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;
1249 u8 nexthdr = 0; /* default to not TCP */
1252 /* this should never happen, but better safe than sorry */
1253 if (max_len < ETH_HLEN)
1256 /* initialize network frame pointer */
1259 /* set first protocol and move network header forward */
1260 protocol = hdr.eth->h_proto;
1261 hdr.network += ETH_HLEN;
1263 /* handle any vlan tag if present */
1264 if (protocol == __constant_htons(ETH_P_8021Q)) {
1265 if ((hdr.network - data) > (max_len - VLAN_HLEN))
1268 protocol = hdr.vlan->h_vlan_encapsulated_proto;
1269 hdr.network += VLAN_HLEN;
1272 /* handle L3 protocols */
1273 if (protocol == __constant_htons(ETH_P_IP)) {
1274 if ((hdr.network - data) > (max_len - sizeof(struct iphdr)))
1277 /* access ihl as a u8 to avoid unaligned access on ia64 */
1278 hlen = (hdr.network[0] & 0x0F) << 2;
1280 /* verify hlen meets minimum size requirements */
1281 if (hlen < sizeof(struct iphdr))
1282 return hdr.network - data;
1284 /* record next protocol */
1285 nexthdr = hdr.ipv4->protocol;
1286 hdr.network += hlen;
1288 } else if (protocol == __constant_htons(ETH_P_FCOE)) {
1289 if ((hdr.network - data) > (max_len - FCOE_HEADER_LEN))
1291 hdr.network += FCOE_HEADER_LEN;
1294 return hdr.network - data;
1297 /* finally sort out TCP */
1298 if (nexthdr == IPPROTO_TCP) {
1299 if ((hdr.network - data) > (max_len - sizeof(struct tcphdr)))
1302 /* access doff as a u8 to avoid unaligned access on ia64 */
1303 hlen = (hdr.network[12] & 0xF0) >> 2;
1305 /* verify hlen meets minimum size requirements */
1306 if (hlen < sizeof(struct tcphdr))
1307 return hdr.network - data;
1309 hdr.network += hlen;
1313 * If everything has gone correctly hdr.network should be the
1314 * data section of the packet and will be the end of the header.
1315 * If not then it probably represents the end of the last recognized
1318 if ((hdr.network - data) < max_len)
1319 return hdr.network - data;
1324 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring *ring,
1325 struct sk_buff *skb)
1327 u16 hdr_len = skb_headlen(skb);
1329 /* set gso_size to avoid messing up TCP MSS */
1330 skb_shinfo(skb)->gso_size = DIV_ROUND_UP((skb->len - hdr_len),
1331 IXGBE_CB(skb)->append_cnt);
1334 static void ixgbe_update_rsc_stats(struct ixgbe_ring *rx_ring,
1335 struct sk_buff *skb)
1337 /* if append_cnt is 0 then frame is not RSC */
1338 if (!IXGBE_CB(skb)->append_cnt)
1341 rx_ring->rx_stats.rsc_count += IXGBE_CB(skb)->append_cnt;
1342 rx_ring->rx_stats.rsc_flush++;
1344 ixgbe_set_rsc_gso_size(rx_ring, skb);
1346 /* gso_size is computed using append_cnt so always clear it last */
1347 IXGBE_CB(skb)->append_cnt = 0;
1351 * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor
1352 * @rx_ring: rx descriptor ring packet is being transacted on
1353 * @rx_desc: pointer to the EOP Rx descriptor
1354 * @skb: pointer to current skb being populated
1356 * This function checks the ring, descriptor, and packet information in
1357 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1358 * other fields within the skb.
1360 static void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring,
1361 union ixgbe_adv_rx_desc *rx_desc,
1362 struct sk_buff *skb)
1364 struct net_device *dev = rx_ring->netdev;
1366 ixgbe_update_rsc_stats(rx_ring, skb);
1368 ixgbe_rx_hash(rx_ring, rx_desc, skb);
1370 ixgbe_rx_checksum(rx_ring, rx_desc, skb);
1372 #ifdef CONFIG_IXGBE_PTP
1373 ixgbe_ptp_rx_hwtstamp(rx_ring->q_vector, rx_desc, skb);
1376 if ((dev->features & NETIF_F_HW_VLAN_RX) &&
1377 ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
1378 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
1379 __vlan_hwaccel_put_tag(skb, vid);
1382 skb_record_rx_queue(skb, rx_ring->queue_index);
1384 skb->protocol = eth_type_trans(skb, dev);
1387 static void ixgbe_rx_skb(struct ixgbe_q_vector *q_vector,
1388 struct sk_buff *skb)
1390 struct ixgbe_adapter *adapter = q_vector->adapter;
1392 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
1393 napi_gro_receive(&q_vector->napi, skb);
1399 * ixgbe_is_non_eop - process handling of non-EOP buffers
1400 * @rx_ring: Rx ring being processed
1401 * @rx_desc: Rx descriptor for current buffer
1402 * @skb: Current socket buffer containing buffer in progress
1404 * This function updates next to clean. If the buffer is an EOP buffer
1405 * this function exits returning false, otherwise it will place the
1406 * sk_buff in the next buffer to be chained and return true indicating
1407 * that this is in fact a non-EOP buffer.
1409 static bool ixgbe_is_non_eop(struct ixgbe_ring *rx_ring,
1410 union ixgbe_adv_rx_desc *rx_desc,
1411 struct sk_buff *skb)
1413 u32 ntc = rx_ring->next_to_clean + 1;
1415 /* fetch, update, and store next to clean */
1416 ntc = (ntc < rx_ring->count) ? ntc : 0;
1417 rx_ring->next_to_clean = ntc;
1419 prefetch(IXGBE_RX_DESC(rx_ring, ntc));
1421 /* update RSC append count if present */
1422 if (ring_is_rsc_enabled(rx_ring)) {
1423 __le32 rsc_enabled = rx_desc->wb.lower.lo_dword.data &
1424 cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK);
1426 if (unlikely(rsc_enabled)) {
1427 u32 rsc_cnt = le32_to_cpu(rsc_enabled);
1429 rsc_cnt >>= IXGBE_RXDADV_RSCCNT_SHIFT;
1430 IXGBE_CB(skb)->append_cnt += rsc_cnt - 1;
1432 /* update ntc based on RSC value */
1433 ntc = le32_to_cpu(rx_desc->wb.upper.status_error);
1434 ntc &= IXGBE_RXDADV_NEXTP_MASK;
1435 ntc >>= IXGBE_RXDADV_NEXTP_SHIFT;
1439 /* if we are the last buffer then there is nothing else to do */
1440 if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1443 /* place skb in next buffer to be received */
1444 rx_ring->rx_buffer_info[ntc].skb = skb;
1445 rx_ring->rx_stats.non_eop_descs++;
1451 * ixgbe_pull_tail - ixgbe specific version of skb_pull_tail
1452 * @rx_ring: rx descriptor ring packet is being transacted on
1453 * @skb: pointer to current skb being adjusted
1455 * This function is an ixgbe specific version of __pskb_pull_tail. The
1456 * main difference between this version and the original function is that
1457 * this function can make several assumptions about the state of things
1458 * that allow for significant optimizations versus the standard function.
1459 * As a result we can do things like drop a frag and maintain an accurate
1460 * truesize for the skb.
1462 static void ixgbe_pull_tail(struct ixgbe_ring *rx_ring,
1463 struct sk_buff *skb)
1465 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1467 unsigned int pull_len;
1470 * it is valid to use page_address instead of kmap since we are
1471 * working with pages allocated out of the lomem pool per
1472 * alloc_page(GFP_ATOMIC)
1474 va = skb_frag_address(frag);
1477 * we need the header to contain the greater of either ETH_HLEN or
1478 * 60 bytes if the skb->len is less than 60 for skb_pad.
1480 pull_len = ixgbe_get_headlen(va, IXGBE_RX_HDR_SIZE);
1482 /* align pull length to size of long to optimize memcpy performance */
1483 skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
1485 /* update all of the pointers */
1486 skb_frag_size_sub(frag, pull_len);
1487 frag->page_offset += pull_len;
1488 skb->data_len -= pull_len;
1489 skb->tail += pull_len;
1493 * ixgbe_dma_sync_frag - perform DMA sync for first frag of SKB
1494 * @rx_ring: rx descriptor ring packet is being transacted on
1495 * @skb: pointer to current skb being updated
1497 * This function provides a basic DMA sync up for the first fragment of an
1498 * skb. The reason for doing this is that the first fragment cannot be
1499 * unmapped until we have reached the end of packet descriptor for a buffer
1502 static void ixgbe_dma_sync_frag(struct ixgbe_ring *rx_ring,
1503 struct sk_buff *skb)
1505 /* if the page was released unmap it, else just sync our portion */
1506 if (unlikely(IXGBE_CB(skb)->page_released)) {
1507 dma_unmap_page(rx_ring->dev, IXGBE_CB(skb)->dma,
1508 ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1509 IXGBE_CB(skb)->page_released = false;
1511 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1513 dma_sync_single_range_for_cpu(rx_ring->dev,
1516 ixgbe_rx_bufsz(rx_ring),
1519 IXGBE_CB(skb)->dma = 0;
1523 * ixgbe_cleanup_headers - Correct corrupted or empty headers
1524 * @rx_ring: rx descriptor ring packet is being transacted on
1525 * @rx_desc: pointer to the EOP Rx descriptor
1526 * @skb: pointer to current skb being fixed
1528 * Check for corrupted packet headers caused by senders on the local L2
1529 * embedded NIC switch not setting up their Tx Descriptors right. These
1530 * should be very rare.
1532 * Also address the case where we are pulling data in on pages only
1533 * and as such no data is present in the skb header.
1535 * In addition if skb is not at least 60 bytes we need to pad it so that
1536 * it is large enough to qualify as a valid Ethernet frame.
1538 * Returns true if an error was encountered and skb was freed.
1540 static bool ixgbe_cleanup_headers(struct ixgbe_ring *rx_ring,
1541 union ixgbe_adv_rx_desc *rx_desc,
1542 struct sk_buff *skb)
1544 struct net_device *netdev = rx_ring->netdev;
1546 /* verify that the packet does not have any known errors */
1547 if (unlikely(ixgbe_test_staterr(rx_desc,
1548 IXGBE_RXDADV_ERR_FRAME_ERR_MASK) &&
1549 !(netdev->features & NETIF_F_RXALL))) {
1550 dev_kfree_skb_any(skb);
1554 /* place header in linear portion of buffer */
1555 if (skb_is_nonlinear(skb))
1556 ixgbe_pull_tail(rx_ring, skb);
1559 /* do not attempt to pad FCoE Frames as this will disrupt DDP */
1560 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc))
1564 /* if skb_pad returns an error the skb was freed */
1565 if (unlikely(skb->len < 60)) {
1566 int pad_len = 60 - skb->len;
1568 if (skb_pad(skb, pad_len))
1570 __skb_put(skb, pad_len);
1577 * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1578 * @rx_ring: rx descriptor ring to store buffers on
1579 * @old_buff: donor buffer to have page reused
1581 * Synchronizes page for reuse by the adapter
1583 static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring,
1584 struct ixgbe_rx_buffer *old_buff)
1586 struct ixgbe_rx_buffer *new_buff;
1587 u16 nta = rx_ring->next_to_alloc;
1589 new_buff = &rx_ring->rx_buffer_info[nta];
1591 /* update, and store next to alloc */
1593 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1595 /* transfer page from old buffer to new buffer */
1596 new_buff->page = old_buff->page;
1597 new_buff->dma = old_buff->dma;
1598 new_buff->page_offset = old_buff->page_offset;
1600 /* sync the buffer for use by the device */
1601 dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
1602 new_buff->page_offset,
1603 ixgbe_rx_bufsz(rx_ring),
1608 * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1609 * @rx_ring: rx descriptor ring to transact packets on
1610 * @rx_buffer: buffer containing page to add
1611 * @rx_desc: descriptor containing length of buffer written by hardware
1612 * @skb: sk_buff to place the data into
1614 * This function will add the data contained in rx_buffer->page to the skb.
1615 * This is done either through a direct copy if the data in the buffer is
1616 * less than the skb header size, otherwise it will just attach the page as
1617 * a frag to the skb.
1619 * The function will then update the page offset if necessary and return
1620 * true if the buffer can be reused by the adapter.
1622 static bool ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
1623 struct ixgbe_rx_buffer *rx_buffer,
1624 union ixgbe_adv_rx_desc *rx_desc,
1625 struct sk_buff *skb)
1627 struct page *page = rx_buffer->page;
1628 unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
1629 #if (PAGE_SIZE < 8192)
1630 unsigned int truesize = ixgbe_rx_bufsz(rx_ring);
1632 unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
1633 unsigned int last_offset = ixgbe_rx_pg_size(rx_ring) -
1634 ixgbe_rx_bufsz(rx_ring);
1637 if ((size <= IXGBE_RX_HDR_SIZE) && !skb_is_nonlinear(skb)) {
1638 unsigned char *va = page_address(page) + rx_buffer->page_offset;
1640 memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
1642 /* we can reuse buffer as-is, just make sure it is local */
1643 if (likely(page_to_nid(page) == numa_node_id()))
1646 /* this page cannot be reused so discard it */
1651 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
1652 rx_buffer->page_offset, size, truesize);
1654 /* avoid re-using remote pages */
1655 if (unlikely(page_to_nid(page) != numa_node_id()))
1658 #if (PAGE_SIZE < 8192)
1659 /* if we are only owner of page we can reuse it */
1660 if (unlikely(page_count(page) != 1))
1663 /* flip page offset to other buffer */
1664 rx_buffer->page_offset ^= truesize;
1667 * since we are the only owner of the page and we need to
1668 * increment it, just set the value to 2 in order to avoid
1669 * an unecessary locked operation
1671 atomic_set(&page->_count, 2);
1673 /* move offset up to the next cache line */
1674 rx_buffer->page_offset += truesize;
1676 if (rx_buffer->page_offset > last_offset)
1679 /* bump ref count on page before it is given to the stack */
1686 static struct sk_buff *ixgbe_fetch_rx_buffer(struct ixgbe_ring *rx_ring,
1687 union ixgbe_adv_rx_desc *rx_desc)
1689 struct ixgbe_rx_buffer *rx_buffer;
1690 struct sk_buff *skb;
1693 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
1694 page = rx_buffer->page;
1697 skb = rx_buffer->skb;
1700 void *page_addr = page_address(page) +
1701 rx_buffer->page_offset;
1703 /* prefetch first cache line of first page */
1704 prefetch(page_addr);
1705 #if L1_CACHE_BYTES < 128
1706 prefetch(page_addr + L1_CACHE_BYTES);
1709 /* allocate a skb to store the frags */
1710 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1712 if (unlikely(!skb)) {
1713 rx_ring->rx_stats.alloc_rx_buff_failed++;
1718 * we will be copying header into skb->data in
1719 * pskb_may_pull so it is in our interest to prefetch
1720 * it now to avoid a possible cache miss
1722 prefetchw(skb->data);
1725 * Delay unmapping of the first packet. It carries the
1726 * header information, HW may still access the header
1727 * after the writeback. Only unmap it when EOP is
1730 if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1733 IXGBE_CB(skb)->dma = rx_buffer->dma;
1735 if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
1736 ixgbe_dma_sync_frag(rx_ring, skb);
1739 /* we are reusing so sync this buffer for CPU use */
1740 dma_sync_single_range_for_cpu(rx_ring->dev,
1742 rx_buffer->page_offset,
1743 ixgbe_rx_bufsz(rx_ring),
1747 /* pull page into skb */
1748 if (ixgbe_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
1749 /* hand second half of page back to the ring */
1750 ixgbe_reuse_rx_page(rx_ring, rx_buffer);
1751 } else if (IXGBE_CB(skb)->dma == rx_buffer->dma) {
1752 /* the page has been released from the ring */
1753 IXGBE_CB(skb)->page_released = true;
1755 /* we are not reusing the buffer so unmap it */
1756 dma_unmap_page(rx_ring->dev, rx_buffer->dma,
1757 ixgbe_rx_pg_size(rx_ring),
1761 /* clear contents of buffer_info */
1762 rx_buffer->skb = NULL;
1764 rx_buffer->page = NULL;
1770 * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
1771 * @q_vector: structure containing interrupt and ring information
1772 * @rx_ring: rx descriptor ring to transact packets on
1773 * @budget: Total limit on number of packets to process
1775 * This function provides a "bounce buffer" approach to Rx interrupt
1776 * processing. The advantage to this is that on systems that have
1777 * expensive overhead for IOMMU access this provides a means of avoiding
1778 * it by maintaining the mapping of the page to the syste.
1780 * Returns true if all work is completed without reaching budget
1782 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1783 struct ixgbe_ring *rx_ring,
1786 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1788 struct ixgbe_adapter *adapter = q_vector->adapter;
1790 unsigned int mss = 0;
1791 #endif /* IXGBE_FCOE */
1792 u16 cleaned_count = ixgbe_desc_unused(rx_ring);
1795 union ixgbe_adv_rx_desc *rx_desc;
1796 struct sk_buff *skb;
1798 /* return some buffers to hardware, one at a time is too slow */
1799 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
1800 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1804 rx_desc = IXGBE_RX_DESC(rx_ring, rx_ring->next_to_clean);
1806 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_DD))
1810 * This memory barrier is needed to keep us from reading
1811 * any other fields out of the rx_desc until we know the
1812 * RXD_STAT_DD bit is set
1816 /* retrieve a buffer from the ring */
1817 skb = ixgbe_fetch_rx_buffer(rx_ring, rx_desc);
1819 /* exit if we failed to retrieve a buffer */
1825 /* place incomplete frames back on ring for completion */
1826 if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
1829 /* verify the packet layout is correct */
1830 if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
1833 /* probably a little skewed due to removing CRC */
1834 total_rx_bytes += skb->len;
1837 /* populate checksum, timestamp, VLAN, and protocol */
1838 ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
1841 /* if ddp, not passing to ULD unless for FCP_RSP or error */
1842 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc)) {
1843 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
1844 /* include DDPed FCoE data */
1845 if (ddp_bytes > 0) {
1847 mss = rx_ring->netdev->mtu -
1848 sizeof(struct fcoe_hdr) -
1849 sizeof(struct fc_frame_header) -
1850 sizeof(struct fcoe_crc_eof);
1854 total_rx_bytes += ddp_bytes;
1855 total_rx_packets += DIV_ROUND_UP(ddp_bytes,
1859 dev_kfree_skb_any(skb);
1864 #endif /* IXGBE_FCOE */
1865 ixgbe_rx_skb(q_vector, skb);
1867 /* update budget accounting */
1869 } while (likely(budget));
1871 u64_stats_update_begin(&rx_ring->syncp);
1872 rx_ring->stats.packets += total_rx_packets;
1873 rx_ring->stats.bytes += total_rx_bytes;
1874 u64_stats_update_end(&rx_ring->syncp);
1875 q_vector->rx.total_packets += total_rx_packets;
1876 q_vector->rx.total_bytes += total_rx_bytes;
1879 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1885 * ixgbe_configure_msix - Configure MSI-X hardware
1886 * @adapter: board private structure
1888 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1891 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1893 struct ixgbe_q_vector *q_vector;
1897 /* Populate MSIX to EITR Select */
1898 if (adapter->num_vfs > 32) {
1899 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
1900 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
1904 * Populate the IVAR table and set the ITR values to the
1905 * corresponding register.
1907 for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
1908 struct ixgbe_ring *ring;
1909 q_vector = adapter->q_vector[v_idx];
1911 ixgbe_for_each_ring(ring, q_vector->rx)
1912 ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1914 ixgbe_for_each_ring(ring, q_vector->tx)
1915 ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1917 if (q_vector->tx.ring && !q_vector->rx.ring) {
1918 /* tx only vector */
1919 if (adapter->tx_itr_setting == 1)
1920 q_vector->itr = IXGBE_10K_ITR;
1922 q_vector->itr = adapter->tx_itr_setting;
1924 /* rx or rx/tx vector */
1925 if (adapter->rx_itr_setting == 1)
1926 q_vector->itr = IXGBE_20K_ITR;
1928 q_vector->itr = adapter->rx_itr_setting;
1931 ixgbe_write_eitr(q_vector);
1934 switch (adapter->hw.mac.type) {
1935 case ixgbe_mac_82598EB:
1936 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1939 case ixgbe_mac_82599EB:
1940 case ixgbe_mac_X540:
1941 ixgbe_set_ivar(adapter, -1, 1, v_idx);
1946 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1948 /* set up to autoclear timer, and the vectors */
1949 mask = IXGBE_EIMS_ENABLE_MASK;
1950 mask &= ~(IXGBE_EIMS_OTHER |
1951 IXGBE_EIMS_MAILBOX |
1954 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1957 enum latency_range {
1961 latency_invalid = 255
1965 * ixgbe_update_itr - update the dynamic ITR value based on statistics
1966 * @q_vector: structure containing interrupt and ring information
1967 * @ring_container: structure containing ring performance data
1969 * Stores a new ITR value based on packets and byte
1970 * counts during the last interrupt. The advantage of per interrupt
1971 * computation is faster updates and more accurate ITR for the current
1972 * traffic pattern. Constants in this function were computed
1973 * based on theoretical maximum wire speed and thresholds were set based
1974 * on testing data as well as attempting to minimize response time
1975 * while increasing bulk throughput.
1976 * this functionality is controlled by the InterruptThrottleRate module
1977 * parameter (see ixgbe_param.c)
1979 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
1980 struct ixgbe_ring_container *ring_container)
1982 int bytes = ring_container->total_bytes;
1983 int packets = ring_container->total_packets;
1986 u8 itr_setting = ring_container->itr;
1991 /* simple throttlerate management
1992 * 0-10MB/s lowest (100000 ints/s)
1993 * 10-20MB/s low (20000 ints/s)
1994 * 20-1249MB/s bulk (8000 ints/s)
1996 /* what was last interrupt timeslice? */
1997 timepassed_us = q_vector->itr >> 2;
1998 bytes_perint = bytes / timepassed_us; /* bytes/usec */
2000 switch (itr_setting) {
2001 case lowest_latency:
2002 if (bytes_perint > 10)
2003 itr_setting = low_latency;
2006 if (bytes_perint > 20)
2007 itr_setting = bulk_latency;
2008 else if (bytes_perint <= 10)
2009 itr_setting = lowest_latency;
2012 if (bytes_perint <= 20)
2013 itr_setting = low_latency;
2017 /* clear work counters since we have the values we need */
2018 ring_container->total_bytes = 0;
2019 ring_container->total_packets = 0;
2021 /* write updated itr to ring container */
2022 ring_container->itr = itr_setting;
2026 * ixgbe_write_eitr - write EITR register in hardware specific way
2027 * @q_vector: structure containing interrupt and ring information
2029 * This function is made to be called by ethtool and by the driver
2030 * when it needs to update EITR registers at runtime. Hardware
2031 * specific quirks/differences are taken care of here.
2033 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
2035 struct ixgbe_adapter *adapter = q_vector->adapter;
2036 struct ixgbe_hw *hw = &adapter->hw;
2037 int v_idx = q_vector->v_idx;
2038 u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
2040 switch (adapter->hw.mac.type) {
2041 case ixgbe_mac_82598EB:
2042 /* must write high and low 16 bits to reset counter */
2043 itr_reg |= (itr_reg << 16);
2045 case ixgbe_mac_82599EB:
2046 case ixgbe_mac_X540:
2048 * set the WDIS bit to not clear the timer bits and cause an
2049 * immediate assertion of the interrupt
2051 itr_reg |= IXGBE_EITR_CNT_WDIS;
2056 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
2059 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
2061 u32 new_itr = q_vector->itr;
2064 ixgbe_update_itr(q_vector, &q_vector->tx);
2065 ixgbe_update_itr(q_vector, &q_vector->rx);
2067 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
2069 switch (current_itr) {
2070 /* counts and packets in update_itr are dependent on these numbers */
2071 case lowest_latency:
2072 new_itr = IXGBE_100K_ITR;
2075 new_itr = IXGBE_20K_ITR;
2078 new_itr = IXGBE_8K_ITR;
2084 if (new_itr != q_vector->itr) {
2085 /* do an exponential smoothing */
2086 new_itr = (10 * new_itr * q_vector->itr) /
2087 ((9 * new_itr) + q_vector->itr);
2089 /* save the algorithm value here */
2090 q_vector->itr = new_itr;
2092 ixgbe_write_eitr(q_vector);
2097 * ixgbe_check_overtemp_subtask - check for over temperature
2098 * @adapter: pointer to adapter
2100 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2102 struct ixgbe_hw *hw = &adapter->hw;
2103 u32 eicr = adapter->interrupt_event;
2105 if (test_bit(__IXGBE_DOWN, &adapter->state))
2108 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2109 !(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2112 adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2114 switch (hw->device_id) {
2115 case IXGBE_DEV_ID_82599_T3_LOM:
2117 * Since the warning interrupt is for both ports
2118 * we don't have to check if:
2119 * - This interrupt wasn't for our port.
2120 * - We may have missed the interrupt so always have to
2121 * check if we got a LSC
2123 if (!(eicr & IXGBE_EICR_GPI_SDP0) &&
2124 !(eicr & IXGBE_EICR_LSC))
2127 if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2129 bool link_up = false;
2131 hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
2137 /* Check if this is not due to overtemp */
2138 if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
2143 if (!(eicr & IXGBE_EICR_GPI_SDP0))
2148 "Network adapter has been stopped because it has over heated. "
2149 "Restart the computer. If the problem persists, "
2150 "power off the system and replace the adapter\n");
2152 adapter->interrupt_event = 0;
2155 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2157 struct ixgbe_hw *hw = &adapter->hw;
2159 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2160 (eicr & IXGBE_EICR_GPI_SDP1)) {
2161 e_crit(probe, "Fan has stopped, replace the adapter\n");
2162 /* write to clear the interrupt */
2163 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
2167 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2169 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2172 switch (adapter->hw.mac.type) {
2173 case ixgbe_mac_82599EB:
2175 * Need to check link state so complete overtemp check
2178 if (((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)) &&
2179 (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2180 adapter->interrupt_event = eicr;
2181 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2182 ixgbe_service_event_schedule(adapter);
2186 case ixgbe_mac_X540:
2187 if (!(eicr & IXGBE_EICR_TS))
2195 "Network adapter has been stopped because it has over heated. "
2196 "Restart the computer. If the problem persists, "
2197 "power off the system and replace the adapter\n");
2200 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2202 struct ixgbe_hw *hw = &adapter->hw;
2204 if (eicr & IXGBE_EICR_GPI_SDP2) {
2205 /* Clear the interrupt */
2206 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
2207 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2208 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2209 ixgbe_service_event_schedule(adapter);
2213 if (eicr & IXGBE_EICR_GPI_SDP1) {
2214 /* Clear the interrupt */
2215 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
2216 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2217 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2218 ixgbe_service_event_schedule(adapter);
2223 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2225 struct ixgbe_hw *hw = &adapter->hw;
2228 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2229 adapter->link_check_timeout = jiffies;
2230 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2231 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2232 IXGBE_WRITE_FLUSH(hw);
2233 ixgbe_service_event_schedule(adapter);
2237 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2241 struct ixgbe_hw *hw = &adapter->hw;
2243 switch (hw->mac.type) {
2244 case ixgbe_mac_82598EB:
2245 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2246 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2248 case ixgbe_mac_82599EB:
2249 case ixgbe_mac_X540:
2250 mask = (qmask & 0xFFFFFFFF);
2252 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2253 mask = (qmask >> 32);
2255 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2260 /* skip the flush */
2263 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
2267 struct ixgbe_hw *hw = &adapter->hw;
2269 switch (hw->mac.type) {
2270 case ixgbe_mac_82598EB:
2271 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2272 IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
2274 case ixgbe_mac_82599EB:
2275 case ixgbe_mac_X540:
2276 mask = (qmask & 0xFFFFFFFF);
2278 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
2279 mask = (qmask >> 32);
2281 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
2286 /* skip the flush */
2290 * ixgbe_irq_enable - Enable default interrupt generation settings
2291 * @adapter: board private structure
2293 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2296 u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2298 /* don't reenable LSC while waiting for link */
2299 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
2300 mask &= ~IXGBE_EIMS_LSC;
2302 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2303 switch (adapter->hw.mac.type) {
2304 case ixgbe_mac_82599EB:
2305 mask |= IXGBE_EIMS_GPI_SDP0;
2307 case ixgbe_mac_X540:
2308 mask |= IXGBE_EIMS_TS;
2313 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2314 mask |= IXGBE_EIMS_GPI_SDP1;
2315 switch (adapter->hw.mac.type) {
2316 case ixgbe_mac_82599EB:
2317 mask |= IXGBE_EIMS_GPI_SDP1;
2318 mask |= IXGBE_EIMS_GPI_SDP2;
2319 case ixgbe_mac_X540:
2320 mask |= IXGBE_EIMS_ECC;
2321 mask |= IXGBE_EIMS_MAILBOX;
2327 #ifdef CONFIG_IXGBE_PTP
2328 if (adapter->hw.mac.type == ixgbe_mac_X540)
2329 mask |= IXGBE_EIMS_TIMESYNC;
2332 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2333 !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
2334 mask |= IXGBE_EIMS_FLOW_DIR;
2336 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2338 ixgbe_irq_enable_queues(adapter, ~0);
2340 IXGBE_WRITE_FLUSH(&adapter->hw);
2343 static irqreturn_t ixgbe_msix_other(int irq, void *data)
2345 struct ixgbe_adapter *adapter = data;
2346 struct ixgbe_hw *hw = &adapter->hw;
2350 * Workaround for Silicon errata. Use clear-by-write instead
2351 * of clear-by-read. Reading with EICS will return the
2352 * interrupt causes without clearing, which later be done
2353 * with the write to EICR.
2355 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
2356 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
2358 if (eicr & IXGBE_EICR_LSC)
2359 ixgbe_check_lsc(adapter);
2361 if (eicr & IXGBE_EICR_MAILBOX)
2362 ixgbe_msg_task(adapter);
2364 switch (hw->mac.type) {
2365 case ixgbe_mac_82599EB:
2366 case ixgbe_mac_X540:
2367 if (eicr & IXGBE_EICR_ECC)
2368 e_info(link, "Received unrecoverable ECC Err, please "
2370 /* Handle Flow Director Full threshold interrupt */
2371 if (eicr & IXGBE_EICR_FLOW_DIR) {
2372 int reinit_count = 0;
2374 for (i = 0; i < adapter->num_tx_queues; i++) {
2375 struct ixgbe_ring *ring = adapter->tx_ring[i];
2376 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
2381 /* no more flow director interrupts until after init */
2382 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
2383 adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
2384 ixgbe_service_event_schedule(adapter);
2387 ixgbe_check_sfp_event(adapter, eicr);
2388 ixgbe_check_overtemp_event(adapter, eicr);
2394 ixgbe_check_fan_failure(adapter, eicr);
2396 #ifdef CONFIG_IXGBE_PTP
2397 if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
2398 ixgbe_ptp_check_pps_event(adapter, eicr);
2401 /* re-enable the original interrupt state, no lsc, no queues */
2402 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2403 ixgbe_irq_enable(adapter, false, false);
2408 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
2410 struct ixgbe_q_vector *q_vector = data;
2412 /* EIAM disabled interrupts (on this vector) for us */
2414 if (q_vector->rx.ring || q_vector->tx.ring)
2415 napi_schedule(&q_vector->napi);
2421 * ixgbe_poll - NAPI Rx polling callback
2422 * @napi: structure for representing this polling device
2423 * @budget: how many packets driver is allowed to clean
2425 * This function is used for legacy and MSI, NAPI mode
2427 int ixgbe_poll(struct napi_struct *napi, int budget)
2429 struct ixgbe_q_vector *q_vector =
2430 container_of(napi, struct ixgbe_q_vector, napi);
2431 struct ixgbe_adapter *adapter = q_vector->adapter;
2432 struct ixgbe_ring *ring;
2433 int per_ring_budget;
2434 bool clean_complete = true;
2436 #ifdef CONFIG_IXGBE_DCA
2437 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2438 ixgbe_update_dca(q_vector);
2441 ixgbe_for_each_ring(ring, q_vector->tx)
2442 clean_complete &= !!ixgbe_clean_tx_irq(q_vector, ring);
2444 /* attempt to distribute budget to each queue fairly, but don't allow
2445 * the budget to go below 1 because we'll exit polling */
2446 if (q_vector->rx.count > 1)
2447 per_ring_budget = max(budget/q_vector->rx.count, 1);
2449 per_ring_budget = budget;
2451 ixgbe_for_each_ring(ring, q_vector->rx)
2452 clean_complete &= ixgbe_clean_rx_irq(q_vector, ring,
2455 /* If all work not completed, return budget and keep polling */
2456 if (!clean_complete)
2459 /* all work done, exit the polling mode */
2460 napi_complete(napi);
2461 if (adapter->rx_itr_setting & 1)
2462 ixgbe_set_itr(q_vector);
2463 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2464 ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
2470 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2471 * @adapter: board private structure
2473 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2474 * interrupts from the kernel.
2476 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2478 struct net_device *netdev = adapter->netdev;
2482 for (vector = 0; vector < adapter->num_q_vectors; vector++) {
2483 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2484 struct msix_entry *entry = &adapter->msix_entries[vector];
2486 if (q_vector->tx.ring && q_vector->rx.ring) {
2487 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2488 "%s-%s-%d", netdev->name, "TxRx", ri++);
2490 } else if (q_vector->rx.ring) {
2491 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2492 "%s-%s-%d", netdev->name, "rx", ri++);
2493 } else if (q_vector->tx.ring) {
2494 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2495 "%s-%s-%d", netdev->name, "tx", ti++);
2497 /* skip this unused q_vector */
2500 err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
2501 q_vector->name, q_vector);
2503 e_err(probe, "request_irq failed for MSIX interrupt "
2504 "Error: %d\n", err);
2505 goto free_queue_irqs;
2507 /* If Flow Director is enabled, set interrupt affinity */
2508 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2509 /* assign the mask for this irq */
2510 irq_set_affinity_hint(entry->vector,
2511 &q_vector->affinity_mask);
2515 err = request_irq(adapter->msix_entries[vector].vector,
2516 ixgbe_msix_other, 0, netdev->name, adapter);
2518 e_err(probe, "request_irq for msix_other failed: %d\n", err);
2519 goto free_queue_irqs;
2527 irq_set_affinity_hint(adapter->msix_entries[vector].vector,
2529 free_irq(adapter->msix_entries[vector].vector,
2530 adapter->q_vector[vector]);
2532 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2533 pci_disable_msix(adapter->pdev);
2534 kfree(adapter->msix_entries);
2535 adapter->msix_entries = NULL;
2540 * ixgbe_intr - legacy mode Interrupt Handler
2541 * @irq: interrupt number
2542 * @data: pointer to a network interface device structure
2544 static irqreturn_t ixgbe_intr(int irq, void *data)
2546 struct ixgbe_adapter *adapter = data;
2547 struct ixgbe_hw *hw = &adapter->hw;
2548 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2552 * Workaround for silicon errata #26 on 82598. Mask the interrupt
2553 * before the read of EICR.
2555 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2557 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2558 * therefore no explicit interrupt disable is necessary */
2559 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2562 * shared interrupt alert!
2563 * make sure interrupts are enabled because the read will
2564 * have disabled interrupts due to EIAM
2565 * finish the workaround of silicon errata on 82598. Unmask
2566 * the interrupt that we masked before the EICR read.
2568 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2569 ixgbe_irq_enable(adapter, true, true);
2570 return IRQ_NONE; /* Not our interrupt */
2573 if (eicr & IXGBE_EICR_LSC)
2574 ixgbe_check_lsc(adapter);
2576 switch (hw->mac.type) {
2577 case ixgbe_mac_82599EB:
2578 ixgbe_check_sfp_event(adapter, eicr);
2580 case ixgbe_mac_X540:
2581 if (eicr & IXGBE_EICR_ECC)
2582 e_info(link, "Received unrecoverable ECC err, please "
2584 ixgbe_check_overtemp_event(adapter, eicr);
2590 ixgbe_check_fan_failure(adapter, eicr);
2591 #ifdef CONFIG_IXGBE_PTP
2592 if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
2593 ixgbe_ptp_check_pps_event(adapter, eicr);
2596 /* would disable interrupts here but EIAM disabled it */
2597 napi_schedule(&q_vector->napi);
2600 * re-enable link(maybe) and non-queue interrupts, no flush.
2601 * ixgbe_poll will re-enable the queue interrupts
2603 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2604 ixgbe_irq_enable(adapter, false, false);
2610 * ixgbe_request_irq - initialize interrupts
2611 * @adapter: board private structure
2613 * Attempts to configure interrupts using the best available
2614 * capabilities of the hardware and kernel.
2616 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2618 struct net_device *netdev = adapter->netdev;
2621 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2622 err = ixgbe_request_msix_irqs(adapter);
2623 else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
2624 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2625 netdev->name, adapter);
2627 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2628 netdev->name, adapter);
2631 e_err(probe, "request_irq failed, Error %d\n", err);
2636 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2640 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
2641 free_irq(adapter->pdev->irq, adapter);
2645 for (vector = 0; vector < adapter->num_q_vectors; vector++) {
2646 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2647 struct msix_entry *entry = &adapter->msix_entries[vector];
2649 /* free only the irqs that were actually requested */
2650 if (!q_vector->rx.ring && !q_vector->tx.ring)
2653 /* clear the affinity_mask in the IRQ descriptor */
2654 irq_set_affinity_hint(entry->vector, NULL);
2656 free_irq(entry->vector, q_vector);
2659 free_irq(adapter->msix_entries[vector++].vector, adapter);
2663 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2664 * @adapter: board private structure
2666 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2668 switch (adapter->hw.mac.type) {
2669 case ixgbe_mac_82598EB:
2670 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2672 case ixgbe_mac_82599EB:
2673 case ixgbe_mac_X540:
2674 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2675 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2676 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2681 IXGBE_WRITE_FLUSH(&adapter->hw);
2682 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2685 for (vector = 0; vector < adapter->num_q_vectors; vector++)
2686 synchronize_irq(adapter->msix_entries[vector].vector);
2688 synchronize_irq(adapter->msix_entries[vector++].vector);
2690 synchronize_irq(adapter->pdev->irq);
2695 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2698 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2700 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2703 if (adapter->rx_itr_setting == 1)
2704 q_vector->itr = IXGBE_20K_ITR;
2706 q_vector->itr = adapter->rx_itr_setting;
2708 ixgbe_write_eitr(q_vector);
2710 ixgbe_set_ivar(adapter, 0, 0, 0);
2711 ixgbe_set_ivar(adapter, 1, 0, 0);
2713 e_info(hw, "Legacy interrupt IVAR setup done\n");
2717 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2718 * @adapter: board private structure
2719 * @ring: structure containing ring specific data
2721 * Configure the Tx descriptor ring after a reset.
2723 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2724 struct ixgbe_ring *ring)
2726 struct ixgbe_hw *hw = &adapter->hw;
2727 u64 tdba = ring->dma;
2729 u32 txdctl = IXGBE_TXDCTL_ENABLE;
2730 u8 reg_idx = ring->reg_idx;
2732 /* disable queue to avoid issues while updating state */
2733 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
2734 IXGBE_WRITE_FLUSH(hw);
2736 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2737 (tdba & DMA_BIT_MASK(32)));
2738 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2739 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2740 ring->count * sizeof(union ixgbe_adv_tx_desc));
2741 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2742 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2743 ring->tail = hw->hw_addr + IXGBE_TDT(reg_idx);
2746 * set WTHRESH to encourage burst writeback, it should not be set
2747 * higher than 1 when ITR is 0 as it could cause false TX hangs
2749 * In order to avoid issues WTHRESH + PTHRESH should always be equal
2750 * to or less than the number of on chip descriptors, which is
2753 if (!ring->q_vector || (ring->q_vector->itr < 8))
2754 txdctl |= (1 << 16); /* WTHRESH = 1 */
2756 txdctl |= (8 << 16); /* WTHRESH = 8 */
2759 * Setting PTHRESH to 32 both improves performance
2760 * and avoids a TX hang with DFP enabled
2762 txdctl |= (1 << 8) | /* HTHRESH = 1 */
2763 32; /* PTHRESH = 32 */
2765 /* reinitialize flowdirector state */
2766 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2767 ring->atr_sample_rate = adapter->atr_sample_rate;
2768 ring->atr_count = 0;
2769 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
2771 ring->atr_sample_rate = 0;
2774 clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
2777 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
2779 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2780 if (hw->mac.type == ixgbe_mac_82598EB &&
2781 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2784 /* poll to verify queue is enabled */
2786 usleep_range(1000, 2000);
2787 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2788 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
2790 e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
2793 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
2795 struct ixgbe_hw *hw = &adapter->hw;
2797 u8 tcs = netdev_get_num_tc(adapter->netdev);
2799 if (hw->mac.type == ixgbe_mac_82598EB)
2802 /* disable the arbiter while setting MTQC */
2803 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2804 rttdcs |= IXGBE_RTTDCS_ARBDIS;
2805 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2807 /* set transmit pool layout */
2808 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
2809 mtqc = IXGBE_MTQC_VT_ENA;
2811 mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
2813 mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
2814 else if (adapter->ring_feature[RING_F_RSS].indices == 4)
2815 mtqc |= IXGBE_MTQC_32VF;
2817 mtqc |= IXGBE_MTQC_64VF;
2820 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
2822 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
2824 mtqc = IXGBE_MTQC_64Q_1PB;
2827 IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
2829 /* Enable Security TX Buffer IFG for multiple pb */
2831 u32 sectx = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
2832 sectx |= IXGBE_SECTX_DCB;
2833 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, sectx);
2836 /* re-enable the arbiter */
2837 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2838 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2842 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2843 * @adapter: board private structure
2845 * Configure the Tx unit of the MAC after a reset.
2847 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2849 struct ixgbe_hw *hw = &adapter->hw;
2853 ixgbe_setup_mtqc(adapter);
2855 if (hw->mac.type != ixgbe_mac_82598EB) {
2856 /* DMATXCTL.EN must be before Tx queues are enabled */
2857 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2858 dmatxctl |= IXGBE_DMATXCTL_TE;
2859 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2862 /* Setup the HW Tx Head and Tail descriptor pointers */
2863 for (i = 0; i < adapter->num_tx_queues; i++)
2864 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
2867 static void ixgbe_enable_rx_drop(struct ixgbe_adapter *adapter,
2868 struct ixgbe_ring *ring)
2870 struct ixgbe_hw *hw = &adapter->hw;
2871 u8 reg_idx = ring->reg_idx;
2872 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
2874 srrctl |= IXGBE_SRRCTL_DROP_EN;
2876 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
2879 static void ixgbe_disable_rx_drop(struct ixgbe_adapter *adapter,
2880 struct ixgbe_ring *ring)
2882 struct ixgbe_hw *hw = &adapter->hw;
2883 u8 reg_idx = ring->reg_idx;
2884 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
2886 srrctl &= ~IXGBE_SRRCTL_DROP_EN;
2888 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
2891 #ifdef CONFIG_IXGBE_DCB
2892 void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
2894 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
2898 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
2900 if (adapter->ixgbe_ieee_pfc)
2901 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
2904 * We should set the drop enable bit if:
2907 * Number of Rx queues > 1 and flow control is disabled
2909 * This allows us to avoid head of line blocking for security
2910 * and performance reasons.
2912 if (adapter->num_vfs || (adapter->num_rx_queues > 1 &&
2913 !(adapter->hw.fc.current_mode & ixgbe_fc_tx_pause) && !pfc_en)) {
2914 for (i = 0; i < adapter->num_rx_queues; i++)
2915 ixgbe_enable_rx_drop(adapter, adapter->rx_ring[i]);
2917 for (i = 0; i < adapter->num_rx_queues; i++)
2918 ixgbe_disable_rx_drop(adapter, adapter->rx_ring[i]);
2922 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2924 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2925 struct ixgbe_ring *rx_ring)
2927 struct ixgbe_hw *hw = &adapter->hw;
2929 u8 reg_idx = rx_ring->reg_idx;
2931 if (hw->mac.type == ixgbe_mac_82598EB) {
2932 u16 mask = adapter->ring_feature[RING_F_RSS].mask;
2935 * if VMDq is not active we must program one srrctl register
2936 * per RSS queue since we have enabled RDRXCTL.MVMEN
2941 /* configure header buffer length, needed for RSC */
2942 srrctl = IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
2944 /* configure the packet buffer length */
2945 srrctl |= ixgbe_rx_bufsz(rx_ring) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2947 /* configure descriptor type */
2948 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2950 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
2953 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2955 struct ixgbe_hw *hw = &adapter->hw;
2956 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2957 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2958 0x6A3E67EA, 0x14364D17, 0x3BED200D};
2959 u32 mrqc = 0, reta = 0;
2962 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
2965 * Program table for at least 2 queues w/ SR-IOV so that VFs can
2966 * make full use of any rings they may have. We will use the
2967 * PSRTYPE register to control how many rings we use within the PF.
2969 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) && (rss_i < 2))
2972 /* Fill out hash function seeds */
2973 for (i = 0; i < 10; i++)
2974 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2976 /* Fill out redirection table */
2977 for (i = 0, j = 0; i < 128; i++, j++) {
2980 /* reta = 4-byte sliding window of
2981 * 0x00..(indices-1)(indices-1)00..etc. */
2982 reta = (reta << 8) | (j * 0x11);
2984 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2987 /* Disable indicating checksum in descriptor, enables RSS hash */
2988 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2989 rxcsum |= IXGBE_RXCSUM_PCSD;
2990 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2992 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2993 if (adapter->ring_feature[RING_F_RSS].mask)
2994 mrqc = IXGBE_MRQC_RSSEN;
2996 u8 tcs = netdev_get_num_tc(adapter->netdev);
2998 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3000 mrqc = IXGBE_MRQC_VMDQRT8TCEN; /* 8 TCs */
3002 mrqc = IXGBE_MRQC_VMDQRT4TCEN; /* 4 TCs */
3003 else if (adapter->ring_feature[RING_F_RSS].indices == 4)
3004 mrqc = IXGBE_MRQC_VMDQRSS32EN;
3006 mrqc = IXGBE_MRQC_VMDQRSS64EN;
3009 mrqc = IXGBE_MRQC_RTRSS8TCEN;
3011 mrqc = IXGBE_MRQC_RTRSS4TCEN;
3013 mrqc = IXGBE_MRQC_RSSEN;
3017 /* Perform hash on these packet types */
3018 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4 |
3019 IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
3020 IXGBE_MRQC_RSS_FIELD_IPV6 |
3021 IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
3023 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP)
3024 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
3025 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP)
3026 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
3028 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3032 * ixgbe_configure_rscctl - enable RSC for the indicated ring
3033 * @adapter: address of board private structure
3034 * @index: index of ring to set
3036 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
3037 struct ixgbe_ring *ring)
3039 struct ixgbe_hw *hw = &adapter->hw;
3041 u8 reg_idx = ring->reg_idx;
3043 if (!ring_is_rsc_enabled(ring))
3046 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
3047 rscctrl |= IXGBE_RSCCTL_RSCEN;
3049 * we must limit the number of descriptors so that the
3050 * total size of max desc * buf_len is not greater
3053 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
3054 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
3057 #define IXGBE_MAX_RX_DESC_POLL 10
3058 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
3059 struct ixgbe_ring *ring)
3061 struct ixgbe_hw *hw = &adapter->hw;
3062 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3064 u8 reg_idx = ring->reg_idx;
3066 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3067 if (hw->mac.type == ixgbe_mac_82598EB &&
3068 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3072 usleep_range(1000, 2000);
3073 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3074 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
3077 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
3078 "the polling period\n", reg_idx);
3082 void ixgbe_disable_rx_queue(struct ixgbe_adapter *adapter,
3083 struct ixgbe_ring *ring)
3085 struct ixgbe_hw *hw = &adapter->hw;
3086 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3088 u8 reg_idx = ring->reg_idx;
3090 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3091 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
3093 /* write value back with RXDCTL.ENABLE bit cleared */
3094 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3096 if (hw->mac.type == ixgbe_mac_82598EB &&
3097 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3100 /* the hardware may take up to 100us to really disable the rx queue */
3103 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3104 } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
3107 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not cleared within "
3108 "the polling period\n", reg_idx);
3112 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
3113 struct ixgbe_ring *ring)
3115 struct ixgbe_hw *hw = &adapter->hw;
3116 u64 rdba = ring->dma;
3118 u8 reg_idx = ring->reg_idx;
3120 /* disable queue to avoid issues while updating state */
3121 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3122 ixgbe_disable_rx_queue(adapter, ring);
3124 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
3125 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
3126 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
3127 ring->count * sizeof(union ixgbe_adv_rx_desc));
3128 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
3129 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
3130 ring->tail = hw->hw_addr + IXGBE_RDT(reg_idx);
3132 ixgbe_configure_srrctl(adapter, ring);
3133 ixgbe_configure_rscctl(adapter, ring);
3135 /* If operating in IOV mode set RLPML for X540 */
3136 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
3137 hw->mac.type == ixgbe_mac_X540) {
3138 rxdctl &= ~IXGBE_RXDCTL_RLPMLMASK;
3139 rxdctl |= ((ring->netdev->mtu + ETH_HLEN +
3140 ETH_FCS_LEN + VLAN_HLEN) | IXGBE_RXDCTL_RLPML_EN);
3143 if (hw->mac.type == ixgbe_mac_82598EB) {
3145 * enable cache line friendly hardware writes:
3146 * PTHRESH=32 descriptors (half the internal cache),
3147 * this also removes ugly rx_no_buffer_count increment
3148 * HTHRESH=4 descriptors (to minimize latency on fetch)
3149 * WTHRESH=8 burst writeback up to two cache lines
3151 rxdctl &= ~0x3FFFFF;
3155 /* enable receive descriptor ring */
3156 rxdctl |= IXGBE_RXDCTL_ENABLE;
3157 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3159 ixgbe_rx_desc_queue_enable(adapter, ring);
3160 ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
3163 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
3165 struct ixgbe_hw *hw = &adapter->hw;
3166 int rss_i = adapter->ring_feature[RING_F_RSS].indices;
3169 /* PSRTYPE must be initialized in non 82598 adapters */
3170 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
3171 IXGBE_PSRTYPE_UDPHDR |
3172 IXGBE_PSRTYPE_IPV4HDR |
3173 IXGBE_PSRTYPE_L2HDR |
3174 IXGBE_PSRTYPE_IPV6HDR;
3176 if (hw->mac.type == ixgbe_mac_82598EB)
3184 for (p = 0; p < adapter->num_rx_pools; p++)
3185 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(p)),
3189 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
3191 struct ixgbe_hw *hw = &adapter->hw;
3192 u32 reg_offset, vf_shift;
3193 u32 gcr_ext, vmdctl;
3196 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
3199 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
3200 vmdctl |= IXGBE_VMD_CTL_VMDQ_EN;
3201 vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
3202 vmdctl |= VMDQ_P(0) << IXGBE_VT_CTL_POOL_SHIFT;
3203 vmdctl |= IXGBE_VT_CTL_REPLEN;
3204 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
3206 vf_shift = VMDQ_P(0) % 32;
3207 reg_offset = (VMDQ_P(0) >= 32) ? 1 : 0;
3209 /* Enable only the PF's pool for Tx/Rx */
3210 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (~0) << vf_shift);
3211 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), reg_offset - 1);
3212 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (~0) << vf_shift);
3213 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), reg_offset - 1);
3214 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3216 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
3217 hw->mac.ops.set_vmdq(hw, 0, VMDQ_P(0));
3220 * Set up VF register offsets for selected VT Mode,
3221 * i.e. 32 or 64 VFs for SR-IOV
3223 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
3224 case IXGBE_82599_VMDQ_8Q_MASK:
3225 gcr_ext = IXGBE_GCR_EXT_VT_MODE_16;
3227 case IXGBE_82599_VMDQ_4Q_MASK:
3228 gcr_ext = IXGBE_GCR_EXT_VT_MODE_32;
3231 gcr_ext = IXGBE_GCR_EXT_VT_MODE_64;
3235 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
3237 /* enable Tx loopback for VF/PF communication */
3238 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3240 /* Enable MAC Anti-Spoofing */
3241 hw->mac.ops.set_mac_anti_spoofing(hw, (adapter->num_vfs != 0),
3243 /* For VFs that have spoof checking turned off */
3244 for (i = 0; i < adapter->num_vfs; i++) {
3245 if (!adapter->vfinfo[i].spoofchk_enabled)
3246 ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i, false);
3250 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
3252 struct ixgbe_hw *hw = &adapter->hw;
3253 struct net_device *netdev = adapter->netdev;
3254 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3255 struct ixgbe_ring *rx_ring;
3260 /* adjust max frame to be able to do baby jumbo for FCoE */
3261 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
3262 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
3263 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3265 #endif /* IXGBE_FCOE */
3266 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
3267 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
3268 mhadd &= ~IXGBE_MHADD_MFS_MASK;
3269 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
3271 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
3274 /* MHADD will allow an extra 4 bytes past for vlan tagged frames */
3275 max_frame += VLAN_HLEN;
3277 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
3278 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
3279 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
3280 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
3283 * Setup the HW Rx Head and Tail Descriptor Pointers and
3284 * the Base and Length of the Rx Descriptor Ring
3286 for (i = 0; i < adapter->num_rx_queues; i++) {
3287 rx_ring = adapter->rx_ring[i];
3288 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
3289 set_ring_rsc_enabled(rx_ring);
3291 clear_ring_rsc_enabled(rx_ring);
3295 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
3297 struct ixgbe_hw *hw = &adapter->hw;
3298 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
3300 switch (hw->mac.type) {
3301 case ixgbe_mac_82598EB:
3303 * For VMDq support of different descriptor types or
3304 * buffer sizes through the use of multiple SRRCTL
3305 * registers, RDRXCTL.MVMEN must be set to 1
3307 * also, the manual doesn't mention it clearly but DCA hints
3308 * will only use queue 0's tags unless this bit is set. Side
3309 * effects of setting this bit are only that SRRCTL must be
3310 * fully programmed [0..15]
3312 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
3314 case ixgbe_mac_82599EB:
3315 case ixgbe_mac_X540:
3316 /* Disable RSC for ACK packets */
3317 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
3318 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
3319 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3320 /* hardware requires some bits to be set by default */
3321 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3322 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3325 /* We should do nothing since we don't know this hardware */
3329 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3333 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3334 * @adapter: board private structure
3336 * Configure the Rx unit of the MAC after a reset.
3338 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3340 struct ixgbe_hw *hw = &adapter->hw;
3344 /* disable receives while setting up the descriptors */
3345 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3346 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3348 ixgbe_setup_psrtype(adapter);
3349 ixgbe_setup_rdrxctl(adapter);
3351 /* Program registers for the distribution of queues */
3352 ixgbe_setup_mrqc(adapter);
3354 /* set_rx_buffer_len must be called before ring initialization */
3355 ixgbe_set_rx_buffer_len(adapter);
3358 * Setup the HW Rx Head and Tail Descriptor Pointers and
3359 * the Base and Length of the Rx Descriptor Ring
3361 for (i = 0; i < adapter->num_rx_queues; i++)
3362 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3364 /* disable drop enable for 82598 parts */
3365 if (hw->mac.type == ixgbe_mac_82598EB)
3366 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3368 /* enable all receives */
3369 rxctrl |= IXGBE_RXCTRL_RXEN;
3370 hw->mac.ops.enable_rx_dma(hw, rxctrl);
3373 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3375 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3376 struct ixgbe_hw *hw = &adapter->hw;
3378 /* add VID to filter table */
3379 hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), true);
3380 set_bit(vid, adapter->active_vlans);
3385 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3387 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3388 struct ixgbe_hw *hw = &adapter->hw;
3390 /* remove VID from filter table */
3391 hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), false);
3392 clear_bit(vid, adapter->active_vlans);
3398 * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3399 * @adapter: driver data
3401 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
3403 struct ixgbe_hw *hw = &adapter->hw;
3406 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3407 vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
3408 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3412 * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3413 * @adapter: driver data
3415 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
3417 struct ixgbe_hw *hw = &adapter->hw;
3420 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3421 vlnctrl |= IXGBE_VLNCTRL_VFE;
3422 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3423 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3427 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3428 * @adapter: driver data
3430 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3432 struct ixgbe_hw *hw = &adapter->hw;
3436 switch (hw->mac.type) {
3437 case ixgbe_mac_82598EB:
3438 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3439 vlnctrl &= ~IXGBE_VLNCTRL_VME;
3440 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3442 case ixgbe_mac_82599EB:
3443 case ixgbe_mac_X540:
3444 for (i = 0; i < adapter->num_rx_queues; i++) {
3445 j = adapter->rx_ring[i]->reg_idx;
3446 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3447 vlnctrl &= ~IXGBE_RXDCTL_VME;
3448 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3457 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3458 * @adapter: driver data
3460 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3462 struct ixgbe_hw *hw = &adapter->hw;
3466 switch (hw->mac.type) {
3467 case ixgbe_mac_82598EB:
3468 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3469 vlnctrl |= IXGBE_VLNCTRL_VME;
3470 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3472 case ixgbe_mac_82599EB:
3473 case ixgbe_mac_X540:
3474 for (i = 0; i < adapter->num_rx_queues; i++) {
3475 j = adapter->rx_ring[i]->reg_idx;
3476 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3477 vlnctrl |= IXGBE_RXDCTL_VME;
3478 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3486 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3490 ixgbe_vlan_rx_add_vid(adapter->netdev, 0);
3492 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3493 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
3497 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3498 * @netdev: network interface device structure
3500 * Writes unicast address list to the RAR table.
3501 * Returns: -ENOMEM on failure/insufficient address space
3502 * 0 on no addresses written
3503 * X on writing X addresses to the RAR table
3505 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3507 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3508 struct ixgbe_hw *hw = &adapter->hw;
3509 unsigned int rar_entries = hw->mac.num_rar_entries - 1;
3512 /* In SR-IOV mode significantly less RAR entries are available */
3513 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3514 rar_entries = IXGBE_MAX_PF_MACVLANS - 1;
3516 /* return ENOMEM indicating insufficient memory for addresses */
3517 if (netdev_uc_count(netdev) > rar_entries)
3520 if (!netdev_uc_empty(netdev)) {
3521 struct netdev_hw_addr *ha;
3522 /* return error if we do not support writing to RAR table */
3523 if (!hw->mac.ops.set_rar)
3526 netdev_for_each_uc_addr(ha, netdev) {
3529 hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3530 VMDQ_P(0), IXGBE_RAH_AV);
3534 /* write the addresses in reverse order to avoid write combining */
3535 for (; rar_entries > 0 ; rar_entries--)
3536 hw->mac.ops.clear_rar(hw, rar_entries);
3542 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3543 * @netdev: network interface device structure
3545 * The set_rx_method entry point is called whenever the unicast/multicast
3546 * address list or the network interface flags are updated. This routine is
3547 * responsible for configuring the hardware for proper unicast, multicast and
3550 void ixgbe_set_rx_mode(struct net_device *netdev)
3552 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3553 struct ixgbe_hw *hw = &adapter->hw;
3554 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3557 /* Check for Promiscuous and All Multicast modes */
3559 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3561 /* set all bits that we expect to always be set */
3562 fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
3563 fctrl |= IXGBE_FCTRL_BAM;
3564 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3565 fctrl |= IXGBE_FCTRL_PMCF;
3567 /* clear the bits we are changing the status of */
3568 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3570 if (netdev->flags & IFF_PROMISC) {
3571 hw->addr_ctrl.user_set_promisc = true;
3572 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3573 vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3574 /* don't hardware filter vlans in promisc mode */
3575 ixgbe_vlan_filter_disable(adapter);
3577 if (netdev->flags & IFF_ALLMULTI) {
3578 fctrl |= IXGBE_FCTRL_MPE;
3579 vmolr |= IXGBE_VMOLR_MPE;
3582 * Write addresses to the MTA, if the attempt fails
3583 * then we should just turn on promiscuous mode so
3584 * that we can at least receive multicast traffic
3586 hw->mac.ops.update_mc_addr_list(hw, netdev);
3587 vmolr |= IXGBE_VMOLR_ROMPE;
3589 ixgbe_vlan_filter_enable(adapter);
3590 hw->addr_ctrl.user_set_promisc = false;
3594 * Write addresses to available RAR registers, if there is not
3595 * sufficient space to store all the addresses then enable
3596 * unicast promiscuous mode
3598 count = ixgbe_write_uc_addr_list(netdev);
3600 fctrl |= IXGBE_FCTRL_UPE;
3601 vmolr |= IXGBE_VMOLR_ROPE;
3604 if (adapter->num_vfs)
3605 ixgbe_restore_vf_multicasts(adapter);
3607 if (hw->mac.type != ixgbe_mac_82598EB) {
3608 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(VMDQ_P(0))) &
3609 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3611 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(0)), vmolr);
3614 /* This is useful for sniffing bad packets. */
3615 if (adapter->netdev->features & NETIF_F_RXALL) {
3616 /* UPE and MPE will be handled by normal PROMISC logic
3617 * in e1000e_set_rx_mode */
3618 fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
3619 IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
3620 IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
3622 fctrl &= ~(IXGBE_FCTRL_DPF);
3623 /* NOTE: VLAN filtering is disabled by setting PROMISC */
3626 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3628 if (netdev->features & NETIF_F_HW_VLAN_RX)
3629 ixgbe_vlan_strip_enable(adapter);
3631 ixgbe_vlan_strip_disable(adapter);
3634 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3638 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
3639 napi_enable(&adapter->q_vector[q_idx]->napi);
3642 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3646 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
3647 napi_disable(&adapter->q_vector[q_idx]->napi);
3650 #ifdef CONFIG_IXGBE_DCB
3652 * ixgbe_configure_dcb - Configure DCB hardware
3653 * @adapter: ixgbe adapter struct
3655 * This is called by the driver on open to configure the DCB hardware.
3656 * This is also called by the gennetlink interface when reconfiguring
3659 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3661 struct ixgbe_hw *hw = &adapter->hw;
3662 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3664 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3665 if (hw->mac.type == ixgbe_mac_82598EB)
3666 netif_set_gso_max_size(adapter->netdev, 65536);
3670 if (hw->mac.type == ixgbe_mac_82598EB)
3671 netif_set_gso_max_size(adapter->netdev, 32768);
3674 if (adapter->netdev->features & NETIF_F_FCOE_MTU)
3675 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
3678 /* reconfigure the hardware */
3679 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
3680 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3682 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3684 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
3685 } else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
3686 ixgbe_dcb_hw_ets(&adapter->hw,
3687 adapter->ixgbe_ieee_ets,
3689 ixgbe_dcb_hw_pfc_config(&adapter->hw,
3690 adapter->ixgbe_ieee_pfc->pfc_en,
3691 adapter->ixgbe_ieee_ets->prio_tc);
3694 /* Enable RSS Hash per TC */
3695 if (hw->mac.type != ixgbe_mac_82598EB) {
3697 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices - 1;
3704 /* write msb to all 8 TCs in one write */
3705 IXGBE_WRITE_REG(hw, IXGBE_RQTC, msb * 0x11111111);
3710 /* Additional bittime to account for IXGBE framing */
3711 #define IXGBE_ETH_FRAMING 20
3714 * ixgbe_hpbthresh - calculate high water mark for flow control
3716 * @adapter: board private structure to calculate for
3717 * @pb: packet buffer to calculate
3719 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
3721 struct ixgbe_hw *hw = &adapter->hw;
3722 struct net_device *dev = adapter->netdev;
3723 int link, tc, kb, marker;
3726 /* Calculate max LAN frame size */
3727 tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
3730 /* FCoE traffic class uses FCOE jumbo frames */
3731 if ((dev->features & NETIF_F_FCOE_MTU) &&
3732 (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
3733 (pb == ixgbe_fcoe_get_tc(adapter)))
3734 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3737 /* Calculate delay value for device */
3738 switch (hw->mac.type) {
3739 case ixgbe_mac_X540:
3740 dv_id = IXGBE_DV_X540(link, tc);
3743 dv_id = IXGBE_DV(link, tc);
3747 /* Loopback switch introduces additional latency */
3748 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3749 dv_id += IXGBE_B2BT(tc);
3751 /* Delay value is calculated in bit times convert to KB */
3752 kb = IXGBE_BT2KB(dv_id);
3753 rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
3755 marker = rx_pba - kb;
3757 /* It is possible that the packet buffer is not large enough
3758 * to provide required headroom. In this case throw an error
3759 * to user and a do the best we can.
3762 e_warn(drv, "Packet Buffer(%i) can not provide enough"
3763 "headroom to support flow control."
3764 "Decrease MTU or number of traffic classes\n", pb);
3772 * ixgbe_lpbthresh - calculate low water mark for for flow control
3774 * @adapter: board private structure to calculate for
3775 * @pb: packet buffer to calculate
3777 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter)
3779 struct ixgbe_hw *hw = &adapter->hw;
3780 struct net_device *dev = adapter->netdev;
3784 /* Calculate max LAN frame size */
3785 tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
3787 /* Calculate delay value for device */
3788 switch (hw->mac.type) {
3789 case ixgbe_mac_X540:
3790 dv_id = IXGBE_LOW_DV_X540(tc);
3793 dv_id = IXGBE_LOW_DV(tc);
3797 /* Delay value is calculated in bit times convert to KB */
3798 return IXGBE_BT2KB(dv_id);
3802 * ixgbe_pbthresh_setup - calculate and setup high low water marks
3804 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
3806 struct ixgbe_hw *hw = &adapter->hw;
3807 int num_tc = netdev_get_num_tc(adapter->netdev);
3813 hw->fc.low_water = ixgbe_lpbthresh(adapter);
3815 for (i = 0; i < num_tc; i++) {
3816 hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
3818 /* Low water marks must not be larger than high water marks */
3819 if (hw->fc.low_water > hw->fc.high_water[i])
3820 hw->fc.low_water = 0;
3824 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
3826 struct ixgbe_hw *hw = &adapter->hw;
3828 u8 tc = netdev_get_num_tc(adapter->netdev);
3830 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3831 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3832 hdrm = 32 << adapter->fdir_pballoc;
3836 hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
3837 ixgbe_pbthresh_setup(adapter);
3840 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
3842 struct ixgbe_hw *hw = &adapter->hw;
3843 struct hlist_node *node, *node2;
3844 struct ixgbe_fdir_filter *filter;
3846 spin_lock(&adapter->fdir_perfect_lock);
3848 if (!hlist_empty(&adapter->fdir_filter_list))
3849 ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
3851 hlist_for_each_entry_safe(filter, node, node2,
3852 &adapter->fdir_filter_list, fdir_node) {
3853 ixgbe_fdir_write_perfect_filter_82599(hw,
3856 (filter->action == IXGBE_FDIR_DROP_QUEUE) ?
3857 IXGBE_FDIR_DROP_QUEUE :
3858 adapter->rx_ring[filter->action]->reg_idx);
3861 spin_unlock(&adapter->fdir_perfect_lock);
3864 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3866 struct ixgbe_hw *hw = &adapter->hw;
3868 ixgbe_configure_pb(adapter);
3869 #ifdef CONFIG_IXGBE_DCB
3870 ixgbe_configure_dcb(adapter);
3873 * We must restore virtualization before VLANs or else
3874 * the VLVF registers will not be populated
3876 ixgbe_configure_virtualization(adapter);
3878 ixgbe_set_rx_mode(adapter->netdev);
3879 ixgbe_restore_vlan(adapter);
3881 switch (hw->mac.type) {
3882 case ixgbe_mac_82599EB:
3883 case ixgbe_mac_X540:
3884 hw->mac.ops.disable_rx_buff(hw);
3890 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3891 ixgbe_init_fdir_signature_82599(&adapter->hw,
3892 adapter->fdir_pballoc);
3893 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3894 ixgbe_init_fdir_perfect_82599(&adapter->hw,
3895 adapter->fdir_pballoc);
3896 ixgbe_fdir_filter_restore(adapter);
3899 switch (hw->mac.type) {
3900 case ixgbe_mac_82599EB:
3901 case ixgbe_mac_X540:
3902 hw->mac.ops.enable_rx_buff(hw);
3909 /* configure FCoE L2 filters, redirection table, and Rx control */
3910 ixgbe_configure_fcoe(adapter);
3912 #endif /* IXGBE_FCOE */
3913 ixgbe_configure_tx(adapter);
3914 ixgbe_configure_rx(adapter);
3917 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3919 switch (hw->phy.type) {
3920 case ixgbe_phy_sfp_avago:
3921 case ixgbe_phy_sfp_ftl:
3922 case ixgbe_phy_sfp_intel:
3923 case ixgbe_phy_sfp_unknown:
3924 case ixgbe_phy_sfp_passive_tyco:
3925 case ixgbe_phy_sfp_passive_unknown:
3926 case ixgbe_phy_sfp_active_unknown:
3927 case ixgbe_phy_sfp_ftl_active:
3930 if (hw->mac.type == ixgbe_mac_82598EB)
3938 * ixgbe_sfp_link_config - set up SFP+ link
3939 * @adapter: pointer to private adapter struct
3941 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3944 * We are assuming the worst case scenario here, and that
3945 * is that an SFP was inserted/removed after the reset
3946 * but before SFP detection was enabled. As such the best
3947 * solution is to just start searching as soon as we start
3949 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3950 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
3952 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
3956 * ixgbe_non_sfp_link_config - set up non-SFP+ link
3957 * @hw: pointer to private hardware struct
3959 * Returns 0 on success, negative on failure
3961 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3964 bool negotiation, link_up = false;
3965 u32 ret = IXGBE_ERR_LINK_SETUP;
3967 if (hw->mac.ops.check_link)
3968 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3973 autoneg = hw->phy.autoneg_advertised;
3974 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
3975 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
3980 if (hw->mac.ops.setup_link)
3981 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
3986 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
3988 struct ixgbe_hw *hw = &adapter->hw;
3991 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3992 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
3994 gpie |= IXGBE_GPIE_EIAME;
3996 * use EIAM to auto-mask when MSI-X interrupt is asserted
3997 * this saves a register write for every interrupt
3999 switch (hw->mac.type) {
4000 case ixgbe_mac_82598EB:
4001 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
4003 case ixgbe_mac_82599EB:
4004 case ixgbe_mac_X540:
4006 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
4007 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
4011 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
4012 * specifically only auto mask tx and rx interrupts */
4013 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
4016 /* XXX: to interrupt immediately for EICS writes, enable this */
4017 /* gpie |= IXGBE_GPIE_EIMEN; */
4019 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
4020 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
4022 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
4023 case IXGBE_82599_VMDQ_8Q_MASK:
4024 gpie |= IXGBE_GPIE_VTMODE_16;
4026 case IXGBE_82599_VMDQ_4Q_MASK:
4027 gpie |= IXGBE_GPIE_VTMODE_32;
4030 gpie |= IXGBE_GPIE_VTMODE_64;
4035 /* Enable Thermal over heat sensor interrupt */
4036 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
4037 switch (adapter->hw.mac.type) {
4038 case ixgbe_mac_82599EB:
4039 gpie |= IXGBE_SDP0_GPIEN;
4041 case ixgbe_mac_X540:
4042 gpie |= IXGBE_EIMS_TS;
4049 /* Enable fan failure interrupt */
4050 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
4051 gpie |= IXGBE_SDP1_GPIEN;
4053 if (hw->mac.type == ixgbe_mac_82599EB) {
4054 gpie |= IXGBE_SDP1_GPIEN;
4055 gpie |= IXGBE_SDP2_GPIEN;
4058 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
4061 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
4063 struct ixgbe_hw *hw = &adapter->hw;
4067 ixgbe_get_hw_control(adapter);
4068 ixgbe_setup_gpie(adapter);
4070 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4071 ixgbe_configure_msix(adapter);
4073 ixgbe_configure_msi_and_legacy(adapter);
4075 /* enable the optics for both mult-speed fiber and 82599 SFP+ fiber */
4076 if (hw->mac.ops.enable_tx_laser &&
4077 ((hw->phy.multispeed_fiber) ||
4078 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
4079 (hw->mac.type == ixgbe_mac_82599EB))))
4080 hw->mac.ops.enable_tx_laser(hw);
4082 clear_bit(__IXGBE_DOWN, &adapter->state);
4083 ixgbe_napi_enable_all(adapter);
4085 if (ixgbe_is_sfp(hw)) {
4086 ixgbe_sfp_link_config(adapter);
4088 err = ixgbe_non_sfp_link_config(hw);
4090 e_err(probe, "link_config FAILED %d\n", err);
4093 /* clear any pending interrupts, may auto mask */
4094 IXGBE_READ_REG(hw, IXGBE_EICR);
4095 ixgbe_irq_enable(adapter, true, true);
4098 * If this adapter has a fan, check to see if we had a failure
4099 * before we enabled the interrupt.
4101 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
4102 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
4103 if (esdp & IXGBE_ESDP_SDP1)
4104 e_crit(drv, "Fan has stopped, replace the adapter\n");
4107 /* enable transmits */
4108 netif_tx_start_all_queues(adapter->netdev);
4110 /* bring the link up in the watchdog, this could race with our first
4111 * link up interrupt but shouldn't be a problem */
4112 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
4113 adapter->link_check_timeout = jiffies;
4114 mod_timer(&adapter->service_timer, jiffies);
4116 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
4117 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
4118 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
4119 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
4122 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
4124 WARN_ON(in_interrupt());
4125 /* put off any impending NetWatchDogTimeout */
4126 adapter->netdev->trans_start = jiffies;
4128 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
4129 usleep_range(1000, 2000);
4130 ixgbe_down(adapter);
4132 * If SR-IOV enabled then wait a bit before bringing the adapter
4133 * back up to give the VFs time to respond to the reset. The
4134 * two second wait is based upon the watchdog timer cycle in
4137 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4140 clear_bit(__IXGBE_RESETTING, &adapter->state);
4143 void ixgbe_up(struct ixgbe_adapter *adapter)
4145 /* hardware has been reset, we need to reload some things */
4146 ixgbe_configure(adapter);
4148 ixgbe_up_complete(adapter);
4151 void ixgbe_reset(struct ixgbe_adapter *adapter)
4153 struct ixgbe_hw *hw = &adapter->hw;
4156 /* lock SFP init bit to prevent race conditions with the watchdog */
4157 while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
4158 usleep_range(1000, 2000);
4160 /* clear all SFP and link config related flags while holding SFP_INIT */
4161 adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
4162 IXGBE_FLAG2_SFP_NEEDS_RESET);
4163 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
4165 err = hw->mac.ops.init_hw(hw);
4168 case IXGBE_ERR_SFP_NOT_PRESENT:
4169 case IXGBE_ERR_SFP_NOT_SUPPORTED:
4171 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
4172 e_dev_err("master disable timed out\n");
4174 case IXGBE_ERR_EEPROM_VERSION:
4175 /* We are running on a pre-production device, log a warning */
4176 e_dev_warn("This device is a pre-production adapter/LOM. "
4177 "Please be aware there may be issues associated with "
4178 "your hardware. If you are experiencing problems "
4179 "please contact your Intel or hardware "
4180 "representative who provided you with this "
4184 e_dev_err("Hardware Error: %d\n", err);
4187 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
4189 /* reprogram the RAR[0] in case user changed it. */
4190 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, VMDQ_P(0), IXGBE_RAH_AV);
4192 /* update SAN MAC vmdq pool selection */
4193 if (hw->mac.san_mac_rar_index)
4194 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
4198 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
4199 * @rx_ring: ring to free buffers from
4201 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
4203 struct device *dev = rx_ring->dev;
4207 /* ring already cleared, nothing to do */
4208 if (!rx_ring->rx_buffer_info)
4211 /* Free all the Rx ring sk_buffs */
4212 for (i = 0; i < rx_ring->count; i++) {
4213 struct ixgbe_rx_buffer *rx_buffer;
4215 rx_buffer = &rx_ring->rx_buffer_info[i];
4216 if (rx_buffer->skb) {
4217 struct sk_buff *skb = rx_buffer->skb;
4218 if (IXGBE_CB(skb)->page_released) {
4221 ixgbe_rx_bufsz(rx_ring),
4223 IXGBE_CB(skb)->page_released = false;
4227 rx_buffer->skb = NULL;
4229 dma_unmap_page(dev, rx_buffer->dma,
4230 ixgbe_rx_pg_size(rx_ring),
4233 if (rx_buffer->page)
4234 __free_pages(rx_buffer->page,
4235 ixgbe_rx_pg_order(rx_ring));
4236 rx_buffer->page = NULL;
4239 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4240 memset(rx_ring->rx_buffer_info, 0, size);
4242 /* Zero out the descriptor ring */
4243 memset(rx_ring->desc, 0, rx_ring->size);
4245 rx_ring->next_to_alloc = 0;
4246 rx_ring->next_to_clean = 0;
4247 rx_ring->next_to_use = 0;
4251 * ixgbe_clean_tx_ring - Free Tx Buffers
4252 * @tx_ring: ring to be cleaned
4254 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
4256 struct ixgbe_tx_buffer *tx_buffer_info;
4260 /* ring already cleared, nothing to do */
4261 if (!tx_ring->tx_buffer_info)
4264 /* Free all the Tx ring sk_buffs */
4265 for (i = 0; i < tx_ring->count; i++) {
4266 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4267 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
4270 netdev_tx_reset_queue(txring_txq(tx_ring));
4272 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4273 memset(tx_ring->tx_buffer_info, 0, size);
4275 /* Zero out the descriptor ring */
4276 memset(tx_ring->desc, 0, tx_ring->size);
4278 tx_ring->next_to_use = 0;
4279 tx_ring->next_to_clean = 0;
4283 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
4284 * @adapter: board private structure
4286 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
4290 for (i = 0; i < adapter->num_rx_queues; i++)
4291 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
4295 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
4296 * @adapter: board private structure
4298 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
4302 for (i = 0; i < adapter->num_tx_queues; i++)
4303 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
4306 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
4308 struct hlist_node *node, *node2;
4309 struct ixgbe_fdir_filter *filter;
4311 spin_lock(&adapter->fdir_perfect_lock);
4313 hlist_for_each_entry_safe(filter, node, node2,
4314 &adapter->fdir_filter_list, fdir_node) {
4315 hlist_del(&filter->fdir_node);
4318 adapter->fdir_filter_count = 0;
4320 spin_unlock(&adapter->fdir_perfect_lock);
4323 void ixgbe_down(struct ixgbe_adapter *adapter)
4325 struct net_device *netdev = adapter->netdev;
4326 struct ixgbe_hw *hw = &adapter->hw;
4330 /* signal that we are down to the interrupt handler */
4331 set_bit(__IXGBE_DOWN, &adapter->state);
4333 /* disable receives */
4334 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4335 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
4337 /* disable all enabled rx queues */
4338 for (i = 0; i < adapter->num_rx_queues; i++)
4339 /* this call also flushes the previous write */
4340 ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]);
4342 usleep_range(10000, 20000);
4344 netif_tx_stop_all_queues(netdev);
4346 /* call carrier off first to avoid false dev_watchdog timeouts */
4347 netif_carrier_off(netdev);
4348 netif_tx_disable(netdev);
4350 ixgbe_irq_disable(adapter);
4352 ixgbe_napi_disable_all(adapter);
4354 adapter->flags2 &= ~(IXGBE_FLAG2_FDIR_REQUIRES_REINIT |
4355 IXGBE_FLAG2_RESET_REQUESTED);
4356 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
4358 del_timer_sync(&adapter->service_timer);
4360 if (adapter->num_vfs) {
4361 /* Clear EITR Select mapping */
4362 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
4364 /* Mark all the VFs as inactive */
4365 for (i = 0 ; i < adapter->num_vfs; i++)
4366 adapter->vfinfo[i].clear_to_send = false;
4368 /* ping all the active vfs to let them know we are going down */
4369 ixgbe_ping_all_vfs(adapter);
4371 /* Disable all VFTE/VFRE TX/RX */
4372 ixgbe_disable_tx_rx(adapter);
4375 /* disable transmits in the hardware now that interrupts are off */
4376 for (i = 0; i < adapter->num_tx_queues; i++) {
4377 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
4378 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
4381 /* Disable the Tx DMA engine on 82599 and X540 */
4382 switch (hw->mac.type) {
4383 case ixgbe_mac_82599EB:
4384 case ixgbe_mac_X540:
4385 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
4386 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
4387 ~IXGBE_DMATXCTL_TE));
4393 if (!pci_channel_offline(adapter->pdev))
4394 ixgbe_reset(adapter);
4396 /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
4397 if (hw->mac.ops.disable_tx_laser &&
4398 ((hw->phy.multispeed_fiber) ||
4399 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
4400 (hw->mac.type == ixgbe_mac_82599EB))))
4401 hw->mac.ops.disable_tx_laser(hw);
4403 ixgbe_clean_all_tx_rings(adapter);
4404 ixgbe_clean_all_rx_rings(adapter);
4406 #ifdef CONFIG_IXGBE_DCA
4407 /* since we reset the hardware DCA settings were cleared */
4408 ixgbe_setup_dca(adapter);
4413 * ixgbe_tx_timeout - Respond to a Tx Hang
4414 * @netdev: network interface device structure
4416 static void ixgbe_tx_timeout(struct net_device *netdev)
4418 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4420 /* Do the reset outside of interrupt context */
4421 ixgbe_tx_timeout_reset(adapter);
4425 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4426 * @adapter: board private structure to initialize
4428 * ixgbe_sw_init initializes the Adapter private data structure.
4429 * Fields are initialized based on PCI device information and
4430 * OS network device settings (MTU size).
4432 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
4434 struct ixgbe_hw *hw = &adapter->hw;
4435 struct pci_dev *pdev = adapter->pdev;
4437 #ifdef CONFIG_IXGBE_DCB
4439 struct tc_configuration *tc;
4442 /* PCI config space info */
4444 hw->vendor_id = pdev->vendor;
4445 hw->device_id = pdev->device;
4446 hw->revision_id = pdev->revision;
4447 hw->subsystem_vendor_id = pdev->subsystem_vendor;
4448 hw->subsystem_device_id = pdev->subsystem_device;
4450 /* Set capability flags */
4451 rss = min_t(int, IXGBE_MAX_RSS_INDICES, num_online_cpus());
4452 adapter->ring_feature[RING_F_RSS].limit = rss;
4453 switch (hw->mac.type) {
4454 case ixgbe_mac_82598EB:
4455 if (hw->device_id == IXGBE_DEV_ID_82598AT)
4456 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4457 adapter->max_q_vectors = MAX_Q_VECTORS_82598;
4459 case ixgbe_mac_X540:
4460 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4461 case ixgbe_mac_82599EB:
4462 adapter->max_q_vectors = MAX_Q_VECTORS_82599;
4463 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4464 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4465 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
4466 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4467 /* Flow Director hash filters enabled */
4468 adapter->atr_sample_rate = 20;
4469 adapter->ring_feature[RING_F_FDIR].limit =
4470 IXGBE_MAX_FDIR_INDICES;
4471 adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
4473 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4474 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4475 #ifdef CONFIG_IXGBE_DCB
4476 /* Default traffic class to use for FCoE */
4477 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
4479 #endif /* IXGBE_FCOE */
4486 /* FCoE support exists, always init the FCoE lock */
4487 spin_lock_init(&adapter->fcoe.lock);
4490 /* n-tuple support exists, always init our spinlock */
4491 spin_lock_init(&adapter->fdir_perfect_lock);
4493 #ifdef CONFIG_IXGBE_DCB
4494 switch (hw->mac.type) {
4495 case ixgbe_mac_X540:
4496 adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
4497 adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
4500 adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
4501 adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
4505 /* Configure DCB traffic classes */
4506 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
4507 tc = &adapter->dcb_cfg.tc_config[j];
4508 tc->path[DCB_TX_CONFIG].bwg_id = 0;
4509 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
4510 tc->path[DCB_RX_CONFIG].bwg_id = 0;
4511 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
4512 tc->dcb_pfc = pfc_disabled;
4515 /* Initialize default user to priority mapping, UPx->TC0 */
4516 tc = &adapter->dcb_cfg.tc_config[0];
4517 tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
4518 tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
4520 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
4521 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
4522 adapter->dcb_cfg.pfc_mode_enable = false;
4523 adapter->dcb_set_bitmap = 0x00;
4524 adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
4525 memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg,
4526 sizeof(adapter->temp_dcb_cfg));
4530 /* default flow control settings */
4531 hw->fc.requested_mode = ixgbe_fc_full;
4532 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
4533 ixgbe_pbthresh_setup(adapter);
4534 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
4535 hw->fc.send_xon = true;
4536 hw->fc.disable_fc_autoneg = false;
4538 #ifdef CONFIG_PCI_IOV
4539 /* assign number of SR-IOV VFs */
4540 if (hw->mac.type != ixgbe_mac_82598EB)
4541 adapter->num_vfs = (max_vfs > 63) ? 0 : max_vfs;
4544 /* enable itr by default in dynamic mode */
4545 adapter->rx_itr_setting = 1;
4546 adapter->tx_itr_setting = 1;
4548 /* set default ring sizes */
4549 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
4550 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
4552 /* set default work limits */
4553 adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
4555 /* initialize eeprom parameters */
4556 if (ixgbe_init_eeprom_params_generic(hw)) {
4557 e_dev_err("EEPROM initialization failed\n");
4561 set_bit(__IXGBE_DOWN, &adapter->state);
4567 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
4568 * @tx_ring: tx descriptor ring (for a specific queue) to setup
4570 * Return 0 on success, negative on failure
4572 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
4574 struct device *dev = tx_ring->dev;
4575 int orig_node = dev_to_node(dev);
4579 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4581 if (tx_ring->q_vector)
4582 numa_node = tx_ring->q_vector->numa_node;
4584 tx_ring->tx_buffer_info = vzalloc_node(size, numa_node);
4585 if (!tx_ring->tx_buffer_info)
4586 tx_ring->tx_buffer_info = vzalloc(size);
4587 if (!tx_ring->tx_buffer_info)
4590 /* round up to nearest 4K */
4591 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
4592 tx_ring->size = ALIGN(tx_ring->size, 4096);
4594 set_dev_node(dev, numa_node);
4595 tx_ring->desc = dma_alloc_coherent(dev,
4599 set_dev_node(dev, orig_node);
4601 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
4602 &tx_ring->dma, GFP_KERNEL);
4606 tx_ring->next_to_use = 0;
4607 tx_ring->next_to_clean = 0;
4611 vfree(tx_ring->tx_buffer_info);
4612 tx_ring->tx_buffer_info = NULL;
4613 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
4618 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
4619 * @adapter: board private structure
4621 * If this function returns with an error, then it's possible one or
4622 * more of the rings is populated (while the rest are not). It is the
4623 * callers duty to clean those orphaned rings.
4625 * Return 0 on success, negative on failure
4627 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
4631 for (i = 0; i < adapter->num_tx_queues; i++) {
4632 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
4636 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
4642 /* rewind the index freeing the rings as we go */
4644 ixgbe_free_tx_resources(adapter->tx_ring[i]);
4649 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
4650 * @rx_ring: rx descriptor ring (for a specific queue) to setup
4652 * Returns 0 on success, negative on failure
4654 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
4656 struct device *dev = rx_ring->dev;
4657 int orig_node = dev_to_node(dev);
4661 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4663 if (rx_ring->q_vector)
4664 numa_node = rx_ring->q_vector->numa_node;
4666 rx_ring->rx_buffer_info = vzalloc_node(size, numa_node);
4667 if (!rx_ring->rx_buffer_info)
4668 rx_ring->rx_buffer_info = vzalloc(size);
4669 if (!rx_ring->rx_buffer_info)
4672 /* Round up to nearest 4K */
4673 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
4674 rx_ring->size = ALIGN(rx_ring->size, 4096);
4676 set_dev_node(dev, numa_node);
4677 rx_ring->desc = dma_alloc_coherent(dev,
4681 set_dev_node(dev, orig_node);
4683 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
4684 &rx_ring->dma, GFP_KERNEL);
4688 rx_ring->next_to_clean = 0;
4689 rx_ring->next_to_use = 0;
4693 vfree(rx_ring->rx_buffer_info);
4694 rx_ring->rx_buffer_info = NULL;
4695 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
4700 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
4701 * @adapter: board private structure
4703 * If this function returns with an error, then it's possible one or
4704 * more of the rings is populated (while the rest are not). It is the
4705 * callers duty to clean those orphaned rings.
4707 * Return 0 on success, negative on failure
4709 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
4713 for (i = 0; i < adapter->num_rx_queues; i++) {
4714 err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
4718 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
4723 err = ixgbe_setup_fcoe_ddp_resources(adapter);
4728 /* rewind the index freeing the rings as we go */
4730 ixgbe_free_rx_resources(adapter->rx_ring[i]);
4735 * ixgbe_free_tx_resources - Free Tx Resources per Queue
4736 * @tx_ring: Tx descriptor ring for a specific queue
4738 * Free all transmit software resources
4740 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
4742 ixgbe_clean_tx_ring(tx_ring);
4744 vfree(tx_ring->tx_buffer_info);
4745 tx_ring->tx_buffer_info = NULL;
4747 /* if not set, then don't free */
4751 dma_free_coherent(tx_ring->dev, tx_ring->size,
4752 tx_ring->desc, tx_ring->dma);
4754 tx_ring->desc = NULL;
4758 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
4759 * @adapter: board private structure
4761 * Free all transmit software resources
4763 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
4767 for (i = 0; i < adapter->num_tx_queues; i++)
4768 if (adapter->tx_ring[i]->desc)
4769 ixgbe_free_tx_resources(adapter->tx_ring[i]);
4773 * ixgbe_free_rx_resources - Free Rx Resources
4774 * @rx_ring: ring to clean the resources from
4776 * Free all receive software resources
4778 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
4780 ixgbe_clean_rx_ring(rx_ring);
4782 vfree(rx_ring->rx_buffer_info);
4783 rx_ring->rx_buffer_info = NULL;
4785 /* if not set, then don't free */
4789 dma_free_coherent(rx_ring->dev, rx_ring->size,
4790 rx_ring->desc, rx_ring->dma);
4792 rx_ring->desc = NULL;
4796 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
4797 * @adapter: board private structure
4799 * Free all receive software resources
4801 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
4806 ixgbe_free_fcoe_ddp_resources(adapter);
4809 for (i = 0; i < adapter->num_rx_queues; i++)
4810 if (adapter->rx_ring[i]->desc)
4811 ixgbe_free_rx_resources(adapter->rx_ring[i]);
4815 * ixgbe_change_mtu - Change the Maximum Transfer Unit
4816 * @netdev: network interface device structure
4817 * @new_mtu: new value for maximum frame size
4819 * Returns 0 on success, negative on failure
4821 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
4823 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4824 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4826 /* MTU < 68 is an error and causes problems on some kernels */
4827 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
4831 * For 82599EB we cannot allow PF to change MTU greater than 1500
4832 * in SR-IOV mode as it may cause buffer overruns in guest VFs that
4833 * don't allocate and chain buffers correctly.
4835 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
4836 (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
4837 (max_frame > MAXIMUM_ETHERNET_VLAN_SIZE))
4840 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4842 /* must set new MTU before calling down or up */
4843 netdev->mtu = new_mtu;
4845 if (netif_running(netdev))
4846 ixgbe_reinit_locked(adapter);
4852 * ixgbe_open - Called when a network interface is made active
4853 * @netdev: network interface device structure
4855 * Returns 0 on success, negative value on failure
4857 * The open entry point is called when a network interface is made
4858 * active by the system (IFF_UP). At this point all resources needed
4859 * for transmit and receive operations are allocated, the interrupt
4860 * handler is registered with the OS, the watchdog timer is started,
4861 * and the stack is notified that the interface is ready.
4863 static int ixgbe_open(struct net_device *netdev)
4865 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4868 /* disallow open during test */
4869 if (test_bit(__IXGBE_TESTING, &adapter->state))
4872 netif_carrier_off(netdev);
4874 /* allocate transmit descriptors */
4875 err = ixgbe_setup_all_tx_resources(adapter);
4879 /* allocate receive descriptors */
4880 err = ixgbe_setup_all_rx_resources(adapter);
4884 ixgbe_configure(adapter);
4886 err = ixgbe_request_irq(adapter);
4890 /* Notify the stack of the actual queue counts. */
4891 err = netif_set_real_num_tx_queues(netdev,
4892 adapter->num_rx_pools > 1 ? 1 :
4893 adapter->num_tx_queues);
4895 goto err_set_queues;
4898 err = netif_set_real_num_rx_queues(netdev,
4899 adapter->num_rx_pools > 1 ? 1 :
4900 adapter->num_rx_queues);
4902 goto err_set_queues;
4904 ixgbe_up_complete(adapter);
4909 ixgbe_free_irq(adapter);
4911 ixgbe_free_all_rx_resources(adapter);
4913 ixgbe_free_all_tx_resources(adapter);
4915 ixgbe_reset(adapter);
4921 * ixgbe_close - Disables a network interface
4922 * @netdev: network interface device structure
4924 * Returns 0, this is not allowed to fail
4926 * The close entry point is called when an interface is de-activated
4927 * by the OS. The hardware is still under the drivers control, but
4928 * needs to be disabled. A global MAC reset is issued to stop the
4929 * hardware, and all transmit and receive resources are freed.
4931 static int ixgbe_close(struct net_device *netdev)
4933 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4935 ixgbe_down(adapter);
4936 ixgbe_free_irq(adapter);
4938 ixgbe_fdir_filter_exit(adapter);
4940 ixgbe_free_all_tx_resources(adapter);
4941 ixgbe_free_all_rx_resources(adapter);
4943 ixgbe_release_hw_control(adapter);
4949 static int ixgbe_resume(struct pci_dev *pdev)
4951 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
4952 struct net_device *netdev = adapter->netdev;
4955 pci_set_power_state(pdev, PCI_D0);
4956 pci_restore_state(pdev);
4958 * pci_restore_state clears dev->state_saved so call
4959 * pci_save_state to restore it.
4961 pci_save_state(pdev);
4963 err = pci_enable_device_mem(pdev);
4965 e_dev_err("Cannot enable PCI device from suspend\n");
4968 pci_set_master(pdev);
4970 pci_wake_from_d3(pdev, false);
4972 ixgbe_reset(adapter);
4974 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
4977 err = ixgbe_init_interrupt_scheme(adapter);
4978 if (!err && netif_running(netdev))
4979 err = ixgbe_open(netdev);
4986 netif_device_attach(netdev);
4990 #endif /* CONFIG_PM */
4992 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
4994 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
4995 struct net_device *netdev = adapter->netdev;
4996 struct ixgbe_hw *hw = &adapter->hw;
4998 u32 wufc = adapter->wol;
5003 netif_device_detach(netdev);
5005 if (netif_running(netdev)) {
5007 ixgbe_down(adapter);
5008 ixgbe_free_irq(adapter);
5009 ixgbe_free_all_tx_resources(adapter);
5010 ixgbe_free_all_rx_resources(adapter);
5014 ixgbe_clear_interrupt_scheme(adapter);
5017 retval = pci_save_state(pdev);
5023 ixgbe_set_rx_mode(netdev);
5026 * enable the optics for both mult-speed fiber and
5027 * 82599 SFP+ fiber as we can WoL.
5029 if (hw->mac.ops.enable_tx_laser &&
5030 (hw->phy.multispeed_fiber ||
5031 (hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber &&
5032 hw->mac.type == ixgbe_mac_82599EB)))
5033 hw->mac.ops.enable_tx_laser(hw);
5035 /* turn on all-multi mode if wake on multicast is enabled */
5036 if (wufc & IXGBE_WUFC_MC) {
5037 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5038 fctrl |= IXGBE_FCTRL_MPE;
5039 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5042 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5043 ctrl |= IXGBE_CTRL_GIO_DIS;
5044 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5046 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5048 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5049 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5052 switch (hw->mac.type) {
5053 case ixgbe_mac_82598EB:
5054 pci_wake_from_d3(pdev, false);
5056 case ixgbe_mac_82599EB:
5057 case ixgbe_mac_X540:
5058 pci_wake_from_d3(pdev, !!wufc);
5064 *enable_wake = !!wufc;
5066 ixgbe_release_hw_control(adapter);
5068 pci_disable_device(pdev);
5074 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5079 retval = __ixgbe_shutdown(pdev, &wake);
5084 pci_prepare_to_sleep(pdev);
5086 pci_wake_from_d3(pdev, false);
5087 pci_set_power_state(pdev, PCI_D3hot);
5092 #endif /* CONFIG_PM */
5094 static void ixgbe_shutdown(struct pci_dev *pdev)
5098 __ixgbe_shutdown(pdev, &wake);
5100 if (system_state == SYSTEM_POWER_OFF) {
5101 pci_wake_from_d3(pdev, wake);
5102 pci_set_power_state(pdev, PCI_D3hot);
5107 * ixgbe_update_stats - Update the board statistics counters.
5108 * @adapter: board private structure
5110 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5112 struct net_device *netdev = adapter->netdev;
5113 struct ixgbe_hw *hw = &adapter->hw;
5114 struct ixgbe_hw_stats *hwstats = &adapter->stats;
5116 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5117 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
5118 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
5119 u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
5121 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5122 test_bit(__IXGBE_RESETTING, &adapter->state))
5125 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5128 for (i = 0; i < adapter->num_rx_queues; i++) {
5129 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
5130 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
5132 adapter->rsc_total_count = rsc_count;
5133 adapter->rsc_total_flush = rsc_flush;
5136 for (i = 0; i < adapter->num_rx_queues; i++) {
5137 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
5138 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
5139 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
5140 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
5141 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
5142 bytes += rx_ring->stats.bytes;
5143 packets += rx_ring->stats.packets;
5145 adapter->non_eop_descs = non_eop_descs;
5146 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5147 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5148 adapter->hw_csum_rx_error = hw_csum_rx_error;
5149 netdev->stats.rx_bytes = bytes;
5150 netdev->stats.rx_packets = packets;
5154 /* gather some stats to the adapter struct that are per queue */
5155 for (i = 0; i < adapter->num_tx_queues; i++) {
5156 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5157 restart_queue += tx_ring->tx_stats.restart_queue;
5158 tx_busy += tx_ring->tx_stats.tx_busy;
5159 bytes += tx_ring->stats.bytes;
5160 packets += tx_ring->stats.packets;
5162 adapter->restart_queue = restart_queue;
5163 adapter->tx_busy = tx_busy;
5164 netdev->stats.tx_bytes = bytes;
5165 netdev->stats.tx_packets = packets;
5167 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5169 /* 8 register reads */
5170 for (i = 0; i < 8; i++) {
5171 /* for packet buffers not used, the register should read 0 */
5172 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5174 hwstats->mpc[i] += mpc;
5175 total_mpc += hwstats->mpc[i];
5176 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5177 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5178 switch (hw->mac.type) {
5179 case ixgbe_mac_82598EB:
5180 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5181 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5182 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5183 hwstats->pxonrxc[i] +=
5184 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5186 case ixgbe_mac_82599EB:
5187 case ixgbe_mac_X540:
5188 hwstats->pxonrxc[i] +=
5189 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5196 /*16 register reads */
5197 for (i = 0; i < 16; i++) {
5198 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5199 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5200 if ((hw->mac.type == ixgbe_mac_82599EB) ||
5201 (hw->mac.type == ixgbe_mac_X540)) {
5202 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
5203 IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
5204 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
5205 IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
5209 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5210 /* work around hardware counting issue */
5211 hwstats->gprc -= missed_rx;
5213 ixgbe_update_xoff_received(adapter);
5215 /* 82598 hardware only has a 32 bit counter in the high register */
5216 switch (hw->mac.type) {
5217 case ixgbe_mac_82598EB:
5218 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5219 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5220 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5221 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5223 case ixgbe_mac_X540:
5224 /* OS2BMC stats are X540 only*/
5225 hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
5226 hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
5227 hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
5228 hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
5229 case ixgbe_mac_82599EB:
5230 for (i = 0; i < 16; i++)
5231 adapter->hw_rx_no_dma_resources +=
5232 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5233 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5234 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
5235 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5236 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
5237 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5238 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5239 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5240 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5241 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5243 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5244 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5245 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5246 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5247 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5248 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5249 /* Add up per cpu counters for total ddp aloc fail */
5250 if (adapter->fcoe.ddp_pool) {
5251 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
5252 struct ixgbe_fcoe_ddp_pool *ddp_pool;
5254 u64 noddp = 0, noddp_ext_buff = 0;
5255 for_each_possible_cpu(cpu) {
5256 ddp_pool = per_cpu_ptr(fcoe->ddp_pool, cpu);
5257 noddp += ddp_pool->noddp;
5258 noddp_ext_buff += ddp_pool->noddp_ext_buff;
5260 hwstats->fcoe_noddp = noddp;
5261 hwstats->fcoe_noddp_ext_buff = noddp_ext_buff;
5263 #endif /* IXGBE_FCOE */
5268 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5269 hwstats->bprc += bprc;
5270 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5271 if (hw->mac.type == ixgbe_mac_82598EB)
5272 hwstats->mprc -= bprc;
5273 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5274 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5275 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5276 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5277 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5278 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5279 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5280 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5281 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5282 hwstats->lxontxc += lxon;
5283 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5284 hwstats->lxofftxc += lxoff;
5285 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5286 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5288 * 82598 errata - tx of flow control packets is included in tx counters
5290 xon_off_tot = lxon + lxoff;
5291 hwstats->gptc -= xon_off_tot;
5292 hwstats->mptc -= xon_off_tot;
5293 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5294 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5295 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5296 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5297 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5298 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5299 hwstats->ptc64 -= xon_off_tot;
5300 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5301 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5302 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5303 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5304 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5305 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5307 /* Fill out the OS statistics structure */
5308 netdev->stats.multicast = hwstats->mprc;
5311 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
5312 netdev->stats.rx_dropped = 0;
5313 netdev->stats.rx_length_errors = hwstats->rlec;
5314 netdev->stats.rx_crc_errors = hwstats->crcerrs;
5315 netdev->stats.rx_missed_errors = total_mpc;
5319 * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
5320 * @adapter: pointer to the device adapter structure
5322 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
5324 struct ixgbe_hw *hw = &adapter->hw;
5327 if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
5330 adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
5332 /* if interface is down do nothing */
5333 if (test_bit(__IXGBE_DOWN, &adapter->state))
5336 /* do nothing if we are not using signature filters */
5337 if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
5340 adapter->fdir_overflow++;
5342 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5343 for (i = 0; i < adapter->num_tx_queues; i++)
5344 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
5345 &(adapter->tx_ring[i]->state));
5346 /* re-enable flow director interrupts */
5347 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
5349 e_err(probe, "failed to finish FDIR re-initialization, "
5350 "ignored adding FDIR ATR filters\n");
5355 * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
5356 * @adapter: pointer to the device adapter structure
5358 * This function serves two purposes. First it strobes the interrupt lines
5359 * in order to make certain interrupts are occurring. Secondly it sets the
5360 * bits needed to check for TX hangs. As a result we should immediately
5361 * determine if a hang has occurred.
5363 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
5365 struct ixgbe_hw *hw = &adapter->hw;
5369 /* If we're down or resetting, just bail */
5370 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5371 test_bit(__IXGBE_RESETTING, &adapter->state))
5374 /* Force detection of hung controller */
5375 if (netif_carrier_ok(adapter->netdev)) {
5376 for (i = 0; i < adapter->num_tx_queues; i++)
5377 set_check_for_tx_hang(adapter->tx_ring[i]);
5380 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5382 * for legacy and MSI interrupts don't set any bits
5383 * that are enabled for EIAM, because this operation
5384 * would set *both* EIMS and EICS for any bit in EIAM
5386 IXGBE_WRITE_REG(hw, IXGBE_EICS,
5387 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5389 /* get one bit for every active tx/rx interrupt vector */
5390 for (i = 0; i < adapter->num_q_vectors; i++) {
5391 struct ixgbe_q_vector *qv = adapter->q_vector[i];
5392 if (qv->rx.ring || qv->tx.ring)
5393 eics |= ((u64)1 << i);
5397 /* Cause software interrupt to ensure rings are cleaned */
5398 ixgbe_irq_rearm_queues(adapter, eics);
5403 * ixgbe_watchdog_update_link - update the link status
5404 * @adapter: pointer to the device adapter structure
5405 * @link_speed: pointer to a u32 to store the link_speed
5407 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
5409 struct ixgbe_hw *hw = &adapter->hw;
5410 u32 link_speed = adapter->link_speed;
5411 bool link_up = adapter->link_up;
5412 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
5414 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
5417 if (hw->mac.ops.check_link) {
5418 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5420 /* always assume link is up, if no check link function */
5421 link_speed = IXGBE_LINK_SPEED_10GB_FULL;
5425 if (adapter->ixgbe_ieee_pfc)
5426 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
5428 if (link_up && !((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && pfc_en)) {
5429 hw->mac.ops.fc_enable(hw);
5430 ixgbe_set_rx_drop_en(adapter);
5434 time_after(jiffies, (adapter->link_check_timeout +
5435 IXGBE_TRY_LINK_TIMEOUT))) {
5436 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5437 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5438 IXGBE_WRITE_FLUSH(hw);
5441 adapter->link_up = link_up;
5442 adapter->link_speed = link_speed;
5446 * ixgbe_watchdog_link_is_up - update netif_carrier status and
5447 * print link up message
5448 * @adapter: pointer to the device adapter structure
5450 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
5452 struct net_device *netdev = adapter->netdev;
5453 struct ixgbe_hw *hw = &adapter->hw;
5454 u32 link_speed = adapter->link_speed;
5455 bool flow_rx, flow_tx;
5457 /* only continue if link was previously down */
5458 if (netif_carrier_ok(netdev))
5461 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
5463 switch (hw->mac.type) {
5464 case ixgbe_mac_82598EB: {
5465 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5466 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5467 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5468 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5471 case ixgbe_mac_X540:
5472 case ixgbe_mac_82599EB: {
5473 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5474 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5475 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5476 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5485 #ifdef CONFIG_IXGBE_PTP
5486 ixgbe_ptp_start_cyclecounter(adapter);
5489 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
5490 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5492 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5494 (link_speed == IXGBE_LINK_SPEED_100_FULL ?
5497 ((flow_rx && flow_tx) ? "RX/TX" :
5499 (flow_tx ? "TX" : "None"))));
5501 netif_carrier_on(netdev);
5502 ixgbe_check_vf_rate_limit(adapter);
5504 /* ping all the active vfs to let them know link has changed */
5505 ixgbe_ping_all_vfs(adapter);
5509 * ixgbe_watchdog_link_is_down - update netif_carrier status and
5510 * print link down message
5511 * @adapter: pointer to the adapter structure
5513 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
5515 struct net_device *netdev = adapter->netdev;
5516 struct ixgbe_hw *hw = &adapter->hw;
5518 adapter->link_up = false;
5519 adapter->link_speed = 0;
5521 /* only continue if link was up previously */
5522 if (!netif_carrier_ok(netdev))
5525 /* poll for SFP+ cable when link is down */
5526 if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
5527 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5529 #ifdef CONFIG_IXGBE_PTP
5530 ixgbe_ptp_start_cyclecounter(adapter);
5533 e_info(drv, "NIC Link is Down\n");
5534 netif_carrier_off(netdev);
5536 /* ping all the active vfs to let them know link has changed */
5537 ixgbe_ping_all_vfs(adapter);
5541 * ixgbe_watchdog_flush_tx - flush queues on link down
5542 * @adapter: pointer to the device adapter structure
5544 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
5547 int some_tx_pending = 0;
5549 if (!netif_carrier_ok(adapter->netdev)) {
5550 for (i = 0; i < adapter->num_tx_queues; i++) {
5551 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5552 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5553 some_tx_pending = 1;
5558 if (some_tx_pending) {
5559 /* We've lost link, so the controller stops DMA,
5560 * but we've got queued Tx work that's never going
5561 * to get done, so reset controller to flush Tx.
5562 * (Do the reset outside of interrupt context).
5564 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
5569 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
5573 /* Do not perform spoof check for 82598 or if not in IOV mode */
5574 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
5575 adapter->num_vfs == 0)
5578 ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
5581 * ssvpc register is cleared on read, if zero then no
5582 * spoofed packets in the last interval.
5587 e_warn(drv, "%u Spoofed packets detected\n", ssvpc);
5591 * ixgbe_watchdog_subtask - check and bring link up
5592 * @adapter: pointer to the device adapter structure
5594 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
5596 /* if interface is down do nothing */
5597 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5598 test_bit(__IXGBE_RESETTING, &adapter->state))
5601 ixgbe_watchdog_update_link(adapter);
5603 if (adapter->link_up)
5604 ixgbe_watchdog_link_is_up(adapter);
5606 ixgbe_watchdog_link_is_down(adapter);
5608 ixgbe_spoof_check(adapter);
5609 ixgbe_update_stats(adapter);
5611 ixgbe_watchdog_flush_tx(adapter);
5615 * ixgbe_sfp_detection_subtask - poll for SFP+ cable
5616 * @adapter: the ixgbe adapter structure
5618 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
5620 struct ixgbe_hw *hw = &adapter->hw;
5623 /* not searching for SFP so there is nothing to do here */
5624 if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
5625 !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
5628 /* someone else is in init, wait until next service event */
5629 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5632 err = hw->phy.ops.identify_sfp(hw);
5633 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
5636 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
5637 /* If no cable is present, then we need to reset
5638 * the next time we find a good cable. */
5639 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
5646 /* exit if reset not needed */
5647 if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
5650 adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
5653 * A module may be identified correctly, but the EEPROM may not have
5654 * support for that module. setup_sfp() will fail in that case, so
5655 * we should not allow that module to load.
5657 if (hw->mac.type == ixgbe_mac_82598EB)
5658 err = hw->phy.ops.reset(hw);
5660 err = hw->mac.ops.setup_sfp(hw);
5662 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
5665 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
5666 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
5669 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5671 if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
5672 (adapter->netdev->reg_state == NETREG_REGISTERED)) {
5673 e_dev_err("failed to initialize because an unsupported "
5674 "SFP+ module type was detected.\n");
5675 e_dev_err("Reload the driver after installing a "
5676 "supported module.\n");
5677 unregister_netdev(adapter->netdev);
5682 * ixgbe_sfp_link_config_subtask - set up link SFP after module install
5683 * @adapter: the ixgbe adapter structure
5685 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
5687 struct ixgbe_hw *hw = &adapter->hw;
5691 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
5694 /* someone else is in init, wait until next service event */
5695 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5698 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
5700 autoneg = hw->phy.autoneg_advertised;
5701 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
5702 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
5703 if (hw->mac.ops.setup_link)
5704 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
5706 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5707 adapter->link_check_timeout = jiffies;
5708 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5711 #ifdef CONFIG_PCI_IOV
5712 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
5715 struct ixgbe_hw *hw = &adapter->hw;
5716 struct net_device *netdev = adapter->netdev;
5720 gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
5721 if (gpc) /* If incrementing then no need for the check below */
5724 * Check to see if a bad DMA write target from an errant or
5725 * malicious VF has caused a PCIe error. If so then we can
5726 * issue a VFLR to the offending VF(s) and then resume without
5727 * requesting a full slot reset.
5730 for (vf = 0; vf < adapter->num_vfs; vf++) {
5731 ciaa = (vf << 16) | 0x80000000;
5732 /* 32 bit read so align, we really want status at offset 6 */
5733 ciaa |= PCI_COMMAND;
5734 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5735 ciad = IXGBE_READ_REG(hw, IXGBE_CIAD_82599);
5737 /* disable debug mode asap after reading data */
5738 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5739 /* Get the upper 16 bits which will be the PCI status reg */
5741 if (ciad & PCI_STATUS_REC_MASTER_ABORT) {
5742 netdev_err(netdev, "VF %d Hung DMA\n", vf);
5744 ciaa = (vf << 16) | 0x80000000;
5746 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5747 ciad = 0x00008000; /* VFLR */
5748 IXGBE_WRITE_REG(hw, IXGBE_CIAD_82599, ciad);
5750 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5757 * ixgbe_service_timer - Timer Call-back
5758 * @data: pointer to adapter cast into an unsigned long
5760 static void ixgbe_service_timer(unsigned long data)
5762 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5763 unsigned long next_event_offset;
5766 /* poll faster when waiting for link */
5767 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
5768 next_event_offset = HZ / 10;
5770 next_event_offset = HZ * 2;
5772 #ifdef CONFIG_PCI_IOV
5774 * don't bother with SR-IOV VF DMA hang check if there are
5775 * no VFs or the link is down
5777 if (!adapter->num_vfs ||
5778 (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
5779 goto normal_timer_service;
5781 /* If we have VFs allocated then we must check for DMA hangs */
5782 ixgbe_check_for_bad_vf(adapter);
5783 next_event_offset = HZ / 50;
5784 adapter->timer_event_accumulator++;
5786 if (adapter->timer_event_accumulator >= 100)
5787 adapter->timer_event_accumulator = 0;
5791 normal_timer_service:
5793 /* Reset the timer */
5794 mod_timer(&adapter->service_timer, next_event_offset + jiffies);
5797 ixgbe_service_event_schedule(adapter);
5800 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
5802 if (!(adapter->flags2 & IXGBE_FLAG2_RESET_REQUESTED))
5805 adapter->flags2 &= ~IXGBE_FLAG2_RESET_REQUESTED;
5807 /* If we're already down or resetting, just bail */
5808 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5809 test_bit(__IXGBE_RESETTING, &adapter->state))
5812 ixgbe_dump(adapter);
5813 netdev_err(adapter->netdev, "Reset adapter\n");
5814 adapter->tx_timeout_count++;
5816 ixgbe_reinit_locked(adapter);
5820 * ixgbe_service_task - manages and runs subtasks
5821 * @work: pointer to work_struct containing our data
5823 static void ixgbe_service_task(struct work_struct *work)
5825 struct ixgbe_adapter *adapter = container_of(work,
5826 struct ixgbe_adapter,
5829 ixgbe_reset_subtask(adapter);
5830 ixgbe_sfp_detection_subtask(adapter);
5831 ixgbe_sfp_link_config_subtask(adapter);
5832 ixgbe_check_overtemp_subtask(adapter);
5833 ixgbe_watchdog_subtask(adapter);
5834 ixgbe_fdir_reinit_subtask(adapter);
5835 ixgbe_check_hang_subtask(adapter);
5836 #ifdef CONFIG_IXGBE_PTP
5837 ixgbe_ptp_overflow_check(adapter);
5840 ixgbe_service_event_complete(adapter);
5843 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
5844 struct ixgbe_tx_buffer *first,
5847 struct sk_buff *skb = first->skb;
5848 u32 vlan_macip_lens, type_tucmd;
5849 u32 mss_l4len_idx, l4len;
5851 if (!skb_is_gso(skb))
5854 if (skb_header_cloned(skb)) {
5855 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5860 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
5861 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
5863 if (first->protocol == __constant_htons(ETH_P_IP)) {
5864 struct iphdr *iph = ip_hdr(skb);
5867 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
5871 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
5872 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
5873 IXGBE_TX_FLAGS_CSUM |
5874 IXGBE_TX_FLAGS_IPV4;
5875 } else if (skb_is_gso_v6(skb)) {
5876 ipv6_hdr(skb)->payload_len = 0;
5877 tcp_hdr(skb)->check =
5878 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5879 &ipv6_hdr(skb)->daddr,
5881 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
5882 IXGBE_TX_FLAGS_CSUM;
5885 /* compute header lengths */
5886 l4len = tcp_hdrlen(skb);
5887 *hdr_len = skb_transport_offset(skb) + l4len;
5889 /* update gso size and bytecount with header size */
5890 first->gso_segs = skb_shinfo(skb)->gso_segs;
5891 first->bytecount += (first->gso_segs - 1) * *hdr_len;
5893 /* mss_l4len_id: use 1 as index for TSO */
5894 mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
5895 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
5896 mss_l4len_idx |= 1 << IXGBE_ADVTXD_IDX_SHIFT;
5898 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
5899 vlan_macip_lens = skb_network_header_len(skb);
5900 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
5901 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
5903 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0, type_tucmd,
5909 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
5910 struct ixgbe_tx_buffer *first)
5912 struct sk_buff *skb = first->skb;
5913 u32 vlan_macip_lens = 0;
5914 u32 mss_l4len_idx = 0;
5917 if (skb->ip_summed != CHECKSUM_PARTIAL) {
5918 if (!(first->tx_flags & IXGBE_TX_FLAGS_HW_VLAN)) {
5919 if (unlikely(skb->no_fcs))
5920 first->tx_flags |= IXGBE_TX_FLAGS_NO_IFCS;
5921 if (!(first->tx_flags & IXGBE_TX_FLAGS_TXSW))
5926 switch (first->protocol) {
5927 case __constant_htons(ETH_P_IP):
5928 vlan_macip_lens |= skb_network_header_len(skb);
5929 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
5930 l4_hdr = ip_hdr(skb)->protocol;
5932 case __constant_htons(ETH_P_IPV6):
5933 vlan_macip_lens |= skb_network_header_len(skb);
5934 l4_hdr = ipv6_hdr(skb)->nexthdr;
5937 if (unlikely(net_ratelimit())) {
5938 dev_warn(tx_ring->dev,
5939 "partial checksum but proto=%x!\n",
5947 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5948 mss_l4len_idx = tcp_hdrlen(skb) <<
5949 IXGBE_ADVTXD_L4LEN_SHIFT;
5952 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5953 mss_l4len_idx = sizeof(struct sctphdr) <<
5954 IXGBE_ADVTXD_L4LEN_SHIFT;
5957 mss_l4len_idx = sizeof(struct udphdr) <<
5958 IXGBE_ADVTXD_L4LEN_SHIFT;
5961 if (unlikely(net_ratelimit())) {
5962 dev_warn(tx_ring->dev,
5963 "partial checksum but l4 proto=%x!\n",
5969 /* update TX checksum flag */
5970 first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
5973 /* vlan_macip_lens: MACLEN, VLAN tag */
5974 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
5975 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
5977 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0,
5978 type_tucmd, mss_l4len_idx);
5981 static __le32 ixgbe_tx_cmd_type(u32 tx_flags)
5983 /* set type for advanced descriptor with frame checksum insertion */
5984 __le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
5985 IXGBE_ADVTXD_DCMD_DEXT);
5987 /* set HW vlan bit if vlan is present */
5988 if (tx_flags & IXGBE_TX_FLAGS_HW_VLAN)
5989 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
5991 #ifdef CONFIG_IXGBE_PTP
5992 if (tx_flags & IXGBE_TX_FLAGS_TSTAMP)
5993 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_MAC_TSTAMP);
5996 /* set segmentation enable bits for TSO/FSO */
5998 if (tx_flags & (IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_FSO))
6000 if (tx_flags & IXGBE_TX_FLAGS_TSO)
6002 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
6004 /* insert frame checksum */
6005 if (!(tx_flags & IXGBE_TX_FLAGS_NO_IFCS))
6006 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_IFCS);
6011 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
6012 u32 tx_flags, unsigned int paylen)
6014 __le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
6016 /* enable L4 checksum for TSO and TX checksum offload */
6017 if (tx_flags & IXGBE_TX_FLAGS_CSUM)
6018 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
6020 /* enble IPv4 checksum for TSO */
6021 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
6022 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
6024 /* use index 1 context for TSO/FSO/FCOE */
6026 if (tx_flags & (IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_FCOE))
6028 if (tx_flags & IXGBE_TX_FLAGS_TSO)
6030 olinfo_status |= cpu_to_le32(1 << IXGBE_ADVTXD_IDX_SHIFT);
6033 * Check Context must be set if Tx switch is enabled, which it
6034 * always is for case where virtual functions are running
6037 if (tx_flags & (IXGBE_TX_FLAGS_TXSW | IXGBE_TX_FLAGS_FCOE))
6039 if (tx_flags & IXGBE_TX_FLAGS_TXSW)
6041 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
6043 tx_desc->read.olinfo_status = olinfo_status;
6046 #define IXGBE_TXD_CMD (IXGBE_TXD_CMD_EOP | \
6049 static void ixgbe_tx_map(struct ixgbe_ring *tx_ring,
6050 struct ixgbe_tx_buffer *first,
6054 struct sk_buff *skb = first->skb;
6055 struct ixgbe_tx_buffer *tx_buffer;
6056 union ixgbe_adv_tx_desc *tx_desc;
6057 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
6058 unsigned int data_len = skb->data_len;
6059 unsigned int size = skb_headlen(skb);
6060 unsigned int paylen = skb->len - hdr_len;
6061 u32 tx_flags = first->tx_flags;
6063 u16 i = tx_ring->next_to_use;
6065 tx_desc = IXGBE_TX_DESC(tx_ring, i);
6067 ixgbe_tx_olinfo_status(tx_desc, tx_flags, paylen);
6068 cmd_type = ixgbe_tx_cmd_type(tx_flags);
6071 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6072 if (data_len < sizeof(struct fcoe_crc_eof)) {
6073 size -= sizeof(struct fcoe_crc_eof) - data_len;
6076 data_len -= sizeof(struct fcoe_crc_eof);
6081 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
6082 if (dma_mapping_error(tx_ring->dev, dma))
6085 /* record length, and DMA address */
6086 dma_unmap_len_set(first, len, size);
6087 dma_unmap_addr_set(first, dma, dma);
6089 tx_desc->read.buffer_addr = cpu_to_le64(dma);
6092 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
6093 tx_desc->read.cmd_type_len =
6094 cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
6098 if (i == tx_ring->count) {
6099 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
6103 dma += IXGBE_MAX_DATA_PER_TXD;
6104 size -= IXGBE_MAX_DATA_PER_TXD;
6106 tx_desc->read.buffer_addr = cpu_to_le64(dma);
6107 tx_desc->read.olinfo_status = 0;
6110 if (likely(!data_len))
6113 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
6117 if (i == tx_ring->count) {
6118 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
6123 size = min_t(unsigned int, data_len, skb_frag_size(frag));
6125 size = skb_frag_size(frag);
6129 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
6131 if (dma_mapping_error(tx_ring->dev, dma))
6134 tx_buffer = &tx_ring->tx_buffer_info[i];
6135 dma_unmap_len_set(tx_buffer, len, size);
6136 dma_unmap_addr_set(tx_buffer, dma, dma);
6138 tx_desc->read.buffer_addr = cpu_to_le64(dma);
6139 tx_desc->read.olinfo_status = 0;
6144 /* write last descriptor with RS and EOP bits */
6145 cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD);
6146 tx_desc->read.cmd_type_len = cmd_type;
6148 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
6150 /* set the timestamp */
6151 first->time_stamp = jiffies;
6154 * Force memory writes to complete before letting h/w know there
6155 * are new descriptors to fetch. (Only applicable for weak-ordered
6156 * memory model archs, such as IA-64).
6158 * We also need this memory barrier to make certain all of the
6159 * status bits have been updated before next_to_watch is written.
6163 /* set next_to_watch value indicating a packet is present */
6164 first->next_to_watch = tx_desc;
6167 if (i == tx_ring->count)
6170 tx_ring->next_to_use = i;
6172 /* notify HW of packet */
6173 writel(i, tx_ring->tail);
6177 dev_err(tx_ring->dev, "TX DMA map failed\n");
6179 /* clear dma mappings for failed tx_buffer_info map */
6181 tx_buffer = &tx_ring->tx_buffer_info[i];
6182 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer);
6183 if (tx_buffer == first)
6190 tx_ring->next_to_use = i;
6193 static void ixgbe_atr(struct ixgbe_ring *ring,
6194 struct ixgbe_tx_buffer *first)
6196 struct ixgbe_q_vector *q_vector = ring->q_vector;
6197 union ixgbe_atr_hash_dword input = { .dword = 0 };
6198 union ixgbe_atr_hash_dword common = { .dword = 0 };
6200 unsigned char *network;
6202 struct ipv6hdr *ipv6;
6207 /* if ring doesn't have a interrupt vector, cannot perform ATR */
6211 /* do nothing if sampling is disabled */
6212 if (!ring->atr_sample_rate)
6217 /* snag network header to get L4 type and address */
6218 hdr.network = skb_network_header(first->skb);
6220 /* Currently only IPv4/IPv6 with TCP is supported */
6221 if ((first->protocol != __constant_htons(ETH_P_IPV6) ||
6222 hdr.ipv6->nexthdr != IPPROTO_TCP) &&
6223 (first->protocol != __constant_htons(ETH_P_IP) ||
6224 hdr.ipv4->protocol != IPPROTO_TCP))
6227 th = tcp_hdr(first->skb);
6229 /* skip this packet since it is invalid or the socket is closing */
6233 /* sample on all syn packets or once every atr sample count */
6234 if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
6237 /* reset sample count */
6238 ring->atr_count = 0;
6240 vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
6243 * src and dst are inverted, think how the receiver sees them
6245 * The input is broken into two sections, a non-compressed section
6246 * containing vm_pool, vlan_id, and flow_type. The rest of the data
6247 * is XORed together and stored in the compressed dword.
6249 input.formatted.vlan_id = vlan_id;
6252 * since src port and flex bytes occupy the same word XOR them together
6253 * and write the value to source port portion of compressed dword
6255 if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
6256 common.port.src ^= th->dest ^ __constant_htons(ETH_P_8021Q);
6258 common.port.src ^= th->dest ^ first->protocol;
6259 common.port.dst ^= th->source;
6261 if (first->protocol == __constant_htons(ETH_P_IP)) {
6262 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
6263 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
6265 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
6266 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
6267 hdr.ipv6->saddr.s6_addr32[1] ^
6268 hdr.ipv6->saddr.s6_addr32[2] ^
6269 hdr.ipv6->saddr.s6_addr32[3] ^
6270 hdr.ipv6->daddr.s6_addr32[0] ^
6271 hdr.ipv6->daddr.s6_addr32[1] ^
6272 hdr.ipv6->daddr.s6_addr32[2] ^
6273 hdr.ipv6->daddr.s6_addr32[3];
6276 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
6277 ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
6278 input, common, ring->queue_index);
6281 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6283 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
6284 /* Herbert's original patch had:
6285 * smp_mb__after_netif_stop_queue();
6286 * but since that doesn't exist yet, just open code it. */
6289 /* We need to check again in a case another CPU has just
6290 * made room available. */
6291 if (likely(ixgbe_desc_unused(tx_ring) < size))
6294 /* A reprieve! - use start_queue because it doesn't call schedule */
6295 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
6296 ++tx_ring->tx_stats.restart_queue;
6300 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6302 if (likely(ixgbe_desc_unused(tx_ring) >= size))
6304 return __ixgbe_maybe_stop_tx(tx_ring, size);
6307 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6309 struct ixgbe_adapter *adapter = netdev_priv(dev);
6310 int txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
6313 __be16 protocol = vlan_get_protocol(skb);
6315 if (((protocol == htons(ETH_P_FCOE)) ||
6316 (protocol == htons(ETH_P_FIP))) &&
6317 (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)) {
6318 struct ixgbe_ring_feature *f;
6320 f = &adapter->ring_feature[RING_F_FCOE];
6322 while (txq >= f->indices)
6324 txq += adapter->ring_feature[RING_F_FCOE].offset;
6330 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6331 while (unlikely(txq >= dev->real_num_tx_queues))
6332 txq -= dev->real_num_tx_queues;
6336 return skb_tx_hash(dev, skb);
6339 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
6340 struct ixgbe_adapter *adapter,
6341 struct ixgbe_ring *tx_ring)
6343 struct ixgbe_tx_buffer *first;
6346 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6349 u16 count = TXD_USE_COUNT(skb_headlen(skb));
6350 __be16 protocol = skb->protocol;
6354 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
6355 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
6356 * + 2 desc gap to keep tail from touching head,
6357 * + 1 desc for context descriptor,
6358 * otherwise try next time
6360 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6361 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6362 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6364 count += skb_shinfo(skb)->nr_frags;
6366 if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
6367 tx_ring->tx_stats.tx_busy++;
6368 return NETDEV_TX_BUSY;
6371 /* record the location of the first descriptor for this packet */
6372 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
6374 first->bytecount = skb->len;
6375 first->gso_segs = 1;
6377 /* if we have a HW VLAN tag being added default to the HW one */
6378 if (vlan_tx_tag_present(skb)) {
6379 tx_flags |= vlan_tx_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
6380 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6381 /* else if it is a SW VLAN check the next protocol and store the tag */
6382 } else if (protocol == __constant_htons(ETH_P_8021Q)) {
6383 struct vlan_hdr *vhdr, _vhdr;
6384 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
6388 protocol = vhdr->h_vlan_encapsulated_proto;
6389 tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
6390 IXGBE_TX_FLAGS_VLAN_SHIFT;
6391 tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
6394 skb_tx_timestamp(skb);
6396 #ifdef CONFIG_IXGBE_PTP
6397 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
6398 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
6399 tx_flags |= IXGBE_TX_FLAGS_TSTAMP;
6403 #ifdef CONFIG_PCI_IOV
6405 * Use the l2switch_enable flag - would be false if the DMA
6406 * Tx switch had been disabled.
6408 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6409 tx_flags |= IXGBE_TX_FLAGS_TXSW;
6412 /* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
6413 if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
6414 ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
6415 (skb->priority != TC_PRIO_CONTROL))) {
6416 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6417 tx_flags |= (skb->priority & 0x7) <<
6418 IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
6419 if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
6420 struct vlan_ethhdr *vhdr;
6421 if (skb_header_cloned(skb) &&
6422 pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
6424 vhdr = (struct vlan_ethhdr *)skb->data;
6425 vhdr->h_vlan_TCI = htons(tx_flags >>
6426 IXGBE_TX_FLAGS_VLAN_SHIFT);
6428 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6432 /* record initial flags and protocol */
6433 first->tx_flags = tx_flags;
6434 first->protocol = protocol;
6437 /* setup tx offload for FCoE */
6438 if ((protocol == __constant_htons(ETH_P_FCOE)) &&
6439 (tx_ring->netdev->features & (NETIF_F_FSO | NETIF_F_FCOE_CRC))) {
6440 tso = ixgbe_fso(tx_ring, first, &hdr_len);
6447 #endif /* IXGBE_FCOE */
6448 tso = ixgbe_tso(tx_ring, first, &hdr_len);
6452 ixgbe_tx_csum(tx_ring, first);
6454 /* add the ATR filter if ATR is on */
6455 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
6456 ixgbe_atr(tx_ring, first);
6460 #endif /* IXGBE_FCOE */
6461 ixgbe_tx_map(tx_ring, first, hdr_len);
6463 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
6465 return NETDEV_TX_OK;
6468 dev_kfree_skb_any(first->skb);
6471 return NETDEV_TX_OK;
6474 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
6475 struct net_device *netdev)
6477 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6478 struct ixgbe_ring *tx_ring;
6481 * The minimum packet size for olinfo paylen is 17 so pad the skb
6482 * in order to meet this minimum size requirement.
6484 if (unlikely(skb->len < 17)) {
6485 if (skb_pad(skb, 17 - skb->len))
6486 return NETDEV_TX_OK;
6490 tx_ring = adapter->tx_ring[skb->queue_mapping];
6491 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
6495 * ixgbe_set_mac - Change the Ethernet Address of the NIC
6496 * @netdev: network interface device structure
6497 * @p: pointer to an address structure
6499 * Returns 0 on success, negative on failure
6501 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6503 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6504 struct ixgbe_hw *hw = &adapter->hw;
6505 struct sockaddr *addr = p;
6507 if (!is_valid_ether_addr(addr->sa_data))
6508 return -EADDRNOTAVAIL;
6510 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6511 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6513 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, VMDQ_P(0), IXGBE_RAH_AV);
6519 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6521 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6522 struct ixgbe_hw *hw = &adapter->hw;
6526 if (prtad != hw->phy.mdio.prtad)
6528 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6534 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6535 u16 addr, u16 value)
6537 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6538 struct ixgbe_hw *hw = &adapter->hw;
6540 if (prtad != hw->phy.mdio.prtad)
6542 return hw->phy.ops.write_reg(hw, addr, devad, value);
6545 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6547 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6550 #ifdef CONFIG_IXGBE_PTP
6552 return ixgbe_ptp_hwtstamp_ioctl(adapter, req, cmd);
6555 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6560 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6562 * @netdev: network interface device structure
6564 * Returns non-zero on failure
6566 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6569 struct ixgbe_adapter *adapter = netdev_priv(dev);
6570 struct ixgbe_hw *hw = &adapter->hw;
6572 if (is_valid_ether_addr(hw->mac.san_addr)) {
6574 err = dev_addr_add(dev, hw->mac.san_addr, NETDEV_HW_ADDR_T_SAN);
6577 /* update SAN MAC vmdq pool selection */
6578 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
6584 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6586 * @netdev: network interface device structure
6588 * Returns non-zero on failure
6590 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6593 struct ixgbe_adapter *adapter = netdev_priv(dev);
6594 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6596 if (is_valid_ether_addr(mac->san_addr)) {
6598 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6604 #ifdef CONFIG_NET_POLL_CONTROLLER
6606 * Polling 'interrupt' - used by things like netconsole to send skbs
6607 * without having to re-enable interrupts. It's not called while
6608 * the interrupt routine is executing.
6610 static void ixgbe_netpoll(struct net_device *netdev)
6612 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6615 /* if interface is down do nothing */
6616 if (test_bit(__IXGBE_DOWN, &adapter->state))
6619 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
6620 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
6621 for (i = 0; i < adapter->num_q_vectors; i++)
6622 ixgbe_msix_clean_rings(0, adapter->q_vector[i]);
6624 ixgbe_intr(adapter->pdev->irq, netdev);
6626 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
6630 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
6631 struct rtnl_link_stats64 *stats)
6633 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6637 for (i = 0; i < adapter->num_rx_queues; i++) {
6638 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
6644 start = u64_stats_fetch_begin_bh(&ring->syncp);
6645 packets = ring->stats.packets;
6646 bytes = ring->stats.bytes;
6647 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6648 stats->rx_packets += packets;
6649 stats->rx_bytes += bytes;
6653 for (i = 0; i < adapter->num_tx_queues; i++) {
6654 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->tx_ring[i]);
6660 start = u64_stats_fetch_begin_bh(&ring->syncp);
6661 packets = ring->stats.packets;
6662 bytes = ring->stats.bytes;
6663 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6664 stats->tx_packets += packets;
6665 stats->tx_bytes += bytes;
6669 /* following stats updated by ixgbe_watchdog_task() */
6670 stats->multicast = netdev->stats.multicast;
6671 stats->rx_errors = netdev->stats.rx_errors;
6672 stats->rx_length_errors = netdev->stats.rx_length_errors;
6673 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
6674 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
6678 #ifdef CONFIG_IXGBE_DCB
6680 * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
6681 * @adapter: pointer to ixgbe_adapter
6682 * @tc: number of traffic classes currently enabled
6684 * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
6685 * 802.1Q priority maps to a packet buffer that exists.
6687 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
6689 struct ixgbe_hw *hw = &adapter->hw;
6693 /* 82598 have a static priority to TC mapping that can not
6694 * be changed so no validation is needed.
6696 if (hw->mac.type == ixgbe_mac_82598EB)
6699 reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
6702 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
6703 u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
6705 /* If up2tc is out of bounds default to zero */
6707 reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
6711 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
6717 * ixgbe_set_prio_tc_map - Configure netdev prio tc map
6718 * @adapter: Pointer to adapter struct
6720 * Populate the netdev user priority to tc map
6722 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter *adapter)
6724 struct net_device *dev = adapter->netdev;
6725 struct ixgbe_dcb_config *dcb_cfg = &adapter->dcb_cfg;
6726 struct ieee_ets *ets = adapter->ixgbe_ieee_ets;
6729 for (prio = 0; prio < MAX_USER_PRIORITY; prio++) {
6732 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE)
6733 tc = ixgbe_dcb_get_tc_from_up(dcb_cfg, 0, prio);
6735 tc = ets->prio_tc[prio];
6737 netdev_set_prio_tc_map(dev, prio, tc);
6742 * ixgbe_setup_tc - configure net_device for multiple traffic classes
6744 * @netdev: net device to configure
6745 * @tc: number of traffic classes to enable
6747 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
6749 struct ixgbe_adapter *adapter = netdev_priv(dev);
6750 struct ixgbe_hw *hw = &adapter->hw;
6752 /* Hardware supports up to 8 traffic classes */
6753 if (tc > adapter->dcb_cfg.num_tcs.pg_tcs ||
6754 (hw->mac.type == ixgbe_mac_82598EB &&
6755 tc < MAX_TRAFFIC_CLASS))
6758 /* Hardware has to reinitialize queues and interrupts to
6759 * match packet buffer alignment. Unfortunately, the
6760 * hardware is not flexible enough to do this dynamically.
6762 if (netif_running(dev))
6764 ixgbe_clear_interrupt_scheme(adapter);
6767 netdev_set_num_tc(dev, tc);
6768 ixgbe_set_prio_tc_map(adapter);
6770 adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
6772 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
6773 adapter->last_lfc_mode = adapter->hw.fc.requested_mode;
6774 adapter->hw.fc.requested_mode = ixgbe_fc_none;
6777 netdev_reset_tc(dev);
6779 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
6780 adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
6782 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
6784 adapter->temp_dcb_cfg.pfc_mode_enable = false;
6785 adapter->dcb_cfg.pfc_mode_enable = false;
6788 ixgbe_init_interrupt_scheme(adapter);
6789 ixgbe_validate_rtr(adapter, tc);
6790 if (netif_running(dev))
6796 #endif /* CONFIG_IXGBE_DCB */
6797 void ixgbe_do_reset(struct net_device *netdev)
6799 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6801 if (netif_running(netdev))
6802 ixgbe_reinit_locked(adapter);
6804 ixgbe_reset(adapter);
6807 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
6808 netdev_features_t features)
6810 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6812 /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
6813 if (!(features & NETIF_F_RXCSUM))
6814 features &= ~NETIF_F_LRO;
6816 /* Turn off LRO if not RSC capable */
6817 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
6818 features &= ~NETIF_F_LRO;
6823 static int ixgbe_set_features(struct net_device *netdev,
6824 netdev_features_t features)
6826 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6827 netdev_features_t changed = netdev->features ^ features;
6828 bool need_reset = false;
6830 /* Make sure RSC matches LRO, reset if change */
6831 if (!(features & NETIF_F_LRO)) {
6832 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
6834 adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
6835 } else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
6836 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
6837 if (adapter->rx_itr_setting == 1 ||
6838 adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
6839 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
6841 } else if ((changed ^ features) & NETIF_F_LRO) {
6842 e_info(probe, "rx-usecs set too low, "
6848 * Check if Flow Director n-tuple support was enabled or disabled. If
6849 * the state changed, we need to reset.
6851 switch (features & NETIF_F_NTUPLE) {
6852 case NETIF_F_NTUPLE:
6853 /* turn off ATR, enable perfect filters and reset */
6854 if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
6857 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
6858 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
6861 /* turn off perfect filters, enable ATR and reset */
6862 if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
6865 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
6867 /* We cannot enable ATR if SR-IOV is enabled */
6868 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6871 /* We cannot enable ATR if we have 2 or more traffic classes */
6872 if (netdev_get_num_tc(netdev) > 1)
6875 /* We cannot enable ATR if RSS is disabled */
6876 if (adapter->ring_feature[RING_F_RSS].limit <= 1)
6879 /* A sample rate of 0 indicates ATR disabled */
6880 if (!adapter->atr_sample_rate)
6883 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
6887 if (features & NETIF_F_HW_VLAN_RX)
6888 ixgbe_vlan_strip_enable(adapter);
6890 ixgbe_vlan_strip_disable(adapter);
6892 if (changed & NETIF_F_RXALL)
6895 netdev->features = features;
6897 ixgbe_do_reset(netdev);
6902 static int ixgbe_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
6903 struct net_device *dev,
6904 const unsigned char *addr,
6907 struct ixgbe_adapter *adapter = netdev_priv(dev);
6910 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
6913 if (ndm->ndm_state & NUD_PERMANENT) {
6914 pr_info("%s: FDB only supports static addresses\n",
6919 if (is_unicast_ether_addr(addr)) {
6920 u32 rar_uc_entries = IXGBE_MAX_PF_MACVLANS;
6922 if (netdev_uc_count(dev) < rar_uc_entries)
6923 err = dev_uc_add_excl(dev, addr);
6926 } else if (is_multicast_ether_addr(addr)) {
6927 err = dev_mc_add_excl(dev, addr);
6932 /* Only return duplicate errors if NLM_F_EXCL is set */
6933 if (err == -EEXIST && !(flags & NLM_F_EXCL))
6939 static int ixgbe_ndo_fdb_del(struct ndmsg *ndm,
6940 struct net_device *dev,
6941 const unsigned char *addr)
6943 struct ixgbe_adapter *adapter = netdev_priv(dev);
6944 int err = -EOPNOTSUPP;
6946 if (ndm->ndm_state & NUD_PERMANENT) {
6947 pr_info("%s: FDB only supports static addresses\n",
6952 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
6953 if (is_unicast_ether_addr(addr))
6954 err = dev_uc_del(dev, addr);
6955 else if (is_multicast_ether_addr(addr))
6956 err = dev_mc_del(dev, addr);
6964 static int ixgbe_ndo_fdb_dump(struct sk_buff *skb,
6965 struct netlink_callback *cb,
6966 struct net_device *dev,
6969 struct ixgbe_adapter *adapter = netdev_priv(dev);
6971 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6972 idx = ndo_dflt_fdb_dump(skb, cb, dev, idx);
6977 static const struct net_device_ops ixgbe_netdev_ops = {
6978 .ndo_open = ixgbe_open,
6979 .ndo_stop = ixgbe_close,
6980 .ndo_start_xmit = ixgbe_xmit_frame,
6981 .ndo_select_queue = ixgbe_select_queue,
6982 .ndo_set_rx_mode = ixgbe_set_rx_mode,
6983 .ndo_validate_addr = eth_validate_addr,
6984 .ndo_set_mac_address = ixgbe_set_mac,
6985 .ndo_change_mtu = ixgbe_change_mtu,
6986 .ndo_tx_timeout = ixgbe_tx_timeout,
6987 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
6988 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
6989 .ndo_do_ioctl = ixgbe_ioctl,
6990 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
6991 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
6992 .ndo_set_vf_tx_rate = ixgbe_ndo_set_vf_bw,
6993 .ndo_set_vf_spoofchk = ixgbe_ndo_set_vf_spoofchk,
6994 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
6995 .ndo_get_stats64 = ixgbe_get_stats64,
6996 #ifdef CONFIG_IXGBE_DCB
6997 .ndo_setup_tc = ixgbe_setup_tc,
6999 #ifdef CONFIG_NET_POLL_CONTROLLER
7000 .ndo_poll_controller = ixgbe_netpoll,
7003 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
7004 .ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
7005 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
7006 .ndo_fcoe_enable = ixgbe_fcoe_enable,
7007 .ndo_fcoe_disable = ixgbe_fcoe_disable,
7008 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
7009 .ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
7010 #endif /* IXGBE_FCOE */
7011 .ndo_set_features = ixgbe_set_features,
7012 .ndo_fix_features = ixgbe_fix_features,
7013 .ndo_fdb_add = ixgbe_ndo_fdb_add,
7014 .ndo_fdb_del = ixgbe_ndo_fdb_del,
7015 .ndo_fdb_dump = ixgbe_ndo_fdb_dump,
7019 * ixgbe_wol_supported - Check whether device supports WoL
7020 * @hw: hw specific details
7021 * @device_id: the device ID
7022 * @subdev_id: the subsystem device ID
7024 * This function is used by probe and ethtool to determine
7025 * which devices have WoL support
7028 int ixgbe_wol_supported(struct ixgbe_adapter *adapter, u16 device_id,
7031 struct ixgbe_hw *hw = &adapter->hw;
7032 u16 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
7033 int is_wol_supported = 0;
7035 switch (device_id) {
7036 case IXGBE_DEV_ID_82599_SFP:
7037 /* Only these subdevices could supports WOL */
7038 switch (subdevice_id) {
7039 case IXGBE_SUBDEV_ID_82599_560FLR:
7040 /* only support first port */
7041 if (hw->bus.func != 0)
7043 case IXGBE_SUBDEV_ID_82599_SFP:
7044 case IXGBE_SUBDEV_ID_82599_RNDC:
7045 is_wol_supported = 1;
7049 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
7050 /* All except this subdevice support WOL */
7051 if (subdevice_id != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
7052 is_wol_supported = 1;
7054 case IXGBE_DEV_ID_82599_KX4:
7055 is_wol_supported = 1;
7057 case IXGBE_DEV_ID_X540T:
7058 case IXGBE_DEV_ID_X540T1:
7059 /* check eeprom to see if enabled wol */
7060 if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
7061 ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
7062 (hw->bus.func == 0))) {
7063 is_wol_supported = 1;
7068 return is_wol_supported;
7072 * ixgbe_probe - Device Initialization Routine
7073 * @pdev: PCI device information struct
7074 * @ent: entry in ixgbe_pci_tbl
7076 * Returns 0 on success, negative on failure
7078 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
7079 * The OS initialization, configuring of the adapter private structure,
7080 * and a hardware reset occur.
7082 static int __devinit ixgbe_probe(struct pci_dev *pdev,
7083 const struct pci_device_id *ent)
7085 struct net_device *netdev;
7086 struct ixgbe_adapter *adapter = NULL;
7087 struct ixgbe_hw *hw;
7088 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
7089 static int cards_found;
7090 int i, err, pci_using_dac;
7091 u8 part_str[IXGBE_PBANUM_LENGTH];
7092 unsigned int indices = num_possible_cpus();
7093 unsigned int dcb_max = 0;
7099 /* Catch broken hardware that put the wrong VF device ID in
7100 * the PCIe SR-IOV capability.
7102 if (pdev->is_virtfn) {
7103 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
7104 pci_name(pdev), pdev->vendor, pdev->device);
7108 err = pci_enable_device_mem(pdev);
7112 if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
7113 !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
7116 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
7118 err = dma_set_coherent_mask(&pdev->dev,
7122 "No usable DMA configuration, aborting\n");
7129 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
7130 IORESOURCE_MEM), ixgbe_driver_name);
7133 "pci_request_selected_regions failed 0x%x\n", err);
7137 pci_enable_pcie_error_reporting(pdev);
7139 pci_set_master(pdev);
7140 pci_save_state(pdev);
7142 #ifdef CONFIG_IXGBE_DCB
7143 if (ii->mac == ixgbe_mac_82598EB)
7144 dcb_max = min_t(unsigned int, indices * MAX_TRAFFIC_CLASS,
7145 IXGBE_MAX_RSS_INDICES);
7147 dcb_max = min_t(unsigned int, indices * MAX_TRAFFIC_CLASS,
7148 IXGBE_MAX_FDIR_INDICES);
7151 if (ii->mac == ixgbe_mac_82598EB)
7152 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
7154 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
7157 indices += min_t(unsigned int, num_possible_cpus(),
7158 IXGBE_MAX_FCOE_INDICES);
7160 indices = max_t(unsigned int, dcb_max, indices);
7161 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
7164 goto err_alloc_etherdev;
7167 SET_NETDEV_DEV(netdev, &pdev->dev);
7169 adapter = netdev_priv(netdev);
7170 pci_set_drvdata(pdev, adapter);
7172 adapter->netdev = netdev;
7173 adapter->pdev = pdev;
7176 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
7178 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
7179 pci_resource_len(pdev, 0));
7185 netdev->netdev_ops = &ixgbe_netdev_ops;
7186 ixgbe_set_ethtool_ops(netdev);
7187 netdev->watchdog_timeo = 5 * HZ;
7188 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
7190 adapter->bd_number = cards_found;
7193 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
7194 hw->mac.type = ii->mac;
7197 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
7198 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
7199 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
7200 if (!(eec & (1 << 8)))
7201 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
7204 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
7205 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
7206 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
7207 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
7208 hw->phy.mdio.mmds = 0;
7209 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
7210 hw->phy.mdio.dev = netdev;
7211 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
7212 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
7214 ii->get_invariants(hw);
7216 /* setup the private structure */
7217 err = ixgbe_sw_init(adapter);
7221 /* Make it possible the adapter to be woken up via WOL */
7222 switch (adapter->hw.mac.type) {
7223 case ixgbe_mac_82599EB:
7224 case ixgbe_mac_X540:
7225 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7232 * If there is a fan on this device and it has failed log the
7235 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
7236 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
7237 if (esdp & IXGBE_ESDP_SDP1)
7238 e_crit(probe, "Fan has stopped, replace the adapter\n");
7241 if (allow_unsupported_sfp)
7242 hw->allow_unsupported_sfp = allow_unsupported_sfp;
7244 /* reset_hw fills in the perm_addr as well */
7245 hw->phy.reset_if_overtemp = true;
7246 err = hw->mac.ops.reset_hw(hw);
7247 hw->phy.reset_if_overtemp = false;
7248 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
7249 hw->mac.type == ixgbe_mac_82598EB) {
7251 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
7252 e_dev_err("failed to load because an unsupported SFP+ "
7253 "module type was detected.\n");
7254 e_dev_err("Reload the driver after installing a supported "
7258 e_dev_err("HW Init failed: %d\n", err);
7262 #ifdef CONFIG_PCI_IOV
7263 ixgbe_enable_sriov(adapter, ii);
7266 netdev->features = NETIF_F_SG |
7269 NETIF_F_HW_VLAN_TX |
7270 NETIF_F_HW_VLAN_RX |
7271 NETIF_F_HW_VLAN_FILTER |
7277 netdev->hw_features = netdev->features;
7279 switch (adapter->hw.mac.type) {
7280 case ixgbe_mac_82599EB:
7281 case ixgbe_mac_X540:
7282 netdev->features |= NETIF_F_SCTP_CSUM;
7283 netdev->hw_features |= NETIF_F_SCTP_CSUM |
7290 netdev->hw_features |= NETIF_F_RXALL;
7292 netdev->vlan_features |= NETIF_F_TSO;
7293 netdev->vlan_features |= NETIF_F_TSO6;
7294 netdev->vlan_features |= NETIF_F_IP_CSUM;
7295 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
7296 netdev->vlan_features |= NETIF_F_SG;
7298 netdev->priv_flags |= IFF_UNICAST_FLT;
7299 netdev->priv_flags |= IFF_SUPP_NOFCS;
7301 #ifdef CONFIG_IXGBE_DCB
7302 netdev->dcbnl_ops = &dcbnl_ops;
7306 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7307 if (hw->mac.ops.get_device_caps) {
7308 hw->mac.ops.get_device_caps(hw, &device_caps);
7309 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
7310 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
7313 adapter->ring_feature[RING_F_FCOE].limit = IXGBE_FCRETA_SIZE;
7315 netdev->features |= NETIF_F_FSO |
7318 netdev->vlan_features |= NETIF_F_FSO |
7322 #endif /* IXGBE_FCOE */
7323 if (pci_using_dac) {
7324 netdev->features |= NETIF_F_HIGHDMA;
7325 netdev->vlan_features |= NETIF_F_HIGHDMA;
7328 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
7329 netdev->hw_features |= NETIF_F_LRO;
7330 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
7331 netdev->features |= NETIF_F_LRO;
7333 /* make sure the EEPROM is good */
7334 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
7335 e_dev_err("The EEPROM Checksum Is Not Valid\n");
7340 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
7341 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
7343 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
7344 e_dev_err("invalid MAC address\n");
7349 setup_timer(&adapter->service_timer, &ixgbe_service_timer,
7350 (unsigned long) adapter);
7352 INIT_WORK(&adapter->service_task, ixgbe_service_task);
7353 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
7355 err = ixgbe_init_interrupt_scheme(adapter);
7359 /* WOL not supported for all devices */
7361 hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
7362 if (ixgbe_wol_supported(adapter, pdev->device, pdev->subsystem_device))
7363 adapter->wol = IXGBE_WUFC_MAG;
7365 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
7367 #ifdef CONFIG_IXGBE_PTP
7368 ixgbe_ptp_init(adapter);
7369 #endif /* CONFIG_IXGBE_PTP*/
7371 /* save off EEPROM version number */
7372 hw->eeprom.ops.read(hw, 0x2e, &adapter->eeprom_verh);
7373 hw->eeprom.ops.read(hw, 0x2d, &adapter->eeprom_verl);
7375 /* pick up the PCI bus settings for reporting later */
7376 hw->mac.ops.get_bus_info(hw);
7378 /* print bus type/speed/width info */
7379 e_dev_info("(PCI Express:%s:%s) %pM\n",
7380 (hw->bus.speed == ixgbe_bus_speed_5000 ? "5.0GT/s" :
7381 hw->bus.speed == ixgbe_bus_speed_2500 ? "2.5GT/s" :
7383 (hw->bus.width == ixgbe_bus_width_pcie_x8 ? "Width x8" :
7384 hw->bus.width == ixgbe_bus_width_pcie_x4 ? "Width x4" :
7385 hw->bus.width == ixgbe_bus_width_pcie_x1 ? "Width x1" :
7389 err = ixgbe_read_pba_string_generic(hw, part_str, IXGBE_PBANUM_LENGTH);
7391 strncpy(part_str, "Unknown", IXGBE_PBANUM_LENGTH);
7392 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
7393 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
7394 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
7397 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
7398 hw->mac.type, hw->phy.type, part_str);
7400 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
7401 e_dev_warn("PCI-Express bandwidth available for this card is "
7402 "not sufficient for optimal performance.\n");
7403 e_dev_warn("For optimal performance a x8 PCI-Express slot "
7407 /* reset the hardware with the new settings */
7408 err = hw->mac.ops.start_hw(hw);
7409 if (err == IXGBE_ERR_EEPROM_VERSION) {
7410 /* We are running on a pre-production device, log a warning */
7411 e_dev_warn("This device is a pre-production adapter/LOM. "
7412 "Please be aware there may be issues associated "
7413 "with your hardware. If you are experiencing "
7414 "problems please contact your Intel or hardware "
7415 "representative who provided you with this "
7418 strcpy(netdev->name, "eth%d");
7419 err = register_netdev(netdev);
7423 /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
7424 if (hw->mac.ops.disable_tx_laser &&
7425 ((hw->phy.multispeed_fiber) ||
7426 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
7427 (hw->mac.type == ixgbe_mac_82599EB))))
7428 hw->mac.ops.disable_tx_laser(hw);
7430 /* carrier off reporting is important to ethtool even BEFORE open */
7431 netif_carrier_off(netdev);
7433 #ifdef CONFIG_IXGBE_DCA
7434 if (dca_add_requester(&pdev->dev) == 0) {
7435 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
7436 ixgbe_setup_dca(adapter);
7439 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7440 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
7441 for (i = 0; i < adapter->num_vfs; i++)
7442 ixgbe_vf_configuration(pdev, (i | 0x10000000));
7445 /* firmware requires driver version to be 0xFFFFFFFF
7446 * since os does not support feature
7448 if (hw->mac.ops.set_fw_drv_ver)
7449 hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF,
7452 /* add san mac addr to netdev */
7453 ixgbe_add_sanmac_netdev(netdev);
7455 e_dev_info("%s\n", ixgbe_default_device_descr);
7458 #ifdef CONFIG_IXGBE_HWMON
7459 if (ixgbe_sysfs_init(adapter))
7460 e_err(probe, "failed to allocate sysfs resources\n");
7461 #endif /* CONFIG_IXGBE_HWMON */
7463 #ifdef CONFIG_DEBUG_FS
7464 ixgbe_dbg_adapter_init(adapter);
7465 #endif /* CONFIG_DEBUG_FS */
7470 ixgbe_release_hw_control(adapter);
7471 ixgbe_clear_interrupt_scheme(adapter);
7473 ixgbe_disable_sriov(adapter);
7474 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
7475 iounmap(hw->hw_addr);
7477 free_netdev(netdev);
7479 pci_release_selected_regions(pdev,
7480 pci_select_bars(pdev, IORESOURCE_MEM));
7483 pci_disable_device(pdev);
7488 * ixgbe_remove - Device Removal Routine
7489 * @pdev: PCI device information struct
7491 * ixgbe_remove is called by the PCI subsystem to alert the driver
7492 * that it should release a PCI device. The could be caused by a
7493 * Hot-Plug event, or because the driver is going to be removed from
7496 static void __devexit ixgbe_remove(struct pci_dev *pdev)
7498 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7499 struct net_device *netdev = adapter->netdev;
7501 #ifdef CONFIG_DEBUG_FS
7502 ixgbe_dbg_adapter_exit(adapter);
7503 #endif /*CONFIG_DEBUG_FS */
7505 set_bit(__IXGBE_DOWN, &adapter->state);
7506 cancel_work_sync(&adapter->service_task);
7508 #ifdef CONFIG_IXGBE_PTP
7509 ixgbe_ptp_stop(adapter);
7512 #ifdef CONFIG_IXGBE_DCA
7513 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
7514 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
7515 dca_remove_requester(&pdev->dev);
7516 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
7520 #ifdef CONFIG_IXGBE_HWMON
7521 ixgbe_sysfs_exit(adapter);
7522 #endif /* CONFIG_IXGBE_HWMON */
7524 /* remove the added san mac */
7525 ixgbe_del_sanmac_netdev(netdev);
7527 if (netdev->reg_state == NETREG_REGISTERED)
7528 unregister_netdev(netdev);
7530 ixgbe_disable_sriov(adapter);
7532 ixgbe_clear_interrupt_scheme(adapter);
7534 ixgbe_release_hw_control(adapter);
7537 kfree(adapter->ixgbe_ieee_pfc);
7538 kfree(adapter->ixgbe_ieee_ets);
7541 iounmap(adapter->hw.hw_addr);
7542 pci_release_selected_regions(pdev, pci_select_bars(pdev,
7545 e_dev_info("complete\n");
7547 free_netdev(netdev);
7549 pci_disable_pcie_error_reporting(pdev);
7551 pci_disable_device(pdev);
7555 * ixgbe_io_error_detected - called when PCI error is detected
7556 * @pdev: Pointer to PCI device
7557 * @state: The current pci connection state
7559 * This function is called after a PCI bus error affecting
7560 * this device has been detected.
7562 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
7563 pci_channel_state_t state)
7565 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7566 struct net_device *netdev = adapter->netdev;
7568 #ifdef CONFIG_PCI_IOV
7569 struct pci_dev *bdev, *vfdev;
7570 u32 dw0, dw1, dw2, dw3;
7572 u16 req_id, pf_func;
7574 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
7575 adapter->num_vfs == 0)
7576 goto skip_bad_vf_detection;
7578 bdev = pdev->bus->self;
7579 while (bdev && (pci_pcie_type(bdev) != PCI_EXP_TYPE_ROOT_PORT))
7580 bdev = bdev->bus->self;
7583 goto skip_bad_vf_detection;
7585 pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
7587 goto skip_bad_vf_detection;
7589 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG, &dw0);
7590 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 4, &dw1);
7591 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 8, &dw2);
7592 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 12, &dw3);
7595 /* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
7596 if (!(req_id & 0x0080))
7597 goto skip_bad_vf_detection;
7599 pf_func = req_id & 0x01;
7600 if ((pf_func & 1) == (pdev->devfn & 1)) {
7601 unsigned int device_id;
7603 vf = (req_id & 0x7F) >> 1;
7604 e_dev_err("VF %d has caused a PCIe error\n", vf);
7605 e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
7606 "%8.8x\tdw3: %8.8x\n",
7607 dw0, dw1, dw2, dw3);
7608 switch (adapter->hw.mac.type) {
7609 case ixgbe_mac_82599EB:
7610 device_id = IXGBE_82599_VF_DEVICE_ID;
7612 case ixgbe_mac_X540:
7613 device_id = IXGBE_X540_VF_DEVICE_ID;
7620 /* Find the pci device of the offending VF */
7621 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, device_id, NULL);
7623 if (vfdev->devfn == (req_id & 0xFF))
7625 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL,
7629 * There's a slim chance the VF could have been hot plugged,
7630 * so if it is no longer present we don't need to issue the
7631 * VFLR. Just clean up the AER in that case.
7634 e_dev_err("Issuing VFLR to VF %d\n", vf);
7635 pci_write_config_dword(vfdev, 0xA8, 0x00008000);
7638 pci_cleanup_aer_uncorrect_error_status(pdev);
7642 * Even though the error may have occurred on the other port
7643 * we still need to increment the vf error reference count for
7644 * both ports because the I/O resume function will be called
7647 adapter->vferr_refcount++;
7649 return PCI_ERS_RESULT_RECOVERED;
7651 skip_bad_vf_detection:
7652 #endif /* CONFIG_PCI_IOV */
7653 netif_device_detach(netdev);
7655 if (state == pci_channel_io_perm_failure)
7656 return PCI_ERS_RESULT_DISCONNECT;
7658 if (netif_running(netdev))
7659 ixgbe_down(adapter);
7660 pci_disable_device(pdev);
7662 /* Request a slot reset. */
7663 return PCI_ERS_RESULT_NEED_RESET;
7667 * ixgbe_io_slot_reset - called after the pci bus has been reset.
7668 * @pdev: Pointer to PCI device
7670 * Restart the card from scratch, as if from a cold-boot.
7672 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7674 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7675 pci_ers_result_t result;
7678 if (pci_enable_device_mem(pdev)) {
7679 e_err(probe, "Cannot re-enable PCI device after reset.\n");
7680 result = PCI_ERS_RESULT_DISCONNECT;
7682 pci_set_master(pdev);
7683 pci_restore_state(pdev);
7684 pci_save_state(pdev);
7686 pci_wake_from_d3(pdev, false);
7688 ixgbe_reset(adapter);
7689 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7690 result = PCI_ERS_RESULT_RECOVERED;
7693 err = pci_cleanup_aer_uncorrect_error_status(pdev);
7695 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7696 "failed 0x%0x\n", err);
7697 /* non-fatal, continue */
7704 * ixgbe_io_resume - called when traffic can start flowing again.
7705 * @pdev: Pointer to PCI device
7707 * This callback is called when the error recovery driver tells us that
7708 * its OK to resume normal operation.
7710 static void ixgbe_io_resume(struct pci_dev *pdev)
7712 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7713 struct net_device *netdev = adapter->netdev;
7715 #ifdef CONFIG_PCI_IOV
7716 if (adapter->vferr_refcount) {
7717 e_info(drv, "Resuming after VF err\n");
7718 adapter->vferr_refcount--;
7723 if (netif_running(netdev))
7726 netif_device_attach(netdev);
7729 static const struct pci_error_handlers ixgbe_err_handler = {
7730 .error_detected = ixgbe_io_error_detected,
7731 .slot_reset = ixgbe_io_slot_reset,
7732 .resume = ixgbe_io_resume,
7735 static struct pci_driver ixgbe_driver = {
7736 .name = ixgbe_driver_name,
7737 .id_table = ixgbe_pci_tbl,
7738 .probe = ixgbe_probe,
7739 .remove = __devexit_p(ixgbe_remove),
7741 .suspend = ixgbe_suspend,
7742 .resume = ixgbe_resume,
7744 .shutdown = ixgbe_shutdown,
7745 .err_handler = &ixgbe_err_handler
7749 * ixgbe_init_module - Driver Registration Routine
7751 * ixgbe_init_module is the first routine called when the driver is
7752 * loaded. All it does is register with the PCI subsystem.
7754 static int __init ixgbe_init_module(void)
7757 pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
7758 pr_info("%s\n", ixgbe_copyright);
7760 #ifdef CONFIG_DEBUG_FS
7762 #endif /* CONFIG_DEBUG_FS */
7764 #ifdef CONFIG_IXGBE_DCA
7765 dca_register_notify(&dca_notifier);
7768 ret = pci_register_driver(&ixgbe_driver);
7772 module_init(ixgbe_init_module);
7775 * ixgbe_exit_module - Driver Exit Cleanup Routine
7777 * ixgbe_exit_module is called just before the driver is removed
7780 static void __exit ixgbe_exit_module(void)
7782 #ifdef CONFIG_IXGBE_DCA
7783 dca_unregister_notify(&dca_notifier);
7785 pci_unregister_driver(&ixgbe_driver);
7787 #ifdef CONFIG_DEBUG_FS
7789 #endif /* CONFIG_DEBUG_FS */
7791 rcu_barrier(); /* Wait for completion of call_rcu()'s */
7794 #ifdef CONFIG_IXGBE_DCA
7795 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
7800 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
7801 __ixgbe_notify_dca);
7803 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7806 #endif /* CONFIG_IXGBE_DCA */
7808 module_exit(ixgbe_exit_module);