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);
187 static const struct dev_pm_ops igb_pm_ops = {
188 SET_SYSTEM_SLEEP_PM_OPS(igb_suspend, igb_resume)
189 SET_RUNTIME_PM_OPS(igb_runtime_suspend, igb_runtime_resume,
193 static void igb_shutdown(struct pci_dev *);
194 static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs);
195 #ifdef CONFIG_IGB_DCA
196 static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
197 static struct notifier_block dca_notifier = {
198 .notifier_call = igb_notify_dca,
203 #ifdef CONFIG_PCI_IOV
204 static unsigned int max_vfs;
205 module_param(max_vfs, uint, 0);
206 MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate per physical function");
207 #endif /* CONFIG_PCI_IOV */
209 static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
210 pci_channel_state_t);
211 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
212 static void igb_io_resume(struct pci_dev *);
214 static const struct pci_error_handlers igb_err_handler = {
215 .error_detected = igb_io_error_detected,
216 .slot_reset = igb_io_slot_reset,
217 .resume = igb_io_resume,
220 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba);
222 static struct pci_driver igb_driver = {
223 .name = igb_driver_name,
224 .id_table = igb_pci_tbl,
226 .remove = igb_remove,
228 .driver.pm = &igb_pm_ops,
230 .shutdown = igb_shutdown,
231 .sriov_configure = igb_pci_sriov_configure,
232 .err_handler = &igb_err_handler
235 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
236 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
237 MODULE_LICENSE("GPL v2");
239 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
240 static int debug = -1;
241 module_param(debug, int, 0);
242 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
244 struct igb_reg_info {
249 static const struct igb_reg_info igb_reg_info_tbl[] = {
251 /* General Registers */
252 {E1000_CTRL, "CTRL"},
253 {E1000_STATUS, "STATUS"},
254 {E1000_CTRL_EXT, "CTRL_EXT"},
256 /* Interrupt Registers */
260 {E1000_RCTL, "RCTL"},
261 {E1000_RDLEN(0), "RDLEN"},
262 {E1000_RDH(0), "RDH"},
263 {E1000_RDT(0), "RDT"},
264 {E1000_RXDCTL(0), "RXDCTL"},
265 {E1000_RDBAL(0), "RDBAL"},
266 {E1000_RDBAH(0), "RDBAH"},
269 {E1000_TCTL, "TCTL"},
270 {E1000_TDBAL(0), "TDBAL"},
271 {E1000_TDBAH(0), "TDBAH"},
272 {E1000_TDLEN(0), "TDLEN"},
273 {E1000_TDH(0), "TDH"},
274 {E1000_TDT(0), "TDT"},
275 {E1000_TXDCTL(0), "TXDCTL"},
276 {E1000_TDFH, "TDFH"},
277 {E1000_TDFT, "TDFT"},
278 {E1000_TDFHS, "TDFHS"},
279 {E1000_TDFPC, "TDFPC"},
281 /* List Terminator */
285 /* igb_regdump - register printout routine */
286 static void igb_regdump(struct e1000_hw *hw, struct igb_reg_info *reginfo)
292 switch (reginfo->ofs) {
294 for (n = 0; n < 4; n++)
295 regs[n] = rd32(E1000_RDLEN(n));
298 for (n = 0; n < 4; n++)
299 regs[n] = rd32(E1000_RDH(n));
302 for (n = 0; n < 4; n++)
303 regs[n] = rd32(E1000_RDT(n));
305 case E1000_RXDCTL(0):
306 for (n = 0; n < 4; n++)
307 regs[n] = rd32(E1000_RXDCTL(n));
310 for (n = 0; n < 4; n++)
311 regs[n] = rd32(E1000_RDBAL(n));
314 for (n = 0; n < 4; n++)
315 regs[n] = rd32(E1000_RDBAH(n));
318 for (n = 0; n < 4; n++)
319 regs[n] = rd32(E1000_TDBAL(n));
322 for (n = 0; n < 4; n++)
323 regs[n] = rd32(E1000_TDBAH(n));
326 for (n = 0; n < 4; n++)
327 regs[n] = rd32(E1000_TDLEN(n));
330 for (n = 0; n < 4; n++)
331 regs[n] = rd32(E1000_TDH(n));
334 for (n = 0; n < 4; n++)
335 regs[n] = rd32(E1000_TDT(n));
337 case E1000_TXDCTL(0):
338 for (n = 0; n < 4; n++)
339 regs[n] = rd32(E1000_TXDCTL(n));
342 pr_info("%-15s %08x\n", reginfo->name, rd32(reginfo->ofs));
346 snprintf(rname, 16, "%s%s", reginfo->name, "[0-3]");
347 pr_info("%-15s %08x %08x %08x %08x\n", rname, regs[0], regs[1],
351 /* igb_dump - Print registers, Tx-rings and Rx-rings */
352 static void igb_dump(struct igb_adapter *adapter)
354 struct net_device *netdev = adapter->netdev;
355 struct e1000_hw *hw = &adapter->hw;
356 struct igb_reg_info *reginfo;
357 struct igb_ring *tx_ring;
358 union e1000_adv_tx_desc *tx_desc;
359 struct my_u0 { __le64 a; __le64 b; } *u0;
360 struct igb_ring *rx_ring;
361 union e1000_adv_rx_desc *rx_desc;
365 if (!netif_msg_hw(adapter))
368 /* Print netdevice Info */
370 dev_info(&adapter->pdev->dev, "Net device Info\n");
371 pr_info("Device Name state trans_start\n");
372 pr_info("%-15s %016lX %016lX\n", netdev->name,
373 netdev->state, dev_trans_start(netdev));
376 /* Print Registers */
377 dev_info(&adapter->pdev->dev, "Register Dump\n");
378 pr_info(" Register Name Value\n");
379 for (reginfo = (struct igb_reg_info *)igb_reg_info_tbl;
380 reginfo->name; reginfo++) {
381 igb_regdump(hw, reginfo);
384 /* Print TX Ring Summary */
385 if (!netdev || !netif_running(netdev))
388 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
389 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
390 for (n = 0; n < adapter->num_tx_queues; n++) {
391 struct igb_tx_buffer *buffer_info;
392 tx_ring = adapter->tx_ring[n];
393 buffer_info = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
394 pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
395 n, tx_ring->next_to_use, tx_ring->next_to_clean,
396 (u64)dma_unmap_addr(buffer_info, dma),
397 dma_unmap_len(buffer_info, len),
398 buffer_info->next_to_watch,
399 (u64)buffer_info->time_stamp);
403 if (!netif_msg_tx_done(adapter))
404 goto rx_ring_summary;
406 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
408 /* Transmit Descriptor Formats
410 * Advanced Transmit Descriptor
411 * +--------------------------------------------------------------+
412 * 0 | Buffer Address [63:0] |
413 * +--------------------------------------------------------------+
414 * 8 | PAYLEN | PORTS |CC|IDX | STA | DCMD |DTYP|MAC|RSV| DTALEN |
415 * +--------------------------------------------------------------+
416 * 63 46 45 40 39 38 36 35 32 31 24 15 0
419 for (n = 0; n < adapter->num_tx_queues; n++) {
420 tx_ring = adapter->tx_ring[n];
421 pr_info("------------------------------------\n");
422 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
423 pr_info("------------------------------------\n");
424 pr_info("T [desc] [address 63:0 ] [PlPOCIStDDM Ln] [bi->dma ] leng ntw timestamp bi->skb\n");
426 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
427 const char *next_desc;
428 struct igb_tx_buffer *buffer_info;
429 tx_desc = IGB_TX_DESC(tx_ring, i);
430 buffer_info = &tx_ring->tx_buffer_info[i];
431 u0 = (struct my_u0 *)tx_desc;
432 if (i == tx_ring->next_to_use &&
433 i == tx_ring->next_to_clean)
434 next_desc = " NTC/U";
435 else if (i == tx_ring->next_to_use)
437 else if (i == tx_ring->next_to_clean)
442 pr_info("T [0x%03X] %016llX %016llX %016llX %04X %p %016llX %p%s\n",
443 i, le64_to_cpu(u0->a),
445 (u64)dma_unmap_addr(buffer_info, dma),
446 dma_unmap_len(buffer_info, len),
447 buffer_info->next_to_watch,
448 (u64)buffer_info->time_stamp,
449 buffer_info->skb, next_desc);
451 if (netif_msg_pktdata(adapter) && buffer_info->skb)
452 print_hex_dump(KERN_INFO, "",
454 16, 1, buffer_info->skb->data,
455 dma_unmap_len(buffer_info, len),
460 /* Print RX Rings Summary */
462 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
463 pr_info("Queue [NTU] [NTC]\n");
464 for (n = 0; n < adapter->num_rx_queues; n++) {
465 rx_ring = adapter->rx_ring[n];
466 pr_info(" %5d %5X %5X\n",
467 n, rx_ring->next_to_use, rx_ring->next_to_clean);
471 if (!netif_msg_rx_status(adapter))
474 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
476 /* Advanced Receive Descriptor (Read) Format
478 * +-----------------------------------------------------+
479 * 0 | Packet Buffer Address [63:1] |A0/NSE|
480 * +----------------------------------------------+------+
481 * 8 | Header Buffer Address [63:1] | DD |
482 * +-----------------------------------------------------+
485 * Advanced Receive Descriptor (Write-Back) Format
487 * 63 48 47 32 31 30 21 20 17 16 4 3 0
488 * +------------------------------------------------------+
489 * 0 | Packet IP |SPH| HDR_LEN | RSV|Packet| RSS |
490 * | Checksum Ident | | | | Type | Type |
491 * +------------------------------------------------------+
492 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
493 * +------------------------------------------------------+
494 * 63 48 47 32 31 20 19 0
497 for (n = 0; n < adapter->num_rx_queues; n++) {
498 rx_ring = adapter->rx_ring[n];
499 pr_info("------------------------------------\n");
500 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
501 pr_info("------------------------------------\n");
502 pr_info("R [desc] [ PktBuf A0] [ HeadBuf DD] [bi->dma ] [bi->skb] <-- Adv Rx Read format\n");
503 pr_info("RWB[desc] [PcsmIpSHl PtRs] [vl er S cks ln] ---------------- [bi->skb] <-- Adv Rx Write-Back format\n");
505 for (i = 0; i < rx_ring->count; i++) {
506 const char *next_desc;
507 struct igb_rx_buffer *buffer_info;
508 buffer_info = &rx_ring->rx_buffer_info[i];
509 rx_desc = IGB_RX_DESC(rx_ring, i);
510 u0 = (struct my_u0 *)rx_desc;
511 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
513 if (i == rx_ring->next_to_use)
515 else if (i == rx_ring->next_to_clean)
520 if (staterr & E1000_RXD_STAT_DD) {
521 /* Descriptor Done */
522 pr_info("%s[0x%03X] %016llX %016llX ---------------- %s\n",
528 pr_info("%s[0x%03X] %016llX %016llX %016llX %s\n",
532 (u64)buffer_info->dma,
535 if (netif_msg_pktdata(adapter) &&
536 buffer_info->dma && buffer_info->page) {
537 print_hex_dump(KERN_INFO, "",
540 page_address(buffer_info->page) +
541 buffer_info->page_offset,
542 igb_rx_bufsz(rx_ring), true);
553 * igb_get_i2c_data - Reads the I2C SDA data bit
554 * @data: opaque pointer to adapter struct
556 * Returns the I2C data bit value
558 static int igb_get_i2c_data(void *data)
560 struct igb_adapter *adapter = (struct igb_adapter *)data;
561 struct e1000_hw *hw = &adapter->hw;
562 s32 i2cctl = rd32(E1000_I2CPARAMS);
564 return !!(i2cctl & E1000_I2C_DATA_IN);
568 * igb_set_i2c_data - Sets the I2C data bit
569 * @data: pointer to hardware structure
570 * @state: I2C data value (0 or 1) to set
572 * Sets the I2C data bit
574 static void igb_set_i2c_data(void *data, int state)
576 struct igb_adapter *adapter = (struct igb_adapter *)data;
577 struct e1000_hw *hw = &adapter->hw;
578 s32 i2cctl = rd32(E1000_I2CPARAMS);
581 i2cctl |= E1000_I2C_DATA_OUT | E1000_I2C_DATA_OE_N;
583 i2cctl &= ~E1000_I2C_DATA_OE_N;
584 i2cctl &= ~E1000_I2C_DATA_OUT;
587 wr32(E1000_I2CPARAMS, i2cctl);
592 * igb_set_i2c_clk - Sets the I2C SCL clock
593 * @data: pointer to hardware structure
594 * @state: state to set clock
596 * Sets the I2C clock line to state
598 static void igb_set_i2c_clk(void *data, int state)
600 struct igb_adapter *adapter = (struct igb_adapter *)data;
601 struct e1000_hw *hw = &adapter->hw;
602 s32 i2cctl = rd32(E1000_I2CPARAMS);
605 i2cctl |= E1000_I2C_CLK_OUT | E1000_I2C_CLK_OE_N;
607 i2cctl &= ~E1000_I2C_CLK_OUT;
608 i2cctl &= ~E1000_I2C_CLK_OE_N;
610 wr32(E1000_I2CPARAMS, i2cctl);
615 * igb_get_i2c_clk - Gets the I2C SCL clock state
616 * @data: pointer to hardware structure
618 * Gets the I2C clock state
620 static int igb_get_i2c_clk(void *data)
622 struct igb_adapter *adapter = (struct igb_adapter *)data;
623 struct e1000_hw *hw = &adapter->hw;
624 s32 i2cctl = rd32(E1000_I2CPARAMS);
626 return !!(i2cctl & E1000_I2C_CLK_IN);
629 static const struct i2c_algo_bit_data igb_i2c_algo = {
630 .setsda = igb_set_i2c_data,
631 .setscl = igb_set_i2c_clk,
632 .getsda = igb_get_i2c_data,
633 .getscl = igb_get_i2c_clk,
639 * igb_get_hw_dev - return device
640 * @hw: pointer to hardware structure
642 * used by hardware layer to print debugging information
644 struct net_device *igb_get_hw_dev(struct e1000_hw *hw)
646 struct igb_adapter *adapter = hw->back;
647 return adapter->netdev;
651 * igb_init_module - Driver Registration Routine
653 * igb_init_module is the first routine called when the driver is
654 * loaded. All it does is register with the PCI subsystem.
656 static int __init igb_init_module(void)
660 pr_info("%s\n", igb_driver_string);
661 pr_info("%s\n", igb_copyright);
663 #ifdef CONFIG_IGB_DCA
664 dca_register_notify(&dca_notifier);
666 ret = pci_register_driver(&igb_driver);
670 module_init(igb_init_module);
673 * igb_exit_module - Driver Exit Cleanup Routine
675 * igb_exit_module is called just before the driver is removed
678 static void __exit igb_exit_module(void)
680 #ifdef CONFIG_IGB_DCA
681 dca_unregister_notify(&dca_notifier);
683 pci_unregister_driver(&igb_driver);
686 module_exit(igb_exit_module);
688 #define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
690 * igb_cache_ring_register - Descriptor ring to register mapping
691 * @adapter: board private structure to initialize
693 * Once we know the feature-set enabled for the device, we'll cache
694 * the register offset the descriptor ring is assigned to.
696 static void igb_cache_ring_register(struct igb_adapter *adapter)
699 u32 rbase_offset = adapter->vfs_allocated_count;
701 switch (adapter->hw.mac.type) {
703 /* The queues are allocated for virtualization such that VF 0
704 * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
705 * In order to avoid collision we start at the first free queue
706 * and continue consuming queues in the same sequence
708 if (adapter->vfs_allocated_count) {
709 for (; i < adapter->rss_queues; i++)
710 adapter->rx_ring[i]->reg_idx = rbase_offset +
721 for (; i < adapter->num_rx_queues; i++)
722 adapter->rx_ring[i]->reg_idx = rbase_offset + i;
723 for (; j < adapter->num_tx_queues; j++)
724 adapter->tx_ring[j]->reg_idx = rbase_offset + j;
729 u32 igb_rd32(struct e1000_hw *hw, u32 reg)
731 struct igb_adapter *igb = container_of(hw, struct igb_adapter, hw);
732 u8 __iomem *hw_addr = READ_ONCE(hw->hw_addr);
735 if (E1000_REMOVED(hw_addr))
738 value = readl(&hw_addr[reg]);
740 /* reads should not return all F's */
741 if (!(~value) && (!reg || !(~readl(hw_addr)))) {
742 struct net_device *netdev = igb->netdev;
744 netdev_err(netdev, "PCIe link lost\n");
745 WARN(pci_device_is_present(igb->pdev),
746 "igb: Failed to read reg 0x%x!\n", reg);
753 * igb_write_ivar - configure ivar for given MSI-X vector
754 * @hw: pointer to the HW structure
755 * @msix_vector: vector number we are allocating to a given ring
756 * @index: row index of IVAR register to write within IVAR table
757 * @offset: column offset of in IVAR, should be multiple of 8
759 * This function is intended to handle the writing of the IVAR register
760 * for adapters 82576 and newer. The IVAR table consists of 2 columns,
761 * each containing an cause allocation for an Rx and Tx ring, and a
762 * variable number of rows depending on the number of queues supported.
764 static void igb_write_ivar(struct e1000_hw *hw, int msix_vector,
765 int index, int offset)
767 u32 ivar = array_rd32(E1000_IVAR0, index);
769 /* clear any bits that are currently set */
770 ivar &= ~((u32)0xFF << offset);
772 /* write vector and valid bit */
773 ivar |= (msix_vector | E1000_IVAR_VALID) << offset;
775 array_wr32(E1000_IVAR0, index, ivar);
778 #define IGB_N0_QUEUE -1
779 static void igb_assign_vector(struct igb_q_vector *q_vector, int msix_vector)
781 struct igb_adapter *adapter = q_vector->adapter;
782 struct e1000_hw *hw = &adapter->hw;
783 int rx_queue = IGB_N0_QUEUE;
784 int tx_queue = IGB_N0_QUEUE;
787 if (q_vector->rx.ring)
788 rx_queue = q_vector->rx.ring->reg_idx;
789 if (q_vector->tx.ring)
790 tx_queue = q_vector->tx.ring->reg_idx;
792 switch (hw->mac.type) {
794 /* The 82575 assigns vectors using a bitmask, which matches the
795 * bitmask for the EICR/EIMS/EIMC registers. To assign one
796 * or more queues to a vector, we write the appropriate bits
797 * into the MSIXBM register for that vector.
799 if (rx_queue > IGB_N0_QUEUE)
800 msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
801 if (tx_queue > IGB_N0_QUEUE)
802 msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
803 if (!(adapter->flags & IGB_FLAG_HAS_MSIX) && msix_vector == 0)
804 msixbm |= E1000_EIMS_OTHER;
805 array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
806 q_vector->eims_value = msixbm;
809 /* 82576 uses a table that essentially consists of 2 columns
810 * with 8 rows. The ordering is column-major so we use the
811 * lower 3 bits as the row index, and the 4th bit as the
814 if (rx_queue > IGB_N0_QUEUE)
815 igb_write_ivar(hw, msix_vector,
817 (rx_queue & 0x8) << 1);
818 if (tx_queue > IGB_N0_QUEUE)
819 igb_write_ivar(hw, msix_vector,
821 ((tx_queue & 0x8) << 1) + 8);
822 q_vector->eims_value = BIT(msix_vector);
829 /* On 82580 and newer adapters the scheme is similar to 82576
830 * however instead of ordering column-major we have things
831 * ordered row-major. So we traverse the table by using
832 * bit 0 as the column offset, and the remaining bits as the
835 if (rx_queue > IGB_N0_QUEUE)
836 igb_write_ivar(hw, msix_vector,
838 (rx_queue & 0x1) << 4);
839 if (tx_queue > IGB_N0_QUEUE)
840 igb_write_ivar(hw, msix_vector,
842 ((tx_queue & 0x1) << 4) + 8);
843 q_vector->eims_value = BIT(msix_vector);
850 /* add q_vector eims value to global eims_enable_mask */
851 adapter->eims_enable_mask |= q_vector->eims_value;
853 /* configure q_vector to set itr on first interrupt */
854 q_vector->set_itr = 1;
858 * igb_configure_msix - Configure MSI-X hardware
859 * @adapter: board private structure to initialize
861 * igb_configure_msix sets up the hardware to properly
862 * generate MSI-X interrupts.
864 static void igb_configure_msix(struct igb_adapter *adapter)
868 struct e1000_hw *hw = &adapter->hw;
870 adapter->eims_enable_mask = 0;
872 /* set vector for other causes, i.e. link changes */
873 switch (hw->mac.type) {
875 tmp = rd32(E1000_CTRL_EXT);
876 /* enable MSI-X PBA support*/
877 tmp |= E1000_CTRL_EXT_PBA_CLR;
879 /* Auto-Mask interrupts upon ICR read. */
880 tmp |= E1000_CTRL_EXT_EIAME;
881 tmp |= E1000_CTRL_EXT_IRCA;
883 wr32(E1000_CTRL_EXT, tmp);
885 /* enable msix_other interrupt */
886 array_wr32(E1000_MSIXBM(0), vector++, E1000_EIMS_OTHER);
887 adapter->eims_other = E1000_EIMS_OTHER;
897 /* Turn on MSI-X capability first, or our settings
898 * won't stick. And it will take days to debug.
900 wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
901 E1000_GPIE_PBA | E1000_GPIE_EIAME |
904 /* enable msix_other interrupt */
905 adapter->eims_other = BIT(vector);
906 tmp = (vector++ | E1000_IVAR_VALID) << 8;
908 wr32(E1000_IVAR_MISC, tmp);
911 /* do nothing, since nothing else supports MSI-X */
913 } /* switch (hw->mac.type) */
915 adapter->eims_enable_mask |= adapter->eims_other;
917 for (i = 0; i < adapter->num_q_vectors; i++)
918 igb_assign_vector(adapter->q_vector[i], vector++);
924 * igb_request_msix - Initialize MSI-X interrupts
925 * @adapter: board private structure to initialize
927 * igb_request_msix allocates MSI-X vectors and requests interrupts from the
930 static int igb_request_msix(struct igb_adapter *adapter)
932 unsigned int num_q_vectors = adapter->num_q_vectors;
933 struct net_device *netdev = adapter->netdev;
934 int i, err = 0, vector = 0, free_vector = 0;
936 err = request_irq(adapter->msix_entries[vector].vector,
937 igb_msix_other, 0, netdev->name, adapter);
941 if (num_q_vectors > MAX_Q_VECTORS) {
942 num_q_vectors = MAX_Q_VECTORS;
943 dev_warn(&adapter->pdev->dev,
944 "The number of queue vectors (%d) is higher than max allowed (%d)\n",
945 adapter->num_q_vectors, MAX_Q_VECTORS);
947 for (i = 0; i < num_q_vectors; i++) {
948 struct igb_q_vector *q_vector = adapter->q_vector[i];
952 q_vector->itr_register = adapter->io_addr + E1000_EITR(vector);
954 if (q_vector->rx.ring && q_vector->tx.ring)
955 sprintf(q_vector->name, "%s-TxRx-%u", netdev->name,
956 q_vector->rx.ring->queue_index);
957 else if (q_vector->tx.ring)
958 sprintf(q_vector->name, "%s-tx-%u", netdev->name,
959 q_vector->tx.ring->queue_index);
960 else if (q_vector->rx.ring)
961 sprintf(q_vector->name, "%s-rx-%u", netdev->name,
962 q_vector->rx.ring->queue_index);
964 sprintf(q_vector->name, "%s-unused", netdev->name);
966 err = request_irq(adapter->msix_entries[vector].vector,
967 igb_msix_ring, 0, q_vector->name,
973 igb_configure_msix(adapter);
977 /* free already assigned IRQs */
978 free_irq(adapter->msix_entries[free_vector++].vector, adapter);
981 for (i = 0; i < vector; i++) {
982 free_irq(adapter->msix_entries[free_vector++].vector,
983 adapter->q_vector[i]);
990 * igb_free_q_vector - Free memory allocated for specific interrupt vector
991 * @adapter: board private structure to initialize
992 * @v_idx: Index of vector to be freed
994 * This function frees the memory allocated to the q_vector.
996 static void igb_free_q_vector(struct igb_adapter *adapter, int v_idx)
998 struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
1000 adapter->q_vector[v_idx] = NULL;
1002 /* igb_get_stats64() might access the rings on this vector,
1003 * we must wait a grace period before freeing it.
1006 kfree_rcu(q_vector, rcu);
1010 * igb_reset_q_vector - Reset config for interrupt vector
1011 * @adapter: board private structure to initialize
1012 * @v_idx: Index of vector to be reset
1014 * If NAPI is enabled it will delete any references to the
1015 * NAPI struct. This is preparation for igb_free_q_vector.
1017 static void igb_reset_q_vector(struct igb_adapter *adapter, int v_idx)
1019 struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
1021 /* Coming from igb_set_interrupt_capability, the vectors are not yet
1022 * allocated. So, q_vector is NULL so we should stop here.
1027 if (q_vector->tx.ring)
1028 adapter->tx_ring[q_vector->tx.ring->queue_index] = NULL;
1030 if (q_vector->rx.ring)
1031 adapter->rx_ring[q_vector->rx.ring->queue_index] = NULL;
1033 netif_napi_del(&q_vector->napi);
1037 static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
1039 int v_idx = adapter->num_q_vectors;
1041 if (adapter->flags & IGB_FLAG_HAS_MSIX)
1042 pci_disable_msix(adapter->pdev);
1043 else if (adapter->flags & IGB_FLAG_HAS_MSI)
1044 pci_disable_msi(adapter->pdev);
1047 igb_reset_q_vector(adapter, v_idx);
1051 * igb_free_q_vectors - Free memory allocated for interrupt vectors
1052 * @adapter: board private structure to initialize
1054 * This function frees the memory allocated to the q_vectors. In addition if
1055 * NAPI is enabled it will delete any references to the NAPI struct prior
1056 * to freeing the q_vector.
1058 static void igb_free_q_vectors(struct igb_adapter *adapter)
1060 int v_idx = adapter->num_q_vectors;
1062 adapter->num_tx_queues = 0;
1063 adapter->num_rx_queues = 0;
1064 adapter->num_q_vectors = 0;
1067 igb_reset_q_vector(adapter, v_idx);
1068 igb_free_q_vector(adapter, v_idx);
1073 * igb_clear_interrupt_scheme - reset the device to a state of no interrupts
1074 * @adapter: board private structure to initialize
1076 * This function resets the device so that it has 0 Rx queues, Tx queues, and
1077 * MSI-X interrupts allocated.
1079 static void igb_clear_interrupt_scheme(struct igb_adapter *adapter)
1081 igb_free_q_vectors(adapter);
1082 igb_reset_interrupt_capability(adapter);
1086 * igb_set_interrupt_capability - set MSI or MSI-X if supported
1087 * @adapter: board private structure to initialize
1088 * @msix: boolean value of MSIX capability
1090 * Attempt to configure interrupts using the best available
1091 * capabilities of the hardware and kernel.
1093 static void igb_set_interrupt_capability(struct igb_adapter *adapter, bool msix)
1100 adapter->flags |= IGB_FLAG_HAS_MSIX;
1102 /* Number of supported queues. */
1103 adapter->num_rx_queues = adapter->rss_queues;
1104 if (adapter->vfs_allocated_count)
1105 adapter->num_tx_queues = 1;
1107 adapter->num_tx_queues = adapter->rss_queues;
1109 /* start with one vector for every Rx queue */
1110 numvecs = adapter->num_rx_queues;
1112 /* if Tx handler is separate add 1 for every Tx queue */
1113 if (!(adapter->flags & IGB_FLAG_QUEUE_PAIRS))
1114 numvecs += adapter->num_tx_queues;
1116 /* store the number of vectors reserved for queues */
1117 adapter->num_q_vectors = numvecs;
1119 /* add 1 vector for link status interrupts */
1121 for (i = 0; i < numvecs; i++)
1122 adapter->msix_entries[i].entry = i;
1124 err = pci_enable_msix_range(adapter->pdev,
1125 adapter->msix_entries,
1131 igb_reset_interrupt_capability(adapter);
1133 /* If we can't do MSI-X, try MSI */
1135 adapter->flags &= ~IGB_FLAG_HAS_MSIX;
1136 #ifdef CONFIG_PCI_IOV
1137 /* disable SR-IOV for non MSI-X configurations */
1138 if (adapter->vf_data) {
1139 struct e1000_hw *hw = &adapter->hw;
1140 /* disable iov and allow time for transactions to clear */
1141 pci_disable_sriov(adapter->pdev);
1144 kfree(adapter->vf_mac_list);
1145 adapter->vf_mac_list = NULL;
1146 kfree(adapter->vf_data);
1147 adapter->vf_data = NULL;
1148 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
1151 dev_info(&adapter->pdev->dev, "IOV Disabled\n");
1154 adapter->vfs_allocated_count = 0;
1155 adapter->rss_queues = 1;
1156 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
1157 adapter->num_rx_queues = 1;
1158 adapter->num_tx_queues = 1;
1159 adapter->num_q_vectors = 1;
1160 if (!pci_enable_msi(adapter->pdev))
1161 adapter->flags |= IGB_FLAG_HAS_MSI;
1164 static void igb_add_ring(struct igb_ring *ring,
1165 struct igb_ring_container *head)
1172 * igb_alloc_q_vector - Allocate memory for a single interrupt vector
1173 * @adapter: board private structure to initialize
1174 * @v_count: q_vectors allocated on adapter, used for ring interleaving
1175 * @v_idx: index of vector in adapter struct
1176 * @txr_count: total number of Tx rings to allocate
1177 * @txr_idx: index of first Tx ring to allocate
1178 * @rxr_count: total number of Rx rings to allocate
1179 * @rxr_idx: index of first Rx ring to allocate
1181 * We allocate one q_vector. If allocation fails we return -ENOMEM.
1183 static int igb_alloc_q_vector(struct igb_adapter *adapter,
1184 int v_count, int v_idx,
1185 int txr_count, int txr_idx,
1186 int rxr_count, int rxr_idx)
1188 struct igb_q_vector *q_vector;
1189 struct igb_ring *ring;
1193 /* igb only supports 1 Tx and/or 1 Rx queue per vector */
1194 if (txr_count > 1 || rxr_count > 1)
1197 ring_count = txr_count + rxr_count;
1198 size = kmalloc_size_roundup(struct_size(q_vector, ring, ring_count));
1200 /* allocate q_vector and rings */
1201 q_vector = adapter->q_vector[v_idx];
1203 q_vector = kzalloc(size, GFP_KERNEL);
1204 } else if (size > ksize(q_vector)) {
1205 struct igb_q_vector *new_q_vector;
1207 new_q_vector = kzalloc(size, GFP_KERNEL);
1209 kfree_rcu(q_vector, rcu);
1210 q_vector = new_q_vector;
1212 memset(q_vector, 0, size);
1217 /* initialize NAPI */
1218 netif_napi_add(adapter->netdev, &q_vector->napi, igb_poll);
1220 /* tie q_vector and adapter together */
1221 adapter->q_vector[v_idx] = q_vector;
1222 q_vector->adapter = adapter;
1224 /* initialize work limits */
1225 q_vector->tx.work_limit = adapter->tx_work_limit;
1227 /* initialize ITR configuration */
1228 q_vector->itr_register = adapter->io_addr + E1000_EITR(0);
1229 q_vector->itr_val = IGB_START_ITR;
1231 /* initialize pointer to rings */
1232 ring = q_vector->ring;
1236 /* rx or rx/tx vector */
1237 if (!adapter->rx_itr_setting || adapter->rx_itr_setting > 3)
1238 q_vector->itr_val = adapter->rx_itr_setting;
1240 /* tx only vector */
1241 if (!adapter->tx_itr_setting || adapter->tx_itr_setting > 3)
1242 q_vector->itr_val = adapter->tx_itr_setting;
1246 /* assign generic ring traits */
1247 ring->dev = &adapter->pdev->dev;
1248 ring->netdev = adapter->netdev;
1250 /* configure backlink on ring */
1251 ring->q_vector = q_vector;
1253 /* update q_vector Tx values */
1254 igb_add_ring(ring, &q_vector->tx);
1256 /* For 82575, context index must be unique per ring. */
1257 if (adapter->hw.mac.type == e1000_82575)
1258 set_bit(IGB_RING_FLAG_TX_CTX_IDX, &ring->flags);
1260 /* apply Tx specific ring traits */
1261 ring->count = adapter->tx_ring_count;
1262 ring->queue_index = txr_idx;
1264 ring->cbs_enable = false;
1265 ring->idleslope = 0;
1266 ring->sendslope = 0;
1270 u64_stats_init(&ring->tx_syncp);
1271 u64_stats_init(&ring->tx_syncp2);
1273 /* assign ring to adapter */
1274 adapter->tx_ring[txr_idx] = ring;
1276 /* push pointer to next ring */
1281 /* assign generic ring traits */
1282 ring->dev = &adapter->pdev->dev;
1283 ring->netdev = adapter->netdev;
1285 /* configure backlink on ring */
1286 ring->q_vector = q_vector;
1288 /* update q_vector Rx values */
1289 igb_add_ring(ring, &q_vector->rx);
1291 /* set flag indicating ring supports SCTP checksum offload */
1292 if (adapter->hw.mac.type >= e1000_82576)
1293 set_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags);
1295 /* On i350, i354, i210, and i211, loopback VLAN packets
1296 * have the tag byte-swapped.
1298 if (adapter->hw.mac.type >= e1000_i350)
1299 set_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &ring->flags);
1301 /* apply Rx specific ring traits */
1302 ring->count = adapter->rx_ring_count;
1303 ring->queue_index = rxr_idx;
1305 u64_stats_init(&ring->rx_syncp);
1307 /* assign ring to adapter */
1308 adapter->rx_ring[rxr_idx] = ring;
1316 * igb_alloc_q_vectors - Allocate memory for interrupt vectors
1317 * @adapter: board private structure to initialize
1319 * We allocate one q_vector per queue interrupt. If allocation fails we
1322 static int igb_alloc_q_vectors(struct igb_adapter *adapter)
1324 int q_vectors = adapter->num_q_vectors;
1325 int rxr_remaining = adapter->num_rx_queues;
1326 int txr_remaining = adapter->num_tx_queues;
1327 int rxr_idx = 0, txr_idx = 0, v_idx = 0;
1330 if (q_vectors >= (rxr_remaining + txr_remaining)) {
1331 for (; rxr_remaining; v_idx++) {
1332 err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1338 /* update counts and index */
1344 for (; v_idx < q_vectors; v_idx++) {
1345 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_idx);
1346 int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_idx);
1348 err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1349 tqpv, txr_idx, rqpv, rxr_idx);
1354 /* update counts and index */
1355 rxr_remaining -= rqpv;
1356 txr_remaining -= tqpv;
1364 adapter->num_tx_queues = 0;
1365 adapter->num_rx_queues = 0;
1366 adapter->num_q_vectors = 0;
1369 igb_free_q_vector(adapter, v_idx);
1375 * igb_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
1376 * @adapter: board private structure to initialize
1377 * @msix: boolean value of MSIX capability
1379 * This function initializes the interrupts and allocates all of the queues.
1381 static int igb_init_interrupt_scheme(struct igb_adapter *adapter, bool msix)
1383 struct pci_dev *pdev = adapter->pdev;
1386 igb_set_interrupt_capability(adapter, msix);
1388 err = igb_alloc_q_vectors(adapter);
1390 dev_err(&pdev->dev, "Unable to allocate memory for vectors\n");
1391 goto err_alloc_q_vectors;
1394 igb_cache_ring_register(adapter);
1398 err_alloc_q_vectors:
1399 igb_reset_interrupt_capability(adapter);
1404 * igb_request_irq - initialize interrupts
1405 * @adapter: board private structure to initialize
1407 * Attempts to configure interrupts using the best available
1408 * capabilities of the hardware and kernel.
1410 static int igb_request_irq(struct igb_adapter *adapter)
1412 struct net_device *netdev = adapter->netdev;
1413 struct pci_dev *pdev = adapter->pdev;
1416 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1417 err = igb_request_msix(adapter);
1420 /* fall back to MSI */
1421 igb_free_all_tx_resources(adapter);
1422 igb_free_all_rx_resources(adapter);
1424 igb_clear_interrupt_scheme(adapter);
1425 err = igb_init_interrupt_scheme(adapter, false);
1429 igb_setup_all_tx_resources(adapter);
1430 igb_setup_all_rx_resources(adapter);
1431 igb_configure(adapter);
1434 igb_assign_vector(adapter->q_vector[0], 0);
1436 if (adapter->flags & IGB_FLAG_HAS_MSI) {
1437 err = request_irq(pdev->irq, igb_intr_msi, 0,
1438 netdev->name, adapter);
1442 /* fall back to legacy interrupts */
1443 igb_reset_interrupt_capability(adapter);
1444 adapter->flags &= ~IGB_FLAG_HAS_MSI;
1447 err = request_irq(pdev->irq, igb_intr, IRQF_SHARED,
1448 netdev->name, adapter);
1451 dev_err(&pdev->dev, "Error %d getting interrupt\n",
1458 static void igb_free_irq(struct igb_adapter *adapter)
1460 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1463 free_irq(adapter->msix_entries[vector++].vector, adapter);
1465 for (i = 0; i < adapter->num_q_vectors; i++)
1466 free_irq(adapter->msix_entries[vector++].vector,
1467 adapter->q_vector[i]);
1469 free_irq(adapter->pdev->irq, adapter);
1474 * igb_irq_disable - Mask off interrupt generation on the NIC
1475 * @adapter: board private structure
1477 static void igb_irq_disable(struct igb_adapter *adapter)
1479 struct e1000_hw *hw = &adapter->hw;
1481 /* we need to be careful when disabling interrupts. The VFs are also
1482 * mapped into these registers and so clearing the bits can cause
1483 * issues on the VF drivers so we only need to clear what we set
1485 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1486 u32 regval = rd32(E1000_EIAM);
1488 wr32(E1000_EIAM, regval & ~adapter->eims_enable_mask);
1489 wr32(E1000_EIMC, adapter->eims_enable_mask);
1490 regval = rd32(E1000_EIAC);
1491 wr32(E1000_EIAC, regval & ~adapter->eims_enable_mask);
1495 wr32(E1000_IMC, ~0);
1497 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1500 for (i = 0; i < adapter->num_q_vectors; i++)
1501 synchronize_irq(adapter->msix_entries[i].vector);
1503 synchronize_irq(adapter->pdev->irq);
1508 * igb_irq_enable - Enable default interrupt generation settings
1509 * @adapter: board private structure
1511 static void igb_irq_enable(struct igb_adapter *adapter)
1513 struct e1000_hw *hw = &adapter->hw;
1515 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1516 u32 ims = E1000_IMS_LSC | E1000_IMS_DOUTSYNC | E1000_IMS_DRSTA;
1517 u32 regval = rd32(E1000_EIAC);
1519 wr32(E1000_EIAC, regval | adapter->eims_enable_mask);
1520 regval = rd32(E1000_EIAM);
1521 wr32(E1000_EIAM, regval | adapter->eims_enable_mask);
1522 wr32(E1000_EIMS, adapter->eims_enable_mask);
1523 if (adapter->vfs_allocated_count) {
1524 wr32(E1000_MBVFIMR, 0xFF);
1525 ims |= E1000_IMS_VMMB;
1527 wr32(E1000_IMS, ims);
1529 wr32(E1000_IMS, IMS_ENABLE_MASK |
1531 wr32(E1000_IAM, IMS_ENABLE_MASK |
1536 static void igb_update_mng_vlan(struct igb_adapter *adapter)
1538 struct e1000_hw *hw = &adapter->hw;
1539 u16 pf_id = adapter->vfs_allocated_count;
1540 u16 vid = adapter->hw.mng_cookie.vlan_id;
1541 u16 old_vid = adapter->mng_vlan_id;
1543 if (hw->mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1544 /* add VID to filter table */
1545 igb_vfta_set(hw, vid, pf_id, true, true);
1546 adapter->mng_vlan_id = vid;
1548 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1551 if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
1553 !test_bit(old_vid, adapter->active_vlans)) {
1554 /* remove VID from filter table */
1555 igb_vfta_set(hw, vid, pf_id, false, true);
1560 * igb_release_hw_control - release control of the h/w to f/w
1561 * @adapter: address of board private structure
1563 * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
1564 * For ASF and Pass Through versions of f/w this means that the
1565 * driver is no longer loaded.
1567 static void igb_release_hw_control(struct igb_adapter *adapter)
1569 struct e1000_hw *hw = &adapter->hw;
1572 /* Let firmware take over control of h/w */
1573 ctrl_ext = rd32(E1000_CTRL_EXT);
1574 wr32(E1000_CTRL_EXT,
1575 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1579 * igb_get_hw_control - get control of the h/w from f/w
1580 * @adapter: address of board private structure
1582 * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
1583 * For ASF and Pass Through versions of f/w this means that
1584 * the driver is loaded.
1586 static void igb_get_hw_control(struct igb_adapter *adapter)
1588 struct e1000_hw *hw = &adapter->hw;
1591 /* Let firmware know the driver has taken over */
1592 ctrl_ext = rd32(E1000_CTRL_EXT);
1593 wr32(E1000_CTRL_EXT,
1594 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1597 static void enable_fqtss(struct igb_adapter *adapter, bool enable)
1599 struct net_device *netdev = adapter->netdev;
1600 struct e1000_hw *hw = &adapter->hw;
1602 WARN_ON(hw->mac.type != e1000_i210);
1605 adapter->flags |= IGB_FLAG_FQTSS;
1607 adapter->flags &= ~IGB_FLAG_FQTSS;
1609 if (netif_running(netdev))
1610 schedule_work(&adapter->reset_task);
1613 static bool is_fqtss_enabled(struct igb_adapter *adapter)
1615 return (adapter->flags & IGB_FLAG_FQTSS) ? true : false;
1618 static void set_tx_desc_fetch_prio(struct e1000_hw *hw, int queue,
1619 enum tx_queue_prio prio)
1623 WARN_ON(hw->mac.type != e1000_i210);
1624 WARN_ON(queue < 0 || queue > 4);
1626 val = rd32(E1000_I210_TXDCTL(queue));
1628 if (prio == TX_QUEUE_PRIO_HIGH)
1629 val |= E1000_TXDCTL_PRIORITY;
1631 val &= ~E1000_TXDCTL_PRIORITY;
1633 wr32(E1000_I210_TXDCTL(queue), val);
1636 static void set_queue_mode(struct e1000_hw *hw, int queue, enum queue_mode mode)
1640 WARN_ON(hw->mac.type != e1000_i210);
1641 WARN_ON(queue < 0 || queue > 1);
1643 val = rd32(E1000_I210_TQAVCC(queue));
1645 if (mode == QUEUE_MODE_STREAM_RESERVATION)
1646 val |= E1000_TQAVCC_QUEUEMODE;
1648 val &= ~E1000_TQAVCC_QUEUEMODE;
1650 wr32(E1000_I210_TQAVCC(queue), val);
1653 static bool is_any_cbs_enabled(struct igb_adapter *adapter)
1657 for (i = 0; i < adapter->num_tx_queues; i++) {
1658 if (adapter->tx_ring[i]->cbs_enable)
1665 static bool is_any_txtime_enabled(struct igb_adapter *adapter)
1669 for (i = 0; i < adapter->num_tx_queues; i++) {
1670 if (adapter->tx_ring[i]->launchtime_enable)
1678 * igb_config_tx_modes - Configure "Qav Tx mode" features on igb
1679 * @adapter: pointer to adapter struct
1680 * @queue: queue number
1682 * Configure CBS and Launchtime for a given hardware queue.
1683 * Parameters are retrieved from the correct Tx ring, so
1684 * igb_save_cbs_params() and igb_save_txtime_params() should be used
1685 * for setting those correctly prior to this function being called.
1687 static void igb_config_tx_modes(struct igb_adapter *adapter, int queue)
1689 struct net_device *netdev = adapter->netdev;
1690 struct e1000_hw *hw = &adapter->hw;
1691 struct igb_ring *ring;
1692 u32 tqavcc, tqavctrl;
1695 WARN_ON(hw->mac.type != e1000_i210);
1696 WARN_ON(queue < 0 || queue > 1);
1697 ring = adapter->tx_ring[queue];
1699 /* If any of the Qav features is enabled, configure queues as SR and
1700 * with HIGH PRIO. If none is, then configure them with LOW PRIO and
1703 if (ring->cbs_enable || ring->launchtime_enable) {
1704 set_tx_desc_fetch_prio(hw, queue, TX_QUEUE_PRIO_HIGH);
1705 set_queue_mode(hw, queue, QUEUE_MODE_STREAM_RESERVATION);
1707 set_tx_desc_fetch_prio(hw, queue, TX_QUEUE_PRIO_LOW);
1708 set_queue_mode(hw, queue, QUEUE_MODE_STRICT_PRIORITY);
1711 /* If CBS is enabled, set DataTranARB and config its parameters. */
1712 if (ring->cbs_enable || queue == 0) {
1713 /* i210 does not allow the queue 0 to be in the Strict
1714 * Priority mode while the Qav mode is enabled, so,
1715 * instead of disabling strict priority mode, we give
1716 * queue 0 the maximum of credits possible.
1718 * See section 8.12.19 of the i210 datasheet, "Note:
1719 * Queue0 QueueMode must be set to 1b when
1720 * TransmitMode is set to Qav."
1722 if (queue == 0 && !ring->cbs_enable) {
1723 /* max "linkspeed" idleslope in kbps */
1724 ring->idleslope = 1000000;
1725 ring->hicredit = ETH_FRAME_LEN;
1728 /* Always set data transfer arbitration to credit-based
1729 * shaper algorithm on TQAVCTRL if CBS is enabled for any of
1732 tqavctrl = rd32(E1000_I210_TQAVCTRL);
1733 tqavctrl |= E1000_TQAVCTRL_DATATRANARB;
1734 wr32(E1000_I210_TQAVCTRL, tqavctrl);
1736 /* According to i210 datasheet section 7.2.7.7, we should set
1737 * the 'idleSlope' field from TQAVCC register following the
1740 * For 100 Mbps link speed:
1742 * value = BW * 0x7735 * 0.2 (E1)
1744 * For 1000Mbps link speed:
1746 * value = BW * 0x7735 * 2 (E2)
1748 * E1 and E2 can be merged into one equation as shown below.
1749 * Note that 'link-speed' is in Mbps.
1751 * value = BW * 0x7735 * 2 * link-speed
1752 * -------------- (E3)
1755 * 'BW' is the percentage bandwidth out of full link speed
1756 * which can be found with the following equation. Note that
1757 * idleSlope here is the parameter from this function which
1761 * ----------------- (E4)
1764 * That said, we can come up with a generic equation to
1765 * calculate the value we should set it TQAVCC register by
1766 * replacing 'BW' in E3 by E4. The resulting equation is:
1768 * value = idleSlope * 0x7735 * 2 * link-speed
1769 * ----------------- -------------- (E5)
1770 * link-speed * 1000 1000
1772 * 'link-speed' is present in both sides of the fraction so
1773 * it is canceled out. The final equation is the following:
1775 * value = idleSlope * 61034
1776 * ----------------- (E6)
1779 * NOTE: For i210, given the above, we can see that idleslope
1780 * is represented in 16.38431 kbps units by the value at
1781 * the TQAVCC register (1Gbps / 61034), which reduces
1782 * the granularity for idleslope increments.
1783 * For instance, if you want to configure a 2576kbps
1784 * idleslope, the value to be written on the register
1785 * would have to be 157.23. If rounded down, you end
1786 * up with less bandwidth available than originally
1787 * required (~2572 kbps). If rounded up, you end up
1788 * with a higher bandwidth (~2589 kbps). Below the
1789 * approach we take is to always round up the
1790 * calculated value, so the resulting bandwidth might
1791 * be slightly higher for some configurations.
1793 value = DIV_ROUND_UP_ULL(ring->idleslope * 61034ULL, 1000000);
1795 tqavcc = rd32(E1000_I210_TQAVCC(queue));
1796 tqavcc &= ~E1000_TQAVCC_IDLESLOPE_MASK;
1798 wr32(E1000_I210_TQAVCC(queue), tqavcc);
1800 wr32(E1000_I210_TQAVHC(queue),
1801 0x80000000 + ring->hicredit * 0x7735);
1804 /* Set idleSlope to zero. */
1805 tqavcc = rd32(E1000_I210_TQAVCC(queue));
1806 tqavcc &= ~E1000_TQAVCC_IDLESLOPE_MASK;
1807 wr32(E1000_I210_TQAVCC(queue), tqavcc);
1809 /* Set hiCredit to zero. */
1810 wr32(E1000_I210_TQAVHC(queue), 0);
1812 /* If CBS is not enabled for any queues anymore, then return to
1813 * the default state of Data Transmission Arbitration on
1816 if (!is_any_cbs_enabled(adapter)) {
1817 tqavctrl = rd32(E1000_I210_TQAVCTRL);
1818 tqavctrl &= ~E1000_TQAVCTRL_DATATRANARB;
1819 wr32(E1000_I210_TQAVCTRL, tqavctrl);
1823 /* If LaunchTime is enabled, set DataTranTIM. */
1824 if (ring->launchtime_enable) {
1825 /* Always set DataTranTIM on TQAVCTRL if LaunchTime is enabled
1826 * for any of the SR queues, and configure fetchtime delta.
1828 * - LaunchTime will be enabled for all SR queues.
1829 * - A fixed offset can be added relative to the launch
1830 * time of all packets if configured at reg LAUNCH_OS0.
1831 * We are keeping it as 0 for now (default value).
1833 tqavctrl = rd32(E1000_I210_TQAVCTRL);
1834 tqavctrl |= E1000_TQAVCTRL_DATATRANTIM |
1835 E1000_TQAVCTRL_FETCHTIME_DELTA;
1836 wr32(E1000_I210_TQAVCTRL, tqavctrl);
1838 /* If Launchtime is not enabled for any SR queues anymore,
1839 * then clear DataTranTIM on TQAVCTRL and clear fetchtime delta,
1840 * effectively disabling Launchtime.
1842 if (!is_any_txtime_enabled(adapter)) {
1843 tqavctrl = rd32(E1000_I210_TQAVCTRL);
1844 tqavctrl &= ~E1000_TQAVCTRL_DATATRANTIM;
1845 tqavctrl &= ~E1000_TQAVCTRL_FETCHTIME_DELTA;
1846 wr32(E1000_I210_TQAVCTRL, tqavctrl);
1850 /* XXX: In i210 controller the sendSlope and loCredit parameters from
1851 * CBS are not configurable by software so we don't do any 'controller
1852 * configuration' in respect to these parameters.
1855 netdev_dbg(netdev, "Qav Tx mode: cbs %s, launchtime %s, queue %d idleslope %d sendslope %d hiCredit %d locredit %d\n",
1856 ring->cbs_enable ? "enabled" : "disabled",
1857 ring->launchtime_enable ? "enabled" : "disabled",
1859 ring->idleslope, ring->sendslope,
1860 ring->hicredit, ring->locredit);
1863 static int igb_save_txtime_params(struct igb_adapter *adapter, int queue,
1866 struct igb_ring *ring;
1868 if (queue < 0 || queue > adapter->num_tx_queues)
1871 ring = adapter->tx_ring[queue];
1872 ring->launchtime_enable = enable;
1877 static int igb_save_cbs_params(struct igb_adapter *adapter, int queue,
1878 bool enable, int idleslope, int sendslope,
1879 int hicredit, int locredit)
1881 struct igb_ring *ring;
1883 if (queue < 0 || queue > adapter->num_tx_queues)
1886 ring = adapter->tx_ring[queue];
1888 ring->cbs_enable = enable;
1889 ring->idleslope = idleslope;
1890 ring->sendslope = sendslope;
1891 ring->hicredit = hicredit;
1892 ring->locredit = locredit;
1898 * igb_setup_tx_mode - Switch to/from Qav Tx mode when applicable
1899 * @adapter: pointer to adapter struct
1901 * Configure TQAVCTRL register switching the controller's Tx mode
1902 * if FQTSS mode is enabled or disabled. Additionally, will issue
1903 * a call to igb_config_tx_modes() per queue so any previously saved
1904 * Tx parameters are applied.
1906 static void igb_setup_tx_mode(struct igb_adapter *adapter)
1908 struct net_device *netdev = adapter->netdev;
1909 struct e1000_hw *hw = &adapter->hw;
1912 /* Only i210 controller supports changing the transmission mode. */
1913 if (hw->mac.type != e1000_i210)
1916 if (is_fqtss_enabled(adapter)) {
1919 /* Configure TQAVCTRL register: set transmit mode to 'Qav',
1920 * set data fetch arbitration to 'round robin', set SP_WAIT_SR
1921 * so SP queues wait for SR ones.
1923 val = rd32(E1000_I210_TQAVCTRL);
1924 val |= E1000_TQAVCTRL_XMIT_MODE | E1000_TQAVCTRL_SP_WAIT_SR;
1925 val &= ~E1000_TQAVCTRL_DATAFETCHARB;
1926 wr32(E1000_I210_TQAVCTRL, val);
1928 /* Configure Tx and Rx packet buffers sizes as described in
1929 * i210 datasheet section 7.2.7.7.
1931 val = rd32(E1000_TXPBS);
1932 val &= ~I210_TXPBSIZE_MASK;
1933 val |= I210_TXPBSIZE_PB0_6KB | I210_TXPBSIZE_PB1_6KB |
1934 I210_TXPBSIZE_PB2_6KB | I210_TXPBSIZE_PB3_6KB;
1935 wr32(E1000_TXPBS, val);
1937 val = rd32(E1000_RXPBS);
1938 val &= ~I210_RXPBSIZE_MASK;
1939 val |= I210_RXPBSIZE_PB_30KB;
1940 wr32(E1000_RXPBS, val);
1942 /* Section 8.12.9 states that MAX_TPKT_SIZE from DTXMXPKTSZ
1943 * register should not exceed the buffer size programmed in
1944 * TXPBS. The smallest buffer size programmed in TXPBS is 4kB
1945 * so according to the datasheet we should set MAX_TPKT_SIZE to
1948 * However, when we do so, no frame from queue 2 and 3 are
1949 * transmitted. It seems the MAX_TPKT_SIZE should not be great
1950 * or _equal_ to the buffer size programmed in TXPBS. For this
1951 * reason, we set MAX_ TPKT_SIZE to (4kB - 1) / 64.
1953 val = (4096 - 1) / 64;
1954 wr32(E1000_I210_DTXMXPKTSZ, val);
1956 /* Since FQTSS mode is enabled, apply any CBS configuration
1957 * previously set. If no previous CBS configuration has been
1958 * done, then the initial configuration is applied, which means
1961 max_queue = (adapter->num_tx_queues < I210_SR_QUEUES_NUM) ?
1962 adapter->num_tx_queues : I210_SR_QUEUES_NUM;
1964 for (i = 0; i < max_queue; i++) {
1965 igb_config_tx_modes(adapter, i);
1968 wr32(E1000_RXPBS, I210_RXPBSIZE_DEFAULT);
1969 wr32(E1000_TXPBS, I210_TXPBSIZE_DEFAULT);
1970 wr32(E1000_I210_DTXMXPKTSZ, I210_DTXMXPKTSZ_DEFAULT);
1972 val = rd32(E1000_I210_TQAVCTRL);
1973 /* According to Section 8.12.21, the other flags we've set when
1974 * enabling FQTSS are not relevant when disabling FQTSS so we
1975 * don't set they here.
1977 val &= ~E1000_TQAVCTRL_XMIT_MODE;
1978 wr32(E1000_I210_TQAVCTRL, val);
1981 netdev_dbg(netdev, "FQTSS %s\n", (is_fqtss_enabled(adapter)) ?
1982 "enabled" : "disabled");
1986 * igb_configure - configure the hardware for RX and TX
1987 * @adapter: private board structure
1989 static void igb_configure(struct igb_adapter *adapter)
1991 struct net_device *netdev = adapter->netdev;
1994 igb_get_hw_control(adapter);
1995 igb_set_rx_mode(netdev);
1996 igb_setup_tx_mode(adapter);
1998 igb_restore_vlan(adapter);
2000 igb_setup_tctl(adapter);
2001 igb_setup_mrqc(adapter);
2002 igb_setup_rctl(adapter);
2004 igb_nfc_filter_restore(adapter);
2005 igb_configure_tx(adapter);
2006 igb_configure_rx(adapter);
2008 igb_rx_fifo_flush_82575(&adapter->hw);
2010 /* call igb_desc_unused which always leaves
2011 * at least 1 descriptor unused to make sure
2012 * next_to_use != next_to_clean
2014 for (i = 0; i < adapter->num_rx_queues; i++) {
2015 struct igb_ring *ring = adapter->rx_ring[i];
2016 igb_alloc_rx_buffers(ring, igb_desc_unused(ring));
2021 * igb_power_up_link - Power up the phy/serdes link
2022 * @adapter: address of board private structure
2024 void igb_power_up_link(struct igb_adapter *adapter)
2026 igb_reset_phy(&adapter->hw);
2028 if (adapter->hw.phy.media_type == e1000_media_type_copper)
2029 igb_power_up_phy_copper(&adapter->hw);
2031 igb_power_up_serdes_link_82575(&adapter->hw);
2033 igb_setup_link(&adapter->hw);
2037 * igb_power_down_link - Power down the phy/serdes link
2038 * @adapter: address of board private structure
2040 static void igb_power_down_link(struct igb_adapter *adapter)
2042 if (adapter->hw.phy.media_type == e1000_media_type_copper)
2043 igb_power_down_phy_copper_82575(&adapter->hw);
2045 igb_shutdown_serdes_link_82575(&adapter->hw);
2049 * igb_check_swap_media - Detect and switch function for Media Auto Sense
2050 * @adapter: address of the board private structure
2052 static void igb_check_swap_media(struct igb_adapter *adapter)
2054 struct e1000_hw *hw = &adapter->hw;
2055 u32 ctrl_ext, connsw;
2056 bool swap_now = false;
2058 ctrl_ext = rd32(E1000_CTRL_EXT);
2059 connsw = rd32(E1000_CONNSW);
2061 /* need to live swap if current media is copper and we have fiber/serdes
2065 if ((hw->phy.media_type == e1000_media_type_copper) &&
2066 (!(connsw & E1000_CONNSW_AUTOSENSE_EN))) {
2068 } else if ((hw->phy.media_type != e1000_media_type_copper) &&
2069 !(connsw & E1000_CONNSW_SERDESD)) {
2070 /* copper signal takes time to appear */
2071 if (adapter->copper_tries < 4) {
2072 adapter->copper_tries++;
2073 connsw |= E1000_CONNSW_AUTOSENSE_CONF;
2074 wr32(E1000_CONNSW, connsw);
2077 adapter->copper_tries = 0;
2078 if ((connsw & E1000_CONNSW_PHYSD) &&
2079 (!(connsw & E1000_CONNSW_PHY_PDN))) {
2081 connsw &= ~E1000_CONNSW_AUTOSENSE_CONF;
2082 wr32(E1000_CONNSW, connsw);
2090 switch (hw->phy.media_type) {
2091 case e1000_media_type_copper:
2092 netdev_info(adapter->netdev,
2093 "MAS: changing media to fiber/serdes\n");
2095 E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
2096 adapter->flags |= IGB_FLAG_MEDIA_RESET;
2097 adapter->copper_tries = 0;
2099 case e1000_media_type_internal_serdes:
2100 case e1000_media_type_fiber:
2101 netdev_info(adapter->netdev,
2102 "MAS: changing media to copper\n");
2104 ~E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
2105 adapter->flags |= IGB_FLAG_MEDIA_RESET;
2108 /* shouldn't get here during regular operation */
2109 netdev_err(adapter->netdev,
2110 "AMS: Invalid media type found, returning\n");
2113 wr32(E1000_CTRL_EXT, ctrl_ext);
2117 * igb_up - Open the interface and prepare it to handle traffic
2118 * @adapter: board private structure
2120 int igb_up(struct igb_adapter *adapter)
2122 struct e1000_hw *hw = &adapter->hw;
2125 /* hardware has been reset, we need to reload some things */
2126 igb_configure(adapter);
2128 clear_bit(__IGB_DOWN, &adapter->state);
2130 for (i = 0; i < adapter->num_q_vectors; i++)
2131 napi_enable(&(adapter->q_vector[i]->napi));
2133 if (adapter->flags & IGB_FLAG_HAS_MSIX)
2134 igb_configure_msix(adapter);
2136 igb_assign_vector(adapter->q_vector[0], 0);
2138 /* Clear any pending interrupts. */
2141 igb_irq_enable(adapter);
2143 /* notify VFs that reset has been completed */
2144 if (adapter->vfs_allocated_count) {
2145 u32 reg_data = rd32(E1000_CTRL_EXT);
2147 reg_data |= E1000_CTRL_EXT_PFRSTD;
2148 wr32(E1000_CTRL_EXT, reg_data);
2151 netif_tx_start_all_queues(adapter->netdev);
2153 /* start the watchdog. */
2154 hw->mac.get_link_status = 1;
2155 schedule_work(&adapter->watchdog_task);
2157 if ((adapter->flags & IGB_FLAG_EEE) &&
2158 (!hw->dev_spec._82575.eee_disable))
2159 adapter->eee_advert = MDIO_EEE_100TX | MDIO_EEE_1000T;
2164 void igb_down(struct igb_adapter *adapter)
2166 struct net_device *netdev = adapter->netdev;
2167 struct e1000_hw *hw = &adapter->hw;
2171 /* signal that we're down so the interrupt handler does not
2172 * reschedule our watchdog timer
2174 set_bit(__IGB_DOWN, &adapter->state);
2176 /* disable receives in the hardware */
2177 rctl = rd32(E1000_RCTL);
2178 wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
2179 /* flush and sleep below */
2181 igb_nfc_filter_exit(adapter);
2183 netif_carrier_off(netdev);
2184 netif_tx_stop_all_queues(netdev);
2186 /* disable transmits in the hardware */
2187 tctl = rd32(E1000_TCTL);
2188 tctl &= ~E1000_TCTL_EN;
2189 wr32(E1000_TCTL, tctl);
2190 /* flush both disables and wait for them to finish */
2192 usleep_range(10000, 11000);
2194 igb_irq_disable(adapter);
2196 adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
2198 for (i = 0; i < adapter->num_q_vectors; i++) {
2199 if (adapter->q_vector[i]) {
2200 napi_synchronize(&adapter->q_vector[i]->napi);
2201 napi_disable(&adapter->q_vector[i]->napi);
2205 del_timer_sync(&adapter->watchdog_timer);
2206 del_timer_sync(&adapter->phy_info_timer);
2208 /* record the stats before reset*/
2209 spin_lock(&adapter->stats64_lock);
2210 igb_update_stats(adapter);
2211 spin_unlock(&adapter->stats64_lock);
2213 adapter->link_speed = 0;
2214 adapter->link_duplex = 0;
2216 if (!pci_channel_offline(adapter->pdev))
2219 /* clear VLAN promisc flag so VFTA will be updated if necessary */
2220 adapter->flags &= ~IGB_FLAG_VLAN_PROMISC;
2222 igb_clean_all_tx_rings(adapter);
2223 igb_clean_all_rx_rings(adapter);
2224 #ifdef CONFIG_IGB_DCA
2226 /* since we reset the hardware DCA settings were cleared */
2227 igb_setup_dca(adapter);
2231 void igb_reinit_locked(struct igb_adapter *adapter)
2233 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
2234 usleep_range(1000, 2000);
2237 clear_bit(__IGB_RESETTING, &adapter->state);
2240 /** igb_enable_mas - Media Autosense re-enable after swap
2242 * @adapter: adapter struct
2244 static void igb_enable_mas(struct igb_adapter *adapter)
2246 struct e1000_hw *hw = &adapter->hw;
2247 u32 connsw = rd32(E1000_CONNSW);
2249 /* configure for SerDes media detect */
2250 if ((hw->phy.media_type == e1000_media_type_copper) &&
2251 (!(connsw & E1000_CONNSW_SERDESD))) {
2252 connsw |= E1000_CONNSW_ENRGSRC;
2253 connsw |= E1000_CONNSW_AUTOSENSE_EN;
2254 wr32(E1000_CONNSW, connsw);
2259 #ifdef CONFIG_IGB_HWMON
2261 * igb_set_i2c_bb - Init I2C interface
2262 * @hw: pointer to hardware structure
2264 static void igb_set_i2c_bb(struct e1000_hw *hw)
2269 ctrl_ext = rd32(E1000_CTRL_EXT);
2270 ctrl_ext |= E1000_CTRL_I2C_ENA;
2271 wr32(E1000_CTRL_EXT, ctrl_ext);
2274 i2cctl = rd32(E1000_I2CPARAMS);
2275 i2cctl |= E1000_I2CBB_EN
2276 | E1000_I2C_CLK_OE_N
2277 | E1000_I2C_DATA_OE_N;
2278 wr32(E1000_I2CPARAMS, i2cctl);
2283 void igb_reset(struct igb_adapter *adapter)
2285 struct pci_dev *pdev = adapter->pdev;
2286 struct e1000_hw *hw = &adapter->hw;
2287 struct e1000_mac_info *mac = &hw->mac;
2288 struct e1000_fc_info *fc = &hw->fc;
2291 /* Repartition Pba for greater than 9k mtu
2292 * To take effect CTRL.RST is required.
2294 switch (mac->type) {
2298 pba = rd32(E1000_RXPBS);
2299 pba = igb_rxpbs_adjust_82580(pba);
2302 pba = rd32(E1000_RXPBS);
2303 pba &= E1000_RXPBS_SIZE_MASK_82576;
2309 pba = E1000_PBA_34K;
2313 if (mac->type == e1000_82575) {
2314 u32 min_rx_space, min_tx_space, needed_tx_space;
2316 /* write Rx PBA so that hardware can report correct Tx PBA */
2317 wr32(E1000_PBA, pba);
2319 /* To maintain wire speed transmits, the Tx FIFO should be
2320 * large enough to accommodate two full transmit packets,
2321 * rounded up to the next 1KB and expressed in KB. Likewise,
2322 * the Rx FIFO should be large enough to accommodate at least
2323 * one full receive packet and is similarly rounded up and
2326 min_rx_space = DIV_ROUND_UP(MAX_JUMBO_FRAME_SIZE, 1024);
2328 /* The Tx FIFO also stores 16 bytes of information about the Tx
2329 * but don't include Ethernet FCS because hardware appends it.
2330 * We only need to round down to the nearest 512 byte block
2331 * count since the value we care about is 2 frames, not 1.
2333 min_tx_space = adapter->max_frame_size;
2334 min_tx_space += sizeof(union e1000_adv_tx_desc) - ETH_FCS_LEN;
2335 min_tx_space = DIV_ROUND_UP(min_tx_space, 512);
2337 /* upper 16 bits has Tx packet buffer allocation size in KB */
2338 needed_tx_space = min_tx_space - (rd32(E1000_PBA) >> 16);
2340 /* If current Tx allocation is less than the min Tx FIFO size,
2341 * and the min Tx FIFO size is less than the current Rx FIFO
2342 * allocation, take space away from current Rx allocation.
2344 if (needed_tx_space < pba) {
2345 pba -= needed_tx_space;
2347 /* if short on Rx space, Rx wins and must trump Tx
2350 if (pba < min_rx_space)
2354 /* adjust PBA for jumbo frames */
2355 wr32(E1000_PBA, pba);
2358 /* flow control settings
2359 * The high water mark must be low enough to fit one full frame
2360 * after transmitting the pause frame. As such we must have enough
2361 * space to allow for us to complete our current transmit and then
2362 * receive the frame that is in progress from the link partner.
2364 * - the full Rx FIFO size minus one full Tx plus one full Rx frame
2366 hwm = (pba << 10) - (adapter->max_frame_size + MAX_JUMBO_FRAME_SIZE);
2368 fc->high_water = hwm & 0xFFFFFFF0; /* 16-byte granularity */
2369 fc->low_water = fc->high_water - 16;
2370 fc->pause_time = 0xFFFF;
2372 fc->current_mode = fc->requested_mode;
2374 /* disable receive for all VFs and wait one second */
2375 if (adapter->vfs_allocated_count) {
2378 for (i = 0 ; i < adapter->vfs_allocated_count; i++)
2379 adapter->vf_data[i].flags &= IGB_VF_FLAG_PF_SET_MAC;
2381 /* ping all the active vfs to let them know we are going down */
2382 igb_ping_all_vfs(adapter);
2384 /* disable transmits and receives */
2385 wr32(E1000_VFRE, 0);
2386 wr32(E1000_VFTE, 0);
2389 /* Allow time for pending master requests to run */
2390 hw->mac.ops.reset_hw(hw);
2393 if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
2394 /* need to resetup here after media swap */
2395 adapter->ei.get_invariants(hw);
2396 adapter->flags &= ~IGB_FLAG_MEDIA_RESET;
2398 if ((mac->type == e1000_82575 || mac->type == e1000_i350) &&
2399 (adapter->flags & IGB_FLAG_MAS_ENABLE)) {
2400 igb_enable_mas(adapter);
2402 if (hw->mac.ops.init_hw(hw))
2403 dev_err(&pdev->dev, "Hardware Error\n");
2405 /* RAR registers were cleared during init_hw, clear mac table */
2406 igb_flush_mac_table(adapter);
2407 __dev_uc_unsync(adapter->netdev, NULL);
2409 /* Recover default RAR entry */
2410 igb_set_default_mac_filter(adapter);
2412 /* Flow control settings reset on hardware reset, so guarantee flow
2413 * control is off when forcing speed.
2415 if (!hw->mac.autoneg)
2416 igb_force_mac_fc(hw);
2418 igb_init_dmac(adapter, pba);
2419 #ifdef CONFIG_IGB_HWMON
2420 /* Re-initialize the thermal sensor on i350 devices. */
2421 if (!test_bit(__IGB_DOWN, &adapter->state)) {
2422 if (mac->type == e1000_i350 && hw->bus.func == 0) {
2423 /* If present, re-initialize the external thermal sensor
2428 mac->ops.init_thermal_sensor_thresh(hw);
2432 /* Re-establish EEE setting */
2433 if (hw->phy.media_type == e1000_media_type_copper) {
2434 switch (mac->type) {
2438 igb_set_eee_i350(hw, true, true);
2441 igb_set_eee_i354(hw, true, true);
2447 if (!netif_running(adapter->netdev))
2448 igb_power_down_link(adapter);
2450 igb_update_mng_vlan(adapter);
2452 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2453 wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
2455 /* Re-enable PTP, where applicable. */
2456 if (adapter->ptp_flags & IGB_PTP_ENABLED)
2457 igb_ptp_reset(adapter);
2459 igb_get_phy_info(hw);
2462 static netdev_features_t igb_fix_features(struct net_device *netdev,
2463 netdev_features_t features)
2465 /* Since there is no support for separate Rx/Tx vlan accel
2466 * enable/disable make sure Tx flag is always in same state as Rx.
2468 if (features & NETIF_F_HW_VLAN_CTAG_RX)
2469 features |= NETIF_F_HW_VLAN_CTAG_TX;
2471 features &= ~NETIF_F_HW_VLAN_CTAG_TX;
2476 static int igb_set_features(struct net_device *netdev,
2477 netdev_features_t features)
2479 netdev_features_t changed = netdev->features ^ features;
2480 struct igb_adapter *adapter = netdev_priv(netdev);
2482 if (changed & NETIF_F_HW_VLAN_CTAG_RX)
2483 igb_vlan_mode(netdev, features);
2485 if (!(changed & (NETIF_F_RXALL | NETIF_F_NTUPLE)))
2488 if (!(features & NETIF_F_NTUPLE)) {
2489 struct hlist_node *node2;
2490 struct igb_nfc_filter *rule;
2492 spin_lock(&adapter->nfc_lock);
2493 hlist_for_each_entry_safe(rule, node2,
2494 &adapter->nfc_filter_list, nfc_node) {
2495 igb_erase_filter(adapter, rule);
2496 hlist_del(&rule->nfc_node);
2499 spin_unlock(&adapter->nfc_lock);
2500 adapter->nfc_filter_count = 0;
2503 netdev->features = features;
2505 if (netif_running(netdev))
2506 igb_reinit_locked(adapter);
2513 static int igb_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
2514 struct net_device *dev,
2515 const unsigned char *addr, u16 vid,
2517 struct netlink_ext_ack *extack)
2519 /* guarantee we can provide a unique filter for the unicast address */
2520 if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
2521 struct igb_adapter *adapter = netdev_priv(dev);
2522 int vfn = adapter->vfs_allocated_count;
2524 if (netdev_uc_count(dev) >= igb_available_rars(adapter, vfn))
2528 return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags);
2531 #define IGB_MAX_MAC_HDR_LEN 127
2532 #define IGB_MAX_NETWORK_HDR_LEN 511
2534 static netdev_features_t
2535 igb_features_check(struct sk_buff *skb, struct net_device *dev,
2536 netdev_features_t features)
2538 unsigned int network_hdr_len, mac_hdr_len;
2540 /* Make certain the headers can be described by a context descriptor */
2541 mac_hdr_len = skb_network_header(skb) - skb->data;
2542 if (unlikely(mac_hdr_len > IGB_MAX_MAC_HDR_LEN))
2543 return features & ~(NETIF_F_HW_CSUM |
2545 NETIF_F_GSO_UDP_L4 |
2546 NETIF_F_HW_VLAN_CTAG_TX |
2550 network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
2551 if (unlikely(network_hdr_len > IGB_MAX_NETWORK_HDR_LEN))
2552 return features & ~(NETIF_F_HW_CSUM |
2554 NETIF_F_GSO_UDP_L4 |
2558 /* We can only support IPV4 TSO in tunnels if we can mangle the
2559 * inner IP ID field, so strip TSO if MANGLEID is not supported.
2561 if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
2562 features &= ~NETIF_F_TSO;
2567 static void igb_offload_apply(struct igb_adapter *adapter, s32 queue)
2569 if (!is_fqtss_enabled(adapter)) {
2570 enable_fqtss(adapter, true);
2574 igb_config_tx_modes(adapter, queue);
2576 if (!is_any_cbs_enabled(adapter) && !is_any_txtime_enabled(adapter))
2577 enable_fqtss(adapter, false);
2580 static int igb_offload_cbs(struct igb_adapter *adapter,
2581 struct tc_cbs_qopt_offload *qopt)
2583 struct e1000_hw *hw = &adapter->hw;
2586 /* CBS offloading is only supported by i210 controller. */
2587 if (hw->mac.type != e1000_i210)
2590 /* CBS offloading is only supported by queue 0 and queue 1. */
2591 if (qopt->queue < 0 || qopt->queue > 1)
2594 err = igb_save_cbs_params(adapter, qopt->queue, qopt->enable,
2595 qopt->idleslope, qopt->sendslope,
2596 qopt->hicredit, qopt->locredit);
2600 igb_offload_apply(adapter, qopt->queue);
2605 #define ETHER_TYPE_FULL_MASK ((__force __be16)~0)
2606 #define VLAN_PRIO_FULL_MASK (0x07)
2608 static int igb_parse_cls_flower(struct igb_adapter *adapter,
2609 struct flow_cls_offload *f,
2611 struct igb_nfc_filter *input)
2613 struct flow_rule *rule = flow_cls_offload_flow_rule(f);
2614 struct flow_dissector *dissector = rule->match.dissector;
2615 struct netlink_ext_ack *extack = f->common.extack;
2617 if (dissector->used_keys &
2618 ~(BIT_ULL(FLOW_DISSECTOR_KEY_BASIC) |
2619 BIT_ULL(FLOW_DISSECTOR_KEY_CONTROL) |
2620 BIT_ULL(FLOW_DISSECTOR_KEY_ETH_ADDRS) |
2621 BIT_ULL(FLOW_DISSECTOR_KEY_VLAN))) {
2622 NL_SET_ERR_MSG_MOD(extack,
2623 "Unsupported key used, only BASIC, CONTROL, ETH_ADDRS and VLAN are supported");
2627 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ETH_ADDRS)) {
2628 struct flow_match_eth_addrs match;
2630 flow_rule_match_eth_addrs(rule, &match);
2631 if (!is_zero_ether_addr(match.mask->dst)) {
2632 if (!is_broadcast_ether_addr(match.mask->dst)) {
2633 NL_SET_ERR_MSG_MOD(extack, "Only full masks are supported for destination MAC address");
2637 input->filter.match_flags |=
2638 IGB_FILTER_FLAG_DST_MAC_ADDR;
2639 ether_addr_copy(input->filter.dst_addr, match.key->dst);
2642 if (!is_zero_ether_addr(match.mask->src)) {
2643 if (!is_broadcast_ether_addr(match.mask->src)) {
2644 NL_SET_ERR_MSG_MOD(extack, "Only full masks are supported for source MAC address");
2648 input->filter.match_flags |=
2649 IGB_FILTER_FLAG_SRC_MAC_ADDR;
2650 ether_addr_copy(input->filter.src_addr, match.key->src);
2654 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_BASIC)) {
2655 struct flow_match_basic match;
2657 flow_rule_match_basic(rule, &match);
2658 if (match.mask->n_proto) {
2659 if (match.mask->n_proto != ETHER_TYPE_FULL_MASK) {
2660 NL_SET_ERR_MSG_MOD(extack, "Only full mask is supported for EtherType filter");
2664 input->filter.match_flags |= IGB_FILTER_FLAG_ETHER_TYPE;
2665 input->filter.etype = match.key->n_proto;
2669 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_VLAN)) {
2670 struct flow_match_vlan match;
2672 flow_rule_match_vlan(rule, &match);
2673 if (match.mask->vlan_priority) {
2674 if (match.mask->vlan_priority != VLAN_PRIO_FULL_MASK) {
2675 NL_SET_ERR_MSG_MOD(extack, "Only full mask is supported for VLAN priority");
2679 input->filter.match_flags |= IGB_FILTER_FLAG_VLAN_TCI;
2680 input->filter.vlan_tci =
2681 (__force __be16)match.key->vlan_priority;
2685 input->action = traffic_class;
2686 input->cookie = f->cookie;
2691 static int igb_configure_clsflower(struct igb_adapter *adapter,
2692 struct flow_cls_offload *cls_flower)
2694 struct netlink_ext_ack *extack = cls_flower->common.extack;
2695 struct igb_nfc_filter *filter, *f;
2698 tc = tc_classid_to_hwtc(adapter->netdev, cls_flower->classid);
2700 NL_SET_ERR_MSG_MOD(extack, "Invalid traffic class");
2704 filter = kzalloc(sizeof(*filter), GFP_KERNEL);
2708 err = igb_parse_cls_flower(adapter, cls_flower, tc, filter);
2712 spin_lock(&adapter->nfc_lock);
2714 hlist_for_each_entry(f, &adapter->nfc_filter_list, nfc_node) {
2715 if (!memcmp(&f->filter, &filter->filter, sizeof(f->filter))) {
2717 NL_SET_ERR_MSG_MOD(extack,
2718 "This filter is already set in ethtool");
2723 hlist_for_each_entry(f, &adapter->cls_flower_list, nfc_node) {
2724 if (!memcmp(&f->filter, &filter->filter, sizeof(f->filter))) {
2726 NL_SET_ERR_MSG_MOD(extack,
2727 "This filter is already set in cls_flower");
2732 err = igb_add_filter(adapter, filter);
2734 NL_SET_ERR_MSG_MOD(extack, "Could not add filter to the adapter");
2738 hlist_add_head(&filter->nfc_node, &adapter->cls_flower_list);
2740 spin_unlock(&adapter->nfc_lock);
2745 spin_unlock(&adapter->nfc_lock);
2753 static int igb_delete_clsflower(struct igb_adapter *adapter,
2754 struct flow_cls_offload *cls_flower)
2756 struct igb_nfc_filter *filter;
2759 spin_lock(&adapter->nfc_lock);
2761 hlist_for_each_entry(filter, &adapter->cls_flower_list, nfc_node)
2762 if (filter->cookie == cls_flower->cookie)
2770 err = igb_erase_filter(adapter, filter);
2774 hlist_del(&filter->nfc_node);
2778 spin_unlock(&adapter->nfc_lock);
2783 static int igb_setup_tc_cls_flower(struct igb_adapter *adapter,
2784 struct flow_cls_offload *cls_flower)
2786 switch (cls_flower->command) {
2787 case FLOW_CLS_REPLACE:
2788 return igb_configure_clsflower(adapter, cls_flower);
2789 case FLOW_CLS_DESTROY:
2790 return igb_delete_clsflower(adapter, cls_flower);
2791 case FLOW_CLS_STATS:
2798 static int igb_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
2801 struct igb_adapter *adapter = cb_priv;
2803 if (!tc_cls_can_offload_and_chain0(adapter->netdev, type_data))
2807 case TC_SETUP_CLSFLOWER:
2808 return igb_setup_tc_cls_flower(adapter, type_data);
2815 static int igb_offload_txtime(struct igb_adapter *adapter,
2816 struct tc_etf_qopt_offload *qopt)
2818 struct e1000_hw *hw = &adapter->hw;
2821 /* Launchtime offloading is only supported by i210 controller. */
2822 if (hw->mac.type != e1000_i210)
2825 /* Launchtime offloading is only supported by queues 0 and 1. */
2826 if (qopt->queue < 0 || qopt->queue > 1)
2829 err = igb_save_txtime_params(adapter, qopt->queue, qopt->enable);
2833 igb_offload_apply(adapter, qopt->queue);
2838 static int igb_tc_query_caps(struct igb_adapter *adapter,
2839 struct tc_query_caps_base *base)
2841 switch (base->type) {
2842 case TC_SETUP_QDISC_TAPRIO: {
2843 struct tc_taprio_caps *caps = base->caps;
2845 caps->broken_mqprio = true;
2854 static LIST_HEAD(igb_block_cb_list);
2856 static int igb_setup_tc(struct net_device *dev, enum tc_setup_type type,
2859 struct igb_adapter *adapter = netdev_priv(dev);
2863 return igb_tc_query_caps(adapter, type_data);
2864 case TC_SETUP_QDISC_CBS:
2865 return igb_offload_cbs(adapter, type_data);
2866 case TC_SETUP_BLOCK:
2867 return flow_block_cb_setup_simple(type_data,
2869 igb_setup_tc_block_cb,
2870 adapter, adapter, true);
2872 case TC_SETUP_QDISC_ETF:
2873 return igb_offload_txtime(adapter, type_data);
2880 static int igb_xdp_setup(struct net_device *dev, struct netdev_bpf *bpf)
2882 int i, frame_size = dev->mtu + IGB_ETH_PKT_HDR_PAD;
2883 struct igb_adapter *adapter = netdev_priv(dev);
2884 struct bpf_prog *prog = bpf->prog, *old_prog;
2885 bool running = netif_running(dev);
2888 /* verify igb ring attributes are sufficient for XDP */
2889 for (i = 0; i < adapter->num_rx_queues; i++) {
2890 struct igb_ring *ring = adapter->rx_ring[i];
2892 if (frame_size > igb_rx_bufsz(ring)) {
2893 NL_SET_ERR_MSG_MOD(bpf->extack,
2894 "The RX buffer size is too small for the frame size");
2895 netdev_warn(dev, "XDP RX buffer size %d is too small for the frame size %d\n",
2896 igb_rx_bufsz(ring), frame_size);
2901 old_prog = xchg(&adapter->xdp_prog, prog);
2902 need_reset = (!!prog != !!old_prog);
2904 /* device is up and bpf is added/removed, must setup the RX queues */
2905 if (need_reset && running) {
2908 for (i = 0; i < adapter->num_rx_queues; i++)
2909 (void)xchg(&adapter->rx_ring[i]->xdp_prog,
2914 bpf_prog_put(old_prog);
2916 /* bpf is just replaced, RXQ and MTU are already setup */
2921 xdp_features_set_redirect_target(dev, true);
2923 xdp_features_clear_redirect_target(dev);
2932 static int igb_xdp(struct net_device *dev, struct netdev_bpf *xdp)
2934 switch (xdp->command) {
2935 case XDP_SETUP_PROG:
2936 return igb_xdp_setup(dev, xdp);
2942 static void igb_xdp_ring_update_tail(struct igb_ring *ring)
2944 /* Force memory writes to complete before letting h/w know there
2945 * are new descriptors to fetch.
2948 writel(ring->next_to_use, ring->tail);
2951 static struct igb_ring *igb_xdp_tx_queue_mapping(struct igb_adapter *adapter)
2953 unsigned int r_idx = smp_processor_id();
2955 if (r_idx >= adapter->num_tx_queues)
2956 r_idx = r_idx % adapter->num_tx_queues;
2958 return adapter->tx_ring[r_idx];
2961 static int igb_xdp_xmit_back(struct igb_adapter *adapter, struct xdp_buff *xdp)
2963 struct xdp_frame *xdpf = xdp_convert_buff_to_frame(xdp);
2964 int cpu = smp_processor_id();
2965 struct igb_ring *tx_ring;
2966 struct netdev_queue *nq;
2969 if (unlikely(!xdpf))
2970 return IGB_XDP_CONSUMED;
2972 /* During program transitions its possible adapter->xdp_prog is assigned
2973 * but ring has not been configured yet. In this case simply abort xmit.
2975 tx_ring = adapter->xdp_prog ? igb_xdp_tx_queue_mapping(adapter) : NULL;
2976 if (unlikely(!tx_ring))
2977 return IGB_XDP_CONSUMED;
2979 nq = txring_txq(tx_ring);
2980 __netif_tx_lock(nq, cpu);
2981 /* Avoid transmit queue timeout since we share it with the slow path */
2982 txq_trans_cond_update(nq);
2983 ret = igb_xmit_xdp_ring(adapter, tx_ring, xdpf);
2984 __netif_tx_unlock(nq);
2989 static int igb_xdp_xmit(struct net_device *dev, int n,
2990 struct xdp_frame **frames, u32 flags)
2992 struct igb_adapter *adapter = netdev_priv(dev);
2993 int cpu = smp_processor_id();
2994 struct igb_ring *tx_ring;
2995 struct netdev_queue *nq;
2999 if (unlikely(test_bit(__IGB_DOWN, &adapter->state)))
3002 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
3005 /* During program transitions its possible adapter->xdp_prog is assigned
3006 * but ring has not been configured yet. In this case simply abort xmit.
3008 tx_ring = adapter->xdp_prog ? igb_xdp_tx_queue_mapping(adapter) : NULL;
3009 if (unlikely(!tx_ring))
3012 nq = txring_txq(tx_ring);
3013 __netif_tx_lock(nq, cpu);
3015 /* Avoid transmit queue timeout since we share it with the slow path */
3016 txq_trans_cond_update(nq);
3018 for (i = 0; i < n; i++) {
3019 struct xdp_frame *xdpf = frames[i];
3022 err = igb_xmit_xdp_ring(adapter, tx_ring, xdpf);
3023 if (err != IGB_XDP_TX)
3028 __netif_tx_unlock(nq);
3030 if (unlikely(flags & XDP_XMIT_FLUSH))
3031 igb_xdp_ring_update_tail(tx_ring);
3036 static const struct net_device_ops igb_netdev_ops = {
3037 .ndo_open = igb_open,
3038 .ndo_stop = igb_close,
3039 .ndo_start_xmit = igb_xmit_frame,
3040 .ndo_get_stats64 = igb_get_stats64,
3041 .ndo_set_rx_mode = igb_set_rx_mode,
3042 .ndo_set_mac_address = igb_set_mac,
3043 .ndo_change_mtu = igb_change_mtu,
3044 .ndo_eth_ioctl = igb_ioctl,
3045 .ndo_tx_timeout = igb_tx_timeout,
3046 .ndo_validate_addr = eth_validate_addr,
3047 .ndo_vlan_rx_add_vid = igb_vlan_rx_add_vid,
3048 .ndo_vlan_rx_kill_vid = igb_vlan_rx_kill_vid,
3049 .ndo_set_vf_mac = igb_ndo_set_vf_mac,
3050 .ndo_set_vf_vlan = igb_ndo_set_vf_vlan,
3051 .ndo_set_vf_rate = igb_ndo_set_vf_bw,
3052 .ndo_set_vf_spoofchk = igb_ndo_set_vf_spoofchk,
3053 .ndo_set_vf_trust = igb_ndo_set_vf_trust,
3054 .ndo_get_vf_config = igb_ndo_get_vf_config,
3055 .ndo_fix_features = igb_fix_features,
3056 .ndo_set_features = igb_set_features,
3057 .ndo_fdb_add = igb_ndo_fdb_add,
3058 .ndo_features_check = igb_features_check,
3059 .ndo_setup_tc = igb_setup_tc,
3061 .ndo_xdp_xmit = igb_xdp_xmit,
3065 * igb_set_fw_version - Configure version string for ethtool
3066 * @adapter: adapter struct
3068 void igb_set_fw_version(struct igb_adapter *adapter)
3070 struct e1000_hw *hw = &adapter->hw;
3071 struct e1000_fw_version fw;
3073 igb_get_fw_version(hw, &fw);
3075 switch (hw->mac.type) {
3078 if (!(igb_get_flash_presence_i210(hw))) {
3079 snprintf(adapter->fw_version,
3080 sizeof(adapter->fw_version),
3082 fw.invm_major, fw.invm_minor,
3088 /* if option is rom valid, display its version too */
3090 snprintf(adapter->fw_version,
3091 sizeof(adapter->fw_version),
3092 "%d.%d, 0x%08x, %d.%d.%d",
3093 fw.eep_major, fw.eep_minor, fw.etrack_id,
3094 fw.or_major, fw.or_build, fw.or_patch);
3096 } else if (fw.etrack_id != 0X0000) {
3097 snprintf(adapter->fw_version,
3098 sizeof(adapter->fw_version),
3100 fw.eep_major, fw.eep_minor, fw.etrack_id);
3102 snprintf(adapter->fw_version,
3103 sizeof(adapter->fw_version),
3105 fw.eep_major, fw.eep_minor, fw.eep_build);
3112 * igb_init_mas - init Media Autosense feature if enabled in the NVM
3114 * @adapter: adapter struct
3116 static void igb_init_mas(struct igb_adapter *adapter)
3118 struct e1000_hw *hw = &adapter->hw;
3121 hw->nvm.ops.read(hw, NVM_COMPAT, 1, &eeprom_data);
3122 switch (hw->bus.func) {
3124 if (eeprom_data & IGB_MAS_ENABLE_0) {
3125 adapter->flags |= IGB_FLAG_MAS_ENABLE;
3126 netdev_info(adapter->netdev,
3127 "MAS: Enabling Media Autosense for port %d\n",
3132 if (eeprom_data & IGB_MAS_ENABLE_1) {
3133 adapter->flags |= IGB_FLAG_MAS_ENABLE;
3134 netdev_info(adapter->netdev,
3135 "MAS: Enabling Media Autosense for port %d\n",
3140 if (eeprom_data & IGB_MAS_ENABLE_2) {
3141 adapter->flags |= IGB_FLAG_MAS_ENABLE;
3142 netdev_info(adapter->netdev,
3143 "MAS: Enabling Media Autosense for port %d\n",
3148 if (eeprom_data & IGB_MAS_ENABLE_3) {
3149 adapter->flags |= IGB_FLAG_MAS_ENABLE;
3150 netdev_info(adapter->netdev,
3151 "MAS: Enabling Media Autosense for port %d\n",
3156 /* Shouldn't get here */
3157 netdev_err(adapter->netdev,
3158 "MAS: Invalid port configuration, returning\n");
3164 * igb_init_i2c - Init I2C interface
3165 * @adapter: pointer to adapter structure
3167 static s32 igb_init_i2c(struct igb_adapter *adapter)
3171 /* I2C interface supported on i350 devices */
3172 if (adapter->hw.mac.type != e1000_i350)
3175 /* Initialize the i2c bus which is controlled by the registers.
3176 * This bus will use the i2c_algo_bit structure that implements
3177 * the protocol through toggling of the 4 bits in the register.
3179 adapter->i2c_adap.owner = THIS_MODULE;
3180 adapter->i2c_algo = igb_i2c_algo;
3181 adapter->i2c_algo.data = adapter;
3182 adapter->i2c_adap.algo_data = &adapter->i2c_algo;
3183 adapter->i2c_adap.dev.parent = &adapter->pdev->dev;
3184 strscpy(adapter->i2c_adap.name, "igb BB",
3185 sizeof(adapter->i2c_adap.name));
3186 status = i2c_bit_add_bus(&adapter->i2c_adap);
3191 * igb_probe - Device Initialization Routine
3192 * @pdev: PCI device information struct
3193 * @ent: entry in igb_pci_tbl
3195 * Returns 0 on success, negative on failure
3197 * igb_probe initializes an adapter identified by a pci_dev structure.
3198 * The OS initialization, configuring of the adapter private structure,
3199 * and a hardware reset occur.
3201 static int igb_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
3203 struct net_device *netdev;
3204 struct igb_adapter *adapter;
3205 struct e1000_hw *hw;
3206 u16 eeprom_data = 0;
3208 static int global_quad_port_a; /* global quad port a indication */
3209 const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
3210 u8 part_str[E1000_PBANUM_LENGTH];
3213 /* Catch broken hardware that put the wrong VF device ID in
3214 * the PCIe SR-IOV capability.
3216 if (pdev->is_virtfn) {
3217 WARN(1, KERN_ERR "%s (%x:%x) should not be a VF!\n",
3218 pci_name(pdev), pdev->vendor, pdev->device);
3222 err = pci_enable_device_mem(pdev);
3226 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
3229 "No usable DMA configuration, aborting\n");
3233 err = pci_request_mem_regions(pdev, igb_driver_name);
3237 pci_set_master(pdev);
3238 pci_save_state(pdev);
3241 netdev = alloc_etherdev_mq(sizeof(struct igb_adapter),
3244 goto err_alloc_etherdev;
3246 SET_NETDEV_DEV(netdev, &pdev->dev);
3248 pci_set_drvdata(pdev, netdev);
3249 adapter = netdev_priv(netdev);
3250 adapter->netdev = netdev;
3251 adapter->pdev = pdev;
3254 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
3257 adapter->io_addr = pci_iomap(pdev, 0, 0);
3258 if (!adapter->io_addr)
3260 /* hw->hw_addr can be altered, we'll use adapter->io_addr for unmap */
3261 hw->hw_addr = adapter->io_addr;
3263 netdev->netdev_ops = &igb_netdev_ops;
3264 igb_set_ethtool_ops(netdev);
3265 netdev->watchdog_timeo = 5 * HZ;
3267 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
3269 netdev->mem_start = pci_resource_start(pdev, 0);
3270 netdev->mem_end = pci_resource_end(pdev, 0);
3272 /* PCI config space info */
3273 hw->vendor_id = pdev->vendor;
3274 hw->device_id = pdev->device;
3275 hw->revision_id = pdev->revision;
3276 hw->subsystem_vendor_id = pdev->subsystem_vendor;
3277 hw->subsystem_device_id = pdev->subsystem_device;
3279 /* Copy the default MAC, PHY and NVM function pointers */
3280 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
3281 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
3282 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
3283 /* Initialize skew-specific constants */
3284 err = ei->get_invariants(hw);
3288 /* setup the private structure */
3289 err = igb_sw_init(adapter);
3293 igb_get_bus_info_pcie(hw);
3295 hw->phy.autoneg_wait_to_complete = false;
3297 /* Copper options */
3298 if (hw->phy.media_type == e1000_media_type_copper) {
3299 hw->phy.mdix = AUTO_ALL_MODES;
3300 hw->phy.disable_polarity_correction = false;
3301 hw->phy.ms_type = e1000_ms_hw_default;
3304 if (igb_check_reset_block(hw))
3305 dev_info(&pdev->dev,
3306 "PHY reset is blocked due to SOL/IDER session.\n");
3308 /* features is initialized to 0 in allocation, it might have bits
3309 * set by igb_sw_init so we should use an or instead of an
3312 netdev->features |= NETIF_F_SG |
3319 if (hw->mac.type >= e1000_82576)
3320 netdev->features |= NETIF_F_SCTP_CRC | NETIF_F_GSO_UDP_L4;
3322 if (hw->mac.type >= e1000_i350)
3323 netdev->features |= NETIF_F_HW_TC;
3325 #define IGB_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
3326 NETIF_F_GSO_GRE_CSUM | \
3327 NETIF_F_GSO_IPXIP4 | \
3328 NETIF_F_GSO_IPXIP6 | \
3329 NETIF_F_GSO_UDP_TUNNEL | \
3330 NETIF_F_GSO_UDP_TUNNEL_CSUM)
3332 netdev->gso_partial_features = IGB_GSO_PARTIAL_FEATURES;
3333 netdev->features |= NETIF_F_GSO_PARTIAL | IGB_GSO_PARTIAL_FEATURES;
3335 /* copy netdev features into list of user selectable features */
3336 netdev->hw_features |= netdev->features |
3337 NETIF_F_HW_VLAN_CTAG_RX |
3338 NETIF_F_HW_VLAN_CTAG_TX |
3341 if (hw->mac.type >= e1000_i350)
3342 netdev->hw_features |= NETIF_F_NTUPLE;
3344 netdev->features |= NETIF_F_HIGHDMA;
3346 netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
3347 netdev->mpls_features |= NETIF_F_HW_CSUM;
3348 netdev->hw_enc_features |= netdev->vlan_features;
3350 /* set this bit last since it cannot be part of vlan_features */
3351 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
3352 NETIF_F_HW_VLAN_CTAG_RX |
3353 NETIF_F_HW_VLAN_CTAG_TX;
3355 netdev->priv_flags |= IFF_SUPP_NOFCS;
3357 netdev->priv_flags |= IFF_UNICAST_FLT;
3358 netdev->xdp_features = NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_REDIRECT;
3360 /* MTU range: 68 - 9216 */
3361 netdev->min_mtu = ETH_MIN_MTU;
3362 netdev->max_mtu = MAX_STD_JUMBO_FRAME_SIZE;
3364 adapter->en_mng_pt = igb_enable_mng_pass_thru(hw);
3366 /* before reading the NVM, reset the controller to put the device in a
3367 * known good starting state
3369 hw->mac.ops.reset_hw(hw);
3371 /* make sure the NVM is good , i211/i210 parts can have special NVM
3372 * that doesn't contain a checksum
3374 switch (hw->mac.type) {
3377 if (igb_get_flash_presence_i210(hw)) {
3378 if (hw->nvm.ops.validate(hw) < 0) {
3380 "The NVM Checksum Is Not Valid\n");
3387 if (hw->nvm.ops.validate(hw) < 0) {
3388 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
3395 if (eth_platform_get_mac_address(&pdev->dev, hw->mac.addr)) {
3396 /* copy the MAC address out of the NVM */
3397 if (hw->mac.ops.read_mac_addr(hw))
3398 dev_err(&pdev->dev, "NVM Read Error\n");
3401 eth_hw_addr_set(netdev, hw->mac.addr);
3403 if (!is_valid_ether_addr(netdev->dev_addr)) {
3404 dev_err(&pdev->dev, "Invalid MAC Address\n");
3409 igb_set_default_mac_filter(adapter);
3411 /* get firmware version for ethtool -i */
3412 igb_set_fw_version(adapter);
3414 /* configure RXPBSIZE and TXPBSIZE */
3415 if (hw->mac.type == e1000_i210) {
3416 wr32(E1000_RXPBS, I210_RXPBSIZE_DEFAULT);
3417 wr32(E1000_TXPBS, I210_TXPBSIZE_DEFAULT);
3420 timer_setup(&adapter->watchdog_timer, igb_watchdog, 0);
3421 timer_setup(&adapter->phy_info_timer, igb_update_phy_info, 0);
3423 INIT_WORK(&adapter->reset_task, igb_reset_task);
3424 INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
3426 /* Initialize link properties that are user-changeable */
3427 adapter->fc_autoneg = true;
3428 hw->mac.autoneg = true;
3429 hw->phy.autoneg_advertised = 0x2f;
3431 hw->fc.requested_mode = e1000_fc_default;
3432 hw->fc.current_mode = e1000_fc_default;
3434 igb_validate_mdi_setting(hw);
3436 /* By default, support wake on port A */
3437 if (hw->bus.func == 0)
3438 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3440 /* Check the NVM for wake support on non-port A ports */
3441 if (hw->mac.type >= e1000_82580)
3442 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A +
3443 NVM_82580_LAN_FUNC_OFFSET(hw->bus.func), 1,
3445 else if (hw->bus.func == 1)
3446 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
3448 if (eeprom_data & IGB_EEPROM_APME)
3449 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3451 /* now that we have the eeprom settings, apply the special cases where
3452 * the eeprom may be wrong or the board simply won't support wake on
3453 * lan on a particular port
3455 switch (pdev->device) {
3456 case E1000_DEV_ID_82575GB_QUAD_COPPER:
3457 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
3459 case E1000_DEV_ID_82575EB_FIBER_SERDES:
3460 case E1000_DEV_ID_82576_FIBER:
3461 case E1000_DEV_ID_82576_SERDES:
3462 /* Wake events only supported on port A for dual fiber
3463 * regardless of eeprom setting
3465 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
3466 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
3468 case E1000_DEV_ID_82576_QUAD_COPPER:
3469 case E1000_DEV_ID_82576_QUAD_COPPER_ET2:
3470 /* if quad port adapter, disable WoL on all but port A */
3471 if (global_quad_port_a != 0)
3472 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
3474 adapter->flags |= IGB_FLAG_QUAD_PORT_A;
3475 /* Reset for multiple quad port adapters */
3476 if (++global_quad_port_a == 4)
3477 global_quad_port_a = 0;
3480 /* If the device can't wake, don't set software support */
3481 if (!device_can_wakeup(&adapter->pdev->dev))
3482 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
3485 /* initialize the wol settings based on the eeprom settings */
3486 if (adapter->flags & IGB_FLAG_WOL_SUPPORTED)
3487 adapter->wol |= E1000_WUFC_MAG;
3489 /* Some vendors want WoL disabled by default, but still supported */
3490 if ((hw->mac.type == e1000_i350) &&
3491 (pdev->subsystem_vendor == PCI_VENDOR_ID_HP)) {
3492 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3496 /* Some vendors want the ability to Use the EEPROM setting as
3497 * enable/disable only, and not for capability
3499 if (((hw->mac.type == e1000_i350) ||
3500 (hw->mac.type == e1000_i354)) &&
3501 (pdev->subsystem_vendor == PCI_VENDOR_ID_DELL)) {
3502 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3505 if (hw->mac.type == e1000_i350) {
3506 if (((pdev->subsystem_device == 0x5001) ||
3507 (pdev->subsystem_device == 0x5002)) &&
3508 (hw->bus.func == 0)) {
3509 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3512 if (pdev->subsystem_device == 0x1F52)
3513 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3516 device_set_wakeup_enable(&adapter->pdev->dev,
3517 adapter->flags & IGB_FLAG_WOL_SUPPORTED);
3519 /* reset the hardware with the new settings */
3522 /* Init the I2C interface */
3523 err = igb_init_i2c(adapter);
3525 dev_err(&pdev->dev, "failed to init i2c interface\n");
3529 /* let the f/w know that the h/w is now under the control of the
3532 igb_get_hw_control(adapter);
3534 strcpy(netdev->name, "eth%d");
3535 err = register_netdev(netdev);
3539 /* carrier off reporting is important to ethtool even BEFORE open */
3540 netif_carrier_off(netdev);
3542 #ifdef CONFIG_IGB_DCA
3543 if (dca_add_requester(&pdev->dev) == 0) {
3544 adapter->flags |= IGB_FLAG_DCA_ENABLED;
3545 dev_info(&pdev->dev, "DCA enabled\n");
3546 igb_setup_dca(adapter);
3550 #ifdef CONFIG_IGB_HWMON
3551 /* Initialize the thermal sensor on i350 devices. */
3552 if (hw->mac.type == e1000_i350 && hw->bus.func == 0) {
3555 /* Read the NVM to determine if this i350 device supports an
3556 * external thermal sensor.
3558 hw->nvm.ops.read(hw, NVM_ETS_CFG, 1, &ets_word);
3559 if (ets_word != 0x0000 && ets_word != 0xFFFF)
3560 adapter->ets = true;
3562 adapter->ets = false;
3563 /* Only enable I2C bit banging if an external thermal
3564 * sensor is supported.
3568 hw->mac.ops.init_thermal_sensor_thresh(hw);
3569 if (igb_sysfs_init(adapter))
3571 "failed to allocate sysfs resources\n");
3573 adapter->ets = false;
3576 /* Check if Media Autosense is enabled */
3578 if (hw->dev_spec._82575.mas_capable)
3579 igb_init_mas(adapter);
3581 /* do hw tstamp init after resetting */
3582 igb_ptp_init(adapter);
3584 dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
3585 /* print bus type/speed/width info, not applicable to i354 */
3586 if (hw->mac.type != e1000_i354) {
3587 dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
3589 ((hw->bus.speed == e1000_bus_speed_2500) ? "2.5Gb/s" :
3590 (hw->bus.speed == e1000_bus_speed_5000) ? "5.0Gb/s" :
3592 ((hw->bus.width == e1000_bus_width_pcie_x4) ?
3594 (hw->bus.width == e1000_bus_width_pcie_x2) ?
3596 (hw->bus.width == e1000_bus_width_pcie_x1) ?
3597 "Width x1" : "unknown"), netdev->dev_addr);
3600 if ((hw->mac.type == e1000_82576 &&
3601 rd32(E1000_EECD) & E1000_EECD_PRES) ||
3602 (hw->mac.type >= e1000_i210 ||
3603 igb_get_flash_presence_i210(hw))) {
3604 ret_val = igb_read_part_string(hw, part_str,
3605 E1000_PBANUM_LENGTH);
3607 ret_val = -E1000_ERR_INVM_VALUE_NOT_FOUND;
3611 strcpy(part_str, "Unknown");
3612 dev_info(&pdev->dev, "%s: PBA No: %s\n", netdev->name, part_str);
3613 dev_info(&pdev->dev,
3614 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
3615 (adapter->flags & IGB_FLAG_HAS_MSIX) ? "MSI-X" :
3616 (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
3617 adapter->num_rx_queues, adapter->num_tx_queues);
3618 if (hw->phy.media_type == e1000_media_type_copper) {
3619 switch (hw->mac.type) {
3623 /* Enable EEE for internal copper PHY devices */
3624 err = igb_set_eee_i350(hw, true, true);
3626 (!hw->dev_spec._82575.eee_disable)) {
3627 adapter->eee_advert =
3628 MDIO_EEE_100TX | MDIO_EEE_1000T;
3629 adapter->flags |= IGB_FLAG_EEE;
3633 if ((rd32(E1000_CTRL_EXT) &
3634 E1000_CTRL_EXT_LINK_MODE_SGMII)) {
3635 err = igb_set_eee_i354(hw, true, true);
3637 (!hw->dev_spec._82575.eee_disable)) {
3638 adapter->eee_advert =
3639 MDIO_EEE_100TX | MDIO_EEE_1000T;
3640 adapter->flags |= IGB_FLAG_EEE;
3649 dev_pm_set_driver_flags(&pdev->dev, DPM_FLAG_NO_DIRECT_COMPLETE);
3651 pm_runtime_put_noidle(&pdev->dev);
3655 igb_release_hw_control(adapter);
3656 memset(&adapter->i2c_adap, 0, sizeof(adapter->i2c_adap));
3658 if (!igb_check_reset_block(hw))
3661 if (hw->flash_address)
3662 iounmap(hw->flash_address);
3664 kfree(adapter->mac_table);
3665 kfree(adapter->shadow_vfta);
3666 igb_clear_interrupt_scheme(adapter);
3667 #ifdef CONFIG_PCI_IOV
3668 igb_disable_sriov(pdev, false);
3670 pci_iounmap(pdev, adapter->io_addr);
3672 free_netdev(netdev);
3674 pci_release_mem_regions(pdev);
3677 pci_disable_device(pdev);
3681 #ifdef CONFIG_PCI_IOV
3682 static int igb_sriov_reinit(struct pci_dev *dev)
3684 struct net_device *netdev = pci_get_drvdata(dev);
3685 struct igb_adapter *adapter = netdev_priv(netdev);
3686 struct pci_dev *pdev = adapter->pdev;
3690 if (netif_running(netdev))
3695 igb_clear_interrupt_scheme(adapter);
3697 igb_init_queue_configuration(adapter);
3699 if (igb_init_interrupt_scheme(adapter, true)) {
3701 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
3705 if (netif_running(netdev))
3713 static int igb_disable_sriov(struct pci_dev *pdev, bool reinit)
3715 struct net_device *netdev = pci_get_drvdata(pdev);
3716 struct igb_adapter *adapter = netdev_priv(netdev);
3717 struct e1000_hw *hw = &adapter->hw;
3718 unsigned long flags;
3720 /* reclaim resources allocated to VFs */
3721 if (adapter->vf_data) {
3722 /* disable iov and allow time for transactions to clear */
3723 if (pci_vfs_assigned(pdev)) {
3724 dev_warn(&pdev->dev,
3725 "Cannot deallocate SR-IOV virtual functions while they are assigned - VFs will not be deallocated\n");
3728 pci_disable_sriov(pdev);
3731 spin_lock_irqsave(&adapter->vfs_lock, flags);
3732 kfree(adapter->vf_mac_list);
3733 adapter->vf_mac_list = NULL;
3734 kfree(adapter->vf_data);
3735 adapter->vf_data = NULL;
3736 adapter->vfs_allocated_count = 0;
3737 spin_unlock_irqrestore(&adapter->vfs_lock, flags);
3738 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
3741 dev_info(&pdev->dev, "IOV Disabled\n");
3743 /* Re-enable DMA Coalescing flag since IOV is turned off */
3744 adapter->flags |= IGB_FLAG_DMAC;
3747 return reinit ? igb_sriov_reinit(pdev) : 0;
3750 static int igb_enable_sriov(struct pci_dev *pdev, int num_vfs, bool reinit)
3752 struct net_device *netdev = pci_get_drvdata(pdev);
3753 struct igb_adapter *adapter = netdev_priv(netdev);
3754 int old_vfs = pci_num_vf(pdev);
3755 struct vf_mac_filter *mac_list;
3757 int num_vf_mac_filters, i;
3759 if (!(adapter->flags & IGB_FLAG_HAS_MSIX) || num_vfs > 7) {
3767 dev_info(&pdev->dev, "%d pre-allocated VFs found - override max_vfs setting of %d\n",
3769 adapter->vfs_allocated_count = old_vfs;
3771 adapter->vfs_allocated_count = num_vfs;
3773 adapter->vf_data = kcalloc(adapter->vfs_allocated_count,
3774 sizeof(struct vf_data_storage), GFP_KERNEL);
3776 /* if allocation failed then we do not support SR-IOV */
3777 if (!adapter->vf_data) {
3778 adapter->vfs_allocated_count = 0;
3783 /* Due to the limited number of RAR entries calculate potential
3784 * number of MAC filters available for the VFs. Reserve entries
3785 * for PF default MAC, PF MAC filters and at least one RAR entry
3786 * for each VF for VF MAC.
3788 num_vf_mac_filters = adapter->hw.mac.rar_entry_count -
3789 (1 + IGB_PF_MAC_FILTERS_RESERVED +
3790 adapter->vfs_allocated_count);
3792 adapter->vf_mac_list = kcalloc(num_vf_mac_filters,
3793 sizeof(struct vf_mac_filter),
3796 mac_list = adapter->vf_mac_list;
3797 INIT_LIST_HEAD(&adapter->vf_macs.l);
3799 if (adapter->vf_mac_list) {
3800 /* Initialize list of VF MAC filters */
3801 for (i = 0; i < num_vf_mac_filters; i++) {
3803 mac_list->free = true;
3804 list_add(&mac_list->l, &adapter->vf_macs.l);
3808 /* If we could not allocate memory for the VF MAC filters
3809 * we can continue without this feature but warn user.
3812 "Unable to allocate memory for VF MAC filter list\n");
3815 dev_info(&pdev->dev, "%d VFs allocated\n",
3816 adapter->vfs_allocated_count);
3817 for (i = 0; i < adapter->vfs_allocated_count; i++)
3818 igb_vf_configure(adapter, i);
3820 /* DMA Coalescing is not supported in IOV mode. */
3821 adapter->flags &= ~IGB_FLAG_DMAC;
3824 err = igb_sriov_reinit(pdev);
3829 /* only call pci_enable_sriov() if no VFs are allocated already */
3831 err = pci_enable_sriov(pdev, adapter->vfs_allocated_count);
3836 kfree(adapter->vf_mac_list);
3837 adapter->vf_mac_list = NULL;
3838 kfree(adapter->vf_data);
3839 adapter->vf_data = NULL;
3840 adapter->vfs_allocated_count = 0;
3847 * igb_remove_i2c - Cleanup I2C interface
3848 * @adapter: pointer to adapter structure
3850 static void igb_remove_i2c(struct igb_adapter *adapter)
3852 /* free the adapter bus structure */
3853 i2c_del_adapter(&adapter->i2c_adap);
3857 * igb_remove - Device Removal Routine
3858 * @pdev: PCI device information struct
3860 * igb_remove is called by the PCI subsystem to alert the driver
3861 * that it should release a PCI device. The could be caused by a
3862 * Hot-Plug event, or because the driver is going to be removed from
3865 static void igb_remove(struct pci_dev *pdev)
3867 struct net_device *netdev = pci_get_drvdata(pdev);
3868 struct igb_adapter *adapter = netdev_priv(netdev);
3869 struct e1000_hw *hw = &adapter->hw;
3871 pm_runtime_get_noresume(&pdev->dev);
3872 #ifdef CONFIG_IGB_HWMON
3873 igb_sysfs_exit(adapter);
3875 igb_remove_i2c(adapter);
3876 igb_ptp_stop(adapter);
3877 /* The watchdog timer may be rescheduled, so explicitly
3878 * disable watchdog from being rescheduled.
3880 set_bit(__IGB_DOWN, &adapter->state);
3881 del_timer_sync(&adapter->watchdog_timer);
3882 del_timer_sync(&adapter->phy_info_timer);
3884 cancel_work_sync(&adapter->reset_task);
3885 cancel_work_sync(&adapter->watchdog_task);
3887 #ifdef CONFIG_IGB_DCA
3888 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
3889 dev_info(&pdev->dev, "DCA disabled\n");
3890 dca_remove_requester(&pdev->dev);
3891 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
3892 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
3896 /* Release control of h/w to f/w. If f/w is AMT enabled, this
3897 * would have already happened in close and is redundant.
3899 igb_release_hw_control(adapter);
3901 #ifdef CONFIG_PCI_IOV
3902 igb_disable_sriov(pdev, false);
3905 unregister_netdev(netdev);
3907 igb_clear_interrupt_scheme(adapter);
3909 pci_iounmap(pdev, adapter->io_addr);
3910 if (hw->flash_address)
3911 iounmap(hw->flash_address);
3912 pci_release_mem_regions(pdev);
3914 kfree(adapter->mac_table);
3915 kfree(adapter->shadow_vfta);
3916 free_netdev(netdev);
3918 pci_disable_device(pdev);
3922 * igb_probe_vfs - Initialize vf data storage and add VFs to pci config space
3923 * @adapter: board private structure to initialize
3925 * This function initializes the vf specific data storage and then attempts to
3926 * allocate the VFs. The reason for ordering it this way is because it is much
3927 * mor expensive time wise to disable SR-IOV than it is to allocate and free
3928 * the memory for the VFs.
3930 static void igb_probe_vfs(struct igb_adapter *adapter)
3932 #ifdef CONFIG_PCI_IOV
3933 struct pci_dev *pdev = adapter->pdev;
3934 struct e1000_hw *hw = &adapter->hw;
3936 /* Virtualization features not supported on i210 family. */
3937 if ((hw->mac.type == e1000_i210) || (hw->mac.type == e1000_i211))
3940 /* Of the below we really only want the effect of getting
3941 * IGB_FLAG_HAS_MSIX set (if available), without which
3942 * igb_enable_sriov() has no effect.
3944 igb_set_interrupt_capability(adapter, true);
3945 igb_reset_interrupt_capability(adapter);
3947 pci_sriov_set_totalvfs(pdev, 7);
3948 igb_enable_sriov(pdev, max_vfs, false);
3950 #endif /* CONFIG_PCI_IOV */
3953 unsigned int igb_get_max_rss_queues(struct igb_adapter *adapter)
3955 struct e1000_hw *hw = &adapter->hw;
3956 unsigned int max_rss_queues;
3958 /* Determine the maximum number of RSS queues supported. */
3959 switch (hw->mac.type) {
3961 max_rss_queues = IGB_MAX_RX_QUEUES_I211;
3965 max_rss_queues = IGB_MAX_RX_QUEUES_82575;
3968 /* I350 cannot do RSS and SR-IOV at the same time */
3969 if (!!adapter->vfs_allocated_count) {
3975 if (!!adapter->vfs_allocated_count) {
3983 max_rss_queues = IGB_MAX_RX_QUEUES;
3987 return max_rss_queues;
3990 static void igb_init_queue_configuration(struct igb_adapter *adapter)
3994 max_rss_queues = igb_get_max_rss_queues(adapter);
3995 adapter->rss_queues = min_t(u32, max_rss_queues, num_online_cpus());
3997 igb_set_flag_queue_pairs(adapter, max_rss_queues);
4000 void igb_set_flag_queue_pairs(struct igb_adapter *adapter,
4001 const u32 max_rss_queues)
4003 struct e1000_hw *hw = &adapter->hw;
4005 /* Determine if we need to pair queues. */
4006 switch (hw->mac.type) {
4009 /* Device supports enough interrupts without queue pairing. */
4017 /* If rss_queues > half of max_rss_queues, pair the queues in
4018 * order to conserve interrupts due to limited supply.
4020 if (adapter->rss_queues > (max_rss_queues / 2))
4021 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
4023 adapter->flags &= ~IGB_FLAG_QUEUE_PAIRS;
4029 * igb_sw_init - Initialize general software structures (struct igb_adapter)
4030 * @adapter: board private structure to initialize
4032 * igb_sw_init initializes the Adapter private data structure.
4033 * Fields are initialized based on PCI device information and
4034 * OS network device settings (MTU size).
4036 static int igb_sw_init(struct igb_adapter *adapter)
4038 struct e1000_hw *hw = &adapter->hw;
4039 struct net_device *netdev = adapter->netdev;
4040 struct pci_dev *pdev = adapter->pdev;
4042 pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
4044 /* set default ring sizes */
4045 adapter->tx_ring_count = IGB_DEFAULT_TXD;
4046 adapter->rx_ring_count = IGB_DEFAULT_RXD;
4048 /* set default ITR values */
4049 adapter->rx_itr_setting = IGB_DEFAULT_ITR;
4050 adapter->tx_itr_setting = IGB_DEFAULT_ITR;
4052 /* set default work limits */
4053 adapter->tx_work_limit = IGB_DEFAULT_TX_WORK;
4055 adapter->max_frame_size = netdev->mtu + IGB_ETH_PKT_HDR_PAD;
4056 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
4058 spin_lock_init(&adapter->nfc_lock);
4059 spin_lock_init(&adapter->stats64_lock);
4061 /* init spinlock to avoid concurrency of VF resources */
4062 spin_lock_init(&adapter->vfs_lock);
4063 #ifdef CONFIG_PCI_IOV
4064 switch (hw->mac.type) {
4068 dev_warn(&pdev->dev,
4069 "Maximum of 7 VFs per PF, using max\n");
4070 max_vfs = adapter->vfs_allocated_count = 7;
4072 adapter->vfs_allocated_count = max_vfs;
4073 if (adapter->vfs_allocated_count)
4074 dev_warn(&pdev->dev,
4075 "Enabling SR-IOV VFs using the module parameter is deprecated - please use the pci sysfs interface.\n");
4080 #endif /* CONFIG_PCI_IOV */
4082 /* Assume MSI-X interrupts, will be checked during IRQ allocation */
4083 adapter->flags |= IGB_FLAG_HAS_MSIX;
4085 adapter->mac_table = kcalloc(hw->mac.rar_entry_count,
4086 sizeof(struct igb_mac_addr),
4088 if (!adapter->mac_table)
4091 igb_probe_vfs(adapter);
4093 igb_init_queue_configuration(adapter);
4095 /* Setup and initialize a copy of the hw vlan table array */
4096 adapter->shadow_vfta = kcalloc(E1000_VLAN_FILTER_TBL_SIZE, sizeof(u32),
4098 if (!adapter->shadow_vfta)
4101 /* This call may decrease the number of queues */
4102 if (igb_init_interrupt_scheme(adapter, true)) {
4103 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
4107 /* Explicitly disable IRQ since the NIC can be in any state. */
4108 igb_irq_disable(adapter);
4110 if (hw->mac.type >= e1000_i350)
4111 adapter->flags &= ~IGB_FLAG_DMAC;
4113 set_bit(__IGB_DOWN, &adapter->state);
4118 * __igb_open - Called when a network interface is made active
4119 * @netdev: network interface device structure
4120 * @resuming: indicates whether we are in a resume call
4122 * Returns 0 on success, negative value on failure
4124 * The open entry point is called when a network interface is made
4125 * active by the system (IFF_UP). At this point all resources needed
4126 * for transmit and receive operations are allocated, the interrupt
4127 * handler is registered with the OS, the watchdog timer is started,
4128 * and the stack is notified that the interface is ready.
4130 static int __igb_open(struct net_device *netdev, bool resuming)
4132 struct igb_adapter *adapter = netdev_priv(netdev);
4133 struct e1000_hw *hw = &adapter->hw;
4134 struct pci_dev *pdev = adapter->pdev;
4138 /* disallow open during test */
4139 if (test_bit(__IGB_TESTING, &adapter->state)) {
4145 pm_runtime_get_sync(&pdev->dev);
4147 netif_carrier_off(netdev);
4149 /* allocate transmit descriptors */
4150 err = igb_setup_all_tx_resources(adapter);
4154 /* allocate receive descriptors */
4155 err = igb_setup_all_rx_resources(adapter);
4159 igb_power_up_link(adapter);
4161 /* before we allocate an interrupt, we must be ready to handle it.
4162 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
4163 * as soon as we call pci_request_irq, so we have to setup our
4164 * clean_rx handler before we do so.
4166 igb_configure(adapter);
4168 err = igb_request_irq(adapter);
4172 /* Notify the stack of the actual queue counts. */
4173 err = netif_set_real_num_tx_queues(adapter->netdev,
4174 adapter->num_tx_queues);
4176 goto err_set_queues;
4178 err = netif_set_real_num_rx_queues(adapter->netdev,
4179 adapter->num_rx_queues);
4181 goto err_set_queues;
4183 /* From here on the code is the same as igb_up() */
4184 clear_bit(__IGB_DOWN, &adapter->state);
4186 for (i = 0; i < adapter->num_q_vectors; i++)
4187 napi_enable(&(adapter->q_vector[i]->napi));
4189 /* Clear any pending interrupts. */
4193 igb_irq_enable(adapter);
4195 /* notify VFs that reset has been completed */
4196 if (adapter->vfs_allocated_count) {
4197 u32 reg_data = rd32(E1000_CTRL_EXT);
4199 reg_data |= E1000_CTRL_EXT_PFRSTD;
4200 wr32(E1000_CTRL_EXT, reg_data);
4203 netif_tx_start_all_queues(netdev);
4206 pm_runtime_put(&pdev->dev);
4208 /* start the watchdog. */
4209 hw->mac.get_link_status = 1;
4210 schedule_work(&adapter->watchdog_task);
4215 igb_free_irq(adapter);
4217 igb_release_hw_control(adapter);
4218 igb_power_down_link(adapter);
4219 igb_free_all_rx_resources(adapter);
4221 igb_free_all_tx_resources(adapter);
4225 pm_runtime_put(&pdev->dev);
4230 int igb_open(struct net_device *netdev)
4232 return __igb_open(netdev, false);
4236 * __igb_close - Disables a network interface
4237 * @netdev: network interface device structure
4238 * @suspending: indicates we are in a suspend call
4240 * Returns 0, this is not allowed to fail
4242 * The close entry point is called when an interface is de-activated
4243 * by the OS. The hardware is still under the driver's control, but
4244 * needs to be disabled. A global MAC reset is issued to stop the
4245 * hardware, and all transmit and receive resources are freed.
4247 static int __igb_close(struct net_device *netdev, bool suspending)
4249 struct igb_adapter *adapter = netdev_priv(netdev);
4250 struct pci_dev *pdev = adapter->pdev;
4252 WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
4255 pm_runtime_get_sync(&pdev->dev);
4258 igb_free_irq(adapter);
4260 igb_free_all_tx_resources(adapter);
4261 igb_free_all_rx_resources(adapter);
4264 pm_runtime_put_sync(&pdev->dev);
4268 int igb_close(struct net_device *netdev)
4270 if (netif_device_present(netdev) || netdev->dismantle)
4271 return __igb_close(netdev, false);
4276 * igb_setup_tx_resources - allocate Tx resources (Descriptors)
4277 * @tx_ring: tx descriptor ring (for a specific queue) to setup
4279 * Return 0 on success, negative on failure
4281 int igb_setup_tx_resources(struct igb_ring *tx_ring)
4283 struct device *dev = tx_ring->dev;
4286 size = sizeof(struct igb_tx_buffer) * tx_ring->count;
4288 tx_ring->tx_buffer_info = vmalloc(size);
4289 if (!tx_ring->tx_buffer_info)
4292 /* round up to nearest 4K */
4293 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
4294 tx_ring->size = ALIGN(tx_ring->size, 4096);
4296 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
4297 &tx_ring->dma, GFP_KERNEL);
4301 tx_ring->next_to_use = 0;
4302 tx_ring->next_to_clean = 0;
4307 vfree(tx_ring->tx_buffer_info);
4308 tx_ring->tx_buffer_info = NULL;
4309 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
4314 * igb_setup_all_tx_resources - wrapper to allocate Tx resources
4315 * (Descriptors) for all queues
4316 * @adapter: board private structure
4318 * Return 0 on success, negative on failure
4320 static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
4322 struct pci_dev *pdev = adapter->pdev;
4325 for (i = 0; i < adapter->num_tx_queues; i++) {
4326 err = igb_setup_tx_resources(adapter->tx_ring[i]);
4329 "Allocation for Tx Queue %u failed\n", i);
4330 for (i--; i >= 0; i--)
4331 igb_free_tx_resources(adapter->tx_ring[i]);
4340 * igb_setup_tctl - configure the transmit control registers
4341 * @adapter: Board private structure
4343 void igb_setup_tctl(struct igb_adapter *adapter)
4345 struct e1000_hw *hw = &adapter->hw;
4348 /* disable queue 0 which is enabled by default on 82575 and 82576 */
4349 wr32(E1000_TXDCTL(0), 0);
4351 /* Program the Transmit Control Register */
4352 tctl = rd32(E1000_TCTL);
4353 tctl &= ~E1000_TCTL_CT;
4354 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
4355 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
4357 igb_config_collision_dist(hw);
4359 /* Enable transmits */
4360 tctl |= E1000_TCTL_EN;
4362 wr32(E1000_TCTL, tctl);
4366 * igb_configure_tx_ring - Configure transmit ring after Reset
4367 * @adapter: board private structure
4368 * @ring: tx ring to configure
4370 * Configure a transmit ring after a reset.
4372 void igb_configure_tx_ring(struct igb_adapter *adapter,
4373 struct igb_ring *ring)
4375 struct e1000_hw *hw = &adapter->hw;
4377 u64 tdba = ring->dma;
4378 int reg_idx = ring->reg_idx;
4380 wr32(E1000_TDLEN(reg_idx),
4381 ring->count * sizeof(union e1000_adv_tx_desc));
4382 wr32(E1000_TDBAL(reg_idx),
4383 tdba & 0x00000000ffffffffULL);
4384 wr32(E1000_TDBAH(reg_idx), tdba >> 32);
4386 ring->tail = adapter->io_addr + E1000_TDT(reg_idx);
4387 wr32(E1000_TDH(reg_idx), 0);
4388 writel(0, ring->tail);
4390 txdctl |= IGB_TX_PTHRESH;
4391 txdctl |= IGB_TX_HTHRESH << 8;
4392 txdctl |= IGB_TX_WTHRESH << 16;
4394 /* reinitialize tx_buffer_info */
4395 memset(ring->tx_buffer_info, 0,
4396 sizeof(struct igb_tx_buffer) * ring->count);
4398 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
4399 wr32(E1000_TXDCTL(reg_idx), txdctl);
4403 * igb_configure_tx - Configure transmit Unit after Reset
4404 * @adapter: board private structure
4406 * Configure the Tx unit of the MAC after a reset.
4408 static void igb_configure_tx(struct igb_adapter *adapter)
4410 struct e1000_hw *hw = &adapter->hw;
4413 /* disable the queues */
4414 for (i = 0; i < adapter->num_tx_queues; i++)
4415 wr32(E1000_TXDCTL(adapter->tx_ring[i]->reg_idx), 0);
4418 usleep_range(10000, 20000);
4420 for (i = 0; i < adapter->num_tx_queues; i++)
4421 igb_configure_tx_ring(adapter, adapter->tx_ring[i]);
4425 * igb_setup_rx_resources - allocate Rx resources (Descriptors)
4426 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
4428 * Returns 0 on success, negative on failure
4430 int igb_setup_rx_resources(struct igb_ring *rx_ring)
4432 struct igb_adapter *adapter = netdev_priv(rx_ring->netdev);
4433 struct device *dev = rx_ring->dev;
4436 /* XDP RX-queue info */
4437 if (xdp_rxq_info_is_reg(&rx_ring->xdp_rxq))
4438 xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
4439 res = xdp_rxq_info_reg(&rx_ring->xdp_rxq, rx_ring->netdev,
4440 rx_ring->queue_index, 0);
4442 dev_err(dev, "Failed to register xdp_rxq index %u\n",
4443 rx_ring->queue_index);
4447 size = sizeof(struct igb_rx_buffer) * rx_ring->count;
4449 rx_ring->rx_buffer_info = vmalloc(size);
4450 if (!rx_ring->rx_buffer_info)
4453 /* Round up to nearest 4K */
4454 rx_ring->size = rx_ring->count * sizeof(union e1000_adv_rx_desc);
4455 rx_ring->size = ALIGN(rx_ring->size, 4096);
4457 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
4458 &rx_ring->dma, GFP_KERNEL);
4462 rx_ring->next_to_alloc = 0;
4463 rx_ring->next_to_clean = 0;
4464 rx_ring->next_to_use = 0;
4466 rx_ring->xdp_prog = adapter->xdp_prog;
4471 xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
4472 vfree(rx_ring->rx_buffer_info);
4473 rx_ring->rx_buffer_info = NULL;
4474 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
4479 * igb_setup_all_rx_resources - wrapper to allocate Rx resources
4480 * (Descriptors) for all queues
4481 * @adapter: board private structure
4483 * Return 0 on success, negative on failure
4485 static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
4487 struct pci_dev *pdev = adapter->pdev;
4490 for (i = 0; i < adapter->num_rx_queues; i++) {
4491 err = igb_setup_rx_resources(adapter->rx_ring[i]);
4494 "Allocation for Rx Queue %u failed\n", i);
4495 for (i--; i >= 0; i--)
4496 igb_free_rx_resources(adapter->rx_ring[i]);
4505 * igb_setup_mrqc - configure the multiple receive queue control registers
4506 * @adapter: Board private structure
4508 static void igb_setup_mrqc(struct igb_adapter *adapter)
4510 struct e1000_hw *hw = &adapter->hw;
4512 u32 j, num_rx_queues;
4515 netdev_rss_key_fill(rss_key, sizeof(rss_key));
4516 for (j = 0; j < 10; j++)
4517 wr32(E1000_RSSRK(j), rss_key[j]);
4519 num_rx_queues = adapter->rss_queues;
4521 switch (hw->mac.type) {
4523 /* 82576 supports 2 RSS queues for SR-IOV */
4524 if (adapter->vfs_allocated_count)
4531 if (adapter->rss_indir_tbl_init != num_rx_queues) {
4532 for (j = 0; j < IGB_RETA_SIZE; j++)
4533 adapter->rss_indir_tbl[j] =
4534 (j * num_rx_queues) / IGB_RETA_SIZE;
4535 adapter->rss_indir_tbl_init = num_rx_queues;
4537 igb_write_rss_indir_tbl(adapter);
4539 /* Disable raw packet checksumming so that RSS hash is placed in
4540 * descriptor on writeback. No need to enable TCP/UDP/IP checksum
4541 * offloads as they are enabled by default
4543 rxcsum = rd32(E1000_RXCSUM);
4544 rxcsum |= E1000_RXCSUM_PCSD;
4546 if (adapter->hw.mac.type >= e1000_82576)
4547 /* Enable Receive Checksum Offload for SCTP */
4548 rxcsum |= E1000_RXCSUM_CRCOFL;
4550 /* Don't need to set TUOFL or IPOFL, they default to 1 */
4551 wr32(E1000_RXCSUM, rxcsum);
4553 /* Generate RSS hash based on packet types, TCP/UDP
4554 * port numbers and/or IPv4/v6 src and dst addresses
4556 mrqc = E1000_MRQC_RSS_FIELD_IPV4 |
4557 E1000_MRQC_RSS_FIELD_IPV4_TCP |
4558 E1000_MRQC_RSS_FIELD_IPV6 |
4559 E1000_MRQC_RSS_FIELD_IPV6_TCP |
4560 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX;
4562 if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV4_UDP)
4563 mrqc |= E1000_MRQC_RSS_FIELD_IPV4_UDP;
4564 if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV6_UDP)
4565 mrqc |= E1000_MRQC_RSS_FIELD_IPV6_UDP;
4567 /* If VMDq is enabled then we set the appropriate mode for that, else
4568 * we default to RSS so that an RSS hash is calculated per packet even
4569 * if we are only using one queue
4571 if (adapter->vfs_allocated_count) {
4572 if (hw->mac.type > e1000_82575) {
4573 /* Set the default pool for the PF's first queue */
4574 u32 vtctl = rd32(E1000_VT_CTL);
4576 vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK |
4577 E1000_VT_CTL_DISABLE_DEF_POOL);
4578 vtctl |= adapter->vfs_allocated_count <<
4579 E1000_VT_CTL_DEFAULT_POOL_SHIFT;
4580 wr32(E1000_VT_CTL, vtctl);
4582 if (adapter->rss_queues > 1)
4583 mrqc |= E1000_MRQC_ENABLE_VMDQ_RSS_MQ;
4585 mrqc |= E1000_MRQC_ENABLE_VMDQ;
4587 mrqc |= E1000_MRQC_ENABLE_RSS_MQ;
4589 igb_vmm_control(adapter);
4591 wr32(E1000_MRQC, mrqc);
4595 * igb_setup_rctl - configure the receive control registers
4596 * @adapter: Board private structure
4598 void igb_setup_rctl(struct igb_adapter *adapter)
4600 struct e1000_hw *hw = &adapter->hw;
4603 rctl = rd32(E1000_RCTL);
4605 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
4606 rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
4608 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
4609 (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
4611 /* enable stripping of CRC. It's unlikely this will break BMC
4612 * redirection as it did with e1000. Newer features require
4613 * that the HW strips the CRC.
4615 rctl |= E1000_RCTL_SECRC;
4617 /* disable store bad packets and clear size bits. */
4618 rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
4620 /* enable LPE to allow for reception of jumbo frames */
4621 rctl |= E1000_RCTL_LPE;
4623 /* disable queue 0 to prevent tail write w/o re-config */
4624 wr32(E1000_RXDCTL(0), 0);
4626 /* Attention!!! For SR-IOV PF driver operations you must enable
4627 * queue drop for all VF and PF queues to prevent head of line blocking
4628 * if an un-trusted VF does not provide descriptors to hardware.
4630 if (adapter->vfs_allocated_count) {
4631 /* set all queue drop enable bits */
4632 wr32(E1000_QDE, ALL_QUEUES);
4635 /* This is useful for sniffing bad packets. */
4636 if (adapter->netdev->features & NETIF_F_RXALL) {
4637 /* UPE and MPE will be handled by normal PROMISC logic
4638 * in e1000e_set_rx_mode
4640 rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
4641 E1000_RCTL_BAM | /* RX All Bcast Pkts */
4642 E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
4644 rctl &= ~(E1000_RCTL_DPF | /* Allow filtered pause */
4645 E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
4646 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
4647 * and that breaks VLANs.
4651 wr32(E1000_RCTL, rctl);
4654 static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
4657 struct e1000_hw *hw = &adapter->hw;
4660 if (size > MAX_JUMBO_FRAME_SIZE)
4661 size = MAX_JUMBO_FRAME_SIZE;
4663 vmolr = rd32(E1000_VMOLR(vfn));
4664 vmolr &= ~E1000_VMOLR_RLPML_MASK;
4665 vmolr |= size | E1000_VMOLR_LPE;
4666 wr32(E1000_VMOLR(vfn), vmolr);
4671 static inline void igb_set_vf_vlan_strip(struct igb_adapter *adapter,
4672 int vfn, bool enable)
4674 struct e1000_hw *hw = &adapter->hw;
4677 if (hw->mac.type < e1000_82576)
4680 if (hw->mac.type == e1000_i350)
4681 reg = E1000_DVMOLR(vfn);
4683 reg = E1000_VMOLR(vfn);
4687 val |= E1000_VMOLR_STRVLAN;
4689 val &= ~(E1000_VMOLR_STRVLAN);
4693 static inline void igb_set_vmolr(struct igb_adapter *adapter,
4696 struct e1000_hw *hw = &adapter->hw;
4699 /* This register exists only on 82576 and newer so if we are older then
4700 * we should exit and do nothing
4702 if (hw->mac.type < e1000_82576)
4705 vmolr = rd32(E1000_VMOLR(vfn));
4707 vmolr |= E1000_VMOLR_AUPE; /* Accept untagged packets */
4709 vmolr &= ~(E1000_VMOLR_AUPE); /* Tagged packets ONLY */
4711 /* clear all bits that might not be set */
4712 vmolr &= ~(E1000_VMOLR_BAM | E1000_VMOLR_RSSE);
4714 if (adapter->rss_queues > 1 && vfn == adapter->vfs_allocated_count)
4715 vmolr |= E1000_VMOLR_RSSE; /* enable RSS */
4716 /* for VMDq only allow the VFs and pool 0 to accept broadcast and
4719 if (vfn <= adapter->vfs_allocated_count)
4720 vmolr |= E1000_VMOLR_BAM; /* Accept broadcast */
4722 wr32(E1000_VMOLR(vfn), vmolr);
4726 * igb_setup_srrctl - configure the split and replication receive control
4728 * @adapter: Board private structure
4729 * @ring: receive ring to be configured
4731 void igb_setup_srrctl(struct igb_adapter *adapter, struct igb_ring *ring)
4733 struct e1000_hw *hw = &adapter->hw;
4734 int reg_idx = ring->reg_idx;
4737 srrctl = IGB_RX_HDR_LEN << E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
4738 if (ring_uses_large_buffer(ring))
4739 srrctl |= IGB_RXBUFFER_3072 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
4741 srrctl |= IGB_RXBUFFER_2048 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
4742 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
4743 if (hw->mac.type >= e1000_82580)
4744 srrctl |= E1000_SRRCTL_TIMESTAMP;
4745 /* Only set Drop Enable if VFs allocated, or we are supporting multiple
4746 * queues and rx flow control is disabled
4748 if (adapter->vfs_allocated_count ||
4749 (!(hw->fc.current_mode & e1000_fc_rx_pause) &&
4750 adapter->num_rx_queues > 1))
4751 srrctl |= E1000_SRRCTL_DROP_EN;
4753 wr32(E1000_SRRCTL(reg_idx), srrctl);
4757 * igb_configure_rx_ring - Configure a receive ring after Reset
4758 * @adapter: board private structure
4759 * @ring: receive ring to be configured
4761 * Configure the Rx unit of the MAC after a reset.
4763 void igb_configure_rx_ring(struct igb_adapter *adapter,
4764 struct igb_ring *ring)
4766 struct e1000_hw *hw = &adapter->hw;
4767 union e1000_adv_rx_desc *rx_desc;
4768 u64 rdba = ring->dma;
4769 int reg_idx = ring->reg_idx;
4772 xdp_rxq_info_unreg_mem_model(&ring->xdp_rxq);
4773 WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
4774 MEM_TYPE_PAGE_SHARED, NULL));
4776 /* disable the queue */
4777 wr32(E1000_RXDCTL(reg_idx), 0);
4779 /* Set DMA base address registers */
4780 wr32(E1000_RDBAL(reg_idx),
4781 rdba & 0x00000000ffffffffULL);
4782 wr32(E1000_RDBAH(reg_idx), rdba >> 32);
4783 wr32(E1000_RDLEN(reg_idx),
4784 ring->count * sizeof(union e1000_adv_rx_desc));
4786 /* initialize head and tail */
4787 ring->tail = adapter->io_addr + E1000_RDT(reg_idx);
4788 wr32(E1000_RDH(reg_idx), 0);
4789 writel(0, ring->tail);
4791 /* set descriptor configuration */
4792 igb_setup_srrctl(adapter, ring);
4794 /* set filtering for VMDQ pools */
4795 igb_set_vmolr(adapter, reg_idx & 0x7, true);
4797 rxdctl |= IGB_RX_PTHRESH;
4798 rxdctl |= IGB_RX_HTHRESH << 8;
4799 rxdctl |= IGB_RX_WTHRESH << 16;
4801 /* initialize rx_buffer_info */
4802 memset(ring->rx_buffer_info, 0,
4803 sizeof(struct igb_rx_buffer) * ring->count);
4805 /* initialize Rx descriptor 0 */
4806 rx_desc = IGB_RX_DESC(ring, 0);
4807 rx_desc->wb.upper.length = 0;
4809 /* enable receive descriptor fetching */
4810 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
4811 wr32(E1000_RXDCTL(reg_idx), rxdctl);
4814 static void igb_set_rx_buffer_len(struct igb_adapter *adapter,
4815 struct igb_ring *rx_ring)
4817 #if (PAGE_SIZE < 8192)
4818 struct e1000_hw *hw = &adapter->hw;
4821 /* set build_skb and buffer size flags */
4822 clear_ring_build_skb_enabled(rx_ring);
4823 clear_ring_uses_large_buffer(rx_ring);
4825 if (adapter->flags & IGB_FLAG_RX_LEGACY)
4828 set_ring_build_skb_enabled(rx_ring);
4830 #if (PAGE_SIZE < 8192)
4831 if (adapter->max_frame_size > IGB_MAX_FRAME_BUILD_SKB ||
4832 rd32(E1000_RCTL) & E1000_RCTL_SBP)
4833 set_ring_uses_large_buffer(rx_ring);
4838 * igb_configure_rx - Configure receive Unit after Reset
4839 * @adapter: board private structure
4841 * Configure the Rx unit of the MAC after a reset.
4843 static void igb_configure_rx(struct igb_adapter *adapter)
4847 /* set the correct pool for the PF default MAC address in entry 0 */
4848 igb_set_default_mac_filter(adapter);
4850 /* Setup the HW Rx Head and Tail Descriptor Pointers and
4851 * the Base and Length of the Rx Descriptor Ring
4853 for (i = 0; i < adapter->num_rx_queues; i++) {
4854 struct igb_ring *rx_ring = adapter->rx_ring[i];
4856 igb_set_rx_buffer_len(adapter, rx_ring);
4857 igb_configure_rx_ring(adapter, rx_ring);
4862 * igb_free_tx_resources - Free Tx Resources per Queue
4863 * @tx_ring: Tx descriptor ring for a specific queue
4865 * Free all transmit software resources
4867 void igb_free_tx_resources(struct igb_ring *tx_ring)
4869 igb_clean_tx_ring(tx_ring);
4871 vfree(tx_ring->tx_buffer_info);
4872 tx_ring->tx_buffer_info = NULL;
4874 /* if not set, then don't free */
4878 dma_free_coherent(tx_ring->dev, tx_ring->size,
4879 tx_ring->desc, tx_ring->dma);
4881 tx_ring->desc = NULL;
4885 * igb_free_all_tx_resources - Free Tx Resources for All Queues
4886 * @adapter: board private structure
4888 * Free all transmit software resources
4890 static void igb_free_all_tx_resources(struct igb_adapter *adapter)
4894 for (i = 0; i < adapter->num_tx_queues; i++)
4895 if (adapter->tx_ring[i])
4896 igb_free_tx_resources(adapter->tx_ring[i]);
4900 * igb_clean_tx_ring - Free Tx Buffers
4901 * @tx_ring: ring to be cleaned
4903 static void igb_clean_tx_ring(struct igb_ring *tx_ring)
4905 u16 i = tx_ring->next_to_clean;
4906 struct igb_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
4908 while (i != tx_ring->next_to_use) {
4909 union e1000_adv_tx_desc *eop_desc, *tx_desc;
4911 /* Free all the Tx ring sk_buffs or xdp frames */
4912 if (tx_buffer->type == IGB_TYPE_SKB)
4913 dev_kfree_skb_any(tx_buffer->skb);
4915 xdp_return_frame(tx_buffer->xdpf);
4917 /* unmap skb header data */
4918 dma_unmap_single(tx_ring->dev,
4919 dma_unmap_addr(tx_buffer, dma),
4920 dma_unmap_len(tx_buffer, len),
4923 /* check for eop_desc to determine the end of the packet */
4924 eop_desc = tx_buffer->next_to_watch;
4925 tx_desc = IGB_TX_DESC(tx_ring, i);
4927 /* unmap remaining buffers */
4928 while (tx_desc != eop_desc) {
4932 if (unlikely(i == tx_ring->count)) {
4934 tx_buffer = tx_ring->tx_buffer_info;
4935 tx_desc = IGB_TX_DESC(tx_ring, 0);
4938 /* unmap any remaining paged data */
4939 if (dma_unmap_len(tx_buffer, len))
4940 dma_unmap_page(tx_ring->dev,
4941 dma_unmap_addr(tx_buffer, dma),
4942 dma_unmap_len(tx_buffer, len),
4946 tx_buffer->next_to_watch = NULL;
4948 /* move us one more past the eop_desc for start of next pkt */
4951 if (unlikely(i == tx_ring->count)) {
4953 tx_buffer = tx_ring->tx_buffer_info;
4957 /* reset BQL for queue */
4958 netdev_tx_reset_queue(txring_txq(tx_ring));
4960 /* reset next_to_use and next_to_clean */
4961 tx_ring->next_to_use = 0;
4962 tx_ring->next_to_clean = 0;
4966 * igb_clean_all_tx_rings - Free Tx Buffers for all queues
4967 * @adapter: board private structure
4969 static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
4973 for (i = 0; i < adapter->num_tx_queues; i++)
4974 if (adapter->tx_ring[i])
4975 igb_clean_tx_ring(adapter->tx_ring[i]);
4979 * igb_free_rx_resources - Free Rx Resources
4980 * @rx_ring: ring to clean the resources from
4982 * Free all receive software resources
4984 void igb_free_rx_resources(struct igb_ring *rx_ring)
4986 igb_clean_rx_ring(rx_ring);
4988 rx_ring->xdp_prog = NULL;
4989 xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
4990 vfree(rx_ring->rx_buffer_info);
4991 rx_ring->rx_buffer_info = NULL;
4993 /* if not set, then don't free */
4997 dma_free_coherent(rx_ring->dev, rx_ring->size,
4998 rx_ring->desc, rx_ring->dma);
5000 rx_ring->desc = NULL;
5004 * igb_free_all_rx_resources - Free Rx Resources for All Queues
5005 * @adapter: board private structure
5007 * Free all receive software resources
5009 static void igb_free_all_rx_resources(struct igb_adapter *adapter)
5013 for (i = 0; i < adapter->num_rx_queues; i++)
5014 if (adapter->rx_ring[i])
5015 igb_free_rx_resources(adapter->rx_ring[i]);
5019 * igb_clean_rx_ring - Free Rx Buffers per Queue
5020 * @rx_ring: ring to free buffers from
5022 static void igb_clean_rx_ring(struct igb_ring *rx_ring)
5024 u16 i = rx_ring->next_to_clean;
5026 dev_kfree_skb(rx_ring->skb);
5027 rx_ring->skb = NULL;
5029 /* Free all the Rx ring sk_buffs */
5030 while (i != rx_ring->next_to_alloc) {
5031 struct igb_rx_buffer *buffer_info = &rx_ring->rx_buffer_info[i];
5033 /* Invalidate cache lines that may have been written to by
5034 * device so that we avoid corrupting memory.
5036 dma_sync_single_range_for_cpu(rx_ring->dev,
5038 buffer_info->page_offset,
5039 igb_rx_bufsz(rx_ring),
5042 /* free resources associated with mapping */
5043 dma_unmap_page_attrs(rx_ring->dev,
5045 igb_rx_pg_size(rx_ring),
5048 __page_frag_cache_drain(buffer_info->page,
5049 buffer_info->pagecnt_bias);
5052 if (i == rx_ring->count)
5056 rx_ring->next_to_alloc = 0;
5057 rx_ring->next_to_clean = 0;
5058 rx_ring->next_to_use = 0;
5062 * igb_clean_all_rx_rings - Free Rx Buffers for all queues
5063 * @adapter: board private structure
5065 static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
5069 for (i = 0; i < adapter->num_rx_queues; i++)
5070 if (adapter->rx_ring[i])
5071 igb_clean_rx_ring(adapter->rx_ring[i]);
5075 * igb_set_mac - Change the Ethernet Address of the NIC
5076 * @netdev: network interface device structure
5077 * @p: pointer to an address structure
5079 * Returns 0 on success, negative on failure
5081 static int igb_set_mac(struct net_device *netdev, void *p)
5083 struct igb_adapter *adapter = netdev_priv(netdev);
5084 struct e1000_hw *hw = &adapter->hw;
5085 struct sockaddr *addr = p;
5087 if (!is_valid_ether_addr(addr->sa_data))
5088 return -EADDRNOTAVAIL;
5090 eth_hw_addr_set(netdev, addr->sa_data);
5091 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
5093 /* set the correct pool for the new PF MAC address in entry 0 */
5094 igb_set_default_mac_filter(adapter);
5100 * igb_write_mc_addr_list - write multicast addresses to MTA
5101 * @netdev: network interface device structure
5103 * Writes multicast address list to the MTA hash table.
5104 * Returns: -ENOMEM on failure
5105 * 0 on no addresses written
5106 * X on writing X addresses to MTA
5108 static int igb_write_mc_addr_list(struct net_device *netdev)
5110 struct igb_adapter *adapter = netdev_priv(netdev);
5111 struct e1000_hw *hw = &adapter->hw;
5112 struct netdev_hw_addr *ha;
5116 if (netdev_mc_empty(netdev)) {
5117 /* nothing to program, so clear mc list */
5118 igb_update_mc_addr_list(hw, NULL, 0);
5119 igb_restore_vf_multicasts(adapter);
5123 mta_list = kcalloc(netdev_mc_count(netdev), 6, GFP_ATOMIC);
5127 /* The shared function expects a packed array of only addresses. */
5129 netdev_for_each_mc_addr(ha, netdev)
5130 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
5132 igb_update_mc_addr_list(hw, mta_list, i);
5135 return netdev_mc_count(netdev);
5138 static int igb_vlan_promisc_enable(struct igb_adapter *adapter)
5140 struct e1000_hw *hw = &adapter->hw;
5143 switch (hw->mac.type) {
5147 /* VLAN filtering needed for VLAN prio filter */
5148 if (adapter->netdev->features & NETIF_F_NTUPLE)
5154 /* VLAN filtering needed for pool filtering */
5155 if (adapter->vfs_allocated_count)
5162 /* We are already in VLAN promisc, nothing to do */
5163 if (adapter->flags & IGB_FLAG_VLAN_PROMISC)
5166 if (!adapter->vfs_allocated_count)
5169 /* Add PF to all active pools */
5170 pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT;
5172 for (i = E1000_VLVF_ARRAY_SIZE; --i;) {
5173 u32 vlvf = rd32(E1000_VLVF(i));
5176 wr32(E1000_VLVF(i), vlvf);
5180 /* Set all bits in the VLAN filter table array */
5181 for (i = E1000_VLAN_FILTER_TBL_SIZE; i--;)
5182 hw->mac.ops.write_vfta(hw, i, ~0U);
5184 /* Set flag so we don't redo unnecessary work */
5185 adapter->flags |= IGB_FLAG_VLAN_PROMISC;
5190 #define VFTA_BLOCK_SIZE 8
5191 static void igb_scrub_vfta(struct igb_adapter *adapter, u32 vfta_offset)
5193 struct e1000_hw *hw = &adapter->hw;
5194 u32 vfta[VFTA_BLOCK_SIZE] = { 0 };
5195 u32 vid_start = vfta_offset * 32;
5196 u32 vid_end = vid_start + (VFTA_BLOCK_SIZE * 32);
5197 u32 i, vid, word, bits, pf_id;
5199 /* guarantee that we don't scrub out management VLAN */
5200 vid = adapter->mng_vlan_id;
5201 if (vid >= vid_start && vid < vid_end)
5202 vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
5204 if (!adapter->vfs_allocated_count)
5207 pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT;
5209 for (i = E1000_VLVF_ARRAY_SIZE; --i;) {
5210 u32 vlvf = rd32(E1000_VLVF(i));
5212 /* pull VLAN ID from VLVF */
5213 vid = vlvf & VLAN_VID_MASK;
5215 /* only concern ourselves with a certain range */
5216 if (vid < vid_start || vid >= vid_end)
5219 if (vlvf & E1000_VLVF_VLANID_ENABLE) {
5220 /* record VLAN ID in VFTA */
5221 vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
5223 /* if PF is part of this then continue */
5224 if (test_bit(vid, adapter->active_vlans))
5228 /* remove PF from the pool */
5230 bits &= rd32(E1000_VLVF(i));
5231 wr32(E1000_VLVF(i), bits);
5235 /* extract values from active_vlans and write back to VFTA */
5236 for (i = VFTA_BLOCK_SIZE; i--;) {
5237 vid = (vfta_offset + i) * 32;
5238 word = vid / BITS_PER_LONG;
5239 bits = vid % BITS_PER_LONG;
5241 vfta[i] |= adapter->active_vlans[word] >> bits;
5243 hw->mac.ops.write_vfta(hw, vfta_offset + i, vfta[i]);
5247 static void igb_vlan_promisc_disable(struct igb_adapter *adapter)
5251 /* We are not in VLAN promisc, nothing to do */
5252 if (!(adapter->flags & IGB_FLAG_VLAN_PROMISC))
5255 /* Set flag so we don't redo unnecessary work */
5256 adapter->flags &= ~IGB_FLAG_VLAN_PROMISC;
5258 for (i = 0; i < E1000_VLAN_FILTER_TBL_SIZE; i += VFTA_BLOCK_SIZE)
5259 igb_scrub_vfta(adapter, i);
5263 * igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
5264 * @netdev: network interface device structure
5266 * The set_rx_mode entry point is called whenever the unicast or multicast
5267 * address lists or the network interface flags are updated. This routine is
5268 * responsible for configuring the hardware for proper unicast, multicast,
5269 * promiscuous mode, and all-multi behavior.
5271 static void igb_set_rx_mode(struct net_device *netdev)
5273 struct igb_adapter *adapter = netdev_priv(netdev);
5274 struct e1000_hw *hw = &adapter->hw;
5275 unsigned int vfn = adapter->vfs_allocated_count;
5276 u32 rctl = 0, vmolr = 0, rlpml = MAX_JUMBO_FRAME_SIZE;
5279 /* Check for Promiscuous and All Multicast modes */
5280 if (netdev->flags & IFF_PROMISC) {
5281 rctl |= E1000_RCTL_UPE | E1000_RCTL_MPE;
5282 vmolr |= E1000_VMOLR_MPME;
5284 /* enable use of UTA filter to force packets to default pool */
5285 if (hw->mac.type == e1000_82576)
5286 vmolr |= E1000_VMOLR_ROPE;
5288 if (netdev->flags & IFF_ALLMULTI) {
5289 rctl |= E1000_RCTL_MPE;
5290 vmolr |= E1000_VMOLR_MPME;
5292 /* Write addresses to the MTA, if the attempt fails
5293 * then we should just turn on promiscuous mode so
5294 * that we can at least receive multicast traffic
5296 count = igb_write_mc_addr_list(netdev);
5298 rctl |= E1000_RCTL_MPE;
5299 vmolr |= E1000_VMOLR_MPME;
5301 vmolr |= E1000_VMOLR_ROMPE;
5306 /* Write addresses to available RAR registers, if there is not
5307 * sufficient space to store all the addresses then enable
5308 * unicast promiscuous mode
5310 if (__dev_uc_sync(netdev, igb_uc_sync, igb_uc_unsync)) {
5311 rctl |= E1000_RCTL_UPE;
5312 vmolr |= E1000_VMOLR_ROPE;
5315 /* enable VLAN filtering by default */
5316 rctl |= E1000_RCTL_VFE;
5318 /* disable VLAN filtering for modes that require it */
5319 if ((netdev->flags & IFF_PROMISC) ||
5320 (netdev->features & NETIF_F_RXALL)) {
5321 /* if we fail to set all rules then just clear VFE */
5322 if (igb_vlan_promisc_enable(adapter))
5323 rctl &= ~E1000_RCTL_VFE;
5325 igb_vlan_promisc_disable(adapter);
5328 /* update state of unicast, multicast, and VLAN filtering modes */
5329 rctl |= rd32(E1000_RCTL) & ~(E1000_RCTL_UPE | E1000_RCTL_MPE |
5331 wr32(E1000_RCTL, rctl);
5333 #if (PAGE_SIZE < 8192)
5334 if (!adapter->vfs_allocated_count) {
5335 if (adapter->max_frame_size <= IGB_MAX_FRAME_BUILD_SKB)
5336 rlpml = IGB_MAX_FRAME_BUILD_SKB;
5339 wr32(E1000_RLPML, rlpml);
5341 /* In order to support SR-IOV and eventually VMDq it is necessary to set
5342 * the VMOLR to enable the appropriate modes. Without this workaround
5343 * we will have issues with VLAN tag stripping not being done for frames
5344 * that are only arriving because we are the default pool
5346 if ((hw->mac.type < e1000_82576) || (hw->mac.type > e1000_i350))
5349 /* set UTA to appropriate mode */
5350 igb_set_uta(adapter, !!(vmolr & E1000_VMOLR_ROPE));
5352 vmolr |= rd32(E1000_VMOLR(vfn)) &
5353 ~(E1000_VMOLR_ROPE | E1000_VMOLR_MPME | E1000_VMOLR_ROMPE);
5355 /* enable Rx jumbo frames, restrict as needed to support build_skb */
5356 vmolr &= ~E1000_VMOLR_RLPML_MASK;
5357 #if (PAGE_SIZE < 8192)
5358 if (adapter->max_frame_size <= IGB_MAX_FRAME_BUILD_SKB)
5359 vmolr |= IGB_MAX_FRAME_BUILD_SKB;
5362 vmolr |= MAX_JUMBO_FRAME_SIZE;
5363 vmolr |= E1000_VMOLR_LPE;
5365 wr32(E1000_VMOLR(vfn), vmolr);
5367 igb_restore_vf_multicasts(adapter);
5370 static void igb_check_wvbr(struct igb_adapter *adapter)
5372 struct e1000_hw *hw = &adapter->hw;
5375 switch (hw->mac.type) {
5378 wvbr = rd32(E1000_WVBR);
5386 adapter->wvbr |= wvbr;
5389 #define IGB_STAGGERED_QUEUE_OFFSET 8
5391 static void igb_spoof_check(struct igb_adapter *adapter)
5398 for (j = 0; j < adapter->vfs_allocated_count; j++) {
5399 if (adapter->wvbr & BIT(j) ||
5400 adapter->wvbr & BIT(j + IGB_STAGGERED_QUEUE_OFFSET)) {
5401 dev_warn(&adapter->pdev->dev,
5402 "Spoof event(s) detected on VF %d\n", j);
5405 BIT(j + IGB_STAGGERED_QUEUE_OFFSET));
5410 /* Need to wait a few seconds after link up to get diagnostic information from
5413 static void igb_update_phy_info(struct timer_list *t)
5415 struct igb_adapter *adapter = from_timer(adapter, t, phy_info_timer);
5416 igb_get_phy_info(&adapter->hw);
5420 * igb_has_link - check shared code for link and determine up/down
5421 * @adapter: pointer to driver private info
5423 bool igb_has_link(struct igb_adapter *adapter)
5425 struct e1000_hw *hw = &adapter->hw;
5426 bool link_active = false;
5428 /* get_link_status is set on LSC (link status) interrupt or
5429 * rx sequence error interrupt. get_link_status will stay
5430 * false until the e1000_check_for_link establishes link
5431 * for copper adapters ONLY
5433 switch (hw->phy.media_type) {
5434 case e1000_media_type_copper:
5435 if (!hw->mac.get_link_status)
5438 case e1000_media_type_internal_serdes:
5439 hw->mac.ops.check_for_link(hw);
5440 link_active = !hw->mac.get_link_status;
5443 case e1000_media_type_unknown:
5447 if (((hw->mac.type == e1000_i210) ||
5448 (hw->mac.type == e1000_i211)) &&
5449 (hw->phy.id == I210_I_PHY_ID)) {
5450 if (!netif_carrier_ok(adapter->netdev)) {
5451 adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
5452 } else if (!(adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)) {
5453 adapter->flags |= IGB_FLAG_NEED_LINK_UPDATE;
5454 adapter->link_check_timeout = jiffies;
5461 static bool igb_thermal_sensor_event(struct e1000_hw *hw, u32 event)
5464 u32 ctrl_ext, thstat;
5466 /* check for thermal sensor event on i350 copper only */
5467 if (hw->mac.type == e1000_i350) {
5468 thstat = rd32(E1000_THSTAT);
5469 ctrl_ext = rd32(E1000_CTRL_EXT);
5471 if ((hw->phy.media_type == e1000_media_type_copper) &&
5472 !(ctrl_ext & E1000_CTRL_EXT_LINK_MODE_SGMII))
5473 ret = !!(thstat & event);
5480 * igb_check_lvmmc - check for malformed packets received
5481 * and indicated in LVMMC register
5482 * @adapter: pointer to adapter
5484 static void igb_check_lvmmc(struct igb_adapter *adapter)
5486 struct e1000_hw *hw = &adapter->hw;
5489 lvmmc = rd32(E1000_LVMMC);
5491 if (unlikely(net_ratelimit())) {
5492 netdev_warn(adapter->netdev,
5493 "malformed Tx packet detected and dropped, LVMMC:0x%08x\n",
5500 * igb_watchdog - Timer Call-back
5501 * @t: pointer to timer_list containing our private info pointer
5503 static void igb_watchdog(struct timer_list *t)
5505 struct igb_adapter *adapter = from_timer(adapter, t, watchdog_timer);
5506 /* Do the rest outside of interrupt context */
5507 schedule_work(&adapter->watchdog_task);
5510 static void igb_watchdog_task(struct work_struct *work)
5512 struct igb_adapter *adapter = container_of(work,
5515 struct e1000_hw *hw = &adapter->hw;
5516 struct e1000_phy_info *phy = &hw->phy;
5517 struct net_device *netdev = adapter->netdev;
5521 u16 phy_data, retry_count = 20;
5523 link = igb_has_link(adapter);
5525 if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE) {
5526 if (time_after(jiffies, (adapter->link_check_timeout + HZ)))
5527 adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
5532 /* Force link down if we have fiber to swap to */
5533 if (adapter->flags & IGB_FLAG_MAS_ENABLE) {
5534 if (hw->phy.media_type == e1000_media_type_copper) {
5535 connsw = rd32(E1000_CONNSW);
5536 if (!(connsw & E1000_CONNSW_AUTOSENSE_EN))
5541 /* Perform a reset if the media type changed. */
5542 if (hw->dev_spec._82575.media_changed) {
5543 hw->dev_spec._82575.media_changed = false;
5544 adapter->flags |= IGB_FLAG_MEDIA_RESET;
5547 /* Cancel scheduled suspend requests. */
5548 pm_runtime_resume(netdev->dev.parent);
5550 if (!netif_carrier_ok(netdev)) {
5553 hw->mac.ops.get_speed_and_duplex(hw,
5554 &adapter->link_speed,
5555 &adapter->link_duplex);
5557 ctrl = rd32(E1000_CTRL);
5558 /* Links status message must follow this format */
5560 "igb: %s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
5562 adapter->link_speed,
5563 adapter->link_duplex == FULL_DUPLEX ?
5565 (ctrl & E1000_CTRL_TFCE) &&
5566 (ctrl & E1000_CTRL_RFCE) ? "RX/TX" :
5567 (ctrl & E1000_CTRL_RFCE) ? "RX" :
5568 (ctrl & E1000_CTRL_TFCE) ? "TX" : "None");
5570 /* disable EEE if enabled */
5571 if ((adapter->flags & IGB_FLAG_EEE) &&
5572 (adapter->link_duplex == HALF_DUPLEX)) {
5573 dev_info(&adapter->pdev->dev,
5574 "EEE Disabled: unsupported at half duplex. Re-enable using ethtool when at full duplex.\n");
5575 adapter->hw.dev_spec._82575.eee_disable = true;
5576 adapter->flags &= ~IGB_FLAG_EEE;
5579 /* check if SmartSpeed worked */
5580 igb_check_downshift(hw);
5581 if (phy->speed_downgraded)
5582 netdev_warn(netdev, "Link Speed was downgraded by SmartSpeed\n");
5584 /* check for thermal sensor event */
5585 if (igb_thermal_sensor_event(hw,
5586 E1000_THSTAT_LINK_THROTTLE))
5587 netdev_info(netdev, "The network adapter link speed was downshifted because it overheated\n");
5589 /* adjust timeout factor according to speed/duplex */
5590 adapter->tx_timeout_factor = 1;
5591 switch (adapter->link_speed) {
5593 adapter->tx_timeout_factor = 14;
5596 /* maybe add some timeout factor ? */
5600 if (adapter->link_speed != SPEED_1000 ||
5601 !hw->phy.ops.read_reg)
5604 /* wait for Remote receiver status OK */
5606 if (!igb_read_phy_reg(hw, PHY_1000T_STATUS,
5608 if (!(phy_data & SR_1000T_REMOTE_RX_STATUS) &&
5612 goto retry_read_status;
5613 } else if (!retry_count) {
5614 dev_err(&adapter->pdev->dev, "exceed max 2 second\n");
5617 dev_err(&adapter->pdev->dev, "read 1000Base-T Status Reg\n");
5620 netif_carrier_on(netdev);
5622 igb_ping_all_vfs(adapter);
5623 igb_check_vf_rate_limit(adapter);
5625 /* link state has changed, schedule phy info update */
5626 if (!test_bit(__IGB_DOWN, &adapter->state))
5627 mod_timer(&adapter->phy_info_timer,
5628 round_jiffies(jiffies + 2 * HZ));
5631 if (netif_carrier_ok(netdev)) {
5632 adapter->link_speed = 0;
5633 adapter->link_duplex = 0;
5635 /* check for thermal sensor event */
5636 if (igb_thermal_sensor_event(hw,
5637 E1000_THSTAT_PWR_DOWN)) {
5638 netdev_err(netdev, "The network adapter was stopped because it overheated\n");
5641 /* Links status message must follow this format */
5642 netdev_info(netdev, "igb: %s NIC Link is Down\n",
5644 netif_carrier_off(netdev);
5646 igb_ping_all_vfs(adapter);
5648 /* link state has changed, schedule phy info update */
5649 if (!test_bit(__IGB_DOWN, &adapter->state))
5650 mod_timer(&adapter->phy_info_timer,
5651 round_jiffies(jiffies + 2 * HZ));
5653 /* link is down, time to check for alternate media */
5654 if (adapter->flags & IGB_FLAG_MAS_ENABLE) {
5655 igb_check_swap_media(adapter);
5656 if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
5657 schedule_work(&adapter->reset_task);
5658 /* return immediately */
5662 pm_schedule_suspend(netdev->dev.parent,
5665 /* also check for alternate media here */
5666 } else if (!netif_carrier_ok(netdev) &&
5667 (adapter->flags & IGB_FLAG_MAS_ENABLE)) {
5668 igb_check_swap_media(adapter);
5669 if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
5670 schedule_work(&adapter->reset_task);
5671 /* return immediately */
5677 spin_lock(&adapter->stats64_lock);
5678 igb_update_stats(adapter);
5679 spin_unlock(&adapter->stats64_lock);
5681 for (i = 0; i < adapter->num_tx_queues; i++) {
5682 struct igb_ring *tx_ring = adapter->tx_ring[i];
5683 if (!netif_carrier_ok(netdev)) {
5684 /* We've lost link, so the controller stops DMA,
5685 * but we've got queued Tx work that's never going
5686 * to get done, so reset controller to flush Tx.
5687 * (Do the reset outside of interrupt context).
5689 if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) {
5690 adapter->tx_timeout_count++;
5691 schedule_work(&adapter->reset_task);
5692 /* return immediately since reset is imminent */
5697 /* Force detection of hung controller every watchdog period */
5698 set_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
5701 /* Cause software interrupt to ensure Rx ring is cleaned */
5702 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
5705 for (i = 0; i < adapter->num_q_vectors; i++)
5706 eics |= adapter->q_vector[i]->eims_value;
5707 wr32(E1000_EICS, eics);
5709 wr32(E1000_ICS, E1000_ICS_RXDMT0);
5712 igb_spoof_check(adapter);
5713 igb_ptp_rx_hang(adapter);
5714 igb_ptp_tx_hang(adapter);
5716 /* Check LVMMC register on i350/i354 only */
5717 if ((adapter->hw.mac.type == e1000_i350) ||
5718 (adapter->hw.mac.type == e1000_i354))
5719 igb_check_lvmmc(adapter);
5721 /* Reset the timer */
5722 if (!test_bit(__IGB_DOWN, &adapter->state)) {
5723 if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)
5724 mod_timer(&adapter->watchdog_timer,
5725 round_jiffies(jiffies + HZ));
5727 mod_timer(&adapter->watchdog_timer,
5728 round_jiffies(jiffies + 2 * HZ));
5732 enum latency_range {
5736 latency_invalid = 255
5740 * igb_update_ring_itr - update the dynamic ITR value based on packet size
5741 * @q_vector: pointer to q_vector
5743 * Stores a new ITR value based on strictly on packet size. This
5744 * algorithm is less sophisticated than that used in igb_update_itr,
5745 * due to the difficulty of synchronizing statistics across multiple
5746 * receive rings. The divisors and thresholds used by this function
5747 * were determined based on theoretical maximum wire speed and testing
5748 * data, in order to minimize response time while increasing bulk
5750 * This functionality is controlled by ethtool's coalescing settings.
5751 * NOTE: This function is called only when operating in a multiqueue
5752 * receive environment.
5754 static void igb_update_ring_itr(struct igb_q_vector *q_vector)
5756 int new_val = q_vector->itr_val;
5757 int avg_wire_size = 0;
5758 struct igb_adapter *adapter = q_vector->adapter;
5759 unsigned int packets;
5761 /* For non-gigabit speeds, just fix the interrupt rate at 4000
5762 * ints/sec - ITR timer value of 120 ticks.
5764 if (adapter->link_speed != SPEED_1000) {
5765 new_val = IGB_4K_ITR;
5769 packets = q_vector->rx.total_packets;
5771 avg_wire_size = q_vector->rx.total_bytes / packets;
5773 packets = q_vector->tx.total_packets;
5775 avg_wire_size = max_t(u32, avg_wire_size,
5776 q_vector->tx.total_bytes / packets);
5778 /* if avg_wire_size isn't set no work was done */
5782 /* Add 24 bytes to size to account for CRC, preamble, and gap */
5783 avg_wire_size += 24;
5785 /* Don't starve jumbo frames */
5786 avg_wire_size = min(avg_wire_size, 3000);
5788 /* Give a little boost to mid-size frames */
5789 if ((avg_wire_size > 300) && (avg_wire_size < 1200))
5790 new_val = avg_wire_size / 3;
5792 new_val = avg_wire_size / 2;
5794 /* conservative mode (itr 3) eliminates the lowest_latency setting */
5795 if (new_val < IGB_20K_ITR &&
5796 ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
5797 (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
5798 new_val = IGB_20K_ITR;
5801 if (new_val != q_vector->itr_val) {
5802 q_vector->itr_val = new_val;
5803 q_vector->set_itr = 1;
5806 q_vector->rx.total_bytes = 0;
5807 q_vector->rx.total_packets = 0;
5808 q_vector->tx.total_bytes = 0;
5809 q_vector->tx.total_packets = 0;
5813 * igb_update_itr - update the dynamic ITR value based on statistics
5814 * @q_vector: pointer to q_vector
5815 * @ring_container: ring info to update the itr for
5817 * Stores a new ITR value based on packets and byte
5818 * counts during the last interrupt. The advantage of per interrupt
5819 * computation is faster updates and more accurate ITR for the current
5820 * traffic pattern. Constants in this function were computed
5821 * based on theoretical maximum wire speed and thresholds were set based
5822 * on testing data as well as attempting to minimize response time
5823 * while increasing bulk throughput.
5824 * This functionality is controlled by ethtool's coalescing settings.
5825 * NOTE: These calculations are only valid when operating in a single-
5826 * queue environment.
5828 static void igb_update_itr(struct igb_q_vector *q_vector,
5829 struct igb_ring_container *ring_container)
5831 unsigned int packets = ring_container->total_packets;
5832 unsigned int bytes = ring_container->total_bytes;
5833 u8 itrval = ring_container->itr;
5835 /* no packets, exit with status unchanged */
5840 case lowest_latency:
5841 /* handle TSO and jumbo frames */
5842 if (bytes/packets > 8000)
5843 itrval = bulk_latency;
5844 else if ((packets < 5) && (bytes > 512))
5845 itrval = low_latency;
5847 case low_latency: /* 50 usec aka 20000 ints/s */
5848 if (bytes > 10000) {
5849 /* this if handles the TSO accounting */
5850 if (bytes/packets > 8000)
5851 itrval = bulk_latency;
5852 else if ((packets < 10) || ((bytes/packets) > 1200))
5853 itrval = bulk_latency;
5854 else if ((packets > 35))
5855 itrval = lowest_latency;
5856 } else if (bytes/packets > 2000) {
5857 itrval = bulk_latency;
5858 } else if (packets <= 2 && bytes < 512) {
5859 itrval = lowest_latency;
5862 case bulk_latency: /* 250 usec aka 4000 ints/s */
5863 if (bytes > 25000) {
5865 itrval = low_latency;
5866 } else if (bytes < 1500) {
5867 itrval = low_latency;
5872 /* clear work counters since we have the values we need */
5873 ring_container->total_bytes = 0;
5874 ring_container->total_packets = 0;
5876 /* write updated itr to ring container */
5877 ring_container->itr = itrval;
5880 static void igb_set_itr(struct igb_q_vector *q_vector)
5882 struct igb_adapter *adapter = q_vector->adapter;
5883 u32 new_itr = q_vector->itr_val;
5886 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
5887 if (adapter->link_speed != SPEED_1000) {
5889 new_itr = IGB_4K_ITR;
5893 igb_update_itr(q_vector, &q_vector->tx);
5894 igb_update_itr(q_vector, &q_vector->rx);
5896 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
5898 /* conservative mode (itr 3) eliminates the lowest_latency setting */
5899 if (current_itr == lowest_latency &&
5900 ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
5901 (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
5902 current_itr = low_latency;
5904 switch (current_itr) {
5905 /* counts and packets in update_itr are dependent on these numbers */
5906 case lowest_latency:
5907 new_itr = IGB_70K_ITR; /* 70,000 ints/sec */
5910 new_itr = IGB_20K_ITR; /* 20,000 ints/sec */
5913 new_itr = IGB_4K_ITR; /* 4,000 ints/sec */
5920 if (new_itr != q_vector->itr_val) {
5921 /* this attempts to bias the interrupt rate towards Bulk
5922 * by adding intermediate steps when interrupt rate is
5925 new_itr = new_itr > q_vector->itr_val ?
5926 max((new_itr * q_vector->itr_val) /
5927 (new_itr + (q_vector->itr_val >> 2)),
5929 /* Don't write the value here; it resets the adapter's
5930 * internal timer, and causes us to delay far longer than
5931 * we should between interrupts. Instead, we write the ITR
5932 * value at the beginning of the next interrupt so the timing
5933 * ends up being correct.
5935 q_vector->itr_val = new_itr;
5936 q_vector->set_itr = 1;
5940 static void igb_tx_ctxtdesc(struct igb_ring *tx_ring,
5941 struct igb_tx_buffer *first,
5942 u32 vlan_macip_lens, u32 type_tucmd,
5945 struct e1000_adv_tx_context_desc *context_desc;
5946 u16 i = tx_ring->next_to_use;
5947 struct timespec64 ts;
5949 context_desc = IGB_TX_CTXTDESC(tx_ring, i);
5952 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
5954 /* set bits to identify this as an advanced context descriptor */
5955 type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
5957 /* For 82575, context index must be unique per ring. */
5958 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
5959 mss_l4len_idx |= tx_ring->reg_idx << 4;
5961 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
5962 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd);
5963 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
5965 /* We assume there is always a valid tx time available. Invalid times
5966 * should have been handled by the upper layers.
5968 if (tx_ring->launchtime_enable) {
5969 ts = ktime_to_timespec64(first->skb->tstamp);
5970 skb_txtime_consumed(first->skb);
5971 context_desc->seqnum_seed = cpu_to_le32(ts.tv_nsec / 32);
5973 context_desc->seqnum_seed = 0;
5977 static int igb_tso(struct igb_ring *tx_ring,
5978 struct igb_tx_buffer *first,
5981 u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
5982 struct sk_buff *skb = first->skb;
5993 u32 paylen, l4_offset;
5996 if (skb->ip_summed != CHECKSUM_PARTIAL)
5999 if (!skb_is_gso(skb))
6002 err = skb_cow_head(skb, 0);
6006 ip.hdr = skb_network_header(skb);
6007 l4.hdr = skb_checksum_start(skb);
6009 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
6010 type_tucmd = (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ?
6011 E1000_ADVTXD_TUCMD_L4T_UDP : E1000_ADVTXD_TUCMD_L4T_TCP;
6013 /* initialize outer IP header fields */
6014 if (ip.v4->version == 4) {
6015 unsigned char *csum_start = skb_checksum_start(skb);
6016 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
6018 /* IP header will have to cancel out any data that
6019 * is not a part of the outer IP header
6021 ip.v4->check = csum_fold(csum_partial(trans_start,
6022 csum_start - trans_start,
6024 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
6027 first->tx_flags |= IGB_TX_FLAGS_TSO |
6031 ip.v6->payload_len = 0;
6032 first->tx_flags |= IGB_TX_FLAGS_TSO |
6036 /* determine offset of inner transport header */
6037 l4_offset = l4.hdr - skb->data;
6039 /* remove payload length from inner checksum */
6040 paylen = skb->len - l4_offset;
6041 if (type_tucmd & E1000_ADVTXD_TUCMD_L4T_TCP) {
6042 /* compute length of segmentation header */
6043 *hdr_len = (l4.tcp->doff * 4) + l4_offset;
6044 csum_replace_by_diff(&l4.tcp->check,
6045 (__force __wsum)htonl(paylen));
6047 /* compute length of segmentation header */
6048 *hdr_len = sizeof(*l4.udp) + l4_offset;
6049 csum_replace_by_diff(&l4.udp->check,
6050 (__force __wsum)htonl(paylen));
6053 /* update gso size and bytecount with header size */
6054 first->gso_segs = skb_shinfo(skb)->gso_segs;
6055 first->bytecount += (first->gso_segs - 1) * *hdr_len;
6058 mss_l4len_idx = (*hdr_len - l4_offset) << E1000_ADVTXD_L4LEN_SHIFT;
6059 mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT;
6061 /* VLAN MACLEN IPLEN */
6062 vlan_macip_lens = l4.hdr - ip.hdr;
6063 vlan_macip_lens |= (ip.hdr - skb->data) << E1000_ADVTXD_MACLEN_SHIFT;
6064 vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
6066 igb_tx_ctxtdesc(tx_ring, first, vlan_macip_lens,
6067 type_tucmd, mss_l4len_idx);
6072 static void igb_tx_csum(struct igb_ring *tx_ring, struct igb_tx_buffer *first)
6074 struct sk_buff *skb = first->skb;
6075 u32 vlan_macip_lens = 0;
6078 if (skb->ip_summed != CHECKSUM_PARTIAL) {
6080 if (!(first->tx_flags & IGB_TX_FLAGS_VLAN) &&
6081 !tx_ring->launchtime_enable)
6086 switch (skb->csum_offset) {
6087 case offsetof(struct tcphdr, check):
6088 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
6090 case offsetof(struct udphdr, check):
6092 case offsetof(struct sctphdr, checksum):
6093 /* validate that this is actually an SCTP request */
6094 if (skb_csum_is_sctp(skb)) {
6095 type_tucmd = E1000_ADVTXD_TUCMD_L4T_SCTP;
6100 skb_checksum_help(skb);
6104 /* update TX checksum flag */
6105 first->tx_flags |= IGB_TX_FLAGS_CSUM;
6106 vlan_macip_lens = skb_checksum_start_offset(skb) -
6107 skb_network_offset(skb);
6109 vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
6110 vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
6112 igb_tx_ctxtdesc(tx_ring, first, vlan_macip_lens, type_tucmd, 0);
6115 #define IGB_SET_FLAG(_input, _flag, _result) \
6116 ((_flag <= _result) ? \
6117 ((u32)(_input & _flag) * (_result / _flag)) : \
6118 ((u32)(_input & _flag) / (_flag / _result)))
6120 static u32 igb_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
6122 /* set type for advanced descriptor with frame checksum insertion */
6123 u32 cmd_type = E1000_ADVTXD_DTYP_DATA |
6124 E1000_ADVTXD_DCMD_DEXT |
6125 E1000_ADVTXD_DCMD_IFCS;
6127 /* set HW vlan bit if vlan is present */
6128 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_VLAN,
6129 (E1000_ADVTXD_DCMD_VLE));
6131 /* set segmentation bits for TSO */
6132 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSO,
6133 (E1000_ADVTXD_DCMD_TSE));
6135 /* set timestamp bit if present */
6136 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSTAMP,
6137 (E1000_ADVTXD_MAC_TSTAMP));
6139 /* insert frame checksum */
6140 cmd_type ^= IGB_SET_FLAG(skb->no_fcs, 1, E1000_ADVTXD_DCMD_IFCS);
6145 static void igb_tx_olinfo_status(struct igb_ring *tx_ring,
6146 union e1000_adv_tx_desc *tx_desc,
6147 u32 tx_flags, unsigned int paylen)
6149 u32 olinfo_status = paylen << E1000_ADVTXD_PAYLEN_SHIFT;
6151 /* 82575 requires a unique index per ring */
6152 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
6153 olinfo_status |= tx_ring->reg_idx << 4;
6155 /* insert L4 checksum */
6156 olinfo_status |= IGB_SET_FLAG(tx_flags,
6158 (E1000_TXD_POPTS_TXSM << 8));
6160 /* insert IPv4 checksum */
6161 olinfo_status |= IGB_SET_FLAG(tx_flags,
6163 (E1000_TXD_POPTS_IXSM << 8));
6165 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
6168 static int __igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
6170 struct net_device *netdev = tx_ring->netdev;
6172 netif_stop_subqueue(netdev, tx_ring->queue_index);
6174 /* Herbert's original patch had:
6175 * smp_mb__after_netif_stop_queue();
6176 * but since that doesn't exist yet, just open code it.
6180 /* We need to check again in a case another CPU has just
6181 * made room available.
6183 if (igb_desc_unused(tx_ring) < size)
6187 netif_wake_subqueue(netdev, tx_ring->queue_index);
6189 u64_stats_update_begin(&tx_ring->tx_syncp2);
6190 tx_ring->tx_stats.restart_queue2++;
6191 u64_stats_update_end(&tx_ring->tx_syncp2);
6196 static inline int igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
6198 if (igb_desc_unused(tx_ring) >= size)
6200 return __igb_maybe_stop_tx(tx_ring, size);
6203 static int igb_tx_map(struct igb_ring *tx_ring,
6204 struct igb_tx_buffer *first,
6207 struct sk_buff *skb = first->skb;
6208 struct igb_tx_buffer *tx_buffer;
6209 union e1000_adv_tx_desc *tx_desc;
6212 unsigned int data_len, size;
6213 u32 tx_flags = first->tx_flags;
6214 u32 cmd_type = igb_tx_cmd_type(skb, tx_flags);
6215 u16 i = tx_ring->next_to_use;
6217 tx_desc = IGB_TX_DESC(tx_ring, i);
6219 igb_tx_olinfo_status(tx_ring, tx_desc, tx_flags, skb->len - hdr_len);
6221 size = skb_headlen(skb);
6222 data_len = skb->data_len;
6224 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
6228 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
6229 if (dma_mapping_error(tx_ring->dev, dma))
6232 /* record length, and DMA address */
6233 dma_unmap_len_set(tx_buffer, len, size);
6234 dma_unmap_addr_set(tx_buffer, dma, dma);
6236 tx_desc->read.buffer_addr = cpu_to_le64(dma);
6238 while (unlikely(size > IGB_MAX_DATA_PER_TXD)) {
6239 tx_desc->read.cmd_type_len =
6240 cpu_to_le32(cmd_type ^ IGB_MAX_DATA_PER_TXD);
6244 if (i == tx_ring->count) {
6245 tx_desc = IGB_TX_DESC(tx_ring, 0);
6248 tx_desc->read.olinfo_status = 0;
6250 dma += IGB_MAX_DATA_PER_TXD;
6251 size -= IGB_MAX_DATA_PER_TXD;
6253 tx_desc->read.buffer_addr = cpu_to_le64(dma);
6256 if (likely(!data_len))
6259 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
6263 if (i == tx_ring->count) {
6264 tx_desc = IGB_TX_DESC(tx_ring, 0);
6267 tx_desc->read.olinfo_status = 0;
6269 size = skb_frag_size(frag);
6272 dma = skb_frag_dma_map(tx_ring->dev, frag, 0,
6273 size, DMA_TO_DEVICE);
6275 tx_buffer = &tx_ring->tx_buffer_info[i];
6278 /* write last descriptor with RS and EOP bits */
6279 cmd_type |= size | IGB_TXD_DCMD;
6280 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
6282 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
6284 /* set the timestamp */
6285 first->time_stamp = jiffies;
6287 skb_tx_timestamp(skb);
6289 /* Force memory writes to complete before letting h/w know there
6290 * are new descriptors to fetch. (Only applicable for weak-ordered
6291 * memory model archs, such as IA-64).
6293 * We also need this memory barrier to make certain all of the
6294 * status bits have been updated before next_to_watch is written.
6298 /* set next_to_watch value indicating a packet is present */
6299 first->next_to_watch = tx_desc;
6302 if (i == tx_ring->count)
6305 tx_ring->next_to_use = i;
6307 /* Make sure there is space in the ring for the next send. */
6308 igb_maybe_stop_tx(tx_ring, DESC_NEEDED);
6310 if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more()) {
6311 writel(i, tx_ring->tail);
6316 dev_err(tx_ring->dev, "TX DMA map failed\n");
6317 tx_buffer = &tx_ring->tx_buffer_info[i];
6319 /* clear dma mappings for failed tx_buffer_info map */
6320 while (tx_buffer != first) {
6321 if (dma_unmap_len(tx_buffer, len))
6322 dma_unmap_page(tx_ring->dev,
6323 dma_unmap_addr(tx_buffer, dma),
6324 dma_unmap_len(tx_buffer, len),
6326 dma_unmap_len_set(tx_buffer, len, 0);
6329 i += tx_ring->count;
6330 tx_buffer = &tx_ring->tx_buffer_info[i];
6333 if (dma_unmap_len(tx_buffer, len))
6334 dma_unmap_single(tx_ring->dev,
6335 dma_unmap_addr(tx_buffer, dma),
6336 dma_unmap_len(tx_buffer, len),
6338 dma_unmap_len_set(tx_buffer, len, 0);
6340 dev_kfree_skb_any(tx_buffer->skb);
6341 tx_buffer->skb = NULL;
6343 tx_ring->next_to_use = i;
6348 int igb_xmit_xdp_ring(struct igb_adapter *adapter,
6349 struct igb_ring *tx_ring,
6350 struct xdp_frame *xdpf)
6352 struct skb_shared_info *sinfo = xdp_get_shared_info_from_frame(xdpf);
6353 u8 nr_frags = unlikely(xdp_frame_has_frags(xdpf)) ? sinfo->nr_frags : 0;
6354 u16 count, i, index = tx_ring->next_to_use;
6355 struct igb_tx_buffer *tx_head = &tx_ring->tx_buffer_info[index];
6356 struct igb_tx_buffer *tx_buffer = tx_head;
6357 union e1000_adv_tx_desc *tx_desc = IGB_TX_DESC(tx_ring, index);
6358 u32 len = xdpf->len, cmd_type, olinfo_status;
6359 void *data = xdpf->data;
6361 count = TXD_USE_COUNT(len);
6362 for (i = 0; i < nr_frags; i++)
6363 count += TXD_USE_COUNT(skb_frag_size(&sinfo->frags[i]));
6365 if (igb_maybe_stop_tx(tx_ring, count + 3))
6366 return IGB_XDP_CONSUMED;
6369 /* record the location of the first descriptor for this packet */
6370 tx_head->bytecount = xdp_get_frame_len(xdpf);
6371 tx_head->type = IGB_TYPE_XDP;
6372 tx_head->gso_segs = 1;
6373 tx_head->xdpf = xdpf;
6375 olinfo_status = tx_head->bytecount << E1000_ADVTXD_PAYLEN_SHIFT;
6376 /* 82575 requires a unique index per ring */
6377 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
6378 olinfo_status |= tx_ring->reg_idx << 4;
6379 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
6384 dma = dma_map_single(tx_ring->dev, data, len, DMA_TO_DEVICE);
6385 if (dma_mapping_error(tx_ring->dev, dma))
6388 /* record length, and DMA address */
6389 dma_unmap_len_set(tx_buffer, len, len);
6390 dma_unmap_addr_set(tx_buffer, dma, dma);
6392 /* put descriptor type bits */
6393 cmd_type = E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_DEXT |
6394 E1000_ADVTXD_DCMD_IFCS | len;
6396 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
6397 tx_desc->read.buffer_addr = cpu_to_le64(dma);
6399 tx_buffer->protocol = 0;
6401 if (++index == tx_ring->count)
6407 tx_buffer = &tx_ring->tx_buffer_info[index];
6408 tx_desc = IGB_TX_DESC(tx_ring, index);
6409 tx_desc->read.olinfo_status = 0;
6411 data = skb_frag_address(&sinfo->frags[i]);
6412 len = skb_frag_size(&sinfo->frags[i]);
6415 tx_desc->read.cmd_type_len |= cpu_to_le32(IGB_TXD_DCMD);
6417 netdev_tx_sent_queue(txring_txq(tx_ring), tx_head->bytecount);
6418 /* set the timestamp */
6419 tx_head->time_stamp = jiffies;
6421 /* Avoid any potential race with xdp_xmit and cleanup */
6424 /* set next_to_watch value indicating a packet is present */
6425 tx_head->next_to_watch = tx_desc;
6426 tx_ring->next_to_use = index;
6428 /* Make sure there is space in the ring for the next send. */
6429 igb_maybe_stop_tx(tx_ring, DESC_NEEDED);
6431 if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more())
6432 writel(index, tx_ring->tail);
6438 tx_buffer = &tx_ring->tx_buffer_info[index];
6439 if (dma_unmap_len(tx_buffer, len))
6440 dma_unmap_page(tx_ring->dev,
6441 dma_unmap_addr(tx_buffer, dma),
6442 dma_unmap_len(tx_buffer, len),
6444 dma_unmap_len_set(tx_buffer, len, 0);
6445 if (tx_buffer == tx_head)
6449 index += tx_ring->count;
6453 return IGB_XDP_CONSUMED;
6456 netdev_tx_t igb_xmit_frame_ring(struct sk_buff *skb,
6457 struct igb_ring *tx_ring)
6459 struct igb_tx_buffer *first;
6463 u16 count = TXD_USE_COUNT(skb_headlen(skb));
6464 __be16 protocol = vlan_get_protocol(skb);
6467 /* need: 1 descriptor per page * PAGE_SIZE/IGB_MAX_DATA_PER_TXD,
6468 * + 1 desc for skb_headlen/IGB_MAX_DATA_PER_TXD,
6469 * + 2 desc gap to keep tail from touching head,
6470 * + 1 desc for context descriptor,
6471 * otherwise try next time
6473 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6474 count += TXD_USE_COUNT(skb_frag_size(
6475 &skb_shinfo(skb)->frags[f]));
6477 if (igb_maybe_stop_tx(tx_ring, count + 3)) {
6478 /* this is a hard error */
6479 return NETDEV_TX_BUSY;
6482 /* record the location of the first descriptor for this packet */
6483 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
6484 first->type = IGB_TYPE_SKB;
6486 first->bytecount = skb->len;
6487 first->gso_segs = 1;
6489 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
6490 struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
6492 if (adapter->tstamp_config.tx_type == HWTSTAMP_TX_ON &&
6493 !test_and_set_bit_lock(__IGB_PTP_TX_IN_PROGRESS,
6495 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
6496 tx_flags |= IGB_TX_FLAGS_TSTAMP;
6498 adapter->ptp_tx_skb = skb_get(skb);
6499 adapter->ptp_tx_start = jiffies;
6500 if (adapter->hw.mac.type == e1000_82576)
6501 schedule_work(&adapter->ptp_tx_work);
6503 adapter->tx_hwtstamp_skipped++;
6507 if (skb_vlan_tag_present(skb)) {
6508 tx_flags |= IGB_TX_FLAGS_VLAN;
6509 tx_flags |= (skb_vlan_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
6512 /* record initial flags and protocol */
6513 first->tx_flags = tx_flags;
6514 first->protocol = protocol;
6516 tso = igb_tso(tx_ring, first, &hdr_len);
6520 igb_tx_csum(tx_ring, first);
6522 if (igb_tx_map(tx_ring, first, hdr_len))
6523 goto cleanup_tx_tstamp;
6525 return NETDEV_TX_OK;
6528 dev_kfree_skb_any(first->skb);
6531 if (unlikely(tx_flags & IGB_TX_FLAGS_TSTAMP)) {
6532 struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
6534 dev_kfree_skb_any(adapter->ptp_tx_skb);
6535 adapter->ptp_tx_skb = NULL;
6536 if (adapter->hw.mac.type == e1000_82576)
6537 cancel_work_sync(&adapter->ptp_tx_work);
6538 clear_bit_unlock(__IGB_PTP_TX_IN_PROGRESS, &adapter->state);
6541 return NETDEV_TX_OK;
6544 static inline struct igb_ring *igb_tx_queue_mapping(struct igb_adapter *adapter,
6545 struct sk_buff *skb)
6547 unsigned int r_idx = skb->queue_mapping;
6549 if (r_idx >= adapter->num_tx_queues)
6550 r_idx = r_idx % adapter->num_tx_queues;
6552 return adapter->tx_ring[r_idx];
6555 static netdev_tx_t igb_xmit_frame(struct sk_buff *skb,
6556 struct net_device *netdev)
6558 struct igb_adapter *adapter = netdev_priv(netdev);
6560 /* The minimum packet size with TCTL.PSP set is 17 so pad the skb
6561 * in order to meet this minimum size requirement.
6563 if (skb_put_padto(skb, 17))
6564 return NETDEV_TX_OK;
6566 return igb_xmit_frame_ring(skb, igb_tx_queue_mapping(adapter, skb));
6570 * igb_tx_timeout - Respond to a Tx Hang
6571 * @netdev: network interface device structure
6572 * @txqueue: number of the Tx queue that hung (unused)
6574 static void igb_tx_timeout(struct net_device *netdev, unsigned int __always_unused txqueue)
6576 struct igb_adapter *adapter = netdev_priv(netdev);
6577 struct e1000_hw *hw = &adapter->hw;
6579 /* Do the reset outside of interrupt context */
6580 adapter->tx_timeout_count++;
6582 if (hw->mac.type >= e1000_82580)
6583 hw->dev_spec._82575.global_device_reset = true;
6585 schedule_work(&adapter->reset_task);
6587 (adapter->eims_enable_mask & ~adapter->eims_other));
6590 static void igb_reset_task(struct work_struct *work)
6592 struct igb_adapter *adapter;
6593 adapter = container_of(work, struct igb_adapter, reset_task);
6596 /* If we're already down or resetting, just bail */
6597 if (test_bit(__IGB_DOWN, &adapter->state) ||
6598 test_bit(__IGB_RESETTING, &adapter->state)) {
6604 netdev_err(adapter->netdev, "Reset adapter\n");
6605 igb_reinit_locked(adapter);
6610 * igb_get_stats64 - Get System Network Statistics
6611 * @netdev: network interface device structure
6612 * @stats: rtnl_link_stats64 pointer
6614 static void igb_get_stats64(struct net_device *netdev,
6615 struct rtnl_link_stats64 *stats)
6617 struct igb_adapter *adapter = netdev_priv(netdev);
6619 spin_lock(&adapter->stats64_lock);
6620 igb_update_stats(adapter);
6621 memcpy(stats, &adapter->stats64, sizeof(*stats));
6622 spin_unlock(&adapter->stats64_lock);
6626 * igb_change_mtu - Change the Maximum Transfer Unit
6627 * @netdev: network interface device structure
6628 * @new_mtu: new value for maximum frame size
6630 * Returns 0 on success, negative on failure
6632 static int igb_change_mtu(struct net_device *netdev, int new_mtu)
6634 struct igb_adapter *adapter = netdev_priv(netdev);
6635 int max_frame = new_mtu + IGB_ETH_PKT_HDR_PAD;
6637 if (adapter->xdp_prog) {
6640 for (i = 0; i < adapter->num_rx_queues; i++) {
6641 struct igb_ring *ring = adapter->rx_ring[i];
6643 if (max_frame > igb_rx_bufsz(ring)) {
6644 netdev_warn(adapter->netdev,
6645 "Requested MTU size is not supported with XDP. Max frame size is %d\n",
6652 /* adjust max frame to be at least the size of a standard frame */
6653 if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
6654 max_frame = ETH_FRAME_LEN + ETH_FCS_LEN;
6656 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
6657 usleep_range(1000, 2000);
6659 /* igb_down has a dependency on max_frame_size */
6660 adapter->max_frame_size = max_frame;
6662 if (netif_running(netdev))
6665 netdev_dbg(netdev, "changing MTU from %d to %d\n",
6666 netdev->mtu, new_mtu);
6667 netdev->mtu = new_mtu;
6669 if (netif_running(netdev))
6674 clear_bit(__IGB_RESETTING, &adapter->state);
6680 * igb_update_stats - Update the board statistics counters
6681 * @adapter: board private structure
6683 void igb_update_stats(struct igb_adapter *adapter)
6685 struct rtnl_link_stats64 *net_stats = &adapter->stats64;
6686 struct e1000_hw *hw = &adapter->hw;
6687 struct pci_dev *pdev = adapter->pdev;
6692 u64 _bytes, _packets;
6694 /* Prevent stats update while adapter is being reset, or if the pci
6695 * connection is down.
6697 if (adapter->link_speed == 0)
6699 if (pci_channel_offline(pdev))
6706 for (i = 0; i < adapter->num_rx_queues; i++) {
6707 struct igb_ring *ring = adapter->rx_ring[i];
6708 u32 rqdpc = rd32(E1000_RQDPC(i));
6709 if (hw->mac.type >= e1000_i210)
6710 wr32(E1000_RQDPC(i), 0);
6713 ring->rx_stats.drops += rqdpc;
6714 net_stats->rx_fifo_errors += rqdpc;
6718 start = u64_stats_fetch_begin(&ring->rx_syncp);
6719 _bytes = ring->rx_stats.bytes;
6720 _packets = ring->rx_stats.packets;
6721 } while (u64_stats_fetch_retry(&ring->rx_syncp, start));
6723 packets += _packets;
6726 net_stats->rx_bytes = bytes;
6727 net_stats->rx_packets = packets;
6731 for (i = 0; i < adapter->num_tx_queues; i++) {
6732 struct igb_ring *ring = adapter->tx_ring[i];
6734 start = u64_stats_fetch_begin(&ring->tx_syncp);
6735 _bytes = ring->tx_stats.bytes;
6736 _packets = ring->tx_stats.packets;
6737 } while (u64_stats_fetch_retry(&ring->tx_syncp, start));
6739 packets += _packets;
6741 net_stats->tx_bytes = bytes;
6742 net_stats->tx_packets = packets;
6745 /* read stats registers */
6746 adapter->stats.crcerrs += rd32(E1000_CRCERRS);
6747 adapter->stats.gprc += rd32(E1000_GPRC);
6748 adapter->stats.gorc += rd32(E1000_GORCL);
6749 rd32(E1000_GORCH); /* clear GORCL */
6750 adapter->stats.bprc += rd32(E1000_BPRC);
6751 adapter->stats.mprc += rd32(E1000_MPRC);
6752 adapter->stats.roc += rd32(E1000_ROC);
6754 adapter->stats.prc64 += rd32(E1000_PRC64);
6755 adapter->stats.prc127 += rd32(E1000_PRC127);
6756 adapter->stats.prc255 += rd32(E1000_PRC255);
6757 adapter->stats.prc511 += rd32(E1000_PRC511);
6758 adapter->stats.prc1023 += rd32(E1000_PRC1023);
6759 adapter->stats.prc1522 += rd32(E1000_PRC1522);
6760 adapter->stats.symerrs += rd32(E1000_SYMERRS);
6761 adapter->stats.sec += rd32(E1000_SEC);
6763 mpc = rd32(E1000_MPC);
6764 adapter->stats.mpc += mpc;
6765 net_stats->rx_fifo_errors += mpc;
6766 adapter->stats.scc += rd32(E1000_SCC);
6767 adapter->stats.ecol += rd32(E1000_ECOL);
6768 adapter->stats.mcc += rd32(E1000_MCC);
6769 adapter->stats.latecol += rd32(E1000_LATECOL);
6770 adapter->stats.dc += rd32(E1000_DC);
6771 adapter->stats.rlec += rd32(E1000_RLEC);
6772 adapter->stats.xonrxc += rd32(E1000_XONRXC);
6773 adapter->stats.xontxc += rd32(E1000_XONTXC);
6774 adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
6775 adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
6776 adapter->stats.fcruc += rd32(E1000_FCRUC);
6777 adapter->stats.gptc += rd32(E1000_GPTC);
6778 adapter->stats.gotc += rd32(E1000_GOTCL);
6779 rd32(E1000_GOTCH); /* clear GOTCL */
6780 adapter->stats.rnbc += rd32(E1000_RNBC);
6781 adapter->stats.ruc += rd32(E1000_RUC);
6782 adapter->stats.rfc += rd32(E1000_RFC);
6783 adapter->stats.rjc += rd32(E1000_RJC);
6784 adapter->stats.tor += rd32(E1000_TORH);
6785 adapter->stats.tot += rd32(E1000_TOTH);
6786 adapter->stats.tpr += rd32(E1000_TPR);
6788 adapter->stats.ptc64 += rd32(E1000_PTC64);
6789 adapter->stats.ptc127 += rd32(E1000_PTC127);
6790 adapter->stats.ptc255 += rd32(E1000_PTC255);
6791 adapter->stats.ptc511 += rd32(E1000_PTC511);
6792 adapter->stats.ptc1023 += rd32(E1000_PTC1023);
6793 adapter->stats.ptc1522 += rd32(E1000_PTC1522);
6795 adapter->stats.mptc += rd32(E1000_MPTC);
6796 adapter->stats.bptc += rd32(E1000_BPTC);
6798 adapter->stats.tpt += rd32(E1000_TPT);
6799 adapter->stats.colc += rd32(E1000_COLC);
6801 adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
6802 /* read internal phy specific stats */
6803 reg = rd32(E1000_CTRL_EXT);
6804 if (!(reg & E1000_CTRL_EXT_LINK_MODE_MASK)) {
6805 adapter->stats.rxerrc += rd32(E1000_RXERRC);
6807 /* this stat has invalid values on i210/i211 */
6808 if ((hw->mac.type != e1000_i210) &&
6809 (hw->mac.type != e1000_i211))
6810 adapter->stats.tncrs += rd32(E1000_TNCRS);
6813 adapter->stats.tsctc += rd32(E1000_TSCTC);
6814 adapter->stats.tsctfc += rd32(E1000_TSCTFC);
6816 adapter->stats.iac += rd32(E1000_IAC);
6817 adapter->stats.icrxoc += rd32(E1000_ICRXOC);
6818 adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
6819 adapter->stats.icrxatc += rd32(E1000_ICRXATC);
6820 adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
6821 adapter->stats.ictxatc += rd32(E1000_ICTXATC);
6822 adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
6823 adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
6824 adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
6826 /* Fill out the OS statistics structure */
6827 net_stats->multicast = adapter->stats.mprc;
6828 net_stats->collisions = adapter->stats.colc;
6832 /* RLEC on some newer hardware can be incorrect so build
6833 * our own version based on RUC and ROC
6835 net_stats->rx_errors = adapter->stats.rxerrc +
6836 adapter->stats.crcerrs + adapter->stats.algnerrc +
6837 adapter->stats.ruc + adapter->stats.roc +
6838 adapter->stats.cexterr;
6839 net_stats->rx_length_errors = adapter->stats.ruc +
6841 net_stats->rx_crc_errors = adapter->stats.crcerrs;
6842 net_stats->rx_frame_errors = adapter->stats.algnerrc;
6843 net_stats->rx_missed_errors = adapter->stats.mpc;
6846 net_stats->tx_errors = adapter->stats.ecol +
6847 adapter->stats.latecol;
6848 net_stats->tx_aborted_errors = adapter->stats.ecol;
6849 net_stats->tx_window_errors = adapter->stats.latecol;
6850 net_stats->tx_carrier_errors = adapter->stats.tncrs;
6852 /* Tx Dropped needs to be maintained elsewhere */
6854 /* Management Stats */
6855 adapter->stats.mgptc += rd32(E1000_MGTPTC);
6856 adapter->stats.mgprc += rd32(E1000_MGTPRC);
6857 adapter->stats.mgpdc += rd32(E1000_MGTPDC);
6860 reg = rd32(E1000_MANC);
6861 if (reg & E1000_MANC_EN_BMC2OS) {
6862 adapter->stats.o2bgptc += rd32(E1000_O2BGPTC);
6863 adapter->stats.o2bspc += rd32(E1000_O2BSPC);
6864 adapter->stats.b2ospc += rd32(E1000_B2OSPC);
6865 adapter->stats.b2ogprc += rd32(E1000_B2OGPRC);
6869 static void igb_perout(struct igb_adapter *adapter, int tsintr_tt)
6871 int pin = ptp_find_pin(adapter->ptp_clock, PTP_PF_PEROUT, tsintr_tt);
6872 struct e1000_hw *hw = &adapter->hw;
6873 struct timespec64 ts;
6876 if (pin < 0 || pin >= IGB_N_SDP)
6879 spin_lock(&adapter->tmreg_lock);
6881 if (hw->mac.type == e1000_82580 ||
6882 hw->mac.type == e1000_i354 ||
6883 hw->mac.type == e1000_i350) {
6884 s64 ns = timespec64_to_ns(&adapter->perout[tsintr_tt].period);
6885 u32 systiml, systimh, level_mask, level, rem;
6888 /* read systim registers in sequence */
6889 rd32(E1000_SYSTIMR);
6890 systiml = rd32(E1000_SYSTIML);
6891 systimh = rd32(E1000_SYSTIMH);
6892 systim = (((u64)(systimh & 0xFF)) << 32) | ((u64)systiml);
6893 now = timecounter_cyc2time(&adapter->tc, systim);
6896 level_mask = (tsintr_tt == 1) ? 0x80000 : 0x40000;
6897 level = (rd32(E1000_CTRL) & level_mask) ? 1 : 0;
6899 level_mask = (tsintr_tt == 1) ? 0x80 : 0x40;
6900 level = (rd32(E1000_CTRL_EXT) & level_mask) ? 1 : 0;
6903 div_u64_rem(now, ns, &rem);
6904 systim = systim + (ns - rem);
6906 /* synchronize pin level with rising/falling edges */
6907 div_u64_rem(now, ns << 1, &rem);
6909 /* first half of period */
6911 /* output is already low, skip this period */
6913 pr_notice("igb: periodic output on %s missed falling edge\n",
6914 adapter->sdp_config[pin].name);
6917 /* second half of period */
6919 /* output is already high, skip this period */
6921 pr_notice("igb: periodic output on %s missed rising edge\n",
6922 adapter->sdp_config[pin].name);
6926 /* for this chip family tv_sec is the upper part of the binary value,
6929 ts.tv_nsec = (u32)systim;
6930 ts.tv_sec = ((u32)(systim >> 32)) & 0xFF;
6932 ts = timespec64_add(adapter->perout[tsintr_tt].start,
6933 adapter->perout[tsintr_tt].period);
6936 /* u32 conversion of tv_sec is safe until y2106 */
6937 wr32((tsintr_tt == 1) ? E1000_TRGTTIML1 : E1000_TRGTTIML0, ts.tv_nsec);
6938 wr32((tsintr_tt == 1) ? E1000_TRGTTIMH1 : E1000_TRGTTIMH0, (u32)ts.tv_sec);
6939 tsauxc = rd32(E1000_TSAUXC);
6940 tsauxc |= TSAUXC_EN_TT0;
6941 wr32(E1000_TSAUXC, tsauxc);
6942 adapter->perout[tsintr_tt].start = ts;
6944 spin_unlock(&adapter->tmreg_lock);
6947 static void igb_extts(struct igb_adapter *adapter, int tsintr_tt)
6949 int pin = ptp_find_pin(adapter->ptp_clock, PTP_PF_EXTTS, tsintr_tt);
6950 int auxstmpl = (tsintr_tt == 1) ? E1000_AUXSTMPL1 : E1000_AUXSTMPL0;
6951 int auxstmph = (tsintr_tt == 1) ? E1000_AUXSTMPH1 : E1000_AUXSTMPH0;
6952 struct e1000_hw *hw = &adapter->hw;
6953 struct ptp_clock_event event;
6954 struct timespec64 ts;
6955 unsigned long flags;
6957 if (pin < 0 || pin >= IGB_N_SDP)
6960 if (hw->mac.type == e1000_82580 ||
6961 hw->mac.type == e1000_i354 ||
6962 hw->mac.type == e1000_i350) {
6963 u64 ns = rd32(auxstmpl);
6965 ns += ((u64)(rd32(auxstmph) & 0xFF)) << 32;
6966 spin_lock_irqsave(&adapter->tmreg_lock, flags);
6967 ns = timecounter_cyc2time(&adapter->tc, ns);
6968 spin_unlock_irqrestore(&adapter->tmreg_lock, flags);
6969 ts = ns_to_timespec64(ns);
6971 ts.tv_nsec = rd32(auxstmpl);
6972 ts.tv_sec = rd32(auxstmph);
6975 event.type = PTP_CLOCK_EXTTS;
6976 event.index = tsintr_tt;
6977 event.timestamp = ts.tv_sec * 1000000000ULL + ts.tv_nsec;
6978 ptp_clock_event(adapter->ptp_clock, &event);
6981 static void igb_tsync_interrupt(struct igb_adapter *adapter)
6983 struct e1000_hw *hw = &adapter->hw;
6984 u32 ack = 0, tsicr = rd32(E1000_TSICR);
6985 struct ptp_clock_event event;
6987 if (tsicr & TSINTR_SYS_WRAP) {
6988 event.type = PTP_CLOCK_PPS;
6989 if (adapter->ptp_caps.pps)
6990 ptp_clock_event(adapter->ptp_clock, &event);
6991 ack |= TSINTR_SYS_WRAP;
6994 if (tsicr & E1000_TSICR_TXTS) {
6995 /* retrieve hardware timestamp */
6996 schedule_work(&adapter->ptp_tx_work);
6997 ack |= E1000_TSICR_TXTS;
7000 if (tsicr & TSINTR_TT0) {
7001 igb_perout(adapter, 0);
7005 if (tsicr & TSINTR_TT1) {
7006 igb_perout(adapter, 1);
7010 if (tsicr & TSINTR_AUTT0) {
7011 igb_extts(adapter, 0);
7012 ack |= TSINTR_AUTT0;
7015 if (tsicr & TSINTR_AUTT1) {
7016 igb_extts(adapter, 1);
7017 ack |= TSINTR_AUTT1;
7020 /* acknowledge the interrupts */
7021 wr32(E1000_TSICR, ack);
7024 static irqreturn_t igb_msix_other(int irq, void *data)
7026 struct igb_adapter *adapter = data;
7027 struct e1000_hw *hw = &adapter->hw;
7028 u32 icr = rd32(E1000_ICR);
7029 /* reading ICR causes bit 31 of EICR to be cleared */
7031 if (icr & E1000_ICR_DRSTA)
7032 schedule_work(&adapter->reset_task);
7034 if (icr & E1000_ICR_DOUTSYNC) {
7035 /* HW is reporting DMA is out of sync */
7036 adapter->stats.doosync++;
7037 /* The DMA Out of Sync is also indication of a spoof event
7038 * in IOV mode. Check the Wrong VM Behavior register to
7039 * see if it is really a spoof event.
7041 igb_check_wvbr(adapter);
7044 /* Check for a mailbox event */
7045 if (icr & E1000_ICR_VMMB)
7046 igb_msg_task(adapter);
7048 if (icr & E1000_ICR_LSC) {
7049 hw->mac.get_link_status = 1;
7050 /* guard against interrupt when we're going down */
7051 if (!test_bit(__IGB_DOWN, &adapter->state))
7052 mod_timer(&adapter->watchdog_timer, jiffies + 1);
7055 if (icr & E1000_ICR_TS)
7056 igb_tsync_interrupt(adapter);
7058 wr32(E1000_EIMS, adapter->eims_other);
7063 static void igb_write_itr(struct igb_q_vector *q_vector)
7065 struct igb_adapter *adapter = q_vector->adapter;
7066 u32 itr_val = q_vector->itr_val & 0x7FFC;
7068 if (!q_vector->set_itr)
7074 if (adapter->hw.mac.type == e1000_82575)
7075 itr_val |= itr_val << 16;
7077 itr_val |= E1000_EITR_CNT_IGNR;
7079 writel(itr_val, q_vector->itr_register);
7080 q_vector->set_itr = 0;
7083 static irqreturn_t igb_msix_ring(int irq, void *data)
7085 struct igb_q_vector *q_vector = data;
7087 /* Write the ITR value calculated from the previous interrupt. */
7088 igb_write_itr(q_vector);
7090 napi_schedule(&q_vector->napi);
7095 #ifdef CONFIG_IGB_DCA
7096 static void igb_update_tx_dca(struct igb_adapter *adapter,
7097 struct igb_ring *tx_ring,
7100 struct e1000_hw *hw = &adapter->hw;
7101 u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
7103 if (hw->mac.type != e1000_82575)
7104 txctrl <<= E1000_DCA_TXCTRL_CPUID_SHIFT;
7106 /* We can enable relaxed ordering for reads, but not writes when
7107 * DCA is enabled. This is due to a known issue in some chipsets
7108 * which will cause the DCA tag to be cleared.
7110 txctrl |= E1000_DCA_TXCTRL_DESC_RRO_EN |
7111 E1000_DCA_TXCTRL_DATA_RRO_EN |
7112 E1000_DCA_TXCTRL_DESC_DCA_EN;
7114 wr32(E1000_DCA_TXCTRL(tx_ring->reg_idx), txctrl);
7117 static void igb_update_rx_dca(struct igb_adapter *adapter,
7118 struct igb_ring *rx_ring,
7121 struct e1000_hw *hw = &adapter->hw;
7122 u32 rxctrl = dca3_get_tag(&adapter->pdev->dev, cpu);
7124 if (hw->mac.type != e1000_82575)
7125 rxctrl <<= E1000_DCA_RXCTRL_CPUID_SHIFT;
7127 /* We can enable relaxed ordering for reads, but not writes when
7128 * DCA is enabled. This is due to a known issue in some chipsets
7129 * which will cause the DCA tag to be cleared.
7131 rxctrl |= E1000_DCA_RXCTRL_DESC_RRO_EN |
7132 E1000_DCA_RXCTRL_DESC_DCA_EN;
7134 wr32(E1000_DCA_RXCTRL(rx_ring->reg_idx), rxctrl);
7137 static void igb_update_dca(struct igb_q_vector *q_vector)
7139 struct igb_adapter *adapter = q_vector->adapter;
7140 int cpu = get_cpu();
7142 if (q_vector->cpu == cpu)
7145 if (q_vector->tx.ring)
7146 igb_update_tx_dca(adapter, q_vector->tx.ring, cpu);
7148 if (q_vector->rx.ring)
7149 igb_update_rx_dca(adapter, q_vector->rx.ring, cpu);
7151 q_vector->cpu = cpu;
7156 static void igb_setup_dca(struct igb_adapter *adapter)
7158 struct e1000_hw *hw = &adapter->hw;
7161 if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
7164 /* Always use CB2 mode, difference is masked in the CB driver. */
7165 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
7167 for (i = 0; i < adapter->num_q_vectors; i++) {
7168 adapter->q_vector[i]->cpu = -1;
7169 igb_update_dca(adapter->q_vector[i]);
7173 static int __igb_notify_dca(struct device *dev, void *data)
7175 struct net_device *netdev = dev_get_drvdata(dev);
7176 struct igb_adapter *adapter = netdev_priv(netdev);
7177 struct pci_dev *pdev = adapter->pdev;
7178 struct e1000_hw *hw = &adapter->hw;
7179 unsigned long event = *(unsigned long *)data;
7182 case DCA_PROVIDER_ADD:
7183 /* if already enabled, don't do it again */
7184 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
7186 if (dca_add_requester(dev) == 0) {
7187 adapter->flags |= IGB_FLAG_DCA_ENABLED;
7188 dev_info(&pdev->dev, "DCA enabled\n");
7189 igb_setup_dca(adapter);
7192 fallthrough; /* since DCA is disabled. */
7193 case DCA_PROVIDER_REMOVE:
7194 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
7195 /* without this a class_device is left
7196 * hanging around in the sysfs model
7198 dca_remove_requester(dev);
7199 dev_info(&pdev->dev, "DCA disabled\n");
7200 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
7201 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
7209 static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
7214 ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
7217 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7219 #endif /* CONFIG_IGB_DCA */
7221 #ifdef CONFIG_PCI_IOV
7222 static int igb_vf_configure(struct igb_adapter *adapter, int vf)
7224 unsigned char mac_addr[ETH_ALEN];
7226 eth_zero_addr(mac_addr);
7227 igb_set_vf_mac(adapter, vf, mac_addr);
7229 /* By default spoof check is enabled for all VFs */
7230 adapter->vf_data[vf].spoofchk_enabled = true;
7232 /* By default VFs are not trusted */
7233 adapter->vf_data[vf].trusted = false;
7239 static void igb_ping_all_vfs(struct igb_adapter *adapter)
7241 struct e1000_hw *hw = &adapter->hw;
7245 for (i = 0 ; i < adapter->vfs_allocated_count; i++) {
7246 ping = E1000_PF_CONTROL_MSG;
7247 if (adapter->vf_data[i].flags & IGB_VF_FLAG_CTS)
7248 ping |= E1000_VT_MSGTYPE_CTS;
7249 igb_write_mbx(hw, &ping, 1, i);
7253 static int igb_set_vf_promisc(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
7255 struct e1000_hw *hw = &adapter->hw;
7256 u32 vmolr = rd32(E1000_VMOLR(vf));
7257 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7259 vf_data->flags &= ~(IGB_VF_FLAG_UNI_PROMISC |
7260 IGB_VF_FLAG_MULTI_PROMISC);
7261 vmolr &= ~(E1000_VMOLR_ROPE | E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
7263 if (*msgbuf & E1000_VF_SET_PROMISC_MULTICAST) {
7264 vmolr |= E1000_VMOLR_MPME;
7265 vf_data->flags |= IGB_VF_FLAG_MULTI_PROMISC;
7266 *msgbuf &= ~E1000_VF_SET_PROMISC_MULTICAST;
7268 /* if we have hashes and we are clearing a multicast promisc
7269 * flag we need to write the hashes to the MTA as this step
7270 * was previously skipped
7272 if (vf_data->num_vf_mc_hashes > 30) {
7273 vmolr |= E1000_VMOLR_MPME;
7274 } else if (vf_data->num_vf_mc_hashes) {
7277 vmolr |= E1000_VMOLR_ROMPE;
7278 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
7279 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
7283 wr32(E1000_VMOLR(vf), vmolr);
7285 /* there are flags left unprocessed, likely not supported */
7286 if (*msgbuf & E1000_VT_MSGINFO_MASK)
7292 static int igb_set_vf_multicasts(struct igb_adapter *adapter,
7293 u32 *msgbuf, u32 vf)
7295 int n = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
7296 u16 *hash_list = (u16 *)&msgbuf[1];
7297 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7300 /* salt away the number of multicast addresses assigned
7301 * to this VF for later use to restore when the PF multi cast
7304 vf_data->num_vf_mc_hashes = n;
7306 /* only up to 30 hash values supported */
7310 /* store the hashes for later use */
7311 for (i = 0; i < n; i++)
7312 vf_data->vf_mc_hashes[i] = hash_list[i];
7314 /* Flush and reset the mta with the new values */
7315 igb_set_rx_mode(adapter->netdev);
7320 static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
7322 struct e1000_hw *hw = &adapter->hw;
7323 struct vf_data_storage *vf_data;
7326 for (i = 0; i < adapter->vfs_allocated_count; i++) {
7327 u32 vmolr = rd32(E1000_VMOLR(i));
7329 vmolr &= ~(E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
7331 vf_data = &adapter->vf_data[i];
7333 if ((vf_data->num_vf_mc_hashes > 30) ||
7334 (vf_data->flags & IGB_VF_FLAG_MULTI_PROMISC)) {
7335 vmolr |= E1000_VMOLR_MPME;
7336 } else if (vf_data->num_vf_mc_hashes) {
7337 vmolr |= E1000_VMOLR_ROMPE;
7338 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
7339 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
7341 wr32(E1000_VMOLR(i), vmolr);
7345 static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
7347 struct e1000_hw *hw = &adapter->hw;
7348 u32 pool_mask, vlvf_mask, i;
7350 /* create mask for VF and other pools */
7351 pool_mask = E1000_VLVF_POOLSEL_MASK;
7352 vlvf_mask = BIT(E1000_VLVF_POOLSEL_SHIFT + vf);
7354 /* drop PF from pool bits */
7355 pool_mask &= ~BIT(E1000_VLVF_POOLSEL_SHIFT +
7356 adapter->vfs_allocated_count);
7358 /* Find the vlan filter for this id */
7359 for (i = E1000_VLVF_ARRAY_SIZE; i--;) {
7360 u32 vlvf = rd32(E1000_VLVF(i));
7361 u32 vfta_mask, vid, vfta;
7363 /* remove the vf from the pool */
7364 if (!(vlvf & vlvf_mask))
7367 /* clear out bit from VLVF */
7370 /* if other pools are present, just remove ourselves */
7371 if (vlvf & pool_mask)
7374 /* if PF is present, leave VFTA */
7375 if (vlvf & E1000_VLVF_POOLSEL_MASK)
7378 vid = vlvf & E1000_VLVF_VLANID_MASK;
7379 vfta_mask = BIT(vid % 32);
7381 /* clear bit from VFTA */
7382 vfta = adapter->shadow_vfta[vid / 32];
7383 if (vfta & vfta_mask)
7384 hw->mac.ops.write_vfta(hw, vid / 32, vfta ^ vfta_mask);
7386 /* clear pool selection enable */
7387 if (adapter->flags & IGB_FLAG_VLAN_PROMISC)
7388 vlvf &= E1000_VLVF_POOLSEL_MASK;
7392 /* clear pool bits */
7393 wr32(E1000_VLVF(i), vlvf);
7397 static int igb_find_vlvf_entry(struct e1000_hw *hw, u32 vlan)
7402 /* short cut the special case */
7406 /* Search for the VLAN id in the VLVF entries */
7407 for (idx = E1000_VLVF_ARRAY_SIZE; --idx;) {
7408 vlvf = rd32(E1000_VLVF(idx));
7409 if ((vlvf & VLAN_VID_MASK) == vlan)
7416 static void igb_update_pf_vlvf(struct igb_adapter *adapter, u32 vid)
7418 struct e1000_hw *hw = &adapter->hw;
7422 idx = igb_find_vlvf_entry(hw, vid);
7426 /* See if any other pools are set for this VLAN filter
7427 * entry other than the PF.
7429 pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT;
7430 bits = ~BIT(pf_id) & E1000_VLVF_POOLSEL_MASK;
7431 bits &= rd32(E1000_VLVF(idx));
7433 /* Disable the filter so this falls into the default pool. */
7435 if (adapter->flags & IGB_FLAG_VLAN_PROMISC)
7436 wr32(E1000_VLVF(idx), BIT(pf_id));
7438 wr32(E1000_VLVF(idx), 0);
7442 static s32 igb_set_vf_vlan(struct igb_adapter *adapter, u32 vid,
7445 int pf_id = adapter->vfs_allocated_count;
7446 struct e1000_hw *hw = &adapter->hw;
7449 /* If VLAN overlaps with one the PF is currently monitoring make
7450 * sure that we are able to allocate a VLVF entry. This may be
7451 * redundant but it guarantees PF will maintain visibility to
7454 if (add && test_bit(vid, adapter->active_vlans)) {
7455 err = igb_vfta_set(hw, vid, pf_id, true, false);
7460 err = igb_vfta_set(hw, vid, vf, add, false);
7465 /* If we failed to add the VF VLAN or we are removing the VF VLAN
7466 * we may need to drop the PF pool bit in order to allow us to free
7467 * up the VLVF resources.
7469 if (test_bit(vid, adapter->active_vlans) ||
7470 (adapter->flags & IGB_FLAG_VLAN_PROMISC))
7471 igb_update_pf_vlvf(adapter, vid);
7476 static void igb_set_vmvir(struct igb_adapter *adapter, u32 vid, u32 vf)
7478 struct e1000_hw *hw = &adapter->hw;
7481 wr32(E1000_VMVIR(vf), (vid | E1000_VMVIR_VLANA_DEFAULT));
7483 wr32(E1000_VMVIR(vf), 0);
7486 static int igb_enable_port_vlan(struct igb_adapter *adapter, int vf,
7491 err = igb_set_vf_vlan(adapter, vlan, true, vf);
7495 igb_set_vmvir(adapter, vlan | (qos << VLAN_PRIO_SHIFT), vf);
7496 igb_set_vmolr(adapter, vf, !vlan);
7498 /* revoke access to previous VLAN */
7499 if (vlan != adapter->vf_data[vf].pf_vlan)
7500 igb_set_vf_vlan(adapter, adapter->vf_data[vf].pf_vlan,
7503 adapter->vf_data[vf].pf_vlan = vlan;
7504 adapter->vf_data[vf].pf_qos = qos;
7505 igb_set_vf_vlan_strip(adapter, vf, true);
7506 dev_info(&adapter->pdev->dev,
7507 "Setting VLAN %d, QOS 0x%x on VF %d\n", vlan, qos, vf);
7508 if (test_bit(__IGB_DOWN, &adapter->state)) {
7509 dev_warn(&adapter->pdev->dev,
7510 "The VF VLAN has been set, but the PF device is not up.\n");
7511 dev_warn(&adapter->pdev->dev,
7512 "Bring the PF device up before attempting to use the VF device.\n");
7518 static int igb_disable_port_vlan(struct igb_adapter *adapter, int vf)
7520 /* Restore tagless access via VLAN 0 */
7521 igb_set_vf_vlan(adapter, 0, true, vf);
7523 igb_set_vmvir(adapter, 0, vf);
7524 igb_set_vmolr(adapter, vf, true);
7526 /* Remove any PF assigned VLAN */
7527 if (adapter->vf_data[vf].pf_vlan)
7528 igb_set_vf_vlan(adapter, adapter->vf_data[vf].pf_vlan,
7531 adapter->vf_data[vf].pf_vlan = 0;
7532 adapter->vf_data[vf].pf_qos = 0;
7533 igb_set_vf_vlan_strip(adapter, vf, false);
7538 static int igb_ndo_set_vf_vlan(struct net_device *netdev, int vf,
7539 u16 vlan, u8 qos, __be16 vlan_proto)
7541 struct igb_adapter *adapter = netdev_priv(netdev);
7543 if ((vf >= adapter->vfs_allocated_count) || (vlan > 4095) || (qos > 7))
7546 if (vlan_proto != htons(ETH_P_8021Q))
7547 return -EPROTONOSUPPORT;
7549 return (vlan || qos) ? igb_enable_port_vlan(adapter, vf, vlan, qos) :
7550 igb_disable_port_vlan(adapter, vf);
7553 static int igb_set_vf_vlan_msg(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
7555 int add = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
7556 int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
7559 if (adapter->vf_data[vf].pf_vlan)
7562 /* VLAN 0 is a special case, don't allow it to be removed */
7566 ret = igb_set_vf_vlan(adapter, vid, !!add, vf);
7568 igb_set_vf_vlan_strip(adapter, vf, !!vid);
7572 static inline void igb_vf_reset(struct igb_adapter *adapter, u32 vf)
7574 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7576 /* clear flags - except flag that indicates PF has set the MAC */
7577 vf_data->flags &= IGB_VF_FLAG_PF_SET_MAC;
7578 vf_data->last_nack = jiffies;
7580 /* reset vlans for device */
7581 igb_clear_vf_vfta(adapter, vf);
7582 igb_set_vf_vlan(adapter, vf_data->pf_vlan, true, vf);
7583 igb_set_vmvir(adapter, vf_data->pf_vlan |
7584 (vf_data->pf_qos << VLAN_PRIO_SHIFT), vf);
7585 igb_set_vmolr(adapter, vf, !vf_data->pf_vlan);
7586 igb_set_vf_vlan_strip(adapter, vf, !!(vf_data->pf_vlan));
7588 /* reset multicast table array for vf */
7589 adapter->vf_data[vf].num_vf_mc_hashes = 0;
7591 /* Flush and reset the mta with the new values */
7592 igb_set_rx_mode(adapter->netdev);
7595 static void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
7597 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
7599 /* clear mac address as we were hotplug removed/added */
7600 if (!(adapter->vf_data[vf].flags & IGB_VF_FLAG_PF_SET_MAC))
7601 eth_zero_addr(vf_mac);
7603 /* process remaining reset events */
7604 igb_vf_reset(adapter, vf);
7607 static void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
7609 struct e1000_hw *hw = &adapter->hw;
7610 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
7611 u32 reg, msgbuf[3] = {};
7612 u8 *addr = (u8 *)(&msgbuf[1]);
7614 /* process all the same items cleared in a function level reset */
7615 igb_vf_reset(adapter, vf);
7617 /* set vf mac address */
7618 igb_set_vf_mac(adapter, vf, vf_mac);
7620 /* enable transmit and receive for vf */
7621 reg = rd32(E1000_VFTE);
7622 wr32(E1000_VFTE, reg | BIT(vf));
7623 reg = rd32(E1000_VFRE);
7624 wr32(E1000_VFRE, reg | BIT(vf));
7626 adapter->vf_data[vf].flags |= IGB_VF_FLAG_CTS;
7628 /* reply to reset with ack and vf mac address */
7629 if (!is_zero_ether_addr(vf_mac)) {
7630 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
7631 memcpy(addr, vf_mac, ETH_ALEN);
7633 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_NACK;
7635 igb_write_mbx(hw, msgbuf, 3, vf);
7638 static void igb_flush_mac_table(struct igb_adapter *adapter)
7640 struct e1000_hw *hw = &adapter->hw;
7643 for (i = 0; i < hw->mac.rar_entry_count; i++) {
7644 adapter->mac_table[i].state &= ~IGB_MAC_STATE_IN_USE;
7645 eth_zero_addr(adapter->mac_table[i].addr);
7646 adapter->mac_table[i].queue = 0;
7647 igb_rar_set_index(adapter, i);
7651 static int igb_available_rars(struct igb_adapter *adapter, u8 queue)
7653 struct e1000_hw *hw = &adapter->hw;
7654 /* do not count rar entries reserved for VFs MAC addresses */
7655 int rar_entries = hw->mac.rar_entry_count -
7656 adapter->vfs_allocated_count;
7659 for (i = 0; i < rar_entries; i++) {
7660 /* do not count default entries */
7661 if (adapter->mac_table[i].state & IGB_MAC_STATE_DEFAULT)
7664 /* do not count "in use" entries for different queues */
7665 if ((adapter->mac_table[i].state & IGB_MAC_STATE_IN_USE) &&
7666 (adapter->mac_table[i].queue != queue))
7675 /* Set default MAC address for the PF in the first RAR entry */
7676 static void igb_set_default_mac_filter(struct igb_adapter *adapter)
7678 struct igb_mac_addr *mac_table = &adapter->mac_table[0];
7680 ether_addr_copy(mac_table->addr, adapter->hw.mac.addr);
7681 mac_table->queue = adapter->vfs_allocated_count;
7682 mac_table->state = IGB_MAC_STATE_DEFAULT | IGB_MAC_STATE_IN_USE;
7684 igb_rar_set_index(adapter, 0);
7687 /* If the filter to be added and an already existing filter express
7688 * the same address and address type, it should be possible to only
7689 * override the other configurations, for example the queue to steer
7692 static bool igb_mac_entry_can_be_used(const struct igb_mac_addr *entry,
7693 const u8 *addr, const u8 flags)
7695 if (!(entry->state & IGB_MAC_STATE_IN_USE))
7698 if ((entry->state & IGB_MAC_STATE_SRC_ADDR) !=
7699 (flags & IGB_MAC_STATE_SRC_ADDR))
7702 if (!ether_addr_equal(addr, entry->addr))
7708 /* Add a MAC filter for 'addr' directing matching traffic to 'queue',
7709 * 'flags' is used to indicate what kind of match is made, match is by
7710 * default for the destination address, if matching by source address
7711 * is desired the flag IGB_MAC_STATE_SRC_ADDR can be used.
7713 static int igb_add_mac_filter_flags(struct igb_adapter *adapter,
7714 const u8 *addr, const u8 queue,
7717 struct e1000_hw *hw = &adapter->hw;
7718 int rar_entries = hw->mac.rar_entry_count -
7719 adapter->vfs_allocated_count;
7722 if (is_zero_ether_addr(addr))
7725 /* Search for the first empty entry in the MAC table.
7726 * Do not touch entries at the end of the table reserved for the VF MAC
7729 for (i = 0; i < rar_entries; i++) {
7730 if (!igb_mac_entry_can_be_used(&adapter->mac_table[i],
7734 ether_addr_copy(adapter->mac_table[i].addr, addr);
7735 adapter->mac_table[i].queue = queue;
7736 adapter->mac_table[i].state |= IGB_MAC_STATE_IN_USE | flags;
7738 igb_rar_set_index(adapter, i);
7745 static int igb_add_mac_filter(struct igb_adapter *adapter, const u8 *addr,
7748 return igb_add_mac_filter_flags(adapter, addr, queue, 0);
7751 /* Remove a MAC filter for 'addr' directing matching traffic to
7752 * 'queue', 'flags' is used to indicate what kind of match need to be
7753 * removed, match is by default for the destination address, if
7754 * matching by source address is to be removed the flag
7755 * IGB_MAC_STATE_SRC_ADDR can be used.
7757 static int igb_del_mac_filter_flags(struct igb_adapter *adapter,
7758 const u8 *addr, const u8 queue,
7761 struct e1000_hw *hw = &adapter->hw;
7762 int rar_entries = hw->mac.rar_entry_count -
7763 adapter->vfs_allocated_count;
7766 if (is_zero_ether_addr(addr))
7769 /* Search for matching entry in the MAC table based on given address
7770 * and queue. Do not touch entries at the end of the table reserved
7771 * for the VF MAC addresses.
7773 for (i = 0; i < rar_entries; i++) {
7774 if (!(adapter->mac_table[i].state & IGB_MAC_STATE_IN_USE))
7776 if ((adapter->mac_table[i].state & flags) != flags)
7778 if (adapter->mac_table[i].queue != queue)
7780 if (!ether_addr_equal(adapter->mac_table[i].addr, addr))
7783 /* When a filter for the default address is "deleted",
7784 * we return it to its initial configuration
7786 if (adapter->mac_table[i].state & IGB_MAC_STATE_DEFAULT) {
7787 adapter->mac_table[i].state =
7788 IGB_MAC_STATE_DEFAULT | IGB_MAC_STATE_IN_USE;
7789 adapter->mac_table[i].queue =
7790 adapter->vfs_allocated_count;
7792 adapter->mac_table[i].state = 0;
7793 adapter->mac_table[i].queue = 0;
7794 eth_zero_addr(adapter->mac_table[i].addr);
7797 igb_rar_set_index(adapter, i);
7804 static int igb_del_mac_filter(struct igb_adapter *adapter, const u8 *addr,
7807 return igb_del_mac_filter_flags(adapter, addr, queue, 0);
7810 int igb_add_mac_steering_filter(struct igb_adapter *adapter,
7811 const u8 *addr, u8 queue, u8 flags)
7813 struct e1000_hw *hw = &adapter->hw;
7815 /* In theory, this should be supported on 82575 as well, but
7816 * that part wasn't easily accessible during development.
7818 if (hw->mac.type != e1000_i210)
7821 return igb_add_mac_filter_flags(adapter, addr, queue,
7822 IGB_MAC_STATE_QUEUE_STEERING | flags);
7825 int igb_del_mac_steering_filter(struct igb_adapter *adapter,
7826 const u8 *addr, u8 queue, u8 flags)
7828 return igb_del_mac_filter_flags(adapter, addr, queue,
7829 IGB_MAC_STATE_QUEUE_STEERING | flags);
7832 static int igb_uc_sync(struct net_device *netdev, const unsigned char *addr)
7834 struct igb_adapter *adapter = netdev_priv(netdev);
7837 ret = igb_add_mac_filter(adapter, addr, adapter->vfs_allocated_count);
7839 return min_t(int, ret, 0);
7842 static int igb_uc_unsync(struct net_device *netdev, const unsigned char *addr)
7844 struct igb_adapter *adapter = netdev_priv(netdev);
7846 igb_del_mac_filter(adapter, addr, adapter->vfs_allocated_count);
7851 static int igb_set_vf_mac_filter(struct igb_adapter *adapter, const int vf,
7852 const u32 info, const u8 *addr)
7854 struct pci_dev *pdev = adapter->pdev;
7855 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7856 struct list_head *pos;
7857 struct vf_mac_filter *entry = NULL;
7860 if ((vf_data->flags & IGB_VF_FLAG_PF_SET_MAC) &&
7861 !vf_data->trusted) {
7862 dev_warn(&pdev->dev,
7863 "VF %d requested MAC filter but is administratively denied\n",
7867 if (!is_valid_ether_addr(addr)) {
7868 dev_warn(&pdev->dev,
7869 "VF %d attempted to set invalid MAC filter\n",
7875 case E1000_VF_MAC_FILTER_CLR:
7876 /* remove all unicast MAC filters related to the current VF */
7877 list_for_each(pos, &adapter->vf_macs.l) {
7878 entry = list_entry(pos, struct vf_mac_filter, l);
7879 if (entry->vf == vf) {
7882 igb_del_mac_filter(adapter, entry->vf_mac, vf);
7886 case E1000_VF_MAC_FILTER_ADD:
7887 /* try to find empty slot in the list */
7888 list_for_each(pos, &adapter->vf_macs.l) {
7889 entry = list_entry(pos, struct vf_mac_filter, l);
7894 if (entry && entry->free) {
7895 entry->free = false;
7897 ether_addr_copy(entry->vf_mac, addr);
7899 ret = igb_add_mac_filter(adapter, addr, vf);
7900 ret = min_t(int, ret, 0);
7906 dev_warn(&pdev->dev,
7907 "VF %d has requested MAC filter but there is no space for it\n",
7918 static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf)
7920 struct pci_dev *pdev = adapter->pdev;
7921 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7922 u32 info = msg[0] & E1000_VT_MSGINFO_MASK;
7924 /* The VF MAC Address is stored in a packed array of bytes
7925 * starting at the second 32 bit word of the msg array
7927 unsigned char *addr = (unsigned char *)&msg[1];
7931 if ((vf_data->flags & IGB_VF_FLAG_PF_SET_MAC) &&
7932 !vf_data->trusted) {
7933 dev_warn(&pdev->dev,
7934 "VF %d attempted to override administratively set MAC address\nReload the VF driver to resume operations\n",
7939 if (!is_valid_ether_addr(addr)) {
7940 dev_warn(&pdev->dev,
7941 "VF %d attempted to set invalid MAC\n",
7946 ret = igb_set_vf_mac(adapter, vf, addr);
7948 ret = igb_set_vf_mac_filter(adapter, vf, info, addr);
7954 static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf)
7956 struct e1000_hw *hw = &adapter->hw;
7957 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7958 u32 msg = E1000_VT_MSGTYPE_NACK;
7960 /* if device isn't clear to send it shouldn't be reading either */
7961 if (!(vf_data->flags & IGB_VF_FLAG_CTS) &&
7962 time_after(jiffies, vf_data->last_nack + (2 * HZ))) {
7963 igb_write_mbx(hw, &msg, 1, vf);
7964 vf_data->last_nack = jiffies;
7968 static void igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
7970 struct pci_dev *pdev = adapter->pdev;
7971 u32 msgbuf[E1000_VFMAILBOX_SIZE];
7972 struct e1000_hw *hw = &adapter->hw;
7973 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7976 retval = igb_read_mbx(hw, msgbuf, E1000_VFMAILBOX_SIZE, vf, false);
7979 /* if receive failed revoke VF CTS stats and restart init */
7980 dev_err(&pdev->dev, "Error receiving message from VF\n");
7981 vf_data->flags &= ~IGB_VF_FLAG_CTS;
7982 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
7987 /* this is a message we already processed, do nothing */
7988 if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
7991 /* until the vf completes a reset it should not be
7992 * allowed to start any configuration.
7994 if (msgbuf[0] == E1000_VF_RESET) {
7995 /* unlocks mailbox */
7996 igb_vf_reset_msg(adapter, vf);
8000 if (!(vf_data->flags & IGB_VF_FLAG_CTS)) {
8001 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
8007 switch ((msgbuf[0] & 0xFFFF)) {
8008 case E1000_VF_SET_MAC_ADDR:
8009 retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
8011 case E1000_VF_SET_PROMISC:
8012 retval = igb_set_vf_promisc(adapter, msgbuf, vf);
8014 case E1000_VF_SET_MULTICAST:
8015 retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
8017 case E1000_VF_SET_LPE:
8018 retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf);
8020 case E1000_VF_SET_VLAN:
8022 if (vf_data->pf_vlan)
8023 dev_warn(&pdev->dev,
8024 "VF %d attempted to override administratively set VLAN tag\nReload the VF driver to resume operations\n",
8027 retval = igb_set_vf_vlan_msg(adapter, msgbuf, vf);
8030 dev_err(&pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
8035 msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
8037 /* notify the VF of the results of what it sent us */
8039 msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
8041 msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
8043 /* unlocks mailbox */
8044 igb_write_mbx(hw, msgbuf, 1, vf);
8048 igb_unlock_mbx(hw, vf);
8051 static void igb_msg_task(struct igb_adapter *adapter)
8053 struct e1000_hw *hw = &adapter->hw;
8054 unsigned long flags;
8057 spin_lock_irqsave(&adapter->vfs_lock, flags);
8058 for (vf = 0; vf < adapter->vfs_allocated_count; vf++) {
8059 /* process any reset requests */
8060 if (!igb_check_for_rst(hw, vf))
8061 igb_vf_reset_event(adapter, vf);
8063 /* process any messages pending */
8064 if (!igb_check_for_msg(hw, vf))
8065 igb_rcv_msg_from_vf(adapter, vf);
8067 /* process any acks */
8068 if (!igb_check_for_ack(hw, vf))
8069 igb_rcv_ack_from_vf(adapter, vf);
8071 spin_unlock_irqrestore(&adapter->vfs_lock, flags);
8075 * igb_set_uta - Set unicast filter table address
8076 * @adapter: board private structure
8077 * @set: boolean indicating if we are setting or clearing bits
8079 * The unicast table address is a register array of 32-bit registers.
8080 * The table is meant to be used in a way similar to how the MTA is used
8081 * however due to certain limitations in the hardware it is necessary to
8082 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
8083 * enable bit to allow vlan tag stripping when promiscuous mode is enabled
8085 static void igb_set_uta(struct igb_adapter *adapter, bool set)
8087 struct e1000_hw *hw = &adapter->hw;
8088 u32 uta = set ? ~0 : 0;
8091 /* we only need to do this if VMDq is enabled */
8092 if (!adapter->vfs_allocated_count)
8095 for (i = hw->mac.uta_reg_count; i--;)
8096 array_wr32(E1000_UTA, i, uta);
8100 * igb_intr_msi - Interrupt Handler
8101 * @irq: interrupt number
8102 * @data: pointer to a network interface device structure
8104 static irqreturn_t igb_intr_msi(int irq, void *data)
8106 struct igb_adapter *adapter = data;
8107 struct igb_q_vector *q_vector = adapter->q_vector[0];
8108 struct e1000_hw *hw = &adapter->hw;
8109 /* read ICR disables interrupts using IAM */
8110 u32 icr = rd32(E1000_ICR);
8112 igb_write_itr(q_vector);
8114 if (icr & E1000_ICR_DRSTA)
8115 schedule_work(&adapter->reset_task);
8117 if (icr & E1000_ICR_DOUTSYNC) {
8118 /* HW is reporting DMA is out of sync */
8119 adapter->stats.doosync++;
8122 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
8123 hw->mac.get_link_status = 1;
8124 if (!test_bit(__IGB_DOWN, &adapter->state))
8125 mod_timer(&adapter->watchdog_timer, jiffies + 1);
8128 if (icr & E1000_ICR_TS)
8129 igb_tsync_interrupt(adapter);
8131 napi_schedule(&q_vector->napi);
8137 * igb_intr - Legacy Interrupt Handler
8138 * @irq: interrupt number
8139 * @data: pointer to a network interface device structure
8141 static irqreturn_t igb_intr(int irq, void *data)
8143 struct igb_adapter *adapter = data;
8144 struct igb_q_vector *q_vector = adapter->q_vector[0];
8145 struct e1000_hw *hw = &adapter->hw;
8146 /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked. No
8147 * need for the IMC write
8149 u32 icr = rd32(E1000_ICR);
8151 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
8152 * not set, then the adapter didn't send an interrupt
8154 if (!(icr & E1000_ICR_INT_ASSERTED))
8157 igb_write_itr(q_vector);
8159 if (icr & E1000_ICR_DRSTA)
8160 schedule_work(&adapter->reset_task);
8162 if (icr & E1000_ICR_DOUTSYNC) {
8163 /* HW is reporting DMA is out of sync */
8164 adapter->stats.doosync++;
8167 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
8168 hw->mac.get_link_status = 1;
8169 /* guard against interrupt when we're going down */
8170 if (!test_bit(__IGB_DOWN, &adapter->state))
8171 mod_timer(&adapter->watchdog_timer, jiffies + 1);
8174 if (icr & E1000_ICR_TS)
8175 igb_tsync_interrupt(adapter);
8177 napi_schedule(&q_vector->napi);
8182 static void igb_ring_irq_enable(struct igb_q_vector *q_vector)
8184 struct igb_adapter *adapter = q_vector->adapter;
8185 struct e1000_hw *hw = &adapter->hw;
8187 if ((q_vector->rx.ring && (adapter->rx_itr_setting & 3)) ||
8188 (!q_vector->rx.ring && (adapter->tx_itr_setting & 3))) {
8189 if ((adapter->num_q_vectors == 1) && !adapter->vf_data)
8190 igb_set_itr(q_vector);
8192 igb_update_ring_itr(q_vector);
8195 if (!test_bit(__IGB_DOWN, &adapter->state)) {
8196 if (adapter->flags & IGB_FLAG_HAS_MSIX)
8197 wr32(E1000_EIMS, q_vector->eims_value);
8199 igb_irq_enable(adapter);
8204 * igb_poll - NAPI Rx polling callback
8205 * @napi: napi polling structure
8206 * @budget: count of how many packets we should handle
8208 static int igb_poll(struct napi_struct *napi, int budget)
8210 struct igb_q_vector *q_vector = container_of(napi,
8211 struct igb_q_vector,
8213 bool clean_complete = true;
8216 #ifdef CONFIG_IGB_DCA
8217 if (q_vector->adapter->flags & IGB_FLAG_DCA_ENABLED)
8218 igb_update_dca(q_vector);
8220 if (q_vector->tx.ring)
8221 clean_complete = igb_clean_tx_irq(q_vector, budget);
8223 if (q_vector->rx.ring) {
8224 int cleaned = igb_clean_rx_irq(q_vector, budget);
8226 work_done += cleaned;
8227 if (cleaned >= budget)
8228 clean_complete = false;
8231 /* If all work not completed, return budget and keep polling */
8232 if (!clean_complete)
8235 /* Exit the polling mode, but don't re-enable interrupts if stack might
8236 * poll us due to busy-polling
8238 if (likely(napi_complete_done(napi, work_done)))
8239 igb_ring_irq_enable(q_vector);
8245 * igb_clean_tx_irq - Reclaim resources after transmit completes
8246 * @q_vector: pointer to q_vector containing needed info
8247 * @napi_budget: Used to determine if we are in netpoll
8249 * returns true if ring is completely cleaned
8251 static bool igb_clean_tx_irq(struct igb_q_vector *q_vector, int napi_budget)
8253 struct igb_adapter *adapter = q_vector->adapter;
8254 struct igb_ring *tx_ring = q_vector->tx.ring;
8255 struct igb_tx_buffer *tx_buffer;
8256 union e1000_adv_tx_desc *tx_desc;
8257 unsigned int total_bytes = 0, total_packets = 0;
8258 unsigned int budget = q_vector->tx.work_limit;
8259 unsigned int i = tx_ring->next_to_clean;
8261 if (test_bit(__IGB_DOWN, &adapter->state))
8264 tx_buffer = &tx_ring->tx_buffer_info[i];
8265 tx_desc = IGB_TX_DESC(tx_ring, i);
8266 i -= tx_ring->count;
8269 union e1000_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
8271 /* if next_to_watch is not set then there is no work pending */
8275 /* prevent any other reads prior to eop_desc */
8278 /* if DD is not set pending work has not been completed */
8279 if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)))
8282 /* clear next_to_watch to prevent false hangs */
8283 tx_buffer->next_to_watch = NULL;
8285 /* update the statistics for this packet */
8286 total_bytes += tx_buffer->bytecount;
8287 total_packets += tx_buffer->gso_segs;
8290 if (tx_buffer->type == IGB_TYPE_SKB)
8291 napi_consume_skb(tx_buffer->skb, napi_budget);
8293 xdp_return_frame(tx_buffer->xdpf);
8295 /* unmap skb header data */
8296 dma_unmap_single(tx_ring->dev,
8297 dma_unmap_addr(tx_buffer, dma),
8298 dma_unmap_len(tx_buffer, len),
8301 /* clear tx_buffer data */
8302 dma_unmap_len_set(tx_buffer, len, 0);
8304 /* clear last DMA location and unmap remaining buffers */
8305 while (tx_desc != eop_desc) {
8310 i -= tx_ring->count;
8311 tx_buffer = tx_ring->tx_buffer_info;
8312 tx_desc = IGB_TX_DESC(tx_ring, 0);
8315 /* unmap any remaining paged data */
8316 if (dma_unmap_len(tx_buffer, len)) {
8317 dma_unmap_page(tx_ring->dev,
8318 dma_unmap_addr(tx_buffer, dma),
8319 dma_unmap_len(tx_buffer, len),
8321 dma_unmap_len_set(tx_buffer, len, 0);
8325 /* move us one more past the eop_desc for start of next pkt */
8330 i -= tx_ring->count;
8331 tx_buffer = tx_ring->tx_buffer_info;
8332 tx_desc = IGB_TX_DESC(tx_ring, 0);
8335 /* issue prefetch for next Tx descriptor */
8338 /* update budget accounting */
8340 } while (likely(budget));
8342 netdev_tx_completed_queue(txring_txq(tx_ring),
8343 total_packets, total_bytes);
8344 i += tx_ring->count;
8345 tx_ring->next_to_clean = i;
8346 u64_stats_update_begin(&tx_ring->tx_syncp);
8347 tx_ring->tx_stats.bytes += total_bytes;
8348 tx_ring->tx_stats.packets += total_packets;
8349 u64_stats_update_end(&tx_ring->tx_syncp);
8350 q_vector->tx.total_bytes += total_bytes;
8351 q_vector->tx.total_packets += total_packets;
8353 if (test_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags)) {
8354 struct e1000_hw *hw = &adapter->hw;
8356 /* Detect a transmit hang in hardware, this serializes the
8357 * check with the clearing of time_stamp and movement of i
8359 clear_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
8360 if (tx_buffer->next_to_watch &&
8361 time_after(jiffies, tx_buffer->time_stamp +
8362 (adapter->tx_timeout_factor * HZ)) &&
8363 !(rd32(E1000_STATUS) & E1000_STATUS_TXOFF)) {
8365 /* detected Tx unit hang */
8366 dev_err(tx_ring->dev,
8367 "Detected Tx Unit Hang\n"
8371 " next_to_use <%x>\n"
8372 " next_to_clean <%x>\n"
8373 "buffer_info[next_to_clean]\n"
8374 " time_stamp <%lx>\n"
8375 " next_to_watch <%p>\n"
8377 " desc.status <%x>\n",
8378 tx_ring->queue_index,
8379 rd32(E1000_TDH(tx_ring->reg_idx)),
8380 readl(tx_ring->tail),
8381 tx_ring->next_to_use,
8382 tx_ring->next_to_clean,
8383 tx_buffer->time_stamp,
8384 tx_buffer->next_to_watch,
8386 tx_buffer->next_to_watch->wb.status);
8387 netif_stop_subqueue(tx_ring->netdev,
8388 tx_ring->queue_index);
8390 /* we are about to reset, no point in enabling stuff */
8395 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
8396 if (unlikely(total_packets &&
8397 netif_carrier_ok(tx_ring->netdev) &&
8398 igb_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD)) {
8399 /* Make sure that anybody stopping the queue after this
8400 * sees the new next_to_clean.
8403 if (__netif_subqueue_stopped(tx_ring->netdev,
8404 tx_ring->queue_index) &&
8405 !(test_bit(__IGB_DOWN, &adapter->state))) {
8406 netif_wake_subqueue(tx_ring->netdev,
8407 tx_ring->queue_index);
8409 u64_stats_update_begin(&tx_ring->tx_syncp);
8410 tx_ring->tx_stats.restart_queue++;
8411 u64_stats_update_end(&tx_ring->tx_syncp);
8419 * igb_reuse_rx_page - page flip buffer and store it back on the ring
8420 * @rx_ring: rx descriptor ring to store buffers on
8421 * @old_buff: donor buffer to have page reused
8423 * Synchronizes page for reuse by the adapter
8425 static void igb_reuse_rx_page(struct igb_ring *rx_ring,
8426 struct igb_rx_buffer *old_buff)
8428 struct igb_rx_buffer *new_buff;
8429 u16 nta = rx_ring->next_to_alloc;
8431 new_buff = &rx_ring->rx_buffer_info[nta];
8433 /* update, and store next to alloc */
8435 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
8437 /* Transfer page from old buffer to new buffer.
8438 * Move each member individually to avoid possible store
8439 * forwarding stalls.
8441 new_buff->dma = old_buff->dma;
8442 new_buff->page = old_buff->page;
8443 new_buff->page_offset = old_buff->page_offset;
8444 new_buff->pagecnt_bias = old_buff->pagecnt_bias;
8447 static bool igb_can_reuse_rx_page(struct igb_rx_buffer *rx_buffer,
8450 unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
8451 struct page *page = rx_buffer->page;
8453 /* avoid re-using remote and pfmemalloc pages */
8454 if (!dev_page_is_reusable(page))
8457 #if (PAGE_SIZE < 8192)
8458 /* if we are only owner of page we can reuse it */
8459 if (unlikely((rx_buf_pgcnt - pagecnt_bias) > 1))
8462 #define IGB_LAST_OFFSET \
8463 (SKB_WITH_OVERHEAD(PAGE_SIZE) - IGB_RXBUFFER_2048)
8465 if (rx_buffer->page_offset > IGB_LAST_OFFSET)
8469 /* If we have drained the page fragment pool we need to update
8470 * the pagecnt_bias and page count so that we fully restock the
8471 * number of references the driver holds.
8473 if (unlikely(pagecnt_bias == 1)) {
8474 page_ref_add(page, USHRT_MAX - 1);
8475 rx_buffer->pagecnt_bias = USHRT_MAX;
8482 * igb_add_rx_frag - Add contents of Rx buffer to sk_buff
8483 * @rx_ring: rx descriptor ring to transact packets on
8484 * @rx_buffer: buffer containing page to add
8485 * @skb: sk_buff to place the data into
8486 * @size: size of buffer to be added
8488 * This function will add the data contained in rx_buffer->page to the skb.
8490 static void igb_add_rx_frag(struct igb_ring *rx_ring,
8491 struct igb_rx_buffer *rx_buffer,
8492 struct sk_buff *skb,
8495 #if (PAGE_SIZE < 8192)
8496 unsigned int truesize = igb_rx_pg_size(rx_ring) / 2;
8498 unsigned int truesize = ring_uses_build_skb(rx_ring) ?
8499 SKB_DATA_ALIGN(IGB_SKB_PAD + size) :
8500 SKB_DATA_ALIGN(size);
8502 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
8503 rx_buffer->page_offset, size, truesize);
8504 #if (PAGE_SIZE < 8192)
8505 rx_buffer->page_offset ^= truesize;
8507 rx_buffer->page_offset += truesize;
8511 static struct sk_buff *igb_construct_skb(struct igb_ring *rx_ring,
8512 struct igb_rx_buffer *rx_buffer,
8513 struct xdp_buff *xdp,
8516 #if (PAGE_SIZE < 8192)
8517 unsigned int truesize = igb_rx_pg_size(rx_ring) / 2;
8519 unsigned int truesize = SKB_DATA_ALIGN(xdp->data_end -
8520 xdp->data_hard_start);
8522 unsigned int size = xdp->data_end - xdp->data;
8523 unsigned int headlen;
8524 struct sk_buff *skb;
8526 /* prefetch first cache line of first page */
8527 net_prefetch(xdp->data);
8529 /* allocate a skb to store the frags */
8530 skb = napi_alloc_skb(&rx_ring->q_vector->napi, IGB_RX_HDR_LEN);
8535 skb_hwtstamps(skb)->hwtstamp = timestamp;
8537 /* Determine available headroom for copy */
8539 if (headlen > IGB_RX_HDR_LEN)
8540 headlen = eth_get_headlen(skb->dev, xdp->data, IGB_RX_HDR_LEN);
8542 /* align pull length to size of long to optimize memcpy performance */
8543 memcpy(__skb_put(skb, headlen), xdp->data, ALIGN(headlen, sizeof(long)));
8545 /* update all of the pointers */
8548 skb_add_rx_frag(skb, 0, rx_buffer->page,
8549 (xdp->data + headlen) - page_address(rx_buffer->page),
8551 #if (PAGE_SIZE < 8192)
8552 rx_buffer->page_offset ^= truesize;
8554 rx_buffer->page_offset += truesize;
8557 rx_buffer->pagecnt_bias++;
8563 static struct sk_buff *igb_build_skb(struct igb_ring *rx_ring,
8564 struct igb_rx_buffer *rx_buffer,
8565 struct xdp_buff *xdp,
8568 #if (PAGE_SIZE < 8192)
8569 unsigned int truesize = igb_rx_pg_size(rx_ring) / 2;
8571 unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
8572 SKB_DATA_ALIGN(xdp->data_end -
8573 xdp->data_hard_start);
8575 unsigned int metasize = xdp->data - xdp->data_meta;
8576 struct sk_buff *skb;
8578 /* prefetch first cache line of first page */
8579 net_prefetch(xdp->data_meta);
8581 /* build an skb around the page buffer */
8582 skb = napi_build_skb(xdp->data_hard_start, truesize);
8586 /* update pointers within the skb to store the data */
8587 skb_reserve(skb, xdp->data - xdp->data_hard_start);
8588 __skb_put(skb, xdp->data_end - xdp->data);
8591 skb_metadata_set(skb, metasize);
8594 skb_hwtstamps(skb)->hwtstamp = timestamp;
8596 /* update buffer offset */
8597 #if (PAGE_SIZE < 8192)
8598 rx_buffer->page_offset ^= truesize;
8600 rx_buffer->page_offset += truesize;
8606 static struct sk_buff *igb_run_xdp(struct igb_adapter *adapter,
8607 struct igb_ring *rx_ring,
8608 struct xdp_buff *xdp)
8610 int err, result = IGB_XDP_PASS;
8611 struct bpf_prog *xdp_prog;
8614 xdp_prog = READ_ONCE(rx_ring->xdp_prog);
8619 prefetchw(xdp->data_hard_start); /* xdp_frame write */
8621 act = bpf_prog_run_xdp(xdp_prog, xdp);
8626 result = igb_xdp_xmit_back(adapter, xdp);
8627 if (result == IGB_XDP_CONSUMED)
8631 err = xdp_do_redirect(adapter->netdev, xdp, xdp_prog);
8634 result = IGB_XDP_REDIR;
8637 bpf_warn_invalid_xdp_action(adapter->netdev, xdp_prog, act);
8641 trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
8644 result = IGB_XDP_CONSUMED;
8648 return ERR_PTR(-result);
8651 static unsigned int igb_rx_frame_truesize(struct igb_ring *rx_ring,
8654 unsigned int truesize;
8656 #if (PAGE_SIZE < 8192)
8657 truesize = igb_rx_pg_size(rx_ring) / 2; /* Must be power-of-2 */
8659 truesize = ring_uses_build_skb(rx_ring) ?
8660 SKB_DATA_ALIGN(IGB_SKB_PAD + size) +
8661 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) :
8662 SKB_DATA_ALIGN(size);
8667 static void igb_rx_buffer_flip(struct igb_ring *rx_ring,
8668 struct igb_rx_buffer *rx_buffer,
8671 unsigned int truesize = igb_rx_frame_truesize(rx_ring, size);
8672 #if (PAGE_SIZE < 8192)
8673 rx_buffer->page_offset ^= truesize;
8675 rx_buffer->page_offset += truesize;
8679 static inline void igb_rx_checksum(struct igb_ring *ring,
8680 union e1000_adv_rx_desc *rx_desc,
8681 struct sk_buff *skb)
8683 skb_checksum_none_assert(skb);
8685 /* Ignore Checksum bit is set */
8686 if (igb_test_staterr(rx_desc, E1000_RXD_STAT_IXSM))
8689 /* Rx checksum disabled via ethtool */
8690 if (!(ring->netdev->features & NETIF_F_RXCSUM))
8693 /* TCP/UDP checksum error bit is set */
8694 if (igb_test_staterr(rx_desc,
8695 E1000_RXDEXT_STATERR_TCPE |
8696 E1000_RXDEXT_STATERR_IPE)) {
8697 /* work around errata with sctp packets where the TCPE aka
8698 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
8699 * packets, (aka let the stack check the crc32c)
8701 if (!((skb->len == 60) &&
8702 test_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags))) {
8703 u64_stats_update_begin(&ring->rx_syncp);
8704 ring->rx_stats.csum_err++;
8705 u64_stats_update_end(&ring->rx_syncp);
8707 /* let the stack verify checksum errors */
8710 /* It must be a TCP or UDP packet with a valid checksum */
8711 if (igb_test_staterr(rx_desc, E1000_RXD_STAT_TCPCS |
8712 E1000_RXD_STAT_UDPCS))
8713 skb->ip_summed = CHECKSUM_UNNECESSARY;
8715 dev_dbg(ring->dev, "cksum success: bits %08X\n",
8716 le32_to_cpu(rx_desc->wb.upper.status_error));
8719 static inline void igb_rx_hash(struct igb_ring *ring,
8720 union e1000_adv_rx_desc *rx_desc,
8721 struct sk_buff *skb)
8723 if (ring->netdev->features & NETIF_F_RXHASH)
8725 le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
8730 * igb_is_non_eop - process handling of non-EOP buffers
8731 * @rx_ring: Rx ring being processed
8732 * @rx_desc: Rx descriptor for current buffer
8734 * This function updates next to clean. If the buffer is an EOP buffer
8735 * this function exits returning false, otherwise it will place the
8736 * sk_buff in the next buffer to be chained and return true indicating
8737 * that this is in fact a non-EOP buffer.
8739 static bool igb_is_non_eop(struct igb_ring *rx_ring,
8740 union e1000_adv_rx_desc *rx_desc)
8742 u32 ntc = rx_ring->next_to_clean + 1;
8744 /* fetch, update, and store next to clean */
8745 ntc = (ntc < rx_ring->count) ? ntc : 0;
8746 rx_ring->next_to_clean = ntc;
8748 prefetch(IGB_RX_DESC(rx_ring, ntc));
8750 if (likely(igb_test_staterr(rx_desc, E1000_RXD_STAT_EOP)))
8757 * igb_cleanup_headers - Correct corrupted or empty headers
8758 * @rx_ring: rx descriptor ring packet is being transacted on
8759 * @rx_desc: pointer to the EOP Rx descriptor
8760 * @skb: pointer to current skb being fixed
8762 * Address the case where we are pulling data in on pages only
8763 * and as such no data is present in the skb header.
8765 * In addition if skb is not at least 60 bytes we need to pad it so that
8766 * it is large enough to qualify as a valid Ethernet frame.
8768 * Returns true if an error was encountered and skb was freed.
8770 static bool igb_cleanup_headers(struct igb_ring *rx_ring,
8771 union e1000_adv_rx_desc *rx_desc,
8772 struct sk_buff *skb)
8774 /* XDP packets use error pointer so abort at this point */
8778 if (unlikely((igb_test_staterr(rx_desc,
8779 E1000_RXDEXT_ERR_FRAME_ERR_MASK)))) {
8780 struct net_device *netdev = rx_ring->netdev;
8781 if (!(netdev->features & NETIF_F_RXALL)) {
8782 dev_kfree_skb_any(skb);
8787 /* if eth_skb_pad returns an error the skb was freed */
8788 if (eth_skb_pad(skb))
8795 * igb_process_skb_fields - Populate skb header fields from Rx descriptor
8796 * @rx_ring: rx descriptor ring packet is being transacted on
8797 * @rx_desc: pointer to the EOP Rx descriptor
8798 * @skb: pointer to current skb being populated
8800 * This function checks the ring, descriptor, and packet information in
8801 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
8802 * other fields within the skb.
8804 static void igb_process_skb_fields(struct igb_ring *rx_ring,
8805 union e1000_adv_rx_desc *rx_desc,
8806 struct sk_buff *skb)
8808 struct net_device *dev = rx_ring->netdev;
8810 igb_rx_hash(rx_ring, rx_desc, skb);
8812 igb_rx_checksum(rx_ring, rx_desc, skb);
8814 if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TS) &&
8815 !igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP))
8816 igb_ptp_rx_rgtstamp(rx_ring->q_vector, skb);
8818 if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
8819 igb_test_staterr(rx_desc, E1000_RXD_STAT_VP)) {
8822 if (igb_test_staterr(rx_desc, E1000_RXDEXT_STATERR_LB) &&
8823 test_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &rx_ring->flags))
8824 vid = be16_to_cpu((__force __be16)rx_desc->wb.upper.vlan);
8826 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
8828 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
8831 skb_record_rx_queue(skb, rx_ring->queue_index);
8833 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
8836 static unsigned int igb_rx_offset(struct igb_ring *rx_ring)
8838 return ring_uses_build_skb(rx_ring) ? IGB_SKB_PAD : 0;
8841 static struct igb_rx_buffer *igb_get_rx_buffer(struct igb_ring *rx_ring,
8842 const unsigned int size, int *rx_buf_pgcnt)
8844 struct igb_rx_buffer *rx_buffer;
8846 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
8848 #if (PAGE_SIZE < 8192)
8849 page_count(rx_buffer->page);
8853 prefetchw(rx_buffer->page);
8855 /* we are reusing so sync this buffer for CPU use */
8856 dma_sync_single_range_for_cpu(rx_ring->dev,
8858 rx_buffer->page_offset,
8862 rx_buffer->pagecnt_bias--;
8867 static void igb_put_rx_buffer(struct igb_ring *rx_ring,
8868 struct igb_rx_buffer *rx_buffer, int rx_buf_pgcnt)
8870 if (igb_can_reuse_rx_page(rx_buffer, rx_buf_pgcnt)) {
8871 /* hand second half of page back to the ring */
8872 igb_reuse_rx_page(rx_ring, rx_buffer);
8874 /* We are not reusing the buffer so unmap it and free
8875 * any references we are holding to it
8877 dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
8878 igb_rx_pg_size(rx_ring), DMA_FROM_DEVICE,
8880 __page_frag_cache_drain(rx_buffer->page,
8881 rx_buffer->pagecnt_bias);
8884 /* clear contents of rx_buffer */
8885 rx_buffer->page = NULL;
8888 static int igb_clean_rx_irq(struct igb_q_vector *q_vector, const int budget)
8890 struct igb_adapter *adapter = q_vector->adapter;
8891 struct igb_ring *rx_ring = q_vector->rx.ring;
8892 struct sk_buff *skb = rx_ring->skb;
8893 unsigned int total_bytes = 0, total_packets = 0;
8894 u16 cleaned_count = igb_desc_unused(rx_ring);
8895 unsigned int xdp_xmit = 0;
8896 struct xdp_buff xdp;
8900 /* Frame size depend on rx_ring setup when PAGE_SIZE=4K */
8901 #if (PAGE_SIZE < 8192)
8902 frame_sz = igb_rx_frame_truesize(rx_ring, 0);
8904 xdp_init_buff(&xdp, frame_sz, &rx_ring->xdp_rxq);
8906 while (likely(total_packets < budget)) {
8907 union e1000_adv_rx_desc *rx_desc;
8908 struct igb_rx_buffer *rx_buffer;
8909 ktime_t timestamp = 0;
8914 /* return some buffers to hardware, one at a time is too slow */
8915 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
8916 igb_alloc_rx_buffers(rx_ring, cleaned_count);
8920 rx_desc = IGB_RX_DESC(rx_ring, rx_ring->next_to_clean);
8921 size = le16_to_cpu(rx_desc->wb.upper.length);
8925 /* This memory barrier is needed to keep us from reading
8926 * any other fields out of the rx_desc until we know the
8927 * descriptor has been written back
8931 rx_buffer = igb_get_rx_buffer(rx_ring, size, &rx_buf_pgcnt);
8932 pktbuf = page_address(rx_buffer->page) + rx_buffer->page_offset;
8934 /* pull rx packet timestamp if available and valid */
8935 if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
8938 ts_hdr_len = igb_ptp_rx_pktstamp(rx_ring->q_vector,
8939 pktbuf, ×tamp);
8941 pkt_offset += ts_hdr_len;
8945 /* retrieve a buffer from the ring */
8947 unsigned char *hard_start = pktbuf - igb_rx_offset(rx_ring);
8948 unsigned int offset = pkt_offset + igb_rx_offset(rx_ring);
8950 xdp_prepare_buff(&xdp, hard_start, offset, size, true);
8951 xdp_buff_clear_frags_flag(&xdp);
8952 #if (PAGE_SIZE > 4096)
8953 /* At larger PAGE_SIZE, frame_sz depend on len size */
8954 xdp.frame_sz = igb_rx_frame_truesize(rx_ring, size);
8956 skb = igb_run_xdp(adapter, rx_ring, &xdp);
8960 unsigned int xdp_res = -PTR_ERR(skb);
8962 if (xdp_res & (IGB_XDP_TX | IGB_XDP_REDIR)) {
8963 xdp_xmit |= xdp_res;
8964 igb_rx_buffer_flip(rx_ring, rx_buffer, size);
8966 rx_buffer->pagecnt_bias++;
8969 total_bytes += size;
8971 igb_add_rx_frag(rx_ring, rx_buffer, skb, size);
8972 else if (ring_uses_build_skb(rx_ring))
8973 skb = igb_build_skb(rx_ring, rx_buffer, &xdp,
8976 skb = igb_construct_skb(rx_ring, rx_buffer,
8979 /* exit if we failed to retrieve a buffer */
8981 rx_ring->rx_stats.alloc_failed++;
8982 rx_buffer->pagecnt_bias++;
8986 igb_put_rx_buffer(rx_ring, rx_buffer, rx_buf_pgcnt);
8989 /* fetch next buffer in frame if non-eop */
8990 if (igb_is_non_eop(rx_ring, rx_desc))
8993 /* verify the packet layout is correct */
8994 if (igb_cleanup_headers(rx_ring, rx_desc, skb)) {
8999 /* probably a little skewed due to removing CRC */
9000 total_bytes += skb->len;
9002 /* populate checksum, timestamp, VLAN, and protocol */
9003 igb_process_skb_fields(rx_ring, rx_desc, skb);
9005 napi_gro_receive(&q_vector->napi, skb);
9007 /* reset skb pointer */
9010 /* update budget accounting */
9014 /* place incomplete frames back on ring for completion */
9017 if (xdp_xmit & IGB_XDP_REDIR)
9020 if (xdp_xmit & IGB_XDP_TX) {
9021 struct igb_ring *tx_ring = igb_xdp_tx_queue_mapping(adapter);
9023 igb_xdp_ring_update_tail(tx_ring);
9026 u64_stats_update_begin(&rx_ring->rx_syncp);
9027 rx_ring->rx_stats.packets += total_packets;
9028 rx_ring->rx_stats.bytes += total_bytes;
9029 u64_stats_update_end(&rx_ring->rx_syncp);
9030 q_vector->rx.total_packets += total_packets;
9031 q_vector->rx.total_bytes += total_bytes;
9034 igb_alloc_rx_buffers(rx_ring, cleaned_count);
9036 return total_packets;
9039 static bool igb_alloc_mapped_page(struct igb_ring *rx_ring,
9040 struct igb_rx_buffer *bi)
9042 struct page *page = bi->page;
9045 /* since we are recycling buffers we should seldom need to alloc */
9049 /* alloc new page for storage */
9050 page = dev_alloc_pages(igb_rx_pg_order(rx_ring));
9051 if (unlikely(!page)) {
9052 rx_ring->rx_stats.alloc_failed++;
9056 /* map page for use */
9057 dma = dma_map_page_attrs(rx_ring->dev, page, 0,
9058 igb_rx_pg_size(rx_ring),
9062 /* if mapping failed free memory back to system since
9063 * there isn't much point in holding memory we can't use
9065 if (dma_mapping_error(rx_ring->dev, dma)) {
9066 __free_pages(page, igb_rx_pg_order(rx_ring));
9068 rx_ring->rx_stats.alloc_failed++;
9074 bi->page_offset = igb_rx_offset(rx_ring);
9075 page_ref_add(page, USHRT_MAX - 1);
9076 bi->pagecnt_bias = USHRT_MAX;
9082 * igb_alloc_rx_buffers - Replace used receive buffers
9083 * @rx_ring: rx descriptor ring to allocate new receive buffers
9084 * @cleaned_count: count of buffers to allocate
9086 void igb_alloc_rx_buffers(struct igb_ring *rx_ring, u16 cleaned_count)
9088 union e1000_adv_rx_desc *rx_desc;
9089 struct igb_rx_buffer *bi;
9090 u16 i = rx_ring->next_to_use;
9097 rx_desc = IGB_RX_DESC(rx_ring, i);
9098 bi = &rx_ring->rx_buffer_info[i];
9099 i -= rx_ring->count;
9101 bufsz = igb_rx_bufsz(rx_ring);
9104 if (!igb_alloc_mapped_page(rx_ring, bi))
9107 /* sync the buffer for use by the device */
9108 dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
9109 bi->page_offset, bufsz,
9112 /* Refresh the desc even if buffer_addrs didn't change
9113 * because each write-back erases this info.
9115 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
9121 rx_desc = IGB_RX_DESC(rx_ring, 0);
9122 bi = rx_ring->rx_buffer_info;
9123 i -= rx_ring->count;
9126 /* clear the length for the next_to_use descriptor */
9127 rx_desc->wb.upper.length = 0;
9130 } while (cleaned_count);
9132 i += rx_ring->count;
9134 if (rx_ring->next_to_use != i) {
9135 /* record the next descriptor to use */
9136 rx_ring->next_to_use = i;
9138 /* update next to alloc since we have filled the ring */
9139 rx_ring->next_to_alloc = i;
9141 /* Force memory writes to complete before letting h/w
9142 * know there are new descriptors to fetch. (Only
9143 * applicable for weak-ordered memory model archs,
9147 writel(i, rx_ring->tail);
9153 * @netdev: pointer to netdev struct
9154 * @ifr: interface structure
9155 * @cmd: ioctl command to execute
9157 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
9159 struct igb_adapter *adapter = netdev_priv(netdev);
9160 struct mii_ioctl_data *data = if_mii(ifr);
9162 if (adapter->hw.phy.media_type != e1000_media_type_copper)
9167 data->phy_id = adapter->hw.phy.addr;
9170 if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
9183 * @netdev: pointer to netdev struct
9184 * @ifr: interface structure
9185 * @cmd: ioctl command to execute
9187 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
9193 return igb_mii_ioctl(netdev, ifr, cmd);
9195 return igb_ptp_get_ts_config(netdev, ifr);
9197 return igb_ptp_set_ts_config(netdev, ifr);
9203 void igb_read_pci_cfg(struct e1000_hw *hw, u32 reg, u16 *value)
9205 struct igb_adapter *adapter = hw->back;
9207 pci_read_config_word(adapter->pdev, reg, value);
9210 void igb_write_pci_cfg(struct e1000_hw *hw, u32 reg, u16 *value)
9212 struct igb_adapter *adapter = hw->back;
9214 pci_write_config_word(adapter->pdev, reg, *value);
9217 s32 igb_read_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
9219 struct igb_adapter *adapter = hw->back;
9221 if (pcie_capability_read_word(adapter->pdev, reg, value))
9222 return -E1000_ERR_CONFIG;
9227 s32 igb_write_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
9229 struct igb_adapter *adapter = hw->back;
9231 if (pcie_capability_write_word(adapter->pdev, reg, *value))
9232 return -E1000_ERR_CONFIG;
9237 static void igb_vlan_mode(struct net_device *netdev, netdev_features_t features)
9239 struct igb_adapter *adapter = netdev_priv(netdev);
9240 struct e1000_hw *hw = &adapter->hw;
9242 bool enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
9245 /* enable VLAN tag insert/strip */
9246 ctrl = rd32(E1000_CTRL);
9247 ctrl |= E1000_CTRL_VME;
9248 wr32(E1000_CTRL, ctrl);
9250 /* Disable CFI check */
9251 rctl = rd32(E1000_RCTL);
9252 rctl &= ~E1000_RCTL_CFIEN;
9253 wr32(E1000_RCTL, rctl);
9255 /* disable VLAN tag insert/strip */
9256 ctrl = rd32(E1000_CTRL);
9257 ctrl &= ~E1000_CTRL_VME;
9258 wr32(E1000_CTRL, ctrl);
9261 igb_set_vf_vlan_strip(adapter, adapter->vfs_allocated_count, enable);
9264 static int igb_vlan_rx_add_vid(struct net_device *netdev,
9265 __be16 proto, u16 vid)
9267 struct igb_adapter *adapter = netdev_priv(netdev);
9268 struct e1000_hw *hw = &adapter->hw;
9269 int pf_id = adapter->vfs_allocated_count;
9271 /* add the filter since PF can receive vlans w/o entry in vlvf */
9272 if (!vid || !(adapter->flags & IGB_FLAG_VLAN_PROMISC))
9273 igb_vfta_set(hw, vid, pf_id, true, !!vid);
9275 set_bit(vid, adapter->active_vlans);
9280 static int igb_vlan_rx_kill_vid(struct net_device *netdev,
9281 __be16 proto, u16 vid)
9283 struct igb_adapter *adapter = netdev_priv(netdev);
9284 int pf_id = adapter->vfs_allocated_count;
9285 struct e1000_hw *hw = &adapter->hw;
9287 /* remove VID from filter table */
9288 if (vid && !(adapter->flags & IGB_FLAG_VLAN_PROMISC))
9289 igb_vfta_set(hw, vid, pf_id, false, true);
9291 clear_bit(vid, adapter->active_vlans);
9296 static void igb_restore_vlan(struct igb_adapter *adapter)
9300 igb_vlan_mode(adapter->netdev, adapter->netdev->features);
9301 igb_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
9303 for_each_set_bit_from(vid, adapter->active_vlans, VLAN_N_VID)
9304 igb_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
9307 int igb_set_spd_dplx(struct igb_adapter *adapter, u32 spd, u8 dplx)
9309 struct pci_dev *pdev = adapter->pdev;
9310 struct e1000_mac_info *mac = &adapter->hw.mac;
9314 /* Make sure dplx is at most 1 bit and lsb of speed is not set
9315 * for the switch() below to work
9317 if ((spd & 1) || (dplx & ~1))
9320 /* Fiber NIC's only allow 1000 gbps Full duplex
9321 * and 100Mbps Full duplex for 100baseFx sfp
9323 if (adapter->hw.phy.media_type == e1000_media_type_internal_serdes) {
9324 switch (spd + dplx) {
9325 case SPEED_10 + DUPLEX_HALF:
9326 case SPEED_10 + DUPLEX_FULL:
9327 case SPEED_100 + DUPLEX_HALF:
9334 switch (spd + dplx) {
9335 case SPEED_10 + DUPLEX_HALF:
9336 mac->forced_speed_duplex = ADVERTISE_10_HALF;
9338 case SPEED_10 + DUPLEX_FULL:
9339 mac->forced_speed_duplex = ADVERTISE_10_FULL;
9341 case SPEED_100 + DUPLEX_HALF:
9342 mac->forced_speed_duplex = ADVERTISE_100_HALF;
9344 case SPEED_100 + DUPLEX_FULL:
9345 mac->forced_speed_duplex = ADVERTISE_100_FULL;
9347 case SPEED_1000 + DUPLEX_FULL:
9349 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
9351 case SPEED_1000 + DUPLEX_HALF: /* not supported */
9356 /* clear MDI, MDI(-X) override is only allowed when autoneg enabled */
9357 adapter->hw.phy.mdix = AUTO_ALL_MODES;
9362 dev_err(&pdev->dev, "Unsupported Speed/Duplex configuration\n");
9366 static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake,
9369 struct net_device *netdev = pci_get_drvdata(pdev);
9370 struct igb_adapter *adapter = netdev_priv(netdev);
9371 struct e1000_hw *hw = &adapter->hw;
9372 u32 ctrl, rctl, status;
9373 u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
9377 netif_device_detach(netdev);
9379 if (netif_running(netdev))
9380 __igb_close(netdev, true);
9382 igb_ptp_suspend(adapter);
9384 igb_clear_interrupt_scheme(adapter);
9387 status = rd32(E1000_STATUS);
9388 if (status & E1000_STATUS_LU)
9389 wufc &= ~E1000_WUFC_LNKC;
9392 igb_setup_rctl(adapter);
9393 igb_set_rx_mode(netdev);
9395 /* turn on all-multi mode if wake on multicast is enabled */
9396 if (wufc & E1000_WUFC_MC) {
9397 rctl = rd32(E1000_RCTL);
9398 rctl |= E1000_RCTL_MPE;
9399 wr32(E1000_RCTL, rctl);
9402 ctrl = rd32(E1000_CTRL);
9403 ctrl |= E1000_CTRL_ADVD3WUC;
9404 wr32(E1000_CTRL, ctrl);
9406 /* Allow time for pending master requests to run */
9407 igb_disable_pcie_master(hw);
9409 wr32(E1000_WUC, E1000_WUC_PME_EN);
9410 wr32(E1000_WUFC, wufc);
9413 wr32(E1000_WUFC, 0);
9416 wake = wufc || adapter->en_mng_pt;
9418 igb_power_down_link(adapter);
9420 igb_power_up_link(adapter);
9423 *enable_wake = wake;
9425 /* Release control of h/w to f/w. If f/w is AMT enabled, this
9426 * would have already happened in close and is redundant.
9428 igb_release_hw_control(adapter);
9430 pci_disable_device(pdev);
9435 static void igb_deliver_wake_packet(struct net_device *netdev)
9437 struct igb_adapter *adapter = netdev_priv(netdev);
9438 struct e1000_hw *hw = &adapter->hw;
9439 struct sk_buff *skb;
9442 wupl = rd32(E1000_WUPL) & E1000_WUPL_MASK;
9444 /* WUPM stores only the first 128 bytes of the wake packet.
9445 * Read the packet only if we have the whole thing.
9447 if ((wupl == 0) || (wupl > E1000_WUPM_BYTES))
9450 skb = netdev_alloc_skb_ip_align(netdev, E1000_WUPM_BYTES);
9456 /* Ensure reads are 32-bit aligned */
9457 wupl = roundup(wupl, 4);
9459 memcpy_fromio(skb->data, hw->hw_addr + E1000_WUPM_REG(0), wupl);
9461 skb->protocol = eth_type_trans(skb, netdev);
9465 static int __maybe_unused igb_suspend(struct device *dev)
9467 return __igb_shutdown(to_pci_dev(dev), NULL, 0);
9470 static int __maybe_unused __igb_resume(struct device *dev, bool rpm)
9472 struct pci_dev *pdev = to_pci_dev(dev);
9473 struct net_device *netdev = pci_get_drvdata(pdev);
9474 struct igb_adapter *adapter = netdev_priv(netdev);
9475 struct e1000_hw *hw = &adapter->hw;
9478 pci_set_power_state(pdev, PCI_D0);
9479 pci_restore_state(pdev);
9480 pci_save_state(pdev);
9482 if (!pci_device_is_present(pdev))
9484 err = pci_enable_device_mem(pdev);
9487 "igb: Cannot enable PCI device from suspend\n");
9490 pci_set_master(pdev);
9492 pci_enable_wake(pdev, PCI_D3hot, 0);
9493 pci_enable_wake(pdev, PCI_D3cold, 0);
9495 if (igb_init_interrupt_scheme(adapter, true)) {
9496 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
9502 /* let the f/w know that the h/w is now under the control of the
9505 igb_get_hw_control(adapter);
9507 val = rd32(E1000_WUS);
9508 if (val & WAKE_PKT_WUS)
9509 igb_deliver_wake_packet(netdev);
9511 wr32(E1000_WUS, ~0);
9515 if (!err && netif_running(netdev))
9516 err = __igb_open(netdev, true);
9519 netif_device_attach(netdev);
9526 static int __maybe_unused igb_resume(struct device *dev)
9528 return __igb_resume(dev, false);
9531 static int __maybe_unused igb_runtime_idle(struct device *dev)
9533 struct net_device *netdev = dev_get_drvdata(dev);
9534 struct igb_adapter *adapter = netdev_priv(netdev);
9536 if (!igb_has_link(adapter))
9537 pm_schedule_suspend(dev, MSEC_PER_SEC * 5);
9542 static int __maybe_unused igb_runtime_suspend(struct device *dev)
9544 return __igb_shutdown(to_pci_dev(dev), NULL, 1);
9547 static int __maybe_unused igb_runtime_resume(struct device *dev)
9549 return __igb_resume(dev, true);
9552 static void igb_shutdown(struct pci_dev *pdev)
9556 __igb_shutdown(pdev, &wake, 0);
9558 if (system_state == SYSTEM_POWER_OFF) {
9559 pci_wake_from_d3(pdev, wake);
9560 pci_set_power_state(pdev, PCI_D3hot);
9564 static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs)
9566 #ifdef CONFIG_PCI_IOV
9570 return igb_disable_sriov(dev, true);
9572 err = igb_enable_sriov(dev, num_vfs, true);
9573 return err ? err : num_vfs;
9580 * igb_io_error_detected - called when PCI error is detected
9581 * @pdev: Pointer to PCI device
9582 * @state: The current pci connection state
9584 * This function is called after a PCI bus error affecting
9585 * this device has been detected.
9587 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
9588 pci_channel_state_t state)
9590 struct net_device *netdev = pci_get_drvdata(pdev);
9591 struct igb_adapter *adapter = netdev_priv(netdev);
9593 if (state == pci_channel_io_normal) {
9594 dev_warn(&pdev->dev, "Non-correctable non-fatal error reported.\n");
9595 return PCI_ERS_RESULT_CAN_RECOVER;
9598 netif_device_detach(netdev);
9600 if (state == pci_channel_io_perm_failure)
9601 return PCI_ERS_RESULT_DISCONNECT;
9603 if (netif_running(netdev))
9605 pci_disable_device(pdev);
9607 /* Request a slot reset. */
9608 return PCI_ERS_RESULT_NEED_RESET;
9612 * igb_io_slot_reset - called after the pci bus has been reset.
9613 * @pdev: Pointer to PCI device
9615 * Restart the card from scratch, as if from a cold-boot. Implementation
9616 * resembles the first-half of the __igb_resume routine.
9618 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
9620 struct net_device *netdev = pci_get_drvdata(pdev);
9621 struct igb_adapter *adapter = netdev_priv(netdev);
9622 struct e1000_hw *hw = &adapter->hw;
9623 pci_ers_result_t result;
9625 if (pci_enable_device_mem(pdev)) {
9627 "Cannot re-enable PCI device after reset.\n");
9628 result = PCI_ERS_RESULT_DISCONNECT;
9630 pci_set_master(pdev);
9631 pci_restore_state(pdev);
9632 pci_save_state(pdev);
9634 pci_enable_wake(pdev, PCI_D3hot, 0);
9635 pci_enable_wake(pdev, PCI_D3cold, 0);
9637 /* In case of PCI error, adapter lose its HW address
9638 * so we should re-assign it here.
9640 hw->hw_addr = adapter->io_addr;
9643 wr32(E1000_WUS, ~0);
9644 result = PCI_ERS_RESULT_RECOVERED;
9651 * igb_io_resume - called when traffic can start flowing again.
9652 * @pdev: Pointer to PCI device
9654 * This callback is called when the error recovery driver tells us that
9655 * its OK to resume normal operation. Implementation resembles the
9656 * second-half of the __igb_resume routine.
9658 static void igb_io_resume(struct pci_dev *pdev)
9660 struct net_device *netdev = pci_get_drvdata(pdev);
9661 struct igb_adapter *adapter = netdev_priv(netdev);
9663 if (netif_running(netdev)) {
9664 if (igb_up(adapter)) {
9665 dev_err(&pdev->dev, "igb_up failed after reset\n");
9670 netif_device_attach(netdev);
9672 /* let the f/w know that the h/w is now under the control of the
9675 igb_get_hw_control(adapter);
9679 * igb_rar_set_index - Sync RAL[index] and RAH[index] registers with MAC table
9680 * @adapter: Pointer to adapter structure
9681 * @index: Index of the RAR entry which need to be synced with MAC table
9683 static void igb_rar_set_index(struct igb_adapter *adapter, u32 index)
9685 struct e1000_hw *hw = &adapter->hw;
9686 u32 rar_low, rar_high;
9687 u8 *addr = adapter->mac_table[index].addr;
9689 /* HW expects these to be in network order when they are plugged
9690 * into the registers which are little endian. In order to guarantee
9691 * that ordering we need to do an leXX_to_cpup here in order to be
9692 * ready for the byteswap that occurs with writel
9694 rar_low = le32_to_cpup((__le32 *)(addr));
9695 rar_high = le16_to_cpup((__le16 *)(addr + 4));
9697 /* Indicate to hardware the Address is Valid. */
9698 if (adapter->mac_table[index].state & IGB_MAC_STATE_IN_USE) {
9699 if (is_valid_ether_addr(addr))
9700 rar_high |= E1000_RAH_AV;
9702 if (adapter->mac_table[index].state & IGB_MAC_STATE_SRC_ADDR)
9703 rar_high |= E1000_RAH_ASEL_SRC_ADDR;
9705 switch (hw->mac.type) {
9708 if (adapter->mac_table[index].state &
9709 IGB_MAC_STATE_QUEUE_STEERING)
9710 rar_high |= E1000_RAH_QSEL_ENABLE;
9712 rar_high |= E1000_RAH_POOL_1 *
9713 adapter->mac_table[index].queue;
9716 rar_high |= E1000_RAH_POOL_1 <<
9717 adapter->mac_table[index].queue;
9722 wr32(E1000_RAL(index), rar_low);
9724 wr32(E1000_RAH(index), rar_high);
9728 static int igb_set_vf_mac(struct igb_adapter *adapter,
9729 int vf, unsigned char *mac_addr)
9731 struct e1000_hw *hw = &adapter->hw;
9732 /* VF MAC addresses start at end of receive addresses and moves
9733 * towards the first, as a result a collision should not be possible
9735 int rar_entry = hw->mac.rar_entry_count - (vf + 1);
9736 unsigned char *vf_mac_addr = adapter->vf_data[vf].vf_mac_addresses;
9738 ether_addr_copy(vf_mac_addr, mac_addr);
9739 ether_addr_copy(adapter->mac_table[rar_entry].addr, mac_addr);
9740 adapter->mac_table[rar_entry].queue = vf;
9741 adapter->mac_table[rar_entry].state |= IGB_MAC_STATE_IN_USE;
9742 igb_rar_set_index(adapter, rar_entry);
9747 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
9749 struct igb_adapter *adapter = netdev_priv(netdev);
9751 if (vf >= adapter->vfs_allocated_count)
9754 /* Setting the VF MAC to 0 reverts the IGB_VF_FLAG_PF_SET_MAC
9755 * flag and allows to overwrite the MAC via VF netdev. This
9756 * is necessary to allow libvirt a way to restore the original
9757 * MAC after unbinding vfio-pci and reloading igbvf after shutting
9760 if (is_zero_ether_addr(mac)) {
9761 adapter->vf_data[vf].flags &= ~IGB_VF_FLAG_PF_SET_MAC;
9762 dev_info(&adapter->pdev->dev,
9763 "remove administratively set MAC on VF %d\n",
9765 } else if (is_valid_ether_addr(mac)) {
9766 adapter->vf_data[vf].flags |= IGB_VF_FLAG_PF_SET_MAC;
9767 dev_info(&adapter->pdev->dev, "setting MAC %pM on VF %d\n",
9769 dev_info(&adapter->pdev->dev,
9770 "Reload the VF driver to make this change effective.");
9771 /* Generate additional warning if PF is down */
9772 if (test_bit(__IGB_DOWN, &adapter->state)) {
9773 dev_warn(&adapter->pdev->dev,
9774 "The VF MAC address has been set, but the PF device is not up.\n");
9775 dev_warn(&adapter->pdev->dev,
9776 "Bring the PF device up before attempting to use the VF device.\n");
9781 return igb_set_vf_mac(adapter, vf, mac);
9784 static int igb_link_mbps(int internal_link_speed)
9786 switch (internal_link_speed) {
9796 static void igb_set_vf_rate_limit(struct e1000_hw *hw, int vf, int tx_rate,
9803 /* Calculate the rate factor values to set */
9804 rf_int = link_speed / tx_rate;
9805 rf_dec = (link_speed - (rf_int * tx_rate));
9806 rf_dec = (rf_dec * BIT(E1000_RTTBCNRC_RF_INT_SHIFT)) /
9809 bcnrc_val = E1000_RTTBCNRC_RS_ENA;
9810 bcnrc_val |= ((rf_int << E1000_RTTBCNRC_RF_INT_SHIFT) &
9811 E1000_RTTBCNRC_RF_INT_MASK);
9812 bcnrc_val |= (rf_dec & E1000_RTTBCNRC_RF_DEC_MASK);
9817 wr32(E1000_RTTDQSEL, vf); /* vf X uses queue X */
9818 /* Set global transmit compensation time to the MMW_SIZE in RTTBCNRM
9819 * register. MMW_SIZE=0x014 if 9728-byte jumbo is supported.
9821 wr32(E1000_RTTBCNRM, 0x14);
9822 wr32(E1000_RTTBCNRC, bcnrc_val);
9825 static void igb_check_vf_rate_limit(struct igb_adapter *adapter)
9827 int actual_link_speed, i;
9828 bool reset_rate = false;
9830 /* VF TX rate limit was not set or not supported */
9831 if ((adapter->vf_rate_link_speed == 0) ||
9832 (adapter->hw.mac.type != e1000_82576))
9835 actual_link_speed = igb_link_mbps(adapter->link_speed);
9836 if (actual_link_speed != adapter->vf_rate_link_speed) {
9838 adapter->vf_rate_link_speed = 0;
9839 dev_info(&adapter->pdev->dev,
9840 "Link speed has been changed. VF Transmit rate is disabled\n");
9843 for (i = 0; i < adapter->vfs_allocated_count; i++) {
9845 adapter->vf_data[i].tx_rate = 0;
9847 igb_set_vf_rate_limit(&adapter->hw, i,
9848 adapter->vf_data[i].tx_rate,
9853 static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf,
9854 int min_tx_rate, int max_tx_rate)
9856 struct igb_adapter *adapter = netdev_priv(netdev);
9857 struct e1000_hw *hw = &adapter->hw;
9858 int actual_link_speed;
9860 if (hw->mac.type != e1000_82576)
9866 actual_link_speed = igb_link_mbps(adapter->link_speed);
9867 if ((vf >= adapter->vfs_allocated_count) ||
9868 (!(rd32(E1000_STATUS) & E1000_STATUS_LU)) ||
9869 (max_tx_rate < 0) ||
9870 (max_tx_rate > actual_link_speed))
9873 adapter->vf_rate_link_speed = actual_link_speed;
9874 adapter->vf_data[vf].tx_rate = (u16)max_tx_rate;
9875 igb_set_vf_rate_limit(hw, vf, max_tx_rate, actual_link_speed);
9880 static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf,
9883 struct igb_adapter *adapter = netdev_priv(netdev);
9884 struct e1000_hw *hw = &adapter->hw;
9885 u32 reg_val, reg_offset;
9887 if (!adapter->vfs_allocated_count)
9890 if (vf >= adapter->vfs_allocated_count)
9893 reg_offset = (hw->mac.type == e1000_82576) ? E1000_DTXSWC : E1000_TXSWC;
9894 reg_val = rd32(reg_offset);
9896 reg_val |= (BIT(vf) |
9897 BIT(vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT));
9899 reg_val &= ~(BIT(vf) |
9900 BIT(vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT));
9901 wr32(reg_offset, reg_val);
9903 adapter->vf_data[vf].spoofchk_enabled = setting;
9907 static int igb_ndo_set_vf_trust(struct net_device *netdev, int vf, bool setting)
9909 struct igb_adapter *adapter = netdev_priv(netdev);
9911 if (vf >= adapter->vfs_allocated_count)
9913 if (adapter->vf_data[vf].trusted == setting)
9916 adapter->vf_data[vf].trusted = setting;
9918 dev_info(&adapter->pdev->dev, "VF %u is %strusted\n",
9919 vf, setting ? "" : "not ");
9923 static int igb_ndo_get_vf_config(struct net_device *netdev,
9924 int vf, struct ifla_vf_info *ivi)
9926 struct igb_adapter *adapter = netdev_priv(netdev);
9927 if (vf >= adapter->vfs_allocated_count)
9930 memcpy(&ivi->mac, adapter->vf_data[vf].vf_mac_addresses, ETH_ALEN);
9931 ivi->max_tx_rate = adapter->vf_data[vf].tx_rate;
9932 ivi->min_tx_rate = 0;
9933 ivi->vlan = adapter->vf_data[vf].pf_vlan;
9934 ivi->qos = adapter->vf_data[vf].pf_qos;
9935 ivi->spoofchk = adapter->vf_data[vf].spoofchk_enabled;
9936 ivi->trusted = adapter->vf_data[vf].trusted;
9940 static void igb_vmm_control(struct igb_adapter *adapter)
9942 struct e1000_hw *hw = &adapter->hw;
9945 switch (hw->mac.type) {
9951 /* replication is not supported for 82575 */
9954 /* notify HW that the MAC is adding vlan tags */
9955 reg = rd32(E1000_DTXCTL);
9956 reg |= E1000_DTXCTL_VLAN_ADDED;
9957 wr32(E1000_DTXCTL, reg);
9960 /* enable replication vlan tag stripping */
9961 reg = rd32(E1000_RPLOLR);
9962 reg |= E1000_RPLOLR_STRVLAN;
9963 wr32(E1000_RPLOLR, reg);
9966 /* none of the above registers are supported by i350 */
9970 if (adapter->vfs_allocated_count) {
9971 igb_vmdq_set_loopback_pf(hw, true);
9972 igb_vmdq_set_replication_pf(hw, true);
9973 igb_vmdq_set_anti_spoofing_pf(hw, true,
9974 adapter->vfs_allocated_count);
9976 igb_vmdq_set_loopback_pf(hw, false);
9977 igb_vmdq_set_replication_pf(hw, false);
9981 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba)
9983 struct e1000_hw *hw = &adapter->hw;
9988 if (hw->mac.type > e1000_82580) {
9989 if (adapter->flags & IGB_FLAG_DMAC) {
9990 /* force threshold to 0. */
9991 wr32(E1000_DMCTXTH, 0);
9993 /* DMA Coalescing high water mark needs to be greater
9994 * than the Rx threshold. Set hwm to PBA - max frame
9995 * size in 16B units, capping it at PBA - 6KB.
9997 hwm = 64 * (pba - 6);
9998 reg = rd32(E1000_FCRTC);
9999 reg &= ~E1000_FCRTC_RTH_COAL_MASK;
10000 reg |= ((hwm << E1000_FCRTC_RTH_COAL_SHIFT)
10001 & E1000_FCRTC_RTH_COAL_MASK);
10002 wr32(E1000_FCRTC, reg);
10004 /* Set the DMA Coalescing Rx threshold to PBA - 2 * max
10005 * frame size, capping it at PBA - 10KB.
10007 dmac_thr = pba - 10;
10008 reg = rd32(E1000_DMACR);
10009 reg &= ~E1000_DMACR_DMACTHR_MASK;
10010 reg |= ((dmac_thr << E1000_DMACR_DMACTHR_SHIFT)
10011 & E1000_DMACR_DMACTHR_MASK);
10013 /* transition to L0x or L1 if available..*/
10014 reg |= (E1000_DMACR_DMAC_EN | E1000_DMACR_DMAC_LX_MASK);
10016 /* watchdog timer= +-1000 usec in 32usec intervals */
10017 reg |= (1000 >> 5);
10019 /* Disable BMC-to-OS Watchdog Enable */
10020 if (hw->mac.type != e1000_i354)
10021 reg &= ~E1000_DMACR_DC_BMC2OSW_EN;
10022 wr32(E1000_DMACR, reg);
10024 /* no lower threshold to disable
10025 * coalescing(smart fifb)-UTRESH=0
10027 wr32(E1000_DMCRTRH, 0);
10029 reg = (IGB_DMCTLX_DCFLUSH_DIS | 0x4);
10031 wr32(E1000_DMCTLX, reg);
10033 /* free space in tx packet buffer to wake from
10036 wr32(E1000_DMCTXTH, (IGB_MIN_TXPBSIZE -
10037 (IGB_TX_BUF_4096 + adapter->max_frame_size)) >> 6);
10040 if (hw->mac.type >= e1000_i210 ||
10041 (adapter->flags & IGB_FLAG_DMAC)) {
10042 reg = rd32(E1000_PCIEMISC);
10043 reg |= E1000_PCIEMISC_LX_DECISION;
10044 wr32(E1000_PCIEMISC, reg);
10045 } /* endif adapter->dmac is not disabled */
10046 } else if (hw->mac.type == e1000_82580) {
10047 u32 reg = rd32(E1000_PCIEMISC);
10049 wr32(E1000_PCIEMISC, reg & ~E1000_PCIEMISC_LX_DECISION);
10050 wr32(E1000_DMACR, 0);
10055 * igb_read_i2c_byte - Reads 8 bit word over I2C
10056 * @hw: pointer to hardware structure
10057 * @byte_offset: byte offset to read
10058 * @dev_addr: device address
10059 * @data: value read
10061 * Performs byte read operation over I2C interface at
10062 * a specified device address.
10064 s32 igb_read_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
10065 u8 dev_addr, u8 *data)
10067 struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
10068 struct i2c_client *this_client = adapter->i2c_client;
10073 return E1000_ERR_I2C;
10075 swfw_mask = E1000_SWFW_PHY0_SM;
10077 if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask))
10078 return E1000_ERR_SWFW_SYNC;
10080 status = i2c_smbus_read_byte_data(this_client, byte_offset);
10081 hw->mac.ops.release_swfw_sync(hw, swfw_mask);
10084 return E1000_ERR_I2C;
10092 * igb_write_i2c_byte - Writes 8 bit word over I2C
10093 * @hw: pointer to hardware structure
10094 * @byte_offset: byte offset to write
10095 * @dev_addr: device address
10096 * @data: value to write
10098 * Performs byte write operation over I2C interface at
10099 * a specified device address.
10101 s32 igb_write_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
10102 u8 dev_addr, u8 data)
10104 struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
10105 struct i2c_client *this_client = adapter->i2c_client;
10107 u16 swfw_mask = E1000_SWFW_PHY0_SM;
10110 return E1000_ERR_I2C;
10112 if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask))
10113 return E1000_ERR_SWFW_SYNC;
10114 status = i2c_smbus_write_byte_data(this_client, byte_offset, data);
10115 hw->mac.ops.release_swfw_sync(hw, swfw_mask);
10118 return E1000_ERR_I2C;
10124 int igb_reinit_queues(struct igb_adapter *adapter)
10126 struct net_device *netdev = adapter->netdev;
10127 struct pci_dev *pdev = adapter->pdev;
10130 if (netif_running(netdev))
10133 igb_reset_interrupt_capability(adapter);
10135 if (igb_init_interrupt_scheme(adapter, true)) {
10136 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
10140 if (netif_running(netdev))
10141 err = igb_open(netdev);
10146 static void igb_nfc_filter_exit(struct igb_adapter *adapter)
10148 struct igb_nfc_filter *rule;
10150 spin_lock(&adapter->nfc_lock);
10152 hlist_for_each_entry(rule, &adapter->nfc_filter_list, nfc_node)
10153 igb_erase_filter(adapter, rule);
10155 hlist_for_each_entry(rule, &adapter->cls_flower_list, nfc_node)
10156 igb_erase_filter(adapter, rule);
10158 spin_unlock(&adapter->nfc_lock);
10161 static void igb_nfc_filter_restore(struct igb_adapter *adapter)
10163 struct igb_nfc_filter *rule;
10165 spin_lock(&adapter->nfc_lock);
10167 hlist_for_each_entry(rule, &adapter->nfc_filter_list, nfc_node)
10168 igb_add_filter(adapter, rule);
10170 spin_unlock(&adapter->nfc_lock);