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/if_bridge.h>
48 #include <linux/prefetch.h>
49 #include <scsi/fc/fc_fcoe.h>
52 #include "ixgbe_common.h"
53 #include "ixgbe_dcb_82599.h"
54 #include "ixgbe_sriov.h"
56 char ixgbe_driver_name[] = "ixgbe";
57 static const char ixgbe_driver_string[] =
58 "Intel(R) 10 Gigabit PCI Express Network Driver";
60 char ixgbe_default_device_descr[] =
61 "Intel(R) 10 Gigabit Network Connection";
63 static char ixgbe_default_device_descr[] =
64 "Intel(R) 10 Gigabit Network Connection";
66 #define DRV_VERSION "3.11.33-k"
67 const char ixgbe_driver_version[] = DRV_VERSION;
68 static const char ixgbe_copyright[] =
69 "Copyright (c) 1999-2012 Intel Corporation.";
71 static const struct ixgbe_info *ixgbe_info_tbl[] = {
72 [board_82598] = &ixgbe_82598_info,
73 [board_82599] = &ixgbe_82599_info,
74 [board_X540] = &ixgbe_X540_info,
77 /* ixgbe_pci_tbl - PCI Device ID Table
79 * Wildcard entries (PCI_ANY_ID) should come last
80 * Last entry must be all 0s
82 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
83 * Class, Class Mask, private data (not used) }
85 static DEFINE_PCI_DEVICE_TABLE(ixgbe_pci_tbl) = {
86 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598), board_82598 },
87 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT), board_82598 },
88 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT), board_82598 },
89 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT), board_82598 },
90 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2), board_82598 },
91 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4), board_82598 },
92 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT), board_82598 },
93 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT), board_82598 },
94 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM), board_82598 },
95 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR), board_82598 },
96 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM), board_82598 },
97 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX), board_82598 },
98 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4), board_82599 },
99 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM), board_82599 },
100 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR), board_82599 },
101 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP), board_82599 },
102 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM), board_82599 },
103 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ), board_82599 },
104 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4), board_82599 },
105 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE), board_82599 },
106 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE), board_82599 },
107 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM), board_82599 },
108 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE), board_82599 },
109 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T), board_X540 },
110 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF2), board_82599 },
111 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_LS), board_82599 },
112 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599EN_SFP), board_82599 },
113 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF_QP), board_82599 },
114 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T1), board_X540 },
115 /* required last entry */
118 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
120 #ifdef CONFIG_IXGBE_DCA
121 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
123 static struct notifier_block dca_notifier = {
124 .notifier_call = ixgbe_notify_dca,
130 #ifdef CONFIG_PCI_IOV
131 static unsigned int max_vfs;
132 module_param(max_vfs, uint, 0);
133 MODULE_PARM_DESC(max_vfs,
134 "Maximum number of virtual functions to allocate per physical function - default is zero and maximum value is 63");
135 #endif /* CONFIG_PCI_IOV */
137 static unsigned int allow_unsupported_sfp;
138 module_param(allow_unsupported_sfp, uint, 0);
139 MODULE_PARM_DESC(allow_unsupported_sfp,
140 "Allow unsupported and untested SFP+ modules on 82599-based adapters");
142 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
143 static int debug = -1;
144 module_param(debug, int, 0);
145 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
147 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
148 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
149 MODULE_LICENSE("GPL");
150 MODULE_VERSION(DRV_VERSION);
152 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
154 if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
155 !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
156 schedule_work(&adapter->service_task);
159 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
161 BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
163 /* flush memory to make sure state is correct before next watchdog */
164 smp_mb__before_clear_bit();
165 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
168 struct ixgbe_reg_info {
173 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
175 /* General Registers */
176 {IXGBE_CTRL, "CTRL"},
177 {IXGBE_STATUS, "STATUS"},
178 {IXGBE_CTRL_EXT, "CTRL_EXT"},
180 /* Interrupt Registers */
181 {IXGBE_EICR, "EICR"},
184 {IXGBE_SRRCTL(0), "SRRCTL"},
185 {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
186 {IXGBE_RDLEN(0), "RDLEN"},
187 {IXGBE_RDH(0), "RDH"},
188 {IXGBE_RDT(0), "RDT"},
189 {IXGBE_RXDCTL(0), "RXDCTL"},
190 {IXGBE_RDBAL(0), "RDBAL"},
191 {IXGBE_RDBAH(0), "RDBAH"},
194 {IXGBE_TDBAL(0), "TDBAL"},
195 {IXGBE_TDBAH(0), "TDBAH"},
196 {IXGBE_TDLEN(0), "TDLEN"},
197 {IXGBE_TDH(0), "TDH"},
198 {IXGBE_TDT(0), "TDT"},
199 {IXGBE_TXDCTL(0), "TXDCTL"},
201 /* List Terminator */
207 * ixgbe_regdump - register printout routine
209 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
215 switch (reginfo->ofs) {
216 case IXGBE_SRRCTL(0):
217 for (i = 0; i < 64; i++)
218 regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
220 case IXGBE_DCA_RXCTRL(0):
221 for (i = 0; i < 64; i++)
222 regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
225 for (i = 0; i < 64; i++)
226 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
229 for (i = 0; i < 64; i++)
230 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
233 for (i = 0; i < 64; i++)
234 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
236 case IXGBE_RXDCTL(0):
237 for (i = 0; i < 64; i++)
238 regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
241 for (i = 0; i < 64; i++)
242 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
245 for (i = 0; i < 64; i++)
246 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
249 for (i = 0; i < 64; i++)
250 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
253 for (i = 0; i < 64; i++)
254 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
257 for (i = 0; i < 64; i++)
258 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
261 for (i = 0; i < 64; i++)
262 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
265 for (i = 0; i < 64; i++)
266 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
268 case IXGBE_TXDCTL(0):
269 for (i = 0; i < 64; i++)
270 regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
273 pr_info("%-15s %08x\n", reginfo->name,
274 IXGBE_READ_REG(hw, reginfo->ofs));
278 for (i = 0; i < 8; i++) {
279 snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i*8, i*8+7);
280 pr_err("%-15s", rname);
281 for (j = 0; j < 8; j++)
282 pr_cont(" %08x", regs[i*8+j]);
289 * ixgbe_dump - Print registers, tx-rings and rx-rings
291 static void ixgbe_dump(struct ixgbe_adapter *adapter)
293 struct net_device *netdev = adapter->netdev;
294 struct ixgbe_hw *hw = &adapter->hw;
295 struct ixgbe_reg_info *reginfo;
297 struct ixgbe_ring *tx_ring;
298 struct ixgbe_tx_buffer *tx_buffer;
299 union ixgbe_adv_tx_desc *tx_desc;
300 struct my_u0 { u64 a; u64 b; } *u0;
301 struct ixgbe_ring *rx_ring;
302 union ixgbe_adv_rx_desc *rx_desc;
303 struct ixgbe_rx_buffer *rx_buffer_info;
307 if (!netif_msg_hw(adapter))
310 /* Print netdevice Info */
312 dev_info(&adapter->pdev->dev, "Net device Info\n");
313 pr_info("Device Name state "
314 "trans_start last_rx\n");
315 pr_info("%-15s %016lX %016lX %016lX\n",
322 /* Print Registers */
323 dev_info(&adapter->pdev->dev, "Register Dump\n");
324 pr_info(" Register Name Value\n");
325 for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
326 reginfo->name; reginfo++) {
327 ixgbe_regdump(hw, reginfo);
330 /* Print TX Ring Summary */
331 if (!netdev || !netif_running(netdev))
334 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
335 pr_info(" %s %s %s %s\n",
336 "Queue [NTU] [NTC] [bi(ntc)->dma ]",
337 "leng", "ntw", "timestamp");
338 for (n = 0; n < adapter->num_tx_queues; n++) {
339 tx_ring = adapter->tx_ring[n];
340 tx_buffer = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
341 pr_info(" %5d %5X %5X %016llX %08X %p %016llX\n",
342 n, tx_ring->next_to_use, tx_ring->next_to_clean,
343 (u64)dma_unmap_addr(tx_buffer, dma),
344 dma_unmap_len(tx_buffer, len),
345 tx_buffer->next_to_watch,
346 (u64)tx_buffer->time_stamp);
350 if (!netif_msg_tx_done(adapter))
351 goto rx_ring_summary;
353 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
355 /* Transmit Descriptor Formats
357 * 82598 Advanced Transmit Descriptor
358 * +--------------------------------------------------------------+
359 * 0 | Buffer Address [63:0] |
360 * +--------------------------------------------------------------+
361 * 8 | PAYLEN | POPTS | IDX | STA | DCMD |DTYP | RSV | DTALEN |
362 * +--------------------------------------------------------------+
363 * 63 46 45 40 39 36 35 32 31 24 23 20 19 0
365 * 82598 Advanced Transmit Descriptor (Write-Back Format)
366 * +--------------------------------------------------------------+
368 * +--------------------------------------------------------------+
369 * 8 | RSV | STA | NXTSEQ |
370 * +--------------------------------------------------------------+
373 * 82599+ Advanced Transmit Descriptor
374 * +--------------------------------------------------------------+
375 * 0 | Buffer Address [63:0] |
376 * +--------------------------------------------------------------+
377 * 8 |PAYLEN |POPTS|CC|IDX |STA |DCMD |DTYP |MAC |RSV |DTALEN |
378 * +--------------------------------------------------------------+
379 * 63 46 45 40 39 38 36 35 32 31 24 23 20 19 18 17 16 15 0
381 * 82599+ Advanced Transmit Descriptor (Write-Back Format)
382 * +--------------------------------------------------------------+
384 * +--------------------------------------------------------------+
385 * 8 | RSV | STA | RSV |
386 * +--------------------------------------------------------------+
390 for (n = 0; n < adapter->num_tx_queues; n++) {
391 tx_ring = adapter->tx_ring[n];
392 pr_info("------------------------------------\n");
393 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
394 pr_info("------------------------------------\n");
395 pr_info("%s%s %s %s %s %s\n",
396 "T [desc] [address 63:0 ] ",
397 "[PlPOIdStDDt Ln] [bi->dma ] ",
398 "leng", "ntw", "timestamp", "bi->skb");
400 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
401 tx_desc = IXGBE_TX_DESC(tx_ring, i);
402 tx_buffer = &tx_ring->tx_buffer_info[i];
403 u0 = (struct my_u0 *)tx_desc;
404 if (dma_unmap_len(tx_buffer, len) > 0) {
405 pr_info("T [0x%03X] %016llX %016llX %016llX %08X %p %016llX %p",
409 (u64)dma_unmap_addr(tx_buffer, dma),
410 dma_unmap_len(tx_buffer, len),
411 tx_buffer->next_to_watch,
412 (u64)tx_buffer->time_stamp,
414 if (i == tx_ring->next_to_use &&
415 i == tx_ring->next_to_clean)
417 else if (i == tx_ring->next_to_use)
419 else if (i == tx_ring->next_to_clean)
424 if (netif_msg_pktdata(adapter) &&
426 print_hex_dump(KERN_INFO, "",
427 DUMP_PREFIX_ADDRESS, 16, 1,
428 tx_buffer->skb->data,
429 dma_unmap_len(tx_buffer, len),
435 /* Print RX Rings Summary */
437 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
438 pr_info("Queue [NTU] [NTC]\n");
439 for (n = 0; n < adapter->num_rx_queues; n++) {
440 rx_ring = adapter->rx_ring[n];
441 pr_info("%5d %5X %5X\n",
442 n, rx_ring->next_to_use, rx_ring->next_to_clean);
446 if (!netif_msg_rx_status(adapter))
449 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
451 /* Receive Descriptor Formats
453 * 82598 Advanced Receive Descriptor (Read) Format
455 * +-----------------------------------------------------+
456 * 0 | Packet Buffer Address [63:1] |A0/NSE|
457 * +----------------------------------------------+------+
458 * 8 | Header Buffer Address [63:1] | DD |
459 * +-----------------------------------------------------+
462 * 82598 Advanced Receive Descriptor (Write-Back) Format
464 * 63 48 47 32 31 30 21 20 16 15 4 3 0
465 * +------------------------------------------------------+
466 * 0 | RSS Hash / |SPH| HDR_LEN | RSV |Packet| RSS |
467 * | Packet | IP | | | | Type | Type |
468 * | Checksum | Ident | | | | | |
469 * +------------------------------------------------------+
470 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
471 * +------------------------------------------------------+
472 * 63 48 47 32 31 20 19 0
474 * 82599+ Advanced Receive Descriptor (Read) Format
476 * +-----------------------------------------------------+
477 * 0 | Packet Buffer Address [63:1] |A0/NSE|
478 * +----------------------------------------------+------+
479 * 8 | Header Buffer Address [63:1] | DD |
480 * +-----------------------------------------------------+
483 * 82599+ Advanced Receive Descriptor (Write-Back) Format
485 * 63 48 47 32 31 30 21 20 17 16 4 3 0
486 * +------------------------------------------------------+
487 * 0 |RSS / Frag Checksum|SPH| HDR_LEN |RSC- |Packet| RSS |
488 * |/ RTT / PCoE_PARAM | | | CNT | Type | Type |
489 * |/ Flow Dir Flt ID | | | | | |
490 * +------------------------------------------------------+
491 * 8 | VLAN Tag | Length |Extended Error| Xtnd Status/NEXTP |
492 * +------------------------------------------------------+
493 * 63 48 47 32 31 20 19 0
496 for (n = 0; n < adapter->num_rx_queues; n++) {
497 rx_ring = adapter->rx_ring[n];
498 pr_info("------------------------------------\n");
499 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
500 pr_info("------------------------------------\n");
502 "R [desc] [ PktBuf A0] ",
503 "[ HeadBuf DD] [bi->dma ] [bi->skb ] ",
504 "<-- Adv Rx Read format\n");
506 "RWB[desc] [PcsmIpSHl PtRs] ",
507 "[vl er S cks ln] ---------------- [bi->skb ] ",
508 "<-- Adv Rx Write-Back format\n");
510 for (i = 0; i < rx_ring->count; i++) {
511 rx_buffer_info = &rx_ring->rx_buffer_info[i];
512 rx_desc = IXGBE_RX_DESC(rx_ring, i);
513 u0 = (struct my_u0 *)rx_desc;
514 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
515 if (staterr & IXGBE_RXD_STAT_DD) {
516 /* Descriptor Done */
517 pr_info("RWB[0x%03X] %016llX "
518 "%016llX ---------------- %p", i,
521 rx_buffer_info->skb);
523 pr_info("R [0x%03X] %016llX "
524 "%016llX %016llX %p", i,
527 (u64)rx_buffer_info->dma,
528 rx_buffer_info->skb);
530 if (netif_msg_pktdata(adapter) &&
531 rx_buffer_info->dma) {
532 print_hex_dump(KERN_INFO, "",
533 DUMP_PREFIX_ADDRESS, 16, 1,
534 page_address(rx_buffer_info->page) +
535 rx_buffer_info->page_offset,
536 ixgbe_rx_bufsz(rx_ring), true);
540 if (i == rx_ring->next_to_use)
542 else if (i == rx_ring->next_to_clean)
554 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
558 /* Let firmware take over control of h/w */
559 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
560 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
561 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
564 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
568 /* Let firmware know the driver has taken over */
569 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
570 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
571 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
575 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
576 * @adapter: pointer to adapter struct
577 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
578 * @queue: queue to map the corresponding interrupt to
579 * @msix_vector: the vector to map to the corresponding queue
582 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
583 u8 queue, u8 msix_vector)
586 struct ixgbe_hw *hw = &adapter->hw;
587 switch (hw->mac.type) {
588 case ixgbe_mac_82598EB:
589 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
592 index = (((direction * 64) + queue) >> 2) & 0x1F;
593 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
594 ivar &= ~(0xFF << (8 * (queue & 0x3)));
595 ivar |= (msix_vector << (8 * (queue & 0x3)));
596 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
598 case ixgbe_mac_82599EB:
600 if (direction == -1) {
602 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
603 index = ((queue & 1) * 8);
604 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
605 ivar &= ~(0xFF << index);
606 ivar |= (msix_vector << index);
607 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
610 /* tx or rx causes */
611 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
612 index = ((16 * (queue & 1)) + (8 * direction));
613 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
614 ivar &= ~(0xFF << index);
615 ivar |= (msix_vector << index);
616 IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
624 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
629 switch (adapter->hw.mac.type) {
630 case ixgbe_mac_82598EB:
631 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
632 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
634 case ixgbe_mac_82599EB:
636 mask = (qmask & 0xFFFFFFFF);
637 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
638 mask = (qmask >> 32);
639 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
646 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_ring *ring,
647 struct ixgbe_tx_buffer *tx_buffer)
649 if (tx_buffer->skb) {
650 dev_kfree_skb_any(tx_buffer->skb);
651 if (dma_unmap_len(tx_buffer, len))
652 dma_unmap_single(ring->dev,
653 dma_unmap_addr(tx_buffer, dma),
654 dma_unmap_len(tx_buffer, len),
656 } else if (dma_unmap_len(tx_buffer, len)) {
657 dma_unmap_page(ring->dev,
658 dma_unmap_addr(tx_buffer, dma),
659 dma_unmap_len(tx_buffer, len),
662 tx_buffer->next_to_watch = NULL;
663 tx_buffer->skb = NULL;
664 dma_unmap_len_set(tx_buffer, len, 0);
665 /* tx_buffer must be completely set up in the transmit path */
668 static void ixgbe_update_xoff_rx_lfc(struct ixgbe_adapter *adapter)
670 struct ixgbe_hw *hw = &adapter->hw;
671 struct ixgbe_hw_stats *hwstats = &adapter->stats;
675 if ((hw->fc.current_mode != ixgbe_fc_full) &&
676 (hw->fc.current_mode != ixgbe_fc_rx_pause))
679 switch (hw->mac.type) {
680 case ixgbe_mac_82598EB:
681 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
684 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
686 hwstats->lxoffrxc += data;
688 /* refill credits (no tx hang) if we received xoff */
692 for (i = 0; i < adapter->num_tx_queues; i++)
693 clear_bit(__IXGBE_HANG_CHECK_ARMED,
694 &adapter->tx_ring[i]->state);
697 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
699 struct ixgbe_hw *hw = &adapter->hw;
700 struct ixgbe_hw_stats *hwstats = &adapter->stats;
704 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
706 if (adapter->ixgbe_ieee_pfc)
707 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
709 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED) || !pfc_en) {
710 ixgbe_update_xoff_rx_lfc(adapter);
714 /* update stats for each tc, only valid with PFC enabled */
715 for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
718 switch (hw->mac.type) {
719 case ixgbe_mac_82598EB:
720 pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
723 pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
725 hwstats->pxoffrxc[i] += pxoffrxc;
726 /* Get the TC for given UP */
727 tc = netdev_get_prio_tc_map(adapter->netdev, i);
728 xoff[tc] += pxoffrxc;
731 /* disarm tx queues that have received xoff frames */
732 for (i = 0; i < adapter->num_tx_queues; i++) {
733 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
735 tc = tx_ring->dcb_tc;
737 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
741 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
743 return ring->stats.packets;
746 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
748 struct ixgbe_adapter *adapter = netdev_priv(ring->netdev);
749 struct ixgbe_hw *hw = &adapter->hw;
751 u32 head = IXGBE_READ_REG(hw, IXGBE_TDH(ring->reg_idx));
752 u32 tail = IXGBE_READ_REG(hw, IXGBE_TDT(ring->reg_idx));
755 return (head < tail) ?
756 tail - head : (tail + ring->count - head);
761 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
763 u32 tx_done = ixgbe_get_tx_completed(tx_ring);
764 u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
765 u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
768 clear_check_for_tx_hang(tx_ring);
771 * Check for a hung queue, but be thorough. This verifies
772 * that a transmit has been completed since the previous
773 * check AND there is at least one packet pending. The
774 * ARMED bit is set to indicate a potential hang. The
775 * bit is cleared if a pause frame is received to remove
776 * false hang detection due to PFC or 802.3x frames. By
777 * requiring this to fail twice we avoid races with
778 * pfc clearing the ARMED bit and conditions where we
779 * run the check_tx_hang logic with a transmit completion
780 * pending but without time to complete it yet.
782 if ((tx_done_old == tx_done) && tx_pending) {
783 /* make sure it is true for two checks in a row */
784 ret = test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
787 /* update completed stats and continue */
788 tx_ring->tx_stats.tx_done_old = tx_done;
789 /* reset the countdown */
790 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
797 * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
798 * @adapter: driver private struct
800 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
803 /* Do the reset outside of interrupt context */
804 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
805 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
806 ixgbe_service_event_schedule(adapter);
811 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
812 * @q_vector: structure containing interrupt and ring information
813 * @tx_ring: tx ring to clean
815 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
816 struct ixgbe_ring *tx_ring)
818 struct ixgbe_adapter *adapter = q_vector->adapter;
819 struct ixgbe_tx_buffer *tx_buffer;
820 union ixgbe_adv_tx_desc *tx_desc;
821 unsigned int total_bytes = 0, total_packets = 0;
822 unsigned int budget = q_vector->tx.work_limit;
823 unsigned int i = tx_ring->next_to_clean;
825 if (test_bit(__IXGBE_DOWN, &adapter->state))
828 tx_buffer = &tx_ring->tx_buffer_info[i];
829 tx_desc = IXGBE_TX_DESC(tx_ring, i);
833 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
835 /* if next_to_watch is not set then there is no work pending */
839 /* prevent any other reads prior to eop_desc */
842 /* if DD is not set pending work has not been completed */
843 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
846 /* clear next_to_watch to prevent false hangs */
847 tx_buffer->next_to_watch = NULL;
849 /* update the statistics for this packet */
850 total_bytes += tx_buffer->bytecount;
851 total_packets += tx_buffer->gso_segs;
853 if (unlikely(tx_buffer->tx_flags & IXGBE_TX_FLAGS_TSTAMP))
854 ixgbe_ptp_tx_hwtstamp(q_vector, tx_buffer->skb);
857 dev_kfree_skb_any(tx_buffer->skb);
859 /* unmap skb header data */
860 dma_unmap_single(tx_ring->dev,
861 dma_unmap_addr(tx_buffer, dma),
862 dma_unmap_len(tx_buffer, len),
865 /* clear tx_buffer data */
866 tx_buffer->skb = NULL;
867 dma_unmap_len_set(tx_buffer, len, 0);
869 /* unmap remaining buffers */
870 while (tx_desc != eop_desc) {
876 tx_buffer = tx_ring->tx_buffer_info;
877 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
880 /* unmap any remaining paged data */
881 if (dma_unmap_len(tx_buffer, len)) {
882 dma_unmap_page(tx_ring->dev,
883 dma_unmap_addr(tx_buffer, dma),
884 dma_unmap_len(tx_buffer, len),
886 dma_unmap_len_set(tx_buffer, len, 0);
890 /* move us one more past the eop_desc for start of next pkt */
896 tx_buffer = tx_ring->tx_buffer_info;
897 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
900 /* issue prefetch for next Tx descriptor */
903 /* update budget accounting */
905 } while (likely(budget));
908 tx_ring->next_to_clean = i;
909 u64_stats_update_begin(&tx_ring->syncp);
910 tx_ring->stats.bytes += total_bytes;
911 tx_ring->stats.packets += total_packets;
912 u64_stats_update_end(&tx_ring->syncp);
913 q_vector->tx.total_bytes += total_bytes;
914 q_vector->tx.total_packets += total_packets;
916 if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
917 /* schedule immediate reset if we believe we hung */
918 struct ixgbe_hw *hw = &adapter->hw;
919 e_err(drv, "Detected Tx Unit Hang\n"
921 " TDH, TDT <%x>, <%x>\n"
922 " next_to_use <%x>\n"
923 " next_to_clean <%x>\n"
924 "tx_buffer_info[next_to_clean]\n"
925 " time_stamp <%lx>\n"
927 tx_ring->queue_index,
928 IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
929 IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
930 tx_ring->next_to_use, i,
931 tx_ring->tx_buffer_info[i].time_stamp, jiffies);
933 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
936 "tx hang %d detected on queue %d, resetting adapter\n",
937 adapter->tx_timeout_count + 1, tx_ring->queue_index);
939 /* schedule immediate reset if we believe we hung */
940 ixgbe_tx_timeout_reset(adapter);
942 /* the adapter is about to reset, no point in enabling stuff */
946 netdev_tx_completed_queue(txring_txq(tx_ring),
947 total_packets, total_bytes);
949 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
950 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
951 (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
952 /* Make sure that anybody stopping the queue after this
953 * sees the new next_to_clean.
956 if (__netif_subqueue_stopped(tx_ring->netdev,
957 tx_ring->queue_index)
958 && !test_bit(__IXGBE_DOWN, &adapter->state)) {
959 netif_wake_subqueue(tx_ring->netdev,
960 tx_ring->queue_index);
961 ++tx_ring->tx_stats.restart_queue;
968 #ifdef CONFIG_IXGBE_DCA
969 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
970 struct ixgbe_ring *tx_ring,
973 struct ixgbe_hw *hw = &adapter->hw;
974 u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
977 switch (hw->mac.type) {
978 case ixgbe_mac_82598EB:
979 reg_offset = IXGBE_DCA_TXCTRL(tx_ring->reg_idx);
981 case ixgbe_mac_82599EB:
983 reg_offset = IXGBE_DCA_TXCTRL_82599(tx_ring->reg_idx);
984 txctrl <<= IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599;
987 /* for unknown hardware do not write register */
992 * We can enable relaxed ordering for reads, but not writes when
993 * DCA is enabled. This is due to a known issue in some chipsets
994 * which will cause the DCA tag to be cleared.
996 txctrl |= IXGBE_DCA_TXCTRL_DESC_RRO_EN |
997 IXGBE_DCA_TXCTRL_DATA_RRO_EN |
998 IXGBE_DCA_TXCTRL_DESC_DCA_EN;
1000 IXGBE_WRITE_REG(hw, reg_offset, txctrl);
1003 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
1004 struct ixgbe_ring *rx_ring,
1007 struct ixgbe_hw *hw = &adapter->hw;
1008 u32 rxctrl = dca3_get_tag(rx_ring->dev, cpu);
1009 u8 reg_idx = rx_ring->reg_idx;
1012 switch (hw->mac.type) {
1013 case ixgbe_mac_82599EB:
1014 case ixgbe_mac_X540:
1015 rxctrl <<= IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599;
1022 * We can enable relaxed ordering for reads, but not writes when
1023 * DCA is enabled. This is due to a known issue in some chipsets
1024 * which will cause the DCA tag to be cleared.
1026 rxctrl |= IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1027 IXGBE_DCA_RXCTRL_DESC_DCA_EN;
1029 IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
1032 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
1034 struct ixgbe_adapter *adapter = q_vector->adapter;
1035 struct ixgbe_ring *ring;
1036 int cpu = get_cpu();
1038 if (q_vector->cpu == cpu)
1041 ixgbe_for_each_ring(ring, q_vector->tx)
1042 ixgbe_update_tx_dca(adapter, ring, cpu);
1044 ixgbe_for_each_ring(ring, q_vector->rx)
1045 ixgbe_update_rx_dca(adapter, ring, cpu);
1047 q_vector->cpu = cpu;
1052 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
1056 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
1059 /* always use CB2 mode, difference is masked in the CB driver */
1060 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
1062 for (i = 0; i < adapter->num_q_vectors; i++) {
1063 adapter->q_vector[i]->cpu = -1;
1064 ixgbe_update_dca(adapter->q_vector[i]);
1068 static int __ixgbe_notify_dca(struct device *dev, void *data)
1070 struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1071 unsigned long event = *(unsigned long *)data;
1073 if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
1077 case DCA_PROVIDER_ADD:
1078 /* if we're already enabled, don't do it again */
1079 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1081 if (dca_add_requester(dev) == 0) {
1082 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1083 ixgbe_setup_dca(adapter);
1086 /* Fall Through since DCA is disabled. */
1087 case DCA_PROVIDER_REMOVE:
1088 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1089 dca_remove_requester(dev);
1090 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1091 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
1099 #endif /* CONFIG_IXGBE_DCA */
1100 static inline void ixgbe_rx_hash(struct ixgbe_ring *ring,
1101 union ixgbe_adv_rx_desc *rx_desc,
1102 struct sk_buff *skb)
1104 if (ring->netdev->features & NETIF_F_RXHASH)
1105 skb->rxhash = le32_to_cpu(rx_desc->wb.lower.hi_dword.rss);
1110 * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1111 * @ring: structure containing ring specific data
1112 * @rx_desc: advanced rx descriptor
1114 * Returns : true if it is FCoE pkt
1116 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_ring *ring,
1117 union ixgbe_adv_rx_desc *rx_desc)
1119 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1121 return test_bit(__IXGBE_RX_FCOE, &ring->state) &&
1122 ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1123 (cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1124 IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1127 #endif /* IXGBE_FCOE */
1129 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1130 * @ring: structure containing ring specific data
1131 * @rx_desc: current Rx descriptor being processed
1132 * @skb: skb currently being received and modified
1134 static inline void ixgbe_rx_checksum(struct ixgbe_ring *ring,
1135 union ixgbe_adv_rx_desc *rx_desc,
1136 struct sk_buff *skb)
1138 skb_checksum_none_assert(skb);
1140 /* Rx csum disabled */
1141 if (!(ring->netdev->features & NETIF_F_RXCSUM))
1144 /* if IP and error */
1145 if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
1146 ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
1147 ring->rx_stats.csum_err++;
1151 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
1154 if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
1155 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1158 * 82599 errata, UDP frames with a 0 checksum can be marked as
1161 if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_UDP)) &&
1162 test_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR, &ring->state))
1165 ring->rx_stats.csum_err++;
1169 /* It must be a TCP or UDP packet with a valid checksum */
1170 skb->ip_summed = CHECKSUM_UNNECESSARY;
1173 static inline void ixgbe_release_rx_desc(struct ixgbe_ring *rx_ring, u32 val)
1175 rx_ring->next_to_use = val;
1177 /* update next to alloc since we have filled the ring */
1178 rx_ring->next_to_alloc = val;
1180 * Force memory writes to complete before letting h/w
1181 * know there are new descriptors to fetch. (Only
1182 * applicable for weak-ordered memory model archs,
1186 writel(val, rx_ring->tail);
1189 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring *rx_ring,
1190 struct ixgbe_rx_buffer *bi)
1192 struct page *page = bi->page;
1193 dma_addr_t dma = bi->dma;
1195 /* since we are recycling buffers we should seldom need to alloc */
1199 /* alloc new page for storage */
1200 if (likely(!page)) {
1201 page = __skb_alloc_pages(GFP_ATOMIC | __GFP_COLD | __GFP_COMP,
1202 bi->skb, ixgbe_rx_pg_order(rx_ring));
1203 if (unlikely(!page)) {
1204 rx_ring->rx_stats.alloc_rx_page_failed++;
1210 /* map page for use */
1211 dma = dma_map_page(rx_ring->dev, page, 0,
1212 ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1215 * if mapping failed free memory back to system since
1216 * there isn't much point in holding memory we can't use
1218 if (dma_mapping_error(rx_ring->dev, dma)) {
1219 __free_pages(page, ixgbe_rx_pg_order(rx_ring));
1222 rx_ring->rx_stats.alloc_rx_page_failed++;
1227 bi->page_offset = 0;
1233 * ixgbe_alloc_rx_buffers - Replace used receive buffers
1234 * @rx_ring: ring to place buffers on
1235 * @cleaned_count: number of buffers to replace
1237 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1239 union ixgbe_adv_rx_desc *rx_desc;
1240 struct ixgbe_rx_buffer *bi;
1241 u16 i = rx_ring->next_to_use;
1247 rx_desc = IXGBE_RX_DESC(rx_ring, i);
1248 bi = &rx_ring->rx_buffer_info[i];
1249 i -= rx_ring->count;
1252 if (!ixgbe_alloc_mapped_page(rx_ring, bi))
1256 * Refresh the desc even if buffer_addrs didn't change
1257 * because each write-back erases this info.
1259 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1265 rx_desc = IXGBE_RX_DESC(rx_ring, 0);
1266 bi = rx_ring->rx_buffer_info;
1267 i -= rx_ring->count;
1270 /* clear the hdr_addr for the next_to_use descriptor */
1271 rx_desc->read.hdr_addr = 0;
1274 } while (cleaned_count);
1276 i += rx_ring->count;
1278 if (rx_ring->next_to_use != i)
1279 ixgbe_release_rx_desc(rx_ring, i);
1283 * ixgbe_get_headlen - determine size of header for RSC/LRO/GRO/FCOE
1284 * @data: pointer to the start of the headers
1285 * @max_len: total length of section to find headers in
1287 * This function is meant to determine the length of headers that will
1288 * be recognized by hardware for LRO, GRO, and RSC offloads. The main
1289 * motivation of doing this is to only perform one pull for IPv4 TCP
1290 * packets so that we can do basic things like calculating the gso_size
1291 * based on the average data per packet.
1293 static unsigned int ixgbe_get_headlen(unsigned char *data,
1294 unsigned int max_len)
1297 unsigned char *network;
1300 struct vlan_hdr *vlan;
1303 struct ipv6hdr *ipv6;
1306 u8 nexthdr = 0; /* default to not TCP */
1309 /* this should never happen, but better safe than sorry */
1310 if (max_len < ETH_HLEN)
1313 /* initialize network frame pointer */
1316 /* set first protocol and move network header forward */
1317 protocol = hdr.eth->h_proto;
1318 hdr.network += ETH_HLEN;
1320 /* handle any vlan tag if present */
1321 if (protocol == __constant_htons(ETH_P_8021Q)) {
1322 if ((hdr.network - data) > (max_len - VLAN_HLEN))
1325 protocol = hdr.vlan->h_vlan_encapsulated_proto;
1326 hdr.network += VLAN_HLEN;
1329 /* handle L3 protocols */
1330 if (protocol == __constant_htons(ETH_P_IP)) {
1331 if ((hdr.network - data) > (max_len - sizeof(struct iphdr)))
1334 /* access ihl as a u8 to avoid unaligned access on ia64 */
1335 hlen = (hdr.network[0] & 0x0F) << 2;
1337 /* verify hlen meets minimum size requirements */
1338 if (hlen < sizeof(struct iphdr))
1339 return hdr.network - data;
1341 /* record next protocol if header is present */
1342 if (!hdr.ipv4->frag_off)
1343 nexthdr = hdr.ipv4->protocol;
1344 } else if (protocol == __constant_htons(ETH_P_IPV6)) {
1345 if ((hdr.network - data) > (max_len - sizeof(struct ipv6hdr)))
1348 /* record next protocol */
1349 nexthdr = hdr.ipv6->nexthdr;
1350 hlen = sizeof(struct ipv6hdr);
1352 } else if (protocol == __constant_htons(ETH_P_FCOE)) {
1353 if ((hdr.network - data) > (max_len - FCOE_HEADER_LEN))
1355 hlen = FCOE_HEADER_LEN;
1358 return hdr.network - data;
1361 /* relocate pointer to start of L4 header */
1362 hdr.network += hlen;
1364 /* finally sort out TCP/UDP */
1365 if (nexthdr == IPPROTO_TCP) {
1366 if ((hdr.network - data) > (max_len - sizeof(struct tcphdr)))
1369 /* access doff as a u8 to avoid unaligned access on ia64 */
1370 hlen = (hdr.network[12] & 0xF0) >> 2;
1372 /* verify hlen meets minimum size requirements */
1373 if (hlen < sizeof(struct tcphdr))
1374 return hdr.network - data;
1376 hdr.network += hlen;
1377 } else if (nexthdr == IPPROTO_UDP) {
1378 if ((hdr.network - data) > (max_len - sizeof(struct udphdr)))
1381 hdr.network += sizeof(struct udphdr);
1385 * If everything has gone correctly hdr.network should be the
1386 * data section of the packet and will be the end of the header.
1387 * If not then it probably represents the end of the last recognized
1390 if ((hdr.network - data) < max_len)
1391 return hdr.network - data;
1396 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring *ring,
1397 struct sk_buff *skb)
1399 u16 hdr_len = skb_headlen(skb);
1401 /* set gso_size to avoid messing up TCP MSS */
1402 skb_shinfo(skb)->gso_size = DIV_ROUND_UP((skb->len - hdr_len),
1403 IXGBE_CB(skb)->append_cnt);
1406 static void ixgbe_update_rsc_stats(struct ixgbe_ring *rx_ring,
1407 struct sk_buff *skb)
1409 /* if append_cnt is 0 then frame is not RSC */
1410 if (!IXGBE_CB(skb)->append_cnt)
1413 rx_ring->rx_stats.rsc_count += IXGBE_CB(skb)->append_cnt;
1414 rx_ring->rx_stats.rsc_flush++;
1416 ixgbe_set_rsc_gso_size(rx_ring, skb);
1418 /* gso_size is computed using append_cnt so always clear it last */
1419 IXGBE_CB(skb)->append_cnt = 0;
1423 * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor
1424 * @rx_ring: rx descriptor ring packet is being transacted on
1425 * @rx_desc: pointer to the EOP Rx descriptor
1426 * @skb: pointer to current skb being populated
1428 * This function checks the ring, descriptor, and packet information in
1429 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1430 * other fields within the skb.
1432 static void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring,
1433 union ixgbe_adv_rx_desc *rx_desc,
1434 struct sk_buff *skb)
1436 struct net_device *dev = rx_ring->netdev;
1438 ixgbe_update_rsc_stats(rx_ring, skb);
1440 ixgbe_rx_hash(rx_ring, rx_desc, skb);
1442 ixgbe_rx_checksum(rx_ring, rx_desc, skb);
1444 ixgbe_ptp_rx_hwtstamp(rx_ring->q_vector, rx_desc, skb);
1446 if ((dev->features & NETIF_F_HW_VLAN_RX) &&
1447 ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
1448 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
1449 __vlan_hwaccel_put_tag(skb, vid);
1452 skb_record_rx_queue(skb, rx_ring->queue_index);
1454 skb->protocol = eth_type_trans(skb, dev);
1457 static void ixgbe_rx_skb(struct ixgbe_q_vector *q_vector,
1458 struct sk_buff *skb)
1460 struct ixgbe_adapter *adapter = q_vector->adapter;
1462 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
1463 napi_gro_receive(&q_vector->napi, skb);
1469 * ixgbe_is_non_eop - process handling of non-EOP buffers
1470 * @rx_ring: Rx ring being processed
1471 * @rx_desc: Rx descriptor for current buffer
1472 * @skb: Current socket buffer containing buffer in progress
1474 * This function updates next to clean. If the buffer is an EOP buffer
1475 * this function exits returning false, otherwise it will place the
1476 * sk_buff in the next buffer to be chained and return true indicating
1477 * that this is in fact a non-EOP buffer.
1479 static bool ixgbe_is_non_eop(struct ixgbe_ring *rx_ring,
1480 union ixgbe_adv_rx_desc *rx_desc,
1481 struct sk_buff *skb)
1483 u32 ntc = rx_ring->next_to_clean + 1;
1485 /* fetch, update, and store next to clean */
1486 ntc = (ntc < rx_ring->count) ? ntc : 0;
1487 rx_ring->next_to_clean = ntc;
1489 prefetch(IXGBE_RX_DESC(rx_ring, ntc));
1491 /* update RSC append count if present */
1492 if (ring_is_rsc_enabled(rx_ring)) {
1493 __le32 rsc_enabled = rx_desc->wb.lower.lo_dword.data &
1494 cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK);
1496 if (unlikely(rsc_enabled)) {
1497 u32 rsc_cnt = le32_to_cpu(rsc_enabled);
1499 rsc_cnt >>= IXGBE_RXDADV_RSCCNT_SHIFT;
1500 IXGBE_CB(skb)->append_cnt += rsc_cnt - 1;
1502 /* update ntc based on RSC value */
1503 ntc = le32_to_cpu(rx_desc->wb.upper.status_error);
1504 ntc &= IXGBE_RXDADV_NEXTP_MASK;
1505 ntc >>= IXGBE_RXDADV_NEXTP_SHIFT;
1509 /* if we are the last buffer then there is nothing else to do */
1510 if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1513 /* place skb in next buffer to be received */
1514 rx_ring->rx_buffer_info[ntc].skb = skb;
1515 rx_ring->rx_stats.non_eop_descs++;
1521 * ixgbe_pull_tail - ixgbe specific version of skb_pull_tail
1522 * @rx_ring: rx descriptor ring packet is being transacted on
1523 * @skb: pointer to current skb being adjusted
1525 * This function is an ixgbe specific version of __pskb_pull_tail. The
1526 * main difference between this version and the original function is that
1527 * this function can make several assumptions about the state of things
1528 * that allow for significant optimizations versus the standard function.
1529 * As a result we can do things like drop a frag and maintain an accurate
1530 * truesize for the skb.
1532 static void ixgbe_pull_tail(struct ixgbe_ring *rx_ring,
1533 struct sk_buff *skb)
1535 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1537 unsigned int pull_len;
1540 * it is valid to use page_address instead of kmap since we are
1541 * working with pages allocated out of the lomem pool per
1542 * alloc_page(GFP_ATOMIC)
1544 va = skb_frag_address(frag);
1547 * we need the header to contain the greater of either ETH_HLEN or
1548 * 60 bytes if the skb->len is less than 60 for skb_pad.
1550 pull_len = ixgbe_get_headlen(va, IXGBE_RX_HDR_SIZE);
1552 /* align pull length to size of long to optimize memcpy performance */
1553 skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
1555 /* update all of the pointers */
1556 skb_frag_size_sub(frag, pull_len);
1557 frag->page_offset += pull_len;
1558 skb->data_len -= pull_len;
1559 skb->tail += pull_len;
1563 * ixgbe_dma_sync_frag - perform DMA sync for first frag of SKB
1564 * @rx_ring: rx descriptor ring packet is being transacted on
1565 * @skb: pointer to current skb being updated
1567 * This function provides a basic DMA sync up for the first fragment of an
1568 * skb. The reason for doing this is that the first fragment cannot be
1569 * unmapped until we have reached the end of packet descriptor for a buffer
1572 static void ixgbe_dma_sync_frag(struct ixgbe_ring *rx_ring,
1573 struct sk_buff *skb)
1575 /* if the page was released unmap it, else just sync our portion */
1576 if (unlikely(IXGBE_CB(skb)->page_released)) {
1577 dma_unmap_page(rx_ring->dev, IXGBE_CB(skb)->dma,
1578 ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1579 IXGBE_CB(skb)->page_released = false;
1581 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1583 dma_sync_single_range_for_cpu(rx_ring->dev,
1586 ixgbe_rx_bufsz(rx_ring),
1589 IXGBE_CB(skb)->dma = 0;
1593 * ixgbe_cleanup_headers - Correct corrupted or empty headers
1594 * @rx_ring: rx descriptor ring packet is being transacted on
1595 * @rx_desc: pointer to the EOP Rx descriptor
1596 * @skb: pointer to current skb being fixed
1598 * Check for corrupted packet headers caused by senders on the local L2
1599 * embedded NIC switch not setting up their Tx Descriptors right. These
1600 * should be very rare.
1602 * Also address the case where we are pulling data in on pages only
1603 * and as such no data is present in the skb header.
1605 * In addition if skb is not at least 60 bytes we need to pad it so that
1606 * it is large enough to qualify as a valid Ethernet frame.
1608 * Returns true if an error was encountered and skb was freed.
1610 static bool ixgbe_cleanup_headers(struct ixgbe_ring *rx_ring,
1611 union ixgbe_adv_rx_desc *rx_desc,
1612 struct sk_buff *skb)
1614 struct net_device *netdev = rx_ring->netdev;
1616 /* verify that the packet does not have any known errors */
1617 if (unlikely(ixgbe_test_staterr(rx_desc,
1618 IXGBE_RXDADV_ERR_FRAME_ERR_MASK) &&
1619 !(netdev->features & NETIF_F_RXALL))) {
1620 dev_kfree_skb_any(skb);
1624 /* place header in linear portion of buffer */
1625 if (skb_is_nonlinear(skb))
1626 ixgbe_pull_tail(rx_ring, skb);
1629 /* do not attempt to pad FCoE Frames as this will disrupt DDP */
1630 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc))
1634 /* if skb_pad returns an error the skb was freed */
1635 if (unlikely(skb->len < 60)) {
1636 int pad_len = 60 - skb->len;
1638 if (skb_pad(skb, pad_len))
1640 __skb_put(skb, pad_len);
1647 * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1648 * @rx_ring: rx descriptor ring to store buffers on
1649 * @old_buff: donor buffer to have page reused
1651 * Synchronizes page for reuse by the adapter
1653 static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring,
1654 struct ixgbe_rx_buffer *old_buff)
1656 struct ixgbe_rx_buffer *new_buff;
1657 u16 nta = rx_ring->next_to_alloc;
1659 new_buff = &rx_ring->rx_buffer_info[nta];
1661 /* update, and store next to alloc */
1663 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1665 /* transfer page from old buffer to new buffer */
1666 new_buff->page = old_buff->page;
1667 new_buff->dma = old_buff->dma;
1668 new_buff->page_offset = old_buff->page_offset;
1670 /* sync the buffer for use by the device */
1671 dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
1672 new_buff->page_offset,
1673 ixgbe_rx_bufsz(rx_ring),
1678 * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1679 * @rx_ring: rx descriptor ring to transact packets on
1680 * @rx_buffer: buffer containing page to add
1681 * @rx_desc: descriptor containing length of buffer written by hardware
1682 * @skb: sk_buff to place the data into
1684 * This function will add the data contained in rx_buffer->page to the skb.
1685 * This is done either through a direct copy if the data in the buffer is
1686 * less than the skb header size, otherwise it will just attach the page as
1687 * a frag to the skb.
1689 * The function will then update the page offset if necessary and return
1690 * true if the buffer can be reused by the adapter.
1692 static bool ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
1693 struct ixgbe_rx_buffer *rx_buffer,
1694 union ixgbe_adv_rx_desc *rx_desc,
1695 struct sk_buff *skb)
1697 struct page *page = rx_buffer->page;
1698 unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
1699 #if (PAGE_SIZE < 8192)
1700 unsigned int truesize = ixgbe_rx_bufsz(rx_ring);
1702 unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
1703 unsigned int last_offset = ixgbe_rx_pg_size(rx_ring) -
1704 ixgbe_rx_bufsz(rx_ring);
1707 if ((size <= IXGBE_RX_HDR_SIZE) && !skb_is_nonlinear(skb)) {
1708 unsigned char *va = page_address(page) + rx_buffer->page_offset;
1710 memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
1712 /* we can reuse buffer as-is, just make sure it is local */
1713 if (likely(page_to_nid(page) == numa_node_id()))
1716 /* this page cannot be reused so discard it */
1721 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
1722 rx_buffer->page_offset, size, truesize);
1724 /* avoid re-using remote pages */
1725 if (unlikely(page_to_nid(page) != numa_node_id()))
1728 #if (PAGE_SIZE < 8192)
1729 /* if we are only owner of page we can reuse it */
1730 if (unlikely(page_count(page) != 1))
1733 /* flip page offset to other buffer */
1734 rx_buffer->page_offset ^= truesize;
1737 * since we are the only owner of the page and we need to
1738 * increment it, just set the value to 2 in order to avoid
1739 * an unecessary locked operation
1741 atomic_set(&page->_count, 2);
1743 /* move offset up to the next cache line */
1744 rx_buffer->page_offset += truesize;
1746 if (rx_buffer->page_offset > last_offset)
1749 /* bump ref count on page before it is given to the stack */
1756 static struct sk_buff *ixgbe_fetch_rx_buffer(struct ixgbe_ring *rx_ring,
1757 union ixgbe_adv_rx_desc *rx_desc)
1759 struct ixgbe_rx_buffer *rx_buffer;
1760 struct sk_buff *skb;
1763 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
1764 page = rx_buffer->page;
1767 skb = rx_buffer->skb;
1770 void *page_addr = page_address(page) +
1771 rx_buffer->page_offset;
1773 /* prefetch first cache line of first page */
1774 prefetch(page_addr);
1775 #if L1_CACHE_BYTES < 128
1776 prefetch(page_addr + L1_CACHE_BYTES);
1779 /* allocate a skb to store the frags */
1780 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1782 if (unlikely(!skb)) {
1783 rx_ring->rx_stats.alloc_rx_buff_failed++;
1788 * we will be copying header into skb->data in
1789 * pskb_may_pull so it is in our interest to prefetch
1790 * it now to avoid a possible cache miss
1792 prefetchw(skb->data);
1795 * Delay unmapping of the first packet. It carries the
1796 * header information, HW may still access the header
1797 * after the writeback. Only unmap it when EOP is
1800 if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1803 IXGBE_CB(skb)->dma = rx_buffer->dma;
1805 if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
1806 ixgbe_dma_sync_frag(rx_ring, skb);
1809 /* we are reusing so sync this buffer for CPU use */
1810 dma_sync_single_range_for_cpu(rx_ring->dev,
1812 rx_buffer->page_offset,
1813 ixgbe_rx_bufsz(rx_ring),
1817 /* pull page into skb */
1818 if (ixgbe_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
1819 /* hand second half of page back to the ring */
1820 ixgbe_reuse_rx_page(rx_ring, rx_buffer);
1821 } else if (IXGBE_CB(skb)->dma == rx_buffer->dma) {
1822 /* the page has been released from the ring */
1823 IXGBE_CB(skb)->page_released = true;
1825 /* we are not reusing the buffer so unmap it */
1826 dma_unmap_page(rx_ring->dev, rx_buffer->dma,
1827 ixgbe_rx_pg_size(rx_ring),
1831 /* clear contents of buffer_info */
1832 rx_buffer->skb = NULL;
1834 rx_buffer->page = NULL;
1840 * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
1841 * @q_vector: structure containing interrupt and ring information
1842 * @rx_ring: rx descriptor ring to transact packets on
1843 * @budget: Total limit on number of packets to process
1845 * This function provides a "bounce buffer" approach to Rx interrupt
1846 * processing. The advantage to this is that on systems that have
1847 * expensive overhead for IOMMU access this provides a means of avoiding
1848 * it by maintaining the mapping of the page to the syste.
1850 * Returns true if all work is completed without reaching budget
1852 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1853 struct ixgbe_ring *rx_ring,
1856 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1858 struct ixgbe_adapter *adapter = q_vector->adapter;
1860 unsigned int mss = 0;
1861 #endif /* IXGBE_FCOE */
1862 u16 cleaned_count = ixgbe_desc_unused(rx_ring);
1865 union ixgbe_adv_rx_desc *rx_desc;
1866 struct sk_buff *skb;
1868 /* return some buffers to hardware, one at a time is too slow */
1869 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
1870 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1874 rx_desc = IXGBE_RX_DESC(rx_ring, rx_ring->next_to_clean);
1876 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_DD))
1880 * This memory barrier is needed to keep us from reading
1881 * any other fields out of the rx_desc until we know the
1882 * RXD_STAT_DD bit is set
1886 /* retrieve a buffer from the ring */
1887 skb = ixgbe_fetch_rx_buffer(rx_ring, rx_desc);
1889 /* exit if we failed to retrieve a buffer */
1895 /* place incomplete frames back on ring for completion */
1896 if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
1899 /* verify the packet layout is correct */
1900 if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
1903 /* probably a little skewed due to removing CRC */
1904 total_rx_bytes += skb->len;
1906 /* populate checksum, timestamp, VLAN, and protocol */
1907 ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
1910 /* if ddp, not passing to ULD unless for FCP_RSP or error */
1911 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc)) {
1912 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
1913 /* include DDPed FCoE data */
1914 if (ddp_bytes > 0) {
1916 mss = rx_ring->netdev->mtu -
1917 sizeof(struct fcoe_hdr) -
1918 sizeof(struct fc_frame_header) -
1919 sizeof(struct fcoe_crc_eof);
1923 total_rx_bytes += ddp_bytes;
1924 total_rx_packets += DIV_ROUND_UP(ddp_bytes,
1928 dev_kfree_skb_any(skb);
1933 #endif /* IXGBE_FCOE */
1934 ixgbe_rx_skb(q_vector, skb);
1936 /* update budget accounting */
1938 } while (likely(total_rx_packets < budget));
1940 u64_stats_update_begin(&rx_ring->syncp);
1941 rx_ring->stats.packets += total_rx_packets;
1942 rx_ring->stats.bytes += total_rx_bytes;
1943 u64_stats_update_end(&rx_ring->syncp);
1944 q_vector->rx.total_packets += total_rx_packets;
1945 q_vector->rx.total_bytes += total_rx_bytes;
1948 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1950 return (total_rx_packets < budget);
1954 * ixgbe_configure_msix - Configure MSI-X hardware
1955 * @adapter: board private structure
1957 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1960 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1962 struct ixgbe_q_vector *q_vector;
1966 /* Populate MSIX to EITR Select */
1967 if (adapter->num_vfs > 32) {
1968 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
1969 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
1973 * Populate the IVAR table and set the ITR values to the
1974 * corresponding register.
1976 for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
1977 struct ixgbe_ring *ring;
1978 q_vector = adapter->q_vector[v_idx];
1980 ixgbe_for_each_ring(ring, q_vector->rx)
1981 ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1983 ixgbe_for_each_ring(ring, q_vector->tx)
1984 ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1986 ixgbe_write_eitr(q_vector);
1989 switch (adapter->hw.mac.type) {
1990 case ixgbe_mac_82598EB:
1991 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1994 case ixgbe_mac_82599EB:
1995 case ixgbe_mac_X540:
1996 ixgbe_set_ivar(adapter, -1, 1, v_idx);
2001 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
2003 /* set up to autoclear timer, and the vectors */
2004 mask = IXGBE_EIMS_ENABLE_MASK;
2005 mask &= ~(IXGBE_EIMS_OTHER |
2006 IXGBE_EIMS_MAILBOX |
2009 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
2012 enum latency_range {
2016 latency_invalid = 255
2020 * ixgbe_update_itr - update the dynamic ITR value based on statistics
2021 * @q_vector: structure containing interrupt and ring information
2022 * @ring_container: structure containing ring performance data
2024 * Stores a new ITR value based on packets and byte
2025 * counts during the last interrupt. The advantage of per interrupt
2026 * computation is faster updates and more accurate ITR for the current
2027 * traffic pattern. Constants in this function were computed
2028 * based on theoretical maximum wire speed and thresholds were set based
2029 * on testing data as well as attempting to minimize response time
2030 * while increasing bulk throughput.
2031 * this functionality is controlled by the InterruptThrottleRate module
2032 * parameter (see ixgbe_param.c)
2034 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
2035 struct ixgbe_ring_container *ring_container)
2037 int bytes = ring_container->total_bytes;
2038 int packets = ring_container->total_packets;
2041 u8 itr_setting = ring_container->itr;
2046 /* simple throttlerate management
2047 * 0-10MB/s lowest (100000 ints/s)
2048 * 10-20MB/s low (20000 ints/s)
2049 * 20-1249MB/s bulk (8000 ints/s)
2051 /* what was last interrupt timeslice? */
2052 timepassed_us = q_vector->itr >> 2;
2053 bytes_perint = bytes / timepassed_us; /* bytes/usec */
2055 switch (itr_setting) {
2056 case lowest_latency:
2057 if (bytes_perint > 10)
2058 itr_setting = low_latency;
2061 if (bytes_perint > 20)
2062 itr_setting = bulk_latency;
2063 else if (bytes_perint <= 10)
2064 itr_setting = lowest_latency;
2067 if (bytes_perint <= 20)
2068 itr_setting = low_latency;
2072 /* clear work counters since we have the values we need */
2073 ring_container->total_bytes = 0;
2074 ring_container->total_packets = 0;
2076 /* write updated itr to ring container */
2077 ring_container->itr = itr_setting;
2081 * ixgbe_write_eitr - write EITR register in hardware specific way
2082 * @q_vector: structure containing interrupt and ring information
2084 * This function is made to be called by ethtool and by the driver
2085 * when it needs to update EITR registers at runtime. Hardware
2086 * specific quirks/differences are taken care of here.
2088 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
2090 struct ixgbe_adapter *adapter = q_vector->adapter;
2091 struct ixgbe_hw *hw = &adapter->hw;
2092 int v_idx = q_vector->v_idx;
2093 u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
2095 switch (adapter->hw.mac.type) {
2096 case ixgbe_mac_82598EB:
2097 /* must write high and low 16 bits to reset counter */
2098 itr_reg |= (itr_reg << 16);
2100 case ixgbe_mac_82599EB:
2101 case ixgbe_mac_X540:
2103 * set the WDIS bit to not clear the timer bits and cause an
2104 * immediate assertion of the interrupt
2106 itr_reg |= IXGBE_EITR_CNT_WDIS;
2111 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
2114 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
2116 u32 new_itr = q_vector->itr;
2119 ixgbe_update_itr(q_vector, &q_vector->tx);
2120 ixgbe_update_itr(q_vector, &q_vector->rx);
2122 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
2124 switch (current_itr) {
2125 /* counts and packets in update_itr are dependent on these numbers */
2126 case lowest_latency:
2127 new_itr = IXGBE_100K_ITR;
2130 new_itr = IXGBE_20K_ITR;
2133 new_itr = IXGBE_8K_ITR;
2139 if (new_itr != q_vector->itr) {
2140 /* do an exponential smoothing */
2141 new_itr = (10 * new_itr * q_vector->itr) /
2142 ((9 * new_itr) + q_vector->itr);
2144 /* save the algorithm value here */
2145 q_vector->itr = new_itr;
2147 ixgbe_write_eitr(q_vector);
2152 * ixgbe_check_overtemp_subtask - check for over temperature
2153 * @adapter: pointer to adapter
2155 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2157 struct ixgbe_hw *hw = &adapter->hw;
2158 u32 eicr = adapter->interrupt_event;
2160 if (test_bit(__IXGBE_DOWN, &adapter->state))
2163 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2164 !(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2167 adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2169 switch (hw->device_id) {
2170 case IXGBE_DEV_ID_82599_T3_LOM:
2172 * Since the warning interrupt is for both ports
2173 * we don't have to check if:
2174 * - This interrupt wasn't for our port.
2175 * - We may have missed the interrupt so always have to
2176 * check if we got a LSC
2178 if (!(eicr & IXGBE_EICR_GPI_SDP0) &&
2179 !(eicr & IXGBE_EICR_LSC))
2182 if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2184 bool link_up = false;
2186 hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
2192 /* Check if this is not due to overtemp */
2193 if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
2198 if (!(eicr & IXGBE_EICR_GPI_SDP0))
2203 "Network adapter has been stopped because it has over heated. "
2204 "Restart the computer. If the problem persists, "
2205 "power off the system and replace the adapter\n");
2207 adapter->interrupt_event = 0;
2210 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2212 struct ixgbe_hw *hw = &adapter->hw;
2214 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2215 (eicr & IXGBE_EICR_GPI_SDP1)) {
2216 e_crit(probe, "Fan has stopped, replace the adapter\n");
2217 /* write to clear the interrupt */
2218 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
2222 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2224 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2227 switch (adapter->hw.mac.type) {
2228 case ixgbe_mac_82599EB:
2230 * Need to check link state so complete overtemp check
2233 if (((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)) &&
2234 (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2235 adapter->interrupt_event = eicr;
2236 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2237 ixgbe_service_event_schedule(adapter);
2241 case ixgbe_mac_X540:
2242 if (!(eicr & IXGBE_EICR_TS))
2250 "Network adapter has been stopped because it has over heated. "
2251 "Restart the computer. If the problem persists, "
2252 "power off the system and replace the adapter\n");
2255 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2257 struct ixgbe_hw *hw = &adapter->hw;
2259 if (eicr & IXGBE_EICR_GPI_SDP2) {
2260 /* Clear the interrupt */
2261 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
2262 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2263 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2264 ixgbe_service_event_schedule(adapter);
2268 if (eicr & IXGBE_EICR_GPI_SDP1) {
2269 /* Clear the interrupt */
2270 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
2271 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2272 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2273 ixgbe_service_event_schedule(adapter);
2278 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2280 struct ixgbe_hw *hw = &adapter->hw;
2283 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2284 adapter->link_check_timeout = jiffies;
2285 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2286 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2287 IXGBE_WRITE_FLUSH(hw);
2288 ixgbe_service_event_schedule(adapter);
2292 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2296 struct ixgbe_hw *hw = &adapter->hw;
2298 switch (hw->mac.type) {
2299 case ixgbe_mac_82598EB:
2300 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2301 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2303 case ixgbe_mac_82599EB:
2304 case ixgbe_mac_X540:
2305 mask = (qmask & 0xFFFFFFFF);
2307 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2308 mask = (qmask >> 32);
2310 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2315 /* skip the flush */
2318 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
2322 struct ixgbe_hw *hw = &adapter->hw;
2324 switch (hw->mac.type) {
2325 case ixgbe_mac_82598EB:
2326 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2327 IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
2329 case ixgbe_mac_82599EB:
2330 case ixgbe_mac_X540:
2331 mask = (qmask & 0xFFFFFFFF);
2333 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
2334 mask = (qmask >> 32);
2336 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
2341 /* skip the flush */
2345 * ixgbe_irq_enable - Enable default interrupt generation settings
2346 * @adapter: board private structure
2348 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2351 u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2353 /* don't reenable LSC while waiting for link */
2354 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
2355 mask &= ~IXGBE_EIMS_LSC;
2357 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2358 switch (adapter->hw.mac.type) {
2359 case ixgbe_mac_82599EB:
2360 mask |= IXGBE_EIMS_GPI_SDP0;
2362 case ixgbe_mac_X540:
2363 mask |= IXGBE_EIMS_TS;
2368 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2369 mask |= IXGBE_EIMS_GPI_SDP1;
2370 switch (adapter->hw.mac.type) {
2371 case ixgbe_mac_82599EB:
2372 mask |= IXGBE_EIMS_GPI_SDP1;
2373 mask |= IXGBE_EIMS_GPI_SDP2;
2374 case ixgbe_mac_X540:
2375 mask |= IXGBE_EIMS_ECC;
2376 mask |= IXGBE_EIMS_MAILBOX;
2382 if (adapter->hw.mac.type == ixgbe_mac_X540)
2383 mask |= IXGBE_EIMS_TIMESYNC;
2385 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2386 !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
2387 mask |= IXGBE_EIMS_FLOW_DIR;
2389 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2391 ixgbe_irq_enable_queues(adapter, ~0);
2393 IXGBE_WRITE_FLUSH(&adapter->hw);
2396 static irqreturn_t ixgbe_msix_other(int irq, void *data)
2398 struct ixgbe_adapter *adapter = data;
2399 struct ixgbe_hw *hw = &adapter->hw;
2403 * Workaround for Silicon errata. Use clear-by-write instead
2404 * of clear-by-read. Reading with EICS will return the
2405 * interrupt causes without clearing, which later be done
2406 * with the write to EICR.
2408 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
2409 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
2411 if (eicr & IXGBE_EICR_LSC)
2412 ixgbe_check_lsc(adapter);
2414 if (eicr & IXGBE_EICR_MAILBOX)
2415 ixgbe_msg_task(adapter);
2417 switch (hw->mac.type) {
2418 case ixgbe_mac_82599EB:
2419 case ixgbe_mac_X540:
2420 if (eicr & IXGBE_EICR_ECC)
2421 e_info(link, "Received unrecoverable ECC Err, please "
2423 /* Handle Flow Director Full threshold interrupt */
2424 if (eicr & IXGBE_EICR_FLOW_DIR) {
2425 int reinit_count = 0;
2427 for (i = 0; i < adapter->num_tx_queues; i++) {
2428 struct ixgbe_ring *ring = adapter->tx_ring[i];
2429 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
2434 /* no more flow director interrupts until after init */
2435 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
2436 adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
2437 ixgbe_service_event_schedule(adapter);
2440 ixgbe_check_sfp_event(adapter, eicr);
2441 ixgbe_check_overtemp_event(adapter, eicr);
2447 ixgbe_check_fan_failure(adapter, eicr);
2449 if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
2450 ixgbe_ptp_check_pps_event(adapter, eicr);
2452 /* re-enable the original interrupt state, no lsc, no queues */
2453 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2454 ixgbe_irq_enable(adapter, false, false);
2459 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
2461 struct ixgbe_q_vector *q_vector = data;
2463 /* EIAM disabled interrupts (on this vector) for us */
2465 if (q_vector->rx.ring || q_vector->tx.ring)
2466 napi_schedule(&q_vector->napi);
2472 * ixgbe_poll - NAPI Rx polling callback
2473 * @napi: structure for representing this polling device
2474 * @budget: how many packets driver is allowed to clean
2476 * This function is used for legacy and MSI, NAPI mode
2478 int ixgbe_poll(struct napi_struct *napi, int budget)
2480 struct ixgbe_q_vector *q_vector =
2481 container_of(napi, struct ixgbe_q_vector, napi);
2482 struct ixgbe_adapter *adapter = q_vector->adapter;
2483 struct ixgbe_ring *ring;
2484 int per_ring_budget;
2485 bool clean_complete = true;
2487 #ifdef CONFIG_IXGBE_DCA
2488 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2489 ixgbe_update_dca(q_vector);
2492 ixgbe_for_each_ring(ring, q_vector->tx)
2493 clean_complete &= !!ixgbe_clean_tx_irq(q_vector, ring);
2495 /* attempt to distribute budget to each queue fairly, but don't allow
2496 * the budget to go below 1 because we'll exit polling */
2497 if (q_vector->rx.count > 1)
2498 per_ring_budget = max(budget/q_vector->rx.count, 1);
2500 per_ring_budget = budget;
2502 ixgbe_for_each_ring(ring, q_vector->rx)
2503 clean_complete &= ixgbe_clean_rx_irq(q_vector, ring,
2506 /* If all work not completed, return budget and keep polling */
2507 if (!clean_complete)
2510 /* all work done, exit the polling mode */
2511 napi_complete(napi);
2512 if (adapter->rx_itr_setting & 1)
2513 ixgbe_set_itr(q_vector);
2514 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2515 ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
2521 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2522 * @adapter: board private structure
2524 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2525 * interrupts from the kernel.
2527 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2529 struct net_device *netdev = adapter->netdev;
2533 for (vector = 0; vector < adapter->num_q_vectors; vector++) {
2534 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2535 struct msix_entry *entry = &adapter->msix_entries[vector];
2537 if (q_vector->tx.ring && q_vector->rx.ring) {
2538 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2539 "%s-%s-%d", netdev->name, "TxRx", ri++);
2541 } else if (q_vector->rx.ring) {
2542 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2543 "%s-%s-%d", netdev->name, "rx", ri++);
2544 } else if (q_vector->tx.ring) {
2545 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2546 "%s-%s-%d", netdev->name, "tx", ti++);
2548 /* skip this unused q_vector */
2551 err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
2552 q_vector->name, q_vector);
2554 e_err(probe, "request_irq failed for MSIX interrupt "
2555 "Error: %d\n", err);
2556 goto free_queue_irqs;
2558 /* If Flow Director is enabled, set interrupt affinity */
2559 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2560 /* assign the mask for this irq */
2561 irq_set_affinity_hint(entry->vector,
2562 &q_vector->affinity_mask);
2566 err = request_irq(adapter->msix_entries[vector].vector,
2567 ixgbe_msix_other, 0, netdev->name, adapter);
2569 e_err(probe, "request_irq for msix_other failed: %d\n", err);
2570 goto free_queue_irqs;
2578 irq_set_affinity_hint(adapter->msix_entries[vector].vector,
2580 free_irq(adapter->msix_entries[vector].vector,
2581 adapter->q_vector[vector]);
2583 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2584 pci_disable_msix(adapter->pdev);
2585 kfree(adapter->msix_entries);
2586 adapter->msix_entries = NULL;
2591 * ixgbe_intr - legacy mode Interrupt Handler
2592 * @irq: interrupt number
2593 * @data: pointer to a network interface device structure
2595 static irqreturn_t ixgbe_intr(int irq, void *data)
2597 struct ixgbe_adapter *adapter = data;
2598 struct ixgbe_hw *hw = &adapter->hw;
2599 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2603 * Workaround for silicon errata #26 on 82598. Mask the interrupt
2604 * before the read of EICR.
2606 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2608 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2609 * therefore no explicit interrupt disable is necessary */
2610 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2613 * shared interrupt alert!
2614 * make sure interrupts are enabled because the read will
2615 * have disabled interrupts due to EIAM
2616 * finish the workaround of silicon errata on 82598. Unmask
2617 * the interrupt that we masked before the EICR read.
2619 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2620 ixgbe_irq_enable(adapter, true, true);
2621 return IRQ_NONE; /* Not our interrupt */
2624 if (eicr & IXGBE_EICR_LSC)
2625 ixgbe_check_lsc(adapter);
2627 switch (hw->mac.type) {
2628 case ixgbe_mac_82599EB:
2629 ixgbe_check_sfp_event(adapter, eicr);
2631 case ixgbe_mac_X540:
2632 if (eicr & IXGBE_EICR_ECC)
2633 e_info(link, "Received unrecoverable ECC err, please "
2635 ixgbe_check_overtemp_event(adapter, eicr);
2641 ixgbe_check_fan_failure(adapter, eicr);
2642 if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
2643 ixgbe_ptp_check_pps_event(adapter, eicr);
2645 /* would disable interrupts here but EIAM disabled it */
2646 napi_schedule(&q_vector->napi);
2649 * re-enable link(maybe) and non-queue interrupts, no flush.
2650 * ixgbe_poll will re-enable the queue interrupts
2652 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2653 ixgbe_irq_enable(adapter, false, false);
2659 * ixgbe_request_irq - initialize interrupts
2660 * @adapter: board private structure
2662 * Attempts to configure interrupts using the best available
2663 * capabilities of the hardware and kernel.
2665 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2667 struct net_device *netdev = adapter->netdev;
2670 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2671 err = ixgbe_request_msix_irqs(adapter);
2672 else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
2673 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2674 netdev->name, adapter);
2676 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2677 netdev->name, adapter);
2680 e_err(probe, "request_irq failed, Error %d\n", err);
2685 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2689 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
2690 free_irq(adapter->pdev->irq, adapter);
2694 for (vector = 0; vector < adapter->num_q_vectors; vector++) {
2695 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2696 struct msix_entry *entry = &adapter->msix_entries[vector];
2698 /* free only the irqs that were actually requested */
2699 if (!q_vector->rx.ring && !q_vector->tx.ring)
2702 /* clear the affinity_mask in the IRQ descriptor */
2703 irq_set_affinity_hint(entry->vector, NULL);
2705 free_irq(entry->vector, q_vector);
2708 free_irq(adapter->msix_entries[vector++].vector, adapter);
2712 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2713 * @adapter: board private structure
2715 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2717 switch (adapter->hw.mac.type) {
2718 case ixgbe_mac_82598EB:
2719 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2721 case ixgbe_mac_82599EB:
2722 case ixgbe_mac_X540:
2723 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2724 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2725 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2730 IXGBE_WRITE_FLUSH(&adapter->hw);
2731 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2734 for (vector = 0; vector < adapter->num_q_vectors; vector++)
2735 synchronize_irq(adapter->msix_entries[vector].vector);
2737 synchronize_irq(adapter->msix_entries[vector++].vector);
2739 synchronize_irq(adapter->pdev->irq);
2744 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2747 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2749 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2751 ixgbe_write_eitr(q_vector);
2753 ixgbe_set_ivar(adapter, 0, 0, 0);
2754 ixgbe_set_ivar(adapter, 1, 0, 0);
2756 e_info(hw, "Legacy interrupt IVAR setup done\n");
2760 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2761 * @adapter: board private structure
2762 * @ring: structure containing ring specific data
2764 * Configure the Tx descriptor ring after a reset.
2766 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2767 struct ixgbe_ring *ring)
2769 struct ixgbe_hw *hw = &adapter->hw;
2770 u64 tdba = ring->dma;
2772 u32 txdctl = IXGBE_TXDCTL_ENABLE;
2773 u8 reg_idx = ring->reg_idx;
2775 /* disable queue to avoid issues while updating state */
2776 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
2777 IXGBE_WRITE_FLUSH(hw);
2779 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2780 (tdba & DMA_BIT_MASK(32)));
2781 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2782 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2783 ring->count * sizeof(union ixgbe_adv_tx_desc));
2784 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2785 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2786 ring->tail = hw->hw_addr + IXGBE_TDT(reg_idx);
2789 * set WTHRESH to encourage burst writeback, it should not be set
2790 * higher than 1 when ITR is 0 as it could cause false TX hangs
2792 * In order to avoid issues WTHRESH + PTHRESH should always be equal
2793 * to or less than the number of on chip descriptors, which is
2796 if (!ring->q_vector || (ring->q_vector->itr < 8))
2797 txdctl |= (1 << 16); /* WTHRESH = 1 */
2799 txdctl |= (8 << 16); /* WTHRESH = 8 */
2802 * Setting PTHRESH to 32 both improves performance
2803 * and avoids a TX hang with DFP enabled
2805 txdctl |= (1 << 8) | /* HTHRESH = 1 */
2806 32; /* PTHRESH = 32 */
2808 /* reinitialize flowdirector state */
2809 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2810 ring->atr_sample_rate = adapter->atr_sample_rate;
2811 ring->atr_count = 0;
2812 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
2814 ring->atr_sample_rate = 0;
2817 clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
2820 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
2822 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2823 if (hw->mac.type == ixgbe_mac_82598EB &&
2824 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2827 /* poll to verify queue is enabled */
2829 usleep_range(1000, 2000);
2830 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2831 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
2833 e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
2836 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
2838 struct ixgbe_hw *hw = &adapter->hw;
2840 u8 tcs = netdev_get_num_tc(adapter->netdev);
2842 if (hw->mac.type == ixgbe_mac_82598EB)
2845 /* disable the arbiter while setting MTQC */
2846 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2847 rttdcs |= IXGBE_RTTDCS_ARBDIS;
2848 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2850 /* set transmit pool layout */
2851 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
2852 mtqc = IXGBE_MTQC_VT_ENA;
2854 mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
2856 mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
2857 else if (adapter->ring_feature[RING_F_RSS].indices == 4)
2858 mtqc |= IXGBE_MTQC_32VF;
2860 mtqc |= IXGBE_MTQC_64VF;
2863 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
2865 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
2867 mtqc = IXGBE_MTQC_64Q_1PB;
2870 IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
2872 /* Enable Security TX Buffer IFG for multiple pb */
2874 u32 sectx = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
2875 sectx |= IXGBE_SECTX_DCB;
2876 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, sectx);
2879 /* re-enable the arbiter */
2880 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2881 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2885 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2886 * @adapter: board private structure
2888 * Configure the Tx unit of the MAC after a reset.
2890 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2892 struct ixgbe_hw *hw = &adapter->hw;
2896 ixgbe_setup_mtqc(adapter);
2898 if (hw->mac.type != ixgbe_mac_82598EB) {
2899 /* DMATXCTL.EN must be before Tx queues are enabled */
2900 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2901 dmatxctl |= IXGBE_DMATXCTL_TE;
2902 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2905 /* Setup the HW Tx Head and Tail descriptor pointers */
2906 for (i = 0; i < adapter->num_tx_queues; i++)
2907 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
2910 static void ixgbe_enable_rx_drop(struct ixgbe_adapter *adapter,
2911 struct ixgbe_ring *ring)
2913 struct ixgbe_hw *hw = &adapter->hw;
2914 u8 reg_idx = ring->reg_idx;
2915 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
2917 srrctl |= IXGBE_SRRCTL_DROP_EN;
2919 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
2922 static void ixgbe_disable_rx_drop(struct ixgbe_adapter *adapter,
2923 struct ixgbe_ring *ring)
2925 struct ixgbe_hw *hw = &adapter->hw;
2926 u8 reg_idx = ring->reg_idx;
2927 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
2929 srrctl &= ~IXGBE_SRRCTL_DROP_EN;
2931 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
2934 #ifdef CONFIG_IXGBE_DCB
2935 void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
2937 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
2941 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
2943 if (adapter->ixgbe_ieee_pfc)
2944 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
2947 * We should set the drop enable bit if:
2950 * Number of Rx queues > 1 and flow control is disabled
2952 * This allows us to avoid head of line blocking for security
2953 * and performance reasons.
2955 if (adapter->num_vfs || (adapter->num_rx_queues > 1 &&
2956 !(adapter->hw.fc.current_mode & ixgbe_fc_tx_pause) && !pfc_en)) {
2957 for (i = 0; i < adapter->num_rx_queues; i++)
2958 ixgbe_enable_rx_drop(adapter, adapter->rx_ring[i]);
2960 for (i = 0; i < adapter->num_rx_queues; i++)
2961 ixgbe_disable_rx_drop(adapter, adapter->rx_ring[i]);
2965 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2967 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2968 struct ixgbe_ring *rx_ring)
2970 struct ixgbe_hw *hw = &adapter->hw;
2972 u8 reg_idx = rx_ring->reg_idx;
2974 if (hw->mac.type == ixgbe_mac_82598EB) {
2975 u16 mask = adapter->ring_feature[RING_F_RSS].mask;
2978 * if VMDq is not active we must program one srrctl register
2979 * per RSS queue since we have enabled RDRXCTL.MVMEN
2984 /* configure header buffer length, needed for RSC */
2985 srrctl = IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
2987 /* configure the packet buffer length */
2988 srrctl |= ixgbe_rx_bufsz(rx_ring) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2990 /* configure descriptor type */
2991 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2993 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
2996 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2998 struct ixgbe_hw *hw = &adapter->hw;
2999 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
3000 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
3001 0x6A3E67EA, 0x14364D17, 0x3BED200D};
3002 u32 mrqc = 0, reta = 0;
3005 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3008 * Program table for at least 2 queues w/ SR-IOV so that VFs can
3009 * make full use of any rings they may have. We will use the
3010 * PSRTYPE register to control how many rings we use within the PF.
3012 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) && (rss_i < 2))
3015 /* Fill out hash function seeds */
3016 for (i = 0; i < 10; i++)
3017 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
3019 /* Fill out redirection table */
3020 for (i = 0, j = 0; i < 128; i++, j++) {
3023 /* reta = 4-byte sliding window of
3024 * 0x00..(indices-1)(indices-1)00..etc. */
3025 reta = (reta << 8) | (j * 0x11);
3027 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
3030 /* Disable indicating checksum in descriptor, enables RSS hash */
3031 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
3032 rxcsum |= IXGBE_RXCSUM_PCSD;
3033 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
3035 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3036 if (adapter->ring_feature[RING_F_RSS].mask)
3037 mrqc = IXGBE_MRQC_RSSEN;
3039 u8 tcs = netdev_get_num_tc(adapter->netdev);
3041 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3043 mrqc = IXGBE_MRQC_VMDQRT8TCEN; /* 8 TCs */
3045 mrqc = IXGBE_MRQC_VMDQRT4TCEN; /* 4 TCs */
3046 else if (adapter->ring_feature[RING_F_RSS].indices == 4)
3047 mrqc = IXGBE_MRQC_VMDQRSS32EN;
3049 mrqc = IXGBE_MRQC_VMDQRSS64EN;
3052 mrqc = IXGBE_MRQC_RTRSS8TCEN;
3054 mrqc = IXGBE_MRQC_RTRSS4TCEN;
3056 mrqc = IXGBE_MRQC_RSSEN;
3060 /* Perform hash on these packet types */
3061 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4 |
3062 IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
3063 IXGBE_MRQC_RSS_FIELD_IPV6 |
3064 IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
3066 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP)
3067 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
3068 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP)
3069 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
3071 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3075 * ixgbe_configure_rscctl - enable RSC for the indicated ring
3076 * @adapter: address of board private structure
3077 * @index: index of ring to set
3079 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
3080 struct ixgbe_ring *ring)
3082 struct ixgbe_hw *hw = &adapter->hw;
3084 u8 reg_idx = ring->reg_idx;
3086 if (!ring_is_rsc_enabled(ring))
3089 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
3090 rscctrl |= IXGBE_RSCCTL_RSCEN;
3092 * we must limit the number of descriptors so that the
3093 * total size of max desc * buf_len is not greater
3096 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
3097 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
3100 #define IXGBE_MAX_RX_DESC_POLL 10
3101 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
3102 struct ixgbe_ring *ring)
3104 struct ixgbe_hw *hw = &adapter->hw;
3105 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3107 u8 reg_idx = ring->reg_idx;
3109 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3110 if (hw->mac.type == ixgbe_mac_82598EB &&
3111 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3115 usleep_range(1000, 2000);
3116 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3117 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
3120 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
3121 "the polling period\n", reg_idx);
3125 void ixgbe_disable_rx_queue(struct ixgbe_adapter *adapter,
3126 struct ixgbe_ring *ring)
3128 struct ixgbe_hw *hw = &adapter->hw;
3129 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3131 u8 reg_idx = ring->reg_idx;
3133 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3134 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
3136 /* write value back with RXDCTL.ENABLE bit cleared */
3137 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3139 if (hw->mac.type == ixgbe_mac_82598EB &&
3140 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3143 /* the hardware may take up to 100us to really disable the rx queue */
3146 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3147 } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
3150 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not cleared within "
3151 "the polling period\n", reg_idx);
3155 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
3156 struct ixgbe_ring *ring)
3158 struct ixgbe_hw *hw = &adapter->hw;
3159 u64 rdba = ring->dma;
3161 u8 reg_idx = ring->reg_idx;
3163 /* disable queue to avoid issues while updating state */
3164 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3165 ixgbe_disable_rx_queue(adapter, ring);
3167 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
3168 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
3169 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
3170 ring->count * sizeof(union ixgbe_adv_rx_desc));
3171 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
3172 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
3173 ring->tail = hw->hw_addr + IXGBE_RDT(reg_idx);
3175 ixgbe_configure_srrctl(adapter, ring);
3176 ixgbe_configure_rscctl(adapter, ring);
3178 if (hw->mac.type == ixgbe_mac_82598EB) {
3180 * enable cache line friendly hardware writes:
3181 * PTHRESH=32 descriptors (half the internal cache),
3182 * this also removes ugly rx_no_buffer_count increment
3183 * HTHRESH=4 descriptors (to minimize latency on fetch)
3184 * WTHRESH=8 burst writeback up to two cache lines
3186 rxdctl &= ~0x3FFFFF;
3190 /* enable receive descriptor ring */
3191 rxdctl |= IXGBE_RXDCTL_ENABLE;
3192 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3194 ixgbe_rx_desc_queue_enable(adapter, ring);
3195 ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
3198 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
3200 struct ixgbe_hw *hw = &adapter->hw;
3201 int rss_i = adapter->ring_feature[RING_F_RSS].indices;
3204 /* PSRTYPE must be initialized in non 82598 adapters */
3205 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
3206 IXGBE_PSRTYPE_UDPHDR |
3207 IXGBE_PSRTYPE_IPV4HDR |
3208 IXGBE_PSRTYPE_L2HDR |
3209 IXGBE_PSRTYPE_IPV6HDR;
3211 if (hw->mac.type == ixgbe_mac_82598EB)
3219 for (p = 0; p < adapter->num_rx_pools; p++)
3220 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(p)),
3224 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
3226 struct ixgbe_hw *hw = &adapter->hw;
3227 u32 reg_offset, vf_shift;
3228 u32 gcr_ext, vmdctl;
3231 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
3234 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
3235 vmdctl |= IXGBE_VMD_CTL_VMDQ_EN;
3236 vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
3237 vmdctl |= VMDQ_P(0) << IXGBE_VT_CTL_POOL_SHIFT;
3238 vmdctl |= IXGBE_VT_CTL_REPLEN;
3239 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
3241 vf_shift = VMDQ_P(0) % 32;
3242 reg_offset = (VMDQ_P(0) >= 32) ? 1 : 0;
3244 /* Enable only the PF's pool for Tx/Rx */
3245 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (~0) << vf_shift);
3246 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), reg_offset - 1);
3247 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (~0) << vf_shift);
3248 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), reg_offset - 1);
3249 if (adapter->flags2 & IXGBE_FLAG2_BRIDGE_MODE_VEB)
3250 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3252 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
3253 hw->mac.ops.set_vmdq(hw, 0, VMDQ_P(0));
3256 * Set up VF register offsets for selected VT Mode,
3257 * i.e. 32 or 64 VFs for SR-IOV
3259 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
3260 case IXGBE_82599_VMDQ_8Q_MASK:
3261 gcr_ext = IXGBE_GCR_EXT_VT_MODE_16;
3263 case IXGBE_82599_VMDQ_4Q_MASK:
3264 gcr_ext = IXGBE_GCR_EXT_VT_MODE_32;
3267 gcr_ext = IXGBE_GCR_EXT_VT_MODE_64;
3271 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
3274 /* Enable MAC Anti-Spoofing */
3275 hw->mac.ops.set_mac_anti_spoofing(hw, (adapter->num_vfs != 0),
3277 /* For VFs that have spoof checking turned off */
3278 for (i = 0; i < adapter->num_vfs; i++) {
3279 if (!adapter->vfinfo[i].spoofchk_enabled)
3280 ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i, false);
3284 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
3286 struct ixgbe_hw *hw = &adapter->hw;
3287 struct net_device *netdev = adapter->netdev;
3288 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3289 struct ixgbe_ring *rx_ring;
3294 /* adjust max frame to be able to do baby jumbo for FCoE */
3295 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
3296 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
3297 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3299 #endif /* IXGBE_FCOE */
3301 /* adjust max frame to be at least the size of a standard frame */
3302 if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
3303 max_frame = (ETH_FRAME_LEN + ETH_FCS_LEN);
3305 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
3306 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
3307 mhadd &= ~IXGBE_MHADD_MFS_MASK;
3308 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
3310 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
3313 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
3314 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
3315 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
3316 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
3319 * Setup the HW Rx Head and Tail Descriptor Pointers and
3320 * the Base and Length of the Rx Descriptor Ring
3322 for (i = 0; i < adapter->num_rx_queues; i++) {
3323 rx_ring = adapter->rx_ring[i];
3324 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
3325 set_ring_rsc_enabled(rx_ring);
3327 clear_ring_rsc_enabled(rx_ring);
3331 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
3333 struct ixgbe_hw *hw = &adapter->hw;
3334 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
3336 switch (hw->mac.type) {
3337 case ixgbe_mac_82598EB:
3339 * For VMDq support of different descriptor types or
3340 * buffer sizes through the use of multiple SRRCTL
3341 * registers, RDRXCTL.MVMEN must be set to 1
3343 * also, the manual doesn't mention it clearly but DCA hints
3344 * will only use queue 0's tags unless this bit is set. Side
3345 * effects of setting this bit are only that SRRCTL must be
3346 * fully programmed [0..15]
3348 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
3350 case ixgbe_mac_82599EB:
3351 case ixgbe_mac_X540:
3352 /* Disable RSC for ACK packets */
3353 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
3354 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
3355 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3356 /* hardware requires some bits to be set by default */
3357 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3358 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3361 /* We should do nothing since we don't know this hardware */
3365 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3369 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3370 * @adapter: board private structure
3372 * Configure the Rx unit of the MAC after a reset.
3374 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3376 struct ixgbe_hw *hw = &adapter->hw;
3380 /* disable receives while setting up the descriptors */
3381 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3382 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3384 ixgbe_setup_psrtype(adapter);
3385 ixgbe_setup_rdrxctl(adapter);
3387 /* Program registers for the distribution of queues */
3388 ixgbe_setup_mrqc(adapter);
3390 /* set_rx_buffer_len must be called before ring initialization */
3391 ixgbe_set_rx_buffer_len(adapter);
3394 * Setup the HW Rx Head and Tail Descriptor Pointers and
3395 * the Base and Length of the Rx Descriptor Ring
3397 for (i = 0; i < adapter->num_rx_queues; i++)
3398 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3400 /* disable drop enable for 82598 parts */
3401 if (hw->mac.type == ixgbe_mac_82598EB)
3402 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3404 /* enable all receives */
3405 rxctrl |= IXGBE_RXCTRL_RXEN;
3406 hw->mac.ops.enable_rx_dma(hw, rxctrl);
3409 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3411 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3412 struct ixgbe_hw *hw = &adapter->hw;
3414 /* add VID to filter table */
3415 hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), true);
3416 set_bit(vid, adapter->active_vlans);
3421 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3423 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3424 struct ixgbe_hw *hw = &adapter->hw;
3426 /* remove VID from filter table */
3427 hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), false);
3428 clear_bit(vid, adapter->active_vlans);
3434 * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3435 * @adapter: driver data
3437 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
3439 struct ixgbe_hw *hw = &adapter->hw;
3442 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3443 vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
3444 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3448 * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3449 * @adapter: driver data
3451 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
3453 struct ixgbe_hw *hw = &adapter->hw;
3456 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3457 vlnctrl |= IXGBE_VLNCTRL_VFE;
3458 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3459 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3463 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3464 * @adapter: driver data
3466 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3468 struct ixgbe_hw *hw = &adapter->hw;
3472 switch (hw->mac.type) {
3473 case ixgbe_mac_82598EB:
3474 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3475 vlnctrl &= ~IXGBE_VLNCTRL_VME;
3476 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3478 case ixgbe_mac_82599EB:
3479 case ixgbe_mac_X540:
3480 for (i = 0; i < adapter->num_rx_queues; i++) {
3481 j = adapter->rx_ring[i]->reg_idx;
3482 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3483 vlnctrl &= ~IXGBE_RXDCTL_VME;
3484 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3493 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3494 * @adapter: driver data
3496 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3498 struct ixgbe_hw *hw = &adapter->hw;
3502 switch (hw->mac.type) {
3503 case ixgbe_mac_82598EB:
3504 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3505 vlnctrl |= IXGBE_VLNCTRL_VME;
3506 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3508 case ixgbe_mac_82599EB:
3509 case ixgbe_mac_X540:
3510 for (i = 0; i < adapter->num_rx_queues; i++) {
3511 j = adapter->rx_ring[i]->reg_idx;
3512 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3513 vlnctrl |= IXGBE_RXDCTL_VME;
3514 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3522 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3526 ixgbe_vlan_rx_add_vid(adapter->netdev, 0);
3528 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3529 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
3533 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3534 * @netdev: network interface device structure
3536 * Writes unicast address list to the RAR table.
3537 * Returns: -ENOMEM on failure/insufficient address space
3538 * 0 on no addresses written
3539 * X on writing X addresses to the RAR table
3541 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3543 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3544 struct ixgbe_hw *hw = &adapter->hw;
3545 unsigned int rar_entries = hw->mac.num_rar_entries - 1;
3548 /* In SR-IOV mode significantly less RAR entries are available */
3549 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3550 rar_entries = IXGBE_MAX_PF_MACVLANS - 1;
3552 /* return ENOMEM indicating insufficient memory for addresses */
3553 if (netdev_uc_count(netdev) > rar_entries)
3556 if (!netdev_uc_empty(netdev)) {
3557 struct netdev_hw_addr *ha;
3558 /* return error if we do not support writing to RAR table */
3559 if (!hw->mac.ops.set_rar)
3562 netdev_for_each_uc_addr(ha, netdev) {
3565 hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3566 VMDQ_P(0), IXGBE_RAH_AV);
3570 /* write the addresses in reverse order to avoid write combining */
3571 for (; rar_entries > 0 ; rar_entries--)
3572 hw->mac.ops.clear_rar(hw, rar_entries);
3578 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3579 * @netdev: network interface device structure
3581 * The set_rx_method entry point is called whenever the unicast/multicast
3582 * address list or the network interface flags are updated. This routine is
3583 * responsible for configuring the hardware for proper unicast, multicast and
3586 void ixgbe_set_rx_mode(struct net_device *netdev)
3588 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3589 struct ixgbe_hw *hw = &adapter->hw;
3590 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3593 /* Check for Promiscuous and All Multicast modes */
3595 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3597 /* set all bits that we expect to always be set */
3598 fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
3599 fctrl |= IXGBE_FCTRL_BAM;
3600 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3601 fctrl |= IXGBE_FCTRL_PMCF;
3603 /* clear the bits we are changing the status of */
3604 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3606 if (netdev->flags & IFF_PROMISC) {
3607 hw->addr_ctrl.user_set_promisc = true;
3608 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3609 vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3610 /* don't hardware filter vlans in promisc mode */
3611 ixgbe_vlan_filter_disable(adapter);
3613 if (netdev->flags & IFF_ALLMULTI) {
3614 fctrl |= IXGBE_FCTRL_MPE;
3615 vmolr |= IXGBE_VMOLR_MPE;
3618 * Write addresses to the MTA, if the attempt fails
3619 * then we should just turn on promiscuous mode so
3620 * that we can at least receive multicast traffic
3622 hw->mac.ops.update_mc_addr_list(hw, netdev);
3623 vmolr |= IXGBE_VMOLR_ROMPE;
3625 ixgbe_vlan_filter_enable(adapter);
3626 hw->addr_ctrl.user_set_promisc = false;
3630 * Write addresses to available RAR registers, if there is not
3631 * sufficient space to store all the addresses then enable
3632 * unicast promiscuous mode
3634 count = ixgbe_write_uc_addr_list(netdev);
3636 fctrl |= IXGBE_FCTRL_UPE;
3637 vmolr |= IXGBE_VMOLR_ROPE;
3640 if (adapter->num_vfs)
3641 ixgbe_restore_vf_multicasts(adapter);
3643 if (hw->mac.type != ixgbe_mac_82598EB) {
3644 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(VMDQ_P(0))) &
3645 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3647 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(0)), vmolr);
3650 /* This is useful for sniffing bad packets. */
3651 if (adapter->netdev->features & NETIF_F_RXALL) {
3652 /* UPE and MPE will be handled by normal PROMISC logic
3653 * in e1000e_set_rx_mode */
3654 fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
3655 IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
3656 IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
3658 fctrl &= ~(IXGBE_FCTRL_DPF);
3659 /* NOTE: VLAN filtering is disabled by setting PROMISC */
3662 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3664 if (netdev->features & NETIF_F_HW_VLAN_RX)
3665 ixgbe_vlan_strip_enable(adapter);
3667 ixgbe_vlan_strip_disable(adapter);
3670 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3674 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
3675 napi_enable(&adapter->q_vector[q_idx]->napi);
3678 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3682 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
3683 napi_disable(&adapter->q_vector[q_idx]->napi);
3686 #ifdef CONFIG_IXGBE_DCB
3688 * ixgbe_configure_dcb - Configure DCB hardware
3689 * @adapter: ixgbe adapter struct
3691 * This is called by the driver on open to configure the DCB hardware.
3692 * This is also called by the gennetlink interface when reconfiguring
3695 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3697 struct ixgbe_hw *hw = &adapter->hw;
3698 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3700 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3701 if (hw->mac.type == ixgbe_mac_82598EB)
3702 netif_set_gso_max_size(adapter->netdev, 65536);
3706 if (hw->mac.type == ixgbe_mac_82598EB)
3707 netif_set_gso_max_size(adapter->netdev, 32768);
3710 if (adapter->netdev->features & NETIF_F_FCOE_MTU)
3711 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
3714 /* reconfigure the hardware */
3715 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
3716 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3718 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3720 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
3721 } else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
3722 ixgbe_dcb_hw_ets(&adapter->hw,
3723 adapter->ixgbe_ieee_ets,
3725 ixgbe_dcb_hw_pfc_config(&adapter->hw,
3726 adapter->ixgbe_ieee_pfc->pfc_en,
3727 adapter->ixgbe_ieee_ets->prio_tc);
3730 /* Enable RSS Hash per TC */
3731 if (hw->mac.type != ixgbe_mac_82598EB) {
3733 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices - 1;
3740 /* write msb to all 8 TCs in one write */
3741 IXGBE_WRITE_REG(hw, IXGBE_RQTC, msb * 0x11111111);
3746 /* Additional bittime to account for IXGBE framing */
3747 #define IXGBE_ETH_FRAMING 20
3750 * ixgbe_hpbthresh - calculate high water mark for flow control
3752 * @adapter: board private structure to calculate for
3753 * @pb: packet buffer to calculate
3755 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
3757 struct ixgbe_hw *hw = &adapter->hw;
3758 struct net_device *dev = adapter->netdev;
3759 int link, tc, kb, marker;
3762 /* Calculate max LAN frame size */
3763 tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
3766 /* FCoE traffic class uses FCOE jumbo frames */
3767 if ((dev->features & NETIF_F_FCOE_MTU) &&
3768 (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
3769 (pb == ixgbe_fcoe_get_tc(adapter)))
3770 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3773 /* Calculate delay value for device */
3774 switch (hw->mac.type) {
3775 case ixgbe_mac_X540:
3776 dv_id = IXGBE_DV_X540(link, tc);
3779 dv_id = IXGBE_DV(link, tc);
3783 /* Loopback switch introduces additional latency */
3784 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3785 dv_id += IXGBE_B2BT(tc);
3787 /* Delay value is calculated in bit times convert to KB */
3788 kb = IXGBE_BT2KB(dv_id);
3789 rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
3791 marker = rx_pba - kb;
3793 /* It is possible that the packet buffer is not large enough
3794 * to provide required headroom. In this case throw an error
3795 * to user and a do the best we can.
3798 e_warn(drv, "Packet Buffer(%i) can not provide enough"
3799 "headroom to support flow control."
3800 "Decrease MTU or number of traffic classes\n", pb);
3808 * ixgbe_lpbthresh - calculate low water mark for for flow control
3810 * @adapter: board private structure to calculate for
3811 * @pb: packet buffer to calculate
3813 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter)
3815 struct ixgbe_hw *hw = &adapter->hw;
3816 struct net_device *dev = adapter->netdev;
3820 /* Calculate max LAN frame size */
3821 tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
3823 /* Calculate delay value for device */
3824 switch (hw->mac.type) {
3825 case ixgbe_mac_X540:
3826 dv_id = IXGBE_LOW_DV_X540(tc);
3829 dv_id = IXGBE_LOW_DV(tc);
3833 /* Delay value is calculated in bit times convert to KB */
3834 return IXGBE_BT2KB(dv_id);
3838 * ixgbe_pbthresh_setup - calculate and setup high low water marks
3840 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
3842 struct ixgbe_hw *hw = &adapter->hw;
3843 int num_tc = netdev_get_num_tc(adapter->netdev);
3849 hw->fc.low_water = ixgbe_lpbthresh(adapter);
3851 for (i = 0; i < num_tc; i++) {
3852 hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
3854 /* Low water marks must not be larger than high water marks */
3855 if (hw->fc.low_water > hw->fc.high_water[i])
3856 hw->fc.low_water = 0;
3860 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
3862 struct ixgbe_hw *hw = &adapter->hw;
3864 u8 tc = netdev_get_num_tc(adapter->netdev);
3866 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3867 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3868 hdrm = 32 << adapter->fdir_pballoc;
3872 hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
3873 ixgbe_pbthresh_setup(adapter);
3876 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
3878 struct ixgbe_hw *hw = &adapter->hw;
3879 struct hlist_node *node, *node2;
3880 struct ixgbe_fdir_filter *filter;
3882 spin_lock(&adapter->fdir_perfect_lock);
3884 if (!hlist_empty(&adapter->fdir_filter_list))
3885 ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
3887 hlist_for_each_entry_safe(filter, node, node2,
3888 &adapter->fdir_filter_list, fdir_node) {
3889 ixgbe_fdir_write_perfect_filter_82599(hw,
3892 (filter->action == IXGBE_FDIR_DROP_QUEUE) ?
3893 IXGBE_FDIR_DROP_QUEUE :
3894 adapter->rx_ring[filter->action]->reg_idx);
3897 spin_unlock(&adapter->fdir_perfect_lock);
3900 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3902 struct ixgbe_hw *hw = &adapter->hw;
3904 ixgbe_configure_pb(adapter);
3905 #ifdef CONFIG_IXGBE_DCB
3906 ixgbe_configure_dcb(adapter);
3909 * We must restore virtualization before VLANs or else
3910 * the VLVF registers will not be populated
3912 ixgbe_configure_virtualization(adapter);
3914 ixgbe_set_rx_mode(adapter->netdev);
3915 ixgbe_restore_vlan(adapter);
3917 switch (hw->mac.type) {
3918 case ixgbe_mac_82599EB:
3919 case ixgbe_mac_X540:
3920 hw->mac.ops.disable_rx_buff(hw);
3926 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3927 ixgbe_init_fdir_signature_82599(&adapter->hw,
3928 adapter->fdir_pballoc);
3929 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3930 ixgbe_init_fdir_perfect_82599(&adapter->hw,
3931 adapter->fdir_pballoc);
3932 ixgbe_fdir_filter_restore(adapter);
3935 switch (hw->mac.type) {
3936 case ixgbe_mac_82599EB:
3937 case ixgbe_mac_X540:
3938 hw->mac.ops.enable_rx_buff(hw);
3945 /* configure FCoE L2 filters, redirection table, and Rx control */
3946 ixgbe_configure_fcoe(adapter);
3948 #endif /* IXGBE_FCOE */
3949 ixgbe_configure_tx(adapter);
3950 ixgbe_configure_rx(adapter);
3953 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3955 switch (hw->phy.type) {
3956 case ixgbe_phy_sfp_avago:
3957 case ixgbe_phy_sfp_ftl:
3958 case ixgbe_phy_sfp_intel:
3959 case ixgbe_phy_sfp_unknown:
3960 case ixgbe_phy_sfp_passive_tyco:
3961 case ixgbe_phy_sfp_passive_unknown:
3962 case ixgbe_phy_sfp_active_unknown:
3963 case ixgbe_phy_sfp_ftl_active:
3966 if (hw->mac.type == ixgbe_mac_82598EB)
3974 * ixgbe_sfp_link_config - set up SFP+ link
3975 * @adapter: pointer to private adapter struct
3977 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3980 * We are assuming the worst case scenario here, and that
3981 * is that an SFP was inserted/removed after the reset
3982 * but before SFP detection was enabled. As such the best
3983 * solution is to just start searching as soon as we start
3985 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3986 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
3988 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
3992 * ixgbe_non_sfp_link_config - set up non-SFP+ link
3993 * @hw: pointer to private hardware struct
3995 * Returns 0 on success, negative on failure
3997 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
4000 bool negotiation, link_up = false;
4001 u32 ret = IXGBE_ERR_LINK_SETUP;
4003 if (hw->mac.ops.check_link)
4004 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
4009 autoneg = hw->phy.autoneg_advertised;
4010 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
4011 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
4016 if (hw->mac.ops.setup_link)
4017 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
4022 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
4024 struct ixgbe_hw *hw = &adapter->hw;
4027 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4028 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
4030 gpie |= IXGBE_GPIE_EIAME;
4032 * use EIAM to auto-mask when MSI-X interrupt is asserted
4033 * this saves a register write for every interrupt
4035 switch (hw->mac.type) {
4036 case ixgbe_mac_82598EB:
4037 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
4039 case ixgbe_mac_82599EB:
4040 case ixgbe_mac_X540:
4042 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
4043 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
4047 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
4048 * specifically only auto mask tx and rx interrupts */
4049 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
4052 /* XXX: to interrupt immediately for EICS writes, enable this */
4053 /* gpie |= IXGBE_GPIE_EIMEN; */
4055 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
4056 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
4058 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
4059 case IXGBE_82599_VMDQ_8Q_MASK:
4060 gpie |= IXGBE_GPIE_VTMODE_16;
4062 case IXGBE_82599_VMDQ_4Q_MASK:
4063 gpie |= IXGBE_GPIE_VTMODE_32;
4066 gpie |= IXGBE_GPIE_VTMODE_64;
4071 /* Enable Thermal over heat sensor interrupt */
4072 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
4073 switch (adapter->hw.mac.type) {
4074 case ixgbe_mac_82599EB:
4075 gpie |= IXGBE_SDP0_GPIEN;
4077 case ixgbe_mac_X540:
4078 gpie |= IXGBE_EIMS_TS;
4085 /* Enable fan failure interrupt */
4086 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
4087 gpie |= IXGBE_SDP1_GPIEN;
4089 if (hw->mac.type == ixgbe_mac_82599EB) {
4090 gpie |= IXGBE_SDP1_GPIEN;
4091 gpie |= IXGBE_SDP2_GPIEN;
4094 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
4097 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
4099 struct ixgbe_hw *hw = &adapter->hw;
4103 ixgbe_get_hw_control(adapter);
4104 ixgbe_setup_gpie(adapter);
4106 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4107 ixgbe_configure_msix(adapter);
4109 ixgbe_configure_msi_and_legacy(adapter);
4111 /* enable the optics for 82599 SFP+ fiber */
4112 if (hw->mac.ops.enable_tx_laser)
4113 hw->mac.ops.enable_tx_laser(hw);
4115 clear_bit(__IXGBE_DOWN, &adapter->state);
4116 ixgbe_napi_enable_all(adapter);
4118 if (ixgbe_is_sfp(hw)) {
4119 ixgbe_sfp_link_config(adapter);
4121 err = ixgbe_non_sfp_link_config(hw);
4123 e_err(probe, "link_config FAILED %d\n", err);
4126 /* clear any pending interrupts, may auto mask */
4127 IXGBE_READ_REG(hw, IXGBE_EICR);
4128 ixgbe_irq_enable(adapter, true, true);
4131 * If this adapter has a fan, check to see if we had a failure
4132 * before we enabled the interrupt.
4134 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
4135 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
4136 if (esdp & IXGBE_ESDP_SDP1)
4137 e_crit(drv, "Fan has stopped, replace the adapter\n");
4140 /* enable transmits */
4141 netif_tx_start_all_queues(adapter->netdev);
4143 /* bring the link up in the watchdog, this could race with our first
4144 * link up interrupt but shouldn't be a problem */
4145 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
4146 adapter->link_check_timeout = jiffies;
4147 mod_timer(&adapter->service_timer, jiffies);
4149 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
4150 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
4151 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
4152 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
4155 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
4157 WARN_ON(in_interrupt());
4158 /* put off any impending NetWatchDogTimeout */
4159 adapter->netdev->trans_start = jiffies;
4161 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
4162 usleep_range(1000, 2000);
4163 ixgbe_down(adapter);
4165 * If SR-IOV enabled then wait a bit before bringing the adapter
4166 * back up to give the VFs time to respond to the reset. The
4167 * two second wait is based upon the watchdog timer cycle in
4170 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4173 clear_bit(__IXGBE_RESETTING, &adapter->state);
4176 void ixgbe_up(struct ixgbe_adapter *adapter)
4178 /* hardware has been reset, we need to reload some things */
4179 ixgbe_configure(adapter);
4181 ixgbe_up_complete(adapter);
4184 void ixgbe_reset(struct ixgbe_adapter *adapter)
4186 struct ixgbe_hw *hw = &adapter->hw;
4189 /* lock SFP init bit to prevent race conditions with the watchdog */
4190 while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
4191 usleep_range(1000, 2000);
4193 /* clear all SFP and link config related flags while holding SFP_INIT */
4194 adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
4195 IXGBE_FLAG2_SFP_NEEDS_RESET);
4196 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
4198 err = hw->mac.ops.init_hw(hw);
4201 case IXGBE_ERR_SFP_NOT_PRESENT:
4202 case IXGBE_ERR_SFP_NOT_SUPPORTED:
4204 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
4205 e_dev_err("master disable timed out\n");
4207 case IXGBE_ERR_EEPROM_VERSION:
4208 /* We are running on a pre-production device, log a warning */
4209 e_dev_warn("This device is a pre-production adapter/LOM. "
4210 "Please be aware there may be issues associated with "
4211 "your hardware. If you are experiencing problems "
4212 "please contact your Intel or hardware "
4213 "representative who provided you with this "
4217 e_dev_err("Hardware Error: %d\n", err);
4220 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
4222 /* reprogram the RAR[0] in case user changed it. */
4223 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, VMDQ_P(0), IXGBE_RAH_AV);
4225 /* update SAN MAC vmdq pool selection */
4226 if (hw->mac.san_mac_rar_index)
4227 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
4229 if (adapter->flags2 & IXGBE_FLAG2_PTP_ENABLED)
4230 ixgbe_ptp_reset(adapter);
4234 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
4235 * @rx_ring: ring to free buffers from
4237 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
4239 struct device *dev = rx_ring->dev;
4243 /* ring already cleared, nothing to do */
4244 if (!rx_ring->rx_buffer_info)
4247 /* Free all the Rx ring sk_buffs */
4248 for (i = 0; i < rx_ring->count; i++) {
4249 struct ixgbe_rx_buffer *rx_buffer;
4251 rx_buffer = &rx_ring->rx_buffer_info[i];
4252 if (rx_buffer->skb) {
4253 struct sk_buff *skb = rx_buffer->skb;
4254 if (IXGBE_CB(skb)->page_released) {
4257 ixgbe_rx_bufsz(rx_ring),
4259 IXGBE_CB(skb)->page_released = false;
4263 rx_buffer->skb = NULL;
4265 dma_unmap_page(dev, rx_buffer->dma,
4266 ixgbe_rx_pg_size(rx_ring),
4269 if (rx_buffer->page)
4270 __free_pages(rx_buffer->page,
4271 ixgbe_rx_pg_order(rx_ring));
4272 rx_buffer->page = NULL;
4275 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4276 memset(rx_ring->rx_buffer_info, 0, size);
4278 /* Zero out the descriptor ring */
4279 memset(rx_ring->desc, 0, rx_ring->size);
4281 rx_ring->next_to_alloc = 0;
4282 rx_ring->next_to_clean = 0;
4283 rx_ring->next_to_use = 0;
4287 * ixgbe_clean_tx_ring - Free Tx Buffers
4288 * @tx_ring: ring to be cleaned
4290 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
4292 struct ixgbe_tx_buffer *tx_buffer_info;
4296 /* ring already cleared, nothing to do */
4297 if (!tx_ring->tx_buffer_info)
4300 /* Free all the Tx ring sk_buffs */
4301 for (i = 0; i < tx_ring->count; i++) {
4302 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4303 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
4306 netdev_tx_reset_queue(txring_txq(tx_ring));
4308 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4309 memset(tx_ring->tx_buffer_info, 0, size);
4311 /* Zero out the descriptor ring */
4312 memset(tx_ring->desc, 0, tx_ring->size);
4314 tx_ring->next_to_use = 0;
4315 tx_ring->next_to_clean = 0;
4319 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
4320 * @adapter: board private structure
4322 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
4326 for (i = 0; i < adapter->num_rx_queues; i++)
4327 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
4331 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
4332 * @adapter: board private structure
4334 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
4338 for (i = 0; i < adapter->num_tx_queues; i++)
4339 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
4342 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
4344 struct hlist_node *node, *node2;
4345 struct ixgbe_fdir_filter *filter;
4347 spin_lock(&adapter->fdir_perfect_lock);
4349 hlist_for_each_entry_safe(filter, node, node2,
4350 &adapter->fdir_filter_list, fdir_node) {
4351 hlist_del(&filter->fdir_node);
4354 adapter->fdir_filter_count = 0;
4356 spin_unlock(&adapter->fdir_perfect_lock);
4359 void ixgbe_down(struct ixgbe_adapter *adapter)
4361 struct net_device *netdev = adapter->netdev;
4362 struct ixgbe_hw *hw = &adapter->hw;
4366 /* signal that we are down to the interrupt handler */
4367 set_bit(__IXGBE_DOWN, &adapter->state);
4369 /* disable receives */
4370 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4371 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
4373 /* disable all enabled rx queues */
4374 for (i = 0; i < adapter->num_rx_queues; i++)
4375 /* this call also flushes the previous write */
4376 ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]);
4378 usleep_range(10000, 20000);
4380 netif_tx_stop_all_queues(netdev);
4382 /* call carrier off first to avoid false dev_watchdog timeouts */
4383 netif_carrier_off(netdev);
4384 netif_tx_disable(netdev);
4386 ixgbe_irq_disable(adapter);
4388 ixgbe_napi_disable_all(adapter);
4390 adapter->flags2 &= ~(IXGBE_FLAG2_FDIR_REQUIRES_REINIT |
4391 IXGBE_FLAG2_RESET_REQUESTED);
4392 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
4394 del_timer_sync(&adapter->service_timer);
4396 if (adapter->num_vfs) {
4397 /* Clear EITR Select mapping */
4398 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
4400 /* Mark all the VFs as inactive */
4401 for (i = 0 ; i < adapter->num_vfs; i++)
4402 adapter->vfinfo[i].clear_to_send = false;
4404 /* ping all the active vfs to let them know we are going down */
4405 ixgbe_ping_all_vfs(adapter);
4407 /* Disable all VFTE/VFRE TX/RX */
4408 ixgbe_disable_tx_rx(adapter);
4411 /* disable transmits in the hardware now that interrupts are off */
4412 for (i = 0; i < adapter->num_tx_queues; i++) {
4413 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
4414 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
4417 /* Disable the Tx DMA engine on 82599 and X540 */
4418 switch (hw->mac.type) {
4419 case ixgbe_mac_82599EB:
4420 case ixgbe_mac_X540:
4421 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
4422 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
4423 ~IXGBE_DMATXCTL_TE));
4429 if (!pci_channel_offline(adapter->pdev))
4430 ixgbe_reset(adapter);
4432 /* power down the optics for 82599 SFP+ fiber */
4433 if (hw->mac.ops.disable_tx_laser)
4434 hw->mac.ops.disable_tx_laser(hw);
4436 ixgbe_clean_all_tx_rings(adapter);
4437 ixgbe_clean_all_rx_rings(adapter);
4439 #ifdef CONFIG_IXGBE_DCA
4440 /* since we reset the hardware DCA settings were cleared */
4441 ixgbe_setup_dca(adapter);
4446 * ixgbe_tx_timeout - Respond to a Tx Hang
4447 * @netdev: network interface device structure
4449 static void ixgbe_tx_timeout(struct net_device *netdev)
4451 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4453 /* Do the reset outside of interrupt context */
4454 ixgbe_tx_timeout_reset(adapter);
4458 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4459 * @adapter: board private structure to initialize
4461 * ixgbe_sw_init initializes the Adapter private data structure.
4462 * Fields are initialized based on PCI device information and
4463 * OS network device settings (MTU size).
4465 static int ixgbe_sw_init(struct ixgbe_adapter *adapter)
4467 struct ixgbe_hw *hw = &adapter->hw;
4468 struct pci_dev *pdev = adapter->pdev;
4471 #ifdef CONFIG_IXGBE_DCB
4473 struct tc_configuration *tc;
4476 /* PCI config space info */
4478 hw->vendor_id = pdev->vendor;
4479 hw->device_id = pdev->device;
4480 hw->revision_id = pdev->revision;
4481 hw->subsystem_vendor_id = pdev->subsystem_vendor;
4482 hw->subsystem_device_id = pdev->subsystem_device;
4484 /* Set capability flags */
4485 rss = min_t(int, IXGBE_MAX_RSS_INDICES, num_online_cpus());
4486 adapter->ring_feature[RING_F_RSS].limit = rss;
4487 switch (hw->mac.type) {
4488 case ixgbe_mac_82598EB:
4489 if (hw->device_id == IXGBE_DEV_ID_82598AT)
4490 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4491 adapter->max_q_vectors = MAX_Q_VECTORS_82598;
4493 case ixgbe_mac_X540:
4494 fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM);
4495 if (fwsm & IXGBE_FWSM_TS_ENABLED)
4496 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4497 case ixgbe_mac_82599EB:
4498 adapter->max_q_vectors = MAX_Q_VECTORS_82599;
4499 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4500 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4501 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
4502 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4503 /* Flow Director hash filters enabled */
4504 adapter->atr_sample_rate = 20;
4505 adapter->ring_feature[RING_F_FDIR].limit =
4506 IXGBE_MAX_FDIR_INDICES;
4507 adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
4509 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4510 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4511 #ifdef CONFIG_IXGBE_DCB
4512 /* Default traffic class to use for FCoE */
4513 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
4515 #endif /* IXGBE_FCOE */
4522 /* FCoE support exists, always init the FCoE lock */
4523 spin_lock_init(&adapter->fcoe.lock);
4526 /* n-tuple support exists, always init our spinlock */
4527 spin_lock_init(&adapter->fdir_perfect_lock);
4529 #ifdef CONFIG_IXGBE_DCB
4530 switch (hw->mac.type) {
4531 case ixgbe_mac_X540:
4532 adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
4533 adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
4536 adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
4537 adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
4541 /* Configure DCB traffic classes */
4542 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
4543 tc = &adapter->dcb_cfg.tc_config[j];
4544 tc->path[DCB_TX_CONFIG].bwg_id = 0;
4545 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
4546 tc->path[DCB_RX_CONFIG].bwg_id = 0;
4547 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
4548 tc->dcb_pfc = pfc_disabled;
4551 /* Initialize default user to priority mapping, UPx->TC0 */
4552 tc = &adapter->dcb_cfg.tc_config[0];
4553 tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
4554 tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
4556 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
4557 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
4558 adapter->dcb_cfg.pfc_mode_enable = false;
4559 adapter->dcb_set_bitmap = 0x00;
4560 adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
4561 memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg,
4562 sizeof(adapter->temp_dcb_cfg));
4566 /* default flow control settings */
4567 hw->fc.requested_mode = ixgbe_fc_full;
4568 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
4569 ixgbe_pbthresh_setup(adapter);
4570 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
4571 hw->fc.send_xon = true;
4572 hw->fc.disable_fc_autoneg =
4573 (ixgbe_device_supports_autoneg_fc(hw) == 0) ? false : true;
4575 #ifdef CONFIG_PCI_IOV
4576 /* assign number of SR-IOV VFs */
4577 if (hw->mac.type != ixgbe_mac_82598EB)
4578 adapter->num_vfs = (max_vfs > 63) ? 0 : max_vfs;
4581 /* enable itr by default in dynamic mode */
4582 adapter->rx_itr_setting = 1;
4583 adapter->tx_itr_setting = 1;
4585 /* set default ring sizes */
4586 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
4587 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
4589 /* set default work limits */
4590 adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
4592 /* initialize eeprom parameters */
4593 if (ixgbe_init_eeprom_params_generic(hw)) {
4594 e_dev_err("EEPROM initialization failed\n");
4598 set_bit(__IXGBE_DOWN, &adapter->state);
4604 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
4605 * @tx_ring: tx descriptor ring (for a specific queue) to setup
4607 * Return 0 on success, negative on failure
4609 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
4611 struct device *dev = tx_ring->dev;
4612 int orig_node = dev_to_node(dev);
4616 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4618 if (tx_ring->q_vector)
4619 numa_node = tx_ring->q_vector->numa_node;
4621 tx_ring->tx_buffer_info = vzalloc_node(size, numa_node);
4622 if (!tx_ring->tx_buffer_info)
4623 tx_ring->tx_buffer_info = vzalloc(size);
4624 if (!tx_ring->tx_buffer_info)
4627 /* round up to nearest 4K */
4628 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
4629 tx_ring->size = ALIGN(tx_ring->size, 4096);
4631 set_dev_node(dev, numa_node);
4632 tx_ring->desc = dma_alloc_coherent(dev,
4636 set_dev_node(dev, orig_node);
4638 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
4639 &tx_ring->dma, GFP_KERNEL);
4643 tx_ring->next_to_use = 0;
4644 tx_ring->next_to_clean = 0;
4648 vfree(tx_ring->tx_buffer_info);
4649 tx_ring->tx_buffer_info = NULL;
4650 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
4655 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
4656 * @adapter: board private structure
4658 * If this function returns with an error, then it's possible one or
4659 * more of the rings is populated (while the rest are not). It is the
4660 * callers duty to clean those orphaned rings.
4662 * Return 0 on success, negative on failure
4664 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
4668 for (i = 0; i < adapter->num_tx_queues; i++) {
4669 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
4673 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
4679 /* rewind the index freeing the rings as we go */
4681 ixgbe_free_tx_resources(adapter->tx_ring[i]);
4686 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
4687 * @rx_ring: rx descriptor ring (for a specific queue) to setup
4689 * Returns 0 on success, negative on failure
4691 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
4693 struct device *dev = rx_ring->dev;
4694 int orig_node = dev_to_node(dev);
4698 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4700 if (rx_ring->q_vector)
4701 numa_node = rx_ring->q_vector->numa_node;
4703 rx_ring->rx_buffer_info = vzalloc_node(size, numa_node);
4704 if (!rx_ring->rx_buffer_info)
4705 rx_ring->rx_buffer_info = vzalloc(size);
4706 if (!rx_ring->rx_buffer_info)
4709 /* Round up to nearest 4K */
4710 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
4711 rx_ring->size = ALIGN(rx_ring->size, 4096);
4713 set_dev_node(dev, numa_node);
4714 rx_ring->desc = dma_alloc_coherent(dev,
4718 set_dev_node(dev, orig_node);
4720 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
4721 &rx_ring->dma, GFP_KERNEL);
4725 rx_ring->next_to_clean = 0;
4726 rx_ring->next_to_use = 0;
4730 vfree(rx_ring->rx_buffer_info);
4731 rx_ring->rx_buffer_info = NULL;
4732 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
4737 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
4738 * @adapter: board private structure
4740 * If this function returns with an error, then it's possible one or
4741 * more of the rings is populated (while the rest are not). It is the
4742 * callers duty to clean those orphaned rings.
4744 * Return 0 on success, negative on failure
4746 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
4750 for (i = 0; i < adapter->num_rx_queues; i++) {
4751 err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
4755 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
4760 err = ixgbe_setup_fcoe_ddp_resources(adapter);
4765 /* rewind the index freeing the rings as we go */
4767 ixgbe_free_rx_resources(adapter->rx_ring[i]);
4772 * ixgbe_free_tx_resources - Free Tx Resources per Queue
4773 * @tx_ring: Tx descriptor ring for a specific queue
4775 * Free all transmit software resources
4777 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
4779 ixgbe_clean_tx_ring(tx_ring);
4781 vfree(tx_ring->tx_buffer_info);
4782 tx_ring->tx_buffer_info = NULL;
4784 /* if not set, then don't free */
4788 dma_free_coherent(tx_ring->dev, tx_ring->size,
4789 tx_ring->desc, tx_ring->dma);
4791 tx_ring->desc = NULL;
4795 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
4796 * @adapter: board private structure
4798 * Free all transmit software resources
4800 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
4804 for (i = 0; i < adapter->num_tx_queues; i++)
4805 if (adapter->tx_ring[i]->desc)
4806 ixgbe_free_tx_resources(adapter->tx_ring[i]);
4810 * ixgbe_free_rx_resources - Free Rx Resources
4811 * @rx_ring: ring to clean the resources from
4813 * Free all receive software resources
4815 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
4817 ixgbe_clean_rx_ring(rx_ring);
4819 vfree(rx_ring->rx_buffer_info);
4820 rx_ring->rx_buffer_info = NULL;
4822 /* if not set, then don't free */
4826 dma_free_coherent(rx_ring->dev, rx_ring->size,
4827 rx_ring->desc, rx_ring->dma);
4829 rx_ring->desc = NULL;
4833 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
4834 * @adapter: board private structure
4836 * Free all receive software resources
4838 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
4843 ixgbe_free_fcoe_ddp_resources(adapter);
4846 for (i = 0; i < adapter->num_rx_queues; i++)
4847 if (adapter->rx_ring[i]->desc)
4848 ixgbe_free_rx_resources(adapter->rx_ring[i]);
4852 * ixgbe_change_mtu - Change the Maximum Transfer Unit
4853 * @netdev: network interface device structure
4854 * @new_mtu: new value for maximum frame size
4856 * Returns 0 on success, negative on failure
4858 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
4860 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4861 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4863 /* MTU < 68 is an error and causes problems on some kernels */
4864 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
4868 * For 82599EB we cannot allow legacy VFs to enable their receive
4869 * paths when MTU greater than 1500 is configured. So display a
4870 * warning that legacy VFs will be disabled.
4872 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
4873 (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
4874 (max_frame > MAXIMUM_ETHERNET_VLAN_SIZE))
4875 e_warn(probe, "Setting MTU > 1500 will disable legacy VFs\n");
4877 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4879 /* must set new MTU before calling down or up */
4880 netdev->mtu = new_mtu;
4882 if (netif_running(netdev))
4883 ixgbe_reinit_locked(adapter);
4889 * ixgbe_open - Called when a network interface is made active
4890 * @netdev: network interface device structure
4892 * Returns 0 on success, negative value on failure
4894 * The open entry point is called when a network interface is made
4895 * active by the system (IFF_UP). At this point all resources needed
4896 * for transmit and receive operations are allocated, the interrupt
4897 * handler is registered with the OS, the watchdog timer is started,
4898 * and the stack is notified that the interface is ready.
4900 static int ixgbe_open(struct net_device *netdev)
4902 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4905 /* disallow open during test */
4906 if (test_bit(__IXGBE_TESTING, &adapter->state))
4909 netif_carrier_off(netdev);
4911 /* allocate transmit descriptors */
4912 err = ixgbe_setup_all_tx_resources(adapter);
4916 /* allocate receive descriptors */
4917 err = ixgbe_setup_all_rx_resources(adapter);
4921 ixgbe_configure(adapter);
4923 err = ixgbe_request_irq(adapter);
4927 /* Notify the stack of the actual queue counts. */
4928 err = netif_set_real_num_tx_queues(netdev,
4929 adapter->num_rx_pools > 1 ? 1 :
4930 adapter->num_tx_queues);
4932 goto err_set_queues;
4935 err = netif_set_real_num_rx_queues(netdev,
4936 adapter->num_rx_pools > 1 ? 1 :
4937 adapter->num_rx_queues);
4939 goto err_set_queues;
4941 ixgbe_ptp_init(adapter);
4943 ixgbe_up_complete(adapter);
4948 ixgbe_free_irq(adapter);
4950 ixgbe_free_all_rx_resources(adapter);
4952 ixgbe_free_all_tx_resources(adapter);
4954 ixgbe_reset(adapter);
4960 * ixgbe_close - Disables a network interface
4961 * @netdev: network interface device structure
4963 * Returns 0, this is not allowed to fail
4965 * The close entry point is called when an interface is de-activated
4966 * by the OS. The hardware is still under the drivers control, but
4967 * needs to be disabled. A global MAC reset is issued to stop the
4968 * hardware, and all transmit and receive resources are freed.
4970 static int ixgbe_close(struct net_device *netdev)
4972 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4974 ixgbe_ptp_stop(adapter);
4976 ixgbe_down(adapter);
4977 ixgbe_free_irq(adapter);
4979 ixgbe_fdir_filter_exit(adapter);
4981 ixgbe_free_all_tx_resources(adapter);
4982 ixgbe_free_all_rx_resources(adapter);
4984 ixgbe_release_hw_control(adapter);
4990 static int ixgbe_resume(struct pci_dev *pdev)
4992 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
4993 struct net_device *netdev = adapter->netdev;
4996 pci_set_power_state(pdev, PCI_D0);
4997 pci_restore_state(pdev);
4999 * pci_restore_state clears dev->state_saved so call
5000 * pci_save_state to restore it.
5002 pci_save_state(pdev);
5004 err = pci_enable_device_mem(pdev);
5006 e_dev_err("Cannot enable PCI device from suspend\n");
5009 pci_set_master(pdev);
5011 pci_wake_from_d3(pdev, false);
5013 ixgbe_reset(adapter);
5015 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5018 err = ixgbe_init_interrupt_scheme(adapter);
5019 if (!err && netif_running(netdev))
5020 err = ixgbe_open(netdev);
5027 netif_device_attach(netdev);
5031 #endif /* CONFIG_PM */
5033 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5035 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5036 struct net_device *netdev = adapter->netdev;
5037 struct ixgbe_hw *hw = &adapter->hw;
5039 u32 wufc = adapter->wol;
5044 netif_device_detach(netdev);
5046 if (netif_running(netdev)) {
5048 ixgbe_down(adapter);
5049 ixgbe_free_irq(adapter);
5050 ixgbe_free_all_tx_resources(adapter);
5051 ixgbe_free_all_rx_resources(adapter);
5055 ixgbe_clear_interrupt_scheme(adapter);
5058 retval = pci_save_state(pdev);
5064 ixgbe_set_rx_mode(netdev);
5066 /* enable the optics for 82599 SFP+ fiber as we can WoL */
5067 if (hw->mac.ops.enable_tx_laser)
5068 hw->mac.ops.enable_tx_laser(hw);
5070 /* turn on all-multi mode if wake on multicast is enabled */
5071 if (wufc & IXGBE_WUFC_MC) {
5072 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5073 fctrl |= IXGBE_FCTRL_MPE;
5074 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5077 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5078 ctrl |= IXGBE_CTRL_GIO_DIS;
5079 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5081 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5083 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5084 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5087 switch (hw->mac.type) {
5088 case ixgbe_mac_82598EB:
5089 pci_wake_from_d3(pdev, false);
5091 case ixgbe_mac_82599EB:
5092 case ixgbe_mac_X540:
5093 pci_wake_from_d3(pdev, !!wufc);
5099 *enable_wake = !!wufc;
5101 ixgbe_release_hw_control(adapter);
5103 pci_disable_device(pdev);
5109 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5114 retval = __ixgbe_shutdown(pdev, &wake);
5119 pci_prepare_to_sleep(pdev);
5121 pci_wake_from_d3(pdev, false);
5122 pci_set_power_state(pdev, PCI_D3hot);
5127 #endif /* CONFIG_PM */
5129 static void ixgbe_shutdown(struct pci_dev *pdev)
5133 __ixgbe_shutdown(pdev, &wake);
5135 if (system_state == SYSTEM_POWER_OFF) {
5136 pci_wake_from_d3(pdev, wake);
5137 pci_set_power_state(pdev, PCI_D3hot);
5142 * ixgbe_update_stats - Update the board statistics counters.
5143 * @adapter: board private structure
5145 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5147 struct net_device *netdev = adapter->netdev;
5148 struct ixgbe_hw *hw = &adapter->hw;
5149 struct ixgbe_hw_stats *hwstats = &adapter->stats;
5151 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5152 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
5153 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
5154 u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
5156 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5157 test_bit(__IXGBE_RESETTING, &adapter->state))
5160 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5163 for (i = 0; i < adapter->num_rx_queues; i++) {
5164 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
5165 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
5167 adapter->rsc_total_count = rsc_count;
5168 adapter->rsc_total_flush = rsc_flush;
5171 for (i = 0; i < adapter->num_rx_queues; i++) {
5172 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
5173 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
5174 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
5175 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
5176 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
5177 bytes += rx_ring->stats.bytes;
5178 packets += rx_ring->stats.packets;
5180 adapter->non_eop_descs = non_eop_descs;
5181 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5182 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5183 adapter->hw_csum_rx_error = hw_csum_rx_error;
5184 netdev->stats.rx_bytes = bytes;
5185 netdev->stats.rx_packets = packets;
5189 /* gather some stats to the adapter struct that are per queue */
5190 for (i = 0; i < adapter->num_tx_queues; i++) {
5191 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5192 restart_queue += tx_ring->tx_stats.restart_queue;
5193 tx_busy += tx_ring->tx_stats.tx_busy;
5194 bytes += tx_ring->stats.bytes;
5195 packets += tx_ring->stats.packets;
5197 adapter->restart_queue = restart_queue;
5198 adapter->tx_busy = tx_busy;
5199 netdev->stats.tx_bytes = bytes;
5200 netdev->stats.tx_packets = packets;
5202 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5204 /* 8 register reads */
5205 for (i = 0; i < 8; i++) {
5206 /* for packet buffers not used, the register should read 0 */
5207 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5209 hwstats->mpc[i] += mpc;
5210 total_mpc += hwstats->mpc[i];
5211 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5212 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5213 switch (hw->mac.type) {
5214 case ixgbe_mac_82598EB:
5215 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5216 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5217 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5218 hwstats->pxonrxc[i] +=
5219 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5221 case ixgbe_mac_82599EB:
5222 case ixgbe_mac_X540:
5223 hwstats->pxonrxc[i] +=
5224 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5231 /*16 register reads */
5232 for (i = 0; i < 16; i++) {
5233 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5234 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5235 if ((hw->mac.type == ixgbe_mac_82599EB) ||
5236 (hw->mac.type == ixgbe_mac_X540)) {
5237 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
5238 IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
5239 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
5240 IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
5244 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5245 /* work around hardware counting issue */
5246 hwstats->gprc -= missed_rx;
5248 ixgbe_update_xoff_received(adapter);
5250 /* 82598 hardware only has a 32 bit counter in the high register */
5251 switch (hw->mac.type) {
5252 case ixgbe_mac_82598EB:
5253 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5254 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5255 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5256 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5258 case ixgbe_mac_X540:
5259 /* OS2BMC stats are X540 only*/
5260 hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
5261 hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
5262 hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
5263 hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
5264 case ixgbe_mac_82599EB:
5265 for (i = 0; i < 16; i++)
5266 adapter->hw_rx_no_dma_resources +=
5267 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5268 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5269 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
5270 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5271 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
5272 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5273 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5274 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5275 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5276 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5278 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5279 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5280 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5281 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5282 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5283 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5284 /* Add up per cpu counters for total ddp aloc fail */
5285 if (adapter->fcoe.ddp_pool) {
5286 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
5287 struct ixgbe_fcoe_ddp_pool *ddp_pool;
5289 u64 noddp = 0, noddp_ext_buff = 0;
5290 for_each_possible_cpu(cpu) {
5291 ddp_pool = per_cpu_ptr(fcoe->ddp_pool, cpu);
5292 noddp += ddp_pool->noddp;
5293 noddp_ext_buff += ddp_pool->noddp_ext_buff;
5295 hwstats->fcoe_noddp = noddp;
5296 hwstats->fcoe_noddp_ext_buff = noddp_ext_buff;
5298 #endif /* IXGBE_FCOE */
5303 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5304 hwstats->bprc += bprc;
5305 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5306 if (hw->mac.type == ixgbe_mac_82598EB)
5307 hwstats->mprc -= bprc;
5308 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5309 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5310 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5311 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5312 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5313 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5314 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5315 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5316 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5317 hwstats->lxontxc += lxon;
5318 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5319 hwstats->lxofftxc += lxoff;
5320 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5321 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5323 * 82598 errata - tx of flow control packets is included in tx counters
5325 xon_off_tot = lxon + lxoff;
5326 hwstats->gptc -= xon_off_tot;
5327 hwstats->mptc -= xon_off_tot;
5328 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5329 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5330 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5331 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5332 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5333 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5334 hwstats->ptc64 -= xon_off_tot;
5335 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5336 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5337 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5338 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5339 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5340 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5342 /* Fill out the OS statistics structure */
5343 netdev->stats.multicast = hwstats->mprc;
5346 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
5347 netdev->stats.rx_dropped = 0;
5348 netdev->stats.rx_length_errors = hwstats->rlec;
5349 netdev->stats.rx_crc_errors = hwstats->crcerrs;
5350 netdev->stats.rx_missed_errors = total_mpc;
5354 * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
5355 * @adapter: pointer to the device adapter structure
5357 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
5359 struct ixgbe_hw *hw = &adapter->hw;
5362 if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
5365 adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
5367 /* if interface is down do nothing */
5368 if (test_bit(__IXGBE_DOWN, &adapter->state))
5371 /* do nothing if we are not using signature filters */
5372 if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
5375 adapter->fdir_overflow++;
5377 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5378 for (i = 0; i < adapter->num_tx_queues; i++)
5379 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
5380 &(adapter->tx_ring[i]->state));
5381 /* re-enable flow director interrupts */
5382 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
5384 e_err(probe, "failed to finish FDIR re-initialization, "
5385 "ignored adding FDIR ATR filters\n");
5390 * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
5391 * @adapter: pointer to the device adapter structure
5393 * This function serves two purposes. First it strobes the interrupt lines
5394 * in order to make certain interrupts are occurring. Secondly it sets the
5395 * bits needed to check for TX hangs. As a result we should immediately
5396 * determine if a hang has occurred.
5398 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
5400 struct ixgbe_hw *hw = &adapter->hw;
5404 /* If we're down or resetting, just bail */
5405 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5406 test_bit(__IXGBE_RESETTING, &adapter->state))
5409 /* Force detection of hung controller */
5410 if (netif_carrier_ok(adapter->netdev)) {
5411 for (i = 0; i < adapter->num_tx_queues; i++)
5412 set_check_for_tx_hang(adapter->tx_ring[i]);
5415 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5417 * for legacy and MSI interrupts don't set any bits
5418 * that are enabled for EIAM, because this operation
5419 * would set *both* EIMS and EICS for any bit in EIAM
5421 IXGBE_WRITE_REG(hw, IXGBE_EICS,
5422 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5424 /* get one bit for every active tx/rx interrupt vector */
5425 for (i = 0; i < adapter->num_q_vectors; i++) {
5426 struct ixgbe_q_vector *qv = adapter->q_vector[i];
5427 if (qv->rx.ring || qv->tx.ring)
5428 eics |= ((u64)1 << i);
5432 /* Cause software interrupt to ensure rings are cleaned */
5433 ixgbe_irq_rearm_queues(adapter, eics);
5438 * ixgbe_watchdog_update_link - update the link status
5439 * @adapter: pointer to the device adapter structure
5440 * @link_speed: pointer to a u32 to store the link_speed
5442 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
5444 struct ixgbe_hw *hw = &adapter->hw;
5445 u32 link_speed = adapter->link_speed;
5446 bool link_up = adapter->link_up;
5447 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
5449 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
5452 if (hw->mac.ops.check_link) {
5453 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5455 /* always assume link is up, if no check link function */
5456 link_speed = IXGBE_LINK_SPEED_10GB_FULL;
5460 if (adapter->ixgbe_ieee_pfc)
5461 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
5463 if (link_up && !((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && pfc_en)) {
5464 hw->mac.ops.fc_enable(hw);
5465 ixgbe_set_rx_drop_en(adapter);
5469 time_after(jiffies, (adapter->link_check_timeout +
5470 IXGBE_TRY_LINK_TIMEOUT))) {
5471 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5472 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5473 IXGBE_WRITE_FLUSH(hw);
5476 adapter->link_up = link_up;
5477 adapter->link_speed = link_speed;
5480 static void ixgbe_update_default_up(struct ixgbe_adapter *adapter)
5482 #ifdef CONFIG_IXGBE_DCB
5483 struct net_device *netdev = adapter->netdev;
5484 struct dcb_app app = {
5485 .selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE,
5490 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_IEEE)
5491 up = dcb_ieee_getapp_mask(netdev, &app);
5493 adapter->default_up = (up > 1) ? (ffs(up) - 1) : 0;
5498 * ixgbe_watchdog_link_is_up - update netif_carrier status and
5499 * print link up message
5500 * @adapter: pointer to the device adapter structure
5502 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
5504 struct net_device *netdev = adapter->netdev;
5505 struct ixgbe_hw *hw = &adapter->hw;
5506 u32 link_speed = adapter->link_speed;
5507 bool flow_rx, flow_tx;
5509 /* only continue if link was previously down */
5510 if (netif_carrier_ok(netdev))
5513 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
5515 switch (hw->mac.type) {
5516 case ixgbe_mac_82598EB: {
5517 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5518 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5519 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5520 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5523 case ixgbe_mac_X540:
5524 case ixgbe_mac_82599EB: {
5525 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5526 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5527 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5528 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5537 if (adapter->flags2 & IXGBE_FLAG2_PTP_ENABLED)
5538 ixgbe_ptp_start_cyclecounter(adapter);
5540 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
5541 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5543 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5545 (link_speed == IXGBE_LINK_SPEED_100_FULL ?
5548 ((flow_rx && flow_tx) ? "RX/TX" :
5550 (flow_tx ? "TX" : "None"))));
5552 netif_carrier_on(netdev);
5553 ixgbe_check_vf_rate_limit(adapter);
5555 /* update the default user priority for VFs */
5556 ixgbe_update_default_up(adapter);
5558 /* ping all the active vfs to let them know link has changed */
5559 ixgbe_ping_all_vfs(adapter);
5563 * ixgbe_watchdog_link_is_down - update netif_carrier status and
5564 * print link down message
5565 * @adapter: pointer to the adapter structure
5567 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
5569 struct net_device *netdev = adapter->netdev;
5570 struct ixgbe_hw *hw = &adapter->hw;
5572 adapter->link_up = false;
5573 adapter->link_speed = 0;
5575 /* only continue if link was up previously */
5576 if (!netif_carrier_ok(netdev))
5579 /* poll for SFP+ cable when link is down */
5580 if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
5581 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5583 if (adapter->flags2 & IXGBE_FLAG2_PTP_ENABLED)
5584 ixgbe_ptp_start_cyclecounter(adapter);
5586 e_info(drv, "NIC Link is Down\n");
5587 netif_carrier_off(netdev);
5589 /* ping all the active vfs to let them know link has changed */
5590 ixgbe_ping_all_vfs(adapter);
5594 * ixgbe_watchdog_flush_tx - flush queues on link down
5595 * @adapter: pointer to the device adapter structure
5597 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
5600 int some_tx_pending = 0;
5602 if (!netif_carrier_ok(adapter->netdev)) {
5603 for (i = 0; i < adapter->num_tx_queues; i++) {
5604 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5605 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5606 some_tx_pending = 1;
5611 if (some_tx_pending) {
5612 /* We've lost link, so the controller stops DMA,
5613 * but we've got queued Tx work that's never going
5614 * to get done, so reset controller to flush Tx.
5615 * (Do the reset outside of interrupt context).
5617 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
5622 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
5626 /* Do not perform spoof check for 82598 or if not in IOV mode */
5627 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
5628 adapter->num_vfs == 0)
5631 ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
5634 * ssvpc register is cleared on read, if zero then no
5635 * spoofed packets in the last interval.
5640 e_warn(drv, "%u Spoofed packets detected\n", ssvpc);
5644 * ixgbe_watchdog_subtask - check and bring link up
5645 * @adapter: pointer to the device adapter structure
5647 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
5649 /* if interface is down do nothing */
5650 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5651 test_bit(__IXGBE_RESETTING, &adapter->state))
5654 ixgbe_watchdog_update_link(adapter);
5656 if (adapter->link_up)
5657 ixgbe_watchdog_link_is_up(adapter);
5659 ixgbe_watchdog_link_is_down(adapter);
5661 ixgbe_spoof_check(adapter);
5662 ixgbe_update_stats(adapter);
5664 ixgbe_watchdog_flush_tx(adapter);
5668 * ixgbe_sfp_detection_subtask - poll for SFP+ cable
5669 * @adapter: the ixgbe adapter structure
5671 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
5673 struct ixgbe_hw *hw = &adapter->hw;
5676 /* not searching for SFP so there is nothing to do here */
5677 if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
5678 !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
5681 /* someone else is in init, wait until next service event */
5682 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5685 err = hw->phy.ops.identify_sfp(hw);
5686 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
5689 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
5690 /* If no cable is present, then we need to reset
5691 * the next time we find a good cable. */
5692 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
5699 /* exit if reset not needed */
5700 if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
5703 adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
5706 * A module may be identified correctly, but the EEPROM may not have
5707 * support for that module. setup_sfp() will fail in that case, so
5708 * we should not allow that module to load.
5710 if (hw->mac.type == ixgbe_mac_82598EB)
5711 err = hw->phy.ops.reset(hw);
5713 err = hw->mac.ops.setup_sfp(hw);
5715 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
5718 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
5719 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
5722 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5724 if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
5725 (adapter->netdev->reg_state == NETREG_REGISTERED)) {
5726 e_dev_err("failed to initialize because an unsupported "
5727 "SFP+ module type was detected.\n");
5728 e_dev_err("Reload the driver after installing a "
5729 "supported module.\n");
5730 unregister_netdev(adapter->netdev);
5735 * ixgbe_sfp_link_config_subtask - set up link SFP after module install
5736 * @adapter: the ixgbe adapter structure
5738 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
5740 struct ixgbe_hw *hw = &adapter->hw;
5744 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
5747 /* someone else is in init, wait until next service event */
5748 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5751 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
5753 autoneg = hw->phy.autoneg_advertised;
5754 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
5755 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
5756 if (hw->mac.ops.setup_link)
5757 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
5759 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5760 adapter->link_check_timeout = jiffies;
5761 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5764 #ifdef CONFIG_PCI_IOV
5765 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
5768 struct ixgbe_hw *hw = &adapter->hw;
5769 struct net_device *netdev = adapter->netdev;
5773 gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
5774 if (gpc) /* If incrementing then no need for the check below */
5777 * Check to see if a bad DMA write target from an errant or
5778 * malicious VF has caused a PCIe error. If so then we can
5779 * issue a VFLR to the offending VF(s) and then resume without
5780 * requesting a full slot reset.
5783 for (vf = 0; vf < adapter->num_vfs; vf++) {
5784 ciaa = (vf << 16) | 0x80000000;
5785 /* 32 bit read so align, we really want status at offset 6 */
5786 ciaa |= PCI_COMMAND;
5787 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5788 ciad = IXGBE_READ_REG(hw, IXGBE_CIAD_82599);
5790 /* disable debug mode asap after reading data */
5791 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5792 /* Get the upper 16 bits which will be the PCI status reg */
5794 if (ciad & PCI_STATUS_REC_MASTER_ABORT) {
5795 netdev_err(netdev, "VF %d Hung DMA\n", vf);
5797 ciaa = (vf << 16) | 0x80000000;
5799 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5800 ciad = 0x00008000; /* VFLR */
5801 IXGBE_WRITE_REG(hw, IXGBE_CIAD_82599, ciad);
5803 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5810 * ixgbe_service_timer - Timer Call-back
5811 * @data: pointer to adapter cast into an unsigned long
5813 static void ixgbe_service_timer(unsigned long data)
5815 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5816 unsigned long next_event_offset;
5819 /* poll faster when waiting for link */
5820 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
5821 next_event_offset = HZ / 10;
5823 next_event_offset = HZ * 2;
5825 #ifdef CONFIG_PCI_IOV
5827 * don't bother with SR-IOV VF DMA hang check if there are
5828 * no VFs or the link is down
5830 if (!adapter->num_vfs ||
5831 (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
5832 goto normal_timer_service;
5834 /* If we have VFs allocated then we must check for DMA hangs */
5835 ixgbe_check_for_bad_vf(adapter);
5836 next_event_offset = HZ / 50;
5837 adapter->timer_event_accumulator++;
5839 if (adapter->timer_event_accumulator >= 100)
5840 adapter->timer_event_accumulator = 0;
5844 normal_timer_service:
5846 /* Reset the timer */
5847 mod_timer(&adapter->service_timer, next_event_offset + jiffies);
5850 ixgbe_service_event_schedule(adapter);
5853 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
5855 if (!(adapter->flags2 & IXGBE_FLAG2_RESET_REQUESTED))
5858 adapter->flags2 &= ~IXGBE_FLAG2_RESET_REQUESTED;
5860 /* If we're already down or resetting, just bail */
5861 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5862 test_bit(__IXGBE_RESETTING, &adapter->state))
5865 ixgbe_dump(adapter);
5866 netdev_err(adapter->netdev, "Reset adapter\n");
5867 adapter->tx_timeout_count++;
5869 ixgbe_reinit_locked(adapter);
5873 * ixgbe_service_task - manages and runs subtasks
5874 * @work: pointer to work_struct containing our data
5876 static void ixgbe_service_task(struct work_struct *work)
5878 struct ixgbe_adapter *adapter = container_of(work,
5879 struct ixgbe_adapter,
5882 ixgbe_reset_subtask(adapter);
5883 ixgbe_sfp_detection_subtask(adapter);
5884 ixgbe_sfp_link_config_subtask(adapter);
5885 ixgbe_check_overtemp_subtask(adapter);
5886 ixgbe_watchdog_subtask(adapter);
5887 ixgbe_fdir_reinit_subtask(adapter);
5888 ixgbe_check_hang_subtask(adapter);
5889 ixgbe_ptp_overflow_check(adapter);
5891 ixgbe_service_event_complete(adapter);
5894 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
5895 struct ixgbe_tx_buffer *first,
5898 struct sk_buff *skb = first->skb;
5899 u32 vlan_macip_lens, type_tucmd;
5900 u32 mss_l4len_idx, l4len;
5902 if (skb->ip_summed != CHECKSUM_PARTIAL)
5905 if (!skb_is_gso(skb))
5908 if (skb_header_cloned(skb)) {
5909 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5914 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
5915 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
5917 if (first->protocol == __constant_htons(ETH_P_IP)) {
5918 struct iphdr *iph = ip_hdr(skb);
5921 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
5925 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
5926 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
5927 IXGBE_TX_FLAGS_CSUM |
5928 IXGBE_TX_FLAGS_IPV4;
5929 } else if (skb_is_gso_v6(skb)) {
5930 ipv6_hdr(skb)->payload_len = 0;
5931 tcp_hdr(skb)->check =
5932 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5933 &ipv6_hdr(skb)->daddr,
5935 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
5936 IXGBE_TX_FLAGS_CSUM;
5939 /* compute header lengths */
5940 l4len = tcp_hdrlen(skb);
5941 *hdr_len = skb_transport_offset(skb) + l4len;
5943 /* update gso size and bytecount with header size */
5944 first->gso_segs = skb_shinfo(skb)->gso_segs;
5945 first->bytecount += (first->gso_segs - 1) * *hdr_len;
5947 /* mss_l4len_id: use 0 as index for TSO */
5948 mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
5949 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
5951 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
5952 vlan_macip_lens = skb_network_header_len(skb);
5953 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
5954 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
5956 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0, type_tucmd,
5962 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
5963 struct ixgbe_tx_buffer *first)
5965 struct sk_buff *skb = first->skb;
5966 u32 vlan_macip_lens = 0;
5967 u32 mss_l4len_idx = 0;
5970 if (skb->ip_summed != CHECKSUM_PARTIAL) {
5971 if (!(first->tx_flags & IXGBE_TX_FLAGS_HW_VLAN) &&
5972 !(first->tx_flags & IXGBE_TX_FLAGS_CC))
5976 switch (first->protocol) {
5977 case __constant_htons(ETH_P_IP):
5978 vlan_macip_lens |= skb_network_header_len(skb);
5979 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
5980 l4_hdr = ip_hdr(skb)->protocol;
5982 case __constant_htons(ETH_P_IPV6):
5983 vlan_macip_lens |= skb_network_header_len(skb);
5984 l4_hdr = ipv6_hdr(skb)->nexthdr;
5987 if (unlikely(net_ratelimit())) {
5988 dev_warn(tx_ring->dev,
5989 "partial checksum but proto=%x!\n",
5997 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5998 mss_l4len_idx = tcp_hdrlen(skb) <<
5999 IXGBE_ADVTXD_L4LEN_SHIFT;
6002 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
6003 mss_l4len_idx = sizeof(struct sctphdr) <<
6004 IXGBE_ADVTXD_L4LEN_SHIFT;
6007 mss_l4len_idx = sizeof(struct udphdr) <<
6008 IXGBE_ADVTXD_L4LEN_SHIFT;
6011 if (unlikely(net_ratelimit())) {
6012 dev_warn(tx_ring->dev,
6013 "partial checksum but l4 proto=%x!\n",
6019 /* update TX checksum flag */
6020 first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
6023 /* vlan_macip_lens: MACLEN, VLAN tag */
6024 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
6025 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
6027 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0,
6028 type_tucmd, mss_l4len_idx);
6031 #define IXGBE_SET_FLAG(_input, _flag, _result) \
6032 ((_flag <= _result) ? \
6033 ((u32)(_input & _flag) * (_result / _flag)) : \
6034 ((u32)(_input & _flag) / (_flag / _result)))
6036 static u32 ixgbe_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
6038 /* set type for advanced descriptor with frame checksum insertion */
6039 u32 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
6040 IXGBE_ADVTXD_DCMD_DEXT |
6041 IXGBE_ADVTXD_DCMD_IFCS;
6043 /* set HW vlan bit if vlan is present */
6044 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_HW_VLAN,
6045 IXGBE_ADVTXD_DCMD_VLE);
6047 /* set segmentation enable bits for TSO/FSO */
6048 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSO,
6049 IXGBE_ADVTXD_DCMD_TSE);
6051 /* set timestamp bit if present */
6052 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSTAMP,
6053 IXGBE_ADVTXD_MAC_TSTAMP);
6055 /* insert frame checksum */
6056 cmd_type ^= IXGBE_SET_FLAG(skb->no_fcs, 1, IXGBE_ADVTXD_DCMD_IFCS);
6061 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
6062 u32 tx_flags, unsigned int paylen)
6064 u32 olinfo_status = paylen << IXGBE_ADVTXD_PAYLEN_SHIFT;
6066 /* enable L4 checksum for TSO and TX checksum offload */
6067 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
6068 IXGBE_TX_FLAGS_CSUM,
6069 IXGBE_ADVTXD_POPTS_TXSM);
6071 /* enble IPv4 checksum for TSO */
6072 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
6073 IXGBE_TX_FLAGS_IPV4,
6074 IXGBE_ADVTXD_POPTS_IXSM);
6077 * Check Context must be set if Tx switch is enabled, which it
6078 * always is for case where virtual functions are running
6080 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
6084 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
6087 #define IXGBE_TXD_CMD (IXGBE_TXD_CMD_EOP | \
6090 static void ixgbe_tx_map(struct ixgbe_ring *tx_ring,
6091 struct ixgbe_tx_buffer *first,
6094 struct sk_buff *skb = first->skb;
6095 struct ixgbe_tx_buffer *tx_buffer;
6096 union ixgbe_adv_tx_desc *tx_desc;
6097 struct skb_frag_struct *frag;
6099 unsigned int data_len, size;
6100 u32 tx_flags = first->tx_flags;
6101 u32 cmd_type = ixgbe_tx_cmd_type(skb, tx_flags);
6102 u16 i = tx_ring->next_to_use;
6104 tx_desc = IXGBE_TX_DESC(tx_ring, i);
6106 ixgbe_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
6108 size = skb_headlen(skb);
6109 data_len = skb->data_len;
6112 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6113 if (data_len < sizeof(struct fcoe_crc_eof)) {
6114 size -= sizeof(struct fcoe_crc_eof) - data_len;
6117 data_len -= sizeof(struct fcoe_crc_eof);
6122 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
6126 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
6127 if (dma_mapping_error(tx_ring->dev, dma))
6130 /* record length, and DMA address */
6131 dma_unmap_len_set(tx_buffer, len, size);
6132 dma_unmap_addr_set(tx_buffer, dma, dma);
6134 tx_desc->read.buffer_addr = cpu_to_le64(dma);
6136 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
6137 tx_desc->read.cmd_type_len =
6138 cpu_to_le32(cmd_type ^ IXGBE_MAX_DATA_PER_TXD);
6142 if (i == tx_ring->count) {
6143 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
6146 tx_desc->read.olinfo_status = 0;
6148 dma += IXGBE_MAX_DATA_PER_TXD;
6149 size -= IXGBE_MAX_DATA_PER_TXD;
6151 tx_desc->read.buffer_addr = cpu_to_le64(dma);
6154 if (likely(!data_len))
6157 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
6161 if (i == tx_ring->count) {
6162 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
6165 tx_desc->read.olinfo_status = 0;
6168 size = min_t(unsigned int, data_len, skb_frag_size(frag));
6170 size = skb_frag_size(frag);
6174 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
6177 tx_buffer = &tx_ring->tx_buffer_info[i];
6180 /* write last descriptor with RS and EOP bits */
6181 cmd_type |= size | IXGBE_TXD_CMD;
6182 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
6184 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
6186 /* set the timestamp */
6187 first->time_stamp = jiffies;
6190 * Force memory writes to complete before letting h/w know there
6191 * are new descriptors to fetch. (Only applicable for weak-ordered
6192 * memory model archs, such as IA-64).
6194 * We also need this memory barrier to make certain all of the
6195 * status bits have been updated before next_to_watch is written.
6199 /* set next_to_watch value indicating a packet is present */
6200 first->next_to_watch = tx_desc;
6203 if (i == tx_ring->count)
6206 tx_ring->next_to_use = i;
6208 /* notify HW of packet */
6209 writel(i, tx_ring->tail);
6213 dev_err(tx_ring->dev, "TX DMA map failed\n");
6215 /* clear dma mappings for failed tx_buffer_info map */
6217 tx_buffer = &tx_ring->tx_buffer_info[i];
6218 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer);
6219 if (tx_buffer == first)
6226 tx_ring->next_to_use = i;
6229 static void ixgbe_atr(struct ixgbe_ring *ring,
6230 struct ixgbe_tx_buffer *first)
6232 struct ixgbe_q_vector *q_vector = ring->q_vector;
6233 union ixgbe_atr_hash_dword input = { .dword = 0 };
6234 union ixgbe_atr_hash_dword common = { .dword = 0 };
6236 unsigned char *network;
6238 struct ipv6hdr *ipv6;
6243 /* if ring doesn't have a interrupt vector, cannot perform ATR */
6247 /* do nothing if sampling is disabled */
6248 if (!ring->atr_sample_rate)
6253 /* snag network header to get L4 type and address */
6254 hdr.network = skb_network_header(first->skb);
6256 /* Currently only IPv4/IPv6 with TCP is supported */
6257 if ((first->protocol != __constant_htons(ETH_P_IPV6) ||
6258 hdr.ipv6->nexthdr != IPPROTO_TCP) &&
6259 (first->protocol != __constant_htons(ETH_P_IP) ||
6260 hdr.ipv4->protocol != IPPROTO_TCP))
6263 th = tcp_hdr(first->skb);
6265 /* skip this packet since it is invalid or the socket is closing */
6269 /* sample on all syn packets or once every atr sample count */
6270 if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
6273 /* reset sample count */
6274 ring->atr_count = 0;
6276 vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
6279 * src and dst are inverted, think how the receiver sees them
6281 * The input is broken into two sections, a non-compressed section
6282 * containing vm_pool, vlan_id, and flow_type. The rest of the data
6283 * is XORed together and stored in the compressed dword.
6285 input.formatted.vlan_id = vlan_id;
6288 * since src port and flex bytes occupy the same word XOR them together
6289 * and write the value to source port portion of compressed dword
6291 if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
6292 common.port.src ^= th->dest ^ __constant_htons(ETH_P_8021Q);
6294 common.port.src ^= th->dest ^ first->protocol;
6295 common.port.dst ^= th->source;
6297 if (first->protocol == __constant_htons(ETH_P_IP)) {
6298 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
6299 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
6301 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
6302 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
6303 hdr.ipv6->saddr.s6_addr32[1] ^
6304 hdr.ipv6->saddr.s6_addr32[2] ^
6305 hdr.ipv6->saddr.s6_addr32[3] ^
6306 hdr.ipv6->daddr.s6_addr32[0] ^
6307 hdr.ipv6->daddr.s6_addr32[1] ^
6308 hdr.ipv6->daddr.s6_addr32[2] ^
6309 hdr.ipv6->daddr.s6_addr32[3];
6312 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
6313 ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
6314 input, common, ring->queue_index);
6317 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6319 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
6320 /* Herbert's original patch had:
6321 * smp_mb__after_netif_stop_queue();
6322 * but since that doesn't exist yet, just open code it. */
6325 /* We need to check again in a case another CPU has just
6326 * made room available. */
6327 if (likely(ixgbe_desc_unused(tx_ring) < size))
6330 /* A reprieve! - use start_queue because it doesn't call schedule */
6331 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
6332 ++tx_ring->tx_stats.restart_queue;
6336 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6338 if (likely(ixgbe_desc_unused(tx_ring) >= size))
6340 return __ixgbe_maybe_stop_tx(tx_ring, size);
6343 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6345 struct ixgbe_adapter *adapter = netdev_priv(dev);
6346 int txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
6349 __be16 protocol = vlan_get_protocol(skb);
6351 if (((protocol == htons(ETH_P_FCOE)) ||
6352 (protocol == htons(ETH_P_FIP))) &&
6353 (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)) {
6354 struct ixgbe_ring_feature *f;
6356 f = &adapter->ring_feature[RING_F_FCOE];
6358 while (txq >= f->indices)
6360 txq += adapter->ring_feature[RING_F_FCOE].offset;
6366 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6367 while (unlikely(txq >= dev->real_num_tx_queues))
6368 txq -= dev->real_num_tx_queues;
6372 return skb_tx_hash(dev, skb);
6375 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
6376 struct ixgbe_adapter *adapter,
6377 struct ixgbe_ring *tx_ring)
6379 struct ixgbe_tx_buffer *first;
6382 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6385 u16 count = TXD_USE_COUNT(skb_headlen(skb));
6386 __be16 protocol = skb->protocol;
6390 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
6391 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
6392 * + 2 desc gap to keep tail from touching head,
6393 * + 1 desc for context descriptor,
6394 * otherwise try next time
6396 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6397 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6398 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6400 count += skb_shinfo(skb)->nr_frags;
6402 if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
6403 tx_ring->tx_stats.tx_busy++;
6404 return NETDEV_TX_BUSY;
6407 /* record the location of the first descriptor for this packet */
6408 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
6410 first->bytecount = skb->len;
6411 first->gso_segs = 1;
6413 /* if we have a HW VLAN tag being added default to the HW one */
6414 if (vlan_tx_tag_present(skb)) {
6415 tx_flags |= vlan_tx_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
6416 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6417 /* else if it is a SW VLAN check the next protocol and store the tag */
6418 } else if (protocol == __constant_htons(ETH_P_8021Q)) {
6419 struct vlan_hdr *vhdr, _vhdr;
6420 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
6424 protocol = vhdr->h_vlan_encapsulated_proto;
6425 tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
6426 IXGBE_TX_FLAGS_VLAN_SHIFT;
6427 tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
6430 skb_tx_timestamp(skb);
6432 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
6433 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
6434 tx_flags |= IXGBE_TX_FLAGS_TSTAMP;
6437 #ifdef CONFIG_PCI_IOV
6439 * Use the l2switch_enable flag - would be false if the DMA
6440 * Tx switch had been disabled.
6442 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6443 tx_flags |= IXGBE_TX_FLAGS_CC;
6446 /* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
6447 if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
6448 ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
6449 (skb->priority != TC_PRIO_CONTROL))) {
6450 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6451 tx_flags |= (skb->priority & 0x7) <<
6452 IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
6453 if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
6454 struct vlan_ethhdr *vhdr;
6455 if (skb_header_cloned(skb) &&
6456 pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
6458 vhdr = (struct vlan_ethhdr *)skb->data;
6459 vhdr->h_vlan_TCI = htons(tx_flags >>
6460 IXGBE_TX_FLAGS_VLAN_SHIFT);
6462 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6466 /* record initial flags and protocol */
6467 first->tx_flags = tx_flags;
6468 first->protocol = protocol;
6471 /* setup tx offload for FCoE */
6472 if ((protocol == __constant_htons(ETH_P_FCOE)) &&
6473 (tx_ring->netdev->features & (NETIF_F_FSO | NETIF_F_FCOE_CRC))) {
6474 tso = ixgbe_fso(tx_ring, first, &hdr_len);
6481 #endif /* IXGBE_FCOE */
6482 tso = ixgbe_tso(tx_ring, first, &hdr_len);
6486 ixgbe_tx_csum(tx_ring, first);
6488 /* add the ATR filter if ATR is on */
6489 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
6490 ixgbe_atr(tx_ring, first);
6494 #endif /* IXGBE_FCOE */
6495 ixgbe_tx_map(tx_ring, first, hdr_len);
6497 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
6499 return NETDEV_TX_OK;
6502 dev_kfree_skb_any(first->skb);
6505 return NETDEV_TX_OK;
6508 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
6509 struct net_device *netdev)
6511 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6512 struct ixgbe_ring *tx_ring;
6515 * The minimum packet size for olinfo paylen is 17 so pad the skb
6516 * in order to meet this minimum size requirement.
6518 if (unlikely(skb->len < 17)) {
6519 if (skb_pad(skb, 17 - skb->len))
6520 return NETDEV_TX_OK;
6522 skb_set_tail_pointer(skb, 17);
6525 tx_ring = adapter->tx_ring[skb->queue_mapping];
6526 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
6530 * ixgbe_set_mac - Change the Ethernet Address of the NIC
6531 * @netdev: network interface device structure
6532 * @p: pointer to an address structure
6534 * Returns 0 on success, negative on failure
6536 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6538 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6539 struct ixgbe_hw *hw = &adapter->hw;
6540 struct sockaddr *addr = p;
6542 if (!is_valid_ether_addr(addr->sa_data))
6543 return -EADDRNOTAVAIL;
6545 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6546 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6548 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, VMDQ_P(0), IXGBE_RAH_AV);
6554 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6556 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6557 struct ixgbe_hw *hw = &adapter->hw;
6561 if (prtad != hw->phy.mdio.prtad)
6563 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6569 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6570 u16 addr, u16 value)
6572 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6573 struct ixgbe_hw *hw = &adapter->hw;
6575 if (prtad != hw->phy.mdio.prtad)
6577 return hw->phy.ops.write_reg(hw, addr, devad, value);
6580 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6582 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6586 return ixgbe_ptp_hwtstamp_ioctl(adapter, req, cmd);
6588 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6593 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6595 * @netdev: network interface device structure
6597 * Returns non-zero on failure
6599 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6602 struct ixgbe_adapter *adapter = netdev_priv(dev);
6603 struct ixgbe_hw *hw = &adapter->hw;
6605 if (is_valid_ether_addr(hw->mac.san_addr)) {
6607 err = dev_addr_add(dev, hw->mac.san_addr, NETDEV_HW_ADDR_T_SAN);
6610 /* update SAN MAC vmdq pool selection */
6611 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
6617 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6619 * @netdev: network interface device structure
6621 * Returns non-zero on failure
6623 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6626 struct ixgbe_adapter *adapter = netdev_priv(dev);
6627 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6629 if (is_valid_ether_addr(mac->san_addr)) {
6631 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6637 #ifdef CONFIG_NET_POLL_CONTROLLER
6639 * Polling 'interrupt' - used by things like netconsole to send skbs
6640 * without having to re-enable interrupts. It's not called while
6641 * the interrupt routine is executing.
6643 static void ixgbe_netpoll(struct net_device *netdev)
6645 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6648 /* if interface is down do nothing */
6649 if (test_bit(__IXGBE_DOWN, &adapter->state))
6652 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
6653 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
6654 for (i = 0; i < adapter->num_q_vectors; i++)
6655 ixgbe_msix_clean_rings(0, adapter->q_vector[i]);
6657 ixgbe_intr(adapter->pdev->irq, netdev);
6659 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
6663 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
6664 struct rtnl_link_stats64 *stats)
6666 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6670 for (i = 0; i < adapter->num_rx_queues; i++) {
6671 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
6677 start = u64_stats_fetch_begin_bh(&ring->syncp);
6678 packets = ring->stats.packets;
6679 bytes = ring->stats.bytes;
6680 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6681 stats->rx_packets += packets;
6682 stats->rx_bytes += bytes;
6686 for (i = 0; i < adapter->num_tx_queues; i++) {
6687 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->tx_ring[i]);
6693 start = u64_stats_fetch_begin_bh(&ring->syncp);
6694 packets = ring->stats.packets;
6695 bytes = ring->stats.bytes;
6696 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6697 stats->tx_packets += packets;
6698 stats->tx_bytes += bytes;
6702 /* following stats updated by ixgbe_watchdog_task() */
6703 stats->multicast = netdev->stats.multicast;
6704 stats->rx_errors = netdev->stats.rx_errors;
6705 stats->rx_length_errors = netdev->stats.rx_length_errors;
6706 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
6707 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
6711 #ifdef CONFIG_IXGBE_DCB
6713 * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
6714 * @adapter: pointer to ixgbe_adapter
6715 * @tc: number of traffic classes currently enabled
6717 * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
6718 * 802.1Q priority maps to a packet buffer that exists.
6720 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
6722 struct ixgbe_hw *hw = &adapter->hw;
6726 /* 82598 have a static priority to TC mapping that can not
6727 * be changed so no validation is needed.
6729 if (hw->mac.type == ixgbe_mac_82598EB)
6732 reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
6735 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
6736 u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
6738 /* If up2tc is out of bounds default to zero */
6740 reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
6744 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
6750 * ixgbe_set_prio_tc_map - Configure netdev prio tc map
6751 * @adapter: Pointer to adapter struct
6753 * Populate the netdev user priority to tc map
6755 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter *adapter)
6757 struct net_device *dev = adapter->netdev;
6758 struct ixgbe_dcb_config *dcb_cfg = &adapter->dcb_cfg;
6759 struct ieee_ets *ets = adapter->ixgbe_ieee_ets;
6762 for (prio = 0; prio < MAX_USER_PRIORITY; prio++) {
6765 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE)
6766 tc = ixgbe_dcb_get_tc_from_up(dcb_cfg, 0, prio);
6768 tc = ets->prio_tc[prio];
6770 netdev_set_prio_tc_map(dev, prio, tc);
6775 * ixgbe_setup_tc - configure net_device for multiple traffic classes
6777 * @netdev: net device to configure
6778 * @tc: number of traffic classes to enable
6780 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
6782 struct ixgbe_adapter *adapter = netdev_priv(dev);
6783 struct ixgbe_hw *hw = &adapter->hw;
6785 /* Hardware supports up to 8 traffic classes */
6786 if (tc > adapter->dcb_cfg.num_tcs.pg_tcs ||
6787 (hw->mac.type == ixgbe_mac_82598EB &&
6788 tc < MAX_TRAFFIC_CLASS))
6791 /* Hardware has to reinitialize queues and interrupts to
6792 * match packet buffer alignment. Unfortunately, the
6793 * hardware is not flexible enough to do this dynamically.
6795 if (netif_running(dev))
6797 ixgbe_clear_interrupt_scheme(adapter);
6800 netdev_set_num_tc(dev, tc);
6801 ixgbe_set_prio_tc_map(adapter);
6803 adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
6805 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
6806 adapter->last_lfc_mode = adapter->hw.fc.requested_mode;
6807 adapter->hw.fc.requested_mode = ixgbe_fc_none;
6810 netdev_reset_tc(dev);
6812 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
6813 adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
6815 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
6817 adapter->temp_dcb_cfg.pfc_mode_enable = false;
6818 adapter->dcb_cfg.pfc_mode_enable = false;
6821 ixgbe_init_interrupt_scheme(adapter);
6822 ixgbe_validate_rtr(adapter, tc);
6823 if (netif_running(dev))
6829 #endif /* CONFIG_IXGBE_DCB */
6830 void ixgbe_do_reset(struct net_device *netdev)
6832 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6834 if (netif_running(netdev))
6835 ixgbe_reinit_locked(adapter);
6837 ixgbe_reset(adapter);
6840 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
6841 netdev_features_t features)
6843 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6845 /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
6846 if (!(features & NETIF_F_RXCSUM))
6847 features &= ~NETIF_F_LRO;
6849 /* Turn off LRO if not RSC capable */
6850 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
6851 features &= ~NETIF_F_LRO;
6856 static int ixgbe_set_features(struct net_device *netdev,
6857 netdev_features_t features)
6859 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6860 netdev_features_t changed = netdev->features ^ features;
6861 bool need_reset = false;
6863 /* Make sure RSC matches LRO, reset if change */
6864 if (!(features & NETIF_F_LRO)) {
6865 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
6867 adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
6868 } else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
6869 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
6870 if (adapter->rx_itr_setting == 1 ||
6871 adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
6872 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
6874 } else if ((changed ^ features) & NETIF_F_LRO) {
6875 e_info(probe, "rx-usecs set too low, "
6881 * Check if Flow Director n-tuple support was enabled or disabled. If
6882 * the state changed, we need to reset.
6884 switch (features & NETIF_F_NTUPLE) {
6885 case NETIF_F_NTUPLE:
6886 /* turn off ATR, enable perfect filters and reset */
6887 if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
6890 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
6891 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
6894 /* turn off perfect filters, enable ATR and reset */
6895 if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
6898 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
6900 /* We cannot enable ATR if SR-IOV is enabled */
6901 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6904 /* We cannot enable ATR if we have 2 or more traffic classes */
6905 if (netdev_get_num_tc(netdev) > 1)
6908 /* We cannot enable ATR if RSS is disabled */
6909 if (adapter->ring_feature[RING_F_RSS].limit <= 1)
6912 /* A sample rate of 0 indicates ATR disabled */
6913 if (!adapter->atr_sample_rate)
6916 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
6920 if (features & NETIF_F_HW_VLAN_RX)
6921 ixgbe_vlan_strip_enable(adapter);
6923 ixgbe_vlan_strip_disable(adapter);
6925 if (changed & NETIF_F_RXALL)
6928 netdev->features = features;
6930 ixgbe_do_reset(netdev);
6935 static int ixgbe_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
6936 struct net_device *dev,
6937 const unsigned char *addr,
6940 struct ixgbe_adapter *adapter = netdev_priv(dev);
6943 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
6946 /* Hardware does not support aging addresses so if a
6947 * ndm_state is given only allow permanent addresses
6949 if (ndm->ndm_state && !(ndm->ndm_state & NUD_PERMANENT)) {
6950 pr_info("%s: FDB only supports static addresses\n",
6955 if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
6956 u32 rar_uc_entries = IXGBE_MAX_PF_MACVLANS;
6958 if (netdev_uc_count(dev) < rar_uc_entries)
6959 err = dev_uc_add_excl(dev, addr);
6962 } else if (is_multicast_ether_addr(addr)) {
6963 err = dev_mc_add_excl(dev, addr);
6968 /* Only return duplicate errors if NLM_F_EXCL is set */
6969 if (err == -EEXIST && !(flags & NLM_F_EXCL))
6975 static int ixgbe_ndo_fdb_del(struct ndmsg *ndm,
6976 struct net_device *dev,
6977 const unsigned char *addr)
6979 struct ixgbe_adapter *adapter = netdev_priv(dev);
6980 int err = -EOPNOTSUPP;
6982 if (ndm->ndm_state & NUD_PERMANENT) {
6983 pr_info("%s: FDB only supports static addresses\n",
6988 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
6989 if (is_unicast_ether_addr(addr))
6990 err = dev_uc_del(dev, addr);
6991 else if (is_multicast_ether_addr(addr))
6992 err = dev_mc_del(dev, addr);
7000 static int ixgbe_ndo_fdb_dump(struct sk_buff *skb,
7001 struct netlink_callback *cb,
7002 struct net_device *dev,
7005 struct ixgbe_adapter *adapter = netdev_priv(dev);
7007 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7008 idx = ndo_dflt_fdb_dump(skb, cb, dev, idx);
7013 static int ixgbe_ndo_bridge_setlink(struct net_device *dev,
7014 struct nlmsghdr *nlh)
7016 struct ixgbe_adapter *adapter = netdev_priv(dev);
7017 struct nlattr *attr, *br_spec;
7020 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
7023 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
7025 nla_for_each_nested(attr, br_spec, rem) {
7029 if (nla_type(attr) != IFLA_BRIDGE_MODE)
7032 mode = nla_get_u16(attr);
7033 if (mode == BRIDGE_MODE_VEPA) {
7035 adapter->flags2 &= ~IXGBE_FLAG2_BRIDGE_MODE_VEB;
7036 } else if (mode == BRIDGE_MODE_VEB) {
7037 reg = IXGBE_PFDTXGSWC_VT_LBEN;
7038 adapter->flags2 |= IXGBE_FLAG2_BRIDGE_MODE_VEB;
7042 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC, reg);
7044 e_info(drv, "enabling bridge mode: %s\n",
7045 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
7051 static int ixgbe_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
7052 struct net_device *dev)
7054 struct ixgbe_adapter *adapter = netdev_priv(dev);
7057 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
7060 if (adapter->flags2 & IXGBE_FLAG2_BRIDGE_MODE_VEB)
7061 mode = BRIDGE_MODE_VEB;
7063 mode = BRIDGE_MODE_VEPA;
7065 return ndo_dflt_bridge_getlink(skb, pid, seq, dev, mode);
7068 static const struct net_device_ops ixgbe_netdev_ops = {
7069 .ndo_open = ixgbe_open,
7070 .ndo_stop = ixgbe_close,
7071 .ndo_start_xmit = ixgbe_xmit_frame,
7072 .ndo_select_queue = ixgbe_select_queue,
7073 .ndo_set_rx_mode = ixgbe_set_rx_mode,
7074 .ndo_validate_addr = eth_validate_addr,
7075 .ndo_set_mac_address = ixgbe_set_mac,
7076 .ndo_change_mtu = ixgbe_change_mtu,
7077 .ndo_tx_timeout = ixgbe_tx_timeout,
7078 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
7079 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
7080 .ndo_do_ioctl = ixgbe_ioctl,
7081 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
7082 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
7083 .ndo_set_vf_tx_rate = ixgbe_ndo_set_vf_bw,
7084 .ndo_set_vf_spoofchk = ixgbe_ndo_set_vf_spoofchk,
7085 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
7086 .ndo_get_stats64 = ixgbe_get_stats64,
7087 #ifdef CONFIG_IXGBE_DCB
7088 .ndo_setup_tc = ixgbe_setup_tc,
7090 #ifdef CONFIG_NET_POLL_CONTROLLER
7091 .ndo_poll_controller = ixgbe_netpoll,
7094 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
7095 .ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
7096 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
7097 .ndo_fcoe_enable = ixgbe_fcoe_enable,
7098 .ndo_fcoe_disable = ixgbe_fcoe_disable,
7099 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
7100 .ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
7101 #endif /* IXGBE_FCOE */
7102 .ndo_set_features = ixgbe_set_features,
7103 .ndo_fix_features = ixgbe_fix_features,
7104 .ndo_fdb_add = ixgbe_ndo_fdb_add,
7105 .ndo_fdb_del = ixgbe_ndo_fdb_del,
7106 .ndo_fdb_dump = ixgbe_ndo_fdb_dump,
7107 .ndo_bridge_setlink = ixgbe_ndo_bridge_setlink,
7108 .ndo_bridge_getlink = ixgbe_ndo_bridge_getlink,
7112 * ixgbe_wol_supported - Check whether device supports WoL
7113 * @hw: hw specific details
7114 * @device_id: the device ID
7115 * @subdev_id: the subsystem device ID
7117 * This function is used by probe and ethtool to determine
7118 * which devices have WoL support
7121 int ixgbe_wol_supported(struct ixgbe_adapter *adapter, u16 device_id,
7124 struct ixgbe_hw *hw = &adapter->hw;
7125 u16 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
7126 int is_wol_supported = 0;
7128 switch (device_id) {
7129 case IXGBE_DEV_ID_82599_SFP:
7130 /* Only these subdevices could supports WOL */
7131 switch (subdevice_id) {
7132 case IXGBE_SUBDEV_ID_82599_560FLR:
7133 /* only support first port */
7134 if (hw->bus.func != 0)
7136 case IXGBE_SUBDEV_ID_82599_SFP:
7137 case IXGBE_SUBDEV_ID_82599_RNDC:
7138 case IXGBE_SUBDEV_ID_82599_ECNA_DP:
7139 is_wol_supported = 1;
7143 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
7144 /* All except this subdevice support WOL */
7145 if (subdevice_id != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
7146 is_wol_supported = 1;
7148 case IXGBE_DEV_ID_82599_KX4:
7149 is_wol_supported = 1;
7151 case IXGBE_DEV_ID_X540T:
7152 case IXGBE_DEV_ID_X540T1:
7153 /* check eeprom to see if enabled wol */
7154 if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
7155 ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
7156 (hw->bus.func == 0))) {
7157 is_wol_supported = 1;
7162 return is_wol_supported;
7166 * ixgbe_probe - Device Initialization Routine
7167 * @pdev: PCI device information struct
7168 * @ent: entry in ixgbe_pci_tbl
7170 * Returns 0 on success, negative on failure
7172 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
7173 * The OS initialization, configuring of the adapter private structure,
7174 * and a hardware reset occur.
7176 static int ixgbe_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
7178 struct net_device *netdev;
7179 struct ixgbe_adapter *adapter = NULL;
7180 struct ixgbe_hw *hw;
7181 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
7182 static int cards_found;
7183 int i, err, pci_using_dac;
7184 u8 part_str[IXGBE_PBANUM_LENGTH];
7185 unsigned int indices = num_possible_cpus();
7186 unsigned int dcb_max = 0;
7192 /* Catch broken hardware that put the wrong VF device ID in
7193 * the PCIe SR-IOV capability.
7195 if (pdev->is_virtfn) {
7196 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
7197 pci_name(pdev), pdev->vendor, pdev->device);
7201 err = pci_enable_device_mem(pdev);
7205 if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
7206 !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
7209 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
7211 err = dma_set_coherent_mask(&pdev->dev,
7215 "No usable DMA configuration, aborting\n");
7222 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
7223 IORESOURCE_MEM), ixgbe_driver_name);
7226 "pci_request_selected_regions failed 0x%x\n", err);
7230 pci_enable_pcie_error_reporting(pdev);
7232 pci_set_master(pdev);
7233 pci_save_state(pdev);
7235 #ifdef CONFIG_IXGBE_DCB
7236 if (ii->mac == ixgbe_mac_82598EB)
7237 dcb_max = min_t(unsigned int, indices * MAX_TRAFFIC_CLASS,
7238 IXGBE_MAX_RSS_INDICES);
7240 dcb_max = min_t(unsigned int, indices * MAX_TRAFFIC_CLASS,
7241 IXGBE_MAX_FDIR_INDICES);
7244 if (ii->mac == ixgbe_mac_82598EB)
7245 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
7247 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
7250 indices += min_t(unsigned int, num_possible_cpus(),
7251 IXGBE_MAX_FCOE_INDICES);
7253 indices = max_t(unsigned int, dcb_max, indices);
7254 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
7257 goto err_alloc_etherdev;
7260 SET_NETDEV_DEV(netdev, &pdev->dev);
7262 adapter = netdev_priv(netdev);
7263 pci_set_drvdata(pdev, adapter);
7265 adapter->netdev = netdev;
7266 adapter->pdev = pdev;
7269 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
7271 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
7272 pci_resource_len(pdev, 0));
7278 netdev->netdev_ops = &ixgbe_netdev_ops;
7279 ixgbe_set_ethtool_ops(netdev);
7280 netdev->watchdog_timeo = 5 * HZ;
7281 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
7283 adapter->bd_number = cards_found;
7286 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
7287 hw->mac.type = ii->mac;
7290 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
7291 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
7292 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
7293 if (!(eec & (1 << 8)))
7294 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
7297 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
7298 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
7299 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
7300 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
7301 hw->phy.mdio.mmds = 0;
7302 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
7303 hw->phy.mdio.dev = netdev;
7304 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
7305 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
7307 ii->get_invariants(hw);
7309 /* setup the private structure */
7310 err = ixgbe_sw_init(adapter);
7314 /* Make it possible the adapter to be woken up via WOL */
7315 switch (adapter->hw.mac.type) {
7316 case ixgbe_mac_82599EB:
7317 case ixgbe_mac_X540:
7318 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7325 * If there is a fan on this device and it has failed log the
7328 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
7329 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
7330 if (esdp & IXGBE_ESDP_SDP1)
7331 e_crit(probe, "Fan has stopped, replace the adapter\n");
7334 if (allow_unsupported_sfp)
7335 hw->allow_unsupported_sfp = allow_unsupported_sfp;
7337 /* reset_hw fills in the perm_addr as well */
7338 hw->phy.reset_if_overtemp = true;
7339 err = hw->mac.ops.reset_hw(hw);
7340 hw->phy.reset_if_overtemp = false;
7341 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
7342 hw->mac.type == ixgbe_mac_82598EB) {
7344 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
7345 e_dev_err("failed to load because an unsupported SFP+ "
7346 "module type was detected.\n");
7347 e_dev_err("Reload the driver after installing a supported "
7351 e_dev_err("HW Init failed: %d\n", err);
7355 #ifdef CONFIG_PCI_IOV
7356 ixgbe_enable_sriov(adapter, ii);
7359 netdev->features = NETIF_F_SG |
7362 NETIF_F_HW_VLAN_TX |
7363 NETIF_F_HW_VLAN_RX |
7364 NETIF_F_HW_VLAN_FILTER |
7370 netdev->hw_features = netdev->features;
7372 switch (adapter->hw.mac.type) {
7373 case ixgbe_mac_82599EB:
7374 case ixgbe_mac_X540:
7375 netdev->features |= NETIF_F_SCTP_CSUM;
7376 netdev->hw_features |= NETIF_F_SCTP_CSUM |
7383 netdev->hw_features |= NETIF_F_RXALL;
7385 netdev->vlan_features |= NETIF_F_TSO;
7386 netdev->vlan_features |= NETIF_F_TSO6;
7387 netdev->vlan_features |= NETIF_F_IP_CSUM;
7388 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
7389 netdev->vlan_features |= NETIF_F_SG;
7391 netdev->priv_flags |= IFF_UNICAST_FLT;
7392 netdev->priv_flags |= IFF_SUPP_NOFCS;
7394 #ifdef CONFIG_IXGBE_DCB
7395 netdev->dcbnl_ops = &dcbnl_ops;
7399 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7400 if (hw->mac.ops.get_device_caps) {
7401 hw->mac.ops.get_device_caps(hw, &device_caps);
7402 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
7403 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
7406 adapter->ring_feature[RING_F_FCOE].limit = IXGBE_FCRETA_SIZE;
7408 netdev->features |= NETIF_F_FSO |
7411 netdev->vlan_features |= NETIF_F_FSO |
7415 #endif /* IXGBE_FCOE */
7416 if (pci_using_dac) {
7417 netdev->features |= NETIF_F_HIGHDMA;
7418 netdev->vlan_features |= NETIF_F_HIGHDMA;
7421 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
7422 netdev->hw_features |= NETIF_F_LRO;
7423 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
7424 netdev->features |= NETIF_F_LRO;
7426 /* make sure the EEPROM is good */
7427 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
7428 e_dev_err("The EEPROM Checksum Is Not Valid\n");
7433 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
7435 if (!is_valid_ether_addr(netdev->dev_addr)) {
7436 e_dev_err("invalid MAC address\n");
7441 setup_timer(&adapter->service_timer, &ixgbe_service_timer,
7442 (unsigned long) adapter);
7444 INIT_WORK(&adapter->service_task, ixgbe_service_task);
7445 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
7447 err = ixgbe_init_interrupt_scheme(adapter);
7451 /* WOL not supported for all devices */
7453 hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
7454 if (ixgbe_wol_supported(adapter, pdev->device, pdev->subsystem_device))
7455 adapter->wol = IXGBE_WUFC_MAG;
7457 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
7459 /* save off EEPROM version number */
7460 hw->eeprom.ops.read(hw, 0x2e, &adapter->eeprom_verh);
7461 hw->eeprom.ops.read(hw, 0x2d, &adapter->eeprom_verl);
7463 /* pick up the PCI bus settings for reporting later */
7464 hw->mac.ops.get_bus_info(hw);
7466 /* print bus type/speed/width info */
7467 e_dev_info("(PCI Express:%s:%s) %pM\n",
7468 (hw->bus.speed == ixgbe_bus_speed_5000 ? "5.0GT/s" :
7469 hw->bus.speed == ixgbe_bus_speed_2500 ? "2.5GT/s" :
7471 (hw->bus.width == ixgbe_bus_width_pcie_x8 ? "Width x8" :
7472 hw->bus.width == ixgbe_bus_width_pcie_x4 ? "Width x4" :
7473 hw->bus.width == ixgbe_bus_width_pcie_x1 ? "Width x1" :
7477 err = ixgbe_read_pba_string_generic(hw, part_str, IXGBE_PBANUM_LENGTH);
7479 strncpy(part_str, "Unknown", IXGBE_PBANUM_LENGTH);
7480 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
7481 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
7482 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
7485 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
7486 hw->mac.type, hw->phy.type, part_str);
7488 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
7489 e_dev_warn("PCI-Express bandwidth available for this card is "
7490 "not sufficient for optimal performance.\n");
7491 e_dev_warn("For optimal performance a x8 PCI-Express slot "
7495 /* reset the hardware with the new settings */
7496 err = hw->mac.ops.start_hw(hw);
7497 if (err == IXGBE_ERR_EEPROM_VERSION) {
7498 /* We are running on a pre-production device, log a warning */
7499 e_dev_warn("This device is a pre-production adapter/LOM. "
7500 "Please be aware there may be issues associated "
7501 "with your hardware. If you are experiencing "
7502 "problems please contact your Intel or hardware "
7503 "representative who provided you with this "
7506 strcpy(netdev->name, "eth%d");
7507 err = register_netdev(netdev);
7511 /* power down the optics for 82599 SFP+ fiber */
7512 if (hw->mac.ops.disable_tx_laser)
7513 hw->mac.ops.disable_tx_laser(hw);
7515 /* carrier off reporting is important to ethtool even BEFORE open */
7516 netif_carrier_off(netdev);
7518 #ifdef CONFIG_IXGBE_DCA
7519 if (dca_add_requester(&pdev->dev) == 0) {
7520 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
7521 ixgbe_setup_dca(adapter);
7524 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7525 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
7526 for (i = 0; i < adapter->num_vfs; i++)
7527 ixgbe_vf_configuration(pdev, (i | 0x10000000));
7530 /* firmware requires driver version to be 0xFFFFFFFF
7531 * since os does not support feature
7533 if (hw->mac.ops.set_fw_drv_ver)
7534 hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF,
7537 /* add san mac addr to netdev */
7538 ixgbe_add_sanmac_netdev(netdev);
7540 e_dev_info("%s\n", ixgbe_default_device_descr);
7543 #ifdef CONFIG_IXGBE_HWMON
7544 if (ixgbe_sysfs_init(adapter))
7545 e_err(probe, "failed to allocate sysfs resources\n");
7546 #endif /* CONFIG_IXGBE_HWMON */
7548 #ifdef CONFIG_DEBUG_FS
7549 ixgbe_dbg_adapter_init(adapter);
7550 #endif /* CONFIG_DEBUG_FS */
7555 ixgbe_release_hw_control(adapter);
7556 ixgbe_clear_interrupt_scheme(adapter);
7558 ixgbe_disable_sriov(adapter);
7559 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
7560 iounmap(hw->hw_addr);
7562 free_netdev(netdev);
7564 pci_release_selected_regions(pdev,
7565 pci_select_bars(pdev, IORESOURCE_MEM));
7568 pci_disable_device(pdev);
7573 * ixgbe_remove - Device Removal Routine
7574 * @pdev: PCI device information struct
7576 * ixgbe_remove is called by the PCI subsystem to alert the driver
7577 * that it should release a PCI device. The could be caused by a
7578 * Hot-Plug event, or because the driver is going to be removed from
7581 static void ixgbe_remove(struct pci_dev *pdev)
7583 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7584 struct net_device *netdev = adapter->netdev;
7586 #ifdef CONFIG_DEBUG_FS
7587 ixgbe_dbg_adapter_exit(adapter);
7588 #endif /*CONFIG_DEBUG_FS */
7590 set_bit(__IXGBE_DOWN, &adapter->state);
7591 cancel_work_sync(&adapter->service_task);
7594 #ifdef CONFIG_IXGBE_DCA
7595 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
7596 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
7597 dca_remove_requester(&pdev->dev);
7598 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
7602 #ifdef CONFIG_IXGBE_HWMON
7603 ixgbe_sysfs_exit(adapter);
7604 #endif /* CONFIG_IXGBE_HWMON */
7606 /* remove the added san mac */
7607 ixgbe_del_sanmac_netdev(netdev);
7609 if (netdev->reg_state == NETREG_REGISTERED)
7610 unregister_netdev(netdev);
7612 ixgbe_disable_sriov(adapter);
7614 ixgbe_clear_interrupt_scheme(adapter);
7616 ixgbe_release_hw_control(adapter);
7619 kfree(adapter->ixgbe_ieee_pfc);
7620 kfree(adapter->ixgbe_ieee_ets);
7623 iounmap(adapter->hw.hw_addr);
7624 pci_release_selected_regions(pdev, pci_select_bars(pdev,
7627 e_dev_info("complete\n");
7629 free_netdev(netdev);
7631 pci_disable_pcie_error_reporting(pdev);
7633 pci_disable_device(pdev);
7637 * ixgbe_io_error_detected - called when PCI error is detected
7638 * @pdev: Pointer to PCI device
7639 * @state: The current pci connection state
7641 * This function is called after a PCI bus error affecting
7642 * this device has been detected.
7644 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
7645 pci_channel_state_t state)
7647 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7648 struct net_device *netdev = adapter->netdev;
7650 #ifdef CONFIG_PCI_IOV
7651 struct pci_dev *bdev, *vfdev;
7652 u32 dw0, dw1, dw2, dw3;
7654 u16 req_id, pf_func;
7656 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
7657 adapter->num_vfs == 0)
7658 goto skip_bad_vf_detection;
7660 bdev = pdev->bus->self;
7661 while (bdev && (pci_pcie_type(bdev) != PCI_EXP_TYPE_ROOT_PORT))
7662 bdev = bdev->bus->self;
7665 goto skip_bad_vf_detection;
7667 pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
7669 goto skip_bad_vf_detection;
7671 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG, &dw0);
7672 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 4, &dw1);
7673 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 8, &dw2);
7674 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 12, &dw3);
7677 /* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
7678 if (!(req_id & 0x0080))
7679 goto skip_bad_vf_detection;
7681 pf_func = req_id & 0x01;
7682 if ((pf_func & 1) == (pdev->devfn & 1)) {
7683 unsigned int device_id;
7685 vf = (req_id & 0x7F) >> 1;
7686 e_dev_err("VF %d has caused a PCIe error\n", vf);
7687 e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
7688 "%8.8x\tdw3: %8.8x\n",
7689 dw0, dw1, dw2, dw3);
7690 switch (adapter->hw.mac.type) {
7691 case ixgbe_mac_82599EB:
7692 device_id = IXGBE_82599_VF_DEVICE_ID;
7694 case ixgbe_mac_X540:
7695 device_id = IXGBE_X540_VF_DEVICE_ID;
7702 /* Find the pci device of the offending VF */
7703 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, device_id, NULL);
7705 if (vfdev->devfn == (req_id & 0xFF))
7707 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL,
7711 * There's a slim chance the VF could have been hot plugged,
7712 * so if it is no longer present we don't need to issue the
7713 * VFLR. Just clean up the AER in that case.
7716 e_dev_err("Issuing VFLR to VF %d\n", vf);
7717 pci_write_config_dword(vfdev, 0xA8, 0x00008000);
7720 pci_cleanup_aer_uncorrect_error_status(pdev);
7724 * Even though the error may have occurred on the other port
7725 * we still need to increment the vf error reference count for
7726 * both ports because the I/O resume function will be called
7729 adapter->vferr_refcount++;
7731 return PCI_ERS_RESULT_RECOVERED;
7733 skip_bad_vf_detection:
7734 #endif /* CONFIG_PCI_IOV */
7735 netif_device_detach(netdev);
7737 if (state == pci_channel_io_perm_failure)
7738 return PCI_ERS_RESULT_DISCONNECT;
7740 if (netif_running(netdev))
7741 ixgbe_down(adapter);
7742 pci_disable_device(pdev);
7744 /* Request a slot reset. */
7745 return PCI_ERS_RESULT_NEED_RESET;
7749 * ixgbe_io_slot_reset - called after the pci bus has been reset.
7750 * @pdev: Pointer to PCI device
7752 * Restart the card from scratch, as if from a cold-boot.
7754 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7756 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7757 pci_ers_result_t result;
7760 if (pci_enable_device_mem(pdev)) {
7761 e_err(probe, "Cannot re-enable PCI device after reset.\n");
7762 result = PCI_ERS_RESULT_DISCONNECT;
7764 pci_set_master(pdev);
7765 pci_restore_state(pdev);
7766 pci_save_state(pdev);
7768 pci_wake_from_d3(pdev, false);
7770 ixgbe_reset(adapter);
7771 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7772 result = PCI_ERS_RESULT_RECOVERED;
7775 err = pci_cleanup_aer_uncorrect_error_status(pdev);
7777 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7778 "failed 0x%0x\n", err);
7779 /* non-fatal, continue */
7786 * ixgbe_io_resume - called when traffic can start flowing again.
7787 * @pdev: Pointer to PCI device
7789 * This callback is called when the error recovery driver tells us that
7790 * its OK to resume normal operation.
7792 static void ixgbe_io_resume(struct pci_dev *pdev)
7794 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7795 struct net_device *netdev = adapter->netdev;
7797 #ifdef CONFIG_PCI_IOV
7798 if (adapter->vferr_refcount) {
7799 e_info(drv, "Resuming after VF err\n");
7800 adapter->vferr_refcount--;
7805 if (netif_running(netdev))
7808 netif_device_attach(netdev);
7811 static const struct pci_error_handlers ixgbe_err_handler = {
7812 .error_detected = ixgbe_io_error_detected,
7813 .slot_reset = ixgbe_io_slot_reset,
7814 .resume = ixgbe_io_resume,
7817 static struct pci_driver ixgbe_driver = {
7818 .name = ixgbe_driver_name,
7819 .id_table = ixgbe_pci_tbl,
7820 .probe = ixgbe_probe,
7821 .remove = ixgbe_remove,
7823 .suspend = ixgbe_suspend,
7824 .resume = ixgbe_resume,
7826 .shutdown = ixgbe_shutdown,
7827 .err_handler = &ixgbe_err_handler
7831 * ixgbe_init_module - Driver Registration Routine
7833 * ixgbe_init_module is the first routine called when the driver is
7834 * loaded. All it does is register with the PCI subsystem.
7836 static int __init ixgbe_init_module(void)
7839 pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
7840 pr_info("%s\n", ixgbe_copyright);
7842 #ifdef CONFIG_DEBUG_FS
7844 #endif /* CONFIG_DEBUG_FS */
7846 #ifdef CONFIG_IXGBE_DCA
7847 dca_register_notify(&dca_notifier);
7850 ret = pci_register_driver(&ixgbe_driver);
7854 module_init(ixgbe_init_module);
7857 * ixgbe_exit_module - Driver Exit Cleanup Routine
7859 * ixgbe_exit_module is called just before the driver is removed
7862 static void __exit ixgbe_exit_module(void)
7864 #ifdef CONFIG_IXGBE_DCA
7865 dca_unregister_notify(&dca_notifier);
7867 pci_unregister_driver(&ixgbe_driver);
7869 #ifdef CONFIG_DEBUG_FS
7871 #endif /* CONFIG_DEBUG_FS */
7873 rcu_barrier(); /* Wait for completion of call_rcu()'s */
7876 #ifdef CONFIG_IXGBE_DCA
7877 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
7882 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
7883 __ixgbe_notify_dca);
7885 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7888 #endif /* CONFIG_IXGBE_DCA */
7890 module_exit(ixgbe_exit_module);