1 /*******************************************************************************
3 Intel 10 Gigabit PCI Express Linux driver
4 Copyright(c) 1999 - 2010 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>
36 #include <linux/tcp.h>
37 #include <linux/pkt_sched.h>
38 #include <linux/ipv6.h>
39 #include <linux/slab.h>
40 #include <net/checksum.h>
41 #include <net/ip6_checksum.h>
42 #include <linux/ethtool.h>
43 #include <linux/if_vlan.h>
44 #include <scsi/fc/fc_fcoe.h>
47 #include "ixgbe_common.h"
48 #include "ixgbe_dcb_82599.h"
49 #include "ixgbe_sriov.h"
51 char ixgbe_driver_name[] = "ixgbe";
52 static const char ixgbe_driver_string[] =
53 "Intel(R) 10 Gigabit PCI Express Network Driver";
55 #define DRV_VERSION "2.0.84-k2"
56 const char ixgbe_driver_version[] = DRV_VERSION;
57 static char ixgbe_copyright[] = "Copyright (c) 1999-2010 Intel Corporation.";
59 static const struct ixgbe_info *ixgbe_info_tbl[] = {
60 [board_82598] = &ixgbe_82598_info,
61 [board_82599] = &ixgbe_82599_info,
64 /* ixgbe_pci_tbl - PCI Device ID Table
66 * Wildcard entries (PCI_ANY_ID) should come last
67 * Last entry must be all 0s
69 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
70 * Class, Class Mask, private data (not used) }
72 static DEFINE_PCI_DEVICE_TABLE(ixgbe_pci_tbl) = {
73 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598),
75 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT),
77 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT),
79 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT),
81 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2),
83 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4),
85 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT),
87 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT),
89 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM),
91 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR),
93 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM),
95 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX),
97 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4),
99 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM),
101 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR),
103 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP),
105 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM),
107 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ),
109 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4),
111 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM),
113 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE),
116 /* required last entry */
119 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
121 #ifdef CONFIG_IXGBE_DCA
122 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
124 static struct notifier_block dca_notifier = {
125 .notifier_call = ixgbe_notify_dca,
131 #ifdef CONFIG_PCI_IOV
132 static unsigned int max_vfs;
133 module_param(max_vfs, uint, 0);
134 MODULE_PARM_DESC(max_vfs,
135 "Maximum number of virtual functions to allocate per physical function");
136 #endif /* CONFIG_PCI_IOV */
138 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
139 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
140 MODULE_LICENSE("GPL");
141 MODULE_VERSION(DRV_VERSION);
143 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
145 static inline void ixgbe_disable_sriov(struct ixgbe_adapter *adapter)
147 struct ixgbe_hw *hw = &adapter->hw;
152 #ifdef CONFIG_PCI_IOV
153 /* disable iov and allow time for transactions to clear */
154 pci_disable_sriov(adapter->pdev);
157 /* turn off device IOV mode */
158 gcr = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
159 gcr &= ~(IXGBE_GCR_EXT_SRIOV);
160 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr);
161 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
162 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
163 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
165 /* set default pool back to 0 */
166 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
167 vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
168 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
170 /* take a breather then clean up driver data */
173 kfree(adapter->vfinfo);
174 adapter->vfinfo = NULL;
176 adapter->num_vfs = 0;
177 adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
180 struct ixgbe_reg_info {
185 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
187 /* General Registers */
188 {IXGBE_CTRL, "CTRL"},
189 {IXGBE_STATUS, "STATUS"},
190 {IXGBE_CTRL_EXT, "CTRL_EXT"},
192 /* Interrupt Registers */
193 {IXGBE_EICR, "EICR"},
196 {IXGBE_SRRCTL(0), "SRRCTL"},
197 {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
198 {IXGBE_RDLEN(0), "RDLEN"},
199 {IXGBE_RDH(0), "RDH"},
200 {IXGBE_RDT(0), "RDT"},
201 {IXGBE_RXDCTL(0), "RXDCTL"},
202 {IXGBE_RDBAL(0), "RDBAL"},
203 {IXGBE_RDBAH(0), "RDBAH"},
206 {IXGBE_TDBAL(0), "TDBAL"},
207 {IXGBE_TDBAH(0), "TDBAH"},
208 {IXGBE_TDLEN(0), "TDLEN"},
209 {IXGBE_TDH(0), "TDH"},
210 {IXGBE_TDT(0), "TDT"},
211 {IXGBE_TXDCTL(0), "TXDCTL"},
213 /* List Terminator */
219 * ixgbe_regdump - register printout routine
221 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
227 switch (reginfo->ofs) {
228 case IXGBE_SRRCTL(0):
229 for (i = 0; i < 64; i++)
230 regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
232 case IXGBE_DCA_RXCTRL(0):
233 for (i = 0; i < 64; i++)
234 regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
237 for (i = 0; i < 64; i++)
238 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
241 for (i = 0; i < 64; i++)
242 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
245 for (i = 0; i < 64; i++)
246 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
248 case IXGBE_RXDCTL(0):
249 for (i = 0; i < 64; i++)
250 regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
253 for (i = 0; i < 64; i++)
254 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
257 for (i = 0; i < 64; i++)
258 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
261 for (i = 0; i < 64; i++)
262 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
265 for (i = 0; i < 64; i++)
266 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
269 for (i = 0; i < 64; i++)
270 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
273 for (i = 0; i < 64; i++)
274 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
277 for (i = 0; i < 64; i++)
278 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
280 case IXGBE_TXDCTL(0):
281 for (i = 0; i < 64; i++)
282 regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
285 pr_info("%-15s %08x\n", reginfo->name,
286 IXGBE_READ_REG(hw, reginfo->ofs));
290 for (i = 0; i < 8; i++) {
291 snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i*8, i*8+7);
292 pr_err("%-15s", rname);
293 for (j = 0; j < 8; j++)
294 pr_cont(" %08x", regs[i*8+j]);
301 * ixgbe_dump - Print registers, tx-rings and rx-rings
303 static void ixgbe_dump(struct ixgbe_adapter *adapter)
305 struct net_device *netdev = adapter->netdev;
306 struct ixgbe_hw *hw = &adapter->hw;
307 struct ixgbe_reg_info *reginfo;
309 struct ixgbe_ring *tx_ring;
310 struct ixgbe_tx_buffer *tx_buffer_info;
311 union ixgbe_adv_tx_desc *tx_desc;
312 struct my_u0 { u64 a; u64 b; } *u0;
313 struct ixgbe_ring *rx_ring;
314 union ixgbe_adv_rx_desc *rx_desc;
315 struct ixgbe_rx_buffer *rx_buffer_info;
319 if (!netif_msg_hw(adapter))
322 /* Print netdevice Info */
324 dev_info(&adapter->pdev->dev, "Net device Info\n");
325 pr_info("Device Name state "
326 "trans_start last_rx\n");
327 pr_info("%-15s %016lX %016lX %016lX\n",
334 /* Print Registers */
335 dev_info(&adapter->pdev->dev, "Register Dump\n");
336 pr_info(" Register Name Value\n");
337 for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
338 reginfo->name; reginfo++) {
339 ixgbe_regdump(hw, reginfo);
342 /* Print TX Ring Summary */
343 if (!netdev || !netif_running(netdev))
346 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
347 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
348 for (n = 0; n < adapter->num_tx_queues; n++) {
349 tx_ring = adapter->tx_ring[n];
351 &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
352 pr_info(" %5d %5X %5X %016llX %04X %3X %016llX\n",
353 n, tx_ring->next_to_use, tx_ring->next_to_clean,
354 (u64)tx_buffer_info->dma,
355 tx_buffer_info->length,
356 tx_buffer_info->next_to_watch,
357 (u64)tx_buffer_info->time_stamp);
361 if (!netif_msg_tx_done(adapter))
362 goto rx_ring_summary;
364 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
366 /* Transmit Descriptor Formats
368 * Advanced Transmit Descriptor
369 * +--------------------------------------------------------------+
370 * 0 | Buffer Address [63:0] |
371 * +--------------------------------------------------------------+
372 * 8 | PAYLEN | PORTS | IDX | STA | DCMD |DTYP | RSV | DTALEN |
373 * +--------------------------------------------------------------+
374 * 63 46 45 40 39 36 35 32 31 24 23 20 19 0
377 for (n = 0; n < adapter->num_tx_queues; n++) {
378 tx_ring = adapter->tx_ring[n];
379 pr_info("------------------------------------\n");
380 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
381 pr_info("------------------------------------\n");
382 pr_info("T [desc] [address 63:0 ] "
383 "[PlPOIdStDDt Ln] [bi->dma ] "
384 "leng ntw timestamp bi->skb\n");
386 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
387 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
388 tx_buffer_info = &tx_ring->tx_buffer_info[i];
389 u0 = (struct my_u0 *)tx_desc;
390 pr_info("T [0x%03X] %016llX %016llX %016llX"
391 " %04X %3X %016llX %p", i,
394 (u64)tx_buffer_info->dma,
395 tx_buffer_info->length,
396 tx_buffer_info->next_to_watch,
397 (u64)tx_buffer_info->time_stamp,
398 tx_buffer_info->skb);
399 if (i == tx_ring->next_to_use &&
400 i == tx_ring->next_to_clean)
402 else if (i == tx_ring->next_to_use)
404 else if (i == tx_ring->next_to_clean)
409 if (netif_msg_pktdata(adapter) &&
410 tx_buffer_info->dma != 0)
411 print_hex_dump(KERN_INFO, "",
412 DUMP_PREFIX_ADDRESS, 16, 1,
413 phys_to_virt(tx_buffer_info->dma),
414 tx_buffer_info->length, true);
418 /* Print RX Rings Summary */
420 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
421 pr_info("Queue [NTU] [NTC]\n");
422 for (n = 0; n < adapter->num_rx_queues; n++) {
423 rx_ring = adapter->rx_ring[n];
424 pr_info("%5d %5X %5X\n",
425 n, rx_ring->next_to_use, rx_ring->next_to_clean);
429 if (!netif_msg_rx_status(adapter))
432 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
434 /* Advanced Receive Descriptor (Read) Format
436 * +-----------------------------------------------------+
437 * 0 | Packet Buffer Address [63:1] |A0/NSE|
438 * +----------------------------------------------+------+
439 * 8 | Header Buffer Address [63:1] | DD |
440 * +-----------------------------------------------------+
443 * Advanced Receive Descriptor (Write-Back) Format
445 * 63 48 47 32 31 30 21 20 16 15 4 3 0
446 * +------------------------------------------------------+
447 * 0 | Packet IP |SPH| HDR_LEN | RSV|Packet| RSS |
448 * | Checksum Ident | | | | Type | Type |
449 * +------------------------------------------------------+
450 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
451 * +------------------------------------------------------+
452 * 63 48 47 32 31 20 19 0
454 for (n = 0; n < adapter->num_rx_queues; n++) {
455 rx_ring = adapter->rx_ring[n];
456 pr_info("------------------------------------\n");
457 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
458 pr_info("------------------------------------\n");
459 pr_info("R [desc] [ PktBuf A0] "
460 "[ HeadBuf DD] [bi->dma ] [bi->skb] "
461 "<-- Adv Rx Read format\n");
462 pr_info("RWB[desc] [PcsmIpSHl PtRs] "
463 "[vl er S cks ln] ---------------- [bi->skb] "
464 "<-- Adv Rx Write-Back format\n");
466 for (i = 0; i < rx_ring->count; i++) {
467 rx_buffer_info = &rx_ring->rx_buffer_info[i];
468 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
469 u0 = (struct my_u0 *)rx_desc;
470 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
471 if (staterr & IXGBE_RXD_STAT_DD) {
472 /* Descriptor Done */
473 pr_info("RWB[0x%03X] %016llX "
474 "%016llX ---------------- %p", i,
477 rx_buffer_info->skb);
479 pr_info("R [0x%03X] %016llX "
480 "%016llX %016llX %p", i,
483 (u64)rx_buffer_info->dma,
484 rx_buffer_info->skb);
486 if (netif_msg_pktdata(adapter)) {
487 print_hex_dump(KERN_INFO, "",
488 DUMP_PREFIX_ADDRESS, 16, 1,
489 phys_to_virt(rx_buffer_info->dma),
490 rx_ring->rx_buf_len, true);
492 if (rx_ring->rx_buf_len
493 < IXGBE_RXBUFFER_2048)
494 print_hex_dump(KERN_INFO, "",
495 DUMP_PREFIX_ADDRESS, 16, 1,
497 rx_buffer_info->page_dma +
498 rx_buffer_info->page_offset
504 if (i == rx_ring->next_to_use)
506 else if (i == rx_ring->next_to_clean)
518 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
522 /* Let firmware take over control of h/w */
523 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
524 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
525 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
528 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
532 /* Let firmware know the driver has taken over */
533 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
534 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
535 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
539 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
540 * @adapter: pointer to adapter struct
541 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
542 * @queue: queue to map the corresponding interrupt to
543 * @msix_vector: the vector to map to the corresponding queue
546 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
547 u8 queue, u8 msix_vector)
550 struct ixgbe_hw *hw = &adapter->hw;
551 switch (hw->mac.type) {
552 case ixgbe_mac_82598EB:
553 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
556 index = (((direction * 64) + queue) >> 2) & 0x1F;
557 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
558 ivar &= ~(0xFF << (8 * (queue & 0x3)));
559 ivar |= (msix_vector << (8 * (queue & 0x3)));
560 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
562 case ixgbe_mac_82599EB:
563 if (direction == -1) {
565 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
566 index = ((queue & 1) * 8);
567 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
568 ivar &= ~(0xFF << index);
569 ivar |= (msix_vector << index);
570 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
573 /* tx or rx causes */
574 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
575 index = ((16 * (queue & 1)) + (8 * direction));
576 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
577 ivar &= ~(0xFF << index);
578 ivar |= (msix_vector << index);
579 IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
587 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
592 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
593 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
594 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
596 mask = (qmask & 0xFFFFFFFF);
597 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
598 mask = (qmask >> 32);
599 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
603 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_ring *tx_ring,
604 struct ixgbe_tx_buffer *tx_buffer_info)
606 if (tx_buffer_info->dma) {
607 if (tx_buffer_info->mapped_as_page)
608 dma_unmap_page(tx_ring->dev,
610 tx_buffer_info->length,
613 dma_unmap_single(tx_ring->dev,
615 tx_buffer_info->length,
617 tx_buffer_info->dma = 0;
619 if (tx_buffer_info->skb) {
620 dev_kfree_skb_any(tx_buffer_info->skb);
621 tx_buffer_info->skb = NULL;
623 tx_buffer_info->time_stamp = 0;
624 /* tx_buffer_info must be completely set up in the transmit path */
628 * ixgbe_tx_xon_state - check the tx ring xon state
629 * @adapter: the ixgbe adapter
630 * @tx_ring: the corresponding tx_ring
632 * If not in DCB mode, checks TFCS.TXOFF, otherwise, find out the
633 * corresponding TC of this tx_ring when checking TFCS.
635 * Returns : true if in xon state (currently not paused)
637 static inline bool ixgbe_tx_xon_state(struct ixgbe_adapter *adapter,
638 struct ixgbe_ring *tx_ring)
640 u32 txoff = IXGBE_TFCS_TXOFF;
642 #ifdef CONFIG_IXGBE_DCB
643 if (adapter->dcb_cfg.pfc_mode_enable) {
645 int reg_idx = tx_ring->reg_idx;
646 int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
648 switch (adapter->hw.mac.type) {
649 case ixgbe_mac_82598EB:
651 txoff = IXGBE_TFCS_TXOFF0;
653 case ixgbe_mac_82599EB:
655 txoff = IXGBE_TFCS_TXOFF;
659 if (tc == 2) /* TC2, TC3 */
660 tc += (reg_idx - 64) >> 4;
661 else if (tc == 3) /* TC4, TC5, TC6, TC7 */
662 tc += 1 + ((reg_idx - 96) >> 3);
663 } else if (dcb_i == 4) {
667 tc += (reg_idx - 64) >> 5;
668 if (tc == 2) /* TC2, TC3 */
669 tc += (reg_idx - 96) >> 4;
679 return IXGBE_READ_REG(&adapter->hw, IXGBE_TFCS) & txoff;
682 static inline bool ixgbe_check_tx_hang(struct ixgbe_adapter *adapter,
683 struct ixgbe_ring *tx_ring,
686 struct ixgbe_hw *hw = &adapter->hw;
688 /* Detect a transmit hang in hardware, this serializes the
689 * check with the clearing of time_stamp and movement of eop */
690 clear_check_for_tx_hang(tx_ring);
691 if (tx_ring->tx_buffer_info[eop].time_stamp &&
692 time_after(jiffies, tx_ring->tx_buffer_info[eop].time_stamp + HZ) &&
693 ixgbe_tx_xon_state(adapter, tx_ring)) {
694 /* detected Tx unit hang */
695 union ixgbe_adv_tx_desc *tx_desc;
696 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, eop);
697 e_err(drv, "Detected Tx Unit Hang\n"
699 " TDH, TDT <%x>, <%x>\n"
700 " next_to_use <%x>\n"
701 " next_to_clean <%x>\n"
702 "tx_buffer_info[next_to_clean]\n"
703 " time_stamp <%lx>\n"
705 tx_ring->queue_index,
706 IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
707 IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
708 tx_ring->next_to_use, eop,
709 tx_ring->tx_buffer_info[eop].time_stamp, jiffies);
716 #define IXGBE_MAX_TXD_PWR 14
717 #define IXGBE_MAX_DATA_PER_TXD (1 << IXGBE_MAX_TXD_PWR)
719 /* Tx Descriptors needed, worst case */
720 #define TXD_USE_COUNT(S) (((S) >> IXGBE_MAX_TXD_PWR) + \
721 (((S) & (IXGBE_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
722 #define DESC_NEEDED (TXD_USE_COUNT(IXGBE_MAX_DATA_PER_TXD) /* skb->data */ + \
723 MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1) /* for context */
725 static void ixgbe_tx_timeout(struct net_device *netdev);
728 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
729 * @q_vector: structure containing interrupt and ring information
730 * @tx_ring: tx ring to clean
732 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
733 struct ixgbe_ring *tx_ring)
735 struct ixgbe_adapter *adapter = q_vector->adapter;
736 union ixgbe_adv_tx_desc *tx_desc, *eop_desc;
737 struct ixgbe_tx_buffer *tx_buffer_info;
738 unsigned int total_bytes = 0, total_packets = 0;
739 u16 i, eop, count = 0;
741 i = tx_ring->next_to_clean;
742 eop = tx_ring->tx_buffer_info[i].next_to_watch;
743 eop_desc = IXGBE_TX_DESC_ADV(tx_ring, eop);
745 while ((eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)) &&
746 (count < tx_ring->work_limit)) {
747 bool cleaned = false;
748 rmb(); /* read buffer_info after eop_desc */
749 for ( ; !cleaned; count++) {
750 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
751 tx_buffer_info = &tx_ring->tx_buffer_info[i];
753 tx_desc->wb.status = 0;
754 cleaned = (i == eop);
757 if (i == tx_ring->count)
760 if (cleaned && tx_buffer_info->skb) {
761 total_bytes += tx_buffer_info->bytecount;
762 total_packets += tx_buffer_info->gso_segs;
765 ixgbe_unmap_and_free_tx_resource(tx_ring,
769 eop = tx_ring->tx_buffer_info[i].next_to_watch;
770 eop_desc = IXGBE_TX_DESC_ADV(tx_ring, eop);
773 tx_ring->next_to_clean = i;
774 tx_ring->total_bytes += total_bytes;
775 tx_ring->total_packets += total_packets;
776 u64_stats_update_begin(&tx_ring->syncp);
777 tx_ring->stats.packets += total_packets;
778 tx_ring->stats.bytes += total_bytes;
779 u64_stats_update_end(&tx_ring->syncp);
781 if (check_for_tx_hang(tx_ring) &&
782 ixgbe_check_tx_hang(adapter, tx_ring, i)) {
783 /* schedule immediate reset if we believe we hung */
784 e_info(probe, "tx hang %d detected, resetting "
785 "adapter\n", adapter->tx_timeout_count + 1);
786 ixgbe_tx_timeout(adapter->netdev);
788 /* the adapter is about to reset, no point in enabling stuff */
792 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
793 if (unlikely(count && netif_carrier_ok(tx_ring->netdev) &&
794 (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
795 /* Make sure that anybody stopping the queue after this
796 * sees the new next_to_clean.
799 if (__netif_subqueue_stopped(tx_ring->netdev, tx_ring->queue_index) &&
800 !test_bit(__IXGBE_DOWN, &adapter->state)) {
801 netif_wake_subqueue(tx_ring->netdev, tx_ring->queue_index);
802 ++tx_ring->tx_stats.restart_queue;
806 return count < tx_ring->work_limit;
809 #ifdef CONFIG_IXGBE_DCA
810 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
811 struct ixgbe_ring *rx_ring,
814 struct ixgbe_hw *hw = &adapter->hw;
816 u8 reg_idx = rx_ring->reg_idx;
818 rxctrl = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(reg_idx));
819 switch (hw->mac.type) {
820 case ixgbe_mac_82598EB:
821 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
822 rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
824 case ixgbe_mac_82599EB:
825 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK_82599;
826 rxctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
827 IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599);
832 rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
833 rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
834 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_RRO_EN);
835 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_WRO_EN |
836 IXGBE_DCA_RXCTRL_DESC_HSRO_EN);
837 IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
840 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
841 struct ixgbe_ring *tx_ring,
844 struct ixgbe_hw *hw = &adapter->hw;
846 u8 reg_idx = tx_ring->reg_idx;
848 switch (hw->mac.type) {
849 case ixgbe_mac_82598EB:
850 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(reg_idx));
851 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
852 txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
853 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
854 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
855 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(reg_idx), txctrl);
857 case ixgbe_mac_82599EB:
858 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(reg_idx));
859 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK_82599;
860 txctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
861 IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599);
862 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
863 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
864 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(reg_idx), txctrl);
871 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
873 struct ixgbe_adapter *adapter = q_vector->adapter;
878 if (q_vector->cpu == cpu)
881 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
882 for (i = 0; i < q_vector->txr_count; i++) {
883 ixgbe_update_tx_dca(adapter, adapter->tx_ring[r_idx], cpu);
884 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
888 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
889 for (i = 0; i < q_vector->rxr_count; i++) {
890 ixgbe_update_rx_dca(adapter, adapter->rx_ring[r_idx], cpu);
891 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
900 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
905 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
908 /* always use CB2 mode, difference is masked in the CB driver */
909 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
911 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
912 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
916 for (i = 0; i < num_q_vectors; i++) {
917 adapter->q_vector[i]->cpu = -1;
918 ixgbe_update_dca(adapter->q_vector[i]);
922 static int __ixgbe_notify_dca(struct device *dev, void *data)
924 struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
925 unsigned long event = *(unsigned long *)data;
927 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
931 case DCA_PROVIDER_ADD:
932 /* if we're already enabled, don't do it again */
933 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
935 if (dca_add_requester(dev) == 0) {
936 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
937 ixgbe_setup_dca(adapter);
940 /* Fall Through since DCA is disabled. */
941 case DCA_PROVIDER_REMOVE:
942 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
943 dca_remove_requester(dev);
944 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
945 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
953 #endif /* CONFIG_IXGBE_DCA */
955 * ixgbe_receive_skb - Send a completed packet up the stack
956 * @adapter: board private structure
957 * @skb: packet to send up
958 * @status: hardware indication of status of receive
959 * @rx_ring: rx descriptor ring (for a specific queue) to setup
960 * @rx_desc: rx descriptor
962 static void ixgbe_receive_skb(struct ixgbe_q_vector *q_vector,
963 struct sk_buff *skb, u8 status,
964 struct ixgbe_ring *ring,
965 union ixgbe_adv_rx_desc *rx_desc)
967 struct ixgbe_adapter *adapter = q_vector->adapter;
968 struct napi_struct *napi = &q_vector->napi;
969 bool is_vlan = (status & IXGBE_RXD_STAT_VP);
970 u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
972 if (is_vlan && (tag & VLAN_VID_MASK))
973 __vlan_hwaccel_put_tag(skb, tag);
975 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
976 napi_gro_receive(napi, skb);
982 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
983 * @adapter: address of board private structure
984 * @status_err: hardware indication of status of receive
985 * @skb: skb currently being received and modified
987 static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
988 union ixgbe_adv_rx_desc *rx_desc,
991 u32 status_err = le32_to_cpu(rx_desc->wb.upper.status_error);
993 skb_checksum_none_assert(skb);
995 /* Rx csum disabled */
996 if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
999 /* if IP and error */
1000 if ((status_err & IXGBE_RXD_STAT_IPCS) &&
1001 (status_err & IXGBE_RXDADV_ERR_IPE)) {
1002 adapter->hw_csum_rx_error++;
1006 if (!(status_err & IXGBE_RXD_STAT_L4CS))
1009 if (status_err & IXGBE_RXDADV_ERR_TCPE) {
1010 u16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1013 * 82599 errata, UDP frames with a 0 checksum can be marked as
1016 if ((pkt_info & IXGBE_RXDADV_PKTTYPE_UDP) &&
1017 (adapter->hw.mac.type == ixgbe_mac_82599EB))
1020 adapter->hw_csum_rx_error++;
1024 /* It must be a TCP or UDP packet with a valid checksum */
1025 skb->ip_summed = CHECKSUM_UNNECESSARY;
1028 static inline void ixgbe_release_rx_desc(struct ixgbe_ring *rx_ring, u32 val)
1031 * Force memory writes to complete before letting h/w
1032 * know there are new descriptors to fetch. (Only
1033 * applicable for weak-ordered memory model archs,
1037 writel(val, rx_ring->tail);
1041 * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
1042 * @rx_ring: ring to place buffers on
1043 * @cleaned_count: number of buffers to replace
1045 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1047 union ixgbe_adv_rx_desc *rx_desc;
1048 struct ixgbe_rx_buffer *bi;
1049 struct sk_buff *skb;
1050 u16 i = rx_ring->next_to_use;
1052 /* do nothing if no valid netdev defined */
1053 if (!rx_ring->netdev)
1056 while (cleaned_count--) {
1057 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1058 bi = &rx_ring->rx_buffer_info[i];
1062 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1063 rx_ring->rx_buf_len);
1065 rx_ring->rx_stats.alloc_rx_buff_failed++;
1068 /* initialize queue mapping */
1069 skb_record_rx_queue(skb, rx_ring->queue_index);
1074 bi->dma = dma_map_single(rx_ring->dev,
1076 rx_ring->rx_buf_len,
1078 if (dma_mapping_error(rx_ring->dev, bi->dma)) {
1079 rx_ring->rx_stats.alloc_rx_buff_failed++;
1085 if (ring_is_ps_enabled(rx_ring)) {
1087 bi->page = netdev_alloc_page(rx_ring->netdev);
1089 rx_ring->rx_stats.alloc_rx_page_failed++;
1094 if (!bi->page_dma) {
1095 /* use a half page if we're re-using */
1096 bi->page_offset ^= PAGE_SIZE / 2;
1097 bi->page_dma = dma_map_page(rx_ring->dev,
1102 if (dma_mapping_error(rx_ring->dev,
1104 rx_ring->rx_stats.alloc_rx_page_failed++;
1110 /* Refresh the desc even if buffer_addrs didn't change
1111 * because each write-back erases this info. */
1112 rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
1113 rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
1115 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
1116 rx_desc->read.hdr_addr = 0;
1120 if (i == rx_ring->count)
1125 if (rx_ring->next_to_use != i) {
1126 rx_ring->next_to_use = i;
1127 ixgbe_release_rx_desc(rx_ring, i);
1131 static inline u16 ixgbe_get_hlen(union ixgbe_adv_rx_desc *rx_desc)
1133 /* HW will not DMA in data larger than the given buffer, even if it
1134 * parses the (NFS, of course) header to be larger. In that case, it
1135 * fills the header buffer and spills the rest into the page.
1137 u16 hdr_info = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info);
1138 u16 hlen = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
1139 IXGBE_RXDADV_HDRBUFLEN_SHIFT;
1140 if (hlen > IXGBE_RX_HDR_SIZE)
1141 hlen = IXGBE_RX_HDR_SIZE;
1145 static inline u32 ixgbe_get_rsc_count(union ixgbe_adv_rx_desc *rx_desc)
1147 return (le32_to_cpu(rx_desc->wb.lower.lo_dword.data) &
1148 IXGBE_RXDADV_RSCCNT_MASK) >>
1149 IXGBE_RXDADV_RSCCNT_SHIFT;
1153 * ixgbe_transform_rsc_queue - change rsc queue into a full packet
1154 * @skb: pointer to the last skb in the rsc queue
1155 * @count: pointer to number of packets coalesced in this context
1157 * This function changes a queue full of hw rsc buffers into a completed
1158 * packet. It uses the ->prev pointers to find the first packet and then
1159 * turns it into the frag list owner.
1161 static inline struct sk_buff *ixgbe_transform_rsc_queue(struct sk_buff *skb,
1164 unsigned int frag_list_size = 0;
1167 struct sk_buff *prev = skb->prev;
1168 frag_list_size += skb->len;
1174 skb_shinfo(skb)->frag_list = skb->next;
1176 skb->len += frag_list_size;
1177 skb->data_len += frag_list_size;
1178 skb->truesize += frag_list_size;
1182 struct ixgbe_rsc_cb {
1187 #define IXGBE_RSC_CB(skb) ((struct ixgbe_rsc_cb *)(skb)->cb)
1189 static void ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1190 struct ixgbe_ring *rx_ring,
1191 int *work_done, int work_to_do)
1193 struct ixgbe_adapter *adapter = q_vector->adapter;
1194 union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
1195 struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
1196 struct sk_buff *skb;
1197 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1198 const int current_node = numa_node_id();
1199 unsigned int rsc_count = 0;
1202 #endif /* IXGBE_FCOE */
1205 u16 cleaned_count = 0;
1207 i = rx_ring->next_to_clean;
1208 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1209 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1211 while (staterr & IXGBE_RXD_STAT_DD) {
1214 rmb(); /* read descriptor and rx_buffer_info after status DD */
1216 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1218 skb = rx_buffer_info->skb;
1219 rx_buffer_info->skb = NULL;
1220 prefetch(skb->data);
1222 if (ring_is_rsc_enabled(rx_ring))
1223 rsc_count = ixgbe_get_rsc_count(rx_desc);
1225 /* if this is a skb from previous receive DMA will be 0 */
1226 if (rx_buffer_info->dma) {
1229 !(staterr & IXGBE_RXD_STAT_EOP) &&
1232 * When HWRSC is enabled, delay unmapping
1233 * of the first packet. It carries the
1234 * header information, HW may still
1235 * access the header after the writeback.
1236 * Only unmap it when EOP is reached
1238 IXGBE_RSC_CB(skb)->delay_unmap = true;
1239 IXGBE_RSC_CB(skb)->dma = rx_buffer_info->dma;
1241 dma_unmap_single(rx_ring->dev,
1242 rx_buffer_info->dma,
1243 rx_ring->rx_buf_len,
1246 rx_buffer_info->dma = 0;
1248 if (ring_is_ps_enabled(rx_ring)) {
1249 hlen = ixgbe_get_hlen(rx_desc);
1250 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1252 hlen = le16_to_cpu(rx_desc->wb.upper.length);
1257 /* assume packet split since header is unmapped */
1258 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1262 dma_unmap_page(rx_ring->dev,
1263 rx_buffer_info->page_dma,
1266 rx_buffer_info->page_dma = 0;
1267 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1268 rx_buffer_info->page,
1269 rx_buffer_info->page_offset,
1272 if ((page_count(rx_buffer_info->page) == 1) &&
1273 (page_to_nid(rx_buffer_info->page) == current_node))
1274 get_page(rx_buffer_info->page);
1276 rx_buffer_info->page = NULL;
1278 skb->len += upper_len;
1279 skb->data_len += upper_len;
1280 skb->truesize += upper_len;
1284 if (i == rx_ring->count)
1287 next_rxd = IXGBE_RX_DESC_ADV(rx_ring, i);
1292 u32 nextp = (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
1293 IXGBE_RXDADV_NEXTP_SHIFT;
1294 next_buffer = &rx_ring->rx_buffer_info[nextp];
1296 next_buffer = &rx_ring->rx_buffer_info[i];
1299 if (!(staterr & IXGBE_RXD_STAT_EOP)) {
1300 if (ring_is_ps_enabled(rx_ring)) {
1301 rx_buffer_info->skb = next_buffer->skb;
1302 rx_buffer_info->dma = next_buffer->dma;
1303 next_buffer->skb = skb;
1304 next_buffer->dma = 0;
1306 skb->next = next_buffer->skb;
1307 skb->next->prev = skb;
1309 rx_ring->rx_stats.non_eop_descs++;
1314 skb = ixgbe_transform_rsc_queue(skb,
1315 &(rx_ring->rx_stats.rsc_count));
1317 if (ring_is_rsc_enabled(rx_ring)) {
1318 if (IXGBE_RSC_CB(skb)->delay_unmap) {
1319 dma_unmap_single(rx_ring->dev,
1320 IXGBE_RSC_CB(skb)->dma,
1321 rx_ring->rx_buf_len,
1323 IXGBE_RSC_CB(skb)->dma = 0;
1324 IXGBE_RSC_CB(skb)->delay_unmap = false;
1326 if (ring_is_ps_enabled(rx_ring))
1327 rx_ring->rx_stats.rsc_count +=
1328 skb_shinfo(skb)->nr_frags;
1330 rx_ring->rx_stats.rsc_count++;
1331 rx_ring->rx_stats.rsc_flush++;
1334 /* ERR_MASK will only have valid bits if EOP set */
1335 if (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) {
1336 /* trim packet back to size 0 and recycle it */
1337 __pskb_trim(skb, 0);
1338 rx_buffer_info->skb = skb;
1342 ixgbe_rx_checksum(adapter, rx_desc, skb);
1344 /* probably a little skewed due to removing CRC */
1345 total_rx_bytes += skb->len;
1348 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1350 /* if ddp, not passing to ULD unless for FCP_RSP or error */
1351 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
1352 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
1356 #endif /* IXGBE_FCOE */
1357 ixgbe_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
1360 rx_desc->wb.upper.status_error = 0;
1363 if (*work_done >= work_to_do)
1366 /* return some buffers to hardware, one at a time is too slow */
1367 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
1368 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1372 /* use prefetched values */
1374 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1377 rx_ring->next_to_clean = i;
1378 cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
1381 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1384 /* include DDPed FCoE data */
1385 if (ddp_bytes > 0) {
1388 mss = rx_ring->netdev->mtu - sizeof(struct fcoe_hdr) -
1389 sizeof(struct fc_frame_header) -
1390 sizeof(struct fcoe_crc_eof);
1393 total_rx_bytes += ddp_bytes;
1394 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
1396 #endif /* IXGBE_FCOE */
1398 rx_ring->total_packets += total_rx_packets;
1399 rx_ring->total_bytes += total_rx_bytes;
1400 u64_stats_update_begin(&rx_ring->syncp);
1401 rx_ring->stats.packets += total_rx_packets;
1402 rx_ring->stats.bytes += total_rx_bytes;
1403 u64_stats_update_end(&rx_ring->syncp);
1406 static int ixgbe_clean_rxonly(struct napi_struct *, int);
1408 * ixgbe_configure_msix - Configure MSI-X hardware
1409 * @adapter: board private structure
1411 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1414 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1416 struct ixgbe_q_vector *q_vector;
1417 int i, j, q_vectors, v_idx, r_idx;
1420 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1423 * Populate the IVAR table and set the ITR values to the
1424 * corresponding register.
1426 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1427 q_vector = adapter->q_vector[v_idx];
1428 /* XXX for_each_set_bit(...) */
1429 r_idx = find_first_bit(q_vector->rxr_idx,
1430 adapter->num_rx_queues);
1432 for (i = 0; i < q_vector->rxr_count; i++) {
1433 j = adapter->rx_ring[r_idx]->reg_idx;
1434 ixgbe_set_ivar(adapter, 0, j, v_idx);
1435 r_idx = find_next_bit(q_vector->rxr_idx,
1436 adapter->num_rx_queues,
1439 r_idx = find_first_bit(q_vector->txr_idx,
1440 adapter->num_tx_queues);
1442 for (i = 0; i < q_vector->txr_count; i++) {
1443 j = adapter->tx_ring[r_idx]->reg_idx;
1444 ixgbe_set_ivar(adapter, 1, j, v_idx);
1445 r_idx = find_next_bit(q_vector->txr_idx,
1446 adapter->num_tx_queues,
1450 if (q_vector->txr_count && !q_vector->rxr_count)
1452 q_vector->eitr = adapter->tx_eitr_param;
1453 else if (q_vector->rxr_count)
1455 q_vector->eitr = adapter->rx_eitr_param;
1457 ixgbe_write_eitr(q_vector);
1458 /* If Flow Director is enabled, set interrupt affinity */
1459 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
1460 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)) {
1462 * Allocate the affinity_hint cpumask, assign the mask
1463 * for this vector, and set our affinity_hint for
1466 if (!alloc_cpumask_var(&q_vector->affinity_mask,
1469 cpumask_set_cpu(v_idx, q_vector->affinity_mask);
1470 irq_set_affinity_hint(adapter->msix_entries[v_idx].vector,
1471 q_vector->affinity_mask);
1475 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
1476 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1478 else if (adapter->hw.mac.type == ixgbe_mac_82599EB)
1479 ixgbe_set_ivar(adapter, -1, 1, v_idx);
1480 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1482 /* set up to autoclear timer, and the vectors */
1483 mask = IXGBE_EIMS_ENABLE_MASK;
1484 if (adapter->num_vfs)
1485 mask &= ~(IXGBE_EIMS_OTHER |
1486 IXGBE_EIMS_MAILBOX |
1489 mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
1490 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1493 enum latency_range {
1497 latency_invalid = 255
1501 * ixgbe_update_itr - update the dynamic ITR value based on statistics
1502 * @adapter: pointer to adapter
1503 * @eitr: eitr setting (ints per sec) to give last timeslice
1504 * @itr_setting: current throttle rate in ints/second
1505 * @packets: the number of packets during this measurement interval
1506 * @bytes: the number of bytes during this measurement interval
1508 * Stores a new ITR value based on packets and byte
1509 * counts during the last interrupt. The advantage of per interrupt
1510 * computation is faster updates and more accurate ITR for the current
1511 * traffic pattern. Constants in this function were computed
1512 * based on theoretical maximum wire speed and thresholds were set based
1513 * on testing data as well as attempting to minimize response time
1514 * while increasing bulk throughput.
1515 * this functionality is controlled by the InterruptThrottleRate module
1516 * parameter (see ixgbe_param.c)
1518 static u8 ixgbe_update_itr(struct ixgbe_adapter *adapter,
1519 u32 eitr, u8 itr_setting,
1520 int packets, int bytes)
1522 unsigned int retval = itr_setting;
1527 goto update_itr_done;
1530 /* simple throttlerate management
1531 * 0-20MB/s lowest (100000 ints/s)
1532 * 20-100MB/s low (20000 ints/s)
1533 * 100-1249MB/s bulk (8000 ints/s)
1535 /* what was last interrupt timeslice? */
1536 timepassed_us = 1000000/eitr;
1537 bytes_perint = bytes / timepassed_us; /* bytes/usec */
1539 switch (itr_setting) {
1540 case lowest_latency:
1541 if (bytes_perint > adapter->eitr_low)
1542 retval = low_latency;
1545 if (bytes_perint > adapter->eitr_high)
1546 retval = bulk_latency;
1547 else if (bytes_perint <= adapter->eitr_low)
1548 retval = lowest_latency;
1551 if (bytes_perint <= adapter->eitr_high)
1552 retval = low_latency;
1561 * ixgbe_write_eitr - write EITR register in hardware specific way
1562 * @q_vector: structure containing interrupt and ring information
1564 * This function is made to be called by ethtool and by the driver
1565 * when it needs to update EITR registers at runtime. Hardware
1566 * specific quirks/differences are taken care of here.
1568 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1570 struct ixgbe_adapter *adapter = q_vector->adapter;
1571 struct ixgbe_hw *hw = &adapter->hw;
1572 int v_idx = q_vector->v_idx;
1573 u32 itr_reg = EITR_INTS_PER_SEC_TO_REG(q_vector->eitr);
1575 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1576 /* must write high and low 16 bits to reset counter */
1577 itr_reg |= (itr_reg << 16);
1578 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1580 * 82599 can support a value of zero, so allow it for
1581 * max interrupt rate, but there is an errata where it can
1582 * not be zero with RSC
1585 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
1589 * set the WDIS bit to not clear the timer bits and cause an
1590 * immediate assertion of the interrupt
1592 itr_reg |= IXGBE_EITR_CNT_WDIS;
1594 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1597 static void ixgbe_set_itr_msix(struct ixgbe_q_vector *q_vector)
1599 struct ixgbe_adapter *adapter = q_vector->adapter;
1601 u8 current_itr, ret_itr;
1603 struct ixgbe_ring *rx_ring, *tx_ring;
1605 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1606 for (i = 0; i < q_vector->txr_count; i++) {
1607 tx_ring = adapter->tx_ring[r_idx];
1608 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1610 tx_ring->total_packets,
1611 tx_ring->total_bytes);
1612 /* if the result for this queue would decrease interrupt
1613 * rate for this vector then use that result */
1614 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
1615 q_vector->tx_itr - 1 : ret_itr);
1616 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1620 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1621 for (i = 0; i < q_vector->rxr_count; i++) {
1622 rx_ring = adapter->rx_ring[r_idx];
1623 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1625 rx_ring->total_packets,
1626 rx_ring->total_bytes);
1627 /* if the result for this queue would decrease interrupt
1628 * rate for this vector then use that result */
1629 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
1630 q_vector->rx_itr - 1 : ret_itr);
1631 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1635 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1637 switch (current_itr) {
1638 /* counts and packets in update_itr are dependent on these numbers */
1639 case lowest_latency:
1643 new_itr = 20000; /* aka hwitr = ~200 */
1651 if (new_itr != q_vector->eitr) {
1652 /* do an exponential smoothing */
1653 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1655 /* save the algorithm value here, not the smoothed one */
1656 q_vector->eitr = new_itr;
1658 ixgbe_write_eitr(q_vector);
1663 * ixgbe_check_overtemp_task - worker thread to check over tempurature
1664 * @work: pointer to work_struct containing our data
1666 static void ixgbe_check_overtemp_task(struct work_struct *work)
1668 struct ixgbe_adapter *adapter = container_of(work,
1669 struct ixgbe_adapter,
1670 check_overtemp_task);
1671 struct ixgbe_hw *hw = &adapter->hw;
1672 u32 eicr = adapter->interrupt_event;
1674 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
1677 switch (hw->device_id) {
1678 case IXGBE_DEV_ID_82599_T3_LOM: {
1680 bool link_up = false;
1682 if (hw->mac.ops.check_link)
1683 hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
1685 if (((eicr & IXGBE_EICR_GPI_SDP0) && (!link_up)) ||
1686 (eicr & IXGBE_EICR_LSC))
1687 /* Check if this is due to overtemp */
1688 if (hw->phy.ops.check_overtemp(hw) == IXGBE_ERR_OVERTEMP)
1693 if (!(eicr & IXGBE_EICR_GPI_SDP0))
1698 "Network adapter has been stopped because it has over heated. "
1699 "Restart the computer. If the problem persists, "
1700 "power off the system and replace the adapter\n");
1701 /* write to clear the interrupt */
1702 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP0);
1705 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
1707 struct ixgbe_hw *hw = &adapter->hw;
1709 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
1710 (eicr & IXGBE_EICR_GPI_SDP1)) {
1711 e_crit(probe, "Fan has stopped, replace the adapter\n");
1712 /* write to clear the interrupt */
1713 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1717 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
1719 struct ixgbe_hw *hw = &adapter->hw;
1721 if (eicr & IXGBE_EICR_GPI_SDP2) {
1722 /* Clear the interrupt */
1723 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
1724 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1725 schedule_work(&adapter->sfp_config_module_task);
1728 if (eicr & IXGBE_EICR_GPI_SDP1) {
1729 /* Clear the interrupt */
1730 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1731 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1732 schedule_work(&adapter->multispeed_fiber_task);
1736 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1738 struct ixgbe_hw *hw = &adapter->hw;
1741 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1742 adapter->link_check_timeout = jiffies;
1743 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1744 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1745 IXGBE_WRITE_FLUSH(hw);
1746 schedule_work(&adapter->watchdog_task);
1750 static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
1752 struct net_device *netdev = data;
1753 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1754 struct ixgbe_hw *hw = &adapter->hw;
1758 * Workaround for Silicon errata. Use clear-by-write instead
1759 * of clear-by-read. Reading with EICS will return the
1760 * interrupt causes without clearing, which later be done
1761 * with the write to EICR.
1763 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1764 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1766 if (eicr & IXGBE_EICR_LSC)
1767 ixgbe_check_lsc(adapter);
1769 if (eicr & IXGBE_EICR_MAILBOX)
1770 ixgbe_msg_task(adapter);
1772 if (hw->mac.type == ixgbe_mac_82598EB)
1773 ixgbe_check_fan_failure(adapter, eicr);
1775 if (hw->mac.type == ixgbe_mac_82599EB) {
1776 ixgbe_check_sfp_event(adapter, eicr);
1777 adapter->interrupt_event = eicr;
1778 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
1779 ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)))
1780 schedule_work(&adapter->check_overtemp_task);
1782 /* Handle Flow Director Full threshold interrupt */
1783 if (eicr & IXGBE_EICR_FLOW_DIR) {
1785 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_FLOW_DIR);
1786 /* Disable transmits before FDIR Re-initialization */
1787 netif_tx_stop_all_queues(netdev);
1788 for (i = 0; i < adapter->num_tx_queues; i++) {
1789 struct ixgbe_ring *tx_ring =
1790 adapter->tx_ring[i];
1791 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
1793 schedule_work(&adapter->fdir_reinit_task);
1797 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1798 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_OTHER);
1803 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
1808 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1809 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1810 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1812 mask = (qmask & 0xFFFFFFFF);
1813 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(0), mask);
1814 mask = (qmask >> 32);
1815 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(1), mask);
1817 /* skip the flush */
1820 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
1825 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1826 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1827 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, mask);
1829 mask = (qmask & 0xFFFFFFFF);
1830 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), mask);
1831 mask = (qmask >> 32);
1832 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), mask);
1834 /* skip the flush */
1837 static irqreturn_t ixgbe_msix_clean_tx(int irq, void *data)
1839 struct ixgbe_q_vector *q_vector = data;
1840 struct ixgbe_adapter *adapter = q_vector->adapter;
1841 struct ixgbe_ring *tx_ring;
1844 if (!q_vector->txr_count)
1847 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1848 for (i = 0; i < q_vector->txr_count; i++) {
1849 tx_ring = adapter->tx_ring[r_idx];
1850 tx_ring->total_bytes = 0;
1851 tx_ring->total_packets = 0;
1852 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1856 /* EIAM disabled interrupts (on this vector) for us */
1857 napi_schedule(&q_vector->napi);
1863 * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
1865 * @data: pointer to our q_vector struct for this interrupt vector
1867 static irqreturn_t ixgbe_msix_clean_rx(int irq, void *data)
1869 struct ixgbe_q_vector *q_vector = data;
1870 struct ixgbe_adapter *adapter = q_vector->adapter;
1871 struct ixgbe_ring *rx_ring;
1875 #ifdef CONFIG_IXGBE_DCA
1876 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1877 ixgbe_update_dca(q_vector);
1880 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1881 for (i = 0; i < q_vector->rxr_count; i++) {
1882 rx_ring = adapter->rx_ring[r_idx];
1883 rx_ring->total_bytes = 0;
1884 rx_ring->total_packets = 0;
1885 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1889 if (!q_vector->rxr_count)
1892 /* EIAM disabled interrupts (on this vector) for us */
1893 napi_schedule(&q_vector->napi);
1898 static irqreturn_t ixgbe_msix_clean_many(int irq, void *data)
1900 struct ixgbe_q_vector *q_vector = data;
1901 struct ixgbe_adapter *adapter = q_vector->adapter;
1902 struct ixgbe_ring *ring;
1906 if (!q_vector->txr_count && !q_vector->rxr_count)
1909 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1910 for (i = 0; i < q_vector->txr_count; i++) {
1911 ring = adapter->tx_ring[r_idx];
1912 ring->total_bytes = 0;
1913 ring->total_packets = 0;
1914 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1918 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1919 for (i = 0; i < q_vector->rxr_count; i++) {
1920 ring = adapter->rx_ring[r_idx];
1921 ring->total_bytes = 0;
1922 ring->total_packets = 0;
1923 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1927 /* EIAM disabled interrupts (on this vector) for us */
1928 napi_schedule(&q_vector->napi);
1934 * ixgbe_clean_rxonly - msix (aka one shot) rx clean routine
1935 * @napi: napi struct with our devices info in it
1936 * @budget: amount of work driver is allowed to do this pass, in packets
1938 * This function is optimized for cleaning one queue only on a single
1941 static int ixgbe_clean_rxonly(struct napi_struct *napi, int budget)
1943 struct ixgbe_q_vector *q_vector =
1944 container_of(napi, struct ixgbe_q_vector, napi);
1945 struct ixgbe_adapter *adapter = q_vector->adapter;
1946 struct ixgbe_ring *rx_ring = NULL;
1950 #ifdef CONFIG_IXGBE_DCA
1951 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1952 ixgbe_update_dca(q_vector);
1955 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1956 rx_ring = adapter->rx_ring[r_idx];
1958 ixgbe_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
1960 /* If all Rx work done, exit the polling mode */
1961 if (work_done < budget) {
1962 napi_complete(napi);
1963 if (adapter->rx_itr_setting & 1)
1964 ixgbe_set_itr_msix(q_vector);
1965 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1966 ixgbe_irq_enable_queues(adapter,
1967 ((u64)1 << q_vector->v_idx));
1974 * ixgbe_clean_rxtx_many - msix (aka one shot) rx clean routine
1975 * @napi: napi struct with our devices info in it
1976 * @budget: amount of work driver is allowed to do this pass, in packets
1978 * This function will clean more than one rx queue associated with a
1981 static int ixgbe_clean_rxtx_many(struct napi_struct *napi, int budget)
1983 struct ixgbe_q_vector *q_vector =
1984 container_of(napi, struct ixgbe_q_vector, napi);
1985 struct ixgbe_adapter *adapter = q_vector->adapter;
1986 struct ixgbe_ring *ring = NULL;
1987 int work_done = 0, i;
1989 bool tx_clean_complete = true;
1991 #ifdef CONFIG_IXGBE_DCA
1992 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1993 ixgbe_update_dca(q_vector);
1996 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1997 for (i = 0; i < q_vector->txr_count; i++) {
1998 ring = adapter->tx_ring[r_idx];
1999 tx_clean_complete &= ixgbe_clean_tx_irq(q_vector, ring);
2000 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
2004 /* attempt to distribute budget to each queue fairly, but don't allow
2005 * the budget to go below 1 because we'll exit polling */
2006 budget /= (q_vector->rxr_count ?: 1);
2007 budget = max(budget, 1);
2008 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2009 for (i = 0; i < q_vector->rxr_count; i++) {
2010 ring = adapter->rx_ring[r_idx];
2011 ixgbe_clean_rx_irq(q_vector, ring, &work_done, budget);
2012 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
2016 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2017 ring = adapter->rx_ring[r_idx];
2018 /* If all Rx work done, exit the polling mode */
2019 if (work_done < budget) {
2020 napi_complete(napi);
2021 if (adapter->rx_itr_setting & 1)
2022 ixgbe_set_itr_msix(q_vector);
2023 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2024 ixgbe_irq_enable_queues(adapter,
2025 ((u64)1 << q_vector->v_idx));
2033 * ixgbe_clean_txonly - msix (aka one shot) tx clean routine
2034 * @napi: napi struct with our devices info in it
2035 * @budget: amount of work driver is allowed to do this pass, in packets
2037 * This function is optimized for cleaning one queue only on a single
2040 static int ixgbe_clean_txonly(struct napi_struct *napi, int budget)
2042 struct ixgbe_q_vector *q_vector =
2043 container_of(napi, struct ixgbe_q_vector, napi);
2044 struct ixgbe_adapter *adapter = q_vector->adapter;
2045 struct ixgbe_ring *tx_ring = NULL;
2049 #ifdef CONFIG_IXGBE_DCA
2050 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2051 ixgbe_update_dca(q_vector);
2054 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
2055 tx_ring = adapter->tx_ring[r_idx];
2057 if (!ixgbe_clean_tx_irq(q_vector, tx_ring))
2060 /* If all Tx work done, exit the polling mode */
2061 if (work_done < budget) {
2062 napi_complete(napi);
2063 if (adapter->tx_itr_setting & 1)
2064 ixgbe_set_itr_msix(q_vector);
2065 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2066 ixgbe_irq_enable_queues(adapter,
2067 ((u64)1 << q_vector->v_idx));
2073 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
2076 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2078 set_bit(r_idx, q_vector->rxr_idx);
2079 q_vector->rxr_count++;
2082 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
2085 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2087 set_bit(t_idx, q_vector->txr_idx);
2088 q_vector->txr_count++;
2092 * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
2093 * @adapter: board private structure to initialize
2094 * @vectors: allotted vector count for descriptor rings
2096 * This function maps descriptor rings to the queue-specific vectors
2097 * we were allotted through the MSI-X enabling code. Ideally, we'd have
2098 * one vector per ring/queue, but on a constrained vector budget, we
2099 * group the rings as "efficiently" as possible. You would add new
2100 * mapping configurations in here.
2102 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter,
2106 int rxr_idx = 0, txr_idx = 0;
2107 int rxr_remaining = adapter->num_rx_queues;
2108 int txr_remaining = adapter->num_tx_queues;
2113 /* No mapping required if MSI-X is disabled. */
2114 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2118 * The ideal configuration...
2119 * We have enough vectors to map one per queue.
2121 if (vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
2122 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
2123 map_vector_to_rxq(adapter, v_start, rxr_idx);
2125 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
2126 map_vector_to_txq(adapter, v_start, txr_idx);
2132 * If we don't have enough vectors for a 1-to-1
2133 * mapping, we'll have to group them so there are
2134 * multiple queues per vector.
2136 /* Re-adjusting *qpv takes care of the remainder. */
2137 for (i = v_start; i < vectors; i++) {
2138 rqpv = DIV_ROUND_UP(rxr_remaining, vectors - i);
2139 for (j = 0; j < rqpv; j++) {
2140 map_vector_to_rxq(adapter, i, rxr_idx);
2145 for (i = v_start; i < vectors; i++) {
2146 tqpv = DIV_ROUND_UP(txr_remaining, vectors - i);
2147 for (j = 0; j < tqpv; j++) {
2148 map_vector_to_txq(adapter, i, txr_idx);
2159 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2160 * @adapter: board private structure
2162 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2163 * interrupts from the kernel.
2165 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2167 struct net_device *netdev = adapter->netdev;
2168 irqreturn_t (*handler)(int, void *);
2169 int i, vector, q_vectors, err;
2172 /* Decrement for Other and TCP Timer vectors */
2173 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2175 /* Map the Tx/Rx rings to the vectors we were allotted. */
2176 err = ixgbe_map_rings_to_vectors(adapter, q_vectors);
2180 #define SET_HANDLER(_v) ((!(_v)->rxr_count) ? &ixgbe_msix_clean_tx : \
2181 (!(_v)->txr_count) ? &ixgbe_msix_clean_rx : \
2182 &ixgbe_msix_clean_many)
2183 for (vector = 0; vector < q_vectors; vector++) {
2184 handler = SET_HANDLER(adapter->q_vector[vector]);
2186 if (handler == &ixgbe_msix_clean_rx) {
2187 sprintf(adapter->name[vector], "%s-%s-%d",
2188 netdev->name, "rx", ri++);
2189 } else if (handler == &ixgbe_msix_clean_tx) {
2190 sprintf(adapter->name[vector], "%s-%s-%d",
2191 netdev->name, "tx", ti++);
2193 sprintf(adapter->name[vector], "%s-%s-%d",
2194 netdev->name, "TxRx", ri++);
2198 err = request_irq(adapter->msix_entries[vector].vector,
2199 handler, 0, adapter->name[vector],
2200 adapter->q_vector[vector]);
2202 e_err(probe, "request_irq failed for MSIX interrupt "
2203 "Error: %d\n", err);
2204 goto free_queue_irqs;
2208 sprintf(adapter->name[vector], "%s:lsc", netdev->name);
2209 err = request_irq(adapter->msix_entries[vector].vector,
2210 ixgbe_msix_lsc, 0, adapter->name[vector], netdev);
2212 e_err(probe, "request_irq for msix_lsc failed: %d\n", err);
2213 goto free_queue_irqs;
2219 for (i = vector - 1; i >= 0; i--)
2220 free_irq(adapter->msix_entries[--vector].vector,
2221 adapter->q_vector[i]);
2222 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2223 pci_disable_msix(adapter->pdev);
2224 kfree(adapter->msix_entries);
2225 adapter->msix_entries = NULL;
2230 static void ixgbe_set_itr(struct ixgbe_adapter *adapter)
2232 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2234 u32 new_itr = q_vector->eitr;
2235 struct ixgbe_ring *rx_ring = adapter->rx_ring[0];
2236 struct ixgbe_ring *tx_ring = adapter->tx_ring[0];
2238 q_vector->tx_itr = ixgbe_update_itr(adapter, new_itr,
2240 tx_ring->total_packets,
2241 tx_ring->total_bytes);
2242 q_vector->rx_itr = ixgbe_update_itr(adapter, new_itr,
2244 rx_ring->total_packets,
2245 rx_ring->total_bytes);
2247 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
2249 switch (current_itr) {
2250 /* counts and packets in update_itr are dependent on these numbers */
2251 case lowest_latency:
2255 new_itr = 20000; /* aka hwitr = ~200 */
2264 if (new_itr != q_vector->eitr) {
2265 /* do an exponential smoothing */
2266 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
2268 /* save the algorithm value here, not the smoothed one */
2269 q_vector->eitr = new_itr;
2271 ixgbe_write_eitr(q_vector);
2276 * ixgbe_irq_enable - Enable default interrupt generation settings
2277 * @adapter: board private structure
2279 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2284 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2285 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2286 mask |= IXGBE_EIMS_GPI_SDP0;
2287 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2288 mask |= IXGBE_EIMS_GPI_SDP1;
2289 if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
2290 mask |= IXGBE_EIMS_ECC;
2291 mask |= IXGBE_EIMS_GPI_SDP1;
2292 mask |= IXGBE_EIMS_GPI_SDP2;
2293 if (adapter->num_vfs)
2294 mask |= IXGBE_EIMS_MAILBOX;
2296 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
2297 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
2298 mask |= IXGBE_EIMS_FLOW_DIR;
2300 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2302 ixgbe_irq_enable_queues(adapter, ~0);
2304 IXGBE_WRITE_FLUSH(&adapter->hw);
2306 if (adapter->num_vfs > 32) {
2307 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
2308 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2313 * ixgbe_intr - legacy mode Interrupt Handler
2314 * @irq: interrupt number
2315 * @data: pointer to a network interface device structure
2317 static irqreturn_t ixgbe_intr(int irq, void *data)
2319 struct net_device *netdev = data;
2320 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2321 struct ixgbe_hw *hw = &adapter->hw;
2322 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2326 * Workaround for silicon errata on 82598. Mask the interrupts
2327 * before the read of EICR.
2329 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2331 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2332 * therefore no explict interrupt disable is necessary */
2333 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2336 * shared interrupt alert!
2337 * make sure interrupts are enabled because the read will
2338 * have disabled interrupts due to EIAM
2339 * finish the workaround of silicon errata on 82598. Unmask
2340 * the interrupt that we masked before the EICR read.
2342 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2343 ixgbe_irq_enable(adapter, true, true);
2344 return IRQ_NONE; /* Not our interrupt */
2347 if (eicr & IXGBE_EICR_LSC)
2348 ixgbe_check_lsc(adapter);
2350 if (hw->mac.type == ixgbe_mac_82599EB)
2351 ixgbe_check_sfp_event(adapter, eicr);
2353 ixgbe_check_fan_failure(adapter, eicr);
2354 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2355 ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)))
2356 schedule_work(&adapter->check_overtemp_task);
2358 if (napi_schedule_prep(&(q_vector->napi))) {
2359 adapter->tx_ring[0]->total_packets = 0;
2360 adapter->tx_ring[0]->total_bytes = 0;
2361 adapter->rx_ring[0]->total_packets = 0;
2362 adapter->rx_ring[0]->total_bytes = 0;
2363 /* would disable interrupts here but EIAM disabled it */
2364 __napi_schedule(&(q_vector->napi));
2368 * re-enable link(maybe) and non-queue interrupts, no flush.
2369 * ixgbe_poll will re-enable the queue interrupts
2372 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2373 ixgbe_irq_enable(adapter, false, false);
2378 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
2380 int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2382 for (i = 0; i < q_vectors; i++) {
2383 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
2384 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
2385 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
2386 q_vector->rxr_count = 0;
2387 q_vector->txr_count = 0;
2392 * ixgbe_request_irq - initialize interrupts
2393 * @adapter: board private structure
2395 * Attempts to configure interrupts using the best available
2396 * capabilities of the hardware and kernel.
2398 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2400 struct net_device *netdev = adapter->netdev;
2403 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2404 err = ixgbe_request_msix_irqs(adapter);
2405 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
2406 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2407 netdev->name, netdev);
2409 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2410 netdev->name, netdev);
2414 e_err(probe, "request_irq failed, Error %d\n", err);
2419 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2421 struct net_device *netdev = adapter->netdev;
2423 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2426 q_vectors = adapter->num_msix_vectors;
2429 free_irq(adapter->msix_entries[i].vector, netdev);
2432 for (; i >= 0; i--) {
2433 free_irq(adapter->msix_entries[i].vector,
2434 adapter->q_vector[i]);
2437 ixgbe_reset_q_vectors(adapter);
2439 free_irq(adapter->pdev->irq, netdev);
2444 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2445 * @adapter: board private structure
2447 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2449 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2450 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2452 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2453 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2454 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2455 if (adapter->num_vfs > 32)
2456 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
2458 IXGBE_WRITE_FLUSH(&adapter->hw);
2459 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2461 for (i = 0; i < adapter->num_msix_vectors; i++)
2462 synchronize_irq(adapter->msix_entries[i].vector);
2464 synchronize_irq(adapter->pdev->irq);
2469 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2472 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2474 struct ixgbe_hw *hw = &adapter->hw;
2476 IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
2477 EITR_INTS_PER_SEC_TO_REG(adapter->rx_eitr_param));
2479 ixgbe_set_ivar(adapter, 0, 0, 0);
2480 ixgbe_set_ivar(adapter, 1, 0, 0);
2482 map_vector_to_rxq(adapter, 0, 0);
2483 map_vector_to_txq(adapter, 0, 0);
2485 e_info(hw, "Legacy interrupt IVAR setup done\n");
2489 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2490 * @adapter: board private structure
2491 * @ring: structure containing ring specific data
2493 * Configure the Tx descriptor ring after a reset.
2495 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2496 struct ixgbe_ring *ring)
2498 struct ixgbe_hw *hw = &adapter->hw;
2499 u64 tdba = ring->dma;
2502 u16 reg_idx = ring->reg_idx;
2504 /* disable queue to avoid issues while updating state */
2505 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2506 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx),
2507 txdctl & ~IXGBE_TXDCTL_ENABLE);
2508 IXGBE_WRITE_FLUSH(hw);
2510 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2511 (tdba & DMA_BIT_MASK(32)));
2512 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2513 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2514 ring->count * sizeof(union ixgbe_adv_tx_desc));
2515 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2516 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2517 ring->tail = hw->hw_addr + IXGBE_TDT(reg_idx);
2519 /* configure fetching thresholds */
2520 if (adapter->rx_itr_setting == 0) {
2521 /* cannot set wthresh when itr==0 */
2522 txdctl &= ~0x007F0000;
2524 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
2525 txdctl |= (8 << 16);
2527 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2528 /* PThresh workaround for Tx hang with DFP enabled. */
2532 /* reinitialize flowdirector state */
2533 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2534 adapter->atr_sample_rate) {
2535 ring->atr_sample_rate = adapter->atr_sample_rate;
2536 ring->atr_count = 0;
2537 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
2539 ring->atr_sample_rate = 0;
2543 txdctl |= IXGBE_TXDCTL_ENABLE;
2544 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
2546 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2547 if (hw->mac.type == ixgbe_mac_82598EB &&
2548 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2551 /* poll to verify queue is enabled */
2554 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2555 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
2557 e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
2560 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
2562 struct ixgbe_hw *hw = &adapter->hw;
2566 if (hw->mac.type == ixgbe_mac_82598EB)
2569 /* disable the arbiter while setting MTQC */
2570 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2571 rttdcs |= IXGBE_RTTDCS_ARBDIS;
2572 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2574 /* set transmit pool layout */
2575 mask = (IXGBE_FLAG_SRIOV_ENABLED | IXGBE_FLAG_DCB_ENABLED);
2576 switch (adapter->flags & mask) {
2578 case (IXGBE_FLAG_SRIOV_ENABLED):
2579 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2580 (IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF));
2583 case (IXGBE_FLAG_DCB_ENABLED):
2584 /* We enable 8 traffic classes, DCB only */
2585 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2586 (IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ));
2590 IXGBE_WRITE_REG(hw, IXGBE_MTQC, IXGBE_MTQC_64Q_1PB);
2594 /* re-enable the arbiter */
2595 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2596 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2600 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2601 * @adapter: board private structure
2603 * Configure the Tx unit of the MAC after a reset.
2605 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2607 struct ixgbe_hw *hw = &adapter->hw;
2611 ixgbe_setup_mtqc(adapter);
2613 if (hw->mac.type != ixgbe_mac_82598EB) {
2614 /* DMATXCTL.EN must be before Tx queues are enabled */
2615 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2616 dmatxctl |= IXGBE_DMATXCTL_TE;
2617 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2620 /* Setup the HW Tx Head and Tail descriptor pointers */
2621 for (i = 0; i < adapter->num_tx_queues; i++)
2622 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
2625 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2627 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2628 struct ixgbe_ring *rx_ring)
2632 struct ixgbe_ring_feature *feature = adapter->ring_feature;
2634 index = rx_ring->reg_idx;
2635 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2637 mask = (unsigned long) feature[RING_F_RSS].mask;
2638 index = index & mask;
2640 srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(index));
2642 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
2643 srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
2644 if (adapter->num_vfs)
2645 srrctl |= IXGBE_SRRCTL_DROP_EN;
2647 srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
2648 IXGBE_SRRCTL_BSIZEHDR_MASK;
2650 if (ring_is_ps_enabled(rx_ring)) {
2651 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
2652 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2654 srrctl |= (PAGE_SIZE / 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2656 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2658 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
2659 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2660 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2663 IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(index), srrctl);
2666 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2668 struct ixgbe_hw *hw = &adapter->hw;
2669 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2670 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2671 0x6A3E67EA, 0x14364D17, 0x3BED200D};
2672 u32 mrqc = 0, reta = 0;
2677 /* Fill out hash function seeds */
2678 for (i = 0; i < 10; i++)
2679 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2681 /* Fill out redirection table */
2682 for (i = 0, j = 0; i < 128; i++, j++) {
2683 if (j == adapter->ring_feature[RING_F_RSS].indices)
2685 /* reta = 4-byte sliding window of
2686 * 0x00..(indices-1)(indices-1)00..etc. */
2687 reta = (reta << 8) | (j * 0x11);
2689 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2692 /* Disable indicating checksum in descriptor, enables RSS hash */
2693 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2694 rxcsum |= IXGBE_RXCSUM_PCSD;
2695 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2697 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
2698 mask = adapter->flags & IXGBE_FLAG_RSS_ENABLED;
2700 mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2701 #ifdef CONFIG_IXGBE_DCB
2702 | IXGBE_FLAG_DCB_ENABLED
2704 | IXGBE_FLAG_SRIOV_ENABLED
2708 case (IXGBE_FLAG_RSS_ENABLED):
2709 mrqc = IXGBE_MRQC_RSSEN;
2711 case (IXGBE_FLAG_SRIOV_ENABLED):
2712 mrqc = IXGBE_MRQC_VMDQEN;
2714 #ifdef CONFIG_IXGBE_DCB
2715 case (IXGBE_FLAG_DCB_ENABLED):
2716 mrqc = IXGBE_MRQC_RT8TCEN;
2718 #endif /* CONFIG_IXGBE_DCB */
2723 /* Perform hash on these packet types */
2724 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2725 | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2726 | IXGBE_MRQC_RSS_FIELD_IPV6
2727 | IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
2729 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2733 * ixgbe_configure_rscctl - enable RSC for the indicated ring
2734 * @adapter: address of board private structure
2735 * @index: index of ring to set
2737 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
2738 struct ixgbe_ring *ring)
2740 struct ixgbe_hw *hw = &adapter->hw;
2743 u16 reg_idx = ring->reg_idx;
2745 if (!ring_is_rsc_enabled(ring))
2748 rx_buf_len = ring->rx_buf_len;
2749 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
2750 rscctrl |= IXGBE_RSCCTL_RSCEN;
2752 * we must limit the number of descriptors so that the
2753 * total size of max desc * buf_len is not greater
2756 if (ring_is_ps_enabled(ring)) {
2757 #if (MAX_SKB_FRAGS > 16)
2758 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2759 #elif (MAX_SKB_FRAGS > 8)
2760 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2761 #elif (MAX_SKB_FRAGS > 4)
2762 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2764 rscctrl |= IXGBE_RSCCTL_MAXDESC_1;
2767 if (rx_buf_len < IXGBE_RXBUFFER_4096)
2768 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2769 else if (rx_buf_len < IXGBE_RXBUFFER_8192)
2770 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2772 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2774 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
2778 * ixgbe_set_uta - Set unicast filter table address
2779 * @adapter: board private structure
2781 * The unicast table address is a register array of 32-bit registers.
2782 * The table is meant to be used in a way similar to how the MTA is used
2783 * however due to certain limitations in the hardware it is necessary to
2784 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
2785 * enable bit to allow vlan tag stripping when promiscuous mode is enabled
2787 static void ixgbe_set_uta(struct ixgbe_adapter *adapter)
2789 struct ixgbe_hw *hw = &adapter->hw;
2792 /* The UTA table only exists on 82599 hardware and newer */
2793 if (hw->mac.type < ixgbe_mac_82599EB)
2796 /* we only need to do this if VMDq is enabled */
2797 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
2800 for (i = 0; i < 128; i++)
2801 IXGBE_WRITE_REG(hw, IXGBE_UTA(i), ~0);
2804 #define IXGBE_MAX_RX_DESC_POLL 10
2805 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
2806 struct ixgbe_ring *ring)
2808 struct ixgbe_hw *hw = &adapter->hw;
2809 int reg_idx = ring->reg_idx;
2810 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
2813 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2814 if (hw->mac.type == ixgbe_mac_82598EB &&
2815 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2820 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2821 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
2824 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
2825 "the polling period\n", reg_idx);
2829 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
2830 struct ixgbe_ring *ring)
2832 struct ixgbe_hw *hw = &adapter->hw;
2833 u64 rdba = ring->dma;
2835 u16 reg_idx = ring->reg_idx;
2837 /* disable queue to avoid issues while updating state */
2838 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2839 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx),
2840 rxdctl & ~IXGBE_RXDCTL_ENABLE);
2841 IXGBE_WRITE_FLUSH(hw);
2843 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
2844 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
2845 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
2846 ring->count * sizeof(union ixgbe_adv_rx_desc));
2847 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
2848 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
2849 ring->tail = hw->hw_addr + IXGBE_RDT(reg_idx);
2851 ixgbe_configure_srrctl(adapter, ring);
2852 ixgbe_configure_rscctl(adapter, ring);
2854 if (hw->mac.type == ixgbe_mac_82598EB) {
2856 * enable cache line friendly hardware writes:
2857 * PTHRESH=32 descriptors (half the internal cache),
2858 * this also removes ugly rx_no_buffer_count increment
2859 * HTHRESH=4 descriptors (to minimize latency on fetch)
2860 * WTHRESH=8 burst writeback up to two cache lines
2862 rxdctl &= ~0x3FFFFF;
2866 /* enable receive descriptor ring */
2867 rxdctl |= IXGBE_RXDCTL_ENABLE;
2868 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
2870 ixgbe_rx_desc_queue_enable(adapter, ring);
2871 ixgbe_alloc_rx_buffers(ring, IXGBE_DESC_UNUSED(ring));
2874 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
2876 struct ixgbe_hw *hw = &adapter->hw;
2879 /* PSRTYPE must be initialized in non 82598 adapters */
2880 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
2881 IXGBE_PSRTYPE_UDPHDR |
2882 IXGBE_PSRTYPE_IPV4HDR |
2883 IXGBE_PSRTYPE_L2HDR |
2884 IXGBE_PSRTYPE_IPV6HDR;
2886 if (hw->mac.type == ixgbe_mac_82598EB)
2889 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED)
2890 psrtype |= (adapter->num_rx_queues_per_pool << 29);
2892 for (p = 0; p < adapter->num_rx_pools; p++)
2893 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(adapter->num_vfs + p),
2897 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
2899 struct ixgbe_hw *hw = &adapter->hw;
2902 u32 reg_offset, vf_shift;
2905 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
2908 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
2909 vt_reg_bits = IXGBE_VMD_CTL_VMDQ_EN | IXGBE_VT_CTL_REPLEN;
2910 vt_reg_bits |= (adapter->num_vfs << IXGBE_VT_CTL_POOL_SHIFT);
2911 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl | vt_reg_bits);
2913 vf_shift = adapter->num_vfs % 32;
2914 reg_offset = (adapter->num_vfs > 32) ? 1 : 0;
2916 /* Enable only the PF's pool for Tx/Rx */
2917 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (1 << vf_shift));
2918 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), 0);
2919 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (1 << vf_shift));
2920 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), 0);
2921 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
2923 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
2924 hw->mac.ops.set_vmdq(hw, 0, adapter->num_vfs);
2927 * Set up VF register offsets for selected VT Mode,
2928 * i.e. 32 or 64 VFs for SR-IOV
2930 gcr_ext = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
2931 gcr_ext |= IXGBE_GCR_EXT_MSIX_EN;
2932 gcr_ext |= IXGBE_GCR_EXT_VT_MODE_64;
2933 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
2935 /* enable Tx loopback for VF/PF communication */
2936 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
2939 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
2941 struct ixgbe_hw *hw = &adapter->hw;
2942 struct net_device *netdev = adapter->netdev;
2943 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2945 struct ixgbe_ring *rx_ring;
2949 /* Decide whether to use packet split mode or not */
2950 /* Do not use packet split if we're in SR-IOV Mode */
2951 if (!adapter->num_vfs)
2952 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
2954 /* Set the RX buffer length according to the mode */
2955 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
2956 rx_buf_len = IXGBE_RX_HDR_SIZE;
2958 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
2959 (netdev->mtu <= ETH_DATA_LEN))
2960 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
2962 rx_buf_len = ALIGN(max_frame + VLAN_HLEN, 1024);
2966 /* adjust max frame to be able to do baby jumbo for FCoE */
2967 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
2968 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
2969 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
2971 #endif /* IXGBE_FCOE */
2972 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
2973 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
2974 mhadd &= ~IXGBE_MHADD_MFS_MASK;
2975 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
2977 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
2980 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2981 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
2982 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2983 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
2986 * Setup the HW Rx Head and Tail Descriptor Pointers and
2987 * the Base and Length of the Rx Descriptor Ring
2989 for (i = 0; i < adapter->num_rx_queues; i++) {
2990 rx_ring = adapter->rx_ring[i];
2991 rx_ring->rx_buf_len = rx_buf_len;
2993 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)
2994 set_ring_ps_enabled(rx_ring);
2996 clear_ring_ps_enabled(rx_ring);
2998 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
2999 set_ring_rsc_enabled(rx_ring);
3001 clear_ring_rsc_enabled(rx_ring);
3004 if (netdev->features & NETIF_F_FCOE_MTU) {
3005 struct ixgbe_ring_feature *f;
3006 f = &adapter->ring_feature[RING_F_FCOE];
3007 if ((i >= f->mask) && (i < f->mask + f->indices)) {
3008 clear_ring_ps_enabled(rx_ring);
3009 if (rx_buf_len < IXGBE_FCOE_JUMBO_FRAME_SIZE)
3010 rx_ring->rx_buf_len =
3011 IXGBE_FCOE_JUMBO_FRAME_SIZE;
3012 } else if (!ring_is_rsc_enabled(rx_ring) &&
3013 !ring_is_ps_enabled(rx_ring)) {
3014 rx_ring->rx_buf_len =
3015 IXGBE_FCOE_JUMBO_FRAME_SIZE;
3018 #endif /* IXGBE_FCOE */
3023 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
3025 struct ixgbe_hw *hw = &adapter->hw;
3026 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
3028 switch (hw->mac.type) {
3029 case ixgbe_mac_82598EB:
3031 * For VMDq support of different descriptor types or
3032 * buffer sizes through the use of multiple SRRCTL
3033 * registers, RDRXCTL.MVMEN must be set to 1
3035 * also, the manual doesn't mention it clearly but DCA hints
3036 * will only use queue 0's tags unless this bit is set. Side
3037 * effects of setting this bit are only that SRRCTL must be
3038 * fully programmed [0..15]
3040 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
3042 case ixgbe_mac_82599EB:
3043 /* Disable RSC for ACK packets */
3044 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
3045 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
3046 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3047 /* hardware requires some bits to be set by default */
3048 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3049 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3052 /* We should do nothing since we don't know this hardware */
3056 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3060 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3061 * @adapter: board private structure
3063 * Configure the Rx unit of the MAC after a reset.
3065 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3067 struct ixgbe_hw *hw = &adapter->hw;
3071 /* disable receives while setting up the descriptors */
3072 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3073 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3075 ixgbe_setup_psrtype(adapter);
3076 ixgbe_setup_rdrxctl(adapter);
3078 /* Program registers for the distribution of queues */
3079 ixgbe_setup_mrqc(adapter);
3081 ixgbe_set_uta(adapter);
3083 /* set_rx_buffer_len must be called before ring initialization */
3084 ixgbe_set_rx_buffer_len(adapter);
3087 * Setup the HW Rx Head and Tail Descriptor Pointers and
3088 * the Base and Length of the Rx Descriptor Ring
3090 for (i = 0; i < adapter->num_rx_queues; i++)
3091 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3093 /* disable drop enable for 82598 parts */
3094 if (hw->mac.type == ixgbe_mac_82598EB)
3095 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3097 /* enable all receives */
3098 rxctrl |= IXGBE_RXCTRL_RXEN;
3099 hw->mac.ops.enable_rx_dma(hw, rxctrl);
3102 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3104 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3105 struct ixgbe_hw *hw = &adapter->hw;
3106 int pool_ndx = adapter->num_vfs;
3108 /* add VID to filter table */
3109 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, true);
3110 set_bit(vid, adapter->active_vlans);
3113 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3115 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3116 struct ixgbe_hw *hw = &adapter->hw;
3117 int pool_ndx = adapter->num_vfs;
3119 /* remove VID from filter table */
3120 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, false);
3121 clear_bit(vid, adapter->active_vlans);
3125 * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3126 * @adapter: driver data
3128 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
3130 struct ixgbe_hw *hw = &adapter->hw;
3133 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3134 vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
3135 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3139 * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3140 * @adapter: driver data
3142 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
3144 struct ixgbe_hw *hw = &adapter->hw;
3147 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3148 vlnctrl |= IXGBE_VLNCTRL_VFE;
3149 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3150 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3154 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3155 * @adapter: driver data
3157 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3159 struct ixgbe_hw *hw = &adapter->hw;
3163 switch (hw->mac.type) {
3164 case ixgbe_mac_82598EB:
3165 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3166 vlnctrl &= ~IXGBE_VLNCTRL_VME;
3167 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3169 case ixgbe_mac_82599EB:
3170 for (i = 0; i < adapter->num_rx_queues; i++) {
3171 j = adapter->rx_ring[i]->reg_idx;
3172 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3173 vlnctrl &= ~IXGBE_RXDCTL_VME;
3174 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3183 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3184 * @adapter: driver data
3186 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3188 struct ixgbe_hw *hw = &adapter->hw;
3192 switch (hw->mac.type) {
3193 case ixgbe_mac_82598EB:
3194 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3195 vlnctrl |= IXGBE_VLNCTRL_VME;
3196 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3198 case ixgbe_mac_82599EB:
3199 for (i = 0; i < adapter->num_rx_queues; i++) {
3200 j = adapter->rx_ring[i]->reg_idx;
3201 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3202 vlnctrl |= IXGBE_RXDCTL_VME;
3203 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3211 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3215 ixgbe_vlan_rx_add_vid(adapter->netdev, 0);
3217 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3218 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
3222 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3223 * @netdev: network interface device structure
3225 * Writes unicast address list to the RAR table.
3226 * Returns: -ENOMEM on failure/insufficient address space
3227 * 0 on no addresses written
3228 * X on writing X addresses to the RAR table
3230 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3232 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3233 struct ixgbe_hw *hw = &adapter->hw;
3234 unsigned int vfn = adapter->num_vfs;
3235 unsigned int rar_entries = hw->mac.num_rar_entries - (vfn + 1);
3238 /* return ENOMEM indicating insufficient memory for addresses */
3239 if (netdev_uc_count(netdev) > rar_entries)
3242 if (!netdev_uc_empty(netdev) && rar_entries) {
3243 struct netdev_hw_addr *ha;
3244 /* return error if we do not support writing to RAR table */
3245 if (!hw->mac.ops.set_rar)
3248 netdev_for_each_uc_addr(ha, netdev) {
3251 hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3256 /* write the addresses in reverse order to avoid write combining */
3257 for (; rar_entries > 0 ; rar_entries--)
3258 hw->mac.ops.clear_rar(hw, rar_entries);
3264 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3265 * @netdev: network interface device structure
3267 * The set_rx_method entry point is called whenever the unicast/multicast
3268 * address list or the network interface flags are updated. This routine is
3269 * responsible for configuring the hardware for proper unicast, multicast and
3272 void ixgbe_set_rx_mode(struct net_device *netdev)
3274 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3275 struct ixgbe_hw *hw = &adapter->hw;
3276 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3279 /* Check for Promiscuous and All Multicast modes */
3281 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3283 /* set all bits that we expect to always be set */
3284 fctrl |= IXGBE_FCTRL_BAM;
3285 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3286 fctrl |= IXGBE_FCTRL_PMCF;
3288 /* clear the bits we are changing the status of */
3289 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3291 if (netdev->flags & IFF_PROMISC) {
3292 hw->addr_ctrl.user_set_promisc = true;
3293 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3294 vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3295 /* don't hardware filter vlans in promisc mode */
3296 ixgbe_vlan_filter_disable(adapter);
3298 if (netdev->flags & IFF_ALLMULTI) {
3299 fctrl |= IXGBE_FCTRL_MPE;
3300 vmolr |= IXGBE_VMOLR_MPE;
3303 * Write addresses to the MTA, if the attempt fails
3304 * then we should just turn on promiscous mode so
3305 * that we can at least receive multicast traffic
3307 hw->mac.ops.update_mc_addr_list(hw, netdev);
3308 vmolr |= IXGBE_VMOLR_ROMPE;
3310 ixgbe_vlan_filter_enable(adapter);
3311 hw->addr_ctrl.user_set_promisc = false;
3313 * Write addresses to available RAR registers, if there is not
3314 * sufficient space to store all the addresses then enable
3315 * unicast promiscous mode
3317 count = ixgbe_write_uc_addr_list(netdev);
3319 fctrl |= IXGBE_FCTRL_UPE;
3320 vmolr |= IXGBE_VMOLR_ROPE;
3324 if (adapter->num_vfs) {
3325 ixgbe_restore_vf_multicasts(adapter);
3326 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(adapter->num_vfs)) &
3327 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3329 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(adapter->num_vfs), vmolr);
3332 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3334 if (netdev->features & NETIF_F_HW_VLAN_RX)
3335 ixgbe_vlan_strip_enable(adapter);
3337 ixgbe_vlan_strip_disable(adapter);
3340 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3343 struct ixgbe_q_vector *q_vector;
3344 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3346 /* legacy and MSI only use one vector */
3347 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3350 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3351 struct napi_struct *napi;
3352 q_vector = adapter->q_vector[q_idx];
3353 napi = &q_vector->napi;
3354 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3355 if (!q_vector->rxr_count || !q_vector->txr_count) {
3356 if (q_vector->txr_count == 1)
3357 napi->poll = &ixgbe_clean_txonly;
3358 else if (q_vector->rxr_count == 1)
3359 napi->poll = &ixgbe_clean_rxonly;
3367 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3370 struct ixgbe_q_vector *q_vector;
3371 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3373 /* legacy and MSI only use one vector */
3374 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3377 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3378 q_vector = adapter->q_vector[q_idx];
3379 napi_disable(&q_vector->napi);
3383 #ifdef CONFIG_IXGBE_DCB
3385 * ixgbe_configure_dcb - Configure DCB hardware
3386 * @adapter: ixgbe adapter struct
3388 * This is called by the driver on open to configure the DCB hardware.
3389 * This is also called by the gennetlink interface when reconfiguring
3392 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3394 struct ixgbe_hw *hw = &adapter->hw;
3395 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3397 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3398 if (hw->mac.type == ixgbe_mac_82598EB)
3399 netif_set_gso_max_size(adapter->netdev, 65536);
3403 if (hw->mac.type == ixgbe_mac_82598EB)
3404 netif_set_gso_max_size(adapter->netdev, 32768);
3407 if (adapter->netdev->features & NETIF_F_FCOE_MTU)
3408 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
3411 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3413 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3416 /* Enable VLAN tag insert/strip */
3417 adapter->netdev->features |= NETIF_F_HW_VLAN_RX;
3419 hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
3421 /* reconfigure the hardware */
3422 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
3426 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3428 struct net_device *netdev = adapter->netdev;
3429 struct ixgbe_hw *hw = &adapter->hw;
3432 #ifdef CONFIG_IXGBE_DCB
3433 ixgbe_configure_dcb(adapter);
3436 ixgbe_set_rx_mode(netdev);
3437 ixgbe_restore_vlan(adapter);
3440 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
3441 ixgbe_configure_fcoe(adapter);
3443 #endif /* IXGBE_FCOE */
3444 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3445 for (i = 0; i < adapter->num_tx_queues; i++)
3446 adapter->tx_ring[i]->atr_sample_rate =
3447 adapter->atr_sample_rate;
3448 ixgbe_init_fdir_signature_82599(hw, adapter->fdir_pballoc);
3449 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3450 ixgbe_init_fdir_perfect_82599(hw, adapter->fdir_pballoc);
3452 ixgbe_configure_virtualization(adapter);
3454 ixgbe_configure_tx(adapter);
3455 ixgbe_configure_rx(adapter);
3458 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3460 switch (hw->phy.type) {
3461 case ixgbe_phy_sfp_avago:
3462 case ixgbe_phy_sfp_ftl:
3463 case ixgbe_phy_sfp_intel:
3464 case ixgbe_phy_sfp_unknown:
3465 case ixgbe_phy_sfp_passive_tyco:
3466 case ixgbe_phy_sfp_passive_unknown:
3467 case ixgbe_phy_sfp_active_unknown:
3468 case ixgbe_phy_sfp_ftl_active:
3476 * ixgbe_sfp_link_config - set up SFP+ link
3477 * @adapter: pointer to private adapter struct
3479 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3481 struct ixgbe_hw *hw = &adapter->hw;
3483 if (hw->phy.multispeed_fiber) {
3485 * In multispeed fiber setups, the device may not have
3486 * had a physical connection when the driver loaded.
3487 * If that's the case, the initial link configuration
3488 * couldn't get the MAC into 10G or 1G mode, so we'll
3489 * never have a link status change interrupt fire.
3490 * We need to try and force an autonegotiation
3491 * session, then bring up link.
3493 hw->mac.ops.setup_sfp(hw);
3494 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
3495 schedule_work(&adapter->multispeed_fiber_task);
3498 * Direct Attach Cu and non-multispeed fiber modules
3499 * still need to be configured properly prior to
3502 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_MOD_TASK))
3503 schedule_work(&adapter->sfp_config_module_task);
3508 * ixgbe_non_sfp_link_config - set up non-SFP+ link
3509 * @hw: pointer to private hardware struct
3511 * Returns 0 on success, negative on failure
3513 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3516 bool negotiation, link_up = false;
3517 u32 ret = IXGBE_ERR_LINK_SETUP;
3519 if (hw->mac.ops.check_link)
3520 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3525 if (hw->mac.ops.get_link_capabilities)
3526 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
3531 if (hw->mac.ops.setup_link)
3532 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
3537 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
3539 struct ixgbe_hw *hw = &adapter->hw;
3542 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3543 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
3545 gpie |= IXGBE_GPIE_EIAME;
3547 * use EIAM to auto-mask when MSI-X interrupt is asserted
3548 * this saves a register write for every interrupt
3550 switch (hw->mac.type) {
3551 case ixgbe_mac_82598EB:
3552 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3555 case ixgbe_mac_82599EB:
3556 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
3557 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
3561 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
3562 * specifically only auto mask tx and rx interrupts */
3563 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3566 /* XXX: to interrupt immediately for EICS writes, enable this */
3567 /* gpie |= IXGBE_GPIE_EIMEN; */
3569 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3570 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
3571 gpie |= IXGBE_GPIE_VTMODE_64;
3574 /* Enable fan failure interrupt */
3575 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3576 gpie |= IXGBE_SDP1_GPIEN;
3578 if (hw->mac.type == ixgbe_mac_82599EB)
3579 gpie |= IXGBE_SDP1_GPIEN;
3580 gpie |= IXGBE_SDP2_GPIEN;
3582 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3585 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
3587 struct ixgbe_hw *hw = &adapter->hw;
3591 ixgbe_get_hw_control(adapter);
3592 ixgbe_setup_gpie(adapter);
3594 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3595 ixgbe_configure_msix(adapter);
3597 ixgbe_configure_msi_and_legacy(adapter);
3599 /* enable the optics */
3600 if (hw->phy.multispeed_fiber)
3601 hw->mac.ops.enable_tx_laser(hw);
3603 clear_bit(__IXGBE_DOWN, &adapter->state);
3604 ixgbe_napi_enable_all(adapter);
3606 if (ixgbe_is_sfp(hw)) {
3607 ixgbe_sfp_link_config(adapter);
3609 err = ixgbe_non_sfp_link_config(hw);
3611 e_err(probe, "link_config FAILED %d\n", err);
3614 /* clear any pending interrupts, may auto mask */
3615 IXGBE_READ_REG(hw, IXGBE_EICR);
3616 ixgbe_irq_enable(adapter, true, true);
3619 * If this adapter has a fan, check to see if we had a failure
3620 * before we enabled the interrupt.
3622 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3623 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
3624 if (esdp & IXGBE_ESDP_SDP1)
3625 e_crit(drv, "Fan has stopped, replace the adapter\n");
3629 * For hot-pluggable SFP+ devices, a new SFP+ module may have
3630 * arrived before interrupts were enabled but after probe. Such
3631 * devices wouldn't have their type identified yet. We need to
3632 * kick off the SFP+ module setup first, then try to bring up link.
3633 * If we're not hot-pluggable SFP+, we just need to configure link
3636 if (hw->phy.type == ixgbe_phy_unknown)
3637 schedule_work(&adapter->sfp_config_module_task);
3639 /* enable transmits */
3640 netif_tx_start_all_queues(adapter->netdev);
3642 /* bring the link up in the watchdog, this could race with our first
3643 * link up interrupt but shouldn't be a problem */
3644 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3645 adapter->link_check_timeout = jiffies;
3646 mod_timer(&adapter->watchdog_timer, jiffies);
3648 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
3649 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
3650 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
3651 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
3656 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
3658 WARN_ON(in_interrupt());
3659 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
3661 ixgbe_down(adapter);
3663 * If SR-IOV enabled then wait a bit before bringing the adapter
3664 * back up to give the VFs time to respond to the reset. The
3665 * two second wait is based upon the watchdog timer cycle in
3668 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3671 clear_bit(__IXGBE_RESETTING, &adapter->state);
3674 int ixgbe_up(struct ixgbe_adapter *adapter)
3676 /* hardware has been reset, we need to reload some things */
3677 ixgbe_configure(adapter);
3679 return ixgbe_up_complete(adapter);
3682 void ixgbe_reset(struct ixgbe_adapter *adapter)
3684 struct ixgbe_hw *hw = &adapter->hw;
3687 err = hw->mac.ops.init_hw(hw);
3690 case IXGBE_ERR_SFP_NOT_PRESENT:
3692 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
3693 e_dev_err("master disable timed out\n");
3695 case IXGBE_ERR_EEPROM_VERSION:
3696 /* We are running on a pre-production device, log a warning */
3697 e_dev_warn("This device is a pre-production adapter/LOM. "
3698 "Please be aware there may be issuesassociated with "
3699 "your hardware. If you are experiencing problems "
3700 "please contact your Intel or hardware "
3701 "representative who provided you with this "
3705 e_dev_err("Hardware Error: %d\n", err);
3708 /* reprogram the RAR[0] in case user changed it. */
3709 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
3714 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
3715 * @rx_ring: ring to free buffers from
3717 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
3719 struct device *dev = rx_ring->dev;
3723 /* ring already cleared, nothing to do */
3724 if (!rx_ring->rx_buffer_info)
3727 /* Free all the Rx ring sk_buffs */
3728 for (i = 0; i < rx_ring->count; i++) {
3729 struct ixgbe_rx_buffer *rx_buffer_info;
3731 rx_buffer_info = &rx_ring->rx_buffer_info[i];
3732 if (rx_buffer_info->dma) {
3733 dma_unmap_single(rx_ring->dev, rx_buffer_info->dma,
3734 rx_ring->rx_buf_len,
3736 rx_buffer_info->dma = 0;
3738 if (rx_buffer_info->skb) {
3739 struct sk_buff *skb = rx_buffer_info->skb;
3740 rx_buffer_info->skb = NULL;
3742 struct sk_buff *this = skb;
3743 if (IXGBE_RSC_CB(this)->delay_unmap) {
3744 dma_unmap_single(dev,
3745 IXGBE_RSC_CB(this)->dma,
3746 rx_ring->rx_buf_len,
3748 IXGBE_RSC_CB(this)->dma = 0;
3749 IXGBE_RSC_CB(skb)->delay_unmap = false;
3752 dev_kfree_skb(this);
3755 if (!rx_buffer_info->page)
3757 if (rx_buffer_info->page_dma) {
3758 dma_unmap_page(dev, rx_buffer_info->page_dma,
3759 PAGE_SIZE / 2, DMA_FROM_DEVICE);
3760 rx_buffer_info->page_dma = 0;
3762 put_page(rx_buffer_info->page);
3763 rx_buffer_info->page = NULL;
3764 rx_buffer_info->page_offset = 0;
3767 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
3768 memset(rx_ring->rx_buffer_info, 0, size);
3770 /* Zero out the descriptor ring */
3771 memset(rx_ring->desc, 0, rx_ring->size);
3773 rx_ring->next_to_clean = 0;
3774 rx_ring->next_to_use = 0;
3778 * ixgbe_clean_tx_ring - Free Tx Buffers
3779 * @tx_ring: ring to be cleaned
3781 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
3783 struct ixgbe_tx_buffer *tx_buffer_info;
3787 /* ring already cleared, nothing to do */
3788 if (!tx_ring->tx_buffer_info)
3791 /* Free all the Tx ring sk_buffs */
3792 for (i = 0; i < tx_ring->count; i++) {
3793 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3794 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
3797 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
3798 memset(tx_ring->tx_buffer_info, 0, size);
3800 /* Zero out the descriptor ring */
3801 memset(tx_ring->desc, 0, tx_ring->size);
3803 tx_ring->next_to_use = 0;
3804 tx_ring->next_to_clean = 0;
3808 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
3809 * @adapter: board private structure
3811 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
3815 for (i = 0; i < adapter->num_rx_queues; i++)
3816 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
3820 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
3821 * @adapter: board private structure
3823 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
3827 for (i = 0; i < adapter->num_tx_queues; i++)
3828 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
3831 void ixgbe_down(struct ixgbe_adapter *adapter)
3833 struct net_device *netdev = adapter->netdev;
3834 struct ixgbe_hw *hw = &adapter->hw;
3838 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3840 /* signal that we are down to the interrupt handler */
3841 set_bit(__IXGBE_DOWN, &adapter->state);
3843 /* disable receive for all VFs and wait one second */
3844 if (adapter->num_vfs) {
3845 /* ping all the active vfs to let them know we are going down */
3846 ixgbe_ping_all_vfs(adapter);
3848 /* Disable all VFTE/VFRE TX/RX */
3849 ixgbe_disable_tx_rx(adapter);
3851 /* Mark all the VFs as inactive */
3852 for (i = 0 ; i < adapter->num_vfs; i++)
3853 adapter->vfinfo[i].clear_to_send = 0;
3856 /* disable receives */
3857 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3858 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3860 IXGBE_WRITE_FLUSH(hw);
3863 netif_tx_stop_all_queues(netdev);
3865 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
3866 del_timer_sync(&adapter->sfp_timer);
3867 del_timer_sync(&adapter->watchdog_timer);
3868 cancel_work_sync(&adapter->watchdog_task);
3870 netif_carrier_off(netdev);
3871 netif_tx_disable(netdev);
3873 ixgbe_irq_disable(adapter);
3875 ixgbe_napi_disable_all(adapter);
3877 /* Cleanup the affinity_hint CPU mask memory and callback */
3878 for (i = 0; i < num_q_vectors; i++) {
3879 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
3880 /* clear the affinity_mask in the IRQ descriptor */
3881 irq_set_affinity_hint(adapter->msix_entries[i]. vector, NULL);
3882 /* release the CPU mask memory */
3883 free_cpumask_var(q_vector->affinity_mask);
3886 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3887 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3888 cancel_work_sync(&adapter->fdir_reinit_task);
3890 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
3891 cancel_work_sync(&adapter->check_overtemp_task);
3893 /* disable transmits in the hardware now that interrupts are off */
3894 for (i = 0; i < adapter->num_tx_queues; i++) {
3895 j = adapter->tx_ring[i]->reg_idx;
3896 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3897 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j),
3898 (txdctl & ~IXGBE_TXDCTL_ENABLE));
3900 /* Disable the Tx DMA engine on 82599 */
3901 if (hw->mac.type == ixgbe_mac_82599EB)
3902 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
3903 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
3904 ~IXGBE_DMATXCTL_TE));
3906 /* power down the optics */
3907 if (hw->phy.multispeed_fiber)
3908 hw->mac.ops.disable_tx_laser(hw);
3910 /* clear n-tuple filters that are cached */
3911 ethtool_ntuple_flush(netdev);
3913 if (!pci_channel_offline(adapter->pdev))
3914 ixgbe_reset(adapter);
3915 ixgbe_clean_all_tx_rings(adapter);
3916 ixgbe_clean_all_rx_rings(adapter);
3918 #ifdef CONFIG_IXGBE_DCA
3919 /* since we reset the hardware DCA settings were cleared */
3920 ixgbe_setup_dca(adapter);
3925 * ixgbe_poll - NAPI Rx polling callback
3926 * @napi: structure for representing this polling device
3927 * @budget: how many packets driver is allowed to clean
3929 * This function is used for legacy and MSI, NAPI mode
3931 static int ixgbe_poll(struct napi_struct *napi, int budget)
3933 struct ixgbe_q_vector *q_vector =
3934 container_of(napi, struct ixgbe_q_vector, napi);
3935 struct ixgbe_adapter *adapter = q_vector->adapter;
3936 int tx_clean_complete, work_done = 0;
3938 #ifdef CONFIG_IXGBE_DCA
3939 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
3940 ixgbe_update_dca(q_vector);
3943 tx_clean_complete = ixgbe_clean_tx_irq(q_vector, adapter->tx_ring[0]);
3944 ixgbe_clean_rx_irq(q_vector, adapter->rx_ring[0], &work_done, budget);
3946 if (!tx_clean_complete)
3949 /* If budget not fully consumed, exit the polling mode */
3950 if (work_done < budget) {
3951 napi_complete(napi);
3952 if (adapter->rx_itr_setting & 1)
3953 ixgbe_set_itr(adapter);
3954 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3955 ixgbe_irq_enable_queues(adapter, IXGBE_EIMS_RTX_QUEUE);
3961 * ixgbe_tx_timeout - Respond to a Tx Hang
3962 * @netdev: network interface device structure
3964 static void ixgbe_tx_timeout(struct net_device *netdev)
3966 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3968 /* Do the reset outside of interrupt context */
3969 schedule_work(&adapter->reset_task);
3972 static void ixgbe_reset_task(struct work_struct *work)
3974 struct ixgbe_adapter *adapter;
3975 adapter = container_of(work, struct ixgbe_adapter, reset_task);
3977 /* If we're already down or resetting, just bail */
3978 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
3979 test_bit(__IXGBE_RESETTING, &adapter->state))
3982 adapter->tx_timeout_count++;
3984 ixgbe_dump(adapter);
3985 netdev_err(adapter->netdev, "Reset adapter\n");
3986 ixgbe_reinit_locked(adapter);
3989 #ifdef CONFIG_IXGBE_DCB
3990 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
3993 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_DCB];
3995 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
3999 adapter->num_rx_queues = f->indices;
4000 adapter->num_tx_queues = f->indices;
4008 * ixgbe_set_rss_queues: Allocate queues for RSS
4009 * @adapter: board private structure to initialize
4011 * This is our "base" multiqueue mode. RSS (Receive Side Scaling) will try
4012 * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
4015 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
4018 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
4020 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4022 adapter->num_rx_queues = f->indices;
4023 adapter->num_tx_queues = f->indices;
4033 * ixgbe_set_fdir_queues: Allocate queues for Flow Director
4034 * @adapter: board private structure to initialize
4036 * Flow Director is an advanced Rx filter, attempting to get Rx flows back
4037 * to the original CPU that initiated the Tx session. This runs in addition
4038 * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
4039 * Rx load across CPUs using RSS.
4042 static inline bool ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
4045 struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
4047 f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
4050 /* Flow Director must have RSS enabled */
4051 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
4052 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
4053 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)))) {
4054 adapter->num_tx_queues = f_fdir->indices;
4055 adapter->num_rx_queues = f_fdir->indices;
4058 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4059 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4066 * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
4067 * @adapter: board private structure to initialize
4069 * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
4070 * The ring feature mask is not used as a mask for FCoE, as it can take any 8
4071 * rx queues out of the max number of rx queues, instead, it is used as the
4072 * index of the first rx queue used by FCoE.
4075 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
4078 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4080 f->indices = min((int)num_online_cpus(), f->indices);
4081 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
4082 adapter->num_rx_queues = 1;
4083 adapter->num_tx_queues = 1;
4084 #ifdef CONFIG_IXGBE_DCB
4085 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4086 e_info(probe, "FCoE enabled with DCB\n");
4087 ixgbe_set_dcb_queues(adapter);
4090 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4091 e_info(probe, "FCoE enabled with RSS\n");
4092 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4093 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
4094 ixgbe_set_fdir_queues(adapter);
4096 ixgbe_set_rss_queues(adapter);
4098 /* adding FCoE rx rings to the end */
4099 f->mask = adapter->num_rx_queues;
4100 adapter->num_rx_queues += f->indices;
4101 adapter->num_tx_queues += f->indices;
4109 #endif /* IXGBE_FCOE */
4111 * ixgbe_set_sriov_queues: Allocate queues for IOV use
4112 * @adapter: board private structure to initialize
4114 * IOV doesn't actually use anything, so just NAK the
4115 * request for now and let the other queue routines
4116 * figure out what to do.
4118 static inline bool ixgbe_set_sriov_queues(struct ixgbe_adapter *adapter)
4124 * ixgbe_set_num_queues: Allocate queues for device, feature dependant
4125 * @adapter: board private structure to initialize
4127 * This is the top level queue allocation routine. The order here is very
4128 * important, starting with the "most" number of features turned on at once,
4129 * and ending with the smallest set of features. This way large combinations
4130 * can be allocated if they're turned on, and smaller combinations are the
4131 * fallthrough conditions.
4134 static int ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
4136 /* Start with base case */
4137 adapter->num_rx_queues = 1;
4138 adapter->num_tx_queues = 1;
4139 adapter->num_rx_pools = adapter->num_rx_queues;
4140 adapter->num_rx_queues_per_pool = 1;
4142 if (ixgbe_set_sriov_queues(adapter))
4146 if (ixgbe_set_fcoe_queues(adapter))
4149 #endif /* IXGBE_FCOE */
4150 #ifdef CONFIG_IXGBE_DCB
4151 if (ixgbe_set_dcb_queues(adapter))
4155 if (ixgbe_set_fdir_queues(adapter))
4158 if (ixgbe_set_rss_queues(adapter))
4161 /* fallback to base case */
4162 adapter->num_rx_queues = 1;
4163 adapter->num_tx_queues = 1;
4166 /* Notify the stack of the (possibly) reduced queue counts. */
4167 netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
4168 return netif_set_real_num_rx_queues(adapter->netdev,
4169 adapter->num_rx_queues);
4172 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
4175 int err, vector_threshold;
4177 /* We'll want at least 3 (vector_threshold):
4180 * 3) Other (Link Status Change, etc.)
4181 * 4) TCP Timer (optional)
4183 vector_threshold = MIN_MSIX_COUNT;
4185 /* The more we get, the more we will assign to Tx/Rx Cleanup
4186 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
4187 * Right now, we simply care about how many we'll get; we'll
4188 * set them up later while requesting irq's.
4190 while (vectors >= vector_threshold) {
4191 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
4193 if (!err) /* Success in acquiring all requested vectors. */
4196 vectors = 0; /* Nasty failure, quit now */
4197 else /* err == number of vectors we should try again with */
4201 if (vectors < vector_threshold) {
4202 /* Can't allocate enough MSI-X interrupts? Oh well.
4203 * This just means we'll go with either a single MSI
4204 * vector or fall back to legacy interrupts.
4206 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4207 "Unable to allocate MSI-X interrupts\n");
4208 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4209 kfree(adapter->msix_entries);
4210 adapter->msix_entries = NULL;
4212 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
4214 * Adjust for only the vectors we'll use, which is minimum
4215 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
4216 * vectors we were allocated.
4218 adapter->num_msix_vectors = min(vectors,
4219 adapter->max_msix_q_vectors + NON_Q_VECTORS);
4224 * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
4225 * @adapter: board private structure to initialize
4227 * Cache the descriptor ring offsets for RSS to the assigned rings.
4230 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
4235 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4236 for (i = 0; i < adapter->num_rx_queues; i++)
4237 adapter->rx_ring[i]->reg_idx = i;
4238 for (i = 0; i < adapter->num_tx_queues; i++)
4239 adapter->tx_ring[i]->reg_idx = i;
4248 #ifdef CONFIG_IXGBE_DCB
4250 * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
4251 * @adapter: board private structure to initialize
4253 * Cache the descriptor ring offsets for DCB to the assigned rings.
4256 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
4260 int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
4262 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4263 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
4264 /* the number of queues is assumed to be symmetric */
4265 for (i = 0; i < dcb_i; i++) {
4266 adapter->rx_ring[i]->reg_idx = i << 3;
4267 adapter->tx_ring[i]->reg_idx = i << 2;
4270 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
4273 * Tx TC0 starts at: descriptor queue 0
4274 * Tx TC1 starts at: descriptor queue 32
4275 * Tx TC2 starts at: descriptor queue 64
4276 * Tx TC3 starts at: descriptor queue 80
4277 * Tx TC4 starts at: descriptor queue 96
4278 * Tx TC5 starts at: descriptor queue 104
4279 * Tx TC6 starts at: descriptor queue 112
4280 * Tx TC7 starts at: descriptor queue 120
4282 * Rx TC0-TC7 are offset by 16 queues each
4284 for (i = 0; i < 3; i++) {
4285 adapter->tx_ring[i]->reg_idx = i << 5;
4286 adapter->rx_ring[i]->reg_idx = i << 4;
4288 for ( ; i < 5; i++) {
4289 adapter->tx_ring[i]->reg_idx =
4291 adapter->rx_ring[i]->reg_idx = i << 4;
4293 for ( ; i < dcb_i; i++) {
4294 adapter->tx_ring[i]->reg_idx =
4296 adapter->rx_ring[i]->reg_idx = i << 4;
4300 } else if (dcb_i == 4) {
4302 * Tx TC0 starts at: descriptor queue 0
4303 * Tx TC1 starts at: descriptor queue 64
4304 * Tx TC2 starts at: descriptor queue 96
4305 * Tx TC3 starts at: descriptor queue 112
4307 * Rx TC0-TC3 are offset by 32 queues each
4309 adapter->tx_ring[0]->reg_idx = 0;
4310 adapter->tx_ring[1]->reg_idx = 64;
4311 adapter->tx_ring[2]->reg_idx = 96;
4312 adapter->tx_ring[3]->reg_idx = 112;
4313 for (i = 0 ; i < dcb_i; i++)
4314 adapter->rx_ring[i]->reg_idx = i << 5;
4332 * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
4333 * @adapter: board private structure to initialize
4335 * Cache the descriptor ring offsets for Flow Director to the assigned rings.
4338 static inline bool ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
4343 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
4344 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4345 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))) {
4346 for (i = 0; i < adapter->num_rx_queues; i++)
4347 adapter->rx_ring[i]->reg_idx = i;
4348 for (i = 0; i < adapter->num_tx_queues; i++)
4349 adapter->tx_ring[i]->reg_idx = i;
4358 * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
4359 * @adapter: board private structure to initialize
4361 * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
4364 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
4366 int i, fcoe_rx_i = 0, fcoe_tx_i = 0;
4368 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4370 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
4371 #ifdef CONFIG_IXGBE_DCB
4372 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4373 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
4375 ixgbe_cache_ring_dcb(adapter);
4376 /* find out queues in TC for FCoE */
4377 fcoe_rx_i = adapter->rx_ring[fcoe->tc]->reg_idx + 1;
4378 fcoe_tx_i = adapter->tx_ring[fcoe->tc]->reg_idx + 1;
4380 * In 82599, the number of Tx queues for each traffic
4381 * class for both 8-TC and 4-TC modes are:
4382 * TCs : TC0 TC1 TC2 TC3 TC4 TC5 TC6 TC7
4383 * 8 TCs: 32 32 16 16 8 8 8 8
4384 * 4 TCs: 64 64 32 32
4385 * We have max 8 queues for FCoE, where 8 the is
4386 * FCoE redirection table size. If TC for FCoE is
4387 * less than or equal to TC3, we have enough queues
4388 * to add max of 8 queues for FCoE, so we start FCoE
4389 * tx descriptor from the next one, i.e., reg_idx + 1.
4390 * If TC for FCoE is above TC3, implying 8 TC mode,
4391 * and we need 8 for FCoE, we have to take all queues
4392 * in that traffic class for FCoE.
4394 if ((f->indices == IXGBE_FCRETA_SIZE) && (fcoe->tc > 3))
4397 #endif /* CONFIG_IXGBE_DCB */
4398 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4399 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4400 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
4401 ixgbe_cache_ring_fdir(adapter);
4403 ixgbe_cache_ring_rss(adapter);
4405 fcoe_rx_i = f->mask;
4406 fcoe_tx_i = f->mask;
4408 for (i = 0; i < f->indices; i++, fcoe_rx_i++, fcoe_tx_i++) {
4409 adapter->rx_ring[f->mask + i]->reg_idx = fcoe_rx_i;
4410 adapter->tx_ring[f->mask + i]->reg_idx = fcoe_tx_i;
4417 #endif /* IXGBE_FCOE */
4419 * ixgbe_cache_ring_sriov - Descriptor ring to register mapping for sriov
4420 * @adapter: board private structure to initialize
4422 * SR-IOV doesn't use any descriptor rings but changes the default if
4423 * no other mapping is used.
4426 static inline bool ixgbe_cache_ring_sriov(struct ixgbe_adapter *adapter)
4428 adapter->rx_ring[0]->reg_idx = adapter->num_vfs * 2;
4429 adapter->tx_ring[0]->reg_idx = adapter->num_vfs * 2;
4430 if (adapter->num_vfs)
4437 * ixgbe_cache_ring_register - Descriptor ring to register mapping
4438 * @adapter: board private structure to initialize
4440 * Once we know the feature-set enabled for the device, we'll cache
4441 * the register offset the descriptor ring is assigned to.
4443 * Note, the order the various feature calls is important. It must start with
4444 * the "most" features enabled at the same time, then trickle down to the
4445 * least amount of features turned on at once.
4447 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
4449 /* start with default case */
4450 adapter->rx_ring[0]->reg_idx = 0;
4451 adapter->tx_ring[0]->reg_idx = 0;
4453 if (ixgbe_cache_ring_sriov(adapter))
4457 if (ixgbe_cache_ring_fcoe(adapter))
4460 #endif /* IXGBE_FCOE */
4461 #ifdef CONFIG_IXGBE_DCB
4462 if (ixgbe_cache_ring_dcb(adapter))
4466 if (ixgbe_cache_ring_fdir(adapter))
4469 if (ixgbe_cache_ring_rss(adapter))
4474 * ixgbe_alloc_queues - Allocate memory for all rings
4475 * @adapter: board private structure to initialize
4477 * We allocate one ring per queue at run-time since we don't know the
4478 * number of queues at compile-time. The polling_netdev array is
4479 * intended for Multiqueue, but should work fine with a single queue.
4481 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
4485 int orig_node = adapter->node;
4487 for (i = 0; i < adapter->num_tx_queues; i++) {
4488 struct ixgbe_ring *ring = adapter->tx_ring[i];
4489 if (orig_node == -1) {
4490 int cur_node = next_online_node(adapter->node);
4491 if (cur_node == MAX_NUMNODES)
4492 cur_node = first_online_node;
4493 adapter->node = cur_node;
4495 ring = kzalloc_node(sizeof(struct ixgbe_ring), GFP_KERNEL,
4498 ring = kzalloc(sizeof(struct ixgbe_ring), GFP_KERNEL);
4500 goto err_tx_ring_allocation;
4501 ring->count = adapter->tx_ring_count;
4502 ring->queue_index = i;
4503 ring->dev = &adapter->pdev->dev;
4504 ring->netdev = adapter->netdev;
4505 ring->numa_node = adapter->node;
4507 adapter->tx_ring[i] = ring;
4510 /* Restore the adapter's original node */
4511 adapter->node = orig_node;
4513 rx_count = adapter->rx_ring_count;
4514 for (i = 0; i < adapter->num_rx_queues; i++) {
4515 struct ixgbe_ring *ring = adapter->rx_ring[i];
4516 if (orig_node == -1) {
4517 int cur_node = next_online_node(adapter->node);
4518 if (cur_node == MAX_NUMNODES)
4519 cur_node = first_online_node;
4520 adapter->node = cur_node;
4522 ring = kzalloc_node(sizeof(struct ixgbe_ring), GFP_KERNEL,
4525 ring = kzalloc(sizeof(struct ixgbe_ring), GFP_KERNEL);
4527 goto err_rx_ring_allocation;
4528 ring->count = rx_count;
4529 ring->queue_index = i;
4530 ring->dev = &adapter->pdev->dev;
4531 ring->netdev = adapter->netdev;
4532 ring->numa_node = adapter->node;
4534 adapter->rx_ring[i] = ring;
4537 /* Restore the adapter's original node */
4538 adapter->node = orig_node;
4540 ixgbe_cache_ring_register(adapter);
4544 err_rx_ring_allocation:
4545 for (i = 0; i < adapter->num_tx_queues; i++)
4546 kfree(adapter->tx_ring[i]);
4547 err_tx_ring_allocation:
4552 * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
4553 * @adapter: board private structure to initialize
4555 * Attempt to configure the interrupts using the best available
4556 * capabilities of the hardware and the kernel.
4558 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
4560 struct ixgbe_hw *hw = &adapter->hw;
4562 int vector, v_budget;
4565 * It's easy to be greedy for MSI-X vectors, but it really
4566 * doesn't do us much good if we have a lot more vectors
4567 * than CPU's. So let's be conservative and only ask for
4568 * (roughly) the same number of vectors as there are CPU's.
4570 v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
4571 (int)num_online_cpus()) + NON_Q_VECTORS;
4574 * At the same time, hardware can only support a maximum of
4575 * hw.mac->max_msix_vectors vectors. With features
4576 * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
4577 * descriptor queues supported by our device. Thus, we cap it off in
4578 * those rare cases where the cpu count also exceeds our vector limit.
4580 v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
4582 /* A failure in MSI-X entry allocation isn't fatal, but it does
4583 * mean we disable MSI-X capabilities of the adapter. */
4584 adapter->msix_entries = kcalloc(v_budget,
4585 sizeof(struct msix_entry), GFP_KERNEL);
4586 if (adapter->msix_entries) {
4587 for (vector = 0; vector < v_budget; vector++)
4588 adapter->msix_entries[vector].entry = vector;
4590 ixgbe_acquire_msix_vectors(adapter, v_budget);
4592 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4596 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
4597 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
4598 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4599 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4600 adapter->atr_sample_rate = 0;
4601 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4602 ixgbe_disable_sriov(adapter);
4604 err = ixgbe_set_num_queues(adapter);
4608 err = pci_enable_msi(adapter->pdev);
4610 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
4612 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4613 "Unable to allocate MSI interrupt, "
4614 "falling back to legacy. Error: %d\n", err);
4624 * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
4625 * @adapter: board private structure to initialize
4627 * We allocate one q_vector per queue interrupt. If allocation fails we
4630 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
4632 int q_idx, num_q_vectors;
4633 struct ixgbe_q_vector *q_vector;
4635 int (*poll)(struct napi_struct *, int);
4637 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4638 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4639 napi_vectors = adapter->num_rx_queues;
4640 poll = &ixgbe_clean_rxtx_many;
4647 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4648 q_vector = kzalloc_node(sizeof(struct ixgbe_q_vector),
4649 GFP_KERNEL, adapter->node);
4651 q_vector = kzalloc(sizeof(struct ixgbe_q_vector),
4655 q_vector->adapter = adapter;
4656 if (q_vector->txr_count && !q_vector->rxr_count)
4657 q_vector->eitr = adapter->tx_eitr_param;
4659 q_vector->eitr = adapter->rx_eitr_param;
4660 q_vector->v_idx = q_idx;
4661 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
4662 adapter->q_vector[q_idx] = q_vector;
4670 q_vector = adapter->q_vector[q_idx];
4671 netif_napi_del(&q_vector->napi);
4673 adapter->q_vector[q_idx] = NULL;
4679 * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
4680 * @adapter: board private structure to initialize
4682 * This function frees the memory allocated to the q_vectors. In addition if
4683 * NAPI is enabled it will delete any references to the NAPI struct prior
4684 * to freeing the q_vector.
4686 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
4688 int q_idx, num_q_vectors;
4690 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4691 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4695 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4696 struct ixgbe_q_vector *q_vector = adapter->q_vector[q_idx];
4697 adapter->q_vector[q_idx] = NULL;
4698 netif_napi_del(&q_vector->napi);
4703 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
4705 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4706 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4707 pci_disable_msix(adapter->pdev);
4708 kfree(adapter->msix_entries);
4709 adapter->msix_entries = NULL;
4710 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
4711 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
4712 pci_disable_msi(adapter->pdev);
4717 * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
4718 * @adapter: board private structure to initialize
4720 * We determine which interrupt scheme to use based on...
4721 * - Kernel support (MSI, MSI-X)
4722 * - which can be user-defined (via MODULE_PARAM)
4723 * - Hardware queue count (num_*_queues)
4724 * - defined by miscellaneous hardware support/features (RSS, etc.)
4726 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
4730 /* Number of supported queues */
4731 err = ixgbe_set_num_queues(adapter);
4735 err = ixgbe_set_interrupt_capability(adapter);
4737 e_dev_err("Unable to setup interrupt capabilities\n");
4738 goto err_set_interrupt;
4741 err = ixgbe_alloc_q_vectors(adapter);
4743 e_dev_err("Unable to allocate memory for queue vectors\n");
4744 goto err_alloc_q_vectors;
4747 err = ixgbe_alloc_queues(adapter);
4749 e_dev_err("Unable to allocate memory for queues\n");
4750 goto err_alloc_queues;
4753 e_dev_info("Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
4754 (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
4755 adapter->num_rx_queues, adapter->num_tx_queues);
4757 set_bit(__IXGBE_DOWN, &adapter->state);
4762 ixgbe_free_q_vectors(adapter);
4763 err_alloc_q_vectors:
4764 ixgbe_reset_interrupt_capability(adapter);
4769 static void ring_free_rcu(struct rcu_head *head)
4771 kfree(container_of(head, struct ixgbe_ring, rcu));
4775 * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
4776 * @adapter: board private structure to clear interrupt scheme on
4778 * We go through and clear interrupt specific resources and reset the structure
4779 * to pre-load conditions
4781 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
4785 for (i = 0; i < adapter->num_tx_queues; i++) {
4786 kfree(adapter->tx_ring[i]);
4787 adapter->tx_ring[i] = NULL;
4789 for (i = 0; i < adapter->num_rx_queues; i++) {
4790 struct ixgbe_ring *ring = adapter->rx_ring[i];
4792 /* ixgbe_get_stats64() might access this ring, we must wait
4793 * a grace period before freeing it.
4795 call_rcu(&ring->rcu, ring_free_rcu);
4796 adapter->rx_ring[i] = NULL;
4799 ixgbe_free_q_vectors(adapter);
4800 ixgbe_reset_interrupt_capability(adapter);
4804 * ixgbe_sfp_timer - worker thread to find a missing module
4805 * @data: pointer to our adapter struct
4807 static void ixgbe_sfp_timer(unsigned long data)
4809 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
4812 * Do the sfp_timer outside of interrupt context due to the
4813 * delays that sfp+ detection requires
4815 schedule_work(&adapter->sfp_task);
4819 * ixgbe_sfp_task - worker thread to find a missing module
4820 * @work: pointer to work_struct containing our data
4822 static void ixgbe_sfp_task(struct work_struct *work)
4824 struct ixgbe_adapter *adapter = container_of(work,
4825 struct ixgbe_adapter,
4827 struct ixgbe_hw *hw = &adapter->hw;
4829 if ((hw->phy.type == ixgbe_phy_nl) &&
4830 (hw->phy.sfp_type == ixgbe_sfp_type_not_present)) {
4831 s32 ret = hw->phy.ops.identify_sfp(hw);
4832 if (ret == IXGBE_ERR_SFP_NOT_PRESENT)
4834 ret = hw->phy.ops.reset(hw);
4835 if (ret == IXGBE_ERR_SFP_NOT_SUPPORTED) {
4836 e_dev_err("failed to initialize because an unsupported "
4837 "SFP+ module type was detected.\n");
4838 e_dev_err("Reload the driver after installing a "
4839 "supported module.\n");
4840 unregister_netdev(adapter->netdev);
4842 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
4844 /* don't need this routine any more */
4845 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
4849 if (test_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state))
4850 mod_timer(&adapter->sfp_timer,
4851 round_jiffies(jiffies + (2 * HZ)));
4855 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4856 * @adapter: board private structure to initialize
4858 * ixgbe_sw_init initializes the Adapter private data structure.
4859 * Fields are initialized based on PCI device information and
4860 * OS network device settings (MTU size).
4862 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
4864 struct ixgbe_hw *hw = &adapter->hw;
4865 struct pci_dev *pdev = adapter->pdev;
4866 struct net_device *dev = adapter->netdev;
4868 #ifdef CONFIG_IXGBE_DCB
4870 struct tc_configuration *tc;
4872 int max_frame = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
4874 /* PCI config space info */
4876 hw->vendor_id = pdev->vendor;
4877 hw->device_id = pdev->device;
4878 hw->revision_id = pdev->revision;
4879 hw->subsystem_vendor_id = pdev->subsystem_vendor;
4880 hw->subsystem_device_id = pdev->subsystem_device;
4882 /* Set capability flags */
4883 rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
4884 adapter->ring_feature[RING_F_RSS].indices = rss;
4885 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
4886 adapter->ring_feature[RING_F_DCB].indices = IXGBE_MAX_DCB_INDICES;
4887 if (hw->mac.type == ixgbe_mac_82598EB) {
4888 if (hw->device_id == IXGBE_DEV_ID_82598AT)
4889 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4890 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
4891 } else if (hw->mac.type == ixgbe_mac_82599EB) {
4892 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
4893 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4894 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4895 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
4896 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4897 if (dev->features & NETIF_F_NTUPLE) {
4898 /* Flow Director perfect filter enabled */
4899 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4900 adapter->atr_sample_rate = 0;
4901 spin_lock_init(&adapter->fdir_perfect_lock);
4903 /* Flow Director hash filters enabled */
4904 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
4905 adapter->atr_sample_rate = 20;
4907 adapter->ring_feature[RING_F_FDIR].indices =
4908 IXGBE_MAX_FDIR_INDICES;
4909 adapter->fdir_pballoc = 0;
4911 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4912 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4913 adapter->ring_feature[RING_F_FCOE].indices = 0;
4914 #ifdef CONFIG_IXGBE_DCB
4915 /* Default traffic class to use for FCoE */
4916 adapter->fcoe.tc = IXGBE_FCOE_DEFTC;
4917 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
4919 #endif /* IXGBE_FCOE */
4922 #ifdef CONFIG_IXGBE_DCB
4923 /* Configure DCB traffic classes */
4924 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
4925 tc = &adapter->dcb_cfg.tc_config[j];
4926 tc->path[DCB_TX_CONFIG].bwg_id = 0;
4927 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
4928 tc->path[DCB_RX_CONFIG].bwg_id = 0;
4929 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
4930 tc->dcb_pfc = pfc_disabled;
4932 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
4933 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
4934 adapter->dcb_cfg.rx_pba_cfg = pba_equal;
4935 adapter->dcb_cfg.pfc_mode_enable = false;
4936 adapter->dcb_cfg.round_robin_enable = false;
4937 adapter->dcb_set_bitmap = 0x00;
4938 ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
4939 adapter->ring_feature[RING_F_DCB].indices);
4943 /* default flow control settings */
4944 hw->fc.requested_mode = ixgbe_fc_full;
4945 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
4947 adapter->last_lfc_mode = hw->fc.current_mode;
4949 hw->fc.high_water = FC_HIGH_WATER(max_frame);
4950 hw->fc.low_water = FC_LOW_WATER(max_frame);
4951 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
4952 hw->fc.send_xon = true;
4953 hw->fc.disable_fc_autoneg = false;
4955 /* enable itr by default in dynamic mode */
4956 adapter->rx_itr_setting = 1;
4957 adapter->rx_eitr_param = 20000;
4958 adapter->tx_itr_setting = 1;
4959 adapter->tx_eitr_param = 10000;
4961 /* set defaults for eitr in MegaBytes */
4962 adapter->eitr_low = 10;
4963 adapter->eitr_high = 20;
4965 /* set default ring sizes */
4966 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
4967 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
4969 /* initialize eeprom parameters */
4970 if (ixgbe_init_eeprom_params_generic(hw)) {
4971 e_dev_err("EEPROM initialization failed\n");
4975 /* enable rx csum by default */
4976 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
4978 /* get assigned NUMA node */
4979 adapter->node = dev_to_node(&pdev->dev);
4981 set_bit(__IXGBE_DOWN, &adapter->state);
4987 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
4988 * @tx_ring: tx descriptor ring (for a specific queue) to setup
4990 * Return 0 on success, negative on failure
4992 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
4994 struct device *dev = tx_ring->dev;
4997 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4998 tx_ring->tx_buffer_info = vmalloc_node(size, tx_ring->numa_node);
4999 if (!tx_ring->tx_buffer_info)
5000 tx_ring->tx_buffer_info = vmalloc(size);
5001 if (!tx_ring->tx_buffer_info)
5003 memset(tx_ring->tx_buffer_info, 0, size);
5005 /* round up to nearest 4K */
5006 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
5007 tx_ring->size = ALIGN(tx_ring->size, 4096);
5009 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
5010 &tx_ring->dma, GFP_KERNEL);
5014 tx_ring->next_to_use = 0;
5015 tx_ring->next_to_clean = 0;
5016 tx_ring->work_limit = tx_ring->count;
5020 vfree(tx_ring->tx_buffer_info);
5021 tx_ring->tx_buffer_info = NULL;
5022 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
5027 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
5028 * @adapter: board private structure
5030 * If this function returns with an error, then it's possible one or
5031 * more of the rings is populated (while the rest are not). It is the
5032 * callers duty to clean those orphaned rings.
5034 * Return 0 on success, negative on failure
5036 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
5040 for (i = 0; i < adapter->num_tx_queues; i++) {
5041 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
5044 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
5052 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
5053 * @rx_ring: rx descriptor ring (for a specific queue) to setup
5055 * Returns 0 on success, negative on failure
5057 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
5059 struct device *dev = rx_ring->dev;
5062 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
5063 rx_ring->rx_buffer_info = vmalloc_node(size, rx_ring->numa_node);
5064 if (!rx_ring->rx_buffer_info)
5065 rx_ring->rx_buffer_info = vmalloc(size);
5066 if (!rx_ring->rx_buffer_info)
5068 memset(rx_ring->rx_buffer_info, 0, size);
5070 /* Round up to nearest 4K */
5071 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
5072 rx_ring->size = ALIGN(rx_ring->size, 4096);
5074 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
5075 &rx_ring->dma, GFP_KERNEL);
5080 rx_ring->next_to_clean = 0;
5081 rx_ring->next_to_use = 0;
5085 vfree(rx_ring->rx_buffer_info);
5086 rx_ring->rx_buffer_info = NULL;
5087 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
5092 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
5093 * @adapter: board private structure
5095 * If this function returns with an error, then it's possible one or
5096 * more of the rings is populated (while the rest are not). It is the
5097 * callers duty to clean those orphaned rings.
5099 * Return 0 on success, negative on failure
5101 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
5105 for (i = 0; i < adapter->num_rx_queues; i++) {
5106 err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
5109 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
5117 * ixgbe_free_tx_resources - Free Tx Resources per Queue
5118 * @tx_ring: Tx descriptor ring for a specific queue
5120 * Free all transmit software resources
5122 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
5124 ixgbe_clean_tx_ring(tx_ring);
5126 vfree(tx_ring->tx_buffer_info);
5127 tx_ring->tx_buffer_info = NULL;
5129 /* if not set, then don't free */
5133 dma_free_coherent(tx_ring->dev, tx_ring->size,
5134 tx_ring->desc, tx_ring->dma);
5136 tx_ring->desc = NULL;
5140 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5141 * @adapter: board private structure
5143 * Free all transmit software resources
5145 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
5149 for (i = 0; i < adapter->num_tx_queues; i++)
5150 if (adapter->tx_ring[i]->desc)
5151 ixgbe_free_tx_resources(adapter->tx_ring[i]);
5155 * ixgbe_free_rx_resources - Free Rx Resources
5156 * @rx_ring: ring to clean the resources from
5158 * Free all receive software resources
5160 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
5162 ixgbe_clean_rx_ring(rx_ring);
5164 vfree(rx_ring->rx_buffer_info);
5165 rx_ring->rx_buffer_info = NULL;
5167 /* if not set, then don't free */
5171 dma_free_coherent(rx_ring->dev, rx_ring->size,
5172 rx_ring->desc, rx_ring->dma);
5174 rx_ring->desc = NULL;
5178 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5179 * @adapter: board private structure
5181 * Free all receive software resources
5183 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5187 for (i = 0; i < adapter->num_rx_queues; i++)
5188 if (adapter->rx_ring[i]->desc)
5189 ixgbe_free_rx_resources(adapter->rx_ring[i]);
5193 * ixgbe_change_mtu - Change the Maximum Transfer Unit
5194 * @netdev: network interface device structure
5195 * @new_mtu: new value for maximum frame size
5197 * Returns 0 on success, negative on failure
5199 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5201 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5202 struct ixgbe_hw *hw = &adapter->hw;
5203 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5205 /* MTU < 68 is an error and causes problems on some kernels */
5206 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5209 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5210 /* must set new MTU before calling down or up */
5211 netdev->mtu = new_mtu;
5213 hw->fc.high_water = FC_HIGH_WATER(max_frame);
5214 hw->fc.low_water = FC_LOW_WATER(max_frame);
5216 if (netif_running(netdev))
5217 ixgbe_reinit_locked(adapter);
5223 * ixgbe_open - Called when a network interface is made active
5224 * @netdev: network interface device structure
5226 * Returns 0 on success, negative value on failure
5228 * The open entry point is called when a network interface is made
5229 * active by the system (IFF_UP). At this point all resources needed
5230 * for transmit and receive operations are allocated, the interrupt
5231 * handler is registered with the OS, the watchdog timer is started,
5232 * and the stack is notified that the interface is ready.
5234 static int ixgbe_open(struct net_device *netdev)
5236 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5239 /* disallow open during test */
5240 if (test_bit(__IXGBE_TESTING, &adapter->state))
5243 netif_carrier_off(netdev);
5245 /* allocate transmit descriptors */
5246 err = ixgbe_setup_all_tx_resources(adapter);
5250 /* allocate receive descriptors */
5251 err = ixgbe_setup_all_rx_resources(adapter);
5255 ixgbe_configure(adapter);
5257 err = ixgbe_request_irq(adapter);
5261 err = ixgbe_up_complete(adapter);
5265 netif_tx_start_all_queues(netdev);
5270 ixgbe_release_hw_control(adapter);
5271 ixgbe_free_irq(adapter);
5274 ixgbe_free_all_rx_resources(adapter);
5276 ixgbe_free_all_tx_resources(adapter);
5277 ixgbe_reset(adapter);
5283 * ixgbe_close - Disables a network interface
5284 * @netdev: network interface device structure
5286 * Returns 0, this is not allowed to fail
5288 * The close entry point is called when an interface is de-activated
5289 * by the OS. The hardware is still under the drivers control, but
5290 * needs to be disabled. A global MAC reset is issued to stop the
5291 * hardware, and all transmit and receive resources are freed.
5293 static int ixgbe_close(struct net_device *netdev)
5295 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5297 ixgbe_down(adapter);
5298 ixgbe_free_irq(adapter);
5300 ixgbe_free_all_tx_resources(adapter);
5301 ixgbe_free_all_rx_resources(adapter);
5303 ixgbe_release_hw_control(adapter);
5309 static int ixgbe_resume(struct pci_dev *pdev)
5311 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5312 struct net_device *netdev = adapter->netdev;
5315 pci_set_power_state(pdev, PCI_D0);
5316 pci_restore_state(pdev);
5318 * pci_restore_state clears dev->state_saved so call
5319 * pci_save_state to restore it.
5321 pci_save_state(pdev);
5323 err = pci_enable_device_mem(pdev);
5325 e_dev_err("Cannot enable PCI device from suspend\n");
5328 pci_set_master(pdev);
5330 pci_wake_from_d3(pdev, false);
5332 err = ixgbe_init_interrupt_scheme(adapter);
5334 e_dev_err("Cannot initialize interrupts for device\n");
5338 ixgbe_reset(adapter);
5340 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5342 if (netif_running(netdev)) {
5343 err = ixgbe_open(netdev);
5348 netif_device_attach(netdev);
5352 #endif /* CONFIG_PM */
5354 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5356 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5357 struct net_device *netdev = adapter->netdev;
5358 struct ixgbe_hw *hw = &adapter->hw;
5360 u32 wufc = adapter->wol;
5365 netif_device_detach(netdev);
5367 if (netif_running(netdev)) {
5368 ixgbe_down(adapter);
5369 ixgbe_free_irq(adapter);
5370 ixgbe_free_all_tx_resources(adapter);
5371 ixgbe_free_all_rx_resources(adapter);
5374 ixgbe_clear_interrupt_scheme(adapter);
5377 retval = pci_save_state(pdev);
5383 ixgbe_set_rx_mode(netdev);
5385 /* turn on all-multi mode if wake on multicast is enabled */
5386 if (wufc & IXGBE_WUFC_MC) {
5387 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5388 fctrl |= IXGBE_FCTRL_MPE;
5389 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5392 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5393 ctrl |= IXGBE_CTRL_GIO_DIS;
5394 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5396 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5398 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5399 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5402 if (wufc && hw->mac.type == ixgbe_mac_82599EB)
5403 pci_wake_from_d3(pdev, true);
5405 pci_wake_from_d3(pdev, false);
5407 *enable_wake = !!wufc;
5409 ixgbe_release_hw_control(adapter);
5411 pci_disable_device(pdev);
5417 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5422 retval = __ixgbe_shutdown(pdev, &wake);
5427 pci_prepare_to_sleep(pdev);
5429 pci_wake_from_d3(pdev, false);
5430 pci_set_power_state(pdev, PCI_D3hot);
5435 #endif /* CONFIG_PM */
5437 static void ixgbe_shutdown(struct pci_dev *pdev)
5441 __ixgbe_shutdown(pdev, &wake);
5443 if (system_state == SYSTEM_POWER_OFF) {
5444 pci_wake_from_d3(pdev, wake);
5445 pci_set_power_state(pdev, PCI_D3hot);
5450 * ixgbe_update_stats - Update the board statistics counters.
5451 * @adapter: board private structure
5453 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5455 struct net_device *netdev = adapter->netdev;
5456 struct ixgbe_hw *hw = &adapter->hw;
5457 struct ixgbe_hw_stats *hwstats = &adapter->stats;
5459 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5460 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
5461 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
5462 u64 bytes = 0, packets = 0;
5464 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5465 test_bit(__IXGBE_RESETTING, &adapter->state))
5468 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5471 for (i = 0; i < 16; i++)
5472 adapter->hw_rx_no_dma_resources +=
5473 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5474 for (i = 0; i < adapter->num_rx_queues; i++) {
5475 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
5476 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
5478 adapter->rsc_total_count = rsc_count;
5479 adapter->rsc_total_flush = rsc_flush;
5482 for (i = 0; i < adapter->num_rx_queues; i++) {
5483 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
5484 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
5485 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
5486 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
5487 bytes += rx_ring->stats.bytes;
5488 packets += rx_ring->stats.packets;
5490 adapter->non_eop_descs = non_eop_descs;
5491 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5492 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5493 netdev->stats.rx_bytes = bytes;
5494 netdev->stats.rx_packets = packets;
5498 /* gather some stats to the adapter struct that are per queue */
5499 for (i = 0; i < adapter->num_tx_queues; i++) {
5500 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5501 restart_queue += tx_ring->tx_stats.restart_queue;
5502 tx_busy += tx_ring->tx_stats.tx_busy;
5503 bytes += tx_ring->stats.bytes;
5504 packets += tx_ring->stats.packets;
5506 adapter->restart_queue = restart_queue;
5507 adapter->tx_busy = tx_busy;
5508 netdev->stats.tx_bytes = bytes;
5509 netdev->stats.tx_packets = packets;
5511 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5512 for (i = 0; i < 8; i++) {
5513 /* for packet buffers not used, the register should read 0 */
5514 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5516 hwstats->mpc[i] += mpc;
5517 total_mpc += hwstats->mpc[i];
5518 if (hw->mac.type == ixgbe_mac_82598EB)
5519 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5520 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5521 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5522 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5523 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5524 if (hw->mac.type == ixgbe_mac_82599EB) {
5525 hwstats->pxonrxc[i] +=
5526 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5527 hwstats->pxoffrxc[i] +=
5528 IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
5529 hwstats->qprdc[i] += IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5531 hwstats->pxonrxc[i] +=
5532 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5533 hwstats->pxoffrxc[i] +=
5534 IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
5536 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5537 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5539 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5540 /* work around hardware counting issue */
5541 hwstats->gprc -= missed_rx;
5543 /* 82598 hardware only has a 32 bit counter in the high register */
5544 if (hw->mac.type == ixgbe_mac_82599EB) {
5546 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5547 tmp = IXGBE_READ_REG(hw, IXGBE_GORCH) & 0xF;
5548 /* 4 high bits of GORC */
5549 hwstats->gorc += (tmp << 32);
5550 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5551 tmp = IXGBE_READ_REG(hw, IXGBE_GOTCH) & 0xF;
5552 /* 4 high bits of GOTC */
5553 hwstats->gotc += (tmp << 32);
5554 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5555 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5556 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5557 hwstats->lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
5558 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5559 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5561 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5562 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5563 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5564 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5565 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5566 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5567 #endif /* IXGBE_FCOE */
5569 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5570 hwstats->lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
5571 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5572 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5573 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5575 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5576 hwstats->bprc += bprc;
5577 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5578 if (hw->mac.type == ixgbe_mac_82598EB)
5579 hwstats->mprc -= bprc;
5580 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5581 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5582 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5583 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5584 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5585 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5586 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5587 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5588 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5589 hwstats->lxontxc += lxon;
5590 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5591 hwstats->lxofftxc += lxoff;
5592 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5593 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5594 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5596 * 82598 errata - tx of flow control packets is included in tx counters
5598 xon_off_tot = lxon + lxoff;
5599 hwstats->gptc -= xon_off_tot;
5600 hwstats->mptc -= xon_off_tot;
5601 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5602 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5603 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5604 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5605 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5606 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5607 hwstats->ptc64 -= xon_off_tot;
5608 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5609 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5610 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5611 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5612 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5613 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5615 /* Fill out the OS statistics structure */
5616 netdev->stats.multicast = hwstats->mprc;
5619 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
5620 netdev->stats.rx_dropped = 0;
5621 netdev->stats.rx_length_errors = hwstats->rlec;
5622 netdev->stats.rx_crc_errors = hwstats->crcerrs;
5623 netdev->stats.rx_missed_errors = total_mpc;
5627 * ixgbe_watchdog - Timer Call-back
5628 * @data: pointer to adapter cast into an unsigned long
5630 static void ixgbe_watchdog(unsigned long data)
5632 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5633 struct ixgbe_hw *hw = &adapter->hw;
5638 * Do the watchdog outside of interrupt context due to the lovely
5639 * delays that some of the newer hardware requires
5642 if (test_bit(__IXGBE_DOWN, &adapter->state))
5643 goto watchdog_short_circuit;
5645 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5647 * for legacy and MSI interrupts don't set any bits
5648 * that are enabled for EIAM, because this operation
5649 * would set *both* EIMS and EICS for any bit in EIAM
5651 IXGBE_WRITE_REG(hw, IXGBE_EICS,
5652 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5653 goto watchdog_reschedule;
5656 /* get one bit for every active tx/rx interrupt vector */
5657 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
5658 struct ixgbe_q_vector *qv = adapter->q_vector[i];
5659 if (qv->rxr_count || qv->txr_count)
5660 eics |= ((u64)1 << i);
5663 /* Cause software interrupt to ensure rx rings are cleaned */
5664 ixgbe_irq_rearm_queues(adapter, eics);
5666 watchdog_reschedule:
5667 /* Reset the timer */
5668 mod_timer(&adapter->watchdog_timer, round_jiffies(jiffies + 2 * HZ));
5670 watchdog_short_circuit:
5671 schedule_work(&adapter->watchdog_task);
5675 * ixgbe_multispeed_fiber_task - worker thread to configure multispeed fiber
5676 * @work: pointer to work_struct containing our data
5678 static void ixgbe_multispeed_fiber_task(struct work_struct *work)
5680 struct ixgbe_adapter *adapter = container_of(work,
5681 struct ixgbe_adapter,
5682 multispeed_fiber_task);
5683 struct ixgbe_hw *hw = &adapter->hw;
5687 adapter->flags |= IXGBE_FLAG_IN_SFP_LINK_TASK;
5688 autoneg = hw->phy.autoneg_advertised;
5689 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
5690 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
5691 hw->mac.autotry_restart = false;
5692 if (hw->mac.ops.setup_link)
5693 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
5694 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5695 adapter->flags &= ~IXGBE_FLAG_IN_SFP_LINK_TASK;
5699 * ixgbe_sfp_config_module_task - worker thread to configure a new SFP+ module
5700 * @work: pointer to work_struct containing our data
5702 static void ixgbe_sfp_config_module_task(struct work_struct *work)
5704 struct ixgbe_adapter *adapter = container_of(work,
5705 struct ixgbe_adapter,
5706 sfp_config_module_task);
5707 struct ixgbe_hw *hw = &adapter->hw;
5710 adapter->flags |= IXGBE_FLAG_IN_SFP_MOD_TASK;
5712 /* Time for electrical oscillations to settle down */
5714 err = hw->phy.ops.identify_sfp(hw);
5716 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
5717 e_dev_err("failed to initialize because an unsupported SFP+ "
5718 "module type was detected.\n");
5719 e_dev_err("Reload the driver after installing a supported "
5721 unregister_netdev(adapter->netdev);
5724 hw->mac.ops.setup_sfp(hw);
5726 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
5727 /* This will also work for DA Twinax connections */
5728 schedule_work(&adapter->multispeed_fiber_task);
5729 adapter->flags &= ~IXGBE_FLAG_IN_SFP_MOD_TASK;
5733 * ixgbe_fdir_reinit_task - worker thread to reinit FDIR filter table
5734 * @work: pointer to work_struct containing our data
5736 static void ixgbe_fdir_reinit_task(struct work_struct *work)
5738 struct ixgbe_adapter *adapter = container_of(work,
5739 struct ixgbe_adapter,
5741 struct ixgbe_hw *hw = &adapter->hw;
5744 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5745 for (i = 0; i < adapter->num_tx_queues; i++)
5746 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
5747 &(adapter->tx_ring[i]->state));
5749 e_err(probe, "failed to finish FDIR re-initialization, "
5750 "ignored adding FDIR ATR filters\n");
5752 /* Done FDIR Re-initialization, enable transmits */
5753 netif_tx_start_all_queues(adapter->netdev);
5756 static DEFINE_MUTEX(ixgbe_watchdog_lock);
5759 * ixgbe_watchdog_task - worker thread to bring link up
5760 * @work: pointer to work_struct containing our data
5762 static void ixgbe_watchdog_task(struct work_struct *work)
5764 struct ixgbe_adapter *adapter = container_of(work,
5765 struct ixgbe_adapter,
5767 struct net_device *netdev = adapter->netdev;
5768 struct ixgbe_hw *hw = &adapter->hw;
5772 struct ixgbe_ring *tx_ring;
5773 int some_tx_pending = 0;
5775 mutex_lock(&ixgbe_watchdog_lock);
5777 link_up = adapter->link_up;
5778 link_speed = adapter->link_speed;
5780 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
5781 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5784 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5785 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
5786 hw->mac.ops.fc_enable(hw, i);
5788 hw->mac.ops.fc_enable(hw, 0);
5791 hw->mac.ops.fc_enable(hw, 0);
5796 time_after(jiffies, (adapter->link_check_timeout +
5797 IXGBE_TRY_LINK_TIMEOUT))) {
5798 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5799 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5801 adapter->link_up = link_up;
5802 adapter->link_speed = link_speed;
5806 if (!netif_carrier_ok(netdev)) {
5807 bool flow_rx, flow_tx;
5809 if (hw->mac.type == ixgbe_mac_82599EB) {
5810 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5811 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5812 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5813 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5815 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5816 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5817 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5818 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5821 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
5822 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5824 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5825 "1 Gbps" : "unknown speed")),
5826 ((flow_rx && flow_tx) ? "RX/TX" :
5828 (flow_tx ? "TX" : "None"))));
5830 netif_carrier_on(netdev);
5832 /* Force detection of hung controller */
5833 for (i = 0; i < adapter->num_tx_queues; i++) {
5834 tx_ring = adapter->tx_ring[i];
5835 set_check_for_tx_hang(tx_ring);
5839 adapter->link_up = false;
5840 adapter->link_speed = 0;
5841 if (netif_carrier_ok(netdev)) {
5842 e_info(drv, "NIC Link is Down\n");
5843 netif_carrier_off(netdev);
5847 if (!netif_carrier_ok(netdev)) {
5848 for (i = 0; i < adapter->num_tx_queues; i++) {
5849 tx_ring = adapter->tx_ring[i];
5850 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5851 some_tx_pending = 1;
5856 if (some_tx_pending) {
5857 /* We've lost link, so the controller stops DMA,
5858 * but we've got queued Tx work that's never going
5859 * to get done, so reset controller to flush Tx.
5860 * (Do the reset outside of interrupt context).
5862 schedule_work(&adapter->reset_task);
5866 ixgbe_update_stats(adapter);
5867 mutex_unlock(&ixgbe_watchdog_lock);
5870 static int ixgbe_tso(struct ixgbe_adapter *adapter,
5871 struct ixgbe_ring *tx_ring, struct sk_buff *skb,
5872 u32 tx_flags, u8 *hdr_len, __be16 protocol)
5874 struct ixgbe_adv_tx_context_desc *context_desc;
5877 struct ixgbe_tx_buffer *tx_buffer_info;
5878 u32 vlan_macip_lens = 0, type_tucmd_mlhl;
5879 u32 mss_l4len_idx, l4len;
5881 if (skb_is_gso(skb)) {
5882 if (skb_header_cloned(skb)) {
5883 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5887 l4len = tcp_hdrlen(skb);
5890 if (protocol == htons(ETH_P_IP)) {
5891 struct iphdr *iph = ip_hdr(skb);
5894 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
5898 } else if (skb_is_gso_v6(skb)) {
5899 ipv6_hdr(skb)->payload_len = 0;
5900 tcp_hdr(skb)->check =
5901 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5902 &ipv6_hdr(skb)->daddr,
5906 i = tx_ring->next_to_use;
5908 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5909 context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
5911 /* VLAN MACLEN IPLEN */
5912 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5914 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
5915 vlan_macip_lens |= ((skb_network_offset(skb)) <<
5916 IXGBE_ADVTXD_MACLEN_SHIFT);
5917 *hdr_len += skb_network_offset(skb);
5919 (skb_transport_header(skb) - skb_network_header(skb));
5921 (skb_transport_header(skb) - skb_network_header(skb));
5922 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
5923 context_desc->seqnum_seed = 0;
5925 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
5926 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
5927 IXGBE_ADVTXD_DTYP_CTXT);
5929 if (protocol == htons(ETH_P_IP))
5930 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
5931 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5932 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
5936 (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
5937 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
5938 /* use index 1 for TSO */
5939 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
5940 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
5942 tx_buffer_info->time_stamp = jiffies;
5943 tx_buffer_info->next_to_watch = i;
5946 if (i == tx_ring->count)
5948 tx_ring->next_to_use = i;
5955 static u32 ixgbe_psum(struct ixgbe_adapter *adapter, struct sk_buff *skb,
5961 case cpu_to_be16(ETH_P_IP):
5962 rtn |= IXGBE_ADVTXD_TUCMD_IPV4;
5963 switch (ip_hdr(skb)->protocol) {
5965 rtn |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5968 rtn |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5972 case cpu_to_be16(ETH_P_IPV6):
5973 /* XXX what about other V6 headers?? */
5974 switch (ipv6_hdr(skb)->nexthdr) {
5976 rtn |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5979 rtn |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5984 if (unlikely(net_ratelimit()))
5985 e_warn(probe, "partial checksum but proto=%x!\n",
5993 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
5994 struct ixgbe_ring *tx_ring,
5995 struct sk_buff *skb, u32 tx_flags,
5998 struct ixgbe_adv_tx_context_desc *context_desc;
6000 struct ixgbe_tx_buffer *tx_buffer_info;
6001 u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
6003 if (skb->ip_summed == CHECKSUM_PARTIAL ||
6004 (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
6005 i = tx_ring->next_to_use;
6006 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6007 context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
6009 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
6011 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
6012 vlan_macip_lens |= (skb_network_offset(skb) <<
6013 IXGBE_ADVTXD_MACLEN_SHIFT);
6014 if (skb->ip_summed == CHECKSUM_PARTIAL)
6015 vlan_macip_lens |= (skb_transport_header(skb) -
6016 skb_network_header(skb));
6018 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
6019 context_desc->seqnum_seed = 0;
6021 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
6022 IXGBE_ADVTXD_DTYP_CTXT);
6024 if (skb->ip_summed == CHECKSUM_PARTIAL)
6025 type_tucmd_mlhl |= ixgbe_psum(adapter, skb, protocol);
6027 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
6028 /* use index zero for tx checksum offload */
6029 context_desc->mss_l4len_idx = 0;
6031 tx_buffer_info->time_stamp = jiffies;
6032 tx_buffer_info->next_to_watch = i;
6035 if (i == tx_ring->count)
6037 tx_ring->next_to_use = i;
6045 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
6046 struct ixgbe_ring *tx_ring,
6047 struct sk_buff *skb, u32 tx_flags,
6048 unsigned int first, const u8 hdr_len)
6050 struct device *dev = tx_ring->dev;
6051 struct ixgbe_tx_buffer *tx_buffer_info;
6053 unsigned int total = skb->len;
6054 unsigned int offset = 0, size, count = 0, i;
6055 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
6057 unsigned int bytecount = skb->len;
6060 i = tx_ring->next_to_use;
6062 if (tx_flags & IXGBE_TX_FLAGS_FCOE)
6063 /* excluding fcoe_crc_eof for FCoE */
6064 total -= sizeof(struct fcoe_crc_eof);
6066 len = min(skb_headlen(skb), total);
6068 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6069 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
6071 tx_buffer_info->length = size;
6072 tx_buffer_info->mapped_as_page = false;
6073 tx_buffer_info->dma = dma_map_single(dev,
6075 size, DMA_TO_DEVICE);
6076 if (dma_mapping_error(dev, tx_buffer_info->dma))
6078 tx_buffer_info->time_stamp = jiffies;
6079 tx_buffer_info->next_to_watch = i;
6088 if (i == tx_ring->count)
6093 for (f = 0; f < nr_frags; f++) {
6094 struct skb_frag_struct *frag;
6096 frag = &skb_shinfo(skb)->frags[f];
6097 len = min((unsigned int)frag->size, total);
6098 offset = frag->page_offset;
6102 if (i == tx_ring->count)
6105 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6106 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
6108 tx_buffer_info->length = size;
6109 tx_buffer_info->dma = dma_map_page(dev,
6113 tx_buffer_info->mapped_as_page = true;
6114 if (dma_mapping_error(dev, tx_buffer_info->dma))
6116 tx_buffer_info->time_stamp = jiffies;
6117 tx_buffer_info->next_to_watch = i;
6128 if (tx_flags & IXGBE_TX_FLAGS_TSO)
6129 gso_segs = skb_shinfo(skb)->gso_segs;
6131 /* adjust for FCoE Sequence Offload */
6132 else if (tx_flags & IXGBE_TX_FLAGS_FSO)
6133 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
6134 skb_shinfo(skb)->gso_size);
6135 #endif /* IXGBE_FCOE */
6136 bytecount += (gso_segs - 1) * hdr_len;
6138 /* multiply data chunks by size of headers */
6139 tx_ring->tx_buffer_info[i].bytecount = bytecount;
6140 tx_ring->tx_buffer_info[i].gso_segs = gso_segs;
6141 tx_ring->tx_buffer_info[i].skb = skb;
6142 tx_ring->tx_buffer_info[first].next_to_watch = i;
6147 e_dev_err("TX DMA map failed\n");
6149 /* clear timestamp and dma mappings for failed tx_buffer_info map */
6150 tx_buffer_info->dma = 0;
6151 tx_buffer_info->time_stamp = 0;
6152 tx_buffer_info->next_to_watch = 0;
6156 /* clear timestamp and dma mappings for remaining portion of packet */
6159 i += tx_ring->count;
6161 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6162 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
6168 static void ixgbe_tx_queue(struct ixgbe_ring *tx_ring,
6169 int tx_flags, int count, u32 paylen, u8 hdr_len)
6171 union ixgbe_adv_tx_desc *tx_desc = NULL;
6172 struct ixgbe_tx_buffer *tx_buffer_info;
6173 u32 olinfo_status = 0, cmd_type_len = 0;
6175 u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
6177 cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
6179 cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
6181 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
6182 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
6184 if (tx_flags & IXGBE_TX_FLAGS_TSO) {
6185 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
6187 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
6188 IXGBE_ADVTXD_POPTS_SHIFT;
6190 /* use index 1 context for tso */
6191 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6192 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
6193 olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
6194 IXGBE_ADVTXD_POPTS_SHIFT;
6196 } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
6197 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
6198 IXGBE_ADVTXD_POPTS_SHIFT;
6200 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6201 olinfo_status |= IXGBE_ADVTXD_CC;
6202 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6203 if (tx_flags & IXGBE_TX_FLAGS_FSO)
6204 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
6207 olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
6209 i = tx_ring->next_to_use;
6211 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6212 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
6213 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
6214 tx_desc->read.cmd_type_len =
6215 cpu_to_le32(cmd_type_len | tx_buffer_info->length);
6216 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
6218 if (i == tx_ring->count)
6222 tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
6225 * Force memory writes to complete before letting h/w
6226 * know there are new descriptors to fetch. (Only
6227 * applicable for weak-ordered memory model archs,
6232 tx_ring->next_to_use = i;
6233 writel(i, tx_ring->tail);
6236 static void ixgbe_atr(struct ixgbe_adapter *adapter, struct sk_buff *skb,
6237 u8 queue, u32 tx_flags, __be16 protocol)
6239 struct ixgbe_atr_input atr_input;
6240 struct iphdr *iph = ip_hdr(skb);
6241 struct ethhdr *eth = (struct ethhdr *)skb->data;
6245 /* Right now, we support IPv4 w/ TCP only */
6246 if (protocol != htons(ETH_P_IP) ||
6247 iph->protocol != IPPROTO_TCP)
6250 memset(&atr_input, 0, sizeof(struct ixgbe_atr_input));
6252 vlan_id = (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK) >>
6253 IXGBE_TX_FLAGS_VLAN_SHIFT;
6257 ixgbe_atr_set_vlan_id_82599(&atr_input, vlan_id);
6258 ixgbe_atr_set_src_port_82599(&atr_input, th->dest);
6259 ixgbe_atr_set_dst_port_82599(&atr_input, th->source);
6260 ixgbe_atr_set_flex_byte_82599(&atr_input, eth->h_proto);
6261 ixgbe_atr_set_l4type_82599(&atr_input, IXGBE_ATR_L4TYPE_TCP);
6262 /* src and dst are inverted, think how the receiver sees them */
6263 ixgbe_atr_set_src_ipv4_82599(&atr_input, iph->daddr);
6264 ixgbe_atr_set_dst_ipv4_82599(&atr_input, iph->saddr);
6266 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
6267 ixgbe_fdir_add_signature_filter_82599(&adapter->hw, &atr_input, queue);
6270 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, int size)
6272 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
6273 /* Herbert's original patch had:
6274 * smp_mb__after_netif_stop_queue();
6275 * but since that doesn't exist yet, just open code it. */
6278 /* We need to check again in a case another CPU has just
6279 * made room available. */
6280 if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
6283 /* A reprieve! - use start_queue because it doesn't call schedule */
6284 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
6285 ++tx_ring->tx_stats.restart_queue;
6289 static int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, int size)
6291 if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
6293 return __ixgbe_maybe_stop_tx(tx_ring, size);
6296 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6298 struct ixgbe_adapter *adapter = netdev_priv(dev);
6299 int txq = smp_processor_id();
6303 protocol = vlan_get_protocol(skb);
6305 if ((protocol == htons(ETH_P_FCOE)) ||
6306 (protocol == htons(ETH_P_FIP))) {
6307 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
6308 txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
6309 txq += adapter->ring_feature[RING_F_FCOE].mask;
6311 #ifdef CONFIG_IXGBE_DCB
6312 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6313 txq = adapter->fcoe.up;
6320 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6321 while (unlikely(txq >= dev->real_num_tx_queues))
6322 txq -= dev->real_num_tx_queues;
6326 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6327 if (skb->priority == TC_PRIO_CONTROL)
6328 txq = adapter->ring_feature[RING_F_DCB].indices-1;
6330 txq = (skb->vlan_tci & IXGBE_TX_FLAGS_VLAN_PRIO_MASK)
6335 return skb_tx_hash(dev, skb);
6338 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
6339 struct ixgbe_adapter *adapter,
6340 struct ixgbe_ring *tx_ring)
6342 struct net_device *netdev = tx_ring->netdev;
6343 struct netdev_queue *txq;
6345 unsigned int tx_flags = 0;
6352 protocol = vlan_get_protocol(skb);
6354 if (vlan_tx_tag_present(skb)) {
6355 tx_flags |= vlan_tx_tag_get(skb);
6356 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6357 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6358 tx_flags |= ((skb->queue_mapping & 0x7) << 13);
6360 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6361 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6362 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED &&
6363 skb->priority != TC_PRIO_CONTROL) {
6364 tx_flags |= ((skb->queue_mapping & 0x7) << 13);
6365 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6366 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6370 /* for FCoE with DCB, we force the priority to what
6371 * was specified by the switch */
6372 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED &&
6373 (protocol == htons(ETH_P_FCOE) ||
6374 protocol == htons(ETH_P_FIP))) {
6375 #ifdef CONFIG_IXGBE_DCB
6376 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6377 tx_flags &= ~(IXGBE_TX_FLAGS_VLAN_PRIO_MASK
6378 << IXGBE_TX_FLAGS_VLAN_SHIFT);
6379 tx_flags |= ((adapter->fcoe.up << 13)
6380 << IXGBE_TX_FLAGS_VLAN_SHIFT);
6383 /* flag for FCoE offloads */
6384 if (protocol == htons(ETH_P_FCOE))
6385 tx_flags |= IXGBE_TX_FLAGS_FCOE;
6389 /* four things can cause us to need a context descriptor */
6390 if (skb_is_gso(skb) ||
6391 (skb->ip_summed == CHECKSUM_PARTIAL) ||
6392 (tx_flags & IXGBE_TX_FLAGS_VLAN) ||
6393 (tx_flags & IXGBE_TX_FLAGS_FCOE))
6396 count += TXD_USE_COUNT(skb_headlen(skb));
6397 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6398 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6400 if (ixgbe_maybe_stop_tx(tx_ring, count)) {
6401 tx_ring->tx_stats.tx_busy++;
6402 return NETDEV_TX_BUSY;
6405 first = tx_ring->next_to_use;
6406 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6408 /* setup tx offload for FCoE */
6409 tso = ixgbe_fso(adapter, tx_ring, skb, tx_flags, &hdr_len);
6411 dev_kfree_skb_any(skb);
6412 return NETDEV_TX_OK;
6415 tx_flags |= IXGBE_TX_FLAGS_FSO;
6416 #endif /* IXGBE_FCOE */
6418 if (protocol == htons(ETH_P_IP))
6419 tx_flags |= IXGBE_TX_FLAGS_IPV4;
6420 tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len,
6423 dev_kfree_skb_any(skb);
6424 return NETDEV_TX_OK;
6428 tx_flags |= IXGBE_TX_FLAGS_TSO;
6429 else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags,
6431 (skb->ip_summed == CHECKSUM_PARTIAL))
6432 tx_flags |= IXGBE_TX_FLAGS_CSUM;
6435 count = ixgbe_tx_map(adapter, tx_ring, skb, tx_flags, first, hdr_len);
6437 /* add the ATR filter if ATR is on */
6438 if (tx_ring->atr_sample_rate) {
6439 ++tx_ring->atr_count;
6440 if ((tx_ring->atr_count >= tx_ring->atr_sample_rate) &&
6441 test_bit(__IXGBE_TX_FDIR_INIT_DONE,
6443 ixgbe_atr(adapter, skb, tx_ring->queue_index,
6444 tx_flags, protocol);
6445 tx_ring->atr_count = 0;
6448 txq = netdev_get_tx_queue(netdev, tx_ring->queue_index);
6449 txq->tx_bytes += skb->len;
6451 ixgbe_tx_queue(tx_ring, tx_flags, count, skb->len, hdr_len);
6452 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
6455 dev_kfree_skb_any(skb);
6456 tx_ring->tx_buffer_info[first].time_stamp = 0;
6457 tx_ring->next_to_use = first;
6460 return NETDEV_TX_OK;
6463 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
6465 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6466 struct ixgbe_ring *tx_ring;
6468 tx_ring = adapter->tx_ring[skb->queue_mapping];
6469 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
6473 * ixgbe_set_mac - Change the Ethernet Address of the NIC
6474 * @netdev: network interface device structure
6475 * @p: pointer to an address structure
6477 * Returns 0 on success, negative on failure
6479 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6481 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6482 struct ixgbe_hw *hw = &adapter->hw;
6483 struct sockaddr *addr = p;
6485 if (!is_valid_ether_addr(addr->sa_data))
6486 return -EADDRNOTAVAIL;
6488 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6489 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6491 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
6498 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6500 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6501 struct ixgbe_hw *hw = &adapter->hw;
6505 if (prtad != hw->phy.mdio.prtad)
6507 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6513 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6514 u16 addr, u16 value)
6516 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6517 struct ixgbe_hw *hw = &adapter->hw;
6519 if (prtad != hw->phy.mdio.prtad)
6521 return hw->phy.ops.write_reg(hw, addr, devad, value);
6524 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6526 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6528 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6532 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6534 * @netdev: network interface device structure
6536 * Returns non-zero on failure
6538 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6541 struct ixgbe_adapter *adapter = netdev_priv(dev);
6542 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6544 if (is_valid_ether_addr(mac->san_addr)) {
6546 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6553 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6555 * @netdev: network interface device structure
6557 * Returns non-zero on failure
6559 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6562 struct ixgbe_adapter *adapter = netdev_priv(dev);
6563 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6565 if (is_valid_ether_addr(mac->san_addr)) {
6567 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6573 #ifdef CONFIG_NET_POLL_CONTROLLER
6575 * Polling 'interrupt' - used by things like netconsole to send skbs
6576 * without having to re-enable interrupts. It's not called while
6577 * the interrupt routine is executing.
6579 static void ixgbe_netpoll(struct net_device *netdev)
6581 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6584 /* if interface is down do nothing */
6585 if (test_bit(__IXGBE_DOWN, &adapter->state))
6588 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
6589 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
6590 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
6591 for (i = 0; i < num_q_vectors; i++) {
6592 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
6593 ixgbe_msix_clean_many(0, q_vector);
6596 ixgbe_intr(adapter->pdev->irq, netdev);
6598 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
6602 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
6603 struct rtnl_link_stats64 *stats)
6605 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6608 /* accurate rx/tx bytes/packets stats */
6609 dev_txq_stats_fold(netdev, stats);
6611 for (i = 0; i < adapter->num_rx_queues; i++) {
6612 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
6618 start = u64_stats_fetch_begin_bh(&ring->syncp);
6619 packets = ring->stats.packets;
6620 bytes = ring->stats.bytes;
6621 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6622 stats->rx_packets += packets;
6623 stats->rx_bytes += bytes;
6627 /* following stats updated by ixgbe_watchdog_task() */
6628 stats->multicast = netdev->stats.multicast;
6629 stats->rx_errors = netdev->stats.rx_errors;
6630 stats->rx_length_errors = netdev->stats.rx_length_errors;
6631 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
6632 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
6637 static const struct net_device_ops ixgbe_netdev_ops = {
6638 .ndo_open = ixgbe_open,
6639 .ndo_stop = ixgbe_close,
6640 .ndo_start_xmit = ixgbe_xmit_frame,
6641 .ndo_select_queue = ixgbe_select_queue,
6642 .ndo_set_rx_mode = ixgbe_set_rx_mode,
6643 .ndo_set_multicast_list = ixgbe_set_rx_mode,
6644 .ndo_validate_addr = eth_validate_addr,
6645 .ndo_set_mac_address = ixgbe_set_mac,
6646 .ndo_change_mtu = ixgbe_change_mtu,
6647 .ndo_tx_timeout = ixgbe_tx_timeout,
6648 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
6649 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
6650 .ndo_do_ioctl = ixgbe_ioctl,
6651 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
6652 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
6653 .ndo_set_vf_tx_rate = ixgbe_ndo_set_vf_bw,
6654 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
6655 .ndo_get_stats64 = ixgbe_get_stats64,
6656 #ifdef CONFIG_NET_POLL_CONTROLLER
6657 .ndo_poll_controller = ixgbe_netpoll,
6660 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
6661 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
6662 .ndo_fcoe_enable = ixgbe_fcoe_enable,
6663 .ndo_fcoe_disable = ixgbe_fcoe_disable,
6664 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
6665 #endif /* IXGBE_FCOE */
6668 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
6669 const struct ixgbe_info *ii)
6671 #ifdef CONFIG_PCI_IOV
6672 struct ixgbe_hw *hw = &adapter->hw;
6675 if (hw->mac.type != ixgbe_mac_82599EB || !max_vfs)
6678 /* The 82599 supports up to 64 VFs per physical function
6679 * but this implementation limits allocation to 63 so that
6680 * basic networking resources are still available to the
6683 adapter->num_vfs = (max_vfs > 63) ? 63 : max_vfs;
6684 adapter->flags |= IXGBE_FLAG_SRIOV_ENABLED;
6685 err = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
6687 e_err(probe, "Failed to enable PCI sriov: %d\n", err);
6690 /* If call to enable VFs succeeded then allocate memory
6691 * for per VF control structures.
6694 kcalloc(adapter->num_vfs,
6695 sizeof(struct vf_data_storage), GFP_KERNEL);
6696 if (adapter->vfinfo) {
6697 /* Now that we're sure SR-IOV is enabled
6698 * and memory allocated set up the mailbox parameters
6700 ixgbe_init_mbx_params_pf(hw);
6701 memcpy(&hw->mbx.ops, ii->mbx_ops,
6702 sizeof(hw->mbx.ops));
6704 /* Disable RSC when in SR-IOV mode */
6705 adapter->flags2 &= ~(IXGBE_FLAG2_RSC_CAPABLE |
6706 IXGBE_FLAG2_RSC_ENABLED);
6711 e_err(probe, "Unable to allocate memory for VF Data Storage - "
6712 "SRIOV disabled\n");
6713 pci_disable_sriov(adapter->pdev);
6716 adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
6717 adapter->num_vfs = 0;
6718 #endif /* CONFIG_PCI_IOV */
6722 * ixgbe_probe - Device Initialization Routine
6723 * @pdev: PCI device information struct
6724 * @ent: entry in ixgbe_pci_tbl
6726 * Returns 0 on success, negative on failure
6728 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
6729 * The OS initialization, configuring of the adapter private structure,
6730 * and a hardware reset occur.
6732 static int __devinit ixgbe_probe(struct pci_dev *pdev,
6733 const struct pci_device_id *ent)
6735 struct net_device *netdev;
6736 struct ixgbe_adapter *adapter = NULL;
6737 struct ixgbe_hw *hw;
6738 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
6739 static int cards_found;
6740 int i, err, pci_using_dac;
6741 unsigned int indices = num_possible_cpus();
6747 /* Catch broken hardware that put the wrong VF device ID in
6748 * the PCIe SR-IOV capability.
6750 if (pdev->is_virtfn) {
6751 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
6752 pci_name(pdev), pdev->vendor, pdev->device);
6756 err = pci_enable_device_mem(pdev);
6760 if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
6761 !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
6764 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
6766 err = dma_set_coherent_mask(&pdev->dev,
6770 "No usable DMA configuration, aborting\n");
6777 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
6778 IORESOURCE_MEM), ixgbe_driver_name);
6781 "pci_request_selected_regions failed 0x%x\n", err);
6785 pci_enable_pcie_error_reporting(pdev);
6787 pci_set_master(pdev);
6788 pci_save_state(pdev);
6790 if (ii->mac == ixgbe_mac_82598EB)
6791 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
6793 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
6795 indices = max_t(unsigned int, indices, IXGBE_MAX_DCB_INDICES);
6797 indices += min_t(unsigned int, num_possible_cpus(),
6798 IXGBE_MAX_FCOE_INDICES);
6800 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
6803 goto err_alloc_etherdev;
6806 SET_NETDEV_DEV(netdev, &pdev->dev);
6808 adapter = netdev_priv(netdev);
6809 pci_set_drvdata(pdev, adapter);
6811 adapter->netdev = netdev;
6812 adapter->pdev = pdev;
6815 adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
6817 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
6818 pci_resource_len(pdev, 0));
6824 for (i = 1; i <= 5; i++) {
6825 if (pci_resource_len(pdev, i) == 0)
6829 netdev->netdev_ops = &ixgbe_netdev_ops;
6830 ixgbe_set_ethtool_ops(netdev);
6831 netdev->watchdog_timeo = 5 * HZ;
6832 strcpy(netdev->name, pci_name(pdev));
6834 adapter->bd_number = cards_found;
6837 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
6838 hw->mac.type = ii->mac;
6841 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
6842 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
6843 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
6844 if (!(eec & (1 << 8)))
6845 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
6848 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
6849 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
6850 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
6851 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
6852 hw->phy.mdio.mmds = 0;
6853 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
6854 hw->phy.mdio.dev = netdev;
6855 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
6856 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
6858 /* set up this timer and work struct before calling get_invariants
6859 * which might start the timer
6861 init_timer(&adapter->sfp_timer);
6862 adapter->sfp_timer.function = ixgbe_sfp_timer;
6863 adapter->sfp_timer.data = (unsigned long) adapter;
6865 INIT_WORK(&adapter->sfp_task, ixgbe_sfp_task);
6867 /* multispeed fiber has its own tasklet, called from GPI SDP1 context */
6868 INIT_WORK(&adapter->multispeed_fiber_task, ixgbe_multispeed_fiber_task);
6870 /* a new SFP+ module arrival, called from GPI SDP2 context */
6871 INIT_WORK(&adapter->sfp_config_module_task,
6872 ixgbe_sfp_config_module_task);
6874 ii->get_invariants(hw);
6876 /* setup the private structure */
6877 err = ixgbe_sw_init(adapter);
6881 /* Make it possible the adapter to be woken up via WOL */
6882 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
6883 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6886 * If there is a fan on this device and it has failed log the
6889 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
6890 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
6891 if (esdp & IXGBE_ESDP_SDP1)
6892 e_crit(probe, "Fan has stopped, replace the adapter\n");
6895 /* reset_hw fills in the perm_addr as well */
6896 hw->phy.reset_if_overtemp = true;
6897 err = hw->mac.ops.reset_hw(hw);
6898 hw->phy.reset_if_overtemp = false;
6899 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
6900 hw->mac.type == ixgbe_mac_82598EB) {
6902 * Start a kernel thread to watch for a module to arrive.
6903 * Only do this for 82598, since 82599 will generate
6904 * interrupts on module arrival.
6906 set_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
6907 mod_timer(&adapter->sfp_timer,
6908 round_jiffies(jiffies + (2 * HZ)));
6910 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
6911 e_dev_err("failed to initialize because an unsupported SFP+ "
6912 "module type was detected.\n");
6913 e_dev_err("Reload the driver after installing a supported "
6917 e_dev_err("HW Init failed: %d\n", err);
6921 ixgbe_probe_vf(adapter, ii);
6923 netdev->features = NETIF_F_SG |
6925 NETIF_F_HW_VLAN_TX |
6926 NETIF_F_HW_VLAN_RX |
6927 NETIF_F_HW_VLAN_FILTER;
6929 netdev->features |= NETIF_F_IPV6_CSUM;
6930 netdev->features |= NETIF_F_TSO;
6931 netdev->features |= NETIF_F_TSO6;
6932 netdev->features |= NETIF_F_GRO;
6934 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
6935 netdev->features |= NETIF_F_SCTP_CSUM;
6937 netdev->vlan_features |= NETIF_F_TSO;
6938 netdev->vlan_features |= NETIF_F_TSO6;
6939 netdev->vlan_features |= NETIF_F_IP_CSUM;
6940 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
6941 netdev->vlan_features |= NETIF_F_SG;
6943 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6944 adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
6945 IXGBE_FLAG_DCB_ENABLED);
6946 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
6947 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
6949 #ifdef CONFIG_IXGBE_DCB
6950 netdev->dcbnl_ops = &dcbnl_ops;
6954 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
6955 if (hw->mac.ops.get_device_caps) {
6956 hw->mac.ops.get_device_caps(hw, &device_caps);
6957 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
6958 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6961 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
6962 netdev->vlan_features |= NETIF_F_FCOE_CRC;
6963 netdev->vlan_features |= NETIF_F_FSO;
6964 netdev->vlan_features |= NETIF_F_FCOE_MTU;
6966 #endif /* IXGBE_FCOE */
6967 if (pci_using_dac) {
6968 netdev->features |= NETIF_F_HIGHDMA;
6969 netdev->vlan_features |= NETIF_F_HIGHDMA;
6972 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
6973 netdev->features |= NETIF_F_LRO;
6975 /* make sure the EEPROM is good */
6976 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
6977 e_dev_err("The EEPROM Checksum Is Not Valid\n");
6982 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
6983 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
6985 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
6986 e_dev_err("invalid MAC address\n");
6991 /* power down the optics */
6992 if (hw->phy.multispeed_fiber)
6993 hw->mac.ops.disable_tx_laser(hw);
6995 init_timer(&adapter->watchdog_timer);
6996 adapter->watchdog_timer.function = ixgbe_watchdog;
6997 adapter->watchdog_timer.data = (unsigned long)adapter;
6999 INIT_WORK(&adapter->reset_task, ixgbe_reset_task);
7000 INIT_WORK(&adapter->watchdog_task, ixgbe_watchdog_task);
7002 err = ixgbe_init_interrupt_scheme(adapter);
7006 switch (pdev->device) {
7007 case IXGBE_DEV_ID_82599_KX4:
7008 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
7009 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
7015 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
7017 /* pick up the PCI bus settings for reporting later */
7018 hw->mac.ops.get_bus_info(hw);
7020 /* print bus type/speed/width info */
7021 e_dev_info("(PCI Express:%s:%s) %pM\n",
7022 (hw->bus.speed == ixgbe_bus_speed_5000 ? "5.0Gb/s" :
7023 hw->bus.speed == ixgbe_bus_speed_2500 ? "2.5Gb/s" :
7025 (hw->bus.width == ixgbe_bus_width_pcie_x8 ? "Width x8" :
7026 hw->bus.width == ixgbe_bus_width_pcie_x4 ? "Width x4" :
7027 hw->bus.width == ixgbe_bus_width_pcie_x1 ? "Width x1" :
7030 ixgbe_read_pba_num_generic(hw, &part_num);
7031 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
7032 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, "
7033 "PBA No: %06x-%03x\n",
7034 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
7035 (part_num >> 8), (part_num & 0xff));
7037 e_dev_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
7038 hw->mac.type, hw->phy.type,
7039 (part_num >> 8), (part_num & 0xff));
7041 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
7042 e_dev_warn("PCI-Express bandwidth available for this card is "
7043 "not sufficient for optimal performance.\n");
7044 e_dev_warn("For optimal performance a x8 PCI-Express slot "
7048 /* save off EEPROM version number */
7049 hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
7051 /* reset the hardware with the new settings */
7052 err = hw->mac.ops.start_hw(hw);
7054 if (err == IXGBE_ERR_EEPROM_VERSION) {
7055 /* We are running on a pre-production device, log a warning */
7056 e_dev_warn("This device is a pre-production adapter/LOM. "
7057 "Please be aware there may be issues associated "
7058 "with your hardware. If you are experiencing "
7059 "problems please contact your Intel or hardware "
7060 "representative who provided you with this "
7063 strcpy(netdev->name, "eth%d");
7064 err = register_netdev(netdev);
7068 /* carrier off reporting is important to ethtool even BEFORE open */
7069 netif_carrier_off(netdev);
7071 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
7072 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
7073 INIT_WORK(&adapter->fdir_reinit_task, ixgbe_fdir_reinit_task);
7075 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
7076 INIT_WORK(&adapter->check_overtemp_task,
7077 ixgbe_check_overtemp_task);
7078 #ifdef CONFIG_IXGBE_DCA
7079 if (dca_add_requester(&pdev->dev) == 0) {
7080 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
7081 ixgbe_setup_dca(adapter);
7084 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7085 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
7086 for (i = 0; i < adapter->num_vfs; i++)
7087 ixgbe_vf_configuration(pdev, (i | 0x10000000));
7090 /* add san mac addr to netdev */
7091 ixgbe_add_sanmac_netdev(netdev);
7093 e_dev_info("Intel(R) 10 Gigabit Network Connection\n");
7098 ixgbe_release_hw_control(adapter);
7099 ixgbe_clear_interrupt_scheme(adapter);
7102 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7103 ixgbe_disable_sriov(adapter);
7104 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
7105 del_timer_sync(&adapter->sfp_timer);
7106 cancel_work_sync(&adapter->sfp_task);
7107 cancel_work_sync(&adapter->multispeed_fiber_task);
7108 cancel_work_sync(&adapter->sfp_config_module_task);
7109 iounmap(hw->hw_addr);
7111 free_netdev(netdev);
7113 pci_release_selected_regions(pdev,
7114 pci_select_bars(pdev, IORESOURCE_MEM));
7117 pci_disable_device(pdev);
7122 * ixgbe_remove - Device Removal Routine
7123 * @pdev: PCI device information struct
7125 * ixgbe_remove is called by the PCI subsystem to alert the driver
7126 * that it should release a PCI device. The could be caused by a
7127 * Hot-Plug event, or because the driver is going to be removed from
7130 static void __devexit ixgbe_remove(struct pci_dev *pdev)
7132 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7133 struct net_device *netdev = adapter->netdev;
7135 set_bit(__IXGBE_DOWN, &adapter->state);
7136 /* clear the module not found bit to make sure the worker won't
7139 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
7140 del_timer_sync(&adapter->watchdog_timer);
7142 del_timer_sync(&adapter->sfp_timer);
7143 cancel_work_sync(&adapter->watchdog_task);
7144 cancel_work_sync(&adapter->sfp_task);
7145 cancel_work_sync(&adapter->multispeed_fiber_task);
7146 cancel_work_sync(&adapter->sfp_config_module_task);
7147 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
7148 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
7149 cancel_work_sync(&adapter->fdir_reinit_task);
7150 flush_scheduled_work();
7152 #ifdef CONFIG_IXGBE_DCA
7153 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
7154 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
7155 dca_remove_requester(&pdev->dev);
7156 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
7161 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
7162 ixgbe_cleanup_fcoe(adapter);
7164 #endif /* IXGBE_FCOE */
7166 /* remove the added san mac */
7167 ixgbe_del_sanmac_netdev(netdev);
7169 if (netdev->reg_state == NETREG_REGISTERED)
7170 unregister_netdev(netdev);
7172 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7173 ixgbe_disable_sriov(adapter);
7175 ixgbe_clear_interrupt_scheme(adapter);
7177 ixgbe_release_hw_control(adapter);
7179 iounmap(adapter->hw.hw_addr);
7180 pci_release_selected_regions(pdev, pci_select_bars(pdev,
7183 e_dev_info("complete\n");
7185 free_netdev(netdev);
7187 pci_disable_pcie_error_reporting(pdev);
7189 pci_disable_device(pdev);
7193 * ixgbe_io_error_detected - called when PCI error is detected
7194 * @pdev: Pointer to PCI device
7195 * @state: The current pci connection state
7197 * This function is called after a PCI bus error affecting
7198 * this device has been detected.
7200 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
7201 pci_channel_state_t state)
7203 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7204 struct net_device *netdev = adapter->netdev;
7206 netif_device_detach(netdev);
7208 if (state == pci_channel_io_perm_failure)
7209 return PCI_ERS_RESULT_DISCONNECT;
7211 if (netif_running(netdev))
7212 ixgbe_down(adapter);
7213 pci_disable_device(pdev);
7215 /* Request a slot reset. */
7216 return PCI_ERS_RESULT_NEED_RESET;
7220 * ixgbe_io_slot_reset - called after the pci bus has been reset.
7221 * @pdev: Pointer to PCI device
7223 * Restart the card from scratch, as if from a cold-boot.
7225 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7227 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7228 pci_ers_result_t result;
7231 if (pci_enable_device_mem(pdev)) {
7232 e_err(probe, "Cannot re-enable PCI device after reset.\n");
7233 result = PCI_ERS_RESULT_DISCONNECT;
7235 pci_set_master(pdev);
7236 pci_restore_state(pdev);
7237 pci_save_state(pdev);
7239 pci_wake_from_d3(pdev, false);
7241 ixgbe_reset(adapter);
7242 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7243 result = PCI_ERS_RESULT_RECOVERED;
7246 err = pci_cleanup_aer_uncorrect_error_status(pdev);
7248 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7249 "failed 0x%0x\n", err);
7250 /* non-fatal, continue */
7257 * ixgbe_io_resume - called when traffic can start flowing again.
7258 * @pdev: Pointer to PCI device
7260 * This callback is called when the error recovery driver tells us that
7261 * its OK to resume normal operation.
7263 static void ixgbe_io_resume(struct pci_dev *pdev)
7265 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7266 struct net_device *netdev = adapter->netdev;
7268 if (netif_running(netdev)) {
7269 if (ixgbe_up(adapter)) {
7270 e_info(probe, "ixgbe_up failed after reset\n");
7275 netif_device_attach(netdev);
7278 static struct pci_error_handlers ixgbe_err_handler = {
7279 .error_detected = ixgbe_io_error_detected,
7280 .slot_reset = ixgbe_io_slot_reset,
7281 .resume = ixgbe_io_resume,
7284 static struct pci_driver ixgbe_driver = {
7285 .name = ixgbe_driver_name,
7286 .id_table = ixgbe_pci_tbl,
7287 .probe = ixgbe_probe,
7288 .remove = __devexit_p(ixgbe_remove),
7290 .suspend = ixgbe_suspend,
7291 .resume = ixgbe_resume,
7293 .shutdown = ixgbe_shutdown,
7294 .err_handler = &ixgbe_err_handler
7298 * ixgbe_init_module - Driver Registration Routine
7300 * ixgbe_init_module is the first routine called when the driver is
7301 * loaded. All it does is register with the PCI subsystem.
7303 static int __init ixgbe_init_module(void)
7306 pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
7307 pr_info("%s\n", ixgbe_copyright);
7309 #ifdef CONFIG_IXGBE_DCA
7310 dca_register_notify(&dca_notifier);
7313 ret = pci_register_driver(&ixgbe_driver);
7317 module_init(ixgbe_init_module);
7320 * ixgbe_exit_module - Driver Exit Cleanup Routine
7322 * ixgbe_exit_module is called just before the driver is removed
7325 static void __exit ixgbe_exit_module(void)
7327 #ifdef CONFIG_IXGBE_DCA
7328 dca_unregister_notify(&dca_notifier);
7330 pci_unregister_driver(&ixgbe_driver);
7331 rcu_barrier(); /* Wait for completion of call_rcu()'s */
7334 #ifdef CONFIG_IXGBE_DCA
7335 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
7340 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
7341 __ixgbe_notify_dca);
7343 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7346 #endif /* CONFIG_IXGBE_DCA */
7349 * ixgbe_get_hw_dev return device
7350 * used by hardware layer to print debugging information
7352 struct net_device *ixgbe_get_hw_dev(struct ixgbe_hw *hw)
7354 struct ixgbe_adapter *adapter = hw->back;
7355 return adapter->netdev;
7358 module_exit(ixgbe_exit_module);