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 switch (adapter->hw.mac.type) {
593 case ixgbe_mac_82598EB:
594 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
595 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
597 case ixgbe_mac_82599EB:
598 mask = (qmask & 0xFFFFFFFF);
599 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
600 mask = (qmask >> 32);
601 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
608 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_ring *tx_ring,
609 struct ixgbe_tx_buffer *tx_buffer_info)
611 if (tx_buffer_info->dma) {
612 if (tx_buffer_info->mapped_as_page)
613 dma_unmap_page(tx_ring->dev,
615 tx_buffer_info->length,
618 dma_unmap_single(tx_ring->dev,
620 tx_buffer_info->length,
622 tx_buffer_info->dma = 0;
624 if (tx_buffer_info->skb) {
625 dev_kfree_skb_any(tx_buffer_info->skb);
626 tx_buffer_info->skb = NULL;
628 tx_buffer_info->time_stamp = 0;
629 /* tx_buffer_info must be completely set up in the transmit path */
633 * ixgbe_tx_xon_state - check the tx ring xon state
634 * @adapter: the ixgbe adapter
635 * @tx_ring: the corresponding tx_ring
637 * If not in DCB mode, checks TFCS.TXOFF, otherwise, find out the
638 * corresponding TC of this tx_ring when checking TFCS.
640 * Returns : true if in xon state (currently not paused)
642 static inline bool ixgbe_tx_xon_state(struct ixgbe_adapter *adapter,
643 struct ixgbe_ring *tx_ring)
645 u32 txoff = IXGBE_TFCS_TXOFF;
647 #ifdef CONFIG_IXGBE_DCB
648 if (adapter->dcb_cfg.pfc_mode_enable) {
650 int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
651 u8 reg_idx = tx_ring->reg_idx;
653 switch (adapter->hw.mac.type) {
654 case ixgbe_mac_82598EB:
656 txoff = IXGBE_TFCS_TXOFF0;
658 case ixgbe_mac_82599EB:
660 txoff = IXGBE_TFCS_TXOFF;
664 if (tc == 2) /* TC2, TC3 */
665 tc += (reg_idx - 64) >> 4;
666 else if (tc == 3) /* TC4, TC5, TC6, TC7 */
667 tc += 1 + ((reg_idx - 96) >> 3);
668 } else if (dcb_i == 4) {
672 tc += (reg_idx - 64) >> 5;
673 if (tc == 2) /* TC2, TC3 */
674 tc += (reg_idx - 96) >> 4;
685 return IXGBE_READ_REG(&adapter->hw, IXGBE_TFCS) & txoff;
688 static inline bool ixgbe_check_tx_hang(struct ixgbe_adapter *adapter,
689 struct ixgbe_ring *tx_ring,
692 struct ixgbe_hw *hw = &adapter->hw;
694 /* Detect a transmit hang in hardware, this serializes the
695 * check with the clearing of time_stamp and movement of eop */
696 clear_check_for_tx_hang(tx_ring);
697 if (tx_ring->tx_buffer_info[eop].time_stamp &&
698 time_after(jiffies, tx_ring->tx_buffer_info[eop].time_stamp + HZ) &&
699 ixgbe_tx_xon_state(adapter, tx_ring)) {
700 /* detected Tx unit hang */
701 union ixgbe_adv_tx_desc *tx_desc;
702 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, eop);
703 e_err(drv, "Detected Tx Unit Hang\n"
705 " TDH, TDT <%x>, <%x>\n"
706 " next_to_use <%x>\n"
707 " next_to_clean <%x>\n"
708 "tx_buffer_info[next_to_clean]\n"
709 " time_stamp <%lx>\n"
711 tx_ring->queue_index,
712 IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
713 IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
714 tx_ring->next_to_use, eop,
715 tx_ring->tx_buffer_info[eop].time_stamp, jiffies);
722 #define IXGBE_MAX_TXD_PWR 14
723 #define IXGBE_MAX_DATA_PER_TXD (1 << IXGBE_MAX_TXD_PWR)
725 /* Tx Descriptors needed, worst case */
726 #define TXD_USE_COUNT(S) (((S) >> IXGBE_MAX_TXD_PWR) + \
727 (((S) & (IXGBE_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
728 #define DESC_NEEDED (TXD_USE_COUNT(IXGBE_MAX_DATA_PER_TXD) /* skb->data */ + \
729 MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1) /* for context */
731 static void ixgbe_tx_timeout(struct net_device *netdev);
734 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
735 * @q_vector: structure containing interrupt and ring information
736 * @tx_ring: tx ring to clean
738 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
739 struct ixgbe_ring *tx_ring)
741 struct ixgbe_adapter *adapter = q_vector->adapter;
742 union ixgbe_adv_tx_desc *tx_desc, *eop_desc;
743 struct ixgbe_tx_buffer *tx_buffer_info;
744 unsigned int total_bytes = 0, total_packets = 0;
745 u16 i, eop, count = 0;
747 i = tx_ring->next_to_clean;
748 eop = tx_ring->tx_buffer_info[i].next_to_watch;
749 eop_desc = IXGBE_TX_DESC_ADV(tx_ring, eop);
751 while ((eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)) &&
752 (count < tx_ring->work_limit)) {
753 bool cleaned = false;
754 rmb(); /* read buffer_info after eop_desc */
755 for ( ; !cleaned; count++) {
756 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
757 tx_buffer_info = &tx_ring->tx_buffer_info[i];
759 tx_desc->wb.status = 0;
760 cleaned = (i == eop);
763 if (i == tx_ring->count)
766 if (cleaned && tx_buffer_info->skb) {
767 total_bytes += tx_buffer_info->bytecount;
768 total_packets += tx_buffer_info->gso_segs;
771 ixgbe_unmap_and_free_tx_resource(tx_ring,
775 eop = tx_ring->tx_buffer_info[i].next_to_watch;
776 eop_desc = IXGBE_TX_DESC_ADV(tx_ring, eop);
779 tx_ring->next_to_clean = i;
780 tx_ring->total_bytes += total_bytes;
781 tx_ring->total_packets += total_packets;
782 u64_stats_update_begin(&tx_ring->syncp);
783 tx_ring->stats.packets += total_packets;
784 tx_ring->stats.bytes += total_bytes;
785 u64_stats_update_end(&tx_ring->syncp);
787 if (check_for_tx_hang(tx_ring) &&
788 ixgbe_check_tx_hang(adapter, tx_ring, i)) {
789 /* schedule immediate reset if we believe we hung */
790 e_info(probe, "tx hang %d detected, resetting "
791 "adapter\n", adapter->tx_timeout_count + 1);
792 ixgbe_tx_timeout(adapter->netdev);
794 /* the adapter is about to reset, no point in enabling stuff */
798 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
799 if (unlikely(count && netif_carrier_ok(tx_ring->netdev) &&
800 (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
801 /* Make sure that anybody stopping the queue after this
802 * sees the new next_to_clean.
805 if (__netif_subqueue_stopped(tx_ring->netdev, tx_ring->queue_index) &&
806 !test_bit(__IXGBE_DOWN, &adapter->state)) {
807 netif_wake_subqueue(tx_ring->netdev, tx_ring->queue_index);
808 ++tx_ring->tx_stats.restart_queue;
812 return count < tx_ring->work_limit;
815 #ifdef CONFIG_IXGBE_DCA
816 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
817 struct ixgbe_ring *rx_ring,
820 struct ixgbe_hw *hw = &adapter->hw;
822 u8 reg_idx = rx_ring->reg_idx;
824 rxctrl = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(reg_idx));
825 switch (hw->mac.type) {
826 case ixgbe_mac_82598EB:
827 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
828 rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
830 case ixgbe_mac_82599EB:
831 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK_82599;
832 rxctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
833 IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599);
838 rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
839 rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
840 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_RRO_EN);
841 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_WRO_EN |
842 IXGBE_DCA_RXCTRL_DESC_HSRO_EN);
843 IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
846 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
847 struct ixgbe_ring *tx_ring,
850 struct ixgbe_hw *hw = &adapter->hw;
852 u8 reg_idx = tx_ring->reg_idx;
854 switch (hw->mac.type) {
855 case ixgbe_mac_82598EB:
856 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(reg_idx));
857 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
858 txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
859 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
860 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
861 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(reg_idx), txctrl);
863 case ixgbe_mac_82599EB:
864 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(reg_idx));
865 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK_82599;
866 txctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
867 IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599);
868 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
869 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
870 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(reg_idx), txctrl);
877 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
879 struct ixgbe_adapter *adapter = q_vector->adapter;
884 if (q_vector->cpu == cpu)
887 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
888 for (i = 0; i < q_vector->txr_count; i++) {
889 ixgbe_update_tx_dca(adapter, adapter->tx_ring[r_idx], cpu);
890 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
894 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
895 for (i = 0; i < q_vector->rxr_count; i++) {
896 ixgbe_update_rx_dca(adapter, adapter->rx_ring[r_idx], cpu);
897 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
906 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
911 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
914 /* always use CB2 mode, difference is masked in the CB driver */
915 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
917 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
918 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
922 for (i = 0; i < num_q_vectors; i++) {
923 adapter->q_vector[i]->cpu = -1;
924 ixgbe_update_dca(adapter->q_vector[i]);
928 static int __ixgbe_notify_dca(struct device *dev, void *data)
930 struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
931 unsigned long event = *(unsigned long *)data;
933 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
937 case DCA_PROVIDER_ADD:
938 /* if we're already enabled, don't do it again */
939 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
941 if (dca_add_requester(dev) == 0) {
942 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
943 ixgbe_setup_dca(adapter);
946 /* Fall Through since DCA is disabled. */
947 case DCA_PROVIDER_REMOVE:
948 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
949 dca_remove_requester(dev);
950 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
951 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
959 #endif /* CONFIG_IXGBE_DCA */
961 * ixgbe_receive_skb - Send a completed packet up the stack
962 * @adapter: board private structure
963 * @skb: packet to send up
964 * @status: hardware indication of status of receive
965 * @rx_ring: rx descriptor ring (for a specific queue) to setup
966 * @rx_desc: rx descriptor
968 static void ixgbe_receive_skb(struct ixgbe_q_vector *q_vector,
969 struct sk_buff *skb, u8 status,
970 struct ixgbe_ring *ring,
971 union ixgbe_adv_rx_desc *rx_desc)
973 struct ixgbe_adapter *adapter = q_vector->adapter;
974 struct napi_struct *napi = &q_vector->napi;
975 bool is_vlan = (status & IXGBE_RXD_STAT_VP);
976 u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
978 if (is_vlan && (tag & VLAN_VID_MASK))
979 __vlan_hwaccel_put_tag(skb, tag);
981 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
982 napi_gro_receive(napi, skb);
988 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
989 * @adapter: address of board private structure
990 * @status_err: hardware indication of status of receive
991 * @skb: skb currently being received and modified
993 static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
994 union ixgbe_adv_rx_desc *rx_desc,
997 u32 status_err = le32_to_cpu(rx_desc->wb.upper.status_error);
999 skb_checksum_none_assert(skb);
1001 /* Rx csum disabled */
1002 if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
1005 /* if IP and error */
1006 if ((status_err & IXGBE_RXD_STAT_IPCS) &&
1007 (status_err & IXGBE_RXDADV_ERR_IPE)) {
1008 adapter->hw_csum_rx_error++;
1012 if (!(status_err & IXGBE_RXD_STAT_L4CS))
1015 if (status_err & IXGBE_RXDADV_ERR_TCPE) {
1016 u16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1019 * 82599 errata, UDP frames with a 0 checksum can be marked as
1022 if ((pkt_info & IXGBE_RXDADV_PKTTYPE_UDP) &&
1023 (adapter->hw.mac.type == ixgbe_mac_82599EB))
1026 adapter->hw_csum_rx_error++;
1030 /* It must be a TCP or UDP packet with a valid checksum */
1031 skb->ip_summed = CHECKSUM_UNNECESSARY;
1034 static inline void ixgbe_release_rx_desc(struct ixgbe_ring *rx_ring, u32 val)
1037 * Force memory writes to complete before letting h/w
1038 * know there are new descriptors to fetch. (Only
1039 * applicable for weak-ordered memory model archs,
1043 writel(val, rx_ring->tail);
1047 * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
1048 * @rx_ring: ring to place buffers on
1049 * @cleaned_count: number of buffers to replace
1051 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1053 union ixgbe_adv_rx_desc *rx_desc;
1054 struct ixgbe_rx_buffer *bi;
1055 struct sk_buff *skb;
1056 u16 i = rx_ring->next_to_use;
1058 /* do nothing if no valid netdev defined */
1059 if (!rx_ring->netdev)
1062 while (cleaned_count--) {
1063 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1064 bi = &rx_ring->rx_buffer_info[i];
1068 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1069 rx_ring->rx_buf_len);
1071 rx_ring->rx_stats.alloc_rx_buff_failed++;
1074 /* initialize queue mapping */
1075 skb_record_rx_queue(skb, rx_ring->queue_index);
1080 bi->dma = dma_map_single(rx_ring->dev,
1082 rx_ring->rx_buf_len,
1084 if (dma_mapping_error(rx_ring->dev, bi->dma)) {
1085 rx_ring->rx_stats.alloc_rx_buff_failed++;
1091 if (ring_is_ps_enabled(rx_ring)) {
1093 bi->page = netdev_alloc_page(rx_ring->netdev);
1095 rx_ring->rx_stats.alloc_rx_page_failed++;
1100 if (!bi->page_dma) {
1101 /* use a half page if we're re-using */
1102 bi->page_offset ^= PAGE_SIZE / 2;
1103 bi->page_dma = dma_map_page(rx_ring->dev,
1108 if (dma_mapping_error(rx_ring->dev,
1110 rx_ring->rx_stats.alloc_rx_page_failed++;
1116 /* Refresh the desc even if buffer_addrs didn't change
1117 * because each write-back erases this info. */
1118 rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
1119 rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
1121 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
1122 rx_desc->read.hdr_addr = 0;
1126 if (i == rx_ring->count)
1131 if (rx_ring->next_to_use != i) {
1132 rx_ring->next_to_use = i;
1133 ixgbe_release_rx_desc(rx_ring, i);
1137 static inline u16 ixgbe_get_hlen(union ixgbe_adv_rx_desc *rx_desc)
1139 /* HW will not DMA in data larger than the given buffer, even if it
1140 * parses the (NFS, of course) header to be larger. In that case, it
1141 * fills the header buffer and spills the rest into the page.
1143 u16 hdr_info = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info);
1144 u16 hlen = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
1145 IXGBE_RXDADV_HDRBUFLEN_SHIFT;
1146 if (hlen > IXGBE_RX_HDR_SIZE)
1147 hlen = IXGBE_RX_HDR_SIZE;
1152 * ixgbe_transform_rsc_queue - change rsc queue into a full packet
1153 * @skb: pointer to the last skb in the rsc queue
1155 * This function changes a queue full of hw rsc buffers into a completed
1156 * packet. It uses the ->prev pointers to find the first packet and then
1157 * turns it into the frag list owner.
1159 static inline struct sk_buff *ixgbe_transform_rsc_queue(struct sk_buff *skb)
1161 unsigned int frag_list_size = 0;
1162 unsigned int skb_cnt = 1;
1165 struct sk_buff *prev = skb->prev;
1166 frag_list_size += skb->len;
1172 skb_shinfo(skb)->frag_list = skb->next;
1174 skb->len += frag_list_size;
1175 skb->data_len += frag_list_size;
1176 skb->truesize += frag_list_size;
1177 IXGBE_RSC_CB(skb)->skb_cnt = skb_cnt;
1182 static inline bool ixgbe_get_rsc_state(union ixgbe_adv_rx_desc *rx_desc)
1184 return !!(le32_to_cpu(rx_desc->wb.lower.lo_dword.data) &
1185 IXGBE_RXDADV_RSCCNT_MASK);
1188 static void ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1189 struct ixgbe_ring *rx_ring,
1190 int *work_done, int work_to_do)
1192 struct ixgbe_adapter *adapter = q_vector->adapter;
1193 union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
1194 struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
1195 struct sk_buff *skb;
1196 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1197 const int current_node = numa_node_id();
1200 #endif /* IXGBE_FCOE */
1203 u16 cleaned_count = 0;
1204 bool pkt_is_rsc = false;
1206 i = rx_ring->next_to_clean;
1207 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1208 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1210 while (staterr & IXGBE_RXD_STAT_DD) {
1213 rmb(); /* read descriptor and rx_buffer_info after status DD */
1215 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1217 skb = rx_buffer_info->skb;
1218 rx_buffer_info->skb = NULL;
1219 prefetch(skb->data);
1221 if (ring_is_rsc_enabled(rx_ring))
1222 pkt_is_rsc = ixgbe_get_rsc_state(rx_desc);
1224 /* if this is a skb from previous receive DMA will be 0 */
1225 if (rx_buffer_info->dma) {
1228 !(staterr & IXGBE_RXD_STAT_EOP) &&
1231 * When HWRSC is enabled, delay unmapping
1232 * of the first packet. It carries the
1233 * header information, HW may still
1234 * access the header after the writeback.
1235 * Only unmap it when EOP is reached
1237 IXGBE_RSC_CB(skb)->delay_unmap = true;
1238 IXGBE_RSC_CB(skb)->dma = rx_buffer_info->dma;
1240 dma_unmap_single(rx_ring->dev,
1241 rx_buffer_info->dma,
1242 rx_ring->rx_buf_len,
1245 rx_buffer_info->dma = 0;
1247 if (ring_is_ps_enabled(rx_ring)) {
1248 hlen = ixgbe_get_hlen(rx_desc);
1249 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1251 hlen = le16_to_cpu(rx_desc->wb.upper.length);
1256 /* assume packet split since header is unmapped */
1257 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1261 dma_unmap_page(rx_ring->dev,
1262 rx_buffer_info->page_dma,
1265 rx_buffer_info->page_dma = 0;
1266 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1267 rx_buffer_info->page,
1268 rx_buffer_info->page_offset,
1271 if ((page_count(rx_buffer_info->page) == 1) &&
1272 (page_to_nid(rx_buffer_info->page) == current_node))
1273 get_page(rx_buffer_info->page);
1275 rx_buffer_info->page = NULL;
1277 skb->len += upper_len;
1278 skb->data_len += upper_len;
1279 skb->truesize += upper_len;
1283 if (i == rx_ring->count)
1286 next_rxd = IXGBE_RX_DESC_ADV(rx_ring, i);
1291 u32 nextp = (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
1292 IXGBE_RXDADV_NEXTP_SHIFT;
1293 next_buffer = &rx_ring->rx_buffer_info[nextp];
1295 next_buffer = &rx_ring->rx_buffer_info[i];
1298 if (!(staterr & IXGBE_RXD_STAT_EOP)) {
1299 if (ring_is_ps_enabled(rx_ring)) {
1300 rx_buffer_info->skb = next_buffer->skb;
1301 rx_buffer_info->dma = next_buffer->dma;
1302 next_buffer->skb = skb;
1303 next_buffer->dma = 0;
1305 skb->next = next_buffer->skb;
1306 skb->next->prev = skb;
1308 rx_ring->rx_stats.non_eop_descs++;
1313 skb = ixgbe_transform_rsc_queue(skb);
1314 /* if we got here without RSC the packet is invalid */
1316 __pskb_trim(skb, 0);
1317 rx_buffer_info->skb = skb;
1322 if (ring_is_rsc_enabled(rx_ring)) {
1323 if (IXGBE_RSC_CB(skb)->delay_unmap) {
1324 dma_unmap_single(rx_ring->dev,
1325 IXGBE_RSC_CB(skb)->dma,
1326 rx_ring->rx_buf_len,
1328 IXGBE_RSC_CB(skb)->dma = 0;
1329 IXGBE_RSC_CB(skb)->delay_unmap = false;
1333 if (ring_is_ps_enabled(rx_ring))
1334 rx_ring->rx_stats.rsc_count +=
1335 skb_shinfo(skb)->nr_frags;
1337 rx_ring->rx_stats.rsc_count +=
1338 IXGBE_RSC_CB(skb)->skb_cnt;
1339 rx_ring->rx_stats.rsc_flush++;
1342 /* ERR_MASK will only have valid bits if EOP set */
1343 if (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) {
1344 /* trim packet back to size 0 and recycle it */
1345 __pskb_trim(skb, 0);
1346 rx_buffer_info->skb = skb;
1350 ixgbe_rx_checksum(adapter, rx_desc, skb);
1352 /* probably a little skewed due to removing CRC */
1353 total_rx_bytes += skb->len;
1356 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1358 /* if ddp, not passing to ULD unless for FCP_RSP or error */
1359 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
1360 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
1364 #endif /* IXGBE_FCOE */
1365 ixgbe_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
1368 rx_desc->wb.upper.status_error = 0;
1371 if (*work_done >= work_to_do)
1374 /* return some buffers to hardware, one at a time is too slow */
1375 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
1376 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1380 /* use prefetched values */
1382 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1385 rx_ring->next_to_clean = i;
1386 cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
1389 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1392 /* include DDPed FCoE data */
1393 if (ddp_bytes > 0) {
1396 mss = rx_ring->netdev->mtu - sizeof(struct fcoe_hdr) -
1397 sizeof(struct fc_frame_header) -
1398 sizeof(struct fcoe_crc_eof);
1401 total_rx_bytes += ddp_bytes;
1402 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
1404 #endif /* IXGBE_FCOE */
1406 rx_ring->total_packets += total_rx_packets;
1407 rx_ring->total_bytes += total_rx_bytes;
1408 u64_stats_update_begin(&rx_ring->syncp);
1409 rx_ring->stats.packets += total_rx_packets;
1410 rx_ring->stats.bytes += total_rx_bytes;
1411 u64_stats_update_end(&rx_ring->syncp);
1414 static int ixgbe_clean_rxonly(struct napi_struct *, int);
1416 * ixgbe_configure_msix - Configure MSI-X hardware
1417 * @adapter: board private structure
1419 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1422 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1424 struct ixgbe_q_vector *q_vector;
1425 int i, q_vectors, v_idx, r_idx;
1428 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1431 * Populate the IVAR table and set the ITR values to the
1432 * corresponding register.
1434 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1435 q_vector = adapter->q_vector[v_idx];
1436 /* XXX for_each_set_bit(...) */
1437 r_idx = find_first_bit(q_vector->rxr_idx,
1438 adapter->num_rx_queues);
1440 for (i = 0; i < q_vector->rxr_count; i++) {
1441 u8 reg_idx = adapter->rx_ring[r_idx]->reg_idx;
1442 ixgbe_set_ivar(adapter, 0, reg_idx, v_idx);
1443 r_idx = find_next_bit(q_vector->rxr_idx,
1444 adapter->num_rx_queues,
1447 r_idx = find_first_bit(q_vector->txr_idx,
1448 adapter->num_tx_queues);
1450 for (i = 0; i < q_vector->txr_count; i++) {
1451 u8 reg_idx = adapter->tx_ring[r_idx]->reg_idx;
1452 ixgbe_set_ivar(adapter, 1, reg_idx, v_idx);
1453 r_idx = find_next_bit(q_vector->txr_idx,
1454 adapter->num_tx_queues,
1458 if (q_vector->txr_count && !q_vector->rxr_count)
1460 q_vector->eitr = adapter->tx_eitr_param;
1461 else if (q_vector->rxr_count)
1463 q_vector->eitr = adapter->rx_eitr_param;
1465 ixgbe_write_eitr(q_vector);
1466 /* If Flow Director is enabled, set interrupt affinity */
1467 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
1468 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)) {
1470 * Allocate the affinity_hint cpumask, assign the mask
1471 * for this vector, and set our affinity_hint for
1474 if (!alloc_cpumask_var(&q_vector->affinity_mask,
1477 cpumask_set_cpu(v_idx, q_vector->affinity_mask);
1478 irq_set_affinity_hint(adapter->msix_entries[v_idx].vector,
1479 q_vector->affinity_mask);
1483 switch (adapter->hw.mac.type) {
1484 case ixgbe_mac_82598EB:
1485 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1488 case ixgbe_mac_82599EB:
1489 ixgbe_set_ivar(adapter, -1, 1, v_idx);
1495 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1497 /* set up to autoclear timer, and the vectors */
1498 mask = IXGBE_EIMS_ENABLE_MASK;
1499 if (adapter->num_vfs)
1500 mask &= ~(IXGBE_EIMS_OTHER |
1501 IXGBE_EIMS_MAILBOX |
1504 mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
1505 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1508 enum latency_range {
1512 latency_invalid = 255
1516 * ixgbe_update_itr - update the dynamic ITR value based on statistics
1517 * @adapter: pointer to adapter
1518 * @eitr: eitr setting (ints per sec) to give last timeslice
1519 * @itr_setting: current throttle rate in ints/second
1520 * @packets: the number of packets during this measurement interval
1521 * @bytes: the number of bytes during this measurement interval
1523 * Stores a new ITR value based on packets and byte
1524 * counts during the last interrupt. The advantage of per interrupt
1525 * computation is faster updates and more accurate ITR for the current
1526 * traffic pattern. Constants in this function were computed
1527 * based on theoretical maximum wire speed and thresholds were set based
1528 * on testing data as well as attempting to minimize response time
1529 * while increasing bulk throughput.
1530 * this functionality is controlled by the InterruptThrottleRate module
1531 * parameter (see ixgbe_param.c)
1533 static u8 ixgbe_update_itr(struct ixgbe_adapter *adapter,
1534 u32 eitr, u8 itr_setting,
1535 int packets, int bytes)
1537 unsigned int retval = itr_setting;
1542 goto update_itr_done;
1545 /* simple throttlerate management
1546 * 0-20MB/s lowest (100000 ints/s)
1547 * 20-100MB/s low (20000 ints/s)
1548 * 100-1249MB/s bulk (8000 ints/s)
1550 /* what was last interrupt timeslice? */
1551 timepassed_us = 1000000/eitr;
1552 bytes_perint = bytes / timepassed_us; /* bytes/usec */
1554 switch (itr_setting) {
1555 case lowest_latency:
1556 if (bytes_perint > adapter->eitr_low)
1557 retval = low_latency;
1560 if (bytes_perint > adapter->eitr_high)
1561 retval = bulk_latency;
1562 else if (bytes_perint <= adapter->eitr_low)
1563 retval = lowest_latency;
1566 if (bytes_perint <= adapter->eitr_high)
1567 retval = low_latency;
1576 * ixgbe_write_eitr - write EITR register in hardware specific way
1577 * @q_vector: structure containing interrupt and ring information
1579 * This function is made to be called by ethtool and by the driver
1580 * when it needs to update EITR registers at runtime. Hardware
1581 * specific quirks/differences are taken care of here.
1583 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1585 struct ixgbe_adapter *adapter = q_vector->adapter;
1586 struct ixgbe_hw *hw = &adapter->hw;
1587 int v_idx = q_vector->v_idx;
1588 u32 itr_reg = EITR_INTS_PER_SEC_TO_REG(q_vector->eitr);
1590 switch (adapter->hw.mac.type) {
1591 case ixgbe_mac_82598EB:
1592 /* must write high and low 16 bits to reset counter */
1593 itr_reg |= (itr_reg << 16);
1595 case ixgbe_mac_82599EB:
1597 * 82599 can support a value of zero, so allow it for
1598 * max interrupt rate, but there is an errata where it can
1599 * not be zero with RSC
1602 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
1606 * set the WDIS bit to not clear the timer bits and cause an
1607 * immediate assertion of the interrupt
1609 itr_reg |= IXGBE_EITR_CNT_WDIS;
1614 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1617 static void ixgbe_set_itr_msix(struct ixgbe_q_vector *q_vector)
1619 struct ixgbe_adapter *adapter = q_vector->adapter;
1622 u8 current_itr, ret_itr;
1624 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1625 for (i = 0; i < q_vector->txr_count; i++) {
1626 struct ixgbe_ring *tx_ring = adapter->tx_ring[r_idx];
1627 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1629 tx_ring->total_packets,
1630 tx_ring->total_bytes);
1631 /* if the result for this queue would decrease interrupt
1632 * rate for this vector then use that result */
1633 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
1634 q_vector->tx_itr - 1 : ret_itr);
1635 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1639 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1640 for (i = 0; i < q_vector->rxr_count; i++) {
1641 struct ixgbe_ring *rx_ring = adapter->rx_ring[r_idx];
1642 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1644 rx_ring->total_packets,
1645 rx_ring->total_bytes);
1646 /* if the result for this queue would decrease interrupt
1647 * rate for this vector then use that result */
1648 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
1649 q_vector->rx_itr - 1 : ret_itr);
1650 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1654 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1656 switch (current_itr) {
1657 /* counts and packets in update_itr are dependent on these numbers */
1658 case lowest_latency:
1662 new_itr = 20000; /* aka hwitr = ~200 */
1670 if (new_itr != q_vector->eitr) {
1671 /* do an exponential smoothing */
1672 new_itr = ((q_vector->eitr * 9) + new_itr)/10;
1674 /* save the algorithm value here, not the smoothed one */
1675 q_vector->eitr = new_itr;
1677 ixgbe_write_eitr(q_vector);
1682 * ixgbe_check_overtemp_task - worker thread to check over tempurature
1683 * @work: pointer to work_struct containing our data
1685 static void ixgbe_check_overtemp_task(struct work_struct *work)
1687 struct ixgbe_adapter *adapter = container_of(work,
1688 struct ixgbe_adapter,
1689 check_overtemp_task);
1690 struct ixgbe_hw *hw = &adapter->hw;
1691 u32 eicr = adapter->interrupt_event;
1693 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
1696 switch (hw->device_id) {
1697 case IXGBE_DEV_ID_82599_T3_LOM: {
1699 bool link_up = false;
1701 if (hw->mac.ops.check_link)
1702 hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
1704 if (((eicr & IXGBE_EICR_GPI_SDP0) && (!link_up)) ||
1705 (eicr & IXGBE_EICR_LSC))
1706 /* Check if this is due to overtemp */
1707 if (hw->phy.ops.check_overtemp(hw) == IXGBE_ERR_OVERTEMP)
1712 if (!(eicr & IXGBE_EICR_GPI_SDP0))
1717 "Network adapter has been stopped because it has over heated. "
1718 "Restart the computer. If the problem persists, "
1719 "power off the system and replace the adapter\n");
1720 /* write to clear the interrupt */
1721 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP0);
1724 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
1726 struct ixgbe_hw *hw = &adapter->hw;
1728 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
1729 (eicr & IXGBE_EICR_GPI_SDP1)) {
1730 e_crit(probe, "Fan has stopped, replace the adapter\n");
1731 /* write to clear the interrupt */
1732 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1736 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
1738 struct ixgbe_hw *hw = &adapter->hw;
1740 if (eicr & IXGBE_EICR_GPI_SDP2) {
1741 /* Clear the interrupt */
1742 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
1743 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1744 schedule_work(&adapter->sfp_config_module_task);
1747 if (eicr & IXGBE_EICR_GPI_SDP1) {
1748 /* Clear the interrupt */
1749 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1750 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1751 schedule_work(&adapter->multispeed_fiber_task);
1755 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1757 struct ixgbe_hw *hw = &adapter->hw;
1760 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1761 adapter->link_check_timeout = jiffies;
1762 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1763 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1764 IXGBE_WRITE_FLUSH(hw);
1765 schedule_work(&adapter->watchdog_task);
1769 static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
1771 struct net_device *netdev = data;
1772 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1773 struct ixgbe_hw *hw = &adapter->hw;
1777 * Workaround for Silicon errata. Use clear-by-write instead
1778 * of clear-by-read. Reading with EICS will return the
1779 * interrupt causes without clearing, which later be done
1780 * with the write to EICR.
1782 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1783 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1785 if (eicr & IXGBE_EICR_LSC)
1786 ixgbe_check_lsc(adapter);
1788 if (eicr & IXGBE_EICR_MAILBOX)
1789 ixgbe_msg_task(adapter);
1791 switch (hw->mac.type) {
1792 case ixgbe_mac_82599EB:
1793 /* Handle Flow Director Full threshold interrupt */
1794 if (eicr & IXGBE_EICR_FLOW_DIR) {
1796 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_FLOW_DIR);
1797 /* Disable transmits before FDIR Re-initialization */
1798 netif_tx_stop_all_queues(netdev);
1799 for (i = 0; i < adapter->num_tx_queues; i++) {
1800 struct ixgbe_ring *tx_ring =
1801 adapter->tx_ring[i];
1802 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
1804 schedule_work(&adapter->fdir_reinit_task);
1807 ixgbe_check_sfp_event(adapter, eicr);
1808 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
1809 ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC))) {
1810 adapter->interrupt_event = eicr;
1811 schedule_work(&adapter->check_overtemp_task);
1818 ixgbe_check_fan_failure(adapter, eicr);
1820 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1821 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_OTHER);
1826 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
1830 struct ixgbe_hw *hw = &adapter->hw;
1832 switch (hw->mac.type) {
1833 case ixgbe_mac_82598EB:
1834 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1835 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
1837 case ixgbe_mac_82599EB:
1838 mask = (qmask & 0xFFFFFFFF);
1840 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
1841 mask = (qmask >> 32);
1843 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
1848 /* skip the flush */
1851 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
1855 struct ixgbe_hw *hw = &adapter->hw;
1857 switch (hw->mac.type) {
1858 case ixgbe_mac_82598EB:
1859 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1860 IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
1862 case ixgbe_mac_82599EB:
1863 mask = (qmask & 0xFFFFFFFF);
1865 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
1866 mask = (qmask >> 32);
1868 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
1873 /* skip the flush */
1876 static irqreturn_t ixgbe_msix_clean_tx(int irq, void *data)
1878 struct ixgbe_q_vector *q_vector = data;
1879 struct ixgbe_adapter *adapter = q_vector->adapter;
1880 struct ixgbe_ring *tx_ring;
1883 if (!q_vector->txr_count)
1886 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1887 for (i = 0; i < q_vector->txr_count; i++) {
1888 tx_ring = adapter->tx_ring[r_idx];
1889 tx_ring->total_bytes = 0;
1890 tx_ring->total_packets = 0;
1891 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1895 /* EIAM disabled interrupts (on this vector) for us */
1896 napi_schedule(&q_vector->napi);
1902 * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
1904 * @data: pointer to our q_vector struct for this interrupt vector
1906 static irqreturn_t ixgbe_msix_clean_rx(int irq, void *data)
1908 struct ixgbe_q_vector *q_vector = data;
1909 struct ixgbe_adapter *adapter = q_vector->adapter;
1910 struct ixgbe_ring *rx_ring;
1914 #ifdef CONFIG_IXGBE_DCA
1915 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1916 ixgbe_update_dca(q_vector);
1919 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1920 for (i = 0; i < q_vector->rxr_count; i++) {
1921 rx_ring = adapter->rx_ring[r_idx];
1922 rx_ring->total_bytes = 0;
1923 rx_ring->total_packets = 0;
1924 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1928 if (!q_vector->rxr_count)
1931 /* EIAM disabled interrupts (on this vector) for us */
1932 napi_schedule(&q_vector->napi);
1937 static irqreturn_t ixgbe_msix_clean_many(int irq, void *data)
1939 struct ixgbe_q_vector *q_vector = data;
1940 struct ixgbe_adapter *adapter = q_vector->adapter;
1941 struct ixgbe_ring *ring;
1945 if (!q_vector->txr_count && !q_vector->rxr_count)
1948 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1949 for (i = 0; i < q_vector->txr_count; i++) {
1950 ring = adapter->tx_ring[r_idx];
1951 ring->total_bytes = 0;
1952 ring->total_packets = 0;
1953 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1957 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1958 for (i = 0; i < q_vector->rxr_count; i++) {
1959 ring = adapter->rx_ring[r_idx];
1960 ring->total_bytes = 0;
1961 ring->total_packets = 0;
1962 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1966 /* EIAM disabled interrupts (on this vector) for us */
1967 napi_schedule(&q_vector->napi);
1973 * ixgbe_clean_rxonly - msix (aka one shot) rx clean routine
1974 * @napi: napi struct with our devices info in it
1975 * @budget: amount of work driver is allowed to do this pass, in packets
1977 * This function is optimized for cleaning one queue only on a single
1980 static int ixgbe_clean_rxonly(struct napi_struct *napi, int budget)
1982 struct ixgbe_q_vector *q_vector =
1983 container_of(napi, struct ixgbe_q_vector, napi);
1984 struct ixgbe_adapter *adapter = q_vector->adapter;
1985 struct ixgbe_ring *rx_ring = NULL;
1989 #ifdef CONFIG_IXGBE_DCA
1990 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1991 ixgbe_update_dca(q_vector);
1994 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1995 rx_ring = adapter->rx_ring[r_idx];
1997 ixgbe_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
1999 /* If all Rx work done, exit the polling mode */
2000 if (work_done < budget) {
2001 napi_complete(napi);
2002 if (adapter->rx_itr_setting & 1)
2003 ixgbe_set_itr_msix(q_vector);
2004 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2005 ixgbe_irq_enable_queues(adapter,
2006 ((u64)1 << q_vector->v_idx));
2013 * ixgbe_clean_rxtx_many - msix (aka one shot) rx clean routine
2014 * @napi: napi struct with our devices info in it
2015 * @budget: amount of work driver is allowed to do this pass, in packets
2017 * This function will clean more than one rx queue associated with a
2020 static int ixgbe_clean_rxtx_many(struct napi_struct *napi, int budget)
2022 struct ixgbe_q_vector *q_vector =
2023 container_of(napi, struct ixgbe_q_vector, napi);
2024 struct ixgbe_adapter *adapter = q_vector->adapter;
2025 struct ixgbe_ring *ring = NULL;
2026 int work_done = 0, i;
2028 bool tx_clean_complete = true;
2030 #ifdef CONFIG_IXGBE_DCA
2031 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2032 ixgbe_update_dca(q_vector);
2035 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
2036 for (i = 0; i < q_vector->txr_count; i++) {
2037 ring = adapter->tx_ring[r_idx];
2038 tx_clean_complete &= ixgbe_clean_tx_irq(q_vector, ring);
2039 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
2043 /* attempt to distribute budget to each queue fairly, but don't allow
2044 * the budget to go below 1 because we'll exit polling */
2045 budget /= (q_vector->rxr_count ?: 1);
2046 budget = max(budget, 1);
2047 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2048 for (i = 0; i < q_vector->rxr_count; i++) {
2049 ring = adapter->rx_ring[r_idx];
2050 ixgbe_clean_rx_irq(q_vector, ring, &work_done, budget);
2051 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
2055 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2056 ring = adapter->rx_ring[r_idx];
2057 /* If all Rx work done, exit the polling mode */
2058 if (work_done < budget) {
2059 napi_complete(napi);
2060 if (adapter->rx_itr_setting & 1)
2061 ixgbe_set_itr_msix(q_vector);
2062 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2063 ixgbe_irq_enable_queues(adapter,
2064 ((u64)1 << q_vector->v_idx));
2072 * ixgbe_clean_txonly - msix (aka one shot) tx clean routine
2073 * @napi: napi struct with our devices info in it
2074 * @budget: amount of work driver is allowed to do this pass, in packets
2076 * This function is optimized for cleaning one queue only on a single
2079 static int ixgbe_clean_txonly(struct napi_struct *napi, int budget)
2081 struct ixgbe_q_vector *q_vector =
2082 container_of(napi, struct ixgbe_q_vector, napi);
2083 struct ixgbe_adapter *adapter = q_vector->adapter;
2084 struct ixgbe_ring *tx_ring = NULL;
2088 #ifdef CONFIG_IXGBE_DCA
2089 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2090 ixgbe_update_dca(q_vector);
2093 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
2094 tx_ring = adapter->tx_ring[r_idx];
2096 if (!ixgbe_clean_tx_irq(q_vector, tx_ring))
2099 /* If all Tx work done, exit the polling mode */
2100 if (work_done < budget) {
2101 napi_complete(napi);
2102 if (adapter->tx_itr_setting & 1)
2103 ixgbe_set_itr_msix(q_vector);
2104 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2105 ixgbe_irq_enable_queues(adapter,
2106 ((u64)1 << q_vector->v_idx));
2112 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
2115 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2117 set_bit(r_idx, q_vector->rxr_idx);
2118 q_vector->rxr_count++;
2121 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
2124 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2126 set_bit(t_idx, q_vector->txr_idx);
2127 q_vector->txr_count++;
2131 * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
2132 * @adapter: board private structure to initialize
2133 * @vectors: allotted vector count for descriptor rings
2135 * This function maps descriptor rings to the queue-specific vectors
2136 * we were allotted through the MSI-X enabling code. Ideally, we'd have
2137 * one vector per ring/queue, but on a constrained vector budget, we
2138 * group the rings as "efficiently" as possible. You would add new
2139 * mapping configurations in here.
2141 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter,
2145 int rxr_idx = 0, txr_idx = 0;
2146 int rxr_remaining = adapter->num_rx_queues;
2147 int txr_remaining = adapter->num_tx_queues;
2152 /* No mapping required if MSI-X is disabled. */
2153 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2157 * The ideal configuration...
2158 * We have enough vectors to map one per queue.
2160 if (vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
2161 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
2162 map_vector_to_rxq(adapter, v_start, rxr_idx);
2164 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
2165 map_vector_to_txq(adapter, v_start, txr_idx);
2171 * If we don't have enough vectors for a 1-to-1
2172 * mapping, we'll have to group them so there are
2173 * multiple queues per vector.
2175 /* Re-adjusting *qpv takes care of the remainder. */
2176 for (i = v_start; i < vectors; i++) {
2177 rqpv = DIV_ROUND_UP(rxr_remaining, vectors - i);
2178 for (j = 0; j < rqpv; j++) {
2179 map_vector_to_rxq(adapter, i, rxr_idx);
2184 for (i = v_start; i < vectors; i++) {
2185 tqpv = DIV_ROUND_UP(txr_remaining, vectors - i);
2186 for (j = 0; j < tqpv; j++) {
2187 map_vector_to_txq(adapter, i, txr_idx);
2198 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2199 * @adapter: board private structure
2201 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2202 * interrupts from the kernel.
2204 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2206 struct net_device *netdev = adapter->netdev;
2207 irqreturn_t (*handler)(int, void *);
2208 int i, vector, q_vectors, err;
2211 /* Decrement for Other and TCP Timer vectors */
2212 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2214 /* Map the Tx/Rx rings to the vectors we were allotted. */
2215 err = ixgbe_map_rings_to_vectors(adapter, q_vectors);
2219 #define SET_HANDLER(_v) ((!(_v)->rxr_count) ? &ixgbe_msix_clean_tx : \
2220 (!(_v)->txr_count) ? &ixgbe_msix_clean_rx : \
2221 &ixgbe_msix_clean_many)
2222 for (vector = 0; vector < q_vectors; vector++) {
2223 handler = SET_HANDLER(adapter->q_vector[vector]);
2225 if (handler == &ixgbe_msix_clean_rx) {
2226 sprintf(adapter->name[vector], "%s-%s-%d",
2227 netdev->name, "rx", ri++);
2228 } else if (handler == &ixgbe_msix_clean_tx) {
2229 sprintf(adapter->name[vector], "%s-%s-%d",
2230 netdev->name, "tx", ti++);
2232 sprintf(adapter->name[vector], "%s-%s-%d",
2233 netdev->name, "TxRx", ri++);
2237 err = request_irq(adapter->msix_entries[vector].vector,
2238 handler, 0, adapter->name[vector],
2239 adapter->q_vector[vector]);
2241 e_err(probe, "request_irq failed for MSIX interrupt "
2242 "Error: %d\n", err);
2243 goto free_queue_irqs;
2247 sprintf(adapter->name[vector], "%s:lsc", netdev->name);
2248 err = request_irq(adapter->msix_entries[vector].vector,
2249 ixgbe_msix_lsc, 0, adapter->name[vector], netdev);
2251 e_err(probe, "request_irq for msix_lsc failed: %d\n", err);
2252 goto free_queue_irqs;
2258 for (i = vector - 1; i >= 0; i--)
2259 free_irq(adapter->msix_entries[--vector].vector,
2260 adapter->q_vector[i]);
2261 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2262 pci_disable_msix(adapter->pdev);
2263 kfree(adapter->msix_entries);
2264 adapter->msix_entries = NULL;
2269 static void ixgbe_set_itr(struct ixgbe_adapter *adapter)
2271 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2272 struct ixgbe_ring *rx_ring = adapter->rx_ring[0];
2273 struct ixgbe_ring *tx_ring = adapter->tx_ring[0];
2274 u32 new_itr = q_vector->eitr;
2277 q_vector->tx_itr = ixgbe_update_itr(adapter, new_itr,
2279 tx_ring->total_packets,
2280 tx_ring->total_bytes);
2281 q_vector->rx_itr = ixgbe_update_itr(adapter, new_itr,
2283 rx_ring->total_packets,
2284 rx_ring->total_bytes);
2286 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
2288 switch (current_itr) {
2289 /* counts and packets in update_itr are dependent on these numbers */
2290 case lowest_latency:
2294 new_itr = 20000; /* aka hwitr = ~200 */
2303 if (new_itr != q_vector->eitr) {
2304 /* do an exponential smoothing */
2305 new_itr = ((q_vector->eitr * 9) + new_itr)/10;
2307 /* save the algorithm value here */
2308 q_vector->eitr = new_itr;
2310 ixgbe_write_eitr(q_vector);
2315 * ixgbe_irq_enable - Enable default interrupt generation settings
2316 * @adapter: board private structure
2318 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2323 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2324 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2325 mask |= IXGBE_EIMS_GPI_SDP0;
2326 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2327 mask |= IXGBE_EIMS_GPI_SDP1;
2328 switch (adapter->hw.mac.type) {
2329 case ixgbe_mac_82599EB:
2330 mask |= IXGBE_EIMS_ECC;
2331 mask |= IXGBE_EIMS_GPI_SDP1;
2332 mask |= IXGBE_EIMS_GPI_SDP2;
2333 if (adapter->num_vfs)
2334 mask |= IXGBE_EIMS_MAILBOX;
2339 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
2340 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
2341 mask |= IXGBE_EIMS_FLOW_DIR;
2343 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2345 ixgbe_irq_enable_queues(adapter, ~0);
2347 IXGBE_WRITE_FLUSH(&adapter->hw);
2349 if (adapter->num_vfs > 32) {
2350 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
2351 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2356 * ixgbe_intr - legacy mode Interrupt Handler
2357 * @irq: interrupt number
2358 * @data: pointer to a network interface device structure
2360 static irqreturn_t ixgbe_intr(int irq, void *data)
2362 struct net_device *netdev = data;
2363 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2364 struct ixgbe_hw *hw = &adapter->hw;
2365 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2369 * Workaround for silicon errata on 82598. Mask the interrupts
2370 * before the read of EICR.
2372 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2374 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2375 * therefore no explict interrupt disable is necessary */
2376 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2379 * shared interrupt alert!
2380 * make sure interrupts are enabled because the read will
2381 * have disabled interrupts due to EIAM
2382 * finish the workaround of silicon errata on 82598. Unmask
2383 * the interrupt that we masked before the EICR read.
2385 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2386 ixgbe_irq_enable(adapter, true, true);
2387 return IRQ_NONE; /* Not our interrupt */
2390 if (eicr & IXGBE_EICR_LSC)
2391 ixgbe_check_lsc(adapter);
2393 switch (hw->mac.type) {
2394 case ixgbe_mac_82599EB:
2395 ixgbe_check_sfp_event(adapter, eicr);
2396 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2397 ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC))) {
2398 adapter->interrupt_event = eicr;
2399 schedule_work(&adapter->check_overtemp_task);
2406 ixgbe_check_fan_failure(adapter, eicr);
2408 if (napi_schedule_prep(&(q_vector->napi))) {
2409 adapter->tx_ring[0]->total_packets = 0;
2410 adapter->tx_ring[0]->total_bytes = 0;
2411 adapter->rx_ring[0]->total_packets = 0;
2412 adapter->rx_ring[0]->total_bytes = 0;
2413 /* would disable interrupts here but EIAM disabled it */
2414 __napi_schedule(&(q_vector->napi));
2418 * re-enable link(maybe) and non-queue interrupts, no flush.
2419 * ixgbe_poll will re-enable the queue interrupts
2422 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2423 ixgbe_irq_enable(adapter, false, false);
2428 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
2430 int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2432 for (i = 0; i < q_vectors; i++) {
2433 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
2434 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
2435 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
2436 q_vector->rxr_count = 0;
2437 q_vector->txr_count = 0;
2442 * ixgbe_request_irq - initialize interrupts
2443 * @adapter: board private structure
2445 * Attempts to configure interrupts using the best available
2446 * capabilities of the hardware and kernel.
2448 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2450 struct net_device *netdev = adapter->netdev;
2453 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2454 err = ixgbe_request_msix_irqs(adapter);
2455 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
2456 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2457 netdev->name, netdev);
2459 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2460 netdev->name, netdev);
2464 e_err(probe, "request_irq failed, Error %d\n", err);
2469 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2471 struct net_device *netdev = adapter->netdev;
2473 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2476 q_vectors = adapter->num_msix_vectors;
2479 free_irq(adapter->msix_entries[i].vector, netdev);
2482 for (; i >= 0; i--) {
2483 free_irq(adapter->msix_entries[i].vector,
2484 adapter->q_vector[i]);
2487 ixgbe_reset_q_vectors(adapter);
2489 free_irq(adapter->pdev->irq, netdev);
2494 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2495 * @adapter: board private structure
2497 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2499 switch (adapter->hw.mac.type) {
2500 case ixgbe_mac_82598EB:
2501 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2503 case ixgbe_mac_82599EB:
2504 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2505 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2506 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2507 if (adapter->num_vfs > 32)
2508 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
2513 IXGBE_WRITE_FLUSH(&adapter->hw);
2514 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2516 for (i = 0; i < adapter->num_msix_vectors; i++)
2517 synchronize_irq(adapter->msix_entries[i].vector);
2519 synchronize_irq(adapter->pdev->irq);
2524 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2527 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2529 struct ixgbe_hw *hw = &adapter->hw;
2531 IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
2532 EITR_INTS_PER_SEC_TO_REG(adapter->rx_eitr_param));
2534 ixgbe_set_ivar(adapter, 0, 0, 0);
2535 ixgbe_set_ivar(adapter, 1, 0, 0);
2537 map_vector_to_rxq(adapter, 0, 0);
2538 map_vector_to_txq(adapter, 0, 0);
2540 e_info(hw, "Legacy interrupt IVAR setup done\n");
2544 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2545 * @adapter: board private structure
2546 * @ring: structure containing ring specific data
2548 * Configure the Tx descriptor ring after a reset.
2550 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2551 struct ixgbe_ring *ring)
2553 struct ixgbe_hw *hw = &adapter->hw;
2554 u64 tdba = ring->dma;
2557 u8 reg_idx = ring->reg_idx;
2559 /* disable queue to avoid issues while updating state */
2560 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2561 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx),
2562 txdctl & ~IXGBE_TXDCTL_ENABLE);
2563 IXGBE_WRITE_FLUSH(hw);
2565 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2566 (tdba & DMA_BIT_MASK(32)));
2567 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2568 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2569 ring->count * sizeof(union ixgbe_adv_tx_desc));
2570 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2571 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2572 ring->tail = hw->hw_addr + IXGBE_TDT(reg_idx);
2574 /* configure fetching thresholds */
2575 if (adapter->rx_itr_setting == 0) {
2576 /* cannot set wthresh when itr==0 */
2577 txdctl &= ~0x007F0000;
2579 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
2580 txdctl |= (8 << 16);
2582 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2583 /* PThresh workaround for Tx hang with DFP enabled. */
2587 /* reinitialize flowdirector state */
2588 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2589 adapter->atr_sample_rate) {
2590 ring->atr_sample_rate = adapter->atr_sample_rate;
2591 ring->atr_count = 0;
2592 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
2594 ring->atr_sample_rate = 0;
2598 txdctl |= IXGBE_TXDCTL_ENABLE;
2599 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
2601 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2602 if (hw->mac.type == ixgbe_mac_82598EB &&
2603 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2606 /* poll to verify queue is enabled */
2609 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2610 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
2612 e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
2615 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
2617 struct ixgbe_hw *hw = &adapter->hw;
2621 if (hw->mac.type == ixgbe_mac_82598EB)
2624 /* disable the arbiter while setting MTQC */
2625 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2626 rttdcs |= IXGBE_RTTDCS_ARBDIS;
2627 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2629 /* set transmit pool layout */
2630 mask = (IXGBE_FLAG_SRIOV_ENABLED | IXGBE_FLAG_DCB_ENABLED);
2631 switch (adapter->flags & mask) {
2633 case (IXGBE_FLAG_SRIOV_ENABLED):
2634 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2635 (IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF));
2638 case (IXGBE_FLAG_DCB_ENABLED):
2639 /* We enable 8 traffic classes, DCB only */
2640 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2641 (IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ));
2645 IXGBE_WRITE_REG(hw, IXGBE_MTQC, IXGBE_MTQC_64Q_1PB);
2649 /* re-enable the arbiter */
2650 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2651 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2655 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2656 * @adapter: board private structure
2658 * Configure the Tx unit of the MAC after a reset.
2660 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2662 struct ixgbe_hw *hw = &adapter->hw;
2666 ixgbe_setup_mtqc(adapter);
2668 if (hw->mac.type != ixgbe_mac_82598EB) {
2669 /* DMATXCTL.EN must be before Tx queues are enabled */
2670 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2671 dmatxctl |= IXGBE_DMATXCTL_TE;
2672 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2675 /* Setup the HW Tx Head and Tail descriptor pointers */
2676 for (i = 0; i < adapter->num_tx_queues; i++)
2677 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
2680 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2682 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2683 struct ixgbe_ring *rx_ring)
2686 u8 reg_idx = rx_ring->reg_idx;
2688 switch (adapter->hw.mac.type) {
2689 case ixgbe_mac_82598EB: {
2690 struct ixgbe_ring_feature *feature = adapter->ring_feature;
2691 const int mask = feature[RING_F_RSS].mask;
2692 reg_idx = reg_idx & mask;
2695 case ixgbe_mac_82599EB:
2700 srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx));
2702 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
2703 srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
2704 if (adapter->num_vfs)
2705 srrctl |= IXGBE_SRRCTL_DROP_EN;
2707 srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
2708 IXGBE_SRRCTL_BSIZEHDR_MASK;
2710 if (ring_is_ps_enabled(rx_ring)) {
2711 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
2712 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2714 srrctl |= (PAGE_SIZE / 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2716 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2718 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
2719 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2720 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2723 IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx), srrctl);
2726 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2728 struct ixgbe_hw *hw = &adapter->hw;
2729 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2730 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2731 0x6A3E67EA, 0x14364D17, 0x3BED200D};
2732 u32 mrqc = 0, reta = 0;
2737 /* Fill out hash function seeds */
2738 for (i = 0; i < 10; i++)
2739 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2741 /* Fill out redirection table */
2742 for (i = 0, j = 0; i < 128; i++, j++) {
2743 if (j == adapter->ring_feature[RING_F_RSS].indices)
2745 /* reta = 4-byte sliding window of
2746 * 0x00..(indices-1)(indices-1)00..etc. */
2747 reta = (reta << 8) | (j * 0x11);
2749 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2752 /* Disable indicating checksum in descriptor, enables RSS hash */
2753 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2754 rxcsum |= IXGBE_RXCSUM_PCSD;
2755 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2757 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
2758 mask = adapter->flags & IXGBE_FLAG_RSS_ENABLED;
2760 mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2761 #ifdef CONFIG_IXGBE_DCB
2762 | IXGBE_FLAG_DCB_ENABLED
2764 | IXGBE_FLAG_SRIOV_ENABLED
2768 case (IXGBE_FLAG_RSS_ENABLED):
2769 mrqc = IXGBE_MRQC_RSSEN;
2771 case (IXGBE_FLAG_SRIOV_ENABLED):
2772 mrqc = IXGBE_MRQC_VMDQEN;
2774 #ifdef CONFIG_IXGBE_DCB
2775 case (IXGBE_FLAG_DCB_ENABLED):
2776 mrqc = IXGBE_MRQC_RT8TCEN;
2778 #endif /* CONFIG_IXGBE_DCB */
2783 /* Perform hash on these packet types */
2784 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2785 | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2786 | IXGBE_MRQC_RSS_FIELD_IPV6
2787 | IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
2789 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2793 * ixgbe_configure_rscctl - enable RSC for the indicated ring
2794 * @adapter: address of board private structure
2795 * @index: index of ring to set
2797 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
2798 struct ixgbe_ring *ring)
2800 struct ixgbe_hw *hw = &adapter->hw;
2803 u8 reg_idx = ring->reg_idx;
2805 if (!ring_is_rsc_enabled(ring))
2808 rx_buf_len = ring->rx_buf_len;
2809 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
2810 rscctrl |= IXGBE_RSCCTL_RSCEN;
2812 * we must limit the number of descriptors so that the
2813 * total size of max desc * buf_len is not greater
2816 if (ring_is_ps_enabled(ring)) {
2817 #if (MAX_SKB_FRAGS > 16)
2818 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2819 #elif (MAX_SKB_FRAGS > 8)
2820 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2821 #elif (MAX_SKB_FRAGS > 4)
2822 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2824 rscctrl |= IXGBE_RSCCTL_MAXDESC_1;
2827 if (rx_buf_len < IXGBE_RXBUFFER_4096)
2828 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2829 else if (rx_buf_len < IXGBE_RXBUFFER_8192)
2830 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2832 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2834 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
2838 * ixgbe_set_uta - Set unicast filter table address
2839 * @adapter: board private structure
2841 * The unicast table address is a register array of 32-bit registers.
2842 * The table is meant to be used in a way similar to how the MTA is used
2843 * however due to certain limitations in the hardware it is necessary to
2844 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
2845 * enable bit to allow vlan tag stripping when promiscuous mode is enabled
2847 static void ixgbe_set_uta(struct ixgbe_adapter *adapter)
2849 struct ixgbe_hw *hw = &adapter->hw;
2852 /* The UTA table only exists on 82599 hardware and newer */
2853 if (hw->mac.type < ixgbe_mac_82599EB)
2856 /* we only need to do this if VMDq is enabled */
2857 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
2860 for (i = 0; i < 128; i++)
2861 IXGBE_WRITE_REG(hw, IXGBE_UTA(i), ~0);
2864 #define IXGBE_MAX_RX_DESC_POLL 10
2865 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
2866 struct ixgbe_ring *ring)
2868 struct ixgbe_hw *hw = &adapter->hw;
2869 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
2871 u8 reg_idx = ring->reg_idx;
2873 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2874 if (hw->mac.type == ixgbe_mac_82598EB &&
2875 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2880 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2881 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
2884 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
2885 "the polling period\n", reg_idx);
2889 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
2890 struct ixgbe_ring *ring)
2892 struct ixgbe_hw *hw = &adapter->hw;
2893 u64 rdba = ring->dma;
2895 u8 reg_idx = ring->reg_idx;
2897 /* disable queue to avoid issues while updating state */
2898 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2899 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx),
2900 rxdctl & ~IXGBE_RXDCTL_ENABLE);
2901 IXGBE_WRITE_FLUSH(hw);
2903 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
2904 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
2905 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
2906 ring->count * sizeof(union ixgbe_adv_rx_desc));
2907 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
2908 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
2909 ring->tail = hw->hw_addr + IXGBE_RDT(reg_idx);
2911 ixgbe_configure_srrctl(adapter, ring);
2912 ixgbe_configure_rscctl(adapter, ring);
2914 if (hw->mac.type == ixgbe_mac_82598EB) {
2916 * enable cache line friendly hardware writes:
2917 * PTHRESH=32 descriptors (half the internal cache),
2918 * this also removes ugly rx_no_buffer_count increment
2919 * HTHRESH=4 descriptors (to minimize latency on fetch)
2920 * WTHRESH=8 burst writeback up to two cache lines
2922 rxdctl &= ~0x3FFFFF;
2926 /* enable receive descriptor ring */
2927 rxdctl |= IXGBE_RXDCTL_ENABLE;
2928 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
2930 ixgbe_rx_desc_queue_enable(adapter, ring);
2931 ixgbe_alloc_rx_buffers(ring, IXGBE_DESC_UNUSED(ring));
2934 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
2936 struct ixgbe_hw *hw = &adapter->hw;
2939 /* PSRTYPE must be initialized in non 82598 adapters */
2940 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
2941 IXGBE_PSRTYPE_UDPHDR |
2942 IXGBE_PSRTYPE_IPV4HDR |
2943 IXGBE_PSRTYPE_L2HDR |
2944 IXGBE_PSRTYPE_IPV6HDR;
2946 if (hw->mac.type == ixgbe_mac_82598EB)
2949 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED)
2950 psrtype |= (adapter->num_rx_queues_per_pool << 29);
2952 for (p = 0; p < adapter->num_rx_pools; p++)
2953 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(adapter->num_vfs + p),
2957 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
2959 struct ixgbe_hw *hw = &adapter->hw;
2962 u32 reg_offset, vf_shift;
2965 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
2968 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
2969 vt_reg_bits = IXGBE_VMD_CTL_VMDQ_EN | IXGBE_VT_CTL_REPLEN;
2970 vt_reg_bits |= (adapter->num_vfs << IXGBE_VT_CTL_POOL_SHIFT);
2971 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl | vt_reg_bits);
2973 vf_shift = adapter->num_vfs % 32;
2974 reg_offset = (adapter->num_vfs > 32) ? 1 : 0;
2976 /* Enable only the PF's pool for Tx/Rx */
2977 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (1 << vf_shift));
2978 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), 0);
2979 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (1 << vf_shift));
2980 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), 0);
2981 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
2983 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
2984 hw->mac.ops.set_vmdq(hw, 0, adapter->num_vfs);
2987 * Set up VF register offsets for selected VT Mode,
2988 * i.e. 32 or 64 VFs for SR-IOV
2990 gcr_ext = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
2991 gcr_ext |= IXGBE_GCR_EXT_MSIX_EN;
2992 gcr_ext |= IXGBE_GCR_EXT_VT_MODE_64;
2993 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
2995 /* enable Tx loopback for VF/PF communication */
2996 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
2999 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
3001 struct ixgbe_hw *hw = &adapter->hw;
3002 struct net_device *netdev = adapter->netdev;
3003 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3005 struct ixgbe_ring *rx_ring;
3009 /* Decide whether to use packet split mode or not */
3010 /* Do not use packet split if we're in SR-IOV Mode */
3011 if (!adapter->num_vfs)
3012 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
3014 /* Set the RX buffer length according to the mode */
3015 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
3016 rx_buf_len = IXGBE_RX_HDR_SIZE;
3018 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
3019 (netdev->mtu <= ETH_DATA_LEN))
3020 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3022 rx_buf_len = ALIGN(max_frame + VLAN_HLEN, 1024);
3026 /* adjust max frame to be able to do baby jumbo for FCoE */
3027 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
3028 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
3029 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3031 #endif /* IXGBE_FCOE */
3032 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
3033 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
3034 mhadd &= ~IXGBE_MHADD_MFS_MASK;
3035 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
3037 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
3040 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
3041 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
3042 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
3043 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
3046 * Setup the HW Rx Head and Tail Descriptor Pointers and
3047 * the Base and Length of the Rx Descriptor Ring
3049 for (i = 0; i < adapter->num_rx_queues; i++) {
3050 rx_ring = adapter->rx_ring[i];
3051 rx_ring->rx_buf_len = rx_buf_len;
3053 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)
3054 set_ring_ps_enabled(rx_ring);
3056 clear_ring_ps_enabled(rx_ring);
3058 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
3059 set_ring_rsc_enabled(rx_ring);
3061 clear_ring_rsc_enabled(rx_ring);
3064 if (netdev->features & NETIF_F_FCOE_MTU) {
3065 struct ixgbe_ring_feature *f;
3066 f = &adapter->ring_feature[RING_F_FCOE];
3067 if ((i >= f->mask) && (i < f->mask + f->indices)) {
3068 clear_ring_ps_enabled(rx_ring);
3069 if (rx_buf_len < IXGBE_FCOE_JUMBO_FRAME_SIZE)
3070 rx_ring->rx_buf_len =
3071 IXGBE_FCOE_JUMBO_FRAME_SIZE;
3072 } else if (!ring_is_rsc_enabled(rx_ring) &&
3073 !ring_is_ps_enabled(rx_ring)) {
3074 rx_ring->rx_buf_len =
3075 IXGBE_FCOE_JUMBO_FRAME_SIZE;
3078 #endif /* IXGBE_FCOE */
3082 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
3084 struct ixgbe_hw *hw = &adapter->hw;
3085 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
3087 switch (hw->mac.type) {
3088 case ixgbe_mac_82598EB:
3090 * For VMDq support of different descriptor types or
3091 * buffer sizes through the use of multiple SRRCTL
3092 * registers, RDRXCTL.MVMEN must be set to 1
3094 * also, the manual doesn't mention it clearly but DCA hints
3095 * will only use queue 0's tags unless this bit is set. Side
3096 * effects of setting this bit are only that SRRCTL must be
3097 * fully programmed [0..15]
3099 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
3101 case ixgbe_mac_82599EB:
3102 /* Disable RSC for ACK packets */
3103 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
3104 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
3105 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3106 /* hardware requires some bits to be set by default */
3107 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3108 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3111 /* We should do nothing since we don't know this hardware */
3115 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3119 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3120 * @adapter: board private structure
3122 * Configure the Rx unit of the MAC after a reset.
3124 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3126 struct ixgbe_hw *hw = &adapter->hw;
3130 /* disable receives while setting up the descriptors */
3131 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3132 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3134 ixgbe_setup_psrtype(adapter);
3135 ixgbe_setup_rdrxctl(adapter);
3137 /* Program registers for the distribution of queues */
3138 ixgbe_setup_mrqc(adapter);
3140 ixgbe_set_uta(adapter);
3142 /* set_rx_buffer_len must be called before ring initialization */
3143 ixgbe_set_rx_buffer_len(adapter);
3146 * Setup the HW Rx Head and Tail Descriptor Pointers and
3147 * the Base and Length of the Rx Descriptor Ring
3149 for (i = 0; i < adapter->num_rx_queues; i++)
3150 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3152 /* disable drop enable for 82598 parts */
3153 if (hw->mac.type == ixgbe_mac_82598EB)
3154 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3156 /* enable all receives */
3157 rxctrl |= IXGBE_RXCTRL_RXEN;
3158 hw->mac.ops.enable_rx_dma(hw, rxctrl);
3161 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3163 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3164 struct ixgbe_hw *hw = &adapter->hw;
3165 int pool_ndx = adapter->num_vfs;
3167 /* add VID to filter table */
3168 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, true);
3169 set_bit(vid, adapter->active_vlans);
3172 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3174 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3175 struct ixgbe_hw *hw = &adapter->hw;
3176 int pool_ndx = adapter->num_vfs;
3178 /* remove VID from filter table */
3179 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, false);
3180 clear_bit(vid, adapter->active_vlans);
3184 * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3185 * @adapter: driver data
3187 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
3189 struct ixgbe_hw *hw = &adapter->hw;
3192 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3193 vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
3194 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3198 * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3199 * @adapter: driver data
3201 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
3203 struct ixgbe_hw *hw = &adapter->hw;
3206 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3207 vlnctrl |= IXGBE_VLNCTRL_VFE;
3208 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3209 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3213 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3214 * @adapter: driver data
3216 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3218 struct ixgbe_hw *hw = &adapter->hw;
3222 switch (hw->mac.type) {
3223 case ixgbe_mac_82598EB:
3224 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3225 vlnctrl &= ~IXGBE_VLNCTRL_VME;
3226 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3228 case ixgbe_mac_82599EB:
3229 for (i = 0; i < adapter->num_rx_queues; i++) {
3230 j = adapter->rx_ring[i]->reg_idx;
3231 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3232 vlnctrl &= ~IXGBE_RXDCTL_VME;
3233 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3242 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3243 * @adapter: driver data
3245 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3247 struct ixgbe_hw *hw = &adapter->hw;
3251 switch (hw->mac.type) {
3252 case ixgbe_mac_82598EB:
3253 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3254 vlnctrl |= IXGBE_VLNCTRL_VME;
3255 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3257 case ixgbe_mac_82599EB:
3258 for (i = 0; i < adapter->num_rx_queues; i++) {
3259 j = adapter->rx_ring[i]->reg_idx;
3260 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3261 vlnctrl |= IXGBE_RXDCTL_VME;
3262 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3270 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3274 ixgbe_vlan_rx_add_vid(adapter->netdev, 0);
3276 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3277 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
3281 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3282 * @netdev: network interface device structure
3284 * Writes unicast address list to the RAR table.
3285 * Returns: -ENOMEM on failure/insufficient address space
3286 * 0 on no addresses written
3287 * X on writing X addresses to the RAR table
3289 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3291 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3292 struct ixgbe_hw *hw = &adapter->hw;
3293 unsigned int vfn = adapter->num_vfs;
3294 unsigned int rar_entries = hw->mac.num_rar_entries - (vfn + 1);
3297 /* return ENOMEM indicating insufficient memory for addresses */
3298 if (netdev_uc_count(netdev) > rar_entries)
3301 if (!netdev_uc_empty(netdev) && rar_entries) {
3302 struct netdev_hw_addr *ha;
3303 /* return error if we do not support writing to RAR table */
3304 if (!hw->mac.ops.set_rar)
3307 netdev_for_each_uc_addr(ha, netdev) {
3310 hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3315 /* write the addresses in reverse order to avoid write combining */
3316 for (; rar_entries > 0 ; rar_entries--)
3317 hw->mac.ops.clear_rar(hw, rar_entries);
3323 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3324 * @netdev: network interface device structure
3326 * The set_rx_method entry point is called whenever the unicast/multicast
3327 * address list or the network interface flags are updated. This routine is
3328 * responsible for configuring the hardware for proper unicast, multicast and
3331 void ixgbe_set_rx_mode(struct net_device *netdev)
3333 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3334 struct ixgbe_hw *hw = &adapter->hw;
3335 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3338 /* Check for Promiscuous and All Multicast modes */
3340 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3342 /* set all bits that we expect to always be set */
3343 fctrl |= IXGBE_FCTRL_BAM;
3344 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3345 fctrl |= IXGBE_FCTRL_PMCF;
3347 /* clear the bits we are changing the status of */
3348 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3350 if (netdev->flags & IFF_PROMISC) {
3351 hw->addr_ctrl.user_set_promisc = true;
3352 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3353 vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3354 /* don't hardware filter vlans in promisc mode */
3355 ixgbe_vlan_filter_disable(adapter);
3357 if (netdev->flags & IFF_ALLMULTI) {
3358 fctrl |= IXGBE_FCTRL_MPE;
3359 vmolr |= IXGBE_VMOLR_MPE;
3362 * Write addresses to the MTA, if the attempt fails
3363 * then we should just turn on promiscous mode so
3364 * that we can at least receive multicast traffic
3366 hw->mac.ops.update_mc_addr_list(hw, netdev);
3367 vmolr |= IXGBE_VMOLR_ROMPE;
3369 ixgbe_vlan_filter_enable(adapter);
3370 hw->addr_ctrl.user_set_promisc = false;
3372 * Write addresses to available RAR registers, if there is not
3373 * sufficient space to store all the addresses then enable
3374 * unicast promiscous mode
3376 count = ixgbe_write_uc_addr_list(netdev);
3378 fctrl |= IXGBE_FCTRL_UPE;
3379 vmolr |= IXGBE_VMOLR_ROPE;
3383 if (adapter->num_vfs) {
3384 ixgbe_restore_vf_multicasts(adapter);
3385 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(adapter->num_vfs)) &
3386 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3388 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(adapter->num_vfs), vmolr);
3391 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3393 if (netdev->features & NETIF_F_HW_VLAN_RX)
3394 ixgbe_vlan_strip_enable(adapter);
3396 ixgbe_vlan_strip_disable(adapter);
3399 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3402 struct ixgbe_q_vector *q_vector;
3403 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3405 /* legacy and MSI only use one vector */
3406 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3409 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3410 struct napi_struct *napi;
3411 q_vector = adapter->q_vector[q_idx];
3412 napi = &q_vector->napi;
3413 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3414 if (!q_vector->rxr_count || !q_vector->txr_count) {
3415 if (q_vector->txr_count == 1)
3416 napi->poll = &ixgbe_clean_txonly;
3417 else if (q_vector->rxr_count == 1)
3418 napi->poll = &ixgbe_clean_rxonly;
3426 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3429 struct ixgbe_q_vector *q_vector;
3430 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3432 /* legacy and MSI only use one vector */
3433 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3436 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3437 q_vector = adapter->q_vector[q_idx];
3438 napi_disable(&q_vector->napi);
3442 #ifdef CONFIG_IXGBE_DCB
3444 * ixgbe_configure_dcb - Configure DCB hardware
3445 * @adapter: ixgbe adapter struct
3447 * This is called by the driver on open to configure the DCB hardware.
3448 * This is also called by the gennetlink interface when reconfiguring
3451 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3453 struct ixgbe_hw *hw = &adapter->hw;
3454 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3456 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3457 if (hw->mac.type == ixgbe_mac_82598EB)
3458 netif_set_gso_max_size(adapter->netdev, 65536);
3462 if (hw->mac.type == ixgbe_mac_82598EB)
3463 netif_set_gso_max_size(adapter->netdev, 32768);
3466 if (adapter->netdev->features & NETIF_F_FCOE_MTU)
3467 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
3470 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3472 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3475 /* Enable VLAN tag insert/strip */
3476 adapter->netdev->features |= NETIF_F_HW_VLAN_RX;
3478 hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
3480 /* reconfigure the hardware */
3481 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
3485 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3487 struct net_device *netdev = adapter->netdev;
3488 struct ixgbe_hw *hw = &adapter->hw;
3491 #ifdef CONFIG_IXGBE_DCB
3492 ixgbe_configure_dcb(adapter);
3495 ixgbe_set_rx_mode(netdev);
3496 ixgbe_restore_vlan(adapter);
3499 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
3500 ixgbe_configure_fcoe(adapter);
3502 #endif /* IXGBE_FCOE */
3503 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3504 for (i = 0; i < adapter->num_tx_queues; i++)
3505 adapter->tx_ring[i]->atr_sample_rate =
3506 adapter->atr_sample_rate;
3507 ixgbe_init_fdir_signature_82599(hw, adapter->fdir_pballoc);
3508 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3509 ixgbe_init_fdir_perfect_82599(hw, adapter->fdir_pballoc);
3511 ixgbe_configure_virtualization(adapter);
3513 ixgbe_configure_tx(adapter);
3514 ixgbe_configure_rx(adapter);
3517 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3519 switch (hw->phy.type) {
3520 case ixgbe_phy_sfp_avago:
3521 case ixgbe_phy_sfp_ftl:
3522 case ixgbe_phy_sfp_intel:
3523 case ixgbe_phy_sfp_unknown:
3524 case ixgbe_phy_sfp_passive_tyco:
3525 case ixgbe_phy_sfp_passive_unknown:
3526 case ixgbe_phy_sfp_active_unknown:
3527 case ixgbe_phy_sfp_ftl_active:
3535 * ixgbe_sfp_link_config - set up SFP+ link
3536 * @adapter: pointer to private adapter struct
3538 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3540 struct ixgbe_hw *hw = &adapter->hw;
3542 if (hw->phy.multispeed_fiber) {
3544 * In multispeed fiber setups, the device may not have
3545 * had a physical connection when the driver loaded.
3546 * If that's the case, the initial link configuration
3547 * couldn't get the MAC into 10G or 1G mode, so we'll
3548 * never have a link status change interrupt fire.
3549 * We need to try and force an autonegotiation
3550 * session, then bring up link.
3552 hw->mac.ops.setup_sfp(hw);
3553 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
3554 schedule_work(&adapter->multispeed_fiber_task);
3557 * Direct Attach Cu and non-multispeed fiber modules
3558 * still need to be configured properly prior to
3561 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_MOD_TASK))
3562 schedule_work(&adapter->sfp_config_module_task);
3567 * ixgbe_non_sfp_link_config - set up non-SFP+ link
3568 * @hw: pointer to private hardware struct
3570 * Returns 0 on success, negative on failure
3572 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3575 bool negotiation, link_up = false;
3576 u32 ret = IXGBE_ERR_LINK_SETUP;
3578 if (hw->mac.ops.check_link)
3579 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3584 if (hw->mac.ops.get_link_capabilities)
3585 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
3590 if (hw->mac.ops.setup_link)
3591 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
3596 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
3598 struct ixgbe_hw *hw = &adapter->hw;
3601 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3602 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
3604 gpie |= IXGBE_GPIE_EIAME;
3606 * use EIAM to auto-mask when MSI-X interrupt is asserted
3607 * this saves a register write for every interrupt
3609 switch (hw->mac.type) {
3610 case ixgbe_mac_82598EB:
3611 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3614 case ixgbe_mac_82599EB:
3615 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
3616 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
3620 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
3621 * specifically only auto mask tx and rx interrupts */
3622 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3625 /* XXX: to interrupt immediately for EICS writes, enable this */
3626 /* gpie |= IXGBE_GPIE_EIMEN; */
3628 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3629 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
3630 gpie |= IXGBE_GPIE_VTMODE_64;
3633 /* Enable fan failure interrupt */
3634 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3635 gpie |= IXGBE_SDP1_GPIEN;
3637 if (hw->mac.type == ixgbe_mac_82599EB)
3638 gpie |= IXGBE_SDP1_GPIEN;
3639 gpie |= IXGBE_SDP2_GPIEN;
3641 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3644 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
3646 struct ixgbe_hw *hw = &adapter->hw;
3650 ixgbe_get_hw_control(adapter);
3651 ixgbe_setup_gpie(adapter);
3653 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3654 ixgbe_configure_msix(adapter);
3656 ixgbe_configure_msi_and_legacy(adapter);
3658 /* enable the optics */
3659 if (hw->phy.multispeed_fiber)
3660 hw->mac.ops.enable_tx_laser(hw);
3662 clear_bit(__IXGBE_DOWN, &adapter->state);
3663 ixgbe_napi_enable_all(adapter);
3665 if (ixgbe_is_sfp(hw)) {
3666 ixgbe_sfp_link_config(adapter);
3668 err = ixgbe_non_sfp_link_config(hw);
3670 e_err(probe, "link_config FAILED %d\n", err);
3673 /* clear any pending interrupts, may auto mask */
3674 IXGBE_READ_REG(hw, IXGBE_EICR);
3675 ixgbe_irq_enable(adapter, true, true);
3678 * If this adapter has a fan, check to see if we had a failure
3679 * before we enabled the interrupt.
3681 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3682 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
3683 if (esdp & IXGBE_ESDP_SDP1)
3684 e_crit(drv, "Fan has stopped, replace the adapter\n");
3688 * For hot-pluggable SFP+ devices, a new SFP+ module may have
3689 * arrived before interrupts were enabled but after probe. Such
3690 * devices wouldn't have their type identified yet. We need to
3691 * kick off the SFP+ module setup first, then try to bring up link.
3692 * If we're not hot-pluggable SFP+, we just need to configure link
3695 if (hw->phy.type == ixgbe_phy_unknown)
3696 schedule_work(&adapter->sfp_config_module_task);
3698 /* enable transmits */
3699 netif_tx_start_all_queues(adapter->netdev);
3701 /* bring the link up in the watchdog, this could race with our first
3702 * link up interrupt but shouldn't be a problem */
3703 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3704 adapter->link_check_timeout = jiffies;
3705 mod_timer(&adapter->watchdog_timer, jiffies);
3707 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
3708 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
3709 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
3710 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
3715 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
3717 WARN_ON(in_interrupt());
3718 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
3720 ixgbe_down(adapter);
3722 * If SR-IOV enabled then wait a bit before bringing the adapter
3723 * back up to give the VFs time to respond to the reset. The
3724 * two second wait is based upon the watchdog timer cycle in
3727 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3730 clear_bit(__IXGBE_RESETTING, &adapter->state);
3733 int ixgbe_up(struct ixgbe_adapter *adapter)
3735 /* hardware has been reset, we need to reload some things */
3736 ixgbe_configure(adapter);
3738 return ixgbe_up_complete(adapter);
3741 void ixgbe_reset(struct ixgbe_adapter *adapter)
3743 struct ixgbe_hw *hw = &adapter->hw;
3746 err = hw->mac.ops.init_hw(hw);
3749 case IXGBE_ERR_SFP_NOT_PRESENT:
3751 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
3752 e_dev_err("master disable timed out\n");
3754 case IXGBE_ERR_EEPROM_VERSION:
3755 /* We are running on a pre-production device, log a warning */
3756 e_dev_warn("This device is a pre-production adapter/LOM. "
3757 "Please be aware there may be issuesassociated with "
3758 "your hardware. If you are experiencing problems "
3759 "please contact your Intel or hardware "
3760 "representative who provided you with this "
3764 e_dev_err("Hardware Error: %d\n", err);
3767 /* reprogram the RAR[0] in case user changed it. */
3768 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
3773 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
3774 * @rx_ring: ring to free buffers from
3776 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
3778 struct device *dev = rx_ring->dev;
3782 /* ring already cleared, nothing to do */
3783 if (!rx_ring->rx_buffer_info)
3786 /* Free all the Rx ring sk_buffs */
3787 for (i = 0; i < rx_ring->count; i++) {
3788 struct ixgbe_rx_buffer *rx_buffer_info;
3790 rx_buffer_info = &rx_ring->rx_buffer_info[i];
3791 if (rx_buffer_info->dma) {
3792 dma_unmap_single(rx_ring->dev, rx_buffer_info->dma,
3793 rx_ring->rx_buf_len,
3795 rx_buffer_info->dma = 0;
3797 if (rx_buffer_info->skb) {
3798 struct sk_buff *skb = rx_buffer_info->skb;
3799 rx_buffer_info->skb = NULL;
3801 struct sk_buff *this = skb;
3802 if (IXGBE_RSC_CB(this)->delay_unmap) {
3803 dma_unmap_single(dev,
3804 IXGBE_RSC_CB(this)->dma,
3805 rx_ring->rx_buf_len,
3807 IXGBE_RSC_CB(this)->dma = 0;
3808 IXGBE_RSC_CB(skb)->delay_unmap = false;
3811 dev_kfree_skb(this);
3814 if (!rx_buffer_info->page)
3816 if (rx_buffer_info->page_dma) {
3817 dma_unmap_page(dev, rx_buffer_info->page_dma,
3818 PAGE_SIZE / 2, DMA_FROM_DEVICE);
3819 rx_buffer_info->page_dma = 0;
3821 put_page(rx_buffer_info->page);
3822 rx_buffer_info->page = NULL;
3823 rx_buffer_info->page_offset = 0;
3826 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
3827 memset(rx_ring->rx_buffer_info, 0, size);
3829 /* Zero out the descriptor ring */
3830 memset(rx_ring->desc, 0, rx_ring->size);
3832 rx_ring->next_to_clean = 0;
3833 rx_ring->next_to_use = 0;
3837 * ixgbe_clean_tx_ring - Free Tx Buffers
3838 * @tx_ring: ring to be cleaned
3840 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
3842 struct ixgbe_tx_buffer *tx_buffer_info;
3846 /* ring already cleared, nothing to do */
3847 if (!tx_ring->tx_buffer_info)
3850 /* Free all the Tx ring sk_buffs */
3851 for (i = 0; i < tx_ring->count; i++) {
3852 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3853 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
3856 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
3857 memset(tx_ring->tx_buffer_info, 0, size);
3859 /* Zero out the descriptor ring */
3860 memset(tx_ring->desc, 0, tx_ring->size);
3862 tx_ring->next_to_use = 0;
3863 tx_ring->next_to_clean = 0;
3867 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
3868 * @adapter: board private structure
3870 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
3874 for (i = 0; i < adapter->num_rx_queues; i++)
3875 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
3879 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
3880 * @adapter: board private structure
3882 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
3886 for (i = 0; i < adapter->num_tx_queues; i++)
3887 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
3890 void ixgbe_down(struct ixgbe_adapter *adapter)
3892 struct net_device *netdev = adapter->netdev;
3893 struct ixgbe_hw *hw = &adapter->hw;
3897 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3899 /* signal that we are down to the interrupt handler */
3900 set_bit(__IXGBE_DOWN, &adapter->state);
3902 /* disable receive for all VFs and wait one second */
3903 if (adapter->num_vfs) {
3904 /* ping all the active vfs to let them know we are going down */
3905 ixgbe_ping_all_vfs(adapter);
3907 /* Disable all VFTE/VFRE TX/RX */
3908 ixgbe_disable_tx_rx(adapter);
3910 /* Mark all the VFs as inactive */
3911 for (i = 0 ; i < adapter->num_vfs; i++)
3912 adapter->vfinfo[i].clear_to_send = 0;
3915 /* disable receives */
3916 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3917 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3919 IXGBE_WRITE_FLUSH(hw);
3922 netif_tx_stop_all_queues(netdev);
3924 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
3925 del_timer_sync(&adapter->sfp_timer);
3926 del_timer_sync(&adapter->watchdog_timer);
3927 cancel_work_sync(&adapter->watchdog_task);
3929 netif_carrier_off(netdev);
3930 netif_tx_disable(netdev);
3932 ixgbe_irq_disable(adapter);
3934 ixgbe_napi_disable_all(adapter);
3936 /* Cleanup the affinity_hint CPU mask memory and callback */
3937 for (i = 0; i < num_q_vectors; i++) {
3938 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
3939 /* clear the affinity_mask in the IRQ descriptor */
3940 irq_set_affinity_hint(adapter->msix_entries[i]. vector, NULL);
3941 /* release the CPU mask memory */
3942 free_cpumask_var(q_vector->affinity_mask);
3945 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3946 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3947 cancel_work_sync(&adapter->fdir_reinit_task);
3949 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
3950 cancel_work_sync(&adapter->check_overtemp_task);
3952 /* disable transmits in the hardware now that interrupts are off */
3953 for (i = 0; i < adapter->num_tx_queues; i++) {
3954 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
3955 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
3956 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx),
3957 (txdctl & ~IXGBE_TXDCTL_ENABLE));
3959 /* Disable the Tx DMA engine on 82599 */
3960 switch (hw->mac.type) {
3961 case ixgbe_mac_82599EB:
3962 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
3963 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
3964 ~IXGBE_DMATXCTL_TE));
3970 /* power down the optics */
3971 if (hw->phy.multispeed_fiber)
3972 hw->mac.ops.disable_tx_laser(hw);
3974 /* clear n-tuple filters that are cached */
3975 ethtool_ntuple_flush(netdev);
3977 if (!pci_channel_offline(adapter->pdev))
3978 ixgbe_reset(adapter);
3979 ixgbe_clean_all_tx_rings(adapter);
3980 ixgbe_clean_all_rx_rings(adapter);
3982 #ifdef CONFIG_IXGBE_DCA
3983 /* since we reset the hardware DCA settings were cleared */
3984 ixgbe_setup_dca(adapter);
3989 * ixgbe_poll - NAPI Rx polling callback
3990 * @napi: structure for representing this polling device
3991 * @budget: how many packets driver is allowed to clean
3993 * This function is used for legacy and MSI, NAPI mode
3995 static int ixgbe_poll(struct napi_struct *napi, int budget)
3997 struct ixgbe_q_vector *q_vector =
3998 container_of(napi, struct ixgbe_q_vector, napi);
3999 struct ixgbe_adapter *adapter = q_vector->adapter;
4000 int tx_clean_complete, work_done = 0;
4002 #ifdef CONFIG_IXGBE_DCA
4003 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
4004 ixgbe_update_dca(q_vector);
4007 tx_clean_complete = ixgbe_clean_tx_irq(q_vector, adapter->tx_ring[0]);
4008 ixgbe_clean_rx_irq(q_vector, adapter->rx_ring[0], &work_done, budget);
4010 if (!tx_clean_complete)
4013 /* If budget not fully consumed, exit the polling mode */
4014 if (work_done < budget) {
4015 napi_complete(napi);
4016 if (adapter->rx_itr_setting & 1)
4017 ixgbe_set_itr(adapter);
4018 if (!test_bit(__IXGBE_DOWN, &adapter->state))
4019 ixgbe_irq_enable_queues(adapter, IXGBE_EIMS_RTX_QUEUE);
4025 * ixgbe_tx_timeout - Respond to a Tx Hang
4026 * @netdev: network interface device structure
4028 static void ixgbe_tx_timeout(struct net_device *netdev)
4030 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4032 /* Do the reset outside of interrupt context */
4033 schedule_work(&adapter->reset_task);
4036 static void ixgbe_reset_task(struct work_struct *work)
4038 struct ixgbe_adapter *adapter;
4039 adapter = container_of(work, struct ixgbe_adapter, reset_task);
4041 /* If we're already down or resetting, just bail */
4042 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
4043 test_bit(__IXGBE_RESETTING, &adapter->state))
4046 adapter->tx_timeout_count++;
4048 ixgbe_dump(adapter);
4049 netdev_err(adapter->netdev, "Reset adapter\n");
4050 ixgbe_reinit_locked(adapter);
4053 #ifdef CONFIG_IXGBE_DCB
4054 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
4057 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_DCB];
4059 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
4063 adapter->num_rx_queues = f->indices;
4064 adapter->num_tx_queues = f->indices;
4072 * ixgbe_set_rss_queues: Allocate queues for RSS
4073 * @adapter: board private structure to initialize
4075 * This is our "base" multiqueue mode. RSS (Receive Side Scaling) will try
4076 * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
4079 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
4082 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
4084 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4086 adapter->num_rx_queues = f->indices;
4087 adapter->num_tx_queues = f->indices;
4097 * ixgbe_set_fdir_queues: Allocate queues for Flow Director
4098 * @adapter: board private structure to initialize
4100 * Flow Director is an advanced Rx filter, attempting to get Rx flows back
4101 * to the original CPU that initiated the Tx session. This runs in addition
4102 * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
4103 * Rx load across CPUs using RSS.
4106 static inline bool ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
4109 struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
4111 f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
4114 /* Flow Director must have RSS enabled */
4115 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
4116 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
4117 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)))) {
4118 adapter->num_tx_queues = f_fdir->indices;
4119 adapter->num_rx_queues = f_fdir->indices;
4122 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4123 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4130 * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
4131 * @adapter: board private structure to initialize
4133 * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
4134 * The ring feature mask is not used as a mask for FCoE, as it can take any 8
4135 * rx queues out of the max number of rx queues, instead, it is used as the
4136 * index of the first rx queue used by FCoE.
4139 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
4142 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4144 f->indices = min((int)num_online_cpus(), f->indices);
4145 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
4146 adapter->num_rx_queues = 1;
4147 adapter->num_tx_queues = 1;
4148 #ifdef CONFIG_IXGBE_DCB
4149 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4150 e_info(probe, "FCoE enabled with DCB\n");
4151 ixgbe_set_dcb_queues(adapter);
4154 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4155 e_info(probe, "FCoE enabled with RSS\n");
4156 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4157 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
4158 ixgbe_set_fdir_queues(adapter);
4160 ixgbe_set_rss_queues(adapter);
4162 /* adding FCoE rx rings to the end */
4163 f->mask = adapter->num_rx_queues;
4164 adapter->num_rx_queues += f->indices;
4165 adapter->num_tx_queues += f->indices;
4173 #endif /* IXGBE_FCOE */
4175 * ixgbe_set_sriov_queues: Allocate queues for IOV use
4176 * @adapter: board private structure to initialize
4178 * IOV doesn't actually use anything, so just NAK the
4179 * request for now and let the other queue routines
4180 * figure out what to do.
4182 static inline bool ixgbe_set_sriov_queues(struct ixgbe_adapter *adapter)
4188 * ixgbe_set_num_queues: Allocate queues for device, feature dependant
4189 * @adapter: board private structure to initialize
4191 * This is the top level queue allocation routine. The order here is very
4192 * important, starting with the "most" number of features turned on at once,
4193 * and ending with the smallest set of features. This way large combinations
4194 * can be allocated if they're turned on, and smaller combinations are the
4195 * fallthrough conditions.
4198 static int ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
4200 /* Start with base case */
4201 adapter->num_rx_queues = 1;
4202 adapter->num_tx_queues = 1;
4203 adapter->num_rx_pools = adapter->num_rx_queues;
4204 adapter->num_rx_queues_per_pool = 1;
4206 if (ixgbe_set_sriov_queues(adapter))
4210 if (ixgbe_set_fcoe_queues(adapter))
4213 #endif /* IXGBE_FCOE */
4214 #ifdef CONFIG_IXGBE_DCB
4215 if (ixgbe_set_dcb_queues(adapter))
4219 if (ixgbe_set_fdir_queues(adapter))
4222 if (ixgbe_set_rss_queues(adapter))
4225 /* fallback to base case */
4226 adapter->num_rx_queues = 1;
4227 adapter->num_tx_queues = 1;
4230 /* Notify the stack of the (possibly) reduced queue counts. */
4231 netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
4232 return netif_set_real_num_rx_queues(adapter->netdev,
4233 adapter->num_rx_queues);
4236 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
4239 int err, vector_threshold;
4241 /* We'll want at least 3 (vector_threshold):
4244 * 3) Other (Link Status Change, etc.)
4245 * 4) TCP Timer (optional)
4247 vector_threshold = MIN_MSIX_COUNT;
4249 /* The more we get, the more we will assign to Tx/Rx Cleanup
4250 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
4251 * Right now, we simply care about how many we'll get; we'll
4252 * set them up later while requesting irq's.
4254 while (vectors >= vector_threshold) {
4255 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
4257 if (!err) /* Success in acquiring all requested vectors. */
4260 vectors = 0; /* Nasty failure, quit now */
4261 else /* err == number of vectors we should try again with */
4265 if (vectors < vector_threshold) {
4266 /* Can't allocate enough MSI-X interrupts? Oh well.
4267 * This just means we'll go with either a single MSI
4268 * vector or fall back to legacy interrupts.
4270 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4271 "Unable to allocate MSI-X interrupts\n");
4272 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4273 kfree(adapter->msix_entries);
4274 adapter->msix_entries = NULL;
4276 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
4278 * Adjust for only the vectors we'll use, which is minimum
4279 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
4280 * vectors we were allocated.
4282 adapter->num_msix_vectors = min(vectors,
4283 adapter->max_msix_q_vectors + NON_Q_VECTORS);
4288 * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
4289 * @adapter: board private structure to initialize
4291 * Cache the descriptor ring offsets for RSS to the assigned rings.
4294 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
4298 if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED))
4301 for (i = 0; i < adapter->num_rx_queues; i++)
4302 adapter->rx_ring[i]->reg_idx = i;
4303 for (i = 0; i < adapter->num_tx_queues; i++)
4304 adapter->tx_ring[i]->reg_idx = i;
4309 #ifdef CONFIG_IXGBE_DCB
4311 * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
4312 * @adapter: board private structure to initialize
4314 * Cache the descriptor ring offsets for DCB to the assigned rings.
4317 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
4321 int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
4323 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
4326 /* the number of queues is assumed to be symmetric */
4327 switch (adapter->hw.mac.type) {
4328 case ixgbe_mac_82598EB:
4329 for (i = 0; i < dcb_i; i++) {
4330 adapter->rx_ring[i]->reg_idx = i << 3;
4331 adapter->tx_ring[i]->reg_idx = i << 2;
4335 case ixgbe_mac_82599EB:
4338 * Tx TC0 starts at: descriptor queue 0
4339 * Tx TC1 starts at: descriptor queue 32
4340 * Tx TC2 starts at: descriptor queue 64
4341 * Tx TC3 starts at: descriptor queue 80
4342 * Tx TC4 starts at: descriptor queue 96
4343 * Tx TC5 starts at: descriptor queue 104
4344 * Tx TC6 starts at: descriptor queue 112
4345 * Tx TC7 starts at: descriptor queue 120
4347 * Rx TC0-TC7 are offset by 16 queues each
4349 for (i = 0; i < 3; i++) {
4350 adapter->tx_ring[i]->reg_idx = i << 5;
4351 adapter->rx_ring[i]->reg_idx = i << 4;
4353 for ( ; i < 5; i++) {
4354 adapter->tx_ring[i]->reg_idx = ((i + 2) << 4);
4355 adapter->rx_ring[i]->reg_idx = i << 4;
4357 for ( ; i < dcb_i; i++) {
4358 adapter->tx_ring[i]->reg_idx = ((i + 8) << 3);
4359 adapter->rx_ring[i]->reg_idx = i << 4;
4362 } else if (dcb_i == 4) {
4364 * Tx TC0 starts at: descriptor queue 0
4365 * Tx TC1 starts at: descriptor queue 64
4366 * Tx TC2 starts at: descriptor queue 96
4367 * Tx TC3 starts at: descriptor queue 112
4369 * Rx TC0-TC3 are offset by 32 queues each
4371 adapter->tx_ring[0]->reg_idx = 0;
4372 adapter->tx_ring[1]->reg_idx = 64;
4373 adapter->tx_ring[2]->reg_idx = 96;
4374 adapter->tx_ring[3]->reg_idx = 112;
4375 for (i = 0 ; i < dcb_i; i++)
4376 adapter->rx_ring[i]->reg_idx = i << 5;
4388 * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
4389 * @adapter: board private structure to initialize
4391 * Cache the descriptor ring offsets for Flow Director to the assigned rings.
4394 static inline bool ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
4399 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
4400 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4401 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))) {
4402 for (i = 0; i < adapter->num_rx_queues; i++)
4403 adapter->rx_ring[i]->reg_idx = i;
4404 for (i = 0; i < adapter->num_tx_queues; i++)
4405 adapter->tx_ring[i]->reg_idx = i;
4414 * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
4415 * @adapter: board private structure to initialize
4417 * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
4420 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
4422 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4424 u8 fcoe_rx_i = 0, fcoe_tx_i = 0;
4426 if (!(adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
4429 #ifdef CONFIG_IXGBE_DCB
4430 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4431 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
4433 ixgbe_cache_ring_dcb(adapter);
4434 /* find out queues in TC for FCoE */
4435 fcoe_rx_i = adapter->rx_ring[fcoe->tc]->reg_idx + 1;
4436 fcoe_tx_i = adapter->tx_ring[fcoe->tc]->reg_idx + 1;
4438 * In 82599, the number of Tx queues for each traffic
4439 * class for both 8-TC and 4-TC modes are:
4440 * TCs : TC0 TC1 TC2 TC3 TC4 TC5 TC6 TC7
4441 * 8 TCs: 32 32 16 16 8 8 8 8
4442 * 4 TCs: 64 64 32 32
4443 * We have max 8 queues for FCoE, where 8 the is
4444 * FCoE redirection table size. If TC for FCoE is
4445 * less than or equal to TC3, we have enough queues
4446 * to add max of 8 queues for FCoE, so we start FCoE
4447 * Tx queue from the next one, i.e., reg_idx + 1.
4448 * If TC for FCoE is above TC3, implying 8 TC mode,
4449 * and we need 8 for FCoE, we have to take all queues
4450 * in that traffic class for FCoE.
4452 if ((f->indices == IXGBE_FCRETA_SIZE) && (fcoe->tc > 3))
4455 #endif /* CONFIG_IXGBE_DCB */
4456 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4457 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4458 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
4459 ixgbe_cache_ring_fdir(adapter);
4461 ixgbe_cache_ring_rss(adapter);
4463 fcoe_rx_i = f->mask;
4464 fcoe_tx_i = f->mask;
4466 for (i = 0; i < f->indices; i++, fcoe_rx_i++, fcoe_tx_i++) {
4467 adapter->rx_ring[f->mask + i]->reg_idx = fcoe_rx_i;
4468 adapter->tx_ring[f->mask + i]->reg_idx = fcoe_tx_i;
4473 #endif /* IXGBE_FCOE */
4475 * ixgbe_cache_ring_sriov - Descriptor ring to register mapping for sriov
4476 * @adapter: board private structure to initialize
4478 * SR-IOV doesn't use any descriptor rings but changes the default if
4479 * no other mapping is used.
4482 static inline bool ixgbe_cache_ring_sriov(struct ixgbe_adapter *adapter)
4484 adapter->rx_ring[0]->reg_idx = adapter->num_vfs * 2;
4485 adapter->tx_ring[0]->reg_idx = adapter->num_vfs * 2;
4486 if (adapter->num_vfs)
4493 * ixgbe_cache_ring_register - Descriptor ring to register mapping
4494 * @adapter: board private structure to initialize
4496 * Once we know the feature-set enabled for the device, we'll cache
4497 * the register offset the descriptor ring is assigned to.
4499 * Note, the order the various feature calls is important. It must start with
4500 * the "most" features enabled at the same time, then trickle down to the
4501 * least amount of features turned on at once.
4503 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
4505 /* start with default case */
4506 adapter->rx_ring[0]->reg_idx = 0;
4507 adapter->tx_ring[0]->reg_idx = 0;
4509 if (ixgbe_cache_ring_sriov(adapter))
4513 if (ixgbe_cache_ring_fcoe(adapter))
4516 #endif /* IXGBE_FCOE */
4517 #ifdef CONFIG_IXGBE_DCB
4518 if (ixgbe_cache_ring_dcb(adapter))
4522 if (ixgbe_cache_ring_fdir(adapter))
4525 if (ixgbe_cache_ring_rss(adapter))
4530 * ixgbe_alloc_queues - Allocate memory for all rings
4531 * @adapter: board private structure to initialize
4533 * We allocate one ring per queue at run-time since we don't know the
4534 * number of queues at compile-time. The polling_netdev array is
4535 * intended for Multiqueue, but should work fine with a single queue.
4537 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
4541 int orig_node = adapter->node;
4543 for (i = 0; i < adapter->num_tx_queues; i++) {
4544 struct ixgbe_ring *ring = adapter->tx_ring[i];
4545 if (orig_node == -1) {
4546 int cur_node = next_online_node(adapter->node);
4547 if (cur_node == MAX_NUMNODES)
4548 cur_node = first_online_node;
4549 adapter->node = cur_node;
4551 ring = kzalloc_node(sizeof(struct ixgbe_ring), GFP_KERNEL,
4554 ring = kzalloc(sizeof(struct ixgbe_ring), GFP_KERNEL);
4556 goto err_tx_ring_allocation;
4557 ring->count = adapter->tx_ring_count;
4558 ring->queue_index = i;
4559 ring->dev = &adapter->pdev->dev;
4560 ring->netdev = adapter->netdev;
4561 ring->numa_node = adapter->node;
4563 adapter->tx_ring[i] = ring;
4566 /* Restore the adapter's original node */
4567 adapter->node = orig_node;
4569 rx_count = adapter->rx_ring_count;
4570 for (i = 0; i < adapter->num_rx_queues; i++) {
4571 struct ixgbe_ring *ring = adapter->rx_ring[i];
4572 if (orig_node == -1) {
4573 int cur_node = next_online_node(adapter->node);
4574 if (cur_node == MAX_NUMNODES)
4575 cur_node = first_online_node;
4576 adapter->node = cur_node;
4578 ring = kzalloc_node(sizeof(struct ixgbe_ring), GFP_KERNEL,
4581 ring = kzalloc(sizeof(struct ixgbe_ring), GFP_KERNEL);
4583 goto err_rx_ring_allocation;
4584 ring->count = rx_count;
4585 ring->queue_index = i;
4586 ring->dev = &adapter->pdev->dev;
4587 ring->netdev = adapter->netdev;
4588 ring->numa_node = adapter->node;
4590 adapter->rx_ring[i] = ring;
4593 /* Restore the adapter's original node */
4594 adapter->node = orig_node;
4596 ixgbe_cache_ring_register(adapter);
4600 err_rx_ring_allocation:
4601 for (i = 0; i < adapter->num_tx_queues; i++)
4602 kfree(adapter->tx_ring[i]);
4603 err_tx_ring_allocation:
4608 * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
4609 * @adapter: board private structure to initialize
4611 * Attempt to configure the interrupts using the best available
4612 * capabilities of the hardware and the kernel.
4614 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
4616 struct ixgbe_hw *hw = &adapter->hw;
4618 int vector, v_budget;
4621 * It's easy to be greedy for MSI-X vectors, but it really
4622 * doesn't do us much good if we have a lot more vectors
4623 * than CPU's. So let's be conservative and only ask for
4624 * (roughly) the same number of vectors as there are CPU's.
4626 v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
4627 (int)num_online_cpus()) + NON_Q_VECTORS;
4630 * At the same time, hardware can only support a maximum of
4631 * hw.mac->max_msix_vectors vectors. With features
4632 * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
4633 * descriptor queues supported by our device. Thus, we cap it off in
4634 * those rare cases where the cpu count also exceeds our vector limit.
4636 v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
4638 /* A failure in MSI-X entry allocation isn't fatal, but it does
4639 * mean we disable MSI-X capabilities of the adapter. */
4640 adapter->msix_entries = kcalloc(v_budget,
4641 sizeof(struct msix_entry), GFP_KERNEL);
4642 if (adapter->msix_entries) {
4643 for (vector = 0; vector < v_budget; vector++)
4644 adapter->msix_entries[vector].entry = vector;
4646 ixgbe_acquire_msix_vectors(adapter, v_budget);
4648 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4652 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
4653 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
4654 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4655 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4656 adapter->atr_sample_rate = 0;
4657 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4658 ixgbe_disable_sriov(adapter);
4660 err = ixgbe_set_num_queues(adapter);
4664 err = pci_enable_msi(adapter->pdev);
4666 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
4668 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4669 "Unable to allocate MSI interrupt, "
4670 "falling back to legacy. Error: %d\n", err);
4680 * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
4681 * @adapter: board private structure to initialize
4683 * We allocate one q_vector per queue interrupt. If allocation fails we
4686 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
4688 int q_idx, num_q_vectors;
4689 struct ixgbe_q_vector *q_vector;
4691 int (*poll)(struct napi_struct *, int);
4693 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4694 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4695 napi_vectors = adapter->num_rx_queues;
4696 poll = &ixgbe_clean_rxtx_many;
4703 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4704 q_vector = kzalloc_node(sizeof(struct ixgbe_q_vector),
4705 GFP_KERNEL, adapter->node);
4707 q_vector = kzalloc(sizeof(struct ixgbe_q_vector),
4711 q_vector->adapter = adapter;
4712 if (q_vector->txr_count && !q_vector->rxr_count)
4713 q_vector->eitr = adapter->tx_eitr_param;
4715 q_vector->eitr = adapter->rx_eitr_param;
4716 q_vector->v_idx = q_idx;
4717 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
4718 adapter->q_vector[q_idx] = q_vector;
4726 q_vector = adapter->q_vector[q_idx];
4727 netif_napi_del(&q_vector->napi);
4729 adapter->q_vector[q_idx] = NULL;
4735 * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
4736 * @adapter: board private structure to initialize
4738 * This function frees the memory allocated to the q_vectors. In addition if
4739 * NAPI is enabled it will delete any references to the NAPI struct prior
4740 * to freeing the q_vector.
4742 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
4744 int q_idx, num_q_vectors;
4746 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4747 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4751 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4752 struct ixgbe_q_vector *q_vector = adapter->q_vector[q_idx];
4753 adapter->q_vector[q_idx] = NULL;
4754 netif_napi_del(&q_vector->napi);
4759 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
4761 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4762 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4763 pci_disable_msix(adapter->pdev);
4764 kfree(adapter->msix_entries);
4765 adapter->msix_entries = NULL;
4766 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
4767 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
4768 pci_disable_msi(adapter->pdev);
4773 * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
4774 * @adapter: board private structure to initialize
4776 * We determine which interrupt scheme to use based on...
4777 * - Kernel support (MSI, MSI-X)
4778 * - which can be user-defined (via MODULE_PARAM)
4779 * - Hardware queue count (num_*_queues)
4780 * - defined by miscellaneous hardware support/features (RSS, etc.)
4782 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
4786 /* Number of supported queues */
4787 err = ixgbe_set_num_queues(adapter);
4791 err = ixgbe_set_interrupt_capability(adapter);
4793 e_dev_err("Unable to setup interrupt capabilities\n");
4794 goto err_set_interrupt;
4797 err = ixgbe_alloc_q_vectors(adapter);
4799 e_dev_err("Unable to allocate memory for queue vectors\n");
4800 goto err_alloc_q_vectors;
4803 err = ixgbe_alloc_queues(adapter);
4805 e_dev_err("Unable to allocate memory for queues\n");
4806 goto err_alloc_queues;
4809 e_dev_info("Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
4810 (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
4811 adapter->num_rx_queues, adapter->num_tx_queues);
4813 set_bit(__IXGBE_DOWN, &adapter->state);
4818 ixgbe_free_q_vectors(adapter);
4819 err_alloc_q_vectors:
4820 ixgbe_reset_interrupt_capability(adapter);
4825 static void ring_free_rcu(struct rcu_head *head)
4827 kfree(container_of(head, struct ixgbe_ring, rcu));
4831 * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
4832 * @adapter: board private structure to clear interrupt scheme on
4834 * We go through and clear interrupt specific resources and reset the structure
4835 * to pre-load conditions
4837 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
4841 for (i = 0; i < adapter->num_tx_queues; i++) {
4842 kfree(adapter->tx_ring[i]);
4843 adapter->tx_ring[i] = NULL;
4845 for (i = 0; i < adapter->num_rx_queues; i++) {
4846 struct ixgbe_ring *ring = adapter->rx_ring[i];
4848 /* ixgbe_get_stats64() might access this ring, we must wait
4849 * a grace period before freeing it.
4851 call_rcu(&ring->rcu, ring_free_rcu);
4852 adapter->rx_ring[i] = NULL;
4855 ixgbe_free_q_vectors(adapter);
4856 ixgbe_reset_interrupt_capability(adapter);
4860 * ixgbe_sfp_timer - worker thread to find a missing module
4861 * @data: pointer to our adapter struct
4863 static void ixgbe_sfp_timer(unsigned long data)
4865 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
4868 * Do the sfp_timer outside of interrupt context due to the
4869 * delays that sfp+ detection requires
4871 schedule_work(&adapter->sfp_task);
4875 * ixgbe_sfp_task - worker thread to find a missing module
4876 * @work: pointer to work_struct containing our data
4878 static void ixgbe_sfp_task(struct work_struct *work)
4880 struct ixgbe_adapter *adapter = container_of(work,
4881 struct ixgbe_adapter,
4883 struct ixgbe_hw *hw = &adapter->hw;
4885 if ((hw->phy.type == ixgbe_phy_nl) &&
4886 (hw->phy.sfp_type == ixgbe_sfp_type_not_present)) {
4887 s32 ret = hw->phy.ops.identify_sfp(hw);
4888 if (ret == IXGBE_ERR_SFP_NOT_PRESENT)
4890 ret = hw->phy.ops.reset(hw);
4891 if (ret == IXGBE_ERR_SFP_NOT_SUPPORTED) {
4892 e_dev_err("failed to initialize because an unsupported "
4893 "SFP+ module type was detected.\n");
4894 e_dev_err("Reload the driver after installing a "
4895 "supported module.\n");
4896 unregister_netdev(adapter->netdev);
4898 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
4900 /* don't need this routine any more */
4901 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
4905 if (test_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state))
4906 mod_timer(&adapter->sfp_timer,
4907 round_jiffies(jiffies + (2 * HZ)));
4911 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4912 * @adapter: board private structure to initialize
4914 * ixgbe_sw_init initializes the Adapter private data structure.
4915 * Fields are initialized based on PCI device information and
4916 * OS network device settings (MTU size).
4918 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
4920 struct ixgbe_hw *hw = &adapter->hw;
4921 struct pci_dev *pdev = adapter->pdev;
4922 struct net_device *dev = adapter->netdev;
4924 #ifdef CONFIG_IXGBE_DCB
4926 struct tc_configuration *tc;
4928 int max_frame = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
4930 /* PCI config space info */
4932 hw->vendor_id = pdev->vendor;
4933 hw->device_id = pdev->device;
4934 hw->revision_id = pdev->revision;
4935 hw->subsystem_vendor_id = pdev->subsystem_vendor;
4936 hw->subsystem_device_id = pdev->subsystem_device;
4938 /* Set capability flags */
4939 rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
4940 adapter->ring_feature[RING_F_RSS].indices = rss;
4941 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
4942 adapter->ring_feature[RING_F_DCB].indices = IXGBE_MAX_DCB_INDICES;
4943 switch (hw->mac.type) {
4944 case ixgbe_mac_82598EB:
4945 if (hw->device_id == IXGBE_DEV_ID_82598AT)
4946 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4947 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
4949 case ixgbe_mac_82599EB:
4950 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
4951 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4952 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4953 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
4954 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4955 if (dev->features & NETIF_F_NTUPLE) {
4956 /* Flow Director perfect filter enabled */
4957 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4958 adapter->atr_sample_rate = 0;
4959 spin_lock_init(&adapter->fdir_perfect_lock);
4961 /* Flow Director hash filters enabled */
4962 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
4963 adapter->atr_sample_rate = 20;
4965 adapter->ring_feature[RING_F_FDIR].indices =
4966 IXGBE_MAX_FDIR_INDICES;
4967 adapter->fdir_pballoc = 0;
4969 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4970 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4971 adapter->ring_feature[RING_F_FCOE].indices = 0;
4972 #ifdef CONFIG_IXGBE_DCB
4973 /* Default traffic class to use for FCoE */
4974 adapter->fcoe.tc = IXGBE_FCOE_DEFTC;
4975 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
4977 #endif /* IXGBE_FCOE */
4983 #ifdef CONFIG_IXGBE_DCB
4984 /* Configure DCB traffic classes */
4985 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
4986 tc = &adapter->dcb_cfg.tc_config[j];
4987 tc->path[DCB_TX_CONFIG].bwg_id = 0;
4988 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
4989 tc->path[DCB_RX_CONFIG].bwg_id = 0;
4990 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
4991 tc->dcb_pfc = pfc_disabled;
4993 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
4994 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
4995 adapter->dcb_cfg.rx_pba_cfg = pba_equal;
4996 adapter->dcb_cfg.pfc_mode_enable = false;
4997 adapter->dcb_cfg.round_robin_enable = false;
4998 adapter->dcb_set_bitmap = 0x00;
4999 ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
5000 adapter->ring_feature[RING_F_DCB].indices);
5004 /* default flow control settings */
5005 hw->fc.requested_mode = ixgbe_fc_full;
5006 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
5008 adapter->last_lfc_mode = hw->fc.current_mode;
5010 hw->fc.high_water = FC_HIGH_WATER(max_frame);
5011 hw->fc.low_water = FC_LOW_WATER(max_frame);
5012 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
5013 hw->fc.send_xon = true;
5014 hw->fc.disable_fc_autoneg = false;
5016 /* enable itr by default in dynamic mode */
5017 adapter->rx_itr_setting = 1;
5018 adapter->rx_eitr_param = 20000;
5019 adapter->tx_itr_setting = 1;
5020 adapter->tx_eitr_param = 10000;
5022 /* set defaults for eitr in MegaBytes */
5023 adapter->eitr_low = 10;
5024 adapter->eitr_high = 20;
5026 /* set default ring sizes */
5027 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
5028 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
5030 /* initialize eeprom parameters */
5031 if (ixgbe_init_eeprom_params_generic(hw)) {
5032 e_dev_err("EEPROM initialization failed\n");
5036 /* enable rx csum by default */
5037 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
5039 /* get assigned NUMA node */
5040 adapter->node = dev_to_node(&pdev->dev);
5042 set_bit(__IXGBE_DOWN, &adapter->state);
5048 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
5049 * @tx_ring: tx descriptor ring (for a specific queue) to setup
5051 * Return 0 on success, negative on failure
5053 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
5055 struct device *dev = tx_ring->dev;
5058 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
5059 tx_ring->tx_buffer_info = vmalloc_node(size, tx_ring->numa_node);
5060 if (!tx_ring->tx_buffer_info)
5061 tx_ring->tx_buffer_info = vmalloc(size);
5062 if (!tx_ring->tx_buffer_info)
5064 memset(tx_ring->tx_buffer_info, 0, size);
5066 /* round up to nearest 4K */
5067 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
5068 tx_ring->size = ALIGN(tx_ring->size, 4096);
5070 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
5071 &tx_ring->dma, GFP_KERNEL);
5075 tx_ring->next_to_use = 0;
5076 tx_ring->next_to_clean = 0;
5077 tx_ring->work_limit = tx_ring->count;
5081 vfree(tx_ring->tx_buffer_info);
5082 tx_ring->tx_buffer_info = NULL;
5083 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
5088 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
5089 * @adapter: board private structure
5091 * If this function returns with an error, then it's possible one or
5092 * more of the rings is populated (while the rest are not). It is the
5093 * callers duty to clean those orphaned rings.
5095 * Return 0 on success, negative on failure
5097 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
5101 for (i = 0; i < adapter->num_tx_queues; i++) {
5102 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
5105 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
5113 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
5114 * @rx_ring: rx descriptor ring (for a specific queue) to setup
5116 * Returns 0 on success, negative on failure
5118 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
5120 struct device *dev = rx_ring->dev;
5123 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
5124 rx_ring->rx_buffer_info = vmalloc_node(size, rx_ring->numa_node);
5125 if (!rx_ring->rx_buffer_info)
5126 rx_ring->rx_buffer_info = vmalloc(size);
5127 if (!rx_ring->rx_buffer_info)
5129 memset(rx_ring->rx_buffer_info, 0, size);
5131 /* Round up to nearest 4K */
5132 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
5133 rx_ring->size = ALIGN(rx_ring->size, 4096);
5135 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
5136 &rx_ring->dma, GFP_KERNEL);
5141 rx_ring->next_to_clean = 0;
5142 rx_ring->next_to_use = 0;
5146 vfree(rx_ring->rx_buffer_info);
5147 rx_ring->rx_buffer_info = NULL;
5148 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
5153 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
5154 * @adapter: board private structure
5156 * If this function returns with an error, then it's possible one or
5157 * more of the rings is populated (while the rest are not). It is the
5158 * callers duty to clean those orphaned rings.
5160 * Return 0 on success, negative on failure
5162 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
5166 for (i = 0; i < adapter->num_rx_queues; i++) {
5167 err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
5170 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
5178 * ixgbe_free_tx_resources - Free Tx Resources per Queue
5179 * @tx_ring: Tx descriptor ring for a specific queue
5181 * Free all transmit software resources
5183 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
5185 ixgbe_clean_tx_ring(tx_ring);
5187 vfree(tx_ring->tx_buffer_info);
5188 tx_ring->tx_buffer_info = NULL;
5190 /* if not set, then don't free */
5194 dma_free_coherent(tx_ring->dev, tx_ring->size,
5195 tx_ring->desc, tx_ring->dma);
5197 tx_ring->desc = NULL;
5201 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5202 * @adapter: board private structure
5204 * Free all transmit software resources
5206 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
5210 for (i = 0; i < adapter->num_tx_queues; i++)
5211 if (adapter->tx_ring[i]->desc)
5212 ixgbe_free_tx_resources(adapter->tx_ring[i]);
5216 * ixgbe_free_rx_resources - Free Rx Resources
5217 * @rx_ring: ring to clean the resources from
5219 * Free all receive software resources
5221 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
5223 ixgbe_clean_rx_ring(rx_ring);
5225 vfree(rx_ring->rx_buffer_info);
5226 rx_ring->rx_buffer_info = NULL;
5228 /* if not set, then don't free */
5232 dma_free_coherent(rx_ring->dev, rx_ring->size,
5233 rx_ring->desc, rx_ring->dma);
5235 rx_ring->desc = NULL;
5239 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5240 * @adapter: board private structure
5242 * Free all receive software resources
5244 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5248 for (i = 0; i < adapter->num_rx_queues; i++)
5249 if (adapter->rx_ring[i]->desc)
5250 ixgbe_free_rx_resources(adapter->rx_ring[i]);
5254 * ixgbe_change_mtu - Change the Maximum Transfer Unit
5255 * @netdev: network interface device structure
5256 * @new_mtu: new value for maximum frame size
5258 * Returns 0 on success, negative on failure
5260 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5262 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5263 struct ixgbe_hw *hw = &adapter->hw;
5264 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5266 /* MTU < 68 is an error and causes problems on some kernels */
5267 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5270 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5271 /* must set new MTU before calling down or up */
5272 netdev->mtu = new_mtu;
5274 hw->fc.high_water = FC_HIGH_WATER(max_frame);
5275 hw->fc.low_water = FC_LOW_WATER(max_frame);
5277 if (netif_running(netdev))
5278 ixgbe_reinit_locked(adapter);
5284 * ixgbe_open - Called when a network interface is made active
5285 * @netdev: network interface device structure
5287 * Returns 0 on success, negative value on failure
5289 * The open entry point is called when a network interface is made
5290 * active by the system (IFF_UP). At this point all resources needed
5291 * for transmit and receive operations are allocated, the interrupt
5292 * handler is registered with the OS, the watchdog timer is started,
5293 * and the stack is notified that the interface is ready.
5295 static int ixgbe_open(struct net_device *netdev)
5297 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5300 /* disallow open during test */
5301 if (test_bit(__IXGBE_TESTING, &adapter->state))
5304 netif_carrier_off(netdev);
5306 /* allocate transmit descriptors */
5307 err = ixgbe_setup_all_tx_resources(adapter);
5311 /* allocate receive descriptors */
5312 err = ixgbe_setup_all_rx_resources(adapter);
5316 ixgbe_configure(adapter);
5318 err = ixgbe_request_irq(adapter);
5322 err = ixgbe_up_complete(adapter);
5326 netif_tx_start_all_queues(netdev);
5331 ixgbe_release_hw_control(adapter);
5332 ixgbe_free_irq(adapter);
5335 ixgbe_free_all_rx_resources(adapter);
5337 ixgbe_free_all_tx_resources(adapter);
5338 ixgbe_reset(adapter);
5344 * ixgbe_close - Disables a network interface
5345 * @netdev: network interface device structure
5347 * Returns 0, this is not allowed to fail
5349 * The close entry point is called when an interface is de-activated
5350 * by the OS. The hardware is still under the drivers control, but
5351 * needs to be disabled. A global MAC reset is issued to stop the
5352 * hardware, and all transmit and receive resources are freed.
5354 static int ixgbe_close(struct net_device *netdev)
5356 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5358 ixgbe_down(adapter);
5359 ixgbe_free_irq(adapter);
5361 ixgbe_free_all_tx_resources(adapter);
5362 ixgbe_free_all_rx_resources(adapter);
5364 ixgbe_release_hw_control(adapter);
5370 static int ixgbe_resume(struct pci_dev *pdev)
5372 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5373 struct net_device *netdev = adapter->netdev;
5376 pci_set_power_state(pdev, PCI_D0);
5377 pci_restore_state(pdev);
5379 * pci_restore_state clears dev->state_saved so call
5380 * pci_save_state to restore it.
5382 pci_save_state(pdev);
5384 err = pci_enable_device_mem(pdev);
5386 e_dev_err("Cannot enable PCI device from suspend\n");
5389 pci_set_master(pdev);
5391 pci_wake_from_d3(pdev, false);
5393 err = ixgbe_init_interrupt_scheme(adapter);
5395 e_dev_err("Cannot initialize interrupts for device\n");
5399 ixgbe_reset(adapter);
5401 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5403 if (netif_running(netdev)) {
5404 err = ixgbe_open(netdev);
5409 netif_device_attach(netdev);
5413 #endif /* CONFIG_PM */
5415 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5417 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5418 struct net_device *netdev = adapter->netdev;
5419 struct ixgbe_hw *hw = &adapter->hw;
5421 u32 wufc = adapter->wol;
5426 netif_device_detach(netdev);
5428 if (netif_running(netdev)) {
5429 ixgbe_down(adapter);
5430 ixgbe_free_irq(adapter);
5431 ixgbe_free_all_tx_resources(adapter);
5432 ixgbe_free_all_rx_resources(adapter);
5435 ixgbe_clear_interrupt_scheme(adapter);
5438 retval = pci_save_state(pdev);
5444 ixgbe_set_rx_mode(netdev);
5446 /* turn on all-multi mode if wake on multicast is enabled */
5447 if (wufc & IXGBE_WUFC_MC) {
5448 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5449 fctrl |= IXGBE_FCTRL_MPE;
5450 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5453 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5454 ctrl |= IXGBE_CTRL_GIO_DIS;
5455 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5457 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5459 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5460 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5463 switch (hw->mac.type) {
5464 case ixgbe_mac_82598EB:
5465 pci_wake_from_d3(pdev, false);
5467 case ixgbe_mac_82599EB:
5468 pci_wake_from_d3(pdev, !!wufc);
5474 *enable_wake = !!wufc;
5476 ixgbe_release_hw_control(adapter);
5478 pci_disable_device(pdev);
5484 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5489 retval = __ixgbe_shutdown(pdev, &wake);
5494 pci_prepare_to_sleep(pdev);
5496 pci_wake_from_d3(pdev, false);
5497 pci_set_power_state(pdev, PCI_D3hot);
5502 #endif /* CONFIG_PM */
5504 static void ixgbe_shutdown(struct pci_dev *pdev)
5508 __ixgbe_shutdown(pdev, &wake);
5510 if (system_state == SYSTEM_POWER_OFF) {
5511 pci_wake_from_d3(pdev, wake);
5512 pci_set_power_state(pdev, PCI_D3hot);
5517 * ixgbe_update_stats - Update the board statistics counters.
5518 * @adapter: board private structure
5520 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5522 struct net_device *netdev = adapter->netdev;
5523 struct ixgbe_hw *hw = &adapter->hw;
5524 struct ixgbe_hw_stats *hwstats = &adapter->stats;
5526 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5527 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
5528 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
5529 u64 bytes = 0, packets = 0;
5531 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5532 test_bit(__IXGBE_RESETTING, &adapter->state))
5535 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5538 for (i = 0; i < 16; i++)
5539 adapter->hw_rx_no_dma_resources +=
5540 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5541 for (i = 0; i < adapter->num_rx_queues; i++) {
5542 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
5543 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
5545 adapter->rsc_total_count = rsc_count;
5546 adapter->rsc_total_flush = rsc_flush;
5549 for (i = 0; i < adapter->num_rx_queues; i++) {
5550 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
5551 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
5552 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
5553 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
5554 bytes += rx_ring->stats.bytes;
5555 packets += rx_ring->stats.packets;
5557 adapter->non_eop_descs = non_eop_descs;
5558 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5559 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5560 netdev->stats.rx_bytes = bytes;
5561 netdev->stats.rx_packets = packets;
5565 /* gather some stats to the adapter struct that are per queue */
5566 for (i = 0; i < adapter->num_tx_queues; i++) {
5567 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5568 restart_queue += tx_ring->tx_stats.restart_queue;
5569 tx_busy += tx_ring->tx_stats.tx_busy;
5570 bytes += tx_ring->stats.bytes;
5571 packets += tx_ring->stats.packets;
5573 adapter->restart_queue = restart_queue;
5574 adapter->tx_busy = tx_busy;
5575 netdev->stats.tx_bytes = bytes;
5576 netdev->stats.tx_packets = packets;
5578 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5579 for (i = 0; i < 8; i++) {
5580 /* for packet buffers not used, the register should read 0 */
5581 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5583 hwstats->mpc[i] += mpc;
5584 total_mpc += hwstats->mpc[i];
5585 if (hw->mac.type == ixgbe_mac_82598EB)
5586 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5587 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5588 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5589 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5590 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5591 switch (hw->mac.type) {
5592 case ixgbe_mac_82598EB:
5593 hwstats->pxonrxc[i] +=
5594 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5595 hwstats->pxoffrxc[i] +=
5596 IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
5598 case ixgbe_mac_82599EB:
5599 hwstats->pxonrxc[i] +=
5600 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5601 hwstats->pxoffrxc[i] +=
5602 IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
5607 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5608 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5610 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5611 /* work around hardware counting issue */
5612 hwstats->gprc -= missed_rx;
5614 /* 82598 hardware only has a 32 bit counter in the high register */
5615 switch (hw->mac.type) {
5616 case ixgbe_mac_82598EB:
5617 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5618 hwstats->lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
5619 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5620 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5621 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5623 case ixgbe_mac_82599EB:
5624 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5625 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
5626 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5627 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
5628 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5629 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5630 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5631 hwstats->lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
5632 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5633 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5635 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5636 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5637 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5638 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5639 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5640 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5641 #endif /* IXGBE_FCOE */
5646 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5647 hwstats->bprc += bprc;
5648 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5649 if (hw->mac.type == ixgbe_mac_82598EB)
5650 hwstats->mprc -= bprc;
5651 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5652 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5653 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5654 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5655 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5656 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5657 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5658 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5659 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5660 hwstats->lxontxc += lxon;
5661 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5662 hwstats->lxofftxc += lxoff;
5663 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5664 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5665 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5667 * 82598 errata - tx of flow control packets is included in tx counters
5669 xon_off_tot = lxon + lxoff;
5670 hwstats->gptc -= xon_off_tot;
5671 hwstats->mptc -= xon_off_tot;
5672 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5673 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5674 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5675 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5676 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5677 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5678 hwstats->ptc64 -= xon_off_tot;
5679 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5680 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5681 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5682 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5683 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5684 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5686 /* Fill out the OS statistics structure */
5687 netdev->stats.multicast = hwstats->mprc;
5690 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
5691 netdev->stats.rx_dropped = 0;
5692 netdev->stats.rx_length_errors = hwstats->rlec;
5693 netdev->stats.rx_crc_errors = hwstats->crcerrs;
5694 netdev->stats.rx_missed_errors = total_mpc;
5698 * ixgbe_watchdog - Timer Call-back
5699 * @data: pointer to adapter cast into an unsigned long
5701 static void ixgbe_watchdog(unsigned long data)
5703 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5704 struct ixgbe_hw *hw = &adapter->hw;
5709 * Do the watchdog outside of interrupt context due to the lovely
5710 * delays that some of the newer hardware requires
5713 if (test_bit(__IXGBE_DOWN, &adapter->state))
5714 goto watchdog_short_circuit;
5716 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5718 * for legacy and MSI interrupts don't set any bits
5719 * that are enabled for EIAM, because this operation
5720 * would set *both* EIMS and EICS for any bit in EIAM
5722 IXGBE_WRITE_REG(hw, IXGBE_EICS,
5723 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5724 goto watchdog_reschedule;
5727 /* get one bit for every active tx/rx interrupt vector */
5728 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
5729 struct ixgbe_q_vector *qv = adapter->q_vector[i];
5730 if (qv->rxr_count || qv->txr_count)
5731 eics |= ((u64)1 << i);
5734 /* Cause software interrupt to ensure rx rings are cleaned */
5735 ixgbe_irq_rearm_queues(adapter, eics);
5737 watchdog_reschedule:
5738 /* Reset the timer */
5739 mod_timer(&adapter->watchdog_timer, round_jiffies(jiffies + 2 * HZ));
5741 watchdog_short_circuit:
5742 schedule_work(&adapter->watchdog_task);
5746 * ixgbe_multispeed_fiber_task - worker thread to configure multispeed fiber
5747 * @work: pointer to work_struct containing our data
5749 static void ixgbe_multispeed_fiber_task(struct work_struct *work)
5751 struct ixgbe_adapter *adapter = container_of(work,
5752 struct ixgbe_adapter,
5753 multispeed_fiber_task);
5754 struct ixgbe_hw *hw = &adapter->hw;
5758 adapter->flags |= IXGBE_FLAG_IN_SFP_LINK_TASK;
5759 autoneg = hw->phy.autoneg_advertised;
5760 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
5761 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
5762 hw->mac.autotry_restart = false;
5763 if (hw->mac.ops.setup_link)
5764 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
5765 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5766 adapter->flags &= ~IXGBE_FLAG_IN_SFP_LINK_TASK;
5770 * ixgbe_sfp_config_module_task - worker thread to configure a new SFP+ module
5771 * @work: pointer to work_struct containing our data
5773 static void ixgbe_sfp_config_module_task(struct work_struct *work)
5775 struct ixgbe_adapter *adapter = container_of(work,
5776 struct ixgbe_adapter,
5777 sfp_config_module_task);
5778 struct ixgbe_hw *hw = &adapter->hw;
5781 adapter->flags |= IXGBE_FLAG_IN_SFP_MOD_TASK;
5783 /* Time for electrical oscillations to settle down */
5785 err = hw->phy.ops.identify_sfp(hw);
5787 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
5788 e_dev_err("failed to initialize because an unsupported SFP+ "
5789 "module type was detected.\n");
5790 e_dev_err("Reload the driver after installing a supported "
5792 unregister_netdev(adapter->netdev);
5795 hw->mac.ops.setup_sfp(hw);
5797 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
5798 /* This will also work for DA Twinax connections */
5799 schedule_work(&adapter->multispeed_fiber_task);
5800 adapter->flags &= ~IXGBE_FLAG_IN_SFP_MOD_TASK;
5804 * ixgbe_fdir_reinit_task - worker thread to reinit FDIR filter table
5805 * @work: pointer to work_struct containing our data
5807 static void ixgbe_fdir_reinit_task(struct work_struct *work)
5809 struct ixgbe_adapter *adapter = container_of(work,
5810 struct ixgbe_adapter,
5812 struct ixgbe_hw *hw = &adapter->hw;
5815 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5816 for (i = 0; i < adapter->num_tx_queues; i++)
5817 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
5818 &(adapter->tx_ring[i]->state));
5820 e_err(probe, "failed to finish FDIR re-initialization, "
5821 "ignored adding FDIR ATR filters\n");
5823 /* Done FDIR Re-initialization, enable transmits */
5824 netif_tx_start_all_queues(adapter->netdev);
5827 static DEFINE_MUTEX(ixgbe_watchdog_lock);
5830 * ixgbe_watchdog_task - worker thread to bring link up
5831 * @work: pointer to work_struct containing our data
5833 static void ixgbe_watchdog_task(struct work_struct *work)
5835 struct ixgbe_adapter *adapter = container_of(work,
5836 struct ixgbe_adapter,
5838 struct net_device *netdev = adapter->netdev;
5839 struct ixgbe_hw *hw = &adapter->hw;
5843 struct ixgbe_ring *tx_ring;
5844 int some_tx_pending = 0;
5846 mutex_lock(&ixgbe_watchdog_lock);
5848 link_up = adapter->link_up;
5849 link_speed = adapter->link_speed;
5851 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
5852 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5855 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5856 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
5857 hw->mac.ops.fc_enable(hw, i);
5859 hw->mac.ops.fc_enable(hw, 0);
5862 hw->mac.ops.fc_enable(hw, 0);
5867 time_after(jiffies, (adapter->link_check_timeout +
5868 IXGBE_TRY_LINK_TIMEOUT))) {
5869 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5870 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5872 adapter->link_up = link_up;
5873 adapter->link_speed = link_speed;
5877 if (!netif_carrier_ok(netdev)) {
5878 bool flow_rx, flow_tx;
5880 switch (hw->mac.type) {
5881 case ixgbe_mac_82598EB: {
5882 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5883 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5884 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5885 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5888 case ixgbe_mac_82599EB: {
5889 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5890 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5891 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5892 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5901 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
5902 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5904 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5905 "1 Gbps" : "unknown speed")),
5906 ((flow_rx && flow_tx) ? "RX/TX" :
5908 (flow_tx ? "TX" : "None"))));
5910 netif_carrier_on(netdev);
5912 /* Force detection of hung controller */
5913 for (i = 0; i < adapter->num_tx_queues; i++) {
5914 tx_ring = adapter->tx_ring[i];
5915 set_check_for_tx_hang(tx_ring);
5919 adapter->link_up = false;
5920 adapter->link_speed = 0;
5921 if (netif_carrier_ok(netdev)) {
5922 e_info(drv, "NIC Link is Down\n");
5923 netif_carrier_off(netdev);
5927 if (!netif_carrier_ok(netdev)) {
5928 for (i = 0; i < adapter->num_tx_queues; i++) {
5929 tx_ring = adapter->tx_ring[i];
5930 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5931 some_tx_pending = 1;
5936 if (some_tx_pending) {
5937 /* We've lost link, so the controller stops DMA,
5938 * but we've got queued Tx work that's never going
5939 * to get done, so reset controller to flush Tx.
5940 * (Do the reset outside of interrupt context).
5942 schedule_work(&adapter->reset_task);
5946 ixgbe_update_stats(adapter);
5947 mutex_unlock(&ixgbe_watchdog_lock);
5950 static int ixgbe_tso(struct ixgbe_adapter *adapter,
5951 struct ixgbe_ring *tx_ring, struct sk_buff *skb,
5952 u32 tx_flags, u8 *hdr_len, __be16 protocol)
5954 struct ixgbe_adv_tx_context_desc *context_desc;
5957 struct ixgbe_tx_buffer *tx_buffer_info;
5958 u32 vlan_macip_lens = 0, type_tucmd_mlhl;
5959 u32 mss_l4len_idx, l4len;
5961 if (skb_is_gso(skb)) {
5962 if (skb_header_cloned(skb)) {
5963 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5967 l4len = tcp_hdrlen(skb);
5970 if (protocol == htons(ETH_P_IP)) {
5971 struct iphdr *iph = ip_hdr(skb);
5974 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
5978 } else if (skb_is_gso_v6(skb)) {
5979 ipv6_hdr(skb)->payload_len = 0;
5980 tcp_hdr(skb)->check =
5981 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5982 &ipv6_hdr(skb)->daddr,
5986 i = tx_ring->next_to_use;
5988 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5989 context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
5991 /* VLAN MACLEN IPLEN */
5992 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5994 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
5995 vlan_macip_lens |= ((skb_network_offset(skb)) <<
5996 IXGBE_ADVTXD_MACLEN_SHIFT);
5997 *hdr_len += skb_network_offset(skb);
5999 (skb_transport_header(skb) - skb_network_header(skb));
6001 (skb_transport_header(skb) - skb_network_header(skb));
6002 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
6003 context_desc->seqnum_seed = 0;
6005 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
6006 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
6007 IXGBE_ADVTXD_DTYP_CTXT);
6009 if (protocol == htons(ETH_P_IP))
6010 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
6011 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6012 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
6016 (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
6017 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
6018 /* use index 1 for TSO */
6019 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6020 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
6022 tx_buffer_info->time_stamp = jiffies;
6023 tx_buffer_info->next_to_watch = i;
6026 if (i == tx_ring->count)
6028 tx_ring->next_to_use = i;
6035 static u32 ixgbe_psum(struct ixgbe_adapter *adapter, struct sk_buff *skb,
6041 case cpu_to_be16(ETH_P_IP):
6042 rtn |= IXGBE_ADVTXD_TUCMD_IPV4;
6043 switch (ip_hdr(skb)->protocol) {
6045 rtn |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6048 rtn |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
6052 case cpu_to_be16(ETH_P_IPV6):
6053 /* XXX what about other V6 headers?? */
6054 switch (ipv6_hdr(skb)->nexthdr) {
6056 rtn |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6059 rtn |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
6064 if (unlikely(net_ratelimit()))
6065 e_warn(probe, "partial checksum but proto=%x!\n",
6073 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
6074 struct ixgbe_ring *tx_ring,
6075 struct sk_buff *skb, u32 tx_flags,
6078 struct ixgbe_adv_tx_context_desc *context_desc;
6080 struct ixgbe_tx_buffer *tx_buffer_info;
6081 u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
6083 if (skb->ip_summed == CHECKSUM_PARTIAL ||
6084 (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
6085 i = tx_ring->next_to_use;
6086 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6087 context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
6089 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
6091 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
6092 vlan_macip_lens |= (skb_network_offset(skb) <<
6093 IXGBE_ADVTXD_MACLEN_SHIFT);
6094 if (skb->ip_summed == CHECKSUM_PARTIAL)
6095 vlan_macip_lens |= (skb_transport_header(skb) -
6096 skb_network_header(skb));
6098 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
6099 context_desc->seqnum_seed = 0;
6101 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
6102 IXGBE_ADVTXD_DTYP_CTXT);
6104 if (skb->ip_summed == CHECKSUM_PARTIAL)
6105 type_tucmd_mlhl |= ixgbe_psum(adapter, skb, protocol);
6107 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
6108 /* use index zero for tx checksum offload */
6109 context_desc->mss_l4len_idx = 0;
6111 tx_buffer_info->time_stamp = jiffies;
6112 tx_buffer_info->next_to_watch = i;
6115 if (i == tx_ring->count)
6117 tx_ring->next_to_use = i;
6125 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
6126 struct ixgbe_ring *tx_ring,
6127 struct sk_buff *skb, u32 tx_flags,
6128 unsigned int first, const u8 hdr_len)
6130 struct device *dev = tx_ring->dev;
6131 struct ixgbe_tx_buffer *tx_buffer_info;
6133 unsigned int total = skb->len;
6134 unsigned int offset = 0, size, count = 0, i;
6135 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
6137 unsigned int bytecount = skb->len;
6140 i = tx_ring->next_to_use;
6142 if (tx_flags & IXGBE_TX_FLAGS_FCOE)
6143 /* excluding fcoe_crc_eof for FCoE */
6144 total -= sizeof(struct fcoe_crc_eof);
6146 len = min(skb_headlen(skb), total);
6148 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6149 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
6151 tx_buffer_info->length = size;
6152 tx_buffer_info->mapped_as_page = false;
6153 tx_buffer_info->dma = dma_map_single(dev,
6155 size, DMA_TO_DEVICE);
6156 if (dma_mapping_error(dev, tx_buffer_info->dma))
6158 tx_buffer_info->time_stamp = jiffies;
6159 tx_buffer_info->next_to_watch = i;
6168 if (i == tx_ring->count)
6173 for (f = 0; f < nr_frags; f++) {
6174 struct skb_frag_struct *frag;
6176 frag = &skb_shinfo(skb)->frags[f];
6177 len = min((unsigned int)frag->size, total);
6178 offset = frag->page_offset;
6182 if (i == tx_ring->count)
6185 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6186 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
6188 tx_buffer_info->length = size;
6189 tx_buffer_info->dma = dma_map_page(dev,
6193 tx_buffer_info->mapped_as_page = true;
6194 if (dma_mapping_error(dev, tx_buffer_info->dma))
6196 tx_buffer_info->time_stamp = jiffies;
6197 tx_buffer_info->next_to_watch = i;
6208 if (tx_flags & IXGBE_TX_FLAGS_TSO)
6209 gso_segs = skb_shinfo(skb)->gso_segs;
6211 /* adjust for FCoE Sequence Offload */
6212 else if (tx_flags & IXGBE_TX_FLAGS_FSO)
6213 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
6214 skb_shinfo(skb)->gso_size);
6215 #endif /* IXGBE_FCOE */
6216 bytecount += (gso_segs - 1) * hdr_len;
6218 /* multiply data chunks by size of headers */
6219 tx_ring->tx_buffer_info[i].bytecount = bytecount;
6220 tx_ring->tx_buffer_info[i].gso_segs = gso_segs;
6221 tx_ring->tx_buffer_info[i].skb = skb;
6222 tx_ring->tx_buffer_info[first].next_to_watch = i;
6227 e_dev_err("TX DMA map failed\n");
6229 /* clear timestamp and dma mappings for failed tx_buffer_info map */
6230 tx_buffer_info->dma = 0;
6231 tx_buffer_info->time_stamp = 0;
6232 tx_buffer_info->next_to_watch = 0;
6236 /* clear timestamp and dma mappings for remaining portion of packet */
6239 i += tx_ring->count;
6241 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6242 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
6248 static void ixgbe_tx_queue(struct ixgbe_ring *tx_ring,
6249 int tx_flags, int count, u32 paylen, u8 hdr_len)
6251 union ixgbe_adv_tx_desc *tx_desc = NULL;
6252 struct ixgbe_tx_buffer *tx_buffer_info;
6253 u32 olinfo_status = 0, cmd_type_len = 0;
6255 u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
6257 cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
6259 cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
6261 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
6262 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
6264 if (tx_flags & IXGBE_TX_FLAGS_TSO) {
6265 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
6267 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
6268 IXGBE_ADVTXD_POPTS_SHIFT;
6270 /* use index 1 context for tso */
6271 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6272 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
6273 olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
6274 IXGBE_ADVTXD_POPTS_SHIFT;
6276 } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
6277 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
6278 IXGBE_ADVTXD_POPTS_SHIFT;
6280 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6281 olinfo_status |= IXGBE_ADVTXD_CC;
6282 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6283 if (tx_flags & IXGBE_TX_FLAGS_FSO)
6284 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
6287 olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
6289 i = tx_ring->next_to_use;
6291 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6292 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
6293 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
6294 tx_desc->read.cmd_type_len =
6295 cpu_to_le32(cmd_type_len | tx_buffer_info->length);
6296 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
6298 if (i == tx_ring->count)
6302 tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
6305 * Force memory writes to complete before letting h/w
6306 * know there are new descriptors to fetch. (Only
6307 * applicable for weak-ordered memory model archs,
6312 tx_ring->next_to_use = i;
6313 writel(i, tx_ring->tail);
6316 static void ixgbe_atr(struct ixgbe_adapter *adapter, struct sk_buff *skb,
6317 u8 queue, u32 tx_flags, __be16 protocol)
6319 struct ixgbe_atr_input atr_input;
6320 struct iphdr *iph = ip_hdr(skb);
6321 struct ethhdr *eth = (struct ethhdr *)skb->data;
6325 /* Right now, we support IPv4 w/ TCP only */
6326 if (protocol != htons(ETH_P_IP) ||
6327 iph->protocol != IPPROTO_TCP)
6330 memset(&atr_input, 0, sizeof(struct ixgbe_atr_input));
6332 vlan_id = (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK) >>
6333 IXGBE_TX_FLAGS_VLAN_SHIFT;
6337 ixgbe_atr_set_vlan_id_82599(&atr_input, vlan_id);
6338 ixgbe_atr_set_src_port_82599(&atr_input, th->dest);
6339 ixgbe_atr_set_dst_port_82599(&atr_input, th->source);
6340 ixgbe_atr_set_flex_byte_82599(&atr_input, eth->h_proto);
6341 ixgbe_atr_set_l4type_82599(&atr_input, IXGBE_ATR_L4TYPE_TCP);
6342 /* src and dst are inverted, think how the receiver sees them */
6343 ixgbe_atr_set_src_ipv4_82599(&atr_input, iph->daddr);
6344 ixgbe_atr_set_dst_ipv4_82599(&atr_input, iph->saddr);
6346 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
6347 ixgbe_fdir_add_signature_filter_82599(&adapter->hw, &atr_input, queue);
6350 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, int size)
6352 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
6353 /* Herbert's original patch had:
6354 * smp_mb__after_netif_stop_queue();
6355 * but since that doesn't exist yet, just open code it. */
6358 /* We need to check again in a case another CPU has just
6359 * made room available. */
6360 if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
6363 /* A reprieve! - use start_queue because it doesn't call schedule */
6364 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
6365 ++tx_ring->tx_stats.restart_queue;
6369 static int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, int size)
6371 if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
6373 return __ixgbe_maybe_stop_tx(tx_ring, size);
6376 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6378 struct ixgbe_adapter *adapter = netdev_priv(dev);
6379 int txq = smp_processor_id();
6383 protocol = vlan_get_protocol(skb);
6385 if ((protocol == htons(ETH_P_FCOE)) ||
6386 (protocol == htons(ETH_P_FIP))) {
6387 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
6388 txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
6389 txq += adapter->ring_feature[RING_F_FCOE].mask;
6391 #ifdef CONFIG_IXGBE_DCB
6392 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6393 txq = adapter->fcoe.up;
6400 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6401 while (unlikely(txq >= dev->real_num_tx_queues))
6402 txq -= dev->real_num_tx_queues;
6406 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6407 if (skb->priority == TC_PRIO_CONTROL)
6408 txq = adapter->ring_feature[RING_F_DCB].indices-1;
6410 txq = (skb->vlan_tci & IXGBE_TX_FLAGS_VLAN_PRIO_MASK)
6415 return skb_tx_hash(dev, skb);
6418 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
6419 struct ixgbe_adapter *adapter,
6420 struct ixgbe_ring *tx_ring)
6422 struct net_device *netdev = tx_ring->netdev;
6423 struct netdev_queue *txq;
6425 unsigned int tx_flags = 0;
6432 protocol = vlan_get_protocol(skb);
6434 if (vlan_tx_tag_present(skb)) {
6435 tx_flags |= vlan_tx_tag_get(skb);
6436 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6437 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6438 tx_flags |= ((skb->queue_mapping & 0x7) << 13);
6440 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6441 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6442 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED &&
6443 skb->priority != TC_PRIO_CONTROL) {
6444 tx_flags |= ((skb->queue_mapping & 0x7) << 13);
6445 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6446 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6450 /* for FCoE with DCB, we force the priority to what
6451 * was specified by the switch */
6452 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED &&
6453 (protocol == htons(ETH_P_FCOE) ||
6454 protocol == htons(ETH_P_FIP))) {
6455 #ifdef CONFIG_IXGBE_DCB
6456 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6457 tx_flags &= ~(IXGBE_TX_FLAGS_VLAN_PRIO_MASK
6458 << IXGBE_TX_FLAGS_VLAN_SHIFT);
6459 tx_flags |= ((adapter->fcoe.up << 13)
6460 << IXGBE_TX_FLAGS_VLAN_SHIFT);
6463 /* flag for FCoE offloads */
6464 if (protocol == htons(ETH_P_FCOE))
6465 tx_flags |= IXGBE_TX_FLAGS_FCOE;
6469 /* four things can cause us to need a context descriptor */
6470 if (skb_is_gso(skb) ||
6471 (skb->ip_summed == CHECKSUM_PARTIAL) ||
6472 (tx_flags & IXGBE_TX_FLAGS_VLAN) ||
6473 (tx_flags & IXGBE_TX_FLAGS_FCOE))
6476 count += TXD_USE_COUNT(skb_headlen(skb));
6477 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6478 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6480 if (ixgbe_maybe_stop_tx(tx_ring, count)) {
6481 tx_ring->tx_stats.tx_busy++;
6482 return NETDEV_TX_BUSY;
6485 first = tx_ring->next_to_use;
6486 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6488 /* setup tx offload for FCoE */
6489 tso = ixgbe_fso(adapter, tx_ring, skb, tx_flags, &hdr_len);
6491 dev_kfree_skb_any(skb);
6492 return NETDEV_TX_OK;
6495 tx_flags |= IXGBE_TX_FLAGS_FSO;
6496 #endif /* IXGBE_FCOE */
6498 if (protocol == htons(ETH_P_IP))
6499 tx_flags |= IXGBE_TX_FLAGS_IPV4;
6500 tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len,
6503 dev_kfree_skb_any(skb);
6504 return NETDEV_TX_OK;
6508 tx_flags |= IXGBE_TX_FLAGS_TSO;
6509 else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags,
6511 (skb->ip_summed == CHECKSUM_PARTIAL))
6512 tx_flags |= IXGBE_TX_FLAGS_CSUM;
6515 count = ixgbe_tx_map(adapter, tx_ring, skb, tx_flags, first, hdr_len);
6517 /* add the ATR filter if ATR is on */
6518 if (tx_ring->atr_sample_rate) {
6519 ++tx_ring->atr_count;
6520 if ((tx_ring->atr_count >= tx_ring->atr_sample_rate) &&
6521 test_bit(__IXGBE_TX_FDIR_INIT_DONE,
6523 ixgbe_atr(adapter, skb, tx_ring->queue_index,
6524 tx_flags, protocol);
6525 tx_ring->atr_count = 0;
6528 txq = netdev_get_tx_queue(netdev, tx_ring->queue_index);
6529 txq->tx_bytes += skb->len;
6531 ixgbe_tx_queue(tx_ring, tx_flags, count, skb->len, hdr_len);
6532 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
6535 dev_kfree_skb_any(skb);
6536 tx_ring->tx_buffer_info[first].time_stamp = 0;
6537 tx_ring->next_to_use = first;
6540 return NETDEV_TX_OK;
6543 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
6545 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6546 struct ixgbe_ring *tx_ring;
6548 tx_ring = adapter->tx_ring[skb->queue_mapping];
6549 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
6553 * ixgbe_set_mac - Change the Ethernet Address of the NIC
6554 * @netdev: network interface device structure
6555 * @p: pointer to an address structure
6557 * Returns 0 on success, negative on failure
6559 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6561 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6562 struct ixgbe_hw *hw = &adapter->hw;
6563 struct sockaddr *addr = p;
6565 if (!is_valid_ether_addr(addr->sa_data))
6566 return -EADDRNOTAVAIL;
6568 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6569 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6571 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
6578 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6580 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6581 struct ixgbe_hw *hw = &adapter->hw;
6585 if (prtad != hw->phy.mdio.prtad)
6587 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6593 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6594 u16 addr, u16 value)
6596 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6597 struct ixgbe_hw *hw = &adapter->hw;
6599 if (prtad != hw->phy.mdio.prtad)
6601 return hw->phy.ops.write_reg(hw, addr, devad, value);
6604 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6606 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6608 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6612 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6614 * @netdev: network interface device structure
6616 * Returns non-zero on failure
6618 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6621 struct ixgbe_adapter *adapter = netdev_priv(dev);
6622 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6624 if (is_valid_ether_addr(mac->san_addr)) {
6626 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6633 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6635 * @netdev: network interface device structure
6637 * Returns non-zero on failure
6639 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6642 struct ixgbe_adapter *adapter = netdev_priv(dev);
6643 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6645 if (is_valid_ether_addr(mac->san_addr)) {
6647 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6653 #ifdef CONFIG_NET_POLL_CONTROLLER
6655 * Polling 'interrupt' - used by things like netconsole to send skbs
6656 * without having to re-enable interrupts. It's not called while
6657 * the interrupt routine is executing.
6659 static void ixgbe_netpoll(struct net_device *netdev)
6661 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6664 /* if interface is down do nothing */
6665 if (test_bit(__IXGBE_DOWN, &adapter->state))
6668 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
6669 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
6670 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
6671 for (i = 0; i < num_q_vectors; i++) {
6672 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
6673 ixgbe_msix_clean_many(0, q_vector);
6676 ixgbe_intr(adapter->pdev->irq, netdev);
6678 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
6682 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
6683 struct rtnl_link_stats64 *stats)
6685 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6688 /* accurate rx/tx bytes/packets stats */
6689 dev_txq_stats_fold(netdev, stats);
6691 for (i = 0; i < adapter->num_rx_queues; i++) {
6692 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
6698 start = u64_stats_fetch_begin_bh(&ring->syncp);
6699 packets = ring->stats.packets;
6700 bytes = ring->stats.bytes;
6701 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6702 stats->rx_packets += packets;
6703 stats->rx_bytes += bytes;
6707 /* following stats updated by ixgbe_watchdog_task() */
6708 stats->multicast = netdev->stats.multicast;
6709 stats->rx_errors = netdev->stats.rx_errors;
6710 stats->rx_length_errors = netdev->stats.rx_length_errors;
6711 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
6712 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
6717 static const struct net_device_ops ixgbe_netdev_ops = {
6718 .ndo_open = ixgbe_open,
6719 .ndo_stop = ixgbe_close,
6720 .ndo_start_xmit = ixgbe_xmit_frame,
6721 .ndo_select_queue = ixgbe_select_queue,
6722 .ndo_set_rx_mode = ixgbe_set_rx_mode,
6723 .ndo_set_multicast_list = ixgbe_set_rx_mode,
6724 .ndo_validate_addr = eth_validate_addr,
6725 .ndo_set_mac_address = ixgbe_set_mac,
6726 .ndo_change_mtu = ixgbe_change_mtu,
6727 .ndo_tx_timeout = ixgbe_tx_timeout,
6728 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
6729 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
6730 .ndo_do_ioctl = ixgbe_ioctl,
6731 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
6732 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
6733 .ndo_set_vf_tx_rate = ixgbe_ndo_set_vf_bw,
6734 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
6735 .ndo_get_stats64 = ixgbe_get_stats64,
6736 #ifdef CONFIG_NET_POLL_CONTROLLER
6737 .ndo_poll_controller = ixgbe_netpoll,
6740 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
6741 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
6742 .ndo_fcoe_enable = ixgbe_fcoe_enable,
6743 .ndo_fcoe_disable = ixgbe_fcoe_disable,
6744 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
6745 #endif /* IXGBE_FCOE */
6748 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
6749 const struct ixgbe_info *ii)
6751 #ifdef CONFIG_PCI_IOV
6752 struct ixgbe_hw *hw = &adapter->hw;
6755 if (hw->mac.type != ixgbe_mac_82599EB || !max_vfs)
6758 /* The 82599 supports up to 64 VFs per physical function
6759 * but this implementation limits allocation to 63 so that
6760 * basic networking resources are still available to the
6763 adapter->num_vfs = (max_vfs > 63) ? 63 : max_vfs;
6764 adapter->flags |= IXGBE_FLAG_SRIOV_ENABLED;
6765 err = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
6767 e_err(probe, "Failed to enable PCI sriov: %d\n", err);
6770 /* If call to enable VFs succeeded then allocate memory
6771 * for per VF control structures.
6774 kcalloc(adapter->num_vfs,
6775 sizeof(struct vf_data_storage), GFP_KERNEL);
6776 if (adapter->vfinfo) {
6777 /* Now that we're sure SR-IOV is enabled
6778 * and memory allocated set up the mailbox parameters
6780 ixgbe_init_mbx_params_pf(hw);
6781 memcpy(&hw->mbx.ops, ii->mbx_ops,
6782 sizeof(hw->mbx.ops));
6784 /* Disable RSC when in SR-IOV mode */
6785 adapter->flags2 &= ~(IXGBE_FLAG2_RSC_CAPABLE |
6786 IXGBE_FLAG2_RSC_ENABLED);
6791 e_err(probe, "Unable to allocate memory for VF Data Storage - "
6792 "SRIOV disabled\n");
6793 pci_disable_sriov(adapter->pdev);
6796 adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
6797 adapter->num_vfs = 0;
6798 #endif /* CONFIG_PCI_IOV */
6802 * ixgbe_probe - Device Initialization Routine
6803 * @pdev: PCI device information struct
6804 * @ent: entry in ixgbe_pci_tbl
6806 * Returns 0 on success, negative on failure
6808 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
6809 * The OS initialization, configuring of the adapter private structure,
6810 * and a hardware reset occur.
6812 static int __devinit ixgbe_probe(struct pci_dev *pdev,
6813 const struct pci_device_id *ent)
6815 struct net_device *netdev;
6816 struct ixgbe_adapter *adapter = NULL;
6817 struct ixgbe_hw *hw;
6818 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
6819 static int cards_found;
6820 int i, err, pci_using_dac;
6821 unsigned int indices = num_possible_cpus();
6827 /* Catch broken hardware that put the wrong VF device ID in
6828 * the PCIe SR-IOV capability.
6830 if (pdev->is_virtfn) {
6831 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
6832 pci_name(pdev), pdev->vendor, pdev->device);
6836 err = pci_enable_device_mem(pdev);
6840 if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
6841 !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
6844 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
6846 err = dma_set_coherent_mask(&pdev->dev,
6850 "No usable DMA configuration, aborting\n");
6857 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
6858 IORESOURCE_MEM), ixgbe_driver_name);
6861 "pci_request_selected_regions failed 0x%x\n", err);
6865 pci_enable_pcie_error_reporting(pdev);
6867 pci_set_master(pdev);
6868 pci_save_state(pdev);
6870 if (ii->mac == ixgbe_mac_82598EB)
6871 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
6873 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
6875 indices = max_t(unsigned int, indices, IXGBE_MAX_DCB_INDICES);
6877 indices += min_t(unsigned int, num_possible_cpus(),
6878 IXGBE_MAX_FCOE_INDICES);
6880 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
6883 goto err_alloc_etherdev;
6886 SET_NETDEV_DEV(netdev, &pdev->dev);
6888 adapter = netdev_priv(netdev);
6889 pci_set_drvdata(pdev, adapter);
6891 adapter->netdev = netdev;
6892 adapter->pdev = pdev;
6895 adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
6897 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
6898 pci_resource_len(pdev, 0));
6904 for (i = 1; i <= 5; i++) {
6905 if (pci_resource_len(pdev, i) == 0)
6909 netdev->netdev_ops = &ixgbe_netdev_ops;
6910 ixgbe_set_ethtool_ops(netdev);
6911 netdev->watchdog_timeo = 5 * HZ;
6912 strcpy(netdev->name, pci_name(pdev));
6914 adapter->bd_number = cards_found;
6917 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
6918 hw->mac.type = ii->mac;
6921 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
6922 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
6923 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
6924 if (!(eec & (1 << 8)))
6925 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
6928 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
6929 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
6930 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
6931 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
6932 hw->phy.mdio.mmds = 0;
6933 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
6934 hw->phy.mdio.dev = netdev;
6935 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
6936 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
6938 /* set up this timer and work struct before calling get_invariants
6939 * which might start the timer
6941 init_timer(&adapter->sfp_timer);
6942 adapter->sfp_timer.function = ixgbe_sfp_timer;
6943 adapter->sfp_timer.data = (unsigned long) adapter;
6945 INIT_WORK(&adapter->sfp_task, ixgbe_sfp_task);
6947 /* multispeed fiber has its own tasklet, called from GPI SDP1 context */
6948 INIT_WORK(&adapter->multispeed_fiber_task, ixgbe_multispeed_fiber_task);
6950 /* a new SFP+ module arrival, called from GPI SDP2 context */
6951 INIT_WORK(&adapter->sfp_config_module_task,
6952 ixgbe_sfp_config_module_task);
6954 ii->get_invariants(hw);
6956 /* setup the private structure */
6957 err = ixgbe_sw_init(adapter);
6961 /* Make it possible the adapter to be woken up via WOL */
6962 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
6963 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6966 * If there is a fan on this device and it has failed log the
6969 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
6970 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
6971 if (esdp & IXGBE_ESDP_SDP1)
6972 e_crit(probe, "Fan has stopped, replace the adapter\n");
6975 /* reset_hw fills in the perm_addr as well */
6976 hw->phy.reset_if_overtemp = true;
6977 err = hw->mac.ops.reset_hw(hw);
6978 hw->phy.reset_if_overtemp = false;
6979 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
6980 hw->mac.type == ixgbe_mac_82598EB) {
6982 * Start a kernel thread to watch for a module to arrive.
6983 * Only do this for 82598, since 82599 will generate
6984 * interrupts on module arrival.
6986 set_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
6987 mod_timer(&adapter->sfp_timer,
6988 round_jiffies(jiffies + (2 * HZ)));
6990 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
6991 e_dev_err("failed to initialize because an unsupported SFP+ "
6992 "module type was detected.\n");
6993 e_dev_err("Reload the driver after installing a supported "
6997 e_dev_err("HW Init failed: %d\n", err);
7001 ixgbe_probe_vf(adapter, ii);
7003 netdev->features = NETIF_F_SG |
7005 NETIF_F_HW_VLAN_TX |
7006 NETIF_F_HW_VLAN_RX |
7007 NETIF_F_HW_VLAN_FILTER;
7009 netdev->features |= NETIF_F_IPV6_CSUM;
7010 netdev->features |= NETIF_F_TSO;
7011 netdev->features |= NETIF_F_TSO6;
7012 netdev->features |= NETIF_F_GRO;
7014 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
7015 netdev->features |= NETIF_F_SCTP_CSUM;
7017 netdev->vlan_features |= NETIF_F_TSO;
7018 netdev->vlan_features |= NETIF_F_TSO6;
7019 netdev->vlan_features |= NETIF_F_IP_CSUM;
7020 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
7021 netdev->vlan_features |= NETIF_F_SG;
7023 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7024 adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
7025 IXGBE_FLAG_DCB_ENABLED);
7026 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
7027 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
7029 #ifdef CONFIG_IXGBE_DCB
7030 netdev->dcbnl_ops = &dcbnl_ops;
7034 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7035 if (hw->mac.ops.get_device_caps) {
7036 hw->mac.ops.get_device_caps(hw, &device_caps);
7037 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
7038 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
7041 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7042 netdev->vlan_features |= NETIF_F_FCOE_CRC;
7043 netdev->vlan_features |= NETIF_F_FSO;
7044 netdev->vlan_features |= NETIF_F_FCOE_MTU;
7046 #endif /* IXGBE_FCOE */
7047 if (pci_using_dac) {
7048 netdev->features |= NETIF_F_HIGHDMA;
7049 netdev->vlan_features |= NETIF_F_HIGHDMA;
7052 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
7053 netdev->features |= NETIF_F_LRO;
7055 /* make sure the EEPROM is good */
7056 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
7057 e_dev_err("The EEPROM Checksum Is Not Valid\n");
7062 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
7063 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
7065 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
7066 e_dev_err("invalid MAC address\n");
7071 /* power down the optics */
7072 if (hw->phy.multispeed_fiber)
7073 hw->mac.ops.disable_tx_laser(hw);
7075 init_timer(&adapter->watchdog_timer);
7076 adapter->watchdog_timer.function = ixgbe_watchdog;
7077 adapter->watchdog_timer.data = (unsigned long)adapter;
7079 INIT_WORK(&adapter->reset_task, ixgbe_reset_task);
7080 INIT_WORK(&adapter->watchdog_task, ixgbe_watchdog_task);
7082 err = ixgbe_init_interrupt_scheme(adapter);
7086 switch (pdev->device) {
7087 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
7088 /* All except this subdevice support WOL */
7089 if (pdev->subsystem_device ==
7090 IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ) {
7094 case IXGBE_DEV_ID_82599_KX4:
7095 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
7096 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
7102 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
7104 /* pick up the PCI bus settings for reporting later */
7105 hw->mac.ops.get_bus_info(hw);
7107 /* print bus type/speed/width info */
7108 e_dev_info("(PCI Express:%s:%s) %pM\n",
7109 (hw->bus.speed == ixgbe_bus_speed_5000 ? "5.0Gb/s" :
7110 hw->bus.speed == ixgbe_bus_speed_2500 ? "2.5Gb/s" :
7112 (hw->bus.width == ixgbe_bus_width_pcie_x8 ? "Width x8" :
7113 hw->bus.width == ixgbe_bus_width_pcie_x4 ? "Width x4" :
7114 hw->bus.width == ixgbe_bus_width_pcie_x1 ? "Width x1" :
7117 ixgbe_read_pba_num_generic(hw, &part_num);
7118 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
7119 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, "
7120 "PBA No: %06x-%03x\n",
7121 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
7122 (part_num >> 8), (part_num & 0xff));
7124 e_dev_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
7125 hw->mac.type, hw->phy.type,
7126 (part_num >> 8), (part_num & 0xff));
7128 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
7129 e_dev_warn("PCI-Express bandwidth available for this card is "
7130 "not sufficient for optimal performance.\n");
7131 e_dev_warn("For optimal performance a x8 PCI-Express slot "
7135 /* save off EEPROM version number */
7136 hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
7138 /* reset the hardware with the new settings */
7139 err = hw->mac.ops.start_hw(hw);
7141 if (err == IXGBE_ERR_EEPROM_VERSION) {
7142 /* We are running on a pre-production device, log a warning */
7143 e_dev_warn("This device is a pre-production adapter/LOM. "
7144 "Please be aware there may be issues associated "
7145 "with your hardware. If you are experiencing "
7146 "problems please contact your Intel or hardware "
7147 "representative who provided you with this "
7150 strcpy(netdev->name, "eth%d");
7151 err = register_netdev(netdev);
7155 /* carrier off reporting is important to ethtool even BEFORE open */
7156 netif_carrier_off(netdev);
7158 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
7159 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
7160 INIT_WORK(&adapter->fdir_reinit_task, ixgbe_fdir_reinit_task);
7162 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
7163 INIT_WORK(&adapter->check_overtemp_task,
7164 ixgbe_check_overtemp_task);
7165 #ifdef CONFIG_IXGBE_DCA
7166 if (dca_add_requester(&pdev->dev) == 0) {
7167 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
7168 ixgbe_setup_dca(adapter);
7171 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7172 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
7173 for (i = 0; i < adapter->num_vfs; i++)
7174 ixgbe_vf_configuration(pdev, (i | 0x10000000));
7177 /* add san mac addr to netdev */
7178 ixgbe_add_sanmac_netdev(netdev);
7180 e_dev_info("Intel(R) 10 Gigabit Network Connection\n");
7185 ixgbe_release_hw_control(adapter);
7186 ixgbe_clear_interrupt_scheme(adapter);
7189 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7190 ixgbe_disable_sriov(adapter);
7191 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
7192 del_timer_sync(&adapter->sfp_timer);
7193 cancel_work_sync(&adapter->sfp_task);
7194 cancel_work_sync(&adapter->multispeed_fiber_task);
7195 cancel_work_sync(&adapter->sfp_config_module_task);
7196 iounmap(hw->hw_addr);
7198 free_netdev(netdev);
7200 pci_release_selected_regions(pdev,
7201 pci_select_bars(pdev, IORESOURCE_MEM));
7204 pci_disable_device(pdev);
7209 * ixgbe_remove - Device Removal Routine
7210 * @pdev: PCI device information struct
7212 * ixgbe_remove is called by the PCI subsystem to alert the driver
7213 * that it should release a PCI device. The could be caused by a
7214 * Hot-Plug event, or because the driver is going to be removed from
7217 static void __devexit ixgbe_remove(struct pci_dev *pdev)
7219 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7220 struct net_device *netdev = adapter->netdev;
7222 set_bit(__IXGBE_DOWN, &adapter->state);
7223 /* clear the module not found bit to make sure the worker won't
7226 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
7227 del_timer_sync(&adapter->watchdog_timer);
7229 del_timer_sync(&adapter->sfp_timer);
7230 cancel_work_sync(&adapter->watchdog_task);
7231 cancel_work_sync(&adapter->sfp_task);
7232 cancel_work_sync(&adapter->multispeed_fiber_task);
7233 cancel_work_sync(&adapter->sfp_config_module_task);
7234 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
7235 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
7236 cancel_work_sync(&adapter->fdir_reinit_task);
7237 flush_scheduled_work();
7239 #ifdef CONFIG_IXGBE_DCA
7240 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
7241 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
7242 dca_remove_requester(&pdev->dev);
7243 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
7248 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
7249 ixgbe_cleanup_fcoe(adapter);
7251 #endif /* IXGBE_FCOE */
7253 /* remove the added san mac */
7254 ixgbe_del_sanmac_netdev(netdev);
7256 if (netdev->reg_state == NETREG_REGISTERED)
7257 unregister_netdev(netdev);
7259 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7260 ixgbe_disable_sriov(adapter);
7262 ixgbe_clear_interrupt_scheme(adapter);
7264 ixgbe_release_hw_control(adapter);
7266 iounmap(adapter->hw.hw_addr);
7267 pci_release_selected_regions(pdev, pci_select_bars(pdev,
7270 e_dev_info("complete\n");
7272 free_netdev(netdev);
7274 pci_disable_pcie_error_reporting(pdev);
7276 pci_disable_device(pdev);
7280 * ixgbe_io_error_detected - called when PCI error is detected
7281 * @pdev: Pointer to PCI device
7282 * @state: The current pci connection state
7284 * This function is called after a PCI bus error affecting
7285 * this device has been detected.
7287 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
7288 pci_channel_state_t state)
7290 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7291 struct net_device *netdev = adapter->netdev;
7293 netif_device_detach(netdev);
7295 if (state == pci_channel_io_perm_failure)
7296 return PCI_ERS_RESULT_DISCONNECT;
7298 if (netif_running(netdev))
7299 ixgbe_down(adapter);
7300 pci_disable_device(pdev);
7302 /* Request a slot reset. */
7303 return PCI_ERS_RESULT_NEED_RESET;
7307 * ixgbe_io_slot_reset - called after the pci bus has been reset.
7308 * @pdev: Pointer to PCI device
7310 * Restart the card from scratch, as if from a cold-boot.
7312 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7314 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7315 pci_ers_result_t result;
7318 if (pci_enable_device_mem(pdev)) {
7319 e_err(probe, "Cannot re-enable PCI device after reset.\n");
7320 result = PCI_ERS_RESULT_DISCONNECT;
7322 pci_set_master(pdev);
7323 pci_restore_state(pdev);
7324 pci_save_state(pdev);
7326 pci_wake_from_d3(pdev, false);
7328 ixgbe_reset(adapter);
7329 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7330 result = PCI_ERS_RESULT_RECOVERED;
7333 err = pci_cleanup_aer_uncorrect_error_status(pdev);
7335 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7336 "failed 0x%0x\n", err);
7337 /* non-fatal, continue */
7344 * ixgbe_io_resume - called when traffic can start flowing again.
7345 * @pdev: Pointer to PCI device
7347 * This callback is called when the error recovery driver tells us that
7348 * its OK to resume normal operation.
7350 static void ixgbe_io_resume(struct pci_dev *pdev)
7352 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7353 struct net_device *netdev = adapter->netdev;
7355 if (netif_running(netdev)) {
7356 if (ixgbe_up(adapter)) {
7357 e_info(probe, "ixgbe_up failed after reset\n");
7362 netif_device_attach(netdev);
7365 static struct pci_error_handlers ixgbe_err_handler = {
7366 .error_detected = ixgbe_io_error_detected,
7367 .slot_reset = ixgbe_io_slot_reset,
7368 .resume = ixgbe_io_resume,
7371 static struct pci_driver ixgbe_driver = {
7372 .name = ixgbe_driver_name,
7373 .id_table = ixgbe_pci_tbl,
7374 .probe = ixgbe_probe,
7375 .remove = __devexit_p(ixgbe_remove),
7377 .suspend = ixgbe_suspend,
7378 .resume = ixgbe_resume,
7380 .shutdown = ixgbe_shutdown,
7381 .err_handler = &ixgbe_err_handler
7385 * ixgbe_init_module - Driver Registration Routine
7387 * ixgbe_init_module is the first routine called when the driver is
7388 * loaded. All it does is register with the PCI subsystem.
7390 static int __init ixgbe_init_module(void)
7393 pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
7394 pr_info("%s\n", ixgbe_copyright);
7396 #ifdef CONFIG_IXGBE_DCA
7397 dca_register_notify(&dca_notifier);
7400 ret = pci_register_driver(&ixgbe_driver);
7404 module_init(ixgbe_init_module);
7407 * ixgbe_exit_module - Driver Exit Cleanup Routine
7409 * ixgbe_exit_module is called just before the driver is removed
7412 static void __exit ixgbe_exit_module(void)
7414 #ifdef CONFIG_IXGBE_DCA
7415 dca_unregister_notify(&dca_notifier);
7417 pci_unregister_driver(&ixgbe_driver);
7418 rcu_barrier(); /* Wait for completion of call_rcu()'s */
7421 #ifdef CONFIG_IXGBE_DCA
7422 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
7427 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
7428 __ixgbe_notify_dca);
7430 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7433 #endif /* CONFIG_IXGBE_DCA */
7436 * ixgbe_get_hw_dev return device
7437 * used by hardware layer to print debugging information
7439 struct net_device *ixgbe_get_hw_dev(struct ixgbe_hw *hw)
7441 struct ixgbe_adapter *adapter = hw->back;
7442 return adapter->netdev;
7445 module_exit(ixgbe_exit_module);