe_mac_82598EB:
3167 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3168 vlnctrl |= IXGBE_VLNCTRL_VME;
3169 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3171 case ixgbe_mac_82599EB:
3172 case ixgbe_mac_X540:
3173 for (i = 0; i < adapter->num_rx_queues; i++) {
3174 j = adapter->rx_ring[i]->reg_idx;
3175 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3176 vlnctrl |= IXGBE_RXDCTL_VME;
3177 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3185 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3189 ixgbe_vlan_rx_add_vid(adapter->netdev, 0);
3191 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3192 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
3196 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3197 * @netdev: network interface device structure
3199 * Writes unicast address list to the RAR table.
3200 * Returns: -ENOMEM on failure/insufficient address space
3201 * 0 on no addresses written
3202 * X on writing X addresses to the RAR table
3204 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3206 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3207 struct ixgbe_hw *hw = &adapter->hw;
3208 unsigned int vfn = adapter->num_vfs;
3209 unsigned int rar_entries = IXGBE_MAX_PF_MACVLANS;
3212 /* return ENOMEM indicating insufficient memory for addresses */
3213 if (netdev_uc_count(netdev) > rar_entries)
3216 if (!netdev_uc_empty(netdev) && rar_entries) {
3217 struct netdev_hw_addr *ha;
3218 /* return error if we do not support writing to RAR table */
3219 if (!hw->mac.ops.set_rar)
3222 netdev_for_each_uc_addr(ha, netdev) {
3225 hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3230 /* write the addresses in reverse order to avoid write combining */
3231 for (; rar_entries > 0 ; rar_entries--)
3232 hw->mac.ops.clear_rar(hw, rar_entries);
3238 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3239 * @netdev: network interface device structure
3241 * The set_rx_method entry point is called whenever the unicast/multicast
3242 * address list or the network interface flags are updated. This routine is
3243 * responsible for configuring the hardware for proper unicast, multicast and
3246 void ixgbe_set_rx_mode(struct net_device *netdev)
3248 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3249 struct ixgbe_hw *hw = &adapter->hw;
3250 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3253 /* Check for Promiscuous and All Multicast modes */
3255 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3257 /* set all bits that we expect to always be set */
3258 fctrl |= IXGBE_FCTRL_BAM;
3259 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3260 fctrl |= IXGBE_FCTRL_PMCF;
3262 /* clear the bits we are changing the status of */
3263 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3265 if (netdev->flags & IFF_PROMISC) {
3266 hw->addr_ctrl.user_set_promisc = true;
3267 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3268 vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3269 /* don't hardware filter vlans in promisc mode */
3270 ixgbe_vlan_filter_disable(adapter);
3272 if (netdev->flags & IFF_ALLMULTI) {
3273 fctrl |= IXGBE_FCTRL_MPE;
3274 vmolr |= IXGBE_VMOLR_MPE;
3277 * Write addresses to the MTA, if the attempt fails
3278 * then we should just turn on promiscuous mode so
3279 * that we can at least receive multicast traffic
3281 hw->mac.ops.update_mc_addr_list(hw, netdev);
3282 vmolr |= IXGBE_VMOLR_ROMPE;
3284 ixgbe_vlan_filter_enable(adapter);
3285 hw->addr_ctrl.user_set_promisc = false;
3287 * Write addresses to available RAR registers, if there is not
3288 * sufficient space to store all the addresses then enable
3289 * unicast promiscuous mode
3291 count = ixgbe_write_uc_addr_list(netdev);
3293 fctrl |= IXGBE_FCTRL_UPE;
3294 vmolr |= IXGBE_VMOLR_ROPE;
3298 if (adapter->num_vfs) {
3299 ixgbe_restore_vf_multicasts(adapter);
3300 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(adapter->num_vfs)) &
3301 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3303 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(adapter->num_vfs), vmolr);
3306 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3308 if (netdev->features & NETIF_F_HW_VLAN_RX)
3309 ixgbe_vlan_strip_enable(adapter);
3311 ixgbe_vlan_strip_disable(adapter);
3314 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3317 struct ixgbe_q_vector *q_vector;
3318 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3320 /* legacy and MSI only use one vector */
3321 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3324 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3325 q_vector = adapter->q_vector[q_idx];
3326 napi_enable(&q_vector->napi);
3330 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3333 struct ixgbe_q_vector *q_vector;
3334 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3336 /* legacy and MSI only use one vector */
3337 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3340 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3341 q_vector = adapter->q_vector[q_idx];
3342 napi_disable(&q_vector->napi);
3346 #ifdef CONFIG_IXGBE_DCB
3348 * ixgbe_configure_dcb - Configure DCB hardware
3349 * @adapter: ixgbe adapter struct
3351 * This is called by the driver on open to configure the DCB hardware.
3352 * This is also called by the gennetlink interface when reconfiguring
3355 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3357 struct ixgbe_hw *hw = &adapter->hw;
3358 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3360 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3361 if (hw->mac.type == ixgbe_mac_82598EB)
3362 netif_set_gso_max_size(adapter->netdev, 65536);
3366 if (hw->mac.type == ixgbe_mac_82598EB)
3367 netif_set_gso_max_size(adapter->netdev, 32768);
3370 /* Enable VLAN tag insert/strip */
3371 adapter->netdev->features |= NETIF_F_HW_VLAN_RX;
3373 hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
3375 /* reconfigure the hardware */
3376 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
3378 if (adapter->netdev->features & NETIF_F_FCOE_MTU)
3379 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
3381 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3383 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3385 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
3387 struct net_device *dev = adapter->netdev;
3389 if (adapter->ixgbe_ieee_ets)
3390 dev->dcbnl_ops->ieee_setets(dev,
3391 adapter->ixgbe_ieee_ets);
3392 if (adapter->ixgbe_ieee_pfc)
3393 dev->dcbnl_ops->ieee_setpfc(dev,
3394 adapter->ixgbe_ieee_pfc);
3397 /* Enable RSS Hash per TC */
3398 if (hw->mac.type != ixgbe_mac_82598EB) {
3402 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
3404 u8 cnt = adapter->netdev->tc_to_txq[i].count;
3409 reg |= msb << IXGBE_RQTC_SHIFT_TC(i);
3411 IXGBE_WRITE_REG(hw, IXGBE_RQTC, reg);
3417 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
3420 int num_tc = netdev_get_num_tc(adapter->netdev);
3421 struct ixgbe_hw *hw = &adapter->hw;
3423 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3424 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3425 hdrm = 64 << adapter->fdir_pballoc;
3427 hw->mac.ops.set_rxpba(&adapter->hw, num_tc, hdrm, PBA_STRATEGY_EQUAL);
3430 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
3432 struct ixgbe_hw *hw = &adapter->hw;
3433 struct hlist_node *node, *node2;
3434 struct ixgbe_fdir_filter *filter;
3436 spin_lock(&adapter->fdir_perfect_lock);
3438 if (!hlist_empty(&adapter->fdir_filter_list))
3439 ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
3441 hlist_for_each_entry_safe(filter, node, node2,
3442 &adapter->fdir_filter_list, fdir_node) {
3443 ixgbe_fdir_write_perfect_filter_82599(hw,
3446 (filter->action == IXGBE_FDIR_DROP_QUEUE) ?
3447 IXGBE_FDIR_DROP_QUEUE :
3448 adapter->rx_ring[filter->action]->reg_idx);
3451 spin_unlock(&adapter->fdir_perfect_lock);
3454 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3456 struct net_device *netdev = adapter->netdev;
3457 struct ixgbe_hw *hw = &adapter->hw;
3460 ixgbe_configure_pb(adapter);
3461 #ifdef CONFIG_IXGBE_DCB
3462 ixgbe_configure_dcb(adapter);
3465 ixgbe_set_rx_mode(netdev);
3466 ixgbe_restore_vlan(adapter);
3469 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
3470 ixgbe_configure_fcoe(adapter);
3472 #endif /* IXGBE_FCOE */
3473 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3474 for (i = 0; i < adapter->num_tx_queues; i++)
3475 adapter->tx_ring[i]->atr_sample_rate =
3476 adapter->atr_sample_rate;
3477 ixgbe_init_fdir_signature_82599(hw, adapter->fdir_pballoc);
3478 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3479 ixgbe_init_fdir_perfect_82599(&adapter->hw,
3480 adapter->fdir_pballoc);
3481 ixgbe_fdir_filter_restore(adapter);
3483 ixgbe_configure_virtualization(adapter);
3485 ixgbe_configure_tx(adapter);
3486 ixgbe_configure_rx(adapter);
3489 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3491 switch (hw->phy.type) {
3492 case ixgbe_phy_sfp_avago:
3493 case ixgbe_phy_sfp_ftl:
3494 case ixgbe_phy_sfp_intel:
3495 case ixgbe_phy_sfp_unknown:
3496 case ixgbe_phy_sfp_passive_tyco:
3497 case ixgbe_phy_sfp_passive_unknown:
3498 case ixgbe_phy_sfp_active_unknown:
3499 case ixgbe_phy_sfp_ftl_active:
3507 * ixgbe_sfp_link_config - set up SFP+ link
3508 * @adapter: pointer to private adapter struct
3510 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3513 * We are assuming the worst case scenerio here, and that
3514 * is that an SFP was inserted/removed after the reset
3515 * but before SFP detection was enabled. As such the best
3516 * solution is to just start searching as soon as we start
3518 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3519 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
3521 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
3525 * ixgbe_non_sfp_link_config - set up non-SFP+ link
3526 * @hw: pointer to private hardware struct
3528 * Returns 0 on success, negative on failure
3530 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3533 bool negotiation, link_up = false;
3534 u32 ret = IXGBE_ERR_LINK_SETUP;
3536 if (hw->mac.ops.check_link)
3537 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3542 autoneg = hw->phy.autoneg_advertised;
3543 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
3544 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
3549 if (hw->mac.ops.setup_link)
3550 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
3555 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
3557 struct ixgbe_hw *hw = &adapter->hw;
3560 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3561 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
3563 gpie |= IXGBE_GPIE_EIAME;
3565 * use EIAM to auto-mask when MSI-X interrupt is asserted
3566 * this saves a register write for every interrupt
3568 switch (hw->mac.type) {
3569 case ixgbe_mac_82598EB:
3570 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3572 case ixgbe_mac_82599EB:
3573 case ixgbe_mac_X540:
3575 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
3576 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
3580 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
3581 * specifically only auto mask tx and rx interrupts */
3582 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3585 /* XXX: to interrupt immediately for EICS writes, enable this */
3586 /* gpie |= IXGBE_GPIE_EIMEN; */
3588 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3589 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
3590 gpie |= IXGBE_GPIE_VTMODE_64;
3593 /* Enable fan failure interrupt */
3594 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3595 gpie |= IXGBE_SDP1_GPIEN;
3597 if (hw->mac.type == ixgbe_mac_82599EB) {
3598 gpie |= IXGBE_SDP1_GPIEN;
3599 gpie |= IXGBE_SDP2_GPIEN;
3602 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3605 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
3607 struct ixgbe_hw *hw = &adapter->hw;
3611 ixgbe_get_hw_control(adapter);
3612 ixgbe_setup_gpie(adapter);
3614 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3615 ixgbe_configure_msix(adapter);
3617 ixgbe_configure_msi_and_legacy(adapter);
3619 /* enable the optics for both mult-speed fiber and 82599 SFP+ fiber */
3620 if (hw->mac.ops.enable_tx_laser &&
3621 ((hw->phy.multispeed_fiber) ||
3622 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
3623 (hw->mac.type == ixgbe_mac_82599EB))))
3624 hw->mac.ops.enable_tx_laser(hw);
3626 clear_bit(__IXGBE_DOWN, &adapter->state);
3627 ixgbe_napi_enable_all(adapter);
3629 if (ixgbe_is_sfp(hw)) {
3630 ixgbe_sfp_link_config(adapter);
3632 err = ixgbe_non_sfp_link_config(hw);
3634 e_err(probe, "link_config FAILED %d\n", err);
3637 /* clear any pending interrupts, may auto mask */
3638 IXGBE_READ_REG(hw, IXGBE_EICR);
3639 ixgbe_irq_enable(adapter, true, true);
3642 * If this adapter has a fan, check to see if we had a failure
3643 * before we enabled the interrupt.
3645 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3646 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
3647 if (esdp & IXGBE_ESDP_SDP1)
3648 e_crit(drv, "Fan has stopped, replace the adapter\n");
3651 /* enable transmits */
3652 netif_tx_start_all_queues(adapter->netdev);
3654 /* bring the link up in the watchdog, this could race with our first
3655 * link up interrupt but shouldn't be a problem */
3656 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3657 adapter->link_check_timeout = jiffies;
3658 mod_timer(&adapter->service_timer, jiffies);
3660 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
3661 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
3662 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
3663 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
3668 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
3670 WARN_ON(in_interrupt());
3671 /* put off any impending NetWatchDogTimeout */
3672 adapter->netdev->trans_start = jiffies;
3674 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
3675 usleep_range(1000, 2000);
3676 ixgbe_down(adapter);
3678 * If SR-IOV enabled then wait a bit before bringing the adapter
3679 * back up to give the VFs time to respond to the reset. The
3680 * two second wait is based upon the watchdog timer cycle in
3683 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3686 clear_bit(__IXGBE_RESETTING, &adapter->state);
3689 int ixgbe_up(struct ixgbe_adapter *adapter)
3691 /* hardware has been reset, we need to reload some things */
3692 ixgbe_configure(adapter);
3694 return ixgbe_up_complete(adapter);
3697 void ixgbe_reset(struct ixgbe_adapter *adapter)
3699 struct ixgbe_hw *hw = &adapter->hw;
3702 /* lock SFP init bit to prevent race conditions with the watchdog */
3703 while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
3704 usleep_range(1000, 2000);
3706 /* clear all SFP and link config related flags while holding SFP_INIT */
3707 adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
3708 IXGBE_FLAG2_SFP_NEEDS_RESET);
3709 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
3711 err = hw->mac.ops.init_hw(hw);
3714 case IXGBE_ERR_SFP_NOT_PRESENT:
3715 case IXGBE_ERR_SFP_NOT_SUPPORTED:
3717 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
3718 e_dev_err("master disable timed out\n");
3720 case IXGBE_ERR_EEPROM_VERSION:
3721 /* We are running on a pre-production device, log a warning */
3722 e_dev_warn("This device is a pre-production adapter/LOM. "
3723 "Please be aware there may be issuesassociated with "
3724 "your hardware. If you are experiencing problems "
3725 "please contact your Intel or hardware "
3726 "representative who provided you with this "
3730 e_dev_err("Hardware Error: %d\n", err);
3733 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
3735 /* reprogram the RAR[0] in case user changed it. */
3736 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
3741 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
3742 * @rx_ring: ring to free buffers from
3744 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
3746 struct device *dev = rx_ring->dev;
3750 /* ring already cleared, nothing to do */
3751 if (!rx_ring->rx_buffer_info)
3754 /* Free all the Rx ring sk_buffs */
3755 for (i = 0; i < rx_ring->count; i++) {
3756 struct ixgbe_rx_buffer *rx_buffer_info;
3758 rx_buffer_info = &rx_ring->rx_buffer_info[i];
3759 if (rx_buffer_info->dma) {
3760 dma_unmap_single(rx_ring->dev, rx_buffer_info->dma,
3761 rx_ring->rx_buf_len,
3763 rx_buffer_info->dma = 0;
3765 if (rx_buffer_info->skb) {
3766 struct sk_buff *skb = rx_buffer_info->skb;
3767 rx_buffer_info->skb = NULL;
3769 struct sk_buff *this = skb;
3770 if (IXGBE_RSC_CB(this)->delay_unmap) {
3771 dma_unmap_single(dev,
3772 IXGBE_RSC_CB(this)->dma,
3773 rx_ring->rx_buf_len,
3775 IXGBE_RSC_CB(this)->dma = 0;
3776 IXGBE_RSC_CB(skb)->delay_unmap = false;
3779 dev_kfree_skb(this);
3782 if (!rx_buffer_info->page)
3784 if (rx_buffer_info->page_dma) {
3785 dma_unmap_page(dev, rx_buffer_info->page_dma,
3786 PAGE_SIZE / 2, DMA_FROM_DEVICE);
3787 rx_buffer_info->page_dma = 0;
3789 put_page(rx_buffer_info->page);
3790 rx_buffer_info->page = NULL;
3791 rx_buffer_info->page_offset = 0;
3794 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
3795 memset(rx_ring->rx_buffer_info, 0, size);
3797 /* Zero out the descriptor ring */
3798 memset(rx_ring->desc, 0, rx_ring->size);
3800 rx_ring->next_to_clean = 0;
3801 rx_ring->next_to_use = 0;
3805 * ixgbe_clean_tx_ring - Free Tx Buffers
3806 * @tx_ring: ring to be cleaned
3808 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
3810 struct ixgbe_tx_buffer *tx_buffer_info;
3814 /* ring already cleared, nothing to do */
3815 if (!tx_ring->tx_buffer_info)
3818 /* Free all the Tx ring sk_buffs */
3819 for (i = 0; i < tx_ring->count; i++) {
3820 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3821 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
3824 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
3825 memset(tx_ring->tx_buffer_info, 0, size);
3827 /* Zero out the descriptor ring */
3828 memset(tx_ring->desc, 0, tx_ring->size);
3830 tx_ring->next_to_use = 0;
3831 tx_ring->next_to_clean = 0;
3835 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
3836 * @adapter: board private structure
3838 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
3842 for (i = 0; i < adapter->num_rx_queues; i++)
3843 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
3847 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
3848 * @adapter: board private structure
3850 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
3854 for (i = 0; i < adapter->num_tx_queues; i++)
3855 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
3858 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
3860 struct hlist_node *node, *node2;
3861 struct ixgbe_fdir_filter *filter;
3863 spin_lock(&adapter->fdir_perfect_lock);
3865 hlist_for_each_entry_safe(filter, node, node2,
3866 &adapter->fdir_filter_list, fdir_node) {
3867 hlist_del(&filter->fdir_node);
3870 adapter->fdir_filter_count = 0;
3872 spin_unlock(&adapter->fdir_perfect_lock);
3875 void ixgbe_down(struct ixgbe_adapter *adapter)
3877 struct net_device *netdev = adapter->netdev;
3878 struct ixgbe_hw *hw = &adapter->hw;
3882 /* signal that we are down to the interrupt handler */
3883 set_bit(__IXGBE_DOWN, &adapter->state);
3885 /* disable receives */
3886 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3887 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3889 /* disable all enabled rx queues */
3890 for (i = 0; i < adapter->num_rx_queues; i++)
3891 /* this call also flushes the previous write */
3892 ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]);
3894 usleep_range(10000, 20000);
3896 netif_tx_stop_all_queues(netdev);
3898 /* call carrier off first to avoid false dev_watchdog timeouts */
3899 netif_carrier_off(netdev);
3900 netif_tx_disable(netdev);
3902 ixgbe_irq_disable(adapter);
3904 ixgbe_napi_disable_all(adapter);
3906 adapter->flags2 &= ~(IXGBE_FLAG2_FDIR_REQUIRES_REINIT |
3907 IXGBE_FLAG2_RESET_REQUESTED);
3908 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
3910 del_timer_sync(&adapter->service_timer);
3912 /* disable receive for all VFs and wait one second */
3913 if (adapter->num_vfs) {
3914 /* ping all the active vfs to let them know we are going down */
3915 ixgbe_ping_all_vfs(adapter);
3917 /* Disable all VFTE/VFRE TX/RX */
3918 ixgbe_disable_tx_rx(adapter);
3920 /* Mark all the VFs as inactive */
3921 for (i = 0 ; i < adapter->num_vfs; i++)
3922 adapter->vfinfo[i].clear_to_send = 0;
3925 /* disable transmits in the hardware now that interrupts are off */
3926 for (i = 0; i < adapter->num_tx_queues; i++) {
3927 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
3928 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
3931 /* Disable the Tx DMA engine on 82599 and X540 */
3932 switch (hw->mac.type) {
3933 case ixgbe_mac_82599EB:
3934 case ixgbe_mac_X540:
3935 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
3936 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
3937 ~IXGBE_DMATXCTL_TE));
3943 if (!pci_channel_offline(adapter->pdev))
3944 ixgbe_reset(adapter);
3946 /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
3947 if (hw->mac.ops.disable_tx_laser &&
3948 ((hw->phy.multispeed_fiber) ||
3949 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
3950 (hw->mac.type == ixgbe_mac_82599EB))))
3951 hw->mac.ops.disable_tx_laser(hw);
3953 ixgbe_clean_all_tx_rings(adapter);
3954 ixgbe_clean_all_rx_rings(adapter);
3956 #ifdef CONFIG_IXGBE_DCA
3957 /* since we reset the hardware DCA settings were cleared */
3958 ixgbe_setup_dca(adapter);
3963 * ixgbe_poll - NAPI Rx polling callback
3964 * @napi: structure for representing this polling device
3965 * @budget: how many packets driver is allowed to clean
3967 * This function is used for legacy and MSI, NAPI mode
3969 static int ixgbe_poll(struct napi_struct *napi, int budget)
3971 struct ixgbe_q_vector *q_vector =
3972 container_of(napi, struct ixgbe_q_vector, napi);
3973 struct ixgbe_adapter *adapter = q_vector->adapter;
3974 struct ixgbe_ring *ring;
3975 int per_ring_budget;
3976 bool clean_complete = true;
3978 #ifdef CONFIG_IXGBE_DCA
3979 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
3980 ixgbe_update_dca(q_vector);
3983 for (ring = q_vector->tx.ring; ring != NULL; ring = ring->next)
3984 clean_complete &= !!ixgbe_clean_tx_irq(q_vector, ring);
3986 /* attempt to distribute budget to each queue fairly, but don't allow
3987 * the budget to go below 1 because we'll exit polling */
3988 if (q_vector->rx.count > 1)
3989 per_ring_budget = max(budget/q_vector->rx.count, 1);
3991 per_ring_budget = budget;
3993 for (ring = q_vector->rx.ring; ring != NULL; ring = ring->next)
3994 clean_complete &= ixgbe_clean_rx_irq(q_vector, ring,
3997 /* If all work not completed, return budget and keep polling */
3998 if (!clean_complete)
4001 /* all work done, exit the polling mode */
4002 napi_complete(napi);
4003 if (adapter->rx_itr_setting & 1)
4004 ixgbe_set_itr(q_vector);
4005 if (!test_bit(__IXGBE_DOWN, &adapter->state))
4006 ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
4012 * ixgbe_tx_timeout - Respond to a Tx Hang
4013 * @netdev: network interface device structure
4015 static void ixgbe_tx_timeout(struct net_device *netdev)
4017 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4019 /* Do the reset outside of interrupt context */
4020 ixgbe_tx_timeout_reset(adapter);
4024 * ixgbe_set_rss_queues: Allocate queues for RSS
4025 * @adapter: board private structure to initialize
4027 * This is our "base" multiqueue mode. RSS (Receive Side Scaling) will try
4028 * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
4031 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
4034 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
4036 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4038 adapter->num_rx_queues = f->indices;
4039 adapter->num_tx_queues = f->indices;
4049 * ixgbe_set_fdir_queues: Allocate queues for Flow Director
4050 * @adapter: board private structure to initialize
4052 * Flow Director is an advanced Rx filter, attempting to get Rx flows back
4053 * to the original CPU that initiated the Tx session. This runs in addition
4054 * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
4055 * Rx load across CPUs using RSS.
4058 static inline bool ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
4061 struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
4063 f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
4066 /* Flow Director must have RSS enabled */
4067 if ((adapter->flags & IXGBE_FLAG_RSS_ENABLED) &&
4068 (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)) {
4069 adapter->num_tx_queues = f_fdir->indices;
4070 adapter->num_rx_queues = f_fdir->indices;
4073 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4080 * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
4081 * @adapter: board private structure to initialize
4083 * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
4084 * The ring feature mask is not used as a mask for FCoE, as it can take any 8
4085 * rx queues out of the max number of rx queues, instead, it is used as the
4086 * index of the first rx queue used by FCoE.
4089 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
4091 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4093 if (!(adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
4096 f->indices = min((int)num_online_cpus(), f->indices);
4098 adapter->num_rx_queues = 1;
4099 adapter->num_tx_queues = 1;
4101 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4102 e_info(probe, "FCoE enabled with RSS\n");
4103 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)
4104 ixgbe_set_fdir_queues(adapter);
4106 ixgbe_set_rss_queues(adapter);
4109 /* adding FCoE rx rings to the end */
4110 f->mask = adapter->num_rx_queues;
4111 adapter->num_rx_queues += f->indices;
4112 adapter->num_tx_queues += f->indices;
4116 #endif /* IXGBE_FCOE */
4118 /* Artificial max queue cap per traffic class in DCB mode */
4119 #define DCB_QUEUE_CAP 8
4121 #ifdef CONFIG_IXGBE_DCB
4122 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
4124 int per_tc_q, q, i, offset = 0;
4125 struct net_device *dev = adapter->netdev;
4126 int tcs = netdev_get_num_tc(dev);
4131 /* Map queue offset and counts onto allocated tx queues */
4132 per_tc_q = min(dev->num_tx_queues / tcs, (unsigned int)DCB_QUEUE_CAP);
4133 q = min((int)num_online_cpus(), per_tc_q);
4135 for (i = 0; i < tcs; i++) {
4136 netdev_set_prio_tc_map(dev, i, i);
4137 netdev_set_tc_queue(dev, i, q, offset);
4141 adapter->num_tx_queues = q * tcs;
4142 adapter->num_rx_queues = q * tcs;
4145 /* FCoE enabled queues require special configuration indexed
4146 * by feature specific indices and mask. Here we map FCoE
4147 * indices onto the DCB queue pairs allowing FCoE to own
4148 * configuration later.
4150 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
4152 struct ixgbe_ring_feature *f =
4153 &adapter->ring_feature[RING_F_FCOE];
4155 tc = netdev_get_prio_tc_map(dev, adapter->fcoe.up);
4156 f->indices = dev->tc_to_txq[tc].count;
4157 f->mask = dev->tc_to_txq[tc].offset;
4166 * ixgbe_set_sriov_queues: Allocate queues for IOV use
4167 * @adapter: board private structure to initialize
4169 * IOV doesn't actually use anything, so just NAK the
4170 * request for now and let the other queue routines
4171 * figure out what to do.
4173 static inline bool ixgbe_set_sriov_queues(struct ixgbe_adapter *adapter)
4179 * ixgbe_set_num_queues: Allocate queues for device, feature dependent
4180 * @adapter: board private structure to initialize
4182 * This is the top level queue allocation routine. The order here is very
4183 * important, starting with the "most" number of features turned on at once,
4184 * and ending with the smallest set of features. This way large combinations
4185 * can be allocated if they're turned on, and smaller combinations are the
4186 * fallthrough conditions.
4189 static int ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
4191 /* Start with base case */
4192 adapter->num_rx_queues = 1;
4193 adapter->num_tx_queues = 1;
4194 adapter->num_rx_pools = adapter->num_rx_queues;
4195 adapter->num_rx_queues_per_pool = 1;
4197 if (ixgbe_set_sriov_queues(adapter))
4200 #ifdef CONFIG_IXGBE_DCB
4201 if (ixgbe_set_dcb_queues(adapter))
4206 if (ixgbe_set_fcoe_queues(adapter))
4209 #endif /* IXGBE_FCOE */
4210 if (ixgbe_set_fdir_queues(adapter))
4213 if (ixgbe_set_rss_queues(adapter))
4216 /* fallback to base case */
4217 adapter->num_rx_queues = 1;
4218 adapter->num_tx_queues = 1;
4221 /* Notify the stack of the (possibly) reduced queue counts. */
4222 netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
4223 return netif_set_real_num_rx_queues(adapter->netdev,
4224 adapter->num_rx_queues);
4227 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
4230 int err, vector_threshold;
4232 /* We'll want at least 3 (vector_threshold):
4235 * 3) Other (Link Status Change, etc.)
4236 * 4) TCP Timer (optional)
4238 vector_threshold = MIN_MSIX_COUNT;
4240 /* The more we get, the more we will assign to Tx/Rx Cleanup
4241 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
4242 * Right now, we simply care about how many we'll get; we'll
4243 * set them up later while requesting irq's.
4245 while (vectors >= vector_threshold) {
4246 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
4248 if (!err) /* Success in acquiring all requested vectors. */
4251 vectors = 0; /* Nasty failure, quit now */
4252 else /* err == number of vectors we should try again with */
4256 if (vectors < vector_threshold) {
4257 /* Can't allocate enough MSI-X interrupts? Oh well.
4258 * This just means we'll go with either a single MSI
4259 * vector or fall back to legacy interrupts.
4261 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4262 "Unable to allocate MSI-X interrupts\n");
4263 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4264 kfree(adapter->msix_entries);
4265 adapter->msix_entries = NULL;
4267 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
4269 * Adjust for only the vectors we'll use, which is minimum
4270 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
4271 * vectors we were allocated.
4273 adapter->num_msix_vectors = min(vectors,
4274 adapter->max_msix_q_vectors + NON_Q_VECTORS);
4279 * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
4280 * @adapter: board private structure to initialize
4282 * Cache the descriptor ring offsets for RSS to the assigned rings.
4285 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
4289 if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED))
4292 for (i = 0; i < adapter->num_rx_queues; i++)
4293 adapter->rx_ring[i]->reg_idx = i;
4294 for (i = 0; i < adapter->num_tx_queues; i++)
4295 adapter->tx_ring[i]->reg_idx = i;
4300 #ifdef CONFIG_IXGBE_DCB
4302 /* ixgbe_get_first_reg_idx - Return first register index associated with ring */
4303 static void ixgbe_get_first_reg_idx(struct ixgbe_adapter *adapter, u8 tc,
4304 unsigned int *tx, unsigned int *rx)
4306 struct net_device *dev = adapter->netdev;
4307 struct ixgbe_hw *hw = &adapter->hw;
4308 u8 num_tcs = netdev_get_num_tc(dev);
4313 switch (hw->mac.type) {
4314 case ixgbe_mac_82598EB:
4318 case ixgbe_mac_82599EB:
4319 case ixgbe_mac_X540:
4324 } else if (tc < 5) {
4325 *tx = ((tc + 2) << 4);
4327 } else if (tc < num_tcs) {
4328 *tx = ((tc + 8) << 3);
4357 * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
4358 * @adapter: board private structure to initialize
4360 * Cache the descriptor ring offsets for DCB to the assigned rings.
4363 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
4365 struct net_device *dev = adapter->netdev;
4367 u8 num_tcs = netdev_get_num_tc(dev);
4372 for (i = 0, k = 0; i < num_tcs; i++) {
4373 unsigned int tx_s, rx_s;
4374 u16 count = dev->tc_to_txq[i].count;
4376 ixgbe_get_first_reg_idx(adapter, i, &tx_s, &rx_s);
4377 for (j = 0; j < count; j++, k++) {
4378 adapter->tx_ring[k]->reg_idx = tx_s + j;
4379 adapter->rx_ring[k]->reg_idx = rx_s + j;
4380 adapter->tx_ring[k]->dcb_tc = i;
4381 adapter->rx_ring[k]->dcb_tc = i;
4390 * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
4391 * @adapter: board private structure to initialize
4393 * Cache the descriptor ring offsets for Flow Director to the assigned rings.
4396 static inline bool ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
4401 if ((adapter->flags & IXGBE_FLAG_RSS_ENABLED) &&
4402 (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)) {
4403 for (i = 0; i < adapter->num_rx_queues; i++)
4404 adapter->rx_ring[i]->reg_idx = i;
4405 for (i = 0; i < adapter->num_tx_queues; i++)
4406 adapter->tx_ring[i]->reg_idx = i;
4415 * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
4416 * @adapter: board private structure to initialize
4418 * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
4421 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
4423 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4425 u8 fcoe_rx_i = 0, fcoe_tx_i = 0;
4427 if (!(adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
4430 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4431 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)
4432 ixgbe_cache_ring_fdir(adapter);
4434 ixgbe_cache_ring_rss(adapter);
4436 fcoe_rx_i = f->mask;
4437 fcoe_tx_i = f->mask;
4439 for (i = 0; i < f->indices; i++, fcoe_rx_i++, fcoe_tx_i++) {
4440 adapter->rx_ring[f->mask + i]->reg_idx = fcoe_rx_i;
4441 adapter->tx_ring[f->mask + i]->reg_idx = fcoe_tx_i;
4446 #endif /* IXGBE_FCOE */
4448 * ixgbe_cache_ring_sriov - Descriptor ring to register mapping for sriov
4449 * @adapter: board private structure to initialize
4451 * SR-IOV doesn't use any descriptor rings but changes the default if
4452 * no other mapping is used.
4455 static inline bool ixgbe_cache_ring_sriov(struct ixgbe_adapter *adapter)
4457 adapter->rx_ring[0]->reg_idx = adapter->num_vfs * 2;
4458 adapter->tx_ring[0]->reg_idx = adapter->num_vfs * 2;
4459 if (adapter->num_vfs)
4466 * ixgbe_cache_ring_register - Descriptor ring to register mapping
4467 * @adapter: board private structure to initialize
4469 * Once we know the feature-set enabled for the device, we'll cache
4470 * the register offset the descriptor ring is assigned to.
4472 * Note, the order the various feature calls is important. It must start with
4473 * the "most" features enabled at the same time, then trickle down to the
4474 * least amount of features turned on at once.
4476 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
4478 /* start with default case */
4479 adapter->rx_ring[0]->reg_idx = 0;
4480 adapter->tx_ring[0]->reg_idx = 0;
4482 if (ixgbe_cache_ring_sriov(adapter))
4485 #ifdef CONFIG_IXGBE_DCB
4486 if (ixgbe_cache_ring_dcb(adapter))
4491 if (ixgbe_cache_ring_fcoe(adapter))
4493 #endif /* IXGBE_FCOE */
4495 if (ixgbe_cache_ring_fdir(adapter))
4498 if (ixgbe_cache_ring_rss(adapter))
4503 * ixgbe_alloc_queues - Allocate memory for all rings
4504 * @adapter: board private structure to initialize
4506 * We allocate one ring per queue at run-time since we don't know the
4507 * number of queues at compile-time. The polling_netdev array is
4508 * intended for Multiqueue, but should work fine with a single queue.
4510 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
4512 int rx = 0, tx = 0, nid = adapter->node;
4514 if (nid < 0 || !node_online(nid))
4515 nid = first_online_node;
4517 for (; tx < adapter->num_tx_queues; tx++) {
4518 struct ixgbe_ring *ring;
4520 ring = kzalloc_node(sizeof(*ring), GFP_KERNEL, nid);
4522 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
4524 goto err_allocation;
4525 ring->count = adapter->tx_ring_count;
4526 ring->queue_index = tx;
4527 ring->numa_node = nid;
4528 ring->dev = &adapter->pdev->dev;
4529 ring->netdev = adapter->netdev;
4531 adapter->tx_ring[tx] = ring;
4534 for (; rx < adapter->num_rx_queues; rx++) {
4535 struct ixgbe_ring *ring;
4537 ring = kzalloc_node(sizeof(*ring), GFP_KERNEL, nid);
4539 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
4541 goto err_allocation;
4542 ring->count = adapter->rx_ring_count;
4543 ring->queue_index = rx;
4544 ring->numa_node = nid;
4545 ring->dev = &adapter->pdev->dev;
4546 ring->netdev = adapter->netdev;
4548 adapter->rx_ring[rx] = ring;
4551 ixgbe_cache_ring_register(adapter);
4557 kfree(adapter->tx_ring[--tx]);
4560 kfree(adapter->rx_ring[--rx]);
4565 * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
4566 * @adapter: board private structure to initialize
4568 * Attempt to configure the interrupts using the best available
4569 * capabilities of the hardware and the kernel.
4571 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
4573 struct ixgbe_hw *hw = &adapter->hw;
4575 int vector, v_budget;
4578 * It's easy to be greedy for MSI-X vectors, but it really
4579 * doesn't do us much good if we have a lot more vectors
4580 * than CPU's. So let's be conservative and only ask for
4581 * (roughly) the same number of vectors as there are CPU's.
4583 v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
4584 (int)num_online_cpus()) + NON_Q_VECTORS;
4587 * At the same time, hardware can only support a maximum of
4588 * hw.mac->max_msix_vectors vectors. With features
4589 * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
4590 * descriptor queues supported by our device. Thus, we cap it off in
4591 * those rare cases where the cpu count also exceeds our vector limit.
4593 v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
4595 /* A failure in MSI-X entry allocation isn't fatal, but it does
4596 * mean we disable MSI-X capabilities of the adapter. */
4597 adapter->msix_entries = kcalloc(v_budget,
4598 sizeof(struct msix_entry), GFP_KERNEL);
4599 if (adapter->msix_entries) {
4600 for (vector = 0; vector < v_budget; vector++)
4601 adapter->msix_entries[vector].entry = vector;
4603 ixgbe_acquire_msix_vectors(adapter, v_budget);
4605 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4609 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
4610 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
4611 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
4613 "ATR is not supported while multiple "
4614 "queues are disabled. Disabling Flow Director\n");
4616 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4617 adapter->atr_sample_rate = 0;
4618 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4619 ixgbe_disable_sriov(adapter);
4621 err = ixgbe_set_num_queues(adapter);
4625 err = pci_enable_msi(adapter->pdev);
4627 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
4629 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4630 "Unable to allocate MSI interrupt, "
4631 "falling back to legacy. Error: %d\n", err);
4641 * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
4642 * @adapter: board private structure to initialize
4644 * We allocate one q_vector per queue interrupt. If allocation fails we
4647 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
4649 int v_idx, num_q_vectors;
4650 struct ixgbe_q_vector *q_vector;
4652 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4653 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4657 for (v_idx = 0; v_idx < num_q_vectors; v_idx++) {
4658 q_vector = kzalloc_node(sizeof(struct ixgbe_q_vector),
4659 GFP_KERNEL, adapter->node);
4661 q_vector = kzalloc(sizeof(struct ixgbe_q_vector),
4666 q_vector->adapter = adapter;
4667 q_vector->v_idx = v_idx;
4669 /* Allocate the affinity_hint cpumask, configure the mask */
4670 if (!alloc_cpumask_var(&q_vector->affinity_mask, GFP_KERNEL))
4672 cpumask_set_cpu(v_idx, q_vector->affinity_mask);
4674 if (q_vector->tx.count && !q_vector->rx.count)
4675 q_vector->eitr = adapter->tx_eitr_param;
4677 q_vector->eitr = adapter->rx_eitr_param;
4679 netif_napi_add(adapter->netdev, &q_vector->napi,
4681 adapter->q_vector[v_idx] = q_vector;
4689 q_vector = adapter->q_vector[v_idx];
4690 netif_napi_del(&q_vector->napi);
4691 free_cpumask_var(q_vector->affinity_mask);
4693 adapter->q_vector[v_idx] = NULL;
4699 * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
4700 * @adapter: board private structure to initialize
4702 * This function frees the memory allocated to the q_vectors. In addition if
4703 * NAPI is enabled it will delete any references to the NAPI struct prior
4704 * to freeing the q_vector.
4706 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
4708 int v_idx, num_q_vectors;
4710 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4711 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4715 for (v_idx = 0; v_idx < num_q_vectors; v_idx++) {
4716 struct ixgbe_q_vector *q_vector = adapter->q_vector[v_idx];
4717 adapter->q_vector[v_idx] = NULL;
4718 netif_napi_del(&q_vector->napi);
4719 free_cpumask_var(q_vector->affinity_mask);
4724 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
4726 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4727 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4728 pci_disable_msix(adapter->pdev);
4729 kfree(adapter->msix_entries);
4730 adapter->msix_entries = NULL;
4731 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
4732 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
4733 pci_disable_msi(adapter->pdev);
4738 * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
4739 * @adapter: board private structure to initialize
4741 * We determine which interrupt scheme to use based on...
4742 * - Kernel support (MSI, MSI-X)
4743 * - which can be user-defined (via MODULE_PARAM)
4744 * - Hardware queue count (num_*_queues)
4745 * - defined by miscellaneous hardware support/features (RSS, etc.)
4747 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
4751 /* Number of supported queues */
4752 err = ixgbe_set_num_queues(adapter);
4756 err = ixgbe_set_interrupt_capability(adapter);
4758 e_dev_err("Unable to setup interrupt capabilities\n");
4759 goto err_set_interrupt;
4762 err = ixgbe_alloc_q_vectors(adapter);
4764 e_dev_err("Unable to allocate memory for queue vectors\n");
4765 goto err_alloc_q_vectors;
4768 err = ixgbe_alloc_queues(adapter);
4770 e_dev_err("Unable to allocate memory for queues\n");
4771 goto err_alloc_queues;
4774 e_dev_info("Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
4775 (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
4776 adapter->num_rx_queues, adapter->num_tx_queues);
4778 set_bit(__IXGBE_DOWN, &adapter->state);
4783 ixgbe_free_q_vectors(adapter);
4784 err_alloc_q_vectors:
4785 ixgbe_reset_interrupt_capability(adapter);
4791 * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
4792 * @adapter: board private structure to clear interrupt scheme on
4794 * We go through and clear interrupt specific resources and reset the structure
4795 * to pre-load conditions
4797 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
4801 for (i = 0; i < adapter->num_tx_queues; i++) {
4802 kfree(adapter->tx_ring[i]);
4803 adapter->tx_ring[i] = NULL;
4805 for (i = 0; i < adapter->num_rx_queues; i++) {
4806 struct ixgbe_ring *ring = adapter->rx_ring[i];
4808 /* ixgbe_get_stats64() might access this ring, we must wait
4809 * a grace period before freeing it.
4811 kfree_rcu(ring, rcu);
4812 adapter->rx_ring[i] = NULL;
4815 adapter->num_tx_queues = 0;
4816 adapter->num_rx_queues = 0;
4818 ixgbe_free_q_vectors(adapter);
4819 ixgbe_reset_interrupt_capability(adapter);
4823 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4824 * @adapter: board private structure to initialize
4826 * ixgbe_sw_init initializes the Adapter private data structure.
4827 * Fields are initialized based on PCI device information and
4828 * OS network device settings (MTU size).
4830 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
4832 struct ixgbe_hw *hw = &adapter->hw;
4833 struct pci_dev *pdev = adapter->pdev;
4834 struct net_device *dev = adapter->netdev;
4836 #ifdef CONFIG_IXGBE_DCB
4838 struct tc_configuration *tc;
4840 int max_frame = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
4842 /* PCI config space info */
4844 hw->vendor_id = pdev->vendor;
4845 hw->device_id = pdev->device;
4846 hw->revision_id = pdev->revision;
4847 hw->subsystem_vendor_id = pdev->subsystem_vendor;
4848 hw->subsystem_device_id = pdev->subsystem_device;
4850 /* Set capability flags */
4851 rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
4852 adapter->ring_feature[RING_F_RSS].indices = rss;
4853 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
4854 switch (hw->mac.type) {
4855 case ixgbe_mac_82598EB:
4856 if (hw->device_id == IXGBE_DEV_ID_82598AT)
4857 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4858 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
4860 case ixgbe_mac_82599EB:
4861 case ixgbe_mac_X540:
4862 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
4863 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4864 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4865 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
4866 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4867 /* Flow Director hash filters enabled */
4868 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
4869 adapter->atr_sample_rate = 20;
4870 adapter->ring_feature[RING_F_FDIR].indices =
4871 IXGBE_MAX_FDIR_INDICES;
4872 adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
4874 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4875 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4876 adapter->ring_feature[RING_F_FCOE].indices = 0;
4877 #ifdef CONFIG_IXGBE_DCB
4878 /* Default traffic class to use for FCoE */
4879 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
4881 #endif /* IXGBE_FCOE */
4887 /* n-tuple support exists, always init our spinlock */
4888 spin_lock_init(&adapter->fdir_perfect_lock);
4890 #ifdef CONFIG_IXGBE_DCB
4891 /* Configure DCB traffic classes */
4892 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
4893 tc = &adapter->dcb_cfg.tc_config[j];
4894 tc->path[DCB_TX_CONFIG].bwg_id = 0;
4895 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
4896 tc->path[DCB_RX_CONFIG].bwg_id = 0;
4897 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
4898 tc->dcb_pfc = pfc_disabled;
4900 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
4901 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
4902 adapter->dcb_cfg.pfc_mode_enable = false;
4903 adapter->dcb_set_bitmap = 0x00;
4904 adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
4905 ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
4910 /* default flow control settings */
4911 hw->fc.requested_mode = ixgbe_fc_full;
4912 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
4914 adapter->last_lfc_mode = hw->fc.current_mode;
4916 hw->fc.high_water = FC_HIGH_WATER(max_frame);
4917 hw->fc.low_water = FC_LOW_WATER(max_frame);
4918 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
4919 hw->fc.send_xon = true;
4920 hw->fc.disable_fc_autoneg = false;
4922 /* enable itr by default in dynamic mode */
4923 adapter->rx_itr_setting = 1;
4924 adapter->rx_eitr_param = 20000;
4925 adapter->tx_itr_setting = 1;
4926 adapter->tx_eitr_param = 10000;
4928 /* set defaults for eitr in MegaBytes */
4929 adapter->eitr_low = 10;
4930 adapter->eitr_high = 20;
4932 /* set default ring sizes */
4933 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
4934 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
4936 /* set default work limits */
4937 adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
4939 /* initialize eeprom parameters */
4940 if (ixgbe_init_eeprom_params_generic(hw)) {
4941 e_dev_err("EEPROM initialization failed\n");
4945 /* enable rx csum by default */
4946 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
4948 /* get assigned NUMA node */
4949 adapter->node = dev_to_node(&pdev->dev);
4951 set_bit(__IXGBE_DOWN, &adapter->state);
4957 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
4958 * @tx_ring: tx descriptor ring (for a specific queue) to setup
4960 * Return 0 on success, negative on failure
4962 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
4964 struct device *dev = tx_ring->dev;
4967 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4968 tx_ring->tx_buffer_info = vzalloc_node(size, tx_ring->numa_node);
4969 if (!tx_ring->tx_buffer_info)
4970 tx_ring->tx_buffer_info = vzalloc(size);
4971 if (!tx_ring->tx_buffer_info)
4974 /* round up to nearest 4K */
4975 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
4976 tx_ring->size = ALIGN(tx_ring->size, 4096);
4978 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
4979 &tx_ring->dma, GFP_KERNEL);
4983 tx_ring->next_to_use = 0;
4984 tx_ring->next_to_clean = 0;
4988 vfree(tx_ring->tx_buffer_info);
4989 tx_ring->tx_buffer_info = NULL;
4990 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
4995 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
4996 * @adapter: board private structure
4998 * If this function returns with an error, then it's possible one or
4999 * more of the rings is populated (while the rest are not). It is the
5000 * callers duty to clean those orphaned rings.
5002 * Return 0 on success, negative on failure
5004 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
5008 for (i = 0; i < adapter->num_tx_queues; i++) {
5009 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
5012 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
5020 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
5021 * @rx_ring: rx descriptor ring (for a specific queue) to setup
5023 * Returns 0 on success, negative on failure
5025 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
5027 struct device *dev = rx_ring->dev;
5030 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
5031 rx_ring->rx_buffer_info = vzalloc_node(size, rx_ring->numa_node);
5032 if (!rx_ring->rx_buffer_info)
5033 rx_ring->rx_buffer_info = vzalloc(size);
5034 if (!rx_ring->rx_buffer_info)
5037 /* Round up to nearest 4K */
5038 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
5039 rx_ring->size = ALIGN(rx_ring->size, 4096);
5041 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
5042 &rx_ring->dma, GFP_KERNEL);
5047 rx_ring->next_to_clean = 0;
5048 rx_ring->next_to_use = 0;
5052 vfree(rx_ring->rx_buffer_info);
5053 rx_ring->rx_buffer_info = NULL;
5054 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
5059 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
5060 * @adapter: board private structure
5062 * If this function returns with an error, then it's possible one or
5063 * more of the rings is populated (while the rest are not). It is the
5064 * callers duty to clean those orphaned rings.
5066 * Return 0 on success, negative on failure
5068 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
5072 for (i = 0; i < adapter->num_rx_queues; i++) {
5073 err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
5076 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
5084 * ixgbe_free_tx_resources - Free Tx Resources per Queue
5085 * @tx_ring: Tx descriptor ring for a specific queue
5087 * Free all transmit software resources
5089 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
5091 ixgbe_clean_tx_ring(tx_ring);
5093 vfree(tx_ring->tx_buffer_info);
5094 tx_ring->tx_buffer_info = NULL;
5096 /* if not set, then don't free */
5100 dma_free_coherent(tx_ring->dev, tx_ring->size,
5101 tx_ring->desc, tx_ring->dma);
5103 tx_ring->desc = NULL;
5107 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5108 * @adapter: board private structure
5110 * Free all transmit software resources
5112 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
5116 for (i = 0; i < adapter->num_tx_queues; i++)
5117 if (adapter->tx_ring[i]->desc)
5118 ixgbe_free_tx_resources(adapter->tx_ring[i]);
5122 * ixgbe_free_rx_resources - Free Rx Resources
5123 * @rx_ring: ring to clean the resources from
5125 * Free all receive software resources
5127 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
5129 ixgbe_clean_rx_ring(rx_ring);
5131 vfree(rx_ring->rx_buffer_info);
5132 rx_ring->rx_buffer_info = NULL;
5134 /* if not set, then don't free */
5138 dma_free_coherent(rx_ring->dev, rx_ring->size,
5139 rx_ring->desc, rx_ring->dma);
5141 rx_ring->desc = NULL;
5145 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5146 * @adapter: board private structure
5148 * Free all receive software resources
5150 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5154 for (i = 0; i < adapter->num_rx_queues; i++)
5155 if (adapter->rx_ring[i]->desc)
5156 ixgbe_free_rx_resources(adapter->rx_ring[i]);
5160 * ixgbe_change_mtu - Change the Maximum Transfer Unit
5161 * @netdev: network interface device structure
5162 * @new_mtu: new value for maximum frame size
5164 * Returns 0 on success, negative on failure
5166 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5168 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5169 struct ixgbe_hw *hw = &adapter->hw;
5170 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5172 /* MTU < 68 is an error and causes problems on some kernels */
5173 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED &&
5174 hw->mac.type != ixgbe_mac_X540) {
5175 if ((new_mtu < 68) || (max_frame > MAXIMUM_ETHERNET_VLAN_SIZE))
5178 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5182 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5183 /* must set new MTU before calling down or up */
5184 netdev->mtu = new_mtu;
5186 hw->fc.high_water = FC_HIGH_WATER(max_frame);
5187 hw->fc.low_water = FC_LOW_WATER(max_frame);
5189 if (netif_running(netdev))
5190 ixgbe_reinit_locked(adapter);
5196 * ixgbe_open - Called when a network interface is made active
5197 * @netdev: network interface device structure
5199 * Returns 0 on success, negative value on failure
5201 * The open entry point is called when a network interface is made
5202 * active by the system (IFF_UP). At this point all resources needed
5203 * for transmit and receive operations are allocated, the interrupt
5204 * handler is registered with the OS, the watchdog timer is started,
5205 * and the stack is notified that the interface is ready.
5207 static int ixgbe_open(struct net_device *netdev)
5209 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5212 /* disallow open during test */
5213 if (test_bit(__IXGBE_TESTING, &adapter->state))
5216 netif_carrier_off(netdev);
5218 /* allocate transmit descriptors */
5219 err = ixgbe_setup_all_tx_resources(adapter);
5223 /* allocate receive descriptors */
5224 err = ixgbe_setup_all_rx_resources(adapter);
5228 ixgbe_configure(adapter);
5230 err = ixgbe_request_irq(adapter);
5234 err = ixgbe_up_complete(adapter);
5238 netif_tx_start_all_queues(netdev);
5243 ixgbe_release_hw_control(adapter);
5244 ixgbe_free_irq(adapter);
5247 ixgbe_free_all_rx_resources(adapter);
5249 ixgbe_free_all_tx_resources(adapter);
5250 ixgbe_reset(adapter);
5256 * ixgbe_close - Disables a network interface
5257 * @netdev: network interface device structure
5259 * Returns 0, this is not allowed to fail
5261 * The close entry point is called when an interface is de-activated
5262 * by the OS. The hardware is still under the drivers control, but
5263 * needs to be disabled. A global MAC reset is issued to stop the
5264 * hardware, and all transmit and receive resources are freed.
5266 static int ixgbe_close(struct net_device *netdev)
5268 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5270 ixgbe_down(adapter);
5271 ixgbe_free_irq(adapter);
5273 ixgbe_fdir_filter_exit(adapter);
5275 ixgbe_free_all_tx_resources(adapter);
5276 ixgbe_free_all_rx_resources(adapter);
5278 ixgbe_release_hw_control(adapter);
5284 static int ixgbe_resume(struct pci_dev *pdev)
5286 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5287 struct net_device *netdev = adapter->netdev;
5290 pci_set_power_state(pdev, PCI_D0);
5291 pci_restore_state(pdev);
5293 * pci_restore_state clears dev->state_saved so call
5294 * pci_save_state to restore it.
5296 pci_save_state(pdev);
5298 err = pci_enable_device_mem(pdev);
5300 e_dev_err("Cannot enable PCI device from suspend\n");
5303 pci_set_master(pdev);
5305 pci_wake_from_d3(pdev, false);
5307 err = ixgbe_init_interrupt_scheme(adapter);
5309 e_dev_err("Cannot initialize interrupts for device\n");
5313 ixgbe_reset(adapter);
5315 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5317 if (netif_running(netdev)) {
5318 err = ixgbe_open(netdev);
5323 netif_device_attach(netdev);
5327 #endif /* CONFIG_PM */
5329 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5331 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5332 struct net_device *netdev = adapter->netdev;
5333 struct ixgbe_hw *hw = &adapter->hw;
5335 u32 wufc = adapter->wol;
5340 netif_device_detach(netdev);
5342 if (netif_running(netdev)) {
5343 ixgbe_down(adapter);
5344 ixgbe_free_irq(adapter);
5345 ixgbe_free_all_tx_resources(adapter);
5346 ixgbe_free_all_rx_resources(adapter);
5349 ixgbe_clear_interrupt_scheme(adapter);
5351 kfree(adapter->ixgbe_ieee_pfc);
5352 kfree(adapter->ixgbe_ieee_ets);
5356 retval = pci_save_state(pdev);
5362 ixgbe_set_rx_mode(netdev);
5364 /* turn on all-multi mode if wake on multicast is enabled */
5365 if (wufc & IXGBE_WUFC_MC) {
5366 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5367 fctrl |= IXGBE_FCTRL_MPE;
5368 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5371 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5372 ctrl |= IXGBE_CTRL_GIO_DIS;
5373 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5375 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5377 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5378 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5381 switch (hw->mac.type) {
5382 case ixgbe_mac_82598EB:
5383 pci_wake_from_d3(pdev, false);
5385 case ixgbe_mac_82599EB:
5386 case ixgbe_mac_X540:
5387 pci_wake_from_d3(pdev, !!wufc);
5393 *enable_wake = !!wufc;
5395 ixgbe_release_hw_control(adapter);
5397 pci_disable_device(pdev);
5403 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5408 retval = __ixgbe_shutdown(pdev, &wake);
5413 pci_prepare_to_sleep(pdev);
5415 pci_wake_from_d3(pdev, false);
5416 pci_set_power_state(pdev, PCI_D3hot);
5421 #endif /* CONFIG_PM */
5423 static void ixgbe_shutdown(struct pci_dev *pdev)
5427 __ixgbe_shutdown(pdev, &wake);
5429 if (system_state == SYSTEM_POWER_OFF) {
5430 pci_wake_from_d3(pdev, wake);
5431 pci_set_power_state(pdev, PCI_D3hot);
5436 * ixgbe_update_stats - Update the board statistics counters.
5437 * @adapter: board private structure
5439 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5441 struct net_device *netdev = adapter->netdev;
5442 struct ixgbe_hw *hw = &adapter->hw;
5443 struct ixgbe_hw_stats *hwstats = &adapter->stats;
5445 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5446 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
5447 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
5448 u64 bytes = 0, packets = 0;
5450 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5451 test_bit(__IXGBE_RESETTING, &adapter->state))
5454 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5457 for (i = 0; i < 16; i++)
5458 adapter->hw_rx_no_dma_resources +=
5459 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5460 for (i = 0; i < adapter->num_rx_queues; i++) {
5461 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
5462 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
5464 adapter->rsc_total_count = rsc_count;
5465 adapter->rsc_total_flush = rsc_flush;
5468 for (i = 0; i < adapter->num_rx_queues; i++) {
5469 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
5470 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
5471 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
5472 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
5473 bytes += rx_ring->stats.bytes;
5474 packets += rx_ring->stats.packets;
5476 adapter->non_eop_descs = non_eop_descs;
5477 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5478 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5479 netdev->stats.rx_bytes = bytes;
5480 netdev->stats.rx_packets = packets;
5484 /* gather some stats to the adapter struct that are per queue */
5485 for (i = 0; i < adapter->num_tx_queues; i++) {
5486 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5487 restart_queue += tx_ring->tx_stats.restart_queue;
5488 tx_busy += tx_ring->tx_stats.tx_busy;
5489 bytes += tx_ring->stats.bytes;
5490 packets += tx_ring->stats.packets;
5492 adapter->restart_queue = restart_queue;
5493 adapter->tx_busy = tx_busy;
5494 netdev->stats.tx_bytes = bytes;
5495 netdev->stats.tx_packets = packets;
5497 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5498 for (i = 0; i < 8; i++) {
5499 /* for packet buffers not used, the register should read 0 */
5500 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5502 hwstats->mpc[i] += mpc;
5503 total_mpc += hwstats->mpc[i];
5504 if (hw->mac.type == ixgbe_mac_82598EB)
5505 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5506 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5507 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5508 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5509 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5510 switch (hw->mac.type) {
5511 case ixgbe_mac_82598EB:
5512 hwstats->pxonrxc[i] +=
5513 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5515 case ixgbe_mac_82599EB:
5516 case ixgbe_mac_X540:
5517 hwstats->pxonrxc[i] +=
5518 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5523 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5524 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5526 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5527 /* work around hardware counting issue */
5528 hwstats->gprc -= missed_rx;
5530 ixgbe_update_xoff_received(adapter);
5532 /* 82598 hardware only has a 32 bit counter in the high register */
5533 switch (hw->mac.type) {
5534 case ixgbe_mac_82598EB:
5535 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5536 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5537 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5538 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5540 case ixgbe_mac_X540:
5541 /* OS2BMC stats are X540 only*/
5542 hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
5543 hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
5544 hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
5545 hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
5546 case ixgbe_mac_82599EB:
5547 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5548 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
5549 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5550 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
5551 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5552 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5553 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5554 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5555 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5557 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5558 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5559 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5560 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5561 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5562 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5563 #endif /* IXGBE_FCOE */
5568 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5569 hwstats->bprc += bprc;
5570 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5571 if (hw->mac.type == ixgbe_mac_82598EB)
5572 hwstats->mprc -= bprc;
5573 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5574 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5575 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5576 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5577 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5578 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5579 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5580 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5581 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5582 hwstats->lxontxc += lxon;
5583 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5584 hwstats->lxofftxc += lxoff;
5585 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5586 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5587 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5589 * 82598 errata - tx of flow control packets is included in tx counters
5591 xon_off_tot = lxon + lxoff;
5592 hwstats->gptc -= xon_off_tot;
5593 hwstats->mptc -= xon_off_tot;
5594 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5595 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5596 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5597 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5598 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5599 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5600 hwstats->ptc64 -= xon_off_tot;
5601 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5602 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5603 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5604 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5605 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5606 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5608 /* Fill out the OS statistics structure */
5609 netdev->stats.multicast = hwstats->mprc;
5612 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
5613 netdev->stats.rx_dropped = 0;
5614 netdev->stats.rx_length_errors = hwstats->rlec;
5615 netdev->stats.rx_crc_errors = hwstats->crcerrs;
5616 netdev->stats.rx_missed_errors = total_mpc;
5620 * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
5621 * @adapter - pointer to the device adapter structure
5623 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
5625 struct ixgbe_hw *hw = &adapter->hw;
5628 if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
5631 adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
5633 /* if interface is down do nothing */
5634 if (test_bit(__IXGBE_DOWN, &adapter->state))
5637 /* do nothing if we are not using signature filters */
5638 if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
5641 adapter->fdir_overflow++;
5643 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5644 for (i = 0; i < adapter->num_tx_queues; i++)
5645 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
5646 &(adapter->tx_ring[i]->state));
5647 /* re-enable flow director interrupts */
5648 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
5650 e_err(probe, "failed to finish FDIR re-initialization, "
5651 "ignored adding FDIR ATR filters\n");
5656 * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
5657 * @adapter - pointer to the device adapter structure
5659 * This function serves two purposes. First it strobes the interrupt lines
5660 * in order to make certain interrupts are occuring. Secondly it sets the
5661 * bits needed to check for TX hangs. As a result we should immediately
5662 * determine if a hang has occured.
5664 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
5666 struct ixgbe_hw *hw = &adapter->hw;
5670 /* If we're down or resetting, just bail */
5671 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5672 test_bit(__IXGBE_RESETTING, &adapter->state))
5675 /* Force detection of hung controller */
5676 if (netif_carrier_ok(adapter->netdev)) {
5677 for (i = 0; i < adapter->num_tx_queues; i++)
5678 set_check_for_tx_hang(adapter->tx_ring[i]);
5681 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5683 * for legacy and MSI interrupts don't set any bits
5684 * that are enabled for EIAM, because this operation
5685 * would set *both* EIMS and EICS for any bit in EIAM
5687 IXGBE_WRITE_REG(hw, IXGBE_EICS,
5688 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5690 /* get one bit for every active tx/rx interrupt vector */
5691 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
5692 struct ixgbe_q_vector *qv = adapter->q_vector[i];
5693 if (qv->rx.ring || qv->tx.ring)
5694 eics |= ((u64)1 << i);
5698 /* Cause software interrupt to ensure rings are cleaned */
5699 ixgbe_irq_rearm_queues(adapter, eics);
5704 * ixgbe_watchdog_update_link - update the link status
5705 * @adapter - pointer to the device adapter structure
5706 * @link_speed - pointer to a u32 to store the link_speed
5708 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
5710 struct ixgbe_hw *hw = &adapter->hw;
5711 u32 link_speed = adapter->link_speed;
5712 bool link_up = adapter->link_up;
5715 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
5718 if (hw->mac.ops.check_link) {
5719 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5721 /* always assume link is up, if no check link function */
5722 link_speed = IXGBE_LINK_SPEED_10GB_FULL;
5726 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5727 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
5728 hw->mac.ops.fc_enable(hw, i);
5730 hw->mac.ops.fc_enable(hw, 0);
5735 time_after(jiffies, (adapter->link_check_timeout +
5736 IXGBE_TRY_LINK_TIMEOUT))) {
5737 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5738 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5739 IXGBE_WRITE_FLUSH(hw);
5742 adapter->link_up = link_up;
5743 adapter->link_speed = link_speed;
5747 * ixgbe_watchdog_link_is_up - update netif_carrier status and
5748 * print link up message
5749 * @adapter - pointer to the device adapter structure
5751 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
5753 struct net_device *netdev = adapter->netdev;
5754 struct ixgbe_hw *hw = &adapter->hw;
5755 u32 link_speed = adapter->link_speed;
5756 bool flow_rx, flow_tx;
5758 /* only continue if link was previously down */
5759 if (netif_carrier_ok(netdev))
5762 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
5764 switch (hw->mac.type) {
5765 case ixgbe_mac_82598EB: {
5766 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5767 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5768 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5769 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5772 case ixgbe_mac_X540:
5773 case ixgbe_mac_82599EB: {
5774 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5775 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5776 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5777 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5785 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
5786 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5788 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5790 (link_speed == IXGBE_LINK_SPEED_100_FULL ?
5793 ((flow_rx && flow_tx) ? "RX/TX" :
5795 (flow_tx ? "TX" : "None"))));
5797 netif_carrier_on(netdev);
5798 ixgbe_check_vf_rate_limit(adapter);
5802 * ixgbe_watchdog_link_is_down - update netif_carrier status and
5803 * print link down message
5804 * @adapter - pointer to the adapter structure
5806 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter* adapter)
5808 struct net_device *netdev = adapter->netdev;
5809 struct ixgbe_hw *hw = &adapter->hw;
5811 adapter->link_up = false;
5812 adapter->link_speed = 0;
5814 /* only continue if link was up previously */
5815 if (!netif_carrier_ok(netdev))
5818 /* poll for SFP+ cable when link is down */
5819 if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
5820 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5822 e_info(drv, "NIC Link is Down\n");
5823 netif_carrier_off(netdev);
5827 * ixgbe_watchdog_flush_tx - flush queues on link down
5828 * @adapter - pointer to the device adapter structure
5830 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
5833 int some_tx_pending = 0;
5835 if (!netif_carrier_ok(adapter->netdev)) {
5836 for (i = 0; i < adapter->num_tx_queues; i++) {
5837 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5838 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5839 some_tx_pending = 1;
5844 if (some_tx_pending) {
5845 /* We've lost link, so the controller stops DMA,
5846 * but we've got queued Tx work that's never going
5847 * to get done, so reset controller to flush Tx.
5848 * (Do the reset outside of interrupt context).
5850 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
5855 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
5859 /* Do not perform spoof check for 82598 */
5860 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
5863 ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
5866 * ssvpc register is cleared on read, if zero then no
5867 * spoofed packets in the last interval.
5872 e_warn(drv, "%d Spoofed packets detected\n", ssvpc);
5876 * ixgbe_watchdog_subtask - check and bring link up
5877 * @adapter - pointer to the device adapter structure
5879 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
5881 /* if interface is down do nothing */
5882 if (test_bit(__IXGBE_DOWN, &adapter->state))
5885 ixgbe_watchdog_update_link(adapter);
5887 if (adapter->link_up)
5888 ixgbe_watchdog_link_is_up(adapter);
5890 ixgbe_watchdog_link_is_down(adapter);
5892 ixgbe_spoof_check(adapter);
5893 ixgbe_update_stats(adapter);
5895 ixgbe_watchdog_flush_tx(adapter);
5899 * ixgbe_sfp_detection_subtask - poll for SFP+ cable
5900 * @adapter - the ixgbe adapter structure
5902 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
5904 struct ixgbe_hw *hw = &adapter->hw;
5907 /* not searching for SFP so there is nothing to do here */
5908 if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
5909 !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
5912 /* someone else is in init, wait until next service event */
5913 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5916 err = hw->phy.ops.identify_sfp(hw);
5917 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
5920 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
5921 /* If no cable is present, then we need to reset
5922 * the next time we find a good cable. */
5923 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
5930 /* exit if reset not needed */
5931 if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
5934 adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
5937 * A module may be identified correctly, but the EEPROM may not have
5938 * support for that module. setup_sfp() will fail in that case, so
5939 * we should not allow that module to load.
5941 if (hw->mac.type == ixgbe_mac_82598EB)
5942 err = hw->phy.ops.reset(hw);
5944 err = hw->mac.ops.setup_sfp(hw);
5946 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
5949 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
5950 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
5953 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5955 if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
5956 (adapter->netdev->reg_state == NETREG_REGISTERED)) {
5957 e_dev_err("failed to initialize because an unsupported "
5958 "SFP+ module type was detected.\n");
5959 e_dev_err("Reload the driver after installing a "
5960 "supported module.\n");
5961 unregister_netdev(adapter->netdev);
5966 * ixgbe_sfp_link_config_subtask - set up link SFP after module install
5967 * @adapter - the ixgbe adapter structure
5969 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
5971 struct ixgbe_hw *hw = &adapter->hw;
5975 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
5978 /* someone else is in init, wait until next service event */
5979 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5982 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
5984 autoneg = hw->phy.autoneg_advertised;
5985 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
5986 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
5987 hw->mac.autotry_restart = false;
5988 if (hw->mac.ops.setup_link)
5989 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
5991 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5992 adapter->link_check_timeout = jiffies;
5993 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5997 * ixgbe_service_timer - Timer Call-back
5998 * @data: pointer to adapter cast into an unsigned long
6000 static void ixgbe_service_timer(unsigned long data)
6002 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
6003 unsigned long next_event_offset;
6005 /* poll faster when waiting for link */
6006 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
6007 next_event_offset = HZ / 10;
6009 next_event_offset = HZ * 2;
6011 /* Reset the timer */
6012 mod_timer(&adapter->service_timer, next_event_offset + jiffies);
6014 ixgbe_service_event_schedule(adapter);
6017 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
6019 if (!(adapter->flags2 & IXGBE_FLAG2_RESET_REQUESTED))
6022 adapter->flags2 &= ~IXGBE_FLAG2_RESET_REQUESTED;
6024 /* If we're already down or resetting, just bail */
6025 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6026 test_bit(__IXGBE_RESETTING, &adapter->state))
6029 ixgbe_dump(adapter);
6030 netdev_err(adapter->netdev, "Reset adapter\n");
6031 adapter->tx_timeout_count++;
6033 ixgbe_reinit_locked(adapter);
6037 * ixgbe_service_task - manages and runs subtasks
6038 * @work: pointer to work_struct containing our data
6040 static void ixgbe_service_task(struct work_struct *work)
6042 struct ixgbe_adapter *adapter = container_of(work,
6043 struct ixgbe_adapter,
6046 ixgbe_reset_subtask(adapter);
6047 ixgbe_sfp_detection_subtask(adapter);
6048 ixgbe_sfp_link_config_subtask(adapter);
6049 ixgbe_check_overtemp_subtask(adapter);
6050 ixgbe_watchdog_subtask(adapter);
6051 ixgbe_fdir_reinit_subtask(adapter);
6052 ixgbe_check_hang_subtask(adapter);
6054 ixgbe_service_event_complete(adapter);
6057 void ixgbe_tx_ctxtdesc(struct ixgbe_ring *tx_ring, u32 vlan_macip_lens,
6058 u32 fcoe_sof_eof, u32 type_tucmd, u32 mss_l4len_idx)
6060 struct ixgbe_adv_tx_context_desc *context_desc;
6061 u16 i = tx_ring->next_to_use;
6063 context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
6066 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
6068 /* set bits to identify this as an advanced context descriptor */
6069 type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
6071 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
6072 context_desc->seqnum_seed = cpu_to_le32(fcoe_sof_eof);
6073 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd);
6074 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
6077 static int ixgbe_tso(struct ixgbe_ring *tx_ring, struct sk_buff *skb,
6078 u32 tx_flags, __be16 protocol, u8 *hdr_len)
6081 u32 vlan_macip_lens, type_tucmd;
6082 u32 mss_l4len_idx, l4len;
6084 if (!skb_is_gso(skb))
6087 if (skb_header_cloned(skb)) {
6088 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
6093 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
6094 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
6096 if (protocol == __constant_htons(ETH_P_IP)) {
6097 struct iphdr *iph = ip_hdr(skb);
6100 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
6104 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
6105 } else if (skb_is_gso_v6(skb)) {
6106 ipv6_hdr(skb)->payload_len = 0;
6107 tcp_hdr(skb)->check =
6108 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
6109 &ipv6_hdr(skb)->daddr,
6113 l4len = tcp_hdrlen(skb);
6114 *hdr_len = skb_transport_offset(skb) + l4len;
6116 /* mss_l4len_id: use 1 as index for TSO */
6117 mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
6118 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
6119 mss_l4len_idx |= 1 << IXGBE_ADVTXD_IDX_SHIFT;
6121 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
6122 vlan_macip_lens = skb_network_header_len(skb);
6123 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
6124 vlan_macip_lens |= tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
6126 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0, type_tucmd,
6132 static bool ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
6133 struct sk_buff *skb, u32 tx_flags,
6136 u32 vlan_macip_lens = 0;
6137 u32 mss_l4len_idx = 0;
6140 if (skb->ip_summed != CHECKSUM_PARTIAL) {
6141 if (!(tx_flags & IXGBE_TX_FLAGS_HW_VLAN) &&
6142 !(tx_flags & IXGBE_TX_FLAGS_TXSW))
6147 case __constant_htons(ETH_P_IP):
6148 vlan_macip_lens |= skb_network_header_len(skb);
6149 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
6150 l4_hdr = ip_hdr(skb)->protocol;
6152 case __constant_htons(ETH_P_IPV6):
6153 vlan_macip_lens |= skb_network_header_len(skb);
6154 l4_hdr = ipv6_hdr(skb)->nexthdr;
6157 if (unlikely(net_ratelimit())) {
6158 dev_warn(tx_ring->dev,
6159 "partial checksum but proto=%x!\n",
6167 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6168 mss_l4len_idx = tcp_hdrlen(skb) <<
6169 IXGBE_ADVTXD_L4LEN_SHIFT;
6172 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
6173 mss_l4len_idx = sizeof(struct sctphdr) <<
6174 IXGBE_ADVTXD_L4LEN_SHIFT;
6177 mss_l4len_idx = sizeof(struct udphdr) <<
6178 IXGBE_ADVTXD_L4LEN_SHIFT;
6181 if (unlikely(net_ratelimit())) {
6182 dev_warn(tx_ring->dev,
6183 "partial checksum but l4 proto=%x!\n",
6190 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
6191 vlan_macip_lens |= tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
6193 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0,
6194 type_tucmd, mss_l4len_idx);
6196 return (skb->ip_summed == CHECKSUM_PARTIAL);
6199 static __le32 ixgbe_tx_cmd_type(u32 tx_flags)
6201 /* set type for advanced descriptor with frame checksum insertion */
6202 __le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
6203 IXGBE_ADVTXD_DCMD_IFCS |
6204 IXGBE_ADVTXD_DCMD_DEXT);
6206 /* set HW vlan bit if vlan is present */
6207 if (tx_flags & IXGBE_TX_FLAGS_HW_VLAN)
6208 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
6210 /* set segmentation enable bits for TSO/FSO */
6212 if ((tx_flags & IXGBE_TX_FLAGS_TSO) || (tx_flags & IXGBE_TX_FLAGS_FSO))
6214 if (tx_flags & IXGBE_TX_FLAGS_TSO)
6216 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
6221 static __le32 ixgbe_tx_olinfo_status(u32 tx_flags, unsigned int paylen)
6223 __le32 olinfo_status =
6224 cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
6226 if (tx_flags & IXGBE_TX_FLAGS_TSO) {
6227 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM |
6228 (1 << IXGBE_ADVTXD_IDX_SHIFT));
6229 /* enble IPv4 checksum for TSO */
6230 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
6231 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
6234 /* enable L4 checksum for TSO and TX checksum offload */
6235 if (tx_flags & IXGBE_TX_FLAGS_CSUM)
6236 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
6239 /* use index 1 context for FCOE/FSO */
6240 if (tx_flags & IXGBE_TX_FLAGS_FCOE)
6241 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC |
6242 (1 << IXGBE_ADVTXD_IDX_SHIFT));
6246 * Check Context must be set if Tx switch is enabled, which it
6247 * always is for case where virtual functions are running
6249 if (tx_flags & IXGBE_TX_FLAGS_TXSW)
6250 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
6252 return olinfo_status;
6255 #define IXGBE_TXD_CMD (IXGBE_TXD_CMD_EOP | \
6258 static void ixgbe_tx_map(struct ixgbe_ring *tx_ring,
6259 struct sk_buff *skb,
6260 struct ixgbe_tx_buffer *first,
6264 struct device *dev = tx_ring->dev;
6265 struct ixgbe_tx_buffer *tx_buffer_info;
6266 union ixgbe_adv_tx_desc *tx_desc;
6268 __le32 cmd_type, olinfo_status;
6269 struct skb_frag_struct *frag;
6271 unsigned int data_len = skb->data_len;
6272 unsigned int size = skb_headlen(skb);
6274 u32 paylen = skb->len - hdr_len;
6275 u16 i = tx_ring->next_to_use;
6279 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6280 if (data_len >= sizeof(struct fcoe_crc_eof)) {
6281 data_len -= sizeof(struct fcoe_crc_eof);
6283 size -= sizeof(struct fcoe_crc_eof) - data_len;
6289 dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE);
6290 if (dma_mapping_error(dev, dma))
6293 cmd_type = ixgbe_tx_cmd_type(tx_flags);
6294 olinfo_status = ixgbe_tx_olinfo_status(tx_flags, paylen);
6296 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
6299 while (size > IXGBE_MAX_DATA_PER_TXD) {
6300 tx_desc->read.buffer_addr = cpu_to_le64(dma + offset);
6301 tx_desc->read.cmd_type_len =
6302 cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
6303 tx_desc->read.olinfo_status = olinfo_status;
6305 offset += IXGBE_MAX_DATA_PER_TXD;
6306 size -= IXGBE_MAX_DATA_PER_TXD;
6310 if (i == tx_ring->count) {
6311 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, 0);
6316 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6317 tx_buffer_info->length = offset + size;
6318 tx_buffer_info->tx_flags = tx_flags;
6319 tx_buffer_info->dma = dma;
6321 tx_desc->read.buffer_addr = cpu_to_le64(dma + offset);
6322 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
6323 tx_desc->read.olinfo_status = olinfo_status;
6328 frag = &skb_shinfo(skb)->frags[f];
6330 size = min_t(unsigned int, data_len, frag->size);
6338 tx_flags |= IXGBE_TX_FLAGS_MAPPED_AS_PAGE;
6340 dma = dma_map_page(dev, frag->page, frag->page_offset,
6341 size, DMA_TO_DEVICE);
6342 if (dma_mapping_error(dev, dma))
6347 if (i == tx_ring->count) {
6348 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, 0);
6353 tx_desc->read.cmd_type_len |= cpu_to_le32(IXGBE_TXD_CMD);
6356 if (i == tx_ring->count)
6359 tx_ring->next_to_use = i;
6361 if (tx_flags & IXGBE_TX_FLAGS_TSO)
6362 gso_segs = skb_shinfo(skb)->gso_segs;
6364 /* adjust for FCoE Sequence Offload */
6365 else if (tx_flags & IXGBE_TX_FLAGS_FSO)
6366 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
6367 skb_shinfo(skb)->gso_size);
6368 #endif /* IXGBE_FCOE */
6372 /* multiply data chunks by size of headers */
6373 tx_buffer_info->bytecount = paylen + (gso_segs * hdr_len);
6374 tx_buffer_info->gso_segs = gso_segs;
6375 tx_buffer_info->skb = skb;
6377 /* set the timestamp */
6378 first->time_stamp = jiffies;
6381 * Force memory writes to complete before letting h/w
6382 * know there are new descriptors to fetch. (Only
6383 * applicable for weak-ordered memory model archs,
6388 /* set next_to_watch value indicating a packet is present */
6389 first->next_to_watch = tx_desc;
6391 /* notify HW of packet */
6392 writel(i, tx_ring->tail);
6396 dev_err(dev, "TX DMA map failed\n");
6398 /* clear dma mappings for failed tx_buffer_info map */
6400 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6401 ixgbe_unmap_tx_resource(tx_ring, tx_buffer_info);
6402 if (tx_buffer_info == first)
6409 dev_kfree_skb_any(skb);
6411 tx_ring->next_to_use = i;
6414 static void ixgbe_atr(struct ixgbe_ring *ring, struct sk_buff *skb,
6415 u32 tx_flags, __be16 protocol)
6417 struct ixgbe_q_vector *q_vector = ring->q_vector;
6418 union ixgbe_atr_hash_dword input = { .dword = 0 };
6419 union ixgbe_atr_hash_dword common = { .dword = 0 };
6421 unsigned char *network;
6423 struct ipv6hdr *ipv6;
6428 /* if ring doesn't have a interrupt vector, cannot perform ATR */
6432 /* do nothing if sampling is disabled */
6433 if (!ring->atr_sample_rate)
6438 /* snag network header to get L4 type and address */
6439 hdr.network = skb_network_header(skb);
6441 /* Currently only IPv4/IPv6 with TCP is supported */
6442 if ((protocol != __constant_htons(ETH_P_IPV6) ||
6443 hdr.ipv6->nexthdr != IPPROTO_TCP) &&
6444 (protocol != __constant_htons(ETH_P_IP) ||
6445 hdr.ipv4->protocol != IPPROTO_TCP))
6450 /* skip this packet since it is invalid or the socket is closing */
6454 /* sample on all syn packets or once every atr sample count */
6455 if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
6458 /* reset sample count */
6459 ring->atr_count = 0;
6461 vlan_id = htons(tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
6464 * src and dst are inverted, think how the receiver sees them
6466 * The input is broken into two sections, a non-compressed section
6467 * containing vm_pool, vlan_id, and flow_type. The rest of the data
6468 * is XORed together and stored in the compressed dword.
6470 input.formatted.vlan_id = vlan_id;
6473 * since src port and flex bytes occupy the same word XOR them together
6474 * and write the value to source port portion of compressed dword
6476 if (tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
6477 common.port.src ^= th->dest ^ __constant_htons(ETH_P_8021Q);
6479 common.port.src ^= th->dest ^ protocol;
6480 common.port.dst ^= th->source;
6482 if (protocol == __constant_htons(ETH_P_IP)) {
6483 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
6484 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
6486 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
6487 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
6488 hdr.ipv6->saddr.s6_addr32[1] ^
6489 hdr.ipv6->saddr.s6_addr32[2] ^
6490 hdr.ipv6->saddr.s6_addr32[3] ^
6491 hdr.ipv6->daddr.s6_addr32[0] ^
6492 hdr.ipv6->daddr.s6_addr32[1] ^
6493 hdr.ipv6->daddr.s6_addr32[2] ^
6494 hdr.ipv6->daddr.s6_addr32[3];
6497 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
6498 ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
6499 input, common, ring->queue_index);
6502 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6504 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
6505 /* Herbert's original patch had:
6506 * smp_mb__after_netif_stop_queue();
6507 * but since that doesn't exist yet, just open code it. */
6510 /* We need to check again in a case another CPU has just
6511 * made room available. */
6512 if (likely(ixgbe_desc_unused(tx_ring) < size))
6515 /* A reprieve! - use start_queue because it doesn't call schedule */
6516 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
6517 ++tx_ring->tx_stats.restart_queue;
6521 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6523 if (likely(ixgbe_desc_unused(tx_ring) >= size))
6525 return __ixgbe_maybe_stop_tx(tx_ring, size);
6528 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6530 struct ixgbe_adapter *adapter = netdev_priv(dev);
6531 int txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
6534 __be16 protocol = vlan_get_protocol(skb);
6536 if (((protocol == htons(ETH_P_FCOE)) ||
6537 (protocol == htons(ETH_P_FIP))) &&
6538 (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)) {
6539 txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
6540 txq += adapter->ring_feature[RING_F_FCOE].mask;
6545 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6546 while (unlikely(txq >= dev->real_num_tx_queues))
6547 txq -= dev->real_num_tx_queues;
6551 return skb_tx_hash(dev, skb);
6554 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
6555 struct ixgbe_adapter *adapter,
6556 struct ixgbe_ring *tx_ring)
6558 struct ixgbe_tx_buffer *first;
6561 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6564 u16 count = TXD_USE_COUNT(skb_headlen(skb));
6565 __be16 protocol = skb->protocol;
6569 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
6570 * + 1 desc for skb_head_len/IXGBE_MAX_DATA_PER_TXD,
6571 * + 2 desc gap to keep tail from touching head,
6572 * + 1 desc for context descriptor,
6573 * otherwise try next time
6575 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6576 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6577 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6579 count += skb_shinfo(skb)->nr_frags;
6581 if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
6582 tx_ring->tx_stats.tx_busy++;
6583 return NETDEV_TX_BUSY;
6586 #ifdef CONFIG_PCI_IOV
6587 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6588 tx_flags |= IXGBE_TX_FLAGS_TXSW;
6591 /* if we have a HW VLAN tag being added default to the HW one */
6592 if (vlan_tx_tag_present(skb)) {
6593 tx_flags |= vlan_tx_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
6594 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6595 /* else if it is a SW VLAN check the next protocol and store the tag */
6596 } else if (protocol == __constant_htons(ETH_P_8021Q)) {
6597 struct vlan_hdr *vhdr, _vhdr;
6598 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
6602 protocol = vhdr->h_vlan_encapsulated_proto;
6603 tx_flags |= ntohs(vhdr->h_vlan_TCI) << IXGBE_TX_FLAGS_VLAN_SHIFT;
6604 tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
6607 if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
6608 ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
6609 (skb->priority != TC_PRIO_CONTROL))) {
6610 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6611 tx_flags |= tx_ring->dcb_tc <<
6612 IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
6613 if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
6614 struct vlan_ethhdr *vhdr;
6615 if (skb_header_cloned(skb) &&
6616 pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
6618 vhdr = (struct vlan_ethhdr *)skb->data;
6619 vhdr->h_vlan_TCI = htons(tx_flags >>
6620 IXGBE_TX_FLAGS_VLAN_SHIFT);
6622 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6626 /* record the location of the first descriptor for this packet */
6627 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
6630 /* setup tx offload for FCoE */
6631 if ((protocol == __constant_htons(ETH_P_FCOE)) &&
6632 (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)) {
6633 tso = ixgbe_fso(tx_ring, skb, tx_flags, &hdr_len);
6637 tx_flags |= IXGBE_TX_FLAGS_FSO |
6638 IXGBE_TX_FLAGS_FCOE;
6640 tx_flags |= IXGBE_TX_FLAGS_FCOE;
6645 #endif /* IXGBE_FCOE */
6646 /* setup IPv4/IPv6 offloads */
6647 if (protocol == __constant_htons(ETH_P_IP))
6648 tx_flags |= IXGBE_TX_FLAGS_IPV4;
6650 tso = ixgbe_tso(tx_ring, skb, tx_flags, protocol, &hdr_len);
6654 tx_flags |= IXGBE_TX_FLAGS_TSO;
6655 else if (ixgbe_tx_csum(tx_ring, skb, tx_flags, protocol))
6656 tx_flags |= IXGBE_TX_FLAGS_CSUM;
6658 /* add the ATR filter if ATR is on */
6659 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
6660 ixgbe_atr(tx_ring, skb, tx_flags, protocol);
6664 #endif /* IXGBE_FCOE */
6665 ixgbe_tx_map(tx_ring, skb, first, tx_flags, hdr_len);
6667 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
6669 return NETDEV_TX_OK;
6672 dev_kfree_skb_any(skb);
6673 return NETDEV_TX_OK;
6676 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
6678 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6679 struct ixgbe_ring *tx_ring;
6681 tx_ring = adapter->tx_ring[skb->queue_mapping];
6682 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
6686 * ixgbe_set_mac - Change the Ethernet Address of the NIC
6687 * @netdev: network interface device structure
6688 * @p: pointer to an address structure
6690 * Returns 0 on success, negative on failure
6692 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6694 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6695 struct ixgbe_hw *hw = &adapter->hw;
6696 struct sockaddr *addr = p;
6698 if (!is_valid_ether_addr(addr->sa_data))
6699 return -EADDRNOTAVAIL;
6701 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6702 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6704 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
6711 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6713 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6714 struct ixgbe_hw *hw = &adapter->hw;
6718 if (prtad != hw->phy.mdio.prtad)
6720 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6726 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6727 u16 addr, u16 value)
6729 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6730 struct ixgbe_hw *hw = &adapter->hw;
6732 if (prtad != hw->phy.mdio.prtad)
6734 return hw->phy.ops.write_reg(hw, addr, devad, value);
6737 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6739 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6741 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6745 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6747 * @netdev: network interface device structure
6749 * Returns non-zero on failure
6751 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6754 struct ixgbe_adapter *adapter = netdev_priv(dev);
6755 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6757 if (is_valid_ether_addr(mac->san_addr)) {
6759 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6766 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6768 * @netdev: network interface device structure
6770 * Returns non-zero on failure
6772 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6775 struct ixgbe_adapter *adapter = netdev_priv(dev);
6776 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6778 if (is_valid_ether_addr(mac->san_addr)) {
6780 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6786 #ifdef CONFIG_NET_POLL_CONTROLLER
6788 * Polling 'interrupt' - used by things like netconsole to send skbs
6789 * without having to re-enable interrupts. It's not called while
6790 * the interrupt routine is executing.
6792 static void ixgbe_netpoll(struct net_device *netdev)
6794 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6797 /* if interface is down do nothing */
6798 if (test_bit(__IXGBE_DOWN, &adapter->state))
6801 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
6802 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
6803 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
6804 for (i = 0; i < num_q_vectors; i++) {
6805 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
6806 ixgbe_msix_clean_rings(0, q_vector);
6809 ixgbe_intr(adapter->pdev->irq, netdev);
6811 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
6815 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
6816 struct rtnl_link_stats64 *stats)
6818 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6822 for (i = 0; i < adapter->num_rx_queues; i++) {
6823 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
6829 start = u64_stats_fetch_begin_bh(&ring->syncp);
6830 packets = ring->stats.packets;
6831 bytes = ring->stats.bytes;
6832 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6833 stats->rx_packets += packets;
6834 stats->rx_bytes += bytes;
6838 for (i = 0; i < adapter->num_tx_queues; i++) {
6839 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->tx_ring[i]);
6845 start = u64_stats_fetch_begin_bh(&ring->syncp);
6846 packets = ring->stats.packets;
6847 bytes = ring->stats.bytes;
6848 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6849 stats->tx_packets += packets;
6850 stats->tx_bytes += bytes;
6854 /* following stats updated by ixgbe_watchdog_task() */
6855 stats->multicast = netdev->stats.multicast;
6856 stats->rx_errors = netdev->stats.rx_errors;
6857 stats->rx_length_errors = netdev->stats.rx_length_errors;
6858 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
6859 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
6863 /* ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
6864 * #adapter: pointer to ixgbe_adapter
6865 * @tc: number of traffic classes currently enabled
6867 * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
6868 * 802.1Q priority maps to a packet buffer that exists.
6870 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
6872 struct ixgbe_hw *hw = &adapter->hw;
6876 /* 82598 have a static priority to TC mapping that can not
6877 * be changed so no validation is needed.
6879 if (hw->mac.type == ixgbe_mac_82598EB)
6882 reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
6885 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
6886 u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
6888 /* If up2tc is out of bounds default to zero */
6890 reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
6894 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
6900 /* ixgbe_setup_tc - routine to configure net_device for multiple traffic
6903 * @netdev: net device to configure
6904 * @tc: number of traffic classes to enable
6906 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
6908 struct ixgbe_adapter *adapter = netdev_priv(dev);
6909 struct ixgbe_hw *hw = &adapter->hw;
6911 /* Multiple traffic classes requires multiple queues */
6912 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
6913 e_err(drv, "Enable failed, needs MSI-X\n");
6917 /* Hardware supports up to 8 traffic classes */
6918 if (tc > MAX_TRAFFIC_CLASS ||
6919 (hw->mac.type == ixgbe_mac_82598EB && tc < MAX_TRAFFIC_CLASS))
6922 /* Hardware has to reinitialize queues and interrupts to
6923 * match packet buffer alignment. Unfortunantly, the
6924 * hardware is not flexible enough to do this dynamically.
6926 if (netif_running(dev))
6928 ixgbe_clear_interrupt_scheme(adapter);
6931 netdev_set_num_tc(dev, tc);
6932 adapter->last_lfc_mode = adapter->hw.fc.current_mode;
6934 adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
6935 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
6937 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
6938 adapter->hw.fc.requested_mode = ixgbe_fc_none;
6940 netdev_reset_tc(dev);
6942 adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
6944 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
6945 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
6947 adapter->temp_dcb_cfg.pfc_mode_enable = false;
6948 adapter->dcb_cfg.pfc_mode_enable = false;
6951 ixgbe_init_interrupt_scheme(adapter);
6952 ixgbe_validate_rtr(adapter, tc);
6953 if (netif_running(dev))
6959 void ixgbe_do_reset(struct net_device *netdev)
6961 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6963 if (netif_running(netdev))
6964 ixgbe_reinit_locked(adapter);
6966 ixgbe_reset(adapter);
6969 static u32 ixgbe_fix_features(struct net_device *netdev, u32 data)
6971 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6974 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
6975 data &= ~NETIF_F_HW_VLAN_RX;
6978 /* return error if RXHASH is being enabled when RSS is not supported */
6979 if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED))
6980 data &= ~NETIF_F_RXHASH;
6982 /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
6983 if (!(data & NETIF_F_RXCSUM))
6984 data &= ~NETIF_F_LRO;
6986 /* Turn off LRO if not RSC capable or invalid ITR settings */
6987 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)) {
6988 data &= ~NETIF_F_LRO;
6989 } else if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
6990 (adapter->rx_itr_setting != 1 &&
6991 adapter->rx_itr_setting > IXGBE_MAX_RSC_INT_RATE)) {
6992 data &= ~NETIF_F_LRO;
6993 e_info(probe, "rx-usecs set too low, not enabling RSC\n");
6999 static int ixgbe_set_features(struct net_device *netdev, u32 data)
7001 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7002 bool need_reset = false;
7004 /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
7005 if (!(data & NETIF_F_RXCSUM))
7006 adapter->flags &= ~IXGBE_FLAG_RX_CSUM_ENABLED;
7008 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
7010 /* Make sure RSC matches LRO, reset if change */
7011 if (!!(data & NETIF_F_LRO) !=
7012 !!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
7013 adapter->flags2 ^= IXGBE_FLAG2_RSC_ENABLED;
7014 switch (adapter->hw.mac.type) {
7015 case ixgbe_mac_X540:
7016 case ixgbe_mac_82599EB:
7025 * Check if Flow Director n-tuple support was enabled or disabled. If
7026 * the state changed, we need to reset.
7028 if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)) {
7029 /* turn off ATR, enable perfect filters and reset */
7030 if (data & NETIF_F_NTUPLE) {
7031 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
7032 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
7035 } else if (!(data & NETIF_F_NTUPLE)) {
7036 /* turn off Flow Director, set ATR and reset */
7037 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
7038 if ((adapter->flags & IXGBE_FLAG_RSS_ENABLED) &&
7039 !(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
7040 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
7045 ixgbe_do_reset(netdev);
7051 static const struct net_device_ops ixgbe_netdev_ops = {
7052 .ndo_open = ixgbe_open,
7053 .ndo_stop = ixgbe_close,
7054 .ndo_start_xmit = ixgbe_xmit_frame,
7055 .ndo_select_queue = ixgbe_select_queue,
7056 .ndo_set_rx_mode = ixgbe_set_rx_mode,
7057 .ndo_validate_addr = eth_validate_addr,
7058 .ndo_set_mac_address = ixgbe_set_mac,
7059 .ndo_change_mtu = ixgbe_change_mtu,
7060 .ndo_tx_timeout = ixgbe_tx_timeout,
7061 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
7062 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
7063 .ndo_do_ioctl = ixgbe_ioctl,
7064 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
7065 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
7066 .ndo_set_vf_tx_rate = ixgbe_ndo_set_vf_bw,
7067 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
7068 .ndo_get_stats64 = ixgbe_get_stats64,
7069 .ndo_setup_tc = ixgbe_setup_tc,
7070 #ifdef CONFIG_NET_POLL_CONTROLLER
7071 .ndo_poll_controller = ixgbe_netpoll,
7074 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
7075 .ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
7076 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
7077 .ndo_fcoe_enable = ixgbe_fcoe_enable,
7078 .ndo_fcoe_disable = ixgbe_fcoe_disable,
7079 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
7080 #endif /* IXGBE_FCOE */
7081 .ndo_set_features = ixgbe_set_features,
7082 .ndo_fix_features = ixgbe_fix_features,
7085 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
7086 const struct ixgbe_info *ii)
7088 #ifdef CONFIG_PCI_IOV
7089 struct ixgbe_hw *hw = &adapter->hw;
7091 int num_vf_macvlans, i;
7092 struct vf_macvlans *mv_list;
7094 if (hw->mac.type == ixgbe_mac_82598EB || !max_vfs)
7097 /* The 82599 supports up to 64 VFs per physical function
7098 * but this implementation limits allocation to 63 so that
7099 * basic networking resources are still available to the
7102 adapter->num_vfs = (max_vfs > 63) ? 63 : max_vfs;
7103 adapter->flags |= IXGBE_FLAG_SRIOV_ENABLED;
7104 err = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
7106 e_err(probe, "Failed to enable PCI sriov: %d\n", err);
7110 num_vf_macvlans = hw->mac.num_rar_entries -
7111 (IXGBE_MAX_PF_MACVLANS + 1 + adapter->num_vfs);
7113 adapter->mv_list = mv_list = kcalloc(num_vf_macvlans,
7114 sizeof(struct vf_macvlans),
7117 /* Initialize list of VF macvlans */
7118 INIT_LIST_HEAD(&adapter->vf_mvs.l);
7119 for (i = 0; i < num_vf_macvlans; i++) {
7121 mv_list->free = true;
7122 mv_list->rar_entry = hw->mac.num_rar_entries -
7123 (i + adapter->num_vfs + 1);
7124 list_add(&mv_list->l, &adapter->vf_mvs.l);
7129 /* If call to enable VFs succeeded then allocate memory
7130 * for per VF control structures.
7133 kcalloc(adapter->num_vfs,
7134 sizeof(struct vf_data_storage), GFP_KERNEL);
7135 if (adapter->vfinfo) {
7136 /* Now that we're sure SR-IOV is enabled
7137 * and memory allocated set up the mailbox parameters
7139 ixgbe_init_mbx_params_pf(hw);
7140 memcpy(&hw->mbx.ops, ii->mbx_ops,
7141 sizeof(hw->mbx.ops));
7143 /* Disable RSC when in SR-IOV mode */
7144 adapter->flags2 &= ~(IXGBE_FLAG2_RSC_CAPABLE |
7145 IXGBE_FLAG2_RSC_ENABLED);
7150 e_err(probe, "Unable to allocate memory for VF Data Storage - "
7151 "SRIOV disabled\n");
7152 pci_disable_sriov(adapter->pdev);
7155 adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
7156 adapter->num_vfs = 0;
7157 #endif /* CONFIG_PCI_IOV */
7161 * ixgbe_probe - Device Initialization Routine
7162 * @pdev: PCI device information struct
7163 * @ent: entry in ixgbe_pci_tbl
7165 * Returns 0 on success, negative on failure
7167 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
7168 * The OS initialization, configuring of the adapter private structure,
7169 * and a hardware reset occur.
7171 static int __devinit ixgbe_probe(struct pci_dev *pdev,
7172 const struct pci_device_id *ent)
7174 struct net_device *netdev;
7175 struct ixgbe_adapter *adapter = NULL;
7176 struct ixgbe_hw *hw;
7177 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
7178 static int cards_found;
7179 int i, err, pci_using_dac;
7180 u8 part_str[IXGBE_PBANUM_LENGTH];
7181 unsigned int indices = num_possible_cpus();
7187 /* Catch broken hardware that put the wrong VF device ID in
7188 * the PCIe SR-IOV capability.
7190 if (pdev->is_virtfn) {
7191 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
7192 pci_name(pdev), pdev->vendor, pdev->device);
7196 err = pci_enable_device_mem(pdev);
7200 if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
7201 !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
7204 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
7206 err = dma_set_coherent_mask(&pdev->dev,
7210 "No usable DMA configuration, aborting\n");
7217 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
7218 IORESOURCE_MEM), ixgbe_driver_name);
7221 "pci_request_selected_regions failed 0x%x\n", err);
7225 pci_enable_pcie_error_reporting(pdev);
7227 pci_set_master(pdev);
7228 pci_save_state(pdev);
7230 #ifdef CONFIG_IXGBE_DCB
7231 indices *= MAX_TRAFFIC_CLASS;
7234 if (ii->mac == ixgbe_mac_82598EB)
7235 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
7237 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
7240 indices += min_t(unsigned int, num_possible_cpus(),
7241 IXGBE_MAX_FCOE_INDICES);
7243 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
7246 goto err_alloc_etherdev;
7249 SET_NETDEV_DEV(netdev, &pdev->dev);
7251 adapter = netdev_priv(netdev);
7252 pci_set_drvdata(pdev, adapter);
7254 adapter->netdev = netdev;
7255 adapter->pdev = pdev;
7258 adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
7260 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
7261 pci_resource_len(pdev, 0));
7267 for (i = 1; i <= 5; i++) {
7268 if (pci_resource_len(pdev, i) == 0)
7272 netdev->netdev_ops = &ixgbe_netdev_ops;
7273 ixgbe_set_ethtool_ops(netdev);
7274 netdev->watchdog_timeo = 5 * HZ;
7275 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
7277 adapter->bd_number = cards_found;
7280 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
7281 hw->mac.type = ii->mac;
7284 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
7285 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
7286 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
7287 if (!(eec & (1 << 8)))
7288 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
7291 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
7292 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
7293 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
7294 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
7295 hw->phy.mdio.mmds = 0;
7296 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
7297 hw->phy.mdio.dev = netdev;
7298 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
7299 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
7301 ii->get_invariants(hw);
7303 /* setup the private structure */
7304 err = ixgbe_sw_init(adapter);
7308 /* Make it possible the adapter to be woken up via WOL */
7309 switch (adapter->hw.mac.type) {
7310 case ixgbe_mac_82599EB:
7311 case ixgbe_mac_X540:
7312 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7319 * If there is a fan on this device and it has failed log the
7322 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
7323 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
7324 if (esdp & IXGBE_ESDP_SDP1)
7325 e_crit(probe, "Fan has stopped, replace the adapter\n");
7328 /* reset_hw fills in the perm_addr as well */
7329 hw->phy.reset_if_overtemp = true;
7330 err = hw->mac.ops.reset_hw(hw);
7331 hw->phy.reset_if_overtemp = false;
7332 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
7333 hw->mac.type == ixgbe_mac_82598EB) {
7335 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
7336 e_dev_err("failed to load because an unsupported SFP+ "
7337 "module type was detected.\n");
7338 e_dev_err("Reload the driver after installing a supported "
7342 e_dev_err("HW Init failed: %d\n", err);
7346 ixgbe_probe_vf(adapter, ii);
7348 netdev->features = NETIF_F_SG |
7351 NETIF_F_HW_VLAN_TX |
7352 NETIF_F_HW_VLAN_RX |
7353 NETIF_F_HW_VLAN_FILTER |
7359 netdev->hw_features = netdev->features;
7361 switch (adapter->hw.mac.type) {
7362 case ixgbe_mac_82599EB:
7363 case ixgbe_mac_X540:
7364 netdev->features |= NETIF_F_SCTP_CSUM;
7365 netdev->hw_features |= NETIF_F_SCTP_CSUM |
7372 netdev->vlan_features |= NETIF_F_TSO;
7373 netdev->vlan_features |= NETIF_F_TSO6;
7374 netdev->vlan_features |= NETIF_F_IP_CSUM;
7375 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
7376 netdev->vlan_features |= NETIF_F_SG;
7378 netdev->priv_flags |= IFF_UNICAST_FLT;
7380 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7381 adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
7382 IXGBE_FLAG_DCB_ENABLED);
7384 #ifdef CONFIG_IXGBE_DCB
7385 netdev->dcbnl_ops = &dcbnl_ops;
7389 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7390 if (hw->mac.ops.get_device_caps) {
7391 hw->mac.ops.get_device_caps(hw, &device_caps);
7392 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
7393 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
7396 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7397 netdev->vlan_features |= NETIF_F_FCOE_CRC;
7398 netdev->vlan_features |= NETIF_F_FSO;
7399 netdev->vlan_features |= NETIF_F_FCOE_MTU;
7401 #endif /* IXGBE_FCOE */
7402 if (pci_using_dac) {
7403 netdev->features |= NETIF_F_HIGHDMA;
7404 netdev->vlan_features |= NETIF_F_HIGHDMA;
7407 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
7408 netdev->hw_features |= NETIF_F_LRO;
7409 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
7410 netdev->features |= NETIF_F_LRO;
7412 /* make sure the EEPROM is good */
7413 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
7414 e_dev_err("The EEPROM Checksum Is Not Valid\n");
7419 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
7420 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
7422 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
7423 e_dev_err("invalid MAC address\n");
7428 /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
7429 if (hw->mac.ops.disable_tx_laser &&
7430 ((hw->phy.multispeed_fiber) ||
7431 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
7432 (hw->mac.type == ixgbe_mac_82599EB))))
7433 hw->mac.ops.disable_tx_laser(hw);
7435 setup_timer(&adapter->service_timer, &ixgbe_service_timer,
7436 (unsigned long) adapter);
7438 INIT_WORK(&adapter->service_task, ixgbe_service_task);
7439 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
7441 err = ixgbe_init_interrupt_scheme(adapter);
7445 if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED)) {
7446 netdev->hw_features &= ~NETIF_F_RXHASH;
7447 netdev->features &= ~NETIF_F_RXHASH;
7450 switch (pdev->device) {
7451 case IXGBE_DEV_ID_82599_SFP:
7452 /* Only this subdevice supports WOL */
7453 if (pdev->subsystem_device == IXGBE_SUBDEV_ID_82599_SFP)
7454 adapter->wol = IXGBE_WUFC_MAG;
7456 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
7457 /* All except this subdevice support WOL */
7458 if (pdev->subsystem_device != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
7459 adapter->wol = IXGBE_WUFC_MAG;
7461 case IXGBE_DEV_ID_82599_KX4:
7462 adapter->wol = IXGBE_WUFC_MAG;
7468 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
7470 /* pick up the PCI bus settings for reporting later */
7471 hw->mac.ops.get_bus_info(hw);
7473 /* print bus type/speed/width info */
7474 e_dev_info("(PCI Express:%s:%s) %pM\n",
7475 (hw->bus.speed == ixgbe_bus_speed_5000 ? "5.0GT/s" :
7476 hw->bus.speed == ixgbe_bus_speed_2500 ? "2.5GT/s" :
7478 (hw->bus.width == ixgbe_bus_width_pcie_x8 ? "Width x8" :
7479 hw->bus.width == ixgbe_bus_width_pcie_x4 ? "Width x4" :
7480 hw->bus.width == ixgbe_bus_width_pcie_x1 ? "Width x1" :
7484 err = ixgbe_read_pba_string_generic(hw, part_str, IXGBE_PBANUM_LENGTH);
7486 strncpy(part_str, "Unknown", IXGBE_PBANUM_LENGTH);
7487 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
7488 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
7489 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
7492 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
7493 hw->mac.type, hw->phy.type, part_str);
7495 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
7496 e_dev_warn("PCI-Express bandwidth available for this card is "
7497 "not sufficient for optimal performance.\n");
7498 e_dev_warn("For optimal performance a x8 PCI-Express slot "
7502 /* save off EEPROM version number */
7503 hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
7505 /* reset the hardware with the new settings */
7506 err = hw->mac.ops.start_hw(hw);
7508 if (err == IXGBE_ERR_EEPROM_VERSION) {
7509 /* We are running on a pre-production device, log a warning */
7510 e_dev_warn("This device is a pre-production adapter/LOM. "
7511 "Please be aware there may be issues associated "
7512 "with your hardware. If you are experiencing "
7513 "problems please contact your Intel or hardware "
7514 "representative who provided you with this "
7517 strcpy(netdev->name, "eth%d");
7518 err = register_netdev(netdev);
7522 /* carrier off reporting is important to ethtool even BEFORE open */
7523 netif_carrier_off(netdev);
7525 #ifdef CONFIG_IXGBE_DCA
7526 if (dca_add_requester(&pdev->dev) == 0) {
7527 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
7528 ixgbe_setup_dca(adapter);
7531 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7532 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
7533 for (i = 0; i < adapter->num_vfs; i++)
7534 ixgbe_vf_configuration(pdev, (i | 0x10000000));
7537 /* Inform firmware of driver version */
7538 if (hw->mac.ops.set_fw_drv_ver)
7539 hw->mac.ops.set_fw_drv_ver(hw, MAJ, MIN, BUILD,
7542 /* add san mac addr to netdev */
7543 ixgbe_add_sanmac_netdev(netdev);
7545 e_dev_info("Intel(R) 10 Gigabit Network Connection\n");
7550 ixgbe_release_hw_control(adapter);
7551 ixgbe_clear_interrupt_scheme(adapter);
7554 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7555 ixgbe_disable_sriov(adapter);
7556 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
7557 iounmap(hw->hw_addr);
7559 free_netdev(netdev);
7561 pci_release_selected_regions(pdev,
7562 pci_select_bars(pdev, IORESOURCE_MEM));
7565 pci_disable_device(pdev);
7570 * ixgbe_remove - Device Removal Routine
7571 * @pdev: PCI device information struct
7573 * ixgbe_remove is called by the PCI subsystem to alert the driver
7574 * that it should release a PCI device. The could be caused by a
7575 * Hot-Plug event, or because the driver is going to be removed from
7578 static void __devexit ixgbe_remove(struct pci_dev *pdev)
7580 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7581 struct net_device *netdev = adapter->netdev;
7583 set_bit(__IXGBE_DOWN, &adapter->state);
7584 cancel_work_sync(&adapter->service_task);
7586 #ifdef CONFIG_IXGBE_DCA
7587 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
7588 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
7589 dca_remove_requester(&pdev->dev);
7590 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
7595 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
7596 ixgbe_cleanup_fcoe(adapter);
7598 #endif /* IXGBE_FCOE */
7600 /* remove the added san mac */
7601 ixgbe_del_sanmac_netdev(netdev);
7603 if (netdev->reg_state == NETREG_REGISTERED)
7604 unregister_netdev(netdev);
7606 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7607 ixgbe_disable_sriov(adapter);
7609 ixgbe_clear_interrupt_scheme(adapter);
7611 ixgbe_release_hw_control(adapter);
7613 iounmap(adapter->hw.hw_addr);
7614 pci_release_selected_regions(pdev, pci_select_bars(pdev,
7617 e_dev_info("complete\n");
7619 free_netdev(netdev);
7621 pci_disable_pcie_error_reporting(pdev);
7623 pci_disable_device(pdev);
7627 * ixgbe_io_error_detected - called when PCI error is detected
7628 * @pdev: Pointer to PCI device
7629 * @state: The current pci connection state
7631 * This function is called after a PCI bus error affecting
7632 * this device has been detected.
7634 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
7635 pci_channel_state_t state)
7637 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7638 struct net_device *netdev = adapter->netdev;
7640 netif_device_detach(netdev);
7642 if (state == pci_channel_io_perm_failure)
7643 return PCI_ERS_RESULT_DISCONNECT;
7645 if (netif_running(netdev))
7646 ixgbe_down(adapter);
7647 pci_disable_device(pdev);
7649 /* Request a slot reset. */
7650 return PCI_ERS_RESULT_NEED_RESET;
7654 * ixgbe_io_slot_reset - called after the pci bus has been reset.
7655 * @pdev: Pointer to PCI device
7657 * Restart the card from scratch, as if from a cold-boot.
7659 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7661 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7662 pci_ers_result_t result;
7665 if (pci_enable_device_mem(pdev)) {
7666 e_err(probe, "Cannot re-enable PCI device after reset.\n");
7667 result = PCI_ERS_RESULT_DISCONNECT;
7669 pci_set_master(pdev);
7670 pci_restore_state(pdev);
7671 pci_save_state(pdev);
7673 pci_wake_from_d3(pdev, false);
7675 ixgbe_reset(adapter);
7676 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7677 result = PCI_ERS_RESULT_RECOVERED;
7680 err = pci_cleanup_aer_uncorrect_error_status(pdev);
7682 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7683 "failed 0x%0x\n", err);
7684 /* non-fatal, continue */
7691 * ixgbe_io_resume - called when traffic can start flowing again.
7692 * @pdev: Pointer to PCI device
7694 * This callback is called when the error recovery driver tells us that
7695 * its OK to resume normal operation.
7697 static void ixgbe_io_resume(struct pci_dev *pdev)
7699 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7700 struct net_device *netdev = adapter->netdev;
7702 if (netif_running(netdev)) {
7703 if (ixgbe_up(adapter)) {
7704 e_info(probe, "ixgbe_up failed after reset\n");
7709 netif_device_attach(netdev);
7712 static struct pci_error_handlers ixgbe_err_handler = {
7713 .error_detected = ixgbe_io_error_detected,
7714 .slot_reset = ixgbe_io_slot_reset,
7715 .resume = ixgbe_io_resume,
7718 static struct pci_driver ixgbe_driver = {
7719 .name = ixgbe_driver_name,
7720 .id_table = ixgbe_pci_tbl,
7721 .probe = ixgbe_probe,
7722 .remove = __devexit_p(ixgbe_remove),
7724 .suspend = ixgbe_suspend,
7725 .resume = ixgbe_resume,
7727 .shutdown = ixgbe_shutdown,
7728 .err_handler = &ixgbe_err_handler
7732 * ixgbe_init_module - Driver Registration Routine
7734 * ixgbe_init_module is the first routine called when the driver is
7735 * loaded. All it does is register with the PCI subsystem.
7737 static int __init ixgbe_init_module(void)
7740 pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
7741 pr_info("%s\n", ixgbe_copyright);
7743 #ifdef CONFIG_IXGBE_DCA
7744 dca_register_notify(&dca_notifier);
7747 ret = pci_register_driver(&ixgbe_driver);
7751 module_init(ixgbe_init_module);
7754 * ixgbe_exit_module - Driver Exit Cleanup Routine
7756 * ixgbe_exit_module is called just before the driver is removed
7759 static void __exit ixgbe_exit_module(void)
7761 #ifdef CONFIG_IXGBE_DCA
7762 dca_unregister_notify(&dca_notifier);
7764 pci_unregister_driver(&ixgbe_driver);
7765 rcu_barrier(); /* Wait for completion of call_rcu()'s */
7768 #ifdef CONFIG_IXGBE_DCA
7769 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
7774 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
7775 __ixgbe_notify_dca);
7777 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7780 #endif /* CONFIG_IXGBE_DCA */
7782 module_exit(ixgbe_exit_module);