1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright(c) 2007 - 2018 Intel Corporation. */
4 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
6 #include <linux/module.h>
7 #include <linux/types.h>
8 #include <linux/init.h>
9 #include <linux/bitops.h>
10 #include <linux/vmalloc.h>
11 #include <linux/pagemap.h>
12 #include <linux/netdevice.h>
13 #include <linux/ipv6.h>
14 #include <linux/slab.h>
15 #include <net/checksum.h>
16 #include <net/ip6_checksum.h>
17 #include <net/pkt_sched.h>
18 #include <net/pkt_cls.h>
19 #include <linux/net_tstamp.h>
20 #include <linux/mii.h>
21 #include <linux/ethtool.h>
23 #include <linux/if_vlan.h>
24 #include <linux/pci.h>
25 #include <linux/delay.h>
26 #include <linux/interrupt.h>
28 #include <linux/tcp.h>
29 #include <linux/sctp.h>
30 #include <linux/if_ether.h>
31 #include <linux/prefetch.h>
32 #include <linux/bpf.h>
33 #include <linux/bpf_trace.h>
34 #include <linux/pm_runtime.h>
35 #include <linux/etherdevice.h>
37 #include <linux/dca.h>
39 #include <linux/i2c.h>
43 QUEUE_MODE_STRICT_PRIORITY,
44 QUEUE_MODE_STREAM_RESERVATION,
52 char igb_driver_name[] = "igb";
53 static const char igb_driver_string[] =
54 "Intel(R) Gigabit Ethernet Network Driver";
55 static const char igb_copyright[] =
56 "Copyright (c) 2007-2014 Intel Corporation.";
58 static const struct e1000_info *igb_info_tbl[] = {
59 [board_82575] = &e1000_82575_info,
62 static const struct pci_device_id igb_pci_tbl[] = {
63 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_BACKPLANE_1GBPS) },
64 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_SGMII) },
65 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_BACKPLANE_2_5GBPS) },
66 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I211_COPPER), board_82575 },
67 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER), board_82575 },
68 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_FIBER), board_82575 },
69 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES), board_82575 },
70 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SGMII), board_82575 },
71 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER_FLASHLESS), board_82575 },
72 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES_FLASHLESS), board_82575 },
73 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_COPPER), board_82575 },
74 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_FIBER), board_82575 },
75 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SERDES), board_82575 },
76 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SGMII), board_82575 },
77 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER), board_82575 },
78 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_FIBER), board_82575 },
79 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_QUAD_FIBER), board_82575 },
80 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SERDES), board_82575 },
81 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SGMII), board_82575 },
82 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER_DUAL), board_82575 },
83 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SGMII), board_82575 },
84 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SERDES), board_82575 },
85 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_BACKPLANE), board_82575 },
86 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SFP), board_82575 },
87 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576), board_82575 },
88 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS), board_82575 },
89 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS_SERDES), board_82575 },
90 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_FIBER), board_82575 },
91 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES), board_82575 },
92 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES_QUAD), board_82575 },
93 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER_ET2), board_82575 },
94 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER), board_82575 },
95 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
96 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
97 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
98 /* required last entry */
102 MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
104 static int igb_setup_all_tx_resources(struct igb_adapter *);
105 static int igb_setup_all_rx_resources(struct igb_adapter *);
106 static void igb_free_all_tx_resources(struct igb_adapter *);
107 static void igb_free_all_rx_resources(struct igb_adapter *);
108 static void igb_setup_mrqc(struct igb_adapter *);
109 static int igb_probe(struct pci_dev *, const struct pci_device_id *);
110 static void igb_remove(struct pci_dev *pdev);
111 static void igb_init_queue_configuration(struct igb_adapter *adapter);
112 static int igb_sw_init(struct igb_adapter *);
113 int igb_open(struct net_device *);
114 int igb_close(struct net_device *);
115 static void igb_configure(struct igb_adapter *);
116 static void igb_configure_tx(struct igb_adapter *);
117 static void igb_configure_rx(struct igb_adapter *);
118 static void igb_clean_all_tx_rings(struct igb_adapter *);
119 static void igb_clean_all_rx_rings(struct igb_adapter *);
120 static void igb_clean_tx_ring(struct igb_ring *);
121 static void igb_clean_rx_ring(struct igb_ring *);
122 static void igb_set_rx_mode(struct net_device *);
123 static void igb_update_phy_info(struct timer_list *);
124 static void igb_watchdog(struct timer_list *);
125 static void igb_watchdog_task(struct work_struct *);
126 static netdev_tx_t igb_xmit_frame(struct sk_buff *skb, struct net_device *);
127 static void igb_get_stats64(struct net_device *dev,
128 struct rtnl_link_stats64 *stats);
129 static int igb_change_mtu(struct net_device *, int);
130 static int igb_set_mac(struct net_device *, void *);
131 static void igb_set_uta(struct igb_adapter *adapter, bool set);
132 static irqreturn_t igb_intr(int irq, void *);
133 static irqreturn_t igb_intr_msi(int irq, void *);
134 static irqreturn_t igb_msix_other(int irq, void *);
135 static irqreturn_t igb_msix_ring(int irq, void *);
136 #ifdef CONFIG_IGB_DCA
137 static void igb_update_dca(struct igb_q_vector *);
138 static void igb_setup_dca(struct igb_adapter *);
139 #endif /* CONFIG_IGB_DCA */
140 static int igb_poll(struct napi_struct *, int);
141 static bool igb_clean_tx_irq(struct igb_q_vector *, int);
142 static int igb_clean_rx_irq(struct igb_q_vector *, int);
143 static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
144 static void igb_tx_timeout(struct net_device *, unsigned int txqueue);
145 static void igb_reset_task(struct work_struct *);
146 static void igb_vlan_mode(struct net_device *netdev,
147 netdev_features_t features);
148 static int igb_vlan_rx_add_vid(struct net_device *, __be16, u16);
149 static int igb_vlan_rx_kill_vid(struct net_device *, __be16, u16);
150 static void igb_restore_vlan(struct igb_adapter *);
151 static void igb_rar_set_index(struct igb_adapter *, u32);
152 static void igb_ping_all_vfs(struct igb_adapter *);
153 static void igb_msg_task(struct igb_adapter *);
154 static void igb_vmm_control(struct igb_adapter *);
155 static int igb_set_vf_mac(struct igb_adapter *, int, unsigned char *);
156 static void igb_flush_mac_table(struct igb_adapter *);
157 static int igb_available_rars(struct igb_adapter *, u8);
158 static void igb_set_default_mac_filter(struct igb_adapter *);
159 static int igb_uc_sync(struct net_device *, const unsigned char *);
160 static int igb_uc_unsync(struct net_device *, const unsigned char *);
161 static void igb_restore_vf_multicasts(struct igb_adapter *adapter);
162 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac);
163 static int igb_ndo_set_vf_vlan(struct net_device *netdev,
164 int vf, u16 vlan, u8 qos, __be16 vlan_proto);
165 static int igb_ndo_set_vf_bw(struct net_device *, int, int, int);
166 static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf,
168 static int igb_ndo_set_vf_trust(struct net_device *netdev, int vf,
170 static int igb_ndo_get_vf_config(struct net_device *netdev, int vf,
171 struct ifla_vf_info *ivi);
172 static void igb_check_vf_rate_limit(struct igb_adapter *);
173 static void igb_nfc_filter_exit(struct igb_adapter *adapter);
174 static void igb_nfc_filter_restore(struct igb_adapter *adapter);
176 #ifdef CONFIG_PCI_IOV
177 static int igb_vf_configure(struct igb_adapter *adapter, int vf);
178 static int igb_disable_sriov(struct pci_dev *dev, bool reinit);
181 static int igb_suspend(struct device *);
182 static int igb_resume(struct device *);
183 static int igb_runtime_suspend(struct device *dev);
184 static int igb_runtime_resume(struct device *dev);
185 static int igb_runtime_idle(struct device *dev);
186 static const struct dev_pm_ops igb_pm_ops = {
187 SET_SYSTEM_SLEEP_PM_OPS(igb_suspend, igb_resume)
188 SET_RUNTIME_PM_OPS(igb_runtime_suspend, igb_runtime_resume,
191 static void igb_shutdown(struct pci_dev *);
192 static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs);
193 #ifdef CONFIG_IGB_DCA
194 static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
195 static struct notifier_block dca_notifier = {
196 .notifier_call = igb_notify_dca,
201 #ifdef CONFIG_PCI_IOV
202 static unsigned int max_vfs;
203 module_param(max_vfs, uint, 0);
204 MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate per physical function");
205 #endif /* CONFIG_PCI_IOV */
207 static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
208 pci_channel_state_t);
209 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
210 static void igb_io_resume(struct pci_dev *);
212 static const struct pci_error_handlers igb_err_handler = {
213 .error_detected = igb_io_error_detected,
214 .slot_reset = igb_io_slot_reset,
215 .resume = igb_io_resume,
218 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba);
220 static struct pci_driver igb_driver = {
221 .name = igb_driver_name,
222 .id_table = igb_pci_tbl,
224 .remove = igb_remove,
226 .driver.pm = &igb_pm_ops,
228 .shutdown = igb_shutdown,
229 .sriov_configure = igb_pci_sriov_configure,
230 .err_handler = &igb_err_handler
233 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
234 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
235 MODULE_LICENSE("GPL v2");
237 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
238 static int debug = -1;
239 module_param(debug, int, 0);
240 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
242 struct igb_reg_info {
247 static const struct igb_reg_info igb_reg_info_tbl[] = {
249 /* General Registers */
250 {E1000_CTRL, "CTRL"},
251 {E1000_STATUS, "STATUS"},
252 {E1000_CTRL_EXT, "CTRL_EXT"},
254 /* Interrupt Registers */
258 {E1000_RCTL, "RCTL"},
259 {E1000_RDLEN(0), "RDLEN"},
260 {E1000_RDH(0), "RDH"},
261 {E1000_RDT(0), "RDT"},
262 {E1000_RXDCTL(0), "RXDCTL"},
263 {E1000_RDBAL(0), "RDBAL"},
264 {E1000_RDBAH(0), "RDBAH"},
267 {E1000_TCTL, "TCTL"},
268 {E1000_TDBAL(0), "TDBAL"},
269 {E1000_TDBAH(0), "TDBAH"},
270 {E1000_TDLEN(0), "TDLEN"},
271 {E1000_TDH(0), "TDH"},
272 {E1000_TDT(0), "TDT"},
273 {E1000_TXDCTL(0), "TXDCTL"},
274 {E1000_TDFH, "TDFH"},
275 {E1000_TDFT, "TDFT"},
276 {E1000_TDFHS, "TDFHS"},
277 {E1000_TDFPC, "TDFPC"},
279 /* List Terminator */
283 /* igb_regdump - register printout routine */
284 static void igb_regdump(struct e1000_hw *hw, struct igb_reg_info *reginfo)
290 switch (reginfo->ofs) {
292 for (n = 0; n < 4; n++)
293 regs[n] = rd32(E1000_RDLEN(n));
296 for (n = 0; n < 4; n++)
297 regs[n] = rd32(E1000_RDH(n));
300 for (n = 0; n < 4; n++)
301 regs[n] = rd32(E1000_RDT(n));
303 case E1000_RXDCTL(0):
304 for (n = 0; n < 4; n++)
305 regs[n] = rd32(E1000_RXDCTL(n));
308 for (n = 0; n < 4; n++)
309 regs[n] = rd32(E1000_RDBAL(n));
312 for (n = 0; n < 4; n++)
313 regs[n] = rd32(E1000_RDBAH(n));
316 for (n = 0; n < 4; n++)
317 regs[n] = rd32(E1000_TDBAL(n));
320 for (n = 0; n < 4; n++)
321 regs[n] = rd32(E1000_TDBAH(n));
324 for (n = 0; n < 4; n++)
325 regs[n] = rd32(E1000_TDLEN(n));
328 for (n = 0; n < 4; n++)
329 regs[n] = rd32(E1000_TDH(n));
332 for (n = 0; n < 4; n++)
333 regs[n] = rd32(E1000_TDT(n));
335 case E1000_TXDCTL(0):
336 for (n = 0; n < 4; n++)
337 regs[n] = rd32(E1000_TXDCTL(n));
340 pr_info("%-15s %08x\n", reginfo->name, rd32(reginfo->ofs));
344 snprintf(rname, 16, "%s%s", reginfo->name, "[0-3]");
345 pr_info("%-15s %08x %08x %08x %08x\n", rname, regs[0], regs[1],
349 /* igb_dump - Print registers, Tx-rings and Rx-rings */
350 static void igb_dump(struct igb_adapter *adapter)
352 struct net_device *netdev = adapter->netdev;
353 struct e1000_hw *hw = &adapter->hw;
354 struct igb_reg_info *reginfo;
355 struct igb_ring *tx_ring;
356 union e1000_adv_tx_desc *tx_desc;
357 struct my_u0 { __le64 a; __le64 b; } *u0;
358 struct igb_ring *rx_ring;
359 union e1000_adv_rx_desc *rx_desc;
363 if (!netif_msg_hw(adapter))
366 /* Print netdevice Info */
368 dev_info(&adapter->pdev->dev, "Net device Info\n");
369 pr_info("Device Name state trans_start\n");
370 pr_info("%-15s %016lX %016lX\n", netdev->name,
371 netdev->state, dev_trans_start(netdev));
374 /* Print Registers */
375 dev_info(&adapter->pdev->dev, "Register Dump\n");
376 pr_info(" Register Name Value\n");
377 for (reginfo = (struct igb_reg_info *)igb_reg_info_tbl;
378 reginfo->name; reginfo++) {
379 igb_regdump(hw, reginfo);
382 /* Print TX Ring Summary */
383 if (!netdev || !netif_running(netdev))
386 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
387 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
388 for (n = 0; n < adapter->num_tx_queues; n++) {
389 struct igb_tx_buffer *buffer_info;
390 tx_ring = adapter->tx_ring[n];
391 buffer_info = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
392 pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
393 n, tx_ring->next_to_use, tx_ring->next_to_clean,
394 (u64)dma_unmap_addr(buffer_info, dma),
395 dma_unmap_len(buffer_info, len),
396 buffer_info->next_to_watch,
397 (u64)buffer_info->time_stamp);
401 if (!netif_msg_tx_done(adapter))
402 goto rx_ring_summary;
404 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
406 /* Transmit Descriptor Formats
408 * Advanced Transmit Descriptor
409 * +--------------------------------------------------------------+
410 * 0 | Buffer Address [63:0] |
411 * +--------------------------------------------------------------+
412 * 8 | PAYLEN | PORTS |CC|IDX | STA | DCMD |DTYP|MAC|RSV| DTALEN |
413 * +--------------------------------------------------------------+
414 * 63 46 45 40 39 38 36 35 32 31 24 15 0
417 for (n = 0; n < adapter->num_tx_queues; n++) {
418 tx_ring = adapter->tx_ring[n];
419 pr_info("------------------------------------\n");
420 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
421 pr_info("------------------------------------\n");
422 pr_info("T [desc] [address 63:0 ] [PlPOCIStDDM Ln] [bi->dma ] leng ntw timestamp bi->skb\n");
424 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
425 const char *next_desc;
426 struct igb_tx_buffer *buffer_info;
427 tx_desc = IGB_TX_DESC(tx_ring, i);
428 buffer_info = &tx_ring->tx_buffer_info[i];
429 u0 = (struct my_u0 *)tx_desc;
430 if (i == tx_ring->next_to_use &&
431 i == tx_ring->next_to_clean)
432 next_desc = " NTC/U";
433 else if (i == tx_ring->next_to_use)
435 else if (i == tx_ring->next_to_clean)
440 pr_info("T [0x%03X] %016llX %016llX %016llX %04X %p %016llX %p%s\n",
441 i, le64_to_cpu(u0->a),
443 (u64)dma_unmap_addr(buffer_info, dma),
444 dma_unmap_len(buffer_info, len),
445 buffer_info->next_to_watch,
446 (u64)buffer_info->time_stamp,
447 buffer_info->skb, next_desc);
449 if (netif_msg_pktdata(adapter) && buffer_info->skb)
450 print_hex_dump(KERN_INFO, "",
452 16, 1, buffer_info->skb->data,
453 dma_unmap_len(buffer_info, len),
458 /* Print RX Rings Summary */
460 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
461 pr_info("Queue [NTU] [NTC]\n");
462 for (n = 0; n < adapter->num_rx_queues; n++) {
463 rx_ring = adapter->rx_ring[n];
464 pr_info(" %5d %5X %5X\n",
465 n, rx_ring->next_to_use, rx_ring->next_to_clean);
469 if (!netif_msg_rx_status(adapter))
472 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
474 /* Advanced Receive Descriptor (Read) Format
476 * +-----------------------------------------------------+
477 * 0 | Packet Buffer Address [63:1] |A0/NSE|
478 * +----------------------------------------------+------+
479 * 8 | Header Buffer Address [63:1] | DD |
480 * +-----------------------------------------------------+
483 * Advanced Receive Descriptor (Write-Back) Format
485 * 63 48 47 32 31 30 21 20 17 16 4 3 0
486 * +------------------------------------------------------+
487 * 0 | Packet IP |SPH| HDR_LEN | RSV|Packet| RSS |
488 * | Checksum Ident | | | | Type | Type |
489 * +------------------------------------------------------+
490 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
491 * +------------------------------------------------------+
492 * 63 48 47 32 31 20 19 0
495 for (n = 0; n < adapter->num_rx_queues; n++) {
496 rx_ring = adapter->rx_ring[n];
497 pr_info("------------------------------------\n");
498 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
499 pr_info("------------------------------------\n");
500 pr_info("R [desc] [ PktBuf A0] [ HeadBuf DD] [bi->dma ] [bi->skb] <-- Adv Rx Read format\n");
501 pr_info("RWB[desc] [PcsmIpSHl PtRs] [vl er S cks ln] ---------------- [bi->skb] <-- Adv Rx Write-Back format\n");
503 for (i = 0; i < rx_ring->count; i++) {
504 const char *next_desc;
505 struct igb_rx_buffer *buffer_info;
506 buffer_info = &rx_ring->rx_buffer_info[i];
507 rx_desc = IGB_RX_DESC(rx_ring, i);
508 u0 = (struct my_u0 *)rx_desc;
509 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
511 if (i == rx_ring->next_to_use)
513 else if (i == rx_ring->next_to_clean)
518 if (staterr & E1000_RXD_STAT_DD) {
519 /* Descriptor Done */
520 pr_info("%s[0x%03X] %016llX %016llX ---------------- %s\n",
526 pr_info("%s[0x%03X] %016llX %016llX %016llX %s\n",
530 (u64)buffer_info->dma,
533 if (netif_msg_pktdata(adapter) &&
534 buffer_info->dma && buffer_info->page) {
535 print_hex_dump(KERN_INFO, "",
538 page_address(buffer_info->page) +
539 buffer_info->page_offset,
540 igb_rx_bufsz(rx_ring), true);
551 * igb_get_i2c_data - Reads the I2C SDA data bit
552 * @data: opaque pointer to adapter struct
554 * Returns the I2C data bit value
556 static int igb_get_i2c_data(void *data)
558 struct igb_adapter *adapter = (struct igb_adapter *)data;
559 struct e1000_hw *hw = &adapter->hw;
560 s32 i2cctl = rd32(E1000_I2CPARAMS);
562 return !!(i2cctl & E1000_I2C_DATA_IN);
566 * igb_set_i2c_data - Sets the I2C data bit
567 * @data: pointer to hardware structure
568 * @state: I2C data value (0 or 1) to set
570 * Sets the I2C data bit
572 static void igb_set_i2c_data(void *data, int state)
574 struct igb_adapter *adapter = (struct igb_adapter *)data;
575 struct e1000_hw *hw = &adapter->hw;
576 s32 i2cctl = rd32(E1000_I2CPARAMS);
579 i2cctl |= E1000_I2C_DATA_OUT | E1000_I2C_DATA_OE_N;
581 i2cctl &= ~E1000_I2C_DATA_OE_N;
582 i2cctl &= ~E1000_I2C_DATA_OUT;
585 wr32(E1000_I2CPARAMS, i2cctl);
590 * igb_set_i2c_clk - Sets the I2C SCL clock
591 * @data: pointer to hardware structure
592 * @state: state to set clock
594 * Sets the I2C clock line to state
596 static void igb_set_i2c_clk(void *data, int state)
598 struct igb_adapter *adapter = (struct igb_adapter *)data;
599 struct e1000_hw *hw = &adapter->hw;
600 s32 i2cctl = rd32(E1000_I2CPARAMS);
603 i2cctl |= E1000_I2C_CLK_OUT | E1000_I2C_CLK_OE_N;
605 i2cctl &= ~E1000_I2C_CLK_OUT;
606 i2cctl &= ~E1000_I2C_CLK_OE_N;
608 wr32(E1000_I2CPARAMS, i2cctl);
613 * igb_get_i2c_clk - Gets the I2C SCL clock state
614 * @data: pointer to hardware structure
616 * Gets the I2C clock state
618 static int igb_get_i2c_clk(void *data)
620 struct igb_adapter *adapter = (struct igb_adapter *)data;
621 struct e1000_hw *hw = &adapter->hw;
622 s32 i2cctl = rd32(E1000_I2CPARAMS);
624 return !!(i2cctl & E1000_I2C_CLK_IN);
627 static const struct i2c_algo_bit_data igb_i2c_algo = {
628 .setsda = igb_set_i2c_data,
629 .setscl = igb_set_i2c_clk,
630 .getsda = igb_get_i2c_data,
631 .getscl = igb_get_i2c_clk,
637 * igb_get_hw_dev - return device
638 * @hw: pointer to hardware structure
640 * used by hardware layer to print debugging information
642 struct net_device *igb_get_hw_dev(struct e1000_hw *hw)
644 struct igb_adapter *adapter = hw->back;
645 return adapter->netdev;
649 * igb_init_module - Driver Registration Routine
651 * igb_init_module is the first routine called when the driver is
652 * loaded. All it does is register with the PCI subsystem.
654 static int __init igb_init_module(void)
658 pr_info("%s\n", igb_driver_string);
659 pr_info("%s\n", igb_copyright);
661 #ifdef CONFIG_IGB_DCA
662 dca_register_notify(&dca_notifier);
664 ret = pci_register_driver(&igb_driver);
668 module_init(igb_init_module);
671 * igb_exit_module - Driver Exit Cleanup Routine
673 * igb_exit_module is called just before the driver is removed
676 static void __exit igb_exit_module(void)
678 #ifdef CONFIG_IGB_DCA
679 dca_unregister_notify(&dca_notifier);
681 pci_unregister_driver(&igb_driver);
684 module_exit(igb_exit_module);
686 #define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
688 * igb_cache_ring_register - Descriptor ring to register mapping
689 * @adapter: board private structure to initialize
691 * Once we know the feature-set enabled for the device, we'll cache
692 * the register offset the descriptor ring is assigned to.
694 static void igb_cache_ring_register(struct igb_adapter *adapter)
697 u32 rbase_offset = adapter->vfs_allocated_count;
699 switch (adapter->hw.mac.type) {
701 /* The queues are allocated for virtualization such that VF 0
702 * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
703 * In order to avoid collision we start at the first free queue
704 * and continue consuming queues in the same sequence
706 if (adapter->vfs_allocated_count) {
707 for (; i < adapter->rss_queues; i++)
708 adapter->rx_ring[i]->reg_idx = rbase_offset +
719 for (; i < adapter->num_rx_queues; i++)
720 adapter->rx_ring[i]->reg_idx = rbase_offset + i;
721 for (; j < adapter->num_tx_queues; j++)
722 adapter->tx_ring[j]->reg_idx = rbase_offset + j;
727 u32 igb_rd32(struct e1000_hw *hw, u32 reg)
729 struct igb_adapter *igb = container_of(hw, struct igb_adapter, hw);
730 u8 __iomem *hw_addr = READ_ONCE(hw->hw_addr);
733 if (E1000_REMOVED(hw_addr))
736 value = readl(&hw_addr[reg]);
738 /* reads should not return all F's */
739 if (!(~value) && (!reg || !(~readl(hw_addr)))) {
740 struct net_device *netdev = igb->netdev;
742 netdev_err(netdev, "PCIe link lost\n");
743 WARN(pci_device_is_present(igb->pdev),
744 "igb: Failed to read reg 0x%x!\n", reg);
751 * igb_write_ivar - configure ivar for given MSI-X vector
752 * @hw: pointer to the HW structure
753 * @msix_vector: vector number we are allocating to a given ring
754 * @index: row index of IVAR register to write within IVAR table
755 * @offset: column offset of in IVAR, should be multiple of 8
757 * This function is intended to handle the writing of the IVAR register
758 * for adapters 82576 and newer. The IVAR table consists of 2 columns,
759 * each containing an cause allocation for an Rx and Tx ring, and a
760 * variable number of rows depending on the number of queues supported.
762 static void igb_write_ivar(struct e1000_hw *hw, int msix_vector,
763 int index, int offset)
765 u32 ivar = array_rd32(E1000_IVAR0, index);
767 /* clear any bits that are currently set */
768 ivar &= ~((u32)0xFF << offset);
770 /* write vector and valid bit */
771 ivar |= (msix_vector | E1000_IVAR_VALID) << offset;
773 array_wr32(E1000_IVAR0, index, ivar);
776 #define IGB_N0_QUEUE -1
777 static void igb_assign_vector(struct igb_q_vector *q_vector, int msix_vector)
779 struct igb_adapter *adapter = q_vector->adapter;
780 struct e1000_hw *hw = &adapter->hw;
781 int rx_queue = IGB_N0_QUEUE;
782 int tx_queue = IGB_N0_QUEUE;
785 if (q_vector->rx.ring)
786 rx_queue = q_vector->rx.ring->reg_idx;
787 if (q_vector->tx.ring)
788 tx_queue = q_vector->tx.ring->reg_idx;
790 switch (hw->mac.type) {
792 /* The 82575 assigns vectors using a bitmask, which matches the
793 * bitmask for the EICR/EIMS/EIMC registers. To assign one
794 * or more queues to a vector, we write the appropriate bits
795 * into the MSIXBM register for that vector.
797 if (rx_queue > IGB_N0_QUEUE)
798 msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
799 if (tx_queue > IGB_N0_QUEUE)
800 msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
801 if (!(adapter->flags & IGB_FLAG_HAS_MSIX) && msix_vector == 0)
802 msixbm |= E1000_EIMS_OTHER;
803 array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
804 q_vector->eims_value = msixbm;
807 /* 82576 uses a table that essentially consists of 2 columns
808 * with 8 rows. The ordering is column-major so we use the
809 * lower 3 bits as the row index, and the 4th bit as the
812 if (rx_queue > IGB_N0_QUEUE)
813 igb_write_ivar(hw, msix_vector,
815 (rx_queue & 0x8) << 1);
816 if (tx_queue > IGB_N0_QUEUE)
817 igb_write_ivar(hw, msix_vector,
819 ((tx_queue & 0x8) << 1) + 8);
820 q_vector->eims_value = BIT(msix_vector);
827 /* On 82580 and newer adapters the scheme is similar to 82576
828 * however instead of ordering column-major we have things
829 * ordered row-major. So we traverse the table by using
830 * bit 0 as the column offset, and the remaining bits as the
833 if (rx_queue > IGB_N0_QUEUE)
834 igb_write_ivar(hw, msix_vector,
836 (rx_queue & 0x1) << 4);
837 if (tx_queue > IGB_N0_QUEUE)
838 igb_write_ivar(hw, msix_vector,
840 ((tx_queue & 0x1) << 4) + 8);
841 q_vector->eims_value = BIT(msix_vector);
848 /* add q_vector eims value to global eims_enable_mask */
849 adapter->eims_enable_mask |= q_vector->eims_value;
851 /* configure q_vector to set itr on first interrupt */
852 q_vector->set_itr = 1;
856 * igb_configure_msix - Configure MSI-X hardware
857 * @adapter: board private structure to initialize
859 * igb_configure_msix sets up the hardware to properly
860 * generate MSI-X interrupts.
862 static void igb_configure_msix(struct igb_adapter *adapter)
866 struct e1000_hw *hw = &adapter->hw;
868 adapter->eims_enable_mask = 0;
870 /* set vector for other causes, i.e. link changes */
871 switch (hw->mac.type) {
873 tmp = rd32(E1000_CTRL_EXT);
874 /* enable MSI-X PBA support*/
875 tmp |= E1000_CTRL_EXT_PBA_CLR;
877 /* Auto-Mask interrupts upon ICR read. */
878 tmp |= E1000_CTRL_EXT_EIAME;
879 tmp |= E1000_CTRL_EXT_IRCA;
881 wr32(E1000_CTRL_EXT, tmp);
883 /* enable msix_other interrupt */
884 array_wr32(E1000_MSIXBM(0), vector++, E1000_EIMS_OTHER);
885 adapter->eims_other = E1000_EIMS_OTHER;
895 /* Turn on MSI-X capability first, or our settings
896 * won't stick. And it will take days to debug.
898 wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
899 E1000_GPIE_PBA | E1000_GPIE_EIAME |
902 /* enable msix_other interrupt */
903 adapter->eims_other = BIT(vector);
904 tmp = (vector++ | E1000_IVAR_VALID) << 8;
906 wr32(E1000_IVAR_MISC, tmp);
909 /* do nothing, since nothing else supports MSI-X */
911 } /* switch (hw->mac.type) */
913 adapter->eims_enable_mask |= adapter->eims_other;
915 for (i = 0; i < adapter->num_q_vectors; i++)
916 igb_assign_vector(adapter->q_vector[i], vector++);
922 * igb_request_msix - Initialize MSI-X interrupts
923 * @adapter: board private structure to initialize
925 * igb_request_msix allocates MSI-X vectors and requests interrupts from the
928 static int igb_request_msix(struct igb_adapter *adapter)
930 unsigned int num_q_vectors = adapter->num_q_vectors;
931 struct net_device *netdev = adapter->netdev;
932 int i, err = 0, vector = 0, free_vector = 0;
934 err = request_irq(adapter->msix_entries[vector].vector,
935 igb_msix_other, 0, netdev->name, adapter);
939 if (num_q_vectors > MAX_Q_VECTORS) {
940 num_q_vectors = MAX_Q_VECTORS;
941 dev_warn(&adapter->pdev->dev,
942 "The number of queue vectors (%d) is higher than max allowed (%d)\n",
943 adapter->num_q_vectors, MAX_Q_VECTORS);
945 for (i = 0; i < num_q_vectors; i++) {
946 struct igb_q_vector *q_vector = adapter->q_vector[i];
950 q_vector->itr_register = adapter->io_addr + E1000_EITR(vector);
952 if (q_vector->rx.ring && q_vector->tx.ring)
953 sprintf(q_vector->name, "%s-TxRx-%u", netdev->name,
954 q_vector->rx.ring->queue_index);
955 else if (q_vector->tx.ring)
956 sprintf(q_vector->name, "%s-tx-%u", netdev->name,
957 q_vector->tx.ring->queue_index);
958 else if (q_vector->rx.ring)
959 sprintf(q_vector->name, "%s-rx-%u", netdev->name,
960 q_vector->rx.ring->queue_index);
962 sprintf(q_vector->name, "%s-unused", netdev->name);
964 err = request_irq(adapter->msix_entries[vector].vector,
965 igb_msix_ring, 0, q_vector->name,
971 igb_configure_msix(adapter);
975 /* free already assigned IRQs */
976 free_irq(adapter->msix_entries[free_vector++].vector, adapter);
979 for (i = 0; i < vector; i++) {
980 free_irq(adapter->msix_entries[free_vector++].vector,
981 adapter->q_vector[i]);
988 * igb_free_q_vector - Free memory allocated for specific interrupt vector
989 * @adapter: board private structure to initialize
990 * @v_idx: Index of vector to be freed
992 * This function frees the memory allocated to the q_vector.
994 static void igb_free_q_vector(struct igb_adapter *adapter, int v_idx)
996 struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
998 adapter->q_vector[v_idx] = NULL;
1000 /* igb_get_stats64() might access the rings on this vector,
1001 * we must wait a grace period before freeing it.
1004 kfree_rcu(q_vector, rcu);
1008 * igb_reset_q_vector - Reset config for interrupt vector
1009 * @adapter: board private structure to initialize
1010 * @v_idx: Index of vector to be reset
1012 * If NAPI is enabled it will delete any references to the
1013 * NAPI struct. This is preparation for igb_free_q_vector.
1015 static void igb_reset_q_vector(struct igb_adapter *adapter, int v_idx)
1017 struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
1019 /* Coming from igb_set_interrupt_capability, the vectors are not yet
1020 * allocated. So, q_vector is NULL so we should stop here.
1025 if (q_vector->tx.ring)
1026 adapter->tx_ring[q_vector->tx.ring->queue_index] = NULL;
1028 if (q_vector->rx.ring)
1029 adapter->rx_ring[q_vector->rx.ring->queue_index] = NULL;
1031 netif_napi_del(&q_vector->napi);
1035 static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
1037 int v_idx = adapter->num_q_vectors;
1039 if (adapter->flags & IGB_FLAG_HAS_MSIX)
1040 pci_disable_msix(adapter->pdev);
1041 else if (adapter->flags & IGB_FLAG_HAS_MSI)
1042 pci_disable_msi(adapter->pdev);
1045 igb_reset_q_vector(adapter, v_idx);
1049 * igb_free_q_vectors - Free memory allocated for interrupt vectors
1050 * @adapter: board private structure to initialize
1052 * This function frees the memory allocated to the q_vectors. In addition if
1053 * NAPI is enabled it will delete any references to the NAPI struct prior
1054 * to freeing the q_vector.
1056 static void igb_free_q_vectors(struct igb_adapter *adapter)
1058 int v_idx = adapter->num_q_vectors;
1060 adapter->num_tx_queues = 0;
1061 adapter->num_rx_queues = 0;
1062 adapter->num_q_vectors = 0;
1065 igb_reset_q_vector(adapter, v_idx);
1066 igb_free_q_vector(adapter, v_idx);
1071 * igb_clear_interrupt_scheme - reset the device to a state of no interrupts
1072 * @adapter: board private structure to initialize
1074 * This function resets the device so that it has 0 Rx queues, Tx queues, and
1075 * MSI-X interrupts allocated.
1077 static void igb_clear_interrupt_scheme(struct igb_adapter *adapter)
1079 igb_free_q_vectors(adapter);
1080 igb_reset_interrupt_capability(adapter);
1084 * igb_set_interrupt_capability - set MSI or MSI-X if supported
1085 * @adapter: board private structure to initialize
1086 * @msix: boolean value of MSIX capability
1088 * Attempt to configure interrupts using the best available
1089 * capabilities of the hardware and kernel.
1091 static void igb_set_interrupt_capability(struct igb_adapter *adapter, bool msix)
1098 adapter->flags |= IGB_FLAG_HAS_MSIX;
1100 /* Number of supported queues. */
1101 adapter->num_rx_queues = adapter->rss_queues;
1102 if (adapter->vfs_allocated_count)
1103 adapter->num_tx_queues = 1;
1105 adapter->num_tx_queues = adapter->rss_queues;
1107 /* start with one vector for every Rx queue */
1108 numvecs = adapter->num_rx_queues;
1110 /* if Tx handler is separate add 1 for every Tx queue */
1111 if (!(adapter->flags & IGB_FLAG_QUEUE_PAIRS))
1112 numvecs += adapter->num_tx_queues;
1114 /* store the number of vectors reserved for queues */
1115 adapter->num_q_vectors = numvecs;
1117 /* add 1 vector for link status interrupts */
1119 for (i = 0; i < numvecs; i++)
1120 adapter->msix_entries[i].entry = i;
1122 err = pci_enable_msix_range(adapter->pdev,
1123 adapter->msix_entries,
1129 igb_reset_interrupt_capability(adapter);
1131 /* If we can't do MSI-X, try MSI */
1133 adapter->flags &= ~IGB_FLAG_HAS_MSIX;
1134 #ifdef CONFIG_PCI_IOV
1135 /* disable SR-IOV for non MSI-X configurations */
1136 if (adapter->vf_data) {
1137 struct e1000_hw *hw = &adapter->hw;
1138 /* disable iov and allow time for transactions to clear */
1139 pci_disable_sriov(adapter->pdev);
1142 kfree(adapter->vf_mac_list);
1143 adapter->vf_mac_list = NULL;
1144 kfree(adapter->vf_data);
1145 adapter->vf_data = NULL;
1146 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
1149 dev_info(&adapter->pdev->dev, "IOV Disabled\n");
1152 adapter->vfs_allocated_count = 0;
1153 adapter->rss_queues = 1;
1154 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
1155 adapter->num_rx_queues = 1;
1156 adapter->num_tx_queues = 1;
1157 adapter->num_q_vectors = 1;
1158 if (!pci_enable_msi(adapter->pdev))
1159 adapter->flags |= IGB_FLAG_HAS_MSI;
1162 static void igb_add_ring(struct igb_ring *ring,
1163 struct igb_ring_container *head)
1170 * igb_alloc_q_vector - Allocate memory for a single interrupt vector
1171 * @adapter: board private structure to initialize
1172 * @v_count: q_vectors allocated on adapter, used for ring interleaving
1173 * @v_idx: index of vector in adapter struct
1174 * @txr_count: total number of Tx rings to allocate
1175 * @txr_idx: index of first Tx ring to allocate
1176 * @rxr_count: total number of Rx rings to allocate
1177 * @rxr_idx: index of first Rx ring to allocate
1179 * We allocate one q_vector. If allocation fails we return -ENOMEM.
1181 static int igb_alloc_q_vector(struct igb_adapter *adapter,
1182 int v_count, int v_idx,
1183 int txr_count, int txr_idx,
1184 int rxr_count, int rxr_idx)
1186 struct igb_q_vector *q_vector;
1187 struct igb_ring *ring;
1191 /* igb only supports 1 Tx and/or 1 Rx queue per vector */
1192 if (txr_count > 1 || rxr_count > 1)
1195 ring_count = txr_count + rxr_count;
1196 size = kmalloc_size_roundup(struct_size(q_vector, ring, ring_count));
1198 /* allocate q_vector and rings */
1199 q_vector = adapter->q_vector[v_idx];
1201 q_vector = kzalloc(size, GFP_KERNEL);
1202 } else if (size > ksize(q_vector)) {
1203 struct igb_q_vector *new_q_vector;
1205 new_q_vector = kzalloc(size, GFP_KERNEL);
1207 kfree_rcu(q_vector, rcu);
1208 q_vector = new_q_vector;
1210 memset(q_vector, 0, size);
1215 /* initialize NAPI */
1216 netif_napi_add(adapter->netdev, &q_vector->napi, igb_poll);
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->io_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 ring->cbs_enable = false;
1263 ring->idleslope = 0;
1264 ring->sendslope = 0;
1268 u64_stats_init(&ring->tx_syncp);
1269 u64_stats_init(&ring->tx_syncp2);
1271 /* assign ring to adapter */
1272 adapter->tx_ring[txr_idx] = ring;
1274 /* push pointer to next ring */
1279 /* assign generic ring traits */
1280 ring->dev = &adapter->pdev->dev;
1281 ring->netdev = adapter->netdev;
1283 /* configure backlink on ring */
1284 ring->q_vector = q_vector;
1286 /* update q_vector Rx values */
1287 igb_add_ring(ring, &q_vector->rx);
1289 /* set flag indicating ring supports SCTP checksum offload */
1290 if (adapter->hw.mac.type >= e1000_82576)
1291 set_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags);
1293 /* On i350, i354, i210, and i211, loopback VLAN packets
1294 * have the tag byte-swapped.
1296 if (adapter->hw.mac.type >= e1000_i350)
1297 set_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &ring->flags);
1299 /* apply Rx specific ring traits */
1300 ring->count = adapter->rx_ring_count;
1301 ring->queue_index = rxr_idx;
1303 u64_stats_init(&ring->rx_syncp);
1305 /* assign ring to adapter */
1306 adapter->rx_ring[rxr_idx] = ring;
1314 * igb_alloc_q_vectors - Allocate memory for interrupt vectors
1315 * @adapter: board private structure to initialize
1317 * We allocate one q_vector per queue interrupt. If allocation fails we
1320 static int igb_alloc_q_vectors(struct igb_adapter *adapter)
1322 int q_vectors = adapter->num_q_vectors;
1323 int rxr_remaining = adapter->num_rx_queues;
1324 int txr_remaining = adapter->num_tx_queues;
1325 int rxr_idx = 0, txr_idx = 0, v_idx = 0;
1328 if (q_vectors >= (rxr_remaining + txr_remaining)) {
1329 for (; rxr_remaining; v_idx++) {
1330 err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1336 /* update counts and index */
1342 for (; v_idx < q_vectors; v_idx++) {
1343 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_idx);
1344 int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_idx);
1346 err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1347 tqpv, txr_idx, rqpv, rxr_idx);
1352 /* update counts and index */
1353 rxr_remaining -= rqpv;
1354 txr_remaining -= tqpv;
1362 adapter->num_tx_queues = 0;
1363 adapter->num_rx_queues = 0;
1364 adapter->num_q_vectors = 0;
1367 igb_free_q_vector(adapter, v_idx);
1373 * igb_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
1374 * @adapter: board private structure to initialize
1375 * @msix: boolean value of MSIX capability
1377 * This function initializes the interrupts and allocates all of the queues.
1379 static int igb_init_interrupt_scheme(struct igb_adapter *adapter, bool msix)
1381 struct pci_dev *pdev = adapter->pdev;
1384 igb_set_interrupt_capability(adapter, msix);
1386 err = igb_alloc_q_vectors(adapter);
1388 dev_err(&pdev->dev, "Unable to allocate memory for vectors\n");
1389 goto err_alloc_q_vectors;
1392 igb_cache_ring_register(adapter);
1396 err_alloc_q_vectors:
1397 igb_reset_interrupt_capability(adapter);
1402 * igb_request_irq - initialize interrupts
1403 * @adapter: board private structure to initialize
1405 * Attempts to configure interrupts using the best available
1406 * capabilities of the hardware and kernel.
1408 static int igb_request_irq(struct igb_adapter *adapter)
1410 struct net_device *netdev = adapter->netdev;
1411 struct pci_dev *pdev = adapter->pdev;
1414 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1415 err = igb_request_msix(adapter);
1418 /* fall back to MSI */
1419 igb_free_all_tx_resources(adapter);
1420 igb_free_all_rx_resources(adapter);
1422 igb_clear_interrupt_scheme(adapter);
1423 err = igb_init_interrupt_scheme(adapter, false);
1427 igb_setup_all_tx_resources(adapter);
1428 igb_setup_all_rx_resources(adapter);
1429 igb_configure(adapter);
1432 igb_assign_vector(adapter->q_vector[0], 0);
1434 if (adapter->flags & IGB_FLAG_HAS_MSI) {
1435 err = request_irq(pdev->irq, igb_intr_msi, 0,
1436 netdev->name, adapter);
1440 /* fall back to legacy interrupts */
1441 igb_reset_interrupt_capability(adapter);
1442 adapter->flags &= ~IGB_FLAG_HAS_MSI;
1445 err = request_irq(pdev->irq, igb_intr, IRQF_SHARED,
1446 netdev->name, adapter);
1449 dev_err(&pdev->dev, "Error %d getting interrupt\n",
1456 static void igb_free_irq(struct igb_adapter *adapter)
1458 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1461 free_irq(adapter->msix_entries[vector++].vector, adapter);
1463 for (i = 0; i < adapter->num_q_vectors; i++)
1464 free_irq(adapter->msix_entries[vector++].vector,
1465 adapter->q_vector[i]);
1467 free_irq(adapter->pdev->irq, adapter);
1472 * igb_irq_disable - Mask off interrupt generation on the NIC
1473 * @adapter: board private structure
1475 static void igb_irq_disable(struct igb_adapter *adapter)
1477 struct e1000_hw *hw = &adapter->hw;
1479 /* we need to be careful when disabling interrupts. The VFs are also
1480 * mapped into these registers and so clearing the bits can cause
1481 * issues on the VF drivers so we only need to clear what we set
1483 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1484 u32 regval = rd32(E1000_EIAM);
1486 wr32(E1000_EIAM, regval & ~adapter->eims_enable_mask);
1487 wr32(E1000_EIMC, adapter->eims_enable_mask);
1488 regval = rd32(E1000_EIAC);
1489 wr32(E1000_EIAC, regval & ~adapter->eims_enable_mask);
1493 wr32(E1000_IMC, ~0);
1495 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1498 for (i = 0; i < adapter->num_q_vectors; i++)
1499 synchronize_irq(adapter->msix_entries[i].vector);
1501 synchronize_irq(adapter->pdev->irq);
1506 * igb_irq_enable - Enable default interrupt generation settings
1507 * @adapter: board private structure
1509 static void igb_irq_enable(struct igb_adapter *adapter)
1511 struct e1000_hw *hw = &adapter->hw;
1513 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1514 u32 ims = E1000_IMS_LSC | E1000_IMS_DOUTSYNC | E1000_IMS_DRSTA;
1515 u32 regval = rd32(E1000_EIAC);
1517 wr32(E1000_EIAC, regval | adapter->eims_enable_mask);
1518 regval = rd32(E1000_EIAM);
1519 wr32(E1000_EIAM, regval | adapter->eims_enable_mask);
1520 wr32(E1000_EIMS, adapter->eims_enable_mask);
1521 if (adapter->vfs_allocated_count) {
1522 wr32(E1000_MBVFIMR, 0xFF);
1523 ims |= E1000_IMS_VMMB;
1525 wr32(E1000_IMS, ims);
1527 wr32(E1000_IMS, IMS_ENABLE_MASK |
1529 wr32(E1000_IAM, IMS_ENABLE_MASK |
1534 static void igb_update_mng_vlan(struct igb_adapter *adapter)
1536 struct e1000_hw *hw = &adapter->hw;
1537 u16 pf_id = adapter->vfs_allocated_count;
1538 u16 vid = adapter->hw.mng_cookie.vlan_id;
1539 u16 old_vid = adapter->mng_vlan_id;
1541 if (hw->mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1542 /* add VID to filter table */
1543 igb_vfta_set(hw, vid, pf_id, true, true);
1544 adapter->mng_vlan_id = vid;
1546 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1549 if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
1551 !test_bit(old_vid, adapter->active_vlans)) {
1552 /* remove VID from filter table */
1553 igb_vfta_set(hw, vid, pf_id, false, true);
1558 * igb_release_hw_control - release control of the h/w to f/w
1559 * @adapter: address of board private structure
1561 * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
1562 * For ASF and Pass Through versions of f/w this means that the
1563 * driver is no longer loaded.
1565 static void igb_release_hw_control(struct igb_adapter *adapter)
1567 struct e1000_hw *hw = &adapter->hw;
1570 /* Let firmware take over control of h/w */
1571 ctrl_ext = rd32(E1000_CTRL_EXT);
1572 wr32(E1000_CTRL_EXT,
1573 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1577 * igb_get_hw_control - get control of the h/w from f/w
1578 * @adapter: address of board private structure
1580 * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
1581 * For ASF and Pass Through versions of f/w this means that
1582 * the driver is loaded.
1584 static void igb_get_hw_control(struct igb_adapter *adapter)
1586 struct e1000_hw *hw = &adapter->hw;
1589 /* Let firmware know the driver has taken over */
1590 ctrl_ext = rd32(E1000_CTRL_EXT);
1591 wr32(E1000_CTRL_EXT,
1592 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1595 static void enable_fqtss(struct igb_adapter *adapter, bool enable)
1597 struct net_device *netdev = adapter->netdev;
1598 struct e1000_hw *hw = &adapter->hw;
1600 WARN_ON(hw->mac.type != e1000_i210);
1603 adapter->flags |= IGB_FLAG_FQTSS;
1605 adapter->flags &= ~IGB_FLAG_FQTSS;
1607 if (netif_running(netdev))
1608 schedule_work(&adapter->reset_task);
1611 static bool is_fqtss_enabled(struct igb_adapter *adapter)
1613 return (adapter->flags & IGB_FLAG_FQTSS) ? true : false;
1616 static void set_tx_desc_fetch_prio(struct e1000_hw *hw, int queue,
1617 enum tx_queue_prio prio)
1621 WARN_ON(hw->mac.type != e1000_i210);
1622 WARN_ON(queue < 0 || queue > 4);
1624 val = rd32(E1000_I210_TXDCTL(queue));
1626 if (prio == TX_QUEUE_PRIO_HIGH)
1627 val |= E1000_TXDCTL_PRIORITY;
1629 val &= ~E1000_TXDCTL_PRIORITY;
1631 wr32(E1000_I210_TXDCTL(queue), val);
1634 static void set_queue_mode(struct e1000_hw *hw, int queue, enum queue_mode mode)
1638 WARN_ON(hw->mac.type != e1000_i210);
1639 WARN_ON(queue < 0 || queue > 1);
1641 val = rd32(E1000_I210_TQAVCC(queue));
1643 if (mode == QUEUE_MODE_STREAM_RESERVATION)
1644 val |= E1000_TQAVCC_QUEUEMODE;
1646 val &= ~E1000_TQAVCC_QUEUEMODE;
1648 wr32(E1000_I210_TQAVCC(queue), val);
1651 static bool is_any_cbs_enabled(struct igb_adapter *adapter)
1655 for (i = 0; i < adapter->num_tx_queues; i++) {
1656 if (adapter->tx_ring[i]->cbs_enable)
1663 static bool is_any_txtime_enabled(struct igb_adapter *adapter)
1667 for (i = 0; i < adapter->num_tx_queues; i++) {
1668 if (adapter->tx_ring[i]->launchtime_enable)
1676 * igb_config_tx_modes - Configure "Qav Tx mode" features on igb
1677 * @adapter: pointer to adapter struct
1678 * @queue: queue number
1680 * Configure CBS and Launchtime for a given hardware queue.
1681 * Parameters are retrieved from the correct Tx ring, so
1682 * igb_save_cbs_params() and igb_save_txtime_params() should be used
1683 * for setting those correctly prior to this function being called.
1685 static void igb_config_tx_modes(struct igb_adapter *adapter, int queue)
1687 struct net_device *netdev = adapter->netdev;
1688 struct e1000_hw *hw = &adapter->hw;
1689 struct igb_ring *ring;
1690 u32 tqavcc, tqavctrl;
1693 WARN_ON(hw->mac.type != e1000_i210);
1694 WARN_ON(queue < 0 || queue > 1);
1695 ring = adapter->tx_ring[queue];
1697 /* If any of the Qav features is enabled, configure queues as SR and
1698 * with HIGH PRIO. If none is, then configure them with LOW PRIO and
1701 if (ring->cbs_enable || ring->launchtime_enable) {
1702 set_tx_desc_fetch_prio(hw, queue, TX_QUEUE_PRIO_HIGH);
1703 set_queue_mode(hw, queue, QUEUE_MODE_STREAM_RESERVATION);
1705 set_tx_desc_fetch_prio(hw, queue, TX_QUEUE_PRIO_LOW);
1706 set_queue_mode(hw, queue, QUEUE_MODE_STRICT_PRIORITY);
1709 /* If CBS is enabled, set DataTranARB and config its parameters. */
1710 if (ring->cbs_enable || queue == 0) {
1711 /* i210 does not allow the queue 0 to be in the Strict
1712 * Priority mode while the Qav mode is enabled, so,
1713 * instead of disabling strict priority mode, we give
1714 * queue 0 the maximum of credits possible.
1716 * See section 8.12.19 of the i210 datasheet, "Note:
1717 * Queue0 QueueMode must be set to 1b when
1718 * TransmitMode is set to Qav."
1720 if (queue == 0 && !ring->cbs_enable) {
1721 /* max "linkspeed" idleslope in kbps */
1722 ring->idleslope = 1000000;
1723 ring->hicredit = ETH_FRAME_LEN;
1726 /* Always set data transfer arbitration to credit-based
1727 * shaper algorithm on TQAVCTRL if CBS is enabled for any of
1730 tqavctrl = rd32(E1000_I210_TQAVCTRL);
1731 tqavctrl |= E1000_TQAVCTRL_DATATRANARB;
1732 wr32(E1000_I210_TQAVCTRL, tqavctrl);
1734 /* According to i210 datasheet section 7.2.7.7, we should set
1735 * the 'idleSlope' field from TQAVCC register following the
1738 * For 100 Mbps link speed:
1740 * value = BW * 0x7735 * 0.2 (E1)
1742 * For 1000Mbps link speed:
1744 * value = BW * 0x7735 * 2 (E2)
1746 * E1 and E2 can be merged into one equation as shown below.
1747 * Note that 'link-speed' is in Mbps.
1749 * value = BW * 0x7735 * 2 * link-speed
1750 * -------------- (E3)
1753 * 'BW' is the percentage bandwidth out of full link speed
1754 * which can be found with the following equation. Note that
1755 * idleSlope here is the parameter from this function which
1759 * ----------------- (E4)
1762 * That said, we can come up with a generic equation to
1763 * calculate the value we should set it TQAVCC register by
1764 * replacing 'BW' in E3 by E4. The resulting equation is:
1766 * value = idleSlope * 0x7735 * 2 * link-speed
1767 * ----------------- -------------- (E5)
1768 * link-speed * 1000 1000
1770 * 'link-speed' is present in both sides of the fraction so
1771 * it is canceled out. The final equation is the following:
1773 * value = idleSlope * 61034
1774 * ----------------- (E6)
1777 * NOTE: For i210, given the above, we can see that idleslope
1778 * is represented in 16.38431 kbps units by the value at
1779 * the TQAVCC register (1Gbps / 61034), which reduces
1780 * the granularity for idleslope increments.
1781 * For instance, if you want to configure a 2576kbps
1782 * idleslope, the value to be written on the register
1783 * would have to be 157.23. If rounded down, you end
1784 * up with less bandwidth available than originally
1785 * required (~2572 kbps). If rounded up, you end up
1786 * with a higher bandwidth (~2589 kbps). Below the
1787 * approach we take is to always round up the
1788 * calculated value, so the resulting bandwidth might
1789 * be slightly higher for some configurations.
1791 value = DIV_ROUND_UP_ULL(ring->idleslope * 61034ULL, 1000000);
1793 tqavcc = rd32(E1000_I210_TQAVCC(queue));
1794 tqavcc &= ~E1000_TQAVCC_IDLESLOPE_MASK;
1796 wr32(E1000_I210_TQAVCC(queue), tqavcc);
1798 wr32(E1000_I210_TQAVHC(queue),
1799 0x80000000 + ring->hicredit * 0x7735);
1802 /* Set idleSlope to zero. */
1803 tqavcc = rd32(E1000_I210_TQAVCC(queue));
1804 tqavcc &= ~E1000_TQAVCC_IDLESLOPE_MASK;
1805 wr32(E1000_I210_TQAVCC(queue), tqavcc);
1807 /* Set hiCredit to zero. */
1808 wr32(E1000_I210_TQAVHC(queue), 0);
1810 /* If CBS is not enabled for any queues anymore, then return to
1811 * the default state of Data Transmission Arbitration on
1814 if (!is_any_cbs_enabled(adapter)) {
1815 tqavctrl = rd32(E1000_I210_TQAVCTRL);
1816 tqavctrl &= ~E1000_TQAVCTRL_DATATRANARB;
1817 wr32(E1000_I210_TQAVCTRL, tqavctrl);
1821 /* If LaunchTime is enabled, set DataTranTIM. */
1822 if (ring->launchtime_enable) {
1823 /* Always set DataTranTIM on TQAVCTRL if LaunchTime is enabled
1824 * for any of the SR queues, and configure fetchtime delta.
1826 * - LaunchTime will be enabled for all SR queues.
1827 * - A fixed offset can be added relative to the launch
1828 * time of all packets if configured at reg LAUNCH_OS0.
1829 * We are keeping it as 0 for now (default value).
1831 tqavctrl = rd32(E1000_I210_TQAVCTRL);
1832 tqavctrl |= E1000_TQAVCTRL_DATATRANTIM |
1833 E1000_TQAVCTRL_FETCHTIME_DELTA;
1834 wr32(E1000_I210_TQAVCTRL, tqavctrl);
1836 /* If Launchtime is not enabled for any SR queues anymore,
1837 * then clear DataTranTIM on TQAVCTRL and clear fetchtime delta,
1838 * effectively disabling Launchtime.
1840 if (!is_any_txtime_enabled(adapter)) {
1841 tqavctrl = rd32(E1000_I210_TQAVCTRL);
1842 tqavctrl &= ~E1000_TQAVCTRL_DATATRANTIM;
1843 tqavctrl &= ~E1000_TQAVCTRL_FETCHTIME_DELTA;
1844 wr32(E1000_I210_TQAVCTRL, tqavctrl);
1848 /* XXX: In i210 controller the sendSlope and loCredit parameters from
1849 * CBS are not configurable by software so we don't do any 'controller
1850 * configuration' in respect to these parameters.
1853 netdev_dbg(netdev, "Qav Tx mode: cbs %s, launchtime %s, queue %d idleslope %d sendslope %d hiCredit %d locredit %d\n",
1854 ring->cbs_enable ? "enabled" : "disabled",
1855 ring->launchtime_enable ? "enabled" : "disabled",
1857 ring->idleslope, ring->sendslope,
1858 ring->hicredit, ring->locredit);
1861 static int igb_save_txtime_params(struct igb_adapter *adapter, int queue,
1864 struct igb_ring *ring;
1866 if (queue < 0 || queue > adapter->num_tx_queues)
1869 ring = adapter->tx_ring[queue];
1870 ring->launchtime_enable = enable;
1875 static int igb_save_cbs_params(struct igb_adapter *adapter, int queue,
1876 bool enable, int idleslope, int sendslope,
1877 int hicredit, int locredit)
1879 struct igb_ring *ring;
1881 if (queue < 0 || queue > adapter->num_tx_queues)
1884 ring = adapter->tx_ring[queue];
1886 ring->cbs_enable = enable;
1887 ring->idleslope = idleslope;
1888 ring->sendslope = sendslope;
1889 ring->hicredit = hicredit;
1890 ring->locredit = locredit;
1896 * igb_setup_tx_mode - Switch to/from Qav Tx mode when applicable
1897 * @adapter: pointer to adapter struct
1899 * Configure TQAVCTRL register switching the controller's Tx mode
1900 * if FQTSS mode is enabled or disabled. Additionally, will issue
1901 * a call to igb_config_tx_modes() per queue so any previously saved
1902 * Tx parameters are applied.
1904 static void igb_setup_tx_mode(struct igb_adapter *adapter)
1906 struct net_device *netdev = adapter->netdev;
1907 struct e1000_hw *hw = &adapter->hw;
1910 /* Only i210 controller supports changing the transmission mode. */
1911 if (hw->mac.type != e1000_i210)
1914 if (is_fqtss_enabled(adapter)) {
1917 /* Configure TQAVCTRL register: set transmit mode to 'Qav',
1918 * set data fetch arbitration to 'round robin', set SP_WAIT_SR
1919 * so SP queues wait for SR ones.
1921 val = rd32(E1000_I210_TQAVCTRL);
1922 val |= E1000_TQAVCTRL_XMIT_MODE | E1000_TQAVCTRL_SP_WAIT_SR;
1923 val &= ~E1000_TQAVCTRL_DATAFETCHARB;
1924 wr32(E1000_I210_TQAVCTRL, val);
1926 /* Configure Tx and Rx packet buffers sizes as described in
1927 * i210 datasheet section 7.2.7.7.
1929 val = rd32(E1000_TXPBS);
1930 val &= ~I210_TXPBSIZE_MASK;
1931 val |= I210_TXPBSIZE_PB0_6KB | I210_TXPBSIZE_PB1_6KB |
1932 I210_TXPBSIZE_PB2_6KB | I210_TXPBSIZE_PB3_6KB;
1933 wr32(E1000_TXPBS, val);
1935 val = rd32(E1000_RXPBS);
1936 val &= ~I210_RXPBSIZE_MASK;
1937 val |= I210_RXPBSIZE_PB_30KB;
1938 wr32(E1000_RXPBS, val);
1940 /* Section 8.12.9 states that MAX_TPKT_SIZE from DTXMXPKTSZ
1941 * register should not exceed the buffer size programmed in
1942 * TXPBS. The smallest buffer size programmed in TXPBS is 4kB
1943 * so according to the datasheet we should set MAX_TPKT_SIZE to
1946 * However, when we do so, no frame from queue 2 and 3 are
1947 * transmitted. It seems the MAX_TPKT_SIZE should not be great
1948 * or _equal_ to the buffer size programmed in TXPBS. For this
1949 * reason, we set MAX_ TPKT_SIZE to (4kB - 1) / 64.
1951 val = (4096 - 1) / 64;
1952 wr32(E1000_I210_DTXMXPKTSZ, val);
1954 /* Since FQTSS mode is enabled, apply any CBS configuration
1955 * previously set. If no previous CBS configuration has been
1956 * done, then the initial configuration is applied, which means
1959 max_queue = (adapter->num_tx_queues < I210_SR_QUEUES_NUM) ?
1960 adapter->num_tx_queues : I210_SR_QUEUES_NUM;
1962 for (i = 0; i < max_queue; i++) {
1963 igb_config_tx_modes(adapter, i);
1966 wr32(E1000_RXPBS, I210_RXPBSIZE_DEFAULT);
1967 wr32(E1000_TXPBS, I210_TXPBSIZE_DEFAULT);
1968 wr32(E1000_I210_DTXMXPKTSZ, I210_DTXMXPKTSZ_DEFAULT);
1970 val = rd32(E1000_I210_TQAVCTRL);
1971 /* According to Section 8.12.21, the other flags we've set when
1972 * enabling FQTSS are not relevant when disabling FQTSS so we
1973 * don't set they here.
1975 val &= ~E1000_TQAVCTRL_XMIT_MODE;
1976 wr32(E1000_I210_TQAVCTRL, val);
1979 netdev_dbg(netdev, "FQTSS %s\n", (is_fqtss_enabled(adapter)) ?
1980 "enabled" : "disabled");
1984 * igb_configure - configure the hardware for RX and TX
1985 * @adapter: private board structure
1987 static void igb_configure(struct igb_adapter *adapter)
1989 struct net_device *netdev = adapter->netdev;
1992 igb_get_hw_control(adapter);
1993 igb_set_rx_mode(netdev);
1994 igb_setup_tx_mode(adapter);
1996 igb_restore_vlan(adapter);
1998 igb_setup_tctl(adapter);
1999 igb_setup_mrqc(adapter);
2000 igb_setup_rctl(adapter);
2002 igb_nfc_filter_restore(adapter);
2003 igb_configure_tx(adapter);
2004 igb_configure_rx(adapter);
2006 igb_rx_fifo_flush_82575(&adapter->hw);
2008 /* call igb_desc_unused which always leaves
2009 * at least 1 descriptor unused to make sure
2010 * next_to_use != next_to_clean
2012 for (i = 0; i < adapter->num_rx_queues; i++) {
2013 struct igb_ring *ring = adapter->rx_ring[i];
2014 igb_alloc_rx_buffers(ring, igb_desc_unused(ring));
2019 * igb_power_up_link - Power up the phy/serdes link
2020 * @adapter: address of board private structure
2022 void igb_power_up_link(struct igb_adapter *adapter)
2024 igb_reset_phy(&adapter->hw);
2026 if (adapter->hw.phy.media_type == e1000_media_type_copper)
2027 igb_power_up_phy_copper(&adapter->hw);
2029 igb_power_up_serdes_link_82575(&adapter->hw);
2031 igb_setup_link(&adapter->hw);
2035 * igb_power_down_link - Power down the phy/serdes link
2036 * @adapter: address of board private structure
2038 static void igb_power_down_link(struct igb_adapter *adapter)
2040 if (adapter->hw.phy.media_type == e1000_media_type_copper)
2041 igb_power_down_phy_copper_82575(&adapter->hw);
2043 igb_shutdown_serdes_link_82575(&adapter->hw);
2047 * igb_check_swap_media - Detect and switch function for Media Auto Sense
2048 * @adapter: address of the board private structure
2050 static void igb_check_swap_media(struct igb_adapter *adapter)
2052 struct e1000_hw *hw = &adapter->hw;
2053 u32 ctrl_ext, connsw;
2054 bool swap_now = false;
2056 ctrl_ext = rd32(E1000_CTRL_EXT);
2057 connsw = rd32(E1000_CONNSW);
2059 /* need to live swap if current media is copper and we have fiber/serdes
2063 if ((hw->phy.media_type == e1000_media_type_copper) &&
2064 (!(connsw & E1000_CONNSW_AUTOSENSE_EN))) {
2066 } else if ((hw->phy.media_type != e1000_media_type_copper) &&
2067 !(connsw & E1000_CONNSW_SERDESD)) {
2068 /* copper signal takes time to appear */
2069 if (adapter->copper_tries < 4) {
2070 adapter->copper_tries++;
2071 connsw |= E1000_CONNSW_AUTOSENSE_CONF;
2072 wr32(E1000_CONNSW, connsw);
2075 adapter->copper_tries = 0;
2076 if ((connsw & E1000_CONNSW_PHYSD) &&
2077 (!(connsw & E1000_CONNSW_PHY_PDN))) {
2079 connsw &= ~E1000_CONNSW_AUTOSENSE_CONF;
2080 wr32(E1000_CONNSW, connsw);
2088 switch (hw->phy.media_type) {
2089 case e1000_media_type_copper:
2090 netdev_info(adapter->netdev,
2091 "MAS: changing media to fiber/serdes\n");
2093 E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
2094 adapter->flags |= IGB_FLAG_MEDIA_RESET;
2095 adapter->copper_tries = 0;
2097 case e1000_media_type_internal_serdes:
2098 case e1000_media_type_fiber:
2099 netdev_info(adapter->netdev,
2100 "MAS: changing media to copper\n");
2102 ~E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
2103 adapter->flags |= IGB_FLAG_MEDIA_RESET;
2106 /* shouldn't get here during regular operation */
2107 netdev_err(adapter->netdev,
2108 "AMS: Invalid media type found, returning\n");
2111 wr32(E1000_CTRL_EXT, ctrl_ext);
2115 * igb_up - Open the interface and prepare it to handle traffic
2116 * @adapter: board private structure
2118 int igb_up(struct igb_adapter *adapter)
2120 struct e1000_hw *hw = &adapter->hw;
2123 /* hardware has been reset, we need to reload some things */
2124 igb_configure(adapter);
2126 clear_bit(__IGB_DOWN, &adapter->state);
2128 for (i = 0; i < adapter->num_q_vectors; i++)
2129 napi_enable(&(adapter->q_vector[i]->napi));
2131 if (adapter->flags & IGB_FLAG_HAS_MSIX)
2132 igb_configure_msix(adapter);
2134 igb_assign_vector(adapter->q_vector[0], 0);
2136 /* Clear any pending interrupts. */
2139 igb_irq_enable(adapter);
2141 /* notify VFs that reset has been completed */
2142 if (adapter->vfs_allocated_count) {
2143 u32 reg_data = rd32(E1000_CTRL_EXT);
2145 reg_data |= E1000_CTRL_EXT_PFRSTD;
2146 wr32(E1000_CTRL_EXT, reg_data);
2149 netif_tx_start_all_queues(adapter->netdev);
2151 /* start the watchdog. */
2152 hw->mac.get_link_status = 1;
2153 schedule_work(&adapter->watchdog_task);
2155 if ((adapter->flags & IGB_FLAG_EEE) &&
2156 (!hw->dev_spec._82575.eee_disable))
2157 adapter->eee_advert = MDIO_EEE_100TX | MDIO_EEE_1000T;
2162 void igb_down(struct igb_adapter *adapter)
2164 struct net_device *netdev = adapter->netdev;
2165 struct e1000_hw *hw = &adapter->hw;
2169 /* signal that we're down so the interrupt handler does not
2170 * reschedule our watchdog timer
2172 set_bit(__IGB_DOWN, &adapter->state);
2174 /* disable receives in the hardware */
2175 rctl = rd32(E1000_RCTL);
2176 wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
2177 /* flush and sleep below */
2179 igb_nfc_filter_exit(adapter);
2181 netif_carrier_off(netdev);
2182 netif_tx_stop_all_queues(netdev);
2184 /* disable transmits in the hardware */
2185 tctl = rd32(E1000_TCTL);
2186 tctl &= ~E1000_TCTL_EN;
2187 wr32(E1000_TCTL, tctl);
2188 /* flush both disables and wait for them to finish */
2190 usleep_range(10000, 11000);
2192 igb_irq_disable(adapter);
2194 adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
2196 for (i = 0; i < adapter->num_q_vectors; i++) {
2197 if (adapter->q_vector[i]) {
2198 napi_synchronize(&adapter->q_vector[i]->napi);
2199 napi_disable(&adapter->q_vector[i]->napi);
2203 del_timer_sync(&adapter->watchdog_timer);
2204 del_timer_sync(&adapter->phy_info_timer);
2206 /* record the stats before reset*/
2207 spin_lock(&adapter->stats64_lock);
2208 igb_update_stats(adapter);
2209 spin_unlock(&adapter->stats64_lock);
2211 adapter->link_speed = 0;
2212 adapter->link_duplex = 0;
2214 if (!pci_channel_offline(adapter->pdev))
2217 /* clear VLAN promisc flag so VFTA will be updated if necessary */
2218 adapter->flags &= ~IGB_FLAG_VLAN_PROMISC;
2220 igb_clean_all_tx_rings(adapter);
2221 igb_clean_all_rx_rings(adapter);
2222 #ifdef CONFIG_IGB_DCA
2224 /* since we reset the hardware DCA settings were cleared */
2225 igb_setup_dca(adapter);
2229 void igb_reinit_locked(struct igb_adapter *adapter)
2231 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
2232 usleep_range(1000, 2000);
2235 clear_bit(__IGB_RESETTING, &adapter->state);
2238 /** igb_enable_mas - Media Autosense re-enable after swap
2240 * @adapter: adapter struct
2242 static void igb_enable_mas(struct igb_adapter *adapter)
2244 struct e1000_hw *hw = &adapter->hw;
2245 u32 connsw = rd32(E1000_CONNSW);
2247 /* configure for SerDes media detect */
2248 if ((hw->phy.media_type == e1000_media_type_copper) &&
2249 (!(connsw & E1000_CONNSW_SERDESD))) {
2250 connsw |= E1000_CONNSW_ENRGSRC;
2251 connsw |= E1000_CONNSW_AUTOSENSE_EN;
2252 wr32(E1000_CONNSW, connsw);
2257 #ifdef CONFIG_IGB_HWMON
2259 * igb_set_i2c_bb - Init I2C interface
2260 * @hw: pointer to hardware structure
2262 static void igb_set_i2c_bb(struct e1000_hw *hw)
2267 ctrl_ext = rd32(E1000_CTRL_EXT);
2268 ctrl_ext |= E1000_CTRL_I2C_ENA;
2269 wr32(E1000_CTRL_EXT, ctrl_ext);
2272 i2cctl = rd32(E1000_I2CPARAMS);
2273 i2cctl |= E1000_I2CBB_EN
2274 | E1000_I2C_CLK_OE_N
2275 | E1000_I2C_DATA_OE_N;
2276 wr32(E1000_I2CPARAMS, i2cctl);
2281 void igb_reset(struct igb_adapter *adapter)
2283 struct pci_dev *pdev = adapter->pdev;
2284 struct e1000_hw *hw = &adapter->hw;
2285 struct e1000_mac_info *mac = &hw->mac;
2286 struct e1000_fc_info *fc = &hw->fc;
2289 /* Repartition Pba for greater than 9k mtu
2290 * To take effect CTRL.RST is required.
2292 switch (mac->type) {
2296 pba = rd32(E1000_RXPBS);
2297 pba = igb_rxpbs_adjust_82580(pba);
2300 pba = rd32(E1000_RXPBS);
2301 pba &= E1000_RXPBS_SIZE_MASK_82576;
2307 pba = E1000_PBA_34K;
2311 if (mac->type == e1000_82575) {
2312 u32 min_rx_space, min_tx_space, needed_tx_space;
2314 /* write Rx PBA so that hardware can report correct Tx PBA */
2315 wr32(E1000_PBA, pba);
2317 /* To maintain wire speed transmits, the Tx FIFO should be
2318 * large enough to accommodate two full transmit packets,
2319 * rounded up to the next 1KB and expressed in KB. Likewise,
2320 * the Rx FIFO should be large enough to accommodate at least
2321 * one full receive packet and is similarly rounded up and
2324 min_rx_space = DIV_ROUND_UP(MAX_JUMBO_FRAME_SIZE, 1024);
2326 /* The Tx FIFO also stores 16 bytes of information about the Tx
2327 * but don't include Ethernet FCS because hardware appends it.
2328 * We only need to round down to the nearest 512 byte block
2329 * count since the value we care about is 2 frames, not 1.
2331 min_tx_space = adapter->max_frame_size;
2332 min_tx_space += sizeof(union e1000_adv_tx_desc) - ETH_FCS_LEN;
2333 min_tx_space = DIV_ROUND_UP(min_tx_space, 512);
2335 /* upper 16 bits has Tx packet buffer allocation size in KB */
2336 needed_tx_space = min_tx_space - (rd32(E1000_PBA) >> 16);
2338 /* If current Tx allocation is less than the min Tx FIFO size,
2339 * and the min Tx FIFO size is less than the current Rx FIFO
2340 * allocation, take space away from current Rx allocation.
2342 if (needed_tx_space < pba) {
2343 pba -= needed_tx_space;
2345 /* if short on Rx space, Rx wins and must trump Tx
2348 if (pba < min_rx_space)
2352 /* adjust PBA for jumbo frames */
2353 wr32(E1000_PBA, pba);
2356 /* flow control settings
2357 * The high water mark must be low enough to fit one full frame
2358 * after transmitting the pause frame. As such we must have enough
2359 * space to allow for us to complete our current transmit and then
2360 * receive the frame that is in progress from the link partner.
2362 * - the full Rx FIFO size minus one full Tx plus one full Rx frame
2364 hwm = (pba << 10) - (adapter->max_frame_size + MAX_JUMBO_FRAME_SIZE);
2366 fc->high_water = hwm & 0xFFFFFFF0; /* 16-byte granularity */
2367 fc->low_water = fc->high_water - 16;
2368 fc->pause_time = 0xFFFF;
2370 fc->current_mode = fc->requested_mode;
2372 /* disable receive for all VFs and wait one second */
2373 if (adapter->vfs_allocated_count) {
2376 for (i = 0 ; i < adapter->vfs_allocated_count; i++)
2377 adapter->vf_data[i].flags &= IGB_VF_FLAG_PF_SET_MAC;
2379 /* ping all the active vfs to let them know we are going down */
2380 igb_ping_all_vfs(adapter);
2382 /* disable transmits and receives */
2383 wr32(E1000_VFRE, 0);
2384 wr32(E1000_VFTE, 0);
2387 /* Allow time for pending master requests to run */
2388 hw->mac.ops.reset_hw(hw);
2391 if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
2392 /* need to resetup here after media swap */
2393 adapter->ei.get_invariants(hw);
2394 adapter->flags &= ~IGB_FLAG_MEDIA_RESET;
2396 if ((mac->type == e1000_82575 || mac->type == e1000_i350) &&
2397 (adapter->flags & IGB_FLAG_MAS_ENABLE)) {
2398 igb_enable_mas(adapter);
2400 if (hw->mac.ops.init_hw(hw))
2401 dev_err(&pdev->dev, "Hardware Error\n");
2403 /* RAR registers were cleared during init_hw, clear mac table */
2404 igb_flush_mac_table(adapter);
2405 __dev_uc_unsync(adapter->netdev, NULL);
2407 /* Recover default RAR entry */
2408 igb_set_default_mac_filter(adapter);
2410 /* Flow control settings reset on hardware reset, so guarantee flow
2411 * control is off when forcing speed.
2413 if (!hw->mac.autoneg)
2414 igb_force_mac_fc(hw);
2416 igb_init_dmac(adapter, pba);
2417 #ifdef CONFIG_IGB_HWMON
2418 /* Re-initialize the thermal sensor on i350 devices. */
2419 if (!test_bit(__IGB_DOWN, &adapter->state)) {
2420 if (mac->type == e1000_i350 && hw->bus.func == 0) {
2421 /* If present, re-initialize the external thermal sensor
2426 mac->ops.init_thermal_sensor_thresh(hw);
2430 /* Re-establish EEE setting */
2431 if (hw->phy.media_type == e1000_media_type_copper) {
2432 switch (mac->type) {
2436 igb_set_eee_i350(hw, true, true);
2439 igb_set_eee_i354(hw, true, true);
2445 if (!netif_running(adapter->netdev))
2446 igb_power_down_link(adapter);
2448 igb_update_mng_vlan(adapter);
2450 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2451 wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
2453 /* Re-enable PTP, where applicable. */
2454 if (adapter->ptp_flags & IGB_PTP_ENABLED)
2455 igb_ptp_reset(adapter);
2457 igb_get_phy_info(hw);
2460 static netdev_features_t igb_fix_features(struct net_device *netdev,
2461 netdev_features_t features)
2463 /* Since there is no support for separate Rx/Tx vlan accel
2464 * enable/disable make sure Tx flag is always in same state as Rx.
2466 if (features & NETIF_F_HW_VLAN_CTAG_RX)
2467 features |= NETIF_F_HW_VLAN_CTAG_TX;
2469 features &= ~NETIF_F_HW_VLAN_CTAG_TX;
2474 static int igb_set_features(struct net_device *netdev,
2475 netdev_features_t features)
2477 netdev_features_t changed = netdev->features ^ features;
2478 struct igb_adapter *adapter = netdev_priv(netdev);
2480 if (changed & NETIF_F_HW_VLAN_CTAG_RX)
2481 igb_vlan_mode(netdev, features);
2483 if (!(changed & (NETIF_F_RXALL | NETIF_F_NTUPLE)))
2486 if (!(features & NETIF_F_NTUPLE)) {
2487 struct hlist_node *node2;
2488 struct igb_nfc_filter *rule;
2490 spin_lock(&adapter->nfc_lock);
2491 hlist_for_each_entry_safe(rule, node2,
2492 &adapter->nfc_filter_list, nfc_node) {
2493 igb_erase_filter(adapter, rule);
2494 hlist_del(&rule->nfc_node);
2497 spin_unlock(&adapter->nfc_lock);
2498 adapter->nfc_filter_count = 0;
2501 netdev->features = features;
2503 if (netif_running(netdev))
2504 igb_reinit_locked(adapter);
2511 static int igb_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
2512 struct net_device *dev,
2513 const unsigned char *addr, u16 vid,
2515 struct netlink_ext_ack *extack)
2517 /* guarantee we can provide a unique filter for the unicast address */
2518 if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
2519 struct igb_adapter *adapter = netdev_priv(dev);
2520 int vfn = adapter->vfs_allocated_count;
2522 if (netdev_uc_count(dev) >= igb_available_rars(adapter, vfn))
2526 return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags);
2529 #define IGB_MAX_MAC_HDR_LEN 127
2530 #define IGB_MAX_NETWORK_HDR_LEN 511
2532 static netdev_features_t
2533 igb_features_check(struct sk_buff *skb, struct net_device *dev,
2534 netdev_features_t features)
2536 unsigned int network_hdr_len, mac_hdr_len;
2538 /* Make certain the headers can be described by a context descriptor */
2539 mac_hdr_len = skb_network_header(skb) - skb->data;
2540 if (unlikely(mac_hdr_len > IGB_MAX_MAC_HDR_LEN))
2541 return features & ~(NETIF_F_HW_CSUM |
2543 NETIF_F_GSO_UDP_L4 |
2544 NETIF_F_HW_VLAN_CTAG_TX |
2548 network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
2549 if (unlikely(network_hdr_len > IGB_MAX_NETWORK_HDR_LEN))
2550 return features & ~(NETIF_F_HW_CSUM |
2552 NETIF_F_GSO_UDP_L4 |
2556 /* We can only support IPV4 TSO in tunnels if we can mangle the
2557 * inner IP ID field, so strip TSO if MANGLEID is not supported.
2559 if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
2560 features &= ~NETIF_F_TSO;
2565 static void igb_offload_apply(struct igb_adapter *adapter, s32 queue)
2567 if (!is_fqtss_enabled(adapter)) {
2568 enable_fqtss(adapter, true);
2572 igb_config_tx_modes(adapter, queue);
2574 if (!is_any_cbs_enabled(adapter) && !is_any_txtime_enabled(adapter))
2575 enable_fqtss(adapter, false);
2578 static int igb_offload_cbs(struct igb_adapter *adapter,
2579 struct tc_cbs_qopt_offload *qopt)
2581 struct e1000_hw *hw = &adapter->hw;
2584 /* CBS offloading is only supported by i210 controller. */
2585 if (hw->mac.type != e1000_i210)
2588 /* CBS offloading is only supported by queue 0 and queue 1. */
2589 if (qopt->queue < 0 || qopt->queue > 1)
2592 err = igb_save_cbs_params(adapter, qopt->queue, qopt->enable,
2593 qopt->idleslope, qopt->sendslope,
2594 qopt->hicredit, qopt->locredit);
2598 igb_offload_apply(adapter, qopt->queue);
2603 #define ETHER_TYPE_FULL_MASK ((__force __be16)~0)
2604 #define VLAN_PRIO_FULL_MASK (0x07)
2606 static int igb_parse_cls_flower(struct igb_adapter *adapter,
2607 struct flow_cls_offload *f,
2609 struct igb_nfc_filter *input)
2611 struct flow_rule *rule = flow_cls_offload_flow_rule(f);
2612 struct flow_dissector *dissector = rule->match.dissector;
2613 struct netlink_ext_ack *extack = f->common.extack;
2615 if (dissector->used_keys &
2616 ~(BIT(FLOW_DISSECTOR_KEY_BASIC) |
2617 BIT(FLOW_DISSECTOR_KEY_CONTROL) |
2618 BIT(FLOW_DISSECTOR_KEY_ETH_ADDRS) |
2619 BIT(FLOW_DISSECTOR_KEY_VLAN))) {
2620 NL_SET_ERR_MSG_MOD(extack,
2621 "Unsupported key used, only BASIC, CONTROL, ETH_ADDRS and VLAN are supported");
2625 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ETH_ADDRS)) {
2626 struct flow_match_eth_addrs match;
2628 flow_rule_match_eth_addrs(rule, &match);
2629 if (!is_zero_ether_addr(match.mask->dst)) {
2630 if (!is_broadcast_ether_addr(match.mask->dst)) {
2631 NL_SET_ERR_MSG_MOD(extack, "Only full masks are supported for destination MAC address");
2635 input->filter.match_flags |=
2636 IGB_FILTER_FLAG_DST_MAC_ADDR;
2637 ether_addr_copy(input->filter.dst_addr, match.key->dst);
2640 if (!is_zero_ether_addr(match.mask->src)) {
2641 if (!is_broadcast_ether_addr(match.mask->src)) {
2642 NL_SET_ERR_MSG_MOD(extack, "Only full masks are supported for source MAC address");
2646 input->filter.match_flags |=
2647 IGB_FILTER_FLAG_SRC_MAC_ADDR;
2648 ether_addr_copy(input->filter.src_addr, match.key->src);
2652 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_BASIC)) {
2653 struct flow_match_basic match;
2655 flow_rule_match_basic(rule, &match);
2656 if (match.mask->n_proto) {
2657 if (match.mask->n_proto != ETHER_TYPE_FULL_MASK) {
2658 NL_SET_ERR_MSG_MOD(extack, "Only full mask is supported for EtherType filter");
2662 input->filter.match_flags |= IGB_FILTER_FLAG_ETHER_TYPE;
2663 input->filter.etype = match.key->n_proto;
2667 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_VLAN)) {
2668 struct flow_match_vlan match;
2670 flow_rule_match_vlan(rule, &match);
2671 if (match.mask->vlan_priority) {
2672 if (match.mask->vlan_priority != VLAN_PRIO_FULL_MASK) {
2673 NL_SET_ERR_MSG_MOD(extack, "Only full mask is supported for VLAN priority");
2677 input->filter.match_flags |= IGB_FILTER_FLAG_VLAN_TCI;
2678 input->filter.vlan_tci =
2679 (__force __be16)match.key->vlan_priority;
2683 input->action = traffic_class;
2684 input->cookie = f->cookie;
2689 static int igb_configure_clsflower(struct igb_adapter *adapter,
2690 struct flow_cls_offload *cls_flower)
2692 struct netlink_ext_ack *extack = cls_flower->common.extack;
2693 struct igb_nfc_filter *filter, *f;
2696 tc = tc_classid_to_hwtc(adapter->netdev, cls_flower->classid);
2698 NL_SET_ERR_MSG_MOD(extack, "Invalid traffic class");
2702 filter = kzalloc(sizeof(*filter), GFP_KERNEL);
2706 err = igb_parse_cls_flower(adapter, cls_flower, tc, filter);
2710 spin_lock(&adapter->nfc_lock);
2712 hlist_for_each_entry(f, &adapter->nfc_filter_list, nfc_node) {
2713 if (!memcmp(&f->filter, &filter->filter, sizeof(f->filter))) {
2715 NL_SET_ERR_MSG_MOD(extack,
2716 "This filter is already set in ethtool");
2721 hlist_for_each_entry(f, &adapter->cls_flower_list, nfc_node) {
2722 if (!memcmp(&f->filter, &filter->filter, sizeof(f->filter))) {
2724 NL_SET_ERR_MSG_MOD(extack,
2725 "This filter is already set in cls_flower");
2730 err = igb_add_filter(adapter, filter);
2732 NL_SET_ERR_MSG_MOD(extack, "Could not add filter to the adapter");
2736 hlist_add_head(&filter->nfc_node, &adapter->cls_flower_list);
2738 spin_unlock(&adapter->nfc_lock);
2743 spin_unlock(&adapter->nfc_lock);
2751 static int igb_delete_clsflower(struct igb_adapter *adapter,
2752 struct flow_cls_offload *cls_flower)
2754 struct igb_nfc_filter *filter;
2757 spin_lock(&adapter->nfc_lock);
2759 hlist_for_each_entry(filter, &adapter->cls_flower_list, nfc_node)
2760 if (filter->cookie == cls_flower->cookie)
2768 err = igb_erase_filter(adapter, filter);
2772 hlist_del(&filter->nfc_node);
2776 spin_unlock(&adapter->nfc_lock);
2781 static int igb_setup_tc_cls_flower(struct igb_adapter *adapter,
2782 struct flow_cls_offload *cls_flower)
2784 switch (cls_flower->command) {
2785 case FLOW_CLS_REPLACE:
2786 return igb_configure_clsflower(adapter, cls_flower);
2787 case FLOW_CLS_DESTROY:
2788 return igb_delete_clsflower(adapter, cls_flower);
2789 case FLOW_CLS_STATS:
2796 static int igb_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
2799 struct igb_adapter *adapter = cb_priv;
2801 if (!tc_cls_can_offload_and_chain0(adapter->netdev, type_data))
2805 case TC_SETUP_CLSFLOWER:
2806 return igb_setup_tc_cls_flower(adapter, type_data);
2813 static int igb_offload_txtime(struct igb_adapter *adapter,
2814 struct tc_etf_qopt_offload *qopt)
2816 struct e1000_hw *hw = &adapter->hw;
2819 /* Launchtime offloading is only supported by i210 controller. */
2820 if (hw->mac.type != e1000_i210)
2823 /* Launchtime offloading is only supported by queues 0 and 1. */
2824 if (qopt->queue < 0 || qopt->queue > 1)
2827 err = igb_save_txtime_params(adapter, qopt->queue, qopt->enable);
2831 igb_offload_apply(adapter, qopt->queue);
2836 static int igb_tc_query_caps(struct igb_adapter *adapter,
2837 struct tc_query_caps_base *base)
2839 switch (base->type) {
2840 case TC_SETUP_QDISC_TAPRIO: {
2841 struct tc_taprio_caps *caps = base->caps;
2843 caps->broken_mqprio = true;
2852 static LIST_HEAD(igb_block_cb_list);
2854 static int igb_setup_tc(struct net_device *dev, enum tc_setup_type type,
2857 struct igb_adapter *adapter = netdev_priv(dev);
2861 return igb_tc_query_caps(adapter, type_data);
2862 case TC_SETUP_QDISC_CBS:
2863 return igb_offload_cbs(adapter, type_data);
2864 case TC_SETUP_BLOCK:
2865 return flow_block_cb_setup_simple(type_data,
2867 igb_setup_tc_block_cb,
2868 adapter, adapter, true);
2870 case TC_SETUP_QDISC_ETF:
2871 return igb_offload_txtime(adapter, type_data);
2878 static int igb_xdp_setup(struct net_device *dev, struct netdev_bpf *bpf)
2880 int i, frame_size = dev->mtu + IGB_ETH_PKT_HDR_PAD;
2881 struct igb_adapter *adapter = netdev_priv(dev);
2882 struct bpf_prog *prog = bpf->prog, *old_prog;
2883 bool running = netif_running(dev);
2886 /* verify igb ring attributes are sufficient for XDP */
2887 for (i = 0; i < adapter->num_rx_queues; i++) {
2888 struct igb_ring *ring = adapter->rx_ring[i];
2890 if (frame_size > igb_rx_bufsz(ring)) {
2891 NL_SET_ERR_MSG_MOD(bpf->extack,
2892 "The RX buffer size is too small for the frame size");
2893 netdev_warn(dev, "XDP RX buffer size %d is too small for the frame size %d\n",
2894 igb_rx_bufsz(ring), frame_size);
2899 old_prog = xchg(&adapter->xdp_prog, prog);
2900 need_reset = (!!prog != !!old_prog);
2902 /* device is up and bpf is added/removed, must setup the RX queues */
2903 if (need_reset && running) {
2906 for (i = 0; i < adapter->num_rx_queues; i++)
2907 (void)xchg(&adapter->rx_ring[i]->xdp_prog,
2912 bpf_prog_put(old_prog);
2914 /* bpf is just replaced, RXQ and MTU are already setup */
2919 xdp_features_set_redirect_target(dev, true);
2921 xdp_features_clear_redirect_target(dev);
2930 static int igb_xdp(struct net_device *dev, struct netdev_bpf *xdp)
2932 switch (xdp->command) {
2933 case XDP_SETUP_PROG:
2934 return igb_xdp_setup(dev, xdp);
2940 static void igb_xdp_ring_update_tail(struct igb_ring *ring)
2942 /* Force memory writes to complete before letting h/w know there
2943 * are new descriptors to fetch.
2946 writel(ring->next_to_use, ring->tail);
2949 static struct igb_ring *igb_xdp_tx_queue_mapping(struct igb_adapter *adapter)
2951 unsigned int r_idx = smp_processor_id();
2953 if (r_idx >= adapter->num_tx_queues)
2954 r_idx = r_idx % adapter->num_tx_queues;
2956 return adapter->tx_ring[r_idx];
2959 static int igb_xdp_xmit_back(struct igb_adapter *adapter, struct xdp_buff *xdp)
2961 struct xdp_frame *xdpf = xdp_convert_buff_to_frame(xdp);
2962 int cpu = smp_processor_id();
2963 struct igb_ring *tx_ring;
2964 struct netdev_queue *nq;
2967 if (unlikely(!xdpf))
2968 return IGB_XDP_CONSUMED;
2970 /* During program transitions its possible adapter->xdp_prog is assigned
2971 * but ring has not been configured yet. In this case simply abort xmit.
2973 tx_ring = adapter->xdp_prog ? igb_xdp_tx_queue_mapping(adapter) : NULL;
2974 if (unlikely(!tx_ring))
2975 return IGB_XDP_CONSUMED;
2977 nq = txring_txq(tx_ring);
2978 __netif_tx_lock(nq, cpu);
2979 /* Avoid transmit queue timeout since we share it with the slow path */
2980 txq_trans_cond_update(nq);
2981 ret = igb_xmit_xdp_ring(adapter, tx_ring, xdpf);
2982 __netif_tx_unlock(nq);
2987 static int igb_xdp_xmit(struct net_device *dev, int n,
2988 struct xdp_frame **frames, u32 flags)
2990 struct igb_adapter *adapter = netdev_priv(dev);
2991 int cpu = smp_processor_id();
2992 struct igb_ring *tx_ring;
2993 struct netdev_queue *nq;
2997 if (unlikely(test_bit(__IGB_DOWN, &adapter->state)))
3000 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
3003 /* During program transitions its possible adapter->xdp_prog is assigned
3004 * but ring has not been configured yet. In this case simply abort xmit.
3006 tx_ring = adapter->xdp_prog ? igb_xdp_tx_queue_mapping(adapter) : NULL;
3007 if (unlikely(!tx_ring))
3010 nq = txring_txq(tx_ring);
3011 __netif_tx_lock(nq, cpu);
3013 /* Avoid transmit queue timeout since we share it with the slow path */
3014 txq_trans_cond_update(nq);
3016 for (i = 0; i < n; i++) {
3017 struct xdp_frame *xdpf = frames[i];
3020 err = igb_xmit_xdp_ring(adapter, tx_ring, xdpf);
3021 if (err != IGB_XDP_TX)
3026 __netif_tx_unlock(nq);
3028 if (unlikely(flags & XDP_XMIT_FLUSH))
3029 igb_xdp_ring_update_tail(tx_ring);
3034 static const struct net_device_ops igb_netdev_ops = {
3035 .ndo_open = igb_open,
3036 .ndo_stop = igb_close,
3037 .ndo_start_xmit = igb_xmit_frame,
3038 .ndo_get_stats64 = igb_get_stats64,
3039 .ndo_set_rx_mode = igb_set_rx_mode,
3040 .ndo_set_mac_address = igb_set_mac,
3041 .ndo_change_mtu = igb_change_mtu,
3042 .ndo_eth_ioctl = igb_ioctl,
3043 .ndo_tx_timeout = igb_tx_timeout,
3044 .ndo_validate_addr = eth_validate_addr,
3045 .ndo_vlan_rx_add_vid = igb_vlan_rx_add_vid,
3046 .ndo_vlan_rx_kill_vid = igb_vlan_rx_kill_vid,
3047 .ndo_set_vf_mac = igb_ndo_set_vf_mac,
3048 .ndo_set_vf_vlan = igb_ndo_set_vf_vlan,
3049 .ndo_set_vf_rate = igb_ndo_set_vf_bw,
3050 .ndo_set_vf_spoofchk = igb_ndo_set_vf_spoofchk,
3051 .ndo_set_vf_trust = igb_ndo_set_vf_trust,
3052 .ndo_get_vf_config = igb_ndo_get_vf_config,
3053 .ndo_fix_features = igb_fix_features,
3054 .ndo_set_features = igb_set_features,
3055 .ndo_fdb_add = igb_ndo_fdb_add,
3056 .ndo_features_check = igb_features_check,
3057 .ndo_setup_tc = igb_setup_tc,
3059 .ndo_xdp_xmit = igb_xdp_xmit,
3063 * igb_set_fw_version - Configure version string for ethtool
3064 * @adapter: adapter struct
3066 void igb_set_fw_version(struct igb_adapter *adapter)
3068 struct e1000_hw *hw = &adapter->hw;
3069 struct e1000_fw_version fw;
3071 igb_get_fw_version(hw, &fw);
3073 switch (hw->mac.type) {
3076 if (!(igb_get_flash_presence_i210(hw))) {
3077 snprintf(adapter->fw_version,
3078 sizeof(adapter->fw_version),
3080 fw.invm_major, fw.invm_minor,
3086 /* if option is rom valid, display its version too */
3088 snprintf(adapter->fw_version,
3089 sizeof(adapter->fw_version),
3090 "%d.%d, 0x%08x, %d.%d.%d",
3091 fw.eep_major, fw.eep_minor, fw.etrack_id,
3092 fw.or_major, fw.or_build, fw.or_patch);
3094 } else if (fw.etrack_id != 0X0000) {
3095 snprintf(adapter->fw_version,
3096 sizeof(adapter->fw_version),
3098 fw.eep_major, fw.eep_minor, fw.etrack_id);
3100 snprintf(adapter->fw_version,
3101 sizeof(adapter->fw_version),
3103 fw.eep_major, fw.eep_minor, fw.eep_build);
3110 * igb_init_mas - init Media Autosense feature if enabled in the NVM
3112 * @adapter: adapter struct
3114 static void igb_init_mas(struct igb_adapter *adapter)
3116 struct e1000_hw *hw = &adapter->hw;
3119 hw->nvm.ops.read(hw, NVM_COMPAT, 1, &eeprom_data);
3120 switch (hw->bus.func) {
3122 if (eeprom_data & IGB_MAS_ENABLE_0) {
3123 adapter->flags |= IGB_FLAG_MAS_ENABLE;
3124 netdev_info(adapter->netdev,
3125 "MAS: Enabling Media Autosense for port %d\n",
3130 if (eeprom_data & IGB_MAS_ENABLE_1) {
3131 adapter->flags |= IGB_FLAG_MAS_ENABLE;
3132 netdev_info(adapter->netdev,
3133 "MAS: Enabling Media Autosense for port %d\n",
3138 if (eeprom_data & IGB_MAS_ENABLE_2) {
3139 adapter->flags |= IGB_FLAG_MAS_ENABLE;
3140 netdev_info(adapter->netdev,
3141 "MAS: Enabling Media Autosense for port %d\n",
3146 if (eeprom_data & IGB_MAS_ENABLE_3) {
3147 adapter->flags |= IGB_FLAG_MAS_ENABLE;
3148 netdev_info(adapter->netdev,
3149 "MAS: Enabling Media Autosense for port %d\n",
3154 /* Shouldn't get here */
3155 netdev_err(adapter->netdev,
3156 "MAS: Invalid port configuration, returning\n");
3162 * igb_init_i2c - Init I2C interface
3163 * @adapter: pointer to adapter structure
3165 static s32 igb_init_i2c(struct igb_adapter *adapter)
3169 /* I2C interface supported on i350 devices */
3170 if (adapter->hw.mac.type != e1000_i350)
3173 /* Initialize the i2c bus which is controlled by the registers.
3174 * This bus will use the i2c_algo_bit structure that implements
3175 * the protocol through toggling of the 4 bits in the register.
3177 adapter->i2c_adap.owner = THIS_MODULE;
3178 adapter->i2c_algo = igb_i2c_algo;
3179 adapter->i2c_algo.data = adapter;
3180 adapter->i2c_adap.algo_data = &adapter->i2c_algo;
3181 adapter->i2c_adap.dev.parent = &adapter->pdev->dev;
3182 strscpy(adapter->i2c_adap.name, "igb BB",
3183 sizeof(adapter->i2c_adap.name));
3184 status = i2c_bit_add_bus(&adapter->i2c_adap);
3189 * igb_probe - Device Initialization Routine
3190 * @pdev: PCI device information struct
3191 * @ent: entry in igb_pci_tbl
3193 * Returns 0 on success, negative on failure
3195 * igb_probe initializes an adapter identified by a pci_dev structure.
3196 * The OS initialization, configuring of the adapter private structure,
3197 * and a hardware reset occur.
3199 static int igb_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
3201 struct net_device *netdev;
3202 struct igb_adapter *adapter;
3203 struct e1000_hw *hw;
3204 u16 eeprom_data = 0;
3206 static int global_quad_port_a; /* global quad port a indication */
3207 const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
3208 u8 part_str[E1000_PBANUM_LENGTH];
3211 /* Catch broken hardware that put the wrong VF device ID in
3212 * the PCIe SR-IOV capability.
3214 if (pdev->is_virtfn) {
3215 WARN(1, KERN_ERR "%s (%x:%x) should not be a VF!\n",
3216 pci_name(pdev), pdev->vendor, pdev->device);
3220 err = pci_enable_device_mem(pdev);
3224 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
3227 "No usable DMA configuration, aborting\n");
3231 err = pci_request_mem_regions(pdev, igb_driver_name);
3235 pci_set_master(pdev);
3236 pci_save_state(pdev);
3239 netdev = alloc_etherdev_mq(sizeof(struct igb_adapter),
3242 goto err_alloc_etherdev;
3244 SET_NETDEV_DEV(netdev, &pdev->dev);
3246 pci_set_drvdata(pdev, netdev);
3247 adapter = netdev_priv(netdev);
3248 adapter->netdev = netdev;
3249 adapter->pdev = pdev;
3252 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
3255 adapter->io_addr = pci_iomap(pdev, 0, 0);
3256 if (!adapter->io_addr)
3258 /* hw->hw_addr can be altered, we'll use adapter->io_addr for unmap */
3259 hw->hw_addr = adapter->io_addr;
3261 netdev->netdev_ops = &igb_netdev_ops;
3262 igb_set_ethtool_ops(netdev);
3263 netdev->watchdog_timeo = 5 * HZ;
3265 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
3267 netdev->mem_start = pci_resource_start(pdev, 0);
3268 netdev->mem_end = pci_resource_end(pdev, 0);
3270 /* PCI config space info */
3271 hw->vendor_id = pdev->vendor;
3272 hw->device_id = pdev->device;
3273 hw->revision_id = pdev->revision;
3274 hw->subsystem_vendor_id = pdev->subsystem_vendor;
3275 hw->subsystem_device_id = pdev->subsystem_device;
3277 /* Copy the default MAC, PHY and NVM function pointers */
3278 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
3279 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
3280 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
3281 /* Initialize skew-specific constants */
3282 err = ei->get_invariants(hw);
3286 /* setup the private structure */
3287 err = igb_sw_init(adapter);
3291 igb_get_bus_info_pcie(hw);
3293 hw->phy.autoneg_wait_to_complete = false;
3295 /* Copper options */
3296 if (hw->phy.media_type == e1000_media_type_copper) {
3297 hw->phy.mdix = AUTO_ALL_MODES;
3298 hw->phy.disable_polarity_correction = false;
3299 hw->phy.ms_type = e1000_ms_hw_default;
3302 if (igb_check_reset_block(hw))
3303 dev_info(&pdev->dev,
3304 "PHY reset is blocked due to SOL/IDER session.\n");
3306 /* features is initialized to 0 in allocation, it might have bits
3307 * set by igb_sw_init so we should use an or instead of an
3310 netdev->features |= NETIF_F_SG |
3317 if (hw->mac.type >= e1000_82576)
3318 netdev->features |= NETIF_F_SCTP_CRC | NETIF_F_GSO_UDP_L4;
3320 if (hw->mac.type >= e1000_i350)
3321 netdev->features |= NETIF_F_HW_TC;
3323 #define IGB_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
3324 NETIF_F_GSO_GRE_CSUM | \
3325 NETIF_F_GSO_IPXIP4 | \
3326 NETIF_F_GSO_IPXIP6 | \
3327 NETIF_F_GSO_UDP_TUNNEL | \
3328 NETIF_F_GSO_UDP_TUNNEL_CSUM)
3330 netdev->gso_partial_features = IGB_GSO_PARTIAL_FEATURES;
3331 netdev->features |= NETIF_F_GSO_PARTIAL | IGB_GSO_PARTIAL_FEATURES;
3333 /* copy netdev features into list of user selectable features */
3334 netdev->hw_features |= netdev->features |
3335 NETIF_F_HW_VLAN_CTAG_RX |
3336 NETIF_F_HW_VLAN_CTAG_TX |
3339 if (hw->mac.type >= e1000_i350)
3340 netdev->hw_features |= NETIF_F_NTUPLE;
3342 netdev->features |= NETIF_F_HIGHDMA;
3344 netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
3345 netdev->mpls_features |= NETIF_F_HW_CSUM;
3346 netdev->hw_enc_features |= netdev->vlan_features;
3348 /* set this bit last since it cannot be part of vlan_features */
3349 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
3350 NETIF_F_HW_VLAN_CTAG_RX |
3351 NETIF_F_HW_VLAN_CTAG_TX;
3353 netdev->priv_flags |= IFF_SUPP_NOFCS;
3355 netdev->priv_flags |= IFF_UNICAST_FLT;
3356 netdev->xdp_features = NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_REDIRECT;
3358 /* MTU range: 68 - 9216 */
3359 netdev->min_mtu = ETH_MIN_MTU;
3360 netdev->max_mtu = MAX_STD_JUMBO_FRAME_SIZE;
3362 adapter->en_mng_pt = igb_enable_mng_pass_thru(hw);
3364 /* before reading the NVM, reset the controller to put the device in a
3365 * known good starting state
3367 hw->mac.ops.reset_hw(hw);
3369 /* make sure the NVM is good , i211/i210 parts can have special NVM
3370 * that doesn't contain a checksum
3372 switch (hw->mac.type) {
3375 if (igb_get_flash_presence_i210(hw)) {
3376 if (hw->nvm.ops.validate(hw) < 0) {
3378 "The NVM Checksum Is Not Valid\n");
3385 if (hw->nvm.ops.validate(hw) < 0) {
3386 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
3393 if (eth_platform_get_mac_address(&pdev->dev, hw->mac.addr)) {
3394 /* copy the MAC address out of the NVM */
3395 if (hw->mac.ops.read_mac_addr(hw))
3396 dev_err(&pdev->dev, "NVM Read Error\n");
3399 eth_hw_addr_set(netdev, hw->mac.addr);
3401 if (!is_valid_ether_addr(netdev->dev_addr)) {
3402 dev_err(&pdev->dev, "Invalid MAC Address\n");
3407 igb_set_default_mac_filter(adapter);
3409 /* get firmware version for ethtool -i */
3410 igb_set_fw_version(adapter);
3412 /* configure RXPBSIZE and TXPBSIZE */
3413 if (hw->mac.type == e1000_i210) {
3414 wr32(E1000_RXPBS, I210_RXPBSIZE_DEFAULT);
3415 wr32(E1000_TXPBS, I210_TXPBSIZE_DEFAULT);
3418 timer_setup(&adapter->watchdog_timer, igb_watchdog, 0);
3419 timer_setup(&adapter->phy_info_timer, igb_update_phy_info, 0);
3421 INIT_WORK(&adapter->reset_task, igb_reset_task);
3422 INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
3424 /* Initialize link properties that are user-changeable */
3425 adapter->fc_autoneg = true;
3426 hw->mac.autoneg = true;
3427 hw->phy.autoneg_advertised = 0x2f;
3429 hw->fc.requested_mode = e1000_fc_default;
3430 hw->fc.current_mode = e1000_fc_default;
3432 igb_validate_mdi_setting(hw);
3434 /* By default, support wake on port A */
3435 if (hw->bus.func == 0)
3436 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3438 /* Check the NVM for wake support on non-port A ports */
3439 if (hw->mac.type >= e1000_82580)
3440 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A +
3441 NVM_82580_LAN_FUNC_OFFSET(hw->bus.func), 1,
3443 else if (hw->bus.func == 1)
3444 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
3446 if (eeprom_data & IGB_EEPROM_APME)
3447 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3449 /* now that we have the eeprom settings, apply the special cases where
3450 * the eeprom may be wrong or the board simply won't support wake on
3451 * lan on a particular port
3453 switch (pdev->device) {
3454 case E1000_DEV_ID_82575GB_QUAD_COPPER:
3455 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
3457 case E1000_DEV_ID_82575EB_FIBER_SERDES:
3458 case E1000_DEV_ID_82576_FIBER:
3459 case E1000_DEV_ID_82576_SERDES:
3460 /* Wake events only supported on port A for dual fiber
3461 * regardless of eeprom setting
3463 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
3464 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
3466 case E1000_DEV_ID_82576_QUAD_COPPER:
3467 case E1000_DEV_ID_82576_QUAD_COPPER_ET2:
3468 /* if quad port adapter, disable WoL on all but port A */
3469 if (global_quad_port_a != 0)
3470 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
3472 adapter->flags |= IGB_FLAG_QUAD_PORT_A;
3473 /* Reset for multiple quad port adapters */
3474 if (++global_quad_port_a == 4)
3475 global_quad_port_a = 0;
3478 /* If the device can't wake, don't set software support */
3479 if (!device_can_wakeup(&adapter->pdev->dev))
3480 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
3483 /* initialize the wol settings based on the eeprom settings */
3484 if (adapter->flags & IGB_FLAG_WOL_SUPPORTED)
3485 adapter->wol |= E1000_WUFC_MAG;
3487 /* Some vendors want WoL disabled by default, but still supported */
3488 if ((hw->mac.type == e1000_i350) &&
3489 (pdev->subsystem_vendor == PCI_VENDOR_ID_HP)) {
3490 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3494 /* Some vendors want the ability to Use the EEPROM setting as
3495 * enable/disable only, and not for capability
3497 if (((hw->mac.type == e1000_i350) ||
3498 (hw->mac.type == e1000_i354)) &&
3499 (pdev->subsystem_vendor == PCI_VENDOR_ID_DELL)) {
3500 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3503 if (hw->mac.type == e1000_i350) {
3504 if (((pdev->subsystem_device == 0x5001) ||
3505 (pdev->subsystem_device == 0x5002)) &&
3506 (hw->bus.func == 0)) {
3507 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3510 if (pdev->subsystem_device == 0x1F52)
3511 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3514 device_set_wakeup_enable(&adapter->pdev->dev,
3515 adapter->flags & IGB_FLAG_WOL_SUPPORTED);
3517 /* reset the hardware with the new settings */
3520 /* Init the I2C interface */
3521 err = igb_init_i2c(adapter);
3523 dev_err(&pdev->dev, "failed to init i2c interface\n");
3527 /* let the f/w know that the h/w is now under the control of the
3530 igb_get_hw_control(adapter);
3532 strcpy(netdev->name, "eth%d");
3533 err = register_netdev(netdev);
3537 /* carrier off reporting is important to ethtool even BEFORE open */
3538 netif_carrier_off(netdev);
3540 #ifdef CONFIG_IGB_DCA
3541 if (dca_add_requester(&pdev->dev) == 0) {
3542 adapter->flags |= IGB_FLAG_DCA_ENABLED;
3543 dev_info(&pdev->dev, "DCA enabled\n");
3544 igb_setup_dca(adapter);
3548 #ifdef CONFIG_IGB_HWMON
3549 /* Initialize the thermal sensor on i350 devices. */
3550 if (hw->mac.type == e1000_i350 && hw->bus.func == 0) {
3553 /* Read the NVM to determine if this i350 device supports an
3554 * external thermal sensor.
3556 hw->nvm.ops.read(hw, NVM_ETS_CFG, 1, &ets_word);
3557 if (ets_word != 0x0000 && ets_word != 0xFFFF)
3558 adapter->ets = true;
3560 adapter->ets = false;
3561 /* Only enable I2C bit banging if an external thermal
3562 * sensor is supported.
3566 hw->mac.ops.init_thermal_sensor_thresh(hw);
3567 if (igb_sysfs_init(adapter))
3569 "failed to allocate sysfs resources\n");
3571 adapter->ets = false;
3574 /* Check if Media Autosense is enabled */
3576 if (hw->dev_spec._82575.mas_capable)
3577 igb_init_mas(adapter);
3579 /* do hw tstamp init after resetting */
3580 igb_ptp_init(adapter);
3582 dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
3583 /* print bus type/speed/width info, not applicable to i354 */
3584 if (hw->mac.type != e1000_i354) {
3585 dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
3587 ((hw->bus.speed == e1000_bus_speed_2500) ? "2.5Gb/s" :
3588 (hw->bus.speed == e1000_bus_speed_5000) ? "5.0Gb/s" :
3590 ((hw->bus.width == e1000_bus_width_pcie_x4) ?
3592 (hw->bus.width == e1000_bus_width_pcie_x2) ?
3594 (hw->bus.width == e1000_bus_width_pcie_x1) ?
3595 "Width x1" : "unknown"), netdev->dev_addr);
3598 if ((hw->mac.type == e1000_82576 &&
3599 rd32(E1000_EECD) & E1000_EECD_PRES) ||
3600 (hw->mac.type >= e1000_i210 ||
3601 igb_get_flash_presence_i210(hw))) {
3602 ret_val = igb_read_part_string(hw, part_str,
3603 E1000_PBANUM_LENGTH);
3605 ret_val = -E1000_ERR_INVM_VALUE_NOT_FOUND;
3609 strcpy(part_str, "Unknown");
3610 dev_info(&pdev->dev, "%s: PBA No: %s\n", netdev->name, part_str);
3611 dev_info(&pdev->dev,
3612 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
3613 (adapter->flags & IGB_FLAG_HAS_MSIX) ? "MSI-X" :
3614 (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
3615 adapter->num_rx_queues, adapter->num_tx_queues);
3616 if (hw->phy.media_type == e1000_media_type_copper) {
3617 switch (hw->mac.type) {
3621 /* Enable EEE for internal copper PHY devices */
3622 err = igb_set_eee_i350(hw, true, true);
3624 (!hw->dev_spec._82575.eee_disable)) {
3625 adapter->eee_advert =
3626 MDIO_EEE_100TX | MDIO_EEE_1000T;
3627 adapter->flags |= IGB_FLAG_EEE;
3631 if ((rd32(E1000_CTRL_EXT) &
3632 E1000_CTRL_EXT_LINK_MODE_SGMII)) {
3633 err = igb_set_eee_i354(hw, true, true);
3635 (!hw->dev_spec._82575.eee_disable)) {
3636 adapter->eee_advert =
3637 MDIO_EEE_100TX | MDIO_EEE_1000T;
3638 adapter->flags |= IGB_FLAG_EEE;
3647 dev_pm_set_driver_flags(&pdev->dev, DPM_FLAG_NO_DIRECT_COMPLETE);
3649 pm_runtime_put_noidle(&pdev->dev);
3653 igb_release_hw_control(adapter);
3654 memset(&adapter->i2c_adap, 0, sizeof(adapter->i2c_adap));
3656 if (!igb_check_reset_block(hw))
3659 if (hw->flash_address)
3660 iounmap(hw->flash_address);
3662 kfree(adapter->mac_table);
3663 kfree(adapter->shadow_vfta);
3664 igb_clear_interrupt_scheme(adapter);
3665 #ifdef CONFIG_PCI_IOV
3666 igb_disable_sriov(pdev, false);
3668 pci_iounmap(pdev, adapter->io_addr);
3670 free_netdev(netdev);
3672 pci_release_mem_regions(pdev);
3675 pci_disable_device(pdev);
3679 #ifdef CONFIG_PCI_IOV
3680 static int igb_sriov_reinit(struct pci_dev *dev)
3682 struct net_device *netdev = pci_get_drvdata(dev);
3683 struct igb_adapter *adapter = netdev_priv(netdev);
3684 struct pci_dev *pdev = adapter->pdev;
3688 if (netif_running(netdev))
3693 igb_clear_interrupt_scheme(adapter);
3695 igb_init_queue_configuration(adapter);
3697 if (igb_init_interrupt_scheme(adapter, true)) {
3699 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
3703 if (netif_running(netdev))
3711 static int igb_disable_sriov(struct pci_dev *pdev, bool reinit)
3713 struct net_device *netdev = pci_get_drvdata(pdev);
3714 struct igb_adapter *adapter = netdev_priv(netdev);
3715 struct e1000_hw *hw = &adapter->hw;
3716 unsigned long flags;
3718 /* reclaim resources allocated to VFs */
3719 if (adapter->vf_data) {
3720 /* disable iov and allow time for transactions to clear */
3721 if (pci_vfs_assigned(pdev)) {
3722 dev_warn(&pdev->dev,
3723 "Cannot deallocate SR-IOV virtual functions while they are assigned - VFs will not be deallocated\n");
3726 pci_disable_sriov(pdev);
3729 spin_lock_irqsave(&adapter->vfs_lock, flags);
3730 kfree(adapter->vf_mac_list);
3731 adapter->vf_mac_list = NULL;
3732 kfree(adapter->vf_data);
3733 adapter->vf_data = NULL;
3734 adapter->vfs_allocated_count = 0;
3735 spin_unlock_irqrestore(&adapter->vfs_lock, flags);
3736 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
3739 dev_info(&pdev->dev, "IOV Disabled\n");
3741 /* Re-enable DMA Coalescing flag since IOV is turned off */
3742 adapter->flags |= IGB_FLAG_DMAC;
3745 return reinit ? igb_sriov_reinit(pdev) : 0;
3748 static int igb_enable_sriov(struct pci_dev *pdev, int num_vfs, bool reinit)
3750 struct net_device *netdev = pci_get_drvdata(pdev);
3751 struct igb_adapter *adapter = netdev_priv(netdev);
3752 int old_vfs = pci_num_vf(pdev);
3753 struct vf_mac_filter *mac_list;
3755 int num_vf_mac_filters, i;
3757 if (!(adapter->flags & IGB_FLAG_HAS_MSIX) || num_vfs > 7) {
3765 dev_info(&pdev->dev, "%d pre-allocated VFs found - override max_vfs setting of %d\n",
3767 adapter->vfs_allocated_count = old_vfs;
3769 adapter->vfs_allocated_count = num_vfs;
3771 adapter->vf_data = kcalloc(adapter->vfs_allocated_count,
3772 sizeof(struct vf_data_storage), GFP_KERNEL);
3774 /* if allocation failed then we do not support SR-IOV */
3775 if (!adapter->vf_data) {
3776 adapter->vfs_allocated_count = 0;
3781 /* Due to the limited number of RAR entries calculate potential
3782 * number of MAC filters available for the VFs. Reserve entries
3783 * for PF default MAC, PF MAC filters and at least one RAR entry
3784 * for each VF for VF MAC.
3786 num_vf_mac_filters = adapter->hw.mac.rar_entry_count -
3787 (1 + IGB_PF_MAC_FILTERS_RESERVED +
3788 adapter->vfs_allocated_count);
3790 adapter->vf_mac_list = kcalloc(num_vf_mac_filters,
3791 sizeof(struct vf_mac_filter),
3794 mac_list = adapter->vf_mac_list;
3795 INIT_LIST_HEAD(&adapter->vf_macs.l);
3797 if (adapter->vf_mac_list) {
3798 /* Initialize list of VF MAC filters */
3799 for (i = 0; i < num_vf_mac_filters; i++) {
3801 mac_list->free = true;
3802 list_add(&mac_list->l, &adapter->vf_macs.l);
3806 /* If we could not allocate memory for the VF MAC filters
3807 * we can continue without this feature but warn user.
3810 "Unable to allocate memory for VF MAC filter list\n");
3813 dev_info(&pdev->dev, "%d VFs allocated\n",
3814 adapter->vfs_allocated_count);
3815 for (i = 0; i < adapter->vfs_allocated_count; i++)
3816 igb_vf_configure(adapter, i);
3818 /* DMA Coalescing is not supported in IOV mode. */
3819 adapter->flags &= ~IGB_FLAG_DMAC;
3822 err = igb_sriov_reinit(pdev);
3827 /* only call pci_enable_sriov() if no VFs are allocated already */
3829 err = pci_enable_sriov(pdev, adapter->vfs_allocated_count);
3834 kfree(adapter->vf_mac_list);
3835 adapter->vf_mac_list = NULL;
3836 kfree(adapter->vf_data);
3837 adapter->vf_data = NULL;
3838 adapter->vfs_allocated_count = 0;
3845 * igb_remove_i2c - Cleanup I2C interface
3846 * @adapter: pointer to adapter structure
3848 static void igb_remove_i2c(struct igb_adapter *adapter)
3850 /* free the adapter bus structure */
3851 i2c_del_adapter(&adapter->i2c_adap);
3855 * igb_remove - Device Removal Routine
3856 * @pdev: PCI device information struct
3858 * igb_remove is called by the PCI subsystem to alert the driver
3859 * that it should release a PCI device. The could be caused by a
3860 * Hot-Plug event, or because the driver is going to be removed from
3863 static void igb_remove(struct pci_dev *pdev)
3865 struct net_device *netdev = pci_get_drvdata(pdev);
3866 struct igb_adapter *adapter = netdev_priv(netdev);
3867 struct e1000_hw *hw = &adapter->hw;
3869 pm_runtime_get_noresume(&pdev->dev);
3870 #ifdef CONFIG_IGB_HWMON
3871 igb_sysfs_exit(adapter);
3873 igb_remove_i2c(adapter);
3874 igb_ptp_stop(adapter);
3875 /* The watchdog timer may be rescheduled, so explicitly
3876 * disable watchdog from being rescheduled.
3878 set_bit(__IGB_DOWN, &adapter->state);
3879 del_timer_sync(&adapter->watchdog_timer);
3880 del_timer_sync(&adapter->phy_info_timer);
3882 cancel_work_sync(&adapter->reset_task);
3883 cancel_work_sync(&adapter->watchdog_task);
3885 #ifdef CONFIG_IGB_DCA
3886 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
3887 dev_info(&pdev->dev, "DCA disabled\n");
3888 dca_remove_requester(&pdev->dev);
3889 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
3890 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
3894 /* Release control of h/w to f/w. If f/w is AMT enabled, this
3895 * would have already happened in close and is redundant.
3897 igb_release_hw_control(adapter);
3899 #ifdef CONFIG_PCI_IOV
3900 igb_disable_sriov(pdev, false);
3903 unregister_netdev(netdev);
3905 igb_clear_interrupt_scheme(adapter);
3907 pci_iounmap(pdev, adapter->io_addr);
3908 if (hw->flash_address)
3909 iounmap(hw->flash_address);
3910 pci_release_mem_regions(pdev);
3912 kfree(adapter->mac_table);
3913 kfree(adapter->shadow_vfta);
3914 free_netdev(netdev);
3916 pci_disable_device(pdev);
3920 * igb_probe_vfs - Initialize vf data storage and add VFs to pci config space
3921 * @adapter: board private structure to initialize
3923 * This function initializes the vf specific data storage and then attempts to
3924 * allocate the VFs. The reason for ordering it this way is because it is much
3925 * mor expensive time wise to disable SR-IOV than it is to allocate and free
3926 * the memory for the VFs.
3928 static void igb_probe_vfs(struct igb_adapter *adapter)
3930 #ifdef CONFIG_PCI_IOV
3931 struct pci_dev *pdev = adapter->pdev;
3932 struct e1000_hw *hw = &adapter->hw;
3934 /* Virtualization features not supported on i210 family. */
3935 if ((hw->mac.type == e1000_i210) || (hw->mac.type == e1000_i211))
3938 /* Of the below we really only want the effect of getting
3939 * IGB_FLAG_HAS_MSIX set (if available), without which
3940 * igb_enable_sriov() has no effect.
3942 igb_set_interrupt_capability(adapter, true);
3943 igb_reset_interrupt_capability(adapter);
3945 pci_sriov_set_totalvfs(pdev, 7);
3946 igb_enable_sriov(pdev, max_vfs, false);
3948 #endif /* CONFIG_PCI_IOV */
3951 unsigned int igb_get_max_rss_queues(struct igb_adapter *adapter)
3953 struct e1000_hw *hw = &adapter->hw;
3954 unsigned int max_rss_queues;
3956 /* Determine the maximum number of RSS queues supported. */
3957 switch (hw->mac.type) {
3959 max_rss_queues = IGB_MAX_RX_QUEUES_I211;
3963 max_rss_queues = IGB_MAX_RX_QUEUES_82575;
3966 /* I350 cannot do RSS and SR-IOV at the same time */
3967 if (!!adapter->vfs_allocated_count) {
3973 if (!!adapter->vfs_allocated_count) {
3981 max_rss_queues = IGB_MAX_RX_QUEUES;
3985 return max_rss_queues;
3988 static void igb_init_queue_configuration(struct igb_adapter *adapter)
3992 max_rss_queues = igb_get_max_rss_queues(adapter);
3993 adapter->rss_queues = min_t(u32, max_rss_queues, num_online_cpus());
3995 igb_set_flag_queue_pairs(adapter, max_rss_queues);
3998 void igb_set_flag_queue_pairs(struct igb_adapter *adapter,
3999 const u32 max_rss_queues)
4001 struct e1000_hw *hw = &adapter->hw;
4003 /* Determine if we need to pair queues. */
4004 switch (hw->mac.type) {
4007 /* Device supports enough interrupts without queue pairing. */
4015 /* If rss_queues > half of max_rss_queues, pair the queues in
4016 * order to conserve interrupts due to limited supply.
4018 if (adapter->rss_queues > (max_rss_queues / 2))
4019 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
4021 adapter->flags &= ~IGB_FLAG_QUEUE_PAIRS;
4027 * igb_sw_init - Initialize general software structures (struct igb_adapter)
4028 * @adapter: board private structure to initialize
4030 * igb_sw_init initializes the Adapter private data structure.
4031 * Fields are initialized based on PCI device information and
4032 * OS network device settings (MTU size).
4034 static int igb_sw_init(struct igb_adapter *adapter)
4036 struct e1000_hw *hw = &adapter->hw;
4037 struct net_device *netdev = adapter->netdev;
4038 struct pci_dev *pdev = adapter->pdev;
4040 pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
4042 /* set default ring sizes */
4043 adapter->tx_ring_count = IGB_DEFAULT_TXD;
4044 adapter->rx_ring_count = IGB_DEFAULT_RXD;
4046 /* set default ITR values */
4047 adapter->rx_itr_setting = IGB_DEFAULT_ITR;
4048 adapter->tx_itr_setting = IGB_DEFAULT_ITR;
4050 /* set default work limits */
4051 adapter->tx_work_limit = IGB_DEFAULT_TX_WORK;
4053 adapter->max_frame_size = netdev->mtu + IGB_ETH_PKT_HDR_PAD;
4054 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
4056 spin_lock_init(&adapter->nfc_lock);
4057 spin_lock_init(&adapter->stats64_lock);
4059 /* init spinlock to avoid concurrency of VF resources */
4060 spin_lock_init(&adapter->vfs_lock);
4061 #ifdef CONFIG_PCI_IOV
4062 switch (hw->mac.type) {
4066 dev_warn(&pdev->dev,
4067 "Maximum of 7 VFs per PF, using max\n");
4068 max_vfs = adapter->vfs_allocated_count = 7;
4070 adapter->vfs_allocated_count = max_vfs;
4071 if (adapter->vfs_allocated_count)
4072 dev_warn(&pdev->dev,
4073 "Enabling SR-IOV VFs using the module parameter is deprecated - please use the pci sysfs interface.\n");
4078 #endif /* CONFIG_PCI_IOV */
4080 /* Assume MSI-X interrupts, will be checked during IRQ allocation */
4081 adapter->flags |= IGB_FLAG_HAS_MSIX;
4083 adapter->mac_table = kcalloc(hw->mac.rar_entry_count,
4084 sizeof(struct igb_mac_addr),
4086 if (!adapter->mac_table)
4089 igb_probe_vfs(adapter);
4091 igb_init_queue_configuration(adapter);
4093 /* Setup and initialize a copy of the hw vlan table array */
4094 adapter->shadow_vfta = kcalloc(E1000_VLAN_FILTER_TBL_SIZE, sizeof(u32),
4096 if (!adapter->shadow_vfta)
4099 /* This call may decrease the number of queues */
4100 if (igb_init_interrupt_scheme(adapter, true)) {
4101 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
4105 /* Explicitly disable IRQ since the NIC can be in any state. */
4106 igb_irq_disable(adapter);
4108 if (hw->mac.type >= e1000_i350)
4109 adapter->flags &= ~IGB_FLAG_DMAC;
4111 set_bit(__IGB_DOWN, &adapter->state);
4116 * __igb_open - Called when a network interface is made active
4117 * @netdev: network interface device structure
4118 * @resuming: indicates whether we are in a resume call
4120 * Returns 0 on success, negative value on failure
4122 * The open entry point is called when a network interface is made
4123 * active by the system (IFF_UP). At this point all resources needed
4124 * for transmit and receive operations are allocated, the interrupt
4125 * handler is registered with the OS, the watchdog timer is started,
4126 * and the stack is notified that the interface is ready.
4128 static int __igb_open(struct net_device *netdev, bool resuming)
4130 struct igb_adapter *adapter = netdev_priv(netdev);
4131 struct e1000_hw *hw = &adapter->hw;
4132 struct pci_dev *pdev = adapter->pdev;
4136 /* disallow open during test */
4137 if (test_bit(__IGB_TESTING, &adapter->state)) {
4143 pm_runtime_get_sync(&pdev->dev);
4145 netif_carrier_off(netdev);
4147 /* allocate transmit descriptors */
4148 err = igb_setup_all_tx_resources(adapter);
4152 /* allocate receive descriptors */
4153 err = igb_setup_all_rx_resources(adapter);
4157 igb_power_up_link(adapter);
4159 /* before we allocate an interrupt, we must be ready to handle it.
4160 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
4161 * as soon as we call pci_request_irq, so we have to setup our
4162 * clean_rx handler before we do so.
4164 igb_configure(adapter);
4166 err = igb_request_irq(adapter);
4170 /* Notify the stack of the actual queue counts. */
4171 err = netif_set_real_num_tx_queues(adapter->netdev,
4172 adapter->num_tx_queues);
4174 goto err_set_queues;
4176 err = netif_set_real_num_rx_queues(adapter->netdev,
4177 adapter->num_rx_queues);
4179 goto err_set_queues;
4181 /* From here on the code is the same as igb_up() */
4182 clear_bit(__IGB_DOWN, &adapter->state);
4184 for (i = 0; i < adapter->num_q_vectors; i++)
4185 napi_enable(&(adapter->q_vector[i]->napi));
4187 /* Clear any pending interrupts. */
4191 igb_irq_enable(adapter);
4193 /* notify VFs that reset has been completed */
4194 if (adapter->vfs_allocated_count) {
4195 u32 reg_data = rd32(E1000_CTRL_EXT);
4197 reg_data |= E1000_CTRL_EXT_PFRSTD;
4198 wr32(E1000_CTRL_EXT, reg_data);
4201 netif_tx_start_all_queues(netdev);
4204 pm_runtime_put(&pdev->dev);
4206 /* start the watchdog. */
4207 hw->mac.get_link_status = 1;
4208 schedule_work(&adapter->watchdog_task);
4213 igb_free_irq(adapter);
4215 igb_release_hw_control(adapter);
4216 igb_power_down_link(adapter);
4217 igb_free_all_rx_resources(adapter);
4219 igb_free_all_tx_resources(adapter);
4223 pm_runtime_put(&pdev->dev);
4228 int igb_open(struct net_device *netdev)
4230 return __igb_open(netdev, false);
4234 * __igb_close - Disables a network interface
4235 * @netdev: network interface device structure
4236 * @suspending: indicates we are in a suspend call
4238 * Returns 0, this is not allowed to fail
4240 * The close entry point is called when an interface is de-activated
4241 * by the OS. The hardware is still under the driver's control, but
4242 * needs to be disabled. A global MAC reset is issued to stop the
4243 * hardware, and all transmit and receive resources are freed.
4245 static int __igb_close(struct net_device *netdev, bool suspending)
4247 struct igb_adapter *adapter = netdev_priv(netdev);
4248 struct pci_dev *pdev = adapter->pdev;
4250 WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
4253 pm_runtime_get_sync(&pdev->dev);
4256 igb_free_irq(adapter);
4258 igb_free_all_tx_resources(adapter);
4259 igb_free_all_rx_resources(adapter);
4262 pm_runtime_put_sync(&pdev->dev);
4266 int igb_close(struct net_device *netdev)
4268 if (netif_device_present(netdev) || netdev->dismantle)
4269 return __igb_close(netdev, false);
4274 * igb_setup_tx_resources - allocate Tx resources (Descriptors)
4275 * @tx_ring: tx descriptor ring (for a specific queue) to setup
4277 * Return 0 on success, negative on failure
4279 int igb_setup_tx_resources(struct igb_ring *tx_ring)
4281 struct device *dev = tx_ring->dev;
4284 size = sizeof(struct igb_tx_buffer) * tx_ring->count;
4286 tx_ring->tx_buffer_info = vmalloc(size);
4287 if (!tx_ring->tx_buffer_info)
4290 /* round up to nearest 4K */
4291 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
4292 tx_ring->size = ALIGN(tx_ring->size, 4096);
4294 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
4295 &tx_ring->dma, GFP_KERNEL);
4299 tx_ring->next_to_use = 0;
4300 tx_ring->next_to_clean = 0;
4305 vfree(tx_ring->tx_buffer_info);
4306 tx_ring->tx_buffer_info = NULL;
4307 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
4312 * igb_setup_all_tx_resources - wrapper to allocate Tx resources
4313 * (Descriptors) for all queues
4314 * @adapter: board private structure
4316 * Return 0 on success, negative on failure
4318 static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
4320 struct pci_dev *pdev = adapter->pdev;
4323 for (i = 0; i < adapter->num_tx_queues; i++) {
4324 err = igb_setup_tx_resources(adapter->tx_ring[i]);
4327 "Allocation for Tx Queue %u failed\n", i);
4328 for (i--; i >= 0; i--)
4329 igb_free_tx_resources(adapter->tx_ring[i]);
4338 * igb_setup_tctl - configure the transmit control registers
4339 * @adapter: Board private structure
4341 void igb_setup_tctl(struct igb_adapter *adapter)
4343 struct e1000_hw *hw = &adapter->hw;
4346 /* disable queue 0 which is enabled by default on 82575 and 82576 */
4347 wr32(E1000_TXDCTL(0), 0);
4349 /* Program the Transmit Control Register */
4350 tctl = rd32(E1000_TCTL);
4351 tctl &= ~E1000_TCTL_CT;
4352 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
4353 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
4355 igb_config_collision_dist(hw);
4357 /* Enable transmits */
4358 tctl |= E1000_TCTL_EN;
4360 wr32(E1000_TCTL, tctl);
4364 * igb_configure_tx_ring - Configure transmit ring after Reset
4365 * @adapter: board private structure
4366 * @ring: tx ring to configure
4368 * Configure a transmit ring after a reset.
4370 void igb_configure_tx_ring(struct igb_adapter *adapter,
4371 struct igb_ring *ring)
4373 struct e1000_hw *hw = &adapter->hw;
4375 u64 tdba = ring->dma;
4376 int reg_idx = ring->reg_idx;
4378 wr32(E1000_TDLEN(reg_idx),
4379 ring->count * sizeof(union e1000_adv_tx_desc));
4380 wr32(E1000_TDBAL(reg_idx),
4381 tdba & 0x00000000ffffffffULL);
4382 wr32(E1000_TDBAH(reg_idx), tdba >> 32);
4384 ring->tail = adapter->io_addr + E1000_TDT(reg_idx);
4385 wr32(E1000_TDH(reg_idx), 0);
4386 writel(0, ring->tail);
4388 txdctl |= IGB_TX_PTHRESH;
4389 txdctl |= IGB_TX_HTHRESH << 8;
4390 txdctl |= IGB_TX_WTHRESH << 16;
4392 /* reinitialize tx_buffer_info */
4393 memset(ring->tx_buffer_info, 0,
4394 sizeof(struct igb_tx_buffer) * ring->count);
4396 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
4397 wr32(E1000_TXDCTL(reg_idx), txdctl);
4401 * igb_configure_tx - Configure transmit Unit after Reset
4402 * @adapter: board private structure
4404 * Configure the Tx unit of the MAC after a reset.
4406 static void igb_configure_tx(struct igb_adapter *adapter)
4408 struct e1000_hw *hw = &adapter->hw;
4411 /* disable the queues */
4412 for (i = 0; i < adapter->num_tx_queues; i++)
4413 wr32(E1000_TXDCTL(adapter->tx_ring[i]->reg_idx), 0);
4416 usleep_range(10000, 20000);
4418 for (i = 0; i < adapter->num_tx_queues; i++)
4419 igb_configure_tx_ring(adapter, adapter->tx_ring[i]);
4423 * igb_setup_rx_resources - allocate Rx resources (Descriptors)
4424 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
4426 * Returns 0 on success, negative on failure
4428 int igb_setup_rx_resources(struct igb_ring *rx_ring)
4430 struct igb_adapter *adapter = netdev_priv(rx_ring->netdev);
4431 struct device *dev = rx_ring->dev;
4434 /* XDP RX-queue info */
4435 if (xdp_rxq_info_is_reg(&rx_ring->xdp_rxq))
4436 xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
4437 res = xdp_rxq_info_reg(&rx_ring->xdp_rxq, rx_ring->netdev,
4438 rx_ring->queue_index, 0);
4440 dev_err(dev, "Failed to register xdp_rxq index %u\n",
4441 rx_ring->queue_index);
4445 size = sizeof(struct igb_rx_buffer) * rx_ring->count;
4447 rx_ring->rx_buffer_info = vmalloc(size);
4448 if (!rx_ring->rx_buffer_info)
4451 /* Round up to nearest 4K */
4452 rx_ring->size = rx_ring->count * sizeof(union e1000_adv_rx_desc);
4453 rx_ring->size = ALIGN(rx_ring->size, 4096);
4455 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
4456 &rx_ring->dma, GFP_KERNEL);
4460 rx_ring->next_to_alloc = 0;
4461 rx_ring->next_to_clean = 0;
4462 rx_ring->next_to_use = 0;
4464 rx_ring->xdp_prog = adapter->xdp_prog;
4469 xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
4470 vfree(rx_ring->rx_buffer_info);
4471 rx_ring->rx_buffer_info = NULL;
4472 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
4477 * igb_setup_all_rx_resources - wrapper to allocate Rx resources
4478 * (Descriptors) for all queues
4479 * @adapter: board private structure
4481 * Return 0 on success, negative on failure
4483 static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
4485 struct pci_dev *pdev = adapter->pdev;
4488 for (i = 0; i < adapter->num_rx_queues; i++) {
4489 err = igb_setup_rx_resources(adapter->rx_ring[i]);
4492 "Allocation for Rx Queue %u failed\n", i);
4493 for (i--; i >= 0; i--)
4494 igb_free_rx_resources(adapter->rx_ring[i]);
4503 * igb_setup_mrqc - configure the multiple receive queue control registers
4504 * @adapter: Board private structure
4506 static void igb_setup_mrqc(struct igb_adapter *adapter)
4508 struct e1000_hw *hw = &adapter->hw;
4510 u32 j, num_rx_queues;
4513 netdev_rss_key_fill(rss_key, sizeof(rss_key));
4514 for (j = 0; j < 10; j++)
4515 wr32(E1000_RSSRK(j), rss_key[j]);
4517 num_rx_queues = adapter->rss_queues;
4519 switch (hw->mac.type) {
4521 /* 82576 supports 2 RSS queues for SR-IOV */
4522 if (adapter->vfs_allocated_count)
4529 if (adapter->rss_indir_tbl_init != num_rx_queues) {
4530 for (j = 0; j < IGB_RETA_SIZE; j++)
4531 adapter->rss_indir_tbl[j] =
4532 (j * num_rx_queues) / IGB_RETA_SIZE;
4533 adapter->rss_indir_tbl_init = num_rx_queues;
4535 igb_write_rss_indir_tbl(adapter);
4537 /* Disable raw packet checksumming so that RSS hash is placed in
4538 * descriptor on writeback. No need to enable TCP/UDP/IP checksum
4539 * offloads as they are enabled by default
4541 rxcsum = rd32(E1000_RXCSUM);
4542 rxcsum |= E1000_RXCSUM_PCSD;
4544 if (adapter->hw.mac.type >= e1000_82576)
4545 /* Enable Receive Checksum Offload for SCTP */
4546 rxcsum |= E1000_RXCSUM_CRCOFL;
4548 /* Don't need to set TUOFL or IPOFL, they default to 1 */
4549 wr32(E1000_RXCSUM, rxcsum);
4551 /* Generate RSS hash based on packet types, TCP/UDP
4552 * port numbers and/or IPv4/v6 src and dst addresses
4554 mrqc = E1000_MRQC_RSS_FIELD_IPV4 |
4555 E1000_MRQC_RSS_FIELD_IPV4_TCP |
4556 E1000_MRQC_RSS_FIELD_IPV6 |
4557 E1000_MRQC_RSS_FIELD_IPV6_TCP |
4558 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX;
4560 if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV4_UDP)
4561 mrqc |= E1000_MRQC_RSS_FIELD_IPV4_UDP;
4562 if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV6_UDP)
4563 mrqc |= E1000_MRQC_RSS_FIELD_IPV6_UDP;
4565 /* If VMDq is enabled then we set the appropriate mode for that, else
4566 * we default to RSS so that an RSS hash is calculated per packet even
4567 * if we are only using one queue
4569 if (adapter->vfs_allocated_count) {
4570 if (hw->mac.type > e1000_82575) {
4571 /* Set the default pool for the PF's first queue */
4572 u32 vtctl = rd32(E1000_VT_CTL);
4574 vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK |
4575 E1000_VT_CTL_DISABLE_DEF_POOL);
4576 vtctl |= adapter->vfs_allocated_count <<
4577 E1000_VT_CTL_DEFAULT_POOL_SHIFT;
4578 wr32(E1000_VT_CTL, vtctl);
4580 if (adapter->rss_queues > 1)
4581 mrqc |= E1000_MRQC_ENABLE_VMDQ_RSS_MQ;
4583 mrqc |= E1000_MRQC_ENABLE_VMDQ;
4585 mrqc |= E1000_MRQC_ENABLE_RSS_MQ;
4587 igb_vmm_control(adapter);
4589 wr32(E1000_MRQC, mrqc);
4593 * igb_setup_rctl - configure the receive control registers
4594 * @adapter: Board private structure
4596 void igb_setup_rctl(struct igb_adapter *adapter)
4598 struct e1000_hw *hw = &adapter->hw;
4601 rctl = rd32(E1000_RCTL);
4603 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
4604 rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
4606 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
4607 (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
4609 /* enable stripping of CRC. It's unlikely this will break BMC
4610 * redirection as it did with e1000. Newer features require
4611 * that the HW strips the CRC.
4613 rctl |= E1000_RCTL_SECRC;
4615 /* disable store bad packets and clear size bits. */
4616 rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
4618 /* enable LPE to allow for reception of jumbo frames */
4619 rctl |= E1000_RCTL_LPE;
4621 /* disable queue 0 to prevent tail write w/o re-config */
4622 wr32(E1000_RXDCTL(0), 0);
4624 /* Attention!!! For SR-IOV PF driver operations you must enable
4625 * queue drop for all VF and PF queues to prevent head of line blocking
4626 * if an un-trusted VF does not provide descriptors to hardware.
4628 if (adapter->vfs_allocated_count) {
4629 /* set all queue drop enable bits */
4630 wr32(E1000_QDE, ALL_QUEUES);
4633 /* This is useful for sniffing bad packets. */
4634 if (adapter->netdev->features & NETIF_F_RXALL) {
4635 /* UPE and MPE will be handled by normal PROMISC logic
4636 * in e1000e_set_rx_mode
4638 rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
4639 E1000_RCTL_BAM | /* RX All Bcast Pkts */
4640 E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
4642 rctl &= ~(E1000_RCTL_DPF | /* Allow filtered pause */
4643 E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
4644 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
4645 * and that breaks VLANs.
4649 wr32(E1000_RCTL, rctl);
4652 static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
4655 struct e1000_hw *hw = &adapter->hw;
4658 if (size > MAX_JUMBO_FRAME_SIZE)
4659 size = MAX_JUMBO_FRAME_SIZE;
4661 vmolr = rd32(E1000_VMOLR(vfn));
4662 vmolr &= ~E1000_VMOLR_RLPML_MASK;
4663 vmolr |= size | E1000_VMOLR_LPE;
4664 wr32(E1000_VMOLR(vfn), vmolr);
4669 static inline void igb_set_vf_vlan_strip(struct igb_adapter *adapter,
4670 int vfn, bool enable)
4672 struct e1000_hw *hw = &adapter->hw;
4675 if (hw->mac.type < e1000_82576)
4678 if (hw->mac.type == e1000_i350)
4679 reg = E1000_DVMOLR(vfn);
4681 reg = E1000_VMOLR(vfn);
4685 val |= E1000_VMOLR_STRVLAN;
4687 val &= ~(E1000_VMOLR_STRVLAN);
4691 static inline void igb_set_vmolr(struct igb_adapter *adapter,
4694 struct e1000_hw *hw = &adapter->hw;
4697 /* This register exists only on 82576 and newer so if we are older then
4698 * we should exit and do nothing
4700 if (hw->mac.type < e1000_82576)
4703 vmolr = rd32(E1000_VMOLR(vfn));
4705 vmolr |= E1000_VMOLR_AUPE; /* Accept untagged packets */
4707 vmolr &= ~(E1000_VMOLR_AUPE); /* Tagged packets ONLY */
4709 /* clear all bits that might not be set */
4710 vmolr &= ~(E1000_VMOLR_BAM | E1000_VMOLR_RSSE);
4712 if (adapter->rss_queues > 1 && vfn == adapter->vfs_allocated_count)
4713 vmolr |= E1000_VMOLR_RSSE; /* enable RSS */
4714 /* for VMDq only allow the VFs and pool 0 to accept broadcast and
4717 if (vfn <= adapter->vfs_allocated_count)
4718 vmolr |= E1000_VMOLR_BAM; /* Accept broadcast */
4720 wr32(E1000_VMOLR(vfn), vmolr);
4724 * igb_setup_srrctl - configure the split and replication receive control
4726 * @adapter: Board private structure
4727 * @ring: receive ring to be configured
4729 void igb_setup_srrctl(struct igb_adapter *adapter, struct igb_ring *ring)
4731 struct e1000_hw *hw = &adapter->hw;
4732 int reg_idx = ring->reg_idx;
4735 srrctl = IGB_RX_HDR_LEN << E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
4736 if (ring_uses_large_buffer(ring))
4737 srrctl |= IGB_RXBUFFER_3072 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
4739 srrctl |= IGB_RXBUFFER_2048 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
4740 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
4741 if (hw->mac.type >= e1000_82580)
4742 srrctl |= E1000_SRRCTL_TIMESTAMP;
4743 /* Only set Drop Enable if VFs allocated, or we are supporting multiple
4744 * queues and rx flow control is disabled
4746 if (adapter->vfs_allocated_count ||
4747 (!(hw->fc.current_mode & e1000_fc_rx_pause) &&
4748 adapter->num_rx_queues > 1))
4749 srrctl |= E1000_SRRCTL_DROP_EN;
4751 wr32(E1000_SRRCTL(reg_idx), srrctl);
4755 * igb_configure_rx_ring - Configure a receive ring after Reset
4756 * @adapter: board private structure
4757 * @ring: receive ring to be configured
4759 * Configure the Rx unit of the MAC after a reset.
4761 void igb_configure_rx_ring(struct igb_adapter *adapter,
4762 struct igb_ring *ring)
4764 struct e1000_hw *hw = &adapter->hw;
4765 union e1000_adv_rx_desc *rx_desc;
4766 u64 rdba = ring->dma;
4767 int reg_idx = ring->reg_idx;
4770 xdp_rxq_info_unreg_mem_model(&ring->xdp_rxq);
4771 WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
4772 MEM_TYPE_PAGE_SHARED, NULL));
4774 /* disable the queue */
4775 wr32(E1000_RXDCTL(reg_idx), 0);
4777 /* Set DMA base address registers */
4778 wr32(E1000_RDBAL(reg_idx),
4779 rdba & 0x00000000ffffffffULL);
4780 wr32(E1000_RDBAH(reg_idx), rdba >> 32);
4781 wr32(E1000_RDLEN(reg_idx),
4782 ring->count * sizeof(union e1000_adv_rx_desc));
4784 /* initialize head and tail */
4785 ring->tail = adapter->io_addr + E1000_RDT(reg_idx);
4786 wr32(E1000_RDH(reg_idx), 0);
4787 writel(0, ring->tail);
4789 /* set descriptor configuration */
4790 igb_setup_srrctl(adapter, ring);
4792 /* set filtering for VMDQ pools */
4793 igb_set_vmolr(adapter, reg_idx & 0x7, true);
4795 rxdctl |= IGB_RX_PTHRESH;
4796 rxdctl |= IGB_RX_HTHRESH << 8;
4797 rxdctl |= IGB_RX_WTHRESH << 16;
4799 /* initialize rx_buffer_info */
4800 memset(ring->rx_buffer_info, 0,
4801 sizeof(struct igb_rx_buffer) * ring->count);
4803 /* initialize Rx descriptor 0 */
4804 rx_desc = IGB_RX_DESC(ring, 0);
4805 rx_desc->wb.upper.length = 0;
4807 /* enable receive descriptor fetching */
4808 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
4809 wr32(E1000_RXDCTL(reg_idx), rxdctl);
4812 static void igb_set_rx_buffer_len(struct igb_adapter *adapter,
4813 struct igb_ring *rx_ring)
4815 /* set build_skb and buffer size flags */
4816 clear_ring_build_skb_enabled(rx_ring);
4817 clear_ring_uses_large_buffer(rx_ring);
4819 if (adapter->flags & IGB_FLAG_RX_LEGACY)
4822 set_ring_build_skb_enabled(rx_ring);
4824 #if (PAGE_SIZE < 8192)
4825 if (adapter->max_frame_size <= IGB_MAX_FRAME_BUILD_SKB)
4828 set_ring_uses_large_buffer(rx_ring);
4833 * igb_configure_rx - Configure receive Unit after Reset
4834 * @adapter: board private structure
4836 * Configure the Rx unit of the MAC after a reset.
4838 static void igb_configure_rx(struct igb_adapter *adapter)
4842 /* set the correct pool for the PF default MAC address in entry 0 */
4843 igb_set_default_mac_filter(adapter);
4845 /* Setup the HW Rx Head and Tail Descriptor Pointers and
4846 * the Base and Length of the Rx Descriptor Ring
4848 for (i = 0; i < adapter->num_rx_queues; i++) {
4849 struct igb_ring *rx_ring = adapter->rx_ring[i];
4851 igb_set_rx_buffer_len(adapter, rx_ring);
4852 igb_configure_rx_ring(adapter, rx_ring);
4857 * igb_free_tx_resources - Free Tx Resources per Queue
4858 * @tx_ring: Tx descriptor ring for a specific queue
4860 * Free all transmit software resources
4862 void igb_free_tx_resources(struct igb_ring *tx_ring)
4864 igb_clean_tx_ring(tx_ring);
4866 vfree(tx_ring->tx_buffer_info);
4867 tx_ring->tx_buffer_info = NULL;
4869 /* if not set, then don't free */
4873 dma_free_coherent(tx_ring->dev, tx_ring->size,
4874 tx_ring->desc, tx_ring->dma);
4876 tx_ring->desc = NULL;
4880 * igb_free_all_tx_resources - Free Tx Resources for All Queues
4881 * @adapter: board private structure
4883 * Free all transmit software resources
4885 static void igb_free_all_tx_resources(struct igb_adapter *adapter)
4889 for (i = 0; i < adapter->num_tx_queues; i++)
4890 if (adapter->tx_ring[i])
4891 igb_free_tx_resources(adapter->tx_ring[i]);
4895 * igb_clean_tx_ring - Free Tx Buffers
4896 * @tx_ring: ring to be cleaned
4898 static void igb_clean_tx_ring(struct igb_ring *tx_ring)
4900 u16 i = tx_ring->next_to_clean;
4901 struct igb_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
4903 while (i != tx_ring->next_to_use) {
4904 union e1000_adv_tx_desc *eop_desc, *tx_desc;
4906 /* Free all the Tx ring sk_buffs or xdp frames */
4907 if (tx_buffer->type == IGB_TYPE_SKB)
4908 dev_kfree_skb_any(tx_buffer->skb);
4910 xdp_return_frame(tx_buffer->xdpf);
4912 /* unmap skb header data */
4913 dma_unmap_single(tx_ring->dev,
4914 dma_unmap_addr(tx_buffer, dma),
4915 dma_unmap_len(tx_buffer, len),
4918 /* check for eop_desc to determine the end of the packet */
4919 eop_desc = tx_buffer->next_to_watch;
4920 tx_desc = IGB_TX_DESC(tx_ring, i);
4922 /* unmap remaining buffers */
4923 while (tx_desc != eop_desc) {
4927 if (unlikely(i == tx_ring->count)) {
4929 tx_buffer = tx_ring->tx_buffer_info;
4930 tx_desc = IGB_TX_DESC(tx_ring, 0);
4933 /* unmap any remaining paged data */
4934 if (dma_unmap_len(tx_buffer, len))
4935 dma_unmap_page(tx_ring->dev,
4936 dma_unmap_addr(tx_buffer, dma),
4937 dma_unmap_len(tx_buffer, len),
4941 tx_buffer->next_to_watch = NULL;
4943 /* move us one more past the eop_desc for start of next pkt */
4946 if (unlikely(i == tx_ring->count)) {
4948 tx_buffer = tx_ring->tx_buffer_info;
4952 /* reset BQL for queue */
4953 netdev_tx_reset_queue(txring_txq(tx_ring));
4955 /* reset next_to_use and next_to_clean */
4956 tx_ring->next_to_use = 0;
4957 tx_ring->next_to_clean = 0;
4961 * igb_clean_all_tx_rings - Free Tx Buffers for all queues
4962 * @adapter: board private structure
4964 static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
4968 for (i = 0; i < adapter->num_tx_queues; i++)
4969 if (adapter->tx_ring[i])
4970 igb_clean_tx_ring(adapter->tx_ring[i]);
4974 * igb_free_rx_resources - Free Rx Resources
4975 * @rx_ring: ring to clean the resources from
4977 * Free all receive software resources
4979 void igb_free_rx_resources(struct igb_ring *rx_ring)
4981 igb_clean_rx_ring(rx_ring);
4983 rx_ring->xdp_prog = NULL;
4984 xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
4985 vfree(rx_ring->rx_buffer_info);
4986 rx_ring->rx_buffer_info = NULL;
4988 /* if not set, then don't free */
4992 dma_free_coherent(rx_ring->dev, rx_ring->size,
4993 rx_ring->desc, rx_ring->dma);
4995 rx_ring->desc = NULL;
4999 * igb_free_all_rx_resources - Free Rx Resources for All Queues
5000 * @adapter: board private structure
5002 * Free all receive software resources
5004 static void igb_free_all_rx_resources(struct igb_adapter *adapter)
5008 for (i = 0; i < adapter->num_rx_queues; i++)
5009 if (adapter->rx_ring[i])
5010 igb_free_rx_resources(adapter->rx_ring[i]);
5014 * igb_clean_rx_ring - Free Rx Buffers per Queue
5015 * @rx_ring: ring to free buffers from
5017 static void igb_clean_rx_ring(struct igb_ring *rx_ring)
5019 u16 i = rx_ring->next_to_clean;
5021 dev_kfree_skb(rx_ring->skb);
5022 rx_ring->skb = NULL;
5024 /* Free all the Rx ring sk_buffs */
5025 while (i != rx_ring->next_to_alloc) {
5026 struct igb_rx_buffer *buffer_info = &rx_ring->rx_buffer_info[i];
5028 /* Invalidate cache lines that may have been written to by
5029 * device so that we avoid corrupting memory.
5031 dma_sync_single_range_for_cpu(rx_ring->dev,
5033 buffer_info->page_offset,
5034 igb_rx_bufsz(rx_ring),
5037 /* free resources associated with mapping */
5038 dma_unmap_page_attrs(rx_ring->dev,
5040 igb_rx_pg_size(rx_ring),
5043 __page_frag_cache_drain(buffer_info->page,
5044 buffer_info->pagecnt_bias);
5047 if (i == rx_ring->count)
5051 rx_ring->next_to_alloc = 0;
5052 rx_ring->next_to_clean = 0;
5053 rx_ring->next_to_use = 0;
5057 * igb_clean_all_rx_rings - Free Rx Buffers for all queues
5058 * @adapter: board private structure
5060 static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
5064 for (i = 0; i < adapter->num_rx_queues; i++)
5065 if (adapter->rx_ring[i])
5066 igb_clean_rx_ring(adapter->rx_ring[i]);
5070 * igb_set_mac - Change the Ethernet Address of the NIC
5071 * @netdev: network interface device structure
5072 * @p: pointer to an address structure
5074 * Returns 0 on success, negative on failure
5076 static int igb_set_mac(struct net_device *netdev, void *p)
5078 struct igb_adapter *adapter = netdev_priv(netdev);
5079 struct e1000_hw *hw = &adapter->hw;
5080 struct sockaddr *addr = p;
5082 if (!is_valid_ether_addr(addr->sa_data))
5083 return -EADDRNOTAVAIL;
5085 eth_hw_addr_set(netdev, addr->sa_data);
5086 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
5088 /* set the correct pool for the new PF MAC address in entry 0 */
5089 igb_set_default_mac_filter(adapter);
5095 * igb_write_mc_addr_list - write multicast addresses to MTA
5096 * @netdev: network interface device structure
5098 * Writes multicast address list to the MTA hash table.
5099 * Returns: -ENOMEM on failure
5100 * 0 on no addresses written
5101 * X on writing X addresses to MTA
5103 static int igb_write_mc_addr_list(struct net_device *netdev)
5105 struct igb_adapter *adapter = netdev_priv(netdev);
5106 struct e1000_hw *hw = &adapter->hw;
5107 struct netdev_hw_addr *ha;
5111 if (netdev_mc_empty(netdev)) {
5112 /* nothing to program, so clear mc list */
5113 igb_update_mc_addr_list(hw, NULL, 0);
5114 igb_restore_vf_multicasts(adapter);
5118 mta_list = kcalloc(netdev_mc_count(netdev), 6, GFP_ATOMIC);
5122 /* The shared function expects a packed array of only addresses. */
5124 netdev_for_each_mc_addr(ha, netdev)
5125 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
5127 igb_update_mc_addr_list(hw, mta_list, i);
5130 return netdev_mc_count(netdev);
5133 static int igb_vlan_promisc_enable(struct igb_adapter *adapter)
5135 struct e1000_hw *hw = &adapter->hw;
5138 switch (hw->mac.type) {
5142 /* VLAN filtering needed for VLAN prio filter */
5143 if (adapter->netdev->features & NETIF_F_NTUPLE)
5149 /* VLAN filtering needed for pool filtering */
5150 if (adapter->vfs_allocated_count)
5157 /* We are already in VLAN promisc, nothing to do */
5158 if (adapter->flags & IGB_FLAG_VLAN_PROMISC)
5161 if (!adapter->vfs_allocated_count)
5164 /* Add PF to all active pools */
5165 pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT;
5167 for (i = E1000_VLVF_ARRAY_SIZE; --i;) {
5168 u32 vlvf = rd32(E1000_VLVF(i));
5171 wr32(E1000_VLVF(i), vlvf);
5175 /* Set all bits in the VLAN filter table array */
5176 for (i = E1000_VLAN_FILTER_TBL_SIZE; i--;)
5177 hw->mac.ops.write_vfta(hw, i, ~0U);
5179 /* Set flag so we don't redo unnecessary work */
5180 adapter->flags |= IGB_FLAG_VLAN_PROMISC;
5185 #define VFTA_BLOCK_SIZE 8
5186 static void igb_scrub_vfta(struct igb_adapter *adapter, u32 vfta_offset)
5188 struct e1000_hw *hw = &adapter->hw;
5189 u32 vfta[VFTA_BLOCK_SIZE] = { 0 };
5190 u32 vid_start = vfta_offset * 32;
5191 u32 vid_end = vid_start + (VFTA_BLOCK_SIZE * 32);
5192 u32 i, vid, word, bits, pf_id;
5194 /* guarantee that we don't scrub out management VLAN */
5195 vid = adapter->mng_vlan_id;
5196 if (vid >= vid_start && vid < vid_end)
5197 vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
5199 if (!adapter->vfs_allocated_count)
5202 pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT;
5204 for (i = E1000_VLVF_ARRAY_SIZE; --i;) {
5205 u32 vlvf = rd32(E1000_VLVF(i));
5207 /* pull VLAN ID from VLVF */
5208 vid = vlvf & VLAN_VID_MASK;
5210 /* only concern ourselves with a certain range */
5211 if (vid < vid_start || vid >= vid_end)
5214 if (vlvf & E1000_VLVF_VLANID_ENABLE) {
5215 /* record VLAN ID in VFTA */
5216 vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
5218 /* if PF is part of this then continue */
5219 if (test_bit(vid, adapter->active_vlans))
5223 /* remove PF from the pool */
5225 bits &= rd32(E1000_VLVF(i));
5226 wr32(E1000_VLVF(i), bits);
5230 /* extract values from active_vlans and write back to VFTA */
5231 for (i = VFTA_BLOCK_SIZE; i--;) {
5232 vid = (vfta_offset + i) * 32;
5233 word = vid / BITS_PER_LONG;
5234 bits = vid % BITS_PER_LONG;
5236 vfta[i] |= adapter->active_vlans[word] >> bits;
5238 hw->mac.ops.write_vfta(hw, vfta_offset + i, vfta[i]);
5242 static void igb_vlan_promisc_disable(struct igb_adapter *adapter)
5246 /* We are not in VLAN promisc, nothing to do */
5247 if (!(adapter->flags & IGB_FLAG_VLAN_PROMISC))
5250 /* Set flag so we don't redo unnecessary work */
5251 adapter->flags &= ~IGB_FLAG_VLAN_PROMISC;
5253 for (i = 0; i < E1000_VLAN_FILTER_TBL_SIZE; i += VFTA_BLOCK_SIZE)
5254 igb_scrub_vfta(adapter, i);
5258 * igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
5259 * @netdev: network interface device structure
5261 * The set_rx_mode entry point is called whenever the unicast or multicast
5262 * address lists or the network interface flags are updated. This routine is
5263 * responsible for configuring the hardware for proper unicast, multicast,
5264 * promiscuous mode, and all-multi behavior.
5266 static void igb_set_rx_mode(struct net_device *netdev)
5268 struct igb_adapter *adapter = netdev_priv(netdev);
5269 struct e1000_hw *hw = &adapter->hw;
5270 unsigned int vfn = adapter->vfs_allocated_count;
5271 u32 rctl = 0, vmolr = 0, rlpml = MAX_JUMBO_FRAME_SIZE;
5274 /* Check for Promiscuous and All Multicast modes */
5275 if (netdev->flags & IFF_PROMISC) {
5276 rctl |= E1000_RCTL_UPE | E1000_RCTL_MPE;
5277 vmolr |= E1000_VMOLR_MPME;
5279 /* enable use of UTA filter to force packets to default pool */
5280 if (hw->mac.type == e1000_82576)
5281 vmolr |= E1000_VMOLR_ROPE;
5283 if (netdev->flags & IFF_ALLMULTI) {
5284 rctl |= E1000_RCTL_MPE;
5285 vmolr |= E1000_VMOLR_MPME;
5287 /* Write addresses to the MTA, if the attempt fails
5288 * then we should just turn on promiscuous mode so
5289 * that we can at least receive multicast traffic
5291 count = igb_write_mc_addr_list(netdev);
5293 rctl |= E1000_RCTL_MPE;
5294 vmolr |= E1000_VMOLR_MPME;
5296 vmolr |= E1000_VMOLR_ROMPE;
5301 /* Write addresses to available RAR registers, if there is not
5302 * sufficient space to store all the addresses then enable
5303 * unicast promiscuous mode
5305 if (__dev_uc_sync(netdev, igb_uc_sync, igb_uc_unsync)) {
5306 rctl |= E1000_RCTL_UPE;
5307 vmolr |= E1000_VMOLR_ROPE;
5310 /* enable VLAN filtering by default */
5311 rctl |= E1000_RCTL_VFE;
5313 /* disable VLAN filtering for modes that require it */
5314 if ((netdev->flags & IFF_PROMISC) ||
5315 (netdev->features & NETIF_F_RXALL)) {
5316 /* if we fail to set all rules then just clear VFE */
5317 if (igb_vlan_promisc_enable(adapter))
5318 rctl &= ~E1000_RCTL_VFE;
5320 igb_vlan_promisc_disable(adapter);
5323 /* update state of unicast, multicast, and VLAN filtering modes */
5324 rctl |= rd32(E1000_RCTL) & ~(E1000_RCTL_UPE | E1000_RCTL_MPE |
5326 wr32(E1000_RCTL, rctl);
5328 #if (PAGE_SIZE < 8192)
5329 if (!adapter->vfs_allocated_count) {
5330 if (adapter->max_frame_size <= IGB_MAX_FRAME_BUILD_SKB)
5331 rlpml = IGB_MAX_FRAME_BUILD_SKB;
5334 wr32(E1000_RLPML, rlpml);
5336 /* In order to support SR-IOV and eventually VMDq it is necessary to set
5337 * the VMOLR to enable the appropriate modes. Without this workaround
5338 * we will have issues with VLAN tag stripping not being done for frames
5339 * that are only arriving because we are the default pool
5341 if ((hw->mac.type < e1000_82576) || (hw->mac.type > e1000_i350))
5344 /* set UTA to appropriate mode */
5345 igb_set_uta(adapter, !!(vmolr & E1000_VMOLR_ROPE));
5347 vmolr |= rd32(E1000_VMOLR(vfn)) &
5348 ~(E1000_VMOLR_ROPE | E1000_VMOLR_MPME | E1000_VMOLR_ROMPE);
5350 /* enable Rx jumbo frames, restrict as needed to support build_skb */
5351 vmolr &= ~E1000_VMOLR_RLPML_MASK;
5352 #if (PAGE_SIZE < 8192)
5353 if (adapter->max_frame_size <= IGB_MAX_FRAME_BUILD_SKB)
5354 vmolr |= IGB_MAX_FRAME_BUILD_SKB;
5357 vmolr |= MAX_JUMBO_FRAME_SIZE;
5358 vmolr |= E1000_VMOLR_LPE;
5360 wr32(E1000_VMOLR(vfn), vmolr);
5362 igb_restore_vf_multicasts(adapter);
5365 static void igb_check_wvbr(struct igb_adapter *adapter)
5367 struct e1000_hw *hw = &adapter->hw;
5370 switch (hw->mac.type) {
5373 wvbr = rd32(E1000_WVBR);
5381 adapter->wvbr |= wvbr;
5384 #define IGB_STAGGERED_QUEUE_OFFSET 8
5386 static void igb_spoof_check(struct igb_adapter *adapter)
5393 for (j = 0; j < adapter->vfs_allocated_count; j++) {
5394 if (adapter->wvbr & BIT(j) ||
5395 adapter->wvbr & BIT(j + IGB_STAGGERED_QUEUE_OFFSET)) {
5396 dev_warn(&adapter->pdev->dev,
5397 "Spoof event(s) detected on VF %d\n", j);
5400 BIT(j + IGB_STAGGERED_QUEUE_OFFSET));
5405 /* Need to wait a few seconds after link up to get diagnostic information from
5408 static void igb_update_phy_info(struct timer_list *t)
5410 struct igb_adapter *adapter = from_timer(adapter, t, phy_info_timer);
5411 igb_get_phy_info(&adapter->hw);
5415 * igb_has_link - check shared code for link and determine up/down
5416 * @adapter: pointer to driver private info
5418 bool igb_has_link(struct igb_adapter *adapter)
5420 struct e1000_hw *hw = &adapter->hw;
5421 bool link_active = false;
5423 /* get_link_status is set on LSC (link status) interrupt or
5424 * rx sequence error interrupt. get_link_status will stay
5425 * false until the e1000_check_for_link establishes link
5426 * for copper adapters ONLY
5428 switch (hw->phy.media_type) {
5429 case e1000_media_type_copper:
5430 if (!hw->mac.get_link_status)
5433 case e1000_media_type_internal_serdes:
5434 hw->mac.ops.check_for_link(hw);
5435 link_active = !hw->mac.get_link_status;
5438 case e1000_media_type_unknown:
5442 if (((hw->mac.type == e1000_i210) ||
5443 (hw->mac.type == e1000_i211)) &&
5444 (hw->phy.id == I210_I_PHY_ID)) {
5445 if (!netif_carrier_ok(adapter->netdev)) {
5446 adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
5447 } else if (!(adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)) {
5448 adapter->flags |= IGB_FLAG_NEED_LINK_UPDATE;
5449 adapter->link_check_timeout = jiffies;
5456 static bool igb_thermal_sensor_event(struct e1000_hw *hw, u32 event)
5459 u32 ctrl_ext, thstat;
5461 /* check for thermal sensor event on i350 copper only */
5462 if (hw->mac.type == e1000_i350) {
5463 thstat = rd32(E1000_THSTAT);
5464 ctrl_ext = rd32(E1000_CTRL_EXT);
5466 if ((hw->phy.media_type == e1000_media_type_copper) &&
5467 !(ctrl_ext & E1000_CTRL_EXT_LINK_MODE_SGMII))
5468 ret = !!(thstat & event);
5475 * igb_check_lvmmc - check for malformed packets received
5476 * and indicated in LVMMC register
5477 * @adapter: pointer to adapter
5479 static void igb_check_lvmmc(struct igb_adapter *adapter)
5481 struct e1000_hw *hw = &adapter->hw;
5484 lvmmc = rd32(E1000_LVMMC);
5486 if (unlikely(net_ratelimit())) {
5487 netdev_warn(adapter->netdev,
5488 "malformed Tx packet detected and dropped, LVMMC:0x%08x\n",
5495 * igb_watchdog - Timer Call-back
5496 * @t: pointer to timer_list containing our private info pointer
5498 static void igb_watchdog(struct timer_list *t)
5500 struct igb_adapter *adapter = from_timer(adapter, t, watchdog_timer);
5501 /* Do the rest outside of interrupt context */
5502 schedule_work(&adapter->watchdog_task);
5505 static void igb_watchdog_task(struct work_struct *work)
5507 struct igb_adapter *adapter = container_of(work,
5510 struct e1000_hw *hw = &adapter->hw;
5511 struct e1000_phy_info *phy = &hw->phy;
5512 struct net_device *netdev = adapter->netdev;
5516 u16 phy_data, retry_count = 20;
5518 link = igb_has_link(adapter);
5520 if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE) {
5521 if (time_after(jiffies, (adapter->link_check_timeout + HZ)))
5522 adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
5527 /* Force link down if we have fiber to swap to */
5528 if (adapter->flags & IGB_FLAG_MAS_ENABLE) {
5529 if (hw->phy.media_type == e1000_media_type_copper) {
5530 connsw = rd32(E1000_CONNSW);
5531 if (!(connsw & E1000_CONNSW_AUTOSENSE_EN))
5536 /* Perform a reset if the media type changed. */
5537 if (hw->dev_spec._82575.media_changed) {
5538 hw->dev_spec._82575.media_changed = false;
5539 adapter->flags |= IGB_FLAG_MEDIA_RESET;
5542 /* Cancel scheduled suspend requests. */
5543 pm_runtime_resume(netdev->dev.parent);
5545 if (!netif_carrier_ok(netdev)) {
5548 hw->mac.ops.get_speed_and_duplex(hw,
5549 &adapter->link_speed,
5550 &adapter->link_duplex);
5552 ctrl = rd32(E1000_CTRL);
5553 /* Links status message must follow this format */
5555 "igb: %s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
5557 adapter->link_speed,
5558 adapter->link_duplex == FULL_DUPLEX ?
5560 (ctrl & E1000_CTRL_TFCE) &&
5561 (ctrl & E1000_CTRL_RFCE) ? "RX/TX" :
5562 (ctrl & E1000_CTRL_RFCE) ? "RX" :
5563 (ctrl & E1000_CTRL_TFCE) ? "TX" : "None");
5565 /* disable EEE if enabled */
5566 if ((adapter->flags & IGB_FLAG_EEE) &&
5567 (adapter->link_duplex == HALF_DUPLEX)) {
5568 dev_info(&adapter->pdev->dev,
5569 "EEE Disabled: unsupported at half duplex. Re-enable using ethtool when at full duplex.\n");
5570 adapter->hw.dev_spec._82575.eee_disable = true;
5571 adapter->flags &= ~IGB_FLAG_EEE;
5574 /* check if SmartSpeed worked */
5575 igb_check_downshift(hw);
5576 if (phy->speed_downgraded)
5577 netdev_warn(netdev, "Link Speed was downgraded by SmartSpeed\n");
5579 /* check for thermal sensor event */
5580 if (igb_thermal_sensor_event(hw,
5581 E1000_THSTAT_LINK_THROTTLE))
5582 netdev_info(netdev, "The network adapter link speed was downshifted because it overheated\n");
5584 /* adjust timeout factor according to speed/duplex */
5585 adapter->tx_timeout_factor = 1;
5586 switch (adapter->link_speed) {
5588 adapter->tx_timeout_factor = 14;
5591 /* maybe add some timeout factor ? */
5595 if (adapter->link_speed != SPEED_1000 ||
5596 !hw->phy.ops.read_reg)
5599 /* wait for Remote receiver status OK */
5601 if (!igb_read_phy_reg(hw, PHY_1000T_STATUS,
5603 if (!(phy_data & SR_1000T_REMOTE_RX_STATUS) &&
5607 goto retry_read_status;
5608 } else if (!retry_count) {
5609 dev_err(&adapter->pdev->dev, "exceed max 2 second\n");
5612 dev_err(&adapter->pdev->dev, "read 1000Base-T Status Reg\n");
5615 netif_carrier_on(netdev);
5617 igb_ping_all_vfs(adapter);
5618 igb_check_vf_rate_limit(adapter);
5620 /* link state has changed, schedule phy info update */
5621 if (!test_bit(__IGB_DOWN, &adapter->state))
5622 mod_timer(&adapter->phy_info_timer,
5623 round_jiffies(jiffies + 2 * HZ));
5626 if (netif_carrier_ok(netdev)) {
5627 adapter->link_speed = 0;
5628 adapter->link_duplex = 0;
5630 /* check for thermal sensor event */
5631 if (igb_thermal_sensor_event(hw,
5632 E1000_THSTAT_PWR_DOWN)) {
5633 netdev_err(netdev, "The network adapter was stopped because it overheated\n");
5636 /* Links status message must follow this format */
5637 netdev_info(netdev, "igb: %s NIC Link is Down\n",
5639 netif_carrier_off(netdev);
5641 igb_ping_all_vfs(adapter);
5643 /* link state has changed, schedule phy info update */
5644 if (!test_bit(__IGB_DOWN, &adapter->state))
5645 mod_timer(&adapter->phy_info_timer,
5646 round_jiffies(jiffies + 2 * HZ));
5648 /* link is down, time to check for alternate media */
5649 if (adapter->flags & IGB_FLAG_MAS_ENABLE) {
5650 igb_check_swap_media(adapter);
5651 if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
5652 schedule_work(&adapter->reset_task);
5653 /* return immediately */
5657 pm_schedule_suspend(netdev->dev.parent,
5660 /* also check for alternate media here */
5661 } else if (!netif_carrier_ok(netdev) &&
5662 (adapter->flags & IGB_FLAG_MAS_ENABLE)) {
5663 igb_check_swap_media(adapter);
5664 if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
5665 schedule_work(&adapter->reset_task);
5666 /* return immediately */
5672 spin_lock(&adapter->stats64_lock);
5673 igb_update_stats(adapter);
5674 spin_unlock(&adapter->stats64_lock);
5676 for (i = 0; i < adapter->num_tx_queues; i++) {
5677 struct igb_ring *tx_ring = adapter->tx_ring[i];
5678 if (!netif_carrier_ok(netdev)) {
5679 /* We've lost link, so the controller stops DMA,
5680 * but we've got queued Tx work that's never going
5681 * to get done, so reset controller to flush Tx.
5682 * (Do the reset outside of interrupt context).
5684 if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) {
5685 adapter->tx_timeout_count++;
5686 schedule_work(&adapter->reset_task);
5687 /* return immediately since reset is imminent */
5692 /* Force detection of hung controller every watchdog period */
5693 set_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
5696 /* Cause software interrupt to ensure Rx ring is cleaned */
5697 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
5700 for (i = 0; i < adapter->num_q_vectors; i++)
5701 eics |= adapter->q_vector[i]->eims_value;
5702 wr32(E1000_EICS, eics);
5704 wr32(E1000_ICS, E1000_ICS_RXDMT0);
5707 igb_spoof_check(adapter);
5708 igb_ptp_rx_hang(adapter);
5709 igb_ptp_tx_hang(adapter);
5711 /* Check LVMMC register on i350/i354 only */
5712 if ((adapter->hw.mac.type == e1000_i350) ||
5713 (adapter->hw.mac.type == e1000_i354))
5714 igb_check_lvmmc(adapter);
5716 /* Reset the timer */
5717 if (!test_bit(__IGB_DOWN, &adapter->state)) {
5718 if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)
5719 mod_timer(&adapter->watchdog_timer,
5720 round_jiffies(jiffies + HZ));
5722 mod_timer(&adapter->watchdog_timer,
5723 round_jiffies(jiffies + 2 * HZ));
5727 enum latency_range {
5731 latency_invalid = 255
5735 * igb_update_ring_itr - update the dynamic ITR value based on packet size
5736 * @q_vector: pointer to q_vector
5738 * Stores a new ITR value based on strictly on packet size. This
5739 * algorithm is less sophisticated than that used in igb_update_itr,
5740 * due to the difficulty of synchronizing statistics across multiple
5741 * receive rings. The divisors and thresholds used by this function
5742 * were determined based on theoretical maximum wire speed and testing
5743 * data, in order to minimize response time while increasing bulk
5745 * This functionality is controlled by ethtool's coalescing settings.
5746 * NOTE: This function is called only when operating in a multiqueue
5747 * receive environment.
5749 static void igb_update_ring_itr(struct igb_q_vector *q_vector)
5751 int new_val = q_vector->itr_val;
5752 int avg_wire_size = 0;
5753 struct igb_adapter *adapter = q_vector->adapter;
5754 unsigned int packets;
5756 /* For non-gigabit speeds, just fix the interrupt rate at 4000
5757 * ints/sec - ITR timer value of 120 ticks.
5759 if (adapter->link_speed != SPEED_1000) {
5760 new_val = IGB_4K_ITR;
5764 packets = q_vector->rx.total_packets;
5766 avg_wire_size = q_vector->rx.total_bytes / packets;
5768 packets = q_vector->tx.total_packets;
5770 avg_wire_size = max_t(u32, avg_wire_size,
5771 q_vector->tx.total_bytes / packets);
5773 /* if avg_wire_size isn't set no work was done */
5777 /* Add 24 bytes to size to account for CRC, preamble, and gap */
5778 avg_wire_size += 24;
5780 /* Don't starve jumbo frames */
5781 avg_wire_size = min(avg_wire_size, 3000);
5783 /* Give a little boost to mid-size frames */
5784 if ((avg_wire_size > 300) && (avg_wire_size < 1200))
5785 new_val = avg_wire_size / 3;
5787 new_val = avg_wire_size / 2;
5789 /* conservative mode (itr 3) eliminates the lowest_latency setting */
5790 if (new_val < IGB_20K_ITR &&
5791 ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
5792 (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
5793 new_val = IGB_20K_ITR;
5796 if (new_val != q_vector->itr_val) {
5797 q_vector->itr_val = new_val;
5798 q_vector->set_itr = 1;
5801 q_vector->rx.total_bytes = 0;
5802 q_vector->rx.total_packets = 0;
5803 q_vector->tx.total_bytes = 0;
5804 q_vector->tx.total_packets = 0;
5808 * igb_update_itr - update the dynamic ITR value based on statistics
5809 * @q_vector: pointer to q_vector
5810 * @ring_container: ring info to update the itr for
5812 * Stores a new ITR value based on packets and byte
5813 * counts during the last interrupt. The advantage of per interrupt
5814 * computation is faster updates and more accurate ITR for the current
5815 * traffic pattern. Constants in this function were computed
5816 * based on theoretical maximum wire speed and thresholds were set based
5817 * on testing data as well as attempting to minimize response time
5818 * while increasing bulk throughput.
5819 * This functionality is controlled by ethtool's coalescing settings.
5820 * NOTE: These calculations are only valid when operating in a single-
5821 * queue environment.
5823 static void igb_update_itr(struct igb_q_vector *q_vector,
5824 struct igb_ring_container *ring_container)
5826 unsigned int packets = ring_container->total_packets;
5827 unsigned int bytes = ring_container->total_bytes;
5828 u8 itrval = ring_container->itr;
5830 /* no packets, exit with status unchanged */
5835 case lowest_latency:
5836 /* handle TSO and jumbo frames */
5837 if (bytes/packets > 8000)
5838 itrval = bulk_latency;
5839 else if ((packets < 5) && (bytes > 512))
5840 itrval = low_latency;
5842 case low_latency: /* 50 usec aka 20000 ints/s */
5843 if (bytes > 10000) {
5844 /* this if handles the TSO accounting */
5845 if (bytes/packets > 8000)
5846 itrval = bulk_latency;
5847 else if ((packets < 10) || ((bytes/packets) > 1200))
5848 itrval = bulk_latency;
5849 else if ((packets > 35))
5850 itrval = lowest_latency;
5851 } else if (bytes/packets > 2000) {
5852 itrval = bulk_latency;
5853 } else if (packets <= 2 && bytes < 512) {
5854 itrval = lowest_latency;
5857 case bulk_latency: /* 250 usec aka 4000 ints/s */
5858 if (bytes > 25000) {
5860 itrval = low_latency;
5861 } else if (bytes < 1500) {
5862 itrval = low_latency;
5867 /* clear work counters since we have the values we need */
5868 ring_container->total_bytes = 0;
5869 ring_container->total_packets = 0;
5871 /* write updated itr to ring container */
5872 ring_container->itr = itrval;
5875 static void igb_set_itr(struct igb_q_vector *q_vector)
5877 struct igb_adapter *adapter = q_vector->adapter;
5878 u32 new_itr = q_vector->itr_val;
5881 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
5882 if (adapter->link_speed != SPEED_1000) {
5884 new_itr = IGB_4K_ITR;
5888 igb_update_itr(q_vector, &q_vector->tx);
5889 igb_update_itr(q_vector, &q_vector->rx);
5891 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
5893 /* conservative mode (itr 3) eliminates the lowest_latency setting */
5894 if (current_itr == lowest_latency &&
5895 ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
5896 (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
5897 current_itr = low_latency;
5899 switch (current_itr) {
5900 /* counts and packets in update_itr are dependent on these numbers */
5901 case lowest_latency:
5902 new_itr = IGB_70K_ITR; /* 70,000 ints/sec */
5905 new_itr = IGB_20K_ITR; /* 20,000 ints/sec */
5908 new_itr = IGB_4K_ITR; /* 4,000 ints/sec */
5915 if (new_itr != q_vector->itr_val) {
5916 /* this attempts to bias the interrupt rate towards Bulk
5917 * by adding intermediate steps when interrupt rate is
5920 new_itr = new_itr > q_vector->itr_val ?
5921 max((new_itr * q_vector->itr_val) /
5922 (new_itr + (q_vector->itr_val >> 2)),
5924 /* Don't write the value here; it resets the adapter's
5925 * internal timer, and causes us to delay far longer than
5926 * we should between interrupts. Instead, we write the ITR
5927 * value at the beginning of the next interrupt so the timing
5928 * ends up being correct.
5930 q_vector->itr_val = new_itr;
5931 q_vector->set_itr = 1;
5935 static void igb_tx_ctxtdesc(struct igb_ring *tx_ring,
5936 struct igb_tx_buffer *first,
5937 u32 vlan_macip_lens, u32 type_tucmd,
5940 struct e1000_adv_tx_context_desc *context_desc;
5941 u16 i = tx_ring->next_to_use;
5942 struct timespec64 ts;
5944 context_desc = IGB_TX_CTXTDESC(tx_ring, i);
5947 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
5949 /* set bits to identify this as an advanced context descriptor */
5950 type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
5952 /* For 82575, context index must be unique per ring. */
5953 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
5954 mss_l4len_idx |= tx_ring->reg_idx << 4;
5956 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
5957 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd);
5958 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
5960 /* We assume there is always a valid tx time available. Invalid times
5961 * should have been handled by the upper layers.
5963 if (tx_ring->launchtime_enable) {
5964 ts = ktime_to_timespec64(first->skb->tstamp);
5965 skb_txtime_consumed(first->skb);
5966 context_desc->seqnum_seed = cpu_to_le32(ts.tv_nsec / 32);
5968 context_desc->seqnum_seed = 0;
5972 static int igb_tso(struct igb_ring *tx_ring,
5973 struct igb_tx_buffer *first,
5976 u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
5977 struct sk_buff *skb = first->skb;
5988 u32 paylen, l4_offset;
5991 if (skb->ip_summed != CHECKSUM_PARTIAL)
5994 if (!skb_is_gso(skb))
5997 err = skb_cow_head(skb, 0);
6001 ip.hdr = skb_network_header(skb);
6002 l4.hdr = skb_checksum_start(skb);
6004 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
6005 type_tucmd = (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ?
6006 E1000_ADVTXD_TUCMD_L4T_UDP : E1000_ADVTXD_TUCMD_L4T_TCP;
6008 /* initialize outer IP header fields */
6009 if (ip.v4->version == 4) {
6010 unsigned char *csum_start = skb_checksum_start(skb);
6011 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
6013 /* IP header will have to cancel out any data that
6014 * is not a part of the outer IP header
6016 ip.v4->check = csum_fold(csum_partial(trans_start,
6017 csum_start - trans_start,
6019 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
6022 first->tx_flags |= IGB_TX_FLAGS_TSO |
6026 ip.v6->payload_len = 0;
6027 first->tx_flags |= IGB_TX_FLAGS_TSO |
6031 /* determine offset of inner transport header */
6032 l4_offset = l4.hdr - skb->data;
6034 /* remove payload length from inner checksum */
6035 paylen = skb->len - l4_offset;
6036 if (type_tucmd & E1000_ADVTXD_TUCMD_L4T_TCP) {
6037 /* compute length of segmentation header */
6038 *hdr_len = (l4.tcp->doff * 4) + l4_offset;
6039 csum_replace_by_diff(&l4.tcp->check,
6040 (__force __wsum)htonl(paylen));
6042 /* compute length of segmentation header */
6043 *hdr_len = sizeof(*l4.udp) + l4_offset;
6044 csum_replace_by_diff(&l4.udp->check,
6045 (__force __wsum)htonl(paylen));
6048 /* update gso size and bytecount with header size */
6049 first->gso_segs = skb_shinfo(skb)->gso_segs;
6050 first->bytecount += (first->gso_segs - 1) * *hdr_len;
6053 mss_l4len_idx = (*hdr_len - l4_offset) << E1000_ADVTXD_L4LEN_SHIFT;
6054 mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT;
6056 /* VLAN MACLEN IPLEN */
6057 vlan_macip_lens = l4.hdr - ip.hdr;
6058 vlan_macip_lens |= (ip.hdr - skb->data) << E1000_ADVTXD_MACLEN_SHIFT;
6059 vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
6061 igb_tx_ctxtdesc(tx_ring, first, vlan_macip_lens,
6062 type_tucmd, mss_l4len_idx);
6067 static void igb_tx_csum(struct igb_ring *tx_ring, struct igb_tx_buffer *first)
6069 struct sk_buff *skb = first->skb;
6070 u32 vlan_macip_lens = 0;
6073 if (skb->ip_summed != CHECKSUM_PARTIAL) {
6075 if (!(first->tx_flags & IGB_TX_FLAGS_VLAN) &&
6076 !tx_ring->launchtime_enable)
6081 switch (skb->csum_offset) {
6082 case offsetof(struct tcphdr, check):
6083 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
6085 case offsetof(struct udphdr, check):
6087 case offsetof(struct sctphdr, checksum):
6088 /* validate that this is actually an SCTP request */
6089 if (skb_csum_is_sctp(skb)) {
6090 type_tucmd = E1000_ADVTXD_TUCMD_L4T_SCTP;
6095 skb_checksum_help(skb);
6099 /* update TX checksum flag */
6100 first->tx_flags |= IGB_TX_FLAGS_CSUM;
6101 vlan_macip_lens = skb_checksum_start_offset(skb) -
6102 skb_network_offset(skb);
6104 vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
6105 vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
6107 igb_tx_ctxtdesc(tx_ring, first, vlan_macip_lens, type_tucmd, 0);
6110 #define IGB_SET_FLAG(_input, _flag, _result) \
6111 ((_flag <= _result) ? \
6112 ((u32)(_input & _flag) * (_result / _flag)) : \
6113 ((u32)(_input & _flag) / (_flag / _result)))
6115 static u32 igb_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
6117 /* set type for advanced descriptor with frame checksum insertion */
6118 u32 cmd_type = E1000_ADVTXD_DTYP_DATA |
6119 E1000_ADVTXD_DCMD_DEXT |
6120 E1000_ADVTXD_DCMD_IFCS;
6122 /* set HW vlan bit if vlan is present */
6123 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_VLAN,
6124 (E1000_ADVTXD_DCMD_VLE));
6126 /* set segmentation bits for TSO */
6127 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSO,
6128 (E1000_ADVTXD_DCMD_TSE));
6130 /* set timestamp bit if present */
6131 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSTAMP,
6132 (E1000_ADVTXD_MAC_TSTAMP));
6134 /* insert frame checksum */
6135 cmd_type ^= IGB_SET_FLAG(skb->no_fcs, 1, E1000_ADVTXD_DCMD_IFCS);
6140 static void igb_tx_olinfo_status(struct igb_ring *tx_ring,
6141 union e1000_adv_tx_desc *tx_desc,
6142 u32 tx_flags, unsigned int paylen)
6144 u32 olinfo_status = paylen << E1000_ADVTXD_PAYLEN_SHIFT;
6146 /* 82575 requires a unique index per ring */
6147 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
6148 olinfo_status |= tx_ring->reg_idx << 4;
6150 /* insert L4 checksum */
6151 olinfo_status |= IGB_SET_FLAG(tx_flags,
6153 (E1000_TXD_POPTS_TXSM << 8));
6155 /* insert IPv4 checksum */
6156 olinfo_status |= IGB_SET_FLAG(tx_flags,
6158 (E1000_TXD_POPTS_IXSM << 8));
6160 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
6163 static int __igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
6165 struct net_device *netdev = tx_ring->netdev;
6167 netif_stop_subqueue(netdev, tx_ring->queue_index);
6169 /* Herbert's original patch had:
6170 * smp_mb__after_netif_stop_queue();
6171 * but since that doesn't exist yet, just open code it.
6175 /* We need to check again in a case another CPU has just
6176 * made room available.
6178 if (igb_desc_unused(tx_ring) < size)
6182 netif_wake_subqueue(netdev, tx_ring->queue_index);
6184 u64_stats_update_begin(&tx_ring->tx_syncp2);
6185 tx_ring->tx_stats.restart_queue2++;
6186 u64_stats_update_end(&tx_ring->tx_syncp2);
6191 static inline int igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
6193 if (igb_desc_unused(tx_ring) >= size)
6195 return __igb_maybe_stop_tx(tx_ring, size);
6198 static int igb_tx_map(struct igb_ring *tx_ring,
6199 struct igb_tx_buffer *first,
6202 struct sk_buff *skb = first->skb;
6203 struct igb_tx_buffer *tx_buffer;
6204 union e1000_adv_tx_desc *tx_desc;
6207 unsigned int data_len, size;
6208 u32 tx_flags = first->tx_flags;
6209 u32 cmd_type = igb_tx_cmd_type(skb, tx_flags);
6210 u16 i = tx_ring->next_to_use;
6212 tx_desc = IGB_TX_DESC(tx_ring, i);
6214 igb_tx_olinfo_status(tx_ring, tx_desc, tx_flags, skb->len - hdr_len);
6216 size = skb_headlen(skb);
6217 data_len = skb->data_len;
6219 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
6223 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
6224 if (dma_mapping_error(tx_ring->dev, dma))
6227 /* record length, and DMA address */
6228 dma_unmap_len_set(tx_buffer, len, size);
6229 dma_unmap_addr_set(tx_buffer, dma, dma);
6231 tx_desc->read.buffer_addr = cpu_to_le64(dma);
6233 while (unlikely(size > IGB_MAX_DATA_PER_TXD)) {
6234 tx_desc->read.cmd_type_len =
6235 cpu_to_le32(cmd_type ^ IGB_MAX_DATA_PER_TXD);
6239 if (i == tx_ring->count) {
6240 tx_desc = IGB_TX_DESC(tx_ring, 0);
6243 tx_desc->read.olinfo_status = 0;
6245 dma += IGB_MAX_DATA_PER_TXD;
6246 size -= IGB_MAX_DATA_PER_TXD;
6248 tx_desc->read.buffer_addr = cpu_to_le64(dma);
6251 if (likely(!data_len))
6254 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
6258 if (i == tx_ring->count) {
6259 tx_desc = IGB_TX_DESC(tx_ring, 0);
6262 tx_desc->read.olinfo_status = 0;
6264 size = skb_frag_size(frag);
6267 dma = skb_frag_dma_map(tx_ring->dev, frag, 0,
6268 size, DMA_TO_DEVICE);
6270 tx_buffer = &tx_ring->tx_buffer_info[i];
6273 /* write last descriptor with RS and EOP bits */
6274 cmd_type |= size | IGB_TXD_DCMD;
6275 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
6277 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
6279 /* set the timestamp */
6280 first->time_stamp = jiffies;
6282 skb_tx_timestamp(skb);
6284 /* Force memory writes to complete before letting h/w know there
6285 * are new descriptors to fetch. (Only applicable for weak-ordered
6286 * memory model archs, such as IA-64).
6288 * We also need this memory barrier to make certain all of the
6289 * status bits have been updated before next_to_watch is written.
6293 /* set next_to_watch value indicating a packet is present */
6294 first->next_to_watch = tx_desc;
6297 if (i == tx_ring->count)
6300 tx_ring->next_to_use = i;
6302 /* Make sure there is space in the ring for the next send. */
6303 igb_maybe_stop_tx(tx_ring, DESC_NEEDED);
6305 if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more()) {
6306 writel(i, tx_ring->tail);
6311 dev_err(tx_ring->dev, "TX DMA map failed\n");
6312 tx_buffer = &tx_ring->tx_buffer_info[i];
6314 /* clear dma mappings for failed tx_buffer_info map */
6315 while (tx_buffer != first) {
6316 if (dma_unmap_len(tx_buffer, len))
6317 dma_unmap_page(tx_ring->dev,
6318 dma_unmap_addr(tx_buffer, dma),
6319 dma_unmap_len(tx_buffer, len),
6321 dma_unmap_len_set(tx_buffer, len, 0);
6324 i += tx_ring->count;
6325 tx_buffer = &tx_ring->tx_buffer_info[i];
6328 if (dma_unmap_len(tx_buffer, len))
6329 dma_unmap_single(tx_ring->dev,
6330 dma_unmap_addr(tx_buffer, dma),
6331 dma_unmap_len(tx_buffer, len),
6333 dma_unmap_len_set(tx_buffer, len, 0);
6335 dev_kfree_skb_any(tx_buffer->skb);
6336 tx_buffer->skb = NULL;
6338 tx_ring->next_to_use = i;
6343 int igb_xmit_xdp_ring(struct igb_adapter *adapter,
6344 struct igb_ring *tx_ring,
6345 struct xdp_frame *xdpf)
6347 struct skb_shared_info *sinfo = xdp_get_shared_info_from_frame(xdpf);
6348 u8 nr_frags = unlikely(xdp_frame_has_frags(xdpf)) ? sinfo->nr_frags : 0;
6349 u16 count, i, index = tx_ring->next_to_use;
6350 struct igb_tx_buffer *tx_head = &tx_ring->tx_buffer_info[index];
6351 struct igb_tx_buffer *tx_buffer = tx_head;
6352 union e1000_adv_tx_desc *tx_desc = IGB_TX_DESC(tx_ring, index);
6353 u32 len = xdpf->len, cmd_type, olinfo_status;
6354 void *data = xdpf->data;
6356 count = TXD_USE_COUNT(len);
6357 for (i = 0; i < nr_frags; i++)
6358 count += TXD_USE_COUNT(skb_frag_size(&sinfo->frags[i]));
6360 if (igb_maybe_stop_tx(tx_ring, count + 3))
6361 return IGB_XDP_CONSUMED;
6364 /* record the location of the first descriptor for this packet */
6365 tx_head->bytecount = xdp_get_frame_len(xdpf);
6366 tx_head->type = IGB_TYPE_XDP;
6367 tx_head->gso_segs = 1;
6368 tx_head->xdpf = xdpf;
6370 olinfo_status = tx_head->bytecount << E1000_ADVTXD_PAYLEN_SHIFT;
6371 /* 82575 requires a unique index per ring */
6372 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
6373 olinfo_status |= tx_ring->reg_idx << 4;
6374 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
6379 dma = dma_map_single(tx_ring->dev, data, len, DMA_TO_DEVICE);
6380 if (dma_mapping_error(tx_ring->dev, dma))
6383 /* record length, and DMA address */
6384 dma_unmap_len_set(tx_buffer, len, len);
6385 dma_unmap_addr_set(tx_buffer, dma, dma);
6387 /* put descriptor type bits */
6388 cmd_type = E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_DEXT |
6389 E1000_ADVTXD_DCMD_IFCS | len;
6391 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
6392 tx_desc->read.buffer_addr = cpu_to_le64(dma);
6394 tx_buffer->protocol = 0;
6396 if (++index == tx_ring->count)
6402 tx_buffer = &tx_ring->tx_buffer_info[index];
6403 tx_desc = IGB_TX_DESC(tx_ring, index);
6404 tx_desc->read.olinfo_status = 0;
6406 data = skb_frag_address(&sinfo->frags[i]);
6407 len = skb_frag_size(&sinfo->frags[i]);
6410 tx_desc->read.cmd_type_len |= cpu_to_le32(IGB_TXD_DCMD);
6412 netdev_tx_sent_queue(txring_txq(tx_ring), tx_head->bytecount);
6413 /* set the timestamp */
6414 tx_head->time_stamp = jiffies;
6416 /* Avoid any potential race with xdp_xmit and cleanup */
6419 /* set next_to_watch value indicating a packet is present */
6420 tx_head->next_to_watch = tx_desc;
6421 tx_ring->next_to_use = index;
6423 /* Make sure there is space in the ring for the next send. */
6424 igb_maybe_stop_tx(tx_ring, DESC_NEEDED);
6426 if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more())
6427 writel(index, tx_ring->tail);
6433 tx_buffer = &tx_ring->tx_buffer_info[index];
6434 if (dma_unmap_len(tx_buffer, len))
6435 dma_unmap_page(tx_ring->dev,
6436 dma_unmap_addr(tx_buffer, dma),
6437 dma_unmap_len(tx_buffer, len),
6439 dma_unmap_len_set(tx_buffer, len, 0);
6440 if (tx_buffer == tx_head)
6444 index += tx_ring->count;
6448 return IGB_XDP_CONSUMED;
6451 netdev_tx_t igb_xmit_frame_ring(struct sk_buff *skb,
6452 struct igb_ring *tx_ring)
6454 struct igb_tx_buffer *first;
6458 u16 count = TXD_USE_COUNT(skb_headlen(skb));
6459 __be16 protocol = vlan_get_protocol(skb);
6462 /* need: 1 descriptor per page * PAGE_SIZE/IGB_MAX_DATA_PER_TXD,
6463 * + 1 desc for skb_headlen/IGB_MAX_DATA_PER_TXD,
6464 * + 2 desc gap to keep tail from touching head,
6465 * + 1 desc for context descriptor,
6466 * otherwise try next time
6468 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6469 count += TXD_USE_COUNT(skb_frag_size(
6470 &skb_shinfo(skb)->frags[f]));
6472 if (igb_maybe_stop_tx(tx_ring, count + 3)) {
6473 /* this is a hard error */
6474 return NETDEV_TX_BUSY;
6477 /* record the location of the first descriptor for this packet */
6478 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
6479 first->type = IGB_TYPE_SKB;
6481 first->bytecount = skb->len;
6482 first->gso_segs = 1;
6484 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
6485 struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
6487 if (adapter->tstamp_config.tx_type == HWTSTAMP_TX_ON &&
6488 !test_and_set_bit_lock(__IGB_PTP_TX_IN_PROGRESS,
6490 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
6491 tx_flags |= IGB_TX_FLAGS_TSTAMP;
6493 adapter->ptp_tx_skb = skb_get(skb);
6494 adapter->ptp_tx_start = jiffies;
6495 if (adapter->hw.mac.type == e1000_82576)
6496 schedule_work(&adapter->ptp_tx_work);
6498 adapter->tx_hwtstamp_skipped++;
6502 if (skb_vlan_tag_present(skb)) {
6503 tx_flags |= IGB_TX_FLAGS_VLAN;
6504 tx_flags |= (skb_vlan_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
6507 /* record initial flags and protocol */
6508 first->tx_flags = tx_flags;
6509 first->protocol = protocol;
6511 tso = igb_tso(tx_ring, first, &hdr_len);
6515 igb_tx_csum(tx_ring, first);
6517 if (igb_tx_map(tx_ring, first, hdr_len))
6518 goto cleanup_tx_tstamp;
6520 return NETDEV_TX_OK;
6523 dev_kfree_skb_any(first->skb);
6526 if (unlikely(tx_flags & IGB_TX_FLAGS_TSTAMP)) {
6527 struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
6529 dev_kfree_skb_any(adapter->ptp_tx_skb);
6530 adapter->ptp_tx_skb = NULL;
6531 if (adapter->hw.mac.type == e1000_82576)
6532 cancel_work_sync(&adapter->ptp_tx_work);
6533 clear_bit_unlock(__IGB_PTP_TX_IN_PROGRESS, &adapter->state);
6536 return NETDEV_TX_OK;
6539 static inline struct igb_ring *igb_tx_queue_mapping(struct igb_adapter *adapter,
6540 struct sk_buff *skb)
6542 unsigned int r_idx = skb->queue_mapping;
6544 if (r_idx >= adapter->num_tx_queues)
6545 r_idx = r_idx % adapter->num_tx_queues;
6547 return adapter->tx_ring[r_idx];
6550 static netdev_tx_t igb_xmit_frame(struct sk_buff *skb,
6551 struct net_device *netdev)
6553 struct igb_adapter *adapter = netdev_priv(netdev);
6555 /* The minimum packet size with TCTL.PSP set is 17 so pad the skb
6556 * in order to meet this minimum size requirement.
6558 if (skb_put_padto(skb, 17))
6559 return NETDEV_TX_OK;
6561 return igb_xmit_frame_ring(skb, igb_tx_queue_mapping(adapter, skb));
6565 * igb_tx_timeout - Respond to a Tx Hang
6566 * @netdev: network interface device structure
6567 * @txqueue: number of the Tx queue that hung (unused)
6569 static void igb_tx_timeout(struct net_device *netdev, unsigned int __always_unused txqueue)
6571 struct igb_adapter *adapter = netdev_priv(netdev);
6572 struct e1000_hw *hw = &adapter->hw;
6574 /* Do the reset outside of interrupt context */
6575 adapter->tx_timeout_count++;
6577 if (hw->mac.type >= e1000_82580)
6578 hw->dev_spec._82575.global_device_reset = true;
6580 schedule_work(&adapter->reset_task);
6582 (adapter->eims_enable_mask & ~adapter->eims_other));
6585 static void igb_reset_task(struct work_struct *work)
6587 struct igb_adapter *adapter;
6588 adapter = container_of(work, struct igb_adapter, reset_task);
6591 /* If we're already down or resetting, just bail */
6592 if (test_bit(__IGB_DOWN, &adapter->state) ||
6593 test_bit(__IGB_RESETTING, &adapter->state)) {
6599 netdev_err(adapter->netdev, "Reset adapter\n");
6600 igb_reinit_locked(adapter);
6605 * igb_get_stats64 - Get System Network Statistics
6606 * @netdev: network interface device structure
6607 * @stats: rtnl_link_stats64 pointer
6609 static void igb_get_stats64(struct net_device *netdev,
6610 struct rtnl_link_stats64 *stats)
6612 struct igb_adapter *adapter = netdev_priv(netdev);
6614 spin_lock(&adapter->stats64_lock);
6615 igb_update_stats(adapter);
6616 memcpy(stats, &adapter->stats64, sizeof(*stats));
6617 spin_unlock(&adapter->stats64_lock);
6621 * igb_change_mtu - Change the Maximum Transfer Unit
6622 * @netdev: network interface device structure
6623 * @new_mtu: new value for maximum frame size
6625 * Returns 0 on success, negative on failure
6627 static int igb_change_mtu(struct net_device *netdev, int new_mtu)
6629 struct igb_adapter *adapter = netdev_priv(netdev);
6630 int max_frame = new_mtu + IGB_ETH_PKT_HDR_PAD;
6632 if (adapter->xdp_prog) {
6635 for (i = 0; i < adapter->num_rx_queues; i++) {
6636 struct igb_ring *ring = adapter->rx_ring[i];
6638 if (max_frame > igb_rx_bufsz(ring)) {
6639 netdev_warn(adapter->netdev,
6640 "Requested MTU size is not supported with XDP. Max frame size is %d\n",
6647 /* adjust max frame to be at least the size of a standard frame */
6648 if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
6649 max_frame = ETH_FRAME_LEN + ETH_FCS_LEN;
6651 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
6652 usleep_range(1000, 2000);
6654 /* igb_down has a dependency on max_frame_size */
6655 adapter->max_frame_size = max_frame;
6657 if (netif_running(netdev))
6660 netdev_dbg(netdev, "changing MTU from %d to %d\n",
6661 netdev->mtu, new_mtu);
6662 netdev->mtu = new_mtu;
6664 if (netif_running(netdev))
6669 clear_bit(__IGB_RESETTING, &adapter->state);
6675 * igb_update_stats - Update the board statistics counters
6676 * @adapter: board private structure
6678 void igb_update_stats(struct igb_adapter *adapter)
6680 struct rtnl_link_stats64 *net_stats = &adapter->stats64;
6681 struct e1000_hw *hw = &adapter->hw;
6682 struct pci_dev *pdev = adapter->pdev;
6687 u64 _bytes, _packets;
6689 /* Prevent stats update while adapter is being reset, or if the pci
6690 * connection is down.
6692 if (adapter->link_speed == 0)
6694 if (pci_channel_offline(pdev))
6701 for (i = 0; i < adapter->num_rx_queues; i++) {
6702 struct igb_ring *ring = adapter->rx_ring[i];
6703 u32 rqdpc = rd32(E1000_RQDPC(i));
6704 if (hw->mac.type >= e1000_i210)
6705 wr32(E1000_RQDPC(i), 0);
6708 ring->rx_stats.drops += rqdpc;
6709 net_stats->rx_fifo_errors += rqdpc;
6713 start = u64_stats_fetch_begin(&ring->rx_syncp);
6714 _bytes = ring->rx_stats.bytes;
6715 _packets = ring->rx_stats.packets;
6716 } while (u64_stats_fetch_retry(&ring->rx_syncp, start));
6718 packets += _packets;
6721 net_stats->rx_bytes = bytes;
6722 net_stats->rx_packets = packets;
6726 for (i = 0; i < adapter->num_tx_queues; i++) {
6727 struct igb_ring *ring = adapter->tx_ring[i];
6729 start = u64_stats_fetch_begin(&ring->tx_syncp);
6730 _bytes = ring->tx_stats.bytes;
6731 _packets = ring->tx_stats.packets;
6732 } while (u64_stats_fetch_retry(&ring->tx_syncp, start));
6734 packets += _packets;
6736 net_stats->tx_bytes = bytes;
6737 net_stats->tx_packets = packets;
6740 /* read stats registers */
6741 adapter->stats.crcerrs += rd32(E1000_CRCERRS);
6742 adapter->stats.gprc += rd32(E1000_GPRC);
6743 adapter->stats.gorc += rd32(E1000_GORCL);
6744 rd32(E1000_GORCH); /* clear GORCL */
6745 adapter->stats.bprc += rd32(E1000_BPRC);
6746 adapter->stats.mprc += rd32(E1000_MPRC);
6747 adapter->stats.roc += rd32(E1000_ROC);
6749 adapter->stats.prc64 += rd32(E1000_PRC64);
6750 adapter->stats.prc127 += rd32(E1000_PRC127);
6751 adapter->stats.prc255 += rd32(E1000_PRC255);
6752 adapter->stats.prc511 += rd32(E1000_PRC511);
6753 adapter->stats.prc1023 += rd32(E1000_PRC1023);
6754 adapter->stats.prc1522 += rd32(E1000_PRC1522);
6755 adapter->stats.symerrs += rd32(E1000_SYMERRS);
6756 adapter->stats.sec += rd32(E1000_SEC);
6758 mpc = rd32(E1000_MPC);
6759 adapter->stats.mpc += mpc;
6760 net_stats->rx_fifo_errors += mpc;
6761 adapter->stats.scc += rd32(E1000_SCC);
6762 adapter->stats.ecol += rd32(E1000_ECOL);
6763 adapter->stats.mcc += rd32(E1000_MCC);
6764 adapter->stats.latecol += rd32(E1000_LATECOL);
6765 adapter->stats.dc += rd32(E1000_DC);
6766 adapter->stats.rlec += rd32(E1000_RLEC);
6767 adapter->stats.xonrxc += rd32(E1000_XONRXC);
6768 adapter->stats.xontxc += rd32(E1000_XONTXC);
6769 adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
6770 adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
6771 adapter->stats.fcruc += rd32(E1000_FCRUC);
6772 adapter->stats.gptc += rd32(E1000_GPTC);
6773 adapter->stats.gotc += rd32(E1000_GOTCL);
6774 rd32(E1000_GOTCH); /* clear GOTCL */
6775 adapter->stats.rnbc += rd32(E1000_RNBC);
6776 adapter->stats.ruc += rd32(E1000_RUC);
6777 adapter->stats.rfc += rd32(E1000_RFC);
6778 adapter->stats.rjc += rd32(E1000_RJC);
6779 adapter->stats.tor += rd32(E1000_TORH);
6780 adapter->stats.tot += rd32(E1000_TOTH);
6781 adapter->stats.tpr += rd32(E1000_TPR);
6783 adapter->stats.ptc64 += rd32(E1000_PTC64);
6784 adapter->stats.ptc127 += rd32(E1000_PTC127);
6785 adapter->stats.ptc255 += rd32(E1000_PTC255);
6786 adapter->stats.ptc511 += rd32(E1000_PTC511);
6787 adapter->stats.ptc1023 += rd32(E1000_PTC1023);
6788 adapter->stats.ptc1522 += rd32(E1000_PTC1522);
6790 adapter->stats.mptc += rd32(E1000_MPTC);
6791 adapter->stats.bptc += rd32(E1000_BPTC);
6793 adapter->stats.tpt += rd32(E1000_TPT);
6794 adapter->stats.colc += rd32(E1000_COLC);
6796 adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
6797 /* read internal phy specific stats */
6798 reg = rd32(E1000_CTRL_EXT);
6799 if (!(reg & E1000_CTRL_EXT_LINK_MODE_MASK)) {
6800 adapter->stats.rxerrc += rd32(E1000_RXERRC);
6802 /* this stat has invalid values on i210/i211 */
6803 if ((hw->mac.type != e1000_i210) &&
6804 (hw->mac.type != e1000_i211))
6805 adapter->stats.tncrs += rd32(E1000_TNCRS);
6808 adapter->stats.tsctc += rd32(E1000_TSCTC);
6809 adapter->stats.tsctfc += rd32(E1000_TSCTFC);
6811 adapter->stats.iac += rd32(E1000_IAC);
6812 adapter->stats.icrxoc += rd32(E1000_ICRXOC);
6813 adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
6814 adapter->stats.icrxatc += rd32(E1000_ICRXATC);
6815 adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
6816 adapter->stats.ictxatc += rd32(E1000_ICTXATC);
6817 adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
6818 adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
6819 adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
6821 /* Fill out the OS statistics structure */
6822 net_stats->multicast = adapter->stats.mprc;
6823 net_stats->collisions = adapter->stats.colc;
6827 /* RLEC on some newer hardware can be incorrect so build
6828 * our own version based on RUC and ROC
6830 net_stats->rx_errors = adapter->stats.rxerrc +
6831 adapter->stats.crcerrs + adapter->stats.algnerrc +
6832 adapter->stats.ruc + adapter->stats.roc +
6833 adapter->stats.cexterr;
6834 net_stats->rx_length_errors = adapter->stats.ruc +
6836 net_stats->rx_crc_errors = adapter->stats.crcerrs;
6837 net_stats->rx_frame_errors = adapter->stats.algnerrc;
6838 net_stats->rx_missed_errors = adapter->stats.mpc;
6841 net_stats->tx_errors = adapter->stats.ecol +
6842 adapter->stats.latecol;
6843 net_stats->tx_aborted_errors = adapter->stats.ecol;
6844 net_stats->tx_window_errors = adapter->stats.latecol;
6845 net_stats->tx_carrier_errors = adapter->stats.tncrs;
6847 /* Tx Dropped needs to be maintained elsewhere */
6849 /* Management Stats */
6850 adapter->stats.mgptc += rd32(E1000_MGTPTC);
6851 adapter->stats.mgprc += rd32(E1000_MGTPRC);
6852 adapter->stats.mgpdc += rd32(E1000_MGTPDC);
6855 reg = rd32(E1000_MANC);
6856 if (reg & E1000_MANC_EN_BMC2OS) {
6857 adapter->stats.o2bgptc += rd32(E1000_O2BGPTC);
6858 adapter->stats.o2bspc += rd32(E1000_O2BSPC);
6859 adapter->stats.b2ospc += rd32(E1000_B2OSPC);
6860 adapter->stats.b2ogprc += rd32(E1000_B2OGPRC);
6864 static void igb_perout(struct igb_adapter *adapter, int tsintr_tt)
6866 int pin = ptp_find_pin(adapter->ptp_clock, PTP_PF_PEROUT, tsintr_tt);
6867 struct e1000_hw *hw = &adapter->hw;
6868 struct timespec64 ts;
6871 if (pin < 0 || pin >= IGB_N_SDP)
6874 spin_lock(&adapter->tmreg_lock);
6876 if (hw->mac.type == e1000_82580 ||
6877 hw->mac.type == e1000_i354 ||
6878 hw->mac.type == e1000_i350) {
6879 s64 ns = timespec64_to_ns(&adapter->perout[tsintr_tt].period);
6880 u32 systiml, systimh, level_mask, level, rem;
6883 /* read systim registers in sequence */
6884 rd32(E1000_SYSTIMR);
6885 systiml = rd32(E1000_SYSTIML);
6886 systimh = rd32(E1000_SYSTIMH);
6887 systim = (((u64)(systimh & 0xFF)) << 32) | ((u64)systiml);
6888 now = timecounter_cyc2time(&adapter->tc, systim);
6891 level_mask = (tsintr_tt == 1) ? 0x80000 : 0x40000;
6892 level = (rd32(E1000_CTRL) & level_mask) ? 1 : 0;
6894 level_mask = (tsintr_tt == 1) ? 0x80 : 0x40;
6895 level = (rd32(E1000_CTRL_EXT) & level_mask) ? 1 : 0;
6898 div_u64_rem(now, ns, &rem);
6899 systim = systim + (ns - rem);
6901 /* synchronize pin level with rising/falling edges */
6902 div_u64_rem(now, ns << 1, &rem);
6904 /* first half of period */
6906 /* output is already low, skip this period */
6908 pr_notice("igb: periodic output on %s missed falling edge\n",
6909 adapter->sdp_config[pin].name);
6912 /* second half of period */
6914 /* output is already high, skip this period */
6916 pr_notice("igb: periodic output on %s missed rising edge\n",
6917 adapter->sdp_config[pin].name);
6921 /* for this chip family tv_sec is the upper part of the binary value,
6924 ts.tv_nsec = (u32)systim;
6925 ts.tv_sec = ((u32)(systim >> 32)) & 0xFF;
6927 ts = timespec64_add(adapter->perout[tsintr_tt].start,
6928 adapter->perout[tsintr_tt].period);
6931 /* u32 conversion of tv_sec is safe until y2106 */
6932 wr32((tsintr_tt == 1) ? E1000_TRGTTIML1 : E1000_TRGTTIML0, ts.tv_nsec);
6933 wr32((tsintr_tt == 1) ? E1000_TRGTTIMH1 : E1000_TRGTTIMH0, (u32)ts.tv_sec);
6934 tsauxc = rd32(E1000_TSAUXC);
6935 tsauxc |= TSAUXC_EN_TT0;
6936 wr32(E1000_TSAUXC, tsauxc);
6937 adapter->perout[tsintr_tt].start = ts;
6939 spin_unlock(&adapter->tmreg_lock);
6942 static void igb_extts(struct igb_adapter *adapter, int tsintr_tt)
6944 int pin = ptp_find_pin(adapter->ptp_clock, PTP_PF_EXTTS, tsintr_tt);
6945 int auxstmpl = (tsintr_tt == 1) ? E1000_AUXSTMPL1 : E1000_AUXSTMPL0;
6946 int auxstmph = (tsintr_tt == 1) ? E1000_AUXSTMPH1 : E1000_AUXSTMPH0;
6947 struct e1000_hw *hw = &adapter->hw;
6948 struct ptp_clock_event event;
6949 struct timespec64 ts;
6951 if (pin < 0 || pin >= IGB_N_SDP)
6954 if (hw->mac.type == e1000_82580 ||
6955 hw->mac.type == e1000_i354 ||
6956 hw->mac.type == e1000_i350) {
6957 s64 ns = rd32(auxstmpl);
6959 ns += ((s64)(rd32(auxstmph) & 0xFF)) << 32;
6960 ts = ns_to_timespec64(ns);
6962 ts.tv_nsec = rd32(auxstmpl);
6963 ts.tv_sec = rd32(auxstmph);
6966 event.type = PTP_CLOCK_EXTTS;
6967 event.index = tsintr_tt;
6968 event.timestamp = ts.tv_sec * 1000000000ULL + ts.tv_nsec;
6969 ptp_clock_event(adapter->ptp_clock, &event);
6972 static void igb_tsync_interrupt(struct igb_adapter *adapter)
6974 struct e1000_hw *hw = &adapter->hw;
6975 u32 ack = 0, tsicr = rd32(E1000_TSICR);
6976 struct ptp_clock_event event;
6978 if (tsicr & TSINTR_SYS_WRAP) {
6979 event.type = PTP_CLOCK_PPS;
6980 if (adapter->ptp_caps.pps)
6981 ptp_clock_event(adapter->ptp_clock, &event);
6982 ack |= TSINTR_SYS_WRAP;
6985 if (tsicr & E1000_TSICR_TXTS) {
6986 /* retrieve hardware timestamp */
6987 schedule_work(&adapter->ptp_tx_work);
6988 ack |= E1000_TSICR_TXTS;
6991 if (tsicr & TSINTR_TT0) {
6992 igb_perout(adapter, 0);
6996 if (tsicr & TSINTR_TT1) {
6997 igb_perout(adapter, 1);
7001 if (tsicr & TSINTR_AUTT0) {
7002 igb_extts(adapter, 0);
7003 ack |= TSINTR_AUTT0;
7006 if (tsicr & TSINTR_AUTT1) {
7007 igb_extts(adapter, 1);
7008 ack |= TSINTR_AUTT1;
7011 /* acknowledge the interrupts */
7012 wr32(E1000_TSICR, ack);
7015 static irqreturn_t igb_msix_other(int irq, void *data)
7017 struct igb_adapter *adapter = data;
7018 struct e1000_hw *hw = &adapter->hw;
7019 u32 icr = rd32(E1000_ICR);
7020 /* reading ICR causes bit 31 of EICR to be cleared */
7022 if (icr & E1000_ICR_DRSTA)
7023 schedule_work(&adapter->reset_task);
7025 if (icr & E1000_ICR_DOUTSYNC) {
7026 /* HW is reporting DMA is out of sync */
7027 adapter->stats.doosync++;
7028 /* The DMA Out of Sync is also indication of a spoof event
7029 * in IOV mode. Check the Wrong VM Behavior register to
7030 * see if it is really a spoof event.
7032 igb_check_wvbr(adapter);
7035 /* Check for a mailbox event */
7036 if (icr & E1000_ICR_VMMB)
7037 igb_msg_task(adapter);
7039 if (icr & E1000_ICR_LSC) {
7040 hw->mac.get_link_status = 1;
7041 /* guard against interrupt when we're going down */
7042 if (!test_bit(__IGB_DOWN, &adapter->state))
7043 mod_timer(&adapter->watchdog_timer, jiffies + 1);
7046 if (icr & E1000_ICR_TS)
7047 igb_tsync_interrupt(adapter);
7049 wr32(E1000_EIMS, adapter->eims_other);
7054 static void igb_write_itr(struct igb_q_vector *q_vector)
7056 struct igb_adapter *adapter = q_vector->adapter;
7057 u32 itr_val = q_vector->itr_val & 0x7FFC;
7059 if (!q_vector->set_itr)
7065 if (adapter->hw.mac.type == e1000_82575)
7066 itr_val |= itr_val << 16;
7068 itr_val |= E1000_EITR_CNT_IGNR;
7070 writel(itr_val, q_vector->itr_register);
7071 q_vector->set_itr = 0;
7074 static irqreturn_t igb_msix_ring(int irq, void *data)
7076 struct igb_q_vector *q_vector = data;
7078 /* Write the ITR value calculated from the previous interrupt. */
7079 igb_write_itr(q_vector);
7081 napi_schedule(&q_vector->napi);
7086 #ifdef CONFIG_IGB_DCA
7087 static void igb_update_tx_dca(struct igb_adapter *adapter,
7088 struct igb_ring *tx_ring,
7091 struct e1000_hw *hw = &adapter->hw;
7092 u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
7094 if (hw->mac.type != e1000_82575)
7095 txctrl <<= E1000_DCA_TXCTRL_CPUID_SHIFT;
7097 /* We can enable relaxed ordering for reads, but not writes when
7098 * DCA is enabled. This is due to a known issue in some chipsets
7099 * which will cause the DCA tag to be cleared.
7101 txctrl |= E1000_DCA_TXCTRL_DESC_RRO_EN |
7102 E1000_DCA_TXCTRL_DATA_RRO_EN |
7103 E1000_DCA_TXCTRL_DESC_DCA_EN;
7105 wr32(E1000_DCA_TXCTRL(tx_ring->reg_idx), txctrl);
7108 static void igb_update_rx_dca(struct igb_adapter *adapter,
7109 struct igb_ring *rx_ring,
7112 struct e1000_hw *hw = &adapter->hw;
7113 u32 rxctrl = dca3_get_tag(&adapter->pdev->dev, cpu);
7115 if (hw->mac.type != e1000_82575)
7116 rxctrl <<= E1000_DCA_RXCTRL_CPUID_SHIFT;
7118 /* We can enable relaxed ordering for reads, but not writes when
7119 * DCA is enabled. This is due to a known issue in some chipsets
7120 * which will cause the DCA tag to be cleared.
7122 rxctrl |= E1000_DCA_RXCTRL_DESC_RRO_EN |
7123 E1000_DCA_RXCTRL_DESC_DCA_EN;
7125 wr32(E1000_DCA_RXCTRL(rx_ring->reg_idx), rxctrl);
7128 static void igb_update_dca(struct igb_q_vector *q_vector)
7130 struct igb_adapter *adapter = q_vector->adapter;
7131 int cpu = get_cpu();
7133 if (q_vector->cpu == cpu)
7136 if (q_vector->tx.ring)
7137 igb_update_tx_dca(adapter, q_vector->tx.ring, cpu);
7139 if (q_vector->rx.ring)
7140 igb_update_rx_dca(adapter, q_vector->rx.ring, cpu);
7142 q_vector->cpu = cpu;
7147 static void igb_setup_dca(struct igb_adapter *adapter)
7149 struct e1000_hw *hw = &adapter->hw;
7152 if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
7155 /* Always use CB2 mode, difference is masked in the CB driver. */
7156 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
7158 for (i = 0; i < adapter->num_q_vectors; i++) {
7159 adapter->q_vector[i]->cpu = -1;
7160 igb_update_dca(adapter->q_vector[i]);
7164 static int __igb_notify_dca(struct device *dev, void *data)
7166 struct net_device *netdev = dev_get_drvdata(dev);
7167 struct igb_adapter *adapter = netdev_priv(netdev);
7168 struct pci_dev *pdev = adapter->pdev;
7169 struct e1000_hw *hw = &adapter->hw;
7170 unsigned long event = *(unsigned long *)data;
7173 case DCA_PROVIDER_ADD:
7174 /* if already enabled, don't do it again */
7175 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
7177 if (dca_add_requester(dev) == 0) {
7178 adapter->flags |= IGB_FLAG_DCA_ENABLED;
7179 dev_info(&pdev->dev, "DCA enabled\n");
7180 igb_setup_dca(adapter);
7183 fallthrough; /* since DCA is disabled. */
7184 case DCA_PROVIDER_REMOVE:
7185 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
7186 /* without this a class_device is left
7187 * hanging around in the sysfs model
7189 dca_remove_requester(dev);
7190 dev_info(&pdev->dev, "DCA disabled\n");
7191 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
7192 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
7200 static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
7205 ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
7208 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7210 #endif /* CONFIG_IGB_DCA */
7212 #ifdef CONFIG_PCI_IOV
7213 static int igb_vf_configure(struct igb_adapter *adapter, int vf)
7215 unsigned char mac_addr[ETH_ALEN];
7217 eth_zero_addr(mac_addr);
7218 igb_set_vf_mac(adapter, vf, mac_addr);
7220 /* By default spoof check is enabled for all VFs */
7221 adapter->vf_data[vf].spoofchk_enabled = true;
7223 /* By default VFs are not trusted */
7224 adapter->vf_data[vf].trusted = false;
7230 static void igb_ping_all_vfs(struct igb_adapter *adapter)
7232 struct e1000_hw *hw = &adapter->hw;
7236 for (i = 0 ; i < adapter->vfs_allocated_count; i++) {
7237 ping = E1000_PF_CONTROL_MSG;
7238 if (adapter->vf_data[i].flags & IGB_VF_FLAG_CTS)
7239 ping |= E1000_VT_MSGTYPE_CTS;
7240 igb_write_mbx(hw, &ping, 1, i);
7244 static int igb_set_vf_promisc(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
7246 struct e1000_hw *hw = &adapter->hw;
7247 u32 vmolr = rd32(E1000_VMOLR(vf));
7248 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7250 vf_data->flags &= ~(IGB_VF_FLAG_UNI_PROMISC |
7251 IGB_VF_FLAG_MULTI_PROMISC);
7252 vmolr &= ~(E1000_VMOLR_ROPE | E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
7254 if (*msgbuf & E1000_VF_SET_PROMISC_MULTICAST) {
7255 vmolr |= E1000_VMOLR_MPME;
7256 vf_data->flags |= IGB_VF_FLAG_MULTI_PROMISC;
7257 *msgbuf &= ~E1000_VF_SET_PROMISC_MULTICAST;
7259 /* if we have hashes and we are clearing a multicast promisc
7260 * flag we need to write the hashes to the MTA as this step
7261 * was previously skipped
7263 if (vf_data->num_vf_mc_hashes > 30) {
7264 vmolr |= E1000_VMOLR_MPME;
7265 } else if (vf_data->num_vf_mc_hashes) {
7268 vmolr |= E1000_VMOLR_ROMPE;
7269 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
7270 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
7274 wr32(E1000_VMOLR(vf), vmolr);
7276 /* there are flags left unprocessed, likely not supported */
7277 if (*msgbuf & E1000_VT_MSGINFO_MASK)
7283 static int igb_set_vf_multicasts(struct igb_adapter *adapter,
7284 u32 *msgbuf, u32 vf)
7286 int n = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
7287 u16 *hash_list = (u16 *)&msgbuf[1];
7288 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7291 /* salt away the number of multicast addresses assigned
7292 * to this VF for later use to restore when the PF multi cast
7295 vf_data->num_vf_mc_hashes = n;
7297 /* only up to 30 hash values supported */
7301 /* store the hashes for later use */
7302 for (i = 0; i < n; i++)
7303 vf_data->vf_mc_hashes[i] = hash_list[i];
7305 /* Flush and reset the mta with the new values */
7306 igb_set_rx_mode(adapter->netdev);
7311 static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
7313 struct e1000_hw *hw = &adapter->hw;
7314 struct vf_data_storage *vf_data;
7317 for (i = 0; i < adapter->vfs_allocated_count; i++) {
7318 u32 vmolr = rd32(E1000_VMOLR(i));
7320 vmolr &= ~(E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
7322 vf_data = &adapter->vf_data[i];
7324 if ((vf_data->num_vf_mc_hashes > 30) ||
7325 (vf_data->flags & IGB_VF_FLAG_MULTI_PROMISC)) {
7326 vmolr |= E1000_VMOLR_MPME;
7327 } else if (vf_data->num_vf_mc_hashes) {
7328 vmolr |= E1000_VMOLR_ROMPE;
7329 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
7330 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
7332 wr32(E1000_VMOLR(i), vmolr);
7336 static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
7338 struct e1000_hw *hw = &adapter->hw;
7339 u32 pool_mask, vlvf_mask, i;
7341 /* create mask for VF and other pools */
7342 pool_mask = E1000_VLVF_POOLSEL_MASK;
7343 vlvf_mask = BIT(E1000_VLVF_POOLSEL_SHIFT + vf);
7345 /* drop PF from pool bits */
7346 pool_mask &= ~BIT(E1000_VLVF_POOLSEL_SHIFT +
7347 adapter->vfs_allocated_count);
7349 /* Find the vlan filter for this id */
7350 for (i = E1000_VLVF_ARRAY_SIZE; i--;) {
7351 u32 vlvf = rd32(E1000_VLVF(i));
7352 u32 vfta_mask, vid, vfta;
7354 /* remove the vf from the pool */
7355 if (!(vlvf & vlvf_mask))
7358 /* clear out bit from VLVF */
7361 /* if other pools are present, just remove ourselves */
7362 if (vlvf & pool_mask)
7365 /* if PF is present, leave VFTA */
7366 if (vlvf & E1000_VLVF_POOLSEL_MASK)
7369 vid = vlvf & E1000_VLVF_VLANID_MASK;
7370 vfta_mask = BIT(vid % 32);
7372 /* clear bit from VFTA */
7373 vfta = adapter->shadow_vfta[vid / 32];
7374 if (vfta & vfta_mask)
7375 hw->mac.ops.write_vfta(hw, vid / 32, vfta ^ vfta_mask);
7377 /* clear pool selection enable */
7378 if (adapter->flags & IGB_FLAG_VLAN_PROMISC)
7379 vlvf &= E1000_VLVF_POOLSEL_MASK;
7383 /* clear pool bits */
7384 wr32(E1000_VLVF(i), vlvf);
7388 static int igb_find_vlvf_entry(struct e1000_hw *hw, u32 vlan)
7393 /* short cut the special case */
7397 /* Search for the VLAN id in the VLVF entries */
7398 for (idx = E1000_VLVF_ARRAY_SIZE; --idx;) {
7399 vlvf = rd32(E1000_VLVF(idx));
7400 if ((vlvf & VLAN_VID_MASK) == vlan)
7407 static void igb_update_pf_vlvf(struct igb_adapter *adapter, u32 vid)
7409 struct e1000_hw *hw = &adapter->hw;
7413 idx = igb_find_vlvf_entry(hw, vid);
7417 /* See if any other pools are set for this VLAN filter
7418 * entry other than the PF.
7420 pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT;
7421 bits = ~BIT(pf_id) & E1000_VLVF_POOLSEL_MASK;
7422 bits &= rd32(E1000_VLVF(idx));
7424 /* Disable the filter so this falls into the default pool. */
7426 if (adapter->flags & IGB_FLAG_VLAN_PROMISC)
7427 wr32(E1000_VLVF(idx), BIT(pf_id));
7429 wr32(E1000_VLVF(idx), 0);
7433 static s32 igb_set_vf_vlan(struct igb_adapter *adapter, u32 vid,
7436 int pf_id = adapter->vfs_allocated_count;
7437 struct e1000_hw *hw = &adapter->hw;
7440 /* If VLAN overlaps with one the PF is currently monitoring make
7441 * sure that we are able to allocate a VLVF entry. This may be
7442 * redundant but it guarantees PF will maintain visibility to
7445 if (add && test_bit(vid, adapter->active_vlans)) {
7446 err = igb_vfta_set(hw, vid, pf_id, true, false);
7451 err = igb_vfta_set(hw, vid, vf, add, false);
7456 /* If we failed to add the VF VLAN or we are removing the VF VLAN
7457 * we may need to drop the PF pool bit in order to allow us to free
7458 * up the VLVF resources.
7460 if (test_bit(vid, adapter->active_vlans) ||
7461 (adapter->flags & IGB_FLAG_VLAN_PROMISC))
7462 igb_update_pf_vlvf(adapter, vid);
7467 static void igb_set_vmvir(struct igb_adapter *adapter, u32 vid, u32 vf)
7469 struct e1000_hw *hw = &adapter->hw;
7472 wr32(E1000_VMVIR(vf), (vid | E1000_VMVIR_VLANA_DEFAULT));
7474 wr32(E1000_VMVIR(vf), 0);
7477 static int igb_enable_port_vlan(struct igb_adapter *adapter, int vf,
7482 err = igb_set_vf_vlan(adapter, vlan, true, vf);
7486 igb_set_vmvir(adapter, vlan | (qos << VLAN_PRIO_SHIFT), vf);
7487 igb_set_vmolr(adapter, vf, !vlan);
7489 /* revoke access to previous VLAN */
7490 if (vlan != adapter->vf_data[vf].pf_vlan)
7491 igb_set_vf_vlan(adapter, adapter->vf_data[vf].pf_vlan,
7494 adapter->vf_data[vf].pf_vlan = vlan;
7495 adapter->vf_data[vf].pf_qos = qos;
7496 igb_set_vf_vlan_strip(adapter, vf, true);
7497 dev_info(&adapter->pdev->dev,
7498 "Setting VLAN %d, QOS 0x%x on VF %d\n", vlan, qos, vf);
7499 if (test_bit(__IGB_DOWN, &adapter->state)) {
7500 dev_warn(&adapter->pdev->dev,
7501 "The VF VLAN has been set, but the PF device is not up.\n");
7502 dev_warn(&adapter->pdev->dev,
7503 "Bring the PF device up before attempting to use the VF device.\n");
7509 static int igb_disable_port_vlan(struct igb_adapter *adapter, int vf)
7511 /* Restore tagless access via VLAN 0 */
7512 igb_set_vf_vlan(adapter, 0, true, vf);
7514 igb_set_vmvir(adapter, 0, vf);
7515 igb_set_vmolr(adapter, vf, true);
7517 /* Remove any PF assigned VLAN */
7518 if (adapter->vf_data[vf].pf_vlan)
7519 igb_set_vf_vlan(adapter, adapter->vf_data[vf].pf_vlan,
7522 adapter->vf_data[vf].pf_vlan = 0;
7523 adapter->vf_data[vf].pf_qos = 0;
7524 igb_set_vf_vlan_strip(adapter, vf, false);
7529 static int igb_ndo_set_vf_vlan(struct net_device *netdev, int vf,
7530 u16 vlan, u8 qos, __be16 vlan_proto)
7532 struct igb_adapter *adapter = netdev_priv(netdev);
7534 if ((vf >= adapter->vfs_allocated_count) || (vlan > 4095) || (qos > 7))
7537 if (vlan_proto != htons(ETH_P_8021Q))
7538 return -EPROTONOSUPPORT;
7540 return (vlan || qos) ? igb_enable_port_vlan(adapter, vf, vlan, qos) :
7541 igb_disable_port_vlan(adapter, vf);
7544 static int igb_set_vf_vlan_msg(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
7546 int add = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
7547 int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
7550 if (adapter->vf_data[vf].pf_vlan)
7553 /* VLAN 0 is a special case, don't allow it to be removed */
7557 ret = igb_set_vf_vlan(adapter, vid, !!add, vf);
7559 igb_set_vf_vlan_strip(adapter, vf, !!vid);
7563 static inline void igb_vf_reset(struct igb_adapter *adapter, u32 vf)
7565 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7567 /* clear flags - except flag that indicates PF has set the MAC */
7568 vf_data->flags &= IGB_VF_FLAG_PF_SET_MAC;
7569 vf_data->last_nack = jiffies;
7571 /* reset vlans for device */
7572 igb_clear_vf_vfta(adapter, vf);
7573 igb_set_vf_vlan(adapter, vf_data->pf_vlan, true, vf);
7574 igb_set_vmvir(adapter, vf_data->pf_vlan |
7575 (vf_data->pf_qos << VLAN_PRIO_SHIFT), vf);
7576 igb_set_vmolr(adapter, vf, !vf_data->pf_vlan);
7577 igb_set_vf_vlan_strip(adapter, vf, !!(vf_data->pf_vlan));
7579 /* reset multicast table array for vf */
7580 adapter->vf_data[vf].num_vf_mc_hashes = 0;
7582 /* Flush and reset the mta with the new values */
7583 igb_set_rx_mode(adapter->netdev);
7586 static void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
7588 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
7590 /* clear mac address as we were hotplug removed/added */
7591 if (!(adapter->vf_data[vf].flags & IGB_VF_FLAG_PF_SET_MAC))
7592 eth_zero_addr(vf_mac);
7594 /* process remaining reset events */
7595 igb_vf_reset(adapter, vf);
7598 static void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
7600 struct e1000_hw *hw = &adapter->hw;
7601 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
7602 u32 reg, msgbuf[3] = {};
7603 u8 *addr = (u8 *)(&msgbuf[1]);
7605 /* process all the same items cleared in a function level reset */
7606 igb_vf_reset(adapter, vf);
7608 /* set vf mac address */
7609 igb_set_vf_mac(adapter, vf, vf_mac);
7611 /* enable transmit and receive for vf */
7612 reg = rd32(E1000_VFTE);
7613 wr32(E1000_VFTE, reg | BIT(vf));
7614 reg = rd32(E1000_VFRE);
7615 wr32(E1000_VFRE, reg | BIT(vf));
7617 adapter->vf_data[vf].flags |= IGB_VF_FLAG_CTS;
7619 /* reply to reset with ack and vf mac address */
7620 if (!is_zero_ether_addr(vf_mac)) {
7621 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
7622 memcpy(addr, vf_mac, ETH_ALEN);
7624 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_NACK;
7626 igb_write_mbx(hw, msgbuf, 3, vf);
7629 static void igb_flush_mac_table(struct igb_adapter *adapter)
7631 struct e1000_hw *hw = &adapter->hw;
7634 for (i = 0; i < hw->mac.rar_entry_count; i++) {
7635 adapter->mac_table[i].state &= ~IGB_MAC_STATE_IN_USE;
7636 eth_zero_addr(adapter->mac_table[i].addr);
7637 adapter->mac_table[i].queue = 0;
7638 igb_rar_set_index(adapter, i);
7642 static int igb_available_rars(struct igb_adapter *adapter, u8 queue)
7644 struct e1000_hw *hw = &adapter->hw;
7645 /* do not count rar entries reserved for VFs MAC addresses */
7646 int rar_entries = hw->mac.rar_entry_count -
7647 adapter->vfs_allocated_count;
7650 for (i = 0; i < rar_entries; i++) {
7651 /* do not count default entries */
7652 if (adapter->mac_table[i].state & IGB_MAC_STATE_DEFAULT)
7655 /* do not count "in use" entries for different queues */
7656 if ((adapter->mac_table[i].state & IGB_MAC_STATE_IN_USE) &&
7657 (adapter->mac_table[i].queue != queue))
7666 /* Set default MAC address for the PF in the first RAR entry */
7667 static void igb_set_default_mac_filter(struct igb_adapter *adapter)
7669 struct igb_mac_addr *mac_table = &adapter->mac_table[0];
7671 ether_addr_copy(mac_table->addr, adapter->hw.mac.addr);
7672 mac_table->queue = adapter->vfs_allocated_count;
7673 mac_table->state = IGB_MAC_STATE_DEFAULT | IGB_MAC_STATE_IN_USE;
7675 igb_rar_set_index(adapter, 0);
7678 /* If the filter to be added and an already existing filter express
7679 * the same address and address type, it should be possible to only
7680 * override the other configurations, for example the queue to steer
7683 static bool igb_mac_entry_can_be_used(const struct igb_mac_addr *entry,
7684 const u8 *addr, const u8 flags)
7686 if (!(entry->state & IGB_MAC_STATE_IN_USE))
7689 if ((entry->state & IGB_MAC_STATE_SRC_ADDR) !=
7690 (flags & IGB_MAC_STATE_SRC_ADDR))
7693 if (!ether_addr_equal(addr, entry->addr))
7699 /* Add a MAC filter for 'addr' directing matching traffic to 'queue',
7700 * 'flags' is used to indicate what kind of match is made, match is by
7701 * default for the destination address, if matching by source address
7702 * is desired the flag IGB_MAC_STATE_SRC_ADDR can be used.
7704 static int igb_add_mac_filter_flags(struct igb_adapter *adapter,
7705 const u8 *addr, const u8 queue,
7708 struct e1000_hw *hw = &adapter->hw;
7709 int rar_entries = hw->mac.rar_entry_count -
7710 adapter->vfs_allocated_count;
7713 if (is_zero_ether_addr(addr))
7716 /* Search for the first empty entry in the MAC table.
7717 * Do not touch entries at the end of the table reserved for the VF MAC
7720 for (i = 0; i < rar_entries; i++) {
7721 if (!igb_mac_entry_can_be_used(&adapter->mac_table[i],
7725 ether_addr_copy(adapter->mac_table[i].addr, addr);
7726 adapter->mac_table[i].queue = queue;
7727 adapter->mac_table[i].state |= IGB_MAC_STATE_IN_USE | flags;
7729 igb_rar_set_index(adapter, i);
7736 static int igb_add_mac_filter(struct igb_adapter *adapter, const u8 *addr,
7739 return igb_add_mac_filter_flags(adapter, addr, queue, 0);
7742 /* Remove a MAC filter for 'addr' directing matching traffic to
7743 * 'queue', 'flags' is used to indicate what kind of match need to be
7744 * removed, match is by default for the destination address, if
7745 * matching by source address is to be removed the flag
7746 * IGB_MAC_STATE_SRC_ADDR can be used.
7748 static int igb_del_mac_filter_flags(struct igb_adapter *adapter,
7749 const u8 *addr, const u8 queue,
7752 struct e1000_hw *hw = &adapter->hw;
7753 int rar_entries = hw->mac.rar_entry_count -
7754 adapter->vfs_allocated_count;
7757 if (is_zero_ether_addr(addr))
7760 /* Search for matching entry in the MAC table based on given address
7761 * and queue. Do not touch entries at the end of the table reserved
7762 * for the VF MAC addresses.
7764 for (i = 0; i < rar_entries; i++) {
7765 if (!(adapter->mac_table[i].state & IGB_MAC_STATE_IN_USE))
7767 if ((adapter->mac_table[i].state & flags) != flags)
7769 if (adapter->mac_table[i].queue != queue)
7771 if (!ether_addr_equal(adapter->mac_table[i].addr, addr))
7774 /* When a filter for the default address is "deleted",
7775 * we return it to its initial configuration
7777 if (adapter->mac_table[i].state & IGB_MAC_STATE_DEFAULT) {
7778 adapter->mac_table[i].state =
7779 IGB_MAC_STATE_DEFAULT | IGB_MAC_STATE_IN_USE;
7780 adapter->mac_table[i].queue =
7781 adapter->vfs_allocated_count;
7783 adapter->mac_table[i].state = 0;
7784 adapter->mac_table[i].queue = 0;
7785 eth_zero_addr(adapter->mac_table[i].addr);
7788 igb_rar_set_index(adapter, i);
7795 static int igb_del_mac_filter(struct igb_adapter *adapter, const u8 *addr,
7798 return igb_del_mac_filter_flags(adapter, addr, queue, 0);
7801 int igb_add_mac_steering_filter(struct igb_adapter *adapter,
7802 const u8 *addr, u8 queue, u8 flags)
7804 struct e1000_hw *hw = &adapter->hw;
7806 /* In theory, this should be supported on 82575 as well, but
7807 * that part wasn't easily accessible during development.
7809 if (hw->mac.type != e1000_i210)
7812 return igb_add_mac_filter_flags(adapter, addr, queue,
7813 IGB_MAC_STATE_QUEUE_STEERING | flags);
7816 int igb_del_mac_steering_filter(struct igb_adapter *adapter,
7817 const u8 *addr, u8 queue, u8 flags)
7819 return igb_del_mac_filter_flags(adapter, addr, queue,
7820 IGB_MAC_STATE_QUEUE_STEERING | flags);
7823 static int igb_uc_sync(struct net_device *netdev, const unsigned char *addr)
7825 struct igb_adapter *adapter = netdev_priv(netdev);
7828 ret = igb_add_mac_filter(adapter, addr, adapter->vfs_allocated_count);
7830 return min_t(int, ret, 0);
7833 static int igb_uc_unsync(struct net_device *netdev, const unsigned char *addr)
7835 struct igb_adapter *adapter = netdev_priv(netdev);
7837 igb_del_mac_filter(adapter, addr, adapter->vfs_allocated_count);
7842 static int igb_set_vf_mac_filter(struct igb_adapter *adapter, const int vf,
7843 const u32 info, const u8 *addr)
7845 struct pci_dev *pdev = adapter->pdev;
7846 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7847 struct list_head *pos;
7848 struct vf_mac_filter *entry = NULL;
7851 if ((vf_data->flags & IGB_VF_FLAG_PF_SET_MAC) &&
7852 !vf_data->trusted) {
7853 dev_warn(&pdev->dev,
7854 "VF %d requested MAC filter but is administratively denied\n",
7858 if (!is_valid_ether_addr(addr)) {
7859 dev_warn(&pdev->dev,
7860 "VF %d attempted to set invalid MAC filter\n",
7866 case E1000_VF_MAC_FILTER_CLR:
7867 /* remove all unicast MAC filters related to the current VF */
7868 list_for_each(pos, &adapter->vf_macs.l) {
7869 entry = list_entry(pos, struct vf_mac_filter, l);
7870 if (entry->vf == vf) {
7873 igb_del_mac_filter(adapter, entry->vf_mac, vf);
7877 case E1000_VF_MAC_FILTER_ADD:
7878 /* try to find empty slot in the list */
7879 list_for_each(pos, &adapter->vf_macs.l) {
7880 entry = list_entry(pos, struct vf_mac_filter, l);
7885 if (entry && entry->free) {
7886 entry->free = false;
7888 ether_addr_copy(entry->vf_mac, addr);
7890 ret = igb_add_mac_filter(adapter, addr, vf);
7891 ret = min_t(int, ret, 0);
7897 dev_warn(&pdev->dev,
7898 "VF %d has requested MAC filter but there is no space for it\n",
7909 static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf)
7911 struct pci_dev *pdev = adapter->pdev;
7912 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7913 u32 info = msg[0] & E1000_VT_MSGINFO_MASK;
7915 /* The VF MAC Address is stored in a packed array of bytes
7916 * starting at the second 32 bit word of the msg array
7918 unsigned char *addr = (unsigned char *)&msg[1];
7922 if ((vf_data->flags & IGB_VF_FLAG_PF_SET_MAC) &&
7923 !vf_data->trusted) {
7924 dev_warn(&pdev->dev,
7925 "VF %d attempted to override administratively set MAC address\nReload the VF driver to resume operations\n",
7930 if (!is_valid_ether_addr(addr)) {
7931 dev_warn(&pdev->dev,
7932 "VF %d attempted to set invalid MAC\n",
7937 ret = igb_set_vf_mac(adapter, vf, addr);
7939 ret = igb_set_vf_mac_filter(adapter, vf, info, addr);
7945 static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf)
7947 struct e1000_hw *hw = &adapter->hw;
7948 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7949 u32 msg = E1000_VT_MSGTYPE_NACK;
7951 /* if device isn't clear to send it shouldn't be reading either */
7952 if (!(vf_data->flags & IGB_VF_FLAG_CTS) &&
7953 time_after(jiffies, vf_data->last_nack + (2 * HZ))) {
7954 igb_write_mbx(hw, &msg, 1, vf);
7955 vf_data->last_nack = jiffies;
7959 static void igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
7961 struct pci_dev *pdev = adapter->pdev;
7962 u32 msgbuf[E1000_VFMAILBOX_SIZE];
7963 struct e1000_hw *hw = &adapter->hw;
7964 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7967 retval = igb_read_mbx(hw, msgbuf, E1000_VFMAILBOX_SIZE, vf, false);
7970 /* if receive failed revoke VF CTS stats and restart init */
7971 dev_err(&pdev->dev, "Error receiving message from VF\n");
7972 vf_data->flags &= ~IGB_VF_FLAG_CTS;
7973 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
7978 /* this is a message we already processed, do nothing */
7979 if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
7982 /* until the vf completes a reset it should not be
7983 * allowed to start any configuration.
7985 if (msgbuf[0] == E1000_VF_RESET) {
7986 /* unlocks mailbox */
7987 igb_vf_reset_msg(adapter, vf);
7991 if (!(vf_data->flags & IGB_VF_FLAG_CTS)) {
7992 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
7998 switch ((msgbuf[0] & 0xFFFF)) {
7999 case E1000_VF_SET_MAC_ADDR:
8000 retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
8002 case E1000_VF_SET_PROMISC:
8003 retval = igb_set_vf_promisc(adapter, msgbuf, vf);
8005 case E1000_VF_SET_MULTICAST:
8006 retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
8008 case E1000_VF_SET_LPE:
8009 retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf);
8011 case E1000_VF_SET_VLAN:
8013 if (vf_data->pf_vlan)
8014 dev_warn(&pdev->dev,
8015 "VF %d attempted to override administratively set VLAN tag\nReload the VF driver to resume operations\n",
8018 retval = igb_set_vf_vlan_msg(adapter, msgbuf, vf);
8021 dev_err(&pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
8026 msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
8028 /* notify the VF of the results of what it sent us */
8030 msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
8032 msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
8034 /* unlocks mailbox */
8035 igb_write_mbx(hw, msgbuf, 1, vf);
8039 igb_unlock_mbx(hw, vf);
8042 static void igb_msg_task(struct igb_adapter *adapter)
8044 struct e1000_hw *hw = &adapter->hw;
8045 unsigned long flags;
8048 spin_lock_irqsave(&adapter->vfs_lock, flags);
8049 for (vf = 0; vf < adapter->vfs_allocated_count; vf++) {
8050 /* process any reset requests */
8051 if (!igb_check_for_rst(hw, vf))
8052 igb_vf_reset_event(adapter, vf);
8054 /* process any messages pending */
8055 if (!igb_check_for_msg(hw, vf))
8056 igb_rcv_msg_from_vf(adapter, vf);
8058 /* process any acks */
8059 if (!igb_check_for_ack(hw, vf))
8060 igb_rcv_ack_from_vf(adapter, vf);
8062 spin_unlock_irqrestore(&adapter->vfs_lock, flags);
8066 * igb_set_uta - Set unicast filter table address
8067 * @adapter: board private structure
8068 * @set: boolean indicating if we are setting or clearing bits
8070 * The unicast table address is a register array of 32-bit registers.
8071 * The table is meant to be used in a way similar to how the MTA is used
8072 * however due to certain limitations in the hardware it is necessary to
8073 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
8074 * enable bit to allow vlan tag stripping when promiscuous mode is enabled
8076 static void igb_set_uta(struct igb_adapter *adapter, bool set)
8078 struct e1000_hw *hw = &adapter->hw;
8079 u32 uta = set ? ~0 : 0;
8082 /* we only need to do this if VMDq is enabled */
8083 if (!adapter->vfs_allocated_count)
8086 for (i = hw->mac.uta_reg_count; i--;)
8087 array_wr32(E1000_UTA, i, uta);
8091 * igb_intr_msi - Interrupt Handler
8092 * @irq: interrupt number
8093 * @data: pointer to a network interface device structure
8095 static irqreturn_t igb_intr_msi(int irq, void *data)
8097 struct igb_adapter *adapter = data;
8098 struct igb_q_vector *q_vector = adapter->q_vector[0];
8099 struct e1000_hw *hw = &adapter->hw;
8100 /* read ICR disables interrupts using IAM */
8101 u32 icr = rd32(E1000_ICR);
8103 igb_write_itr(q_vector);
8105 if (icr & E1000_ICR_DRSTA)
8106 schedule_work(&adapter->reset_task);
8108 if (icr & E1000_ICR_DOUTSYNC) {
8109 /* HW is reporting DMA is out of sync */
8110 adapter->stats.doosync++;
8113 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
8114 hw->mac.get_link_status = 1;
8115 if (!test_bit(__IGB_DOWN, &adapter->state))
8116 mod_timer(&adapter->watchdog_timer, jiffies + 1);
8119 if (icr & E1000_ICR_TS)
8120 igb_tsync_interrupt(adapter);
8122 napi_schedule(&q_vector->napi);
8128 * igb_intr - Legacy Interrupt Handler
8129 * @irq: interrupt number
8130 * @data: pointer to a network interface device structure
8132 static irqreturn_t igb_intr(int irq, void *data)
8134 struct igb_adapter *adapter = data;
8135 struct igb_q_vector *q_vector = adapter->q_vector[0];
8136 struct e1000_hw *hw = &adapter->hw;
8137 /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked. No
8138 * need for the IMC write
8140 u32 icr = rd32(E1000_ICR);
8142 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
8143 * not set, then the adapter didn't send an interrupt
8145 if (!(icr & E1000_ICR_INT_ASSERTED))
8148 igb_write_itr(q_vector);
8150 if (icr & E1000_ICR_DRSTA)
8151 schedule_work(&adapter->reset_task);
8153 if (icr & E1000_ICR_DOUTSYNC) {
8154 /* HW is reporting DMA is out of sync */
8155 adapter->stats.doosync++;
8158 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
8159 hw->mac.get_link_status = 1;
8160 /* guard against interrupt when we're going down */
8161 if (!test_bit(__IGB_DOWN, &adapter->state))
8162 mod_timer(&adapter->watchdog_timer, jiffies + 1);
8165 if (icr & E1000_ICR_TS)
8166 igb_tsync_interrupt(adapter);
8168 napi_schedule(&q_vector->napi);
8173 static void igb_ring_irq_enable(struct igb_q_vector *q_vector)
8175 struct igb_adapter *adapter = q_vector->adapter;
8176 struct e1000_hw *hw = &adapter->hw;
8178 if ((q_vector->rx.ring && (adapter->rx_itr_setting & 3)) ||
8179 (!q_vector->rx.ring && (adapter->tx_itr_setting & 3))) {
8180 if ((adapter->num_q_vectors == 1) && !adapter->vf_data)
8181 igb_set_itr(q_vector);
8183 igb_update_ring_itr(q_vector);
8186 if (!test_bit(__IGB_DOWN, &adapter->state)) {
8187 if (adapter->flags & IGB_FLAG_HAS_MSIX)
8188 wr32(E1000_EIMS, q_vector->eims_value);
8190 igb_irq_enable(adapter);
8195 * igb_poll - NAPI Rx polling callback
8196 * @napi: napi polling structure
8197 * @budget: count of how many packets we should handle
8199 static int igb_poll(struct napi_struct *napi, int budget)
8201 struct igb_q_vector *q_vector = container_of(napi,
8202 struct igb_q_vector,
8204 bool clean_complete = true;
8207 #ifdef CONFIG_IGB_DCA
8208 if (q_vector->adapter->flags & IGB_FLAG_DCA_ENABLED)
8209 igb_update_dca(q_vector);
8211 if (q_vector->tx.ring)
8212 clean_complete = igb_clean_tx_irq(q_vector, budget);
8214 if (q_vector->rx.ring) {
8215 int cleaned = igb_clean_rx_irq(q_vector, budget);
8217 work_done += cleaned;
8218 if (cleaned >= budget)
8219 clean_complete = false;
8222 /* If all work not completed, return budget and keep polling */
8223 if (!clean_complete)
8226 /* Exit the polling mode, but don't re-enable interrupts if stack might
8227 * poll us due to busy-polling
8229 if (likely(napi_complete_done(napi, work_done)))
8230 igb_ring_irq_enable(q_vector);
8236 * igb_clean_tx_irq - Reclaim resources after transmit completes
8237 * @q_vector: pointer to q_vector containing needed info
8238 * @napi_budget: Used to determine if we are in netpoll
8240 * returns true if ring is completely cleaned
8242 static bool igb_clean_tx_irq(struct igb_q_vector *q_vector, int napi_budget)
8244 struct igb_adapter *adapter = q_vector->adapter;
8245 struct igb_ring *tx_ring = q_vector->tx.ring;
8246 struct igb_tx_buffer *tx_buffer;
8247 union e1000_adv_tx_desc *tx_desc;
8248 unsigned int total_bytes = 0, total_packets = 0;
8249 unsigned int budget = q_vector->tx.work_limit;
8250 unsigned int i = tx_ring->next_to_clean;
8252 if (test_bit(__IGB_DOWN, &adapter->state))
8255 tx_buffer = &tx_ring->tx_buffer_info[i];
8256 tx_desc = IGB_TX_DESC(tx_ring, i);
8257 i -= tx_ring->count;
8260 union e1000_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
8262 /* if next_to_watch is not set then there is no work pending */
8266 /* prevent any other reads prior to eop_desc */
8269 /* if DD is not set pending work has not been completed */
8270 if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)))
8273 /* clear next_to_watch to prevent false hangs */
8274 tx_buffer->next_to_watch = NULL;
8276 /* update the statistics for this packet */
8277 total_bytes += tx_buffer->bytecount;
8278 total_packets += tx_buffer->gso_segs;
8281 if (tx_buffer->type == IGB_TYPE_SKB)
8282 napi_consume_skb(tx_buffer->skb, napi_budget);
8284 xdp_return_frame(tx_buffer->xdpf);
8286 /* unmap skb header data */
8287 dma_unmap_single(tx_ring->dev,
8288 dma_unmap_addr(tx_buffer, dma),
8289 dma_unmap_len(tx_buffer, len),
8292 /* clear tx_buffer data */
8293 dma_unmap_len_set(tx_buffer, len, 0);
8295 /* clear last DMA location and unmap remaining buffers */
8296 while (tx_desc != eop_desc) {
8301 i -= tx_ring->count;
8302 tx_buffer = tx_ring->tx_buffer_info;
8303 tx_desc = IGB_TX_DESC(tx_ring, 0);
8306 /* unmap any remaining paged data */
8307 if (dma_unmap_len(tx_buffer, len)) {
8308 dma_unmap_page(tx_ring->dev,
8309 dma_unmap_addr(tx_buffer, dma),
8310 dma_unmap_len(tx_buffer, len),
8312 dma_unmap_len_set(tx_buffer, len, 0);
8316 /* move us one more past the eop_desc for start of next pkt */
8321 i -= tx_ring->count;
8322 tx_buffer = tx_ring->tx_buffer_info;
8323 tx_desc = IGB_TX_DESC(tx_ring, 0);
8326 /* issue prefetch for next Tx descriptor */
8329 /* update budget accounting */
8331 } while (likely(budget));
8333 netdev_tx_completed_queue(txring_txq(tx_ring),
8334 total_packets, total_bytes);
8335 i += tx_ring->count;
8336 tx_ring->next_to_clean = i;
8337 u64_stats_update_begin(&tx_ring->tx_syncp);
8338 tx_ring->tx_stats.bytes += total_bytes;
8339 tx_ring->tx_stats.packets += total_packets;
8340 u64_stats_update_end(&tx_ring->tx_syncp);
8341 q_vector->tx.total_bytes += total_bytes;
8342 q_vector->tx.total_packets += total_packets;
8344 if (test_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags)) {
8345 struct e1000_hw *hw = &adapter->hw;
8347 /* Detect a transmit hang in hardware, this serializes the
8348 * check with the clearing of time_stamp and movement of i
8350 clear_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
8351 if (tx_buffer->next_to_watch &&
8352 time_after(jiffies, tx_buffer->time_stamp +
8353 (adapter->tx_timeout_factor * HZ)) &&
8354 !(rd32(E1000_STATUS) & E1000_STATUS_TXOFF)) {
8356 /* detected Tx unit hang */
8357 dev_err(tx_ring->dev,
8358 "Detected Tx Unit Hang\n"
8362 " next_to_use <%x>\n"
8363 " next_to_clean <%x>\n"
8364 "buffer_info[next_to_clean]\n"
8365 " time_stamp <%lx>\n"
8366 " next_to_watch <%p>\n"
8368 " desc.status <%x>\n",
8369 tx_ring->queue_index,
8370 rd32(E1000_TDH(tx_ring->reg_idx)),
8371 readl(tx_ring->tail),
8372 tx_ring->next_to_use,
8373 tx_ring->next_to_clean,
8374 tx_buffer->time_stamp,
8375 tx_buffer->next_to_watch,
8377 tx_buffer->next_to_watch->wb.status);
8378 netif_stop_subqueue(tx_ring->netdev,
8379 tx_ring->queue_index);
8381 /* we are about to reset, no point in enabling stuff */
8386 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
8387 if (unlikely(total_packets &&
8388 netif_carrier_ok(tx_ring->netdev) &&
8389 igb_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD)) {
8390 /* Make sure that anybody stopping the queue after this
8391 * sees the new next_to_clean.
8394 if (__netif_subqueue_stopped(tx_ring->netdev,
8395 tx_ring->queue_index) &&
8396 !(test_bit(__IGB_DOWN, &adapter->state))) {
8397 netif_wake_subqueue(tx_ring->netdev,
8398 tx_ring->queue_index);
8400 u64_stats_update_begin(&tx_ring->tx_syncp);
8401 tx_ring->tx_stats.restart_queue++;
8402 u64_stats_update_end(&tx_ring->tx_syncp);
8410 * igb_reuse_rx_page - page flip buffer and store it back on the ring
8411 * @rx_ring: rx descriptor ring to store buffers on
8412 * @old_buff: donor buffer to have page reused
8414 * Synchronizes page for reuse by the adapter
8416 static void igb_reuse_rx_page(struct igb_ring *rx_ring,
8417 struct igb_rx_buffer *old_buff)
8419 struct igb_rx_buffer *new_buff;
8420 u16 nta = rx_ring->next_to_alloc;
8422 new_buff = &rx_ring->rx_buffer_info[nta];
8424 /* update, and store next to alloc */
8426 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
8428 /* Transfer page from old buffer to new buffer.
8429 * Move each member individually to avoid possible store
8430 * forwarding stalls.
8432 new_buff->dma = old_buff->dma;
8433 new_buff->page = old_buff->page;
8434 new_buff->page_offset = old_buff->page_offset;
8435 new_buff->pagecnt_bias = old_buff->pagecnt_bias;
8438 static bool igb_can_reuse_rx_page(struct igb_rx_buffer *rx_buffer,
8441 unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
8442 struct page *page = rx_buffer->page;
8444 /* avoid re-using remote and pfmemalloc pages */
8445 if (!dev_page_is_reusable(page))
8448 #if (PAGE_SIZE < 8192)
8449 /* if we are only owner of page we can reuse it */
8450 if (unlikely((rx_buf_pgcnt - pagecnt_bias) > 1))
8453 #define IGB_LAST_OFFSET \
8454 (SKB_WITH_OVERHEAD(PAGE_SIZE) - IGB_RXBUFFER_2048)
8456 if (rx_buffer->page_offset > IGB_LAST_OFFSET)
8460 /* If we have drained the page fragment pool we need to update
8461 * the pagecnt_bias and page count so that we fully restock the
8462 * number of references the driver holds.
8464 if (unlikely(pagecnt_bias == 1)) {
8465 page_ref_add(page, USHRT_MAX - 1);
8466 rx_buffer->pagecnt_bias = USHRT_MAX;
8473 * igb_add_rx_frag - Add contents of Rx buffer to sk_buff
8474 * @rx_ring: rx descriptor ring to transact packets on
8475 * @rx_buffer: buffer containing page to add
8476 * @skb: sk_buff to place the data into
8477 * @size: size of buffer to be added
8479 * This function will add the data contained in rx_buffer->page to the skb.
8481 static void igb_add_rx_frag(struct igb_ring *rx_ring,
8482 struct igb_rx_buffer *rx_buffer,
8483 struct sk_buff *skb,
8486 #if (PAGE_SIZE < 8192)
8487 unsigned int truesize = igb_rx_pg_size(rx_ring) / 2;
8489 unsigned int truesize = ring_uses_build_skb(rx_ring) ?
8490 SKB_DATA_ALIGN(IGB_SKB_PAD + size) :
8491 SKB_DATA_ALIGN(size);
8493 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
8494 rx_buffer->page_offset, size, truesize);
8495 #if (PAGE_SIZE < 8192)
8496 rx_buffer->page_offset ^= truesize;
8498 rx_buffer->page_offset += truesize;
8502 static struct sk_buff *igb_construct_skb(struct igb_ring *rx_ring,
8503 struct igb_rx_buffer *rx_buffer,
8504 struct xdp_buff *xdp,
8507 #if (PAGE_SIZE < 8192)
8508 unsigned int truesize = igb_rx_pg_size(rx_ring) / 2;
8510 unsigned int truesize = SKB_DATA_ALIGN(xdp->data_end -
8511 xdp->data_hard_start);
8513 unsigned int size = xdp->data_end - xdp->data;
8514 unsigned int headlen;
8515 struct sk_buff *skb;
8517 /* prefetch first cache line of first page */
8518 net_prefetch(xdp->data);
8520 /* allocate a skb to store the frags */
8521 skb = napi_alloc_skb(&rx_ring->q_vector->napi, IGB_RX_HDR_LEN);
8526 skb_hwtstamps(skb)->hwtstamp = timestamp;
8528 /* Determine available headroom for copy */
8530 if (headlen > IGB_RX_HDR_LEN)
8531 headlen = eth_get_headlen(skb->dev, xdp->data, IGB_RX_HDR_LEN);
8533 /* align pull length to size of long to optimize memcpy performance */
8534 memcpy(__skb_put(skb, headlen), xdp->data, ALIGN(headlen, sizeof(long)));
8536 /* update all of the pointers */
8539 skb_add_rx_frag(skb, 0, rx_buffer->page,
8540 (xdp->data + headlen) - page_address(rx_buffer->page),
8542 #if (PAGE_SIZE < 8192)
8543 rx_buffer->page_offset ^= truesize;
8545 rx_buffer->page_offset += truesize;
8548 rx_buffer->pagecnt_bias++;
8554 static struct sk_buff *igb_build_skb(struct igb_ring *rx_ring,
8555 struct igb_rx_buffer *rx_buffer,
8556 struct xdp_buff *xdp,
8559 #if (PAGE_SIZE < 8192)
8560 unsigned int truesize = igb_rx_pg_size(rx_ring) / 2;
8562 unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
8563 SKB_DATA_ALIGN(xdp->data_end -
8564 xdp->data_hard_start);
8566 unsigned int metasize = xdp->data - xdp->data_meta;
8567 struct sk_buff *skb;
8569 /* prefetch first cache line of first page */
8570 net_prefetch(xdp->data_meta);
8572 /* build an skb around the page buffer */
8573 skb = napi_build_skb(xdp->data_hard_start, truesize);
8577 /* update pointers within the skb to store the data */
8578 skb_reserve(skb, xdp->data - xdp->data_hard_start);
8579 __skb_put(skb, xdp->data_end - xdp->data);
8582 skb_metadata_set(skb, metasize);
8585 skb_hwtstamps(skb)->hwtstamp = timestamp;
8587 /* update buffer offset */
8588 #if (PAGE_SIZE < 8192)
8589 rx_buffer->page_offset ^= truesize;
8591 rx_buffer->page_offset += truesize;
8597 static struct sk_buff *igb_run_xdp(struct igb_adapter *adapter,
8598 struct igb_ring *rx_ring,
8599 struct xdp_buff *xdp)
8601 int err, result = IGB_XDP_PASS;
8602 struct bpf_prog *xdp_prog;
8605 xdp_prog = READ_ONCE(rx_ring->xdp_prog);
8610 prefetchw(xdp->data_hard_start); /* xdp_frame write */
8612 act = bpf_prog_run_xdp(xdp_prog, xdp);
8617 result = igb_xdp_xmit_back(adapter, xdp);
8618 if (result == IGB_XDP_CONSUMED)
8622 err = xdp_do_redirect(adapter->netdev, xdp, xdp_prog);
8625 result = IGB_XDP_REDIR;
8628 bpf_warn_invalid_xdp_action(adapter->netdev, xdp_prog, act);
8632 trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
8635 result = IGB_XDP_CONSUMED;
8639 return ERR_PTR(-result);
8642 static unsigned int igb_rx_frame_truesize(struct igb_ring *rx_ring,
8645 unsigned int truesize;
8647 #if (PAGE_SIZE < 8192)
8648 truesize = igb_rx_pg_size(rx_ring) / 2; /* Must be power-of-2 */
8650 truesize = ring_uses_build_skb(rx_ring) ?
8651 SKB_DATA_ALIGN(IGB_SKB_PAD + size) +
8652 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) :
8653 SKB_DATA_ALIGN(size);
8658 static void igb_rx_buffer_flip(struct igb_ring *rx_ring,
8659 struct igb_rx_buffer *rx_buffer,
8662 unsigned int truesize = igb_rx_frame_truesize(rx_ring, size);
8663 #if (PAGE_SIZE < 8192)
8664 rx_buffer->page_offset ^= truesize;
8666 rx_buffer->page_offset += truesize;
8670 static inline void igb_rx_checksum(struct igb_ring *ring,
8671 union e1000_adv_rx_desc *rx_desc,
8672 struct sk_buff *skb)
8674 skb_checksum_none_assert(skb);
8676 /* Ignore Checksum bit is set */
8677 if (igb_test_staterr(rx_desc, E1000_RXD_STAT_IXSM))
8680 /* Rx checksum disabled via ethtool */
8681 if (!(ring->netdev->features & NETIF_F_RXCSUM))
8684 /* TCP/UDP checksum error bit is set */
8685 if (igb_test_staterr(rx_desc,
8686 E1000_RXDEXT_STATERR_TCPE |
8687 E1000_RXDEXT_STATERR_IPE)) {
8688 /* work around errata with sctp packets where the TCPE aka
8689 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
8690 * packets, (aka let the stack check the crc32c)
8692 if (!((skb->len == 60) &&
8693 test_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags))) {
8694 u64_stats_update_begin(&ring->rx_syncp);
8695 ring->rx_stats.csum_err++;
8696 u64_stats_update_end(&ring->rx_syncp);
8698 /* let the stack verify checksum errors */
8701 /* It must be a TCP or UDP packet with a valid checksum */
8702 if (igb_test_staterr(rx_desc, E1000_RXD_STAT_TCPCS |
8703 E1000_RXD_STAT_UDPCS))
8704 skb->ip_summed = CHECKSUM_UNNECESSARY;
8706 dev_dbg(ring->dev, "cksum success: bits %08X\n",
8707 le32_to_cpu(rx_desc->wb.upper.status_error));
8710 static inline void igb_rx_hash(struct igb_ring *ring,
8711 union e1000_adv_rx_desc *rx_desc,
8712 struct sk_buff *skb)
8714 if (ring->netdev->features & NETIF_F_RXHASH)
8716 le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
8721 * igb_is_non_eop - process handling of non-EOP buffers
8722 * @rx_ring: Rx ring being processed
8723 * @rx_desc: Rx descriptor for current buffer
8725 * This function updates next to clean. If the buffer is an EOP buffer
8726 * this function exits returning false, otherwise it will place the
8727 * sk_buff in the next buffer to be chained and return true indicating
8728 * that this is in fact a non-EOP buffer.
8730 static bool igb_is_non_eop(struct igb_ring *rx_ring,
8731 union e1000_adv_rx_desc *rx_desc)
8733 u32 ntc = rx_ring->next_to_clean + 1;
8735 /* fetch, update, and store next to clean */
8736 ntc = (ntc < rx_ring->count) ? ntc : 0;
8737 rx_ring->next_to_clean = ntc;
8739 prefetch(IGB_RX_DESC(rx_ring, ntc));
8741 if (likely(igb_test_staterr(rx_desc, E1000_RXD_STAT_EOP)))
8748 * igb_cleanup_headers - Correct corrupted or empty headers
8749 * @rx_ring: rx descriptor ring packet is being transacted on
8750 * @rx_desc: pointer to the EOP Rx descriptor
8751 * @skb: pointer to current skb being fixed
8753 * Address the case where we are pulling data in on pages only
8754 * and as such no data is present in the skb header.
8756 * In addition if skb is not at least 60 bytes we need to pad it so that
8757 * it is large enough to qualify as a valid Ethernet frame.
8759 * Returns true if an error was encountered and skb was freed.
8761 static bool igb_cleanup_headers(struct igb_ring *rx_ring,
8762 union e1000_adv_rx_desc *rx_desc,
8763 struct sk_buff *skb)
8765 /* XDP packets use error pointer so abort at this point */
8769 if (unlikely((igb_test_staterr(rx_desc,
8770 E1000_RXDEXT_ERR_FRAME_ERR_MASK)))) {
8771 struct net_device *netdev = rx_ring->netdev;
8772 if (!(netdev->features & NETIF_F_RXALL)) {
8773 dev_kfree_skb_any(skb);
8778 /* if eth_skb_pad returns an error the skb was freed */
8779 if (eth_skb_pad(skb))
8786 * igb_process_skb_fields - Populate skb header fields from Rx descriptor
8787 * @rx_ring: rx descriptor ring packet is being transacted on
8788 * @rx_desc: pointer to the EOP Rx descriptor
8789 * @skb: pointer to current skb being populated
8791 * This function checks the ring, descriptor, and packet information in
8792 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
8793 * other fields within the skb.
8795 static void igb_process_skb_fields(struct igb_ring *rx_ring,
8796 union e1000_adv_rx_desc *rx_desc,
8797 struct sk_buff *skb)
8799 struct net_device *dev = rx_ring->netdev;
8801 igb_rx_hash(rx_ring, rx_desc, skb);
8803 igb_rx_checksum(rx_ring, rx_desc, skb);
8805 if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TS) &&
8806 !igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP))
8807 igb_ptp_rx_rgtstamp(rx_ring->q_vector, skb);
8809 if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
8810 igb_test_staterr(rx_desc, E1000_RXD_STAT_VP)) {
8813 if (igb_test_staterr(rx_desc, E1000_RXDEXT_STATERR_LB) &&
8814 test_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &rx_ring->flags))
8815 vid = be16_to_cpu((__force __be16)rx_desc->wb.upper.vlan);
8817 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
8819 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
8822 skb_record_rx_queue(skb, rx_ring->queue_index);
8824 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
8827 static unsigned int igb_rx_offset(struct igb_ring *rx_ring)
8829 return ring_uses_build_skb(rx_ring) ? IGB_SKB_PAD : 0;
8832 static struct igb_rx_buffer *igb_get_rx_buffer(struct igb_ring *rx_ring,
8833 const unsigned int size, int *rx_buf_pgcnt)
8835 struct igb_rx_buffer *rx_buffer;
8837 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
8839 #if (PAGE_SIZE < 8192)
8840 page_count(rx_buffer->page);
8844 prefetchw(rx_buffer->page);
8846 /* we are reusing so sync this buffer for CPU use */
8847 dma_sync_single_range_for_cpu(rx_ring->dev,
8849 rx_buffer->page_offset,
8853 rx_buffer->pagecnt_bias--;
8858 static void igb_put_rx_buffer(struct igb_ring *rx_ring,
8859 struct igb_rx_buffer *rx_buffer, int rx_buf_pgcnt)
8861 if (igb_can_reuse_rx_page(rx_buffer, rx_buf_pgcnt)) {
8862 /* hand second half of page back to the ring */
8863 igb_reuse_rx_page(rx_ring, rx_buffer);
8865 /* We are not reusing the buffer so unmap it and free
8866 * any references we are holding to it
8868 dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
8869 igb_rx_pg_size(rx_ring), DMA_FROM_DEVICE,
8871 __page_frag_cache_drain(rx_buffer->page,
8872 rx_buffer->pagecnt_bias);
8875 /* clear contents of rx_buffer */
8876 rx_buffer->page = NULL;
8879 static int igb_clean_rx_irq(struct igb_q_vector *q_vector, const int budget)
8881 struct igb_adapter *adapter = q_vector->adapter;
8882 struct igb_ring *rx_ring = q_vector->rx.ring;
8883 struct sk_buff *skb = rx_ring->skb;
8884 unsigned int total_bytes = 0, total_packets = 0;
8885 u16 cleaned_count = igb_desc_unused(rx_ring);
8886 unsigned int xdp_xmit = 0;
8887 struct xdp_buff xdp;
8891 /* Frame size depend on rx_ring setup when PAGE_SIZE=4K */
8892 #if (PAGE_SIZE < 8192)
8893 frame_sz = igb_rx_frame_truesize(rx_ring, 0);
8895 xdp_init_buff(&xdp, frame_sz, &rx_ring->xdp_rxq);
8897 while (likely(total_packets < budget)) {
8898 union e1000_adv_rx_desc *rx_desc;
8899 struct igb_rx_buffer *rx_buffer;
8900 ktime_t timestamp = 0;
8905 /* return some buffers to hardware, one at a time is too slow */
8906 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
8907 igb_alloc_rx_buffers(rx_ring, cleaned_count);
8911 rx_desc = IGB_RX_DESC(rx_ring, rx_ring->next_to_clean);
8912 size = le16_to_cpu(rx_desc->wb.upper.length);
8916 /* This memory barrier is needed to keep us from reading
8917 * any other fields out of the rx_desc until we know the
8918 * descriptor has been written back
8922 rx_buffer = igb_get_rx_buffer(rx_ring, size, &rx_buf_pgcnt);
8923 pktbuf = page_address(rx_buffer->page) + rx_buffer->page_offset;
8925 /* pull rx packet timestamp if available and valid */
8926 if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
8929 ts_hdr_len = igb_ptp_rx_pktstamp(rx_ring->q_vector,
8930 pktbuf, ×tamp);
8932 pkt_offset += ts_hdr_len;
8936 /* retrieve a buffer from the ring */
8938 unsigned char *hard_start = pktbuf - igb_rx_offset(rx_ring);
8939 unsigned int offset = pkt_offset + igb_rx_offset(rx_ring);
8941 xdp_prepare_buff(&xdp, hard_start, offset, size, true);
8942 xdp_buff_clear_frags_flag(&xdp);
8943 #if (PAGE_SIZE > 4096)
8944 /* At larger PAGE_SIZE, frame_sz depend on len size */
8945 xdp.frame_sz = igb_rx_frame_truesize(rx_ring, size);
8947 skb = igb_run_xdp(adapter, rx_ring, &xdp);
8951 unsigned int xdp_res = -PTR_ERR(skb);
8953 if (xdp_res & (IGB_XDP_TX | IGB_XDP_REDIR)) {
8954 xdp_xmit |= xdp_res;
8955 igb_rx_buffer_flip(rx_ring, rx_buffer, size);
8957 rx_buffer->pagecnt_bias++;
8960 total_bytes += size;
8962 igb_add_rx_frag(rx_ring, rx_buffer, skb, size);
8963 else if (ring_uses_build_skb(rx_ring))
8964 skb = igb_build_skb(rx_ring, rx_buffer, &xdp,
8967 skb = igb_construct_skb(rx_ring, rx_buffer,
8970 /* exit if we failed to retrieve a buffer */
8972 rx_ring->rx_stats.alloc_failed++;
8973 rx_buffer->pagecnt_bias++;
8977 igb_put_rx_buffer(rx_ring, rx_buffer, rx_buf_pgcnt);
8980 /* fetch next buffer in frame if non-eop */
8981 if (igb_is_non_eop(rx_ring, rx_desc))
8984 /* verify the packet layout is correct */
8985 if (igb_cleanup_headers(rx_ring, rx_desc, skb)) {
8990 /* probably a little skewed due to removing CRC */
8991 total_bytes += skb->len;
8993 /* populate checksum, timestamp, VLAN, and protocol */
8994 igb_process_skb_fields(rx_ring, rx_desc, skb);
8996 napi_gro_receive(&q_vector->napi, skb);
8998 /* reset skb pointer */
9001 /* update budget accounting */
9005 /* place incomplete frames back on ring for completion */
9008 if (xdp_xmit & IGB_XDP_REDIR)
9011 if (xdp_xmit & IGB_XDP_TX) {
9012 struct igb_ring *tx_ring = igb_xdp_tx_queue_mapping(adapter);
9014 igb_xdp_ring_update_tail(tx_ring);
9017 u64_stats_update_begin(&rx_ring->rx_syncp);
9018 rx_ring->rx_stats.packets += total_packets;
9019 rx_ring->rx_stats.bytes += total_bytes;
9020 u64_stats_update_end(&rx_ring->rx_syncp);
9021 q_vector->rx.total_packets += total_packets;
9022 q_vector->rx.total_bytes += total_bytes;
9025 igb_alloc_rx_buffers(rx_ring, cleaned_count);
9027 return total_packets;
9030 static bool igb_alloc_mapped_page(struct igb_ring *rx_ring,
9031 struct igb_rx_buffer *bi)
9033 struct page *page = bi->page;
9036 /* since we are recycling buffers we should seldom need to alloc */
9040 /* alloc new page for storage */
9041 page = dev_alloc_pages(igb_rx_pg_order(rx_ring));
9042 if (unlikely(!page)) {
9043 rx_ring->rx_stats.alloc_failed++;
9047 /* map page for use */
9048 dma = dma_map_page_attrs(rx_ring->dev, page, 0,
9049 igb_rx_pg_size(rx_ring),
9053 /* if mapping failed free memory back to system since
9054 * there isn't much point in holding memory we can't use
9056 if (dma_mapping_error(rx_ring->dev, dma)) {
9057 __free_pages(page, igb_rx_pg_order(rx_ring));
9059 rx_ring->rx_stats.alloc_failed++;
9065 bi->page_offset = igb_rx_offset(rx_ring);
9066 page_ref_add(page, USHRT_MAX - 1);
9067 bi->pagecnt_bias = USHRT_MAX;
9073 * igb_alloc_rx_buffers - Replace used receive buffers
9074 * @rx_ring: rx descriptor ring to allocate new receive buffers
9075 * @cleaned_count: count of buffers to allocate
9077 void igb_alloc_rx_buffers(struct igb_ring *rx_ring, u16 cleaned_count)
9079 union e1000_adv_rx_desc *rx_desc;
9080 struct igb_rx_buffer *bi;
9081 u16 i = rx_ring->next_to_use;
9088 rx_desc = IGB_RX_DESC(rx_ring, i);
9089 bi = &rx_ring->rx_buffer_info[i];
9090 i -= rx_ring->count;
9092 bufsz = igb_rx_bufsz(rx_ring);
9095 if (!igb_alloc_mapped_page(rx_ring, bi))
9098 /* sync the buffer for use by the device */
9099 dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
9100 bi->page_offset, bufsz,
9103 /* Refresh the desc even if buffer_addrs didn't change
9104 * because each write-back erases this info.
9106 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
9112 rx_desc = IGB_RX_DESC(rx_ring, 0);
9113 bi = rx_ring->rx_buffer_info;
9114 i -= rx_ring->count;
9117 /* clear the length for the next_to_use descriptor */
9118 rx_desc->wb.upper.length = 0;
9121 } while (cleaned_count);
9123 i += rx_ring->count;
9125 if (rx_ring->next_to_use != i) {
9126 /* record the next descriptor to use */
9127 rx_ring->next_to_use = i;
9129 /* update next to alloc since we have filled the ring */
9130 rx_ring->next_to_alloc = i;
9132 /* Force memory writes to complete before letting h/w
9133 * know there are new descriptors to fetch. (Only
9134 * applicable for weak-ordered memory model archs,
9138 writel(i, rx_ring->tail);
9144 * @netdev: pointer to netdev struct
9145 * @ifr: interface structure
9146 * @cmd: ioctl command to execute
9148 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
9150 struct igb_adapter *adapter = netdev_priv(netdev);
9151 struct mii_ioctl_data *data = if_mii(ifr);
9153 if (adapter->hw.phy.media_type != e1000_media_type_copper)
9158 data->phy_id = adapter->hw.phy.addr;
9161 if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
9174 * @netdev: pointer to netdev struct
9175 * @ifr: interface structure
9176 * @cmd: ioctl command to execute
9178 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
9184 return igb_mii_ioctl(netdev, ifr, cmd);
9186 return igb_ptp_get_ts_config(netdev, ifr);
9188 return igb_ptp_set_ts_config(netdev, ifr);
9194 void igb_read_pci_cfg(struct e1000_hw *hw, u32 reg, u16 *value)
9196 struct igb_adapter *adapter = hw->back;
9198 pci_read_config_word(adapter->pdev, reg, value);
9201 void igb_write_pci_cfg(struct e1000_hw *hw, u32 reg, u16 *value)
9203 struct igb_adapter *adapter = hw->back;
9205 pci_write_config_word(adapter->pdev, reg, *value);
9208 s32 igb_read_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
9210 struct igb_adapter *adapter = hw->back;
9212 if (pcie_capability_read_word(adapter->pdev, reg, value))
9213 return -E1000_ERR_CONFIG;
9218 s32 igb_write_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
9220 struct igb_adapter *adapter = hw->back;
9222 if (pcie_capability_write_word(adapter->pdev, reg, *value))
9223 return -E1000_ERR_CONFIG;
9228 static void igb_vlan_mode(struct net_device *netdev, netdev_features_t features)
9230 struct igb_adapter *adapter = netdev_priv(netdev);
9231 struct e1000_hw *hw = &adapter->hw;
9233 bool enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
9236 /* enable VLAN tag insert/strip */
9237 ctrl = rd32(E1000_CTRL);
9238 ctrl |= E1000_CTRL_VME;
9239 wr32(E1000_CTRL, ctrl);
9241 /* Disable CFI check */
9242 rctl = rd32(E1000_RCTL);
9243 rctl &= ~E1000_RCTL_CFIEN;
9244 wr32(E1000_RCTL, rctl);
9246 /* disable VLAN tag insert/strip */
9247 ctrl = rd32(E1000_CTRL);
9248 ctrl &= ~E1000_CTRL_VME;
9249 wr32(E1000_CTRL, ctrl);
9252 igb_set_vf_vlan_strip(adapter, adapter->vfs_allocated_count, enable);
9255 static int igb_vlan_rx_add_vid(struct net_device *netdev,
9256 __be16 proto, u16 vid)
9258 struct igb_adapter *adapter = netdev_priv(netdev);
9259 struct e1000_hw *hw = &adapter->hw;
9260 int pf_id = adapter->vfs_allocated_count;
9262 /* add the filter since PF can receive vlans w/o entry in vlvf */
9263 if (!vid || !(adapter->flags & IGB_FLAG_VLAN_PROMISC))
9264 igb_vfta_set(hw, vid, pf_id, true, !!vid);
9266 set_bit(vid, adapter->active_vlans);
9271 static int igb_vlan_rx_kill_vid(struct net_device *netdev,
9272 __be16 proto, u16 vid)
9274 struct igb_adapter *adapter = netdev_priv(netdev);
9275 int pf_id = adapter->vfs_allocated_count;
9276 struct e1000_hw *hw = &adapter->hw;
9278 /* remove VID from filter table */
9279 if (vid && !(adapter->flags & IGB_FLAG_VLAN_PROMISC))
9280 igb_vfta_set(hw, vid, pf_id, false, true);
9282 clear_bit(vid, adapter->active_vlans);
9287 static void igb_restore_vlan(struct igb_adapter *adapter)
9291 igb_vlan_mode(adapter->netdev, adapter->netdev->features);
9292 igb_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
9294 for_each_set_bit_from(vid, adapter->active_vlans, VLAN_N_VID)
9295 igb_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
9298 int igb_set_spd_dplx(struct igb_adapter *adapter, u32 spd, u8 dplx)
9300 struct pci_dev *pdev = adapter->pdev;
9301 struct e1000_mac_info *mac = &adapter->hw.mac;
9305 /* Make sure dplx is at most 1 bit and lsb of speed is not set
9306 * for the switch() below to work
9308 if ((spd & 1) || (dplx & ~1))
9311 /* Fiber NIC's only allow 1000 gbps Full duplex
9312 * and 100Mbps Full duplex for 100baseFx sfp
9314 if (adapter->hw.phy.media_type == e1000_media_type_internal_serdes) {
9315 switch (spd + dplx) {
9316 case SPEED_10 + DUPLEX_HALF:
9317 case SPEED_10 + DUPLEX_FULL:
9318 case SPEED_100 + DUPLEX_HALF:
9325 switch (spd + dplx) {
9326 case SPEED_10 + DUPLEX_HALF:
9327 mac->forced_speed_duplex = ADVERTISE_10_HALF;
9329 case SPEED_10 + DUPLEX_FULL:
9330 mac->forced_speed_duplex = ADVERTISE_10_FULL;
9332 case SPEED_100 + DUPLEX_HALF:
9333 mac->forced_speed_duplex = ADVERTISE_100_HALF;
9335 case SPEED_100 + DUPLEX_FULL:
9336 mac->forced_speed_duplex = ADVERTISE_100_FULL;
9338 case SPEED_1000 + DUPLEX_FULL:
9340 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
9342 case SPEED_1000 + DUPLEX_HALF: /* not supported */
9347 /* clear MDI, MDI(-X) override is only allowed when autoneg enabled */
9348 adapter->hw.phy.mdix = AUTO_ALL_MODES;
9353 dev_err(&pdev->dev, "Unsupported Speed/Duplex configuration\n");
9357 static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake,
9360 struct net_device *netdev = pci_get_drvdata(pdev);
9361 struct igb_adapter *adapter = netdev_priv(netdev);
9362 struct e1000_hw *hw = &adapter->hw;
9363 u32 ctrl, rctl, status;
9364 u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
9368 netif_device_detach(netdev);
9370 if (netif_running(netdev))
9371 __igb_close(netdev, true);
9373 igb_ptp_suspend(adapter);
9375 igb_clear_interrupt_scheme(adapter);
9378 status = rd32(E1000_STATUS);
9379 if (status & E1000_STATUS_LU)
9380 wufc &= ~E1000_WUFC_LNKC;
9383 igb_setup_rctl(adapter);
9384 igb_set_rx_mode(netdev);
9386 /* turn on all-multi mode if wake on multicast is enabled */
9387 if (wufc & E1000_WUFC_MC) {
9388 rctl = rd32(E1000_RCTL);
9389 rctl |= E1000_RCTL_MPE;
9390 wr32(E1000_RCTL, rctl);
9393 ctrl = rd32(E1000_CTRL);
9394 ctrl |= E1000_CTRL_ADVD3WUC;
9395 wr32(E1000_CTRL, ctrl);
9397 /* Allow time for pending master requests to run */
9398 igb_disable_pcie_master(hw);
9400 wr32(E1000_WUC, E1000_WUC_PME_EN);
9401 wr32(E1000_WUFC, wufc);
9404 wr32(E1000_WUFC, 0);
9407 wake = wufc || adapter->en_mng_pt;
9409 igb_power_down_link(adapter);
9411 igb_power_up_link(adapter);
9414 *enable_wake = wake;
9416 /* Release control of h/w to f/w. If f/w is AMT enabled, this
9417 * would have already happened in close and is redundant.
9419 igb_release_hw_control(adapter);
9421 pci_disable_device(pdev);
9426 static void igb_deliver_wake_packet(struct net_device *netdev)
9428 struct igb_adapter *adapter = netdev_priv(netdev);
9429 struct e1000_hw *hw = &adapter->hw;
9430 struct sk_buff *skb;
9433 wupl = rd32(E1000_WUPL) & E1000_WUPL_MASK;
9435 /* WUPM stores only the first 128 bytes of the wake packet.
9436 * Read the packet only if we have the whole thing.
9438 if ((wupl == 0) || (wupl > E1000_WUPM_BYTES))
9441 skb = netdev_alloc_skb_ip_align(netdev, E1000_WUPM_BYTES);
9447 /* Ensure reads are 32-bit aligned */
9448 wupl = roundup(wupl, 4);
9450 memcpy_fromio(skb->data, hw->hw_addr + E1000_WUPM_REG(0), wupl);
9452 skb->protocol = eth_type_trans(skb, netdev);
9456 static int __maybe_unused igb_suspend(struct device *dev)
9458 return __igb_shutdown(to_pci_dev(dev), NULL, 0);
9461 static int __maybe_unused __igb_resume(struct device *dev, bool rpm)
9463 struct pci_dev *pdev = to_pci_dev(dev);
9464 struct net_device *netdev = pci_get_drvdata(pdev);
9465 struct igb_adapter *adapter = netdev_priv(netdev);
9466 struct e1000_hw *hw = &adapter->hw;
9469 pci_set_power_state(pdev, PCI_D0);
9470 pci_restore_state(pdev);
9471 pci_save_state(pdev);
9473 if (!pci_device_is_present(pdev))
9475 err = pci_enable_device_mem(pdev);
9478 "igb: Cannot enable PCI device from suspend\n");
9481 pci_set_master(pdev);
9483 pci_enable_wake(pdev, PCI_D3hot, 0);
9484 pci_enable_wake(pdev, PCI_D3cold, 0);
9486 if (igb_init_interrupt_scheme(adapter, true)) {
9487 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
9493 /* let the f/w know that the h/w is now under the control of the
9496 igb_get_hw_control(adapter);
9498 val = rd32(E1000_WUS);
9499 if (val & WAKE_PKT_WUS)
9500 igb_deliver_wake_packet(netdev);
9502 wr32(E1000_WUS, ~0);
9506 if (!err && netif_running(netdev))
9507 err = __igb_open(netdev, true);
9510 netif_device_attach(netdev);
9517 static int __maybe_unused igb_resume(struct device *dev)
9519 return __igb_resume(dev, false);
9522 static int __maybe_unused igb_runtime_idle(struct device *dev)
9524 struct net_device *netdev = dev_get_drvdata(dev);
9525 struct igb_adapter *adapter = netdev_priv(netdev);
9527 if (!igb_has_link(adapter))
9528 pm_schedule_suspend(dev, MSEC_PER_SEC * 5);
9533 static int __maybe_unused igb_runtime_suspend(struct device *dev)
9535 return __igb_shutdown(to_pci_dev(dev), NULL, 1);
9538 static int __maybe_unused igb_runtime_resume(struct device *dev)
9540 return __igb_resume(dev, true);
9543 static void igb_shutdown(struct pci_dev *pdev)
9547 __igb_shutdown(pdev, &wake, 0);
9549 if (system_state == SYSTEM_POWER_OFF) {
9550 pci_wake_from_d3(pdev, wake);
9551 pci_set_power_state(pdev, PCI_D3hot);
9555 static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs)
9557 #ifdef CONFIG_PCI_IOV
9561 return igb_disable_sriov(dev, true);
9563 err = igb_enable_sriov(dev, num_vfs, true);
9564 return err ? err : num_vfs;
9571 * igb_io_error_detected - called when PCI error is detected
9572 * @pdev: Pointer to PCI device
9573 * @state: The current pci connection state
9575 * This function is called after a PCI bus error affecting
9576 * this device has been detected.
9578 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
9579 pci_channel_state_t state)
9581 struct net_device *netdev = pci_get_drvdata(pdev);
9582 struct igb_adapter *adapter = netdev_priv(netdev);
9584 netif_device_detach(netdev);
9586 if (state == pci_channel_io_perm_failure)
9587 return PCI_ERS_RESULT_DISCONNECT;
9589 if (netif_running(netdev))
9591 pci_disable_device(pdev);
9593 /* Request a slot reset. */
9594 return PCI_ERS_RESULT_NEED_RESET;
9598 * igb_io_slot_reset - called after the pci bus has been reset.
9599 * @pdev: Pointer to PCI device
9601 * Restart the card from scratch, as if from a cold-boot. Implementation
9602 * resembles the first-half of the __igb_resume routine.
9604 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
9606 struct net_device *netdev = pci_get_drvdata(pdev);
9607 struct igb_adapter *adapter = netdev_priv(netdev);
9608 struct e1000_hw *hw = &adapter->hw;
9609 pci_ers_result_t result;
9611 if (pci_enable_device_mem(pdev)) {
9613 "Cannot re-enable PCI device after reset.\n");
9614 result = PCI_ERS_RESULT_DISCONNECT;
9616 pci_set_master(pdev);
9617 pci_restore_state(pdev);
9618 pci_save_state(pdev);
9620 pci_enable_wake(pdev, PCI_D3hot, 0);
9621 pci_enable_wake(pdev, PCI_D3cold, 0);
9623 /* In case of PCI error, adapter lose its HW address
9624 * so we should re-assign it here.
9626 hw->hw_addr = adapter->io_addr;
9629 wr32(E1000_WUS, ~0);
9630 result = PCI_ERS_RESULT_RECOVERED;
9637 * igb_io_resume - called when traffic can start flowing again.
9638 * @pdev: Pointer to PCI device
9640 * This callback is called when the error recovery driver tells us that
9641 * its OK to resume normal operation. Implementation resembles the
9642 * second-half of the __igb_resume routine.
9644 static void igb_io_resume(struct pci_dev *pdev)
9646 struct net_device *netdev = pci_get_drvdata(pdev);
9647 struct igb_adapter *adapter = netdev_priv(netdev);
9649 if (netif_running(netdev)) {
9650 if (igb_up(adapter)) {
9651 dev_err(&pdev->dev, "igb_up failed after reset\n");
9656 netif_device_attach(netdev);
9658 /* let the f/w know that the h/w is now under the control of the
9661 igb_get_hw_control(adapter);
9665 * igb_rar_set_index - Sync RAL[index] and RAH[index] registers with MAC table
9666 * @adapter: Pointer to adapter structure
9667 * @index: Index of the RAR entry which need to be synced with MAC table
9669 static void igb_rar_set_index(struct igb_adapter *adapter, u32 index)
9671 struct e1000_hw *hw = &adapter->hw;
9672 u32 rar_low, rar_high;
9673 u8 *addr = adapter->mac_table[index].addr;
9675 /* HW expects these to be in network order when they are plugged
9676 * into the registers which are little endian. In order to guarantee
9677 * that ordering we need to do an leXX_to_cpup here in order to be
9678 * ready for the byteswap that occurs with writel
9680 rar_low = le32_to_cpup((__le32 *)(addr));
9681 rar_high = le16_to_cpup((__le16 *)(addr + 4));
9683 /* Indicate to hardware the Address is Valid. */
9684 if (adapter->mac_table[index].state & IGB_MAC_STATE_IN_USE) {
9685 if (is_valid_ether_addr(addr))
9686 rar_high |= E1000_RAH_AV;
9688 if (adapter->mac_table[index].state & IGB_MAC_STATE_SRC_ADDR)
9689 rar_high |= E1000_RAH_ASEL_SRC_ADDR;
9691 switch (hw->mac.type) {
9694 if (adapter->mac_table[index].state &
9695 IGB_MAC_STATE_QUEUE_STEERING)
9696 rar_high |= E1000_RAH_QSEL_ENABLE;
9698 rar_high |= E1000_RAH_POOL_1 *
9699 adapter->mac_table[index].queue;
9702 rar_high |= E1000_RAH_POOL_1 <<
9703 adapter->mac_table[index].queue;
9708 wr32(E1000_RAL(index), rar_low);
9710 wr32(E1000_RAH(index), rar_high);
9714 static int igb_set_vf_mac(struct igb_adapter *adapter,
9715 int vf, unsigned char *mac_addr)
9717 struct e1000_hw *hw = &adapter->hw;
9718 /* VF MAC addresses start at end of receive addresses and moves
9719 * towards the first, as a result a collision should not be possible
9721 int rar_entry = hw->mac.rar_entry_count - (vf + 1);
9722 unsigned char *vf_mac_addr = adapter->vf_data[vf].vf_mac_addresses;
9724 ether_addr_copy(vf_mac_addr, mac_addr);
9725 ether_addr_copy(adapter->mac_table[rar_entry].addr, mac_addr);
9726 adapter->mac_table[rar_entry].queue = vf;
9727 adapter->mac_table[rar_entry].state |= IGB_MAC_STATE_IN_USE;
9728 igb_rar_set_index(adapter, rar_entry);
9733 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
9735 struct igb_adapter *adapter = netdev_priv(netdev);
9737 if (vf >= adapter->vfs_allocated_count)
9740 /* Setting the VF MAC to 0 reverts the IGB_VF_FLAG_PF_SET_MAC
9741 * flag and allows to overwrite the MAC via VF netdev. This
9742 * is necessary to allow libvirt a way to restore the original
9743 * MAC after unbinding vfio-pci and reloading igbvf after shutting
9746 if (is_zero_ether_addr(mac)) {
9747 adapter->vf_data[vf].flags &= ~IGB_VF_FLAG_PF_SET_MAC;
9748 dev_info(&adapter->pdev->dev,
9749 "remove administratively set MAC on VF %d\n",
9751 } else if (is_valid_ether_addr(mac)) {
9752 adapter->vf_data[vf].flags |= IGB_VF_FLAG_PF_SET_MAC;
9753 dev_info(&adapter->pdev->dev, "setting MAC %pM on VF %d\n",
9755 dev_info(&adapter->pdev->dev,
9756 "Reload the VF driver to make this change effective.");
9757 /* Generate additional warning if PF is down */
9758 if (test_bit(__IGB_DOWN, &adapter->state)) {
9759 dev_warn(&adapter->pdev->dev,
9760 "The VF MAC address has been set, but the PF device is not up.\n");
9761 dev_warn(&adapter->pdev->dev,
9762 "Bring the PF device up before attempting to use the VF device.\n");
9767 return igb_set_vf_mac(adapter, vf, mac);
9770 static int igb_link_mbps(int internal_link_speed)
9772 switch (internal_link_speed) {
9782 static void igb_set_vf_rate_limit(struct e1000_hw *hw, int vf, int tx_rate,
9789 /* Calculate the rate factor values to set */
9790 rf_int = link_speed / tx_rate;
9791 rf_dec = (link_speed - (rf_int * tx_rate));
9792 rf_dec = (rf_dec * BIT(E1000_RTTBCNRC_RF_INT_SHIFT)) /
9795 bcnrc_val = E1000_RTTBCNRC_RS_ENA;
9796 bcnrc_val |= ((rf_int << E1000_RTTBCNRC_RF_INT_SHIFT) &
9797 E1000_RTTBCNRC_RF_INT_MASK);
9798 bcnrc_val |= (rf_dec & E1000_RTTBCNRC_RF_DEC_MASK);
9803 wr32(E1000_RTTDQSEL, vf); /* vf X uses queue X */
9804 /* Set global transmit compensation time to the MMW_SIZE in RTTBCNRM
9805 * register. MMW_SIZE=0x014 if 9728-byte jumbo is supported.
9807 wr32(E1000_RTTBCNRM, 0x14);
9808 wr32(E1000_RTTBCNRC, bcnrc_val);
9811 static void igb_check_vf_rate_limit(struct igb_adapter *adapter)
9813 int actual_link_speed, i;
9814 bool reset_rate = false;
9816 /* VF TX rate limit was not set or not supported */
9817 if ((adapter->vf_rate_link_speed == 0) ||
9818 (adapter->hw.mac.type != e1000_82576))
9821 actual_link_speed = igb_link_mbps(adapter->link_speed);
9822 if (actual_link_speed != adapter->vf_rate_link_speed) {
9824 adapter->vf_rate_link_speed = 0;
9825 dev_info(&adapter->pdev->dev,
9826 "Link speed has been changed. VF Transmit rate is disabled\n");
9829 for (i = 0; i < adapter->vfs_allocated_count; i++) {
9831 adapter->vf_data[i].tx_rate = 0;
9833 igb_set_vf_rate_limit(&adapter->hw, i,
9834 adapter->vf_data[i].tx_rate,
9839 static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf,
9840 int min_tx_rate, int max_tx_rate)
9842 struct igb_adapter *adapter = netdev_priv(netdev);
9843 struct e1000_hw *hw = &adapter->hw;
9844 int actual_link_speed;
9846 if (hw->mac.type != e1000_82576)
9852 actual_link_speed = igb_link_mbps(adapter->link_speed);
9853 if ((vf >= adapter->vfs_allocated_count) ||
9854 (!(rd32(E1000_STATUS) & E1000_STATUS_LU)) ||
9855 (max_tx_rate < 0) ||
9856 (max_tx_rate > actual_link_speed))
9859 adapter->vf_rate_link_speed = actual_link_speed;
9860 adapter->vf_data[vf].tx_rate = (u16)max_tx_rate;
9861 igb_set_vf_rate_limit(hw, vf, max_tx_rate, actual_link_speed);
9866 static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf,
9869 struct igb_adapter *adapter = netdev_priv(netdev);
9870 struct e1000_hw *hw = &adapter->hw;
9871 u32 reg_val, reg_offset;
9873 if (!adapter->vfs_allocated_count)
9876 if (vf >= adapter->vfs_allocated_count)
9879 reg_offset = (hw->mac.type == e1000_82576) ? E1000_DTXSWC : E1000_TXSWC;
9880 reg_val = rd32(reg_offset);
9882 reg_val |= (BIT(vf) |
9883 BIT(vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT));
9885 reg_val &= ~(BIT(vf) |
9886 BIT(vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT));
9887 wr32(reg_offset, reg_val);
9889 adapter->vf_data[vf].spoofchk_enabled = setting;
9893 static int igb_ndo_set_vf_trust(struct net_device *netdev, int vf, bool setting)
9895 struct igb_adapter *adapter = netdev_priv(netdev);
9897 if (vf >= adapter->vfs_allocated_count)
9899 if (adapter->vf_data[vf].trusted == setting)
9902 adapter->vf_data[vf].trusted = setting;
9904 dev_info(&adapter->pdev->dev, "VF %u is %strusted\n",
9905 vf, setting ? "" : "not ");
9909 static int igb_ndo_get_vf_config(struct net_device *netdev,
9910 int vf, struct ifla_vf_info *ivi)
9912 struct igb_adapter *adapter = netdev_priv(netdev);
9913 if (vf >= adapter->vfs_allocated_count)
9916 memcpy(&ivi->mac, adapter->vf_data[vf].vf_mac_addresses, ETH_ALEN);
9917 ivi->max_tx_rate = adapter->vf_data[vf].tx_rate;
9918 ivi->min_tx_rate = 0;
9919 ivi->vlan = adapter->vf_data[vf].pf_vlan;
9920 ivi->qos = adapter->vf_data[vf].pf_qos;
9921 ivi->spoofchk = adapter->vf_data[vf].spoofchk_enabled;
9922 ivi->trusted = adapter->vf_data[vf].trusted;
9926 static void igb_vmm_control(struct igb_adapter *adapter)
9928 struct e1000_hw *hw = &adapter->hw;
9931 switch (hw->mac.type) {
9937 /* replication is not supported for 82575 */
9940 /* notify HW that the MAC is adding vlan tags */
9941 reg = rd32(E1000_DTXCTL);
9942 reg |= E1000_DTXCTL_VLAN_ADDED;
9943 wr32(E1000_DTXCTL, reg);
9946 /* enable replication vlan tag stripping */
9947 reg = rd32(E1000_RPLOLR);
9948 reg |= E1000_RPLOLR_STRVLAN;
9949 wr32(E1000_RPLOLR, reg);
9952 /* none of the above registers are supported by i350 */
9956 if (adapter->vfs_allocated_count) {
9957 igb_vmdq_set_loopback_pf(hw, true);
9958 igb_vmdq_set_replication_pf(hw, true);
9959 igb_vmdq_set_anti_spoofing_pf(hw, true,
9960 adapter->vfs_allocated_count);
9962 igb_vmdq_set_loopback_pf(hw, false);
9963 igb_vmdq_set_replication_pf(hw, false);
9967 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba)
9969 struct e1000_hw *hw = &adapter->hw;
9974 if (hw->mac.type > e1000_82580) {
9975 if (adapter->flags & IGB_FLAG_DMAC) {
9976 /* force threshold to 0. */
9977 wr32(E1000_DMCTXTH, 0);
9979 /* DMA Coalescing high water mark needs to be greater
9980 * than the Rx threshold. Set hwm to PBA - max frame
9981 * size in 16B units, capping it at PBA - 6KB.
9983 hwm = 64 * (pba - 6);
9984 reg = rd32(E1000_FCRTC);
9985 reg &= ~E1000_FCRTC_RTH_COAL_MASK;
9986 reg |= ((hwm << E1000_FCRTC_RTH_COAL_SHIFT)
9987 & E1000_FCRTC_RTH_COAL_MASK);
9988 wr32(E1000_FCRTC, reg);
9990 /* Set the DMA Coalescing Rx threshold to PBA - 2 * max
9991 * frame size, capping it at PBA - 10KB.
9993 dmac_thr = pba - 10;
9994 reg = rd32(E1000_DMACR);
9995 reg &= ~E1000_DMACR_DMACTHR_MASK;
9996 reg |= ((dmac_thr << E1000_DMACR_DMACTHR_SHIFT)
9997 & E1000_DMACR_DMACTHR_MASK);
9999 /* transition to L0x or L1 if available..*/
10000 reg |= (E1000_DMACR_DMAC_EN | E1000_DMACR_DMAC_LX_MASK);
10002 /* watchdog timer= +-1000 usec in 32usec intervals */
10003 reg |= (1000 >> 5);
10005 /* Disable BMC-to-OS Watchdog Enable */
10006 if (hw->mac.type != e1000_i354)
10007 reg &= ~E1000_DMACR_DC_BMC2OSW_EN;
10008 wr32(E1000_DMACR, reg);
10010 /* no lower threshold to disable
10011 * coalescing(smart fifb)-UTRESH=0
10013 wr32(E1000_DMCRTRH, 0);
10015 reg = (IGB_DMCTLX_DCFLUSH_DIS | 0x4);
10017 wr32(E1000_DMCTLX, reg);
10019 /* free space in tx packet buffer to wake from
10022 wr32(E1000_DMCTXTH, (IGB_MIN_TXPBSIZE -
10023 (IGB_TX_BUF_4096 + adapter->max_frame_size)) >> 6);
10026 if (hw->mac.type >= e1000_i210 ||
10027 (adapter->flags & IGB_FLAG_DMAC)) {
10028 reg = rd32(E1000_PCIEMISC);
10029 reg |= E1000_PCIEMISC_LX_DECISION;
10030 wr32(E1000_PCIEMISC, reg);
10031 } /* endif adapter->dmac is not disabled */
10032 } else if (hw->mac.type == e1000_82580) {
10033 u32 reg = rd32(E1000_PCIEMISC);
10035 wr32(E1000_PCIEMISC, reg & ~E1000_PCIEMISC_LX_DECISION);
10036 wr32(E1000_DMACR, 0);
10041 * igb_read_i2c_byte - Reads 8 bit word over I2C
10042 * @hw: pointer to hardware structure
10043 * @byte_offset: byte offset to read
10044 * @dev_addr: device address
10045 * @data: value read
10047 * Performs byte read operation over I2C interface at
10048 * a specified device address.
10050 s32 igb_read_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
10051 u8 dev_addr, u8 *data)
10053 struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
10054 struct i2c_client *this_client = adapter->i2c_client;
10059 return E1000_ERR_I2C;
10061 swfw_mask = E1000_SWFW_PHY0_SM;
10063 if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask))
10064 return E1000_ERR_SWFW_SYNC;
10066 status = i2c_smbus_read_byte_data(this_client, byte_offset);
10067 hw->mac.ops.release_swfw_sync(hw, swfw_mask);
10070 return E1000_ERR_I2C;
10078 * igb_write_i2c_byte - Writes 8 bit word over I2C
10079 * @hw: pointer to hardware structure
10080 * @byte_offset: byte offset to write
10081 * @dev_addr: device address
10082 * @data: value to write
10084 * Performs byte write operation over I2C interface at
10085 * a specified device address.
10087 s32 igb_write_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
10088 u8 dev_addr, u8 data)
10090 struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
10091 struct i2c_client *this_client = adapter->i2c_client;
10093 u16 swfw_mask = E1000_SWFW_PHY0_SM;
10096 return E1000_ERR_I2C;
10098 if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask))
10099 return E1000_ERR_SWFW_SYNC;
10100 status = i2c_smbus_write_byte_data(this_client, byte_offset, data);
10101 hw->mac.ops.release_swfw_sync(hw, swfw_mask);
10104 return E1000_ERR_I2C;
10110 int igb_reinit_queues(struct igb_adapter *adapter)
10112 struct net_device *netdev = adapter->netdev;
10113 struct pci_dev *pdev = adapter->pdev;
10116 if (netif_running(netdev))
10119 igb_reset_interrupt_capability(adapter);
10121 if (igb_init_interrupt_scheme(adapter, true)) {
10122 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
10126 if (netif_running(netdev))
10127 err = igb_open(netdev);
10132 static void igb_nfc_filter_exit(struct igb_adapter *adapter)
10134 struct igb_nfc_filter *rule;
10136 spin_lock(&adapter->nfc_lock);
10138 hlist_for_each_entry(rule, &adapter->nfc_filter_list, nfc_node)
10139 igb_erase_filter(adapter, rule);
10141 hlist_for_each_entry(rule, &adapter->cls_flower_list, nfc_node)
10142 igb_erase_filter(adapter, rule);
10144 spin_unlock(&adapter->nfc_lock);
10147 static void igb_nfc_filter_restore(struct igb_adapter *adapter)
10149 struct igb_nfc_filter *rule;
10151 spin_lock(&adapter->nfc_lock);
10153 hlist_for_each_entry(rule, &adapter->nfc_filter_list, nfc_node)
10154 igb_add_filter(adapter, rule);
10156 spin_unlock(&adapter->nfc_lock);