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_hdr_info(union ixgbe_adv_rx_desc *rx_desc)
1133 return rx_desc->wb.lower.lo_dword.hs_rss.hdr_info;
1136 static inline u16 ixgbe_get_pkt_info(union ixgbe_adv_rx_desc *rx_desc)
1138 return rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1141 static inline u32 ixgbe_get_rsc_count(union ixgbe_adv_rx_desc *rx_desc)
1143 return (le32_to_cpu(rx_desc->wb.lower.lo_dword.data) &
1144 IXGBE_RXDADV_RSCCNT_MASK) >>
1145 IXGBE_RXDADV_RSCCNT_SHIFT;
1149 * ixgbe_transform_rsc_queue - change rsc queue into a full packet
1150 * @skb: pointer to the last skb in the rsc queue
1151 * @count: pointer to number of packets coalesced in this context
1153 * This function changes a queue full of hw rsc buffers into a completed
1154 * packet. It uses the ->prev pointers to find the first packet and then
1155 * turns it into the frag list owner.
1157 static inline struct sk_buff *ixgbe_transform_rsc_queue(struct sk_buff *skb,
1160 unsigned int frag_list_size = 0;
1163 struct sk_buff *prev = skb->prev;
1164 frag_list_size += skb->len;
1170 skb_shinfo(skb)->frag_list = skb->next;
1172 skb->len += frag_list_size;
1173 skb->data_len += frag_list_size;
1174 skb->truesize += frag_list_size;
1178 struct ixgbe_rsc_cb {
1183 #define IXGBE_RSC_CB(skb) ((struct ixgbe_rsc_cb *)(skb)->cb)
1185 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1186 struct ixgbe_ring *rx_ring,
1187 int *work_done, int work_to_do)
1189 struct ixgbe_adapter *adapter = q_vector->adapter;
1190 union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
1191 struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
1192 struct sk_buff *skb;
1193 unsigned int i, rsc_count = 0;
1196 bool cleaned = false;
1197 int cleaned_count = 0;
1198 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1201 #endif /* IXGBE_FCOE */
1203 i = rx_ring->next_to_clean;
1204 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1205 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1206 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1208 while (staterr & IXGBE_RXD_STAT_DD) {
1210 if (*work_done >= work_to_do)
1214 rmb(); /* read descriptor and rx_buffer_info after status DD */
1215 if (ring_is_ps_enabled(rx_ring)) {
1216 hdr_info = le16_to_cpu(ixgbe_get_hdr_info(rx_desc));
1217 len = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
1218 IXGBE_RXDADV_HDRBUFLEN_SHIFT;
1219 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1220 if ((len > IXGBE_RX_HDR_SIZE) ||
1221 (upper_len && !(hdr_info & IXGBE_RXDADV_SPH)))
1222 len = IXGBE_RX_HDR_SIZE;
1224 len = le16_to_cpu(rx_desc->wb.upper.length);
1228 skb = rx_buffer_info->skb;
1229 prefetch(skb->data);
1230 rx_buffer_info->skb = NULL;
1232 if (rx_buffer_info->dma) {
1233 if ((adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
1234 (!(staterr & IXGBE_RXD_STAT_EOP)) &&
1237 * When HWRSC is enabled, delay unmapping
1238 * of the first packet. It carries the
1239 * header information, HW may still
1240 * access the header after the writeback.
1241 * Only unmap it when EOP is reached
1243 IXGBE_RSC_CB(skb)->delay_unmap = true;
1244 IXGBE_RSC_CB(skb)->dma = rx_buffer_info->dma;
1246 dma_unmap_single(rx_ring->dev,
1247 rx_buffer_info->dma,
1248 rx_ring->rx_buf_len,
1251 rx_buffer_info->dma = 0;
1256 dma_unmap_page(rx_ring->dev,
1257 rx_buffer_info->page_dma,
1260 rx_buffer_info->page_dma = 0;
1261 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1262 rx_buffer_info->page,
1263 rx_buffer_info->page_offset,
1266 if ((rx_ring->rx_buf_len > (PAGE_SIZE / 2)) ||
1267 (page_count(rx_buffer_info->page) != 1))
1268 rx_buffer_info->page = NULL;
1270 get_page(rx_buffer_info->page);
1272 skb->len += upper_len;
1273 skb->data_len += upper_len;
1274 skb->truesize += upper_len;
1278 if (i == rx_ring->count)
1281 next_rxd = IXGBE_RX_DESC_ADV(rx_ring, i);
1285 if (ring_is_rsc_enabled(rx_ring))
1286 rsc_count = ixgbe_get_rsc_count(rx_desc);
1289 u32 nextp = (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
1290 IXGBE_RXDADV_NEXTP_SHIFT;
1291 next_buffer = &rx_ring->rx_buffer_info[nextp];
1293 next_buffer = &rx_ring->rx_buffer_info[i];
1296 if (staterr & IXGBE_RXD_STAT_EOP) {
1298 skb = ixgbe_transform_rsc_queue(skb,
1299 &(rx_ring->rx_stats.rsc_count));
1300 if (ring_is_rsc_enabled(rx_ring)) {
1301 if (IXGBE_RSC_CB(skb)->delay_unmap) {
1302 dma_unmap_single(rx_ring->dev,
1303 IXGBE_RSC_CB(skb)->dma,
1304 rx_ring->rx_buf_len,
1306 IXGBE_RSC_CB(skb)->dma = 0;
1307 IXGBE_RSC_CB(skb)->delay_unmap = false;
1309 if (ring_is_ps_enabled(rx_ring))
1310 rx_ring->rx_stats.rsc_count +=
1311 skb_shinfo(skb)->nr_frags;
1313 rx_ring->rx_stats.rsc_count++;
1314 rx_ring->rx_stats.rsc_flush++;
1316 u64_stats_update_begin(&rx_ring->syncp);
1317 rx_ring->stats.packets++;
1318 rx_ring->stats.bytes += skb->len;
1319 u64_stats_update_end(&rx_ring->syncp);
1321 if (ring_is_ps_enabled(rx_ring)) {
1322 rx_buffer_info->skb = next_buffer->skb;
1323 rx_buffer_info->dma = next_buffer->dma;
1324 next_buffer->skb = skb;
1325 next_buffer->dma = 0;
1327 skb->next = next_buffer->skb;
1328 skb->next->prev = skb;
1330 rx_ring->rx_stats.non_eop_descs++;
1334 if (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) {
1335 dev_kfree_skb_irq(skb);
1339 ixgbe_rx_checksum(adapter, rx_desc, skb);
1341 /* probably a little skewed due to removing CRC */
1342 total_rx_bytes += skb->len;
1345 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1347 /* if ddp, not passing to ULD unless for FCP_RSP or error */
1348 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
1349 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
1353 #endif /* IXGBE_FCOE */
1354 ixgbe_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
1357 rx_desc->wb.upper.status_error = 0;
1359 /* return some buffers to hardware, one at a time is too slow */
1360 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
1361 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1365 /* use prefetched values */
1367 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1369 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1372 rx_ring->next_to_clean = i;
1373 cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
1376 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1379 /* include DDPed FCoE data */
1380 if (ddp_bytes > 0) {
1383 mss = rx_ring->netdev->mtu - sizeof(struct fcoe_hdr) -
1384 sizeof(struct fc_frame_header) -
1385 sizeof(struct fcoe_crc_eof);
1388 total_rx_bytes += ddp_bytes;
1389 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
1391 #endif /* IXGBE_FCOE */
1393 rx_ring->total_packets += total_rx_packets;
1394 rx_ring->total_bytes += total_rx_bytes;
1399 static int ixgbe_clean_rxonly(struct napi_struct *, int);
1401 * ixgbe_configure_msix - Configure MSI-X hardware
1402 * @adapter: board private structure
1404 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1407 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1409 struct ixgbe_q_vector *q_vector;
1410 int i, j, q_vectors, v_idx, r_idx;
1413 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1416 * Populate the IVAR table and set the ITR values to the
1417 * corresponding register.
1419 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1420 q_vector = adapter->q_vector[v_idx];
1421 /* XXX for_each_set_bit(...) */
1422 r_idx = find_first_bit(q_vector->rxr_idx,
1423 adapter->num_rx_queues);
1425 for (i = 0; i < q_vector->rxr_count; i++) {
1426 j = adapter->rx_ring[r_idx]->reg_idx;
1427 ixgbe_set_ivar(adapter, 0, j, v_idx);
1428 r_idx = find_next_bit(q_vector->rxr_idx,
1429 adapter->num_rx_queues,
1432 r_idx = find_first_bit(q_vector->txr_idx,
1433 adapter->num_tx_queues);
1435 for (i = 0; i < q_vector->txr_count; i++) {
1436 j = adapter->tx_ring[r_idx]->reg_idx;
1437 ixgbe_set_ivar(adapter, 1, j, v_idx);
1438 r_idx = find_next_bit(q_vector->txr_idx,
1439 adapter->num_tx_queues,
1443 if (q_vector->txr_count && !q_vector->rxr_count)
1445 q_vector->eitr = adapter->tx_eitr_param;
1446 else if (q_vector->rxr_count)
1448 q_vector->eitr = adapter->rx_eitr_param;
1450 ixgbe_write_eitr(q_vector);
1451 /* If Flow Director is enabled, set interrupt affinity */
1452 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
1453 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)) {
1455 * Allocate the affinity_hint cpumask, assign the mask
1456 * for this vector, and set our affinity_hint for
1459 if (!alloc_cpumask_var(&q_vector->affinity_mask,
1462 cpumask_set_cpu(v_idx, q_vector->affinity_mask);
1463 irq_set_affinity_hint(adapter->msix_entries[v_idx].vector,
1464 q_vector->affinity_mask);
1468 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
1469 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1471 else if (adapter->hw.mac.type == ixgbe_mac_82599EB)
1472 ixgbe_set_ivar(adapter, -1, 1, v_idx);
1473 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1475 /* set up to autoclear timer, and the vectors */
1476 mask = IXGBE_EIMS_ENABLE_MASK;
1477 if (adapter->num_vfs)
1478 mask &= ~(IXGBE_EIMS_OTHER |
1479 IXGBE_EIMS_MAILBOX |
1482 mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
1483 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1486 enum latency_range {
1490 latency_invalid = 255
1494 * ixgbe_update_itr - update the dynamic ITR value based on statistics
1495 * @adapter: pointer to adapter
1496 * @eitr: eitr setting (ints per sec) to give last timeslice
1497 * @itr_setting: current throttle rate in ints/second
1498 * @packets: the number of packets during this measurement interval
1499 * @bytes: the number of bytes during this measurement interval
1501 * Stores a new ITR value based on packets and byte
1502 * counts during the last interrupt. The advantage of per interrupt
1503 * computation is faster updates and more accurate ITR for the current
1504 * traffic pattern. Constants in this function were computed
1505 * based on theoretical maximum wire speed and thresholds were set based
1506 * on testing data as well as attempting to minimize response time
1507 * while increasing bulk throughput.
1508 * this functionality is controlled by the InterruptThrottleRate module
1509 * parameter (see ixgbe_param.c)
1511 static u8 ixgbe_update_itr(struct ixgbe_adapter *adapter,
1512 u32 eitr, u8 itr_setting,
1513 int packets, int bytes)
1515 unsigned int retval = itr_setting;
1520 goto update_itr_done;
1523 /* simple throttlerate management
1524 * 0-20MB/s lowest (100000 ints/s)
1525 * 20-100MB/s low (20000 ints/s)
1526 * 100-1249MB/s bulk (8000 ints/s)
1528 /* what was last interrupt timeslice? */
1529 timepassed_us = 1000000/eitr;
1530 bytes_perint = bytes / timepassed_us; /* bytes/usec */
1532 switch (itr_setting) {
1533 case lowest_latency:
1534 if (bytes_perint > adapter->eitr_low)
1535 retval = low_latency;
1538 if (bytes_perint > adapter->eitr_high)
1539 retval = bulk_latency;
1540 else if (bytes_perint <= adapter->eitr_low)
1541 retval = lowest_latency;
1544 if (bytes_perint <= adapter->eitr_high)
1545 retval = low_latency;
1554 * ixgbe_write_eitr - write EITR register in hardware specific way
1555 * @q_vector: structure containing interrupt and ring information
1557 * This function is made to be called by ethtool and by the driver
1558 * when it needs to update EITR registers at runtime. Hardware
1559 * specific quirks/differences are taken care of here.
1561 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1563 struct ixgbe_adapter *adapter = q_vector->adapter;
1564 struct ixgbe_hw *hw = &adapter->hw;
1565 int v_idx = q_vector->v_idx;
1566 u32 itr_reg = EITR_INTS_PER_SEC_TO_REG(q_vector->eitr);
1568 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1569 /* must write high and low 16 bits to reset counter */
1570 itr_reg |= (itr_reg << 16);
1571 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1573 * 82599 can support a value of zero, so allow it for
1574 * max interrupt rate, but there is an errata where it can
1575 * not be zero with RSC
1578 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
1582 * set the WDIS bit to not clear the timer bits and cause an
1583 * immediate assertion of the interrupt
1585 itr_reg |= IXGBE_EITR_CNT_WDIS;
1587 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1590 static void ixgbe_set_itr_msix(struct ixgbe_q_vector *q_vector)
1592 struct ixgbe_adapter *adapter = q_vector->adapter;
1594 u8 current_itr, ret_itr;
1596 struct ixgbe_ring *rx_ring, *tx_ring;
1598 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1599 for (i = 0; i < q_vector->txr_count; i++) {
1600 tx_ring = adapter->tx_ring[r_idx];
1601 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1603 tx_ring->total_packets,
1604 tx_ring->total_bytes);
1605 /* if the result for this queue would decrease interrupt
1606 * rate for this vector then use that result */
1607 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
1608 q_vector->tx_itr - 1 : ret_itr);
1609 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1613 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1614 for (i = 0; i < q_vector->rxr_count; i++) {
1615 rx_ring = adapter->rx_ring[r_idx];
1616 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1618 rx_ring->total_packets,
1619 rx_ring->total_bytes);
1620 /* if the result for this queue would decrease interrupt
1621 * rate for this vector then use that result */
1622 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
1623 q_vector->rx_itr - 1 : ret_itr);
1624 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1628 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1630 switch (current_itr) {
1631 /* counts and packets in update_itr are dependent on these numbers */
1632 case lowest_latency:
1636 new_itr = 20000; /* aka hwitr = ~200 */
1644 if (new_itr != q_vector->eitr) {
1645 /* do an exponential smoothing */
1646 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1648 /* save the algorithm value here, not the smoothed one */
1649 q_vector->eitr = new_itr;
1651 ixgbe_write_eitr(q_vector);
1656 * ixgbe_check_overtemp_task - worker thread to check over tempurature
1657 * @work: pointer to work_struct containing our data
1659 static void ixgbe_check_overtemp_task(struct work_struct *work)
1661 struct ixgbe_adapter *adapter = container_of(work,
1662 struct ixgbe_adapter,
1663 check_overtemp_task);
1664 struct ixgbe_hw *hw = &adapter->hw;
1665 u32 eicr = adapter->interrupt_event;
1667 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
1670 switch (hw->device_id) {
1671 case IXGBE_DEV_ID_82599_T3_LOM: {
1673 bool link_up = false;
1675 if (hw->mac.ops.check_link)
1676 hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
1678 if (((eicr & IXGBE_EICR_GPI_SDP0) && (!link_up)) ||
1679 (eicr & IXGBE_EICR_LSC))
1680 /* Check if this is due to overtemp */
1681 if (hw->phy.ops.check_overtemp(hw) == IXGBE_ERR_OVERTEMP)
1686 if (!(eicr & IXGBE_EICR_GPI_SDP0))
1691 "Network adapter has been stopped because it has over heated. "
1692 "Restart the computer. If the problem persists, "
1693 "power off the system and replace the adapter\n");
1694 /* write to clear the interrupt */
1695 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP0);
1698 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
1700 struct ixgbe_hw *hw = &adapter->hw;
1702 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
1703 (eicr & IXGBE_EICR_GPI_SDP1)) {
1704 e_crit(probe, "Fan has stopped, replace the adapter\n");
1705 /* write to clear the interrupt */
1706 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1710 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
1712 struct ixgbe_hw *hw = &adapter->hw;
1714 if (eicr & IXGBE_EICR_GPI_SDP2) {
1715 /* Clear the interrupt */
1716 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
1717 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1718 schedule_work(&adapter->sfp_config_module_task);
1721 if (eicr & IXGBE_EICR_GPI_SDP1) {
1722 /* Clear the interrupt */
1723 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1724 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1725 schedule_work(&adapter->multispeed_fiber_task);
1729 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1731 struct ixgbe_hw *hw = &adapter->hw;
1734 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1735 adapter->link_check_timeout = jiffies;
1736 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1737 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1738 IXGBE_WRITE_FLUSH(hw);
1739 schedule_work(&adapter->watchdog_task);
1743 static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
1745 struct net_device *netdev = data;
1746 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1747 struct ixgbe_hw *hw = &adapter->hw;
1751 * Workaround for Silicon errata. Use clear-by-write instead
1752 * of clear-by-read. Reading with EICS will return the
1753 * interrupt causes without clearing, which later be done
1754 * with the write to EICR.
1756 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1757 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1759 if (eicr & IXGBE_EICR_LSC)
1760 ixgbe_check_lsc(adapter);
1762 if (eicr & IXGBE_EICR_MAILBOX)
1763 ixgbe_msg_task(adapter);
1765 if (hw->mac.type == ixgbe_mac_82598EB)
1766 ixgbe_check_fan_failure(adapter, eicr);
1768 if (hw->mac.type == ixgbe_mac_82599EB) {
1769 ixgbe_check_sfp_event(adapter, eicr);
1770 adapter->interrupt_event = eicr;
1771 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
1772 ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)))
1773 schedule_work(&adapter->check_overtemp_task);
1775 /* Handle Flow Director Full threshold interrupt */
1776 if (eicr & IXGBE_EICR_FLOW_DIR) {
1778 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_FLOW_DIR);
1779 /* Disable transmits before FDIR Re-initialization */
1780 netif_tx_stop_all_queues(netdev);
1781 for (i = 0; i < adapter->num_tx_queues; i++) {
1782 struct ixgbe_ring *tx_ring =
1783 adapter->tx_ring[i];
1784 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
1786 schedule_work(&adapter->fdir_reinit_task);
1790 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1791 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_OTHER);
1796 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
1801 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1802 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1803 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1805 mask = (qmask & 0xFFFFFFFF);
1806 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(0), mask);
1807 mask = (qmask >> 32);
1808 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(1), mask);
1810 /* skip the flush */
1813 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
1818 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1819 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1820 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, mask);
1822 mask = (qmask & 0xFFFFFFFF);
1823 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), mask);
1824 mask = (qmask >> 32);
1825 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), mask);
1827 /* skip the flush */
1830 static irqreturn_t ixgbe_msix_clean_tx(int irq, void *data)
1832 struct ixgbe_q_vector *q_vector = data;
1833 struct ixgbe_adapter *adapter = q_vector->adapter;
1834 struct ixgbe_ring *tx_ring;
1837 if (!q_vector->txr_count)
1840 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1841 for (i = 0; i < q_vector->txr_count; i++) {
1842 tx_ring = adapter->tx_ring[r_idx];
1843 tx_ring->total_bytes = 0;
1844 tx_ring->total_packets = 0;
1845 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1849 /* EIAM disabled interrupts (on this vector) for us */
1850 napi_schedule(&q_vector->napi);
1856 * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
1858 * @data: pointer to our q_vector struct for this interrupt vector
1860 static irqreturn_t ixgbe_msix_clean_rx(int irq, void *data)
1862 struct ixgbe_q_vector *q_vector = data;
1863 struct ixgbe_adapter *adapter = q_vector->adapter;
1864 struct ixgbe_ring *rx_ring;
1868 #ifdef CONFIG_IXGBE_DCA
1869 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1870 ixgbe_update_dca(q_vector);
1873 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1874 for (i = 0; i < q_vector->rxr_count; i++) {
1875 rx_ring = adapter->rx_ring[r_idx];
1876 rx_ring->total_bytes = 0;
1877 rx_ring->total_packets = 0;
1878 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1882 if (!q_vector->rxr_count)
1885 /* EIAM disabled interrupts (on this vector) for us */
1886 napi_schedule(&q_vector->napi);
1891 static irqreturn_t ixgbe_msix_clean_many(int irq, void *data)
1893 struct ixgbe_q_vector *q_vector = data;
1894 struct ixgbe_adapter *adapter = q_vector->adapter;
1895 struct ixgbe_ring *ring;
1899 if (!q_vector->txr_count && !q_vector->rxr_count)
1902 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1903 for (i = 0; i < q_vector->txr_count; i++) {
1904 ring = adapter->tx_ring[r_idx];
1905 ring->total_bytes = 0;
1906 ring->total_packets = 0;
1907 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1911 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1912 for (i = 0; i < q_vector->rxr_count; i++) {
1913 ring = adapter->rx_ring[r_idx];
1914 ring->total_bytes = 0;
1915 ring->total_packets = 0;
1916 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1920 /* EIAM disabled interrupts (on this vector) for us */
1921 napi_schedule(&q_vector->napi);
1927 * ixgbe_clean_rxonly - msix (aka one shot) rx clean routine
1928 * @napi: napi struct with our devices info in it
1929 * @budget: amount of work driver is allowed to do this pass, in packets
1931 * This function is optimized for cleaning one queue only on a single
1934 static int ixgbe_clean_rxonly(struct napi_struct *napi, int budget)
1936 struct ixgbe_q_vector *q_vector =
1937 container_of(napi, struct ixgbe_q_vector, napi);
1938 struct ixgbe_adapter *adapter = q_vector->adapter;
1939 struct ixgbe_ring *rx_ring = NULL;
1943 #ifdef CONFIG_IXGBE_DCA
1944 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1945 ixgbe_update_dca(q_vector);
1948 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1949 rx_ring = adapter->rx_ring[r_idx];
1951 ixgbe_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
1953 /* If all Rx work done, exit the polling mode */
1954 if (work_done < budget) {
1955 napi_complete(napi);
1956 if (adapter->rx_itr_setting & 1)
1957 ixgbe_set_itr_msix(q_vector);
1958 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1959 ixgbe_irq_enable_queues(adapter,
1960 ((u64)1 << q_vector->v_idx));
1967 * ixgbe_clean_rxtx_many - msix (aka one shot) rx clean routine
1968 * @napi: napi struct with our devices info in it
1969 * @budget: amount of work driver is allowed to do this pass, in packets
1971 * This function will clean more than one rx queue associated with a
1974 static int ixgbe_clean_rxtx_many(struct napi_struct *napi, int budget)
1976 struct ixgbe_q_vector *q_vector =
1977 container_of(napi, struct ixgbe_q_vector, napi);
1978 struct ixgbe_adapter *adapter = q_vector->adapter;
1979 struct ixgbe_ring *ring = NULL;
1980 int work_done = 0, i;
1982 bool tx_clean_complete = true;
1984 #ifdef CONFIG_IXGBE_DCA
1985 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1986 ixgbe_update_dca(q_vector);
1989 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1990 for (i = 0; i < q_vector->txr_count; i++) {
1991 ring = adapter->tx_ring[r_idx];
1992 tx_clean_complete &= ixgbe_clean_tx_irq(q_vector, ring);
1993 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1997 /* attempt to distribute budget to each queue fairly, but don't allow
1998 * the budget to go below 1 because we'll exit polling */
1999 budget /= (q_vector->rxr_count ?: 1);
2000 budget = max(budget, 1);
2001 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2002 for (i = 0; i < q_vector->rxr_count; i++) {
2003 ring = adapter->rx_ring[r_idx];
2004 ixgbe_clean_rx_irq(q_vector, ring, &work_done, budget);
2005 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
2009 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2010 ring = adapter->rx_ring[r_idx];
2011 /* If all Rx work done, exit the polling mode */
2012 if (work_done < budget) {
2013 napi_complete(napi);
2014 if (adapter->rx_itr_setting & 1)
2015 ixgbe_set_itr_msix(q_vector);
2016 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2017 ixgbe_irq_enable_queues(adapter,
2018 ((u64)1 << q_vector->v_idx));
2026 * ixgbe_clean_txonly - msix (aka one shot) tx clean routine
2027 * @napi: napi struct with our devices info in it
2028 * @budget: amount of work driver is allowed to do this pass, in packets
2030 * This function is optimized for cleaning one queue only on a single
2033 static int ixgbe_clean_txonly(struct napi_struct *napi, int budget)
2035 struct ixgbe_q_vector *q_vector =
2036 container_of(napi, struct ixgbe_q_vector, napi);
2037 struct ixgbe_adapter *adapter = q_vector->adapter;
2038 struct ixgbe_ring *tx_ring = NULL;
2042 #ifdef CONFIG_IXGBE_DCA
2043 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2044 ixgbe_update_dca(q_vector);
2047 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
2048 tx_ring = adapter->tx_ring[r_idx];
2050 if (!ixgbe_clean_tx_irq(q_vector, tx_ring))
2053 /* If all Tx work done, exit the polling mode */
2054 if (work_done < budget) {
2055 napi_complete(napi);
2056 if (adapter->tx_itr_setting & 1)
2057 ixgbe_set_itr_msix(q_vector);
2058 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2059 ixgbe_irq_enable_queues(adapter,
2060 ((u64)1 << q_vector->v_idx));
2066 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
2069 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2071 set_bit(r_idx, q_vector->rxr_idx);
2072 q_vector->rxr_count++;
2075 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
2078 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2080 set_bit(t_idx, q_vector->txr_idx);
2081 q_vector->txr_count++;
2085 * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
2086 * @adapter: board private structure to initialize
2087 * @vectors: allotted vector count for descriptor rings
2089 * This function maps descriptor rings to the queue-specific vectors
2090 * we were allotted through the MSI-X enabling code. Ideally, we'd have
2091 * one vector per ring/queue, but on a constrained vector budget, we
2092 * group the rings as "efficiently" as possible. You would add new
2093 * mapping configurations in here.
2095 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter,
2099 int rxr_idx = 0, txr_idx = 0;
2100 int rxr_remaining = adapter->num_rx_queues;
2101 int txr_remaining = adapter->num_tx_queues;
2106 /* No mapping required if MSI-X is disabled. */
2107 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2111 * The ideal configuration...
2112 * We have enough vectors to map one per queue.
2114 if (vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
2115 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
2116 map_vector_to_rxq(adapter, v_start, rxr_idx);
2118 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
2119 map_vector_to_txq(adapter, v_start, txr_idx);
2125 * If we don't have enough vectors for a 1-to-1
2126 * mapping, we'll have to group them so there are
2127 * multiple queues per vector.
2129 /* Re-adjusting *qpv takes care of the remainder. */
2130 for (i = v_start; i < vectors; i++) {
2131 rqpv = DIV_ROUND_UP(rxr_remaining, vectors - i);
2132 for (j = 0; j < rqpv; j++) {
2133 map_vector_to_rxq(adapter, i, rxr_idx);
2138 for (i = v_start; i < vectors; i++) {
2139 tqpv = DIV_ROUND_UP(txr_remaining, vectors - i);
2140 for (j = 0; j < tqpv; j++) {
2141 map_vector_to_txq(adapter, i, txr_idx);
2152 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2153 * @adapter: board private structure
2155 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2156 * interrupts from the kernel.
2158 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2160 struct net_device *netdev = adapter->netdev;
2161 irqreturn_t (*handler)(int, void *);
2162 int i, vector, q_vectors, err;
2165 /* Decrement for Other and TCP Timer vectors */
2166 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2168 /* Map the Tx/Rx rings to the vectors we were allotted. */
2169 err = ixgbe_map_rings_to_vectors(adapter, q_vectors);
2173 #define SET_HANDLER(_v) ((!(_v)->rxr_count) ? &ixgbe_msix_clean_tx : \
2174 (!(_v)->txr_count) ? &ixgbe_msix_clean_rx : \
2175 &ixgbe_msix_clean_many)
2176 for (vector = 0; vector < q_vectors; vector++) {
2177 handler = SET_HANDLER(adapter->q_vector[vector]);
2179 if (handler == &ixgbe_msix_clean_rx) {
2180 sprintf(adapter->name[vector], "%s-%s-%d",
2181 netdev->name, "rx", ri++);
2182 } else if (handler == &ixgbe_msix_clean_tx) {
2183 sprintf(adapter->name[vector], "%s-%s-%d",
2184 netdev->name, "tx", ti++);
2186 sprintf(adapter->name[vector], "%s-%s-%d",
2187 netdev->name, "TxRx", vector);
2189 err = request_irq(adapter->msix_entries[vector].vector,
2190 handler, 0, adapter->name[vector],
2191 adapter->q_vector[vector]);
2193 e_err(probe, "request_irq failed for MSIX interrupt "
2194 "Error: %d\n", err);
2195 goto free_queue_irqs;
2199 sprintf(adapter->name[vector], "%s:lsc", netdev->name);
2200 err = request_irq(adapter->msix_entries[vector].vector,
2201 ixgbe_msix_lsc, 0, adapter->name[vector], netdev);
2203 e_err(probe, "request_irq for msix_lsc failed: %d\n", err);
2204 goto free_queue_irqs;
2210 for (i = vector - 1; i >= 0; i--)
2211 free_irq(adapter->msix_entries[--vector].vector,
2212 adapter->q_vector[i]);
2213 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2214 pci_disable_msix(adapter->pdev);
2215 kfree(adapter->msix_entries);
2216 adapter->msix_entries = NULL;
2221 static void ixgbe_set_itr(struct ixgbe_adapter *adapter)
2223 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2225 u32 new_itr = q_vector->eitr;
2226 struct ixgbe_ring *rx_ring = adapter->rx_ring[0];
2227 struct ixgbe_ring *tx_ring = adapter->tx_ring[0];
2229 q_vector->tx_itr = ixgbe_update_itr(adapter, new_itr,
2231 tx_ring->total_packets,
2232 tx_ring->total_bytes);
2233 q_vector->rx_itr = ixgbe_update_itr(adapter, new_itr,
2235 rx_ring->total_packets,
2236 rx_ring->total_bytes);
2238 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
2240 switch (current_itr) {
2241 /* counts and packets in update_itr are dependent on these numbers */
2242 case lowest_latency:
2246 new_itr = 20000; /* aka hwitr = ~200 */
2255 if (new_itr != q_vector->eitr) {
2256 /* do an exponential smoothing */
2257 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
2259 /* save the algorithm value here, not the smoothed one */
2260 q_vector->eitr = new_itr;
2262 ixgbe_write_eitr(q_vector);
2267 * ixgbe_irq_enable - Enable default interrupt generation settings
2268 * @adapter: board private structure
2270 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2275 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2276 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2277 mask |= IXGBE_EIMS_GPI_SDP0;
2278 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2279 mask |= IXGBE_EIMS_GPI_SDP1;
2280 if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
2281 mask |= IXGBE_EIMS_ECC;
2282 mask |= IXGBE_EIMS_GPI_SDP1;
2283 mask |= IXGBE_EIMS_GPI_SDP2;
2284 if (adapter->num_vfs)
2285 mask |= IXGBE_EIMS_MAILBOX;
2287 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
2288 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
2289 mask |= IXGBE_EIMS_FLOW_DIR;
2291 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2293 ixgbe_irq_enable_queues(adapter, ~0);
2295 IXGBE_WRITE_FLUSH(&adapter->hw);
2297 if (adapter->num_vfs > 32) {
2298 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
2299 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2304 * ixgbe_intr - legacy mode Interrupt Handler
2305 * @irq: interrupt number
2306 * @data: pointer to a network interface device structure
2308 static irqreturn_t ixgbe_intr(int irq, void *data)
2310 struct net_device *netdev = data;
2311 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2312 struct ixgbe_hw *hw = &adapter->hw;
2313 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2317 * Workaround for silicon errata on 82598. Mask the interrupts
2318 * before the read of EICR.
2320 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2322 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2323 * therefore no explict interrupt disable is necessary */
2324 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2327 * shared interrupt alert!
2328 * make sure interrupts are enabled because the read will
2329 * have disabled interrupts due to EIAM
2330 * finish the workaround of silicon errata on 82598. Unmask
2331 * the interrupt that we masked before the EICR read.
2333 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2334 ixgbe_irq_enable(adapter, true, true);
2335 return IRQ_NONE; /* Not our interrupt */
2338 if (eicr & IXGBE_EICR_LSC)
2339 ixgbe_check_lsc(adapter);
2341 if (hw->mac.type == ixgbe_mac_82599EB)
2342 ixgbe_check_sfp_event(adapter, eicr);
2344 ixgbe_check_fan_failure(adapter, eicr);
2345 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2346 ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)))
2347 schedule_work(&adapter->check_overtemp_task);
2349 if (napi_schedule_prep(&(q_vector->napi))) {
2350 adapter->tx_ring[0]->total_packets = 0;
2351 adapter->tx_ring[0]->total_bytes = 0;
2352 adapter->rx_ring[0]->total_packets = 0;
2353 adapter->rx_ring[0]->total_bytes = 0;
2354 /* would disable interrupts here but EIAM disabled it */
2355 __napi_schedule(&(q_vector->napi));
2359 * re-enable link(maybe) and non-queue interrupts, no flush.
2360 * ixgbe_poll will re-enable the queue interrupts
2363 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2364 ixgbe_irq_enable(adapter, false, false);
2369 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
2371 int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2373 for (i = 0; i < q_vectors; i++) {
2374 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
2375 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
2376 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
2377 q_vector->rxr_count = 0;
2378 q_vector->txr_count = 0;
2383 * ixgbe_request_irq - initialize interrupts
2384 * @adapter: board private structure
2386 * Attempts to configure interrupts using the best available
2387 * capabilities of the hardware and kernel.
2389 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2391 struct net_device *netdev = adapter->netdev;
2394 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2395 err = ixgbe_request_msix_irqs(adapter);
2396 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
2397 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2398 netdev->name, netdev);
2400 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2401 netdev->name, netdev);
2405 e_err(probe, "request_irq failed, Error %d\n", err);
2410 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2412 struct net_device *netdev = adapter->netdev;
2414 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2417 q_vectors = adapter->num_msix_vectors;
2420 free_irq(adapter->msix_entries[i].vector, netdev);
2423 for (; i >= 0; i--) {
2424 free_irq(adapter->msix_entries[i].vector,
2425 adapter->q_vector[i]);
2428 ixgbe_reset_q_vectors(adapter);
2430 free_irq(adapter->pdev->irq, netdev);
2435 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2436 * @adapter: board private structure
2438 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2440 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2441 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2443 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2444 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2445 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2446 if (adapter->num_vfs > 32)
2447 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
2449 IXGBE_WRITE_FLUSH(&adapter->hw);
2450 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2452 for (i = 0; i < adapter->num_msix_vectors; i++)
2453 synchronize_irq(adapter->msix_entries[i].vector);
2455 synchronize_irq(adapter->pdev->irq);
2460 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2463 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2465 struct ixgbe_hw *hw = &adapter->hw;
2467 IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
2468 EITR_INTS_PER_SEC_TO_REG(adapter->rx_eitr_param));
2470 ixgbe_set_ivar(adapter, 0, 0, 0);
2471 ixgbe_set_ivar(adapter, 1, 0, 0);
2473 map_vector_to_rxq(adapter, 0, 0);
2474 map_vector_to_txq(adapter, 0, 0);
2476 e_info(hw, "Legacy interrupt IVAR setup done\n");
2480 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2481 * @adapter: board private structure
2482 * @ring: structure containing ring specific data
2484 * Configure the Tx descriptor ring after a reset.
2486 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2487 struct ixgbe_ring *ring)
2489 struct ixgbe_hw *hw = &adapter->hw;
2490 u64 tdba = ring->dma;
2493 u16 reg_idx = ring->reg_idx;
2495 /* disable queue to avoid issues while updating state */
2496 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2497 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx),
2498 txdctl & ~IXGBE_TXDCTL_ENABLE);
2499 IXGBE_WRITE_FLUSH(hw);
2501 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2502 (tdba & DMA_BIT_MASK(32)));
2503 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2504 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2505 ring->count * sizeof(union ixgbe_adv_tx_desc));
2506 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2507 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2508 ring->tail = hw->hw_addr + IXGBE_TDT(reg_idx);
2510 /* configure fetching thresholds */
2511 if (adapter->rx_itr_setting == 0) {
2512 /* cannot set wthresh when itr==0 */
2513 txdctl &= ~0x007F0000;
2515 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
2516 txdctl |= (8 << 16);
2518 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2519 /* PThresh workaround for Tx hang with DFP enabled. */
2523 /* reinitialize flowdirector state */
2524 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
2527 txdctl |= IXGBE_TXDCTL_ENABLE;
2528 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
2530 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2531 if (hw->mac.type == ixgbe_mac_82598EB &&
2532 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2535 /* poll to verify queue is enabled */
2538 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2539 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
2541 e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
2544 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
2546 struct ixgbe_hw *hw = &adapter->hw;
2550 if (hw->mac.type == ixgbe_mac_82598EB)
2553 /* disable the arbiter while setting MTQC */
2554 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2555 rttdcs |= IXGBE_RTTDCS_ARBDIS;
2556 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2558 /* set transmit pool layout */
2559 mask = (IXGBE_FLAG_SRIOV_ENABLED | IXGBE_FLAG_DCB_ENABLED);
2560 switch (adapter->flags & mask) {
2562 case (IXGBE_FLAG_SRIOV_ENABLED):
2563 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2564 (IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF));
2567 case (IXGBE_FLAG_DCB_ENABLED):
2568 /* We enable 8 traffic classes, DCB only */
2569 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2570 (IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ));
2574 IXGBE_WRITE_REG(hw, IXGBE_MTQC, IXGBE_MTQC_64Q_1PB);
2578 /* re-enable the arbiter */
2579 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2580 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2584 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2585 * @adapter: board private structure
2587 * Configure the Tx unit of the MAC after a reset.
2589 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2591 struct ixgbe_hw *hw = &adapter->hw;
2595 ixgbe_setup_mtqc(adapter);
2597 if (hw->mac.type != ixgbe_mac_82598EB) {
2598 /* DMATXCTL.EN must be before Tx queues are enabled */
2599 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2600 dmatxctl |= IXGBE_DMATXCTL_TE;
2601 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2604 /* Setup the HW Tx Head and Tail descriptor pointers */
2605 for (i = 0; i < adapter->num_tx_queues; i++)
2606 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
2609 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2611 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2612 struct ixgbe_ring *rx_ring)
2616 struct ixgbe_ring_feature *feature = adapter->ring_feature;
2618 index = rx_ring->reg_idx;
2619 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2621 mask = (unsigned long) feature[RING_F_RSS].mask;
2622 index = index & mask;
2624 srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(index));
2626 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
2627 srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
2628 if (adapter->num_vfs)
2629 srrctl |= IXGBE_SRRCTL_DROP_EN;
2631 srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
2632 IXGBE_SRRCTL_BSIZEHDR_MASK;
2634 if (ring_is_ps_enabled(rx_ring)) {
2635 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
2636 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2638 srrctl |= (PAGE_SIZE / 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2640 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2642 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
2643 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2644 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2647 IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(index), srrctl);
2650 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2652 struct ixgbe_hw *hw = &adapter->hw;
2653 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2654 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2655 0x6A3E67EA, 0x14364D17, 0x3BED200D};
2656 u32 mrqc = 0, reta = 0;
2661 /* Fill out hash function seeds */
2662 for (i = 0; i < 10; i++)
2663 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2665 /* Fill out redirection table */
2666 for (i = 0, j = 0; i < 128; i++, j++) {
2667 if (j == adapter->ring_feature[RING_F_RSS].indices)
2669 /* reta = 4-byte sliding window of
2670 * 0x00..(indices-1)(indices-1)00..etc. */
2671 reta = (reta << 8) | (j * 0x11);
2673 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2676 /* Disable indicating checksum in descriptor, enables RSS hash */
2677 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2678 rxcsum |= IXGBE_RXCSUM_PCSD;
2679 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2681 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
2682 mask = adapter->flags & IXGBE_FLAG_RSS_ENABLED;
2684 mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2685 #ifdef CONFIG_IXGBE_DCB
2686 | IXGBE_FLAG_DCB_ENABLED
2688 | IXGBE_FLAG_SRIOV_ENABLED
2692 case (IXGBE_FLAG_RSS_ENABLED):
2693 mrqc = IXGBE_MRQC_RSSEN;
2695 case (IXGBE_FLAG_SRIOV_ENABLED):
2696 mrqc = IXGBE_MRQC_VMDQEN;
2698 #ifdef CONFIG_IXGBE_DCB
2699 case (IXGBE_FLAG_DCB_ENABLED):
2700 mrqc = IXGBE_MRQC_RT8TCEN;
2702 #endif /* CONFIG_IXGBE_DCB */
2707 /* Perform hash on these packet types */
2708 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2709 | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2710 | IXGBE_MRQC_RSS_FIELD_IPV6
2711 | IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
2713 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2717 * ixgbe_configure_rscctl - enable RSC for the indicated ring
2718 * @adapter: address of board private structure
2719 * @index: index of ring to set
2721 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
2722 struct ixgbe_ring *ring)
2724 struct ixgbe_hw *hw = &adapter->hw;
2727 u16 reg_idx = ring->reg_idx;
2729 if (!ring_is_rsc_enabled(ring))
2732 rx_buf_len = ring->rx_buf_len;
2733 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
2734 rscctrl |= IXGBE_RSCCTL_RSCEN;
2736 * we must limit the number of descriptors so that the
2737 * total size of max desc * buf_len is not greater
2740 if (ring_is_ps_enabled(ring)) {
2741 #if (MAX_SKB_FRAGS > 16)
2742 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2743 #elif (MAX_SKB_FRAGS > 8)
2744 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2745 #elif (MAX_SKB_FRAGS > 4)
2746 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2748 rscctrl |= IXGBE_RSCCTL_MAXDESC_1;
2751 if (rx_buf_len < IXGBE_RXBUFFER_4096)
2752 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2753 else if (rx_buf_len < IXGBE_RXBUFFER_8192)
2754 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2756 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2758 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
2762 * ixgbe_set_uta - Set unicast filter table address
2763 * @adapter: board private structure
2765 * The unicast table address is a register array of 32-bit registers.
2766 * The table is meant to be used in a way similar to how the MTA is used
2767 * however due to certain limitations in the hardware it is necessary to
2768 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
2769 * enable bit to allow vlan tag stripping when promiscuous mode is enabled
2771 static void ixgbe_set_uta(struct ixgbe_adapter *adapter)
2773 struct ixgbe_hw *hw = &adapter->hw;
2776 /* The UTA table only exists on 82599 hardware and newer */
2777 if (hw->mac.type < ixgbe_mac_82599EB)
2780 /* we only need to do this if VMDq is enabled */
2781 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
2784 for (i = 0; i < 128; i++)
2785 IXGBE_WRITE_REG(hw, IXGBE_UTA(i), ~0);
2788 #define IXGBE_MAX_RX_DESC_POLL 10
2789 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
2790 struct ixgbe_ring *ring)
2792 struct ixgbe_hw *hw = &adapter->hw;
2793 int reg_idx = ring->reg_idx;
2794 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
2797 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2798 if (hw->mac.type == ixgbe_mac_82598EB &&
2799 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2804 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2805 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
2808 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
2809 "the polling period\n", reg_idx);
2813 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
2814 struct ixgbe_ring *ring)
2816 struct ixgbe_hw *hw = &adapter->hw;
2817 u64 rdba = ring->dma;
2819 u16 reg_idx = ring->reg_idx;
2821 /* disable queue to avoid issues while updating state */
2822 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2823 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx),
2824 rxdctl & ~IXGBE_RXDCTL_ENABLE);
2825 IXGBE_WRITE_FLUSH(hw);
2827 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
2828 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
2829 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
2830 ring->count * sizeof(union ixgbe_adv_rx_desc));
2831 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
2832 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
2833 ring->tail = hw->hw_addr + IXGBE_RDT(reg_idx);
2835 ixgbe_configure_srrctl(adapter, ring);
2836 ixgbe_configure_rscctl(adapter, ring);
2838 if (hw->mac.type == ixgbe_mac_82598EB) {
2840 * enable cache line friendly hardware writes:
2841 * PTHRESH=32 descriptors (half the internal cache),
2842 * this also removes ugly rx_no_buffer_count increment
2843 * HTHRESH=4 descriptors (to minimize latency on fetch)
2844 * WTHRESH=8 burst writeback up to two cache lines
2846 rxdctl &= ~0x3FFFFF;
2850 /* enable receive descriptor ring */
2851 rxdctl |= IXGBE_RXDCTL_ENABLE;
2852 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
2854 ixgbe_rx_desc_queue_enable(adapter, ring);
2855 ixgbe_alloc_rx_buffers(ring, IXGBE_DESC_UNUSED(ring));
2858 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
2860 struct ixgbe_hw *hw = &adapter->hw;
2863 /* PSRTYPE must be initialized in non 82598 adapters */
2864 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
2865 IXGBE_PSRTYPE_UDPHDR |
2866 IXGBE_PSRTYPE_IPV4HDR |
2867 IXGBE_PSRTYPE_L2HDR |
2868 IXGBE_PSRTYPE_IPV6HDR;
2870 if (hw->mac.type == ixgbe_mac_82598EB)
2873 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED)
2874 psrtype |= (adapter->num_rx_queues_per_pool << 29);
2876 for (p = 0; p < adapter->num_rx_pools; p++)
2877 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(adapter->num_vfs + p),
2881 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
2883 struct ixgbe_hw *hw = &adapter->hw;
2886 u32 reg_offset, vf_shift;
2889 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
2892 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
2893 vt_reg_bits = IXGBE_VMD_CTL_VMDQ_EN | IXGBE_VT_CTL_REPLEN;
2894 vt_reg_bits |= (adapter->num_vfs << IXGBE_VT_CTL_POOL_SHIFT);
2895 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl | vt_reg_bits);
2897 vf_shift = adapter->num_vfs % 32;
2898 reg_offset = (adapter->num_vfs > 32) ? 1 : 0;
2900 /* Enable only the PF's pool for Tx/Rx */
2901 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (1 << vf_shift));
2902 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), 0);
2903 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (1 << vf_shift));
2904 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), 0);
2905 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
2907 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
2908 hw->mac.ops.set_vmdq(hw, 0, adapter->num_vfs);
2911 * Set up VF register offsets for selected VT Mode,
2912 * i.e. 32 or 64 VFs for SR-IOV
2914 gcr_ext = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
2915 gcr_ext |= IXGBE_GCR_EXT_MSIX_EN;
2916 gcr_ext |= IXGBE_GCR_EXT_VT_MODE_64;
2917 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
2919 /* enable Tx loopback for VF/PF communication */
2920 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
2923 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
2925 struct ixgbe_hw *hw = &adapter->hw;
2926 struct net_device *netdev = adapter->netdev;
2927 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2929 struct ixgbe_ring *rx_ring;
2933 /* Decide whether to use packet split mode or not */
2934 /* Do not use packet split if we're in SR-IOV Mode */
2935 if (!adapter->num_vfs)
2936 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
2938 /* Set the RX buffer length according to the mode */
2939 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
2940 rx_buf_len = IXGBE_RX_HDR_SIZE;
2942 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
2943 (netdev->mtu <= ETH_DATA_LEN))
2944 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
2946 rx_buf_len = ALIGN(max_frame + VLAN_HLEN, 1024);
2950 /* adjust max frame to be able to do baby jumbo for FCoE */
2951 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
2952 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
2953 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
2955 #endif /* IXGBE_FCOE */
2956 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
2957 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
2958 mhadd &= ~IXGBE_MHADD_MFS_MASK;
2959 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
2961 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
2964 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2965 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
2966 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2967 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
2970 * Setup the HW Rx Head and Tail Descriptor Pointers and
2971 * the Base and Length of the Rx Descriptor Ring
2973 for (i = 0; i < adapter->num_rx_queues; i++) {
2974 rx_ring = adapter->rx_ring[i];
2975 rx_ring->rx_buf_len = rx_buf_len;
2977 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)
2978 set_ring_ps_enabled(rx_ring);
2980 clear_ring_ps_enabled(rx_ring);
2982 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
2983 set_ring_rsc_enabled(rx_ring);
2985 clear_ring_rsc_enabled(rx_ring);
2988 if (netdev->features & NETIF_F_FCOE_MTU) {
2989 struct ixgbe_ring_feature *f;
2990 f = &adapter->ring_feature[RING_F_FCOE];
2991 if ((i >= f->mask) && (i < f->mask + f->indices)) {
2992 clear_ring_ps_enabled(rx_ring);
2993 if (rx_buf_len < IXGBE_FCOE_JUMBO_FRAME_SIZE)
2994 rx_ring->rx_buf_len =
2995 IXGBE_FCOE_JUMBO_FRAME_SIZE;
2996 } else if (!ring_is_rsc_enabled(rx_ring) &&
2997 !ring_is_ps_enabled(rx_ring)) {
2998 rx_ring->rx_buf_len =
2999 IXGBE_FCOE_JUMBO_FRAME_SIZE;
3002 #endif /* IXGBE_FCOE */
3007 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
3009 struct ixgbe_hw *hw = &adapter->hw;
3010 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
3012 switch (hw->mac.type) {
3013 case ixgbe_mac_82598EB:
3015 * For VMDq support of different descriptor types or
3016 * buffer sizes through the use of multiple SRRCTL
3017 * registers, RDRXCTL.MVMEN must be set to 1
3019 * also, the manual doesn't mention it clearly but DCA hints
3020 * will only use queue 0's tags unless this bit is set. Side
3021 * effects of setting this bit are only that SRRCTL must be
3022 * fully programmed [0..15]
3024 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
3026 case ixgbe_mac_82599EB:
3027 /* Disable RSC for ACK packets */
3028 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
3029 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
3030 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3031 /* hardware requires some bits to be set by default */
3032 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3033 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3036 /* We should do nothing since we don't know this hardware */
3040 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3044 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3045 * @adapter: board private structure
3047 * Configure the Rx unit of the MAC after a reset.
3049 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3051 struct ixgbe_hw *hw = &adapter->hw;
3055 /* disable receives while setting up the descriptors */
3056 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3057 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3059 ixgbe_setup_psrtype(adapter);
3060 ixgbe_setup_rdrxctl(adapter);
3062 /* Program registers for the distribution of queues */
3063 ixgbe_setup_mrqc(adapter);
3065 ixgbe_set_uta(adapter);
3067 /* set_rx_buffer_len must be called before ring initialization */
3068 ixgbe_set_rx_buffer_len(adapter);
3071 * Setup the HW Rx Head and Tail Descriptor Pointers and
3072 * the Base and Length of the Rx Descriptor Ring
3074 for (i = 0; i < adapter->num_rx_queues; i++)
3075 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3077 /* disable drop enable for 82598 parts */
3078 if (hw->mac.type == ixgbe_mac_82598EB)
3079 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3081 /* enable all receives */
3082 rxctrl |= IXGBE_RXCTRL_RXEN;
3083 hw->mac.ops.enable_rx_dma(hw, rxctrl);
3086 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3088 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3089 struct ixgbe_hw *hw = &adapter->hw;
3090 int pool_ndx = adapter->num_vfs;
3092 /* add VID to filter table */
3093 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, true);
3094 set_bit(vid, adapter->active_vlans);
3097 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3099 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3100 struct ixgbe_hw *hw = &adapter->hw;
3101 int pool_ndx = adapter->num_vfs;
3103 /* remove VID from filter table */
3104 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, false);
3105 clear_bit(vid, adapter->active_vlans);
3109 * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3110 * @adapter: driver data
3112 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
3114 struct ixgbe_hw *hw = &adapter->hw;
3117 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3118 vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
3119 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3123 * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3124 * @adapter: driver data
3126 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
3128 struct ixgbe_hw *hw = &adapter->hw;
3131 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3132 vlnctrl |= IXGBE_VLNCTRL_VFE;
3133 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3134 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3138 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3139 * @adapter: driver data
3141 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3143 struct ixgbe_hw *hw = &adapter->hw;
3147 switch (hw->mac.type) {
3148 case ixgbe_mac_82598EB:
3149 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3150 vlnctrl &= ~IXGBE_VLNCTRL_VME;
3151 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3153 case ixgbe_mac_82599EB:
3154 for (i = 0; i < adapter->num_rx_queues; i++) {
3155 j = adapter->rx_ring[i]->reg_idx;
3156 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3157 vlnctrl &= ~IXGBE_RXDCTL_VME;
3158 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3167 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3168 * @adapter: driver data
3170 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3172 struct ixgbe_hw *hw = &adapter->hw;
3176 switch (hw->mac.type) {
3177 case ixgbe_mac_82598EB:
3178 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3179 vlnctrl |= IXGBE_VLNCTRL_VME;
3180 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3182 case ixgbe_mac_82599EB:
3183 for (i = 0; i < adapter->num_rx_queues; i++) {
3184 j = adapter->rx_ring[i]->reg_idx;
3185 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3186 vlnctrl |= IXGBE_RXDCTL_VME;
3187 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3195 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3199 ixgbe_vlan_rx_add_vid(adapter->netdev, 0);
3201 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3202 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
3206 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3207 * @netdev: network interface device structure
3209 * Writes unicast address list to the RAR table.
3210 * Returns: -ENOMEM on failure/insufficient address space
3211 * 0 on no addresses written
3212 * X on writing X addresses to the RAR table
3214 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3216 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3217 struct ixgbe_hw *hw = &adapter->hw;
3218 unsigned int vfn = adapter->num_vfs;
3219 unsigned int rar_entries = hw->mac.num_rar_entries - (vfn + 1);
3222 /* return ENOMEM indicating insufficient memory for addresses */
3223 if (netdev_uc_count(netdev) > rar_entries)
3226 if (!netdev_uc_empty(netdev) && rar_entries) {
3227 struct netdev_hw_addr *ha;
3228 /* return error if we do not support writing to RAR table */
3229 if (!hw->mac.ops.set_rar)
3232 netdev_for_each_uc_addr(ha, netdev) {
3235 hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3240 /* write the addresses in reverse order to avoid write combining */
3241 for (; rar_entries > 0 ; rar_entries--)
3242 hw->mac.ops.clear_rar(hw, rar_entries);
3248 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3249 * @netdev: network interface device structure
3251 * The set_rx_method entry point is called whenever the unicast/multicast
3252 * address list or the network interface flags are updated. This routine is
3253 * responsible for configuring the hardware for proper unicast, multicast and
3256 void ixgbe_set_rx_mode(struct net_device *netdev)
3258 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3259 struct ixgbe_hw *hw = &adapter->hw;
3260 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3263 /* Check for Promiscuous and All Multicast modes */
3265 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3267 /* set all bits that we expect to always be set */
3268 fctrl |= IXGBE_FCTRL_BAM;
3269 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3270 fctrl |= IXGBE_FCTRL_PMCF;
3272 /* clear the bits we are changing the status of */
3273 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3275 if (netdev->flags & IFF_PROMISC) {
3276 hw->addr_ctrl.user_set_promisc = true;
3277 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3278 vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3279 /* don't hardware filter vlans in promisc mode */
3280 ixgbe_vlan_filter_disable(adapter);
3282 if (netdev->flags & IFF_ALLMULTI) {
3283 fctrl |= IXGBE_FCTRL_MPE;
3284 vmolr |= IXGBE_VMOLR_MPE;
3287 * Write addresses to the MTA, if the attempt fails
3288 * then we should just turn on promiscous mode so
3289 * that we can at least receive multicast traffic
3291 hw->mac.ops.update_mc_addr_list(hw, netdev);
3292 vmolr |= IXGBE_VMOLR_ROMPE;
3294 ixgbe_vlan_filter_enable(adapter);
3295 hw->addr_ctrl.user_set_promisc = false;
3297 * Write addresses to available RAR registers, if there is not
3298 * sufficient space to store all the addresses then enable
3299 * unicast promiscous mode
3301 count = ixgbe_write_uc_addr_list(netdev);
3303 fctrl |= IXGBE_FCTRL_UPE;
3304 vmolr |= IXGBE_VMOLR_ROPE;
3308 if (adapter->num_vfs) {
3309 ixgbe_restore_vf_multicasts(adapter);
3310 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(adapter->num_vfs)) &
3311 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3313 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(adapter->num_vfs), vmolr);
3316 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3318 if (netdev->features & NETIF_F_HW_VLAN_RX)
3319 ixgbe_vlan_strip_enable(adapter);
3321 ixgbe_vlan_strip_disable(adapter);
3324 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3327 struct ixgbe_q_vector *q_vector;
3328 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3330 /* legacy and MSI only use one vector */
3331 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3334 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3335 struct napi_struct *napi;
3336 q_vector = adapter->q_vector[q_idx];
3337 napi = &q_vector->napi;
3338 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3339 if (!q_vector->rxr_count || !q_vector->txr_count) {
3340 if (q_vector->txr_count == 1)
3341 napi->poll = &ixgbe_clean_txonly;
3342 else if (q_vector->rxr_count == 1)
3343 napi->poll = &ixgbe_clean_rxonly;
3351 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3354 struct ixgbe_q_vector *q_vector;
3355 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3357 /* legacy and MSI only use one vector */
3358 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3361 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3362 q_vector = adapter->q_vector[q_idx];
3363 napi_disable(&q_vector->napi);
3367 #ifdef CONFIG_IXGBE_DCB
3369 * ixgbe_configure_dcb - Configure DCB hardware
3370 * @adapter: ixgbe adapter struct
3372 * This is called by the driver on open to configure the DCB hardware.
3373 * This is also called by the gennetlink interface when reconfiguring
3376 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3378 struct ixgbe_hw *hw = &adapter->hw;
3379 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3381 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3382 if (hw->mac.type == ixgbe_mac_82598EB)
3383 netif_set_gso_max_size(adapter->netdev, 65536);
3387 if (hw->mac.type == ixgbe_mac_82598EB)
3388 netif_set_gso_max_size(adapter->netdev, 32768);
3391 if (adapter->netdev->features & NETIF_F_FCOE_MTU)
3392 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
3395 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3397 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3400 /* Enable VLAN tag insert/strip */
3401 adapter->netdev->features |= NETIF_F_HW_VLAN_RX;
3403 hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
3405 /* reconfigure the hardware */
3406 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
3410 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3412 struct net_device *netdev = adapter->netdev;
3413 struct ixgbe_hw *hw = &adapter->hw;
3416 #ifdef CONFIG_IXGBE_DCB
3417 ixgbe_configure_dcb(adapter);
3420 ixgbe_set_rx_mode(netdev);
3421 ixgbe_restore_vlan(adapter);
3424 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
3425 ixgbe_configure_fcoe(adapter);
3427 #endif /* IXGBE_FCOE */
3428 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3429 for (i = 0; i < adapter->num_tx_queues; i++)
3430 adapter->tx_ring[i]->atr_sample_rate =
3431 adapter->atr_sample_rate;
3432 ixgbe_init_fdir_signature_82599(hw, adapter->fdir_pballoc);
3433 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3434 ixgbe_init_fdir_perfect_82599(hw, adapter->fdir_pballoc);
3436 ixgbe_configure_virtualization(adapter);
3438 ixgbe_configure_tx(adapter);
3439 ixgbe_configure_rx(adapter);
3442 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3444 switch (hw->phy.type) {
3445 case ixgbe_phy_sfp_avago:
3446 case ixgbe_phy_sfp_ftl:
3447 case ixgbe_phy_sfp_intel:
3448 case ixgbe_phy_sfp_unknown:
3449 case ixgbe_phy_sfp_passive_tyco:
3450 case ixgbe_phy_sfp_passive_unknown:
3451 case ixgbe_phy_sfp_active_unknown:
3452 case ixgbe_phy_sfp_ftl_active:
3460 * ixgbe_sfp_link_config - set up SFP+ link
3461 * @adapter: pointer to private adapter struct
3463 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3465 struct ixgbe_hw *hw = &adapter->hw;
3467 if (hw->phy.multispeed_fiber) {
3469 * In multispeed fiber setups, the device may not have
3470 * had a physical connection when the driver loaded.
3471 * If that's the case, the initial link configuration
3472 * couldn't get the MAC into 10G or 1G mode, so we'll
3473 * never have a link status change interrupt fire.
3474 * We need to try and force an autonegotiation
3475 * session, then bring up link.
3477 hw->mac.ops.setup_sfp(hw);
3478 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
3479 schedule_work(&adapter->multispeed_fiber_task);
3482 * Direct Attach Cu and non-multispeed fiber modules
3483 * still need to be configured properly prior to
3486 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_MOD_TASK))
3487 schedule_work(&adapter->sfp_config_module_task);
3492 * ixgbe_non_sfp_link_config - set up non-SFP+ link
3493 * @hw: pointer to private hardware struct
3495 * Returns 0 on success, negative on failure
3497 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3500 bool negotiation, link_up = false;
3501 u32 ret = IXGBE_ERR_LINK_SETUP;
3503 if (hw->mac.ops.check_link)
3504 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3509 if (hw->mac.ops.get_link_capabilities)
3510 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
3515 if (hw->mac.ops.setup_link)
3516 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
3521 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
3523 struct ixgbe_hw *hw = &adapter->hw;
3526 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3527 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
3529 gpie |= IXGBE_GPIE_EIAME;
3531 * use EIAM to auto-mask when MSI-X interrupt is asserted
3532 * this saves a register write for every interrupt
3534 switch (hw->mac.type) {
3535 case ixgbe_mac_82598EB:
3536 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3539 case ixgbe_mac_82599EB:
3540 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
3541 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
3545 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
3546 * specifically only auto mask tx and rx interrupts */
3547 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3550 /* XXX: to interrupt immediately for EICS writes, enable this */
3551 /* gpie |= IXGBE_GPIE_EIMEN; */
3553 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3554 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
3555 gpie |= IXGBE_GPIE_VTMODE_64;
3558 /* Enable fan failure interrupt */
3559 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3560 gpie |= IXGBE_SDP1_GPIEN;
3562 if (hw->mac.type == ixgbe_mac_82599EB)
3563 gpie |= IXGBE_SDP1_GPIEN;
3564 gpie |= IXGBE_SDP2_GPIEN;
3566 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3569 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
3571 struct ixgbe_hw *hw = &adapter->hw;
3575 ixgbe_get_hw_control(adapter);
3576 ixgbe_setup_gpie(adapter);
3578 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3579 ixgbe_configure_msix(adapter);
3581 ixgbe_configure_msi_and_legacy(adapter);
3583 /* enable the optics */
3584 if (hw->phy.multispeed_fiber)
3585 hw->mac.ops.enable_tx_laser(hw);
3587 clear_bit(__IXGBE_DOWN, &adapter->state);
3588 ixgbe_napi_enable_all(adapter);
3590 if (ixgbe_is_sfp(hw)) {
3591 ixgbe_sfp_link_config(adapter);
3593 err = ixgbe_non_sfp_link_config(hw);
3595 e_err(probe, "link_config FAILED %d\n", err);
3598 /* clear any pending interrupts, may auto mask */
3599 IXGBE_READ_REG(hw, IXGBE_EICR);
3600 ixgbe_irq_enable(adapter, true, true);
3603 * If this adapter has a fan, check to see if we had a failure
3604 * before we enabled the interrupt.
3606 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3607 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
3608 if (esdp & IXGBE_ESDP_SDP1)
3609 e_crit(drv, "Fan has stopped, replace the adapter\n");
3613 * For hot-pluggable SFP+ devices, a new SFP+ module may have
3614 * arrived before interrupts were enabled but after probe. Such
3615 * devices wouldn't have their type identified yet. We need to
3616 * kick off the SFP+ module setup first, then try to bring up link.
3617 * If we're not hot-pluggable SFP+, we just need to configure link
3620 if (hw->phy.type == ixgbe_phy_unknown)
3621 schedule_work(&adapter->sfp_config_module_task);
3623 /* enable transmits */
3624 netif_tx_start_all_queues(adapter->netdev);
3626 /* bring the link up in the watchdog, this could race with our first
3627 * link up interrupt but shouldn't be a problem */
3628 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3629 adapter->link_check_timeout = jiffies;
3630 mod_timer(&adapter->watchdog_timer, jiffies);
3632 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
3633 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
3634 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
3635 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
3640 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
3642 WARN_ON(in_interrupt());
3643 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
3645 ixgbe_down(adapter);
3647 * If SR-IOV enabled then wait a bit before bringing the adapter
3648 * back up to give the VFs time to respond to the reset. The
3649 * two second wait is based upon the watchdog timer cycle in
3652 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3655 clear_bit(__IXGBE_RESETTING, &adapter->state);
3658 int ixgbe_up(struct ixgbe_adapter *adapter)
3660 /* hardware has been reset, we need to reload some things */
3661 ixgbe_configure(adapter);
3663 return ixgbe_up_complete(adapter);
3666 void ixgbe_reset(struct ixgbe_adapter *adapter)
3668 struct ixgbe_hw *hw = &adapter->hw;
3671 err = hw->mac.ops.init_hw(hw);
3674 case IXGBE_ERR_SFP_NOT_PRESENT:
3676 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
3677 e_dev_err("master disable timed out\n");
3679 case IXGBE_ERR_EEPROM_VERSION:
3680 /* We are running on a pre-production device, log a warning */
3681 e_dev_warn("This device is a pre-production adapter/LOM. "
3682 "Please be aware there may be issuesassociated with "
3683 "your hardware. If you are experiencing problems "
3684 "please contact your Intel or hardware "
3685 "representative who provided you with this "
3689 e_dev_err("Hardware Error: %d\n", err);
3692 /* reprogram the RAR[0] in case user changed it. */
3693 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
3698 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
3699 * @rx_ring: ring to free buffers from
3701 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
3703 struct device *dev = rx_ring->dev;
3707 /* ring already cleared, nothing to do */
3708 if (!rx_ring->rx_buffer_info)
3711 /* Free all the Rx ring sk_buffs */
3712 for (i = 0; i < rx_ring->count; i++) {
3713 struct ixgbe_rx_buffer *rx_buffer_info;
3715 rx_buffer_info = &rx_ring->rx_buffer_info[i];
3716 if (rx_buffer_info->dma) {
3717 dma_unmap_single(rx_ring->dev, rx_buffer_info->dma,
3718 rx_ring->rx_buf_len,
3720 rx_buffer_info->dma = 0;
3722 if (rx_buffer_info->skb) {
3723 struct sk_buff *skb = rx_buffer_info->skb;
3724 rx_buffer_info->skb = NULL;
3726 struct sk_buff *this = skb;
3727 if (IXGBE_RSC_CB(this)->delay_unmap) {
3728 dma_unmap_single(dev,
3729 IXGBE_RSC_CB(this)->dma,
3730 rx_ring->rx_buf_len,
3732 IXGBE_RSC_CB(this)->dma = 0;
3733 IXGBE_RSC_CB(skb)->delay_unmap = false;
3736 dev_kfree_skb(this);
3739 if (!rx_buffer_info->page)
3741 if (rx_buffer_info->page_dma) {
3742 dma_unmap_page(dev, rx_buffer_info->page_dma,
3743 PAGE_SIZE / 2, DMA_FROM_DEVICE);
3744 rx_buffer_info->page_dma = 0;
3746 put_page(rx_buffer_info->page);
3747 rx_buffer_info->page = NULL;
3748 rx_buffer_info->page_offset = 0;
3751 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
3752 memset(rx_ring->rx_buffer_info, 0, size);
3754 /* Zero out the descriptor ring */
3755 memset(rx_ring->desc, 0, rx_ring->size);
3757 rx_ring->next_to_clean = 0;
3758 rx_ring->next_to_use = 0;
3762 * ixgbe_clean_tx_ring - Free Tx Buffers
3763 * @tx_ring: ring to be cleaned
3765 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
3767 struct ixgbe_tx_buffer *tx_buffer_info;
3771 /* ring already cleared, nothing to do */
3772 if (!tx_ring->tx_buffer_info)
3775 /* Free all the Tx ring sk_buffs */
3776 for (i = 0; i < tx_ring->count; i++) {
3777 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3778 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
3781 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
3782 memset(tx_ring->tx_buffer_info, 0, size);
3784 /* Zero out the descriptor ring */
3785 memset(tx_ring->desc, 0, tx_ring->size);
3787 tx_ring->next_to_use = 0;
3788 tx_ring->next_to_clean = 0;
3792 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
3793 * @adapter: board private structure
3795 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
3799 for (i = 0; i < adapter->num_rx_queues; i++)
3800 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
3804 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
3805 * @adapter: board private structure
3807 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
3811 for (i = 0; i < adapter->num_tx_queues; i++)
3812 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
3815 void ixgbe_down(struct ixgbe_adapter *adapter)
3817 struct net_device *netdev = adapter->netdev;
3818 struct ixgbe_hw *hw = &adapter->hw;
3822 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3824 /* signal that we are down to the interrupt handler */
3825 set_bit(__IXGBE_DOWN, &adapter->state);
3827 /* disable receive for all VFs and wait one second */
3828 if (adapter->num_vfs) {
3829 /* ping all the active vfs to let them know we are going down */
3830 ixgbe_ping_all_vfs(adapter);
3832 /* Disable all VFTE/VFRE TX/RX */
3833 ixgbe_disable_tx_rx(adapter);
3835 /* Mark all the VFs as inactive */
3836 for (i = 0 ; i < adapter->num_vfs; i++)
3837 adapter->vfinfo[i].clear_to_send = 0;
3840 /* disable receives */
3841 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3842 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3844 IXGBE_WRITE_FLUSH(hw);
3847 netif_tx_stop_all_queues(netdev);
3849 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
3850 del_timer_sync(&adapter->sfp_timer);
3851 del_timer_sync(&adapter->watchdog_timer);
3852 cancel_work_sync(&adapter->watchdog_task);
3854 netif_carrier_off(netdev);
3855 netif_tx_disable(netdev);
3857 ixgbe_irq_disable(adapter);
3859 ixgbe_napi_disable_all(adapter);
3861 /* Cleanup the affinity_hint CPU mask memory and callback */
3862 for (i = 0; i < num_q_vectors; i++) {
3863 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
3864 /* clear the affinity_mask in the IRQ descriptor */
3865 irq_set_affinity_hint(adapter->msix_entries[i]. vector, NULL);
3866 /* release the CPU mask memory */
3867 free_cpumask_var(q_vector->affinity_mask);
3870 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3871 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3872 cancel_work_sync(&adapter->fdir_reinit_task);
3874 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
3875 cancel_work_sync(&adapter->check_overtemp_task);
3877 /* disable transmits in the hardware now that interrupts are off */
3878 for (i = 0; i < adapter->num_tx_queues; i++) {
3879 j = adapter->tx_ring[i]->reg_idx;
3880 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3881 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j),
3882 (txdctl & ~IXGBE_TXDCTL_ENABLE));
3884 /* Disable the Tx DMA engine on 82599 */
3885 if (hw->mac.type == ixgbe_mac_82599EB)
3886 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
3887 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
3888 ~IXGBE_DMATXCTL_TE));
3890 /* power down the optics */
3891 if (hw->phy.multispeed_fiber)
3892 hw->mac.ops.disable_tx_laser(hw);
3894 /* clear n-tuple filters that are cached */
3895 ethtool_ntuple_flush(netdev);
3897 if (!pci_channel_offline(adapter->pdev))
3898 ixgbe_reset(adapter);
3899 ixgbe_clean_all_tx_rings(adapter);
3900 ixgbe_clean_all_rx_rings(adapter);
3902 #ifdef CONFIG_IXGBE_DCA
3903 /* since we reset the hardware DCA settings were cleared */
3904 ixgbe_setup_dca(adapter);
3909 * ixgbe_poll - NAPI Rx polling callback
3910 * @napi: structure for representing this polling device
3911 * @budget: how many packets driver is allowed to clean
3913 * This function is used for legacy and MSI, NAPI mode
3915 static int ixgbe_poll(struct napi_struct *napi, int budget)
3917 struct ixgbe_q_vector *q_vector =
3918 container_of(napi, struct ixgbe_q_vector, napi);
3919 struct ixgbe_adapter *adapter = q_vector->adapter;
3920 int tx_clean_complete, work_done = 0;
3922 #ifdef CONFIG_IXGBE_DCA
3923 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
3924 ixgbe_update_dca(q_vector);
3927 tx_clean_complete = ixgbe_clean_tx_irq(q_vector, adapter->tx_ring[0]);
3928 ixgbe_clean_rx_irq(q_vector, adapter->rx_ring[0], &work_done, budget);
3930 if (!tx_clean_complete)
3933 /* If budget not fully consumed, exit the polling mode */
3934 if (work_done < budget) {
3935 napi_complete(napi);
3936 if (adapter->rx_itr_setting & 1)
3937 ixgbe_set_itr(adapter);
3938 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3939 ixgbe_irq_enable_queues(adapter, IXGBE_EIMS_RTX_QUEUE);
3945 * ixgbe_tx_timeout - Respond to a Tx Hang
3946 * @netdev: network interface device structure
3948 static void ixgbe_tx_timeout(struct net_device *netdev)
3950 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3952 /* Do the reset outside of interrupt context */
3953 schedule_work(&adapter->reset_task);
3956 static void ixgbe_reset_task(struct work_struct *work)
3958 struct ixgbe_adapter *adapter;
3959 adapter = container_of(work, struct ixgbe_adapter, reset_task);
3961 /* If we're already down or resetting, just bail */
3962 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
3963 test_bit(__IXGBE_RESETTING, &adapter->state))
3966 adapter->tx_timeout_count++;
3968 ixgbe_dump(adapter);
3969 netdev_err(adapter->netdev, "Reset adapter\n");
3970 ixgbe_reinit_locked(adapter);
3973 #ifdef CONFIG_IXGBE_DCB
3974 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
3977 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_DCB];
3979 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
3983 adapter->num_rx_queues = f->indices;
3984 adapter->num_tx_queues = f->indices;
3992 * ixgbe_set_rss_queues: Allocate queues for RSS
3993 * @adapter: board private structure to initialize
3995 * This is our "base" multiqueue mode. RSS (Receive Side Scaling) will try
3996 * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
3999 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
4002 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
4004 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4006 adapter->num_rx_queues = f->indices;
4007 adapter->num_tx_queues = f->indices;
4017 * ixgbe_set_fdir_queues: Allocate queues for Flow Director
4018 * @adapter: board private structure to initialize
4020 * Flow Director is an advanced Rx filter, attempting to get Rx flows back
4021 * to the original CPU that initiated the Tx session. This runs in addition
4022 * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
4023 * Rx load across CPUs using RSS.
4026 static inline bool ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
4029 struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
4031 f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
4034 /* Flow Director must have RSS enabled */
4035 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
4036 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
4037 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)))) {
4038 adapter->num_tx_queues = f_fdir->indices;
4039 adapter->num_rx_queues = f_fdir->indices;
4042 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4043 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4050 * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
4051 * @adapter: board private structure to initialize
4053 * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
4054 * The ring feature mask is not used as a mask for FCoE, as it can take any 8
4055 * rx queues out of the max number of rx queues, instead, it is used as the
4056 * index of the first rx queue used by FCoE.
4059 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
4062 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4064 f->indices = min((int)num_online_cpus(), f->indices);
4065 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
4066 adapter->num_rx_queues = 1;
4067 adapter->num_tx_queues = 1;
4068 #ifdef CONFIG_IXGBE_DCB
4069 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4070 e_info(probe, "FCoE enabled with DCB\n");
4071 ixgbe_set_dcb_queues(adapter);
4074 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4075 e_info(probe, "FCoE enabled with RSS\n");
4076 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4077 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
4078 ixgbe_set_fdir_queues(adapter);
4080 ixgbe_set_rss_queues(adapter);
4082 /* adding FCoE rx rings to the end */
4083 f->mask = adapter->num_rx_queues;
4084 adapter->num_rx_queues += f->indices;
4085 adapter->num_tx_queues += f->indices;
4093 #endif /* IXGBE_FCOE */
4095 * ixgbe_set_sriov_queues: Allocate queues for IOV use
4096 * @adapter: board private structure to initialize
4098 * IOV doesn't actually use anything, so just NAK the
4099 * request for now and let the other queue routines
4100 * figure out what to do.
4102 static inline bool ixgbe_set_sriov_queues(struct ixgbe_adapter *adapter)
4108 * ixgbe_set_num_queues: Allocate queues for device, feature dependant
4109 * @adapter: board private structure to initialize
4111 * This is the top level queue allocation routine. The order here is very
4112 * important, starting with the "most" number of features turned on at once,
4113 * and ending with the smallest set of features. This way large combinations
4114 * can be allocated if they're turned on, and smaller combinations are the
4115 * fallthrough conditions.
4118 static int ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
4120 /* Start with base case */
4121 adapter->num_rx_queues = 1;
4122 adapter->num_tx_queues = 1;
4123 adapter->num_rx_pools = adapter->num_rx_queues;
4124 adapter->num_rx_queues_per_pool = 1;
4126 if (ixgbe_set_sriov_queues(adapter))
4130 if (ixgbe_set_fcoe_queues(adapter))
4133 #endif /* IXGBE_FCOE */
4134 #ifdef CONFIG_IXGBE_DCB
4135 if (ixgbe_set_dcb_queues(adapter))
4139 if (ixgbe_set_fdir_queues(adapter))
4142 if (ixgbe_set_rss_queues(adapter))
4145 /* fallback to base case */
4146 adapter->num_rx_queues = 1;
4147 adapter->num_tx_queues = 1;
4150 /* Notify the stack of the (possibly) reduced queue counts. */
4151 netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
4152 return netif_set_real_num_rx_queues(adapter->netdev,
4153 adapter->num_rx_queues);
4156 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
4159 int err, vector_threshold;
4161 /* We'll want at least 3 (vector_threshold):
4164 * 3) Other (Link Status Change, etc.)
4165 * 4) TCP Timer (optional)
4167 vector_threshold = MIN_MSIX_COUNT;
4169 /* The more we get, the more we will assign to Tx/Rx Cleanup
4170 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
4171 * Right now, we simply care about how many we'll get; we'll
4172 * set them up later while requesting irq's.
4174 while (vectors >= vector_threshold) {
4175 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
4177 if (!err) /* Success in acquiring all requested vectors. */
4180 vectors = 0; /* Nasty failure, quit now */
4181 else /* err == number of vectors we should try again with */
4185 if (vectors < vector_threshold) {
4186 /* Can't allocate enough MSI-X interrupts? Oh well.
4187 * This just means we'll go with either a single MSI
4188 * vector or fall back to legacy interrupts.
4190 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4191 "Unable to allocate MSI-X interrupts\n");
4192 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4193 kfree(adapter->msix_entries);
4194 adapter->msix_entries = NULL;
4196 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
4198 * Adjust for only the vectors we'll use, which is minimum
4199 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
4200 * vectors we were allocated.
4202 adapter->num_msix_vectors = min(vectors,
4203 adapter->max_msix_q_vectors + NON_Q_VECTORS);
4208 * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
4209 * @adapter: board private structure to initialize
4211 * Cache the descriptor ring offsets for RSS to the assigned rings.
4214 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
4219 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4220 for (i = 0; i < adapter->num_rx_queues; i++)
4221 adapter->rx_ring[i]->reg_idx = i;
4222 for (i = 0; i < adapter->num_tx_queues; i++)
4223 adapter->tx_ring[i]->reg_idx = i;
4232 #ifdef CONFIG_IXGBE_DCB
4234 * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
4235 * @adapter: board private structure to initialize
4237 * Cache the descriptor ring offsets for DCB to the assigned rings.
4240 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
4244 int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
4246 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4247 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
4248 /* the number of queues is assumed to be symmetric */
4249 for (i = 0; i < dcb_i; i++) {
4250 adapter->rx_ring[i]->reg_idx = i << 3;
4251 adapter->tx_ring[i]->reg_idx = i << 2;
4254 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
4257 * Tx TC0 starts at: descriptor queue 0
4258 * Tx TC1 starts at: descriptor queue 32
4259 * Tx TC2 starts at: descriptor queue 64
4260 * Tx TC3 starts at: descriptor queue 80
4261 * Tx TC4 starts at: descriptor queue 96
4262 * Tx TC5 starts at: descriptor queue 104
4263 * Tx TC6 starts at: descriptor queue 112
4264 * Tx TC7 starts at: descriptor queue 120
4266 * Rx TC0-TC7 are offset by 16 queues each
4268 for (i = 0; i < 3; i++) {
4269 adapter->tx_ring[i]->reg_idx = i << 5;
4270 adapter->rx_ring[i]->reg_idx = i << 4;
4272 for ( ; i < 5; i++) {
4273 adapter->tx_ring[i]->reg_idx =
4275 adapter->rx_ring[i]->reg_idx = i << 4;
4277 for ( ; i < dcb_i; i++) {
4278 adapter->tx_ring[i]->reg_idx =
4280 adapter->rx_ring[i]->reg_idx = i << 4;
4284 } else if (dcb_i == 4) {
4286 * Tx TC0 starts at: descriptor queue 0
4287 * Tx TC1 starts at: descriptor queue 64
4288 * Tx TC2 starts at: descriptor queue 96
4289 * Tx TC3 starts at: descriptor queue 112
4291 * Rx TC0-TC3 are offset by 32 queues each
4293 adapter->tx_ring[0]->reg_idx = 0;
4294 adapter->tx_ring[1]->reg_idx = 64;
4295 adapter->tx_ring[2]->reg_idx = 96;
4296 adapter->tx_ring[3]->reg_idx = 112;
4297 for (i = 0 ; i < dcb_i; i++)
4298 adapter->rx_ring[i]->reg_idx = i << 5;
4316 * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
4317 * @adapter: board private structure to initialize
4319 * Cache the descriptor ring offsets for Flow Director to the assigned rings.
4322 static inline bool ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
4327 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
4328 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4329 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))) {
4330 for (i = 0; i < adapter->num_rx_queues; i++)
4331 adapter->rx_ring[i]->reg_idx = i;
4332 for (i = 0; i < adapter->num_tx_queues; i++)
4333 adapter->tx_ring[i]->reg_idx = i;
4342 * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
4343 * @adapter: board private structure to initialize
4345 * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
4348 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
4350 int i, fcoe_rx_i = 0, fcoe_tx_i = 0;
4352 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4354 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
4355 #ifdef CONFIG_IXGBE_DCB
4356 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4357 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
4359 ixgbe_cache_ring_dcb(adapter);
4360 /* find out queues in TC for FCoE */
4361 fcoe_rx_i = adapter->rx_ring[fcoe->tc]->reg_idx + 1;
4362 fcoe_tx_i = adapter->tx_ring[fcoe->tc]->reg_idx + 1;
4364 * In 82599, the number of Tx queues for each traffic
4365 * class for both 8-TC and 4-TC modes are:
4366 * TCs : TC0 TC1 TC2 TC3 TC4 TC5 TC6 TC7
4367 * 8 TCs: 32 32 16 16 8 8 8 8
4368 * 4 TCs: 64 64 32 32
4369 * We have max 8 queues for FCoE, where 8 the is
4370 * FCoE redirection table size. If TC for FCoE is
4371 * less than or equal to TC3, we have enough queues
4372 * to add max of 8 queues for FCoE, so we start FCoE
4373 * tx descriptor from the next one, i.e., reg_idx + 1.
4374 * If TC for FCoE is above TC3, implying 8 TC mode,
4375 * and we need 8 for FCoE, we have to take all queues
4376 * in that traffic class for FCoE.
4378 if ((f->indices == IXGBE_FCRETA_SIZE) && (fcoe->tc > 3))
4381 #endif /* CONFIG_IXGBE_DCB */
4382 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4383 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4384 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
4385 ixgbe_cache_ring_fdir(adapter);
4387 ixgbe_cache_ring_rss(adapter);
4389 fcoe_rx_i = f->mask;
4390 fcoe_tx_i = f->mask;
4392 for (i = 0; i < f->indices; i++, fcoe_rx_i++, fcoe_tx_i++) {
4393 adapter->rx_ring[f->mask + i]->reg_idx = fcoe_rx_i;
4394 adapter->tx_ring[f->mask + i]->reg_idx = fcoe_tx_i;
4401 #endif /* IXGBE_FCOE */
4403 * ixgbe_cache_ring_sriov - Descriptor ring to register mapping for sriov
4404 * @adapter: board private structure to initialize
4406 * SR-IOV doesn't use any descriptor rings but changes the default if
4407 * no other mapping is used.
4410 static inline bool ixgbe_cache_ring_sriov(struct ixgbe_adapter *adapter)
4412 adapter->rx_ring[0]->reg_idx = adapter->num_vfs * 2;
4413 adapter->tx_ring[0]->reg_idx = adapter->num_vfs * 2;
4414 if (adapter->num_vfs)
4421 * ixgbe_cache_ring_register - Descriptor ring to register mapping
4422 * @adapter: board private structure to initialize
4424 * Once we know the feature-set enabled for the device, we'll cache
4425 * the register offset the descriptor ring is assigned to.
4427 * Note, the order the various feature calls is important. It must start with
4428 * the "most" features enabled at the same time, then trickle down to the
4429 * least amount of features turned on at once.
4431 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
4433 /* start with default case */
4434 adapter->rx_ring[0]->reg_idx = 0;
4435 adapter->tx_ring[0]->reg_idx = 0;
4437 if (ixgbe_cache_ring_sriov(adapter))
4441 if (ixgbe_cache_ring_fcoe(adapter))
4444 #endif /* IXGBE_FCOE */
4445 #ifdef CONFIG_IXGBE_DCB
4446 if (ixgbe_cache_ring_dcb(adapter))
4450 if (ixgbe_cache_ring_fdir(adapter))
4453 if (ixgbe_cache_ring_rss(adapter))
4458 * ixgbe_alloc_queues - Allocate memory for all rings
4459 * @adapter: board private structure to initialize
4461 * We allocate one ring per queue at run-time since we don't know the
4462 * number of queues at compile-time. The polling_netdev array is
4463 * intended for Multiqueue, but should work fine with a single queue.
4465 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
4469 int orig_node = adapter->node;
4471 for (i = 0; i < adapter->num_tx_queues; i++) {
4472 struct ixgbe_ring *ring = adapter->tx_ring[i];
4473 if (orig_node == -1) {
4474 int cur_node = next_online_node(adapter->node);
4475 if (cur_node == MAX_NUMNODES)
4476 cur_node = first_online_node;
4477 adapter->node = cur_node;
4479 ring = kzalloc_node(sizeof(struct ixgbe_ring), GFP_KERNEL,
4482 ring = kzalloc(sizeof(struct ixgbe_ring), GFP_KERNEL);
4484 goto err_tx_ring_allocation;
4485 ring->count = adapter->tx_ring_count;
4486 ring->queue_index = i;
4487 ring->dev = &adapter->pdev->dev;
4488 ring->netdev = adapter->netdev;
4489 ring->numa_node = adapter->node;
4491 adapter->tx_ring[i] = ring;
4494 /* Restore the adapter's original node */
4495 adapter->node = orig_node;
4497 rx_count = adapter->rx_ring_count;
4498 for (i = 0; i < adapter->num_rx_queues; i++) {
4499 struct ixgbe_ring *ring = adapter->rx_ring[i];
4500 if (orig_node == -1) {
4501 int cur_node = next_online_node(adapter->node);
4502 if (cur_node == MAX_NUMNODES)
4503 cur_node = first_online_node;
4504 adapter->node = cur_node;
4506 ring = kzalloc_node(sizeof(struct ixgbe_ring), GFP_KERNEL,
4509 ring = kzalloc(sizeof(struct ixgbe_ring), GFP_KERNEL);
4511 goto err_rx_ring_allocation;
4512 ring->count = rx_count;
4513 ring->queue_index = i;
4514 ring->dev = &adapter->pdev->dev;
4515 ring->netdev = adapter->netdev;
4516 ring->numa_node = adapter->node;
4518 adapter->rx_ring[i] = ring;
4521 /* Restore the adapter's original node */
4522 adapter->node = orig_node;
4524 ixgbe_cache_ring_register(adapter);
4528 err_rx_ring_allocation:
4529 for (i = 0; i < adapter->num_tx_queues; i++)
4530 kfree(adapter->tx_ring[i]);
4531 err_tx_ring_allocation:
4536 * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
4537 * @adapter: board private structure to initialize
4539 * Attempt to configure the interrupts using the best available
4540 * capabilities of the hardware and the kernel.
4542 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
4544 struct ixgbe_hw *hw = &adapter->hw;
4546 int vector, v_budget;
4549 * It's easy to be greedy for MSI-X vectors, but it really
4550 * doesn't do us much good if we have a lot more vectors
4551 * than CPU's. So let's be conservative and only ask for
4552 * (roughly) the same number of vectors as there are CPU's.
4554 v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
4555 (int)num_online_cpus()) + NON_Q_VECTORS;
4558 * At the same time, hardware can only support a maximum of
4559 * hw.mac->max_msix_vectors vectors. With features
4560 * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
4561 * descriptor queues supported by our device. Thus, we cap it off in
4562 * those rare cases where the cpu count also exceeds our vector limit.
4564 v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
4566 /* A failure in MSI-X entry allocation isn't fatal, but it does
4567 * mean we disable MSI-X capabilities of the adapter. */
4568 adapter->msix_entries = kcalloc(v_budget,
4569 sizeof(struct msix_entry), GFP_KERNEL);
4570 if (adapter->msix_entries) {
4571 for (vector = 0; vector < v_budget; vector++)
4572 adapter->msix_entries[vector].entry = vector;
4574 ixgbe_acquire_msix_vectors(adapter, v_budget);
4576 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4580 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
4581 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
4582 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4583 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4584 adapter->atr_sample_rate = 0;
4585 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4586 ixgbe_disable_sriov(adapter);
4588 err = ixgbe_set_num_queues(adapter);
4592 err = pci_enable_msi(adapter->pdev);
4594 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
4596 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4597 "Unable to allocate MSI interrupt, "
4598 "falling back to legacy. Error: %d\n", err);
4608 * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
4609 * @adapter: board private structure to initialize
4611 * We allocate one q_vector per queue interrupt. If allocation fails we
4614 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
4616 int q_idx, num_q_vectors;
4617 struct ixgbe_q_vector *q_vector;
4619 int (*poll)(struct napi_struct *, int);
4621 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4622 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4623 napi_vectors = adapter->num_rx_queues;
4624 poll = &ixgbe_clean_rxtx_many;
4631 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4632 q_vector = kzalloc_node(sizeof(struct ixgbe_q_vector),
4633 GFP_KERNEL, adapter->node);
4635 q_vector = kzalloc(sizeof(struct ixgbe_q_vector),
4639 q_vector->adapter = adapter;
4640 if (q_vector->txr_count && !q_vector->rxr_count)
4641 q_vector->eitr = adapter->tx_eitr_param;
4643 q_vector->eitr = adapter->rx_eitr_param;
4644 q_vector->v_idx = q_idx;
4645 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
4646 adapter->q_vector[q_idx] = q_vector;
4654 q_vector = adapter->q_vector[q_idx];
4655 netif_napi_del(&q_vector->napi);
4657 adapter->q_vector[q_idx] = NULL;
4663 * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
4664 * @adapter: board private structure to initialize
4666 * This function frees the memory allocated to the q_vectors. In addition if
4667 * NAPI is enabled it will delete any references to the NAPI struct prior
4668 * to freeing the q_vector.
4670 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
4672 int q_idx, num_q_vectors;
4674 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4675 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4679 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4680 struct ixgbe_q_vector *q_vector = adapter->q_vector[q_idx];
4681 adapter->q_vector[q_idx] = NULL;
4682 netif_napi_del(&q_vector->napi);
4687 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
4689 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4690 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4691 pci_disable_msix(adapter->pdev);
4692 kfree(adapter->msix_entries);
4693 adapter->msix_entries = NULL;
4694 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
4695 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
4696 pci_disable_msi(adapter->pdev);
4701 * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
4702 * @adapter: board private structure to initialize
4704 * We determine which interrupt scheme to use based on...
4705 * - Kernel support (MSI, MSI-X)
4706 * - which can be user-defined (via MODULE_PARAM)
4707 * - Hardware queue count (num_*_queues)
4708 * - defined by miscellaneous hardware support/features (RSS, etc.)
4710 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
4714 /* Number of supported queues */
4715 err = ixgbe_set_num_queues(adapter);
4719 err = ixgbe_set_interrupt_capability(adapter);
4721 e_dev_err("Unable to setup interrupt capabilities\n");
4722 goto err_set_interrupt;
4725 err = ixgbe_alloc_q_vectors(adapter);
4727 e_dev_err("Unable to allocate memory for queue vectors\n");
4728 goto err_alloc_q_vectors;
4731 err = ixgbe_alloc_queues(adapter);
4733 e_dev_err("Unable to allocate memory for queues\n");
4734 goto err_alloc_queues;
4737 e_dev_info("Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
4738 (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
4739 adapter->num_rx_queues, adapter->num_tx_queues);
4741 set_bit(__IXGBE_DOWN, &adapter->state);
4746 ixgbe_free_q_vectors(adapter);
4747 err_alloc_q_vectors:
4748 ixgbe_reset_interrupt_capability(adapter);
4753 static void ring_free_rcu(struct rcu_head *head)
4755 kfree(container_of(head, struct ixgbe_ring, rcu));
4759 * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
4760 * @adapter: board private structure to clear interrupt scheme on
4762 * We go through and clear interrupt specific resources and reset the structure
4763 * to pre-load conditions
4765 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
4769 for (i = 0; i < adapter->num_tx_queues; i++) {
4770 kfree(adapter->tx_ring[i]);
4771 adapter->tx_ring[i] = NULL;
4773 for (i = 0; i < adapter->num_rx_queues; i++) {
4774 struct ixgbe_ring *ring = adapter->rx_ring[i];
4776 /* ixgbe_get_stats64() might access this ring, we must wait
4777 * a grace period before freeing it.
4779 call_rcu(&ring->rcu, ring_free_rcu);
4780 adapter->rx_ring[i] = NULL;
4783 ixgbe_free_q_vectors(adapter);
4784 ixgbe_reset_interrupt_capability(adapter);
4788 * ixgbe_sfp_timer - worker thread to find a missing module
4789 * @data: pointer to our adapter struct
4791 static void ixgbe_sfp_timer(unsigned long data)
4793 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
4796 * Do the sfp_timer outside of interrupt context due to the
4797 * delays that sfp+ detection requires
4799 schedule_work(&adapter->sfp_task);
4803 * ixgbe_sfp_task - worker thread to find a missing module
4804 * @work: pointer to work_struct containing our data
4806 static void ixgbe_sfp_task(struct work_struct *work)
4808 struct ixgbe_adapter *adapter = container_of(work,
4809 struct ixgbe_adapter,
4811 struct ixgbe_hw *hw = &adapter->hw;
4813 if ((hw->phy.type == ixgbe_phy_nl) &&
4814 (hw->phy.sfp_type == ixgbe_sfp_type_not_present)) {
4815 s32 ret = hw->phy.ops.identify_sfp(hw);
4816 if (ret == IXGBE_ERR_SFP_NOT_PRESENT)
4818 ret = hw->phy.ops.reset(hw);
4819 if (ret == IXGBE_ERR_SFP_NOT_SUPPORTED) {
4820 e_dev_err("failed to initialize because an unsupported "
4821 "SFP+ module type was detected.\n");
4822 e_dev_err("Reload the driver after installing a "
4823 "supported module.\n");
4824 unregister_netdev(adapter->netdev);
4826 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
4828 /* don't need this routine any more */
4829 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
4833 if (test_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state))
4834 mod_timer(&adapter->sfp_timer,
4835 round_jiffies(jiffies + (2 * HZ)));
4839 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4840 * @adapter: board private structure to initialize
4842 * ixgbe_sw_init initializes the Adapter private data structure.
4843 * Fields are initialized based on PCI device information and
4844 * OS network device settings (MTU size).
4846 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
4848 struct ixgbe_hw *hw = &adapter->hw;
4849 struct pci_dev *pdev = adapter->pdev;
4850 struct net_device *dev = adapter->netdev;
4852 #ifdef CONFIG_IXGBE_DCB
4854 struct tc_configuration *tc;
4856 int max_frame = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
4858 /* PCI config space info */
4860 hw->vendor_id = pdev->vendor;
4861 hw->device_id = pdev->device;
4862 hw->revision_id = pdev->revision;
4863 hw->subsystem_vendor_id = pdev->subsystem_vendor;
4864 hw->subsystem_device_id = pdev->subsystem_device;
4866 /* Set capability flags */
4867 rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
4868 adapter->ring_feature[RING_F_RSS].indices = rss;
4869 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
4870 adapter->ring_feature[RING_F_DCB].indices = IXGBE_MAX_DCB_INDICES;
4871 if (hw->mac.type == ixgbe_mac_82598EB) {
4872 if (hw->device_id == IXGBE_DEV_ID_82598AT)
4873 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4874 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
4875 } else if (hw->mac.type == ixgbe_mac_82599EB) {
4876 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
4877 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4878 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4879 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
4880 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4881 if (dev->features & NETIF_F_NTUPLE) {
4882 /* Flow Director perfect filter enabled */
4883 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4884 adapter->atr_sample_rate = 0;
4885 spin_lock_init(&adapter->fdir_perfect_lock);
4887 /* Flow Director hash filters enabled */
4888 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
4889 adapter->atr_sample_rate = 20;
4891 adapter->ring_feature[RING_F_FDIR].indices =
4892 IXGBE_MAX_FDIR_INDICES;
4893 adapter->fdir_pballoc = 0;
4895 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4896 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4897 adapter->ring_feature[RING_F_FCOE].indices = 0;
4898 #ifdef CONFIG_IXGBE_DCB
4899 /* Default traffic class to use for FCoE */
4900 adapter->fcoe.tc = IXGBE_FCOE_DEFTC;
4901 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
4903 #endif /* IXGBE_FCOE */
4906 #ifdef CONFIG_IXGBE_DCB
4907 /* Configure DCB traffic classes */
4908 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
4909 tc = &adapter->dcb_cfg.tc_config[j];
4910 tc->path[DCB_TX_CONFIG].bwg_id = 0;
4911 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
4912 tc->path[DCB_RX_CONFIG].bwg_id = 0;
4913 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
4914 tc->dcb_pfc = pfc_disabled;
4916 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
4917 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
4918 adapter->dcb_cfg.rx_pba_cfg = pba_equal;
4919 adapter->dcb_cfg.pfc_mode_enable = false;
4920 adapter->dcb_cfg.round_robin_enable = false;
4921 adapter->dcb_set_bitmap = 0x00;
4922 ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
4923 adapter->ring_feature[RING_F_DCB].indices);
4927 /* default flow control settings */
4928 hw->fc.requested_mode = ixgbe_fc_full;
4929 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
4931 adapter->last_lfc_mode = hw->fc.current_mode;
4933 hw->fc.high_water = FC_HIGH_WATER(max_frame);
4934 hw->fc.low_water = FC_LOW_WATER(max_frame);
4935 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
4936 hw->fc.send_xon = true;
4937 hw->fc.disable_fc_autoneg = false;
4939 /* enable itr by default in dynamic mode */
4940 adapter->rx_itr_setting = 1;
4941 adapter->rx_eitr_param = 20000;
4942 adapter->tx_itr_setting = 1;
4943 adapter->tx_eitr_param = 10000;
4945 /* set defaults for eitr in MegaBytes */
4946 adapter->eitr_low = 10;
4947 adapter->eitr_high = 20;
4949 /* set default ring sizes */
4950 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
4951 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
4953 /* initialize eeprom parameters */
4954 if (ixgbe_init_eeprom_params_generic(hw)) {
4955 e_dev_err("EEPROM initialization failed\n");
4959 /* enable rx csum by default */
4960 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
4962 /* get assigned NUMA node */
4963 adapter->node = dev_to_node(&pdev->dev);
4965 set_bit(__IXGBE_DOWN, &adapter->state);
4971 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
4972 * @tx_ring: tx descriptor ring (for a specific queue) to setup
4974 * Return 0 on success, negative on failure
4976 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
4978 struct device *dev = tx_ring->dev;
4981 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4982 tx_ring->tx_buffer_info = vmalloc_node(size, tx_ring->numa_node);
4983 if (!tx_ring->tx_buffer_info)
4984 tx_ring->tx_buffer_info = vmalloc(size);
4985 if (!tx_ring->tx_buffer_info)
4987 memset(tx_ring->tx_buffer_info, 0, size);
4989 /* round up to nearest 4K */
4990 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
4991 tx_ring->size = ALIGN(tx_ring->size, 4096);
4993 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
4994 &tx_ring->dma, GFP_KERNEL);
4998 tx_ring->next_to_use = 0;
4999 tx_ring->next_to_clean = 0;
5000 tx_ring->work_limit = tx_ring->count;
5004 vfree(tx_ring->tx_buffer_info);
5005 tx_ring->tx_buffer_info = NULL;
5006 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
5011 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
5012 * @adapter: board private structure
5014 * If this function returns with an error, then it's possible one or
5015 * more of the rings is populated (while the rest are not). It is the
5016 * callers duty to clean those orphaned rings.
5018 * Return 0 on success, negative on failure
5020 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
5024 for (i = 0; i < adapter->num_tx_queues; i++) {
5025 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
5028 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
5036 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
5037 * @rx_ring: rx descriptor ring (for a specific queue) to setup
5039 * Returns 0 on success, negative on failure
5041 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
5043 struct device *dev = rx_ring->dev;
5046 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
5047 rx_ring->rx_buffer_info = vmalloc_node(size, rx_ring->numa_node);
5048 if (!rx_ring->rx_buffer_info)
5049 rx_ring->rx_buffer_info = vmalloc(size);
5050 if (!rx_ring->rx_buffer_info)
5052 memset(rx_ring->rx_buffer_info, 0, size);
5054 /* Round up to nearest 4K */
5055 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
5056 rx_ring->size = ALIGN(rx_ring->size, 4096);
5058 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
5059 &rx_ring->dma, GFP_KERNEL);
5064 rx_ring->next_to_clean = 0;
5065 rx_ring->next_to_use = 0;
5069 vfree(rx_ring->rx_buffer_info);
5070 rx_ring->rx_buffer_info = NULL;
5071 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
5076 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
5077 * @adapter: board private structure
5079 * If this function returns with an error, then it's possible one or
5080 * more of the rings is populated (while the rest are not). It is the
5081 * callers duty to clean those orphaned rings.
5083 * Return 0 on success, negative on failure
5085 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
5089 for (i = 0; i < adapter->num_rx_queues; i++) {
5090 err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
5093 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
5101 * ixgbe_free_tx_resources - Free Tx Resources per Queue
5102 * @tx_ring: Tx descriptor ring for a specific queue
5104 * Free all transmit software resources
5106 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
5108 ixgbe_clean_tx_ring(tx_ring);
5110 vfree(tx_ring->tx_buffer_info);
5111 tx_ring->tx_buffer_info = NULL;
5113 /* if not set, then don't free */
5117 dma_free_coherent(tx_ring->dev, tx_ring->size,
5118 tx_ring->desc, tx_ring->dma);
5120 tx_ring->desc = NULL;
5124 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5125 * @adapter: board private structure
5127 * Free all transmit software resources
5129 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
5133 for (i = 0; i < adapter->num_tx_queues; i++)
5134 if (adapter->tx_ring[i]->desc)
5135 ixgbe_free_tx_resources(adapter->tx_ring[i]);
5139 * ixgbe_free_rx_resources - Free Rx Resources
5140 * @rx_ring: ring to clean the resources from
5142 * Free all receive software resources
5144 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
5146 ixgbe_clean_rx_ring(rx_ring);
5148 vfree(rx_ring->rx_buffer_info);
5149 rx_ring->rx_buffer_info = NULL;
5151 /* if not set, then don't free */
5155 dma_free_coherent(rx_ring->dev, rx_ring->size,
5156 rx_ring->desc, rx_ring->dma);
5158 rx_ring->desc = NULL;
5162 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5163 * @adapter: board private structure
5165 * Free all receive software resources
5167 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5171 for (i = 0; i < adapter->num_rx_queues; i++)
5172 if (adapter->rx_ring[i]->desc)
5173 ixgbe_free_rx_resources(adapter->rx_ring[i]);
5177 * ixgbe_change_mtu - Change the Maximum Transfer Unit
5178 * @netdev: network interface device structure
5179 * @new_mtu: new value for maximum frame size
5181 * Returns 0 on success, negative on failure
5183 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5185 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5186 struct ixgbe_hw *hw = &adapter->hw;
5187 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5189 /* MTU < 68 is an error and causes problems on some kernels */
5190 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5193 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5194 /* must set new MTU before calling down or up */
5195 netdev->mtu = new_mtu;
5197 hw->fc.high_water = FC_HIGH_WATER(max_frame);
5198 hw->fc.low_water = FC_LOW_WATER(max_frame);
5200 if (netif_running(netdev))
5201 ixgbe_reinit_locked(adapter);
5207 * ixgbe_open - Called when a network interface is made active
5208 * @netdev: network interface device structure
5210 * Returns 0 on success, negative value on failure
5212 * The open entry point is called when a network interface is made
5213 * active by the system (IFF_UP). At this point all resources needed
5214 * for transmit and receive operations are allocated, the interrupt
5215 * handler is registered with the OS, the watchdog timer is started,
5216 * and the stack is notified that the interface is ready.
5218 static int ixgbe_open(struct net_device *netdev)
5220 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5223 /* disallow open during test */
5224 if (test_bit(__IXGBE_TESTING, &adapter->state))
5227 netif_carrier_off(netdev);
5229 /* allocate transmit descriptors */
5230 err = ixgbe_setup_all_tx_resources(adapter);
5234 /* allocate receive descriptors */
5235 err = ixgbe_setup_all_rx_resources(adapter);
5239 ixgbe_configure(adapter);
5241 err = ixgbe_request_irq(adapter);
5245 err = ixgbe_up_complete(adapter);
5249 netif_tx_start_all_queues(netdev);
5254 ixgbe_release_hw_control(adapter);
5255 ixgbe_free_irq(adapter);
5258 ixgbe_free_all_rx_resources(adapter);
5260 ixgbe_free_all_tx_resources(adapter);
5261 ixgbe_reset(adapter);
5267 * ixgbe_close - Disables a network interface
5268 * @netdev: network interface device structure
5270 * Returns 0, this is not allowed to fail
5272 * The close entry point is called when an interface is de-activated
5273 * by the OS. The hardware is still under the drivers control, but
5274 * needs to be disabled. A global MAC reset is issued to stop the
5275 * hardware, and all transmit and receive resources are freed.
5277 static int ixgbe_close(struct net_device *netdev)
5279 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5281 ixgbe_down(adapter);
5282 ixgbe_free_irq(adapter);
5284 ixgbe_free_all_tx_resources(adapter);
5285 ixgbe_free_all_rx_resources(adapter);
5287 ixgbe_release_hw_control(adapter);
5293 static int ixgbe_resume(struct pci_dev *pdev)
5295 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5296 struct net_device *netdev = adapter->netdev;
5299 pci_set_power_state(pdev, PCI_D0);
5300 pci_restore_state(pdev);
5302 * pci_restore_state clears dev->state_saved so call
5303 * pci_save_state to restore it.
5305 pci_save_state(pdev);
5307 err = pci_enable_device_mem(pdev);
5309 e_dev_err("Cannot enable PCI device from suspend\n");
5312 pci_set_master(pdev);
5314 pci_wake_from_d3(pdev, false);
5316 err = ixgbe_init_interrupt_scheme(adapter);
5318 e_dev_err("Cannot initialize interrupts for device\n");
5322 ixgbe_reset(adapter);
5324 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5326 if (netif_running(netdev)) {
5327 err = ixgbe_open(netdev);
5332 netif_device_attach(netdev);
5336 #endif /* CONFIG_PM */
5338 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5340 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5341 struct net_device *netdev = adapter->netdev;
5342 struct ixgbe_hw *hw = &adapter->hw;
5344 u32 wufc = adapter->wol;
5349 netif_device_detach(netdev);
5351 if (netif_running(netdev)) {
5352 ixgbe_down(adapter);
5353 ixgbe_free_irq(adapter);
5354 ixgbe_free_all_tx_resources(adapter);
5355 ixgbe_free_all_rx_resources(adapter);
5358 ixgbe_clear_interrupt_scheme(adapter);
5361 retval = pci_save_state(pdev);
5367 ixgbe_set_rx_mode(netdev);
5369 /* turn on all-multi mode if wake on multicast is enabled */
5370 if (wufc & IXGBE_WUFC_MC) {
5371 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5372 fctrl |= IXGBE_FCTRL_MPE;
5373 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5376 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5377 ctrl |= IXGBE_CTRL_GIO_DIS;
5378 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5380 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5382 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5383 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5386 if (wufc && hw->mac.type == ixgbe_mac_82599EB)
5387 pci_wake_from_d3(pdev, true);
5389 pci_wake_from_d3(pdev, false);
5391 *enable_wake = !!wufc;
5393 ixgbe_release_hw_control(adapter);
5395 pci_disable_device(pdev);
5401 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5406 retval = __ixgbe_shutdown(pdev, &wake);
5411 pci_prepare_to_sleep(pdev);
5413 pci_wake_from_d3(pdev, false);
5414 pci_set_power_state(pdev, PCI_D3hot);
5419 #endif /* CONFIG_PM */
5421 static void ixgbe_shutdown(struct pci_dev *pdev)
5425 __ixgbe_shutdown(pdev, &wake);
5427 if (system_state == SYSTEM_POWER_OFF) {
5428 pci_wake_from_d3(pdev, wake);
5429 pci_set_power_state(pdev, PCI_D3hot);
5434 * ixgbe_update_stats - Update the board statistics counters.
5435 * @adapter: board private structure
5437 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5439 struct net_device *netdev = adapter->netdev;
5440 struct ixgbe_hw *hw = &adapter->hw;
5441 struct ixgbe_hw_stats *hwstats = &adapter->stats;
5443 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5444 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
5445 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
5446 u64 bytes = 0, packets = 0;
5448 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5449 test_bit(__IXGBE_RESETTING, &adapter->state))
5452 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5455 for (i = 0; i < 16; i++)
5456 adapter->hw_rx_no_dma_resources +=
5457 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5458 for (i = 0; i < adapter->num_rx_queues; i++) {
5459 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
5460 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
5462 adapter->rsc_total_count = rsc_count;
5463 adapter->rsc_total_flush = rsc_flush;
5466 for (i = 0; i < adapter->num_rx_queues; i++) {
5467 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
5468 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
5469 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
5470 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
5471 bytes += rx_ring->stats.bytes;
5472 packets += rx_ring->stats.packets;
5474 adapter->non_eop_descs = non_eop_descs;
5475 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5476 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5477 netdev->stats.rx_bytes = bytes;
5478 netdev->stats.rx_packets = packets;
5482 /* gather some stats to the adapter struct that are per queue */
5483 for (i = 0; i < adapter->num_tx_queues; i++) {
5484 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5485 restart_queue += tx_ring->tx_stats.restart_queue;
5486 tx_busy += tx_ring->tx_stats.tx_busy;
5487 bytes += tx_ring->stats.bytes;
5488 packets += tx_ring->stats.packets;
5490 adapter->restart_queue = restart_queue;
5491 adapter->tx_busy = tx_busy;
5492 netdev->stats.tx_bytes = bytes;
5493 netdev->stats.tx_packets = packets;
5495 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5496 for (i = 0; i < 8; i++) {
5497 /* for packet buffers not used, the register should read 0 */
5498 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5500 hwstats->mpc[i] += mpc;
5501 total_mpc += hwstats->mpc[i];
5502 if (hw->mac.type == ixgbe_mac_82598EB)
5503 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5504 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5505 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5506 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5507 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5508 if (hw->mac.type == ixgbe_mac_82599EB) {
5509 hwstats->pxonrxc[i] +=
5510 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5511 hwstats->pxoffrxc[i] +=
5512 IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
5513 hwstats->qprdc[i] += IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5515 hwstats->pxonrxc[i] +=
5516 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5517 hwstats->pxoffrxc[i] +=
5518 IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
5520 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5521 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5523 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5524 /* work around hardware counting issue */
5525 hwstats->gprc -= missed_rx;
5527 /* 82598 hardware only has a 32 bit counter in the high register */
5528 if (hw->mac.type == ixgbe_mac_82599EB) {
5530 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5531 tmp = IXGBE_READ_REG(hw, IXGBE_GORCH) & 0xF;
5532 /* 4 high bits of GORC */
5533 hwstats->gorc += (tmp << 32);
5534 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5535 tmp = IXGBE_READ_REG(hw, IXGBE_GOTCH) & 0xF;
5536 /* 4 high bits of GOTC */
5537 hwstats->gotc += (tmp << 32);
5538 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5539 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5540 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5541 hwstats->lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
5542 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5543 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5545 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5546 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5547 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5548 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5549 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5550 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5551 #endif /* IXGBE_FCOE */
5553 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5554 hwstats->lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
5555 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5556 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5557 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5559 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5560 hwstats->bprc += bprc;
5561 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5562 if (hw->mac.type == ixgbe_mac_82598EB)
5563 hwstats->mprc -= bprc;
5564 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5565 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5566 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5567 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5568 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5569 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5570 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5571 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5572 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5573 hwstats->lxontxc += lxon;
5574 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5575 hwstats->lxofftxc += lxoff;
5576 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5577 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5578 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5580 * 82598 errata - tx of flow control packets is included in tx counters
5582 xon_off_tot = lxon + lxoff;
5583 hwstats->gptc -= xon_off_tot;
5584 hwstats->mptc -= xon_off_tot;
5585 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5586 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5587 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5588 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5589 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5590 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5591 hwstats->ptc64 -= xon_off_tot;
5592 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5593 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5594 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5595 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5596 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5597 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5599 /* Fill out the OS statistics structure */
5600 netdev->stats.multicast = hwstats->mprc;
5603 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
5604 netdev->stats.rx_dropped = 0;
5605 netdev->stats.rx_length_errors = hwstats->rlec;
5606 netdev->stats.rx_crc_errors = hwstats->crcerrs;
5607 netdev->stats.rx_missed_errors = total_mpc;
5611 * ixgbe_watchdog - Timer Call-back
5612 * @data: pointer to adapter cast into an unsigned long
5614 static void ixgbe_watchdog(unsigned long data)
5616 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5617 struct ixgbe_hw *hw = &adapter->hw;
5622 * Do the watchdog outside of interrupt context due to the lovely
5623 * delays that some of the newer hardware requires
5626 if (test_bit(__IXGBE_DOWN, &adapter->state))
5627 goto watchdog_short_circuit;
5629 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5631 * for legacy and MSI interrupts don't set any bits
5632 * that are enabled for EIAM, because this operation
5633 * would set *both* EIMS and EICS for any bit in EIAM
5635 IXGBE_WRITE_REG(hw, IXGBE_EICS,
5636 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5637 goto watchdog_reschedule;
5640 /* get one bit for every active tx/rx interrupt vector */
5641 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
5642 struct ixgbe_q_vector *qv = adapter->q_vector[i];
5643 if (qv->rxr_count || qv->txr_count)
5644 eics |= ((u64)1 << i);
5647 /* Cause software interrupt to ensure rx rings are cleaned */
5648 ixgbe_irq_rearm_queues(adapter, eics);
5650 watchdog_reschedule:
5651 /* Reset the timer */
5652 mod_timer(&adapter->watchdog_timer, round_jiffies(jiffies + 2 * HZ));
5654 watchdog_short_circuit:
5655 schedule_work(&adapter->watchdog_task);
5659 * ixgbe_multispeed_fiber_task - worker thread to configure multispeed fiber
5660 * @work: pointer to work_struct containing our data
5662 static void ixgbe_multispeed_fiber_task(struct work_struct *work)
5664 struct ixgbe_adapter *adapter = container_of(work,
5665 struct ixgbe_adapter,
5666 multispeed_fiber_task);
5667 struct ixgbe_hw *hw = &adapter->hw;
5671 adapter->flags |= IXGBE_FLAG_IN_SFP_LINK_TASK;
5672 autoneg = hw->phy.autoneg_advertised;
5673 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
5674 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
5675 hw->mac.autotry_restart = false;
5676 if (hw->mac.ops.setup_link)
5677 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
5678 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5679 adapter->flags &= ~IXGBE_FLAG_IN_SFP_LINK_TASK;
5683 * ixgbe_sfp_config_module_task - worker thread to configure a new SFP+ module
5684 * @work: pointer to work_struct containing our data
5686 static void ixgbe_sfp_config_module_task(struct work_struct *work)
5688 struct ixgbe_adapter *adapter = container_of(work,
5689 struct ixgbe_adapter,
5690 sfp_config_module_task);
5691 struct ixgbe_hw *hw = &adapter->hw;
5694 adapter->flags |= IXGBE_FLAG_IN_SFP_MOD_TASK;
5696 /* Time for electrical oscillations to settle down */
5698 err = hw->phy.ops.identify_sfp(hw);
5700 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
5701 e_dev_err("failed to initialize because an unsupported SFP+ "
5702 "module type was detected.\n");
5703 e_dev_err("Reload the driver after installing a supported "
5705 unregister_netdev(adapter->netdev);
5708 hw->mac.ops.setup_sfp(hw);
5710 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
5711 /* This will also work for DA Twinax connections */
5712 schedule_work(&adapter->multispeed_fiber_task);
5713 adapter->flags &= ~IXGBE_FLAG_IN_SFP_MOD_TASK;
5717 * ixgbe_fdir_reinit_task - worker thread to reinit FDIR filter table
5718 * @work: pointer to work_struct containing our data
5720 static void ixgbe_fdir_reinit_task(struct work_struct *work)
5722 struct ixgbe_adapter *adapter = container_of(work,
5723 struct ixgbe_adapter,
5725 struct ixgbe_hw *hw = &adapter->hw;
5728 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5729 for (i = 0; i < adapter->num_tx_queues; i++)
5730 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
5731 &(adapter->tx_ring[i]->state));
5733 e_err(probe, "failed to finish FDIR re-initialization, "
5734 "ignored adding FDIR ATR filters\n");
5736 /* Done FDIR Re-initialization, enable transmits */
5737 netif_tx_start_all_queues(adapter->netdev);
5740 static DEFINE_MUTEX(ixgbe_watchdog_lock);
5743 * ixgbe_watchdog_task - worker thread to bring link up
5744 * @work: pointer to work_struct containing our data
5746 static void ixgbe_watchdog_task(struct work_struct *work)
5748 struct ixgbe_adapter *adapter = container_of(work,
5749 struct ixgbe_adapter,
5751 struct net_device *netdev = adapter->netdev;
5752 struct ixgbe_hw *hw = &adapter->hw;
5756 struct ixgbe_ring *tx_ring;
5757 int some_tx_pending = 0;
5759 mutex_lock(&ixgbe_watchdog_lock);
5761 link_up = adapter->link_up;
5762 link_speed = adapter->link_speed;
5764 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
5765 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5768 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5769 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
5770 hw->mac.ops.fc_enable(hw, i);
5772 hw->mac.ops.fc_enable(hw, 0);
5775 hw->mac.ops.fc_enable(hw, 0);
5780 time_after(jiffies, (adapter->link_check_timeout +
5781 IXGBE_TRY_LINK_TIMEOUT))) {
5782 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5783 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5785 adapter->link_up = link_up;
5786 adapter->link_speed = link_speed;
5790 if (!netif_carrier_ok(netdev)) {
5791 bool flow_rx, flow_tx;
5793 if (hw->mac.type == ixgbe_mac_82599EB) {
5794 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5795 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5796 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5797 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5799 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5800 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5801 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5802 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5805 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
5806 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5808 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5809 "1 Gbps" : "unknown speed")),
5810 ((flow_rx && flow_tx) ? "RX/TX" :
5812 (flow_tx ? "TX" : "None"))));
5814 netif_carrier_on(netdev);
5816 /* Force detection of hung controller */
5817 for (i = 0; i < adapter->num_tx_queues; i++) {
5818 tx_ring = adapter->tx_ring[i];
5819 set_check_for_tx_hang(tx_ring);
5823 adapter->link_up = false;
5824 adapter->link_speed = 0;
5825 if (netif_carrier_ok(netdev)) {
5826 e_info(drv, "NIC Link is Down\n");
5827 netif_carrier_off(netdev);
5831 if (!netif_carrier_ok(netdev)) {
5832 for (i = 0; i < adapter->num_tx_queues; i++) {
5833 tx_ring = adapter->tx_ring[i];
5834 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5835 some_tx_pending = 1;
5840 if (some_tx_pending) {
5841 /* We've lost link, so the controller stops DMA,
5842 * but we've got queued Tx work that's never going
5843 * to get done, so reset controller to flush Tx.
5844 * (Do the reset outside of interrupt context).
5846 schedule_work(&adapter->reset_task);
5850 ixgbe_update_stats(adapter);
5851 mutex_unlock(&ixgbe_watchdog_lock);
5854 static int ixgbe_tso(struct ixgbe_adapter *adapter,
5855 struct ixgbe_ring *tx_ring, struct sk_buff *skb,
5856 u32 tx_flags, u8 *hdr_len, __be16 protocol)
5858 struct ixgbe_adv_tx_context_desc *context_desc;
5861 struct ixgbe_tx_buffer *tx_buffer_info;
5862 u32 vlan_macip_lens = 0, type_tucmd_mlhl;
5863 u32 mss_l4len_idx, l4len;
5865 if (skb_is_gso(skb)) {
5866 if (skb_header_cloned(skb)) {
5867 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5871 l4len = tcp_hdrlen(skb);
5874 if (protocol == htons(ETH_P_IP)) {
5875 struct iphdr *iph = ip_hdr(skb);
5878 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
5882 } else if (skb_is_gso_v6(skb)) {
5883 ipv6_hdr(skb)->payload_len = 0;
5884 tcp_hdr(skb)->check =
5885 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5886 &ipv6_hdr(skb)->daddr,
5890 i = tx_ring->next_to_use;
5892 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5893 context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
5895 /* VLAN MACLEN IPLEN */
5896 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5898 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
5899 vlan_macip_lens |= ((skb_network_offset(skb)) <<
5900 IXGBE_ADVTXD_MACLEN_SHIFT);
5901 *hdr_len += skb_network_offset(skb);
5903 (skb_transport_header(skb) - skb_network_header(skb));
5905 (skb_transport_header(skb) - skb_network_header(skb));
5906 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
5907 context_desc->seqnum_seed = 0;
5909 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
5910 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
5911 IXGBE_ADVTXD_DTYP_CTXT);
5913 if (protocol == htons(ETH_P_IP))
5914 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
5915 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5916 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
5920 (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
5921 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
5922 /* use index 1 for TSO */
5923 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
5924 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
5926 tx_buffer_info->time_stamp = jiffies;
5927 tx_buffer_info->next_to_watch = i;
5930 if (i == tx_ring->count)
5932 tx_ring->next_to_use = i;
5939 static u32 ixgbe_psum(struct ixgbe_adapter *adapter, struct sk_buff *skb,
5945 case cpu_to_be16(ETH_P_IP):
5946 rtn |= IXGBE_ADVTXD_TUCMD_IPV4;
5947 switch (ip_hdr(skb)->protocol) {
5949 rtn |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5952 rtn |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5956 case cpu_to_be16(ETH_P_IPV6):
5957 /* XXX what about other V6 headers?? */
5958 switch (ipv6_hdr(skb)->nexthdr) {
5960 rtn |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5963 rtn |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5968 if (unlikely(net_ratelimit()))
5969 e_warn(probe, "partial checksum but proto=%x!\n",
5977 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
5978 struct ixgbe_ring *tx_ring,
5979 struct sk_buff *skb, u32 tx_flags,
5982 struct ixgbe_adv_tx_context_desc *context_desc;
5984 struct ixgbe_tx_buffer *tx_buffer_info;
5985 u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
5987 if (skb->ip_summed == CHECKSUM_PARTIAL ||
5988 (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
5989 i = tx_ring->next_to_use;
5990 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5991 context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
5993 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5995 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
5996 vlan_macip_lens |= (skb_network_offset(skb) <<
5997 IXGBE_ADVTXD_MACLEN_SHIFT);
5998 if (skb->ip_summed == CHECKSUM_PARTIAL)
5999 vlan_macip_lens |= (skb_transport_header(skb) -
6000 skb_network_header(skb));
6002 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
6003 context_desc->seqnum_seed = 0;
6005 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
6006 IXGBE_ADVTXD_DTYP_CTXT);
6008 if (skb->ip_summed == CHECKSUM_PARTIAL)
6009 type_tucmd_mlhl |= ixgbe_psum(adapter, skb, protocol);
6011 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
6012 /* use index zero for tx checksum offload */
6013 context_desc->mss_l4len_idx = 0;
6015 tx_buffer_info->time_stamp = jiffies;
6016 tx_buffer_info->next_to_watch = i;
6019 if (i == tx_ring->count)
6021 tx_ring->next_to_use = i;
6029 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
6030 struct ixgbe_ring *tx_ring,
6031 struct sk_buff *skb, u32 tx_flags,
6032 unsigned int first, const u8 hdr_len)
6034 struct device *dev = tx_ring->dev;
6035 struct ixgbe_tx_buffer *tx_buffer_info;
6037 unsigned int total = skb->len;
6038 unsigned int offset = 0, size, count = 0, i;
6039 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
6041 unsigned int bytecount = skb->len;
6044 i = tx_ring->next_to_use;
6046 if (tx_flags & IXGBE_TX_FLAGS_FCOE)
6047 /* excluding fcoe_crc_eof for FCoE */
6048 total -= sizeof(struct fcoe_crc_eof);
6050 len = min(skb_headlen(skb), total);
6052 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6053 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
6055 tx_buffer_info->length = size;
6056 tx_buffer_info->mapped_as_page = false;
6057 tx_buffer_info->dma = dma_map_single(dev,
6059 size, DMA_TO_DEVICE);
6060 if (dma_mapping_error(dev, tx_buffer_info->dma))
6062 tx_buffer_info->time_stamp = jiffies;
6063 tx_buffer_info->next_to_watch = i;
6072 if (i == tx_ring->count)
6077 for (f = 0; f < nr_frags; f++) {
6078 struct skb_frag_struct *frag;
6080 frag = &skb_shinfo(skb)->frags[f];
6081 len = min((unsigned int)frag->size, total);
6082 offset = frag->page_offset;
6086 if (i == tx_ring->count)
6089 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6090 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
6092 tx_buffer_info->length = size;
6093 tx_buffer_info->dma = dma_map_page(dev,
6097 tx_buffer_info->mapped_as_page = true;
6098 if (dma_mapping_error(dev, tx_buffer_info->dma))
6100 tx_buffer_info->time_stamp = jiffies;
6101 tx_buffer_info->next_to_watch = i;
6112 if (tx_flags & IXGBE_TX_FLAGS_TSO)
6113 gso_segs = skb_shinfo(skb)->gso_segs;
6115 /* adjust for FCoE Sequence Offload */
6116 else if (tx_flags & IXGBE_TX_FLAGS_FSO)
6117 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
6118 skb_shinfo(skb)->gso_size);
6119 #endif /* IXGBE_FCOE */
6120 bytecount += (gso_segs - 1) * hdr_len;
6122 /* multiply data chunks by size of headers */
6123 tx_ring->tx_buffer_info[i].bytecount = bytecount;
6124 tx_ring->tx_buffer_info[i].gso_segs = gso_segs;
6125 tx_ring->tx_buffer_info[i].skb = skb;
6126 tx_ring->tx_buffer_info[first].next_to_watch = i;
6131 e_dev_err("TX DMA map failed\n");
6133 /* clear timestamp and dma mappings for failed tx_buffer_info map */
6134 tx_buffer_info->dma = 0;
6135 tx_buffer_info->time_stamp = 0;
6136 tx_buffer_info->next_to_watch = 0;
6140 /* clear timestamp and dma mappings for remaining portion of packet */
6143 i += tx_ring->count;
6145 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6146 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
6152 static void ixgbe_tx_queue(struct ixgbe_ring *tx_ring,
6153 int tx_flags, int count, u32 paylen, u8 hdr_len)
6155 union ixgbe_adv_tx_desc *tx_desc = NULL;
6156 struct ixgbe_tx_buffer *tx_buffer_info;
6157 u32 olinfo_status = 0, cmd_type_len = 0;
6159 u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
6161 cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
6163 cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
6165 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
6166 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
6168 if (tx_flags & IXGBE_TX_FLAGS_TSO) {
6169 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
6171 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
6172 IXGBE_ADVTXD_POPTS_SHIFT;
6174 /* use index 1 context for tso */
6175 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6176 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
6177 olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
6178 IXGBE_ADVTXD_POPTS_SHIFT;
6180 } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
6181 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
6182 IXGBE_ADVTXD_POPTS_SHIFT;
6184 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6185 olinfo_status |= IXGBE_ADVTXD_CC;
6186 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6187 if (tx_flags & IXGBE_TX_FLAGS_FSO)
6188 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
6191 olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
6193 i = tx_ring->next_to_use;
6195 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6196 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
6197 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
6198 tx_desc->read.cmd_type_len =
6199 cpu_to_le32(cmd_type_len | tx_buffer_info->length);
6200 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
6202 if (i == tx_ring->count)
6206 tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
6209 * Force memory writes to complete before letting h/w
6210 * know there are new descriptors to fetch. (Only
6211 * applicable for weak-ordered memory model archs,
6216 tx_ring->next_to_use = i;
6217 writel(i, tx_ring->tail);
6220 static void ixgbe_atr(struct ixgbe_adapter *adapter, struct sk_buff *skb,
6221 int queue, u32 tx_flags, __be16 protocol)
6223 struct ixgbe_atr_input atr_input;
6225 struct iphdr *iph = ip_hdr(skb);
6226 struct ethhdr *eth = (struct ethhdr *)skb->data;
6227 u16 vlan_id, src_port, dst_port, flex_bytes;
6228 u32 src_ipv4_addr, dst_ipv4_addr;
6231 /* Right now, we support IPv4 only */
6232 if (protocol != htons(ETH_P_IP))
6234 /* check if we're UDP or TCP */
6235 if (iph->protocol == IPPROTO_TCP) {
6237 src_port = th->source;
6238 dst_port = th->dest;
6239 l4type |= IXGBE_ATR_L4TYPE_TCP;
6240 /* l4type IPv4 type is 0, no need to assign */
6242 /* Unsupported L4 header, just bail here */
6246 memset(&atr_input, 0, sizeof(struct ixgbe_atr_input));
6248 vlan_id = (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK) >>
6249 IXGBE_TX_FLAGS_VLAN_SHIFT;
6250 src_ipv4_addr = iph->saddr;
6251 dst_ipv4_addr = iph->daddr;
6252 flex_bytes = eth->h_proto;
6254 ixgbe_atr_set_vlan_id_82599(&atr_input, vlan_id);
6255 ixgbe_atr_set_src_port_82599(&atr_input, dst_port);
6256 ixgbe_atr_set_dst_port_82599(&atr_input, src_port);
6257 ixgbe_atr_set_flex_byte_82599(&atr_input, flex_bytes);
6258 ixgbe_atr_set_l4type_82599(&atr_input, l4type);
6259 /* src and dst are inverted, think how the receiver sees them */
6260 ixgbe_atr_set_src_ipv4_82599(&atr_input, dst_ipv4_addr);
6261 ixgbe_atr_set_dst_ipv4_82599(&atr_input, src_ipv4_addr);
6263 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
6264 ixgbe_fdir_add_signature_filter_82599(&adapter->hw, &atr_input, queue);
6267 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, int size)
6269 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
6270 /* Herbert's original patch had:
6271 * smp_mb__after_netif_stop_queue();
6272 * but since that doesn't exist yet, just open code it. */
6275 /* We need to check again in a case another CPU has just
6276 * made room available. */
6277 if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
6280 /* A reprieve! - use start_queue because it doesn't call schedule */
6281 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
6282 ++tx_ring->tx_stats.restart_queue;
6286 static int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, int size)
6288 if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
6290 return __ixgbe_maybe_stop_tx(tx_ring, size);
6293 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6295 struct ixgbe_adapter *adapter = netdev_priv(dev);
6296 int txq = smp_processor_id();
6300 protocol = vlan_get_protocol(skb);
6302 if ((protocol == htons(ETH_P_FCOE)) ||
6303 (protocol == htons(ETH_P_FIP))) {
6304 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
6305 txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
6306 txq += adapter->ring_feature[RING_F_FCOE].mask;
6308 #ifdef CONFIG_IXGBE_DCB
6309 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6310 txq = adapter->fcoe.up;
6317 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6318 while (unlikely(txq >= dev->real_num_tx_queues))
6319 txq -= dev->real_num_tx_queues;
6323 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6324 if (skb->priority == TC_PRIO_CONTROL)
6325 txq = adapter->ring_feature[RING_F_DCB].indices-1;
6327 txq = (skb->vlan_tci & IXGBE_TX_FLAGS_VLAN_PRIO_MASK)
6332 return skb_tx_hash(dev, skb);
6335 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
6336 struct ixgbe_adapter *adapter,
6337 struct ixgbe_ring *tx_ring)
6339 struct net_device *netdev = tx_ring->netdev;
6340 struct netdev_queue *txq;
6342 unsigned int tx_flags = 0;
6349 protocol = vlan_get_protocol(skb);
6351 if (vlan_tx_tag_present(skb)) {
6352 tx_flags |= vlan_tx_tag_get(skb);
6353 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6354 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6355 tx_flags |= ((skb->queue_mapping & 0x7) << 13);
6357 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6358 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6359 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED &&
6360 skb->priority != TC_PRIO_CONTROL) {
6361 tx_flags |= ((skb->queue_mapping & 0x7) << 13);
6362 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6363 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6367 /* for FCoE with DCB, we force the priority to what
6368 * was specified by the switch */
6369 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED &&
6370 (protocol == htons(ETH_P_FCOE) ||
6371 protocol == htons(ETH_P_FIP))) {
6372 #ifdef CONFIG_IXGBE_DCB
6373 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6374 tx_flags &= ~(IXGBE_TX_FLAGS_VLAN_PRIO_MASK
6375 << IXGBE_TX_FLAGS_VLAN_SHIFT);
6376 tx_flags |= ((adapter->fcoe.up << 13)
6377 << IXGBE_TX_FLAGS_VLAN_SHIFT);
6380 /* flag for FCoE offloads */
6381 if (protocol == htons(ETH_P_FCOE))
6382 tx_flags |= IXGBE_TX_FLAGS_FCOE;
6386 /* four things can cause us to need a context descriptor */
6387 if (skb_is_gso(skb) ||
6388 (skb->ip_summed == CHECKSUM_PARTIAL) ||
6389 (tx_flags & IXGBE_TX_FLAGS_VLAN) ||
6390 (tx_flags & IXGBE_TX_FLAGS_FCOE))
6393 count += TXD_USE_COUNT(skb_headlen(skb));
6394 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6395 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6397 if (ixgbe_maybe_stop_tx(tx_ring, count)) {
6398 tx_ring->tx_stats.tx_busy++;
6399 return NETDEV_TX_BUSY;
6402 first = tx_ring->next_to_use;
6403 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6405 /* setup tx offload for FCoE */
6406 tso = ixgbe_fso(adapter, tx_ring, skb, tx_flags, &hdr_len);
6408 dev_kfree_skb_any(skb);
6409 return NETDEV_TX_OK;
6412 tx_flags |= IXGBE_TX_FLAGS_FSO;
6413 #endif /* IXGBE_FCOE */
6415 if (protocol == htons(ETH_P_IP))
6416 tx_flags |= IXGBE_TX_FLAGS_IPV4;
6417 tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len,
6420 dev_kfree_skb_any(skb);
6421 return NETDEV_TX_OK;
6425 tx_flags |= IXGBE_TX_FLAGS_TSO;
6426 else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags,
6428 (skb->ip_summed == CHECKSUM_PARTIAL))
6429 tx_flags |= IXGBE_TX_FLAGS_CSUM;
6432 count = ixgbe_tx_map(adapter, tx_ring, skb, tx_flags, first, hdr_len);
6434 /* add the ATR filter if ATR is on */
6435 if (tx_ring->atr_sample_rate) {
6436 ++tx_ring->atr_count;
6437 if ((tx_ring->atr_count >= tx_ring->atr_sample_rate) &&
6438 test_bit(__IXGBE_TX_FDIR_INIT_DONE,
6440 ixgbe_atr(adapter, skb, tx_ring->queue_index,
6441 tx_flags, protocol);
6442 tx_ring->atr_count = 0;
6445 txq = netdev_get_tx_queue(netdev, tx_ring->queue_index);
6446 txq->tx_bytes += skb->len;
6448 ixgbe_tx_queue(tx_ring, tx_flags, count, skb->len, hdr_len);
6449 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
6452 dev_kfree_skb_any(skb);
6453 tx_ring->tx_buffer_info[first].time_stamp = 0;
6454 tx_ring->next_to_use = first;
6457 return NETDEV_TX_OK;
6460 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
6462 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6463 struct ixgbe_ring *tx_ring;
6465 tx_ring = adapter->tx_ring[skb->queue_mapping];
6466 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
6470 * ixgbe_set_mac - Change the Ethernet Address of the NIC
6471 * @netdev: network interface device structure
6472 * @p: pointer to an address structure
6474 * Returns 0 on success, negative on failure
6476 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6478 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6479 struct ixgbe_hw *hw = &adapter->hw;
6480 struct sockaddr *addr = p;
6482 if (!is_valid_ether_addr(addr->sa_data))
6483 return -EADDRNOTAVAIL;
6485 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6486 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6488 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
6495 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6497 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6498 struct ixgbe_hw *hw = &adapter->hw;
6502 if (prtad != hw->phy.mdio.prtad)
6504 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6510 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6511 u16 addr, u16 value)
6513 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6514 struct ixgbe_hw *hw = &adapter->hw;
6516 if (prtad != hw->phy.mdio.prtad)
6518 return hw->phy.ops.write_reg(hw, addr, devad, value);
6521 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6523 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6525 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6529 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6531 * @netdev: network interface device structure
6533 * Returns non-zero on failure
6535 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6538 struct ixgbe_adapter *adapter = netdev_priv(dev);
6539 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6541 if (is_valid_ether_addr(mac->san_addr)) {
6543 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6550 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6552 * @netdev: network interface device structure
6554 * Returns non-zero on failure
6556 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6559 struct ixgbe_adapter *adapter = netdev_priv(dev);
6560 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6562 if (is_valid_ether_addr(mac->san_addr)) {
6564 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6570 #ifdef CONFIG_NET_POLL_CONTROLLER
6572 * Polling 'interrupt' - used by things like netconsole to send skbs
6573 * without having to re-enable interrupts. It's not called while
6574 * the interrupt routine is executing.
6576 static void ixgbe_netpoll(struct net_device *netdev)
6578 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6581 /* if interface is down do nothing */
6582 if (test_bit(__IXGBE_DOWN, &adapter->state))
6585 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
6586 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
6587 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
6588 for (i = 0; i < num_q_vectors; i++) {
6589 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
6590 ixgbe_msix_clean_many(0, q_vector);
6593 ixgbe_intr(adapter->pdev->irq, netdev);
6595 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
6599 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
6600 struct rtnl_link_stats64 *stats)
6602 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6605 /* accurate rx/tx bytes/packets stats */
6606 dev_txq_stats_fold(netdev, stats);
6608 for (i = 0; i < adapter->num_rx_queues; i++) {
6609 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
6615 start = u64_stats_fetch_begin_bh(&ring->syncp);
6616 packets = ring->stats.packets;
6617 bytes = ring->stats.bytes;
6618 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6619 stats->rx_packets += packets;
6620 stats->rx_bytes += bytes;
6624 /* following stats updated by ixgbe_watchdog_task() */
6625 stats->multicast = netdev->stats.multicast;
6626 stats->rx_errors = netdev->stats.rx_errors;
6627 stats->rx_length_errors = netdev->stats.rx_length_errors;
6628 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
6629 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
6634 static const struct net_device_ops ixgbe_netdev_ops = {
6635 .ndo_open = ixgbe_open,
6636 .ndo_stop = ixgbe_close,
6637 .ndo_start_xmit = ixgbe_xmit_frame,
6638 .ndo_select_queue = ixgbe_select_queue,
6639 .ndo_set_rx_mode = ixgbe_set_rx_mode,
6640 .ndo_set_multicast_list = ixgbe_set_rx_mode,
6641 .ndo_validate_addr = eth_validate_addr,
6642 .ndo_set_mac_address = ixgbe_set_mac,
6643 .ndo_change_mtu = ixgbe_change_mtu,
6644 .ndo_tx_timeout = ixgbe_tx_timeout,
6645 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
6646 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
6647 .ndo_do_ioctl = ixgbe_ioctl,
6648 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
6649 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
6650 .ndo_set_vf_tx_rate = ixgbe_ndo_set_vf_bw,
6651 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
6652 .ndo_get_stats64 = ixgbe_get_stats64,
6653 #ifdef CONFIG_NET_POLL_CONTROLLER
6654 .ndo_poll_controller = ixgbe_netpoll,
6657 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
6658 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
6659 .ndo_fcoe_enable = ixgbe_fcoe_enable,
6660 .ndo_fcoe_disable = ixgbe_fcoe_disable,
6661 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
6662 #endif /* IXGBE_FCOE */
6665 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
6666 const struct ixgbe_info *ii)
6668 #ifdef CONFIG_PCI_IOV
6669 struct ixgbe_hw *hw = &adapter->hw;
6672 if (hw->mac.type != ixgbe_mac_82599EB || !max_vfs)
6675 /* The 82599 supports up to 64 VFs per physical function
6676 * but this implementation limits allocation to 63 so that
6677 * basic networking resources are still available to the
6680 adapter->num_vfs = (max_vfs > 63) ? 63 : max_vfs;
6681 adapter->flags |= IXGBE_FLAG_SRIOV_ENABLED;
6682 err = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
6684 e_err(probe, "Failed to enable PCI sriov: %d\n", err);
6687 /* If call to enable VFs succeeded then allocate memory
6688 * for per VF control structures.
6691 kcalloc(adapter->num_vfs,
6692 sizeof(struct vf_data_storage), GFP_KERNEL);
6693 if (adapter->vfinfo) {
6694 /* Now that we're sure SR-IOV is enabled
6695 * and memory allocated set up the mailbox parameters
6697 ixgbe_init_mbx_params_pf(hw);
6698 memcpy(&hw->mbx.ops, ii->mbx_ops,
6699 sizeof(hw->mbx.ops));
6701 /* Disable RSC when in SR-IOV mode */
6702 adapter->flags2 &= ~(IXGBE_FLAG2_RSC_CAPABLE |
6703 IXGBE_FLAG2_RSC_ENABLED);
6708 e_err(probe, "Unable to allocate memory for VF Data Storage - "
6709 "SRIOV disabled\n");
6710 pci_disable_sriov(adapter->pdev);
6713 adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
6714 adapter->num_vfs = 0;
6715 #endif /* CONFIG_PCI_IOV */
6719 * ixgbe_probe - Device Initialization Routine
6720 * @pdev: PCI device information struct
6721 * @ent: entry in ixgbe_pci_tbl
6723 * Returns 0 on success, negative on failure
6725 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
6726 * The OS initialization, configuring of the adapter private structure,
6727 * and a hardware reset occur.
6729 static int __devinit ixgbe_probe(struct pci_dev *pdev,
6730 const struct pci_device_id *ent)
6732 struct net_device *netdev;
6733 struct ixgbe_adapter *adapter = NULL;
6734 struct ixgbe_hw *hw;
6735 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
6736 static int cards_found;
6737 int i, err, pci_using_dac;
6738 unsigned int indices = num_possible_cpus();
6744 /* Catch broken hardware that put the wrong VF device ID in
6745 * the PCIe SR-IOV capability.
6747 if (pdev->is_virtfn) {
6748 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
6749 pci_name(pdev), pdev->vendor, pdev->device);
6753 err = pci_enable_device_mem(pdev);
6757 if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
6758 !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
6761 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
6763 err = dma_set_coherent_mask(&pdev->dev,
6767 "No usable DMA configuration, aborting\n");
6774 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
6775 IORESOURCE_MEM), ixgbe_driver_name);
6778 "pci_request_selected_regions failed 0x%x\n", err);
6782 pci_enable_pcie_error_reporting(pdev);
6784 pci_set_master(pdev);
6785 pci_save_state(pdev);
6787 if (ii->mac == ixgbe_mac_82598EB)
6788 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
6790 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
6792 indices = max_t(unsigned int, indices, IXGBE_MAX_DCB_INDICES);
6794 indices += min_t(unsigned int, num_possible_cpus(),
6795 IXGBE_MAX_FCOE_INDICES);
6797 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
6800 goto err_alloc_etherdev;
6803 SET_NETDEV_DEV(netdev, &pdev->dev);
6805 adapter = netdev_priv(netdev);
6806 pci_set_drvdata(pdev, adapter);
6808 adapter->netdev = netdev;
6809 adapter->pdev = pdev;
6812 adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
6814 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
6815 pci_resource_len(pdev, 0));
6821 for (i = 1; i <= 5; i++) {
6822 if (pci_resource_len(pdev, i) == 0)
6826 netdev->netdev_ops = &ixgbe_netdev_ops;
6827 ixgbe_set_ethtool_ops(netdev);
6828 netdev->watchdog_timeo = 5 * HZ;
6829 strcpy(netdev->name, pci_name(pdev));
6831 adapter->bd_number = cards_found;
6834 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
6835 hw->mac.type = ii->mac;
6838 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
6839 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
6840 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
6841 if (!(eec & (1 << 8)))
6842 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
6845 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
6846 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
6847 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
6848 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
6849 hw->phy.mdio.mmds = 0;
6850 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
6851 hw->phy.mdio.dev = netdev;
6852 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
6853 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
6855 /* set up this timer and work struct before calling get_invariants
6856 * which might start the timer
6858 init_timer(&adapter->sfp_timer);
6859 adapter->sfp_timer.function = ixgbe_sfp_timer;
6860 adapter->sfp_timer.data = (unsigned long) adapter;
6862 INIT_WORK(&adapter->sfp_task, ixgbe_sfp_task);
6864 /* multispeed fiber has its own tasklet, called from GPI SDP1 context */
6865 INIT_WORK(&adapter->multispeed_fiber_task, ixgbe_multispeed_fiber_task);
6867 /* a new SFP+ module arrival, called from GPI SDP2 context */
6868 INIT_WORK(&adapter->sfp_config_module_task,
6869 ixgbe_sfp_config_module_task);
6871 ii->get_invariants(hw);
6873 /* setup the private structure */
6874 err = ixgbe_sw_init(adapter);
6878 /* Make it possible the adapter to be woken up via WOL */
6879 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
6880 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6883 * If there is a fan on this device and it has failed log the
6886 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
6887 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
6888 if (esdp & IXGBE_ESDP_SDP1)
6889 e_crit(probe, "Fan has stopped, replace the adapter\n");
6892 /* reset_hw fills in the perm_addr as well */
6893 hw->phy.reset_if_overtemp = true;
6894 err = hw->mac.ops.reset_hw(hw);
6895 hw->phy.reset_if_overtemp = false;
6896 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
6897 hw->mac.type == ixgbe_mac_82598EB) {
6899 * Start a kernel thread to watch for a module to arrive.
6900 * Only do this for 82598, since 82599 will generate
6901 * interrupts on module arrival.
6903 set_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
6904 mod_timer(&adapter->sfp_timer,
6905 round_jiffies(jiffies + (2 * HZ)));
6907 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
6908 e_dev_err("failed to initialize because an unsupported SFP+ "
6909 "module type was detected.\n");
6910 e_dev_err("Reload the driver after installing a supported "
6914 e_dev_err("HW Init failed: %d\n", err);
6918 ixgbe_probe_vf(adapter, ii);
6920 netdev->features = NETIF_F_SG |
6922 NETIF_F_HW_VLAN_TX |
6923 NETIF_F_HW_VLAN_RX |
6924 NETIF_F_HW_VLAN_FILTER;
6926 netdev->features |= NETIF_F_IPV6_CSUM;
6927 netdev->features |= NETIF_F_TSO;
6928 netdev->features |= NETIF_F_TSO6;
6929 netdev->features |= NETIF_F_GRO;
6931 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
6932 netdev->features |= NETIF_F_SCTP_CSUM;
6934 netdev->vlan_features |= NETIF_F_TSO;
6935 netdev->vlan_features |= NETIF_F_TSO6;
6936 netdev->vlan_features |= NETIF_F_IP_CSUM;
6937 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
6938 netdev->vlan_features |= NETIF_F_SG;
6940 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6941 adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
6942 IXGBE_FLAG_DCB_ENABLED);
6943 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
6944 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
6946 #ifdef CONFIG_IXGBE_DCB
6947 netdev->dcbnl_ops = &dcbnl_ops;
6951 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
6952 if (hw->mac.ops.get_device_caps) {
6953 hw->mac.ops.get_device_caps(hw, &device_caps);
6954 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
6955 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6958 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
6959 netdev->vlan_features |= NETIF_F_FCOE_CRC;
6960 netdev->vlan_features |= NETIF_F_FSO;
6961 netdev->vlan_features |= NETIF_F_FCOE_MTU;
6963 #endif /* IXGBE_FCOE */
6964 if (pci_using_dac) {
6965 netdev->features |= NETIF_F_HIGHDMA;
6966 netdev->vlan_features |= NETIF_F_HIGHDMA;
6969 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
6970 netdev->features |= NETIF_F_LRO;
6972 /* make sure the EEPROM is good */
6973 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
6974 e_dev_err("The EEPROM Checksum Is Not Valid\n");
6979 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
6980 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
6982 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
6983 e_dev_err("invalid MAC address\n");
6988 /* power down the optics */
6989 if (hw->phy.multispeed_fiber)
6990 hw->mac.ops.disable_tx_laser(hw);
6992 init_timer(&adapter->watchdog_timer);
6993 adapter->watchdog_timer.function = ixgbe_watchdog;
6994 adapter->watchdog_timer.data = (unsigned long)adapter;
6996 INIT_WORK(&adapter->reset_task, ixgbe_reset_task);
6997 INIT_WORK(&adapter->watchdog_task, ixgbe_watchdog_task);
6999 err = ixgbe_init_interrupt_scheme(adapter);
7003 switch (pdev->device) {
7004 case IXGBE_DEV_ID_82599_KX4:
7005 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
7006 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
7012 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
7014 /* pick up the PCI bus settings for reporting later */
7015 hw->mac.ops.get_bus_info(hw);
7017 /* print bus type/speed/width info */
7018 e_dev_info("(PCI Express:%s:%s) %pM\n",
7019 (hw->bus.speed == ixgbe_bus_speed_5000 ? "5.0Gb/s" :
7020 hw->bus.speed == ixgbe_bus_speed_2500 ? "2.5Gb/s" :
7022 (hw->bus.width == ixgbe_bus_width_pcie_x8 ? "Width x8" :
7023 hw->bus.width == ixgbe_bus_width_pcie_x4 ? "Width x4" :
7024 hw->bus.width == ixgbe_bus_width_pcie_x1 ? "Width x1" :
7027 ixgbe_read_pba_num_generic(hw, &part_num);
7028 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
7029 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, "
7030 "PBA No: %06x-%03x\n",
7031 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
7032 (part_num >> 8), (part_num & 0xff));
7034 e_dev_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
7035 hw->mac.type, hw->phy.type,
7036 (part_num >> 8), (part_num & 0xff));
7038 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
7039 e_dev_warn("PCI-Express bandwidth available for this card is "
7040 "not sufficient for optimal performance.\n");
7041 e_dev_warn("For optimal performance a x8 PCI-Express slot "
7045 /* save off EEPROM version number */
7046 hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
7048 /* reset the hardware with the new settings */
7049 err = hw->mac.ops.start_hw(hw);
7051 if (err == IXGBE_ERR_EEPROM_VERSION) {
7052 /* We are running on a pre-production device, log a warning */
7053 e_dev_warn("This device is a pre-production adapter/LOM. "
7054 "Please be aware there may be issues associated "
7055 "with your hardware. If you are experiencing "
7056 "problems please contact your Intel or hardware "
7057 "representative who provided you with this "
7060 strcpy(netdev->name, "eth%d");
7061 err = register_netdev(netdev);
7065 /* carrier off reporting is important to ethtool even BEFORE open */
7066 netif_carrier_off(netdev);
7068 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
7069 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
7070 INIT_WORK(&adapter->fdir_reinit_task, ixgbe_fdir_reinit_task);
7072 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
7073 INIT_WORK(&adapter->check_overtemp_task,
7074 ixgbe_check_overtemp_task);
7075 #ifdef CONFIG_IXGBE_DCA
7076 if (dca_add_requester(&pdev->dev) == 0) {
7077 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
7078 ixgbe_setup_dca(adapter);
7081 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7082 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
7083 for (i = 0; i < adapter->num_vfs; i++)
7084 ixgbe_vf_configuration(pdev, (i | 0x10000000));
7087 /* add san mac addr to netdev */
7088 ixgbe_add_sanmac_netdev(netdev);
7090 e_dev_info("Intel(R) 10 Gigabit Network Connection\n");
7095 ixgbe_release_hw_control(adapter);
7096 ixgbe_clear_interrupt_scheme(adapter);
7099 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7100 ixgbe_disable_sriov(adapter);
7101 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
7102 del_timer_sync(&adapter->sfp_timer);
7103 cancel_work_sync(&adapter->sfp_task);
7104 cancel_work_sync(&adapter->multispeed_fiber_task);
7105 cancel_work_sync(&adapter->sfp_config_module_task);
7106 iounmap(hw->hw_addr);
7108 free_netdev(netdev);
7110 pci_release_selected_regions(pdev,
7111 pci_select_bars(pdev, IORESOURCE_MEM));
7114 pci_disable_device(pdev);
7119 * ixgbe_remove - Device Removal Routine
7120 * @pdev: PCI device information struct
7122 * ixgbe_remove is called by the PCI subsystem to alert the driver
7123 * that it should release a PCI device. The could be caused by a
7124 * Hot-Plug event, or because the driver is going to be removed from
7127 static void __devexit ixgbe_remove(struct pci_dev *pdev)
7129 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7130 struct net_device *netdev = adapter->netdev;
7132 set_bit(__IXGBE_DOWN, &adapter->state);
7133 /* clear the module not found bit to make sure the worker won't
7136 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
7137 del_timer_sync(&adapter->watchdog_timer);
7139 del_timer_sync(&adapter->sfp_timer);
7140 cancel_work_sync(&adapter->watchdog_task);
7141 cancel_work_sync(&adapter->sfp_task);
7142 cancel_work_sync(&adapter->multispeed_fiber_task);
7143 cancel_work_sync(&adapter->sfp_config_module_task);
7144 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
7145 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
7146 cancel_work_sync(&adapter->fdir_reinit_task);
7147 flush_scheduled_work();
7149 #ifdef CONFIG_IXGBE_DCA
7150 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
7151 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
7152 dca_remove_requester(&pdev->dev);
7153 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
7158 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
7159 ixgbe_cleanup_fcoe(adapter);
7161 #endif /* IXGBE_FCOE */
7163 /* remove the added san mac */
7164 ixgbe_del_sanmac_netdev(netdev);
7166 if (netdev->reg_state == NETREG_REGISTERED)
7167 unregister_netdev(netdev);
7169 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7170 ixgbe_disable_sriov(adapter);
7172 ixgbe_clear_interrupt_scheme(adapter);
7174 ixgbe_release_hw_control(adapter);
7176 iounmap(adapter->hw.hw_addr);
7177 pci_release_selected_regions(pdev, pci_select_bars(pdev,
7180 e_dev_info("complete\n");
7182 free_netdev(netdev);
7184 pci_disable_pcie_error_reporting(pdev);
7186 pci_disable_device(pdev);
7190 * ixgbe_io_error_detected - called when PCI error is detected
7191 * @pdev: Pointer to PCI device
7192 * @state: The current pci connection state
7194 * This function is called after a PCI bus error affecting
7195 * this device has been detected.
7197 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
7198 pci_channel_state_t state)
7200 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7201 struct net_device *netdev = adapter->netdev;
7203 netif_device_detach(netdev);
7205 if (state == pci_channel_io_perm_failure)
7206 return PCI_ERS_RESULT_DISCONNECT;
7208 if (netif_running(netdev))
7209 ixgbe_down(adapter);
7210 pci_disable_device(pdev);
7212 /* Request a slot reset. */
7213 return PCI_ERS_RESULT_NEED_RESET;
7217 * ixgbe_io_slot_reset - called after the pci bus has been reset.
7218 * @pdev: Pointer to PCI device
7220 * Restart the card from scratch, as if from a cold-boot.
7222 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7224 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7225 pci_ers_result_t result;
7228 if (pci_enable_device_mem(pdev)) {
7229 e_err(probe, "Cannot re-enable PCI device after reset.\n");
7230 result = PCI_ERS_RESULT_DISCONNECT;
7232 pci_set_master(pdev);
7233 pci_restore_state(pdev);
7234 pci_save_state(pdev);
7236 pci_wake_from_d3(pdev, false);
7238 ixgbe_reset(adapter);
7239 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7240 result = PCI_ERS_RESULT_RECOVERED;
7243 err = pci_cleanup_aer_uncorrect_error_status(pdev);
7245 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7246 "failed 0x%0x\n", err);
7247 /* non-fatal, continue */
7254 * ixgbe_io_resume - called when traffic can start flowing again.
7255 * @pdev: Pointer to PCI device
7257 * This callback is called when the error recovery driver tells us that
7258 * its OK to resume normal operation.
7260 static void ixgbe_io_resume(struct pci_dev *pdev)
7262 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7263 struct net_device *netdev = adapter->netdev;
7265 if (netif_running(netdev)) {
7266 if (ixgbe_up(adapter)) {
7267 e_info(probe, "ixgbe_up failed after reset\n");
7272 netif_device_attach(netdev);
7275 static struct pci_error_handlers ixgbe_err_handler = {
7276 .error_detected = ixgbe_io_error_detected,
7277 .slot_reset = ixgbe_io_slot_reset,
7278 .resume = ixgbe_io_resume,
7281 static struct pci_driver ixgbe_driver = {
7282 .name = ixgbe_driver_name,
7283 .id_table = ixgbe_pci_tbl,
7284 .probe = ixgbe_probe,
7285 .remove = __devexit_p(ixgbe_remove),
7287 .suspend = ixgbe_suspend,
7288 .resume = ixgbe_resume,
7290 .shutdown = ixgbe_shutdown,
7291 .err_handler = &ixgbe_err_handler
7295 * ixgbe_init_module - Driver Registration Routine
7297 * ixgbe_init_module is the first routine called when the driver is
7298 * loaded. All it does is register with the PCI subsystem.
7300 static int __init ixgbe_init_module(void)
7303 pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
7304 pr_info("%s\n", ixgbe_copyright);
7306 #ifdef CONFIG_IXGBE_DCA
7307 dca_register_notify(&dca_notifier);
7310 ret = pci_register_driver(&ixgbe_driver);
7314 module_init(ixgbe_init_module);
7317 * ixgbe_exit_module - Driver Exit Cleanup Routine
7319 * ixgbe_exit_module is called just before the driver is removed
7322 static void __exit ixgbe_exit_module(void)
7324 #ifdef CONFIG_IXGBE_DCA
7325 dca_unregister_notify(&dca_notifier);
7327 pci_unregister_driver(&ixgbe_driver);
7328 rcu_barrier(); /* Wait for completion of call_rcu()'s */
7331 #ifdef CONFIG_IXGBE_DCA
7332 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
7337 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
7338 __ixgbe_notify_dca);
7340 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7343 #endif /* CONFIG_IXGBE_DCA */
7346 * ixgbe_get_hw_dev return device
7347 * used by hardware layer to print debugging information
7349 struct net_device *ixgbe_get_hw_dev(struct ixgbe_hw *hw)
7351 struct ixgbe_adapter *adapter = hw->back;
7352 return adapter->netdev;
7355 module_exit(ixgbe_exit_module);