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(rx_ring->dev, cpu);
929 case ixgbe_mac_82599EB:
931 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK_82599;
932 rxctrl |= (dca3_get_tag(rx_ring->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(tx_ring->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(tx_ring->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 void ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter)
2019 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2020 int rxr_remaining = adapter->num_rx_queues, rxr_idx = 0;
2021 int txr_remaining = adapter->num_tx_queues, txr_idx = 0;
2024 /* only one q_vector if MSI-X is disabled. */
2025 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2029 * If we don't have enough vectors for a 1-to-1 mapping, we'll have to
2030 * group them so there are multiple queues per vector.
2032 * Re-adjusting *qpv takes care of the remainder.
2034 for (; v_start < q_vectors && rxr_remaining; v_start++) {
2035 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_start);
2036 for (; rqpv; rqpv--, rxr_idx++, rxr_remaining--)
2037 map_vector_to_rxq(adapter, v_start, rxr_idx);
2041 * If there are not enough q_vectors for each ring to have it's own
2042 * vector then we must pair up Rx/Tx on a each vector
2044 if ((v_start + txr_remaining) > q_vectors)
2047 for (; v_start < q_vectors && txr_remaining; v_start++) {
2048 int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_start);
2049 for (; tqpv; tqpv--, txr_idx++, txr_remaining--)
2050 map_vector_to_txq(adapter, v_start, txr_idx);
2055 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2056 * @adapter: board private structure
2058 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2059 * interrupts from the kernel.
2061 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2063 struct net_device *netdev = adapter->netdev;
2064 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2068 for (vector = 0; vector < q_vectors; vector++) {
2069 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2070 struct msix_entry *entry = &adapter->msix_entries[vector];
2072 if (q_vector->tx.ring && q_vector->rx.ring) {
2073 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2074 "%s-%s-%d", netdev->name, "TxRx", ri++);
2076 } else if (q_vector->rx.ring) {
2077 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2078 "%s-%s-%d", netdev->name, "rx", ri++);
2079 } else if (q_vector->tx.ring) {
2080 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2081 "%s-%s-%d", netdev->name, "tx", ti++);
2083 /* skip this unused q_vector */
2086 err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
2087 q_vector->name, q_vector);
2089 e_err(probe, "request_irq failed for MSIX interrupt "
2090 "Error: %d\n", err);
2091 goto free_queue_irqs;
2093 /* If Flow Director is enabled, set interrupt affinity */
2094 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2095 /* assign the mask for this irq */
2096 irq_set_affinity_hint(entry->vector,
2097 q_vector->affinity_mask);
2101 sprintf(adapter->lsc_int_name, "%s:lsc", netdev->name);
2102 err = request_irq(adapter->msix_entries[vector].vector,
2103 ixgbe_msix_lsc, 0, adapter->lsc_int_name, adapter);
2105 e_err(probe, "request_irq for msix_lsc failed: %d\n", err);
2106 goto free_queue_irqs;
2114 irq_set_affinity_hint(adapter->msix_entries[vector].vector,
2116 free_irq(adapter->msix_entries[vector].vector,
2117 adapter->q_vector[vector]);
2119 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2120 pci_disable_msix(adapter->pdev);
2121 kfree(adapter->msix_entries);
2122 adapter->msix_entries = NULL;
2127 * ixgbe_irq_enable - Enable default interrupt generation settings
2128 * @adapter: board private structure
2130 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2135 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2136 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2137 mask |= IXGBE_EIMS_GPI_SDP0;
2138 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2139 mask |= IXGBE_EIMS_GPI_SDP1;
2140 switch (adapter->hw.mac.type) {
2141 case ixgbe_mac_82599EB:
2142 case ixgbe_mac_X540:
2143 mask |= IXGBE_EIMS_ECC;
2144 mask |= IXGBE_EIMS_GPI_SDP1;
2145 mask |= IXGBE_EIMS_GPI_SDP2;
2146 if (adapter->num_vfs)
2147 mask |= IXGBE_EIMS_MAILBOX;
2152 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)
2153 mask |= IXGBE_EIMS_FLOW_DIR;
2155 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2157 ixgbe_irq_enable_queues(adapter, ~0);
2159 IXGBE_WRITE_FLUSH(&adapter->hw);
2161 if (adapter->num_vfs > 32) {
2162 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
2163 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2168 * ixgbe_intr - legacy mode Interrupt Handler
2169 * @irq: interrupt number
2170 * @data: pointer to a network interface device structure
2172 static irqreturn_t ixgbe_intr(int irq, void *data)
2174 struct ixgbe_adapter *adapter = data;
2175 struct ixgbe_hw *hw = &adapter->hw;
2176 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2180 * Workaround for silicon errata on 82598. Mask the interrupts
2181 * before the read of EICR.
2183 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2185 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2186 * therefore no explict interrupt disable is necessary */
2187 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2190 * shared interrupt alert!
2191 * make sure interrupts are enabled because the read will
2192 * have disabled interrupts due to EIAM
2193 * finish the workaround of silicon errata on 82598. Unmask
2194 * the interrupt that we masked before the EICR read.
2196 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2197 ixgbe_irq_enable(adapter, true, true);
2198 return IRQ_NONE; /* Not our interrupt */
2201 if (eicr & IXGBE_EICR_LSC)
2202 ixgbe_check_lsc(adapter);
2204 switch (hw->mac.type) {
2205 case ixgbe_mac_82599EB:
2206 ixgbe_check_sfp_event(adapter, eicr);
2207 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2208 ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC))) {
2209 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2210 adapter->interrupt_event = eicr;
2211 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2212 ixgbe_service_event_schedule(adapter);
2220 ixgbe_check_fan_failure(adapter, eicr);
2222 if (napi_schedule_prep(&(q_vector->napi))) {
2223 /* would disable interrupts here but EIAM disabled it */
2224 __napi_schedule(&(q_vector->napi));
2228 * re-enable link(maybe) and non-queue interrupts, no flush.
2229 * ixgbe_poll will re-enable the queue interrupts
2232 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2233 ixgbe_irq_enable(adapter, false, false);
2238 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
2240 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2243 /* legacy and MSI only use one vector */
2244 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2247 for (i = 0; i < adapter->num_rx_queues; i++) {
2248 adapter->rx_ring[i]->q_vector = NULL;
2249 adapter->rx_ring[i]->next = NULL;
2251 for (i = 0; i < adapter->num_tx_queues; i++) {
2252 adapter->tx_ring[i]->q_vector = NULL;
2253 adapter->tx_ring[i]->next = NULL;
2256 for (i = 0; i < q_vectors; i++) {
2257 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
2258 memset(&q_vector->rx, 0, sizeof(struct ixgbe_ring_container));
2259 memset(&q_vector->tx, 0, sizeof(struct ixgbe_ring_container));
2264 * ixgbe_request_irq - initialize interrupts
2265 * @adapter: board private structure
2267 * Attempts to configure interrupts using the best available
2268 * capabilities of the hardware and kernel.
2270 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2272 struct net_device *netdev = adapter->netdev;
2275 /* map all of the rings to the q_vectors */
2276 ixgbe_map_rings_to_vectors(adapter);
2278 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2279 err = ixgbe_request_msix_irqs(adapter);
2280 else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
2281 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2282 netdev->name, adapter);
2284 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2285 netdev->name, adapter);
2288 e_err(probe, "request_irq failed, Error %d\n", err);
2290 /* place q_vectors and rings back into a known good state */
2291 ixgbe_reset_q_vectors(adapter);
2297 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2299 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2302 q_vectors = adapter->num_msix_vectors;
2304 free_irq(adapter->msix_entries[i].vector, adapter);
2307 for (; i >= 0; i--) {
2308 /* free only the irqs that were actually requested */
2309 if (!adapter->q_vector[i]->rx.ring &&
2310 !adapter->q_vector[i]->tx.ring)
2313 /* clear the affinity_mask in the IRQ descriptor */
2314 irq_set_affinity_hint(adapter->msix_entries[i].vector,
2317 free_irq(adapter->msix_entries[i].vector,
2318 adapter->q_vector[i]);
2321 free_irq(adapter->pdev->irq, adapter);
2324 /* clear q_vector state information */
2325 ixgbe_reset_q_vectors(adapter);
2329 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2330 * @adapter: board private structure
2332 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2334 switch (adapter->hw.mac.type) {
2335 case ixgbe_mac_82598EB:
2336 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2338 case ixgbe_mac_82599EB:
2339 case ixgbe_mac_X540:
2340 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2341 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2342 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2343 if (adapter->num_vfs > 32)
2344 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
2349 IXGBE_WRITE_FLUSH(&adapter->hw);
2350 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2352 for (i = 0; i < adapter->num_msix_vectors; i++)
2353 synchronize_irq(adapter->msix_entries[i].vector);
2355 synchronize_irq(adapter->pdev->irq);
2360 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2363 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2365 struct ixgbe_hw *hw = &adapter->hw;
2367 IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
2368 EITR_INTS_PER_SEC_TO_REG(adapter->rx_eitr_param));
2370 ixgbe_set_ivar(adapter, 0, 0, 0);
2371 ixgbe_set_ivar(adapter, 1, 0, 0);
2373 e_info(hw, "Legacy interrupt IVAR setup done\n");
2377 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2378 * @adapter: board private structure
2379 * @ring: structure containing ring specific data
2381 * Configure the Tx descriptor ring after a reset.
2383 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2384 struct ixgbe_ring *ring)
2386 struct ixgbe_hw *hw = &adapter->hw;
2387 u64 tdba = ring->dma;
2390 u8 reg_idx = ring->reg_idx;
2392 /* disable queue to avoid issues while updating state */
2393 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2394 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx),
2395 txdctl & ~IXGBE_TXDCTL_ENABLE);
2396 IXGBE_WRITE_FLUSH(hw);
2398 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2399 (tdba & DMA_BIT_MASK(32)));
2400 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2401 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2402 ring->count * sizeof(union ixgbe_adv_tx_desc));
2403 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2404 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2405 ring->tail = hw->hw_addr + IXGBE_TDT(reg_idx);
2407 /* configure fetching thresholds */
2408 if (adapter->rx_itr_setting == 0) {
2409 /* cannot set wthresh when itr==0 */
2410 txdctl &= ~0x007F0000;
2412 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
2413 txdctl |= (8 << 16);
2415 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2416 /* PThresh workaround for Tx hang with DFP enabled. */
2420 /* reinitialize flowdirector state */
2421 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2422 adapter->atr_sample_rate) {
2423 ring->atr_sample_rate = adapter->atr_sample_rate;
2424 ring->atr_count = 0;
2425 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
2427 ring->atr_sample_rate = 0;
2430 clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
2433 txdctl |= IXGBE_TXDCTL_ENABLE;
2434 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
2436 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2437 if (hw->mac.type == ixgbe_mac_82598EB &&
2438 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2441 /* poll to verify queue is enabled */
2443 usleep_range(1000, 2000);
2444 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2445 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
2447 e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
2450 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
2452 struct ixgbe_hw *hw = &adapter->hw;
2455 u8 tcs = netdev_get_num_tc(adapter->netdev);
2457 if (hw->mac.type == ixgbe_mac_82598EB)
2460 /* disable the arbiter while setting MTQC */
2461 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2462 rttdcs |= IXGBE_RTTDCS_ARBDIS;
2463 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2465 /* set transmit pool layout */
2466 switch (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
2467 case (IXGBE_FLAG_SRIOV_ENABLED):
2468 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2469 (IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF));
2473 reg = IXGBE_MTQC_64Q_1PB;
2475 reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
2477 reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
2479 IXGBE_WRITE_REG(hw, IXGBE_MTQC, reg);
2481 /* Enable Security TX Buffer IFG for multiple pb */
2483 reg = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
2484 reg |= IXGBE_SECTX_DCB;
2485 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, reg);
2490 /* re-enable the arbiter */
2491 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2492 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2496 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2497 * @adapter: board private structure
2499 * Configure the Tx unit of the MAC after a reset.
2501 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2503 struct ixgbe_hw *hw = &adapter->hw;
2507 ixgbe_setup_mtqc(adapter);
2509 if (hw->mac.type != ixgbe_mac_82598EB) {
2510 /* DMATXCTL.EN must be before Tx queues are enabled */
2511 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2512 dmatxctl |= IXGBE_DMATXCTL_TE;
2513 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2516 /* Setup the HW Tx Head and Tail descriptor pointers */
2517 for (i = 0; i < adapter->num_tx_queues; i++)
2518 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
2521 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2523 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2524 struct ixgbe_ring *rx_ring)
2527 u8 reg_idx = rx_ring->reg_idx;
2529 switch (adapter->hw.mac.type) {
2530 case ixgbe_mac_82598EB: {
2531 struct ixgbe_ring_feature *feature = adapter->ring_feature;
2532 const int mask = feature[RING_F_RSS].mask;
2533 reg_idx = reg_idx & mask;
2536 case ixgbe_mac_82599EB:
2537 case ixgbe_mac_X540:
2542 srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx));
2544 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
2545 srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
2546 if (adapter->num_vfs)
2547 srrctl |= IXGBE_SRRCTL_DROP_EN;
2549 srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
2550 IXGBE_SRRCTL_BSIZEHDR_MASK;
2552 if (ring_is_ps_enabled(rx_ring)) {
2553 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
2554 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2556 srrctl |= (PAGE_SIZE / 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2558 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2560 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
2561 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2562 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2565 IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx), srrctl);
2568 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2570 struct ixgbe_hw *hw = &adapter->hw;
2571 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2572 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2573 0x6A3E67EA, 0x14364D17, 0x3BED200D};
2574 u32 mrqc = 0, reta = 0;
2577 u8 tcs = netdev_get_num_tc(adapter->netdev);
2578 int maxq = adapter->ring_feature[RING_F_RSS].indices;
2581 maxq = min(maxq, adapter->num_tx_queues / tcs);
2583 /* Fill out hash function seeds */
2584 for (i = 0; i < 10; i++)
2585 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2587 /* Fill out redirection table */
2588 for (i = 0, j = 0; i < 128; i++, j++) {
2591 /* reta = 4-byte sliding window of
2592 * 0x00..(indices-1)(indices-1)00..etc. */
2593 reta = (reta << 8) | (j * 0x11);
2595 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2598 /* Disable indicating checksum in descriptor, enables RSS hash */
2599 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2600 rxcsum |= IXGBE_RXCSUM_PCSD;
2601 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2603 if (adapter->hw.mac.type == ixgbe_mac_82598EB &&
2604 (adapter->flags & IXGBE_FLAG_RSS_ENABLED)) {
2605 mrqc = IXGBE_MRQC_RSSEN;
2607 int mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2608 | IXGBE_FLAG_SRIOV_ENABLED);
2611 case (IXGBE_FLAG_RSS_ENABLED):
2613 mrqc = IXGBE_MRQC_RSSEN;
2615 mrqc = IXGBE_MRQC_RTRSS4TCEN;
2617 mrqc = IXGBE_MRQC_RTRSS8TCEN;
2619 case (IXGBE_FLAG_SRIOV_ENABLED):
2620 mrqc = IXGBE_MRQC_VMDQEN;
2627 /* Perform hash on these packet types */
2628 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2629 | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2630 | IXGBE_MRQC_RSS_FIELD_IPV6
2631 | IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
2633 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2637 * ixgbe_configure_rscctl - enable RSC for the indicated ring
2638 * @adapter: address of board private structure
2639 * @index: index of ring to set
2641 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
2642 struct ixgbe_ring *ring)
2644 struct ixgbe_hw *hw = &adapter->hw;
2647 u8 reg_idx = ring->reg_idx;
2649 if (!ring_is_rsc_enabled(ring))
2652 rx_buf_len = ring->rx_buf_len;
2653 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
2654 rscctrl |= IXGBE_RSCCTL_RSCEN;
2656 * we must limit the number of descriptors so that the
2657 * total size of max desc * buf_len is not greater
2660 if (ring_is_ps_enabled(ring)) {
2661 #if (MAX_SKB_FRAGS > 16)
2662 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2663 #elif (MAX_SKB_FRAGS > 8)
2664 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2665 #elif (MAX_SKB_FRAGS > 4)
2666 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2668 rscctrl |= IXGBE_RSCCTL_MAXDESC_1;
2671 if (rx_buf_len < IXGBE_RXBUFFER_4096)
2672 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2673 else if (rx_buf_len < IXGBE_RXBUFFER_8192)
2674 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2676 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2678 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
2682 * ixgbe_set_uta - Set unicast filter table address
2683 * @adapter: board private structure
2685 * The unicast table address is a register array of 32-bit registers.
2686 * The table is meant to be used in a way similar to how the MTA is used
2687 * however due to certain limitations in the hardware it is necessary to
2688 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
2689 * enable bit to allow vlan tag stripping when promiscuous mode is enabled
2691 static void ixgbe_set_uta(struct ixgbe_adapter *adapter)
2693 struct ixgbe_hw *hw = &adapter->hw;
2696 /* The UTA table only exists on 82599 hardware and newer */
2697 if (hw->mac.type < ixgbe_mac_82599EB)
2700 /* we only need to do this if VMDq is enabled */
2701 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
2704 for (i = 0; i < 128; i++)
2705 IXGBE_WRITE_REG(hw, IXGBE_UTA(i), ~0);
2708 #define IXGBE_MAX_RX_DESC_POLL 10
2709 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
2710 struct ixgbe_ring *ring)
2712 struct ixgbe_hw *hw = &adapter->hw;
2713 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
2715 u8 reg_idx = ring->reg_idx;
2717 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2718 if (hw->mac.type == ixgbe_mac_82598EB &&
2719 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2723 usleep_range(1000, 2000);
2724 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2725 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
2728 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
2729 "the polling period\n", reg_idx);
2733 void ixgbe_disable_rx_queue(struct ixgbe_adapter *adapter,
2734 struct ixgbe_ring *ring)
2736 struct ixgbe_hw *hw = &adapter->hw;
2737 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
2739 u8 reg_idx = ring->reg_idx;
2741 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2742 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
2744 /* write value back with RXDCTL.ENABLE bit cleared */
2745 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
2747 if (hw->mac.type == ixgbe_mac_82598EB &&
2748 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2751 /* the hardware may take up to 100us to really disable the rx queue */
2754 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2755 } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
2758 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not cleared within "
2759 "the polling period\n", reg_idx);
2763 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
2764 struct ixgbe_ring *ring)
2766 struct ixgbe_hw *hw = &adapter->hw;
2767 u64 rdba = ring->dma;
2769 u8 reg_idx = ring->reg_idx;
2771 /* disable queue to avoid issues while updating state */
2772 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2773 ixgbe_disable_rx_queue(adapter, ring);
2775 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
2776 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
2777 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
2778 ring->count * sizeof(union ixgbe_adv_rx_desc));
2779 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
2780 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
2781 ring->tail = hw->hw_addr + IXGBE_RDT(reg_idx);
2783 ixgbe_configure_srrctl(adapter, ring);
2784 ixgbe_configure_rscctl(adapter, ring);
2786 /* If operating in IOV mode set RLPML for X540 */
2787 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
2788 hw->mac.type == ixgbe_mac_X540) {
2789 rxdctl &= ~IXGBE_RXDCTL_RLPMLMASK;
2790 rxdctl |= ((ring->netdev->mtu + ETH_HLEN +
2791 ETH_FCS_LEN + VLAN_HLEN) | IXGBE_RXDCTL_RLPML_EN);
2794 if (hw->mac.type == ixgbe_mac_82598EB) {
2796 * enable cache line friendly hardware writes:
2797 * PTHRESH=32 descriptors (half the internal cache),
2798 * this also removes ugly rx_no_buffer_count increment
2799 * HTHRESH=4 descriptors (to minimize latency on fetch)
2800 * WTHRESH=8 burst writeback up to two cache lines
2802 rxdctl &= ~0x3FFFFF;
2806 /* enable receive descriptor ring */
2807 rxdctl |= IXGBE_RXDCTL_ENABLE;
2808 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
2810 ixgbe_rx_desc_queue_enable(adapter, ring);
2811 ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
2814 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
2816 struct ixgbe_hw *hw = &adapter->hw;
2819 /* PSRTYPE must be initialized in non 82598 adapters */
2820 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
2821 IXGBE_PSRTYPE_UDPHDR |
2822 IXGBE_PSRTYPE_IPV4HDR |
2823 IXGBE_PSRTYPE_L2HDR |
2824 IXGBE_PSRTYPE_IPV6HDR;
2826 if (hw->mac.type == ixgbe_mac_82598EB)
2829 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED)
2830 psrtype |= (adapter->num_rx_queues_per_pool << 29);
2832 for (p = 0; p < adapter->num_rx_pools; p++)
2833 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(adapter->num_vfs + p),
2837 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
2839 struct ixgbe_hw *hw = &adapter->hw;
2842 u32 reg_offset, vf_shift;
2845 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
2848 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
2849 vt_reg_bits = IXGBE_VMD_CTL_VMDQ_EN | IXGBE_VT_CTL_REPLEN;
2850 vt_reg_bits |= (adapter->num_vfs << IXGBE_VT_CTL_POOL_SHIFT);
2851 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl | vt_reg_bits);
2853 vf_shift = adapter->num_vfs % 32;
2854 reg_offset = (adapter->num_vfs > 32) ? 1 : 0;
2856 /* Enable only the PF's pool for Tx/Rx */
2857 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (1 << vf_shift));
2858 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), 0);
2859 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (1 << vf_shift));
2860 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), 0);
2861 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
2863 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
2864 hw->mac.ops.set_vmdq(hw, 0, adapter->num_vfs);
2867 * Set up VF register offsets for selected VT Mode,
2868 * i.e. 32 or 64 VFs for SR-IOV
2870 gcr_ext = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
2871 gcr_ext |= IXGBE_GCR_EXT_MSIX_EN;
2872 gcr_ext |= IXGBE_GCR_EXT_VT_MODE_64;
2873 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
2875 /* enable Tx loopback for VF/PF communication */
2876 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
2877 /* Enable MAC Anti-Spoofing */
2878 hw->mac.ops.set_mac_anti_spoofing(hw,
2879 (adapter->antispoofing_enabled =
2880 (adapter->num_vfs != 0)),
2884 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
2886 struct ixgbe_hw *hw = &adapter->hw;
2887 struct net_device *netdev = adapter->netdev;
2888 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2890 struct ixgbe_ring *rx_ring;
2894 /* Decide whether to use packet split mode or not */
2896 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
2898 /* Do not use packet split if we're in SR-IOV Mode */
2899 if (adapter->num_vfs)
2900 adapter->flags &= ~IXGBE_FLAG_RX_PS_ENABLED;
2902 /* Disable packet split due to 82599 erratum #45 */
2903 if (hw->mac.type == ixgbe_mac_82599EB)
2904 adapter->flags &= ~IXGBE_FLAG_RX_PS_ENABLED;
2906 /* Set the RX buffer length according to the mode */
2907 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
2908 rx_buf_len = IXGBE_RX_HDR_SIZE;
2910 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
2911 (netdev->mtu <= ETH_DATA_LEN))
2912 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
2914 rx_buf_len = ALIGN(max_frame + VLAN_HLEN, 1024);
2918 /* adjust max frame to be able to do baby jumbo for FCoE */
2919 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
2920 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
2921 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
2923 #endif /* IXGBE_FCOE */
2924 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
2925 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
2926 mhadd &= ~IXGBE_MHADD_MFS_MASK;
2927 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
2929 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
2932 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2933 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
2934 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2935 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
2938 * Setup the HW Rx Head and Tail Descriptor Pointers and
2939 * the Base and Length of the Rx Descriptor Ring
2941 for (i = 0; i < adapter->num_rx_queues; i++) {
2942 rx_ring = adapter->rx_ring[i];
2943 rx_ring->rx_buf_len = rx_buf_len;
2945 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)
2946 set_ring_ps_enabled(rx_ring);
2948 clear_ring_ps_enabled(rx_ring);
2950 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
2951 set_ring_rsc_enabled(rx_ring);
2953 clear_ring_rsc_enabled(rx_ring);
2956 if (netdev->features & NETIF_F_FCOE_MTU) {
2957 struct ixgbe_ring_feature *f;
2958 f = &adapter->ring_feature[RING_F_FCOE];
2959 if ((i >= f->mask) && (i < f->mask + f->indices)) {
2960 clear_ring_ps_enabled(rx_ring);
2961 if (rx_buf_len < IXGBE_FCOE_JUMBO_FRAME_SIZE)
2962 rx_ring->rx_buf_len =
2963 IXGBE_FCOE_JUMBO_FRAME_SIZE;
2964 } else if (!ring_is_rsc_enabled(rx_ring) &&
2965 !ring_is_ps_enabled(rx_ring)) {
2966 rx_ring->rx_buf_len =
2967 IXGBE_FCOE_JUMBO_FRAME_SIZE;
2970 #endif /* IXGBE_FCOE */
2974 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
2976 struct ixgbe_hw *hw = &adapter->hw;
2977 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2979 switch (hw->mac.type) {
2980 case ixgbe_mac_82598EB:
2982 * For VMDq support of different descriptor types or
2983 * buffer sizes through the use of multiple SRRCTL
2984 * registers, RDRXCTL.MVMEN must be set to 1
2986 * also, the manual doesn't mention it clearly but DCA hints
2987 * will only use queue 0's tags unless this bit is set. Side
2988 * effects of setting this bit are only that SRRCTL must be
2989 * fully programmed [0..15]
2991 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
2993 case ixgbe_mac_82599EB:
2994 case ixgbe_mac_X540:
2995 /* Disable RSC for ACK packets */
2996 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
2997 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
2998 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
2999 /* hardware requires some bits to be set by default */
3000 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3001 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3004 /* We should do nothing since we don't know this hardware */
3008 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3012 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3013 * @adapter: board private structure
3015 * Configure the Rx unit of the MAC after a reset.
3017 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3019 struct ixgbe_hw *hw = &adapter->hw;
3023 /* disable receives while setting up the descriptors */
3024 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3025 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3027 ixgbe_setup_psrtype(adapter);
3028 ixgbe_setup_rdrxctl(adapter);
3030 /* Program registers for the distribution of queues */
3031 ixgbe_setup_mrqc(adapter);
3033 ixgbe_set_uta(adapter);
3035 /* set_rx_buffer_len must be called before ring initialization */
3036 ixgbe_set_rx_buffer_len(adapter);
3039 * Setup the HW Rx Head and Tail Descriptor Pointers and
3040 * the Base and Length of the Rx Descriptor Ring
3042 for (i = 0; i < adapter->num_rx_queues; i++)
3043 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3045 /* disable drop enable for 82598 parts */
3046 if (hw->mac.type == ixgbe_mac_82598EB)
3047 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3049 /* enable all receives */
3050 rxctrl |= IXGBE_RXCTRL_RXEN;
3051 hw->mac.ops.enable_rx_dma(hw, rxctrl);
3054 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3056 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3057 struct ixgbe_hw *hw = &adapter->hw;
3058 int pool_ndx = adapter->num_vfs;
3060 /* add VID to filter table */
3061 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, true);
3062 set_bit(vid, adapter->active_vlans);
3065 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3067 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3068 struct ixgbe_hw *hw = &adapter->hw;
3069 int pool_ndx = adapter->num_vfs;
3071 /* remove VID from filter table */
3072 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, false);
3073 clear_bit(vid, adapter->active_vlans);
3077 * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3078 * @adapter: driver data
3080 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
3082 struct ixgbe_hw *hw = &adapter->hw;
3085 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3086 vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
3087 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3091 * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3092 * @adapter: driver data
3094 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
3096 struct ixgbe_hw *hw = &adapter->hw;
3099 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3100 vlnctrl |= IXGBE_VLNCTRL_VFE;
3101 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3102 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3106 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3107 * @adapter: driver data
3109 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3111 struct ixgbe_hw *hw = &adapter->hw;
3115 switch (hw->mac.type) {
3116 case ixgbe_mac_82598EB:
3117 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3118 vlnctrl &= ~IXGBE_VLNCTRL_VME;
3119 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3121 case ixgbe_mac_82599EB:
3122 case ixgbe_mac_X540:
3123 for (i = 0; i < adapter->num_rx_queues; i++) {
3124 j = adapter->rx_ring[i]->reg_idx;
3125 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3126 vlnctrl &= ~IXGBE_RXDCTL_VME;
3127 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3136 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3137 * @adapter: driver data
3139 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3141 struct ixgbe_hw *hw = &adapter->hw;
3145 switch (hw->mac.type) {
3146 case ixgbe_mac_82598EB:
3147 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3148 vlnctrl |= IXGBE_VLNCTRL_VME;
3149 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3151 case ixgbe_mac_82599EB:
3152 case ixgbe_mac_X540:
3153 for (i = 0; i < adapter->num_rx_queues; i++) {
3154 j = adapter->rx_ring[i]->reg_idx;
3155 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3156 vlnctrl |= IXGBE_RXDCTL_VME;
3157 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3165 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3169 ixgbe_vlan_rx_add_vid(adapter->netdev, 0);
3171 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3172 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
3176 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3177 * @netdev: network interface device structure
3179 * Writes unicast address list to the RAR table.
3180 * Returns: -ENOMEM on failure/insufficient address space
3181 * 0 on no addresses written
3182 * X on writing X addresses to the RAR table
3184 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3186 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3187 struct ixgbe_hw *hw = &adapter->hw;
3188 unsigned int vfn = adapter->num_vfs;
3189 unsigned int rar_entries = IXGBE_MAX_PF_MACVLANS;
3192 /* return ENOMEM indicating insufficient memory for addresses */
3193 if (netdev_uc_count(netdev) > rar_entries)
3196 if (!netdev_uc_empty(netdev) && rar_entries) {
3197 struct netdev_hw_addr *ha;
3198 /* return error if we do not support writing to RAR table */
3199 if (!hw->mac.ops.set_rar)
3202 netdev_for_each_uc_addr(ha, netdev) {
3205 hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3210 /* write the addresses in reverse order to avoid write combining */
3211 for (; rar_entries > 0 ; rar_entries--)
3212 hw->mac.ops.clear_rar(hw, rar_entries);
3218 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3219 * @netdev: network interface device structure
3221 * The set_rx_method entry point is called whenever the unicast/multicast
3222 * address list or the network interface flags are updated. This routine is
3223 * responsible for configuring the hardware for proper unicast, multicast and
3226 void ixgbe_set_rx_mode(struct net_device *netdev)
3228 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3229 struct ixgbe_hw *hw = &adapter->hw;
3230 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3233 /* Check for Promiscuous and All Multicast modes */
3235 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3237 /* set all bits that we expect to always be set */
3238 fctrl |= IXGBE_FCTRL_BAM;
3239 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3240 fctrl |= IXGBE_FCTRL_PMCF;
3242 /* clear the bits we are changing the status of */
3243 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3245 if (netdev->flags & IFF_PROMISC) {
3246 hw->addr_ctrl.user_set_promisc = true;
3247 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3248 vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3249 /* don't hardware filter vlans in promisc mode */
3250 ixgbe_vlan_filter_disable(adapter);
3252 if (netdev->flags & IFF_ALLMULTI) {
3253 fctrl |= IXGBE_FCTRL_MPE;
3254 vmolr |= IXGBE_VMOLR_MPE;
3257 * Write addresses to the MTA, if the attempt fails
3258 * then we should just turn on promiscuous mode so
3259 * that we can at least receive multicast traffic
3261 hw->mac.ops.update_mc_addr_list(hw, netdev);
3262 vmolr |= IXGBE_VMOLR_ROMPE;
3264 ixgbe_vlan_filter_enable(adapter);
3265 hw->addr_ctrl.user_set_promisc = false;
3267 * Write addresses to available RAR registers, if there is not
3268 * sufficient space to store all the addresses then enable
3269 * unicast promiscuous mode
3271 count = ixgbe_write_uc_addr_list(netdev);
3273 fctrl |= IXGBE_FCTRL_UPE;
3274 vmolr |= IXGBE_VMOLR_ROPE;
3278 if (adapter->num_vfs) {
3279 ixgbe_restore_vf_multicasts(adapter);
3280 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(adapter->num_vfs)) &
3281 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3283 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(adapter->num_vfs), vmolr);
3286 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3288 if (netdev->features & NETIF_F_HW_VLAN_RX)
3289 ixgbe_vlan_strip_enable(adapter);
3291 ixgbe_vlan_strip_disable(adapter);
3294 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3297 struct ixgbe_q_vector *q_vector;
3298 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3300 /* legacy and MSI only use one vector */
3301 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3304 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3305 q_vector = adapter->q_vector[q_idx];
3306 napi_enable(&q_vector->napi);
3310 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3313 struct ixgbe_q_vector *q_vector;
3314 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3316 /* legacy and MSI only use one vector */
3317 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3320 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3321 q_vector = adapter->q_vector[q_idx];
3322 napi_disable(&q_vector->napi);
3326 #ifdef CONFIG_IXGBE_DCB
3328 * ixgbe_configure_dcb - Configure DCB hardware
3329 * @adapter: ixgbe adapter struct
3331 * This is called by the driver on open to configure the DCB hardware.
3332 * This is also called by the gennetlink interface when reconfiguring
3335 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3337 struct ixgbe_hw *hw = &adapter->hw;
3338 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3340 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3341 if (hw->mac.type == ixgbe_mac_82598EB)
3342 netif_set_gso_max_size(adapter->netdev, 65536);
3346 if (hw->mac.type == ixgbe_mac_82598EB)
3347 netif_set_gso_max_size(adapter->netdev, 32768);
3350 /* Enable VLAN tag insert/strip */
3351 adapter->netdev->features |= NETIF_F_HW_VLAN_RX;
3353 hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
3355 /* reconfigure the hardware */
3356 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
3358 if (adapter->netdev->features & NETIF_F_FCOE_MTU)
3359 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
3361 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3363 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3365 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
3367 struct net_device *dev = adapter->netdev;
3369 if (adapter->ixgbe_ieee_ets)
3370 dev->dcbnl_ops->ieee_setets(dev,
3371 adapter->ixgbe_ieee_ets);
3372 if (adapter->ixgbe_ieee_pfc)
3373 dev->dcbnl_ops->ieee_setpfc(dev,
3374 adapter->ixgbe_ieee_pfc);
3377 /* Enable RSS Hash per TC */
3378 if (hw->mac.type != ixgbe_mac_82598EB) {
3382 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
3384 u8 cnt = adapter->netdev->tc_to_txq[i].count;
3389 reg |= msb << IXGBE_RQTC_SHIFT_TC(i);
3391 IXGBE_WRITE_REG(hw, IXGBE_RQTC, reg);
3397 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
3400 int num_tc = netdev_get_num_tc(adapter->netdev);
3401 struct ixgbe_hw *hw = &adapter->hw;
3403 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3404 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3405 hdrm = 64 << adapter->fdir_pballoc;
3407 hw->mac.ops.set_rxpba(&adapter->hw, num_tc, hdrm, PBA_STRATEGY_EQUAL);
3410 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
3412 struct ixgbe_hw *hw = &adapter->hw;
3413 struct hlist_node *node, *node2;
3414 struct ixgbe_fdir_filter *filter;
3416 spin_lock(&adapter->fdir_perfect_lock);
3418 if (!hlist_empty(&adapter->fdir_filter_list))
3419 ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
3421 hlist_for_each_entry_safe(filter, node, node2,
3422 &adapter->fdir_filter_list, fdir_node) {
3423 ixgbe_fdir_write_perfect_filter_82599(hw,
3426 (filter->action == IXGBE_FDIR_DROP_QUEUE) ?
3427 IXGBE_FDIR_DROP_QUEUE :
3428 adapter->rx_ring[filter->action]->reg_idx);
3431 spin_unlock(&adapter->fdir_perfect_lock);
3434 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3436 struct net_device *netdev = adapter->netdev;
3437 struct ixgbe_hw *hw = &adapter->hw;
3440 ixgbe_configure_pb(adapter);
3441 #ifdef CONFIG_IXGBE_DCB
3442 ixgbe_configure_dcb(adapter);
3445 ixgbe_set_rx_mode(netdev);
3446 ixgbe_restore_vlan(adapter);
3449 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
3450 ixgbe_configure_fcoe(adapter);
3452 #endif /* IXGBE_FCOE */
3453 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3454 for (i = 0; i < adapter->num_tx_queues; i++)
3455 adapter->tx_ring[i]->atr_sample_rate =
3456 adapter->atr_sample_rate;
3457 ixgbe_init_fdir_signature_82599(hw, adapter->fdir_pballoc);
3458 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3459 ixgbe_init_fdir_perfect_82599(&adapter->hw,
3460 adapter->fdir_pballoc);
3461 ixgbe_fdir_filter_restore(adapter);
3463 ixgbe_configure_virtualization(adapter);
3465 ixgbe_configure_tx(adapter);
3466 ixgbe_configure_rx(adapter);
3469 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3471 switch (hw->phy.type) {
3472 case ixgbe_phy_sfp_avago:
3473 case ixgbe_phy_sfp_ftl:
3474 case ixgbe_phy_sfp_intel:
3475 case ixgbe_phy_sfp_unknown:
3476 case ixgbe_phy_sfp_passive_tyco:
3477 case ixgbe_phy_sfp_passive_unknown:
3478 case ixgbe_phy_sfp_active_unknown:
3479 case ixgbe_phy_sfp_ftl_active:
3487 * ixgbe_sfp_link_config - set up SFP+ link
3488 * @adapter: pointer to private adapter struct
3490 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3493 * We are assuming the worst case scenerio here, and that
3494 * is that an SFP was inserted/removed after the reset
3495 * but before SFP detection was enabled. As such the best
3496 * solution is to just start searching as soon as we start
3498 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3499 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
3501 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
3505 * ixgbe_non_sfp_link_config - set up non-SFP+ link
3506 * @hw: pointer to private hardware struct
3508 * Returns 0 on success, negative on failure
3510 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3513 bool negotiation, link_up = false;
3514 u32 ret = IXGBE_ERR_LINK_SETUP;
3516 if (hw->mac.ops.check_link)
3517 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3522 autoneg = hw->phy.autoneg_advertised;
3523 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
3524 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
3529 if (hw->mac.ops.setup_link)
3530 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
3535 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
3537 struct ixgbe_hw *hw = &adapter->hw;
3540 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3541 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
3543 gpie |= IXGBE_GPIE_EIAME;
3545 * use EIAM to auto-mask when MSI-X interrupt is asserted
3546 * this saves a register write for every interrupt
3548 switch (hw->mac.type) {
3549 case ixgbe_mac_82598EB:
3550 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3552 case ixgbe_mac_82599EB:
3553 case ixgbe_mac_X540:
3555 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
3556 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
3560 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
3561 * specifically only auto mask tx and rx interrupts */
3562 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3565 /* XXX: to interrupt immediately for EICS writes, enable this */
3566 /* gpie |= IXGBE_GPIE_EIMEN; */
3568 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3569 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
3570 gpie |= IXGBE_GPIE_VTMODE_64;
3573 /* Enable fan failure interrupt */
3574 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3575 gpie |= IXGBE_SDP1_GPIEN;
3577 if (hw->mac.type == ixgbe_mac_82599EB) {
3578 gpie |= IXGBE_SDP1_GPIEN;
3579 gpie |= IXGBE_SDP2_GPIEN;
3582 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3585 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
3587 struct ixgbe_hw *hw = &adapter->hw;
3591 ixgbe_get_hw_control(adapter);
3592 ixgbe_setup_gpie(adapter);
3594 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3595 ixgbe_configure_msix(adapter);
3597 ixgbe_configure_msi_and_legacy(adapter);
3599 /* enable the optics for both mult-speed fiber and 82599 SFP+ fiber */
3600 if (hw->mac.ops.enable_tx_laser &&
3601 ((hw->phy.multispeed_fiber) ||
3602 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
3603 (hw->mac.type == ixgbe_mac_82599EB))))
3604 hw->mac.ops.enable_tx_laser(hw);
3606 clear_bit(__IXGBE_DOWN, &adapter->state);
3607 ixgbe_napi_enable_all(adapter);
3609 if (ixgbe_is_sfp(hw)) {
3610 ixgbe_sfp_link_config(adapter);
3612 err = ixgbe_non_sfp_link_config(hw);
3614 e_err(probe, "link_config FAILED %d\n", err);
3617 /* clear any pending interrupts, may auto mask */
3618 IXGBE_READ_REG(hw, IXGBE_EICR);
3619 ixgbe_irq_enable(adapter, true, true);
3622 * If this adapter has a fan, check to see if we had a failure
3623 * before we enabled the interrupt.
3625 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3626 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
3627 if (esdp & IXGBE_ESDP_SDP1)
3628 e_crit(drv, "Fan has stopped, replace the adapter\n");
3631 /* enable transmits */
3632 netif_tx_start_all_queues(adapter->netdev);
3634 /* bring the link up in the watchdog, this could race with our first
3635 * link up interrupt but shouldn't be a problem */
3636 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3637 adapter->link_check_timeout = jiffies;
3638 mod_timer(&adapter->service_timer, jiffies);
3640 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
3641 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
3642 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
3643 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
3648 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
3650 WARN_ON(in_interrupt());
3651 /* put off any impending NetWatchDogTimeout */
3652 adapter->netdev->trans_start = jiffies;
3654 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
3655 usleep_range(1000, 2000);
3656 ixgbe_down(adapter);
3658 * If SR-IOV enabled then wait a bit before bringing the adapter
3659 * back up to give the VFs time to respond to the reset. The
3660 * two second wait is based upon the watchdog timer cycle in
3663 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3666 clear_bit(__IXGBE_RESETTING, &adapter->state);
3669 int ixgbe_up(struct ixgbe_adapter *adapter)
3671 /* hardware has been reset, we need to reload some things */
3672 ixgbe_configure(adapter);
3674 return ixgbe_up_complete(adapter);
3677 void ixgbe_reset(struct ixgbe_adapter *adapter)
3679 struct ixgbe_hw *hw = &adapter->hw;
3682 /* lock SFP init bit to prevent race conditions with the watchdog */
3683 while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
3684 usleep_range(1000, 2000);
3686 /* clear all SFP and link config related flags while holding SFP_INIT */
3687 adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
3688 IXGBE_FLAG2_SFP_NEEDS_RESET);
3689 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
3691 err = hw->mac.ops.init_hw(hw);
3694 case IXGBE_ERR_SFP_NOT_PRESENT:
3695 case IXGBE_ERR_SFP_NOT_SUPPORTED:
3697 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
3698 e_dev_err("master disable timed out\n");
3700 case IXGBE_ERR_EEPROM_VERSION:
3701 /* We are running on a pre-production device, log a warning */
3702 e_dev_warn("This device is a pre-production adapter/LOM. "
3703 "Please be aware there may be issuesassociated with "
3704 "your hardware. If you are experiencing problems "
3705 "please contact your Intel or hardware "
3706 "representative who provided you with this "
3710 e_dev_err("Hardware Error: %d\n", err);
3713 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
3715 /* reprogram the RAR[0] in case user changed it. */
3716 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
3721 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
3722 * @rx_ring: ring to free buffers from
3724 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
3726 struct device *dev = rx_ring->dev;
3730 /* ring already cleared, nothing to do */
3731 if (!rx_ring->rx_buffer_info)
3734 /* Free all the Rx ring sk_buffs */
3735 for (i = 0; i < rx_ring->count; i++) {
3736 struct ixgbe_rx_buffer *rx_buffer_info;
3738 rx_buffer_info = &rx_ring->rx_buffer_info[i];
3739 if (rx_buffer_info->dma) {
3740 dma_unmap_single(rx_ring->dev, rx_buffer_info->dma,
3741 rx_ring->rx_buf_len,
3743 rx_buffer_info->dma = 0;
3745 if (rx_buffer_info->skb) {
3746 struct sk_buff *skb = rx_buffer_info->skb;
3747 rx_buffer_info->skb = NULL;
3749 struct sk_buff *this = skb;
3750 if (IXGBE_RSC_CB(this)->delay_unmap) {
3751 dma_unmap_single(dev,
3752 IXGBE_RSC_CB(this)->dma,
3753 rx_ring->rx_buf_len,
3755 IXGBE_RSC_CB(this)->dma = 0;
3756 IXGBE_RSC_CB(skb)->delay_unmap = false;
3759 dev_kfree_skb(this);
3762 if (!rx_buffer_info->page)
3764 if (rx_buffer_info->page_dma) {
3765 dma_unmap_page(dev, rx_buffer_info->page_dma,
3766 PAGE_SIZE / 2, DMA_FROM_DEVICE);
3767 rx_buffer_info->page_dma = 0;
3769 put_page(rx_buffer_info->page);
3770 rx_buffer_info->page = NULL;
3771 rx_buffer_info->page_offset = 0;
3774 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
3775 memset(rx_ring->rx_buffer_info, 0, size);
3777 /* Zero out the descriptor ring */
3778 memset(rx_ring->desc, 0, rx_ring->size);
3780 rx_ring->next_to_clean = 0;
3781 rx_ring->next_to_use = 0;
3785 * ixgbe_clean_tx_ring - Free Tx Buffers
3786 * @tx_ring: ring to be cleaned
3788 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
3790 struct ixgbe_tx_buffer *tx_buffer_info;
3794 /* ring already cleared, nothing to do */
3795 if (!tx_ring->tx_buffer_info)
3798 /* Free all the Tx ring sk_buffs */
3799 for (i = 0; i < tx_ring->count; i++) {
3800 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3801 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
3804 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
3805 memset(tx_ring->tx_buffer_info, 0, size);
3807 /* Zero out the descriptor ring */
3808 memset(tx_ring->desc, 0, tx_ring->size);
3810 tx_ring->next_to_use = 0;
3811 tx_ring->next_to_clean = 0;
3815 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
3816 * @adapter: board private structure
3818 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
3822 for (i = 0; i < adapter->num_rx_queues; i++)
3823 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
3827 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
3828 * @adapter: board private structure
3830 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
3834 for (i = 0; i < adapter->num_tx_queues; i++)
3835 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
3838 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
3840 struct hlist_node *node, *node2;
3841 struct ixgbe_fdir_filter *filter;
3843 spin_lock(&adapter->fdir_perfect_lock);
3845 hlist_for_each_entry_safe(filter, node, node2,
3846 &adapter->fdir_filter_list, fdir_node) {
3847 hlist_del(&filter->fdir_node);
3850 adapter->fdir_filter_count = 0;
3852 spin_unlock(&adapter->fdir_perfect_lock);
3855 void ixgbe_down(struct ixgbe_adapter *adapter)
3857 struct net_device *netdev = adapter->netdev;
3858 struct ixgbe_hw *hw = &adapter->hw;
3862 /* signal that we are down to the interrupt handler */
3863 set_bit(__IXGBE_DOWN, &adapter->state);
3865 /* disable receives */
3866 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3867 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3869 /* disable all enabled rx queues */
3870 for (i = 0; i < adapter->num_rx_queues; i++)
3871 /* this call also flushes the previous write */
3872 ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]);
3874 usleep_range(10000, 20000);
3876 netif_tx_stop_all_queues(netdev);
3878 /* call carrier off first to avoid false dev_watchdog timeouts */
3879 netif_carrier_off(netdev);
3880 netif_tx_disable(netdev);
3882 ixgbe_irq_disable(adapter);
3884 ixgbe_napi_disable_all(adapter);
3886 adapter->flags2 &= ~(IXGBE_FLAG2_FDIR_REQUIRES_REINIT |
3887 IXGBE_FLAG2_RESET_REQUESTED);
3888 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
3890 del_timer_sync(&adapter->service_timer);
3892 /* disable receive for all VFs and wait one second */
3893 if (adapter->num_vfs) {
3894 /* ping all the active vfs to let them know we are going down */
3895 ixgbe_ping_all_vfs(adapter);
3897 /* Disable all VFTE/VFRE TX/RX */
3898 ixgbe_disable_tx_rx(adapter);
3900 /* Mark all the VFs as inactive */
3901 for (i = 0 ; i < adapter->num_vfs; i++)
3902 adapter->vfinfo[i].clear_to_send = 0;
3905 /* disable transmits in the hardware now that interrupts are off */
3906 for (i = 0; i < adapter->num_tx_queues; i++) {
3907 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
3908 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
3911 /* Disable the Tx DMA engine on 82599 and X540 */
3912 switch (hw->mac.type) {
3913 case ixgbe_mac_82599EB:
3914 case ixgbe_mac_X540:
3915 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
3916 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
3917 ~IXGBE_DMATXCTL_TE));
3923 if (!pci_channel_offline(adapter->pdev))
3924 ixgbe_reset(adapter);
3926 /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
3927 if (hw->mac.ops.disable_tx_laser &&
3928 ((hw->phy.multispeed_fiber) ||
3929 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
3930 (hw->mac.type == ixgbe_mac_82599EB))))
3931 hw->mac.ops.disable_tx_laser(hw);
3933 ixgbe_clean_all_tx_rings(adapter);
3934 ixgbe_clean_all_rx_rings(adapter);
3936 #ifdef CONFIG_IXGBE_DCA
3937 /* since we reset the hardware DCA settings were cleared */
3938 ixgbe_setup_dca(adapter);
3943 * ixgbe_poll - NAPI Rx polling callback
3944 * @napi: structure for representing this polling device
3945 * @budget: how many packets driver is allowed to clean
3947 * This function is used for legacy and MSI, NAPI mode
3949 static int ixgbe_poll(struct napi_struct *napi, int budget)
3951 struct ixgbe_q_vector *q_vector =
3952 container_of(napi, struct ixgbe_q_vector, napi);
3953 struct ixgbe_adapter *adapter = q_vector->adapter;
3954 struct ixgbe_ring *ring;
3955 int per_ring_budget;
3956 bool clean_complete = true;
3958 #ifdef CONFIG_IXGBE_DCA
3959 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
3960 ixgbe_update_dca(q_vector);
3963 for (ring = q_vector->tx.ring; ring != NULL; ring = ring->next)
3964 clean_complete &= !!ixgbe_clean_tx_irq(q_vector, ring);
3966 /* attempt to distribute budget to each queue fairly, but don't allow
3967 * the budget to go below 1 because we'll exit polling */
3968 if (q_vector->rx.count > 1)
3969 per_ring_budget = max(budget/q_vector->rx.count, 1);
3971 per_ring_budget = budget;
3973 for (ring = q_vector->rx.ring; ring != NULL; ring = ring->next)
3974 clean_complete &= ixgbe_clean_rx_irq(q_vector, ring,
3977 /* If all work not completed, return budget and keep polling */
3978 if (!clean_complete)
3981 /* all work done, exit the polling mode */
3982 napi_complete(napi);
3983 if (adapter->rx_itr_setting & 1)
3984 ixgbe_set_itr(q_vector);
3985 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3986 ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
3992 * ixgbe_tx_timeout - Respond to a Tx Hang
3993 * @netdev: network interface device structure
3995 static void ixgbe_tx_timeout(struct net_device *netdev)
3997 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3999 /* Do the reset outside of interrupt context */
4000 ixgbe_tx_timeout_reset(adapter);
4004 * ixgbe_set_rss_queues: Allocate queues for RSS
4005 * @adapter: board private structure to initialize
4007 * This is our "base" multiqueue mode. RSS (Receive Side Scaling) will try
4008 * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
4011 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
4014 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
4016 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4018 adapter->num_rx_queues = f->indices;
4019 adapter->num_tx_queues = f->indices;
4029 * ixgbe_set_fdir_queues: Allocate queues for Flow Director
4030 * @adapter: board private structure to initialize
4032 * Flow Director is an advanced Rx filter, attempting to get Rx flows back
4033 * to the original CPU that initiated the Tx session. This runs in addition
4034 * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
4035 * Rx load across CPUs using RSS.
4038 static inline bool ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
4041 struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
4043 f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
4046 /* Flow Director must have RSS enabled */
4047 if ((adapter->flags & IXGBE_FLAG_RSS_ENABLED) &&
4048 (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)) {
4049 adapter->num_tx_queues = f_fdir->indices;
4050 adapter->num_rx_queues = f_fdir->indices;
4053 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4060 * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
4061 * @adapter: board private structure to initialize
4063 * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
4064 * The ring feature mask is not used as a mask for FCoE, as it can take any 8
4065 * rx queues out of the max number of rx queues, instead, it is used as the
4066 * index of the first rx queue used by FCoE.
4069 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
4071 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4073 if (!(adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
4076 f->indices = min((int)num_online_cpus(), f->indices);
4078 adapter->num_rx_queues = 1;
4079 adapter->num_tx_queues = 1;
4081 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4082 e_info(probe, "FCoE enabled with RSS\n");
4083 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)
4084 ixgbe_set_fdir_queues(adapter);
4086 ixgbe_set_rss_queues(adapter);
4089 /* adding FCoE rx rings to the end */
4090 f->mask = adapter->num_rx_queues;
4091 adapter->num_rx_queues += f->indices;
4092 adapter->num_tx_queues += f->indices;
4096 #endif /* IXGBE_FCOE */
4098 /* Artificial max queue cap per traffic class in DCB mode */
4099 #define DCB_QUEUE_CAP 8
4101 #ifdef CONFIG_IXGBE_DCB
4102 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
4104 int per_tc_q, q, i, offset = 0;
4105 struct net_device *dev = adapter->netdev;
4106 int tcs = netdev_get_num_tc(dev);
4111 /* Map queue offset and counts onto allocated tx queues */
4112 per_tc_q = min(dev->num_tx_queues / tcs, (unsigned int)DCB_QUEUE_CAP);
4113 q = min((int)num_online_cpus(), per_tc_q);
4115 for (i = 0; i < tcs; i++) {
4116 netdev_set_prio_tc_map(dev, i, i);
4117 netdev_set_tc_queue(dev, i, q, offset);
4121 adapter->num_tx_queues = q * tcs;
4122 adapter->num_rx_queues = q * tcs;
4125 /* FCoE enabled queues require special configuration indexed
4126 * by feature specific indices and mask. Here we map FCoE
4127 * indices onto the DCB queue pairs allowing FCoE to own
4128 * configuration later.
4130 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
4132 struct ixgbe_ring_feature *f =
4133 &adapter->ring_feature[RING_F_FCOE];
4135 tc = netdev_get_prio_tc_map(dev, adapter->fcoe.up);
4136 f->indices = dev->tc_to_txq[tc].count;
4137 f->mask = dev->tc_to_txq[tc].offset;
4146 * ixgbe_set_sriov_queues: Allocate queues for IOV use
4147 * @adapter: board private structure to initialize
4149 * IOV doesn't actually use anything, so just NAK the
4150 * request for now and let the other queue routines
4151 * figure out what to do.
4153 static inline bool ixgbe_set_sriov_queues(struct ixgbe_adapter *adapter)
4159 * ixgbe_set_num_queues: Allocate queues for device, feature dependent
4160 * @adapter: board private structure to initialize
4162 * This is the top level queue allocation routine. The order here is very
4163 * important, starting with the "most" number of features turned on at once,
4164 * and ending with the smallest set of features. This way large combinations
4165 * can be allocated if they're turned on, and smaller combinations are the
4166 * fallthrough conditions.
4169 static int ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
4171 /* Start with base case */
4172 adapter->num_rx_queues = 1;
4173 adapter->num_tx_queues = 1;
4174 adapter->num_rx_pools = adapter->num_rx_queues;
4175 adapter->num_rx_queues_per_pool = 1;
4177 if (ixgbe_set_sriov_queues(adapter))
4180 #ifdef CONFIG_IXGBE_DCB
4181 if (ixgbe_set_dcb_queues(adapter))
4186 if (ixgbe_set_fcoe_queues(adapter))
4189 #endif /* IXGBE_FCOE */
4190 if (ixgbe_set_fdir_queues(adapter))
4193 if (ixgbe_set_rss_queues(adapter))
4196 /* fallback to base case */
4197 adapter->num_rx_queues = 1;
4198 adapter->num_tx_queues = 1;
4201 /* Notify the stack of the (possibly) reduced queue counts. */
4202 netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
4203 return netif_set_real_num_rx_queues(adapter->netdev,
4204 adapter->num_rx_queues);
4207 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
4210 int err, vector_threshold;
4212 /* We'll want at least 3 (vector_threshold):
4215 * 3) Other (Link Status Change, etc.)
4216 * 4) TCP Timer (optional)
4218 vector_threshold = MIN_MSIX_COUNT;
4220 /* The more we get, the more we will assign to Tx/Rx Cleanup
4221 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
4222 * Right now, we simply care about how many we'll get; we'll
4223 * set them up later while requesting irq's.
4225 while (vectors >= vector_threshold) {
4226 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
4228 if (!err) /* Success in acquiring all requested vectors. */
4231 vectors = 0; /* Nasty failure, quit now */
4232 else /* err == number of vectors we should try again with */
4236 if (vectors < vector_threshold) {
4237 /* Can't allocate enough MSI-X interrupts? Oh well.
4238 * This just means we'll go with either a single MSI
4239 * vector or fall back to legacy interrupts.
4241 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4242 "Unable to allocate MSI-X interrupts\n");
4243 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4244 kfree(adapter->msix_entries);
4245 adapter->msix_entries = NULL;
4247 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
4249 * Adjust for only the vectors we'll use, which is minimum
4250 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
4251 * vectors we were allocated.
4253 adapter->num_msix_vectors = min(vectors,
4254 adapter->max_msix_q_vectors + NON_Q_VECTORS);
4259 * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
4260 * @adapter: board private structure to initialize
4262 * Cache the descriptor ring offsets for RSS to the assigned rings.
4265 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
4269 if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED))
4272 for (i = 0; i < adapter->num_rx_queues; i++)
4273 adapter->rx_ring[i]->reg_idx = i;
4274 for (i = 0; i < adapter->num_tx_queues; i++)
4275 adapter->tx_ring[i]->reg_idx = i;
4280 #ifdef CONFIG_IXGBE_DCB
4282 /* ixgbe_get_first_reg_idx - Return first register index associated with ring */
4283 static void ixgbe_get_first_reg_idx(struct ixgbe_adapter *adapter, u8 tc,
4284 unsigned int *tx, unsigned int *rx)
4286 struct net_device *dev = adapter->netdev;
4287 struct ixgbe_hw *hw = &adapter->hw;
4288 u8 num_tcs = netdev_get_num_tc(dev);
4293 switch (hw->mac.type) {
4294 case ixgbe_mac_82598EB:
4298 case ixgbe_mac_82599EB:
4299 case ixgbe_mac_X540:
4304 } else if (tc < 5) {
4305 *tx = ((tc + 2) << 4);
4307 } else if (tc < num_tcs) {
4308 *tx = ((tc + 8) << 3);
4337 * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
4338 * @adapter: board private structure to initialize
4340 * Cache the descriptor ring offsets for DCB to the assigned rings.
4343 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
4345 struct net_device *dev = adapter->netdev;
4347 u8 num_tcs = netdev_get_num_tc(dev);
4352 for (i = 0, k = 0; i < num_tcs; i++) {
4353 unsigned int tx_s, rx_s;
4354 u16 count = dev->tc_to_txq[i].count;
4356 ixgbe_get_first_reg_idx(adapter, i, &tx_s, &rx_s);
4357 for (j = 0; j < count; j++, k++) {
4358 adapter->tx_ring[k]->reg_idx = tx_s + j;
4359 adapter->rx_ring[k]->reg_idx = rx_s + j;
4360 adapter->tx_ring[k]->dcb_tc = i;
4361 adapter->rx_ring[k]->dcb_tc = i;
4370 * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
4371 * @adapter: board private structure to initialize
4373 * Cache the descriptor ring offsets for Flow Director to the assigned rings.
4376 static inline bool ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
4381 if ((adapter->flags & IXGBE_FLAG_RSS_ENABLED) &&
4382 (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)) {
4383 for (i = 0; i < adapter->num_rx_queues; i++)
4384 adapter->rx_ring[i]->reg_idx = i;
4385 for (i = 0; i < adapter->num_tx_queues; i++)
4386 adapter->tx_ring[i]->reg_idx = i;
4395 * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
4396 * @adapter: board private structure to initialize
4398 * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
4401 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
4403 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4405 u8 fcoe_rx_i = 0, fcoe_tx_i = 0;
4407 if (!(adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
4410 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4411 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)
4412 ixgbe_cache_ring_fdir(adapter);
4414 ixgbe_cache_ring_rss(adapter);
4416 fcoe_rx_i = f->mask;
4417 fcoe_tx_i = f->mask;
4419 for (i = 0; i < f->indices; i++, fcoe_rx_i++, fcoe_tx_i++) {
4420 adapter->rx_ring[f->mask + i]->reg_idx = fcoe_rx_i;
4421 adapter->tx_ring[f->mask + i]->reg_idx = fcoe_tx_i;
4426 #endif /* IXGBE_FCOE */
4428 * ixgbe_cache_ring_sriov - Descriptor ring to register mapping for sriov
4429 * @adapter: board private structure to initialize
4431 * SR-IOV doesn't use any descriptor rings but changes the default if
4432 * no other mapping is used.
4435 static inline bool ixgbe_cache_ring_sriov(struct ixgbe_adapter *adapter)
4437 adapter->rx_ring[0]->reg_idx = adapter->num_vfs * 2;
4438 adapter->tx_ring[0]->reg_idx = adapter->num_vfs * 2;
4439 if (adapter->num_vfs)
4446 * ixgbe_cache_ring_register - Descriptor ring to register mapping
4447 * @adapter: board private structure to initialize
4449 * Once we know the feature-set enabled for the device, we'll cache
4450 * the register offset the descriptor ring is assigned to.
4452 * Note, the order the various feature calls is important. It must start with
4453 * the "most" features enabled at the same time, then trickle down to the
4454 * least amount of features turned on at once.
4456 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
4458 /* start with default case */
4459 adapter->rx_ring[0]->reg_idx = 0;
4460 adapter->tx_ring[0]->reg_idx = 0;
4462 if (ixgbe_cache_ring_sriov(adapter))
4465 #ifdef CONFIG_IXGBE_DCB
4466 if (ixgbe_cache_ring_dcb(adapter))
4471 if (ixgbe_cache_ring_fcoe(adapter))
4473 #endif /* IXGBE_FCOE */
4475 if (ixgbe_cache_ring_fdir(adapter))
4478 if (ixgbe_cache_ring_rss(adapter))
4483 * ixgbe_alloc_queues - Allocate memory for all rings
4484 * @adapter: board private structure to initialize
4486 * We allocate one ring per queue at run-time since we don't know the
4487 * number of queues at compile-time. The polling_netdev array is
4488 * intended for Multiqueue, but should work fine with a single queue.
4490 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
4492 int rx = 0, tx = 0, nid = adapter->node;
4494 if (nid < 0 || !node_online(nid))
4495 nid = first_online_node;
4497 for (; tx < adapter->num_tx_queues; tx++) {
4498 struct ixgbe_ring *ring;
4500 ring = kzalloc_node(sizeof(*ring), GFP_KERNEL, nid);
4502 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
4504 goto err_allocation;
4505 ring->count = adapter->tx_ring_count;
4506 ring->queue_index = tx;
4507 ring->numa_node = nid;
4508 ring->dev = &adapter->pdev->dev;
4509 ring->netdev = adapter->netdev;
4511 adapter->tx_ring[tx] = ring;
4514 for (; rx < adapter->num_rx_queues; rx++) {
4515 struct ixgbe_ring *ring;
4517 ring = kzalloc_node(sizeof(*ring), GFP_KERNEL, nid);
4519 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
4521 goto err_allocation;
4522 ring->count = adapter->rx_ring_count;
4523 ring->queue_index = rx;
4524 ring->numa_node = nid;
4525 ring->dev = &adapter->pdev->dev;
4526 ring->netdev = adapter->netdev;
4528 adapter->rx_ring[rx] = ring;
4531 ixgbe_cache_ring_register(adapter);
4537 kfree(adapter->tx_ring[--tx]);
4540 kfree(adapter->rx_ring[--rx]);
4545 * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
4546 * @adapter: board private structure to initialize
4548 * Attempt to configure the interrupts using the best available
4549 * capabilities of the hardware and the kernel.
4551 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
4553 struct ixgbe_hw *hw = &adapter->hw;
4555 int vector, v_budget;
4558 * It's easy to be greedy for MSI-X vectors, but it really
4559 * doesn't do us much good if we have a lot more vectors
4560 * than CPU's. So let's be conservative and only ask for
4561 * (roughly) the same number of vectors as there are CPU's.
4563 v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
4564 (int)num_online_cpus()) + NON_Q_VECTORS;
4567 * At the same time, hardware can only support a maximum of
4568 * hw.mac->max_msix_vectors vectors. With features
4569 * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
4570 * descriptor queues supported by our device. Thus, we cap it off in
4571 * those rare cases where the cpu count also exceeds our vector limit.
4573 v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
4575 /* A failure in MSI-X entry allocation isn't fatal, but it does
4576 * mean we disable MSI-X capabilities of the adapter. */
4577 adapter->msix_entries = kcalloc(v_budget,
4578 sizeof(struct msix_entry), GFP_KERNEL);
4579 if (adapter->msix_entries) {
4580 for (vector = 0; vector < v_budget; vector++)
4581 adapter->msix_entries[vector].entry = vector;
4583 ixgbe_acquire_msix_vectors(adapter, v_budget);
4585 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4589 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
4590 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
4591 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
4593 "ATR is not supported while multiple "
4594 "queues are disabled. Disabling Flow Director\n");
4596 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4597 adapter->atr_sample_rate = 0;
4598 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4599 ixgbe_disable_sriov(adapter);
4601 err = ixgbe_set_num_queues(adapter);
4605 err = pci_enable_msi(adapter->pdev);
4607 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
4609 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4610 "Unable to allocate MSI interrupt, "
4611 "falling back to legacy. Error: %d\n", err);
4621 * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
4622 * @adapter: board private structure to initialize
4624 * We allocate one q_vector per queue interrupt. If allocation fails we
4627 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
4629 int v_idx, num_q_vectors;
4630 struct ixgbe_q_vector *q_vector;
4632 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4633 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4637 for (v_idx = 0; v_idx < num_q_vectors; v_idx++) {
4638 q_vector = kzalloc_node(sizeof(struct ixgbe_q_vector),
4639 GFP_KERNEL, adapter->node);
4641 q_vector = kzalloc(sizeof(struct ixgbe_q_vector),
4646 q_vector->adapter = adapter;
4647 q_vector->v_idx = v_idx;
4649 /* Allocate the affinity_hint cpumask, configure the mask */
4650 if (!alloc_cpumask_var(&q_vector->affinity_mask, GFP_KERNEL))
4652 cpumask_set_cpu(v_idx, q_vector->affinity_mask);
4654 if (q_vector->tx.count && !q_vector->rx.count)
4655 q_vector->eitr = adapter->tx_eitr_param;
4657 q_vector->eitr = adapter->rx_eitr_param;
4659 netif_napi_add(adapter->netdev, &q_vector->napi,
4661 adapter->q_vector[v_idx] = q_vector;
4669 q_vector = adapter->q_vector[v_idx];
4670 netif_napi_del(&q_vector->napi);
4671 free_cpumask_var(q_vector->affinity_mask);
4673 adapter->q_vector[v_idx] = NULL;
4679 * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
4680 * @adapter: board private structure to initialize
4682 * This function frees the memory allocated to the q_vectors. In addition if
4683 * NAPI is enabled it will delete any references to the NAPI struct prior
4684 * to freeing the q_vector.
4686 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
4688 int v_idx, num_q_vectors;
4690 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4691 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4695 for (v_idx = 0; v_idx < num_q_vectors; v_idx++) {
4696 struct ixgbe_q_vector *q_vector = adapter->q_vector[v_idx];
4697 adapter->q_vector[v_idx] = NULL;
4698 netif_napi_del(&q_vector->napi);
4699 free_cpumask_var(q_vector->affinity_mask);
4704 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
4706 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4707 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4708 pci_disable_msix(adapter->pdev);
4709 kfree(adapter->msix_entries);
4710 adapter->msix_entries = NULL;
4711 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
4712 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
4713 pci_disable_msi(adapter->pdev);
4718 * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
4719 * @adapter: board private structure to initialize
4721 * We determine which interrupt scheme to use based on...
4722 * - Kernel support (MSI, MSI-X)
4723 * - which can be user-defined (via MODULE_PARAM)
4724 * - Hardware queue count (num_*_queues)
4725 * - defined by miscellaneous hardware support/features (RSS, etc.)
4727 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
4731 /* Number of supported queues */
4732 err = ixgbe_set_num_queues(adapter);
4736 err = ixgbe_set_interrupt_capability(adapter);
4738 e_dev_err("Unable to setup interrupt capabilities\n");
4739 goto err_set_interrupt;
4742 err = ixgbe_alloc_q_vectors(adapter);
4744 e_dev_err("Unable to allocate memory for queue vectors\n");
4745 goto err_alloc_q_vectors;
4748 err = ixgbe_alloc_queues(adapter);
4750 e_dev_err("Unable to allocate memory for queues\n");
4751 goto err_alloc_queues;
4754 e_dev_info("Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
4755 (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
4756 adapter->num_rx_queues, adapter->num_tx_queues);
4758 set_bit(__IXGBE_DOWN, &adapter->state);
4763 ixgbe_free_q_vectors(adapter);
4764 err_alloc_q_vectors:
4765 ixgbe_reset_interrupt_capability(adapter);
4771 * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
4772 * @adapter: board private structure to clear interrupt scheme on
4774 * We go through and clear interrupt specific resources and reset the structure
4775 * to pre-load conditions
4777 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
4781 for (i = 0; i < adapter->num_tx_queues; i++) {
4782 kfree(adapter->tx_ring[i]);
4783 adapter->tx_ring[i] = NULL;
4785 for (i = 0; i < adapter->num_rx_queues; i++) {
4786 struct ixgbe_ring *ring = adapter->rx_ring[i];
4788 /* ixgbe_get_stats64() might access this ring, we must wait
4789 * a grace period before freeing it.
4791 kfree_rcu(ring, rcu);
4792 adapter->rx_ring[i] = NULL;
4795 adapter->num_tx_queues = 0;
4796 adapter->num_rx_queues = 0;
4798 ixgbe_free_q_vectors(adapter);
4799 ixgbe_reset_interrupt_capability(adapter);
4803 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4804 * @adapter: board private structure to initialize
4806 * ixgbe_sw_init initializes the Adapter private data structure.
4807 * Fields are initialized based on PCI device information and
4808 * OS network device settings (MTU size).
4810 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
4812 struct ixgbe_hw *hw = &adapter->hw;
4813 struct pci_dev *pdev = adapter->pdev;
4814 struct net_device *dev = adapter->netdev;
4816 #ifdef CONFIG_IXGBE_DCB
4818 struct tc_configuration *tc;
4820 int max_frame = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
4822 /* PCI config space info */
4824 hw->vendor_id = pdev->vendor;
4825 hw->device_id = pdev->device;
4826 hw->revision_id = pdev->revision;
4827 hw->subsystem_vendor_id = pdev->subsystem_vendor;
4828 hw->subsystem_device_id = pdev->subsystem_device;
4830 /* Set capability flags */
4831 rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
4832 adapter->ring_feature[RING_F_RSS].indices = rss;
4833 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
4834 switch (hw->mac.type) {
4835 case ixgbe_mac_82598EB:
4836 if (hw->device_id == IXGBE_DEV_ID_82598AT)
4837 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4838 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
4840 case ixgbe_mac_82599EB:
4841 case ixgbe_mac_X540:
4842 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
4843 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4844 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4845 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
4846 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4847 /* Flow Director hash filters enabled */
4848 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
4849 adapter->atr_sample_rate = 20;
4850 adapter->ring_feature[RING_F_FDIR].indices =
4851 IXGBE_MAX_FDIR_INDICES;
4852 adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
4854 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4855 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4856 adapter->ring_feature[RING_F_FCOE].indices = 0;
4857 #ifdef CONFIG_IXGBE_DCB
4858 /* Default traffic class to use for FCoE */
4859 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
4861 #endif /* IXGBE_FCOE */
4867 /* n-tuple support exists, always init our spinlock */
4868 spin_lock_init(&adapter->fdir_perfect_lock);
4870 #ifdef CONFIG_IXGBE_DCB
4871 /* Configure DCB traffic classes */
4872 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
4873 tc = &adapter->dcb_cfg.tc_config[j];
4874 tc->path[DCB_TX_CONFIG].bwg_id = 0;
4875 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
4876 tc->path[DCB_RX_CONFIG].bwg_id = 0;
4877 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
4878 tc->dcb_pfc = pfc_disabled;
4880 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
4881 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
4882 adapter->dcb_cfg.pfc_mode_enable = false;
4883 adapter->dcb_set_bitmap = 0x00;
4884 adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
4885 ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
4890 /* default flow control settings */
4891 hw->fc.requested_mode = ixgbe_fc_full;
4892 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
4894 adapter->last_lfc_mode = hw->fc.current_mode;
4896 hw->fc.high_water = FC_HIGH_WATER(max_frame);
4897 hw->fc.low_water = FC_LOW_WATER(max_frame);
4898 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
4899 hw->fc.send_xon = true;
4900 hw->fc.disable_fc_autoneg = false;
4902 /* enable itr by default in dynamic mode */
4903 adapter->rx_itr_setting = 1;
4904 adapter->rx_eitr_param = 20000;
4905 adapter->tx_itr_setting = 1;
4906 adapter->tx_eitr_param = 10000;
4908 /* set defaults for eitr in MegaBytes */
4909 adapter->eitr_low = 10;
4910 adapter->eitr_high = 20;
4912 /* set default ring sizes */
4913 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
4914 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
4916 /* set default work limits */
4917 adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
4919 /* initialize eeprom parameters */
4920 if (ixgbe_init_eeprom_params_generic(hw)) {
4921 e_dev_err("EEPROM initialization failed\n");
4925 /* enable rx csum by default */
4926 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
4928 /* get assigned NUMA node */
4929 adapter->node = dev_to_node(&pdev->dev);
4931 set_bit(__IXGBE_DOWN, &adapter->state);
4937 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
4938 * @tx_ring: tx descriptor ring (for a specific queue) to setup
4940 * Return 0 on success, negative on failure
4942 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
4944 struct device *dev = tx_ring->dev;
4947 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4948 tx_ring->tx_buffer_info = vzalloc_node(size, tx_ring->numa_node);
4949 if (!tx_ring->tx_buffer_info)
4950 tx_ring->tx_buffer_info = vzalloc(size);
4951 if (!tx_ring->tx_buffer_info)
4954 /* round up to nearest 4K */
4955 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
4956 tx_ring->size = ALIGN(tx_ring->size, 4096);
4958 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
4959 &tx_ring->dma, GFP_KERNEL);
4963 tx_ring->next_to_use = 0;
4964 tx_ring->next_to_clean = 0;
4968 vfree(tx_ring->tx_buffer_info);
4969 tx_ring->tx_buffer_info = NULL;
4970 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
4975 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
4976 * @adapter: board private structure
4978 * If this function returns with an error, then it's possible one or
4979 * more of the rings is populated (while the rest are not). It is the
4980 * callers duty to clean those orphaned rings.
4982 * Return 0 on success, negative on failure
4984 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
4988 for (i = 0; i < adapter->num_tx_queues; i++) {
4989 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
4992 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
5000 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
5001 * @rx_ring: rx descriptor ring (for a specific queue) to setup
5003 * Returns 0 on success, negative on failure
5005 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
5007 struct device *dev = rx_ring->dev;
5010 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
5011 rx_ring->rx_buffer_info = vzalloc_node(size, rx_ring->numa_node);
5012 if (!rx_ring->rx_buffer_info)
5013 rx_ring->rx_buffer_info = vzalloc(size);
5014 if (!rx_ring->rx_buffer_info)
5017 /* Round up to nearest 4K */
5018 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
5019 rx_ring->size = ALIGN(rx_ring->size, 4096);
5021 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
5022 &rx_ring->dma, GFP_KERNEL);
5027 rx_ring->next_to_clean = 0;
5028 rx_ring->next_to_use = 0;
5032 vfree(rx_ring->rx_buffer_info);
5033 rx_ring->rx_buffer_info = NULL;
5034 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
5039 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
5040 * @adapter: board private structure
5042 * If this function returns with an error, then it's possible one or
5043 * more of the rings is populated (while the rest are not). It is the
5044 * callers duty to clean those orphaned rings.
5046 * Return 0 on success, negative on failure
5048 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
5052 for (i = 0; i < adapter->num_rx_queues; i++) {
5053 err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
5056 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
5064 * ixgbe_free_tx_resources - Free Tx Resources per Queue
5065 * @tx_ring: Tx descriptor ring for a specific queue
5067 * Free all transmit software resources
5069 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
5071 ixgbe_clean_tx_ring(tx_ring);
5073 vfree(tx_ring->tx_buffer_info);
5074 tx_ring->tx_buffer_info = NULL;
5076 /* if not set, then don't free */
5080 dma_free_coherent(tx_ring->dev, tx_ring->size,
5081 tx_ring->desc, tx_ring->dma);
5083 tx_ring->desc = NULL;
5087 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5088 * @adapter: board private structure
5090 * Free all transmit software resources
5092 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
5096 for (i = 0; i < adapter->num_tx_queues; i++)
5097 if (adapter->tx_ring[i]->desc)
5098 ixgbe_free_tx_resources(adapter->tx_ring[i]);
5102 * ixgbe_free_rx_resources - Free Rx Resources
5103 * @rx_ring: ring to clean the resources from
5105 * Free all receive software resources
5107 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
5109 ixgbe_clean_rx_ring(rx_ring);
5111 vfree(rx_ring->rx_buffer_info);
5112 rx_ring->rx_buffer_info = NULL;
5114 /* if not set, then don't free */
5118 dma_free_coherent(rx_ring->dev, rx_ring->size,
5119 rx_ring->desc, rx_ring->dma);
5121 rx_ring->desc = NULL;
5125 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5126 * @adapter: board private structure
5128 * Free all receive software resources
5130 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5134 for (i = 0; i < adapter->num_rx_queues; i++)
5135 if (adapter->rx_ring[i]->desc)
5136 ixgbe_free_rx_resources(adapter->rx_ring[i]);
5140 * ixgbe_change_mtu - Change the Maximum Transfer Unit
5141 * @netdev: network interface device structure
5142 * @new_mtu: new value for maximum frame size
5144 * Returns 0 on success, negative on failure
5146 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5148 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5149 struct ixgbe_hw *hw = &adapter->hw;
5150 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5152 /* MTU < 68 is an error and causes problems on some kernels */
5153 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED &&
5154 hw->mac.type != ixgbe_mac_X540) {
5155 if ((new_mtu < 68) || (max_frame > MAXIMUM_ETHERNET_VLAN_SIZE))
5158 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5162 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5163 /* must set new MTU before calling down or up */
5164 netdev->mtu = new_mtu;
5166 hw->fc.high_water = FC_HIGH_WATER(max_frame);
5167 hw->fc.low_water = FC_LOW_WATER(max_frame);
5169 if (netif_running(netdev))
5170 ixgbe_reinit_locked(adapter);
5176 * ixgbe_open - Called when a network interface is made active
5177 * @netdev: network interface device structure
5179 * Returns 0 on success, negative value on failure
5181 * The open entry point is called when a network interface is made
5182 * active by the system (IFF_UP). At this point all resources needed
5183 * for transmit and receive operations are allocated, the interrupt
5184 * handler is registered with the OS, the watchdog timer is started,
5185 * and the stack is notified that the interface is ready.
5187 static int ixgbe_open(struct net_device *netdev)
5189 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5192 /* disallow open during test */
5193 if (test_bit(__IXGBE_TESTING, &adapter->state))
5196 netif_carrier_off(netdev);
5198 /* allocate transmit descriptors */
5199 err = ixgbe_setup_all_tx_resources(adapter);
5203 /* allocate receive descriptors */
5204 err = ixgbe_setup_all_rx_resources(adapter);
5208 ixgbe_configure(adapter);
5210 err = ixgbe_request_irq(adapter);
5214 err = ixgbe_up_complete(adapter);
5218 netif_tx_start_all_queues(netdev);
5223 ixgbe_release_hw_control(adapter);
5224 ixgbe_free_irq(adapter);
5227 ixgbe_free_all_rx_resources(adapter);
5229 ixgbe_free_all_tx_resources(adapter);
5230 ixgbe_reset(adapter);
5236 * ixgbe_close - Disables a network interface
5237 * @netdev: network interface device structure
5239 * Returns 0, this is not allowed to fail
5241 * The close entry point is called when an interface is de-activated
5242 * by the OS. The hardware is still under the drivers control, but
5243 * needs to be disabled. A global MAC reset is issued to stop the
5244 * hardware, and all transmit and receive resources are freed.
5246 static int ixgbe_close(struct net_device *netdev)
5248 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5250 ixgbe_down(adapter);
5251 ixgbe_free_irq(adapter);
5253 ixgbe_fdir_filter_exit(adapter);
5255 ixgbe_free_all_tx_resources(adapter);
5256 ixgbe_free_all_rx_resources(adapter);
5258 ixgbe_release_hw_control(adapter);
5264 static int ixgbe_resume(struct pci_dev *pdev)
5266 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5267 struct net_device *netdev = adapter->netdev;
5270 pci_set_power_state(pdev, PCI_D0);
5271 pci_restore_state(pdev);
5273 * pci_restore_state clears dev->state_saved so call
5274 * pci_save_state to restore it.
5276 pci_save_state(pdev);
5278 err = pci_enable_device_mem(pdev);
5280 e_dev_err("Cannot enable PCI device from suspend\n");
5283 pci_set_master(pdev);
5285 pci_wake_from_d3(pdev, false);
5287 err = ixgbe_init_interrupt_scheme(adapter);
5289 e_dev_err("Cannot initialize interrupts for device\n");
5293 ixgbe_reset(adapter);
5295 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5297 if (netif_running(netdev)) {
5298 err = ixgbe_open(netdev);
5303 netif_device_attach(netdev);
5307 #endif /* CONFIG_PM */
5309 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5311 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5312 struct net_device *netdev = adapter->netdev;
5313 struct ixgbe_hw *hw = &adapter->hw;
5315 u32 wufc = adapter->wol;
5320 netif_device_detach(netdev);
5322 if (netif_running(netdev)) {
5323 ixgbe_down(adapter);
5324 ixgbe_free_irq(adapter);
5325 ixgbe_free_all_tx_resources(adapter);
5326 ixgbe_free_all_rx_resources(adapter);
5329 ixgbe_clear_interrupt_scheme(adapter);
5331 kfree(adapter->ixgbe_ieee_pfc);
5332 kfree(adapter->ixgbe_ieee_ets);
5336 retval = pci_save_state(pdev);
5342 ixgbe_set_rx_mode(netdev);
5344 /* turn on all-multi mode if wake on multicast is enabled */
5345 if (wufc & IXGBE_WUFC_MC) {
5346 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5347 fctrl |= IXGBE_FCTRL_MPE;
5348 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5351 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5352 ctrl |= IXGBE_CTRL_GIO_DIS;
5353 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5355 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5357 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5358 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5361 switch (hw->mac.type) {
5362 case ixgbe_mac_82598EB:
5363 pci_wake_from_d3(pdev, false);
5365 case ixgbe_mac_82599EB:
5366 case ixgbe_mac_X540:
5367 pci_wake_from_d3(pdev, !!wufc);
5373 *enable_wake = !!wufc;
5375 ixgbe_release_hw_control(adapter);
5377 pci_disable_device(pdev);
5383 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5388 retval = __ixgbe_shutdown(pdev, &wake);
5393 pci_prepare_to_sleep(pdev);
5395 pci_wake_from_d3(pdev, false);
5396 pci_set_power_state(pdev, PCI_D3hot);
5401 #endif /* CONFIG_PM */
5403 static void ixgbe_shutdown(struct pci_dev *pdev)
5407 __ixgbe_shutdown(pdev, &wake);
5409 if (system_state == SYSTEM_POWER_OFF) {
5410 pci_wake_from_d3(pdev, wake);
5411 pci_set_power_state(pdev, PCI_D3hot);
5416 * ixgbe_update_stats - Update the board statistics counters.
5417 * @adapter: board private structure
5419 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5421 struct net_device *netdev = adapter->netdev;
5422 struct ixgbe_hw *hw = &adapter->hw;
5423 struct ixgbe_hw_stats *hwstats = &adapter->stats;
5425 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5426 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
5427 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
5428 u64 bytes = 0, packets = 0;
5430 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5431 test_bit(__IXGBE_RESETTING, &adapter->state))
5434 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5437 for (i = 0; i < 16; i++)
5438 adapter->hw_rx_no_dma_resources +=
5439 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5440 for (i = 0; i < adapter->num_rx_queues; i++) {
5441 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
5442 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
5444 adapter->rsc_total_count = rsc_count;
5445 adapter->rsc_total_flush = rsc_flush;
5448 for (i = 0; i < adapter->num_rx_queues; i++) {
5449 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
5450 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
5451 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
5452 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
5453 bytes += rx_ring->stats.bytes;
5454 packets += rx_ring->stats.packets;
5456 adapter->non_eop_descs = non_eop_descs;
5457 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5458 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5459 netdev->stats.rx_bytes = bytes;
5460 netdev->stats.rx_packets = packets;
5464 /* gather some stats to the adapter struct that are per queue */
5465 for (i = 0; i < adapter->num_tx_queues; i++) {
5466 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5467 restart_queue += tx_ring->tx_stats.restart_queue;
5468 tx_busy += tx_ring->tx_stats.tx_busy;
5469 bytes += tx_ring->stats.bytes;
5470 packets += tx_ring->stats.packets;
5472 adapter->restart_queue = restart_queue;
5473 adapter->tx_busy = tx_busy;
5474 netdev->stats.tx_bytes = bytes;
5475 netdev->stats.tx_packets = packets;
5477 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5478 for (i = 0; i < 8; i++) {
5479 /* for packet buffers not used, the register should read 0 */
5480 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5482 hwstats->mpc[i] += mpc;
5483 total_mpc += hwstats->mpc[i];
5484 if (hw->mac.type == ixgbe_mac_82598EB)
5485 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5486 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5487 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5488 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5489 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5490 switch (hw->mac.type) {
5491 case ixgbe_mac_82598EB:
5492 hwstats->pxonrxc[i] +=
5493 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5495 case ixgbe_mac_82599EB:
5496 case ixgbe_mac_X540:
5497 hwstats->pxonrxc[i] +=
5498 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5503 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5504 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5506 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5507 /* work around hardware counting issue */
5508 hwstats->gprc -= missed_rx;
5510 ixgbe_update_xoff_received(adapter);
5512 /* 82598 hardware only has a 32 bit counter in the high register */
5513 switch (hw->mac.type) {
5514 case ixgbe_mac_82598EB:
5515 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5516 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5517 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5518 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5520 case ixgbe_mac_X540:
5521 /* OS2BMC stats are X540 only*/
5522 hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
5523 hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
5524 hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
5525 hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
5526 case ixgbe_mac_82599EB:
5527 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5528 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
5529 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5530 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
5531 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5532 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5533 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5534 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5535 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5537 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5538 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5539 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5540 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5541 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5542 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5543 #endif /* IXGBE_FCOE */
5548 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5549 hwstats->bprc += bprc;
5550 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5551 if (hw->mac.type == ixgbe_mac_82598EB)
5552 hwstats->mprc -= bprc;
5553 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5554 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5555 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5556 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5557 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5558 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5559 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5560 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5561 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5562 hwstats->lxontxc += lxon;
5563 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5564 hwstats->lxofftxc += lxoff;
5565 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5566 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5567 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5569 * 82598 errata - tx of flow control packets is included in tx counters
5571 xon_off_tot = lxon + lxoff;
5572 hwstats->gptc -= xon_off_tot;
5573 hwstats->mptc -= xon_off_tot;
5574 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5575 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5576 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5577 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5578 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5579 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5580 hwstats->ptc64 -= xon_off_tot;
5581 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5582 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5583 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5584 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5585 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5586 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5588 /* Fill out the OS statistics structure */
5589 netdev->stats.multicast = hwstats->mprc;
5592 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
5593 netdev->stats.rx_dropped = 0;
5594 netdev->stats.rx_length_errors = hwstats->rlec;
5595 netdev->stats.rx_crc_errors = hwstats->crcerrs;
5596 netdev->stats.rx_missed_errors = total_mpc;
5600 * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
5601 * @adapter - pointer to the device adapter structure
5603 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
5605 struct ixgbe_hw *hw = &adapter->hw;
5608 if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
5611 adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
5613 /* if interface is down do nothing */
5614 if (test_bit(__IXGBE_DOWN, &adapter->state))
5617 /* do nothing if we are not using signature filters */
5618 if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
5621 adapter->fdir_overflow++;
5623 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5624 for (i = 0; i < adapter->num_tx_queues; i++)
5625 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
5626 &(adapter->tx_ring[i]->state));
5627 /* re-enable flow director interrupts */
5628 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
5630 e_err(probe, "failed to finish FDIR re-initialization, "
5631 "ignored adding FDIR ATR filters\n");
5636 * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
5637 * @adapter - pointer to the device adapter structure
5639 * This function serves two purposes. First it strobes the interrupt lines
5640 * in order to make certain interrupts are occuring. Secondly it sets the
5641 * bits needed to check for TX hangs. As a result we should immediately
5642 * determine if a hang has occured.
5644 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
5646 struct ixgbe_hw *hw = &adapter->hw;
5650 /* If we're down or resetting, just bail */
5651 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5652 test_bit(__IXGBE_RESETTING, &adapter->state))
5655 /* Force detection of hung controller */
5656 if (netif_carrier_ok(adapter->netdev)) {
5657 for (i = 0; i < adapter->num_tx_queues; i++)
5658 set_check_for_tx_hang(adapter->tx_ring[i]);
5661 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5663 * for legacy and MSI interrupts don't set any bits
5664 * that are enabled for EIAM, because this operation
5665 * would set *both* EIMS and EICS for any bit in EIAM
5667 IXGBE_WRITE_REG(hw, IXGBE_EICS,
5668 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5670 /* get one bit for every active tx/rx interrupt vector */
5671 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
5672 struct ixgbe_q_vector *qv = adapter->q_vector[i];
5673 if (qv->rx.ring || qv->tx.ring)
5674 eics |= ((u64)1 << i);
5678 /* Cause software interrupt to ensure rings are cleaned */
5679 ixgbe_irq_rearm_queues(adapter, eics);
5684 * ixgbe_watchdog_update_link - update the link status
5685 * @adapter - pointer to the device adapter structure
5686 * @link_speed - pointer to a u32 to store the link_speed
5688 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
5690 struct ixgbe_hw *hw = &adapter->hw;
5691 u32 link_speed = adapter->link_speed;
5692 bool link_up = adapter->link_up;
5695 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
5698 if (hw->mac.ops.check_link) {
5699 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5701 /* always assume link is up, if no check link function */
5702 link_speed = IXGBE_LINK_SPEED_10GB_FULL;
5706 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5707 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
5708 hw->mac.ops.fc_enable(hw, i);
5710 hw->mac.ops.fc_enable(hw, 0);
5715 time_after(jiffies, (adapter->link_check_timeout +
5716 IXGBE_TRY_LINK_TIMEOUT))) {
5717 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5718 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5719 IXGBE_WRITE_FLUSH(hw);
5722 adapter->link_up = link_up;
5723 adapter->link_speed = link_speed;
5727 * ixgbe_watchdog_link_is_up - update netif_carrier status and
5728 * print link up message
5729 * @adapter - pointer to the device adapter structure
5731 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
5733 struct net_device *netdev = adapter->netdev;
5734 struct ixgbe_hw *hw = &adapter->hw;
5735 u32 link_speed = adapter->link_speed;
5736 bool flow_rx, flow_tx;
5738 /* only continue if link was previously down */
5739 if (netif_carrier_ok(netdev))
5742 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
5744 switch (hw->mac.type) {
5745 case ixgbe_mac_82598EB: {
5746 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5747 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5748 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5749 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5752 case ixgbe_mac_X540:
5753 case ixgbe_mac_82599EB: {
5754 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5755 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5756 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5757 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5765 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
5766 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5768 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5770 (link_speed == IXGBE_LINK_SPEED_100_FULL ?
5773 ((flow_rx && flow_tx) ? "RX/TX" :
5775 (flow_tx ? "TX" : "None"))));
5777 netif_carrier_on(netdev);
5778 ixgbe_check_vf_rate_limit(adapter);
5782 * ixgbe_watchdog_link_is_down - update netif_carrier status and
5783 * print link down message
5784 * @adapter - pointer to the adapter structure
5786 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter* adapter)
5788 struct net_device *netdev = adapter->netdev;
5789 struct ixgbe_hw *hw = &adapter->hw;
5791 adapter->link_up = false;
5792 adapter->link_speed = 0;
5794 /* only continue if link was up previously */
5795 if (!netif_carrier_ok(netdev))
5798 /* poll for SFP+ cable when link is down */
5799 if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
5800 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5802 e_info(drv, "NIC Link is Down\n");
5803 netif_carrier_off(netdev);
5807 * ixgbe_watchdog_flush_tx - flush queues on link down
5808 * @adapter - pointer to the device adapter structure
5810 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
5813 int some_tx_pending = 0;
5815 if (!netif_carrier_ok(adapter->netdev)) {
5816 for (i = 0; i < adapter->num_tx_queues; i++) {
5817 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5818 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5819 some_tx_pending = 1;
5824 if (some_tx_pending) {
5825 /* We've lost link, so the controller stops DMA,
5826 * but we've got queued Tx work that's never going
5827 * to get done, so reset controller to flush Tx.
5828 * (Do the reset outside of interrupt context).
5830 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
5835 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
5839 /* Do not perform spoof check for 82598 */
5840 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
5843 ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
5846 * ssvpc register is cleared on read, if zero then no
5847 * spoofed packets in the last interval.
5852 e_warn(drv, "%d Spoofed packets detected\n", ssvpc);
5856 * ixgbe_watchdog_subtask - check and bring link up
5857 * @adapter - pointer to the device adapter structure
5859 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
5861 /* if interface is down do nothing */
5862 if (test_bit(__IXGBE_DOWN, &adapter->state))
5865 ixgbe_watchdog_update_link(adapter);
5867 if (adapter->link_up)
5868 ixgbe_watchdog_link_is_up(adapter);
5870 ixgbe_watchdog_link_is_down(adapter);
5872 ixgbe_spoof_check(adapter);
5873 ixgbe_update_stats(adapter);
5875 ixgbe_watchdog_flush_tx(adapter);
5879 * ixgbe_sfp_detection_subtask - poll for SFP+ cable
5880 * @adapter - the ixgbe adapter structure
5882 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
5884 struct ixgbe_hw *hw = &adapter->hw;
5887 /* not searching for SFP so there is nothing to do here */
5888 if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
5889 !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
5892 /* someone else is in init, wait until next service event */
5893 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5896 err = hw->phy.ops.identify_sfp(hw);
5897 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
5900 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
5901 /* If no cable is present, then we need to reset
5902 * the next time we find a good cable. */
5903 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
5910 /* exit if reset not needed */
5911 if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
5914 adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
5917 * A module may be identified correctly, but the EEPROM may not have
5918 * support for that module. setup_sfp() will fail in that case, so
5919 * we should not allow that module to load.
5921 if (hw->mac.type == ixgbe_mac_82598EB)
5922 err = hw->phy.ops.reset(hw);
5924 err = hw->mac.ops.setup_sfp(hw);
5926 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
5929 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
5930 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
5933 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5935 if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
5936 (adapter->netdev->reg_state == NETREG_REGISTERED)) {
5937 e_dev_err("failed to initialize because an unsupported "
5938 "SFP+ module type was detected.\n");
5939 e_dev_err("Reload the driver after installing a "
5940 "supported module.\n");
5941 unregister_netdev(adapter->netdev);
5946 * ixgbe_sfp_link_config_subtask - set up link SFP after module install
5947 * @adapter - the ixgbe adapter structure
5949 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
5951 struct ixgbe_hw *hw = &adapter->hw;
5955 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
5958 /* someone else is in init, wait until next service event */
5959 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5962 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
5964 autoneg = hw->phy.autoneg_advertised;
5965 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
5966 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
5967 hw->mac.autotry_restart = false;
5968 if (hw->mac.ops.setup_link)
5969 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
5971 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5972 adapter->link_check_timeout = jiffies;
5973 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5977 * ixgbe_service_timer - Timer Call-back
5978 * @data: pointer to adapter cast into an unsigned long
5980 static void ixgbe_service_timer(unsigned long data)
5982 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5983 unsigned long next_event_offset;
5985 /* poll faster when waiting for link */
5986 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
5987 next_event_offset = HZ / 10;
5989 next_event_offset = HZ * 2;
5991 /* Reset the timer */
5992 mod_timer(&adapter->service_timer, next_event_offset + jiffies);
5994 ixgbe_service_event_schedule(adapter);
5997 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
5999 if (!(adapter->flags2 & IXGBE_FLAG2_RESET_REQUESTED))
6002 adapter->flags2 &= ~IXGBE_FLAG2_RESET_REQUESTED;
6004 /* If we're already down or resetting, just bail */
6005 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6006 test_bit(__IXGBE_RESETTING, &adapter->state))
6009 ixgbe_dump(adapter);
6010 netdev_err(adapter->netdev, "Reset adapter\n");
6011 adapter->tx_timeout_count++;
6013 ixgbe_reinit_locked(adapter);
6017 * ixgbe_service_task - manages and runs subtasks
6018 * @work: pointer to work_struct containing our data
6020 static void ixgbe_service_task(struct work_struct *work)
6022 struct ixgbe_adapter *adapter = container_of(work,
6023 struct ixgbe_adapter,
6026 ixgbe_reset_subtask(adapter);
6027 ixgbe_sfp_detection_subtask(adapter);
6028 ixgbe_sfp_link_config_subtask(adapter);
6029 ixgbe_check_overtemp_subtask(adapter);
6030 ixgbe_watchdog_subtask(adapter);
6031 ixgbe_fdir_reinit_subtask(adapter);
6032 ixgbe_check_hang_subtask(adapter);
6034 ixgbe_service_event_complete(adapter);
6037 void ixgbe_tx_ctxtdesc(struct ixgbe_ring *tx_ring, u32 vlan_macip_lens,
6038 u32 fcoe_sof_eof, u32 type_tucmd, u32 mss_l4len_idx)
6040 struct ixgbe_adv_tx_context_desc *context_desc;
6041 u16 i = tx_ring->next_to_use;
6043 context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
6046 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
6048 /* set bits to identify this as an advanced context descriptor */
6049 type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
6051 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
6052 context_desc->seqnum_seed = cpu_to_le32(fcoe_sof_eof);
6053 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd);
6054 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
6057 static int ixgbe_tso(struct ixgbe_ring *tx_ring, struct sk_buff *skb,
6058 u32 tx_flags, __be16 protocol, u8 *hdr_len)
6061 u32 vlan_macip_lens, type_tucmd;
6062 u32 mss_l4len_idx, l4len;
6064 if (!skb_is_gso(skb))
6067 if (skb_header_cloned(skb)) {
6068 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
6073 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
6074 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
6076 if (protocol == __constant_htons(ETH_P_IP)) {
6077 struct iphdr *iph = ip_hdr(skb);
6080 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
6084 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
6085 } else if (skb_is_gso_v6(skb)) {
6086 ipv6_hdr(skb)->payload_len = 0;
6087 tcp_hdr(skb)->check =
6088 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
6089 &ipv6_hdr(skb)->daddr,
6093 l4len = tcp_hdrlen(skb);
6094 *hdr_len = skb_transport_offset(skb) + l4len;
6096 /* mss_l4len_id: use 1 as index for TSO */
6097 mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
6098 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
6099 mss_l4len_idx |= 1 << IXGBE_ADVTXD_IDX_SHIFT;
6101 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
6102 vlan_macip_lens = skb_network_header_len(skb);
6103 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
6104 vlan_macip_lens |= tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
6106 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0, type_tucmd,
6112 static bool ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
6113 struct sk_buff *skb, u32 tx_flags,
6116 u32 vlan_macip_lens = 0;
6117 u32 mss_l4len_idx = 0;
6120 if (skb->ip_summed != CHECKSUM_PARTIAL) {
6121 if (!(tx_flags & IXGBE_TX_FLAGS_HW_VLAN) &&
6122 !(tx_flags & IXGBE_TX_FLAGS_TXSW))
6127 case __constant_htons(ETH_P_IP):
6128 vlan_macip_lens |= skb_network_header_len(skb);
6129 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
6130 l4_hdr = ip_hdr(skb)->protocol;
6132 case __constant_htons(ETH_P_IPV6):
6133 vlan_macip_lens |= skb_network_header_len(skb);
6134 l4_hdr = ipv6_hdr(skb)->nexthdr;
6137 if (unlikely(net_ratelimit())) {
6138 dev_warn(tx_ring->dev,
6139 "partial checksum but proto=%x!\n",
6147 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6148 mss_l4len_idx = tcp_hdrlen(skb) <<
6149 IXGBE_ADVTXD_L4LEN_SHIFT;
6152 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
6153 mss_l4len_idx = sizeof(struct sctphdr) <<
6154 IXGBE_ADVTXD_L4LEN_SHIFT;
6157 mss_l4len_idx = sizeof(struct udphdr) <<
6158 IXGBE_ADVTXD_L4LEN_SHIFT;
6161 if (unlikely(net_ratelimit())) {
6162 dev_warn(tx_ring->dev,
6163 "partial checksum but l4 proto=%x!\n",
6170 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
6171 vlan_macip_lens |= tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
6173 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0,
6174 type_tucmd, mss_l4len_idx);
6176 return (skb->ip_summed == CHECKSUM_PARTIAL);
6179 static __le32 ixgbe_tx_cmd_type(u32 tx_flags)
6181 /* set type for advanced descriptor with frame checksum insertion */
6182 __le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
6183 IXGBE_ADVTXD_DCMD_IFCS |
6184 IXGBE_ADVTXD_DCMD_DEXT);
6186 /* set HW vlan bit if vlan is present */
6187 if (tx_flags & IXGBE_TX_FLAGS_HW_VLAN)
6188 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
6190 /* set segmentation enable bits for TSO/FSO */
6192 if ((tx_flags & IXGBE_TX_FLAGS_TSO) || (tx_flags & IXGBE_TX_FLAGS_FSO))
6194 if (tx_flags & IXGBE_TX_FLAGS_TSO)
6196 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
6201 static __le32 ixgbe_tx_olinfo_status(u32 tx_flags, unsigned int paylen)
6203 __le32 olinfo_status =
6204 cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
6206 if (tx_flags & IXGBE_TX_FLAGS_TSO) {
6207 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM |
6208 (1 << IXGBE_ADVTXD_IDX_SHIFT));
6209 /* enble IPv4 checksum for TSO */
6210 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
6211 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
6214 /* enable L4 checksum for TSO and TX checksum offload */
6215 if (tx_flags & IXGBE_TX_FLAGS_CSUM)
6216 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
6219 /* use index 1 context for FCOE/FSO */
6220 if (tx_flags & IXGBE_TX_FLAGS_FCOE)
6221 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC |
6222 (1 << IXGBE_ADVTXD_IDX_SHIFT));
6226 * Check Context must be set if Tx switch is enabled, which it
6227 * always is for case where virtual functions are running
6229 if (tx_flags & IXGBE_TX_FLAGS_TXSW)
6230 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
6232 return olinfo_status;
6235 #define IXGBE_TXD_CMD (IXGBE_TXD_CMD_EOP | \
6238 static void ixgbe_tx_map(struct ixgbe_ring *tx_ring,
6239 struct sk_buff *skb,
6240 struct ixgbe_tx_buffer *first,
6244 struct device *dev = tx_ring->dev;
6245 struct ixgbe_tx_buffer *tx_buffer_info;
6246 union ixgbe_adv_tx_desc *tx_desc;
6248 __le32 cmd_type, olinfo_status;
6249 struct skb_frag_struct *frag;
6251 unsigned int data_len = skb->data_len;
6252 unsigned int size = skb_headlen(skb);
6254 u32 paylen = skb->len - hdr_len;
6255 u16 i = tx_ring->next_to_use;
6259 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6260 if (data_len >= sizeof(struct fcoe_crc_eof)) {
6261 data_len -= sizeof(struct fcoe_crc_eof);
6263 size -= sizeof(struct fcoe_crc_eof) - data_len;
6269 dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE);
6270 if (dma_mapping_error(dev, dma))
6273 cmd_type = ixgbe_tx_cmd_type(tx_flags);
6274 olinfo_status = ixgbe_tx_olinfo_status(tx_flags, paylen);
6276 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
6279 while (size > IXGBE_MAX_DATA_PER_TXD) {
6280 tx_desc->read.buffer_addr = cpu_to_le64(dma + offset);
6281 tx_desc->read.cmd_type_len =
6282 cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
6283 tx_desc->read.olinfo_status = olinfo_status;
6285 offset += IXGBE_MAX_DATA_PER_TXD;
6286 size -= IXGBE_MAX_DATA_PER_TXD;
6290 if (i == tx_ring->count) {
6291 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, 0);
6296 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6297 tx_buffer_info->length = offset + size;
6298 tx_buffer_info->tx_flags = tx_flags;
6299 tx_buffer_info->dma = dma;
6301 tx_desc->read.buffer_addr = cpu_to_le64(dma + offset);
6302 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
6303 tx_desc->read.olinfo_status = olinfo_status;
6308 frag = &skb_shinfo(skb)->frags[f];
6310 size = min_t(unsigned int, data_len, frag->size);
6318 tx_flags |= IXGBE_TX_FLAGS_MAPPED_AS_PAGE;
6320 dma = dma_map_page(dev, frag->page, frag->page_offset,
6321 size, DMA_TO_DEVICE);
6322 if (dma_mapping_error(dev, dma))
6327 if (i == tx_ring->count) {
6328 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, 0);
6333 tx_desc->read.cmd_type_len |= cpu_to_le32(IXGBE_TXD_CMD);
6336 if (i == tx_ring->count)
6339 tx_ring->next_to_use = i;
6341 if (tx_flags & IXGBE_TX_FLAGS_TSO)
6342 gso_segs = skb_shinfo(skb)->gso_segs;
6344 /* adjust for FCoE Sequence Offload */
6345 else if (tx_flags & IXGBE_TX_FLAGS_FSO)
6346 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
6347 skb_shinfo(skb)->gso_size);
6348 #endif /* IXGBE_FCOE */
6352 /* multiply data chunks by size of headers */
6353 tx_buffer_info->bytecount = paylen + (gso_segs * hdr_len);
6354 tx_buffer_info->gso_segs = gso_segs;
6355 tx_buffer_info->skb = skb;
6357 /* set the timestamp */
6358 first->time_stamp = jiffies;
6361 * Force memory writes to complete before letting h/w
6362 * know there are new descriptors to fetch. (Only
6363 * applicable for weak-ordered memory model archs,
6368 /* set next_to_watch value indicating a packet is present */
6369 first->next_to_watch = tx_desc;
6371 /* notify HW of packet */
6372 writel(i, tx_ring->tail);
6376 dev_err(dev, "TX DMA map failed\n");
6378 /* clear dma mappings for failed tx_buffer_info map */
6380 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6381 ixgbe_unmap_tx_resource(tx_ring, tx_buffer_info);
6382 if (tx_buffer_info == first)
6389 dev_kfree_skb_any(skb);
6391 tx_ring->next_to_use = i;
6394 static void ixgbe_atr(struct ixgbe_ring *ring, struct sk_buff *skb,
6395 u32 tx_flags, __be16 protocol)
6397 struct ixgbe_q_vector *q_vector = ring->q_vector;
6398 union ixgbe_atr_hash_dword input = { .dword = 0 };
6399 union ixgbe_atr_hash_dword common = { .dword = 0 };
6401 unsigned char *network;
6403 struct ipv6hdr *ipv6;
6408 /* if ring doesn't have a interrupt vector, cannot perform ATR */
6412 /* do nothing if sampling is disabled */
6413 if (!ring->atr_sample_rate)
6418 /* snag network header to get L4 type and address */
6419 hdr.network = skb_network_header(skb);
6421 /* Currently only IPv4/IPv6 with TCP is supported */
6422 if ((protocol != __constant_htons(ETH_P_IPV6) ||
6423 hdr.ipv6->nexthdr != IPPROTO_TCP) &&
6424 (protocol != __constant_htons(ETH_P_IP) ||
6425 hdr.ipv4->protocol != IPPROTO_TCP))
6430 /* skip this packet since it is invalid or the socket is closing */
6434 /* sample on all syn packets or once every atr sample count */
6435 if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
6438 /* reset sample count */
6439 ring->atr_count = 0;
6441 vlan_id = htons(tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
6444 * src and dst are inverted, think how the receiver sees them
6446 * The input is broken into two sections, a non-compressed section
6447 * containing vm_pool, vlan_id, and flow_type. The rest of the data
6448 * is XORed together and stored in the compressed dword.
6450 input.formatted.vlan_id = vlan_id;
6453 * since src port and flex bytes occupy the same word XOR them together
6454 * and write the value to source port portion of compressed dword
6456 if (tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
6457 common.port.src ^= th->dest ^ __constant_htons(ETH_P_8021Q);
6459 common.port.src ^= th->dest ^ protocol;
6460 common.port.dst ^= th->source;
6462 if (protocol == __constant_htons(ETH_P_IP)) {
6463 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
6464 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
6466 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
6467 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
6468 hdr.ipv6->saddr.s6_addr32[1] ^
6469 hdr.ipv6->saddr.s6_addr32[2] ^
6470 hdr.ipv6->saddr.s6_addr32[3] ^
6471 hdr.ipv6->daddr.s6_addr32[0] ^
6472 hdr.ipv6->daddr.s6_addr32[1] ^
6473 hdr.ipv6->daddr.s6_addr32[2] ^
6474 hdr.ipv6->daddr.s6_addr32[3];
6477 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
6478 ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
6479 input, common, ring->queue_index);
6482 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6484 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
6485 /* Herbert's original patch had:
6486 * smp_mb__after_netif_stop_queue();
6487 * but since that doesn't exist yet, just open code it. */
6490 /* We need to check again in a case another CPU has just
6491 * made room available. */
6492 if (likely(ixgbe_desc_unused(tx_ring) < size))
6495 /* A reprieve! - use start_queue because it doesn't call schedule */
6496 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
6497 ++tx_ring->tx_stats.restart_queue;
6501 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6503 if (likely(ixgbe_desc_unused(tx_ring) >= size))
6505 return __ixgbe_maybe_stop_tx(tx_ring, size);
6508 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6510 struct ixgbe_adapter *adapter = netdev_priv(dev);
6511 int txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
6514 __be16 protocol = vlan_get_protocol(skb);
6516 if (((protocol == htons(ETH_P_FCOE)) ||
6517 (protocol == htons(ETH_P_FIP))) &&
6518 (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)) {
6519 txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
6520 txq += adapter->ring_feature[RING_F_FCOE].mask;
6525 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6526 while (unlikely(txq >= dev->real_num_tx_queues))
6527 txq -= dev->real_num_tx_queues;
6531 return skb_tx_hash(dev, skb);
6534 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
6535 struct ixgbe_adapter *adapter,
6536 struct ixgbe_ring *tx_ring)
6538 struct ixgbe_tx_buffer *first;
6541 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6544 u16 count = TXD_USE_COUNT(skb_headlen(skb));
6545 __be16 protocol = skb->protocol;
6549 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
6550 * + 1 desc for skb_head_len/IXGBE_MAX_DATA_PER_TXD,
6551 * + 2 desc gap to keep tail from touching head,
6552 * + 1 desc for context descriptor,
6553 * otherwise try next time
6555 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6556 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6557 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6559 count += skb_shinfo(skb)->nr_frags;
6561 if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
6562 tx_ring->tx_stats.tx_busy++;
6563 return NETDEV_TX_BUSY;
6566 #ifdef CONFIG_PCI_IOV
6567 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6568 tx_flags |= IXGBE_TX_FLAGS_TXSW;
6571 /* if we have a HW VLAN tag being added default to the HW one */
6572 if (vlan_tx_tag_present(skb)) {
6573 tx_flags |= vlan_tx_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
6574 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6575 /* else if it is a SW VLAN check the next protocol and store the tag */
6576 } else if (protocol == __constant_htons(ETH_P_8021Q)) {
6577 struct vlan_hdr *vhdr, _vhdr;
6578 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
6582 protocol = vhdr->h_vlan_encapsulated_proto;
6583 tx_flags |= ntohs(vhdr->h_vlan_TCI) << IXGBE_TX_FLAGS_VLAN_SHIFT;
6584 tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
6587 if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
6588 ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
6589 (skb->priority != TC_PRIO_CONTROL))) {
6590 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6591 tx_flags |= tx_ring->dcb_tc <<
6592 IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
6593 if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
6594 struct vlan_ethhdr *vhdr;
6595 if (skb_header_cloned(skb) &&
6596 pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
6598 vhdr = (struct vlan_ethhdr *)skb->data;
6599 vhdr->h_vlan_TCI = htons(tx_flags >>
6600 IXGBE_TX_FLAGS_VLAN_SHIFT);
6602 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6606 /* record the location of the first descriptor for this packet */
6607 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
6610 /* setup tx offload for FCoE */
6611 if ((protocol == __constant_htons(ETH_P_FCOE)) &&
6612 (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)) {
6613 tso = ixgbe_fso(tx_ring, skb, tx_flags, &hdr_len);
6617 tx_flags |= IXGBE_TX_FLAGS_FSO |
6618 IXGBE_TX_FLAGS_FCOE;
6620 tx_flags |= IXGBE_TX_FLAGS_FCOE;
6625 #endif /* IXGBE_FCOE */
6626 /* setup IPv4/IPv6 offloads */
6627 if (protocol == __constant_htons(ETH_P_IP))
6628 tx_flags |= IXGBE_TX_FLAGS_IPV4;
6630 tso = ixgbe_tso(tx_ring, skb, tx_flags, protocol, &hdr_len);
6634 tx_flags |= IXGBE_TX_FLAGS_TSO;
6635 else if (ixgbe_tx_csum(tx_ring, skb, tx_flags, protocol))
6636 tx_flags |= IXGBE_TX_FLAGS_CSUM;
6638 /* add the ATR filter if ATR is on */
6639 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
6640 ixgbe_atr(tx_ring, skb, tx_flags, protocol);
6644 #endif /* IXGBE_FCOE */
6645 ixgbe_tx_map(tx_ring, skb, first, tx_flags, hdr_len);
6647 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
6649 return NETDEV_TX_OK;
6652 dev_kfree_skb_any(skb);
6653 return NETDEV_TX_OK;
6656 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
6658 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6659 struct ixgbe_ring *tx_ring;
6661 tx_ring = adapter->tx_ring[skb->queue_mapping];
6662 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
6666 * ixgbe_set_mac - Change the Ethernet Address of the NIC
6667 * @netdev: network interface device structure
6668 * @p: pointer to an address structure
6670 * Returns 0 on success, negative on failure
6672 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6674 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6675 struct ixgbe_hw *hw = &adapter->hw;
6676 struct sockaddr *addr = p;
6678 if (!is_valid_ether_addr(addr->sa_data))
6679 return -EADDRNOTAVAIL;
6681 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6682 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6684 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
6691 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6693 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6694 struct ixgbe_hw *hw = &adapter->hw;
6698 if (prtad != hw->phy.mdio.prtad)
6700 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6706 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6707 u16 addr, u16 value)
6709 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6710 struct ixgbe_hw *hw = &adapter->hw;
6712 if (prtad != hw->phy.mdio.prtad)
6714 return hw->phy.ops.write_reg(hw, addr, devad, value);
6717 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6719 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6721 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6725 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6727 * @netdev: network interface device structure
6729 * Returns non-zero on failure
6731 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6734 struct ixgbe_adapter *adapter = netdev_priv(dev);
6735 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6737 if (is_valid_ether_addr(mac->san_addr)) {
6739 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6746 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6748 * @netdev: network interface device structure
6750 * Returns non-zero on failure
6752 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6755 struct ixgbe_adapter *adapter = netdev_priv(dev);
6756 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6758 if (is_valid_ether_addr(mac->san_addr)) {
6760 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6766 #ifdef CONFIG_NET_POLL_CONTROLLER
6768 * Polling 'interrupt' - used by things like netconsole to send skbs
6769 * without having to re-enable interrupts. It's not called while
6770 * the interrupt routine is executing.
6772 static void ixgbe_netpoll(struct net_device *netdev)
6774 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6777 /* if interface is down do nothing */
6778 if (test_bit(__IXGBE_DOWN, &adapter->state))
6781 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
6782 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
6783 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
6784 for (i = 0; i < num_q_vectors; i++) {
6785 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
6786 ixgbe_msix_clean_rings(0, q_vector);
6789 ixgbe_intr(adapter->pdev->irq, netdev);
6791 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
6795 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
6796 struct rtnl_link_stats64 *stats)
6798 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6802 for (i = 0; i < adapter->num_rx_queues; i++) {
6803 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
6809 start = u64_stats_fetch_begin_bh(&ring->syncp);
6810 packets = ring->stats.packets;
6811 bytes = ring->stats.bytes;
6812 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6813 stats->rx_packets += packets;
6814 stats->rx_bytes += bytes;
6818 for (i = 0; i < adapter->num_tx_queues; i++) {
6819 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->tx_ring[i]);
6825 start = u64_stats_fetch_begin_bh(&ring->syncp);
6826 packets = ring->stats.packets;
6827 bytes = ring->stats.bytes;
6828 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6829 stats->tx_packets += packets;
6830 stats->tx_bytes += bytes;
6834 /* following stats updated by ixgbe_watchdog_task() */
6835 stats->multicast = netdev->stats.multicast;
6836 stats->rx_errors = netdev->stats.rx_errors;
6837 stats->rx_length_errors = netdev->stats.rx_length_errors;
6838 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
6839 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
6843 /* ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
6844 * #adapter: pointer to ixgbe_adapter
6845 * @tc: number of traffic classes currently enabled
6847 * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
6848 * 802.1Q priority maps to a packet buffer that exists.
6850 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
6852 struct ixgbe_hw *hw = &adapter->hw;
6856 /* 82598 have a static priority to TC mapping that can not
6857 * be changed so no validation is needed.
6859 if (hw->mac.type == ixgbe_mac_82598EB)
6862 reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
6865 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
6866 u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
6868 /* If up2tc is out of bounds default to zero */
6870 reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
6874 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
6880 /* ixgbe_setup_tc - routine to configure net_device for multiple traffic
6883 * @netdev: net device to configure
6884 * @tc: number of traffic classes to enable
6886 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
6888 struct ixgbe_adapter *adapter = netdev_priv(dev);
6889 struct ixgbe_hw *hw = &adapter->hw;
6891 /* Multiple traffic classes requires multiple queues */
6892 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
6893 e_err(drv, "Enable failed, needs MSI-X\n");
6897 /* Hardware supports up to 8 traffic classes */
6898 if (tc > MAX_TRAFFIC_CLASS ||
6899 (hw->mac.type == ixgbe_mac_82598EB && tc < MAX_TRAFFIC_CLASS))
6902 /* Hardware has to reinitialize queues and interrupts to
6903 * match packet buffer alignment. Unfortunantly, the
6904 * hardware is not flexible enough to do this dynamically.
6906 if (netif_running(dev))
6908 ixgbe_clear_interrupt_scheme(adapter);
6911 netdev_set_num_tc(dev, tc);
6912 adapter->last_lfc_mode = adapter->hw.fc.current_mode;
6914 adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
6915 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
6917 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
6918 adapter->hw.fc.requested_mode = ixgbe_fc_none;
6920 netdev_reset_tc(dev);
6922 adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
6924 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
6925 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
6927 adapter->temp_dcb_cfg.pfc_mode_enable = false;
6928 adapter->dcb_cfg.pfc_mode_enable = false;
6931 ixgbe_init_interrupt_scheme(adapter);
6932 ixgbe_validate_rtr(adapter, tc);
6933 if (netif_running(dev))
6939 void ixgbe_do_reset(struct net_device *netdev)
6941 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6943 if (netif_running(netdev))
6944 ixgbe_reinit_locked(adapter);
6946 ixgbe_reset(adapter);
6949 static u32 ixgbe_fix_features(struct net_device *netdev, u32 data)
6951 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6954 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
6955 data &= ~NETIF_F_HW_VLAN_RX;
6958 /* return error if RXHASH is being enabled when RSS is not supported */
6959 if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED))
6960 data &= ~NETIF_F_RXHASH;
6962 /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
6963 if (!(data & NETIF_F_RXCSUM))
6964 data &= ~NETIF_F_LRO;
6966 /* Turn off LRO if not RSC capable or invalid ITR settings */
6967 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)) {
6968 data &= ~NETIF_F_LRO;
6969 } else if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
6970 (adapter->rx_itr_setting != 1 &&
6971 adapter->rx_itr_setting > IXGBE_MAX_RSC_INT_RATE)) {
6972 data &= ~NETIF_F_LRO;
6973 e_info(probe, "rx-usecs set too low, not enabling RSC\n");
6979 static int ixgbe_set_features(struct net_device *netdev, u32 data)
6981 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6982 bool need_reset = false;
6984 /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
6985 if (!(data & NETIF_F_RXCSUM))
6986 adapter->flags &= ~IXGBE_FLAG_RX_CSUM_ENABLED;
6988 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
6990 /* Make sure RSC matches LRO, reset if change */
6991 if (!!(data & NETIF_F_LRO) !=
6992 !!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
6993 adapter->flags2 ^= IXGBE_FLAG2_RSC_ENABLED;
6994 switch (adapter->hw.mac.type) {
6995 case ixgbe_mac_X540:
6996 case ixgbe_mac_82599EB:
7005 * Check if Flow Director n-tuple support was enabled or disabled. If
7006 * the state changed, we need to reset.
7008 if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)) {
7009 /* turn off ATR, enable perfect filters and reset */
7010 if (data & NETIF_F_NTUPLE) {
7011 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
7012 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
7015 } else if (!(data & NETIF_F_NTUPLE)) {
7016 /* turn off Flow Director, set ATR and reset */
7017 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
7018 if ((adapter->flags & IXGBE_FLAG_RSS_ENABLED) &&
7019 !(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
7020 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
7025 ixgbe_do_reset(netdev);
7031 static const struct net_device_ops ixgbe_netdev_ops = {
7032 .ndo_open = ixgbe_open,
7033 .ndo_stop = ixgbe_close,
7034 .ndo_start_xmit = ixgbe_xmit_frame,
7035 .ndo_select_queue = ixgbe_select_queue,
7036 .ndo_set_rx_mode = ixgbe_set_rx_mode,
7037 .ndo_validate_addr = eth_validate_addr,
7038 .ndo_set_mac_address = ixgbe_set_mac,
7039 .ndo_change_mtu = ixgbe_change_mtu,
7040 .ndo_tx_timeout = ixgbe_tx_timeout,
7041 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
7042 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
7043 .ndo_do_ioctl = ixgbe_ioctl,
7044 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
7045 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
7046 .ndo_set_vf_tx_rate = ixgbe_ndo_set_vf_bw,
7047 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
7048 .ndo_get_stats64 = ixgbe_get_stats64,
7049 .ndo_setup_tc = ixgbe_setup_tc,
7050 #ifdef CONFIG_NET_POLL_CONTROLLER
7051 .ndo_poll_controller = ixgbe_netpoll,
7054 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
7055 .ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
7056 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
7057 .ndo_fcoe_enable = ixgbe_fcoe_enable,
7058 .ndo_fcoe_disable = ixgbe_fcoe_disable,
7059 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
7060 #endif /* IXGBE_FCOE */
7061 .ndo_set_features = ixgbe_set_features,
7062 .ndo_fix_features = ixgbe_fix_features,
7065 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
7066 const struct ixgbe_info *ii)
7068 #ifdef CONFIG_PCI_IOV
7069 struct ixgbe_hw *hw = &adapter->hw;
7071 int num_vf_macvlans, i;
7072 struct vf_macvlans *mv_list;
7074 if (hw->mac.type == ixgbe_mac_82598EB || !max_vfs)
7077 /* The 82599 supports up to 64 VFs per physical function
7078 * but this implementation limits allocation to 63 so that
7079 * basic networking resources are still available to the
7082 adapter->num_vfs = (max_vfs > 63) ? 63 : max_vfs;
7083 adapter->flags |= IXGBE_FLAG_SRIOV_ENABLED;
7084 err = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
7086 e_err(probe, "Failed to enable PCI sriov: %d\n", err);
7090 num_vf_macvlans = hw->mac.num_rar_entries -
7091 (IXGBE_MAX_PF_MACVLANS + 1 + adapter->num_vfs);
7093 adapter->mv_list = mv_list = kcalloc(num_vf_macvlans,
7094 sizeof(struct vf_macvlans),
7097 /* Initialize list of VF macvlans */
7098 INIT_LIST_HEAD(&adapter->vf_mvs.l);
7099 for (i = 0; i < num_vf_macvlans; i++) {
7101 mv_list->free = true;
7102 mv_list->rar_entry = hw->mac.num_rar_entries -
7103 (i + adapter->num_vfs + 1);
7104 list_add(&mv_list->l, &adapter->vf_mvs.l);
7109 /* If call to enable VFs succeeded then allocate memory
7110 * for per VF control structures.
7113 kcalloc(adapter->num_vfs,
7114 sizeof(struct vf_data_storage), GFP_KERNEL);
7115 if (adapter->vfinfo) {
7116 /* Now that we're sure SR-IOV is enabled
7117 * and memory allocated set up the mailbox parameters
7119 ixgbe_init_mbx_params_pf(hw);
7120 memcpy(&hw->mbx.ops, ii->mbx_ops,
7121 sizeof(hw->mbx.ops));
7123 /* Disable RSC when in SR-IOV mode */
7124 adapter->flags2 &= ~(IXGBE_FLAG2_RSC_CAPABLE |
7125 IXGBE_FLAG2_RSC_ENABLED);
7130 e_err(probe, "Unable to allocate memory for VF Data Storage - "
7131 "SRIOV disabled\n");
7132 pci_disable_sriov(adapter->pdev);
7135 adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
7136 adapter->num_vfs = 0;
7137 #endif /* CONFIG_PCI_IOV */
7141 * ixgbe_probe - Device Initialization Routine
7142 * @pdev: PCI device information struct
7143 * @ent: entry in ixgbe_pci_tbl
7145 * Returns 0 on success, negative on failure
7147 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
7148 * The OS initialization, configuring of the adapter private structure,
7149 * and a hardware reset occur.
7151 static int __devinit ixgbe_probe(struct pci_dev *pdev,
7152 const struct pci_device_id *ent)
7154 struct net_device *netdev;
7155 struct ixgbe_adapter *adapter = NULL;
7156 struct ixgbe_hw *hw;
7157 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
7158 static int cards_found;
7159 int i, err, pci_using_dac;
7160 u8 part_str[IXGBE_PBANUM_LENGTH];
7161 unsigned int indices = num_possible_cpus();
7167 /* Catch broken hardware that put the wrong VF device ID in
7168 * the PCIe SR-IOV capability.
7170 if (pdev->is_virtfn) {
7171 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
7172 pci_name(pdev), pdev->vendor, pdev->device);
7176 err = pci_enable_device_mem(pdev);
7180 if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
7181 !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
7184 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
7186 err = dma_set_coherent_mask(&pdev->dev,
7190 "No usable DMA configuration, aborting\n");
7197 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
7198 IORESOURCE_MEM), ixgbe_driver_name);
7201 "pci_request_selected_regions failed 0x%x\n", err);
7205 pci_enable_pcie_error_reporting(pdev);
7207 pci_set_master(pdev);
7208 pci_save_state(pdev);
7210 #ifdef CONFIG_IXGBE_DCB
7211 indices *= MAX_TRAFFIC_CLASS;
7214 if (ii->mac == ixgbe_mac_82598EB)
7215 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
7217 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
7220 indices += min_t(unsigned int, num_possible_cpus(),
7221 IXGBE_MAX_FCOE_INDICES);
7223 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
7226 goto err_alloc_etherdev;
7229 SET_NETDEV_DEV(netdev, &pdev->dev);
7231 adapter = netdev_priv(netdev);
7232 pci_set_drvdata(pdev, adapter);
7234 adapter->netdev = netdev;
7235 adapter->pdev = pdev;
7238 adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
7240 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
7241 pci_resource_len(pdev, 0));
7247 for (i = 1; i <= 5; i++) {
7248 if (pci_resource_len(pdev, i) == 0)
7252 netdev->netdev_ops = &ixgbe_netdev_ops;
7253 ixgbe_set_ethtool_ops(netdev);
7254 netdev->watchdog_timeo = 5 * HZ;
7255 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
7257 adapter->bd_number = cards_found;
7260 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
7261 hw->mac.type = ii->mac;
7264 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
7265 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
7266 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
7267 if (!(eec & (1 << 8)))
7268 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
7271 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
7272 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
7273 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
7274 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
7275 hw->phy.mdio.mmds = 0;
7276 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
7277 hw->phy.mdio.dev = netdev;
7278 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
7279 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
7281 ii->get_invariants(hw);
7283 /* setup the private structure */
7284 err = ixgbe_sw_init(adapter);
7288 /* Make it possible the adapter to be woken up via WOL */
7289 switch (adapter->hw.mac.type) {
7290 case ixgbe_mac_82599EB:
7291 case ixgbe_mac_X540:
7292 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7299 * If there is a fan on this device and it has failed log the
7302 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
7303 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
7304 if (esdp & IXGBE_ESDP_SDP1)
7305 e_crit(probe, "Fan has stopped, replace the adapter\n");
7308 /* reset_hw fills in the perm_addr as well */
7309 hw->phy.reset_if_overtemp = true;
7310 err = hw->mac.ops.reset_hw(hw);
7311 hw->phy.reset_if_overtemp = false;
7312 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
7313 hw->mac.type == ixgbe_mac_82598EB) {
7315 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
7316 e_dev_err("failed to load because an unsupported SFP+ "
7317 "module type was detected.\n");
7318 e_dev_err("Reload the driver after installing a supported "
7322 e_dev_err("HW Init failed: %d\n", err);
7326 ixgbe_probe_vf(adapter, ii);
7328 netdev->features = NETIF_F_SG |
7331 NETIF_F_HW_VLAN_TX |
7332 NETIF_F_HW_VLAN_RX |
7333 NETIF_F_HW_VLAN_FILTER |
7339 netdev->hw_features = netdev->features;
7341 switch (adapter->hw.mac.type) {
7342 case ixgbe_mac_82599EB:
7343 case ixgbe_mac_X540:
7344 netdev->features |= NETIF_F_SCTP_CSUM;
7345 netdev->hw_features |= NETIF_F_SCTP_CSUM |
7352 netdev->vlan_features |= NETIF_F_TSO;
7353 netdev->vlan_features |= NETIF_F_TSO6;
7354 netdev->vlan_features |= NETIF_F_IP_CSUM;
7355 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
7356 netdev->vlan_features |= NETIF_F_SG;
7358 netdev->priv_flags |= IFF_UNICAST_FLT;
7360 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7361 adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
7362 IXGBE_FLAG_DCB_ENABLED);
7364 #ifdef CONFIG_IXGBE_DCB
7365 netdev->dcbnl_ops = &dcbnl_ops;
7369 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7370 if (hw->mac.ops.get_device_caps) {
7371 hw->mac.ops.get_device_caps(hw, &device_caps);
7372 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
7373 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
7376 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7377 netdev->vlan_features |= NETIF_F_FCOE_CRC;
7378 netdev->vlan_features |= NETIF_F_FSO;
7379 netdev->vlan_features |= NETIF_F_FCOE_MTU;
7381 #endif /* IXGBE_FCOE */
7382 if (pci_using_dac) {
7383 netdev->features |= NETIF_F_HIGHDMA;
7384 netdev->vlan_features |= NETIF_F_HIGHDMA;
7387 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
7388 netdev->hw_features |= NETIF_F_LRO;
7389 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
7390 netdev->features |= NETIF_F_LRO;
7392 /* make sure the EEPROM is good */
7393 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
7394 e_dev_err("The EEPROM Checksum Is Not Valid\n");
7399 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
7400 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
7402 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
7403 e_dev_err("invalid MAC address\n");
7408 /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
7409 if (hw->mac.ops.disable_tx_laser &&
7410 ((hw->phy.multispeed_fiber) ||
7411 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
7412 (hw->mac.type == ixgbe_mac_82599EB))))
7413 hw->mac.ops.disable_tx_laser(hw);
7415 setup_timer(&adapter->service_timer, &ixgbe_service_timer,
7416 (unsigned long) adapter);
7418 INIT_WORK(&adapter->service_task, ixgbe_service_task);
7419 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
7421 err = ixgbe_init_interrupt_scheme(adapter);
7425 if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED)) {
7426 netdev->hw_features &= ~NETIF_F_RXHASH;
7427 netdev->features &= ~NETIF_F_RXHASH;
7430 switch (pdev->device) {
7431 case IXGBE_DEV_ID_82599_SFP:
7432 /* Only this subdevice supports WOL */
7433 if (pdev->subsystem_device == IXGBE_SUBDEV_ID_82599_SFP)
7434 adapter->wol = IXGBE_WUFC_MAG;
7436 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
7437 /* All except this subdevice support WOL */
7438 if (pdev->subsystem_device != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
7439 adapter->wol = IXGBE_WUFC_MAG;
7441 case IXGBE_DEV_ID_82599_KX4:
7442 adapter->wol = IXGBE_WUFC_MAG;
7448 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
7450 /* pick up the PCI bus settings for reporting later */
7451 hw->mac.ops.get_bus_info(hw);
7453 /* print bus type/speed/width info */
7454 e_dev_info("(PCI Express:%s:%s) %pM\n",
7455 (hw->bus.speed == ixgbe_bus_speed_5000 ? "5.0GT/s" :
7456 hw->bus.speed == ixgbe_bus_speed_2500 ? "2.5GT/s" :
7458 (hw->bus.width == ixgbe_bus_width_pcie_x8 ? "Width x8" :
7459 hw->bus.width == ixgbe_bus_width_pcie_x4 ? "Width x4" :
7460 hw->bus.width == ixgbe_bus_width_pcie_x1 ? "Width x1" :
7464 err = ixgbe_read_pba_string_generic(hw, part_str, IXGBE_PBANUM_LENGTH);
7466 strncpy(part_str, "Unknown", IXGBE_PBANUM_LENGTH);
7467 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
7468 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
7469 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
7472 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
7473 hw->mac.type, hw->phy.type, part_str);
7475 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
7476 e_dev_warn("PCI-Express bandwidth available for this card is "
7477 "not sufficient for optimal performance.\n");
7478 e_dev_warn("For optimal performance a x8 PCI-Express slot "
7482 /* save off EEPROM version number */
7483 hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
7485 /* reset the hardware with the new settings */
7486 err = hw->mac.ops.start_hw(hw);
7488 if (err == IXGBE_ERR_EEPROM_VERSION) {
7489 /* We are running on a pre-production device, log a warning */
7490 e_dev_warn("This device is a pre-production adapter/LOM. "
7491 "Please be aware there may be issues associated "
7492 "with your hardware. If you are experiencing "
7493 "problems please contact your Intel or hardware "
7494 "representative who provided you with this "
7497 strcpy(netdev->name, "eth%d");
7498 err = register_netdev(netdev);
7502 /* carrier off reporting is important to ethtool even BEFORE open */
7503 netif_carrier_off(netdev);
7505 #ifdef CONFIG_IXGBE_DCA
7506 if (dca_add_requester(&pdev->dev) == 0) {
7507 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
7508 ixgbe_setup_dca(adapter);
7511 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7512 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
7513 for (i = 0; i < adapter->num_vfs; i++)
7514 ixgbe_vf_configuration(pdev, (i | 0x10000000));
7517 /* Inform firmware of driver version */
7518 if (hw->mac.ops.set_fw_drv_ver)
7519 hw->mac.ops.set_fw_drv_ver(hw, MAJ, MIN, BUILD,
7522 /* add san mac addr to netdev */
7523 ixgbe_add_sanmac_netdev(netdev);
7525 e_dev_info("Intel(R) 10 Gigabit Network Connection\n");
7530 ixgbe_release_hw_control(adapter);
7531 ixgbe_clear_interrupt_scheme(adapter);
7534 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7535 ixgbe_disable_sriov(adapter);
7536 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
7537 iounmap(hw->hw_addr);
7539 free_netdev(netdev);
7541 pci_release_selected_regions(pdev,
7542 pci_select_bars(pdev, IORESOURCE_MEM));
7545 pci_disable_device(pdev);
7550 * ixgbe_remove - Device Removal Routine
7551 * @pdev: PCI device information struct
7553 * ixgbe_remove is called by the PCI subsystem to alert the driver
7554 * that it should release a PCI device. The could be caused by a
7555 * Hot-Plug event, or because the driver is going to be removed from
7558 static void __devexit ixgbe_remove(struct pci_dev *pdev)
7560 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7561 struct net_device *netdev = adapter->netdev;
7563 set_bit(__IXGBE_DOWN, &adapter->state);
7564 cancel_work_sync(&adapter->service_task);
7566 #ifdef CONFIG_IXGBE_DCA
7567 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
7568 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
7569 dca_remove_requester(&pdev->dev);
7570 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
7575 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
7576 ixgbe_cleanup_fcoe(adapter);
7578 #endif /* IXGBE_FCOE */
7580 /* remove the added san mac */
7581 ixgbe_del_sanmac_netdev(netdev);
7583 if (netdev->reg_state == NETREG_REGISTERED)
7584 unregister_netdev(netdev);
7586 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7587 ixgbe_disable_sriov(adapter);
7589 ixgbe_clear_interrupt_scheme(adapter);
7591 ixgbe_release_hw_control(adapter);
7593 iounmap(adapter->hw.hw_addr);
7594 pci_release_selected_regions(pdev, pci_select_bars(pdev,
7597 e_dev_info("complete\n");
7599 free_netdev(netdev);
7601 pci_disable_pcie_error_reporting(pdev);
7603 pci_disable_device(pdev);
7607 * ixgbe_io_error_detected - called when PCI error is detected
7608 * @pdev: Pointer to PCI device
7609 * @state: The current pci connection state
7611 * This function is called after a PCI bus error affecting
7612 * this device has been detected.
7614 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
7615 pci_channel_state_t state)
7617 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7618 struct net_device *netdev = adapter->netdev;
7620 netif_device_detach(netdev);
7622 if (state == pci_channel_io_perm_failure)
7623 return PCI_ERS_RESULT_DISCONNECT;
7625 if (netif_running(netdev))
7626 ixgbe_down(adapter);
7627 pci_disable_device(pdev);
7629 /* Request a slot reset. */
7630 return PCI_ERS_RESULT_NEED_RESET;
7634 * ixgbe_io_slot_reset - called after the pci bus has been reset.
7635 * @pdev: Pointer to PCI device
7637 * Restart the card from scratch, as if from a cold-boot.
7639 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7641 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7642 pci_ers_result_t result;
7645 if (pci_enable_device_mem(pdev)) {
7646 e_err(probe, "Cannot re-enable PCI device after reset.\n");
7647 result = PCI_ERS_RESULT_DISCONNECT;
7649 pci_set_master(pdev);
7650 pci_restore_state(pdev);
7651 pci_save_state(pdev);
7653 pci_wake_from_d3(pdev, false);
7655 ixgbe_reset(adapter);
7656 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7657 result = PCI_ERS_RESULT_RECOVERED;
7660 err = pci_cleanup_aer_uncorrect_error_status(pdev);
7662 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7663 "failed 0x%0x\n", err);
7664 /* non-fatal, continue */
7671 * ixgbe_io_resume - called when traffic can start flowing again.
7672 * @pdev: Pointer to PCI device
7674 * This callback is called when the error recovery driver tells us that
7675 * its OK to resume normal operation.
7677 static void ixgbe_io_resume(struct pci_dev *pdev)
7679 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7680 struct net_device *netdev = adapter->netdev;
7682 if (netif_running(netdev)) {
7683 if (ixgbe_up(adapter)) {
7684 e_info(probe, "ixgbe_up failed after reset\n");
7689 netif_device_attach(netdev);
7692 static struct pci_error_handlers ixgbe_err_handler = {
7693 .error_detected = ixgbe_io_error_detected,
7694 .slot_reset = ixgbe_io_slot_reset,
7695 .resume = ixgbe_io_resume,
7698 static struct pci_driver ixgbe_driver = {
7699 .name = ixgbe_driver_name,
7700 .id_table = ixgbe_pci_tbl,
7701 .probe = ixgbe_probe,
7702 .remove = __devexit_p(ixgbe_remove),
7704 .suspend = ixgbe_suspend,
7705 .resume = ixgbe_resume,
7707 .shutdown = ixgbe_shutdown,
7708 .err_handler = &ixgbe_err_handler
7712 * ixgbe_init_module - Driver Registration Routine
7714 * ixgbe_init_module is the first routine called when the driver is
7715 * loaded. All it does is register with the PCI subsystem.
7717 static int __init ixgbe_init_module(void)
7720 pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
7721 pr_info("%s\n", ixgbe_copyright);
7723 #ifdef CONFIG_IXGBE_DCA
7724 dca_register_notify(&dca_notifier);
7727 ret = pci_register_driver(&ixgbe_driver);
7731 module_init(ixgbe_init_module);
7734 * ixgbe_exit_module - Driver Exit Cleanup Routine
7736 * ixgbe_exit_module is called just before the driver is removed
7739 static void __exit ixgbe_exit_module(void)
7741 #ifdef CONFIG_IXGBE_DCA
7742 dca_unregister_notify(&dca_notifier);
7744 pci_unregister_driver(&ixgbe_driver);
7745 rcu_barrier(); /* Wait for completion of call_rcu()'s */
7748 #ifdef CONFIG_IXGBE_DCA
7749 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
7754 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
7755 __ixgbe_notify_dca);
7757 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7760 #endif /* CONFIG_IXGBE_DCA */
7762 module_exit(ixgbe_exit_module);