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, "Maximum number of virtual functions to allocate "
135 "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 printk(KERN_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 printk(KERN_ERR "%-15s ", rname);
293 for (j = 0; j < 8; j++)
294 printk(KERN_CONT "%08x ", regs[i*8+j]);
295 printk(KERN_CONT "\n");
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 printk(KERN_INFO "Device Name state "
326 "trans_start last_rx\n");
327 printk(KERN_INFO "%-15s %016lX %016lX %016lX\n",
334 /* Print Registers */
335 dev_info(&adapter->pdev->dev, "Register Dump\n");
336 printk(KERN_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 printk(KERN_INFO "Queue [NTU] [NTC] [bi(ntc)->dma ] "
348 "leng ntw timestamp\n");
349 for (n = 0; n < adapter->num_tx_queues; n++) {
350 tx_ring = adapter->tx_ring[n];
352 &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
353 printk(KERN_INFO " %5d %5X %5X %016llX %04X %3X %016llX\n",
354 n, tx_ring->next_to_use, tx_ring->next_to_clean,
355 (u64)tx_buffer_info->dma,
356 tx_buffer_info->length,
357 tx_buffer_info->next_to_watch,
358 (u64)tx_buffer_info->time_stamp);
362 if (!netif_msg_tx_done(adapter))
363 goto rx_ring_summary;
365 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
367 /* Transmit Descriptor Formats
369 * Advanced Transmit Descriptor
370 * +--------------------------------------------------------------+
371 * 0 | Buffer Address [63:0] |
372 * +--------------------------------------------------------------+
373 * 8 | PAYLEN | PORTS | IDX | STA | DCMD |DTYP | RSV | DTALEN |
374 * +--------------------------------------------------------------+
375 * 63 46 45 40 39 36 35 32 31 24 23 20 19 0
378 for (n = 0; n < adapter->num_tx_queues; n++) {
379 tx_ring = adapter->tx_ring[n];
380 printk(KERN_INFO "------------------------------------\n");
381 printk(KERN_INFO "TX QUEUE INDEX = %d\n", tx_ring->queue_index);
382 printk(KERN_INFO "------------------------------------\n");
383 printk(KERN_INFO "T [desc] [address 63:0 ] "
384 "[PlPOIdStDDt Ln] [bi->dma ] "
385 "leng ntw timestamp bi->skb\n");
387 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
388 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
389 tx_buffer_info = &tx_ring->tx_buffer_info[i];
390 u0 = (struct my_u0 *)tx_desc;
391 printk(KERN_INFO "T [0x%03X] %016llX %016llX %016llX"
392 " %04X %3X %016llX %p", i,
395 (u64)tx_buffer_info->dma,
396 tx_buffer_info->length,
397 tx_buffer_info->next_to_watch,
398 (u64)tx_buffer_info->time_stamp,
399 tx_buffer_info->skb);
400 if (i == tx_ring->next_to_use &&
401 i == tx_ring->next_to_clean)
402 printk(KERN_CONT " NTC/U\n");
403 else if (i == tx_ring->next_to_use)
404 printk(KERN_CONT " NTU\n");
405 else if (i == tx_ring->next_to_clean)
406 printk(KERN_CONT " NTC\n");
408 printk(KERN_CONT "\n");
410 if (netif_msg_pktdata(adapter) &&
411 tx_buffer_info->dma != 0)
412 print_hex_dump(KERN_INFO, "",
413 DUMP_PREFIX_ADDRESS, 16, 1,
414 phys_to_virt(tx_buffer_info->dma),
415 tx_buffer_info->length, true);
419 /* Print RX Rings Summary */
421 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
422 printk(KERN_INFO "Queue [NTU] [NTC]\n");
423 for (n = 0; n < adapter->num_rx_queues; n++) {
424 rx_ring = adapter->rx_ring[n];
425 printk(KERN_INFO "%5d %5X %5X\n", n,
426 rx_ring->next_to_use, rx_ring->next_to_clean);
430 if (!netif_msg_rx_status(adapter))
433 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
435 /* Advanced Receive Descriptor (Read) Format
437 * +-----------------------------------------------------+
438 * 0 | Packet Buffer Address [63:1] |A0/NSE|
439 * +----------------------------------------------+------+
440 * 8 | Header Buffer Address [63:1] | DD |
441 * +-----------------------------------------------------+
444 * Advanced Receive Descriptor (Write-Back) Format
446 * 63 48 47 32 31 30 21 20 16 15 4 3 0
447 * +------------------------------------------------------+
448 * 0 | Packet IP |SPH| HDR_LEN | RSV|Packet| RSS |
449 * | Checksum Ident | | | | Type | Type |
450 * +------------------------------------------------------+
451 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
452 * +------------------------------------------------------+
453 * 63 48 47 32 31 20 19 0
455 for (n = 0; n < adapter->num_rx_queues; n++) {
456 rx_ring = adapter->rx_ring[n];
457 printk(KERN_INFO "------------------------------------\n");
458 printk(KERN_INFO "RX QUEUE INDEX = %d\n", rx_ring->queue_index);
459 printk(KERN_INFO "------------------------------------\n");
460 printk(KERN_INFO "R [desc] [ PktBuf A0] "
461 "[ HeadBuf DD] [bi->dma ] [bi->skb] "
462 "<-- Adv Rx Read format\n");
463 printk(KERN_INFO "RWB[desc] [PcsmIpSHl PtRs] "
464 "[vl er S cks ln] ---------------- [bi->skb] "
465 "<-- Adv Rx Write-Back format\n");
467 for (i = 0; i < rx_ring->count; i++) {
468 rx_buffer_info = &rx_ring->rx_buffer_info[i];
469 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
470 u0 = (struct my_u0 *)rx_desc;
471 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
472 if (staterr & IXGBE_RXD_STAT_DD) {
473 /* Descriptor Done */
474 printk(KERN_INFO "RWB[0x%03X] %016llX "
475 "%016llX ---------------- %p", i,
478 rx_buffer_info->skb);
480 printk(KERN_INFO "R [0x%03X] %016llX "
481 "%016llX %016llX %p", i,
484 (u64)rx_buffer_info->dma,
485 rx_buffer_info->skb);
487 if (netif_msg_pktdata(adapter)) {
488 print_hex_dump(KERN_INFO, "",
489 DUMP_PREFIX_ADDRESS, 16, 1,
490 phys_to_virt(rx_buffer_info->dma),
491 rx_ring->rx_buf_len, true);
493 if (rx_ring->rx_buf_len
494 < IXGBE_RXBUFFER_2048)
495 print_hex_dump(KERN_INFO, "",
496 DUMP_PREFIX_ADDRESS, 16, 1,
498 rx_buffer_info->page_dma +
499 rx_buffer_info->page_offset
505 if (i == rx_ring->next_to_use)
506 printk(KERN_CONT " NTU\n");
507 else if (i == rx_ring->next_to_clean)
508 printk(KERN_CONT " NTC\n");
510 printk(KERN_CONT "\n");
519 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
523 /* Let firmware take over control of h/w */
524 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
525 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
526 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
529 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
533 /* Let firmware know the driver has taken over */
534 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
535 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
536 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
540 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
541 * @adapter: pointer to adapter struct
542 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
543 * @queue: queue to map the corresponding interrupt to
544 * @msix_vector: the vector to map to the corresponding queue
547 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
548 u8 queue, u8 msix_vector)
551 struct ixgbe_hw *hw = &adapter->hw;
552 switch (hw->mac.type) {
553 case ixgbe_mac_82598EB:
554 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
557 index = (((direction * 64) + queue) >> 2) & 0x1F;
558 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
559 ivar &= ~(0xFF << (8 * (queue & 0x3)));
560 ivar |= (msix_vector << (8 * (queue & 0x3)));
561 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
563 case ixgbe_mac_82599EB:
564 if (direction == -1) {
566 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
567 index = ((queue & 1) * 8);
568 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
569 ivar &= ~(0xFF << index);
570 ivar |= (msix_vector << index);
571 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
574 /* tx or rx causes */
575 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
576 index = ((16 * (queue & 1)) + (8 * direction));
577 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
578 ivar &= ~(0xFF << index);
579 ivar |= (msix_vector << index);
580 IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
588 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
593 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
594 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
595 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
597 mask = (qmask & 0xFFFFFFFF);
598 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
599 mask = (qmask >> 32);
600 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
604 static void ixgbe_unmap_and_free_tx_resource(struct ixgbe_adapter *adapter,
605 struct ixgbe_tx_buffer
608 if (tx_buffer_info->dma) {
609 if (tx_buffer_info->mapped_as_page)
610 dma_unmap_page(&adapter->pdev->dev,
612 tx_buffer_info->length,
615 dma_unmap_single(&adapter->pdev->dev,
617 tx_buffer_info->length,
619 tx_buffer_info->dma = 0;
621 if (tx_buffer_info->skb) {
622 dev_kfree_skb_any(tx_buffer_info->skb);
623 tx_buffer_info->skb = NULL;
625 tx_buffer_info->time_stamp = 0;
626 /* tx_buffer_info must be completely set up in the transmit path */
630 * ixgbe_tx_xon_state - check the tx ring xon state
631 * @adapter: the ixgbe adapter
632 * @tx_ring: the corresponding tx_ring
634 * If not in DCB mode, checks TFCS.TXOFF, otherwise, find out the
635 * corresponding TC of this tx_ring when checking TFCS.
637 * Returns : true if in xon state (currently not paused)
639 static inline bool ixgbe_tx_xon_state(struct ixgbe_adapter *adapter,
640 struct ixgbe_ring *tx_ring)
642 u32 txoff = IXGBE_TFCS_TXOFF;
644 #ifdef CONFIG_IXGBE_DCB
645 if (adapter->dcb_cfg.pfc_mode_enable) {
647 int reg_idx = tx_ring->reg_idx;
648 int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
650 switch (adapter->hw.mac.type) {
651 case ixgbe_mac_82598EB:
653 txoff = IXGBE_TFCS_TXOFF0;
655 case ixgbe_mac_82599EB:
657 txoff = IXGBE_TFCS_TXOFF;
661 if (tc == 2) /* TC2, TC3 */
662 tc += (reg_idx - 64) >> 4;
663 else if (tc == 3) /* TC4, TC5, TC6, TC7 */
664 tc += 1 + ((reg_idx - 96) >> 3);
665 } else if (dcb_i == 4) {
669 tc += (reg_idx - 64) >> 5;
670 if (tc == 2) /* TC2, TC3 */
671 tc += (reg_idx - 96) >> 4;
681 return IXGBE_READ_REG(&adapter->hw, IXGBE_TFCS) & txoff;
684 static inline bool ixgbe_check_tx_hang(struct ixgbe_adapter *adapter,
685 struct ixgbe_ring *tx_ring,
688 struct ixgbe_hw *hw = &adapter->hw;
690 /* Detect a transmit hang in hardware, this serializes the
691 * check with the clearing of time_stamp and movement of eop */
692 adapter->detect_tx_hung = false;
693 if (tx_ring->tx_buffer_info[eop].time_stamp &&
694 time_after(jiffies, tx_ring->tx_buffer_info[eop].time_stamp + HZ) &&
695 ixgbe_tx_xon_state(adapter, tx_ring)) {
696 /* detected Tx unit hang */
697 union ixgbe_adv_tx_desc *tx_desc;
698 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
699 e_err(drv, "Detected Tx Unit Hang\n"
701 " TDH, TDT <%x>, <%x>\n"
702 " next_to_use <%x>\n"
703 " next_to_clean <%x>\n"
704 "tx_buffer_info[next_to_clean]\n"
705 " time_stamp <%lx>\n"
707 tx_ring->queue_index,
708 IXGBE_READ_REG(hw, tx_ring->head),
709 IXGBE_READ_REG(hw, tx_ring->tail),
710 tx_ring->next_to_use, eop,
711 tx_ring->tx_buffer_info[eop].time_stamp, jiffies);
718 #define IXGBE_MAX_TXD_PWR 14
719 #define IXGBE_MAX_DATA_PER_TXD (1 << IXGBE_MAX_TXD_PWR)
721 /* Tx Descriptors needed, worst case */
722 #define TXD_USE_COUNT(S) (((S) >> IXGBE_MAX_TXD_PWR) + \
723 (((S) & (IXGBE_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
724 #define DESC_NEEDED (TXD_USE_COUNT(IXGBE_MAX_DATA_PER_TXD) /* skb->data */ + \
725 MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1) /* for context */
727 static void ixgbe_tx_timeout(struct net_device *netdev);
730 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
731 * @q_vector: structure containing interrupt and ring information
732 * @tx_ring: tx ring to clean
734 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
735 struct ixgbe_ring *tx_ring)
737 struct ixgbe_adapter *adapter = q_vector->adapter;
738 struct net_device *netdev = adapter->netdev;
739 union ixgbe_adv_tx_desc *tx_desc, *eop_desc;
740 struct ixgbe_tx_buffer *tx_buffer_info;
741 unsigned int i, eop, count = 0;
742 unsigned int total_bytes = 0, total_packets = 0;
744 i = tx_ring->next_to_clean;
745 eop = tx_ring->tx_buffer_info[i].next_to_watch;
746 eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
748 while ((eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)) &&
749 (count < tx_ring->work_limit)) {
750 bool cleaned = false;
751 rmb(); /* read buffer_info after eop_desc */
752 for ( ; !cleaned; count++) {
754 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
755 tx_buffer_info = &tx_ring->tx_buffer_info[i];
756 cleaned = (i == eop);
757 skb = tx_buffer_info->skb;
759 if (cleaned && skb) {
760 unsigned int segs, bytecount;
761 unsigned int hlen = skb_headlen(skb);
763 /* gso_segs is currently only valid for tcp */
764 segs = skb_shinfo(skb)->gso_segs ?: 1;
766 /* adjust for FCoE Sequence Offload */
767 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
768 && (skb->protocol == htons(ETH_P_FCOE)) &&
770 hlen = skb_transport_offset(skb) +
771 sizeof(struct fc_frame_header) +
772 sizeof(struct fcoe_crc_eof);
773 segs = DIV_ROUND_UP(skb->len - hlen,
774 skb_shinfo(skb)->gso_size);
776 #endif /* IXGBE_FCOE */
777 /* multiply data chunks by size of headers */
778 bytecount = ((segs - 1) * hlen) + skb->len;
779 total_packets += segs;
780 total_bytes += bytecount;
783 ixgbe_unmap_and_free_tx_resource(adapter,
786 tx_desc->wb.status = 0;
789 if (i == tx_ring->count)
793 eop = tx_ring->tx_buffer_info[i].next_to_watch;
794 eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
797 tx_ring->next_to_clean = i;
799 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
800 if (unlikely(count && netif_carrier_ok(netdev) &&
801 (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
802 /* Make sure that anybody stopping the queue after this
803 * sees the new next_to_clean.
806 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
807 !test_bit(__IXGBE_DOWN, &adapter->state)) {
808 netif_wake_subqueue(netdev, tx_ring->queue_index);
809 ++tx_ring->restart_queue;
813 if (adapter->detect_tx_hung) {
814 if (ixgbe_check_tx_hang(adapter, tx_ring, i)) {
815 /* schedule immediate reset if we believe we hung */
816 e_info(probe, "tx hang %d detected, resetting "
817 "adapter\n", adapter->tx_timeout_count + 1);
818 ixgbe_tx_timeout(adapter->netdev);
822 /* re-arm the interrupt */
823 if (count >= tx_ring->work_limit)
824 ixgbe_irq_rearm_queues(adapter, ((u64)1 << q_vector->v_idx));
826 tx_ring->total_bytes += total_bytes;
827 tx_ring->total_packets += total_packets;
828 tx_ring->stats.packets += total_packets;
829 tx_ring->stats.bytes += total_bytes;
830 return (count < tx_ring->work_limit);
833 #ifdef CONFIG_IXGBE_DCA
834 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
835 struct ixgbe_ring *rx_ring)
839 int q = rx_ring->reg_idx;
841 if (rx_ring->cpu != cpu) {
842 rxctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q));
843 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
844 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
845 rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
846 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
847 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK_82599;
848 rxctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
849 IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599);
851 rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
852 rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
853 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_RRO_EN);
854 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_WRO_EN |
855 IXGBE_DCA_RXCTRL_DESC_HSRO_EN);
856 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q), rxctrl);
862 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
863 struct ixgbe_ring *tx_ring)
867 int q = tx_ring->reg_idx;
868 struct ixgbe_hw *hw = &adapter->hw;
870 if (tx_ring->cpu != cpu) {
871 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
872 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(q));
873 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
874 txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
875 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
876 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(q), txctrl);
877 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
878 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(q));
879 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK_82599;
880 txctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
881 IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599);
882 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
883 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(q), txctrl);
890 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
894 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
897 /* always use CB2 mode, difference is masked in the CB driver */
898 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
900 for (i = 0; i < adapter->num_tx_queues; i++) {
901 adapter->tx_ring[i]->cpu = -1;
902 ixgbe_update_tx_dca(adapter, adapter->tx_ring[i]);
904 for (i = 0; i < adapter->num_rx_queues; i++) {
905 adapter->rx_ring[i]->cpu = -1;
906 ixgbe_update_rx_dca(adapter, adapter->rx_ring[i]);
910 static int __ixgbe_notify_dca(struct device *dev, void *data)
912 struct net_device *netdev = dev_get_drvdata(dev);
913 struct ixgbe_adapter *adapter = netdev_priv(netdev);
914 unsigned long event = *(unsigned long *)data;
917 case DCA_PROVIDER_ADD:
918 /* if we're already enabled, don't do it again */
919 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
921 if (dca_add_requester(dev) == 0) {
922 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
923 ixgbe_setup_dca(adapter);
926 /* Fall Through since DCA is disabled. */
927 case DCA_PROVIDER_REMOVE:
928 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
929 dca_remove_requester(dev);
930 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
931 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
939 #endif /* CONFIG_IXGBE_DCA */
941 * ixgbe_receive_skb - Send a completed packet up the stack
942 * @adapter: board private structure
943 * @skb: packet to send up
944 * @status: hardware indication of status of receive
945 * @rx_ring: rx descriptor ring (for a specific queue) to setup
946 * @rx_desc: rx descriptor
948 static void ixgbe_receive_skb(struct ixgbe_q_vector *q_vector,
949 struct sk_buff *skb, u8 status,
950 struct ixgbe_ring *ring,
951 union ixgbe_adv_rx_desc *rx_desc)
953 struct ixgbe_adapter *adapter = q_vector->adapter;
954 struct napi_struct *napi = &q_vector->napi;
955 bool is_vlan = (status & IXGBE_RXD_STAT_VP);
956 u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
958 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL)) {
959 if (adapter->vlgrp && is_vlan && (tag & VLAN_VID_MASK))
960 vlan_gro_receive(napi, adapter->vlgrp, tag, skb);
962 napi_gro_receive(napi, skb);
964 if (adapter->vlgrp && is_vlan && (tag & VLAN_VID_MASK))
965 vlan_hwaccel_rx(skb, adapter->vlgrp, tag);
972 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
973 * @adapter: address of board private structure
974 * @status_err: hardware indication of status of receive
975 * @skb: skb currently being received and modified
977 static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
978 union ixgbe_adv_rx_desc *rx_desc,
981 u32 status_err = le32_to_cpu(rx_desc->wb.upper.status_error);
983 skb->ip_summed = CHECKSUM_NONE;
985 /* Rx csum disabled */
986 if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
989 /* if IP and error */
990 if ((status_err & IXGBE_RXD_STAT_IPCS) &&
991 (status_err & IXGBE_RXDADV_ERR_IPE)) {
992 adapter->hw_csum_rx_error++;
996 if (!(status_err & IXGBE_RXD_STAT_L4CS))
999 if (status_err & IXGBE_RXDADV_ERR_TCPE) {
1000 u16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1003 * 82599 errata, UDP frames with a 0 checksum can be marked as
1006 if ((pkt_info & IXGBE_RXDADV_PKTTYPE_UDP) &&
1007 (adapter->hw.mac.type == ixgbe_mac_82599EB))
1010 adapter->hw_csum_rx_error++;
1014 /* It must be a TCP or UDP packet with a valid checksum */
1015 skb->ip_summed = CHECKSUM_UNNECESSARY;
1018 static inline void ixgbe_release_rx_desc(struct ixgbe_hw *hw,
1019 struct ixgbe_ring *rx_ring, u32 val)
1022 * Force memory writes to complete before letting h/w
1023 * know there are new descriptors to fetch. (Only
1024 * applicable for weak-ordered memory model archs,
1028 IXGBE_WRITE_REG(hw, IXGBE_RDT(rx_ring->reg_idx), val);
1032 * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
1033 * @adapter: address of board private structure
1035 static void ixgbe_alloc_rx_buffers(struct ixgbe_adapter *adapter,
1036 struct ixgbe_ring *rx_ring,
1039 struct net_device *netdev = adapter->netdev;
1040 struct pci_dev *pdev = adapter->pdev;
1041 union ixgbe_adv_rx_desc *rx_desc;
1042 struct ixgbe_rx_buffer *bi;
1044 unsigned int bufsz = rx_ring->rx_buf_len;
1046 i = rx_ring->next_to_use;
1047 bi = &rx_ring->rx_buffer_info[i];
1049 while (cleaned_count--) {
1050 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
1052 if (!bi->page_dma &&
1053 (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED)) {
1055 bi->page = netdev_alloc_page(netdev);
1057 adapter->alloc_rx_page_failed++;
1060 bi->page_offset = 0;
1062 /* use a half page if we're re-using */
1063 bi->page_offset ^= (PAGE_SIZE / 2);
1066 bi->page_dma = dma_map_page(&pdev->dev, bi->page,
1073 struct sk_buff *skb = netdev_alloc_skb_ip_align(netdev,
1078 adapter->alloc_rx_buff_failed++;
1081 /* initialize queue mapping */
1082 skb_record_rx_queue(skb, rx_ring->queue_index);
1086 bi->dma = dma_map_single(&pdev->dev,
1088 rx_ring->rx_buf_len,
1091 /* Refresh the desc even if buffer_addrs didn't change because
1092 * each write-back erases this info. */
1093 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
1094 rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
1095 rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
1097 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
1101 if (i == rx_ring->count)
1103 bi = &rx_ring->rx_buffer_info[i];
1107 if (rx_ring->next_to_use != i) {
1108 rx_ring->next_to_use = i;
1110 i = (rx_ring->count - 1);
1112 ixgbe_release_rx_desc(&adapter->hw, rx_ring, i);
1116 static inline u16 ixgbe_get_hdr_info(union ixgbe_adv_rx_desc *rx_desc)
1118 return rx_desc->wb.lower.lo_dword.hs_rss.hdr_info;
1121 static inline u16 ixgbe_get_pkt_info(union ixgbe_adv_rx_desc *rx_desc)
1123 return rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1126 static inline u32 ixgbe_get_rsc_count(union ixgbe_adv_rx_desc *rx_desc)
1128 return (le32_to_cpu(rx_desc->wb.lower.lo_dword.data) &
1129 IXGBE_RXDADV_RSCCNT_MASK) >>
1130 IXGBE_RXDADV_RSCCNT_SHIFT;
1134 * ixgbe_transform_rsc_queue - change rsc queue into a full packet
1135 * @skb: pointer to the last skb in the rsc queue
1136 * @count: pointer to number of packets coalesced in this context
1138 * This function changes a queue full of hw rsc buffers into a completed
1139 * packet. It uses the ->prev pointers to find the first packet and then
1140 * turns it into the frag list owner.
1142 static inline struct sk_buff *ixgbe_transform_rsc_queue(struct sk_buff *skb,
1145 unsigned int frag_list_size = 0;
1148 struct sk_buff *prev = skb->prev;
1149 frag_list_size += skb->len;
1155 skb_shinfo(skb)->frag_list = skb->next;
1157 skb->len += frag_list_size;
1158 skb->data_len += frag_list_size;
1159 skb->truesize += frag_list_size;
1163 struct ixgbe_rsc_cb {
1168 #define IXGBE_RSC_CB(skb) ((struct ixgbe_rsc_cb *)(skb)->cb)
1170 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1171 struct ixgbe_ring *rx_ring,
1172 int *work_done, int work_to_do)
1174 struct ixgbe_adapter *adapter = q_vector->adapter;
1175 struct net_device *netdev = adapter->netdev;
1176 struct pci_dev *pdev = adapter->pdev;
1177 union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
1178 struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
1179 struct sk_buff *skb;
1180 unsigned int i, rsc_count = 0;
1183 bool cleaned = false;
1184 int cleaned_count = 0;
1185 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1188 #endif /* IXGBE_FCOE */
1190 i = rx_ring->next_to_clean;
1191 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
1192 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1193 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1195 while (staterr & IXGBE_RXD_STAT_DD) {
1197 if (*work_done >= work_to_do)
1201 rmb(); /* read descriptor and rx_buffer_info after status DD */
1202 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
1203 hdr_info = le16_to_cpu(ixgbe_get_hdr_info(rx_desc));
1204 len = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
1205 IXGBE_RXDADV_HDRBUFLEN_SHIFT;
1206 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1207 if ((len > IXGBE_RX_HDR_SIZE) ||
1208 (upper_len && !(hdr_info & IXGBE_RXDADV_SPH)))
1209 len = IXGBE_RX_HDR_SIZE;
1211 len = le16_to_cpu(rx_desc->wb.upper.length);
1215 skb = rx_buffer_info->skb;
1216 prefetch(skb->data);
1217 rx_buffer_info->skb = NULL;
1219 if (rx_buffer_info->dma) {
1220 if ((adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
1221 (!(staterr & IXGBE_RXD_STAT_EOP)) &&
1224 * When HWRSC is enabled, delay unmapping
1225 * of the first packet. It carries the
1226 * header information, HW may still
1227 * access the header after the writeback.
1228 * Only unmap it when EOP is reached
1230 IXGBE_RSC_CB(skb)->delay_unmap = true;
1231 IXGBE_RSC_CB(skb)->dma = rx_buffer_info->dma;
1233 dma_unmap_single(&pdev->dev,
1234 rx_buffer_info->dma,
1235 rx_ring->rx_buf_len,
1238 rx_buffer_info->dma = 0;
1243 dma_unmap_page(&pdev->dev, rx_buffer_info->page_dma,
1244 PAGE_SIZE / 2, DMA_FROM_DEVICE);
1245 rx_buffer_info->page_dma = 0;
1246 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1247 rx_buffer_info->page,
1248 rx_buffer_info->page_offset,
1251 if ((rx_ring->rx_buf_len > (PAGE_SIZE / 2)) ||
1252 (page_count(rx_buffer_info->page) != 1))
1253 rx_buffer_info->page = NULL;
1255 get_page(rx_buffer_info->page);
1257 skb->len += upper_len;
1258 skb->data_len += upper_len;
1259 skb->truesize += upper_len;
1263 if (i == rx_ring->count)
1266 next_rxd = IXGBE_RX_DESC_ADV(*rx_ring, i);
1270 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
1271 rsc_count = ixgbe_get_rsc_count(rx_desc);
1274 u32 nextp = (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
1275 IXGBE_RXDADV_NEXTP_SHIFT;
1276 next_buffer = &rx_ring->rx_buffer_info[nextp];
1278 next_buffer = &rx_ring->rx_buffer_info[i];
1281 if (staterr & IXGBE_RXD_STAT_EOP) {
1283 skb = ixgbe_transform_rsc_queue(skb, &(rx_ring->rsc_count));
1284 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
1285 if (IXGBE_RSC_CB(skb)->delay_unmap) {
1286 dma_unmap_single(&pdev->dev,
1287 IXGBE_RSC_CB(skb)->dma,
1288 rx_ring->rx_buf_len,
1290 IXGBE_RSC_CB(skb)->dma = 0;
1291 IXGBE_RSC_CB(skb)->delay_unmap = false;
1293 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED)
1294 rx_ring->rsc_count += skb_shinfo(skb)->nr_frags;
1296 rx_ring->rsc_count++;
1297 rx_ring->rsc_flush++;
1299 rx_ring->stats.packets++;
1300 rx_ring->stats.bytes += skb->len;
1302 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
1303 rx_buffer_info->skb = next_buffer->skb;
1304 rx_buffer_info->dma = next_buffer->dma;
1305 next_buffer->skb = skb;
1306 next_buffer->dma = 0;
1308 skb->next = next_buffer->skb;
1309 skb->next->prev = skb;
1311 rx_ring->non_eop_descs++;
1315 if (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) {
1316 dev_kfree_skb_irq(skb);
1320 ixgbe_rx_checksum(adapter, rx_desc, skb);
1322 /* probably a little skewed due to removing CRC */
1323 total_rx_bytes += skb->len;
1326 skb->protocol = eth_type_trans(skb, adapter->netdev);
1328 /* if ddp, not passing to ULD unless for FCP_RSP or error */
1329 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
1330 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
1334 #endif /* IXGBE_FCOE */
1335 ixgbe_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
1338 rx_desc->wb.upper.status_error = 0;
1340 /* return some buffers to hardware, one at a time is too slow */
1341 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
1342 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
1346 /* use prefetched values */
1348 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1350 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1353 rx_ring->next_to_clean = i;
1354 cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
1357 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
1360 /* include DDPed FCoE data */
1361 if (ddp_bytes > 0) {
1364 mss = adapter->netdev->mtu - sizeof(struct fcoe_hdr) -
1365 sizeof(struct fc_frame_header) -
1366 sizeof(struct fcoe_crc_eof);
1369 total_rx_bytes += ddp_bytes;
1370 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
1372 #endif /* IXGBE_FCOE */
1374 rx_ring->total_packets += total_rx_packets;
1375 rx_ring->total_bytes += total_rx_bytes;
1376 netdev->stats.rx_bytes += total_rx_bytes;
1377 netdev->stats.rx_packets += total_rx_packets;
1382 static int ixgbe_clean_rxonly(struct napi_struct *, int);
1384 * ixgbe_configure_msix - Configure MSI-X hardware
1385 * @adapter: board private structure
1387 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1390 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1392 struct ixgbe_q_vector *q_vector;
1393 int i, j, q_vectors, v_idx, r_idx;
1396 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1399 * Populate the IVAR table and set the ITR values to the
1400 * corresponding register.
1402 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1403 q_vector = adapter->q_vector[v_idx];
1404 /* XXX for_each_set_bit(...) */
1405 r_idx = find_first_bit(q_vector->rxr_idx,
1406 adapter->num_rx_queues);
1408 for (i = 0; i < q_vector->rxr_count; i++) {
1409 j = adapter->rx_ring[r_idx]->reg_idx;
1410 ixgbe_set_ivar(adapter, 0, j, v_idx);
1411 r_idx = find_next_bit(q_vector->rxr_idx,
1412 adapter->num_rx_queues,
1415 r_idx = find_first_bit(q_vector->txr_idx,
1416 adapter->num_tx_queues);
1418 for (i = 0; i < q_vector->txr_count; i++) {
1419 j = adapter->tx_ring[r_idx]->reg_idx;
1420 ixgbe_set_ivar(adapter, 1, j, v_idx);
1421 r_idx = find_next_bit(q_vector->txr_idx,
1422 adapter->num_tx_queues,
1426 if (q_vector->txr_count && !q_vector->rxr_count)
1428 q_vector->eitr = adapter->tx_eitr_param;
1429 else if (q_vector->rxr_count)
1431 q_vector->eitr = adapter->rx_eitr_param;
1433 ixgbe_write_eitr(q_vector);
1436 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
1437 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1439 else if (adapter->hw.mac.type == ixgbe_mac_82599EB)
1440 ixgbe_set_ivar(adapter, -1, 1, v_idx);
1441 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1443 /* set up to autoclear timer, and the vectors */
1444 mask = IXGBE_EIMS_ENABLE_MASK;
1445 if (adapter->num_vfs)
1446 mask &= ~(IXGBE_EIMS_OTHER |
1447 IXGBE_EIMS_MAILBOX |
1450 mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
1451 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1454 enum latency_range {
1458 latency_invalid = 255
1462 * ixgbe_update_itr - update the dynamic ITR value based on statistics
1463 * @adapter: pointer to adapter
1464 * @eitr: eitr setting (ints per sec) to give last timeslice
1465 * @itr_setting: current throttle rate in ints/second
1466 * @packets: the number of packets during this measurement interval
1467 * @bytes: the number of bytes during this measurement interval
1469 * Stores a new ITR value based on packets and byte
1470 * counts during the last interrupt. The advantage of per interrupt
1471 * computation is faster updates and more accurate ITR for the current
1472 * traffic pattern. Constants in this function were computed
1473 * based on theoretical maximum wire speed and thresholds were set based
1474 * on testing data as well as attempting to minimize response time
1475 * while increasing bulk throughput.
1476 * this functionality is controlled by the InterruptThrottleRate module
1477 * parameter (see ixgbe_param.c)
1479 static u8 ixgbe_update_itr(struct ixgbe_adapter *adapter,
1480 u32 eitr, u8 itr_setting,
1481 int packets, int bytes)
1483 unsigned int retval = itr_setting;
1488 goto update_itr_done;
1491 /* simple throttlerate management
1492 * 0-20MB/s lowest (100000 ints/s)
1493 * 20-100MB/s low (20000 ints/s)
1494 * 100-1249MB/s bulk (8000 ints/s)
1496 /* what was last interrupt timeslice? */
1497 timepassed_us = 1000000/eitr;
1498 bytes_perint = bytes / timepassed_us; /* bytes/usec */
1500 switch (itr_setting) {
1501 case lowest_latency:
1502 if (bytes_perint > adapter->eitr_low)
1503 retval = low_latency;
1506 if (bytes_perint > adapter->eitr_high)
1507 retval = bulk_latency;
1508 else if (bytes_perint <= adapter->eitr_low)
1509 retval = lowest_latency;
1512 if (bytes_perint <= adapter->eitr_high)
1513 retval = low_latency;
1522 * ixgbe_write_eitr - write EITR register in hardware specific way
1523 * @q_vector: structure containing interrupt and ring information
1525 * This function is made to be called by ethtool and by the driver
1526 * when it needs to update EITR registers at runtime. Hardware
1527 * specific quirks/differences are taken care of here.
1529 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1531 struct ixgbe_adapter *adapter = q_vector->adapter;
1532 struct ixgbe_hw *hw = &adapter->hw;
1533 int v_idx = q_vector->v_idx;
1534 u32 itr_reg = EITR_INTS_PER_SEC_TO_REG(q_vector->eitr);
1536 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1537 /* must write high and low 16 bits to reset counter */
1538 itr_reg |= (itr_reg << 16);
1539 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1541 * 82599 can support a value of zero, so allow it for
1542 * max interrupt rate, but there is an errata where it can
1543 * not be zero with RSC
1546 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
1550 * set the WDIS bit to not clear the timer bits and cause an
1551 * immediate assertion of the interrupt
1553 itr_reg |= IXGBE_EITR_CNT_WDIS;
1555 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1558 static void ixgbe_set_itr_msix(struct ixgbe_q_vector *q_vector)
1560 struct ixgbe_adapter *adapter = q_vector->adapter;
1562 u8 current_itr, ret_itr;
1564 struct ixgbe_ring *rx_ring, *tx_ring;
1566 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1567 for (i = 0; i < q_vector->txr_count; i++) {
1568 tx_ring = adapter->tx_ring[r_idx];
1569 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1571 tx_ring->total_packets,
1572 tx_ring->total_bytes);
1573 /* if the result for this queue would decrease interrupt
1574 * rate for this vector then use that result */
1575 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
1576 q_vector->tx_itr - 1 : ret_itr);
1577 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1581 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1582 for (i = 0; i < q_vector->rxr_count; i++) {
1583 rx_ring = adapter->rx_ring[r_idx];
1584 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1586 rx_ring->total_packets,
1587 rx_ring->total_bytes);
1588 /* if the result for this queue would decrease interrupt
1589 * rate for this vector then use that result */
1590 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
1591 q_vector->rx_itr - 1 : ret_itr);
1592 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1596 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1598 switch (current_itr) {
1599 /* counts and packets in update_itr are dependent on these numbers */
1600 case lowest_latency:
1604 new_itr = 20000; /* aka hwitr = ~200 */
1612 if (new_itr != q_vector->eitr) {
1613 /* do an exponential smoothing */
1614 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1616 /* save the algorithm value here, not the smoothed one */
1617 q_vector->eitr = new_itr;
1619 ixgbe_write_eitr(q_vector);
1624 * ixgbe_check_overtemp_task - worker thread to check over tempurature
1625 * @work: pointer to work_struct containing our data
1627 static void ixgbe_check_overtemp_task(struct work_struct *work)
1629 struct ixgbe_adapter *adapter = container_of(work,
1630 struct ixgbe_adapter,
1631 check_overtemp_task);
1632 struct ixgbe_hw *hw = &adapter->hw;
1633 u32 eicr = adapter->interrupt_event;
1635 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
1636 switch (hw->device_id) {
1637 case IXGBE_DEV_ID_82599_T3_LOM: {
1639 bool link_up = false;
1641 if (hw->mac.ops.check_link)
1642 hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
1644 if (((eicr & IXGBE_EICR_GPI_SDP0) && (!link_up)) ||
1645 (eicr & IXGBE_EICR_LSC))
1646 /* Check if this is due to overtemp */
1647 if (hw->phy.ops.check_overtemp(hw) == IXGBE_ERR_OVERTEMP)
1652 if (!(eicr & IXGBE_EICR_GPI_SDP0))
1656 e_crit(drv, "Network adapter has been stopped because it has "
1657 "over heated. Restart the computer. If the problem "
1658 "persists, power off the system and replace the "
1660 /* write to clear the interrupt */
1661 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP0);
1665 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
1667 struct ixgbe_hw *hw = &adapter->hw;
1669 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
1670 (eicr & IXGBE_EICR_GPI_SDP1)) {
1671 e_crit(probe, "Fan has stopped, replace the adapter\n");
1672 /* write to clear the interrupt */
1673 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1677 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
1679 struct ixgbe_hw *hw = &adapter->hw;
1681 if (eicr & IXGBE_EICR_GPI_SDP1) {
1682 /* Clear the interrupt */
1683 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1684 schedule_work(&adapter->multispeed_fiber_task);
1685 } else if (eicr & IXGBE_EICR_GPI_SDP2) {
1686 /* Clear the interrupt */
1687 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
1688 schedule_work(&adapter->sfp_config_module_task);
1690 /* Interrupt isn't for us... */
1695 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1697 struct ixgbe_hw *hw = &adapter->hw;
1700 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1701 adapter->link_check_timeout = jiffies;
1702 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1703 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1704 IXGBE_WRITE_FLUSH(hw);
1705 schedule_work(&adapter->watchdog_task);
1709 static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
1711 struct net_device *netdev = data;
1712 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1713 struct ixgbe_hw *hw = &adapter->hw;
1717 * Workaround for Silicon errata. Use clear-by-write instead
1718 * of clear-by-read. Reading with EICS will return the
1719 * interrupt causes without clearing, which later be done
1720 * with the write to EICR.
1722 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1723 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1725 if (eicr & IXGBE_EICR_LSC)
1726 ixgbe_check_lsc(adapter);
1728 if (eicr & IXGBE_EICR_MAILBOX)
1729 ixgbe_msg_task(adapter);
1731 if (hw->mac.type == ixgbe_mac_82598EB)
1732 ixgbe_check_fan_failure(adapter, eicr);
1734 if (hw->mac.type == ixgbe_mac_82599EB) {
1735 ixgbe_check_sfp_event(adapter, eicr);
1736 adapter->interrupt_event = eicr;
1737 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
1738 ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)))
1739 schedule_work(&adapter->check_overtemp_task);
1741 /* Handle Flow Director Full threshold interrupt */
1742 if (eicr & IXGBE_EICR_FLOW_DIR) {
1744 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_FLOW_DIR);
1745 /* Disable transmits before FDIR Re-initialization */
1746 netif_tx_stop_all_queues(netdev);
1747 for (i = 0; i < adapter->num_tx_queues; i++) {
1748 struct ixgbe_ring *tx_ring =
1749 adapter->tx_ring[i];
1750 if (test_and_clear_bit(__IXGBE_FDIR_INIT_DONE,
1751 &tx_ring->reinit_state))
1752 schedule_work(&adapter->fdir_reinit_task);
1756 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1757 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_OTHER);
1762 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
1767 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1768 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1769 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1771 mask = (qmask & 0xFFFFFFFF);
1772 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(0), mask);
1773 mask = (qmask >> 32);
1774 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(1), mask);
1776 /* skip the flush */
1779 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
1784 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1785 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1786 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, mask);
1788 mask = (qmask & 0xFFFFFFFF);
1789 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), mask);
1790 mask = (qmask >> 32);
1791 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), mask);
1793 /* skip the flush */
1796 static irqreturn_t ixgbe_msix_clean_tx(int irq, void *data)
1798 struct ixgbe_q_vector *q_vector = data;
1799 struct ixgbe_adapter *adapter = q_vector->adapter;
1800 struct ixgbe_ring *tx_ring;
1803 if (!q_vector->txr_count)
1806 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1807 for (i = 0; i < q_vector->txr_count; i++) {
1808 tx_ring = adapter->tx_ring[r_idx];
1809 tx_ring->total_bytes = 0;
1810 tx_ring->total_packets = 0;
1811 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1815 /* EIAM disabled interrupts (on this vector) for us */
1816 napi_schedule(&q_vector->napi);
1822 * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
1824 * @data: pointer to our q_vector struct for this interrupt vector
1826 static irqreturn_t ixgbe_msix_clean_rx(int irq, void *data)
1828 struct ixgbe_q_vector *q_vector = data;
1829 struct ixgbe_adapter *adapter = q_vector->adapter;
1830 struct ixgbe_ring *rx_ring;
1834 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1835 for (i = 0; i < q_vector->rxr_count; i++) {
1836 rx_ring = adapter->rx_ring[r_idx];
1837 rx_ring->total_bytes = 0;
1838 rx_ring->total_packets = 0;
1839 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1843 if (!q_vector->rxr_count)
1846 /* disable interrupts on this vector only */
1847 /* EIAM disabled interrupts (on this vector) for us */
1848 napi_schedule(&q_vector->napi);
1853 static irqreturn_t ixgbe_msix_clean_many(int irq, void *data)
1855 struct ixgbe_q_vector *q_vector = data;
1856 struct ixgbe_adapter *adapter = q_vector->adapter;
1857 struct ixgbe_ring *ring;
1861 if (!q_vector->txr_count && !q_vector->rxr_count)
1864 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1865 for (i = 0; i < q_vector->txr_count; i++) {
1866 ring = adapter->tx_ring[r_idx];
1867 ring->total_bytes = 0;
1868 ring->total_packets = 0;
1869 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1873 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1874 for (i = 0; i < q_vector->rxr_count; i++) {
1875 ring = adapter->rx_ring[r_idx];
1876 ring->total_bytes = 0;
1877 ring->total_packets = 0;
1878 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1882 /* EIAM disabled interrupts (on this vector) for us */
1883 napi_schedule(&q_vector->napi);
1889 * ixgbe_clean_rxonly - msix (aka one shot) rx clean routine
1890 * @napi: napi struct with our devices info in it
1891 * @budget: amount of work driver is allowed to do this pass, in packets
1893 * This function is optimized for cleaning one queue only on a single
1896 static int ixgbe_clean_rxonly(struct napi_struct *napi, int budget)
1898 struct ixgbe_q_vector *q_vector =
1899 container_of(napi, struct ixgbe_q_vector, napi);
1900 struct ixgbe_adapter *adapter = q_vector->adapter;
1901 struct ixgbe_ring *rx_ring = NULL;
1905 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1906 rx_ring = adapter->rx_ring[r_idx];
1907 #ifdef CONFIG_IXGBE_DCA
1908 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1909 ixgbe_update_rx_dca(adapter, rx_ring);
1912 ixgbe_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
1914 /* If all Rx work done, exit the polling mode */
1915 if (work_done < budget) {
1916 napi_complete(napi);
1917 if (adapter->rx_itr_setting & 1)
1918 ixgbe_set_itr_msix(q_vector);
1919 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1920 ixgbe_irq_enable_queues(adapter,
1921 ((u64)1 << q_vector->v_idx));
1928 * ixgbe_clean_rxtx_many - msix (aka one shot) rx clean routine
1929 * @napi: napi struct with our devices info in it
1930 * @budget: amount of work driver is allowed to do this pass, in packets
1932 * This function will clean more than one rx queue associated with a
1935 static int ixgbe_clean_rxtx_many(struct napi_struct *napi, int budget)
1937 struct ixgbe_q_vector *q_vector =
1938 container_of(napi, struct ixgbe_q_vector, napi);
1939 struct ixgbe_adapter *adapter = q_vector->adapter;
1940 struct ixgbe_ring *ring = NULL;
1941 int work_done = 0, i;
1943 bool tx_clean_complete = true;
1945 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1946 for (i = 0; i < q_vector->txr_count; i++) {
1947 ring = adapter->tx_ring[r_idx];
1948 #ifdef CONFIG_IXGBE_DCA
1949 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1950 ixgbe_update_tx_dca(adapter, ring);
1952 tx_clean_complete &= ixgbe_clean_tx_irq(q_vector, ring);
1953 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1957 /* attempt to distribute budget to each queue fairly, but don't allow
1958 * the budget to go below 1 because we'll exit polling */
1959 budget /= (q_vector->rxr_count ?: 1);
1960 budget = max(budget, 1);
1961 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1962 for (i = 0; i < q_vector->rxr_count; i++) {
1963 ring = adapter->rx_ring[r_idx];
1964 #ifdef CONFIG_IXGBE_DCA
1965 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1966 ixgbe_update_rx_dca(adapter, ring);
1968 ixgbe_clean_rx_irq(q_vector, ring, &work_done, budget);
1969 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1973 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1974 ring = adapter->rx_ring[r_idx];
1975 /* If all Rx work done, exit the polling mode */
1976 if (work_done < budget) {
1977 napi_complete(napi);
1978 if (adapter->rx_itr_setting & 1)
1979 ixgbe_set_itr_msix(q_vector);
1980 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1981 ixgbe_irq_enable_queues(adapter,
1982 ((u64)1 << q_vector->v_idx));
1990 * ixgbe_clean_txonly - msix (aka one shot) tx clean routine
1991 * @napi: napi struct with our devices info in it
1992 * @budget: amount of work driver is allowed to do this pass, in packets
1994 * This function is optimized for cleaning one queue only on a single
1997 static int ixgbe_clean_txonly(struct napi_struct *napi, int budget)
1999 struct ixgbe_q_vector *q_vector =
2000 container_of(napi, struct ixgbe_q_vector, napi);
2001 struct ixgbe_adapter *adapter = q_vector->adapter;
2002 struct ixgbe_ring *tx_ring = NULL;
2006 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
2007 tx_ring = adapter->tx_ring[r_idx];
2008 #ifdef CONFIG_IXGBE_DCA
2009 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2010 ixgbe_update_tx_dca(adapter, tx_ring);
2013 if (!ixgbe_clean_tx_irq(q_vector, tx_ring))
2016 /* If all Tx work done, exit the polling mode */
2017 if (work_done < budget) {
2018 napi_complete(napi);
2019 if (adapter->tx_itr_setting & 1)
2020 ixgbe_set_itr_msix(q_vector);
2021 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2022 ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
2028 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
2031 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2033 set_bit(r_idx, q_vector->rxr_idx);
2034 q_vector->rxr_count++;
2037 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
2040 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2042 set_bit(t_idx, q_vector->txr_idx);
2043 q_vector->txr_count++;
2047 * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
2048 * @adapter: board private structure to initialize
2049 * @vectors: allotted vector count for descriptor rings
2051 * This function maps descriptor rings to the queue-specific vectors
2052 * we were allotted through the MSI-X enabling code. Ideally, we'd have
2053 * one vector per ring/queue, but on a constrained vector budget, we
2054 * group the rings as "efficiently" as possible. You would add new
2055 * mapping configurations in here.
2057 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter,
2061 int rxr_idx = 0, txr_idx = 0;
2062 int rxr_remaining = adapter->num_rx_queues;
2063 int txr_remaining = adapter->num_tx_queues;
2068 /* No mapping required if MSI-X is disabled. */
2069 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2073 * The ideal configuration...
2074 * We have enough vectors to map one per queue.
2076 if (vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
2077 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
2078 map_vector_to_rxq(adapter, v_start, rxr_idx);
2080 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
2081 map_vector_to_txq(adapter, v_start, txr_idx);
2087 * If we don't have enough vectors for a 1-to-1
2088 * mapping, we'll have to group them so there are
2089 * multiple queues per vector.
2091 /* Re-adjusting *qpv takes care of the remainder. */
2092 for (i = v_start; i < vectors; i++) {
2093 rqpv = DIV_ROUND_UP(rxr_remaining, vectors - i);
2094 for (j = 0; j < rqpv; j++) {
2095 map_vector_to_rxq(adapter, i, rxr_idx);
2100 for (i = v_start; i < vectors; i++) {
2101 tqpv = DIV_ROUND_UP(txr_remaining, vectors - i);
2102 for (j = 0; j < tqpv; j++) {
2103 map_vector_to_txq(adapter, i, txr_idx);
2114 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2115 * @adapter: board private structure
2117 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2118 * interrupts from the kernel.
2120 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2122 struct net_device *netdev = adapter->netdev;
2123 irqreturn_t (*handler)(int, void *);
2124 int i, vector, q_vectors, err;
2127 /* Decrement for Other and TCP Timer vectors */
2128 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2130 /* Map the Tx/Rx rings to the vectors we were allotted. */
2131 err = ixgbe_map_rings_to_vectors(adapter, q_vectors);
2135 #define SET_HANDLER(_v) ((!(_v)->rxr_count) ? &ixgbe_msix_clean_tx : \
2136 (!(_v)->txr_count) ? &ixgbe_msix_clean_rx : \
2137 &ixgbe_msix_clean_many)
2138 for (vector = 0; vector < q_vectors; vector++) {
2139 handler = SET_HANDLER(adapter->q_vector[vector]);
2141 if(handler == &ixgbe_msix_clean_rx) {
2142 sprintf(adapter->name[vector], "%s-%s-%d",
2143 netdev->name, "rx", ri++);
2145 else if(handler == &ixgbe_msix_clean_tx) {
2146 sprintf(adapter->name[vector], "%s-%s-%d",
2147 netdev->name, "tx", ti++);
2150 sprintf(adapter->name[vector], "%s-%s-%d",
2151 netdev->name, "TxRx", vector);
2153 err = request_irq(adapter->msix_entries[vector].vector,
2154 handler, 0, adapter->name[vector],
2155 adapter->q_vector[vector]);
2157 e_err(probe, "request_irq failed for MSIX interrupt "
2158 "Error: %d\n", err);
2159 goto free_queue_irqs;
2163 sprintf(adapter->name[vector], "%s:lsc", netdev->name);
2164 err = request_irq(adapter->msix_entries[vector].vector,
2165 ixgbe_msix_lsc, 0, adapter->name[vector], netdev);
2167 e_err(probe, "request_irq for msix_lsc failed: %d\n", err);
2168 goto free_queue_irqs;
2174 for (i = vector - 1; i >= 0; i--)
2175 free_irq(adapter->msix_entries[--vector].vector,
2176 adapter->q_vector[i]);
2177 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2178 pci_disable_msix(adapter->pdev);
2179 kfree(adapter->msix_entries);
2180 adapter->msix_entries = NULL;
2185 static void ixgbe_set_itr(struct ixgbe_adapter *adapter)
2187 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2189 u32 new_itr = q_vector->eitr;
2190 struct ixgbe_ring *rx_ring = adapter->rx_ring[0];
2191 struct ixgbe_ring *tx_ring = adapter->tx_ring[0];
2193 q_vector->tx_itr = ixgbe_update_itr(adapter, new_itr,
2195 tx_ring->total_packets,
2196 tx_ring->total_bytes);
2197 q_vector->rx_itr = ixgbe_update_itr(adapter, new_itr,
2199 rx_ring->total_packets,
2200 rx_ring->total_bytes);
2202 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
2204 switch (current_itr) {
2205 /* counts and packets in update_itr are dependent on these numbers */
2206 case lowest_latency:
2210 new_itr = 20000; /* aka hwitr = ~200 */
2219 if (new_itr != q_vector->eitr) {
2220 /* do an exponential smoothing */
2221 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
2223 /* save the algorithm value here, not the smoothed one */
2224 q_vector->eitr = new_itr;
2226 ixgbe_write_eitr(q_vector);
2231 * ixgbe_irq_enable - Enable default interrupt generation settings
2232 * @adapter: board private structure
2234 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter)
2238 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2239 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2240 mask |= IXGBE_EIMS_GPI_SDP0;
2241 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2242 mask |= IXGBE_EIMS_GPI_SDP1;
2243 if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
2244 mask |= IXGBE_EIMS_ECC;
2245 mask |= IXGBE_EIMS_GPI_SDP1;
2246 mask |= IXGBE_EIMS_GPI_SDP2;
2247 if (adapter->num_vfs)
2248 mask |= IXGBE_EIMS_MAILBOX;
2250 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
2251 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
2252 mask |= IXGBE_EIMS_FLOW_DIR;
2254 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2255 ixgbe_irq_enable_queues(adapter, ~0);
2256 IXGBE_WRITE_FLUSH(&adapter->hw);
2258 if (adapter->num_vfs > 32) {
2259 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
2260 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2265 * ixgbe_intr - legacy mode Interrupt Handler
2266 * @irq: interrupt number
2267 * @data: pointer to a network interface device structure
2269 static irqreturn_t ixgbe_intr(int irq, void *data)
2271 struct net_device *netdev = data;
2272 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2273 struct ixgbe_hw *hw = &adapter->hw;
2274 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2278 * Workaround for silicon errata. Mask the interrupts
2279 * before the read of EICR.
2281 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2283 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2284 * therefore no explict interrupt disable is necessary */
2285 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2287 /* shared interrupt alert!
2288 * make sure interrupts are enabled because the read will
2289 * have disabled interrupts due to EIAM */
2290 ixgbe_irq_enable(adapter);
2291 return IRQ_NONE; /* Not our interrupt */
2294 if (eicr & IXGBE_EICR_LSC)
2295 ixgbe_check_lsc(adapter);
2297 if (hw->mac.type == ixgbe_mac_82599EB)
2298 ixgbe_check_sfp_event(adapter, eicr);
2300 ixgbe_check_fan_failure(adapter, eicr);
2301 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2302 ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)))
2303 schedule_work(&adapter->check_overtemp_task);
2305 if (napi_schedule_prep(&(q_vector->napi))) {
2306 adapter->tx_ring[0]->total_packets = 0;
2307 adapter->tx_ring[0]->total_bytes = 0;
2308 adapter->rx_ring[0]->total_packets = 0;
2309 adapter->rx_ring[0]->total_bytes = 0;
2310 /* would disable interrupts here but EIAM disabled it */
2311 __napi_schedule(&(q_vector->napi));
2317 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
2319 int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2321 for (i = 0; i < q_vectors; i++) {
2322 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
2323 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
2324 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
2325 q_vector->rxr_count = 0;
2326 q_vector->txr_count = 0;
2331 * ixgbe_request_irq - initialize interrupts
2332 * @adapter: board private structure
2334 * Attempts to configure interrupts using the best available
2335 * capabilities of the hardware and kernel.
2337 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2339 struct net_device *netdev = adapter->netdev;
2342 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2343 err = ixgbe_request_msix_irqs(adapter);
2344 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
2345 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2346 netdev->name, netdev);
2348 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2349 netdev->name, netdev);
2353 e_err(probe, "request_irq failed, Error %d\n", err);
2358 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2360 struct net_device *netdev = adapter->netdev;
2362 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2365 q_vectors = adapter->num_msix_vectors;
2368 free_irq(adapter->msix_entries[i].vector, netdev);
2371 for (; i >= 0; i--) {
2372 free_irq(adapter->msix_entries[i].vector,
2373 adapter->q_vector[i]);
2376 ixgbe_reset_q_vectors(adapter);
2378 free_irq(adapter->pdev->irq, netdev);
2383 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2384 * @adapter: board private structure
2386 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2388 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2389 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2391 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2392 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2393 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2394 if (adapter->num_vfs > 32)
2395 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
2397 IXGBE_WRITE_FLUSH(&adapter->hw);
2398 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2400 for (i = 0; i < adapter->num_msix_vectors; i++)
2401 synchronize_irq(adapter->msix_entries[i].vector);
2403 synchronize_irq(adapter->pdev->irq);
2408 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2411 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2413 struct ixgbe_hw *hw = &adapter->hw;
2415 IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
2416 EITR_INTS_PER_SEC_TO_REG(adapter->rx_eitr_param));
2418 ixgbe_set_ivar(adapter, 0, 0, 0);
2419 ixgbe_set_ivar(adapter, 1, 0, 0);
2421 map_vector_to_rxq(adapter, 0, 0);
2422 map_vector_to_txq(adapter, 0, 0);
2424 e_info(hw, "Legacy interrupt IVAR setup done\n");
2428 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2429 * @adapter: board private structure
2430 * @ring: structure containing ring specific data
2432 * Configure the Tx descriptor ring after a reset.
2434 static void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2435 struct ixgbe_ring *ring)
2437 struct ixgbe_hw *hw = &adapter->hw;
2438 u64 tdba = ring->dma;
2439 u16 reg_idx = ring->reg_idx;
2441 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2442 (tdba & DMA_BIT_MASK(32)));
2443 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2444 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2445 ring->count * sizeof(union ixgbe_adv_tx_desc));
2446 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2447 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2448 ring->head = IXGBE_TDH(reg_idx);
2449 ring->tail = IXGBE_TDT(reg_idx);
2453 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
2455 struct ixgbe_hw *hw = &adapter->hw;
2459 if (hw->mac.type == ixgbe_mac_82598EB)
2462 /* disable the arbiter while setting MTQC */
2463 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2464 rttdcs |= IXGBE_RTTDCS_ARBDIS;
2465 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2467 /* set transmit pool layout */
2468 mask = (IXGBE_FLAG_SRIOV_ENABLED | IXGBE_FLAG_DCB_ENABLED);
2469 switch (adapter->flags & mask) {
2471 case (IXGBE_FLAG_SRIOV_ENABLED):
2472 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2473 (IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF));
2476 case (IXGBE_FLAG_DCB_ENABLED):
2477 /* We enable 8 traffic classes, DCB only */
2478 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2479 (IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ));
2483 IXGBE_WRITE_REG(hw, IXGBE_MTQC, IXGBE_MTQC_64Q_1PB);
2487 /* re-enable the arbiter */
2488 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2489 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2493 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2494 * @adapter: board private structure
2496 * Configure the Tx unit of the MAC after a reset.
2498 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2502 /* Setup the HW Tx Head and Tail descriptor pointers */
2503 for (i = 0; i < adapter->num_tx_queues; i++)
2504 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
2506 ixgbe_setup_mtqc(adapter);
2509 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2511 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2512 struct ixgbe_ring *rx_ring)
2516 struct ixgbe_ring_feature *feature = adapter->ring_feature;
2518 index = rx_ring->reg_idx;
2519 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2521 mask = (unsigned long) feature[RING_F_RSS].mask;
2522 index = index & mask;
2524 srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(index));
2526 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
2527 srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
2529 srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
2530 IXGBE_SRRCTL_BSIZEHDR_MASK;
2532 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
2533 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
2534 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2536 srrctl |= (PAGE_SIZE / 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2538 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2540 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
2541 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2542 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2545 IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(index), srrctl);
2548 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2550 struct ixgbe_hw *hw = &adapter->hw;
2551 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2552 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2553 0x6A3E67EA, 0x14364D17, 0x3BED200D};
2554 u32 mrqc = 0, reta = 0;
2559 /* Fill out hash function seeds */
2560 for (i = 0; i < 10; i++)
2561 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2563 /* Fill out redirection table */
2564 for (i = 0, j = 0; i < 128; i++, j++) {
2565 if (j == adapter->ring_feature[RING_F_RSS].indices)
2567 /* reta = 4-byte sliding window of
2568 * 0x00..(indices-1)(indices-1)00..etc. */
2569 reta = (reta << 8) | (j * 0x11);
2571 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2574 /* Disable indicating checksum in descriptor, enables RSS hash */
2575 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2576 rxcsum |= IXGBE_RXCSUM_PCSD;
2577 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2579 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
2580 mask = adapter->flags & IXGBE_FLAG_RSS_ENABLED;
2582 mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2583 #ifdef CONFIG_IXGBE_DCB
2584 | IXGBE_FLAG_DCB_ENABLED
2586 | IXGBE_FLAG_SRIOV_ENABLED
2590 case (IXGBE_FLAG_RSS_ENABLED):
2591 mrqc = IXGBE_MRQC_RSSEN;
2593 case (IXGBE_FLAG_SRIOV_ENABLED):
2594 mrqc = IXGBE_MRQC_VMDQEN;
2596 #ifdef CONFIG_IXGBE_DCB
2597 case (IXGBE_FLAG_DCB_ENABLED):
2598 mrqc = IXGBE_MRQC_RT8TCEN;
2600 #endif /* CONFIG_IXGBE_DCB */
2605 /* Perform hash on these packet types */
2606 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2607 | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2608 | IXGBE_MRQC_RSS_FIELD_IPV6
2609 | IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
2611 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2615 * ixgbe_configure_rscctl - enable RSC for the indicated ring
2616 * @adapter: address of board private structure
2617 * @index: index of ring to set
2619 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter, int index)
2621 struct ixgbe_ring *rx_ring;
2622 struct ixgbe_hw *hw = &adapter->hw;
2627 rx_ring = adapter->rx_ring[index];
2628 j = rx_ring->reg_idx;
2629 rx_buf_len = rx_ring->rx_buf_len;
2630 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(j));
2631 rscctrl |= IXGBE_RSCCTL_RSCEN;
2633 * we must limit the number of descriptors so that the
2634 * total size of max desc * buf_len is not greater
2637 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
2638 #if (MAX_SKB_FRAGS > 16)
2639 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2640 #elif (MAX_SKB_FRAGS > 8)
2641 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2642 #elif (MAX_SKB_FRAGS > 4)
2643 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2645 rscctrl |= IXGBE_RSCCTL_MAXDESC_1;
2648 if (rx_buf_len < IXGBE_RXBUFFER_4096)
2649 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2650 else if (rx_buf_len < IXGBE_RXBUFFER_8192)
2651 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2653 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2655 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(j), rscctrl);
2658 static void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
2659 struct ixgbe_ring *ring)
2661 struct ixgbe_hw *hw = &adapter->hw;
2662 u64 rdba = ring->dma;
2663 u16 reg_idx = ring->reg_idx;
2665 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
2666 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
2667 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
2668 ring->count * sizeof(union ixgbe_adv_rx_desc));
2669 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
2670 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
2671 ring->head = IXGBE_RDH(reg_idx);
2672 ring->tail = IXGBE_RDT(reg_idx);
2675 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
2677 struct ixgbe_hw *hw = &adapter->hw;
2680 /* PSRTYPE must be initialized in non 82598 adapters */
2681 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
2682 IXGBE_PSRTYPE_UDPHDR |
2683 IXGBE_PSRTYPE_IPV4HDR |
2684 IXGBE_PSRTYPE_L2HDR |
2685 IXGBE_PSRTYPE_IPV6HDR;
2687 if (hw->mac.type == ixgbe_mac_82598EB)
2690 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED)
2691 psrtype |= (adapter->num_rx_queues_per_pool << 29);
2693 for (p = 0; p < adapter->num_rx_pools; p++)
2694 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(adapter->num_vfs + p),
2699 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
2700 * @adapter: board private structure
2702 * Configure the Rx unit of the MAC after a reset.
2704 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
2706 struct ixgbe_hw *hw = &adapter->hw;
2707 struct ixgbe_ring *rx_ring;
2708 struct net_device *netdev = adapter->netdev;
2709 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2716 ixgbe_setup_psrtype(adapter);
2718 /* Decide whether to use packet split mode or not */
2719 /* Do not use packet split if we're in SR-IOV Mode */
2720 if (!adapter->num_vfs)
2721 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
2723 /* Set the RX buffer length according to the mode */
2724 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
2725 rx_buf_len = IXGBE_RX_HDR_SIZE;
2727 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
2728 (netdev->mtu <= ETH_DATA_LEN))
2729 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
2731 rx_buf_len = ALIGN(max_frame, 1024);
2734 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2735 if (adapter->netdev->mtu <= ETH_DATA_LEN)
2736 hlreg0 &= ~IXGBE_HLREG0_JUMBOEN;
2738 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2740 if (netdev->features & NETIF_F_FCOE_MTU)
2741 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2743 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
2745 /* disable receives while setting up the descriptors */
2746 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2747 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
2750 * Setup the HW Rx Head and Tail Descriptor Pointers and
2751 * the Base and Length of the Rx Descriptor Ring
2753 for (i = 0; i < adapter->num_rx_queues; i++) {
2754 rx_ring = adapter->rx_ring[i];
2755 rx_ring->rx_buf_len = rx_buf_len;
2757 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)
2758 rx_ring->flags |= IXGBE_RING_RX_PS_ENABLED;
2760 rx_ring->flags &= ~IXGBE_RING_RX_PS_ENABLED;
2763 if (netdev->features & NETIF_F_FCOE_MTU) {
2764 struct ixgbe_ring_feature *f;
2765 f = &adapter->ring_feature[RING_F_FCOE];
2766 if ((i >= f->mask) && (i < f->mask + f->indices)) {
2767 rx_ring->flags &= ~IXGBE_RING_RX_PS_ENABLED;
2768 if (rx_buf_len < IXGBE_FCOE_JUMBO_FRAME_SIZE)
2769 rx_ring->rx_buf_len =
2770 IXGBE_FCOE_JUMBO_FRAME_SIZE;
2774 #endif /* IXGBE_FCOE */
2775 ixgbe_configure_rx_ring(adapter, rx_ring);
2776 ixgbe_configure_srrctl(adapter, rx_ring);
2779 if (hw->mac.type == ixgbe_mac_82598EB) {
2781 * For VMDq support of different descriptor types or
2782 * buffer sizes through the use of multiple SRRCTL
2783 * registers, RDRXCTL.MVMEN must be set to 1
2785 * also, the manual doesn't mention it clearly but DCA hints
2786 * will only use queue 0's tags unless this bit is set. Side
2787 * effects of setting this bit are only that SRRCTL must be
2788 * fully programmed [0..15]
2790 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2791 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
2792 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
2795 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
2797 u32 reg_offset, vf_shift;
2798 u32 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
2799 vt_reg_bits = IXGBE_VMD_CTL_VMDQ_EN
2800 | IXGBE_VT_CTL_REPLEN;
2801 vt_reg_bits |= (adapter->num_vfs <<
2802 IXGBE_VT_CTL_POOL_SHIFT);
2803 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl | vt_reg_bits);
2804 IXGBE_WRITE_REG(hw, IXGBE_MRQC, 0);
2806 vf_shift = adapter->num_vfs % 32;
2807 reg_offset = adapter->num_vfs / 32;
2808 IXGBE_WRITE_REG(hw, IXGBE_VFRE(0), 0);
2809 IXGBE_WRITE_REG(hw, IXGBE_VFRE(1), 0);
2810 IXGBE_WRITE_REG(hw, IXGBE_VFTE(0), 0);
2811 IXGBE_WRITE_REG(hw, IXGBE_VFTE(1), 0);
2812 /* Enable only the PF's pool for Tx/Rx */
2813 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (1 << vf_shift));
2814 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (1 << vf_shift));
2815 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
2816 ixgbe_set_vmolr(hw, adapter->num_vfs, true);
2819 /* Program MRQC for the distribution of queues */
2820 ixgbe_setup_mrqc(adapter);
2822 if (adapter->num_vfs) {
2825 /* Map PF MAC address in RAR Entry 0 to first pool
2827 hw->mac.ops.set_vmdq(hw, 0, adapter->num_vfs);
2829 /* Set up VF register offsets for selected VT Mode, i.e.
2830 * 64 VFs for SR-IOV */
2831 reg = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
2832 reg |= IXGBE_GCR_EXT_SRIOV;
2833 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, reg);
2836 if (hw->mac.type == ixgbe_mac_82599EB) {
2837 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2838 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
2839 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
2840 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
2843 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
2844 /* Enable 82599 HW-RSC */
2845 for (i = 0; i < adapter->num_rx_queues; i++)
2846 ixgbe_configure_rscctl(adapter, i);
2848 /* Disable RSC for ACK packets */
2849 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
2850 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
2854 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2856 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2857 struct ixgbe_hw *hw = &adapter->hw;
2858 int pool_ndx = adapter->num_vfs;
2860 /* add VID to filter table */
2861 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, true);
2864 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2866 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2867 struct ixgbe_hw *hw = &adapter->hw;
2868 int pool_ndx = adapter->num_vfs;
2870 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2871 ixgbe_irq_disable(adapter);
2873 vlan_group_set_device(adapter->vlgrp, vid, NULL);
2875 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2876 ixgbe_irq_enable(adapter);
2878 /* remove VID from filter table */
2879 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, false);
2883 * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
2884 * @adapter: driver data
2886 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
2888 struct ixgbe_hw *hw = &adapter->hw;
2889 u32 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2892 switch (hw->mac.type) {
2893 case ixgbe_mac_82598EB:
2894 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
2895 #ifdef CONFIG_IXGBE_DCB
2896 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
2897 vlnctrl &= ~IXGBE_VLNCTRL_VME;
2899 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2900 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2902 case ixgbe_mac_82599EB:
2903 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
2904 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2905 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2906 #ifdef CONFIG_IXGBE_DCB
2907 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
2910 for (i = 0; i < adapter->num_rx_queues; i++) {
2911 j = adapter->rx_ring[i]->reg_idx;
2912 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
2913 vlnctrl &= ~IXGBE_RXDCTL_VME;
2914 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
2923 * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
2924 * @adapter: driver data
2926 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
2928 struct ixgbe_hw *hw = &adapter->hw;
2929 u32 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2932 switch (hw->mac.type) {
2933 case ixgbe_mac_82598EB:
2934 vlnctrl |= IXGBE_VLNCTRL_VME | IXGBE_VLNCTRL_VFE;
2935 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2936 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2938 case ixgbe_mac_82599EB:
2939 vlnctrl |= IXGBE_VLNCTRL_VFE;
2940 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2941 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2942 for (i = 0; i < adapter->num_rx_queues; i++) {
2943 j = adapter->rx_ring[i]->reg_idx;
2944 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
2945 vlnctrl |= IXGBE_RXDCTL_VME;
2946 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
2954 static void ixgbe_vlan_rx_register(struct net_device *netdev,
2955 struct vlan_group *grp)
2957 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2959 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2960 ixgbe_irq_disable(adapter);
2961 adapter->vlgrp = grp;
2964 * For a DCB driver, always enable VLAN tag stripping so we can
2965 * still receive traffic from a DCB-enabled host even if we're
2968 ixgbe_vlan_filter_enable(adapter);
2970 ixgbe_vlan_rx_add_vid(netdev, 0);
2972 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2973 ixgbe_irq_enable(adapter);
2976 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
2978 ixgbe_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2980 if (adapter->vlgrp) {
2982 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2983 if (!vlan_group_get_device(adapter->vlgrp, vid))
2985 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
2991 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
2992 * @netdev: network interface device structure
2994 * Writes unicast address list to the RAR table.
2995 * Returns: -ENOMEM on failure/insufficient address space
2996 * 0 on no addresses written
2997 * X on writing X addresses to the RAR table
2999 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3001 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3002 struct ixgbe_hw *hw = &adapter->hw;
3003 unsigned int vfn = adapter->num_vfs;
3004 unsigned int rar_entries = hw->mac.num_rar_entries - (vfn + 1);
3007 /* return ENOMEM indicating insufficient memory for addresses */
3008 if (netdev_uc_count(netdev) > rar_entries)
3011 if (!netdev_uc_empty(netdev) && rar_entries) {
3012 struct netdev_hw_addr *ha;
3013 /* return error if we do not support writing to RAR table */
3014 if (!hw->mac.ops.set_rar)
3017 netdev_for_each_uc_addr(ha, netdev) {
3020 hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3025 /* write the addresses in reverse order to avoid write combining */
3026 for (; rar_entries > 0 ; rar_entries--)
3027 hw->mac.ops.clear_rar(hw, rar_entries);
3033 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3034 * @netdev: network interface device structure
3036 * The set_rx_method entry point is called whenever the unicast/multicast
3037 * address list or the network interface flags are updated. This routine is
3038 * responsible for configuring the hardware for proper unicast, multicast and
3041 void ixgbe_set_rx_mode(struct net_device *netdev)
3043 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3044 struct ixgbe_hw *hw = &adapter->hw;
3045 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3048 /* Check for Promiscuous and All Multicast modes */
3050 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3052 /* set all bits that we expect to always be set */
3053 fctrl |= IXGBE_FCTRL_BAM;
3054 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3055 fctrl |= IXGBE_FCTRL_PMCF;
3057 /* clear the bits we are changing the status of */
3058 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3060 if (netdev->flags & IFF_PROMISC) {
3061 hw->addr_ctrl.user_set_promisc = true;
3062 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3063 vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3064 /* don't hardware filter vlans in promisc mode */
3065 ixgbe_vlan_filter_disable(adapter);
3067 if (netdev->flags & IFF_ALLMULTI) {
3068 fctrl |= IXGBE_FCTRL_MPE;
3069 vmolr |= IXGBE_VMOLR_MPE;
3072 * Write addresses to the MTA, if the attempt fails
3073 * then we should just turn on promiscous mode so
3074 * that we can at least receive multicast traffic
3076 hw->mac.ops.update_mc_addr_list(hw, netdev);
3077 vmolr |= IXGBE_VMOLR_ROMPE;
3079 ixgbe_vlan_filter_enable(adapter);
3080 hw->addr_ctrl.user_set_promisc = false;
3082 * Write addresses to available RAR registers, if there is not
3083 * sufficient space to store all the addresses then enable
3084 * unicast promiscous mode
3086 count = ixgbe_write_uc_addr_list(netdev);
3088 fctrl |= IXGBE_FCTRL_UPE;
3089 vmolr |= IXGBE_VMOLR_ROPE;
3093 if (adapter->num_vfs) {
3094 ixgbe_restore_vf_multicasts(adapter);
3095 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(adapter->num_vfs)) &
3096 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3098 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(adapter->num_vfs), vmolr);
3101 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3104 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3107 struct ixgbe_q_vector *q_vector;
3108 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3110 /* legacy and MSI only use one vector */
3111 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3114 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3115 struct napi_struct *napi;
3116 q_vector = adapter->q_vector[q_idx];
3117 napi = &q_vector->napi;
3118 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3119 if (!q_vector->rxr_count || !q_vector->txr_count) {
3120 if (q_vector->txr_count == 1)
3121 napi->poll = &ixgbe_clean_txonly;
3122 else if (q_vector->rxr_count == 1)
3123 napi->poll = &ixgbe_clean_rxonly;
3131 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3134 struct ixgbe_q_vector *q_vector;
3135 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3137 /* legacy and MSI only use one vector */
3138 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3141 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3142 q_vector = adapter->q_vector[q_idx];
3143 napi_disable(&q_vector->napi);
3147 #ifdef CONFIG_IXGBE_DCB
3149 * ixgbe_configure_dcb - Configure DCB hardware
3150 * @adapter: ixgbe adapter struct
3152 * This is called by the driver on open to configure the DCB hardware.
3153 * This is also called by the gennetlink interface when reconfiguring
3156 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3158 struct ixgbe_hw *hw = &adapter->hw;
3162 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3163 if (hw->mac.type == ixgbe_mac_82598EB)
3164 netif_set_gso_max_size(adapter->netdev, 65536);
3168 if (hw->mac.type == ixgbe_mac_82598EB)
3169 netif_set_gso_max_size(adapter->netdev, 32768);
3171 ixgbe_dcb_check_config(&adapter->dcb_cfg);
3172 ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_TX_CONFIG);
3173 ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_RX_CONFIG);
3175 /* reconfigure the hardware */
3176 ixgbe_dcb_hw_config(&adapter->hw, &adapter->dcb_cfg);
3178 for (i = 0; i < adapter->num_tx_queues; i++) {
3179 j = adapter->tx_ring[i]->reg_idx;
3180 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3181 /* PThresh workaround for Tx hang with DFP enabled. */
3183 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
3185 /* Enable VLAN tag insert/strip */
3186 ixgbe_vlan_filter_enable(adapter);
3188 hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
3192 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3194 struct net_device *netdev = adapter->netdev;
3195 struct ixgbe_hw *hw = &adapter->hw;
3198 ixgbe_set_rx_mode(netdev);
3200 ixgbe_restore_vlan(adapter);
3201 #ifdef CONFIG_IXGBE_DCB
3202 ixgbe_configure_dcb(adapter);
3206 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
3207 ixgbe_configure_fcoe(adapter);
3209 #endif /* IXGBE_FCOE */
3210 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3211 for (i = 0; i < adapter->num_tx_queues; i++)
3212 adapter->tx_ring[i]->atr_sample_rate =
3213 adapter->atr_sample_rate;
3214 ixgbe_init_fdir_signature_82599(hw, adapter->fdir_pballoc);
3215 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3216 ixgbe_init_fdir_perfect_82599(hw, adapter->fdir_pballoc);
3219 ixgbe_configure_tx(adapter);
3220 ixgbe_configure_rx(adapter);
3221 for (i = 0; i < adapter->num_rx_queues; i++)
3222 ixgbe_alloc_rx_buffers(adapter, adapter->rx_ring[i],
3223 (adapter->rx_ring[i]->count - 1));
3226 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3228 switch (hw->phy.type) {
3229 case ixgbe_phy_sfp_avago:
3230 case ixgbe_phy_sfp_ftl:
3231 case ixgbe_phy_sfp_intel:
3232 case ixgbe_phy_sfp_unknown:
3233 case ixgbe_phy_sfp_passive_tyco:
3234 case ixgbe_phy_sfp_passive_unknown:
3235 case ixgbe_phy_sfp_active_unknown:
3236 case ixgbe_phy_sfp_ftl_active:
3244 * ixgbe_sfp_link_config - set up SFP+ link
3245 * @adapter: pointer to private adapter struct
3247 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3249 struct ixgbe_hw *hw = &adapter->hw;
3251 if (hw->phy.multispeed_fiber) {
3253 * In multispeed fiber setups, the device may not have
3254 * had a physical connection when the driver loaded.
3255 * If that's the case, the initial link configuration
3256 * couldn't get the MAC into 10G or 1G mode, so we'll
3257 * never have a link status change interrupt fire.
3258 * We need to try and force an autonegotiation
3259 * session, then bring up link.
3261 hw->mac.ops.setup_sfp(hw);
3262 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
3263 schedule_work(&adapter->multispeed_fiber_task);
3266 * Direct Attach Cu and non-multispeed fiber modules
3267 * still need to be configured properly prior to
3270 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_MOD_TASK))
3271 schedule_work(&adapter->sfp_config_module_task);
3276 * ixgbe_non_sfp_link_config - set up non-SFP+ link
3277 * @hw: pointer to private hardware struct
3279 * Returns 0 on success, negative on failure
3281 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3284 bool negotiation, link_up = false;
3285 u32 ret = IXGBE_ERR_LINK_SETUP;
3287 if (hw->mac.ops.check_link)
3288 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3293 if (hw->mac.ops.get_link_capabilities)
3294 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
3298 if (hw->mac.ops.setup_link)
3299 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
3304 #define IXGBE_MAX_RX_DESC_POLL 10
3305 static inline void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
3308 int j = adapter->rx_ring[rxr]->reg_idx;
3311 for (k = 0; k < IXGBE_MAX_RX_DESC_POLL; k++) {
3312 if (IXGBE_READ_REG(&adapter->hw,
3313 IXGBE_RXDCTL(j)) & IXGBE_RXDCTL_ENABLE)
3318 if (k >= IXGBE_MAX_RX_DESC_POLL) {
3319 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
3320 "the polling period\n", rxr);
3322 ixgbe_release_rx_desc(&adapter->hw, adapter->rx_ring[rxr],
3323 (adapter->rx_ring[rxr]->count - 1));
3326 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
3328 struct net_device *netdev = adapter->netdev;
3329 struct ixgbe_hw *hw = &adapter->hw;
3331 int num_rx_rings = adapter->num_rx_queues;
3333 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3334 u32 txdctl, rxdctl, mhadd;
3339 ixgbe_get_hw_control(adapter);
3341 if ((adapter->flags & IXGBE_FLAG_MSIX_ENABLED) ||
3342 (adapter->flags & IXGBE_FLAG_MSI_ENABLED)) {
3343 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3344 gpie = (IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_EIAME |
3345 IXGBE_GPIE_PBA_SUPPORT | IXGBE_GPIE_OCD);
3350 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3351 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
3352 gpie |= IXGBE_GPIE_VTMODE_64;
3354 /* XXX: to interrupt immediately for EICS writes, enable this */
3355 /* gpie |= IXGBE_GPIE_EIMEN; */
3356 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3359 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3361 * use EIAM to auto-mask when MSI-X interrupt is asserted
3362 * this saves a register write for every interrupt
3364 switch (hw->mac.type) {
3365 case ixgbe_mac_82598EB:
3366 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3369 case ixgbe_mac_82599EB:
3370 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
3371 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
3375 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
3376 * specifically only auto mask tx and rx interrupts */
3377 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3380 /* Enable Thermal over heat sensor interrupt */
3381 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
3382 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
3383 gpie |= IXGBE_SDP0_GPIEN;
3384 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3387 /* Enable fan failure interrupt if media type is copper */
3388 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3389 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
3390 gpie |= IXGBE_SDP1_GPIEN;
3391 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3394 if (hw->mac.type == ixgbe_mac_82599EB) {
3395 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
3396 gpie |= IXGBE_SDP1_GPIEN;
3397 gpie |= IXGBE_SDP2_GPIEN;
3398 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3402 /* adjust max frame to be able to do baby jumbo for FCoE */
3403 if ((netdev->features & NETIF_F_FCOE_MTU) &&
3404 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
3405 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3407 #endif /* IXGBE_FCOE */
3408 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
3409 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
3410 mhadd &= ~IXGBE_MHADD_MFS_MASK;
3411 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
3413 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
3416 if (hw->mac.type == ixgbe_mac_82599EB) {
3417 /* DMATXCTL.EN must be set after all Tx queue config is done */
3418 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
3419 dmatxctl |= IXGBE_DMATXCTL_TE;
3420 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
3422 for (i = 0; i < adapter->num_tx_queues; i++) {
3423 j = adapter->tx_ring[i]->reg_idx;
3424 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3425 if (adapter->rx_itr_setting == 0) {
3426 /* cannot set wthresh when itr==0 */
3427 txdctl &= ~0x007F0000;
3429 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
3430 txdctl |= (8 << 16);
3432 txdctl |= IXGBE_TXDCTL_ENABLE;
3433 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
3434 if (hw->mac.type == ixgbe_mac_82599EB) {
3436 /* poll for Tx Enable ready */
3439 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3440 } while (--wait_loop &&
3441 !(txdctl & IXGBE_TXDCTL_ENABLE));
3443 e_err(drv, "Could not enable Tx Queue %d\n", j);
3447 for (i = 0; i < num_rx_rings; i++) {
3448 j = adapter->rx_ring[i]->reg_idx;
3449 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3450 /* enable PTHRESH=32 descriptors (half the internal cache)
3451 * and HTHRESH=0 descriptors (to minimize latency on fetch),
3452 * this also removes a pesky rx_no_buffer_count increment */
3454 rxdctl |= IXGBE_RXDCTL_ENABLE;
3455 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), rxdctl);
3456 if (hw->mac.type == ixgbe_mac_82599EB)
3457 ixgbe_rx_desc_queue_enable(adapter, i);
3459 /* enable all receives */
3460 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3461 if (hw->mac.type == ixgbe_mac_82598EB)
3462 rxdctl |= (IXGBE_RXCTRL_DMBYPS | IXGBE_RXCTRL_RXEN);
3464 rxdctl |= IXGBE_RXCTRL_RXEN;
3465 hw->mac.ops.enable_rx_dma(hw, rxdctl);
3467 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3468 ixgbe_configure_msix(adapter);
3470 ixgbe_configure_msi_and_legacy(adapter);
3472 /* enable the optics */
3473 if (hw->phy.multispeed_fiber)
3474 hw->mac.ops.enable_tx_laser(hw);
3476 clear_bit(__IXGBE_DOWN, &adapter->state);
3477 ixgbe_napi_enable_all(adapter);
3479 /* clear any pending interrupts, may auto mask */
3480 IXGBE_READ_REG(hw, IXGBE_EICR);
3482 ixgbe_irq_enable(adapter);
3485 * If this adapter has a fan, check to see if we had a failure
3486 * before we enabled the interrupt.
3488 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3489 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
3490 if (esdp & IXGBE_ESDP_SDP1)
3491 e_crit(drv, "Fan has stopped, replace the adapter\n");
3495 * For hot-pluggable SFP+ devices, a new SFP+ module may have
3496 * arrived before interrupts were enabled but after probe. Such
3497 * devices wouldn't have their type identified yet. We need to
3498 * kick off the SFP+ module setup first, then try to bring up link.
3499 * If we're not hot-pluggable SFP+, we just need to configure link
3502 if (hw->phy.type == ixgbe_phy_unknown) {
3503 err = hw->phy.ops.identify(hw);
3504 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
3506 * Take the device down and schedule the sfp tasklet
3507 * which will unregister_netdev and log it.
3509 ixgbe_down(adapter);
3510 schedule_work(&adapter->sfp_config_module_task);
3515 if (ixgbe_is_sfp(hw)) {
3516 ixgbe_sfp_link_config(adapter);
3518 err = ixgbe_non_sfp_link_config(hw);
3520 e_err(probe, "link_config FAILED %d\n", err);
3523 for (i = 0; i < adapter->num_tx_queues; i++)
3524 set_bit(__IXGBE_FDIR_INIT_DONE,
3525 &(adapter->tx_ring[i]->reinit_state));
3527 /* enable transmits */
3528 netif_tx_start_all_queues(netdev);
3530 /* bring the link up in the watchdog, this could race with our first
3531 * link up interrupt but shouldn't be a problem */
3532 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3533 adapter->link_check_timeout = jiffies;
3534 mod_timer(&adapter->watchdog_timer, jiffies);
3536 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
3537 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
3538 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
3539 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
3544 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
3546 WARN_ON(in_interrupt());
3547 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
3549 ixgbe_down(adapter);
3551 * If SR-IOV enabled then wait a bit before bringing the adapter
3552 * back up to give the VFs time to respond to the reset. The
3553 * two second wait is based upon the watchdog timer cycle in
3556 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3559 clear_bit(__IXGBE_RESETTING, &adapter->state);
3562 int ixgbe_up(struct ixgbe_adapter *adapter)
3564 /* hardware has been reset, we need to reload some things */
3565 ixgbe_configure(adapter);
3567 return ixgbe_up_complete(adapter);
3570 void ixgbe_reset(struct ixgbe_adapter *adapter)
3572 struct ixgbe_hw *hw = &adapter->hw;
3575 err = hw->mac.ops.init_hw(hw);
3578 case IXGBE_ERR_SFP_NOT_PRESENT:
3580 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
3581 e_dev_err("master disable timed out\n");
3583 case IXGBE_ERR_EEPROM_VERSION:
3584 /* We are running on a pre-production device, log a warning */
3585 e_dev_warn("This device is a pre-production adapter/LOM. "
3586 "Please be aware there may be issuesassociated with "
3587 "your hardware. If you are experiencing problems "
3588 "please contact your Intel or hardware "
3589 "representative who provided you with this "
3593 e_dev_err("Hardware Error: %d\n", err);
3596 /* reprogram the RAR[0] in case user changed it. */
3597 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
3602 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
3603 * @adapter: board private structure
3604 * @rx_ring: ring to free buffers from
3606 static void ixgbe_clean_rx_ring(struct ixgbe_adapter *adapter,
3607 struct ixgbe_ring *rx_ring)
3609 struct pci_dev *pdev = adapter->pdev;
3613 /* Free all the Rx ring sk_buffs */
3615 for (i = 0; i < rx_ring->count; i++) {
3616 struct ixgbe_rx_buffer *rx_buffer_info;
3618 rx_buffer_info = &rx_ring->rx_buffer_info[i];
3619 if (rx_buffer_info->dma) {
3620 dma_unmap_single(&pdev->dev, rx_buffer_info->dma,
3621 rx_ring->rx_buf_len,
3623 rx_buffer_info->dma = 0;
3625 if (rx_buffer_info->skb) {
3626 struct sk_buff *skb = rx_buffer_info->skb;
3627 rx_buffer_info->skb = NULL;
3629 struct sk_buff *this = skb;
3630 if (IXGBE_RSC_CB(this)->delay_unmap) {
3631 dma_unmap_single(&pdev->dev,
3632 IXGBE_RSC_CB(this)->dma,
3633 rx_ring->rx_buf_len,
3635 IXGBE_RSC_CB(this)->dma = 0;
3636 IXGBE_RSC_CB(skb)->delay_unmap = false;
3639 dev_kfree_skb(this);
3642 if (!rx_buffer_info->page)
3644 if (rx_buffer_info->page_dma) {
3645 dma_unmap_page(&pdev->dev, rx_buffer_info->page_dma,
3646 PAGE_SIZE / 2, DMA_FROM_DEVICE);
3647 rx_buffer_info->page_dma = 0;
3649 put_page(rx_buffer_info->page);
3650 rx_buffer_info->page = NULL;
3651 rx_buffer_info->page_offset = 0;
3654 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
3655 memset(rx_ring->rx_buffer_info, 0, size);
3657 /* Zero out the descriptor ring */
3658 memset(rx_ring->desc, 0, rx_ring->size);
3660 rx_ring->next_to_clean = 0;
3661 rx_ring->next_to_use = 0;
3664 writel(0, adapter->hw.hw_addr + rx_ring->head);
3666 writel(0, adapter->hw.hw_addr + rx_ring->tail);
3670 * ixgbe_clean_tx_ring - Free Tx Buffers
3671 * @adapter: board private structure
3672 * @tx_ring: ring to be cleaned
3674 static void ixgbe_clean_tx_ring(struct ixgbe_adapter *adapter,
3675 struct ixgbe_ring *tx_ring)
3677 struct ixgbe_tx_buffer *tx_buffer_info;
3681 /* Free all the Tx ring sk_buffs */
3683 for (i = 0; i < tx_ring->count; i++) {
3684 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3685 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
3688 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
3689 memset(tx_ring->tx_buffer_info, 0, size);
3691 /* Zero out the descriptor ring */
3692 memset(tx_ring->desc, 0, tx_ring->size);
3694 tx_ring->next_to_use = 0;
3695 tx_ring->next_to_clean = 0;
3698 writel(0, adapter->hw.hw_addr + tx_ring->head);
3700 writel(0, adapter->hw.hw_addr + tx_ring->tail);
3704 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
3705 * @adapter: board private structure
3707 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
3711 for (i = 0; i < adapter->num_rx_queues; i++)
3712 ixgbe_clean_rx_ring(adapter, adapter->rx_ring[i]);
3716 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
3717 * @adapter: board private structure
3719 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
3723 for (i = 0; i < adapter->num_tx_queues; i++)
3724 ixgbe_clean_tx_ring(adapter, adapter->tx_ring[i]);
3727 void ixgbe_down(struct ixgbe_adapter *adapter)
3729 struct net_device *netdev = adapter->netdev;
3730 struct ixgbe_hw *hw = &adapter->hw;
3735 /* signal that we are down to the interrupt handler */
3736 set_bit(__IXGBE_DOWN, &adapter->state);
3738 /* disable receive for all VFs and wait one second */
3739 if (adapter->num_vfs) {
3740 /* ping all the active vfs to let them know we are going down */
3741 ixgbe_ping_all_vfs(adapter);
3743 /* Disable all VFTE/VFRE TX/RX */
3744 ixgbe_disable_tx_rx(adapter);
3746 /* Mark all the VFs as inactive */
3747 for (i = 0 ; i < adapter->num_vfs; i++)
3748 adapter->vfinfo[i].clear_to_send = 0;
3751 /* disable receives */
3752 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3753 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3755 IXGBE_WRITE_FLUSH(hw);
3758 netif_tx_stop_all_queues(netdev);
3760 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
3761 del_timer_sync(&adapter->sfp_timer);
3762 del_timer_sync(&adapter->watchdog_timer);
3763 cancel_work_sync(&adapter->watchdog_task);
3765 netif_carrier_off(netdev);
3766 netif_tx_disable(netdev);
3768 ixgbe_irq_disable(adapter);
3770 ixgbe_napi_disable_all(adapter);
3772 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3773 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3774 cancel_work_sync(&adapter->fdir_reinit_task);
3776 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
3777 cancel_work_sync(&adapter->check_overtemp_task);
3779 /* disable transmits in the hardware now that interrupts are off */
3780 for (i = 0; i < adapter->num_tx_queues; i++) {
3781 j = adapter->tx_ring[i]->reg_idx;
3782 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3783 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j),
3784 (txdctl & ~IXGBE_TXDCTL_ENABLE));
3786 /* Disable the Tx DMA engine on 82599 */
3787 if (hw->mac.type == ixgbe_mac_82599EB)
3788 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
3789 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
3790 ~IXGBE_DMATXCTL_TE));
3792 /* power down the optics */
3793 if (hw->phy.multispeed_fiber)
3794 hw->mac.ops.disable_tx_laser(hw);
3796 /* clear n-tuple filters that are cached */
3797 ethtool_ntuple_flush(netdev);
3799 if (!pci_channel_offline(adapter->pdev))
3800 ixgbe_reset(adapter);
3801 ixgbe_clean_all_tx_rings(adapter);
3802 ixgbe_clean_all_rx_rings(adapter);
3804 #ifdef CONFIG_IXGBE_DCA
3805 /* since we reset the hardware DCA settings were cleared */
3806 ixgbe_setup_dca(adapter);
3811 * ixgbe_poll - NAPI Rx polling callback
3812 * @napi: structure for representing this polling device
3813 * @budget: how many packets driver is allowed to clean
3815 * This function is used for legacy and MSI, NAPI mode
3817 static int ixgbe_poll(struct napi_struct *napi, int budget)
3819 struct ixgbe_q_vector *q_vector =
3820 container_of(napi, struct ixgbe_q_vector, napi);
3821 struct ixgbe_adapter *adapter = q_vector->adapter;
3822 int tx_clean_complete, work_done = 0;
3824 #ifdef CONFIG_IXGBE_DCA
3825 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
3826 ixgbe_update_tx_dca(adapter, adapter->tx_ring[0]);
3827 ixgbe_update_rx_dca(adapter, adapter->rx_ring[0]);
3831 tx_clean_complete = ixgbe_clean_tx_irq(q_vector, adapter->tx_ring[0]);
3832 ixgbe_clean_rx_irq(q_vector, adapter->rx_ring[0], &work_done, budget);
3834 if (!tx_clean_complete)
3837 /* If budget not fully consumed, exit the polling mode */
3838 if (work_done < budget) {
3839 napi_complete(napi);
3840 if (adapter->rx_itr_setting & 1)
3841 ixgbe_set_itr(adapter);
3842 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3843 ixgbe_irq_enable_queues(adapter, IXGBE_EIMS_RTX_QUEUE);
3849 * ixgbe_tx_timeout - Respond to a Tx Hang
3850 * @netdev: network interface device structure
3852 static void ixgbe_tx_timeout(struct net_device *netdev)
3854 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3856 /* Do the reset outside of interrupt context */
3857 schedule_work(&adapter->reset_task);
3860 static void ixgbe_reset_task(struct work_struct *work)
3862 struct ixgbe_adapter *adapter;
3863 adapter = container_of(work, struct ixgbe_adapter, reset_task);
3865 /* If we're already down or resetting, just bail */
3866 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
3867 test_bit(__IXGBE_RESETTING, &adapter->state))
3870 adapter->tx_timeout_count++;
3872 ixgbe_dump(adapter);
3873 netdev_err(adapter->netdev, "Reset adapter\n");
3874 ixgbe_reinit_locked(adapter);
3877 #ifdef CONFIG_IXGBE_DCB
3878 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
3881 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_DCB];
3883 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
3887 adapter->num_rx_queues = f->indices;
3888 adapter->num_tx_queues = f->indices;
3896 * ixgbe_set_rss_queues: Allocate queues for RSS
3897 * @adapter: board private structure to initialize
3899 * This is our "base" multiqueue mode. RSS (Receive Side Scaling) will try
3900 * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
3903 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
3906 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
3908 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3910 adapter->num_rx_queues = f->indices;
3911 adapter->num_tx_queues = f->indices;
3921 * ixgbe_set_fdir_queues: Allocate queues for Flow Director
3922 * @adapter: board private structure to initialize
3924 * Flow Director is an advanced Rx filter, attempting to get Rx flows back
3925 * to the original CPU that initiated the Tx session. This runs in addition
3926 * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
3927 * Rx load across CPUs using RSS.
3930 static bool inline ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
3933 struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
3935 f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
3938 /* Flow Director must have RSS enabled */
3939 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
3940 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3941 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)))) {
3942 adapter->num_tx_queues = f_fdir->indices;
3943 adapter->num_rx_queues = f_fdir->indices;
3946 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
3947 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
3954 * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
3955 * @adapter: board private structure to initialize
3957 * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
3958 * The ring feature mask is not used as a mask for FCoE, as it can take any 8
3959 * rx queues out of the max number of rx queues, instead, it is used as the
3960 * index of the first rx queue used by FCoE.
3963 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
3966 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
3968 f->indices = min((int)num_online_cpus(), f->indices);
3969 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
3970 adapter->num_rx_queues = 1;
3971 adapter->num_tx_queues = 1;
3972 #ifdef CONFIG_IXGBE_DCB
3973 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3974 e_info(probe, "FCoE enabled with DCB\n");
3975 ixgbe_set_dcb_queues(adapter);
3978 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3979 e_info(probe, "FCoE enabled with RSS\n");
3980 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
3981 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
3982 ixgbe_set_fdir_queues(adapter);
3984 ixgbe_set_rss_queues(adapter);
3986 /* adding FCoE rx rings to the end */
3987 f->mask = adapter->num_rx_queues;
3988 adapter->num_rx_queues += f->indices;
3989 adapter->num_tx_queues += f->indices;
3997 #endif /* IXGBE_FCOE */
3999 * ixgbe_set_sriov_queues: Allocate queues for IOV use
4000 * @adapter: board private structure to initialize
4002 * IOV doesn't actually use anything, so just NAK the
4003 * request for now and let the other queue routines
4004 * figure out what to do.
4006 static inline bool ixgbe_set_sriov_queues(struct ixgbe_adapter *adapter)
4012 * ixgbe_set_num_queues: Allocate queues for device, feature dependant
4013 * @adapter: board private structure to initialize
4015 * This is the top level queue allocation routine. The order here is very
4016 * important, starting with the "most" number of features turned on at once,
4017 * and ending with the smallest set of features. This way large combinations
4018 * can be allocated if they're turned on, and smaller combinations are the
4019 * fallthrough conditions.
4022 static void ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
4024 /* Start with base case */
4025 adapter->num_rx_queues = 1;
4026 adapter->num_tx_queues = 1;
4027 adapter->num_rx_pools = adapter->num_rx_queues;
4028 adapter->num_rx_queues_per_pool = 1;
4030 if (ixgbe_set_sriov_queues(adapter))
4034 if (ixgbe_set_fcoe_queues(adapter))
4037 #endif /* IXGBE_FCOE */
4038 #ifdef CONFIG_IXGBE_DCB
4039 if (ixgbe_set_dcb_queues(adapter))
4043 if (ixgbe_set_fdir_queues(adapter))
4046 if (ixgbe_set_rss_queues(adapter))
4049 /* fallback to base case */
4050 adapter->num_rx_queues = 1;
4051 adapter->num_tx_queues = 1;
4054 /* Notify the stack of the (possibly) reduced Tx Queue count. */
4055 netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
4058 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
4061 int err, vector_threshold;
4063 /* We'll want at least 3 (vector_threshold):
4066 * 3) Other (Link Status Change, etc.)
4067 * 4) TCP Timer (optional)
4069 vector_threshold = MIN_MSIX_COUNT;
4071 /* The more we get, the more we will assign to Tx/Rx Cleanup
4072 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
4073 * Right now, we simply care about how many we'll get; we'll
4074 * set them up later while requesting irq's.
4076 while (vectors >= vector_threshold) {
4077 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
4079 if (!err) /* Success in acquiring all requested vectors. */
4082 vectors = 0; /* Nasty failure, quit now */
4083 else /* err == number of vectors we should try again with */
4087 if (vectors < vector_threshold) {
4088 /* Can't allocate enough MSI-X interrupts? Oh well.
4089 * This just means we'll go with either a single MSI
4090 * vector or fall back to legacy interrupts.
4092 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4093 "Unable to allocate MSI-X interrupts\n");
4094 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4095 kfree(adapter->msix_entries);
4096 adapter->msix_entries = NULL;
4098 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
4100 * Adjust for only the vectors we'll use, which is minimum
4101 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
4102 * vectors we were allocated.
4104 adapter->num_msix_vectors = min(vectors,
4105 adapter->max_msix_q_vectors + NON_Q_VECTORS);
4110 * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
4111 * @adapter: board private structure to initialize
4113 * Cache the descriptor ring offsets for RSS to the assigned rings.
4116 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
4121 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4122 for (i = 0; i < adapter->num_rx_queues; i++)
4123 adapter->rx_ring[i]->reg_idx = i;
4124 for (i = 0; i < adapter->num_tx_queues; i++)
4125 adapter->tx_ring[i]->reg_idx = i;
4134 #ifdef CONFIG_IXGBE_DCB
4136 * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
4137 * @adapter: board private structure to initialize
4139 * Cache the descriptor ring offsets for DCB to the assigned rings.
4142 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
4146 int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
4148 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4149 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
4150 /* the number of queues is assumed to be symmetric */
4151 for (i = 0; i < dcb_i; i++) {
4152 adapter->rx_ring[i]->reg_idx = i << 3;
4153 adapter->tx_ring[i]->reg_idx = i << 2;
4156 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
4159 * Tx TC0 starts at: descriptor queue 0
4160 * Tx TC1 starts at: descriptor queue 32
4161 * Tx TC2 starts at: descriptor queue 64
4162 * Tx TC3 starts at: descriptor queue 80
4163 * Tx TC4 starts at: descriptor queue 96
4164 * Tx TC5 starts at: descriptor queue 104
4165 * Tx TC6 starts at: descriptor queue 112
4166 * Tx TC7 starts at: descriptor queue 120
4168 * Rx TC0-TC7 are offset by 16 queues each
4170 for (i = 0; i < 3; i++) {
4171 adapter->tx_ring[i]->reg_idx = i << 5;
4172 adapter->rx_ring[i]->reg_idx = i << 4;
4174 for ( ; i < 5; i++) {
4175 adapter->tx_ring[i]->reg_idx =
4177 adapter->rx_ring[i]->reg_idx = i << 4;
4179 for ( ; i < dcb_i; i++) {
4180 adapter->tx_ring[i]->reg_idx =
4182 adapter->rx_ring[i]->reg_idx = i << 4;
4186 } else if (dcb_i == 4) {
4188 * Tx TC0 starts at: descriptor queue 0
4189 * Tx TC1 starts at: descriptor queue 64
4190 * Tx TC2 starts at: descriptor queue 96
4191 * Tx TC3 starts at: descriptor queue 112
4193 * Rx TC0-TC3 are offset by 32 queues each
4195 adapter->tx_ring[0]->reg_idx = 0;
4196 adapter->tx_ring[1]->reg_idx = 64;
4197 adapter->tx_ring[2]->reg_idx = 96;
4198 adapter->tx_ring[3]->reg_idx = 112;
4199 for (i = 0 ; i < dcb_i; i++)
4200 adapter->rx_ring[i]->reg_idx = i << 5;
4218 * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
4219 * @adapter: board private structure to initialize
4221 * Cache the descriptor ring offsets for Flow Director to the assigned rings.
4224 static bool inline ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
4229 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
4230 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4231 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))) {
4232 for (i = 0; i < adapter->num_rx_queues; i++)
4233 adapter->rx_ring[i]->reg_idx = i;
4234 for (i = 0; i < adapter->num_tx_queues; i++)
4235 adapter->tx_ring[i]->reg_idx = i;
4244 * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
4245 * @adapter: board private structure to initialize
4247 * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
4250 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
4252 int i, fcoe_rx_i = 0, fcoe_tx_i = 0;
4254 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4256 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
4257 #ifdef CONFIG_IXGBE_DCB
4258 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4259 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
4261 ixgbe_cache_ring_dcb(adapter);
4262 /* find out queues in TC for FCoE */
4263 fcoe_rx_i = adapter->rx_ring[fcoe->tc]->reg_idx + 1;
4264 fcoe_tx_i = adapter->tx_ring[fcoe->tc]->reg_idx + 1;
4266 * In 82599, the number of Tx queues for each traffic
4267 * class for both 8-TC and 4-TC modes are:
4268 * TCs : TC0 TC1 TC2 TC3 TC4 TC5 TC6 TC7
4269 * 8 TCs: 32 32 16 16 8 8 8 8
4270 * 4 TCs: 64 64 32 32
4271 * We have max 8 queues for FCoE, where 8 the is
4272 * FCoE redirection table size. If TC for FCoE is
4273 * less than or equal to TC3, we have enough queues
4274 * to add max of 8 queues for FCoE, so we start FCoE
4275 * tx descriptor from the next one, i.e., reg_idx + 1.
4276 * If TC for FCoE is above TC3, implying 8 TC mode,
4277 * and we need 8 for FCoE, we have to take all queues
4278 * in that traffic class for FCoE.
4280 if ((f->indices == IXGBE_FCRETA_SIZE) && (fcoe->tc > 3))
4283 #endif /* CONFIG_IXGBE_DCB */
4284 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4285 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4286 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
4287 ixgbe_cache_ring_fdir(adapter);
4289 ixgbe_cache_ring_rss(adapter);
4291 fcoe_rx_i = f->mask;
4292 fcoe_tx_i = f->mask;
4294 for (i = 0; i < f->indices; i++, fcoe_rx_i++, fcoe_tx_i++) {
4295 adapter->rx_ring[f->mask + i]->reg_idx = fcoe_rx_i;
4296 adapter->tx_ring[f->mask + i]->reg_idx = fcoe_tx_i;
4303 #endif /* IXGBE_FCOE */
4305 * ixgbe_cache_ring_sriov - Descriptor ring to register mapping for sriov
4306 * @adapter: board private structure to initialize
4308 * SR-IOV doesn't use any descriptor rings but changes the default if
4309 * no other mapping is used.
4312 static inline bool ixgbe_cache_ring_sriov(struct ixgbe_adapter *adapter)
4314 adapter->rx_ring[0]->reg_idx = adapter->num_vfs * 2;
4315 adapter->tx_ring[0]->reg_idx = adapter->num_vfs * 2;
4316 if (adapter->num_vfs)
4323 * ixgbe_cache_ring_register - Descriptor ring to register mapping
4324 * @adapter: board private structure to initialize
4326 * Once we know the feature-set enabled for the device, we'll cache
4327 * the register offset the descriptor ring is assigned to.
4329 * Note, the order the various feature calls is important. It must start with
4330 * the "most" features enabled at the same time, then trickle down to the
4331 * least amount of features turned on at once.
4333 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
4335 /* start with default case */
4336 adapter->rx_ring[0]->reg_idx = 0;
4337 adapter->tx_ring[0]->reg_idx = 0;
4339 if (ixgbe_cache_ring_sriov(adapter))
4343 if (ixgbe_cache_ring_fcoe(adapter))
4346 #endif /* IXGBE_FCOE */
4347 #ifdef CONFIG_IXGBE_DCB
4348 if (ixgbe_cache_ring_dcb(adapter))
4352 if (ixgbe_cache_ring_fdir(adapter))
4355 if (ixgbe_cache_ring_rss(adapter))
4360 * ixgbe_alloc_queues - Allocate memory for all rings
4361 * @adapter: board private structure to initialize
4363 * We allocate one ring per queue at run-time since we don't know the
4364 * number of queues at compile-time. The polling_netdev array is
4365 * intended for Multiqueue, but should work fine with a single queue.
4367 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
4370 int orig_node = adapter->node;
4372 for (i = 0; i < adapter->num_tx_queues; i++) {
4373 struct ixgbe_ring *ring = adapter->tx_ring[i];
4374 if (orig_node == -1) {
4375 int cur_node = next_online_node(adapter->node);
4376 if (cur_node == MAX_NUMNODES)
4377 cur_node = first_online_node;
4378 adapter->node = cur_node;
4380 ring = kzalloc_node(sizeof(struct ixgbe_ring), GFP_KERNEL,
4383 ring = kzalloc(sizeof(struct ixgbe_ring), GFP_KERNEL);
4385 goto err_tx_ring_allocation;
4386 ring->count = adapter->tx_ring_count;
4387 ring->queue_index = i;
4388 ring->numa_node = adapter->node;
4390 adapter->tx_ring[i] = ring;
4393 /* Restore the adapter's original node */
4394 adapter->node = orig_node;
4396 for (i = 0; i < adapter->num_rx_queues; i++) {
4397 struct ixgbe_ring *ring = adapter->rx_ring[i];
4398 if (orig_node == -1) {
4399 int cur_node = next_online_node(adapter->node);
4400 if (cur_node == MAX_NUMNODES)
4401 cur_node = first_online_node;
4402 adapter->node = cur_node;
4404 ring = kzalloc_node(sizeof(struct ixgbe_ring), GFP_KERNEL,
4407 ring = kzalloc(sizeof(struct ixgbe_ring), GFP_KERNEL);
4409 goto err_rx_ring_allocation;
4410 ring->count = adapter->rx_ring_count;
4411 ring->queue_index = i;
4412 ring->numa_node = adapter->node;
4414 adapter->rx_ring[i] = ring;
4417 /* Restore the adapter's original node */
4418 adapter->node = orig_node;
4420 ixgbe_cache_ring_register(adapter);
4424 err_rx_ring_allocation:
4425 for (i = 0; i < adapter->num_tx_queues; i++)
4426 kfree(adapter->tx_ring[i]);
4427 err_tx_ring_allocation:
4432 * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
4433 * @adapter: board private structure to initialize
4435 * Attempt to configure the interrupts using the best available
4436 * capabilities of the hardware and the kernel.
4438 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
4440 struct ixgbe_hw *hw = &adapter->hw;
4442 int vector, v_budget;
4445 * It's easy to be greedy for MSI-X vectors, but it really
4446 * doesn't do us much good if we have a lot more vectors
4447 * than CPU's. So let's be conservative and only ask for
4448 * (roughly) the same number of vectors as there are CPU's.
4450 v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
4451 (int)num_online_cpus()) + NON_Q_VECTORS;
4454 * At the same time, hardware can only support a maximum of
4455 * hw.mac->max_msix_vectors vectors. With features
4456 * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
4457 * descriptor queues supported by our device. Thus, we cap it off in
4458 * those rare cases where the cpu count also exceeds our vector limit.
4460 v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
4462 /* A failure in MSI-X entry allocation isn't fatal, but it does
4463 * mean we disable MSI-X capabilities of the adapter. */
4464 adapter->msix_entries = kcalloc(v_budget,
4465 sizeof(struct msix_entry), GFP_KERNEL);
4466 if (adapter->msix_entries) {
4467 for (vector = 0; vector < v_budget; vector++)
4468 adapter->msix_entries[vector].entry = vector;
4470 ixgbe_acquire_msix_vectors(adapter, v_budget);
4472 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4476 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
4477 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
4478 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4479 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4480 adapter->atr_sample_rate = 0;
4481 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4482 ixgbe_disable_sriov(adapter);
4484 ixgbe_set_num_queues(adapter);
4486 err = pci_enable_msi(adapter->pdev);
4488 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
4490 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4491 "Unable to allocate MSI interrupt, "
4492 "falling back to legacy. Error: %d\n", err);
4502 * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
4503 * @adapter: board private structure to initialize
4505 * We allocate one q_vector per queue interrupt. If allocation fails we
4508 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
4510 int q_idx, num_q_vectors;
4511 struct ixgbe_q_vector *q_vector;
4513 int (*poll)(struct napi_struct *, int);
4515 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4516 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4517 napi_vectors = adapter->num_rx_queues;
4518 poll = &ixgbe_clean_rxtx_many;
4525 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4526 q_vector = kzalloc_node(sizeof(struct ixgbe_q_vector),
4527 GFP_KERNEL, adapter->node);
4529 q_vector = kzalloc(sizeof(struct ixgbe_q_vector),
4533 q_vector->adapter = adapter;
4534 if (q_vector->txr_count && !q_vector->rxr_count)
4535 q_vector->eitr = adapter->tx_eitr_param;
4537 q_vector->eitr = adapter->rx_eitr_param;
4538 q_vector->v_idx = q_idx;
4539 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
4540 adapter->q_vector[q_idx] = q_vector;
4548 q_vector = adapter->q_vector[q_idx];
4549 netif_napi_del(&q_vector->napi);
4551 adapter->q_vector[q_idx] = NULL;
4557 * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
4558 * @adapter: board private structure to initialize
4560 * This function frees the memory allocated to the q_vectors. In addition if
4561 * NAPI is enabled it will delete any references to the NAPI struct prior
4562 * to freeing the q_vector.
4564 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
4566 int q_idx, num_q_vectors;
4568 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4569 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4573 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4574 struct ixgbe_q_vector *q_vector = adapter->q_vector[q_idx];
4575 adapter->q_vector[q_idx] = NULL;
4576 netif_napi_del(&q_vector->napi);
4581 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
4583 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4584 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4585 pci_disable_msix(adapter->pdev);
4586 kfree(adapter->msix_entries);
4587 adapter->msix_entries = NULL;
4588 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
4589 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
4590 pci_disable_msi(adapter->pdev);
4595 * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
4596 * @adapter: board private structure to initialize
4598 * We determine which interrupt scheme to use based on...
4599 * - Kernel support (MSI, MSI-X)
4600 * - which can be user-defined (via MODULE_PARAM)
4601 * - Hardware queue count (num_*_queues)
4602 * - defined by miscellaneous hardware support/features (RSS, etc.)
4604 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
4608 /* Number of supported queues */
4609 ixgbe_set_num_queues(adapter);
4611 err = ixgbe_set_interrupt_capability(adapter);
4613 e_dev_err("Unable to setup interrupt capabilities\n");
4614 goto err_set_interrupt;
4617 err = ixgbe_alloc_q_vectors(adapter);
4619 e_dev_err("Unable to allocate memory for queue vectors\n");
4620 goto err_alloc_q_vectors;
4623 err = ixgbe_alloc_queues(adapter);
4625 e_dev_err("Unable to allocate memory for queues\n");
4626 goto err_alloc_queues;
4629 e_dev_info("Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
4630 (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
4631 adapter->num_rx_queues, adapter->num_tx_queues);
4633 set_bit(__IXGBE_DOWN, &adapter->state);
4638 ixgbe_free_q_vectors(adapter);
4639 err_alloc_q_vectors:
4640 ixgbe_reset_interrupt_capability(adapter);
4646 * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
4647 * @adapter: board private structure to clear interrupt scheme on
4649 * We go through and clear interrupt specific resources and reset the structure
4650 * to pre-load conditions
4652 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
4656 for (i = 0; i < adapter->num_tx_queues; i++) {
4657 kfree(adapter->tx_ring[i]);
4658 adapter->tx_ring[i] = NULL;
4660 for (i = 0; i < adapter->num_rx_queues; i++) {
4661 kfree(adapter->rx_ring[i]);
4662 adapter->rx_ring[i] = NULL;
4665 ixgbe_free_q_vectors(adapter);
4666 ixgbe_reset_interrupt_capability(adapter);
4670 * ixgbe_sfp_timer - worker thread to find a missing module
4671 * @data: pointer to our adapter struct
4673 static void ixgbe_sfp_timer(unsigned long data)
4675 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
4678 * Do the sfp_timer outside of interrupt context due to the
4679 * delays that sfp+ detection requires
4681 schedule_work(&adapter->sfp_task);
4685 * ixgbe_sfp_task - worker thread to find a missing module
4686 * @work: pointer to work_struct containing our data
4688 static void ixgbe_sfp_task(struct work_struct *work)
4690 struct ixgbe_adapter *adapter = container_of(work,
4691 struct ixgbe_adapter,
4693 struct ixgbe_hw *hw = &adapter->hw;
4695 if ((hw->phy.type == ixgbe_phy_nl) &&
4696 (hw->phy.sfp_type == ixgbe_sfp_type_not_present)) {
4697 s32 ret = hw->phy.ops.identify_sfp(hw);
4698 if (ret == IXGBE_ERR_SFP_NOT_PRESENT)
4700 ret = hw->phy.ops.reset(hw);
4701 if (ret == IXGBE_ERR_SFP_NOT_SUPPORTED) {
4702 e_dev_err("failed to initialize because an unsupported "
4703 "SFP+ module type was detected.\n");
4704 e_dev_err("Reload the driver after installing a "
4705 "supported module.\n");
4706 unregister_netdev(adapter->netdev);
4708 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
4710 /* don't need this routine any more */
4711 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
4715 if (test_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state))
4716 mod_timer(&adapter->sfp_timer,
4717 round_jiffies(jiffies + (2 * HZ)));
4721 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4722 * @adapter: board private structure to initialize
4724 * ixgbe_sw_init initializes the Adapter private data structure.
4725 * Fields are initialized based on PCI device information and
4726 * OS network device settings (MTU size).
4728 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
4730 struct ixgbe_hw *hw = &adapter->hw;
4731 struct pci_dev *pdev = adapter->pdev;
4732 struct net_device *dev = adapter->netdev;
4734 #ifdef CONFIG_IXGBE_DCB
4736 struct tc_configuration *tc;
4739 /* PCI config space info */
4741 hw->vendor_id = pdev->vendor;
4742 hw->device_id = pdev->device;
4743 hw->revision_id = pdev->revision;
4744 hw->subsystem_vendor_id = pdev->subsystem_vendor;
4745 hw->subsystem_device_id = pdev->subsystem_device;
4747 /* Set capability flags */
4748 rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
4749 adapter->ring_feature[RING_F_RSS].indices = rss;
4750 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
4751 adapter->ring_feature[RING_F_DCB].indices = IXGBE_MAX_DCB_INDICES;
4752 if (hw->mac.type == ixgbe_mac_82598EB) {
4753 if (hw->device_id == IXGBE_DEV_ID_82598AT)
4754 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4755 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
4756 } else if (hw->mac.type == ixgbe_mac_82599EB) {
4757 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
4758 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4759 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4760 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
4761 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4762 if (dev->features & NETIF_F_NTUPLE) {
4763 /* Flow Director perfect filter enabled */
4764 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4765 adapter->atr_sample_rate = 0;
4766 spin_lock_init(&adapter->fdir_perfect_lock);
4768 /* Flow Director hash filters enabled */
4769 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
4770 adapter->atr_sample_rate = 20;
4772 adapter->ring_feature[RING_F_FDIR].indices =
4773 IXGBE_MAX_FDIR_INDICES;
4774 adapter->fdir_pballoc = 0;
4776 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4777 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4778 adapter->ring_feature[RING_F_FCOE].indices = 0;
4779 #ifdef CONFIG_IXGBE_DCB
4780 /* Default traffic class to use for FCoE */
4781 adapter->fcoe.tc = IXGBE_FCOE_DEFTC;
4782 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
4784 #endif /* IXGBE_FCOE */
4787 #ifdef CONFIG_IXGBE_DCB
4788 /* Configure DCB traffic classes */
4789 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
4790 tc = &adapter->dcb_cfg.tc_config[j];
4791 tc->path[DCB_TX_CONFIG].bwg_id = 0;
4792 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
4793 tc->path[DCB_RX_CONFIG].bwg_id = 0;
4794 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
4795 tc->dcb_pfc = pfc_disabled;
4797 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
4798 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
4799 adapter->dcb_cfg.rx_pba_cfg = pba_equal;
4800 adapter->dcb_cfg.pfc_mode_enable = false;
4801 adapter->dcb_cfg.round_robin_enable = false;
4802 adapter->dcb_set_bitmap = 0x00;
4803 ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
4804 adapter->ring_feature[RING_F_DCB].indices);
4808 /* default flow control settings */
4809 hw->fc.requested_mode = ixgbe_fc_full;
4810 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
4812 adapter->last_lfc_mode = hw->fc.current_mode;
4814 hw->fc.high_water = IXGBE_DEFAULT_FCRTH;
4815 hw->fc.low_water = IXGBE_DEFAULT_FCRTL;
4816 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
4817 hw->fc.send_xon = true;
4818 hw->fc.disable_fc_autoneg = false;
4820 /* enable itr by default in dynamic mode */
4821 adapter->rx_itr_setting = 1;
4822 adapter->rx_eitr_param = 20000;
4823 adapter->tx_itr_setting = 1;
4824 adapter->tx_eitr_param = 10000;
4826 /* set defaults for eitr in MegaBytes */
4827 adapter->eitr_low = 10;
4828 adapter->eitr_high = 20;
4830 /* set default ring sizes */
4831 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
4832 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
4834 /* initialize eeprom parameters */
4835 if (ixgbe_init_eeprom_params_generic(hw)) {
4836 e_dev_err("EEPROM initialization failed\n");
4840 /* enable rx csum by default */
4841 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
4843 /* get assigned NUMA node */
4844 adapter->node = dev_to_node(&pdev->dev);
4846 set_bit(__IXGBE_DOWN, &adapter->state);
4852 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
4853 * @adapter: board private structure
4854 * @tx_ring: tx descriptor ring (for a specific queue) to setup
4856 * Return 0 on success, negative on failure
4858 int ixgbe_setup_tx_resources(struct ixgbe_adapter *adapter,
4859 struct ixgbe_ring *tx_ring)
4861 struct pci_dev *pdev = adapter->pdev;
4864 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4865 tx_ring->tx_buffer_info = vmalloc_node(size, tx_ring->numa_node);
4866 if (!tx_ring->tx_buffer_info)
4867 tx_ring->tx_buffer_info = vmalloc(size);
4868 if (!tx_ring->tx_buffer_info)
4870 memset(tx_ring->tx_buffer_info, 0, size);
4872 /* round up to nearest 4K */
4873 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
4874 tx_ring->size = ALIGN(tx_ring->size, 4096);
4876 tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
4877 &tx_ring->dma, GFP_KERNEL);
4881 tx_ring->next_to_use = 0;
4882 tx_ring->next_to_clean = 0;
4883 tx_ring->work_limit = tx_ring->count;
4887 vfree(tx_ring->tx_buffer_info);
4888 tx_ring->tx_buffer_info = NULL;
4889 e_err(probe, "Unable to allocate memory for the Tx descriptor ring\n");
4894 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
4895 * @adapter: board private structure
4897 * If this function returns with an error, then it's possible one or
4898 * more of the rings is populated (while the rest are not). It is the
4899 * callers duty to clean those orphaned rings.
4901 * Return 0 on success, negative on failure
4903 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
4907 for (i = 0; i < adapter->num_tx_queues; i++) {
4908 err = ixgbe_setup_tx_resources(adapter, adapter->tx_ring[i]);
4911 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
4919 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
4920 * @adapter: board private structure
4921 * @rx_ring: rx descriptor ring (for a specific queue) to setup
4923 * Returns 0 on success, negative on failure
4925 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
4926 struct ixgbe_ring *rx_ring)
4928 struct pci_dev *pdev = adapter->pdev;
4931 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4932 rx_ring->rx_buffer_info = vmalloc_node(size, adapter->node);
4933 if (!rx_ring->rx_buffer_info)
4934 rx_ring->rx_buffer_info = vmalloc(size);
4935 if (!rx_ring->rx_buffer_info) {
4936 e_err(probe, "vmalloc allocation failed for the Rx "
4937 "descriptor ring\n");
4940 memset(rx_ring->rx_buffer_info, 0, size);
4942 /* Round up to nearest 4K */
4943 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
4944 rx_ring->size = ALIGN(rx_ring->size, 4096);
4946 rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
4947 &rx_ring->dma, GFP_KERNEL);
4949 if (!rx_ring->desc) {
4950 e_err(probe, "Memory allocation failed for the Rx "
4951 "descriptor ring\n");
4952 vfree(rx_ring->rx_buffer_info);
4956 rx_ring->next_to_clean = 0;
4957 rx_ring->next_to_use = 0;
4966 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
4967 * @adapter: board private structure
4969 * If this function returns with an error, then it's possible one or
4970 * more of the rings is populated (while the rest are not). It is the
4971 * callers duty to clean those orphaned rings.
4973 * Return 0 on success, negative on failure
4976 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
4980 for (i = 0; i < adapter->num_rx_queues; i++) {
4981 err = ixgbe_setup_rx_resources(adapter, adapter->rx_ring[i]);
4984 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
4992 * ixgbe_free_tx_resources - Free Tx Resources per Queue
4993 * @adapter: board private structure
4994 * @tx_ring: Tx descriptor ring for a specific queue
4996 * Free all transmit software resources
4998 void ixgbe_free_tx_resources(struct ixgbe_adapter *adapter,
4999 struct ixgbe_ring *tx_ring)
5001 struct pci_dev *pdev = adapter->pdev;
5003 ixgbe_clean_tx_ring(adapter, tx_ring);
5005 vfree(tx_ring->tx_buffer_info);
5006 tx_ring->tx_buffer_info = NULL;
5008 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
5011 tx_ring->desc = NULL;
5015 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5016 * @adapter: board private structure
5018 * Free all transmit software resources
5020 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
5024 for (i = 0; i < adapter->num_tx_queues; i++)
5025 if (adapter->tx_ring[i]->desc)
5026 ixgbe_free_tx_resources(adapter, adapter->tx_ring[i]);
5030 * ixgbe_free_rx_resources - Free Rx Resources
5031 * @adapter: board private structure
5032 * @rx_ring: ring to clean the resources from
5034 * Free all receive software resources
5036 void ixgbe_free_rx_resources(struct ixgbe_adapter *adapter,
5037 struct ixgbe_ring *rx_ring)
5039 struct pci_dev *pdev = adapter->pdev;
5041 ixgbe_clean_rx_ring(adapter, rx_ring);
5043 vfree(rx_ring->rx_buffer_info);
5044 rx_ring->rx_buffer_info = NULL;
5046 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
5049 rx_ring->desc = NULL;
5053 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5054 * @adapter: board private structure
5056 * Free all receive software resources
5058 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5062 for (i = 0; i < adapter->num_rx_queues; i++)
5063 if (adapter->rx_ring[i]->desc)
5064 ixgbe_free_rx_resources(adapter, adapter->rx_ring[i]);
5068 * ixgbe_change_mtu - Change the Maximum Transfer Unit
5069 * @netdev: network interface device structure
5070 * @new_mtu: new value for maximum frame size
5072 * Returns 0 on success, negative on failure
5074 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5076 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5077 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5079 /* MTU < 68 is an error and causes problems on some kernels */
5080 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5083 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5084 /* must set new MTU before calling down or up */
5085 netdev->mtu = new_mtu;
5087 if (netif_running(netdev))
5088 ixgbe_reinit_locked(adapter);
5094 * ixgbe_open - Called when a network interface is made active
5095 * @netdev: network interface device structure
5097 * Returns 0 on success, negative value on failure
5099 * The open entry point is called when a network interface is made
5100 * active by the system (IFF_UP). At this point all resources needed
5101 * for transmit and receive operations are allocated, the interrupt
5102 * handler is registered with the OS, the watchdog timer is started,
5103 * and the stack is notified that the interface is ready.
5105 static int ixgbe_open(struct net_device *netdev)
5107 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5110 /* disallow open during test */
5111 if (test_bit(__IXGBE_TESTING, &adapter->state))
5114 netif_carrier_off(netdev);
5116 /* allocate transmit descriptors */
5117 err = ixgbe_setup_all_tx_resources(adapter);
5121 /* allocate receive descriptors */
5122 err = ixgbe_setup_all_rx_resources(adapter);
5126 ixgbe_configure(adapter);
5128 err = ixgbe_request_irq(adapter);
5132 err = ixgbe_up_complete(adapter);
5136 netif_tx_start_all_queues(netdev);
5141 ixgbe_release_hw_control(adapter);
5142 ixgbe_free_irq(adapter);
5145 ixgbe_free_all_rx_resources(adapter);
5147 ixgbe_free_all_tx_resources(adapter);
5148 ixgbe_reset(adapter);
5154 * ixgbe_close - Disables a network interface
5155 * @netdev: network interface device structure
5157 * Returns 0, this is not allowed to fail
5159 * The close entry point is called when an interface is de-activated
5160 * by the OS. The hardware is still under the drivers control, but
5161 * needs to be disabled. A global MAC reset is issued to stop the
5162 * hardware, and all transmit and receive resources are freed.
5164 static int ixgbe_close(struct net_device *netdev)
5166 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5168 ixgbe_down(adapter);
5169 ixgbe_free_irq(adapter);
5171 ixgbe_free_all_tx_resources(adapter);
5172 ixgbe_free_all_rx_resources(adapter);
5174 ixgbe_release_hw_control(adapter);
5180 static int ixgbe_resume(struct pci_dev *pdev)
5182 struct net_device *netdev = pci_get_drvdata(pdev);
5183 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5186 pci_set_power_state(pdev, PCI_D0);
5187 pci_restore_state(pdev);
5189 * pci_restore_state clears dev->state_saved so call
5190 * pci_save_state to restore it.
5192 pci_save_state(pdev);
5194 err = pci_enable_device_mem(pdev);
5196 e_dev_err("Cannot enable PCI device from suspend\n");
5199 pci_set_master(pdev);
5201 pci_wake_from_d3(pdev, false);
5203 err = ixgbe_init_interrupt_scheme(adapter);
5205 e_dev_err("Cannot initialize interrupts for device\n");
5209 ixgbe_reset(adapter);
5211 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5213 if (netif_running(netdev)) {
5214 err = ixgbe_open(adapter->netdev);
5219 netif_device_attach(netdev);
5223 #endif /* CONFIG_PM */
5225 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5227 struct net_device *netdev = pci_get_drvdata(pdev);
5228 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5229 struct ixgbe_hw *hw = &adapter->hw;
5231 u32 wufc = adapter->wol;
5236 netif_device_detach(netdev);
5238 if (netif_running(netdev)) {
5239 ixgbe_down(adapter);
5240 ixgbe_free_irq(adapter);
5241 ixgbe_free_all_tx_resources(adapter);
5242 ixgbe_free_all_rx_resources(adapter);
5246 retval = pci_save_state(pdev);
5252 ixgbe_set_rx_mode(netdev);
5254 /* turn on all-multi mode if wake on multicast is enabled */
5255 if (wufc & IXGBE_WUFC_MC) {
5256 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5257 fctrl |= IXGBE_FCTRL_MPE;
5258 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5261 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5262 ctrl |= IXGBE_CTRL_GIO_DIS;
5263 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5265 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5267 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5268 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5271 if (wufc && hw->mac.type == ixgbe_mac_82599EB)
5272 pci_wake_from_d3(pdev, true);
5274 pci_wake_from_d3(pdev, false);
5276 *enable_wake = !!wufc;
5278 ixgbe_clear_interrupt_scheme(adapter);
5280 ixgbe_release_hw_control(adapter);
5282 pci_disable_device(pdev);
5288 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5293 retval = __ixgbe_shutdown(pdev, &wake);
5298 pci_prepare_to_sleep(pdev);
5300 pci_wake_from_d3(pdev, false);
5301 pci_set_power_state(pdev, PCI_D3hot);
5306 #endif /* CONFIG_PM */
5308 static void ixgbe_shutdown(struct pci_dev *pdev)
5312 __ixgbe_shutdown(pdev, &wake);
5314 if (system_state == SYSTEM_POWER_OFF) {
5315 pci_wake_from_d3(pdev, wake);
5316 pci_set_power_state(pdev, PCI_D3hot);
5321 * ixgbe_update_stats - Update the board statistics counters.
5322 * @adapter: board private structure
5324 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5326 struct net_device *netdev = adapter->netdev;
5327 struct ixgbe_hw *hw = &adapter->hw;
5329 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5330 u64 non_eop_descs = 0, restart_queue = 0;
5332 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5333 test_bit(__IXGBE_RESETTING, &adapter->state))
5336 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5339 for (i = 0; i < 16; i++)
5340 adapter->hw_rx_no_dma_resources +=
5341 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5342 for (i = 0; i < adapter->num_rx_queues; i++) {
5343 rsc_count += adapter->rx_ring[i]->rsc_count;
5344 rsc_flush += adapter->rx_ring[i]->rsc_flush;
5346 adapter->rsc_total_count = rsc_count;
5347 adapter->rsc_total_flush = rsc_flush;
5350 /* gather some stats to the adapter struct that are per queue */
5351 for (i = 0; i < adapter->num_tx_queues; i++)
5352 restart_queue += adapter->tx_ring[i]->restart_queue;
5353 adapter->restart_queue = restart_queue;
5355 for (i = 0; i < adapter->num_rx_queues; i++)
5356 non_eop_descs += adapter->rx_ring[i]->non_eop_descs;
5357 adapter->non_eop_descs = non_eop_descs;
5359 adapter->stats.crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5360 for (i = 0; i < 8; i++) {
5361 /* for packet buffers not used, the register should read 0 */
5362 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5364 adapter->stats.mpc[i] += mpc;
5365 total_mpc += adapter->stats.mpc[i];
5366 if (hw->mac.type == ixgbe_mac_82598EB)
5367 adapter->stats.rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5368 adapter->stats.qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5369 adapter->stats.qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5370 adapter->stats.qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5371 adapter->stats.qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5372 if (hw->mac.type == ixgbe_mac_82599EB) {
5373 adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
5374 IXGBE_PXONRXCNT(i));
5375 adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
5376 IXGBE_PXOFFRXCNT(i));
5377 adapter->stats.qprdc[i] += IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5379 adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
5381 adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
5384 adapter->stats.pxontxc[i] += IXGBE_READ_REG(hw,
5386 adapter->stats.pxofftxc[i] += IXGBE_READ_REG(hw,
5389 adapter->stats.gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5390 /* work around hardware counting issue */
5391 adapter->stats.gprc -= missed_rx;
5393 /* 82598 hardware only has a 32 bit counter in the high register */
5394 if (hw->mac.type == ixgbe_mac_82599EB) {
5396 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5397 tmp = IXGBE_READ_REG(hw, IXGBE_GORCH) & 0xF; /* 4 high bits of GORC */
5398 adapter->stats.gorc += (tmp << 32);
5399 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5400 tmp = IXGBE_READ_REG(hw, IXGBE_GOTCH) & 0xF; /* 4 high bits of GOTC */
5401 adapter->stats.gotc += (tmp << 32);
5402 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5403 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5404 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5405 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
5406 adapter->stats.fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5407 adapter->stats.fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5409 adapter->stats.fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5410 adapter->stats.fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5411 adapter->stats.fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5412 adapter->stats.fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5413 adapter->stats.fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5414 adapter->stats.fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5415 #endif /* IXGBE_FCOE */
5417 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5418 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
5419 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5420 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5421 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5423 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5424 adapter->stats.bprc += bprc;
5425 adapter->stats.mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5426 if (hw->mac.type == ixgbe_mac_82598EB)
5427 adapter->stats.mprc -= bprc;
5428 adapter->stats.roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5429 adapter->stats.prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5430 adapter->stats.prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5431 adapter->stats.prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5432 adapter->stats.prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5433 adapter->stats.prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5434 adapter->stats.prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5435 adapter->stats.rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5436 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5437 adapter->stats.lxontxc += lxon;
5438 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5439 adapter->stats.lxofftxc += lxoff;
5440 adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5441 adapter->stats.gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5442 adapter->stats.mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5444 * 82598 errata - tx of flow control packets is included in tx counters
5446 xon_off_tot = lxon + lxoff;
5447 adapter->stats.gptc -= xon_off_tot;
5448 adapter->stats.mptc -= xon_off_tot;
5449 adapter->stats.gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5450 adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5451 adapter->stats.rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5452 adapter->stats.rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5453 adapter->stats.tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5454 adapter->stats.ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5455 adapter->stats.ptc64 -= xon_off_tot;
5456 adapter->stats.ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5457 adapter->stats.ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5458 adapter->stats.ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5459 adapter->stats.ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5460 adapter->stats.ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5461 adapter->stats.bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5463 /* Fill out the OS statistics structure */
5464 netdev->stats.multicast = adapter->stats.mprc;
5467 netdev->stats.rx_errors = adapter->stats.crcerrs +
5468 adapter->stats.rlec;
5469 netdev->stats.rx_dropped = 0;
5470 netdev->stats.rx_length_errors = adapter->stats.rlec;
5471 netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
5472 netdev->stats.rx_missed_errors = total_mpc;
5476 * ixgbe_watchdog - Timer Call-back
5477 * @data: pointer to adapter cast into an unsigned long
5479 static void ixgbe_watchdog(unsigned long data)
5481 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5482 struct ixgbe_hw *hw = &adapter->hw;
5487 * Do the watchdog outside of interrupt context due to the lovely
5488 * delays that some of the newer hardware requires
5491 if (test_bit(__IXGBE_DOWN, &adapter->state))
5492 goto watchdog_short_circuit;
5494 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5496 * for legacy and MSI interrupts don't set any bits
5497 * that are enabled for EIAM, because this operation
5498 * would set *both* EIMS and EICS for any bit in EIAM
5500 IXGBE_WRITE_REG(hw, IXGBE_EICS,
5501 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5502 goto watchdog_reschedule;
5505 /* get one bit for every active tx/rx interrupt vector */
5506 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
5507 struct ixgbe_q_vector *qv = adapter->q_vector[i];
5508 if (qv->rxr_count || qv->txr_count)
5509 eics |= ((u64)1 << i);
5512 /* Cause software interrupt to ensure rx rings are cleaned */
5513 ixgbe_irq_rearm_queues(adapter, eics);
5515 watchdog_reschedule:
5516 /* Reset the timer */
5517 mod_timer(&adapter->watchdog_timer, round_jiffies(jiffies + 2 * HZ));
5519 watchdog_short_circuit:
5520 schedule_work(&adapter->watchdog_task);
5524 * ixgbe_multispeed_fiber_task - worker thread to configure multispeed fiber
5525 * @work: pointer to work_struct containing our data
5527 static void ixgbe_multispeed_fiber_task(struct work_struct *work)
5529 struct ixgbe_adapter *adapter = container_of(work,
5530 struct ixgbe_adapter,
5531 multispeed_fiber_task);
5532 struct ixgbe_hw *hw = &adapter->hw;
5536 adapter->flags |= IXGBE_FLAG_IN_SFP_LINK_TASK;
5537 autoneg = hw->phy.autoneg_advertised;
5538 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
5539 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
5540 hw->mac.autotry_restart = false;
5541 if (hw->mac.ops.setup_link)
5542 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
5543 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5544 adapter->flags &= ~IXGBE_FLAG_IN_SFP_LINK_TASK;
5548 * ixgbe_sfp_config_module_task - worker thread to configure a new SFP+ module
5549 * @work: pointer to work_struct containing our data
5551 static void ixgbe_sfp_config_module_task(struct work_struct *work)
5553 struct ixgbe_adapter *adapter = container_of(work,
5554 struct ixgbe_adapter,
5555 sfp_config_module_task);
5556 struct ixgbe_hw *hw = &adapter->hw;
5559 adapter->flags |= IXGBE_FLAG_IN_SFP_MOD_TASK;
5561 /* Time for electrical oscillations to settle down */
5563 err = hw->phy.ops.identify_sfp(hw);
5565 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
5566 e_dev_err("failed to initialize because an unsupported SFP+ "
5567 "module type was detected.\n");
5568 e_dev_err("Reload the driver after installing a supported "
5570 unregister_netdev(adapter->netdev);
5573 hw->mac.ops.setup_sfp(hw);
5575 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
5576 /* This will also work for DA Twinax connections */
5577 schedule_work(&adapter->multispeed_fiber_task);
5578 adapter->flags &= ~IXGBE_FLAG_IN_SFP_MOD_TASK;
5582 * ixgbe_fdir_reinit_task - worker thread to reinit FDIR filter table
5583 * @work: pointer to work_struct containing our data
5585 static void ixgbe_fdir_reinit_task(struct work_struct *work)
5587 struct ixgbe_adapter *adapter = container_of(work,
5588 struct ixgbe_adapter,
5590 struct ixgbe_hw *hw = &adapter->hw;
5593 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5594 for (i = 0; i < adapter->num_tx_queues; i++)
5595 set_bit(__IXGBE_FDIR_INIT_DONE,
5596 &(adapter->tx_ring[i]->reinit_state));
5598 e_err(probe, "failed to finish FDIR re-initialization, "
5599 "ignored adding FDIR ATR filters\n");
5601 /* Done FDIR Re-initialization, enable transmits */
5602 netif_tx_start_all_queues(adapter->netdev);
5605 static DEFINE_MUTEX(ixgbe_watchdog_lock);
5608 * ixgbe_watchdog_task - worker thread to bring link up
5609 * @work: pointer to work_struct containing our data
5611 static void ixgbe_watchdog_task(struct work_struct *work)
5613 struct ixgbe_adapter *adapter = container_of(work,
5614 struct ixgbe_adapter,
5616 struct net_device *netdev = adapter->netdev;
5617 struct ixgbe_hw *hw = &adapter->hw;
5621 struct ixgbe_ring *tx_ring;
5622 int some_tx_pending = 0;
5624 mutex_lock(&ixgbe_watchdog_lock);
5626 link_up = adapter->link_up;
5627 link_speed = adapter->link_speed;
5629 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
5630 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5633 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5634 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
5635 hw->mac.ops.fc_enable(hw, i);
5637 hw->mac.ops.fc_enable(hw, 0);
5640 hw->mac.ops.fc_enable(hw, 0);
5645 time_after(jiffies, (adapter->link_check_timeout +
5646 IXGBE_TRY_LINK_TIMEOUT))) {
5647 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5648 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5650 adapter->link_up = link_up;
5651 adapter->link_speed = link_speed;
5655 if (!netif_carrier_ok(netdev)) {
5656 bool flow_rx, flow_tx;
5658 if (hw->mac.type == ixgbe_mac_82599EB) {
5659 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5660 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5661 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5662 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5664 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5665 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5666 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5667 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5670 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
5671 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5673 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5674 "1 Gbps" : "unknown speed")),
5675 ((flow_rx && flow_tx) ? "RX/TX" :
5677 (flow_tx ? "TX" : "None"))));
5679 netif_carrier_on(netdev);
5681 /* Force detection of hung controller */
5682 adapter->detect_tx_hung = true;
5685 adapter->link_up = false;
5686 adapter->link_speed = 0;
5687 if (netif_carrier_ok(netdev)) {
5688 e_info(drv, "NIC Link is Down\n");
5689 netif_carrier_off(netdev);
5693 if (!netif_carrier_ok(netdev)) {
5694 for (i = 0; i < adapter->num_tx_queues; i++) {
5695 tx_ring = adapter->tx_ring[i];
5696 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5697 some_tx_pending = 1;
5702 if (some_tx_pending) {
5703 /* We've lost link, so the controller stops DMA,
5704 * but we've got queued Tx work that's never going
5705 * to get done, so reset controller to flush Tx.
5706 * (Do the reset outside of interrupt context).
5708 schedule_work(&adapter->reset_task);
5712 ixgbe_update_stats(adapter);
5713 mutex_unlock(&ixgbe_watchdog_lock);
5716 static int ixgbe_tso(struct ixgbe_adapter *adapter,
5717 struct ixgbe_ring *tx_ring, struct sk_buff *skb,
5718 u32 tx_flags, u8 *hdr_len)
5720 struct ixgbe_adv_tx_context_desc *context_desc;
5723 struct ixgbe_tx_buffer *tx_buffer_info;
5724 u32 vlan_macip_lens = 0, type_tucmd_mlhl;
5725 u32 mss_l4len_idx, l4len;
5727 if (skb_is_gso(skb)) {
5728 if (skb_header_cloned(skb)) {
5729 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5733 l4len = tcp_hdrlen(skb);
5736 if (skb->protocol == htons(ETH_P_IP)) {
5737 struct iphdr *iph = ip_hdr(skb);
5740 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
5744 } else if (skb_is_gso_v6(skb)) {
5745 ipv6_hdr(skb)->payload_len = 0;
5746 tcp_hdr(skb)->check =
5747 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5748 &ipv6_hdr(skb)->daddr,
5752 i = tx_ring->next_to_use;
5754 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5755 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
5757 /* VLAN MACLEN IPLEN */
5758 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5760 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
5761 vlan_macip_lens |= ((skb_network_offset(skb)) <<
5762 IXGBE_ADVTXD_MACLEN_SHIFT);
5763 *hdr_len += skb_network_offset(skb);
5765 (skb_transport_header(skb) - skb_network_header(skb));
5767 (skb_transport_header(skb) - skb_network_header(skb));
5768 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
5769 context_desc->seqnum_seed = 0;
5771 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
5772 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
5773 IXGBE_ADVTXD_DTYP_CTXT);
5775 if (skb->protocol == htons(ETH_P_IP))
5776 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
5777 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5778 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
5782 (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
5783 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
5784 /* use index 1 for TSO */
5785 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
5786 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
5788 tx_buffer_info->time_stamp = jiffies;
5789 tx_buffer_info->next_to_watch = i;
5792 if (i == tx_ring->count)
5794 tx_ring->next_to_use = i;
5801 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
5802 struct ixgbe_ring *tx_ring,
5803 struct sk_buff *skb, u32 tx_flags)
5805 struct ixgbe_adv_tx_context_desc *context_desc;
5807 struct ixgbe_tx_buffer *tx_buffer_info;
5808 u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
5810 if (skb->ip_summed == CHECKSUM_PARTIAL ||
5811 (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
5812 i = tx_ring->next_to_use;
5813 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5814 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
5816 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5818 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
5819 vlan_macip_lens |= (skb_network_offset(skb) <<
5820 IXGBE_ADVTXD_MACLEN_SHIFT);
5821 if (skb->ip_summed == CHECKSUM_PARTIAL)
5822 vlan_macip_lens |= (skb_transport_header(skb) -
5823 skb_network_header(skb));
5825 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
5826 context_desc->seqnum_seed = 0;
5828 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
5829 IXGBE_ADVTXD_DTYP_CTXT);
5831 if (skb->ip_summed == CHECKSUM_PARTIAL) {
5834 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
5835 const struct vlan_ethhdr *vhdr =
5836 (const struct vlan_ethhdr *)skb->data;
5838 protocol = vhdr->h_vlan_encapsulated_proto;
5840 protocol = skb->protocol;
5844 case cpu_to_be16(ETH_P_IP):
5845 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
5846 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
5848 IXGBE_ADVTXD_TUCMD_L4T_TCP;
5849 else if (ip_hdr(skb)->protocol == IPPROTO_SCTP)
5851 IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5853 case cpu_to_be16(ETH_P_IPV6):
5854 /* XXX what about other V6 headers?? */
5855 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
5857 IXGBE_ADVTXD_TUCMD_L4T_TCP;
5858 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_SCTP)
5860 IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5863 if (unlikely(net_ratelimit())) {
5864 e_warn(probe, "partial checksum "
5872 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
5873 /* use index zero for tx checksum offload */
5874 context_desc->mss_l4len_idx = 0;
5876 tx_buffer_info->time_stamp = jiffies;
5877 tx_buffer_info->next_to_watch = i;
5880 if (i == tx_ring->count)
5882 tx_ring->next_to_use = i;
5890 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
5891 struct ixgbe_ring *tx_ring,
5892 struct sk_buff *skb, u32 tx_flags,
5895 struct pci_dev *pdev = adapter->pdev;
5896 struct ixgbe_tx_buffer *tx_buffer_info;
5898 unsigned int total = skb->len;
5899 unsigned int offset = 0, size, count = 0, i;
5900 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
5903 i = tx_ring->next_to_use;
5905 if (tx_flags & IXGBE_TX_FLAGS_FCOE)
5906 /* excluding fcoe_crc_eof for FCoE */
5907 total -= sizeof(struct fcoe_crc_eof);
5909 len = min(skb_headlen(skb), total);
5911 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5912 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
5914 tx_buffer_info->length = size;
5915 tx_buffer_info->mapped_as_page = false;
5916 tx_buffer_info->dma = dma_map_single(&pdev->dev,
5918 size, DMA_TO_DEVICE);
5919 if (dma_mapping_error(&pdev->dev, tx_buffer_info->dma))
5921 tx_buffer_info->time_stamp = jiffies;
5922 tx_buffer_info->next_to_watch = i;
5931 if (i == tx_ring->count)
5936 for (f = 0; f < nr_frags; f++) {
5937 struct skb_frag_struct *frag;
5939 frag = &skb_shinfo(skb)->frags[f];
5940 len = min((unsigned int)frag->size, total);
5941 offset = frag->page_offset;
5945 if (i == tx_ring->count)
5948 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5949 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
5951 tx_buffer_info->length = size;
5952 tx_buffer_info->dma = dma_map_page(&adapter->pdev->dev,
5956 tx_buffer_info->mapped_as_page = true;
5957 if (dma_mapping_error(&pdev->dev, tx_buffer_info->dma))
5959 tx_buffer_info->time_stamp = jiffies;
5960 tx_buffer_info->next_to_watch = i;
5971 tx_ring->tx_buffer_info[i].skb = skb;
5972 tx_ring->tx_buffer_info[first].next_to_watch = i;
5977 e_dev_err("TX DMA map failed\n");
5979 /* clear timestamp and dma mappings for failed tx_buffer_info map */
5980 tx_buffer_info->dma = 0;
5981 tx_buffer_info->time_stamp = 0;
5982 tx_buffer_info->next_to_watch = 0;
5986 /* clear timestamp and dma mappings for remaining portion of packet */
5989 i += tx_ring->count;
5991 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5992 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
5998 static void ixgbe_tx_queue(struct ixgbe_adapter *adapter,
5999 struct ixgbe_ring *tx_ring,
6000 int tx_flags, int count, u32 paylen, u8 hdr_len)
6002 union ixgbe_adv_tx_desc *tx_desc = NULL;
6003 struct ixgbe_tx_buffer *tx_buffer_info;
6004 u32 olinfo_status = 0, cmd_type_len = 0;
6006 u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
6008 cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
6010 cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
6012 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
6013 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
6015 if (tx_flags & IXGBE_TX_FLAGS_TSO) {
6016 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
6018 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
6019 IXGBE_ADVTXD_POPTS_SHIFT;
6021 /* use index 1 context for tso */
6022 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6023 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
6024 olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
6025 IXGBE_ADVTXD_POPTS_SHIFT;
6027 } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
6028 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
6029 IXGBE_ADVTXD_POPTS_SHIFT;
6031 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6032 olinfo_status |= IXGBE_ADVTXD_CC;
6033 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6034 if (tx_flags & IXGBE_TX_FLAGS_FSO)
6035 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
6038 olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
6040 i = tx_ring->next_to_use;
6042 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6043 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
6044 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
6045 tx_desc->read.cmd_type_len =
6046 cpu_to_le32(cmd_type_len | tx_buffer_info->length);
6047 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
6049 if (i == tx_ring->count)
6053 tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
6056 * Force memory writes to complete before letting h/w
6057 * know there are new descriptors to fetch. (Only
6058 * applicable for weak-ordered memory model archs,
6063 tx_ring->next_to_use = i;
6064 writel(i, adapter->hw.hw_addr + tx_ring->tail);
6067 static void ixgbe_atr(struct ixgbe_adapter *adapter, struct sk_buff *skb,
6068 int queue, u32 tx_flags)
6070 struct ixgbe_atr_input atr_input;
6072 struct iphdr *iph = ip_hdr(skb);
6073 struct ethhdr *eth = (struct ethhdr *)skb->data;
6074 u16 vlan_id, src_port, dst_port, flex_bytes;
6075 u32 src_ipv4_addr, dst_ipv4_addr;
6078 /* Right now, we support IPv4 only */
6079 if (skb->protocol != htons(ETH_P_IP))
6081 /* check if we're UDP or TCP */
6082 if (iph->protocol == IPPROTO_TCP) {
6084 src_port = th->source;
6085 dst_port = th->dest;
6086 l4type |= IXGBE_ATR_L4TYPE_TCP;
6087 /* l4type IPv4 type is 0, no need to assign */
6089 /* Unsupported L4 header, just bail here */
6093 memset(&atr_input, 0, sizeof(struct ixgbe_atr_input));
6095 vlan_id = (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK) >>
6096 IXGBE_TX_FLAGS_VLAN_SHIFT;
6097 src_ipv4_addr = iph->saddr;
6098 dst_ipv4_addr = iph->daddr;
6099 flex_bytes = eth->h_proto;
6101 ixgbe_atr_set_vlan_id_82599(&atr_input, vlan_id);
6102 ixgbe_atr_set_src_port_82599(&atr_input, dst_port);
6103 ixgbe_atr_set_dst_port_82599(&atr_input, src_port);
6104 ixgbe_atr_set_flex_byte_82599(&atr_input, flex_bytes);
6105 ixgbe_atr_set_l4type_82599(&atr_input, l4type);
6106 /* src and dst are inverted, think how the receiver sees them */
6107 ixgbe_atr_set_src_ipv4_82599(&atr_input, dst_ipv4_addr);
6108 ixgbe_atr_set_dst_ipv4_82599(&atr_input, src_ipv4_addr);
6110 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
6111 ixgbe_fdir_add_signature_filter_82599(&adapter->hw, &atr_input, queue);
6114 static int __ixgbe_maybe_stop_tx(struct net_device *netdev,
6115 struct ixgbe_ring *tx_ring, int size)
6117 netif_stop_subqueue(netdev, tx_ring->queue_index);
6118 /* Herbert's original patch had:
6119 * smp_mb__after_netif_stop_queue();
6120 * but since that doesn't exist yet, just open code it. */
6123 /* We need to check again in a case another CPU has just
6124 * made room available. */
6125 if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
6128 /* A reprieve! - use start_queue because it doesn't call schedule */
6129 netif_start_subqueue(netdev, tx_ring->queue_index);
6130 ++tx_ring->restart_queue;
6134 static int ixgbe_maybe_stop_tx(struct net_device *netdev,
6135 struct ixgbe_ring *tx_ring, int size)
6137 if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
6139 return __ixgbe_maybe_stop_tx(netdev, tx_ring, size);
6142 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6144 struct ixgbe_adapter *adapter = netdev_priv(dev);
6145 int txq = smp_processor_id();
6148 if ((skb->protocol == htons(ETH_P_FCOE)) ||
6149 (skb->protocol == htons(ETH_P_FIP))) {
6150 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
6151 txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
6152 txq += adapter->ring_feature[RING_F_FCOE].mask;
6154 #ifdef CONFIG_IXGBE_DCB
6155 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6156 txq = adapter->fcoe.up;
6163 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6164 while (unlikely(txq >= dev->real_num_tx_queues))
6165 txq -= dev->real_num_tx_queues;
6169 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6170 if (skb->priority == TC_PRIO_CONTROL)
6171 txq = adapter->ring_feature[RING_F_DCB].indices-1;
6173 txq = (skb->vlan_tci & IXGBE_TX_FLAGS_VLAN_PRIO_MASK)
6178 return skb_tx_hash(dev, skb);
6181 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
6182 struct net_device *netdev)
6184 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6185 struct ixgbe_ring *tx_ring;
6186 struct netdev_queue *txq;
6188 unsigned int tx_flags = 0;
6194 if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
6195 tx_flags |= vlan_tx_tag_get(skb);
6196 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6197 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6198 tx_flags |= ((skb->queue_mapping & 0x7) << 13);
6200 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6201 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6202 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED &&
6203 skb->priority != TC_PRIO_CONTROL) {
6204 tx_flags |= ((skb->queue_mapping & 0x7) << 13);
6205 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6206 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6209 tx_ring = adapter->tx_ring[skb->queue_mapping];
6212 /* for FCoE with DCB, we force the priority to what
6213 * was specified by the switch */
6214 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED &&
6215 (skb->protocol == htons(ETH_P_FCOE) ||
6216 skb->protocol == htons(ETH_P_FIP))) {
6217 #ifdef CONFIG_IXGBE_DCB
6218 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6219 tx_flags &= ~(IXGBE_TX_FLAGS_VLAN_PRIO_MASK
6220 << IXGBE_TX_FLAGS_VLAN_SHIFT);
6221 tx_flags |= ((adapter->fcoe.up << 13)
6222 << IXGBE_TX_FLAGS_VLAN_SHIFT);
6225 /* flag for FCoE offloads */
6226 if (skb->protocol == htons(ETH_P_FCOE))
6227 tx_flags |= IXGBE_TX_FLAGS_FCOE;
6231 /* four things can cause us to need a context descriptor */
6232 if (skb_is_gso(skb) ||
6233 (skb->ip_summed == CHECKSUM_PARTIAL) ||
6234 (tx_flags & IXGBE_TX_FLAGS_VLAN) ||
6235 (tx_flags & IXGBE_TX_FLAGS_FCOE))
6238 count += TXD_USE_COUNT(skb_headlen(skb));
6239 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6240 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6242 if (ixgbe_maybe_stop_tx(netdev, tx_ring, count)) {
6244 return NETDEV_TX_BUSY;
6247 first = tx_ring->next_to_use;
6248 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6250 /* setup tx offload for FCoE */
6251 tso = ixgbe_fso(adapter, tx_ring, skb, tx_flags, &hdr_len);
6253 dev_kfree_skb_any(skb);
6254 return NETDEV_TX_OK;
6257 tx_flags |= IXGBE_TX_FLAGS_FSO;
6258 #endif /* IXGBE_FCOE */
6260 if (skb->protocol == htons(ETH_P_IP))
6261 tx_flags |= IXGBE_TX_FLAGS_IPV4;
6262 tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len);
6264 dev_kfree_skb_any(skb);
6265 return NETDEV_TX_OK;
6269 tx_flags |= IXGBE_TX_FLAGS_TSO;
6270 else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags) &&
6271 (skb->ip_summed == CHECKSUM_PARTIAL))
6272 tx_flags |= IXGBE_TX_FLAGS_CSUM;
6275 count = ixgbe_tx_map(adapter, tx_ring, skb, tx_flags, first);
6277 /* add the ATR filter if ATR is on */
6278 if (tx_ring->atr_sample_rate) {
6279 ++tx_ring->atr_count;
6280 if ((tx_ring->atr_count >= tx_ring->atr_sample_rate) &&
6281 test_bit(__IXGBE_FDIR_INIT_DONE,
6282 &tx_ring->reinit_state)) {
6283 ixgbe_atr(adapter, skb, tx_ring->queue_index,
6285 tx_ring->atr_count = 0;
6288 txq = netdev_get_tx_queue(netdev, tx_ring->queue_index);
6289 txq->tx_bytes += skb->len;
6291 ixgbe_tx_queue(adapter, tx_ring, tx_flags, count, skb->len,
6293 ixgbe_maybe_stop_tx(netdev, tx_ring, DESC_NEEDED);
6296 dev_kfree_skb_any(skb);
6297 tx_ring->tx_buffer_info[first].time_stamp = 0;
6298 tx_ring->next_to_use = first;
6301 return NETDEV_TX_OK;
6305 * ixgbe_set_mac - Change the Ethernet Address of the NIC
6306 * @netdev: network interface device structure
6307 * @p: pointer to an address structure
6309 * Returns 0 on success, negative on failure
6311 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6313 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6314 struct ixgbe_hw *hw = &adapter->hw;
6315 struct sockaddr *addr = p;
6317 if (!is_valid_ether_addr(addr->sa_data))
6318 return -EADDRNOTAVAIL;
6320 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6321 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6323 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
6330 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6332 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6333 struct ixgbe_hw *hw = &adapter->hw;
6337 if (prtad != hw->phy.mdio.prtad)
6339 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6345 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6346 u16 addr, u16 value)
6348 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6349 struct ixgbe_hw *hw = &adapter->hw;
6351 if (prtad != hw->phy.mdio.prtad)
6353 return hw->phy.ops.write_reg(hw, addr, devad, value);
6356 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6358 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6360 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6364 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6366 * @netdev: network interface device structure
6368 * Returns non-zero on failure
6370 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6373 struct ixgbe_adapter *adapter = netdev_priv(dev);
6374 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6376 if (is_valid_ether_addr(mac->san_addr)) {
6378 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6385 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6387 * @netdev: network interface device structure
6389 * Returns non-zero on failure
6391 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6394 struct ixgbe_adapter *adapter = netdev_priv(dev);
6395 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6397 if (is_valid_ether_addr(mac->san_addr)) {
6399 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6405 #ifdef CONFIG_NET_POLL_CONTROLLER
6407 * Polling 'interrupt' - used by things like netconsole to send skbs
6408 * without having to re-enable interrupts. It's not called while
6409 * the interrupt routine is executing.
6411 static void ixgbe_netpoll(struct net_device *netdev)
6413 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6416 /* if interface is down do nothing */
6417 if (test_bit(__IXGBE_DOWN, &adapter->state))
6420 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
6421 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
6422 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
6423 for (i = 0; i < num_q_vectors; i++) {
6424 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
6425 ixgbe_msix_clean_many(0, q_vector);
6428 ixgbe_intr(adapter->pdev->irq, netdev);
6430 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
6434 static const struct net_device_ops ixgbe_netdev_ops = {
6435 .ndo_open = ixgbe_open,
6436 .ndo_stop = ixgbe_close,
6437 .ndo_start_xmit = ixgbe_xmit_frame,
6438 .ndo_select_queue = ixgbe_select_queue,
6439 .ndo_set_rx_mode = ixgbe_set_rx_mode,
6440 .ndo_set_multicast_list = ixgbe_set_rx_mode,
6441 .ndo_validate_addr = eth_validate_addr,
6442 .ndo_set_mac_address = ixgbe_set_mac,
6443 .ndo_change_mtu = ixgbe_change_mtu,
6444 .ndo_tx_timeout = ixgbe_tx_timeout,
6445 .ndo_vlan_rx_register = ixgbe_vlan_rx_register,
6446 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
6447 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
6448 .ndo_do_ioctl = ixgbe_ioctl,
6449 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
6450 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
6451 .ndo_set_vf_tx_rate = ixgbe_ndo_set_vf_bw,
6452 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
6453 #ifdef CONFIG_NET_POLL_CONTROLLER
6454 .ndo_poll_controller = ixgbe_netpoll,
6457 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
6458 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
6459 .ndo_fcoe_enable = ixgbe_fcoe_enable,
6460 .ndo_fcoe_disable = ixgbe_fcoe_disable,
6461 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
6462 #endif /* IXGBE_FCOE */
6465 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
6466 const struct ixgbe_info *ii)
6468 #ifdef CONFIG_PCI_IOV
6469 struct ixgbe_hw *hw = &adapter->hw;
6472 if (hw->mac.type != ixgbe_mac_82599EB || !max_vfs)
6475 /* The 82599 supports up to 64 VFs per physical function
6476 * but this implementation limits allocation to 63 so that
6477 * basic networking resources are still available to the
6480 adapter->num_vfs = (max_vfs > 63) ? 63 : max_vfs;
6481 adapter->flags |= IXGBE_FLAG_SRIOV_ENABLED;
6482 err = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
6484 e_err(probe, "Failed to enable PCI sriov: %d\n", err);
6487 /* If call to enable VFs succeeded then allocate memory
6488 * for per VF control structures.
6491 kcalloc(adapter->num_vfs,
6492 sizeof(struct vf_data_storage), GFP_KERNEL);
6493 if (adapter->vfinfo) {
6494 /* Now that we're sure SR-IOV is enabled
6495 * and memory allocated set up the mailbox parameters
6497 ixgbe_init_mbx_params_pf(hw);
6498 memcpy(&hw->mbx.ops, ii->mbx_ops,
6499 sizeof(hw->mbx.ops));
6501 /* Disable RSC when in SR-IOV mode */
6502 adapter->flags2 &= ~(IXGBE_FLAG2_RSC_CAPABLE |
6503 IXGBE_FLAG2_RSC_ENABLED);
6508 e_err(probe, "Unable to allocate memory for VF Data Storage - "
6509 "SRIOV disabled\n");
6510 pci_disable_sriov(adapter->pdev);
6513 adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
6514 adapter->num_vfs = 0;
6515 #endif /* CONFIG_PCI_IOV */
6519 * ixgbe_probe - Device Initialization Routine
6520 * @pdev: PCI device information struct
6521 * @ent: entry in ixgbe_pci_tbl
6523 * Returns 0 on success, negative on failure
6525 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
6526 * The OS initialization, configuring of the adapter private structure,
6527 * and a hardware reset occur.
6529 static int __devinit ixgbe_probe(struct pci_dev *pdev,
6530 const struct pci_device_id *ent)
6532 struct net_device *netdev;
6533 struct ixgbe_adapter *adapter = NULL;
6534 struct ixgbe_hw *hw;
6535 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
6536 static int cards_found;
6537 int i, err, pci_using_dac;
6538 unsigned int indices = num_possible_cpus();
6544 /* Catch broken hardware that put the wrong VF device ID in
6545 * the PCIe SR-IOV capability.
6547 if (pdev->is_virtfn) {
6548 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
6549 pci_name(pdev), pdev->vendor, pdev->device);
6553 err = pci_enable_device_mem(pdev);
6557 if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
6558 !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
6561 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
6563 err = dma_set_coherent_mask(&pdev->dev,
6567 "No usable DMA configuration, aborting\n");
6574 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
6575 IORESOURCE_MEM), ixgbe_driver_name);
6578 "pci_request_selected_regions failed 0x%x\n", err);
6582 pci_enable_pcie_error_reporting(pdev);
6584 pci_set_master(pdev);
6585 pci_save_state(pdev);
6587 if (ii->mac == ixgbe_mac_82598EB)
6588 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
6590 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
6592 indices = max_t(unsigned int, indices, IXGBE_MAX_DCB_INDICES);
6594 indices += min_t(unsigned int, num_possible_cpus(),
6595 IXGBE_MAX_FCOE_INDICES);
6597 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
6600 goto err_alloc_etherdev;
6603 SET_NETDEV_DEV(netdev, &pdev->dev);
6605 pci_set_drvdata(pdev, netdev);
6606 adapter = netdev_priv(netdev);
6608 adapter->netdev = netdev;
6609 adapter->pdev = pdev;
6612 adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
6614 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
6615 pci_resource_len(pdev, 0));
6621 for (i = 1; i <= 5; i++) {
6622 if (pci_resource_len(pdev, i) == 0)
6626 netdev->netdev_ops = &ixgbe_netdev_ops;
6627 ixgbe_set_ethtool_ops(netdev);
6628 netdev->watchdog_timeo = 5 * HZ;
6629 strcpy(netdev->name, pci_name(pdev));
6631 adapter->bd_number = cards_found;
6634 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
6635 hw->mac.type = ii->mac;
6638 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
6639 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
6640 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
6641 if (!(eec & (1 << 8)))
6642 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
6645 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
6646 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
6647 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
6648 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
6649 hw->phy.mdio.mmds = 0;
6650 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
6651 hw->phy.mdio.dev = netdev;
6652 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
6653 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
6655 /* set up this timer and work struct before calling get_invariants
6656 * which might start the timer
6658 init_timer(&adapter->sfp_timer);
6659 adapter->sfp_timer.function = &ixgbe_sfp_timer;
6660 adapter->sfp_timer.data = (unsigned long) adapter;
6662 INIT_WORK(&adapter->sfp_task, ixgbe_sfp_task);
6664 /* multispeed fiber has its own tasklet, called from GPI SDP1 context */
6665 INIT_WORK(&adapter->multispeed_fiber_task, ixgbe_multispeed_fiber_task);
6667 /* a new SFP+ module arrival, called from GPI SDP2 context */
6668 INIT_WORK(&adapter->sfp_config_module_task,
6669 ixgbe_sfp_config_module_task);
6671 ii->get_invariants(hw);
6673 /* setup the private structure */
6674 err = ixgbe_sw_init(adapter);
6678 /* Make it possible the adapter to be woken up via WOL */
6679 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
6680 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6683 * If there is a fan on this device and it has failed log the
6686 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
6687 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
6688 if (esdp & IXGBE_ESDP_SDP1)
6689 e_crit(probe, "Fan has stopped, replace the adapter\n");
6692 /* reset_hw fills in the perm_addr as well */
6693 hw->phy.reset_if_overtemp = true;
6694 err = hw->mac.ops.reset_hw(hw);
6695 hw->phy.reset_if_overtemp = false;
6696 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
6697 hw->mac.type == ixgbe_mac_82598EB) {
6699 * Start a kernel thread to watch for a module to arrive.
6700 * Only do this for 82598, since 82599 will generate
6701 * interrupts on module arrival.
6703 set_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
6704 mod_timer(&adapter->sfp_timer,
6705 round_jiffies(jiffies + (2 * HZ)));
6707 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
6708 e_dev_err("failed to initialize because an unsupported SFP+ "
6709 "module type was detected.\n");
6710 e_dev_err("Reload the driver after installing a supported "
6714 e_dev_err("HW Init failed: %d\n", err);
6718 ixgbe_probe_vf(adapter, ii);
6720 netdev->features = NETIF_F_SG |
6722 NETIF_F_HW_VLAN_TX |
6723 NETIF_F_HW_VLAN_RX |
6724 NETIF_F_HW_VLAN_FILTER;
6726 netdev->features |= NETIF_F_IPV6_CSUM;
6727 netdev->features |= NETIF_F_TSO;
6728 netdev->features |= NETIF_F_TSO6;
6729 netdev->features |= NETIF_F_GRO;
6731 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
6732 netdev->features |= NETIF_F_SCTP_CSUM;
6734 netdev->vlan_features |= NETIF_F_TSO;
6735 netdev->vlan_features |= NETIF_F_TSO6;
6736 netdev->vlan_features |= NETIF_F_IP_CSUM;
6737 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
6738 netdev->vlan_features |= NETIF_F_SG;
6740 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6741 adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
6742 IXGBE_FLAG_DCB_ENABLED);
6743 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
6744 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
6746 #ifdef CONFIG_IXGBE_DCB
6747 netdev->dcbnl_ops = &dcbnl_ops;
6751 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
6752 if (hw->mac.ops.get_device_caps) {
6753 hw->mac.ops.get_device_caps(hw, &device_caps);
6754 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
6755 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6758 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
6759 netdev->vlan_features |= NETIF_F_FCOE_CRC;
6760 netdev->vlan_features |= NETIF_F_FSO;
6761 netdev->vlan_features |= NETIF_F_FCOE_MTU;
6763 #endif /* IXGBE_FCOE */
6765 netdev->features |= NETIF_F_HIGHDMA;
6767 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
6768 netdev->features |= NETIF_F_LRO;
6770 /* make sure the EEPROM is good */
6771 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
6772 e_dev_err("The EEPROM Checksum Is Not Valid\n");
6777 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
6778 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
6780 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
6781 e_dev_err("invalid MAC address\n");
6786 /* power down the optics */
6787 if (hw->phy.multispeed_fiber)
6788 hw->mac.ops.disable_tx_laser(hw);
6790 init_timer(&adapter->watchdog_timer);
6791 adapter->watchdog_timer.function = &ixgbe_watchdog;
6792 adapter->watchdog_timer.data = (unsigned long)adapter;
6794 INIT_WORK(&adapter->reset_task, ixgbe_reset_task);
6795 INIT_WORK(&adapter->watchdog_task, ixgbe_watchdog_task);
6797 err = ixgbe_init_interrupt_scheme(adapter);
6801 switch (pdev->device) {
6802 case IXGBE_DEV_ID_82599_KX4:
6803 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
6804 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
6810 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
6812 /* pick up the PCI bus settings for reporting later */
6813 hw->mac.ops.get_bus_info(hw);
6815 /* print bus type/speed/width info */
6816 e_dev_info("(PCI Express:%s:%s) %pM\n",
6817 ((hw->bus.speed == ixgbe_bus_speed_5000) ? "5.0Gb/s":
6818 (hw->bus.speed == ixgbe_bus_speed_2500) ? "2.5Gb/s":"Unknown"),
6819 ((hw->bus.width == ixgbe_bus_width_pcie_x8) ? "Width x8" :
6820 (hw->bus.width == ixgbe_bus_width_pcie_x4) ? "Width x4" :
6821 (hw->bus.width == ixgbe_bus_width_pcie_x1) ? "Width x1" :
6824 ixgbe_read_pba_num_generic(hw, &part_num);
6825 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
6826 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, "
6827 "PBA No: %06x-%03x\n",
6828 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
6829 (part_num >> 8), (part_num & 0xff));
6831 e_dev_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
6832 hw->mac.type, hw->phy.type,
6833 (part_num >> 8), (part_num & 0xff));
6835 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
6836 e_dev_warn("PCI-Express bandwidth available for this card is "
6837 "not sufficient for optimal performance.\n");
6838 e_dev_warn("For optimal performance a x8 PCI-Express slot "
6842 /* save off EEPROM version number */
6843 hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
6845 /* reset the hardware with the new settings */
6846 err = hw->mac.ops.start_hw(hw);
6848 if (err == IXGBE_ERR_EEPROM_VERSION) {
6849 /* We are running on a pre-production device, log a warning */
6850 e_dev_warn("This device is a pre-production adapter/LOM. "
6851 "Please be aware there may be issues associated "
6852 "with your hardware. If you are experiencing "
6853 "problems please contact your Intel or hardware "
6854 "representative who provided you with this "
6857 strcpy(netdev->name, "eth%d");
6858 err = register_netdev(netdev);
6862 /* carrier off reporting is important to ethtool even BEFORE open */
6863 netif_carrier_off(netdev);
6865 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
6866 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
6867 INIT_WORK(&adapter->fdir_reinit_task, ixgbe_fdir_reinit_task);
6869 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
6870 INIT_WORK(&adapter->check_overtemp_task, ixgbe_check_overtemp_task);
6871 #ifdef CONFIG_IXGBE_DCA
6872 if (dca_add_requester(&pdev->dev) == 0) {
6873 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
6874 ixgbe_setup_dca(adapter);
6877 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
6878 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
6879 for (i = 0; i < adapter->num_vfs; i++)
6880 ixgbe_vf_configuration(pdev, (i | 0x10000000));
6883 /* add san mac addr to netdev */
6884 ixgbe_add_sanmac_netdev(netdev);
6886 e_dev_info("Intel(R) 10 Gigabit Network Connection\n");
6891 ixgbe_release_hw_control(adapter);
6892 ixgbe_clear_interrupt_scheme(adapter);
6895 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6896 ixgbe_disable_sriov(adapter);
6897 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
6898 del_timer_sync(&adapter->sfp_timer);
6899 cancel_work_sync(&adapter->sfp_task);
6900 cancel_work_sync(&adapter->multispeed_fiber_task);
6901 cancel_work_sync(&adapter->sfp_config_module_task);
6902 iounmap(hw->hw_addr);
6904 free_netdev(netdev);
6906 pci_release_selected_regions(pdev, pci_select_bars(pdev,
6910 pci_disable_device(pdev);
6915 * ixgbe_remove - Device Removal Routine
6916 * @pdev: PCI device information struct
6918 * ixgbe_remove is called by the PCI subsystem to alert the driver
6919 * that it should release a PCI device. The could be caused by a
6920 * Hot-Plug event, or because the driver is going to be removed from
6923 static void __devexit ixgbe_remove(struct pci_dev *pdev)
6925 struct net_device *netdev = pci_get_drvdata(pdev);
6926 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6928 set_bit(__IXGBE_DOWN, &adapter->state);
6929 /* clear the module not found bit to make sure the worker won't
6932 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
6933 del_timer_sync(&adapter->watchdog_timer);
6935 del_timer_sync(&adapter->sfp_timer);
6936 cancel_work_sync(&adapter->watchdog_task);
6937 cancel_work_sync(&adapter->sfp_task);
6938 cancel_work_sync(&adapter->multispeed_fiber_task);
6939 cancel_work_sync(&adapter->sfp_config_module_task);
6940 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
6941 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
6942 cancel_work_sync(&adapter->fdir_reinit_task);
6943 flush_scheduled_work();
6945 #ifdef CONFIG_IXGBE_DCA
6946 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
6947 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
6948 dca_remove_requester(&pdev->dev);
6949 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
6954 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
6955 ixgbe_cleanup_fcoe(adapter);
6957 #endif /* IXGBE_FCOE */
6959 /* remove the added san mac */
6960 ixgbe_del_sanmac_netdev(netdev);
6962 if (netdev->reg_state == NETREG_REGISTERED)
6963 unregister_netdev(netdev);
6965 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6966 ixgbe_disable_sriov(adapter);
6968 ixgbe_clear_interrupt_scheme(adapter);
6970 ixgbe_release_hw_control(adapter);
6972 iounmap(adapter->hw.hw_addr);
6973 pci_release_selected_regions(pdev, pci_select_bars(pdev,
6976 e_dev_info("complete\n");
6978 free_netdev(netdev);
6980 pci_disable_pcie_error_reporting(pdev);
6982 pci_disable_device(pdev);
6986 * ixgbe_io_error_detected - called when PCI error is detected
6987 * @pdev: Pointer to PCI device
6988 * @state: The current pci connection state
6990 * This function is called after a PCI bus error affecting
6991 * this device has been detected.
6993 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
6994 pci_channel_state_t state)
6996 struct net_device *netdev = pci_get_drvdata(pdev);
6997 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6999 netif_device_detach(netdev);
7001 if (state == pci_channel_io_perm_failure)
7002 return PCI_ERS_RESULT_DISCONNECT;
7004 if (netif_running(netdev))
7005 ixgbe_down(adapter);
7006 pci_disable_device(pdev);
7008 /* Request a slot reset. */
7009 return PCI_ERS_RESULT_NEED_RESET;
7013 * ixgbe_io_slot_reset - called after the pci bus has been reset.
7014 * @pdev: Pointer to PCI device
7016 * Restart the card from scratch, as if from a cold-boot.
7018 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7020 struct net_device *netdev = pci_get_drvdata(pdev);
7021 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7022 pci_ers_result_t result;
7025 if (pci_enable_device_mem(pdev)) {
7026 e_err(probe, "Cannot re-enable PCI device after reset.\n");
7027 result = PCI_ERS_RESULT_DISCONNECT;
7029 pci_set_master(pdev);
7030 pci_restore_state(pdev);
7031 pci_save_state(pdev);
7033 pci_wake_from_d3(pdev, false);
7035 ixgbe_reset(adapter);
7036 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7037 result = PCI_ERS_RESULT_RECOVERED;
7040 err = pci_cleanup_aer_uncorrect_error_status(pdev);
7042 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7043 "failed 0x%0x\n", err);
7044 /* non-fatal, continue */
7051 * ixgbe_io_resume - called when traffic can start flowing again.
7052 * @pdev: Pointer to PCI device
7054 * This callback is called when the error recovery driver tells us that
7055 * its OK to resume normal operation.
7057 static void ixgbe_io_resume(struct pci_dev *pdev)
7059 struct net_device *netdev = pci_get_drvdata(pdev);
7060 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7062 if (netif_running(netdev)) {
7063 if (ixgbe_up(adapter)) {
7064 e_info(probe, "ixgbe_up failed after reset\n");
7069 netif_device_attach(netdev);
7072 static struct pci_error_handlers ixgbe_err_handler = {
7073 .error_detected = ixgbe_io_error_detected,
7074 .slot_reset = ixgbe_io_slot_reset,
7075 .resume = ixgbe_io_resume,
7078 static struct pci_driver ixgbe_driver = {
7079 .name = ixgbe_driver_name,
7080 .id_table = ixgbe_pci_tbl,
7081 .probe = ixgbe_probe,
7082 .remove = __devexit_p(ixgbe_remove),
7084 .suspend = ixgbe_suspend,
7085 .resume = ixgbe_resume,
7087 .shutdown = ixgbe_shutdown,
7088 .err_handler = &ixgbe_err_handler
7092 * ixgbe_init_module - Driver Registration Routine
7094 * ixgbe_init_module is the first routine called when the driver is
7095 * loaded. All it does is register with the PCI subsystem.
7097 static int __init ixgbe_init_module(void)
7100 pr_info("%s - version %s\n", ixgbe_driver_string,
7101 ixgbe_driver_version);
7102 pr_info("%s\n", ixgbe_copyright);
7104 #ifdef CONFIG_IXGBE_DCA
7105 dca_register_notify(&dca_notifier);
7108 ret = pci_register_driver(&ixgbe_driver);
7112 module_init(ixgbe_init_module);
7115 * ixgbe_exit_module - Driver Exit Cleanup Routine
7117 * ixgbe_exit_module is called just before the driver is removed
7120 static void __exit ixgbe_exit_module(void)
7122 #ifdef CONFIG_IXGBE_DCA
7123 dca_unregister_notify(&dca_notifier);
7125 pci_unregister_driver(&ixgbe_driver);
7128 #ifdef CONFIG_IXGBE_DCA
7129 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
7134 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
7135 __ixgbe_notify_dca);
7137 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7140 #endif /* CONFIG_IXGBE_DCA */
7143 * ixgbe_get_hw_dev return device
7144 * used by hardware layer to print debugging information
7146 struct net_device *ixgbe_get_hw_dev(struct ixgbe_hw *hw)
7148 struct ixgbe_adapter *adapter = hw->back;
7149 return adapter->netdev;
7152 module_exit(ixgbe_exit_module);