1 /* Intel(R) Gigabit Ethernet Linux driver
2 * Copyright(c) 2007-2014 Intel Corporation.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * You should have received a copy of the GNU General Public License along with
14 * this program; if not, see <http://www.gnu.org/licenses/>.
16 * The full GNU General Public License is included in this distribution in
17 * the file called "COPYING".
19 * Contact Information:
20 * e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
21 * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
24 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
26 #include <linux/module.h>
27 #include <linux/types.h>
28 #include <linux/init.h>
29 #include <linux/bitops.h>
30 #include <linux/vmalloc.h>
31 #include <linux/pagemap.h>
32 #include <linux/netdevice.h>
33 #include <linux/ipv6.h>
34 #include <linux/slab.h>
35 #include <net/checksum.h>
36 #include <net/ip6_checksum.h>
37 #include <linux/net_tstamp.h>
38 #include <linux/mii.h>
39 #include <linux/ethtool.h>
41 #include <linux/if_vlan.h>
42 #include <linux/pci.h>
43 #include <linux/pci-aspm.h>
44 #include <linux/delay.h>
45 #include <linux/interrupt.h>
47 #include <linux/tcp.h>
48 #include <linux/sctp.h>
49 #include <linux/if_ether.h>
50 #include <linux/aer.h>
51 #include <linux/prefetch.h>
52 #include <linux/pm_runtime.h>
54 #include <linux/dca.h>
56 #include <linux/i2c.h>
62 #define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) "." \
63 __stringify(BUILD) "-k"
64 char igb_driver_name[] = "igb";
65 char igb_driver_version[] = DRV_VERSION;
66 static const char igb_driver_string[] =
67 "Intel(R) Gigabit Ethernet Network Driver";
68 static const char igb_copyright[] =
69 "Copyright (c) 2007-2014 Intel Corporation.";
71 static const struct e1000_info *igb_info_tbl[] = {
72 [board_82575] = &e1000_82575_info,
75 static const struct pci_device_id igb_pci_tbl[] = {
76 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_BACKPLANE_1GBPS) },
77 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_SGMII) },
78 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_BACKPLANE_2_5GBPS) },
79 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I211_COPPER), board_82575 },
80 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER), board_82575 },
81 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_FIBER), board_82575 },
82 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES), board_82575 },
83 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SGMII), board_82575 },
84 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER_FLASHLESS), board_82575 },
85 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES_FLASHLESS), board_82575 },
86 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_COPPER), board_82575 },
87 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_FIBER), board_82575 },
88 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SERDES), board_82575 },
89 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SGMII), board_82575 },
90 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER), board_82575 },
91 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_FIBER), board_82575 },
92 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_QUAD_FIBER), board_82575 },
93 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SERDES), board_82575 },
94 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SGMII), board_82575 },
95 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER_DUAL), board_82575 },
96 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SGMII), board_82575 },
97 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SERDES), board_82575 },
98 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_BACKPLANE), board_82575 },
99 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SFP), board_82575 },
100 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576), board_82575 },
101 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS), board_82575 },
102 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS_SERDES), board_82575 },
103 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_FIBER), board_82575 },
104 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES), board_82575 },
105 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES_QUAD), board_82575 },
106 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER_ET2), board_82575 },
107 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER), board_82575 },
108 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
109 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
110 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
111 /* required last entry */
115 MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
117 static int igb_setup_all_tx_resources(struct igb_adapter *);
118 static int igb_setup_all_rx_resources(struct igb_adapter *);
119 static void igb_free_all_tx_resources(struct igb_adapter *);
120 static void igb_free_all_rx_resources(struct igb_adapter *);
121 static void igb_setup_mrqc(struct igb_adapter *);
122 static int igb_probe(struct pci_dev *, const struct pci_device_id *);
123 static void igb_remove(struct pci_dev *pdev);
124 static int igb_sw_init(struct igb_adapter *);
125 static int igb_open(struct net_device *);
126 static int igb_close(struct net_device *);
127 static void igb_configure(struct igb_adapter *);
128 static void igb_configure_tx(struct igb_adapter *);
129 static void igb_configure_rx(struct igb_adapter *);
130 static void igb_clean_all_tx_rings(struct igb_adapter *);
131 static void igb_clean_all_rx_rings(struct igb_adapter *);
132 static void igb_clean_tx_ring(struct igb_ring *);
133 static void igb_clean_rx_ring(struct igb_ring *);
134 static void igb_set_rx_mode(struct net_device *);
135 static void igb_update_phy_info(unsigned long);
136 static void igb_watchdog(unsigned long);
137 static void igb_watchdog_task(struct work_struct *);
138 static netdev_tx_t igb_xmit_frame(struct sk_buff *skb, struct net_device *);
139 static struct rtnl_link_stats64 *igb_get_stats64(struct net_device *dev,
140 struct rtnl_link_stats64 *stats);
141 static int igb_change_mtu(struct net_device *, int);
142 static int igb_set_mac(struct net_device *, void *);
143 static void igb_set_uta(struct igb_adapter *adapter);
144 static irqreturn_t igb_intr(int irq, void *);
145 static irqreturn_t igb_intr_msi(int irq, void *);
146 static irqreturn_t igb_msix_other(int irq, void *);
147 static irqreturn_t igb_msix_ring(int irq, void *);
148 #ifdef CONFIG_IGB_DCA
149 static void igb_update_dca(struct igb_q_vector *);
150 static void igb_setup_dca(struct igb_adapter *);
151 #endif /* CONFIG_IGB_DCA */
152 static int igb_poll(struct napi_struct *, int);
153 static bool igb_clean_tx_irq(struct igb_q_vector *);
154 static bool igb_clean_rx_irq(struct igb_q_vector *, int);
155 static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
156 static void igb_tx_timeout(struct net_device *);
157 static void igb_reset_task(struct work_struct *);
158 static void igb_vlan_mode(struct net_device *netdev,
159 netdev_features_t features);
160 static int igb_vlan_rx_add_vid(struct net_device *, __be16, u16);
161 static int igb_vlan_rx_kill_vid(struct net_device *, __be16, u16);
162 static void igb_restore_vlan(struct igb_adapter *);
163 static void igb_rar_set_qsel(struct igb_adapter *, u8 *, u32 , u8);
164 static void igb_ping_all_vfs(struct igb_adapter *);
165 static void igb_msg_task(struct igb_adapter *);
166 static void igb_vmm_control(struct igb_adapter *);
167 static int igb_set_vf_mac(struct igb_adapter *, int, unsigned char *);
168 static void igb_restore_vf_multicasts(struct igb_adapter *adapter);
169 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac);
170 static int igb_ndo_set_vf_vlan(struct net_device *netdev,
171 int vf, u16 vlan, u8 qos);
172 static int igb_ndo_set_vf_bw(struct net_device *, int, int, int);
173 static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf,
175 static int igb_ndo_get_vf_config(struct net_device *netdev, int vf,
176 struct ifla_vf_info *ivi);
177 static void igb_check_vf_rate_limit(struct igb_adapter *);
179 #ifdef CONFIG_PCI_IOV
180 static int igb_vf_configure(struct igb_adapter *adapter, int vf);
181 static int igb_pci_enable_sriov(struct pci_dev *dev, int num_vfs);
185 #ifdef CONFIG_PM_SLEEP
186 static int igb_suspend(struct device *);
188 static int igb_resume(struct device *);
189 #ifdef CONFIG_PM_RUNTIME
190 static int igb_runtime_suspend(struct device *dev);
191 static int igb_runtime_resume(struct device *dev);
192 static int igb_runtime_idle(struct device *dev);
194 static const struct dev_pm_ops igb_pm_ops = {
195 SET_SYSTEM_SLEEP_PM_OPS(igb_suspend, igb_resume)
196 SET_RUNTIME_PM_OPS(igb_runtime_suspend, igb_runtime_resume,
200 static void igb_shutdown(struct pci_dev *);
201 static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs);
202 #ifdef CONFIG_IGB_DCA
203 static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
204 static struct notifier_block dca_notifier = {
205 .notifier_call = igb_notify_dca,
210 #ifdef CONFIG_NET_POLL_CONTROLLER
211 /* for netdump / net console */
212 static void igb_netpoll(struct net_device *);
214 #ifdef CONFIG_PCI_IOV
215 static unsigned int max_vfs;
216 module_param(max_vfs, uint, 0);
217 MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate per physical function");
218 #endif /* CONFIG_PCI_IOV */
220 static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
221 pci_channel_state_t);
222 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
223 static void igb_io_resume(struct pci_dev *);
225 static const struct pci_error_handlers igb_err_handler = {
226 .error_detected = igb_io_error_detected,
227 .slot_reset = igb_io_slot_reset,
228 .resume = igb_io_resume,
231 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba);
233 static struct pci_driver igb_driver = {
234 .name = igb_driver_name,
235 .id_table = igb_pci_tbl,
237 .remove = igb_remove,
239 .driver.pm = &igb_pm_ops,
241 .shutdown = igb_shutdown,
242 .sriov_configure = igb_pci_sriov_configure,
243 .err_handler = &igb_err_handler
246 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
247 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
248 MODULE_LICENSE("GPL");
249 MODULE_VERSION(DRV_VERSION);
251 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
252 static int debug = -1;
253 module_param(debug, int, 0);
254 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
256 struct igb_reg_info {
261 static const struct igb_reg_info igb_reg_info_tbl[] = {
263 /* General Registers */
264 {E1000_CTRL, "CTRL"},
265 {E1000_STATUS, "STATUS"},
266 {E1000_CTRL_EXT, "CTRL_EXT"},
268 /* Interrupt Registers */
272 {E1000_RCTL, "RCTL"},
273 {E1000_RDLEN(0), "RDLEN"},
274 {E1000_RDH(0), "RDH"},
275 {E1000_RDT(0), "RDT"},
276 {E1000_RXDCTL(0), "RXDCTL"},
277 {E1000_RDBAL(0), "RDBAL"},
278 {E1000_RDBAH(0), "RDBAH"},
281 {E1000_TCTL, "TCTL"},
282 {E1000_TDBAL(0), "TDBAL"},
283 {E1000_TDBAH(0), "TDBAH"},
284 {E1000_TDLEN(0), "TDLEN"},
285 {E1000_TDH(0), "TDH"},
286 {E1000_TDT(0), "TDT"},
287 {E1000_TXDCTL(0), "TXDCTL"},
288 {E1000_TDFH, "TDFH"},
289 {E1000_TDFT, "TDFT"},
290 {E1000_TDFHS, "TDFHS"},
291 {E1000_TDFPC, "TDFPC"},
293 /* List Terminator */
297 /* igb_regdump - register printout routine */
298 static void igb_regdump(struct e1000_hw *hw, struct igb_reg_info *reginfo)
304 switch (reginfo->ofs) {
306 for (n = 0; n < 4; n++)
307 regs[n] = rd32(E1000_RDLEN(n));
310 for (n = 0; n < 4; n++)
311 regs[n] = rd32(E1000_RDH(n));
314 for (n = 0; n < 4; n++)
315 regs[n] = rd32(E1000_RDT(n));
317 case E1000_RXDCTL(0):
318 for (n = 0; n < 4; n++)
319 regs[n] = rd32(E1000_RXDCTL(n));
322 for (n = 0; n < 4; n++)
323 regs[n] = rd32(E1000_RDBAL(n));
326 for (n = 0; n < 4; n++)
327 regs[n] = rd32(E1000_RDBAH(n));
330 for (n = 0; n < 4; n++)
331 regs[n] = rd32(E1000_RDBAL(n));
334 for (n = 0; n < 4; n++)
335 regs[n] = rd32(E1000_TDBAH(n));
338 for (n = 0; n < 4; n++)
339 regs[n] = rd32(E1000_TDLEN(n));
342 for (n = 0; n < 4; n++)
343 regs[n] = rd32(E1000_TDH(n));
346 for (n = 0; n < 4; n++)
347 regs[n] = rd32(E1000_TDT(n));
349 case E1000_TXDCTL(0):
350 for (n = 0; n < 4; n++)
351 regs[n] = rd32(E1000_TXDCTL(n));
354 pr_info("%-15s %08x\n", reginfo->name, rd32(reginfo->ofs));
358 snprintf(rname, 16, "%s%s", reginfo->name, "[0-3]");
359 pr_info("%-15s %08x %08x %08x %08x\n", rname, regs[0], regs[1],
363 /* igb_dump - Print registers, Tx-rings and Rx-rings */
364 static void igb_dump(struct igb_adapter *adapter)
366 struct net_device *netdev = adapter->netdev;
367 struct e1000_hw *hw = &adapter->hw;
368 struct igb_reg_info *reginfo;
369 struct igb_ring *tx_ring;
370 union e1000_adv_tx_desc *tx_desc;
371 struct my_u0 { u64 a; u64 b; } *u0;
372 struct igb_ring *rx_ring;
373 union e1000_adv_rx_desc *rx_desc;
377 if (!netif_msg_hw(adapter))
380 /* Print netdevice Info */
382 dev_info(&adapter->pdev->dev, "Net device Info\n");
383 pr_info("Device Name state trans_start last_rx\n");
384 pr_info("%-15s %016lX %016lX %016lX\n", netdev->name,
385 netdev->state, netdev->trans_start, netdev->last_rx);
388 /* Print Registers */
389 dev_info(&adapter->pdev->dev, "Register Dump\n");
390 pr_info(" Register Name Value\n");
391 for (reginfo = (struct igb_reg_info *)igb_reg_info_tbl;
392 reginfo->name; reginfo++) {
393 igb_regdump(hw, reginfo);
396 /* Print TX Ring Summary */
397 if (!netdev || !netif_running(netdev))
400 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
401 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
402 for (n = 0; n < adapter->num_tx_queues; n++) {
403 struct igb_tx_buffer *buffer_info;
404 tx_ring = adapter->tx_ring[n];
405 buffer_info = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
406 pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
407 n, tx_ring->next_to_use, tx_ring->next_to_clean,
408 (u64)dma_unmap_addr(buffer_info, dma),
409 dma_unmap_len(buffer_info, len),
410 buffer_info->next_to_watch,
411 (u64)buffer_info->time_stamp);
415 if (!netif_msg_tx_done(adapter))
416 goto rx_ring_summary;
418 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
420 /* Transmit Descriptor Formats
422 * Advanced Transmit Descriptor
423 * +--------------------------------------------------------------+
424 * 0 | Buffer Address [63:0] |
425 * +--------------------------------------------------------------+
426 * 8 | PAYLEN | PORTS |CC|IDX | STA | DCMD |DTYP|MAC|RSV| DTALEN |
427 * +--------------------------------------------------------------+
428 * 63 46 45 40 39 38 36 35 32 31 24 15 0
431 for (n = 0; n < adapter->num_tx_queues; n++) {
432 tx_ring = adapter->tx_ring[n];
433 pr_info("------------------------------------\n");
434 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
435 pr_info("------------------------------------\n");
436 pr_info("T [desc] [address 63:0 ] [PlPOCIStDDM Ln] [bi->dma ] leng ntw timestamp bi->skb\n");
438 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
439 const char *next_desc;
440 struct igb_tx_buffer *buffer_info;
441 tx_desc = IGB_TX_DESC(tx_ring, i);
442 buffer_info = &tx_ring->tx_buffer_info[i];
443 u0 = (struct my_u0 *)tx_desc;
444 if (i == tx_ring->next_to_use &&
445 i == tx_ring->next_to_clean)
446 next_desc = " NTC/U";
447 else if (i == tx_ring->next_to_use)
449 else if (i == tx_ring->next_to_clean)
454 pr_info("T [0x%03X] %016llX %016llX %016llX %04X %p %016llX %p%s\n",
455 i, le64_to_cpu(u0->a),
457 (u64)dma_unmap_addr(buffer_info, dma),
458 dma_unmap_len(buffer_info, len),
459 buffer_info->next_to_watch,
460 (u64)buffer_info->time_stamp,
461 buffer_info->skb, next_desc);
463 if (netif_msg_pktdata(adapter) && buffer_info->skb)
464 print_hex_dump(KERN_INFO, "",
466 16, 1, buffer_info->skb->data,
467 dma_unmap_len(buffer_info, len),
472 /* Print RX Rings Summary */
474 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
475 pr_info("Queue [NTU] [NTC]\n");
476 for (n = 0; n < adapter->num_rx_queues; n++) {
477 rx_ring = adapter->rx_ring[n];
478 pr_info(" %5d %5X %5X\n",
479 n, rx_ring->next_to_use, rx_ring->next_to_clean);
483 if (!netif_msg_rx_status(adapter))
486 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
488 /* Advanced Receive Descriptor (Read) Format
490 * +-----------------------------------------------------+
491 * 0 | Packet Buffer Address [63:1] |A0/NSE|
492 * +----------------------------------------------+------+
493 * 8 | Header Buffer Address [63:1] | DD |
494 * +-----------------------------------------------------+
497 * Advanced Receive Descriptor (Write-Back) Format
499 * 63 48 47 32 31 30 21 20 17 16 4 3 0
500 * +------------------------------------------------------+
501 * 0 | Packet IP |SPH| HDR_LEN | RSV|Packet| RSS |
502 * | Checksum Ident | | | | Type | Type |
503 * +------------------------------------------------------+
504 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
505 * +------------------------------------------------------+
506 * 63 48 47 32 31 20 19 0
509 for (n = 0; n < adapter->num_rx_queues; n++) {
510 rx_ring = adapter->rx_ring[n];
511 pr_info("------------------------------------\n");
512 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
513 pr_info("------------------------------------\n");
514 pr_info("R [desc] [ PktBuf A0] [ HeadBuf DD] [bi->dma ] [bi->skb] <-- Adv Rx Read format\n");
515 pr_info("RWB[desc] [PcsmIpSHl PtRs] [vl er S cks ln] ---------------- [bi->skb] <-- Adv Rx Write-Back format\n");
517 for (i = 0; i < rx_ring->count; i++) {
518 const char *next_desc;
519 struct igb_rx_buffer *buffer_info;
520 buffer_info = &rx_ring->rx_buffer_info[i];
521 rx_desc = IGB_RX_DESC(rx_ring, i);
522 u0 = (struct my_u0 *)rx_desc;
523 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
525 if (i == rx_ring->next_to_use)
527 else if (i == rx_ring->next_to_clean)
532 if (staterr & E1000_RXD_STAT_DD) {
533 /* Descriptor Done */
534 pr_info("%s[0x%03X] %016llX %016llX ---------------- %s\n",
540 pr_info("%s[0x%03X] %016llX %016llX %016llX %s\n",
544 (u64)buffer_info->dma,
547 if (netif_msg_pktdata(adapter) &&
548 buffer_info->dma && buffer_info->page) {
549 print_hex_dump(KERN_INFO, "",
552 page_address(buffer_info->page) +
553 buffer_info->page_offset,
565 * igb_get_i2c_data - Reads the I2C SDA data bit
566 * @hw: pointer to hardware structure
567 * @i2cctl: Current value of I2CCTL register
569 * Returns the I2C data bit value
571 static int igb_get_i2c_data(void *data)
573 struct igb_adapter *adapter = (struct igb_adapter *)data;
574 struct e1000_hw *hw = &adapter->hw;
575 s32 i2cctl = rd32(E1000_I2CPARAMS);
577 return !!(i2cctl & E1000_I2C_DATA_IN);
581 * igb_set_i2c_data - Sets the I2C data bit
582 * @data: pointer to hardware structure
583 * @state: I2C data value (0 or 1) to set
585 * Sets the I2C data bit
587 static void igb_set_i2c_data(void *data, int state)
589 struct igb_adapter *adapter = (struct igb_adapter *)data;
590 struct e1000_hw *hw = &adapter->hw;
591 s32 i2cctl = rd32(E1000_I2CPARAMS);
594 i2cctl |= E1000_I2C_DATA_OUT;
596 i2cctl &= ~E1000_I2C_DATA_OUT;
598 i2cctl &= ~E1000_I2C_DATA_OE_N;
599 i2cctl |= E1000_I2C_CLK_OE_N;
600 wr32(E1000_I2CPARAMS, i2cctl);
606 * igb_set_i2c_clk - Sets the I2C SCL clock
607 * @data: pointer to hardware structure
608 * @state: state to set clock
610 * Sets the I2C clock line to state
612 static void igb_set_i2c_clk(void *data, int state)
614 struct igb_adapter *adapter = (struct igb_adapter *)data;
615 struct e1000_hw *hw = &adapter->hw;
616 s32 i2cctl = rd32(E1000_I2CPARAMS);
619 i2cctl |= E1000_I2C_CLK_OUT;
620 i2cctl &= ~E1000_I2C_CLK_OE_N;
622 i2cctl &= ~E1000_I2C_CLK_OUT;
623 i2cctl &= ~E1000_I2C_CLK_OE_N;
625 wr32(E1000_I2CPARAMS, i2cctl);
630 * igb_get_i2c_clk - Gets the I2C SCL clock state
631 * @data: pointer to hardware structure
633 * Gets the I2C clock state
635 static int igb_get_i2c_clk(void *data)
637 struct igb_adapter *adapter = (struct igb_adapter *)data;
638 struct e1000_hw *hw = &adapter->hw;
639 s32 i2cctl = rd32(E1000_I2CPARAMS);
641 return !!(i2cctl & E1000_I2C_CLK_IN);
644 static const struct i2c_algo_bit_data igb_i2c_algo = {
645 .setsda = igb_set_i2c_data,
646 .setscl = igb_set_i2c_clk,
647 .getsda = igb_get_i2c_data,
648 .getscl = igb_get_i2c_clk,
654 * igb_get_hw_dev - return device
655 * @hw: pointer to hardware structure
657 * used by hardware layer to print debugging information
659 struct net_device *igb_get_hw_dev(struct e1000_hw *hw)
661 struct igb_adapter *adapter = hw->back;
662 return adapter->netdev;
666 * igb_init_module - Driver Registration Routine
668 * igb_init_module is the first routine called when the driver is
669 * loaded. All it does is register with the PCI subsystem.
671 static int __init igb_init_module(void)
675 pr_info("%s - version %s\n",
676 igb_driver_string, igb_driver_version);
677 pr_info("%s\n", igb_copyright);
679 #ifdef CONFIG_IGB_DCA
680 dca_register_notify(&dca_notifier);
682 ret = pci_register_driver(&igb_driver);
686 module_init(igb_init_module);
689 * igb_exit_module - Driver Exit Cleanup Routine
691 * igb_exit_module is called just before the driver is removed
694 static void __exit igb_exit_module(void)
696 #ifdef CONFIG_IGB_DCA
697 dca_unregister_notify(&dca_notifier);
699 pci_unregister_driver(&igb_driver);
702 module_exit(igb_exit_module);
704 #define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
706 * igb_cache_ring_register - Descriptor ring to register mapping
707 * @adapter: board private structure to initialize
709 * Once we know the feature-set enabled for the device, we'll cache
710 * the register offset the descriptor ring is assigned to.
712 static void igb_cache_ring_register(struct igb_adapter *adapter)
715 u32 rbase_offset = adapter->vfs_allocated_count;
717 switch (adapter->hw.mac.type) {
719 /* The queues are allocated for virtualization such that VF 0
720 * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
721 * In order to avoid collision we start at the first free queue
722 * and continue consuming queues in the same sequence
724 if (adapter->vfs_allocated_count) {
725 for (; i < adapter->rss_queues; i++)
726 adapter->rx_ring[i]->reg_idx = rbase_offset +
738 for (; i < adapter->num_rx_queues; i++)
739 adapter->rx_ring[i]->reg_idx = rbase_offset + i;
740 for (; j < adapter->num_tx_queues; j++)
741 adapter->tx_ring[j]->reg_idx = rbase_offset + j;
746 u32 igb_rd32(struct e1000_hw *hw, u32 reg)
748 struct igb_adapter *igb = container_of(hw, struct igb_adapter, hw);
749 u8 __iomem *hw_addr = ACCESS_ONCE(hw->hw_addr);
752 if (E1000_REMOVED(hw_addr))
755 value = readl(&hw_addr[reg]);
757 /* reads should not return all F's */
758 if (!(~value) && (!reg || !(~readl(hw_addr)))) {
759 struct net_device *netdev = igb->netdev;
761 netif_device_detach(netdev);
762 netdev_err(netdev, "PCIe link lost, device now detached\n");
769 * igb_write_ivar - configure ivar for given MSI-X vector
770 * @hw: pointer to the HW structure
771 * @msix_vector: vector number we are allocating to a given ring
772 * @index: row index of IVAR register to write within IVAR table
773 * @offset: column offset of in IVAR, should be multiple of 8
775 * This function is intended to handle the writing of the IVAR register
776 * for adapters 82576 and newer. The IVAR table consists of 2 columns,
777 * each containing an cause allocation for an Rx and Tx ring, and a
778 * variable number of rows depending on the number of queues supported.
780 static void igb_write_ivar(struct e1000_hw *hw, int msix_vector,
781 int index, int offset)
783 u32 ivar = array_rd32(E1000_IVAR0, index);
785 /* clear any bits that are currently set */
786 ivar &= ~((u32)0xFF << offset);
788 /* write vector and valid bit */
789 ivar |= (msix_vector | E1000_IVAR_VALID) << offset;
791 array_wr32(E1000_IVAR0, index, ivar);
794 #define IGB_N0_QUEUE -1
795 static void igb_assign_vector(struct igb_q_vector *q_vector, int msix_vector)
797 struct igb_adapter *adapter = q_vector->adapter;
798 struct e1000_hw *hw = &adapter->hw;
799 int rx_queue = IGB_N0_QUEUE;
800 int tx_queue = IGB_N0_QUEUE;
803 if (q_vector->rx.ring)
804 rx_queue = q_vector->rx.ring->reg_idx;
805 if (q_vector->tx.ring)
806 tx_queue = q_vector->tx.ring->reg_idx;
808 switch (hw->mac.type) {
810 /* The 82575 assigns vectors using a bitmask, which matches the
811 * bitmask for the EICR/EIMS/EIMC registers. To assign one
812 * or more queues to a vector, we write the appropriate bits
813 * into the MSIXBM register for that vector.
815 if (rx_queue > IGB_N0_QUEUE)
816 msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
817 if (tx_queue > IGB_N0_QUEUE)
818 msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
819 if (!(adapter->flags & IGB_FLAG_HAS_MSIX) && msix_vector == 0)
820 msixbm |= E1000_EIMS_OTHER;
821 array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
822 q_vector->eims_value = msixbm;
825 /* 82576 uses a table that essentially consists of 2 columns
826 * with 8 rows. The ordering is column-major so we use the
827 * lower 3 bits as the row index, and the 4th bit as the
830 if (rx_queue > IGB_N0_QUEUE)
831 igb_write_ivar(hw, msix_vector,
833 (rx_queue & 0x8) << 1);
834 if (tx_queue > IGB_N0_QUEUE)
835 igb_write_ivar(hw, msix_vector,
837 ((tx_queue & 0x8) << 1) + 8);
838 q_vector->eims_value = 1 << msix_vector;
845 /* On 82580 and newer adapters the scheme is similar to 82576
846 * however instead of ordering column-major we have things
847 * ordered row-major. So we traverse the table by using
848 * bit 0 as the column offset, and the remaining bits as the
851 if (rx_queue > IGB_N0_QUEUE)
852 igb_write_ivar(hw, msix_vector,
854 (rx_queue & 0x1) << 4);
855 if (tx_queue > IGB_N0_QUEUE)
856 igb_write_ivar(hw, msix_vector,
858 ((tx_queue & 0x1) << 4) + 8);
859 q_vector->eims_value = 1 << msix_vector;
866 /* add q_vector eims value to global eims_enable_mask */
867 adapter->eims_enable_mask |= q_vector->eims_value;
869 /* configure q_vector to set itr on first interrupt */
870 q_vector->set_itr = 1;
874 * igb_configure_msix - Configure MSI-X hardware
875 * @adapter: board private structure to initialize
877 * igb_configure_msix sets up the hardware to properly
878 * generate MSI-X interrupts.
880 static void igb_configure_msix(struct igb_adapter *adapter)
884 struct e1000_hw *hw = &adapter->hw;
886 adapter->eims_enable_mask = 0;
888 /* set vector for other causes, i.e. link changes */
889 switch (hw->mac.type) {
891 tmp = rd32(E1000_CTRL_EXT);
892 /* enable MSI-X PBA support*/
893 tmp |= E1000_CTRL_EXT_PBA_CLR;
895 /* Auto-Mask interrupts upon ICR read. */
896 tmp |= E1000_CTRL_EXT_EIAME;
897 tmp |= E1000_CTRL_EXT_IRCA;
899 wr32(E1000_CTRL_EXT, tmp);
901 /* enable msix_other interrupt */
902 array_wr32(E1000_MSIXBM(0), vector++, E1000_EIMS_OTHER);
903 adapter->eims_other = E1000_EIMS_OTHER;
913 /* Turn on MSI-X capability first, or our settings
914 * won't stick. And it will take days to debug.
916 wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
917 E1000_GPIE_PBA | E1000_GPIE_EIAME |
920 /* enable msix_other interrupt */
921 adapter->eims_other = 1 << vector;
922 tmp = (vector++ | E1000_IVAR_VALID) << 8;
924 wr32(E1000_IVAR_MISC, tmp);
927 /* do nothing, since nothing else supports MSI-X */
929 } /* switch (hw->mac.type) */
931 adapter->eims_enable_mask |= adapter->eims_other;
933 for (i = 0; i < adapter->num_q_vectors; i++)
934 igb_assign_vector(adapter->q_vector[i], vector++);
940 * igb_request_msix - Initialize MSI-X interrupts
941 * @adapter: board private structure to initialize
943 * igb_request_msix allocates MSI-X vectors and requests interrupts from the
946 static int igb_request_msix(struct igb_adapter *adapter)
948 struct net_device *netdev = adapter->netdev;
949 struct e1000_hw *hw = &adapter->hw;
950 int i, err = 0, vector = 0, free_vector = 0;
952 err = request_irq(adapter->msix_entries[vector].vector,
953 igb_msix_other, 0, netdev->name, adapter);
957 for (i = 0; i < adapter->num_q_vectors; i++) {
958 struct igb_q_vector *q_vector = adapter->q_vector[i];
962 q_vector->itr_register = hw->hw_addr + E1000_EITR(vector);
964 if (q_vector->rx.ring && q_vector->tx.ring)
965 sprintf(q_vector->name, "%s-TxRx-%u", netdev->name,
966 q_vector->rx.ring->queue_index);
967 else if (q_vector->tx.ring)
968 sprintf(q_vector->name, "%s-tx-%u", netdev->name,
969 q_vector->tx.ring->queue_index);
970 else if (q_vector->rx.ring)
971 sprintf(q_vector->name, "%s-rx-%u", netdev->name,
972 q_vector->rx.ring->queue_index);
974 sprintf(q_vector->name, "%s-unused", netdev->name);
976 err = request_irq(adapter->msix_entries[vector].vector,
977 igb_msix_ring, 0, q_vector->name,
983 igb_configure_msix(adapter);
987 /* free already assigned IRQs */
988 free_irq(adapter->msix_entries[free_vector++].vector, adapter);
991 for (i = 0; i < vector; i++) {
992 free_irq(adapter->msix_entries[free_vector++].vector,
993 adapter->q_vector[i]);
1000 * igb_free_q_vector - Free memory allocated for specific interrupt vector
1001 * @adapter: board private structure to initialize
1002 * @v_idx: Index of vector to be freed
1004 * This function frees the memory allocated to the q_vector.
1006 static void igb_free_q_vector(struct igb_adapter *adapter, int v_idx)
1008 struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
1010 adapter->q_vector[v_idx] = NULL;
1012 /* igb_get_stats64() might access the rings on this vector,
1013 * we must wait a grace period before freeing it.
1015 kfree_rcu(q_vector, rcu);
1019 * igb_reset_q_vector - Reset config for interrupt vector
1020 * @adapter: board private structure to initialize
1021 * @v_idx: Index of vector to be reset
1023 * If NAPI is enabled it will delete any references to the
1024 * NAPI struct. This is preparation for igb_free_q_vector.
1026 static void igb_reset_q_vector(struct igb_adapter *adapter, int v_idx)
1028 struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
1030 /* Coming from igb_set_interrupt_capability, the vectors are not yet
1031 * allocated. So, q_vector is NULL so we should stop here.
1036 if (q_vector->tx.ring)
1037 adapter->tx_ring[q_vector->tx.ring->queue_index] = NULL;
1039 if (q_vector->rx.ring)
1040 adapter->tx_ring[q_vector->rx.ring->queue_index] = NULL;
1042 netif_napi_del(&q_vector->napi);
1046 static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
1048 int v_idx = adapter->num_q_vectors;
1050 if (adapter->flags & IGB_FLAG_HAS_MSIX)
1051 pci_disable_msix(adapter->pdev);
1052 else if (adapter->flags & IGB_FLAG_HAS_MSI)
1053 pci_disable_msi(adapter->pdev);
1056 igb_reset_q_vector(adapter, v_idx);
1060 * igb_free_q_vectors - Free memory allocated for interrupt vectors
1061 * @adapter: board private structure to initialize
1063 * This function frees the memory allocated to the q_vectors. In addition if
1064 * NAPI is enabled it will delete any references to the NAPI struct prior
1065 * to freeing the q_vector.
1067 static void igb_free_q_vectors(struct igb_adapter *adapter)
1069 int v_idx = adapter->num_q_vectors;
1071 adapter->num_tx_queues = 0;
1072 adapter->num_rx_queues = 0;
1073 adapter->num_q_vectors = 0;
1076 igb_reset_q_vector(adapter, v_idx);
1077 igb_free_q_vector(adapter, v_idx);
1082 * igb_clear_interrupt_scheme - reset the device to a state of no interrupts
1083 * @adapter: board private structure to initialize
1085 * This function resets the device so that it has 0 Rx queues, Tx queues, and
1086 * MSI-X interrupts allocated.
1088 static void igb_clear_interrupt_scheme(struct igb_adapter *adapter)
1090 igb_free_q_vectors(adapter);
1091 igb_reset_interrupt_capability(adapter);
1095 * igb_set_interrupt_capability - set MSI or MSI-X if supported
1096 * @adapter: board private structure to initialize
1097 * @msix: boolean value of MSIX capability
1099 * Attempt to configure interrupts using the best available
1100 * capabilities of the hardware and kernel.
1102 static void igb_set_interrupt_capability(struct igb_adapter *adapter, bool msix)
1109 adapter->flags |= IGB_FLAG_HAS_MSIX;
1111 /* Number of supported queues. */
1112 adapter->num_rx_queues = adapter->rss_queues;
1113 if (adapter->vfs_allocated_count)
1114 adapter->num_tx_queues = 1;
1116 adapter->num_tx_queues = adapter->rss_queues;
1118 /* start with one vector for every Rx queue */
1119 numvecs = adapter->num_rx_queues;
1121 /* if Tx handler is separate add 1 for every Tx queue */
1122 if (!(adapter->flags & IGB_FLAG_QUEUE_PAIRS))
1123 numvecs += adapter->num_tx_queues;
1125 /* store the number of vectors reserved for queues */
1126 adapter->num_q_vectors = numvecs;
1128 /* add 1 vector for link status interrupts */
1130 for (i = 0; i < numvecs; i++)
1131 adapter->msix_entries[i].entry = i;
1133 err = pci_enable_msix_range(adapter->pdev,
1134 adapter->msix_entries,
1140 igb_reset_interrupt_capability(adapter);
1142 /* If we can't do MSI-X, try MSI */
1144 adapter->flags &= ~IGB_FLAG_HAS_MSIX;
1145 #ifdef CONFIG_PCI_IOV
1146 /* disable SR-IOV for non MSI-X configurations */
1147 if (adapter->vf_data) {
1148 struct e1000_hw *hw = &adapter->hw;
1149 /* disable iov and allow time for transactions to clear */
1150 pci_disable_sriov(adapter->pdev);
1153 kfree(adapter->vf_data);
1154 adapter->vf_data = NULL;
1155 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
1158 dev_info(&adapter->pdev->dev, "IOV Disabled\n");
1161 adapter->vfs_allocated_count = 0;
1162 adapter->rss_queues = 1;
1163 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
1164 adapter->num_rx_queues = 1;
1165 adapter->num_tx_queues = 1;
1166 adapter->num_q_vectors = 1;
1167 if (!pci_enable_msi(adapter->pdev))
1168 adapter->flags |= IGB_FLAG_HAS_MSI;
1171 static void igb_add_ring(struct igb_ring *ring,
1172 struct igb_ring_container *head)
1179 * igb_alloc_q_vector - Allocate memory for a single interrupt vector
1180 * @adapter: board private structure to initialize
1181 * @v_count: q_vectors allocated on adapter, used for ring interleaving
1182 * @v_idx: index of vector in adapter struct
1183 * @txr_count: total number of Tx rings to allocate
1184 * @txr_idx: index of first Tx ring to allocate
1185 * @rxr_count: total number of Rx rings to allocate
1186 * @rxr_idx: index of first Rx ring to allocate
1188 * We allocate one q_vector. If allocation fails we return -ENOMEM.
1190 static int igb_alloc_q_vector(struct igb_adapter *adapter,
1191 int v_count, int v_idx,
1192 int txr_count, int txr_idx,
1193 int rxr_count, int rxr_idx)
1195 struct igb_q_vector *q_vector;
1196 struct igb_ring *ring;
1197 int ring_count, size;
1199 /* igb only supports 1 Tx and/or 1 Rx queue per vector */
1200 if (txr_count > 1 || rxr_count > 1)
1203 ring_count = txr_count + rxr_count;
1204 size = sizeof(struct igb_q_vector) +
1205 (sizeof(struct igb_ring) * ring_count);
1207 /* allocate q_vector and rings */
1208 q_vector = adapter->q_vector[v_idx];
1210 q_vector = kzalloc(size, GFP_KERNEL);
1214 /* initialize NAPI */
1215 netif_napi_add(adapter->netdev, &q_vector->napi,
1218 /* tie q_vector and adapter together */
1219 adapter->q_vector[v_idx] = q_vector;
1220 q_vector->adapter = adapter;
1222 /* initialize work limits */
1223 q_vector->tx.work_limit = adapter->tx_work_limit;
1225 /* initialize ITR configuration */
1226 q_vector->itr_register = adapter->hw.hw_addr + E1000_EITR(0);
1227 q_vector->itr_val = IGB_START_ITR;
1229 /* initialize pointer to rings */
1230 ring = q_vector->ring;
1234 /* rx or rx/tx vector */
1235 if (!adapter->rx_itr_setting || adapter->rx_itr_setting > 3)
1236 q_vector->itr_val = adapter->rx_itr_setting;
1238 /* tx only vector */
1239 if (!adapter->tx_itr_setting || adapter->tx_itr_setting > 3)
1240 q_vector->itr_val = adapter->tx_itr_setting;
1244 /* assign generic ring traits */
1245 ring->dev = &adapter->pdev->dev;
1246 ring->netdev = adapter->netdev;
1248 /* configure backlink on ring */
1249 ring->q_vector = q_vector;
1251 /* update q_vector Tx values */
1252 igb_add_ring(ring, &q_vector->tx);
1254 /* For 82575, context index must be unique per ring. */
1255 if (adapter->hw.mac.type == e1000_82575)
1256 set_bit(IGB_RING_FLAG_TX_CTX_IDX, &ring->flags);
1258 /* apply Tx specific ring traits */
1259 ring->count = adapter->tx_ring_count;
1260 ring->queue_index = txr_idx;
1262 u64_stats_init(&ring->tx_syncp);
1263 u64_stats_init(&ring->tx_syncp2);
1265 /* assign ring to adapter */
1266 adapter->tx_ring[txr_idx] = ring;
1268 /* push pointer to next ring */
1273 /* assign generic ring traits */
1274 ring->dev = &adapter->pdev->dev;
1275 ring->netdev = adapter->netdev;
1277 /* configure backlink on ring */
1278 ring->q_vector = q_vector;
1280 /* update q_vector Rx values */
1281 igb_add_ring(ring, &q_vector->rx);
1283 /* set flag indicating ring supports SCTP checksum offload */
1284 if (adapter->hw.mac.type >= e1000_82576)
1285 set_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags);
1287 /* On i350, i354, i210, and i211, loopback VLAN packets
1288 * have the tag byte-swapped.
1290 if (adapter->hw.mac.type >= e1000_i350)
1291 set_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &ring->flags);
1293 /* apply Rx specific ring traits */
1294 ring->count = adapter->rx_ring_count;
1295 ring->queue_index = rxr_idx;
1297 u64_stats_init(&ring->rx_syncp);
1299 /* assign ring to adapter */
1300 adapter->rx_ring[rxr_idx] = ring;
1308 * igb_alloc_q_vectors - Allocate memory for interrupt vectors
1309 * @adapter: board private structure to initialize
1311 * We allocate one q_vector per queue interrupt. If allocation fails we
1314 static int igb_alloc_q_vectors(struct igb_adapter *adapter)
1316 int q_vectors = adapter->num_q_vectors;
1317 int rxr_remaining = adapter->num_rx_queues;
1318 int txr_remaining = adapter->num_tx_queues;
1319 int rxr_idx = 0, txr_idx = 0, v_idx = 0;
1322 if (q_vectors >= (rxr_remaining + txr_remaining)) {
1323 for (; rxr_remaining; v_idx++) {
1324 err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1330 /* update counts and index */
1336 for (; v_idx < q_vectors; v_idx++) {
1337 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_idx);
1338 int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_idx);
1340 err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1341 tqpv, txr_idx, rqpv, rxr_idx);
1346 /* update counts and index */
1347 rxr_remaining -= rqpv;
1348 txr_remaining -= tqpv;
1356 adapter->num_tx_queues = 0;
1357 adapter->num_rx_queues = 0;
1358 adapter->num_q_vectors = 0;
1361 igb_free_q_vector(adapter, v_idx);
1367 * igb_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
1368 * @adapter: board private structure to initialize
1369 * @msix: boolean value of MSIX capability
1371 * This function initializes the interrupts and allocates all of the queues.
1373 static int igb_init_interrupt_scheme(struct igb_adapter *adapter, bool msix)
1375 struct pci_dev *pdev = adapter->pdev;
1378 igb_set_interrupt_capability(adapter, msix);
1380 err = igb_alloc_q_vectors(adapter);
1382 dev_err(&pdev->dev, "Unable to allocate memory for vectors\n");
1383 goto err_alloc_q_vectors;
1386 igb_cache_ring_register(adapter);
1390 err_alloc_q_vectors:
1391 igb_reset_interrupt_capability(adapter);
1396 * igb_request_irq - initialize interrupts
1397 * @adapter: board private structure to initialize
1399 * Attempts to configure interrupts using the best available
1400 * capabilities of the hardware and kernel.
1402 static int igb_request_irq(struct igb_adapter *adapter)
1404 struct net_device *netdev = adapter->netdev;
1405 struct pci_dev *pdev = adapter->pdev;
1408 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1409 err = igb_request_msix(adapter);
1412 /* fall back to MSI */
1413 igb_free_all_tx_resources(adapter);
1414 igb_free_all_rx_resources(adapter);
1416 igb_clear_interrupt_scheme(adapter);
1417 err = igb_init_interrupt_scheme(adapter, false);
1421 igb_setup_all_tx_resources(adapter);
1422 igb_setup_all_rx_resources(adapter);
1423 igb_configure(adapter);
1426 igb_assign_vector(adapter->q_vector[0], 0);
1428 if (adapter->flags & IGB_FLAG_HAS_MSI) {
1429 err = request_irq(pdev->irq, igb_intr_msi, 0,
1430 netdev->name, adapter);
1434 /* fall back to legacy interrupts */
1435 igb_reset_interrupt_capability(adapter);
1436 adapter->flags &= ~IGB_FLAG_HAS_MSI;
1439 err = request_irq(pdev->irq, igb_intr, IRQF_SHARED,
1440 netdev->name, adapter);
1443 dev_err(&pdev->dev, "Error %d getting interrupt\n",
1450 static void igb_free_irq(struct igb_adapter *adapter)
1452 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1455 free_irq(adapter->msix_entries[vector++].vector, adapter);
1457 for (i = 0; i < adapter->num_q_vectors; i++)
1458 free_irq(adapter->msix_entries[vector++].vector,
1459 adapter->q_vector[i]);
1461 free_irq(adapter->pdev->irq, adapter);
1466 * igb_irq_disable - Mask off interrupt generation on the NIC
1467 * @adapter: board private structure
1469 static void igb_irq_disable(struct igb_adapter *adapter)
1471 struct e1000_hw *hw = &adapter->hw;
1473 /* we need to be careful when disabling interrupts. The VFs are also
1474 * mapped into these registers and so clearing the bits can cause
1475 * issues on the VF drivers so we only need to clear what we set
1477 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1478 u32 regval = rd32(E1000_EIAM);
1480 wr32(E1000_EIAM, regval & ~adapter->eims_enable_mask);
1481 wr32(E1000_EIMC, adapter->eims_enable_mask);
1482 regval = rd32(E1000_EIAC);
1483 wr32(E1000_EIAC, regval & ~adapter->eims_enable_mask);
1487 wr32(E1000_IMC, ~0);
1489 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1492 for (i = 0; i < adapter->num_q_vectors; i++)
1493 synchronize_irq(adapter->msix_entries[i].vector);
1495 synchronize_irq(adapter->pdev->irq);
1500 * igb_irq_enable - Enable default interrupt generation settings
1501 * @adapter: board private structure
1503 static void igb_irq_enable(struct igb_adapter *adapter)
1505 struct e1000_hw *hw = &adapter->hw;
1507 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1508 u32 ims = E1000_IMS_LSC | E1000_IMS_DOUTSYNC | E1000_IMS_DRSTA;
1509 u32 regval = rd32(E1000_EIAC);
1511 wr32(E1000_EIAC, regval | adapter->eims_enable_mask);
1512 regval = rd32(E1000_EIAM);
1513 wr32(E1000_EIAM, regval | adapter->eims_enable_mask);
1514 wr32(E1000_EIMS, adapter->eims_enable_mask);
1515 if (adapter->vfs_allocated_count) {
1516 wr32(E1000_MBVFIMR, 0xFF);
1517 ims |= E1000_IMS_VMMB;
1519 wr32(E1000_IMS, ims);
1521 wr32(E1000_IMS, IMS_ENABLE_MASK |
1523 wr32(E1000_IAM, IMS_ENABLE_MASK |
1528 static void igb_update_mng_vlan(struct igb_adapter *adapter)
1530 struct e1000_hw *hw = &adapter->hw;
1531 u16 vid = adapter->hw.mng_cookie.vlan_id;
1532 u16 old_vid = adapter->mng_vlan_id;
1534 if (hw->mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1535 /* add VID to filter table */
1536 igb_vfta_set(hw, vid, true);
1537 adapter->mng_vlan_id = vid;
1539 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1542 if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
1544 !test_bit(old_vid, adapter->active_vlans)) {
1545 /* remove VID from filter table */
1546 igb_vfta_set(hw, old_vid, false);
1551 * igb_release_hw_control - release control of the h/w to f/w
1552 * @adapter: address of board private structure
1554 * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
1555 * For ASF and Pass Through versions of f/w this means that the
1556 * driver is no longer loaded.
1558 static void igb_release_hw_control(struct igb_adapter *adapter)
1560 struct e1000_hw *hw = &adapter->hw;
1563 /* Let firmware take over control of h/w */
1564 ctrl_ext = rd32(E1000_CTRL_EXT);
1565 wr32(E1000_CTRL_EXT,
1566 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1570 * igb_get_hw_control - get control of the h/w from f/w
1571 * @adapter: address of board private structure
1573 * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
1574 * For ASF and Pass Through versions of f/w this means that
1575 * the driver is loaded.
1577 static void igb_get_hw_control(struct igb_adapter *adapter)
1579 struct e1000_hw *hw = &adapter->hw;
1582 /* Let firmware know the driver has taken over */
1583 ctrl_ext = rd32(E1000_CTRL_EXT);
1584 wr32(E1000_CTRL_EXT,
1585 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1589 * igb_configure - configure the hardware for RX and TX
1590 * @adapter: private board structure
1592 static void igb_configure(struct igb_adapter *adapter)
1594 struct net_device *netdev = adapter->netdev;
1597 igb_get_hw_control(adapter);
1598 igb_set_rx_mode(netdev);
1600 igb_restore_vlan(adapter);
1602 igb_setup_tctl(adapter);
1603 igb_setup_mrqc(adapter);
1604 igb_setup_rctl(adapter);
1606 igb_configure_tx(adapter);
1607 igb_configure_rx(adapter);
1609 igb_rx_fifo_flush_82575(&adapter->hw);
1611 /* call igb_desc_unused which always leaves
1612 * at least 1 descriptor unused to make sure
1613 * next_to_use != next_to_clean
1615 for (i = 0; i < adapter->num_rx_queues; i++) {
1616 struct igb_ring *ring = adapter->rx_ring[i];
1617 igb_alloc_rx_buffers(ring, igb_desc_unused(ring));
1622 * igb_power_up_link - Power up the phy/serdes link
1623 * @adapter: address of board private structure
1625 void igb_power_up_link(struct igb_adapter *adapter)
1627 igb_reset_phy(&adapter->hw);
1629 if (adapter->hw.phy.media_type == e1000_media_type_copper)
1630 igb_power_up_phy_copper(&adapter->hw);
1632 igb_power_up_serdes_link_82575(&adapter->hw);
1634 igb_setup_link(&adapter->hw);
1638 * igb_power_down_link - Power down the phy/serdes link
1639 * @adapter: address of board private structure
1641 static void igb_power_down_link(struct igb_adapter *adapter)
1643 if (adapter->hw.phy.media_type == e1000_media_type_copper)
1644 igb_power_down_phy_copper_82575(&adapter->hw);
1646 igb_shutdown_serdes_link_82575(&adapter->hw);
1650 * Detect and switch function for Media Auto Sense
1651 * @adapter: address of the board private structure
1653 static void igb_check_swap_media(struct igb_adapter *adapter)
1655 struct e1000_hw *hw = &adapter->hw;
1656 u32 ctrl_ext, connsw;
1657 bool swap_now = false;
1659 ctrl_ext = rd32(E1000_CTRL_EXT);
1660 connsw = rd32(E1000_CONNSW);
1662 /* need to live swap if current media is copper and we have fiber/serdes
1666 if ((hw->phy.media_type == e1000_media_type_copper) &&
1667 (!(connsw & E1000_CONNSW_AUTOSENSE_EN))) {
1669 } else if (!(connsw & E1000_CONNSW_SERDESD)) {
1670 /* copper signal takes time to appear */
1671 if (adapter->copper_tries < 4) {
1672 adapter->copper_tries++;
1673 connsw |= E1000_CONNSW_AUTOSENSE_CONF;
1674 wr32(E1000_CONNSW, connsw);
1677 adapter->copper_tries = 0;
1678 if ((connsw & E1000_CONNSW_PHYSD) &&
1679 (!(connsw & E1000_CONNSW_PHY_PDN))) {
1681 connsw &= ~E1000_CONNSW_AUTOSENSE_CONF;
1682 wr32(E1000_CONNSW, connsw);
1690 switch (hw->phy.media_type) {
1691 case e1000_media_type_copper:
1692 netdev_info(adapter->netdev,
1693 "MAS: changing media to fiber/serdes\n");
1695 E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
1696 adapter->flags |= IGB_FLAG_MEDIA_RESET;
1697 adapter->copper_tries = 0;
1699 case e1000_media_type_internal_serdes:
1700 case e1000_media_type_fiber:
1701 netdev_info(adapter->netdev,
1702 "MAS: changing media to copper\n");
1704 ~E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
1705 adapter->flags |= IGB_FLAG_MEDIA_RESET;
1708 /* shouldn't get here during regular operation */
1709 netdev_err(adapter->netdev,
1710 "AMS: Invalid media type found, returning\n");
1713 wr32(E1000_CTRL_EXT, ctrl_ext);
1717 * igb_up - Open the interface and prepare it to handle traffic
1718 * @adapter: board private structure
1720 int igb_up(struct igb_adapter *adapter)
1722 struct e1000_hw *hw = &adapter->hw;
1725 /* hardware has been reset, we need to reload some things */
1726 igb_configure(adapter);
1728 clear_bit(__IGB_DOWN, &adapter->state);
1730 for (i = 0; i < adapter->num_q_vectors; i++)
1731 napi_enable(&(adapter->q_vector[i]->napi));
1733 if (adapter->flags & IGB_FLAG_HAS_MSIX)
1734 igb_configure_msix(adapter);
1736 igb_assign_vector(adapter->q_vector[0], 0);
1738 /* Clear any pending interrupts. */
1740 igb_irq_enable(adapter);
1742 /* notify VFs that reset has been completed */
1743 if (adapter->vfs_allocated_count) {
1744 u32 reg_data = rd32(E1000_CTRL_EXT);
1746 reg_data |= E1000_CTRL_EXT_PFRSTD;
1747 wr32(E1000_CTRL_EXT, reg_data);
1750 netif_tx_start_all_queues(adapter->netdev);
1752 /* start the watchdog. */
1753 hw->mac.get_link_status = 1;
1754 schedule_work(&adapter->watchdog_task);
1756 if ((adapter->flags & IGB_FLAG_EEE) &&
1757 (!hw->dev_spec._82575.eee_disable))
1758 adapter->eee_advert = MDIO_EEE_100TX | MDIO_EEE_1000T;
1763 void igb_down(struct igb_adapter *adapter)
1765 struct net_device *netdev = adapter->netdev;
1766 struct e1000_hw *hw = &adapter->hw;
1770 /* signal that we're down so the interrupt handler does not
1771 * reschedule our watchdog timer
1773 set_bit(__IGB_DOWN, &adapter->state);
1775 /* disable receives in the hardware */
1776 rctl = rd32(E1000_RCTL);
1777 wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
1778 /* flush and sleep below */
1780 netif_tx_stop_all_queues(netdev);
1782 /* disable transmits in the hardware */
1783 tctl = rd32(E1000_TCTL);
1784 tctl &= ~E1000_TCTL_EN;
1785 wr32(E1000_TCTL, tctl);
1786 /* flush both disables and wait for them to finish */
1788 usleep_range(10000, 11000);
1790 igb_irq_disable(adapter);
1792 adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
1794 for (i = 0; i < adapter->num_q_vectors; i++) {
1795 napi_synchronize(&(adapter->q_vector[i]->napi));
1796 napi_disable(&(adapter->q_vector[i]->napi));
1800 del_timer_sync(&adapter->watchdog_timer);
1801 del_timer_sync(&adapter->phy_info_timer);
1803 netif_carrier_off(netdev);
1805 /* record the stats before reset*/
1806 spin_lock(&adapter->stats64_lock);
1807 igb_update_stats(adapter, &adapter->stats64);
1808 spin_unlock(&adapter->stats64_lock);
1810 adapter->link_speed = 0;
1811 adapter->link_duplex = 0;
1813 if (!pci_channel_offline(adapter->pdev))
1815 igb_clean_all_tx_rings(adapter);
1816 igb_clean_all_rx_rings(adapter);
1817 #ifdef CONFIG_IGB_DCA
1819 /* since we reset the hardware DCA settings were cleared */
1820 igb_setup_dca(adapter);
1824 void igb_reinit_locked(struct igb_adapter *adapter)
1826 WARN_ON(in_interrupt());
1827 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
1828 usleep_range(1000, 2000);
1831 clear_bit(__IGB_RESETTING, &adapter->state);
1834 /** igb_enable_mas - Media Autosense re-enable after swap
1836 * @adapter: adapter struct
1838 static s32 igb_enable_mas(struct igb_adapter *adapter)
1840 struct e1000_hw *hw = &adapter->hw;
1844 connsw = rd32(E1000_CONNSW);
1845 if (!(hw->phy.media_type == e1000_media_type_copper))
1848 /* configure for SerDes media detect */
1849 if (!(connsw & E1000_CONNSW_SERDESD)) {
1850 connsw |= E1000_CONNSW_ENRGSRC;
1851 connsw |= E1000_CONNSW_AUTOSENSE_EN;
1852 wr32(E1000_CONNSW, connsw);
1854 } else if (connsw & E1000_CONNSW_SERDESD) {
1855 /* already SerDes, no need to enable anything */
1858 netdev_info(adapter->netdev,
1859 "MAS: Unable to configure feature, disabling..\n");
1860 adapter->flags &= ~IGB_FLAG_MAS_ENABLE;
1865 void igb_reset(struct igb_adapter *adapter)
1867 struct pci_dev *pdev = adapter->pdev;
1868 struct e1000_hw *hw = &adapter->hw;
1869 struct e1000_mac_info *mac = &hw->mac;
1870 struct e1000_fc_info *fc = &hw->fc;
1871 u32 pba = 0, tx_space, min_tx_space, min_rx_space, hwm;
1873 /* Repartition Pba for greater than 9k mtu
1874 * To take effect CTRL.RST is required.
1876 switch (mac->type) {
1880 pba = rd32(E1000_RXPBS);
1881 pba = igb_rxpbs_adjust_82580(pba);
1884 pba = rd32(E1000_RXPBS);
1885 pba &= E1000_RXPBS_SIZE_MASK_82576;
1891 pba = E1000_PBA_34K;
1895 if ((adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) &&
1896 (mac->type < e1000_82576)) {
1897 /* adjust PBA for jumbo frames */
1898 wr32(E1000_PBA, pba);
1900 /* To maintain wire speed transmits, the Tx FIFO should be
1901 * large enough to accommodate two full transmit packets,
1902 * rounded up to the next 1KB and expressed in KB. Likewise,
1903 * the Rx FIFO should be large enough to accommodate at least
1904 * one full receive packet and is similarly rounded up and
1907 pba = rd32(E1000_PBA);
1908 /* upper 16 bits has Tx packet buffer allocation size in KB */
1909 tx_space = pba >> 16;
1910 /* lower 16 bits has Rx packet buffer allocation size in KB */
1912 /* the Tx fifo also stores 16 bytes of information about the Tx
1913 * but don't include ethernet FCS because hardware appends it
1915 min_tx_space = (adapter->max_frame_size +
1916 sizeof(union e1000_adv_tx_desc) -
1918 min_tx_space = ALIGN(min_tx_space, 1024);
1919 min_tx_space >>= 10;
1920 /* software strips receive CRC, so leave room for it */
1921 min_rx_space = adapter->max_frame_size;
1922 min_rx_space = ALIGN(min_rx_space, 1024);
1923 min_rx_space >>= 10;
1925 /* If current Tx allocation is less than the min Tx FIFO size,
1926 * and the min Tx FIFO size is less than the current Rx FIFO
1927 * allocation, take space away from current Rx allocation
1929 if (tx_space < min_tx_space &&
1930 ((min_tx_space - tx_space) < pba)) {
1931 pba = pba - (min_tx_space - tx_space);
1933 /* if short on Rx space, Rx wins and must trump Tx
1936 if (pba < min_rx_space)
1939 wr32(E1000_PBA, pba);
1942 /* flow control settings */
1943 /* The high water mark must be low enough to fit one full frame
1944 * (or the size used for early receive) above it in the Rx FIFO.
1945 * Set it to the lower of:
1946 * - 90% of the Rx FIFO size, or
1947 * - the full Rx FIFO size minus one full frame
1949 hwm = min(((pba << 10) * 9 / 10),
1950 ((pba << 10) - 2 * adapter->max_frame_size));
1952 fc->high_water = hwm & 0xFFFFFFF0; /* 16-byte granularity */
1953 fc->low_water = fc->high_water - 16;
1954 fc->pause_time = 0xFFFF;
1956 fc->current_mode = fc->requested_mode;
1958 /* disable receive for all VFs and wait one second */
1959 if (adapter->vfs_allocated_count) {
1962 for (i = 0 ; i < adapter->vfs_allocated_count; i++)
1963 adapter->vf_data[i].flags &= IGB_VF_FLAG_PF_SET_MAC;
1965 /* ping all the active vfs to let them know we are going down */
1966 igb_ping_all_vfs(adapter);
1968 /* disable transmits and receives */
1969 wr32(E1000_VFRE, 0);
1970 wr32(E1000_VFTE, 0);
1973 /* Allow time for pending master requests to run */
1974 hw->mac.ops.reset_hw(hw);
1977 if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
1978 /* need to resetup here after media swap */
1979 adapter->ei.get_invariants(hw);
1980 adapter->flags &= ~IGB_FLAG_MEDIA_RESET;
1982 if (adapter->flags & IGB_FLAG_MAS_ENABLE) {
1983 if (igb_enable_mas(adapter))
1985 "Error enabling Media Auto Sense\n");
1987 if (hw->mac.ops.init_hw(hw))
1988 dev_err(&pdev->dev, "Hardware Error\n");
1990 /* Flow control settings reset on hardware reset, so guarantee flow
1991 * control is off when forcing speed.
1993 if (!hw->mac.autoneg)
1994 igb_force_mac_fc(hw);
1996 igb_init_dmac(adapter, pba);
1997 #ifdef CONFIG_IGB_HWMON
1998 /* Re-initialize the thermal sensor on i350 devices. */
1999 if (!test_bit(__IGB_DOWN, &adapter->state)) {
2000 if (mac->type == e1000_i350 && hw->bus.func == 0) {
2001 /* If present, re-initialize the external thermal sensor
2005 mac->ops.init_thermal_sensor_thresh(hw);
2009 /* Re-establish EEE setting */
2010 if (hw->phy.media_type == e1000_media_type_copper) {
2011 switch (mac->type) {
2015 igb_set_eee_i350(hw, true, true);
2018 igb_set_eee_i354(hw, true, true);
2024 if (!netif_running(adapter->netdev))
2025 igb_power_down_link(adapter);
2027 igb_update_mng_vlan(adapter);
2029 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2030 wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
2032 /* Re-enable PTP, where applicable. */
2033 igb_ptp_reset(adapter);
2035 igb_get_phy_info(hw);
2038 static netdev_features_t igb_fix_features(struct net_device *netdev,
2039 netdev_features_t features)
2041 /* Since there is no support for separate Rx/Tx vlan accel
2042 * enable/disable make sure Tx flag is always in same state as Rx.
2044 if (features & NETIF_F_HW_VLAN_CTAG_RX)
2045 features |= NETIF_F_HW_VLAN_CTAG_TX;
2047 features &= ~NETIF_F_HW_VLAN_CTAG_TX;
2052 static int igb_set_features(struct net_device *netdev,
2053 netdev_features_t features)
2055 netdev_features_t changed = netdev->features ^ features;
2056 struct igb_adapter *adapter = netdev_priv(netdev);
2058 if (changed & NETIF_F_HW_VLAN_CTAG_RX)
2059 igb_vlan_mode(netdev, features);
2061 if (!(changed & NETIF_F_RXALL))
2064 netdev->features = features;
2066 if (netif_running(netdev))
2067 igb_reinit_locked(adapter);
2074 static const struct net_device_ops igb_netdev_ops = {
2075 .ndo_open = igb_open,
2076 .ndo_stop = igb_close,
2077 .ndo_start_xmit = igb_xmit_frame,
2078 .ndo_get_stats64 = igb_get_stats64,
2079 .ndo_set_rx_mode = igb_set_rx_mode,
2080 .ndo_set_mac_address = igb_set_mac,
2081 .ndo_change_mtu = igb_change_mtu,
2082 .ndo_do_ioctl = igb_ioctl,
2083 .ndo_tx_timeout = igb_tx_timeout,
2084 .ndo_validate_addr = eth_validate_addr,
2085 .ndo_vlan_rx_add_vid = igb_vlan_rx_add_vid,
2086 .ndo_vlan_rx_kill_vid = igb_vlan_rx_kill_vid,
2087 .ndo_set_vf_mac = igb_ndo_set_vf_mac,
2088 .ndo_set_vf_vlan = igb_ndo_set_vf_vlan,
2089 .ndo_set_vf_rate = igb_ndo_set_vf_bw,
2090 .ndo_set_vf_spoofchk = igb_ndo_set_vf_spoofchk,
2091 .ndo_get_vf_config = igb_ndo_get_vf_config,
2092 #ifdef CONFIG_NET_POLL_CONTROLLER
2093 .ndo_poll_controller = igb_netpoll,
2095 .ndo_fix_features = igb_fix_features,
2096 .ndo_set_features = igb_set_features,
2100 * igb_set_fw_version - Configure version string for ethtool
2101 * @adapter: adapter struct
2103 void igb_set_fw_version(struct igb_adapter *adapter)
2105 struct e1000_hw *hw = &adapter->hw;
2106 struct e1000_fw_version fw;
2108 igb_get_fw_version(hw, &fw);
2110 switch (hw->mac.type) {
2113 if (!(igb_get_flash_presence_i210(hw))) {
2114 snprintf(adapter->fw_version,
2115 sizeof(adapter->fw_version),
2117 fw.invm_major, fw.invm_minor,
2123 /* if option is rom valid, display its version too */
2125 snprintf(adapter->fw_version,
2126 sizeof(adapter->fw_version),
2127 "%d.%d, 0x%08x, %d.%d.%d",
2128 fw.eep_major, fw.eep_minor, fw.etrack_id,
2129 fw.or_major, fw.or_build, fw.or_patch);
2131 } else if (fw.etrack_id != 0X0000) {
2132 snprintf(adapter->fw_version,
2133 sizeof(adapter->fw_version),
2135 fw.eep_major, fw.eep_minor, fw.etrack_id);
2137 snprintf(adapter->fw_version,
2138 sizeof(adapter->fw_version),
2140 fw.eep_major, fw.eep_minor, fw.eep_build);
2147 * igb_init_mas - init Media Autosense feature if enabled in the NVM
2149 * @adapter: adapter struct
2151 static void igb_init_mas(struct igb_adapter *adapter)
2153 struct e1000_hw *hw = &adapter->hw;
2156 hw->nvm.ops.read(hw, NVM_COMPAT, 1, &eeprom_data);
2157 switch (hw->bus.func) {
2159 if (eeprom_data & IGB_MAS_ENABLE_0) {
2160 adapter->flags |= IGB_FLAG_MAS_ENABLE;
2161 netdev_info(adapter->netdev,
2162 "MAS: Enabling Media Autosense for port %d\n",
2167 if (eeprom_data & IGB_MAS_ENABLE_1) {
2168 adapter->flags |= IGB_FLAG_MAS_ENABLE;
2169 netdev_info(adapter->netdev,
2170 "MAS: Enabling Media Autosense for port %d\n",
2175 if (eeprom_data & IGB_MAS_ENABLE_2) {
2176 adapter->flags |= IGB_FLAG_MAS_ENABLE;
2177 netdev_info(adapter->netdev,
2178 "MAS: Enabling Media Autosense for port %d\n",
2183 if (eeprom_data & IGB_MAS_ENABLE_3) {
2184 adapter->flags |= IGB_FLAG_MAS_ENABLE;
2185 netdev_info(adapter->netdev,
2186 "MAS: Enabling Media Autosense for port %d\n",
2191 /* Shouldn't get here */
2192 netdev_err(adapter->netdev,
2193 "MAS: Invalid port configuration, returning\n");
2199 * igb_init_i2c - Init I2C interface
2200 * @adapter: pointer to adapter structure
2202 static s32 igb_init_i2c(struct igb_adapter *adapter)
2206 /* I2C interface supported on i350 devices */
2207 if (adapter->hw.mac.type != e1000_i350)
2210 /* Initialize the i2c bus which is controlled by the registers.
2211 * This bus will use the i2c_algo_bit structue that implements
2212 * the protocol through toggling of the 4 bits in the register.
2214 adapter->i2c_adap.owner = THIS_MODULE;
2215 adapter->i2c_algo = igb_i2c_algo;
2216 adapter->i2c_algo.data = adapter;
2217 adapter->i2c_adap.algo_data = &adapter->i2c_algo;
2218 adapter->i2c_adap.dev.parent = &adapter->pdev->dev;
2219 strlcpy(adapter->i2c_adap.name, "igb BB",
2220 sizeof(adapter->i2c_adap.name));
2221 status = i2c_bit_add_bus(&adapter->i2c_adap);
2226 * igb_probe - Device Initialization Routine
2227 * @pdev: PCI device information struct
2228 * @ent: entry in igb_pci_tbl
2230 * Returns 0 on success, negative on failure
2232 * igb_probe initializes an adapter identified by a pci_dev structure.
2233 * The OS initialization, configuring of the adapter private structure,
2234 * and a hardware reset occur.
2236 static int igb_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2238 struct net_device *netdev;
2239 struct igb_adapter *adapter;
2240 struct e1000_hw *hw;
2241 u16 eeprom_data = 0;
2243 static int global_quad_port_a; /* global quad port a indication */
2244 const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
2245 int err, pci_using_dac;
2246 u8 part_str[E1000_PBANUM_LENGTH];
2248 /* Catch broken hardware that put the wrong VF device ID in
2249 * the PCIe SR-IOV capability.
2251 if (pdev->is_virtfn) {
2252 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
2253 pci_name(pdev), pdev->vendor, pdev->device);
2257 err = pci_enable_device_mem(pdev);
2262 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
2266 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
2269 "No usable DMA configuration, aborting\n");
2274 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
2280 pci_enable_pcie_error_reporting(pdev);
2282 pci_set_master(pdev);
2283 pci_save_state(pdev);
2286 netdev = alloc_etherdev_mq(sizeof(struct igb_adapter),
2289 goto err_alloc_etherdev;
2291 SET_NETDEV_DEV(netdev, &pdev->dev);
2293 pci_set_drvdata(pdev, netdev);
2294 adapter = netdev_priv(netdev);
2295 adapter->netdev = netdev;
2296 adapter->pdev = pdev;
2299 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
2302 hw->hw_addr = pci_iomap(pdev, 0, 0);
2306 netdev->netdev_ops = &igb_netdev_ops;
2307 igb_set_ethtool_ops(netdev);
2308 netdev->watchdog_timeo = 5 * HZ;
2310 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2312 netdev->mem_start = pci_resource_start(pdev, 0);
2313 netdev->mem_end = pci_resource_end(pdev, 0);
2315 /* PCI config space info */
2316 hw->vendor_id = pdev->vendor;
2317 hw->device_id = pdev->device;
2318 hw->revision_id = pdev->revision;
2319 hw->subsystem_vendor_id = pdev->subsystem_vendor;
2320 hw->subsystem_device_id = pdev->subsystem_device;
2322 /* Copy the default MAC, PHY and NVM function pointers */
2323 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
2324 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
2325 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
2326 /* Initialize skew-specific constants */
2327 err = ei->get_invariants(hw);
2331 /* setup the private structure */
2332 err = igb_sw_init(adapter);
2336 igb_get_bus_info_pcie(hw);
2338 hw->phy.autoneg_wait_to_complete = false;
2340 /* Copper options */
2341 if (hw->phy.media_type == e1000_media_type_copper) {
2342 hw->phy.mdix = AUTO_ALL_MODES;
2343 hw->phy.disable_polarity_correction = false;
2344 hw->phy.ms_type = e1000_ms_hw_default;
2347 if (igb_check_reset_block(hw))
2348 dev_info(&pdev->dev,
2349 "PHY reset is blocked due to SOL/IDER session.\n");
2351 /* features is initialized to 0 in allocation, it might have bits
2352 * set by igb_sw_init so we should use an or instead of an
2355 netdev->features |= NETIF_F_SG |
2362 NETIF_F_HW_VLAN_CTAG_RX |
2363 NETIF_F_HW_VLAN_CTAG_TX;
2365 /* copy netdev features into list of user selectable features */
2366 netdev->hw_features |= netdev->features;
2367 netdev->hw_features |= NETIF_F_RXALL;
2369 /* set this bit last since it cannot be part of hw_features */
2370 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
2372 netdev->vlan_features |= NETIF_F_TSO |
2378 netdev->priv_flags |= IFF_SUPP_NOFCS;
2380 if (pci_using_dac) {
2381 netdev->features |= NETIF_F_HIGHDMA;
2382 netdev->vlan_features |= NETIF_F_HIGHDMA;
2385 if (hw->mac.type >= e1000_82576) {
2386 netdev->hw_features |= NETIF_F_SCTP_CSUM;
2387 netdev->features |= NETIF_F_SCTP_CSUM;
2390 netdev->priv_flags |= IFF_UNICAST_FLT;
2392 adapter->en_mng_pt = igb_enable_mng_pass_thru(hw);
2394 /* before reading the NVM, reset the controller to put the device in a
2395 * known good starting state
2397 hw->mac.ops.reset_hw(hw);
2399 /* make sure the NVM is good , i211/i210 parts can have special NVM
2400 * that doesn't contain a checksum
2402 switch (hw->mac.type) {
2405 if (igb_get_flash_presence_i210(hw)) {
2406 if (hw->nvm.ops.validate(hw) < 0) {
2408 "The NVM Checksum Is Not Valid\n");
2415 if (hw->nvm.ops.validate(hw) < 0) {
2416 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
2423 /* copy the MAC address out of the NVM */
2424 if (hw->mac.ops.read_mac_addr(hw))
2425 dev_err(&pdev->dev, "NVM Read Error\n");
2427 memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
2429 if (!is_valid_ether_addr(netdev->dev_addr)) {
2430 dev_err(&pdev->dev, "Invalid MAC Address\n");
2435 /* get firmware version for ethtool -i */
2436 igb_set_fw_version(adapter);
2438 /* configure RXPBSIZE and TXPBSIZE */
2439 if (hw->mac.type == e1000_i210) {
2440 wr32(E1000_RXPBS, I210_RXPBSIZE_DEFAULT);
2441 wr32(E1000_TXPBS, I210_TXPBSIZE_DEFAULT);
2444 setup_timer(&adapter->watchdog_timer, igb_watchdog,
2445 (unsigned long) adapter);
2446 setup_timer(&adapter->phy_info_timer, igb_update_phy_info,
2447 (unsigned long) adapter);
2449 INIT_WORK(&adapter->reset_task, igb_reset_task);
2450 INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
2452 /* Initialize link properties that are user-changeable */
2453 adapter->fc_autoneg = true;
2454 hw->mac.autoneg = true;
2455 hw->phy.autoneg_advertised = 0x2f;
2457 hw->fc.requested_mode = e1000_fc_default;
2458 hw->fc.current_mode = e1000_fc_default;
2460 igb_validate_mdi_setting(hw);
2462 /* By default, support wake on port A */
2463 if (hw->bus.func == 0)
2464 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2466 /* Check the NVM for wake support on non-port A ports */
2467 if (hw->mac.type >= e1000_82580)
2468 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A +
2469 NVM_82580_LAN_FUNC_OFFSET(hw->bus.func), 1,
2471 else if (hw->bus.func == 1)
2472 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
2474 if (eeprom_data & IGB_EEPROM_APME)
2475 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2477 /* now that we have the eeprom settings, apply the special cases where
2478 * the eeprom may be wrong or the board simply won't support wake on
2479 * lan on a particular port
2481 switch (pdev->device) {
2482 case E1000_DEV_ID_82575GB_QUAD_COPPER:
2483 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2485 case E1000_DEV_ID_82575EB_FIBER_SERDES:
2486 case E1000_DEV_ID_82576_FIBER:
2487 case E1000_DEV_ID_82576_SERDES:
2488 /* Wake events only supported on port A for dual fiber
2489 * regardless of eeprom setting
2491 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
2492 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2494 case E1000_DEV_ID_82576_QUAD_COPPER:
2495 case E1000_DEV_ID_82576_QUAD_COPPER_ET2:
2496 /* if quad port adapter, disable WoL on all but port A */
2497 if (global_quad_port_a != 0)
2498 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2500 adapter->flags |= IGB_FLAG_QUAD_PORT_A;
2501 /* Reset for multiple quad port adapters */
2502 if (++global_quad_port_a == 4)
2503 global_quad_port_a = 0;
2506 /* If the device can't wake, don't set software support */
2507 if (!device_can_wakeup(&adapter->pdev->dev))
2508 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2511 /* initialize the wol settings based on the eeprom settings */
2512 if (adapter->flags & IGB_FLAG_WOL_SUPPORTED)
2513 adapter->wol |= E1000_WUFC_MAG;
2515 /* Some vendors want WoL disabled by default, but still supported */
2516 if ((hw->mac.type == e1000_i350) &&
2517 (pdev->subsystem_vendor == PCI_VENDOR_ID_HP)) {
2518 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2522 device_set_wakeup_enable(&adapter->pdev->dev,
2523 adapter->flags & IGB_FLAG_WOL_SUPPORTED);
2525 /* reset the hardware with the new settings */
2528 /* Init the I2C interface */
2529 err = igb_init_i2c(adapter);
2531 dev_err(&pdev->dev, "failed to init i2c interface\n");
2535 /* let the f/w know that the h/w is now under the control of the
2538 igb_get_hw_control(adapter);
2540 strcpy(netdev->name, "eth%d");
2541 err = register_netdev(netdev);
2545 /* carrier off reporting is important to ethtool even BEFORE open */
2546 netif_carrier_off(netdev);
2548 #ifdef CONFIG_IGB_DCA
2549 if (dca_add_requester(&pdev->dev) == 0) {
2550 adapter->flags |= IGB_FLAG_DCA_ENABLED;
2551 dev_info(&pdev->dev, "DCA enabled\n");
2552 igb_setup_dca(adapter);
2556 #ifdef CONFIG_IGB_HWMON
2557 /* Initialize the thermal sensor on i350 devices. */
2558 if (hw->mac.type == e1000_i350 && hw->bus.func == 0) {
2561 /* Read the NVM to determine if this i350 device supports an
2562 * external thermal sensor.
2564 hw->nvm.ops.read(hw, NVM_ETS_CFG, 1, &ets_word);
2565 if (ets_word != 0x0000 && ets_word != 0xFFFF)
2566 adapter->ets = true;
2568 adapter->ets = false;
2569 if (igb_sysfs_init(adapter))
2571 "failed to allocate sysfs resources\n");
2573 adapter->ets = false;
2576 /* Check if Media Autosense is enabled */
2578 if (hw->dev_spec._82575.mas_capable)
2579 igb_init_mas(adapter);
2581 /* do hw tstamp init after resetting */
2582 igb_ptp_init(adapter);
2584 dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
2585 /* print bus type/speed/width info, not applicable to i354 */
2586 if (hw->mac.type != e1000_i354) {
2587 dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
2589 ((hw->bus.speed == e1000_bus_speed_2500) ? "2.5Gb/s" :
2590 (hw->bus.speed == e1000_bus_speed_5000) ? "5.0Gb/s" :
2592 ((hw->bus.width == e1000_bus_width_pcie_x4) ?
2594 (hw->bus.width == e1000_bus_width_pcie_x2) ?
2596 (hw->bus.width == e1000_bus_width_pcie_x1) ?
2597 "Width x1" : "unknown"), netdev->dev_addr);
2600 if ((hw->mac.type >= e1000_i210 ||
2601 igb_get_flash_presence_i210(hw))) {
2602 ret_val = igb_read_part_string(hw, part_str,
2603 E1000_PBANUM_LENGTH);
2605 ret_val = -E1000_ERR_INVM_VALUE_NOT_FOUND;
2609 strcpy(part_str, "Unknown");
2610 dev_info(&pdev->dev, "%s: PBA No: %s\n", netdev->name, part_str);
2611 dev_info(&pdev->dev,
2612 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
2613 (adapter->flags & IGB_FLAG_HAS_MSIX) ? "MSI-X" :
2614 (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
2615 adapter->num_rx_queues, adapter->num_tx_queues);
2616 if (hw->phy.media_type == e1000_media_type_copper) {
2617 switch (hw->mac.type) {
2621 /* Enable EEE for internal copper PHY devices */
2622 err = igb_set_eee_i350(hw, true, true);
2624 (!hw->dev_spec._82575.eee_disable)) {
2625 adapter->eee_advert =
2626 MDIO_EEE_100TX | MDIO_EEE_1000T;
2627 adapter->flags |= IGB_FLAG_EEE;
2631 if ((rd32(E1000_CTRL_EXT) &
2632 E1000_CTRL_EXT_LINK_MODE_SGMII)) {
2633 err = igb_set_eee_i354(hw, true, true);
2635 (!hw->dev_spec._82575.eee_disable)) {
2636 adapter->eee_advert =
2637 MDIO_EEE_100TX | MDIO_EEE_1000T;
2638 adapter->flags |= IGB_FLAG_EEE;
2646 pm_runtime_put_noidle(&pdev->dev);
2650 igb_release_hw_control(adapter);
2651 memset(&adapter->i2c_adap, 0, sizeof(adapter->i2c_adap));
2653 if (!igb_check_reset_block(hw))
2656 if (hw->flash_address)
2657 iounmap(hw->flash_address);
2659 igb_clear_interrupt_scheme(adapter);
2660 pci_iounmap(pdev, hw->hw_addr);
2662 free_netdev(netdev);
2664 pci_release_selected_regions(pdev,
2665 pci_select_bars(pdev, IORESOURCE_MEM));
2668 pci_disable_device(pdev);
2672 #ifdef CONFIG_PCI_IOV
2673 static int igb_disable_sriov(struct pci_dev *pdev)
2675 struct net_device *netdev = pci_get_drvdata(pdev);
2676 struct igb_adapter *adapter = netdev_priv(netdev);
2677 struct e1000_hw *hw = &adapter->hw;
2679 /* reclaim resources allocated to VFs */
2680 if (adapter->vf_data) {
2681 /* disable iov and allow time for transactions to clear */
2682 if (pci_vfs_assigned(pdev)) {
2683 dev_warn(&pdev->dev,
2684 "Cannot deallocate SR-IOV virtual functions while they are assigned - VFs will not be deallocated\n");
2687 pci_disable_sriov(pdev);
2691 kfree(adapter->vf_data);
2692 adapter->vf_data = NULL;
2693 adapter->vfs_allocated_count = 0;
2694 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
2697 dev_info(&pdev->dev, "IOV Disabled\n");
2699 /* Re-enable DMA Coalescing flag since IOV is turned off */
2700 adapter->flags |= IGB_FLAG_DMAC;
2706 static int igb_enable_sriov(struct pci_dev *pdev, int num_vfs)
2708 struct net_device *netdev = pci_get_drvdata(pdev);
2709 struct igb_adapter *adapter = netdev_priv(netdev);
2710 int old_vfs = pci_num_vf(pdev);
2714 if (!(adapter->flags & IGB_FLAG_HAS_MSIX) || num_vfs > 7) {
2722 dev_info(&pdev->dev, "%d pre-allocated VFs found - override max_vfs setting of %d\n",
2724 adapter->vfs_allocated_count = old_vfs;
2726 adapter->vfs_allocated_count = num_vfs;
2728 adapter->vf_data = kcalloc(adapter->vfs_allocated_count,
2729 sizeof(struct vf_data_storage), GFP_KERNEL);
2731 /* if allocation failed then we do not support SR-IOV */
2732 if (!adapter->vf_data) {
2733 adapter->vfs_allocated_count = 0;
2735 "Unable to allocate memory for VF Data Storage\n");
2740 /* only call pci_enable_sriov() if no VFs are allocated already */
2742 err = pci_enable_sriov(pdev, adapter->vfs_allocated_count);
2746 dev_info(&pdev->dev, "%d VFs allocated\n",
2747 adapter->vfs_allocated_count);
2748 for (i = 0; i < adapter->vfs_allocated_count; i++)
2749 igb_vf_configure(adapter, i);
2751 /* DMA Coalescing is not supported in IOV mode. */
2752 adapter->flags &= ~IGB_FLAG_DMAC;
2756 kfree(adapter->vf_data);
2757 adapter->vf_data = NULL;
2758 adapter->vfs_allocated_count = 0;
2765 * igb_remove_i2c - Cleanup I2C interface
2766 * @adapter: pointer to adapter structure
2768 static void igb_remove_i2c(struct igb_adapter *adapter)
2770 /* free the adapter bus structure */
2771 i2c_del_adapter(&adapter->i2c_adap);
2775 * igb_remove - Device Removal Routine
2776 * @pdev: PCI device information struct
2778 * igb_remove is called by the PCI subsystem to alert the driver
2779 * that it should release a PCI device. The could be caused by a
2780 * Hot-Plug event, or because the driver is going to be removed from
2783 static void igb_remove(struct pci_dev *pdev)
2785 struct net_device *netdev = pci_get_drvdata(pdev);
2786 struct igb_adapter *adapter = netdev_priv(netdev);
2787 struct e1000_hw *hw = &adapter->hw;
2789 pm_runtime_get_noresume(&pdev->dev);
2790 #ifdef CONFIG_IGB_HWMON
2791 igb_sysfs_exit(adapter);
2793 igb_remove_i2c(adapter);
2794 igb_ptp_stop(adapter);
2795 /* The watchdog timer may be rescheduled, so explicitly
2796 * disable watchdog from being rescheduled.
2798 set_bit(__IGB_DOWN, &adapter->state);
2799 del_timer_sync(&adapter->watchdog_timer);
2800 del_timer_sync(&adapter->phy_info_timer);
2802 cancel_work_sync(&adapter->reset_task);
2803 cancel_work_sync(&adapter->watchdog_task);
2805 #ifdef CONFIG_IGB_DCA
2806 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
2807 dev_info(&pdev->dev, "DCA disabled\n");
2808 dca_remove_requester(&pdev->dev);
2809 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
2810 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
2814 /* Release control of h/w to f/w. If f/w is AMT enabled, this
2815 * would have already happened in close and is redundant.
2817 igb_release_hw_control(adapter);
2819 unregister_netdev(netdev);
2821 igb_clear_interrupt_scheme(adapter);
2823 #ifdef CONFIG_PCI_IOV
2824 igb_disable_sriov(pdev);
2827 pci_iounmap(pdev, hw->hw_addr);
2828 if (hw->flash_address)
2829 iounmap(hw->flash_address);
2830 pci_release_selected_regions(pdev,
2831 pci_select_bars(pdev, IORESOURCE_MEM));
2833 kfree(adapter->shadow_vfta);
2834 free_netdev(netdev);
2836 pci_disable_pcie_error_reporting(pdev);
2838 pci_disable_device(pdev);
2842 * igb_probe_vfs - Initialize vf data storage and add VFs to pci config space
2843 * @adapter: board private structure to initialize
2845 * This function initializes the vf specific data storage and then attempts to
2846 * allocate the VFs. The reason for ordering it this way is because it is much
2847 * mor expensive time wise to disable SR-IOV than it is to allocate and free
2848 * the memory for the VFs.
2850 static void igb_probe_vfs(struct igb_adapter *adapter)
2852 #ifdef CONFIG_PCI_IOV
2853 struct pci_dev *pdev = adapter->pdev;
2854 struct e1000_hw *hw = &adapter->hw;
2856 /* Virtualization features not supported on i210 family. */
2857 if ((hw->mac.type == e1000_i210) || (hw->mac.type == e1000_i211))
2860 pci_sriov_set_totalvfs(pdev, 7);
2861 igb_pci_enable_sriov(pdev, max_vfs);
2863 #endif /* CONFIG_PCI_IOV */
2866 static void igb_init_queue_configuration(struct igb_adapter *adapter)
2868 struct e1000_hw *hw = &adapter->hw;
2871 /* Determine the maximum number of RSS queues supported. */
2872 switch (hw->mac.type) {
2874 max_rss_queues = IGB_MAX_RX_QUEUES_I211;
2878 max_rss_queues = IGB_MAX_RX_QUEUES_82575;
2881 /* I350 cannot do RSS and SR-IOV at the same time */
2882 if (!!adapter->vfs_allocated_count) {
2888 if (!!adapter->vfs_allocated_count) {
2896 max_rss_queues = IGB_MAX_RX_QUEUES;
2900 adapter->rss_queues = min_t(u32, max_rss_queues, num_online_cpus());
2902 /* Determine if we need to pair queues. */
2903 switch (hw->mac.type) {
2906 /* Device supports enough interrupts without queue pairing. */
2909 /* If VFs are going to be allocated with RSS queues then we
2910 * should pair the queues in order to conserve interrupts due
2911 * to limited supply.
2913 if ((adapter->rss_queues > 1) &&
2914 (adapter->vfs_allocated_count > 6))
2915 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
2922 /* If rss_queues > half of max_rss_queues, pair the queues in
2923 * order to conserve interrupts due to limited supply.
2925 if (adapter->rss_queues > (max_rss_queues / 2))
2926 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
2932 * igb_sw_init - Initialize general software structures (struct igb_adapter)
2933 * @adapter: board private structure to initialize
2935 * igb_sw_init initializes the Adapter private data structure.
2936 * Fields are initialized based on PCI device information and
2937 * OS network device settings (MTU size).
2939 static int igb_sw_init(struct igb_adapter *adapter)
2941 struct e1000_hw *hw = &adapter->hw;
2942 struct net_device *netdev = adapter->netdev;
2943 struct pci_dev *pdev = adapter->pdev;
2945 pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
2947 /* set default ring sizes */
2948 adapter->tx_ring_count = IGB_DEFAULT_TXD;
2949 adapter->rx_ring_count = IGB_DEFAULT_RXD;
2951 /* set default ITR values */
2952 adapter->rx_itr_setting = IGB_DEFAULT_ITR;
2953 adapter->tx_itr_setting = IGB_DEFAULT_ITR;
2955 /* set default work limits */
2956 adapter->tx_work_limit = IGB_DEFAULT_TX_WORK;
2958 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN +
2960 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2962 spin_lock_init(&adapter->stats64_lock);
2963 #ifdef CONFIG_PCI_IOV
2964 switch (hw->mac.type) {
2968 dev_warn(&pdev->dev,
2969 "Maximum of 7 VFs per PF, using max\n");
2970 max_vfs = adapter->vfs_allocated_count = 7;
2972 adapter->vfs_allocated_count = max_vfs;
2973 if (adapter->vfs_allocated_count)
2974 dev_warn(&pdev->dev,
2975 "Enabling SR-IOV VFs using the module parameter is deprecated - please use the pci sysfs interface.\n");
2980 #endif /* CONFIG_PCI_IOV */
2982 igb_init_queue_configuration(adapter);
2984 /* Setup and initialize a copy of the hw vlan table array */
2985 adapter->shadow_vfta = kcalloc(E1000_VLAN_FILTER_TBL_SIZE, sizeof(u32),
2988 /* This call may decrease the number of queues */
2989 if (igb_init_interrupt_scheme(adapter, true)) {
2990 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
2994 igb_probe_vfs(adapter);
2996 /* Explicitly disable IRQ since the NIC can be in any state. */
2997 igb_irq_disable(adapter);
2999 if (hw->mac.type >= e1000_i350)
3000 adapter->flags &= ~IGB_FLAG_DMAC;
3002 set_bit(__IGB_DOWN, &adapter->state);
3007 * igb_open - Called when a network interface is made active
3008 * @netdev: network interface device structure
3010 * Returns 0 on success, negative value on failure
3012 * The open entry point is called when a network interface is made
3013 * active by the system (IFF_UP). At this point all resources needed
3014 * for transmit and receive operations are allocated, the interrupt
3015 * handler is registered with the OS, the watchdog timer is started,
3016 * and the stack is notified that the interface is ready.
3018 static int __igb_open(struct net_device *netdev, bool resuming)
3020 struct igb_adapter *adapter = netdev_priv(netdev);
3021 struct e1000_hw *hw = &adapter->hw;
3022 struct pci_dev *pdev = adapter->pdev;
3026 /* disallow open during test */
3027 if (test_bit(__IGB_TESTING, &adapter->state)) {
3033 pm_runtime_get_sync(&pdev->dev);
3035 netif_carrier_off(netdev);
3037 /* allocate transmit descriptors */
3038 err = igb_setup_all_tx_resources(adapter);
3042 /* allocate receive descriptors */
3043 err = igb_setup_all_rx_resources(adapter);
3047 igb_power_up_link(adapter);
3049 /* before we allocate an interrupt, we must be ready to handle it.
3050 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3051 * as soon as we call pci_request_irq, so we have to setup our
3052 * clean_rx handler before we do so.
3054 igb_configure(adapter);
3056 err = igb_request_irq(adapter);
3060 /* Notify the stack of the actual queue counts. */
3061 err = netif_set_real_num_tx_queues(adapter->netdev,
3062 adapter->num_tx_queues);
3064 goto err_set_queues;
3066 err = netif_set_real_num_rx_queues(adapter->netdev,
3067 adapter->num_rx_queues);
3069 goto err_set_queues;
3071 /* From here on the code is the same as igb_up() */
3072 clear_bit(__IGB_DOWN, &adapter->state);
3074 for (i = 0; i < adapter->num_q_vectors; i++)
3075 napi_enable(&(adapter->q_vector[i]->napi));
3077 /* Clear any pending interrupts. */
3080 igb_irq_enable(adapter);
3082 /* notify VFs that reset has been completed */
3083 if (adapter->vfs_allocated_count) {
3084 u32 reg_data = rd32(E1000_CTRL_EXT);
3086 reg_data |= E1000_CTRL_EXT_PFRSTD;
3087 wr32(E1000_CTRL_EXT, reg_data);
3090 netif_tx_start_all_queues(netdev);
3093 pm_runtime_put(&pdev->dev);
3095 /* start the watchdog. */
3096 hw->mac.get_link_status = 1;
3097 schedule_work(&adapter->watchdog_task);
3102 igb_free_irq(adapter);
3104 igb_release_hw_control(adapter);
3105 igb_power_down_link(adapter);
3106 igb_free_all_rx_resources(adapter);
3108 igb_free_all_tx_resources(adapter);
3112 pm_runtime_put(&pdev->dev);
3117 static int igb_open(struct net_device *netdev)
3119 return __igb_open(netdev, false);
3123 * igb_close - Disables a network interface
3124 * @netdev: network interface device structure
3126 * Returns 0, this is not allowed to fail
3128 * The close entry point is called when an interface is de-activated
3129 * by the OS. The hardware is still under the driver's control, but
3130 * needs to be disabled. A global MAC reset is issued to stop the
3131 * hardware, and all transmit and receive resources are freed.
3133 static int __igb_close(struct net_device *netdev, bool suspending)
3135 struct igb_adapter *adapter = netdev_priv(netdev);
3136 struct pci_dev *pdev = adapter->pdev;
3138 WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
3141 pm_runtime_get_sync(&pdev->dev);
3144 igb_free_irq(adapter);
3146 igb_free_all_tx_resources(adapter);
3147 igb_free_all_rx_resources(adapter);
3150 pm_runtime_put_sync(&pdev->dev);
3154 static int igb_close(struct net_device *netdev)
3156 return __igb_close(netdev, false);
3160 * igb_setup_tx_resources - allocate Tx resources (Descriptors)
3161 * @tx_ring: tx descriptor ring (for a specific queue) to setup
3163 * Return 0 on success, negative on failure
3165 int igb_setup_tx_resources(struct igb_ring *tx_ring)
3167 struct device *dev = tx_ring->dev;
3170 size = sizeof(struct igb_tx_buffer) * tx_ring->count;
3172 tx_ring->tx_buffer_info = vzalloc(size);
3173 if (!tx_ring->tx_buffer_info)
3176 /* round up to nearest 4K */
3177 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
3178 tx_ring->size = ALIGN(tx_ring->size, 4096);
3180 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
3181 &tx_ring->dma, GFP_KERNEL);
3185 tx_ring->next_to_use = 0;
3186 tx_ring->next_to_clean = 0;
3191 vfree(tx_ring->tx_buffer_info);
3192 tx_ring->tx_buffer_info = NULL;
3193 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
3198 * igb_setup_all_tx_resources - wrapper to allocate Tx resources
3199 * (Descriptors) for all queues
3200 * @adapter: board private structure
3202 * Return 0 on success, negative on failure
3204 static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
3206 struct pci_dev *pdev = adapter->pdev;
3209 for (i = 0; i < adapter->num_tx_queues; i++) {
3210 err = igb_setup_tx_resources(adapter->tx_ring[i]);
3213 "Allocation for Tx Queue %u failed\n", i);
3214 for (i--; i >= 0; i--)
3215 igb_free_tx_resources(adapter->tx_ring[i]);
3224 * igb_setup_tctl - configure the transmit control registers
3225 * @adapter: Board private structure
3227 void igb_setup_tctl(struct igb_adapter *adapter)
3229 struct e1000_hw *hw = &adapter->hw;
3232 /* disable queue 0 which is enabled by default on 82575 and 82576 */
3233 wr32(E1000_TXDCTL(0), 0);
3235 /* Program the Transmit Control Register */
3236 tctl = rd32(E1000_TCTL);
3237 tctl &= ~E1000_TCTL_CT;
3238 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
3239 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
3241 igb_config_collision_dist(hw);
3243 /* Enable transmits */
3244 tctl |= E1000_TCTL_EN;
3246 wr32(E1000_TCTL, tctl);
3250 * igb_configure_tx_ring - Configure transmit ring after Reset
3251 * @adapter: board private structure
3252 * @ring: tx ring to configure
3254 * Configure a transmit ring after a reset.
3256 void igb_configure_tx_ring(struct igb_adapter *adapter,
3257 struct igb_ring *ring)
3259 struct e1000_hw *hw = &adapter->hw;
3261 u64 tdba = ring->dma;
3262 int reg_idx = ring->reg_idx;
3264 /* disable the queue */
3265 wr32(E1000_TXDCTL(reg_idx), 0);
3269 wr32(E1000_TDLEN(reg_idx),
3270 ring->count * sizeof(union e1000_adv_tx_desc));
3271 wr32(E1000_TDBAL(reg_idx),
3272 tdba & 0x00000000ffffffffULL);
3273 wr32(E1000_TDBAH(reg_idx), tdba >> 32);
3275 ring->tail = hw->hw_addr + E1000_TDT(reg_idx);
3276 wr32(E1000_TDH(reg_idx), 0);
3277 writel(0, ring->tail);
3279 txdctl |= IGB_TX_PTHRESH;
3280 txdctl |= IGB_TX_HTHRESH << 8;
3281 txdctl |= IGB_TX_WTHRESH << 16;
3283 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
3284 wr32(E1000_TXDCTL(reg_idx), txdctl);
3288 * igb_configure_tx - Configure transmit Unit after Reset
3289 * @adapter: board private structure
3291 * Configure the Tx unit of the MAC after a reset.
3293 static void igb_configure_tx(struct igb_adapter *adapter)
3297 for (i = 0; i < adapter->num_tx_queues; i++)
3298 igb_configure_tx_ring(adapter, adapter->tx_ring[i]);
3302 * igb_setup_rx_resources - allocate Rx resources (Descriptors)
3303 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
3305 * Returns 0 on success, negative on failure
3307 int igb_setup_rx_resources(struct igb_ring *rx_ring)
3309 struct device *dev = rx_ring->dev;
3312 size = sizeof(struct igb_rx_buffer) * rx_ring->count;
3314 rx_ring->rx_buffer_info = vzalloc(size);
3315 if (!rx_ring->rx_buffer_info)
3318 /* Round up to nearest 4K */
3319 rx_ring->size = rx_ring->count * sizeof(union e1000_adv_rx_desc);
3320 rx_ring->size = ALIGN(rx_ring->size, 4096);
3322 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
3323 &rx_ring->dma, GFP_KERNEL);
3327 rx_ring->next_to_alloc = 0;
3328 rx_ring->next_to_clean = 0;
3329 rx_ring->next_to_use = 0;
3334 vfree(rx_ring->rx_buffer_info);
3335 rx_ring->rx_buffer_info = NULL;
3336 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
3341 * igb_setup_all_rx_resources - wrapper to allocate Rx resources
3342 * (Descriptors) for all queues
3343 * @adapter: board private structure
3345 * Return 0 on success, negative on failure
3347 static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
3349 struct pci_dev *pdev = adapter->pdev;
3352 for (i = 0; i < adapter->num_rx_queues; i++) {
3353 err = igb_setup_rx_resources(adapter->rx_ring[i]);
3356 "Allocation for Rx Queue %u failed\n", i);
3357 for (i--; i >= 0; i--)
3358 igb_free_rx_resources(adapter->rx_ring[i]);
3367 * igb_setup_mrqc - configure the multiple receive queue control registers
3368 * @adapter: Board private structure
3370 static void igb_setup_mrqc(struct igb_adapter *adapter)
3372 struct e1000_hw *hw = &adapter->hw;
3374 u32 j, num_rx_queues;
3375 static const u32 rsskey[10] = { 0xDA565A6D, 0xC20E5B25, 0x3D256741,
3376 0xB08FA343, 0xCB2BCAD0, 0xB4307BAE,
3377 0xA32DCB77, 0x0CF23080, 0x3BB7426A,
3380 /* Fill out hash function seeds */
3381 for (j = 0; j < 10; j++)
3382 wr32(E1000_RSSRK(j), rsskey[j]);
3384 num_rx_queues = adapter->rss_queues;
3386 switch (hw->mac.type) {
3388 /* 82576 supports 2 RSS queues for SR-IOV */
3389 if (adapter->vfs_allocated_count)
3396 if (adapter->rss_indir_tbl_init != num_rx_queues) {
3397 for (j = 0; j < IGB_RETA_SIZE; j++)
3398 adapter->rss_indir_tbl[j] =
3399 (j * num_rx_queues) / IGB_RETA_SIZE;
3400 adapter->rss_indir_tbl_init = num_rx_queues;
3402 igb_write_rss_indir_tbl(adapter);
3404 /* Disable raw packet checksumming so that RSS hash is placed in
3405 * descriptor on writeback. No need to enable TCP/UDP/IP checksum
3406 * offloads as they are enabled by default
3408 rxcsum = rd32(E1000_RXCSUM);
3409 rxcsum |= E1000_RXCSUM_PCSD;
3411 if (adapter->hw.mac.type >= e1000_82576)
3412 /* Enable Receive Checksum Offload for SCTP */
3413 rxcsum |= E1000_RXCSUM_CRCOFL;
3415 /* Don't need to set TUOFL or IPOFL, they default to 1 */
3416 wr32(E1000_RXCSUM, rxcsum);
3418 /* Generate RSS hash based on packet types, TCP/UDP
3419 * port numbers and/or IPv4/v6 src and dst addresses
3421 mrqc = E1000_MRQC_RSS_FIELD_IPV4 |
3422 E1000_MRQC_RSS_FIELD_IPV4_TCP |
3423 E1000_MRQC_RSS_FIELD_IPV6 |
3424 E1000_MRQC_RSS_FIELD_IPV6_TCP |
3425 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX;
3427 if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV4_UDP)
3428 mrqc |= E1000_MRQC_RSS_FIELD_IPV4_UDP;
3429 if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV6_UDP)
3430 mrqc |= E1000_MRQC_RSS_FIELD_IPV6_UDP;
3432 /* If VMDq is enabled then we set the appropriate mode for that, else
3433 * we default to RSS so that an RSS hash is calculated per packet even
3434 * if we are only using one queue
3436 if (adapter->vfs_allocated_count) {
3437 if (hw->mac.type > e1000_82575) {
3438 /* Set the default pool for the PF's first queue */
3439 u32 vtctl = rd32(E1000_VT_CTL);
3441 vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK |
3442 E1000_VT_CTL_DISABLE_DEF_POOL);
3443 vtctl |= adapter->vfs_allocated_count <<
3444 E1000_VT_CTL_DEFAULT_POOL_SHIFT;
3445 wr32(E1000_VT_CTL, vtctl);
3447 if (adapter->rss_queues > 1)
3448 mrqc |= E1000_MRQC_ENABLE_VMDQ_RSS_2Q;
3450 mrqc |= E1000_MRQC_ENABLE_VMDQ;
3452 if (hw->mac.type != e1000_i211)
3453 mrqc |= E1000_MRQC_ENABLE_RSS_4Q;
3455 igb_vmm_control(adapter);
3457 wr32(E1000_MRQC, mrqc);
3461 * igb_setup_rctl - configure the receive control registers
3462 * @adapter: Board private structure
3464 void igb_setup_rctl(struct igb_adapter *adapter)
3466 struct e1000_hw *hw = &adapter->hw;
3469 rctl = rd32(E1000_RCTL);
3471 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
3472 rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
3474 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
3475 (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
3477 /* enable stripping of CRC. It's unlikely this will break BMC
3478 * redirection as it did with e1000. Newer features require
3479 * that the HW strips the CRC.
3481 rctl |= E1000_RCTL_SECRC;
3483 /* disable store bad packets and clear size bits. */
3484 rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
3486 /* enable LPE to prevent packets larger than max_frame_size */
3487 rctl |= E1000_RCTL_LPE;
3489 /* disable queue 0 to prevent tail write w/o re-config */
3490 wr32(E1000_RXDCTL(0), 0);
3492 /* Attention!!! For SR-IOV PF driver operations you must enable
3493 * queue drop for all VF and PF queues to prevent head of line blocking
3494 * if an un-trusted VF does not provide descriptors to hardware.
3496 if (adapter->vfs_allocated_count) {
3497 /* set all queue drop enable bits */
3498 wr32(E1000_QDE, ALL_QUEUES);
3501 /* This is useful for sniffing bad packets. */
3502 if (adapter->netdev->features & NETIF_F_RXALL) {
3503 /* UPE and MPE will be handled by normal PROMISC logic
3504 * in e1000e_set_rx_mode
3506 rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
3507 E1000_RCTL_BAM | /* RX All Bcast Pkts */
3508 E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
3510 rctl &= ~(E1000_RCTL_VFE | /* Disable VLAN filter */
3511 E1000_RCTL_DPF | /* Allow filtered pause */
3512 E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
3513 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
3514 * and that breaks VLANs.
3518 wr32(E1000_RCTL, rctl);
3521 static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
3524 struct e1000_hw *hw = &adapter->hw;
3527 /* if it isn't the PF check to see if VFs are enabled and
3528 * increase the size to support vlan tags
3530 if (vfn < adapter->vfs_allocated_count &&
3531 adapter->vf_data[vfn].vlans_enabled)
3532 size += VLAN_TAG_SIZE;
3534 vmolr = rd32(E1000_VMOLR(vfn));
3535 vmolr &= ~E1000_VMOLR_RLPML_MASK;
3536 vmolr |= size | E1000_VMOLR_LPE;
3537 wr32(E1000_VMOLR(vfn), vmolr);
3543 * igb_rlpml_set - set maximum receive packet size
3544 * @adapter: board private structure
3546 * Configure maximum receivable packet size.
3548 static void igb_rlpml_set(struct igb_adapter *adapter)
3550 u32 max_frame_size = adapter->max_frame_size;
3551 struct e1000_hw *hw = &adapter->hw;
3552 u16 pf_id = adapter->vfs_allocated_count;
3555 igb_set_vf_rlpml(adapter, max_frame_size, pf_id);
3556 /* If we're in VMDQ or SR-IOV mode, then set global RLPML
3557 * to our max jumbo frame size, in case we need to enable
3558 * jumbo frames on one of the rings later.
3559 * This will not pass over-length frames into the default
3560 * queue because it's gated by the VMOLR.RLPML.
3562 max_frame_size = MAX_JUMBO_FRAME_SIZE;
3565 wr32(E1000_RLPML, max_frame_size);
3568 static inline void igb_set_vmolr(struct igb_adapter *adapter,
3571 struct e1000_hw *hw = &adapter->hw;
3574 /* This register exists only on 82576 and newer so if we are older then
3575 * we should exit and do nothing
3577 if (hw->mac.type < e1000_82576)
3580 vmolr = rd32(E1000_VMOLR(vfn));
3581 vmolr |= E1000_VMOLR_STRVLAN; /* Strip vlan tags */
3582 if (hw->mac.type == e1000_i350) {
3585 dvmolr = rd32(E1000_DVMOLR(vfn));
3586 dvmolr |= E1000_DVMOLR_STRVLAN;
3587 wr32(E1000_DVMOLR(vfn), dvmolr);
3590 vmolr |= E1000_VMOLR_AUPE; /* Accept untagged packets */
3592 vmolr &= ~(E1000_VMOLR_AUPE); /* Tagged packets ONLY */
3594 /* clear all bits that might not be set */
3595 vmolr &= ~(E1000_VMOLR_BAM | E1000_VMOLR_RSSE);
3597 if (adapter->rss_queues > 1 && vfn == adapter->vfs_allocated_count)
3598 vmolr |= E1000_VMOLR_RSSE; /* enable RSS */
3599 /* for VMDq only allow the VFs and pool 0 to accept broadcast and
3602 if (vfn <= adapter->vfs_allocated_count)
3603 vmolr |= E1000_VMOLR_BAM; /* Accept broadcast */
3605 wr32(E1000_VMOLR(vfn), vmolr);
3609 * igb_configure_rx_ring - Configure a receive ring after Reset
3610 * @adapter: board private structure
3611 * @ring: receive ring to be configured
3613 * Configure the Rx unit of the MAC after a reset.
3615 void igb_configure_rx_ring(struct igb_adapter *adapter,
3616 struct igb_ring *ring)
3618 struct e1000_hw *hw = &adapter->hw;
3619 u64 rdba = ring->dma;
3620 int reg_idx = ring->reg_idx;
3621 u32 srrctl = 0, rxdctl = 0;
3623 /* disable the queue */
3624 wr32(E1000_RXDCTL(reg_idx), 0);
3626 /* Set DMA base address registers */
3627 wr32(E1000_RDBAL(reg_idx),
3628 rdba & 0x00000000ffffffffULL);
3629 wr32(E1000_RDBAH(reg_idx), rdba >> 32);
3630 wr32(E1000_RDLEN(reg_idx),
3631 ring->count * sizeof(union e1000_adv_rx_desc));
3633 /* initialize head and tail */
3634 ring->tail = hw->hw_addr + E1000_RDT(reg_idx);
3635 wr32(E1000_RDH(reg_idx), 0);
3636 writel(0, ring->tail);
3638 /* set descriptor configuration */
3639 srrctl = IGB_RX_HDR_LEN << E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
3640 srrctl |= IGB_RX_BUFSZ >> E1000_SRRCTL_BSIZEPKT_SHIFT;
3641 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
3642 if (hw->mac.type >= e1000_82580)
3643 srrctl |= E1000_SRRCTL_TIMESTAMP;
3644 /* Only set Drop Enable if we are supporting multiple queues */
3645 if (adapter->vfs_allocated_count || adapter->num_rx_queues > 1)
3646 srrctl |= E1000_SRRCTL_DROP_EN;
3648 wr32(E1000_SRRCTL(reg_idx), srrctl);
3650 /* set filtering for VMDQ pools */
3651 igb_set_vmolr(adapter, reg_idx & 0x7, true);
3653 rxdctl |= IGB_RX_PTHRESH;
3654 rxdctl |= IGB_RX_HTHRESH << 8;
3655 rxdctl |= IGB_RX_WTHRESH << 16;
3657 /* enable receive descriptor fetching */
3658 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
3659 wr32(E1000_RXDCTL(reg_idx), rxdctl);
3663 * igb_configure_rx - Configure receive Unit after Reset
3664 * @adapter: board private structure
3666 * Configure the Rx unit of the MAC after a reset.
3668 static void igb_configure_rx(struct igb_adapter *adapter)
3672 /* set UTA to appropriate mode */
3673 igb_set_uta(adapter);
3675 /* set the correct pool for the PF default MAC address in entry 0 */
3676 igb_rar_set_qsel(adapter, adapter->hw.mac.addr, 0,
3677 adapter->vfs_allocated_count);
3679 /* Setup the HW Rx Head and Tail Descriptor Pointers and
3680 * the Base and Length of the Rx Descriptor Ring
3682 for (i = 0; i < adapter->num_rx_queues; i++)
3683 igb_configure_rx_ring(adapter, adapter->rx_ring[i]);
3687 * igb_free_tx_resources - Free Tx Resources per Queue
3688 * @tx_ring: Tx descriptor ring for a specific queue
3690 * Free all transmit software resources
3692 void igb_free_tx_resources(struct igb_ring *tx_ring)
3694 igb_clean_tx_ring(tx_ring);
3696 vfree(tx_ring->tx_buffer_info);
3697 tx_ring->tx_buffer_info = NULL;
3699 /* if not set, then don't free */
3703 dma_free_coherent(tx_ring->dev, tx_ring->size,
3704 tx_ring->desc, tx_ring->dma);
3706 tx_ring->desc = NULL;
3710 * igb_free_all_tx_resources - Free Tx Resources for All Queues
3711 * @adapter: board private structure
3713 * Free all transmit software resources
3715 static void igb_free_all_tx_resources(struct igb_adapter *adapter)
3719 for (i = 0; i < adapter->num_tx_queues; i++)
3720 igb_free_tx_resources(adapter->tx_ring[i]);
3723 void igb_unmap_and_free_tx_resource(struct igb_ring *ring,
3724 struct igb_tx_buffer *tx_buffer)
3726 if (tx_buffer->skb) {
3727 dev_kfree_skb_any(tx_buffer->skb);
3728 if (dma_unmap_len(tx_buffer, len))
3729 dma_unmap_single(ring->dev,
3730 dma_unmap_addr(tx_buffer, dma),
3731 dma_unmap_len(tx_buffer, len),
3733 } else if (dma_unmap_len(tx_buffer, len)) {
3734 dma_unmap_page(ring->dev,
3735 dma_unmap_addr(tx_buffer, dma),
3736 dma_unmap_len(tx_buffer, len),
3739 tx_buffer->next_to_watch = NULL;
3740 tx_buffer->skb = NULL;
3741 dma_unmap_len_set(tx_buffer, len, 0);
3742 /* buffer_info must be completely set up in the transmit path */
3746 * igb_clean_tx_ring - Free Tx Buffers
3747 * @tx_ring: ring to be cleaned
3749 static void igb_clean_tx_ring(struct igb_ring *tx_ring)
3751 struct igb_tx_buffer *buffer_info;
3755 if (!tx_ring->tx_buffer_info)
3757 /* Free all the Tx ring sk_buffs */
3759 for (i = 0; i < tx_ring->count; i++) {
3760 buffer_info = &tx_ring->tx_buffer_info[i];
3761 igb_unmap_and_free_tx_resource(tx_ring, buffer_info);
3764 netdev_tx_reset_queue(txring_txq(tx_ring));
3766 size = sizeof(struct igb_tx_buffer) * tx_ring->count;
3767 memset(tx_ring->tx_buffer_info, 0, size);
3769 /* Zero out the descriptor ring */
3770 memset(tx_ring->desc, 0, tx_ring->size);
3772 tx_ring->next_to_use = 0;
3773 tx_ring->next_to_clean = 0;
3777 * igb_clean_all_tx_rings - Free Tx Buffers for all queues
3778 * @adapter: board private structure
3780 static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
3784 for (i = 0; i < adapter->num_tx_queues; i++)
3785 igb_clean_tx_ring(adapter->tx_ring[i]);
3789 * igb_free_rx_resources - Free Rx Resources
3790 * @rx_ring: ring to clean the resources from
3792 * Free all receive software resources
3794 void igb_free_rx_resources(struct igb_ring *rx_ring)
3796 igb_clean_rx_ring(rx_ring);
3798 vfree(rx_ring->rx_buffer_info);
3799 rx_ring->rx_buffer_info = NULL;
3801 /* if not set, then don't free */
3805 dma_free_coherent(rx_ring->dev, rx_ring->size,
3806 rx_ring->desc, rx_ring->dma);
3808 rx_ring->desc = NULL;
3812 * igb_free_all_rx_resources - Free Rx Resources for All Queues
3813 * @adapter: board private structure
3815 * Free all receive software resources
3817 static void igb_free_all_rx_resources(struct igb_adapter *adapter)
3821 for (i = 0; i < adapter->num_rx_queues; i++)
3822 igb_free_rx_resources(adapter->rx_ring[i]);
3826 * igb_clean_rx_ring - Free Rx Buffers per Queue
3827 * @rx_ring: ring to free buffers from
3829 static void igb_clean_rx_ring(struct igb_ring *rx_ring)
3835 dev_kfree_skb(rx_ring->skb);
3836 rx_ring->skb = NULL;
3838 if (!rx_ring->rx_buffer_info)
3841 /* Free all the Rx ring sk_buffs */
3842 for (i = 0; i < rx_ring->count; i++) {
3843 struct igb_rx_buffer *buffer_info = &rx_ring->rx_buffer_info[i];
3845 if (!buffer_info->page)
3848 dma_unmap_page(rx_ring->dev,
3852 __free_page(buffer_info->page);
3854 buffer_info->page = NULL;
3857 size = sizeof(struct igb_rx_buffer) * rx_ring->count;
3858 memset(rx_ring->rx_buffer_info, 0, size);
3860 /* Zero out the descriptor ring */
3861 memset(rx_ring->desc, 0, rx_ring->size);
3863 rx_ring->next_to_alloc = 0;
3864 rx_ring->next_to_clean = 0;
3865 rx_ring->next_to_use = 0;
3869 * igb_clean_all_rx_rings - Free Rx Buffers for all queues
3870 * @adapter: board private structure
3872 static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
3876 for (i = 0; i < adapter->num_rx_queues; i++)
3877 igb_clean_rx_ring(adapter->rx_ring[i]);
3881 * igb_set_mac - Change the Ethernet Address of the NIC
3882 * @netdev: network interface device structure
3883 * @p: pointer to an address structure
3885 * Returns 0 on success, negative on failure
3887 static int igb_set_mac(struct net_device *netdev, void *p)
3889 struct igb_adapter *adapter = netdev_priv(netdev);
3890 struct e1000_hw *hw = &adapter->hw;
3891 struct sockaddr *addr = p;
3893 if (!is_valid_ether_addr(addr->sa_data))
3894 return -EADDRNOTAVAIL;
3896 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3897 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
3899 /* set the correct pool for the new PF MAC address in entry 0 */
3900 igb_rar_set_qsel(adapter, hw->mac.addr, 0,
3901 adapter->vfs_allocated_count);
3907 * igb_write_mc_addr_list - write multicast addresses to MTA
3908 * @netdev: network interface device structure
3910 * Writes multicast address list to the MTA hash table.
3911 * Returns: -ENOMEM on failure
3912 * 0 on no addresses written
3913 * X on writing X addresses to MTA
3915 static int igb_write_mc_addr_list(struct net_device *netdev)
3917 struct igb_adapter *adapter = netdev_priv(netdev);
3918 struct e1000_hw *hw = &adapter->hw;
3919 struct netdev_hw_addr *ha;
3923 if (netdev_mc_empty(netdev)) {
3924 /* nothing to program, so clear mc list */
3925 igb_update_mc_addr_list(hw, NULL, 0);
3926 igb_restore_vf_multicasts(adapter);
3930 mta_list = kzalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
3934 /* The shared function expects a packed array of only addresses. */
3936 netdev_for_each_mc_addr(ha, netdev)
3937 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3939 igb_update_mc_addr_list(hw, mta_list, i);
3942 return netdev_mc_count(netdev);
3946 * igb_write_uc_addr_list - write unicast addresses to RAR table
3947 * @netdev: network interface device structure
3949 * Writes unicast address list to the RAR table.
3950 * Returns: -ENOMEM on failure/insufficient address space
3951 * 0 on no addresses written
3952 * X on writing X addresses to the RAR table
3954 static int igb_write_uc_addr_list(struct net_device *netdev)
3956 struct igb_adapter *adapter = netdev_priv(netdev);
3957 struct e1000_hw *hw = &adapter->hw;
3958 unsigned int vfn = adapter->vfs_allocated_count;
3959 unsigned int rar_entries = hw->mac.rar_entry_count - (vfn + 1);
3962 /* return ENOMEM indicating insufficient memory for addresses */
3963 if (netdev_uc_count(netdev) > rar_entries)
3966 if (!netdev_uc_empty(netdev) && rar_entries) {
3967 struct netdev_hw_addr *ha;
3969 netdev_for_each_uc_addr(ha, netdev) {
3972 igb_rar_set_qsel(adapter, ha->addr,
3978 /* write the addresses in reverse order to avoid write combining */
3979 for (; rar_entries > 0 ; rar_entries--) {
3980 wr32(E1000_RAH(rar_entries), 0);
3981 wr32(E1000_RAL(rar_entries), 0);
3989 * igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
3990 * @netdev: network interface device structure
3992 * The set_rx_mode entry point is called whenever the unicast or multicast
3993 * address lists or the network interface flags are updated. This routine is
3994 * responsible for configuring the hardware for proper unicast, multicast,
3995 * promiscuous mode, and all-multi behavior.
3997 static void igb_set_rx_mode(struct net_device *netdev)
3999 struct igb_adapter *adapter = netdev_priv(netdev);
4000 struct e1000_hw *hw = &adapter->hw;
4001 unsigned int vfn = adapter->vfs_allocated_count;
4002 u32 rctl, vmolr = 0;
4005 /* Check for Promiscuous and All Multicast modes */
4006 rctl = rd32(E1000_RCTL);
4008 /* clear the effected bits */
4009 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE | E1000_RCTL_VFE);
4011 if (netdev->flags & IFF_PROMISC) {
4012 /* retain VLAN HW filtering if in VT mode */
4013 if (adapter->vfs_allocated_count)
4014 rctl |= E1000_RCTL_VFE;
4015 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
4016 vmolr |= (E1000_VMOLR_ROPE | E1000_VMOLR_MPME);
4018 if (netdev->flags & IFF_ALLMULTI) {
4019 rctl |= E1000_RCTL_MPE;
4020 vmolr |= E1000_VMOLR_MPME;
4022 /* Write addresses to the MTA, if the attempt fails
4023 * then we should just turn on promiscuous mode so
4024 * that we can at least receive multicast traffic
4026 count = igb_write_mc_addr_list(netdev);
4028 rctl |= E1000_RCTL_MPE;
4029 vmolr |= E1000_VMOLR_MPME;
4031 vmolr |= E1000_VMOLR_ROMPE;
4034 /* Write addresses to available RAR registers, if there is not
4035 * sufficient space to store all the addresses then enable
4036 * unicast promiscuous mode
4038 count = igb_write_uc_addr_list(netdev);
4040 rctl |= E1000_RCTL_UPE;
4041 vmolr |= E1000_VMOLR_ROPE;
4043 rctl |= E1000_RCTL_VFE;
4045 wr32(E1000_RCTL, rctl);
4047 /* In order to support SR-IOV and eventually VMDq it is necessary to set
4048 * the VMOLR to enable the appropriate modes. Without this workaround
4049 * we will have issues with VLAN tag stripping not being done for frames
4050 * that are only arriving because we are the default pool
4052 if ((hw->mac.type < e1000_82576) || (hw->mac.type > e1000_i350))
4055 vmolr |= rd32(E1000_VMOLR(vfn)) &
4056 ~(E1000_VMOLR_ROPE | E1000_VMOLR_MPME | E1000_VMOLR_ROMPE);
4057 wr32(E1000_VMOLR(vfn), vmolr);
4058 igb_restore_vf_multicasts(adapter);
4061 static void igb_check_wvbr(struct igb_adapter *adapter)
4063 struct e1000_hw *hw = &adapter->hw;
4066 switch (hw->mac.type) {
4069 wvbr = rd32(E1000_WVBR);
4077 adapter->wvbr |= wvbr;
4080 #define IGB_STAGGERED_QUEUE_OFFSET 8
4082 static void igb_spoof_check(struct igb_adapter *adapter)
4089 for (j = 0; j < adapter->vfs_allocated_count; j++) {
4090 if (adapter->wvbr & (1 << j) ||
4091 adapter->wvbr & (1 << (j + IGB_STAGGERED_QUEUE_OFFSET))) {
4092 dev_warn(&adapter->pdev->dev,
4093 "Spoof event(s) detected on VF %d\n", j);
4096 (1 << (j + IGB_STAGGERED_QUEUE_OFFSET)));
4101 /* Need to wait a few seconds after link up to get diagnostic information from
4104 static void igb_update_phy_info(unsigned long data)
4106 struct igb_adapter *adapter = (struct igb_adapter *) data;
4107 igb_get_phy_info(&adapter->hw);
4111 * igb_has_link - check shared code for link and determine up/down
4112 * @adapter: pointer to driver private info
4114 bool igb_has_link(struct igb_adapter *adapter)
4116 struct e1000_hw *hw = &adapter->hw;
4117 bool link_active = false;
4119 /* get_link_status is set on LSC (link status) interrupt or
4120 * rx sequence error interrupt. get_link_status will stay
4121 * false until the e1000_check_for_link establishes link
4122 * for copper adapters ONLY
4124 switch (hw->phy.media_type) {
4125 case e1000_media_type_copper:
4126 if (!hw->mac.get_link_status)
4128 case e1000_media_type_internal_serdes:
4129 hw->mac.ops.check_for_link(hw);
4130 link_active = !hw->mac.get_link_status;
4133 case e1000_media_type_unknown:
4137 if (((hw->mac.type == e1000_i210) ||
4138 (hw->mac.type == e1000_i211)) &&
4139 (hw->phy.id == I210_I_PHY_ID)) {
4140 if (!netif_carrier_ok(adapter->netdev)) {
4141 adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
4142 } else if (!(adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)) {
4143 adapter->flags |= IGB_FLAG_NEED_LINK_UPDATE;
4144 adapter->link_check_timeout = jiffies;
4151 static bool igb_thermal_sensor_event(struct e1000_hw *hw, u32 event)
4154 u32 ctrl_ext, thstat;
4156 /* check for thermal sensor event on i350 copper only */
4157 if (hw->mac.type == e1000_i350) {
4158 thstat = rd32(E1000_THSTAT);
4159 ctrl_ext = rd32(E1000_CTRL_EXT);
4161 if ((hw->phy.media_type == e1000_media_type_copper) &&
4162 !(ctrl_ext & E1000_CTRL_EXT_LINK_MODE_SGMII))
4163 ret = !!(thstat & event);
4170 * igb_check_lvmmc - check for malformed packets received
4171 * and indicated in LVMMC register
4172 * @adapter: pointer to adapter
4174 static void igb_check_lvmmc(struct igb_adapter *adapter)
4176 struct e1000_hw *hw = &adapter->hw;
4179 lvmmc = rd32(E1000_LVMMC);
4181 if (unlikely(net_ratelimit())) {
4182 netdev_warn(adapter->netdev,
4183 "malformed Tx packet detected and dropped, LVMMC:0x%08x\n",
4190 * igb_watchdog - Timer Call-back
4191 * @data: pointer to adapter cast into an unsigned long
4193 static void igb_watchdog(unsigned long data)
4195 struct igb_adapter *adapter = (struct igb_adapter *)data;
4196 /* Do the rest outside of interrupt context */
4197 schedule_work(&adapter->watchdog_task);
4200 static void igb_watchdog_task(struct work_struct *work)
4202 struct igb_adapter *adapter = container_of(work,
4205 struct e1000_hw *hw = &adapter->hw;
4206 struct e1000_phy_info *phy = &hw->phy;
4207 struct net_device *netdev = adapter->netdev;
4212 link = igb_has_link(adapter);
4214 if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE) {
4215 if (time_after(jiffies, (adapter->link_check_timeout + HZ)))
4216 adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
4221 /* Force link down if we have fiber to swap to */
4222 if (adapter->flags & IGB_FLAG_MAS_ENABLE) {
4223 if (hw->phy.media_type == e1000_media_type_copper) {
4224 connsw = rd32(E1000_CONNSW);
4225 if (!(connsw & E1000_CONNSW_AUTOSENSE_EN))
4230 /* Perform a reset if the media type changed. */
4231 if (hw->dev_spec._82575.media_changed) {
4232 hw->dev_spec._82575.media_changed = false;
4233 adapter->flags |= IGB_FLAG_MEDIA_RESET;
4236 /* Cancel scheduled suspend requests. */
4237 pm_runtime_resume(netdev->dev.parent);
4239 if (!netif_carrier_ok(netdev)) {
4242 hw->mac.ops.get_speed_and_duplex(hw,
4243 &adapter->link_speed,
4244 &adapter->link_duplex);
4246 ctrl = rd32(E1000_CTRL);
4247 /* Links status message must follow this format */
4249 "igb: %s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
4251 adapter->link_speed,
4252 adapter->link_duplex == FULL_DUPLEX ?
4254 (ctrl & E1000_CTRL_TFCE) &&
4255 (ctrl & E1000_CTRL_RFCE) ? "RX/TX" :
4256 (ctrl & E1000_CTRL_RFCE) ? "RX" :
4257 (ctrl & E1000_CTRL_TFCE) ? "TX" : "None");
4259 /* disable EEE if enabled */
4260 if ((adapter->flags & IGB_FLAG_EEE) &&
4261 (adapter->link_duplex == HALF_DUPLEX)) {
4262 dev_info(&adapter->pdev->dev,
4263 "EEE Disabled: unsupported at half duplex. Re-enable using ethtool when at full duplex.\n");
4264 adapter->hw.dev_spec._82575.eee_disable = true;
4265 adapter->flags &= ~IGB_FLAG_EEE;
4268 /* check if SmartSpeed worked */
4269 igb_check_downshift(hw);
4270 if (phy->speed_downgraded)
4271 netdev_warn(netdev, "Link Speed was downgraded by SmartSpeed\n");
4273 /* check for thermal sensor event */
4274 if (igb_thermal_sensor_event(hw,
4275 E1000_THSTAT_LINK_THROTTLE))
4276 netdev_info(netdev, "The network adapter link speed was downshifted because it overheated\n");
4278 /* adjust timeout factor according to speed/duplex */
4279 adapter->tx_timeout_factor = 1;
4280 switch (adapter->link_speed) {
4282 adapter->tx_timeout_factor = 14;
4285 /* maybe add some timeout factor ? */
4289 netif_carrier_on(netdev);
4291 igb_ping_all_vfs(adapter);
4292 igb_check_vf_rate_limit(adapter);
4294 /* link state has changed, schedule phy info update */
4295 if (!test_bit(__IGB_DOWN, &adapter->state))
4296 mod_timer(&adapter->phy_info_timer,
4297 round_jiffies(jiffies + 2 * HZ));
4300 if (netif_carrier_ok(netdev)) {
4301 adapter->link_speed = 0;
4302 adapter->link_duplex = 0;
4304 /* check for thermal sensor event */
4305 if (igb_thermal_sensor_event(hw,
4306 E1000_THSTAT_PWR_DOWN)) {
4307 netdev_err(netdev, "The network adapter was stopped because it overheated\n");
4310 /* Links status message must follow this format */
4311 netdev_info(netdev, "igb: %s NIC Link is Down\n",
4313 netif_carrier_off(netdev);
4315 igb_ping_all_vfs(adapter);
4317 /* link state has changed, schedule phy info update */
4318 if (!test_bit(__IGB_DOWN, &adapter->state))
4319 mod_timer(&adapter->phy_info_timer,
4320 round_jiffies(jiffies + 2 * HZ));
4322 /* link is down, time to check for alternate media */
4323 if (adapter->flags & IGB_FLAG_MAS_ENABLE) {
4324 igb_check_swap_media(adapter);
4325 if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
4326 schedule_work(&adapter->reset_task);
4327 /* return immediately */
4331 pm_schedule_suspend(netdev->dev.parent,
4334 /* also check for alternate media here */
4335 } else if (!netif_carrier_ok(netdev) &&
4336 (adapter->flags & IGB_FLAG_MAS_ENABLE)) {
4337 igb_check_swap_media(adapter);
4338 if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
4339 schedule_work(&adapter->reset_task);
4340 /* return immediately */
4346 spin_lock(&adapter->stats64_lock);
4347 igb_update_stats(adapter, &adapter->stats64);
4348 spin_unlock(&adapter->stats64_lock);
4350 for (i = 0; i < adapter->num_tx_queues; i++) {
4351 struct igb_ring *tx_ring = adapter->tx_ring[i];
4352 if (!netif_carrier_ok(netdev)) {
4353 /* We've lost link, so the controller stops DMA,
4354 * but we've got queued Tx work that's never going
4355 * to get done, so reset controller to flush Tx.
4356 * (Do the reset outside of interrupt context).
4358 if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) {
4359 adapter->tx_timeout_count++;
4360 schedule_work(&adapter->reset_task);
4361 /* return immediately since reset is imminent */
4366 /* Force detection of hung controller every watchdog period */
4367 set_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
4370 /* Cause software interrupt to ensure Rx ring is cleaned */
4371 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
4374 for (i = 0; i < adapter->num_q_vectors; i++)
4375 eics |= adapter->q_vector[i]->eims_value;
4376 wr32(E1000_EICS, eics);
4378 wr32(E1000_ICS, E1000_ICS_RXDMT0);
4381 igb_spoof_check(adapter);
4382 igb_ptp_rx_hang(adapter);
4384 /* Check LVMMC register on i350/i354 only */
4385 if ((adapter->hw.mac.type == e1000_i350) ||
4386 (adapter->hw.mac.type == e1000_i354))
4387 igb_check_lvmmc(adapter);
4389 /* Reset the timer */
4390 if (!test_bit(__IGB_DOWN, &adapter->state)) {
4391 if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)
4392 mod_timer(&adapter->watchdog_timer,
4393 round_jiffies(jiffies + HZ));
4395 mod_timer(&adapter->watchdog_timer,
4396 round_jiffies(jiffies + 2 * HZ));
4400 enum latency_range {
4404 latency_invalid = 255
4408 * igb_update_ring_itr - update the dynamic ITR value based on packet size
4409 * @q_vector: pointer to q_vector
4411 * Stores a new ITR value based on strictly on packet size. This
4412 * algorithm is less sophisticated than that used in igb_update_itr,
4413 * due to the difficulty of synchronizing statistics across multiple
4414 * receive rings. The divisors and thresholds used by this function
4415 * were determined based on theoretical maximum wire speed and testing
4416 * data, in order to minimize response time while increasing bulk
4418 * This functionality is controlled by ethtool's coalescing settings.
4419 * NOTE: This function is called only when operating in a multiqueue
4420 * receive environment.
4422 static void igb_update_ring_itr(struct igb_q_vector *q_vector)
4424 int new_val = q_vector->itr_val;
4425 int avg_wire_size = 0;
4426 struct igb_adapter *adapter = q_vector->adapter;
4427 unsigned int packets;
4429 /* For non-gigabit speeds, just fix the interrupt rate at 4000
4430 * ints/sec - ITR timer value of 120 ticks.
4432 if (adapter->link_speed != SPEED_1000) {
4433 new_val = IGB_4K_ITR;
4437 packets = q_vector->rx.total_packets;
4439 avg_wire_size = q_vector->rx.total_bytes / packets;
4441 packets = q_vector->tx.total_packets;
4443 avg_wire_size = max_t(u32, avg_wire_size,
4444 q_vector->tx.total_bytes / packets);
4446 /* if avg_wire_size isn't set no work was done */
4450 /* Add 24 bytes to size to account for CRC, preamble, and gap */
4451 avg_wire_size += 24;
4453 /* Don't starve jumbo frames */
4454 avg_wire_size = min(avg_wire_size, 3000);
4456 /* Give a little boost to mid-size frames */
4457 if ((avg_wire_size > 300) && (avg_wire_size < 1200))
4458 new_val = avg_wire_size / 3;
4460 new_val = avg_wire_size / 2;
4462 /* conservative mode (itr 3) eliminates the lowest_latency setting */
4463 if (new_val < IGB_20K_ITR &&
4464 ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
4465 (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
4466 new_val = IGB_20K_ITR;
4469 if (new_val != q_vector->itr_val) {
4470 q_vector->itr_val = new_val;
4471 q_vector->set_itr = 1;
4474 q_vector->rx.total_bytes = 0;
4475 q_vector->rx.total_packets = 0;
4476 q_vector->tx.total_bytes = 0;
4477 q_vector->tx.total_packets = 0;
4481 * igb_update_itr - update the dynamic ITR value based on statistics
4482 * @q_vector: pointer to q_vector
4483 * @ring_container: ring info to update the itr for
4485 * Stores a new ITR value based on packets and byte
4486 * counts during the last interrupt. The advantage of per interrupt
4487 * computation is faster updates and more accurate ITR for the current
4488 * traffic pattern. Constants in this function were computed
4489 * based on theoretical maximum wire speed and thresholds were set based
4490 * on testing data as well as attempting to minimize response time
4491 * while increasing bulk throughput.
4492 * This functionality is controlled by ethtool's coalescing settings.
4493 * NOTE: These calculations are only valid when operating in a single-
4494 * queue environment.
4496 static void igb_update_itr(struct igb_q_vector *q_vector,
4497 struct igb_ring_container *ring_container)
4499 unsigned int packets = ring_container->total_packets;
4500 unsigned int bytes = ring_container->total_bytes;
4501 u8 itrval = ring_container->itr;
4503 /* no packets, exit with status unchanged */
4508 case lowest_latency:
4509 /* handle TSO and jumbo frames */
4510 if (bytes/packets > 8000)
4511 itrval = bulk_latency;
4512 else if ((packets < 5) && (bytes > 512))
4513 itrval = low_latency;
4515 case low_latency: /* 50 usec aka 20000 ints/s */
4516 if (bytes > 10000) {
4517 /* this if handles the TSO accounting */
4518 if (bytes/packets > 8000)
4519 itrval = bulk_latency;
4520 else if ((packets < 10) || ((bytes/packets) > 1200))
4521 itrval = bulk_latency;
4522 else if ((packets > 35))
4523 itrval = lowest_latency;
4524 } else if (bytes/packets > 2000) {
4525 itrval = bulk_latency;
4526 } else if (packets <= 2 && bytes < 512) {
4527 itrval = lowest_latency;
4530 case bulk_latency: /* 250 usec aka 4000 ints/s */
4531 if (bytes > 25000) {
4533 itrval = low_latency;
4534 } else if (bytes < 1500) {
4535 itrval = low_latency;
4540 /* clear work counters since we have the values we need */
4541 ring_container->total_bytes = 0;
4542 ring_container->total_packets = 0;
4544 /* write updated itr to ring container */
4545 ring_container->itr = itrval;
4548 static void igb_set_itr(struct igb_q_vector *q_vector)
4550 struct igb_adapter *adapter = q_vector->adapter;
4551 u32 new_itr = q_vector->itr_val;
4554 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
4555 if (adapter->link_speed != SPEED_1000) {
4557 new_itr = IGB_4K_ITR;
4561 igb_update_itr(q_vector, &q_vector->tx);
4562 igb_update_itr(q_vector, &q_vector->rx);
4564 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
4566 /* conservative mode (itr 3) eliminates the lowest_latency setting */
4567 if (current_itr == lowest_latency &&
4568 ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
4569 (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
4570 current_itr = low_latency;
4572 switch (current_itr) {
4573 /* counts and packets in update_itr are dependent on these numbers */
4574 case lowest_latency:
4575 new_itr = IGB_70K_ITR; /* 70,000 ints/sec */
4578 new_itr = IGB_20K_ITR; /* 20,000 ints/sec */
4581 new_itr = IGB_4K_ITR; /* 4,000 ints/sec */
4588 if (new_itr != q_vector->itr_val) {
4589 /* this attempts to bias the interrupt rate towards Bulk
4590 * by adding intermediate steps when interrupt rate is
4593 new_itr = new_itr > q_vector->itr_val ?
4594 max((new_itr * q_vector->itr_val) /
4595 (new_itr + (q_vector->itr_val >> 2)),
4597 /* Don't write the value here; it resets the adapter's
4598 * internal timer, and causes us to delay far longer than
4599 * we should between interrupts. Instead, we write the ITR
4600 * value at the beginning of the next interrupt so the timing
4601 * ends up being correct.
4603 q_vector->itr_val = new_itr;
4604 q_vector->set_itr = 1;
4608 static void igb_tx_ctxtdesc(struct igb_ring *tx_ring, u32 vlan_macip_lens,
4609 u32 type_tucmd, u32 mss_l4len_idx)
4611 struct e1000_adv_tx_context_desc *context_desc;
4612 u16 i = tx_ring->next_to_use;
4614 context_desc = IGB_TX_CTXTDESC(tx_ring, i);
4617 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
4619 /* set bits to identify this as an advanced context descriptor */
4620 type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
4622 /* For 82575, context index must be unique per ring. */
4623 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
4624 mss_l4len_idx |= tx_ring->reg_idx << 4;
4626 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
4627 context_desc->seqnum_seed = 0;
4628 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd);
4629 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
4632 static int igb_tso(struct igb_ring *tx_ring,
4633 struct igb_tx_buffer *first,
4636 struct sk_buff *skb = first->skb;
4637 u32 vlan_macip_lens, type_tucmd;
4638 u32 mss_l4len_idx, l4len;
4641 if (skb->ip_summed != CHECKSUM_PARTIAL)
4644 if (!skb_is_gso(skb))
4647 err = skb_cow_head(skb, 0);
4651 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
4652 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
4654 if (first->protocol == htons(ETH_P_IP)) {
4655 struct iphdr *iph = ip_hdr(skb);
4658 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
4662 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
4663 first->tx_flags |= IGB_TX_FLAGS_TSO |
4666 } else if (skb_is_gso_v6(skb)) {
4667 ipv6_hdr(skb)->payload_len = 0;
4668 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4669 &ipv6_hdr(skb)->daddr,
4671 first->tx_flags |= IGB_TX_FLAGS_TSO |
4675 /* compute header lengths */
4676 l4len = tcp_hdrlen(skb);
4677 *hdr_len = skb_transport_offset(skb) + l4len;
4679 /* update gso size and bytecount with header size */
4680 first->gso_segs = skb_shinfo(skb)->gso_segs;
4681 first->bytecount += (first->gso_segs - 1) * *hdr_len;
4684 mss_l4len_idx = l4len << E1000_ADVTXD_L4LEN_SHIFT;
4685 mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT;
4687 /* VLAN MACLEN IPLEN */
4688 vlan_macip_lens = skb_network_header_len(skb);
4689 vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
4690 vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
4692 igb_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
4697 static void igb_tx_csum(struct igb_ring *tx_ring, struct igb_tx_buffer *first)
4699 struct sk_buff *skb = first->skb;
4700 u32 vlan_macip_lens = 0;
4701 u32 mss_l4len_idx = 0;
4704 if (skb->ip_summed != CHECKSUM_PARTIAL) {
4705 if (!(first->tx_flags & IGB_TX_FLAGS_VLAN))
4710 switch (first->protocol) {
4711 case htons(ETH_P_IP):
4712 vlan_macip_lens |= skb_network_header_len(skb);
4713 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
4714 l4_hdr = ip_hdr(skb)->protocol;
4716 case htons(ETH_P_IPV6):
4717 vlan_macip_lens |= skb_network_header_len(skb);
4718 l4_hdr = ipv6_hdr(skb)->nexthdr;
4721 if (unlikely(net_ratelimit())) {
4722 dev_warn(tx_ring->dev,
4723 "partial checksum but proto=%x!\n",
4731 type_tucmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
4732 mss_l4len_idx = tcp_hdrlen(skb) <<
4733 E1000_ADVTXD_L4LEN_SHIFT;
4736 type_tucmd |= E1000_ADVTXD_TUCMD_L4T_SCTP;
4737 mss_l4len_idx = sizeof(struct sctphdr) <<
4738 E1000_ADVTXD_L4LEN_SHIFT;
4741 mss_l4len_idx = sizeof(struct udphdr) <<
4742 E1000_ADVTXD_L4LEN_SHIFT;
4745 if (unlikely(net_ratelimit())) {
4746 dev_warn(tx_ring->dev,
4747 "partial checksum but l4 proto=%x!\n",
4753 /* update TX checksum flag */
4754 first->tx_flags |= IGB_TX_FLAGS_CSUM;
4757 vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
4758 vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
4760 igb_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
4763 #define IGB_SET_FLAG(_input, _flag, _result) \
4764 ((_flag <= _result) ? \
4765 ((u32)(_input & _flag) * (_result / _flag)) : \
4766 ((u32)(_input & _flag) / (_flag / _result)))
4768 static u32 igb_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
4770 /* set type for advanced descriptor with frame checksum insertion */
4771 u32 cmd_type = E1000_ADVTXD_DTYP_DATA |
4772 E1000_ADVTXD_DCMD_DEXT |
4773 E1000_ADVTXD_DCMD_IFCS;
4775 /* set HW vlan bit if vlan is present */
4776 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_VLAN,
4777 (E1000_ADVTXD_DCMD_VLE));
4779 /* set segmentation bits for TSO */
4780 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSO,
4781 (E1000_ADVTXD_DCMD_TSE));
4783 /* set timestamp bit if present */
4784 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSTAMP,
4785 (E1000_ADVTXD_MAC_TSTAMP));
4787 /* insert frame checksum */
4788 cmd_type ^= IGB_SET_FLAG(skb->no_fcs, 1, E1000_ADVTXD_DCMD_IFCS);
4793 static void igb_tx_olinfo_status(struct igb_ring *tx_ring,
4794 union e1000_adv_tx_desc *tx_desc,
4795 u32 tx_flags, unsigned int paylen)
4797 u32 olinfo_status = paylen << E1000_ADVTXD_PAYLEN_SHIFT;
4799 /* 82575 requires a unique index per ring */
4800 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
4801 olinfo_status |= tx_ring->reg_idx << 4;
4803 /* insert L4 checksum */
4804 olinfo_status |= IGB_SET_FLAG(tx_flags,
4806 (E1000_TXD_POPTS_TXSM << 8));
4808 /* insert IPv4 checksum */
4809 olinfo_status |= IGB_SET_FLAG(tx_flags,
4811 (E1000_TXD_POPTS_IXSM << 8));
4813 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
4816 static int __igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
4818 struct net_device *netdev = tx_ring->netdev;
4820 netif_stop_subqueue(netdev, tx_ring->queue_index);
4822 /* Herbert's original patch had:
4823 * smp_mb__after_netif_stop_queue();
4824 * but since that doesn't exist yet, just open code it.
4828 /* We need to check again in a case another CPU has just
4829 * made room available.
4831 if (igb_desc_unused(tx_ring) < size)
4835 netif_wake_subqueue(netdev, tx_ring->queue_index);
4837 u64_stats_update_begin(&tx_ring->tx_syncp2);
4838 tx_ring->tx_stats.restart_queue2++;
4839 u64_stats_update_end(&tx_ring->tx_syncp2);
4844 static inline int igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
4846 if (igb_desc_unused(tx_ring) >= size)
4848 return __igb_maybe_stop_tx(tx_ring, size);
4851 static void igb_tx_map(struct igb_ring *tx_ring,
4852 struct igb_tx_buffer *first,
4855 struct sk_buff *skb = first->skb;
4856 struct igb_tx_buffer *tx_buffer;
4857 union e1000_adv_tx_desc *tx_desc;
4858 struct skb_frag_struct *frag;
4860 unsigned int data_len, size;
4861 u32 tx_flags = first->tx_flags;
4862 u32 cmd_type = igb_tx_cmd_type(skb, tx_flags);
4863 u16 i = tx_ring->next_to_use;
4865 tx_desc = IGB_TX_DESC(tx_ring, i);
4867 igb_tx_olinfo_status(tx_ring, tx_desc, tx_flags, skb->len - hdr_len);
4869 size = skb_headlen(skb);
4870 data_len = skb->data_len;
4872 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
4876 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
4877 if (dma_mapping_error(tx_ring->dev, dma))
4880 /* record length, and DMA address */
4881 dma_unmap_len_set(tx_buffer, len, size);
4882 dma_unmap_addr_set(tx_buffer, dma, dma);
4884 tx_desc->read.buffer_addr = cpu_to_le64(dma);
4886 while (unlikely(size > IGB_MAX_DATA_PER_TXD)) {
4887 tx_desc->read.cmd_type_len =
4888 cpu_to_le32(cmd_type ^ IGB_MAX_DATA_PER_TXD);
4892 if (i == tx_ring->count) {
4893 tx_desc = IGB_TX_DESC(tx_ring, 0);
4896 tx_desc->read.olinfo_status = 0;
4898 dma += IGB_MAX_DATA_PER_TXD;
4899 size -= IGB_MAX_DATA_PER_TXD;
4901 tx_desc->read.buffer_addr = cpu_to_le64(dma);
4904 if (likely(!data_len))
4907 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
4911 if (i == tx_ring->count) {
4912 tx_desc = IGB_TX_DESC(tx_ring, 0);
4915 tx_desc->read.olinfo_status = 0;
4917 size = skb_frag_size(frag);
4920 dma = skb_frag_dma_map(tx_ring->dev, frag, 0,
4921 size, DMA_TO_DEVICE);
4923 tx_buffer = &tx_ring->tx_buffer_info[i];
4926 /* write last descriptor with RS and EOP bits */
4927 cmd_type |= size | IGB_TXD_DCMD;
4928 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
4930 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
4932 /* set the timestamp */
4933 first->time_stamp = jiffies;
4935 /* Force memory writes to complete before letting h/w know there
4936 * are new descriptors to fetch. (Only applicable for weak-ordered
4937 * memory model archs, such as IA-64).
4939 * We also need this memory barrier to make certain all of the
4940 * status bits have been updated before next_to_watch is written.
4944 /* set next_to_watch value indicating a packet is present */
4945 first->next_to_watch = tx_desc;
4948 if (i == tx_ring->count)
4951 tx_ring->next_to_use = i;
4953 /* Make sure there is space in the ring for the next send. */
4954 igb_maybe_stop_tx(tx_ring, DESC_NEEDED);
4956 if (netif_xmit_stopped(txring_txq(tx_ring)) || !skb->xmit_more) {
4957 writel(i, tx_ring->tail);
4959 /* we need this if more than one processor can write to our tail
4960 * at a time, it synchronizes IO on IA64/Altix systems
4967 dev_err(tx_ring->dev, "TX DMA map failed\n");
4969 /* clear dma mappings for failed tx_buffer_info map */
4971 tx_buffer = &tx_ring->tx_buffer_info[i];
4972 igb_unmap_and_free_tx_resource(tx_ring, tx_buffer);
4973 if (tx_buffer == first)
4980 tx_ring->next_to_use = i;
4983 netdev_tx_t igb_xmit_frame_ring(struct sk_buff *skb,
4984 struct igb_ring *tx_ring)
4986 struct igb_tx_buffer *first;
4989 u16 count = TXD_USE_COUNT(skb_headlen(skb));
4990 __be16 protocol = vlan_get_protocol(skb);
4993 /* need: 1 descriptor per page * PAGE_SIZE/IGB_MAX_DATA_PER_TXD,
4994 * + 1 desc for skb_headlen/IGB_MAX_DATA_PER_TXD,
4995 * + 2 desc gap to keep tail from touching head,
4996 * + 1 desc for context descriptor,
4997 * otherwise try next time
4999 if (NETDEV_FRAG_PAGE_MAX_SIZE > IGB_MAX_DATA_PER_TXD) {
5002 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
5003 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
5005 count += skb_shinfo(skb)->nr_frags;
5008 if (igb_maybe_stop_tx(tx_ring, count + 3)) {
5009 /* this is a hard error */
5010 return NETDEV_TX_BUSY;
5013 /* record the location of the first descriptor for this packet */
5014 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
5016 first->bytecount = skb->len;
5017 first->gso_segs = 1;
5019 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
5020 struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
5022 if (!test_and_set_bit_lock(__IGB_PTP_TX_IN_PROGRESS,
5024 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
5025 tx_flags |= IGB_TX_FLAGS_TSTAMP;
5027 adapter->ptp_tx_skb = skb_get(skb);
5028 adapter->ptp_tx_start = jiffies;
5029 if (adapter->hw.mac.type == e1000_82576)
5030 schedule_work(&adapter->ptp_tx_work);
5034 skb_tx_timestamp(skb);
5036 if (vlan_tx_tag_present(skb)) {
5037 tx_flags |= IGB_TX_FLAGS_VLAN;
5038 tx_flags |= (vlan_tx_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
5041 /* record initial flags and protocol */
5042 first->tx_flags = tx_flags;
5043 first->protocol = protocol;
5045 tso = igb_tso(tx_ring, first, &hdr_len);
5049 igb_tx_csum(tx_ring, first);
5051 igb_tx_map(tx_ring, first, hdr_len);
5053 return NETDEV_TX_OK;
5056 igb_unmap_and_free_tx_resource(tx_ring, first);
5058 return NETDEV_TX_OK;
5061 static inline struct igb_ring *igb_tx_queue_mapping(struct igb_adapter *adapter,
5062 struct sk_buff *skb)
5064 unsigned int r_idx = skb->queue_mapping;
5066 if (r_idx >= adapter->num_tx_queues)
5067 r_idx = r_idx % adapter->num_tx_queues;
5069 return adapter->tx_ring[r_idx];
5072 static netdev_tx_t igb_xmit_frame(struct sk_buff *skb,
5073 struct net_device *netdev)
5075 struct igb_adapter *adapter = netdev_priv(netdev);
5077 if (test_bit(__IGB_DOWN, &adapter->state)) {
5078 dev_kfree_skb_any(skb);
5079 return NETDEV_TX_OK;
5082 if (skb->len <= 0) {
5083 dev_kfree_skb_any(skb);
5084 return NETDEV_TX_OK;
5087 /* The minimum packet size with TCTL.PSP set is 17 so pad the skb
5088 * in order to meet this minimum size requirement.
5090 if (unlikely(skb->len < 17)) {
5091 if (skb_pad(skb, 17 - skb->len))
5092 return NETDEV_TX_OK;
5094 skb_set_tail_pointer(skb, 17);
5097 return igb_xmit_frame_ring(skb, igb_tx_queue_mapping(adapter, skb));
5101 * igb_tx_timeout - Respond to a Tx Hang
5102 * @netdev: network interface device structure
5104 static void igb_tx_timeout(struct net_device *netdev)
5106 struct igb_adapter *adapter = netdev_priv(netdev);
5107 struct e1000_hw *hw = &adapter->hw;
5109 /* Do the reset outside of interrupt context */
5110 adapter->tx_timeout_count++;
5112 if (hw->mac.type >= e1000_82580)
5113 hw->dev_spec._82575.global_device_reset = true;
5115 schedule_work(&adapter->reset_task);
5117 (adapter->eims_enable_mask & ~adapter->eims_other));
5120 static void igb_reset_task(struct work_struct *work)
5122 struct igb_adapter *adapter;
5123 adapter = container_of(work, struct igb_adapter, reset_task);
5126 netdev_err(adapter->netdev, "Reset adapter\n");
5127 igb_reinit_locked(adapter);
5131 * igb_get_stats64 - Get System Network Statistics
5132 * @netdev: network interface device structure
5133 * @stats: rtnl_link_stats64 pointer
5135 static struct rtnl_link_stats64 *igb_get_stats64(struct net_device *netdev,
5136 struct rtnl_link_stats64 *stats)
5138 struct igb_adapter *adapter = netdev_priv(netdev);
5140 spin_lock(&adapter->stats64_lock);
5141 igb_update_stats(adapter, &adapter->stats64);
5142 memcpy(stats, &adapter->stats64, sizeof(*stats));
5143 spin_unlock(&adapter->stats64_lock);
5149 * igb_change_mtu - Change the Maximum Transfer Unit
5150 * @netdev: network interface device structure
5151 * @new_mtu: new value for maximum frame size
5153 * Returns 0 on success, negative on failure
5155 static int igb_change_mtu(struct net_device *netdev, int new_mtu)
5157 struct igb_adapter *adapter = netdev_priv(netdev);
5158 struct pci_dev *pdev = adapter->pdev;
5159 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
5161 if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
5162 dev_err(&pdev->dev, "Invalid MTU setting\n");
5166 #define MAX_STD_JUMBO_FRAME_SIZE 9238
5167 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
5168 dev_err(&pdev->dev, "MTU > 9216 not supported.\n");
5172 /* adjust max frame to be at least the size of a standard frame */
5173 if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
5174 max_frame = ETH_FRAME_LEN + ETH_FCS_LEN;
5176 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
5177 usleep_range(1000, 2000);
5179 /* igb_down has a dependency on max_frame_size */
5180 adapter->max_frame_size = max_frame;
5182 if (netif_running(netdev))
5185 dev_info(&pdev->dev, "changing MTU from %d to %d\n",
5186 netdev->mtu, new_mtu);
5187 netdev->mtu = new_mtu;
5189 if (netif_running(netdev))
5194 clear_bit(__IGB_RESETTING, &adapter->state);
5200 * igb_update_stats - Update the board statistics counters
5201 * @adapter: board private structure
5203 void igb_update_stats(struct igb_adapter *adapter,
5204 struct rtnl_link_stats64 *net_stats)
5206 struct e1000_hw *hw = &adapter->hw;
5207 struct pci_dev *pdev = adapter->pdev;
5212 u64 _bytes, _packets;
5214 /* Prevent stats update while adapter is being reset, or if the pci
5215 * connection is down.
5217 if (adapter->link_speed == 0)
5219 if (pci_channel_offline(pdev))
5226 for (i = 0; i < adapter->num_rx_queues; i++) {
5227 struct igb_ring *ring = adapter->rx_ring[i];
5228 u32 rqdpc = rd32(E1000_RQDPC(i));
5229 if (hw->mac.type >= e1000_i210)
5230 wr32(E1000_RQDPC(i), 0);
5233 ring->rx_stats.drops += rqdpc;
5234 net_stats->rx_fifo_errors += rqdpc;
5238 start = u64_stats_fetch_begin_irq(&ring->rx_syncp);
5239 _bytes = ring->rx_stats.bytes;
5240 _packets = ring->rx_stats.packets;
5241 } while (u64_stats_fetch_retry_irq(&ring->rx_syncp, start));
5243 packets += _packets;
5246 net_stats->rx_bytes = bytes;
5247 net_stats->rx_packets = packets;
5251 for (i = 0; i < adapter->num_tx_queues; i++) {
5252 struct igb_ring *ring = adapter->tx_ring[i];
5254 start = u64_stats_fetch_begin_irq(&ring->tx_syncp);
5255 _bytes = ring->tx_stats.bytes;
5256 _packets = ring->tx_stats.packets;
5257 } while (u64_stats_fetch_retry_irq(&ring->tx_syncp, start));
5259 packets += _packets;
5261 net_stats->tx_bytes = bytes;
5262 net_stats->tx_packets = packets;
5265 /* read stats registers */
5266 adapter->stats.crcerrs += rd32(E1000_CRCERRS);
5267 adapter->stats.gprc += rd32(E1000_GPRC);
5268 adapter->stats.gorc += rd32(E1000_GORCL);
5269 rd32(E1000_GORCH); /* clear GORCL */
5270 adapter->stats.bprc += rd32(E1000_BPRC);
5271 adapter->stats.mprc += rd32(E1000_MPRC);
5272 adapter->stats.roc += rd32(E1000_ROC);
5274 adapter->stats.prc64 += rd32(E1000_PRC64);
5275 adapter->stats.prc127 += rd32(E1000_PRC127);
5276 adapter->stats.prc255 += rd32(E1000_PRC255);
5277 adapter->stats.prc511 += rd32(E1000_PRC511);
5278 adapter->stats.prc1023 += rd32(E1000_PRC1023);
5279 adapter->stats.prc1522 += rd32(E1000_PRC1522);
5280 adapter->stats.symerrs += rd32(E1000_SYMERRS);
5281 adapter->stats.sec += rd32(E1000_SEC);
5283 mpc = rd32(E1000_MPC);
5284 adapter->stats.mpc += mpc;
5285 net_stats->rx_fifo_errors += mpc;
5286 adapter->stats.scc += rd32(E1000_SCC);
5287 adapter->stats.ecol += rd32(E1000_ECOL);
5288 adapter->stats.mcc += rd32(E1000_MCC);
5289 adapter->stats.latecol += rd32(E1000_LATECOL);
5290 adapter->stats.dc += rd32(E1000_DC);
5291 adapter->stats.rlec += rd32(E1000_RLEC);
5292 adapter->stats.xonrxc += rd32(E1000_XONRXC);
5293 adapter->stats.xontxc += rd32(E1000_XONTXC);
5294 adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
5295 adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
5296 adapter->stats.fcruc += rd32(E1000_FCRUC);
5297 adapter->stats.gptc += rd32(E1000_GPTC);
5298 adapter->stats.gotc += rd32(E1000_GOTCL);
5299 rd32(E1000_GOTCH); /* clear GOTCL */
5300 adapter->stats.rnbc += rd32(E1000_RNBC);
5301 adapter->stats.ruc += rd32(E1000_RUC);
5302 adapter->stats.rfc += rd32(E1000_RFC);
5303 adapter->stats.rjc += rd32(E1000_RJC);
5304 adapter->stats.tor += rd32(E1000_TORH);
5305 adapter->stats.tot += rd32(E1000_TOTH);
5306 adapter->stats.tpr += rd32(E1000_TPR);
5308 adapter->stats.ptc64 += rd32(E1000_PTC64);
5309 adapter->stats.ptc127 += rd32(E1000_PTC127);
5310 adapter->stats.ptc255 += rd32(E1000_PTC255);
5311 adapter->stats.ptc511 += rd32(E1000_PTC511);
5312 adapter->stats.ptc1023 += rd32(E1000_PTC1023);
5313 adapter->stats.ptc1522 += rd32(E1000_PTC1522);
5315 adapter->stats.mptc += rd32(E1000_MPTC);
5316 adapter->stats.bptc += rd32(E1000_BPTC);
5318 adapter->stats.tpt += rd32(E1000_TPT);
5319 adapter->stats.colc += rd32(E1000_COLC);
5321 adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
5322 /* read internal phy specific stats */
5323 reg = rd32(E1000_CTRL_EXT);
5324 if (!(reg & E1000_CTRL_EXT_LINK_MODE_MASK)) {
5325 adapter->stats.rxerrc += rd32(E1000_RXERRC);
5327 /* this stat has invalid values on i210/i211 */
5328 if ((hw->mac.type != e1000_i210) &&
5329 (hw->mac.type != e1000_i211))
5330 adapter->stats.tncrs += rd32(E1000_TNCRS);
5333 adapter->stats.tsctc += rd32(E1000_TSCTC);
5334 adapter->stats.tsctfc += rd32(E1000_TSCTFC);
5336 adapter->stats.iac += rd32(E1000_IAC);
5337 adapter->stats.icrxoc += rd32(E1000_ICRXOC);
5338 adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
5339 adapter->stats.icrxatc += rd32(E1000_ICRXATC);
5340 adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
5341 adapter->stats.ictxatc += rd32(E1000_ICTXATC);
5342 adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
5343 adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
5344 adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
5346 /* Fill out the OS statistics structure */
5347 net_stats->multicast = adapter->stats.mprc;
5348 net_stats->collisions = adapter->stats.colc;
5352 /* RLEC on some newer hardware can be incorrect so build
5353 * our own version based on RUC and ROC
5355 net_stats->rx_errors = adapter->stats.rxerrc +
5356 adapter->stats.crcerrs + adapter->stats.algnerrc +
5357 adapter->stats.ruc + adapter->stats.roc +
5358 adapter->stats.cexterr;
5359 net_stats->rx_length_errors = adapter->stats.ruc +
5361 net_stats->rx_crc_errors = adapter->stats.crcerrs;
5362 net_stats->rx_frame_errors = adapter->stats.algnerrc;
5363 net_stats->rx_missed_errors = adapter->stats.mpc;
5366 net_stats->tx_errors = adapter->stats.ecol +
5367 adapter->stats.latecol;
5368 net_stats->tx_aborted_errors = adapter->stats.ecol;
5369 net_stats->tx_window_errors = adapter->stats.latecol;
5370 net_stats->tx_carrier_errors = adapter->stats.tncrs;
5372 /* Tx Dropped needs to be maintained elsewhere */
5374 /* Management Stats */
5375 adapter->stats.mgptc += rd32(E1000_MGTPTC);
5376 adapter->stats.mgprc += rd32(E1000_MGTPRC);
5377 adapter->stats.mgpdc += rd32(E1000_MGTPDC);
5380 reg = rd32(E1000_MANC);
5381 if (reg & E1000_MANC_EN_BMC2OS) {
5382 adapter->stats.o2bgptc += rd32(E1000_O2BGPTC);
5383 adapter->stats.o2bspc += rd32(E1000_O2BSPC);
5384 adapter->stats.b2ospc += rd32(E1000_B2OSPC);
5385 adapter->stats.b2ogprc += rd32(E1000_B2OGPRC);
5389 static irqreturn_t igb_msix_other(int irq, void *data)
5391 struct igb_adapter *adapter = data;
5392 struct e1000_hw *hw = &adapter->hw;
5393 u32 icr = rd32(E1000_ICR);
5394 /* reading ICR causes bit 31 of EICR to be cleared */
5396 if (icr & E1000_ICR_DRSTA)
5397 schedule_work(&adapter->reset_task);
5399 if (icr & E1000_ICR_DOUTSYNC) {
5400 /* HW is reporting DMA is out of sync */
5401 adapter->stats.doosync++;
5402 /* The DMA Out of Sync is also indication of a spoof event
5403 * in IOV mode. Check the Wrong VM Behavior register to
5404 * see if it is really a spoof event.
5406 igb_check_wvbr(adapter);
5409 /* Check for a mailbox event */
5410 if (icr & E1000_ICR_VMMB)
5411 igb_msg_task(adapter);
5413 if (icr & E1000_ICR_LSC) {
5414 hw->mac.get_link_status = 1;
5415 /* guard against interrupt when we're going down */
5416 if (!test_bit(__IGB_DOWN, &adapter->state))
5417 mod_timer(&adapter->watchdog_timer, jiffies + 1);
5420 if (icr & E1000_ICR_TS) {
5421 u32 tsicr = rd32(E1000_TSICR);
5423 if (tsicr & E1000_TSICR_TXTS) {
5424 /* acknowledge the interrupt */
5425 wr32(E1000_TSICR, E1000_TSICR_TXTS);
5426 /* retrieve hardware timestamp */
5427 schedule_work(&adapter->ptp_tx_work);
5431 wr32(E1000_EIMS, adapter->eims_other);
5436 static void igb_write_itr(struct igb_q_vector *q_vector)
5438 struct igb_adapter *adapter = q_vector->adapter;
5439 u32 itr_val = q_vector->itr_val & 0x7FFC;
5441 if (!q_vector->set_itr)
5447 if (adapter->hw.mac.type == e1000_82575)
5448 itr_val |= itr_val << 16;
5450 itr_val |= E1000_EITR_CNT_IGNR;
5452 writel(itr_val, q_vector->itr_register);
5453 q_vector->set_itr = 0;
5456 static irqreturn_t igb_msix_ring(int irq, void *data)
5458 struct igb_q_vector *q_vector = data;
5460 /* Write the ITR value calculated from the previous interrupt. */
5461 igb_write_itr(q_vector);
5463 napi_schedule(&q_vector->napi);
5468 #ifdef CONFIG_IGB_DCA
5469 static void igb_update_tx_dca(struct igb_adapter *adapter,
5470 struct igb_ring *tx_ring,
5473 struct e1000_hw *hw = &adapter->hw;
5474 u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
5476 if (hw->mac.type != e1000_82575)
5477 txctrl <<= E1000_DCA_TXCTRL_CPUID_SHIFT;
5479 /* We can enable relaxed ordering for reads, but not writes when
5480 * DCA is enabled. This is due to a known issue in some chipsets
5481 * which will cause the DCA tag to be cleared.
5483 txctrl |= E1000_DCA_TXCTRL_DESC_RRO_EN |
5484 E1000_DCA_TXCTRL_DATA_RRO_EN |
5485 E1000_DCA_TXCTRL_DESC_DCA_EN;
5487 wr32(E1000_DCA_TXCTRL(tx_ring->reg_idx), txctrl);
5490 static void igb_update_rx_dca(struct igb_adapter *adapter,
5491 struct igb_ring *rx_ring,
5494 struct e1000_hw *hw = &adapter->hw;
5495 u32 rxctrl = dca3_get_tag(&adapter->pdev->dev, cpu);
5497 if (hw->mac.type != e1000_82575)
5498 rxctrl <<= E1000_DCA_RXCTRL_CPUID_SHIFT;
5500 /* We can enable relaxed ordering for reads, but not writes when
5501 * DCA is enabled. This is due to a known issue in some chipsets
5502 * which will cause the DCA tag to be cleared.
5504 rxctrl |= E1000_DCA_RXCTRL_DESC_RRO_EN |
5505 E1000_DCA_RXCTRL_DESC_DCA_EN;
5507 wr32(E1000_DCA_RXCTRL(rx_ring->reg_idx), rxctrl);
5510 static void igb_update_dca(struct igb_q_vector *q_vector)
5512 struct igb_adapter *adapter = q_vector->adapter;
5513 int cpu = get_cpu();
5515 if (q_vector->cpu == cpu)
5518 if (q_vector->tx.ring)
5519 igb_update_tx_dca(adapter, q_vector->tx.ring, cpu);
5521 if (q_vector->rx.ring)
5522 igb_update_rx_dca(adapter, q_vector->rx.ring, cpu);
5524 q_vector->cpu = cpu;
5529 static void igb_setup_dca(struct igb_adapter *adapter)
5531 struct e1000_hw *hw = &adapter->hw;
5534 if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
5537 /* Always use CB2 mode, difference is masked in the CB driver. */
5538 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
5540 for (i = 0; i < adapter->num_q_vectors; i++) {
5541 adapter->q_vector[i]->cpu = -1;
5542 igb_update_dca(adapter->q_vector[i]);
5546 static int __igb_notify_dca(struct device *dev, void *data)
5548 struct net_device *netdev = dev_get_drvdata(dev);
5549 struct igb_adapter *adapter = netdev_priv(netdev);
5550 struct pci_dev *pdev = adapter->pdev;
5551 struct e1000_hw *hw = &adapter->hw;
5552 unsigned long event = *(unsigned long *)data;
5555 case DCA_PROVIDER_ADD:
5556 /* if already enabled, don't do it again */
5557 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
5559 if (dca_add_requester(dev) == 0) {
5560 adapter->flags |= IGB_FLAG_DCA_ENABLED;
5561 dev_info(&pdev->dev, "DCA enabled\n");
5562 igb_setup_dca(adapter);
5565 /* Fall Through since DCA is disabled. */
5566 case DCA_PROVIDER_REMOVE:
5567 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
5568 /* without this a class_device is left
5569 * hanging around in the sysfs model
5571 dca_remove_requester(dev);
5572 dev_info(&pdev->dev, "DCA disabled\n");
5573 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
5574 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
5582 static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
5587 ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
5590 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
5592 #endif /* CONFIG_IGB_DCA */
5594 #ifdef CONFIG_PCI_IOV
5595 static int igb_vf_configure(struct igb_adapter *adapter, int vf)
5597 unsigned char mac_addr[ETH_ALEN];
5599 eth_zero_addr(mac_addr);
5600 igb_set_vf_mac(adapter, vf, mac_addr);
5602 /* By default spoof check is enabled for all VFs */
5603 adapter->vf_data[vf].spoofchk_enabled = true;
5609 static void igb_ping_all_vfs(struct igb_adapter *adapter)
5611 struct e1000_hw *hw = &adapter->hw;
5615 for (i = 0 ; i < adapter->vfs_allocated_count; i++) {
5616 ping = E1000_PF_CONTROL_MSG;
5617 if (adapter->vf_data[i].flags & IGB_VF_FLAG_CTS)
5618 ping |= E1000_VT_MSGTYPE_CTS;
5619 igb_write_mbx(hw, &ping, 1, i);
5623 static int igb_set_vf_promisc(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
5625 struct e1000_hw *hw = &adapter->hw;
5626 u32 vmolr = rd32(E1000_VMOLR(vf));
5627 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5629 vf_data->flags &= ~(IGB_VF_FLAG_UNI_PROMISC |
5630 IGB_VF_FLAG_MULTI_PROMISC);
5631 vmolr &= ~(E1000_VMOLR_ROPE | E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
5633 if (*msgbuf & E1000_VF_SET_PROMISC_MULTICAST) {
5634 vmolr |= E1000_VMOLR_MPME;
5635 vf_data->flags |= IGB_VF_FLAG_MULTI_PROMISC;
5636 *msgbuf &= ~E1000_VF_SET_PROMISC_MULTICAST;
5638 /* if we have hashes and we are clearing a multicast promisc
5639 * flag we need to write the hashes to the MTA as this step
5640 * was previously skipped
5642 if (vf_data->num_vf_mc_hashes > 30) {
5643 vmolr |= E1000_VMOLR_MPME;
5644 } else if (vf_data->num_vf_mc_hashes) {
5647 vmolr |= E1000_VMOLR_ROMPE;
5648 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
5649 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
5653 wr32(E1000_VMOLR(vf), vmolr);
5655 /* there are flags left unprocessed, likely not supported */
5656 if (*msgbuf & E1000_VT_MSGINFO_MASK)
5662 static int igb_set_vf_multicasts(struct igb_adapter *adapter,
5663 u32 *msgbuf, u32 vf)
5665 int n = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
5666 u16 *hash_list = (u16 *)&msgbuf[1];
5667 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5670 /* salt away the number of multicast addresses assigned
5671 * to this VF for later use to restore when the PF multi cast
5674 vf_data->num_vf_mc_hashes = n;
5676 /* only up to 30 hash values supported */
5680 /* store the hashes for later use */
5681 for (i = 0; i < n; i++)
5682 vf_data->vf_mc_hashes[i] = hash_list[i];
5684 /* Flush and reset the mta with the new values */
5685 igb_set_rx_mode(adapter->netdev);
5690 static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
5692 struct e1000_hw *hw = &adapter->hw;
5693 struct vf_data_storage *vf_data;
5696 for (i = 0; i < adapter->vfs_allocated_count; i++) {
5697 u32 vmolr = rd32(E1000_VMOLR(i));
5699 vmolr &= ~(E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
5701 vf_data = &adapter->vf_data[i];
5703 if ((vf_data->num_vf_mc_hashes > 30) ||
5704 (vf_data->flags & IGB_VF_FLAG_MULTI_PROMISC)) {
5705 vmolr |= E1000_VMOLR_MPME;
5706 } else if (vf_data->num_vf_mc_hashes) {
5707 vmolr |= E1000_VMOLR_ROMPE;
5708 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
5709 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
5711 wr32(E1000_VMOLR(i), vmolr);
5715 static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
5717 struct e1000_hw *hw = &adapter->hw;
5718 u32 pool_mask, reg, vid;
5721 pool_mask = 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
5723 /* Find the vlan filter for this id */
5724 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5725 reg = rd32(E1000_VLVF(i));
5727 /* remove the vf from the pool */
5730 /* if pool is empty then remove entry from vfta */
5731 if (!(reg & E1000_VLVF_POOLSEL_MASK) &&
5732 (reg & E1000_VLVF_VLANID_ENABLE)) {
5734 vid = reg & E1000_VLVF_VLANID_MASK;
5735 igb_vfta_set(hw, vid, false);
5738 wr32(E1000_VLVF(i), reg);
5741 adapter->vf_data[vf].vlans_enabled = 0;
5744 static s32 igb_vlvf_set(struct igb_adapter *adapter, u32 vid, bool add, u32 vf)
5746 struct e1000_hw *hw = &adapter->hw;
5749 /* The vlvf table only exists on 82576 hardware and newer */
5750 if (hw->mac.type < e1000_82576)
5753 /* we only need to do this if VMDq is enabled */
5754 if (!adapter->vfs_allocated_count)
5757 /* Find the vlan filter for this id */
5758 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5759 reg = rd32(E1000_VLVF(i));
5760 if ((reg & E1000_VLVF_VLANID_ENABLE) &&
5761 vid == (reg & E1000_VLVF_VLANID_MASK))
5766 if (i == E1000_VLVF_ARRAY_SIZE) {
5767 /* Did not find a matching VLAN ID entry that was
5768 * enabled. Search for a free filter entry, i.e.
5769 * one without the enable bit set
5771 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5772 reg = rd32(E1000_VLVF(i));
5773 if (!(reg & E1000_VLVF_VLANID_ENABLE))
5777 if (i < E1000_VLVF_ARRAY_SIZE) {
5778 /* Found an enabled/available entry */
5779 reg |= 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
5781 /* if !enabled we need to set this up in vfta */
5782 if (!(reg & E1000_VLVF_VLANID_ENABLE)) {
5783 /* add VID to filter table */
5784 igb_vfta_set(hw, vid, true);
5785 reg |= E1000_VLVF_VLANID_ENABLE;
5787 reg &= ~E1000_VLVF_VLANID_MASK;
5789 wr32(E1000_VLVF(i), reg);
5791 /* do not modify RLPML for PF devices */
5792 if (vf >= adapter->vfs_allocated_count)
5795 if (!adapter->vf_data[vf].vlans_enabled) {
5798 reg = rd32(E1000_VMOLR(vf));
5799 size = reg & E1000_VMOLR_RLPML_MASK;
5801 reg &= ~E1000_VMOLR_RLPML_MASK;
5803 wr32(E1000_VMOLR(vf), reg);
5806 adapter->vf_data[vf].vlans_enabled++;
5809 if (i < E1000_VLVF_ARRAY_SIZE) {
5810 /* remove vf from the pool */
5811 reg &= ~(1 << (E1000_VLVF_POOLSEL_SHIFT + vf));
5812 /* if pool is empty then remove entry from vfta */
5813 if (!(reg & E1000_VLVF_POOLSEL_MASK)) {
5815 igb_vfta_set(hw, vid, false);
5817 wr32(E1000_VLVF(i), reg);
5819 /* do not modify RLPML for PF devices */
5820 if (vf >= adapter->vfs_allocated_count)
5823 adapter->vf_data[vf].vlans_enabled--;
5824 if (!adapter->vf_data[vf].vlans_enabled) {
5827 reg = rd32(E1000_VMOLR(vf));
5828 size = reg & E1000_VMOLR_RLPML_MASK;
5830 reg &= ~E1000_VMOLR_RLPML_MASK;
5832 wr32(E1000_VMOLR(vf), reg);
5839 static void igb_set_vmvir(struct igb_adapter *adapter, u32 vid, u32 vf)
5841 struct e1000_hw *hw = &adapter->hw;
5844 wr32(E1000_VMVIR(vf), (vid | E1000_VMVIR_VLANA_DEFAULT));
5846 wr32(E1000_VMVIR(vf), 0);
5849 static int igb_ndo_set_vf_vlan(struct net_device *netdev,
5850 int vf, u16 vlan, u8 qos)
5853 struct igb_adapter *adapter = netdev_priv(netdev);
5855 if ((vf >= adapter->vfs_allocated_count) || (vlan > 4095) || (qos > 7))
5858 err = igb_vlvf_set(adapter, vlan, !!vlan, vf);
5861 igb_set_vmvir(adapter, vlan | (qos << VLAN_PRIO_SHIFT), vf);
5862 igb_set_vmolr(adapter, vf, !vlan);
5863 adapter->vf_data[vf].pf_vlan = vlan;
5864 adapter->vf_data[vf].pf_qos = qos;
5865 dev_info(&adapter->pdev->dev,
5866 "Setting VLAN %d, QOS 0x%x on VF %d\n", vlan, qos, vf);
5867 if (test_bit(__IGB_DOWN, &adapter->state)) {
5868 dev_warn(&adapter->pdev->dev,
5869 "The VF VLAN has been set, but the PF device is not up.\n");
5870 dev_warn(&adapter->pdev->dev,
5871 "Bring the PF device up before attempting to use the VF device.\n");
5874 igb_vlvf_set(adapter, adapter->vf_data[vf].pf_vlan,
5876 igb_set_vmvir(adapter, vlan, vf);
5877 igb_set_vmolr(adapter, vf, true);
5878 adapter->vf_data[vf].pf_vlan = 0;
5879 adapter->vf_data[vf].pf_qos = 0;
5885 static int igb_find_vlvf_entry(struct igb_adapter *adapter, int vid)
5887 struct e1000_hw *hw = &adapter->hw;
5891 /* Find the vlan filter for this id */
5892 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5893 reg = rd32(E1000_VLVF(i));
5894 if ((reg & E1000_VLVF_VLANID_ENABLE) &&
5895 vid == (reg & E1000_VLVF_VLANID_MASK))
5899 if (i >= E1000_VLVF_ARRAY_SIZE)
5905 static int igb_set_vf_vlan(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
5907 struct e1000_hw *hw = &adapter->hw;
5908 int add = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
5909 int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
5912 /* If in promiscuous mode we need to make sure the PF also has
5913 * the VLAN filter set.
5915 if (add && (adapter->netdev->flags & IFF_PROMISC))
5916 err = igb_vlvf_set(adapter, vid, add,
5917 adapter->vfs_allocated_count);
5921 err = igb_vlvf_set(adapter, vid, add, vf);
5926 /* Go through all the checks to see if the VLAN filter should
5927 * be wiped completely.
5929 if (!add && (adapter->netdev->flags & IFF_PROMISC)) {
5931 int regndx = igb_find_vlvf_entry(adapter, vid);
5935 /* See if any other pools are set for this VLAN filter
5936 * entry other than the PF.
5938 vlvf = bits = rd32(E1000_VLVF(regndx));
5939 bits &= 1 << (E1000_VLVF_POOLSEL_SHIFT +
5940 adapter->vfs_allocated_count);
5941 /* If the filter was removed then ensure PF pool bit
5942 * is cleared if the PF only added itself to the pool
5943 * because the PF is in promiscuous mode.
5945 if ((vlvf & VLAN_VID_MASK) == vid &&
5946 !test_bit(vid, adapter->active_vlans) &&
5948 igb_vlvf_set(adapter, vid, add,
5949 adapter->vfs_allocated_count);
5956 static inline void igb_vf_reset(struct igb_adapter *adapter, u32 vf)
5958 /* clear flags - except flag that indicates PF has set the MAC */
5959 adapter->vf_data[vf].flags &= IGB_VF_FLAG_PF_SET_MAC;
5960 adapter->vf_data[vf].last_nack = jiffies;
5962 /* reset offloads to defaults */
5963 igb_set_vmolr(adapter, vf, true);
5965 /* reset vlans for device */
5966 igb_clear_vf_vfta(adapter, vf);
5967 if (adapter->vf_data[vf].pf_vlan)
5968 igb_ndo_set_vf_vlan(adapter->netdev, vf,
5969 adapter->vf_data[vf].pf_vlan,
5970 adapter->vf_data[vf].pf_qos);
5972 igb_clear_vf_vfta(adapter, vf);
5974 /* reset multicast table array for vf */
5975 adapter->vf_data[vf].num_vf_mc_hashes = 0;
5977 /* Flush and reset the mta with the new values */
5978 igb_set_rx_mode(adapter->netdev);
5981 static void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
5983 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
5985 /* clear mac address as we were hotplug removed/added */
5986 if (!(adapter->vf_data[vf].flags & IGB_VF_FLAG_PF_SET_MAC))
5987 eth_zero_addr(vf_mac);
5989 /* process remaining reset events */
5990 igb_vf_reset(adapter, vf);
5993 static void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
5995 struct e1000_hw *hw = &adapter->hw;
5996 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
5997 int rar_entry = hw->mac.rar_entry_count - (vf + 1);
5999 u8 *addr = (u8 *)(&msgbuf[1]);
6001 /* process all the same items cleared in a function level reset */
6002 igb_vf_reset(adapter, vf);
6004 /* set vf mac address */
6005 igb_rar_set_qsel(adapter, vf_mac, rar_entry, vf);
6007 /* enable transmit and receive for vf */
6008 reg = rd32(E1000_VFTE);
6009 wr32(E1000_VFTE, reg | (1 << vf));
6010 reg = rd32(E1000_VFRE);
6011 wr32(E1000_VFRE, reg | (1 << vf));
6013 adapter->vf_data[vf].flags |= IGB_VF_FLAG_CTS;
6015 /* reply to reset with ack and vf mac address */
6016 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
6017 memcpy(addr, vf_mac, ETH_ALEN);
6018 igb_write_mbx(hw, msgbuf, 3, vf);
6021 static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf)
6023 /* The VF MAC Address is stored in a packed array of bytes
6024 * starting at the second 32 bit word of the msg array
6026 unsigned char *addr = (char *)&msg[1];
6029 if (is_valid_ether_addr(addr))
6030 err = igb_set_vf_mac(adapter, vf, addr);
6035 static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf)
6037 struct e1000_hw *hw = &adapter->hw;
6038 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
6039 u32 msg = E1000_VT_MSGTYPE_NACK;
6041 /* if device isn't clear to send it shouldn't be reading either */
6042 if (!(vf_data->flags & IGB_VF_FLAG_CTS) &&
6043 time_after(jiffies, vf_data->last_nack + (2 * HZ))) {
6044 igb_write_mbx(hw, &msg, 1, vf);
6045 vf_data->last_nack = jiffies;
6049 static void igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
6051 struct pci_dev *pdev = adapter->pdev;
6052 u32 msgbuf[E1000_VFMAILBOX_SIZE];
6053 struct e1000_hw *hw = &adapter->hw;
6054 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
6057 retval = igb_read_mbx(hw, msgbuf, E1000_VFMAILBOX_SIZE, vf);
6060 /* if receive failed revoke VF CTS stats and restart init */
6061 dev_err(&pdev->dev, "Error receiving message from VF\n");
6062 vf_data->flags &= ~IGB_VF_FLAG_CTS;
6063 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
6068 /* this is a message we already processed, do nothing */
6069 if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
6072 /* until the vf completes a reset it should not be
6073 * allowed to start any configuration.
6075 if (msgbuf[0] == E1000_VF_RESET) {
6076 igb_vf_reset_msg(adapter, vf);
6080 if (!(vf_data->flags & IGB_VF_FLAG_CTS)) {
6081 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
6087 switch ((msgbuf[0] & 0xFFFF)) {
6088 case E1000_VF_SET_MAC_ADDR:
6090 if (!(vf_data->flags & IGB_VF_FLAG_PF_SET_MAC))
6091 retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
6093 dev_warn(&pdev->dev,
6094 "VF %d attempted to override administratively set MAC address\nReload the VF driver to resume operations\n",
6097 case E1000_VF_SET_PROMISC:
6098 retval = igb_set_vf_promisc(adapter, msgbuf, vf);
6100 case E1000_VF_SET_MULTICAST:
6101 retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
6103 case E1000_VF_SET_LPE:
6104 retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf);
6106 case E1000_VF_SET_VLAN:
6108 if (vf_data->pf_vlan)
6109 dev_warn(&pdev->dev,
6110 "VF %d attempted to override administratively set VLAN tag\nReload the VF driver to resume operations\n",
6113 retval = igb_set_vf_vlan(adapter, msgbuf, vf);
6116 dev_err(&pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
6121 msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
6123 /* notify the VF of the results of what it sent us */
6125 msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
6127 msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
6129 igb_write_mbx(hw, msgbuf, 1, vf);
6132 static void igb_msg_task(struct igb_adapter *adapter)
6134 struct e1000_hw *hw = &adapter->hw;
6137 for (vf = 0; vf < adapter->vfs_allocated_count; vf++) {
6138 /* process any reset requests */
6139 if (!igb_check_for_rst(hw, vf))
6140 igb_vf_reset_event(adapter, vf);
6142 /* process any messages pending */
6143 if (!igb_check_for_msg(hw, vf))
6144 igb_rcv_msg_from_vf(adapter, vf);
6146 /* process any acks */
6147 if (!igb_check_for_ack(hw, vf))
6148 igb_rcv_ack_from_vf(adapter, vf);
6153 * igb_set_uta - Set unicast filter table address
6154 * @adapter: board private structure
6156 * The unicast table address is a register array of 32-bit registers.
6157 * The table is meant to be used in a way similar to how the MTA is used
6158 * however due to certain limitations in the hardware it is necessary to
6159 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
6160 * enable bit to allow vlan tag stripping when promiscuous mode is enabled
6162 static void igb_set_uta(struct igb_adapter *adapter)
6164 struct e1000_hw *hw = &adapter->hw;
6167 /* The UTA table only exists on 82576 hardware and newer */
6168 if (hw->mac.type < e1000_82576)
6171 /* we only need to do this if VMDq is enabled */
6172 if (!adapter->vfs_allocated_count)
6175 for (i = 0; i < hw->mac.uta_reg_count; i++)
6176 array_wr32(E1000_UTA, i, ~0);
6180 * igb_intr_msi - Interrupt Handler
6181 * @irq: interrupt number
6182 * @data: pointer to a network interface device structure
6184 static irqreturn_t igb_intr_msi(int irq, void *data)
6186 struct igb_adapter *adapter = data;
6187 struct igb_q_vector *q_vector = adapter->q_vector[0];
6188 struct e1000_hw *hw = &adapter->hw;
6189 /* read ICR disables interrupts using IAM */
6190 u32 icr = rd32(E1000_ICR);
6192 igb_write_itr(q_vector);
6194 if (icr & E1000_ICR_DRSTA)
6195 schedule_work(&adapter->reset_task);
6197 if (icr & E1000_ICR_DOUTSYNC) {
6198 /* HW is reporting DMA is out of sync */
6199 adapter->stats.doosync++;
6202 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
6203 hw->mac.get_link_status = 1;
6204 if (!test_bit(__IGB_DOWN, &adapter->state))
6205 mod_timer(&adapter->watchdog_timer, jiffies + 1);
6208 if (icr & E1000_ICR_TS) {
6209 u32 tsicr = rd32(E1000_TSICR);
6211 if (tsicr & E1000_TSICR_TXTS) {
6212 /* acknowledge the interrupt */
6213 wr32(E1000_TSICR, E1000_TSICR_TXTS);
6214 /* retrieve hardware timestamp */
6215 schedule_work(&adapter->ptp_tx_work);
6219 napi_schedule(&q_vector->napi);
6225 * igb_intr - Legacy Interrupt Handler
6226 * @irq: interrupt number
6227 * @data: pointer to a network interface device structure
6229 static irqreturn_t igb_intr(int irq, void *data)
6231 struct igb_adapter *adapter = data;
6232 struct igb_q_vector *q_vector = adapter->q_vector[0];
6233 struct e1000_hw *hw = &adapter->hw;
6234 /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked. No
6235 * need for the IMC write
6237 u32 icr = rd32(E1000_ICR);
6239 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
6240 * not set, then the adapter didn't send an interrupt
6242 if (!(icr & E1000_ICR_INT_ASSERTED))
6245 igb_write_itr(q_vector);
6247 if (icr & E1000_ICR_DRSTA)
6248 schedule_work(&adapter->reset_task);
6250 if (icr & E1000_ICR_DOUTSYNC) {
6251 /* HW is reporting DMA is out of sync */
6252 adapter->stats.doosync++;
6255 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
6256 hw->mac.get_link_status = 1;
6257 /* guard against interrupt when we're going down */
6258 if (!test_bit(__IGB_DOWN, &adapter->state))
6259 mod_timer(&adapter->watchdog_timer, jiffies + 1);
6262 if (icr & E1000_ICR_TS) {
6263 u32 tsicr = rd32(E1000_TSICR);
6265 if (tsicr & E1000_TSICR_TXTS) {
6266 /* acknowledge the interrupt */
6267 wr32(E1000_TSICR, E1000_TSICR_TXTS);
6268 /* retrieve hardware timestamp */
6269 schedule_work(&adapter->ptp_tx_work);
6273 napi_schedule(&q_vector->napi);
6278 static void igb_ring_irq_enable(struct igb_q_vector *q_vector)
6280 struct igb_adapter *adapter = q_vector->adapter;
6281 struct e1000_hw *hw = &adapter->hw;
6283 if ((q_vector->rx.ring && (adapter->rx_itr_setting & 3)) ||
6284 (!q_vector->rx.ring && (adapter->tx_itr_setting & 3))) {
6285 if ((adapter->num_q_vectors == 1) && !adapter->vf_data)
6286 igb_set_itr(q_vector);
6288 igb_update_ring_itr(q_vector);
6291 if (!test_bit(__IGB_DOWN, &adapter->state)) {
6292 if (adapter->flags & IGB_FLAG_HAS_MSIX)
6293 wr32(E1000_EIMS, q_vector->eims_value);
6295 igb_irq_enable(adapter);
6300 * igb_poll - NAPI Rx polling callback
6301 * @napi: napi polling structure
6302 * @budget: count of how many packets we should handle
6304 static int igb_poll(struct napi_struct *napi, int budget)
6306 struct igb_q_vector *q_vector = container_of(napi,
6307 struct igb_q_vector,
6309 bool clean_complete = true;
6311 #ifdef CONFIG_IGB_DCA
6312 if (q_vector->adapter->flags & IGB_FLAG_DCA_ENABLED)
6313 igb_update_dca(q_vector);
6315 if (q_vector->tx.ring)
6316 clean_complete = igb_clean_tx_irq(q_vector);
6318 if (q_vector->rx.ring)
6319 clean_complete &= igb_clean_rx_irq(q_vector, budget);
6321 /* If all work not completed, return budget and keep polling */
6322 if (!clean_complete)
6325 /* If not enough Rx work done, exit the polling mode */
6326 napi_complete(napi);
6327 igb_ring_irq_enable(q_vector);
6333 * igb_clean_tx_irq - Reclaim resources after transmit completes
6334 * @q_vector: pointer to q_vector containing needed info
6336 * returns true if ring is completely cleaned
6338 static bool igb_clean_tx_irq(struct igb_q_vector *q_vector)
6340 struct igb_adapter *adapter = q_vector->adapter;
6341 struct igb_ring *tx_ring = q_vector->tx.ring;
6342 struct igb_tx_buffer *tx_buffer;
6343 union e1000_adv_tx_desc *tx_desc;
6344 unsigned int total_bytes = 0, total_packets = 0;
6345 unsigned int budget = q_vector->tx.work_limit;
6346 unsigned int i = tx_ring->next_to_clean;
6348 if (test_bit(__IGB_DOWN, &adapter->state))
6351 tx_buffer = &tx_ring->tx_buffer_info[i];
6352 tx_desc = IGB_TX_DESC(tx_ring, i);
6353 i -= tx_ring->count;
6356 union e1000_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
6358 /* if next_to_watch is not set then there is no work pending */
6362 /* prevent any other reads prior to eop_desc */
6363 read_barrier_depends();
6365 /* if DD is not set pending work has not been completed */
6366 if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)))
6369 /* clear next_to_watch to prevent false hangs */
6370 tx_buffer->next_to_watch = NULL;
6372 /* update the statistics for this packet */
6373 total_bytes += tx_buffer->bytecount;
6374 total_packets += tx_buffer->gso_segs;
6377 dev_consume_skb_any(tx_buffer->skb);
6379 /* unmap skb header data */
6380 dma_unmap_single(tx_ring->dev,
6381 dma_unmap_addr(tx_buffer, dma),
6382 dma_unmap_len(tx_buffer, len),
6385 /* clear tx_buffer data */
6386 tx_buffer->skb = NULL;
6387 dma_unmap_len_set(tx_buffer, len, 0);
6389 /* clear last DMA location and unmap remaining buffers */
6390 while (tx_desc != eop_desc) {
6395 i -= tx_ring->count;
6396 tx_buffer = tx_ring->tx_buffer_info;
6397 tx_desc = IGB_TX_DESC(tx_ring, 0);
6400 /* unmap any remaining paged data */
6401 if (dma_unmap_len(tx_buffer, len)) {
6402 dma_unmap_page(tx_ring->dev,
6403 dma_unmap_addr(tx_buffer, dma),
6404 dma_unmap_len(tx_buffer, len),
6406 dma_unmap_len_set(tx_buffer, len, 0);
6410 /* move us one more past the eop_desc for start of next pkt */
6415 i -= tx_ring->count;
6416 tx_buffer = tx_ring->tx_buffer_info;
6417 tx_desc = IGB_TX_DESC(tx_ring, 0);
6420 /* issue prefetch for next Tx descriptor */
6423 /* update budget accounting */
6425 } while (likely(budget));
6427 netdev_tx_completed_queue(txring_txq(tx_ring),
6428 total_packets, total_bytes);
6429 i += tx_ring->count;
6430 tx_ring->next_to_clean = i;
6431 u64_stats_update_begin(&tx_ring->tx_syncp);
6432 tx_ring->tx_stats.bytes += total_bytes;
6433 tx_ring->tx_stats.packets += total_packets;
6434 u64_stats_update_end(&tx_ring->tx_syncp);
6435 q_vector->tx.total_bytes += total_bytes;
6436 q_vector->tx.total_packets += total_packets;
6438 if (test_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags)) {
6439 struct e1000_hw *hw = &adapter->hw;
6441 /* Detect a transmit hang in hardware, this serializes the
6442 * check with the clearing of time_stamp and movement of i
6444 clear_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
6445 if (tx_buffer->next_to_watch &&
6446 time_after(jiffies, tx_buffer->time_stamp +
6447 (adapter->tx_timeout_factor * HZ)) &&
6448 !(rd32(E1000_STATUS) & E1000_STATUS_TXOFF)) {
6450 /* detected Tx unit hang */
6451 dev_err(tx_ring->dev,
6452 "Detected Tx Unit Hang\n"
6456 " next_to_use <%x>\n"
6457 " next_to_clean <%x>\n"
6458 "buffer_info[next_to_clean]\n"
6459 " time_stamp <%lx>\n"
6460 " next_to_watch <%p>\n"
6462 " desc.status <%x>\n",
6463 tx_ring->queue_index,
6464 rd32(E1000_TDH(tx_ring->reg_idx)),
6465 readl(tx_ring->tail),
6466 tx_ring->next_to_use,
6467 tx_ring->next_to_clean,
6468 tx_buffer->time_stamp,
6469 tx_buffer->next_to_watch,
6471 tx_buffer->next_to_watch->wb.status);
6472 netif_stop_subqueue(tx_ring->netdev,
6473 tx_ring->queue_index);
6475 /* we are about to reset, no point in enabling stuff */
6480 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
6481 if (unlikely(total_packets &&
6482 netif_carrier_ok(tx_ring->netdev) &&
6483 igb_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD)) {
6484 /* Make sure that anybody stopping the queue after this
6485 * sees the new next_to_clean.
6488 if (__netif_subqueue_stopped(tx_ring->netdev,
6489 tx_ring->queue_index) &&
6490 !(test_bit(__IGB_DOWN, &adapter->state))) {
6491 netif_wake_subqueue(tx_ring->netdev,
6492 tx_ring->queue_index);
6494 u64_stats_update_begin(&tx_ring->tx_syncp);
6495 tx_ring->tx_stats.restart_queue++;
6496 u64_stats_update_end(&tx_ring->tx_syncp);
6504 * igb_reuse_rx_page - page flip buffer and store it back on the ring
6505 * @rx_ring: rx descriptor ring to store buffers on
6506 * @old_buff: donor buffer to have page reused
6508 * Synchronizes page for reuse by the adapter
6510 static void igb_reuse_rx_page(struct igb_ring *rx_ring,
6511 struct igb_rx_buffer *old_buff)
6513 struct igb_rx_buffer *new_buff;
6514 u16 nta = rx_ring->next_to_alloc;
6516 new_buff = &rx_ring->rx_buffer_info[nta];
6518 /* update, and store next to alloc */
6520 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
6522 /* transfer page from old buffer to new buffer */
6523 *new_buff = *old_buff;
6525 /* sync the buffer for use by the device */
6526 dma_sync_single_range_for_device(rx_ring->dev, old_buff->dma,
6527 old_buff->page_offset,
6532 static bool igb_can_reuse_rx_page(struct igb_rx_buffer *rx_buffer,
6534 unsigned int truesize)
6536 /* avoid re-using remote pages */
6537 if (unlikely(page_to_nid(page) != numa_node_id()))
6540 if (unlikely(page->pfmemalloc))
6543 #if (PAGE_SIZE < 8192)
6544 /* if we are only owner of page we can reuse it */
6545 if (unlikely(page_count(page) != 1))
6548 /* flip page offset to other buffer */
6549 rx_buffer->page_offset ^= IGB_RX_BUFSZ;
6551 /* Even if we own the page, we are not allowed to use atomic_set()
6552 * This would break get_page_unless_zero() users.
6554 atomic_inc(&page->_count);
6556 /* move offset up to the next cache line */
6557 rx_buffer->page_offset += truesize;
6559 if (rx_buffer->page_offset > (PAGE_SIZE - IGB_RX_BUFSZ))
6562 /* bump ref count on page before it is given to the stack */
6570 * igb_add_rx_frag - Add contents of Rx buffer to sk_buff
6571 * @rx_ring: rx descriptor ring to transact packets on
6572 * @rx_buffer: buffer containing page to add
6573 * @rx_desc: descriptor containing length of buffer written by hardware
6574 * @skb: sk_buff to place the data into
6576 * This function will add the data contained in rx_buffer->page to the skb.
6577 * This is done either through a direct copy if the data in the buffer is
6578 * less than the skb header size, otherwise it will just attach the page as
6579 * a frag to the skb.
6581 * The function will then update the page offset if necessary and return
6582 * true if the buffer can be reused by the adapter.
6584 static bool igb_add_rx_frag(struct igb_ring *rx_ring,
6585 struct igb_rx_buffer *rx_buffer,
6586 union e1000_adv_rx_desc *rx_desc,
6587 struct sk_buff *skb)
6589 struct page *page = rx_buffer->page;
6590 unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
6591 #if (PAGE_SIZE < 8192)
6592 unsigned int truesize = IGB_RX_BUFSZ;
6594 unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
6597 if ((size <= IGB_RX_HDR_LEN) && !skb_is_nonlinear(skb)) {
6598 unsigned char *va = page_address(page) + rx_buffer->page_offset;
6600 if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
6601 igb_ptp_rx_pktstamp(rx_ring->q_vector, va, skb);
6602 va += IGB_TS_HDR_LEN;
6603 size -= IGB_TS_HDR_LEN;
6606 memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
6608 /* we can reuse buffer as-is, just make sure it is local */
6609 if (likely((page_to_nid(page) == numa_node_id()) &&
6613 /* this page cannot be reused so discard it */
6618 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
6619 rx_buffer->page_offset, size, truesize);
6621 return igb_can_reuse_rx_page(rx_buffer, page, truesize);
6624 static struct sk_buff *igb_fetch_rx_buffer(struct igb_ring *rx_ring,
6625 union e1000_adv_rx_desc *rx_desc,
6626 struct sk_buff *skb)
6628 struct igb_rx_buffer *rx_buffer;
6631 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
6633 page = rx_buffer->page;
6637 void *page_addr = page_address(page) +
6638 rx_buffer->page_offset;
6640 /* prefetch first cache line of first page */
6641 prefetch(page_addr);
6642 #if L1_CACHE_BYTES < 128
6643 prefetch(page_addr + L1_CACHE_BYTES);
6646 /* allocate a skb to store the frags */
6647 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
6649 if (unlikely(!skb)) {
6650 rx_ring->rx_stats.alloc_failed++;
6654 /* we will be copying header into skb->data in
6655 * pskb_may_pull so it is in our interest to prefetch
6656 * it now to avoid a possible cache miss
6658 prefetchw(skb->data);
6661 /* we are reusing so sync this buffer for CPU use */
6662 dma_sync_single_range_for_cpu(rx_ring->dev,
6664 rx_buffer->page_offset,
6668 /* pull page into skb */
6669 if (igb_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
6670 /* hand second half of page back to the ring */
6671 igb_reuse_rx_page(rx_ring, rx_buffer);
6673 /* we are not reusing the buffer so unmap it */
6674 dma_unmap_page(rx_ring->dev, rx_buffer->dma,
6675 PAGE_SIZE, DMA_FROM_DEVICE);
6678 /* clear contents of rx_buffer */
6679 rx_buffer->page = NULL;
6684 static inline void igb_rx_checksum(struct igb_ring *ring,
6685 union e1000_adv_rx_desc *rx_desc,
6686 struct sk_buff *skb)
6688 skb_checksum_none_assert(skb);
6690 /* Ignore Checksum bit is set */
6691 if (igb_test_staterr(rx_desc, E1000_RXD_STAT_IXSM))
6694 /* Rx checksum disabled via ethtool */
6695 if (!(ring->netdev->features & NETIF_F_RXCSUM))
6698 /* TCP/UDP checksum error bit is set */
6699 if (igb_test_staterr(rx_desc,
6700 E1000_RXDEXT_STATERR_TCPE |
6701 E1000_RXDEXT_STATERR_IPE)) {
6702 /* work around errata with sctp packets where the TCPE aka
6703 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
6704 * packets, (aka let the stack check the crc32c)
6706 if (!((skb->len == 60) &&
6707 test_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags))) {
6708 u64_stats_update_begin(&ring->rx_syncp);
6709 ring->rx_stats.csum_err++;
6710 u64_stats_update_end(&ring->rx_syncp);
6712 /* let the stack verify checksum errors */
6715 /* It must be a TCP or UDP packet with a valid checksum */
6716 if (igb_test_staterr(rx_desc, E1000_RXD_STAT_TCPCS |
6717 E1000_RXD_STAT_UDPCS))
6718 skb->ip_summed = CHECKSUM_UNNECESSARY;
6720 dev_dbg(ring->dev, "cksum success: bits %08X\n",
6721 le32_to_cpu(rx_desc->wb.upper.status_error));
6724 static inline void igb_rx_hash(struct igb_ring *ring,
6725 union e1000_adv_rx_desc *rx_desc,
6726 struct sk_buff *skb)
6728 if (ring->netdev->features & NETIF_F_RXHASH)
6730 le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
6735 * igb_is_non_eop - process handling of non-EOP buffers
6736 * @rx_ring: Rx ring being processed
6737 * @rx_desc: Rx descriptor for current buffer
6738 * @skb: current socket buffer containing buffer in progress
6740 * This function updates next to clean. If the buffer is an EOP buffer
6741 * this function exits returning false, otherwise it will place the
6742 * sk_buff in the next buffer to be chained and return true indicating
6743 * that this is in fact a non-EOP buffer.
6745 static bool igb_is_non_eop(struct igb_ring *rx_ring,
6746 union e1000_adv_rx_desc *rx_desc)
6748 u32 ntc = rx_ring->next_to_clean + 1;
6750 /* fetch, update, and store next to clean */
6751 ntc = (ntc < rx_ring->count) ? ntc : 0;
6752 rx_ring->next_to_clean = ntc;
6754 prefetch(IGB_RX_DESC(rx_ring, ntc));
6756 if (likely(igb_test_staterr(rx_desc, E1000_RXD_STAT_EOP)))
6763 * igb_pull_tail - igb specific version of skb_pull_tail
6764 * @rx_ring: rx descriptor ring packet is being transacted on
6765 * @rx_desc: pointer to the EOP Rx descriptor
6766 * @skb: pointer to current skb being adjusted
6768 * This function is an igb specific version of __pskb_pull_tail. The
6769 * main difference between this version and the original function is that
6770 * this function can make several assumptions about the state of things
6771 * that allow for significant optimizations versus the standard function.
6772 * As a result we can do things like drop a frag and maintain an accurate
6773 * truesize for the skb.
6775 static void igb_pull_tail(struct igb_ring *rx_ring,
6776 union e1000_adv_rx_desc *rx_desc,
6777 struct sk_buff *skb)
6779 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
6781 unsigned int pull_len;
6783 /* it is valid to use page_address instead of kmap since we are
6784 * working with pages allocated out of the lomem pool per
6785 * alloc_page(GFP_ATOMIC)
6787 va = skb_frag_address(frag);
6789 if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
6790 /* retrieve timestamp from buffer */
6791 igb_ptp_rx_pktstamp(rx_ring->q_vector, va, skb);
6793 /* update pointers to remove timestamp header */
6794 skb_frag_size_sub(frag, IGB_TS_HDR_LEN);
6795 frag->page_offset += IGB_TS_HDR_LEN;
6796 skb->data_len -= IGB_TS_HDR_LEN;
6797 skb->len -= IGB_TS_HDR_LEN;
6799 /* move va to start of packet data */
6800 va += IGB_TS_HDR_LEN;
6803 /* we need the header to contain the greater of either ETH_HLEN or
6804 * 60 bytes if the skb->len is less than 60 for skb_pad.
6806 pull_len = eth_get_headlen(va, IGB_RX_HDR_LEN);
6808 /* align pull length to size of long to optimize memcpy performance */
6809 skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
6811 /* update all of the pointers */
6812 skb_frag_size_sub(frag, pull_len);
6813 frag->page_offset += pull_len;
6814 skb->data_len -= pull_len;
6815 skb->tail += pull_len;
6819 * igb_cleanup_headers - Correct corrupted or empty headers
6820 * @rx_ring: rx descriptor ring packet is being transacted on
6821 * @rx_desc: pointer to the EOP Rx descriptor
6822 * @skb: pointer to current skb being fixed
6824 * Address the case where we are pulling data in on pages only
6825 * and as such no data is present in the skb header.
6827 * In addition if skb is not at least 60 bytes we need to pad it so that
6828 * it is large enough to qualify as a valid Ethernet frame.
6830 * Returns true if an error was encountered and skb was freed.
6832 static bool igb_cleanup_headers(struct igb_ring *rx_ring,
6833 union e1000_adv_rx_desc *rx_desc,
6834 struct sk_buff *skb)
6836 if (unlikely((igb_test_staterr(rx_desc,
6837 E1000_RXDEXT_ERR_FRAME_ERR_MASK)))) {
6838 struct net_device *netdev = rx_ring->netdev;
6839 if (!(netdev->features & NETIF_F_RXALL)) {
6840 dev_kfree_skb_any(skb);
6845 /* place header in linear portion of buffer */
6846 if (skb_is_nonlinear(skb))
6847 igb_pull_tail(rx_ring, rx_desc, skb);
6849 /* if skb_pad returns an error the skb was freed */
6850 if (unlikely(skb->len < 60)) {
6851 int pad_len = 60 - skb->len;
6853 if (skb_pad(skb, pad_len))
6855 __skb_put(skb, pad_len);
6862 * igb_process_skb_fields - Populate skb header fields from Rx descriptor
6863 * @rx_ring: rx descriptor ring packet is being transacted on
6864 * @rx_desc: pointer to the EOP Rx descriptor
6865 * @skb: pointer to current skb being populated
6867 * This function checks the ring, descriptor, and packet information in
6868 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
6869 * other fields within the skb.
6871 static void igb_process_skb_fields(struct igb_ring *rx_ring,
6872 union e1000_adv_rx_desc *rx_desc,
6873 struct sk_buff *skb)
6875 struct net_device *dev = rx_ring->netdev;
6877 igb_rx_hash(rx_ring, rx_desc, skb);
6879 igb_rx_checksum(rx_ring, rx_desc, skb);
6881 if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TS) &&
6882 !igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP))
6883 igb_ptp_rx_rgtstamp(rx_ring->q_vector, skb);
6885 if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
6886 igb_test_staterr(rx_desc, E1000_RXD_STAT_VP)) {
6889 if (igb_test_staterr(rx_desc, E1000_RXDEXT_STATERR_LB) &&
6890 test_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &rx_ring->flags))
6891 vid = be16_to_cpu(rx_desc->wb.upper.vlan);
6893 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
6895 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
6898 skb_record_rx_queue(skb, rx_ring->queue_index);
6900 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
6903 static bool igb_clean_rx_irq(struct igb_q_vector *q_vector, const int budget)
6905 struct igb_ring *rx_ring = q_vector->rx.ring;
6906 struct sk_buff *skb = rx_ring->skb;
6907 unsigned int total_bytes = 0, total_packets = 0;
6908 u16 cleaned_count = igb_desc_unused(rx_ring);
6910 while (likely(total_packets < budget)) {
6911 union e1000_adv_rx_desc *rx_desc;
6913 /* return some buffers to hardware, one at a time is too slow */
6914 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
6915 igb_alloc_rx_buffers(rx_ring, cleaned_count);
6919 rx_desc = IGB_RX_DESC(rx_ring, rx_ring->next_to_clean);
6921 if (!igb_test_staterr(rx_desc, E1000_RXD_STAT_DD))
6924 /* This memory barrier is needed to keep us from reading
6925 * any other fields out of the rx_desc until we know the
6926 * RXD_STAT_DD bit is set
6930 /* retrieve a buffer from the ring */
6931 skb = igb_fetch_rx_buffer(rx_ring, rx_desc, skb);
6933 /* exit if we failed to retrieve a buffer */
6939 /* fetch next buffer in frame if non-eop */
6940 if (igb_is_non_eop(rx_ring, rx_desc))
6943 /* verify the packet layout is correct */
6944 if (igb_cleanup_headers(rx_ring, rx_desc, skb)) {
6949 /* probably a little skewed due to removing CRC */
6950 total_bytes += skb->len;
6952 /* populate checksum, timestamp, VLAN, and protocol */
6953 igb_process_skb_fields(rx_ring, rx_desc, skb);
6955 napi_gro_receive(&q_vector->napi, skb);
6957 /* reset skb pointer */
6960 /* update budget accounting */
6964 /* place incomplete frames back on ring for completion */
6967 u64_stats_update_begin(&rx_ring->rx_syncp);
6968 rx_ring->rx_stats.packets += total_packets;
6969 rx_ring->rx_stats.bytes += total_bytes;
6970 u64_stats_update_end(&rx_ring->rx_syncp);
6971 q_vector->rx.total_packets += total_packets;
6972 q_vector->rx.total_bytes += total_bytes;
6975 igb_alloc_rx_buffers(rx_ring, cleaned_count);
6977 return total_packets < budget;
6980 static bool igb_alloc_mapped_page(struct igb_ring *rx_ring,
6981 struct igb_rx_buffer *bi)
6983 struct page *page = bi->page;
6986 /* since we are recycling buffers we should seldom need to alloc */
6990 /* alloc new page for storage */
6991 page = dev_alloc_page();
6992 if (unlikely(!page)) {
6993 rx_ring->rx_stats.alloc_failed++;
6997 /* map page for use */
6998 dma = dma_map_page(rx_ring->dev, page, 0, PAGE_SIZE, DMA_FROM_DEVICE);
7000 /* if mapping failed free memory back to system since
7001 * there isn't much point in holding memory we can't use
7003 if (dma_mapping_error(rx_ring->dev, dma)) {
7006 rx_ring->rx_stats.alloc_failed++;
7012 bi->page_offset = 0;
7018 * igb_alloc_rx_buffers - Replace used receive buffers; packet split
7019 * @adapter: address of board private structure
7021 void igb_alloc_rx_buffers(struct igb_ring *rx_ring, u16 cleaned_count)
7023 union e1000_adv_rx_desc *rx_desc;
7024 struct igb_rx_buffer *bi;
7025 u16 i = rx_ring->next_to_use;
7031 rx_desc = IGB_RX_DESC(rx_ring, i);
7032 bi = &rx_ring->rx_buffer_info[i];
7033 i -= rx_ring->count;
7036 if (!igb_alloc_mapped_page(rx_ring, bi))
7039 /* Refresh the desc even if buffer_addrs didn't change
7040 * because each write-back erases this info.
7042 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
7048 rx_desc = IGB_RX_DESC(rx_ring, 0);
7049 bi = rx_ring->rx_buffer_info;
7050 i -= rx_ring->count;
7053 /* clear the hdr_addr for the next_to_use descriptor */
7054 rx_desc->read.hdr_addr = 0;
7057 } while (cleaned_count);
7059 i += rx_ring->count;
7061 if (rx_ring->next_to_use != i) {
7062 /* record the next descriptor to use */
7063 rx_ring->next_to_use = i;
7065 /* update next to alloc since we have filled the ring */
7066 rx_ring->next_to_alloc = i;
7068 /* Force memory writes to complete before letting h/w
7069 * know there are new descriptors to fetch. (Only
7070 * applicable for weak-ordered memory model archs,
7074 writel(i, rx_ring->tail);
7084 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
7086 struct igb_adapter *adapter = netdev_priv(netdev);
7087 struct mii_ioctl_data *data = if_mii(ifr);
7089 if (adapter->hw.phy.media_type != e1000_media_type_copper)
7094 data->phy_id = adapter->hw.phy.addr;
7097 if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
7114 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
7120 return igb_mii_ioctl(netdev, ifr, cmd);
7122 return igb_ptp_get_ts_config(netdev, ifr);
7124 return igb_ptp_set_ts_config(netdev, ifr);
7130 void igb_read_pci_cfg(struct e1000_hw *hw, u32 reg, u16 *value)
7132 struct igb_adapter *adapter = hw->back;
7134 pci_read_config_word(adapter->pdev, reg, value);
7137 void igb_write_pci_cfg(struct e1000_hw *hw, u32 reg, u16 *value)
7139 struct igb_adapter *adapter = hw->back;
7141 pci_write_config_word(adapter->pdev, reg, *value);
7144 s32 igb_read_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
7146 struct igb_adapter *adapter = hw->back;
7148 if (pcie_capability_read_word(adapter->pdev, reg, value))
7149 return -E1000_ERR_CONFIG;
7154 s32 igb_write_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
7156 struct igb_adapter *adapter = hw->back;
7158 if (pcie_capability_write_word(adapter->pdev, reg, *value))
7159 return -E1000_ERR_CONFIG;
7164 static void igb_vlan_mode(struct net_device *netdev, netdev_features_t features)
7166 struct igb_adapter *adapter = netdev_priv(netdev);
7167 struct e1000_hw *hw = &adapter->hw;
7169 bool enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
7172 /* enable VLAN tag insert/strip */
7173 ctrl = rd32(E1000_CTRL);
7174 ctrl |= E1000_CTRL_VME;
7175 wr32(E1000_CTRL, ctrl);
7177 /* Disable CFI check */
7178 rctl = rd32(E1000_RCTL);
7179 rctl &= ~E1000_RCTL_CFIEN;
7180 wr32(E1000_RCTL, rctl);
7182 /* disable VLAN tag insert/strip */
7183 ctrl = rd32(E1000_CTRL);
7184 ctrl &= ~E1000_CTRL_VME;
7185 wr32(E1000_CTRL, ctrl);
7188 igb_rlpml_set(adapter);
7191 static int igb_vlan_rx_add_vid(struct net_device *netdev,
7192 __be16 proto, u16 vid)
7194 struct igb_adapter *adapter = netdev_priv(netdev);
7195 struct e1000_hw *hw = &adapter->hw;
7196 int pf_id = adapter->vfs_allocated_count;
7198 /* attempt to add filter to vlvf array */
7199 igb_vlvf_set(adapter, vid, true, pf_id);
7201 /* add the filter since PF can receive vlans w/o entry in vlvf */
7202 igb_vfta_set(hw, vid, true);
7204 set_bit(vid, adapter->active_vlans);
7209 static int igb_vlan_rx_kill_vid(struct net_device *netdev,
7210 __be16 proto, u16 vid)
7212 struct igb_adapter *adapter = netdev_priv(netdev);
7213 struct e1000_hw *hw = &adapter->hw;
7214 int pf_id = adapter->vfs_allocated_count;
7217 /* remove vlan from VLVF table array */
7218 err = igb_vlvf_set(adapter, vid, false, pf_id);
7220 /* if vid was not present in VLVF just remove it from table */
7222 igb_vfta_set(hw, vid, false);
7224 clear_bit(vid, adapter->active_vlans);
7229 static void igb_restore_vlan(struct igb_adapter *adapter)
7233 igb_vlan_mode(adapter->netdev, adapter->netdev->features);
7235 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
7236 igb_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
7239 int igb_set_spd_dplx(struct igb_adapter *adapter, u32 spd, u8 dplx)
7241 struct pci_dev *pdev = adapter->pdev;
7242 struct e1000_mac_info *mac = &adapter->hw.mac;
7246 /* Make sure dplx is at most 1 bit and lsb of speed is not set
7247 * for the switch() below to work
7249 if ((spd & 1) || (dplx & ~1))
7252 /* Fiber NIC's only allow 1000 gbps Full duplex
7253 * and 100Mbps Full duplex for 100baseFx sfp
7255 if (adapter->hw.phy.media_type == e1000_media_type_internal_serdes) {
7256 switch (spd + dplx) {
7257 case SPEED_10 + DUPLEX_HALF:
7258 case SPEED_10 + DUPLEX_FULL:
7259 case SPEED_100 + DUPLEX_HALF:
7266 switch (spd + dplx) {
7267 case SPEED_10 + DUPLEX_HALF:
7268 mac->forced_speed_duplex = ADVERTISE_10_HALF;
7270 case SPEED_10 + DUPLEX_FULL:
7271 mac->forced_speed_duplex = ADVERTISE_10_FULL;
7273 case SPEED_100 + DUPLEX_HALF:
7274 mac->forced_speed_duplex = ADVERTISE_100_HALF;
7276 case SPEED_100 + DUPLEX_FULL:
7277 mac->forced_speed_duplex = ADVERTISE_100_FULL;
7279 case SPEED_1000 + DUPLEX_FULL:
7281 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
7283 case SPEED_1000 + DUPLEX_HALF: /* not supported */
7288 /* clear MDI, MDI(-X) override is only allowed when autoneg enabled */
7289 adapter->hw.phy.mdix = AUTO_ALL_MODES;
7294 dev_err(&pdev->dev, "Unsupported Speed/Duplex configuration\n");
7298 static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake,
7301 struct net_device *netdev = pci_get_drvdata(pdev);
7302 struct igb_adapter *adapter = netdev_priv(netdev);
7303 struct e1000_hw *hw = &adapter->hw;
7304 u32 ctrl, rctl, status;
7305 u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
7310 netif_device_detach(netdev);
7312 if (netif_running(netdev))
7313 __igb_close(netdev, true);
7315 igb_clear_interrupt_scheme(adapter);
7318 retval = pci_save_state(pdev);
7323 status = rd32(E1000_STATUS);
7324 if (status & E1000_STATUS_LU)
7325 wufc &= ~E1000_WUFC_LNKC;
7328 igb_setup_rctl(adapter);
7329 igb_set_rx_mode(netdev);
7331 /* turn on all-multi mode if wake on multicast is enabled */
7332 if (wufc & E1000_WUFC_MC) {
7333 rctl = rd32(E1000_RCTL);
7334 rctl |= E1000_RCTL_MPE;
7335 wr32(E1000_RCTL, rctl);
7338 ctrl = rd32(E1000_CTRL);
7339 /* advertise wake from D3Cold */
7340 #define E1000_CTRL_ADVD3WUC 0x00100000
7341 /* phy power management enable */
7342 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
7343 ctrl |= E1000_CTRL_ADVD3WUC;
7344 wr32(E1000_CTRL, ctrl);
7346 /* Allow time for pending master requests to run */
7347 igb_disable_pcie_master(hw);
7349 wr32(E1000_WUC, E1000_WUC_PME_EN);
7350 wr32(E1000_WUFC, wufc);
7353 wr32(E1000_WUFC, 0);
7356 *enable_wake = wufc || adapter->en_mng_pt;
7358 igb_power_down_link(adapter);
7360 igb_power_up_link(adapter);
7362 /* Release control of h/w to f/w. If f/w is AMT enabled, this
7363 * would have already happened in close and is redundant.
7365 igb_release_hw_control(adapter);
7367 pci_disable_device(pdev);
7373 #ifdef CONFIG_PM_SLEEP
7374 static int igb_suspend(struct device *dev)
7378 struct pci_dev *pdev = to_pci_dev(dev);
7380 retval = __igb_shutdown(pdev, &wake, 0);
7385 pci_prepare_to_sleep(pdev);
7387 pci_wake_from_d3(pdev, false);
7388 pci_set_power_state(pdev, PCI_D3hot);
7393 #endif /* CONFIG_PM_SLEEP */
7395 static int igb_resume(struct device *dev)
7397 struct pci_dev *pdev = to_pci_dev(dev);
7398 struct net_device *netdev = pci_get_drvdata(pdev);
7399 struct igb_adapter *adapter = netdev_priv(netdev);
7400 struct e1000_hw *hw = &adapter->hw;
7403 pci_set_power_state(pdev, PCI_D0);
7404 pci_restore_state(pdev);
7405 pci_save_state(pdev);
7407 err = pci_enable_device_mem(pdev);
7410 "igb: Cannot enable PCI device from suspend\n");
7413 pci_set_master(pdev);
7415 pci_enable_wake(pdev, PCI_D3hot, 0);
7416 pci_enable_wake(pdev, PCI_D3cold, 0);
7418 if (igb_init_interrupt_scheme(adapter, true)) {
7419 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
7425 /* let the f/w know that the h/w is now under the control of the
7428 igb_get_hw_control(adapter);
7430 wr32(E1000_WUS, ~0);
7432 if (netdev->flags & IFF_UP) {
7434 err = __igb_open(netdev, true);
7440 netif_device_attach(netdev);
7444 #ifdef CONFIG_PM_RUNTIME
7445 static int igb_runtime_idle(struct device *dev)
7447 struct pci_dev *pdev = to_pci_dev(dev);
7448 struct net_device *netdev = pci_get_drvdata(pdev);
7449 struct igb_adapter *adapter = netdev_priv(netdev);
7451 if (!igb_has_link(adapter))
7452 pm_schedule_suspend(dev, MSEC_PER_SEC * 5);
7457 static int igb_runtime_suspend(struct device *dev)
7459 struct pci_dev *pdev = to_pci_dev(dev);
7463 retval = __igb_shutdown(pdev, &wake, 1);
7468 pci_prepare_to_sleep(pdev);
7470 pci_wake_from_d3(pdev, false);
7471 pci_set_power_state(pdev, PCI_D3hot);
7477 static int igb_runtime_resume(struct device *dev)
7479 return igb_resume(dev);
7481 #endif /* CONFIG_PM_RUNTIME */
7484 static void igb_shutdown(struct pci_dev *pdev)
7488 __igb_shutdown(pdev, &wake, 0);
7490 if (system_state == SYSTEM_POWER_OFF) {
7491 pci_wake_from_d3(pdev, wake);
7492 pci_set_power_state(pdev, PCI_D3hot);
7496 #ifdef CONFIG_PCI_IOV
7497 static int igb_sriov_reinit(struct pci_dev *dev)
7499 struct net_device *netdev = pci_get_drvdata(dev);
7500 struct igb_adapter *adapter = netdev_priv(netdev);
7501 struct pci_dev *pdev = adapter->pdev;
7505 if (netif_running(netdev))
7510 igb_clear_interrupt_scheme(adapter);
7512 igb_init_queue_configuration(adapter);
7514 if (igb_init_interrupt_scheme(adapter, true)) {
7515 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
7519 if (netif_running(netdev))
7527 static int igb_pci_disable_sriov(struct pci_dev *dev)
7529 int err = igb_disable_sriov(dev);
7532 err = igb_sriov_reinit(dev);
7537 static int igb_pci_enable_sriov(struct pci_dev *dev, int num_vfs)
7539 int err = igb_enable_sriov(dev, num_vfs);
7544 err = igb_sriov_reinit(dev);
7553 static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs)
7555 #ifdef CONFIG_PCI_IOV
7557 return igb_pci_disable_sriov(dev);
7559 return igb_pci_enable_sriov(dev, num_vfs);
7564 #ifdef CONFIG_NET_POLL_CONTROLLER
7565 /* Polling 'interrupt' - used by things like netconsole to send skbs
7566 * without having to re-enable interrupts. It's not called while
7567 * the interrupt routine is executing.
7569 static void igb_netpoll(struct net_device *netdev)
7571 struct igb_adapter *adapter = netdev_priv(netdev);
7572 struct e1000_hw *hw = &adapter->hw;
7573 struct igb_q_vector *q_vector;
7576 for (i = 0; i < adapter->num_q_vectors; i++) {
7577 q_vector = adapter->q_vector[i];
7578 if (adapter->flags & IGB_FLAG_HAS_MSIX)
7579 wr32(E1000_EIMC, q_vector->eims_value);
7581 igb_irq_disable(adapter);
7582 napi_schedule(&q_vector->napi);
7585 #endif /* CONFIG_NET_POLL_CONTROLLER */
7588 * igb_io_error_detected - called when PCI error is detected
7589 * @pdev: Pointer to PCI device
7590 * @state: The current pci connection state
7592 * This function is called after a PCI bus error affecting
7593 * this device has been detected.
7595 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
7596 pci_channel_state_t state)
7598 struct net_device *netdev = pci_get_drvdata(pdev);
7599 struct igb_adapter *adapter = netdev_priv(netdev);
7601 netif_device_detach(netdev);
7603 if (state == pci_channel_io_perm_failure)
7604 return PCI_ERS_RESULT_DISCONNECT;
7606 if (netif_running(netdev))
7608 pci_disable_device(pdev);
7610 /* Request a slot slot reset. */
7611 return PCI_ERS_RESULT_NEED_RESET;
7615 * igb_io_slot_reset - called after the pci bus has been reset.
7616 * @pdev: Pointer to PCI device
7618 * Restart the card from scratch, as if from a cold-boot. Implementation
7619 * resembles the first-half of the igb_resume routine.
7621 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
7623 struct net_device *netdev = pci_get_drvdata(pdev);
7624 struct igb_adapter *adapter = netdev_priv(netdev);
7625 struct e1000_hw *hw = &adapter->hw;
7626 pci_ers_result_t result;
7629 if (pci_enable_device_mem(pdev)) {
7631 "Cannot re-enable PCI device after reset.\n");
7632 result = PCI_ERS_RESULT_DISCONNECT;
7634 pci_set_master(pdev);
7635 pci_restore_state(pdev);
7636 pci_save_state(pdev);
7638 pci_enable_wake(pdev, PCI_D3hot, 0);
7639 pci_enable_wake(pdev, PCI_D3cold, 0);
7642 wr32(E1000_WUS, ~0);
7643 result = PCI_ERS_RESULT_RECOVERED;
7646 err = pci_cleanup_aer_uncorrect_error_status(pdev);
7649 "pci_cleanup_aer_uncorrect_error_status failed 0x%0x\n",
7651 /* non-fatal, continue */
7658 * igb_io_resume - called when traffic can start flowing again.
7659 * @pdev: Pointer to PCI device
7661 * This callback is called when the error recovery driver tells us that
7662 * its OK to resume normal operation. Implementation resembles the
7663 * second-half of the igb_resume routine.
7665 static void igb_io_resume(struct pci_dev *pdev)
7667 struct net_device *netdev = pci_get_drvdata(pdev);
7668 struct igb_adapter *adapter = netdev_priv(netdev);
7670 if (netif_running(netdev)) {
7671 if (igb_up(adapter)) {
7672 dev_err(&pdev->dev, "igb_up failed after reset\n");
7677 netif_device_attach(netdev);
7679 /* let the f/w know that the h/w is now under the control of the
7682 igb_get_hw_control(adapter);
7685 static void igb_rar_set_qsel(struct igb_adapter *adapter, u8 *addr, u32 index,
7688 u32 rar_low, rar_high;
7689 struct e1000_hw *hw = &adapter->hw;
7691 /* HW expects these in little endian so we reverse the byte order
7692 * from network order (big endian) to little endian
7694 rar_low = ((u32) addr[0] | ((u32) addr[1] << 8) |
7695 ((u32) addr[2] << 16) | ((u32) addr[3] << 24));
7696 rar_high = ((u32) addr[4] | ((u32) addr[5] << 8));
7698 /* Indicate to hardware the Address is Valid. */
7699 rar_high |= E1000_RAH_AV;
7701 if (hw->mac.type == e1000_82575)
7702 rar_high |= E1000_RAH_POOL_1 * qsel;
7704 rar_high |= E1000_RAH_POOL_1 << qsel;
7706 wr32(E1000_RAL(index), rar_low);
7708 wr32(E1000_RAH(index), rar_high);
7712 static int igb_set_vf_mac(struct igb_adapter *adapter,
7713 int vf, unsigned char *mac_addr)
7715 struct e1000_hw *hw = &adapter->hw;
7716 /* VF MAC addresses start at end of receive addresses and moves
7717 * towards the first, as a result a collision should not be possible
7719 int rar_entry = hw->mac.rar_entry_count - (vf + 1);
7721 memcpy(adapter->vf_data[vf].vf_mac_addresses, mac_addr, ETH_ALEN);
7723 igb_rar_set_qsel(adapter, mac_addr, rar_entry, vf);
7728 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
7730 struct igb_adapter *adapter = netdev_priv(netdev);
7731 if (!is_valid_ether_addr(mac) || (vf >= adapter->vfs_allocated_count))
7733 adapter->vf_data[vf].flags |= IGB_VF_FLAG_PF_SET_MAC;
7734 dev_info(&adapter->pdev->dev, "setting MAC %pM on VF %d\n", mac, vf);
7735 dev_info(&adapter->pdev->dev,
7736 "Reload the VF driver to make this change effective.");
7737 if (test_bit(__IGB_DOWN, &adapter->state)) {
7738 dev_warn(&adapter->pdev->dev,
7739 "The VF MAC address has been set, but the PF device is not up.\n");
7740 dev_warn(&adapter->pdev->dev,
7741 "Bring the PF device up before attempting to use the VF device.\n");
7743 return igb_set_vf_mac(adapter, vf, mac);
7746 static int igb_link_mbps(int internal_link_speed)
7748 switch (internal_link_speed) {
7758 static void igb_set_vf_rate_limit(struct e1000_hw *hw, int vf, int tx_rate,
7765 /* Calculate the rate factor values to set */
7766 rf_int = link_speed / tx_rate;
7767 rf_dec = (link_speed - (rf_int * tx_rate));
7768 rf_dec = (rf_dec * (1 << E1000_RTTBCNRC_RF_INT_SHIFT)) /
7771 bcnrc_val = E1000_RTTBCNRC_RS_ENA;
7772 bcnrc_val |= ((rf_int << E1000_RTTBCNRC_RF_INT_SHIFT) &
7773 E1000_RTTBCNRC_RF_INT_MASK);
7774 bcnrc_val |= (rf_dec & E1000_RTTBCNRC_RF_DEC_MASK);
7779 wr32(E1000_RTTDQSEL, vf); /* vf X uses queue X */
7780 /* Set global transmit compensation time to the MMW_SIZE in RTTBCNRM
7781 * register. MMW_SIZE=0x014 if 9728-byte jumbo is supported.
7783 wr32(E1000_RTTBCNRM, 0x14);
7784 wr32(E1000_RTTBCNRC, bcnrc_val);
7787 static void igb_check_vf_rate_limit(struct igb_adapter *adapter)
7789 int actual_link_speed, i;
7790 bool reset_rate = false;
7792 /* VF TX rate limit was not set or not supported */
7793 if ((adapter->vf_rate_link_speed == 0) ||
7794 (adapter->hw.mac.type != e1000_82576))
7797 actual_link_speed = igb_link_mbps(adapter->link_speed);
7798 if (actual_link_speed != adapter->vf_rate_link_speed) {
7800 adapter->vf_rate_link_speed = 0;
7801 dev_info(&adapter->pdev->dev,
7802 "Link speed has been changed. VF Transmit rate is disabled\n");
7805 for (i = 0; i < adapter->vfs_allocated_count; i++) {
7807 adapter->vf_data[i].tx_rate = 0;
7809 igb_set_vf_rate_limit(&adapter->hw, i,
7810 adapter->vf_data[i].tx_rate,
7815 static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf,
7816 int min_tx_rate, int max_tx_rate)
7818 struct igb_adapter *adapter = netdev_priv(netdev);
7819 struct e1000_hw *hw = &adapter->hw;
7820 int actual_link_speed;
7822 if (hw->mac.type != e1000_82576)
7828 actual_link_speed = igb_link_mbps(adapter->link_speed);
7829 if ((vf >= adapter->vfs_allocated_count) ||
7830 (!(rd32(E1000_STATUS) & E1000_STATUS_LU)) ||
7831 (max_tx_rate < 0) ||
7832 (max_tx_rate > actual_link_speed))
7835 adapter->vf_rate_link_speed = actual_link_speed;
7836 adapter->vf_data[vf].tx_rate = (u16)max_tx_rate;
7837 igb_set_vf_rate_limit(hw, vf, max_tx_rate, actual_link_speed);
7842 static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf,
7845 struct igb_adapter *adapter = netdev_priv(netdev);
7846 struct e1000_hw *hw = &adapter->hw;
7847 u32 reg_val, reg_offset;
7849 if (!adapter->vfs_allocated_count)
7852 if (vf >= adapter->vfs_allocated_count)
7855 reg_offset = (hw->mac.type == e1000_82576) ? E1000_DTXSWC : E1000_TXSWC;
7856 reg_val = rd32(reg_offset);
7858 reg_val |= ((1 << vf) |
7859 (1 << (vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT)));
7861 reg_val &= ~((1 << vf) |
7862 (1 << (vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT)));
7863 wr32(reg_offset, reg_val);
7865 adapter->vf_data[vf].spoofchk_enabled = setting;
7869 static int igb_ndo_get_vf_config(struct net_device *netdev,
7870 int vf, struct ifla_vf_info *ivi)
7872 struct igb_adapter *adapter = netdev_priv(netdev);
7873 if (vf >= adapter->vfs_allocated_count)
7876 memcpy(&ivi->mac, adapter->vf_data[vf].vf_mac_addresses, ETH_ALEN);
7877 ivi->max_tx_rate = adapter->vf_data[vf].tx_rate;
7878 ivi->min_tx_rate = 0;
7879 ivi->vlan = adapter->vf_data[vf].pf_vlan;
7880 ivi->qos = adapter->vf_data[vf].pf_qos;
7881 ivi->spoofchk = adapter->vf_data[vf].spoofchk_enabled;
7885 static void igb_vmm_control(struct igb_adapter *adapter)
7887 struct e1000_hw *hw = &adapter->hw;
7890 switch (hw->mac.type) {
7896 /* replication is not supported for 82575 */
7899 /* notify HW that the MAC is adding vlan tags */
7900 reg = rd32(E1000_DTXCTL);
7901 reg |= E1000_DTXCTL_VLAN_ADDED;
7902 wr32(E1000_DTXCTL, reg);
7905 /* enable replication vlan tag stripping */
7906 reg = rd32(E1000_RPLOLR);
7907 reg |= E1000_RPLOLR_STRVLAN;
7908 wr32(E1000_RPLOLR, reg);
7911 /* none of the above registers are supported by i350 */
7915 if (adapter->vfs_allocated_count) {
7916 igb_vmdq_set_loopback_pf(hw, true);
7917 igb_vmdq_set_replication_pf(hw, true);
7918 igb_vmdq_set_anti_spoofing_pf(hw, true,
7919 adapter->vfs_allocated_count);
7921 igb_vmdq_set_loopback_pf(hw, false);
7922 igb_vmdq_set_replication_pf(hw, false);
7926 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba)
7928 struct e1000_hw *hw = &adapter->hw;
7932 if (hw->mac.type > e1000_82580) {
7933 if (adapter->flags & IGB_FLAG_DMAC) {
7936 /* force threshold to 0. */
7937 wr32(E1000_DMCTXTH, 0);
7939 /* DMA Coalescing high water mark needs to be greater
7940 * than the Rx threshold. Set hwm to PBA - max frame
7941 * size in 16B units, capping it at PBA - 6KB.
7943 hwm = 64 * pba - adapter->max_frame_size / 16;
7944 if (hwm < 64 * (pba - 6))
7945 hwm = 64 * (pba - 6);
7946 reg = rd32(E1000_FCRTC);
7947 reg &= ~E1000_FCRTC_RTH_COAL_MASK;
7948 reg |= ((hwm << E1000_FCRTC_RTH_COAL_SHIFT)
7949 & E1000_FCRTC_RTH_COAL_MASK);
7950 wr32(E1000_FCRTC, reg);
7952 /* Set the DMA Coalescing Rx threshold to PBA - 2 * max
7953 * frame size, capping it at PBA - 10KB.
7955 dmac_thr = pba - adapter->max_frame_size / 512;
7956 if (dmac_thr < pba - 10)
7957 dmac_thr = pba - 10;
7958 reg = rd32(E1000_DMACR);
7959 reg &= ~E1000_DMACR_DMACTHR_MASK;
7960 reg |= ((dmac_thr << E1000_DMACR_DMACTHR_SHIFT)
7961 & E1000_DMACR_DMACTHR_MASK);
7963 /* transition to L0x or L1 if available..*/
7964 reg |= (E1000_DMACR_DMAC_EN | E1000_DMACR_DMAC_LX_MASK);
7966 /* watchdog timer= +-1000 usec in 32usec intervals */
7969 /* Disable BMC-to-OS Watchdog Enable */
7970 if (hw->mac.type != e1000_i354)
7971 reg &= ~E1000_DMACR_DC_BMC2OSW_EN;
7973 wr32(E1000_DMACR, reg);
7975 /* no lower threshold to disable
7976 * coalescing(smart fifb)-UTRESH=0
7978 wr32(E1000_DMCRTRH, 0);
7980 reg = (IGB_DMCTLX_DCFLUSH_DIS | 0x4);
7982 wr32(E1000_DMCTLX, reg);
7984 /* free space in tx packet buffer to wake from
7987 wr32(E1000_DMCTXTH, (IGB_MIN_TXPBSIZE -
7988 (IGB_TX_BUF_4096 + adapter->max_frame_size)) >> 6);
7990 /* make low power state decision controlled
7993 reg = rd32(E1000_PCIEMISC);
7994 reg &= ~E1000_PCIEMISC_LX_DECISION;
7995 wr32(E1000_PCIEMISC, reg);
7996 } /* endif adapter->dmac is not disabled */
7997 } else if (hw->mac.type == e1000_82580) {
7998 u32 reg = rd32(E1000_PCIEMISC);
8000 wr32(E1000_PCIEMISC, reg & ~E1000_PCIEMISC_LX_DECISION);
8001 wr32(E1000_DMACR, 0);
8006 * igb_read_i2c_byte - Reads 8 bit word over I2C
8007 * @hw: pointer to hardware structure
8008 * @byte_offset: byte offset to read
8009 * @dev_addr: device address
8012 * Performs byte read operation over I2C interface at
8013 * a specified device address.
8015 s32 igb_read_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
8016 u8 dev_addr, u8 *data)
8018 struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
8019 struct i2c_client *this_client = adapter->i2c_client;
8024 return E1000_ERR_I2C;
8026 swfw_mask = E1000_SWFW_PHY0_SM;
8028 if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask))
8029 return E1000_ERR_SWFW_SYNC;
8031 status = i2c_smbus_read_byte_data(this_client, byte_offset);
8032 hw->mac.ops.release_swfw_sync(hw, swfw_mask);
8035 return E1000_ERR_I2C;
8043 * igb_write_i2c_byte - Writes 8 bit word over I2C
8044 * @hw: pointer to hardware structure
8045 * @byte_offset: byte offset to write
8046 * @dev_addr: device address
8047 * @data: value to write
8049 * Performs byte write operation over I2C interface at
8050 * a specified device address.
8052 s32 igb_write_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
8053 u8 dev_addr, u8 data)
8055 struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
8056 struct i2c_client *this_client = adapter->i2c_client;
8058 u16 swfw_mask = E1000_SWFW_PHY0_SM;
8061 return E1000_ERR_I2C;
8063 if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask))
8064 return E1000_ERR_SWFW_SYNC;
8065 status = i2c_smbus_write_byte_data(this_client, byte_offset, data);
8066 hw->mac.ops.release_swfw_sync(hw, swfw_mask);
8069 return E1000_ERR_I2C;
8075 int igb_reinit_queues(struct igb_adapter *adapter)
8077 struct net_device *netdev = adapter->netdev;
8078 struct pci_dev *pdev = adapter->pdev;
8081 if (netif_running(netdev))
8084 igb_reset_interrupt_capability(adapter);
8086 if (igb_init_interrupt_scheme(adapter, true)) {
8087 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
8091 if (netif_running(netdev))
8092 err = igb_open(netdev);