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;
1146 * ixgbe_transform_rsc_queue - change rsc queue into a full packet
1147 * @skb: pointer to the last skb in the rsc queue
1149 * This function changes a queue full of hw rsc buffers into a completed
1150 * packet. It uses the ->prev pointers to find the first packet and then
1151 * turns it into the frag list owner.
1153 static inline struct sk_buff *ixgbe_transform_rsc_queue(struct sk_buff *skb)
1155 unsigned int frag_list_size = 0;
1156 unsigned int skb_cnt = 1;
1159 struct sk_buff *prev = skb->prev;
1160 frag_list_size += skb->len;
1166 skb_shinfo(skb)->frag_list = skb->next;
1168 skb->len += frag_list_size;
1169 skb->data_len += frag_list_size;
1170 skb->truesize += frag_list_size;
1171 IXGBE_RSC_CB(skb)->skb_cnt = skb_cnt;
1176 static inline bool ixgbe_get_rsc_state(union ixgbe_adv_rx_desc *rx_desc)
1178 return !!(le32_to_cpu(rx_desc->wb.lower.lo_dword.data) &
1179 IXGBE_RXDADV_RSCCNT_MASK);
1182 static void ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1183 struct ixgbe_ring *rx_ring,
1184 int *work_done, int work_to_do)
1186 struct ixgbe_adapter *adapter = q_vector->adapter;
1187 union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
1188 struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
1189 struct sk_buff *skb;
1190 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1191 const int current_node = numa_node_id();
1194 #endif /* IXGBE_FCOE */
1197 u16 cleaned_count = 0;
1198 bool pkt_is_rsc = false;
1200 i = rx_ring->next_to_clean;
1201 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1202 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1204 while (staterr & IXGBE_RXD_STAT_DD) {
1207 rmb(); /* read descriptor and rx_buffer_info after status DD */
1209 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1211 skb = rx_buffer_info->skb;
1212 rx_buffer_info->skb = NULL;
1213 prefetch(skb->data);
1215 if (ring_is_rsc_enabled(rx_ring))
1216 pkt_is_rsc = ixgbe_get_rsc_state(rx_desc);
1218 /* if this is a skb from previous receive DMA will be 0 */
1219 if (rx_buffer_info->dma) {
1222 !(staterr & IXGBE_RXD_STAT_EOP) &&
1225 * When HWRSC is enabled, delay unmapping
1226 * of the first packet. It carries the
1227 * header information, HW may still
1228 * access the header after the writeback.
1229 * Only unmap it when EOP is reached
1231 IXGBE_RSC_CB(skb)->delay_unmap = true;
1232 IXGBE_RSC_CB(skb)->dma = rx_buffer_info->dma;
1234 dma_unmap_single(rx_ring->dev,
1235 rx_buffer_info->dma,
1236 rx_ring->rx_buf_len,
1239 rx_buffer_info->dma = 0;
1241 if (ring_is_ps_enabled(rx_ring)) {
1242 hlen = ixgbe_get_hlen(rx_desc);
1243 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1245 hlen = le16_to_cpu(rx_desc->wb.upper.length);
1250 /* assume packet split since header is unmapped */
1251 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1255 dma_unmap_page(rx_ring->dev,
1256 rx_buffer_info->page_dma,
1259 rx_buffer_info->page_dma = 0;
1260 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1261 rx_buffer_info->page,
1262 rx_buffer_info->page_offset,
1265 if ((page_count(rx_buffer_info->page) == 1) &&
1266 (page_to_nid(rx_buffer_info->page) == current_node))
1267 get_page(rx_buffer_info->page);
1269 rx_buffer_info->page = NULL;
1271 skb->len += upper_len;
1272 skb->data_len += upper_len;
1273 skb->truesize += upper_len;
1277 if (i == rx_ring->count)
1280 next_rxd = IXGBE_RX_DESC_ADV(rx_ring, i);
1285 u32 nextp = (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
1286 IXGBE_RXDADV_NEXTP_SHIFT;
1287 next_buffer = &rx_ring->rx_buffer_info[nextp];
1289 next_buffer = &rx_ring->rx_buffer_info[i];
1292 if (!(staterr & IXGBE_RXD_STAT_EOP)) {
1293 if (ring_is_ps_enabled(rx_ring)) {
1294 rx_buffer_info->skb = next_buffer->skb;
1295 rx_buffer_info->dma = next_buffer->dma;
1296 next_buffer->skb = skb;
1297 next_buffer->dma = 0;
1299 skb->next = next_buffer->skb;
1300 skb->next->prev = skb;
1302 rx_ring->rx_stats.non_eop_descs++;
1307 skb = ixgbe_transform_rsc_queue(skb);
1308 /* if we got here without RSC the packet is invalid */
1310 __pskb_trim(skb, 0);
1311 rx_buffer_info->skb = skb;
1316 if (ring_is_rsc_enabled(rx_ring)) {
1317 if (IXGBE_RSC_CB(skb)->delay_unmap) {
1318 dma_unmap_single(rx_ring->dev,
1319 IXGBE_RSC_CB(skb)->dma,
1320 rx_ring->rx_buf_len,
1322 IXGBE_RSC_CB(skb)->dma = 0;
1323 IXGBE_RSC_CB(skb)->delay_unmap = false;
1327 if (ring_is_ps_enabled(rx_ring))
1328 rx_ring->rx_stats.rsc_count +=
1329 skb_shinfo(skb)->nr_frags;
1331 rx_ring->rx_stats.rsc_count +=
1332 IXGBE_RSC_CB(skb)->skb_cnt;
1333 rx_ring->rx_stats.rsc_flush++;
1336 /* ERR_MASK will only have valid bits if EOP set */
1337 if (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) {
1338 /* trim packet back to size 0 and recycle it */
1339 __pskb_trim(skb, 0);
1340 rx_buffer_info->skb = skb;
1344 ixgbe_rx_checksum(adapter, rx_desc, skb);
1346 /* probably a little skewed due to removing CRC */
1347 total_rx_bytes += skb->len;
1350 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1352 /* if ddp, not passing to ULD unless for FCP_RSP or error */
1353 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
1354 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
1358 #endif /* IXGBE_FCOE */
1359 ixgbe_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
1362 rx_desc->wb.upper.status_error = 0;
1365 if (*work_done >= work_to_do)
1368 /* return some buffers to hardware, one at a time is too slow */
1369 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
1370 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1374 /* use prefetched values */
1376 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1379 rx_ring->next_to_clean = i;
1380 cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
1383 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1386 /* include DDPed FCoE data */
1387 if (ddp_bytes > 0) {
1390 mss = rx_ring->netdev->mtu - sizeof(struct fcoe_hdr) -
1391 sizeof(struct fc_frame_header) -
1392 sizeof(struct fcoe_crc_eof);
1395 total_rx_bytes += ddp_bytes;
1396 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
1398 #endif /* IXGBE_FCOE */
1400 rx_ring->total_packets += total_rx_packets;
1401 rx_ring->total_bytes += total_rx_bytes;
1402 u64_stats_update_begin(&rx_ring->syncp);
1403 rx_ring->stats.packets += total_rx_packets;
1404 rx_ring->stats.bytes += total_rx_bytes;
1405 u64_stats_update_end(&rx_ring->syncp);
1408 static int ixgbe_clean_rxonly(struct napi_struct *, int);
1410 * ixgbe_configure_msix - Configure MSI-X hardware
1411 * @adapter: board private structure
1413 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1416 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1418 struct ixgbe_q_vector *q_vector;
1419 int i, j, q_vectors, v_idx, r_idx;
1422 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1425 * Populate the IVAR table and set the ITR values to the
1426 * corresponding register.
1428 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1429 q_vector = adapter->q_vector[v_idx];
1430 /* XXX for_each_set_bit(...) */
1431 r_idx = find_first_bit(q_vector->rxr_idx,
1432 adapter->num_rx_queues);
1434 for (i = 0; i < q_vector->rxr_count; i++) {
1435 j = adapter->rx_ring[r_idx]->reg_idx;
1436 ixgbe_set_ivar(adapter, 0, j, v_idx);
1437 r_idx = find_next_bit(q_vector->rxr_idx,
1438 adapter->num_rx_queues,
1441 r_idx = find_first_bit(q_vector->txr_idx,
1442 adapter->num_tx_queues);
1444 for (i = 0; i < q_vector->txr_count; i++) {
1445 j = adapter->tx_ring[r_idx]->reg_idx;
1446 ixgbe_set_ivar(adapter, 1, j, v_idx);
1447 r_idx = find_next_bit(q_vector->txr_idx,
1448 adapter->num_tx_queues,
1452 if (q_vector->txr_count && !q_vector->rxr_count)
1454 q_vector->eitr = adapter->tx_eitr_param;
1455 else if (q_vector->rxr_count)
1457 q_vector->eitr = adapter->rx_eitr_param;
1459 ixgbe_write_eitr(q_vector);
1460 /* If Flow Director is enabled, set interrupt affinity */
1461 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
1462 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)) {
1464 * Allocate the affinity_hint cpumask, assign the mask
1465 * for this vector, and set our affinity_hint for
1468 if (!alloc_cpumask_var(&q_vector->affinity_mask,
1471 cpumask_set_cpu(v_idx, q_vector->affinity_mask);
1472 irq_set_affinity_hint(adapter->msix_entries[v_idx].vector,
1473 q_vector->affinity_mask);
1477 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
1478 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1480 else if (adapter->hw.mac.type == ixgbe_mac_82599EB)
1481 ixgbe_set_ivar(adapter, -1, 1, v_idx);
1482 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1484 /* set up to autoclear timer, and the vectors */
1485 mask = IXGBE_EIMS_ENABLE_MASK;
1486 if (adapter->num_vfs)
1487 mask &= ~(IXGBE_EIMS_OTHER |
1488 IXGBE_EIMS_MAILBOX |
1491 mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
1492 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1495 enum latency_range {
1499 latency_invalid = 255
1503 * ixgbe_update_itr - update the dynamic ITR value based on statistics
1504 * @adapter: pointer to adapter
1505 * @eitr: eitr setting (ints per sec) to give last timeslice
1506 * @itr_setting: current throttle rate in ints/second
1507 * @packets: the number of packets during this measurement interval
1508 * @bytes: the number of bytes during this measurement interval
1510 * Stores a new ITR value based on packets and byte
1511 * counts during the last interrupt. The advantage of per interrupt
1512 * computation is faster updates and more accurate ITR for the current
1513 * traffic pattern. Constants in this function were computed
1514 * based on theoretical maximum wire speed and thresholds were set based
1515 * on testing data as well as attempting to minimize response time
1516 * while increasing bulk throughput.
1517 * this functionality is controlled by the InterruptThrottleRate module
1518 * parameter (see ixgbe_param.c)
1520 static u8 ixgbe_update_itr(struct ixgbe_adapter *adapter,
1521 u32 eitr, u8 itr_setting,
1522 int packets, int bytes)
1524 unsigned int retval = itr_setting;
1529 goto update_itr_done;
1532 /* simple throttlerate management
1533 * 0-20MB/s lowest (100000 ints/s)
1534 * 20-100MB/s low (20000 ints/s)
1535 * 100-1249MB/s bulk (8000 ints/s)
1537 /* what was last interrupt timeslice? */
1538 timepassed_us = 1000000/eitr;
1539 bytes_perint = bytes / timepassed_us; /* bytes/usec */
1541 switch (itr_setting) {
1542 case lowest_latency:
1543 if (bytes_perint > adapter->eitr_low)
1544 retval = low_latency;
1547 if (bytes_perint > adapter->eitr_high)
1548 retval = bulk_latency;
1549 else if (bytes_perint <= adapter->eitr_low)
1550 retval = lowest_latency;
1553 if (bytes_perint <= adapter->eitr_high)
1554 retval = low_latency;
1563 * ixgbe_write_eitr - write EITR register in hardware specific way
1564 * @q_vector: structure containing interrupt and ring information
1566 * This function is made to be called by ethtool and by the driver
1567 * when it needs to update EITR registers at runtime. Hardware
1568 * specific quirks/differences are taken care of here.
1570 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1572 struct ixgbe_adapter *adapter = q_vector->adapter;
1573 struct ixgbe_hw *hw = &adapter->hw;
1574 int v_idx = q_vector->v_idx;
1575 u32 itr_reg = EITR_INTS_PER_SEC_TO_REG(q_vector->eitr);
1577 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1578 /* must write high and low 16 bits to reset counter */
1579 itr_reg |= (itr_reg << 16);
1580 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1582 * 82599 can support a value of zero, so allow it for
1583 * max interrupt rate, but there is an errata where it can
1584 * not be zero with RSC
1587 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
1591 * set the WDIS bit to not clear the timer bits and cause an
1592 * immediate assertion of the interrupt
1594 itr_reg |= IXGBE_EITR_CNT_WDIS;
1596 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1599 static void ixgbe_set_itr_msix(struct ixgbe_q_vector *q_vector)
1601 struct ixgbe_adapter *adapter = q_vector->adapter;
1603 u8 current_itr, ret_itr;
1605 struct ixgbe_ring *rx_ring, *tx_ring;
1607 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1608 for (i = 0; i < q_vector->txr_count; i++) {
1609 tx_ring = adapter->tx_ring[r_idx];
1610 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1612 tx_ring->total_packets,
1613 tx_ring->total_bytes);
1614 /* if the result for this queue would decrease interrupt
1615 * rate for this vector then use that result */
1616 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
1617 q_vector->tx_itr - 1 : ret_itr);
1618 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1622 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1623 for (i = 0; i < q_vector->rxr_count; i++) {
1624 rx_ring = adapter->rx_ring[r_idx];
1625 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1627 rx_ring->total_packets,
1628 rx_ring->total_bytes);
1629 /* if the result for this queue would decrease interrupt
1630 * rate for this vector then use that result */
1631 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
1632 q_vector->rx_itr - 1 : ret_itr);
1633 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1637 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1639 switch (current_itr) {
1640 /* counts and packets in update_itr are dependent on these numbers */
1641 case lowest_latency:
1645 new_itr = 20000; /* aka hwitr = ~200 */
1653 if (new_itr != q_vector->eitr) {
1654 /* do an exponential smoothing */
1655 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1657 /* save the algorithm value here, not the smoothed one */
1658 q_vector->eitr = new_itr;
1660 ixgbe_write_eitr(q_vector);
1665 * ixgbe_check_overtemp_task - worker thread to check over tempurature
1666 * @work: pointer to work_struct containing our data
1668 static void ixgbe_check_overtemp_task(struct work_struct *work)
1670 struct ixgbe_adapter *adapter = container_of(work,
1671 struct ixgbe_adapter,
1672 check_overtemp_task);
1673 struct ixgbe_hw *hw = &adapter->hw;
1674 u32 eicr = adapter->interrupt_event;
1676 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
1679 switch (hw->device_id) {
1680 case IXGBE_DEV_ID_82599_T3_LOM: {
1682 bool link_up = false;
1684 if (hw->mac.ops.check_link)
1685 hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
1687 if (((eicr & IXGBE_EICR_GPI_SDP0) && (!link_up)) ||
1688 (eicr & IXGBE_EICR_LSC))
1689 /* Check if this is due to overtemp */
1690 if (hw->phy.ops.check_overtemp(hw) == IXGBE_ERR_OVERTEMP)
1695 if (!(eicr & IXGBE_EICR_GPI_SDP0))
1700 "Network adapter has been stopped because it has over heated. "
1701 "Restart the computer. If the problem persists, "
1702 "power off the system and replace the adapter\n");
1703 /* write to clear the interrupt */
1704 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP0);
1707 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
1709 struct ixgbe_hw *hw = &adapter->hw;
1711 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
1712 (eicr & IXGBE_EICR_GPI_SDP1)) {
1713 e_crit(probe, "Fan has stopped, replace the adapter\n");
1714 /* write to clear the interrupt */
1715 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1719 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
1721 struct ixgbe_hw *hw = &adapter->hw;
1723 if (eicr & IXGBE_EICR_GPI_SDP2) {
1724 /* Clear the interrupt */
1725 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
1726 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1727 schedule_work(&adapter->sfp_config_module_task);
1730 if (eicr & IXGBE_EICR_GPI_SDP1) {
1731 /* Clear the interrupt */
1732 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1733 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1734 schedule_work(&adapter->multispeed_fiber_task);
1738 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1740 struct ixgbe_hw *hw = &adapter->hw;
1743 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1744 adapter->link_check_timeout = jiffies;
1745 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1746 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1747 IXGBE_WRITE_FLUSH(hw);
1748 schedule_work(&adapter->watchdog_task);
1752 static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
1754 struct net_device *netdev = data;
1755 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1756 struct ixgbe_hw *hw = &adapter->hw;
1760 * Workaround for Silicon errata. Use clear-by-write instead
1761 * of clear-by-read. Reading with EICS will return the
1762 * interrupt causes without clearing, which later be done
1763 * with the write to EICR.
1765 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1766 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1768 if (eicr & IXGBE_EICR_LSC)
1769 ixgbe_check_lsc(adapter);
1771 if (eicr & IXGBE_EICR_MAILBOX)
1772 ixgbe_msg_task(adapter);
1774 if (hw->mac.type == ixgbe_mac_82598EB)
1775 ixgbe_check_fan_failure(adapter, eicr);
1777 if (hw->mac.type == ixgbe_mac_82599EB) {
1778 ixgbe_check_sfp_event(adapter, eicr);
1779 adapter->interrupt_event = eicr;
1780 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
1781 ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)))
1782 schedule_work(&adapter->check_overtemp_task);
1784 /* Handle Flow Director Full threshold interrupt */
1785 if (eicr & IXGBE_EICR_FLOW_DIR) {
1787 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_FLOW_DIR);
1788 /* Disable transmits before FDIR Re-initialization */
1789 netif_tx_stop_all_queues(netdev);
1790 for (i = 0; i < adapter->num_tx_queues; i++) {
1791 struct ixgbe_ring *tx_ring =
1792 adapter->tx_ring[i];
1793 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
1795 schedule_work(&adapter->fdir_reinit_task);
1799 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1800 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_OTHER);
1805 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
1810 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1811 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1812 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1814 mask = (qmask & 0xFFFFFFFF);
1815 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(0), mask);
1816 mask = (qmask >> 32);
1817 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(1), mask);
1819 /* skip the flush */
1822 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
1827 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1828 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1829 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, mask);
1831 mask = (qmask & 0xFFFFFFFF);
1832 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), mask);
1833 mask = (qmask >> 32);
1834 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), mask);
1836 /* skip the flush */
1839 static irqreturn_t ixgbe_msix_clean_tx(int irq, void *data)
1841 struct ixgbe_q_vector *q_vector = data;
1842 struct ixgbe_adapter *adapter = q_vector->adapter;
1843 struct ixgbe_ring *tx_ring;
1846 if (!q_vector->txr_count)
1849 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1850 for (i = 0; i < q_vector->txr_count; i++) {
1851 tx_ring = adapter->tx_ring[r_idx];
1852 tx_ring->total_bytes = 0;
1853 tx_ring->total_packets = 0;
1854 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1858 /* EIAM disabled interrupts (on this vector) for us */
1859 napi_schedule(&q_vector->napi);
1865 * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
1867 * @data: pointer to our q_vector struct for this interrupt vector
1869 static irqreturn_t ixgbe_msix_clean_rx(int irq, void *data)
1871 struct ixgbe_q_vector *q_vector = data;
1872 struct ixgbe_adapter *adapter = q_vector->adapter;
1873 struct ixgbe_ring *rx_ring;
1877 #ifdef CONFIG_IXGBE_DCA
1878 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1879 ixgbe_update_dca(q_vector);
1882 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1883 for (i = 0; i < q_vector->rxr_count; i++) {
1884 rx_ring = adapter->rx_ring[r_idx];
1885 rx_ring->total_bytes = 0;
1886 rx_ring->total_packets = 0;
1887 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1891 if (!q_vector->rxr_count)
1894 /* EIAM disabled interrupts (on this vector) for us */
1895 napi_schedule(&q_vector->napi);
1900 static irqreturn_t ixgbe_msix_clean_many(int irq, void *data)
1902 struct ixgbe_q_vector *q_vector = data;
1903 struct ixgbe_adapter *adapter = q_vector->adapter;
1904 struct ixgbe_ring *ring;
1908 if (!q_vector->txr_count && !q_vector->rxr_count)
1911 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1912 for (i = 0; i < q_vector->txr_count; i++) {
1913 ring = adapter->tx_ring[r_idx];
1914 ring->total_bytes = 0;
1915 ring->total_packets = 0;
1916 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1920 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1921 for (i = 0; i < q_vector->rxr_count; i++) {
1922 ring = adapter->rx_ring[r_idx];
1923 ring->total_bytes = 0;
1924 ring->total_packets = 0;
1925 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1929 /* EIAM disabled interrupts (on this vector) for us */
1930 napi_schedule(&q_vector->napi);
1936 * ixgbe_clean_rxonly - msix (aka one shot) rx clean routine
1937 * @napi: napi struct with our devices info in it
1938 * @budget: amount of work driver is allowed to do this pass, in packets
1940 * This function is optimized for cleaning one queue only on a single
1943 static int ixgbe_clean_rxonly(struct napi_struct *napi, int budget)
1945 struct ixgbe_q_vector *q_vector =
1946 container_of(napi, struct ixgbe_q_vector, napi);
1947 struct ixgbe_adapter *adapter = q_vector->adapter;
1948 struct ixgbe_ring *rx_ring = NULL;
1952 #ifdef CONFIG_IXGBE_DCA
1953 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1954 ixgbe_update_dca(q_vector);
1957 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1958 rx_ring = adapter->rx_ring[r_idx];
1960 ixgbe_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
1962 /* If all Rx work done, exit the polling mode */
1963 if (work_done < budget) {
1964 napi_complete(napi);
1965 if (adapter->rx_itr_setting & 1)
1966 ixgbe_set_itr_msix(q_vector);
1967 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1968 ixgbe_irq_enable_queues(adapter,
1969 ((u64)1 << q_vector->v_idx));
1976 * ixgbe_clean_rxtx_many - msix (aka one shot) rx clean routine
1977 * @napi: napi struct with our devices info in it
1978 * @budget: amount of work driver is allowed to do this pass, in packets
1980 * This function will clean more than one rx queue associated with a
1983 static int ixgbe_clean_rxtx_many(struct napi_struct *napi, int budget)
1985 struct ixgbe_q_vector *q_vector =
1986 container_of(napi, struct ixgbe_q_vector, napi);
1987 struct ixgbe_adapter *adapter = q_vector->adapter;
1988 struct ixgbe_ring *ring = NULL;
1989 int work_done = 0, i;
1991 bool tx_clean_complete = true;
1993 #ifdef CONFIG_IXGBE_DCA
1994 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1995 ixgbe_update_dca(q_vector);
1998 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1999 for (i = 0; i < q_vector->txr_count; i++) {
2000 ring = adapter->tx_ring[r_idx];
2001 tx_clean_complete &= ixgbe_clean_tx_irq(q_vector, ring);
2002 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
2006 /* attempt to distribute budget to each queue fairly, but don't allow
2007 * the budget to go below 1 because we'll exit polling */
2008 budget /= (q_vector->rxr_count ?: 1);
2009 budget = max(budget, 1);
2010 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2011 for (i = 0; i < q_vector->rxr_count; i++) {
2012 ring = adapter->rx_ring[r_idx];
2013 ixgbe_clean_rx_irq(q_vector, ring, &work_done, budget);
2014 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
2018 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2019 ring = adapter->rx_ring[r_idx];
2020 /* If all Rx work done, exit the polling mode */
2021 if (work_done < budget) {
2022 napi_complete(napi);
2023 if (adapter->rx_itr_setting & 1)
2024 ixgbe_set_itr_msix(q_vector);
2025 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2026 ixgbe_irq_enable_queues(adapter,
2027 ((u64)1 << q_vector->v_idx));
2035 * ixgbe_clean_txonly - msix (aka one shot) tx clean routine
2036 * @napi: napi struct with our devices info in it
2037 * @budget: amount of work driver is allowed to do this pass, in packets
2039 * This function is optimized for cleaning one queue only on a single
2042 static int ixgbe_clean_txonly(struct napi_struct *napi, int budget)
2044 struct ixgbe_q_vector *q_vector =
2045 container_of(napi, struct ixgbe_q_vector, napi);
2046 struct ixgbe_adapter *adapter = q_vector->adapter;
2047 struct ixgbe_ring *tx_ring = NULL;
2051 #ifdef CONFIG_IXGBE_DCA
2052 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2053 ixgbe_update_dca(q_vector);
2056 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
2057 tx_ring = adapter->tx_ring[r_idx];
2059 if (!ixgbe_clean_tx_irq(q_vector, tx_ring))
2062 /* If all Tx work done, exit the polling mode */
2063 if (work_done < budget) {
2064 napi_complete(napi);
2065 if (adapter->tx_itr_setting & 1)
2066 ixgbe_set_itr_msix(q_vector);
2067 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2068 ixgbe_irq_enable_queues(adapter,
2069 ((u64)1 << q_vector->v_idx));
2075 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
2078 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2080 set_bit(r_idx, q_vector->rxr_idx);
2081 q_vector->rxr_count++;
2084 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
2087 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2089 set_bit(t_idx, q_vector->txr_idx);
2090 q_vector->txr_count++;
2094 * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
2095 * @adapter: board private structure to initialize
2096 * @vectors: allotted vector count for descriptor rings
2098 * This function maps descriptor rings to the queue-specific vectors
2099 * we were allotted through the MSI-X enabling code. Ideally, we'd have
2100 * one vector per ring/queue, but on a constrained vector budget, we
2101 * group the rings as "efficiently" as possible. You would add new
2102 * mapping configurations in here.
2104 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter,
2108 int rxr_idx = 0, txr_idx = 0;
2109 int rxr_remaining = adapter->num_rx_queues;
2110 int txr_remaining = adapter->num_tx_queues;
2115 /* No mapping required if MSI-X is disabled. */
2116 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2120 * The ideal configuration...
2121 * We have enough vectors to map one per queue.
2123 if (vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
2124 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
2125 map_vector_to_rxq(adapter, v_start, rxr_idx);
2127 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
2128 map_vector_to_txq(adapter, v_start, txr_idx);
2134 * If we don't have enough vectors for a 1-to-1
2135 * mapping, we'll have to group them so there are
2136 * multiple queues per vector.
2138 /* Re-adjusting *qpv takes care of the remainder. */
2139 for (i = v_start; i < vectors; i++) {
2140 rqpv = DIV_ROUND_UP(rxr_remaining, vectors - i);
2141 for (j = 0; j < rqpv; j++) {
2142 map_vector_to_rxq(adapter, i, rxr_idx);
2147 for (i = v_start; i < vectors; i++) {
2148 tqpv = DIV_ROUND_UP(txr_remaining, vectors - i);
2149 for (j = 0; j < tqpv; j++) {
2150 map_vector_to_txq(adapter, i, txr_idx);
2161 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2162 * @adapter: board private structure
2164 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2165 * interrupts from the kernel.
2167 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2169 struct net_device *netdev = adapter->netdev;
2170 irqreturn_t (*handler)(int, void *);
2171 int i, vector, q_vectors, err;
2174 /* Decrement for Other and TCP Timer vectors */
2175 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2177 /* Map the Tx/Rx rings to the vectors we were allotted. */
2178 err = ixgbe_map_rings_to_vectors(adapter, q_vectors);
2182 #define SET_HANDLER(_v) ((!(_v)->rxr_count) ? &ixgbe_msix_clean_tx : \
2183 (!(_v)->txr_count) ? &ixgbe_msix_clean_rx : \
2184 &ixgbe_msix_clean_many)
2185 for (vector = 0; vector < q_vectors; vector++) {
2186 handler = SET_HANDLER(adapter->q_vector[vector]);
2188 if (handler == &ixgbe_msix_clean_rx) {
2189 sprintf(adapter->name[vector], "%s-%s-%d",
2190 netdev->name, "rx", ri++);
2191 } else if (handler == &ixgbe_msix_clean_tx) {
2192 sprintf(adapter->name[vector], "%s-%s-%d",
2193 netdev->name, "tx", ti++);
2195 sprintf(adapter->name[vector], "%s-%s-%d",
2196 netdev->name, "TxRx", ri++);
2200 err = request_irq(adapter->msix_entries[vector].vector,
2201 handler, 0, adapter->name[vector],
2202 adapter->q_vector[vector]);
2204 e_err(probe, "request_irq failed for MSIX interrupt "
2205 "Error: %d\n", err);
2206 goto free_queue_irqs;
2210 sprintf(adapter->name[vector], "%s:lsc", netdev->name);
2211 err = request_irq(adapter->msix_entries[vector].vector,
2212 ixgbe_msix_lsc, 0, adapter->name[vector], netdev);
2214 e_err(probe, "request_irq for msix_lsc failed: %d\n", err);
2215 goto free_queue_irqs;
2221 for (i = vector - 1; i >= 0; i--)
2222 free_irq(adapter->msix_entries[--vector].vector,
2223 adapter->q_vector[i]);
2224 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2225 pci_disable_msix(adapter->pdev);
2226 kfree(adapter->msix_entries);
2227 adapter->msix_entries = NULL;
2232 static void ixgbe_set_itr(struct ixgbe_adapter *adapter)
2234 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2236 u32 new_itr = q_vector->eitr;
2237 struct ixgbe_ring *rx_ring = adapter->rx_ring[0];
2238 struct ixgbe_ring *tx_ring = adapter->tx_ring[0];
2240 q_vector->tx_itr = ixgbe_update_itr(adapter, new_itr,
2242 tx_ring->total_packets,
2243 tx_ring->total_bytes);
2244 q_vector->rx_itr = ixgbe_update_itr(adapter, new_itr,
2246 rx_ring->total_packets,
2247 rx_ring->total_bytes);
2249 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
2251 switch (current_itr) {
2252 /* counts and packets in update_itr are dependent on these numbers */
2253 case lowest_latency:
2257 new_itr = 20000; /* aka hwitr = ~200 */
2266 if (new_itr != q_vector->eitr) {
2267 /* do an exponential smoothing */
2268 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
2270 /* save the algorithm value here, not the smoothed one */
2271 q_vector->eitr = new_itr;
2273 ixgbe_write_eitr(q_vector);
2278 * ixgbe_irq_enable - Enable default interrupt generation settings
2279 * @adapter: board private structure
2281 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2286 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2287 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2288 mask |= IXGBE_EIMS_GPI_SDP0;
2289 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2290 mask |= IXGBE_EIMS_GPI_SDP1;
2291 if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
2292 mask |= IXGBE_EIMS_ECC;
2293 mask |= IXGBE_EIMS_GPI_SDP1;
2294 mask |= IXGBE_EIMS_GPI_SDP2;
2295 if (adapter->num_vfs)
2296 mask |= IXGBE_EIMS_MAILBOX;
2298 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
2299 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
2300 mask |= IXGBE_EIMS_FLOW_DIR;
2302 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2304 ixgbe_irq_enable_queues(adapter, ~0);
2306 IXGBE_WRITE_FLUSH(&adapter->hw);
2308 if (adapter->num_vfs > 32) {
2309 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
2310 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2315 * ixgbe_intr - legacy mode Interrupt Handler
2316 * @irq: interrupt number
2317 * @data: pointer to a network interface device structure
2319 static irqreturn_t ixgbe_intr(int irq, void *data)
2321 struct net_device *netdev = data;
2322 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2323 struct ixgbe_hw *hw = &adapter->hw;
2324 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2328 * Workaround for silicon errata on 82598. Mask the interrupts
2329 * before the read of EICR.
2331 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2333 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2334 * therefore no explict interrupt disable is necessary */
2335 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2338 * shared interrupt alert!
2339 * make sure interrupts are enabled because the read will
2340 * have disabled interrupts due to EIAM
2341 * finish the workaround of silicon errata on 82598. Unmask
2342 * the interrupt that we masked before the EICR read.
2344 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2345 ixgbe_irq_enable(adapter, true, true);
2346 return IRQ_NONE; /* Not our interrupt */
2349 if (eicr & IXGBE_EICR_LSC)
2350 ixgbe_check_lsc(adapter);
2352 if (hw->mac.type == ixgbe_mac_82599EB)
2353 ixgbe_check_sfp_event(adapter, eicr);
2355 ixgbe_check_fan_failure(adapter, eicr);
2356 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2357 ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)))
2358 schedule_work(&adapter->check_overtemp_task);
2360 if (napi_schedule_prep(&(q_vector->napi))) {
2361 adapter->tx_ring[0]->total_packets = 0;
2362 adapter->tx_ring[0]->total_bytes = 0;
2363 adapter->rx_ring[0]->total_packets = 0;
2364 adapter->rx_ring[0]->total_bytes = 0;
2365 /* would disable interrupts here but EIAM disabled it */
2366 __napi_schedule(&(q_vector->napi));
2370 * re-enable link(maybe) and non-queue interrupts, no flush.
2371 * ixgbe_poll will re-enable the queue interrupts
2374 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2375 ixgbe_irq_enable(adapter, false, false);
2380 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
2382 int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2384 for (i = 0; i < q_vectors; i++) {
2385 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
2386 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
2387 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
2388 q_vector->rxr_count = 0;
2389 q_vector->txr_count = 0;
2394 * ixgbe_request_irq - initialize interrupts
2395 * @adapter: board private structure
2397 * Attempts to configure interrupts using the best available
2398 * capabilities of the hardware and kernel.
2400 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2402 struct net_device *netdev = adapter->netdev;
2405 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2406 err = ixgbe_request_msix_irqs(adapter);
2407 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
2408 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2409 netdev->name, netdev);
2411 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2412 netdev->name, netdev);
2416 e_err(probe, "request_irq failed, Error %d\n", err);
2421 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2423 struct net_device *netdev = adapter->netdev;
2425 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2428 q_vectors = adapter->num_msix_vectors;
2431 free_irq(adapter->msix_entries[i].vector, netdev);
2434 for (; i >= 0; i--) {
2435 free_irq(adapter->msix_entries[i].vector,
2436 adapter->q_vector[i]);
2439 ixgbe_reset_q_vectors(adapter);
2441 free_irq(adapter->pdev->irq, netdev);
2446 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2447 * @adapter: board private structure
2449 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2451 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2452 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2454 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2455 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2456 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2457 if (adapter->num_vfs > 32)
2458 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
2460 IXGBE_WRITE_FLUSH(&adapter->hw);
2461 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2463 for (i = 0; i < adapter->num_msix_vectors; i++)
2464 synchronize_irq(adapter->msix_entries[i].vector);
2466 synchronize_irq(adapter->pdev->irq);
2471 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2474 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2476 struct ixgbe_hw *hw = &adapter->hw;
2478 IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
2479 EITR_INTS_PER_SEC_TO_REG(adapter->rx_eitr_param));
2481 ixgbe_set_ivar(adapter, 0, 0, 0);
2482 ixgbe_set_ivar(adapter, 1, 0, 0);
2484 map_vector_to_rxq(adapter, 0, 0);
2485 map_vector_to_txq(adapter, 0, 0);
2487 e_info(hw, "Legacy interrupt IVAR setup done\n");
2491 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2492 * @adapter: board private structure
2493 * @ring: structure containing ring specific data
2495 * Configure the Tx descriptor ring after a reset.
2497 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2498 struct ixgbe_ring *ring)
2500 struct ixgbe_hw *hw = &adapter->hw;
2501 u64 tdba = ring->dma;
2504 u16 reg_idx = ring->reg_idx;
2506 /* disable queue to avoid issues while updating state */
2507 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2508 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx),
2509 txdctl & ~IXGBE_TXDCTL_ENABLE);
2510 IXGBE_WRITE_FLUSH(hw);
2512 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2513 (tdba & DMA_BIT_MASK(32)));
2514 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2515 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2516 ring->count * sizeof(union ixgbe_adv_tx_desc));
2517 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2518 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2519 ring->tail = hw->hw_addr + IXGBE_TDT(reg_idx);
2521 /* configure fetching thresholds */
2522 if (adapter->rx_itr_setting == 0) {
2523 /* cannot set wthresh when itr==0 */
2524 txdctl &= ~0x007F0000;
2526 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
2527 txdctl |= (8 << 16);
2529 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2530 /* PThresh workaround for Tx hang with DFP enabled. */
2534 /* reinitialize flowdirector state */
2535 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2536 adapter->atr_sample_rate) {
2537 ring->atr_sample_rate = adapter->atr_sample_rate;
2538 ring->atr_count = 0;
2539 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
2541 ring->atr_sample_rate = 0;
2545 txdctl |= IXGBE_TXDCTL_ENABLE;
2546 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
2548 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2549 if (hw->mac.type == ixgbe_mac_82598EB &&
2550 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2553 /* poll to verify queue is enabled */
2556 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2557 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
2559 e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
2562 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
2564 struct ixgbe_hw *hw = &adapter->hw;
2568 if (hw->mac.type == ixgbe_mac_82598EB)
2571 /* disable the arbiter while setting MTQC */
2572 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2573 rttdcs |= IXGBE_RTTDCS_ARBDIS;
2574 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2576 /* set transmit pool layout */
2577 mask = (IXGBE_FLAG_SRIOV_ENABLED | IXGBE_FLAG_DCB_ENABLED);
2578 switch (adapter->flags & mask) {
2580 case (IXGBE_FLAG_SRIOV_ENABLED):
2581 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2582 (IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF));
2585 case (IXGBE_FLAG_DCB_ENABLED):
2586 /* We enable 8 traffic classes, DCB only */
2587 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2588 (IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ));
2592 IXGBE_WRITE_REG(hw, IXGBE_MTQC, IXGBE_MTQC_64Q_1PB);
2596 /* re-enable the arbiter */
2597 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2598 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2602 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2603 * @adapter: board private structure
2605 * Configure the Tx unit of the MAC after a reset.
2607 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2609 struct ixgbe_hw *hw = &adapter->hw;
2613 ixgbe_setup_mtqc(adapter);
2615 if (hw->mac.type != ixgbe_mac_82598EB) {
2616 /* DMATXCTL.EN must be before Tx queues are enabled */
2617 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2618 dmatxctl |= IXGBE_DMATXCTL_TE;
2619 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2622 /* Setup the HW Tx Head and Tail descriptor pointers */
2623 for (i = 0; i < adapter->num_tx_queues; i++)
2624 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
2627 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2629 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2630 struct ixgbe_ring *rx_ring)
2634 struct ixgbe_ring_feature *feature = adapter->ring_feature;
2636 index = rx_ring->reg_idx;
2637 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2639 mask = (unsigned long) feature[RING_F_RSS].mask;
2640 index = index & mask;
2642 srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(index));
2644 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
2645 srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
2646 if (adapter->num_vfs)
2647 srrctl |= IXGBE_SRRCTL_DROP_EN;
2649 srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
2650 IXGBE_SRRCTL_BSIZEHDR_MASK;
2652 if (ring_is_ps_enabled(rx_ring)) {
2653 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
2654 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2656 srrctl |= (PAGE_SIZE / 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2658 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2660 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
2661 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2662 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2665 IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(index), srrctl);
2668 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2670 struct ixgbe_hw *hw = &adapter->hw;
2671 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2672 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2673 0x6A3E67EA, 0x14364D17, 0x3BED200D};
2674 u32 mrqc = 0, reta = 0;
2679 /* Fill out hash function seeds */
2680 for (i = 0; i < 10; i++)
2681 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2683 /* Fill out redirection table */
2684 for (i = 0, j = 0; i < 128; i++, j++) {
2685 if (j == adapter->ring_feature[RING_F_RSS].indices)
2687 /* reta = 4-byte sliding window of
2688 * 0x00..(indices-1)(indices-1)00..etc. */
2689 reta = (reta << 8) | (j * 0x11);
2691 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2694 /* Disable indicating checksum in descriptor, enables RSS hash */
2695 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2696 rxcsum |= IXGBE_RXCSUM_PCSD;
2697 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2699 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
2700 mask = adapter->flags & IXGBE_FLAG_RSS_ENABLED;
2702 mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2703 #ifdef CONFIG_IXGBE_DCB
2704 | IXGBE_FLAG_DCB_ENABLED
2706 | IXGBE_FLAG_SRIOV_ENABLED
2710 case (IXGBE_FLAG_RSS_ENABLED):
2711 mrqc = IXGBE_MRQC_RSSEN;
2713 case (IXGBE_FLAG_SRIOV_ENABLED):
2714 mrqc = IXGBE_MRQC_VMDQEN;
2716 #ifdef CONFIG_IXGBE_DCB
2717 case (IXGBE_FLAG_DCB_ENABLED):
2718 mrqc = IXGBE_MRQC_RT8TCEN;
2720 #endif /* CONFIG_IXGBE_DCB */
2725 /* Perform hash on these packet types */
2726 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2727 | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2728 | IXGBE_MRQC_RSS_FIELD_IPV6
2729 | IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
2731 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2735 * ixgbe_configure_rscctl - enable RSC for the indicated ring
2736 * @adapter: address of board private structure
2737 * @index: index of ring to set
2739 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
2740 struct ixgbe_ring *ring)
2742 struct ixgbe_hw *hw = &adapter->hw;
2745 u16 reg_idx = ring->reg_idx;
2747 if (!ring_is_rsc_enabled(ring))
2750 rx_buf_len = ring->rx_buf_len;
2751 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
2752 rscctrl |= IXGBE_RSCCTL_RSCEN;
2754 * we must limit the number of descriptors so that the
2755 * total size of max desc * buf_len is not greater
2758 if (ring_is_ps_enabled(ring)) {
2759 #if (MAX_SKB_FRAGS > 16)
2760 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2761 #elif (MAX_SKB_FRAGS > 8)
2762 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2763 #elif (MAX_SKB_FRAGS > 4)
2764 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2766 rscctrl |= IXGBE_RSCCTL_MAXDESC_1;
2769 if (rx_buf_len < IXGBE_RXBUFFER_4096)
2770 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2771 else if (rx_buf_len < IXGBE_RXBUFFER_8192)
2772 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2774 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2776 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
2780 * ixgbe_set_uta - Set unicast filter table address
2781 * @adapter: board private structure
2783 * The unicast table address is a register array of 32-bit registers.
2784 * The table is meant to be used in a way similar to how the MTA is used
2785 * however due to certain limitations in the hardware it is necessary to
2786 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
2787 * enable bit to allow vlan tag stripping when promiscuous mode is enabled
2789 static void ixgbe_set_uta(struct ixgbe_adapter *adapter)
2791 struct ixgbe_hw *hw = &adapter->hw;
2794 /* The UTA table only exists on 82599 hardware and newer */
2795 if (hw->mac.type < ixgbe_mac_82599EB)
2798 /* we only need to do this if VMDq is enabled */
2799 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
2802 for (i = 0; i < 128; i++)
2803 IXGBE_WRITE_REG(hw, IXGBE_UTA(i), ~0);
2806 #define IXGBE_MAX_RX_DESC_POLL 10
2807 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
2808 struct ixgbe_ring *ring)
2810 struct ixgbe_hw *hw = &adapter->hw;
2811 int reg_idx = ring->reg_idx;
2812 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
2815 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2816 if (hw->mac.type == ixgbe_mac_82598EB &&
2817 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2822 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2823 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
2826 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
2827 "the polling period\n", reg_idx);
2831 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
2832 struct ixgbe_ring *ring)
2834 struct ixgbe_hw *hw = &adapter->hw;
2835 u64 rdba = ring->dma;
2837 u16 reg_idx = ring->reg_idx;
2839 /* disable queue to avoid issues while updating state */
2840 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2841 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx),
2842 rxdctl & ~IXGBE_RXDCTL_ENABLE);
2843 IXGBE_WRITE_FLUSH(hw);
2845 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
2846 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
2847 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
2848 ring->count * sizeof(union ixgbe_adv_rx_desc));
2849 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
2850 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
2851 ring->tail = hw->hw_addr + IXGBE_RDT(reg_idx);
2853 ixgbe_configure_srrctl(adapter, ring);
2854 ixgbe_configure_rscctl(adapter, ring);
2856 if (hw->mac.type == ixgbe_mac_82598EB) {
2858 * enable cache line friendly hardware writes:
2859 * PTHRESH=32 descriptors (half the internal cache),
2860 * this also removes ugly rx_no_buffer_count increment
2861 * HTHRESH=4 descriptors (to minimize latency on fetch)
2862 * WTHRESH=8 burst writeback up to two cache lines
2864 rxdctl &= ~0x3FFFFF;
2868 /* enable receive descriptor ring */
2869 rxdctl |= IXGBE_RXDCTL_ENABLE;
2870 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
2872 ixgbe_rx_desc_queue_enable(adapter, ring);
2873 ixgbe_alloc_rx_buffers(ring, IXGBE_DESC_UNUSED(ring));
2876 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
2878 struct ixgbe_hw *hw = &adapter->hw;
2881 /* PSRTYPE must be initialized in non 82598 adapters */
2882 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
2883 IXGBE_PSRTYPE_UDPHDR |
2884 IXGBE_PSRTYPE_IPV4HDR |
2885 IXGBE_PSRTYPE_L2HDR |
2886 IXGBE_PSRTYPE_IPV6HDR;
2888 if (hw->mac.type == ixgbe_mac_82598EB)
2891 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED)
2892 psrtype |= (adapter->num_rx_queues_per_pool << 29);
2894 for (p = 0; p < adapter->num_rx_pools; p++)
2895 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(adapter->num_vfs + p),
2899 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
2901 struct ixgbe_hw *hw = &adapter->hw;
2904 u32 reg_offset, vf_shift;
2907 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
2910 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
2911 vt_reg_bits = IXGBE_VMD_CTL_VMDQ_EN | IXGBE_VT_CTL_REPLEN;
2912 vt_reg_bits |= (adapter->num_vfs << IXGBE_VT_CTL_POOL_SHIFT);
2913 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl | vt_reg_bits);
2915 vf_shift = adapter->num_vfs % 32;
2916 reg_offset = (adapter->num_vfs > 32) ? 1 : 0;
2918 /* Enable only the PF's pool for Tx/Rx */
2919 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (1 << vf_shift));
2920 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), 0);
2921 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (1 << vf_shift));
2922 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), 0);
2923 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
2925 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
2926 hw->mac.ops.set_vmdq(hw, 0, adapter->num_vfs);
2929 * Set up VF register offsets for selected VT Mode,
2930 * i.e. 32 or 64 VFs for SR-IOV
2932 gcr_ext = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
2933 gcr_ext |= IXGBE_GCR_EXT_MSIX_EN;
2934 gcr_ext |= IXGBE_GCR_EXT_VT_MODE_64;
2935 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
2937 /* enable Tx loopback for VF/PF communication */
2938 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
2941 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
2943 struct ixgbe_hw *hw = &adapter->hw;
2944 struct net_device *netdev = adapter->netdev;
2945 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2947 struct ixgbe_ring *rx_ring;
2951 /* Decide whether to use packet split mode or not */
2952 /* Do not use packet split if we're in SR-IOV Mode */
2953 if (!adapter->num_vfs)
2954 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
2956 /* Set the RX buffer length according to the mode */
2957 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
2958 rx_buf_len = IXGBE_RX_HDR_SIZE;
2960 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
2961 (netdev->mtu <= ETH_DATA_LEN))
2962 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
2964 rx_buf_len = ALIGN(max_frame + VLAN_HLEN, 1024);
2968 /* adjust max frame to be able to do baby jumbo for FCoE */
2969 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
2970 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
2971 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
2973 #endif /* IXGBE_FCOE */
2974 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
2975 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
2976 mhadd &= ~IXGBE_MHADD_MFS_MASK;
2977 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
2979 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
2982 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2983 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
2984 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2985 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
2988 * Setup the HW Rx Head and Tail Descriptor Pointers and
2989 * the Base and Length of the Rx Descriptor Ring
2991 for (i = 0; i < adapter->num_rx_queues; i++) {
2992 rx_ring = adapter->rx_ring[i];
2993 rx_ring->rx_buf_len = rx_buf_len;
2995 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)
2996 set_ring_ps_enabled(rx_ring);
2998 clear_ring_ps_enabled(rx_ring);
3000 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
3001 set_ring_rsc_enabled(rx_ring);
3003 clear_ring_rsc_enabled(rx_ring);
3006 if (netdev->features & NETIF_F_FCOE_MTU) {
3007 struct ixgbe_ring_feature *f;
3008 f = &adapter->ring_feature[RING_F_FCOE];
3009 if ((i >= f->mask) && (i < f->mask + f->indices)) {
3010 clear_ring_ps_enabled(rx_ring);
3011 if (rx_buf_len < IXGBE_FCOE_JUMBO_FRAME_SIZE)
3012 rx_ring->rx_buf_len =
3013 IXGBE_FCOE_JUMBO_FRAME_SIZE;
3014 } else if (!ring_is_rsc_enabled(rx_ring) &&
3015 !ring_is_ps_enabled(rx_ring)) {
3016 rx_ring->rx_buf_len =
3017 IXGBE_FCOE_JUMBO_FRAME_SIZE;
3020 #endif /* IXGBE_FCOE */
3024 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
3026 struct ixgbe_hw *hw = &adapter->hw;
3027 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
3029 switch (hw->mac.type) {
3030 case ixgbe_mac_82598EB:
3032 * For VMDq support of different descriptor types or
3033 * buffer sizes through the use of multiple SRRCTL
3034 * registers, RDRXCTL.MVMEN must be set to 1
3036 * also, the manual doesn't mention it clearly but DCA hints
3037 * will only use queue 0's tags unless this bit is set. Side
3038 * effects of setting this bit are only that SRRCTL must be
3039 * fully programmed [0..15]
3041 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
3043 case ixgbe_mac_82599EB:
3044 /* Disable RSC for ACK packets */
3045 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
3046 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
3047 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3048 /* hardware requires some bits to be set by default */
3049 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3050 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3053 /* We should do nothing since we don't know this hardware */
3057 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3061 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3062 * @adapter: board private structure
3064 * Configure the Rx unit of the MAC after a reset.
3066 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3068 struct ixgbe_hw *hw = &adapter->hw;
3072 /* disable receives while setting up the descriptors */
3073 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3074 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3076 ixgbe_setup_psrtype(adapter);
3077 ixgbe_setup_rdrxctl(adapter);
3079 /* Program registers for the distribution of queues */
3080 ixgbe_setup_mrqc(adapter);
3082 ixgbe_set_uta(adapter);
3084 /* set_rx_buffer_len must be called before ring initialization */
3085 ixgbe_set_rx_buffer_len(adapter);
3088 * Setup the HW Rx Head and Tail Descriptor Pointers and
3089 * the Base and Length of the Rx Descriptor Ring
3091 for (i = 0; i < adapter->num_rx_queues; i++)
3092 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3094 /* disable drop enable for 82598 parts */
3095 if (hw->mac.type == ixgbe_mac_82598EB)
3096 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3098 /* enable all receives */
3099 rxctrl |= IXGBE_RXCTRL_RXEN;
3100 hw->mac.ops.enable_rx_dma(hw, rxctrl);
3103 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3105 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3106 struct ixgbe_hw *hw = &adapter->hw;
3107 int pool_ndx = adapter->num_vfs;
3109 /* add VID to filter table */
3110 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, true);
3111 set_bit(vid, adapter->active_vlans);
3114 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3116 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3117 struct ixgbe_hw *hw = &adapter->hw;
3118 int pool_ndx = adapter->num_vfs;
3120 /* remove VID from filter table */
3121 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, false);
3122 clear_bit(vid, adapter->active_vlans);
3126 * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3127 * @adapter: driver data
3129 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
3131 struct ixgbe_hw *hw = &adapter->hw;
3134 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3135 vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
3136 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3140 * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3141 * @adapter: driver data
3143 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
3145 struct ixgbe_hw *hw = &adapter->hw;
3148 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3149 vlnctrl |= IXGBE_VLNCTRL_VFE;
3150 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3151 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3155 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3156 * @adapter: driver data
3158 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3160 struct ixgbe_hw *hw = &adapter->hw;
3164 switch (hw->mac.type) {
3165 case ixgbe_mac_82598EB:
3166 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3167 vlnctrl &= ~IXGBE_VLNCTRL_VME;
3168 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3170 case ixgbe_mac_82599EB:
3171 for (i = 0; i < adapter->num_rx_queues; i++) {
3172 j = adapter->rx_ring[i]->reg_idx;
3173 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3174 vlnctrl &= ~IXGBE_RXDCTL_VME;
3175 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3184 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3185 * @adapter: driver data
3187 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3189 struct ixgbe_hw *hw = &adapter->hw;
3193 switch (hw->mac.type) {
3194 case ixgbe_mac_82598EB:
3195 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3196 vlnctrl |= IXGBE_VLNCTRL_VME;
3197 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3199 case ixgbe_mac_82599EB:
3200 for (i = 0; i < adapter->num_rx_queues; i++) {
3201 j = adapter->rx_ring[i]->reg_idx;
3202 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3203 vlnctrl |= IXGBE_RXDCTL_VME;
3204 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3212 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3216 ixgbe_vlan_rx_add_vid(adapter->netdev, 0);
3218 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3219 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
3223 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3224 * @netdev: network interface device structure
3226 * Writes unicast address list to the RAR table.
3227 * Returns: -ENOMEM on failure/insufficient address space
3228 * 0 on no addresses written
3229 * X on writing X addresses to the RAR table
3231 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3233 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3234 struct ixgbe_hw *hw = &adapter->hw;
3235 unsigned int vfn = adapter->num_vfs;
3236 unsigned int rar_entries = hw->mac.num_rar_entries - (vfn + 1);
3239 /* return ENOMEM indicating insufficient memory for addresses */
3240 if (netdev_uc_count(netdev) > rar_entries)
3243 if (!netdev_uc_empty(netdev) && rar_entries) {
3244 struct netdev_hw_addr *ha;
3245 /* return error if we do not support writing to RAR table */
3246 if (!hw->mac.ops.set_rar)
3249 netdev_for_each_uc_addr(ha, netdev) {
3252 hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3257 /* write the addresses in reverse order to avoid write combining */
3258 for (; rar_entries > 0 ; rar_entries--)
3259 hw->mac.ops.clear_rar(hw, rar_entries);
3265 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3266 * @netdev: network interface device structure
3268 * The set_rx_method entry point is called whenever the unicast/multicast
3269 * address list or the network interface flags are updated. This routine is
3270 * responsible for configuring the hardware for proper unicast, multicast and
3273 void ixgbe_set_rx_mode(struct net_device *netdev)
3275 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3276 struct ixgbe_hw *hw = &adapter->hw;
3277 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3280 /* Check for Promiscuous and All Multicast modes */
3282 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3284 /* set all bits that we expect to always be set */
3285 fctrl |= IXGBE_FCTRL_BAM;
3286 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3287 fctrl |= IXGBE_FCTRL_PMCF;
3289 /* clear the bits we are changing the status of */
3290 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3292 if (netdev->flags & IFF_PROMISC) {
3293 hw->addr_ctrl.user_set_promisc = true;
3294 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3295 vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3296 /* don't hardware filter vlans in promisc mode */
3297 ixgbe_vlan_filter_disable(adapter);
3299 if (netdev->flags & IFF_ALLMULTI) {
3300 fctrl |= IXGBE_FCTRL_MPE;
3301 vmolr |= IXGBE_VMOLR_MPE;
3304 * Write addresses to the MTA, if the attempt fails
3305 * then we should just turn on promiscous mode so
3306 * that we can at least receive multicast traffic
3308 hw->mac.ops.update_mc_addr_list(hw, netdev);
3309 vmolr |= IXGBE_VMOLR_ROMPE;
3311 ixgbe_vlan_filter_enable(adapter);
3312 hw->addr_ctrl.user_set_promisc = false;
3314 * Write addresses to available RAR registers, if there is not
3315 * sufficient space to store all the addresses then enable
3316 * unicast promiscous mode
3318 count = ixgbe_write_uc_addr_list(netdev);
3320 fctrl |= IXGBE_FCTRL_UPE;
3321 vmolr |= IXGBE_VMOLR_ROPE;
3325 if (adapter->num_vfs) {
3326 ixgbe_restore_vf_multicasts(adapter);
3327 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(adapter->num_vfs)) &
3328 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3330 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(adapter->num_vfs), vmolr);
3333 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3335 if (netdev->features & NETIF_F_HW_VLAN_RX)
3336 ixgbe_vlan_strip_enable(adapter);
3338 ixgbe_vlan_strip_disable(adapter);
3341 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3344 struct ixgbe_q_vector *q_vector;
3345 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3347 /* legacy and MSI only use one vector */
3348 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3351 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3352 struct napi_struct *napi;
3353 q_vector = adapter->q_vector[q_idx];
3354 napi = &q_vector->napi;
3355 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3356 if (!q_vector->rxr_count || !q_vector->txr_count) {
3357 if (q_vector->txr_count == 1)
3358 napi->poll = &ixgbe_clean_txonly;
3359 else if (q_vector->rxr_count == 1)
3360 napi->poll = &ixgbe_clean_rxonly;
3368 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3371 struct ixgbe_q_vector *q_vector;
3372 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3374 /* legacy and MSI only use one vector */
3375 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3378 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3379 q_vector = adapter->q_vector[q_idx];
3380 napi_disable(&q_vector->napi);
3384 #ifdef CONFIG_IXGBE_DCB
3386 * ixgbe_configure_dcb - Configure DCB hardware
3387 * @adapter: ixgbe adapter struct
3389 * This is called by the driver on open to configure the DCB hardware.
3390 * This is also called by the gennetlink interface when reconfiguring
3393 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3395 struct ixgbe_hw *hw = &adapter->hw;
3396 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3398 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3399 if (hw->mac.type == ixgbe_mac_82598EB)
3400 netif_set_gso_max_size(adapter->netdev, 65536);
3404 if (hw->mac.type == ixgbe_mac_82598EB)
3405 netif_set_gso_max_size(adapter->netdev, 32768);
3408 if (adapter->netdev->features & NETIF_F_FCOE_MTU)
3409 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
3412 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3414 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3417 /* Enable VLAN tag insert/strip */
3418 adapter->netdev->features |= NETIF_F_HW_VLAN_RX;
3420 hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
3422 /* reconfigure the hardware */
3423 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
3427 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3429 struct net_device *netdev = adapter->netdev;
3430 struct ixgbe_hw *hw = &adapter->hw;
3433 #ifdef CONFIG_IXGBE_DCB
3434 ixgbe_configure_dcb(adapter);
3437 ixgbe_set_rx_mode(netdev);
3438 ixgbe_restore_vlan(adapter);
3441 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
3442 ixgbe_configure_fcoe(adapter);
3444 #endif /* IXGBE_FCOE */
3445 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3446 for (i = 0; i < adapter->num_tx_queues; i++)
3447 adapter->tx_ring[i]->atr_sample_rate =
3448 adapter->atr_sample_rate;
3449 ixgbe_init_fdir_signature_82599(hw, adapter->fdir_pballoc);
3450 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3451 ixgbe_init_fdir_perfect_82599(hw, adapter->fdir_pballoc);
3453 ixgbe_configure_virtualization(adapter);
3455 ixgbe_configure_tx(adapter);
3456 ixgbe_configure_rx(adapter);
3459 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3461 switch (hw->phy.type) {
3462 case ixgbe_phy_sfp_avago:
3463 case ixgbe_phy_sfp_ftl:
3464 case ixgbe_phy_sfp_intel:
3465 case ixgbe_phy_sfp_unknown:
3466 case ixgbe_phy_sfp_passive_tyco:
3467 case ixgbe_phy_sfp_passive_unknown:
3468 case ixgbe_phy_sfp_active_unknown:
3469 case ixgbe_phy_sfp_ftl_active:
3477 * ixgbe_sfp_link_config - set up SFP+ link
3478 * @adapter: pointer to private adapter struct
3480 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3482 struct ixgbe_hw *hw = &adapter->hw;
3484 if (hw->phy.multispeed_fiber) {
3486 * In multispeed fiber setups, the device may not have
3487 * had a physical connection when the driver loaded.
3488 * If that's the case, the initial link configuration
3489 * couldn't get the MAC into 10G or 1G mode, so we'll
3490 * never have a link status change interrupt fire.
3491 * We need to try and force an autonegotiation
3492 * session, then bring up link.
3494 hw->mac.ops.setup_sfp(hw);
3495 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
3496 schedule_work(&adapter->multispeed_fiber_task);
3499 * Direct Attach Cu and non-multispeed fiber modules
3500 * still need to be configured properly prior to
3503 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_MOD_TASK))
3504 schedule_work(&adapter->sfp_config_module_task);
3509 * ixgbe_non_sfp_link_config - set up non-SFP+ link
3510 * @hw: pointer to private hardware struct
3512 * Returns 0 on success, negative on failure
3514 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3517 bool negotiation, link_up = false;
3518 u32 ret = IXGBE_ERR_LINK_SETUP;
3520 if (hw->mac.ops.check_link)
3521 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3526 if (hw->mac.ops.get_link_capabilities)
3527 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
3532 if (hw->mac.ops.setup_link)
3533 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
3538 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
3540 struct ixgbe_hw *hw = &adapter->hw;
3543 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3544 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
3546 gpie |= IXGBE_GPIE_EIAME;
3548 * use EIAM to auto-mask when MSI-X interrupt is asserted
3549 * this saves a register write for every interrupt
3551 switch (hw->mac.type) {
3552 case ixgbe_mac_82598EB:
3553 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3556 case ixgbe_mac_82599EB:
3557 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
3558 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
3562 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
3563 * specifically only auto mask tx and rx interrupts */
3564 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3567 /* XXX: to interrupt immediately for EICS writes, enable this */
3568 /* gpie |= IXGBE_GPIE_EIMEN; */
3570 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3571 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
3572 gpie |= IXGBE_GPIE_VTMODE_64;
3575 /* Enable fan failure interrupt */
3576 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3577 gpie |= IXGBE_SDP1_GPIEN;
3579 if (hw->mac.type == ixgbe_mac_82599EB)
3580 gpie |= IXGBE_SDP1_GPIEN;
3581 gpie |= IXGBE_SDP2_GPIEN;
3583 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3586 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
3588 struct ixgbe_hw *hw = &adapter->hw;
3592 ixgbe_get_hw_control(adapter);
3593 ixgbe_setup_gpie(adapter);
3595 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3596 ixgbe_configure_msix(adapter);
3598 ixgbe_configure_msi_and_legacy(adapter);
3600 /* enable the optics */
3601 if (hw->phy.multispeed_fiber)
3602 hw->mac.ops.enable_tx_laser(hw);
3604 clear_bit(__IXGBE_DOWN, &adapter->state);
3605 ixgbe_napi_enable_all(adapter);
3607 if (ixgbe_is_sfp(hw)) {
3608 ixgbe_sfp_link_config(adapter);
3610 err = ixgbe_non_sfp_link_config(hw);
3612 e_err(probe, "link_config FAILED %d\n", err);
3615 /* clear any pending interrupts, may auto mask */
3616 IXGBE_READ_REG(hw, IXGBE_EICR);
3617 ixgbe_irq_enable(adapter, true, true);
3620 * If this adapter has a fan, check to see if we had a failure
3621 * before we enabled the interrupt.
3623 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3624 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
3625 if (esdp & IXGBE_ESDP_SDP1)
3626 e_crit(drv, "Fan has stopped, replace the adapter\n");
3630 * For hot-pluggable SFP+ devices, a new SFP+ module may have
3631 * arrived before interrupts were enabled but after probe. Such
3632 * devices wouldn't have their type identified yet. We need to
3633 * kick off the SFP+ module setup first, then try to bring up link.
3634 * If we're not hot-pluggable SFP+, we just need to configure link
3637 if (hw->phy.type == ixgbe_phy_unknown)
3638 schedule_work(&adapter->sfp_config_module_task);
3640 /* enable transmits */
3641 netif_tx_start_all_queues(adapter->netdev);
3643 /* bring the link up in the watchdog, this could race with our first
3644 * link up interrupt but shouldn't be a problem */
3645 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3646 adapter->link_check_timeout = jiffies;
3647 mod_timer(&adapter->watchdog_timer, jiffies);
3649 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
3650 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
3651 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
3652 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
3657 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
3659 WARN_ON(in_interrupt());
3660 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
3662 ixgbe_down(adapter);
3664 * If SR-IOV enabled then wait a bit before bringing the adapter
3665 * back up to give the VFs time to respond to the reset. The
3666 * two second wait is based upon the watchdog timer cycle in
3669 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3672 clear_bit(__IXGBE_RESETTING, &adapter->state);
3675 int ixgbe_up(struct ixgbe_adapter *adapter)
3677 /* hardware has been reset, we need to reload some things */
3678 ixgbe_configure(adapter);
3680 return ixgbe_up_complete(adapter);
3683 void ixgbe_reset(struct ixgbe_adapter *adapter)
3685 struct ixgbe_hw *hw = &adapter->hw;
3688 err = hw->mac.ops.init_hw(hw);
3691 case IXGBE_ERR_SFP_NOT_PRESENT:
3693 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
3694 e_dev_err("master disable timed out\n");
3696 case IXGBE_ERR_EEPROM_VERSION:
3697 /* We are running on a pre-production device, log a warning */
3698 e_dev_warn("This device is a pre-production adapter/LOM. "
3699 "Please be aware there may be issuesassociated with "
3700 "your hardware. If you are experiencing problems "
3701 "please contact your Intel or hardware "
3702 "representative who provided you with this "
3706 e_dev_err("Hardware Error: %d\n", err);
3709 /* reprogram the RAR[0] in case user changed it. */
3710 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
3715 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
3716 * @rx_ring: ring to free buffers from
3718 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
3720 struct device *dev = rx_ring->dev;
3724 /* ring already cleared, nothing to do */
3725 if (!rx_ring->rx_buffer_info)
3728 /* Free all the Rx ring sk_buffs */
3729 for (i = 0; i < rx_ring->count; i++) {
3730 struct ixgbe_rx_buffer *rx_buffer_info;
3732 rx_buffer_info = &rx_ring->rx_buffer_info[i];
3733 if (rx_buffer_info->dma) {
3734 dma_unmap_single(rx_ring->dev, rx_buffer_info->dma,
3735 rx_ring->rx_buf_len,
3737 rx_buffer_info->dma = 0;
3739 if (rx_buffer_info->skb) {
3740 struct sk_buff *skb = rx_buffer_info->skb;
3741 rx_buffer_info->skb = NULL;
3743 struct sk_buff *this = skb;
3744 if (IXGBE_RSC_CB(this)->delay_unmap) {
3745 dma_unmap_single(dev,
3746 IXGBE_RSC_CB(this)->dma,
3747 rx_ring->rx_buf_len,
3749 IXGBE_RSC_CB(this)->dma = 0;
3750 IXGBE_RSC_CB(skb)->delay_unmap = false;
3753 dev_kfree_skb(this);
3756 if (!rx_buffer_info->page)
3758 if (rx_buffer_info->page_dma) {
3759 dma_unmap_page(dev, rx_buffer_info->page_dma,
3760 PAGE_SIZE / 2, DMA_FROM_DEVICE);
3761 rx_buffer_info->page_dma = 0;
3763 put_page(rx_buffer_info->page);
3764 rx_buffer_info->page = NULL;
3765 rx_buffer_info->page_offset = 0;
3768 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
3769 memset(rx_ring->rx_buffer_info, 0, size);
3771 /* Zero out the descriptor ring */
3772 memset(rx_ring->desc, 0, rx_ring->size);
3774 rx_ring->next_to_clean = 0;
3775 rx_ring->next_to_use = 0;
3779 * ixgbe_clean_tx_ring - Free Tx Buffers
3780 * @tx_ring: ring to be cleaned
3782 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
3784 struct ixgbe_tx_buffer *tx_buffer_info;
3788 /* ring already cleared, nothing to do */
3789 if (!tx_ring->tx_buffer_info)
3792 /* Free all the Tx ring sk_buffs */
3793 for (i = 0; i < tx_ring->count; i++) {
3794 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3795 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
3798 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
3799 memset(tx_ring->tx_buffer_info, 0, size);
3801 /* Zero out the descriptor ring */
3802 memset(tx_ring->desc, 0, tx_ring->size);
3804 tx_ring->next_to_use = 0;
3805 tx_ring->next_to_clean = 0;
3809 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
3810 * @adapter: board private structure
3812 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
3816 for (i = 0; i < adapter->num_rx_queues; i++)
3817 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
3821 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
3822 * @adapter: board private structure
3824 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
3828 for (i = 0; i < adapter->num_tx_queues; i++)
3829 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
3832 void ixgbe_down(struct ixgbe_adapter *adapter)
3834 struct net_device *netdev = adapter->netdev;
3835 struct ixgbe_hw *hw = &adapter->hw;
3839 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3841 /* signal that we are down to the interrupt handler */
3842 set_bit(__IXGBE_DOWN, &adapter->state);
3844 /* disable receive for all VFs and wait one second */
3845 if (adapter->num_vfs) {
3846 /* ping all the active vfs to let them know we are going down */
3847 ixgbe_ping_all_vfs(adapter);
3849 /* Disable all VFTE/VFRE TX/RX */
3850 ixgbe_disable_tx_rx(adapter);
3852 /* Mark all the VFs as inactive */
3853 for (i = 0 ; i < adapter->num_vfs; i++)
3854 adapter->vfinfo[i].clear_to_send = 0;
3857 /* disable receives */
3858 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3859 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3861 IXGBE_WRITE_FLUSH(hw);
3864 netif_tx_stop_all_queues(netdev);
3866 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
3867 del_timer_sync(&adapter->sfp_timer);
3868 del_timer_sync(&adapter->watchdog_timer);
3869 cancel_work_sync(&adapter->watchdog_task);
3871 netif_carrier_off(netdev);
3872 netif_tx_disable(netdev);
3874 ixgbe_irq_disable(adapter);
3876 ixgbe_napi_disable_all(adapter);
3878 /* Cleanup the affinity_hint CPU mask memory and callback */
3879 for (i = 0; i < num_q_vectors; i++) {
3880 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
3881 /* clear the affinity_mask in the IRQ descriptor */
3882 irq_set_affinity_hint(adapter->msix_entries[i]. vector, NULL);
3883 /* release the CPU mask memory */
3884 free_cpumask_var(q_vector->affinity_mask);
3887 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3888 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3889 cancel_work_sync(&adapter->fdir_reinit_task);
3891 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
3892 cancel_work_sync(&adapter->check_overtemp_task);
3894 /* disable transmits in the hardware now that interrupts are off */
3895 for (i = 0; i < adapter->num_tx_queues; i++) {
3896 j = adapter->tx_ring[i]->reg_idx;
3897 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3898 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j),
3899 (txdctl & ~IXGBE_TXDCTL_ENABLE));
3901 /* Disable the Tx DMA engine on 82599 */
3902 if (hw->mac.type == ixgbe_mac_82599EB)
3903 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
3904 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
3905 ~IXGBE_DMATXCTL_TE));
3907 /* power down the optics */
3908 if (hw->phy.multispeed_fiber)
3909 hw->mac.ops.disable_tx_laser(hw);
3911 /* clear n-tuple filters that are cached */
3912 ethtool_ntuple_flush(netdev);
3914 if (!pci_channel_offline(adapter->pdev))
3915 ixgbe_reset(adapter);
3916 ixgbe_clean_all_tx_rings(adapter);
3917 ixgbe_clean_all_rx_rings(adapter);
3919 #ifdef CONFIG_IXGBE_DCA
3920 /* since we reset the hardware DCA settings were cleared */
3921 ixgbe_setup_dca(adapter);
3926 * ixgbe_poll - NAPI Rx polling callback
3927 * @napi: structure for representing this polling device
3928 * @budget: how many packets driver is allowed to clean
3930 * This function is used for legacy and MSI, NAPI mode
3932 static int ixgbe_poll(struct napi_struct *napi, int budget)
3934 struct ixgbe_q_vector *q_vector =
3935 container_of(napi, struct ixgbe_q_vector, napi);
3936 struct ixgbe_adapter *adapter = q_vector->adapter;
3937 int tx_clean_complete, work_done = 0;
3939 #ifdef CONFIG_IXGBE_DCA
3940 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
3941 ixgbe_update_dca(q_vector);
3944 tx_clean_complete = ixgbe_clean_tx_irq(q_vector, adapter->tx_ring[0]);
3945 ixgbe_clean_rx_irq(q_vector, adapter->rx_ring[0], &work_done, budget);
3947 if (!tx_clean_complete)
3950 /* If budget not fully consumed, exit the polling mode */
3951 if (work_done < budget) {
3952 napi_complete(napi);
3953 if (adapter->rx_itr_setting & 1)
3954 ixgbe_set_itr(adapter);
3955 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3956 ixgbe_irq_enable_queues(adapter, IXGBE_EIMS_RTX_QUEUE);
3962 * ixgbe_tx_timeout - Respond to a Tx Hang
3963 * @netdev: network interface device structure
3965 static void ixgbe_tx_timeout(struct net_device *netdev)
3967 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3969 /* Do the reset outside of interrupt context */
3970 schedule_work(&adapter->reset_task);
3973 static void ixgbe_reset_task(struct work_struct *work)
3975 struct ixgbe_adapter *adapter;
3976 adapter = container_of(work, struct ixgbe_adapter, reset_task);
3978 /* If we're already down or resetting, just bail */
3979 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
3980 test_bit(__IXGBE_RESETTING, &adapter->state))
3983 adapter->tx_timeout_count++;
3985 ixgbe_dump(adapter);
3986 netdev_err(adapter->netdev, "Reset adapter\n");
3987 ixgbe_reinit_locked(adapter);
3990 #ifdef CONFIG_IXGBE_DCB
3991 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
3994 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_DCB];
3996 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
4000 adapter->num_rx_queues = f->indices;
4001 adapter->num_tx_queues = f->indices;
4009 * ixgbe_set_rss_queues: Allocate queues for RSS
4010 * @adapter: board private structure to initialize
4012 * This is our "base" multiqueue mode. RSS (Receive Side Scaling) will try
4013 * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
4016 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
4019 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
4021 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4023 adapter->num_rx_queues = f->indices;
4024 adapter->num_tx_queues = f->indices;
4034 * ixgbe_set_fdir_queues: Allocate queues for Flow Director
4035 * @adapter: board private structure to initialize
4037 * Flow Director is an advanced Rx filter, attempting to get Rx flows back
4038 * to the original CPU that initiated the Tx session. This runs in addition
4039 * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
4040 * Rx load across CPUs using RSS.
4043 static inline bool ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
4046 struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
4048 f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
4051 /* Flow Director must have RSS enabled */
4052 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
4053 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
4054 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)))) {
4055 adapter->num_tx_queues = f_fdir->indices;
4056 adapter->num_rx_queues = f_fdir->indices;
4059 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4060 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4067 * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
4068 * @adapter: board private structure to initialize
4070 * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
4071 * The ring feature mask is not used as a mask for FCoE, as it can take any 8
4072 * rx queues out of the max number of rx queues, instead, it is used as the
4073 * index of the first rx queue used by FCoE.
4076 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
4079 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4081 f->indices = min((int)num_online_cpus(), f->indices);
4082 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
4083 adapter->num_rx_queues = 1;
4084 adapter->num_tx_queues = 1;
4085 #ifdef CONFIG_IXGBE_DCB
4086 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4087 e_info(probe, "FCoE enabled with DCB\n");
4088 ixgbe_set_dcb_queues(adapter);
4091 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4092 e_info(probe, "FCoE enabled with RSS\n");
4093 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4094 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
4095 ixgbe_set_fdir_queues(adapter);
4097 ixgbe_set_rss_queues(adapter);
4099 /* adding FCoE rx rings to the end */
4100 f->mask = adapter->num_rx_queues;
4101 adapter->num_rx_queues += f->indices;
4102 adapter->num_tx_queues += f->indices;
4110 #endif /* IXGBE_FCOE */
4112 * ixgbe_set_sriov_queues: Allocate queues for IOV use
4113 * @adapter: board private structure to initialize
4115 * IOV doesn't actually use anything, so just NAK the
4116 * request for now and let the other queue routines
4117 * figure out what to do.
4119 static inline bool ixgbe_set_sriov_queues(struct ixgbe_adapter *adapter)
4125 * ixgbe_set_num_queues: Allocate queues for device, feature dependant
4126 * @adapter: board private structure to initialize
4128 * This is the top level queue allocation routine. The order here is very
4129 * important, starting with the "most" number of features turned on at once,
4130 * and ending with the smallest set of features. This way large combinations
4131 * can be allocated if they're turned on, and smaller combinations are the
4132 * fallthrough conditions.
4135 static int ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
4137 /* Start with base case */
4138 adapter->num_rx_queues = 1;
4139 adapter->num_tx_queues = 1;
4140 adapter->num_rx_pools = adapter->num_rx_queues;
4141 adapter->num_rx_queues_per_pool = 1;
4143 if (ixgbe_set_sriov_queues(adapter))
4147 if (ixgbe_set_fcoe_queues(adapter))
4150 #endif /* IXGBE_FCOE */
4151 #ifdef CONFIG_IXGBE_DCB
4152 if (ixgbe_set_dcb_queues(adapter))
4156 if (ixgbe_set_fdir_queues(adapter))
4159 if (ixgbe_set_rss_queues(adapter))
4162 /* fallback to base case */
4163 adapter->num_rx_queues = 1;
4164 adapter->num_tx_queues = 1;
4167 /* Notify the stack of the (possibly) reduced queue counts. */
4168 netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
4169 return netif_set_real_num_rx_queues(adapter->netdev,
4170 adapter->num_rx_queues);
4173 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
4176 int err, vector_threshold;
4178 /* We'll want at least 3 (vector_threshold):
4181 * 3) Other (Link Status Change, etc.)
4182 * 4) TCP Timer (optional)
4184 vector_threshold = MIN_MSIX_COUNT;
4186 /* The more we get, the more we will assign to Tx/Rx Cleanup
4187 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
4188 * Right now, we simply care about how many we'll get; we'll
4189 * set them up later while requesting irq's.
4191 while (vectors >= vector_threshold) {
4192 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
4194 if (!err) /* Success in acquiring all requested vectors. */
4197 vectors = 0; /* Nasty failure, quit now */
4198 else /* err == number of vectors we should try again with */
4202 if (vectors < vector_threshold) {
4203 /* Can't allocate enough MSI-X interrupts? Oh well.
4204 * This just means we'll go with either a single MSI
4205 * vector or fall back to legacy interrupts.
4207 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4208 "Unable to allocate MSI-X interrupts\n");
4209 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4210 kfree(adapter->msix_entries);
4211 adapter->msix_entries = NULL;
4213 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
4215 * Adjust for only the vectors we'll use, which is minimum
4216 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
4217 * vectors we were allocated.
4219 adapter->num_msix_vectors = min(vectors,
4220 adapter->max_msix_q_vectors + NON_Q_VECTORS);
4225 * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
4226 * @adapter: board private structure to initialize
4228 * Cache the descriptor ring offsets for RSS to the assigned rings.
4231 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
4236 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4237 for (i = 0; i < adapter->num_rx_queues; i++)
4238 adapter->rx_ring[i]->reg_idx = i;
4239 for (i = 0; i < adapter->num_tx_queues; i++)
4240 adapter->tx_ring[i]->reg_idx = i;
4249 #ifdef CONFIG_IXGBE_DCB
4251 * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
4252 * @adapter: board private structure to initialize
4254 * Cache the descriptor ring offsets for DCB to the assigned rings.
4257 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
4261 int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
4263 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4264 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
4265 /* the number of queues is assumed to be symmetric */
4266 for (i = 0; i < dcb_i; i++) {
4267 adapter->rx_ring[i]->reg_idx = i << 3;
4268 adapter->tx_ring[i]->reg_idx = i << 2;
4271 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
4274 * Tx TC0 starts at: descriptor queue 0
4275 * Tx TC1 starts at: descriptor queue 32
4276 * Tx TC2 starts at: descriptor queue 64
4277 * Tx TC3 starts at: descriptor queue 80
4278 * Tx TC4 starts at: descriptor queue 96
4279 * Tx TC5 starts at: descriptor queue 104
4280 * Tx TC6 starts at: descriptor queue 112
4281 * Tx TC7 starts at: descriptor queue 120
4283 * Rx TC0-TC7 are offset by 16 queues each
4285 for (i = 0; i < 3; i++) {
4286 adapter->tx_ring[i]->reg_idx = i << 5;
4287 adapter->rx_ring[i]->reg_idx = i << 4;
4289 for ( ; i < 5; i++) {
4290 adapter->tx_ring[i]->reg_idx =
4292 adapter->rx_ring[i]->reg_idx = i << 4;
4294 for ( ; i < dcb_i; i++) {
4295 adapter->tx_ring[i]->reg_idx =
4297 adapter->rx_ring[i]->reg_idx = i << 4;
4301 } else if (dcb_i == 4) {
4303 * Tx TC0 starts at: descriptor queue 0
4304 * Tx TC1 starts at: descriptor queue 64
4305 * Tx TC2 starts at: descriptor queue 96
4306 * Tx TC3 starts at: descriptor queue 112
4308 * Rx TC0-TC3 are offset by 32 queues each
4310 adapter->tx_ring[0]->reg_idx = 0;
4311 adapter->tx_ring[1]->reg_idx = 64;
4312 adapter->tx_ring[2]->reg_idx = 96;
4313 adapter->tx_ring[3]->reg_idx = 112;
4314 for (i = 0 ; i < dcb_i; i++)
4315 adapter->rx_ring[i]->reg_idx = i << 5;
4333 * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
4334 * @adapter: board private structure to initialize
4336 * Cache the descriptor ring offsets for Flow Director to the assigned rings.
4339 static inline bool ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
4344 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
4345 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4346 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))) {
4347 for (i = 0; i < adapter->num_rx_queues; i++)
4348 adapter->rx_ring[i]->reg_idx = i;
4349 for (i = 0; i < adapter->num_tx_queues; i++)
4350 adapter->tx_ring[i]->reg_idx = i;
4359 * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
4360 * @adapter: board private structure to initialize
4362 * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
4365 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
4367 int i, fcoe_rx_i = 0, fcoe_tx_i = 0;
4369 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4371 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
4372 #ifdef CONFIG_IXGBE_DCB
4373 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4374 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
4376 ixgbe_cache_ring_dcb(adapter);
4377 /* find out queues in TC for FCoE */
4378 fcoe_rx_i = adapter->rx_ring[fcoe->tc]->reg_idx + 1;
4379 fcoe_tx_i = adapter->tx_ring[fcoe->tc]->reg_idx + 1;
4381 * In 82599, the number of Tx queues for each traffic
4382 * class for both 8-TC and 4-TC modes are:
4383 * TCs : TC0 TC1 TC2 TC3 TC4 TC5 TC6 TC7
4384 * 8 TCs: 32 32 16 16 8 8 8 8
4385 * 4 TCs: 64 64 32 32
4386 * We have max 8 queues for FCoE, where 8 the is
4387 * FCoE redirection table size. If TC for FCoE is
4388 * less than or equal to TC3, we have enough queues
4389 * to add max of 8 queues for FCoE, so we start FCoE
4390 * tx descriptor from the next one, i.e., reg_idx + 1.
4391 * If TC for FCoE is above TC3, implying 8 TC mode,
4392 * and we need 8 for FCoE, we have to take all queues
4393 * in that traffic class for FCoE.
4395 if ((f->indices == IXGBE_FCRETA_SIZE) && (fcoe->tc > 3))
4398 #endif /* CONFIG_IXGBE_DCB */
4399 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4400 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4401 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
4402 ixgbe_cache_ring_fdir(adapter);
4404 ixgbe_cache_ring_rss(adapter);
4406 fcoe_rx_i = f->mask;
4407 fcoe_tx_i = f->mask;
4409 for (i = 0; i < f->indices; i++, fcoe_rx_i++, fcoe_tx_i++) {
4410 adapter->rx_ring[f->mask + i]->reg_idx = fcoe_rx_i;
4411 adapter->tx_ring[f->mask + i]->reg_idx = fcoe_tx_i;
4418 #endif /* IXGBE_FCOE */
4420 * ixgbe_cache_ring_sriov - Descriptor ring to register mapping for sriov
4421 * @adapter: board private structure to initialize
4423 * SR-IOV doesn't use any descriptor rings but changes the default if
4424 * no other mapping is used.
4427 static inline bool ixgbe_cache_ring_sriov(struct ixgbe_adapter *adapter)
4429 adapter->rx_ring[0]->reg_idx = adapter->num_vfs * 2;
4430 adapter->tx_ring[0]->reg_idx = adapter->num_vfs * 2;
4431 if (adapter->num_vfs)
4438 * ixgbe_cache_ring_register - Descriptor ring to register mapping
4439 * @adapter: board private structure to initialize
4441 * Once we know the feature-set enabled for the device, we'll cache
4442 * the register offset the descriptor ring is assigned to.
4444 * Note, the order the various feature calls is important. It must start with
4445 * the "most" features enabled at the same time, then trickle down to the
4446 * least amount of features turned on at once.
4448 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
4450 /* start with default case */
4451 adapter->rx_ring[0]->reg_idx = 0;
4452 adapter->tx_ring[0]->reg_idx = 0;
4454 if (ixgbe_cache_ring_sriov(adapter))
4458 if (ixgbe_cache_ring_fcoe(adapter))
4461 #endif /* IXGBE_FCOE */
4462 #ifdef CONFIG_IXGBE_DCB
4463 if (ixgbe_cache_ring_dcb(adapter))
4467 if (ixgbe_cache_ring_fdir(adapter))
4470 if (ixgbe_cache_ring_rss(adapter))
4475 * ixgbe_alloc_queues - Allocate memory for all rings
4476 * @adapter: board private structure to initialize
4478 * We allocate one ring per queue at run-time since we don't know the
4479 * number of queues at compile-time. The polling_netdev array is
4480 * intended for Multiqueue, but should work fine with a single queue.
4482 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
4486 int orig_node = adapter->node;
4488 for (i = 0; i < adapter->num_tx_queues; i++) {
4489 struct ixgbe_ring *ring = adapter->tx_ring[i];
4490 if (orig_node == -1) {
4491 int cur_node = next_online_node(adapter->node);
4492 if (cur_node == MAX_NUMNODES)
4493 cur_node = first_online_node;
4494 adapter->node = cur_node;
4496 ring = kzalloc_node(sizeof(struct ixgbe_ring), GFP_KERNEL,
4499 ring = kzalloc(sizeof(struct ixgbe_ring), GFP_KERNEL);
4501 goto err_tx_ring_allocation;
4502 ring->count = adapter->tx_ring_count;
4503 ring->queue_index = i;
4504 ring->dev = &adapter->pdev->dev;
4505 ring->netdev = adapter->netdev;
4506 ring->numa_node = adapter->node;
4508 adapter->tx_ring[i] = ring;
4511 /* Restore the adapter's original node */
4512 adapter->node = orig_node;
4514 rx_count = adapter->rx_ring_count;
4515 for (i = 0; i < adapter->num_rx_queues; i++) {
4516 struct ixgbe_ring *ring = adapter->rx_ring[i];
4517 if (orig_node == -1) {
4518 int cur_node = next_online_node(adapter->node);
4519 if (cur_node == MAX_NUMNODES)
4520 cur_node = first_online_node;
4521 adapter->node = cur_node;
4523 ring = kzalloc_node(sizeof(struct ixgbe_ring), GFP_KERNEL,
4526 ring = kzalloc(sizeof(struct ixgbe_ring), GFP_KERNEL);
4528 goto err_rx_ring_allocation;
4529 ring->count = rx_count;
4530 ring->queue_index = i;
4531 ring->dev = &adapter->pdev->dev;
4532 ring->netdev = adapter->netdev;
4533 ring->numa_node = adapter->node;
4535 adapter->rx_ring[i] = ring;
4538 /* Restore the adapter's original node */
4539 adapter->node = orig_node;
4541 ixgbe_cache_ring_register(adapter);
4545 err_rx_ring_allocation:
4546 for (i = 0; i < adapter->num_tx_queues; i++)
4547 kfree(adapter->tx_ring[i]);
4548 err_tx_ring_allocation:
4553 * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
4554 * @adapter: board private structure to initialize
4556 * Attempt to configure the interrupts using the best available
4557 * capabilities of the hardware and the kernel.
4559 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
4561 struct ixgbe_hw *hw = &adapter->hw;
4563 int vector, v_budget;
4566 * It's easy to be greedy for MSI-X vectors, but it really
4567 * doesn't do us much good if we have a lot more vectors
4568 * than CPU's. So let's be conservative and only ask for
4569 * (roughly) the same number of vectors as there are CPU's.
4571 v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
4572 (int)num_online_cpus()) + NON_Q_VECTORS;
4575 * At the same time, hardware can only support a maximum of
4576 * hw.mac->max_msix_vectors vectors. With features
4577 * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
4578 * descriptor queues supported by our device. Thus, we cap it off in
4579 * those rare cases where the cpu count also exceeds our vector limit.
4581 v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
4583 /* A failure in MSI-X entry allocation isn't fatal, but it does
4584 * mean we disable MSI-X capabilities of the adapter. */
4585 adapter->msix_entries = kcalloc(v_budget,
4586 sizeof(struct msix_entry), GFP_KERNEL);
4587 if (adapter->msix_entries) {
4588 for (vector = 0; vector < v_budget; vector++)
4589 adapter->msix_entries[vector].entry = vector;
4591 ixgbe_acquire_msix_vectors(adapter, v_budget);
4593 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4597 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
4598 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
4599 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4600 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4601 adapter->atr_sample_rate = 0;
4602 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4603 ixgbe_disable_sriov(adapter);
4605 err = ixgbe_set_num_queues(adapter);
4609 err = pci_enable_msi(adapter->pdev);
4611 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
4613 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4614 "Unable to allocate MSI interrupt, "
4615 "falling back to legacy. Error: %d\n", err);
4625 * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
4626 * @adapter: board private structure to initialize
4628 * We allocate one q_vector per queue interrupt. If allocation fails we
4631 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
4633 int q_idx, num_q_vectors;
4634 struct ixgbe_q_vector *q_vector;
4636 int (*poll)(struct napi_struct *, int);
4638 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4639 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4640 napi_vectors = adapter->num_rx_queues;
4641 poll = &ixgbe_clean_rxtx_many;
4648 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4649 q_vector = kzalloc_node(sizeof(struct ixgbe_q_vector),
4650 GFP_KERNEL, adapter->node);
4652 q_vector = kzalloc(sizeof(struct ixgbe_q_vector),
4656 q_vector->adapter = adapter;
4657 if (q_vector->txr_count && !q_vector->rxr_count)
4658 q_vector->eitr = adapter->tx_eitr_param;
4660 q_vector->eitr = adapter->rx_eitr_param;
4661 q_vector->v_idx = q_idx;
4662 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
4663 adapter->q_vector[q_idx] = q_vector;
4671 q_vector = adapter->q_vector[q_idx];
4672 netif_napi_del(&q_vector->napi);
4674 adapter->q_vector[q_idx] = NULL;
4680 * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
4681 * @adapter: board private structure to initialize
4683 * This function frees the memory allocated to the q_vectors. In addition if
4684 * NAPI is enabled it will delete any references to the NAPI struct prior
4685 * to freeing the q_vector.
4687 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
4689 int q_idx, num_q_vectors;
4691 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4692 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4696 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4697 struct ixgbe_q_vector *q_vector = adapter->q_vector[q_idx];
4698 adapter->q_vector[q_idx] = NULL;
4699 netif_napi_del(&q_vector->napi);
4704 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
4706 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4707 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4708 pci_disable_msix(adapter->pdev);
4709 kfree(adapter->msix_entries);
4710 adapter->msix_entries = NULL;
4711 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
4712 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
4713 pci_disable_msi(adapter->pdev);
4718 * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
4719 * @adapter: board private structure to initialize
4721 * We determine which interrupt scheme to use based on...
4722 * - Kernel support (MSI, MSI-X)
4723 * - which can be user-defined (via MODULE_PARAM)
4724 * - Hardware queue count (num_*_queues)
4725 * - defined by miscellaneous hardware support/features (RSS, etc.)
4727 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
4731 /* Number of supported queues */
4732 err = ixgbe_set_num_queues(adapter);
4736 err = ixgbe_set_interrupt_capability(adapter);
4738 e_dev_err("Unable to setup interrupt capabilities\n");
4739 goto err_set_interrupt;
4742 err = ixgbe_alloc_q_vectors(adapter);
4744 e_dev_err("Unable to allocate memory for queue vectors\n");
4745 goto err_alloc_q_vectors;
4748 err = ixgbe_alloc_queues(adapter);
4750 e_dev_err("Unable to allocate memory for queues\n");
4751 goto err_alloc_queues;
4754 e_dev_info("Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
4755 (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
4756 adapter->num_rx_queues, adapter->num_tx_queues);
4758 set_bit(__IXGBE_DOWN, &adapter->state);
4763 ixgbe_free_q_vectors(adapter);
4764 err_alloc_q_vectors:
4765 ixgbe_reset_interrupt_capability(adapter);
4770 static void ring_free_rcu(struct rcu_head *head)
4772 kfree(container_of(head, struct ixgbe_ring, rcu));
4776 * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
4777 * @adapter: board private structure to clear interrupt scheme on
4779 * We go through and clear interrupt specific resources and reset the structure
4780 * to pre-load conditions
4782 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
4786 for (i = 0; i < adapter->num_tx_queues; i++) {
4787 kfree(adapter->tx_ring[i]);
4788 adapter->tx_ring[i] = NULL;
4790 for (i = 0; i < adapter->num_rx_queues; i++) {
4791 struct ixgbe_ring *ring = adapter->rx_ring[i];
4793 /* ixgbe_get_stats64() might access this ring, we must wait
4794 * a grace period before freeing it.
4796 call_rcu(&ring->rcu, ring_free_rcu);
4797 adapter->rx_ring[i] = NULL;
4800 ixgbe_free_q_vectors(adapter);
4801 ixgbe_reset_interrupt_capability(adapter);
4805 * ixgbe_sfp_timer - worker thread to find a missing module
4806 * @data: pointer to our adapter struct
4808 static void ixgbe_sfp_timer(unsigned long data)
4810 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
4813 * Do the sfp_timer outside of interrupt context due to the
4814 * delays that sfp+ detection requires
4816 schedule_work(&adapter->sfp_task);
4820 * ixgbe_sfp_task - worker thread to find a missing module
4821 * @work: pointer to work_struct containing our data
4823 static void ixgbe_sfp_task(struct work_struct *work)
4825 struct ixgbe_adapter *adapter = container_of(work,
4826 struct ixgbe_adapter,
4828 struct ixgbe_hw *hw = &adapter->hw;
4830 if ((hw->phy.type == ixgbe_phy_nl) &&
4831 (hw->phy.sfp_type == ixgbe_sfp_type_not_present)) {
4832 s32 ret = hw->phy.ops.identify_sfp(hw);
4833 if (ret == IXGBE_ERR_SFP_NOT_PRESENT)
4835 ret = hw->phy.ops.reset(hw);
4836 if (ret == IXGBE_ERR_SFP_NOT_SUPPORTED) {
4837 e_dev_err("failed to initialize because an unsupported "
4838 "SFP+ module type was detected.\n");
4839 e_dev_err("Reload the driver after installing a "
4840 "supported module.\n");
4841 unregister_netdev(adapter->netdev);
4843 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
4845 /* don't need this routine any more */
4846 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
4850 if (test_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state))
4851 mod_timer(&adapter->sfp_timer,
4852 round_jiffies(jiffies + (2 * HZ)));
4856 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4857 * @adapter: board private structure to initialize
4859 * ixgbe_sw_init initializes the Adapter private data structure.
4860 * Fields are initialized based on PCI device information and
4861 * OS network device settings (MTU size).
4863 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
4865 struct ixgbe_hw *hw = &adapter->hw;
4866 struct pci_dev *pdev = adapter->pdev;
4867 struct net_device *dev = adapter->netdev;
4869 #ifdef CONFIG_IXGBE_DCB
4871 struct tc_configuration *tc;
4873 int max_frame = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
4875 /* PCI config space info */
4877 hw->vendor_id = pdev->vendor;
4878 hw->device_id = pdev->device;
4879 hw->revision_id = pdev->revision;
4880 hw->subsystem_vendor_id = pdev->subsystem_vendor;
4881 hw->subsystem_device_id = pdev->subsystem_device;
4883 /* Set capability flags */
4884 rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
4885 adapter->ring_feature[RING_F_RSS].indices = rss;
4886 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
4887 adapter->ring_feature[RING_F_DCB].indices = IXGBE_MAX_DCB_INDICES;
4888 if (hw->mac.type == ixgbe_mac_82598EB) {
4889 if (hw->device_id == IXGBE_DEV_ID_82598AT)
4890 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4891 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
4892 } else if (hw->mac.type == ixgbe_mac_82599EB) {
4893 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
4894 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4895 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4896 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
4897 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4898 if (dev->features & NETIF_F_NTUPLE) {
4899 /* Flow Director perfect filter enabled */
4900 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4901 adapter->atr_sample_rate = 0;
4902 spin_lock_init(&adapter->fdir_perfect_lock);
4904 /* Flow Director hash filters enabled */
4905 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
4906 adapter->atr_sample_rate = 20;
4908 adapter->ring_feature[RING_F_FDIR].indices =
4909 IXGBE_MAX_FDIR_INDICES;
4910 adapter->fdir_pballoc = 0;
4912 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4913 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4914 adapter->ring_feature[RING_F_FCOE].indices = 0;
4915 #ifdef CONFIG_IXGBE_DCB
4916 /* Default traffic class to use for FCoE */
4917 adapter->fcoe.tc = IXGBE_FCOE_DEFTC;
4918 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
4920 #endif /* IXGBE_FCOE */
4923 #ifdef CONFIG_IXGBE_DCB
4924 /* Configure DCB traffic classes */
4925 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
4926 tc = &adapter->dcb_cfg.tc_config[j];
4927 tc->path[DCB_TX_CONFIG].bwg_id = 0;
4928 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
4929 tc->path[DCB_RX_CONFIG].bwg_id = 0;
4930 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
4931 tc->dcb_pfc = pfc_disabled;
4933 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
4934 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
4935 adapter->dcb_cfg.rx_pba_cfg = pba_equal;
4936 adapter->dcb_cfg.pfc_mode_enable = false;
4937 adapter->dcb_cfg.round_robin_enable = false;
4938 adapter->dcb_set_bitmap = 0x00;
4939 ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
4940 adapter->ring_feature[RING_F_DCB].indices);
4944 /* default flow control settings */
4945 hw->fc.requested_mode = ixgbe_fc_full;
4946 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
4948 adapter->last_lfc_mode = hw->fc.current_mode;
4950 hw->fc.high_water = FC_HIGH_WATER(max_frame);
4951 hw->fc.low_water = FC_LOW_WATER(max_frame);
4952 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
4953 hw->fc.send_xon = true;
4954 hw->fc.disable_fc_autoneg = false;
4956 /* enable itr by default in dynamic mode */
4957 adapter->rx_itr_setting = 1;
4958 adapter->rx_eitr_param = 20000;
4959 adapter->tx_itr_setting = 1;
4960 adapter->tx_eitr_param = 10000;
4962 /* set defaults for eitr in MegaBytes */
4963 adapter->eitr_low = 10;
4964 adapter->eitr_high = 20;
4966 /* set default ring sizes */
4967 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
4968 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
4970 /* initialize eeprom parameters */
4971 if (ixgbe_init_eeprom_params_generic(hw)) {
4972 e_dev_err("EEPROM initialization failed\n");
4976 /* enable rx csum by default */
4977 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
4979 /* get assigned NUMA node */
4980 adapter->node = dev_to_node(&pdev->dev);
4982 set_bit(__IXGBE_DOWN, &adapter->state);
4988 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
4989 * @tx_ring: tx descriptor ring (for a specific queue) to setup
4991 * Return 0 on success, negative on failure
4993 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
4995 struct device *dev = tx_ring->dev;
4998 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4999 tx_ring->tx_buffer_info = vmalloc_node(size, tx_ring->numa_node);
5000 if (!tx_ring->tx_buffer_info)
5001 tx_ring->tx_buffer_info = vmalloc(size);
5002 if (!tx_ring->tx_buffer_info)
5004 memset(tx_ring->tx_buffer_info, 0, size);
5006 /* round up to nearest 4K */
5007 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
5008 tx_ring->size = ALIGN(tx_ring->size, 4096);
5010 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
5011 &tx_ring->dma, GFP_KERNEL);
5015 tx_ring->next_to_use = 0;
5016 tx_ring->next_to_clean = 0;
5017 tx_ring->work_limit = tx_ring->count;
5021 vfree(tx_ring->tx_buffer_info);
5022 tx_ring->tx_buffer_info = NULL;
5023 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
5028 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
5029 * @adapter: board private structure
5031 * If this function returns with an error, then it's possible one or
5032 * more of the rings is populated (while the rest are not). It is the
5033 * callers duty to clean those orphaned rings.
5035 * Return 0 on success, negative on failure
5037 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
5041 for (i = 0; i < adapter->num_tx_queues; i++) {
5042 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
5045 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
5053 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
5054 * @rx_ring: rx descriptor ring (for a specific queue) to setup
5056 * Returns 0 on success, negative on failure
5058 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
5060 struct device *dev = rx_ring->dev;
5063 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
5064 rx_ring->rx_buffer_info = vmalloc_node(size, rx_ring->numa_node);
5065 if (!rx_ring->rx_buffer_info)
5066 rx_ring->rx_buffer_info = vmalloc(size);
5067 if (!rx_ring->rx_buffer_info)
5069 memset(rx_ring->rx_buffer_info, 0, size);
5071 /* Round up to nearest 4K */
5072 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
5073 rx_ring->size = ALIGN(rx_ring->size, 4096);
5075 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
5076 &rx_ring->dma, GFP_KERNEL);
5081 rx_ring->next_to_clean = 0;
5082 rx_ring->next_to_use = 0;
5086 vfree(rx_ring->rx_buffer_info);
5087 rx_ring->rx_buffer_info = NULL;
5088 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
5093 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
5094 * @adapter: board private structure
5096 * If this function returns with an error, then it's possible one or
5097 * more of the rings is populated (while the rest are not). It is the
5098 * callers duty to clean those orphaned rings.
5100 * Return 0 on success, negative on failure
5102 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
5106 for (i = 0; i < adapter->num_rx_queues; i++) {
5107 err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
5110 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
5118 * ixgbe_free_tx_resources - Free Tx Resources per Queue
5119 * @tx_ring: Tx descriptor ring for a specific queue
5121 * Free all transmit software resources
5123 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
5125 ixgbe_clean_tx_ring(tx_ring);
5127 vfree(tx_ring->tx_buffer_info);
5128 tx_ring->tx_buffer_info = NULL;
5130 /* if not set, then don't free */
5134 dma_free_coherent(tx_ring->dev, tx_ring->size,
5135 tx_ring->desc, tx_ring->dma);
5137 tx_ring->desc = NULL;
5141 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5142 * @adapter: board private structure
5144 * Free all transmit software resources
5146 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
5150 for (i = 0; i < adapter->num_tx_queues; i++)
5151 if (adapter->tx_ring[i]->desc)
5152 ixgbe_free_tx_resources(adapter->tx_ring[i]);
5156 * ixgbe_free_rx_resources - Free Rx Resources
5157 * @rx_ring: ring to clean the resources from
5159 * Free all receive software resources
5161 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
5163 ixgbe_clean_rx_ring(rx_ring);
5165 vfree(rx_ring->rx_buffer_info);
5166 rx_ring->rx_buffer_info = NULL;
5168 /* if not set, then don't free */
5172 dma_free_coherent(rx_ring->dev, rx_ring->size,
5173 rx_ring->desc, rx_ring->dma);
5175 rx_ring->desc = NULL;
5179 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5180 * @adapter: board private structure
5182 * Free all receive software resources
5184 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5188 for (i = 0; i < adapter->num_rx_queues; i++)
5189 if (adapter->rx_ring[i]->desc)
5190 ixgbe_free_rx_resources(adapter->rx_ring[i]);
5194 * ixgbe_change_mtu - Change the Maximum Transfer Unit
5195 * @netdev: network interface device structure
5196 * @new_mtu: new value for maximum frame size
5198 * Returns 0 on success, negative on failure
5200 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5202 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5203 struct ixgbe_hw *hw = &adapter->hw;
5204 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5206 /* MTU < 68 is an error and causes problems on some kernels */
5207 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5210 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5211 /* must set new MTU before calling down or up */
5212 netdev->mtu = new_mtu;
5214 hw->fc.high_water = FC_HIGH_WATER(max_frame);
5215 hw->fc.low_water = FC_LOW_WATER(max_frame);
5217 if (netif_running(netdev))
5218 ixgbe_reinit_locked(adapter);
5224 * ixgbe_open - Called when a network interface is made active
5225 * @netdev: network interface device structure
5227 * Returns 0 on success, negative value on failure
5229 * The open entry point is called when a network interface is made
5230 * active by the system (IFF_UP). At this point all resources needed
5231 * for transmit and receive operations are allocated, the interrupt
5232 * handler is registered with the OS, the watchdog timer is started,
5233 * and the stack is notified that the interface is ready.
5235 static int ixgbe_open(struct net_device *netdev)
5237 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5240 /* disallow open during test */
5241 if (test_bit(__IXGBE_TESTING, &adapter->state))
5244 netif_carrier_off(netdev);
5246 /* allocate transmit descriptors */
5247 err = ixgbe_setup_all_tx_resources(adapter);
5251 /* allocate receive descriptors */
5252 err = ixgbe_setup_all_rx_resources(adapter);
5256 ixgbe_configure(adapter);
5258 err = ixgbe_request_irq(adapter);
5262 err = ixgbe_up_complete(adapter);
5266 netif_tx_start_all_queues(netdev);
5271 ixgbe_release_hw_control(adapter);
5272 ixgbe_free_irq(adapter);
5275 ixgbe_free_all_rx_resources(adapter);
5277 ixgbe_free_all_tx_resources(adapter);
5278 ixgbe_reset(adapter);
5284 * ixgbe_close - Disables a network interface
5285 * @netdev: network interface device structure
5287 * Returns 0, this is not allowed to fail
5289 * The close entry point is called when an interface is de-activated
5290 * by the OS. The hardware is still under the drivers control, but
5291 * needs to be disabled. A global MAC reset is issued to stop the
5292 * hardware, and all transmit and receive resources are freed.
5294 static int ixgbe_close(struct net_device *netdev)
5296 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5298 ixgbe_down(adapter);
5299 ixgbe_free_irq(adapter);
5301 ixgbe_free_all_tx_resources(adapter);
5302 ixgbe_free_all_rx_resources(adapter);
5304 ixgbe_release_hw_control(adapter);
5310 static int ixgbe_resume(struct pci_dev *pdev)
5312 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5313 struct net_device *netdev = adapter->netdev;
5316 pci_set_power_state(pdev, PCI_D0);
5317 pci_restore_state(pdev);
5319 * pci_restore_state clears dev->state_saved so call
5320 * pci_save_state to restore it.
5322 pci_save_state(pdev);
5324 err = pci_enable_device_mem(pdev);
5326 e_dev_err("Cannot enable PCI device from suspend\n");
5329 pci_set_master(pdev);
5331 pci_wake_from_d3(pdev, false);
5333 err = ixgbe_init_interrupt_scheme(adapter);
5335 e_dev_err("Cannot initialize interrupts for device\n");
5339 ixgbe_reset(adapter);
5341 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5343 if (netif_running(netdev)) {
5344 err = ixgbe_open(netdev);
5349 netif_device_attach(netdev);
5353 #endif /* CONFIG_PM */
5355 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5357 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5358 struct net_device *netdev = adapter->netdev;
5359 struct ixgbe_hw *hw = &adapter->hw;
5361 u32 wufc = adapter->wol;
5366 netif_device_detach(netdev);
5368 if (netif_running(netdev)) {
5369 ixgbe_down(adapter);
5370 ixgbe_free_irq(adapter);
5371 ixgbe_free_all_tx_resources(adapter);
5372 ixgbe_free_all_rx_resources(adapter);
5375 ixgbe_clear_interrupt_scheme(adapter);
5378 retval = pci_save_state(pdev);
5384 ixgbe_set_rx_mode(netdev);
5386 /* turn on all-multi mode if wake on multicast is enabled */
5387 if (wufc & IXGBE_WUFC_MC) {
5388 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5389 fctrl |= IXGBE_FCTRL_MPE;
5390 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5393 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5394 ctrl |= IXGBE_CTRL_GIO_DIS;
5395 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5397 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5399 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5400 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5403 if (wufc && hw->mac.type == ixgbe_mac_82599EB)
5404 pci_wake_from_d3(pdev, true);
5406 pci_wake_from_d3(pdev, false);
5408 *enable_wake = !!wufc;
5410 ixgbe_release_hw_control(adapter);
5412 pci_disable_device(pdev);
5418 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5423 retval = __ixgbe_shutdown(pdev, &wake);
5428 pci_prepare_to_sleep(pdev);
5430 pci_wake_from_d3(pdev, false);
5431 pci_set_power_state(pdev, PCI_D3hot);
5436 #endif /* CONFIG_PM */
5438 static void ixgbe_shutdown(struct pci_dev *pdev)
5442 __ixgbe_shutdown(pdev, &wake);
5444 if (system_state == SYSTEM_POWER_OFF) {
5445 pci_wake_from_d3(pdev, wake);
5446 pci_set_power_state(pdev, PCI_D3hot);
5451 * ixgbe_update_stats - Update the board statistics counters.
5452 * @adapter: board private structure
5454 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5456 struct net_device *netdev = adapter->netdev;
5457 struct ixgbe_hw *hw = &adapter->hw;
5458 struct ixgbe_hw_stats *hwstats = &adapter->stats;
5460 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5461 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
5462 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
5463 u64 bytes = 0, packets = 0;
5465 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5466 test_bit(__IXGBE_RESETTING, &adapter->state))
5469 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5472 for (i = 0; i < 16; i++)
5473 adapter->hw_rx_no_dma_resources +=
5474 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5475 for (i = 0; i < adapter->num_rx_queues; i++) {
5476 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
5477 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
5479 adapter->rsc_total_count = rsc_count;
5480 adapter->rsc_total_flush = rsc_flush;
5483 for (i = 0; i < adapter->num_rx_queues; i++) {
5484 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
5485 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
5486 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
5487 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
5488 bytes += rx_ring->stats.bytes;
5489 packets += rx_ring->stats.packets;
5491 adapter->non_eop_descs = non_eop_descs;
5492 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5493 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5494 netdev->stats.rx_bytes = bytes;
5495 netdev->stats.rx_packets = packets;
5499 /* gather some stats to the adapter struct that are per queue */
5500 for (i = 0; i < adapter->num_tx_queues; i++) {
5501 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5502 restart_queue += tx_ring->tx_stats.restart_queue;
5503 tx_busy += tx_ring->tx_stats.tx_busy;
5504 bytes += tx_ring->stats.bytes;
5505 packets += tx_ring->stats.packets;
5507 adapter->restart_queue = restart_queue;
5508 adapter->tx_busy = tx_busy;
5509 netdev->stats.tx_bytes = bytes;
5510 netdev->stats.tx_packets = packets;
5512 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5513 for (i = 0; i < 8; i++) {
5514 /* for packet buffers not used, the register should read 0 */
5515 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5517 hwstats->mpc[i] += mpc;
5518 total_mpc += hwstats->mpc[i];
5519 if (hw->mac.type == ixgbe_mac_82598EB)
5520 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5521 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5522 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5523 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5524 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5525 if (hw->mac.type == ixgbe_mac_82599EB) {
5526 hwstats->pxonrxc[i] +=
5527 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5528 hwstats->pxoffrxc[i] +=
5529 IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
5530 hwstats->qprdc[i] += IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5532 hwstats->pxonrxc[i] +=
5533 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5534 hwstats->pxoffrxc[i] +=
5535 IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
5537 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5538 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5540 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5541 /* work around hardware counting issue */
5542 hwstats->gprc -= missed_rx;
5544 /* 82598 hardware only has a 32 bit counter in the high register */
5545 if (hw->mac.type == ixgbe_mac_82599EB) {
5547 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5548 tmp = IXGBE_READ_REG(hw, IXGBE_GORCH) & 0xF;
5549 /* 4 high bits of GORC */
5550 hwstats->gorc += (tmp << 32);
5551 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5552 tmp = IXGBE_READ_REG(hw, IXGBE_GOTCH) & 0xF;
5553 /* 4 high bits of GOTC */
5554 hwstats->gotc += (tmp << 32);
5555 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5556 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5557 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5558 hwstats->lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
5559 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5560 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5562 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5563 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5564 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5565 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5566 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5567 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5568 #endif /* IXGBE_FCOE */
5570 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5571 hwstats->lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
5572 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5573 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5574 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5576 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5577 hwstats->bprc += bprc;
5578 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5579 if (hw->mac.type == ixgbe_mac_82598EB)
5580 hwstats->mprc -= bprc;
5581 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5582 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5583 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5584 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5585 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5586 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5587 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5588 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5589 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5590 hwstats->lxontxc += lxon;
5591 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5592 hwstats->lxofftxc += lxoff;
5593 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5594 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5595 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5597 * 82598 errata - tx of flow control packets is included in tx counters
5599 xon_off_tot = lxon + lxoff;
5600 hwstats->gptc -= xon_off_tot;
5601 hwstats->mptc -= xon_off_tot;
5602 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5603 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5604 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5605 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5606 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5607 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5608 hwstats->ptc64 -= xon_off_tot;
5609 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5610 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5611 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5612 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5613 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5614 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5616 /* Fill out the OS statistics structure */
5617 netdev->stats.multicast = hwstats->mprc;
5620 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
5621 netdev->stats.rx_dropped = 0;
5622 netdev->stats.rx_length_errors = hwstats->rlec;
5623 netdev->stats.rx_crc_errors = hwstats->crcerrs;
5624 netdev->stats.rx_missed_errors = total_mpc;
5628 * ixgbe_watchdog - Timer Call-back
5629 * @data: pointer to adapter cast into an unsigned long
5631 static void ixgbe_watchdog(unsigned long data)
5633 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5634 struct ixgbe_hw *hw = &adapter->hw;
5639 * Do the watchdog outside of interrupt context due to the lovely
5640 * delays that some of the newer hardware requires
5643 if (test_bit(__IXGBE_DOWN, &adapter->state))
5644 goto watchdog_short_circuit;
5646 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5648 * for legacy and MSI interrupts don't set any bits
5649 * that are enabled for EIAM, because this operation
5650 * would set *both* EIMS and EICS for any bit in EIAM
5652 IXGBE_WRITE_REG(hw, IXGBE_EICS,
5653 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5654 goto watchdog_reschedule;
5657 /* get one bit for every active tx/rx interrupt vector */
5658 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
5659 struct ixgbe_q_vector *qv = adapter->q_vector[i];
5660 if (qv->rxr_count || qv->txr_count)
5661 eics |= ((u64)1 << i);
5664 /* Cause software interrupt to ensure rx rings are cleaned */
5665 ixgbe_irq_rearm_queues(adapter, eics);
5667 watchdog_reschedule:
5668 /* Reset the timer */
5669 mod_timer(&adapter->watchdog_timer, round_jiffies(jiffies + 2 * HZ));
5671 watchdog_short_circuit:
5672 schedule_work(&adapter->watchdog_task);
5676 * ixgbe_multispeed_fiber_task - worker thread to configure multispeed fiber
5677 * @work: pointer to work_struct containing our data
5679 static void ixgbe_multispeed_fiber_task(struct work_struct *work)
5681 struct ixgbe_adapter *adapter = container_of(work,
5682 struct ixgbe_adapter,
5683 multispeed_fiber_task);
5684 struct ixgbe_hw *hw = &adapter->hw;
5688 adapter->flags |= IXGBE_FLAG_IN_SFP_LINK_TASK;
5689 autoneg = hw->phy.autoneg_advertised;
5690 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
5691 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
5692 hw->mac.autotry_restart = false;
5693 if (hw->mac.ops.setup_link)
5694 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
5695 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5696 adapter->flags &= ~IXGBE_FLAG_IN_SFP_LINK_TASK;
5700 * ixgbe_sfp_config_module_task - worker thread to configure a new SFP+ module
5701 * @work: pointer to work_struct containing our data
5703 static void ixgbe_sfp_config_module_task(struct work_struct *work)
5705 struct ixgbe_adapter *adapter = container_of(work,
5706 struct ixgbe_adapter,
5707 sfp_config_module_task);
5708 struct ixgbe_hw *hw = &adapter->hw;
5711 adapter->flags |= IXGBE_FLAG_IN_SFP_MOD_TASK;
5713 /* Time for electrical oscillations to settle down */
5715 err = hw->phy.ops.identify_sfp(hw);
5717 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
5718 e_dev_err("failed to initialize because an unsupported SFP+ "
5719 "module type was detected.\n");
5720 e_dev_err("Reload the driver after installing a supported "
5722 unregister_netdev(adapter->netdev);
5725 hw->mac.ops.setup_sfp(hw);
5727 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
5728 /* This will also work for DA Twinax connections */
5729 schedule_work(&adapter->multispeed_fiber_task);
5730 adapter->flags &= ~IXGBE_FLAG_IN_SFP_MOD_TASK;
5734 * ixgbe_fdir_reinit_task - worker thread to reinit FDIR filter table
5735 * @work: pointer to work_struct containing our data
5737 static void ixgbe_fdir_reinit_task(struct work_struct *work)
5739 struct ixgbe_adapter *adapter = container_of(work,
5740 struct ixgbe_adapter,
5742 struct ixgbe_hw *hw = &adapter->hw;
5745 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5746 for (i = 0; i < adapter->num_tx_queues; i++)
5747 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
5748 &(adapter->tx_ring[i]->state));
5750 e_err(probe, "failed to finish FDIR re-initialization, "
5751 "ignored adding FDIR ATR filters\n");
5753 /* Done FDIR Re-initialization, enable transmits */
5754 netif_tx_start_all_queues(adapter->netdev);
5757 static DEFINE_MUTEX(ixgbe_watchdog_lock);
5760 * ixgbe_watchdog_task - worker thread to bring link up
5761 * @work: pointer to work_struct containing our data
5763 static void ixgbe_watchdog_task(struct work_struct *work)
5765 struct ixgbe_adapter *adapter = container_of(work,
5766 struct ixgbe_adapter,
5768 struct net_device *netdev = adapter->netdev;
5769 struct ixgbe_hw *hw = &adapter->hw;
5773 struct ixgbe_ring *tx_ring;
5774 int some_tx_pending = 0;
5776 mutex_lock(&ixgbe_watchdog_lock);
5778 link_up = adapter->link_up;
5779 link_speed = adapter->link_speed;
5781 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
5782 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5785 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5786 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
5787 hw->mac.ops.fc_enable(hw, i);
5789 hw->mac.ops.fc_enable(hw, 0);
5792 hw->mac.ops.fc_enable(hw, 0);
5797 time_after(jiffies, (adapter->link_check_timeout +
5798 IXGBE_TRY_LINK_TIMEOUT))) {
5799 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5800 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5802 adapter->link_up = link_up;
5803 adapter->link_speed = link_speed;
5807 if (!netif_carrier_ok(netdev)) {
5808 bool flow_rx, flow_tx;
5810 if (hw->mac.type == ixgbe_mac_82599EB) {
5811 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5812 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5813 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5814 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5816 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5817 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5818 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5819 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5822 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
5823 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5825 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5826 "1 Gbps" : "unknown speed")),
5827 ((flow_rx && flow_tx) ? "RX/TX" :
5829 (flow_tx ? "TX" : "None"))));
5831 netif_carrier_on(netdev);
5833 /* Force detection of hung controller */
5834 for (i = 0; i < adapter->num_tx_queues; i++) {
5835 tx_ring = adapter->tx_ring[i];
5836 set_check_for_tx_hang(tx_ring);
5840 adapter->link_up = false;
5841 adapter->link_speed = 0;
5842 if (netif_carrier_ok(netdev)) {
5843 e_info(drv, "NIC Link is Down\n");
5844 netif_carrier_off(netdev);
5848 if (!netif_carrier_ok(netdev)) {
5849 for (i = 0; i < adapter->num_tx_queues; i++) {
5850 tx_ring = adapter->tx_ring[i];
5851 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5852 some_tx_pending = 1;
5857 if (some_tx_pending) {
5858 /* We've lost link, so the controller stops DMA,
5859 * but we've got queued Tx work that's never going
5860 * to get done, so reset controller to flush Tx.
5861 * (Do the reset outside of interrupt context).
5863 schedule_work(&adapter->reset_task);
5867 ixgbe_update_stats(adapter);
5868 mutex_unlock(&ixgbe_watchdog_lock);
5871 static int ixgbe_tso(struct ixgbe_adapter *adapter,
5872 struct ixgbe_ring *tx_ring, struct sk_buff *skb,
5873 u32 tx_flags, u8 *hdr_len, __be16 protocol)
5875 struct ixgbe_adv_tx_context_desc *context_desc;
5878 struct ixgbe_tx_buffer *tx_buffer_info;
5879 u32 vlan_macip_lens = 0, type_tucmd_mlhl;
5880 u32 mss_l4len_idx, l4len;
5882 if (skb_is_gso(skb)) {
5883 if (skb_header_cloned(skb)) {
5884 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5888 l4len = tcp_hdrlen(skb);
5891 if (protocol == htons(ETH_P_IP)) {
5892 struct iphdr *iph = ip_hdr(skb);
5895 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
5899 } else if (skb_is_gso_v6(skb)) {
5900 ipv6_hdr(skb)->payload_len = 0;
5901 tcp_hdr(skb)->check =
5902 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5903 &ipv6_hdr(skb)->daddr,
5907 i = tx_ring->next_to_use;
5909 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5910 context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
5912 /* VLAN MACLEN IPLEN */
5913 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5915 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
5916 vlan_macip_lens |= ((skb_network_offset(skb)) <<
5917 IXGBE_ADVTXD_MACLEN_SHIFT);
5918 *hdr_len += skb_network_offset(skb);
5920 (skb_transport_header(skb) - skb_network_header(skb));
5922 (skb_transport_header(skb) - skb_network_header(skb));
5923 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
5924 context_desc->seqnum_seed = 0;
5926 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
5927 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
5928 IXGBE_ADVTXD_DTYP_CTXT);
5930 if (protocol == htons(ETH_P_IP))
5931 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
5932 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5933 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
5937 (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
5938 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
5939 /* use index 1 for TSO */
5940 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
5941 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
5943 tx_buffer_info->time_stamp = jiffies;
5944 tx_buffer_info->next_to_watch = i;
5947 if (i == tx_ring->count)
5949 tx_ring->next_to_use = i;
5956 static u32 ixgbe_psum(struct ixgbe_adapter *adapter, struct sk_buff *skb,
5962 case cpu_to_be16(ETH_P_IP):
5963 rtn |= IXGBE_ADVTXD_TUCMD_IPV4;
5964 switch (ip_hdr(skb)->protocol) {
5966 rtn |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5969 rtn |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5973 case cpu_to_be16(ETH_P_IPV6):
5974 /* XXX what about other V6 headers?? */
5975 switch (ipv6_hdr(skb)->nexthdr) {
5977 rtn |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5980 rtn |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5985 if (unlikely(net_ratelimit()))
5986 e_warn(probe, "partial checksum but proto=%x!\n",
5994 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
5995 struct ixgbe_ring *tx_ring,
5996 struct sk_buff *skb, u32 tx_flags,
5999 struct ixgbe_adv_tx_context_desc *context_desc;
6001 struct ixgbe_tx_buffer *tx_buffer_info;
6002 u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
6004 if (skb->ip_summed == CHECKSUM_PARTIAL ||
6005 (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
6006 i = tx_ring->next_to_use;
6007 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6008 context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
6010 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
6012 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
6013 vlan_macip_lens |= (skb_network_offset(skb) <<
6014 IXGBE_ADVTXD_MACLEN_SHIFT);
6015 if (skb->ip_summed == CHECKSUM_PARTIAL)
6016 vlan_macip_lens |= (skb_transport_header(skb) -
6017 skb_network_header(skb));
6019 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
6020 context_desc->seqnum_seed = 0;
6022 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
6023 IXGBE_ADVTXD_DTYP_CTXT);
6025 if (skb->ip_summed == CHECKSUM_PARTIAL)
6026 type_tucmd_mlhl |= ixgbe_psum(adapter, skb, protocol);
6028 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
6029 /* use index zero for tx checksum offload */
6030 context_desc->mss_l4len_idx = 0;
6032 tx_buffer_info->time_stamp = jiffies;
6033 tx_buffer_info->next_to_watch = i;
6036 if (i == tx_ring->count)
6038 tx_ring->next_to_use = i;
6046 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
6047 struct ixgbe_ring *tx_ring,
6048 struct sk_buff *skb, u32 tx_flags,
6049 unsigned int first, const u8 hdr_len)
6051 struct device *dev = tx_ring->dev;
6052 struct ixgbe_tx_buffer *tx_buffer_info;
6054 unsigned int total = skb->len;
6055 unsigned int offset = 0, size, count = 0, i;
6056 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
6058 unsigned int bytecount = skb->len;
6061 i = tx_ring->next_to_use;
6063 if (tx_flags & IXGBE_TX_FLAGS_FCOE)
6064 /* excluding fcoe_crc_eof for FCoE */
6065 total -= sizeof(struct fcoe_crc_eof);
6067 len = min(skb_headlen(skb), total);
6069 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6070 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
6072 tx_buffer_info->length = size;
6073 tx_buffer_info->mapped_as_page = false;
6074 tx_buffer_info->dma = dma_map_single(dev,
6076 size, DMA_TO_DEVICE);
6077 if (dma_mapping_error(dev, tx_buffer_info->dma))
6079 tx_buffer_info->time_stamp = jiffies;
6080 tx_buffer_info->next_to_watch = i;
6089 if (i == tx_ring->count)
6094 for (f = 0; f < nr_frags; f++) {
6095 struct skb_frag_struct *frag;
6097 frag = &skb_shinfo(skb)->frags[f];
6098 len = min((unsigned int)frag->size, total);
6099 offset = frag->page_offset;
6103 if (i == tx_ring->count)
6106 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6107 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
6109 tx_buffer_info->length = size;
6110 tx_buffer_info->dma = dma_map_page(dev,
6114 tx_buffer_info->mapped_as_page = true;
6115 if (dma_mapping_error(dev, tx_buffer_info->dma))
6117 tx_buffer_info->time_stamp = jiffies;
6118 tx_buffer_info->next_to_watch = i;
6129 if (tx_flags & IXGBE_TX_FLAGS_TSO)
6130 gso_segs = skb_shinfo(skb)->gso_segs;
6132 /* adjust for FCoE Sequence Offload */
6133 else if (tx_flags & IXGBE_TX_FLAGS_FSO)
6134 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
6135 skb_shinfo(skb)->gso_size);
6136 #endif /* IXGBE_FCOE */
6137 bytecount += (gso_segs - 1) * hdr_len;
6139 /* multiply data chunks by size of headers */
6140 tx_ring->tx_buffer_info[i].bytecount = bytecount;
6141 tx_ring->tx_buffer_info[i].gso_segs = gso_segs;
6142 tx_ring->tx_buffer_info[i].skb = skb;
6143 tx_ring->tx_buffer_info[first].next_to_watch = i;
6148 e_dev_err("TX DMA map failed\n");
6150 /* clear timestamp and dma mappings for failed tx_buffer_info map */
6151 tx_buffer_info->dma = 0;
6152 tx_buffer_info->time_stamp = 0;
6153 tx_buffer_info->next_to_watch = 0;
6157 /* clear timestamp and dma mappings for remaining portion of packet */
6160 i += tx_ring->count;
6162 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6163 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
6169 static void ixgbe_tx_queue(struct ixgbe_ring *tx_ring,
6170 int tx_flags, int count, u32 paylen, u8 hdr_len)
6172 union ixgbe_adv_tx_desc *tx_desc = NULL;
6173 struct ixgbe_tx_buffer *tx_buffer_info;
6174 u32 olinfo_status = 0, cmd_type_len = 0;
6176 u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
6178 cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
6180 cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
6182 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
6183 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
6185 if (tx_flags & IXGBE_TX_FLAGS_TSO) {
6186 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
6188 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
6189 IXGBE_ADVTXD_POPTS_SHIFT;
6191 /* use index 1 context for tso */
6192 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6193 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
6194 olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
6195 IXGBE_ADVTXD_POPTS_SHIFT;
6197 } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
6198 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
6199 IXGBE_ADVTXD_POPTS_SHIFT;
6201 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6202 olinfo_status |= IXGBE_ADVTXD_CC;
6203 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6204 if (tx_flags & IXGBE_TX_FLAGS_FSO)
6205 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
6208 olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
6210 i = tx_ring->next_to_use;
6212 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6213 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
6214 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
6215 tx_desc->read.cmd_type_len =
6216 cpu_to_le32(cmd_type_len | tx_buffer_info->length);
6217 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
6219 if (i == tx_ring->count)
6223 tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
6226 * Force memory writes to complete before letting h/w
6227 * know there are new descriptors to fetch. (Only
6228 * applicable for weak-ordered memory model archs,
6233 tx_ring->next_to_use = i;
6234 writel(i, tx_ring->tail);
6237 static void ixgbe_atr(struct ixgbe_adapter *adapter, struct sk_buff *skb,
6238 u8 queue, u32 tx_flags, __be16 protocol)
6240 struct ixgbe_atr_input atr_input;
6241 struct iphdr *iph = ip_hdr(skb);
6242 struct ethhdr *eth = (struct ethhdr *)skb->data;
6246 /* Right now, we support IPv4 w/ TCP only */
6247 if (protocol != htons(ETH_P_IP) ||
6248 iph->protocol != IPPROTO_TCP)
6251 memset(&atr_input, 0, sizeof(struct ixgbe_atr_input));
6253 vlan_id = (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK) >>
6254 IXGBE_TX_FLAGS_VLAN_SHIFT;
6258 ixgbe_atr_set_vlan_id_82599(&atr_input, vlan_id);
6259 ixgbe_atr_set_src_port_82599(&atr_input, th->dest);
6260 ixgbe_atr_set_dst_port_82599(&atr_input, th->source);
6261 ixgbe_atr_set_flex_byte_82599(&atr_input, eth->h_proto);
6262 ixgbe_atr_set_l4type_82599(&atr_input, IXGBE_ATR_L4TYPE_TCP);
6263 /* src and dst are inverted, think how the receiver sees them */
6264 ixgbe_atr_set_src_ipv4_82599(&atr_input, iph->daddr);
6265 ixgbe_atr_set_dst_ipv4_82599(&atr_input, iph->saddr);
6267 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
6268 ixgbe_fdir_add_signature_filter_82599(&adapter->hw, &atr_input, queue);
6271 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, int size)
6273 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
6274 /* Herbert's original patch had:
6275 * smp_mb__after_netif_stop_queue();
6276 * but since that doesn't exist yet, just open code it. */
6279 /* We need to check again in a case another CPU has just
6280 * made room available. */
6281 if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
6284 /* A reprieve! - use start_queue because it doesn't call schedule */
6285 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
6286 ++tx_ring->tx_stats.restart_queue;
6290 static int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, int size)
6292 if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
6294 return __ixgbe_maybe_stop_tx(tx_ring, size);
6297 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6299 struct ixgbe_adapter *adapter = netdev_priv(dev);
6300 int txq = smp_processor_id();
6304 protocol = vlan_get_protocol(skb);
6306 if ((protocol == htons(ETH_P_FCOE)) ||
6307 (protocol == htons(ETH_P_FIP))) {
6308 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
6309 txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
6310 txq += adapter->ring_feature[RING_F_FCOE].mask;
6312 #ifdef CONFIG_IXGBE_DCB
6313 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6314 txq = adapter->fcoe.up;
6321 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6322 while (unlikely(txq >= dev->real_num_tx_queues))
6323 txq -= dev->real_num_tx_queues;
6327 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6328 if (skb->priority == TC_PRIO_CONTROL)
6329 txq = adapter->ring_feature[RING_F_DCB].indices-1;
6331 txq = (skb->vlan_tci & IXGBE_TX_FLAGS_VLAN_PRIO_MASK)
6336 return skb_tx_hash(dev, skb);
6339 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
6340 struct ixgbe_adapter *adapter,
6341 struct ixgbe_ring *tx_ring)
6343 struct net_device *netdev = tx_ring->netdev;
6344 struct netdev_queue *txq;
6346 unsigned int tx_flags = 0;
6353 protocol = vlan_get_protocol(skb);
6355 if (vlan_tx_tag_present(skb)) {
6356 tx_flags |= vlan_tx_tag_get(skb);
6357 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6358 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6359 tx_flags |= ((skb->queue_mapping & 0x7) << 13);
6361 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6362 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6363 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED &&
6364 skb->priority != TC_PRIO_CONTROL) {
6365 tx_flags |= ((skb->queue_mapping & 0x7) << 13);
6366 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6367 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6371 /* for FCoE with DCB, we force the priority to what
6372 * was specified by the switch */
6373 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED &&
6374 (protocol == htons(ETH_P_FCOE) ||
6375 protocol == htons(ETH_P_FIP))) {
6376 #ifdef CONFIG_IXGBE_DCB
6377 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6378 tx_flags &= ~(IXGBE_TX_FLAGS_VLAN_PRIO_MASK
6379 << IXGBE_TX_FLAGS_VLAN_SHIFT);
6380 tx_flags |= ((adapter->fcoe.up << 13)
6381 << IXGBE_TX_FLAGS_VLAN_SHIFT);
6384 /* flag for FCoE offloads */
6385 if (protocol == htons(ETH_P_FCOE))
6386 tx_flags |= IXGBE_TX_FLAGS_FCOE;
6390 /* four things can cause us to need a context descriptor */
6391 if (skb_is_gso(skb) ||
6392 (skb->ip_summed == CHECKSUM_PARTIAL) ||
6393 (tx_flags & IXGBE_TX_FLAGS_VLAN) ||
6394 (tx_flags & IXGBE_TX_FLAGS_FCOE))
6397 count += TXD_USE_COUNT(skb_headlen(skb));
6398 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6399 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6401 if (ixgbe_maybe_stop_tx(tx_ring, count)) {
6402 tx_ring->tx_stats.tx_busy++;
6403 return NETDEV_TX_BUSY;
6406 first = tx_ring->next_to_use;
6407 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6409 /* setup tx offload for FCoE */
6410 tso = ixgbe_fso(adapter, tx_ring, skb, tx_flags, &hdr_len);
6412 dev_kfree_skb_any(skb);
6413 return NETDEV_TX_OK;
6416 tx_flags |= IXGBE_TX_FLAGS_FSO;
6417 #endif /* IXGBE_FCOE */
6419 if (protocol == htons(ETH_P_IP))
6420 tx_flags |= IXGBE_TX_FLAGS_IPV4;
6421 tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len,
6424 dev_kfree_skb_any(skb);
6425 return NETDEV_TX_OK;
6429 tx_flags |= IXGBE_TX_FLAGS_TSO;
6430 else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags,
6432 (skb->ip_summed == CHECKSUM_PARTIAL))
6433 tx_flags |= IXGBE_TX_FLAGS_CSUM;
6436 count = ixgbe_tx_map(adapter, tx_ring, skb, tx_flags, first, hdr_len);
6438 /* add the ATR filter if ATR is on */
6439 if (tx_ring->atr_sample_rate) {
6440 ++tx_ring->atr_count;
6441 if ((tx_ring->atr_count >= tx_ring->atr_sample_rate) &&
6442 test_bit(__IXGBE_TX_FDIR_INIT_DONE,
6444 ixgbe_atr(adapter, skb, tx_ring->queue_index,
6445 tx_flags, protocol);
6446 tx_ring->atr_count = 0;
6449 txq = netdev_get_tx_queue(netdev, tx_ring->queue_index);
6450 txq->tx_bytes += skb->len;
6452 ixgbe_tx_queue(tx_ring, tx_flags, count, skb->len, hdr_len);
6453 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
6456 dev_kfree_skb_any(skb);
6457 tx_ring->tx_buffer_info[first].time_stamp = 0;
6458 tx_ring->next_to_use = first;
6461 return NETDEV_TX_OK;
6464 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
6466 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6467 struct ixgbe_ring *tx_ring;
6469 tx_ring = adapter->tx_ring[skb->queue_mapping];
6470 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
6474 * ixgbe_set_mac - Change the Ethernet Address of the NIC
6475 * @netdev: network interface device structure
6476 * @p: pointer to an address structure
6478 * Returns 0 on success, negative on failure
6480 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6482 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6483 struct ixgbe_hw *hw = &adapter->hw;
6484 struct sockaddr *addr = p;
6486 if (!is_valid_ether_addr(addr->sa_data))
6487 return -EADDRNOTAVAIL;
6489 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6490 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6492 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
6499 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6501 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6502 struct ixgbe_hw *hw = &adapter->hw;
6506 if (prtad != hw->phy.mdio.prtad)
6508 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6514 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6515 u16 addr, u16 value)
6517 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6518 struct ixgbe_hw *hw = &adapter->hw;
6520 if (prtad != hw->phy.mdio.prtad)
6522 return hw->phy.ops.write_reg(hw, addr, devad, value);
6525 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6527 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6529 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6533 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6535 * @netdev: network interface device structure
6537 * Returns non-zero on failure
6539 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6542 struct ixgbe_adapter *adapter = netdev_priv(dev);
6543 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6545 if (is_valid_ether_addr(mac->san_addr)) {
6547 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6554 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6556 * @netdev: network interface device structure
6558 * Returns non-zero on failure
6560 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6563 struct ixgbe_adapter *adapter = netdev_priv(dev);
6564 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6566 if (is_valid_ether_addr(mac->san_addr)) {
6568 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6574 #ifdef CONFIG_NET_POLL_CONTROLLER
6576 * Polling 'interrupt' - used by things like netconsole to send skbs
6577 * without having to re-enable interrupts. It's not called while
6578 * the interrupt routine is executing.
6580 static void ixgbe_netpoll(struct net_device *netdev)
6582 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6585 /* if interface is down do nothing */
6586 if (test_bit(__IXGBE_DOWN, &adapter->state))
6589 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
6590 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
6591 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
6592 for (i = 0; i < num_q_vectors; i++) {
6593 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
6594 ixgbe_msix_clean_many(0, q_vector);
6597 ixgbe_intr(adapter->pdev->irq, netdev);
6599 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
6603 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
6604 struct rtnl_link_stats64 *stats)
6606 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6609 /* accurate rx/tx bytes/packets stats */
6610 dev_txq_stats_fold(netdev, stats);
6612 for (i = 0; i < adapter->num_rx_queues; i++) {
6613 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
6619 start = u64_stats_fetch_begin_bh(&ring->syncp);
6620 packets = ring->stats.packets;
6621 bytes = ring->stats.bytes;
6622 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6623 stats->rx_packets += packets;
6624 stats->rx_bytes += bytes;
6628 /* following stats updated by ixgbe_watchdog_task() */
6629 stats->multicast = netdev->stats.multicast;
6630 stats->rx_errors = netdev->stats.rx_errors;
6631 stats->rx_length_errors = netdev->stats.rx_length_errors;
6632 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
6633 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
6638 static const struct net_device_ops ixgbe_netdev_ops = {
6639 .ndo_open = ixgbe_open,
6640 .ndo_stop = ixgbe_close,
6641 .ndo_start_xmit = ixgbe_xmit_frame,
6642 .ndo_select_queue = ixgbe_select_queue,
6643 .ndo_set_rx_mode = ixgbe_set_rx_mode,
6644 .ndo_set_multicast_list = ixgbe_set_rx_mode,
6645 .ndo_validate_addr = eth_validate_addr,
6646 .ndo_set_mac_address = ixgbe_set_mac,
6647 .ndo_change_mtu = ixgbe_change_mtu,
6648 .ndo_tx_timeout = ixgbe_tx_timeout,
6649 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
6650 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
6651 .ndo_do_ioctl = ixgbe_ioctl,
6652 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
6653 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
6654 .ndo_set_vf_tx_rate = ixgbe_ndo_set_vf_bw,
6655 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
6656 .ndo_get_stats64 = ixgbe_get_stats64,
6657 #ifdef CONFIG_NET_POLL_CONTROLLER
6658 .ndo_poll_controller = ixgbe_netpoll,
6661 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
6662 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
6663 .ndo_fcoe_enable = ixgbe_fcoe_enable,
6664 .ndo_fcoe_disable = ixgbe_fcoe_disable,
6665 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
6666 #endif /* IXGBE_FCOE */
6669 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
6670 const struct ixgbe_info *ii)
6672 #ifdef CONFIG_PCI_IOV
6673 struct ixgbe_hw *hw = &adapter->hw;
6676 if (hw->mac.type != ixgbe_mac_82599EB || !max_vfs)
6679 /* The 82599 supports up to 64 VFs per physical function
6680 * but this implementation limits allocation to 63 so that
6681 * basic networking resources are still available to the
6684 adapter->num_vfs = (max_vfs > 63) ? 63 : max_vfs;
6685 adapter->flags |= IXGBE_FLAG_SRIOV_ENABLED;
6686 err = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
6688 e_err(probe, "Failed to enable PCI sriov: %d\n", err);
6691 /* If call to enable VFs succeeded then allocate memory
6692 * for per VF control structures.
6695 kcalloc(adapter->num_vfs,
6696 sizeof(struct vf_data_storage), GFP_KERNEL);
6697 if (adapter->vfinfo) {
6698 /* Now that we're sure SR-IOV is enabled
6699 * and memory allocated set up the mailbox parameters
6701 ixgbe_init_mbx_params_pf(hw);
6702 memcpy(&hw->mbx.ops, ii->mbx_ops,
6703 sizeof(hw->mbx.ops));
6705 /* Disable RSC when in SR-IOV mode */
6706 adapter->flags2 &= ~(IXGBE_FLAG2_RSC_CAPABLE |
6707 IXGBE_FLAG2_RSC_ENABLED);
6712 e_err(probe, "Unable to allocate memory for VF Data Storage - "
6713 "SRIOV disabled\n");
6714 pci_disable_sriov(adapter->pdev);
6717 adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
6718 adapter->num_vfs = 0;
6719 #endif /* CONFIG_PCI_IOV */
6723 * ixgbe_probe - Device Initialization Routine
6724 * @pdev: PCI device information struct
6725 * @ent: entry in ixgbe_pci_tbl
6727 * Returns 0 on success, negative on failure
6729 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
6730 * The OS initialization, configuring of the adapter private structure,
6731 * and a hardware reset occur.
6733 static int __devinit ixgbe_probe(struct pci_dev *pdev,
6734 const struct pci_device_id *ent)
6736 struct net_device *netdev;
6737 struct ixgbe_adapter *adapter = NULL;
6738 struct ixgbe_hw *hw;
6739 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
6740 static int cards_found;
6741 int i, err, pci_using_dac;
6742 unsigned int indices = num_possible_cpus();
6748 /* Catch broken hardware that put the wrong VF device ID in
6749 * the PCIe SR-IOV capability.
6751 if (pdev->is_virtfn) {
6752 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
6753 pci_name(pdev), pdev->vendor, pdev->device);
6757 err = pci_enable_device_mem(pdev);
6761 if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
6762 !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
6765 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
6767 err = dma_set_coherent_mask(&pdev->dev,
6771 "No usable DMA configuration, aborting\n");
6778 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
6779 IORESOURCE_MEM), ixgbe_driver_name);
6782 "pci_request_selected_regions failed 0x%x\n", err);
6786 pci_enable_pcie_error_reporting(pdev);
6788 pci_set_master(pdev);
6789 pci_save_state(pdev);
6791 if (ii->mac == ixgbe_mac_82598EB)
6792 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
6794 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
6796 indices = max_t(unsigned int, indices, IXGBE_MAX_DCB_INDICES);
6798 indices += min_t(unsigned int, num_possible_cpus(),
6799 IXGBE_MAX_FCOE_INDICES);
6801 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
6804 goto err_alloc_etherdev;
6807 SET_NETDEV_DEV(netdev, &pdev->dev);
6809 adapter = netdev_priv(netdev);
6810 pci_set_drvdata(pdev, adapter);
6812 adapter->netdev = netdev;
6813 adapter->pdev = pdev;
6816 adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
6818 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
6819 pci_resource_len(pdev, 0));
6825 for (i = 1; i <= 5; i++) {
6826 if (pci_resource_len(pdev, i) == 0)
6830 netdev->netdev_ops = &ixgbe_netdev_ops;
6831 ixgbe_set_ethtool_ops(netdev);
6832 netdev->watchdog_timeo = 5 * HZ;
6833 strcpy(netdev->name, pci_name(pdev));
6835 adapter->bd_number = cards_found;
6838 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
6839 hw->mac.type = ii->mac;
6842 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
6843 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
6844 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
6845 if (!(eec & (1 << 8)))
6846 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
6849 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
6850 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
6851 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
6852 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
6853 hw->phy.mdio.mmds = 0;
6854 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
6855 hw->phy.mdio.dev = netdev;
6856 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
6857 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
6859 /* set up this timer and work struct before calling get_invariants
6860 * which might start the timer
6862 init_timer(&adapter->sfp_timer);
6863 adapter->sfp_timer.function = ixgbe_sfp_timer;
6864 adapter->sfp_timer.data = (unsigned long) adapter;
6866 INIT_WORK(&adapter->sfp_task, ixgbe_sfp_task);
6868 /* multispeed fiber has its own tasklet, called from GPI SDP1 context */
6869 INIT_WORK(&adapter->multispeed_fiber_task, ixgbe_multispeed_fiber_task);
6871 /* a new SFP+ module arrival, called from GPI SDP2 context */
6872 INIT_WORK(&adapter->sfp_config_module_task,
6873 ixgbe_sfp_config_module_task);
6875 ii->get_invariants(hw);
6877 /* setup the private structure */
6878 err = ixgbe_sw_init(adapter);
6882 /* Make it possible the adapter to be woken up via WOL */
6883 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
6884 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6887 * If there is a fan on this device and it has failed log the
6890 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
6891 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
6892 if (esdp & IXGBE_ESDP_SDP1)
6893 e_crit(probe, "Fan has stopped, replace the adapter\n");
6896 /* reset_hw fills in the perm_addr as well */
6897 hw->phy.reset_if_overtemp = true;
6898 err = hw->mac.ops.reset_hw(hw);
6899 hw->phy.reset_if_overtemp = false;
6900 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
6901 hw->mac.type == ixgbe_mac_82598EB) {
6903 * Start a kernel thread to watch for a module to arrive.
6904 * Only do this for 82598, since 82599 will generate
6905 * interrupts on module arrival.
6907 set_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
6908 mod_timer(&adapter->sfp_timer,
6909 round_jiffies(jiffies + (2 * HZ)));
6911 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
6912 e_dev_err("failed to initialize because an unsupported SFP+ "
6913 "module type was detected.\n");
6914 e_dev_err("Reload the driver after installing a supported "
6918 e_dev_err("HW Init failed: %d\n", err);
6922 ixgbe_probe_vf(adapter, ii);
6924 netdev->features = NETIF_F_SG |
6926 NETIF_F_HW_VLAN_TX |
6927 NETIF_F_HW_VLAN_RX |
6928 NETIF_F_HW_VLAN_FILTER;
6930 netdev->features |= NETIF_F_IPV6_CSUM;
6931 netdev->features |= NETIF_F_TSO;
6932 netdev->features |= NETIF_F_TSO6;
6933 netdev->features |= NETIF_F_GRO;
6935 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
6936 netdev->features |= NETIF_F_SCTP_CSUM;
6938 netdev->vlan_features |= NETIF_F_TSO;
6939 netdev->vlan_features |= NETIF_F_TSO6;
6940 netdev->vlan_features |= NETIF_F_IP_CSUM;
6941 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
6942 netdev->vlan_features |= NETIF_F_SG;
6944 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6945 adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
6946 IXGBE_FLAG_DCB_ENABLED);
6947 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
6948 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
6950 #ifdef CONFIG_IXGBE_DCB
6951 netdev->dcbnl_ops = &dcbnl_ops;
6955 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
6956 if (hw->mac.ops.get_device_caps) {
6957 hw->mac.ops.get_device_caps(hw, &device_caps);
6958 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
6959 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6962 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
6963 netdev->vlan_features |= NETIF_F_FCOE_CRC;
6964 netdev->vlan_features |= NETIF_F_FSO;
6965 netdev->vlan_features |= NETIF_F_FCOE_MTU;
6967 #endif /* IXGBE_FCOE */
6968 if (pci_using_dac) {
6969 netdev->features |= NETIF_F_HIGHDMA;
6970 netdev->vlan_features |= NETIF_F_HIGHDMA;
6973 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
6974 netdev->features |= NETIF_F_LRO;
6976 /* make sure the EEPROM is good */
6977 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
6978 e_dev_err("The EEPROM Checksum Is Not Valid\n");
6983 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
6984 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
6986 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
6987 e_dev_err("invalid MAC address\n");
6992 /* power down the optics */
6993 if (hw->phy.multispeed_fiber)
6994 hw->mac.ops.disable_tx_laser(hw);
6996 init_timer(&adapter->watchdog_timer);
6997 adapter->watchdog_timer.function = ixgbe_watchdog;
6998 adapter->watchdog_timer.data = (unsigned long)adapter;
7000 INIT_WORK(&adapter->reset_task, ixgbe_reset_task);
7001 INIT_WORK(&adapter->watchdog_task, ixgbe_watchdog_task);
7003 err = ixgbe_init_interrupt_scheme(adapter);
7007 switch (pdev->device) {
7008 case IXGBE_DEV_ID_82599_KX4:
7009 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
7010 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
7016 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
7018 /* pick up the PCI bus settings for reporting later */
7019 hw->mac.ops.get_bus_info(hw);
7021 /* print bus type/speed/width info */
7022 e_dev_info("(PCI Express:%s:%s) %pM\n",
7023 (hw->bus.speed == ixgbe_bus_speed_5000 ? "5.0Gb/s" :
7024 hw->bus.speed == ixgbe_bus_speed_2500 ? "2.5Gb/s" :
7026 (hw->bus.width == ixgbe_bus_width_pcie_x8 ? "Width x8" :
7027 hw->bus.width == ixgbe_bus_width_pcie_x4 ? "Width x4" :
7028 hw->bus.width == ixgbe_bus_width_pcie_x1 ? "Width x1" :
7031 ixgbe_read_pba_num_generic(hw, &part_num);
7032 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
7033 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, "
7034 "PBA No: %06x-%03x\n",
7035 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
7036 (part_num >> 8), (part_num & 0xff));
7038 e_dev_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
7039 hw->mac.type, hw->phy.type,
7040 (part_num >> 8), (part_num & 0xff));
7042 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
7043 e_dev_warn("PCI-Express bandwidth available for this card is "
7044 "not sufficient for optimal performance.\n");
7045 e_dev_warn("For optimal performance a x8 PCI-Express slot "
7049 /* save off EEPROM version number */
7050 hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
7052 /* reset the hardware with the new settings */
7053 err = hw->mac.ops.start_hw(hw);
7055 if (err == IXGBE_ERR_EEPROM_VERSION) {
7056 /* We are running on a pre-production device, log a warning */
7057 e_dev_warn("This device is a pre-production adapter/LOM. "
7058 "Please be aware there may be issues associated "
7059 "with your hardware. If you are experiencing "
7060 "problems please contact your Intel or hardware "
7061 "representative who provided you with this "
7064 strcpy(netdev->name, "eth%d");
7065 err = register_netdev(netdev);
7069 /* carrier off reporting is important to ethtool even BEFORE open */
7070 netif_carrier_off(netdev);
7072 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
7073 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
7074 INIT_WORK(&adapter->fdir_reinit_task, ixgbe_fdir_reinit_task);
7076 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
7077 INIT_WORK(&adapter->check_overtemp_task,
7078 ixgbe_check_overtemp_task);
7079 #ifdef CONFIG_IXGBE_DCA
7080 if (dca_add_requester(&pdev->dev) == 0) {
7081 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
7082 ixgbe_setup_dca(adapter);
7085 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7086 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
7087 for (i = 0; i < adapter->num_vfs; i++)
7088 ixgbe_vf_configuration(pdev, (i | 0x10000000));
7091 /* add san mac addr to netdev */
7092 ixgbe_add_sanmac_netdev(netdev);
7094 e_dev_info("Intel(R) 10 Gigabit Network Connection\n");
7099 ixgbe_release_hw_control(adapter);
7100 ixgbe_clear_interrupt_scheme(adapter);
7103 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7104 ixgbe_disable_sriov(adapter);
7105 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
7106 del_timer_sync(&adapter->sfp_timer);
7107 cancel_work_sync(&adapter->sfp_task);
7108 cancel_work_sync(&adapter->multispeed_fiber_task);
7109 cancel_work_sync(&adapter->sfp_config_module_task);
7110 iounmap(hw->hw_addr);
7112 free_netdev(netdev);
7114 pci_release_selected_regions(pdev,
7115 pci_select_bars(pdev, IORESOURCE_MEM));
7118 pci_disable_device(pdev);
7123 * ixgbe_remove - Device Removal Routine
7124 * @pdev: PCI device information struct
7126 * ixgbe_remove is called by the PCI subsystem to alert the driver
7127 * that it should release a PCI device. The could be caused by a
7128 * Hot-Plug event, or because the driver is going to be removed from
7131 static void __devexit ixgbe_remove(struct pci_dev *pdev)
7133 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7134 struct net_device *netdev = adapter->netdev;
7136 set_bit(__IXGBE_DOWN, &adapter->state);
7137 /* clear the module not found bit to make sure the worker won't
7140 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
7141 del_timer_sync(&adapter->watchdog_timer);
7143 del_timer_sync(&adapter->sfp_timer);
7144 cancel_work_sync(&adapter->watchdog_task);
7145 cancel_work_sync(&adapter->sfp_task);
7146 cancel_work_sync(&adapter->multispeed_fiber_task);
7147 cancel_work_sync(&adapter->sfp_config_module_task);
7148 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
7149 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
7150 cancel_work_sync(&adapter->fdir_reinit_task);
7151 flush_scheduled_work();
7153 #ifdef CONFIG_IXGBE_DCA
7154 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
7155 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
7156 dca_remove_requester(&pdev->dev);
7157 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
7162 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
7163 ixgbe_cleanup_fcoe(adapter);
7165 #endif /* IXGBE_FCOE */
7167 /* remove the added san mac */
7168 ixgbe_del_sanmac_netdev(netdev);
7170 if (netdev->reg_state == NETREG_REGISTERED)
7171 unregister_netdev(netdev);
7173 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7174 ixgbe_disable_sriov(adapter);
7176 ixgbe_clear_interrupt_scheme(adapter);
7178 ixgbe_release_hw_control(adapter);
7180 iounmap(adapter->hw.hw_addr);
7181 pci_release_selected_regions(pdev, pci_select_bars(pdev,
7184 e_dev_info("complete\n");
7186 free_netdev(netdev);
7188 pci_disable_pcie_error_reporting(pdev);
7190 pci_disable_device(pdev);
7194 * ixgbe_io_error_detected - called when PCI error is detected
7195 * @pdev: Pointer to PCI device
7196 * @state: The current pci connection state
7198 * This function is called after a PCI bus error affecting
7199 * this device has been detected.
7201 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
7202 pci_channel_state_t state)
7204 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7205 struct net_device *netdev = adapter->netdev;
7207 netif_device_detach(netdev);
7209 if (state == pci_channel_io_perm_failure)
7210 return PCI_ERS_RESULT_DISCONNECT;
7212 if (netif_running(netdev))
7213 ixgbe_down(adapter);
7214 pci_disable_device(pdev);
7216 /* Request a slot reset. */
7217 return PCI_ERS_RESULT_NEED_RESET;
7221 * ixgbe_io_slot_reset - called after the pci bus has been reset.
7222 * @pdev: Pointer to PCI device
7224 * Restart the card from scratch, as if from a cold-boot.
7226 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7228 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7229 pci_ers_result_t result;
7232 if (pci_enable_device_mem(pdev)) {
7233 e_err(probe, "Cannot re-enable PCI device after reset.\n");
7234 result = PCI_ERS_RESULT_DISCONNECT;
7236 pci_set_master(pdev);
7237 pci_restore_state(pdev);
7238 pci_save_state(pdev);
7240 pci_wake_from_d3(pdev, false);
7242 ixgbe_reset(adapter);
7243 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7244 result = PCI_ERS_RESULT_RECOVERED;
7247 err = pci_cleanup_aer_uncorrect_error_status(pdev);
7249 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7250 "failed 0x%0x\n", err);
7251 /* non-fatal, continue */
7258 * ixgbe_io_resume - called when traffic can start flowing again.
7259 * @pdev: Pointer to PCI device
7261 * This callback is called when the error recovery driver tells us that
7262 * its OK to resume normal operation.
7264 static void ixgbe_io_resume(struct pci_dev *pdev)
7266 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7267 struct net_device *netdev = adapter->netdev;
7269 if (netif_running(netdev)) {
7270 if (ixgbe_up(adapter)) {
7271 e_info(probe, "ixgbe_up failed after reset\n");
7276 netif_device_attach(netdev);
7279 static struct pci_error_handlers ixgbe_err_handler = {
7280 .error_detected = ixgbe_io_error_detected,
7281 .slot_reset = ixgbe_io_slot_reset,
7282 .resume = ixgbe_io_resume,
7285 static struct pci_driver ixgbe_driver = {
7286 .name = ixgbe_driver_name,
7287 .id_table = ixgbe_pci_tbl,
7288 .probe = ixgbe_probe,
7289 .remove = __devexit_p(ixgbe_remove),
7291 .suspend = ixgbe_suspend,
7292 .resume = ixgbe_resume,
7294 .shutdown = ixgbe_shutdown,
7295 .err_handler = &ixgbe_err_handler
7299 * ixgbe_init_module - Driver Registration Routine
7301 * ixgbe_init_module is the first routine called when the driver is
7302 * loaded. All it does is register with the PCI subsystem.
7304 static int __init ixgbe_init_module(void)
7307 pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
7308 pr_info("%s\n", ixgbe_copyright);
7310 #ifdef CONFIG_IXGBE_DCA
7311 dca_register_notify(&dca_notifier);
7314 ret = pci_register_driver(&ixgbe_driver);
7318 module_init(ixgbe_init_module);
7321 * ixgbe_exit_module - Driver Exit Cleanup Routine
7323 * ixgbe_exit_module is called just before the driver is removed
7326 static void __exit ixgbe_exit_module(void)
7328 #ifdef CONFIG_IXGBE_DCA
7329 dca_unregister_notify(&dca_notifier);
7331 pci_unregister_driver(&ixgbe_driver);
7332 rcu_barrier(); /* Wait for completion of call_rcu()'s */
7335 #ifdef CONFIG_IXGBE_DCA
7336 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
7341 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
7342 __ixgbe_notify_dca);
7344 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7347 #endif /* CONFIG_IXGBE_DCA */
7350 * ixgbe_get_hw_dev return device
7351 * used by hardware layer to print debugging information
7353 struct net_device *ixgbe_get_hw_dev(struct ixgbe_hw *hw)
7355 struct ixgbe_adapter *adapter = hw->back;
7356 return adapter->netdev;
7359 module_exit(ixgbe_exit_module);