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 i, eop, count = 0;
739 unsigned int total_bytes = 0, total_packets = 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;
775 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
776 if (unlikely(count && netif_carrier_ok(tx_ring->netdev) &&
777 (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
778 /* Make sure that anybody stopping the queue after this
779 * sees the new next_to_clean.
782 if (__netif_subqueue_stopped(tx_ring->netdev, tx_ring->queue_index) &&
783 !test_bit(__IXGBE_DOWN, &adapter->state)) {
784 netif_wake_subqueue(tx_ring->netdev, tx_ring->queue_index);
785 ++tx_ring->tx_stats.restart_queue;
789 if (check_for_tx_hang(tx_ring) &&
790 ixgbe_check_tx_hang(adapter, tx_ring, i)) {
791 /* schedule immediate reset if we believe we hung */
792 e_info(probe, "tx hang %d detected, resetting "
793 "adapter\n", adapter->tx_timeout_count + 1);
794 ixgbe_tx_timeout(adapter->netdev);
797 /* re-arm the interrupt */
798 if (count >= tx_ring->work_limit)
799 ixgbe_irq_rearm_queues(adapter, ((u64)1 << q_vector->v_idx));
801 tx_ring->total_bytes += total_bytes;
802 tx_ring->total_packets += total_packets;
803 u64_stats_update_begin(&tx_ring->syncp);
804 tx_ring->stats.packets += total_packets;
805 tx_ring->stats.bytes += total_bytes;
806 u64_stats_update_end(&tx_ring->syncp);
807 return count < tx_ring->work_limit;
810 #ifdef CONFIG_IXGBE_DCA
811 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
812 struct ixgbe_ring *rx_ring,
815 struct ixgbe_hw *hw = &adapter->hw;
817 u8 reg_idx = rx_ring->reg_idx;
819 rxctrl = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(reg_idx));
820 switch (hw->mac.type) {
821 case ixgbe_mac_82598EB:
822 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
823 rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
825 case ixgbe_mac_82599EB:
826 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK_82599;
827 rxctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
828 IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599);
833 rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
834 rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
835 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_RRO_EN);
836 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_WRO_EN |
837 IXGBE_DCA_RXCTRL_DESC_HSRO_EN);
838 IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
841 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
842 struct ixgbe_ring *tx_ring,
845 struct ixgbe_hw *hw = &adapter->hw;
847 u8 reg_idx = tx_ring->reg_idx;
849 switch (hw->mac.type) {
850 case ixgbe_mac_82598EB:
851 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(reg_idx));
852 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
853 txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
854 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
855 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
856 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(reg_idx), txctrl);
858 case ixgbe_mac_82599EB:
859 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(reg_idx));
860 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK_82599;
861 txctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
862 IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599);
863 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
864 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
865 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(reg_idx), txctrl);
872 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
874 struct ixgbe_adapter *adapter = q_vector->adapter;
879 if (q_vector->cpu == cpu)
882 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
883 for (i = 0; i < q_vector->txr_count; i++) {
884 ixgbe_update_tx_dca(adapter, adapter->tx_ring[r_idx], cpu);
885 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
889 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
890 for (i = 0; i < q_vector->rxr_count; i++) {
891 ixgbe_update_rx_dca(adapter, adapter->rx_ring[r_idx], cpu);
892 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
901 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
906 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
909 /* always use CB2 mode, difference is masked in the CB driver */
910 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
912 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
913 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
917 for (i = 0; i < num_q_vectors; i++) {
918 adapter->q_vector[i]->cpu = -1;
919 ixgbe_update_dca(adapter->q_vector[i]);
923 static int __ixgbe_notify_dca(struct device *dev, void *data)
925 struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
926 unsigned long event = *(unsigned long *)data;
928 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
932 case DCA_PROVIDER_ADD:
933 /* if we're already enabled, don't do it again */
934 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
936 if (dca_add_requester(dev) == 0) {
937 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
938 ixgbe_setup_dca(adapter);
941 /* Fall Through since DCA is disabled. */
942 case DCA_PROVIDER_REMOVE:
943 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
944 dca_remove_requester(dev);
945 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
946 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
954 #endif /* CONFIG_IXGBE_DCA */
956 * ixgbe_receive_skb - Send a completed packet up the stack
957 * @adapter: board private structure
958 * @skb: packet to send up
959 * @status: hardware indication of status of receive
960 * @rx_ring: rx descriptor ring (for a specific queue) to setup
961 * @rx_desc: rx descriptor
963 static void ixgbe_receive_skb(struct ixgbe_q_vector *q_vector,
964 struct sk_buff *skb, u8 status,
965 struct ixgbe_ring *ring,
966 union ixgbe_adv_rx_desc *rx_desc)
968 struct ixgbe_adapter *adapter = q_vector->adapter;
969 struct napi_struct *napi = &q_vector->napi;
970 bool is_vlan = (status & IXGBE_RXD_STAT_VP);
971 u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
973 if (is_vlan && (tag & VLAN_VID_MASK))
974 __vlan_hwaccel_put_tag(skb, tag);
976 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
977 napi_gro_receive(napi, skb);
983 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
984 * @adapter: address of board private structure
985 * @status_err: hardware indication of status of receive
986 * @skb: skb currently being received and modified
988 static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
989 union ixgbe_adv_rx_desc *rx_desc,
992 u32 status_err = le32_to_cpu(rx_desc->wb.upper.status_error);
994 skb_checksum_none_assert(skb);
996 /* Rx csum disabled */
997 if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
1000 /* if IP and error */
1001 if ((status_err & IXGBE_RXD_STAT_IPCS) &&
1002 (status_err & IXGBE_RXDADV_ERR_IPE)) {
1003 adapter->hw_csum_rx_error++;
1007 if (!(status_err & IXGBE_RXD_STAT_L4CS))
1010 if (status_err & IXGBE_RXDADV_ERR_TCPE) {
1011 u16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1014 * 82599 errata, UDP frames with a 0 checksum can be marked as
1017 if ((pkt_info & IXGBE_RXDADV_PKTTYPE_UDP) &&
1018 (adapter->hw.mac.type == ixgbe_mac_82599EB))
1021 adapter->hw_csum_rx_error++;
1025 /* It must be a TCP or UDP packet with a valid checksum */
1026 skb->ip_summed = CHECKSUM_UNNECESSARY;
1029 static inline void ixgbe_release_rx_desc(struct ixgbe_ring *rx_ring, u32 val)
1032 * Force memory writes to complete before letting h/w
1033 * know there are new descriptors to fetch. (Only
1034 * applicable for weak-ordered memory model archs,
1038 writel(val, rx_ring->tail);
1042 * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
1043 * @rx_ring: ring to place buffers on
1044 * @cleaned_count: number of buffers to replace
1046 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1048 union ixgbe_adv_rx_desc *rx_desc;
1049 struct ixgbe_rx_buffer *bi;
1050 struct sk_buff *skb;
1051 u16 i = rx_ring->next_to_use;
1053 /* do nothing if no valid netdev defined */
1054 if (!rx_ring->netdev)
1057 while (cleaned_count--) {
1058 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1059 bi = &rx_ring->rx_buffer_info[i];
1063 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1064 rx_ring->rx_buf_len);
1066 rx_ring->rx_stats.alloc_rx_buff_failed++;
1069 /* initialize queue mapping */
1070 skb_record_rx_queue(skb, rx_ring->queue_index);
1075 bi->dma = dma_map_single(rx_ring->dev,
1077 rx_ring->rx_buf_len,
1079 if (dma_mapping_error(rx_ring->dev, bi->dma)) {
1080 rx_ring->rx_stats.alloc_rx_buff_failed++;
1086 if (ring_is_ps_enabled(rx_ring)) {
1088 bi->page = netdev_alloc_page(rx_ring->netdev);
1090 rx_ring->rx_stats.alloc_rx_page_failed++;
1095 if (!bi->page_dma) {
1096 /* use a half page if we're re-using */
1097 bi->page_offset ^= PAGE_SIZE / 2;
1098 bi->page_dma = dma_map_page(rx_ring->dev,
1103 if (dma_mapping_error(rx_ring->dev,
1105 rx_ring->rx_stats.alloc_rx_page_failed++;
1111 /* Refresh the desc even if buffer_addrs didn't change
1112 * because each write-back erases this info. */
1113 rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
1114 rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
1116 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
1117 rx_desc->read.hdr_addr = 0;
1121 if (i == rx_ring->count)
1126 if (rx_ring->next_to_use != i) {
1127 rx_ring->next_to_use = i;
1128 ixgbe_release_rx_desc(rx_ring, i);
1132 static inline u16 ixgbe_get_hdr_info(union ixgbe_adv_rx_desc *rx_desc)
1134 return rx_desc->wb.lower.lo_dword.hs_rss.hdr_info;
1137 static inline u16 ixgbe_get_pkt_info(union ixgbe_adv_rx_desc *rx_desc)
1139 return rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1142 static inline u32 ixgbe_get_rsc_count(union ixgbe_adv_rx_desc *rx_desc)
1144 return (le32_to_cpu(rx_desc->wb.lower.lo_dword.data) &
1145 IXGBE_RXDADV_RSCCNT_MASK) >>
1146 IXGBE_RXDADV_RSCCNT_SHIFT;
1150 * ixgbe_transform_rsc_queue - change rsc queue into a full packet
1151 * @skb: pointer to the last skb in the rsc queue
1152 * @count: pointer to number of packets coalesced in this context
1154 * This function changes a queue full of hw rsc buffers into a completed
1155 * packet. It uses the ->prev pointers to find the first packet and then
1156 * turns it into the frag list owner.
1158 static inline struct sk_buff *ixgbe_transform_rsc_queue(struct sk_buff *skb,
1161 unsigned int frag_list_size = 0;
1164 struct sk_buff *prev = skb->prev;
1165 frag_list_size += skb->len;
1171 skb_shinfo(skb)->frag_list = skb->next;
1173 skb->len += frag_list_size;
1174 skb->data_len += frag_list_size;
1175 skb->truesize += frag_list_size;
1179 struct ixgbe_rsc_cb {
1184 #define IXGBE_RSC_CB(skb) ((struct ixgbe_rsc_cb *)(skb)->cb)
1186 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1187 struct ixgbe_ring *rx_ring,
1188 int *work_done, int work_to_do)
1190 struct ixgbe_adapter *adapter = q_vector->adapter;
1191 union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
1192 struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
1193 struct sk_buff *skb;
1194 unsigned int i, rsc_count = 0;
1197 bool cleaned = false;
1198 int cleaned_count = 0;
1199 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1202 #endif /* IXGBE_FCOE */
1204 i = rx_ring->next_to_clean;
1205 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1206 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1207 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1209 while (staterr & IXGBE_RXD_STAT_DD) {
1211 if (*work_done >= work_to_do)
1215 rmb(); /* read descriptor and rx_buffer_info after status DD */
1216 if (ring_is_ps_enabled(rx_ring)) {
1217 hdr_info = le16_to_cpu(ixgbe_get_hdr_info(rx_desc));
1218 len = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
1219 IXGBE_RXDADV_HDRBUFLEN_SHIFT;
1220 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1221 if ((len > IXGBE_RX_HDR_SIZE) ||
1222 (upper_len && !(hdr_info & IXGBE_RXDADV_SPH)))
1223 len = IXGBE_RX_HDR_SIZE;
1225 len = le16_to_cpu(rx_desc->wb.upper.length);
1229 skb = rx_buffer_info->skb;
1230 prefetch(skb->data);
1231 rx_buffer_info->skb = NULL;
1233 if (rx_buffer_info->dma) {
1234 if ((adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
1235 (!(staterr & IXGBE_RXD_STAT_EOP)) &&
1238 * When HWRSC is enabled, delay unmapping
1239 * of the first packet. It carries the
1240 * header information, HW may still
1241 * access the header after the writeback.
1242 * Only unmap it when EOP is reached
1244 IXGBE_RSC_CB(skb)->delay_unmap = true;
1245 IXGBE_RSC_CB(skb)->dma = rx_buffer_info->dma;
1247 dma_unmap_single(rx_ring->dev,
1248 rx_buffer_info->dma,
1249 rx_ring->rx_buf_len,
1252 rx_buffer_info->dma = 0;
1257 dma_unmap_page(rx_ring->dev,
1258 rx_buffer_info->page_dma,
1261 rx_buffer_info->page_dma = 0;
1262 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1263 rx_buffer_info->page,
1264 rx_buffer_info->page_offset,
1267 if ((rx_ring->rx_buf_len > (PAGE_SIZE / 2)) ||
1268 (page_count(rx_buffer_info->page) != 1))
1269 rx_buffer_info->page = NULL;
1271 get_page(rx_buffer_info->page);
1273 skb->len += upper_len;
1274 skb->data_len += upper_len;
1275 skb->truesize += upper_len;
1279 if (i == rx_ring->count)
1282 next_rxd = IXGBE_RX_DESC_ADV(rx_ring, i);
1286 if (ring_is_rsc_enabled(rx_ring))
1287 rsc_count = ixgbe_get_rsc_count(rx_desc);
1290 u32 nextp = (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
1291 IXGBE_RXDADV_NEXTP_SHIFT;
1292 next_buffer = &rx_ring->rx_buffer_info[nextp];
1294 next_buffer = &rx_ring->rx_buffer_info[i];
1297 if (staterr & IXGBE_RXD_STAT_EOP) {
1299 skb = ixgbe_transform_rsc_queue(skb,
1300 &(rx_ring->rx_stats.rsc_count));
1301 if (ring_is_rsc_enabled(rx_ring)) {
1302 if (IXGBE_RSC_CB(skb)->delay_unmap) {
1303 dma_unmap_single(rx_ring->dev,
1304 IXGBE_RSC_CB(skb)->dma,
1305 rx_ring->rx_buf_len,
1307 IXGBE_RSC_CB(skb)->dma = 0;
1308 IXGBE_RSC_CB(skb)->delay_unmap = false;
1310 if (ring_is_ps_enabled(rx_ring))
1311 rx_ring->rx_stats.rsc_count +=
1312 skb_shinfo(skb)->nr_frags;
1314 rx_ring->rx_stats.rsc_count++;
1315 rx_ring->rx_stats.rsc_flush++;
1317 u64_stats_update_begin(&rx_ring->syncp);
1318 rx_ring->stats.packets++;
1319 rx_ring->stats.bytes += skb->len;
1320 u64_stats_update_end(&rx_ring->syncp);
1322 if (ring_is_ps_enabled(rx_ring)) {
1323 rx_buffer_info->skb = next_buffer->skb;
1324 rx_buffer_info->dma = next_buffer->dma;
1325 next_buffer->skb = skb;
1326 next_buffer->dma = 0;
1328 skb->next = next_buffer->skb;
1329 skb->next->prev = skb;
1331 rx_ring->rx_stats.non_eop_descs++;
1335 if (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) {
1336 dev_kfree_skb_irq(skb);
1340 ixgbe_rx_checksum(adapter, rx_desc, skb);
1342 /* probably a little skewed due to removing CRC */
1343 total_rx_bytes += skb->len;
1346 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1348 /* if ddp, not passing to ULD unless for FCP_RSP or error */
1349 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
1350 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
1354 #endif /* IXGBE_FCOE */
1355 ixgbe_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
1358 rx_desc->wb.upper.status_error = 0;
1360 /* return some buffers to hardware, one at a time is too slow */
1361 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
1362 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1366 /* use prefetched values */
1368 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1370 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1373 rx_ring->next_to_clean = i;
1374 cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
1377 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1380 /* include DDPed FCoE data */
1381 if (ddp_bytes > 0) {
1384 mss = rx_ring->netdev->mtu - sizeof(struct fcoe_hdr) -
1385 sizeof(struct fc_frame_header) -
1386 sizeof(struct fcoe_crc_eof);
1389 total_rx_bytes += ddp_bytes;
1390 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
1392 #endif /* IXGBE_FCOE */
1394 rx_ring->total_packets += total_rx_packets;
1395 rx_ring->total_bytes += total_rx_bytes;
1400 static int ixgbe_clean_rxonly(struct napi_struct *, int);
1402 * ixgbe_configure_msix - Configure MSI-X hardware
1403 * @adapter: board private structure
1405 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1408 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1410 struct ixgbe_q_vector *q_vector;
1411 int i, j, q_vectors, v_idx, r_idx;
1414 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1417 * Populate the IVAR table and set the ITR values to the
1418 * corresponding register.
1420 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1421 q_vector = adapter->q_vector[v_idx];
1422 /* XXX for_each_set_bit(...) */
1423 r_idx = find_first_bit(q_vector->rxr_idx,
1424 adapter->num_rx_queues);
1426 for (i = 0; i < q_vector->rxr_count; i++) {
1427 j = adapter->rx_ring[r_idx]->reg_idx;
1428 ixgbe_set_ivar(adapter, 0, j, v_idx);
1429 r_idx = find_next_bit(q_vector->rxr_idx,
1430 adapter->num_rx_queues,
1433 r_idx = find_first_bit(q_vector->txr_idx,
1434 adapter->num_tx_queues);
1436 for (i = 0; i < q_vector->txr_count; i++) {
1437 j = adapter->tx_ring[r_idx]->reg_idx;
1438 ixgbe_set_ivar(adapter, 1, j, v_idx);
1439 r_idx = find_next_bit(q_vector->txr_idx,
1440 adapter->num_tx_queues,
1444 if (q_vector->txr_count && !q_vector->rxr_count)
1446 q_vector->eitr = adapter->tx_eitr_param;
1447 else if (q_vector->rxr_count)
1449 q_vector->eitr = adapter->rx_eitr_param;
1451 ixgbe_write_eitr(q_vector);
1452 /* If Flow Director is enabled, set interrupt affinity */
1453 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
1454 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)) {
1456 * Allocate the affinity_hint cpumask, assign the mask
1457 * for this vector, and set our affinity_hint for
1460 if (!alloc_cpumask_var(&q_vector->affinity_mask,
1463 cpumask_set_cpu(v_idx, q_vector->affinity_mask);
1464 irq_set_affinity_hint(adapter->msix_entries[v_idx].vector,
1465 q_vector->affinity_mask);
1469 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
1470 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1472 else if (adapter->hw.mac.type == ixgbe_mac_82599EB)
1473 ixgbe_set_ivar(adapter, -1, 1, v_idx);
1474 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1476 /* set up to autoclear timer, and the vectors */
1477 mask = IXGBE_EIMS_ENABLE_MASK;
1478 if (adapter->num_vfs)
1479 mask &= ~(IXGBE_EIMS_OTHER |
1480 IXGBE_EIMS_MAILBOX |
1483 mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
1484 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1487 enum latency_range {
1491 latency_invalid = 255
1495 * ixgbe_update_itr - update the dynamic ITR value based on statistics
1496 * @adapter: pointer to adapter
1497 * @eitr: eitr setting (ints per sec) to give last timeslice
1498 * @itr_setting: current throttle rate in ints/second
1499 * @packets: the number of packets during this measurement interval
1500 * @bytes: the number of bytes during this measurement interval
1502 * Stores a new ITR value based on packets and byte
1503 * counts during the last interrupt. The advantage of per interrupt
1504 * computation is faster updates and more accurate ITR for the current
1505 * traffic pattern. Constants in this function were computed
1506 * based on theoretical maximum wire speed and thresholds were set based
1507 * on testing data as well as attempting to minimize response time
1508 * while increasing bulk throughput.
1509 * this functionality is controlled by the InterruptThrottleRate module
1510 * parameter (see ixgbe_param.c)
1512 static u8 ixgbe_update_itr(struct ixgbe_adapter *adapter,
1513 u32 eitr, u8 itr_setting,
1514 int packets, int bytes)
1516 unsigned int retval = itr_setting;
1521 goto update_itr_done;
1524 /* simple throttlerate management
1525 * 0-20MB/s lowest (100000 ints/s)
1526 * 20-100MB/s low (20000 ints/s)
1527 * 100-1249MB/s bulk (8000 ints/s)
1529 /* what was last interrupt timeslice? */
1530 timepassed_us = 1000000/eitr;
1531 bytes_perint = bytes / timepassed_us; /* bytes/usec */
1533 switch (itr_setting) {
1534 case lowest_latency:
1535 if (bytes_perint > adapter->eitr_low)
1536 retval = low_latency;
1539 if (bytes_perint > adapter->eitr_high)
1540 retval = bulk_latency;
1541 else if (bytes_perint <= adapter->eitr_low)
1542 retval = lowest_latency;
1545 if (bytes_perint <= adapter->eitr_high)
1546 retval = low_latency;
1555 * ixgbe_write_eitr - write EITR register in hardware specific way
1556 * @q_vector: structure containing interrupt and ring information
1558 * This function is made to be called by ethtool and by the driver
1559 * when it needs to update EITR registers at runtime. Hardware
1560 * specific quirks/differences are taken care of here.
1562 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1564 struct ixgbe_adapter *adapter = q_vector->adapter;
1565 struct ixgbe_hw *hw = &adapter->hw;
1566 int v_idx = q_vector->v_idx;
1567 u32 itr_reg = EITR_INTS_PER_SEC_TO_REG(q_vector->eitr);
1569 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1570 /* must write high and low 16 bits to reset counter */
1571 itr_reg |= (itr_reg << 16);
1572 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1574 * 82599 can support a value of zero, so allow it for
1575 * max interrupt rate, but there is an errata where it can
1576 * not be zero with RSC
1579 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
1583 * set the WDIS bit to not clear the timer bits and cause an
1584 * immediate assertion of the interrupt
1586 itr_reg |= IXGBE_EITR_CNT_WDIS;
1588 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1591 static void ixgbe_set_itr_msix(struct ixgbe_q_vector *q_vector)
1593 struct ixgbe_adapter *adapter = q_vector->adapter;
1595 u8 current_itr, ret_itr;
1597 struct ixgbe_ring *rx_ring, *tx_ring;
1599 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1600 for (i = 0; i < q_vector->txr_count; i++) {
1601 tx_ring = adapter->tx_ring[r_idx];
1602 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1604 tx_ring->total_packets,
1605 tx_ring->total_bytes);
1606 /* if the result for this queue would decrease interrupt
1607 * rate for this vector then use that result */
1608 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
1609 q_vector->tx_itr - 1 : ret_itr);
1610 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1614 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1615 for (i = 0; i < q_vector->rxr_count; i++) {
1616 rx_ring = adapter->rx_ring[r_idx];
1617 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1619 rx_ring->total_packets,
1620 rx_ring->total_bytes);
1621 /* if the result for this queue would decrease interrupt
1622 * rate for this vector then use that result */
1623 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
1624 q_vector->rx_itr - 1 : ret_itr);
1625 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1629 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1631 switch (current_itr) {
1632 /* counts and packets in update_itr are dependent on these numbers */
1633 case lowest_latency:
1637 new_itr = 20000; /* aka hwitr = ~200 */
1645 if (new_itr != q_vector->eitr) {
1646 /* do an exponential smoothing */
1647 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1649 /* save the algorithm value here, not the smoothed one */
1650 q_vector->eitr = new_itr;
1652 ixgbe_write_eitr(q_vector);
1657 * ixgbe_check_overtemp_task - worker thread to check over tempurature
1658 * @work: pointer to work_struct containing our data
1660 static void ixgbe_check_overtemp_task(struct work_struct *work)
1662 struct ixgbe_adapter *adapter = container_of(work,
1663 struct ixgbe_adapter,
1664 check_overtemp_task);
1665 struct ixgbe_hw *hw = &adapter->hw;
1666 u32 eicr = adapter->interrupt_event;
1668 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
1671 switch (hw->device_id) {
1672 case IXGBE_DEV_ID_82599_T3_LOM: {
1674 bool link_up = false;
1676 if (hw->mac.ops.check_link)
1677 hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
1679 if (((eicr & IXGBE_EICR_GPI_SDP0) && (!link_up)) ||
1680 (eicr & IXGBE_EICR_LSC))
1681 /* Check if this is due to overtemp */
1682 if (hw->phy.ops.check_overtemp(hw) == IXGBE_ERR_OVERTEMP)
1687 if (!(eicr & IXGBE_EICR_GPI_SDP0))
1692 "Network adapter has been stopped because it has over heated. "
1693 "Restart the computer. If the problem persists, "
1694 "power off the system and replace the adapter\n");
1695 /* write to clear the interrupt */
1696 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP0);
1699 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
1701 struct ixgbe_hw *hw = &adapter->hw;
1703 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
1704 (eicr & IXGBE_EICR_GPI_SDP1)) {
1705 e_crit(probe, "Fan has stopped, replace the adapter\n");
1706 /* write to clear the interrupt */
1707 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1711 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
1713 struct ixgbe_hw *hw = &adapter->hw;
1715 if (eicr & IXGBE_EICR_GPI_SDP2) {
1716 /* Clear the interrupt */
1717 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
1718 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1719 schedule_work(&adapter->sfp_config_module_task);
1722 if (eicr & IXGBE_EICR_GPI_SDP1) {
1723 /* Clear the interrupt */
1724 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1725 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1726 schedule_work(&adapter->multispeed_fiber_task);
1730 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1732 struct ixgbe_hw *hw = &adapter->hw;
1735 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1736 adapter->link_check_timeout = jiffies;
1737 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1738 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1739 IXGBE_WRITE_FLUSH(hw);
1740 schedule_work(&adapter->watchdog_task);
1744 static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
1746 struct net_device *netdev = data;
1747 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1748 struct ixgbe_hw *hw = &adapter->hw;
1752 * Workaround for Silicon errata. Use clear-by-write instead
1753 * of clear-by-read. Reading with EICS will return the
1754 * interrupt causes without clearing, which later be done
1755 * with the write to EICR.
1757 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1758 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1760 if (eicr & IXGBE_EICR_LSC)
1761 ixgbe_check_lsc(adapter);
1763 if (eicr & IXGBE_EICR_MAILBOX)
1764 ixgbe_msg_task(adapter);
1766 if (hw->mac.type == ixgbe_mac_82598EB)
1767 ixgbe_check_fan_failure(adapter, eicr);
1769 if (hw->mac.type == ixgbe_mac_82599EB) {
1770 ixgbe_check_sfp_event(adapter, eicr);
1771 adapter->interrupt_event = eicr;
1772 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
1773 ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)))
1774 schedule_work(&adapter->check_overtemp_task);
1776 /* Handle Flow Director Full threshold interrupt */
1777 if (eicr & IXGBE_EICR_FLOW_DIR) {
1779 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_FLOW_DIR);
1780 /* Disable transmits before FDIR Re-initialization */
1781 netif_tx_stop_all_queues(netdev);
1782 for (i = 0; i < adapter->num_tx_queues; i++) {
1783 struct ixgbe_ring *tx_ring =
1784 adapter->tx_ring[i];
1785 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
1787 schedule_work(&adapter->fdir_reinit_task);
1791 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1792 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_OTHER);
1797 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
1802 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1803 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1804 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1806 mask = (qmask & 0xFFFFFFFF);
1807 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(0), mask);
1808 mask = (qmask >> 32);
1809 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(1), mask);
1811 /* skip the flush */
1814 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
1819 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1820 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1821 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, mask);
1823 mask = (qmask & 0xFFFFFFFF);
1824 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), mask);
1825 mask = (qmask >> 32);
1826 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), mask);
1828 /* skip the flush */
1831 static irqreturn_t ixgbe_msix_clean_tx(int irq, void *data)
1833 struct ixgbe_q_vector *q_vector = data;
1834 struct ixgbe_adapter *adapter = q_vector->adapter;
1835 struct ixgbe_ring *tx_ring;
1838 if (!q_vector->txr_count)
1841 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1842 for (i = 0; i < q_vector->txr_count; i++) {
1843 tx_ring = adapter->tx_ring[r_idx];
1844 tx_ring->total_bytes = 0;
1845 tx_ring->total_packets = 0;
1846 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1850 /* EIAM disabled interrupts (on this vector) for us */
1851 napi_schedule(&q_vector->napi);
1857 * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
1859 * @data: pointer to our q_vector struct for this interrupt vector
1861 static irqreturn_t ixgbe_msix_clean_rx(int irq, void *data)
1863 struct ixgbe_q_vector *q_vector = data;
1864 struct ixgbe_adapter *adapter = q_vector->adapter;
1865 struct ixgbe_ring *rx_ring;
1869 #ifdef CONFIG_IXGBE_DCA
1870 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1871 ixgbe_update_dca(q_vector);
1874 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1875 for (i = 0; i < q_vector->rxr_count; i++) {
1876 rx_ring = adapter->rx_ring[r_idx];
1877 rx_ring->total_bytes = 0;
1878 rx_ring->total_packets = 0;
1879 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1883 if (!q_vector->rxr_count)
1886 /* EIAM disabled interrupts (on this vector) for us */
1887 napi_schedule(&q_vector->napi);
1892 static irqreturn_t ixgbe_msix_clean_many(int irq, void *data)
1894 struct ixgbe_q_vector *q_vector = data;
1895 struct ixgbe_adapter *adapter = q_vector->adapter;
1896 struct ixgbe_ring *ring;
1900 if (!q_vector->txr_count && !q_vector->rxr_count)
1903 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1904 for (i = 0; i < q_vector->txr_count; i++) {
1905 ring = adapter->tx_ring[r_idx];
1906 ring->total_bytes = 0;
1907 ring->total_packets = 0;
1908 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1912 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1913 for (i = 0; i < q_vector->rxr_count; i++) {
1914 ring = adapter->rx_ring[r_idx];
1915 ring->total_bytes = 0;
1916 ring->total_packets = 0;
1917 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1921 /* EIAM disabled interrupts (on this vector) for us */
1922 napi_schedule(&q_vector->napi);
1928 * ixgbe_clean_rxonly - msix (aka one shot) rx clean routine
1929 * @napi: napi struct with our devices info in it
1930 * @budget: amount of work driver is allowed to do this pass, in packets
1932 * This function is optimized for cleaning one queue only on a single
1935 static int ixgbe_clean_rxonly(struct napi_struct *napi, int budget)
1937 struct ixgbe_q_vector *q_vector =
1938 container_of(napi, struct ixgbe_q_vector, napi);
1939 struct ixgbe_adapter *adapter = q_vector->adapter;
1940 struct ixgbe_ring *rx_ring = NULL;
1944 #ifdef CONFIG_IXGBE_DCA
1945 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1946 ixgbe_update_dca(q_vector);
1949 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1950 rx_ring = adapter->rx_ring[r_idx];
1952 ixgbe_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
1954 /* If all Rx work done, exit the polling mode */
1955 if (work_done < budget) {
1956 napi_complete(napi);
1957 if (adapter->rx_itr_setting & 1)
1958 ixgbe_set_itr_msix(q_vector);
1959 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1960 ixgbe_irq_enable_queues(adapter,
1961 ((u64)1 << q_vector->v_idx));
1968 * ixgbe_clean_rxtx_many - msix (aka one shot) rx clean routine
1969 * @napi: napi struct with our devices info in it
1970 * @budget: amount of work driver is allowed to do this pass, in packets
1972 * This function will clean more than one rx queue associated with a
1975 static int ixgbe_clean_rxtx_many(struct napi_struct *napi, int budget)
1977 struct ixgbe_q_vector *q_vector =
1978 container_of(napi, struct ixgbe_q_vector, napi);
1979 struct ixgbe_adapter *adapter = q_vector->adapter;
1980 struct ixgbe_ring *ring = NULL;
1981 int work_done = 0, i;
1983 bool tx_clean_complete = true;
1985 #ifdef CONFIG_IXGBE_DCA
1986 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1987 ixgbe_update_dca(q_vector);
1990 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1991 for (i = 0; i < q_vector->txr_count; i++) {
1992 ring = adapter->tx_ring[r_idx];
1993 tx_clean_complete &= ixgbe_clean_tx_irq(q_vector, ring);
1994 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1998 /* attempt to distribute budget to each queue fairly, but don't allow
1999 * the budget to go below 1 because we'll exit polling */
2000 budget /= (q_vector->rxr_count ?: 1);
2001 budget = max(budget, 1);
2002 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2003 for (i = 0; i < q_vector->rxr_count; i++) {
2004 ring = adapter->rx_ring[r_idx];
2005 ixgbe_clean_rx_irq(q_vector, ring, &work_done, budget);
2006 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
2010 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2011 ring = adapter->rx_ring[r_idx];
2012 /* If all Rx work done, exit the polling mode */
2013 if (work_done < budget) {
2014 napi_complete(napi);
2015 if (adapter->rx_itr_setting & 1)
2016 ixgbe_set_itr_msix(q_vector);
2017 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2018 ixgbe_irq_enable_queues(adapter,
2019 ((u64)1 << q_vector->v_idx));
2027 * ixgbe_clean_txonly - msix (aka one shot) tx clean routine
2028 * @napi: napi struct with our devices info in it
2029 * @budget: amount of work driver is allowed to do this pass, in packets
2031 * This function is optimized for cleaning one queue only on a single
2034 static int ixgbe_clean_txonly(struct napi_struct *napi, int budget)
2036 struct ixgbe_q_vector *q_vector =
2037 container_of(napi, struct ixgbe_q_vector, napi);
2038 struct ixgbe_adapter *adapter = q_vector->adapter;
2039 struct ixgbe_ring *tx_ring = NULL;
2043 #ifdef CONFIG_IXGBE_DCA
2044 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2045 ixgbe_update_dca(q_vector);
2048 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
2049 tx_ring = adapter->tx_ring[r_idx];
2051 if (!ixgbe_clean_tx_irq(q_vector, tx_ring))
2054 /* If all Tx work done, exit the polling mode */
2055 if (work_done < budget) {
2056 napi_complete(napi);
2057 if (adapter->tx_itr_setting & 1)
2058 ixgbe_set_itr_msix(q_vector);
2059 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2060 ixgbe_irq_enable_queues(adapter,
2061 ((u64)1 << q_vector->v_idx));
2067 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
2070 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2072 set_bit(r_idx, q_vector->rxr_idx);
2073 q_vector->rxr_count++;
2076 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
2079 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2081 set_bit(t_idx, q_vector->txr_idx);
2082 q_vector->txr_count++;
2086 * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
2087 * @adapter: board private structure to initialize
2088 * @vectors: allotted vector count for descriptor rings
2090 * This function maps descriptor rings to the queue-specific vectors
2091 * we were allotted through the MSI-X enabling code. Ideally, we'd have
2092 * one vector per ring/queue, but on a constrained vector budget, we
2093 * group the rings as "efficiently" as possible. You would add new
2094 * mapping configurations in here.
2096 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter,
2100 int rxr_idx = 0, txr_idx = 0;
2101 int rxr_remaining = adapter->num_rx_queues;
2102 int txr_remaining = adapter->num_tx_queues;
2107 /* No mapping required if MSI-X is disabled. */
2108 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2112 * The ideal configuration...
2113 * We have enough vectors to map one per queue.
2115 if (vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
2116 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
2117 map_vector_to_rxq(adapter, v_start, rxr_idx);
2119 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
2120 map_vector_to_txq(adapter, v_start, txr_idx);
2126 * If we don't have enough vectors for a 1-to-1
2127 * mapping, we'll have to group them so there are
2128 * multiple queues per vector.
2130 /* Re-adjusting *qpv takes care of the remainder. */
2131 for (i = v_start; i < vectors; i++) {
2132 rqpv = DIV_ROUND_UP(rxr_remaining, vectors - i);
2133 for (j = 0; j < rqpv; j++) {
2134 map_vector_to_rxq(adapter, i, rxr_idx);
2139 for (i = v_start; i < vectors; i++) {
2140 tqpv = DIV_ROUND_UP(txr_remaining, vectors - i);
2141 for (j = 0; j < tqpv; j++) {
2142 map_vector_to_txq(adapter, i, txr_idx);
2153 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2154 * @adapter: board private structure
2156 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2157 * interrupts from the kernel.
2159 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2161 struct net_device *netdev = adapter->netdev;
2162 irqreturn_t (*handler)(int, void *);
2163 int i, vector, q_vectors, err;
2166 /* Decrement for Other and TCP Timer vectors */
2167 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2169 /* Map the Tx/Rx rings to the vectors we were allotted. */
2170 err = ixgbe_map_rings_to_vectors(adapter, q_vectors);
2174 #define SET_HANDLER(_v) ((!(_v)->rxr_count) ? &ixgbe_msix_clean_tx : \
2175 (!(_v)->txr_count) ? &ixgbe_msix_clean_rx : \
2176 &ixgbe_msix_clean_many)
2177 for (vector = 0; vector < q_vectors; vector++) {
2178 handler = SET_HANDLER(adapter->q_vector[vector]);
2180 if (handler == &ixgbe_msix_clean_rx) {
2181 sprintf(adapter->name[vector], "%s-%s-%d",
2182 netdev->name, "rx", ri++);
2183 } else if (handler == &ixgbe_msix_clean_tx) {
2184 sprintf(adapter->name[vector], "%s-%s-%d",
2185 netdev->name, "tx", ti++);
2187 sprintf(adapter->name[vector], "%s-%s-%d",
2188 netdev->name, "TxRx", vector);
2190 err = request_irq(adapter->msix_entries[vector].vector,
2191 handler, 0, adapter->name[vector],
2192 adapter->q_vector[vector]);
2194 e_err(probe, "request_irq failed for MSIX interrupt "
2195 "Error: %d\n", err);
2196 goto free_queue_irqs;
2200 sprintf(adapter->name[vector], "%s:lsc", netdev->name);
2201 err = request_irq(adapter->msix_entries[vector].vector,
2202 ixgbe_msix_lsc, 0, adapter->name[vector], netdev);
2204 e_err(probe, "request_irq for msix_lsc failed: %d\n", err);
2205 goto free_queue_irqs;
2211 for (i = vector - 1; i >= 0; i--)
2212 free_irq(adapter->msix_entries[--vector].vector,
2213 adapter->q_vector[i]);
2214 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2215 pci_disable_msix(adapter->pdev);
2216 kfree(adapter->msix_entries);
2217 adapter->msix_entries = NULL;
2222 static void ixgbe_set_itr(struct ixgbe_adapter *adapter)
2224 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2226 u32 new_itr = q_vector->eitr;
2227 struct ixgbe_ring *rx_ring = adapter->rx_ring[0];
2228 struct ixgbe_ring *tx_ring = adapter->tx_ring[0];
2230 q_vector->tx_itr = ixgbe_update_itr(adapter, new_itr,
2232 tx_ring->total_packets,
2233 tx_ring->total_bytes);
2234 q_vector->rx_itr = ixgbe_update_itr(adapter, new_itr,
2236 rx_ring->total_packets,
2237 rx_ring->total_bytes);
2239 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
2241 switch (current_itr) {
2242 /* counts and packets in update_itr are dependent on these numbers */
2243 case lowest_latency:
2247 new_itr = 20000; /* aka hwitr = ~200 */
2256 if (new_itr != q_vector->eitr) {
2257 /* do an exponential smoothing */
2258 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
2260 /* save the algorithm value here, not the smoothed one */
2261 q_vector->eitr = new_itr;
2263 ixgbe_write_eitr(q_vector);
2268 * ixgbe_irq_enable - Enable default interrupt generation settings
2269 * @adapter: board private structure
2271 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2276 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2277 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2278 mask |= IXGBE_EIMS_GPI_SDP0;
2279 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2280 mask |= IXGBE_EIMS_GPI_SDP1;
2281 if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
2282 mask |= IXGBE_EIMS_ECC;
2283 mask |= IXGBE_EIMS_GPI_SDP1;
2284 mask |= IXGBE_EIMS_GPI_SDP2;
2285 if (adapter->num_vfs)
2286 mask |= IXGBE_EIMS_MAILBOX;
2288 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
2289 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
2290 mask |= IXGBE_EIMS_FLOW_DIR;
2292 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2294 ixgbe_irq_enable_queues(adapter, ~0);
2296 IXGBE_WRITE_FLUSH(&adapter->hw);
2298 if (adapter->num_vfs > 32) {
2299 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
2300 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2305 * ixgbe_intr - legacy mode Interrupt Handler
2306 * @irq: interrupt number
2307 * @data: pointer to a network interface device structure
2309 static irqreturn_t ixgbe_intr(int irq, void *data)
2311 struct net_device *netdev = data;
2312 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2313 struct ixgbe_hw *hw = &adapter->hw;
2314 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2318 * Workaround for silicon errata on 82598. Mask the interrupts
2319 * before the read of EICR.
2321 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2323 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2324 * therefore no explict interrupt disable is necessary */
2325 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2328 * shared interrupt alert!
2329 * make sure interrupts are enabled because the read will
2330 * have disabled interrupts due to EIAM
2331 * finish the workaround of silicon errata on 82598. Unmask
2332 * the interrupt that we masked before the EICR read.
2334 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2335 ixgbe_irq_enable(adapter, true, true);
2336 return IRQ_NONE; /* Not our interrupt */
2339 if (eicr & IXGBE_EICR_LSC)
2340 ixgbe_check_lsc(adapter);
2342 if (hw->mac.type == ixgbe_mac_82599EB)
2343 ixgbe_check_sfp_event(adapter, eicr);
2345 ixgbe_check_fan_failure(adapter, eicr);
2346 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2347 ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)))
2348 schedule_work(&adapter->check_overtemp_task);
2350 if (napi_schedule_prep(&(q_vector->napi))) {
2351 adapter->tx_ring[0]->total_packets = 0;
2352 adapter->tx_ring[0]->total_bytes = 0;
2353 adapter->rx_ring[0]->total_packets = 0;
2354 adapter->rx_ring[0]->total_bytes = 0;
2355 /* would disable interrupts here but EIAM disabled it */
2356 __napi_schedule(&(q_vector->napi));
2360 * re-enable link(maybe) and non-queue interrupts, no flush.
2361 * ixgbe_poll will re-enable the queue interrupts
2364 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2365 ixgbe_irq_enable(adapter, false, false);
2370 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
2372 int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2374 for (i = 0; i < q_vectors; i++) {
2375 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
2376 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
2377 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
2378 q_vector->rxr_count = 0;
2379 q_vector->txr_count = 0;
2384 * ixgbe_request_irq - initialize interrupts
2385 * @adapter: board private structure
2387 * Attempts to configure interrupts using the best available
2388 * capabilities of the hardware and kernel.
2390 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2392 struct net_device *netdev = adapter->netdev;
2395 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2396 err = ixgbe_request_msix_irqs(adapter);
2397 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
2398 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2399 netdev->name, netdev);
2401 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2402 netdev->name, netdev);
2406 e_err(probe, "request_irq failed, Error %d\n", err);
2411 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2413 struct net_device *netdev = adapter->netdev;
2415 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2418 q_vectors = adapter->num_msix_vectors;
2421 free_irq(adapter->msix_entries[i].vector, netdev);
2424 for (; i >= 0; i--) {
2425 free_irq(adapter->msix_entries[i].vector,
2426 adapter->q_vector[i]);
2429 ixgbe_reset_q_vectors(adapter);
2431 free_irq(adapter->pdev->irq, netdev);
2436 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2437 * @adapter: board private structure
2439 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2441 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2442 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2444 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2445 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2446 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2447 if (adapter->num_vfs > 32)
2448 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
2450 IXGBE_WRITE_FLUSH(&adapter->hw);
2451 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2453 for (i = 0; i < adapter->num_msix_vectors; i++)
2454 synchronize_irq(adapter->msix_entries[i].vector);
2456 synchronize_irq(adapter->pdev->irq);
2461 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2464 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2466 struct ixgbe_hw *hw = &adapter->hw;
2468 IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
2469 EITR_INTS_PER_SEC_TO_REG(adapter->rx_eitr_param));
2471 ixgbe_set_ivar(adapter, 0, 0, 0);
2472 ixgbe_set_ivar(adapter, 1, 0, 0);
2474 map_vector_to_rxq(adapter, 0, 0);
2475 map_vector_to_txq(adapter, 0, 0);
2477 e_info(hw, "Legacy interrupt IVAR setup done\n");
2481 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2482 * @adapter: board private structure
2483 * @ring: structure containing ring specific data
2485 * Configure the Tx descriptor ring after a reset.
2487 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2488 struct ixgbe_ring *ring)
2490 struct ixgbe_hw *hw = &adapter->hw;
2491 u64 tdba = ring->dma;
2494 u16 reg_idx = ring->reg_idx;
2496 /* disable queue to avoid issues while updating state */
2497 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2498 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx),
2499 txdctl & ~IXGBE_TXDCTL_ENABLE);
2500 IXGBE_WRITE_FLUSH(hw);
2502 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2503 (tdba & DMA_BIT_MASK(32)));
2504 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2505 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2506 ring->count * sizeof(union ixgbe_adv_tx_desc));
2507 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2508 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2509 ring->tail = hw->hw_addr + IXGBE_TDT(reg_idx);
2511 /* configure fetching thresholds */
2512 if (adapter->rx_itr_setting == 0) {
2513 /* cannot set wthresh when itr==0 */
2514 txdctl &= ~0x007F0000;
2516 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
2517 txdctl |= (8 << 16);
2519 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2520 /* PThresh workaround for Tx hang with DFP enabled. */
2524 /* reinitialize flowdirector state */
2525 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
2528 txdctl |= IXGBE_TXDCTL_ENABLE;
2529 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
2531 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2532 if (hw->mac.type == ixgbe_mac_82598EB &&
2533 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2536 /* poll to verify queue is enabled */
2539 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2540 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
2542 e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
2545 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
2547 struct ixgbe_hw *hw = &adapter->hw;
2551 if (hw->mac.type == ixgbe_mac_82598EB)
2554 /* disable the arbiter while setting MTQC */
2555 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2556 rttdcs |= IXGBE_RTTDCS_ARBDIS;
2557 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2559 /* set transmit pool layout */
2560 mask = (IXGBE_FLAG_SRIOV_ENABLED | IXGBE_FLAG_DCB_ENABLED);
2561 switch (adapter->flags & mask) {
2563 case (IXGBE_FLAG_SRIOV_ENABLED):
2564 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2565 (IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF));
2568 case (IXGBE_FLAG_DCB_ENABLED):
2569 /* We enable 8 traffic classes, DCB only */
2570 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2571 (IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ));
2575 IXGBE_WRITE_REG(hw, IXGBE_MTQC, IXGBE_MTQC_64Q_1PB);
2579 /* re-enable the arbiter */
2580 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2581 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2585 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2586 * @adapter: board private structure
2588 * Configure the Tx unit of the MAC after a reset.
2590 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2592 struct ixgbe_hw *hw = &adapter->hw;
2596 ixgbe_setup_mtqc(adapter);
2598 if (hw->mac.type != ixgbe_mac_82598EB) {
2599 /* DMATXCTL.EN must be before Tx queues are enabled */
2600 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2601 dmatxctl |= IXGBE_DMATXCTL_TE;
2602 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2605 /* Setup the HW Tx Head and Tail descriptor pointers */
2606 for (i = 0; i < adapter->num_tx_queues; i++)
2607 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
2610 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2612 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2613 struct ixgbe_ring *rx_ring)
2617 struct ixgbe_ring_feature *feature = adapter->ring_feature;
2619 index = rx_ring->reg_idx;
2620 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2622 mask = (unsigned long) feature[RING_F_RSS].mask;
2623 index = index & mask;
2625 srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(index));
2627 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
2628 srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
2629 if (adapter->num_vfs)
2630 srrctl |= IXGBE_SRRCTL_DROP_EN;
2632 srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
2633 IXGBE_SRRCTL_BSIZEHDR_MASK;
2635 if (ring_is_ps_enabled(rx_ring)) {
2636 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
2637 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2639 srrctl |= (PAGE_SIZE / 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2641 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2643 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
2644 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2645 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2648 IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(index), srrctl);
2651 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2653 struct ixgbe_hw *hw = &adapter->hw;
2654 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2655 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2656 0x6A3E67EA, 0x14364D17, 0x3BED200D};
2657 u32 mrqc = 0, reta = 0;
2662 /* Fill out hash function seeds */
2663 for (i = 0; i < 10; i++)
2664 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2666 /* Fill out redirection table */
2667 for (i = 0, j = 0; i < 128; i++, j++) {
2668 if (j == adapter->ring_feature[RING_F_RSS].indices)
2670 /* reta = 4-byte sliding window of
2671 * 0x00..(indices-1)(indices-1)00..etc. */
2672 reta = (reta << 8) | (j * 0x11);
2674 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2677 /* Disable indicating checksum in descriptor, enables RSS hash */
2678 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2679 rxcsum |= IXGBE_RXCSUM_PCSD;
2680 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2682 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
2683 mask = adapter->flags & IXGBE_FLAG_RSS_ENABLED;
2685 mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2686 #ifdef CONFIG_IXGBE_DCB
2687 | IXGBE_FLAG_DCB_ENABLED
2689 | IXGBE_FLAG_SRIOV_ENABLED
2693 case (IXGBE_FLAG_RSS_ENABLED):
2694 mrqc = IXGBE_MRQC_RSSEN;
2696 case (IXGBE_FLAG_SRIOV_ENABLED):
2697 mrqc = IXGBE_MRQC_VMDQEN;
2699 #ifdef CONFIG_IXGBE_DCB
2700 case (IXGBE_FLAG_DCB_ENABLED):
2701 mrqc = IXGBE_MRQC_RT8TCEN;
2703 #endif /* CONFIG_IXGBE_DCB */
2708 /* Perform hash on these packet types */
2709 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2710 | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2711 | IXGBE_MRQC_RSS_FIELD_IPV6
2712 | IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
2714 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2718 * ixgbe_configure_rscctl - enable RSC for the indicated ring
2719 * @adapter: address of board private structure
2720 * @index: index of ring to set
2722 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
2723 struct ixgbe_ring *ring)
2725 struct ixgbe_hw *hw = &adapter->hw;
2728 u16 reg_idx = ring->reg_idx;
2730 if (!ring_is_rsc_enabled(ring))
2733 rx_buf_len = ring->rx_buf_len;
2734 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
2735 rscctrl |= IXGBE_RSCCTL_RSCEN;
2737 * we must limit the number of descriptors so that the
2738 * total size of max desc * buf_len is not greater
2741 if (ring_is_ps_enabled(ring)) {
2742 #if (MAX_SKB_FRAGS > 16)
2743 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2744 #elif (MAX_SKB_FRAGS > 8)
2745 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2746 #elif (MAX_SKB_FRAGS > 4)
2747 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2749 rscctrl |= IXGBE_RSCCTL_MAXDESC_1;
2752 if (rx_buf_len < IXGBE_RXBUFFER_4096)
2753 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2754 else if (rx_buf_len < IXGBE_RXBUFFER_8192)
2755 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2757 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2759 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
2763 * ixgbe_set_uta - Set unicast filter table address
2764 * @adapter: board private structure
2766 * The unicast table address is a register array of 32-bit registers.
2767 * The table is meant to be used in a way similar to how the MTA is used
2768 * however due to certain limitations in the hardware it is necessary to
2769 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
2770 * enable bit to allow vlan tag stripping when promiscuous mode is enabled
2772 static void ixgbe_set_uta(struct ixgbe_adapter *adapter)
2774 struct ixgbe_hw *hw = &adapter->hw;
2777 /* The UTA table only exists on 82599 hardware and newer */
2778 if (hw->mac.type < ixgbe_mac_82599EB)
2781 /* we only need to do this if VMDq is enabled */
2782 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
2785 for (i = 0; i < 128; i++)
2786 IXGBE_WRITE_REG(hw, IXGBE_UTA(i), ~0);
2789 #define IXGBE_MAX_RX_DESC_POLL 10
2790 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
2791 struct ixgbe_ring *ring)
2793 struct ixgbe_hw *hw = &adapter->hw;
2794 int reg_idx = ring->reg_idx;
2795 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
2798 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2799 if (hw->mac.type == ixgbe_mac_82598EB &&
2800 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2805 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2806 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
2809 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
2810 "the polling period\n", reg_idx);
2814 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
2815 struct ixgbe_ring *ring)
2817 struct ixgbe_hw *hw = &adapter->hw;
2818 u64 rdba = ring->dma;
2820 u16 reg_idx = ring->reg_idx;
2822 /* disable queue to avoid issues while updating state */
2823 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2824 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx),
2825 rxdctl & ~IXGBE_RXDCTL_ENABLE);
2826 IXGBE_WRITE_FLUSH(hw);
2828 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
2829 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
2830 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
2831 ring->count * sizeof(union ixgbe_adv_rx_desc));
2832 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
2833 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
2834 ring->tail = hw->hw_addr + IXGBE_RDT(reg_idx);
2836 ixgbe_configure_srrctl(adapter, ring);
2837 ixgbe_configure_rscctl(adapter, ring);
2839 if (hw->mac.type == ixgbe_mac_82598EB) {
2841 * enable cache line friendly hardware writes:
2842 * PTHRESH=32 descriptors (half the internal cache),
2843 * this also removes ugly rx_no_buffer_count increment
2844 * HTHRESH=4 descriptors (to minimize latency on fetch)
2845 * WTHRESH=8 burst writeback up to two cache lines
2847 rxdctl &= ~0x3FFFFF;
2851 /* enable receive descriptor ring */
2852 rxdctl |= IXGBE_RXDCTL_ENABLE;
2853 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
2855 ixgbe_rx_desc_queue_enable(adapter, ring);
2856 ixgbe_alloc_rx_buffers(ring, IXGBE_DESC_UNUSED(ring));
2859 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
2861 struct ixgbe_hw *hw = &adapter->hw;
2864 /* PSRTYPE must be initialized in non 82598 adapters */
2865 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
2866 IXGBE_PSRTYPE_UDPHDR |
2867 IXGBE_PSRTYPE_IPV4HDR |
2868 IXGBE_PSRTYPE_L2HDR |
2869 IXGBE_PSRTYPE_IPV6HDR;
2871 if (hw->mac.type == ixgbe_mac_82598EB)
2874 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED)
2875 psrtype |= (adapter->num_rx_queues_per_pool << 29);
2877 for (p = 0; p < adapter->num_rx_pools; p++)
2878 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(adapter->num_vfs + p),
2882 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
2884 struct ixgbe_hw *hw = &adapter->hw;
2887 u32 reg_offset, vf_shift;
2890 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
2893 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
2894 vt_reg_bits = IXGBE_VMD_CTL_VMDQ_EN | IXGBE_VT_CTL_REPLEN;
2895 vt_reg_bits |= (adapter->num_vfs << IXGBE_VT_CTL_POOL_SHIFT);
2896 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl | vt_reg_bits);
2898 vf_shift = adapter->num_vfs % 32;
2899 reg_offset = (adapter->num_vfs > 32) ? 1 : 0;
2901 /* Enable only the PF's pool for Tx/Rx */
2902 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (1 << vf_shift));
2903 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), 0);
2904 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (1 << vf_shift));
2905 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), 0);
2906 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
2908 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
2909 hw->mac.ops.set_vmdq(hw, 0, adapter->num_vfs);
2912 * Set up VF register offsets for selected VT Mode,
2913 * i.e. 32 or 64 VFs for SR-IOV
2915 gcr_ext = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
2916 gcr_ext |= IXGBE_GCR_EXT_MSIX_EN;
2917 gcr_ext |= IXGBE_GCR_EXT_VT_MODE_64;
2918 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
2920 /* enable Tx loopback for VF/PF communication */
2921 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
2924 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
2926 struct ixgbe_hw *hw = &adapter->hw;
2927 struct net_device *netdev = adapter->netdev;
2928 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2930 struct ixgbe_ring *rx_ring;
2934 /* Decide whether to use packet split mode or not */
2935 /* Do not use packet split if we're in SR-IOV Mode */
2936 if (!adapter->num_vfs)
2937 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
2939 /* Set the RX buffer length according to the mode */
2940 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
2941 rx_buf_len = IXGBE_RX_HDR_SIZE;
2943 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
2944 (netdev->mtu <= ETH_DATA_LEN))
2945 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
2947 rx_buf_len = ALIGN(max_frame + VLAN_HLEN, 1024);
2951 /* adjust max frame to be able to do baby jumbo for FCoE */
2952 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
2953 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
2954 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
2956 #endif /* IXGBE_FCOE */
2957 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
2958 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
2959 mhadd &= ~IXGBE_MHADD_MFS_MASK;
2960 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
2962 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
2965 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2966 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
2967 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2968 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
2971 * Setup the HW Rx Head and Tail Descriptor Pointers and
2972 * the Base and Length of the Rx Descriptor Ring
2974 for (i = 0; i < adapter->num_rx_queues; i++) {
2975 rx_ring = adapter->rx_ring[i];
2976 rx_ring->rx_buf_len = rx_buf_len;
2978 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)
2979 set_ring_ps_enabled(rx_ring);
2981 clear_ring_ps_enabled(rx_ring);
2983 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
2984 set_ring_rsc_enabled(rx_ring);
2986 clear_ring_rsc_enabled(rx_ring);
2989 if (netdev->features & NETIF_F_FCOE_MTU) {
2990 struct ixgbe_ring_feature *f;
2991 f = &adapter->ring_feature[RING_F_FCOE];
2992 if ((i >= f->mask) && (i < f->mask + f->indices)) {
2993 clear_ring_ps_enabled(rx_ring);
2994 if (rx_buf_len < IXGBE_FCOE_JUMBO_FRAME_SIZE)
2995 rx_ring->rx_buf_len =
2996 IXGBE_FCOE_JUMBO_FRAME_SIZE;
2997 } else if (!ring_is_rsc_enabled(rx_ring) &&
2998 !ring_is_ps_enabled(rx_ring)) {
2999 rx_ring->rx_buf_len =
3000 IXGBE_FCOE_JUMBO_FRAME_SIZE;
3003 #endif /* IXGBE_FCOE */
3008 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
3010 struct ixgbe_hw *hw = &adapter->hw;
3011 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
3013 switch (hw->mac.type) {
3014 case ixgbe_mac_82598EB:
3016 * For VMDq support of different descriptor types or
3017 * buffer sizes through the use of multiple SRRCTL
3018 * registers, RDRXCTL.MVMEN must be set to 1
3020 * also, the manual doesn't mention it clearly but DCA hints
3021 * will only use queue 0's tags unless this bit is set. Side
3022 * effects of setting this bit are only that SRRCTL must be
3023 * fully programmed [0..15]
3025 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
3027 case ixgbe_mac_82599EB:
3028 /* Disable RSC for ACK packets */
3029 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
3030 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
3031 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3032 /* hardware requires some bits to be set by default */
3033 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3034 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3037 /* We should do nothing since we don't know this hardware */
3041 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3045 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3046 * @adapter: board private structure
3048 * Configure the Rx unit of the MAC after a reset.
3050 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3052 struct ixgbe_hw *hw = &adapter->hw;
3056 /* disable receives while setting up the descriptors */
3057 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3058 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3060 ixgbe_setup_psrtype(adapter);
3061 ixgbe_setup_rdrxctl(adapter);
3063 /* Program registers for the distribution of queues */
3064 ixgbe_setup_mrqc(adapter);
3066 ixgbe_set_uta(adapter);
3068 /* set_rx_buffer_len must be called before ring initialization */
3069 ixgbe_set_rx_buffer_len(adapter);
3072 * Setup the HW Rx Head and Tail Descriptor Pointers and
3073 * the Base and Length of the Rx Descriptor Ring
3075 for (i = 0; i < adapter->num_rx_queues; i++)
3076 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3078 /* disable drop enable for 82598 parts */
3079 if (hw->mac.type == ixgbe_mac_82598EB)
3080 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3082 /* enable all receives */
3083 rxctrl |= IXGBE_RXCTRL_RXEN;
3084 hw->mac.ops.enable_rx_dma(hw, rxctrl);
3087 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3089 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3090 struct ixgbe_hw *hw = &adapter->hw;
3091 int pool_ndx = adapter->num_vfs;
3093 /* add VID to filter table */
3094 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, true);
3095 set_bit(vid, adapter->active_vlans);
3098 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3100 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3101 struct ixgbe_hw *hw = &adapter->hw;
3102 int pool_ndx = adapter->num_vfs;
3104 /* remove VID from filter table */
3105 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, false);
3106 clear_bit(vid, adapter->active_vlans);
3110 * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3111 * @adapter: driver data
3113 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
3115 struct ixgbe_hw *hw = &adapter->hw;
3118 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3119 vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
3120 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3124 * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3125 * @adapter: driver data
3127 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
3129 struct ixgbe_hw *hw = &adapter->hw;
3132 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3133 vlnctrl |= IXGBE_VLNCTRL_VFE;
3134 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3135 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3139 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3140 * @adapter: driver data
3142 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3144 struct ixgbe_hw *hw = &adapter->hw;
3148 switch (hw->mac.type) {
3149 case ixgbe_mac_82598EB:
3150 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3151 vlnctrl &= ~IXGBE_VLNCTRL_VME;
3152 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3154 case ixgbe_mac_82599EB:
3155 for (i = 0; i < adapter->num_rx_queues; i++) {
3156 j = adapter->rx_ring[i]->reg_idx;
3157 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3158 vlnctrl &= ~IXGBE_RXDCTL_VME;
3159 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3168 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3169 * @adapter: driver data
3171 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3173 struct ixgbe_hw *hw = &adapter->hw;
3177 switch (hw->mac.type) {
3178 case ixgbe_mac_82598EB:
3179 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3180 vlnctrl |= IXGBE_VLNCTRL_VME;
3181 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3183 case ixgbe_mac_82599EB:
3184 for (i = 0; i < adapter->num_rx_queues; i++) {
3185 j = adapter->rx_ring[i]->reg_idx;
3186 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3187 vlnctrl |= IXGBE_RXDCTL_VME;
3188 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3196 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3200 ixgbe_vlan_rx_add_vid(adapter->netdev, 0);
3202 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3203 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
3207 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3208 * @netdev: network interface device structure
3210 * Writes unicast address list to the RAR table.
3211 * Returns: -ENOMEM on failure/insufficient address space
3212 * 0 on no addresses written
3213 * X on writing X addresses to the RAR table
3215 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3217 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3218 struct ixgbe_hw *hw = &adapter->hw;
3219 unsigned int vfn = adapter->num_vfs;
3220 unsigned int rar_entries = hw->mac.num_rar_entries - (vfn + 1);
3223 /* return ENOMEM indicating insufficient memory for addresses */
3224 if (netdev_uc_count(netdev) > rar_entries)
3227 if (!netdev_uc_empty(netdev) && rar_entries) {
3228 struct netdev_hw_addr *ha;
3229 /* return error if we do not support writing to RAR table */
3230 if (!hw->mac.ops.set_rar)
3233 netdev_for_each_uc_addr(ha, netdev) {
3236 hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3241 /* write the addresses in reverse order to avoid write combining */
3242 for (; rar_entries > 0 ; rar_entries--)
3243 hw->mac.ops.clear_rar(hw, rar_entries);
3249 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3250 * @netdev: network interface device structure
3252 * The set_rx_method entry point is called whenever the unicast/multicast
3253 * address list or the network interface flags are updated. This routine is
3254 * responsible for configuring the hardware for proper unicast, multicast and
3257 void ixgbe_set_rx_mode(struct net_device *netdev)
3259 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3260 struct ixgbe_hw *hw = &adapter->hw;
3261 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3264 /* Check for Promiscuous and All Multicast modes */
3266 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3268 /* set all bits that we expect to always be set */
3269 fctrl |= IXGBE_FCTRL_BAM;
3270 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3271 fctrl |= IXGBE_FCTRL_PMCF;
3273 /* clear the bits we are changing the status of */
3274 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3276 if (netdev->flags & IFF_PROMISC) {
3277 hw->addr_ctrl.user_set_promisc = true;
3278 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3279 vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3280 /* don't hardware filter vlans in promisc mode */
3281 ixgbe_vlan_filter_disable(adapter);
3283 if (netdev->flags & IFF_ALLMULTI) {
3284 fctrl |= IXGBE_FCTRL_MPE;
3285 vmolr |= IXGBE_VMOLR_MPE;
3288 * Write addresses to the MTA, if the attempt fails
3289 * then we should just turn on promiscous mode so
3290 * that we can at least receive multicast traffic
3292 hw->mac.ops.update_mc_addr_list(hw, netdev);
3293 vmolr |= IXGBE_VMOLR_ROMPE;
3295 ixgbe_vlan_filter_enable(adapter);
3296 hw->addr_ctrl.user_set_promisc = false;
3298 * Write addresses to available RAR registers, if there is not
3299 * sufficient space to store all the addresses then enable
3300 * unicast promiscous mode
3302 count = ixgbe_write_uc_addr_list(netdev);
3304 fctrl |= IXGBE_FCTRL_UPE;
3305 vmolr |= IXGBE_VMOLR_ROPE;
3309 if (adapter->num_vfs) {
3310 ixgbe_restore_vf_multicasts(adapter);
3311 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(adapter->num_vfs)) &
3312 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3314 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(adapter->num_vfs), vmolr);
3317 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3319 if (netdev->features & NETIF_F_HW_VLAN_RX)
3320 ixgbe_vlan_strip_enable(adapter);
3322 ixgbe_vlan_strip_disable(adapter);
3325 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3328 struct ixgbe_q_vector *q_vector;
3329 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3331 /* legacy and MSI only use one vector */
3332 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3335 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3336 struct napi_struct *napi;
3337 q_vector = adapter->q_vector[q_idx];
3338 napi = &q_vector->napi;
3339 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3340 if (!q_vector->rxr_count || !q_vector->txr_count) {
3341 if (q_vector->txr_count == 1)
3342 napi->poll = &ixgbe_clean_txonly;
3343 else if (q_vector->rxr_count == 1)
3344 napi->poll = &ixgbe_clean_rxonly;
3352 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3355 struct ixgbe_q_vector *q_vector;
3356 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3358 /* legacy and MSI only use one vector */
3359 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3362 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3363 q_vector = adapter->q_vector[q_idx];
3364 napi_disable(&q_vector->napi);
3368 #ifdef CONFIG_IXGBE_DCB
3370 * ixgbe_configure_dcb - Configure DCB hardware
3371 * @adapter: ixgbe adapter struct
3373 * This is called by the driver on open to configure the DCB hardware.
3374 * This is also called by the gennetlink interface when reconfiguring
3377 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3379 struct ixgbe_hw *hw = &adapter->hw;
3380 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3382 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3383 if (hw->mac.type == ixgbe_mac_82598EB)
3384 netif_set_gso_max_size(adapter->netdev, 65536);
3388 if (hw->mac.type == ixgbe_mac_82598EB)
3389 netif_set_gso_max_size(adapter->netdev, 32768);
3392 if (adapter->netdev->features & NETIF_F_FCOE_MTU)
3393 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
3396 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3398 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3401 /* Enable VLAN tag insert/strip */
3402 adapter->netdev->features |= NETIF_F_HW_VLAN_RX;
3404 hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
3406 /* reconfigure the hardware */
3407 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
3411 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3413 struct net_device *netdev = adapter->netdev;
3414 struct ixgbe_hw *hw = &adapter->hw;
3417 #ifdef CONFIG_IXGBE_DCB
3418 ixgbe_configure_dcb(adapter);
3421 ixgbe_set_rx_mode(netdev);
3422 ixgbe_restore_vlan(adapter);
3425 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
3426 ixgbe_configure_fcoe(adapter);
3428 #endif /* IXGBE_FCOE */
3429 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3430 for (i = 0; i < adapter->num_tx_queues; i++)
3431 adapter->tx_ring[i]->atr_sample_rate =
3432 adapter->atr_sample_rate;
3433 ixgbe_init_fdir_signature_82599(hw, adapter->fdir_pballoc);
3434 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3435 ixgbe_init_fdir_perfect_82599(hw, adapter->fdir_pballoc);
3437 ixgbe_configure_virtualization(adapter);
3439 ixgbe_configure_tx(adapter);
3440 ixgbe_configure_rx(adapter);
3443 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3445 switch (hw->phy.type) {
3446 case ixgbe_phy_sfp_avago:
3447 case ixgbe_phy_sfp_ftl:
3448 case ixgbe_phy_sfp_intel:
3449 case ixgbe_phy_sfp_unknown:
3450 case ixgbe_phy_sfp_passive_tyco:
3451 case ixgbe_phy_sfp_passive_unknown:
3452 case ixgbe_phy_sfp_active_unknown:
3453 case ixgbe_phy_sfp_ftl_active:
3461 * ixgbe_sfp_link_config - set up SFP+ link
3462 * @adapter: pointer to private adapter struct
3464 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3466 struct ixgbe_hw *hw = &adapter->hw;
3468 if (hw->phy.multispeed_fiber) {
3470 * In multispeed fiber setups, the device may not have
3471 * had a physical connection when the driver loaded.
3472 * If that's the case, the initial link configuration
3473 * couldn't get the MAC into 10G or 1G mode, so we'll
3474 * never have a link status change interrupt fire.
3475 * We need to try and force an autonegotiation
3476 * session, then bring up link.
3478 hw->mac.ops.setup_sfp(hw);
3479 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
3480 schedule_work(&adapter->multispeed_fiber_task);
3483 * Direct Attach Cu and non-multispeed fiber modules
3484 * still need to be configured properly prior to
3487 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_MOD_TASK))
3488 schedule_work(&adapter->sfp_config_module_task);
3493 * ixgbe_non_sfp_link_config - set up non-SFP+ link
3494 * @hw: pointer to private hardware struct
3496 * Returns 0 on success, negative on failure
3498 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3501 bool negotiation, link_up = false;
3502 u32 ret = IXGBE_ERR_LINK_SETUP;
3504 if (hw->mac.ops.check_link)
3505 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3510 if (hw->mac.ops.get_link_capabilities)
3511 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
3516 if (hw->mac.ops.setup_link)
3517 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
3522 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
3524 struct ixgbe_hw *hw = &adapter->hw;
3527 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3528 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
3530 gpie |= IXGBE_GPIE_EIAME;
3532 * use EIAM to auto-mask when MSI-X interrupt is asserted
3533 * this saves a register write for every interrupt
3535 switch (hw->mac.type) {
3536 case ixgbe_mac_82598EB:
3537 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3540 case ixgbe_mac_82599EB:
3541 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
3542 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
3546 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
3547 * specifically only auto mask tx and rx interrupts */
3548 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3551 /* XXX: to interrupt immediately for EICS writes, enable this */
3552 /* gpie |= IXGBE_GPIE_EIMEN; */
3554 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3555 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
3556 gpie |= IXGBE_GPIE_VTMODE_64;
3559 /* Enable fan failure interrupt */
3560 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3561 gpie |= IXGBE_SDP1_GPIEN;
3563 if (hw->mac.type == ixgbe_mac_82599EB)
3564 gpie |= IXGBE_SDP1_GPIEN;
3565 gpie |= IXGBE_SDP2_GPIEN;
3567 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3570 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
3572 struct ixgbe_hw *hw = &adapter->hw;
3576 ixgbe_get_hw_control(adapter);
3577 ixgbe_setup_gpie(adapter);
3579 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3580 ixgbe_configure_msix(adapter);
3582 ixgbe_configure_msi_and_legacy(adapter);
3584 /* enable the optics */
3585 if (hw->phy.multispeed_fiber)
3586 hw->mac.ops.enable_tx_laser(hw);
3588 clear_bit(__IXGBE_DOWN, &adapter->state);
3589 ixgbe_napi_enable_all(adapter);
3591 if (ixgbe_is_sfp(hw)) {
3592 ixgbe_sfp_link_config(adapter);
3594 err = ixgbe_non_sfp_link_config(hw);
3596 e_err(probe, "link_config FAILED %d\n", err);
3599 /* clear any pending interrupts, may auto mask */
3600 IXGBE_READ_REG(hw, IXGBE_EICR);
3601 ixgbe_irq_enable(adapter, true, true);
3604 * If this adapter has a fan, check to see if we had a failure
3605 * before we enabled the interrupt.
3607 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3608 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
3609 if (esdp & IXGBE_ESDP_SDP1)
3610 e_crit(drv, "Fan has stopped, replace the adapter\n");
3614 * For hot-pluggable SFP+ devices, a new SFP+ module may have
3615 * arrived before interrupts were enabled but after probe. Such
3616 * devices wouldn't have their type identified yet. We need to
3617 * kick off the SFP+ module setup first, then try to bring up link.
3618 * If we're not hot-pluggable SFP+, we just need to configure link
3621 if (hw->phy.type == ixgbe_phy_unknown)
3622 schedule_work(&adapter->sfp_config_module_task);
3624 /* enable transmits */
3625 netif_tx_start_all_queues(adapter->netdev);
3627 /* bring the link up in the watchdog, this could race with our first
3628 * link up interrupt but shouldn't be a problem */
3629 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3630 adapter->link_check_timeout = jiffies;
3631 mod_timer(&adapter->watchdog_timer, jiffies);
3633 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
3634 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
3635 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
3636 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
3641 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
3643 WARN_ON(in_interrupt());
3644 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
3646 ixgbe_down(adapter);
3648 * If SR-IOV enabled then wait a bit before bringing the adapter
3649 * back up to give the VFs time to respond to the reset. The
3650 * two second wait is based upon the watchdog timer cycle in
3653 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3656 clear_bit(__IXGBE_RESETTING, &adapter->state);
3659 int ixgbe_up(struct ixgbe_adapter *adapter)
3661 /* hardware has been reset, we need to reload some things */
3662 ixgbe_configure(adapter);
3664 return ixgbe_up_complete(adapter);
3667 void ixgbe_reset(struct ixgbe_adapter *adapter)
3669 struct ixgbe_hw *hw = &adapter->hw;
3672 err = hw->mac.ops.init_hw(hw);
3675 case IXGBE_ERR_SFP_NOT_PRESENT:
3677 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
3678 e_dev_err("master disable timed out\n");
3680 case IXGBE_ERR_EEPROM_VERSION:
3681 /* We are running on a pre-production device, log a warning */
3682 e_dev_warn("This device is a pre-production adapter/LOM. "
3683 "Please be aware there may be issuesassociated with "
3684 "your hardware. If you are experiencing problems "
3685 "please contact your Intel or hardware "
3686 "representative who provided you with this "
3690 e_dev_err("Hardware Error: %d\n", err);
3693 /* reprogram the RAR[0] in case user changed it. */
3694 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
3699 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
3700 * @rx_ring: ring to free buffers from
3702 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
3704 struct device *dev = rx_ring->dev;
3708 /* ring already cleared, nothing to do */
3709 if (!rx_ring->rx_buffer_info)
3712 /* Free all the Rx ring sk_buffs */
3713 for (i = 0; i < rx_ring->count; i++) {
3714 struct ixgbe_rx_buffer *rx_buffer_info;
3716 rx_buffer_info = &rx_ring->rx_buffer_info[i];
3717 if (rx_buffer_info->dma) {
3718 dma_unmap_single(rx_ring->dev, rx_buffer_info->dma,
3719 rx_ring->rx_buf_len,
3721 rx_buffer_info->dma = 0;
3723 if (rx_buffer_info->skb) {
3724 struct sk_buff *skb = rx_buffer_info->skb;
3725 rx_buffer_info->skb = NULL;
3727 struct sk_buff *this = skb;
3728 if (IXGBE_RSC_CB(this)->delay_unmap) {
3729 dma_unmap_single(dev,
3730 IXGBE_RSC_CB(this)->dma,
3731 rx_ring->rx_buf_len,
3733 IXGBE_RSC_CB(this)->dma = 0;
3734 IXGBE_RSC_CB(skb)->delay_unmap = false;
3737 dev_kfree_skb(this);
3740 if (!rx_buffer_info->page)
3742 if (rx_buffer_info->page_dma) {
3743 dma_unmap_page(dev, rx_buffer_info->page_dma,
3744 PAGE_SIZE / 2, DMA_FROM_DEVICE);
3745 rx_buffer_info->page_dma = 0;
3747 put_page(rx_buffer_info->page);
3748 rx_buffer_info->page = NULL;
3749 rx_buffer_info->page_offset = 0;
3752 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
3753 memset(rx_ring->rx_buffer_info, 0, size);
3755 /* Zero out the descriptor ring */
3756 memset(rx_ring->desc, 0, rx_ring->size);
3758 rx_ring->next_to_clean = 0;
3759 rx_ring->next_to_use = 0;
3763 * ixgbe_clean_tx_ring - Free Tx Buffers
3764 * @tx_ring: ring to be cleaned
3766 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
3768 struct ixgbe_tx_buffer *tx_buffer_info;
3772 /* ring already cleared, nothing to do */
3773 if (!tx_ring->tx_buffer_info)
3776 /* Free all the Tx ring sk_buffs */
3777 for (i = 0; i < tx_ring->count; i++) {
3778 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3779 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
3782 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
3783 memset(tx_ring->tx_buffer_info, 0, size);
3785 /* Zero out the descriptor ring */
3786 memset(tx_ring->desc, 0, tx_ring->size);
3788 tx_ring->next_to_use = 0;
3789 tx_ring->next_to_clean = 0;
3793 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
3794 * @adapter: board private structure
3796 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
3800 for (i = 0; i < adapter->num_rx_queues; i++)
3801 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
3805 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
3806 * @adapter: board private structure
3808 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
3812 for (i = 0; i < adapter->num_tx_queues; i++)
3813 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
3816 void ixgbe_down(struct ixgbe_adapter *adapter)
3818 struct net_device *netdev = adapter->netdev;
3819 struct ixgbe_hw *hw = &adapter->hw;
3823 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3825 /* signal that we are down to the interrupt handler */
3826 set_bit(__IXGBE_DOWN, &adapter->state);
3828 /* disable receive for all VFs and wait one second */
3829 if (adapter->num_vfs) {
3830 /* ping all the active vfs to let them know we are going down */
3831 ixgbe_ping_all_vfs(adapter);
3833 /* Disable all VFTE/VFRE TX/RX */
3834 ixgbe_disable_tx_rx(adapter);
3836 /* Mark all the VFs as inactive */
3837 for (i = 0 ; i < adapter->num_vfs; i++)
3838 adapter->vfinfo[i].clear_to_send = 0;
3841 /* disable receives */
3842 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3843 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3845 IXGBE_WRITE_FLUSH(hw);
3848 netif_tx_stop_all_queues(netdev);
3850 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
3851 del_timer_sync(&adapter->sfp_timer);
3852 del_timer_sync(&adapter->watchdog_timer);
3853 cancel_work_sync(&adapter->watchdog_task);
3855 netif_carrier_off(netdev);
3856 netif_tx_disable(netdev);
3858 ixgbe_irq_disable(adapter);
3860 ixgbe_napi_disable_all(adapter);
3862 /* Cleanup the affinity_hint CPU mask memory and callback */
3863 for (i = 0; i < num_q_vectors; i++) {
3864 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
3865 /* clear the affinity_mask in the IRQ descriptor */
3866 irq_set_affinity_hint(adapter->msix_entries[i]. vector, NULL);
3867 /* release the CPU mask memory */
3868 free_cpumask_var(q_vector->affinity_mask);
3871 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3872 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3873 cancel_work_sync(&adapter->fdir_reinit_task);
3875 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
3876 cancel_work_sync(&adapter->check_overtemp_task);
3878 /* disable transmits in the hardware now that interrupts are off */
3879 for (i = 0; i < adapter->num_tx_queues; i++) {
3880 j = adapter->tx_ring[i]->reg_idx;
3881 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3882 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j),
3883 (txdctl & ~IXGBE_TXDCTL_ENABLE));
3885 /* Disable the Tx DMA engine on 82599 */
3886 if (hw->mac.type == ixgbe_mac_82599EB)
3887 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
3888 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
3889 ~IXGBE_DMATXCTL_TE));
3891 /* power down the optics */
3892 if (hw->phy.multispeed_fiber)
3893 hw->mac.ops.disable_tx_laser(hw);
3895 /* clear n-tuple filters that are cached */
3896 ethtool_ntuple_flush(netdev);
3898 if (!pci_channel_offline(adapter->pdev))
3899 ixgbe_reset(adapter);
3900 ixgbe_clean_all_tx_rings(adapter);
3901 ixgbe_clean_all_rx_rings(adapter);
3903 #ifdef CONFIG_IXGBE_DCA
3904 /* since we reset the hardware DCA settings were cleared */
3905 ixgbe_setup_dca(adapter);
3910 * ixgbe_poll - NAPI Rx polling callback
3911 * @napi: structure for representing this polling device
3912 * @budget: how many packets driver is allowed to clean
3914 * This function is used for legacy and MSI, NAPI mode
3916 static int ixgbe_poll(struct napi_struct *napi, int budget)
3918 struct ixgbe_q_vector *q_vector =
3919 container_of(napi, struct ixgbe_q_vector, napi);
3920 struct ixgbe_adapter *adapter = q_vector->adapter;
3921 int tx_clean_complete, work_done = 0;
3923 #ifdef CONFIG_IXGBE_DCA
3924 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
3925 ixgbe_update_dca(q_vector);
3928 tx_clean_complete = ixgbe_clean_tx_irq(q_vector, adapter->tx_ring[0]);
3929 ixgbe_clean_rx_irq(q_vector, adapter->rx_ring[0], &work_done, budget);
3931 if (!tx_clean_complete)
3934 /* If budget not fully consumed, exit the polling mode */
3935 if (work_done < budget) {
3936 napi_complete(napi);
3937 if (adapter->rx_itr_setting & 1)
3938 ixgbe_set_itr(adapter);
3939 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3940 ixgbe_irq_enable_queues(adapter, IXGBE_EIMS_RTX_QUEUE);
3946 * ixgbe_tx_timeout - Respond to a Tx Hang
3947 * @netdev: network interface device structure
3949 static void ixgbe_tx_timeout(struct net_device *netdev)
3951 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3953 /* Do the reset outside of interrupt context */
3954 schedule_work(&adapter->reset_task);
3957 static void ixgbe_reset_task(struct work_struct *work)
3959 struct ixgbe_adapter *adapter;
3960 adapter = container_of(work, struct ixgbe_adapter, reset_task);
3962 /* If we're already down or resetting, just bail */
3963 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
3964 test_bit(__IXGBE_RESETTING, &adapter->state))
3967 adapter->tx_timeout_count++;
3969 ixgbe_dump(adapter);
3970 netdev_err(adapter->netdev, "Reset adapter\n");
3971 ixgbe_reinit_locked(adapter);
3974 #ifdef CONFIG_IXGBE_DCB
3975 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
3978 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_DCB];
3980 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
3984 adapter->num_rx_queues = f->indices;
3985 adapter->num_tx_queues = f->indices;
3993 * ixgbe_set_rss_queues: Allocate queues for RSS
3994 * @adapter: board private structure to initialize
3996 * This is our "base" multiqueue mode. RSS (Receive Side Scaling) will try
3997 * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
4000 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
4003 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
4005 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4007 adapter->num_rx_queues = f->indices;
4008 adapter->num_tx_queues = f->indices;
4018 * ixgbe_set_fdir_queues: Allocate queues for Flow Director
4019 * @adapter: board private structure to initialize
4021 * Flow Director is an advanced Rx filter, attempting to get Rx flows back
4022 * to the original CPU that initiated the Tx session. This runs in addition
4023 * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
4024 * Rx load across CPUs using RSS.
4027 static inline bool ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
4030 struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
4032 f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
4035 /* Flow Director must have RSS enabled */
4036 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
4037 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
4038 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)))) {
4039 adapter->num_tx_queues = f_fdir->indices;
4040 adapter->num_rx_queues = f_fdir->indices;
4043 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4044 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4051 * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
4052 * @adapter: board private structure to initialize
4054 * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
4055 * The ring feature mask is not used as a mask for FCoE, as it can take any 8
4056 * rx queues out of the max number of rx queues, instead, it is used as the
4057 * index of the first rx queue used by FCoE.
4060 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
4063 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4065 f->indices = min((int)num_online_cpus(), f->indices);
4066 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
4067 adapter->num_rx_queues = 1;
4068 adapter->num_tx_queues = 1;
4069 #ifdef CONFIG_IXGBE_DCB
4070 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4071 e_info(probe, "FCoE enabled with DCB\n");
4072 ixgbe_set_dcb_queues(adapter);
4075 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4076 e_info(probe, "FCoE enabled with RSS\n");
4077 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4078 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
4079 ixgbe_set_fdir_queues(adapter);
4081 ixgbe_set_rss_queues(adapter);
4083 /* adding FCoE rx rings to the end */
4084 f->mask = adapter->num_rx_queues;
4085 adapter->num_rx_queues += f->indices;
4086 adapter->num_tx_queues += f->indices;
4094 #endif /* IXGBE_FCOE */
4096 * ixgbe_set_sriov_queues: Allocate queues for IOV use
4097 * @adapter: board private structure to initialize
4099 * IOV doesn't actually use anything, so just NAK the
4100 * request for now and let the other queue routines
4101 * figure out what to do.
4103 static inline bool ixgbe_set_sriov_queues(struct ixgbe_adapter *adapter)
4109 * ixgbe_set_num_queues: Allocate queues for device, feature dependant
4110 * @adapter: board private structure to initialize
4112 * This is the top level queue allocation routine. The order here is very
4113 * important, starting with the "most" number of features turned on at once,
4114 * and ending with the smallest set of features. This way large combinations
4115 * can be allocated if they're turned on, and smaller combinations are the
4116 * fallthrough conditions.
4119 static int ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
4121 /* Start with base case */
4122 adapter->num_rx_queues = 1;
4123 adapter->num_tx_queues = 1;
4124 adapter->num_rx_pools = adapter->num_rx_queues;
4125 adapter->num_rx_queues_per_pool = 1;
4127 if (ixgbe_set_sriov_queues(adapter))
4131 if (ixgbe_set_fcoe_queues(adapter))
4134 #endif /* IXGBE_FCOE */
4135 #ifdef CONFIG_IXGBE_DCB
4136 if (ixgbe_set_dcb_queues(adapter))
4140 if (ixgbe_set_fdir_queues(adapter))
4143 if (ixgbe_set_rss_queues(adapter))
4146 /* fallback to base case */
4147 adapter->num_rx_queues = 1;
4148 adapter->num_tx_queues = 1;
4151 /* Notify the stack of the (possibly) reduced queue counts. */
4152 netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
4153 return netif_set_real_num_rx_queues(adapter->netdev,
4154 adapter->num_rx_queues);
4157 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
4160 int err, vector_threshold;
4162 /* We'll want at least 3 (vector_threshold):
4165 * 3) Other (Link Status Change, etc.)
4166 * 4) TCP Timer (optional)
4168 vector_threshold = MIN_MSIX_COUNT;
4170 /* The more we get, the more we will assign to Tx/Rx Cleanup
4171 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
4172 * Right now, we simply care about how many we'll get; we'll
4173 * set them up later while requesting irq's.
4175 while (vectors >= vector_threshold) {
4176 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
4178 if (!err) /* Success in acquiring all requested vectors. */
4181 vectors = 0; /* Nasty failure, quit now */
4182 else /* err == number of vectors we should try again with */
4186 if (vectors < vector_threshold) {
4187 /* Can't allocate enough MSI-X interrupts? Oh well.
4188 * This just means we'll go with either a single MSI
4189 * vector or fall back to legacy interrupts.
4191 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4192 "Unable to allocate MSI-X interrupts\n");
4193 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4194 kfree(adapter->msix_entries);
4195 adapter->msix_entries = NULL;
4197 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
4199 * Adjust for only the vectors we'll use, which is minimum
4200 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
4201 * vectors we were allocated.
4203 adapter->num_msix_vectors = min(vectors,
4204 adapter->max_msix_q_vectors + NON_Q_VECTORS);
4209 * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
4210 * @adapter: board private structure to initialize
4212 * Cache the descriptor ring offsets for RSS to the assigned rings.
4215 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
4220 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4221 for (i = 0; i < adapter->num_rx_queues; i++)
4222 adapter->rx_ring[i]->reg_idx = i;
4223 for (i = 0; i < adapter->num_tx_queues; i++)
4224 adapter->tx_ring[i]->reg_idx = i;
4233 #ifdef CONFIG_IXGBE_DCB
4235 * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
4236 * @adapter: board private structure to initialize
4238 * Cache the descriptor ring offsets for DCB to the assigned rings.
4241 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
4245 int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
4247 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4248 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
4249 /* the number of queues is assumed to be symmetric */
4250 for (i = 0; i < dcb_i; i++) {
4251 adapter->rx_ring[i]->reg_idx = i << 3;
4252 adapter->tx_ring[i]->reg_idx = i << 2;
4255 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
4258 * Tx TC0 starts at: descriptor queue 0
4259 * Tx TC1 starts at: descriptor queue 32
4260 * Tx TC2 starts at: descriptor queue 64
4261 * Tx TC3 starts at: descriptor queue 80
4262 * Tx TC4 starts at: descriptor queue 96
4263 * Tx TC5 starts at: descriptor queue 104
4264 * Tx TC6 starts at: descriptor queue 112
4265 * Tx TC7 starts at: descriptor queue 120
4267 * Rx TC0-TC7 are offset by 16 queues each
4269 for (i = 0; i < 3; i++) {
4270 adapter->tx_ring[i]->reg_idx = i << 5;
4271 adapter->rx_ring[i]->reg_idx = i << 4;
4273 for ( ; i < 5; i++) {
4274 adapter->tx_ring[i]->reg_idx =
4276 adapter->rx_ring[i]->reg_idx = i << 4;
4278 for ( ; i < dcb_i; i++) {
4279 adapter->tx_ring[i]->reg_idx =
4281 adapter->rx_ring[i]->reg_idx = i << 4;
4285 } else if (dcb_i == 4) {
4287 * Tx TC0 starts at: descriptor queue 0
4288 * Tx TC1 starts at: descriptor queue 64
4289 * Tx TC2 starts at: descriptor queue 96
4290 * Tx TC3 starts at: descriptor queue 112
4292 * Rx TC0-TC3 are offset by 32 queues each
4294 adapter->tx_ring[0]->reg_idx = 0;
4295 adapter->tx_ring[1]->reg_idx = 64;
4296 adapter->tx_ring[2]->reg_idx = 96;
4297 adapter->tx_ring[3]->reg_idx = 112;
4298 for (i = 0 ; i < dcb_i; i++)
4299 adapter->rx_ring[i]->reg_idx = i << 5;
4317 * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
4318 * @adapter: board private structure to initialize
4320 * Cache the descriptor ring offsets for Flow Director to the assigned rings.
4323 static inline bool ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
4328 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
4329 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4330 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))) {
4331 for (i = 0; i < adapter->num_rx_queues; i++)
4332 adapter->rx_ring[i]->reg_idx = i;
4333 for (i = 0; i < adapter->num_tx_queues; i++)
4334 adapter->tx_ring[i]->reg_idx = i;
4343 * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
4344 * @adapter: board private structure to initialize
4346 * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
4349 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
4351 int i, fcoe_rx_i = 0, fcoe_tx_i = 0;
4353 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4355 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
4356 #ifdef CONFIG_IXGBE_DCB
4357 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4358 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
4360 ixgbe_cache_ring_dcb(adapter);
4361 /* find out queues in TC for FCoE */
4362 fcoe_rx_i = adapter->rx_ring[fcoe->tc]->reg_idx + 1;
4363 fcoe_tx_i = adapter->tx_ring[fcoe->tc]->reg_idx + 1;
4365 * In 82599, the number of Tx queues for each traffic
4366 * class for both 8-TC and 4-TC modes are:
4367 * TCs : TC0 TC1 TC2 TC3 TC4 TC5 TC6 TC7
4368 * 8 TCs: 32 32 16 16 8 8 8 8
4369 * 4 TCs: 64 64 32 32
4370 * We have max 8 queues for FCoE, where 8 the is
4371 * FCoE redirection table size. If TC for FCoE is
4372 * less than or equal to TC3, we have enough queues
4373 * to add max of 8 queues for FCoE, so we start FCoE
4374 * tx descriptor from the next one, i.e., reg_idx + 1.
4375 * If TC for FCoE is above TC3, implying 8 TC mode,
4376 * and we need 8 for FCoE, we have to take all queues
4377 * in that traffic class for FCoE.
4379 if ((f->indices == IXGBE_FCRETA_SIZE) && (fcoe->tc > 3))
4382 #endif /* CONFIG_IXGBE_DCB */
4383 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4384 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4385 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
4386 ixgbe_cache_ring_fdir(adapter);
4388 ixgbe_cache_ring_rss(adapter);
4390 fcoe_rx_i = f->mask;
4391 fcoe_tx_i = f->mask;
4393 for (i = 0; i < f->indices; i++, fcoe_rx_i++, fcoe_tx_i++) {
4394 adapter->rx_ring[f->mask + i]->reg_idx = fcoe_rx_i;
4395 adapter->tx_ring[f->mask + i]->reg_idx = fcoe_tx_i;
4402 #endif /* IXGBE_FCOE */
4404 * ixgbe_cache_ring_sriov - Descriptor ring to register mapping for sriov
4405 * @adapter: board private structure to initialize
4407 * SR-IOV doesn't use any descriptor rings but changes the default if
4408 * no other mapping is used.
4411 static inline bool ixgbe_cache_ring_sriov(struct ixgbe_adapter *adapter)
4413 adapter->rx_ring[0]->reg_idx = adapter->num_vfs * 2;
4414 adapter->tx_ring[0]->reg_idx = adapter->num_vfs * 2;
4415 if (adapter->num_vfs)
4422 * ixgbe_cache_ring_register - Descriptor ring to register mapping
4423 * @adapter: board private structure to initialize
4425 * Once we know the feature-set enabled for the device, we'll cache
4426 * the register offset the descriptor ring is assigned to.
4428 * Note, the order the various feature calls is important. It must start with
4429 * the "most" features enabled at the same time, then trickle down to the
4430 * least amount of features turned on at once.
4432 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
4434 /* start with default case */
4435 adapter->rx_ring[0]->reg_idx = 0;
4436 adapter->tx_ring[0]->reg_idx = 0;
4438 if (ixgbe_cache_ring_sriov(adapter))
4442 if (ixgbe_cache_ring_fcoe(adapter))
4445 #endif /* IXGBE_FCOE */
4446 #ifdef CONFIG_IXGBE_DCB
4447 if (ixgbe_cache_ring_dcb(adapter))
4451 if (ixgbe_cache_ring_fdir(adapter))
4454 if (ixgbe_cache_ring_rss(adapter))
4459 * ixgbe_alloc_queues - Allocate memory for all rings
4460 * @adapter: board private structure to initialize
4462 * We allocate one ring per queue at run-time since we don't know the
4463 * number of queues at compile-time. The polling_netdev array is
4464 * intended for Multiqueue, but should work fine with a single queue.
4466 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
4470 int orig_node = adapter->node;
4472 for (i = 0; i < adapter->num_tx_queues; i++) {
4473 struct ixgbe_ring *ring = adapter->tx_ring[i];
4474 if (orig_node == -1) {
4475 int cur_node = next_online_node(adapter->node);
4476 if (cur_node == MAX_NUMNODES)
4477 cur_node = first_online_node;
4478 adapter->node = cur_node;
4480 ring = kzalloc_node(sizeof(struct ixgbe_ring), GFP_KERNEL,
4483 ring = kzalloc(sizeof(struct ixgbe_ring), GFP_KERNEL);
4485 goto err_tx_ring_allocation;
4486 ring->count = adapter->tx_ring_count;
4487 ring->queue_index = i;
4488 ring->dev = &adapter->pdev->dev;
4489 ring->netdev = adapter->netdev;
4490 ring->numa_node = adapter->node;
4492 adapter->tx_ring[i] = ring;
4495 /* Restore the adapter's original node */
4496 adapter->node = orig_node;
4498 rx_count = adapter->rx_ring_count;
4499 for (i = 0; i < adapter->num_rx_queues; i++) {
4500 struct ixgbe_ring *ring = adapter->rx_ring[i];
4501 if (orig_node == -1) {
4502 int cur_node = next_online_node(adapter->node);
4503 if (cur_node == MAX_NUMNODES)
4504 cur_node = first_online_node;
4505 adapter->node = cur_node;
4507 ring = kzalloc_node(sizeof(struct ixgbe_ring), GFP_KERNEL,
4510 ring = kzalloc(sizeof(struct ixgbe_ring), GFP_KERNEL);
4512 goto err_rx_ring_allocation;
4513 ring->count = rx_count;
4514 ring->queue_index = i;
4515 ring->dev = &adapter->pdev->dev;
4516 ring->netdev = adapter->netdev;
4517 ring->numa_node = adapter->node;
4519 adapter->rx_ring[i] = ring;
4522 /* Restore the adapter's original node */
4523 adapter->node = orig_node;
4525 ixgbe_cache_ring_register(adapter);
4529 err_rx_ring_allocation:
4530 for (i = 0; i < adapter->num_tx_queues; i++)
4531 kfree(adapter->tx_ring[i]);
4532 err_tx_ring_allocation:
4537 * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
4538 * @adapter: board private structure to initialize
4540 * Attempt to configure the interrupts using the best available
4541 * capabilities of the hardware and the kernel.
4543 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
4545 struct ixgbe_hw *hw = &adapter->hw;
4547 int vector, v_budget;
4550 * It's easy to be greedy for MSI-X vectors, but it really
4551 * doesn't do us much good if we have a lot more vectors
4552 * than CPU's. So let's be conservative and only ask for
4553 * (roughly) the same number of vectors as there are CPU's.
4555 v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
4556 (int)num_online_cpus()) + NON_Q_VECTORS;
4559 * At the same time, hardware can only support a maximum of
4560 * hw.mac->max_msix_vectors vectors. With features
4561 * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
4562 * descriptor queues supported by our device. Thus, we cap it off in
4563 * those rare cases where the cpu count also exceeds our vector limit.
4565 v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
4567 /* A failure in MSI-X entry allocation isn't fatal, but it does
4568 * mean we disable MSI-X capabilities of the adapter. */
4569 adapter->msix_entries = kcalloc(v_budget,
4570 sizeof(struct msix_entry), GFP_KERNEL);
4571 if (adapter->msix_entries) {
4572 for (vector = 0; vector < v_budget; vector++)
4573 adapter->msix_entries[vector].entry = vector;
4575 ixgbe_acquire_msix_vectors(adapter, v_budget);
4577 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4581 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
4582 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
4583 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4584 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4585 adapter->atr_sample_rate = 0;
4586 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4587 ixgbe_disable_sriov(adapter);
4589 err = ixgbe_set_num_queues(adapter);
4593 err = pci_enable_msi(adapter->pdev);
4595 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
4597 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4598 "Unable to allocate MSI interrupt, "
4599 "falling back to legacy. Error: %d\n", err);
4609 * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
4610 * @adapter: board private structure to initialize
4612 * We allocate one q_vector per queue interrupt. If allocation fails we
4615 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
4617 int q_idx, num_q_vectors;
4618 struct ixgbe_q_vector *q_vector;
4620 int (*poll)(struct napi_struct *, int);
4622 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4623 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4624 napi_vectors = adapter->num_rx_queues;
4625 poll = &ixgbe_clean_rxtx_many;
4632 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4633 q_vector = kzalloc_node(sizeof(struct ixgbe_q_vector),
4634 GFP_KERNEL, adapter->node);
4636 q_vector = kzalloc(sizeof(struct ixgbe_q_vector),
4640 q_vector->adapter = adapter;
4641 if (q_vector->txr_count && !q_vector->rxr_count)
4642 q_vector->eitr = adapter->tx_eitr_param;
4644 q_vector->eitr = adapter->rx_eitr_param;
4645 q_vector->v_idx = q_idx;
4646 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
4647 adapter->q_vector[q_idx] = q_vector;
4655 q_vector = adapter->q_vector[q_idx];
4656 netif_napi_del(&q_vector->napi);
4658 adapter->q_vector[q_idx] = NULL;
4664 * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
4665 * @adapter: board private structure to initialize
4667 * This function frees the memory allocated to the q_vectors. In addition if
4668 * NAPI is enabled it will delete any references to the NAPI struct prior
4669 * to freeing the q_vector.
4671 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
4673 int q_idx, num_q_vectors;
4675 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4676 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4680 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4681 struct ixgbe_q_vector *q_vector = adapter->q_vector[q_idx];
4682 adapter->q_vector[q_idx] = NULL;
4683 netif_napi_del(&q_vector->napi);
4688 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
4690 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4691 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4692 pci_disable_msix(adapter->pdev);
4693 kfree(adapter->msix_entries);
4694 adapter->msix_entries = NULL;
4695 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
4696 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
4697 pci_disable_msi(adapter->pdev);
4702 * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
4703 * @adapter: board private structure to initialize
4705 * We determine which interrupt scheme to use based on...
4706 * - Kernel support (MSI, MSI-X)
4707 * - which can be user-defined (via MODULE_PARAM)
4708 * - Hardware queue count (num_*_queues)
4709 * - defined by miscellaneous hardware support/features (RSS, etc.)
4711 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
4715 /* Number of supported queues */
4716 err = ixgbe_set_num_queues(adapter);
4720 err = ixgbe_set_interrupt_capability(adapter);
4722 e_dev_err("Unable to setup interrupt capabilities\n");
4723 goto err_set_interrupt;
4726 err = ixgbe_alloc_q_vectors(adapter);
4728 e_dev_err("Unable to allocate memory for queue vectors\n");
4729 goto err_alloc_q_vectors;
4732 err = ixgbe_alloc_queues(adapter);
4734 e_dev_err("Unable to allocate memory for queues\n");
4735 goto err_alloc_queues;
4738 e_dev_info("Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
4739 (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
4740 adapter->num_rx_queues, adapter->num_tx_queues);
4742 set_bit(__IXGBE_DOWN, &adapter->state);
4747 ixgbe_free_q_vectors(adapter);
4748 err_alloc_q_vectors:
4749 ixgbe_reset_interrupt_capability(adapter);
4754 static void ring_free_rcu(struct rcu_head *head)
4756 kfree(container_of(head, struct ixgbe_ring, rcu));
4760 * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
4761 * @adapter: board private structure to clear interrupt scheme on
4763 * We go through and clear interrupt specific resources and reset the structure
4764 * to pre-load conditions
4766 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
4770 for (i = 0; i < adapter->num_tx_queues; i++) {
4771 kfree(adapter->tx_ring[i]);
4772 adapter->tx_ring[i] = NULL;
4774 for (i = 0; i < adapter->num_rx_queues; i++) {
4775 struct ixgbe_ring *ring = adapter->rx_ring[i];
4777 /* ixgbe_get_stats64() might access this ring, we must wait
4778 * a grace period before freeing it.
4780 call_rcu(&ring->rcu, ring_free_rcu);
4781 adapter->rx_ring[i] = NULL;
4784 ixgbe_free_q_vectors(adapter);
4785 ixgbe_reset_interrupt_capability(adapter);
4789 * ixgbe_sfp_timer - worker thread to find a missing module
4790 * @data: pointer to our adapter struct
4792 static void ixgbe_sfp_timer(unsigned long data)
4794 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
4797 * Do the sfp_timer outside of interrupt context due to the
4798 * delays that sfp+ detection requires
4800 schedule_work(&adapter->sfp_task);
4804 * ixgbe_sfp_task - worker thread to find a missing module
4805 * @work: pointer to work_struct containing our data
4807 static void ixgbe_sfp_task(struct work_struct *work)
4809 struct ixgbe_adapter *adapter = container_of(work,
4810 struct ixgbe_adapter,
4812 struct ixgbe_hw *hw = &adapter->hw;
4814 if ((hw->phy.type == ixgbe_phy_nl) &&
4815 (hw->phy.sfp_type == ixgbe_sfp_type_not_present)) {
4816 s32 ret = hw->phy.ops.identify_sfp(hw);
4817 if (ret == IXGBE_ERR_SFP_NOT_PRESENT)
4819 ret = hw->phy.ops.reset(hw);
4820 if (ret == IXGBE_ERR_SFP_NOT_SUPPORTED) {
4821 e_dev_err("failed to initialize because an unsupported "
4822 "SFP+ module type was detected.\n");
4823 e_dev_err("Reload the driver after installing a "
4824 "supported module.\n");
4825 unregister_netdev(adapter->netdev);
4827 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
4829 /* don't need this routine any more */
4830 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
4834 if (test_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state))
4835 mod_timer(&adapter->sfp_timer,
4836 round_jiffies(jiffies + (2 * HZ)));
4840 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4841 * @adapter: board private structure to initialize
4843 * ixgbe_sw_init initializes the Adapter private data structure.
4844 * Fields are initialized based on PCI device information and
4845 * OS network device settings (MTU size).
4847 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
4849 struct ixgbe_hw *hw = &adapter->hw;
4850 struct pci_dev *pdev = adapter->pdev;
4851 struct net_device *dev = adapter->netdev;
4853 #ifdef CONFIG_IXGBE_DCB
4855 struct tc_configuration *tc;
4857 int max_frame = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
4859 /* PCI config space info */
4861 hw->vendor_id = pdev->vendor;
4862 hw->device_id = pdev->device;
4863 hw->revision_id = pdev->revision;
4864 hw->subsystem_vendor_id = pdev->subsystem_vendor;
4865 hw->subsystem_device_id = pdev->subsystem_device;
4867 /* Set capability flags */
4868 rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
4869 adapter->ring_feature[RING_F_RSS].indices = rss;
4870 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
4871 adapter->ring_feature[RING_F_DCB].indices = IXGBE_MAX_DCB_INDICES;
4872 if (hw->mac.type == ixgbe_mac_82598EB) {
4873 if (hw->device_id == IXGBE_DEV_ID_82598AT)
4874 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4875 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
4876 } else if (hw->mac.type == ixgbe_mac_82599EB) {
4877 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
4878 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4879 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4880 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
4881 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4882 if (dev->features & NETIF_F_NTUPLE) {
4883 /* Flow Director perfect filter enabled */
4884 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4885 adapter->atr_sample_rate = 0;
4886 spin_lock_init(&adapter->fdir_perfect_lock);
4888 /* Flow Director hash filters enabled */
4889 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
4890 adapter->atr_sample_rate = 20;
4892 adapter->ring_feature[RING_F_FDIR].indices =
4893 IXGBE_MAX_FDIR_INDICES;
4894 adapter->fdir_pballoc = 0;
4896 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4897 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4898 adapter->ring_feature[RING_F_FCOE].indices = 0;
4899 #ifdef CONFIG_IXGBE_DCB
4900 /* Default traffic class to use for FCoE */
4901 adapter->fcoe.tc = IXGBE_FCOE_DEFTC;
4902 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
4904 #endif /* IXGBE_FCOE */
4907 #ifdef CONFIG_IXGBE_DCB
4908 /* Configure DCB traffic classes */
4909 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
4910 tc = &adapter->dcb_cfg.tc_config[j];
4911 tc->path[DCB_TX_CONFIG].bwg_id = 0;
4912 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
4913 tc->path[DCB_RX_CONFIG].bwg_id = 0;
4914 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
4915 tc->dcb_pfc = pfc_disabled;
4917 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
4918 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
4919 adapter->dcb_cfg.rx_pba_cfg = pba_equal;
4920 adapter->dcb_cfg.pfc_mode_enable = false;
4921 adapter->dcb_cfg.round_robin_enable = false;
4922 adapter->dcb_set_bitmap = 0x00;
4923 ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
4924 adapter->ring_feature[RING_F_DCB].indices);
4928 /* default flow control settings */
4929 hw->fc.requested_mode = ixgbe_fc_full;
4930 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
4932 adapter->last_lfc_mode = hw->fc.current_mode;
4934 hw->fc.high_water = FC_HIGH_WATER(max_frame);
4935 hw->fc.low_water = FC_LOW_WATER(max_frame);
4936 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
4937 hw->fc.send_xon = true;
4938 hw->fc.disable_fc_autoneg = false;
4940 /* enable itr by default in dynamic mode */
4941 adapter->rx_itr_setting = 1;
4942 adapter->rx_eitr_param = 20000;
4943 adapter->tx_itr_setting = 1;
4944 adapter->tx_eitr_param = 10000;
4946 /* set defaults for eitr in MegaBytes */
4947 adapter->eitr_low = 10;
4948 adapter->eitr_high = 20;
4950 /* set default ring sizes */
4951 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
4952 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
4954 /* initialize eeprom parameters */
4955 if (ixgbe_init_eeprom_params_generic(hw)) {
4956 e_dev_err("EEPROM initialization failed\n");
4960 /* enable rx csum by default */
4961 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
4963 /* get assigned NUMA node */
4964 adapter->node = dev_to_node(&pdev->dev);
4966 set_bit(__IXGBE_DOWN, &adapter->state);
4972 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
4973 * @tx_ring: tx descriptor ring (for a specific queue) to setup
4975 * Return 0 on success, negative on failure
4977 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
4979 struct device *dev = tx_ring->dev;
4982 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4983 tx_ring->tx_buffer_info = vmalloc_node(size, tx_ring->numa_node);
4984 if (!tx_ring->tx_buffer_info)
4985 tx_ring->tx_buffer_info = vmalloc(size);
4986 if (!tx_ring->tx_buffer_info)
4988 memset(tx_ring->tx_buffer_info, 0, size);
4990 /* round up to nearest 4K */
4991 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
4992 tx_ring->size = ALIGN(tx_ring->size, 4096);
4994 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
4995 &tx_ring->dma, GFP_KERNEL);
4999 tx_ring->next_to_use = 0;
5000 tx_ring->next_to_clean = 0;
5001 tx_ring->work_limit = tx_ring->count;
5005 vfree(tx_ring->tx_buffer_info);
5006 tx_ring->tx_buffer_info = NULL;
5007 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
5012 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
5013 * @adapter: board private structure
5015 * If this function returns with an error, then it's possible one or
5016 * more of the rings is populated (while the rest are not). It is the
5017 * callers duty to clean those orphaned rings.
5019 * Return 0 on success, negative on failure
5021 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
5025 for (i = 0; i < adapter->num_tx_queues; i++) {
5026 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
5029 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
5037 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
5038 * @rx_ring: rx descriptor ring (for a specific queue) to setup
5040 * Returns 0 on success, negative on failure
5042 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
5044 struct device *dev = rx_ring->dev;
5047 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
5048 rx_ring->rx_buffer_info = vmalloc_node(size, rx_ring->numa_node);
5049 if (!rx_ring->rx_buffer_info)
5050 rx_ring->rx_buffer_info = vmalloc(size);
5051 if (!rx_ring->rx_buffer_info)
5053 memset(rx_ring->rx_buffer_info, 0, size);
5055 /* Round up to nearest 4K */
5056 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
5057 rx_ring->size = ALIGN(rx_ring->size, 4096);
5059 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
5060 &rx_ring->dma, GFP_KERNEL);
5065 rx_ring->next_to_clean = 0;
5066 rx_ring->next_to_use = 0;
5070 vfree(rx_ring->rx_buffer_info);
5071 rx_ring->rx_buffer_info = NULL;
5072 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
5077 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
5078 * @adapter: board private structure
5080 * If this function returns with an error, then it's possible one or
5081 * more of the rings is populated (while the rest are not). It is the
5082 * callers duty to clean those orphaned rings.
5084 * Return 0 on success, negative on failure
5086 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
5090 for (i = 0; i < adapter->num_rx_queues; i++) {
5091 err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
5094 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
5102 * ixgbe_free_tx_resources - Free Tx Resources per Queue
5103 * @tx_ring: Tx descriptor ring for a specific queue
5105 * Free all transmit software resources
5107 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
5109 ixgbe_clean_tx_ring(tx_ring);
5111 vfree(tx_ring->tx_buffer_info);
5112 tx_ring->tx_buffer_info = NULL;
5114 /* if not set, then don't free */
5118 dma_free_coherent(tx_ring->dev, tx_ring->size,
5119 tx_ring->desc, tx_ring->dma);
5121 tx_ring->desc = NULL;
5125 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5126 * @adapter: board private structure
5128 * Free all transmit software resources
5130 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
5134 for (i = 0; i < adapter->num_tx_queues; i++)
5135 if (adapter->tx_ring[i]->desc)
5136 ixgbe_free_tx_resources(adapter->tx_ring[i]);
5140 * ixgbe_free_rx_resources - Free Rx Resources
5141 * @rx_ring: ring to clean the resources from
5143 * Free all receive software resources
5145 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
5147 ixgbe_clean_rx_ring(rx_ring);
5149 vfree(rx_ring->rx_buffer_info);
5150 rx_ring->rx_buffer_info = NULL;
5152 /* if not set, then don't free */
5156 dma_free_coherent(rx_ring->dev, rx_ring->size,
5157 rx_ring->desc, rx_ring->dma);
5159 rx_ring->desc = NULL;
5163 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5164 * @adapter: board private structure
5166 * Free all receive software resources
5168 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5172 for (i = 0; i < adapter->num_rx_queues; i++)
5173 if (adapter->rx_ring[i]->desc)
5174 ixgbe_free_rx_resources(adapter->rx_ring[i]);
5178 * ixgbe_change_mtu - Change the Maximum Transfer Unit
5179 * @netdev: network interface device structure
5180 * @new_mtu: new value for maximum frame size
5182 * Returns 0 on success, negative on failure
5184 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5186 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5187 struct ixgbe_hw *hw = &adapter->hw;
5188 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5190 /* MTU < 68 is an error and causes problems on some kernels */
5191 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5194 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5195 /* must set new MTU before calling down or up */
5196 netdev->mtu = new_mtu;
5198 hw->fc.high_water = FC_HIGH_WATER(max_frame);
5199 hw->fc.low_water = FC_LOW_WATER(max_frame);
5201 if (netif_running(netdev))
5202 ixgbe_reinit_locked(adapter);
5208 * ixgbe_open - Called when a network interface is made active
5209 * @netdev: network interface device structure
5211 * Returns 0 on success, negative value on failure
5213 * The open entry point is called when a network interface is made
5214 * active by the system (IFF_UP). At this point all resources needed
5215 * for transmit and receive operations are allocated, the interrupt
5216 * handler is registered with the OS, the watchdog timer is started,
5217 * and the stack is notified that the interface is ready.
5219 static int ixgbe_open(struct net_device *netdev)
5221 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5224 /* disallow open during test */
5225 if (test_bit(__IXGBE_TESTING, &adapter->state))
5228 netif_carrier_off(netdev);
5230 /* allocate transmit descriptors */
5231 err = ixgbe_setup_all_tx_resources(adapter);
5235 /* allocate receive descriptors */
5236 err = ixgbe_setup_all_rx_resources(adapter);
5240 ixgbe_configure(adapter);
5242 err = ixgbe_request_irq(adapter);
5246 err = ixgbe_up_complete(adapter);
5250 netif_tx_start_all_queues(netdev);
5255 ixgbe_release_hw_control(adapter);
5256 ixgbe_free_irq(adapter);
5259 ixgbe_free_all_rx_resources(adapter);
5261 ixgbe_free_all_tx_resources(adapter);
5262 ixgbe_reset(adapter);
5268 * ixgbe_close - Disables a network interface
5269 * @netdev: network interface device structure
5271 * Returns 0, this is not allowed to fail
5273 * The close entry point is called when an interface is de-activated
5274 * by the OS. The hardware is still under the drivers control, but
5275 * needs to be disabled. A global MAC reset is issued to stop the
5276 * hardware, and all transmit and receive resources are freed.
5278 static int ixgbe_close(struct net_device *netdev)
5280 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5282 ixgbe_down(adapter);
5283 ixgbe_free_irq(adapter);
5285 ixgbe_free_all_tx_resources(adapter);
5286 ixgbe_free_all_rx_resources(adapter);
5288 ixgbe_release_hw_control(adapter);
5294 static int ixgbe_resume(struct pci_dev *pdev)
5296 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5297 struct net_device *netdev = adapter->netdev;
5300 pci_set_power_state(pdev, PCI_D0);
5301 pci_restore_state(pdev);
5303 * pci_restore_state clears dev->state_saved so call
5304 * pci_save_state to restore it.
5306 pci_save_state(pdev);
5308 err = pci_enable_device_mem(pdev);
5310 e_dev_err("Cannot enable PCI device from suspend\n");
5313 pci_set_master(pdev);
5315 pci_wake_from_d3(pdev, false);
5317 err = ixgbe_init_interrupt_scheme(adapter);
5319 e_dev_err("Cannot initialize interrupts for device\n");
5323 ixgbe_reset(adapter);
5325 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5327 if (netif_running(netdev)) {
5328 err = ixgbe_open(netdev);
5333 netif_device_attach(netdev);
5337 #endif /* CONFIG_PM */
5339 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5341 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5342 struct net_device *netdev = adapter->netdev;
5343 struct ixgbe_hw *hw = &adapter->hw;
5345 u32 wufc = adapter->wol;
5350 netif_device_detach(netdev);
5352 if (netif_running(netdev)) {
5353 ixgbe_down(adapter);
5354 ixgbe_free_irq(adapter);
5355 ixgbe_free_all_tx_resources(adapter);
5356 ixgbe_free_all_rx_resources(adapter);
5359 ixgbe_clear_interrupt_scheme(adapter);
5362 retval = pci_save_state(pdev);
5368 ixgbe_set_rx_mode(netdev);
5370 /* turn on all-multi mode if wake on multicast is enabled */
5371 if (wufc & IXGBE_WUFC_MC) {
5372 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5373 fctrl |= IXGBE_FCTRL_MPE;
5374 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5377 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5378 ctrl |= IXGBE_CTRL_GIO_DIS;
5379 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5381 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5383 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5384 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5387 if (wufc && hw->mac.type == ixgbe_mac_82599EB)
5388 pci_wake_from_d3(pdev, true);
5390 pci_wake_from_d3(pdev, false);
5392 *enable_wake = !!wufc;
5394 ixgbe_release_hw_control(adapter);
5396 pci_disable_device(pdev);
5402 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5407 retval = __ixgbe_shutdown(pdev, &wake);
5412 pci_prepare_to_sleep(pdev);
5414 pci_wake_from_d3(pdev, false);
5415 pci_set_power_state(pdev, PCI_D3hot);
5420 #endif /* CONFIG_PM */
5422 static void ixgbe_shutdown(struct pci_dev *pdev)
5426 __ixgbe_shutdown(pdev, &wake);
5428 if (system_state == SYSTEM_POWER_OFF) {
5429 pci_wake_from_d3(pdev, wake);
5430 pci_set_power_state(pdev, PCI_D3hot);
5435 * ixgbe_update_stats - Update the board statistics counters.
5436 * @adapter: board private structure
5438 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5440 struct net_device *netdev = adapter->netdev;
5441 struct ixgbe_hw *hw = &adapter->hw;
5442 struct ixgbe_hw_stats *hwstats = &adapter->stats;
5444 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5445 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
5446 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
5447 u64 bytes = 0, packets = 0;
5449 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5450 test_bit(__IXGBE_RESETTING, &adapter->state))
5453 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5456 for (i = 0; i < 16; i++)
5457 adapter->hw_rx_no_dma_resources +=
5458 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5459 for (i = 0; i < adapter->num_rx_queues; i++) {
5460 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
5461 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
5463 adapter->rsc_total_count = rsc_count;
5464 adapter->rsc_total_flush = rsc_flush;
5467 for (i = 0; i < adapter->num_rx_queues; i++) {
5468 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
5469 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
5470 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
5471 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
5472 bytes += rx_ring->stats.bytes;
5473 packets += rx_ring->stats.packets;
5475 adapter->non_eop_descs = non_eop_descs;
5476 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5477 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5478 netdev->stats.rx_bytes = bytes;
5479 netdev->stats.rx_packets = packets;
5483 /* gather some stats to the adapter struct that are per queue */
5484 for (i = 0; i < adapter->num_tx_queues; i++) {
5485 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5486 restart_queue += tx_ring->tx_stats.restart_queue;
5487 tx_busy += tx_ring->tx_stats.tx_busy;
5488 bytes += tx_ring->stats.bytes;
5489 packets += tx_ring->stats.packets;
5491 adapter->restart_queue = restart_queue;
5492 adapter->tx_busy = tx_busy;
5493 netdev->stats.tx_bytes = bytes;
5494 netdev->stats.tx_packets = packets;
5496 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5497 for (i = 0; i < 8; i++) {
5498 /* for packet buffers not used, the register should read 0 */
5499 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5501 hwstats->mpc[i] += mpc;
5502 total_mpc += hwstats->mpc[i];
5503 if (hw->mac.type == ixgbe_mac_82598EB)
5504 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5505 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5506 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5507 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5508 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5509 if (hw->mac.type == ixgbe_mac_82599EB) {
5510 hwstats->pxonrxc[i] +=
5511 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5512 hwstats->pxoffrxc[i] +=
5513 IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
5514 hwstats->qprdc[i] += IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5516 hwstats->pxonrxc[i] +=
5517 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5518 hwstats->pxoffrxc[i] +=
5519 IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
5521 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5522 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5524 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5525 /* work around hardware counting issue */
5526 hwstats->gprc -= missed_rx;
5528 /* 82598 hardware only has a 32 bit counter in the high register */
5529 if (hw->mac.type == ixgbe_mac_82599EB) {
5531 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5532 tmp = IXGBE_READ_REG(hw, IXGBE_GORCH) & 0xF;
5533 /* 4 high bits of GORC */
5534 hwstats->gorc += (tmp << 32);
5535 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5536 tmp = IXGBE_READ_REG(hw, IXGBE_GOTCH) & 0xF;
5537 /* 4 high bits of GOTC */
5538 hwstats->gotc += (tmp << 32);
5539 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5540 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5541 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5542 hwstats->lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
5543 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5544 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5546 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5547 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5548 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5549 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5550 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5551 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5552 #endif /* IXGBE_FCOE */
5554 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5555 hwstats->lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
5556 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5557 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5558 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5560 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5561 hwstats->bprc += bprc;
5562 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5563 if (hw->mac.type == ixgbe_mac_82598EB)
5564 hwstats->mprc -= bprc;
5565 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5566 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5567 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5568 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5569 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5570 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5571 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5572 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5573 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5574 hwstats->lxontxc += lxon;
5575 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5576 hwstats->lxofftxc += lxoff;
5577 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5578 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5579 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5581 * 82598 errata - tx of flow control packets is included in tx counters
5583 xon_off_tot = lxon + lxoff;
5584 hwstats->gptc -= xon_off_tot;
5585 hwstats->mptc -= xon_off_tot;
5586 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5587 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5588 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5589 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5590 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5591 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5592 hwstats->ptc64 -= xon_off_tot;
5593 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5594 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5595 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5596 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5597 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5598 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5600 /* Fill out the OS statistics structure */
5601 netdev->stats.multicast = hwstats->mprc;
5604 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
5605 netdev->stats.rx_dropped = 0;
5606 netdev->stats.rx_length_errors = hwstats->rlec;
5607 netdev->stats.rx_crc_errors = hwstats->crcerrs;
5608 netdev->stats.rx_missed_errors = total_mpc;
5612 * ixgbe_watchdog - Timer Call-back
5613 * @data: pointer to adapter cast into an unsigned long
5615 static void ixgbe_watchdog(unsigned long data)
5617 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5618 struct ixgbe_hw *hw = &adapter->hw;
5623 * Do the watchdog outside of interrupt context due to the lovely
5624 * delays that some of the newer hardware requires
5627 if (test_bit(__IXGBE_DOWN, &adapter->state))
5628 goto watchdog_short_circuit;
5630 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5632 * for legacy and MSI interrupts don't set any bits
5633 * that are enabled for EIAM, because this operation
5634 * would set *both* EIMS and EICS for any bit in EIAM
5636 IXGBE_WRITE_REG(hw, IXGBE_EICS,
5637 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5638 goto watchdog_reschedule;
5641 /* get one bit for every active tx/rx interrupt vector */
5642 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
5643 struct ixgbe_q_vector *qv = adapter->q_vector[i];
5644 if (qv->rxr_count || qv->txr_count)
5645 eics |= ((u64)1 << i);
5648 /* Cause software interrupt to ensure rx rings are cleaned */
5649 ixgbe_irq_rearm_queues(adapter, eics);
5651 watchdog_reschedule:
5652 /* Reset the timer */
5653 mod_timer(&adapter->watchdog_timer, round_jiffies(jiffies + 2 * HZ));
5655 watchdog_short_circuit:
5656 schedule_work(&adapter->watchdog_task);
5660 * ixgbe_multispeed_fiber_task - worker thread to configure multispeed fiber
5661 * @work: pointer to work_struct containing our data
5663 static void ixgbe_multispeed_fiber_task(struct work_struct *work)
5665 struct ixgbe_adapter *adapter = container_of(work,
5666 struct ixgbe_adapter,
5667 multispeed_fiber_task);
5668 struct ixgbe_hw *hw = &adapter->hw;
5672 adapter->flags |= IXGBE_FLAG_IN_SFP_LINK_TASK;
5673 autoneg = hw->phy.autoneg_advertised;
5674 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
5675 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
5676 hw->mac.autotry_restart = false;
5677 if (hw->mac.ops.setup_link)
5678 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
5679 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5680 adapter->flags &= ~IXGBE_FLAG_IN_SFP_LINK_TASK;
5684 * ixgbe_sfp_config_module_task - worker thread to configure a new SFP+ module
5685 * @work: pointer to work_struct containing our data
5687 static void ixgbe_sfp_config_module_task(struct work_struct *work)
5689 struct ixgbe_adapter *adapter = container_of(work,
5690 struct ixgbe_adapter,
5691 sfp_config_module_task);
5692 struct ixgbe_hw *hw = &adapter->hw;
5695 adapter->flags |= IXGBE_FLAG_IN_SFP_MOD_TASK;
5697 /* Time for electrical oscillations to settle down */
5699 err = hw->phy.ops.identify_sfp(hw);
5701 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
5702 e_dev_err("failed to initialize because an unsupported SFP+ "
5703 "module type was detected.\n");
5704 e_dev_err("Reload the driver after installing a supported "
5706 unregister_netdev(adapter->netdev);
5709 hw->mac.ops.setup_sfp(hw);
5711 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
5712 /* This will also work for DA Twinax connections */
5713 schedule_work(&adapter->multispeed_fiber_task);
5714 adapter->flags &= ~IXGBE_FLAG_IN_SFP_MOD_TASK;
5718 * ixgbe_fdir_reinit_task - worker thread to reinit FDIR filter table
5719 * @work: pointer to work_struct containing our data
5721 static void ixgbe_fdir_reinit_task(struct work_struct *work)
5723 struct ixgbe_adapter *adapter = container_of(work,
5724 struct ixgbe_adapter,
5726 struct ixgbe_hw *hw = &adapter->hw;
5729 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5730 for (i = 0; i < adapter->num_tx_queues; i++)
5731 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
5732 &(adapter->tx_ring[i]->state));
5734 e_err(probe, "failed to finish FDIR re-initialization, "
5735 "ignored adding FDIR ATR filters\n");
5737 /* Done FDIR Re-initialization, enable transmits */
5738 netif_tx_start_all_queues(adapter->netdev);
5741 static DEFINE_MUTEX(ixgbe_watchdog_lock);
5744 * ixgbe_watchdog_task - worker thread to bring link up
5745 * @work: pointer to work_struct containing our data
5747 static void ixgbe_watchdog_task(struct work_struct *work)
5749 struct ixgbe_adapter *adapter = container_of(work,
5750 struct ixgbe_adapter,
5752 struct net_device *netdev = adapter->netdev;
5753 struct ixgbe_hw *hw = &adapter->hw;
5757 struct ixgbe_ring *tx_ring;
5758 int some_tx_pending = 0;
5760 mutex_lock(&ixgbe_watchdog_lock);
5762 link_up = adapter->link_up;
5763 link_speed = adapter->link_speed;
5765 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
5766 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5769 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5770 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
5771 hw->mac.ops.fc_enable(hw, i);
5773 hw->mac.ops.fc_enable(hw, 0);
5776 hw->mac.ops.fc_enable(hw, 0);
5781 time_after(jiffies, (adapter->link_check_timeout +
5782 IXGBE_TRY_LINK_TIMEOUT))) {
5783 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5784 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5786 adapter->link_up = link_up;
5787 adapter->link_speed = link_speed;
5791 if (!netif_carrier_ok(netdev)) {
5792 bool flow_rx, flow_tx;
5794 if (hw->mac.type == ixgbe_mac_82599EB) {
5795 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5796 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5797 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5798 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5800 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5801 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5802 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5803 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5806 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
5807 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5809 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5810 "1 Gbps" : "unknown speed")),
5811 ((flow_rx && flow_tx) ? "RX/TX" :
5813 (flow_tx ? "TX" : "None"))));
5815 netif_carrier_on(netdev);
5817 /* Force detection of hung controller */
5818 for (i = 0; i < adapter->num_tx_queues; i++) {
5819 tx_ring = adapter->tx_ring[i];
5820 set_check_for_tx_hang(tx_ring);
5824 adapter->link_up = false;
5825 adapter->link_speed = 0;
5826 if (netif_carrier_ok(netdev)) {
5827 e_info(drv, "NIC Link is Down\n");
5828 netif_carrier_off(netdev);
5832 if (!netif_carrier_ok(netdev)) {
5833 for (i = 0; i < adapter->num_tx_queues; i++) {
5834 tx_ring = adapter->tx_ring[i];
5835 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5836 some_tx_pending = 1;
5841 if (some_tx_pending) {
5842 /* We've lost link, so the controller stops DMA,
5843 * but we've got queued Tx work that's never going
5844 * to get done, so reset controller to flush Tx.
5845 * (Do the reset outside of interrupt context).
5847 schedule_work(&adapter->reset_task);
5851 ixgbe_update_stats(adapter);
5852 mutex_unlock(&ixgbe_watchdog_lock);
5855 static int ixgbe_tso(struct ixgbe_adapter *adapter,
5856 struct ixgbe_ring *tx_ring, struct sk_buff *skb,
5857 u32 tx_flags, u8 *hdr_len, __be16 protocol)
5859 struct ixgbe_adv_tx_context_desc *context_desc;
5862 struct ixgbe_tx_buffer *tx_buffer_info;
5863 u32 vlan_macip_lens = 0, type_tucmd_mlhl;
5864 u32 mss_l4len_idx, l4len;
5866 if (skb_is_gso(skb)) {
5867 if (skb_header_cloned(skb)) {
5868 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5872 l4len = tcp_hdrlen(skb);
5875 if (protocol == htons(ETH_P_IP)) {
5876 struct iphdr *iph = ip_hdr(skb);
5879 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
5883 } else if (skb_is_gso_v6(skb)) {
5884 ipv6_hdr(skb)->payload_len = 0;
5885 tcp_hdr(skb)->check =
5886 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5887 &ipv6_hdr(skb)->daddr,
5891 i = tx_ring->next_to_use;
5893 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5894 context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
5896 /* VLAN MACLEN IPLEN */
5897 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5899 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
5900 vlan_macip_lens |= ((skb_network_offset(skb)) <<
5901 IXGBE_ADVTXD_MACLEN_SHIFT);
5902 *hdr_len += skb_network_offset(skb);
5904 (skb_transport_header(skb) - skb_network_header(skb));
5906 (skb_transport_header(skb) - skb_network_header(skb));
5907 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
5908 context_desc->seqnum_seed = 0;
5910 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
5911 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
5912 IXGBE_ADVTXD_DTYP_CTXT);
5914 if (protocol == htons(ETH_P_IP))
5915 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
5916 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5917 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
5921 (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
5922 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
5923 /* use index 1 for TSO */
5924 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
5925 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
5927 tx_buffer_info->time_stamp = jiffies;
5928 tx_buffer_info->next_to_watch = i;
5931 if (i == tx_ring->count)
5933 tx_ring->next_to_use = i;
5940 static u32 ixgbe_psum(struct ixgbe_adapter *adapter, struct sk_buff *skb,
5946 case cpu_to_be16(ETH_P_IP):
5947 rtn |= IXGBE_ADVTXD_TUCMD_IPV4;
5948 switch (ip_hdr(skb)->protocol) {
5950 rtn |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5953 rtn |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5957 case cpu_to_be16(ETH_P_IPV6):
5958 /* XXX what about other V6 headers?? */
5959 switch (ipv6_hdr(skb)->nexthdr) {
5961 rtn |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5964 rtn |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5969 if (unlikely(net_ratelimit()))
5970 e_warn(probe, "partial checksum but proto=%x!\n",
5978 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
5979 struct ixgbe_ring *tx_ring,
5980 struct sk_buff *skb, u32 tx_flags,
5983 struct ixgbe_adv_tx_context_desc *context_desc;
5985 struct ixgbe_tx_buffer *tx_buffer_info;
5986 u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
5988 if (skb->ip_summed == CHECKSUM_PARTIAL ||
5989 (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
5990 i = tx_ring->next_to_use;
5991 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5992 context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
5994 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5996 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
5997 vlan_macip_lens |= (skb_network_offset(skb) <<
5998 IXGBE_ADVTXD_MACLEN_SHIFT);
5999 if (skb->ip_summed == CHECKSUM_PARTIAL)
6000 vlan_macip_lens |= (skb_transport_header(skb) -
6001 skb_network_header(skb));
6003 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
6004 context_desc->seqnum_seed = 0;
6006 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
6007 IXGBE_ADVTXD_DTYP_CTXT);
6009 if (skb->ip_summed == CHECKSUM_PARTIAL)
6010 type_tucmd_mlhl |= ixgbe_psum(adapter, skb, protocol);
6012 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
6013 /* use index zero for tx checksum offload */
6014 context_desc->mss_l4len_idx = 0;
6016 tx_buffer_info->time_stamp = jiffies;
6017 tx_buffer_info->next_to_watch = i;
6020 if (i == tx_ring->count)
6022 tx_ring->next_to_use = i;
6030 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
6031 struct ixgbe_ring *tx_ring,
6032 struct sk_buff *skb, u32 tx_flags,
6033 unsigned int first, const u8 hdr_len)
6035 struct device *dev = tx_ring->dev;
6036 struct ixgbe_tx_buffer *tx_buffer_info;
6038 unsigned int total = skb->len;
6039 unsigned int offset = 0, size, count = 0, i;
6040 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
6042 unsigned int bytecount = skb->len;
6045 i = tx_ring->next_to_use;
6047 if (tx_flags & IXGBE_TX_FLAGS_FCOE)
6048 /* excluding fcoe_crc_eof for FCoE */
6049 total -= sizeof(struct fcoe_crc_eof);
6051 len = min(skb_headlen(skb), total);
6053 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6054 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
6056 tx_buffer_info->length = size;
6057 tx_buffer_info->mapped_as_page = false;
6058 tx_buffer_info->dma = dma_map_single(dev,
6060 size, DMA_TO_DEVICE);
6061 if (dma_mapping_error(dev, tx_buffer_info->dma))
6063 tx_buffer_info->time_stamp = jiffies;
6064 tx_buffer_info->next_to_watch = i;
6073 if (i == tx_ring->count)
6078 for (f = 0; f < nr_frags; f++) {
6079 struct skb_frag_struct *frag;
6081 frag = &skb_shinfo(skb)->frags[f];
6082 len = min((unsigned int)frag->size, total);
6083 offset = frag->page_offset;
6087 if (i == tx_ring->count)
6090 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6091 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
6093 tx_buffer_info->length = size;
6094 tx_buffer_info->dma = dma_map_page(dev,
6098 tx_buffer_info->mapped_as_page = true;
6099 if (dma_mapping_error(dev, tx_buffer_info->dma))
6101 tx_buffer_info->time_stamp = jiffies;
6102 tx_buffer_info->next_to_watch = i;
6113 if (tx_flags & IXGBE_TX_FLAGS_TSO)
6114 gso_segs = skb_shinfo(skb)->gso_segs;
6116 /* adjust for FCoE Sequence Offload */
6117 else if (tx_flags & IXGBE_TX_FLAGS_FSO)
6118 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
6119 skb_shinfo(skb)->gso_size);
6120 #endif /* IXGBE_FCOE */
6121 bytecount += (gso_segs - 1) * hdr_len;
6123 /* multiply data chunks by size of headers */
6124 tx_ring->tx_buffer_info[i].bytecount = bytecount;
6125 tx_ring->tx_buffer_info[i].gso_segs = gso_segs;
6126 tx_ring->tx_buffer_info[i].skb = skb;
6127 tx_ring->tx_buffer_info[first].next_to_watch = i;
6132 e_dev_err("TX DMA map failed\n");
6134 /* clear timestamp and dma mappings for failed tx_buffer_info map */
6135 tx_buffer_info->dma = 0;
6136 tx_buffer_info->time_stamp = 0;
6137 tx_buffer_info->next_to_watch = 0;
6141 /* clear timestamp and dma mappings for remaining portion of packet */
6144 i += tx_ring->count;
6146 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6147 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
6153 static void ixgbe_tx_queue(struct ixgbe_ring *tx_ring,
6154 int tx_flags, int count, u32 paylen, u8 hdr_len)
6156 union ixgbe_adv_tx_desc *tx_desc = NULL;
6157 struct ixgbe_tx_buffer *tx_buffer_info;
6158 u32 olinfo_status = 0, cmd_type_len = 0;
6160 u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
6162 cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
6164 cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
6166 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
6167 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
6169 if (tx_flags & IXGBE_TX_FLAGS_TSO) {
6170 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
6172 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
6173 IXGBE_ADVTXD_POPTS_SHIFT;
6175 /* use index 1 context for tso */
6176 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6177 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
6178 olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
6179 IXGBE_ADVTXD_POPTS_SHIFT;
6181 } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
6182 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
6183 IXGBE_ADVTXD_POPTS_SHIFT;
6185 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6186 olinfo_status |= IXGBE_ADVTXD_CC;
6187 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6188 if (tx_flags & IXGBE_TX_FLAGS_FSO)
6189 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
6192 olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
6194 i = tx_ring->next_to_use;
6196 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6197 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
6198 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
6199 tx_desc->read.cmd_type_len =
6200 cpu_to_le32(cmd_type_len | tx_buffer_info->length);
6201 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
6203 if (i == tx_ring->count)
6207 tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
6210 * Force memory writes to complete before letting h/w
6211 * know there are new descriptors to fetch. (Only
6212 * applicable for weak-ordered memory model archs,
6217 tx_ring->next_to_use = i;
6218 writel(i, tx_ring->tail);
6221 static void ixgbe_atr(struct ixgbe_adapter *adapter, struct sk_buff *skb,
6222 int queue, u32 tx_flags, __be16 protocol)
6224 struct ixgbe_atr_input atr_input;
6226 struct iphdr *iph = ip_hdr(skb);
6227 struct ethhdr *eth = (struct ethhdr *)skb->data;
6228 u16 vlan_id, src_port, dst_port, flex_bytes;
6229 u32 src_ipv4_addr, dst_ipv4_addr;
6232 /* Right now, we support IPv4 only */
6233 if (protocol != htons(ETH_P_IP))
6235 /* check if we're UDP or TCP */
6236 if (iph->protocol == IPPROTO_TCP) {
6238 src_port = th->source;
6239 dst_port = th->dest;
6240 l4type |= IXGBE_ATR_L4TYPE_TCP;
6241 /* l4type IPv4 type is 0, no need to assign */
6243 /* Unsupported L4 header, just bail here */
6247 memset(&atr_input, 0, sizeof(struct ixgbe_atr_input));
6249 vlan_id = (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK) >>
6250 IXGBE_TX_FLAGS_VLAN_SHIFT;
6251 src_ipv4_addr = iph->saddr;
6252 dst_ipv4_addr = iph->daddr;
6253 flex_bytes = eth->h_proto;
6255 ixgbe_atr_set_vlan_id_82599(&atr_input, vlan_id);
6256 ixgbe_atr_set_src_port_82599(&atr_input, dst_port);
6257 ixgbe_atr_set_dst_port_82599(&atr_input, src_port);
6258 ixgbe_atr_set_flex_byte_82599(&atr_input, flex_bytes);
6259 ixgbe_atr_set_l4type_82599(&atr_input, l4type);
6260 /* src and dst are inverted, think how the receiver sees them */
6261 ixgbe_atr_set_src_ipv4_82599(&atr_input, dst_ipv4_addr);
6262 ixgbe_atr_set_dst_ipv4_82599(&atr_input, src_ipv4_addr);
6264 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
6265 ixgbe_fdir_add_signature_filter_82599(&adapter->hw, &atr_input, queue);
6268 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, int size)
6270 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
6271 /* Herbert's original patch had:
6272 * smp_mb__after_netif_stop_queue();
6273 * but since that doesn't exist yet, just open code it. */
6276 /* We need to check again in a case another CPU has just
6277 * made room available. */
6278 if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
6281 /* A reprieve! - use start_queue because it doesn't call schedule */
6282 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
6283 ++tx_ring->tx_stats.restart_queue;
6287 static int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, int size)
6289 if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
6291 return __ixgbe_maybe_stop_tx(tx_ring, size);
6294 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6296 struct ixgbe_adapter *adapter = netdev_priv(dev);
6297 int txq = smp_processor_id();
6301 protocol = vlan_get_protocol(skb);
6303 if ((protocol == htons(ETH_P_FCOE)) ||
6304 (protocol == htons(ETH_P_FIP))) {
6305 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
6306 txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
6307 txq += adapter->ring_feature[RING_F_FCOE].mask;
6309 #ifdef CONFIG_IXGBE_DCB
6310 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6311 txq = adapter->fcoe.up;
6318 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6319 while (unlikely(txq >= dev->real_num_tx_queues))
6320 txq -= dev->real_num_tx_queues;
6324 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6325 if (skb->priority == TC_PRIO_CONTROL)
6326 txq = adapter->ring_feature[RING_F_DCB].indices-1;
6328 txq = (skb->vlan_tci & IXGBE_TX_FLAGS_VLAN_PRIO_MASK)
6333 return skb_tx_hash(dev, skb);
6336 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
6337 struct ixgbe_adapter *adapter,
6338 struct ixgbe_ring *tx_ring)
6340 struct net_device *netdev = tx_ring->netdev;
6341 struct netdev_queue *txq;
6343 unsigned int tx_flags = 0;
6350 protocol = vlan_get_protocol(skb);
6352 if (vlan_tx_tag_present(skb)) {
6353 tx_flags |= vlan_tx_tag_get(skb);
6354 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6355 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6356 tx_flags |= ((skb->queue_mapping & 0x7) << 13);
6358 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6359 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6360 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED &&
6361 skb->priority != TC_PRIO_CONTROL) {
6362 tx_flags |= ((skb->queue_mapping & 0x7) << 13);
6363 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6364 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6368 /* for FCoE with DCB, we force the priority to what
6369 * was specified by the switch */
6370 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED &&
6371 (protocol == htons(ETH_P_FCOE) ||
6372 protocol == htons(ETH_P_FIP))) {
6373 #ifdef CONFIG_IXGBE_DCB
6374 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6375 tx_flags &= ~(IXGBE_TX_FLAGS_VLAN_PRIO_MASK
6376 << IXGBE_TX_FLAGS_VLAN_SHIFT);
6377 tx_flags |= ((adapter->fcoe.up << 13)
6378 << IXGBE_TX_FLAGS_VLAN_SHIFT);
6381 /* flag for FCoE offloads */
6382 if (protocol == htons(ETH_P_FCOE))
6383 tx_flags |= IXGBE_TX_FLAGS_FCOE;
6387 /* four things can cause us to need a context descriptor */
6388 if (skb_is_gso(skb) ||
6389 (skb->ip_summed == CHECKSUM_PARTIAL) ||
6390 (tx_flags & IXGBE_TX_FLAGS_VLAN) ||
6391 (tx_flags & IXGBE_TX_FLAGS_FCOE))
6394 count += TXD_USE_COUNT(skb_headlen(skb));
6395 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6396 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6398 if (ixgbe_maybe_stop_tx(tx_ring, count)) {
6399 tx_ring->tx_stats.tx_busy++;
6400 return NETDEV_TX_BUSY;
6403 first = tx_ring->next_to_use;
6404 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6406 /* setup tx offload for FCoE */
6407 tso = ixgbe_fso(adapter, tx_ring, skb, tx_flags, &hdr_len);
6409 dev_kfree_skb_any(skb);
6410 return NETDEV_TX_OK;
6413 tx_flags |= IXGBE_TX_FLAGS_FSO;
6414 #endif /* IXGBE_FCOE */
6416 if (protocol == htons(ETH_P_IP))
6417 tx_flags |= IXGBE_TX_FLAGS_IPV4;
6418 tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len,
6421 dev_kfree_skb_any(skb);
6422 return NETDEV_TX_OK;
6426 tx_flags |= IXGBE_TX_FLAGS_TSO;
6427 else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags,
6429 (skb->ip_summed == CHECKSUM_PARTIAL))
6430 tx_flags |= IXGBE_TX_FLAGS_CSUM;
6433 count = ixgbe_tx_map(adapter, tx_ring, skb, tx_flags, first, hdr_len);
6435 /* add the ATR filter if ATR is on */
6436 if (tx_ring->atr_sample_rate) {
6437 ++tx_ring->atr_count;
6438 if ((tx_ring->atr_count >= tx_ring->atr_sample_rate) &&
6439 test_bit(__IXGBE_TX_FDIR_INIT_DONE,
6441 ixgbe_atr(adapter, skb, tx_ring->queue_index,
6442 tx_flags, protocol);
6443 tx_ring->atr_count = 0;
6446 txq = netdev_get_tx_queue(netdev, tx_ring->queue_index);
6447 txq->tx_bytes += skb->len;
6449 ixgbe_tx_queue(tx_ring, tx_flags, count, skb->len, hdr_len);
6450 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
6453 dev_kfree_skb_any(skb);
6454 tx_ring->tx_buffer_info[first].time_stamp = 0;
6455 tx_ring->next_to_use = first;
6458 return NETDEV_TX_OK;
6461 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
6463 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6464 struct ixgbe_ring *tx_ring;
6466 tx_ring = adapter->tx_ring[skb->queue_mapping];
6467 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
6471 * ixgbe_set_mac - Change the Ethernet Address of the NIC
6472 * @netdev: network interface device structure
6473 * @p: pointer to an address structure
6475 * Returns 0 on success, negative on failure
6477 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6479 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6480 struct ixgbe_hw *hw = &adapter->hw;
6481 struct sockaddr *addr = p;
6483 if (!is_valid_ether_addr(addr->sa_data))
6484 return -EADDRNOTAVAIL;
6486 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6487 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6489 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
6496 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6498 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6499 struct ixgbe_hw *hw = &adapter->hw;
6503 if (prtad != hw->phy.mdio.prtad)
6505 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6511 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6512 u16 addr, u16 value)
6514 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6515 struct ixgbe_hw *hw = &adapter->hw;
6517 if (prtad != hw->phy.mdio.prtad)
6519 return hw->phy.ops.write_reg(hw, addr, devad, value);
6522 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6524 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6526 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6530 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6532 * @netdev: network interface device structure
6534 * Returns non-zero on failure
6536 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6539 struct ixgbe_adapter *adapter = netdev_priv(dev);
6540 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6542 if (is_valid_ether_addr(mac->san_addr)) {
6544 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6551 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6553 * @netdev: network interface device structure
6555 * Returns non-zero on failure
6557 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6560 struct ixgbe_adapter *adapter = netdev_priv(dev);
6561 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6563 if (is_valid_ether_addr(mac->san_addr)) {
6565 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6571 #ifdef CONFIG_NET_POLL_CONTROLLER
6573 * Polling 'interrupt' - used by things like netconsole to send skbs
6574 * without having to re-enable interrupts. It's not called while
6575 * the interrupt routine is executing.
6577 static void ixgbe_netpoll(struct net_device *netdev)
6579 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6582 /* if interface is down do nothing */
6583 if (test_bit(__IXGBE_DOWN, &adapter->state))
6586 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
6587 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
6588 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
6589 for (i = 0; i < num_q_vectors; i++) {
6590 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
6591 ixgbe_msix_clean_many(0, q_vector);
6594 ixgbe_intr(adapter->pdev->irq, netdev);
6596 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
6600 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
6601 struct rtnl_link_stats64 *stats)
6603 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6606 /* accurate rx/tx bytes/packets stats */
6607 dev_txq_stats_fold(netdev, stats);
6609 for (i = 0; i < adapter->num_rx_queues; i++) {
6610 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
6616 start = u64_stats_fetch_begin_bh(&ring->syncp);
6617 packets = ring->stats.packets;
6618 bytes = ring->stats.bytes;
6619 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6620 stats->rx_packets += packets;
6621 stats->rx_bytes += bytes;
6625 /* following stats updated by ixgbe_watchdog_task() */
6626 stats->multicast = netdev->stats.multicast;
6627 stats->rx_errors = netdev->stats.rx_errors;
6628 stats->rx_length_errors = netdev->stats.rx_length_errors;
6629 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
6630 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
6635 static const struct net_device_ops ixgbe_netdev_ops = {
6636 .ndo_open = ixgbe_open,
6637 .ndo_stop = ixgbe_close,
6638 .ndo_start_xmit = ixgbe_xmit_frame,
6639 .ndo_select_queue = ixgbe_select_queue,
6640 .ndo_set_rx_mode = ixgbe_set_rx_mode,
6641 .ndo_set_multicast_list = ixgbe_set_rx_mode,
6642 .ndo_validate_addr = eth_validate_addr,
6643 .ndo_set_mac_address = ixgbe_set_mac,
6644 .ndo_change_mtu = ixgbe_change_mtu,
6645 .ndo_tx_timeout = ixgbe_tx_timeout,
6646 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
6647 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
6648 .ndo_do_ioctl = ixgbe_ioctl,
6649 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
6650 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
6651 .ndo_set_vf_tx_rate = ixgbe_ndo_set_vf_bw,
6652 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
6653 .ndo_get_stats64 = ixgbe_get_stats64,
6654 #ifdef CONFIG_NET_POLL_CONTROLLER
6655 .ndo_poll_controller = ixgbe_netpoll,
6658 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
6659 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
6660 .ndo_fcoe_enable = ixgbe_fcoe_enable,
6661 .ndo_fcoe_disable = ixgbe_fcoe_disable,
6662 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
6663 #endif /* IXGBE_FCOE */
6666 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
6667 const struct ixgbe_info *ii)
6669 #ifdef CONFIG_PCI_IOV
6670 struct ixgbe_hw *hw = &adapter->hw;
6673 if (hw->mac.type != ixgbe_mac_82599EB || !max_vfs)
6676 /* The 82599 supports up to 64 VFs per physical function
6677 * but this implementation limits allocation to 63 so that
6678 * basic networking resources are still available to the
6681 adapter->num_vfs = (max_vfs > 63) ? 63 : max_vfs;
6682 adapter->flags |= IXGBE_FLAG_SRIOV_ENABLED;
6683 err = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
6685 e_err(probe, "Failed to enable PCI sriov: %d\n", err);
6688 /* If call to enable VFs succeeded then allocate memory
6689 * for per VF control structures.
6692 kcalloc(adapter->num_vfs,
6693 sizeof(struct vf_data_storage), GFP_KERNEL);
6694 if (adapter->vfinfo) {
6695 /* Now that we're sure SR-IOV is enabled
6696 * and memory allocated set up the mailbox parameters
6698 ixgbe_init_mbx_params_pf(hw);
6699 memcpy(&hw->mbx.ops, ii->mbx_ops,
6700 sizeof(hw->mbx.ops));
6702 /* Disable RSC when in SR-IOV mode */
6703 adapter->flags2 &= ~(IXGBE_FLAG2_RSC_CAPABLE |
6704 IXGBE_FLAG2_RSC_ENABLED);
6709 e_err(probe, "Unable to allocate memory for VF Data Storage - "
6710 "SRIOV disabled\n");
6711 pci_disable_sriov(adapter->pdev);
6714 adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
6715 adapter->num_vfs = 0;
6716 #endif /* CONFIG_PCI_IOV */
6720 * ixgbe_probe - Device Initialization Routine
6721 * @pdev: PCI device information struct
6722 * @ent: entry in ixgbe_pci_tbl
6724 * Returns 0 on success, negative on failure
6726 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
6727 * The OS initialization, configuring of the adapter private structure,
6728 * and a hardware reset occur.
6730 static int __devinit ixgbe_probe(struct pci_dev *pdev,
6731 const struct pci_device_id *ent)
6733 struct net_device *netdev;
6734 struct ixgbe_adapter *adapter = NULL;
6735 struct ixgbe_hw *hw;
6736 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
6737 static int cards_found;
6738 int i, err, pci_using_dac;
6739 unsigned int indices = num_possible_cpus();
6745 /* Catch broken hardware that put the wrong VF device ID in
6746 * the PCIe SR-IOV capability.
6748 if (pdev->is_virtfn) {
6749 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
6750 pci_name(pdev), pdev->vendor, pdev->device);
6754 err = pci_enable_device_mem(pdev);
6758 if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
6759 !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
6762 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
6764 err = dma_set_coherent_mask(&pdev->dev,
6768 "No usable DMA configuration, aborting\n");
6775 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
6776 IORESOURCE_MEM), ixgbe_driver_name);
6779 "pci_request_selected_regions failed 0x%x\n", err);
6783 pci_enable_pcie_error_reporting(pdev);
6785 pci_set_master(pdev);
6786 pci_save_state(pdev);
6788 if (ii->mac == ixgbe_mac_82598EB)
6789 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
6791 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
6793 indices = max_t(unsigned int, indices, IXGBE_MAX_DCB_INDICES);
6795 indices += min_t(unsigned int, num_possible_cpus(),
6796 IXGBE_MAX_FCOE_INDICES);
6798 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
6801 goto err_alloc_etherdev;
6804 SET_NETDEV_DEV(netdev, &pdev->dev);
6806 adapter = netdev_priv(netdev);
6807 pci_set_drvdata(pdev, adapter);
6809 adapter->netdev = netdev;
6810 adapter->pdev = pdev;
6813 adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
6815 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
6816 pci_resource_len(pdev, 0));
6822 for (i = 1; i <= 5; i++) {
6823 if (pci_resource_len(pdev, i) == 0)
6827 netdev->netdev_ops = &ixgbe_netdev_ops;
6828 ixgbe_set_ethtool_ops(netdev);
6829 netdev->watchdog_timeo = 5 * HZ;
6830 strcpy(netdev->name, pci_name(pdev));
6832 adapter->bd_number = cards_found;
6835 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
6836 hw->mac.type = ii->mac;
6839 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
6840 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
6841 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
6842 if (!(eec & (1 << 8)))
6843 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
6846 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
6847 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
6848 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
6849 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
6850 hw->phy.mdio.mmds = 0;
6851 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
6852 hw->phy.mdio.dev = netdev;
6853 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
6854 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
6856 /* set up this timer and work struct before calling get_invariants
6857 * which might start the timer
6859 init_timer(&adapter->sfp_timer);
6860 adapter->sfp_timer.function = ixgbe_sfp_timer;
6861 adapter->sfp_timer.data = (unsigned long) adapter;
6863 INIT_WORK(&adapter->sfp_task, ixgbe_sfp_task);
6865 /* multispeed fiber has its own tasklet, called from GPI SDP1 context */
6866 INIT_WORK(&adapter->multispeed_fiber_task, ixgbe_multispeed_fiber_task);
6868 /* a new SFP+ module arrival, called from GPI SDP2 context */
6869 INIT_WORK(&adapter->sfp_config_module_task,
6870 ixgbe_sfp_config_module_task);
6872 ii->get_invariants(hw);
6874 /* setup the private structure */
6875 err = ixgbe_sw_init(adapter);
6879 /* Make it possible the adapter to be woken up via WOL */
6880 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
6881 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6884 * If there is a fan on this device and it has failed log the
6887 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
6888 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
6889 if (esdp & IXGBE_ESDP_SDP1)
6890 e_crit(probe, "Fan has stopped, replace the adapter\n");
6893 /* reset_hw fills in the perm_addr as well */
6894 hw->phy.reset_if_overtemp = true;
6895 err = hw->mac.ops.reset_hw(hw);
6896 hw->phy.reset_if_overtemp = false;
6897 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
6898 hw->mac.type == ixgbe_mac_82598EB) {
6900 * Start a kernel thread to watch for a module to arrive.
6901 * Only do this for 82598, since 82599 will generate
6902 * interrupts on module arrival.
6904 set_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
6905 mod_timer(&adapter->sfp_timer,
6906 round_jiffies(jiffies + (2 * HZ)));
6908 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
6909 e_dev_err("failed to initialize because an unsupported SFP+ "
6910 "module type was detected.\n");
6911 e_dev_err("Reload the driver after installing a supported "
6915 e_dev_err("HW Init failed: %d\n", err);
6919 ixgbe_probe_vf(adapter, ii);
6921 netdev->features = NETIF_F_SG |
6923 NETIF_F_HW_VLAN_TX |
6924 NETIF_F_HW_VLAN_RX |
6925 NETIF_F_HW_VLAN_FILTER;
6927 netdev->features |= NETIF_F_IPV6_CSUM;
6928 netdev->features |= NETIF_F_TSO;
6929 netdev->features |= NETIF_F_TSO6;
6930 netdev->features |= NETIF_F_GRO;
6932 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
6933 netdev->features |= NETIF_F_SCTP_CSUM;
6935 netdev->vlan_features |= NETIF_F_TSO;
6936 netdev->vlan_features |= NETIF_F_TSO6;
6937 netdev->vlan_features |= NETIF_F_IP_CSUM;
6938 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
6939 netdev->vlan_features |= NETIF_F_SG;
6941 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6942 adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
6943 IXGBE_FLAG_DCB_ENABLED);
6944 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
6945 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
6947 #ifdef CONFIG_IXGBE_DCB
6948 netdev->dcbnl_ops = &dcbnl_ops;
6952 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
6953 if (hw->mac.ops.get_device_caps) {
6954 hw->mac.ops.get_device_caps(hw, &device_caps);
6955 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
6956 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6959 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
6960 netdev->vlan_features |= NETIF_F_FCOE_CRC;
6961 netdev->vlan_features |= NETIF_F_FSO;
6962 netdev->vlan_features |= NETIF_F_FCOE_MTU;
6964 #endif /* IXGBE_FCOE */
6965 if (pci_using_dac) {
6966 netdev->features |= NETIF_F_HIGHDMA;
6967 netdev->vlan_features |= NETIF_F_HIGHDMA;
6970 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
6971 netdev->features |= NETIF_F_LRO;
6973 /* make sure the EEPROM is good */
6974 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
6975 e_dev_err("The EEPROM Checksum Is Not Valid\n");
6980 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
6981 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
6983 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
6984 e_dev_err("invalid MAC address\n");
6989 /* power down the optics */
6990 if (hw->phy.multispeed_fiber)
6991 hw->mac.ops.disable_tx_laser(hw);
6993 init_timer(&adapter->watchdog_timer);
6994 adapter->watchdog_timer.function = ixgbe_watchdog;
6995 adapter->watchdog_timer.data = (unsigned long)adapter;
6997 INIT_WORK(&adapter->reset_task, ixgbe_reset_task);
6998 INIT_WORK(&adapter->watchdog_task, ixgbe_watchdog_task);
7000 err = ixgbe_init_interrupt_scheme(adapter);
7004 switch (pdev->device) {
7005 case IXGBE_DEV_ID_82599_KX4:
7006 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
7007 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
7013 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
7015 /* pick up the PCI bus settings for reporting later */
7016 hw->mac.ops.get_bus_info(hw);
7018 /* print bus type/speed/width info */
7019 e_dev_info("(PCI Express:%s:%s) %pM\n",
7020 (hw->bus.speed == ixgbe_bus_speed_5000 ? "5.0Gb/s" :
7021 hw->bus.speed == ixgbe_bus_speed_2500 ? "2.5Gb/s" :
7023 (hw->bus.width == ixgbe_bus_width_pcie_x8 ? "Width x8" :
7024 hw->bus.width == ixgbe_bus_width_pcie_x4 ? "Width x4" :
7025 hw->bus.width == ixgbe_bus_width_pcie_x1 ? "Width x1" :
7028 ixgbe_read_pba_num_generic(hw, &part_num);
7029 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
7030 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, "
7031 "PBA No: %06x-%03x\n",
7032 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
7033 (part_num >> 8), (part_num & 0xff));
7035 e_dev_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
7036 hw->mac.type, hw->phy.type,
7037 (part_num >> 8), (part_num & 0xff));
7039 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
7040 e_dev_warn("PCI-Express bandwidth available for this card is "
7041 "not sufficient for optimal performance.\n");
7042 e_dev_warn("For optimal performance a x8 PCI-Express slot "
7046 /* save off EEPROM version number */
7047 hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
7049 /* reset the hardware with the new settings */
7050 err = hw->mac.ops.start_hw(hw);
7052 if (err == IXGBE_ERR_EEPROM_VERSION) {
7053 /* We are running on a pre-production device, log a warning */
7054 e_dev_warn("This device is a pre-production adapter/LOM. "
7055 "Please be aware there may be issues associated "
7056 "with your hardware. If you are experiencing "
7057 "problems please contact your Intel or hardware "
7058 "representative who provided you with this "
7061 strcpy(netdev->name, "eth%d");
7062 err = register_netdev(netdev);
7066 /* carrier off reporting is important to ethtool even BEFORE open */
7067 netif_carrier_off(netdev);
7069 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
7070 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
7071 INIT_WORK(&adapter->fdir_reinit_task, ixgbe_fdir_reinit_task);
7073 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
7074 INIT_WORK(&adapter->check_overtemp_task,
7075 ixgbe_check_overtemp_task);
7076 #ifdef CONFIG_IXGBE_DCA
7077 if (dca_add_requester(&pdev->dev) == 0) {
7078 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
7079 ixgbe_setup_dca(adapter);
7082 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7083 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
7084 for (i = 0; i < adapter->num_vfs; i++)
7085 ixgbe_vf_configuration(pdev, (i | 0x10000000));
7088 /* add san mac addr to netdev */
7089 ixgbe_add_sanmac_netdev(netdev);
7091 e_dev_info("Intel(R) 10 Gigabit Network Connection\n");
7096 ixgbe_release_hw_control(adapter);
7097 ixgbe_clear_interrupt_scheme(adapter);
7100 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7101 ixgbe_disable_sriov(adapter);
7102 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
7103 del_timer_sync(&adapter->sfp_timer);
7104 cancel_work_sync(&adapter->sfp_task);
7105 cancel_work_sync(&adapter->multispeed_fiber_task);
7106 cancel_work_sync(&adapter->sfp_config_module_task);
7107 iounmap(hw->hw_addr);
7109 free_netdev(netdev);
7111 pci_release_selected_regions(pdev,
7112 pci_select_bars(pdev, IORESOURCE_MEM));
7115 pci_disable_device(pdev);
7120 * ixgbe_remove - Device Removal Routine
7121 * @pdev: PCI device information struct
7123 * ixgbe_remove is called by the PCI subsystem to alert the driver
7124 * that it should release a PCI device. The could be caused by a
7125 * Hot-Plug event, or because the driver is going to be removed from
7128 static void __devexit ixgbe_remove(struct pci_dev *pdev)
7130 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7131 struct net_device *netdev = adapter->netdev;
7133 set_bit(__IXGBE_DOWN, &adapter->state);
7134 /* clear the module not found bit to make sure the worker won't
7137 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
7138 del_timer_sync(&adapter->watchdog_timer);
7140 del_timer_sync(&adapter->sfp_timer);
7141 cancel_work_sync(&adapter->watchdog_task);
7142 cancel_work_sync(&adapter->sfp_task);
7143 cancel_work_sync(&adapter->multispeed_fiber_task);
7144 cancel_work_sync(&adapter->sfp_config_module_task);
7145 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
7146 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
7147 cancel_work_sync(&adapter->fdir_reinit_task);
7148 flush_scheduled_work();
7150 #ifdef CONFIG_IXGBE_DCA
7151 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
7152 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
7153 dca_remove_requester(&pdev->dev);
7154 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
7159 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
7160 ixgbe_cleanup_fcoe(adapter);
7162 #endif /* IXGBE_FCOE */
7164 /* remove the added san mac */
7165 ixgbe_del_sanmac_netdev(netdev);
7167 if (netdev->reg_state == NETREG_REGISTERED)
7168 unregister_netdev(netdev);
7170 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7171 ixgbe_disable_sriov(adapter);
7173 ixgbe_clear_interrupt_scheme(adapter);
7175 ixgbe_release_hw_control(adapter);
7177 iounmap(adapter->hw.hw_addr);
7178 pci_release_selected_regions(pdev, pci_select_bars(pdev,
7181 e_dev_info("complete\n");
7183 free_netdev(netdev);
7185 pci_disable_pcie_error_reporting(pdev);
7187 pci_disable_device(pdev);
7191 * ixgbe_io_error_detected - called when PCI error is detected
7192 * @pdev: Pointer to PCI device
7193 * @state: The current pci connection state
7195 * This function is called after a PCI bus error affecting
7196 * this device has been detected.
7198 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
7199 pci_channel_state_t state)
7201 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7202 struct net_device *netdev = adapter->netdev;
7204 netif_device_detach(netdev);
7206 if (state == pci_channel_io_perm_failure)
7207 return PCI_ERS_RESULT_DISCONNECT;
7209 if (netif_running(netdev))
7210 ixgbe_down(adapter);
7211 pci_disable_device(pdev);
7213 /* Request a slot reset. */
7214 return PCI_ERS_RESULT_NEED_RESET;
7218 * ixgbe_io_slot_reset - called after the pci bus has been reset.
7219 * @pdev: Pointer to PCI device
7221 * Restart the card from scratch, as if from a cold-boot.
7223 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7225 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7226 pci_ers_result_t result;
7229 if (pci_enable_device_mem(pdev)) {
7230 e_err(probe, "Cannot re-enable PCI device after reset.\n");
7231 result = PCI_ERS_RESULT_DISCONNECT;
7233 pci_set_master(pdev);
7234 pci_restore_state(pdev);
7235 pci_save_state(pdev);
7237 pci_wake_from_d3(pdev, false);
7239 ixgbe_reset(adapter);
7240 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7241 result = PCI_ERS_RESULT_RECOVERED;
7244 err = pci_cleanup_aer_uncorrect_error_status(pdev);
7246 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7247 "failed 0x%0x\n", err);
7248 /* non-fatal, continue */
7255 * ixgbe_io_resume - called when traffic can start flowing again.
7256 * @pdev: Pointer to PCI device
7258 * This callback is called when the error recovery driver tells us that
7259 * its OK to resume normal operation.
7261 static void ixgbe_io_resume(struct pci_dev *pdev)
7263 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7264 struct net_device *netdev = adapter->netdev;
7266 if (netif_running(netdev)) {
7267 if (ixgbe_up(adapter)) {
7268 e_info(probe, "ixgbe_up failed after reset\n");
7273 netif_device_attach(netdev);
7276 static struct pci_error_handlers ixgbe_err_handler = {
7277 .error_detected = ixgbe_io_error_detected,
7278 .slot_reset = ixgbe_io_slot_reset,
7279 .resume = ixgbe_io_resume,
7282 static struct pci_driver ixgbe_driver = {
7283 .name = ixgbe_driver_name,
7284 .id_table = ixgbe_pci_tbl,
7285 .probe = ixgbe_probe,
7286 .remove = __devexit_p(ixgbe_remove),
7288 .suspend = ixgbe_suspend,
7289 .resume = ixgbe_resume,
7291 .shutdown = ixgbe_shutdown,
7292 .err_handler = &ixgbe_err_handler
7296 * ixgbe_init_module - Driver Registration Routine
7298 * ixgbe_init_module is the first routine called when the driver is
7299 * loaded. All it does is register with the PCI subsystem.
7301 static int __init ixgbe_init_module(void)
7304 pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
7305 pr_info("%s\n", ixgbe_copyright);
7307 #ifdef CONFIG_IXGBE_DCA
7308 dca_register_notify(&dca_notifier);
7311 ret = pci_register_driver(&ixgbe_driver);
7315 module_init(ixgbe_init_module);
7318 * ixgbe_exit_module - Driver Exit Cleanup Routine
7320 * ixgbe_exit_module is called just before the driver is removed
7323 static void __exit ixgbe_exit_module(void)
7325 #ifdef CONFIG_IXGBE_DCA
7326 dca_unregister_notify(&dca_notifier);
7328 pci_unregister_driver(&ixgbe_driver);
7329 rcu_barrier(); /* Wait for completion of call_rcu()'s */
7332 #ifdef CONFIG_IXGBE_DCA
7333 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
7338 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
7339 __ixgbe_notify_dca);
7341 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7344 #endif /* CONFIG_IXGBE_DCA */
7347 * ixgbe_get_hw_dev return device
7348 * used by hardware layer to print debugging information
7350 struct net_device *ixgbe_get_hw_dev(struct ixgbe_hw *hw)
7352 struct ixgbe_adapter *adapter = hw->back;
7353 return adapter->netdev;
7356 module_exit(ixgbe_exit_module);