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("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("tx hang %d detected, resetting adapter\n",
816 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("Network adapter has been stopped because it "
1656 "has 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("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("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("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("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("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_set_rx_mode - Unicast, Multicast and Promiscuous mode set
2996 * @netdev: network interface device structure
2998 * The set_rx_method entry point is called whenever the unicast/multicast
2999 * address list or the network interface flags are updated. This routine is
3000 * responsible for configuring the hardware for proper unicast, multicast and
3003 void ixgbe_set_rx_mode(struct net_device *netdev)
3005 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3006 struct ixgbe_hw *hw = &adapter->hw;
3009 /* Check for Promiscuous and All Multicast modes */
3011 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3013 if (netdev->flags & IFF_PROMISC) {
3014 hw->addr_ctrl.user_set_promisc = true;
3015 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3016 /* don't hardware filter vlans in promisc mode */
3017 ixgbe_vlan_filter_disable(adapter);
3019 if (netdev->flags & IFF_ALLMULTI) {
3020 fctrl |= IXGBE_FCTRL_MPE;
3021 fctrl &= ~IXGBE_FCTRL_UPE;
3022 } else if (!hw->addr_ctrl.uc_set_promisc) {
3023 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3025 ixgbe_vlan_filter_enable(adapter);
3026 hw->addr_ctrl.user_set_promisc = false;
3029 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3031 /* reprogram secondary unicast list */
3032 hw->mac.ops.update_uc_addr_list(hw, netdev);
3034 /* reprogram multicast list */
3035 hw->mac.ops.update_mc_addr_list(hw, netdev);
3037 if (adapter->num_vfs)
3038 ixgbe_restore_vf_multicasts(adapter);
3041 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3044 struct ixgbe_q_vector *q_vector;
3045 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3047 /* legacy and MSI only use one vector */
3048 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3051 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3052 struct napi_struct *napi;
3053 q_vector = adapter->q_vector[q_idx];
3054 napi = &q_vector->napi;
3055 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3056 if (!q_vector->rxr_count || !q_vector->txr_count) {
3057 if (q_vector->txr_count == 1)
3058 napi->poll = &ixgbe_clean_txonly;
3059 else if (q_vector->rxr_count == 1)
3060 napi->poll = &ixgbe_clean_rxonly;
3068 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3071 struct ixgbe_q_vector *q_vector;
3072 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3074 /* legacy and MSI only use one vector */
3075 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3078 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3079 q_vector = adapter->q_vector[q_idx];
3080 napi_disable(&q_vector->napi);
3084 #ifdef CONFIG_IXGBE_DCB
3086 * ixgbe_configure_dcb - Configure DCB hardware
3087 * @adapter: ixgbe adapter struct
3089 * This is called by the driver on open to configure the DCB hardware.
3090 * This is also called by the gennetlink interface when reconfiguring
3093 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3095 struct ixgbe_hw *hw = &adapter->hw;
3099 ixgbe_dcb_check_config(&adapter->dcb_cfg);
3100 ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_TX_CONFIG);
3101 ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_RX_CONFIG);
3103 /* reconfigure the hardware */
3104 ixgbe_dcb_hw_config(&adapter->hw, &adapter->dcb_cfg);
3106 for (i = 0; i < adapter->num_tx_queues; i++) {
3107 j = adapter->tx_ring[i]->reg_idx;
3108 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3109 /* PThresh workaround for Tx hang with DFP enabled. */
3111 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
3113 /* Enable VLAN tag insert/strip */
3114 ixgbe_vlan_filter_enable(adapter);
3116 hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
3120 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3122 struct net_device *netdev = adapter->netdev;
3123 struct ixgbe_hw *hw = &adapter->hw;
3126 ixgbe_set_rx_mode(netdev);
3128 ixgbe_restore_vlan(adapter);
3129 #ifdef CONFIG_IXGBE_DCB
3130 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3131 if (hw->mac.type == ixgbe_mac_82598EB)
3132 netif_set_gso_max_size(netdev, 32768);
3134 netif_set_gso_max_size(netdev, 65536);
3135 ixgbe_configure_dcb(adapter);
3137 netif_set_gso_max_size(netdev, 65536);
3140 netif_set_gso_max_size(netdev, 65536);
3144 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
3145 ixgbe_configure_fcoe(adapter);
3147 #endif /* IXGBE_FCOE */
3148 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3149 for (i = 0; i < adapter->num_tx_queues; i++)
3150 adapter->tx_ring[i]->atr_sample_rate =
3151 adapter->atr_sample_rate;
3152 ixgbe_init_fdir_signature_82599(hw, adapter->fdir_pballoc);
3153 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3154 ixgbe_init_fdir_perfect_82599(hw, adapter->fdir_pballoc);
3157 ixgbe_configure_tx(adapter);
3158 ixgbe_configure_rx(adapter);
3159 for (i = 0; i < adapter->num_rx_queues; i++)
3160 ixgbe_alloc_rx_buffers(adapter, adapter->rx_ring[i],
3161 (adapter->rx_ring[i]->count - 1));
3164 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3166 switch (hw->phy.type) {
3167 case ixgbe_phy_sfp_avago:
3168 case ixgbe_phy_sfp_ftl:
3169 case ixgbe_phy_sfp_intel:
3170 case ixgbe_phy_sfp_unknown:
3171 case ixgbe_phy_sfp_passive_tyco:
3172 case ixgbe_phy_sfp_passive_unknown:
3173 case ixgbe_phy_sfp_active_unknown:
3174 case ixgbe_phy_sfp_ftl_active:
3182 * ixgbe_sfp_link_config - set up SFP+ link
3183 * @adapter: pointer to private adapter struct
3185 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3187 struct ixgbe_hw *hw = &adapter->hw;
3189 if (hw->phy.multispeed_fiber) {
3191 * In multispeed fiber setups, the device may not have
3192 * had a physical connection when the driver loaded.
3193 * If that's the case, the initial link configuration
3194 * couldn't get the MAC into 10G or 1G mode, so we'll
3195 * never have a link status change interrupt fire.
3196 * We need to try and force an autonegotiation
3197 * session, then bring up link.
3199 hw->mac.ops.setup_sfp(hw);
3200 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
3201 schedule_work(&adapter->multispeed_fiber_task);
3204 * Direct Attach Cu and non-multispeed fiber modules
3205 * still need to be configured properly prior to
3208 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_MOD_TASK))
3209 schedule_work(&adapter->sfp_config_module_task);
3214 * ixgbe_non_sfp_link_config - set up non-SFP+ link
3215 * @hw: pointer to private hardware struct
3217 * Returns 0 on success, negative on failure
3219 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3222 bool negotiation, link_up = false;
3223 u32 ret = IXGBE_ERR_LINK_SETUP;
3225 if (hw->mac.ops.check_link)
3226 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3231 if (hw->mac.ops.get_link_capabilities)
3232 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
3236 if (hw->mac.ops.setup_link)
3237 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
3242 #define IXGBE_MAX_RX_DESC_POLL 10
3243 static inline void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
3246 int j = adapter->rx_ring[rxr]->reg_idx;
3249 for (k = 0; k < IXGBE_MAX_RX_DESC_POLL; k++) {
3250 if (IXGBE_READ_REG(&adapter->hw,
3251 IXGBE_RXDCTL(j)) & IXGBE_RXDCTL_ENABLE)
3256 if (k >= IXGBE_MAX_RX_DESC_POLL) {
3257 e_err("RXDCTL.ENABLE on Rx queue %d not set within "
3258 "the polling period\n", rxr);
3260 ixgbe_release_rx_desc(&adapter->hw, adapter->rx_ring[rxr],
3261 (adapter->rx_ring[rxr]->count - 1));
3264 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
3266 struct net_device *netdev = adapter->netdev;
3267 struct ixgbe_hw *hw = &adapter->hw;
3269 int num_rx_rings = adapter->num_rx_queues;
3271 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3272 u32 txdctl, rxdctl, mhadd;
3277 ixgbe_get_hw_control(adapter);
3279 if ((adapter->flags & IXGBE_FLAG_MSIX_ENABLED) ||
3280 (adapter->flags & IXGBE_FLAG_MSI_ENABLED)) {
3281 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3282 gpie = (IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_EIAME |
3283 IXGBE_GPIE_PBA_SUPPORT | IXGBE_GPIE_OCD);
3288 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3289 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
3290 gpie |= IXGBE_GPIE_VTMODE_64;
3292 /* XXX: to interrupt immediately for EICS writes, enable this */
3293 /* gpie |= IXGBE_GPIE_EIMEN; */
3294 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3297 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3299 * use EIAM to auto-mask when MSI-X interrupt is asserted
3300 * this saves a register write for every interrupt
3302 switch (hw->mac.type) {
3303 case ixgbe_mac_82598EB:
3304 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3307 case ixgbe_mac_82599EB:
3308 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
3309 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
3313 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
3314 * specifically only auto mask tx and rx interrupts */
3315 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3318 /* Enable Thermal over heat sensor interrupt */
3319 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
3320 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
3321 gpie |= IXGBE_SDP0_GPIEN;
3322 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3325 /* Enable fan failure interrupt if media type is copper */
3326 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3327 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
3328 gpie |= IXGBE_SDP1_GPIEN;
3329 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3332 if (hw->mac.type == ixgbe_mac_82599EB) {
3333 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
3334 gpie |= IXGBE_SDP1_GPIEN;
3335 gpie |= IXGBE_SDP2_GPIEN;
3336 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3340 /* adjust max frame to be able to do baby jumbo for FCoE */
3341 if ((netdev->features & NETIF_F_FCOE_MTU) &&
3342 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
3343 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3345 #endif /* IXGBE_FCOE */
3346 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
3347 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
3348 mhadd &= ~IXGBE_MHADD_MFS_MASK;
3349 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
3351 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
3354 for (i = 0; i < adapter->num_tx_queues; i++) {
3355 j = adapter->tx_ring[i]->reg_idx;
3356 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3357 if (adapter->rx_itr_setting == 0) {
3358 /* cannot set wthresh when itr==0 */
3359 txdctl &= ~0x007F0000;
3361 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
3362 txdctl |= (8 << 16);
3364 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
3367 if (hw->mac.type == ixgbe_mac_82599EB) {
3368 /* DMATXCTL.EN must be set after all Tx queue config is done */
3369 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
3370 dmatxctl |= IXGBE_DMATXCTL_TE;
3371 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
3373 for (i = 0; i < adapter->num_tx_queues; i++) {
3374 j = adapter->tx_ring[i]->reg_idx;
3375 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3376 txdctl |= IXGBE_TXDCTL_ENABLE;
3377 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
3378 if (hw->mac.type == ixgbe_mac_82599EB) {
3380 /* poll for Tx Enable ready */
3383 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3384 } while (--wait_loop &&
3385 !(txdctl & IXGBE_TXDCTL_ENABLE));
3387 e_err("Could not enable Tx Queue %d\n", j);
3391 for (i = 0; i < num_rx_rings; i++) {
3392 j = adapter->rx_ring[i]->reg_idx;
3393 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3394 /* enable PTHRESH=32 descriptors (half the internal cache)
3395 * and HTHRESH=0 descriptors (to minimize latency on fetch),
3396 * this also removes a pesky rx_no_buffer_count increment */
3398 rxdctl |= IXGBE_RXDCTL_ENABLE;
3399 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), rxdctl);
3400 if (hw->mac.type == ixgbe_mac_82599EB)
3401 ixgbe_rx_desc_queue_enable(adapter, i);
3403 /* enable all receives */
3404 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3405 if (hw->mac.type == ixgbe_mac_82598EB)
3406 rxdctl |= (IXGBE_RXCTRL_DMBYPS | IXGBE_RXCTRL_RXEN);
3408 rxdctl |= IXGBE_RXCTRL_RXEN;
3409 hw->mac.ops.enable_rx_dma(hw, rxdctl);
3411 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3412 ixgbe_configure_msix(adapter);
3414 ixgbe_configure_msi_and_legacy(adapter);
3416 /* enable the optics */
3417 if (hw->phy.multispeed_fiber)
3418 hw->mac.ops.enable_tx_laser(hw);
3420 clear_bit(__IXGBE_DOWN, &adapter->state);
3421 ixgbe_napi_enable_all(adapter);
3423 /* clear any pending interrupts, may auto mask */
3424 IXGBE_READ_REG(hw, IXGBE_EICR);
3426 ixgbe_irq_enable(adapter);
3429 * If this adapter has a fan, check to see if we had a failure
3430 * before we enabled the interrupt.
3432 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3433 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
3434 if (esdp & IXGBE_ESDP_SDP1)
3435 e_crit("Fan has stopped, replace the adapter\n");
3439 * For hot-pluggable SFP+ devices, a new SFP+ module may have
3440 * arrived before interrupts were enabled but after probe. Such
3441 * devices wouldn't have their type identified yet. We need to
3442 * kick off the SFP+ module setup first, then try to bring up link.
3443 * If we're not hot-pluggable SFP+, we just need to configure link
3446 if (hw->phy.type == ixgbe_phy_unknown) {
3447 err = hw->phy.ops.identify(hw);
3448 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
3450 * Take the device down and schedule the sfp tasklet
3451 * which will unregister_netdev and log it.
3453 ixgbe_down(adapter);
3454 schedule_work(&adapter->sfp_config_module_task);
3459 if (ixgbe_is_sfp(hw)) {
3460 ixgbe_sfp_link_config(adapter);
3462 err = ixgbe_non_sfp_link_config(hw);
3464 e_err("link_config FAILED %d\n", err);
3467 for (i = 0; i < adapter->num_tx_queues; i++)
3468 set_bit(__IXGBE_FDIR_INIT_DONE,
3469 &(adapter->tx_ring[i]->reinit_state));
3471 /* enable transmits */
3472 netif_tx_start_all_queues(netdev);
3474 /* bring the link up in the watchdog, this could race with our first
3475 * link up interrupt but shouldn't be a problem */
3476 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3477 adapter->link_check_timeout = jiffies;
3478 mod_timer(&adapter->watchdog_timer, jiffies);
3480 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
3481 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
3482 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
3483 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
3488 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
3490 WARN_ON(in_interrupt());
3491 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
3493 ixgbe_down(adapter);
3495 * If SR-IOV enabled then wait a bit before bringing the adapter
3496 * back up to give the VFs time to respond to the reset. The
3497 * two second wait is based upon the watchdog timer cycle in
3500 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3503 clear_bit(__IXGBE_RESETTING, &adapter->state);
3506 int ixgbe_up(struct ixgbe_adapter *adapter)
3508 /* hardware has been reset, we need to reload some things */
3509 ixgbe_configure(adapter);
3511 return ixgbe_up_complete(adapter);
3514 void ixgbe_reset(struct ixgbe_adapter *adapter)
3516 struct ixgbe_hw *hw = &adapter->hw;
3519 err = hw->mac.ops.init_hw(hw);
3522 case IXGBE_ERR_SFP_NOT_PRESENT:
3524 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
3525 e_dev_err("master disable timed out\n");
3527 case IXGBE_ERR_EEPROM_VERSION:
3528 /* We are running on a pre-production device, log a warning */
3529 e_dev_warn("This device is a pre-production adapter/LOM. "
3530 "Please be aware there may be issuesassociated with "
3531 "your hardware. If you are experiencing problems "
3532 "please contact your Intel or hardware "
3533 "representative who provided you with this "
3537 e_dev_err("Hardware Error: %d\n", err);
3540 /* reprogram the RAR[0] in case user changed it. */
3541 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
3546 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
3547 * @adapter: board private structure
3548 * @rx_ring: ring to free buffers from
3550 static void ixgbe_clean_rx_ring(struct ixgbe_adapter *adapter,
3551 struct ixgbe_ring *rx_ring)
3553 struct pci_dev *pdev = adapter->pdev;
3557 /* Free all the Rx ring sk_buffs */
3559 for (i = 0; i < rx_ring->count; i++) {
3560 struct ixgbe_rx_buffer *rx_buffer_info;
3562 rx_buffer_info = &rx_ring->rx_buffer_info[i];
3563 if (rx_buffer_info->dma) {
3564 dma_unmap_single(&pdev->dev, rx_buffer_info->dma,
3565 rx_ring->rx_buf_len,
3567 rx_buffer_info->dma = 0;
3569 if (rx_buffer_info->skb) {
3570 struct sk_buff *skb = rx_buffer_info->skb;
3571 rx_buffer_info->skb = NULL;
3573 struct sk_buff *this = skb;
3574 if (IXGBE_RSC_CB(this)->delay_unmap) {
3575 dma_unmap_single(&pdev->dev,
3576 IXGBE_RSC_CB(this)->dma,
3577 rx_ring->rx_buf_len,
3579 IXGBE_RSC_CB(this)->dma = 0;
3580 IXGBE_RSC_CB(skb)->delay_unmap = false;
3583 dev_kfree_skb(this);
3586 if (!rx_buffer_info->page)
3588 if (rx_buffer_info->page_dma) {
3589 dma_unmap_page(&pdev->dev, rx_buffer_info->page_dma,
3590 PAGE_SIZE / 2, DMA_FROM_DEVICE);
3591 rx_buffer_info->page_dma = 0;
3593 put_page(rx_buffer_info->page);
3594 rx_buffer_info->page = NULL;
3595 rx_buffer_info->page_offset = 0;
3598 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
3599 memset(rx_ring->rx_buffer_info, 0, size);
3601 /* Zero out the descriptor ring */
3602 memset(rx_ring->desc, 0, rx_ring->size);
3604 rx_ring->next_to_clean = 0;
3605 rx_ring->next_to_use = 0;
3608 writel(0, adapter->hw.hw_addr + rx_ring->head);
3610 writel(0, adapter->hw.hw_addr + rx_ring->tail);
3614 * ixgbe_clean_tx_ring - Free Tx Buffers
3615 * @adapter: board private structure
3616 * @tx_ring: ring to be cleaned
3618 static void ixgbe_clean_tx_ring(struct ixgbe_adapter *adapter,
3619 struct ixgbe_ring *tx_ring)
3621 struct ixgbe_tx_buffer *tx_buffer_info;
3625 /* Free all the Tx ring sk_buffs */
3627 for (i = 0; i < tx_ring->count; i++) {
3628 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3629 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
3632 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
3633 memset(tx_ring->tx_buffer_info, 0, size);
3635 /* Zero out the descriptor ring */
3636 memset(tx_ring->desc, 0, tx_ring->size);
3638 tx_ring->next_to_use = 0;
3639 tx_ring->next_to_clean = 0;
3642 writel(0, adapter->hw.hw_addr + tx_ring->head);
3644 writel(0, adapter->hw.hw_addr + tx_ring->tail);
3648 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
3649 * @adapter: board private structure
3651 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
3655 for (i = 0; i < adapter->num_rx_queues; i++)
3656 ixgbe_clean_rx_ring(adapter, adapter->rx_ring[i]);
3660 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
3661 * @adapter: board private structure
3663 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
3667 for (i = 0; i < adapter->num_tx_queues; i++)
3668 ixgbe_clean_tx_ring(adapter, adapter->tx_ring[i]);
3671 void ixgbe_down(struct ixgbe_adapter *adapter)
3673 struct net_device *netdev = adapter->netdev;
3674 struct ixgbe_hw *hw = &adapter->hw;
3679 /* signal that we are down to the interrupt handler */
3680 set_bit(__IXGBE_DOWN, &adapter->state);
3682 /* power down the optics */
3683 if (hw->phy.multispeed_fiber)
3684 hw->mac.ops.disable_tx_laser(hw);
3686 /* disable receive for all VFs and wait one second */
3687 if (adapter->num_vfs) {
3688 /* ping all the active vfs to let them know we are going down */
3689 ixgbe_ping_all_vfs(adapter);
3691 /* Disable all VFTE/VFRE TX/RX */
3692 ixgbe_disable_tx_rx(adapter);
3694 /* Mark all the VFs as inactive */
3695 for (i = 0 ; i < adapter->num_vfs; i++)
3696 adapter->vfinfo[i].clear_to_send = 0;
3699 /* disable receives */
3700 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3701 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3703 IXGBE_WRITE_FLUSH(hw);
3706 netif_tx_stop_all_queues(netdev);
3708 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
3709 del_timer_sync(&adapter->sfp_timer);
3710 del_timer_sync(&adapter->watchdog_timer);
3711 cancel_work_sync(&adapter->watchdog_task);
3713 netif_carrier_off(netdev);
3714 netif_tx_disable(netdev);
3716 ixgbe_irq_disable(adapter);
3718 ixgbe_napi_disable_all(adapter);
3720 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3721 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3722 cancel_work_sync(&adapter->fdir_reinit_task);
3724 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
3725 cancel_work_sync(&adapter->check_overtemp_task);
3727 /* disable transmits in the hardware now that interrupts are off */
3728 for (i = 0; i < adapter->num_tx_queues; i++) {
3729 j = adapter->tx_ring[i]->reg_idx;
3730 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3731 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j),
3732 (txdctl & ~IXGBE_TXDCTL_ENABLE));
3734 /* Disable the Tx DMA engine on 82599 */
3735 if (hw->mac.type == ixgbe_mac_82599EB)
3736 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
3737 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
3738 ~IXGBE_DMATXCTL_TE));
3740 /* clear n-tuple filters that are cached */
3741 ethtool_ntuple_flush(netdev);
3743 if (!pci_channel_offline(adapter->pdev))
3744 ixgbe_reset(adapter);
3745 ixgbe_clean_all_tx_rings(adapter);
3746 ixgbe_clean_all_rx_rings(adapter);
3748 #ifdef CONFIG_IXGBE_DCA
3749 /* since we reset the hardware DCA settings were cleared */
3750 ixgbe_setup_dca(adapter);
3755 * ixgbe_poll - NAPI Rx polling callback
3756 * @napi: structure for representing this polling device
3757 * @budget: how many packets driver is allowed to clean
3759 * This function is used for legacy and MSI, NAPI mode
3761 static int ixgbe_poll(struct napi_struct *napi, int budget)
3763 struct ixgbe_q_vector *q_vector =
3764 container_of(napi, struct ixgbe_q_vector, napi);
3765 struct ixgbe_adapter *adapter = q_vector->adapter;
3766 int tx_clean_complete, work_done = 0;
3768 #ifdef CONFIG_IXGBE_DCA
3769 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
3770 ixgbe_update_tx_dca(adapter, adapter->tx_ring[0]);
3771 ixgbe_update_rx_dca(adapter, adapter->rx_ring[0]);
3775 tx_clean_complete = ixgbe_clean_tx_irq(q_vector, adapter->tx_ring[0]);
3776 ixgbe_clean_rx_irq(q_vector, adapter->rx_ring[0], &work_done, budget);
3778 if (!tx_clean_complete)
3781 /* If budget not fully consumed, exit the polling mode */
3782 if (work_done < budget) {
3783 napi_complete(napi);
3784 if (adapter->rx_itr_setting & 1)
3785 ixgbe_set_itr(adapter);
3786 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3787 ixgbe_irq_enable_queues(adapter, IXGBE_EIMS_RTX_QUEUE);
3793 * ixgbe_tx_timeout - Respond to a Tx Hang
3794 * @netdev: network interface device structure
3796 static void ixgbe_tx_timeout(struct net_device *netdev)
3798 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3800 /* Do the reset outside of interrupt context */
3801 schedule_work(&adapter->reset_task);
3804 static void ixgbe_reset_task(struct work_struct *work)
3806 struct ixgbe_adapter *adapter;
3807 adapter = container_of(work, struct ixgbe_adapter, reset_task);
3809 /* If we're already down or resetting, just bail */
3810 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
3811 test_bit(__IXGBE_RESETTING, &adapter->state))
3814 adapter->tx_timeout_count++;
3816 ixgbe_dump(adapter);
3817 netdev_err(adapter->netdev, "Reset adapter\n");
3818 ixgbe_reinit_locked(adapter);
3821 #ifdef CONFIG_IXGBE_DCB
3822 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
3825 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_DCB];
3827 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
3831 adapter->num_rx_queues = f->indices;
3832 adapter->num_tx_queues = f->indices;
3840 * ixgbe_set_rss_queues: Allocate queues for RSS
3841 * @adapter: board private structure to initialize
3843 * This is our "base" multiqueue mode. RSS (Receive Side Scaling) will try
3844 * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
3847 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
3850 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
3852 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3854 adapter->num_rx_queues = f->indices;
3855 adapter->num_tx_queues = f->indices;
3865 * ixgbe_set_fdir_queues: Allocate queues for Flow Director
3866 * @adapter: board private structure to initialize
3868 * Flow Director is an advanced Rx filter, attempting to get Rx flows back
3869 * to the original CPU that initiated the Tx session. This runs in addition
3870 * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
3871 * Rx load across CPUs using RSS.
3874 static bool inline ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
3877 struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
3879 f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
3882 /* Flow Director must have RSS enabled */
3883 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
3884 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3885 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)))) {
3886 adapter->num_tx_queues = f_fdir->indices;
3887 adapter->num_rx_queues = f_fdir->indices;
3890 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
3891 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
3898 * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
3899 * @adapter: board private structure to initialize
3901 * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
3902 * The ring feature mask is not used as a mask for FCoE, as it can take any 8
3903 * rx queues out of the max number of rx queues, instead, it is used as the
3904 * index of the first rx queue used by FCoE.
3907 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
3910 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
3912 f->indices = min((int)num_online_cpus(), f->indices);
3913 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
3914 adapter->num_rx_queues = 1;
3915 adapter->num_tx_queues = 1;
3916 #ifdef CONFIG_IXGBE_DCB
3917 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3918 e_info("FCoE enabled with DCB\n");
3919 ixgbe_set_dcb_queues(adapter);
3922 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3923 e_info("FCoE enabled with RSS\n");
3924 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
3925 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
3926 ixgbe_set_fdir_queues(adapter);
3928 ixgbe_set_rss_queues(adapter);
3930 /* adding FCoE rx rings to the end */
3931 f->mask = adapter->num_rx_queues;
3932 adapter->num_rx_queues += f->indices;
3933 adapter->num_tx_queues += f->indices;
3941 #endif /* IXGBE_FCOE */
3943 * ixgbe_set_sriov_queues: Allocate queues for IOV use
3944 * @adapter: board private structure to initialize
3946 * IOV doesn't actually use anything, so just NAK the
3947 * request for now and let the other queue routines
3948 * figure out what to do.
3950 static inline bool ixgbe_set_sriov_queues(struct ixgbe_adapter *adapter)
3956 * ixgbe_set_num_queues: Allocate queues for device, feature dependant
3957 * @adapter: board private structure to initialize
3959 * This is the top level queue allocation routine. The order here is very
3960 * important, starting with the "most" number of features turned on at once,
3961 * and ending with the smallest set of features. This way large combinations
3962 * can be allocated if they're turned on, and smaller combinations are the
3963 * fallthrough conditions.
3966 static void ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
3968 /* Start with base case */
3969 adapter->num_rx_queues = 1;
3970 adapter->num_tx_queues = 1;
3971 adapter->num_rx_pools = adapter->num_rx_queues;
3972 adapter->num_rx_queues_per_pool = 1;
3974 if (ixgbe_set_sriov_queues(adapter))
3978 if (ixgbe_set_fcoe_queues(adapter))
3981 #endif /* IXGBE_FCOE */
3982 #ifdef CONFIG_IXGBE_DCB
3983 if (ixgbe_set_dcb_queues(adapter))
3987 if (ixgbe_set_fdir_queues(adapter))
3990 if (ixgbe_set_rss_queues(adapter))
3993 /* fallback to base case */
3994 adapter->num_rx_queues = 1;
3995 adapter->num_tx_queues = 1;
3998 /* Notify the stack of the (possibly) reduced Tx Queue count. */
3999 adapter->netdev->real_num_tx_queues = adapter->num_tx_queues;
4002 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
4005 int err, vector_threshold;
4007 /* We'll want at least 3 (vector_threshold):
4010 * 3) Other (Link Status Change, etc.)
4011 * 4) TCP Timer (optional)
4013 vector_threshold = MIN_MSIX_COUNT;
4015 /* The more we get, the more we will assign to Tx/Rx Cleanup
4016 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
4017 * Right now, we simply care about how many we'll get; we'll
4018 * set them up later while requesting irq's.
4020 while (vectors >= vector_threshold) {
4021 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
4023 if (!err) /* Success in acquiring all requested vectors. */
4026 vectors = 0; /* Nasty failure, quit now */
4027 else /* err == number of vectors we should try again with */
4031 if (vectors < vector_threshold) {
4032 /* Can't allocate enough MSI-X interrupts? Oh well.
4033 * This just means we'll go with either a single MSI
4034 * vector or fall back to legacy interrupts.
4036 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4037 "Unable to allocate MSI-X interrupts\n");
4038 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4039 kfree(adapter->msix_entries);
4040 adapter->msix_entries = NULL;
4042 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
4044 * Adjust for only the vectors we'll use, which is minimum
4045 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
4046 * vectors we were allocated.
4048 adapter->num_msix_vectors = min(vectors,
4049 adapter->max_msix_q_vectors + NON_Q_VECTORS);
4054 * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
4055 * @adapter: board private structure to initialize
4057 * Cache the descriptor ring offsets for RSS to the assigned rings.
4060 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
4065 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4066 for (i = 0; i < adapter->num_rx_queues; i++)
4067 adapter->rx_ring[i]->reg_idx = i;
4068 for (i = 0; i < adapter->num_tx_queues; i++)
4069 adapter->tx_ring[i]->reg_idx = i;
4078 #ifdef CONFIG_IXGBE_DCB
4080 * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
4081 * @adapter: board private structure to initialize
4083 * Cache the descriptor ring offsets for DCB to the assigned rings.
4086 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
4090 int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
4092 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4093 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
4094 /* the number of queues is assumed to be symmetric */
4095 for (i = 0; i < dcb_i; i++) {
4096 adapter->rx_ring[i]->reg_idx = i << 3;
4097 adapter->tx_ring[i]->reg_idx = i << 2;
4100 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
4103 * Tx TC0 starts at: descriptor queue 0
4104 * Tx TC1 starts at: descriptor queue 32
4105 * Tx TC2 starts at: descriptor queue 64
4106 * Tx TC3 starts at: descriptor queue 80
4107 * Tx TC4 starts at: descriptor queue 96
4108 * Tx TC5 starts at: descriptor queue 104
4109 * Tx TC6 starts at: descriptor queue 112
4110 * Tx TC7 starts at: descriptor queue 120
4112 * Rx TC0-TC7 are offset by 16 queues each
4114 for (i = 0; i < 3; i++) {
4115 adapter->tx_ring[i]->reg_idx = i << 5;
4116 adapter->rx_ring[i]->reg_idx = i << 4;
4118 for ( ; i < 5; i++) {
4119 adapter->tx_ring[i]->reg_idx =
4121 adapter->rx_ring[i]->reg_idx = i << 4;
4123 for ( ; i < dcb_i; i++) {
4124 adapter->tx_ring[i]->reg_idx =
4126 adapter->rx_ring[i]->reg_idx = i << 4;
4130 } else if (dcb_i == 4) {
4132 * Tx TC0 starts at: descriptor queue 0
4133 * Tx TC1 starts at: descriptor queue 64
4134 * Tx TC2 starts at: descriptor queue 96
4135 * Tx TC3 starts at: descriptor queue 112
4137 * Rx TC0-TC3 are offset by 32 queues each
4139 adapter->tx_ring[0]->reg_idx = 0;
4140 adapter->tx_ring[1]->reg_idx = 64;
4141 adapter->tx_ring[2]->reg_idx = 96;
4142 adapter->tx_ring[3]->reg_idx = 112;
4143 for (i = 0 ; i < dcb_i; i++)
4144 adapter->rx_ring[i]->reg_idx = i << 5;
4162 * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
4163 * @adapter: board private structure to initialize
4165 * Cache the descriptor ring offsets for Flow Director to the assigned rings.
4168 static bool inline ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
4173 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
4174 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4175 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))) {
4176 for (i = 0; i < adapter->num_rx_queues; i++)
4177 adapter->rx_ring[i]->reg_idx = i;
4178 for (i = 0; i < adapter->num_tx_queues; i++)
4179 adapter->tx_ring[i]->reg_idx = i;
4188 * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
4189 * @adapter: board private structure to initialize
4191 * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
4194 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
4196 int i, fcoe_rx_i = 0, fcoe_tx_i = 0;
4198 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4200 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
4201 #ifdef CONFIG_IXGBE_DCB
4202 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4203 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
4205 ixgbe_cache_ring_dcb(adapter);
4206 /* find out queues in TC for FCoE */
4207 fcoe_rx_i = adapter->rx_ring[fcoe->tc]->reg_idx + 1;
4208 fcoe_tx_i = adapter->tx_ring[fcoe->tc]->reg_idx + 1;
4210 * In 82599, the number of Tx queues for each traffic
4211 * class for both 8-TC and 4-TC modes are:
4212 * TCs : TC0 TC1 TC2 TC3 TC4 TC5 TC6 TC7
4213 * 8 TCs: 32 32 16 16 8 8 8 8
4214 * 4 TCs: 64 64 32 32
4215 * We have max 8 queues for FCoE, where 8 the is
4216 * FCoE redirection table size. If TC for FCoE is
4217 * less than or equal to TC3, we have enough queues
4218 * to add max of 8 queues for FCoE, so we start FCoE
4219 * tx descriptor from the next one, i.e., reg_idx + 1.
4220 * If TC for FCoE is above TC3, implying 8 TC mode,
4221 * and we need 8 for FCoE, we have to take all queues
4222 * in that traffic class for FCoE.
4224 if ((f->indices == IXGBE_FCRETA_SIZE) && (fcoe->tc > 3))
4227 #endif /* CONFIG_IXGBE_DCB */
4228 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4229 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4230 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
4231 ixgbe_cache_ring_fdir(adapter);
4233 ixgbe_cache_ring_rss(adapter);
4235 fcoe_rx_i = f->mask;
4236 fcoe_tx_i = f->mask;
4238 for (i = 0; i < f->indices; i++, fcoe_rx_i++, fcoe_tx_i++) {
4239 adapter->rx_ring[f->mask + i]->reg_idx = fcoe_rx_i;
4240 adapter->tx_ring[f->mask + i]->reg_idx = fcoe_tx_i;
4247 #endif /* IXGBE_FCOE */
4249 * ixgbe_cache_ring_sriov - Descriptor ring to register mapping for sriov
4250 * @adapter: board private structure to initialize
4252 * SR-IOV doesn't use any descriptor rings but changes the default if
4253 * no other mapping is used.
4256 static inline bool ixgbe_cache_ring_sriov(struct ixgbe_adapter *adapter)
4258 adapter->rx_ring[0]->reg_idx = adapter->num_vfs * 2;
4259 adapter->tx_ring[0]->reg_idx = adapter->num_vfs * 2;
4260 if (adapter->num_vfs)
4267 * ixgbe_cache_ring_register - Descriptor ring to register mapping
4268 * @adapter: board private structure to initialize
4270 * Once we know the feature-set enabled for the device, we'll cache
4271 * the register offset the descriptor ring is assigned to.
4273 * Note, the order the various feature calls is important. It must start with
4274 * the "most" features enabled at the same time, then trickle down to the
4275 * least amount of features turned on at once.
4277 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
4279 /* start with default case */
4280 adapter->rx_ring[0]->reg_idx = 0;
4281 adapter->tx_ring[0]->reg_idx = 0;
4283 if (ixgbe_cache_ring_sriov(adapter))
4287 if (ixgbe_cache_ring_fcoe(adapter))
4290 #endif /* IXGBE_FCOE */
4291 #ifdef CONFIG_IXGBE_DCB
4292 if (ixgbe_cache_ring_dcb(adapter))
4296 if (ixgbe_cache_ring_fdir(adapter))
4299 if (ixgbe_cache_ring_rss(adapter))
4304 * ixgbe_alloc_queues - Allocate memory for all rings
4305 * @adapter: board private structure to initialize
4307 * We allocate one ring per queue at run-time since we don't know the
4308 * number of queues at compile-time. The polling_netdev array is
4309 * intended for Multiqueue, but should work fine with a single queue.
4311 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
4314 int orig_node = adapter->node;
4316 for (i = 0; i < adapter->num_tx_queues; i++) {
4317 struct ixgbe_ring *ring = adapter->tx_ring[i];
4318 if (orig_node == -1) {
4319 int cur_node = next_online_node(adapter->node);
4320 if (cur_node == MAX_NUMNODES)
4321 cur_node = first_online_node;
4322 adapter->node = cur_node;
4324 ring = kzalloc_node(sizeof(struct ixgbe_ring), GFP_KERNEL,
4327 ring = kzalloc(sizeof(struct ixgbe_ring), GFP_KERNEL);
4329 goto err_tx_ring_allocation;
4330 ring->count = adapter->tx_ring_count;
4331 ring->queue_index = i;
4332 ring->numa_node = adapter->node;
4334 adapter->tx_ring[i] = ring;
4337 /* Restore the adapter's original node */
4338 adapter->node = orig_node;
4340 for (i = 0; i < adapter->num_rx_queues; i++) {
4341 struct ixgbe_ring *ring = adapter->rx_ring[i];
4342 if (orig_node == -1) {
4343 int cur_node = next_online_node(adapter->node);
4344 if (cur_node == MAX_NUMNODES)
4345 cur_node = first_online_node;
4346 adapter->node = cur_node;
4348 ring = kzalloc_node(sizeof(struct ixgbe_ring), GFP_KERNEL,
4351 ring = kzalloc(sizeof(struct ixgbe_ring), GFP_KERNEL);
4353 goto err_rx_ring_allocation;
4354 ring->count = adapter->rx_ring_count;
4355 ring->queue_index = i;
4356 ring->numa_node = adapter->node;
4358 adapter->rx_ring[i] = ring;
4361 /* Restore the adapter's original node */
4362 adapter->node = orig_node;
4364 ixgbe_cache_ring_register(adapter);
4368 err_rx_ring_allocation:
4369 for (i = 0; i < adapter->num_tx_queues; i++)
4370 kfree(adapter->tx_ring[i]);
4371 err_tx_ring_allocation:
4376 * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
4377 * @adapter: board private structure to initialize
4379 * Attempt to configure the interrupts using the best available
4380 * capabilities of the hardware and the kernel.
4382 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
4384 struct ixgbe_hw *hw = &adapter->hw;
4386 int vector, v_budget;
4389 * It's easy to be greedy for MSI-X vectors, but it really
4390 * doesn't do us much good if we have a lot more vectors
4391 * than CPU's. So let's be conservative and only ask for
4392 * (roughly) the same number of vectors as there are CPU's.
4394 v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
4395 (int)num_online_cpus()) + NON_Q_VECTORS;
4398 * At the same time, hardware can only support a maximum of
4399 * hw.mac->max_msix_vectors vectors. With features
4400 * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
4401 * descriptor queues supported by our device. Thus, we cap it off in
4402 * those rare cases where the cpu count also exceeds our vector limit.
4404 v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
4406 /* A failure in MSI-X entry allocation isn't fatal, but it does
4407 * mean we disable MSI-X capabilities of the adapter. */
4408 adapter->msix_entries = kcalloc(v_budget,
4409 sizeof(struct msix_entry), GFP_KERNEL);
4410 if (adapter->msix_entries) {
4411 for (vector = 0; vector < v_budget; vector++)
4412 adapter->msix_entries[vector].entry = vector;
4414 ixgbe_acquire_msix_vectors(adapter, v_budget);
4416 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4420 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
4421 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
4422 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4423 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4424 adapter->atr_sample_rate = 0;
4425 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4426 ixgbe_disable_sriov(adapter);
4428 ixgbe_set_num_queues(adapter);
4430 err = pci_enable_msi(adapter->pdev);
4432 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
4434 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4435 "Unable to allocate MSI interrupt, "
4436 "falling back to legacy. Error: %d\n", err);
4446 * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
4447 * @adapter: board private structure to initialize
4449 * We allocate one q_vector per queue interrupt. If allocation fails we
4452 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
4454 int q_idx, num_q_vectors;
4455 struct ixgbe_q_vector *q_vector;
4457 int (*poll)(struct napi_struct *, int);
4459 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4460 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4461 napi_vectors = adapter->num_rx_queues;
4462 poll = &ixgbe_clean_rxtx_many;
4469 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4470 q_vector = kzalloc_node(sizeof(struct ixgbe_q_vector),
4471 GFP_KERNEL, adapter->node);
4473 q_vector = kzalloc(sizeof(struct ixgbe_q_vector),
4477 q_vector->adapter = adapter;
4478 if (q_vector->txr_count && !q_vector->rxr_count)
4479 q_vector->eitr = adapter->tx_eitr_param;
4481 q_vector->eitr = adapter->rx_eitr_param;
4482 q_vector->v_idx = q_idx;
4483 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
4484 adapter->q_vector[q_idx] = q_vector;
4492 q_vector = adapter->q_vector[q_idx];
4493 netif_napi_del(&q_vector->napi);
4495 adapter->q_vector[q_idx] = NULL;
4501 * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
4502 * @adapter: board private structure to initialize
4504 * This function frees the memory allocated to the q_vectors. In addition if
4505 * NAPI is enabled it will delete any references to the NAPI struct prior
4506 * to freeing the q_vector.
4508 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
4510 int q_idx, num_q_vectors;
4512 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4513 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4517 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4518 struct ixgbe_q_vector *q_vector = adapter->q_vector[q_idx];
4519 adapter->q_vector[q_idx] = NULL;
4520 netif_napi_del(&q_vector->napi);
4525 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
4527 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4528 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4529 pci_disable_msix(adapter->pdev);
4530 kfree(adapter->msix_entries);
4531 adapter->msix_entries = NULL;
4532 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
4533 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
4534 pci_disable_msi(adapter->pdev);
4539 * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
4540 * @adapter: board private structure to initialize
4542 * We determine which interrupt scheme to use based on...
4543 * - Kernel support (MSI, MSI-X)
4544 * - which can be user-defined (via MODULE_PARAM)
4545 * - Hardware queue count (num_*_queues)
4546 * - defined by miscellaneous hardware support/features (RSS, etc.)
4548 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
4552 /* Number of supported queues */
4553 ixgbe_set_num_queues(adapter);
4555 err = ixgbe_set_interrupt_capability(adapter);
4557 e_dev_err("Unable to setup interrupt capabilities\n");
4558 goto err_set_interrupt;
4561 err = ixgbe_alloc_q_vectors(adapter);
4563 e_dev_err("Unable to allocate memory for queue vectors\n");
4564 goto err_alloc_q_vectors;
4567 err = ixgbe_alloc_queues(adapter);
4569 e_dev_err("Unable to allocate memory for queues\n");
4570 goto err_alloc_queues;
4573 e_dev_info("Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
4574 (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
4575 adapter->num_rx_queues, adapter->num_tx_queues);
4577 set_bit(__IXGBE_DOWN, &adapter->state);
4582 ixgbe_free_q_vectors(adapter);
4583 err_alloc_q_vectors:
4584 ixgbe_reset_interrupt_capability(adapter);
4590 * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
4591 * @adapter: board private structure to clear interrupt scheme on
4593 * We go through and clear interrupt specific resources and reset the structure
4594 * to pre-load conditions
4596 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
4600 for (i = 0; i < adapter->num_tx_queues; i++) {
4601 kfree(adapter->tx_ring[i]);
4602 adapter->tx_ring[i] = NULL;
4604 for (i = 0; i < adapter->num_rx_queues; i++) {
4605 kfree(adapter->rx_ring[i]);
4606 adapter->rx_ring[i] = NULL;
4609 ixgbe_free_q_vectors(adapter);
4610 ixgbe_reset_interrupt_capability(adapter);
4614 * ixgbe_sfp_timer - worker thread to find a missing module
4615 * @data: pointer to our adapter struct
4617 static void ixgbe_sfp_timer(unsigned long data)
4619 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
4622 * Do the sfp_timer outside of interrupt context due to the
4623 * delays that sfp+ detection requires
4625 schedule_work(&adapter->sfp_task);
4629 * ixgbe_sfp_task - worker thread to find a missing module
4630 * @work: pointer to work_struct containing our data
4632 static void ixgbe_sfp_task(struct work_struct *work)
4634 struct ixgbe_adapter *adapter = container_of(work,
4635 struct ixgbe_adapter,
4637 struct ixgbe_hw *hw = &adapter->hw;
4639 if ((hw->phy.type == ixgbe_phy_nl) &&
4640 (hw->phy.sfp_type == ixgbe_sfp_type_not_present)) {
4641 s32 ret = hw->phy.ops.identify_sfp(hw);
4642 if (ret == IXGBE_ERR_SFP_NOT_PRESENT)
4644 ret = hw->phy.ops.reset(hw);
4645 if (ret == IXGBE_ERR_SFP_NOT_SUPPORTED) {
4646 e_dev_err("failed to initialize because an unsupported "
4647 "SFP+ module type was detected.\n");
4648 e_dev_err("Reload the driver after installing a "
4649 "supported module.\n");
4650 unregister_netdev(adapter->netdev);
4652 e_info("detected SFP+: %d\n", hw->phy.sfp_type);
4654 /* don't need this routine any more */
4655 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
4659 if (test_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state))
4660 mod_timer(&adapter->sfp_timer,
4661 round_jiffies(jiffies + (2 * HZ)));
4665 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4666 * @adapter: board private structure to initialize
4668 * ixgbe_sw_init initializes the Adapter private data structure.
4669 * Fields are initialized based on PCI device information and
4670 * OS network device settings (MTU size).
4672 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
4674 struct ixgbe_hw *hw = &adapter->hw;
4675 struct pci_dev *pdev = adapter->pdev;
4676 struct net_device *dev = adapter->netdev;
4678 #ifdef CONFIG_IXGBE_DCB
4680 struct tc_configuration *tc;
4683 /* PCI config space info */
4685 hw->vendor_id = pdev->vendor;
4686 hw->device_id = pdev->device;
4687 hw->revision_id = pdev->revision;
4688 hw->subsystem_vendor_id = pdev->subsystem_vendor;
4689 hw->subsystem_device_id = pdev->subsystem_device;
4691 /* Set capability flags */
4692 rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
4693 adapter->ring_feature[RING_F_RSS].indices = rss;
4694 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
4695 adapter->ring_feature[RING_F_DCB].indices = IXGBE_MAX_DCB_INDICES;
4696 if (hw->mac.type == ixgbe_mac_82598EB) {
4697 if (hw->device_id == IXGBE_DEV_ID_82598AT)
4698 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4699 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
4700 } else if (hw->mac.type == ixgbe_mac_82599EB) {
4701 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
4702 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4703 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4704 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
4705 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4706 if (dev->features & NETIF_F_NTUPLE) {
4707 /* Flow Director perfect filter enabled */
4708 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4709 adapter->atr_sample_rate = 0;
4710 spin_lock_init(&adapter->fdir_perfect_lock);
4712 /* Flow Director hash filters enabled */
4713 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
4714 adapter->atr_sample_rate = 20;
4716 adapter->ring_feature[RING_F_FDIR].indices =
4717 IXGBE_MAX_FDIR_INDICES;
4718 adapter->fdir_pballoc = 0;
4720 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4721 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4722 adapter->ring_feature[RING_F_FCOE].indices = 0;
4723 #ifdef CONFIG_IXGBE_DCB
4724 /* Default traffic class to use for FCoE */
4725 adapter->fcoe.tc = IXGBE_FCOE_DEFTC;
4727 #endif /* IXGBE_FCOE */
4730 #ifdef CONFIG_IXGBE_DCB
4731 /* Configure DCB traffic classes */
4732 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
4733 tc = &adapter->dcb_cfg.tc_config[j];
4734 tc->path[DCB_TX_CONFIG].bwg_id = 0;
4735 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
4736 tc->path[DCB_RX_CONFIG].bwg_id = 0;
4737 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
4738 tc->dcb_pfc = pfc_disabled;
4740 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
4741 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
4742 adapter->dcb_cfg.rx_pba_cfg = pba_equal;
4743 adapter->dcb_cfg.pfc_mode_enable = false;
4744 adapter->dcb_cfg.round_robin_enable = false;
4745 adapter->dcb_set_bitmap = 0x00;
4746 ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
4747 adapter->ring_feature[RING_F_DCB].indices);
4751 /* default flow control settings */
4752 hw->fc.requested_mode = ixgbe_fc_full;
4753 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
4755 adapter->last_lfc_mode = hw->fc.current_mode;
4757 hw->fc.high_water = IXGBE_DEFAULT_FCRTH;
4758 hw->fc.low_water = IXGBE_DEFAULT_FCRTL;
4759 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
4760 hw->fc.send_xon = true;
4761 hw->fc.disable_fc_autoneg = false;
4763 /* enable itr by default in dynamic mode */
4764 adapter->rx_itr_setting = 1;
4765 adapter->rx_eitr_param = 20000;
4766 adapter->tx_itr_setting = 1;
4767 adapter->tx_eitr_param = 10000;
4769 /* set defaults for eitr in MegaBytes */
4770 adapter->eitr_low = 10;
4771 adapter->eitr_high = 20;
4773 /* set default ring sizes */
4774 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
4775 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
4777 /* initialize eeprom parameters */
4778 if (ixgbe_init_eeprom_params_generic(hw)) {
4779 e_dev_err("EEPROM initialization failed\n");
4783 /* enable rx csum by default */
4784 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
4786 /* get assigned NUMA node */
4787 adapter->node = dev_to_node(&pdev->dev);
4789 set_bit(__IXGBE_DOWN, &adapter->state);
4795 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
4796 * @adapter: board private structure
4797 * @tx_ring: tx descriptor ring (for a specific queue) to setup
4799 * Return 0 on success, negative on failure
4801 int ixgbe_setup_tx_resources(struct ixgbe_adapter *adapter,
4802 struct ixgbe_ring *tx_ring)
4804 struct pci_dev *pdev = adapter->pdev;
4807 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4808 tx_ring->tx_buffer_info = vmalloc_node(size, tx_ring->numa_node);
4809 if (!tx_ring->tx_buffer_info)
4810 tx_ring->tx_buffer_info = vmalloc(size);
4811 if (!tx_ring->tx_buffer_info)
4813 memset(tx_ring->tx_buffer_info, 0, size);
4815 /* round up to nearest 4K */
4816 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
4817 tx_ring->size = ALIGN(tx_ring->size, 4096);
4819 tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
4820 &tx_ring->dma, GFP_KERNEL);
4824 tx_ring->next_to_use = 0;
4825 tx_ring->next_to_clean = 0;
4826 tx_ring->work_limit = tx_ring->count;
4830 vfree(tx_ring->tx_buffer_info);
4831 tx_ring->tx_buffer_info = NULL;
4832 e_err("Unable to allocate memory for the Tx descriptor ring\n");
4837 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
4838 * @adapter: board private structure
4840 * If this function returns with an error, then it's possible one or
4841 * more of the rings is populated (while the rest are not). It is the
4842 * callers duty to clean those orphaned rings.
4844 * Return 0 on success, negative on failure
4846 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
4850 for (i = 0; i < adapter->num_tx_queues; i++) {
4851 err = ixgbe_setup_tx_resources(adapter, adapter->tx_ring[i]);
4854 e_err("Allocation for Tx Queue %u failed\n", i);
4862 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
4863 * @adapter: board private structure
4864 * @rx_ring: rx descriptor ring (for a specific queue) to setup
4866 * Returns 0 on success, negative on failure
4868 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
4869 struct ixgbe_ring *rx_ring)
4871 struct pci_dev *pdev = adapter->pdev;
4874 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4875 rx_ring->rx_buffer_info = vmalloc_node(size, adapter->node);
4876 if (!rx_ring->rx_buffer_info)
4877 rx_ring->rx_buffer_info = vmalloc(size);
4878 if (!rx_ring->rx_buffer_info) {
4879 e_err("vmalloc allocation failed for the Rx desc ring\n");
4882 memset(rx_ring->rx_buffer_info, 0, size);
4884 /* Round up to nearest 4K */
4885 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
4886 rx_ring->size = ALIGN(rx_ring->size, 4096);
4888 rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
4889 &rx_ring->dma, GFP_KERNEL);
4891 if (!rx_ring->desc) {
4892 e_err("Memory allocation failed for the Rx desc ring\n");
4893 vfree(rx_ring->rx_buffer_info);
4897 rx_ring->next_to_clean = 0;
4898 rx_ring->next_to_use = 0;
4907 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
4908 * @adapter: board private structure
4910 * If this function returns with an error, then it's possible one or
4911 * more of the rings is populated (while the rest are not). It is the
4912 * callers duty to clean those orphaned rings.
4914 * Return 0 on success, negative on failure
4917 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
4921 for (i = 0; i < adapter->num_rx_queues; i++) {
4922 err = ixgbe_setup_rx_resources(adapter, adapter->rx_ring[i]);
4925 e_err("Allocation for Rx Queue %u failed\n", i);
4933 * ixgbe_free_tx_resources - Free Tx Resources per Queue
4934 * @adapter: board private structure
4935 * @tx_ring: Tx descriptor ring for a specific queue
4937 * Free all transmit software resources
4939 void ixgbe_free_tx_resources(struct ixgbe_adapter *adapter,
4940 struct ixgbe_ring *tx_ring)
4942 struct pci_dev *pdev = adapter->pdev;
4944 ixgbe_clean_tx_ring(adapter, tx_ring);
4946 vfree(tx_ring->tx_buffer_info);
4947 tx_ring->tx_buffer_info = NULL;
4949 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
4952 tx_ring->desc = NULL;
4956 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
4957 * @adapter: board private structure
4959 * Free all transmit software resources
4961 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
4965 for (i = 0; i < adapter->num_tx_queues; i++)
4966 if (adapter->tx_ring[i]->desc)
4967 ixgbe_free_tx_resources(adapter, adapter->tx_ring[i]);
4971 * ixgbe_free_rx_resources - Free Rx Resources
4972 * @adapter: board private structure
4973 * @rx_ring: ring to clean the resources from
4975 * Free all receive software resources
4977 void ixgbe_free_rx_resources(struct ixgbe_adapter *adapter,
4978 struct ixgbe_ring *rx_ring)
4980 struct pci_dev *pdev = adapter->pdev;
4982 ixgbe_clean_rx_ring(adapter, rx_ring);
4984 vfree(rx_ring->rx_buffer_info);
4985 rx_ring->rx_buffer_info = NULL;
4987 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
4990 rx_ring->desc = NULL;
4994 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
4995 * @adapter: board private structure
4997 * Free all receive software resources
4999 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5003 for (i = 0; i < adapter->num_rx_queues; i++)
5004 if (adapter->rx_ring[i]->desc)
5005 ixgbe_free_rx_resources(adapter, adapter->rx_ring[i]);
5009 * ixgbe_change_mtu - Change the Maximum Transfer Unit
5010 * @netdev: network interface device structure
5011 * @new_mtu: new value for maximum frame size
5013 * Returns 0 on success, negative on failure
5015 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5017 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5018 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5020 /* MTU < 68 is an error and causes problems on some kernels */
5021 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5024 e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5025 /* must set new MTU before calling down or up */
5026 netdev->mtu = new_mtu;
5028 if (netif_running(netdev))
5029 ixgbe_reinit_locked(adapter);
5035 * ixgbe_open - Called when a network interface is made active
5036 * @netdev: network interface device structure
5038 * Returns 0 on success, negative value on failure
5040 * The open entry point is called when a network interface is made
5041 * active by the system (IFF_UP). At this point all resources needed
5042 * for transmit and receive operations are allocated, the interrupt
5043 * handler is registered with the OS, the watchdog timer is started,
5044 * and the stack is notified that the interface is ready.
5046 static int ixgbe_open(struct net_device *netdev)
5048 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5051 /* disallow open during test */
5052 if (test_bit(__IXGBE_TESTING, &adapter->state))
5055 netif_carrier_off(netdev);
5057 /* allocate transmit descriptors */
5058 err = ixgbe_setup_all_tx_resources(adapter);
5062 /* allocate receive descriptors */
5063 err = ixgbe_setup_all_rx_resources(adapter);
5067 ixgbe_configure(adapter);
5069 err = ixgbe_request_irq(adapter);
5073 err = ixgbe_up_complete(adapter);
5077 netif_tx_start_all_queues(netdev);
5082 ixgbe_release_hw_control(adapter);
5083 ixgbe_free_irq(adapter);
5086 ixgbe_free_all_rx_resources(adapter);
5088 ixgbe_free_all_tx_resources(adapter);
5089 ixgbe_reset(adapter);
5095 * ixgbe_close - Disables a network interface
5096 * @netdev: network interface device structure
5098 * Returns 0, this is not allowed to fail
5100 * The close entry point is called when an interface is de-activated
5101 * by the OS. The hardware is still under the drivers control, but
5102 * needs to be disabled. A global MAC reset is issued to stop the
5103 * hardware, and all transmit and receive resources are freed.
5105 static int ixgbe_close(struct net_device *netdev)
5107 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5109 ixgbe_down(adapter);
5110 ixgbe_free_irq(adapter);
5112 ixgbe_free_all_tx_resources(adapter);
5113 ixgbe_free_all_rx_resources(adapter);
5115 ixgbe_release_hw_control(adapter);
5121 static int ixgbe_resume(struct pci_dev *pdev)
5123 struct net_device *netdev = pci_get_drvdata(pdev);
5124 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5127 pci_set_power_state(pdev, PCI_D0);
5128 pci_restore_state(pdev);
5130 * pci_restore_state clears dev->state_saved so call
5131 * pci_save_state to restore it.
5133 pci_save_state(pdev);
5135 err = pci_enable_device_mem(pdev);
5137 e_dev_err("Cannot enable PCI device from suspend\n");
5140 pci_set_master(pdev);
5142 pci_wake_from_d3(pdev, false);
5144 err = ixgbe_init_interrupt_scheme(adapter);
5146 e_dev_err("Cannot initialize interrupts for device\n");
5150 ixgbe_reset(adapter);
5152 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5154 if (netif_running(netdev)) {
5155 err = ixgbe_open(adapter->netdev);
5160 netif_device_attach(netdev);
5164 #endif /* CONFIG_PM */
5166 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5168 struct net_device *netdev = pci_get_drvdata(pdev);
5169 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5170 struct ixgbe_hw *hw = &adapter->hw;
5172 u32 wufc = adapter->wol;
5177 netif_device_detach(netdev);
5179 if (netif_running(netdev)) {
5180 ixgbe_down(adapter);
5181 ixgbe_free_irq(adapter);
5182 ixgbe_free_all_tx_resources(adapter);
5183 ixgbe_free_all_rx_resources(adapter);
5185 ixgbe_clear_interrupt_scheme(adapter);
5188 retval = pci_save_state(pdev);
5194 ixgbe_set_rx_mode(netdev);
5196 /* turn on all-multi mode if wake on multicast is enabled */
5197 if (wufc & IXGBE_WUFC_MC) {
5198 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5199 fctrl |= IXGBE_FCTRL_MPE;
5200 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5203 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5204 ctrl |= IXGBE_CTRL_GIO_DIS;
5205 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5207 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5209 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5210 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5213 if (wufc && hw->mac.type == ixgbe_mac_82599EB)
5214 pci_wake_from_d3(pdev, true);
5216 pci_wake_from_d3(pdev, false);
5218 *enable_wake = !!wufc;
5220 ixgbe_release_hw_control(adapter);
5222 pci_disable_device(pdev);
5228 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5233 retval = __ixgbe_shutdown(pdev, &wake);
5238 pci_prepare_to_sleep(pdev);
5240 pci_wake_from_d3(pdev, false);
5241 pci_set_power_state(pdev, PCI_D3hot);
5246 #endif /* CONFIG_PM */
5248 static void ixgbe_shutdown(struct pci_dev *pdev)
5252 __ixgbe_shutdown(pdev, &wake);
5254 if (system_state == SYSTEM_POWER_OFF) {
5255 pci_wake_from_d3(pdev, wake);
5256 pci_set_power_state(pdev, PCI_D3hot);
5261 * ixgbe_update_stats - Update the board statistics counters.
5262 * @adapter: board private structure
5264 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5266 struct net_device *netdev = adapter->netdev;
5267 struct ixgbe_hw *hw = &adapter->hw;
5269 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5270 u64 non_eop_descs = 0, restart_queue = 0;
5272 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5273 test_bit(__IXGBE_RESETTING, &adapter->state))
5276 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5279 for (i = 0; i < 16; i++)
5280 adapter->hw_rx_no_dma_resources +=
5281 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5282 for (i = 0; i < adapter->num_rx_queues; i++) {
5283 rsc_count += adapter->rx_ring[i]->rsc_count;
5284 rsc_flush += adapter->rx_ring[i]->rsc_flush;
5286 adapter->rsc_total_count = rsc_count;
5287 adapter->rsc_total_flush = rsc_flush;
5290 /* gather some stats to the adapter struct that are per queue */
5291 for (i = 0; i < adapter->num_tx_queues; i++)
5292 restart_queue += adapter->tx_ring[i]->restart_queue;
5293 adapter->restart_queue = restart_queue;
5295 for (i = 0; i < adapter->num_rx_queues; i++)
5296 non_eop_descs += adapter->rx_ring[i]->non_eop_descs;
5297 adapter->non_eop_descs = non_eop_descs;
5299 adapter->stats.crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5300 for (i = 0; i < 8; i++) {
5301 /* for packet buffers not used, the register should read 0 */
5302 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5304 adapter->stats.mpc[i] += mpc;
5305 total_mpc += adapter->stats.mpc[i];
5306 if (hw->mac.type == ixgbe_mac_82598EB)
5307 adapter->stats.rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5308 adapter->stats.qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5309 adapter->stats.qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5310 adapter->stats.qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5311 adapter->stats.qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5312 if (hw->mac.type == ixgbe_mac_82599EB) {
5313 adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
5314 IXGBE_PXONRXCNT(i));
5315 adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
5316 IXGBE_PXOFFRXCNT(i));
5317 adapter->stats.qprdc[i] += IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5319 adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
5321 adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
5324 adapter->stats.pxontxc[i] += IXGBE_READ_REG(hw,
5326 adapter->stats.pxofftxc[i] += IXGBE_READ_REG(hw,
5329 adapter->stats.gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5330 /* work around hardware counting issue */
5331 adapter->stats.gprc -= missed_rx;
5333 /* 82598 hardware only has a 32 bit counter in the high register */
5334 if (hw->mac.type == ixgbe_mac_82599EB) {
5336 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5337 tmp = IXGBE_READ_REG(hw, IXGBE_GORCH) & 0xF; /* 4 high bits of GORC */
5338 adapter->stats.gorc += (tmp << 32);
5339 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5340 tmp = IXGBE_READ_REG(hw, IXGBE_GOTCH) & 0xF; /* 4 high bits of GOTC */
5341 adapter->stats.gotc += (tmp << 32);
5342 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5343 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5344 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5345 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
5346 adapter->stats.fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5347 adapter->stats.fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5349 adapter->stats.fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5350 adapter->stats.fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5351 adapter->stats.fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5352 adapter->stats.fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5353 adapter->stats.fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5354 adapter->stats.fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5355 #endif /* IXGBE_FCOE */
5357 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5358 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
5359 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5360 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5361 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5363 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5364 adapter->stats.bprc += bprc;
5365 adapter->stats.mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5366 if (hw->mac.type == ixgbe_mac_82598EB)
5367 adapter->stats.mprc -= bprc;
5368 adapter->stats.roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5369 adapter->stats.prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5370 adapter->stats.prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5371 adapter->stats.prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5372 adapter->stats.prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5373 adapter->stats.prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5374 adapter->stats.prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5375 adapter->stats.rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5376 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5377 adapter->stats.lxontxc += lxon;
5378 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5379 adapter->stats.lxofftxc += lxoff;
5380 adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5381 adapter->stats.gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5382 adapter->stats.mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5384 * 82598 errata - tx of flow control packets is included in tx counters
5386 xon_off_tot = lxon + lxoff;
5387 adapter->stats.gptc -= xon_off_tot;
5388 adapter->stats.mptc -= xon_off_tot;
5389 adapter->stats.gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5390 adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5391 adapter->stats.rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5392 adapter->stats.rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5393 adapter->stats.tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5394 adapter->stats.ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5395 adapter->stats.ptc64 -= xon_off_tot;
5396 adapter->stats.ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5397 adapter->stats.ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5398 adapter->stats.ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5399 adapter->stats.ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5400 adapter->stats.ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5401 adapter->stats.bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5403 /* Fill out the OS statistics structure */
5404 netdev->stats.multicast = adapter->stats.mprc;
5407 netdev->stats.rx_errors = adapter->stats.crcerrs +
5408 adapter->stats.rlec;
5409 netdev->stats.rx_dropped = 0;
5410 netdev->stats.rx_length_errors = adapter->stats.rlec;
5411 netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
5412 netdev->stats.rx_missed_errors = total_mpc;
5416 * ixgbe_watchdog - Timer Call-back
5417 * @data: pointer to adapter cast into an unsigned long
5419 static void ixgbe_watchdog(unsigned long data)
5421 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5422 struct ixgbe_hw *hw = &adapter->hw;
5427 * Do the watchdog outside of interrupt context due to the lovely
5428 * delays that some of the newer hardware requires
5431 if (test_bit(__IXGBE_DOWN, &adapter->state))
5432 goto watchdog_short_circuit;
5434 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5436 * for legacy and MSI interrupts don't set any bits
5437 * that are enabled for EIAM, because this operation
5438 * would set *both* EIMS and EICS for any bit in EIAM
5440 IXGBE_WRITE_REG(hw, IXGBE_EICS,
5441 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5442 goto watchdog_reschedule;
5445 /* get one bit for every active tx/rx interrupt vector */
5446 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
5447 struct ixgbe_q_vector *qv = adapter->q_vector[i];
5448 if (qv->rxr_count || qv->txr_count)
5449 eics |= ((u64)1 << i);
5452 /* Cause software interrupt to ensure rx rings are cleaned */
5453 ixgbe_irq_rearm_queues(adapter, eics);
5455 watchdog_reschedule:
5456 /* Reset the timer */
5457 mod_timer(&adapter->watchdog_timer, round_jiffies(jiffies + 2 * HZ));
5459 watchdog_short_circuit:
5460 schedule_work(&adapter->watchdog_task);
5464 * ixgbe_multispeed_fiber_task - worker thread to configure multispeed fiber
5465 * @work: pointer to work_struct containing our data
5467 static void ixgbe_multispeed_fiber_task(struct work_struct *work)
5469 struct ixgbe_adapter *adapter = container_of(work,
5470 struct ixgbe_adapter,
5471 multispeed_fiber_task);
5472 struct ixgbe_hw *hw = &adapter->hw;
5476 adapter->flags |= IXGBE_FLAG_IN_SFP_LINK_TASK;
5477 autoneg = hw->phy.autoneg_advertised;
5478 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
5479 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
5480 hw->mac.autotry_restart = false;
5481 if (hw->mac.ops.setup_link)
5482 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
5483 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5484 adapter->flags &= ~IXGBE_FLAG_IN_SFP_LINK_TASK;
5488 * ixgbe_sfp_config_module_task - worker thread to configure a new SFP+ module
5489 * @work: pointer to work_struct containing our data
5491 static void ixgbe_sfp_config_module_task(struct work_struct *work)
5493 struct ixgbe_adapter *adapter = container_of(work,
5494 struct ixgbe_adapter,
5495 sfp_config_module_task);
5496 struct ixgbe_hw *hw = &adapter->hw;
5499 adapter->flags |= IXGBE_FLAG_IN_SFP_MOD_TASK;
5501 /* Time for electrical oscillations to settle down */
5503 err = hw->phy.ops.identify_sfp(hw);
5505 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
5506 e_dev_err("failed to initialize because an unsupported SFP+ "
5507 "module type was detected.\n");
5508 e_dev_err("Reload the driver after installing a supported "
5510 unregister_netdev(adapter->netdev);
5513 hw->mac.ops.setup_sfp(hw);
5515 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
5516 /* This will also work for DA Twinax connections */
5517 schedule_work(&adapter->multispeed_fiber_task);
5518 adapter->flags &= ~IXGBE_FLAG_IN_SFP_MOD_TASK;
5522 * ixgbe_fdir_reinit_task - worker thread to reinit FDIR filter table
5523 * @work: pointer to work_struct containing our data
5525 static void ixgbe_fdir_reinit_task(struct work_struct *work)
5527 struct ixgbe_adapter *adapter = container_of(work,
5528 struct ixgbe_adapter,
5530 struct ixgbe_hw *hw = &adapter->hw;
5533 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5534 for (i = 0; i < adapter->num_tx_queues; i++)
5535 set_bit(__IXGBE_FDIR_INIT_DONE,
5536 &(adapter->tx_ring[i]->reinit_state));
5538 e_err("failed to finish FDIR re-initialization, "
5539 "ignored adding FDIR ATR filters\n");
5541 /* Done FDIR Re-initialization, enable transmits */
5542 netif_tx_start_all_queues(adapter->netdev);
5545 static DEFINE_MUTEX(ixgbe_watchdog_lock);
5548 * ixgbe_watchdog_task - worker thread to bring link up
5549 * @work: pointer to work_struct containing our data
5551 static void ixgbe_watchdog_task(struct work_struct *work)
5553 struct ixgbe_adapter *adapter = container_of(work,
5554 struct ixgbe_adapter,
5556 struct net_device *netdev = adapter->netdev;
5557 struct ixgbe_hw *hw = &adapter->hw;
5561 struct ixgbe_ring *tx_ring;
5562 int some_tx_pending = 0;
5564 mutex_lock(&ixgbe_watchdog_lock);
5566 link_up = adapter->link_up;
5567 link_speed = adapter->link_speed;
5569 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
5570 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5573 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5574 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
5575 hw->mac.ops.fc_enable(hw, i);
5577 hw->mac.ops.fc_enable(hw, 0);
5580 hw->mac.ops.fc_enable(hw, 0);
5585 time_after(jiffies, (adapter->link_check_timeout +
5586 IXGBE_TRY_LINK_TIMEOUT))) {
5587 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5588 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5590 adapter->link_up = link_up;
5591 adapter->link_speed = link_speed;
5595 if (!netif_carrier_ok(netdev)) {
5596 bool flow_rx, flow_tx;
5598 if (hw->mac.type == ixgbe_mac_82599EB) {
5599 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5600 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5601 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5602 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5604 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5605 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5606 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5607 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5610 e_info("NIC Link is Up %s, Flow Control: %s\n",
5611 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5613 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5614 "1 Gbps" : "unknown speed")),
5615 ((flow_rx && flow_tx) ? "RX/TX" :
5617 (flow_tx ? "TX" : "None"))));
5619 netif_carrier_on(netdev);
5621 /* Force detection of hung controller */
5622 adapter->detect_tx_hung = true;
5625 adapter->link_up = false;
5626 adapter->link_speed = 0;
5627 if (netif_carrier_ok(netdev)) {
5628 e_info("NIC Link is Down\n");
5629 netif_carrier_off(netdev);
5633 if (!netif_carrier_ok(netdev)) {
5634 for (i = 0; i < adapter->num_tx_queues; i++) {
5635 tx_ring = adapter->tx_ring[i];
5636 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5637 some_tx_pending = 1;
5642 if (some_tx_pending) {
5643 /* We've lost link, so the controller stops DMA,
5644 * but we've got queued Tx work that's never going
5645 * to get done, so reset controller to flush Tx.
5646 * (Do the reset outside of interrupt context).
5648 schedule_work(&adapter->reset_task);
5652 ixgbe_update_stats(adapter);
5653 mutex_unlock(&ixgbe_watchdog_lock);
5656 static int ixgbe_tso(struct ixgbe_adapter *adapter,
5657 struct ixgbe_ring *tx_ring, struct sk_buff *skb,
5658 u32 tx_flags, u8 *hdr_len)
5660 struct ixgbe_adv_tx_context_desc *context_desc;
5663 struct ixgbe_tx_buffer *tx_buffer_info;
5664 u32 vlan_macip_lens = 0, type_tucmd_mlhl;
5665 u32 mss_l4len_idx, l4len;
5667 if (skb_is_gso(skb)) {
5668 if (skb_header_cloned(skb)) {
5669 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5673 l4len = tcp_hdrlen(skb);
5676 if (skb->protocol == htons(ETH_P_IP)) {
5677 struct iphdr *iph = ip_hdr(skb);
5680 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
5684 } else if (skb_is_gso_v6(skb)) {
5685 ipv6_hdr(skb)->payload_len = 0;
5686 tcp_hdr(skb)->check =
5687 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5688 &ipv6_hdr(skb)->daddr,
5692 i = tx_ring->next_to_use;
5694 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5695 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
5697 /* VLAN MACLEN IPLEN */
5698 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5700 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
5701 vlan_macip_lens |= ((skb_network_offset(skb)) <<
5702 IXGBE_ADVTXD_MACLEN_SHIFT);
5703 *hdr_len += skb_network_offset(skb);
5705 (skb_transport_header(skb) - skb_network_header(skb));
5707 (skb_transport_header(skb) - skb_network_header(skb));
5708 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
5709 context_desc->seqnum_seed = 0;
5711 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
5712 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
5713 IXGBE_ADVTXD_DTYP_CTXT);
5715 if (skb->protocol == htons(ETH_P_IP))
5716 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
5717 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5718 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
5722 (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
5723 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
5724 /* use index 1 for TSO */
5725 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
5726 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
5728 tx_buffer_info->time_stamp = jiffies;
5729 tx_buffer_info->next_to_watch = i;
5732 if (i == tx_ring->count)
5734 tx_ring->next_to_use = i;
5741 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
5742 struct ixgbe_ring *tx_ring,
5743 struct sk_buff *skb, u32 tx_flags)
5745 struct ixgbe_adv_tx_context_desc *context_desc;
5747 struct ixgbe_tx_buffer *tx_buffer_info;
5748 u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
5750 if (skb->ip_summed == CHECKSUM_PARTIAL ||
5751 (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
5752 i = tx_ring->next_to_use;
5753 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5754 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
5756 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5758 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
5759 vlan_macip_lens |= (skb_network_offset(skb) <<
5760 IXGBE_ADVTXD_MACLEN_SHIFT);
5761 if (skb->ip_summed == CHECKSUM_PARTIAL)
5762 vlan_macip_lens |= (skb_transport_header(skb) -
5763 skb_network_header(skb));
5765 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
5766 context_desc->seqnum_seed = 0;
5768 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
5769 IXGBE_ADVTXD_DTYP_CTXT);
5771 if (skb->ip_summed == CHECKSUM_PARTIAL) {
5774 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
5775 const struct vlan_ethhdr *vhdr =
5776 (const struct vlan_ethhdr *)skb->data;
5778 protocol = vhdr->h_vlan_encapsulated_proto;
5780 protocol = skb->protocol;
5784 case cpu_to_be16(ETH_P_IP):
5785 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
5786 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
5788 IXGBE_ADVTXD_TUCMD_L4T_TCP;
5789 else if (ip_hdr(skb)->protocol == IPPROTO_SCTP)
5791 IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5793 case cpu_to_be16(ETH_P_IPV6):
5794 /* XXX what about other V6 headers?? */
5795 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
5797 IXGBE_ADVTXD_TUCMD_L4T_TCP;
5798 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_SCTP)
5800 IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5803 if (unlikely(net_ratelimit())) {
5804 e_warn("partial checksum but "
5805 "proto=%x!\n", skb->protocol);
5811 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
5812 /* use index zero for tx checksum offload */
5813 context_desc->mss_l4len_idx = 0;
5815 tx_buffer_info->time_stamp = jiffies;
5816 tx_buffer_info->next_to_watch = i;
5819 if (i == tx_ring->count)
5821 tx_ring->next_to_use = i;
5829 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
5830 struct ixgbe_ring *tx_ring,
5831 struct sk_buff *skb, u32 tx_flags,
5834 struct pci_dev *pdev = adapter->pdev;
5835 struct ixgbe_tx_buffer *tx_buffer_info;
5837 unsigned int total = skb->len;
5838 unsigned int offset = 0, size, count = 0, i;
5839 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
5842 i = tx_ring->next_to_use;
5844 if (tx_flags & IXGBE_TX_FLAGS_FCOE)
5845 /* excluding fcoe_crc_eof for FCoE */
5846 total -= sizeof(struct fcoe_crc_eof);
5848 len = min(skb_headlen(skb), total);
5850 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5851 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
5853 tx_buffer_info->length = size;
5854 tx_buffer_info->mapped_as_page = false;
5855 tx_buffer_info->dma = dma_map_single(&pdev->dev,
5857 size, DMA_TO_DEVICE);
5858 if (dma_mapping_error(&pdev->dev, tx_buffer_info->dma))
5860 tx_buffer_info->time_stamp = jiffies;
5861 tx_buffer_info->next_to_watch = i;
5870 if (i == tx_ring->count)
5875 for (f = 0; f < nr_frags; f++) {
5876 struct skb_frag_struct *frag;
5878 frag = &skb_shinfo(skb)->frags[f];
5879 len = min((unsigned int)frag->size, total);
5880 offset = frag->page_offset;
5884 if (i == tx_ring->count)
5887 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5888 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
5890 tx_buffer_info->length = size;
5891 tx_buffer_info->dma = dma_map_page(&adapter->pdev->dev,
5895 tx_buffer_info->mapped_as_page = true;
5896 if (dma_mapping_error(&pdev->dev, tx_buffer_info->dma))
5898 tx_buffer_info->time_stamp = jiffies;
5899 tx_buffer_info->next_to_watch = i;
5910 tx_ring->tx_buffer_info[i].skb = skb;
5911 tx_ring->tx_buffer_info[first].next_to_watch = i;
5916 e_dev_err("TX DMA map failed\n");
5918 /* clear timestamp and dma mappings for failed tx_buffer_info map */
5919 tx_buffer_info->dma = 0;
5920 tx_buffer_info->time_stamp = 0;
5921 tx_buffer_info->next_to_watch = 0;
5925 /* clear timestamp and dma mappings for remaining portion of packet */
5928 i += tx_ring->count;
5930 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5931 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
5937 static void ixgbe_tx_queue(struct ixgbe_adapter *adapter,
5938 struct ixgbe_ring *tx_ring,
5939 int tx_flags, int count, u32 paylen, u8 hdr_len)
5941 union ixgbe_adv_tx_desc *tx_desc = NULL;
5942 struct ixgbe_tx_buffer *tx_buffer_info;
5943 u32 olinfo_status = 0, cmd_type_len = 0;
5945 u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
5947 cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
5949 cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
5951 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5952 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
5954 if (tx_flags & IXGBE_TX_FLAGS_TSO) {
5955 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
5957 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
5958 IXGBE_ADVTXD_POPTS_SHIFT;
5960 /* use index 1 context for tso */
5961 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
5962 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
5963 olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
5964 IXGBE_ADVTXD_POPTS_SHIFT;
5966 } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
5967 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
5968 IXGBE_ADVTXD_POPTS_SHIFT;
5970 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
5971 olinfo_status |= IXGBE_ADVTXD_CC;
5972 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
5973 if (tx_flags & IXGBE_TX_FLAGS_FSO)
5974 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
5977 olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
5979 i = tx_ring->next_to_use;
5981 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5982 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
5983 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
5984 tx_desc->read.cmd_type_len =
5985 cpu_to_le32(cmd_type_len | tx_buffer_info->length);
5986 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
5988 if (i == tx_ring->count)
5992 tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
5995 * Force memory writes to complete before letting h/w
5996 * know there are new descriptors to fetch. (Only
5997 * applicable for weak-ordered memory model archs,
6002 tx_ring->next_to_use = i;
6003 writel(i, adapter->hw.hw_addr + tx_ring->tail);
6006 static void ixgbe_atr(struct ixgbe_adapter *adapter, struct sk_buff *skb,
6007 int queue, u32 tx_flags)
6009 /* Right now, we support IPv4 only */
6010 struct ixgbe_atr_input atr_input;
6012 struct iphdr *iph = ip_hdr(skb);
6013 struct ethhdr *eth = (struct ethhdr *)skb->data;
6014 u16 vlan_id, src_port, dst_port, flex_bytes;
6015 u32 src_ipv4_addr, dst_ipv4_addr;
6018 /* check if we're UDP or TCP */
6019 if (iph->protocol == IPPROTO_TCP) {
6021 src_port = th->source;
6022 dst_port = th->dest;
6023 l4type |= IXGBE_ATR_L4TYPE_TCP;
6024 /* l4type IPv4 type is 0, no need to assign */
6026 /* Unsupported L4 header, just bail here */
6030 memset(&atr_input, 0, sizeof(struct ixgbe_atr_input));
6032 vlan_id = (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK) >>
6033 IXGBE_TX_FLAGS_VLAN_SHIFT;
6034 src_ipv4_addr = iph->saddr;
6035 dst_ipv4_addr = iph->daddr;
6036 flex_bytes = eth->h_proto;
6038 ixgbe_atr_set_vlan_id_82599(&atr_input, vlan_id);
6039 ixgbe_atr_set_src_port_82599(&atr_input, dst_port);
6040 ixgbe_atr_set_dst_port_82599(&atr_input, src_port);
6041 ixgbe_atr_set_flex_byte_82599(&atr_input, flex_bytes);
6042 ixgbe_atr_set_l4type_82599(&atr_input, l4type);
6043 /* src and dst are inverted, think how the receiver sees them */
6044 ixgbe_atr_set_src_ipv4_82599(&atr_input, dst_ipv4_addr);
6045 ixgbe_atr_set_dst_ipv4_82599(&atr_input, src_ipv4_addr);
6047 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
6048 ixgbe_fdir_add_signature_filter_82599(&adapter->hw, &atr_input, queue);
6051 static int __ixgbe_maybe_stop_tx(struct net_device *netdev,
6052 struct ixgbe_ring *tx_ring, int size)
6054 netif_stop_subqueue(netdev, tx_ring->queue_index);
6055 /* Herbert's original patch had:
6056 * smp_mb__after_netif_stop_queue();
6057 * but since that doesn't exist yet, just open code it. */
6060 /* We need to check again in a case another CPU has just
6061 * made room available. */
6062 if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
6065 /* A reprieve! - use start_queue because it doesn't call schedule */
6066 netif_start_subqueue(netdev, tx_ring->queue_index);
6067 ++tx_ring->restart_queue;
6071 static int ixgbe_maybe_stop_tx(struct net_device *netdev,
6072 struct ixgbe_ring *tx_ring, int size)
6074 if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
6076 return __ixgbe_maybe_stop_tx(netdev, tx_ring, size);
6079 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6081 struct ixgbe_adapter *adapter = netdev_priv(dev);
6082 int txq = smp_processor_id();
6084 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6085 while (unlikely(txq >= dev->real_num_tx_queues))
6086 txq -= dev->real_num_tx_queues;
6091 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
6092 ((skb->protocol == htons(ETH_P_FCOE)) ||
6093 (skb->protocol == htons(ETH_P_FIP)))) {
6094 txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
6095 txq += adapter->ring_feature[RING_F_FCOE].mask;
6099 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6100 if (skb->priority == TC_PRIO_CONTROL)
6101 txq = adapter->ring_feature[RING_F_DCB].indices-1;
6103 txq = (skb->vlan_tci & IXGBE_TX_FLAGS_VLAN_PRIO_MASK)
6108 return skb_tx_hash(dev, skb);
6111 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
6112 struct net_device *netdev)
6114 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6115 struct ixgbe_ring *tx_ring;
6116 struct netdev_queue *txq;
6118 unsigned int tx_flags = 0;
6124 if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
6125 tx_flags |= vlan_tx_tag_get(skb);
6126 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6127 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6128 tx_flags |= ((skb->queue_mapping & 0x7) << 13);
6130 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6131 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6132 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED &&
6133 skb->priority != TC_PRIO_CONTROL) {
6134 tx_flags |= ((skb->queue_mapping & 0x7) << 13);
6135 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6136 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6139 tx_ring = adapter->tx_ring[skb->queue_mapping];
6142 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
6143 #ifdef CONFIG_IXGBE_DCB
6144 /* for FCoE with DCB, we force the priority to what
6145 * was specified by the switch */
6146 if ((skb->protocol == htons(ETH_P_FCOE)) ||
6147 (skb->protocol == htons(ETH_P_FIP))) {
6148 tx_flags &= ~(IXGBE_TX_FLAGS_VLAN_PRIO_MASK
6149 << IXGBE_TX_FLAGS_VLAN_SHIFT);
6150 tx_flags |= ((adapter->fcoe.up << 13)
6151 << IXGBE_TX_FLAGS_VLAN_SHIFT);
6154 /* flag for FCoE offloads */
6155 if (skb->protocol == htons(ETH_P_FCOE))
6156 tx_flags |= IXGBE_TX_FLAGS_FCOE;
6160 /* four things can cause us to need a context descriptor */
6161 if (skb_is_gso(skb) ||
6162 (skb->ip_summed == CHECKSUM_PARTIAL) ||
6163 (tx_flags & IXGBE_TX_FLAGS_VLAN) ||
6164 (tx_flags & IXGBE_TX_FLAGS_FCOE))
6167 count += TXD_USE_COUNT(skb_headlen(skb));
6168 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6169 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6171 if (ixgbe_maybe_stop_tx(netdev, tx_ring, count)) {
6173 return NETDEV_TX_BUSY;
6176 first = tx_ring->next_to_use;
6177 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6179 /* setup tx offload for FCoE */
6180 tso = ixgbe_fso(adapter, tx_ring, skb, tx_flags, &hdr_len);
6182 dev_kfree_skb_any(skb);
6183 return NETDEV_TX_OK;
6186 tx_flags |= IXGBE_TX_FLAGS_FSO;
6187 #endif /* IXGBE_FCOE */
6189 if (skb->protocol == htons(ETH_P_IP))
6190 tx_flags |= IXGBE_TX_FLAGS_IPV4;
6191 tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len);
6193 dev_kfree_skb_any(skb);
6194 return NETDEV_TX_OK;
6198 tx_flags |= IXGBE_TX_FLAGS_TSO;
6199 else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags) &&
6200 (skb->ip_summed == CHECKSUM_PARTIAL))
6201 tx_flags |= IXGBE_TX_FLAGS_CSUM;
6204 count = ixgbe_tx_map(adapter, tx_ring, skb, tx_flags, first);
6206 /* add the ATR filter if ATR is on */
6207 if (tx_ring->atr_sample_rate) {
6208 ++tx_ring->atr_count;
6209 if ((tx_ring->atr_count >= tx_ring->atr_sample_rate) &&
6210 test_bit(__IXGBE_FDIR_INIT_DONE,
6211 &tx_ring->reinit_state)) {
6212 ixgbe_atr(adapter, skb, tx_ring->queue_index,
6214 tx_ring->atr_count = 0;
6217 txq = netdev_get_tx_queue(netdev, tx_ring->queue_index);
6218 txq->tx_bytes += skb->len;
6220 ixgbe_tx_queue(adapter, tx_ring, tx_flags, count, skb->len,
6222 ixgbe_maybe_stop_tx(netdev, tx_ring, DESC_NEEDED);
6225 dev_kfree_skb_any(skb);
6226 tx_ring->tx_buffer_info[first].time_stamp = 0;
6227 tx_ring->next_to_use = first;
6230 return NETDEV_TX_OK;
6234 * ixgbe_set_mac - Change the Ethernet Address of the NIC
6235 * @netdev: network interface device structure
6236 * @p: pointer to an address structure
6238 * Returns 0 on success, negative on failure
6240 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6242 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6243 struct ixgbe_hw *hw = &adapter->hw;
6244 struct sockaddr *addr = p;
6246 if (!is_valid_ether_addr(addr->sa_data))
6247 return -EADDRNOTAVAIL;
6249 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6250 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6252 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
6259 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6261 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6262 struct ixgbe_hw *hw = &adapter->hw;
6266 if (prtad != hw->phy.mdio.prtad)
6268 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6274 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6275 u16 addr, u16 value)
6277 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6278 struct ixgbe_hw *hw = &adapter->hw;
6280 if (prtad != hw->phy.mdio.prtad)
6282 return hw->phy.ops.write_reg(hw, addr, devad, value);
6285 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6287 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6289 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6293 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6295 * @netdev: network interface device structure
6297 * Returns non-zero on failure
6299 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6302 struct ixgbe_adapter *adapter = netdev_priv(dev);
6303 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6305 if (is_valid_ether_addr(mac->san_addr)) {
6307 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6314 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6316 * @netdev: network interface device structure
6318 * Returns non-zero on failure
6320 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6323 struct ixgbe_adapter *adapter = netdev_priv(dev);
6324 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6326 if (is_valid_ether_addr(mac->san_addr)) {
6328 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6334 #ifdef CONFIG_NET_POLL_CONTROLLER
6336 * Polling 'interrupt' - used by things like netconsole to send skbs
6337 * without having to re-enable interrupts. It's not called while
6338 * the interrupt routine is executing.
6340 static void ixgbe_netpoll(struct net_device *netdev)
6342 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6345 /* if interface is down do nothing */
6346 if (test_bit(__IXGBE_DOWN, &adapter->state))
6349 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
6350 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
6351 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
6352 for (i = 0; i < num_q_vectors; i++) {
6353 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
6354 ixgbe_msix_clean_many(0, q_vector);
6357 ixgbe_intr(adapter->pdev->irq, netdev);
6359 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
6363 static const struct net_device_ops ixgbe_netdev_ops = {
6364 .ndo_open = ixgbe_open,
6365 .ndo_stop = ixgbe_close,
6366 .ndo_start_xmit = ixgbe_xmit_frame,
6367 .ndo_select_queue = ixgbe_select_queue,
6368 .ndo_set_rx_mode = ixgbe_set_rx_mode,
6369 .ndo_set_multicast_list = ixgbe_set_rx_mode,
6370 .ndo_validate_addr = eth_validate_addr,
6371 .ndo_set_mac_address = ixgbe_set_mac,
6372 .ndo_change_mtu = ixgbe_change_mtu,
6373 .ndo_tx_timeout = ixgbe_tx_timeout,
6374 .ndo_vlan_rx_register = ixgbe_vlan_rx_register,
6375 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
6376 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
6377 .ndo_do_ioctl = ixgbe_ioctl,
6378 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
6379 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
6380 .ndo_set_vf_tx_rate = ixgbe_ndo_set_vf_bw,
6381 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
6382 #ifdef CONFIG_NET_POLL_CONTROLLER
6383 .ndo_poll_controller = ixgbe_netpoll,
6386 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
6387 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
6388 .ndo_fcoe_enable = ixgbe_fcoe_enable,
6389 .ndo_fcoe_disable = ixgbe_fcoe_disable,
6390 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
6391 #endif /* IXGBE_FCOE */
6394 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
6395 const struct ixgbe_info *ii)
6397 #ifdef CONFIG_PCI_IOV
6398 struct ixgbe_hw *hw = &adapter->hw;
6401 if (hw->mac.type != ixgbe_mac_82599EB || !max_vfs)
6404 /* The 82599 supports up to 64 VFs per physical function
6405 * but this implementation limits allocation to 63 so that
6406 * basic networking resources are still available to the
6409 adapter->num_vfs = (max_vfs > 63) ? 63 : max_vfs;
6410 adapter->flags |= IXGBE_FLAG_SRIOV_ENABLED;
6411 err = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
6413 e_err("Failed to enable PCI sriov: %d\n", err);
6416 /* If call to enable VFs succeeded then allocate memory
6417 * for per VF control structures.
6420 kcalloc(adapter->num_vfs,
6421 sizeof(struct vf_data_storage), GFP_KERNEL);
6422 if (adapter->vfinfo) {
6423 /* Now that we're sure SR-IOV is enabled
6424 * and memory allocated set up the mailbox parameters
6426 ixgbe_init_mbx_params_pf(hw);
6427 memcpy(&hw->mbx.ops, ii->mbx_ops,
6428 sizeof(hw->mbx.ops));
6430 /* Disable RSC when in SR-IOV mode */
6431 adapter->flags2 &= ~(IXGBE_FLAG2_RSC_CAPABLE |
6432 IXGBE_FLAG2_RSC_ENABLED);
6437 e_err("Unable to allocate memory for VF Data Storage - SRIOV "
6439 pci_disable_sriov(adapter->pdev);
6442 adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
6443 adapter->num_vfs = 0;
6444 #endif /* CONFIG_PCI_IOV */
6448 * ixgbe_probe - Device Initialization Routine
6449 * @pdev: PCI device information struct
6450 * @ent: entry in ixgbe_pci_tbl
6452 * Returns 0 on success, negative on failure
6454 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
6455 * The OS initialization, configuring of the adapter private structure,
6456 * and a hardware reset occur.
6458 static int __devinit ixgbe_probe(struct pci_dev *pdev,
6459 const struct pci_device_id *ent)
6461 struct net_device *netdev;
6462 struct ixgbe_adapter *adapter = NULL;
6463 struct ixgbe_hw *hw;
6464 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
6465 static int cards_found;
6466 int i, err, pci_using_dac;
6467 unsigned int indices = num_possible_cpus();
6473 err = pci_enable_device_mem(pdev);
6477 if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
6478 !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
6481 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
6483 err = dma_set_coherent_mask(&pdev->dev,
6486 e_dev_err("No usable DMA configuration, "
6494 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
6495 IORESOURCE_MEM), ixgbe_driver_name);
6497 e_dev_err("pci_request_selected_regions failed 0x%x\n", err);
6501 pci_enable_pcie_error_reporting(pdev);
6503 pci_set_master(pdev);
6504 pci_save_state(pdev);
6506 if (ii->mac == ixgbe_mac_82598EB)
6507 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
6509 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
6511 indices = max_t(unsigned int, indices, IXGBE_MAX_DCB_INDICES);
6513 indices += min_t(unsigned int, num_possible_cpus(),
6514 IXGBE_MAX_FCOE_INDICES);
6516 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
6519 goto err_alloc_etherdev;
6522 SET_NETDEV_DEV(netdev, &pdev->dev);
6524 pci_set_drvdata(pdev, netdev);
6525 adapter = netdev_priv(netdev);
6527 adapter->netdev = netdev;
6528 adapter->pdev = pdev;
6531 adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
6533 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
6534 pci_resource_len(pdev, 0));
6540 for (i = 1; i <= 5; i++) {
6541 if (pci_resource_len(pdev, i) == 0)
6545 netdev->netdev_ops = &ixgbe_netdev_ops;
6546 ixgbe_set_ethtool_ops(netdev);
6547 netdev->watchdog_timeo = 5 * HZ;
6548 strcpy(netdev->name, pci_name(pdev));
6550 adapter->bd_number = cards_found;
6553 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
6554 hw->mac.type = ii->mac;
6557 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
6558 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
6559 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
6560 if (!(eec & (1 << 8)))
6561 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
6564 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
6565 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
6566 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
6567 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
6568 hw->phy.mdio.mmds = 0;
6569 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
6570 hw->phy.mdio.dev = netdev;
6571 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
6572 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
6574 /* set up this timer and work struct before calling get_invariants
6575 * which might start the timer
6577 init_timer(&adapter->sfp_timer);
6578 adapter->sfp_timer.function = &ixgbe_sfp_timer;
6579 adapter->sfp_timer.data = (unsigned long) adapter;
6581 INIT_WORK(&adapter->sfp_task, ixgbe_sfp_task);
6583 /* multispeed fiber has its own tasklet, called from GPI SDP1 context */
6584 INIT_WORK(&adapter->multispeed_fiber_task, ixgbe_multispeed_fiber_task);
6586 /* a new SFP+ module arrival, called from GPI SDP2 context */
6587 INIT_WORK(&adapter->sfp_config_module_task,
6588 ixgbe_sfp_config_module_task);
6590 ii->get_invariants(hw);
6592 /* setup the private structure */
6593 err = ixgbe_sw_init(adapter);
6597 /* Make it possible the adapter to be woken up via WOL */
6598 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
6599 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6602 * If there is a fan on this device and it has failed log the
6605 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
6606 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
6607 if (esdp & IXGBE_ESDP_SDP1)
6608 e_crit("Fan has stopped, replace the adapter\n");
6611 /* reset_hw fills in the perm_addr as well */
6612 hw->phy.reset_if_overtemp = true;
6613 err = hw->mac.ops.reset_hw(hw);
6614 hw->phy.reset_if_overtemp = false;
6615 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
6616 hw->mac.type == ixgbe_mac_82598EB) {
6618 * Start a kernel thread to watch for a module to arrive.
6619 * Only do this for 82598, since 82599 will generate
6620 * interrupts on module arrival.
6622 set_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
6623 mod_timer(&adapter->sfp_timer,
6624 round_jiffies(jiffies + (2 * HZ)));
6626 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
6627 e_dev_err("failed to initialize because an unsupported SFP+ "
6628 "module type was detected.\n");
6629 e_dev_err("Reload the driver after installing a supported "
6633 e_dev_err("HW Init failed: %d\n", err);
6637 ixgbe_probe_vf(adapter, ii);
6639 netdev->features = NETIF_F_SG |
6641 NETIF_F_HW_VLAN_TX |
6642 NETIF_F_HW_VLAN_RX |
6643 NETIF_F_HW_VLAN_FILTER;
6645 netdev->features |= NETIF_F_IPV6_CSUM;
6646 netdev->features |= NETIF_F_TSO;
6647 netdev->features |= NETIF_F_TSO6;
6648 netdev->features |= NETIF_F_GRO;
6650 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
6651 netdev->features |= NETIF_F_SCTP_CSUM;
6653 netdev->vlan_features |= NETIF_F_TSO;
6654 netdev->vlan_features |= NETIF_F_TSO6;
6655 netdev->vlan_features |= NETIF_F_IP_CSUM;
6656 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
6657 netdev->vlan_features |= NETIF_F_SG;
6659 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6660 adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
6661 IXGBE_FLAG_DCB_ENABLED);
6662 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
6663 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
6665 #ifdef CONFIG_IXGBE_DCB
6666 netdev->dcbnl_ops = &dcbnl_ops;
6670 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
6671 if (hw->mac.ops.get_device_caps) {
6672 hw->mac.ops.get_device_caps(hw, &device_caps);
6673 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
6674 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6677 #endif /* IXGBE_FCOE */
6679 netdev->features |= NETIF_F_HIGHDMA;
6681 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
6682 netdev->features |= NETIF_F_LRO;
6684 /* make sure the EEPROM is good */
6685 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
6686 e_dev_err("The EEPROM Checksum Is Not Valid\n");
6691 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
6692 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
6694 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
6695 e_dev_err("invalid MAC address\n");
6700 /* power down the optics */
6701 if (hw->phy.multispeed_fiber)
6702 hw->mac.ops.disable_tx_laser(hw);
6704 init_timer(&adapter->watchdog_timer);
6705 adapter->watchdog_timer.function = &ixgbe_watchdog;
6706 adapter->watchdog_timer.data = (unsigned long)adapter;
6708 INIT_WORK(&adapter->reset_task, ixgbe_reset_task);
6709 INIT_WORK(&adapter->watchdog_task, ixgbe_watchdog_task);
6711 err = ixgbe_init_interrupt_scheme(adapter);
6715 switch (pdev->device) {
6716 case IXGBE_DEV_ID_82599_KX4:
6717 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
6718 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
6724 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
6726 /* pick up the PCI bus settings for reporting later */
6727 hw->mac.ops.get_bus_info(hw);
6729 /* print bus type/speed/width info */
6730 e_dev_info("(PCI Express:%s:%s) %pM\n",
6731 ((hw->bus.speed == ixgbe_bus_speed_5000) ? "5.0Gb/s":
6732 (hw->bus.speed == ixgbe_bus_speed_2500) ? "2.5Gb/s":"Unknown"),
6733 ((hw->bus.width == ixgbe_bus_width_pcie_x8) ? "Width x8" :
6734 (hw->bus.width == ixgbe_bus_width_pcie_x4) ? "Width x4" :
6735 (hw->bus.width == ixgbe_bus_width_pcie_x1) ? "Width x1" :
6738 ixgbe_read_pba_num_generic(hw, &part_num);
6739 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
6740 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, "
6741 "PBA No: %06x-%03x\n",
6742 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
6743 (part_num >> 8), (part_num & 0xff));
6745 e_dev_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
6746 hw->mac.type, hw->phy.type,
6747 (part_num >> 8), (part_num & 0xff));
6749 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
6750 e_dev_warn("PCI-Express bandwidth available for this card is "
6751 "not sufficient for optimal performance.\n");
6752 e_dev_warn("For optimal performance a x8 PCI-Express slot "
6756 /* save off EEPROM version number */
6757 hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
6759 /* reset the hardware with the new settings */
6760 err = hw->mac.ops.start_hw(hw);
6762 if (err == IXGBE_ERR_EEPROM_VERSION) {
6763 /* We are running on a pre-production device, log a warning */
6764 e_dev_warn("This device is a pre-production adapter/LOM. "
6765 "Please be aware there may be issues associated "
6766 "with your hardware. If you are experiencing "
6767 "problems please contact your Intel or hardware "
6768 "representative who provided you with this "
6771 strcpy(netdev->name, "eth%d");
6772 err = register_netdev(netdev);
6776 /* carrier off reporting is important to ethtool even BEFORE open */
6777 netif_carrier_off(netdev);
6779 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
6780 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
6781 INIT_WORK(&adapter->fdir_reinit_task, ixgbe_fdir_reinit_task);
6783 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
6784 INIT_WORK(&adapter->check_overtemp_task, ixgbe_check_overtemp_task);
6785 #ifdef CONFIG_IXGBE_DCA
6786 if (dca_add_requester(&pdev->dev) == 0) {
6787 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
6788 ixgbe_setup_dca(adapter);
6791 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
6792 e_info("IOV is enabled with %d VFs\n", adapter->num_vfs);
6793 for (i = 0; i < adapter->num_vfs; i++)
6794 ixgbe_vf_configuration(pdev, (i | 0x10000000));
6797 /* add san mac addr to netdev */
6798 ixgbe_add_sanmac_netdev(netdev);
6800 e_dev_info("Intel(R) 10 Gigabit Network Connection\n");
6805 ixgbe_release_hw_control(adapter);
6806 ixgbe_clear_interrupt_scheme(adapter);
6809 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6810 ixgbe_disable_sriov(adapter);
6811 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
6812 del_timer_sync(&adapter->sfp_timer);
6813 cancel_work_sync(&adapter->sfp_task);
6814 cancel_work_sync(&adapter->multispeed_fiber_task);
6815 cancel_work_sync(&adapter->sfp_config_module_task);
6816 iounmap(hw->hw_addr);
6818 free_netdev(netdev);
6820 pci_release_selected_regions(pdev, pci_select_bars(pdev,
6824 pci_disable_device(pdev);
6829 * ixgbe_remove - Device Removal Routine
6830 * @pdev: PCI device information struct
6832 * ixgbe_remove is called by the PCI subsystem to alert the driver
6833 * that it should release a PCI device. The could be caused by a
6834 * Hot-Plug event, or because the driver is going to be removed from
6837 static void __devexit ixgbe_remove(struct pci_dev *pdev)
6839 struct net_device *netdev = pci_get_drvdata(pdev);
6840 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6842 set_bit(__IXGBE_DOWN, &adapter->state);
6843 /* clear the module not found bit to make sure the worker won't
6846 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
6847 del_timer_sync(&adapter->watchdog_timer);
6849 del_timer_sync(&adapter->sfp_timer);
6850 cancel_work_sync(&adapter->watchdog_task);
6851 cancel_work_sync(&adapter->sfp_task);
6852 cancel_work_sync(&adapter->multispeed_fiber_task);
6853 cancel_work_sync(&adapter->sfp_config_module_task);
6854 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
6855 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
6856 cancel_work_sync(&adapter->fdir_reinit_task);
6857 flush_scheduled_work();
6859 #ifdef CONFIG_IXGBE_DCA
6860 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
6861 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
6862 dca_remove_requester(&pdev->dev);
6863 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
6868 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
6869 ixgbe_cleanup_fcoe(adapter);
6871 #endif /* IXGBE_FCOE */
6873 /* remove the added san mac */
6874 ixgbe_del_sanmac_netdev(netdev);
6876 if (netdev->reg_state == NETREG_REGISTERED)
6877 unregister_netdev(netdev);
6879 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6880 ixgbe_disable_sriov(adapter);
6882 ixgbe_clear_interrupt_scheme(adapter);
6884 ixgbe_release_hw_control(adapter);
6886 iounmap(adapter->hw.hw_addr);
6887 pci_release_selected_regions(pdev, pci_select_bars(pdev,
6890 e_dev_info("complete\n");
6892 free_netdev(netdev);
6894 pci_disable_pcie_error_reporting(pdev);
6896 pci_disable_device(pdev);
6900 * ixgbe_io_error_detected - called when PCI error is detected
6901 * @pdev: Pointer to PCI device
6902 * @state: The current pci connection state
6904 * This function is called after a PCI bus error affecting
6905 * this device has been detected.
6907 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
6908 pci_channel_state_t state)
6910 struct net_device *netdev = pci_get_drvdata(pdev);
6911 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6913 netif_device_detach(netdev);
6915 if (state == pci_channel_io_perm_failure)
6916 return PCI_ERS_RESULT_DISCONNECT;
6918 if (netif_running(netdev))
6919 ixgbe_down(adapter);
6920 pci_disable_device(pdev);
6922 /* Request a slot reset. */
6923 return PCI_ERS_RESULT_NEED_RESET;
6927 * ixgbe_io_slot_reset - called after the pci bus has been reset.
6928 * @pdev: Pointer to PCI device
6930 * Restart the card from scratch, as if from a cold-boot.
6932 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
6934 struct net_device *netdev = pci_get_drvdata(pdev);
6935 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6936 pci_ers_result_t result;
6939 if (pci_enable_device_mem(pdev)) {
6940 e_err("Cannot re-enable PCI device after reset.\n");
6941 result = PCI_ERS_RESULT_DISCONNECT;
6943 pci_set_master(pdev);
6944 pci_restore_state(pdev);
6945 pci_save_state(pdev);
6947 pci_wake_from_d3(pdev, false);
6949 ixgbe_reset(adapter);
6950 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6951 result = PCI_ERS_RESULT_RECOVERED;
6954 err = pci_cleanup_aer_uncorrect_error_status(pdev);
6956 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
6957 "failed 0x%0x\n", err);
6958 /* non-fatal, continue */
6965 * ixgbe_io_resume - called when traffic can start flowing again.
6966 * @pdev: Pointer to PCI device
6968 * This callback is called when the error recovery driver tells us that
6969 * its OK to resume normal operation.
6971 static void ixgbe_io_resume(struct pci_dev *pdev)
6973 struct net_device *netdev = pci_get_drvdata(pdev);
6974 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6976 if (netif_running(netdev)) {
6977 if (ixgbe_up(adapter)) {
6978 e_info("ixgbe_up failed after reset\n");
6983 netif_device_attach(netdev);
6986 static struct pci_error_handlers ixgbe_err_handler = {
6987 .error_detected = ixgbe_io_error_detected,
6988 .slot_reset = ixgbe_io_slot_reset,
6989 .resume = ixgbe_io_resume,
6992 static struct pci_driver ixgbe_driver = {
6993 .name = ixgbe_driver_name,
6994 .id_table = ixgbe_pci_tbl,
6995 .probe = ixgbe_probe,
6996 .remove = __devexit_p(ixgbe_remove),
6998 .suspend = ixgbe_suspend,
6999 .resume = ixgbe_resume,
7001 .shutdown = ixgbe_shutdown,
7002 .err_handler = &ixgbe_err_handler
7006 * ixgbe_init_module - Driver Registration Routine
7008 * ixgbe_init_module is the first routine called when the driver is
7009 * loaded. All it does is register with the PCI subsystem.
7011 static int __init ixgbe_init_module(void)
7014 pr_info("%s - version %s\n", ixgbe_driver_string,
7015 ixgbe_driver_version);
7016 pr_info("%s\n", ixgbe_copyright);
7018 #ifdef CONFIG_IXGBE_DCA
7019 dca_register_notify(&dca_notifier);
7022 ret = pci_register_driver(&ixgbe_driver);
7026 module_init(ixgbe_init_module);
7029 * ixgbe_exit_module - Driver Exit Cleanup Routine
7031 * ixgbe_exit_module is called just before the driver is removed
7034 static void __exit ixgbe_exit_module(void)
7036 #ifdef CONFIG_IXGBE_DCA
7037 dca_unregister_notify(&dca_notifier);
7039 pci_unregister_driver(&ixgbe_driver);
7042 #ifdef CONFIG_IXGBE_DCA
7043 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
7048 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
7049 __ixgbe_notify_dca);
7051 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7054 #endif /* CONFIG_IXGBE_DCA */
7057 * ixgbe_get_hw_dev return device
7058 * used by hardware layer to print debugging information
7060 struct net_device *ixgbe_get_hw_dev(struct ixgbe_hw *hw)
7062 struct ixgbe_adapter *adapter = hw->back;
7063 return adapter->netdev;
7066 module_exit(ixgbe_exit_module);