1 /*******************************************************************************
3 Intel 10 Gigabit PCI Express Linux driver
4 Copyright(c) 1999 - 2012 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";
59 char ixgbe_default_device_descr[] =
60 "Intel(R) 10 Gigabit Network Connection";
62 static char ixgbe_default_device_descr[] =
63 "Intel(R) 10 Gigabit Network Connection";
68 #define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) "." \
69 __stringify(BUILD) "-k"
70 const char ixgbe_driver_version[] = DRV_VERSION;
71 static const char ixgbe_copyright[] =
72 "Copyright (c) 1999-2012 Intel Corporation.";
74 static const struct ixgbe_info *ixgbe_info_tbl[] = {
75 [board_82598] = &ixgbe_82598_info,
76 [board_82599] = &ixgbe_82599_info,
77 [board_X540] = &ixgbe_X540_info,
80 /* ixgbe_pci_tbl - PCI Device ID Table
82 * Wildcard entries (PCI_ANY_ID) should come last
83 * Last entry must be all 0s
85 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
86 * Class, Class Mask, private data (not used) }
88 static DEFINE_PCI_DEVICE_TABLE(ixgbe_pci_tbl) = {
89 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598), board_82598 },
90 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT), board_82598 },
91 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT), board_82598 },
92 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT), board_82598 },
93 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2), board_82598 },
94 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4), board_82598 },
95 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT), board_82598 },
96 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT), board_82598 },
97 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM), board_82598 },
98 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR), board_82598 },
99 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM), board_82598 },
100 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX), board_82598 },
101 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4), board_82599 },
102 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM), board_82599 },
103 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR), board_82599 },
104 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP), board_82599 },
105 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM), board_82599 },
106 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ), board_82599 },
107 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4), board_82599 },
108 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE), board_82599 },
109 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE), board_82599 },
110 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM), board_82599 },
111 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE), board_82599 },
112 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T), board_X540 },
113 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF2), board_82599 },
114 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_LS), board_82599 },
115 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599EN_SFP), board_82599 },
116 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF_QP), board_82599 },
117 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T1), board_X540 },
118 /* required last entry */
121 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
123 #ifdef CONFIG_IXGBE_DCA
124 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
126 static struct notifier_block dca_notifier = {
127 .notifier_call = ixgbe_notify_dca,
133 #ifdef CONFIG_PCI_IOV
134 static unsigned int max_vfs;
135 module_param(max_vfs, uint, 0);
136 MODULE_PARM_DESC(max_vfs,
137 "Maximum number of virtual functions to allocate per physical function - default is zero and maximum value is 63");
138 #endif /* CONFIG_PCI_IOV */
140 static unsigned int allow_unsupported_sfp;
141 module_param(allow_unsupported_sfp, uint, 0);
142 MODULE_PARM_DESC(allow_unsupported_sfp,
143 "Allow unsupported and untested SFP+ modules on 82599-based adapters");
145 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
146 static int debug = -1;
147 module_param(debug, int, 0);
148 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
150 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
151 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
152 MODULE_LICENSE("GPL");
153 MODULE_VERSION(DRV_VERSION);
155 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
157 if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
158 !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
159 schedule_work(&adapter->service_task);
162 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
164 BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
166 /* flush memory to make sure state is correct before next watchdog */
167 smp_mb__before_clear_bit();
168 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
171 struct ixgbe_reg_info {
176 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
178 /* General Registers */
179 {IXGBE_CTRL, "CTRL"},
180 {IXGBE_STATUS, "STATUS"},
181 {IXGBE_CTRL_EXT, "CTRL_EXT"},
183 /* Interrupt Registers */
184 {IXGBE_EICR, "EICR"},
187 {IXGBE_SRRCTL(0), "SRRCTL"},
188 {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
189 {IXGBE_RDLEN(0), "RDLEN"},
190 {IXGBE_RDH(0), "RDH"},
191 {IXGBE_RDT(0), "RDT"},
192 {IXGBE_RXDCTL(0), "RXDCTL"},
193 {IXGBE_RDBAL(0), "RDBAL"},
194 {IXGBE_RDBAH(0), "RDBAH"},
197 {IXGBE_TDBAL(0), "TDBAL"},
198 {IXGBE_TDBAH(0), "TDBAH"},
199 {IXGBE_TDLEN(0), "TDLEN"},
200 {IXGBE_TDH(0), "TDH"},
201 {IXGBE_TDT(0), "TDT"},
202 {IXGBE_TXDCTL(0), "TXDCTL"},
204 /* List Terminator */
210 * ixgbe_regdump - register printout routine
212 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
218 switch (reginfo->ofs) {
219 case IXGBE_SRRCTL(0):
220 for (i = 0; i < 64; i++)
221 regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
223 case IXGBE_DCA_RXCTRL(0):
224 for (i = 0; i < 64; i++)
225 regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
228 for (i = 0; i < 64; i++)
229 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
232 for (i = 0; i < 64; i++)
233 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
236 for (i = 0; i < 64; i++)
237 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
239 case IXGBE_RXDCTL(0):
240 for (i = 0; i < 64; i++)
241 regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
244 for (i = 0; i < 64; i++)
245 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
248 for (i = 0; i < 64; i++)
249 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
252 for (i = 0; i < 64; i++)
253 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
256 for (i = 0; i < 64; i++)
257 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
260 for (i = 0; i < 64; i++)
261 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
264 for (i = 0; i < 64; i++)
265 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
268 for (i = 0; i < 64; i++)
269 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
271 case IXGBE_TXDCTL(0):
272 for (i = 0; i < 64; i++)
273 regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
276 pr_info("%-15s %08x\n", reginfo->name,
277 IXGBE_READ_REG(hw, reginfo->ofs));
281 for (i = 0; i < 8; i++) {
282 snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i*8, i*8+7);
283 pr_err("%-15s", rname);
284 for (j = 0; j < 8; j++)
285 pr_cont(" %08x", regs[i*8+j]);
292 * ixgbe_dump - Print registers, tx-rings and rx-rings
294 static void ixgbe_dump(struct ixgbe_adapter *adapter)
296 struct net_device *netdev = adapter->netdev;
297 struct ixgbe_hw *hw = &adapter->hw;
298 struct ixgbe_reg_info *reginfo;
300 struct ixgbe_ring *tx_ring;
301 struct ixgbe_tx_buffer *tx_buffer;
302 union ixgbe_adv_tx_desc *tx_desc;
303 struct my_u0 { u64 a; u64 b; } *u0;
304 struct ixgbe_ring *rx_ring;
305 union ixgbe_adv_rx_desc *rx_desc;
306 struct ixgbe_rx_buffer *rx_buffer_info;
310 if (!netif_msg_hw(adapter))
313 /* Print netdevice Info */
315 dev_info(&adapter->pdev->dev, "Net device Info\n");
316 pr_info("Device Name state "
317 "trans_start last_rx\n");
318 pr_info("%-15s %016lX %016lX %016lX\n",
325 /* Print Registers */
326 dev_info(&adapter->pdev->dev, "Register Dump\n");
327 pr_info(" Register Name Value\n");
328 for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
329 reginfo->name; reginfo++) {
330 ixgbe_regdump(hw, reginfo);
333 /* Print TX Ring Summary */
334 if (!netdev || !netif_running(netdev))
337 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
338 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
339 for (n = 0; n < adapter->num_tx_queues; n++) {
340 tx_ring = adapter->tx_ring[n];
341 tx_buffer = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
342 pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
343 n, tx_ring->next_to_use, tx_ring->next_to_clean,
344 (u64)dma_unmap_addr(tx_buffer, dma),
345 dma_unmap_len(tx_buffer, len),
346 tx_buffer->next_to_watch,
347 (u64)tx_buffer->time_stamp);
351 if (!netif_msg_tx_done(adapter))
352 goto rx_ring_summary;
354 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
356 /* Transmit Descriptor Formats
358 * Advanced Transmit Descriptor
359 * +--------------------------------------------------------------+
360 * 0 | Buffer Address [63:0] |
361 * +--------------------------------------------------------------+
362 * 8 | PAYLEN | PORTS | IDX | STA | DCMD |DTYP | RSV | DTALEN |
363 * +--------------------------------------------------------------+
364 * 63 46 45 40 39 36 35 32 31 24 23 20 19 0
367 for (n = 0; n < adapter->num_tx_queues; n++) {
368 tx_ring = adapter->tx_ring[n];
369 pr_info("------------------------------------\n");
370 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
371 pr_info("------------------------------------\n");
372 pr_info("T [desc] [address 63:0 ] "
373 "[PlPOIdStDDt Ln] [bi->dma ] "
374 "leng ntw timestamp bi->skb\n");
376 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
377 tx_desc = IXGBE_TX_DESC(tx_ring, i);
378 tx_buffer = &tx_ring->tx_buffer_info[i];
379 u0 = (struct my_u0 *)tx_desc;
380 pr_info("T [0x%03X] %016llX %016llX %016llX"
381 " %04X %p %016llX %p", i,
384 (u64)dma_unmap_addr(tx_buffer, dma),
385 dma_unmap_len(tx_buffer, len),
386 tx_buffer->next_to_watch,
387 (u64)tx_buffer->time_stamp,
389 if (i == tx_ring->next_to_use &&
390 i == tx_ring->next_to_clean)
392 else if (i == tx_ring->next_to_use)
394 else if (i == tx_ring->next_to_clean)
399 if (netif_msg_pktdata(adapter) &&
401 print_hex_dump(KERN_INFO, "",
402 DUMP_PREFIX_ADDRESS, 16, 1,
403 tx_buffer->skb->data,
404 dma_unmap_len(tx_buffer, len),
409 /* Print RX Rings Summary */
411 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
412 pr_info("Queue [NTU] [NTC]\n");
413 for (n = 0; n < adapter->num_rx_queues; n++) {
414 rx_ring = adapter->rx_ring[n];
415 pr_info("%5d %5X %5X\n",
416 n, rx_ring->next_to_use, rx_ring->next_to_clean);
420 if (!netif_msg_rx_status(adapter))
423 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
425 /* Advanced Receive Descriptor (Read) Format
427 * +-----------------------------------------------------+
428 * 0 | Packet Buffer Address [63:1] |A0/NSE|
429 * +----------------------------------------------+------+
430 * 8 | Header Buffer Address [63:1] | DD |
431 * +-----------------------------------------------------+
434 * Advanced Receive Descriptor (Write-Back) Format
436 * 63 48 47 32 31 30 21 20 16 15 4 3 0
437 * +------------------------------------------------------+
438 * 0 | Packet IP |SPH| HDR_LEN | RSV|Packet| RSS |
439 * | Checksum Ident | | | | Type | Type |
440 * +------------------------------------------------------+
441 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
442 * +------------------------------------------------------+
443 * 63 48 47 32 31 20 19 0
445 for (n = 0; n < adapter->num_rx_queues; n++) {
446 rx_ring = adapter->rx_ring[n];
447 pr_info("------------------------------------\n");
448 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
449 pr_info("------------------------------------\n");
450 pr_info("R [desc] [ PktBuf A0] "
451 "[ HeadBuf DD] [bi->dma ] [bi->skb] "
452 "<-- Adv Rx Read format\n");
453 pr_info("RWB[desc] [PcsmIpSHl PtRs] "
454 "[vl er S cks ln] ---------------- [bi->skb] "
455 "<-- Adv Rx Write-Back format\n");
457 for (i = 0; i < rx_ring->count; i++) {
458 rx_buffer_info = &rx_ring->rx_buffer_info[i];
459 rx_desc = IXGBE_RX_DESC(rx_ring, i);
460 u0 = (struct my_u0 *)rx_desc;
461 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
462 if (staterr & IXGBE_RXD_STAT_DD) {
463 /* Descriptor Done */
464 pr_info("RWB[0x%03X] %016llX "
465 "%016llX ---------------- %p", i,
468 rx_buffer_info->skb);
470 pr_info("R [0x%03X] %016llX "
471 "%016llX %016llX %p", i,
474 (u64)rx_buffer_info->dma,
475 rx_buffer_info->skb);
477 if (netif_msg_pktdata(adapter) &&
478 rx_buffer_info->dma) {
479 print_hex_dump(KERN_INFO, "",
480 DUMP_PREFIX_ADDRESS, 16, 1,
481 page_address(rx_buffer_info->page) +
482 rx_buffer_info->page_offset,
483 ixgbe_rx_bufsz(rx_ring), true);
487 if (i == rx_ring->next_to_use)
489 else if (i == rx_ring->next_to_clean)
501 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
505 /* Let firmware take over control of h/w */
506 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
507 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
508 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
511 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
515 /* Let firmware know the driver has taken over */
516 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
517 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
518 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
522 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
523 * @adapter: pointer to adapter struct
524 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
525 * @queue: queue to map the corresponding interrupt to
526 * @msix_vector: the vector to map to the corresponding queue
529 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
530 u8 queue, u8 msix_vector)
533 struct ixgbe_hw *hw = &adapter->hw;
534 switch (hw->mac.type) {
535 case ixgbe_mac_82598EB:
536 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
539 index = (((direction * 64) + queue) >> 2) & 0x1F;
540 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
541 ivar &= ~(0xFF << (8 * (queue & 0x3)));
542 ivar |= (msix_vector << (8 * (queue & 0x3)));
543 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
545 case ixgbe_mac_82599EB:
547 if (direction == -1) {
549 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
550 index = ((queue & 1) * 8);
551 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
552 ivar &= ~(0xFF << index);
553 ivar |= (msix_vector << index);
554 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
557 /* tx or rx causes */
558 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
559 index = ((16 * (queue & 1)) + (8 * direction));
560 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
561 ivar &= ~(0xFF << index);
562 ivar |= (msix_vector << index);
563 IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
571 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
576 switch (adapter->hw.mac.type) {
577 case ixgbe_mac_82598EB:
578 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
579 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
581 case ixgbe_mac_82599EB:
583 mask = (qmask & 0xFFFFFFFF);
584 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
585 mask = (qmask >> 32);
586 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
593 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_ring *ring,
594 struct ixgbe_tx_buffer *tx_buffer)
596 if (tx_buffer->skb) {
597 dev_kfree_skb_any(tx_buffer->skb);
598 if (dma_unmap_len(tx_buffer, len))
599 dma_unmap_single(ring->dev,
600 dma_unmap_addr(tx_buffer, dma),
601 dma_unmap_len(tx_buffer, len),
603 } else if (dma_unmap_len(tx_buffer, len)) {
604 dma_unmap_page(ring->dev,
605 dma_unmap_addr(tx_buffer, dma),
606 dma_unmap_len(tx_buffer, len),
609 tx_buffer->next_to_watch = NULL;
610 tx_buffer->skb = NULL;
611 dma_unmap_len_set(tx_buffer, len, 0);
612 /* tx_buffer must be completely set up in the transmit path */
615 static void ixgbe_update_xoff_rx_lfc(struct ixgbe_adapter *adapter)
617 struct ixgbe_hw *hw = &adapter->hw;
618 struct ixgbe_hw_stats *hwstats = &adapter->stats;
622 if ((hw->fc.current_mode != ixgbe_fc_full) &&
623 (hw->fc.current_mode != ixgbe_fc_rx_pause))
626 switch (hw->mac.type) {
627 case ixgbe_mac_82598EB:
628 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
631 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
633 hwstats->lxoffrxc += data;
635 /* refill credits (no tx hang) if we received xoff */
639 for (i = 0; i < adapter->num_tx_queues; i++)
640 clear_bit(__IXGBE_HANG_CHECK_ARMED,
641 &adapter->tx_ring[i]->state);
644 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
646 struct ixgbe_hw *hw = &adapter->hw;
647 struct ixgbe_hw_stats *hwstats = &adapter->stats;
650 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
652 if (adapter->ixgbe_ieee_pfc)
653 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
655 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED) || !pfc_en) {
656 ixgbe_update_xoff_rx_lfc(adapter);
660 /* update stats for each tc, only valid with PFC enabled */
661 for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
662 switch (hw->mac.type) {
663 case ixgbe_mac_82598EB:
664 xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
667 xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
669 hwstats->pxoffrxc[i] += xoff[i];
672 /* disarm tx queues that have received xoff frames */
673 for (i = 0; i < adapter->num_tx_queues; i++) {
674 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
675 u8 tc = tx_ring->dcb_tc;
678 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
682 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
684 return ring->stats.packets;
687 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
689 struct ixgbe_adapter *adapter = netdev_priv(ring->netdev);
690 struct ixgbe_hw *hw = &adapter->hw;
692 u32 head = IXGBE_READ_REG(hw, IXGBE_TDH(ring->reg_idx));
693 u32 tail = IXGBE_READ_REG(hw, IXGBE_TDT(ring->reg_idx));
696 return (head < tail) ?
697 tail - head : (tail + ring->count - head);
702 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
704 u32 tx_done = ixgbe_get_tx_completed(tx_ring);
705 u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
706 u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
709 clear_check_for_tx_hang(tx_ring);
712 * Check for a hung queue, but be thorough. This verifies
713 * that a transmit has been completed since the previous
714 * check AND there is at least one packet pending. The
715 * ARMED bit is set to indicate a potential hang. The
716 * bit is cleared if a pause frame is received to remove
717 * false hang detection due to PFC or 802.3x frames. By
718 * requiring this to fail twice we avoid races with
719 * pfc clearing the ARMED bit and conditions where we
720 * run the check_tx_hang logic with a transmit completion
721 * pending but without time to complete it yet.
723 if ((tx_done_old == tx_done) && tx_pending) {
724 /* make sure it is true for two checks in a row */
725 ret = test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
728 /* update completed stats and continue */
729 tx_ring->tx_stats.tx_done_old = tx_done;
730 /* reset the countdown */
731 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
738 * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
739 * @adapter: driver private struct
741 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
744 /* Do the reset outside of interrupt context */
745 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
746 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
747 ixgbe_service_event_schedule(adapter);
752 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
753 * @q_vector: structure containing interrupt and ring information
754 * @tx_ring: tx ring to clean
756 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
757 struct ixgbe_ring *tx_ring)
759 struct ixgbe_adapter *adapter = q_vector->adapter;
760 struct ixgbe_tx_buffer *tx_buffer;
761 union ixgbe_adv_tx_desc *tx_desc;
762 unsigned int total_bytes = 0, total_packets = 0;
763 unsigned int budget = q_vector->tx.work_limit;
764 unsigned int i = tx_ring->next_to_clean;
766 if (test_bit(__IXGBE_DOWN, &adapter->state))
769 tx_buffer = &tx_ring->tx_buffer_info[i];
770 tx_desc = IXGBE_TX_DESC(tx_ring, i);
774 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
776 /* if next_to_watch is not set then there is no work pending */
780 /* prevent any other reads prior to eop_desc */
783 /* if DD is not set pending work has not been completed */
784 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
787 /* clear next_to_watch to prevent false hangs */
788 tx_buffer->next_to_watch = NULL;
790 /* update the statistics for this packet */
791 total_bytes += tx_buffer->bytecount;
792 total_packets += tx_buffer->gso_segs;
794 #ifdef CONFIG_IXGBE_PTP
795 if (unlikely(tx_buffer->tx_flags & IXGBE_TX_FLAGS_TSTAMP))
796 ixgbe_ptp_tx_hwtstamp(q_vector, tx_buffer->skb);
800 dev_kfree_skb_any(tx_buffer->skb);
802 /* unmap skb header data */
803 dma_unmap_single(tx_ring->dev,
804 dma_unmap_addr(tx_buffer, dma),
805 dma_unmap_len(tx_buffer, len),
808 /* clear tx_buffer data */
809 tx_buffer->skb = NULL;
810 dma_unmap_len_set(tx_buffer, len, 0);
812 /* unmap remaining buffers */
813 while (tx_desc != eop_desc) {
819 tx_buffer = tx_ring->tx_buffer_info;
820 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
823 /* unmap any remaining paged data */
824 if (dma_unmap_len(tx_buffer, len)) {
825 dma_unmap_page(tx_ring->dev,
826 dma_unmap_addr(tx_buffer, dma),
827 dma_unmap_len(tx_buffer, len),
829 dma_unmap_len_set(tx_buffer, len, 0);
833 /* move us one more past the eop_desc for start of next pkt */
839 tx_buffer = tx_ring->tx_buffer_info;
840 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
843 /* issue prefetch for next Tx descriptor */
846 /* update budget accounting */
848 } while (likely(budget));
851 tx_ring->next_to_clean = i;
852 u64_stats_update_begin(&tx_ring->syncp);
853 tx_ring->stats.bytes += total_bytes;
854 tx_ring->stats.packets += total_packets;
855 u64_stats_update_end(&tx_ring->syncp);
856 q_vector->tx.total_bytes += total_bytes;
857 q_vector->tx.total_packets += total_packets;
859 if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
860 /* schedule immediate reset if we believe we hung */
861 struct ixgbe_hw *hw = &adapter->hw;
862 e_err(drv, "Detected Tx Unit Hang\n"
864 " TDH, TDT <%x>, <%x>\n"
865 " next_to_use <%x>\n"
866 " next_to_clean <%x>\n"
867 "tx_buffer_info[next_to_clean]\n"
868 " time_stamp <%lx>\n"
870 tx_ring->queue_index,
871 IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
872 IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
873 tx_ring->next_to_use, i,
874 tx_ring->tx_buffer_info[i].time_stamp, jiffies);
876 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
879 "tx hang %d detected on queue %d, resetting adapter\n",
880 adapter->tx_timeout_count + 1, tx_ring->queue_index);
882 /* schedule immediate reset if we believe we hung */
883 ixgbe_tx_timeout_reset(adapter);
885 /* the adapter is about to reset, no point in enabling stuff */
889 netdev_tx_completed_queue(txring_txq(tx_ring),
890 total_packets, total_bytes);
892 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
893 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
894 (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
895 /* Make sure that anybody stopping the queue after this
896 * sees the new next_to_clean.
899 if (__netif_subqueue_stopped(tx_ring->netdev,
900 tx_ring->queue_index)
901 && !test_bit(__IXGBE_DOWN, &adapter->state)) {
902 netif_wake_subqueue(tx_ring->netdev,
903 tx_ring->queue_index);
904 ++tx_ring->tx_stats.restart_queue;
911 #ifdef CONFIG_IXGBE_DCA
912 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
913 struct ixgbe_ring *tx_ring,
916 struct ixgbe_hw *hw = &adapter->hw;
917 u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
920 switch (hw->mac.type) {
921 case ixgbe_mac_82598EB:
922 reg_offset = IXGBE_DCA_TXCTRL(tx_ring->reg_idx);
924 case ixgbe_mac_82599EB:
926 reg_offset = IXGBE_DCA_TXCTRL_82599(tx_ring->reg_idx);
927 txctrl <<= IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599;
930 /* for unknown hardware do not write register */
935 * We can enable relaxed ordering for reads, but not writes when
936 * DCA is enabled. This is due to a known issue in some chipsets
937 * which will cause the DCA tag to be cleared.
939 txctrl |= IXGBE_DCA_TXCTRL_DESC_RRO_EN |
940 IXGBE_DCA_TXCTRL_DATA_RRO_EN |
941 IXGBE_DCA_TXCTRL_DESC_DCA_EN;
943 IXGBE_WRITE_REG(hw, reg_offset, txctrl);
946 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
947 struct ixgbe_ring *rx_ring,
950 struct ixgbe_hw *hw = &adapter->hw;
951 u32 rxctrl = dca3_get_tag(rx_ring->dev, cpu);
952 u8 reg_idx = rx_ring->reg_idx;
955 switch (hw->mac.type) {
956 case ixgbe_mac_82599EB:
958 rxctrl <<= IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599;
965 * We can enable relaxed ordering for reads, but not writes when
966 * DCA is enabled. This is due to a known issue in some chipsets
967 * which will cause the DCA tag to be cleared.
969 rxctrl |= IXGBE_DCA_RXCTRL_DESC_RRO_EN |
970 IXGBE_DCA_RXCTRL_DATA_DCA_EN |
971 IXGBE_DCA_RXCTRL_DESC_DCA_EN;
973 IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
976 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
978 struct ixgbe_adapter *adapter = q_vector->adapter;
979 struct ixgbe_ring *ring;
982 if (q_vector->cpu == cpu)
985 ixgbe_for_each_ring(ring, q_vector->tx)
986 ixgbe_update_tx_dca(adapter, ring, cpu);
988 ixgbe_for_each_ring(ring, q_vector->rx)
989 ixgbe_update_rx_dca(adapter, ring, cpu);
996 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
1000 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
1003 /* always use CB2 mode, difference is masked in the CB driver */
1004 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
1006 for (i = 0; i < adapter->num_q_vectors; i++) {
1007 adapter->q_vector[i]->cpu = -1;
1008 ixgbe_update_dca(adapter->q_vector[i]);
1012 static int __ixgbe_notify_dca(struct device *dev, void *data)
1014 struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1015 unsigned long event = *(unsigned long *)data;
1017 if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
1021 case DCA_PROVIDER_ADD:
1022 /* if we're already enabled, don't do it again */
1023 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1025 if (dca_add_requester(dev) == 0) {
1026 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1027 ixgbe_setup_dca(adapter);
1030 /* Fall Through since DCA is disabled. */
1031 case DCA_PROVIDER_REMOVE:
1032 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1033 dca_remove_requester(dev);
1034 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1035 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
1043 #endif /* CONFIG_IXGBE_DCA */
1044 static inline void ixgbe_rx_hash(struct ixgbe_ring *ring,
1045 union ixgbe_adv_rx_desc *rx_desc,
1046 struct sk_buff *skb)
1048 if (ring->netdev->features & NETIF_F_RXHASH)
1049 skb->rxhash = le32_to_cpu(rx_desc->wb.lower.hi_dword.rss);
1054 * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1055 * @ring: structure containing ring specific data
1056 * @rx_desc: advanced rx descriptor
1058 * Returns : true if it is FCoE pkt
1060 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_ring *ring,
1061 union ixgbe_adv_rx_desc *rx_desc)
1063 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1065 return test_bit(__IXGBE_RX_FCOE, &ring->state) &&
1066 ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1067 (cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1068 IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1071 #endif /* IXGBE_FCOE */
1073 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1074 * @ring: structure containing ring specific data
1075 * @rx_desc: current Rx descriptor being processed
1076 * @skb: skb currently being received and modified
1078 static inline void ixgbe_rx_checksum(struct ixgbe_ring *ring,
1079 union ixgbe_adv_rx_desc *rx_desc,
1080 struct sk_buff *skb)
1082 skb_checksum_none_assert(skb);
1084 /* Rx csum disabled */
1085 if (!(ring->netdev->features & NETIF_F_RXCSUM))
1088 /* if IP and error */
1089 if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
1090 ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
1091 ring->rx_stats.csum_err++;
1095 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
1098 if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
1099 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1102 * 82599 errata, UDP frames with a 0 checksum can be marked as
1105 if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_UDP)) &&
1106 test_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR, &ring->state))
1109 ring->rx_stats.csum_err++;
1113 /* It must be a TCP or UDP packet with a valid checksum */
1114 skb->ip_summed = CHECKSUM_UNNECESSARY;
1117 static inline void ixgbe_release_rx_desc(struct ixgbe_ring *rx_ring, u32 val)
1119 rx_ring->next_to_use = val;
1121 /* update next to alloc since we have filled the ring */
1122 rx_ring->next_to_alloc = val;
1124 * Force memory writes to complete before letting h/w
1125 * know there are new descriptors to fetch. (Only
1126 * applicable for weak-ordered memory model archs,
1130 writel(val, rx_ring->tail);
1133 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring *rx_ring,
1134 struct ixgbe_rx_buffer *bi)
1136 struct page *page = bi->page;
1137 dma_addr_t dma = bi->dma;
1139 /* since we are recycling buffers we should seldom need to alloc */
1143 /* alloc new page for storage */
1144 if (likely(!page)) {
1145 page = __skb_alloc_pages(GFP_ATOMIC | __GFP_COLD | __GFP_COMP,
1146 bi->skb, ixgbe_rx_pg_order(rx_ring));
1147 if (unlikely(!page)) {
1148 rx_ring->rx_stats.alloc_rx_page_failed++;
1154 /* map page for use */
1155 dma = dma_map_page(rx_ring->dev, page, 0,
1156 ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1159 * if mapping failed free memory back to system since
1160 * there isn't much point in holding memory we can't use
1162 if (dma_mapping_error(rx_ring->dev, dma)) {
1163 __free_pages(page, ixgbe_rx_pg_order(rx_ring));
1166 rx_ring->rx_stats.alloc_rx_page_failed++;
1171 bi->page_offset = 0;
1177 * ixgbe_alloc_rx_buffers - Replace used receive buffers
1178 * @rx_ring: ring to place buffers on
1179 * @cleaned_count: number of buffers to replace
1181 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1183 union ixgbe_adv_rx_desc *rx_desc;
1184 struct ixgbe_rx_buffer *bi;
1185 u16 i = rx_ring->next_to_use;
1191 rx_desc = IXGBE_RX_DESC(rx_ring, i);
1192 bi = &rx_ring->rx_buffer_info[i];
1193 i -= rx_ring->count;
1196 if (!ixgbe_alloc_mapped_page(rx_ring, bi))
1200 * Refresh the desc even if buffer_addrs didn't change
1201 * because each write-back erases this info.
1203 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1209 rx_desc = IXGBE_RX_DESC(rx_ring, 0);
1210 bi = rx_ring->rx_buffer_info;
1211 i -= rx_ring->count;
1214 /* clear the hdr_addr for the next_to_use descriptor */
1215 rx_desc->read.hdr_addr = 0;
1218 } while (cleaned_count);
1220 i += rx_ring->count;
1222 if (rx_ring->next_to_use != i)
1223 ixgbe_release_rx_desc(rx_ring, i);
1227 * ixgbe_get_headlen - determine size of header for RSC/LRO/GRO/FCOE
1228 * @data: pointer to the start of the headers
1229 * @max_len: total length of section to find headers in
1231 * This function is meant to determine the length of headers that will
1232 * be recognized by hardware for LRO, GRO, and RSC offloads. The main
1233 * motivation of doing this is to only perform one pull for IPv4 TCP
1234 * packets so that we can do basic things like calculating the gso_size
1235 * based on the average data per packet.
1237 static unsigned int ixgbe_get_headlen(unsigned char *data,
1238 unsigned int max_len)
1241 unsigned char *network;
1244 struct vlan_hdr *vlan;
1247 struct ipv6hdr *ipv6;
1250 u8 nexthdr = 0; /* default to not TCP */
1253 /* this should never happen, but better safe than sorry */
1254 if (max_len < ETH_HLEN)
1257 /* initialize network frame pointer */
1260 /* set first protocol and move network header forward */
1261 protocol = hdr.eth->h_proto;
1262 hdr.network += ETH_HLEN;
1264 /* handle any vlan tag if present */
1265 if (protocol == __constant_htons(ETH_P_8021Q)) {
1266 if ((hdr.network - data) > (max_len - VLAN_HLEN))
1269 protocol = hdr.vlan->h_vlan_encapsulated_proto;
1270 hdr.network += VLAN_HLEN;
1273 /* handle L3 protocols */
1274 if (protocol == __constant_htons(ETH_P_IP)) {
1275 if ((hdr.network - data) > (max_len - sizeof(struct iphdr)))
1278 /* access ihl as a u8 to avoid unaligned access on ia64 */
1279 hlen = (hdr.network[0] & 0x0F) << 2;
1281 /* verify hlen meets minimum size requirements */
1282 if (hlen < sizeof(struct iphdr))
1283 return hdr.network - data;
1285 /* record next protocol */
1286 nexthdr = hdr.ipv4->protocol;
1287 hdr.network += hlen;
1288 } else if (protocol == __constant_htons(ETH_P_IPV6)) {
1289 if ((hdr.network - data) > (max_len - sizeof(struct ipv6hdr)))
1292 /* record next protocol */
1293 nexthdr = hdr.ipv6->nexthdr;
1294 hdr.network += sizeof(struct ipv6hdr);
1296 } else if (protocol == __constant_htons(ETH_P_FCOE)) {
1297 if ((hdr.network - data) > (max_len - FCOE_HEADER_LEN))
1299 hdr.network += FCOE_HEADER_LEN;
1302 return hdr.network - data;
1305 /* finally sort out TCP/UDP */
1306 if (nexthdr == IPPROTO_TCP) {
1307 if ((hdr.network - data) > (max_len - sizeof(struct tcphdr)))
1310 /* access doff as a u8 to avoid unaligned access on ia64 */
1311 hlen = (hdr.network[12] & 0xF0) >> 2;
1313 /* verify hlen meets minimum size requirements */
1314 if (hlen < sizeof(struct tcphdr))
1315 return hdr.network - data;
1317 hdr.network += hlen;
1318 } else if (nexthdr == IPPROTO_UDP) {
1319 if ((hdr.network - data) > (max_len - sizeof(struct udphdr)))
1322 hdr.network += sizeof(struct udphdr);
1326 * If everything has gone correctly hdr.network should be the
1327 * data section of the packet and will be the end of the header.
1328 * If not then it probably represents the end of the last recognized
1331 if ((hdr.network - data) < max_len)
1332 return hdr.network - data;
1337 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring *ring,
1338 struct sk_buff *skb)
1340 u16 hdr_len = skb_headlen(skb);
1342 /* set gso_size to avoid messing up TCP MSS */
1343 skb_shinfo(skb)->gso_size = DIV_ROUND_UP((skb->len - hdr_len),
1344 IXGBE_CB(skb)->append_cnt);
1347 static void ixgbe_update_rsc_stats(struct ixgbe_ring *rx_ring,
1348 struct sk_buff *skb)
1350 /* if append_cnt is 0 then frame is not RSC */
1351 if (!IXGBE_CB(skb)->append_cnt)
1354 rx_ring->rx_stats.rsc_count += IXGBE_CB(skb)->append_cnt;
1355 rx_ring->rx_stats.rsc_flush++;
1357 ixgbe_set_rsc_gso_size(rx_ring, skb);
1359 /* gso_size is computed using append_cnt so always clear it last */
1360 IXGBE_CB(skb)->append_cnt = 0;
1364 * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor
1365 * @rx_ring: rx descriptor ring packet is being transacted on
1366 * @rx_desc: pointer to the EOP Rx descriptor
1367 * @skb: pointer to current skb being populated
1369 * This function checks the ring, descriptor, and packet information in
1370 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1371 * other fields within the skb.
1373 static void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring,
1374 union ixgbe_adv_rx_desc *rx_desc,
1375 struct sk_buff *skb)
1377 struct net_device *dev = rx_ring->netdev;
1379 ixgbe_update_rsc_stats(rx_ring, skb);
1381 ixgbe_rx_hash(rx_ring, rx_desc, skb);
1383 ixgbe_rx_checksum(rx_ring, rx_desc, skb);
1385 #ifdef CONFIG_IXGBE_PTP
1386 ixgbe_ptp_rx_hwtstamp(rx_ring->q_vector, rx_desc, skb);
1389 if ((dev->features & NETIF_F_HW_VLAN_RX) &&
1390 ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
1391 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
1392 __vlan_hwaccel_put_tag(skb, vid);
1395 skb_record_rx_queue(skb, rx_ring->queue_index);
1397 skb->protocol = eth_type_trans(skb, dev);
1400 static void ixgbe_rx_skb(struct ixgbe_q_vector *q_vector,
1401 struct sk_buff *skb)
1403 struct ixgbe_adapter *adapter = q_vector->adapter;
1405 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
1406 napi_gro_receive(&q_vector->napi, skb);
1412 * ixgbe_is_non_eop - process handling of non-EOP buffers
1413 * @rx_ring: Rx ring being processed
1414 * @rx_desc: Rx descriptor for current buffer
1415 * @skb: Current socket buffer containing buffer in progress
1417 * This function updates next to clean. If the buffer is an EOP buffer
1418 * this function exits returning false, otherwise it will place the
1419 * sk_buff in the next buffer to be chained and return true indicating
1420 * that this is in fact a non-EOP buffer.
1422 static bool ixgbe_is_non_eop(struct ixgbe_ring *rx_ring,
1423 union ixgbe_adv_rx_desc *rx_desc,
1424 struct sk_buff *skb)
1426 u32 ntc = rx_ring->next_to_clean + 1;
1428 /* fetch, update, and store next to clean */
1429 ntc = (ntc < rx_ring->count) ? ntc : 0;
1430 rx_ring->next_to_clean = ntc;
1432 prefetch(IXGBE_RX_DESC(rx_ring, ntc));
1434 /* update RSC append count if present */
1435 if (ring_is_rsc_enabled(rx_ring)) {
1436 __le32 rsc_enabled = rx_desc->wb.lower.lo_dword.data &
1437 cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK);
1439 if (unlikely(rsc_enabled)) {
1440 u32 rsc_cnt = le32_to_cpu(rsc_enabled);
1442 rsc_cnt >>= IXGBE_RXDADV_RSCCNT_SHIFT;
1443 IXGBE_CB(skb)->append_cnt += rsc_cnt - 1;
1445 /* update ntc based on RSC value */
1446 ntc = le32_to_cpu(rx_desc->wb.upper.status_error);
1447 ntc &= IXGBE_RXDADV_NEXTP_MASK;
1448 ntc >>= IXGBE_RXDADV_NEXTP_SHIFT;
1452 /* if we are the last buffer then there is nothing else to do */
1453 if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1456 /* place skb in next buffer to be received */
1457 rx_ring->rx_buffer_info[ntc].skb = skb;
1458 rx_ring->rx_stats.non_eop_descs++;
1464 * ixgbe_pull_tail - ixgbe specific version of skb_pull_tail
1465 * @rx_ring: rx descriptor ring packet is being transacted on
1466 * @skb: pointer to current skb being adjusted
1468 * This function is an ixgbe specific version of __pskb_pull_tail. The
1469 * main difference between this version and the original function is that
1470 * this function can make several assumptions about the state of things
1471 * that allow for significant optimizations versus the standard function.
1472 * As a result we can do things like drop a frag and maintain an accurate
1473 * truesize for the skb.
1475 static void ixgbe_pull_tail(struct ixgbe_ring *rx_ring,
1476 struct sk_buff *skb)
1478 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1480 unsigned int pull_len;
1483 * it is valid to use page_address instead of kmap since we are
1484 * working with pages allocated out of the lomem pool per
1485 * alloc_page(GFP_ATOMIC)
1487 va = skb_frag_address(frag);
1490 * we need the header to contain the greater of either ETH_HLEN or
1491 * 60 bytes if the skb->len is less than 60 for skb_pad.
1493 pull_len = ixgbe_get_headlen(va, IXGBE_RX_HDR_SIZE);
1495 /* align pull length to size of long to optimize memcpy performance */
1496 skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
1498 /* update all of the pointers */
1499 skb_frag_size_sub(frag, pull_len);
1500 frag->page_offset += pull_len;
1501 skb->data_len -= pull_len;
1502 skb->tail += pull_len;
1506 * ixgbe_dma_sync_frag - perform DMA sync for first frag of SKB
1507 * @rx_ring: rx descriptor ring packet is being transacted on
1508 * @skb: pointer to current skb being updated
1510 * This function provides a basic DMA sync up for the first fragment of an
1511 * skb. The reason for doing this is that the first fragment cannot be
1512 * unmapped until we have reached the end of packet descriptor for a buffer
1515 static void ixgbe_dma_sync_frag(struct ixgbe_ring *rx_ring,
1516 struct sk_buff *skb)
1518 /* if the page was released unmap it, else just sync our portion */
1519 if (unlikely(IXGBE_CB(skb)->page_released)) {
1520 dma_unmap_page(rx_ring->dev, IXGBE_CB(skb)->dma,
1521 ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1522 IXGBE_CB(skb)->page_released = false;
1524 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1526 dma_sync_single_range_for_cpu(rx_ring->dev,
1529 ixgbe_rx_bufsz(rx_ring),
1532 IXGBE_CB(skb)->dma = 0;
1536 * ixgbe_cleanup_headers - Correct corrupted or empty headers
1537 * @rx_ring: rx descriptor ring packet is being transacted on
1538 * @rx_desc: pointer to the EOP Rx descriptor
1539 * @skb: pointer to current skb being fixed
1541 * Check for corrupted packet headers caused by senders on the local L2
1542 * embedded NIC switch not setting up their Tx Descriptors right. These
1543 * should be very rare.
1545 * Also address the case where we are pulling data in on pages only
1546 * and as such no data is present in the skb header.
1548 * In addition if skb is not at least 60 bytes we need to pad it so that
1549 * it is large enough to qualify as a valid Ethernet frame.
1551 * Returns true if an error was encountered and skb was freed.
1553 static bool ixgbe_cleanup_headers(struct ixgbe_ring *rx_ring,
1554 union ixgbe_adv_rx_desc *rx_desc,
1555 struct sk_buff *skb)
1557 struct net_device *netdev = rx_ring->netdev;
1559 /* verify that the packet does not have any known errors */
1560 if (unlikely(ixgbe_test_staterr(rx_desc,
1561 IXGBE_RXDADV_ERR_FRAME_ERR_MASK) &&
1562 !(netdev->features & NETIF_F_RXALL))) {
1563 dev_kfree_skb_any(skb);
1567 /* place header in linear portion of buffer */
1568 if (skb_is_nonlinear(skb))
1569 ixgbe_pull_tail(rx_ring, skb);
1572 /* do not attempt to pad FCoE Frames as this will disrupt DDP */
1573 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc))
1577 /* if skb_pad returns an error the skb was freed */
1578 if (unlikely(skb->len < 60)) {
1579 int pad_len = 60 - skb->len;
1581 if (skb_pad(skb, pad_len))
1583 __skb_put(skb, pad_len);
1590 * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1591 * @rx_ring: rx descriptor ring to store buffers on
1592 * @old_buff: donor buffer to have page reused
1594 * Synchronizes page for reuse by the adapter
1596 static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring,
1597 struct ixgbe_rx_buffer *old_buff)
1599 struct ixgbe_rx_buffer *new_buff;
1600 u16 nta = rx_ring->next_to_alloc;
1602 new_buff = &rx_ring->rx_buffer_info[nta];
1604 /* update, and store next to alloc */
1606 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1608 /* transfer page from old buffer to new buffer */
1609 new_buff->page = old_buff->page;
1610 new_buff->dma = old_buff->dma;
1611 new_buff->page_offset = old_buff->page_offset;
1613 /* sync the buffer for use by the device */
1614 dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
1615 new_buff->page_offset,
1616 ixgbe_rx_bufsz(rx_ring),
1621 * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1622 * @rx_ring: rx descriptor ring to transact packets on
1623 * @rx_buffer: buffer containing page to add
1624 * @rx_desc: descriptor containing length of buffer written by hardware
1625 * @skb: sk_buff to place the data into
1627 * This function will add the data contained in rx_buffer->page to the skb.
1628 * This is done either through a direct copy if the data in the buffer is
1629 * less than the skb header size, otherwise it will just attach the page as
1630 * a frag to the skb.
1632 * The function will then update the page offset if necessary and return
1633 * true if the buffer can be reused by the adapter.
1635 static bool ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
1636 struct ixgbe_rx_buffer *rx_buffer,
1637 union ixgbe_adv_rx_desc *rx_desc,
1638 struct sk_buff *skb)
1640 struct page *page = rx_buffer->page;
1641 unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
1642 #if (PAGE_SIZE < 8192)
1643 unsigned int truesize = ixgbe_rx_bufsz(rx_ring);
1645 unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
1646 unsigned int last_offset = ixgbe_rx_pg_size(rx_ring) -
1647 ixgbe_rx_bufsz(rx_ring);
1650 if ((size <= IXGBE_RX_HDR_SIZE) && !skb_is_nonlinear(skb)) {
1651 unsigned char *va = page_address(page) + rx_buffer->page_offset;
1653 memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
1655 /* we can reuse buffer as-is, just make sure it is local */
1656 if (likely(page_to_nid(page) == numa_node_id()))
1659 /* this page cannot be reused so discard it */
1664 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
1665 rx_buffer->page_offset, size, truesize);
1667 /* avoid re-using remote pages */
1668 if (unlikely(page_to_nid(page) != numa_node_id()))
1671 #if (PAGE_SIZE < 8192)
1672 /* if we are only owner of page we can reuse it */
1673 if (unlikely(page_count(page) != 1))
1676 /* flip page offset to other buffer */
1677 rx_buffer->page_offset ^= truesize;
1680 * since we are the only owner of the page and we need to
1681 * increment it, just set the value to 2 in order to avoid
1682 * an unecessary locked operation
1684 atomic_set(&page->_count, 2);
1686 /* move offset up to the next cache line */
1687 rx_buffer->page_offset += truesize;
1689 if (rx_buffer->page_offset > last_offset)
1692 /* bump ref count on page before it is given to the stack */
1699 static struct sk_buff *ixgbe_fetch_rx_buffer(struct ixgbe_ring *rx_ring,
1700 union ixgbe_adv_rx_desc *rx_desc)
1702 struct ixgbe_rx_buffer *rx_buffer;
1703 struct sk_buff *skb;
1706 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
1707 page = rx_buffer->page;
1710 skb = rx_buffer->skb;
1713 void *page_addr = page_address(page) +
1714 rx_buffer->page_offset;
1716 /* prefetch first cache line of first page */
1717 prefetch(page_addr);
1718 #if L1_CACHE_BYTES < 128
1719 prefetch(page_addr + L1_CACHE_BYTES);
1722 /* allocate a skb to store the frags */
1723 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1725 if (unlikely(!skb)) {
1726 rx_ring->rx_stats.alloc_rx_buff_failed++;
1731 * we will be copying header into skb->data in
1732 * pskb_may_pull so it is in our interest to prefetch
1733 * it now to avoid a possible cache miss
1735 prefetchw(skb->data);
1738 * Delay unmapping of the first packet. It carries the
1739 * header information, HW may still access the header
1740 * after the writeback. Only unmap it when EOP is
1743 if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1746 IXGBE_CB(skb)->dma = rx_buffer->dma;
1748 if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
1749 ixgbe_dma_sync_frag(rx_ring, skb);
1752 /* we are reusing so sync this buffer for CPU use */
1753 dma_sync_single_range_for_cpu(rx_ring->dev,
1755 rx_buffer->page_offset,
1756 ixgbe_rx_bufsz(rx_ring),
1760 /* pull page into skb */
1761 if (ixgbe_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
1762 /* hand second half of page back to the ring */
1763 ixgbe_reuse_rx_page(rx_ring, rx_buffer);
1764 } else if (IXGBE_CB(skb)->dma == rx_buffer->dma) {
1765 /* the page has been released from the ring */
1766 IXGBE_CB(skb)->page_released = true;
1768 /* we are not reusing the buffer so unmap it */
1769 dma_unmap_page(rx_ring->dev, rx_buffer->dma,
1770 ixgbe_rx_pg_size(rx_ring),
1774 /* clear contents of buffer_info */
1775 rx_buffer->skb = NULL;
1777 rx_buffer->page = NULL;
1783 * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
1784 * @q_vector: structure containing interrupt and ring information
1785 * @rx_ring: rx descriptor ring to transact packets on
1786 * @budget: Total limit on number of packets to process
1788 * This function provides a "bounce buffer" approach to Rx interrupt
1789 * processing. The advantage to this is that on systems that have
1790 * expensive overhead for IOMMU access this provides a means of avoiding
1791 * it by maintaining the mapping of the page to the syste.
1793 * Returns true if all work is completed without reaching budget
1795 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1796 struct ixgbe_ring *rx_ring,
1799 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1801 struct ixgbe_adapter *adapter = q_vector->adapter;
1803 unsigned int mss = 0;
1804 #endif /* IXGBE_FCOE */
1805 u16 cleaned_count = ixgbe_desc_unused(rx_ring);
1808 union ixgbe_adv_rx_desc *rx_desc;
1809 struct sk_buff *skb;
1811 /* return some buffers to hardware, one at a time is too slow */
1812 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
1813 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1817 rx_desc = IXGBE_RX_DESC(rx_ring, rx_ring->next_to_clean);
1819 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_DD))
1823 * This memory barrier is needed to keep us from reading
1824 * any other fields out of the rx_desc until we know the
1825 * RXD_STAT_DD bit is set
1829 /* retrieve a buffer from the ring */
1830 skb = ixgbe_fetch_rx_buffer(rx_ring, rx_desc);
1832 /* exit if we failed to retrieve a buffer */
1838 /* place incomplete frames back on ring for completion */
1839 if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
1842 /* verify the packet layout is correct */
1843 if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
1846 /* probably a little skewed due to removing CRC */
1847 total_rx_bytes += skb->len;
1850 /* populate checksum, timestamp, VLAN, and protocol */
1851 ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
1854 /* if ddp, not passing to ULD unless for FCP_RSP or error */
1855 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc)) {
1856 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
1857 /* include DDPed FCoE data */
1858 if (ddp_bytes > 0) {
1860 mss = rx_ring->netdev->mtu -
1861 sizeof(struct fcoe_hdr) -
1862 sizeof(struct fc_frame_header) -
1863 sizeof(struct fcoe_crc_eof);
1867 total_rx_bytes += ddp_bytes;
1868 total_rx_packets += DIV_ROUND_UP(ddp_bytes,
1872 dev_kfree_skb_any(skb);
1877 #endif /* IXGBE_FCOE */
1878 ixgbe_rx_skb(q_vector, skb);
1880 /* update budget accounting */
1882 } while (likely(budget));
1884 u64_stats_update_begin(&rx_ring->syncp);
1885 rx_ring->stats.packets += total_rx_packets;
1886 rx_ring->stats.bytes += total_rx_bytes;
1887 u64_stats_update_end(&rx_ring->syncp);
1888 q_vector->rx.total_packets += total_rx_packets;
1889 q_vector->rx.total_bytes += total_rx_bytes;
1892 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1898 * ixgbe_configure_msix - Configure MSI-X hardware
1899 * @adapter: board private structure
1901 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1904 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1906 struct ixgbe_q_vector *q_vector;
1910 /* Populate MSIX to EITR Select */
1911 if (adapter->num_vfs > 32) {
1912 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
1913 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
1917 * Populate the IVAR table and set the ITR values to the
1918 * corresponding register.
1920 for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
1921 struct ixgbe_ring *ring;
1922 q_vector = adapter->q_vector[v_idx];
1924 ixgbe_for_each_ring(ring, q_vector->rx)
1925 ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1927 ixgbe_for_each_ring(ring, q_vector->tx)
1928 ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1930 if (q_vector->tx.ring && !q_vector->rx.ring) {
1931 /* tx only vector */
1932 if (adapter->tx_itr_setting == 1)
1933 q_vector->itr = IXGBE_10K_ITR;
1935 q_vector->itr = adapter->tx_itr_setting;
1937 /* rx or rx/tx vector */
1938 if (adapter->rx_itr_setting == 1)
1939 q_vector->itr = IXGBE_20K_ITR;
1941 q_vector->itr = adapter->rx_itr_setting;
1944 ixgbe_write_eitr(q_vector);
1947 switch (adapter->hw.mac.type) {
1948 case ixgbe_mac_82598EB:
1949 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1952 case ixgbe_mac_82599EB:
1953 case ixgbe_mac_X540:
1954 ixgbe_set_ivar(adapter, -1, 1, v_idx);
1959 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1961 /* set up to autoclear timer, and the vectors */
1962 mask = IXGBE_EIMS_ENABLE_MASK;
1963 mask &= ~(IXGBE_EIMS_OTHER |
1964 IXGBE_EIMS_MAILBOX |
1967 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1970 enum latency_range {
1974 latency_invalid = 255
1978 * ixgbe_update_itr - update the dynamic ITR value based on statistics
1979 * @q_vector: structure containing interrupt and ring information
1980 * @ring_container: structure containing ring performance data
1982 * Stores a new ITR value based on packets and byte
1983 * counts during the last interrupt. The advantage of per interrupt
1984 * computation is faster updates and more accurate ITR for the current
1985 * traffic pattern. Constants in this function were computed
1986 * based on theoretical maximum wire speed and thresholds were set based
1987 * on testing data as well as attempting to minimize response time
1988 * while increasing bulk throughput.
1989 * this functionality is controlled by the InterruptThrottleRate module
1990 * parameter (see ixgbe_param.c)
1992 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
1993 struct ixgbe_ring_container *ring_container)
1995 int bytes = ring_container->total_bytes;
1996 int packets = ring_container->total_packets;
1999 u8 itr_setting = ring_container->itr;
2004 /* simple throttlerate management
2005 * 0-10MB/s lowest (100000 ints/s)
2006 * 10-20MB/s low (20000 ints/s)
2007 * 20-1249MB/s bulk (8000 ints/s)
2009 /* what was last interrupt timeslice? */
2010 timepassed_us = q_vector->itr >> 2;
2011 bytes_perint = bytes / timepassed_us; /* bytes/usec */
2013 switch (itr_setting) {
2014 case lowest_latency:
2015 if (bytes_perint > 10)
2016 itr_setting = low_latency;
2019 if (bytes_perint > 20)
2020 itr_setting = bulk_latency;
2021 else if (bytes_perint <= 10)
2022 itr_setting = lowest_latency;
2025 if (bytes_perint <= 20)
2026 itr_setting = low_latency;
2030 /* clear work counters since we have the values we need */
2031 ring_container->total_bytes = 0;
2032 ring_container->total_packets = 0;
2034 /* write updated itr to ring container */
2035 ring_container->itr = itr_setting;
2039 * ixgbe_write_eitr - write EITR register in hardware specific way
2040 * @q_vector: structure containing interrupt and ring information
2042 * This function is made to be called by ethtool and by the driver
2043 * when it needs to update EITR registers at runtime. Hardware
2044 * specific quirks/differences are taken care of here.
2046 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
2048 struct ixgbe_adapter *adapter = q_vector->adapter;
2049 struct ixgbe_hw *hw = &adapter->hw;
2050 int v_idx = q_vector->v_idx;
2051 u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
2053 switch (adapter->hw.mac.type) {
2054 case ixgbe_mac_82598EB:
2055 /* must write high and low 16 bits to reset counter */
2056 itr_reg |= (itr_reg << 16);
2058 case ixgbe_mac_82599EB:
2059 case ixgbe_mac_X540:
2061 * set the WDIS bit to not clear the timer bits and cause an
2062 * immediate assertion of the interrupt
2064 itr_reg |= IXGBE_EITR_CNT_WDIS;
2069 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
2072 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
2074 u32 new_itr = q_vector->itr;
2077 ixgbe_update_itr(q_vector, &q_vector->tx);
2078 ixgbe_update_itr(q_vector, &q_vector->rx);
2080 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
2082 switch (current_itr) {
2083 /* counts and packets in update_itr are dependent on these numbers */
2084 case lowest_latency:
2085 new_itr = IXGBE_100K_ITR;
2088 new_itr = IXGBE_20K_ITR;
2091 new_itr = IXGBE_8K_ITR;
2097 if (new_itr != q_vector->itr) {
2098 /* do an exponential smoothing */
2099 new_itr = (10 * new_itr * q_vector->itr) /
2100 ((9 * new_itr) + q_vector->itr);
2102 /* save the algorithm value here */
2103 q_vector->itr = new_itr;
2105 ixgbe_write_eitr(q_vector);
2110 * ixgbe_check_overtemp_subtask - check for over temperature
2111 * @adapter: pointer to adapter
2113 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2115 struct ixgbe_hw *hw = &adapter->hw;
2116 u32 eicr = adapter->interrupt_event;
2118 if (test_bit(__IXGBE_DOWN, &adapter->state))
2121 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2122 !(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2125 adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2127 switch (hw->device_id) {
2128 case IXGBE_DEV_ID_82599_T3_LOM:
2130 * Since the warning interrupt is for both ports
2131 * we don't have to check if:
2132 * - This interrupt wasn't for our port.
2133 * - We may have missed the interrupt so always have to
2134 * check if we got a LSC
2136 if (!(eicr & IXGBE_EICR_GPI_SDP0) &&
2137 !(eicr & IXGBE_EICR_LSC))
2140 if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2142 bool link_up = false;
2144 hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
2150 /* Check if this is not due to overtemp */
2151 if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
2156 if (!(eicr & IXGBE_EICR_GPI_SDP0))
2161 "Network adapter has been stopped because it has over heated. "
2162 "Restart the computer. If the problem persists, "
2163 "power off the system and replace the adapter\n");
2165 adapter->interrupt_event = 0;
2168 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2170 struct ixgbe_hw *hw = &adapter->hw;
2172 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2173 (eicr & IXGBE_EICR_GPI_SDP1)) {
2174 e_crit(probe, "Fan has stopped, replace the adapter\n");
2175 /* write to clear the interrupt */
2176 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
2180 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2182 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2185 switch (adapter->hw.mac.type) {
2186 case ixgbe_mac_82599EB:
2188 * Need to check link state so complete overtemp check
2191 if (((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)) &&
2192 (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2193 adapter->interrupt_event = eicr;
2194 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2195 ixgbe_service_event_schedule(adapter);
2199 case ixgbe_mac_X540:
2200 if (!(eicr & IXGBE_EICR_TS))
2208 "Network adapter has been stopped because it has over heated. "
2209 "Restart the computer. If the problem persists, "
2210 "power off the system and replace the adapter\n");
2213 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2215 struct ixgbe_hw *hw = &adapter->hw;
2217 if (eicr & IXGBE_EICR_GPI_SDP2) {
2218 /* Clear the interrupt */
2219 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
2220 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2221 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2222 ixgbe_service_event_schedule(adapter);
2226 if (eicr & IXGBE_EICR_GPI_SDP1) {
2227 /* Clear the interrupt */
2228 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
2229 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2230 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2231 ixgbe_service_event_schedule(adapter);
2236 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2238 struct ixgbe_hw *hw = &adapter->hw;
2241 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2242 adapter->link_check_timeout = jiffies;
2243 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2244 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2245 IXGBE_WRITE_FLUSH(hw);
2246 ixgbe_service_event_schedule(adapter);
2250 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2254 struct ixgbe_hw *hw = &adapter->hw;
2256 switch (hw->mac.type) {
2257 case ixgbe_mac_82598EB:
2258 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2259 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2261 case ixgbe_mac_82599EB:
2262 case ixgbe_mac_X540:
2263 mask = (qmask & 0xFFFFFFFF);
2265 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2266 mask = (qmask >> 32);
2268 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2273 /* skip the flush */
2276 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
2280 struct ixgbe_hw *hw = &adapter->hw;
2282 switch (hw->mac.type) {
2283 case ixgbe_mac_82598EB:
2284 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2285 IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
2287 case ixgbe_mac_82599EB:
2288 case ixgbe_mac_X540:
2289 mask = (qmask & 0xFFFFFFFF);
2291 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
2292 mask = (qmask >> 32);
2294 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
2299 /* skip the flush */
2303 * ixgbe_irq_enable - Enable default interrupt generation settings
2304 * @adapter: board private structure
2306 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2309 u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2311 /* don't reenable LSC while waiting for link */
2312 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
2313 mask &= ~IXGBE_EIMS_LSC;
2315 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2316 switch (adapter->hw.mac.type) {
2317 case ixgbe_mac_82599EB:
2318 mask |= IXGBE_EIMS_GPI_SDP0;
2320 case ixgbe_mac_X540:
2321 mask |= IXGBE_EIMS_TS;
2326 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2327 mask |= IXGBE_EIMS_GPI_SDP1;
2328 switch (adapter->hw.mac.type) {
2329 case ixgbe_mac_82599EB:
2330 mask |= IXGBE_EIMS_GPI_SDP1;
2331 mask |= IXGBE_EIMS_GPI_SDP2;
2332 case ixgbe_mac_X540:
2333 mask |= IXGBE_EIMS_ECC;
2334 mask |= IXGBE_EIMS_MAILBOX;
2340 #ifdef CONFIG_IXGBE_PTP
2341 if (adapter->hw.mac.type == ixgbe_mac_X540)
2342 mask |= IXGBE_EIMS_TIMESYNC;
2345 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2346 !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
2347 mask |= IXGBE_EIMS_FLOW_DIR;
2349 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2351 ixgbe_irq_enable_queues(adapter, ~0);
2353 IXGBE_WRITE_FLUSH(&adapter->hw);
2356 static irqreturn_t ixgbe_msix_other(int irq, void *data)
2358 struct ixgbe_adapter *adapter = data;
2359 struct ixgbe_hw *hw = &adapter->hw;
2363 * Workaround for Silicon errata. Use clear-by-write instead
2364 * of clear-by-read. Reading with EICS will return the
2365 * interrupt causes without clearing, which later be done
2366 * with the write to EICR.
2368 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
2369 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
2371 if (eicr & IXGBE_EICR_LSC)
2372 ixgbe_check_lsc(adapter);
2374 if (eicr & IXGBE_EICR_MAILBOX)
2375 ixgbe_msg_task(adapter);
2377 switch (hw->mac.type) {
2378 case ixgbe_mac_82599EB:
2379 case ixgbe_mac_X540:
2380 if (eicr & IXGBE_EICR_ECC)
2381 e_info(link, "Received unrecoverable ECC Err, please "
2383 /* Handle Flow Director Full threshold interrupt */
2384 if (eicr & IXGBE_EICR_FLOW_DIR) {
2385 int reinit_count = 0;
2387 for (i = 0; i < adapter->num_tx_queues; i++) {
2388 struct ixgbe_ring *ring = adapter->tx_ring[i];
2389 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
2394 /* no more flow director interrupts until after init */
2395 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
2396 adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
2397 ixgbe_service_event_schedule(adapter);
2400 ixgbe_check_sfp_event(adapter, eicr);
2401 ixgbe_check_overtemp_event(adapter, eicr);
2407 ixgbe_check_fan_failure(adapter, eicr);
2409 #ifdef CONFIG_IXGBE_PTP
2410 if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
2411 ixgbe_ptp_check_pps_event(adapter, eicr);
2414 /* re-enable the original interrupt state, no lsc, no queues */
2415 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2416 ixgbe_irq_enable(adapter, false, false);
2421 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
2423 struct ixgbe_q_vector *q_vector = data;
2425 /* EIAM disabled interrupts (on this vector) for us */
2427 if (q_vector->rx.ring || q_vector->tx.ring)
2428 napi_schedule(&q_vector->napi);
2434 * ixgbe_poll - NAPI Rx polling callback
2435 * @napi: structure for representing this polling device
2436 * @budget: how many packets driver is allowed to clean
2438 * This function is used for legacy and MSI, NAPI mode
2440 int ixgbe_poll(struct napi_struct *napi, int budget)
2442 struct ixgbe_q_vector *q_vector =
2443 container_of(napi, struct ixgbe_q_vector, napi);
2444 struct ixgbe_adapter *adapter = q_vector->adapter;
2445 struct ixgbe_ring *ring;
2446 int per_ring_budget;
2447 bool clean_complete = true;
2449 #ifdef CONFIG_IXGBE_DCA
2450 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2451 ixgbe_update_dca(q_vector);
2454 ixgbe_for_each_ring(ring, q_vector->tx)
2455 clean_complete &= !!ixgbe_clean_tx_irq(q_vector, ring);
2457 /* attempt to distribute budget to each queue fairly, but don't allow
2458 * the budget to go below 1 because we'll exit polling */
2459 if (q_vector->rx.count > 1)
2460 per_ring_budget = max(budget/q_vector->rx.count, 1);
2462 per_ring_budget = budget;
2464 ixgbe_for_each_ring(ring, q_vector->rx)
2465 clean_complete &= ixgbe_clean_rx_irq(q_vector, ring,
2468 /* If all work not completed, return budget and keep polling */
2469 if (!clean_complete)
2472 /* all work done, exit the polling mode */
2473 napi_complete(napi);
2474 if (adapter->rx_itr_setting & 1)
2475 ixgbe_set_itr(q_vector);
2476 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2477 ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
2483 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2484 * @adapter: board private structure
2486 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2487 * interrupts from the kernel.
2489 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2491 struct net_device *netdev = adapter->netdev;
2495 for (vector = 0; vector < adapter->num_q_vectors; vector++) {
2496 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2497 struct msix_entry *entry = &adapter->msix_entries[vector];
2499 if (q_vector->tx.ring && q_vector->rx.ring) {
2500 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2501 "%s-%s-%d", netdev->name, "TxRx", ri++);
2503 } else if (q_vector->rx.ring) {
2504 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2505 "%s-%s-%d", netdev->name, "rx", ri++);
2506 } else if (q_vector->tx.ring) {
2507 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2508 "%s-%s-%d", netdev->name, "tx", ti++);
2510 /* skip this unused q_vector */
2513 err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
2514 q_vector->name, q_vector);
2516 e_err(probe, "request_irq failed for MSIX interrupt "
2517 "Error: %d\n", err);
2518 goto free_queue_irqs;
2520 /* If Flow Director is enabled, set interrupt affinity */
2521 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2522 /* assign the mask for this irq */
2523 irq_set_affinity_hint(entry->vector,
2524 &q_vector->affinity_mask);
2528 err = request_irq(adapter->msix_entries[vector].vector,
2529 ixgbe_msix_other, 0, netdev->name, adapter);
2531 e_err(probe, "request_irq for msix_other failed: %d\n", err);
2532 goto free_queue_irqs;
2540 irq_set_affinity_hint(adapter->msix_entries[vector].vector,
2542 free_irq(adapter->msix_entries[vector].vector,
2543 adapter->q_vector[vector]);
2545 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2546 pci_disable_msix(adapter->pdev);
2547 kfree(adapter->msix_entries);
2548 adapter->msix_entries = NULL;
2553 * ixgbe_intr - legacy mode Interrupt Handler
2554 * @irq: interrupt number
2555 * @data: pointer to a network interface device structure
2557 static irqreturn_t ixgbe_intr(int irq, void *data)
2559 struct ixgbe_adapter *adapter = data;
2560 struct ixgbe_hw *hw = &adapter->hw;
2561 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2565 * Workaround for silicon errata #26 on 82598. Mask the interrupt
2566 * before the read of EICR.
2568 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2570 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2571 * therefore no explicit interrupt disable is necessary */
2572 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2575 * shared interrupt alert!
2576 * make sure interrupts are enabled because the read will
2577 * have disabled interrupts due to EIAM
2578 * finish the workaround of silicon errata on 82598. Unmask
2579 * the interrupt that we masked before the EICR read.
2581 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2582 ixgbe_irq_enable(adapter, true, true);
2583 return IRQ_NONE; /* Not our interrupt */
2586 if (eicr & IXGBE_EICR_LSC)
2587 ixgbe_check_lsc(adapter);
2589 switch (hw->mac.type) {
2590 case ixgbe_mac_82599EB:
2591 ixgbe_check_sfp_event(adapter, eicr);
2593 case ixgbe_mac_X540:
2594 if (eicr & IXGBE_EICR_ECC)
2595 e_info(link, "Received unrecoverable ECC err, please "
2597 ixgbe_check_overtemp_event(adapter, eicr);
2603 ixgbe_check_fan_failure(adapter, eicr);
2604 #ifdef CONFIG_IXGBE_PTP
2605 if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
2606 ixgbe_ptp_check_pps_event(adapter, eicr);
2609 /* would disable interrupts here but EIAM disabled it */
2610 napi_schedule(&q_vector->napi);
2613 * re-enable link(maybe) and non-queue interrupts, no flush.
2614 * ixgbe_poll will re-enable the queue interrupts
2616 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2617 ixgbe_irq_enable(adapter, false, false);
2623 * ixgbe_request_irq - initialize interrupts
2624 * @adapter: board private structure
2626 * Attempts to configure interrupts using the best available
2627 * capabilities of the hardware and kernel.
2629 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2631 struct net_device *netdev = adapter->netdev;
2634 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2635 err = ixgbe_request_msix_irqs(adapter);
2636 else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
2637 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2638 netdev->name, adapter);
2640 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2641 netdev->name, adapter);
2644 e_err(probe, "request_irq failed, Error %d\n", err);
2649 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2653 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
2654 free_irq(adapter->pdev->irq, adapter);
2658 for (vector = 0; vector < adapter->num_q_vectors; vector++) {
2659 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2660 struct msix_entry *entry = &adapter->msix_entries[vector];
2662 /* free only the irqs that were actually requested */
2663 if (!q_vector->rx.ring && !q_vector->tx.ring)
2666 /* clear the affinity_mask in the IRQ descriptor */
2667 irq_set_affinity_hint(entry->vector, NULL);
2669 free_irq(entry->vector, q_vector);
2672 free_irq(adapter->msix_entries[vector++].vector, adapter);
2676 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2677 * @adapter: board private structure
2679 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2681 switch (adapter->hw.mac.type) {
2682 case ixgbe_mac_82598EB:
2683 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2685 case ixgbe_mac_82599EB:
2686 case ixgbe_mac_X540:
2687 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2688 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2689 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2694 IXGBE_WRITE_FLUSH(&adapter->hw);
2695 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2698 for (vector = 0; vector < adapter->num_q_vectors; vector++)
2699 synchronize_irq(adapter->msix_entries[vector].vector);
2701 synchronize_irq(adapter->msix_entries[vector++].vector);
2703 synchronize_irq(adapter->pdev->irq);
2708 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2711 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2713 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2716 if (adapter->rx_itr_setting == 1)
2717 q_vector->itr = IXGBE_20K_ITR;
2719 q_vector->itr = adapter->rx_itr_setting;
2721 ixgbe_write_eitr(q_vector);
2723 ixgbe_set_ivar(adapter, 0, 0, 0);
2724 ixgbe_set_ivar(adapter, 1, 0, 0);
2726 e_info(hw, "Legacy interrupt IVAR setup done\n");
2730 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2731 * @adapter: board private structure
2732 * @ring: structure containing ring specific data
2734 * Configure the Tx descriptor ring after a reset.
2736 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2737 struct ixgbe_ring *ring)
2739 struct ixgbe_hw *hw = &adapter->hw;
2740 u64 tdba = ring->dma;
2742 u32 txdctl = IXGBE_TXDCTL_ENABLE;
2743 u8 reg_idx = ring->reg_idx;
2745 /* disable queue to avoid issues while updating state */
2746 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
2747 IXGBE_WRITE_FLUSH(hw);
2749 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2750 (tdba & DMA_BIT_MASK(32)));
2751 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2752 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2753 ring->count * sizeof(union ixgbe_adv_tx_desc));
2754 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2755 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2756 ring->tail = hw->hw_addr + IXGBE_TDT(reg_idx);
2759 * set WTHRESH to encourage burst writeback, it should not be set
2760 * higher than 1 when ITR is 0 as it could cause false TX hangs
2762 * In order to avoid issues WTHRESH + PTHRESH should always be equal
2763 * to or less than the number of on chip descriptors, which is
2766 if (!ring->q_vector || (ring->q_vector->itr < 8))
2767 txdctl |= (1 << 16); /* WTHRESH = 1 */
2769 txdctl |= (8 << 16); /* WTHRESH = 8 */
2772 * Setting PTHRESH to 32 both improves performance
2773 * and avoids a TX hang with DFP enabled
2775 txdctl |= (1 << 8) | /* HTHRESH = 1 */
2776 32; /* PTHRESH = 32 */
2778 /* reinitialize flowdirector state */
2779 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2780 ring->atr_sample_rate = adapter->atr_sample_rate;
2781 ring->atr_count = 0;
2782 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
2784 ring->atr_sample_rate = 0;
2787 clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
2790 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
2792 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2793 if (hw->mac.type == ixgbe_mac_82598EB &&
2794 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2797 /* poll to verify queue is enabled */
2799 usleep_range(1000, 2000);
2800 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2801 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
2803 e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
2806 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
2808 struct ixgbe_hw *hw = &adapter->hw;
2810 u8 tcs = netdev_get_num_tc(adapter->netdev);
2812 if (hw->mac.type == ixgbe_mac_82598EB)
2815 /* disable the arbiter while setting MTQC */
2816 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2817 rttdcs |= IXGBE_RTTDCS_ARBDIS;
2818 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2820 /* set transmit pool layout */
2821 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
2822 mtqc = IXGBE_MTQC_VT_ENA;
2824 mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
2826 mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
2827 else if (adapter->ring_feature[RING_F_RSS].indices == 4)
2828 mtqc |= IXGBE_MTQC_32VF;
2830 mtqc |= IXGBE_MTQC_64VF;
2833 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
2835 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
2837 mtqc = IXGBE_MTQC_64Q_1PB;
2840 IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
2842 /* Enable Security TX Buffer IFG for multiple pb */
2844 u32 sectx = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
2845 sectx |= IXGBE_SECTX_DCB;
2846 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, sectx);
2849 /* re-enable the arbiter */
2850 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2851 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2855 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2856 * @adapter: board private structure
2858 * Configure the Tx unit of the MAC after a reset.
2860 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2862 struct ixgbe_hw *hw = &adapter->hw;
2866 ixgbe_setup_mtqc(adapter);
2868 if (hw->mac.type != ixgbe_mac_82598EB) {
2869 /* DMATXCTL.EN must be before Tx queues are enabled */
2870 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2871 dmatxctl |= IXGBE_DMATXCTL_TE;
2872 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2875 /* Setup the HW Tx Head and Tail descriptor pointers */
2876 for (i = 0; i < adapter->num_tx_queues; i++)
2877 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
2880 static void ixgbe_enable_rx_drop(struct ixgbe_adapter *adapter,
2881 struct ixgbe_ring *ring)
2883 struct ixgbe_hw *hw = &adapter->hw;
2884 u8 reg_idx = ring->reg_idx;
2885 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
2887 srrctl |= IXGBE_SRRCTL_DROP_EN;
2889 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
2892 static void ixgbe_disable_rx_drop(struct ixgbe_adapter *adapter,
2893 struct ixgbe_ring *ring)
2895 struct ixgbe_hw *hw = &adapter->hw;
2896 u8 reg_idx = ring->reg_idx;
2897 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
2899 srrctl &= ~IXGBE_SRRCTL_DROP_EN;
2901 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
2904 #ifdef CONFIG_IXGBE_DCB
2905 void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
2907 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
2911 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
2913 if (adapter->ixgbe_ieee_pfc)
2914 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
2917 * We should set the drop enable bit if:
2920 * Number of Rx queues > 1 and flow control is disabled
2922 * This allows us to avoid head of line blocking for security
2923 * and performance reasons.
2925 if (adapter->num_vfs || (adapter->num_rx_queues > 1 &&
2926 !(adapter->hw.fc.current_mode & ixgbe_fc_tx_pause) && !pfc_en)) {
2927 for (i = 0; i < adapter->num_rx_queues; i++)
2928 ixgbe_enable_rx_drop(adapter, adapter->rx_ring[i]);
2930 for (i = 0; i < adapter->num_rx_queues; i++)
2931 ixgbe_disable_rx_drop(adapter, adapter->rx_ring[i]);
2935 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2937 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2938 struct ixgbe_ring *rx_ring)
2940 struct ixgbe_hw *hw = &adapter->hw;
2942 u8 reg_idx = rx_ring->reg_idx;
2944 if (hw->mac.type == ixgbe_mac_82598EB) {
2945 u16 mask = adapter->ring_feature[RING_F_RSS].mask;
2948 * if VMDq is not active we must program one srrctl register
2949 * per RSS queue since we have enabled RDRXCTL.MVMEN
2954 /* configure header buffer length, needed for RSC */
2955 srrctl = IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
2957 /* configure the packet buffer length */
2958 srrctl |= ixgbe_rx_bufsz(rx_ring) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2960 /* configure descriptor type */
2961 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2963 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
2966 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2968 struct ixgbe_hw *hw = &adapter->hw;
2969 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2970 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2971 0x6A3E67EA, 0x14364D17, 0x3BED200D};
2972 u32 mrqc = 0, reta = 0;
2975 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
2978 * Program table for at least 2 queues w/ SR-IOV so that VFs can
2979 * make full use of any rings they may have. We will use the
2980 * PSRTYPE register to control how many rings we use within the PF.
2982 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) && (rss_i < 2))
2985 /* Fill out hash function seeds */
2986 for (i = 0; i < 10; i++)
2987 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2989 /* Fill out redirection table */
2990 for (i = 0, j = 0; i < 128; i++, j++) {
2993 /* reta = 4-byte sliding window of
2994 * 0x00..(indices-1)(indices-1)00..etc. */
2995 reta = (reta << 8) | (j * 0x11);
2997 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
3000 /* Disable indicating checksum in descriptor, enables RSS hash */
3001 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
3002 rxcsum |= IXGBE_RXCSUM_PCSD;
3003 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
3005 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3006 if (adapter->ring_feature[RING_F_RSS].mask)
3007 mrqc = IXGBE_MRQC_RSSEN;
3009 u8 tcs = netdev_get_num_tc(adapter->netdev);
3011 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3013 mrqc = IXGBE_MRQC_VMDQRT8TCEN; /* 8 TCs */
3015 mrqc = IXGBE_MRQC_VMDQRT4TCEN; /* 4 TCs */
3016 else if (adapter->ring_feature[RING_F_RSS].indices == 4)
3017 mrqc = IXGBE_MRQC_VMDQRSS32EN;
3019 mrqc = IXGBE_MRQC_VMDQRSS64EN;
3022 mrqc = IXGBE_MRQC_RTRSS8TCEN;
3024 mrqc = IXGBE_MRQC_RTRSS4TCEN;
3026 mrqc = IXGBE_MRQC_RSSEN;
3030 /* Perform hash on these packet types */
3031 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4 |
3032 IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
3033 IXGBE_MRQC_RSS_FIELD_IPV6 |
3034 IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
3036 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP)
3037 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
3038 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP)
3039 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
3041 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3045 * ixgbe_configure_rscctl - enable RSC for the indicated ring
3046 * @adapter: address of board private structure
3047 * @index: index of ring to set
3049 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
3050 struct ixgbe_ring *ring)
3052 struct ixgbe_hw *hw = &adapter->hw;
3054 u8 reg_idx = ring->reg_idx;
3056 if (!ring_is_rsc_enabled(ring))
3059 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
3060 rscctrl |= IXGBE_RSCCTL_RSCEN;
3062 * we must limit the number of descriptors so that the
3063 * total size of max desc * buf_len is not greater
3066 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
3067 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
3070 #define IXGBE_MAX_RX_DESC_POLL 10
3071 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
3072 struct ixgbe_ring *ring)
3074 struct ixgbe_hw *hw = &adapter->hw;
3075 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3077 u8 reg_idx = ring->reg_idx;
3079 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3080 if (hw->mac.type == ixgbe_mac_82598EB &&
3081 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3085 usleep_range(1000, 2000);
3086 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3087 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
3090 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
3091 "the polling period\n", reg_idx);
3095 void ixgbe_disable_rx_queue(struct ixgbe_adapter *adapter,
3096 struct ixgbe_ring *ring)
3098 struct ixgbe_hw *hw = &adapter->hw;
3099 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3101 u8 reg_idx = ring->reg_idx;
3103 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3104 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
3106 /* write value back with RXDCTL.ENABLE bit cleared */
3107 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3109 if (hw->mac.type == ixgbe_mac_82598EB &&
3110 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3113 /* the hardware may take up to 100us to really disable the rx queue */
3116 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3117 } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
3120 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not cleared within "
3121 "the polling period\n", reg_idx);
3125 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
3126 struct ixgbe_ring *ring)
3128 struct ixgbe_hw *hw = &adapter->hw;
3129 u64 rdba = ring->dma;
3131 u8 reg_idx = ring->reg_idx;
3133 /* disable queue to avoid issues while updating state */
3134 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3135 ixgbe_disable_rx_queue(adapter, ring);
3137 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
3138 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
3139 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
3140 ring->count * sizeof(union ixgbe_adv_rx_desc));
3141 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
3142 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
3143 ring->tail = hw->hw_addr + IXGBE_RDT(reg_idx);
3145 ixgbe_configure_srrctl(adapter, ring);
3146 ixgbe_configure_rscctl(adapter, ring);
3148 /* If operating in IOV mode set RLPML for X540 */
3149 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
3150 hw->mac.type == ixgbe_mac_X540) {
3151 rxdctl &= ~IXGBE_RXDCTL_RLPMLMASK;
3152 rxdctl |= ((ring->netdev->mtu + ETH_HLEN +
3153 ETH_FCS_LEN + VLAN_HLEN) | IXGBE_RXDCTL_RLPML_EN);
3156 if (hw->mac.type == ixgbe_mac_82598EB) {
3158 * enable cache line friendly hardware writes:
3159 * PTHRESH=32 descriptors (half the internal cache),
3160 * this also removes ugly rx_no_buffer_count increment
3161 * HTHRESH=4 descriptors (to minimize latency on fetch)
3162 * WTHRESH=8 burst writeback up to two cache lines
3164 rxdctl &= ~0x3FFFFF;
3168 /* enable receive descriptor ring */
3169 rxdctl |= IXGBE_RXDCTL_ENABLE;
3170 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3172 ixgbe_rx_desc_queue_enable(adapter, ring);
3173 ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
3176 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
3178 struct ixgbe_hw *hw = &adapter->hw;
3179 int rss_i = adapter->ring_feature[RING_F_RSS].indices;
3182 /* PSRTYPE must be initialized in non 82598 adapters */
3183 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
3184 IXGBE_PSRTYPE_UDPHDR |
3185 IXGBE_PSRTYPE_IPV4HDR |
3186 IXGBE_PSRTYPE_L2HDR |
3187 IXGBE_PSRTYPE_IPV6HDR;
3189 if (hw->mac.type == ixgbe_mac_82598EB)
3197 for (p = 0; p < adapter->num_rx_pools; p++)
3198 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(p)),
3202 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
3204 struct ixgbe_hw *hw = &adapter->hw;
3205 u32 reg_offset, vf_shift;
3206 u32 gcr_ext, vmdctl;
3209 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
3212 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
3213 vmdctl |= IXGBE_VMD_CTL_VMDQ_EN;
3214 vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
3215 vmdctl |= VMDQ_P(0) << IXGBE_VT_CTL_POOL_SHIFT;
3216 vmdctl |= IXGBE_VT_CTL_REPLEN;
3217 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
3219 vf_shift = VMDQ_P(0) % 32;
3220 reg_offset = (VMDQ_P(0) >= 32) ? 1 : 0;
3222 /* Enable only the PF's pool for Tx/Rx */
3223 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (~0) << vf_shift);
3224 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), reg_offset - 1);
3225 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (~0) << vf_shift);
3226 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), reg_offset - 1);
3227 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3229 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
3230 hw->mac.ops.set_vmdq(hw, 0, VMDQ_P(0));
3233 * Set up VF register offsets for selected VT Mode,
3234 * i.e. 32 or 64 VFs for SR-IOV
3236 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
3237 case IXGBE_82599_VMDQ_8Q_MASK:
3238 gcr_ext = IXGBE_GCR_EXT_VT_MODE_16;
3240 case IXGBE_82599_VMDQ_4Q_MASK:
3241 gcr_ext = IXGBE_GCR_EXT_VT_MODE_32;
3244 gcr_ext = IXGBE_GCR_EXT_VT_MODE_64;
3248 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
3250 /* enable Tx loopback for VF/PF communication */
3251 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3253 /* Enable MAC Anti-Spoofing */
3254 hw->mac.ops.set_mac_anti_spoofing(hw, (adapter->num_vfs != 0),
3256 /* For VFs that have spoof checking turned off */
3257 for (i = 0; i < adapter->num_vfs; i++) {
3258 if (!adapter->vfinfo[i].spoofchk_enabled)
3259 ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i, false);
3263 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
3265 struct ixgbe_hw *hw = &adapter->hw;
3266 struct net_device *netdev = adapter->netdev;
3267 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3268 struct ixgbe_ring *rx_ring;
3273 /* adjust max frame to be able to do baby jumbo for FCoE */
3274 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
3275 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
3276 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3278 #endif /* IXGBE_FCOE */
3280 /* adjust max frame to be at least the size of a standard frame */
3281 if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
3282 max_frame = (ETH_FRAME_LEN + ETH_FCS_LEN);
3284 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
3285 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
3286 mhadd &= ~IXGBE_MHADD_MFS_MASK;
3287 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
3289 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
3292 /* MHADD will allow an extra 4 bytes past for vlan tagged frames */
3293 max_frame += VLAN_HLEN;
3295 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
3296 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
3297 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
3298 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
3301 * Setup the HW Rx Head and Tail Descriptor Pointers and
3302 * the Base and Length of the Rx Descriptor Ring
3304 for (i = 0; i < adapter->num_rx_queues; i++) {
3305 rx_ring = adapter->rx_ring[i];
3306 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
3307 set_ring_rsc_enabled(rx_ring);
3309 clear_ring_rsc_enabled(rx_ring);
3313 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
3315 struct ixgbe_hw *hw = &adapter->hw;
3316 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
3318 switch (hw->mac.type) {
3319 case ixgbe_mac_82598EB:
3321 * For VMDq support of different descriptor types or
3322 * buffer sizes through the use of multiple SRRCTL
3323 * registers, RDRXCTL.MVMEN must be set to 1
3325 * also, the manual doesn't mention it clearly but DCA hints
3326 * will only use queue 0's tags unless this bit is set. Side
3327 * effects of setting this bit are only that SRRCTL must be
3328 * fully programmed [0..15]
3330 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
3332 case ixgbe_mac_82599EB:
3333 case ixgbe_mac_X540:
3334 /* Disable RSC for ACK packets */
3335 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
3336 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
3337 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3338 /* hardware requires some bits to be set by default */
3339 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3340 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3343 /* We should do nothing since we don't know this hardware */
3347 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3351 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3352 * @adapter: board private structure
3354 * Configure the Rx unit of the MAC after a reset.
3356 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3358 struct ixgbe_hw *hw = &adapter->hw;
3362 /* disable receives while setting up the descriptors */
3363 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3364 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3366 ixgbe_setup_psrtype(adapter);
3367 ixgbe_setup_rdrxctl(adapter);
3369 /* Program registers for the distribution of queues */
3370 ixgbe_setup_mrqc(adapter);
3372 /* set_rx_buffer_len must be called before ring initialization */
3373 ixgbe_set_rx_buffer_len(adapter);
3376 * Setup the HW Rx Head and Tail Descriptor Pointers and
3377 * the Base and Length of the Rx Descriptor Ring
3379 for (i = 0; i < adapter->num_rx_queues; i++)
3380 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3382 /* disable drop enable for 82598 parts */
3383 if (hw->mac.type == ixgbe_mac_82598EB)
3384 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3386 /* enable all receives */
3387 rxctrl |= IXGBE_RXCTRL_RXEN;
3388 hw->mac.ops.enable_rx_dma(hw, rxctrl);
3391 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3393 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3394 struct ixgbe_hw *hw = &adapter->hw;
3396 /* add VID to filter table */
3397 hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), true);
3398 set_bit(vid, adapter->active_vlans);
3403 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3405 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3406 struct ixgbe_hw *hw = &adapter->hw;
3408 /* remove VID from filter table */
3409 hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), false);
3410 clear_bit(vid, adapter->active_vlans);
3416 * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3417 * @adapter: driver data
3419 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
3421 struct ixgbe_hw *hw = &adapter->hw;
3424 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3425 vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
3426 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3430 * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3431 * @adapter: driver data
3433 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
3435 struct ixgbe_hw *hw = &adapter->hw;
3438 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3439 vlnctrl |= IXGBE_VLNCTRL_VFE;
3440 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3441 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3445 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3446 * @adapter: driver data
3448 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3450 struct ixgbe_hw *hw = &adapter->hw;
3454 switch (hw->mac.type) {
3455 case ixgbe_mac_82598EB:
3456 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3457 vlnctrl &= ~IXGBE_VLNCTRL_VME;
3458 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3460 case ixgbe_mac_82599EB:
3461 case ixgbe_mac_X540:
3462 for (i = 0; i < adapter->num_rx_queues; i++) {
3463 j = adapter->rx_ring[i]->reg_idx;
3464 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3465 vlnctrl &= ~IXGBE_RXDCTL_VME;
3466 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3475 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3476 * @adapter: driver data
3478 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3480 struct ixgbe_hw *hw = &adapter->hw;
3484 switch (hw->mac.type) {
3485 case ixgbe_mac_82598EB:
3486 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3487 vlnctrl |= IXGBE_VLNCTRL_VME;
3488 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3490 case ixgbe_mac_82599EB:
3491 case ixgbe_mac_X540:
3492 for (i = 0; i < adapter->num_rx_queues; i++) {
3493 j = adapter->rx_ring[i]->reg_idx;
3494 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3495 vlnctrl |= IXGBE_RXDCTL_VME;
3496 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3504 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3508 ixgbe_vlan_rx_add_vid(adapter->netdev, 0);
3510 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3511 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
3515 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3516 * @netdev: network interface device structure
3518 * Writes unicast address list to the RAR table.
3519 * Returns: -ENOMEM on failure/insufficient address space
3520 * 0 on no addresses written
3521 * X on writing X addresses to the RAR table
3523 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3525 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3526 struct ixgbe_hw *hw = &adapter->hw;
3527 unsigned int rar_entries = hw->mac.num_rar_entries - 1;
3530 /* In SR-IOV mode significantly less RAR entries are available */
3531 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3532 rar_entries = IXGBE_MAX_PF_MACVLANS - 1;
3534 /* return ENOMEM indicating insufficient memory for addresses */
3535 if (netdev_uc_count(netdev) > rar_entries)
3538 if (!netdev_uc_empty(netdev)) {
3539 struct netdev_hw_addr *ha;
3540 /* return error if we do not support writing to RAR table */
3541 if (!hw->mac.ops.set_rar)
3544 netdev_for_each_uc_addr(ha, netdev) {
3547 hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3548 VMDQ_P(0), IXGBE_RAH_AV);
3552 /* write the addresses in reverse order to avoid write combining */
3553 for (; rar_entries > 0 ; rar_entries--)
3554 hw->mac.ops.clear_rar(hw, rar_entries);
3560 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3561 * @netdev: network interface device structure
3563 * The set_rx_method entry point is called whenever the unicast/multicast
3564 * address list or the network interface flags are updated. This routine is
3565 * responsible for configuring the hardware for proper unicast, multicast and
3568 void ixgbe_set_rx_mode(struct net_device *netdev)
3570 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3571 struct ixgbe_hw *hw = &adapter->hw;
3572 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3575 /* Check for Promiscuous and All Multicast modes */
3577 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3579 /* set all bits that we expect to always be set */
3580 fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
3581 fctrl |= IXGBE_FCTRL_BAM;
3582 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3583 fctrl |= IXGBE_FCTRL_PMCF;
3585 /* clear the bits we are changing the status of */
3586 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3588 if (netdev->flags & IFF_PROMISC) {
3589 hw->addr_ctrl.user_set_promisc = true;
3590 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3591 vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3592 /* don't hardware filter vlans in promisc mode */
3593 ixgbe_vlan_filter_disable(adapter);
3595 if (netdev->flags & IFF_ALLMULTI) {
3596 fctrl |= IXGBE_FCTRL_MPE;
3597 vmolr |= IXGBE_VMOLR_MPE;
3600 * Write addresses to the MTA, if the attempt fails
3601 * then we should just turn on promiscuous mode so
3602 * that we can at least receive multicast traffic
3604 hw->mac.ops.update_mc_addr_list(hw, netdev);
3605 vmolr |= IXGBE_VMOLR_ROMPE;
3607 ixgbe_vlan_filter_enable(adapter);
3608 hw->addr_ctrl.user_set_promisc = false;
3612 * Write addresses to available RAR registers, if there is not
3613 * sufficient space to store all the addresses then enable
3614 * unicast promiscuous mode
3616 count = ixgbe_write_uc_addr_list(netdev);
3618 fctrl |= IXGBE_FCTRL_UPE;
3619 vmolr |= IXGBE_VMOLR_ROPE;
3622 if (adapter->num_vfs)
3623 ixgbe_restore_vf_multicasts(adapter);
3625 if (hw->mac.type != ixgbe_mac_82598EB) {
3626 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(VMDQ_P(0))) &
3627 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3629 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(0)), vmolr);
3632 /* This is useful for sniffing bad packets. */
3633 if (adapter->netdev->features & NETIF_F_RXALL) {
3634 /* UPE and MPE will be handled by normal PROMISC logic
3635 * in e1000e_set_rx_mode */
3636 fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
3637 IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
3638 IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
3640 fctrl &= ~(IXGBE_FCTRL_DPF);
3641 /* NOTE: VLAN filtering is disabled by setting PROMISC */
3644 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3646 if (netdev->features & NETIF_F_HW_VLAN_RX)
3647 ixgbe_vlan_strip_enable(adapter);
3649 ixgbe_vlan_strip_disable(adapter);
3652 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3656 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
3657 napi_enable(&adapter->q_vector[q_idx]->napi);
3660 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3664 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
3665 napi_disable(&adapter->q_vector[q_idx]->napi);
3668 #ifdef CONFIG_IXGBE_DCB
3670 * ixgbe_configure_dcb - Configure DCB hardware
3671 * @adapter: ixgbe adapter struct
3673 * This is called by the driver on open to configure the DCB hardware.
3674 * This is also called by the gennetlink interface when reconfiguring
3677 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3679 struct ixgbe_hw *hw = &adapter->hw;
3680 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3682 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3683 if (hw->mac.type == ixgbe_mac_82598EB)
3684 netif_set_gso_max_size(adapter->netdev, 65536);
3688 if (hw->mac.type == ixgbe_mac_82598EB)
3689 netif_set_gso_max_size(adapter->netdev, 32768);
3692 if (adapter->netdev->features & NETIF_F_FCOE_MTU)
3693 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
3696 /* reconfigure the hardware */
3697 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
3698 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3700 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3702 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
3703 } else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
3704 ixgbe_dcb_hw_ets(&adapter->hw,
3705 adapter->ixgbe_ieee_ets,
3707 ixgbe_dcb_hw_pfc_config(&adapter->hw,
3708 adapter->ixgbe_ieee_pfc->pfc_en,
3709 adapter->ixgbe_ieee_ets->prio_tc);
3712 /* Enable RSS Hash per TC */
3713 if (hw->mac.type != ixgbe_mac_82598EB) {
3715 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices - 1;
3722 /* write msb to all 8 TCs in one write */
3723 IXGBE_WRITE_REG(hw, IXGBE_RQTC, msb * 0x11111111);
3728 /* Additional bittime to account for IXGBE framing */
3729 #define IXGBE_ETH_FRAMING 20
3732 * ixgbe_hpbthresh - calculate high water mark for flow control
3734 * @adapter: board private structure to calculate for
3735 * @pb: packet buffer to calculate
3737 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
3739 struct ixgbe_hw *hw = &adapter->hw;
3740 struct net_device *dev = adapter->netdev;
3741 int link, tc, kb, marker;
3744 /* Calculate max LAN frame size */
3745 tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
3748 /* FCoE traffic class uses FCOE jumbo frames */
3749 if ((dev->features & NETIF_F_FCOE_MTU) &&
3750 (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
3751 (pb == ixgbe_fcoe_get_tc(adapter)))
3752 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3755 /* Calculate delay value for device */
3756 switch (hw->mac.type) {
3757 case ixgbe_mac_X540:
3758 dv_id = IXGBE_DV_X540(link, tc);
3761 dv_id = IXGBE_DV(link, tc);
3765 /* Loopback switch introduces additional latency */
3766 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3767 dv_id += IXGBE_B2BT(tc);
3769 /* Delay value is calculated in bit times convert to KB */
3770 kb = IXGBE_BT2KB(dv_id);
3771 rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
3773 marker = rx_pba - kb;
3775 /* It is possible that the packet buffer is not large enough
3776 * to provide required headroom. In this case throw an error
3777 * to user and a do the best we can.
3780 e_warn(drv, "Packet Buffer(%i) can not provide enough"
3781 "headroom to support flow control."
3782 "Decrease MTU or number of traffic classes\n", pb);
3790 * ixgbe_lpbthresh - calculate low water mark for for flow control
3792 * @adapter: board private structure to calculate for
3793 * @pb: packet buffer to calculate
3795 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter)
3797 struct ixgbe_hw *hw = &adapter->hw;
3798 struct net_device *dev = adapter->netdev;
3802 /* Calculate max LAN frame size */
3803 tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
3805 /* Calculate delay value for device */
3806 switch (hw->mac.type) {
3807 case ixgbe_mac_X540:
3808 dv_id = IXGBE_LOW_DV_X540(tc);
3811 dv_id = IXGBE_LOW_DV(tc);
3815 /* Delay value is calculated in bit times convert to KB */
3816 return IXGBE_BT2KB(dv_id);
3820 * ixgbe_pbthresh_setup - calculate and setup high low water marks
3822 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
3824 struct ixgbe_hw *hw = &adapter->hw;
3825 int num_tc = netdev_get_num_tc(adapter->netdev);
3831 hw->fc.low_water = ixgbe_lpbthresh(adapter);
3833 for (i = 0; i < num_tc; i++) {
3834 hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
3836 /* Low water marks must not be larger than high water marks */
3837 if (hw->fc.low_water > hw->fc.high_water[i])
3838 hw->fc.low_water = 0;
3842 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
3844 struct ixgbe_hw *hw = &adapter->hw;
3846 u8 tc = netdev_get_num_tc(adapter->netdev);
3848 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3849 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3850 hdrm = 32 << adapter->fdir_pballoc;
3854 hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
3855 ixgbe_pbthresh_setup(adapter);
3858 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
3860 struct ixgbe_hw *hw = &adapter->hw;
3861 struct hlist_node *node, *node2;
3862 struct ixgbe_fdir_filter *filter;
3864 spin_lock(&adapter->fdir_perfect_lock);
3866 if (!hlist_empty(&adapter->fdir_filter_list))
3867 ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
3869 hlist_for_each_entry_safe(filter, node, node2,
3870 &adapter->fdir_filter_list, fdir_node) {
3871 ixgbe_fdir_write_perfect_filter_82599(hw,
3874 (filter->action == IXGBE_FDIR_DROP_QUEUE) ?
3875 IXGBE_FDIR_DROP_QUEUE :
3876 adapter->rx_ring[filter->action]->reg_idx);
3879 spin_unlock(&adapter->fdir_perfect_lock);
3882 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3884 struct ixgbe_hw *hw = &adapter->hw;
3886 ixgbe_configure_pb(adapter);
3887 #ifdef CONFIG_IXGBE_DCB
3888 ixgbe_configure_dcb(adapter);
3891 * We must restore virtualization before VLANs or else
3892 * the VLVF registers will not be populated
3894 ixgbe_configure_virtualization(adapter);
3896 ixgbe_set_rx_mode(adapter->netdev);
3897 ixgbe_restore_vlan(adapter);
3899 switch (hw->mac.type) {
3900 case ixgbe_mac_82599EB:
3901 case ixgbe_mac_X540:
3902 hw->mac.ops.disable_rx_buff(hw);
3908 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3909 ixgbe_init_fdir_signature_82599(&adapter->hw,
3910 adapter->fdir_pballoc);
3911 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3912 ixgbe_init_fdir_perfect_82599(&adapter->hw,
3913 adapter->fdir_pballoc);
3914 ixgbe_fdir_filter_restore(adapter);
3917 switch (hw->mac.type) {
3918 case ixgbe_mac_82599EB:
3919 case ixgbe_mac_X540:
3920 hw->mac.ops.enable_rx_buff(hw);
3927 /* configure FCoE L2 filters, redirection table, and Rx control */
3928 ixgbe_configure_fcoe(adapter);
3930 #endif /* IXGBE_FCOE */
3931 ixgbe_configure_tx(adapter);
3932 ixgbe_configure_rx(adapter);
3935 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3937 switch (hw->phy.type) {
3938 case ixgbe_phy_sfp_avago:
3939 case ixgbe_phy_sfp_ftl:
3940 case ixgbe_phy_sfp_intel:
3941 case ixgbe_phy_sfp_unknown:
3942 case ixgbe_phy_sfp_passive_tyco:
3943 case ixgbe_phy_sfp_passive_unknown:
3944 case ixgbe_phy_sfp_active_unknown:
3945 case ixgbe_phy_sfp_ftl_active:
3948 if (hw->mac.type == ixgbe_mac_82598EB)
3956 * ixgbe_sfp_link_config - set up SFP+ link
3957 * @adapter: pointer to private adapter struct
3959 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3962 * We are assuming the worst case scenario here, and that
3963 * is that an SFP was inserted/removed after the reset
3964 * but before SFP detection was enabled. As such the best
3965 * solution is to just start searching as soon as we start
3967 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3968 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
3970 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
3974 * ixgbe_non_sfp_link_config - set up non-SFP+ link
3975 * @hw: pointer to private hardware struct
3977 * Returns 0 on success, negative on failure
3979 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3982 bool negotiation, link_up = false;
3983 u32 ret = IXGBE_ERR_LINK_SETUP;
3985 if (hw->mac.ops.check_link)
3986 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3991 autoneg = hw->phy.autoneg_advertised;
3992 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
3993 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
3998 if (hw->mac.ops.setup_link)
3999 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
4004 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
4006 struct ixgbe_hw *hw = &adapter->hw;
4009 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4010 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
4012 gpie |= IXGBE_GPIE_EIAME;
4014 * use EIAM to auto-mask when MSI-X interrupt is asserted
4015 * this saves a register write for every interrupt
4017 switch (hw->mac.type) {
4018 case ixgbe_mac_82598EB:
4019 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
4021 case ixgbe_mac_82599EB:
4022 case ixgbe_mac_X540:
4024 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
4025 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
4029 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
4030 * specifically only auto mask tx and rx interrupts */
4031 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
4034 /* XXX: to interrupt immediately for EICS writes, enable this */
4035 /* gpie |= IXGBE_GPIE_EIMEN; */
4037 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
4038 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
4040 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
4041 case IXGBE_82599_VMDQ_8Q_MASK:
4042 gpie |= IXGBE_GPIE_VTMODE_16;
4044 case IXGBE_82599_VMDQ_4Q_MASK:
4045 gpie |= IXGBE_GPIE_VTMODE_32;
4048 gpie |= IXGBE_GPIE_VTMODE_64;
4053 /* Enable Thermal over heat sensor interrupt */
4054 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
4055 switch (adapter->hw.mac.type) {
4056 case ixgbe_mac_82599EB:
4057 gpie |= IXGBE_SDP0_GPIEN;
4059 case ixgbe_mac_X540:
4060 gpie |= IXGBE_EIMS_TS;
4067 /* Enable fan failure interrupt */
4068 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
4069 gpie |= IXGBE_SDP1_GPIEN;
4071 if (hw->mac.type == ixgbe_mac_82599EB) {
4072 gpie |= IXGBE_SDP1_GPIEN;
4073 gpie |= IXGBE_SDP2_GPIEN;
4076 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
4079 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
4081 struct ixgbe_hw *hw = &adapter->hw;
4085 ixgbe_get_hw_control(adapter);
4086 ixgbe_setup_gpie(adapter);
4088 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4089 ixgbe_configure_msix(adapter);
4091 ixgbe_configure_msi_and_legacy(adapter);
4093 /* enable the optics for both mult-speed fiber and 82599 SFP+ fiber */
4094 if (hw->mac.ops.enable_tx_laser &&
4095 ((hw->phy.multispeed_fiber) ||
4096 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
4097 (hw->mac.type == ixgbe_mac_82599EB))))
4098 hw->mac.ops.enable_tx_laser(hw);
4100 clear_bit(__IXGBE_DOWN, &adapter->state);
4101 ixgbe_napi_enable_all(adapter);
4103 if (ixgbe_is_sfp(hw)) {
4104 ixgbe_sfp_link_config(adapter);
4106 err = ixgbe_non_sfp_link_config(hw);
4108 e_err(probe, "link_config FAILED %d\n", err);
4111 /* clear any pending interrupts, may auto mask */
4112 IXGBE_READ_REG(hw, IXGBE_EICR);
4113 ixgbe_irq_enable(adapter, true, true);
4116 * If this adapter has a fan, check to see if we had a failure
4117 * before we enabled the interrupt.
4119 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
4120 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
4121 if (esdp & IXGBE_ESDP_SDP1)
4122 e_crit(drv, "Fan has stopped, replace the adapter\n");
4125 /* enable transmits */
4126 netif_tx_start_all_queues(adapter->netdev);
4128 /* bring the link up in the watchdog, this could race with our first
4129 * link up interrupt but shouldn't be a problem */
4130 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
4131 adapter->link_check_timeout = jiffies;
4132 mod_timer(&adapter->service_timer, jiffies);
4134 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
4135 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
4136 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
4137 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
4140 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
4142 WARN_ON(in_interrupt());
4143 /* put off any impending NetWatchDogTimeout */
4144 adapter->netdev->trans_start = jiffies;
4146 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
4147 usleep_range(1000, 2000);
4148 ixgbe_down(adapter);
4150 * If SR-IOV enabled then wait a bit before bringing the adapter
4151 * back up to give the VFs time to respond to the reset. The
4152 * two second wait is based upon the watchdog timer cycle in
4155 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4158 clear_bit(__IXGBE_RESETTING, &adapter->state);
4161 void ixgbe_up(struct ixgbe_adapter *adapter)
4163 /* hardware has been reset, we need to reload some things */
4164 ixgbe_configure(adapter);
4166 ixgbe_up_complete(adapter);
4169 void ixgbe_reset(struct ixgbe_adapter *adapter)
4171 struct ixgbe_hw *hw = &adapter->hw;
4174 /* lock SFP init bit to prevent race conditions with the watchdog */
4175 while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
4176 usleep_range(1000, 2000);
4178 /* clear all SFP and link config related flags while holding SFP_INIT */
4179 adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
4180 IXGBE_FLAG2_SFP_NEEDS_RESET);
4181 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
4183 err = hw->mac.ops.init_hw(hw);
4186 case IXGBE_ERR_SFP_NOT_PRESENT:
4187 case IXGBE_ERR_SFP_NOT_SUPPORTED:
4189 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
4190 e_dev_err("master disable timed out\n");
4192 case IXGBE_ERR_EEPROM_VERSION:
4193 /* We are running on a pre-production device, log a warning */
4194 e_dev_warn("This device is a pre-production adapter/LOM. "
4195 "Please be aware there may be issues associated with "
4196 "your hardware. If you are experiencing problems "
4197 "please contact your Intel or hardware "
4198 "representative who provided you with this "
4202 e_dev_err("Hardware Error: %d\n", err);
4205 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
4207 /* reprogram the RAR[0] in case user changed it. */
4208 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, VMDQ_P(0), IXGBE_RAH_AV);
4210 /* update SAN MAC vmdq pool selection */
4211 if (hw->mac.san_mac_rar_index)
4212 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
4216 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
4217 * @rx_ring: ring to free buffers from
4219 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
4221 struct device *dev = rx_ring->dev;
4225 /* ring already cleared, nothing to do */
4226 if (!rx_ring->rx_buffer_info)
4229 /* Free all the Rx ring sk_buffs */
4230 for (i = 0; i < rx_ring->count; i++) {
4231 struct ixgbe_rx_buffer *rx_buffer;
4233 rx_buffer = &rx_ring->rx_buffer_info[i];
4234 if (rx_buffer->skb) {
4235 struct sk_buff *skb = rx_buffer->skb;
4236 if (IXGBE_CB(skb)->page_released) {
4239 ixgbe_rx_bufsz(rx_ring),
4241 IXGBE_CB(skb)->page_released = false;
4245 rx_buffer->skb = NULL;
4247 dma_unmap_page(dev, rx_buffer->dma,
4248 ixgbe_rx_pg_size(rx_ring),
4251 if (rx_buffer->page)
4252 __free_pages(rx_buffer->page,
4253 ixgbe_rx_pg_order(rx_ring));
4254 rx_buffer->page = NULL;
4257 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4258 memset(rx_ring->rx_buffer_info, 0, size);
4260 /* Zero out the descriptor ring */
4261 memset(rx_ring->desc, 0, rx_ring->size);
4263 rx_ring->next_to_alloc = 0;
4264 rx_ring->next_to_clean = 0;
4265 rx_ring->next_to_use = 0;
4269 * ixgbe_clean_tx_ring - Free Tx Buffers
4270 * @tx_ring: ring to be cleaned
4272 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
4274 struct ixgbe_tx_buffer *tx_buffer_info;
4278 /* ring already cleared, nothing to do */
4279 if (!tx_ring->tx_buffer_info)
4282 /* Free all the Tx ring sk_buffs */
4283 for (i = 0; i < tx_ring->count; i++) {
4284 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4285 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
4288 netdev_tx_reset_queue(txring_txq(tx_ring));
4290 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4291 memset(tx_ring->tx_buffer_info, 0, size);
4293 /* Zero out the descriptor ring */
4294 memset(tx_ring->desc, 0, tx_ring->size);
4296 tx_ring->next_to_use = 0;
4297 tx_ring->next_to_clean = 0;
4301 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
4302 * @adapter: board private structure
4304 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
4308 for (i = 0; i < adapter->num_rx_queues; i++)
4309 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
4313 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
4314 * @adapter: board private structure
4316 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
4320 for (i = 0; i < adapter->num_tx_queues; i++)
4321 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
4324 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
4326 struct hlist_node *node, *node2;
4327 struct ixgbe_fdir_filter *filter;
4329 spin_lock(&adapter->fdir_perfect_lock);
4331 hlist_for_each_entry_safe(filter, node, node2,
4332 &adapter->fdir_filter_list, fdir_node) {
4333 hlist_del(&filter->fdir_node);
4336 adapter->fdir_filter_count = 0;
4338 spin_unlock(&adapter->fdir_perfect_lock);
4341 void ixgbe_down(struct ixgbe_adapter *adapter)
4343 struct net_device *netdev = adapter->netdev;
4344 struct ixgbe_hw *hw = &adapter->hw;
4348 /* signal that we are down to the interrupt handler */
4349 set_bit(__IXGBE_DOWN, &adapter->state);
4351 /* disable receives */
4352 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4353 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
4355 /* disable all enabled rx queues */
4356 for (i = 0; i < adapter->num_rx_queues; i++)
4357 /* this call also flushes the previous write */
4358 ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]);
4360 usleep_range(10000, 20000);
4362 netif_tx_stop_all_queues(netdev);
4364 /* call carrier off first to avoid false dev_watchdog timeouts */
4365 netif_carrier_off(netdev);
4366 netif_tx_disable(netdev);
4368 ixgbe_irq_disable(adapter);
4370 ixgbe_napi_disable_all(adapter);
4372 adapter->flags2 &= ~(IXGBE_FLAG2_FDIR_REQUIRES_REINIT |
4373 IXGBE_FLAG2_RESET_REQUESTED);
4374 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
4376 del_timer_sync(&adapter->service_timer);
4378 if (adapter->num_vfs) {
4379 /* Clear EITR Select mapping */
4380 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
4382 /* Mark all the VFs as inactive */
4383 for (i = 0 ; i < adapter->num_vfs; i++)
4384 adapter->vfinfo[i].clear_to_send = false;
4386 /* ping all the active vfs to let them know we are going down */
4387 ixgbe_ping_all_vfs(adapter);
4389 /* Disable all VFTE/VFRE TX/RX */
4390 ixgbe_disable_tx_rx(adapter);
4393 /* disable transmits in the hardware now that interrupts are off */
4394 for (i = 0; i < adapter->num_tx_queues; i++) {
4395 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
4396 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
4399 /* Disable the Tx DMA engine on 82599 and X540 */
4400 switch (hw->mac.type) {
4401 case ixgbe_mac_82599EB:
4402 case ixgbe_mac_X540:
4403 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
4404 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
4405 ~IXGBE_DMATXCTL_TE));
4411 if (!pci_channel_offline(adapter->pdev))
4412 ixgbe_reset(adapter);
4414 /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
4415 if (hw->mac.ops.disable_tx_laser &&
4416 ((hw->phy.multispeed_fiber) ||
4417 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
4418 (hw->mac.type == ixgbe_mac_82599EB))))
4419 hw->mac.ops.disable_tx_laser(hw);
4421 ixgbe_clean_all_tx_rings(adapter);
4422 ixgbe_clean_all_rx_rings(adapter);
4424 #ifdef CONFIG_IXGBE_DCA
4425 /* since we reset the hardware DCA settings were cleared */
4426 ixgbe_setup_dca(adapter);
4431 * ixgbe_tx_timeout - Respond to a Tx Hang
4432 * @netdev: network interface device structure
4434 static void ixgbe_tx_timeout(struct net_device *netdev)
4436 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4438 /* Do the reset outside of interrupt context */
4439 ixgbe_tx_timeout_reset(adapter);
4443 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4444 * @adapter: board private structure to initialize
4446 * ixgbe_sw_init initializes the Adapter private data structure.
4447 * Fields are initialized based on PCI device information and
4448 * OS network device settings (MTU size).
4450 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
4452 struct ixgbe_hw *hw = &adapter->hw;
4453 struct pci_dev *pdev = adapter->pdev;
4455 #ifdef CONFIG_IXGBE_DCB
4457 struct tc_configuration *tc;
4460 /* PCI config space info */
4462 hw->vendor_id = pdev->vendor;
4463 hw->device_id = pdev->device;
4464 hw->revision_id = pdev->revision;
4465 hw->subsystem_vendor_id = pdev->subsystem_vendor;
4466 hw->subsystem_device_id = pdev->subsystem_device;
4468 /* Set capability flags */
4469 rss = min_t(int, IXGBE_MAX_RSS_INDICES, num_online_cpus());
4470 adapter->ring_feature[RING_F_RSS].limit = rss;
4471 switch (hw->mac.type) {
4472 case ixgbe_mac_82598EB:
4473 if (hw->device_id == IXGBE_DEV_ID_82598AT)
4474 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4475 adapter->max_q_vectors = MAX_Q_VECTORS_82598;
4477 case ixgbe_mac_X540:
4478 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4479 case ixgbe_mac_82599EB:
4480 adapter->max_q_vectors = MAX_Q_VECTORS_82599;
4481 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4482 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4483 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
4484 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4485 /* Flow Director hash filters enabled */
4486 adapter->atr_sample_rate = 20;
4487 adapter->ring_feature[RING_F_FDIR].limit =
4488 IXGBE_MAX_FDIR_INDICES;
4489 adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
4491 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4492 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4493 #ifdef CONFIG_IXGBE_DCB
4494 /* Default traffic class to use for FCoE */
4495 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
4497 #endif /* IXGBE_FCOE */
4504 /* FCoE support exists, always init the FCoE lock */
4505 spin_lock_init(&adapter->fcoe.lock);
4508 /* n-tuple support exists, always init our spinlock */
4509 spin_lock_init(&adapter->fdir_perfect_lock);
4511 #ifdef CONFIG_IXGBE_DCB
4512 switch (hw->mac.type) {
4513 case ixgbe_mac_X540:
4514 adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
4515 adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
4518 adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
4519 adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
4523 /* Configure DCB traffic classes */
4524 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
4525 tc = &adapter->dcb_cfg.tc_config[j];
4526 tc->path[DCB_TX_CONFIG].bwg_id = 0;
4527 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
4528 tc->path[DCB_RX_CONFIG].bwg_id = 0;
4529 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
4530 tc->dcb_pfc = pfc_disabled;
4533 /* Initialize default user to priority mapping, UPx->TC0 */
4534 tc = &adapter->dcb_cfg.tc_config[0];
4535 tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
4536 tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
4538 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
4539 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
4540 adapter->dcb_cfg.pfc_mode_enable = false;
4541 adapter->dcb_set_bitmap = 0x00;
4542 adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
4543 memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg,
4544 sizeof(adapter->temp_dcb_cfg));
4548 /* default flow control settings */
4549 hw->fc.requested_mode = ixgbe_fc_full;
4550 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
4551 ixgbe_pbthresh_setup(adapter);
4552 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
4553 hw->fc.send_xon = true;
4554 hw->fc.disable_fc_autoneg = false;
4556 #ifdef CONFIG_PCI_IOV
4557 /* assign number of SR-IOV VFs */
4558 if (hw->mac.type != ixgbe_mac_82598EB)
4559 adapter->num_vfs = (max_vfs > 63) ? 0 : max_vfs;
4562 /* enable itr by default in dynamic mode */
4563 adapter->rx_itr_setting = 1;
4564 adapter->tx_itr_setting = 1;
4566 /* set default ring sizes */
4567 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
4568 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
4570 /* set default work limits */
4571 adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
4573 /* initialize eeprom parameters */
4574 if (ixgbe_init_eeprom_params_generic(hw)) {
4575 e_dev_err("EEPROM initialization failed\n");
4579 set_bit(__IXGBE_DOWN, &adapter->state);
4585 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
4586 * @tx_ring: tx descriptor ring (for a specific queue) to setup
4588 * Return 0 on success, negative on failure
4590 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
4592 struct device *dev = tx_ring->dev;
4593 int orig_node = dev_to_node(dev);
4597 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4599 if (tx_ring->q_vector)
4600 numa_node = tx_ring->q_vector->numa_node;
4602 tx_ring->tx_buffer_info = vzalloc_node(size, numa_node);
4603 if (!tx_ring->tx_buffer_info)
4604 tx_ring->tx_buffer_info = vzalloc(size);
4605 if (!tx_ring->tx_buffer_info)
4608 /* round up to nearest 4K */
4609 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
4610 tx_ring->size = ALIGN(tx_ring->size, 4096);
4612 set_dev_node(dev, numa_node);
4613 tx_ring->desc = dma_alloc_coherent(dev,
4617 set_dev_node(dev, orig_node);
4619 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
4620 &tx_ring->dma, GFP_KERNEL);
4624 tx_ring->next_to_use = 0;
4625 tx_ring->next_to_clean = 0;
4629 vfree(tx_ring->tx_buffer_info);
4630 tx_ring->tx_buffer_info = NULL;
4631 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
4636 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
4637 * @adapter: board private structure
4639 * If this function returns with an error, then it's possible one or
4640 * more of the rings is populated (while the rest are not). It is the
4641 * callers duty to clean those orphaned rings.
4643 * Return 0 on success, negative on failure
4645 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
4649 for (i = 0; i < adapter->num_tx_queues; i++) {
4650 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
4654 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
4660 /* rewind the index freeing the rings as we go */
4662 ixgbe_free_tx_resources(adapter->tx_ring[i]);
4667 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
4668 * @rx_ring: rx descriptor ring (for a specific queue) to setup
4670 * Returns 0 on success, negative on failure
4672 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
4674 struct device *dev = rx_ring->dev;
4675 int orig_node = dev_to_node(dev);
4679 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4681 if (rx_ring->q_vector)
4682 numa_node = rx_ring->q_vector->numa_node;
4684 rx_ring->rx_buffer_info = vzalloc_node(size, numa_node);
4685 if (!rx_ring->rx_buffer_info)
4686 rx_ring->rx_buffer_info = vzalloc(size);
4687 if (!rx_ring->rx_buffer_info)
4690 /* Round up to nearest 4K */
4691 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
4692 rx_ring->size = ALIGN(rx_ring->size, 4096);
4694 set_dev_node(dev, numa_node);
4695 rx_ring->desc = dma_alloc_coherent(dev,
4699 set_dev_node(dev, orig_node);
4701 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
4702 &rx_ring->dma, GFP_KERNEL);
4706 rx_ring->next_to_clean = 0;
4707 rx_ring->next_to_use = 0;
4711 vfree(rx_ring->rx_buffer_info);
4712 rx_ring->rx_buffer_info = NULL;
4713 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
4718 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
4719 * @adapter: board private structure
4721 * If this function returns with an error, then it's possible one or
4722 * more of the rings is populated (while the rest are not). It is the
4723 * callers duty to clean those orphaned rings.
4725 * Return 0 on success, negative on failure
4727 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
4731 for (i = 0; i < adapter->num_rx_queues; i++) {
4732 err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
4736 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
4741 err = ixgbe_setup_fcoe_ddp_resources(adapter);
4746 /* rewind the index freeing the rings as we go */
4748 ixgbe_free_rx_resources(adapter->rx_ring[i]);
4753 * ixgbe_free_tx_resources - Free Tx Resources per Queue
4754 * @tx_ring: Tx descriptor ring for a specific queue
4756 * Free all transmit software resources
4758 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
4760 ixgbe_clean_tx_ring(tx_ring);
4762 vfree(tx_ring->tx_buffer_info);
4763 tx_ring->tx_buffer_info = NULL;
4765 /* if not set, then don't free */
4769 dma_free_coherent(tx_ring->dev, tx_ring->size,
4770 tx_ring->desc, tx_ring->dma);
4772 tx_ring->desc = NULL;
4776 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
4777 * @adapter: board private structure
4779 * Free all transmit software resources
4781 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
4785 for (i = 0; i < adapter->num_tx_queues; i++)
4786 if (adapter->tx_ring[i]->desc)
4787 ixgbe_free_tx_resources(adapter->tx_ring[i]);
4791 * ixgbe_free_rx_resources - Free Rx Resources
4792 * @rx_ring: ring to clean the resources from
4794 * Free all receive software resources
4796 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
4798 ixgbe_clean_rx_ring(rx_ring);
4800 vfree(rx_ring->rx_buffer_info);
4801 rx_ring->rx_buffer_info = NULL;
4803 /* if not set, then don't free */
4807 dma_free_coherent(rx_ring->dev, rx_ring->size,
4808 rx_ring->desc, rx_ring->dma);
4810 rx_ring->desc = NULL;
4814 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
4815 * @adapter: board private structure
4817 * Free all receive software resources
4819 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
4824 ixgbe_free_fcoe_ddp_resources(adapter);
4827 for (i = 0; i < adapter->num_rx_queues; i++)
4828 if (adapter->rx_ring[i]->desc)
4829 ixgbe_free_rx_resources(adapter->rx_ring[i]);
4833 * ixgbe_change_mtu - Change the Maximum Transfer Unit
4834 * @netdev: network interface device structure
4835 * @new_mtu: new value for maximum frame size
4837 * Returns 0 on success, negative on failure
4839 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
4841 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4842 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4844 /* MTU < 68 is an error and causes problems on some kernels */
4845 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
4849 * For 82599EB we cannot allow legacy VFs to enable their receive
4850 * paths when MTU greater than 1500 is configured. So display a
4851 * warning that legacy VFs will be disabled.
4853 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
4854 (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
4855 (max_frame > MAXIMUM_ETHERNET_VLAN_SIZE))
4856 e_warn(probe, "Setting MTU > 1500 will disable legacy VFs\n");
4858 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4860 /* must set new MTU before calling down or up */
4861 netdev->mtu = new_mtu;
4863 if (netif_running(netdev))
4864 ixgbe_reinit_locked(adapter);
4870 * ixgbe_open - Called when a network interface is made active
4871 * @netdev: network interface device structure
4873 * Returns 0 on success, negative value on failure
4875 * The open entry point is called when a network interface is made
4876 * active by the system (IFF_UP). At this point all resources needed
4877 * for transmit and receive operations are allocated, the interrupt
4878 * handler is registered with the OS, the watchdog timer is started,
4879 * and the stack is notified that the interface is ready.
4881 static int ixgbe_open(struct net_device *netdev)
4883 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4886 /* disallow open during test */
4887 if (test_bit(__IXGBE_TESTING, &adapter->state))
4890 netif_carrier_off(netdev);
4892 /* allocate transmit descriptors */
4893 err = ixgbe_setup_all_tx_resources(adapter);
4897 /* allocate receive descriptors */
4898 err = ixgbe_setup_all_rx_resources(adapter);
4902 ixgbe_configure(adapter);
4904 err = ixgbe_request_irq(adapter);
4908 /* Notify the stack of the actual queue counts. */
4909 err = netif_set_real_num_tx_queues(netdev,
4910 adapter->num_rx_pools > 1 ? 1 :
4911 adapter->num_tx_queues);
4913 goto err_set_queues;
4916 err = netif_set_real_num_rx_queues(netdev,
4917 adapter->num_rx_pools > 1 ? 1 :
4918 adapter->num_rx_queues);
4920 goto err_set_queues;
4922 ixgbe_up_complete(adapter);
4927 ixgbe_free_irq(adapter);
4929 ixgbe_free_all_rx_resources(adapter);
4931 ixgbe_free_all_tx_resources(adapter);
4933 ixgbe_reset(adapter);
4939 * ixgbe_close - Disables a network interface
4940 * @netdev: network interface device structure
4942 * Returns 0, this is not allowed to fail
4944 * The close entry point is called when an interface is de-activated
4945 * by the OS. The hardware is still under the drivers control, but
4946 * needs to be disabled. A global MAC reset is issued to stop the
4947 * hardware, and all transmit and receive resources are freed.
4949 static int ixgbe_close(struct net_device *netdev)
4951 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4953 ixgbe_down(adapter);
4954 ixgbe_free_irq(adapter);
4956 ixgbe_fdir_filter_exit(adapter);
4958 ixgbe_free_all_tx_resources(adapter);
4959 ixgbe_free_all_rx_resources(adapter);
4961 ixgbe_release_hw_control(adapter);
4967 static int ixgbe_resume(struct pci_dev *pdev)
4969 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
4970 struct net_device *netdev = adapter->netdev;
4973 pci_set_power_state(pdev, PCI_D0);
4974 pci_restore_state(pdev);
4976 * pci_restore_state clears dev->state_saved so call
4977 * pci_save_state to restore it.
4979 pci_save_state(pdev);
4981 err = pci_enable_device_mem(pdev);
4983 e_dev_err("Cannot enable PCI device from suspend\n");
4986 pci_set_master(pdev);
4988 pci_wake_from_d3(pdev, false);
4990 ixgbe_reset(adapter);
4992 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
4995 err = ixgbe_init_interrupt_scheme(adapter);
4996 if (!err && netif_running(netdev))
4997 err = ixgbe_open(netdev);
5004 netif_device_attach(netdev);
5008 #endif /* CONFIG_PM */
5010 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5012 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5013 struct net_device *netdev = adapter->netdev;
5014 struct ixgbe_hw *hw = &adapter->hw;
5016 u32 wufc = adapter->wol;
5021 netif_device_detach(netdev);
5023 if (netif_running(netdev)) {
5025 ixgbe_down(adapter);
5026 ixgbe_free_irq(adapter);
5027 ixgbe_free_all_tx_resources(adapter);
5028 ixgbe_free_all_rx_resources(adapter);
5032 ixgbe_clear_interrupt_scheme(adapter);
5035 retval = pci_save_state(pdev);
5041 ixgbe_set_rx_mode(netdev);
5044 * enable the optics for both mult-speed fiber and
5045 * 82599 SFP+ fiber as we can WoL.
5047 if (hw->mac.ops.enable_tx_laser &&
5048 (hw->phy.multispeed_fiber ||
5049 (hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber &&
5050 hw->mac.type == ixgbe_mac_82599EB)))
5051 hw->mac.ops.enable_tx_laser(hw);
5053 /* turn on all-multi mode if wake on multicast is enabled */
5054 if (wufc & IXGBE_WUFC_MC) {
5055 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5056 fctrl |= IXGBE_FCTRL_MPE;
5057 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5060 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5061 ctrl |= IXGBE_CTRL_GIO_DIS;
5062 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5064 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5066 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5067 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5070 switch (hw->mac.type) {
5071 case ixgbe_mac_82598EB:
5072 pci_wake_from_d3(pdev, false);
5074 case ixgbe_mac_82599EB:
5075 case ixgbe_mac_X540:
5076 pci_wake_from_d3(pdev, !!wufc);
5082 *enable_wake = !!wufc;
5084 ixgbe_release_hw_control(adapter);
5086 pci_disable_device(pdev);
5092 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5097 retval = __ixgbe_shutdown(pdev, &wake);
5102 pci_prepare_to_sleep(pdev);
5104 pci_wake_from_d3(pdev, false);
5105 pci_set_power_state(pdev, PCI_D3hot);
5110 #endif /* CONFIG_PM */
5112 static void ixgbe_shutdown(struct pci_dev *pdev)
5116 __ixgbe_shutdown(pdev, &wake);
5118 if (system_state == SYSTEM_POWER_OFF) {
5119 pci_wake_from_d3(pdev, wake);
5120 pci_set_power_state(pdev, PCI_D3hot);
5125 * ixgbe_update_stats - Update the board statistics counters.
5126 * @adapter: board private structure
5128 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5130 struct net_device *netdev = adapter->netdev;
5131 struct ixgbe_hw *hw = &adapter->hw;
5132 struct ixgbe_hw_stats *hwstats = &adapter->stats;
5134 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5135 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
5136 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
5137 u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
5139 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5140 test_bit(__IXGBE_RESETTING, &adapter->state))
5143 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5146 for (i = 0; i < adapter->num_rx_queues; i++) {
5147 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
5148 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
5150 adapter->rsc_total_count = rsc_count;
5151 adapter->rsc_total_flush = rsc_flush;
5154 for (i = 0; i < adapter->num_rx_queues; i++) {
5155 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
5156 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
5157 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
5158 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
5159 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
5160 bytes += rx_ring->stats.bytes;
5161 packets += rx_ring->stats.packets;
5163 adapter->non_eop_descs = non_eop_descs;
5164 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5165 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5166 adapter->hw_csum_rx_error = hw_csum_rx_error;
5167 netdev->stats.rx_bytes = bytes;
5168 netdev->stats.rx_packets = packets;
5172 /* gather some stats to the adapter struct that are per queue */
5173 for (i = 0; i < adapter->num_tx_queues; i++) {
5174 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5175 restart_queue += tx_ring->tx_stats.restart_queue;
5176 tx_busy += tx_ring->tx_stats.tx_busy;
5177 bytes += tx_ring->stats.bytes;
5178 packets += tx_ring->stats.packets;
5180 adapter->restart_queue = restart_queue;
5181 adapter->tx_busy = tx_busy;
5182 netdev->stats.tx_bytes = bytes;
5183 netdev->stats.tx_packets = packets;
5185 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5187 /* 8 register reads */
5188 for (i = 0; i < 8; i++) {
5189 /* for packet buffers not used, the register should read 0 */
5190 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5192 hwstats->mpc[i] += mpc;
5193 total_mpc += hwstats->mpc[i];
5194 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5195 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5196 switch (hw->mac.type) {
5197 case ixgbe_mac_82598EB:
5198 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5199 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5200 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5201 hwstats->pxonrxc[i] +=
5202 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5204 case ixgbe_mac_82599EB:
5205 case ixgbe_mac_X540:
5206 hwstats->pxonrxc[i] +=
5207 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5214 /*16 register reads */
5215 for (i = 0; i < 16; i++) {
5216 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5217 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5218 if ((hw->mac.type == ixgbe_mac_82599EB) ||
5219 (hw->mac.type == ixgbe_mac_X540)) {
5220 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
5221 IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
5222 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
5223 IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
5227 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5228 /* work around hardware counting issue */
5229 hwstats->gprc -= missed_rx;
5231 ixgbe_update_xoff_received(adapter);
5233 /* 82598 hardware only has a 32 bit counter in the high register */
5234 switch (hw->mac.type) {
5235 case ixgbe_mac_82598EB:
5236 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5237 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5238 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5239 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5241 case ixgbe_mac_X540:
5242 /* OS2BMC stats are X540 only*/
5243 hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
5244 hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
5245 hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
5246 hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
5247 case ixgbe_mac_82599EB:
5248 for (i = 0; i < 16; i++)
5249 adapter->hw_rx_no_dma_resources +=
5250 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5251 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5252 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
5253 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5254 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
5255 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5256 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5257 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5258 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5259 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5261 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5262 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5263 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5264 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5265 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5266 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5267 /* Add up per cpu counters for total ddp aloc fail */
5268 if (adapter->fcoe.ddp_pool) {
5269 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
5270 struct ixgbe_fcoe_ddp_pool *ddp_pool;
5272 u64 noddp = 0, noddp_ext_buff = 0;
5273 for_each_possible_cpu(cpu) {
5274 ddp_pool = per_cpu_ptr(fcoe->ddp_pool, cpu);
5275 noddp += ddp_pool->noddp;
5276 noddp_ext_buff += ddp_pool->noddp_ext_buff;
5278 hwstats->fcoe_noddp = noddp;
5279 hwstats->fcoe_noddp_ext_buff = noddp_ext_buff;
5281 #endif /* IXGBE_FCOE */
5286 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5287 hwstats->bprc += bprc;
5288 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5289 if (hw->mac.type == ixgbe_mac_82598EB)
5290 hwstats->mprc -= bprc;
5291 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5292 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5293 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5294 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5295 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5296 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5297 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5298 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5299 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5300 hwstats->lxontxc += lxon;
5301 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5302 hwstats->lxofftxc += lxoff;
5303 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5304 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5306 * 82598 errata - tx of flow control packets is included in tx counters
5308 xon_off_tot = lxon + lxoff;
5309 hwstats->gptc -= xon_off_tot;
5310 hwstats->mptc -= xon_off_tot;
5311 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5312 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5313 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5314 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5315 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5316 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5317 hwstats->ptc64 -= xon_off_tot;
5318 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5319 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5320 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5321 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5322 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5323 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5325 /* Fill out the OS statistics structure */
5326 netdev->stats.multicast = hwstats->mprc;
5329 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
5330 netdev->stats.rx_dropped = 0;
5331 netdev->stats.rx_length_errors = hwstats->rlec;
5332 netdev->stats.rx_crc_errors = hwstats->crcerrs;
5333 netdev->stats.rx_missed_errors = total_mpc;
5337 * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
5338 * @adapter: pointer to the device adapter structure
5340 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
5342 struct ixgbe_hw *hw = &adapter->hw;
5345 if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
5348 adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
5350 /* if interface is down do nothing */
5351 if (test_bit(__IXGBE_DOWN, &adapter->state))
5354 /* do nothing if we are not using signature filters */
5355 if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
5358 adapter->fdir_overflow++;
5360 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5361 for (i = 0; i < adapter->num_tx_queues; i++)
5362 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
5363 &(adapter->tx_ring[i]->state));
5364 /* re-enable flow director interrupts */
5365 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
5367 e_err(probe, "failed to finish FDIR re-initialization, "
5368 "ignored adding FDIR ATR filters\n");
5373 * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
5374 * @adapter: pointer to the device adapter structure
5376 * This function serves two purposes. First it strobes the interrupt lines
5377 * in order to make certain interrupts are occurring. Secondly it sets the
5378 * bits needed to check for TX hangs. As a result we should immediately
5379 * determine if a hang has occurred.
5381 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
5383 struct ixgbe_hw *hw = &adapter->hw;
5387 /* If we're down or resetting, just bail */
5388 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5389 test_bit(__IXGBE_RESETTING, &adapter->state))
5392 /* Force detection of hung controller */
5393 if (netif_carrier_ok(adapter->netdev)) {
5394 for (i = 0; i < adapter->num_tx_queues; i++)
5395 set_check_for_tx_hang(adapter->tx_ring[i]);
5398 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5400 * for legacy and MSI interrupts don't set any bits
5401 * that are enabled for EIAM, because this operation
5402 * would set *both* EIMS and EICS for any bit in EIAM
5404 IXGBE_WRITE_REG(hw, IXGBE_EICS,
5405 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5407 /* get one bit for every active tx/rx interrupt vector */
5408 for (i = 0; i < adapter->num_q_vectors; i++) {
5409 struct ixgbe_q_vector *qv = adapter->q_vector[i];
5410 if (qv->rx.ring || qv->tx.ring)
5411 eics |= ((u64)1 << i);
5415 /* Cause software interrupt to ensure rings are cleaned */
5416 ixgbe_irq_rearm_queues(adapter, eics);
5421 * ixgbe_watchdog_update_link - update the link status
5422 * @adapter: pointer to the device adapter structure
5423 * @link_speed: pointer to a u32 to store the link_speed
5425 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
5427 struct ixgbe_hw *hw = &adapter->hw;
5428 u32 link_speed = adapter->link_speed;
5429 bool link_up = adapter->link_up;
5430 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
5432 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
5435 if (hw->mac.ops.check_link) {
5436 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5438 /* always assume link is up, if no check link function */
5439 link_speed = IXGBE_LINK_SPEED_10GB_FULL;
5443 if (adapter->ixgbe_ieee_pfc)
5444 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
5446 if (link_up && !((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && pfc_en)) {
5447 hw->mac.ops.fc_enable(hw);
5448 ixgbe_set_rx_drop_en(adapter);
5452 time_after(jiffies, (adapter->link_check_timeout +
5453 IXGBE_TRY_LINK_TIMEOUT))) {
5454 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5455 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5456 IXGBE_WRITE_FLUSH(hw);
5459 adapter->link_up = link_up;
5460 adapter->link_speed = link_speed;
5463 static void ixgbe_update_default_up(struct ixgbe_adapter *adapter)
5465 #ifdef CONFIG_IXGBE_DCB
5466 struct net_device *netdev = adapter->netdev;
5467 struct dcb_app app = {
5468 .selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE,
5473 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_IEEE)
5474 up = dcb_ieee_getapp_mask(netdev, &app);
5476 adapter->default_up = (up > 1) ? (ffs(up) - 1) : 0;
5481 * ixgbe_watchdog_link_is_up - update netif_carrier status and
5482 * print link up message
5483 * @adapter: pointer to the device adapter structure
5485 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
5487 struct net_device *netdev = adapter->netdev;
5488 struct ixgbe_hw *hw = &adapter->hw;
5489 u32 link_speed = adapter->link_speed;
5490 bool flow_rx, flow_tx;
5492 /* only continue if link was previously down */
5493 if (netif_carrier_ok(netdev))
5496 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
5498 switch (hw->mac.type) {
5499 case ixgbe_mac_82598EB: {
5500 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5501 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5502 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5503 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5506 case ixgbe_mac_X540:
5507 case ixgbe_mac_82599EB: {
5508 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5509 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5510 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5511 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5520 #ifdef CONFIG_IXGBE_PTP
5521 ixgbe_ptp_start_cyclecounter(adapter);
5524 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
5525 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5527 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5529 (link_speed == IXGBE_LINK_SPEED_100_FULL ?
5532 ((flow_rx && flow_tx) ? "RX/TX" :
5534 (flow_tx ? "TX" : "None"))));
5536 netif_carrier_on(netdev);
5537 ixgbe_check_vf_rate_limit(adapter);
5539 /* update the default user priority for VFs */
5540 ixgbe_update_default_up(adapter);
5542 /* ping all the active vfs to let them know link has changed */
5543 ixgbe_ping_all_vfs(adapter);
5547 * ixgbe_watchdog_link_is_down - update netif_carrier status and
5548 * print link down message
5549 * @adapter: pointer to the adapter structure
5551 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
5553 struct net_device *netdev = adapter->netdev;
5554 struct ixgbe_hw *hw = &adapter->hw;
5556 adapter->link_up = false;
5557 adapter->link_speed = 0;
5559 /* only continue if link was up previously */
5560 if (!netif_carrier_ok(netdev))
5563 /* poll for SFP+ cable when link is down */
5564 if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
5565 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5567 #ifdef CONFIG_IXGBE_PTP
5568 ixgbe_ptp_start_cyclecounter(adapter);
5571 e_info(drv, "NIC Link is Down\n");
5572 netif_carrier_off(netdev);
5574 /* ping all the active vfs to let them know link has changed */
5575 ixgbe_ping_all_vfs(adapter);
5579 * ixgbe_watchdog_flush_tx - flush queues on link down
5580 * @adapter: pointer to the device adapter structure
5582 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
5585 int some_tx_pending = 0;
5587 if (!netif_carrier_ok(adapter->netdev)) {
5588 for (i = 0; i < adapter->num_tx_queues; i++) {
5589 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5590 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5591 some_tx_pending = 1;
5596 if (some_tx_pending) {
5597 /* We've lost link, so the controller stops DMA,
5598 * but we've got queued Tx work that's never going
5599 * to get done, so reset controller to flush Tx.
5600 * (Do the reset outside of interrupt context).
5602 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
5607 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
5611 /* Do not perform spoof check for 82598 or if not in IOV mode */
5612 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
5613 adapter->num_vfs == 0)
5616 ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
5619 * ssvpc register is cleared on read, if zero then no
5620 * spoofed packets in the last interval.
5625 e_warn(drv, "%u Spoofed packets detected\n", ssvpc);
5629 * ixgbe_watchdog_subtask - check and bring link up
5630 * @adapter: pointer to the device adapter structure
5632 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
5634 /* if interface is down do nothing */
5635 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5636 test_bit(__IXGBE_RESETTING, &adapter->state))
5639 ixgbe_watchdog_update_link(adapter);
5641 if (adapter->link_up)
5642 ixgbe_watchdog_link_is_up(adapter);
5644 ixgbe_watchdog_link_is_down(adapter);
5646 ixgbe_spoof_check(adapter);
5647 ixgbe_update_stats(adapter);
5649 ixgbe_watchdog_flush_tx(adapter);
5653 * ixgbe_sfp_detection_subtask - poll for SFP+ cable
5654 * @adapter: the ixgbe adapter structure
5656 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
5658 struct ixgbe_hw *hw = &adapter->hw;
5661 /* not searching for SFP so there is nothing to do here */
5662 if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
5663 !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
5666 /* someone else is in init, wait until next service event */
5667 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5670 err = hw->phy.ops.identify_sfp(hw);
5671 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
5674 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
5675 /* If no cable is present, then we need to reset
5676 * the next time we find a good cable. */
5677 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
5684 /* exit if reset not needed */
5685 if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
5688 adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
5691 * A module may be identified correctly, but the EEPROM may not have
5692 * support for that module. setup_sfp() will fail in that case, so
5693 * we should not allow that module to load.
5695 if (hw->mac.type == ixgbe_mac_82598EB)
5696 err = hw->phy.ops.reset(hw);
5698 err = hw->mac.ops.setup_sfp(hw);
5700 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
5703 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
5704 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
5707 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5709 if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
5710 (adapter->netdev->reg_state == NETREG_REGISTERED)) {
5711 e_dev_err("failed to initialize because an unsupported "
5712 "SFP+ module type was detected.\n");
5713 e_dev_err("Reload the driver after installing a "
5714 "supported module.\n");
5715 unregister_netdev(adapter->netdev);
5720 * ixgbe_sfp_link_config_subtask - set up link SFP after module install
5721 * @adapter: the ixgbe adapter structure
5723 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
5725 struct ixgbe_hw *hw = &adapter->hw;
5729 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
5732 /* someone else is in init, wait until next service event */
5733 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5736 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
5738 autoneg = hw->phy.autoneg_advertised;
5739 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
5740 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
5741 if (hw->mac.ops.setup_link)
5742 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
5744 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5745 adapter->link_check_timeout = jiffies;
5746 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5749 #ifdef CONFIG_PCI_IOV
5750 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
5753 struct ixgbe_hw *hw = &adapter->hw;
5754 struct net_device *netdev = adapter->netdev;
5758 gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
5759 if (gpc) /* If incrementing then no need for the check below */
5762 * Check to see if a bad DMA write target from an errant or
5763 * malicious VF has caused a PCIe error. If so then we can
5764 * issue a VFLR to the offending VF(s) and then resume without
5765 * requesting a full slot reset.
5768 for (vf = 0; vf < adapter->num_vfs; vf++) {
5769 ciaa = (vf << 16) | 0x80000000;
5770 /* 32 bit read so align, we really want status at offset 6 */
5771 ciaa |= PCI_COMMAND;
5772 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5773 ciad = IXGBE_READ_REG(hw, IXGBE_CIAD_82599);
5775 /* disable debug mode asap after reading data */
5776 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5777 /* Get the upper 16 bits which will be the PCI status reg */
5779 if (ciad & PCI_STATUS_REC_MASTER_ABORT) {
5780 netdev_err(netdev, "VF %d Hung DMA\n", vf);
5782 ciaa = (vf << 16) | 0x80000000;
5784 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5785 ciad = 0x00008000; /* VFLR */
5786 IXGBE_WRITE_REG(hw, IXGBE_CIAD_82599, ciad);
5788 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5795 * ixgbe_service_timer - Timer Call-back
5796 * @data: pointer to adapter cast into an unsigned long
5798 static void ixgbe_service_timer(unsigned long data)
5800 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5801 unsigned long next_event_offset;
5804 /* poll faster when waiting for link */
5805 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
5806 next_event_offset = HZ / 10;
5808 next_event_offset = HZ * 2;
5810 #ifdef CONFIG_PCI_IOV
5812 * don't bother with SR-IOV VF DMA hang check if there are
5813 * no VFs or the link is down
5815 if (!adapter->num_vfs ||
5816 (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
5817 goto normal_timer_service;
5819 /* If we have VFs allocated then we must check for DMA hangs */
5820 ixgbe_check_for_bad_vf(adapter);
5821 next_event_offset = HZ / 50;
5822 adapter->timer_event_accumulator++;
5824 if (adapter->timer_event_accumulator >= 100)
5825 adapter->timer_event_accumulator = 0;
5829 normal_timer_service:
5831 /* Reset the timer */
5832 mod_timer(&adapter->service_timer, next_event_offset + jiffies);
5835 ixgbe_service_event_schedule(adapter);
5838 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
5840 if (!(adapter->flags2 & IXGBE_FLAG2_RESET_REQUESTED))
5843 adapter->flags2 &= ~IXGBE_FLAG2_RESET_REQUESTED;
5845 /* If we're already down or resetting, just bail */
5846 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5847 test_bit(__IXGBE_RESETTING, &adapter->state))
5850 ixgbe_dump(adapter);
5851 netdev_err(adapter->netdev, "Reset adapter\n");
5852 adapter->tx_timeout_count++;
5854 ixgbe_reinit_locked(adapter);
5858 * ixgbe_service_task - manages and runs subtasks
5859 * @work: pointer to work_struct containing our data
5861 static void ixgbe_service_task(struct work_struct *work)
5863 struct ixgbe_adapter *adapter = container_of(work,
5864 struct ixgbe_adapter,
5867 ixgbe_reset_subtask(adapter);
5868 ixgbe_sfp_detection_subtask(adapter);
5869 ixgbe_sfp_link_config_subtask(adapter);
5870 ixgbe_check_overtemp_subtask(adapter);
5871 ixgbe_watchdog_subtask(adapter);
5872 ixgbe_fdir_reinit_subtask(adapter);
5873 ixgbe_check_hang_subtask(adapter);
5874 #ifdef CONFIG_IXGBE_PTP
5875 ixgbe_ptp_overflow_check(adapter);
5878 ixgbe_service_event_complete(adapter);
5881 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
5882 struct ixgbe_tx_buffer *first,
5885 struct sk_buff *skb = first->skb;
5886 u32 vlan_macip_lens, type_tucmd;
5887 u32 mss_l4len_idx, l4len;
5889 if (!skb_is_gso(skb))
5892 if (skb_header_cloned(skb)) {
5893 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5898 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
5899 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
5901 if (first->protocol == __constant_htons(ETH_P_IP)) {
5902 struct iphdr *iph = ip_hdr(skb);
5905 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
5909 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
5910 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
5911 IXGBE_TX_FLAGS_CSUM |
5912 IXGBE_TX_FLAGS_IPV4;
5913 } else if (skb_is_gso_v6(skb)) {
5914 ipv6_hdr(skb)->payload_len = 0;
5915 tcp_hdr(skb)->check =
5916 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5917 &ipv6_hdr(skb)->daddr,
5919 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
5920 IXGBE_TX_FLAGS_CSUM;
5923 /* compute header lengths */
5924 l4len = tcp_hdrlen(skb);
5925 *hdr_len = skb_transport_offset(skb) + l4len;
5927 /* update gso size and bytecount with header size */
5928 first->gso_segs = skb_shinfo(skb)->gso_segs;
5929 first->bytecount += (first->gso_segs - 1) * *hdr_len;
5931 /* mss_l4len_id: use 1 as index for TSO */
5932 mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
5933 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
5934 mss_l4len_idx |= 1 << IXGBE_ADVTXD_IDX_SHIFT;
5936 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
5937 vlan_macip_lens = skb_network_header_len(skb);
5938 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
5939 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
5941 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0, type_tucmd,
5947 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
5948 struct ixgbe_tx_buffer *first)
5950 struct sk_buff *skb = first->skb;
5951 u32 vlan_macip_lens = 0;
5952 u32 mss_l4len_idx = 0;
5955 if (skb->ip_summed != CHECKSUM_PARTIAL) {
5956 if (!(first->tx_flags & IXGBE_TX_FLAGS_HW_VLAN)) {
5957 if (unlikely(skb->no_fcs))
5958 first->tx_flags |= IXGBE_TX_FLAGS_NO_IFCS;
5959 if (!(first->tx_flags & IXGBE_TX_FLAGS_TXSW))
5964 switch (first->protocol) {
5965 case __constant_htons(ETH_P_IP):
5966 vlan_macip_lens |= skb_network_header_len(skb);
5967 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
5968 l4_hdr = ip_hdr(skb)->protocol;
5970 case __constant_htons(ETH_P_IPV6):
5971 vlan_macip_lens |= skb_network_header_len(skb);
5972 l4_hdr = ipv6_hdr(skb)->nexthdr;
5975 if (unlikely(net_ratelimit())) {
5976 dev_warn(tx_ring->dev,
5977 "partial checksum but proto=%x!\n",
5985 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5986 mss_l4len_idx = tcp_hdrlen(skb) <<
5987 IXGBE_ADVTXD_L4LEN_SHIFT;
5990 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5991 mss_l4len_idx = sizeof(struct sctphdr) <<
5992 IXGBE_ADVTXD_L4LEN_SHIFT;
5995 mss_l4len_idx = sizeof(struct udphdr) <<
5996 IXGBE_ADVTXD_L4LEN_SHIFT;
5999 if (unlikely(net_ratelimit())) {
6000 dev_warn(tx_ring->dev,
6001 "partial checksum but l4 proto=%x!\n",
6007 /* update TX checksum flag */
6008 first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
6011 /* vlan_macip_lens: MACLEN, VLAN tag */
6012 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
6013 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
6015 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0,
6016 type_tucmd, mss_l4len_idx);
6019 static __le32 ixgbe_tx_cmd_type(u32 tx_flags)
6021 /* set type for advanced descriptor with frame checksum insertion */
6022 __le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
6023 IXGBE_ADVTXD_DCMD_DEXT);
6025 /* set HW vlan bit if vlan is present */
6026 if (tx_flags & IXGBE_TX_FLAGS_HW_VLAN)
6027 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
6029 #ifdef CONFIG_IXGBE_PTP
6030 if (tx_flags & IXGBE_TX_FLAGS_TSTAMP)
6031 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_MAC_TSTAMP);
6034 /* set segmentation enable bits for TSO/FSO */
6036 if (tx_flags & (IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_FSO))
6038 if (tx_flags & IXGBE_TX_FLAGS_TSO)
6040 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
6042 /* insert frame checksum */
6043 if (!(tx_flags & IXGBE_TX_FLAGS_NO_IFCS))
6044 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_IFCS);
6049 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
6050 u32 tx_flags, unsigned int paylen)
6052 __le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
6054 /* enable L4 checksum for TSO and TX checksum offload */
6055 if (tx_flags & IXGBE_TX_FLAGS_CSUM)
6056 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
6058 /* enble IPv4 checksum for TSO */
6059 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
6060 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
6062 /* use index 1 context for TSO/FSO/FCOE */
6064 if (tx_flags & (IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_FCOE))
6066 if (tx_flags & IXGBE_TX_FLAGS_TSO)
6068 olinfo_status |= cpu_to_le32(1 << IXGBE_ADVTXD_IDX_SHIFT);
6071 * Check Context must be set if Tx switch is enabled, which it
6072 * always is for case where virtual functions are running
6075 if (tx_flags & (IXGBE_TX_FLAGS_TXSW | IXGBE_TX_FLAGS_FCOE))
6077 if (tx_flags & IXGBE_TX_FLAGS_TXSW)
6079 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
6081 tx_desc->read.olinfo_status = olinfo_status;
6084 #define IXGBE_TXD_CMD (IXGBE_TXD_CMD_EOP | \
6087 static void ixgbe_tx_map(struct ixgbe_ring *tx_ring,
6088 struct ixgbe_tx_buffer *first,
6092 struct sk_buff *skb = first->skb;
6093 struct ixgbe_tx_buffer *tx_buffer;
6094 union ixgbe_adv_tx_desc *tx_desc;
6095 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
6096 unsigned int data_len = skb->data_len;
6097 unsigned int size = skb_headlen(skb);
6098 unsigned int paylen = skb->len - hdr_len;
6099 u32 tx_flags = first->tx_flags;
6101 u16 i = tx_ring->next_to_use;
6103 tx_desc = IXGBE_TX_DESC(tx_ring, i);
6105 ixgbe_tx_olinfo_status(tx_desc, tx_flags, paylen);
6106 cmd_type = ixgbe_tx_cmd_type(tx_flags);
6109 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6110 if (data_len < sizeof(struct fcoe_crc_eof)) {
6111 size -= sizeof(struct fcoe_crc_eof) - data_len;
6114 data_len -= sizeof(struct fcoe_crc_eof);
6119 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
6120 if (dma_mapping_error(tx_ring->dev, dma))
6123 /* record length, and DMA address */
6124 dma_unmap_len_set(first, len, size);
6125 dma_unmap_addr_set(first, dma, dma);
6127 tx_desc->read.buffer_addr = cpu_to_le64(dma);
6130 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
6131 tx_desc->read.cmd_type_len =
6132 cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
6136 if (i == tx_ring->count) {
6137 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
6141 dma += IXGBE_MAX_DATA_PER_TXD;
6142 size -= IXGBE_MAX_DATA_PER_TXD;
6144 tx_desc->read.buffer_addr = cpu_to_le64(dma);
6145 tx_desc->read.olinfo_status = 0;
6148 if (likely(!data_len))
6151 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
6155 if (i == tx_ring->count) {
6156 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
6161 size = min_t(unsigned int, data_len, skb_frag_size(frag));
6163 size = skb_frag_size(frag);
6167 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
6169 if (dma_mapping_error(tx_ring->dev, dma))
6172 tx_buffer = &tx_ring->tx_buffer_info[i];
6173 dma_unmap_len_set(tx_buffer, len, size);
6174 dma_unmap_addr_set(tx_buffer, dma, dma);
6176 tx_desc->read.buffer_addr = cpu_to_le64(dma);
6177 tx_desc->read.olinfo_status = 0;
6182 /* write last descriptor with RS and EOP bits */
6183 cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD);
6184 tx_desc->read.cmd_type_len = cmd_type;
6186 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
6188 /* set the timestamp */
6189 first->time_stamp = jiffies;
6192 * Force memory writes to complete before letting h/w know there
6193 * are new descriptors to fetch. (Only applicable for weak-ordered
6194 * memory model archs, such as IA-64).
6196 * We also need this memory barrier to make certain all of the
6197 * status bits have been updated before next_to_watch is written.
6201 /* set next_to_watch value indicating a packet is present */
6202 first->next_to_watch = tx_desc;
6205 if (i == tx_ring->count)
6208 tx_ring->next_to_use = i;
6210 /* notify HW of packet */
6211 writel(i, tx_ring->tail);
6215 dev_err(tx_ring->dev, "TX DMA map failed\n");
6217 /* clear dma mappings for failed tx_buffer_info map */
6219 tx_buffer = &tx_ring->tx_buffer_info[i];
6220 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer);
6221 if (tx_buffer == first)
6228 tx_ring->next_to_use = i;
6231 static void ixgbe_atr(struct ixgbe_ring *ring,
6232 struct ixgbe_tx_buffer *first)
6234 struct ixgbe_q_vector *q_vector = ring->q_vector;
6235 union ixgbe_atr_hash_dword input = { .dword = 0 };
6236 union ixgbe_atr_hash_dword common = { .dword = 0 };
6238 unsigned char *network;
6240 struct ipv6hdr *ipv6;
6245 /* if ring doesn't have a interrupt vector, cannot perform ATR */
6249 /* do nothing if sampling is disabled */
6250 if (!ring->atr_sample_rate)
6255 /* snag network header to get L4 type and address */
6256 hdr.network = skb_network_header(first->skb);
6258 /* Currently only IPv4/IPv6 with TCP is supported */
6259 if ((first->protocol != __constant_htons(ETH_P_IPV6) ||
6260 hdr.ipv6->nexthdr != IPPROTO_TCP) &&
6261 (first->protocol != __constant_htons(ETH_P_IP) ||
6262 hdr.ipv4->protocol != IPPROTO_TCP))
6265 th = tcp_hdr(first->skb);
6267 /* skip this packet since it is invalid or the socket is closing */
6271 /* sample on all syn packets or once every atr sample count */
6272 if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
6275 /* reset sample count */
6276 ring->atr_count = 0;
6278 vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
6281 * src and dst are inverted, think how the receiver sees them
6283 * The input is broken into two sections, a non-compressed section
6284 * containing vm_pool, vlan_id, and flow_type. The rest of the data
6285 * is XORed together and stored in the compressed dword.
6287 input.formatted.vlan_id = vlan_id;
6290 * since src port and flex bytes occupy the same word XOR them together
6291 * and write the value to source port portion of compressed dword
6293 if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
6294 common.port.src ^= th->dest ^ __constant_htons(ETH_P_8021Q);
6296 common.port.src ^= th->dest ^ first->protocol;
6297 common.port.dst ^= th->source;
6299 if (first->protocol == __constant_htons(ETH_P_IP)) {
6300 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
6301 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
6303 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
6304 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
6305 hdr.ipv6->saddr.s6_addr32[1] ^
6306 hdr.ipv6->saddr.s6_addr32[2] ^
6307 hdr.ipv6->saddr.s6_addr32[3] ^
6308 hdr.ipv6->daddr.s6_addr32[0] ^
6309 hdr.ipv6->daddr.s6_addr32[1] ^
6310 hdr.ipv6->daddr.s6_addr32[2] ^
6311 hdr.ipv6->daddr.s6_addr32[3];
6314 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
6315 ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
6316 input, common, ring->queue_index);
6319 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6321 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
6322 /* Herbert's original patch had:
6323 * smp_mb__after_netif_stop_queue();
6324 * but since that doesn't exist yet, just open code it. */
6327 /* We need to check again in a case another CPU has just
6328 * made room available. */
6329 if (likely(ixgbe_desc_unused(tx_ring) < size))
6332 /* A reprieve! - use start_queue because it doesn't call schedule */
6333 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
6334 ++tx_ring->tx_stats.restart_queue;
6338 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6340 if (likely(ixgbe_desc_unused(tx_ring) >= size))
6342 return __ixgbe_maybe_stop_tx(tx_ring, size);
6345 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6347 struct ixgbe_adapter *adapter = netdev_priv(dev);
6348 int txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
6351 __be16 protocol = vlan_get_protocol(skb);
6353 if (((protocol == htons(ETH_P_FCOE)) ||
6354 (protocol == htons(ETH_P_FIP))) &&
6355 (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)) {
6356 struct ixgbe_ring_feature *f;
6358 f = &adapter->ring_feature[RING_F_FCOE];
6360 while (txq >= f->indices)
6362 txq += adapter->ring_feature[RING_F_FCOE].offset;
6368 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6369 while (unlikely(txq >= dev->real_num_tx_queues))
6370 txq -= dev->real_num_tx_queues;
6374 return skb_tx_hash(dev, skb);
6377 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
6378 struct ixgbe_adapter *adapter,
6379 struct ixgbe_ring *tx_ring)
6381 struct ixgbe_tx_buffer *first;
6384 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6387 u16 count = TXD_USE_COUNT(skb_headlen(skb));
6388 __be16 protocol = skb->protocol;
6392 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
6393 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
6394 * + 2 desc gap to keep tail from touching head,
6395 * + 1 desc for context descriptor,
6396 * otherwise try next time
6398 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6399 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6400 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6402 count += skb_shinfo(skb)->nr_frags;
6404 if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
6405 tx_ring->tx_stats.tx_busy++;
6406 return NETDEV_TX_BUSY;
6409 /* record the location of the first descriptor for this packet */
6410 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
6412 first->bytecount = skb->len;
6413 first->gso_segs = 1;
6415 /* if we have a HW VLAN tag being added default to the HW one */
6416 if (vlan_tx_tag_present(skb)) {
6417 tx_flags |= vlan_tx_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
6418 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6419 /* else if it is a SW VLAN check the next protocol and store the tag */
6420 } else if (protocol == __constant_htons(ETH_P_8021Q)) {
6421 struct vlan_hdr *vhdr, _vhdr;
6422 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
6426 protocol = vhdr->h_vlan_encapsulated_proto;
6427 tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
6428 IXGBE_TX_FLAGS_VLAN_SHIFT;
6429 tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
6432 skb_tx_timestamp(skb);
6434 #ifdef CONFIG_IXGBE_PTP
6435 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
6436 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
6437 tx_flags |= IXGBE_TX_FLAGS_TSTAMP;
6441 #ifdef CONFIG_PCI_IOV
6443 * Use the l2switch_enable flag - would be false if the DMA
6444 * Tx switch had been disabled.
6446 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6447 tx_flags |= IXGBE_TX_FLAGS_TXSW;
6450 /* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
6451 if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
6452 ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
6453 (skb->priority != TC_PRIO_CONTROL))) {
6454 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6455 tx_flags |= (skb->priority & 0x7) <<
6456 IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
6457 if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
6458 struct vlan_ethhdr *vhdr;
6459 if (skb_header_cloned(skb) &&
6460 pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
6462 vhdr = (struct vlan_ethhdr *)skb->data;
6463 vhdr->h_vlan_TCI = htons(tx_flags >>
6464 IXGBE_TX_FLAGS_VLAN_SHIFT);
6466 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6470 /* record initial flags and protocol */
6471 first->tx_flags = tx_flags;
6472 first->protocol = protocol;
6475 /* setup tx offload for FCoE */
6476 if ((protocol == __constant_htons(ETH_P_FCOE)) &&
6477 (tx_ring->netdev->features & (NETIF_F_FSO | NETIF_F_FCOE_CRC))) {
6478 tso = ixgbe_fso(tx_ring, first, &hdr_len);
6485 #endif /* IXGBE_FCOE */
6486 tso = ixgbe_tso(tx_ring, first, &hdr_len);
6490 ixgbe_tx_csum(tx_ring, first);
6492 /* add the ATR filter if ATR is on */
6493 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
6494 ixgbe_atr(tx_ring, first);
6498 #endif /* IXGBE_FCOE */
6499 ixgbe_tx_map(tx_ring, first, hdr_len);
6501 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
6503 return NETDEV_TX_OK;
6506 dev_kfree_skb_any(first->skb);
6509 return NETDEV_TX_OK;
6512 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
6513 struct net_device *netdev)
6515 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6516 struct ixgbe_ring *tx_ring;
6519 * The minimum packet size for olinfo paylen is 17 so pad the skb
6520 * in order to meet this minimum size requirement.
6522 if (unlikely(skb->len < 17)) {
6523 if (skb_pad(skb, 17 - skb->len))
6524 return NETDEV_TX_OK;
6528 tx_ring = adapter->tx_ring[skb->queue_mapping];
6529 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
6533 * ixgbe_set_mac - Change the Ethernet Address of the NIC
6534 * @netdev: network interface device structure
6535 * @p: pointer to an address structure
6537 * Returns 0 on success, negative on failure
6539 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6541 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6542 struct ixgbe_hw *hw = &adapter->hw;
6543 struct sockaddr *addr = p;
6545 if (!is_valid_ether_addr(addr->sa_data))
6546 return -EADDRNOTAVAIL;
6548 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6549 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6551 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, VMDQ_P(0), IXGBE_RAH_AV);
6557 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6559 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6560 struct ixgbe_hw *hw = &adapter->hw;
6564 if (prtad != hw->phy.mdio.prtad)
6566 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6572 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6573 u16 addr, u16 value)
6575 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6576 struct ixgbe_hw *hw = &adapter->hw;
6578 if (prtad != hw->phy.mdio.prtad)
6580 return hw->phy.ops.write_reg(hw, addr, devad, value);
6583 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6585 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6588 #ifdef CONFIG_IXGBE_PTP
6590 return ixgbe_ptp_hwtstamp_ioctl(adapter, req, cmd);
6593 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6598 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6600 * @netdev: network interface device structure
6602 * Returns non-zero on failure
6604 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6607 struct ixgbe_adapter *adapter = netdev_priv(dev);
6608 struct ixgbe_hw *hw = &adapter->hw;
6610 if (is_valid_ether_addr(hw->mac.san_addr)) {
6612 err = dev_addr_add(dev, hw->mac.san_addr, NETDEV_HW_ADDR_T_SAN);
6615 /* update SAN MAC vmdq pool selection */
6616 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
6622 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6624 * @netdev: network interface device structure
6626 * Returns non-zero on failure
6628 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6631 struct ixgbe_adapter *adapter = netdev_priv(dev);
6632 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6634 if (is_valid_ether_addr(mac->san_addr)) {
6636 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6642 #ifdef CONFIG_NET_POLL_CONTROLLER
6644 * Polling 'interrupt' - used by things like netconsole to send skbs
6645 * without having to re-enable interrupts. It's not called while
6646 * the interrupt routine is executing.
6648 static void ixgbe_netpoll(struct net_device *netdev)
6650 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6653 /* if interface is down do nothing */
6654 if (test_bit(__IXGBE_DOWN, &adapter->state))
6657 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
6658 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
6659 for (i = 0; i < adapter->num_q_vectors; i++)
6660 ixgbe_msix_clean_rings(0, adapter->q_vector[i]);
6662 ixgbe_intr(adapter->pdev->irq, netdev);
6664 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
6668 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
6669 struct rtnl_link_stats64 *stats)
6671 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6675 for (i = 0; i < adapter->num_rx_queues; i++) {
6676 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
6682 start = u64_stats_fetch_begin_bh(&ring->syncp);
6683 packets = ring->stats.packets;
6684 bytes = ring->stats.bytes;
6685 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6686 stats->rx_packets += packets;
6687 stats->rx_bytes += bytes;
6691 for (i = 0; i < adapter->num_tx_queues; i++) {
6692 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->tx_ring[i]);
6698 start = u64_stats_fetch_begin_bh(&ring->syncp);
6699 packets = ring->stats.packets;
6700 bytes = ring->stats.bytes;
6701 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6702 stats->tx_packets += packets;
6703 stats->tx_bytes += bytes;
6707 /* following stats updated by ixgbe_watchdog_task() */
6708 stats->multicast = netdev->stats.multicast;
6709 stats->rx_errors = netdev->stats.rx_errors;
6710 stats->rx_length_errors = netdev->stats.rx_length_errors;
6711 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
6712 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
6716 #ifdef CONFIG_IXGBE_DCB
6718 * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
6719 * @adapter: pointer to ixgbe_adapter
6720 * @tc: number of traffic classes currently enabled
6722 * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
6723 * 802.1Q priority maps to a packet buffer that exists.
6725 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
6727 struct ixgbe_hw *hw = &adapter->hw;
6731 /* 82598 have a static priority to TC mapping that can not
6732 * be changed so no validation is needed.
6734 if (hw->mac.type == ixgbe_mac_82598EB)
6737 reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
6740 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
6741 u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
6743 /* If up2tc is out of bounds default to zero */
6745 reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
6749 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
6755 * ixgbe_set_prio_tc_map - Configure netdev prio tc map
6756 * @adapter: Pointer to adapter struct
6758 * Populate the netdev user priority to tc map
6760 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter *adapter)
6762 struct net_device *dev = adapter->netdev;
6763 struct ixgbe_dcb_config *dcb_cfg = &adapter->dcb_cfg;
6764 struct ieee_ets *ets = adapter->ixgbe_ieee_ets;
6767 for (prio = 0; prio < MAX_USER_PRIORITY; prio++) {
6770 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE)
6771 tc = ixgbe_dcb_get_tc_from_up(dcb_cfg, 0, prio);
6773 tc = ets->prio_tc[prio];
6775 netdev_set_prio_tc_map(dev, prio, tc);
6780 * ixgbe_setup_tc - configure net_device for multiple traffic classes
6782 * @netdev: net device to configure
6783 * @tc: number of traffic classes to enable
6785 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
6787 struct ixgbe_adapter *adapter = netdev_priv(dev);
6788 struct ixgbe_hw *hw = &adapter->hw;
6790 /* Hardware supports up to 8 traffic classes */
6791 if (tc > adapter->dcb_cfg.num_tcs.pg_tcs ||
6792 (hw->mac.type == ixgbe_mac_82598EB &&
6793 tc < MAX_TRAFFIC_CLASS))
6796 /* Hardware has to reinitialize queues and interrupts to
6797 * match packet buffer alignment. Unfortunately, the
6798 * hardware is not flexible enough to do this dynamically.
6800 if (netif_running(dev))
6802 ixgbe_clear_interrupt_scheme(adapter);
6805 netdev_set_num_tc(dev, tc);
6806 ixgbe_set_prio_tc_map(adapter);
6808 adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
6810 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
6811 adapter->last_lfc_mode = adapter->hw.fc.requested_mode;
6812 adapter->hw.fc.requested_mode = ixgbe_fc_none;
6815 netdev_reset_tc(dev);
6817 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
6818 adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
6820 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
6822 adapter->temp_dcb_cfg.pfc_mode_enable = false;
6823 adapter->dcb_cfg.pfc_mode_enable = false;
6826 ixgbe_init_interrupt_scheme(adapter);
6827 ixgbe_validate_rtr(adapter, tc);
6828 if (netif_running(dev))
6834 #endif /* CONFIG_IXGBE_DCB */
6835 void ixgbe_do_reset(struct net_device *netdev)
6837 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6839 if (netif_running(netdev))
6840 ixgbe_reinit_locked(adapter);
6842 ixgbe_reset(adapter);
6845 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
6846 netdev_features_t features)
6848 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6850 /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
6851 if (!(features & NETIF_F_RXCSUM))
6852 features &= ~NETIF_F_LRO;
6854 /* Turn off LRO if not RSC capable */
6855 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
6856 features &= ~NETIF_F_LRO;
6861 static int ixgbe_set_features(struct net_device *netdev,
6862 netdev_features_t features)
6864 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6865 netdev_features_t changed = netdev->features ^ features;
6866 bool need_reset = false;
6868 /* Make sure RSC matches LRO, reset if change */
6869 if (!(features & NETIF_F_LRO)) {
6870 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
6872 adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
6873 } else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
6874 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
6875 if (adapter->rx_itr_setting == 1 ||
6876 adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
6877 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
6879 } else if ((changed ^ features) & NETIF_F_LRO) {
6880 e_info(probe, "rx-usecs set too low, "
6886 * Check if Flow Director n-tuple support was enabled or disabled. If
6887 * the state changed, we need to reset.
6889 switch (features & NETIF_F_NTUPLE) {
6890 case NETIF_F_NTUPLE:
6891 /* turn off ATR, enable perfect filters and reset */
6892 if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
6895 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
6896 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
6899 /* turn off perfect filters, enable ATR and reset */
6900 if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
6903 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
6905 /* We cannot enable ATR if SR-IOV is enabled */
6906 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6909 /* We cannot enable ATR if we have 2 or more traffic classes */
6910 if (netdev_get_num_tc(netdev) > 1)
6913 /* We cannot enable ATR if RSS is disabled */
6914 if (adapter->ring_feature[RING_F_RSS].limit <= 1)
6917 /* A sample rate of 0 indicates ATR disabled */
6918 if (!adapter->atr_sample_rate)
6921 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
6925 if (features & NETIF_F_HW_VLAN_RX)
6926 ixgbe_vlan_strip_enable(adapter);
6928 ixgbe_vlan_strip_disable(adapter);
6930 if (changed & NETIF_F_RXALL)
6933 netdev->features = features;
6935 ixgbe_do_reset(netdev);
6940 static int ixgbe_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
6941 struct net_device *dev,
6942 const unsigned char *addr,
6945 struct ixgbe_adapter *adapter = netdev_priv(dev);
6948 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
6951 if (ndm->ndm_state & NUD_PERMANENT) {
6952 pr_info("%s: FDB only supports static addresses\n",
6957 if (is_unicast_ether_addr(addr)) {
6958 u32 rar_uc_entries = IXGBE_MAX_PF_MACVLANS;
6960 if (netdev_uc_count(dev) < rar_uc_entries)
6961 err = dev_uc_add_excl(dev, addr);
6964 } else if (is_multicast_ether_addr(addr)) {
6965 err = dev_mc_add_excl(dev, addr);
6970 /* Only return duplicate errors if NLM_F_EXCL is set */
6971 if (err == -EEXIST && !(flags & NLM_F_EXCL))
6977 static int ixgbe_ndo_fdb_del(struct ndmsg *ndm,
6978 struct net_device *dev,
6979 const unsigned char *addr)
6981 struct ixgbe_adapter *adapter = netdev_priv(dev);
6982 int err = -EOPNOTSUPP;
6984 if (ndm->ndm_state & NUD_PERMANENT) {
6985 pr_info("%s: FDB only supports static addresses\n",
6990 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
6991 if (is_unicast_ether_addr(addr))
6992 err = dev_uc_del(dev, addr);
6993 else if (is_multicast_ether_addr(addr))
6994 err = dev_mc_del(dev, addr);
7002 static int ixgbe_ndo_fdb_dump(struct sk_buff *skb,
7003 struct netlink_callback *cb,
7004 struct net_device *dev,
7007 struct ixgbe_adapter *adapter = netdev_priv(dev);
7009 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7010 idx = ndo_dflt_fdb_dump(skb, cb, dev, idx);
7015 static const struct net_device_ops ixgbe_netdev_ops = {
7016 .ndo_open = ixgbe_open,
7017 .ndo_stop = ixgbe_close,
7018 .ndo_start_xmit = ixgbe_xmit_frame,
7019 .ndo_select_queue = ixgbe_select_queue,
7020 .ndo_set_rx_mode = ixgbe_set_rx_mode,
7021 .ndo_validate_addr = eth_validate_addr,
7022 .ndo_set_mac_address = ixgbe_set_mac,
7023 .ndo_change_mtu = ixgbe_change_mtu,
7024 .ndo_tx_timeout = ixgbe_tx_timeout,
7025 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
7026 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
7027 .ndo_do_ioctl = ixgbe_ioctl,
7028 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
7029 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
7030 .ndo_set_vf_tx_rate = ixgbe_ndo_set_vf_bw,
7031 .ndo_set_vf_spoofchk = ixgbe_ndo_set_vf_spoofchk,
7032 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
7033 .ndo_get_stats64 = ixgbe_get_stats64,
7034 #ifdef CONFIG_IXGBE_DCB
7035 .ndo_setup_tc = ixgbe_setup_tc,
7037 #ifdef CONFIG_NET_POLL_CONTROLLER
7038 .ndo_poll_controller = ixgbe_netpoll,
7041 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
7042 .ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
7043 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
7044 .ndo_fcoe_enable = ixgbe_fcoe_enable,
7045 .ndo_fcoe_disable = ixgbe_fcoe_disable,
7046 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
7047 .ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
7048 #endif /* IXGBE_FCOE */
7049 .ndo_set_features = ixgbe_set_features,
7050 .ndo_fix_features = ixgbe_fix_features,
7051 .ndo_fdb_add = ixgbe_ndo_fdb_add,
7052 .ndo_fdb_del = ixgbe_ndo_fdb_del,
7053 .ndo_fdb_dump = ixgbe_ndo_fdb_dump,
7057 * ixgbe_wol_supported - Check whether device supports WoL
7058 * @hw: hw specific details
7059 * @device_id: the device ID
7060 * @subdev_id: the subsystem device ID
7062 * This function is used by probe and ethtool to determine
7063 * which devices have WoL support
7066 int ixgbe_wol_supported(struct ixgbe_adapter *adapter, u16 device_id,
7069 struct ixgbe_hw *hw = &adapter->hw;
7070 u16 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
7071 int is_wol_supported = 0;
7073 switch (device_id) {
7074 case IXGBE_DEV_ID_82599_SFP:
7075 /* Only these subdevices could supports WOL */
7076 switch (subdevice_id) {
7077 case IXGBE_SUBDEV_ID_82599_560FLR:
7078 /* only support first port */
7079 if (hw->bus.func != 0)
7081 case IXGBE_SUBDEV_ID_82599_SFP:
7082 case IXGBE_SUBDEV_ID_82599_RNDC:
7083 case IXGBE_SUBDEV_ID_82599_ECNA_DP:
7084 is_wol_supported = 1;
7088 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
7089 /* All except this subdevice support WOL */
7090 if (subdevice_id != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
7091 is_wol_supported = 1;
7093 case IXGBE_DEV_ID_82599_KX4:
7094 is_wol_supported = 1;
7096 case IXGBE_DEV_ID_X540T:
7097 case IXGBE_DEV_ID_X540T1:
7098 /* check eeprom to see if enabled wol */
7099 if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
7100 ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
7101 (hw->bus.func == 0))) {
7102 is_wol_supported = 1;
7107 return is_wol_supported;
7111 * ixgbe_probe - Device Initialization Routine
7112 * @pdev: PCI device information struct
7113 * @ent: entry in ixgbe_pci_tbl
7115 * Returns 0 on success, negative on failure
7117 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
7118 * The OS initialization, configuring of the adapter private structure,
7119 * and a hardware reset occur.
7121 static int __devinit ixgbe_probe(struct pci_dev *pdev,
7122 const struct pci_device_id *ent)
7124 struct net_device *netdev;
7125 struct ixgbe_adapter *adapter = NULL;
7126 struct ixgbe_hw *hw;
7127 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
7128 static int cards_found;
7129 int i, err, pci_using_dac;
7130 u8 part_str[IXGBE_PBANUM_LENGTH];
7131 unsigned int indices = num_possible_cpus();
7132 unsigned int dcb_max = 0;
7138 /* Catch broken hardware that put the wrong VF device ID in
7139 * the PCIe SR-IOV capability.
7141 if (pdev->is_virtfn) {
7142 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
7143 pci_name(pdev), pdev->vendor, pdev->device);
7147 err = pci_enable_device_mem(pdev);
7151 if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
7152 !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
7155 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
7157 err = dma_set_coherent_mask(&pdev->dev,
7161 "No usable DMA configuration, aborting\n");
7168 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
7169 IORESOURCE_MEM), ixgbe_driver_name);
7172 "pci_request_selected_regions failed 0x%x\n", err);
7176 pci_enable_pcie_error_reporting(pdev);
7178 pci_set_master(pdev);
7179 pci_save_state(pdev);
7181 #ifdef CONFIG_IXGBE_DCB
7182 if (ii->mac == ixgbe_mac_82598EB)
7183 dcb_max = min_t(unsigned int, indices * MAX_TRAFFIC_CLASS,
7184 IXGBE_MAX_RSS_INDICES);
7186 dcb_max = min_t(unsigned int, indices * MAX_TRAFFIC_CLASS,
7187 IXGBE_MAX_FDIR_INDICES);
7190 if (ii->mac == ixgbe_mac_82598EB)
7191 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
7193 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
7196 indices += min_t(unsigned int, num_possible_cpus(),
7197 IXGBE_MAX_FCOE_INDICES);
7199 indices = max_t(unsigned int, dcb_max, indices);
7200 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
7203 goto err_alloc_etherdev;
7206 SET_NETDEV_DEV(netdev, &pdev->dev);
7208 adapter = netdev_priv(netdev);
7209 pci_set_drvdata(pdev, adapter);
7211 adapter->netdev = netdev;
7212 adapter->pdev = pdev;
7215 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
7217 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
7218 pci_resource_len(pdev, 0));
7224 netdev->netdev_ops = &ixgbe_netdev_ops;
7225 ixgbe_set_ethtool_ops(netdev);
7226 netdev->watchdog_timeo = 5 * HZ;
7227 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
7229 adapter->bd_number = cards_found;
7232 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
7233 hw->mac.type = ii->mac;
7236 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
7237 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
7238 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
7239 if (!(eec & (1 << 8)))
7240 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
7243 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
7244 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
7245 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
7246 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
7247 hw->phy.mdio.mmds = 0;
7248 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
7249 hw->phy.mdio.dev = netdev;
7250 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
7251 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
7253 ii->get_invariants(hw);
7255 /* setup the private structure */
7256 err = ixgbe_sw_init(adapter);
7260 /* Make it possible the adapter to be woken up via WOL */
7261 switch (adapter->hw.mac.type) {
7262 case ixgbe_mac_82599EB:
7263 case ixgbe_mac_X540:
7264 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7271 * If there is a fan on this device and it has failed log the
7274 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
7275 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
7276 if (esdp & IXGBE_ESDP_SDP1)
7277 e_crit(probe, "Fan has stopped, replace the adapter\n");
7280 if (allow_unsupported_sfp)
7281 hw->allow_unsupported_sfp = allow_unsupported_sfp;
7283 /* reset_hw fills in the perm_addr as well */
7284 hw->phy.reset_if_overtemp = true;
7285 err = hw->mac.ops.reset_hw(hw);
7286 hw->phy.reset_if_overtemp = false;
7287 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
7288 hw->mac.type == ixgbe_mac_82598EB) {
7290 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
7291 e_dev_err("failed to load because an unsupported SFP+ "
7292 "module type was detected.\n");
7293 e_dev_err("Reload the driver after installing a supported "
7297 e_dev_err("HW Init failed: %d\n", err);
7301 #ifdef CONFIG_PCI_IOV
7302 ixgbe_enable_sriov(adapter, ii);
7305 netdev->features = NETIF_F_SG |
7308 NETIF_F_HW_VLAN_TX |
7309 NETIF_F_HW_VLAN_RX |
7310 NETIF_F_HW_VLAN_FILTER |
7316 netdev->hw_features = netdev->features;
7318 switch (adapter->hw.mac.type) {
7319 case ixgbe_mac_82599EB:
7320 case ixgbe_mac_X540:
7321 netdev->features |= NETIF_F_SCTP_CSUM;
7322 netdev->hw_features |= NETIF_F_SCTP_CSUM |
7329 netdev->hw_features |= NETIF_F_RXALL;
7331 netdev->vlan_features |= NETIF_F_TSO;
7332 netdev->vlan_features |= NETIF_F_TSO6;
7333 netdev->vlan_features |= NETIF_F_IP_CSUM;
7334 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
7335 netdev->vlan_features |= NETIF_F_SG;
7337 netdev->priv_flags |= IFF_UNICAST_FLT;
7338 netdev->priv_flags |= IFF_SUPP_NOFCS;
7340 #ifdef CONFIG_IXGBE_DCB
7341 netdev->dcbnl_ops = &dcbnl_ops;
7345 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7346 if (hw->mac.ops.get_device_caps) {
7347 hw->mac.ops.get_device_caps(hw, &device_caps);
7348 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
7349 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
7352 adapter->ring_feature[RING_F_FCOE].limit = IXGBE_FCRETA_SIZE;
7354 netdev->features |= NETIF_F_FSO |
7357 netdev->vlan_features |= NETIF_F_FSO |
7361 #endif /* IXGBE_FCOE */
7362 if (pci_using_dac) {
7363 netdev->features |= NETIF_F_HIGHDMA;
7364 netdev->vlan_features |= NETIF_F_HIGHDMA;
7367 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
7368 netdev->hw_features |= NETIF_F_LRO;
7369 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
7370 netdev->features |= NETIF_F_LRO;
7372 /* make sure the EEPROM is good */
7373 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
7374 e_dev_err("The EEPROM Checksum Is Not Valid\n");
7379 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
7380 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
7382 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
7383 e_dev_err("invalid MAC address\n");
7388 setup_timer(&adapter->service_timer, &ixgbe_service_timer,
7389 (unsigned long) adapter);
7391 INIT_WORK(&adapter->service_task, ixgbe_service_task);
7392 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
7394 err = ixgbe_init_interrupt_scheme(adapter);
7398 /* WOL not supported for all devices */
7400 hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
7401 if (ixgbe_wol_supported(adapter, pdev->device, pdev->subsystem_device))
7402 adapter->wol = IXGBE_WUFC_MAG;
7404 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
7406 #ifdef CONFIG_IXGBE_PTP
7407 ixgbe_ptp_init(adapter);
7408 #endif /* CONFIG_IXGBE_PTP*/
7410 /* save off EEPROM version number */
7411 hw->eeprom.ops.read(hw, 0x2e, &adapter->eeprom_verh);
7412 hw->eeprom.ops.read(hw, 0x2d, &adapter->eeprom_verl);
7414 /* pick up the PCI bus settings for reporting later */
7415 hw->mac.ops.get_bus_info(hw);
7417 /* print bus type/speed/width info */
7418 e_dev_info("(PCI Express:%s:%s) %pM\n",
7419 (hw->bus.speed == ixgbe_bus_speed_5000 ? "5.0GT/s" :
7420 hw->bus.speed == ixgbe_bus_speed_2500 ? "2.5GT/s" :
7422 (hw->bus.width == ixgbe_bus_width_pcie_x8 ? "Width x8" :
7423 hw->bus.width == ixgbe_bus_width_pcie_x4 ? "Width x4" :
7424 hw->bus.width == ixgbe_bus_width_pcie_x1 ? "Width x1" :
7428 err = ixgbe_read_pba_string_generic(hw, part_str, IXGBE_PBANUM_LENGTH);
7430 strncpy(part_str, "Unknown", IXGBE_PBANUM_LENGTH);
7431 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
7432 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
7433 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
7436 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
7437 hw->mac.type, hw->phy.type, part_str);
7439 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
7440 e_dev_warn("PCI-Express bandwidth available for this card is "
7441 "not sufficient for optimal performance.\n");
7442 e_dev_warn("For optimal performance a x8 PCI-Express slot "
7446 /* reset the hardware with the new settings */
7447 err = hw->mac.ops.start_hw(hw);
7448 if (err == IXGBE_ERR_EEPROM_VERSION) {
7449 /* We are running on a pre-production device, log a warning */
7450 e_dev_warn("This device is a pre-production adapter/LOM. "
7451 "Please be aware there may be issues associated "
7452 "with your hardware. If you are experiencing "
7453 "problems please contact your Intel or hardware "
7454 "representative who provided you with this "
7457 strcpy(netdev->name, "eth%d");
7458 err = register_netdev(netdev);
7462 /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
7463 if (hw->mac.ops.disable_tx_laser &&
7464 ((hw->phy.multispeed_fiber) ||
7465 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
7466 (hw->mac.type == ixgbe_mac_82599EB))))
7467 hw->mac.ops.disable_tx_laser(hw);
7469 /* carrier off reporting is important to ethtool even BEFORE open */
7470 netif_carrier_off(netdev);
7472 #ifdef CONFIG_IXGBE_DCA
7473 if (dca_add_requester(&pdev->dev) == 0) {
7474 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
7475 ixgbe_setup_dca(adapter);
7478 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7479 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
7480 for (i = 0; i < adapter->num_vfs; i++)
7481 ixgbe_vf_configuration(pdev, (i | 0x10000000));
7484 /* firmware requires driver version to be 0xFFFFFFFF
7485 * since os does not support feature
7487 if (hw->mac.ops.set_fw_drv_ver)
7488 hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF,
7491 /* add san mac addr to netdev */
7492 ixgbe_add_sanmac_netdev(netdev);
7494 e_dev_info("%s\n", ixgbe_default_device_descr);
7497 #ifdef CONFIG_IXGBE_HWMON
7498 if (ixgbe_sysfs_init(adapter))
7499 e_err(probe, "failed to allocate sysfs resources\n");
7500 #endif /* CONFIG_IXGBE_HWMON */
7502 #ifdef CONFIG_DEBUG_FS
7503 ixgbe_dbg_adapter_init(adapter);
7504 #endif /* CONFIG_DEBUG_FS */
7509 ixgbe_release_hw_control(adapter);
7510 ixgbe_clear_interrupt_scheme(adapter);
7512 ixgbe_disable_sriov(adapter);
7513 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
7514 iounmap(hw->hw_addr);
7516 free_netdev(netdev);
7518 pci_release_selected_regions(pdev,
7519 pci_select_bars(pdev, IORESOURCE_MEM));
7522 pci_disable_device(pdev);
7527 * ixgbe_remove - Device Removal Routine
7528 * @pdev: PCI device information struct
7530 * ixgbe_remove is called by the PCI subsystem to alert the driver
7531 * that it should release a PCI device. The could be caused by a
7532 * Hot-Plug event, or because the driver is going to be removed from
7535 static void __devexit ixgbe_remove(struct pci_dev *pdev)
7537 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7538 struct net_device *netdev = adapter->netdev;
7540 #ifdef CONFIG_DEBUG_FS
7541 ixgbe_dbg_adapter_exit(adapter);
7542 #endif /*CONFIG_DEBUG_FS */
7544 set_bit(__IXGBE_DOWN, &adapter->state);
7545 cancel_work_sync(&adapter->service_task);
7547 #ifdef CONFIG_IXGBE_PTP
7548 ixgbe_ptp_stop(adapter);
7551 #ifdef CONFIG_IXGBE_DCA
7552 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
7553 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
7554 dca_remove_requester(&pdev->dev);
7555 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
7559 #ifdef CONFIG_IXGBE_HWMON
7560 ixgbe_sysfs_exit(adapter);
7561 #endif /* CONFIG_IXGBE_HWMON */
7563 /* remove the added san mac */
7564 ixgbe_del_sanmac_netdev(netdev);
7566 if (netdev->reg_state == NETREG_REGISTERED)
7567 unregister_netdev(netdev);
7569 ixgbe_disable_sriov(adapter);
7571 ixgbe_clear_interrupt_scheme(adapter);
7573 ixgbe_release_hw_control(adapter);
7576 kfree(adapter->ixgbe_ieee_pfc);
7577 kfree(adapter->ixgbe_ieee_ets);
7580 iounmap(adapter->hw.hw_addr);
7581 pci_release_selected_regions(pdev, pci_select_bars(pdev,
7584 e_dev_info("complete\n");
7586 free_netdev(netdev);
7588 pci_disable_pcie_error_reporting(pdev);
7590 pci_disable_device(pdev);
7594 * ixgbe_io_error_detected - called when PCI error is detected
7595 * @pdev: Pointer to PCI device
7596 * @state: The current pci connection state
7598 * This function is called after a PCI bus error affecting
7599 * this device has been detected.
7601 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
7602 pci_channel_state_t state)
7604 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7605 struct net_device *netdev = adapter->netdev;
7607 #ifdef CONFIG_PCI_IOV
7608 struct pci_dev *bdev, *vfdev;
7609 u32 dw0, dw1, dw2, dw3;
7611 u16 req_id, pf_func;
7613 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
7614 adapter->num_vfs == 0)
7615 goto skip_bad_vf_detection;
7617 bdev = pdev->bus->self;
7618 while (bdev && (pci_pcie_type(bdev) != PCI_EXP_TYPE_ROOT_PORT))
7619 bdev = bdev->bus->self;
7622 goto skip_bad_vf_detection;
7624 pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
7626 goto skip_bad_vf_detection;
7628 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG, &dw0);
7629 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 4, &dw1);
7630 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 8, &dw2);
7631 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 12, &dw3);
7634 /* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
7635 if (!(req_id & 0x0080))
7636 goto skip_bad_vf_detection;
7638 pf_func = req_id & 0x01;
7639 if ((pf_func & 1) == (pdev->devfn & 1)) {
7640 unsigned int device_id;
7642 vf = (req_id & 0x7F) >> 1;
7643 e_dev_err("VF %d has caused a PCIe error\n", vf);
7644 e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
7645 "%8.8x\tdw3: %8.8x\n",
7646 dw0, dw1, dw2, dw3);
7647 switch (adapter->hw.mac.type) {
7648 case ixgbe_mac_82599EB:
7649 device_id = IXGBE_82599_VF_DEVICE_ID;
7651 case ixgbe_mac_X540:
7652 device_id = IXGBE_X540_VF_DEVICE_ID;
7659 /* Find the pci device of the offending VF */
7660 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, device_id, NULL);
7662 if (vfdev->devfn == (req_id & 0xFF))
7664 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL,
7668 * There's a slim chance the VF could have been hot plugged,
7669 * so if it is no longer present we don't need to issue the
7670 * VFLR. Just clean up the AER in that case.
7673 e_dev_err("Issuing VFLR to VF %d\n", vf);
7674 pci_write_config_dword(vfdev, 0xA8, 0x00008000);
7677 pci_cleanup_aer_uncorrect_error_status(pdev);
7681 * Even though the error may have occurred on the other port
7682 * we still need to increment the vf error reference count for
7683 * both ports because the I/O resume function will be called
7686 adapter->vferr_refcount++;
7688 return PCI_ERS_RESULT_RECOVERED;
7690 skip_bad_vf_detection:
7691 #endif /* CONFIG_PCI_IOV */
7692 netif_device_detach(netdev);
7694 if (state == pci_channel_io_perm_failure)
7695 return PCI_ERS_RESULT_DISCONNECT;
7697 if (netif_running(netdev))
7698 ixgbe_down(adapter);
7699 pci_disable_device(pdev);
7701 /* Request a slot reset. */
7702 return PCI_ERS_RESULT_NEED_RESET;
7706 * ixgbe_io_slot_reset - called after the pci bus has been reset.
7707 * @pdev: Pointer to PCI device
7709 * Restart the card from scratch, as if from a cold-boot.
7711 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7713 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7714 pci_ers_result_t result;
7717 if (pci_enable_device_mem(pdev)) {
7718 e_err(probe, "Cannot re-enable PCI device after reset.\n");
7719 result = PCI_ERS_RESULT_DISCONNECT;
7721 pci_set_master(pdev);
7722 pci_restore_state(pdev);
7723 pci_save_state(pdev);
7725 pci_wake_from_d3(pdev, false);
7727 ixgbe_reset(adapter);
7728 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7729 result = PCI_ERS_RESULT_RECOVERED;
7732 err = pci_cleanup_aer_uncorrect_error_status(pdev);
7734 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7735 "failed 0x%0x\n", err);
7736 /* non-fatal, continue */
7743 * ixgbe_io_resume - called when traffic can start flowing again.
7744 * @pdev: Pointer to PCI device
7746 * This callback is called when the error recovery driver tells us that
7747 * its OK to resume normal operation.
7749 static void ixgbe_io_resume(struct pci_dev *pdev)
7751 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7752 struct net_device *netdev = adapter->netdev;
7754 #ifdef CONFIG_PCI_IOV
7755 if (adapter->vferr_refcount) {
7756 e_info(drv, "Resuming after VF err\n");
7757 adapter->vferr_refcount--;
7762 if (netif_running(netdev))
7765 netif_device_attach(netdev);
7768 static const struct pci_error_handlers ixgbe_err_handler = {
7769 .error_detected = ixgbe_io_error_detected,
7770 .slot_reset = ixgbe_io_slot_reset,
7771 .resume = ixgbe_io_resume,
7774 static struct pci_driver ixgbe_driver = {
7775 .name = ixgbe_driver_name,
7776 .id_table = ixgbe_pci_tbl,
7777 .probe = ixgbe_probe,
7778 .remove = __devexit_p(ixgbe_remove),
7780 .suspend = ixgbe_suspend,
7781 .resume = ixgbe_resume,
7783 .shutdown = ixgbe_shutdown,
7784 .err_handler = &ixgbe_err_handler
7788 * ixgbe_init_module - Driver Registration Routine
7790 * ixgbe_init_module is the first routine called when the driver is
7791 * loaded. All it does is register with the PCI subsystem.
7793 static int __init ixgbe_init_module(void)
7796 pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
7797 pr_info("%s\n", ixgbe_copyright);
7799 #ifdef CONFIG_DEBUG_FS
7801 #endif /* CONFIG_DEBUG_FS */
7803 #ifdef CONFIG_IXGBE_DCA
7804 dca_register_notify(&dca_notifier);
7807 ret = pci_register_driver(&ixgbe_driver);
7811 module_init(ixgbe_init_module);
7814 * ixgbe_exit_module - Driver Exit Cleanup Routine
7816 * ixgbe_exit_module is called just before the driver is removed
7819 static void __exit ixgbe_exit_module(void)
7821 #ifdef CONFIG_IXGBE_DCA
7822 dca_unregister_notify(&dca_notifier);
7824 pci_unregister_driver(&ixgbe_driver);
7826 #ifdef CONFIG_DEBUG_FS
7828 #endif /* CONFIG_DEBUG_FS */
7830 rcu_barrier(); /* Wait for completion of call_rcu()'s */
7833 #ifdef CONFIG_IXGBE_DCA
7834 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
7839 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
7840 __ixgbe_notify_dca);
7842 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7845 #endif /* CONFIG_IXGBE_DCA */
7847 module_exit(ixgbe_exit_module);