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 "2.0.62-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 for ( ; !cleaned; count++) {
753 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
754 tx_buffer_info = &tx_ring->tx_buffer_info[i];
755 cleaned = (i == eop);
756 skb = tx_buffer_info->skb;
758 if (cleaned && skb) {
759 unsigned int segs, bytecount;
760 unsigned int hlen = skb_headlen(skb);
762 /* gso_segs is currently only valid for tcp */
763 segs = skb_shinfo(skb)->gso_segs ?: 1;
765 /* adjust for FCoE Sequence Offload */
766 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
767 && (skb->protocol == htons(ETH_P_FCOE)) &&
769 hlen = skb_transport_offset(skb) +
770 sizeof(struct fc_frame_header) +
771 sizeof(struct fcoe_crc_eof);
772 segs = DIV_ROUND_UP(skb->len - hlen,
773 skb_shinfo(skb)->gso_size);
775 #endif /* IXGBE_FCOE */
776 /* multiply data chunks by size of headers */
777 bytecount = ((segs - 1) * hlen) + skb->len;
778 total_packets += segs;
779 total_bytes += bytecount;
782 ixgbe_unmap_and_free_tx_resource(adapter,
785 tx_desc->wb.status = 0;
788 if (i == tx_ring->count)
792 eop = tx_ring->tx_buffer_info[i].next_to_watch;
793 eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
796 tx_ring->next_to_clean = i;
798 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
799 if (unlikely(count && netif_carrier_ok(netdev) &&
800 (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
801 /* Make sure that anybody stopping the queue after this
802 * sees the new next_to_clean.
805 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
806 !test_bit(__IXGBE_DOWN, &adapter->state)) {
807 netif_wake_subqueue(netdev, tx_ring->queue_index);
808 ++tx_ring->restart_queue;
812 if (adapter->detect_tx_hung) {
813 if (ixgbe_check_tx_hang(adapter, tx_ring, i)) {
814 /* schedule immediate reset if we believe we hung */
815 e_info(probe, "tx hang %d detected, resetting "
816 "adapter\n", adapter->tx_timeout_count + 1);
817 ixgbe_tx_timeout(adapter->netdev);
821 /* re-arm the interrupt */
822 if (count >= tx_ring->work_limit)
823 ixgbe_irq_rearm_queues(adapter, ((u64)1 << q_vector->v_idx));
825 tx_ring->total_bytes += total_bytes;
826 tx_ring->total_packets += total_packets;
827 tx_ring->stats.packets += total_packets;
828 tx_ring->stats.bytes += total_bytes;
829 return (count < tx_ring->work_limit);
832 #ifdef CONFIG_IXGBE_DCA
833 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
834 struct ixgbe_ring *rx_ring)
838 int q = rx_ring->reg_idx;
840 if (rx_ring->cpu != cpu) {
841 rxctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q));
842 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
843 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
844 rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
845 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
846 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK_82599;
847 rxctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
848 IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599);
850 rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
851 rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
852 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_RRO_EN);
853 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_WRO_EN |
854 IXGBE_DCA_RXCTRL_DESC_HSRO_EN);
855 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q), rxctrl);
861 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
862 struct ixgbe_ring *tx_ring)
866 int q = tx_ring->reg_idx;
867 struct ixgbe_hw *hw = &adapter->hw;
869 if (tx_ring->cpu != cpu) {
870 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
871 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(q));
872 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
873 txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
874 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
875 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(q), txctrl);
876 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
877 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(q));
878 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK_82599;
879 txctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
880 IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599);
881 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
882 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(q), txctrl);
889 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
893 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
896 /* always use CB2 mode, difference is masked in the CB driver */
897 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
899 for (i = 0; i < adapter->num_tx_queues; i++) {
900 adapter->tx_ring[i]->cpu = -1;
901 ixgbe_update_tx_dca(adapter, adapter->tx_ring[i]);
903 for (i = 0; i < adapter->num_rx_queues; i++) {
904 adapter->rx_ring[i]->cpu = -1;
905 ixgbe_update_rx_dca(adapter, adapter->rx_ring[i]);
909 static int __ixgbe_notify_dca(struct device *dev, void *data)
911 struct net_device *netdev = dev_get_drvdata(dev);
912 struct ixgbe_adapter *adapter = netdev_priv(netdev);
913 unsigned long event = *(unsigned long *)data;
916 case DCA_PROVIDER_ADD:
917 /* if we're already enabled, don't do it again */
918 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
920 if (dca_add_requester(dev) == 0) {
921 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
922 ixgbe_setup_dca(adapter);
925 /* Fall Through since DCA is disabled. */
926 case DCA_PROVIDER_REMOVE:
927 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
928 dca_remove_requester(dev);
929 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
930 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
938 #endif /* CONFIG_IXGBE_DCA */
940 * ixgbe_receive_skb - Send a completed packet up the stack
941 * @adapter: board private structure
942 * @skb: packet to send up
943 * @status: hardware indication of status of receive
944 * @rx_ring: rx descriptor ring (for a specific queue) to setup
945 * @rx_desc: rx descriptor
947 static void ixgbe_receive_skb(struct ixgbe_q_vector *q_vector,
948 struct sk_buff *skb, u8 status,
949 struct ixgbe_ring *ring,
950 union ixgbe_adv_rx_desc *rx_desc)
952 struct ixgbe_adapter *adapter = q_vector->adapter;
953 struct napi_struct *napi = &q_vector->napi;
954 bool is_vlan = (status & IXGBE_RXD_STAT_VP);
955 u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
957 skb_record_rx_queue(skb, ring->queue_index);
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 pci_dev *pdev = adapter->pdev;
1040 union ixgbe_adv_rx_desc *rx_desc;
1041 struct ixgbe_rx_buffer *bi;
1044 i = rx_ring->next_to_use;
1045 bi = &rx_ring->rx_buffer_info[i];
1047 while (cleaned_count--) {
1048 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
1050 if (!bi->page_dma &&
1051 (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED)) {
1053 bi->page = alloc_page(GFP_ATOMIC);
1055 adapter->alloc_rx_page_failed++;
1058 bi->page_offset = 0;
1060 /* use a half page if we're re-using */
1061 bi->page_offset ^= (PAGE_SIZE / 2);
1064 bi->page_dma = dma_map_page(&pdev->dev, bi->page,
1071 struct sk_buff *skb;
1072 /* netdev_alloc_skb reserves 32 bytes up front!! */
1073 uint bufsz = rx_ring->rx_buf_len + SMP_CACHE_BYTES;
1074 skb = netdev_alloc_skb(adapter->netdev, bufsz);
1077 adapter->alloc_rx_buff_failed++;
1081 /* advance the data pointer to the next cache line */
1082 skb_reserve(skb, (PTR_ALIGN(skb->data, SMP_CACHE_BYTES)
1086 bi->dma = dma_map_single(&pdev->dev, skb->data,
1087 rx_ring->rx_buf_len,
1090 /* Refresh the desc even if buffer_addrs didn't change because
1091 * each write-back erases this info. */
1092 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
1093 rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
1094 rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
1096 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
1100 if (i == rx_ring->count)
1102 bi = &rx_ring->rx_buffer_info[i];
1106 if (rx_ring->next_to_use != i) {
1107 rx_ring->next_to_use = i;
1109 i = (rx_ring->count - 1);
1111 ixgbe_release_rx_desc(&adapter->hw, rx_ring, i);
1115 static inline u16 ixgbe_get_hdr_info(union ixgbe_adv_rx_desc *rx_desc)
1117 return rx_desc->wb.lower.lo_dword.hs_rss.hdr_info;
1120 static inline u16 ixgbe_get_pkt_info(union ixgbe_adv_rx_desc *rx_desc)
1122 return rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1125 static inline u32 ixgbe_get_rsc_count(union ixgbe_adv_rx_desc *rx_desc)
1127 return (le32_to_cpu(rx_desc->wb.lower.lo_dword.data) &
1128 IXGBE_RXDADV_RSCCNT_MASK) >>
1129 IXGBE_RXDADV_RSCCNT_SHIFT;
1133 * ixgbe_transform_rsc_queue - change rsc queue into a full packet
1134 * @skb: pointer to the last skb in the rsc queue
1135 * @count: pointer to number of packets coalesced in this context
1137 * This function changes a queue full of hw rsc buffers into a completed
1138 * packet. It uses the ->prev pointers to find the first packet and then
1139 * turns it into the frag list owner.
1141 static inline struct sk_buff *ixgbe_transform_rsc_queue(struct sk_buff *skb,
1144 unsigned int frag_list_size = 0;
1147 struct sk_buff *prev = skb->prev;
1148 frag_list_size += skb->len;
1154 skb_shinfo(skb)->frag_list = skb->next;
1156 skb->len += frag_list_size;
1157 skb->data_len += frag_list_size;
1158 skb->truesize += frag_list_size;
1162 struct ixgbe_rsc_cb {
1167 #define IXGBE_RSC_CB(skb) ((struct ixgbe_rsc_cb *)(skb)->cb)
1169 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1170 struct ixgbe_ring *rx_ring,
1171 int *work_done, int work_to_do)
1173 struct ixgbe_adapter *adapter = q_vector->adapter;
1174 struct net_device *netdev = adapter->netdev;
1175 struct pci_dev *pdev = adapter->pdev;
1176 union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
1177 struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
1178 struct sk_buff *skb;
1179 unsigned int i, rsc_count = 0;
1182 bool cleaned = false;
1183 int cleaned_count = 0;
1184 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1187 #endif /* IXGBE_FCOE */
1189 i = rx_ring->next_to_clean;
1190 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
1191 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1192 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1194 while (staterr & IXGBE_RXD_STAT_DD) {
1196 if (*work_done >= work_to_do)
1200 rmb(); /* read descriptor and rx_buffer_info after status DD */
1201 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
1202 hdr_info = le16_to_cpu(ixgbe_get_hdr_info(rx_desc));
1203 len = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
1204 IXGBE_RXDADV_HDRBUFLEN_SHIFT;
1205 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1206 if ((len > IXGBE_RX_HDR_SIZE) ||
1207 (upper_len && !(hdr_info & IXGBE_RXDADV_SPH)))
1208 len = IXGBE_RX_HDR_SIZE;
1210 len = le16_to_cpu(rx_desc->wb.upper.length);
1214 skb = rx_buffer_info->skb;
1215 prefetch(skb->data);
1216 rx_buffer_info->skb = NULL;
1218 if (rx_buffer_info->dma) {
1219 if ((adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
1220 (!(staterr & IXGBE_RXD_STAT_EOP)) &&
1223 * When HWRSC is enabled, delay unmapping
1224 * of the first packet. It carries the
1225 * header information, HW may still
1226 * access the header after the writeback.
1227 * Only unmap it when EOP is reached
1229 IXGBE_RSC_CB(skb)->delay_unmap = true;
1230 IXGBE_RSC_CB(skb)->dma = rx_buffer_info->dma;
1232 dma_unmap_single(&pdev->dev,
1233 rx_buffer_info->dma,
1234 rx_ring->rx_buf_len,
1237 rx_buffer_info->dma = 0;
1242 dma_unmap_page(&pdev->dev, rx_buffer_info->page_dma,
1243 PAGE_SIZE / 2, DMA_FROM_DEVICE);
1244 rx_buffer_info->page_dma = 0;
1245 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1246 rx_buffer_info->page,
1247 rx_buffer_info->page_offset,
1250 if ((rx_ring->rx_buf_len > (PAGE_SIZE / 2)) ||
1251 (page_count(rx_buffer_info->page) != 1))
1252 rx_buffer_info->page = NULL;
1254 get_page(rx_buffer_info->page);
1256 skb->len += upper_len;
1257 skb->data_len += upper_len;
1258 skb->truesize += upper_len;
1262 if (i == rx_ring->count)
1265 next_rxd = IXGBE_RX_DESC_ADV(*rx_ring, i);
1269 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
1270 rsc_count = ixgbe_get_rsc_count(rx_desc);
1273 u32 nextp = (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
1274 IXGBE_RXDADV_NEXTP_SHIFT;
1275 next_buffer = &rx_ring->rx_buffer_info[nextp];
1277 next_buffer = &rx_ring->rx_buffer_info[i];
1280 if (staterr & IXGBE_RXD_STAT_EOP) {
1282 skb = ixgbe_transform_rsc_queue(skb, &(rx_ring->rsc_count));
1283 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
1284 if (IXGBE_RSC_CB(skb)->delay_unmap) {
1285 dma_unmap_single(&pdev->dev,
1286 IXGBE_RSC_CB(skb)->dma,
1287 rx_ring->rx_buf_len,
1289 IXGBE_RSC_CB(skb)->dma = 0;
1290 IXGBE_RSC_CB(skb)->delay_unmap = false;
1292 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED)
1293 rx_ring->rsc_count += skb_shinfo(skb)->nr_frags;
1295 rx_ring->rsc_count++;
1296 rx_ring->rsc_flush++;
1298 rx_ring->stats.packets++;
1299 rx_ring->stats.bytes += skb->len;
1301 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
1302 rx_buffer_info->skb = next_buffer->skb;
1303 rx_buffer_info->dma = next_buffer->dma;
1304 next_buffer->skb = skb;
1305 next_buffer->dma = 0;
1307 skb->next = next_buffer->skb;
1308 skb->next->prev = skb;
1310 rx_ring->non_eop_descs++;
1314 if (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) {
1315 dev_kfree_skb_irq(skb);
1319 ixgbe_rx_checksum(adapter, rx_desc, skb);
1321 /* probably a little skewed due to removing CRC */
1322 total_rx_bytes += skb->len;
1325 skb->protocol = eth_type_trans(skb, adapter->netdev);
1327 /* if ddp, not passing to ULD unless for FCP_RSP or error */
1328 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
1329 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
1333 #endif /* IXGBE_FCOE */
1334 ixgbe_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
1337 rx_desc->wb.upper.status_error = 0;
1339 /* return some buffers to hardware, one at a time is too slow */
1340 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
1341 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
1345 /* use prefetched values */
1347 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1349 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1352 rx_ring->next_to_clean = i;
1353 cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
1356 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
1359 /* include DDPed FCoE data */
1360 if (ddp_bytes > 0) {
1363 mss = adapter->netdev->mtu - sizeof(struct fcoe_hdr) -
1364 sizeof(struct fc_frame_header) -
1365 sizeof(struct fcoe_crc_eof);
1368 total_rx_bytes += ddp_bytes;
1369 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
1371 #endif /* IXGBE_FCOE */
1373 rx_ring->total_packets += total_rx_packets;
1374 rx_ring->total_bytes += total_rx_bytes;
1375 netdev->stats.rx_bytes += total_rx_bytes;
1376 netdev->stats.rx_packets += total_rx_packets;
1381 static int ixgbe_clean_rxonly(struct napi_struct *, int);
1383 * ixgbe_configure_msix - Configure MSI-X hardware
1384 * @adapter: board private structure
1386 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1389 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1391 struct ixgbe_q_vector *q_vector;
1392 int i, j, q_vectors, v_idx, r_idx;
1395 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1398 * Populate the IVAR table and set the ITR values to the
1399 * corresponding register.
1401 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1402 q_vector = adapter->q_vector[v_idx];
1403 /* XXX for_each_set_bit(...) */
1404 r_idx = find_first_bit(q_vector->rxr_idx,
1405 adapter->num_rx_queues);
1407 for (i = 0; i < q_vector->rxr_count; i++) {
1408 j = adapter->rx_ring[r_idx]->reg_idx;
1409 ixgbe_set_ivar(adapter, 0, j, v_idx);
1410 r_idx = find_next_bit(q_vector->rxr_idx,
1411 adapter->num_rx_queues,
1414 r_idx = find_first_bit(q_vector->txr_idx,
1415 adapter->num_tx_queues);
1417 for (i = 0; i < q_vector->txr_count; i++) {
1418 j = adapter->tx_ring[r_idx]->reg_idx;
1419 ixgbe_set_ivar(adapter, 1, j, v_idx);
1420 r_idx = find_next_bit(q_vector->txr_idx,
1421 adapter->num_tx_queues,
1425 if (q_vector->txr_count && !q_vector->rxr_count)
1427 q_vector->eitr = adapter->tx_eitr_param;
1428 else if (q_vector->rxr_count)
1430 q_vector->eitr = adapter->rx_eitr_param;
1432 ixgbe_write_eitr(q_vector);
1435 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
1436 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1438 else if (adapter->hw.mac.type == ixgbe_mac_82599EB)
1439 ixgbe_set_ivar(adapter, -1, 1, v_idx);
1440 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1442 /* set up to autoclear timer, and the vectors */
1443 mask = IXGBE_EIMS_ENABLE_MASK;
1444 if (adapter->num_vfs)
1445 mask &= ~(IXGBE_EIMS_OTHER |
1446 IXGBE_EIMS_MAILBOX |
1449 mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
1450 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1453 enum latency_range {
1457 latency_invalid = 255
1461 * ixgbe_update_itr - update the dynamic ITR value based on statistics
1462 * @adapter: pointer to adapter
1463 * @eitr: eitr setting (ints per sec) to give last timeslice
1464 * @itr_setting: current throttle rate in ints/second
1465 * @packets: the number of packets during this measurement interval
1466 * @bytes: the number of bytes during this measurement interval
1468 * Stores a new ITR value based on packets and byte
1469 * counts during the last interrupt. The advantage of per interrupt
1470 * computation is faster updates and more accurate ITR for the current
1471 * traffic pattern. Constants in this function were computed
1472 * based on theoretical maximum wire speed and thresholds were set based
1473 * on testing data as well as attempting to minimize response time
1474 * while increasing bulk throughput.
1475 * this functionality is controlled by the InterruptThrottleRate module
1476 * parameter (see ixgbe_param.c)
1478 static u8 ixgbe_update_itr(struct ixgbe_adapter *adapter,
1479 u32 eitr, u8 itr_setting,
1480 int packets, int bytes)
1482 unsigned int retval = itr_setting;
1487 goto update_itr_done;
1490 /* simple throttlerate management
1491 * 0-20MB/s lowest (100000 ints/s)
1492 * 20-100MB/s low (20000 ints/s)
1493 * 100-1249MB/s bulk (8000 ints/s)
1495 /* what was last interrupt timeslice? */
1496 timepassed_us = 1000000/eitr;
1497 bytes_perint = bytes / timepassed_us; /* bytes/usec */
1499 switch (itr_setting) {
1500 case lowest_latency:
1501 if (bytes_perint > adapter->eitr_low)
1502 retval = low_latency;
1505 if (bytes_perint > adapter->eitr_high)
1506 retval = bulk_latency;
1507 else if (bytes_perint <= adapter->eitr_low)
1508 retval = lowest_latency;
1511 if (bytes_perint <= adapter->eitr_high)
1512 retval = low_latency;
1521 * ixgbe_write_eitr - write EITR register in hardware specific way
1522 * @q_vector: structure containing interrupt and ring information
1524 * This function is made to be called by ethtool and by the driver
1525 * when it needs to update EITR registers at runtime. Hardware
1526 * specific quirks/differences are taken care of here.
1528 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1530 struct ixgbe_adapter *adapter = q_vector->adapter;
1531 struct ixgbe_hw *hw = &adapter->hw;
1532 int v_idx = q_vector->v_idx;
1533 u32 itr_reg = EITR_INTS_PER_SEC_TO_REG(q_vector->eitr);
1535 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1536 /* must write high and low 16 bits to reset counter */
1537 itr_reg |= (itr_reg << 16);
1538 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1540 * 82599 can support a value of zero, so allow it for
1541 * max interrupt rate, but there is an errata where it can
1542 * not be zero with RSC
1545 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
1549 * set the WDIS bit to not clear the timer bits and cause an
1550 * immediate assertion of the interrupt
1552 itr_reg |= IXGBE_EITR_CNT_WDIS;
1554 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1557 static void ixgbe_set_itr_msix(struct ixgbe_q_vector *q_vector)
1559 struct ixgbe_adapter *adapter = q_vector->adapter;
1561 u8 current_itr, ret_itr;
1563 struct ixgbe_ring *rx_ring, *tx_ring;
1565 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1566 for (i = 0; i < q_vector->txr_count; i++) {
1567 tx_ring = adapter->tx_ring[r_idx];
1568 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1570 tx_ring->total_packets,
1571 tx_ring->total_bytes);
1572 /* if the result for this queue would decrease interrupt
1573 * rate for this vector then use that result */
1574 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
1575 q_vector->tx_itr - 1 : ret_itr);
1576 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1580 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1581 for (i = 0; i < q_vector->rxr_count; i++) {
1582 rx_ring = adapter->rx_ring[r_idx];
1583 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1585 rx_ring->total_packets,
1586 rx_ring->total_bytes);
1587 /* if the result for this queue would decrease interrupt
1588 * rate for this vector then use that result */
1589 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
1590 q_vector->rx_itr - 1 : ret_itr);
1591 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1595 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1597 switch (current_itr) {
1598 /* counts and packets in update_itr are dependent on these numbers */
1599 case lowest_latency:
1603 new_itr = 20000; /* aka hwitr = ~200 */
1611 if (new_itr != q_vector->eitr) {
1612 /* do an exponential smoothing */
1613 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1615 /* save the algorithm value here, not the smoothed one */
1616 q_vector->eitr = new_itr;
1618 ixgbe_write_eitr(q_vector);
1623 * ixgbe_check_overtemp_task - worker thread to check over tempurature
1624 * @work: pointer to work_struct containing our data
1626 static void ixgbe_check_overtemp_task(struct work_struct *work)
1628 struct ixgbe_adapter *adapter = container_of(work,
1629 struct ixgbe_adapter,
1630 check_overtemp_task);
1631 struct ixgbe_hw *hw = &adapter->hw;
1632 u32 eicr = adapter->interrupt_event;
1634 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
1635 switch (hw->device_id) {
1636 case IXGBE_DEV_ID_82599_T3_LOM: {
1638 bool link_up = false;
1640 if (hw->mac.ops.check_link)
1641 hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
1643 if (((eicr & IXGBE_EICR_GPI_SDP0) && (!link_up)) ||
1644 (eicr & IXGBE_EICR_LSC))
1645 /* Check if this is due to overtemp */
1646 if (hw->phy.ops.check_overtemp(hw) == IXGBE_ERR_OVERTEMP)
1651 if (!(eicr & IXGBE_EICR_GPI_SDP0))
1655 e_crit(drv, "Network adapter has been stopped because it has "
1656 "over heated. Restart the computer. If the problem "
1657 "persists, power off the system and replace the "
1659 /* write to clear the interrupt */
1660 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP0);
1664 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
1666 struct ixgbe_hw *hw = &adapter->hw;
1668 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
1669 (eicr & IXGBE_EICR_GPI_SDP1)) {
1670 e_crit(probe, "Fan has stopped, replace the adapter\n");
1671 /* write to clear the interrupt */
1672 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1676 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
1678 struct ixgbe_hw *hw = &adapter->hw;
1680 if (eicr & IXGBE_EICR_GPI_SDP1) {
1681 /* Clear the interrupt */
1682 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1683 schedule_work(&adapter->multispeed_fiber_task);
1684 } else if (eicr & IXGBE_EICR_GPI_SDP2) {
1685 /* Clear the interrupt */
1686 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
1687 schedule_work(&adapter->sfp_config_module_task);
1689 /* Interrupt isn't for us... */
1694 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1696 struct ixgbe_hw *hw = &adapter->hw;
1699 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1700 adapter->link_check_timeout = jiffies;
1701 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1702 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1703 IXGBE_WRITE_FLUSH(hw);
1704 schedule_work(&adapter->watchdog_task);
1708 static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
1710 struct net_device *netdev = data;
1711 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1712 struct ixgbe_hw *hw = &adapter->hw;
1716 * Workaround for Silicon errata. Use clear-by-write instead
1717 * of clear-by-read. Reading with EICS will return the
1718 * interrupt causes without clearing, which later be done
1719 * with the write to EICR.
1721 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1722 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1724 if (eicr & IXGBE_EICR_LSC)
1725 ixgbe_check_lsc(adapter);
1727 if (eicr & IXGBE_EICR_MAILBOX)
1728 ixgbe_msg_task(adapter);
1730 if (hw->mac.type == ixgbe_mac_82598EB)
1731 ixgbe_check_fan_failure(adapter, eicr);
1733 if (hw->mac.type == ixgbe_mac_82599EB) {
1734 ixgbe_check_sfp_event(adapter, eicr);
1735 adapter->interrupt_event = eicr;
1736 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
1737 ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)))
1738 schedule_work(&adapter->check_overtemp_task);
1740 /* Handle Flow Director Full threshold interrupt */
1741 if (eicr & IXGBE_EICR_FLOW_DIR) {
1743 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_FLOW_DIR);
1744 /* Disable transmits before FDIR Re-initialization */
1745 netif_tx_stop_all_queues(netdev);
1746 for (i = 0; i < adapter->num_tx_queues; i++) {
1747 struct ixgbe_ring *tx_ring =
1748 adapter->tx_ring[i];
1749 if (test_and_clear_bit(__IXGBE_FDIR_INIT_DONE,
1750 &tx_ring->reinit_state))
1751 schedule_work(&adapter->fdir_reinit_task);
1755 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1756 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_OTHER);
1761 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
1766 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1767 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1768 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1770 mask = (qmask & 0xFFFFFFFF);
1771 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(0), mask);
1772 mask = (qmask >> 32);
1773 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(1), mask);
1775 /* skip the flush */
1778 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
1783 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1784 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1785 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, mask);
1787 mask = (qmask & 0xFFFFFFFF);
1788 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), mask);
1789 mask = (qmask >> 32);
1790 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), mask);
1792 /* skip the flush */
1795 static irqreturn_t ixgbe_msix_clean_tx(int irq, void *data)
1797 struct ixgbe_q_vector *q_vector = data;
1798 struct ixgbe_adapter *adapter = q_vector->adapter;
1799 struct ixgbe_ring *tx_ring;
1802 if (!q_vector->txr_count)
1805 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1806 for (i = 0; i < q_vector->txr_count; i++) {
1807 tx_ring = adapter->tx_ring[r_idx];
1808 tx_ring->total_bytes = 0;
1809 tx_ring->total_packets = 0;
1810 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1814 /* EIAM disabled interrupts (on this vector) for us */
1815 napi_schedule(&q_vector->napi);
1821 * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
1823 * @data: pointer to our q_vector struct for this interrupt vector
1825 static irqreturn_t ixgbe_msix_clean_rx(int irq, void *data)
1827 struct ixgbe_q_vector *q_vector = data;
1828 struct ixgbe_adapter *adapter = q_vector->adapter;
1829 struct ixgbe_ring *rx_ring;
1833 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1834 for (i = 0; i < q_vector->rxr_count; i++) {
1835 rx_ring = adapter->rx_ring[r_idx];
1836 rx_ring->total_bytes = 0;
1837 rx_ring->total_packets = 0;
1838 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1842 if (!q_vector->rxr_count)
1845 /* disable interrupts on this vector only */
1846 /* EIAM disabled interrupts (on this vector) for us */
1847 napi_schedule(&q_vector->napi);
1852 static irqreturn_t ixgbe_msix_clean_many(int irq, void *data)
1854 struct ixgbe_q_vector *q_vector = data;
1855 struct ixgbe_adapter *adapter = q_vector->adapter;
1856 struct ixgbe_ring *ring;
1860 if (!q_vector->txr_count && !q_vector->rxr_count)
1863 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1864 for (i = 0; i < q_vector->txr_count; i++) {
1865 ring = adapter->tx_ring[r_idx];
1866 ring->total_bytes = 0;
1867 ring->total_packets = 0;
1868 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1872 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1873 for (i = 0; i < q_vector->rxr_count; i++) {
1874 ring = adapter->rx_ring[r_idx];
1875 ring->total_bytes = 0;
1876 ring->total_packets = 0;
1877 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1881 /* EIAM disabled interrupts (on this vector) for us */
1882 napi_schedule(&q_vector->napi);
1888 * ixgbe_clean_rxonly - msix (aka one shot) rx clean routine
1889 * @napi: napi struct with our devices info in it
1890 * @budget: amount of work driver is allowed to do this pass, in packets
1892 * This function is optimized for cleaning one queue only on a single
1895 static int ixgbe_clean_rxonly(struct napi_struct *napi, int budget)
1897 struct ixgbe_q_vector *q_vector =
1898 container_of(napi, struct ixgbe_q_vector, napi);
1899 struct ixgbe_adapter *adapter = q_vector->adapter;
1900 struct ixgbe_ring *rx_ring = NULL;
1904 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1905 rx_ring = adapter->rx_ring[r_idx];
1906 #ifdef CONFIG_IXGBE_DCA
1907 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1908 ixgbe_update_rx_dca(adapter, rx_ring);
1911 ixgbe_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
1913 /* If all Rx work done, exit the polling mode */
1914 if (work_done < budget) {
1915 napi_complete(napi);
1916 if (adapter->rx_itr_setting & 1)
1917 ixgbe_set_itr_msix(q_vector);
1918 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1919 ixgbe_irq_enable_queues(adapter,
1920 ((u64)1 << q_vector->v_idx));
1927 * ixgbe_clean_rxtx_many - msix (aka one shot) rx clean routine
1928 * @napi: napi struct with our devices info in it
1929 * @budget: amount of work driver is allowed to do this pass, in packets
1931 * This function will clean more than one rx queue associated with a
1934 static int ixgbe_clean_rxtx_many(struct napi_struct *napi, int budget)
1936 struct ixgbe_q_vector *q_vector =
1937 container_of(napi, struct ixgbe_q_vector, napi);
1938 struct ixgbe_adapter *adapter = q_vector->adapter;
1939 struct ixgbe_ring *ring = NULL;
1940 int work_done = 0, i;
1942 bool tx_clean_complete = true;
1944 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1945 for (i = 0; i < q_vector->txr_count; i++) {
1946 ring = adapter->tx_ring[r_idx];
1947 #ifdef CONFIG_IXGBE_DCA
1948 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1949 ixgbe_update_tx_dca(adapter, ring);
1951 tx_clean_complete &= ixgbe_clean_tx_irq(q_vector, ring);
1952 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1956 /* attempt to distribute budget to each queue fairly, but don't allow
1957 * the budget to go below 1 because we'll exit polling */
1958 budget /= (q_vector->rxr_count ?: 1);
1959 budget = max(budget, 1);
1960 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1961 for (i = 0; i < q_vector->rxr_count; i++) {
1962 ring = adapter->rx_ring[r_idx];
1963 #ifdef CONFIG_IXGBE_DCA
1964 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1965 ixgbe_update_rx_dca(adapter, ring);
1967 ixgbe_clean_rx_irq(q_vector, ring, &work_done, budget);
1968 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1972 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1973 ring = adapter->rx_ring[r_idx];
1974 /* If all Rx work done, exit the polling mode */
1975 if (work_done < budget) {
1976 napi_complete(napi);
1977 if (adapter->rx_itr_setting & 1)
1978 ixgbe_set_itr_msix(q_vector);
1979 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1980 ixgbe_irq_enable_queues(adapter,
1981 ((u64)1 << q_vector->v_idx));
1989 * ixgbe_clean_txonly - msix (aka one shot) tx clean routine
1990 * @napi: napi struct with our devices info in it
1991 * @budget: amount of work driver is allowed to do this pass, in packets
1993 * This function is optimized for cleaning one queue only on a single
1996 static int ixgbe_clean_txonly(struct napi_struct *napi, int budget)
1998 struct ixgbe_q_vector *q_vector =
1999 container_of(napi, struct ixgbe_q_vector, napi);
2000 struct ixgbe_adapter *adapter = q_vector->adapter;
2001 struct ixgbe_ring *tx_ring = NULL;
2005 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
2006 tx_ring = adapter->tx_ring[r_idx];
2007 #ifdef CONFIG_IXGBE_DCA
2008 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2009 ixgbe_update_tx_dca(adapter, tx_ring);
2012 if (!ixgbe_clean_tx_irq(q_vector, tx_ring))
2015 /* If all Tx work done, exit the polling mode */
2016 if (work_done < budget) {
2017 napi_complete(napi);
2018 if (adapter->tx_itr_setting & 1)
2019 ixgbe_set_itr_msix(q_vector);
2020 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2021 ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
2027 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
2030 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2032 set_bit(r_idx, q_vector->rxr_idx);
2033 q_vector->rxr_count++;
2036 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
2039 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2041 set_bit(t_idx, q_vector->txr_idx);
2042 q_vector->txr_count++;
2046 * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
2047 * @adapter: board private structure to initialize
2048 * @vectors: allotted vector count for descriptor rings
2050 * This function maps descriptor rings to the queue-specific vectors
2051 * we were allotted through the MSI-X enabling code. Ideally, we'd have
2052 * one vector per ring/queue, but on a constrained vector budget, we
2053 * group the rings as "efficiently" as possible. You would add new
2054 * mapping configurations in here.
2056 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter,
2060 int rxr_idx = 0, txr_idx = 0;
2061 int rxr_remaining = adapter->num_rx_queues;
2062 int txr_remaining = adapter->num_tx_queues;
2067 /* No mapping required if MSI-X is disabled. */
2068 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2072 * The ideal configuration...
2073 * We have enough vectors to map one per queue.
2075 if (vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
2076 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
2077 map_vector_to_rxq(adapter, v_start, rxr_idx);
2079 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
2080 map_vector_to_txq(adapter, v_start, txr_idx);
2086 * If we don't have enough vectors for a 1-to-1
2087 * mapping, we'll have to group them so there are
2088 * multiple queues per vector.
2090 /* Re-adjusting *qpv takes care of the remainder. */
2091 for (i = v_start; i < vectors; i++) {
2092 rqpv = DIV_ROUND_UP(rxr_remaining, vectors - i);
2093 for (j = 0; j < rqpv; j++) {
2094 map_vector_to_rxq(adapter, i, rxr_idx);
2099 for (i = v_start; i < vectors; i++) {
2100 tqpv = DIV_ROUND_UP(txr_remaining, vectors - i);
2101 for (j = 0; j < tqpv; j++) {
2102 map_vector_to_txq(adapter, i, txr_idx);
2113 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2114 * @adapter: board private structure
2116 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2117 * interrupts from the kernel.
2119 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2121 struct net_device *netdev = adapter->netdev;
2122 irqreturn_t (*handler)(int, void *);
2123 int i, vector, q_vectors, err;
2126 /* Decrement for Other and TCP Timer vectors */
2127 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2129 /* Map the Tx/Rx rings to the vectors we were allotted. */
2130 err = ixgbe_map_rings_to_vectors(adapter, q_vectors);
2134 #define SET_HANDLER(_v) ((!(_v)->rxr_count) ? &ixgbe_msix_clean_tx : \
2135 (!(_v)->txr_count) ? &ixgbe_msix_clean_rx : \
2136 &ixgbe_msix_clean_many)
2137 for (vector = 0; vector < q_vectors; vector++) {
2138 handler = SET_HANDLER(adapter->q_vector[vector]);
2140 if(handler == &ixgbe_msix_clean_rx) {
2141 sprintf(adapter->name[vector], "%s-%s-%d",
2142 netdev->name, "rx", ri++);
2144 else if(handler == &ixgbe_msix_clean_tx) {
2145 sprintf(adapter->name[vector], "%s-%s-%d",
2146 netdev->name, "tx", ti++);
2149 sprintf(adapter->name[vector], "%s-%s-%d",
2150 netdev->name, "TxRx", vector);
2152 err = request_irq(adapter->msix_entries[vector].vector,
2153 handler, 0, adapter->name[vector],
2154 adapter->q_vector[vector]);
2156 e_err(probe, "request_irq failed for MSIX interrupt "
2157 "Error: %d\n", err);
2158 goto free_queue_irqs;
2162 sprintf(adapter->name[vector], "%s:lsc", netdev->name);
2163 err = request_irq(adapter->msix_entries[vector].vector,
2164 ixgbe_msix_lsc, 0, adapter->name[vector], netdev);
2166 e_err(probe, "request_irq for msix_lsc failed: %d\n", err);
2167 goto free_queue_irqs;
2173 for (i = vector - 1; i >= 0; i--)
2174 free_irq(adapter->msix_entries[--vector].vector,
2175 adapter->q_vector[i]);
2176 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2177 pci_disable_msix(adapter->pdev);
2178 kfree(adapter->msix_entries);
2179 adapter->msix_entries = NULL;
2184 static void ixgbe_set_itr(struct ixgbe_adapter *adapter)
2186 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2188 u32 new_itr = q_vector->eitr;
2189 struct ixgbe_ring *rx_ring = adapter->rx_ring[0];
2190 struct ixgbe_ring *tx_ring = adapter->tx_ring[0];
2192 q_vector->tx_itr = ixgbe_update_itr(adapter, new_itr,
2194 tx_ring->total_packets,
2195 tx_ring->total_bytes);
2196 q_vector->rx_itr = ixgbe_update_itr(adapter, new_itr,
2198 rx_ring->total_packets,
2199 rx_ring->total_bytes);
2201 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
2203 switch (current_itr) {
2204 /* counts and packets in update_itr are dependent on these numbers */
2205 case lowest_latency:
2209 new_itr = 20000; /* aka hwitr = ~200 */
2218 if (new_itr != q_vector->eitr) {
2219 /* do an exponential smoothing */
2220 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
2222 /* save the algorithm value here, not the smoothed one */
2223 q_vector->eitr = new_itr;
2225 ixgbe_write_eitr(q_vector);
2230 * ixgbe_irq_enable - Enable default interrupt generation settings
2231 * @adapter: board private structure
2233 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter)
2237 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2238 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2239 mask |= IXGBE_EIMS_GPI_SDP0;
2240 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2241 mask |= IXGBE_EIMS_GPI_SDP1;
2242 if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
2243 mask |= IXGBE_EIMS_ECC;
2244 mask |= IXGBE_EIMS_GPI_SDP1;
2245 mask |= IXGBE_EIMS_GPI_SDP2;
2246 if (adapter->num_vfs)
2247 mask |= IXGBE_EIMS_MAILBOX;
2249 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
2250 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
2251 mask |= IXGBE_EIMS_FLOW_DIR;
2253 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2254 ixgbe_irq_enable_queues(adapter, ~0);
2255 IXGBE_WRITE_FLUSH(&adapter->hw);
2257 if (adapter->num_vfs > 32) {
2258 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
2259 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2264 * ixgbe_intr - legacy mode Interrupt Handler
2265 * @irq: interrupt number
2266 * @data: pointer to a network interface device structure
2268 static irqreturn_t ixgbe_intr(int irq, void *data)
2270 struct net_device *netdev = data;
2271 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2272 struct ixgbe_hw *hw = &adapter->hw;
2273 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2277 * Workaround for silicon errata. Mask the interrupts
2278 * before the read of EICR.
2280 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2282 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2283 * therefore no explict interrupt disable is necessary */
2284 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2286 /* shared interrupt alert!
2287 * make sure interrupts are enabled because the read will
2288 * have disabled interrupts due to EIAM */
2289 ixgbe_irq_enable(adapter);
2290 return IRQ_NONE; /* Not our interrupt */
2293 if (eicr & IXGBE_EICR_LSC)
2294 ixgbe_check_lsc(adapter);
2296 if (hw->mac.type == ixgbe_mac_82599EB)
2297 ixgbe_check_sfp_event(adapter, eicr);
2299 ixgbe_check_fan_failure(adapter, eicr);
2300 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2301 ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)))
2302 schedule_work(&adapter->check_overtemp_task);
2304 if (napi_schedule_prep(&(q_vector->napi))) {
2305 adapter->tx_ring[0]->total_packets = 0;
2306 adapter->tx_ring[0]->total_bytes = 0;
2307 adapter->rx_ring[0]->total_packets = 0;
2308 adapter->rx_ring[0]->total_bytes = 0;
2309 /* would disable interrupts here but EIAM disabled it */
2310 __napi_schedule(&(q_vector->napi));
2316 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
2318 int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2320 for (i = 0; i < q_vectors; i++) {
2321 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
2322 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
2323 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
2324 q_vector->rxr_count = 0;
2325 q_vector->txr_count = 0;
2330 * ixgbe_request_irq - initialize interrupts
2331 * @adapter: board private structure
2333 * Attempts to configure interrupts using the best available
2334 * capabilities of the hardware and kernel.
2336 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2338 struct net_device *netdev = adapter->netdev;
2341 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2342 err = ixgbe_request_msix_irqs(adapter);
2343 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
2344 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2345 netdev->name, netdev);
2347 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2348 netdev->name, netdev);
2352 e_err(probe, "request_irq failed, Error %d\n", err);
2357 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2359 struct net_device *netdev = adapter->netdev;
2361 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2364 q_vectors = adapter->num_msix_vectors;
2367 free_irq(adapter->msix_entries[i].vector, netdev);
2370 for (; i >= 0; i--) {
2371 free_irq(adapter->msix_entries[i].vector,
2372 adapter->q_vector[i]);
2375 ixgbe_reset_q_vectors(adapter);
2377 free_irq(adapter->pdev->irq, netdev);
2382 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2383 * @adapter: board private structure
2385 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2387 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2388 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2390 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2391 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2392 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2393 if (adapter->num_vfs > 32)
2394 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
2396 IXGBE_WRITE_FLUSH(&adapter->hw);
2397 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2399 for (i = 0; i < adapter->num_msix_vectors; i++)
2400 synchronize_irq(adapter->msix_entries[i].vector);
2402 synchronize_irq(adapter->pdev->irq);
2407 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2410 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2412 struct ixgbe_hw *hw = &adapter->hw;
2414 IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
2415 EITR_INTS_PER_SEC_TO_REG(adapter->rx_eitr_param));
2417 ixgbe_set_ivar(adapter, 0, 0, 0);
2418 ixgbe_set_ivar(adapter, 1, 0, 0);
2420 map_vector_to_rxq(adapter, 0, 0);
2421 map_vector_to_txq(adapter, 0, 0);
2423 e_info(hw, "Legacy interrupt IVAR setup done\n");
2427 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2428 * @adapter: board private structure
2430 * Configure the Tx unit of the MAC after a reset.
2432 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2435 struct ixgbe_hw *hw = &adapter->hw;
2436 u32 i, j, tdlen, txctrl;
2438 /* Setup the HW Tx Head and Tail descriptor pointers */
2439 for (i = 0; i < adapter->num_tx_queues; i++) {
2440 struct ixgbe_ring *ring = adapter->tx_ring[i];
2443 tdlen = ring->count * sizeof(union ixgbe_adv_tx_desc);
2444 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(j),
2445 (tdba & DMA_BIT_MASK(32)));
2446 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(j), (tdba >> 32));
2447 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(j), tdlen);
2448 IXGBE_WRITE_REG(hw, IXGBE_TDH(j), 0);
2449 IXGBE_WRITE_REG(hw, IXGBE_TDT(j), 0);
2450 adapter->tx_ring[i]->head = IXGBE_TDH(j);
2451 adapter->tx_ring[i]->tail = IXGBE_TDT(j);
2453 * Disable Tx Head Writeback RO bit, since this hoses
2454 * bookkeeping if things aren't delivered in order.
2456 switch (hw->mac.type) {
2457 case ixgbe_mac_82598EB:
2458 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(j));
2460 case ixgbe_mac_82599EB:
2462 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(j));
2465 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
2466 switch (hw->mac.type) {
2467 case ixgbe_mac_82598EB:
2468 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(j), txctrl);
2470 case ixgbe_mac_82599EB:
2472 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(j), txctrl);
2477 if (hw->mac.type == ixgbe_mac_82599EB) {
2481 /* disable the arbiter while setting MTQC */
2482 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2483 rttdcs |= IXGBE_RTTDCS_ARBDIS;
2484 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2486 /* set transmit pool layout */
2487 mask = (IXGBE_FLAG_SRIOV_ENABLED | IXGBE_FLAG_DCB_ENABLED);
2488 switch (adapter->flags & mask) {
2490 case (IXGBE_FLAG_SRIOV_ENABLED):
2491 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2492 (IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF));
2495 case (IXGBE_FLAG_DCB_ENABLED):
2496 /* We enable 8 traffic classes, DCB only */
2497 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2498 (IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ));
2502 IXGBE_WRITE_REG(hw, IXGBE_MTQC, IXGBE_MTQC_64Q_1PB);
2506 /* re-eable the arbiter */
2507 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2508 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2512 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2514 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2515 struct ixgbe_ring *rx_ring)
2519 struct ixgbe_ring_feature *feature = adapter->ring_feature;
2521 index = rx_ring->reg_idx;
2522 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2524 mask = (unsigned long) feature[RING_F_RSS].mask;
2525 index = index & mask;
2527 srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(index));
2529 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
2530 srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
2532 srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
2533 IXGBE_SRRCTL_BSIZEHDR_MASK;
2535 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
2536 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
2537 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2539 srrctl |= (PAGE_SIZE / 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2541 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2543 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
2544 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2545 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2548 IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(index), srrctl);
2551 static u32 ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2556 if (!(adapter->hw.mac.type == ixgbe_mac_82599EB))
2559 mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2560 #ifdef CONFIG_IXGBE_DCB
2561 | IXGBE_FLAG_DCB_ENABLED
2563 | IXGBE_FLAG_SRIOV_ENABLED
2567 case (IXGBE_FLAG_RSS_ENABLED):
2568 mrqc = IXGBE_MRQC_RSSEN;
2570 case (IXGBE_FLAG_SRIOV_ENABLED):
2571 mrqc = IXGBE_MRQC_VMDQEN;
2573 #ifdef CONFIG_IXGBE_DCB
2574 case (IXGBE_FLAG_DCB_ENABLED):
2575 mrqc = IXGBE_MRQC_RT8TCEN;
2577 #endif /* CONFIG_IXGBE_DCB */
2586 * ixgbe_configure_rscctl - enable RSC for the indicated ring
2587 * @adapter: address of board private structure
2588 * @index: index of ring to set
2590 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter, int index)
2592 struct ixgbe_ring *rx_ring;
2593 struct ixgbe_hw *hw = &adapter->hw;
2598 rx_ring = adapter->rx_ring[index];
2599 j = rx_ring->reg_idx;
2600 rx_buf_len = rx_ring->rx_buf_len;
2601 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(j));
2602 rscctrl |= IXGBE_RSCCTL_RSCEN;
2604 * we must limit the number of descriptors so that the
2605 * total size of max desc * buf_len is not greater
2608 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
2609 #if (MAX_SKB_FRAGS > 16)
2610 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2611 #elif (MAX_SKB_FRAGS > 8)
2612 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2613 #elif (MAX_SKB_FRAGS > 4)
2614 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2616 rscctrl |= IXGBE_RSCCTL_MAXDESC_1;
2619 if (rx_buf_len < IXGBE_RXBUFFER_4096)
2620 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2621 else if (rx_buf_len < IXGBE_RXBUFFER_8192)
2622 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2624 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2626 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(j), rscctrl);
2630 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
2631 * @adapter: board private structure
2633 * Configure the Rx unit of the MAC after a reset.
2635 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
2638 struct ixgbe_hw *hw = &adapter->hw;
2639 struct ixgbe_ring *rx_ring;
2640 struct net_device *netdev = adapter->netdev;
2641 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2643 u32 rdlen, rxctrl, rxcsum;
2644 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2645 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2646 0x6A3E67EA, 0x14364D17, 0x3BED200D};
2648 u32 reta = 0, mrqc = 0;
2652 /* Decide whether to use packet split mode or not */
2653 /* Do not use packet split if we're in SR-IOV Mode */
2654 if (!adapter->num_vfs)
2655 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
2657 /* Set the RX buffer length according to the mode */
2658 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
2659 rx_buf_len = IXGBE_RX_HDR_SIZE;
2660 if (hw->mac.type == ixgbe_mac_82599EB) {
2661 /* PSRTYPE must be initialized in 82599 */
2662 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
2663 IXGBE_PSRTYPE_UDPHDR |
2664 IXGBE_PSRTYPE_IPV4HDR |
2665 IXGBE_PSRTYPE_IPV6HDR |
2666 IXGBE_PSRTYPE_L2HDR;
2668 IXGBE_PSRTYPE(adapter->num_vfs),
2672 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
2673 (netdev->mtu <= ETH_DATA_LEN))
2674 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
2676 rx_buf_len = ALIGN(max_frame, 1024);
2679 fctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_FCTRL);
2680 fctrl |= IXGBE_FCTRL_BAM;
2681 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
2682 fctrl |= IXGBE_FCTRL_PMCF;
2683 IXGBE_WRITE_REG(&adapter->hw, IXGBE_FCTRL, fctrl);
2685 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2686 if (adapter->netdev->mtu <= ETH_DATA_LEN)
2687 hlreg0 &= ~IXGBE_HLREG0_JUMBOEN;
2689 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2691 if (netdev->features & NETIF_F_FCOE_MTU)
2692 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2694 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
2696 rdlen = adapter->rx_ring[0]->count * sizeof(union ixgbe_adv_rx_desc);
2697 /* disable receives while setting up the descriptors */
2698 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2699 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
2702 * Setup the HW Rx Head and Tail Descriptor Pointers and
2703 * the Base and Length of the Rx Descriptor Ring
2705 for (i = 0; i < adapter->num_rx_queues; i++) {
2706 rx_ring = adapter->rx_ring[i];
2707 rdba = rx_ring->dma;
2708 j = rx_ring->reg_idx;
2709 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(j), (rdba & DMA_BIT_MASK(32)));
2710 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(j), (rdba >> 32));
2711 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(j), rdlen);
2712 IXGBE_WRITE_REG(hw, IXGBE_RDH(j), 0);
2713 IXGBE_WRITE_REG(hw, IXGBE_RDT(j), 0);
2714 rx_ring->head = IXGBE_RDH(j);
2715 rx_ring->tail = IXGBE_RDT(j);
2716 rx_ring->rx_buf_len = rx_buf_len;
2718 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)
2719 rx_ring->flags |= IXGBE_RING_RX_PS_ENABLED;
2721 rx_ring->flags &= ~IXGBE_RING_RX_PS_ENABLED;
2724 if (netdev->features & NETIF_F_FCOE_MTU) {
2725 struct ixgbe_ring_feature *f;
2726 f = &adapter->ring_feature[RING_F_FCOE];
2727 if ((i >= f->mask) && (i < f->mask + f->indices)) {
2728 rx_ring->flags &= ~IXGBE_RING_RX_PS_ENABLED;
2729 if (rx_buf_len < IXGBE_FCOE_JUMBO_FRAME_SIZE)
2730 rx_ring->rx_buf_len =
2731 IXGBE_FCOE_JUMBO_FRAME_SIZE;
2735 #endif /* IXGBE_FCOE */
2736 ixgbe_configure_srrctl(adapter, rx_ring);
2739 if (hw->mac.type == ixgbe_mac_82598EB) {
2741 * For VMDq support of different descriptor types or
2742 * buffer sizes through the use of multiple SRRCTL
2743 * registers, RDRXCTL.MVMEN must be set to 1
2745 * also, the manual doesn't mention it clearly but DCA hints
2746 * will only use queue 0's tags unless this bit is set. Side
2747 * effects of setting this bit are only that SRRCTL must be
2748 * fully programmed [0..15]
2750 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2751 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
2752 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
2755 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
2757 u32 reg_offset, vf_shift;
2758 u32 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
2759 vt_reg_bits = IXGBE_VMD_CTL_VMDQ_EN
2760 | IXGBE_VT_CTL_REPLEN;
2761 vt_reg_bits |= (adapter->num_vfs <<
2762 IXGBE_VT_CTL_POOL_SHIFT);
2763 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl | vt_reg_bits);
2764 IXGBE_WRITE_REG(hw, IXGBE_MRQC, 0);
2766 vf_shift = adapter->num_vfs % 32;
2767 reg_offset = adapter->num_vfs / 32;
2768 IXGBE_WRITE_REG(hw, IXGBE_VFRE(0), 0);
2769 IXGBE_WRITE_REG(hw, IXGBE_VFRE(1), 0);
2770 IXGBE_WRITE_REG(hw, IXGBE_VFTE(0), 0);
2771 IXGBE_WRITE_REG(hw, IXGBE_VFTE(1), 0);
2772 /* Enable only the PF's pool for Tx/Rx */
2773 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (1 << vf_shift));
2774 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (1 << vf_shift));
2775 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
2776 ixgbe_set_vmolr(hw, adapter->num_vfs, true);
2779 /* Program MRQC for the distribution of queues */
2780 mrqc = ixgbe_setup_mrqc(adapter);
2782 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
2783 /* Fill out redirection table */
2784 for (i = 0, j = 0; i < 128; i++, j++) {
2785 if (j == adapter->ring_feature[RING_F_RSS].indices)
2787 /* reta = 4-byte sliding window of
2788 * 0x00..(indices-1)(indices-1)00..etc. */
2789 reta = (reta << 8) | (j * 0x11);
2791 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2794 /* Fill out hash function seeds */
2795 for (i = 0; i < 10; i++)
2796 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2798 if (hw->mac.type == ixgbe_mac_82598EB)
2799 mrqc |= IXGBE_MRQC_RSSEN;
2800 /* Perform hash on these packet types */
2801 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2802 | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2803 | IXGBE_MRQC_RSS_FIELD_IPV4_UDP
2804 | IXGBE_MRQC_RSS_FIELD_IPV6
2805 | IXGBE_MRQC_RSS_FIELD_IPV6_TCP
2806 | IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
2808 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2810 if (adapter->num_vfs) {
2813 /* Map PF MAC address in RAR Entry 0 to first pool
2815 hw->mac.ops.set_vmdq(hw, 0, adapter->num_vfs);
2817 /* Set up VF register offsets for selected VT Mode, i.e.
2818 * 64 VFs for SR-IOV */
2819 reg = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
2820 reg |= IXGBE_GCR_EXT_SRIOV;
2821 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, reg);
2824 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2826 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED ||
2827 adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED) {
2828 /* Disable indicating checksum in descriptor, enables
2830 rxcsum |= IXGBE_RXCSUM_PCSD;
2832 if (!(rxcsum & IXGBE_RXCSUM_PCSD)) {
2833 /* Enable IPv4 payload checksum for UDP fragments
2834 * if PCSD is not set */
2835 rxcsum |= IXGBE_RXCSUM_IPPCSE;
2838 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2840 if (hw->mac.type == ixgbe_mac_82599EB) {
2841 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2842 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
2843 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
2844 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
2847 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
2848 /* Enable 82599 HW-RSC */
2849 for (i = 0; i < adapter->num_rx_queues; i++)
2850 ixgbe_configure_rscctl(adapter, i);
2852 /* Disable RSC for ACK packets */
2853 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
2854 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
2858 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2860 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2861 struct ixgbe_hw *hw = &adapter->hw;
2862 int pool_ndx = adapter->num_vfs;
2864 /* add VID to filter table */
2865 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, true);
2868 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2870 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2871 struct ixgbe_hw *hw = &adapter->hw;
2872 int pool_ndx = adapter->num_vfs;
2874 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2875 ixgbe_irq_disable(adapter);
2877 vlan_group_set_device(adapter->vlgrp, vid, NULL);
2879 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2880 ixgbe_irq_enable(adapter);
2882 /* remove VID from filter table */
2883 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, false);
2887 * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
2888 * @adapter: driver data
2890 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
2892 struct ixgbe_hw *hw = &adapter->hw;
2893 u32 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2896 switch (hw->mac.type) {
2897 case ixgbe_mac_82598EB:
2898 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
2899 #ifdef CONFIG_IXGBE_DCB
2900 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
2901 vlnctrl &= ~IXGBE_VLNCTRL_VME;
2903 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2904 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2906 case ixgbe_mac_82599EB:
2907 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
2908 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2909 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2910 #ifdef CONFIG_IXGBE_DCB
2911 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
2914 for (i = 0; i < adapter->num_rx_queues; i++) {
2915 j = adapter->rx_ring[i]->reg_idx;
2916 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
2917 vlnctrl &= ~IXGBE_RXDCTL_VME;
2918 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
2927 * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
2928 * @adapter: driver data
2930 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
2932 struct ixgbe_hw *hw = &adapter->hw;
2933 u32 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2936 switch (hw->mac.type) {
2937 case ixgbe_mac_82598EB:
2938 vlnctrl |= IXGBE_VLNCTRL_VME | IXGBE_VLNCTRL_VFE;
2939 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2940 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2942 case ixgbe_mac_82599EB:
2943 vlnctrl |= IXGBE_VLNCTRL_VFE;
2944 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2945 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2946 for (i = 0; i < adapter->num_rx_queues; i++) {
2947 j = adapter->rx_ring[i]->reg_idx;
2948 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
2949 vlnctrl |= IXGBE_RXDCTL_VME;
2950 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
2958 static void ixgbe_vlan_rx_register(struct net_device *netdev,
2959 struct vlan_group *grp)
2961 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2963 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2964 ixgbe_irq_disable(adapter);
2965 adapter->vlgrp = grp;
2968 * For a DCB driver, always enable VLAN tag stripping so we can
2969 * still receive traffic from a DCB-enabled host even if we're
2972 ixgbe_vlan_filter_enable(adapter);
2974 ixgbe_vlan_rx_add_vid(netdev, 0);
2976 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2977 ixgbe_irq_enable(adapter);
2980 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
2982 ixgbe_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2984 if (adapter->vlgrp) {
2986 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2987 if (!vlan_group_get_device(adapter->vlgrp, vid))
2989 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
2995 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
2996 * @netdev: network interface device structure
2998 * Writes unicast address list to the RAR table.
2999 * Returns: -ENOMEM on failure/insufficient address space
3000 * 0 on no addresses written
3001 * X on writing X addresses to the RAR table
3003 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3005 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3006 struct ixgbe_hw *hw = &adapter->hw;
3007 unsigned int vfn = adapter->num_vfs;
3008 unsigned int rar_entries = hw->mac.num_rar_entries - (vfn + 1);
3011 /* return ENOMEM indicating insufficient memory for addresses */
3012 if (netdev_uc_count(netdev) > rar_entries)
3015 if (!netdev_uc_empty(netdev) && rar_entries) {
3016 struct netdev_hw_addr *ha;
3017 /* return error if we do not support writing to RAR table */
3018 if (!hw->mac.ops.set_rar)
3021 netdev_for_each_uc_addr(ha, netdev) {
3024 hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3029 /* write the addresses in reverse order to avoid write combining */
3030 for (; rar_entries > 0 ; rar_entries--)
3031 hw->mac.ops.clear_rar(hw, rar_entries);
3037 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3038 * @netdev: network interface device structure
3040 * The set_rx_method entry point is called whenever the unicast/multicast
3041 * address list or the network interface flags are updated. This routine is
3042 * responsible for configuring the hardware for proper unicast, multicast and
3045 void ixgbe_set_rx_mode(struct net_device *netdev)
3047 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3048 struct ixgbe_hw *hw = &adapter->hw;
3049 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3052 /* Check for Promiscuous and All Multicast modes */
3054 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3056 /* clear the bits we are changing the status of */
3057 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3059 if (netdev->flags & IFF_PROMISC) {
3060 hw->addr_ctrl.user_set_promisc = true;
3061 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3062 vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3063 /* don't hardware filter vlans in promisc mode */
3064 ixgbe_vlan_filter_disable(adapter);
3066 if (netdev->flags & IFF_ALLMULTI) {
3067 fctrl |= IXGBE_FCTRL_MPE;
3068 vmolr |= IXGBE_VMOLR_MPE;
3071 * Write addresses to the MTA, if the attempt fails
3072 * then we should just turn on promiscous mode so
3073 * that we can at least receive multicast traffic
3075 hw->mac.ops.update_mc_addr_list(hw, netdev);
3076 vmolr |= IXGBE_VMOLR_ROMPE;
3078 ixgbe_vlan_filter_enable(adapter);
3079 hw->addr_ctrl.user_set_promisc = false;
3081 * Write addresses to available RAR registers, if there is not
3082 * sufficient space to store all the addresses then enable
3083 * unicast promiscous mode
3085 count = ixgbe_write_uc_addr_list(netdev);
3087 fctrl |= IXGBE_FCTRL_UPE;
3088 vmolr |= IXGBE_VMOLR_ROPE;
3092 if (adapter->num_vfs) {
3093 ixgbe_restore_vf_multicasts(adapter);
3094 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(adapter->num_vfs)) &
3095 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3097 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(adapter->num_vfs), vmolr);
3100 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3103 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3106 struct ixgbe_q_vector *q_vector;
3107 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3109 /* legacy and MSI only use one vector */
3110 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3113 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3114 struct napi_struct *napi;
3115 q_vector = adapter->q_vector[q_idx];
3116 napi = &q_vector->napi;
3117 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3118 if (!q_vector->rxr_count || !q_vector->txr_count) {
3119 if (q_vector->txr_count == 1)
3120 napi->poll = &ixgbe_clean_txonly;
3121 else if (q_vector->rxr_count == 1)
3122 napi->poll = &ixgbe_clean_rxonly;
3130 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3133 struct ixgbe_q_vector *q_vector;
3134 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3136 /* legacy and MSI only use one vector */
3137 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3140 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3141 q_vector = adapter->q_vector[q_idx];
3142 napi_disable(&q_vector->napi);
3146 #ifdef CONFIG_IXGBE_DCB
3148 * ixgbe_configure_dcb - Configure DCB hardware
3149 * @adapter: ixgbe adapter struct
3151 * This is called by the driver on open to configure the DCB hardware.
3152 * This is also called by the gennetlink interface when reconfiguring
3155 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3157 struct ixgbe_hw *hw = &adapter->hw;
3161 ixgbe_dcb_check_config(&adapter->dcb_cfg);
3162 ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_TX_CONFIG);
3163 ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_RX_CONFIG);
3165 /* reconfigure the hardware */
3166 ixgbe_dcb_hw_config(&adapter->hw, &adapter->dcb_cfg);
3168 for (i = 0; i < adapter->num_tx_queues; i++) {
3169 j = adapter->tx_ring[i]->reg_idx;
3170 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3171 /* PThresh workaround for Tx hang with DFP enabled. */
3173 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
3175 /* Enable VLAN tag insert/strip */
3176 ixgbe_vlan_filter_enable(adapter);
3178 hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
3182 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3184 struct net_device *netdev = adapter->netdev;
3185 struct ixgbe_hw *hw = &adapter->hw;
3188 ixgbe_set_rx_mode(netdev);
3190 ixgbe_restore_vlan(adapter);
3191 #ifdef CONFIG_IXGBE_DCB
3192 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3193 if (hw->mac.type == ixgbe_mac_82598EB)
3194 netif_set_gso_max_size(netdev, 32768);
3196 netif_set_gso_max_size(netdev, 65536);
3197 ixgbe_configure_dcb(adapter);
3199 netif_set_gso_max_size(netdev, 65536);
3202 netif_set_gso_max_size(netdev, 65536);
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 for (i = 0; i < adapter->num_tx_queues; i++) {
3417 j = adapter->tx_ring[i]->reg_idx;
3418 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3419 if (adapter->rx_itr_setting == 0) {
3420 /* cannot set wthresh when itr==0 */
3421 txdctl &= ~0x007F0000;
3423 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
3424 txdctl |= (8 << 16);
3426 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
3429 if (hw->mac.type == ixgbe_mac_82599EB) {
3430 /* DMATXCTL.EN must be set after all Tx queue config is done */
3431 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
3432 dmatxctl |= IXGBE_DMATXCTL_TE;
3433 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
3435 for (i = 0; i < adapter->num_tx_queues; i++) {
3436 j = adapter->tx_ring[i]->reg_idx;
3437 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3438 txdctl |= IXGBE_TXDCTL_ENABLE;
3439 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
3440 if (hw->mac.type == ixgbe_mac_82599EB) {
3442 /* poll for Tx Enable ready */
3445 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3446 } while (--wait_loop &&
3447 !(txdctl & IXGBE_TXDCTL_ENABLE));
3449 e_err(drv, "Could not enable Tx Queue %d\n", j);
3453 for (i = 0; i < num_rx_rings; i++) {
3454 j = adapter->rx_ring[i]->reg_idx;
3455 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3456 /* enable PTHRESH=32 descriptors (half the internal cache)
3457 * and HTHRESH=0 descriptors (to minimize latency on fetch),
3458 * this also removes a pesky rx_no_buffer_count increment */
3460 rxdctl |= IXGBE_RXDCTL_ENABLE;
3461 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), rxdctl);
3462 if (hw->mac.type == ixgbe_mac_82599EB)
3463 ixgbe_rx_desc_queue_enable(adapter, i);
3465 /* enable all receives */
3466 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3467 if (hw->mac.type == ixgbe_mac_82598EB)
3468 rxdctl |= (IXGBE_RXCTRL_DMBYPS | IXGBE_RXCTRL_RXEN);
3470 rxdctl |= IXGBE_RXCTRL_RXEN;
3471 hw->mac.ops.enable_rx_dma(hw, rxdctl);
3473 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3474 ixgbe_configure_msix(adapter);
3476 ixgbe_configure_msi_and_legacy(adapter);
3478 /* enable the optics */
3479 if (hw->phy.multispeed_fiber)
3480 hw->mac.ops.enable_tx_laser(hw);
3482 clear_bit(__IXGBE_DOWN, &adapter->state);
3483 ixgbe_napi_enable_all(adapter);
3485 /* clear any pending interrupts, may auto mask */
3486 IXGBE_READ_REG(hw, IXGBE_EICR);
3488 ixgbe_irq_enable(adapter);
3491 * If this adapter has a fan, check to see if we had a failure
3492 * before we enabled the interrupt.
3494 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3495 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
3496 if (esdp & IXGBE_ESDP_SDP1)
3497 e_crit(drv, "Fan has stopped, replace the adapter\n");
3501 * For hot-pluggable SFP+ devices, a new SFP+ module may have
3502 * arrived before interrupts were enabled but after probe. Such
3503 * devices wouldn't have their type identified yet. We need to
3504 * kick off the SFP+ module setup first, then try to bring up link.
3505 * If we're not hot-pluggable SFP+, we just need to configure link
3508 if (hw->phy.type == ixgbe_phy_unknown) {
3509 err = hw->phy.ops.identify(hw);
3510 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
3512 * Take the device down and schedule the sfp tasklet
3513 * which will unregister_netdev and log it.
3515 ixgbe_down(adapter);
3516 schedule_work(&adapter->sfp_config_module_task);
3521 if (ixgbe_is_sfp(hw)) {
3522 ixgbe_sfp_link_config(adapter);
3524 err = ixgbe_non_sfp_link_config(hw);
3526 e_err(probe, "link_config FAILED %d\n", err);
3529 for (i = 0; i < adapter->num_tx_queues; i++)
3530 set_bit(__IXGBE_FDIR_INIT_DONE,
3531 &(adapter->tx_ring[i]->reinit_state));
3533 /* enable transmits */
3534 netif_tx_start_all_queues(netdev);
3536 /* bring the link up in the watchdog, this could race with our first
3537 * link up interrupt but shouldn't be a problem */
3538 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3539 adapter->link_check_timeout = jiffies;
3540 mod_timer(&adapter->watchdog_timer, jiffies);
3542 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
3543 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
3544 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
3545 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
3550 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
3552 WARN_ON(in_interrupt());
3553 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
3555 ixgbe_down(adapter);
3557 * If SR-IOV enabled then wait a bit before bringing the adapter
3558 * back up to give the VFs time to respond to the reset. The
3559 * two second wait is based upon the watchdog timer cycle in
3562 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3565 clear_bit(__IXGBE_RESETTING, &adapter->state);
3568 int ixgbe_up(struct ixgbe_adapter *adapter)
3570 /* hardware has been reset, we need to reload some things */
3571 ixgbe_configure(adapter);
3573 return ixgbe_up_complete(adapter);
3576 void ixgbe_reset(struct ixgbe_adapter *adapter)
3578 struct ixgbe_hw *hw = &adapter->hw;
3581 err = hw->mac.ops.init_hw(hw);
3584 case IXGBE_ERR_SFP_NOT_PRESENT:
3586 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
3587 e_dev_err("master disable timed out\n");
3589 case IXGBE_ERR_EEPROM_VERSION:
3590 /* We are running on a pre-production device, log a warning */
3591 e_dev_warn("This device is a pre-production adapter/LOM. "
3592 "Please be aware there may be issuesassociated with "
3593 "your hardware. If you are experiencing problems "
3594 "please contact your Intel or hardware "
3595 "representative who provided you with this "
3599 e_dev_err("Hardware Error: %d\n", err);
3602 /* reprogram the RAR[0] in case user changed it. */
3603 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
3608 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
3609 * @adapter: board private structure
3610 * @rx_ring: ring to free buffers from
3612 static void ixgbe_clean_rx_ring(struct ixgbe_adapter *adapter,
3613 struct ixgbe_ring *rx_ring)
3615 struct pci_dev *pdev = adapter->pdev;
3619 /* Free all the Rx ring sk_buffs */
3621 for (i = 0; i < rx_ring->count; i++) {
3622 struct ixgbe_rx_buffer *rx_buffer_info;
3624 rx_buffer_info = &rx_ring->rx_buffer_info[i];
3625 if (rx_buffer_info->dma) {
3626 dma_unmap_single(&pdev->dev, rx_buffer_info->dma,
3627 rx_ring->rx_buf_len,
3629 rx_buffer_info->dma = 0;
3631 if (rx_buffer_info->skb) {
3632 struct sk_buff *skb = rx_buffer_info->skb;
3633 rx_buffer_info->skb = NULL;
3635 struct sk_buff *this = skb;
3636 if (IXGBE_RSC_CB(this)->delay_unmap) {
3637 dma_unmap_single(&pdev->dev,
3638 IXGBE_RSC_CB(this)->dma,
3639 rx_ring->rx_buf_len,
3641 IXGBE_RSC_CB(this)->dma = 0;
3642 IXGBE_RSC_CB(skb)->delay_unmap = false;
3645 dev_kfree_skb(this);
3648 if (!rx_buffer_info->page)
3650 if (rx_buffer_info->page_dma) {
3651 dma_unmap_page(&pdev->dev, rx_buffer_info->page_dma,
3652 PAGE_SIZE / 2, DMA_FROM_DEVICE);
3653 rx_buffer_info->page_dma = 0;
3655 put_page(rx_buffer_info->page);
3656 rx_buffer_info->page = NULL;
3657 rx_buffer_info->page_offset = 0;
3660 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
3661 memset(rx_ring->rx_buffer_info, 0, size);
3663 /* Zero out the descriptor ring */
3664 memset(rx_ring->desc, 0, rx_ring->size);
3666 rx_ring->next_to_clean = 0;
3667 rx_ring->next_to_use = 0;
3670 writel(0, adapter->hw.hw_addr + rx_ring->head);
3672 writel(0, adapter->hw.hw_addr + rx_ring->tail);
3676 * ixgbe_clean_tx_ring - Free Tx Buffers
3677 * @adapter: board private structure
3678 * @tx_ring: ring to be cleaned
3680 static void ixgbe_clean_tx_ring(struct ixgbe_adapter *adapter,
3681 struct ixgbe_ring *tx_ring)
3683 struct ixgbe_tx_buffer *tx_buffer_info;
3687 /* Free all the Tx ring sk_buffs */
3689 for (i = 0; i < tx_ring->count; i++) {
3690 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3691 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
3694 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
3695 memset(tx_ring->tx_buffer_info, 0, size);
3697 /* Zero out the descriptor ring */
3698 memset(tx_ring->desc, 0, tx_ring->size);
3700 tx_ring->next_to_use = 0;
3701 tx_ring->next_to_clean = 0;
3704 writel(0, adapter->hw.hw_addr + tx_ring->head);
3706 writel(0, adapter->hw.hw_addr + tx_ring->tail);
3710 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
3711 * @adapter: board private structure
3713 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
3717 for (i = 0; i < adapter->num_rx_queues; i++)
3718 ixgbe_clean_rx_ring(adapter, adapter->rx_ring[i]);
3722 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
3723 * @adapter: board private structure
3725 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
3729 for (i = 0; i < adapter->num_tx_queues; i++)
3730 ixgbe_clean_tx_ring(adapter, adapter->tx_ring[i]);
3733 void ixgbe_down(struct ixgbe_adapter *adapter)
3735 struct net_device *netdev = adapter->netdev;
3736 struct ixgbe_hw *hw = &adapter->hw;
3741 /* signal that we are down to the interrupt handler */
3742 set_bit(__IXGBE_DOWN, &adapter->state);
3744 /* disable receive for all VFs and wait one second */
3745 if (adapter->num_vfs) {
3746 /* ping all the active vfs to let them know we are going down */
3747 ixgbe_ping_all_vfs(adapter);
3749 /* Disable all VFTE/VFRE TX/RX */
3750 ixgbe_disable_tx_rx(adapter);
3752 /* Mark all the VFs as inactive */
3753 for (i = 0 ; i < adapter->num_vfs; i++)
3754 adapter->vfinfo[i].clear_to_send = 0;
3757 /* disable receives */
3758 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3759 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3761 IXGBE_WRITE_FLUSH(hw);
3764 netif_tx_stop_all_queues(netdev);
3766 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
3767 del_timer_sync(&adapter->sfp_timer);
3768 del_timer_sync(&adapter->watchdog_timer);
3769 cancel_work_sync(&adapter->watchdog_task);
3771 netif_carrier_off(netdev);
3772 netif_tx_disable(netdev);
3774 ixgbe_irq_disable(adapter);
3776 ixgbe_napi_disable_all(adapter);
3778 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3779 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3780 cancel_work_sync(&adapter->fdir_reinit_task);
3782 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
3783 cancel_work_sync(&adapter->check_overtemp_task);
3785 /* disable transmits in the hardware now that interrupts are off */
3786 for (i = 0; i < adapter->num_tx_queues; i++) {
3787 j = adapter->tx_ring[i]->reg_idx;
3788 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3789 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j),
3790 (txdctl & ~IXGBE_TXDCTL_ENABLE));
3792 /* Disable the Tx DMA engine on 82599 */
3793 if (hw->mac.type == ixgbe_mac_82599EB)
3794 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
3795 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
3796 ~IXGBE_DMATXCTL_TE));
3798 /* power down the optics */
3799 if (hw->phy.multispeed_fiber)
3800 hw->mac.ops.disable_tx_laser(hw);
3802 /* clear n-tuple filters that are cached */
3803 ethtool_ntuple_flush(netdev);
3805 if (!pci_channel_offline(adapter->pdev))
3806 ixgbe_reset(adapter);
3807 ixgbe_clean_all_tx_rings(adapter);
3808 ixgbe_clean_all_rx_rings(adapter);
3810 #ifdef CONFIG_IXGBE_DCA
3811 /* since we reset the hardware DCA settings were cleared */
3812 ixgbe_setup_dca(adapter);
3817 * ixgbe_poll - NAPI Rx polling callback
3818 * @napi: structure for representing this polling device
3819 * @budget: how many packets driver is allowed to clean
3821 * This function is used for legacy and MSI, NAPI mode
3823 static int ixgbe_poll(struct napi_struct *napi, int budget)
3825 struct ixgbe_q_vector *q_vector =
3826 container_of(napi, struct ixgbe_q_vector, napi);
3827 struct ixgbe_adapter *adapter = q_vector->adapter;
3828 int tx_clean_complete, work_done = 0;
3830 #ifdef CONFIG_IXGBE_DCA
3831 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
3832 ixgbe_update_tx_dca(adapter, adapter->tx_ring[0]);
3833 ixgbe_update_rx_dca(adapter, adapter->rx_ring[0]);
3837 tx_clean_complete = ixgbe_clean_tx_irq(q_vector, adapter->tx_ring[0]);
3838 ixgbe_clean_rx_irq(q_vector, adapter->rx_ring[0], &work_done, budget);
3840 if (!tx_clean_complete)
3843 /* If budget not fully consumed, exit the polling mode */
3844 if (work_done < budget) {
3845 napi_complete(napi);
3846 if (adapter->rx_itr_setting & 1)
3847 ixgbe_set_itr(adapter);
3848 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3849 ixgbe_irq_enable_queues(adapter, IXGBE_EIMS_RTX_QUEUE);
3855 * ixgbe_tx_timeout - Respond to a Tx Hang
3856 * @netdev: network interface device structure
3858 static void ixgbe_tx_timeout(struct net_device *netdev)
3860 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3862 /* Do the reset outside of interrupt context */
3863 schedule_work(&adapter->reset_task);
3866 static void ixgbe_reset_task(struct work_struct *work)
3868 struct ixgbe_adapter *adapter;
3869 adapter = container_of(work, struct ixgbe_adapter, reset_task);
3871 /* If we're already down or resetting, just bail */
3872 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
3873 test_bit(__IXGBE_RESETTING, &adapter->state))
3876 adapter->tx_timeout_count++;
3878 ixgbe_dump(adapter);
3879 netdev_err(adapter->netdev, "Reset adapter\n");
3880 ixgbe_reinit_locked(adapter);
3883 #ifdef CONFIG_IXGBE_DCB
3884 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
3887 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_DCB];
3889 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
3893 adapter->num_rx_queues = f->indices;
3894 adapter->num_tx_queues = f->indices;
3902 * ixgbe_set_rss_queues: Allocate queues for RSS
3903 * @adapter: board private structure to initialize
3905 * This is our "base" multiqueue mode. RSS (Receive Side Scaling) will try
3906 * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
3909 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
3912 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
3914 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3916 adapter->num_rx_queues = f->indices;
3917 adapter->num_tx_queues = f->indices;
3927 * ixgbe_set_fdir_queues: Allocate queues for Flow Director
3928 * @adapter: board private structure to initialize
3930 * Flow Director is an advanced Rx filter, attempting to get Rx flows back
3931 * to the original CPU that initiated the Tx session. This runs in addition
3932 * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
3933 * Rx load across CPUs using RSS.
3936 static bool inline ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
3939 struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
3941 f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
3944 /* Flow Director must have RSS enabled */
3945 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
3946 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3947 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)))) {
3948 adapter->num_tx_queues = f_fdir->indices;
3949 adapter->num_rx_queues = f_fdir->indices;
3952 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
3953 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
3960 * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
3961 * @adapter: board private structure to initialize
3963 * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
3964 * The ring feature mask is not used as a mask for FCoE, as it can take any 8
3965 * rx queues out of the max number of rx queues, instead, it is used as the
3966 * index of the first rx queue used by FCoE.
3969 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
3972 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
3974 f->indices = min((int)num_online_cpus(), f->indices);
3975 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
3976 adapter->num_rx_queues = 1;
3977 adapter->num_tx_queues = 1;
3978 #ifdef CONFIG_IXGBE_DCB
3979 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3980 e_info(probe, "FCoE enabled with DCB\n");
3981 ixgbe_set_dcb_queues(adapter);
3984 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3985 e_info(probe, "FCoE enabled with RSS\n");
3986 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
3987 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
3988 ixgbe_set_fdir_queues(adapter);
3990 ixgbe_set_rss_queues(adapter);
3992 /* adding FCoE rx rings to the end */
3993 f->mask = adapter->num_rx_queues;
3994 adapter->num_rx_queues += f->indices;
3995 adapter->num_tx_queues += f->indices;
4003 #endif /* IXGBE_FCOE */
4005 * ixgbe_set_sriov_queues: Allocate queues for IOV use
4006 * @adapter: board private structure to initialize
4008 * IOV doesn't actually use anything, so just NAK the
4009 * request for now and let the other queue routines
4010 * figure out what to do.
4012 static inline bool ixgbe_set_sriov_queues(struct ixgbe_adapter *adapter)
4018 * ixgbe_set_num_queues: Allocate queues for device, feature dependant
4019 * @adapter: board private structure to initialize
4021 * This is the top level queue allocation routine. The order here is very
4022 * important, starting with the "most" number of features turned on at once,
4023 * and ending with the smallest set of features. This way large combinations
4024 * can be allocated if they're turned on, and smaller combinations are the
4025 * fallthrough conditions.
4028 static void ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
4030 /* Start with base case */
4031 adapter->num_rx_queues = 1;
4032 adapter->num_tx_queues = 1;
4033 adapter->num_rx_pools = adapter->num_rx_queues;
4034 adapter->num_rx_queues_per_pool = 1;
4036 if (ixgbe_set_sriov_queues(adapter))
4040 if (ixgbe_set_fcoe_queues(adapter))
4043 #endif /* IXGBE_FCOE */
4044 #ifdef CONFIG_IXGBE_DCB
4045 if (ixgbe_set_dcb_queues(adapter))
4049 if (ixgbe_set_fdir_queues(adapter))
4052 if (ixgbe_set_rss_queues(adapter))
4055 /* fallback to base case */
4056 adapter->num_rx_queues = 1;
4057 adapter->num_tx_queues = 1;
4060 /* Notify the stack of the (possibly) reduced Tx Queue count. */
4061 netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
4064 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
4067 int err, vector_threshold;
4069 /* We'll want at least 3 (vector_threshold):
4072 * 3) Other (Link Status Change, etc.)
4073 * 4) TCP Timer (optional)
4075 vector_threshold = MIN_MSIX_COUNT;
4077 /* The more we get, the more we will assign to Tx/Rx Cleanup
4078 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
4079 * Right now, we simply care about how many we'll get; we'll
4080 * set them up later while requesting irq's.
4082 while (vectors >= vector_threshold) {
4083 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
4085 if (!err) /* Success in acquiring all requested vectors. */
4088 vectors = 0; /* Nasty failure, quit now */
4089 else /* err == number of vectors we should try again with */
4093 if (vectors < vector_threshold) {
4094 /* Can't allocate enough MSI-X interrupts? Oh well.
4095 * This just means we'll go with either a single MSI
4096 * vector or fall back to legacy interrupts.
4098 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4099 "Unable to allocate MSI-X interrupts\n");
4100 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4101 kfree(adapter->msix_entries);
4102 adapter->msix_entries = NULL;
4104 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
4106 * Adjust for only the vectors we'll use, which is minimum
4107 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
4108 * vectors we were allocated.
4110 adapter->num_msix_vectors = min(vectors,
4111 adapter->max_msix_q_vectors + NON_Q_VECTORS);
4116 * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
4117 * @adapter: board private structure to initialize
4119 * Cache the descriptor ring offsets for RSS to the assigned rings.
4122 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
4127 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4128 for (i = 0; i < adapter->num_rx_queues; i++)
4129 adapter->rx_ring[i]->reg_idx = i;
4130 for (i = 0; i < adapter->num_tx_queues; i++)
4131 adapter->tx_ring[i]->reg_idx = i;
4140 #ifdef CONFIG_IXGBE_DCB
4142 * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
4143 * @adapter: board private structure to initialize
4145 * Cache the descriptor ring offsets for DCB to the assigned rings.
4148 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
4152 int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
4154 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4155 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
4156 /* the number of queues is assumed to be symmetric */
4157 for (i = 0; i < dcb_i; i++) {
4158 adapter->rx_ring[i]->reg_idx = i << 3;
4159 adapter->tx_ring[i]->reg_idx = i << 2;
4162 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
4165 * Tx TC0 starts at: descriptor queue 0
4166 * Tx TC1 starts at: descriptor queue 32
4167 * Tx TC2 starts at: descriptor queue 64
4168 * Tx TC3 starts at: descriptor queue 80
4169 * Tx TC4 starts at: descriptor queue 96
4170 * Tx TC5 starts at: descriptor queue 104
4171 * Tx TC6 starts at: descriptor queue 112
4172 * Tx TC7 starts at: descriptor queue 120
4174 * Rx TC0-TC7 are offset by 16 queues each
4176 for (i = 0; i < 3; i++) {
4177 adapter->tx_ring[i]->reg_idx = i << 5;
4178 adapter->rx_ring[i]->reg_idx = i << 4;
4180 for ( ; i < 5; i++) {
4181 adapter->tx_ring[i]->reg_idx =
4183 adapter->rx_ring[i]->reg_idx = i << 4;
4185 for ( ; i < dcb_i; i++) {
4186 adapter->tx_ring[i]->reg_idx =
4188 adapter->rx_ring[i]->reg_idx = i << 4;
4192 } else if (dcb_i == 4) {
4194 * Tx TC0 starts at: descriptor queue 0
4195 * Tx TC1 starts at: descriptor queue 64
4196 * Tx TC2 starts at: descriptor queue 96
4197 * Tx TC3 starts at: descriptor queue 112
4199 * Rx TC0-TC3 are offset by 32 queues each
4201 adapter->tx_ring[0]->reg_idx = 0;
4202 adapter->tx_ring[1]->reg_idx = 64;
4203 adapter->tx_ring[2]->reg_idx = 96;
4204 adapter->tx_ring[3]->reg_idx = 112;
4205 for (i = 0 ; i < dcb_i; i++)
4206 adapter->rx_ring[i]->reg_idx = i << 5;
4224 * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
4225 * @adapter: board private structure to initialize
4227 * Cache the descriptor ring offsets for Flow Director to the assigned rings.
4230 static bool inline ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
4235 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
4236 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4237 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))) {
4238 for (i = 0; i < adapter->num_rx_queues; i++)
4239 adapter->rx_ring[i]->reg_idx = i;
4240 for (i = 0; i < adapter->num_tx_queues; i++)
4241 adapter->tx_ring[i]->reg_idx = i;
4250 * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
4251 * @adapter: board private structure to initialize
4253 * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
4256 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
4258 int i, fcoe_rx_i = 0, fcoe_tx_i = 0;
4260 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4262 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
4263 #ifdef CONFIG_IXGBE_DCB
4264 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4265 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
4267 ixgbe_cache_ring_dcb(adapter);
4268 /* find out queues in TC for FCoE */
4269 fcoe_rx_i = adapter->rx_ring[fcoe->tc]->reg_idx + 1;
4270 fcoe_tx_i = adapter->tx_ring[fcoe->tc]->reg_idx + 1;
4272 * In 82599, the number of Tx queues for each traffic
4273 * class for both 8-TC and 4-TC modes are:
4274 * TCs : TC0 TC1 TC2 TC3 TC4 TC5 TC6 TC7
4275 * 8 TCs: 32 32 16 16 8 8 8 8
4276 * 4 TCs: 64 64 32 32
4277 * We have max 8 queues for FCoE, where 8 the is
4278 * FCoE redirection table size. If TC for FCoE is
4279 * less than or equal to TC3, we have enough queues
4280 * to add max of 8 queues for FCoE, so we start FCoE
4281 * tx descriptor from the next one, i.e., reg_idx + 1.
4282 * If TC for FCoE is above TC3, implying 8 TC mode,
4283 * and we need 8 for FCoE, we have to take all queues
4284 * in that traffic class for FCoE.
4286 if ((f->indices == IXGBE_FCRETA_SIZE) && (fcoe->tc > 3))
4289 #endif /* CONFIG_IXGBE_DCB */
4290 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4291 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4292 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
4293 ixgbe_cache_ring_fdir(adapter);
4295 ixgbe_cache_ring_rss(adapter);
4297 fcoe_rx_i = f->mask;
4298 fcoe_tx_i = f->mask;
4300 for (i = 0; i < f->indices; i++, fcoe_rx_i++, fcoe_tx_i++) {
4301 adapter->rx_ring[f->mask + i]->reg_idx = fcoe_rx_i;
4302 adapter->tx_ring[f->mask + i]->reg_idx = fcoe_tx_i;
4309 #endif /* IXGBE_FCOE */
4311 * ixgbe_cache_ring_sriov - Descriptor ring to register mapping for sriov
4312 * @adapter: board private structure to initialize
4314 * SR-IOV doesn't use any descriptor rings but changes the default if
4315 * no other mapping is used.
4318 static inline bool ixgbe_cache_ring_sriov(struct ixgbe_adapter *adapter)
4320 adapter->rx_ring[0]->reg_idx = adapter->num_vfs * 2;
4321 adapter->tx_ring[0]->reg_idx = adapter->num_vfs * 2;
4322 if (adapter->num_vfs)
4329 * ixgbe_cache_ring_register - Descriptor ring to register mapping
4330 * @adapter: board private structure to initialize
4332 * Once we know the feature-set enabled for the device, we'll cache
4333 * the register offset the descriptor ring is assigned to.
4335 * Note, the order the various feature calls is important. It must start with
4336 * the "most" features enabled at the same time, then trickle down to the
4337 * least amount of features turned on at once.
4339 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
4341 /* start with default case */
4342 adapter->rx_ring[0]->reg_idx = 0;
4343 adapter->tx_ring[0]->reg_idx = 0;
4345 if (ixgbe_cache_ring_sriov(adapter))
4349 if (ixgbe_cache_ring_fcoe(adapter))
4352 #endif /* IXGBE_FCOE */
4353 #ifdef CONFIG_IXGBE_DCB
4354 if (ixgbe_cache_ring_dcb(adapter))
4358 if (ixgbe_cache_ring_fdir(adapter))
4361 if (ixgbe_cache_ring_rss(adapter))
4366 * ixgbe_alloc_queues - Allocate memory for all rings
4367 * @adapter: board private structure to initialize
4369 * We allocate one ring per queue at run-time since we don't know the
4370 * number of queues at compile-time. The polling_netdev array is
4371 * intended for Multiqueue, but should work fine with a single queue.
4373 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
4376 int orig_node = adapter->node;
4378 for (i = 0; i < adapter->num_tx_queues; i++) {
4379 struct ixgbe_ring *ring = adapter->tx_ring[i];
4380 if (orig_node == -1) {
4381 int cur_node = next_online_node(adapter->node);
4382 if (cur_node == MAX_NUMNODES)
4383 cur_node = first_online_node;
4384 adapter->node = cur_node;
4386 ring = kzalloc_node(sizeof(struct ixgbe_ring), GFP_KERNEL,
4389 ring = kzalloc(sizeof(struct ixgbe_ring), GFP_KERNEL);
4391 goto err_tx_ring_allocation;
4392 ring->count = adapter->tx_ring_count;
4393 ring->queue_index = i;
4394 ring->numa_node = adapter->node;
4396 adapter->tx_ring[i] = ring;
4399 /* Restore the adapter's original node */
4400 adapter->node = orig_node;
4402 for (i = 0; i < adapter->num_rx_queues; i++) {
4403 struct ixgbe_ring *ring = adapter->rx_ring[i];
4404 if (orig_node == -1) {
4405 int cur_node = next_online_node(adapter->node);
4406 if (cur_node == MAX_NUMNODES)
4407 cur_node = first_online_node;
4408 adapter->node = cur_node;
4410 ring = kzalloc_node(sizeof(struct ixgbe_ring), GFP_KERNEL,
4413 ring = kzalloc(sizeof(struct ixgbe_ring), GFP_KERNEL);
4415 goto err_rx_ring_allocation;
4416 ring->count = adapter->rx_ring_count;
4417 ring->queue_index = i;
4418 ring->numa_node = adapter->node;
4420 adapter->rx_ring[i] = ring;
4423 /* Restore the adapter's original node */
4424 adapter->node = orig_node;
4426 ixgbe_cache_ring_register(adapter);
4430 err_rx_ring_allocation:
4431 for (i = 0; i < adapter->num_tx_queues; i++)
4432 kfree(adapter->tx_ring[i]);
4433 err_tx_ring_allocation:
4438 * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
4439 * @adapter: board private structure to initialize
4441 * Attempt to configure the interrupts using the best available
4442 * capabilities of the hardware and the kernel.
4444 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
4446 struct ixgbe_hw *hw = &adapter->hw;
4448 int vector, v_budget;
4451 * It's easy to be greedy for MSI-X vectors, but it really
4452 * doesn't do us much good if we have a lot more vectors
4453 * than CPU's. So let's be conservative and only ask for
4454 * (roughly) the same number of vectors as there are CPU's.
4456 v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
4457 (int)num_online_cpus()) + NON_Q_VECTORS;
4460 * At the same time, hardware can only support a maximum of
4461 * hw.mac->max_msix_vectors vectors. With features
4462 * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
4463 * descriptor queues supported by our device. Thus, we cap it off in
4464 * those rare cases where the cpu count also exceeds our vector limit.
4466 v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
4468 /* A failure in MSI-X entry allocation isn't fatal, but it does
4469 * mean we disable MSI-X capabilities of the adapter. */
4470 adapter->msix_entries = kcalloc(v_budget,
4471 sizeof(struct msix_entry), GFP_KERNEL);
4472 if (adapter->msix_entries) {
4473 for (vector = 0; vector < v_budget; vector++)
4474 adapter->msix_entries[vector].entry = vector;
4476 ixgbe_acquire_msix_vectors(adapter, v_budget);
4478 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4482 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
4483 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
4484 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4485 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4486 adapter->atr_sample_rate = 0;
4487 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4488 ixgbe_disable_sriov(adapter);
4490 ixgbe_set_num_queues(adapter);
4492 err = pci_enable_msi(adapter->pdev);
4494 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
4496 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4497 "Unable to allocate MSI interrupt, "
4498 "falling back to legacy. Error: %d\n", err);
4508 * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
4509 * @adapter: board private structure to initialize
4511 * We allocate one q_vector per queue interrupt. If allocation fails we
4514 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
4516 int q_idx, num_q_vectors;
4517 struct ixgbe_q_vector *q_vector;
4519 int (*poll)(struct napi_struct *, int);
4521 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4522 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4523 napi_vectors = adapter->num_rx_queues;
4524 poll = &ixgbe_clean_rxtx_many;
4531 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4532 q_vector = kzalloc_node(sizeof(struct ixgbe_q_vector),
4533 GFP_KERNEL, adapter->node);
4535 q_vector = kzalloc(sizeof(struct ixgbe_q_vector),
4539 q_vector->adapter = adapter;
4540 if (q_vector->txr_count && !q_vector->rxr_count)
4541 q_vector->eitr = adapter->tx_eitr_param;
4543 q_vector->eitr = adapter->rx_eitr_param;
4544 q_vector->v_idx = q_idx;
4545 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
4546 adapter->q_vector[q_idx] = q_vector;
4554 q_vector = adapter->q_vector[q_idx];
4555 netif_napi_del(&q_vector->napi);
4557 adapter->q_vector[q_idx] = NULL;
4563 * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
4564 * @adapter: board private structure to initialize
4566 * This function frees the memory allocated to the q_vectors. In addition if
4567 * NAPI is enabled it will delete any references to the NAPI struct prior
4568 * to freeing the q_vector.
4570 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
4572 int q_idx, num_q_vectors;
4574 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4575 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4579 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4580 struct ixgbe_q_vector *q_vector = adapter->q_vector[q_idx];
4581 adapter->q_vector[q_idx] = NULL;
4582 netif_napi_del(&q_vector->napi);
4587 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
4589 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4590 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4591 pci_disable_msix(adapter->pdev);
4592 kfree(adapter->msix_entries);
4593 adapter->msix_entries = NULL;
4594 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
4595 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
4596 pci_disable_msi(adapter->pdev);
4601 * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
4602 * @adapter: board private structure to initialize
4604 * We determine which interrupt scheme to use based on...
4605 * - Kernel support (MSI, MSI-X)
4606 * - which can be user-defined (via MODULE_PARAM)
4607 * - Hardware queue count (num_*_queues)
4608 * - defined by miscellaneous hardware support/features (RSS, etc.)
4610 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
4614 /* Number of supported queues */
4615 ixgbe_set_num_queues(adapter);
4617 err = ixgbe_set_interrupt_capability(adapter);
4619 e_dev_err("Unable to setup interrupt capabilities\n");
4620 goto err_set_interrupt;
4623 err = ixgbe_alloc_q_vectors(adapter);
4625 e_dev_err("Unable to allocate memory for queue vectors\n");
4626 goto err_alloc_q_vectors;
4629 err = ixgbe_alloc_queues(adapter);
4631 e_dev_err("Unable to allocate memory for queues\n");
4632 goto err_alloc_queues;
4635 e_dev_info("Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
4636 (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
4637 adapter->num_rx_queues, adapter->num_tx_queues);
4639 set_bit(__IXGBE_DOWN, &adapter->state);
4644 ixgbe_free_q_vectors(adapter);
4645 err_alloc_q_vectors:
4646 ixgbe_reset_interrupt_capability(adapter);
4652 * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
4653 * @adapter: board private structure to clear interrupt scheme on
4655 * We go through and clear interrupt specific resources and reset the structure
4656 * to pre-load conditions
4658 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
4662 for (i = 0; i < adapter->num_tx_queues; i++) {
4663 kfree(adapter->tx_ring[i]);
4664 adapter->tx_ring[i] = NULL;
4666 for (i = 0; i < adapter->num_rx_queues; i++) {
4667 kfree(adapter->rx_ring[i]);
4668 adapter->rx_ring[i] = NULL;
4671 ixgbe_free_q_vectors(adapter);
4672 ixgbe_reset_interrupt_capability(adapter);
4676 * ixgbe_sfp_timer - worker thread to find a missing module
4677 * @data: pointer to our adapter struct
4679 static void ixgbe_sfp_timer(unsigned long data)
4681 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
4684 * Do the sfp_timer outside of interrupt context due to the
4685 * delays that sfp+ detection requires
4687 schedule_work(&adapter->sfp_task);
4691 * ixgbe_sfp_task - worker thread to find a missing module
4692 * @work: pointer to work_struct containing our data
4694 static void ixgbe_sfp_task(struct work_struct *work)
4696 struct ixgbe_adapter *adapter = container_of(work,
4697 struct ixgbe_adapter,
4699 struct ixgbe_hw *hw = &adapter->hw;
4701 if ((hw->phy.type == ixgbe_phy_nl) &&
4702 (hw->phy.sfp_type == ixgbe_sfp_type_not_present)) {
4703 s32 ret = hw->phy.ops.identify_sfp(hw);
4704 if (ret == IXGBE_ERR_SFP_NOT_PRESENT)
4706 ret = hw->phy.ops.reset(hw);
4707 if (ret == IXGBE_ERR_SFP_NOT_SUPPORTED) {
4708 e_dev_err("failed to initialize because an unsupported "
4709 "SFP+ module type was detected.\n");
4710 e_dev_err("Reload the driver after installing a "
4711 "supported module.\n");
4712 unregister_netdev(adapter->netdev);
4714 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
4716 /* don't need this routine any more */
4717 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
4721 if (test_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state))
4722 mod_timer(&adapter->sfp_timer,
4723 round_jiffies(jiffies + (2 * HZ)));
4727 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4728 * @adapter: board private structure to initialize
4730 * ixgbe_sw_init initializes the Adapter private data structure.
4731 * Fields are initialized based on PCI device information and
4732 * OS network device settings (MTU size).
4734 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
4736 struct ixgbe_hw *hw = &adapter->hw;
4737 struct pci_dev *pdev = adapter->pdev;
4738 struct net_device *dev = adapter->netdev;
4740 #ifdef CONFIG_IXGBE_DCB
4742 struct tc_configuration *tc;
4745 /* PCI config space info */
4747 hw->vendor_id = pdev->vendor;
4748 hw->device_id = pdev->device;
4749 hw->revision_id = pdev->revision;
4750 hw->subsystem_vendor_id = pdev->subsystem_vendor;
4751 hw->subsystem_device_id = pdev->subsystem_device;
4753 /* Set capability flags */
4754 rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
4755 adapter->ring_feature[RING_F_RSS].indices = rss;
4756 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
4757 adapter->ring_feature[RING_F_DCB].indices = IXGBE_MAX_DCB_INDICES;
4758 if (hw->mac.type == ixgbe_mac_82598EB) {
4759 if (hw->device_id == IXGBE_DEV_ID_82598AT)
4760 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4761 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
4762 } else if (hw->mac.type == ixgbe_mac_82599EB) {
4763 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
4764 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4765 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4766 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
4767 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4768 if (dev->features & NETIF_F_NTUPLE) {
4769 /* Flow Director perfect filter enabled */
4770 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4771 adapter->atr_sample_rate = 0;
4772 spin_lock_init(&adapter->fdir_perfect_lock);
4774 /* Flow Director hash filters enabled */
4775 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
4776 adapter->atr_sample_rate = 20;
4778 adapter->ring_feature[RING_F_FDIR].indices =
4779 IXGBE_MAX_FDIR_INDICES;
4780 adapter->fdir_pballoc = 0;
4782 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4783 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4784 adapter->ring_feature[RING_F_FCOE].indices = 0;
4785 #ifdef CONFIG_IXGBE_DCB
4786 /* Default traffic class to use for FCoE */
4787 adapter->fcoe.tc = IXGBE_FCOE_DEFTC;
4789 #endif /* IXGBE_FCOE */
4792 #ifdef CONFIG_IXGBE_DCB
4793 /* Configure DCB traffic classes */
4794 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
4795 tc = &adapter->dcb_cfg.tc_config[j];
4796 tc->path[DCB_TX_CONFIG].bwg_id = 0;
4797 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
4798 tc->path[DCB_RX_CONFIG].bwg_id = 0;
4799 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
4800 tc->dcb_pfc = pfc_disabled;
4802 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
4803 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
4804 adapter->dcb_cfg.rx_pba_cfg = pba_equal;
4805 adapter->dcb_cfg.pfc_mode_enable = false;
4806 adapter->dcb_cfg.round_robin_enable = false;
4807 adapter->dcb_set_bitmap = 0x00;
4808 ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
4809 adapter->ring_feature[RING_F_DCB].indices);
4813 /* default flow control settings */
4814 hw->fc.requested_mode = ixgbe_fc_full;
4815 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
4817 adapter->last_lfc_mode = hw->fc.current_mode;
4819 hw->fc.high_water = IXGBE_DEFAULT_FCRTH;
4820 hw->fc.low_water = IXGBE_DEFAULT_FCRTL;
4821 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
4822 hw->fc.send_xon = true;
4823 hw->fc.disable_fc_autoneg = false;
4825 /* enable itr by default in dynamic mode */
4826 adapter->rx_itr_setting = 1;
4827 adapter->rx_eitr_param = 20000;
4828 adapter->tx_itr_setting = 1;
4829 adapter->tx_eitr_param = 10000;
4831 /* set defaults for eitr in MegaBytes */
4832 adapter->eitr_low = 10;
4833 adapter->eitr_high = 20;
4835 /* set default ring sizes */
4836 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
4837 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
4839 /* initialize eeprom parameters */
4840 if (ixgbe_init_eeprom_params_generic(hw)) {
4841 e_dev_err("EEPROM initialization failed\n");
4845 /* enable rx csum by default */
4846 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
4848 /* get assigned NUMA node */
4849 adapter->node = dev_to_node(&pdev->dev);
4851 set_bit(__IXGBE_DOWN, &adapter->state);
4857 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
4858 * @adapter: board private structure
4859 * @tx_ring: tx descriptor ring (for a specific queue) to setup
4861 * Return 0 on success, negative on failure
4863 int ixgbe_setup_tx_resources(struct ixgbe_adapter *adapter,
4864 struct ixgbe_ring *tx_ring)
4866 struct pci_dev *pdev = adapter->pdev;
4869 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4870 tx_ring->tx_buffer_info = vmalloc_node(size, tx_ring->numa_node);
4871 if (!tx_ring->tx_buffer_info)
4872 tx_ring->tx_buffer_info = vmalloc(size);
4873 if (!tx_ring->tx_buffer_info)
4875 memset(tx_ring->tx_buffer_info, 0, size);
4877 /* round up to nearest 4K */
4878 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
4879 tx_ring->size = ALIGN(tx_ring->size, 4096);
4881 tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
4882 &tx_ring->dma, GFP_KERNEL);
4886 tx_ring->next_to_use = 0;
4887 tx_ring->next_to_clean = 0;
4888 tx_ring->work_limit = tx_ring->count;
4892 vfree(tx_ring->tx_buffer_info);
4893 tx_ring->tx_buffer_info = NULL;
4894 e_err(probe, "Unable to allocate memory for the Tx descriptor ring\n");
4899 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
4900 * @adapter: board private structure
4902 * If this function returns with an error, then it's possible one or
4903 * more of the rings is populated (while the rest are not). It is the
4904 * callers duty to clean those orphaned rings.
4906 * Return 0 on success, negative on failure
4908 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
4912 for (i = 0; i < adapter->num_tx_queues; i++) {
4913 err = ixgbe_setup_tx_resources(adapter, adapter->tx_ring[i]);
4916 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
4924 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
4925 * @adapter: board private structure
4926 * @rx_ring: rx descriptor ring (for a specific queue) to setup
4928 * Returns 0 on success, negative on failure
4930 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
4931 struct ixgbe_ring *rx_ring)
4933 struct pci_dev *pdev = adapter->pdev;
4936 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4937 rx_ring->rx_buffer_info = vmalloc_node(size, adapter->node);
4938 if (!rx_ring->rx_buffer_info)
4939 rx_ring->rx_buffer_info = vmalloc(size);
4940 if (!rx_ring->rx_buffer_info) {
4941 e_err(probe, "vmalloc allocation failed for the Rx "
4942 "descriptor ring\n");
4945 memset(rx_ring->rx_buffer_info, 0, size);
4947 /* Round up to nearest 4K */
4948 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
4949 rx_ring->size = ALIGN(rx_ring->size, 4096);
4951 rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
4952 &rx_ring->dma, GFP_KERNEL);
4954 if (!rx_ring->desc) {
4955 e_err(probe, "Memory allocation failed for the Rx "
4956 "descriptor ring\n");
4957 vfree(rx_ring->rx_buffer_info);
4961 rx_ring->next_to_clean = 0;
4962 rx_ring->next_to_use = 0;
4971 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
4972 * @adapter: board private structure
4974 * If this function returns with an error, then it's possible one or
4975 * more of the rings is populated (while the rest are not). It is the
4976 * callers duty to clean those orphaned rings.
4978 * Return 0 on success, negative on failure
4981 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
4985 for (i = 0; i < adapter->num_rx_queues; i++) {
4986 err = ixgbe_setup_rx_resources(adapter, adapter->rx_ring[i]);
4989 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
4997 * ixgbe_free_tx_resources - Free Tx Resources per Queue
4998 * @adapter: board private structure
4999 * @tx_ring: Tx descriptor ring for a specific queue
5001 * Free all transmit software resources
5003 void ixgbe_free_tx_resources(struct ixgbe_adapter *adapter,
5004 struct ixgbe_ring *tx_ring)
5006 struct pci_dev *pdev = adapter->pdev;
5008 ixgbe_clean_tx_ring(adapter, tx_ring);
5010 vfree(tx_ring->tx_buffer_info);
5011 tx_ring->tx_buffer_info = NULL;
5013 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
5016 tx_ring->desc = NULL;
5020 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5021 * @adapter: board private structure
5023 * Free all transmit software resources
5025 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
5029 for (i = 0; i < adapter->num_tx_queues; i++)
5030 if (adapter->tx_ring[i]->desc)
5031 ixgbe_free_tx_resources(adapter, adapter->tx_ring[i]);
5035 * ixgbe_free_rx_resources - Free Rx Resources
5036 * @adapter: board private structure
5037 * @rx_ring: ring to clean the resources from
5039 * Free all receive software resources
5041 void ixgbe_free_rx_resources(struct ixgbe_adapter *adapter,
5042 struct ixgbe_ring *rx_ring)
5044 struct pci_dev *pdev = adapter->pdev;
5046 ixgbe_clean_rx_ring(adapter, rx_ring);
5048 vfree(rx_ring->rx_buffer_info);
5049 rx_ring->rx_buffer_info = NULL;
5051 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
5054 rx_ring->desc = NULL;
5058 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5059 * @adapter: board private structure
5061 * Free all receive software resources
5063 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5067 for (i = 0; i < adapter->num_rx_queues; i++)
5068 if (adapter->rx_ring[i]->desc)
5069 ixgbe_free_rx_resources(adapter, adapter->rx_ring[i]);
5073 * ixgbe_change_mtu - Change the Maximum Transfer Unit
5074 * @netdev: network interface device structure
5075 * @new_mtu: new value for maximum frame size
5077 * Returns 0 on success, negative on failure
5079 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5081 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5082 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5084 /* MTU < 68 is an error and causes problems on some kernels */
5085 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5088 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5089 /* must set new MTU before calling down or up */
5090 netdev->mtu = new_mtu;
5092 if (netif_running(netdev))
5093 ixgbe_reinit_locked(adapter);
5099 * ixgbe_open - Called when a network interface is made active
5100 * @netdev: network interface device structure
5102 * Returns 0 on success, negative value on failure
5104 * The open entry point is called when a network interface is made
5105 * active by the system (IFF_UP). At this point all resources needed
5106 * for transmit and receive operations are allocated, the interrupt
5107 * handler is registered with the OS, the watchdog timer is started,
5108 * and the stack is notified that the interface is ready.
5110 static int ixgbe_open(struct net_device *netdev)
5112 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5115 /* disallow open during test */
5116 if (test_bit(__IXGBE_TESTING, &adapter->state))
5119 netif_carrier_off(netdev);
5121 /* allocate transmit descriptors */
5122 err = ixgbe_setup_all_tx_resources(adapter);
5126 /* allocate receive descriptors */
5127 err = ixgbe_setup_all_rx_resources(adapter);
5131 ixgbe_configure(adapter);
5133 err = ixgbe_request_irq(adapter);
5137 err = ixgbe_up_complete(adapter);
5141 netif_tx_start_all_queues(netdev);
5146 ixgbe_release_hw_control(adapter);
5147 ixgbe_free_irq(adapter);
5150 ixgbe_free_all_rx_resources(adapter);
5152 ixgbe_free_all_tx_resources(adapter);
5153 ixgbe_reset(adapter);
5159 * ixgbe_close - Disables a network interface
5160 * @netdev: network interface device structure
5162 * Returns 0, this is not allowed to fail
5164 * The close entry point is called when an interface is de-activated
5165 * by the OS. The hardware is still under the drivers control, but
5166 * needs to be disabled. A global MAC reset is issued to stop the
5167 * hardware, and all transmit and receive resources are freed.
5169 static int ixgbe_close(struct net_device *netdev)
5171 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5173 ixgbe_down(adapter);
5174 ixgbe_free_irq(adapter);
5176 ixgbe_free_all_tx_resources(adapter);
5177 ixgbe_free_all_rx_resources(adapter);
5179 ixgbe_release_hw_control(adapter);
5185 static int ixgbe_resume(struct pci_dev *pdev)
5187 struct net_device *netdev = pci_get_drvdata(pdev);
5188 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5191 pci_set_power_state(pdev, PCI_D0);
5192 pci_restore_state(pdev);
5194 * pci_restore_state clears dev->state_saved so call
5195 * pci_save_state to restore it.
5197 pci_save_state(pdev);
5199 err = pci_enable_device_mem(pdev);
5201 e_dev_err("Cannot enable PCI device from suspend\n");
5204 pci_set_master(pdev);
5206 pci_wake_from_d3(pdev, false);
5208 err = ixgbe_init_interrupt_scheme(adapter);
5210 e_dev_err("Cannot initialize interrupts for device\n");
5214 ixgbe_reset(adapter);
5216 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5218 if (netif_running(netdev)) {
5219 err = ixgbe_open(adapter->netdev);
5224 netif_device_attach(netdev);
5228 #endif /* CONFIG_PM */
5230 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5232 struct net_device *netdev = pci_get_drvdata(pdev);
5233 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5234 struct ixgbe_hw *hw = &adapter->hw;
5236 u32 wufc = adapter->wol;
5241 netif_device_detach(netdev);
5243 if (netif_running(netdev)) {
5244 ixgbe_down(adapter);
5245 ixgbe_free_irq(adapter);
5246 ixgbe_free_all_tx_resources(adapter);
5247 ixgbe_free_all_rx_resources(adapter);
5251 retval = pci_save_state(pdev);
5257 ixgbe_set_rx_mode(netdev);
5259 /* turn on all-multi mode if wake on multicast is enabled */
5260 if (wufc & IXGBE_WUFC_MC) {
5261 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5262 fctrl |= IXGBE_FCTRL_MPE;
5263 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5266 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5267 ctrl |= IXGBE_CTRL_GIO_DIS;
5268 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5270 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5272 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5273 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5276 if (wufc && hw->mac.type == ixgbe_mac_82599EB)
5277 pci_wake_from_d3(pdev, true);
5279 pci_wake_from_d3(pdev, false);
5281 *enable_wake = !!wufc;
5283 ixgbe_clear_interrupt_scheme(adapter);
5285 ixgbe_release_hw_control(adapter);
5287 pci_disable_device(pdev);
5293 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5298 retval = __ixgbe_shutdown(pdev, &wake);
5303 pci_prepare_to_sleep(pdev);
5305 pci_wake_from_d3(pdev, false);
5306 pci_set_power_state(pdev, PCI_D3hot);
5311 #endif /* CONFIG_PM */
5313 static void ixgbe_shutdown(struct pci_dev *pdev)
5317 __ixgbe_shutdown(pdev, &wake);
5319 if (system_state == SYSTEM_POWER_OFF) {
5320 pci_wake_from_d3(pdev, wake);
5321 pci_set_power_state(pdev, PCI_D3hot);
5326 * ixgbe_update_stats - Update the board statistics counters.
5327 * @adapter: board private structure
5329 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5331 struct net_device *netdev = adapter->netdev;
5332 struct ixgbe_hw *hw = &adapter->hw;
5334 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5335 u64 non_eop_descs = 0, restart_queue = 0;
5337 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5338 test_bit(__IXGBE_RESETTING, &adapter->state))
5341 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5344 for (i = 0; i < 16; i++)
5345 adapter->hw_rx_no_dma_resources +=
5346 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5347 for (i = 0; i < adapter->num_rx_queues; i++) {
5348 rsc_count += adapter->rx_ring[i]->rsc_count;
5349 rsc_flush += adapter->rx_ring[i]->rsc_flush;
5351 adapter->rsc_total_count = rsc_count;
5352 adapter->rsc_total_flush = rsc_flush;
5355 /* gather some stats to the adapter struct that are per queue */
5356 for (i = 0; i < adapter->num_tx_queues; i++)
5357 restart_queue += adapter->tx_ring[i]->restart_queue;
5358 adapter->restart_queue = restart_queue;
5360 for (i = 0; i < adapter->num_rx_queues; i++)
5361 non_eop_descs += adapter->rx_ring[i]->non_eop_descs;
5362 adapter->non_eop_descs = non_eop_descs;
5364 adapter->stats.crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5365 for (i = 0; i < 8; i++) {
5366 /* for packet buffers not used, the register should read 0 */
5367 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5369 adapter->stats.mpc[i] += mpc;
5370 total_mpc += adapter->stats.mpc[i];
5371 if (hw->mac.type == ixgbe_mac_82598EB)
5372 adapter->stats.rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5373 adapter->stats.qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5374 adapter->stats.qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5375 adapter->stats.qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5376 adapter->stats.qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5377 if (hw->mac.type == ixgbe_mac_82599EB) {
5378 adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
5379 IXGBE_PXONRXCNT(i));
5380 adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
5381 IXGBE_PXOFFRXCNT(i));
5382 adapter->stats.qprdc[i] += IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5384 adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
5386 adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
5389 adapter->stats.pxontxc[i] += IXGBE_READ_REG(hw,
5391 adapter->stats.pxofftxc[i] += IXGBE_READ_REG(hw,
5394 adapter->stats.gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5395 /* work around hardware counting issue */
5396 adapter->stats.gprc -= missed_rx;
5398 /* 82598 hardware only has a 32 bit counter in the high register */
5399 if (hw->mac.type == ixgbe_mac_82599EB) {
5401 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5402 tmp = IXGBE_READ_REG(hw, IXGBE_GORCH) & 0xF; /* 4 high bits of GORC */
5403 adapter->stats.gorc += (tmp << 32);
5404 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5405 tmp = IXGBE_READ_REG(hw, IXGBE_GOTCH) & 0xF; /* 4 high bits of GOTC */
5406 adapter->stats.gotc += (tmp << 32);
5407 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5408 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5409 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5410 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
5411 adapter->stats.fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5412 adapter->stats.fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5414 adapter->stats.fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5415 adapter->stats.fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5416 adapter->stats.fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5417 adapter->stats.fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5418 adapter->stats.fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5419 adapter->stats.fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5420 #endif /* IXGBE_FCOE */
5422 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5423 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
5424 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5425 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5426 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5428 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5429 adapter->stats.bprc += bprc;
5430 adapter->stats.mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5431 if (hw->mac.type == ixgbe_mac_82598EB)
5432 adapter->stats.mprc -= bprc;
5433 adapter->stats.roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5434 adapter->stats.prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5435 adapter->stats.prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5436 adapter->stats.prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5437 adapter->stats.prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5438 adapter->stats.prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5439 adapter->stats.prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5440 adapter->stats.rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5441 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5442 adapter->stats.lxontxc += lxon;
5443 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5444 adapter->stats.lxofftxc += lxoff;
5445 adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5446 adapter->stats.gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5447 adapter->stats.mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5449 * 82598 errata - tx of flow control packets is included in tx counters
5451 xon_off_tot = lxon + lxoff;
5452 adapter->stats.gptc -= xon_off_tot;
5453 adapter->stats.mptc -= xon_off_tot;
5454 adapter->stats.gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5455 adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5456 adapter->stats.rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5457 adapter->stats.rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5458 adapter->stats.tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5459 adapter->stats.ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5460 adapter->stats.ptc64 -= xon_off_tot;
5461 adapter->stats.ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5462 adapter->stats.ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5463 adapter->stats.ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5464 adapter->stats.ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5465 adapter->stats.ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5466 adapter->stats.bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5468 /* Fill out the OS statistics structure */
5469 netdev->stats.multicast = adapter->stats.mprc;
5472 netdev->stats.rx_errors = adapter->stats.crcerrs +
5473 adapter->stats.rlec;
5474 netdev->stats.rx_dropped = 0;
5475 netdev->stats.rx_length_errors = adapter->stats.rlec;
5476 netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
5477 netdev->stats.rx_missed_errors = total_mpc;
5481 * ixgbe_watchdog - Timer Call-back
5482 * @data: pointer to adapter cast into an unsigned long
5484 static void ixgbe_watchdog(unsigned long data)
5486 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5487 struct ixgbe_hw *hw = &adapter->hw;
5492 * Do the watchdog outside of interrupt context due to the lovely
5493 * delays that some of the newer hardware requires
5496 if (test_bit(__IXGBE_DOWN, &adapter->state))
5497 goto watchdog_short_circuit;
5499 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5501 * for legacy and MSI interrupts don't set any bits
5502 * that are enabled for EIAM, because this operation
5503 * would set *both* EIMS and EICS for any bit in EIAM
5505 IXGBE_WRITE_REG(hw, IXGBE_EICS,
5506 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5507 goto watchdog_reschedule;
5510 /* get one bit for every active tx/rx interrupt vector */
5511 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
5512 struct ixgbe_q_vector *qv = adapter->q_vector[i];
5513 if (qv->rxr_count || qv->txr_count)
5514 eics |= ((u64)1 << i);
5517 /* Cause software interrupt to ensure rx rings are cleaned */
5518 ixgbe_irq_rearm_queues(adapter, eics);
5520 watchdog_reschedule:
5521 /* Reset the timer */
5522 mod_timer(&adapter->watchdog_timer, round_jiffies(jiffies + 2 * HZ));
5524 watchdog_short_circuit:
5525 schedule_work(&adapter->watchdog_task);
5529 * ixgbe_multispeed_fiber_task - worker thread to configure multispeed fiber
5530 * @work: pointer to work_struct containing our data
5532 static void ixgbe_multispeed_fiber_task(struct work_struct *work)
5534 struct ixgbe_adapter *adapter = container_of(work,
5535 struct ixgbe_adapter,
5536 multispeed_fiber_task);
5537 struct ixgbe_hw *hw = &adapter->hw;
5541 adapter->flags |= IXGBE_FLAG_IN_SFP_LINK_TASK;
5542 autoneg = hw->phy.autoneg_advertised;
5543 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
5544 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
5545 hw->mac.autotry_restart = false;
5546 if (hw->mac.ops.setup_link)
5547 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
5548 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5549 adapter->flags &= ~IXGBE_FLAG_IN_SFP_LINK_TASK;
5553 * ixgbe_sfp_config_module_task - worker thread to configure a new SFP+ module
5554 * @work: pointer to work_struct containing our data
5556 static void ixgbe_sfp_config_module_task(struct work_struct *work)
5558 struct ixgbe_adapter *adapter = container_of(work,
5559 struct ixgbe_adapter,
5560 sfp_config_module_task);
5561 struct ixgbe_hw *hw = &adapter->hw;
5564 adapter->flags |= IXGBE_FLAG_IN_SFP_MOD_TASK;
5566 /* Time for electrical oscillations to settle down */
5568 err = hw->phy.ops.identify_sfp(hw);
5570 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
5571 e_dev_err("failed to initialize because an unsupported SFP+ "
5572 "module type was detected.\n");
5573 e_dev_err("Reload the driver after installing a supported "
5575 unregister_netdev(adapter->netdev);
5578 hw->mac.ops.setup_sfp(hw);
5580 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
5581 /* This will also work for DA Twinax connections */
5582 schedule_work(&adapter->multispeed_fiber_task);
5583 adapter->flags &= ~IXGBE_FLAG_IN_SFP_MOD_TASK;
5587 * ixgbe_fdir_reinit_task - worker thread to reinit FDIR filter table
5588 * @work: pointer to work_struct containing our data
5590 static void ixgbe_fdir_reinit_task(struct work_struct *work)
5592 struct ixgbe_adapter *adapter = container_of(work,
5593 struct ixgbe_adapter,
5595 struct ixgbe_hw *hw = &adapter->hw;
5598 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5599 for (i = 0; i < adapter->num_tx_queues; i++)
5600 set_bit(__IXGBE_FDIR_INIT_DONE,
5601 &(adapter->tx_ring[i]->reinit_state));
5603 e_err(probe, "failed to finish FDIR re-initialization, "
5604 "ignored adding FDIR ATR filters\n");
5606 /* Done FDIR Re-initialization, enable transmits */
5607 netif_tx_start_all_queues(adapter->netdev);
5610 static DEFINE_MUTEX(ixgbe_watchdog_lock);
5613 * ixgbe_watchdog_task - worker thread to bring link up
5614 * @work: pointer to work_struct containing our data
5616 static void ixgbe_watchdog_task(struct work_struct *work)
5618 struct ixgbe_adapter *adapter = container_of(work,
5619 struct ixgbe_adapter,
5621 struct net_device *netdev = adapter->netdev;
5622 struct ixgbe_hw *hw = &adapter->hw;
5626 struct ixgbe_ring *tx_ring;
5627 int some_tx_pending = 0;
5629 mutex_lock(&ixgbe_watchdog_lock);
5631 link_up = adapter->link_up;
5632 link_speed = adapter->link_speed;
5634 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
5635 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5638 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5639 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
5640 hw->mac.ops.fc_enable(hw, i);
5642 hw->mac.ops.fc_enable(hw, 0);
5645 hw->mac.ops.fc_enable(hw, 0);
5650 time_after(jiffies, (adapter->link_check_timeout +
5651 IXGBE_TRY_LINK_TIMEOUT))) {
5652 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5653 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5655 adapter->link_up = link_up;
5656 adapter->link_speed = link_speed;
5660 if (!netif_carrier_ok(netdev)) {
5661 bool flow_rx, flow_tx;
5663 if (hw->mac.type == ixgbe_mac_82599EB) {
5664 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5665 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5666 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5667 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5669 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5670 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5671 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5672 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5675 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
5676 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5678 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5679 "1 Gbps" : "unknown speed")),
5680 ((flow_rx && flow_tx) ? "RX/TX" :
5682 (flow_tx ? "TX" : "None"))));
5684 netif_carrier_on(netdev);
5686 /* Force detection of hung controller */
5687 adapter->detect_tx_hung = true;
5690 adapter->link_up = false;
5691 adapter->link_speed = 0;
5692 if (netif_carrier_ok(netdev)) {
5693 e_info(drv, "NIC Link is Down\n");
5694 netif_carrier_off(netdev);
5698 if (!netif_carrier_ok(netdev)) {
5699 for (i = 0; i < adapter->num_tx_queues; i++) {
5700 tx_ring = adapter->tx_ring[i];
5701 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5702 some_tx_pending = 1;
5707 if (some_tx_pending) {
5708 /* We've lost link, so the controller stops DMA,
5709 * but we've got queued Tx work that's never going
5710 * to get done, so reset controller to flush Tx.
5711 * (Do the reset outside of interrupt context).
5713 schedule_work(&adapter->reset_task);
5717 ixgbe_update_stats(adapter);
5718 mutex_unlock(&ixgbe_watchdog_lock);
5721 static int ixgbe_tso(struct ixgbe_adapter *adapter,
5722 struct ixgbe_ring *tx_ring, struct sk_buff *skb,
5723 u32 tx_flags, u8 *hdr_len)
5725 struct ixgbe_adv_tx_context_desc *context_desc;
5728 struct ixgbe_tx_buffer *tx_buffer_info;
5729 u32 vlan_macip_lens = 0, type_tucmd_mlhl;
5730 u32 mss_l4len_idx, l4len;
5732 if (skb_is_gso(skb)) {
5733 if (skb_header_cloned(skb)) {
5734 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5738 l4len = tcp_hdrlen(skb);
5741 if (skb->protocol == htons(ETH_P_IP)) {
5742 struct iphdr *iph = ip_hdr(skb);
5745 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
5749 } else if (skb_is_gso_v6(skb)) {
5750 ipv6_hdr(skb)->payload_len = 0;
5751 tcp_hdr(skb)->check =
5752 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5753 &ipv6_hdr(skb)->daddr,
5757 i = tx_ring->next_to_use;
5759 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5760 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
5762 /* VLAN MACLEN IPLEN */
5763 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5765 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
5766 vlan_macip_lens |= ((skb_network_offset(skb)) <<
5767 IXGBE_ADVTXD_MACLEN_SHIFT);
5768 *hdr_len += skb_network_offset(skb);
5770 (skb_transport_header(skb) - skb_network_header(skb));
5772 (skb_transport_header(skb) - skb_network_header(skb));
5773 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
5774 context_desc->seqnum_seed = 0;
5776 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
5777 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
5778 IXGBE_ADVTXD_DTYP_CTXT);
5780 if (skb->protocol == htons(ETH_P_IP))
5781 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
5782 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5783 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
5787 (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
5788 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
5789 /* use index 1 for TSO */
5790 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
5791 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
5793 tx_buffer_info->time_stamp = jiffies;
5794 tx_buffer_info->next_to_watch = i;
5797 if (i == tx_ring->count)
5799 tx_ring->next_to_use = i;
5806 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
5807 struct ixgbe_ring *tx_ring,
5808 struct sk_buff *skb, u32 tx_flags)
5810 struct ixgbe_adv_tx_context_desc *context_desc;
5812 struct ixgbe_tx_buffer *tx_buffer_info;
5813 u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
5815 if (skb->ip_summed == CHECKSUM_PARTIAL ||
5816 (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
5817 i = tx_ring->next_to_use;
5818 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5819 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
5821 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5823 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
5824 vlan_macip_lens |= (skb_network_offset(skb) <<
5825 IXGBE_ADVTXD_MACLEN_SHIFT);
5826 if (skb->ip_summed == CHECKSUM_PARTIAL)
5827 vlan_macip_lens |= (skb_transport_header(skb) -
5828 skb_network_header(skb));
5830 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
5831 context_desc->seqnum_seed = 0;
5833 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
5834 IXGBE_ADVTXD_DTYP_CTXT);
5836 if (skb->ip_summed == CHECKSUM_PARTIAL) {
5839 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
5840 const struct vlan_ethhdr *vhdr =
5841 (const struct vlan_ethhdr *)skb->data;
5843 protocol = vhdr->h_vlan_encapsulated_proto;
5845 protocol = skb->protocol;
5849 case cpu_to_be16(ETH_P_IP):
5850 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
5851 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
5853 IXGBE_ADVTXD_TUCMD_L4T_TCP;
5854 else if (ip_hdr(skb)->protocol == IPPROTO_SCTP)
5856 IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5858 case cpu_to_be16(ETH_P_IPV6):
5859 /* XXX what about other V6 headers?? */
5860 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
5862 IXGBE_ADVTXD_TUCMD_L4T_TCP;
5863 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_SCTP)
5865 IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5868 if (unlikely(net_ratelimit())) {
5869 e_warn(probe, "partial checksum "
5877 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
5878 /* use index zero for tx checksum offload */
5879 context_desc->mss_l4len_idx = 0;
5881 tx_buffer_info->time_stamp = jiffies;
5882 tx_buffer_info->next_to_watch = i;
5885 if (i == tx_ring->count)
5887 tx_ring->next_to_use = i;
5895 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
5896 struct ixgbe_ring *tx_ring,
5897 struct sk_buff *skb, u32 tx_flags,
5900 struct pci_dev *pdev = adapter->pdev;
5901 struct ixgbe_tx_buffer *tx_buffer_info;
5903 unsigned int total = skb->len;
5904 unsigned int offset = 0, size, count = 0, i;
5905 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
5908 i = tx_ring->next_to_use;
5910 if (tx_flags & IXGBE_TX_FLAGS_FCOE)
5911 /* excluding fcoe_crc_eof for FCoE */
5912 total -= sizeof(struct fcoe_crc_eof);
5914 len = min(skb_headlen(skb), total);
5916 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5917 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
5919 tx_buffer_info->length = size;
5920 tx_buffer_info->mapped_as_page = false;
5921 tx_buffer_info->dma = dma_map_single(&pdev->dev,
5923 size, DMA_TO_DEVICE);
5924 if (dma_mapping_error(&pdev->dev, tx_buffer_info->dma))
5926 tx_buffer_info->time_stamp = jiffies;
5927 tx_buffer_info->next_to_watch = i;
5936 if (i == tx_ring->count)
5941 for (f = 0; f < nr_frags; f++) {
5942 struct skb_frag_struct *frag;
5944 frag = &skb_shinfo(skb)->frags[f];
5945 len = min((unsigned int)frag->size, total);
5946 offset = frag->page_offset;
5950 if (i == tx_ring->count)
5953 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5954 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
5956 tx_buffer_info->length = size;
5957 tx_buffer_info->dma = dma_map_page(&adapter->pdev->dev,
5961 tx_buffer_info->mapped_as_page = true;
5962 if (dma_mapping_error(&pdev->dev, tx_buffer_info->dma))
5964 tx_buffer_info->time_stamp = jiffies;
5965 tx_buffer_info->next_to_watch = i;
5976 tx_ring->tx_buffer_info[i].skb = skb;
5977 tx_ring->tx_buffer_info[first].next_to_watch = i;
5982 e_dev_err("TX DMA map failed\n");
5984 /* clear timestamp and dma mappings for failed tx_buffer_info map */
5985 tx_buffer_info->dma = 0;
5986 tx_buffer_info->time_stamp = 0;
5987 tx_buffer_info->next_to_watch = 0;
5991 /* clear timestamp and dma mappings for remaining portion of packet */
5994 i += tx_ring->count;
5996 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5997 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
6003 static void ixgbe_tx_queue(struct ixgbe_adapter *adapter,
6004 struct ixgbe_ring *tx_ring,
6005 int tx_flags, int count, u32 paylen, u8 hdr_len)
6007 union ixgbe_adv_tx_desc *tx_desc = NULL;
6008 struct ixgbe_tx_buffer *tx_buffer_info;
6009 u32 olinfo_status = 0, cmd_type_len = 0;
6011 u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
6013 cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
6015 cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
6017 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
6018 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
6020 if (tx_flags & IXGBE_TX_FLAGS_TSO) {
6021 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
6023 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
6024 IXGBE_ADVTXD_POPTS_SHIFT;
6026 /* use index 1 context for tso */
6027 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6028 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
6029 olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
6030 IXGBE_ADVTXD_POPTS_SHIFT;
6032 } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
6033 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
6034 IXGBE_ADVTXD_POPTS_SHIFT;
6036 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6037 olinfo_status |= IXGBE_ADVTXD_CC;
6038 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6039 if (tx_flags & IXGBE_TX_FLAGS_FSO)
6040 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
6043 olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
6045 i = tx_ring->next_to_use;
6047 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6048 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
6049 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
6050 tx_desc->read.cmd_type_len =
6051 cpu_to_le32(cmd_type_len | tx_buffer_info->length);
6052 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
6054 if (i == tx_ring->count)
6058 tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
6061 * Force memory writes to complete before letting h/w
6062 * know there are new descriptors to fetch. (Only
6063 * applicable for weak-ordered memory model archs,
6068 tx_ring->next_to_use = i;
6069 writel(i, adapter->hw.hw_addr + tx_ring->tail);
6072 static void ixgbe_atr(struct ixgbe_adapter *adapter, struct sk_buff *skb,
6073 int queue, u32 tx_flags)
6075 struct ixgbe_atr_input atr_input;
6077 struct iphdr *iph = ip_hdr(skb);
6078 struct ethhdr *eth = (struct ethhdr *)skb->data;
6079 u16 vlan_id, src_port, dst_port, flex_bytes;
6080 u32 src_ipv4_addr, dst_ipv4_addr;
6083 /* Right now, we support IPv4 only */
6084 if (skb->protocol != htons(ETH_P_IP))
6086 /* check if we're UDP or TCP */
6087 if (iph->protocol == IPPROTO_TCP) {
6089 src_port = th->source;
6090 dst_port = th->dest;
6091 l4type |= IXGBE_ATR_L4TYPE_TCP;
6092 /* l4type IPv4 type is 0, no need to assign */
6094 /* Unsupported L4 header, just bail here */
6098 memset(&atr_input, 0, sizeof(struct ixgbe_atr_input));
6100 vlan_id = (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK) >>
6101 IXGBE_TX_FLAGS_VLAN_SHIFT;
6102 src_ipv4_addr = iph->saddr;
6103 dst_ipv4_addr = iph->daddr;
6104 flex_bytes = eth->h_proto;
6106 ixgbe_atr_set_vlan_id_82599(&atr_input, vlan_id);
6107 ixgbe_atr_set_src_port_82599(&atr_input, dst_port);
6108 ixgbe_atr_set_dst_port_82599(&atr_input, src_port);
6109 ixgbe_atr_set_flex_byte_82599(&atr_input, flex_bytes);
6110 ixgbe_atr_set_l4type_82599(&atr_input, l4type);
6111 /* src and dst are inverted, think how the receiver sees them */
6112 ixgbe_atr_set_src_ipv4_82599(&atr_input, dst_ipv4_addr);
6113 ixgbe_atr_set_dst_ipv4_82599(&atr_input, src_ipv4_addr);
6115 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
6116 ixgbe_fdir_add_signature_filter_82599(&adapter->hw, &atr_input, queue);
6119 static int __ixgbe_maybe_stop_tx(struct net_device *netdev,
6120 struct ixgbe_ring *tx_ring, int size)
6122 netif_stop_subqueue(netdev, tx_ring->queue_index);
6123 /* Herbert's original patch had:
6124 * smp_mb__after_netif_stop_queue();
6125 * but since that doesn't exist yet, just open code it. */
6128 /* We need to check again in a case another CPU has just
6129 * made room available. */
6130 if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
6133 /* A reprieve! - use start_queue because it doesn't call schedule */
6134 netif_start_subqueue(netdev, tx_ring->queue_index);
6135 ++tx_ring->restart_queue;
6139 static int ixgbe_maybe_stop_tx(struct net_device *netdev,
6140 struct ixgbe_ring *tx_ring, int size)
6142 if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
6144 return __ixgbe_maybe_stop_tx(netdev, tx_ring, size);
6147 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6149 struct ixgbe_adapter *adapter = netdev_priv(dev);
6150 int txq = smp_processor_id();
6152 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6153 while (unlikely(txq >= dev->real_num_tx_queues))
6154 txq -= dev->real_num_tx_queues;
6159 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
6160 ((skb->protocol == htons(ETH_P_FCOE)) ||
6161 (skb->protocol == htons(ETH_P_FIP)))) {
6162 txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
6163 txq += adapter->ring_feature[RING_F_FCOE].mask;
6167 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6168 if (skb->priority == TC_PRIO_CONTROL)
6169 txq = adapter->ring_feature[RING_F_DCB].indices-1;
6171 txq = (skb->vlan_tci & IXGBE_TX_FLAGS_VLAN_PRIO_MASK)
6176 return skb_tx_hash(dev, skb);
6179 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
6180 struct net_device *netdev)
6182 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6183 struct ixgbe_ring *tx_ring;
6184 struct netdev_queue *txq;
6186 unsigned int tx_flags = 0;
6192 if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
6193 tx_flags |= vlan_tx_tag_get(skb);
6194 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6195 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6196 tx_flags |= ((skb->queue_mapping & 0x7) << 13);
6198 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6199 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6200 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED &&
6201 skb->priority != TC_PRIO_CONTROL) {
6202 tx_flags |= ((skb->queue_mapping & 0x7) << 13);
6203 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6204 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6207 tx_ring = adapter->tx_ring[skb->queue_mapping];
6210 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
6211 #ifdef CONFIG_IXGBE_DCB
6212 /* for FCoE with DCB, we force the priority to what
6213 * was specified by the switch */
6214 if ((skb->protocol == htons(ETH_P_FCOE)) ||
6215 (skb->protocol == htons(ETH_P_FIP))) {
6216 tx_flags &= ~(IXGBE_TX_FLAGS_VLAN_PRIO_MASK
6217 << IXGBE_TX_FLAGS_VLAN_SHIFT);
6218 tx_flags |= ((adapter->fcoe.up << 13)
6219 << IXGBE_TX_FLAGS_VLAN_SHIFT);
6222 /* flag for FCoE offloads */
6223 if (skb->protocol == htons(ETH_P_FCOE))
6224 tx_flags |= IXGBE_TX_FLAGS_FCOE;
6228 /* four things can cause us to need a context descriptor */
6229 if (skb_is_gso(skb) ||
6230 (skb->ip_summed == CHECKSUM_PARTIAL) ||
6231 (tx_flags & IXGBE_TX_FLAGS_VLAN) ||
6232 (tx_flags & IXGBE_TX_FLAGS_FCOE))
6235 count += TXD_USE_COUNT(skb_headlen(skb));
6236 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6237 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6239 if (ixgbe_maybe_stop_tx(netdev, tx_ring, count)) {
6241 return NETDEV_TX_BUSY;
6244 first = tx_ring->next_to_use;
6245 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6247 /* setup tx offload for FCoE */
6248 tso = ixgbe_fso(adapter, tx_ring, skb, tx_flags, &hdr_len);
6250 dev_kfree_skb_any(skb);
6251 return NETDEV_TX_OK;
6254 tx_flags |= IXGBE_TX_FLAGS_FSO;
6255 #endif /* IXGBE_FCOE */
6257 if (skb->protocol == htons(ETH_P_IP))
6258 tx_flags |= IXGBE_TX_FLAGS_IPV4;
6259 tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len);
6261 dev_kfree_skb_any(skb);
6262 return NETDEV_TX_OK;
6266 tx_flags |= IXGBE_TX_FLAGS_TSO;
6267 else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags) &&
6268 (skb->ip_summed == CHECKSUM_PARTIAL))
6269 tx_flags |= IXGBE_TX_FLAGS_CSUM;
6272 count = ixgbe_tx_map(adapter, tx_ring, skb, tx_flags, first);
6274 /* add the ATR filter if ATR is on */
6275 if (tx_ring->atr_sample_rate) {
6276 ++tx_ring->atr_count;
6277 if ((tx_ring->atr_count >= tx_ring->atr_sample_rate) &&
6278 test_bit(__IXGBE_FDIR_INIT_DONE,
6279 &tx_ring->reinit_state)) {
6280 ixgbe_atr(adapter, skb, tx_ring->queue_index,
6282 tx_ring->atr_count = 0;
6285 txq = netdev_get_tx_queue(netdev, tx_ring->queue_index);
6286 txq->tx_bytes += skb->len;
6288 ixgbe_tx_queue(adapter, tx_ring, tx_flags, count, skb->len,
6290 ixgbe_maybe_stop_tx(netdev, tx_ring, DESC_NEEDED);
6293 dev_kfree_skb_any(skb);
6294 tx_ring->tx_buffer_info[first].time_stamp = 0;
6295 tx_ring->next_to_use = first;
6298 return NETDEV_TX_OK;
6302 * ixgbe_set_mac - Change the Ethernet Address of the NIC
6303 * @netdev: network interface device structure
6304 * @p: pointer to an address structure
6306 * Returns 0 on success, negative on failure
6308 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6310 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6311 struct ixgbe_hw *hw = &adapter->hw;
6312 struct sockaddr *addr = p;
6314 if (!is_valid_ether_addr(addr->sa_data))
6315 return -EADDRNOTAVAIL;
6317 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6318 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6320 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
6327 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6329 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6330 struct ixgbe_hw *hw = &adapter->hw;
6334 if (prtad != hw->phy.mdio.prtad)
6336 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6342 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6343 u16 addr, u16 value)
6345 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6346 struct ixgbe_hw *hw = &adapter->hw;
6348 if (prtad != hw->phy.mdio.prtad)
6350 return hw->phy.ops.write_reg(hw, addr, devad, value);
6353 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6355 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6357 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6361 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6363 * @netdev: network interface device structure
6365 * Returns non-zero on failure
6367 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6370 struct ixgbe_adapter *adapter = netdev_priv(dev);
6371 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6373 if (is_valid_ether_addr(mac->san_addr)) {
6375 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6382 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6384 * @netdev: network interface device structure
6386 * Returns non-zero on failure
6388 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6391 struct ixgbe_adapter *adapter = netdev_priv(dev);
6392 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6394 if (is_valid_ether_addr(mac->san_addr)) {
6396 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6402 #ifdef CONFIG_NET_POLL_CONTROLLER
6404 * Polling 'interrupt' - used by things like netconsole to send skbs
6405 * without having to re-enable interrupts. It's not called while
6406 * the interrupt routine is executing.
6408 static void ixgbe_netpoll(struct net_device *netdev)
6410 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6413 /* if interface is down do nothing */
6414 if (test_bit(__IXGBE_DOWN, &adapter->state))
6417 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
6418 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
6419 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
6420 for (i = 0; i < num_q_vectors; i++) {
6421 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
6422 ixgbe_msix_clean_many(0, q_vector);
6425 ixgbe_intr(adapter->pdev->irq, netdev);
6427 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
6431 static const struct net_device_ops ixgbe_netdev_ops = {
6432 .ndo_open = ixgbe_open,
6433 .ndo_stop = ixgbe_close,
6434 .ndo_start_xmit = ixgbe_xmit_frame,
6435 .ndo_select_queue = ixgbe_select_queue,
6436 .ndo_set_rx_mode = ixgbe_set_rx_mode,
6437 .ndo_set_multicast_list = ixgbe_set_rx_mode,
6438 .ndo_validate_addr = eth_validate_addr,
6439 .ndo_set_mac_address = ixgbe_set_mac,
6440 .ndo_change_mtu = ixgbe_change_mtu,
6441 .ndo_tx_timeout = ixgbe_tx_timeout,
6442 .ndo_vlan_rx_register = ixgbe_vlan_rx_register,
6443 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
6444 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
6445 .ndo_do_ioctl = ixgbe_ioctl,
6446 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
6447 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
6448 .ndo_set_vf_tx_rate = ixgbe_ndo_set_vf_bw,
6449 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
6450 #ifdef CONFIG_NET_POLL_CONTROLLER
6451 .ndo_poll_controller = ixgbe_netpoll,
6454 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
6455 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
6456 .ndo_fcoe_enable = ixgbe_fcoe_enable,
6457 .ndo_fcoe_disable = ixgbe_fcoe_disable,
6458 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
6459 #endif /* IXGBE_FCOE */
6462 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
6463 const struct ixgbe_info *ii)
6465 #ifdef CONFIG_PCI_IOV
6466 struct ixgbe_hw *hw = &adapter->hw;
6469 if (hw->mac.type != ixgbe_mac_82599EB || !max_vfs)
6472 /* The 82599 supports up to 64 VFs per physical function
6473 * but this implementation limits allocation to 63 so that
6474 * basic networking resources are still available to the
6477 adapter->num_vfs = (max_vfs > 63) ? 63 : max_vfs;
6478 adapter->flags |= IXGBE_FLAG_SRIOV_ENABLED;
6479 err = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
6481 e_err(probe, "Failed to enable PCI sriov: %d\n", err);
6484 /* If call to enable VFs succeeded then allocate memory
6485 * for per VF control structures.
6488 kcalloc(adapter->num_vfs,
6489 sizeof(struct vf_data_storage), GFP_KERNEL);
6490 if (adapter->vfinfo) {
6491 /* Now that we're sure SR-IOV is enabled
6492 * and memory allocated set up the mailbox parameters
6494 ixgbe_init_mbx_params_pf(hw);
6495 memcpy(&hw->mbx.ops, ii->mbx_ops,
6496 sizeof(hw->mbx.ops));
6498 /* Disable RSC when in SR-IOV mode */
6499 adapter->flags2 &= ~(IXGBE_FLAG2_RSC_CAPABLE |
6500 IXGBE_FLAG2_RSC_ENABLED);
6505 e_err(probe, "Unable to allocate memory for VF Data Storage - "
6506 "SRIOV disabled\n");
6507 pci_disable_sriov(adapter->pdev);
6510 adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
6511 adapter->num_vfs = 0;
6512 #endif /* CONFIG_PCI_IOV */
6516 * ixgbe_probe - Device Initialization Routine
6517 * @pdev: PCI device information struct
6518 * @ent: entry in ixgbe_pci_tbl
6520 * Returns 0 on success, negative on failure
6522 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
6523 * The OS initialization, configuring of the adapter private structure,
6524 * and a hardware reset occur.
6526 static int __devinit ixgbe_probe(struct pci_dev *pdev,
6527 const struct pci_device_id *ent)
6529 struct net_device *netdev;
6530 struct ixgbe_adapter *adapter = NULL;
6531 struct ixgbe_hw *hw;
6532 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
6533 static int cards_found;
6534 int i, err, pci_using_dac;
6535 unsigned int indices = num_possible_cpus();
6541 err = pci_enable_device_mem(pdev);
6545 if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
6546 !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
6549 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
6551 err = dma_set_coherent_mask(&pdev->dev,
6554 e_dev_err("No usable DMA configuration, "
6562 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
6563 IORESOURCE_MEM), ixgbe_driver_name);
6565 e_dev_err("pci_request_selected_regions failed 0x%x\n", err);
6569 pci_enable_pcie_error_reporting(pdev);
6571 pci_set_master(pdev);
6572 pci_save_state(pdev);
6574 if (ii->mac == ixgbe_mac_82598EB)
6575 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
6577 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
6579 indices = max_t(unsigned int, indices, IXGBE_MAX_DCB_INDICES);
6581 indices += min_t(unsigned int, num_possible_cpus(),
6582 IXGBE_MAX_FCOE_INDICES);
6584 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
6587 goto err_alloc_etherdev;
6590 SET_NETDEV_DEV(netdev, &pdev->dev);
6592 pci_set_drvdata(pdev, netdev);
6593 adapter = netdev_priv(netdev);
6595 adapter->netdev = netdev;
6596 adapter->pdev = pdev;
6599 adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
6601 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
6602 pci_resource_len(pdev, 0));
6608 for (i = 1; i <= 5; i++) {
6609 if (pci_resource_len(pdev, i) == 0)
6613 netdev->netdev_ops = &ixgbe_netdev_ops;
6614 ixgbe_set_ethtool_ops(netdev);
6615 netdev->watchdog_timeo = 5 * HZ;
6616 strcpy(netdev->name, pci_name(pdev));
6618 adapter->bd_number = cards_found;
6621 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
6622 hw->mac.type = ii->mac;
6625 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
6626 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
6627 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
6628 if (!(eec & (1 << 8)))
6629 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
6632 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
6633 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
6634 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
6635 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
6636 hw->phy.mdio.mmds = 0;
6637 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
6638 hw->phy.mdio.dev = netdev;
6639 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
6640 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
6642 /* set up this timer and work struct before calling get_invariants
6643 * which might start the timer
6645 init_timer(&adapter->sfp_timer);
6646 adapter->sfp_timer.function = &ixgbe_sfp_timer;
6647 adapter->sfp_timer.data = (unsigned long) adapter;
6649 INIT_WORK(&adapter->sfp_task, ixgbe_sfp_task);
6651 /* multispeed fiber has its own tasklet, called from GPI SDP1 context */
6652 INIT_WORK(&adapter->multispeed_fiber_task, ixgbe_multispeed_fiber_task);
6654 /* a new SFP+ module arrival, called from GPI SDP2 context */
6655 INIT_WORK(&adapter->sfp_config_module_task,
6656 ixgbe_sfp_config_module_task);
6658 ii->get_invariants(hw);
6660 /* setup the private structure */
6661 err = ixgbe_sw_init(adapter);
6665 /* Make it possible the adapter to be woken up via WOL */
6666 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
6667 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6670 * If there is a fan on this device and it has failed log the
6673 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
6674 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
6675 if (esdp & IXGBE_ESDP_SDP1)
6676 e_crit(probe, "Fan has stopped, replace the adapter\n");
6679 /* reset_hw fills in the perm_addr as well */
6680 hw->phy.reset_if_overtemp = true;
6681 err = hw->mac.ops.reset_hw(hw);
6682 hw->phy.reset_if_overtemp = false;
6683 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
6684 hw->mac.type == ixgbe_mac_82598EB) {
6686 * Start a kernel thread to watch for a module to arrive.
6687 * Only do this for 82598, since 82599 will generate
6688 * interrupts on module arrival.
6690 set_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
6691 mod_timer(&adapter->sfp_timer,
6692 round_jiffies(jiffies + (2 * HZ)));
6694 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
6695 e_dev_err("failed to initialize because an unsupported SFP+ "
6696 "module type was detected.\n");
6697 e_dev_err("Reload the driver after installing a supported "
6701 e_dev_err("HW Init failed: %d\n", err);
6705 ixgbe_probe_vf(adapter, ii);
6707 netdev->features = NETIF_F_SG |
6709 NETIF_F_HW_VLAN_TX |
6710 NETIF_F_HW_VLAN_RX |
6711 NETIF_F_HW_VLAN_FILTER;
6713 netdev->features |= NETIF_F_IPV6_CSUM;
6714 netdev->features |= NETIF_F_TSO;
6715 netdev->features |= NETIF_F_TSO6;
6716 netdev->features |= NETIF_F_GRO;
6718 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
6719 netdev->features |= NETIF_F_SCTP_CSUM;
6721 netdev->vlan_features |= NETIF_F_TSO;
6722 netdev->vlan_features |= NETIF_F_TSO6;
6723 netdev->vlan_features |= NETIF_F_IP_CSUM;
6724 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
6725 netdev->vlan_features |= NETIF_F_SG;
6727 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6728 adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
6729 IXGBE_FLAG_DCB_ENABLED);
6730 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
6731 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
6733 #ifdef CONFIG_IXGBE_DCB
6734 netdev->dcbnl_ops = &dcbnl_ops;
6738 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
6739 if (hw->mac.ops.get_device_caps) {
6740 hw->mac.ops.get_device_caps(hw, &device_caps);
6741 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
6742 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6745 #endif /* IXGBE_FCOE */
6747 netdev->features |= NETIF_F_HIGHDMA;
6749 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
6750 netdev->features |= NETIF_F_LRO;
6752 /* make sure the EEPROM is good */
6753 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
6754 e_dev_err("The EEPROM Checksum Is Not Valid\n");
6759 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
6760 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
6762 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
6763 e_dev_err("invalid MAC address\n");
6768 /* power down the optics */
6769 if (hw->phy.multispeed_fiber)
6770 hw->mac.ops.disable_tx_laser(hw);
6772 init_timer(&adapter->watchdog_timer);
6773 adapter->watchdog_timer.function = &ixgbe_watchdog;
6774 adapter->watchdog_timer.data = (unsigned long)adapter;
6776 INIT_WORK(&adapter->reset_task, ixgbe_reset_task);
6777 INIT_WORK(&adapter->watchdog_task, ixgbe_watchdog_task);
6779 err = ixgbe_init_interrupt_scheme(adapter);
6783 switch (pdev->device) {
6784 case IXGBE_DEV_ID_82599_KX4:
6785 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
6786 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
6792 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
6794 /* pick up the PCI bus settings for reporting later */
6795 hw->mac.ops.get_bus_info(hw);
6797 /* print bus type/speed/width info */
6798 e_dev_info("(PCI Express:%s:%s) %pM\n",
6799 ((hw->bus.speed == ixgbe_bus_speed_5000) ? "5.0Gb/s":
6800 (hw->bus.speed == ixgbe_bus_speed_2500) ? "2.5Gb/s":"Unknown"),
6801 ((hw->bus.width == ixgbe_bus_width_pcie_x8) ? "Width x8" :
6802 (hw->bus.width == ixgbe_bus_width_pcie_x4) ? "Width x4" :
6803 (hw->bus.width == ixgbe_bus_width_pcie_x1) ? "Width x1" :
6806 ixgbe_read_pba_num_generic(hw, &part_num);
6807 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
6808 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, "
6809 "PBA No: %06x-%03x\n",
6810 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
6811 (part_num >> 8), (part_num & 0xff));
6813 e_dev_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
6814 hw->mac.type, hw->phy.type,
6815 (part_num >> 8), (part_num & 0xff));
6817 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
6818 e_dev_warn("PCI-Express bandwidth available for this card is "
6819 "not sufficient for optimal performance.\n");
6820 e_dev_warn("For optimal performance a x8 PCI-Express slot "
6824 /* save off EEPROM version number */
6825 hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
6827 /* reset the hardware with the new settings */
6828 err = hw->mac.ops.start_hw(hw);
6830 if (err == IXGBE_ERR_EEPROM_VERSION) {
6831 /* We are running on a pre-production device, log a warning */
6832 e_dev_warn("This device is a pre-production adapter/LOM. "
6833 "Please be aware there may be issues associated "
6834 "with your hardware. If you are experiencing "
6835 "problems please contact your Intel or hardware "
6836 "representative who provided you with this "
6839 strcpy(netdev->name, "eth%d");
6840 err = register_netdev(netdev);
6844 /* carrier off reporting is important to ethtool even BEFORE open */
6845 netif_carrier_off(netdev);
6847 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
6848 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
6849 INIT_WORK(&adapter->fdir_reinit_task, ixgbe_fdir_reinit_task);
6851 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
6852 INIT_WORK(&adapter->check_overtemp_task, ixgbe_check_overtemp_task);
6853 #ifdef CONFIG_IXGBE_DCA
6854 if (dca_add_requester(&pdev->dev) == 0) {
6855 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
6856 ixgbe_setup_dca(adapter);
6859 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
6860 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
6861 for (i = 0; i < adapter->num_vfs; i++)
6862 ixgbe_vf_configuration(pdev, (i | 0x10000000));
6865 /* add san mac addr to netdev */
6866 ixgbe_add_sanmac_netdev(netdev);
6868 e_dev_info("Intel(R) 10 Gigabit Network Connection\n");
6873 ixgbe_release_hw_control(adapter);
6874 ixgbe_clear_interrupt_scheme(adapter);
6877 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6878 ixgbe_disable_sriov(adapter);
6879 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
6880 del_timer_sync(&adapter->sfp_timer);
6881 cancel_work_sync(&adapter->sfp_task);
6882 cancel_work_sync(&adapter->multispeed_fiber_task);
6883 cancel_work_sync(&adapter->sfp_config_module_task);
6884 iounmap(hw->hw_addr);
6886 free_netdev(netdev);
6888 pci_release_selected_regions(pdev, pci_select_bars(pdev,
6892 pci_disable_device(pdev);
6897 * ixgbe_remove - Device Removal Routine
6898 * @pdev: PCI device information struct
6900 * ixgbe_remove is called by the PCI subsystem to alert the driver
6901 * that it should release a PCI device. The could be caused by a
6902 * Hot-Plug event, or because the driver is going to be removed from
6905 static void __devexit ixgbe_remove(struct pci_dev *pdev)
6907 struct net_device *netdev = pci_get_drvdata(pdev);
6908 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6910 set_bit(__IXGBE_DOWN, &adapter->state);
6911 /* clear the module not found bit to make sure the worker won't
6914 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
6915 del_timer_sync(&adapter->watchdog_timer);
6917 del_timer_sync(&adapter->sfp_timer);
6918 cancel_work_sync(&adapter->watchdog_task);
6919 cancel_work_sync(&adapter->sfp_task);
6920 cancel_work_sync(&adapter->multispeed_fiber_task);
6921 cancel_work_sync(&adapter->sfp_config_module_task);
6922 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
6923 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
6924 cancel_work_sync(&adapter->fdir_reinit_task);
6925 flush_scheduled_work();
6927 #ifdef CONFIG_IXGBE_DCA
6928 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
6929 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
6930 dca_remove_requester(&pdev->dev);
6931 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
6936 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
6937 ixgbe_cleanup_fcoe(adapter);
6939 #endif /* IXGBE_FCOE */
6941 /* remove the added san mac */
6942 ixgbe_del_sanmac_netdev(netdev);
6944 if (netdev->reg_state == NETREG_REGISTERED)
6945 unregister_netdev(netdev);
6947 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6948 ixgbe_disable_sriov(adapter);
6950 ixgbe_clear_interrupt_scheme(adapter);
6952 ixgbe_release_hw_control(adapter);
6954 iounmap(adapter->hw.hw_addr);
6955 pci_release_selected_regions(pdev, pci_select_bars(pdev,
6958 e_dev_info("complete\n");
6960 free_netdev(netdev);
6962 pci_disable_pcie_error_reporting(pdev);
6964 pci_disable_device(pdev);
6968 * ixgbe_io_error_detected - called when PCI error is detected
6969 * @pdev: Pointer to PCI device
6970 * @state: The current pci connection state
6972 * This function is called after a PCI bus error affecting
6973 * this device has been detected.
6975 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
6976 pci_channel_state_t state)
6978 struct net_device *netdev = pci_get_drvdata(pdev);
6979 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6981 netif_device_detach(netdev);
6983 if (state == pci_channel_io_perm_failure)
6984 return PCI_ERS_RESULT_DISCONNECT;
6986 if (netif_running(netdev))
6987 ixgbe_down(adapter);
6988 pci_disable_device(pdev);
6990 /* Request a slot reset. */
6991 return PCI_ERS_RESULT_NEED_RESET;
6995 * ixgbe_io_slot_reset - called after the pci bus has been reset.
6996 * @pdev: Pointer to PCI device
6998 * Restart the card from scratch, as if from a cold-boot.
7000 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7002 struct net_device *netdev = pci_get_drvdata(pdev);
7003 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7004 pci_ers_result_t result;
7007 if (pci_enable_device_mem(pdev)) {
7008 e_err(probe, "Cannot re-enable PCI device after reset.\n");
7009 result = PCI_ERS_RESULT_DISCONNECT;
7011 pci_set_master(pdev);
7012 pci_restore_state(pdev);
7013 pci_save_state(pdev);
7015 pci_wake_from_d3(pdev, false);
7017 ixgbe_reset(adapter);
7018 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7019 result = PCI_ERS_RESULT_RECOVERED;
7022 err = pci_cleanup_aer_uncorrect_error_status(pdev);
7024 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7025 "failed 0x%0x\n", err);
7026 /* non-fatal, continue */
7033 * ixgbe_io_resume - called when traffic can start flowing again.
7034 * @pdev: Pointer to PCI device
7036 * This callback is called when the error recovery driver tells us that
7037 * its OK to resume normal operation.
7039 static void ixgbe_io_resume(struct pci_dev *pdev)
7041 struct net_device *netdev = pci_get_drvdata(pdev);
7042 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7044 if (netif_running(netdev)) {
7045 if (ixgbe_up(adapter)) {
7046 e_info(probe, "ixgbe_up failed after reset\n");
7051 netif_device_attach(netdev);
7054 static struct pci_error_handlers ixgbe_err_handler = {
7055 .error_detected = ixgbe_io_error_detected,
7056 .slot_reset = ixgbe_io_slot_reset,
7057 .resume = ixgbe_io_resume,
7060 static struct pci_driver ixgbe_driver = {
7061 .name = ixgbe_driver_name,
7062 .id_table = ixgbe_pci_tbl,
7063 .probe = ixgbe_probe,
7064 .remove = __devexit_p(ixgbe_remove),
7066 .suspend = ixgbe_suspend,
7067 .resume = ixgbe_resume,
7069 .shutdown = ixgbe_shutdown,
7070 .err_handler = &ixgbe_err_handler
7074 * ixgbe_init_module - Driver Registration Routine
7076 * ixgbe_init_module is the first routine called when the driver is
7077 * loaded. All it does is register with the PCI subsystem.
7079 static int __init ixgbe_init_module(void)
7082 pr_info("%s - version %s\n", ixgbe_driver_string,
7083 ixgbe_driver_version);
7084 pr_info("%s\n", ixgbe_copyright);
7086 #ifdef CONFIG_IXGBE_DCA
7087 dca_register_notify(&dca_notifier);
7090 ret = pci_register_driver(&ixgbe_driver);
7094 module_init(ixgbe_init_module);
7097 * ixgbe_exit_module - Driver Exit Cleanup Routine
7099 * ixgbe_exit_module is called just before the driver is removed
7102 static void __exit ixgbe_exit_module(void)
7104 #ifdef CONFIG_IXGBE_DCA
7105 dca_unregister_notify(&dca_notifier);
7107 pci_unregister_driver(&ixgbe_driver);
7110 #ifdef CONFIG_IXGBE_DCA
7111 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
7116 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
7117 __ixgbe_notify_dca);
7119 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7122 #endif /* CONFIG_IXGBE_DCA */
7125 * ixgbe_get_hw_dev return device
7126 * used by hardware layer to print debugging information
7128 struct net_device *ixgbe_get_hw_dev(struct ixgbe_hw *hw)
7130 struct ixgbe_adapter *adapter = hw->back;
7131 return adapter->netdev;
7134 module_exit(ixgbe_exit_module);