1 /*******************************************************************************
3 Intel(R) Gigabit Ethernet Linux driver
4 Copyright(c) 2007-2012 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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
30 #include <linux/module.h>
31 #include <linux/types.h>
32 #include <linux/init.h>
33 #include <linux/bitops.h>
34 #include <linux/vmalloc.h>
35 #include <linux/pagemap.h>
36 #include <linux/netdevice.h>
37 #include <linux/ipv6.h>
38 #include <linux/slab.h>
39 #include <net/checksum.h>
40 #include <net/ip6_checksum.h>
41 #include <linux/net_tstamp.h>
42 #include <linux/mii.h>
43 #include <linux/ethtool.h>
45 #include <linux/if_vlan.h>
46 #include <linux/pci.h>
47 #include <linux/pci-aspm.h>
48 #include <linux/delay.h>
49 #include <linux/interrupt.h>
51 #include <linux/tcp.h>
52 #include <linux/sctp.h>
53 #include <linux/if_ether.h>
54 #include <linux/aer.h>
55 #include <linux/prefetch.h>
56 #include <linux/pm_runtime.h>
58 #include <linux/dca.h>
65 #define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) "." \
66 __stringify(BUILD) "-k"
67 char igb_driver_name[] = "igb";
68 char igb_driver_version[] = DRV_VERSION;
69 static const char igb_driver_string[] =
70 "Intel(R) Gigabit Ethernet Network Driver";
71 static const char igb_copyright[] = "Copyright (c) 2007-2012 Intel Corporation.";
73 static const struct e1000_info *igb_info_tbl[] = {
74 [board_82575] = &e1000_82575_info,
77 static DEFINE_PCI_DEVICE_TABLE(igb_pci_tbl) = {
78 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I211_COPPER), board_82575 },
79 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER), board_82575 },
80 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_FIBER), board_82575 },
81 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES), board_82575 },
82 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SGMII), board_82575 },
83 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_COPPER), board_82575 },
84 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_FIBER), board_82575 },
85 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SERDES), board_82575 },
86 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SGMII), board_82575 },
87 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER), board_82575 },
88 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_FIBER), board_82575 },
89 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_QUAD_FIBER), board_82575 },
90 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SERDES), board_82575 },
91 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SGMII), board_82575 },
92 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER_DUAL), board_82575 },
93 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SGMII), board_82575 },
94 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SERDES), board_82575 },
95 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_BACKPLANE), board_82575 },
96 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SFP), board_82575 },
97 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576), board_82575 },
98 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS), board_82575 },
99 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS_SERDES), board_82575 },
100 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_FIBER), board_82575 },
101 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES), board_82575 },
102 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES_QUAD), board_82575 },
103 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER_ET2), board_82575 },
104 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER), board_82575 },
105 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
106 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
107 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
108 /* required last entry */
112 MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
114 void igb_reset(struct igb_adapter *);
115 static int igb_setup_all_tx_resources(struct igb_adapter *);
116 static int igb_setup_all_rx_resources(struct igb_adapter *);
117 static void igb_free_all_tx_resources(struct igb_adapter *);
118 static void igb_free_all_rx_resources(struct igb_adapter *);
119 static void igb_setup_mrqc(struct igb_adapter *);
120 static int igb_probe(struct pci_dev *, const struct pci_device_id *);
121 static void igb_remove(struct pci_dev *pdev);
122 static int igb_sw_init(struct igb_adapter *);
123 static int igb_open(struct net_device *);
124 static int igb_close(struct net_device *);
125 static void igb_configure(struct igb_adapter *);
126 static void igb_configure_tx(struct igb_adapter *);
127 static void igb_configure_rx(struct igb_adapter *);
128 static void igb_clean_all_tx_rings(struct igb_adapter *);
129 static void igb_clean_all_rx_rings(struct igb_adapter *);
130 static void igb_clean_tx_ring(struct igb_ring *);
131 static void igb_clean_rx_ring(struct igb_ring *);
132 static void igb_set_rx_mode(struct net_device *);
133 static void igb_update_phy_info(unsigned long);
134 static void igb_watchdog(unsigned long);
135 static void igb_watchdog_task(struct work_struct *);
136 static netdev_tx_t igb_xmit_frame(struct sk_buff *skb, struct net_device *);
137 static struct rtnl_link_stats64 *igb_get_stats64(struct net_device *dev,
138 struct rtnl_link_stats64 *stats);
139 static int igb_change_mtu(struct net_device *, int);
140 static int igb_set_mac(struct net_device *, void *);
141 static void igb_set_uta(struct igb_adapter *adapter);
142 static irqreturn_t igb_intr(int irq, void *);
143 static irqreturn_t igb_intr_msi(int irq, void *);
144 static irqreturn_t igb_msix_other(int irq, void *);
145 static irqreturn_t igb_msix_ring(int irq, void *);
146 #ifdef CONFIG_IGB_DCA
147 static void igb_update_dca(struct igb_q_vector *);
148 static void igb_setup_dca(struct igb_adapter *);
149 #endif /* CONFIG_IGB_DCA */
150 static int igb_poll(struct napi_struct *, int);
151 static bool igb_clean_tx_irq(struct igb_q_vector *);
152 static bool igb_clean_rx_irq(struct igb_q_vector *, int);
153 static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
154 static void igb_tx_timeout(struct net_device *);
155 static void igb_reset_task(struct work_struct *);
156 static void igb_vlan_mode(struct net_device *netdev, netdev_features_t features);
157 static int igb_vlan_rx_add_vid(struct net_device *, u16);
158 static int igb_vlan_rx_kill_vid(struct net_device *, u16);
159 static void igb_restore_vlan(struct igb_adapter *);
160 static void igb_rar_set_qsel(struct igb_adapter *, u8 *, u32 , u8);
161 static void igb_ping_all_vfs(struct igb_adapter *);
162 static void igb_msg_task(struct igb_adapter *);
163 static void igb_vmm_control(struct igb_adapter *);
164 static int igb_set_vf_mac(struct igb_adapter *, int, unsigned char *);
165 static void igb_restore_vf_multicasts(struct igb_adapter *adapter);
166 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac);
167 static int igb_ndo_set_vf_vlan(struct net_device *netdev,
168 int vf, u16 vlan, u8 qos);
169 static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf, int tx_rate);
170 static int igb_ndo_get_vf_config(struct net_device *netdev, int vf,
171 struct ifla_vf_info *ivi);
172 static void igb_check_vf_rate_limit(struct igb_adapter *);
174 #ifdef CONFIG_PCI_IOV
175 static int igb_vf_configure(struct igb_adapter *adapter, int vf);
176 static bool igb_vfs_are_assigned(struct igb_adapter *adapter);
180 #ifdef CONFIG_PM_SLEEP
181 static int igb_suspend(struct device *);
183 static int igb_resume(struct device *);
184 #ifdef CONFIG_PM_RUNTIME
185 static int igb_runtime_suspend(struct device *dev);
186 static int igb_runtime_resume(struct device *dev);
187 static int igb_runtime_idle(struct device *dev);
189 static const struct dev_pm_ops igb_pm_ops = {
190 SET_SYSTEM_SLEEP_PM_OPS(igb_suspend, igb_resume)
191 SET_RUNTIME_PM_OPS(igb_runtime_suspend, igb_runtime_resume,
195 static void igb_shutdown(struct pci_dev *);
196 #ifdef CONFIG_IGB_DCA
197 static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
198 static struct notifier_block dca_notifier = {
199 .notifier_call = igb_notify_dca,
204 #ifdef CONFIG_NET_POLL_CONTROLLER
205 /* for netdump / net console */
206 static void igb_netpoll(struct net_device *);
208 #ifdef CONFIG_PCI_IOV
209 static unsigned int max_vfs = 0;
210 module_param(max_vfs, uint, 0);
211 MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate "
212 "per physical function");
213 #endif /* CONFIG_PCI_IOV */
215 static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
216 pci_channel_state_t);
217 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
218 static void igb_io_resume(struct pci_dev *);
220 static const struct pci_error_handlers igb_err_handler = {
221 .error_detected = igb_io_error_detected,
222 .slot_reset = igb_io_slot_reset,
223 .resume = igb_io_resume,
226 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba);
228 static struct pci_driver igb_driver = {
229 .name = igb_driver_name,
230 .id_table = igb_pci_tbl,
232 .remove = igb_remove,
234 .driver.pm = &igb_pm_ops,
236 .shutdown = igb_shutdown,
237 .err_handler = &igb_err_handler
240 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
241 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
242 MODULE_LICENSE("GPL");
243 MODULE_VERSION(DRV_VERSION);
245 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
246 static int debug = -1;
247 module_param(debug, int, 0);
248 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
250 struct igb_reg_info {
255 static const struct igb_reg_info igb_reg_info_tbl[] = {
257 /* General Registers */
258 {E1000_CTRL, "CTRL"},
259 {E1000_STATUS, "STATUS"},
260 {E1000_CTRL_EXT, "CTRL_EXT"},
262 /* Interrupt Registers */
266 {E1000_RCTL, "RCTL"},
267 {E1000_RDLEN(0), "RDLEN"},
268 {E1000_RDH(0), "RDH"},
269 {E1000_RDT(0), "RDT"},
270 {E1000_RXDCTL(0), "RXDCTL"},
271 {E1000_RDBAL(0), "RDBAL"},
272 {E1000_RDBAH(0), "RDBAH"},
275 {E1000_TCTL, "TCTL"},
276 {E1000_TDBAL(0), "TDBAL"},
277 {E1000_TDBAH(0), "TDBAH"},
278 {E1000_TDLEN(0), "TDLEN"},
279 {E1000_TDH(0), "TDH"},
280 {E1000_TDT(0), "TDT"},
281 {E1000_TXDCTL(0), "TXDCTL"},
282 {E1000_TDFH, "TDFH"},
283 {E1000_TDFT, "TDFT"},
284 {E1000_TDFHS, "TDFHS"},
285 {E1000_TDFPC, "TDFPC"},
287 /* List Terminator */
292 * igb_regdump - register printout routine
294 static void igb_regdump(struct e1000_hw *hw, struct igb_reg_info *reginfo)
300 switch (reginfo->ofs) {
302 for (n = 0; n < 4; n++)
303 regs[n] = rd32(E1000_RDLEN(n));
306 for (n = 0; n < 4; n++)
307 regs[n] = rd32(E1000_RDH(n));
310 for (n = 0; n < 4; n++)
311 regs[n] = rd32(E1000_RDT(n));
313 case E1000_RXDCTL(0):
314 for (n = 0; n < 4; n++)
315 regs[n] = rd32(E1000_RXDCTL(n));
318 for (n = 0; n < 4; n++)
319 regs[n] = rd32(E1000_RDBAL(n));
322 for (n = 0; n < 4; n++)
323 regs[n] = rd32(E1000_RDBAH(n));
326 for (n = 0; n < 4; n++)
327 regs[n] = rd32(E1000_RDBAL(n));
330 for (n = 0; n < 4; n++)
331 regs[n] = rd32(E1000_TDBAH(n));
334 for (n = 0; n < 4; n++)
335 regs[n] = rd32(E1000_TDLEN(n));
338 for (n = 0; n < 4; n++)
339 regs[n] = rd32(E1000_TDH(n));
342 for (n = 0; n < 4; n++)
343 regs[n] = rd32(E1000_TDT(n));
345 case E1000_TXDCTL(0):
346 for (n = 0; n < 4; n++)
347 regs[n] = rd32(E1000_TXDCTL(n));
350 pr_info("%-15s %08x\n", reginfo->name, rd32(reginfo->ofs));
354 snprintf(rname, 16, "%s%s", reginfo->name, "[0-3]");
355 pr_info("%-15s %08x %08x %08x %08x\n", rname, regs[0], regs[1],
360 * igb_dump - Print registers, tx-rings and rx-rings
362 static void igb_dump(struct igb_adapter *adapter)
364 struct net_device *netdev = adapter->netdev;
365 struct e1000_hw *hw = &adapter->hw;
366 struct igb_reg_info *reginfo;
367 struct igb_ring *tx_ring;
368 union e1000_adv_tx_desc *tx_desc;
369 struct my_u0 { u64 a; u64 b; } *u0;
370 struct igb_ring *rx_ring;
371 union e1000_adv_rx_desc *rx_desc;
375 if (!netif_msg_hw(adapter))
378 /* Print netdevice Info */
380 dev_info(&adapter->pdev->dev, "Net device Info\n");
381 pr_info("Device Name state trans_start "
383 pr_info("%-15s %016lX %016lX %016lX\n", netdev->name,
384 netdev->state, netdev->trans_start, netdev->last_rx);
387 /* Print Registers */
388 dev_info(&adapter->pdev->dev, "Register Dump\n");
389 pr_info(" Register Name Value\n");
390 for (reginfo = (struct igb_reg_info *)igb_reg_info_tbl;
391 reginfo->name; reginfo++) {
392 igb_regdump(hw, reginfo);
395 /* Print TX Ring Summary */
396 if (!netdev || !netif_running(netdev))
399 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
400 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
401 for (n = 0; n < adapter->num_tx_queues; n++) {
402 struct igb_tx_buffer *buffer_info;
403 tx_ring = adapter->tx_ring[n];
404 buffer_info = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
405 pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
406 n, tx_ring->next_to_use, tx_ring->next_to_clean,
407 (u64)dma_unmap_addr(buffer_info, dma),
408 dma_unmap_len(buffer_info, len),
409 buffer_info->next_to_watch,
410 (u64)buffer_info->time_stamp);
414 if (!netif_msg_tx_done(adapter))
415 goto rx_ring_summary;
417 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
419 /* Transmit Descriptor Formats
421 * Advanced Transmit Descriptor
422 * +--------------------------------------------------------------+
423 * 0 | Buffer Address [63:0] |
424 * +--------------------------------------------------------------+
425 * 8 | PAYLEN | PORTS |CC|IDX | STA | DCMD |DTYP|MAC|RSV| DTALEN |
426 * +--------------------------------------------------------------+
427 * 63 46 45 40 39 38 36 35 32 31 24 15 0
430 for (n = 0; n < adapter->num_tx_queues; n++) {
431 tx_ring = adapter->tx_ring[n];
432 pr_info("------------------------------------\n");
433 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
434 pr_info("------------------------------------\n");
435 pr_info("T [desc] [address 63:0 ] [PlPOCIStDDM Ln] "
436 "[bi->dma ] leng ntw timestamp "
439 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
440 const char *next_desc;
441 struct igb_tx_buffer *buffer_info;
442 tx_desc = IGB_TX_DESC(tx_ring, i);
443 buffer_info = &tx_ring->tx_buffer_info[i];
444 u0 = (struct my_u0 *)tx_desc;
445 if (i == tx_ring->next_to_use &&
446 i == tx_ring->next_to_clean)
447 next_desc = " NTC/U";
448 else if (i == tx_ring->next_to_use)
450 else if (i == tx_ring->next_to_clean)
455 pr_info("T [0x%03X] %016llX %016llX %016llX"
456 " %04X %p %016llX %p%s\n", i,
459 (u64)dma_unmap_addr(buffer_info, dma),
460 dma_unmap_len(buffer_info, len),
461 buffer_info->next_to_watch,
462 (u64)buffer_info->time_stamp,
463 buffer_info->skb, next_desc);
465 if (netif_msg_pktdata(adapter) && buffer_info->skb)
466 print_hex_dump(KERN_INFO, "",
468 16, 1, buffer_info->skb->data,
469 dma_unmap_len(buffer_info, len),
474 /* Print RX Rings Summary */
476 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
477 pr_info("Queue [NTU] [NTC]\n");
478 for (n = 0; n < adapter->num_rx_queues; n++) {
479 rx_ring = adapter->rx_ring[n];
480 pr_info(" %5d %5X %5X\n",
481 n, rx_ring->next_to_use, rx_ring->next_to_clean);
485 if (!netif_msg_rx_status(adapter))
488 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
490 /* Advanced Receive Descriptor (Read) Format
492 * +-----------------------------------------------------+
493 * 0 | Packet Buffer Address [63:1] |A0/NSE|
494 * +----------------------------------------------+------+
495 * 8 | Header Buffer Address [63:1] | DD |
496 * +-----------------------------------------------------+
499 * Advanced Receive Descriptor (Write-Back) Format
501 * 63 48 47 32 31 30 21 20 17 16 4 3 0
502 * +------------------------------------------------------+
503 * 0 | Packet IP |SPH| HDR_LEN | RSV|Packet| RSS |
504 * | Checksum Ident | | | | Type | Type |
505 * +------------------------------------------------------+
506 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
507 * +------------------------------------------------------+
508 * 63 48 47 32 31 20 19 0
511 for (n = 0; n < adapter->num_rx_queues; n++) {
512 rx_ring = adapter->rx_ring[n];
513 pr_info("------------------------------------\n");
514 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
515 pr_info("------------------------------------\n");
516 pr_info("R [desc] [ PktBuf A0] [ HeadBuf DD] "
517 "[bi->dma ] [bi->skb] <-- Adv Rx Read format\n");
518 pr_info("RWB[desc] [PcsmIpSHl PtRs] [vl er S cks ln] -----"
519 "----------- [bi->skb] <-- Adv Rx Write-Back format\n");
521 for (i = 0; i < rx_ring->count; i++) {
522 const char *next_desc;
523 struct igb_rx_buffer *buffer_info;
524 buffer_info = &rx_ring->rx_buffer_info[i];
525 rx_desc = IGB_RX_DESC(rx_ring, i);
526 u0 = (struct my_u0 *)rx_desc;
527 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
529 if (i == rx_ring->next_to_use)
531 else if (i == rx_ring->next_to_clean)
536 if (staterr & E1000_RXD_STAT_DD) {
537 /* Descriptor Done */
538 pr_info("%s[0x%03X] %016llX %016llX ---------------- %s\n",
544 pr_info("%s[0x%03X] %016llX %016llX %016llX %s\n",
548 (u64)buffer_info->dma,
551 if (netif_msg_pktdata(adapter) &&
552 buffer_info->dma && buffer_info->page) {
553 print_hex_dump(KERN_INFO, "",
556 page_address(buffer_info->page) +
557 buffer_info->page_offset,
569 * igb_get_hw_dev - return device
570 * used by hardware layer to print debugging information
572 struct net_device *igb_get_hw_dev(struct e1000_hw *hw)
574 struct igb_adapter *adapter = hw->back;
575 return adapter->netdev;
579 * igb_init_module - Driver Registration Routine
581 * igb_init_module is the first routine called when the driver is
582 * loaded. All it does is register with the PCI subsystem.
584 static int __init igb_init_module(void)
587 pr_info("%s - version %s\n",
588 igb_driver_string, igb_driver_version);
590 pr_info("%s\n", igb_copyright);
592 #ifdef CONFIG_IGB_DCA
593 dca_register_notify(&dca_notifier);
595 ret = pci_register_driver(&igb_driver);
599 module_init(igb_init_module);
602 * igb_exit_module - Driver Exit Cleanup Routine
604 * igb_exit_module is called just before the driver is removed
607 static void __exit igb_exit_module(void)
609 #ifdef CONFIG_IGB_DCA
610 dca_unregister_notify(&dca_notifier);
612 pci_unregister_driver(&igb_driver);
615 module_exit(igb_exit_module);
617 #define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
619 * igb_cache_ring_register - Descriptor ring to register mapping
620 * @adapter: board private structure to initialize
622 * Once we know the feature-set enabled for the device, we'll cache
623 * the register offset the descriptor ring is assigned to.
625 static void igb_cache_ring_register(struct igb_adapter *adapter)
628 u32 rbase_offset = adapter->vfs_allocated_count;
630 switch (adapter->hw.mac.type) {
632 /* The queues are allocated for virtualization such that VF 0
633 * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
634 * In order to avoid collision we start at the first free queue
635 * and continue consuming queues in the same sequence
637 if (adapter->vfs_allocated_count) {
638 for (; i < adapter->rss_queues; i++)
639 adapter->rx_ring[i]->reg_idx = rbase_offset +
648 for (; i < adapter->num_rx_queues; i++)
649 adapter->rx_ring[i]->reg_idx = rbase_offset + i;
650 for (; j < adapter->num_tx_queues; j++)
651 adapter->tx_ring[j]->reg_idx = rbase_offset + j;
657 * igb_write_ivar - configure ivar for given MSI-X vector
658 * @hw: pointer to the HW structure
659 * @msix_vector: vector number we are allocating to a given ring
660 * @index: row index of IVAR register to write within IVAR table
661 * @offset: column offset of in IVAR, should be multiple of 8
663 * This function is intended to handle the writing of the IVAR register
664 * for adapters 82576 and newer. The IVAR table consists of 2 columns,
665 * each containing an cause allocation for an Rx and Tx ring, and a
666 * variable number of rows depending on the number of queues supported.
668 static void igb_write_ivar(struct e1000_hw *hw, int msix_vector,
669 int index, int offset)
671 u32 ivar = array_rd32(E1000_IVAR0, index);
673 /* clear any bits that are currently set */
674 ivar &= ~((u32)0xFF << offset);
676 /* write vector and valid bit */
677 ivar |= (msix_vector | E1000_IVAR_VALID) << offset;
679 array_wr32(E1000_IVAR0, index, ivar);
682 #define IGB_N0_QUEUE -1
683 static void igb_assign_vector(struct igb_q_vector *q_vector, int msix_vector)
685 struct igb_adapter *adapter = q_vector->adapter;
686 struct e1000_hw *hw = &adapter->hw;
687 int rx_queue = IGB_N0_QUEUE;
688 int tx_queue = IGB_N0_QUEUE;
691 if (q_vector->rx.ring)
692 rx_queue = q_vector->rx.ring->reg_idx;
693 if (q_vector->tx.ring)
694 tx_queue = q_vector->tx.ring->reg_idx;
696 switch (hw->mac.type) {
698 /* The 82575 assigns vectors using a bitmask, which matches the
699 bitmask for the EICR/EIMS/EIMC registers. To assign one
700 or more queues to a vector, we write the appropriate bits
701 into the MSIXBM register for that vector. */
702 if (rx_queue > IGB_N0_QUEUE)
703 msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
704 if (tx_queue > IGB_N0_QUEUE)
705 msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
706 if (!adapter->msix_entries && msix_vector == 0)
707 msixbm |= E1000_EIMS_OTHER;
708 array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
709 q_vector->eims_value = msixbm;
713 * 82576 uses a table that essentially consists of 2 columns
714 * with 8 rows. The ordering is column-major so we use the
715 * lower 3 bits as the row index, and the 4th bit as the
718 if (rx_queue > IGB_N0_QUEUE)
719 igb_write_ivar(hw, msix_vector,
721 (rx_queue & 0x8) << 1);
722 if (tx_queue > IGB_N0_QUEUE)
723 igb_write_ivar(hw, msix_vector,
725 ((tx_queue & 0x8) << 1) + 8);
726 q_vector->eims_value = 1 << msix_vector;
733 * On 82580 and newer adapters the scheme is similar to 82576
734 * however instead of ordering column-major we have things
735 * ordered row-major. So we traverse the table by using
736 * bit 0 as the column offset, and the remaining bits as the
739 if (rx_queue > IGB_N0_QUEUE)
740 igb_write_ivar(hw, msix_vector,
742 (rx_queue & 0x1) << 4);
743 if (tx_queue > IGB_N0_QUEUE)
744 igb_write_ivar(hw, msix_vector,
746 ((tx_queue & 0x1) << 4) + 8);
747 q_vector->eims_value = 1 << msix_vector;
754 /* add q_vector eims value to global eims_enable_mask */
755 adapter->eims_enable_mask |= q_vector->eims_value;
757 /* configure q_vector to set itr on first interrupt */
758 q_vector->set_itr = 1;
762 * igb_configure_msix - Configure MSI-X hardware
764 * igb_configure_msix sets up the hardware to properly
765 * generate MSI-X interrupts.
767 static void igb_configure_msix(struct igb_adapter *adapter)
771 struct e1000_hw *hw = &adapter->hw;
773 adapter->eims_enable_mask = 0;
775 /* set vector for other causes, i.e. link changes */
776 switch (hw->mac.type) {
778 tmp = rd32(E1000_CTRL_EXT);
779 /* enable MSI-X PBA support*/
780 tmp |= E1000_CTRL_EXT_PBA_CLR;
782 /* Auto-Mask interrupts upon ICR read. */
783 tmp |= E1000_CTRL_EXT_EIAME;
784 tmp |= E1000_CTRL_EXT_IRCA;
786 wr32(E1000_CTRL_EXT, tmp);
788 /* enable msix_other interrupt */
789 array_wr32(E1000_MSIXBM(0), vector++,
791 adapter->eims_other = E1000_EIMS_OTHER;
800 /* Turn on MSI-X capability first, or our settings
801 * won't stick. And it will take days to debug. */
802 wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
803 E1000_GPIE_PBA | E1000_GPIE_EIAME |
806 /* enable msix_other interrupt */
807 adapter->eims_other = 1 << vector;
808 tmp = (vector++ | E1000_IVAR_VALID) << 8;
810 wr32(E1000_IVAR_MISC, tmp);
813 /* do nothing, since nothing else supports MSI-X */
815 } /* switch (hw->mac.type) */
817 adapter->eims_enable_mask |= adapter->eims_other;
819 for (i = 0; i < adapter->num_q_vectors; i++)
820 igb_assign_vector(adapter->q_vector[i], vector++);
826 * igb_request_msix - Initialize MSI-X interrupts
828 * igb_request_msix allocates MSI-X vectors and requests interrupts from the
831 static int igb_request_msix(struct igb_adapter *adapter)
833 struct net_device *netdev = adapter->netdev;
834 struct e1000_hw *hw = &adapter->hw;
835 int i, err = 0, vector = 0, free_vector = 0;
837 err = request_irq(adapter->msix_entries[vector].vector,
838 igb_msix_other, 0, netdev->name, adapter);
842 for (i = 0; i < adapter->num_q_vectors; i++) {
843 struct igb_q_vector *q_vector = adapter->q_vector[i];
847 q_vector->itr_register = hw->hw_addr + E1000_EITR(vector);
849 if (q_vector->rx.ring && q_vector->tx.ring)
850 sprintf(q_vector->name, "%s-TxRx-%u", netdev->name,
851 q_vector->rx.ring->queue_index);
852 else if (q_vector->tx.ring)
853 sprintf(q_vector->name, "%s-tx-%u", netdev->name,
854 q_vector->tx.ring->queue_index);
855 else if (q_vector->rx.ring)
856 sprintf(q_vector->name, "%s-rx-%u", netdev->name,
857 q_vector->rx.ring->queue_index);
859 sprintf(q_vector->name, "%s-unused", netdev->name);
861 err = request_irq(adapter->msix_entries[vector].vector,
862 igb_msix_ring, 0, q_vector->name,
868 igb_configure_msix(adapter);
872 /* free already assigned IRQs */
873 free_irq(adapter->msix_entries[free_vector++].vector, adapter);
876 for (i = 0; i < vector; i++) {
877 free_irq(adapter->msix_entries[free_vector++].vector,
878 adapter->q_vector[i]);
884 static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
886 if (adapter->msix_entries) {
887 pci_disable_msix(adapter->pdev);
888 kfree(adapter->msix_entries);
889 adapter->msix_entries = NULL;
890 } else if (adapter->flags & IGB_FLAG_HAS_MSI) {
891 pci_disable_msi(adapter->pdev);
896 * igb_free_q_vector - Free memory allocated for specific interrupt vector
897 * @adapter: board private structure to initialize
898 * @v_idx: Index of vector to be freed
900 * This function frees the memory allocated to the q_vector. In addition if
901 * NAPI is enabled it will delete any references to the NAPI struct prior
902 * to freeing the q_vector.
904 static void igb_free_q_vector(struct igb_adapter *adapter, int v_idx)
906 struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
908 if (q_vector->tx.ring)
909 adapter->tx_ring[q_vector->tx.ring->queue_index] = NULL;
911 if (q_vector->rx.ring)
912 adapter->tx_ring[q_vector->rx.ring->queue_index] = NULL;
914 adapter->q_vector[v_idx] = NULL;
915 netif_napi_del(&q_vector->napi);
918 * ixgbe_get_stats64() might access the rings on this vector,
919 * we must wait a grace period before freeing it.
921 kfree_rcu(q_vector, rcu);
925 * igb_free_q_vectors - Free memory allocated for interrupt vectors
926 * @adapter: board private structure to initialize
928 * This function frees the memory allocated to the q_vectors. In addition if
929 * NAPI is enabled it will delete any references to the NAPI struct prior
930 * to freeing the q_vector.
932 static void igb_free_q_vectors(struct igb_adapter *adapter)
934 int v_idx = adapter->num_q_vectors;
936 adapter->num_tx_queues = 0;
937 adapter->num_rx_queues = 0;
938 adapter->num_q_vectors = 0;
941 igb_free_q_vector(adapter, v_idx);
945 * igb_clear_interrupt_scheme - reset the device to a state of no interrupts
947 * This function resets the device so that it has 0 rx queues, tx queues, and
948 * MSI-X interrupts allocated.
950 static void igb_clear_interrupt_scheme(struct igb_adapter *adapter)
952 igb_free_q_vectors(adapter);
953 igb_reset_interrupt_capability(adapter);
957 * igb_set_interrupt_capability - set MSI or MSI-X if supported
959 * Attempt to configure interrupts using the best available
960 * capabilities of the hardware and kernel.
962 static void igb_set_interrupt_capability(struct igb_adapter *adapter, bool msix)
970 /* Number of supported queues. */
971 adapter->num_rx_queues = adapter->rss_queues;
972 if (adapter->vfs_allocated_count)
973 adapter->num_tx_queues = 1;
975 adapter->num_tx_queues = adapter->rss_queues;
977 /* start with one vector for every rx queue */
978 numvecs = adapter->num_rx_queues;
980 /* if tx handler is separate add 1 for every tx queue */
981 if (!(adapter->flags & IGB_FLAG_QUEUE_PAIRS))
982 numvecs += adapter->num_tx_queues;
984 /* store the number of vectors reserved for queues */
985 adapter->num_q_vectors = numvecs;
987 /* add 1 vector for link status interrupts */
989 adapter->msix_entries = kcalloc(numvecs, sizeof(struct msix_entry),
992 if (!adapter->msix_entries)
995 for (i = 0; i < numvecs; i++)
996 adapter->msix_entries[i].entry = i;
998 err = pci_enable_msix(adapter->pdev,
999 adapter->msix_entries,
1004 igb_reset_interrupt_capability(adapter);
1006 /* If we can't do MSI-X, try MSI */
1008 #ifdef CONFIG_PCI_IOV
1009 /* disable SR-IOV for non MSI-X configurations */
1010 if (adapter->vf_data) {
1011 struct e1000_hw *hw = &adapter->hw;
1012 /* disable iov and allow time for transactions to clear */
1013 pci_disable_sriov(adapter->pdev);
1016 kfree(adapter->vf_data);
1017 adapter->vf_data = NULL;
1018 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
1021 dev_info(&adapter->pdev->dev, "IOV Disabled\n");
1024 adapter->vfs_allocated_count = 0;
1025 adapter->rss_queues = 1;
1026 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
1027 adapter->num_rx_queues = 1;
1028 adapter->num_tx_queues = 1;
1029 adapter->num_q_vectors = 1;
1030 if (!pci_enable_msi(adapter->pdev))
1031 adapter->flags |= IGB_FLAG_HAS_MSI;
1034 static void igb_add_ring(struct igb_ring *ring,
1035 struct igb_ring_container *head)
1042 * igb_alloc_q_vector - Allocate memory for a single interrupt vector
1043 * @adapter: board private structure to initialize
1044 * @v_count: q_vectors allocated on adapter, used for ring interleaving
1045 * @v_idx: index of vector in adapter struct
1046 * @txr_count: total number of Tx rings to allocate
1047 * @txr_idx: index of first Tx ring to allocate
1048 * @rxr_count: total number of Rx rings to allocate
1049 * @rxr_idx: index of first Rx ring to allocate
1051 * We allocate one q_vector. If allocation fails we return -ENOMEM.
1053 static int igb_alloc_q_vector(struct igb_adapter *adapter,
1054 int v_count, int v_idx,
1055 int txr_count, int txr_idx,
1056 int rxr_count, int rxr_idx)
1058 struct igb_q_vector *q_vector;
1059 struct igb_ring *ring;
1060 int ring_count, size;
1062 /* igb only supports 1 Tx and/or 1 Rx queue per vector */
1063 if (txr_count > 1 || rxr_count > 1)
1066 ring_count = txr_count + rxr_count;
1067 size = sizeof(struct igb_q_vector) +
1068 (sizeof(struct igb_ring) * ring_count);
1070 /* allocate q_vector and rings */
1071 q_vector = kzalloc(size, GFP_KERNEL);
1075 /* initialize NAPI */
1076 netif_napi_add(adapter->netdev, &q_vector->napi,
1079 /* tie q_vector and adapter together */
1080 adapter->q_vector[v_idx] = q_vector;
1081 q_vector->adapter = adapter;
1083 /* initialize work limits */
1084 q_vector->tx.work_limit = adapter->tx_work_limit;
1086 /* initialize ITR configuration */
1087 q_vector->itr_register = adapter->hw.hw_addr + E1000_EITR(0);
1088 q_vector->itr_val = IGB_START_ITR;
1090 /* initialize pointer to rings */
1091 ring = q_vector->ring;
1094 /* assign generic ring traits */
1095 ring->dev = &adapter->pdev->dev;
1096 ring->netdev = adapter->netdev;
1098 /* configure backlink on ring */
1099 ring->q_vector = q_vector;
1101 /* update q_vector Tx values */
1102 igb_add_ring(ring, &q_vector->tx);
1104 /* For 82575, context index must be unique per ring. */
1105 if (adapter->hw.mac.type == e1000_82575)
1106 set_bit(IGB_RING_FLAG_TX_CTX_IDX, &ring->flags);
1108 /* apply Tx specific ring traits */
1109 ring->count = adapter->tx_ring_count;
1110 ring->queue_index = txr_idx;
1112 /* assign ring to adapter */
1113 adapter->tx_ring[txr_idx] = ring;
1115 /* push pointer to next ring */
1120 /* assign generic ring traits */
1121 ring->dev = &adapter->pdev->dev;
1122 ring->netdev = adapter->netdev;
1124 /* configure backlink on ring */
1125 ring->q_vector = q_vector;
1127 /* update q_vector Rx values */
1128 igb_add_ring(ring, &q_vector->rx);
1130 /* set flag indicating ring supports SCTP checksum offload */
1131 if (adapter->hw.mac.type >= e1000_82576)
1132 set_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags);
1135 * On i350, i210, and i211, loopback VLAN packets
1136 * have the tag byte-swapped.
1138 if (adapter->hw.mac.type >= e1000_i350)
1139 set_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &ring->flags);
1141 /* apply Rx specific ring traits */
1142 ring->count = adapter->rx_ring_count;
1143 ring->queue_index = rxr_idx;
1145 /* assign ring to adapter */
1146 adapter->rx_ring[rxr_idx] = ring;
1154 * igb_alloc_q_vectors - Allocate memory for interrupt vectors
1155 * @adapter: board private structure to initialize
1157 * We allocate one q_vector per queue interrupt. If allocation fails we
1160 static int igb_alloc_q_vectors(struct igb_adapter *adapter)
1162 int q_vectors = adapter->num_q_vectors;
1163 int rxr_remaining = adapter->num_rx_queues;
1164 int txr_remaining = adapter->num_tx_queues;
1165 int rxr_idx = 0, txr_idx = 0, v_idx = 0;
1168 if (q_vectors >= (rxr_remaining + txr_remaining)) {
1169 for (; rxr_remaining; v_idx++) {
1170 err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1176 /* update counts and index */
1182 for (; v_idx < q_vectors; v_idx++) {
1183 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_idx);
1184 int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_idx);
1185 err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1186 tqpv, txr_idx, rqpv, rxr_idx);
1191 /* update counts and index */
1192 rxr_remaining -= rqpv;
1193 txr_remaining -= tqpv;
1201 adapter->num_tx_queues = 0;
1202 adapter->num_rx_queues = 0;
1203 adapter->num_q_vectors = 0;
1206 igb_free_q_vector(adapter, v_idx);
1212 * igb_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
1214 * This function initializes the interrupts and allocates all of the queues.
1216 static int igb_init_interrupt_scheme(struct igb_adapter *adapter, bool msix)
1218 struct pci_dev *pdev = adapter->pdev;
1221 igb_set_interrupt_capability(adapter, msix);
1223 err = igb_alloc_q_vectors(adapter);
1225 dev_err(&pdev->dev, "Unable to allocate memory for vectors\n");
1226 goto err_alloc_q_vectors;
1229 igb_cache_ring_register(adapter);
1233 err_alloc_q_vectors:
1234 igb_reset_interrupt_capability(adapter);
1239 * igb_request_irq - initialize interrupts
1241 * Attempts to configure interrupts using the best available
1242 * capabilities of the hardware and kernel.
1244 static int igb_request_irq(struct igb_adapter *adapter)
1246 struct net_device *netdev = adapter->netdev;
1247 struct pci_dev *pdev = adapter->pdev;
1250 if (adapter->msix_entries) {
1251 err = igb_request_msix(adapter);
1254 /* fall back to MSI */
1255 igb_free_all_tx_resources(adapter);
1256 igb_free_all_rx_resources(adapter);
1258 igb_clear_interrupt_scheme(adapter);
1259 err = igb_init_interrupt_scheme(adapter, false);
1263 igb_setup_all_tx_resources(adapter);
1264 igb_setup_all_rx_resources(adapter);
1265 igb_configure(adapter);
1268 igb_assign_vector(adapter->q_vector[0], 0);
1270 if (adapter->flags & IGB_FLAG_HAS_MSI) {
1271 err = request_irq(pdev->irq, igb_intr_msi, 0,
1272 netdev->name, adapter);
1276 /* fall back to legacy interrupts */
1277 igb_reset_interrupt_capability(adapter);
1278 adapter->flags &= ~IGB_FLAG_HAS_MSI;
1281 err = request_irq(pdev->irq, igb_intr, IRQF_SHARED,
1282 netdev->name, adapter);
1285 dev_err(&pdev->dev, "Error %d getting interrupt\n",
1292 static void igb_free_irq(struct igb_adapter *adapter)
1294 if (adapter->msix_entries) {
1297 free_irq(adapter->msix_entries[vector++].vector, adapter);
1299 for (i = 0; i < adapter->num_q_vectors; i++)
1300 free_irq(adapter->msix_entries[vector++].vector,
1301 adapter->q_vector[i]);
1303 free_irq(adapter->pdev->irq, adapter);
1308 * igb_irq_disable - Mask off interrupt generation on the NIC
1309 * @adapter: board private structure
1311 static void igb_irq_disable(struct igb_adapter *adapter)
1313 struct e1000_hw *hw = &adapter->hw;
1316 * we need to be careful when disabling interrupts. The VFs are also
1317 * mapped into these registers and so clearing the bits can cause
1318 * issues on the VF drivers so we only need to clear what we set
1320 if (adapter->msix_entries) {
1321 u32 regval = rd32(E1000_EIAM);
1322 wr32(E1000_EIAM, regval & ~adapter->eims_enable_mask);
1323 wr32(E1000_EIMC, adapter->eims_enable_mask);
1324 regval = rd32(E1000_EIAC);
1325 wr32(E1000_EIAC, regval & ~adapter->eims_enable_mask);
1329 wr32(E1000_IMC, ~0);
1331 if (adapter->msix_entries) {
1333 for (i = 0; i < adapter->num_q_vectors; i++)
1334 synchronize_irq(adapter->msix_entries[i].vector);
1336 synchronize_irq(adapter->pdev->irq);
1341 * igb_irq_enable - Enable default interrupt generation settings
1342 * @adapter: board private structure
1344 static void igb_irq_enable(struct igb_adapter *adapter)
1346 struct e1000_hw *hw = &adapter->hw;
1348 if (adapter->msix_entries) {
1349 u32 ims = E1000_IMS_LSC | E1000_IMS_DOUTSYNC | E1000_IMS_DRSTA;
1350 u32 regval = rd32(E1000_EIAC);
1351 wr32(E1000_EIAC, regval | adapter->eims_enable_mask);
1352 regval = rd32(E1000_EIAM);
1353 wr32(E1000_EIAM, regval | adapter->eims_enable_mask);
1354 wr32(E1000_EIMS, adapter->eims_enable_mask);
1355 if (adapter->vfs_allocated_count) {
1356 wr32(E1000_MBVFIMR, 0xFF);
1357 ims |= E1000_IMS_VMMB;
1359 wr32(E1000_IMS, ims);
1361 wr32(E1000_IMS, IMS_ENABLE_MASK |
1363 wr32(E1000_IAM, IMS_ENABLE_MASK |
1368 static void igb_update_mng_vlan(struct igb_adapter *adapter)
1370 struct e1000_hw *hw = &adapter->hw;
1371 u16 vid = adapter->hw.mng_cookie.vlan_id;
1372 u16 old_vid = adapter->mng_vlan_id;
1374 if (hw->mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1375 /* add VID to filter table */
1376 igb_vfta_set(hw, vid, true);
1377 adapter->mng_vlan_id = vid;
1379 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1382 if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
1384 !test_bit(old_vid, adapter->active_vlans)) {
1385 /* remove VID from filter table */
1386 igb_vfta_set(hw, old_vid, false);
1391 * igb_release_hw_control - release control of the h/w to f/w
1392 * @adapter: address of board private structure
1394 * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
1395 * For ASF and Pass Through versions of f/w this means that the
1396 * driver is no longer loaded.
1399 static void igb_release_hw_control(struct igb_adapter *adapter)
1401 struct e1000_hw *hw = &adapter->hw;
1404 /* Let firmware take over control of h/w */
1405 ctrl_ext = rd32(E1000_CTRL_EXT);
1406 wr32(E1000_CTRL_EXT,
1407 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1411 * igb_get_hw_control - get control of the h/w from f/w
1412 * @adapter: address of board private structure
1414 * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
1415 * For ASF and Pass Through versions of f/w this means that
1416 * the driver is loaded.
1419 static void igb_get_hw_control(struct igb_adapter *adapter)
1421 struct e1000_hw *hw = &adapter->hw;
1424 /* Let firmware know the driver has taken over */
1425 ctrl_ext = rd32(E1000_CTRL_EXT);
1426 wr32(E1000_CTRL_EXT,
1427 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1431 * igb_configure - configure the hardware for RX and TX
1432 * @adapter: private board structure
1434 static void igb_configure(struct igb_adapter *adapter)
1436 struct net_device *netdev = adapter->netdev;
1439 igb_get_hw_control(adapter);
1440 igb_set_rx_mode(netdev);
1442 igb_restore_vlan(adapter);
1444 igb_setup_tctl(adapter);
1445 igb_setup_mrqc(adapter);
1446 igb_setup_rctl(adapter);
1448 igb_configure_tx(adapter);
1449 igb_configure_rx(adapter);
1451 igb_rx_fifo_flush_82575(&adapter->hw);
1453 /* call igb_desc_unused which always leaves
1454 * at least 1 descriptor unused to make sure
1455 * next_to_use != next_to_clean */
1456 for (i = 0; i < adapter->num_rx_queues; i++) {
1457 struct igb_ring *ring = adapter->rx_ring[i];
1458 igb_alloc_rx_buffers(ring, igb_desc_unused(ring));
1463 * igb_power_up_link - Power up the phy/serdes link
1464 * @adapter: address of board private structure
1466 void igb_power_up_link(struct igb_adapter *adapter)
1468 igb_reset_phy(&adapter->hw);
1470 if (adapter->hw.phy.media_type == e1000_media_type_copper)
1471 igb_power_up_phy_copper(&adapter->hw);
1473 igb_power_up_serdes_link_82575(&adapter->hw);
1477 * igb_power_down_link - Power down the phy/serdes link
1478 * @adapter: address of board private structure
1480 static void igb_power_down_link(struct igb_adapter *adapter)
1482 if (adapter->hw.phy.media_type == e1000_media_type_copper)
1483 igb_power_down_phy_copper_82575(&adapter->hw);
1485 igb_shutdown_serdes_link_82575(&adapter->hw);
1489 * igb_up - Open the interface and prepare it to handle traffic
1490 * @adapter: board private structure
1492 int igb_up(struct igb_adapter *adapter)
1494 struct e1000_hw *hw = &adapter->hw;
1497 /* hardware has been reset, we need to reload some things */
1498 igb_configure(adapter);
1500 clear_bit(__IGB_DOWN, &adapter->state);
1502 for (i = 0; i < adapter->num_q_vectors; i++)
1503 napi_enable(&(adapter->q_vector[i]->napi));
1505 if (adapter->msix_entries)
1506 igb_configure_msix(adapter);
1508 igb_assign_vector(adapter->q_vector[0], 0);
1510 /* Clear any pending interrupts. */
1512 igb_irq_enable(adapter);
1514 /* notify VFs that reset has been completed */
1515 if (adapter->vfs_allocated_count) {
1516 u32 reg_data = rd32(E1000_CTRL_EXT);
1517 reg_data |= E1000_CTRL_EXT_PFRSTD;
1518 wr32(E1000_CTRL_EXT, reg_data);
1521 netif_tx_start_all_queues(adapter->netdev);
1523 /* start the watchdog. */
1524 hw->mac.get_link_status = 1;
1525 schedule_work(&adapter->watchdog_task);
1530 void igb_down(struct igb_adapter *adapter)
1532 struct net_device *netdev = adapter->netdev;
1533 struct e1000_hw *hw = &adapter->hw;
1537 /* signal that we're down so the interrupt handler does not
1538 * reschedule our watchdog timer */
1539 set_bit(__IGB_DOWN, &adapter->state);
1541 /* disable receives in the hardware */
1542 rctl = rd32(E1000_RCTL);
1543 wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
1544 /* flush and sleep below */
1546 netif_tx_stop_all_queues(netdev);
1548 /* disable transmits in the hardware */
1549 tctl = rd32(E1000_TCTL);
1550 tctl &= ~E1000_TCTL_EN;
1551 wr32(E1000_TCTL, tctl);
1552 /* flush both disables and wait for them to finish */
1556 for (i = 0; i < adapter->num_q_vectors; i++)
1557 napi_disable(&(adapter->q_vector[i]->napi));
1559 igb_irq_disable(adapter);
1561 del_timer_sync(&adapter->watchdog_timer);
1562 del_timer_sync(&adapter->phy_info_timer);
1564 netif_carrier_off(netdev);
1566 /* record the stats before reset*/
1567 spin_lock(&adapter->stats64_lock);
1568 igb_update_stats(adapter, &adapter->stats64);
1569 spin_unlock(&adapter->stats64_lock);
1571 adapter->link_speed = 0;
1572 adapter->link_duplex = 0;
1574 if (!pci_channel_offline(adapter->pdev))
1576 igb_clean_all_tx_rings(adapter);
1577 igb_clean_all_rx_rings(adapter);
1578 #ifdef CONFIG_IGB_DCA
1580 /* since we reset the hardware DCA settings were cleared */
1581 igb_setup_dca(adapter);
1585 void igb_reinit_locked(struct igb_adapter *adapter)
1587 WARN_ON(in_interrupt());
1588 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
1592 clear_bit(__IGB_RESETTING, &adapter->state);
1595 void igb_reset(struct igb_adapter *adapter)
1597 struct pci_dev *pdev = adapter->pdev;
1598 struct e1000_hw *hw = &adapter->hw;
1599 struct e1000_mac_info *mac = &hw->mac;
1600 struct e1000_fc_info *fc = &hw->fc;
1601 u32 pba = 0, tx_space, min_tx_space, min_rx_space, hwm;
1603 /* Repartition Pba for greater than 9k mtu
1604 * To take effect CTRL.RST is required.
1606 switch (mac->type) {
1609 pba = rd32(E1000_RXPBS);
1610 pba = igb_rxpbs_adjust_82580(pba);
1613 pba = rd32(E1000_RXPBS);
1614 pba &= E1000_RXPBS_SIZE_MASK_82576;
1620 pba = E1000_PBA_34K;
1624 if ((adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) &&
1625 (mac->type < e1000_82576)) {
1626 /* adjust PBA for jumbo frames */
1627 wr32(E1000_PBA, pba);
1629 /* To maintain wire speed transmits, the Tx FIFO should be
1630 * large enough to accommodate two full transmit packets,
1631 * rounded up to the next 1KB and expressed in KB. Likewise,
1632 * the Rx FIFO should be large enough to accommodate at least
1633 * one full receive packet and is similarly rounded up and
1634 * expressed in KB. */
1635 pba = rd32(E1000_PBA);
1636 /* upper 16 bits has Tx packet buffer allocation size in KB */
1637 tx_space = pba >> 16;
1638 /* lower 16 bits has Rx packet buffer allocation size in KB */
1640 /* the tx fifo also stores 16 bytes of information about the tx
1641 * but don't include ethernet FCS because hardware appends it */
1642 min_tx_space = (adapter->max_frame_size +
1643 sizeof(union e1000_adv_tx_desc) -
1645 min_tx_space = ALIGN(min_tx_space, 1024);
1646 min_tx_space >>= 10;
1647 /* software strips receive CRC, so leave room for it */
1648 min_rx_space = adapter->max_frame_size;
1649 min_rx_space = ALIGN(min_rx_space, 1024);
1650 min_rx_space >>= 10;
1652 /* If current Tx allocation is less than the min Tx FIFO size,
1653 * and the min Tx FIFO size is less than the current Rx FIFO
1654 * allocation, take space away from current Rx allocation */
1655 if (tx_space < min_tx_space &&
1656 ((min_tx_space - tx_space) < pba)) {
1657 pba = pba - (min_tx_space - tx_space);
1659 /* if short on rx space, rx wins and must trump tx
1661 if (pba < min_rx_space)
1664 wr32(E1000_PBA, pba);
1667 /* flow control settings */
1668 /* The high water mark must be low enough to fit one full frame
1669 * (or the size used for early receive) above it in the Rx FIFO.
1670 * Set it to the lower of:
1671 * - 90% of the Rx FIFO size, or
1672 * - the full Rx FIFO size minus one full frame */
1673 hwm = min(((pba << 10) * 9 / 10),
1674 ((pba << 10) - 2 * adapter->max_frame_size));
1676 fc->high_water = hwm & 0xFFFFFFF0; /* 16-byte granularity */
1677 fc->low_water = fc->high_water - 16;
1678 fc->pause_time = 0xFFFF;
1680 fc->current_mode = fc->requested_mode;
1682 /* disable receive for all VFs and wait one second */
1683 if (adapter->vfs_allocated_count) {
1685 for (i = 0 ; i < adapter->vfs_allocated_count; i++)
1686 adapter->vf_data[i].flags &= IGB_VF_FLAG_PF_SET_MAC;
1688 /* ping all the active vfs to let them know we are going down */
1689 igb_ping_all_vfs(adapter);
1691 /* disable transmits and receives */
1692 wr32(E1000_VFRE, 0);
1693 wr32(E1000_VFTE, 0);
1696 /* Allow time for pending master requests to run */
1697 hw->mac.ops.reset_hw(hw);
1700 if (hw->mac.ops.init_hw(hw))
1701 dev_err(&pdev->dev, "Hardware Error\n");
1704 * Flow control settings reset on hardware reset, so guarantee flow
1705 * control is off when forcing speed.
1707 if (!hw->mac.autoneg)
1708 igb_force_mac_fc(hw);
1710 igb_init_dmac(adapter, pba);
1711 if (!netif_running(adapter->netdev))
1712 igb_power_down_link(adapter);
1714 igb_update_mng_vlan(adapter);
1716 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
1717 wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
1719 /* Re-enable PTP, where applicable. */
1720 igb_ptp_reset(adapter);
1722 igb_get_phy_info(hw);
1725 static netdev_features_t igb_fix_features(struct net_device *netdev,
1726 netdev_features_t features)
1729 * Since there is no support for separate rx/tx vlan accel
1730 * enable/disable make sure tx flag is always in same state as rx.
1732 if (features & NETIF_F_HW_VLAN_RX)
1733 features |= NETIF_F_HW_VLAN_TX;
1735 features &= ~NETIF_F_HW_VLAN_TX;
1740 static int igb_set_features(struct net_device *netdev,
1741 netdev_features_t features)
1743 netdev_features_t changed = netdev->features ^ features;
1744 struct igb_adapter *adapter = netdev_priv(netdev);
1746 if (changed & NETIF_F_HW_VLAN_RX)
1747 igb_vlan_mode(netdev, features);
1749 if (!(changed & NETIF_F_RXALL))
1752 netdev->features = features;
1754 if (netif_running(netdev))
1755 igb_reinit_locked(adapter);
1762 static const struct net_device_ops igb_netdev_ops = {
1763 .ndo_open = igb_open,
1764 .ndo_stop = igb_close,
1765 .ndo_start_xmit = igb_xmit_frame,
1766 .ndo_get_stats64 = igb_get_stats64,
1767 .ndo_set_rx_mode = igb_set_rx_mode,
1768 .ndo_set_mac_address = igb_set_mac,
1769 .ndo_change_mtu = igb_change_mtu,
1770 .ndo_do_ioctl = igb_ioctl,
1771 .ndo_tx_timeout = igb_tx_timeout,
1772 .ndo_validate_addr = eth_validate_addr,
1773 .ndo_vlan_rx_add_vid = igb_vlan_rx_add_vid,
1774 .ndo_vlan_rx_kill_vid = igb_vlan_rx_kill_vid,
1775 .ndo_set_vf_mac = igb_ndo_set_vf_mac,
1776 .ndo_set_vf_vlan = igb_ndo_set_vf_vlan,
1777 .ndo_set_vf_tx_rate = igb_ndo_set_vf_bw,
1778 .ndo_get_vf_config = igb_ndo_get_vf_config,
1779 #ifdef CONFIG_NET_POLL_CONTROLLER
1780 .ndo_poll_controller = igb_netpoll,
1782 .ndo_fix_features = igb_fix_features,
1783 .ndo_set_features = igb_set_features,
1787 * igb_set_fw_version - Configure version string for ethtool
1788 * @adapter: adapter struct
1791 void igb_set_fw_version(struct igb_adapter *adapter)
1793 struct e1000_hw *hw = &adapter->hw;
1794 struct e1000_fw_version fw;
1796 igb_get_fw_version(hw, &fw);
1798 switch (hw->mac.type) {
1800 snprintf(adapter->fw_version, sizeof(adapter->fw_version),
1802 fw.invm_major, fw.invm_minor, fw.invm_img_type);
1806 /* if option is rom valid, display its version too */
1808 snprintf(adapter->fw_version,
1809 sizeof(adapter->fw_version),
1810 "%d.%d, 0x%08x, %d.%d.%d",
1811 fw.eep_major, fw.eep_minor, fw.etrack_id,
1812 fw.or_major, fw.or_build, fw.or_patch);
1815 snprintf(adapter->fw_version,
1816 sizeof(adapter->fw_version),
1818 fw.eep_major, fw.eep_minor, fw.etrack_id);
1826 * igb_probe - Device Initialization Routine
1827 * @pdev: PCI device information struct
1828 * @ent: entry in igb_pci_tbl
1830 * Returns 0 on success, negative on failure
1832 * igb_probe initializes an adapter identified by a pci_dev structure.
1833 * The OS initialization, configuring of the adapter private structure,
1834 * and a hardware reset occur.
1836 static int igb_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
1838 struct net_device *netdev;
1839 struct igb_adapter *adapter;
1840 struct e1000_hw *hw;
1841 u16 eeprom_data = 0;
1843 static int global_quad_port_a; /* global quad port a indication */
1844 const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
1845 unsigned long mmio_start, mmio_len;
1846 int err, pci_using_dac;
1847 u8 part_str[E1000_PBANUM_LENGTH];
1849 /* Catch broken hardware that put the wrong VF device ID in
1850 * the PCIe SR-IOV capability.
1852 if (pdev->is_virtfn) {
1853 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
1854 pci_name(pdev), pdev->vendor, pdev->device);
1858 err = pci_enable_device_mem(pdev);
1863 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
1865 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
1869 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
1871 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(32));
1873 dev_err(&pdev->dev, "No usable DMA "
1874 "configuration, aborting\n");
1880 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
1886 pci_enable_pcie_error_reporting(pdev);
1888 pci_set_master(pdev);
1889 pci_save_state(pdev);
1892 netdev = alloc_etherdev_mq(sizeof(struct igb_adapter),
1895 goto err_alloc_etherdev;
1897 SET_NETDEV_DEV(netdev, &pdev->dev);
1899 pci_set_drvdata(pdev, netdev);
1900 adapter = netdev_priv(netdev);
1901 adapter->netdev = netdev;
1902 adapter->pdev = pdev;
1905 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
1907 mmio_start = pci_resource_start(pdev, 0);
1908 mmio_len = pci_resource_len(pdev, 0);
1911 hw->hw_addr = ioremap(mmio_start, mmio_len);
1915 netdev->netdev_ops = &igb_netdev_ops;
1916 igb_set_ethtool_ops(netdev);
1917 netdev->watchdog_timeo = 5 * HZ;
1919 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
1921 netdev->mem_start = mmio_start;
1922 netdev->mem_end = mmio_start + mmio_len;
1924 /* PCI config space info */
1925 hw->vendor_id = pdev->vendor;
1926 hw->device_id = pdev->device;
1927 hw->revision_id = pdev->revision;
1928 hw->subsystem_vendor_id = pdev->subsystem_vendor;
1929 hw->subsystem_device_id = pdev->subsystem_device;
1931 /* Copy the default MAC, PHY and NVM function pointers */
1932 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
1933 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
1934 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
1935 /* Initialize skew-specific constants */
1936 err = ei->get_invariants(hw);
1940 /* setup the private structure */
1941 err = igb_sw_init(adapter);
1945 igb_get_bus_info_pcie(hw);
1947 hw->phy.autoneg_wait_to_complete = false;
1949 /* Copper options */
1950 if (hw->phy.media_type == e1000_media_type_copper) {
1951 hw->phy.mdix = AUTO_ALL_MODES;
1952 hw->phy.disable_polarity_correction = false;
1953 hw->phy.ms_type = e1000_ms_hw_default;
1956 if (igb_check_reset_block(hw))
1957 dev_info(&pdev->dev,
1958 "PHY reset is blocked due to SOL/IDER session.\n");
1961 * features is initialized to 0 in allocation, it might have bits
1962 * set by igb_sw_init so we should use an or instead of an
1965 netdev->features |= NETIF_F_SG |
1972 NETIF_F_HW_VLAN_RX |
1975 /* copy netdev features into list of user selectable features */
1976 netdev->hw_features |= netdev->features;
1977 netdev->hw_features |= NETIF_F_RXALL;
1979 /* set this bit last since it cannot be part of hw_features */
1980 netdev->features |= NETIF_F_HW_VLAN_FILTER;
1982 netdev->vlan_features |= NETIF_F_TSO |
1988 netdev->priv_flags |= IFF_SUPP_NOFCS;
1990 if (pci_using_dac) {
1991 netdev->features |= NETIF_F_HIGHDMA;
1992 netdev->vlan_features |= NETIF_F_HIGHDMA;
1995 if (hw->mac.type >= e1000_82576) {
1996 netdev->hw_features |= NETIF_F_SCTP_CSUM;
1997 netdev->features |= NETIF_F_SCTP_CSUM;
2000 netdev->priv_flags |= IFF_UNICAST_FLT;
2002 adapter->en_mng_pt = igb_enable_mng_pass_thru(hw);
2004 /* before reading the NVM, reset the controller to put the device in a
2005 * known good starting state */
2006 hw->mac.ops.reset_hw(hw);
2009 * make sure the NVM is good , i211 parts have special NVM that
2010 * doesn't contain a checksum
2012 if (hw->mac.type != e1000_i211) {
2013 if (hw->nvm.ops.validate(hw) < 0) {
2014 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
2020 /* copy the MAC address out of the NVM */
2021 if (hw->mac.ops.read_mac_addr(hw))
2022 dev_err(&pdev->dev, "NVM Read Error\n");
2024 memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
2026 if (!is_valid_ether_addr(netdev->dev_addr)) {
2027 dev_err(&pdev->dev, "Invalid MAC Address\n");
2032 /* get firmware version for ethtool -i */
2033 igb_set_fw_version(adapter);
2035 setup_timer(&adapter->watchdog_timer, igb_watchdog,
2036 (unsigned long) adapter);
2037 setup_timer(&adapter->phy_info_timer, igb_update_phy_info,
2038 (unsigned long) adapter);
2040 INIT_WORK(&adapter->reset_task, igb_reset_task);
2041 INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
2043 /* Initialize link properties that are user-changeable */
2044 adapter->fc_autoneg = true;
2045 hw->mac.autoneg = true;
2046 hw->phy.autoneg_advertised = 0x2f;
2048 hw->fc.requested_mode = e1000_fc_default;
2049 hw->fc.current_mode = e1000_fc_default;
2051 igb_validate_mdi_setting(hw);
2053 /* By default, support wake on port A */
2054 if (hw->bus.func == 0)
2055 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2057 /* Check the NVM for wake support on non-port A ports */
2058 if (hw->mac.type >= e1000_82580)
2059 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A +
2060 NVM_82580_LAN_FUNC_OFFSET(hw->bus.func), 1,
2062 else if (hw->bus.func == 1)
2063 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
2065 if (eeprom_data & IGB_EEPROM_APME)
2066 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2068 /* now that we have the eeprom settings, apply the special cases where
2069 * the eeprom may be wrong or the board simply won't support wake on
2070 * lan on a particular port */
2071 switch (pdev->device) {
2072 case E1000_DEV_ID_82575GB_QUAD_COPPER:
2073 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2075 case E1000_DEV_ID_82575EB_FIBER_SERDES:
2076 case E1000_DEV_ID_82576_FIBER:
2077 case E1000_DEV_ID_82576_SERDES:
2078 /* Wake events only supported on port A for dual fiber
2079 * regardless of eeprom setting */
2080 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
2081 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2083 case E1000_DEV_ID_82576_QUAD_COPPER:
2084 case E1000_DEV_ID_82576_QUAD_COPPER_ET2:
2085 /* if quad port adapter, disable WoL on all but port A */
2086 if (global_quad_port_a != 0)
2087 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2089 adapter->flags |= IGB_FLAG_QUAD_PORT_A;
2090 /* Reset for multiple quad port adapters */
2091 if (++global_quad_port_a == 4)
2092 global_quad_port_a = 0;
2095 /* If the device can't wake, don't set software support */
2096 if (!device_can_wakeup(&adapter->pdev->dev))
2097 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2100 /* initialize the wol settings based on the eeprom settings */
2101 if (adapter->flags & IGB_FLAG_WOL_SUPPORTED)
2102 adapter->wol |= E1000_WUFC_MAG;
2104 /* Some vendors want WoL disabled by default, but still supported */
2105 if ((hw->mac.type == e1000_i350) &&
2106 (pdev->subsystem_vendor == PCI_VENDOR_ID_HP)) {
2107 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2111 device_set_wakeup_enable(&adapter->pdev->dev,
2112 adapter->flags & IGB_FLAG_WOL_SUPPORTED);
2114 /* reset the hardware with the new settings */
2117 /* let the f/w know that the h/w is now under the control of the
2119 igb_get_hw_control(adapter);
2121 strcpy(netdev->name, "eth%d");
2122 err = register_netdev(netdev);
2126 /* carrier off reporting is important to ethtool even BEFORE open */
2127 netif_carrier_off(netdev);
2129 #ifdef CONFIG_IGB_DCA
2130 if (dca_add_requester(&pdev->dev) == 0) {
2131 adapter->flags |= IGB_FLAG_DCA_ENABLED;
2132 dev_info(&pdev->dev, "DCA enabled\n");
2133 igb_setup_dca(adapter);
2138 /* do hw tstamp init after resetting */
2139 igb_ptp_init(adapter);
2141 dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
2142 /* print bus type/speed/width info */
2143 dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
2145 ((hw->bus.speed == e1000_bus_speed_2500) ? "2.5Gb/s" :
2146 (hw->bus.speed == e1000_bus_speed_5000) ? "5.0Gb/s" :
2148 ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
2149 (hw->bus.width == e1000_bus_width_pcie_x2) ? "Width x2" :
2150 (hw->bus.width == e1000_bus_width_pcie_x1) ? "Width x1" :
2154 ret_val = igb_read_part_string(hw, part_str, E1000_PBANUM_LENGTH);
2156 strcpy(part_str, "Unknown");
2157 dev_info(&pdev->dev, "%s: PBA No: %s\n", netdev->name, part_str);
2158 dev_info(&pdev->dev,
2159 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
2160 adapter->msix_entries ? "MSI-X" :
2161 (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
2162 adapter->num_rx_queues, adapter->num_tx_queues);
2163 switch (hw->mac.type) {
2167 igb_set_eee_i350(hw);
2173 pm_runtime_put_noidle(&pdev->dev);
2177 igb_release_hw_control(adapter);
2179 if (!igb_check_reset_block(hw))
2182 if (hw->flash_address)
2183 iounmap(hw->flash_address);
2185 igb_clear_interrupt_scheme(adapter);
2186 iounmap(hw->hw_addr);
2188 free_netdev(netdev);
2190 pci_release_selected_regions(pdev,
2191 pci_select_bars(pdev, IORESOURCE_MEM));
2194 pci_disable_device(pdev);
2199 * igb_remove - Device Removal Routine
2200 * @pdev: PCI device information struct
2202 * igb_remove is called by the PCI subsystem to alert the driver
2203 * that it should release a PCI device. The could be caused by a
2204 * Hot-Plug event, or because the driver is going to be removed from
2207 static void igb_remove(struct pci_dev *pdev)
2209 struct net_device *netdev = pci_get_drvdata(pdev);
2210 struct igb_adapter *adapter = netdev_priv(netdev);
2211 struct e1000_hw *hw = &adapter->hw;
2213 pm_runtime_get_noresume(&pdev->dev);
2214 igb_ptp_stop(adapter);
2217 * The watchdog timer may be rescheduled, so explicitly
2218 * disable watchdog from being rescheduled.
2220 set_bit(__IGB_DOWN, &adapter->state);
2221 del_timer_sync(&adapter->watchdog_timer);
2222 del_timer_sync(&adapter->phy_info_timer);
2224 cancel_work_sync(&adapter->reset_task);
2225 cancel_work_sync(&adapter->watchdog_task);
2227 #ifdef CONFIG_IGB_DCA
2228 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
2229 dev_info(&pdev->dev, "DCA disabled\n");
2230 dca_remove_requester(&pdev->dev);
2231 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
2232 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
2236 /* Release control of h/w to f/w. If f/w is AMT enabled, this
2237 * would have already happened in close and is redundant. */
2238 igb_release_hw_control(adapter);
2240 unregister_netdev(netdev);
2242 igb_clear_interrupt_scheme(adapter);
2244 #ifdef CONFIG_PCI_IOV
2245 /* reclaim resources allocated to VFs */
2246 if (adapter->vf_data) {
2247 /* disable iov and allow time for transactions to clear */
2248 if (igb_vfs_are_assigned(adapter)) {
2249 dev_info(&pdev->dev, "Unloading driver while VFs are assigned - VFs will not be deallocated\n");
2251 pci_disable_sriov(pdev);
2255 kfree(adapter->vf_data);
2256 adapter->vf_data = NULL;
2257 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
2260 dev_info(&pdev->dev, "IOV Disabled\n");
2264 iounmap(hw->hw_addr);
2265 if (hw->flash_address)
2266 iounmap(hw->flash_address);
2267 pci_release_selected_regions(pdev,
2268 pci_select_bars(pdev, IORESOURCE_MEM));
2270 kfree(adapter->shadow_vfta);
2271 free_netdev(netdev);
2273 pci_disable_pcie_error_reporting(pdev);
2275 pci_disable_device(pdev);
2279 * igb_probe_vfs - Initialize vf data storage and add VFs to pci config space
2280 * @adapter: board private structure to initialize
2282 * This function initializes the vf specific data storage and then attempts to
2283 * allocate the VFs. The reason for ordering it this way is because it is much
2284 * mor expensive time wise to disable SR-IOV than it is to allocate and free
2285 * the memory for the VFs.
2287 static void igb_probe_vfs(struct igb_adapter *adapter)
2289 #ifdef CONFIG_PCI_IOV
2290 struct pci_dev *pdev = adapter->pdev;
2291 struct e1000_hw *hw = &adapter->hw;
2292 int old_vfs = pci_num_vf(adapter->pdev);
2295 /* Virtualization features not supported on i210 family. */
2296 if ((hw->mac.type == e1000_i210) || (hw->mac.type == e1000_i211))
2300 dev_info(&pdev->dev, "%d pre-allocated VFs found - override "
2301 "max_vfs setting of %d\n", old_vfs, max_vfs);
2302 adapter->vfs_allocated_count = old_vfs;
2305 if (!adapter->vfs_allocated_count)
2308 adapter->vf_data = kcalloc(adapter->vfs_allocated_count,
2309 sizeof(struct vf_data_storage), GFP_KERNEL);
2311 /* if allocation failed then we do not support SR-IOV */
2312 if (!adapter->vf_data) {
2313 adapter->vfs_allocated_count = 0;
2314 dev_err(&pdev->dev, "Unable to allocate memory for VF "
2320 if (pci_enable_sriov(pdev, adapter->vfs_allocated_count))
2323 dev_info(&pdev->dev, "%d VFs allocated\n",
2324 adapter->vfs_allocated_count);
2325 for (i = 0; i < adapter->vfs_allocated_count; i++)
2326 igb_vf_configure(adapter, i);
2328 /* DMA Coalescing is not supported in IOV mode. */
2329 adapter->flags &= ~IGB_FLAG_DMAC;
2332 kfree(adapter->vf_data);
2333 adapter->vf_data = NULL;
2334 adapter->vfs_allocated_count = 0;
2337 #endif /* CONFIG_PCI_IOV */
2341 * igb_sw_init - Initialize general software structures (struct igb_adapter)
2342 * @adapter: board private structure to initialize
2344 * igb_sw_init initializes the Adapter private data structure.
2345 * Fields are initialized based on PCI device information and
2346 * OS network device settings (MTU size).
2348 static int igb_sw_init(struct igb_adapter *adapter)
2350 struct e1000_hw *hw = &adapter->hw;
2351 struct net_device *netdev = adapter->netdev;
2352 struct pci_dev *pdev = adapter->pdev;
2355 pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
2357 /* set default ring sizes */
2358 adapter->tx_ring_count = IGB_DEFAULT_TXD;
2359 adapter->rx_ring_count = IGB_DEFAULT_RXD;
2361 /* set default ITR values */
2362 adapter->rx_itr_setting = IGB_DEFAULT_ITR;
2363 adapter->tx_itr_setting = IGB_DEFAULT_ITR;
2365 /* set default work limits */
2366 adapter->tx_work_limit = IGB_DEFAULT_TX_WORK;
2368 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN +
2370 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2372 spin_lock_init(&adapter->stats64_lock);
2373 #ifdef CONFIG_PCI_IOV
2374 switch (hw->mac.type) {
2378 dev_warn(&pdev->dev,
2379 "Maximum of 7 VFs per PF, using max\n");
2380 adapter->vfs_allocated_count = 7;
2382 adapter->vfs_allocated_count = max_vfs;
2387 #endif /* CONFIG_PCI_IOV */
2389 /* Determine the maximum number of RSS queues supported. */
2390 switch (hw->mac.type) {
2392 max_rss_queues = IGB_MAX_RX_QUEUES_I211;
2396 max_rss_queues = IGB_MAX_RX_QUEUES_82575;
2399 /* I350 cannot do RSS and SR-IOV at the same time */
2400 if (!!adapter->vfs_allocated_count) {
2406 if (!!adapter->vfs_allocated_count) {
2413 max_rss_queues = IGB_MAX_RX_QUEUES;
2417 adapter->rss_queues = min_t(u32, max_rss_queues, num_online_cpus());
2419 /* Determine if we need to pair queues. */
2420 switch (hw->mac.type) {
2423 /* Device supports enough interrupts without queue pairing. */
2427 * If VFs are going to be allocated with RSS queues then we
2428 * should pair the queues in order to conserve interrupts due
2429 * to limited supply.
2431 if ((adapter->rss_queues > 1) &&
2432 (adapter->vfs_allocated_count > 6))
2433 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
2440 * If rss_queues > half of max_rss_queues, pair the queues in
2441 * order to conserve interrupts due to limited supply.
2443 if (adapter->rss_queues > (max_rss_queues / 2))
2444 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
2448 /* Setup and initialize a copy of the hw vlan table array */
2449 adapter->shadow_vfta = kzalloc(sizeof(u32) *
2450 E1000_VLAN_FILTER_TBL_SIZE,
2453 /* This call may decrease the number of queues */
2454 if (igb_init_interrupt_scheme(adapter, true)) {
2455 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
2459 igb_probe_vfs(adapter);
2461 /* Explicitly disable IRQ since the NIC can be in any state. */
2462 igb_irq_disable(adapter);
2464 if (hw->mac.type >= e1000_i350)
2465 adapter->flags &= ~IGB_FLAG_DMAC;
2467 set_bit(__IGB_DOWN, &adapter->state);
2472 * igb_open - Called when a network interface is made active
2473 * @netdev: network interface device structure
2475 * Returns 0 on success, negative value on failure
2477 * The open entry point is called when a network interface is made
2478 * active by the system (IFF_UP). At this point all resources needed
2479 * for transmit and receive operations are allocated, the interrupt
2480 * handler is registered with the OS, the watchdog timer is started,
2481 * and the stack is notified that the interface is ready.
2483 static int __igb_open(struct net_device *netdev, bool resuming)
2485 struct igb_adapter *adapter = netdev_priv(netdev);
2486 struct e1000_hw *hw = &adapter->hw;
2487 struct pci_dev *pdev = adapter->pdev;
2491 /* disallow open during test */
2492 if (test_bit(__IGB_TESTING, &adapter->state)) {
2498 pm_runtime_get_sync(&pdev->dev);
2500 netif_carrier_off(netdev);
2502 /* allocate transmit descriptors */
2503 err = igb_setup_all_tx_resources(adapter);
2507 /* allocate receive descriptors */
2508 err = igb_setup_all_rx_resources(adapter);
2512 igb_power_up_link(adapter);
2514 /* before we allocate an interrupt, we must be ready to handle it.
2515 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
2516 * as soon as we call pci_request_irq, so we have to setup our
2517 * clean_rx handler before we do so. */
2518 igb_configure(adapter);
2520 err = igb_request_irq(adapter);
2524 /* Notify the stack of the actual queue counts. */
2525 err = netif_set_real_num_tx_queues(adapter->netdev,
2526 adapter->num_tx_queues);
2528 goto err_set_queues;
2530 err = netif_set_real_num_rx_queues(adapter->netdev,
2531 adapter->num_rx_queues);
2533 goto err_set_queues;
2535 /* From here on the code is the same as igb_up() */
2536 clear_bit(__IGB_DOWN, &adapter->state);
2538 for (i = 0; i < adapter->num_q_vectors; i++)
2539 napi_enable(&(adapter->q_vector[i]->napi));
2541 /* Clear any pending interrupts. */
2544 igb_irq_enable(adapter);
2546 /* notify VFs that reset has been completed */
2547 if (adapter->vfs_allocated_count) {
2548 u32 reg_data = rd32(E1000_CTRL_EXT);
2549 reg_data |= E1000_CTRL_EXT_PFRSTD;
2550 wr32(E1000_CTRL_EXT, reg_data);
2553 netif_tx_start_all_queues(netdev);
2556 pm_runtime_put(&pdev->dev);
2558 /* start the watchdog. */
2559 hw->mac.get_link_status = 1;
2560 schedule_work(&adapter->watchdog_task);
2565 igb_free_irq(adapter);
2567 igb_release_hw_control(adapter);
2568 igb_power_down_link(adapter);
2569 igb_free_all_rx_resources(adapter);
2571 igb_free_all_tx_resources(adapter);
2575 pm_runtime_put(&pdev->dev);
2580 static int igb_open(struct net_device *netdev)
2582 return __igb_open(netdev, false);
2586 * igb_close - Disables a network interface
2587 * @netdev: network interface device structure
2589 * Returns 0, this is not allowed to fail
2591 * The close entry point is called when an interface is de-activated
2592 * by the OS. The hardware is still under the driver's control, but
2593 * needs to be disabled. A global MAC reset is issued to stop the
2594 * hardware, and all transmit and receive resources are freed.
2596 static int __igb_close(struct net_device *netdev, bool suspending)
2598 struct igb_adapter *adapter = netdev_priv(netdev);
2599 struct pci_dev *pdev = adapter->pdev;
2601 WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
2604 pm_runtime_get_sync(&pdev->dev);
2607 igb_free_irq(adapter);
2609 igb_free_all_tx_resources(adapter);
2610 igb_free_all_rx_resources(adapter);
2613 pm_runtime_put_sync(&pdev->dev);
2617 static int igb_close(struct net_device *netdev)
2619 return __igb_close(netdev, false);
2623 * igb_setup_tx_resources - allocate Tx resources (Descriptors)
2624 * @tx_ring: tx descriptor ring (for a specific queue) to setup
2626 * Return 0 on success, negative on failure
2628 int igb_setup_tx_resources(struct igb_ring *tx_ring)
2630 struct device *dev = tx_ring->dev;
2633 size = sizeof(struct igb_tx_buffer) * tx_ring->count;
2635 tx_ring->tx_buffer_info = vzalloc(size);
2636 if (!tx_ring->tx_buffer_info)
2639 /* round up to nearest 4K */
2640 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
2641 tx_ring->size = ALIGN(tx_ring->size, 4096);
2643 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
2644 &tx_ring->dma, GFP_KERNEL);
2648 tx_ring->next_to_use = 0;
2649 tx_ring->next_to_clean = 0;
2654 vfree(tx_ring->tx_buffer_info);
2655 tx_ring->tx_buffer_info = NULL;
2656 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
2661 * igb_setup_all_tx_resources - wrapper to allocate Tx resources
2662 * (Descriptors) for all queues
2663 * @adapter: board private structure
2665 * Return 0 on success, negative on failure
2667 static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
2669 struct pci_dev *pdev = adapter->pdev;
2672 for (i = 0; i < adapter->num_tx_queues; i++) {
2673 err = igb_setup_tx_resources(adapter->tx_ring[i]);
2676 "Allocation for Tx Queue %u failed\n", i);
2677 for (i--; i >= 0; i--)
2678 igb_free_tx_resources(adapter->tx_ring[i]);
2687 * igb_setup_tctl - configure the transmit control registers
2688 * @adapter: Board private structure
2690 void igb_setup_tctl(struct igb_adapter *adapter)
2692 struct e1000_hw *hw = &adapter->hw;
2695 /* disable queue 0 which is enabled by default on 82575 and 82576 */
2696 wr32(E1000_TXDCTL(0), 0);
2698 /* Program the Transmit Control Register */
2699 tctl = rd32(E1000_TCTL);
2700 tctl &= ~E1000_TCTL_CT;
2701 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2702 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2704 igb_config_collision_dist(hw);
2706 /* Enable transmits */
2707 tctl |= E1000_TCTL_EN;
2709 wr32(E1000_TCTL, tctl);
2713 * igb_configure_tx_ring - Configure transmit ring after Reset
2714 * @adapter: board private structure
2715 * @ring: tx ring to configure
2717 * Configure a transmit ring after a reset.
2719 void igb_configure_tx_ring(struct igb_adapter *adapter,
2720 struct igb_ring *ring)
2722 struct e1000_hw *hw = &adapter->hw;
2724 u64 tdba = ring->dma;
2725 int reg_idx = ring->reg_idx;
2727 /* disable the queue */
2728 wr32(E1000_TXDCTL(reg_idx), 0);
2732 wr32(E1000_TDLEN(reg_idx),
2733 ring->count * sizeof(union e1000_adv_tx_desc));
2734 wr32(E1000_TDBAL(reg_idx),
2735 tdba & 0x00000000ffffffffULL);
2736 wr32(E1000_TDBAH(reg_idx), tdba >> 32);
2738 ring->tail = hw->hw_addr + E1000_TDT(reg_idx);
2739 wr32(E1000_TDH(reg_idx), 0);
2740 writel(0, ring->tail);
2742 txdctl |= IGB_TX_PTHRESH;
2743 txdctl |= IGB_TX_HTHRESH << 8;
2744 txdctl |= IGB_TX_WTHRESH << 16;
2746 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
2747 wr32(E1000_TXDCTL(reg_idx), txdctl);
2751 * igb_configure_tx - Configure transmit Unit after Reset
2752 * @adapter: board private structure
2754 * Configure the Tx unit of the MAC after a reset.
2756 static void igb_configure_tx(struct igb_adapter *adapter)
2760 for (i = 0; i < adapter->num_tx_queues; i++)
2761 igb_configure_tx_ring(adapter, adapter->tx_ring[i]);
2765 * igb_setup_rx_resources - allocate Rx resources (Descriptors)
2766 * @rx_ring: rx descriptor ring (for a specific queue) to setup
2768 * Returns 0 on success, negative on failure
2770 int igb_setup_rx_resources(struct igb_ring *rx_ring)
2772 struct device *dev = rx_ring->dev;
2775 size = sizeof(struct igb_rx_buffer) * rx_ring->count;
2777 rx_ring->rx_buffer_info = vzalloc(size);
2778 if (!rx_ring->rx_buffer_info)
2781 /* Round up to nearest 4K */
2782 rx_ring->size = rx_ring->count * sizeof(union e1000_adv_rx_desc);
2783 rx_ring->size = ALIGN(rx_ring->size, 4096);
2785 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
2786 &rx_ring->dma, GFP_KERNEL);
2790 rx_ring->next_to_alloc = 0;
2791 rx_ring->next_to_clean = 0;
2792 rx_ring->next_to_use = 0;
2797 vfree(rx_ring->rx_buffer_info);
2798 rx_ring->rx_buffer_info = NULL;
2799 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
2804 * igb_setup_all_rx_resources - wrapper to allocate Rx resources
2805 * (Descriptors) for all queues
2806 * @adapter: board private structure
2808 * Return 0 on success, negative on failure
2810 static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
2812 struct pci_dev *pdev = adapter->pdev;
2815 for (i = 0; i < adapter->num_rx_queues; i++) {
2816 err = igb_setup_rx_resources(adapter->rx_ring[i]);
2819 "Allocation for Rx Queue %u failed\n", i);
2820 for (i--; i >= 0; i--)
2821 igb_free_rx_resources(adapter->rx_ring[i]);
2830 * igb_setup_mrqc - configure the multiple receive queue control registers
2831 * @adapter: Board private structure
2833 static void igb_setup_mrqc(struct igb_adapter *adapter)
2835 struct e1000_hw *hw = &adapter->hw;
2837 u32 j, num_rx_queues, shift = 0;
2838 static const u32 rsskey[10] = { 0xDA565A6D, 0xC20E5B25, 0x3D256741,
2839 0xB08FA343, 0xCB2BCAD0, 0xB4307BAE,
2840 0xA32DCB77, 0x0CF23080, 0x3BB7426A,
2843 /* Fill out hash function seeds */
2844 for (j = 0; j < 10; j++)
2845 wr32(E1000_RSSRK(j), rsskey[j]);
2847 num_rx_queues = adapter->rss_queues;
2849 switch (hw->mac.type) {
2854 /* 82576 supports 2 RSS queues for SR-IOV */
2855 if (adapter->vfs_allocated_count) {
2865 * Populate the indirection table 4 entries at a time. To do this
2866 * we are generating the results for n and n+2 and then interleaving
2867 * those with the results with n+1 and n+3.
2869 for (j = 0; j < 32; j++) {
2870 /* first pass generates n and n+2 */
2871 u32 base = ((j * 0x00040004) + 0x00020000) * num_rx_queues;
2872 u32 reta = (base & 0x07800780) >> (7 - shift);
2874 /* second pass generates n+1 and n+3 */
2875 base += 0x00010001 * num_rx_queues;
2876 reta |= (base & 0x07800780) << (1 + shift);
2878 wr32(E1000_RETA(j), reta);
2882 * Disable raw packet checksumming so that RSS hash is placed in
2883 * descriptor on writeback. No need to enable TCP/UDP/IP checksum
2884 * offloads as they are enabled by default
2886 rxcsum = rd32(E1000_RXCSUM);
2887 rxcsum |= E1000_RXCSUM_PCSD;
2889 if (adapter->hw.mac.type >= e1000_82576)
2890 /* Enable Receive Checksum Offload for SCTP */
2891 rxcsum |= E1000_RXCSUM_CRCOFL;
2893 /* Don't need to set TUOFL or IPOFL, they default to 1 */
2894 wr32(E1000_RXCSUM, rxcsum);
2896 /* Generate RSS hash based on packet types, TCP/UDP
2897 * port numbers and/or IPv4/v6 src and dst addresses
2899 mrqc = E1000_MRQC_RSS_FIELD_IPV4 |
2900 E1000_MRQC_RSS_FIELD_IPV4_TCP |
2901 E1000_MRQC_RSS_FIELD_IPV6 |
2902 E1000_MRQC_RSS_FIELD_IPV6_TCP |
2903 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX;
2905 if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV4_UDP)
2906 mrqc |= E1000_MRQC_RSS_FIELD_IPV4_UDP;
2907 if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV6_UDP)
2908 mrqc |= E1000_MRQC_RSS_FIELD_IPV6_UDP;
2910 /* If VMDq is enabled then we set the appropriate mode for that, else
2911 * we default to RSS so that an RSS hash is calculated per packet even
2912 * if we are only using one queue */
2913 if (adapter->vfs_allocated_count) {
2914 if (hw->mac.type > e1000_82575) {
2915 /* Set the default pool for the PF's first queue */
2916 u32 vtctl = rd32(E1000_VT_CTL);
2917 vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK |
2918 E1000_VT_CTL_DISABLE_DEF_POOL);
2919 vtctl |= adapter->vfs_allocated_count <<
2920 E1000_VT_CTL_DEFAULT_POOL_SHIFT;
2921 wr32(E1000_VT_CTL, vtctl);
2923 if (adapter->rss_queues > 1)
2924 mrqc |= E1000_MRQC_ENABLE_VMDQ_RSS_2Q;
2926 mrqc |= E1000_MRQC_ENABLE_VMDQ;
2928 if (hw->mac.type != e1000_i211)
2929 mrqc |= E1000_MRQC_ENABLE_RSS_4Q;
2931 igb_vmm_control(adapter);
2933 wr32(E1000_MRQC, mrqc);
2937 * igb_setup_rctl - configure the receive control registers
2938 * @adapter: Board private structure
2940 void igb_setup_rctl(struct igb_adapter *adapter)
2942 struct e1000_hw *hw = &adapter->hw;
2945 rctl = rd32(E1000_RCTL);
2947 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2948 rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
2950 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
2951 (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2954 * enable stripping of CRC. It's unlikely this will break BMC
2955 * redirection as it did with e1000. Newer features require
2956 * that the HW strips the CRC.
2958 rctl |= E1000_RCTL_SECRC;
2960 /* disable store bad packets and clear size bits. */
2961 rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
2963 /* enable LPE to prevent packets larger than max_frame_size */
2964 rctl |= E1000_RCTL_LPE;
2966 /* disable queue 0 to prevent tail write w/o re-config */
2967 wr32(E1000_RXDCTL(0), 0);
2969 /* Attention!!! For SR-IOV PF driver operations you must enable
2970 * queue drop for all VF and PF queues to prevent head of line blocking
2971 * if an un-trusted VF does not provide descriptors to hardware.
2973 if (adapter->vfs_allocated_count) {
2974 /* set all queue drop enable bits */
2975 wr32(E1000_QDE, ALL_QUEUES);
2978 /* This is useful for sniffing bad packets. */
2979 if (adapter->netdev->features & NETIF_F_RXALL) {
2980 /* UPE and MPE will be handled by normal PROMISC logic
2981 * in e1000e_set_rx_mode */
2982 rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
2983 E1000_RCTL_BAM | /* RX All Bcast Pkts */
2984 E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
2986 rctl &= ~(E1000_RCTL_VFE | /* Disable VLAN filter */
2987 E1000_RCTL_DPF | /* Allow filtered pause */
2988 E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
2989 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
2990 * and that breaks VLANs.
2994 wr32(E1000_RCTL, rctl);
2997 static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
3000 struct e1000_hw *hw = &adapter->hw;
3003 /* if it isn't the PF check to see if VFs are enabled and
3004 * increase the size to support vlan tags */
3005 if (vfn < adapter->vfs_allocated_count &&
3006 adapter->vf_data[vfn].vlans_enabled)
3007 size += VLAN_TAG_SIZE;
3009 vmolr = rd32(E1000_VMOLR(vfn));
3010 vmolr &= ~E1000_VMOLR_RLPML_MASK;
3011 vmolr |= size | E1000_VMOLR_LPE;
3012 wr32(E1000_VMOLR(vfn), vmolr);
3018 * igb_rlpml_set - set maximum receive packet size
3019 * @adapter: board private structure
3021 * Configure maximum receivable packet size.
3023 static void igb_rlpml_set(struct igb_adapter *adapter)
3025 u32 max_frame_size = adapter->max_frame_size;
3026 struct e1000_hw *hw = &adapter->hw;
3027 u16 pf_id = adapter->vfs_allocated_count;
3030 igb_set_vf_rlpml(adapter, max_frame_size, pf_id);
3032 * If we're in VMDQ or SR-IOV mode, then set global RLPML
3033 * to our max jumbo frame size, in case we need to enable
3034 * jumbo frames on one of the rings later.
3035 * This will not pass over-length frames into the default
3036 * queue because it's gated by the VMOLR.RLPML.
3038 max_frame_size = MAX_JUMBO_FRAME_SIZE;
3041 wr32(E1000_RLPML, max_frame_size);
3044 static inline void igb_set_vmolr(struct igb_adapter *adapter,
3047 struct e1000_hw *hw = &adapter->hw;
3051 * This register exists only on 82576 and newer so if we are older then
3052 * we should exit and do nothing
3054 if (hw->mac.type < e1000_82576)
3057 vmolr = rd32(E1000_VMOLR(vfn));
3058 vmolr |= E1000_VMOLR_STRVLAN; /* Strip vlan tags */
3060 vmolr |= E1000_VMOLR_AUPE; /* Accept untagged packets */
3062 vmolr &= ~(E1000_VMOLR_AUPE); /* Tagged packets ONLY */
3064 /* clear all bits that might not be set */
3065 vmolr &= ~(E1000_VMOLR_BAM | E1000_VMOLR_RSSE);
3067 if (adapter->rss_queues > 1 && vfn == adapter->vfs_allocated_count)
3068 vmolr |= E1000_VMOLR_RSSE; /* enable RSS */
3070 * for VMDq only allow the VFs and pool 0 to accept broadcast and
3073 if (vfn <= adapter->vfs_allocated_count)
3074 vmolr |= E1000_VMOLR_BAM; /* Accept broadcast */
3076 wr32(E1000_VMOLR(vfn), vmolr);
3080 * igb_configure_rx_ring - Configure a receive ring after Reset
3081 * @adapter: board private structure
3082 * @ring: receive ring to be configured
3084 * Configure the Rx unit of the MAC after a reset.
3086 void igb_configure_rx_ring(struct igb_adapter *adapter,
3087 struct igb_ring *ring)
3089 struct e1000_hw *hw = &adapter->hw;
3090 u64 rdba = ring->dma;
3091 int reg_idx = ring->reg_idx;
3092 u32 srrctl = 0, rxdctl = 0;
3094 /* disable the queue */
3095 wr32(E1000_RXDCTL(reg_idx), 0);
3097 /* Set DMA base address registers */
3098 wr32(E1000_RDBAL(reg_idx),
3099 rdba & 0x00000000ffffffffULL);
3100 wr32(E1000_RDBAH(reg_idx), rdba >> 32);
3101 wr32(E1000_RDLEN(reg_idx),
3102 ring->count * sizeof(union e1000_adv_rx_desc));
3104 /* initialize head and tail */
3105 ring->tail = hw->hw_addr + E1000_RDT(reg_idx);
3106 wr32(E1000_RDH(reg_idx), 0);
3107 writel(0, ring->tail);
3109 /* set descriptor configuration */
3110 srrctl = IGB_RX_HDR_LEN << E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
3111 srrctl |= IGB_RX_BUFSZ >> E1000_SRRCTL_BSIZEPKT_SHIFT;
3112 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
3113 if (hw->mac.type >= e1000_82580)
3114 srrctl |= E1000_SRRCTL_TIMESTAMP;
3115 /* Only set Drop Enable if we are supporting multiple queues */
3116 if (adapter->vfs_allocated_count || adapter->num_rx_queues > 1)
3117 srrctl |= E1000_SRRCTL_DROP_EN;
3119 wr32(E1000_SRRCTL(reg_idx), srrctl);
3121 /* set filtering for VMDQ pools */
3122 igb_set_vmolr(adapter, reg_idx & 0x7, true);
3124 rxdctl |= IGB_RX_PTHRESH;
3125 rxdctl |= IGB_RX_HTHRESH << 8;
3126 rxdctl |= IGB_RX_WTHRESH << 16;
3128 /* enable receive descriptor fetching */
3129 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
3130 wr32(E1000_RXDCTL(reg_idx), rxdctl);
3134 * igb_configure_rx - Configure receive Unit after Reset
3135 * @adapter: board private structure
3137 * Configure the Rx unit of the MAC after a reset.
3139 static void igb_configure_rx(struct igb_adapter *adapter)
3143 /* set UTA to appropriate mode */
3144 igb_set_uta(adapter);
3146 /* set the correct pool for the PF default MAC address in entry 0 */
3147 igb_rar_set_qsel(adapter, adapter->hw.mac.addr, 0,
3148 adapter->vfs_allocated_count);
3150 /* Setup the HW Rx Head and Tail Descriptor Pointers and
3151 * the Base and Length of the Rx Descriptor Ring */
3152 for (i = 0; i < adapter->num_rx_queues; i++)
3153 igb_configure_rx_ring(adapter, adapter->rx_ring[i]);
3157 * igb_free_tx_resources - Free Tx Resources per Queue
3158 * @tx_ring: Tx descriptor ring for a specific queue
3160 * Free all transmit software resources
3162 void igb_free_tx_resources(struct igb_ring *tx_ring)
3164 igb_clean_tx_ring(tx_ring);
3166 vfree(tx_ring->tx_buffer_info);
3167 tx_ring->tx_buffer_info = NULL;
3169 /* if not set, then don't free */
3173 dma_free_coherent(tx_ring->dev, tx_ring->size,
3174 tx_ring->desc, tx_ring->dma);
3176 tx_ring->desc = NULL;
3180 * igb_free_all_tx_resources - Free Tx Resources for All Queues
3181 * @adapter: board private structure
3183 * Free all transmit software resources
3185 static void igb_free_all_tx_resources(struct igb_adapter *adapter)
3189 for (i = 0; i < adapter->num_tx_queues; i++)
3190 igb_free_tx_resources(adapter->tx_ring[i]);
3193 void igb_unmap_and_free_tx_resource(struct igb_ring *ring,
3194 struct igb_tx_buffer *tx_buffer)
3196 if (tx_buffer->skb) {
3197 dev_kfree_skb_any(tx_buffer->skb);
3198 if (dma_unmap_len(tx_buffer, len))
3199 dma_unmap_single(ring->dev,
3200 dma_unmap_addr(tx_buffer, dma),
3201 dma_unmap_len(tx_buffer, len),
3203 } else if (dma_unmap_len(tx_buffer, len)) {
3204 dma_unmap_page(ring->dev,
3205 dma_unmap_addr(tx_buffer, dma),
3206 dma_unmap_len(tx_buffer, len),
3209 tx_buffer->next_to_watch = NULL;
3210 tx_buffer->skb = NULL;
3211 dma_unmap_len_set(tx_buffer, len, 0);
3212 /* buffer_info must be completely set up in the transmit path */
3216 * igb_clean_tx_ring - Free Tx Buffers
3217 * @tx_ring: ring to be cleaned
3219 static void igb_clean_tx_ring(struct igb_ring *tx_ring)
3221 struct igb_tx_buffer *buffer_info;
3225 if (!tx_ring->tx_buffer_info)
3227 /* Free all the Tx ring sk_buffs */
3229 for (i = 0; i < tx_ring->count; i++) {
3230 buffer_info = &tx_ring->tx_buffer_info[i];
3231 igb_unmap_and_free_tx_resource(tx_ring, buffer_info);
3234 netdev_tx_reset_queue(txring_txq(tx_ring));
3236 size = sizeof(struct igb_tx_buffer) * tx_ring->count;
3237 memset(tx_ring->tx_buffer_info, 0, size);
3239 /* Zero out the descriptor ring */
3240 memset(tx_ring->desc, 0, tx_ring->size);
3242 tx_ring->next_to_use = 0;
3243 tx_ring->next_to_clean = 0;
3247 * igb_clean_all_tx_rings - Free Tx Buffers for all queues
3248 * @adapter: board private structure
3250 static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
3254 for (i = 0; i < adapter->num_tx_queues; i++)
3255 igb_clean_tx_ring(adapter->tx_ring[i]);
3259 * igb_free_rx_resources - Free Rx Resources
3260 * @rx_ring: ring to clean the resources from
3262 * Free all receive software resources
3264 void igb_free_rx_resources(struct igb_ring *rx_ring)
3266 igb_clean_rx_ring(rx_ring);
3268 vfree(rx_ring->rx_buffer_info);
3269 rx_ring->rx_buffer_info = NULL;
3271 /* if not set, then don't free */
3275 dma_free_coherent(rx_ring->dev, rx_ring->size,
3276 rx_ring->desc, rx_ring->dma);
3278 rx_ring->desc = NULL;
3282 * igb_free_all_rx_resources - Free Rx Resources for All Queues
3283 * @adapter: board private structure
3285 * Free all receive software resources
3287 static void igb_free_all_rx_resources(struct igb_adapter *adapter)
3291 for (i = 0; i < adapter->num_rx_queues; i++)
3292 igb_free_rx_resources(adapter->rx_ring[i]);
3296 * igb_clean_rx_ring - Free Rx Buffers per Queue
3297 * @rx_ring: ring to free buffers from
3299 static void igb_clean_rx_ring(struct igb_ring *rx_ring)
3305 dev_kfree_skb(rx_ring->skb);
3306 rx_ring->skb = NULL;
3308 if (!rx_ring->rx_buffer_info)
3311 /* Free all the Rx ring sk_buffs */
3312 for (i = 0; i < rx_ring->count; i++) {
3313 struct igb_rx_buffer *buffer_info = &rx_ring->rx_buffer_info[i];
3315 if (!buffer_info->page)
3318 dma_unmap_page(rx_ring->dev,
3322 __free_page(buffer_info->page);
3324 buffer_info->page = NULL;
3327 size = sizeof(struct igb_rx_buffer) * rx_ring->count;
3328 memset(rx_ring->rx_buffer_info, 0, size);
3330 /* Zero out the descriptor ring */
3331 memset(rx_ring->desc, 0, rx_ring->size);
3333 rx_ring->next_to_alloc = 0;
3334 rx_ring->next_to_clean = 0;
3335 rx_ring->next_to_use = 0;
3339 * igb_clean_all_rx_rings - Free Rx Buffers for all queues
3340 * @adapter: board private structure
3342 static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
3346 for (i = 0; i < adapter->num_rx_queues; i++)
3347 igb_clean_rx_ring(adapter->rx_ring[i]);
3351 * igb_set_mac - Change the Ethernet Address of the NIC
3352 * @netdev: network interface device structure
3353 * @p: pointer to an address structure
3355 * Returns 0 on success, negative on failure
3357 static int igb_set_mac(struct net_device *netdev, void *p)
3359 struct igb_adapter *adapter = netdev_priv(netdev);
3360 struct e1000_hw *hw = &adapter->hw;
3361 struct sockaddr *addr = p;
3363 if (!is_valid_ether_addr(addr->sa_data))
3364 return -EADDRNOTAVAIL;
3366 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3367 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
3369 /* set the correct pool for the new PF MAC address in entry 0 */
3370 igb_rar_set_qsel(adapter, hw->mac.addr, 0,
3371 adapter->vfs_allocated_count);
3377 * igb_write_mc_addr_list - write multicast addresses to MTA
3378 * @netdev: network interface device structure
3380 * Writes multicast address list to the MTA hash table.
3381 * Returns: -ENOMEM on failure
3382 * 0 on no addresses written
3383 * X on writing X addresses to MTA
3385 static int igb_write_mc_addr_list(struct net_device *netdev)
3387 struct igb_adapter *adapter = netdev_priv(netdev);
3388 struct e1000_hw *hw = &adapter->hw;
3389 struct netdev_hw_addr *ha;
3393 if (netdev_mc_empty(netdev)) {
3394 /* nothing to program, so clear mc list */
3395 igb_update_mc_addr_list(hw, NULL, 0);
3396 igb_restore_vf_multicasts(adapter);
3400 mta_list = kzalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
3404 /* The shared function expects a packed array of only addresses. */
3406 netdev_for_each_mc_addr(ha, netdev)
3407 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3409 igb_update_mc_addr_list(hw, mta_list, i);
3412 return netdev_mc_count(netdev);
3416 * igb_write_uc_addr_list - write unicast addresses to RAR table
3417 * @netdev: network interface device structure
3419 * Writes unicast address list to the RAR table.
3420 * Returns: -ENOMEM on failure/insufficient address space
3421 * 0 on no addresses written
3422 * X on writing X addresses to the RAR table
3424 static int igb_write_uc_addr_list(struct net_device *netdev)
3426 struct igb_adapter *adapter = netdev_priv(netdev);
3427 struct e1000_hw *hw = &adapter->hw;
3428 unsigned int vfn = adapter->vfs_allocated_count;
3429 unsigned int rar_entries = hw->mac.rar_entry_count - (vfn + 1);
3432 /* return ENOMEM indicating insufficient memory for addresses */
3433 if (netdev_uc_count(netdev) > rar_entries)
3436 if (!netdev_uc_empty(netdev) && rar_entries) {
3437 struct netdev_hw_addr *ha;
3439 netdev_for_each_uc_addr(ha, netdev) {
3442 igb_rar_set_qsel(adapter, ha->addr,
3448 /* write the addresses in reverse order to avoid write combining */
3449 for (; rar_entries > 0 ; rar_entries--) {
3450 wr32(E1000_RAH(rar_entries), 0);
3451 wr32(E1000_RAL(rar_entries), 0);
3459 * igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
3460 * @netdev: network interface device structure
3462 * The set_rx_mode entry point is called whenever the unicast or multicast
3463 * address lists or the network interface flags are updated. This routine is
3464 * responsible for configuring the hardware for proper unicast, multicast,
3465 * promiscuous mode, and all-multi behavior.
3467 static void igb_set_rx_mode(struct net_device *netdev)
3469 struct igb_adapter *adapter = netdev_priv(netdev);
3470 struct e1000_hw *hw = &adapter->hw;
3471 unsigned int vfn = adapter->vfs_allocated_count;
3472 u32 rctl, vmolr = 0;
3475 /* Check for Promiscuous and All Multicast modes */
3476 rctl = rd32(E1000_RCTL);
3478 /* clear the effected bits */
3479 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE | E1000_RCTL_VFE);
3481 if (netdev->flags & IFF_PROMISC) {
3482 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3483 vmolr |= (E1000_VMOLR_ROPE | E1000_VMOLR_MPME);
3485 if (netdev->flags & IFF_ALLMULTI) {
3486 rctl |= E1000_RCTL_MPE;
3487 vmolr |= E1000_VMOLR_MPME;
3490 * Write addresses to the MTA, if the attempt fails
3491 * then we should just turn on promiscuous mode so
3492 * that we can at least receive multicast traffic
3494 count = igb_write_mc_addr_list(netdev);
3496 rctl |= E1000_RCTL_MPE;
3497 vmolr |= E1000_VMOLR_MPME;
3499 vmolr |= E1000_VMOLR_ROMPE;
3503 * Write addresses to available RAR registers, if there is not
3504 * sufficient space to store all the addresses then enable
3505 * unicast promiscuous mode
3507 count = igb_write_uc_addr_list(netdev);
3509 rctl |= E1000_RCTL_UPE;
3510 vmolr |= E1000_VMOLR_ROPE;
3512 rctl |= E1000_RCTL_VFE;
3514 wr32(E1000_RCTL, rctl);
3517 * In order to support SR-IOV and eventually VMDq it is necessary to set
3518 * the VMOLR to enable the appropriate modes. Without this workaround
3519 * we will have issues with VLAN tag stripping not being done for frames
3520 * that are only arriving because we are the default pool
3522 if ((hw->mac.type < e1000_82576) || (hw->mac.type > e1000_i350))
3525 vmolr |= rd32(E1000_VMOLR(vfn)) &
3526 ~(E1000_VMOLR_ROPE | E1000_VMOLR_MPME | E1000_VMOLR_ROMPE);
3527 wr32(E1000_VMOLR(vfn), vmolr);
3528 igb_restore_vf_multicasts(adapter);
3531 static void igb_check_wvbr(struct igb_adapter *adapter)
3533 struct e1000_hw *hw = &adapter->hw;
3536 switch (hw->mac.type) {
3539 if (!(wvbr = rd32(E1000_WVBR)))
3546 adapter->wvbr |= wvbr;
3549 #define IGB_STAGGERED_QUEUE_OFFSET 8
3551 static void igb_spoof_check(struct igb_adapter *adapter)
3558 for(j = 0; j < adapter->vfs_allocated_count; j++) {
3559 if (adapter->wvbr & (1 << j) ||
3560 adapter->wvbr & (1 << (j + IGB_STAGGERED_QUEUE_OFFSET))) {
3561 dev_warn(&adapter->pdev->dev,
3562 "Spoof event(s) detected on VF %d\n", j);
3565 (1 << (j + IGB_STAGGERED_QUEUE_OFFSET)));
3570 /* Need to wait a few seconds after link up to get diagnostic information from
3572 static void igb_update_phy_info(unsigned long data)
3574 struct igb_adapter *adapter = (struct igb_adapter *) data;
3575 igb_get_phy_info(&adapter->hw);
3579 * igb_has_link - check shared code for link and determine up/down
3580 * @adapter: pointer to driver private info
3582 bool igb_has_link(struct igb_adapter *adapter)
3584 struct e1000_hw *hw = &adapter->hw;
3585 bool link_active = false;
3588 /* get_link_status is set on LSC (link status) interrupt or
3589 * rx sequence error interrupt. get_link_status will stay
3590 * false until the e1000_check_for_link establishes link
3591 * for copper adapters ONLY
3593 switch (hw->phy.media_type) {
3594 case e1000_media_type_copper:
3595 if (hw->mac.get_link_status) {
3596 ret_val = hw->mac.ops.check_for_link(hw);
3597 link_active = !hw->mac.get_link_status;
3602 case e1000_media_type_internal_serdes:
3603 ret_val = hw->mac.ops.check_for_link(hw);
3604 link_active = hw->mac.serdes_has_link;
3607 case e1000_media_type_unknown:
3614 static bool igb_thermal_sensor_event(struct e1000_hw *hw, u32 event)
3617 u32 ctrl_ext, thstat;
3619 /* check for thermal sensor event on i350 copper only */
3620 if (hw->mac.type == e1000_i350) {
3621 thstat = rd32(E1000_THSTAT);
3622 ctrl_ext = rd32(E1000_CTRL_EXT);
3624 if ((hw->phy.media_type == e1000_media_type_copper) &&
3625 !(ctrl_ext & E1000_CTRL_EXT_LINK_MODE_SGMII)) {
3626 ret = !!(thstat & event);
3634 * igb_watchdog - Timer Call-back
3635 * @data: pointer to adapter cast into an unsigned long
3637 static void igb_watchdog(unsigned long data)
3639 struct igb_adapter *adapter = (struct igb_adapter *)data;
3640 /* Do the rest outside of interrupt context */
3641 schedule_work(&adapter->watchdog_task);
3644 static void igb_watchdog_task(struct work_struct *work)
3646 struct igb_adapter *adapter = container_of(work,
3649 struct e1000_hw *hw = &adapter->hw;
3650 struct net_device *netdev = adapter->netdev;
3654 link = igb_has_link(adapter);
3656 /* Cancel scheduled suspend requests. */
3657 pm_runtime_resume(netdev->dev.parent);
3659 if (!netif_carrier_ok(netdev)) {
3661 hw->mac.ops.get_speed_and_duplex(hw,
3662 &adapter->link_speed,
3663 &adapter->link_duplex);
3665 ctrl = rd32(E1000_CTRL);
3666 /* Links status message must follow this format */
3667 printk(KERN_INFO "igb: %s NIC Link is Up %d Mbps %s "
3668 "Duplex, Flow Control: %s\n",
3670 adapter->link_speed,
3671 adapter->link_duplex == FULL_DUPLEX ?
3673 (ctrl & E1000_CTRL_TFCE) &&
3674 (ctrl & E1000_CTRL_RFCE) ? "RX/TX" :
3675 (ctrl & E1000_CTRL_RFCE) ? "RX" :
3676 (ctrl & E1000_CTRL_TFCE) ? "TX" : "None");
3678 /* check for thermal sensor event */
3679 if (igb_thermal_sensor_event(hw,
3680 E1000_THSTAT_LINK_THROTTLE)) {
3681 netdev_info(netdev, "The network adapter link "
3682 "speed was downshifted because it "
3686 /* adjust timeout factor according to speed/duplex */
3687 adapter->tx_timeout_factor = 1;
3688 switch (adapter->link_speed) {
3690 adapter->tx_timeout_factor = 14;
3693 /* maybe add some timeout factor ? */
3697 netif_carrier_on(netdev);
3699 igb_ping_all_vfs(adapter);
3700 igb_check_vf_rate_limit(adapter);
3702 /* link state has changed, schedule phy info update */
3703 if (!test_bit(__IGB_DOWN, &adapter->state))
3704 mod_timer(&adapter->phy_info_timer,
3705 round_jiffies(jiffies + 2 * HZ));
3708 if (netif_carrier_ok(netdev)) {
3709 adapter->link_speed = 0;
3710 adapter->link_duplex = 0;
3712 /* check for thermal sensor event */
3713 if (igb_thermal_sensor_event(hw,
3714 E1000_THSTAT_PWR_DOWN)) {
3715 netdev_err(netdev, "The network adapter was "
3716 "stopped because it overheated\n");
3719 /* Links status message must follow this format */
3720 printk(KERN_INFO "igb: %s NIC Link is Down\n",
3722 netif_carrier_off(netdev);
3724 igb_ping_all_vfs(adapter);
3726 /* link state has changed, schedule phy info update */
3727 if (!test_bit(__IGB_DOWN, &adapter->state))
3728 mod_timer(&adapter->phy_info_timer,
3729 round_jiffies(jiffies + 2 * HZ));
3731 pm_schedule_suspend(netdev->dev.parent,
3736 spin_lock(&adapter->stats64_lock);
3737 igb_update_stats(adapter, &adapter->stats64);
3738 spin_unlock(&adapter->stats64_lock);
3740 for (i = 0; i < adapter->num_tx_queues; i++) {
3741 struct igb_ring *tx_ring = adapter->tx_ring[i];
3742 if (!netif_carrier_ok(netdev)) {
3743 /* We've lost link, so the controller stops DMA,
3744 * but we've got queued Tx work that's never going
3745 * to get done, so reset controller to flush Tx.
3746 * (Do the reset outside of interrupt context). */
3747 if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) {
3748 adapter->tx_timeout_count++;
3749 schedule_work(&adapter->reset_task);
3750 /* return immediately since reset is imminent */
3755 /* Force detection of hung controller every watchdog period */
3756 set_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
3759 /* Cause software interrupt to ensure rx ring is cleaned */
3760 if (adapter->msix_entries) {
3762 for (i = 0; i < adapter->num_q_vectors; i++)
3763 eics |= adapter->q_vector[i]->eims_value;
3764 wr32(E1000_EICS, eics);
3766 wr32(E1000_ICS, E1000_ICS_RXDMT0);
3769 igb_spoof_check(adapter);
3771 /* Reset the timer */
3772 if (!test_bit(__IGB_DOWN, &adapter->state))
3773 mod_timer(&adapter->watchdog_timer,
3774 round_jiffies(jiffies + 2 * HZ));
3777 enum latency_range {
3781 latency_invalid = 255
3785 * igb_update_ring_itr - update the dynamic ITR value based on packet size
3787 * Stores a new ITR value based on strictly on packet size. This
3788 * algorithm is less sophisticated than that used in igb_update_itr,
3789 * due to the difficulty of synchronizing statistics across multiple
3790 * receive rings. The divisors and thresholds used by this function
3791 * were determined based on theoretical maximum wire speed and testing
3792 * data, in order to minimize response time while increasing bulk
3794 * This functionality is controlled by the InterruptThrottleRate module
3795 * parameter (see igb_param.c)
3796 * NOTE: This function is called only when operating in a multiqueue
3797 * receive environment.
3798 * @q_vector: pointer to q_vector
3800 static void igb_update_ring_itr(struct igb_q_vector *q_vector)
3802 int new_val = q_vector->itr_val;
3803 int avg_wire_size = 0;
3804 struct igb_adapter *adapter = q_vector->adapter;
3805 unsigned int packets;
3807 /* For non-gigabit speeds, just fix the interrupt rate at 4000
3808 * ints/sec - ITR timer value of 120 ticks.
3810 if (adapter->link_speed != SPEED_1000) {
3811 new_val = IGB_4K_ITR;
3815 packets = q_vector->rx.total_packets;
3817 avg_wire_size = q_vector->rx.total_bytes / packets;
3819 packets = q_vector->tx.total_packets;
3821 avg_wire_size = max_t(u32, avg_wire_size,
3822 q_vector->tx.total_bytes / packets);
3824 /* if avg_wire_size isn't set no work was done */
3828 /* Add 24 bytes to size to account for CRC, preamble, and gap */
3829 avg_wire_size += 24;
3831 /* Don't starve jumbo frames */
3832 avg_wire_size = min(avg_wire_size, 3000);
3834 /* Give a little boost to mid-size frames */
3835 if ((avg_wire_size > 300) && (avg_wire_size < 1200))
3836 new_val = avg_wire_size / 3;
3838 new_val = avg_wire_size / 2;
3840 /* conservative mode (itr 3) eliminates the lowest_latency setting */
3841 if (new_val < IGB_20K_ITR &&
3842 ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
3843 (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
3844 new_val = IGB_20K_ITR;
3847 if (new_val != q_vector->itr_val) {
3848 q_vector->itr_val = new_val;
3849 q_vector->set_itr = 1;
3852 q_vector->rx.total_bytes = 0;
3853 q_vector->rx.total_packets = 0;
3854 q_vector->tx.total_bytes = 0;
3855 q_vector->tx.total_packets = 0;
3859 * igb_update_itr - update the dynamic ITR value based on statistics
3860 * Stores a new ITR value based on packets and byte
3861 * counts during the last interrupt. The advantage of per interrupt
3862 * computation is faster updates and more accurate ITR for the current
3863 * traffic pattern. Constants in this function were computed
3864 * based on theoretical maximum wire speed and thresholds were set based
3865 * on testing data as well as attempting to minimize response time
3866 * while increasing bulk throughput.
3867 * this functionality is controlled by the InterruptThrottleRate module
3868 * parameter (see igb_param.c)
3869 * NOTE: These calculations are only valid when operating in a single-
3870 * queue environment.
3871 * @q_vector: pointer to q_vector
3872 * @ring_container: ring info to update the itr for
3874 static void igb_update_itr(struct igb_q_vector *q_vector,
3875 struct igb_ring_container *ring_container)
3877 unsigned int packets = ring_container->total_packets;
3878 unsigned int bytes = ring_container->total_bytes;
3879 u8 itrval = ring_container->itr;
3881 /* no packets, exit with status unchanged */
3886 case lowest_latency:
3887 /* handle TSO and jumbo frames */
3888 if (bytes/packets > 8000)
3889 itrval = bulk_latency;
3890 else if ((packets < 5) && (bytes > 512))
3891 itrval = low_latency;
3893 case low_latency: /* 50 usec aka 20000 ints/s */
3894 if (bytes > 10000) {
3895 /* this if handles the TSO accounting */
3896 if (bytes/packets > 8000) {
3897 itrval = bulk_latency;
3898 } else if ((packets < 10) || ((bytes/packets) > 1200)) {
3899 itrval = bulk_latency;
3900 } else if ((packets > 35)) {
3901 itrval = lowest_latency;
3903 } else if (bytes/packets > 2000) {
3904 itrval = bulk_latency;
3905 } else if (packets <= 2 && bytes < 512) {
3906 itrval = lowest_latency;
3909 case bulk_latency: /* 250 usec aka 4000 ints/s */
3910 if (bytes > 25000) {
3912 itrval = low_latency;
3913 } else if (bytes < 1500) {
3914 itrval = low_latency;
3919 /* clear work counters since we have the values we need */
3920 ring_container->total_bytes = 0;
3921 ring_container->total_packets = 0;
3923 /* write updated itr to ring container */
3924 ring_container->itr = itrval;
3927 static void igb_set_itr(struct igb_q_vector *q_vector)
3929 struct igb_adapter *adapter = q_vector->adapter;
3930 u32 new_itr = q_vector->itr_val;
3933 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
3934 if (adapter->link_speed != SPEED_1000) {
3936 new_itr = IGB_4K_ITR;
3940 igb_update_itr(q_vector, &q_vector->tx);
3941 igb_update_itr(q_vector, &q_vector->rx);
3943 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
3945 /* conservative mode (itr 3) eliminates the lowest_latency setting */
3946 if (current_itr == lowest_latency &&
3947 ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
3948 (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
3949 current_itr = low_latency;
3951 switch (current_itr) {
3952 /* counts and packets in update_itr are dependent on these numbers */
3953 case lowest_latency:
3954 new_itr = IGB_70K_ITR; /* 70,000 ints/sec */
3957 new_itr = IGB_20K_ITR; /* 20,000 ints/sec */
3960 new_itr = IGB_4K_ITR; /* 4,000 ints/sec */
3967 if (new_itr != q_vector->itr_val) {
3968 /* this attempts to bias the interrupt rate towards Bulk
3969 * by adding intermediate steps when interrupt rate is
3971 new_itr = new_itr > q_vector->itr_val ?
3972 max((new_itr * q_vector->itr_val) /
3973 (new_itr + (q_vector->itr_val >> 2)),
3976 /* Don't write the value here; it resets the adapter's
3977 * internal timer, and causes us to delay far longer than
3978 * we should between interrupts. Instead, we write the ITR
3979 * value at the beginning of the next interrupt so the timing
3980 * ends up being correct.
3982 q_vector->itr_val = new_itr;
3983 q_vector->set_itr = 1;
3987 static void igb_tx_ctxtdesc(struct igb_ring *tx_ring, u32 vlan_macip_lens,
3988 u32 type_tucmd, u32 mss_l4len_idx)
3990 struct e1000_adv_tx_context_desc *context_desc;
3991 u16 i = tx_ring->next_to_use;
3993 context_desc = IGB_TX_CTXTDESC(tx_ring, i);
3996 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
3998 /* set bits to identify this as an advanced context descriptor */
3999 type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
4001 /* For 82575, context index must be unique per ring. */
4002 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
4003 mss_l4len_idx |= tx_ring->reg_idx << 4;
4005 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
4006 context_desc->seqnum_seed = 0;
4007 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd);
4008 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
4011 static int igb_tso(struct igb_ring *tx_ring,
4012 struct igb_tx_buffer *first,
4015 struct sk_buff *skb = first->skb;
4016 u32 vlan_macip_lens, type_tucmd;
4017 u32 mss_l4len_idx, l4len;
4019 if (skb->ip_summed != CHECKSUM_PARTIAL)
4022 if (!skb_is_gso(skb))
4025 if (skb_header_cloned(skb)) {
4026 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4031 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
4032 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
4034 if (first->protocol == __constant_htons(ETH_P_IP)) {
4035 struct iphdr *iph = ip_hdr(skb);
4038 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
4042 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
4043 first->tx_flags |= IGB_TX_FLAGS_TSO |
4046 } else if (skb_is_gso_v6(skb)) {
4047 ipv6_hdr(skb)->payload_len = 0;
4048 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4049 &ipv6_hdr(skb)->daddr,
4051 first->tx_flags |= IGB_TX_FLAGS_TSO |
4055 /* compute header lengths */
4056 l4len = tcp_hdrlen(skb);
4057 *hdr_len = skb_transport_offset(skb) + l4len;
4059 /* update gso size and bytecount with header size */
4060 first->gso_segs = skb_shinfo(skb)->gso_segs;
4061 first->bytecount += (first->gso_segs - 1) * *hdr_len;
4064 mss_l4len_idx = l4len << E1000_ADVTXD_L4LEN_SHIFT;
4065 mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT;
4067 /* VLAN MACLEN IPLEN */
4068 vlan_macip_lens = skb_network_header_len(skb);
4069 vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
4070 vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
4072 igb_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
4077 static void igb_tx_csum(struct igb_ring *tx_ring, struct igb_tx_buffer *first)
4079 struct sk_buff *skb = first->skb;
4080 u32 vlan_macip_lens = 0;
4081 u32 mss_l4len_idx = 0;
4084 if (skb->ip_summed != CHECKSUM_PARTIAL) {
4085 if (!(first->tx_flags & IGB_TX_FLAGS_VLAN))
4089 switch (first->protocol) {
4090 case __constant_htons(ETH_P_IP):
4091 vlan_macip_lens |= skb_network_header_len(skb);
4092 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
4093 l4_hdr = ip_hdr(skb)->protocol;
4095 case __constant_htons(ETH_P_IPV6):
4096 vlan_macip_lens |= skb_network_header_len(skb);
4097 l4_hdr = ipv6_hdr(skb)->nexthdr;
4100 if (unlikely(net_ratelimit())) {
4101 dev_warn(tx_ring->dev,
4102 "partial checksum but proto=%x!\n",
4110 type_tucmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
4111 mss_l4len_idx = tcp_hdrlen(skb) <<
4112 E1000_ADVTXD_L4LEN_SHIFT;
4115 type_tucmd |= E1000_ADVTXD_TUCMD_L4T_SCTP;
4116 mss_l4len_idx = sizeof(struct sctphdr) <<
4117 E1000_ADVTXD_L4LEN_SHIFT;
4120 mss_l4len_idx = sizeof(struct udphdr) <<
4121 E1000_ADVTXD_L4LEN_SHIFT;
4124 if (unlikely(net_ratelimit())) {
4125 dev_warn(tx_ring->dev,
4126 "partial checksum but l4 proto=%x!\n",
4132 /* update TX checksum flag */
4133 first->tx_flags |= IGB_TX_FLAGS_CSUM;
4136 vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
4137 vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
4139 igb_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
4142 #define IGB_SET_FLAG(_input, _flag, _result) \
4143 ((_flag <= _result) ? \
4144 ((u32)(_input & _flag) * (_result / _flag)) : \
4145 ((u32)(_input & _flag) / (_flag / _result)))
4147 static u32 igb_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
4149 /* set type for advanced descriptor with frame checksum insertion */
4150 u32 cmd_type = E1000_ADVTXD_DTYP_DATA |
4151 E1000_ADVTXD_DCMD_DEXT |
4152 E1000_ADVTXD_DCMD_IFCS;
4154 /* set HW vlan bit if vlan is present */
4155 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_VLAN,
4156 (E1000_ADVTXD_DCMD_VLE));
4158 /* set segmentation bits for TSO */
4159 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSO,
4160 (E1000_ADVTXD_DCMD_TSE));
4162 /* set timestamp bit if present */
4163 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSTAMP,
4164 (E1000_ADVTXD_MAC_TSTAMP));
4166 /* insert frame checksum */
4167 cmd_type ^= IGB_SET_FLAG(skb->no_fcs, 1, E1000_ADVTXD_DCMD_IFCS);
4172 static void igb_tx_olinfo_status(struct igb_ring *tx_ring,
4173 union e1000_adv_tx_desc *tx_desc,
4174 u32 tx_flags, unsigned int paylen)
4176 u32 olinfo_status = paylen << E1000_ADVTXD_PAYLEN_SHIFT;
4178 /* 82575 requires a unique index per ring */
4179 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
4180 olinfo_status |= tx_ring->reg_idx << 4;
4182 /* insert L4 checksum */
4183 olinfo_status |= IGB_SET_FLAG(tx_flags,
4185 (E1000_TXD_POPTS_TXSM << 8));
4187 /* insert IPv4 checksum */
4188 olinfo_status |= IGB_SET_FLAG(tx_flags,
4190 (E1000_TXD_POPTS_IXSM << 8));
4192 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
4196 * The largest size we can write to the descriptor is 65535. In order to
4197 * maintain a power of two alignment we have to limit ourselves to 32K.
4199 #define IGB_MAX_TXD_PWR 15
4200 #define IGB_MAX_DATA_PER_TXD (1<<IGB_MAX_TXD_PWR)
4202 static void igb_tx_map(struct igb_ring *tx_ring,
4203 struct igb_tx_buffer *first,
4206 struct sk_buff *skb = first->skb;
4207 struct igb_tx_buffer *tx_buffer;
4208 union e1000_adv_tx_desc *tx_desc;
4209 struct skb_frag_struct *frag;
4211 unsigned int data_len, size;
4212 u32 tx_flags = first->tx_flags;
4213 u32 cmd_type = igb_tx_cmd_type(skb, tx_flags);
4214 u16 i = tx_ring->next_to_use;
4216 tx_desc = IGB_TX_DESC(tx_ring, i);
4218 igb_tx_olinfo_status(tx_ring, tx_desc, tx_flags, skb->len - hdr_len);
4220 size = skb_headlen(skb);
4221 data_len = skb->data_len;
4223 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
4227 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
4228 if (dma_mapping_error(tx_ring->dev, dma))
4231 /* record length, and DMA address */
4232 dma_unmap_len_set(tx_buffer, len, size);
4233 dma_unmap_addr_set(tx_buffer, dma, dma);
4235 tx_desc->read.buffer_addr = cpu_to_le64(dma);
4237 while (unlikely(size > IGB_MAX_DATA_PER_TXD)) {
4238 tx_desc->read.cmd_type_len =
4239 cpu_to_le32(cmd_type ^ IGB_MAX_DATA_PER_TXD);
4243 if (i == tx_ring->count) {
4244 tx_desc = IGB_TX_DESC(tx_ring, 0);
4247 tx_desc->read.olinfo_status = 0;
4249 dma += IGB_MAX_DATA_PER_TXD;
4250 size -= IGB_MAX_DATA_PER_TXD;
4252 tx_desc->read.buffer_addr = cpu_to_le64(dma);
4255 if (likely(!data_len))
4258 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
4262 if (i == tx_ring->count) {
4263 tx_desc = IGB_TX_DESC(tx_ring, 0);
4266 tx_desc->read.olinfo_status = 0;
4268 size = skb_frag_size(frag);
4271 dma = skb_frag_dma_map(tx_ring->dev, frag, 0,
4272 size, DMA_TO_DEVICE);
4274 tx_buffer = &tx_ring->tx_buffer_info[i];
4277 /* write last descriptor with RS and EOP bits */
4278 cmd_type |= size | IGB_TXD_DCMD;
4279 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
4281 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
4283 /* set the timestamp */
4284 first->time_stamp = jiffies;
4287 * Force memory writes to complete before letting h/w know there
4288 * are new descriptors to fetch. (Only applicable for weak-ordered
4289 * memory model archs, such as IA-64).
4291 * We also need this memory barrier to make certain all of the
4292 * status bits have been updated before next_to_watch is written.
4296 /* set next_to_watch value indicating a packet is present */
4297 first->next_to_watch = tx_desc;
4300 if (i == tx_ring->count)
4303 tx_ring->next_to_use = i;
4305 writel(i, tx_ring->tail);
4307 /* we need this if more than one processor can write to our tail
4308 * at a time, it syncronizes IO on IA64/Altix systems */
4314 dev_err(tx_ring->dev, "TX DMA map failed\n");
4316 /* clear dma mappings for failed tx_buffer_info map */
4318 tx_buffer = &tx_ring->tx_buffer_info[i];
4319 igb_unmap_and_free_tx_resource(tx_ring, tx_buffer);
4320 if (tx_buffer == first)
4327 tx_ring->next_to_use = i;
4330 static int __igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
4332 struct net_device *netdev = tx_ring->netdev;
4334 netif_stop_subqueue(netdev, tx_ring->queue_index);
4336 /* Herbert's original patch had:
4337 * smp_mb__after_netif_stop_queue();
4338 * but since that doesn't exist yet, just open code it. */
4341 /* We need to check again in a case another CPU has just
4342 * made room available. */
4343 if (igb_desc_unused(tx_ring) < size)
4347 netif_wake_subqueue(netdev, tx_ring->queue_index);
4349 u64_stats_update_begin(&tx_ring->tx_syncp2);
4350 tx_ring->tx_stats.restart_queue2++;
4351 u64_stats_update_end(&tx_ring->tx_syncp2);
4356 static inline int igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
4358 if (igb_desc_unused(tx_ring) >= size)
4360 return __igb_maybe_stop_tx(tx_ring, size);
4363 netdev_tx_t igb_xmit_frame_ring(struct sk_buff *skb,
4364 struct igb_ring *tx_ring)
4366 struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
4367 struct igb_tx_buffer *first;
4370 __be16 protocol = vlan_get_protocol(skb);
4373 /* need: 1 descriptor per page,
4374 * + 2 desc gap to keep tail from touching head,
4375 * + 1 desc for skb->data,
4376 * + 1 desc for context descriptor,
4377 * otherwise try next time */
4378 if (igb_maybe_stop_tx(tx_ring, skb_shinfo(skb)->nr_frags + 4)) {
4379 /* this is a hard error */
4380 return NETDEV_TX_BUSY;
4383 /* record the location of the first descriptor for this packet */
4384 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
4386 first->bytecount = skb->len;
4387 first->gso_segs = 1;
4389 if (unlikely((skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
4390 !(adapter->ptp_tx_skb))) {
4391 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
4392 tx_flags |= IGB_TX_FLAGS_TSTAMP;
4394 adapter->ptp_tx_skb = skb_get(skb);
4395 if (adapter->hw.mac.type == e1000_82576)
4396 schedule_work(&adapter->ptp_tx_work);
4399 if (vlan_tx_tag_present(skb)) {
4400 tx_flags |= IGB_TX_FLAGS_VLAN;
4401 tx_flags |= (vlan_tx_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
4404 /* record initial flags and protocol */
4405 first->tx_flags = tx_flags;
4406 first->protocol = protocol;
4408 tso = igb_tso(tx_ring, first, &hdr_len);
4412 igb_tx_csum(tx_ring, first);
4414 igb_tx_map(tx_ring, first, hdr_len);
4416 /* Make sure there is space in the ring for the next send. */
4417 igb_maybe_stop_tx(tx_ring, MAX_SKB_FRAGS + 4);
4419 return NETDEV_TX_OK;
4422 igb_unmap_and_free_tx_resource(tx_ring, first);
4424 return NETDEV_TX_OK;
4427 static inline struct igb_ring *igb_tx_queue_mapping(struct igb_adapter *adapter,
4428 struct sk_buff *skb)
4430 unsigned int r_idx = skb->queue_mapping;
4432 if (r_idx >= adapter->num_tx_queues)
4433 r_idx = r_idx % adapter->num_tx_queues;
4435 return adapter->tx_ring[r_idx];
4438 static netdev_tx_t igb_xmit_frame(struct sk_buff *skb,
4439 struct net_device *netdev)
4441 struct igb_adapter *adapter = netdev_priv(netdev);
4443 if (test_bit(__IGB_DOWN, &adapter->state)) {
4444 dev_kfree_skb_any(skb);
4445 return NETDEV_TX_OK;
4448 if (skb->len <= 0) {
4449 dev_kfree_skb_any(skb);
4450 return NETDEV_TX_OK;
4454 * The minimum packet size with TCTL.PSP set is 17 so pad the skb
4455 * in order to meet this minimum size requirement.
4457 if (unlikely(skb->len < 17)) {
4458 if (skb_pad(skb, 17 - skb->len))
4459 return NETDEV_TX_OK;
4461 skb_set_tail_pointer(skb, 17);
4464 return igb_xmit_frame_ring(skb, igb_tx_queue_mapping(adapter, skb));
4468 * igb_tx_timeout - Respond to a Tx Hang
4469 * @netdev: network interface device structure
4471 static void igb_tx_timeout(struct net_device *netdev)
4473 struct igb_adapter *adapter = netdev_priv(netdev);
4474 struct e1000_hw *hw = &adapter->hw;
4476 /* Do the reset outside of interrupt context */
4477 adapter->tx_timeout_count++;
4479 if (hw->mac.type >= e1000_82580)
4480 hw->dev_spec._82575.global_device_reset = true;
4482 schedule_work(&adapter->reset_task);
4484 (adapter->eims_enable_mask & ~adapter->eims_other));
4487 static void igb_reset_task(struct work_struct *work)
4489 struct igb_adapter *adapter;
4490 adapter = container_of(work, struct igb_adapter, reset_task);
4493 netdev_err(adapter->netdev, "Reset adapter\n");
4494 igb_reinit_locked(adapter);
4498 * igb_get_stats64 - Get System Network Statistics
4499 * @netdev: network interface device structure
4500 * @stats: rtnl_link_stats64 pointer
4503 static struct rtnl_link_stats64 *igb_get_stats64(struct net_device *netdev,
4504 struct rtnl_link_stats64 *stats)
4506 struct igb_adapter *adapter = netdev_priv(netdev);
4508 spin_lock(&adapter->stats64_lock);
4509 igb_update_stats(adapter, &adapter->stats64);
4510 memcpy(stats, &adapter->stats64, sizeof(*stats));
4511 spin_unlock(&adapter->stats64_lock);
4517 * igb_change_mtu - Change the Maximum Transfer Unit
4518 * @netdev: network interface device structure
4519 * @new_mtu: new value for maximum frame size
4521 * Returns 0 on success, negative on failure
4523 static int igb_change_mtu(struct net_device *netdev, int new_mtu)
4525 struct igb_adapter *adapter = netdev_priv(netdev);
4526 struct pci_dev *pdev = adapter->pdev;
4527 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
4529 if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
4530 dev_err(&pdev->dev, "Invalid MTU setting\n");
4534 #define MAX_STD_JUMBO_FRAME_SIZE 9238
4535 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
4536 dev_err(&pdev->dev, "MTU > 9216 not supported.\n");
4540 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
4543 /* igb_down has a dependency on max_frame_size */
4544 adapter->max_frame_size = max_frame;
4546 if (netif_running(netdev))
4549 dev_info(&pdev->dev, "changing MTU from %d to %d\n",
4550 netdev->mtu, new_mtu);
4551 netdev->mtu = new_mtu;
4553 if (netif_running(netdev))
4558 clear_bit(__IGB_RESETTING, &adapter->state);
4564 * igb_update_stats - Update the board statistics counters
4565 * @adapter: board private structure
4568 void igb_update_stats(struct igb_adapter *adapter,
4569 struct rtnl_link_stats64 *net_stats)
4571 struct e1000_hw *hw = &adapter->hw;
4572 struct pci_dev *pdev = adapter->pdev;
4578 u64 _bytes, _packets;
4580 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
4583 * Prevent stats update while adapter is being reset, or if the pci
4584 * connection is down.
4586 if (adapter->link_speed == 0)
4588 if (pci_channel_offline(pdev))
4593 for (i = 0; i < adapter->num_rx_queues; i++) {
4594 u32 rqdpc = rd32(E1000_RQDPC(i));
4595 struct igb_ring *ring = adapter->rx_ring[i];
4598 ring->rx_stats.drops += rqdpc;
4599 net_stats->rx_fifo_errors += rqdpc;
4603 start = u64_stats_fetch_begin_bh(&ring->rx_syncp);
4604 _bytes = ring->rx_stats.bytes;
4605 _packets = ring->rx_stats.packets;
4606 } while (u64_stats_fetch_retry_bh(&ring->rx_syncp, start));
4608 packets += _packets;
4611 net_stats->rx_bytes = bytes;
4612 net_stats->rx_packets = packets;
4616 for (i = 0; i < adapter->num_tx_queues; i++) {
4617 struct igb_ring *ring = adapter->tx_ring[i];
4619 start = u64_stats_fetch_begin_bh(&ring->tx_syncp);
4620 _bytes = ring->tx_stats.bytes;
4621 _packets = ring->tx_stats.packets;
4622 } while (u64_stats_fetch_retry_bh(&ring->tx_syncp, start));
4624 packets += _packets;
4626 net_stats->tx_bytes = bytes;
4627 net_stats->tx_packets = packets;
4629 /* read stats registers */
4630 adapter->stats.crcerrs += rd32(E1000_CRCERRS);
4631 adapter->stats.gprc += rd32(E1000_GPRC);
4632 adapter->stats.gorc += rd32(E1000_GORCL);
4633 rd32(E1000_GORCH); /* clear GORCL */
4634 adapter->stats.bprc += rd32(E1000_BPRC);
4635 adapter->stats.mprc += rd32(E1000_MPRC);
4636 adapter->stats.roc += rd32(E1000_ROC);
4638 adapter->stats.prc64 += rd32(E1000_PRC64);
4639 adapter->stats.prc127 += rd32(E1000_PRC127);
4640 adapter->stats.prc255 += rd32(E1000_PRC255);
4641 adapter->stats.prc511 += rd32(E1000_PRC511);
4642 adapter->stats.prc1023 += rd32(E1000_PRC1023);
4643 adapter->stats.prc1522 += rd32(E1000_PRC1522);
4644 adapter->stats.symerrs += rd32(E1000_SYMERRS);
4645 adapter->stats.sec += rd32(E1000_SEC);
4647 mpc = rd32(E1000_MPC);
4648 adapter->stats.mpc += mpc;
4649 net_stats->rx_fifo_errors += mpc;
4650 adapter->stats.scc += rd32(E1000_SCC);
4651 adapter->stats.ecol += rd32(E1000_ECOL);
4652 adapter->stats.mcc += rd32(E1000_MCC);
4653 adapter->stats.latecol += rd32(E1000_LATECOL);
4654 adapter->stats.dc += rd32(E1000_DC);
4655 adapter->stats.rlec += rd32(E1000_RLEC);
4656 adapter->stats.xonrxc += rd32(E1000_XONRXC);
4657 adapter->stats.xontxc += rd32(E1000_XONTXC);
4658 adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
4659 adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
4660 adapter->stats.fcruc += rd32(E1000_FCRUC);
4661 adapter->stats.gptc += rd32(E1000_GPTC);
4662 adapter->stats.gotc += rd32(E1000_GOTCL);
4663 rd32(E1000_GOTCH); /* clear GOTCL */
4664 adapter->stats.rnbc += rd32(E1000_RNBC);
4665 adapter->stats.ruc += rd32(E1000_RUC);
4666 adapter->stats.rfc += rd32(E1000_RFC);
4667 adapter->stats.rjc += rd32(E1000_RJC);
4668 adapter->stats.tor += rd32(E1000_TORH);
4669 adapter->stats.tot += rd32(E1000_TOTH);
4670 adapter->stats.tpr += rd32(E1000_TPR);
4672 adapter->stats.ptc64 += rd32(E1000_PTC64);
4673 adapter->stats.ptc127 += rd32(E1000_PTC127);
4674 adapter->stats.ptc255 += rd32(E1000_PTC255);
4675 adapter->stats.ptc511 += rd32(E1000_PTC511);
4676 adapter->stats.ptc1023 += rd32(E1000_PTC1023);
4677 adapter->stats.ptc1522 += rd32(E1000_PTC1522);
4679 adapter->stats.mptc += rd32(E1000_MPTC);
4680 adapter->stats.bptc += rd32(E1000_BPTC);
4682 adapter->stats.tpt += rd32(E1000_TPT);
4683 adapter->stats.colc += rd32(E1000_COLC);
4685 adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
4686 /* read internal phy specific stats */
4687 reg = rd32(E1000_CTRL_EXT);
4688 if (!(reg & E1000_CTRL_EXT_LINK_MODE_MASK)) {
4689 adapter->stats.rxerrc += rd32(E1000_RXERRC);
4691 /* this stat has invalid values on i210/i211 */
4692 if ((hw->mac.type != e1000_i210) &&
4693 (hw->mac.type != e1000_i211))
4694 adapter->stats.tncrs += rd32(E1000_TNCRS);
4697 adapter->stats.tsctc += rd32(E1000_TSCTC);
4698 adapter->stats.tsctfc += rd32(E1000_TSCTFC);
4700 adapter->stats.iac += rd32(E1000_IAC);
4701 adapter->stats.icrxoc += rd32(E1000_ICRXOC);
4702 adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
4703 adapter->stats.icrxatc += rd32(E1000_ICRXATC);
4704 adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
4705 adapter->stats.ictxatc += rd32(E1000_ICTXATC);
4706 adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
4707 adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
4708 adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
4710 /* Fill out the OS statistics structure */
4711 net_stats->multicast = adapter->stats.mprc;
4712 net_stats->collisions = adapter->stats.colc;
4716 /* RLEC on some newer hardware can be incorrect so build
4717 * our own version based on RUC and ROC */
4718 net_stats->rx_errors = adapter->stats.rxerrc +
4719 adapter->stats.crcerrs + adapter->stats.algnerrc +
4720 adapter->stats.ruc + adapter->stats.roc +
4721 adapter->stats.cexterr;
4722 net_stats->rx_length_errors = adapter->stats.ruc +
4724 net_stats->rx_crc_errors = adapter->stats.crcerrs;
4725 net_stats->rx_frame_errors = adapter->stats.algnerrc;
4726 net_stats->rx_missed_errors = adapter->stats.mpc;
4729 net_stats->tx_errors = adapter->stats.ecol +
4730 adapter->stats.latecol;
4731 net_stats->tx_aborted_errors = adapter->stats.ecol;
4732 net_stats->tx_window_errors = adapter->stats.latecol;
4733 net_stats->tx_carrier_errors = adapter->stats.tncrs;
4735 /* Tx Dropped needs to be maintained elsewhere */
4738 if (hw->phy.media_type == e1000_media_type_copper) {
4739 if ((adapter->link_speed == SPEED_1000) &&
4740 (!igb_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
4741 phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
4742 adapter->phy_stats.idle_errors += phy_tmp;
4746 /* Management Stats */
4747 adapter->stats.mgptc += rd32(E1000_MGTPTC);
4748 adapter->stats.mgprc += rd32(E1000_MGTPRC);
4749 adapter->stats.mgpdc += rd32(E1000_MGTPDC);
4752 reg = rd32(E1000_MANC);
4753 if (reg & E1000_MANC_EN_BMC2OS) {
4754 adapter->stats.o2bgptc += rd32(E1000_O2BGPTC);
4755 adapter->stats.o2bspc += rd32(E1000_O2BSPC);
4756 adapter->stats.b2ospc += rd32(E1000_B2OSPC);
4757 adapter->stats.b2ogprc += rd32(E1000_B2OGPRC);
4761 static irqreturn_t igb_msix_other(int irq, void *data)
4763 struct igb_adapter *adapter = data;
4764 struct e1000_hw *hw = &adapter->hw;
4765 u32 icr = rd32(E1000_ICR);
4766 /* reading ICR causes bit 31 of EICR to be cleared */
4768 if (icr & E1000_ICR_DRSTA)
4769 schedule_work(&adapter->reset_task);
4771 if (icr & E1000_ICR_DOUTSYNC) {
4772 /* HW is reporting DMA is out of sync */
4773 adapter->stats.doosync++;
4774 /* The DMA Out of Sync is also indication of a spoof event
4775 * in IOV mode. Check the Wrong VM Behavior register to
4776 * see if it is really a spoof event. */
4777 igb_check_wvbr(adapter);
4780 /* Check for a mailbox event */
4781 if (icr & E1000_ICR_VMMB)
4782 igb_msg_task(adapter);
4784 if (icr & E1000_ICR_LSC) {
4785 hw->mac.get_link_status = 1;
4786 /* guard against interrupt when we're going down */
4787 if (!test_bit(__IGB_DOWN, &adapter->state))
4788 mod_timer(&adapter->watchdog_timer, jiffies + 1);
4791 if (icr & E1000_ICR_TS) {
4792 u32 tsicr = rd32(E1000_TSICR);
4794 if (tsicr & E1000_TSICR_TXTS) {
4795 /* acknowledge the interrupt */
4796 wr32(E1000_TSICR, E1000_TSICR_TXTS);
4797 /* retrieve hardware timestamp */
4798 schedule_work(&adapter->ptp_tx_work);
4802 wr32(E1000_EIMS, adapter->eims_other);
4807 static void igb_write_itr(struct igb_q_vector *q_vector)
4809 struct igb_adapter *adapter = q_vector->adapter;
4810 u32 itr_val = q_vector->itr_val & 0x7FFC;
4812 if (!q_vector->set_itr)
4818 if (adapter->hw.mac.type == e1000_82575)
4819 itr_val |= itr_val << 16;
4821 itr_val |= E1000_EITR_CNT_IGNR;
4823 writel(itr_val, q_vector->itr_register);
4824 q_vector->set_itr = 0;
4827 static irqreturn_t igb_msix_ring(int irq, void *data)
4829 struct igb_q_vector *q_vector = data;
4831 /* Write the ITR value calculated from the previous interrupt. */
4832 igb_write_itr(q_vector);
4834 napi_schedule(&q_vector->napi);
4839 #ifdef CONFIG_IGB_DCA
4840 static void igb_update_tx_dca(struct igb_adapter *adapter,
4841 struct igb_ring *tx_ring,
4844 struct e1000_hw *hw = &adapter->hw;
4845 u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
4847 if (hw->mac.type != e1000_82575)
4848 txctrl <<= E1000_DCA_TXCTRL_CPUID_SHIFT;
4851 * We can enable relaxed ordering for reads, but not writes when
4852 * DCA is enabled. This is due to a known issue in some chipsets
4853 * which will cause the DCA tag to be cleared.
4855 txctrl |= E1000_DCA_TXCTRL_DESC_RRO_EN |
4856 E1000_DCA_TXCTRL_DATA_RRO_EN |
4857 E1000_DCA_TXCTRL_DESC_DCA_EN;
4859 wr32(E1000_DCA_TXCTRL(tx_ring->reg_idx), txctrl);
4862 static void igb_update_rx_dca(struct igb_adapter *adapter,
4863 struct igb_ring *rx_ring,
4866 struct e1000_hw *hw = &adapter->hw;
4867 u32 rxctrl = dca3_get_tag(&adapter->pdev->dev, cpu);
4869 if (hw->mac.type != e1000_82575)
4870 rxctrl <<= E1000_DCA_RXCTRL_CPUID_SHIFT;
4873 * We can enable relaxed ordering for reads, but not writes when
4874 * DCA is enabled. This is due to a known issue in some chipsets
4875 * which will cause the DCA tag to be cleared.
4877 rxctrl |= E1000_DCA_RXCTRL_DESC_RRO_EN |
4878 E1000_DCA_RXCTRL_DESC_DCA_EN;
4880 wr32(E1000_DCA_RXCTRL(rx_ring->reg_idx), rxctrl);
4883 static void igb_update_dca(struct igb_q_vector *q_vector)
4885 struct igb_adapter *adapter = q_vector->adapter;
4886 int cpu = get_cpu();
4888 if (q_vector->cpu == cpu)
4891 if (q_vector->tx.ring)
4892 igb_update_tx_dca(adapter, q_vector->tx.ring, cpu);
4894 if (q_vector->rx.ring)
4895 igb_update_rx_dca(adapter, q_vector->rx.ring, cpu);
4897 q_vector->cpu = cpu;
4902 static void igb_setup_dca(struct igb_adapter *adapter)
4904 struct e1000_hw *hw = &adapter->hw;
4907 if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
4910 /* Always use CB2 mode, difference is masked in the CB driver. */
4911 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
4913 for (i = 0; i < adapter->num_q_vectors; i++) {
4914 adapter->q_vector[i]->cpu = -1;
4915 igb_update_dca(adapter->q_vector[i]);
4919 static int __igb_notify_dca(struct device *dev, void *data)
4921 struct net_device *netdev = dev_get_drvdata(dev);
4922 struct igb_adapter *adapter = netdev_priv(netdev);
4923 struct pci_dev *pdev = adapter->pdev;
4924 struct e1000_hw *hw = &adapter->hw;
4925 unsigned long event = *(unsigned long *)data;
4928 case DCA_PROVIDER_ADD:
4929 /* if already enabled, don't do it again */
4930 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
4932 if (dca_add_requester(dev) == 0) {
4933 adapter->flags |= IGB_FLAG_DCA_ENABLED;
4934 dev_info(&pdev->dev, "DCA enabled\n");
4935 igb_setup_dca(adapter);
4938 /* Fall Through since DCA is disabled. */
4939 case DCA_PROVIDER_REMOVE:
4940 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
4941 /* without this a class_device is left
4942 * hanging around in the sysfs model */
4943 dca_remove_requester(dev);
4944 dev_info(&pdev->dev, "DCA disabled\n");
4945 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
4946 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
4954 static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
4959 ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
4962 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
4964 #endif /* CONFIG_IGB_DCA */
4966 #ifdef CONFIG_PCI_IOV
4967 static int igb_vf_configure(struct igb_adapter *adapter, int vf)
4969 unsigned char mac_addr[ETH_ALEN];
4971 eth_random_addr(mac_addr);
4972 igb_set_vf_mac(adapter, vf, mac_addr);
4977 static bool igb_vfs_are_assigned(struct igb_adapter *adapter)
4979 struct pci_dev *pdev = adapter->pdev;
4980 struct pci_dev *vfdev;
4983 switch (adapter->hw.mac.type) {
4985 dev_id = IGB_82576_VF_DEV_ID;
4988 dev_id = IGB_I350_VF_DEV_ID;
4994 /* loop through all the VFs to see if we own any that are assigned */
4995 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, dev_id, NULL);
4997 /* if we don't own it we don't care */
4998 if (vfdev->is_virtfn && vfdev->physfn == pdev) {
4999 /* if it is assigned we cannot release it */
5000 if (vfdev->dev_flags & PCI_DEV_FLAGS_ASSIGNED)
5004 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, dev_id, vfdev);
5011 static void igb_ping_all_vfs(struct igb_adapter *adapter)
5013 struct e1000_hw *hw = &adapter->hw;
5017 for (i = 0 ; i < adapter->vfs_allocated_count; i++) {
5018 ping = E1000_PF_CONTROL_MSG;
5019 if (adapter->vf_data[i].flags & IGB_VF_FLAG_CTS)
5020 ping |= E1000_VT_MSGTYPE_CTS;
5021 igb_write_mbx(hw, &ping, 1, i);
5025 static int igb_set_vf_promisc(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
5027 struct e1000_hw *hw = &adapter->hw;
5028 u32 vmolr = rd32(E1000_VMOLR(vf));
5029 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5031 vf_data->flags &= ~(IGB_VF_FLAG_UNI_PROMISC |
5032 IGB_VF_FLAG_MULTI_PROMISC);
5033 vmolr &= ~(E1000_VMOLR_ROPE | E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
5035 if (*msgbuf & E1000_VF_SET_PROMISC_MULTICAST) {
5036 vmolr |= E1000_VMOLR_MPME;
5037 vf_data->flags |= IGB_VF_FLAG_MULTI_PROMISC;
5038 *msgbuf &= ~E1000_VF_SET_PROMISC_MULTICAST;
5041 * if we have hashes and we are clearing a multicast promisc
5042 * flag we need to write the hashes to the MTA as this step
5043 * was previously skipped
5045 if (vf_data->num_vf_mc_hashes > 30) {
5046 vmolr |= E1000_VMOLR_MPME;
5047 } else if (vf_data->num_vf_mc_hashes) {
5049 vmolr |= E1000_VMOLR_ROMPE;
5050 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
5051 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
5055 wr32(E1000_VMOLR(vf), vmolr);
5057 /* there are flags left unprocessed, likely not supported */
5058 if (*msgbuf & E1000_VT_MSGINFO_MASK)
5065 static int igb_set_vf_multicasts(struct igb_adapter *adapter,
5066 u32 *msgbuf, u32 vf)
5068 int n = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
5069 u16 *hash_list = (u16 *)&msgbuf[1];
5070 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5073 /* salt away the number of multicast addresses assigned
5074 * to this VF for later use to restore when the PF multi cast
5077 vf_data->num_vf_mc_hashes = n;
5079 /* only up to 30 hash values supported */
5083 /* store the hashes for later use */
5084 for (i = 0; i < n; i++)
5085 vf_data->vf_mc_hashes[i] = hash_list[i];
5087 /* Flush and reset the mta with the new values */
5088 igb_set_rx_mode(adapter->netdev);
5093 static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
5095 struct e1000_hw *hw = &adapter->hw;
5096 struct vf_data_storage *vf_data;
5099 for (i = 0; i < adapter->vfs_allocated_count; i++) {
5100 u32 vmolr = rd32(E1000_VMOLR(i));
5101 vmolr &= ~(E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
5103 vf_data = &adapter->vf_data[i];
5105 if ((vf_data->num_vf_mc_hashes > 30) ||
5106 (vf_data->flags & IGB_VF_FLAG_MULTI_PROMISC)) {
5107 vmolr |= E1000_VMOLR_MPME;
5108 } else if (vf_data->num_vf_mc_hashes) {
5109 vmolr |= E1000_VMOLR_ROMPE;
5110 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
5111 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
5113 wr32(E1000_VMOLR(i), vmolr);
5117 static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
5119 struct e1000_hw *hw = &adapter->hw;
5120 u32 pool_mask, reg, vid;
5123 pool_mask = 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
5125 /* Find the vlan filter for this id */
5126 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5127 reg = rd32(E1000_VLVF(i));
5129 /* remove the vf from the pool */
5132 /* if pool is empty then remove entry from vfta */
5133 if (!(reg & E1000_VLVF_POOLSEL_MASK) &&
5134 (reg & E1000_VLVF_VLANID_ENABLE)) {
5136 vid = reg & E1000_VLVF_VLANID_MASK;
5137 igb_vfta_set(hw, vid, false);
5140 wr32(E1000_VLVF(i), reg);
5143 adapter->vf_data[vf].vlans_enabled = 0;
5146 static s32 igb_vlvf_set(struct igb_adapter *adapter, u32 vid, bool add, u32 vf)
5148 struct e1000_hw *hw = &adapter->hw;
5151 /* The vlvf table only exists on 82576 hardware and newer */
5152 if (hw->mac.type < e1000_82576)
5155 /* we only need to do this if VMDq is enabled */
5156 if (!adapter->vfs_allocated_count)
5159 /* Find the vlan filter for this id */
5160 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5161 reg = rd32(E1000_VLVF(i));
5162 if ((reg & E1000_VLVF_VLANID_ENABLE) &&
5163 vid == (reg & E1000_VLVF_VLANID_MASK))
5168 if (i == E1000_VLVF_ARRAY_SIZE) {
5169 /* Did not find a matching VLAN ID entry that was
5170 * enabled. Search for a free filter entry, i.e.
5171 * one without the enable bit set
5173 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5174 reg = rd32(E1000_VLVF(i));
5175 if (!(reg & E1000_VLVF_VLANID_ENABLE))
5179 if (i < E1000_VLVF_ARRAY_SIZE) {
5180 /* Found an enabled/available entry */
5181 reg |= 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
5183 /* if !enabled we need to set this up in vfta */
5184 if (!(reg & E1000_VLVF_VLANID_ENABLE)) {
5185 /* add VID to filter table */
5186 igb_vfta_set(hw, vid, true);
5187 reg |= E1000_VLVF_VLANID_ENABLE;
5189 reg &= ~E1000_VLVF_VLANID_MASK;
5191 wr32(E1000_VLVF(i), reg);
5193 /* do not modify RLPML for PF devices */
5194 if (vf >= adapter->vfs_allocated_count)
5197 if (!adapter->vf_data[vf].vlans_enabled) {
5199 reg = rd32(E1000_VMOLR(vf));
5200 size = reg & E1000_VMOLR_RLPML_MASK;
5202 reg &= ~E1000_VMOLR_RLPML_MASK;
5204 wr32(E1000_VMOLR(vf), reg);
5207 adapter->vf_data[vf].vlans_enabled++;
5210 if (i < E1000_VLVF_ARRAY_SIZE) {
5211 /* remove vf from the pool */
5212 reg &= ~(1 << (E1000_VLVF_POOLSEL_SHIFT + vf));
5213 /* if pool is empty then remove entry from vfta */
5214 if (!(reg & E1000_VLVF_POOLSEL_MASK)) {
5216 igb_vfta_set(hw, vid, false);
5218 wr32(E1000_VLVF(i), reg);
5220 /* do not modify RLPML for PF devices */
5221 if (vf >= adapter->vfs_allocated_count)
5224 adapter->vf_data[vf].vlans_enabled--;
5225 if (!adapter->vf_data[vf].vlans_enabled) {
5227 reg = rd32(E1000_VMOLR(vf));
5228 size = reg & E1000_VMOLR_RLPML_MASK;
5230 reg &= ~E1000_VMOLR_RLPML_MASK;
5232 wr32(E1000_VMOLR(vf), reg);
5239 static void igb_set_vmvir(struct igb_adapter *adapter, u32 vid, u32 vf)
5241 struct e1000_hw *hw = &adapter->hw;
5244 wr32(E1000_VMVIR(vf), (vid | E1000_VMVIR_VLANA_DEFAULT));
5246 wr32(E1000_VMVIR(vf), 0);
5249 static int igb_ndo_set_vf_vlan(struct net_device *netdev,
5250 int vf, u16 vlan, u8 qos)
5253 struct igb_adapter *adapter = netdev_priv(netdev);
5255 if ((vf >= adapter->vfs_allocated_count) || (vlan > 4095) || (qos > 7))
5258 err = igb_vlvf_set(adapter, vlan, !!vlan, vf);
5261 igb_set_vmvir(adapter, vlan | (qos << VLAN_PRIO_SHIFT), vf);
5262 igb_set_vmolr(adapter, vf, !vlan);
5263 adapter->vf_data[vf].pf_vlan = vlan;
5264 adapter->vf_data[vf].pf_qos = qos;
5265 dev_info(&adapter->pdev->dev,
5266 "Setting VLAN %d, QOS 0x%x on VF %d\n", vlan, qos, vf);
5267 if (test_bit(__IGB_DOWN, &adapter->state)) {
5268 dev_warn(&adapter->pdev->dev,
5269 "The VF VLAN has been set,"
5270 " but the PF device is not up.\n");
5271 dev_warn(&adapter->pdev->dev,
5272 "Bring the PF device up before"
5273 " attempting to use the VF device.\n");
5276 igb_vlvf_set(adapter, adapter->vf_data[vf].pf_vlan,
5278 igb_set_vmvir(adapter, vlan, vf);
5279 igb_set_vmolr(adapter, vf, true);
5280 adapter->vf_data[vf].pf_vlan = 0;
5281 adapter->vf_data[vf].pf_qos = 0;
5287 static int igb_set_vf_vlan(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
5289 int add = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
5290 int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
5292 return igb_vlvf_set(adapter, vid, add, vf);
5295 static inline void igb_vf_reset(struct igb_adapter *adapter, u32 vf)
5297 /* clear flags - except flag that indicates PF has set the MAC */
5298 adapter->vf_data[vf].flags &= IGB_VF_FLAG_PF_SET_MAC;
5299 adapter->vf_data[vf].last_nack = jiffies;
5301 /* reset offloads to defaults */
5302 igb_set_vmolr(adapter, vf, true);
5304 /* reset vlans for device */
5305 igb_clear_vf_vfta(adapter, vf);
5306 if (adapter->vf_data[vf].pf_vlan)
5307 igb_ndo_set_vf_vlan(adapter->netdev, vf,
5308 adapter->vf_data[vf].pf_vlan,
5309 adapter->vf_data[vf].pf_qos);
5311 igb_clear_vf_vfta(adapter, vf);
5313 /* reset multicast table array for vf */
5314 adapter->vf_data[vf].num_vf_mc_hashes = 0;
5316 /* Flush and reset the mta with the new values */
5317 igb_set_rx_mode(adapter->netdev);
5320 static void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
5322 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
5324 /* generate a new mac address as we were hotplug removed/added */
5325 if (!(adapter->vf_data[vf].flags & IGB_VF_FLAG_PF_SET_MAC))
5326 eth_random_addr(vf_mac);
5328 /* process remaining reset events */
5329 igb_vf_reset(adapter, vf);
5332 static void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
5334 struct e1000_hw *hw = &adapter->hw;
5335 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
5336 int rar_entry = hw->mac.rar_entry_count - (vf + 1);
5338 u8 *addr = (u8 *)(&msgbuf[1]);
5340 /* process all the same items cleared in a function level reset */
5341 igb_vf_reset(adapter, vf);
5343 /* set vf mac address */
5344 igb_rar_set_qsel(adapter, vf_mac, rar_entry, vf);
5346 /* enable transmit and receive for vf */
5347 reg = rd32(E1000_VFTE);
5348 wr32(E1000_VFTE, reg | (1 << vf));
5349 reg = rd32(E1000_VFRE);
5350 wr32(E1000_VFRE, reg | (1 << vf));
5352 adapter->vf_data[vf].flags |= IGB_VF_FLAG_CTS;
5354 /* reply to reset with ack and vf mac address */
5355 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
5356 memcpy(addr, vf_mac, 6);
5357 igb_write_mbx(hw, msgbuf, 3, vf);
5360 static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf)
5363 * The VF MAC Address is stored in a packed array of bytes
5364 * starting at the second 32 bit word of the msg array
5366 unsigned char *addr = (char *)&msg[1];
5369 if (is_valid_ether_addr(addr))
5370 err = igb_set_vf_mac(adapter, vf, addr);
5375 static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf)
5377 struct e1000_hw *hw = &adapter->hw;
5378 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5379 u32 msg = E1000_VT_MSGTYPE_NACK;
5381 /* if device isn't clear to send it shouldn't be reading either */
5382 if (!(vf_data->flags & IGB_VF_FLAG_CTS) &&
5383 time_after(jiffies, vf_data->last_nack + (2 * HZ))) {
5384 igb_write_mbx(hw, &msg, 1, vf);
5385 vf_data->last_nack = jiffies;
5389 static void igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
5391 struct pci_dev *pdev = adapter->pdev;
5392 u32 msgbuf[E1000_VFMAILBOX_SIZE];
5393 struct e1000_hw *hw = &adapter->hw;
5394 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5397 retval = igb_read_mbx(hw, msgbuf, E1000_VFMAILBOX_SIZE, vf);
5400 /* if receive failed revoke VF CTS stats and restart init */
5401 dev_err(&pdev->dev, "Error receiving message from VF\n");
5402 vf_data->flags &= ~IGB_VF_FLAG_CTS;
5403 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
5408 /* this is a message we already processed, do nothing */
5409 if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
5413 * until the vf completes a reset it should not be
5414 * allowed to start any configuration.
5417 if (msgbuf[0] == E1000_VF_RESET) {
5418 igb_vf_reset_msg(adapter, vf);
5422 if (!(vf_data->flags & IGB_VF_FLAG_CTS)) {
5423 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
5429 switch ((msgbuf[0] & 0xFFFF)) {
5430 case E1000_VF_SET_MAC_ADDR:
5432 if (!(vf_data->flags & IGB_VF_FLAG_PF_SET_MAC))
5433 retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
5435 dev_warn(&pdev->dev,
5436 "VF %d attempted to override administratively "
5437 "set MAC address\nReload the VF driver to "
5438 "resume operations\n", vf);
5440 case E1000_VF_SET_PROMISC:
5441 retval = igb_set_vf_promisc(adapter, msgbuf, vf);
5443 case E1000_VF_SET_MULTICAST:
5444 retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
5446 case E1000_VF_SET_LPE:
5447 retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf);
5449 case E1000_VF_SET_VLAN:
5451 if (vf_data->pf_vlan)
5452 dev_warn(&pdev->dev,
5453 "VF %d attempted to override administratively "
5454 "set VLAN tag\nReload the VF driver to "
5455 "resume operations\n", vf);
5457 retval = igb_set_vf_vlan(adapter, msgbuf, vf);
5460 dev_err(&pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
5465 msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
5467 /* notify the VF of the results of what it sent us */
5469 msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
5471 msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
5473 igb_write_mbx(hw, msgbuf, 1, vf);
5476 static void igb_msg_task(struct igb_adapter *adapter)
5478 struct e1000_hw *hw = &adapter->hw;
5481 for (vf = 0; vf < adapter->vfs_allocated_count; vf++) {
5482 /* process any reset requests */
5483 if (!igb_check_for_rst(hw, vf))
5484 igb_vf_reset_event(adapter, vf);
5486 /* process any messages pending */
5487 if (!igb_check_for_msg(hw, vf))
5488 igb_rcv_msg_from_vf(adapter, vf);
5490 /* process any acks */
5491 if (!igb_check_for_ack(hw, vf))
5492 igb_rcv_ack_from_vf(adapter, vf);
5497 * igb_set_uta - Set unicast filter table address
5498 * @adapter: board private structure
5500 * The unicast table address is a register array of 32-bit registers.
5501 * The table is meant to be used in a way similar to how the MTA is used
5502 * however due to certain limitations in the hardware it is necessary to
5503 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
5504 * enable bit to allow vlan tag stripping when promiscuous mode is enabled
5506 static void igb_set_uta(struct igb_adapter *adapter)
5508 struct e1000_hw *hw = &adapter->hw;
5511 /* The UTA table only exists on 82576 hardware and newer */
5512 if (hw->mac.type < e1000_82576)
5515 /* we only need to do this if VMDq is enabled */
5516 if (!adapter->vfs_allocated_count)
5519 for (i = 0; i < hw->mac.uta_reg_count; i++)
5520 array_wr32(E1000_UTA, i, ~0);
5524 * igb_intr_msi - Interrupt Handler
5525 * @irq: interrupt number
5526 * @data: pointer to a network interface device structure
5528 static irqreturn_t igb_intr_msi(int irq, void *data)
5530 struct igb_adapter *adapter = data;
5531 struct igb_q_vector *q_vector = adapter->q_vector[0];
5532 struct e1000_hw *hw = &adapter->hw;
5533 /* read ICR disables interrupts using IAM */
5534 u32 icr = rd32(E1000_ICR);
5536 igb_write_itr(q_vector);
5538 if (icr & E1000_ICR_DRSTA)
5539 schedule_work(&adapter->reset_task);
5541 if (icr & E1000_ICR_DOUTSYNC) {
5542 /* HW is reporting DMA is out of sync */
5543 adapter->stats.doosync++;
5546 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
5547 hw->mac.get_link_status = 1;
5548 if (!test_bit(__IGB_DOWN, &adapter->state))
5549 mod_timer(&adapter->watchdog_timer, jiffies + 1);
5552 if (icr & E1000_ICR_TS) {
5553 u32 tsicr = rd32(E1000_TSICR);
5555 if (tsicr & E1000_TSICR_TXTS) {
5556 /* acknowledge the interrupt */
5557 wr32(E1000_TSICR, E1000_TSICR_TXTS);
5558 /* retrieve hardware timestamp */
5559 schedule_work(&adapter->ptp_tx_work);
5563 napi_schedule(&q_vector->napi);
5569 * igb_intr - Legacy Interrupt Handler
5570 * @irq: interrupt number
5571 * @data: pointer to a network interface device structure
5573 static irqreturn_t igb_intr(int irq, void *data)
5575 struct igb_adapter *adapter = data;
5576 struct igb_q_vector *q_vector = adapter->q_vector[0];
5577 struct e1000_hw *hw = &adapter->hw;
5578 /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked. No
5579 * need for the IMC write */
5580 u32 icr = rd32(E1000_ICR);
5582 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
5583 * not set, then the adapter didn't send an interrupt */
5584 if (!(icr & E1000_ICR_INT_ASSERTED))
5587 igb_write_itr(q_vector);
5589 if (icr & E1000_ICR_DRSTA)
5590 schedule_work(&adapter->reset_task);
5592 if (icr & E1000_ICR_DOUTSYNC) {
5593 /* HW is reporting DMA is out of sync */
5594 adapter->stats.doosync++;
5597 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
5598 hw->mac.get_link_status = 1;
5599 /* guard against interrupt when we're going down */
5600 if (!test_bit(__IGB_DOWN, &adapter->state))
5601 mod_timer(&adapter->watchdog_timer, jiffies + 1);
5604 if (icr & E1000_ICR_TS) {
5605 u32 tsicr = rd32(E1000_TSICR);
5607 if (tsicr & E1000_TSICR_TXTS) {
5608 /* acknowledge the interrupt */
5609 wr32(E1000_TSICR, E1000_TSICR_TXTS);
5610 /* retrieve hardware timestamp */
5611 schedule_work(&adapter->ptp_tx_work);
5615 napi_schedule(&q_vector->napi);
5620 static void igb_ring_irq_enable(struct igb_q_vector *q_vector)
5622 struct igb_adapter *adapter = q_vector->adapter;
5623 struct e1000_hw *hw = &adapter->hw;
5625 if ((q_vector->rx.ring && (adapter->rx_itr_setting & 3)) ||
5626 (!q_vector->rx.ring && (adapter->tx_itr_setting & 3))) {
5627 if ((adapter->num_q_vectors == 1) && !adapter->vf_data)
5628 igb_set_itr(q_vector);
5630 igb_update_ring_itr(q_vector);
5633 if (!test_bit(__IGB_DOWN, &adapter->state)) {
5634 if (adapter->msix_entries)
5635 wr32(E1000_EIMS, q_vector->eims_value);
5637 igb_irq_enable(adapter);
5642 * igb_poll - NAPI Rx polling callback
5643 * @napi: napi polling structure
5644 * @budget: count of how many packets we should handle
5646 static int igb_poll(struct napi_struct *napi, int budget)
5648 struct igb_q_vector *q_vector = container_of(napi,
5649 struct igb_q_vector,
5651 bool clean_complete = true;
5653 #ifdef CONFIG_IGB_DCA
5654 if (q_vector->adapter->flags & IGB_FLAG_DCA_ENABLED)
5655 igb_update_dca(q_vector);
5657 if (q_vector->tx.ring)
5658 clean_complete = igb_clean_tx_irq(q_vector);
5660 if (q_vector->rx.ring)
5661 clean_complete &= igb_clean_rx_irq(q_vector, budget);
5663 /* If all work not completed, return budget and keep polling */
5664 if (!clean_complete)
5667 /* If not enough Rx work done, exit the polling mode */
5668 napi_complete(napi);
5669 igb_ring_irq_enable(q_vector);
5675 * igb_clean_tx_irq - Reclaim resources after transmit completes
5676 * @q_vector: pointer to q_vector containing needed info
5678 * returns true if ring is completely cleaned
5680 static bool igb_clean_tx_irq(struct igb_q_vector *q_vector)
5682 struct igb_adapter *adapter = q_vector->adapter;
5683 struct igb_ring *tx_ring = q_vector->tx.ring;
5684 struct igb_tx_buffer *tx_buffer;
5685 union e1000_adv_tx_desc *tx_desc;
5686 unsigned int total_bytes = 0, total_packets = 0;
5687 unsigned int budget = q_vector->tx.work_limit;
5688 unsigned int i = tx_ring->next_to_clean;
5690 if (test_bit(__IGB_DOWN, &adapter->state))
5693 tx_buffer = &tx_ring->tx_buffer_info[i];
5694 tx_desc = IGB_TX_DESC(tx_ring, i);
5695 i -= tx_ring->count;
5698 union e1000_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
5700 /* if next_to_watch is not set then there is no work pending */
5704 /* prevent any other reads prior to eop_desc */
5707 /* if DD is not set pending work has not been completed */
5708 if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)))
5711 /* clear next_to_watch to prevent false hangs */
5712 tx_buffer->next_to_watch = NULL;
5714 /* update the statistics for this packet */
5715 total_bytes += tx_buffer->bytecount;
5716 total_packets += tx_buffer->gso_segs;
5719 dev_kfree_skb_any(tx_buffer->skb);
5721 /* unmap skb header data */
5722 dma_unmap_single(tx_ring->dev,
5723 dma_unmap_addr(tx_buffer, dma),
5724 dma_unmap_len(tx_buffer, len),
5727 /* clear tx_buffer data */
5728 tx_buffer->skb = NULL;
5729 dma_unmap_len_set(tx_buffer, len, 0);
5731 /* clear last DMA location and unmap remaining buffers */
5732 while (tx_desc != eop_desc) {
5737 i -= tx_ring->count;
5738 tx_buffer = tx_ring->tx_buffer_info;
5739 tx_desc = IGB_TX_DESC(tx_ring, 0);
5742 /* unmap any remaining paged data */
5743 if (dma_unmap_len(tx_buffer, len)) {
5744 dma_unmap_page(tx_ring->dev,
5745 dma_unmap_addr(tx_buffer, dma),
5746 dma_unmap_len(tx_buffer, len),
5748 dma_unmap_len_set(tx_buffer, len, 0);
5752 /* move us one more past the eop_desc for start of next pkt */
5757 i -= tx_ring->count;
5758 tx_buffer = tx_ring->tx_buffer_info;
5759 tx_desc = IGB_TX_DESC(tx_ring, 0);
5762 /* issue prefetch for next Tx descriptor */
5765 /* update budget accounting */
5767 } while (likely(budget));
5769 netdev_tx_completed_queue(txring_txq(tx_ring),
5770 total_packets, total_bytes);
5771 i += tx_ring->count;
5772 tx_ring->next_to_clean = i;
5773 u64_stats_update_begin(&tx_ring->tx_syncp);
5774 tx_ring->tx_stats.bytes += total_bytes;
5775 tx_ring->tx_stats.packets += total_packets;
5776 u64_stats_update_end(&tx_ring->tx_syncp);
5777 q_vector->tx.total_bytes += total_bytes;
5778 q_vector->tx.total_packets += total_packets;
5780 if (test_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags)) {
5781 struct e1000_hw *hw = &adapter->hw;
5783 /* Detect a transmit hang in hardware, this serializes the
5784 * check with the clearing of time_stamp and movement of i */
5785 clear_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
5786 if (tx_buffer->next_to_watch &&
5787 time_after(jiffies, tx_buffer->time_stamp +
5788 (adapter->tx_timeout_factor * HZ)) &&
5789 !(rd32(E1000_STATUS) & E1000_STATUS_TXOFF)) {
5791 /* detected Tx unit hang */
5792 dev_err(tx_ring->dev,
5793 "Detected Tx Unit Hang\n"
5797 " next_to_use <%x>\n"
5798 " next_to_clean <%x>\n"
5799 "buffer_info[next_to_clean]\n"
5800 " time_stamp <%lx>\n"
5801 " next_to_watch <%p>\n"
5803 " desc.status <%x>\n",
5804 tx_ring->queue_index,
5805 rd32(E1000_TDH(tx_ring->reg_idx)),
5806 readl(tx_ring->tail),
5807 tx_ring->next_to_use,
5808 tx_ring->next_to_clean,
5809 tx_buffer->time_stamp,
5810 tx_buffer->next_to_watch,
5812 tx_buffer->next_to_watch->wb.status);
5813 netif_stop_subqueue(tx_ring->netdev,
5814 tx_ring->queue_index);
5816 /* we are about to reset, no point in enabling stuff */
5821 if (unlikely(total_packets &&
5822 netif_carrier_ok(tx_ring->netdev) &&
5823 igb_desc_unused(tx_ring) >= IGB_TX_QUEUE_WAKE)) {
5824 /* Make sure that anybody stopping the queue after this
5825 * sees the new next_to_clean.
5828 if (__netif_subqueue_stopped(tx_ring->netdev,
5829 tx_ring->queue_index) &&
5830 !(test_bit(__IGB_DOWN, &adapter->state))) {
5831 netif_wake_subqueue(tx_ring->netdev,
5832 tx_ring->queue_index);
5834 u64_stats_update_begin(&tx_ring->tx_syncp);
5835 tx_ring->tx_stats.restart_queue++;
5836 u64_stats_update_end(&tx_ring->tx_syncp);
5844 * igb_reuse_rx_page - page flip buffer and store it back on the ring
5845 * @rx_ring: rx descriptor ring to store buffers on
5846 * @old_buff: donor buffer to have page reused
5848 * Synchronizes page for reuse by the adapter
5850 static void igb_reuse_rx_page(struct igb_ring *rx_ring,
5851 struct igb_rx_buffer *old_buff)
5853 struct igb_rx_buffer *new_buff;
5854 u16 nta = rx_ring->next_to_alloc;
5856 new_buff = &rx_ring->rx_buffer_info[nta];
5858 /* update, and store next to alloc */
5860 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
5862 /* transfer page from old buffer to new buffer */
5863 memcpy(new_buff, old_buff, sizeof(struct igb_rx_buffer));
5865 /* sync the buffer for use by the device */
5866 dma_sync_single_range_for_device(rx_ring->dev, old_buff->dma,
5867 old_buff->page_offset,
5873 * igb_add_rx_frag - Add contents of Rx buffer to sk_buff
5874 * @rx_ring: rx descriptor ring to transact packets on
5875 * @rx_buffer: buffer containing page to add
5876 * @rx_desc: descriptor containing length of buffer written by hardware
5877 * @skb: sk_buff to place the data into
5879 * This function will add the data contained in rx_buffer->page to the skb.
5880 * This is done either through a direct copy if the data in the buffer is
5881 * less than the skb header size, otherwise it will just attach the page as
5882 * a frag to the skb.
5884 * The function will then update the page offset if necessary and return
5885 * true if the buffer can be reused by the adapter.
5887 static bool igb_add_rx_frag(struct igb_ring *rx_ring,
5888 struct igb_rx_buffer *rx_buffer,
5889 union e1000_adv_rx_desc *rx_desc,
5890 struct sk_buff *skb)
5892 struct page *page = rx_buffer->page;
5893 unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
5895 if ((size <= IGB_RX_HDR_LEN) && !skb_is_nonlinear(skb)) {
5896 unsigned char *va = page_address(page) + rx_buffer->page_offset;
5898 if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
5899 igb_ptp_rx_pktstamp(rx_ring->q_vector, va, skb);
5900 va += IGB_TS_HDR_LEN;
5901 size -= IGB_TS_HDR_LEN;
5904 memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
5906 /* we can reuse buffer as-is, just make sure it is local */
5907 if (likely(page_to_nid(page) == numa_node_id()))
5910 /* this page cannot be reused so discard it */
5915 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
5916 rx_buffer->page_offset, size, IGB_RX_BUFSZ);
5918 /* avoid re-using remote pages */
5919 if (unlikely(page_to_nid(page) != numa_node_id()))
5922 #if (PAGE_SIZE < 8192)
5923 /* if we are only owner of page we can reuse it */
5924 if (unlikely(page_count(page) != 1))
5927 /* flip page offset to other buffer */
5928 rx_buffer->page_offset ^= IGB_RX_BUFSZ;
5931 * since we are the only owner of the page and we need to
5932 * increment it, just set the value to 2 in order to avoid
5933 * an unnecessary locked operation
5935 atomic_set(&page->_count, 2);
5937 /* move offset up to the next cache line */
5938 rx_buffer->page_offset += SKB_DATA_ALIGN(size);
5940 if (rx_buffer->page_offset > (PAGE_SIZE - IGB_RX_BUFSZ))
5943 /* bump ref count on page before it is given to the stack */
5950 static struct sk_buff *igb_fetch_rx_buffer(struct igb_ring *rx_ring,
5951 union e1000_adv_rx_desc *rx_desc,
5952 struct sk_buff *skb)
5954 struct igb_rx_buffer *rx_buffer;
5957 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
5960 * This memory barrier is needed to keep us from reading
5961 * any other fields out of the rx_desc until we know the
5962 * RXD_STAT_DD bit is set
5966 page = rx_buffer->page;
5970 void *page_addr = page_address(page) +
5971 rx_buffer->page_offset;
5973 /* prefetch first cache line of first page */
5974 prefetch(page_addr);
5975 #if L1_CACHE_BYTES < 128
5976 prefetch(page_addr + L1_CACHE_BYTES);
5979 /* allocate a skb to store the frags */
5980 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
5982 if (unlikely(!skb)) {
5983 rx_ring->rx_stats.alloc_failed++;
5988 * we will be copying header into skb->data in
5989 * pskb_may_pull so it is in our interest to prefetch
5990 * it now to avoid a possible cache miss
5992 prefetchw(skb->data);
5995 /* we are reusing so sync this buffer for CPU use */
5996 dma_sync_single_range_for_cpu(rx_ring->dev,
5998 rx_buffer->page_offset,
6002 /* pull page into skb */
6003 if (igb_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
6004 /* hand second half of page back to the ring */
6005 igb_reuse_rx_page(rx_ring, rx_buffer);
6007 /* we are not reusing the buffer so unmap it */
6008 dma_unmap_page(rx_ring->dev, rx_buffer->dma,
6009 PAGE_SIZE, DMA_FROM_DEVICE);
6012 /* clear contents of rx_buffer */
6013 rx_buffer->page = NULL;
6018 static inline void igb_rx_checksum(struct igb_ring *ring,
6019 union e1000_adv_rx_desc *rx_desc,
6020 struct sk_buff *skb)
6022 skb_checksum_none_assert(skb);
6024 /* Ignore Checksum bit is set */
6025 if (igb_test_staterr(rx_desc, E1000_RXD_STAT_IXSM))
6028 /* Rx checksum disabled via ethtool */
6029 if (!(ring->netdev->features & NETIF_F_RXCSUM))
6032 /* TCP/UDP checksum error bit is set */
6033 if (igb_test_staterr(rx_desc,
6034 E1000_RXDEXT_STATERR_TCPE |
6035 E1000_RXDEXT_STATERR_IPE)) {
6037 * work around errata with sctp packets where the TCPE aka
6038 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
6039 * packets, (aka let the stack check the crc32c)
6041 if (!((skb->len == 60) &&
6042 test_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags))) {
6043 u64_stats_update_begin(&ring->rx_syncp);
6044 ring->rx_stats.csum_err++;
6045 u64_stats_update_end(&ring->rx_syncp);
6047 /* let the stack verify checksum errors */
6050 /* It must be a TCP or UDP packet with a valid checksum */
6051 if (igb_test_staterr(rx_desc, E1000_RXD_STAT_TCPCS |
6052 E1000_RXD_STAT_UDPCS))
6053 skb->ip_summed = CHECKSUM_UNNECESSARY;
6055 dev_dbg(ring->dev, "cksum success: bits %08X\n",
6056 le32_to_cpu(rx_desc->wb.upper.status_error));
6059 static inline void igb_rx_hash(struct igb_ring *ring,
6060 union e1000_adv_rx_desc *rx_desc,
6061 struct sk_buff *skb)
6063 if (ring->netdev->features & NETIF_F_RXHASH)
6064 skb->rxhash = le32_to_cpu(rx_desc->wb.lower.hi_dword.rss);
6068 * igb_is_non_eop - process handling of non-EOP buffers
6069 * @rx_ring: Rx ring being processed
6070 * @rx_desc: Rx descriptor for current buffer
6071 * @skb: current socket buffer containing buffer in progress
6073 * This function updates next to clean. If the buffer is an EOP buffer
6074 * this function exits returning false, otherwise it will place the
6075 * sk_buff in the next buffer to be chained and return true indicating
6076 * that this is in fact a non-EOP buffer.
6078 static bool igb_is_non_eop(struct igb_ring *rx_ring,
6079 union e1000_adv_rx_desc *rx_desc)
6081 u32 ntc = rx_ring->next_to_clean + 1;
6083 /* fetch, update, and store next to clean */
6084 ntc = (ntc < rx_ring->count) ? ntc : 0;
6085 rx_ring->next_to_clean = ntc;
6087 prefetch(IGB_RX_DESC(rx_ring, ntc));
6089 if (likely(igb_test_staterr(rx_desc, E1000_RXD_STAT_EOP)))
6096 * igb_get_headlen - determine size of header for LRO/GRO
6097 * @data: pointer to the start of the headers
6098 * @max_len: total length of section to find headers in
6100 * This function is meant to determine the length of headers that will
6101 * be recognized by hardware for LRO, and GRO offloads. The main
6102 * motivation of doing this is to only perform one pull for IPv4 TCP
6103 * packets so that we can do basic things like calculating the gso_size
6104 * based on the average data per packet.
6106 static unsigned int igb_get_headlen(unsigned char *data,
6107 unsigned int max_len)
6110 unsigned char *network;
6113 struct vlan_hdr *vlan;
6116 struct ipv6hdr *ipv6;
6119 u8 nexthdr = 0; /* default to not TCP */
6122 /* this should never happen, but better safe than sorry */
6123 if (max_len < ETH_HLEN)
6126 /* initialize network frame pointer */
6129 /* set first protocol and move network header forward */
6130 protocol = hdr.eth->h_proto;
6131 hdr.network += ETH_HLEN;
6133 /* handle any vlan tag if present */
6134 if (protocol == __constant_htons(ETH_P_8021Q)) {
6135 if ((hdr.network - data) > (max_len - VLAN_HLEN))
6138 protocol = hdr.vlan->h_vlan_encapsulated_proto;
6139 hdr.network += VLAN_HLEN;
6142 /* handle L3 protocols */
6143 if (protocol == __constant_htons(ETH_P_IP)) {
6144 if ((hdr.network - data) > (max_len - sizeof(struct iphdr)))
6147 /* access ihl as a u8 to avoid unaligned access on ia64 */
6148 hlen = (hdr.network[0] & 0x0F) << 2;
6150 /* verify hlen meets minimum size requirements */
6151 if (hlen < sizeof(struct iphdr))
6152 return hdr.network - data;
6154 /* record next protocol if header is present */
6155 if (!hdr.ipv4->frag_off)
6156 nexthdr = hdr.ipv4->protocol;
6157 } else if (protocol == __constant_htons(ETH_P_IPV6)) {
6158 if ((hdr.network - data) > (max_len - sizeof(struct ipv6hdr)))
6161 /* record next protocol */
6162 nexthdr = hdr.ipv6->nexthdr;
6163 hlen = sizeof(struct ipv6hdr);
6165 return hdr.network - data;
6168 /* relocate pointer to start of L4 header */
6169 hdr.network += hlen;
6171 /* finally sort out TCP */
6172 if (nexthdr == IPPROTO_TCP) {
6173 if ((hdr.network - data) > (max_len - sizeof(struct tcphdr)))
6176 /* access doff as a u8 to avoid unaligned access on ia64 */
6177 hlen = (hdr.network[12] & 0xF0) >> 2;
6179 /* verify hlen meets minimum size requirements */
6180 if (hlen < sizeof(struct tcphdr))
6181 return hdr.network - data;
6183 hdr.network += hlen;
6184 } else if (nexthdr == IPPROTO_UDP) {
6185 if ((hdr.network - data) > (max_len - sizeof(struct udphdr)))
6188 hdr.network += sizeof(struct udphdr);
6192 * If everything has gone correctly hdr.network should be the
6193 * data section of the packet and will be the end of the header.
6194 * If not then it probably represents the end of the last recognized
6197 if ((hdr.network - data) < max_len)
6198 return hdr.network - data;
6204 * igb_pull_tail - igb specific version of skb_pull_tail
6205 * @rx_ring: rx descriptor ring packet is being transacted on
6206 * @rx_desc: pointer to the EOP Rx descriptor
6207 * @skb: pointer to current skb being adjusted
6209 * This function is an igb specific version of __pskb_pull_tail. The
6210 * main difference between this version and the original function is that
6211 * this function can make several assumptions about the state of things
6212 * that allow for significant optimizations versus the standard function.
6213 * As a result we can do things like drop a frag and maintain an accurate
6214 * truesize for the skb.
6216 static void igb_pull_tail(struct igb_ring *rx_ring,
6217 union e1000_adv_rx_desc *rx_desc,
6218 struct sk_buff *skb)
6220 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
6222 unsigned int pull_len;
6225 * it is valid to use page_address instead of kmap since we are
6226 * working with pages allocated out of the lomem pool per
6227 * alloc_page(GFP_ATOMIC)
6229 va = skb_frag_address(frag);
6231 if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
6232 /* retrieve timestamp from buffer */
6233 igb_ptp_rx_pktstamp(rx_ring->q_vector, va, skb);
6235 /* update pointers to remove timestamp header */
6236 skb_frag_size_sub(frag, IGB_TS_HDR_LEN);
6237 frag->page_offset += IGB_TS_HDR_LEN;
6238 skb->data_len -= IGB_TS_HDR_LEN;
6239 skb->len -= IGB_TS_HDR_LEN;
6241 /* move va to start of packet data */
6242 va += IGB_TS_HDR_LEN;
6246 * we need the header to contain the greater of either ETH_HLEN or
6247 * 60 bytes if the skb->len is less than 60 for skb_pad.
6249 pull_len = igb_get_headlen(va, IGB_RX_HDR_LEN);
6251 /* align pull length to size of long to optimize memcpy performance */
6252 skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
6254 /* update all of the pointers */
6255 skb_frag_size_sub(frag, pull_len);
6256 frag->page_offset += pull_len;
6257 skb->data_len -= pull_len;
6258 skb->tail += pull_len;
6262 * igb_cleanup_headers - Correct corrupted or empty headers
6263 * @rx_ring: rx descriptor ring packet is being transacted on
6264 * @rx_desc: pointer to the EOP Rx descriptor
6265 * @skb: pointer to current skb being fixed
6267 * Address the case where we are pulling data in on pages only
6268 * and as such no data is present in the skb header.
6270 * In addition if skb is not at least 60 bytes we need to pad it so that
6271 * it is large enough to qualify as a valid Ethernet frame.
6273 * Returns true if an error was encountered and skb was freed.
6275 static bool igb_cleanup_headers(struct igb_ring *rx_ring,
6276 union e1000_adv_rx_desc *rx_desc,
6277 struct sk_buff *skb)
6280 if (unlikely((igb_test_staterr(rx_desc,
6281 E1000_RXDEXT_ERR_FRAME_ERR_MASK)))) {
6282 struct net_device *netdev = rx_ring->netdev;
6283 if (!(netdev->features & NETIF_F_RXALL)) {
6284 dev_kfree_skb_any(skb);
6289 /* place header in linear portion of buffer */
6290 if (skb_is_nonlinear(skb))
6291 igb_pull_tail(rx_ring, rx_desc, skb);
6293 /* if skb_pad returns an error the skb was freed */
6294 if (unlikely(skb->len < 60)) {
6295 int pad_len = 60 - skb->len;
6297 if (skb_pad(skb, pad_len))
6299 __skb_put(skb, pad_len);
6306 * igb_process_skb_fields - Populate skb header fields from Rx descriptor
6307 * @rx_ring: rx descriptor ring packet is being transacted on
6308 * @rx_desc: pointer to the EOP Rx descriptor
6309 * @skb: pointer to current skb being populated
6311 * This function checks the ring, descriptor, and packet information in
6312 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
6313 * other fields within the skb.
6315 static void igb_process_skb_fields(struct igb_ring *rx_ring,
6316 union e1000_adv_rx_desc *rx_desc,
6317 struct sk_buff *skb)
6319 struct net_device *dev = rx_ring->netdev;
6321 igb_rx_hash(rx_ring, rx_desc, skb);
6323 igb_rx_checksum(rx_ring, rx_desc, skb);
6325 igb_ptp_rx_hwtstamp(rx_ring->q_vector, rx_desc, skb);
6327 if ((dev->features & NETIF_F_HW_VLAN_RX) &&
6328 igb_test_staterr(rx_desc, E1000_RXD_STAT_VP)) {
6330 if (igb_test_staterr(rx_desc, E1000_RXDEXT_STATERR_LB) &&
6331 test_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &rx_ring->flags))
6332 vid = be16_to_cpu(rx_desc->wb.upper.vlan);
6334 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
6336 __vlan_hwaccel_put_tag(skb, vid);
6339 skb_record_rx_queue(skb, rx_ring->queue_index);
6341 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
6344 static bool igb_clean_rx_irq(struct igb_q_vector *q_vector, const int budget)
6346 struct igb_ring *rx_ring = q_vector->rx.ring;
6347 struct sk_buff *skb = rx_ring->skb;
6348 unsigned int total_bytes = 0, total_packets = 0;
6349 u16 cleaned_count = igb_desc_unused(rx_ring);
6352 union e1000_adv_rx_desc *rx_desc;
6354 /* return some buffers to hardware, one at a time is too slow */
6355 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
6356 igb_alloc_rx_buffers(rx_ring, cleaned_count);
6360 rx_desc = IGB_RX_DESC(rx_ring, rx_ring->next_to_clean);
6362 if (!igb_test_staterr(rx_desc, E1000_RXD_STAT_DD))
6365 /* retrieve a buffer from the ring */
6366 skb = igb_fetch_rx_buffer(rx_ring, rx_desc, skb);
6368 /* exit if we failed to retrieve a buffer */
6374 /* fetch next buffer in frame if non-eop */
6375 if (igb_is_non_eop(rx_ring, rx_desc))
6378 /* verify the packet layout is correct */
6379 if (igb_cleanup_headers(rx_ring, rx_desc, skb)) {
6384 /* probably a little skewed due to removing CRC */
6385 total_bytes += skb->len;
6387 /* populate checksum, timestamp, VLAN, and protocol */
6388 igb_process_skb_fields(rx_ring, rx_desc, skb);
6390 napi_gro_receive(&q_vector->napi, skb);
6392 /* reset skb pointer */
6395 /* update budget accounting */
6397 } while (likely(total_packets < budget));
6399 /* place incomplete frames back on ring for completion */
6402 u64_stats_update_begin(&rx_ring->rx_syncp);
6403 rx_ring->rx_stats.packets += total_packets;
6404 rx_ring->rx_stats.bytes += total_bytes;
6405 u64_stats_update_end(&rx_ring->rx_syncp);
6406 q_vector->rx.total_packets += total_packets;
6407 q_vector->rx.total_bytes += total_bytes;
6410 igb_alloc_rx_buffers(rx_ring, cleaned_count);
6412 return (total_packets < budget);
6415 static bool igb_alloc_mapped_page(struct igb_ring *rx_ring,
6416 struct igb_rx_buffer *bi)
6418 struct page *page = bi->page;
6421 /* since we are recycling buffers we should seldom need to alloc */
6425 /* alloc new page for storage */
6426 page = __skb_alloc_page(GFP_ATOMIC | __GFP_COLD, NULL);
6427 if (unlikely(!page)) {
6428 rx_ring->rx_stats.alloc_failed++;
6432 /* map page for use */
6433 dma = dma_map_page(rx_ring->dev, page, 0, PAGE_SIZE, DMA_FROM_DEVICE);
6436 * if mapping failed free memory back to system since
6437 * there isn't much point in holding memory we can't use
6439 if (dma_mapping_error(rx_ring->dev, dma)) {
6442 rx_ring->rx_stats.alloc_failed++;
6448 bi->page_offset = 0;
6454 * igb_alloc_rx_buffers - Replace used receive buffers; packet split
6455 * @adapter: address of board private structure
6457 void igb_alloc_rx_buffers(struct igb_ring *rx_ring, u16 cleaned_count)
6459 union e1000_adv_rx_desc *rx_desc;
6460 struct igb_rx_buffer *bi;
6461 u16 i = rx_ring->next_to_use;
6467 rx_desc = IGB_RX_DESC(rx_ring, i);
6468 bi = &rx_ring->rx_buffer_info[i];
6469 i -= rx_ring->count;
6472 if (!igb_alloc_mapped_page(rx_ring, bi))
6476 * Refresh the desc even if buffer_addrs didn't change
6477 * because each write-back erases this info.
6479 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
6485 rx_desc = IGB_RX_DESC(rx_ring, 0);
6486 bi = rx_ring->rx_buffer_info;
6487 i -= rx_ring->count;
6490 /* clear the hdr_addr for the next_to_use descriptor */
6491 rx_desc->read.hdr_addr = 0;
6494 } while (cleaned_count);
6496 i += rx_ring->count;
6498 if (rx_ring->next_to_use != i) {
6499 /* record the next descriptor to use */
6500 rx_ring->next_to_use = i;
6502 /* update next to alloc since we have filled the ring */
6503 rx_ring->next_to_alloc = i;
6506 * Force memory writes to complete before letting h/w
6507 * know there are new descriptors to fetch. (Only
6508 * applicable for weak-ordered memory model archs,
6512 writel(i, rx_ring->tail);
6522 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
6524 struct igb_adapter *adapter = netdev_priv(netdev);
6525 struct mii_ioctl_data *data = if_mii(ifr);
6527 if (adapter->hw.phy.media_type != e1000_media_type_copper)
6532 data->phy_id = adapter->hw.phy.addr;
6535 if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
6552 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
6558 return igb_mii_ioctl(netdev, ifr, cmd);
6560 return igb_ptp_hwtstamp_ioctl(netdev, ifr, cmd);
6566 s32 igb_read_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
6568 struct igb_adapter *adapter = hw->back;
6570 if (pcie_capability_read_word(adapter->pdev, reg, value))
6571 return -E1000_ERR_CONFIG;
6576 s32 igb_write_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
6578 struct igb_adapter *adapter = hw->back;
6580 if (pcie_capability_write_word(adapter->pdev, reg, *value))
6581 return -E1000_ERR_CONFIG;
6586 static void igb_vlan_mode(struct net_device *netdev, netdev_features_t features)
6588 struct igb_adapter *adapter = netdev_priv(netdev);
6589 struct e1000_hw *hw = &adapter->hw;
6591 bool enable = !!(features & NETIF_F_HW_VLAN_RX);
6594 /* enable VLAN tag insert/strip */
6595 ctrl = rd32(E1000_CTRL);
6596 ctrl |= E1000_CTRL_VME;
6597 wr32(E1000_CTRL, ctrl);
6599 /* Disable CFI check */
6600 rctl = rd32(E1000_RCTL);
6601 rctl &= ~E1000_RCTL_CFIEN;
6602 wr32(E1000_RCTL, rctl);
6604 /* disable VLAN tag insert/strip */
6605 ctrl = rd32(E1000_CTRL);
6606 ctrl &= ~E1000_CTRL_VME;
6607 wr32(E1000_CTRL, ctrl);
6610 igb_rlpml_set(adapter);
6613 static int igb_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
6615 struct igb_adapter *adapter = netdev_priv(netdev);
6616 struct e1000_hw *hw = &adapter->hw;
6617 int pf_id = adapter->vfs_allocated_count;
6619 /* attempt to add filter to vlvf array */
6620 igb_vlvf_set(adapter, vid, true, pf_id);
6622 /* add the filter since PF can receive vlans w/o entry in vlvf */
6623 igb_vfta_set(hw, vid, true);
6625 set_bit(vid, adapter->active_vlans);
6630 static int igb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
6632 struct igb_adapter *adapter = netdev_priv(netdev);
6633 struct e1000_hw *hw = &adapter->hw;
6634 int pf_id = adapter->vfs_allocated_count;
6637 /* remove vlan from VLVF table array */
6638 err = igb_vlvf_set(adapter, vid, false, pf_id);
6640 /* if vid was not present in VLVF just remove it from table */
6642 igb_vfta_set(hw, vid, false);
6644 clear_bit(vid, adapter->active_vlans);
6649 static void igb_restore_vlan(struct igb_adapter *adapter)
6653 igb_vlan_mode(adapter->netdev, adapter->netdev->features);
6655 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
6656 igb_vlan_rx_add_vid(adapter->netdev, vid);
6659 int igb_set_spd_dplx(struct igb_adapter *adapter, u32 spd, u8 dplx)
6661 struct pci_dev *pdev = adapter->pdev;
6662 struct e1000_mac_info *mac = &adapter->hw.mac;
6666 /* Make sure dplx is at most 1 bit and lsb of speed is not set
6667 * for the switch() below to work */
6668 if ((spd & 1) || (dplx & ~1))
6671 /* Fiber NIC's only allow 1000 Gbps Full duplex */
6672 if ((adapter->hw.phy.media_type == e1000_media_type_internal_serdes) &&
6673 spd != SPEED_1000 &&
6674 dplx != DUPLEX_FULL)
6677 switch (spd + dplx) {
6678 case SPEED_10 + DUPLEX_HALF:
6679 mac->forced_speed_duplex = ADVERTISE_10_HALF;
6681 case SPEED_10 + DUPLEX_FULL:
6682 mac->forced_speed_duplex = ADVERTISE_10_FULL;
6684 case SPEED_100 + DUPLEX_HALF:
6685 mac->forced_speed_duplex = ADVERTISE_100_HALF;
6687 case SPEED_100 + DUPLEX_FULL:
6688 mac->forced_speed_duplex = ADVERTISE_100_FULL;
6690 case SPEED_1000 + DUPLEX_FULL:
6692 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
6694 case SPEED_1000 + DUPLEX_HALF: /* not supported */
6699 /* clear MDI, MDI(-X) override is only allowed when autoneg enabled */
6700 adapter->hw.phy.mdix = AUTO_ALL_MODES;
6705 dev_err(&pdev->dev, "Unsupported Speed/Duplex configuration\n");
6709 static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake,
6712 struct net_device *netdev = pci_get_drvdata(pdev);
6713 struct igb_adapter *adapter = netdev_priv(netdev);
6714 struct e1000_hw *hw = &adapter->hw;
6715 u32 ctrl, rctl, status;
6716 u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
6721 netif_device_detach(netdev);
6723 if (netif_running(netdev))
6724 __igb_close(netdev, true);
6726 igb_clear_interrupt_scheme(adapter);
6729 retval = pci_save_state(pdev);
6734 status = rd32(E1000_STATUS);
6735 if (status & E1000_STATUS_LU)
6736 wufc &= ~E1000_WUFC_LNKC;
6739 igb_setup_rctl(adapter);
6740 igb_set_rx_mode(netdev);
6742 /* turn on all-multi mode if wake on multicast is enabled */
6743 if (wufc & E1000_WUFC_MC) {
6744 rctl = rd32(E1000_RCTL);
6745 rctl |= E1000_RCTL_MPE;
6746 wr32(E1000_RCTL, rctl);
6749 ctrl = rd32(E1000_CTRL);
6750 /* advertise wake from D3Cold */
6751 #define E1000_CTRL_ADVD3WUC 0x00100000
6752 /* phy power management enable */
6753 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
6754 ctrl |= E1000_CTRL_ADVD3WUC;
6755 wr32(E1000_CTRL, ctrl);
6757 /* Allow time for pending master requests to run */
6758 igb_disable_pcie_master(hw);
6760 wr32(E1000_WUC, E1000_WUC_PME_EN);
6761 wr32(E1000_WUFC, wufc);
6764 wr32(E1000_WUFC, 0);
6767 *enable_wake = wufc || adapter->en_mng_pt;
6769 igb_power_down_link(adapter);
6771 igb_power_up_link(adapter);
6773 /* Release control of h/w to f/w. If f/w is AMT enabled, this
6774 * would have already happened in close and is redundant. */
6775 igb_release_hw_control(adapter);
6777 pci_disable_device(pdev);
6783 #ifdef CONFIG_PM_SLEEP
6784 static int igb_suspend(struct device *dev)
6788 struct pci_dev *pdev = to_pci_dev(dev);
6790 retval = __igb_shutdown(pdev, &wake, 0);
6795 pci_prepare_to_sleep(pdev);
6797 pci_wake_from_d3(pdev, false);
6798 pci_set_power_state(pdev, PCI_D3hot);
6803 #endif /* CONFIG_PM_SLEEP */
6805 static int igb_resume(struct device *dev)
6807 struct pci_dev *pdev = to_pci_dev(dev);
6808 struct net_device *netdev = pci_get_drvdata(pdev);
6809 struct igb_adapter *adapter = netdev_priv(netdev);
6810 struct e1000_hw *hw = &adapter->hw;
6813 pci_set_power_state(pdev, PCI_D0);
6814 pci_restore_state(pdev);
6815 pci_save_state(pdev);
6817 err = pci_enable_device_mem(pdev);
6820 "igb: Cannot enable PCI device from suspend\n");
6823 pci_set_master(pdev);
6825 pci_enable_wake(pdev, PCI_D3hot, 0);
6826 pci_enable_wake(pdev, PCI_D3cold, 0);
6828 if (igb_init_interrupt_scheme(adapter, true)) {
6829 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
6835 /* let the f/w know that the h/w is now under the control of the
6837 igb_get_hw_control(adapter);
6839 wr32(E1000_WUS, ~0);
6841 if (netdev->flags & IFF_UP) {
6843 err = __igb_open(netdev, true);
6849 netif_device_attach(netdev);
6853 #ifdef CONFIG_PM_RUNTIME
6854 static int igb_runtime_idle(struct device *dev)
6856 struct pci_dev *pdev = to_pci_dev(dev);
6857 struct net_device *netdev = pci_get_drvdata(pdev);
6858 struct igb_adapter *adapter = netdev_priv(netdev);
6860 if (!igb_has_link(adapter))
6861 pm_schedule_suspend(dev, MSEC_PER_SEC * 5);
6866 static int igb_runtime_suspend(struct device *dev)
6868 struct pci_dev *pdev = to_pci_dev(dev);
6872 retval = __igb_shutdown(pdev, &wake, 1);
6877 pci_prepare_to_sleep(pdev);
6879 pci_wake_from_d3(pdev, false);
6880 pci_set_power_state(pdev, PCI_D3hot);
6886 static int igb_runtime_resume(struct device *dev)
6888 return igb_resume(dev);
6890 #endif /* CONFIG_PM_RUNTIME */
6893 static void igb_shutdown(struct pci_dev *pdev)
6897 __igb_shutdown(pdev, &wake, 0);
6899 if (system_state == SYSTEM_POWER_OFF) {
6900 pci_wake_from_d3(pdev, wake);
6901 pci_set_power_state(pdev, PCI_D3hot);
6905 #ifdef CONFIG_NET_POLL_CONTROLLER
6907 * Polling 'interrupt' - used by things like netconsole to send skbs
6908 * without having to re-enable interrupts. It's not called while
6909 * the interrupt routine is executing.
6911 static void igb_netpoll(struct net_device *netdev)
6913 struct igb_adapter *adapter = netdev_priv(netdev);
6914 struct e1000_hw *hw = &adapter->hw;
6915 struct igb_q_vector *q_vector;
6918 for (i = 0; i < adapter->num_q_vectors; i++) {
6919 q_vector = adapter->q_vector[i];
6920 if (adapter->msix_entries)
6921 wr32(E1000_EIMC, q_vector->eims_value);
6923 igb_irq_disable(adapter);
6924 napi_schedule(&q_vector->napi);
6927 #endif /* CONFIG_NET_POLL_CONTROLLER */
6930 * igb_io_error_detected - called when PCI error is detected
6931 * @pdev: Pointer to PCI device
6932 * @state: The current pci connection state
6934 * This function is called after a PCI bus error affecting
6935 * this device has been detected.
6937 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
6938 pci_channel_state_t state)
6940 struct net_device *netdev = pci_get_drvdata(pdev);
6941 struct igb_adapter *adapter = netdev_priv(netdev);
6943 netif_device_detach(netdev);
6945 if (state == pci_channel_io_perm_failure)
6946 return PCI_ERS_RESULT_DISCONNECT;
6948 if (netif_running(netdev))
6950 pci_disable_device(pdev);
6952 /* Request a slot slot reset. */
6953 return PCI_ERS_RESULT_NEED_RESET;
6957 * igb_io_slot_reset - called after the pci bus has been reset.
6958 * @pdev: Pointer to PCI device
6960 * Restart the card from scratch, as if from a cold-boot. Implementation
6961 * resembles the first-half of the igb_resume routine.
6963 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
6965 struct net_device *netdev = pci_get_drvdata(pdev);
6966 struct igb_adapter *adapter = netdev_priv(netdev);
6967 struct e1000_hw *hw = &adapter->hw;
6968 pci_ers_result_t result;
6971 if (pci_enable_device_mem(pdev)) {
6973 "Cannot re-enable PCI device after reset.\n");
6974 result = PCI_ERS_RESULT_DISCONNECT;
6976 pci_set_master(pdev);
6977 pci_restore_state(pdev);
6978 pci_save_state(pdev);
6980 pci_enable_wake(pdev, PCI_D3hot, 0);
6981 pci_enable_wake(pdev, PCI_D3cold, 0);
6984 wr32(E1000_WUS, ~0);
6985 result = PCI_ERS_RESULT_RECOVERED;
6988 err = pci_cleanup_aer_uncorrect_error_status(pdev);
6990 dev_err(&pdev->dev, "pci_cleanup_aer_uncorrect_error_status "
6991 "failed 0x%0x\n", err);
6992 /* non-fatal, continue */
6999 * igb_io_resume - called when traffic can start flowing again.
7000 * @pdev: Pointer to PCI device
7002 * This callback is called when the error recovery driver tells us that
7003 * its OK to resume normal operation. Implementation resembles the
7004 * second-half of the igb_resume routine.
7006 static void igb_io_resume(struct pci_dev *pdev)
7008 struct net_device *netdev = pci_get_drvdata(pdev);
7009 struct igb_adapter *adapter = netdev_priv(netdev);
7011 if (netif_running(netdev)) {
7012 if (igb_up(adapter)) {
7013 dev_err(&pdev->dev, "igb_up failed after reset\n");
7018 netif_device_attach(netdev);
7020 /* let the f/w know that the h/w is now under the control of the
7022 igb_get_hw_control(adapter);
7025 static void igb_rar_set_qsel(struct igb_adapter *adapter, u8 *addr, u32 index,
7028 u32 rar_low, rar_high;
7029 struct e1000_hw *hw = &adapter->hw;
7031 /* HW expects these in little endian so we reverse the byte order
7032 * from network order (big endian) to little endian
7034 rar_low = ((u32) addr[0] | ((u32) addr[1] << 8) |
7035 ((u32) addr[2] << 16) | ((u32) addr[3] << 24));
7036 rar_high = ((u32) addr[4] | ((u32) addr[5] << 8));
7038 /* Indicate to hardware the Address is Valid. */
7039 rar_high |= E1000_RAH_AV;
7041 if (hw->mac.type == e1000_82575)
7042 rar_high |= E1000_RAH_POOL_1 * qsel;
7044 rar_high |= E1000_RAH_POOL_1 << qsel;
7046 wr32(E1000_RAL(index), rar_low);
7048 wr32(E1000_RAH(index), rar_high);
7052 static int igb_set_vf_mac(struct igb_adapter *adapter,
7053 int vf, unsigned char *mac_addr)
7055 struct e1000_hw *hw = &adapter->hw;
7056 /* VF MAC addresses start at end of receive addresses and moves
7057 * torwards the first, as a result a collision should not be possible */
7058 int rar_entry = hw->mac.rar_entry_count - (vf + 1);
7060 memcpy(adapter->vf_data[vf].vf_mac_addresses, mac_addr, ETH_ALEN);
7062 igb_rar_set_qsel(adapter, mac_addr, rar_entry, vf);
7067 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
7069 struct igb_adapter *adapter = netdev_priv(netdev);
7070 if (!is_valid_ether_addr(mac) || (vf >= adapter->vfs_allocated_count))
7072 adapter->vf_data[vf].flags |= IGB_VF_FLAG_PF_SET_MAC;
7073 dev_info(&adapter->pdev->dev, "setting MAC %pM on VF %d\n", mac, vf);
7074 dev_info(&adapter->pdev->dev, "Reload the VF driver to make this"
7075 " change effective.");
7076 if (test_bit(__IGB_DOWN, &adapter->state)) {
7077 dev_warn(&adapter->pdev->dev, "The VF MAC address has been set,"
7078 " but the PF device is not up.\n");
7079 dev_warn(&adapter->pdev->dev, "Bring the PF device up before"
7080 " attempting to use the VF device.\n");
7082 return igb_set_vf_mac(adapter, vf, mac);
7085 static int igb_link_mbps(int internal_link_speed)
7087 switch (internal_link_speed) {
7097 static void igb_set_vf_rate_limit(struct e1000_hw *hw, int vf, int tx_rate,
7104 /* Calculate the rate factor values to set */
7105 rf_int = link_speed / tx_rate;
7106 rf_dec = (link_speed - (rf_int * tx_rate));
7107 rf_dec = (rf_dec * (1<<E1000_RTTBCNRC_RF_INT_SHIFT)) / tx_rate;
7109 bcnrc_val = E1000_RTTBCNRC_RS_ENA;
7110 bcnrc_val |= ((rf_int<<E1000_RTTBCNRC_RF_INT_SHIFT) &
7111 E1000_RTTBCNRC_RF_INT_MASK);
7112 bcnrc_val |= (rf_dec & E1000_RTTBCNRC_RF_DEC_MASK);
7117 wr32(E1000_RTTDQSEL, vf); /* vf X uses queue X */
7119 * Set global transmit compensation time to the MMW_SIZE in RTTBCNRM
7120 * register. MMW_SIZE=0x014 if 9728-byte jumbo is supported.
7122 wr32(E1000_RTTBCNRM, 0x14);
7123 wr32(E1000_RTTBCNRC, bcnrc_val);
7126 static void igb_check_vf_rate_limit(struct igb_adapter *adapter)
7128 int actual_link_speed, i;
7129 bool reset_rate = false;
7131 /* VF TX rate limit was not set or not supported */
7132 if ((adapter->vf_rate_link_speed == 0) ||
7133 (adapter->hw.mac.type != e1000_82576))
7136 actual_link_speed = igb_link_mbps(adapter->link_speed);
7137 if (actual_link_speed != adapter->vf_rate_link_speed) {
7139 adapter->vf_rate_link_speed = 0;
7140 dev_info(&adapter->pdev->dev,
7141 "Link speed has been changed. VF Transmit "
7142 "rate is disabled\n");
7145 for (i = 0; i < adapter->vfs_allocated_count; i++) {
7147 adapter->vf_data[i].tx_rate = 0;
7149 igb_set_vf_rate_limit(&adapter->hw, i,
7150 adapter->vf_data[i].tx_rate,
7155 static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf, int tx_rate)
7157 struct igb_adapter *adapter = netdev_priv(netdev);
7158 struct e1000_hw *hw = &adapter->hw;
7159 int actual_link_speed;
7161 if (hw->mac.type != e1000_82576)
7164 actual_link_speed = igb_link_mbps(adapter->link_speed);
7165 if ((vf >= adapter->vfs_allocated_count) ||
7166 (!(rd32(E1000_STATUS) & E1000_STATUS_LU)) ||
7167 (tx_rate < 0) || (tx_rate > actual_link_speed))
7170 adapter->vf_rate_link_speed = actual_link_speed;
7171 adapter->vf_data[vf].tx_rate = (u16)tx_rate;
7172 igb_set_vf_rate_limit(hw, vf, tx_rate, actual_link_speed);
7177 static int igb_ndo_get_vf_config(struct net_device *netdev,
7178 int vf, struct ifla_vf_info *ivi)
7180 struct igb_adapter *adapter = netdev_priv(netdev);
7181 if (vf >= adapter->vfs_allocated_count)
7184 memcpy(&ivi->mac, adapter->vf_data[vf].vf_mac_addresses, ETH_ALEN);
7185 ivi->tx_rate = adapter->vf_data[vf].tx_rate;
7186 ivi->vlan = adapter->vf_data[vf].pf_vlan;
7187 ivi->qos = adapter->vf_data[vf].pf_qos;
7191 static void igb_vmm_control(struct igb_adapter *adapter)
7193 struct e1000_hw *hw = &adapter->hw;
7196 switch (hw->mac.type) {
7201 /* replication is not supported for 82575 */
7204 /* notify HW that the MAC is adding vlan tags */
7205 reg = rd32(E1000_DTXCTL);
7206 reg |= E1000_DTXCTL_VLAN_ADDED;
7207 wr32(E1000_DTXCTL, reg);
7209 /* enable replication vlan tag stripping */
7210 reg = rd32(E1000_RPLOLR);
7211 reg |= E1000_RPLOLR_STRVLAN;
7212 wr32(E1000_RPLOLR, reg);
7214 /* none of the above registers are supported by i350 */
7218 if (adapter->vfs_allocated_count) {
7219 igb_vmdq_set_loopback_pf(hw, true);
7220 igb_vmdq_set_replication_pf(hw, true);
7221 igb_vmdq_set_anti_spoofing_pf(hw, true,
7222 adapter->vfs_allocated_count);
7224 igb_vmdq_set_loopback_pf(hw, false);
7225 igb_vmdq_set_replication_pf(hw, false);
7229 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba)
7231 struct e1000_hw *hw = &adapter->hw;
7235 if (hw->mac.type > e1000_82580) {
7236 if (adapter->flags & IGB_FLAG_DMAC) {
7239 /* force threshold to 0. */
7240 wr32(E1000_DMCTXTH, 0);
7243 * DMA Coalescing high water mark needs to be greater
7244 * than the Rx threshold. Set hwm to PBA - max frame
7245 * size in 16B units, capping it at PBA - 6KB.
7247 hwm = 64 * pba - adapter->max_frame_size / 16;
7248 if (hwm < 64 * (pba - 6))
7249 hwm = 64 * (pba - 6);
7250 reg = rd32(E1000_FCRTC);
7251 reg &= ~E1000_FCRTC_RTH_COAL_MASK;
7252 reg |= ((hwm << E1000_FCRTC_RTH_COAL_SHIFT)
7253 & E1000_FCRTC_RTH_COAL_MASK);
7254 wr32(E1000_FCRTC, reg);
7257 * Set the DMA Coalescing Rx threshold to PBA - 2 * max
7258 * frame size, capping it at PBA - 10KB.
7260 dmac_thr = pba - adapter->max_frame_size / 512;
7261 if (dmac_thr < pba - 10)
7262 dmac_thr = pba - 10;
7263 reg = rd32(E1000_DMACR);
7264 reg &= ~E1000_DMACR_DMACTHR_MASK;
7265 reg |= ((dmac_thr << E1000_DMACR_DMACTHR_SHIFT)
7266 & E1000_DMACR_DMACTHR_MASK);
7268 /* transition to L0x or L1 if available..*/
7269 reg |= (E1000_DMACR_DMAC_EN | E1000_DMACR_DMAC_LX_MASK);
7271 /* watchdog timer= +-1000 usec in 32usec intervals */
7274 /* Disable BMC-to-OS Watchdog Enable */
7275 reg &= ~E1000_DMACR_DC_BMC2OSW_EN;
7276 wr32(E1000_DMACR, reg);
7279 * no lower threshold to disable
7280 * coalescing(smart fifb)-UTRESH=0
7282 wr32(E1000_DMCRTRH, 0);
7284 reg = (IGB_DMCTLX_DCFLUSH_DIS | 0x4);
7286 wr32(E1000_DMCTLX, reg);
7289 * free space in tx packet buffer to wake from
7292 wr32(E1000_DMCTXTH, (IGB_MIN_TXPBSIZE -
7293 (IGB_TX_BUF_4096 + adapter->max_frame_size)) >> 6);
7296 * make low power state decision controlled
7299 reg = rd32(E1000_PCIEMISC);
7300 reg &= ~E1000_PCIEMISC_LX_DECISION;
7301 wr32(E1000_PCIEMISC, reg);
7302 } /* endif adapter->dmac is not disabled */
7303 } else if (hw->mac.type == e1000_82580) {
7304 u32 reg = rd32(E1000_PCIEMISC);
7305 wr32(E1000_PCIEMISC, reg & ~E1000_PCIEMISC_LX_DECISION);
7306 wr32(E1000_DMACR, 0);