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/aer.h>
32 #include <linux/prefetch.h>
33 #include <linux/bpf.h>
34 #include <linux/bpf_trace.h>
35 #include <linux/pm_runtime.h>
36 #include <linux/etherdevice.h>
38 #include <linux/dca.h>
40 #include <linux/i2c.h>
44 QUEUE_MODE_STRICT_PRIORITY,
45 QUEUE_MODE_STREAM_RESERVATION,
53 char igb_driver_name[] = "igb";
54 static const char igb_driver_string[] =
55 "Intel(R) Gigabit Ethernet Network Driver";
56 static const char igb_copyright[] =
57 "Copyright (c) 2007-2014 Intel Corporation.";
59 static const struct e1000_info *igb_info_tbl[] = {
60 [board_82575] = &e1000_82575_info,
63 static const struct pci_device_id igb_pci_tbl[] = {
64 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_BACKPLANE_1GBPS) },
65 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_SGMII) },
66 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_BACKPLANE_2_5GBPS) },
67 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I211_COPPER), board_82575 },
68 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER), board_82575 },
69 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_FIBER), board_82575 },
70 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES), board_82575 },
71 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SGMII), board_82575 },
72 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER_FLASHLESS), board_82575 },
73 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES_FLASHLESS), board_82575 },
74 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_COPPER), board_82575 },
75 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_FIBER), board_82575 },
76 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SERDES), board_82575 },
77 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SGMII), board_82575 },
78 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER), board_82575 },
79 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_FIBER), board_82575 },
80 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_QUAD_FIBER), board_82575 },
81 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SERDES), board_82575 },
82 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SGMII), board_82575 },
83 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER_DUAL), board_82575 },
84 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SGMII), board_82575 },
85 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SERDES), board_82575 },
86 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_BACKPLANE), board_82575 },
87 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SFP), board_82575 },
88 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576), board_82575 },
89 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS), board_82575 },
90 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS_SERDES), board_82575 },
91 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_FIBER), board_82575 },
92 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES), board_82575 },
93 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES_QUAD), board_82575 },
94 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER_ET2), board_82575 },
95 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER), board_82575 },
96 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
97 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
98 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
99 /* required last entry */
103 MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
105 static int igb_setup_all_tx_resources(struct igb_adapter *);
106 static int igb_setup_all_rx_resources(struct igb_adapter *);
107 static void igb_free_all_tx_resources(struct igb_adapter *);
108 static void igb_free_all_rx_resources(struct igb_adapter *);
109 static void igb_setup_mrqc(struct igb_adapter *);
110 static int igb_probe(struct pci_dev *, const struct pci_device_id *);
111 static void igb_remove(struct pci_dev *pdev);
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_pci_enable_sriov(struct pci_dev *dev, int num_vfs);
179 static int igb_disable_sriov(struct pci_dev *dev);
180 static int igb_pci_disable_sriov(struct pci_dev *dev);
183 static int igb_suspend(struct device *);
184 static int igb_resume(struct device *);
185 static int igb_runtime_suspend(struct device *dev);
186 static int igb_runtime_resume(struct device *dev);
187 static int igb_runtime_idle(struct device *dev);
188 static const struct dev_pm_ops igb_pm_ops = {
189 SET_SYSTEM_SLEEP_PM_OPS(igb_suspend, igb_resume)
190 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 = 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 kfree_rcu(q_vector, rcu);
1206 q_vector = kzalloc(size, GFP_KERNEL);
1208 memset(q_vector, 0, size);
1213 /* initialize NAPI */
1214 netif_napi_add(adapter->netdev, &q_vector->napi,
1217 /* tie q_vector and adapter together */
1218 adapter->q_vector[v_idx] = q_vector;
1219 q_vector->adapter = adapter;
1221 /* initialize work limits */
1222 q_vector->tx.work_limit = adapter->tx_work_limit;
1224 /* initialize ITR configuration */
1225 q_vector->itr_register = adapter->io_addr + E1000_EITR(0);
1226 q_vector->itr_val = IGB_START_ITR;
1228 /* initialize pointer to rings */
1229 ring = q_vector->ring;
1233 /* rx or rx/tx vector */
1234 if (!adapter->rx_itr_setting || adapter->rx_itr_setting > 3)
1235 q_vector->itr_val = adapter->rx_itr_setting;
1237 /* tx only vector */
1238 if (!adapter->tx_itr_setting || adapter->tx_itr_setting > 3)
1239 q_vector->itr_val = adapter->tx_itr_setting;
1243 /* assign generic ring traits */
1244 ring->dev = &adapter->pdev->dev;
1245 ring->netdev = adapter->netdev;
1247 /* configure backlink on ring */
1248 ring->q_vector = q_vector;
1250 /* update q_vector Tx values */
1251 igb_add_ring(ring, &q_vector->tx);
1253 /* For 82575, context index must be unique per ring. */
1254 if (adapter->hw.mac.type == e1000_82575)
1255 set_bit(IGB_RING_FLAG_TX_CTX_IDX, &ring->flags);
1257 /* apply Tx specific ring traits */
1258 ring->count = adapter->tx_ring_count;
1259 ring->queue_index = txr_idx;
1261 ring->cbs_enable = false;
1262 ring->idleslope = 0;
1263 ring->sendslope = 0;
1267 u64_stats_init(&ring->tx_syncp);
1268 u64_stats_init(&ring->tx_syncp2);
1270 /* assign ring to adapter */
1271 adapter->tx_ring[txr_idx] = ring;
1273 /* push pointer to next ring */
1278 /* assign generic ring traits */
1279 ring->dev = &adapter->pdev->dev;
1280 ring->netdev = adapter->netdev;
1282 /* configure backlink on ring */
1283 ring->q_vector = q_vector;
1285 /* update q_vector Rx values */
1286 igb_add_ring(ring, &q_vector->rx);
1288 /* set flag indicating ring supports SCTP checksum offload */
1289 if (adapter->hw.mac.type >= e1000_82576)
1290 set_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags);
1292 /* On i350, i354, i210, and i211, loopback VLAN packets
1293 * have the tag byte-swapped.
1295 if (adapter->hw.mac.type >= e1000_i350)
1296 set_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &ring->flags);
1298 /* apply Rx specific ring traits */
1299 ring->count = adapter->rx_ring_count;
1300 ring->queue_index = rxr_idx;
1302 u64_stats_init(&ring->rx_syncp);
1304 /* assign ring to adapter */
1305 adapter->rx_ring[rxr_idx] = ring;
1313 * igb_alloc_q_vectors - Allocate memory for interrupt vectors
1314 * @adapter: board private structure to initialize
1316 * We allocate one q_vector per queue interrupt. If allocation fails we
1319 static int igb_alloc_q_vectors(struct igb_adapter *adapter)
1321 int q_vectors = adapter->num_q_vectors;
1322 int rxr_remaining = adapter->num_rx_queues;
1323 int txr_remaining = adapter->num_tx_queues;
1324 int rxr_idx = 0, txr_idx = 0, v_idx = 0;
1327 if (q_vectors >= (rxr_remaining + txr_remaining)) {
1328 for (; rxr_remaining; v_idx++) {
1329 err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1335 /* update counts and index */
1341 for (; v_idx < q_vectors; v_idx++) {
1342 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_idx);
1343 int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_idx);
1345 err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1346 tqpv, txr_idx, rqpv, rxr_idx);
1351 /* update counts and index */
1352 rxr_remaining -= rqpv;
1353 txr_remaining -= tqpv;
1361 adapter->num_tx_queues = 0;
1362 adapter->num_rx_queues = 0;
1363 adapter->num_q_vectors = 0;
1366 igb_free_q_vector(adapter, v_idx);
1372 * igb_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
1373 * @adapter: board private structure to initialize
1374 * @msix: boolean value of MSIX capability
1376 * This function initializes the interrupts and allocates all of the queues.
1378 static int igb_init_interrupt_scheme(struct igb_adapter *adapter, bool msix)
1380 struct pci_dev *pdev = adapter->pdev;
1383 igb_set_interrupt_capability(adapter, msix);
1385 err = igb_alloc_q_vectors(adapter);
1387 dev_err(&pdev->dev, "Unable to allocate memory for vectors\n");
1388 goto err_alloc_q_vectors;
1391 igb_cache_ring_register(adapter);
1395 err_alloc_q_vectors:
1396 igb_reset_interrupt_capability(adapter);
1401 * igb_request_irq - initialize interrupts
1402 * @adapter: board private structure to initialize
1404 * Attempts to configure interrupts using the best available
1405 * capabilities of the hardware and kernel.
1407 static int igb_request_irq(struct igb_adapter *adapter)
1409 struct net_device *netdev = adapter->netdev;
1410 struct pci_dev *pdev = adapter->pdev;
1413 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1414 err = igb_request_msix(adapter);
1417 /* fall back to MSI */
1418 igb_free_all_tx_resources(adapter);
1419 igb_free_all_rx_resources(adapter);
1421 igb_clear_interrupt_scheme(adapter);
1422 err = igb_init_interrupt_scheme(adapter, false);
1426 igb_setup_all_tx_resources(adapter);
1427 igb_setup_all_rx_resources(adapter);
1428 igb_configure(adapter);
1431 igb_assign_vector(adapter->q_vector[0], 0);
1433 if (adapter->flags & IGB_FLAG_HAS_MSI) {
1434 err = request_irq(pdev->irq, igb_intr_msi, 0,
1435 netdev->name, adapter);
1439 /* fall back to legacy interrupts */
1440 igb_reset_interrupt_capability(adapter);
1441 adapter->flags &= ~IGB_FLAG_HAS_MSI;
1444 err = request_irq(pdev->irq, igb_intr, IRQF_SHARED,
1445 netdev->name, adapter);
1448 dev_err(&pdev->dev, "Error %d getting interrupt\n",
1455 static void igb_free_irq(struct igb_adapter *adapter)
1457 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1460 free_irq(adapter->msix_entries[vector++].vector, adapter);
1462 for (i = 0; i < adapter->num_q_vectors; i++)
1463 free_irq(adapter->msix_entries[vector++].vector,
1464 adapter->q_vector[i]);
1466 free_irq(adapter->pdev->irq, adapter);
1471 * igb_irq_disable - Mask off interrupt generation on the NIC
1472 * @adapter: board private structure
1474 static void igb_irq_disable(struct igb_adapter *adapter)
1476 struct e1000_hw *hw = &adapter->hw;
1478 /* we need to be careful when disabling interrupts. The VFs are also
1479 * mapped into these registers and so clearing the bits can cause
1480 * issues on the VF drivers so we only need to clear what we set
1482 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1483 u32 regval = rd32(E1000_EIAM);
1485 wr32(E1000_EIAM, regval & ~adapter->eims_enable_mask);
1486 wr32(E1000_EIMC, adapter->eims_enable_mask);
1487 regval = rd32(E1000_EIAC);
1488 wr32(E1000_EIAC, regval & ~adapter->eims_enable_mask);
1492 wr32(E1000_IMC, ~0);
1494 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1497 for (i = 0; i < adapter->num_q_vectors; i++)
1498 synchronize_irq(adapter->msix_entries[i].vector);
1500 synchronize_irq(adapter->pdev->irq);
1505 * igb_irq_enable - Enable default interrupt generation settings
1506 * @adapter: board private structure
1508 static void igb_irq_enable(struct igb_adapter *adapter)
1510 struct e1000_hw *hw = &adapter->hw;
1512 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1513 u32 ims = E1000_IMS_LSC | E1000_IMS_DOUTSYNC | E1000_IMS_DRSTA;
1514 u32 regval = rd32(E1000_EIAC);
1516 wr32(E1000_EIAC, regval | adapter->eims_enable_mask);
1517 regval = rd32(E1000_EIAM);
1518 wr32(E1000_EIAM, regval | adapter->eims_enable_mask);
1519 wr32(E1000_EIMS, adapter->eims_enable_mask);
1520 if (adapter->vfs_allocated_count) {
1521 wr32(E1000_MBVFIMR, 0xFF);
1522 ims |= E1000_IMS_VMMB;
1524 wr32(E1000_IMS, ims);
1526 wr32(E1000_IMS, IMS_ENABLE_MASK |
1528 wr32(E1000_IAM, IMS_ENABLE_MASK |
1533 static void igb_update_mng_vlan(struct igb_adapter *adapter)
1535 struct e1000_hw *hw = &adapter->hw;
1536 u16 pf_id = adapter->vfs_allocated_count;
1537 u16 vid = adapter->hw.mng_cookie.vlan_id;
1538 u16 old_vid = adapter->mng_vlan_id;
1540 if (hw->mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1541 /* add VID to filter table */
1542 igb_vfta_set(hw, vid, pf_id, true, true);
1543 adapter->mng_vlan_id = vid;
1545 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1548 if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
1550 !test_bit(old_vid, adapter->active_vlans)) {
1551 /* remove VID from filter table */
1552 igb_vfta_set(hw, vid, pf_id, false, true);
1557 * igb_release_hw_control - release control of the h/w to f/w
1558 * @adapter: address of board private structure
1560 * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
1561 * For ASF and Pass Through versions of f/w this means that the
1562 * driver is no longer loaded.
1564 static void igb_release_hw_control(struct igb_adapter *adapter)
1566 struct e1000_hw *hw = &adapter->hw;
1569 /* Let firmware take over control of h/w */
1570 ctrl_ext = rd32(E1000_CTRL_EXT);
1571 wr32(E1000_CTRL_EXT,
1572 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1576 * igb_get_hw_control - get control of the h/w from f/w
1577 * @adapter: address of board private structure
1579 * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
1580 * For ASF and Pass Through versions of f/w this means that
1581 * the driver is loaded.
1583 static void igb_get_hw_control(struct igb_adapter *adapter)
1585 struct e1000_hw *hw = &adapter->hw;
1588 /* Let firmware know the driver has taken over */
1589 ctrl_ext = rd32(E1000_CTRL_EXT);
1590 wr32(E1000_CTRL_EXT,
1591 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1594 static void enable_fqtss(struct igb_adapter *adapter, bool enable)
1596 struct net_device *netdev = adapter->netdev;
1597 struct e1000_hw *hw = &adapter->hw;
1599 WARN_ON(hw->mac.type != e1000_i210);
1602 adapter->flags |= IGB_FLAG_FQTSS;
1604 adapter->flags &= ~IGB_FLAG_FQTSS;
1606 if (netif_running(netdev))
1607 schedule_work(&adapter->reset_task);
1610 static bool is_fqtss_enabled(struct igb_adapter *adapter)
1612 return (adapter->flags & IGB_FLAG_FQTSS) ? true : false;
1615 static void set_tx_desc_fetch_prio(struct e1000_hw *hw, int queue,
1616 enum tx_queue_prio prio)
1620 WARN_ON(hw->mac.type != e1000_i210);
1621 WARN_ON(queue < 0 || queue > 4);
1623 val = rd32(E1000_I210_TXDCTL(queue));
1625 if (prio == TX_QUEUE_PRIO_HIGH)
1626 val |= E1000_TXDCTL_PRIORITY;
1628 val &= ~E1000_TXDCTL_PRIORITY;
1630 wr32(E1000_I210_TXDCTL(queue), val);
1633 static void set_queue_mode(struct e1000_hw *hw, int queue, enum queue_mode mode)
1637 WARN_ON(hw->mac.type != e1000_i210);
1638 WARN_ON(queue < 0 || queue > 1);
1640 val = rd32(E1000_I210_TQAVCC(queue));
1642 if (mode == QUEUE_MODE_STREAM_RESERVATION)
1643 val |= E1000_TQAVCC_QUEUEMODE;
1645 val &= ~E1000_TQAVCC_QUEUEMODE;
1647 wr32(E1000_I210_TQAVCC(queue), val);
1650 static bool is_any_cbs_enabled(struct igb_adapter *adapter)
1654 for (i = 0; i < adapter->num_tx_queues; i++) {
1655 if (adapter->tx_ring[i]->cbs_enable)
1662 static bool is_any_txtime_enabled(struct igb_adapter *adapter)
1666 for (i = 0; i < adapter->num_tx_queues; i++) {
1667 if (adapter->tx_ring[i]->launchtime_enable)
1675 * igb_config_tx_modes - Configure "Qav Tx mode" features on igb
1676 * @adapter: pointer to adapter struct
1677 * @queue: queue number
1679 * Configure CBS and Launchtime for a given hardware queue.
1680 * Parameters are retrieved from the correct Tx ring, so
1681 * igb_save_cbs_params() and igb_save_txtime_params() should be used
1682 * for setting those correctly prior to this function being called.
1684 static void igb_config_tx_modes(struct igb_adapter *adapter, int queue)
1686 struct net_device *netdev = adapter->netdev;
1687 struct e1000_hw *hw = &adapter->hw;
1688 struct igb_ring *ring;
1689 u32 tqavcc, tqavctrl;
1692 WARN_ON(hw->mac.type != e1000_i210);
1693 WARN_ON(queue < 0 || queue > 1);
1694 ring = adapter->tx_ring[queue];
1696 /* If any of the Qav features is enabled, configure queues as SR and
1697 * with HIGH PRIO. If none is, then configure them with LOW PRIO and
1700 if (ring->cbs_enable || ring->launchtime_enable) {
1701 set_tx_desc_fetch_prio(hw, queue, TX_QUEUE_PRIO_HIGH);
1702 set_queue_mode(hw, queue, QUEUE_MODE_STREAM_RESERVATION);
1704 set_tx_desc_fetch_prio(hw, queue, TX_QUEUE_PRIO_LOW);
1705 set_queue_mode(hw, queue, QUEUE_MODE_STRICT_PRIORITY);
1708 /* If CBS is enabled, set DataTranARB and config its parameters. */
1709 if (ring->cbs_enable || queue == 0) {
1710 /* i210 does not allow the queue 0 to be in the Strict
1711 * Priority mode while the Qav mode is enabled, so,
1712 * instead of disabling strict priority mode, we give
1713 * queue 0 the maximum of credits possible.
1715 * See section 8.12.19 of the i210 datasheet, "Note:
1716 * Queue0 QueueMode must be set to 1b when
1717 * TransmitMode is set to Qav."
1719 if (queue == 0 && !ring->cbs_enable) {
1720 /* max "linkspeed" idleslope in kbps */
1721 ring->idleslope = 1000000;
1722 ring->hicredit = ETH_FRAME_LEN;
1725 /* Always set data transfer arbitration to credit-based
1726 * shaper algorithm on TQAVCTRL if CBS is enabled for any of
1729 tqavctrl = rd32(E1000_I210_TQAVCTRL);
1730 tqavctrl |= E1000_TQAVCTRL_DATATRANARB;
1731 wr32(E1000_I210_TQAVCTRL, tqavctrl);
1733 /* According to i210 datasheet section 7.2.7.7, we should set
1734 * the 'idleSlope' field from TQAVCC register following the
1737 * For 100 Mbps link speed:
1739 * value = BW * 0x7735 * 0.2 (E1)
1741 * For 1000Mbps link speed:
1743 * value = BW * 0x7735 * 2 (E2)
1745 * E1 and E2 can be merged into one equation as shown below.
1746 * Note that 'link-speed' is in Mbps.
1748 * value = BW * 0x7735 * 2 * link-speed
1749 * -------------- (E3)
1752 * 'BW' is the percentage bandwidth out of full link speed
1753 * which can be found with the following equation. Note that
1754 * idleSlope here is the parameter from this function which
1758 * ----------------- (E4)
1761 * That said, we can come up with a generic equation to
1762 * calculate the value we should set it TQAVCC register by
1763 * replacing 'BW' in E3 by E4. The resulting equation is:
1765 * value = idleSlope * 0x7735 * 2 * link-speed
1766 * ----------------- -------------- (E5)
1767 * link-speed * 1000 1000
1769 * 'link-speed' is present in both sides of the fraction so
1770 * it is canceled out. The final equation is the following:
1772 * value = idleSlope * 61034
1773 * ----------------- (E6)
1776 * NOTE: For i210, given the above, we can see that idleslope
1777 * is represented in 16.38431 kbps units by the value at
1778 * the TQAVCC register (1Gbps / 61034), which reduces
1779 * the granularity for idleslope increments.
1780 * For instance, if you want to configure a 2576kbps
1781 * idleslope, the value to be written on the register
1782 * would have to be 157.23. If rounded down, you end
1783 * up with less bandwidth available than originally
1784 * required (~2572 kbps). If rounded up, you end up
1785 * with a higher bandwidth (~2589 kbps). Below the
1786 * approach we take is to always round up the
1787 * calculated value, so the resulting bandwidth might
1788 * be slightly higher for some configurations.
1790 value = DIV_ROUND_UP_ULL(ring->idleslope * 61034ULL, 1000000);
1792 tqavcc = rd32(E1000_I210_TQAVCC(queue));
1793 tqavcc &= ~E1000_TQAVCC_IDLESLOPE_MASK;
1795 wr32(E1000_I210_TQAVCC(queue), tqavcc);
1797 wr32(E1000_I210_TQAVHC(queue),
1798 0x80000000 + ring->hicredit * 0x7735);
1801 /* Set idleSlope to zero. */
1802 tqavcc = rd32(E1000_I210_TQAVCC(queue));
1803 tqavcc &= ~E1000_TQAVCC_IDLESLOPE_MASK;
1804 wr32(E1000_I210_TQAVCC(queue), tqavcc);
1806 /* Set hiCredit to zero. */
1807 wr32(E1000_I210_TQAVHC(queue), 0);
1809 /* If CBS is not enabled for any queues anymore, then return to
1810 * the default state of Data Transmission Arbitration on
1813 if (!is_any_cbs_enabled(adapter)) {
1814 tqavctrl = rd32(E1000_I210_TQAVCTRL);
1815 tqavctrl &= ~E1000_TQAVCTRL_DATATRANARB;
1816 wr32(E1000_I210_TQAVCTRL, tqavctrl);
1820 /* If LaunchTime is enabled, set DataTranTIM. */
1821 if (ring->launchtime_enable) {
1822 /* Always set DataTranTIM on TQAVCTRL if LaunchTime is enabled
1823 * for any of the SR queues, and configure fetchtime delta.
1825 * - LaunchTime will be enabled for all SR queues.
1826 * - A fixed offset can be added relative to the launch
1827 * time of all packets if configured at reg LAUNCH_OS0.
1828 * We are keeping it as 0 for now (default value).
1830 tqavctrl = rd32(E1000_I210_TQAVCTRL);
1831 tqavctrl |= E1000_TQAVCTRL_DATATRANTIM |
1832 E1000_TQAVCTRL_FETCHTIME_DELTA;
1833 wr32(E1000_I210_TQAVCTRL, tqavctrl);
1835 /* If Launchtime is not enabled for any SR queues anymore,
1836 * then clear DataTranTIM on TQAVCTRL and clear fetchtime delta,
1837 * effectively disabling Launchtime.
1839 if (!is_any_txtime_enabled(adapter)) {
1840 tqavctrl = rd32(E1000_I210_TQAVCTRL);
1841 tqavctrl &= ~E1000_TQAVCTRL_DATATRANTIM;
1842 tqavctrl &= ~E1000_TQAVCTRL_FETCHTIME_DELTA;
1843 wr32(E1000_I210_TQAVCTRL, tqavctrl);
1847 /* XXX: In i210 controller the sendSlope and loCredit parameters from
1848 * CBS are not configurable by software so we don't do any 'controller
1849 * configuration' in respect to these parameters.
1852 netdev_dbg(netdev, "Qav Tx mode: cbs %s, launchtime %s, queue %d idleslope %d sendslope %d hiCredit %d locredit %d\n",
1853 ring->cbs_enable ? "enabled" : "disabled",
1854 ring->launchtime_enable ? "enabled" : "disabled",
1856 ring->idleslope, ring->sendslope,
1857 ring->hicredit, ring->locredit);
1860 static int igb_save_txtime_params(struct igb_adapter *adapter, int queue,
1863 struct igb_ring *ring;
1865 if (queue < 0 || queue > adapter->num_tx_queues)
1868 ring = adapter->tx_ring[queue];
1869 ring->launchtime_enable = enable;
1874 static int igb_save_cbs_params(struct igb_adapter *adapter, int queue,
1875 bool enable, int idleslope, int sendslope,
1876 int hicredit, int locredit)
1878 struct igb_ring *ring;
1880 if (queue < 0 || queue > adapter->num_tx_queues)
1883 ring = adapter->tx_ring[queue];
1885 ring->cbs_enable = enable;
1886 ring->idleslope = idleslope;
1887 ring->sendslope = sendslope;
1888 ring->hicredit = hicredit;
1889 ring->locredit = locredit;
1895 * igb_setup_tx_mode - Switch to/from Qav Tx mode when applicable
1896 * @adapter: pointer to adapter struct
1898 * Configure TQAVCTRL register switching the controller's Tx mode
1899 * if FQTSS mode is enabled or disabled. Additionally, will issue
1900 * a call to igb_config_tx_modes() per queue so any previously saved
1901 * Tx parameters are applied.
1903 static void igb_setup_tx_mode(struct igb_adapter *adapter)
1905 struct net_device *netdev = adapter->netdev;
1906 struct e1000_hw *hw = &adapter->hw;
1909 /* Only i210 controller supports changing the transmission mode. */
1910 if (hw->mac.type != e1000_i210)
1913 if (is_fqtss_enabled(adapter)) {
1916 /* Configure TQAVCTRL register: set transmit mode to 'Qav',
1917 * set data fetch arbitration to 'round robin', set SP_WAIT_SR
1918 * so SP queues wait for SR ones.
1920 val = rd32(E1000_I210_TQAVCTRL);
1921 val |= E1000_TQAVCTRL_XMIT_MODE | E1000_TQAVCTRL_SP_WAIT_SR;
1922 val &= ~E1000_TQAVCTRL_DATAFETCHARB;
1923 wr32(E1000_I210_TQAVCTRL, val);
1925 /* Configure Tx and Rx packet buffers sizes as described in
1926 * i210 datasheet section 7.2.7.7.
1928 val = rd32(E1000_TXPBS);
1929 val &= ~I210_TXPBSIZE_MASK;
1930 val |= I210_TXPBSIZE_PB0_6KB | I210_TXPBSIZE_PB1_6KB |
1931 I210_TXPBSIZE_PB2_6KB | I210_TXPBSIZE_PB3_6KB;
1932 wr32(E1000_TXPBS, val);
1934 val = rd32(E1000_RXPBS);
1935 val &= ~I210_RXPBSIZE_MASK;
1936 val |= I210_RXPBSIZE_PB_30KB;
1937 wr32(E1000_RXPBS, val);
1939 /* Section 8.12.9 states that MAX_TPKT_SIZE from DTXMXPKTSZ
1940 * register should not exceed the buffer size programmed in
1941 * TXPBS. The smallest buffer size programmed in TXPBS is 4kB
1942 * so according to the datasheet we should set MAX_TPKT_SIZE to
1945 * However, when we do so, no frame from queue 2 and 3 are
1946 * transmitted. It seems the MAX_TPKT_SIZE should not be great
1947 * or _equal_ to the buffer size programmed in TXPBS. For this
1948 * reason, we set MAX_ TPKT_SIZE to (4kB - 1) / 64.
1950 val = (4096 - 1) / 64;
1951 wr32(E1000_I210_DTXMXPKTSZ, val);
1953 /* Since FQTSS mode is enabled, apply any CBS configuration
1954 * previously set. If no previous CBS configuration has been
1955 * done, then the initial configuration is applied, which means
1958 max_queue = (adapter->num_tx_queues < I210_SR_QUEUES_NUM) ?
1959 adapter->num_tx_queues : I210_SR_QUEUES_NUM;
1961 for (i = 0; i < max_queue; i++) {
1962 igb_config_tx_modes(adapter, i);
1965 wr32(E1000_RXPBS, I210_RXPBSIZE_DEFAULT);
1966 wr32(E1000_TXPBS, I210_TXPBSIZE_DEFAULT);
1967 wr32(E1000_I210_DTXMXPKTSZ, I210_DTXMXPKTSZ_DEFAULT);
1969 val = rd32(E1000_I210_TQAVCTRL);
1970 /* According to Section 8.12.21, the other flags we've set when
1971 * enabling FQTSS are not relevant when disabling FQTSS so we
1972 * don't set they here.
1974 val &= ~E1000_TQAVCTRL_XMIT_MODE;
1975 wr32(E1000_I210_TQAVCTRL, val);
1978 netdev_dbg(netdev, "FQTSS %s\n", (is_fqtss_enabled(adapter)) ?
1979 "enabled" : "disabled");
1983 * igb_configure - configure the hardware for RX and TX
1984 * @adapter: private board structure
1986 static void igb_configure(struct igb_adapter *adapter)
1988 struct net_device *netdev = adapter->netdev;
1991 igb_get_hw_control(adapter);
1992 igb_set_rx_mode(netdev);
1993 igb_setup_tx_mode(adapter);
1995 igb_restore_vlan(adapter);
1997 igb_setup_tctl(adapter);
1998 igb_setup_mrqc(adapter);
1999 igb_setup_rctl(adapter);
2001 igb_nfc_filter_restore(adapter);
2002 igb_configure_tx(adapter);
2003 igb_configure_rx(adapter);
2005 igb_rx_fifo_flush_82575(&adapter->hw);
2007 /* call igb_desc_unused which always leaves
2008 * at least 1 descriptor unused to make sure
2009 * next_to_use != next_to_clean
2011 for (i = 0; i < adapter->num_rx_queues; i++) {
2012 struct igb_ring *ring = adapter->rx_ring[i];
2013 igb_alloc_rx_buffers(ring, igb_desc_unused(ring));
2018 * igb_power_up_link - Power up the phy/serdes link
2019 * @adapter: address of board private structure
2021 void igb_power_up_link(struct igb_adapter *adapter)
2023 igb_reset_phy(&adapter->hw);
2025 if (adapter->hw.phy.media_type == e1000_media_type_copper)
2026 igb_power_up_phy_copper(&adapter->hw);
2028 igb_power_up_serdes_link_82575(&adapter->hw);
2030 igb_setup_link(&adapter->hw);
2034 * igb_power_down_link - Power down the phy/serdes link
2035 * @adapter: address of board private structure
2037 static void igb_power_down_link(struct igb_adapter *adapter)
2039 if (adapter->hw.phy.media_type == e1000_media_type_copper)
2040 igb_power_down_phy_copper_82575(&adapter->hw);
2042 igb_shutdown_serdes_link_82575(&adapter->hw);
2046 * igb_check_swap_media - Detect and switch function for Media Auto Sense
2047 * @adapter: address of the board private structure
2049 static void igb_check_swap_media(struct igb_adapter *adapter)
2051 struct e1000_hw *hw = &adapter->hw;
2052 u32 ctrl_ext, connsw;
2053 bool swap_now = false;
2055 ctrl_ext = rd32(E1000_CTRL_EXT);
2056 connsw = rd32(E1000_CONNSW);
2058 /* need to live swap if current media is copper and we have fiber/serdes
2062 if ((hw->phy.media_type == e1000_media_type_copper) &&
2063 (!(connsw & E1000_CONNSW_AUTOSENSE_EN))) {
2065 } else if ((hw->phy.media_type != e1000_media_type_copper) &&
2066 !(connsw & E1000_CONNSW_SERDESD)) {
2067 /* copper signal takes time to appear */
2068 if (adapter->copper_tries < 4) {
2069 adapter->copper_tries++;
2070 connsw |= E1000_CONNSW_AUTOSENSE_CONF;
2071 wr32(E1000_CONNSW, connsw);
2074 adapter->copper_tries = 0;
2075 if ((connsw & E1000_CONNSW_PHYSD) &&
2076 (!(connsw & E1000_CONNSW_PHY_PDN))) {
2078 connsw &= ~E1000_CONNSW_AUTOSENSE_CONF;
2079 wr32(E1000_CONNSW, connsw);
2087 switch (hw->phy.media_type) {
2088 case e1000_media_type_copper:
2089 netdev_info(adapter->netdev,
2090 "MAS: changing media to fiber/serdes\n");
2092 E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
2093 adapter->flags |= IGB_FLAG_MEDIA_RESET;
2094 adapter->copper_tries = 0;
2096 case e1000_media_type_internal_serdes:
2097 case e1000_media_type_fiber:
2098 netdev_info(adapter->netdev,
2099 "MAS: changing media to copper\n");
2101 ~E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
2102 adapter->flags |= IGB_FLAG_MEDIA_RESET;
2105 /* shouldn't get here during regular operation */
2106 netdev_err(adapter->netdev,
2107 "AMS: Invalid media type found, returning\n");
2110 wr32(E1000_CTRL_EXT, ctrl_ext);
2114 * igb_up - Open the interface and prepare it to handle traffic
2115 * @adapter: board private structure
2117 int igb_up(struct igb_adapter *adapter)
2119 struct e1000_hw *hw = &adapter->hw;
2122 /* hardware has been reset, we need to reload some things */
2123 igb_configure(adapter);
2125 clear_bit(__IGB_DOWN, &adapter->state);
2127 for (i = 0; i < adapter->num_q_vectors; i++)
2128 napi_enable(&(adapter->q_vector[i]->napi));
2130 if (adapter->flags & IGB_FLAG_HAS_MSIX)
2131 igb_configure_msix(adapter);
2133 igb_assign_vector(adapter->q_vector[0], 0);
2135 /* Clear any pending interrupts. */
2138 igb_irq_enable(adapter);
2140 /* notify VFs that reset has been completed */
2141 if (adapter->vfs_allocated_count) {
2142 u32 reg_data = rd32(E1000_CTRL_EXT);
2144 reg_data |= E1000_CTRL_EXT_PFRSTD;
2145 wr32(E1000_CTRL_EXT, reg_data);
2148 netif_tx_start_all_queues(adapter->netdev);
2150 /* start the watchdog. */
2151 hw->mac.get_link_status = 1;
2152 schedule_work(&adapter->watchdog_task);
2154 if ((adapter->flags & IGB_FLAG_EEE) &&
2155 (!hw->dev_spec._82575.eee_disable))
2156 adapter->eee_advert = MDIO_EEE_100TX | MDIO_EEE_1000T;
2161 void igb_down(struct igb_adapter *adapter)
2163 struct net_device *netdev = adapter->netdev;
2164 struct e1000_hw *hw = &adapter->hw;
2168 /* signal that we're down so the interrupt handler does not
2169 * reschedule our watchdog timer
2171 set_bit(__IGB_DOWN, &adapter->state);
2173 /* disable receives in the hardware */
2174 rctl = rd32(E1000_RCTL);
2175 wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
2176 /* flush and sleep below */
2178 igb_nfc_filter_exit(adapter);
2180 netif_carrier_off(netdev);
2181 netif_tx_stop_all_queues(netdev);
2183 /* disable transmits in the hardware */
2184 tctl = rd32(E1000_TCTL);
2185 tctl &= ~E1000_TCTL_EN;
2186 wr32(E1000_TCTL, tctl);
2187 /* flush both disables and wait for them to finish */
2189 usleep_range(10000, 11000);
2191 igb_irq_disable(adapter);
2193 adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
2195 for (i = 0; i < adapter->num_q_vectors; i++) {
2196 if (adapter->q_vector[i]) {
2197 napi_synchronize(&adapter->q_vector[i]->napi);
2198 napi_disable(&adapter->q_vector[i]->napi);
2202 del_timer_sync(&adapter->watchdog_timer);
2203 del_timer_sync(&adapter->phy_info_timer);
2205 /* record the stats before reset*/
2206 spin_lock(&adapter->stats64_lock);
2207 igb_update_stats(adapter);
2208 spin_unlock(&adapter->stats64_lock);
2210 adapter->link_speed = 0;
2211 adapter->link_duplex = 0;
2213 if (!pci_channel_offline(adapter->pdev))
2216 /* clear VLAN promisc flag so VFTA will be updated if necessary */
2217 adapter->flags &= ~IGB_FLAG_VLAN_PROMISC;
2219 igb_clean_all_tx_rings(adapter);
2220 igb_clean_all_rx_rings(adapter);
2221 #ifdef CONFIG_IGB_DCA
2223 /* since we reset the hardware DCA settings were cleared */
2224 igb_setup_dca(adapter);
2228 void igb_reinit_locked(struct igb_adapter *adapter)
2230 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
2231 usleep_range(1000, 2000);
2234 clear_bit(__IGB_RESETTING, &adapter->state);
2237 /** igb_enable_mas - Media Autosense re-enable after swap
2239 * @adapter: adapter struct
2241 static void igb_enable_mas(struct igb_adapter *adapter)
2243 struct e1000_hw *hw = &adapter->hw;
2244 u32 connsw = rd32(E1000_CONNSW);
2246 /* configure for SerDes media detect */
2247 if ((hw->phy.media_type == e1000_media_type_copper) &&
2248 (!(connsw & E1000_CONNSW_SERDESD))) {
2249 connsw |= E1000_CONNSW_ENRGSRC;
2250 connsw |= E1000_CONNSW_AUTOSENSE_EN;
2251 wr32(E1000_CONNSW, connsw);
2256 void igb_reset(struct igb_adapter *adapter)
2258 struct pci_dev *pdev = adapter->pdev;
2259 struct e1000_hw *hw = &adapter->hw;
2260 struct e1000_mac_info *mac = &hw->mac;
2261 struct e1000_fc_info *fc = &hw->fc;
2264 /* Repartition Pba for greater than 9k mtu
2265 * To take effect CTRL.RST is required.
2267 switch (mac->type) {
2271 pba = rd32(E1000_RXPBS);
2272 pba = igb_rxpbs_adjust_82580(pba);
2275 pba = rd32(E1000_RXPBS);
2276 pba &= E1000_RXPBS_SIZE_MASK_82576;
2282 pba = E1000_PBA_34K;
2286 if (mac->type == e1000_82575) {
2287 u32 min_rx_space, min_tx_space, needed_tx_space;
2289 /* write Rx PBA so that hardware can report correct Tx PBA */
2290 wr32(E1000_PBA, pba);
2292 /* To maintain wire speed transmits, the Tx FIFO should be
2293 * large enough to accommodate two full transmit packets,
2294 * rounded up to the next 1KB and expressed in KB. Likewise,
2295 * the Rx FIFO should be large enough to accommodate at least
2296 * one full receive packet and is similarly rounded up and
2299 min_rx_space = DIV_ROUND_UP(MAX_JUMBO_FRAME_SIZE, 1024);
2301 /* The Tx FIFO also stores 16 bytes of information about the Tx
2302 * but don't include Ethernet FCS because hardware appends it.
2303 * We only need to round down to the nearest 512 byte block
2304 * count since the value we care about is 2 frames, not 1.
2306 min_tx_space = adapter->max_frame_size;
2307 min_tx_space += sizeof(union e1000_adv_tx_desc) - ETH_FCS_LEN;
2308 min_tx_space = DIV_ROUND_UP(min_tx_space, 512);
2310 /* upper 16 bits has Tx packet buffer allocation size in KB */
2311 needed_tx_space = min_tx_space - (rd32(E1000_PBA) >> 16);
2313 /* If current Tx allocation is less than the min Tx FIFO size,
2314 * and the min Tx FIFO size is less than the current Rx FIFO
2315 * allocation, take space away from current Rx allocation.
2317 if (needed_tx_space < pba) {
2318 pba -= needed_tx_space;
2320 /* if short on Rx space, Rx wins and must trump Tx
2323 if (pba < min_rx_space)
2327 /* adjust PBA for jumbo frames */
2328 wr32(E1000_PBA, pba);
2331 /* flow control settings
2332 * The high water mark must be low enough to fit one full frame
2333 * after transmitting the pause frame. As such we must have enough
2334 * space to allow for us to complete our current transmit and then
2335 * receive the frame that is in progress from the link partner.
2337 * - the full Rx FIFO size minus one full Tx plus one full Rx frame
2339 hwm = (pba << 10) - (adapter->max_frame_size + MAX_JUMBO_FRAME_SIZE);
2341 fc->high_water = hwm & 0xFFFFFFF0; /* 16-byte granularity */
2342 fc->low_water = fc->high_water - 16;
2343 fc->pause_time = 0xFFFF;
2345 fc->current_mode = fc->requested_mode;
2347 /* disable receive for all VFs and wait one second */
2348 if (adapter->vfs_allocated_count) {
2351 for (i = 0 ; i < adapter->vfs_allocated_count; i++)
2352 adapter->vf_data[i].flags &= IGB_VF_FLAG_PF_SET_MAC;
2354 /* ping all the active vfs to let them know we are going down */
2355 igb_ping_all_vfs(adapter);
2357 /* disable transmits and receives */
2358 wr32(E1000_VFRE, 0);
2359 wr32(E1000_VFTE, 0);
2362 /* Allow time for pending master requests to run */
2363 hw->mac.ops.reset_hw(hw);
2366 if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
2367 /* need to resetup here after media swap */
2368 adapter->ei.get_invariants(hw);
2369 adapter->flags &= ~IGB_FLAG_MEDIA_RESET;
2371 if ((mac->type == e1000_82575 || mac->type == e1000_i350) &&
2372 (adapter->flags & IGB_FLAG_MAS_ENABLE)) {
2373 igb_enable_mas(adapter);
2375 if (hw->mac.ops.init_hw(hw))
2376 dev_err(&pdev->dev, "Hardware Error\n");
2378 /* RAR registers were cleared during init_hw, clear mac table */
2379 igb_flush_mac_table(adapter);
2380 __dev_uc_unsync(adapter->netdev, NULL);
2382 /* Recover default RAR entry */
2383 igb_set_default_mac_filter(adapter);
2385 /* Flow control settings reset on hardware reset, so guarantee flow
2386 * control is off when forcing speed.
2388 if (!hw->mac.autoneg)
2389 igb_force_mac_fc(hw);
2391 igb_init_dmac(adapter, pba);
2392 #ifdef CONFIG_IGB_HWMON
2393 /* Re-initialize the thermal sensor on i350 devices. */
2394 if (!test_bit(__IGB_DOWN, &adapter->state)) {
2395 if (mac->type == e1000_i350 && hw->bus.func == 0) {
2396 /* If present, re-initialize the external thermal sensor
2400 mac->ops.init_thermal_sensor_thresh(hw);
2404 /* Re-establish EEE setting */
2405 if (hw->phy.media_type == e1000_media_type_copper) {
2406 switch (mac->type) {
2410 igb_set_eee_i350(hw, true, true);
2413 igb_set_eee_i354(hw, true, true);
2419 if (!netif_running(adapter->netdev))
2420 igb_power_down_link(adapter);
2422 igb_update_mng_vlan(adapter);
2424 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2425 wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
2427 /* Re-enable PTP, where applicable. */
2428 if (adapter->ptp_flags & IGB_PTP_ENABLED)
2429 igb_ptp_reset(adapter);
2431 igb_get_phy_info(hw);
2434 static netdev_features_t igb_fix_features(struct net_device *netdev,
2435 netdev_features_t features)
2437 /* Since there is no support for separate Rx/Tx vlan accel
2438 * enable/disable make sure Tx flag is always in same state as Rx.
2440 if (features & NETIF_F_HW_VLAN_CTAG_RX)
2441 features |= NETIF_F_HW_VLAN_CTAG_TX;
2443 features &= ~NETIF_F_HW_VLAN_CTAG_TX;
2448 static int igb_set_features(struct net_device *netdev,
2449 netdev_features_t features)
2451 netdev_features_t changed = netdev->features ^ features;
2452 struct igb_adapter *adapter = netdev_priv(netdev);
2454 if (changed & NETIF_F_HW_VLAN_CTAG_RX)
2455 igb_vlan_mode(netdev, features);
2457 if (!(changed & (NETIF_F_RXALL | NETIF_F_NTUPLE)))
2460 if (!(features & NETIF_F_NTUPLE)) {
2461 struct hlist_node *node2;
2462 struct igb_nfc_filter *rule;
2464 spin_lock(&adapter->nfc_lock);
2465 hlist_for_each_entry_safe(rule, node2,
2466 &adapter->nfc_filter_list, nfc_node) {
2467 igb_erase_filter(adapter, rule);
2468 hlist_del(&rule->nfc_node);
2471 spin_unlock(&adapter->nfc_lock);
2472 adapter->nfc_filter_count = 0;
2475 netdev->features = features;
2477 if (netif_running(netdev))
2478 igb_reinit_locked(adapter);
2485 static int igb_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
2486 struct net_device *dev,
2487 const unsigned char *addr, u16 vid,
2489 struct netlink_ext_ack *extack)
2491 /* guarantee we can provide a unique filter for the unicast address */
2492 if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
2493 struct igb_adapter *adapter = netdev_priv(dev);
2494 int vfn = adapter->vfs_allocated_count;
2496 if (netdev_uc_count(dev) >= igb_available_rars(adapter, vfn))
2500 return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags);
2503 #define IGB_MAX_MAC_HDR_LEN 127
2504 #define IGB_MAX_NETWORK_HDR_LEN 511
2506 static netdev_features_t
2507 igb_features_check(struct sk_buff *skb, struct net_device *dev,
2508 netdev_features_t features)
2510 unsigned int network_hdr_len, mac_hdr_len;
2512 /* Make certain the headers can be described by a context descriptor */
2513 mac_hdr_len = skb_network_header(skb) - skb->data;
2514 if (unlikely(mac_hdr_len > IGB_MAX_MAC_HDR_LEN))
2515 return features & ~(NETIF_F_HW_CSUM |
2517 NETIF_F_GSO_UDP_L4 |
2518 NETIF_F_HW_VLAN_CTAG_TX |
2522 network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
2523 if (unlikely(network_hdr_len > IGB_MAX_NETWORK_HDR_LEN))
2524 return features & ~(NETIF_F_HW_CSUM |
2526 NETIF_F_GSO_UDP_L4 |
2530 /* We can only support IPV4 TSO in tunnels if we can mangle the
2531 * inner IP ID field, so strip TSO if MANGLEID is not supported.
2533 if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
2534 features &= ~NETIF_F_TSO;
2539 static void igb_offload_apply(struct igb_adapter *adapter, s32 queue)
2541 if (!is_fqtss_enabled(adapter)) {
2542 enable_fqtss(adapter, true);
2546 igb_config_tx_modes(adapter, queue);
2548 if (!is_any_cbs_enabled(adapter) && !is_any_txtime_enabled(adapter))
2549 enable_fqtss(adapter, false);
2552 static int igb_offload_cbs(struct igb_adapter *adapter,
2553 struct tc_cbs_qopt_offload *qopt)
2555 struct e1000_hw *hw = &adapter->hw;
2558 /* CBS offloading is only supported by i210 controller. */
2559 if (hw->mac.type != e1000_i210)
2562 /* CBS offloading is only supported by queue 0 and queue 1. */
2563 if (qopt->queue < 0 || qopt->queue > 1)
2566 err = igb_save_cbs_params(adapter, qopt->queue, qopt->enable,
2567 qopt->idleslope, qopt->sendslope,
2568 qopt->hicredit, qopt->locredit);
2572 igb_offload_apply(adapter, qopt->queue);
2577 #define ETHER_TYPE_FULL_MASK ((__force __be16)~0)
2578 #define VLAN_PRIO_FULL_MASK (0x07)
2580 static int igb_parse_cls_flower(struct igb_adapter *adapter,
2581 struct flow_cls_offload *f,
2583 struct igb_nfc_filter *input)
2585 struct flow_rule *rule = flow_cls_offload_flow_rule(f);
2586 struct flow_dissector *dissector = rule->match.dissector;
2587 struct netlink_ext_ack *extack = f->common.extack;
2589 if (dissector->used_keys &
2590 ~(BIT(FLOW_DISSECTOR_KEY_BASIC) |
2591 BIT(FLOW_DISSECTOR_KEY_CONTROL) |
2592 BIT(FLOW_DISSECTOR_KEY_ETH_ADDRS) |
2593 BIT(FLOW_DISSECTOR_KEY_VLAN))) {
2594 NL_SET_ERR_MSG_MOD(extack,
2595 "Unsupported key used, only BASIC, CONTROL, ETH_ADDRS and VLAN are supported");
2599 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ETH_ADDRS)) {
2600 struct flow_match_eth_addrs match;
2602 flow_rule_match_eth_addrs(rule, &match);
2603 if (!is_zero_ether_addr(match.mask->dst)) {
2604 if (!is_broadcast_ether_addr(match.mask->dst)) {
2605 NL_SET_ERR_MSG_MOD(extack, "Only full masks are supported for destination MAC address");
2609 input->filter.match_flags |=
2610 IGB_FILTER_FLAG_DST_MAC_ADDR;
2611 ether_addr_copy(input->filter.dst_addr, match.key->dst);
2614 if (!is_zero_ether_addr(match.mask->src)) {
2615 if (!is_broadcast_ether_addr(match.mask->src)) {
2616 NL_SET_ERR_MSG_MOD(extack, "Only full masks are supported for source MAC address");
2620 input->filter.match_flags |=
2621 IGB_FILTER_FLAG_SRC_MAC_ADDR;
2622 ether_addr_copy(input->filter.src_addr, match.key->src);
2626 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_BASIC)) {
2627 struct flow_match_basic match;
2629 flow_rule_match_basic(rule, &match);
2630 if (match.mask->n_proto) {
2631 if (match.mask->n_proto != ETHER_TYPE_FULL_MASK) {
2632 NL_SET_ERR_MSG_MOD(extack, "Only full mask is supported for EtherType filter");
2636 input->filter.match_flags |= IGB_FILTER_FLAG_ETHER_TYPE;
2637 input->filter.etype = match.key->n_proto;
2641 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_VLAN)) {
2642 struct flow_match_vlan match;
2644 flow_rule_match_vlan(rule, &match);
2645 if (match.mask->vlan_priority) {
2646 if (match.mask->vlan_priority != VLAN_PRIO_FULL_MASK) {
2647 NL_SET_ERR_MSG_MOD(extack, "Only full mask is supported for VLAN priority");
2651 input->filter.match_flags |= IGB_FILTER_FLAG_VLAN_TCI;
2652 input->filter.vlan_tci =
2653 (__force __be16)match.key->vlan_priority;
2657 input->action = traffic_class;
2658 input->cookie = f->cookie;
2663 static int igb_configure_clsflower(struct igb_adapter *adapter,
2664 struct flow_cls_offload *cls_flower)
2666 struct netlink_ext_ack *extack = cls_flower->common.extack;
2667 struct igb_nfc_filter *filter, *f;
2670 tc = tc_classid_to_hwtc(adapter->netdev, cls_flower->classid);
2672 NL_SET_ERR_MSG_MOD(extack, "Invalid traffic class");
2676 filter = kzalloc(sizeof(*filter), GFP_KERNEL);
2680 err = igb_parse_cls_flower(adapter, cls_flower, tc, filter);
2684 spin_lock(&adapter->nfc_lock);
2686 hlist_for_each_entry(f, &adapter->nfc_filter_list, nfc_node) {
2687 if (!memcmp(&f->filter, &filter->filter, sizeof(f->filter))) {
2689 NL_SET_ERR_MSG_MOD(extack,
2690 "This filter is already set in ethtool");
2695 hlist_for_each_entry(f, &adapter->cls_flower_list, nfc_node) {
2696 if (!memcmp(&f->filter, &filter->filter, sizeof(f->filter))) {
2698 NL_SET_ERR_MSG_MOD(extack,
2699 "This filter is already set in cls_flower");
2704 err = igb_add_filter(adapter, filter);
2706 NL_SET_ERR_MSG_MOD(extack, "Could not add filter to the adapter");
2710 hlist_add_head(&filter->nfc_node, &adapter->cls_flower_list);
2712 spin_unlock(&adapter->nfc_lock);
2717 spin_unlock(&adapter->nfc_lock);
2725 static int igb_delete_clsflower(struct igb_adapter *adapter,
2726 struct flow_cls_offload *cls_flower)
2728 struct igb_nfc_filter *filter;
2731 spin_lock(&adapter->nfc_lock);
2733 hlist_for_each_entry(filter, &adapter->cls_flower_list, nfc_node)
2734 if (filter->cookie == cls_flower->cookie)
2742 err = igb_erase_filter(adapter, filter);
2746 hlist_del(&filter->nfc_node);
2750 spin_unlock(&adapter->nfc_lock);
2755 static int igb_setup_tc_cls_flower(struct igb_adapter *adapter,
2756 struct flow_cls_offload *cls_flower)
2758 switch (cls_flower->command) {
2759 case FLOW_CLS_REPLACE:
2760 return igb_configure_clsflower(adapter, cls_flower);
2761 case FLOW_CLS_DESTROY:
2762 return igb_delete_clsflower(adapter, cls_flower);
2763 case FLOW_CLS_STATS:
2770 static int igb_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
2773 struct igb_adapter *adapter = cb_priv;
2775 if (!tc_cls_can_offload_and_chain0(adapter->netdev, type_data))
2779 case TC_SETUP_CLSFLOWER:
2780 return igb_setup_tc_cls_flower(adapter, type_data);
2787 static int igb_offload_txtime(struct igb_adapter *adapter,
2788 struct tc_etf_qopt_offload *qopt)
2790 struct e1000_hw *hw = &adapter->hw;
2793 /* Launchtime offloading is only supported by i210 controller. */
2794 if (hw->mac.type != e1000_i210)
2797 /* Launchtime offloading is only supported by queues 0 and 1. */
2798 if (qopt->queue < 0 || qopt->queue > 1)
2801 err = igb_save_txtime_params(adapter, qopt->queue, qopt->enable);
2805 igb_offload_apply(adapter, qopt->queue);
2810 static LIST_HEAD(igb_block_cb_list);
2812 static int igb_setup_tc(struct net_device *dev, enum tc_setup_type type,
2815 struct igb_adapter *adapter = netdev_priv(dev);
2818 case TC_SETUP_QDISC_CBS:
2819 return igb_offload_cbs(adapter, type_data);
2820 case TC_SETUP_BLOCK:
2821 return flow_block_cb_setup_simple(type_data,
2823 igb_setup_tc_block_cb,
2824 adapter, adapter, true);
2826 case TC_SETUP_QDISC_ETF:
2827 return igb_offload_txtime(adapter, type_data);
2834 static int igb_xdp_setup(struct net_device *dev, struct netdev_bpf *bpf)
2836 int i, frame_size = dev->mtu + IGB_ETH_PKT_HDR_PAD;
2837 struct igb_adapter *adapter = netdev_priv(dev);
2838 struct bpf_prog *prog = bpf->prog, *old_prog;
2839 bool running = netif_running(dev);
2842 /* verify igb ring attributes are sufficient for XDP */
2843 for (i = 0; i < adapter->num_rx_queues; i++) {
2844 struct igb_ring *ring = adapter->rx_ring[i];
2846 if (frame_size > igb_rx_bufsz(ring)) {
2847 NL_SET_ERR_MSG_MOD(bpf->extack,
2848 "The RX buffer size is too small for the frame size");
2849 netdev_warn(dev, "XDP RX buffer size %d is too small for the frame size %d\n",
2850 igb_rx_bufsz(ring), frame_size);
2855 old_prog = xchg(&adapter->xdp_prog, prog);
2856 need_reset = (!!prog != !!old_prog);
2858 /* device is up and bpf is added/removed, must setup the RX queues */
2859 if (need_reset && running) {
2862 for (i = 0; i < adapter->num_rx_queues; i++)
2863 (void)xchg(&adapter->rx_ring[i]->xdp_prog,
2868 bpf_prog_put(old_prog);
2870 /* bpf is just replaced, RXQ and MTU are already setup */
2880 static int igb_xdp(struct net_device *dev, struct netdev_bpf *xdp)
2882 switch (xdp->command) {
2883 case XDP_SETUP_PROG:
2884 return igb_xdp_setup(dev, xdp);
2890 static void igb_xdp_ring_update_tail(struct igb_ring *ring)
2892 /* Force memory writes to complete before letting h/w know there
2893 * are new descriptors to fetch.
2896 writel(ring->next_to_use, ring->tail);
2899 static struct igb_ring *igb_xdp_tx_queue_mapping(struct igb_adapter *adapter)
2901 unsigned int r_idx = smp_processor_id();
2903 if (r_idx >= adapter->num_tx_queues)
2904 r_idx = r_idx % adapter->num_tx_queues;
2906 return adapter->tx_ring[r_idx];
2909 static int igb_xdp_xmit_back(struct igb_adapter *adapter, struct xdp_buff *xdp)
2911 struct xdp_frame *xdpf = xdp_convert_buff_to_frame(xdp);
2912 int cpu = smp_processor_id();
2913 struct igb_ring *tx_ring;
2914 struct netdev_queue *nq;
2917 if (unlikely(!xdpf))
2918 return IGB_XDP_CONSUMED;
2920 /* During program transitions its possible adapter->xdp_prog is assigned
2921 * but ring has not been configured yet. In this case simply abort xmit.
2923 tx_ring = adapter->xdp_prog ? igb_xdp_tx_queue_mapping(adapter) : NULL;
2924 if (unlikely(!tx_ring))
2925 return IGB_XDP_CONSUMED;
2927 nq = txring_txq(tx_ring);
2928 __netif_tx_lock(nq, cpu);
2929 /* Avoid transmit queue timeout since we share it with the slow path */
2930 txq_trans_cond_update(nq);
2931 ret = igb_xmit_xdp_ring(adapter, tx_ring, xdpf);
2932 __netif_tx_unlock(nq);
2937 static int igb_xdp_xmit(struct net_device *dev, int n,
2938 struct xdp_frame **frames, u32 flags)
2940 struct igb_adapter *adapter = netdev_priv(dev);
2941 int cpu = smp_processor_id();
2942 struct igb_ring *tx_ring;
2943 struct netdev_queue *nq;
2947 if (unlikely(test_bit(__IGB_DOWN, &adapter->state)))
2950 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
2953 /* During program transitions its possible adapter->xdp_prog is assigned
2954 * but ring has not been configured yet. In this case simply abort xmit.
2956 tx_ring = adapter->xdp_prog ? igb_xdp_tx_queue_mapping(adapter) : NULL;
2957 if (unlikely(!tx_ring))
2960 nq = txring_txq(tx_ring);
2961 __netif_tx_lock(nq, cpu);
2963 /* Avoid transmit queue timeout since we share it with the slow path */
2964 txq_trans_cond_update(nq);
2966 for (i = 0; i < n; i++) {
2967 struct xdp_frame *xdpf = frames[i];
2970 err = igb_xmit_xdp_ring(adapter, tx_ring, xdpf);
2971 if (err != IGB_XDP_TX)
2976 __netif_tx_unlock(nq);
2978 if (unlikely(flags & XDP_XMIT_FLUSH))
2979 igb_xdp_ring_update_tail(tx_ring);
2984 static const struct net_device_ops igb_netdev_ops = {
2985 .ndo_open = igb_open,
2986 .ndo_stop = igb_close,
2987 .ndo_start_xmit = igb_xmit_frame,
2988 .ndo_get_stats64 = igb_get_stats64,
2989 .ndo_set_rx_mode = igb_set_rx_mode,
2990 .ndo_set_mac_address = igb_set_mac,
2991 .ndo_change_mtu = igb_change_mtu,
2992 .ndo_eth_ioctl = igb_ioctl,
2993 .ndo_tx_timeout = igb_tx_timeout,
2994 .ndo_validate_addr = eth_validate_addr,
2995 .ndo_vlan_rx_add_vid = igb_vlan_rx_add_vid,
2996 .ndo_vlan_rx_kill_vid = igb_vlan_rx_kill_vid,
2997 .ndo_set_vf_mac = igb_ndo_set_vf_mac,
2998 .ndo_set_vf_vlan = igb_ndo_set_vf_vlan,
2999 .ndo_set_vf_rate = igb_ndo_set_vf_bw,
3000 .ndo_set_vf_spoofchk = igb_ndo_set_vf_spoofchk,
3001 .ndo_set_vf_trust = igb_ndo_set_vf_trust,
3002 .ndo_get_vf_config = igb_ndo_get_vf_config,
3003 .ndo_fix_features = igb_fix_features,
3004 .ndo_set_features = igb_set_features,
3005 .ndo_fdb_add = igb_ndo_fdb_add,
3006 .ndo_features_check = igb_features_check,
3007 .ndo_setup_tc = igb_setup_tc,
3009 .ndo_xdp_xmit = igb_xdp_xmit,
3013 * igb_set_fw_version - Configure version string for ethtool
3014 * @adapter: adapter struct
3016 void igb_set_fw_version(struct igb_adapter *adapter)
3018 struct e1000_hw *hw = &adapter->hw;
3019 struct e1000_fw_version fw;
3021 igb_get_fw_version(hw, &fw);
3023 switch (hw->mac.type) {
3026 if (!(igb_get_flash_presence_i210(hw))) {
3027 snprintf(adapter->fw_version,
3028 sizeof(adapter->fw_version),
3030 fw.invm_major, fw.invm_minor,
3036 /* if option is rom valid, display its version too */
3038 snprintf(adapter->fw_version,
3039 sizeof(adapter->fw_version),
3040 "%d.%d, 0x%08x, %d.%d.%d",
3041 fw.eep_major, fw.eep_minor, fw.etrack_id,
3042 fw.or_major, fw.or_build, fw.or_patch);
3044 } else if (fw.etrack_id != 0X0000) {
3045 snprintf(adapter->fw_version,
3046 sizeof(adapter->fw_version),
3048 fw.eep_major, fw.eep_minor, fw.etrack_id);
3050 snprintf(adapter->fw_version,
3051 sizeof(adapter->fw_version),
3053 fw.eep_major, fw.eep_minor, fw.eep_build);
3060 * igb_init_mas - init Media Autosense feature if enabled in the NVM
3062 * @adapter: adapter struct
3064 static void igb_init_mas(struct igb_adapter *adapter)
3066 struct e1000_hw *hw = &adapter->hw;
3069 hw->nvm.ops.read(hw, NVM_COMPAT, 1, &eeprom_data);
3070 switch (hw->bus.func) {
3072 if (eeprom_data & IGB_MAS_ENABLE_0) {
3073 adapter->flags |= IGB_FLAG_MAS_ENABLE;
3074 netdev_info(adapter->netdev,
3075 "MAS: Enabling Media Autosense for port %d\n",
3080 if (eeprom_data & IGB_MAS_ENABLE_1) {
3081 adapter->flags |= IGB_FLAG_MAS_ENABLE;
3082 netdev_info(adapter->netdev,
3083 "MAS: Enabling Media Autosense for port %d\n",
3088 if (eeprom_data & IGB_MAS_ENABLE_2) {
3089 adapter->flags |= IGB_FLAG_MAS_ENABLE;
3090 netdev_info(adapter->netdev,
3091 "MAS: Enabling Media Autosense for port %d\n",
3096 if (eeprom_data & IGB_MAS_ENABLE_3) {
3097 adapter->flags |= IGB_FLAG_MAS_ENABLE;
3098 netdev_info(adapter->netdev,
3099 "MAS: Enabling Media Autosense for port %d\n",
3104 /* Shouldn't get here */
3105 netdev_err(adapter->netdev,
3106 "MAS: Invalid port configuration, returning\n");
3112 * igb_init_i2c - Init I2C interface
3113 * @adapter: pointer to adapter structure
3115 static s32 igb_init_i2c(struct igb_adapter *adapter)
3117 struct e1000_hw *hw = &adapter->hw;
3121 /* I2C interface supported on i350 devices */
3122 if (adapter->hw.mac.type != e1000_i350)
3125 i2cctl = rd32(E1000_I2CPARAMS);
3126 i2cctl |= E1000_I2CBB_EN
3127 | E1000_I2C_CLK_OUT | E1000_I2C_CLK_OE_N
3128 | E1000_I2C_DATA_OUT | E1000_I2C_DATA_OE_N;
3129 wr32(E1000_I2CPARAMS, i2cctl);
3132 /* Initialize the i2c bus which is controlled by the registers.
3133 * This bus will use the i2c_algo_bit structure that implements
3134 * the protocol through toggling of the 4 bits in the register.
3136 adapter->i2c_adap.owner = THIS_MODULE;
3137 adapter->i2c_algo = igb_i2c_algo;
3138 adapter->i2c_algo.data = adapter;
3139 adapter->i2c_adap.algo_data = &adapter->i2c_algo;
3140 adapter->i2c_adap.dev.parent = &adapter->pdev->dev;
3141 strlcpy(adapter->i2c_adap.name, "igb BB",
3142 sizeof(adapter->i2c_adap.name));
3143 status = i2c_bit_add_bus(&adapter->i2c_adap);
3148 * igb_probe - Device Initialization Routine
3149 * @pdev: PCI device information struct
3150 * @ent: entry in igb_pci_tbl
3152 * Returns 0 on success, negative on failure
3154 * igb_probe initializes an adapter identified by a pci_dev structure.
3155 * The OS initialization, configuring of the adapter private structure,
3156 * and a hardware reset occur.
3158 static int igb_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
3160 struct net_device *netdev;
3161 struct igb_adapter *adapter;
3162 struct e1000_hw *hw;
3163 u16 eeprom_data = 0;
3165 static int global_quad_port_a; /* global quad port a indication */
3166 const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
3167 u8 part_str[E1000_PBANUM_LENGTH];
3170 /* Catch broken hardware that put the wrong VF device ID in
3171 * the PCIe SR-IOV capability.
3173 if (pdev->is_virtfn) {
3174 WARN(1, KERN_ERR "%s (%x:%x) should not be a VF!\n",
3175 pci_name(pdev), pdev->vendor, pdev->device);
3179 err = pci_enable_device_mem(pdev);
3183 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
3186 "No usable DMA configuration, aborting\n");
3190 err = pci_request_mem_regions(pdev, igb_driver_name);
3194 pci_enable_pcie_error_reporting(pdev);
3196 pci_set_master(pdev);
3197 pci_save_state(pdev);
3200 netdev = alloc_etherdev_mq(sizeof(struct igb_adapter),
3203 goto err_alloc_etherdev;
3205 SET_NETDEV_DEV(netdev, &pdev->dev);
3207 pci_set_drvdata(pdev, netdev);
3208 adapter = netdev_priv(netdev);
3209 adapter->netdev = netdev;
3210 adapter->pdev = pdev;
3213 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
3216 adapter->io_addr = pci_iomap(pdev, 0, 0);
3217 if (!adapter->io_addr)
3219 /* hw->hw_addr can be altered, we'll use adapter->io_addr for unmap */
3220 hw->hw_addr = adapter->io_addr;
3222 netdev->netdev_ops = &igb_netdev_ops;
3223 igb_set_ethtool_ops(netdev);
3224 netdev->watchdog_timeo = 5 * HZ;
3226 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
3228 netdev->mem_start = pci_resource_start(pdev, 0);
3229 netdev->mem_end = pci_resource_end(pdev, 0);
3231 /* PCI config space info */
3232 hw->vendor_id = pdev->vendor;
3233 hw->device_id = pdev->device;
3234 hw->revision_id = pdev->revision;
3235 hw->subsystem_vendor_id = pdev->subsystem_vendor;
3236 hw->subsystem_device_id = pdev->subsystem_device;
3238 /* Copy the default MAC, PHY and NVM function pointers */
3239 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
3240 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
3241 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
3242 /* Initialize skew-specific constants */
3243 err = ei->get_invariants(hw);
3247 /* setup the private structure */
3248 err = igb_sw_init(adapter);
3252 igb_get_bus_info_pcie(hw);
3254 hw->phy.autoneg_wait_to_complete = false;
3256 /* Copper options */
3257 if (hw->phy.media_type == e1000_media_type_copper) {
3258 hw->phy.mdix = AUTO_ALL_MODES;
3259 hw->phy.disable_polarity_correction = false;
3260 hw->phy.ms_type = e1000_ms_hw_default;
3263 if (igb_check_reset_block(hw))
3264 dev_info(&pdev->dev,
3265 "PHY reset is blocked due to SOL/IDER session.\n");
3267 /* features is initialized to 0 in allocation, it might have bits
3268 * set by igb_sw_init so we should use an or instead of an
3271 netdev->features |= NETIF_F_SG |
3278 if (hw->mac.type >= e1000_82576)
3279 netdev->features |= NETIF_F_SCTP_CRC | NETIF_F_GSO_UDP_L4;
3281 if (hw->mac.type >= e1000_i350)
3282 netdev->features |= NETIF_F_HW_TC;
3284 #define IGB_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
3285 NETIF_F_GSO_GRE_CSUM | \
3286 NETIF_F_GSO_IPXIP4 | \
3287 NETIF_F_GSO_IPXIP6 | \
3288 NETIF_F_GSO_UDP_TUNNEL | \
3289 NETIF_F_GSO_UDP_TUNNEL_CSUM)
3291 netdev->gso_partial_features = IGB_GSO_PARTIAL_FEATURES;
3292 netdev->features |= NETIF_F_GSO_PARTIAL | IGB_GSO_PARTIAL_FEATURES;
3294 /* copy netdev features into list of user selectable features */
3295 netdev->hw_features |= netdev->features |
3296 NETIF_F_HW_VLAN_CTAG_RX |
3297 NETIF_F_HW_VLAN_CTAG_TX |
3300 if (hw->mac.type >= e1000_i350)
3301 netdev->hw_features |= NETIF_F_NTUPLE;
3303 netdev->features |= NETIF_F_HIGHDMA;
3305 netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
3306 netdev->mpls_features |= NETIF_F_HW_CSUM;
3307 netdev->hw_enc_features |= netdev->vlan_features;
3309 /* set this bit last since it cannot be part of vlan_features */
3310 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
3311 NETIF_F_HW_VLAN_CTAG_RX |
3312 NETIF_F_HW_VLAN_CTAG_TX;
3314 netdev->priv_flags |= IFF_SUPP_NOFCS;
3316 netdev->priv_flags |= IFF_UNICAST_FLT;
3318 /* MTU range: 68 - 9216 */
3319 netdev->min_mtu = ETH_MIN_MTU;
3320 netdev->max_mtu = MAX_STD_JUMBO_FRAME_SIZE;
3322 adapter->en_mng_pt = igb_enable_mng_pass_thru(hw);
3324 /* before reading the NVM, reset the controller to put the device in a
3325 * known good starting state
3327 hw->mac.ops.reset_hw(hw);
3329 /* make sure the NVM is good , i211/i210 parts can have special NVM
3330 * that doesn't contain a checksum
3332 switch (hw->mac.type) {
3335 if (igb_get_flash_presence_i210(hw)) {
3336 if (hw->nvm.ops.validate(hw) < 0) {
3338 "The NVM Checksum Is Not Valid\n");
3345 if (hw->nvm.ops.validate(hw) < 0) {
3346 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
3353 if (eth_platform_get_mac_address(&pdev->dev, hw->mac.addr)) {
3354 /* copy the MAC address out of the NVM */
3355 if (hw->mac.ops.read_mac_addr(hw))
3356 dev_err(&pdev->dev, "NVM Read Error\n");
3359 eth_hw_addr_set(netdev, hw->mac.addr);
3361 if (!is_valid_ether_addr(netdev->dev_addr)) {
3362 dev_err(&pdev->dev, "Invalid MAC Address\n");
3367 igb_set_default_mac_filter(adapter);
3369 /* get firmware version for ethtool -i */
3370 igb_set_fw_version(adapter);
3372 /* configure RXPBSIZE and TXPBSIZE */
3373 if (hw->mac.type == e1000_i210) {
3374 wr32(E1000_RXPBS, I210_RXPBSIZE_DEFAULT);
3375 wr32(E1000_TXPBS, I210_TXPBSIZE_DEFAULT);
3378 timer_setup(&adapter->watchdog_timer, igb_watchdog, 0);
3379 timer_setup(&adapter->phy_info_timer, igb_update_phy_info, 0);
3381 INIT_WORK(&adapter->reset_task, igb_reset_task);
3382 INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
3384 /* Initialize link properties that are user-changeable */
3385 adapter->fc_autoneg = true;
3386 hw->mac.autoneg = true;
3387 hw->phy.autoneg_advertised = 0x2f;
3389 hw->fc.requested_mode = e1000_fc_default;
3390 hw->fc.current_mode = e1000_fc_default;
3392 igb_validate_mdi_setting(hw);
3394 /* By default, support wake on port A */
3395 if (hw->bus.func == 0)
3396 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3398 /* Check the NVM for wake support on non-port A ports */
3399 if (hw->mac.type >= e1000_82580)
3400 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A +
3401 NVM_82580_LAN_FUNC_OFFSET(hw->bus.func), 1,
3403 else if (hw->bus.func == 1)
3404 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
3406 if (eeprom_data & IGB_EEPROM_APME)
3407 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3409 /* now that we have the eeprom settings, apply the special cases where
3410 * the eeprom may be wrong or the board simply won't support wake on
3411 * lan on a particular port
3413 switch (pdev->device) {
3414 case E1000_DEV_ID_82575GB_QUAD_COPPER:
3415 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
3417 case E1000_DEV_ID_82575EB_FIBER_SERDES:
3418 case E1000_DEV_ID_82576_FIBER:
3419 case E1000_DEV_ID_82576_SERDES:
3420 /* Wake events only supported on port A for dual fiber
3421 * regardless of eeprom setting
3423 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
3424 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
3426 case E1000_DEV_ID_82576_QUAD_COPPER:
3427 case E1000_DEV_ID_82576_QUAD_COPPER_ET2:
3428 /* if quad port adapter, disable WoL on all but port A */
3429 if (global_quad_port_a != 0)
3430 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
3432 adapter->flags |= IGB_FLAG_QUAD_PORT_A;
3433 /* Reset for multiple quad port adapters */
3434 if (++global_quad_port_a == 4)
3435 global_quad_port_a = 0;
3438 /* If the device can't wake, don't set software support */
3439 if (!device_can_wakeup(&adapter->pdev->dev))
3440 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
3443 /* initialize the wol settings based on the eeprom settings */
3444 if (adapter->flags & IGB_FLAG_WOL_SUPPORTED)
3445 adapter->wol |= E1000_WUFC_MAG;
3447 /* Some vendors want WoL disabled by default, but still supported */
3448 if ((hw->mac.type == e1000_i350) &&
3449 (pdev->subsystem_vendor == PCI_VENDOR_ID_HP)) {
3450 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3454 /* Some vendors want the ability to Use the EEPROM setting as
3455 * enable/disable only, and not for capability
3457 if (((hw->mac.type == e1000_i350) ||
3458 (hw->mac.type == e1000_i354)) &&
3459 (pdev->subsystem_vendor == PCI_VENDOR_ID_DELL)) {
3460 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3463 if (hw->mac.type == e1000_i350) {
3464 if (((pdev->subsystem_device == 0x5001) ||
3465 (pdev->subsystem_device == 0x5002)) &&
3466 (hw->bus.func == 0)) {
3467 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3470 if (pdev->subsystem_device == 0x1F52)
3471 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3474 device_set_wakeup_enable(&adapter->pdev->dev,
3475 adapter->flags & IGB_FLAG_WOL_SUPPORTED);
3477 /* reset the hardware with the new settings */
3480 /* Init the I2C interface */
3481 err = igb_init_i2c(adapter);
3483 dev_err(&pdev->dev, "failed to init i2c interface\n");
3487 /* let the f/w know that the h/w is now under the control of the
3490 igb_get_hw_control(adapter);
3492 strcpy(netdev->name, "eth%d");
3493 err = register_netdev(netdev);
3497 /* carrier off reporting is important to ethtool even BEFORE open */
3498 netif_carrier_off(netdev);
3500 #ifdef CONFIG_IGB_DCA
3501 if (dca_add_requester(&pdev->dev) == 0) {
3502 adapter->flags |= IGB_FLAG_DCA_ENABLED;
3503 dev_info(&pdev->dev, "DCA enabled\n");
3504 igb_setup_dca(adapter);
3508 #ifdef CONFIG_IGB_HWMON
3509 /* Initialize the thermal sensor on i350 devices. */
3510 if (hw->mac.type == e1000_i350 && hw->bus.func == 0) {
3513 /* Read the NVM to determine if this i350 device supports an
3514 * external thermal sensor.
3516 hw->nvm.ops.read(hw, NVM_ETS_CFG, 1, &ets_word);
3517 if (ets_word != 0x0000 && ets_word != 0xFFFF)
3518 adapter->ets = true;
3520 adapter->ets = false;
3521 if (igb_sysfs_init(adapter))
3523 "failed to allocate sysfs resources\n");
3525 adapter->ets = false;
3528 /* Check if Media Autosense is enabled */
3530 if (hw->dev_spec._82575.mas_capable)
3531 igb_init_mas(adapter);
3533 /* do hw tstamp init after resetting */
3534 igb_ptp_init(adapter);
3536 dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
3537 /* print bus type/speed/width info, not applicable to i354 */
3538 if (hw->mac.type != e1000_i354) {
3539 dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
3541 ((hw->bus.speed == e1000_bus_speed_2500) ? "2.5Gb/s" :
3542 (hw->bus.speed == e1000_bus_speed_5000) ? "5.0Gb/s" :
3544 ((hw->bus.width == e1000_bus_width_pcie_x4) ?
3546 (hw->bus.width == e1000_bus_width_pcie_x2) ?
3548 (hw->bus.width == e1000_bus_width_pcie_x1) ?
3549 "Width x1" : "unknown"), netdev->dev_addr);
3552 if ((hw->mac.type == e1000_82576 &&
3553 rd32(E1000_EECD) & E1000_EECD_PRES) ||
3554 (hw->mac.type >= e1000_i210 ||
3555 igb_get_flash_presence_i210(hw))) {
3556 ret_val = igb_read_part_string(hw, part_str,
3557 E1000_PBANUM_LENGTH);
3559 ret_val = -E1000_ERR_INVM_VALUE_NOT_FOUND;
3563 strcpy(part_str, "Unknown");
3564 dev_info(&pdev->dev, "%s: PBA No: %s\n", netdev->name, part_str);
3565 dev_info(&pdev->dev,
3566 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
3567 (adapter->flags & IGB_FLAG_HAS_MSIX) ? "MSI-X" :
3568 (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
3569 adapter->num_rx_queues, adapter->num_tx_queues);
3570 if (hw->phy.media_type == e1000_media_type_copper) {
3571 switch (hw->mac.type) {
3575 /* Enable EEE for internal copper PHY devices */
3576 err = igb_set_eee_i350(hw, true, true);
3578 (!hw->dev_spec._82575.eee_disable)) {
3579 adapter->eee_advert =
3580 MDIO_EEE_100TX | MDIO_EEE_1000T;
3581 adapter->flags |= IGB_FLAG_EEE;
3585 if ((rd32(E1000_CTRL_EXT) &
3586 E1000_CTRL_EXT_LINK_MODE_SGMII)) {
3587 err = igb_set_eee_i354(hw, true, true);
3589 (!hw->dev_spec._82575.eee_disable)) {
3590 adapter->eee_advert =
3591 MDIO_EEE_100TX | MDIO_EEE_1000T;
3592 adapter->flags |= IGB_FLAG_EEE;
3601 dev_pm_set_driver_flags(&pdev->dev, DPM_FLAG_NO_DIRECT_COMPLETE);
3603 pm_runtime_put_noidle(&pdev->dev);
3607 igb_release_hw_control(adapter);
3608 memset(&adapter->i2c_adap, 0, sizeof(adapter->i2c_adap));
3610 if (!igb_check_reset_block(hw))
3613 if (hw->flash_address)
3614 iounmap(hw->flash_address);
3616 kfree(adapter->mac_table);
3617 kfree(adapter->shadow_vfta);
3618 igb_clear_interrupt_scheme(adapter);
3619 #ifdef CONFIG_PCI_IOV
3620 igb_disable_sriov(pdev);
3622 pci_iounmap(pdev, adapter->io_addr);
3624 free_netdev(netdev);
3626 pci_disable_pcie_error_reporting(pdev);
3627 pci_release_mem_regions(pdev);
3630 pci_disable_device(pdev);
3634 #ifdef CONFIG_PCI_IOV
3635 static int igb_disable_sriov(struct pci_dev *pdev)
3637 struct net_device *netdev = pci_get_drvdata(pdev);
3638 struct igb_adapter *adapter = netdev_priv(netdev);
3639 struct e1000_hw *hw = &adapter->hw;
3640 unsigned long flags;
3642 /* reclaim resources allocated to VFs */
3643 if (adapter->vf_data) {
3644 /* disable iov and allow time for transactions to clear */
3645 if (pci_vfs_assigned(pdev)) {
3646 dev_warn(&pdev->dev,
3647 "Cannot deallocate SR-IOV virtual functions while they are assigned - VFs will not be deallocated\n");
3650 pci_disable_sriov(pdev);
3653 spin_lock_irqsave(&adapter->vfs_lock, flags);
3654 kfree(adapter->vf_mac_list);
3655 adapter->vf_mac_list = NULL;
3656 kfree(adapter->vf_data);
3657 adapter->vf_data = NULL;
3658 adapter->vfs_allocated_count = 0;
3659 spin_unlock_irqrestore(&adapter->vfs_lock, flags);
3660 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
3663 dev_info(&pdev->dev, "IOV Disabled\n");
3665 /* Re-enable DMA Coalescing flag since IOV is turned off */
3666 adapter->flags |= IGB_FLAG_DMAC;
3672 static int igb_enable_sriov(struct pci_dev *pdev, int num_vfs)
3674 struct net_device *netdev = pci_get_drvdata(pdev);
3675 struct igb_adapter *adapter = netdev_priv(netdev);
3676 int old_vfs = pci_num_vf(pdev);
3677 struct vf_mac_filter *mac_list;
3679 int num_vf_mac_filters, i;
3681 if (!(adapter->flags & IGB_FLAG_HAS_MSIX) || num_vfs > 7) {
3689 dev_info(&pdev->dev, "%d pre-allocated VFs found - override max_vfs setting of %d\n",
3691 adapter->vfs_allocated_count = old_vfs;
3693 adapter->vfs_allocated_count = num_vfs;
3695 adapter->vf_data = kcalloc(adapter->vfs_allocated_count,
3696 sizeof(struct vf_data_storage), GFP_KERNEL);
3698 /* if allocation failed then we do not support SR-IOV */
3699 if (!adapter->vf_data) {
3700 adapter->vfs_allocated_count = 0;
3705 /* Due to the limited number of RAR entries calculate potential
3706 * number of MAC filters available for the VFs. Reserve entries
3707 * for PF default MAC, PF MAC filters and at least one RAR entry
3708 * for each VF for VF MAC.
3710 num_vf_mac_filters = adapter->hw.mac.rar_entry_count -
3711 (1 + IGB_PF_MAC_FILTERS_RESERVED +
3712 adapter->vfs_allocated_count);
3714 adapter->vf_mac_list = kcalloc(num_vf_mac_filters,
3715 sizeof(struct vf_mac_filter),
3718 mac_list = adapter->vf_mac_list;
3719 INIT_LIST_HEAD(&adapter->vf_macs.l);
3721 if (adapter->vf_mac_list) {
3722 /* Initialize list of VF MAC filters */
3723 for (i = 0; i < num_vf_mac_filters; i++) {
3725 mac_list->free = true;
3726 list_add(&mac_list->l, &adapter->vf_macs.l);
3730 /* If we could not allocate memory for the VF MAC filters
3731 * we can continue without this feature but warn user.
3734 "Unable to allocate memory for VF MAC filter list\n");
3737 /* only call pci_enable_sriov() if no VFs are allocated already */
3739 err = pci_enable_sriov(pdev, adapter->vfs_allocated_count);
3743 dev_info(&pdev->dev, "%d VFs allocated\n",
3744 adapter->vfs_allocated_count);
3745 for (i = 0; i < adapter->vfs_allocated_count; i++)
3746 igb_vf_configure(adapter, i);
3748 /* DMA Coalescing is not supported in IOV mode. */
3749 adapter->flags &= ~IGB_FLAG_DMAC;
3753 kfree(adapter->vf_mac_list);
3754 adapter->vf_mac_list = NULL;
3755 kfree(adapter->vf_data);
3756 adapter->vf_data = NULL;
3757 adapter->vfs_allocated_count = 0;
3764 * igb_remove_i2c - Cleanup I2C interface
3765 * @adapter: pointer to adapter structure
3767 static void igb_remove_i2c(struct igb_adapter *adapter)
3769 /* free the adapter bus structure */
3770 i2c_del_adapter(&adapter->i2c_adap);
3774 * igb_remove - Device Removal Routine
3775 * @pdev: PCI device information struct
3777 * igb_remove is called by the PCI subsystem to alert the driver
3778 * that it should release a PCI device. The could be caused by a
3779 * Hot-Plug event, or because the driver is going to be removed from
3782 static void igb_remove(struct pci_dev *pdev)
3784 struct net_device *netdev = pci_get_drvdata(pdev);
3785 struct igb_adapter *adapter = netdev_priv(netdev);
3786 struct e1000_hw *hw = &adapter->hw;
3788 pm_runtime_get_noresume(&pdev->dev);
3789 #ifdef CONFIG_IGB_HWMON
3790 igb_sysfs_exit(adapter);
3792 igb_remove_i2c(adapter);
3793 igb_ptp_stop(adapter);
3794 /* The watchdog timer may be rescheduled, so explicitly
3795 * disable watchdog from being rescheduled.
3797 set_bit(__IGB_DOWN, &adapter->state);
3798 del_timer_sync(&adapter->watchdog_timer);
3799 del_timer_sync(&adapter->phy_info_timer);
3801 cancel_work_sync(&adapter->reset_task);
3802 cancel_work_sync(&adapter->watchdog_task);
3804 #ifdef CONFIG_IGB_DCA
3805 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
3806 dev_info(&pdev->dev, "DCA disabled\n");
3807 dca_remove_requester(&pdev->dev);
3808 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
3809 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
3813 /* Release control of h/w to f/w. If f/w is AMT enabled, this
3814 * would have already happened in close and is redundant.
3816 igb_release_hw_control(adapter);
3818 #ifdef CONFIG_PCI_IOV
3820 igb_disable_sriov(pdev);
3824 unregister_netdev(netdev);
3826 igb_clear_interrupt_scheme(adapter);
3828 pci_iounmap(pdev, adapter->io_addr);
3829 if (hw->flash_address)
3830 iounmap(hw->flash_address);
3831 pci_release_mem_regions(pdev);
3833 kfree(adapter->mac_table);
3834 kfree(adapter->shadow_vfta);
3835 free_netdev(netdev);
3837 pci_disable_pcie_error_reporting(pdev);
3839 pci_disable_device(pdev);
3843 * igb_probe_vfs - Initialize vf data storage and add VFs to pci config space
3844 * @adapter: board private structure to initialize
3846 * This function initializes the vf specific data storage and then attempts to
3847 * allocate the VFs. The reason for ordering it this way is because it is much
3848 * mor expensive time wise to disable SR-IOV than it is to allocate and free
3849 * the memory for the VFs.
3851 static void igb_probe_vfs(struct igb_adapter *adapter)
3853 #ifdef CONFIG_PCI_IOV
3854 struct pci_dev *pdev = adapter->pdev;
3855 struct e1000_hw *hw = &adapter->hw;
3857 /* Virtualization features not supported on i210 family. */
3858 if ((hw->mac.type == e1000_i210) || (hw->mac.type == e1000_i211))
3861 /* Of the below we really only want the effect of getting
3862 * IGB_FLAG_HAS_MSIX set (if available), without which
3863 * igb_enable_sriov() has no effect.
3865 igb_set_interrupt_capability(adapter, true);
3866 igb_reset_interrupt_capability(adapter);
3868 pci_sriov_set_totalvfs(pdev, 7);
3869 igb_enable_sriov(pdev, max_vfs);
3871 #endif /* CONFIG_PCI_IOV */
3874 unsigned int igb_get_max_rss_queues(struct igb_adapter *adapter)
3876 struct e1000_hw *hw = &adapter->hw;
3877 unsigned int max_rss_queues;
3879 /* Determine the maximum number of RSS queues supported. */
3880 switch (hw->mac.type) {
3882 max_rss_queues = IGB_MAX_RX_QUEUES_I211;
3886 max_rss_queues = IGB_MAX_RX_QUEUES_82575;
3889 /* I350 cannot do RSS and SR-IOV at the same time */
3890 if (!!adapter->vfs_allocated_count) {
3896 if (!!adapter->vfs_allocated_count) {
3904 max_rss_queues = IGB_MAX_RX_QUEUES;
3908 return max_rss_queues;
3911 static void igb_init_queue_configuration(struct igb_adapter *adapter)
3915 max_rss_queues = igb_get_max_rss_queues(adapter);
3916 adapter->rss_queues = min_t(u32, max_rss_queues, num_online_cpus());
3918 igb_set_flag_queue_pairs(adapter, max_rss_queues);
3921 void igb_set_flag_queue_pairs(struct igb_adapter *adapter,
3922 const u32 max_rss_queues)
3924 struct e1000_hw *hw = &adapter->hw;
3926 /* Determine if we need to pair queues. */
3927 switch (hw->mac.type) {
3930 /* Device supports enough interrupts without queue pairing. */
3938 /* If rss_queues > half of max_rss_queues, pair the queues in
3939 * order to conserve interrupts due to limited supply.
3941 if (adapter->rss_queues > (max_rss_queues / 2))
3942 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
3944 adapter->flags &= ~IGB_FLAG_QUEUE_PAIRS;
3950 * igb_sw_init - Initialize general software structures (struct igb_adapter)
3951 * @adapter: board private structure to initialize
3953 * igb_sw_init initializes the Adapter private data structure.
3954 * Fields are initialized based on PCI device information and
3955 * OS network device settings (MTU size).
3957 static int igb_sw_init(struct igb_adapter *adapter)
3959 struct e1000_hw *hw = &adapter->hw;
3960 struct net_device *netdev = adapter->netdev;
3961 struct pci_dev *pdev = adapter->pdev;
3963 pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
3965 /* set default ring sizes */
3966 adapter->tx_ring_count = IGB_DEFAULT_TXD;
3967 adapter->rx_ring_count = IGB_DEFAULT_RXD;
3969 /* set default ITR values */
3970 adapter->rx_itr_setting = IGB_DEFAULT_ITR;
3971 adapter->tx_itr_setting = IGB_DEFAULT_ITR;
3973 /* set default work limits */
3974 adapter->tx_work_limit = IGB_DEFAULT_TX_WORK;
3976 adapter->max_frame_size = netdev->mtu + IGB_ETH_PKT_HDR_PAD;
3977 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3979 spin_lock_init(&adapter->nfc_lock);
3980 spin_lock_init(&adapter->stats64_lock);
3982 /* init spinlock to avoid concurrency of VF resources */
3983 spin_lock_init(&adapter->vfs_lock);
3984 #ifdef CONFIG_PCI_IOV
3985 switch (hw->mac.type) {
3989 dev_warn(&pdev->dev,
3990 "Maximum of 7 VFs per PF, using max\n");
3991 max_vfs = adapter->vfs_allocated_count = 7;
3993 adapter->vfs_allocated_count = max_vfs;
3994 if (adapter->vfs_allocated_count)
3995 dev_warn(&pdev->dev,
3996 "Enabling SR-IOV VFs using the module parameter is deprecated - please use the pci sysfs interface.\n");
4001 #endif /* CONFIG_PCI_IOV */
4003 /* Assume MSI-X interrupts, will be checked during IRQ allocation */
4004 adapter->flags |= IGB_FLAG_HAS_MSIX;
4006 adapter->mac_table = kcalloc(hw->mac.rar_entry_count,
4007 sizeof(struct igb_mac_addr),
4009 if (!adapter->mac_table)
4012 igb_probe_vfs(adapter);
4014 igb_init_queue_configuration(adapter);
4016 /* Setup and initialize a copy of the hw vlan table array */
4017 adapter->shadow_vfta = kcalloc(E1000_VLAN_FILTER_TBL_SIZE, sizeof(u32),
4019 if (!adapter->shadow_vfta)
4022 /* This call may decrease the number of queues */
4023 if (igb_init_interrupt_scheme(adapter, true)) {
4024 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
4028 /* Explicitly disable IRQ since the NIC can be in any state. */
4029 igb_irq_disable(adapter);
4031 if (hw->mac.type >= e1000_i350)
4032 adapter->flags &= ~IGB_FLAG_DMAC;
4034 set_bit(__IGB_DOWN, &adapter->state);
4039 * __igb_open - Called when a network interface is made active
4040 * @netdev: network interface device structure
4041 * @resuming: indicates whether we are in a resume call
4043 * Returns 0 on success, negative value on failure
4045 * The open entry point is called when a network interface is made
4046 * active by the system (IFF_UP). At this point all resources needed
4047 * for transmit and receive operations are allocated, the interrupt
4048 * handler is registered with the OS, the watchdog timer is started,
4049 * and the stack is notified that the interface is ready.
4051 static int __igb_open(struct net_device *netdev, bool resuming)
4053 struct igb_adapter *adapter = netdev_priv(netdev);
4054 struct e1000_hw *hw = &adapter->hw;
4055 struct pci_dev *pdev = adapter->pdev;
4059 /* disallow open during test */
4060 if (test_bit(__IGB_TESTING, &adapter->state)) {
4066 pm_runtime_get_sync(&pdev->dev);
4068 netif_carrier_off(netdev);
4070 /* allocate transmit descriptors */
4071 err = igb_setup_all_tx_resources(adapter);
4075 /* allocate receive descriptors */
4076 err = igb_setup_all_rx_resources(adapter);
4080 igb_power_up_link(adapter);
4082 /* before we allocate an interrupt, we must be ready to handle it.
4083 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
4084 * as soon as we call pci_request_irq, so we have to setup our
4085 * clean_rx handler before we do so.
4087 igb_configure(adapter);
4089 err = igb_request_irq(adapter);
4093 /* Notify the stack of the actual queue counts. */
4094 err = netif_set_real_num_tx_queues(adapter->netdev,
4095 adapter->num_tx_queues);
4097 goto err_set_queues;
4099 err = netif_set_real_num_rx_queues(adapter->netdev,
4100 adapter->num_rx_queues);
4102 goto err_set_queues;
4104 /* From here on the code is the same as igb_up() */
4105 clear_bit(__IGB_DOWN, &adapter->state);
4107 for (i = 0; i < adapter->num_q_vectors; i++)
4108 napi_enable(&(adapter->q_vector[i]->napi));
4110 /* Clear any pending interrupts. */
4114 igb_irq_enable(adapter);
4116 /* notify VFs that reset has been completed */
4117 if (adapter->vfs_allocated_count) {
4118 u32 reg_data = rd32(E1000_CTRL_EXT);
4120 reg_data |= E1000_CTRL_EXT_PFRSTD;
4121 wr32(E1000_CTRL_EXT, reg_data);
4124 netif_tx_start_all_queues(netdev);
4127 pm_runtime_put(&pdev->dev);
4129 /* start the watchdog. */
4130 hw->mac.get_link_status = 1;
4131 schedule_work(&adapter->watchdog_task);
4136 igb_free_irq(adapter);
4138 igb_release_hw_control(adapter);
4139 igb_power_down_link(adapter);
4140 igb_free_all_rx_resources(adapter);
4142 igb_free_all_tx_resources(adapter);
4146 pm_runtime_put(&pdev->dev);
4151 int igb_open(struct net_device *netdev)
4153 return __igb_open(netdev, false);
4157 * __igb_close - Disables a network interface
4158 * @netdev: network interface device structure
4159 * @suspending: indicates we are in a suspend call
4161 * Returns 0, this is not allowed to fail
4163 * The close entry point is called when an interface is de-activated
4164 * by the OS. The hardware is still under the driver's control, but
4165 * needs to be disabled. A global MAC reset is issued to stop the
4166 * hardware, and all transmit and receive resources are freed.
4168 static int __igb_close(struct net_device *netdev, bool suspending)
4170 struct igb_adapter *adapter = netdev_priv(netdev);
4171 struct pci_dev *pdev = adapter->pdev;
4173 WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
4176 pm_runtime_get_sync(&pdev->dev);
4179 igb_free_irq(adapter);
4181 igb_free_all_tx_resources(adapter);
4182 igb_free_all_rx_resources(adapter);
4185 pm_runtime_put_sync(&pdev->dev);
4189 int igb_close(struct net_device *netdev)
4191 if (netif_device_present(netdev) || netdev->dismantle)
4192 return __igb_close(netdev, false);
4197 * igb_setup_tx_resources - allocate Tx resources (Descriptors)
4198 * @tx_ring: tx descriptor ring (for a specific queue) to setup
4200 * Return 0 on success, negative on failure
4202 int igb_setup_tx_resources(struct igb_ring *tx_ring)
4204 struct device *dev = tx_ring->dev;
4207 size = sizeof(struct igb_tx_buffer) * tx_ring->count;
4209 tx_ring->tx_buffer_info = vmalloc(size);
4210 if (!tx_ring->tx_buffer_info)
4213 /* round up to nearest 4K */
4214 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
4215 tx_ring->size = ALIGN(tx_ring->size, 4096);
4217 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
4218 &tx_ring->dma, GFP_KERNEL);
4222 tx_ring->next_to_use = 0;
4223 tx_ring->next_to_clean = 0;
4228 vfree(tx_ring->tx_buffer_info);
4229 tx_ring->tx_buffer_info = NULL;
4230 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
4235 * igb_setup_all_tx_resources - wrapper to allocate Tx resources
4236 * (Descriptors) for all queues
4237 * @adapter: board private structure
4239 * Return 0 on success, negative on failure
4241 static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
4243 struct pci_dev *pdev = adapter->pdev;
4246 for (i = 0; i < adapter->num_tx_queues; i++) {
4247 err = igb_setup_tx_resources(adapter->tx_ring[i]);
4250 "Allocation for Tx Queue %u failed\n", i);
4251 for (i--; i >= 0; i--)
4252 igb_free_tx_resources(adapter->tx_ring[i]);
4261 * igb_setup_tctl - configure the transmit control registers
4262 * @adapter: Board private structure
4264 void igb_setup_tctl(struct igb_adapter *adapter)
4266 struct e1000_hw *hw = &adapter->hw;
4269 /* disable queue 0 which is enabled by default on 82575 and 82576 */
4270 wr32(E1000_TXDCTL(0), 0);
4272 /* Program the Transmit Control Register */
4273 tctl = rd32(E1000_TCTL);
4274 tctl &= ~E1000_TCTL_CT;
4275 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
4276 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
4278 igb_config_collision_dist(hw);
4280 /* Enable transmits */
4281 tctl |= E1000_TCTL_EN;
4283 wr32(E1000_TCTL, tctl);
4287 * igb_configure_tx_ring - Configure transmit ring after Reset
4288 * @adapter: board private structure
4289 * @ring: tx ring to configure
4291 * Configure a transmit ring after a reset.
4293 void igb_configure_tx_ring(struct igb_adapter *adapter,
4294 struct igb_ring *ring)
4296 struct e1000_hw *hw = &adapter->hw;
4298 u64 tdba = ring->dma;
4299 int reg_idx = ring->reg_idx;
4301 wr32(E1000_TDLEN(reg_idx),
4302 ring->count * sizeof(union e1000_adv_tx_desc));
4303 wr32(E1000_TDBAL(reg_idx),
4304 tdba & 0x00000000ffffffffULL);
4305 wr32(E1000_TDBAH(reg_idx), tdba >> 32);
4307 ring->tail = adapter->io_addr + E1000_TDT(reg_idx);
4308 wr32(E1000_TDH(reg_idx), 0);
4309 writel(0, ring->tail);
4311 txdctl |= IGB_TX_PTHRESH;
4312 txdctl |= IGB_TX_HTHRESH << 8;
4313 txdctl |= IGB_TX_WTHRESH << 16;
4315 /* reinitialize tx_buffer_info */
4316 memset(ring->tx_buffer_info, 0,
4317 sizeof(struct igb_tx_buffer) * ring->count);
4319 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
4320 wr32(E1000_TXDCTL(reg_idx), txdctl);
4324 * igb_configure_tx - Configure transmit Unit after Reset
4325 * @adapter: board private structure
4327 * Configure the Tx unit of the MAC after a reset.
4329 static void igb_configure_tx(struct igb_adapter *adapter)
4331 struct e1000_hw *hw = &adapter->hw;
4334 /* disable the queues */
4335 for (i = 0; i < adapter->num_tx_queues; i++)
4336 wr32(E1000_TXDCTL(adapter->tx_ring[i]->reg_idx), 0);
4339 usleep_range(10000, 20000);
4341 for (i = 0; i < adapter->num_tx_queues; i++)
4342 igb_configure_tx_ring(adapter, adapter->tx_ring[i]);
4346 * igb_setup_rx_resources - allocate Rx resources (Descriptors)
4347 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
4349 * Returns 0 on success, negative on failure
4351 int igb_setup_rx_resources(struct igb_ring *rx_ring)
4353 struct igb_adapter *adapter = netdev_priv(rx_ring->netdev);
4354 struct device *dev = rx_ring->dev;
4357 /* XDP RX-queue info */
4358 if (xdp_rxq_info_is_reg(&rx_ring->xdp_rxq))
4359 xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
4360 res = xdp_rxq_info_reg(&rx_ring->xdp_rxq, rx_ring->netdev,
4361 rx_ring->queue_index, 0);
4363 dev_err(dev, "Failed to register xdp_rxq index %u\n",
4364 rx_ring->queue_index);
4368 size = sizeof(struct igb_rx_buffer) * rx_ring->count;
4370 rx_ring->rx_buffer_info = vmalloc(size);
4371 if (!rx_ring->rx_buffer_info)
4374 /* Round up to nearest 4K */
4375 rx_ring->size = rx_ring->count * sizeof(union e1000_adv_rx_desc);
4376 rx_ring->size = ALIGN(rx_ring->size, 4096);
4378 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
4379 &rx_ring->dma, GFP_KERNEL);
4383 rx_ring->next_to_alloc = 0;
4384 rx_ring->next_to_clean = 0;
4385 rx_ring->next_to_use = 0;
4387 rx_ring->xdp_prog = adapter->xdp_prog;
4392 xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
4393 vfree(rx_ring->rx_buffer_info);
4394 rx_ring->rx_buffer_info = NULL;
4395 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
4400 * igb_setup_all_rx_resources - wrapper to allocate Rx resources
4401 * (Descriptors) for all queues
4402 * @adapter: board private structure
4404 * Return 0 on success, negative on failure
4406 static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
4408 struct pci_dev *pdev = adapter->pdev;
4411 for (i = 0; i < adapter->num_rx_queues; i++) {
4412 err = igb_setup_rx_resources(adapter->rx_ring[i]);
4415 "Allocation for Rx Queue %u failed\n", i);
4416 for (i--; i >= 0; i--)
4417 igb_free_rx_resources(adapter->rx_ring[i]);
4426 * igb_setup_mrqc - configure the multiple receive queue control registers
4427 * @adapter: Board private structure
4429 static void igb_setup_mrqc(struct igb_adapter *adapter)
4431 struct e1000_hw *hw = &adapter->hw;
4433 u32 j, num_rx_queues;
4436 netdev_rss_key_fill(rss_key, sizeof(rss_key));
4437 for (j = 0; j < 10; j++)
4438 wr32(E1000_RSSRK(j), rss_key[j]);
4440 num_rx_queues = adapter->rss_queues;
4442 switch (hw->mac.type) {
4444 /* 82576 supports 2 RSS queues for SR-IOV */
4445 if (adapter->vfs_allocated_count)
4452 if (adapter->rss_indir_tbl_init != num_rx_queues) {
4453 for (j = 0; j < IGB_RETA_SIZE; j++)
4454 adapter->rss_indir_tbl[j] =
4455 (j * num_rx_queues) / IGB_RETA_SIZE;
4456 adapter->rss_indir_tbl_init = num_rx_queues;
4458 igb_write_rss_indir_tbl(adapter);
4460 /* Disable raw packet checksumming so that RSS hash is placed in
4461 * descriptor on writeback. No need to enable TCP/UDP/IP checksum
4462 * offloads as they are enabled by default
4464 rxcsum = rd32(E1000_RXCSUM);
4465 rxcsum |= E1000_RXCSUM_PCSD;
4467 if (adapter->hw.mac.type >= e1000_82576)
4468 /* Enable Receive Checksum Offload for SCTP */
4469 rxcsum |= E1000_RXCSUM_CRCOFL;
4471 /* Don't need to set TUOFL or IPOFL, they default to 1 */
4472 wr32(E1000_RXCSUM, rxcsum);
4474 /* Generate RSS hash based on packet types, TCP/UDP
4475 * port numbers and/or IPv4/v6 src and dst addresses
4477 mrqc = E1000_MRQC_RSS_FIELD_IPV4 |
4478 E1000_MRQC_RSS_FIELD_IPV4_TCP |
4479 E1000_MRQC_RSS_FIELD_IPV6 |
4480 E1000_MRQC_RSS_FIELD_IPV6_TCP |
4481 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX;
4483 if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV4_UDP)
4484 mrqc |= E1000_MRQC_RSS_FIELD_IPV4_UDP;
4485 if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV6_UDP)
4486 mrqc |= E1000_MRQC_RSS_FIELD_IPV6_UDP;
4488 /* If VMDq is enabled then we set the appropriate mode for that, else
4489 * we default to RSS so that an RSS hash is calculated per packet even
4490 * if we are only using one queue
4492 if (adapter->vfs_allocated_count) {
4493 if (hw->mac.type > e1000_82575) {
4494 /* Set the default pool for the PF's first queue */
4495 u32 vtctl = rd32(E1000_VT_CTL);
4497 vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK |
4498 E1000_VT_CTL_DISABLE_DEF_POOL);
4499 vtctl |= adapter->vfs_allocated_count <<
4500 E1000_VT_CTL_DEFAULT_POOL_SHIFT;
4501 wr32(E1000_VT_CTL, vtctl);
4503 if (adapter->rss_queues > 1)
4504 mrqc |= E1000_MRQC_ENABLE_VMDQ_RSS_MQ;
4506 mrqc |= E1000_MRQC_ENABLE_VMDQ;
4508 mrqc |= E1000_MRQC_ENABLE_RSS_MQ;
4510 igb_vmm_control(adapter);
4512 wr32(E1000_MRQC, mrqc);
4516 * igb_setup_rctl - configure the receive control registers
4517 * @adapter: Board private structure
4519 void igb_setup_rctl(struct igb_adapter *adapter)
4521 struct e1000_hw *hw = &adapter->hw;
4524 rctl = rd32(E1000_RCTL);
4526 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
4527 rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
4529 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
4530 (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
4532 /* enable stripping of CRC. It's unlikely this will break BMC
4533 * redirection as it did with e1000. Newer features require
4534 * that the HW strips the CRC.
4536 rctl |= E1000_RCTL_SECRC;
4538 /* disable store bad packets and clear size bits. */
4539 rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
4541 /* enable LPE to allow for reception of jumbo frames */
4542 rctl |= E1000_RCTL_LPE;
4544 /* disable queue 0 to prevent tail write w/o re-config */
4545 wr32(E1000_RXDCTL(0), 0);
4547 /* Attention!!! For SR-IOV PF driver operations you must enable
4548 * queue drop for all VF and PF queues to prevent head of line blocking
4549 * if an un-trusted VF does not provide descriptors to hardware.
4551 if (adapter->vfs_allocated_count) {
4552 /* set all queue drop enable bits */
4553 wr32(E1000_QDE, ALL_QUEUES);
4556 /* This is useful for sniffing bad packets. */
4557 if (adapter->netdev->features & NETIF_F_RXALL) {
4558 /* UPE and MPE will be handled by normal PROMISC logic
4559 * in e1000e_set_rx_mode
4561 rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
4562 E1000_RCTL_BAM | /* RX All Bcast Pkts */
4563 E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
4565 rctl &= ~(E1000_RCTL_DPF | /* Allow filtered pause */
4566 E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
4567 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
4568 * and that breaks VLANs.
4572 wr32(E1000_RCTL, rctl);
4575 static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
4578 struct e1000_hw *hw = &adapter->hw;
4581 if (size > MAX_JUMBO_FRAME_SIZE)
4582 size = MAX_JUMBO_FRAME_SIZE;
4584 vmolr = rd32(E1000_VMOLR(vfn));
4585 vmolr &= ~E1000_VMOLR_RLPML_MASK;
4586 vmolr |= size | E1000_VMOLR_LPE;
4587 wr32(E1000_VMOLR(vfn), vmolr);
4592 static inline void igb_set_vf_vlan_strip(struct igb_adapter *adapter,
4593 int vfn, bool enable)
4595 struct e1000_hw *hw = &adapter->hw;
4598 if (hw->mac.type < e1000_82576)
4601 if (hw->mac.type == e1000_i350)
4602 reg = E1000_DVMOLR(vfn);
4604 reg = E1000_VMOLR(vfn);
4608 val |= E1000_VMOLR_STRVLAN;
4610 val &= ~(E1000_VMOLR_STRVLAN);
4614 static inline void igb_set_vmolr(struct igb_adapter *adapter,
4617 struct e1000_hw *hw = &adapter->hw;
4620 /* This register exists only on 82576 and newer so if we are older then
4621 * we should exit and do nothing
4623 if (hw->mac.type < e1000_82576)
4626 vmolr = rd32(E1000_VMOLR(vfn));
4628 vmolr |= E1000_VMOLR_AUPE; /* Accept untagged packets */
4630 vmolr &= ~(E1000_VMOLR_AUPE); /* Tagged packets ONLY */
4632 /* clear all bits that might not be set */
4633 vmolr &= ~(E1000_VMOLR_BAM | E1000_VMOLR_RSSE);
4635 if (adapter->rss_queues > 1 && vfn == adapter->vfs_allocated_count)
4636 vmolr |= E1000_VMOLR_RSSE; /* enable RSS */
4637 /* for VMDq only allow the VFs and pool 0 to accept broadcast and
4640 if (vfn <= adapter->vfs_allocated_count)
4641 vmolr |= E1000_VMOLR_BAM; /* Accept broadcast */
4643 wr32(E1000_VMOLR(vfn), vmolr);
4647 * igb_setup_srrctl - configure the split and replication receive control
4649 * @adapter: Board private structure
4650 * @ring: receive ring to be configured
4652 void igb_setup_srrctl(struct igb_adapter *adapter, struct igb_ring *ring)
4654 struct e1000_hw *hw = &adapter->hw;
4655 int reg_idx = ring->reg_idx;
4658 srrctl = IGB_RX_HDR_LEN << E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
4659 if (ring_uses_large_buffer(ring))
4660 srrctl |= IGB_RXBUFFER_3072 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
4662 srrctl |= IGB_RXBUFFER_2048 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
4663 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
4664 if (hw->mac.type >= e1000_82580)
4665 srrctl |= E1000_SRRCTL_TIMESTAMP;
4666 /* Only set Drop Enable if VFs allocated, or we are supporting multiple
4667 * queues and rx flow control is disabled
4669 if (adapter->vfs_allocated_count ||
4670 (!(hw->fc.current_mode & e1000_fc_rx_pause) &&
4671 adapter->num_rx_queues > 1))
4672 srrctl |= E1000_SRRCTL_DROP_EN;
4674 wr32(E1000_SRRCTL(reg_idx), srrctl);
4678 * igb_configure_rx_ring - Configure a receive ring after Reset
4679 * @adapter: board private structure
4680 * @ring: receive ring to be configured
4682 * Configure the Rx unit of the MAC after a reset.
4684 void igb_configure_rx_ring(struct igb_adapter *adapter,
4685 struct igb_ring *ring)
4687 struct e1000_hw *hw = &adapter->hw;
4688 union e1000_adv_rx_desc *rx_desc;
4689 u64 rdba = ring->dma;
4690 int reg_idx = ring->reg_idx;
4693 xdp_rxq_info_unreg_mem_model(&ring->xdp_rxq);
4694 WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
4695 MEM_TYPE_PAGE_SHARED, NULL));
4697 /* disable the queue */
4698 wr32(E1000_RXDCTL(reg_idx), 0);
4700 /* Set DMA base address registers */
4701 wr32(E1000_RDBAL(reg_idx),
4702 rdba & 0x00000000ffffffffULL);
4703 wr32(E1000_RDBAH(reg_idx), rdba >> 32);
4704 wr32(E1000_RDLEN(reg_idx),
4705 ring->count * sizeof(union e1000_adv_rx_desc));
4707 /* initialize head and tail */
4708 ring->tail = adapter->io_addr + E1000_RDT(reg_idx);
4709 wr32(E1000_RDH(reg_idx), 0);
4710 writel(0, ring->tail);
4712 /* set descriptor configuration */
4713 igb_setup_srrctl(adapter, ring);
4715 /* set filtering for VMDQ pools */
4716 igb_set_vmolr(adapter, reg_idx & 0x7, true);
4718 rxdctl |= IGB_RX_PTHRESH;
4719 rxdctl |= IGB_RX_HTHRESH << 8;
4720 rxdctl |= IGB_RX_WTHRESH << 16;
4722 /* initialize rx_buffer_info */
4723 memset(ring->rx_buffer_info, 0,
4724 sizeof(struct igb_rx_buffer) * ring->count);
4726 /* initialize Rx descriptor 0 */
4727 rx_desc = IGB_RX_DESC(ring, 0);
4728 rx_desc->wb.upper.length = 0;
4730 /* enable receive descriptor fetching */
4731 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
4732 wr32(E1000_RXDCTL(reg_idx), rxdctl);
4735 static void igb_set_rx_buffer_len(struct igb_adapter *adapter,
4736 struct igb_ring *rx_ring)
4738 /* set build_skb and buffer size flags */
4739 clear_ring_build_skb_enabled(rx_ring);
4740 clear_ring_uses_large_buffer(rx_ring);
4742 if (adapter->flags & IGB_FLAG_RX_LEGACY)
4745 set_ring_build_skb_enabled(rx_ring);
4747 #if (PAGE_SIZE < 8192)
4748 if (adapter->max_frame_size <= IGB_MAX_FRAME_BUILD_SKB)
4751 set_ring_uses_large_buffer(rx_ring);
4756 * igb_configure_rx - Configure receive Unit after Reset
4757 * @adapter: board private structure
4759 * Configure the Rx unit of the MAC after a reset.
4761 static void igb_configure_rx(struct igb_adapter *adapter)
4765 /* set the correct pool for the PF default MAC address in entry 0 */
4766 igb_set_default_mac_filter(adapter);
4768 /* Setup the HW Rx Head and Tail Descriptor Pointers and
4769 * the Base and Length of the Rx Descriptor Ring
4771 for (i = 0; i < adapter->num_rx_queues; i++) {
4772 struct igb_ring *rx_ring = adapter->rx_ring[i];
4774 igb_set_rx_buffer_len(adapter, rx_ring);
4775 igb_configure_rx_ring(adapter, rx_ring);
4780 * igb_free_tx_resources - Free Tx Resources per Queue
4781 * @tx_ring: Tx descriptor ring for a specific queue
4783 * Free all transmit software resources
4785 void igb_free_tx_resources(struct igb_ring *tx_ring)
4787 igb_clean_tx_ring(tx_ring);
4789 vfree(tx_ring->tx_buffer_info);
4790 tx_ring->tx_buffer_info = NULL;
4792 /* if not set, then don't free */
4796 dma_free_coherent(tx_ring->dev, tx_ring->size,
4797 tx_ring->desc, tx_ring->dma);
4799 tx_ring->desc = NULL;
4803 * igb_free_all_tx_resources - Free Tx Resources for All Queues
4804 * @adapter: board private structure
4806 * Free all transmit software resources
4808 static void igb_free_all_tx_resources(struct igb_adapter *adapter)
4812 for (i = 0; i < adapter->num_tx_queues; i++)
4813 if (adapter->tx_ring[i])
4814 igb_free_tx_resources(adapter->tx_ring[i]);
4818 * igb_clean_tx_ring - Free Tx Buffers
4819 * @tx_ring: ring to be cleaned
4821 static void igb_clean_tx_ring(struct igb_ring *tx_ring)
4823 u16 i = tx_ring->next_to_clean;
4824 struct igb_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
4826 while (i != tx_ring->next_to_use) {
4827 union e1000_adv_tx_desc *eop_desc, *tx_desc;
4829 /* Free all the Tx ring sk_buffs or xdp frames */
4830 if (tx_buffer->type == IGB_TYPE_SKB)
4831 dev_kfree_skb_any(tx_buffer->skb);
4833 xdp_return_frame(tx_buffer->xdpf);
4835 /* unmap skb header data */
4836 dma_unmap_single(tx_ring->dev,
4837 dma_unmap_addr(tx_buffer, dma),
4838 dma_unmap_len(tx_buffer, len),
4841 /* check for eop_desc to determine the end of the packet */
4842 eop_desc = tx_buffer->next_to_watch;
4843 tx_desc = IGB_TX_DESC(tx_ring, i);
4845 /* unmap remaining buffers */
4846 while (tx_desc != eop_desc) {
4850 if (unlikely(i == tx_ring->count)) {
4852 tx_buffer = tx_ring->tx_buffer_info;
4853 tx_desc = IGB_TX_DESC(tx_ring, 0);
4856 /* unmap any remaining paged data */
4857 if (dma_unmap_len(tx_buffer, len))
4858 dma_unmap_page(tx_ring->dev,
4859 dma_unmap_addr(tx_buffer, dma),
4860 dma_unmap_len(tx_buffer, len),
4864 tx_buffer->next_to_watch = NULL;
4866 /* move us one more past the eop_desc for start of next pkt */
4869 if (unlikely(i == tx_ring->count)) {
4871 tx_buffer = tx_ring->tx_buffer_info;
4875 /* reset BQL for queue */
4876 netdev_tx_reset_queue(txring_txq(tx_ring));
4878 /* reset next_to_use and next_to_clean */
4879 tx_ring->next_to_use = 0;
4880 tx_ring->next_to_clean = 0;
4884 * igb_clean_all_tx_rings - Free Tx Buffers for all queues
4885 * @adapter: board private structure
4887 static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
4891 for (i = 0; i < adapter->num_tx_queues; i++)
4892 if (adapter->tx_ring[i])
4893 igb_clean_tx_ring(adapter->tx_ring[i]);
4897 * igb_free_rx_resources - Free Rx Resources
4898 * @rx_ring: ring to clean the resources from
4900 * Free all receive software resources
4902 void igb_free_rx_resources(struct igb_ring *rx_ring)
4904 igb_clean_rx_ring(rx_ring);
4906 rx_ring->xdp_prog = NULL;
4907 xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
4908 vfree(rx_ring->rx_buffer_info);
4909 rx_ring->rx_buffer_info = NULL;
4911 /* if not set, then don't free */
4915 dma_free_coherent(rx_ring->dev, rx_ring->size,
4916 rx_ring->desc, rx_ring->dma);
4918 rx_ring->desc = NULL;
4922 * igb_free_all_rx_resources - Free Rx Resources for All Queues
4923 * @adapter: board private structure
4925 * Free all receive software resources
4927 static void igb_free_all_rx_resources(struct igb_adapter *adapter)
4931 for (i = 0; i < adapter->num_rx_queues; i++)
4932 if (adapter->rx_ring[i])
4933 igb_free_rx_resources(adapter->rx_ring[i]);
4937 * igb_clean_rx_ring - Free Rx Buffers per Queue
4938 * @rx_ring: ring to free buffers from
4940 static void igb_clean_rx_ring(struct igb_ring *rx_ring)
4942 u16 i = rx_ring->next_to_clean;
4944 dev_kfree_skb(rx_ring->skb);
4945 rx_ring->skb = NULL;
4947 /* Free all the Rx ring sk_buffs */
4948 while (i != rx_ring->next_to_alloc) {
4949 struct igb_rx_buffer *buffer_info = &rx_ring->rx_buffer_info[i];
4951 /* Invalidate cache lines that may have been written to by
4952 * device so that we avoid corrupting memory.
4954 dma_sync_single_range_for_cpu(rx_ring->dev,
4956 buffer_info->page_offset,
4957 igb_rx_bufsz(rx_ring),
4960 /* free resources associated with mapping */
4961 dma_unmap_page_attrs(rx_ring->dev,
4963 igb_rx_pg_size(rx_ring),
4966 __page_frag_cache_drain(buffer_info->page,
4967 buffer_info->pagecnt_bias);
4970 if (i == rx_ring->count)
4974 rx_ring->next_to_alloc = 0;
4975 rx_ring->next_to_clean = 0;
4976 rx_ring->next_to_use = 0;
4980 * igb_clean_all_rx_rings - Free Rx Buffers for all queues
4981 * @adapter: board private structure
4983 static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
4987 for (i = 0; i < adapter->num_rx_queues; i++)
4988 if (adapter->rx_ring[i])
4989 igb_clean_rx_ring(adapter->rx_ring[i]);
4993 * igb_set_mac - Change the Ethernet Address of the NIC
4994 * @netdev: network interface device structure
4995 * @p: pointer to an address structure
4997 * Returns 0 on success, negative on failure
4999 static int igb_set_mac(struct net_device *netdev, void *p)
5001 struct igb_adapter *adapter = netdev_priv(netdev);
5002 struct e1000_hw *hw = &adapter->hw;
5003 struct sockaddr *addr = p;
5005 if (!is_valid_ether_addr(addr->sa_data))
5006 return -EADDRNOTAVAIL;
5008 eth_hw_addr_set(netdev, addr->sa_data);
5009 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
5011 /* set the correct pool for the new PF MAC address in entry 0 */
5012 igb_set_default_mac_filter(adapter);
5018 * igb_write_mc_addr_list - write multicast addresses to MTA
5019 * @netdev: network interface device structure
5021 * Writes multicast address list to the MTA hash table.
5022 * Returns: -ENOMEM on failure
5023 * 0 on no addresses written
5024 * X on writing X addresses to MTA
5026 static int igb_write_mc_addr_list(struct net_device *netdev)
5028 struct igb_adapter *adapter = netdev_priv(netdev);
5029 struct e1000_hw *hw = &adapter->hw;
5030 struct netdev_hw_addr *ha;
5034 if (netdev_mc_empty(netdev)) {
5035 /* nothing to program, so clear mc list */
5036 igb_update_mc_addr_list(hw, NULL, 0);
5037 igb_restore_vf_multicasts(adapter);
5041 mta_list = kcalloc(netdev_mc_count(netdev), 6, GFP_ATOMIC);
5045 /* The shared function expects a packed array of only addresses. */
5047 netdev_for_each_mc_addr(ha, netdev)
5048 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
5050 igb_update_mc_addr_list(hw, mta_list, i);
5053 return netdev_mc_count(netdev);
5056 static int igb_vlan_promisc_enable(struct igb_adapter *adapter)
5058 struct e1000_hw *hw = &adapter->hw;
5061 switch (hw->mac.type) {
5065 /* VLAN filtering needed for VLAN prio filter */
5066 if (adapter->netdev->features & NETIF_F_NTUPLE)
5072 /* VLAN filtering needed for pool filtering */
5073 if (adapter->vfs_allocated_count)
5080 /* We are already in VLAN promisc, nothing to do */
5081 if (adapter->flags & IGB_FLAG_VLAN_PROMISC)
5084 if (!adapter->vfs_allocated_count)
5087 /* Add PF to all active pools */
5088 pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT;
5090 for (i = E1000_VLVF_ARRAY_SIZE; --i;) {
5091 u32 vlvf = rd32(E1000_VLVF(i));
5094 wr32(E1000_VLVF(i), vlvf);
5098 /* Set all bits in the VLAN filter table array */
5099 for (i = E1000_VLAN_FILTER_TBL_SIZE; i--;)
5100 hw->mac.ops.write_vfta(hw, i, ~0U);
5102 /* Set flag so we don't redo unnecessary work */
5103 adapter->flags |= IGB_FLAG_VLAN_PROMISC;
5108 #define VFTA_BLOCK_SIZE 8
5109 static void igb_scrub_vfta(struct igb_adapter *adapter, u32 vfta_offset)
5111 struct e1000_hw *hw = &adapter->hw;
5112 u32 vfta[VFTA_BLOCK_SIZE] = { 0 };
5113 u32 vid_start = vfta_offset * 32;
5114 u32 vid_end = vid_start + (VFTA_BLOCK_SIZE * 32);
5115 u32 i, vid, word, bits, pf_id;
5117 /* guarantee that we don't scrub out management VLAN */
5118 vid = adapter->mng_vlan_id;
5119 if (vid >= vid_start && vid < vid_end)
5120 vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
5122 if (!adapter->vfs_allocated_count)
5125 pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT;
5127 for (i = E1000_VLVF_ARRAY_SIZE; --i;) {
5128 u32 vlvf = rd32(E1000_VLVF(i));
5130 /* pull VLAN ID from VLVF */
5131 vid = vlvf & VLAN_VID_MASK;
5133 /* only concern ourselves with a certain range */
5134 if (vid < vid_start || vid >= vid_end)
5137 if (vlvf & E1000_VLVF_VLANID_ENABLE) {
5138 /* record VLAN ID in VFTA */
5139 vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
5141 /* if PF is part of this then continue */
5142 if (test_bit(vid, adapter->active_vlans))
5146 /* remove PF from the pool */
5148 bits &= rd32(E1000_VLVF(i));
5149 wr32(E1000_VLVF(i), bits);
5153 /* extract values from active_vlans and write back to VFTA */
5154 for (i = VFTA_BLOCK_SIZE; i--;) {
5155 vid = (vfta_offset + i) * 32;
5156 word = vid / BITS_PER_LONG;
5157 bits = vid % BITS_PER_LONG;
5159 vfta[i] |= adapter->active_vlans[word] >> bits;
5161 hw->mac.ops.write_vfta(hw, vfta_offset + i, vfta[i]);
5165 static void igb_vlan_promisc_disable(struct igb_adapter *adapter)
5169 /* We are not in VLAN promisc, nothing to do */
5170 if (!(adapter->flags & IGB_FLAG_VLAN_PROMISC))
5173 /* Set flag so we don't redo unnecessary work */
5174 adapter->flags &= ~IGB_FLAG_VLAN_PROMISC;
5176 for (i = 0; i < E1000_VLAN_FILTER_TBL_SIZE; i += VFTA_BLOCK_SIZE)
5177 igb_scrub_vfta(adapter, i);
5181 * igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
5182 * @netdev: network interface device structure
5184 * The set_rx_mode entry point is called whenever the unicast or multicast
5185 * address lists or the network interface flags are updated. This routine is
5186 * responsible for configuring the hardware for proper unicast, multicast,
5187 * promiscuous mode, and all-multi behavior.
5189 static void igb_set_rx_mode(struct net_device *netdev)
5191 struct igb_adapter *adapter = netdev_priv(netdev);
5192 struct e1000_hw *hw = &adapter->hw;
5193 unsigned int vfn = adapter->vfs_allocated_count;
5194 u32 rctl = 0, vmolr = 0, rlpml = MAX_JUMBO_FRAME_SIZE;
5197 /* Check for Promiscuous and All Multicast modes */
5198 if (netdev->flags & IFF_PROMISC) {
5199 rctl |= E1000_RCTL_UPE | E1000_RCTL_MPE;
5200 vmolr |= E1000_VMOLR_MPME;
5202 /* enable use of UTA filter to force packets to default pool */
5203 if (hw->mac.type == e1000_82576)
5204 vmolr |= E1000_VMOLR_ROPE;
5206 if (netdev->flags & IFF_ALLMULTI) {
5207 rctl |= E1000_RCTL_MPE;
5208 vmolr |= E1000_VMOLR_MPME;
5210 /* Write addresses to the MTA, if the attempt fails
5211 * then we should just turn on promiscuous mode so
5212 * that we can at least receive multicast traffic
5214 count = igb_write_mc_addr_list(netdev);
5216 rctl |= E1000_RCTL_MPE;
5217 vmolr |= E1000_VMOLR_MPME;
5219 vmolr |= E1000_VMOLR_ROMPE;
5224 /* Write addresses to available RAR registers, if there is not
5225 * sufficient space to store all the addresses then enable
5226 * unicast promiscuous mode
5228 if (__dev_uc_sync(netdev, igb_uc_sync, igb_uc_unsync)) {
5229 rctl |= E1000_RCTL_UPE;
5230 vmolr |= E1000_VMOLR_ROPE;
5233 /* enable VLAN filtering by default */
5234 rctl |= E1000_RCTL_VFE;
5236 /* disable VLAN filtering for modes that require it */
5237 if ((netdev->flags & IFF_PROMISC) ||
5238 (netdev->features & NETIF_F_RXALL)) {
5239 /* if we fail to set all rules then just clear VFE */
5240 if (igb_vlan_promisc_enable(adapter))
5241 rctl &= ~E1000_RCTL_VFE;
5243 igb_vlan_promisc_disable(adapter);
5246 /* update state of unicast, multicast, and VLAN filtering modes */
5247 rctl |= rd32(E1000_RCTL) & ~(E1000_RCTL_UPE | E1000_RCTL_MPE |
5249 wr32(E1000_RCTL, rctl);
5251 #if (PAGE_SIZE < 8192)
5252 if (!adapter->vfs_allocated_count) {
5253 if (adapter->max_frame_size <= IGB_MAX_FRAME_BUILD_SKB)
5254 rlpml = IGB_MAX_FRAME_BUILD_SKB;
5257 wr32(E1000_RLPML, rlpml);
5259 /* In order to support SR-IOV and eventually VMDq it is necessary to set
5260 * the VMOLR to enable the appropriate modes. Without this workaround
5261 * we will have issues with VLAN tag stripping not being done for frames
5262 * that are only arriving because we are the default pool
5264 if ((hw->mac.type < e1000_82576) || (hw->mac.type > e1000_i350))
5267 /* set UTA to appropriate mode */
5268 igb_set_uta(adapter, !!(vmolr & E1000_VMOLR_ROPE));
5270 vmolr |= rd32(E1000_VMOLR(vfn)) &
5271 ~(E1000_VMOLR_ROPE | E1000_VMOLR_MPME | E1000_VMOLR_ROMPE);
5273 /* enable Rx jumbo frames, restrict as needed to support build_skb */
5274 vmolr &= ~E1000_VMOLR_RLPML_MASK;
5275 #if (PAGE_SIZE < 8192)
5276 if (adapter->max_frame_size <= IGB_MAX_FRAME_BUILD_SKB)
5277 vmolr |= IGB_MAX_FRAME_BUILD_SKB;
5280 vmolr |= MAX_JUMBO_FRAME_SIZE;
5281 vmolr |= E1000_VMOLR_LPE;
5283 wr32(E1000_VMOLR(vfn), vmolr);
5285 igb_restore_vf_multicasts(adapter);
5288 static void igb_check_wvbr(struct igb_adapter *adapter)
5290 struct e1000_hw *hw = &adapter->hw;
5293 switch (hw->mac.type) {
5296 wvbr = rd32(E1000_WVBR);
5304 adapter->wvbr |= wvbr;
5307 #define IGB_STAGGERED_QUEUE_OFFSET 8
5309 static void igb_spoof_check(struct igb_adapter *adapter)
5316 for (j = 0; j < adapter->vfs_allocated_count; j++) {
5317 if (adapter->wvbr & BIT(j) ||
5318 adapter->wvbr & BIT(j + IGB_STAGGERED_QUEUE_OFFSET)) {
5319 dev_warn(&adapter->pdev->dev,
5320 "Spoof event(s) detected on VF %d\n", j);
5323 BIT(j + IGB_STAGGERED_QUEUE_OFFSET));
5328 /* Need to wait a few seconds after link up to get diagnostic information from
5331 static void igb_update_phy_info(struct timer_list *t)
5333 struct igb_adapter *adapter = from_timer(adapter, t, phy_info_timer);
5334 igb_get_phy_info(&adapter->hw);
5338 * igb_has_link - check shared code for link and determine up/down
5339 * @adapter: pointer to driver private info
5341 bool igb_has_link(struct igb_adapter *adapter)
5343 struct e1000_hw *hw = &adapter->hw;
5344 bool link_active = false;
5346 /* get_link_status is set on LSC (link status) interrupt or
5347 * rx sequence error interrupt. get_link_status will stay
5348 * false until the e1000_check_for_link establishes link
5349 * for copper adapters ONLY
5351 switch (hw->phy.media_type) {
5352 case e1000_media_type_copper:
5353 if (!hw->mac.get_link_status)
5356 case e1000_media_type_internal_serdes:
5357 hw->mac.ops.check_for_link(hw);
5358 link_active = !hw->mac.get_link_status;
5361 case e1000_media_type_unknown:
5365 if (((hw->mac.type == e1000_i210) ||
5366 (hw->mac.type == e1000_i211)) &&
5367 (hw->phy.id == I210_I_PHY_ID)) {
5368 if (!netif_carrier_ok(adapter->netdev)) {
5369 adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
5370 } else if (!(adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)) {
5371 adapter->flags |= IGB_FLAG_NEED_LINK_UPDATE;
5372 adapter->link_check_timeout = jiffies;
5379 static bool igb_thermal_sensor_event(struct e1000_hw *hw, u32 event)
5382 u32 ctrl_ext, thstat;
5384 /* check for thermal sensor event on i350 copper only */
5385 if (hw->mac.type == e1000_i350) {
5386 thstat = rd32(E1000_THSTAT);
5387 ctrl_ext = rd32(E1000_CTRL_EXT);
5389 if ((hw->phy.media_type == e1000_media_type_copper) &&
5390 !(ctrl_ext & E1000_CTRL_EXT_LINK_MODE_SGMII))
5391 ret = !!(thstat & event);
5398 * igb_check_lvmmc - check for malformed packets received
5399 * and indicated in LVMMC register
5400 * @adapter: pointer to adapter
5402 static void igb_check_lvmmc(struct igb_adapter *adapter)
5404 struct e1000_hw *hw = &adapter->hw;
5407 lvmmc = rd32(E1000_LVMMC);
5409 if (unlikely(net_ratelimit())) {
5410 netdev_warn(adapter->netdev,
5411 "malformed Tx packet detected and dropped, LVMMC:0x%08x\n",
5418 * igb_watchdog - Timer Call-back
5419 * @t: pointer to timer_list containing our private info pointer
5421 static void igb_watchdog(struct timer_list *t)
5423 struct igb_adapter *adapter = from_timer(adapter, t, watchdog_timer);
5424 /* Do the rest outside of interrupt context */
5425 schedule_work(&adapter->watchdog_task);
5428 static void igb_watchdog_task(struct work_struct *work)
5430 struct igb_adapter *adapter = container_of(work,
5433 struct e1000_hw *hw = &adapter->hw;
5434 struct e1000_phy_info *phy = &hw->phy;
5435 struct net_device *netdev = adapter->netdev;
5439 u16 phy_data, retry_count = 20;
5441 link = igb_has_link(adapter);
5443 if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE) {
5444 if (time_after(jiffies, (adapter->link_check_timeout + HZ)))
5445 adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
5450 /* Force link down if we have fiber to swap to */
5451 if (adapter->flags & IGB_FLAG_MAS_ENABLE) {
5452 if (hw->phy.media_type == e1000_media_type_copper) {
5453 connsw = rd32(E1000_CONNSW);
5454 if (!(connsw & E1000_CONNSW_AUTOSENSE_EN))
5459 /* Perform a reset if the media type changed. */
5460 if (hw->dev_spec._82575.media_changed) {
5461 hw->dev_spec._82575.media_changed = false;
5462 adapter->flags |= IGB_FLAG_MEDIA_RESET;
5465 /* Cancel scheduled suspend requests. */
5466 pm_runtime_resume(netdev->dev.parent);
5468 if (!netif_carrier_ok(netdev)) {
5471 hw->mac.ops.get_speed_and_duplex(hw,
5472 &adapter->link_speed,
5473 &adapter->link_duplex);
5475 ctrl = rd32(E1000_CTRL);
5476 /* Links status message must follow this format */
5478 "igb: %s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
5480 adapter->link_speed,
5481 adapter->link_duplex == FULL_DUPLEX ?
5483 (ctrl & E1000_CTRL_TFCE) &&
5484 (ctrl & E1000_CTRL_RFCE) ? "RX/TX" :
5485 (ctrl & E1000_CTRL_RFCE) ? "RX" :
5486 (ctrl & E1000_CTRL_TFCE) ? "TX" : "None");
5488 /* disable EEE if enabled */
5489 if ((adapter->flags & IGB_FLAG_EEE) &&
5490 (adapter->link_duplex == HALF_DUPLEX)) {
5491 dev_info(&adapter->pdev->dev,
5492 "EEE Disabled: unsupported at half duplex. Re-enable using ethtool when at full duplex.\n");
5493 adapter->hw.dev_spec._82575.eee_disable = true;
5494 adapter->flags &= ~IGB_FLAG_EEE;
5497 /* check if SmartSpeed worked */
5498 igb_check_downshift(hw);
5499 if (phy->speed_downgraded)
5500 netdev_warn(netdev, "Link Speed was downgraded by SmartSpeed\n");
5502 /* check for thermal sensor event */
5503 if (igb_thermal_sensor_event(hw,
5504 E1000_THSTAT_LINK_THROTTLE))
5505 netdev_info(netdev, "The network adapter link speed was downshifted because it overheated\n");
5507 /* adjust timeout factor according to speed/duplex */
5508 adapter->tx_timeout_factor = 1;
5509 switch (adapter->link_speed) {
5511 adapter->tx_timeout_factor = 14;
5514 /* maybe add some timeout factor ? */
5518 if (adapter->link_speed != SPEED_1000 ||
5519 !hw->phy.ops.read_reg)
5522 /* wait for Remote receiver status OK */
5524 if (!igb_read_phy_reg(hw, PHY_1000T_STATUS,
5526 if (!(phy_data & SR_1000T_REMOTE_RX_STATUS) &&
5530 goto retry_read_status;
5531 } else if (!retry_count) {
5532 dev_err(&adapter->pdev->dev, "exceed max 2 second\n");
5535 dev_err(&adapter->pdev->dev, "read 1000Base-T Status Reg\n");
5538 netif_carrier_on(netdev);
5540 igb_ping_all_vfs(adapter);
5541 igb_check_vf_rate_limit(adapter);
5543 /* link state has changed, schedule phy info update */
5544 if (!test_bit(__IGB_DOWN, &adapter->state))
5545 mod_timer(&adapter->phy_info_timer,
5546 round_jiffies(jiffies + 2 * HZ));
5549 if (netif_carrier_ok(netdev)) {
5550 adapter->link_speed = 0;
5551 adapter->link_duplex = 0;
5553 /* check for thermal sensor event */
5554 if (igb_thermal_sensor_event(hw,
5555 E1000_THSTAT_PWR_DOWN)) {
5556 netdev_err(netdev, "The network adapter was stopped because it overheated\n");
5559 /* Links status message must follow this format */
5560 netdev_info(netdev, "igb: %s NIC Link is Down\n",
5562 netif_carrier_off(netdev);
5564 igb_ping_all_vfs(adapter);
5566 /* link state has changed, schedule phy info update */
5567 if (!test_bit(__IGB_DOWN, &adapter->state))
5568 mod_timer(&adapter->phy_info_timer,
5569 round_jiffies(jiffies + 2 * HZ));
5571 /* link is down, time to check for alternate media */
5572 if (adapter->flags & IGB_FLAG_MAS_ENABLE) {
5573 igb_check_swap_media(adapter);
5574 if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
5575 schedule_work(&adapter->reset_task);
5576 /* return immediately */
5580 pm_schedule_suspend(netdev->dev.parent,
5583 /* also check for alternate media here */
5584 } else if (!netif_carrier_ok(netdev) &&
5585 (adapter->flags & IGB_FLAG_MAS_ENABLE)) {
5586 igb_check_swap_media(adapter);
5587 if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
5588 schedule_work(&adapter->reset_task);
5589 /* return immediately */
5595 spin_lock(&adapter->stats64_lock);
5596 igb_update_stats(adapter);
5597 spin_unlock(&adapter->stats64_lock);
5599 for (i = 0; i < adapter->num_tx_queues; i++) {
5600 struct igb_ring *tx_ring = adapter->tx_ring[i];
5601 if (!netif_carrier_ok(netdev)) {
5602 /* We've lost link, so the controller stops DMA,
5603 * but we've got queued Tx work that's never going
5604 * to get done, so reset controller to flush Tx.
5605 * (Do the reset outside of interrupt context).
5607 if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) {
5608 adapter->tx_timeout_count++;
5609 schedule_work(&adapter->reset_task);
5610 /* return immediately since reset is imminent */
5615 /* Force detection of hung controller every watchdog period */
5616 set_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
5619 /* Cause software interrupt to ensure Rx ring is cleaned */
5620 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
5623 for (i = 0; i < adapter->num_q_vectors; i++)
5624 eics |= adapter->q_vector[i]->eims_value;
5625 wr32(E1000_EICS, eics);
5627 wr32(E1000_ICS, E1000_ICS_RXDMT0);
5630 igb_spoof_check(adapter);
5631 igb_ptp_rx_hang(adapter);
5632 igb_ptp_tx_hang(adapter);
5634 /* Check LVMMC register on i350/i354 only */
5635 if ((adapter->hw.mac.type == e1000_i350) ||
5636 (adapter->hw.mac.type == e1000_i354))
5637 igb_check_lvmmc(adapter);
5639 /* Reset the timer */
5640 if (!test_bit(__IGB_DOWN, &adapter->state)) {
5641 if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)
5642 mod_timer(&adapter->watchdog_timer,
5643 round_jiffies(jiffies + HZ));
5645 mod_timer(&adapter->watchdog_timer,
5646 round_jiffies(jiffies + 2 * HZ));
5650 enum latency_range {
5654 latency_invalid = 255
5658 * igb_update_ring_itr - update the dynamic ITR value based on packet size
5659 * @q_vector: pointer to q_vector
5661 * Stores a new ITR value based on strictly on packet size. This
5662 * algorithm is less sophisticated than that used in igb_update_itr,
5663 * due to the difficulty of synchronizing statistics across multiple
5664 * receive rings. The divisors and thresholds used by this function
5665 * were determined based on theoretical maximum wire speed and testing
5666 * data, in order to minimize response time while increasing bulk
5668 * This functionality is controlled by ethtool's coalescing settings.
5669 * NOTE: This function is called only when operating in a multiqueue
5670 * receive environment.
5672 static void igb_update_ring_itr(struct igb_q_vector *q_vector)
5674 int new_val = q_vector->itr_val;
5675 int avg_wire_size = 0;
5676 struct igb_adapter *adapter = q_vector->adapter;
5677 unsigned int packets;
5679 /* For non-gigabit speeds, just fix the interrupt rate at 4000
5680 * ints/sec - ITR timer value of 120 ticks.
5682 if (adapter->link_speed != SPEED_1000) {
5683 new_val = IGB_4K_ITR;
5687 packets = q_vector->rx.total_packets;
5689 avg_wire_size = q_vector->rx.total_bytes / packets;
5691 packets = q_vector->tx.total_packets;
5693 avg_wire_size = max_t(u32, avg_wire_size,
5694 q_vector->tx.total_bytes / packets);
5696 /* if avg_wire_size isn't set no work was done */
5700 /* Add 24 bytes to size to account for CRC, preamble, and gap */
5701 avg_wire_size += 24;
5703 /* Don't starve jumbo frames */
5704 avg_wire_size = min(avg_wire_size, 3000);
5706 /* Give a little boost to mid-size frames */
5707 if ((avg_wire_size > 300) && (avg_wire_size < 1200))
5708 new_val = avg_wire_size / 3;
5710 new_val = avg_wire_size / 2;
5712 /* conservative mode (itr 3) eliminates the lowest_latency setting */
5713 if (new_val < IGB_20K_ITR &&
5714 ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
5715 (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
5716 new_val = IGB_20K_ITR;
5719 if (new_val != q_vector->itr_val) {
5720 q_vector->itr_val = new_val;
5721 q_vector->set_itr = 1;
5724 q_vector->rx.total_bytes = 0;
5725 q_vector->rx.total_packets = 0;
5726 q_vector->tx.total_bytes = 0;
5727 q_vector->tx.total_packets = 0;
5731 * igb_update_itr - update the dynamic ITR value based on statistics
5732 * @q_vector: pointer to q_vector
5733 * @ring_container: ring info to update the itr for
5735 * Stores a new ITR value based on packets and byte
5736 * counts during the last interrupt. The advantage of per interrupt
5737 * computation is faster updates and more accurate ITR for the current
5738 * traffic pattern. Constants in this function were computed
5739 * based on theoretical maximum wire speed and thresholds were set based
5740 * on testing data as well as attempting to minimize response time
5741 * while increasing bulk throughput.
5742 * This functionality is controlled by ethtool's coalescing settings.
5743 * NOTE: These calculations are only valid when operating in a single-
5744 * queue environment.
5746 static void igb_update_itr(struct igb_q_vector *q_vector,
5747 struct igb_ring_container *ring_container)
5749 unsigned int packets = ring_container->total_packets;
5750 unsigned int bytes = ring_container->total_bytes;
5751 u8 itrval = ring_container->itr;
5753 /* no packets, exit with status unchanged */
5758 case lowest_latency:
5759 /* handle TSO and jumbo frames */
5760 if (bytes/packets > 8000)
5761 itrval = bulk_latency;
5762 else if ((packets < 5) && (bytes > 512))
5763 itrval = low_latency;
5765 case low_latency: /* 50 usec aka 20000 ints/s */
5766 if (bytes > 10000) {
5767 /* this if handles the TSO accounting */
5768 if (bytes/packets > 8000)
5769 itrval = bulk_latency;
5770 else if ((packets < 10) || ((bytes/packets) > 1200))
5771 itrval = bulk_latency;
5772 else if ((packets > 35))
5773 itrval = lowest_latency;
5774 } else if (bytes/packets > 2000) {
5775 itrval = bulk_latency;
5776 } else if (packets <= 2 && bytes < 512) {
5777 itrval = lowest_latency;
5780 case bulk_latency: /* 250 usec aka 4000 ints/s */
5781 if (bytes > 25000) {
5783 itrval = low_latency;
5784 } else if (bytes < 1500) {
5785 itrval = low_latency;
5790 /* clear work counters since we have the values we need */
5791 ring_container->total_bytes = 0;
5792 ring_container->total_packets = 0;
5794 /* write updated itr to ring container */
5795 ring_container->itr = itrval;
5798 static void igb_set_itr(struct igb_q_vector *q_vector)
5800 struct igb_adapter *adapter = q_vector->adapter;
5801 u32 new_itr = q_vector->itr_val;
5804 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
5805 if (adapter->link_speed != SPEED_1000) {
5807 new_itr = IGB_4K_ITR;
5811 igb_update_itr(q_vector, &q_vector->tx);
5812 igb_update_itr(q_vector, &q_vector->rx);
5814 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
5816 /* conservative mode (itr 3) eliminates the lowest_latency setting */
5817 if (current_itr == lowest_latency &&
5818 ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
5819 (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
5820 current_itr = low_latency;
5822 switch (current_itr) {
5823 /* counts and packets in update_itr are dependent on these numbers */
5824 case lowest_latency:
5825 new_itr = IGB_70K_ITR; /* 70,000 ints/sec */
5828 new_itr = IGB_20K_ITR; /* 20,000 ints/sec */
5831 new_itr = IGB_4K_ITR; /* 4,000 ints/sec */
5838 if (new_itr != q_vector->itr_val) {
5839 /* this attempts to bias the interrupt rate towards Bulk
5840 * by adding intermediate steps when interrupt rate is
5843 new_itr = new_itr > q_vector->itr_val ?
5844 max((new_itr * q_vector->itr_val) /
5845 (new_itr + (q_vector->itr_val >> 2)),
5847 /* Don't write the value here; it resets the adapter's
5848 * internal timer, and causes us to delay far longer than
5849 * we should between interrupts. Instead, we write the ITR
5850 * value at the beginning of the next interrupt so the timing
5851 * ends up being correct.
5853 q_vector->itr_val = new_itr;
5854 q_vector->set_itr = 1;
5858 static void igb_tx_ctxtdesc(struct igb_ring *tx_ring,
5859 struct igb_tx_buffer *first,
5860 u32 vlan_macip_lens, u32 type_tucmd,
5863 struct e1000_adv_tx_context_desc *context_desc;
5864 u16 i = tx_ring->next_to_use;
5865 struct timespec64 ts;
5867 context_desc = IGB_TX_CTXTDESC(tx_ring, i);
5870 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
5872 /* set bits to identify this as an advanced context descriptor */
5873 type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
5875 /* For 82575, context index must be unique per ring. */
5876 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
5877 mss_l4len_idx |= tx_ring->reg_idx << 4;
5879 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
5880 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd);
5881 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
5883 /* We assume there is always a valid tx time available. Invalid times
5884 * should have been handled by the upper layers.
5886 if (tx_ring->launchtime_enable) {
5887 ts = ktime_to_timespec64(first->skb->tstamp);
5888 skb_txtime_consumed(first->skb);
5889 context_desc->seqnum_seed = cpu_to_le32(ts.tv_nsec / 32);
5891 context_desc->seqnum_seed = 0;
5895 static int igb_tso(struct igb_ring *tx_ring,
5896 struct igb_tx_buffer *first,
5899 u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
5900 struct sk_buff *skb = first->skb;
5911 u32 paylen, l4_offset;
5914 if (skb->ip_summed != CHECKSUM_PARTIAL)
5917 if (!skb_is_gso(skb))
5920 err = skb_cow_head(skb, 0);
5924 ip.hdr = skb_network_header(skb);
5925 l4.hdr = skb_checksum_start(skb);
5927 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
5928 type_tucmd = (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ?
5929 E1000_ADVTXD_TUCMD_L4T_UDP : E1000_ADVTXD_TUCMD_L4T_TCP;
5931 /* initialize outer IP header fields */
5932 if (ip.v4->version == 4) {
5933 unsigned char *csum_start = skb_checksum_start(skb);
5934 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
5936 /* IP header will have to cancel out any data that
5937 * is not a part of the outer IP header
5939 ip.v4->check = csum_fold(csum_partial(trans_start,
5940 csum_start - trans_start,
5942 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
5945 first->tx_flags |= IGB_TX_FLAGS_TSO |
5949 ip.v6->payload_len = 0;
5950 first->tx_flags |= IGB_TX_FLAGS_TSO |
5954 /* determine offset of inner transport header */
5955 l4_offset = l4.hdr - skb->data;
5957 /* remove payload length from inner checksum */
5958 paylen = skb->len - l4_offset;
5959 if (type_tucmd & E1000_ADVTXD_TUCMD_L4T_TCP) {
5960 /* compute length of segmentation header */
5961 *hdr_len = (l4.tcp->doff * 4) + l4_offset;
5962 csum_replace_by_diff(&l4.tcp->check,
5963 (__force __wsum)htonl(paylen));
5965 /* compute length of segmentation header */
5966 *hdr_len = sizeof(*l4.udp) + l4_offset;
5967 csum_replace_by_diff(&l4.udp->check,
5968 (__force __wsum)htonl(paylen));
5971 /* update gso size and bytecount with header size */
5972 first->gso_segs = skb_shinfo(skb)->gso_segs;
5973 first->bytecount += (first->gso_segs - 1) * *hdr_len;
5976 mss_l4len_idx = (*hdr_len - l4_offset) << E1000_ADVTXD_L4LEN_SHIFT;
5977 mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT;
5979 /* VLAN MACLEN IPLEN */
5980 vlan_macip_lens = l4.hdr - ip.hdr;
5981 vlan_macip_lens |= (ip.hdr - skb->data) << E1000_ADVTXD_MACLEN_SHIFT;
5982 vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
5984 igb_tx_ctxtdesc(tx_ring, first, vlan_macip_lens,
5985 type_tucmd, mss_l4len_idx);
5990 static void igb_tx_csum(struct igb_ring *tx_ring, struct igb_tx_buffer *first)
5992 struct sk_buff *skb = first->skb;
5993 u32 vlan_macip_lens = 0;
5996 if (skb->ip_summed != CHECKSUM_PARTIAL) {
5998 if (!(first->tx_flags & IGB_TX_FLAGS_VLAN) &&
5999 !tx_ring->launchtime_enable)
6004 switch (skb->csum_offset) {
6005 case offsetof(struct tcphdr, check):
6006 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
6008 case offsetof(struct udphdr, check):
6010 case offsetof(struct sctphdr, checksum):
6011 /* validate that this is actually an SCTP request */
6012 if (skb_csum_is_sctp(skb)) {
6013 type_tucmd = E1000_ADVTXD_TUCMD_L4T_SCTP;
6018 skb_checksum_help(skb);
6022 /* update TX checksum flag */
6023 first->tx_flags |= IGB_TX_FLAGS_CSUM;
6024 vlan_macip_lens = skb_checksum_start_offset(skb) -
6025 skb_network_offset(skb);
6027 vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
6028 vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
6030 igb_tx_ctxtdesc(tx_ring, first, vlan_macip_lens, type_tucmd, 0);
6033 #define IGB_SET_FLAG(_input, _flag, _result) \
6034 ((_flag <= _result) ? \
6035 ((u32)(_input & _flag) * (_result / _flag)) : \
6036 ((u32)(_input & _flag) / (_flag / _result)))
6038 static u32 igb_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
6040 /* set type for advanced descriptor with frame checksum insertion */
6041 u32 cmd_type = E1000_ADVTXD_DTYP_DATA |
6042 E1000_ADVTXD_DCMD_DEXT |
6043 E1000_ADVTXD_DCMD_IFCS;
6045 /* set HW vlan bit if vlan is present */
6046 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_VLAN,
6047 (E1000_ADVTXD_DCMD_VLE));
6049 /* set segmentation bits for TSO */
6050 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSO,
6051 (E1000_ADVTXD_DCMD_TSE));
6053 /* set timestamp bit if present */
6054 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSTAMP,
6055 (E1000_ADVTXD_MAC_TSTAMP));
6057 /* insert frame checksum */
6058 cmd_type ^= IGB_SET_FLAG(skb->no_fcs, 1, E1000_ADVTXD_DCMD_IFCS);
6063 static void igb_tx_olinfo_status(struct igb_ring *tx_ring,
6064 union e1000_adv_tx_desc *tx_desc,
6065 u32 tx_flags, unsigned int paylen)
6067 u32 olinfo_status = paylen << E1000_ADVTXD_PAYLEN_SHIFT;
6069 /* 82575 requires a unique index per ring */
6070 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
6071 olinfo_status |= tx_ring->reg_idx << 4;
6073 /* insert L4 checksum */
6074 olinfo_status |= IGB_SET_FLAG(tx_flags,
6076 (E1000_TXD_POPTS_TXSM << 8));
6078 /* insert IPv4 checksum */
6079 olinfo_status |= IGB_SET_FLAG(tx_flags,
6081 (E1000_TXD_POPTS_IXSM << 8));
6083 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
6086 static int __igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
6088 struct net_device *netdev = tx_ring->netdev;
6090 netif_stop_subqueue(netdev, tx_ring->queue_index);
6092 /* Herbert's original patch had:
6093 * smp_mb__after_netif_stop_queue();
6094 * but since that doesn't exist yet, just open code it.
6098 /* We need to check again in a case another CPU has just
6099 * made room available.
6101 if (igb_desc_unused(tx_ring) < size)
6105 netif_wake_subqueue(netdev, tx_ring->queue_index);
6107 u64_stats_update_begin(&tx_ring->tx_syncp2);
6108 tx_ring->tx_stats.restart_queue2++;
6109 u64_stats_update_end(&tx_ring->tx_syncp2);
6114 static inline int igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
6116 if (igb_desc_unused(tx_ring) >= size)
6118 return __igb_maybe_stop_tx(tx_ring, size);
6121 static int igb_tx_map(struct igb_ring *tx_ring,
6122 struct igb_tx_buffer *first,
6125 struct sk_buff *skb = first->skb;
6126 struct igb_tx_buffer *tx_buffer;
6127 union e1000_adv_tx_desc *tx_desc;
6130 unsigned int data_len, size;
6131 u32 tx_flags = first->tx_flags;
6132 u32 cmd_type = igb_tx_cmd_type(skb, tx_flags);
6133 u16 i = tx_ring->next_to_use;
6135 tx_desc = IGB_TX_DESC(tx_ring, i);
6137 igb_tx_olinfo_status(tx_ring, tx_desc, tx_flags, skb->len - hdr_len);
6139 size = skb_headlen(skb);
6140 data_len = skb->data_len;
6142 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
6146 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
6147 if (dma_mapping_error(tx_ring->dev, dma))
6150 /* record length, and DMA address */
6151 dma_unmap_len_set(tx_buffer, len, size);
6152 dma_unmap_addr_set(tx_buffer, dma, dma);
6154 tx_desc->read.buffer_addr = cpu_to_le64(dma);
6156 while (unlikely(size > IGB_MAX_DATA_PER_TXD)) {
6157 tx_desc->read.cmd_type_len =
6158 cpu_to_le32(cmd_type ^ IGB_MAX_DATA_PER_TXD);
6162 if (i == tx_ring->count) {
6163 tx_desc = IGB_TX_DESC(tx_ring, 0);
6166 tx_desc->read.olinfo_status = 0;
6168 dma += IGB_MAX_DATA_PER_TXD;
6169 size -= IGB_MAX_DATA_PER_TXD;
6171 tx_desc->read.buffer_addr = cpu_to_le64(dma);
6174 if (likely(!data_len))
6177 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
6181 if (i == tx_ring->count) {
6182 tx_desc = IGB_TX_DESC(tx_ring, 0);
6185 tx_desc->read.olinfo_status = 0;
6187 size = skb_frag_size(frag);
6190 dma = skb_frag_dma_map(tx_ring->dev, frag, 0,
6191 size, DMA_TO_DEVICE);
6193 tx_buffer = &tx_ring->tx_buffer_info[i];
6196 /* write last descriptor with RS and EOP bits */
6197 cmd_type |= size | IGB_TXD_DCMD;
6198 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
6200 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
6202 /* set the timestamp */
6203 first->time_stamp = jiffies;
6205 skb_tx_timestamp(skb);
6207 /* Force memory writes to complete before letting h/w know there
6208 * are new descriptors to fetch. (Only applicable for weak-ordered
6209 * memory model archs, such as IA-64).
6211 * We also need this memory barrier to make certain all of the
6212 * status bits have been updated before next_to_watch is written.
6216 /* set next_to_watch value indicating a packet is present */
6217 first->next_to_watch = tx_desc;
6220 if (i == tx_ring->count)
6223 tx_ring->next_to_use = i;
6225 /* Make sure there is space in the ring for the next send. */
6226 igb_maybe_stop_tx(tx_ring, DESC_NEEDED);
6228 if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more()) {
6229 writel(i, tx_ring->tail);
6234 dev_err(tx_ring->dev, "TX DMA map failed\n");
6235 tx_buffer = &tx_ring->tx_buffer_info[i];
6237 /* clear dma mappings for failed tx_buffer_info map */
6238 while (tx_buffer != first) {
6239 if (dma_unmap_len(tx_buffer, len))
6240 dma_unmap_page(tx_ring->dev,
6241 dma_unmap_addr(tx_buffer, dma),
6242 dma_unmap_len(tx_buffer, len),
6244 dma_unmap_len_set(tx_buffer, len, 0);
6247 i += tx_ring->count;
6248 tx_buffer = &tx_ring->tx_buffer_info[i];
6251 if (dma_unmap_len(tx_buffer, len))
6252 dma_unmap_single(tx_ring->dev,
6253 dma_unmap_addr(tx_buffer, dma),
6254 dma_unmap_len(tx_buffer, len),
6256 dma_unmap_len_set(tx_buffer, len, 0);
6258 dev_kfree_skb_any(tx_buffer->skb);
6259 tx_buffer->skb = NULL;
6261 tx_ring->next_to_use = i;
6266 int igb_xmit_xdp_ring(struct igb_adapter *adapter,
6267 struct igb_ring *tx_ring,
6268 struct xdp_frame *xdpf)
6270 struct skb_shared_info *sinfo = xdp_get_shared_info_from_frame(xdpf);
6271 u8 nr_frags = unlikely(xdp_frame_has_frags(xdpf)) ? sinfo->nr_frags : 0;
6272 u16 count, i, index = tx_ring->next_to_use;
6273 struct igb_tx_buffer *tx_head = &tx_ring->tx_buffer_info[index];
6274 struct igb_tx_buffer *tx_buffer = tx_head;
6275 union e1000_adv_tx_desc *tx_desc = IGB_TX_DESC(tx_ring, index);
6276 u32 len = xdpf->len, cmd_type, olinfo_status;
6277 void *data = xdpf->data;
6279 count = TXD_USE_COUNT(len);
6280 for (i = 0; i < nr_frags; i++)
6281 count += TXD_USE_COUNT(skb_frag_size(&sinfo->frags[i]));
6283 if (igb_maybe_stop_tx(tx_ring, count + 3))
6284 return IGB_XDP_CONSUMED;
6287 /* record the location of the first descriptor for this packet */
6288 tx_head->bytecount = xdp_get_frame_len(xdpf);
6289 tx_head->type = IGB_TYPE_XDP;
6290 tx_head->gso_segs = 1;
6291 tx_head->xdpf = xdpf;
6293 olinfo_status = tx_head->bytecount << E1000_ADVTXD_PAYLEN_SHIFT;
6294 /* 82575 requires a unique index per ring */
6295 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
6296 olinfo_status |= tx_ring->reg_idx << 4;
6297 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
6302 dma = dma_map_single(tx_ring->dev, data, len, DMA_TO_DEVICE);
6303 if (dma_mapping_error(tx_ring->dev, dma))
6306 /* record length, and DMA address */
6307 dma_unmap_len_set(tx_buffer, len, len);
6308 dma_unmap_addr_set(tx_buffer, dma, dma);
6310 /* put descriptor type bits */
6311 cmd_type = E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_DEXT |
6312 E1000_ADVTXD_DCMD_IFCS | len;
6314 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
6315 tx_desc->read.buffer_addr = cpu_to_le64(dma);
6317 tx_buffer->protocol = 0;
6319 if (++index == tx_ring->count)
6325 tx_buffer = &tx_ring->tx_buffer_info[index];
6326 tx_desc = IGB_TX_DESC(tx_ring, index);
6327 tx_desc->read.olinfo_status = 0;
6329 data = skb_frag_address(&sinfo->frags[i]);
6330 len = skb_frag_size(&sinfo->frags[i]);
6333 tx_desc->read.cmd_type_len |= cpu_to_le32(IGB_TXD_DCMD);
6335 netdev_tx_sent_queue(txring_txq(tx_ring), tx_head->bytecount);
6336 /* set the timestamp */
6337 tx_head->time_stamp = jiffies;
6339 /* Avoid any potential race with xdp_xmit and cleanup */
6342 /* set next_to_watch value indicating a packet is present */
6343 tx_head->next_to_watch = tx_desc;
6344 tx_ring->next_to_use = index;
6346 /* Make sure there is space in the ring for the next send. */
6347 igb_maybe_stop_tx(tx_ring, DESC_NEEDED);
6349 if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more())
6350 writel(index, tx_ring->tail);
6356 tx_buffer = &tx_ring->tx_buffer_info[index];
6357 if (dma_unmap_len(tx_buffer, len))
6358 dma_unmap_page(tx_ring->dev,
6359 dma_unmap_addr(tx_buffer, dma),
6360 dma_unmap_len(tx_buffer, len),
6362 dma_unmap_len_set(tx_buffer, len, 0);
6363 if (tx_buffer == tx_head)
6367 index += tx_ring->count;
6371 return IGB_XDP_CONSUMED;
6374 netdev_tx_t igb_xmit_frame_ring(struct sk_buff *skb,
6375 struct igb_ring *tx_ring)
6377 struct igb_tx_buffer *first;
6381 u16 count = TXD_USE_COUNT(skb_headlen(skb));
6382 __be16 protocol = vlan_get_protocol(skb);
6385 /* need: 1 descriptor per page * PAGE_SIZE/IGB_MAX_DATA_PER_TXD,
6386 * + 1 desc for skb_headlen/IGB_MAX_DATA_PER_TXD,
6387 * + 2 desc gap to keep tail from touching head,
6388 * + 1 desc for context descriptor,
6389 * otherwise try next time
6391 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6392 count += TXD_USE_COUNT(skb_frag_size(
6393 &skb_shinfo(skb)->frags[f]));
6395 if (igb_maybe_stop_tx(tx_ring, count + 3)) {
6396 /* this is a hard error */
6397 return NETDEV_TX_BUSY;
6400 /* record the location of the first descriptor for this packet */
6401 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
6402 first->type = IGB_TYPE_SKB;
6404 first->bytecount = skb->len;
6405 first->gso_segs = 1;
6407 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
6408 struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
6410 if (adapter->tstamp_config.tx_type == HWTSTAMP_TX_ON &&
6411 !test_and_set_bit_lock(__IGB_PTP_TX_IN_PROGRESS,
6413 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
6414 tx_flags |= IGB_TX_FLAGS_TSTAMP;
6416 adapter->ptp_tx_skb = skb_get(skb);
6417 adapter->ptp_tx_start = jiffies;
6418 if (adapter->hw.mac.type == e1000_82576)
6419 schedule_work(&adapter->ptp_tx_work);
6421 adapter->tx_hwtstamp_skipped++;
6425 if (skb_vlan_tag_present(skb)) {
6426 tx_flags |= IGB_TX_FLAGS_VLAN;
6427 tx_flags |= (skb_vlan_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
6430 /* record initial flags and protocol */
6431 first->tx_flags = tx_flags;
6432 first->protocol = protocol;
6434 tso = igb_tso(tx_ring, first, &hdr_len);
6438 igb_tx_csum(tx_ring, first);
6440 if (igb_tx_map(tx_ring, first, hdr_len))
6441 goto cleanup_tx_tstamp;
6443 return NETDEV_TX_OK;
6446 dev_kfree_skb_any(first->skb);
6449 if (unlikely(tx_flags & IGB_TX_FLAGS_TSTAMP)) {
6450 struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
6452 dev_kfree_skb_any(adapter->ptp_tx_skb);
6453 adapter->ptp_tx_skb = NULL;
6454 if (adapter->hw.mac.type == e1000_82576)
6455 cancel_work_sync(&adapter->ptp_tx_work);
6456 clear_bit_unlock(__IGB_PTP_TX_IN_PROGRESS, &adapter->state);
6459 return NETDEV_TX_OK;
6462 static inline struct igb_ring *igb_tx_queue_mapping(struct igb_adapter *adapter,
6463 struct sk_buff *skb)
6465 unsigned int r_idx = skb->queue_mapping;
6467 if (r_idx >= adapter->num_tx_queues)
6468 r_idx = r_idx % adapter->num_tx_queues;
6470 return adapter->tx_ring[r_idx];
6473 static netdev_tx_t igb_xmit_frame(struct sk_buff *skb,
6474 struct net_device *netdev)
6476 struct igb_adapter *adapter = netdev_priv(netdev);
6478 /* The minimum packet size with TCTL.PSP set is 17 so pad the skb
6479 * in order to meet this minimum size requirement.
6481 if (skb_put_padto(skb, 17))
6482 return NETDEV_TX_OK;
6484 return igb_xmit_frame_ring(skb, igb_tx_queue_mapping(adapter, skb));
6488 * igb_tx_timeout - Respond to a Tx Hang
6489 * @netdev: network interface device structure
6490 * @txqueue: number of the Tx queue that hung (unused)
6492 static void igb_tx_timeout(struct net_device *netdev, unsigned int __always_unused txqueue)
6494 struct igb_adapter *adapter = netdev_priv(netdev);
6495 struct e1000_hw *hw = &adapter->hw;
6497 /* Do the reset outside of interrupt context */
6498 adapter->tx_timeout_count++;
6500 if (hw->mac.type >= e1000_82580)
6501 hw->dev_spec._82575.global_device_reset = true;
6503 schedule_work(&adapter->reset_task);
6505 (adapter->eims_enable_mask & ~adapter->eims_other));
6508 static void igb_reset_task(struct work_struct *work)
6510 struct igb_adapter *adapter;
6511 adapter = container_of(work, struct igb_adapter, reset_task);
6514 /* If we're already down or resetting, just bail */
6515 if (test_bit(__IGB_DOWN, &adapter->state) ||
6516 test_bit(__IGB_RESETTING, &adapter->state)) {
6522 netdev_err(adapter->netdev, "Reset adapter\n");
6523 igb_reinit_locked(adapter);
6528 * igb_get_stats64 - Get System Network Statistics
6529 * @netdev: network interface device structure
6530 * @stats: rtnl_link_stats64 pointer
6532 static void igb_get_stats64(struct net_device *netdev,
6533 struct rtnl_link_stats64 *stats)
6535 struct igb_adapter *adapter = netdev_priv(netdev);
6537 spin_lock(&adapter->stats64_lock);
6538 igb_update_stats(adapter);
6539 memcpy(stats, &adapter->stats64, sizeof(*stats));
6540 spin_unlock(&adapter->stats64_lock);
6544 * igb_change_mtu - Change the Maximum Transfer Unit
6545 * @netdev: network interface device structure
6546 * @new_mtu: new value for maximum frame size
6548 * Returns 0 on success, negative on failure
6550 static int igb_change_mtu(struct net_device *netdev, int new_mtu)
6552 struct igb_adapter *adapter = netdev_priv(netdev);
6553 int max_frame = new_mtu + IGB_ETH_PKT_HDR_PAD;
6555 if (adapter->xdp_prog) {
6558 for (i = 0; i < adapter->num_rx_queues; i++) {
6559 struct igb_ring *ring = adapter->rx_ring[i];
6561 if (max_frame > igb_rx_bufsz(ring)) {
6562 netdev_warn(adapter->netdev,
6563 "Requested MTU size is not supported with XDP. Max frame size is %d\n",
6570 /* adjust max frame to be at least the size of a standard frame */
6571 if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
6572 max_frame = ETH_FRAME_LEN + ETH_FCS_LEN;
6574 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
6575 usleep_range(1000, 2000);
6577 /* igb_down has a dependency on max_frame_size */
6578 adapter->max_frame_size = max_frame;
6580 if (netif_running(netdev))
6583 netdev_dbg(netdev, "changing MTU from %d to %d\n",
6584 netdev->mtu, new_mtu);
6585 netdev->mtu = new_mtu;
6587 if (netif_running(netdev))
6592 clear_bit(__IGB_RESETTING, &adapter->state);
6598 * igb_update_stats - Update the board statistics counters
6599 * @adapter: board private structure
6601 void igb_update_stats(struct igb_adapter *adapter)
6603 struct rtnl_link_stats64 *net_stats = &adapter->stats64;
6604 struct e1000_hw *hw = &adapter->hw;
6605 struct pci_dev *pdev = adapter->pdev;
6610 u64 _bytes, _packets;
6612 /* Prevent stats update while adapter is being reset, or if the pci
6613 * connection is down.
6615 if (adapter->link_speed == 0)
6617 if (pci_channel_offline(pdev))
6624 for (i = 0; i < adapter->num_rx_queues; i++) {
6625 struct igb_ring *ring = adapter->rx_ring[i];
6626 u32 rqdpc = rd32(E1000_RQDPC(i));
6627 if (hw->mac.type >= e1000_i210)
6628 wr32(E1000_RQDPC(i), 0);
6631 ring->rx_stats.drops += rqdpc;
6632 net_stats->rx_fifo_errors += rqdpc;
6636 start = u64_stats_fetch_begin_irq(&ring->rx_syncp);
6637 _bytes = ring->rx_stats.bytes;
6638 _packets = ring->rx_stats.packets;
6639 } while (u64_stats_fetch_retry_irq(&ring->rx_syncp, start));
6641 packets += _packets;
6644 net_stats->rx_bytes = bytes;
6645 net_stats->rx_packets = packets;
6649 for (i = 0; i < adapter->num_tx_queues; i++) {
6650 struct igb_ring *ring = adapter->tx_ring[i];
6652 start = u64_stats_fetch_begin_irq(&ring->tx_syncp);
6653 _bytes = ring->tx_stats.bytes;
6654 _packets = ring->tx_stats.packets;
6655 } while (u64_stats_fetch_retry_irq(&ring->tx_syncp, start));
6657 packets += _packets;
6659 net_stats->tx_bytes = bytes;
6660 net_stats->tx_packets = packets;
6663 /* read stats registers */
6664 adapter->stats.crcerrs += rd32(E1000_CRCERRS);
6665 adapter->stats.gprc += rd32(E1000_GPRC);
6666 adapter->stats.gorc += rd32(E1000_GORCL);
6667 rd32(E1000_GORCH); /* clear GORCL */
6668 adapter->stats.bprc += rd32(E1000_BPRC);
6669 adapter->stats.mprc += rd32(E1000_MPRC);
6670 adapter->stats.roc += rd32(E1000_ROC);
6672 adapter->stats.prc64 += rd32(E1000_PRC64);
6673 adapter->stats.prc127 += rd32(E1000_PRC127);
6674 adapter->stats.prc255 += rd32(E1000_PRC255);
6675 adapter->stats.prc511 += rd32(E1000_PRC511);
6676 adapter->stats.prc1023 += rd32(E1000_PRC1023);
6677 adapter->stats.prc1522 += rd32(E1000_PRC1522);
6678 adapter->stats.symerrs += rd32(E1000_SYMERRS);
6679 adapter->stats.sec += rd32(E1000_SEC);
6681 mpc = rd32(E1000_MPC);
6682 adapter->stats.mpc += mpc;
6683 net_stats->rx_fifo_errors += mpc;
6684 adapter->stats.scc += rd32(E1000_SCC);
6685 adapter->stats.ecol += rd32(E1000_ECOL);
6686 adapter->stats.mcc += rd32(E1000_MCC);
6687 adapter->stats.latecol += rd32(E1000_LATECOL);
6688 adapter->stats.dc += rd32(E1000_DC);
6689 adapter->stats.rlec += rd32(E1000_RLEC);
6690 adapter->stats.xonrxc += rd32(E1000_XONRXC);
6691 adapter->stats.xontxc += rd32(E1000_XONTXC);
6692 adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
6693 adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
6694 adapter->stats.fcruc += rd32(E1000_FCRUC);
6695 adapter->stats.gptc += rd32(E1000_GPTC);
6696 adapter->stats.gotc += rd32(E1000_GOTCL);
6697 rd32(E1000_GOTCH); /* clear GOTCL */
6698 adapter->stats.rnbc += rd32(E1000_RNBC);
6699 adapter->stats.ruc += rd32(E1000_RUC);
6700 adapter->stats.rfc += rd32(E1000_RFC);
6701 adapter->stats.rjc += rd32(E1000_RJC);
6702 adapter->stats.tor += rd32(E1000_TORH);
6703 adapter->stats.tot += rd32(E1000_TOTH);
6704 adapter->stats.tpr += rd32(E1000_TPR);
6706 adapter->stats.ptc64 += rd32(E1000_PTC64);
6707 adapter->stats.ptc127 += rd32(E1000_PTC127);
6708 adapter->stats.ptc255 += rd32(E1000_PTC255);
6709 adapter->stats.ptc511 += rd32(E1000_PTC511);
6710 adapter->stats.ptc1023 += rd32(E1000_PTC1023);
6711 adapter->stats.ptc1522 += rd32(E1000_PTC1522);
6713 adapter->stats.mptc += rd32(E1000_MPTC);
6714 adapter->stats.bptc += rd32(E1000_BPTC);
6716 adapter->stats.tpt += rd32(E1000_TPT);
6717 adapter->stats.colc += rd32(E1000_COLC);
6719 adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
6720 /* read internal phy specific stats */
6721 reg = rd32(E1000_CTRL_EXT);
6722 if (!(reg & E1000_CTRL_EXT_LINK_MODE_MASK)) {
6723 adapter->stats.rxerrc += rd32(E1000_RXERRC);
6725 /* this stat has invalid values on i210/i211 */
6726 if ((hw->mac.type != e1000_i210) &&
6727 (hw->mac.type != e1000_i211))
6728 adapter->stats.tncrs += rd32(E1000_TNCRS);
6731 adapter->stats.tsctc += rd32(E1000_TSCTC);
6732 adapter->stats.tsctfc += rd32(E1000_TSCTFC);
6734 adapter->stats.iac += rd32(E1000_IAC);
6735 adapter->stats.icrxoc += rd32(E1000_ICRXOC);
6736 adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
6737 adapter->stats.icrxatc += rd32(E1000_ICRXATC);
6738 adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
6739 adapter->stats.ictxatc += rd32(E1000_ICTXATC);
6740 adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
6741 adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
6742 adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
6744 /* Fill out the OS statistics structure */
6745 net_stats->multicast = adapter->stats.mprc;
6746 net_stats->collisions = adapter->stats.colc;
6750 /* RLEC on some newer hardware can be incorrect so build
6751 * our own version based on RUC and ROC
6753 net_stats->rx_errors = adapter->stats.rxerrc +
6754 adapter->stats.crcerrs + adapter->stats.algnerrc +
6755 adapter->stats.ruc + adapter->stats.roc +
6756 adapter->stats.cexterr;
6757 net_stats->rx_length_errors = adapter->stats.ruc +
6759 net_stats->rx_crc_errors = adapter->stats.crcerrs;
6760 net_stats->rx_frame_errors = adapter->stats.algnerrc;
6761 net_stats->rx_missed_errors = adapter->stats.mpc;
6764 net_stats->tx_errors = adapter->stats.ecol +
6765 adapter->stats.latecol;
6766 net_stats->tx_aborted_errors = adapter->stats.ecol;
6767 net_stats->tx_window_errors = adapter->stats.latecol;
6768 net_stats->tx_carrier_errors = adapter->stats.tncrs;
6770 /* Tx Dropped needs to be maintained elsewhere */
6772 /* Management Stats */
6773 adapter->stats.mgptc += rd32(E1000_MGTPTC);
6774 adapter->stats.mgprc += rd32(E1000_MGTPRC);
6775 adapter->stats.mgpdc += rd32(E1000_MGTPDC);
6778 reg = rd32(E1000_MANC);
6779 if (reg & E1000_MANC_EN_BMC2OS) {
6780 adapter->stats.o2bgptc += rd32(E1000_O2BGPTC);
6781 adapter->stats.o2bspc += rd32(E1000_O2BSPC);
6782 adapter->stats.b2ospc += rd32(E1000_B2OSPC);
6783 adapter->stats.b2ogprc += rd32(E1000_B2OGPRC);
6787 static void igb_perout(struct igb_adapter *adapter, int tsintr_tt)
6789 int pin = ptp_find_pin(adapter->ptp_clock, PTP_PF_PEROUT, tsintr_tt);
6790 struct e1000_hw *hw = &adapter->hw;
6791 struct timespec64 ts;
6794 if (pin < 0 || pin >= IGB_N_PEROUT)
6797 spin_lock(&adapter->tmreg_lock);
6799 if (hw->mac.type == e1000_82580 ||
6800 hw->mac.type == e1000_i354 ||
6801 hw->mac.type == e1000_i350) {
6802 s64 ns = timespec64_to_ns(&adapter->perout[pin].period);
6803 u32 systiml, systimh, level_mask, level, rem;
6806 /* read systim registers in sequence */
6807 rd32(E1000_SYSTIMR);
6808 systiml = rd32(E1000_SYSTIML);
6809 systimh = rd32(E1000_SYSTIMH);
6810 systim = (((u64)(systimh & 0xFF)) << 32) | ((u64)systiml);
6811 now = timecounter_cyc2time(&adapter->tc, systim);
6814 level_mask = (tsintr_tt == 1) ? 0x80000 : 0x40000;
6815 level = (rd32(E1000_CTRL) & level_mask) ? 1 : 0;
6817 level_mask = (tsintr_tt == 1) ? 0x80 : 0x40;
6818 level = (rd32(E1000_CTRL_EXT) & level_mask) ? 1 : 0;
6821 div_u64_rem(now, ns, &rem);
6822 systim = systim + (ns - rem);
6824 /* synchronize pin level with rising/falling edges */
6825 div_u64_rem(now, ns << 1, &rem);
6827 /* first half of period */
6829 /* output is already low, skip this period */
6831 pr_notice("igb: periodic output on %s missed falling edge\n",
6832 adapter->sdp_config[pin].name);
6835 /* second half of period */
6837 /* output is already high, skip this period */
6839 pr_notice("igb: periodic output on %s missed rising edge\n",
6840 adapter->sdp_config[pin].name);
6844 /* for this chip family tv_sec is the upper part of the binary value,
6847 ts.tv_nsec = (u32)systim;
6848 ts.tv_sec = ((u32)(systim >> 32)) & 0xFF;
6850 ts = timespec64_add(adapter->perout[pin].start,
6851 adapter->perout[pin].period);
6854 /* u32 conversion of tv_sec is safe until y2106 */
6855 wr32((tsintr_tt == 1) ? E1000_TRGTTIML1 : E1000_TRGTTIML0, ts.tv_nsec);
6856 wr32((tsintr_tt == 1) ? E1000_TRGTTIMH1 : E1000_TRGTTIMH0, (u32)ts.tv_sec);
6857 tsauxc = rd32(E1000_TSAUXC);
6858 tsauxc |= TSAUXC_EN_TT0;
6859 wr32(E1000_TSAUXC, tsauxc);
6860 adapter->perout[pin].start = ts;
6862 spin_unlock(&adapter->tmreg_lock);
6865 static void igb_extts(struct igb_adapter *adapter, int tsintr_tt)
6867 int pin = ptp_find_pin(adapter->ptp_clock, PTP_PF_EXTTS, tsintr_tt);
6868 int auxstmpl = (tsintr_tt == 1) ? E1000_AUXSTMPL1 : E1000_AUXSTMPL0;
6869 int auxstmph = (tsintr_tt == 1) ? E1000_AUXSTMPH1 : E1000_AUXSTMPH0;
6870 struct e1000_hw *hw = &adapter->hw;
6871 struct ptp_clock_event event;
6872 struct timespec64 ts;
6874 if (pin < 0 || pin >= IGB_N_EXTTS)
6877 if (hw->mac.type == e1000_82580 ||
6878 hw->mac.type == e1000_i354 ||
6879 hw->mac.type == e1000_i350) {
6880 s64 ns = rd32(auxstmpl);
6882 ns += ((s64)(rd32(auxstmph) & 0xFF)) << 32;
6883 ts = ns_to_timespec64(ns);
6885 ts.tv_nsec = rd32(auxstmpl);
6886 ts.tv_sec = rd32(auxstmph);
6889 event.type = PTP_CLOCK_EXTTS;
6890 event.index = tsintr_tt;
6891 event.timestamp = ts.tv_sec * 1000000000ULL + ts.tv_nsec;
6892 ptp_clock_event(adapter->ptp_clock, &event);
6895 static void igb_tsync_interrupt(struct igb_adapter *adapter)
6897 struct e1000_hw *hw = &adapter->hw;
6898 u32 ack = 0, tsicr = rd32(E1000_TSICR);
6899 struct ptp_clock_event event;
6901 if (tsicr & TSINTR_SYS_WRAP) {
6902 event.type = PTP_CLOCK_PPS;
6903 if (adapter->ptp_caps.pps)
6904 ptp_clock_event(adapter->ptp_clock, &event);
6905 ack |= TSINTR_SYS_WRAP;
6908 if (tsicr & E1000_TSICR_TXTS) {
6909 /* retrieve hardware timestamp */
6910 schedule_work(&adapter->ptp_tx_work);
6911 ack |= E1000_TSICR_TXTS;
6914 if (tsicr & TSINTR_TT0) {
6915 igb_perout(adapter, 0);
6919 if (tsicr & TSINTR_TT1) {
6920 igb_perout(adapter, 1);
6924 if (tsicr & TSINTR_AUTT0) {
6925 igb_extts(adapter, 0);
6926 ack |= TSINTR_AUTT0;
6929 if (tsicr & TSINTR_AUTT1) {
6930 igb_extts(adapter, 1);
6931 ack |= TSINTR_AUTT1;
6934 /* acknowledge the interrupts */
6935 wr32(E1000_TSICR, ack);
6938 static irqreturn_t igb_msix_other(int irq, void *data)
6940 struct igb_adapter *adapter = data;
6941 struct e1000_hw *hw = &adapter->hw;
6942 u32 icr = rd32(E1000_ICR);
6943 /* reading ICR causes bit 31 of EICR to be cleared */
6945 if (icr & E1000_ICR_DRSTA)
6946 schedule_work(&adapter->reset_task);
6948 if (icr & E1000_ICR_DOUTSYNC) {
6949 /* HW is reporting DMA is out of sync */
6950 adapter->stats.doosync++;
6951 /* The DMA Out of Sync is also indication of a spoof event
6952 * in IOV mode. Check the Wrong VM Behavior register to
6953 * see if it is really a spoof event.
6955 igb_check_wvbr(adapter);
6958 /* Check for a mailbox event */
6959 if (icr & E1000_ICR_VMMB)
6960 igb_msg_task(adapter);
6962 if (icr & E1000_ICR_LSC) {
6963 hw->mac.get_link_status = 1;
6964 /* guard against interrupt when we're going down */
6965 if (!test_bit(__IGB_DOWN, &adapter->state))
6966 mod_timer(&adapter->watchdog_timer, jiffies + 1);
6969 if (icr & E1000_ICR_TS)
6970 igb_tsync_interrupt(adapter);
6972 wr32(E1000_EIMS, adapter->eims_other);
6977 static void igb_write_itr(struct igb_q_vector *q_vector)
6979 struct igb_adapter *adapter = q_vector->adapter;
6980 u32 itr_val = q_vector->itr_val & 0x7FFC;
6982 if (!q_vector->set_itr)
6988 if (adapter->hw.mac.type == e1000_82575)
6989 itr_val |= itr_val << 16;
6991 itr_val |= E1000_EITR_CNT_IGNR;
6993 writel(itr_val, q_vector->itr_register);
6994 q_vector->set_itr = 0;
6997 static irqreturn_t igb_msix_ring(int irq, void *data)
6999 struct igb_q_vector *q_vector = data;
7001 /* Write the ITR value calculated from the previous interrupt. */
7002 igb_write_itr(q_vector);
7004 napi_schedule(&q_vector->napi);
7009 #ifdef CONFIG_IGB_DCA
7010 static void igb_update_tx_dca(struct igb_adapter *adapter,
7011 struct igb_ring *tx_ring,
7014 struct e1000_hw *hw = &adapter->hw;
7015 u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
7017 if (hw->mac.type != e1000_82575)
7018 txctrl <<= E1000_DCA_TXCTRL_CPUID_SHIFT;
7020 /* We can enable relaxed ordering for reads, but not writes when
7021 * DCA is enabled. This is due to a known issue in some chipsets
7022 * which will cause the DCA tag to be cleared.
7024 txctrl |= E1000_DCA_TXCTRL_DESC_RRO_EN |
7025 E1000_DCA_TXCTRL_DATA_RRO_EN |
7026 E1000_DCA_TXCTRL_DESC_DCA_EN;
7028 wr32(E1000_DCA_TXCTRL(tx_ring->reg_idx), txctrl);
7031 static void igb_update_rx_dca(struct igb_adapter *adapter,
7032 struct igb_ring *rx_ring,
7035 struct e1000_hw *hw = &adapter->hw;
7036 u32 rxctrl = dca3_get_tag(&adapter->pdev->dev, cpu);
7038 if (hw->mac.type != e1000_82575)
7039 rxctrl <<= E1000_DCA_RXCTRL_CPUID_SHIFT;
7041 /* We can enable relaxed ordering for reads, but not writes when
7042 * DCA is enabled. This is due to a known issue in some chipsets
7043 * which will cause the DCA tag to be cleared.
7045 rxctrl |= E1000_DCA_RXCTRL_DESC_RRO_EN |
7046 E1000_DCA_RXCTRL_DESC_DCA_EN;
7048 wr32(E1000_DCA_RXCTRL(rx_ring->reg_idx), rxctrl);
7051 static void igb_update_dca(struct igb_q_vector *q_vector)
7053 struct igb_adapter *adapter = q_vector->adapter;
7054 int cpu = get_cpu();
7056 if (q_vector->cpu == cpu)
7059 if (q_vector->tx.ring)
7060 igb_update_tx_dca(adapter, q_vector->tx.ring, cpu);
7062 if (q_vector->rx.ring)
7063 igb_update_rx_dca(adapter, q_vector->rx.ring, cpu);
7065 q_vector->cpu = cpu;
7070 static void igb_setup_dca(struct igb_adapter *adapter)
7072 struct e1000_hw *hw = &adapter->hw;
7075 if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
7078 /* Always use CB2 mode, difference is masked in the CB driver. */
7079 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
7081 for (i = 0; i < adapter->num_q_vectors; i++) {
7082 adapter->q_vector[i]->cpu = -1;
7083 igb_update_dca(adapter->q_vector[i]);
7087 static int __igb_notify_dca(struct device *dev, void *data)
7089 struct net_device *netdev = dev_get_drvdata(dev);
7090 struct igb_adapter *adapter = netdev_priv(netdev);
7091 struct pci_dev *pdev = adapter->pdev;
7092 struct e1000_hw *hw = &adapter->hw;
7093 unsigned long event = *(unsigned long *)data;
7096 case DCA_PROVIDER_ADD:
7097 /* if already enabled, don't do it again */
7098 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
7100 if (dca_add_requester(dev) == 0) {
7101 adapter->flags |= IGB_FLAG_DCA_ENABLED;
7102 dev_info(&pdev->dev, "DCA enabled\n");
7103 igb_setup_dca(adapter);
7106 fallthrough; /* since DCA is disabled. */
7107 case DCA_PROVIDER_REMOVE:
7108 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
7109 /* without this a class_device is left
7110 * hanging around in the sysfs model
7112 dca_remove_requester(dev);
7113 dev_info(&pdev->dev, "DCA disabled\n");
7114 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
7115 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
7123 static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
7128 ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
7131 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7133 #endif /* CONFIG_IGB_DCA */
7135 #ifdef CONFIG_PCI_IOV
7136 static int igb_vf_configure(struct igb_adapter *adapter, int vf)
7138 unsigned char mac_addr[ETH_ALEN];
7140 eth_zero_addr(mac_addr);
7141 igb_set_vf_mac(adapter, vf, mac_addr);
7143 /* By default spoof check is enabled for all VFs */
7144 adapter->vf_data[vf].spoofchk_enabled = true;
7146 /* By default VFs are not trusted */
7147 adapter->vf_data[vf].trusted = false;
7153 static void igb_ping_all_vfs(struct igb_adapter *adapter)
7155 struct e1000_hw *hw = &adapter->hw;
7159 for (i = 0 ; i < adapter->vfs_allocated_count; i++) {
7160 ping = E1000_PF_CONTROL_MSG;
7161 if (adapter->vf_data[i].flags & IGB_VF_FLAG_CTS)
7162 ping |= E1000_VT_MSGTYPE_CTS;
7163 igb_write_mbx(hw, &ping, 1, i);
7167 static int igb_set_vf_promisc(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
7169 struct e1000_hw *hw = &adapter->hw;
7170 u32 vmolr = rd32(E1000_VMOLR(vf));
7171 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7173 vf_data->flags &= ~(IGB_VF_FLAG_UNI_PROMISC |
7174 IGB_VF_FLAG_MULTI_PROMISC);
7175 vmolr &= ~(E1000_VMOLR_ROPE | E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
7177 if (*msgbuf & E1000_VF_SET_PROMISC_MULTICAST) {
7178 vmolr |= E1000_VMOLR_MPME;
7179 vf_data->flags |= IGB_VF_FLAG_MULTI_PROMISC;
7180 *msgbuf &= ~E1000_VF_SET_PROMISC_MULTICAST;
7182 /* if we have hashes and we are clearing a multicast promisc
7183 * flag we need to write the hashes to the MTA as this step
7184 * was previously skipped
7186 if (vf_data->num_vf_mc_hashes > 30) {
7187 vmolr |= E1000_VMOLR_MPME;
7188 } else if (vf_data->num_vf_mc_hashes) {
7191 vmolr |= E1000_VMOLR_ROMPE;
7192 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
7193 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
7197 wr32(E1000_VMOLR(vf), vmolr);
7199 /* there are flags left unprocessed, likely not supported */
7200 if (*msgbuf & E1000_VT_MSGINFO_MASK)
7206 static int igb_set_vf_multicasts(struct igb_adapter *adapter,
7207 u32 *msgbuf, u32 vf)
7209 int n = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
7210 u16 *hash_list = (u16 *)&msgbuf[1];
7211 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7214 /* salt away the number of multicast addresses assigned
7215 * to this VF for later use to restore when the PF multi cast
7218 vf_data->num_vf_mc_hashes = n;
7220 /* only up to 30 hash values supported */
7224 /* store the hashes for later use */
7225 for (i = 0; i < n; i++)
7226 vf_data->vf_mc_hashes[i] = hash_list[i];
7228 /* Flush and reset the mta with the new values */
7229 igb_set_rx_mode(adapter->netdev);
7234 static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
7236 struct e1000_hw *hw = &adapter->hw;
7237 struct vf_data_storage *vf_data;
7240 for (i = 0; i < adapter->vfs_allocated_count; i++) {
7241 u32 vmolr = rd32(E1000_VMOLR(i));
7243 vmolr &= ~(E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
7245 vf_data = &adapter->vf_data[i];
7247 if ((vf_data->num_vf_mc_hashes > 30) ||
7248 (vf_data->flags & IGB_VF_FLAG_MULTI_PROMISC)) {
7249 vmolr |= E1000_VMOLR_MPME;
7250 } else if (vf_data->num_vf_mc_hashes) {
7251 vmolr |= E1000_VMOLR_ROMPE;
7252 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
7253 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
7255 wr32(E1000_VMOLR(i), vmolr);
7259 static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
7261 struct e1000_hw *hw = &adapter->hw;
7262 u32 pool_mask, vlvf_mask, i;
7264 /* create mask for VF and other pools */
7265 pool_mask = E1000_VLVF_POOLSEL_MASK;
7266 vlvf_mask = BIT(E1000_VLVF_POOLSEL_SHIFT + vf);
7268 /* drop PF from pool bits */
7269 pool_mask &= ~BIT(E1000_VLVF_POOLSEL_SHIFT +
7270 adapter->vfs_allocated_count);
7272 /* Find the vlan filter for this id */
7273 for (i = E1000_VLVF_ARRAY_SIZE; i--;) {
7274 u32 vlvf = rd32(E1000_VLVF(i));
7275 u32 vfta_mask, vid, vfta;
7277 /* remove the vf from the pool */
7278 if (!(vlvf & vlvf_mask))
7281 /* clear out bit from VLVF */
7284 /* if other pools are present, just remove ourselves */
7285 if (vlvf & pool_mask)
7288 /* if PF is present, leave VFTA */
7289 if (vlvf & E1000_VLVF_POOLSEL_MASK)
7292 vid = vlvf & E1000_VLVF_VLANID_MASK;
7293 vfta_mask = BIT(vid % 32);
7295 /* clear bit from VFTA */
7296 vfta = adapter->shadow_vfta[vid / 32];
7297 if (vfta & vfta_mask)
7298 hw->mac.ops.write_vfta(hw, vid / 32, vfta ^ vfta_mask);
7300 /* clear pool selection enable */
7301 if (adapter->flags & IGB_FLAG_VLAN_PROMISC)
7302 vlvf &= E1000_VLVF_POOLSEL_MASK;
7306 /* clear pool bits */
7307 wr32(E1000_VLVF(i), vlvf);
7311 static int igb_find_vlvf_entry(struct e1000_hw *hw, u32 vlan)
7316 /* short cut the special case */
7320 /* Search for the VLAN id in the VLVF entries */
7321 for (idx = E1000_VLVF_ARRAY_SIZE; --idx;) {
7322 vlvf = rd32(E1000_VLVF(idx));
7323 if ((vlvf & VLAN_VID_MASK) == vlan)
7330 static void igb_update_pf_vlvf(struct igb_adapter *adapter, u32 vid)
7332 struct e1000_hw *hw = &adapter->hw;
7336 idx = igb_find_vlvf_entry(hw, vid);
7340 /* See if any other pools are set for this VLAN filter
7341 * entry other than the PF.
7343 pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT;
7344 bits = ~BIT(pf_id) & E1000_VLVF_POOLSEL_MASK;
7345 bits &= rd32(E1000_VLVF(idx));
7347 /* Disable the filter so this falls into the default pool. */
7349 if (adapter->flags & IGB_FLAG_VLAN_PROMISC)
7350 wr32(E1000_VLVF(idx), BIT(pf_id));
7352 wr32(E1000_VLVF(idx), 0);
7356 static s32 igb_set_vf_vlan(struct igb_adapter *adapter, u32 vid,
7359 int pf_id = adapter->vfs_allocated_count;
7360 struct e1000_hw *hw = &adapter->hw;
7363 /* If VLAN overlaps with one the PF is currently monitoring make
7364 * sure that we are able to allocate a VLVF entry. This may be
7365 * redundant but it guarantees PF will maintain visibility to
7368 if (add && test_bit(vid, adapter->active_vlans)) {
7369 err = igb_vfta_set(hw, vid, pf_id, true, false);
7374 err = igb_vfta_set(hw, vid, vf, add, false);
7379 /* If we failed to add the VF VLAN or we are removing the VF VLAN
7380 * we may need to drop the PF pool bit in order to allow us to free
7381 * up the VLVF resources.
7383 if (test_bit(vid, adapter->active_vlans) ||
7384 (adapter->flags & IGB_FLAG_VLAN_PROMISC))
7385 igb_update_pf_vlvf(adapter, vid);
7390 static void igb_set_vmvir(struct igb_adapter *adapter, u32 vid, u32 vf)
7392 struct e1000_hw *hw = &adapter->hw;
7395 wr32(E1000_VMVIR(vf), (vid | E1000_VMVIR_VLANA_DEFAULT));
7397 wr32(E1000_VMVIR(vf), 0);
7400 static int igb_enable_port_vlan(struct igb_adapter *adapter, int vf,
7405 err = igb_set_vf_vlan(adapter, vlan, true, vf);
7409 igb_set_vmvir(adapter, vlan | (qos << VLAN_PRIO_SHIFT), vf);
7410 igb_set_vmolr(adapter, vf, !vlan);
7412 /* revoke access to previous VLAN */
7413 if (vlan != adapter->vf_data[vf].pf_vlan)
7414 igb_set_vf_vlan(adapter, adapter->vf_data[vf].pf_vlan,
7417 adapter->vf_data[vf].pf_vlan = vlan;
7418 adapter->vf_data[vf].pf_qos = qos;
7419 igb_set_vf_vlan_strip(adapter, vf, true);
7420 dev_info(&adapter->pdev->dev,
7421 "Setting VLAN %d, QOS 0x%x on VF %d\n", vlan, qos, vf);
7422 if (test_bit(__IGB_DOWN, &adapter->state)) {
7423 dev_warn(&adapter->pdev->dev,
7424 "The VF VLAN has been set, but the PF device is not up.\n");
7425 dev_warn(&adapter->pdev->dev,
7426 "Bring the PF device up before attempting to use the VF device.\n");
7432 static int igb_disable_port_vlan(struct igb_adapter *adapter, int vf)
7434 /* Restore tagless access via VLAN 0 */
7435 igb_set_vf_vlan(adapter, 0, true, vf);
7437 igb_set_vmvir(adapter, 0, vf);
7438 igb_set_vmolr(adapter, vf, true);
7440 /* Remove any PF assigned VLAN */
7441 if (adapter->vf_data[vf].pf_vlan)
7442 igb_set_vf_vlan(adapter, adapter->vf_data[vf].pf_vlan,
7445 adapter->vf_data[vf].pf_vlan = 0;
7446 adapter->vf_data[vf].pf_qos = 0;
7447 igb_set_vf_vlan_strip(adapter, vf, false);
7452 static int igb_ndo_set_vf_vlan(struct net_device *netdev, int vf,
7453 u16 vlan, u8 qos, __be16 vlan_proto)
7455 struct igb_adapter *adapter = netdev_priv(netdev);
7457 if ((vf >= adapter->vfs_allocated_count) || (vlan > 4095) || (qos > 7))
7460 if (vlan_proto != htons(ETH_P_8021Q))
7461 return -EPROTONOSUPPORT;
7463 return (vlan || qos) ? igb_enable_port_vlan(adapter, vf, vlan, qos) :
7464 igb_disable_port_vlan(adapter, vf);
7467 static int igb_set_vf_vlan_msg(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
7469 int add = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
7470 int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
7473 if (adapter->vf_data[vf].pf_vlan)
7476 /* VLAN 0 is a special case, don't allow it to be removed */
7480 ret = igb_set_vf_vlan(adapter, vid, !!add, vf);
7482 igb_set_vf_vlan_strip(adapter, vf, !!vid);
7486 static inline void igb_vf_reset(struct igb_adapter *adapter, u32 vf)
7488 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7490 /* clear flags - except flag that indicates PF has set the MAC */
7491 vf_data->flags &= IGB_VF_FLAG_PF_SET_MAC;
7492 vf_data->last_nack = jiffies;
7494 /* reset vlans for device */
7495 igb_clear_vf_vfta(adapter, vf);
7496 igb_set_vf_vlan(adapter, vf_data->pf_vlan, true, vf);
7497 igb_set_vmvir(adapter, vf_data->pf_vlan |
7498 (vf_data->pf_qos << VLAN_PRIO_SHIFT), vf);
7499 igb_set_vmolr(adapter, vf, !vf_data->pf_vlan);
7500 igb_set_vf_vlan_strip(adapter, vf, !!(vf_data->pf_vlan));
7502 /* reset multicast table array for vf */
7503 adapter->vf_data[vf].num_vf_mc_hashes = 0;
7505 /* Flush and reset the mta with the new values */
7506 igb_set_rx_mode(adapter->netdev);
7509 static void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
7511 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
7513 /* clear mac address as we were hotplug removed/added */
7514 if (!(adapter->vf_data[vf].flags & IGB_VF_FLAG_PF_SET_MAC))
7515 eth_zero_addr(vf_mac);
7517 /* process remaining reset events */
7518 igb_vf_reset(adapter, vf);
7521 static void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
7523 struct e1000_hw *hw = &adapter->hw;
7524 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
7526 u8 *addr = (u8 *)(&msgbuf[1]);
7528 /* process all the same items cleared in a function level reset */
7529 igb_vf_reset(adapter, vf);
7531 /* set vf mac address */
7532 igb_set_vf_mac(adapter, vf, vf_mac);
7534 /* enable transmit and receive for vf */
7535 reg = rd32(E1000_VFTE);
7536 wr32(E1000_VFTE, reg | BIT(vf));
7537 reg = rd32(E1000_VFRE);
7538 wr32(E1000_VFRE, reg | BIT(vf));
7540 adapter->vf_data[vf].flags |= IGB_VF_FLAG_CTS;
7542 /* reply to reset with ack and vf mac address */
7543 if (!is_zero_ether_addr(vf_mac)) {
7544 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
7545 memcpy(addr, vf_mac, ETH_ALEN);
7547 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_NACK;
7549 igb_write_mbx(hw, msgbuf, 3, vf);
7552 static void igb_flush_mac_table(struct igb_adapter *adapter)
7554 struct e1000_hw *hw = &adapter->hw;
7557 for (i = 0; i < hw->mac.rar_entry_count; i++) {
7558 adapter->mac_table[i].state &= ~IGB_MAC_STATE_IN_USE;
7559 eth_zero_addr(adapter->mac_table[i].addr);
7560 adapter->mac_table[i].queue = 0;
7561 igb_rar_set_index(adapter, i);
7565 static int igb_available_rars(struct igb_adapter *adapter, u8 queue)
7567 struct e1000_hw *hw = &adapter->hw;
7568 /* do not count rar entries reserved for VFs MAC addresses */
7569 int rar_entries = hw->mac.rar_entry_count -
7570 adapter->vfs_allocated_count;
7573 for (i = 0; i < rar_entries; i++) {
7574 /* do not count default entries */
7575 if (adapter->mac_table[i].state & IGB_MAC_STATE_DEFAULT)
7578 /* do not count "in use" entries for different queues */
7579 if ((adapter->mac_table[i].state & IGB_MAC_STATE_IN_USE) &&
7580 (adapter->mac_table[i].queue != queue))
7589 /* Set default MAC address for the PF in the first RAR entry */
7590 static void igb_set_default_mac_filter(struct igb_adapter *adapter)
7592 struct igb_mac_addr *mac_table = &adapter->mac_table[0];
7594 ether_addr_copy(mac_table->addr, adapter->hw.mac.addr);
7595 mac_table->queue = adapter->vfs_allocated_count;
7596 mac_table->state = IGB_MAC_STATE_DEFAULT | IGB_MAC_STATE_IN_USE;
7598 igb_rar_set_index(adapter, 0);
7601 /* If the filter to be added and an already existing filter express
7602 * the same address and address type, it should be possible to only
7603 * override the other configurations, for example the queue to steer
7606 static bool igb_mac_entry_can_be_used(const struct igb_mac_addr *entry,
7607 const u8 *addr, const u8 flags)
7609 if (!(entry->state & IGB_MAC_STATE_IN_USE))
7612 if ((entry->state & IGB_MAC_STATE_SRC_ADDR) !=
7613 (flags & IGB_MAC_STATE_SRC_ADDR))
7616 if (!ether_addr_equal(addr, entry->addr))
7622 /* Add a MAC filter for 'addr' directing matching traffic to 'queue',
7623 * 'flags' is used to indicate what kind of match is made, match is by
7624 * default for the destination address, if matching by source address
7625 * is desired the flag IGB_MAC_STATE_SRC_ADDR can be used.
7627 static int igb_add_mac_filter_flags(struct igb_adapter *adapter,
7628 const u8 *addr, const u8 queue,
7631 struct e1000_hw *hw = &adapter->hw;
7632 int rar_entries = hw->mac.rar_entry_count -
7633 adapter->vfs_allocated_count;
7636 if (is_zero_ether_addr(addr))
7639 /* Search for the first empty entry in the MAC table.
7640 * Do not touch entries at the end of the table reserved for the VF MAC
7643 for (i = 0; i < rar_entries; i++) {
7644 if (!igb_mac_entry_can_be_used(&adapter->mac_table[i],
7648 ether_addr_copy(adapter->mac_table[i].addr, addr);
7649 adapter->mac_table[i].queue = queue;
7650 adapter->mac_table[i].state |= IGB_MAC_STATE_IN_USE | flags;
7652 igb_rar_set_index(adapter, i);
7659 static int igb_add_mac_filter(struct igb_adapter *adapter, const u8 *addr,
7662 return igb_add_mac_filter_flags(adapter, addr, queue, 0);
7665 /* Remove a MAC filter for 'addr' directing matching traffic to
7666 * 'queue', 'flags' is used to indicate what kind of match need to be
7667 * removed, match is by default for the destination address, if
7668 * matching by source address is to be removed the flag
7669 * IGB_MAC_STATE_SRC_ADDR can be used.
7671 static int igb_del_mac_filter_flags(struct igb_adapter *adapter,
7672 const u8 *addr, const u8 queue,
7675 struct e1000_hw *hw = &adapter->hw;
7676 int rar_entries = hw->mac.rar_entry_count -
7677 adapter->vfs_allocated_count;
7680 if (is_zero_ether_addr(addr))
7683 /* Search for matching entry in the MAC table based on given address
7684 * and queue. Do not touch entries at the end of the table reserved
7685 * for the VF MAC addresses.
7687 for (i = 0; i < rar_entries; i++) {
7688 if (!(adapter->mac_table[i].state & IGB_MAC_STATE_IN_USE))
7690 if ((adapter->mac_table[i].state & flags) != flags)
7692 if (adapter->mac_table[i].queue != queue)
7694 if (!ether_addr_equal(adapter->mac_table[i].addr, addr))
7697 /* When a filter for the default address is "deleted",
7698 * we return it to its initial configuration
7700 if (adapter->mac_table[i].state & IGB_MAC_STATE_DEFAULT) {
7701 adapter->mac_table[i].state =
7702 IGB_MAC_STATE_DEFAULT | IGB_MAC_STATE_IN_USE;
7703 adapter->mac_table[i].queue =
7704 adapter->vfs_allocated_count;
7706 adapter->mac_table[i].state = 0;
7707 adapter->mac_table[i].queue = 0;
7708 eth_zero_addr(adapter->mac_table[i].addr);
7711 igb_rar_set_index(adapter, i);
7718 static int igb_del_mac_filter(struct igb_adapter *adapter, const u8 *addr,
7721 return igb_del_mac_filter_flags(adapter, addr, queue, 0);
7724 int igb_add_mac_steering_filter(struct igb_adapter *adapter,
7725 const u8 *addr, u8 queue, u8 flags)
7727 struct e1000_hw *hw = &adapter->hw;
7729 /* In theory, this should be supported on 82575 as well, but
7730 * that part wasn't easily accessible during development.
7732 if (hw->mac.type != e1000_i210)
7735 return igb_add_mac_filter_flags(adapter, addr, queue,
7736 IGB_MAC_STATE_QUEUE_STEERING | flags);
7739 int igb_del_mac_steering_filter(struct igb_adapter *adapter,
7740 const u8 *addr, u8 queue, u8 flags)
7742 return igb_del_mac_filter_flags(adapter, addr, queue,
7743 IGB_MAC_STATE_QUEUE_STEERING | flags);
7746 static int igb_uc_sync(struct net_device *netdev, const unsigned char *addr)
7748 struct igb_adapter *adapter = netdev_priv(netdev);
7751 ret = igb_add_mac_filter(adapter, addr, adapter->vfs_allocated_count);
7753 return min_t(int, ret, 0);
7756 static int igb_uc_unsync(struct net_device *netdev, const unsigned char *addr)
7758 struct igb_adapter *adapter = netdev_priv(netdev);
7760 igb_del_mac_filter(adapter, addr, adapter->vfs_allocated_count);
7765 static int igb_set_vf_mac_filter(struct igb_adapter *adapter, const int vf,
7766 const u32 info, const u8 *addr)
7768 struct pci_dev *pdev = adapter->pdev;
7769 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7770 struct list_head *pos;
7771 struct vf_mac_filter *entry = NULL;
7774 if ((vf_data->flags & IGB_VF_FLAG_PF_SET_MAC) &&
7775 !vf_data->trusted) {
7776 dev_warn(&pdev->dev,
7777 "VF %d requested MAC filter but is administratively denied\n",
7781 if (!is_valid_ether_addr(addr)) {
7782 dev_warn(&pdev->dev,
7783 "VF %d attempted to set invalid MAC filter\n",
7789 case E1000_VF_MAC_FILTER_CLR:
7790 /* remove all unicast MAC filters related to the current VF */
7791 list_for_each(pos, &adapter->vf_macs.l) {
7792 entry = list_entry(pos, struct vf_mac_filter, l);
7793 if (entry->vf == vf) {
7796 igb_del_mac_filter(adapter, entry->vf_mac, vf);
7800 case E1000_VF_MAC_FILTER_ADD:
7801 /* try to find empty slot in the list */
7802 list_for_each(pos, &adapter->vf_macs.l) {
7803 entry = list_entry(pos, struct vf_mac_filter, l);
7808 if (entry && entry->free) {
7809 entry->free = false;
7811 ether_addr_copy(entry->vf_mac, addr);
7813 ret = igb_add_mac_filter(adapter, addr, vf);
7814 ret = min_t(int, ret, 0);
7820 dev_warn(&pdev->dev,
7821 "VF %d has requested MAC filter but there is no space for it\n",
7832 static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf)
7834 struct pci_dev *pdev = adapter->pdev;
7835 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7836 u32 info = msg[0] & E1000_VT_MSGINFO_MASK;
7838 /* The VF MAC Address is stored in a packed array of bytes
7839 * starting at the second 32 bit word of the msg array
7841 unsigned char *addr = (unsigned char *)&msg[1];
7845 if ((vf_data->flags & IGB_VF_FLAG_PF_SET_MAC) &&
7846 !vf_data->trusted) {
7847 dev_warn(&pdev->dev,
7848 "VF %d attempted to override administratively set MAC address\nReload the VF driver to resume operations\n",
7853 if (!is_valid_ether_addr(addr)) {
7854 dev_warn(&pdev->dev,
7855 "VF %d attempted to set invalid MAC\n",
7860 ret = igb_set_vf_mac(adapter, vf, addr);
7862 ret = igb_set_vf_mac_filter(adapter, vf, info, addr);
7868 static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf)
7870 struct e1000_hw *hw = &adapter->hw;
7871 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7872 u32 msg = E1000_VT_MSGTYPE_NACK;
7874 /* if device isn't clear to send it shouldn't be reading either */
7875 if (!(vf_data->flags & IGB_VF_FLAG_CTS) &&
7876 time_after(jiffies, vf_data->last_nack + (2 * HZ))) {
7877 igb_write_mbx(hw, &msg, 1, vf);
7878 vf_data->last_nack = jiffies;
7882 static void igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
7884 struct pci_dev *pdev = adapter->pdev;
7885 u32 msgbuf[E1000_VFMAILBOX_SIZE];
7886 struct e1000_hw *hw = &adapter->hw;
7887 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7890 retval = igb_read_mbx(hw, msgbuf, E1000_VFMAILBOX_SIZE, vf, false);
7893 /* if receive failed revoke VF CTS stats and restart init */
7894 dev_err(&pdev->dev, "Error receiving message from VF\n");
7895 vf_data->flags &= ~IGB_VF_FLAG_CTS;
7896 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
7901 /* this is a message we already processed, do nothing */
7902 if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
7905 /* until the vf completes a reset it should not be
7906 * allowed to start any configuration.
7908 if (msgbuf[0] == E1000_VF_RESET) {
7909 /* unlocks mailbox */
7910 igb_vf_reset_msg(adapter, vf);
7914 if (!(vf_data->flags & IGB_VF_FLAG_CTS)) {
7915 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
7921 switch ((msgbuf[0] & 0xFFFF)) {
7922 case E1000_VF_SET_MAC_ADDR:
7923 retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
7925 case E1000_VF_SET_PROMISC:
7926 retval = igb_set_vf_promisc(adapter, msgbuf, vf);
7928 case E1000_VF_SET_MULTICAST:
7929 retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
7931 case E1000_VF_SET_LPE:
7932 retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf);
7934 case E1000_VF_SET_VLAN:
7936 if (vf_data->pf_vlan)
7937 dev_warn(&pdev->dev,
7938 "VF %d attempted to override administratively set VLAN tag\nReload the VF driver to resume operations\n",
7941 retval = igb_set_vf_vlan_msg(adapter, msgbuf, vf);
7944 dev_err(&pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
7949 msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
7951 /* notify the VF of the results of what it sent us */
7953 msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
7955 msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
7957 /* unlocks mailbox */
7958 igb_write_mbx(hw, msgbuf, 1, vf);
7962 igb_unlock_mbx(hw, vf);
7965 static void igb_msg_task(struct igb_adapter *adapter)
7967 struct e1000_hw *hw = &adapter->hw;
7968 unsigned long flags;
7971 spin_lock_irqsave(&adapter->vfs_lock, flags);
7972 for (vf = 0; vf < adapter->vfs_allocated_count; vf++) {
7973 /* process any reset requests */
7974 if (!igb_check_for_rst(hw, vf))
7975 igb_vf_reset_event(adapter, vf);
7977 /* process any messages pending */
7978 if (!igb_check_for_msg(hw, vf))
7979 igb_rcv_msg_from_vf(adapter, vf);
7981 /* process any acks */
7982 if (!igb_check_for_ack(hw, vf))
7983 igb_rcv_ack_from_vf(adapter, vf);
7985 spin_unlock_irqrestore(&adapter->vfs_lock, flags);
7989 * igb_set_uta - Set unicast filter table address
7990 * @adapter: board private structure
7991 * @set: boolean indicating if we are setting or clearing bits
7993 * The unicast table address is a register array of 32-bit registers.
7994 * The table is meant to be used in a way similar to how the MTA is used
7995 * however due to certain limitations in the hardware it is necessary to
7996 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
7997 * enable bit to allow vlan tag stripping when promiscuous mode is enabled
7999 static void igb_set_uta(struct igb_adapter *adapter, bool set)
8001 struct e1000_hw *hw = &adapter->hw;
8002 u32 uta = set ? ~0 : 0;
8005 /* we only need to do this if VMDq is enabled */
8006 if (!adapter->vfs_allocated_count)
8009 for (i = hw->mac.uta_reg_count; i--;)
8010 array_wr32(E1000_UTA, i, uta);
8014 * igb_intr_msi - Interrupt Handler
8015 * @irq: interrupt number
8016 * @data: pointer to a network interface device structure
8018 static irqreturn_t igb_intr_msi(int irq, void *data)
8020 struct igb_adapter *adapter = data;
8021 struct igb_q_vector *q_vector = adapter->q_vector[0];
8022 struct e1000_hw *hw = &adapter->hw;
8023 /* read ICR disables interrupts using IAM */
8024 u32 icr = rd32(E1000_ICR);
8026 igb_write_itr(q_vector);
8028 if (icr & E1000_ICR_DRSTA)
8029 schedule_work(&adapter->reset_task);
8031 if (icr & E1000_ICR_DOUTSYNC) {
8032 /* HW is reporting DMA is out of sync */
8033 adapter->stats.doosync++;
8036 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
8037 hw->mac.get_link_status = 1;
8038 if (!test_bit(__IGB_DOWN, &adapter->state))
8039 mod_timer(&adapter->watchdog_timer, jiffies + 1);
8042 if (icr & E1000_ICR_TS)
8043 igb_tsync_interrupt(adapter);
8045 napi_schedule(&q_vector->napi);
8051 * igb_intr - Legacy Interrupt Handler
8052 * @irq: interrupt number
8053 * @data: pointer to a network interface device structure
8055 static irqreturn_t igb_intr(int irq, void *data)
8057 struct igb_adapter *adapter = data;
8058 struct igb_q_vector *q_vector = adapter->q_vector[0];
8059 struct e1000_hw *hw = &adapter->hw;
8060 /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked. No
8061 * need for the IMC write
8063 u32 icr = rd32(E1000_ICR);
8065 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
8066 * not set, then the adapter didn't send an interrupt
8068 if (!(icr & E1000_ICR_INT_ASSERTED))
8071 igb_write_itr(q_vector);
8073 if (icr & E1000_ICR_DRSTA)
8074 schedule_work(&adapter->reset_task);
8076 if (icr & E1000_ICR_DOUTSYNC) {
8077 /* HW is reporting DMA is out of sync */
8078 adapter->stats.doosync++;
8081 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
8082 hw->mac.get_link_status = 1;
8083 /* guard against interrupt when we're going down */
8084 if (!test_bit(__IGB_DOWN, &adapter->state))
8085 mod_timer(&adapter->watchdog_timer, jiffies + 1);
8088 if (icr & E1000_ICR_TS)
8089 igb_tsync_interrupt(adapter);
8091 napi_schedule(&q_vector->napi);
8096 static void igb_ring_irq_enable(struct igb_q_vector *q_vector)
8098 struct igb_adapter *adapter = q_vector->adapter;
8099 struct e1000_hw *hw = &adapter->hw;
8101 if ((q_vector->rx.ring && (adapter->rx_itr_setting & 3)) ||
8102 (!q_vector->rx.ring && (adapter->tx_itr_setting & 3))) {
8103 if ((adapter->num_q_vectors == 1) && !adapter->vf_data)
8104 igb_set_itr(q_vector);
8106 igb_update_ring_itr(q_vector);
8109 if (!test_bit(__IGB_DOWN, &adapter->state)) {
8110 if (adapter->flags & IGB_FLAG_HAS_MSIX)
8111 wr32(E1000_EIMS, q_vector->eims_value);
8113 igb_irq_enable(adapter);
8118 * igb_poll - NAPI Rx polling callback
8119 * @napi: napi polling structure
8120 * @budget: count of how many packets we should handle
8122 static int igb_poll(struct napi_struct *napi, int budget)
8124 struct igb_q_vector *q_vector = container_of(napi,
8125 struct igb_q_vector,
8127 bool clean_complete = true;
8130 #ifdef CONFIG_IGB_DCA
8131 if (q_vector->adapter->flags & IGB_FLAG_DCA_ENABLED)
8132 igb_update_dca(q_vector);
8134 if (q_vector->tx.ring)
8135 clean_complete = igb_clean_tx_irq(q_vector, budget);
8137 if (q_vector->rx.ring) {
8138 int cleaned = igb_clean_rx_irq(q_vector, budget);
8140 work_done += cleaned;
8141 if (cleaned >= budget)
8142 clean_complete = false;
8145 /* If all work not completed, return budget and keep polling */
8146 if (!clean_complete)
8149 /* Exit the polling mode, but don't re-enable interrupts if stack might
8150 * poll us due to busy-polling
8152 if (likely(napi_complete_done(napi, work_done)))
8153 igb_ring_irq_enable(q_vector);
8159 * igb_clean_tx_irq - Reclaim resources after transmit completes
8160 * @q_vector: pointer to q_vector containing needed info
8161 * @napi_budget: Used to determine if we are in netpoll
8163 * returns true if ring is completely cleaned
8165 static bool igb_clean_tx_irq(struct igb_q_vector *q_vector, int napi_budget)
8167 struct igb_adapter *adapter = q_vector->adapter;
8168 struct igb_ring *tx_ring = q_vector->tx.ring;
8169 struct igb_tx_buffer *tx_buffer;
8170 union e1000_adv_tx_desc *tx_desc;
8171 unsigned int total_bytes = 0, total_packets = 0;
8172 unsigned int budget = q_vector->tx.work_limit;
8173 unsigned int i = tx_ring->next_to_clean;
8175 if (test_bit(__IGB_DOWN, &adapter->state))
8178 tx_buffer = &tx_ring->tx_buffer_info[i];
8179 tx_desc = IGB_TX_DESC(tx_ring, i);
8180 i -= tx_ring->count;
8183 union e1000_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
8185 /* if next_to_watch is not set then there is no work pending */
8189 /* prevent any other reads prior to eop_desc */
8192 /* if DD is not set pending work has not been completed */
8193 if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)))
8196 /* clear next_to_watch to prevent false hangs */
8197 tx_buffer->next_to_watch = NULL;
8199 /* update the statistics for this packet */
8200 total_bytes += tx_buffer->bytecount;
8201 total_packets += tx_buffer->gso_segs;
8204 if (tx_buffer->type == IGB_TYPE_SKB)
8205 napi_consume_skb(tx_buffer->skb, napi_budget);
8207 xdp_return_frame(tx_buffer->xdpf);
8209 /* unmap skb header data */
8210 dma_unmap_single(tx_ring->dev,
8211 dma_unmap_addr(tx_buffer, dma),
8212 dma_unmap_len(tx_buffer, len),
8215 /* clear tx_buffer data */
8216 dma_unmap_len_set(tx_buffer, len, 0);
8218 /* clear last DMA location and unmap remaining buffers */
8219 while (tx_desc != eop_desc) {
8224 i -= tx_ring->count;
8225 tx_buffer = tx_ring->tx_buffer_info;
8226 tx_desc = IGB_TX_DESC(tx_ring, 0);
8229 /* unmap any remaining paged data */
8230 if (dma_unmap_len(tx_buffer, len)) {
8231 dma_unmap_page(tx_ring->dev,
8232 dma_unmap_addr(tx_buffer, dma),
8233 dma_unmap_len(tx_buffer, len),
8235 dma_unmap_len_set(tx_buffer, len, 0);
8239 /* move us one more past the eop_desc for start of next pkt */
8244 i -= tx_ring->count;
8245 tx_buffer = tx_ring->tx_buffer_info;
8246 tx_desc = IGB_TX_DESC(tx_ring, 0);
8249 /* issue prefetch for next Tx descriptor */
8252 /* update budget accounting */
8254 } while (likely(budget));
8256 netdev_tx_completed_queue(txring_txq(tx_ring),
8257 total_packets, total_bytes);
8258 i += tx_ring->count;
8259 tx_ring->next_to_clean = i;
8260 u64_stats_update_begin(&tx_ring->tx_syncp);
8261 tx_ring->tx_stats.bytes += total_bytes;
8262 tx_ring->tx_stats.packets += total_packets;
8263 u64_stats_update_end(&tx_ring->tx_syncp);
8264 q_vector->tx.total_bytes += total_bytes;
8265 q_vector->tx.total_packets += total_packets;
8267 if (test_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags)) {
8268 struct e1000_hw *hw = &adapter->hw;
8270 /* Detect a transmit hang in hardware, this serializes the
8271 * check with the clearing of time_stamp and movement of i
8273 clear_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
8274 if (tx_buffer->next_to_watch &&
8275 time_after(jiffies, tx_buffer->time_stamp +
8276 (adapter->tx_timeout_factor * HZ)) &&
8277 !(rd32(E1000_STATUS) & E1000_STATUS_TXOFF)) {
8279 /* detected Tx unit hang */
8280 dev_err(tx_ring->dev,
8281 "Detected Tx Unit Hang\n"
8285 " next_to_use <%x>\n"
8286 " next_to_clean <%x>\n"
8287 "buffer_info[next_to_clean]\n"
8288 " time_stamp <%lx>\n"
8289 " next_to_watch <%p>\n"
8291 " desc.status <%x>\n",
8292 tx_ring->queue_index,
8293 rd32(E1000_TDH(tx_ring->reg_idx)),
8294 readl(tx_ring->tail),
8295 tx_ring->next_to_use,
8296 tx_ring->next_to_clean,
8297 tx_buffer->time_stamp,
8298 tx_buffer->next_to_watch,
8300 tx_buffer->next_to_watch->wb.status);
8301 netif_stop_subqueue(tx_ring->netdev,
8302 tx_ring->queue_index);
8304 /* we are about to reset, no point in enabling stuff */
8309 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
8310 if (unlikely(total_packets &&
8311 netif_carrier_ok(tx_ring->netdev) &&
8312 igb_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD)) {
8313 /* Make sure that anybody stopping the queue after this
8314 * sees the new next_to_clean.
8317 if (__netif_subqueue_stopped(tx_ring->netdev,
8318 tx_ring->queue_index) &&
8319 !(test_bit(__IGB_DOWN, &adapter->state))) {
8320 netif_wake_subqueue(tx_ring->netdev,
8321 tx_ring->queue_index);
8323 u64_stats_update_begin(&tx_ring->tx_syncp);
8324 tx_ring->tx_stats.restart_queue++;
8325 u64_stats_update_end(&tx_ring->tx_syncp);
8333 * igb_reuse_rx_page - page flip buffer and store it back on the ring
8334 * @rx_ring: rx descriptor ring to store buffers on
8335 * @old_buff: donor buffer to have page reused
8337 * Synchronizes page for reuse by the adapter
8339 static void igb_reuse_rx_page(struct igb_ring *rx_ring,
8340 struct igb_rx_buffer *old_buff)
8342 struct igb_rx_buffer *new_buff;
8343 u16 nta = rx_ring->next_to_alloc;
8345 new_buff = &rx_ring->rx_buffer_info[nta];
8347 /* update, and store next to alloc */
8349 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
8351 /* Transfer page from old buffer to new buffer.
8352 * Move each member individually to avoid possible store
8353 * forwarding stalls.
8355 new_buff->dma = old_buff->dma;
8356 new_buff->page = old_buff->page;
8357 new_buff->page_offset = old_buff->page_offset;
8358 new_buff->pagecnt_bias = old_buff->pagecnt_bias;
8361 static bool igb_can_reuse_rx_page(struct igb_rx_buffer *rx_buffer,
8364 unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
8365 struct page *page = rx_buffer->page;
8367 /* avoid re-using remote and pfmemalloc pages */
8368 if (!dev_page_is_reusable(page))
8371 #if (PAGE_SIZE < 8192)
8372 /* if we are only owner of page we can reuse it */
8373 if (unlikely((rx_buf_pgcnt - pagecnt_bias) > 1))
8376 #define IGB_LAST_OFFSET \
8377 (SKB_WITH_OVERHEAD(PAGE_SIZE) - IGB_RXBUFFER_2048)
8379 if (rx_buffer->page_offset > IGB_LAST_OFFSET)
8383 /* If we have drained the page fragment pool we need to update
8384 * the pagecnt_bias and page count so that we fully restock the
8385 * number of references the driver holds.
8387 if (unlikely(pagecnt_bias == 1)) {
8388 page_ref_add(page, USHRT_MAX - 1);
8389 rx_buffer->pagecnt_bias = USHRT_MAX;
8396 * igb_add_rx_frag - Add contents of Rx buffer to sk_buff
8397 * @rx_ring: rx descriptor ring to transact packets on
8398 * @rx_buffer: buffer containing page to add
8399 * @skb: sk_buff to place the data into
8400 * @size: size of buffer to be added
8402 * This function will add the data contained in rx_buffer->page to the skb.
8404 static void igb_add_rx_frag(struct igb_ring *rx_ring,
8405 struct igb_rx_buffer *rx_buffer,
8406 struct sk_buff *skb,
8409 #if (PAGE_SIZE < 8192)
8410 unsigned int truesize = igb_rx_pg_size(rx_ring) / 2;
8412 unsigned int truesize = ring_uses_build_skb(rx_ring) ?
8413 SKB_DATA_ALIGN(IGB_SKB_PAD + size) :
8414 SKB_DATA_ALIGN(size);
8416 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
8417 rx_buffer->page_offset, size, truesize);
8418 #if (PAGE_SIZE < 8192)
8419 rx_buffer->page_offset ^= truesize;
8421 rx_buffer->page_offset += truesize;
8425 static struct sk_buff *igb_construct_skb(struct igb_ring *rx_ring,
8426 struct igb_rx_buffer *rx_buffer,
8427 struct xdp_buff *xdp,
8430 #if (PAGE_SIZE < 8192)
8431 unsigned int truesize = igb_rx_pg_size(rx_ring) / 2;
8433 unsigned int truesize = SKB_DATA_ALIGN(xdp->data_end -
8434 xdp->data_hard_start);
8436 unsigned int size = xdp->data_end - xdp->data;
8437 unsigned int headlen;
8438 struct sk_buff *skb;
8440 /* prefetch first cache line of first page */
8441 net_prefetch(xdp->data);
8443 /* allocate a skb to store the frags */
8444 skb = napi_alloc_skb(&rx_ring->q_vector->napi, IGB_RX_HDR_LEN);
8449 skb_hwtstamps(skb)->hwtstamp = timestamp;
8451 /* Determine available headroom for copy */
8453 if (headlen > IGB_RX_HDR_LEN)
8454 headlen = eth_get_headlen(skb->dev, xdp->data, IGB_RX_HDR_LEN);
8456 /* align pull length to size of long to optimize memcpy performance */
8457 memcpy(__skb_put(skb, headlen), xdp->data, ALIGN(headlen, sizeof(long)));
8459 /* update all of the pointers */
8462 skb_add_rx_frag(skb, 0, rx_buffer->page,
8463 (xdp->data + headlen) - page_address(rx_buffer->page),
8465 #if (PAGE_SIZE < 8192)
8466 rx_buffer->page_offset ^= truesize;
8468 rx_buffer->page_offset += truesize;
8471 rx_buffer->pagecnt_bias++;
8477 static struct sk_buff *igb_build_skb(struct igb_ring *rx_ring,
8478 struct igb_rx_buffer *rx_buffer,
8479 struct xdp_buff *xdp,
8482 #if (PAGE_SIZE < 8192)
8483 unsigned int truesize = igb_rx_pg_size(rx_ring) / 2;
8485 unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
8486 SKB_DATA_ALIGN(xdp->data_end -
8487 xdp->data_hard_start);
8489 unsigned int metasize = xdp->data - xdp->data_meta;
8490 struct sk_buff *skb;
8492 /* prefetch first cache line of first page */
8493 net_prefetch(xdp->data_meta);
8495 /* build an skb around the page buffer */
8496 skb = napi_build_skb(xdp->data_hard_start, truesize);
8500 /* update pointers within the skb to store the data */
8501 skb_reserve(skb, xdp->data - xdp->data_hard_start);
8502 __skb_put(skb, xdp->data_end - xdp->data);
8505 skb_metadata_set(skb, metasize);
8508 skb_hwtstamps(skb)->hwtstamp = timestamp;
8510 /* update buffer offset */
8511 #if (PAGE_SIZE < 8192)
8512 rx_buffer->page_offset ^= truesize;
8514 rx_buffer->page_offset += truesize;
8520 static struct sk_buff *igb_run_xdp(struct igb_adapter *adapter,
8521 struct igb_ring *rx_ring,
8522 struct xdp_buff *xdp)
8524 int err, result = IGB_XDP_PASS;
8525 struct bpf_prog *xdp_prog;
8528 xdp_prog = READ_ONCE(rx_ring->xdp_prog);
8533 prefetchw(xdp->data_hard_start); /* xdp_frame write */
8535 act = bpf_prog_run_xdp(xdp_prog, xdp);
8540 result = igb_xdp_xmit_back(adapter, xdp);
8541 if (result == IGB_XDP_CONSUMED)
8545 err = xdp_do_redirect(adapter->netdev, xdp, xdp_prog);
8548 result = IGB_XDP_REDIR;
8551 bpf_warn_invalid_xdp_action(adapter->netdev, xdp_prog, act);
8555 trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
8558 result = IGB_XDP_CONSUMED;
8562 return ERR_PTR(-result);
8565 static unsigned int igb_rx_frame_truesize(struct igb_ring *rx_ring,
8568 unsigned int truesize;
8570 #if (PAGE_SIZE < 8192)
8571 truesize = igb_rx_pg_size(rx_ring) / 2; /* Must be power-of-2 */
8573 truesize = ring_uses_build_skb(rx_ring) ?
8574 SKB_DATA_ALIGN(IGB_SKB_PAD + size) +
8575 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) :
8576 SKB_DATA_ALIGN(size);
8581 static void igb_rx_buffer_flip(struct igb_ring *rx_ring,
8582 struct igb_rx_buffer *rx_buffer,
8585 unsigned int truesize = igb_rx_frame_truesize(rx_ring, size);
8586 #if (PAGE_SIZE < 8192)
8587 rx_buffer->page_offset ^= truesize;
8589 rx_buffer->page_offset += truesize;
8593 static inline void igb_rx_checksum(struct igb_ring *ring,
8594 union e1000_adv_rx_desc *rx_desc,
8595 struct sk_buff *skb)
8597 skb_checksum_none_assert(skb);
8599 /* Ignore Checksum bit is set */
8600 if (igb_test_staterr(rx_desc, E1000_RXD_STAT_IXSM))
8603 /* Rx checksum disabled via ethtool */
8604 if (!(ring->netdev->features & NETIF_F_RXCSUM))
8607 /* TCP/UDP checksum error bit is set */
8608 if (igb_test_staterr(rx_desc,
8609 E1000_RXDEXT_STATERR_TCPE |
8610 E1000_RXDEXT_STATERR_IPE)) {
8611 /* work around errata with sctp packets where the TCPE aka
8612 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
8613 * packets, (aka let the stack check the crc32c)
8615 if (!((skb->len == 60) &&
8616 test_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags))) {
8617 u64_stats_update_begin(&ring->rx_syncp);
8618 ring->rx_stats.csum_err++;
8619 u64_stats_update_end(&ring->rx_syncp);
8621 /* let the stack verify checksum errors */
8624 /* It must be a TCP or UDP packet with a valid checksum */
8625 if (igb_test_staterr(rx_desc, E1000_RXD_STAT_TCPCS |
8626 E1000_RXD_STAT_UDPCS))
8627 skb->ip_summed = CHECKSUM_UNNECESSARY;
8629 dev_dbg(ring->dev, "cksum success: bits %08X\n",
8630 le32_to_cpu(rx_desc->wb.upper.status_error));
8633 static inline void igb_rx_hash(struct igb_ring *ring,
8634 union e1000_adv_rx_desc *rx_desc,
8635 struct sk_buff *skb)
8637 if (ring->netdev->features & NETIF_F_RXHASH)
8639 le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
8644 * igb_is_non_eop - process handling of non-EOP buffers
8645 * @rx_ring: Rx ring being processed
8646 * @rx_desc: Rx descriptor for current buffer
8648 * This function updates next to clean. If the buffer is an EOP buffer
8649 * this function exits returning false, otherwise it will place the
8650 * sk_buff in the next buffer to be chained and return true indicating
8651 * that this is in fact a non-EOP buffer.
8653 static bool igb_is_non_eop(struct igb_ring *rx_ring,
8654 union e1000_adv_rx_desc *rx_desc)
8656 u32 ntc = rx_ring->next_to_clean + 1;
8658 /* fetch, update, and store next to clean */
8659 ntc = (ntc < rx_ring->count) ? ntc : 0;
8660 rx_ring->next_to_clean = ntc;
8662 prefetch(IGB_RX_DESC(rx_ring, ntc));
8664 if (likely(igb_test_staterr(rx_desc, E1000_RXD_STAT_EOP)))
8671 * igb_cleanup_headers - Correct corrupted or empty headers
8672 * @rx_ring: rx descriptor ring packet is being transacted on
8673 * @rx_desc: pointer to the EOP Rx descriptor
8674 * @skb: pointer to current skb being fixed
8676 * Address the case where we are pulling data in on pages only
8677 * and as such no data is present in the skb header.
8679 * In addition if skb is not at least 60 bytes we need to pad it so that
8680 * it is large enough to qualify as a valid Ethernet frame.
8682 * Returns true if an error was encountered and skb was freed.
8684 static bool igb_cleanup_headers(struct igb_ring *rx_ring,
8685 union e1000_adv_rx_desc *rx_desc,
8686 struct sk_buff *skb)
8688 /* XDP packets use error pointer so abort at this point */
8692 if (unlikely((igb_test_staterr(rx_desc,
8693 E1000_RXDEXT_ERR_FRAME_ERR_MASK)))) {
8694 struct net_device *netdev = rx_ring->netdev;
8695 if (!(netdev->features & NETIF_F_RXALL)) {
8696 dev_kfree_skb_any(skb);
8701 /* if eth_skb_pad returns an error the skb was freed */
8702 if (eth_skb_pad(skb))
8709 * igb_process_skb_fields - Populate skb header fields from Rx descriptor
8710 * @rx_ring: rx descriptor ring packet is being transacted on
8711 * @rx_desc: pointer to the EOP Rx descriptor
8712 * @skb: pointer to current skb being populated
8714 * This function checks the ring, descriptor, and packet information in
8715 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
8716 * other fields within the skb.
8718 static void igb_process_skb_fields(struct igb_ring *rx_ring,
8719 union e1000_adv_rx_desc *rx_desc,
8720 struct sk_buff *skb)
8722 struct net_device *dev = rx_ring->netdev;
8724 igb_rx_hash(rx_ring, rx_desc, skb);
8726 igb_rx_checksum(rx_ring, rx_desc, skb);
8728 if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TS) &&
8729 !igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP))
8730 igb_ptp_rx_rgtstamp(rx_ring->q_vector, skb);
8732 if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
8733 igb_test_staterr(rx_desc, E1000_RXD_STAT_VP)) {
8736 if (igb_test_staterr(rx_desc, E1000_RXDEXT_STATERR_LB) &&
8737 test_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &rx_ring->flags))
8738 vid = be16_to_cpu((__force __be16)rx_desc->wb.upper.vlan);
8740 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
8742 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
8745 skb_record_rx_queue(skb, rx_ring->queue_index);
8747 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
8750 static unsigned int igb_rx_offset(struct igb_ring *rx_ring)
8752 return ring_uses_build_skb(rx_ring) ? IGB_SKB_PAD : 0;
8755 static struct igb_rx_buffer *igb_get_rx_buffer(struct igb_ring *rx_ring,
8756 const unsigned int size, int *rx_buf_pgcnt)
8758 struct igb_rx_buffer *rx_buffer;
8760 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
8762 #if (PAGE_SIZE < 8192)
8763 page_count(rx_buffer->page);
8767 prefetchw(rx_buffer->page);
8769 /* we are reusing so sync this buffer for CPU use */
8770 dma_sync_single_range_for_cpu(rx_ring->dev,
8772 rx_buffer->page_offset,
8776 rx_buffer->pagecnt_bias--;
8781 static void igb_put_rx_buffer(struct igb_ring *rx_ring,
8782 struct igb_rx_buffer *rx_buffer, int rx_buf_pgcnt)
8784 if (igb_can_reuse_rx_page(rx_buffer, rx_buf_pgcnt)) {
8785 /* hand second half of page back to the ring */
8786 igb_reuse_rx_page(rx_ring, rx_buffer);
8788 /* We are not reusing the buffer so unmap it and free
8789 * any references we are holding to it
8791 dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
8792 igb_rx_pg_size(rx_ring), DMA_FROM_DEVICE,
8794 __page_frag_cache_drain(rx_buffer->page,
8795 rx_buffer->pagecnt_bias);
8798 /* clear contents of rx_buffer */
8799 rx_buffer->page = NULL;
8802 static int igb_clean_rx_irq(struct igb_q_vector *q_vector, const int budget)
8804 struct igb_adapter *adapter = q_vector->adapter;
8805 struct igb_ring *rx_ring = q_vector->rx.ring;
8806 struct sk_buff *skb = rx_ring->skb;
8807 unsigned int total_bytes = 0, total_packets = 0;
8808 u16 cleaned_count = igb_desc_unused(rx_ring);
8809 unsigned int xdp_xmit = 0;
8810 struct xdp_buff xdp;
8814 /* Frame size depend on rx_ring setup when PAGE_SIZE=4K */
8815 #if (PAGE_SIZE < 8192)
8816 frame_sz = igb_rx_frame_truesize(rx_ring, 0);
8818 xdp_init_buff(&xdp, frame_sz, &rx_ring->xdp_rxq);
8820 while (likely(total_packets < budget)) {
8821 union e1000_adv_rx_desc *rx_desc;
8822 struct igb_rx_buffer *rx_buffer;
8823 ktime_t timestamp = 0;
8828 /* return some buffers to hardware, one at a time is too slow */
8829 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
8830 igb_alloc_rx_buffers(rx_ring, cleaned_count);
8834 rx_desc = IGB_RX_DESC(rx_ring, rx_ring->next_to_clean);
8835 size = le16_to_cpu(rx_desc->wb.upper.length);
8839 /* This memory barrier is needed to keep us from reading
8840 * any other fields out of the rx_desc until we know the
8841 * descriptor has been written back
8845 rx_buffer = igb_get_rx_buffer(rx_ring, size, &rx_buf_pgcnt);
8846 pktbuf = page_address(rx_buffer->page) + rx_buffer->page_offset;
8848 /* pull rx packet timestamp if available and valid */
8849 if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
8852 ts_hdr_len = igb_ptp_rx_pktstamp(rx_ring->q_vector,
8853 pktbuf, ×tamp);
8855 pkt_offset += ts_hdr_len;
8859 /* retrieve a buffer from the ring */
8861 unsigned char *hard_start = pktbuf - igb_rx_offset(rx_ring);
8862 unsigned int offset = pkt_offset + igb_rx_offset(rx_ring);
8864 xdp_prepare_buff(&xdp, hard_start, offset, size, true);
8865 xdp_buff_clear_frags_flag(&xdp);
8866 #if (PAGE_SIZE > 4096)
8867 /* At larger PAGE_SIZE, frame_sz depend on len size */
8868 xdp.frame_sz = igb_rx_frame_truesize(rx_ring, size);
8870 skb = igb_run_xdp(adapter, rx_ring, &xdp);
8874 unsigned int xdp_res = -PTR_ERR(skb);
8876 if (xdp_res & (IGB_XDP_TX | IGB_XDP_REDIR)) {
8877 xdp_xmit |= xdp_res;
8878 igb_rx_buffer_flip(rx_ring, rx_buffer, size);
8880 rx_buffer->pagecnt_bias++;
8883 total_bytes += size;
8885 igb_add_rx_frag(rx_ring, rx_buffer, skb, size);
8886 else if (ring_uses_build_skb(rx_ring))
8887 skb = igb_build_skb(rx_ring, rx_buffer, &xdp,
8890 skb = igb_construct_skb(rx_ring, rx_buffer,
8893 /* exit if we failed to retrieve a buffer */
8895 rx_ring->rx_stats.alloc_failed++;
8896 rx_buffer->pagecnt_bias++;
8900 igb_put_rx_buffer(rx_ring, rx_buffer, rx_buf_pgcnt);
8903 /* fetch next buffer in frame if non-eop */
8904 if (igb_is_non_eop(rx_ring, rx_desc))
8907 /* verify the packet layout is correct */
8908 if (igb_cleanup_headers(rx_ring, rx_desc, skb)) {
8913 /* probably a little skewed due to removing CRC */
8914 total_bytes += skb->len;
8916 /* populate checksum, timestamp, VLAN, and protocol */
8917 igb_process_skb_fields(rx_ring, rx_desc, skb);
8919 napi_gro_receive(&q_vector->napi, skb);
8921 /* reset skb pointer */
8924 /* update budget accounting */
8928 /* place incomplete frames back on ring for completion */
8931 if (xdp_xmit & IGB_XDP_REDIR)
8934 if (xdp_xmit & IGB_XDP_TX) {
8935 struct igb_ring *tx_ring = igb_xdp_tx_queue_mapping(adapter);
8937 igb_xdp_ring_update_tail(tx_ring);
8940 u64_stats_update_begin(&rx_ring->rx_syncp);
8941 rx_ring->rx_stats.packets += total_packets;
8942 rx_ring->rx_stats.bytes += total_bytes;
8943 u64_stats_update_end(&rx_ring->rx_syncp);
8944 q_vector->rx.total_packets += total_packets;
8945 q_vector->rx.total_bytes += total_bytes;
8948 igb_alloc_rx_buffers(rx_ring, cleaned_count);
8950 return total_packets;
8953 static bool igb_alloc_mapped_page(struct igb_ring *rx_ring,
8954 struct igb_rx_buffer *bi)
8956 struct page *page = bi->page;
8959 /* since we are recycling buffers we should seldom need to alloc */
8963 /* alloc new page for storage */
8964 page = dev_alloc_pages(igb_rx_pg_order(rx_ring));
8965 if (unlikely(!page)) {
8966 rx_ring->rx_stats.alloc_failed++;
8970 /* map page for use */
8971 dma = dma_map_page_attrs(rx_ring->dev, page, 0,
8972 igb_rx_pg_size(rx_ring),
8976 /* if mapping failed free memory back to system since
8977 * there isn't much point in holding memory we can't use
8979 if (dma_mapping_error(rx_ring->dev, dma)) {
8980 __free_pages(page, igb_rx_pg_order(rx_ring));
8982 rx_ring->rx_stats.alloc_failed++;
8988 bi->page_offset = igb_rx_offset(rx_ring);
8989 page_ref_add(page, USHRT_MAX - 1);
8990 bi->pagecnt_bias = USHRT_MAX;
8996 * igb_alloc_rx_buffers - Replace used receive buffers
8997 * @rx_ring: rx descriptor ring to allocate new receive buffers
8998 * @cleaned_count: count of buffers to allocate
9000 void igb_alloc_rx_buffers(struct igb_ring *rx_ring, u16 cleaned_count)
9002 union e1000_adv_rx_desc *rx_desc;
9003 struct igb_rx_buffer *bi;
9004 u16 i = rx_ring->next_to_use;
9011 rx_desc = IGB_RX_DESC(rx_ring, i);
9012 bi = &rx_ring->rx_buffer_info[i];
9013 i -= rx_ring->count;
9015 bufsz = igb_rx_bufsz(rx_ring);
9018 if (!igb_alloc_mapped_page(rx_ring, bi))
9021 /* sync the buffer for use by the device */
9022 dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
9023 bi->page_offset, bufsz,
9026 /* Refresh the desc even if buffer_addrs didn't change
9027 * because each write-back erases this info.
9029 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
9035 rx_desc = IGB_RX_DESC(rx_ring, 0);
9036 bi = rx_ring->rx_buffer_info;
9037 i -= rx_ring->count;
9040 /* clear the length for the next_to_use descriptor */
9041 rx_desc->wb.upper.length = 0;
9044 } while (cleaned_count);
9046 i += rx_ring->count;
9048 if (rx_ring->next_to_use != i) {
9049 /* record the next descriptor to use */
9050 rx_ring->next_to_use = i;
9052 /* update next to alloc since we have filled the ring */
9053 rx_ring->next_to_alloc = i;
9055 /* Force memory writes to complete before letting h/w
9056 * know there are new descriptors to fetch. (Only
9057 * applicable for weak-ordered memory model archs,
9061 writel(i, rx_ring->tail);
9067 * @netdev: pointer to netdev struct
9068 * @ifr: interface structure
9069 * @cmd: ioctl command to execute
9071 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
9073 struct igb_adapter *adapter = netdev_priv(netdev);
9074 struct mii_ioctl_data *data = if_mii(ifr);
9076 if (adapter->hw.phy.media_type != e1000_media_type_copper)
9081 data->phy_id = adapter->hw.phy.addr;
9084 if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
9097 * @netdev: pointer to netdev struct
9098 * @ifr: interface structure
9099 * @cmd: ioctl command to execute
9101 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
9107 return igb_mii_ioctl(netdev, ifr, cmd);
9109 return igb_ptp_get_ts_config(netdev, ifr);
9111 return igb_ptp_set_ts_config(netdev, ifr);
9117 void igb_read_pci_cfg(struct e1000_hw *hw, u32 reg, u16 *value)
9119 struct igb_adapter *adapter = hw->back;
9121 pci_read_config_word(adapter->pdev, reg, value);
9124 void igb_write_pci_cfg(struct e1000_hw *hw, u32 reg, u16 *value)
9126 struct igb_adapter *adapter = hw->back;
9128 pci_write_config_word(adapter->pdev, reg, *value);
9131 s32 igb_read_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
9133 struct igb_adapter *adapter = hw->back;
9135 if (pcie_capability_read_word(adapter->pdev, reg, value))
9136 return -E1000_ERR_CONFIG;
9141 s32 igb_write_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
9143 struct igb_adapter *adapter = hw->back;
9145 if (pcie_capability_write_word(adapter->pdev, reg, *value))
9146 return -E1000_ERR_CONFIG;
9151 static void igb_vlan_mode(struct net_device *netdev, netdev_features_t features)
9153 struct igb_adapter *adapter = netdev_priv(netdev);
9154 struct e1000_hw *hw = &adapter->hw;
9156 bool enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
9159 /* enable VLAN tag insert/strip */
9160 ctrl = rd32(E1000_CTRL);
9161 ctrl |= E1000_CTRL_VME;
9162 wr32(E1000_CTRL, ctrl);
9164 /* Disable CFI check */
9165 rctl = rd32(E1000_RCTL);
9166 rctl &= ~E1000_RCTL_CFIEN;
9167 wr32(E1000_RCTL, rctl);
9169 /* disable VLAN tag insert/strip */
9170 ctrl = rd32(E1000_CTRL);
9171 ctrl &= ~E1000_CTRL_VME;
9172 wr32(E1000_CTRL, ctrl);
9175 igb_set_vf_vlan_strip(adapter, adapter->vfs_allocated_count, enable);
9178 static int igb_vlan_rx_add_vid(struct net_device *netdev,
9179 __be16 proto, u16 vid)
9181 struct igb_adapter *adapter = netdev_priv(netdev);
9182 struct e1000_hw *hw = &adapter->hw;
9183 int pf_id = adapter->vfs_allocated_count;
9185 /* add the filter since PF can receive vlans w/o entry in vlvf */
9186 if (!vid || !(adapter->flags & IGB_FLAG_VLAN_PROMISC))
9187 igb_vfta_set(hw, vid, pf_id, true, !!vid);
9189 set_bit(vid, adapter->active_vlans);
9194 static int igb_vlan_rx_kill_vid(struct net_device *netdev,
9195 __be16 proto, u16 vid)
9197 struct igb_adapter *adapter = netdev_priv(netdev);
9198 int pf_id = adapter->vfs_allocated_count;
9199 struct e1000_hw *hw = &adapter->hw;
9201 /* remove VID from filter table */
9202 if (vid && !(adapter->flags & IGB_FLAG_VLAN_PROMISC))
9203 igb_vfta_set(hw, vid, pf_id, false, true);
9205 clear_bit(vid, adapter->active_vlans);
9210 static void igb_restore_vlan(struct igb_adapter *adapter)
9214 igb_vlan_mode(adapter->netdev, adapter->netdev->features);
9215 igb_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
9217 for_each_set_bit_from(vid, adapter->active_vlans, VLAN_N_VID)
9218 igb_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
9221 int igb_set_spd_dplx(struct igb_adapter *adapter, u32 spd, u8 dplx)
9223 struct pci_dev *pdev = adapter->pdev;
9224 struct e1000_mac_info *mac = &adapter->hw.mac;
9228 /* Make sure dplx is at most 1 bit and lsb of speed is not set
9229 * for the switch() below to work
9231 if ((spd & 1) || (dplx & ~1))
9234 /* Fiber NIC's only allow 1000 gbps Full duplex
9235 * and 100Mbps Full duplex for 100baseFx sfp
9237 if (adapter->hw.phy.media_type == e1000_media_type_internal_serdes) {
9238 switch (spd + dplx) {
9239 case SPEED_10 + DUPLEX_HALF:
9240 case SPEED_10 + DUPLEX_FULL:
9241 case SPEED_100 + DUPLEX_HALF:
9248 switch (spd + dplx) {
9249 case SPEED_10 + DUPLEX_HALF:
9250 mac->forced_speed_duplex = ADVERTISE_10_HALF;
9252 case SPEED_10 + DUPLEX_FULL:
9253 mac->forced_speed_duplex = ADVERTISE_10_FULL;
9255 case SPEED_100 + DUPLEX_HALF:
9256 mac->forced_speed_duplex = ADVERTISE_100_HALF;
9258 case SPEED_100 + DUPLEX_FULL:
9259 mac->forced_speed_duplex = ADVERTISE_100_FULL;
9261 case SPEED_1000 + DUPLEX_FULL:
9263 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
9265 case SPEED_1000 + DUPLEX_HALF: /* not supported */
9270 /* clear MDI, MDI(-X) override is only allowed when autoneg enabled */
9271 adapter->hw.phy.mdix = AUTO_ALL_MODES;
9276 dev_err(&pdev->dev, "Unsupported Speed/Duplex configuration\n");
9280 static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake,
9283 struct net_device *netdev = pci_get_drvdata(pdev);
9284 struct igb_adapter *adapter = netdev_priv(netdev);
9285 struct e1000_hw *hw = &adapter->hw;
9286 u32 ctrl, rctl, status;
9287 u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
9291 netif_device_detach(netdev);
9293 if (netif_running(netdev))
9294 __igb_close(netdev, true);
9296 igb_ptp_suspend(adapter);
9298 igb_clear_interrupt_scheme(adapter);
9301 status = rd32(E1000_STATUS);
9302 if (status & E1000_STATUS_LU)
9303 wufc &= ~E1000_WUFC_LNKC;
9306 igb_setup_rctl(adapter);
9307 igb_set_rx_mode(netdev);
9309 /* turn on all-multi mode if wake on multicast is enabled */
9310 if (wufc & E1000_WUFC_MC) {
9311 rctl = rd32(E1000_RCTL);
9312 rctl |= E1000_RCTL_MPE;
9313 wr32(E1000_RCTL, rctl);
9316 ctrl = rd32(E1000_CTRL);
9317 ctrl |= E1000_CTRL_ADVD3WUC;
9318 wr32(E1000_CTRL, ctrl);
9320 /* Allow time for pending master requests to run */
9321 igb_disable_pcie_master(hw);
9323 wr32(E1000_WUC, E1000_WUC_PME_EN);
9324 wr32(E1000_WUFC, wufc);
9327 wr32(E1000_WUFC, 0);
9330 wake = wufc || adapter->en_mng_pt;
9332 igb_power_down_link(adapter);
9334 igb_power_up_link(adapter);
9337 *enable_wake = wake;
9339 /* Release control of h/w to f/w. If f/w is AMT enabled, this
9340 * would have already happened in close and is redundant.
9342 igb_release_hw_control(adapter);
9344 pci_disable_device(pdev);
9349 static void igb_deliver_wake_packet(struct net_device *netdev)
9351 struct igb_adapter *adapter = netdev_priv(netdev);
9352 struct e1000_hw *hw = &adapter->hw;
9353 struct sk_buff *skb;
9356 wupl = rd32(E1000_WUPL) & E1000_WUPL_MASK;
9358 /* WUPM stores only the first 128 bytes of the wake packet.
9359 * Read the packet only if we have the whole thing.
9361 if ((wupl == 0) || (wupl > E1000_WUPM_BYTES))
9364 skb = netdev_alloc_skb_ip_align(netdev, E1000_WUPM_BYTES);
9370 /* Ensure reads are 32-bit aligned */
9371 wupl = roundup(wupl, 4);
9373 memcpy_fromio(skb->data, hw->hw_addr + E1000_WUPM_REG(0), wupl);
9375 skb->protocol = eth_type_trans(skb, netdev);
9379 static int __maybe_unused igb_suspend(struct device *dev)
9381 return __igb_shutdown(to_pci_dev(dev), NULL, 0);
9384 static int __maybe_unused __igb_resume(struct device *dev, bool rpm)
9386 struct pci_dev *pdev = to_pci_dev(dev);
9387 struct net_device *netdev = pci_get_drvdata(pdev);
9388 struct igb_adapter *adapter = netdev_priv(netdev);
9389 struct e1000_hw *hw = &adapter->hw;
9392 pci_set_power_state(pdev, PCI_D0);
9393 pci_restore_state(pdev);
9394 pci_save_state(pdev);
9396 if (!pci_device_is_present(pdev))
9398 err = pci_enable_device_mem(pdev);
9401 "igb: Cannot enable PCI device from suspend\n");
9404 pci_set_master(pdev);
9406 pci_enable_wake(pdev, PCI_D3hot, 0);
9407 pci_enable_wake(pdev, PCI_D3cold, 0);
9409 if (igb_init_interrupt_scheme(adapter, true)) {
9410 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
9416 /* let the f/w know that the h/w is now under the control of the
9419 igb_get_hw_control(adapter);
9421 val = rd32(E1000_WUS);
9422 if (val & WAKE_PKT_WUS)
9423 igb_deliver_wake_packet(netdev);
9425 wr32(E1000_WUS, ~0);
9429 if (!err && netif_running(netdev))
9430 err = __igb_open(netdev, true);
9433 netif_device_attach(netdev);
9440 static int __maybe_unused igb_resume(struct device *dev)
9442 return __igb_resume(dev, false);
9445 static int __maybe_unused igb_runtime_idle(struct device *dev)
9447 struct net_device *netdev = dev_get_drvdata(dev);
9448 struct igb_adapter *adapter = netdev_priv(netdev);
9450 if (!igb_has_link(adapter))
9451 pm_schedule_suspend(dev, MSEC_PER_SEC * 5);
9456 static int __maybe_unused igb_runtime_suspend(struct device *dev)
9458 return __igb_shutdown(to_pci_dev(dev), NULL, 1);
9461 static int __maybe_unused igb_runtime_resume(struct device *dev)
9463 return __igb_resume(dev, true);
9466 static void igb_shutdown(struct pci_dev *pdev)
9470 __igb_shutdown(pdev, &wake, 0);
9472 if (system_state == SYSTEM_POWER_OFF) {
9473 pci_wake_from_d3(pdev, wake);
9474 pci_set_power_state(pdev, PCI_D3hot);
9478 #ifdef CONFIG_PCI_IOV
9479 static int igb_sriov_reinit(struct pci_dev *dev)
9481 struct net_device *netdev = pci_get_drvdata(dev);
9482 struct igb_adapter *adapter = netdev_priv(netdev);
9483 struct pci_dev *pdev = adapter->pdev;
9487 if (netif_running(netdev))
9492 igb_clear_interrupt_scheme(adapter);
9494 igb_init_queue_configuration(adapter);
9496 if (igb_init_interrupt_scheme(adapter, true)) {
9498 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
9502 if (netif_running(netdev))
9510 static int igb_pci_disable_sriov(struct pci_dev *dev)
9512 int err = igb_disable_sriov(dev);
9515 err = igb_sriov_reinit(dev);
9520 static int igb_pci_enable_sriov(struct pci_dev *dev, int num_vfs)
9522 int err = igb_enable_sriov(dev, num_vfs);
9527 err = igb_sriov_reinit(dev);
9536 static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs)
9538 #ifdef CONFIG_PCI_IOV
9540 return igb_pci_disable_sriov(dev);
9542 return igb_pci_enable_sriov(dev, num_vfs);
9548 * igb_io_error_detected - called when PCI error is detected
9549 * @pdev: Pointer to PCI device
9550 * @state: The current pci connection state
9552 * This function is called after a PCI bus error affecting
9553 * this device has been detected.
9555 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
9556 pci_channel_state_t state)
9558 struct net_device *netdev = pci_get_drvdata(pdev);
9559 struct igb_adapter *adapter = netdev_priv(netdev);
9561 netif_device_detach(netdev);
9563 if (state == pci_channel_io_perm_failure)
9564 return PCI_ERS_RESULT_DISCONNECT;
9566 if (netif_running(netdev))
9568 pci_disable_device(pdev);
9570 /* Request a slot reset. */
9571 return PCI_ERS_RESULT_NEED_RESET;
9575 * igb_io_slot_reset - called after the pci bus has been reset.
9576 * @pdev: Pointer to PCI device
9578 * Restart the card from scratch, as if from a cold-boot. Implementation
9579 * resembles the first-half of the __igb_resume routine.
9581 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
9583 struct net_device *netdev = pci_get_drvdata(pdev);
9584 struct igb_adapter *adapter = netdev_priv(netdev);
9585 struct e1000_hw *hw = &adapter->hw;
9586 pci_ers_result_t result;
9588 if (pci_enable_device_mem(pdev)) {
9590 "Cannot re-enable PCI device after reset.\n");
9591 result = PCI_ERS_RESULT_DISCONNECT;
9593 pci_set_master(pdev);
9594 pci_restore_state(pdev);
9595 pci_save_state(pdev);
9597 pci_enable_wake(pdev, PCI_D3hot, 0);
9598 pci_enable_wake(pdev, PCI_D3cold, 0);
9600 /* In case of PCI error, adapter lose its HW address
9601 * so we should re-assign it here.
9603 hw->hw_addr = adapter->io_addr;
9606 wr32(E1000_WUS, ~0);
9607 result = PCI_ERS_RESULT_RECOVERED;
9614 * igb_io_resume - called when traffic can start flowing again.
9615 * @pdev: Pointer to PCI device
9617 * This callback is called when the error recovery driver tells us that
9618 * its OK to resume normal operation. Implementation resembles the
9619 * second-half of the __igb_resume routine.
9621 static void igb_io_resume(struct pci_dev *pdev)
9623 struct net_device *netdev = pci_get_drvdata(pdev);
9624 struct igb_adapter *adapter = netdev_priv(netdev);
9626 if (netif_running(netdev)) {
9627 if (igb_up(adapter)) {
9628 dev_err(&pdev->dev, "igb_up failed after reset\n");
9633 netif_device_attach(netdev);
9635 /* let the f/w know that the h/w is now under the control of the
9638 igb_get_hw_control(adapter);
9642 * igb_rar_set_index - Sync RAL[index] and RAH[index] registers with MAC table
9643 * @adapter: Pointer to adapter structure
9644 * @index: Index of the RAR entry which need to be synced with MAC table
9646 static void igb_rar_set_index(struct igb_adapter *adapter, u32 index)
9648 struct e1000_hw *hw = &adapter->hw;
9649 u32 rar_low, rar_high;
9650 u8 *addr = adapter->mac_table[index].addr;
9652 /* HW expects these to be in network order when they are plugged
9653 * into the registers which are little endian. In order to guarantee
9654 * that ordering we need to do an leXX_to_cpup here in order to be
9655 * ready for the byteswap that occurs with writel
9657 rar_low = le32_to_cpup((__le32 *)(addr));
9658 rar_high = le16_to_cpup((__le16 *)(addr + 4));
9660 /* Indicate to hardware the Address is Valid. */
9661 if (adapter->mac_table[index].state & IGB_MAC_STATE_IN_USE) {
9662 if (is_valid_ether_addr(addr))
9663 rar_high |= E1000_RAH_AV;
9665 if (adapter->mac_table[index].state & IGB_MAC_STATE_SRC_ADDR)
9666 rar_high |= E1000_RAH_ASEL_SRC_ADDR;
9668 switch (hw->mac.type) {
9671 if (adapter->mac_table[index].state &
9672 IGB_MAC_STATE_QUEUE_STEERING)
9673 rar_high |= E1000_RAH_QSEL_ENABLE;
9675 rar_high |= E1000_RAH_POOL_1 *
9676 adapter->mac_table[index].queue;
9679 rar_high |= E1000_RAH_POOL_1 <<
9680 adapter->mac_table[index].queue;
9685 wr32(E1000_RAL(index), rar_low);
9687 wr32(E1000_RAH(index), rar_high);
9691 static int igb_set_vf_mac(struct igb_adapter *adapter,
9692 int vf, unsigned char *mac_addr)
9694 struct e1000_hw *hw = &adapter->hw;
9695 /* VF MAC addresses start at end of receive addresses and moves
9696 * towards the first, as a result a collision should not be possible
9698 int rar_entry = hw->mac.rar_entry_count - (vf + 1);
9699 unsigned char *vf_mac_addr = adapter->vf_data[vf].vf_mac_addresses;
9701 ether_addr_copy(vf_mac_addr, mac_addr);
9702 ether_addr_copy(adapter->mac_table[rar_entry].addr, mac_addr);
9703 adapter->mac_table[rar_entry].queue = vf;
9704 adapter->mac_table[rar_entry].state |= IGB_MAC_STATE_IN_USE;
9705 igb_rar_set_index(adapter, rar_entry);
9710 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
9712 struct igb_adapter *adapter = netdev_priv(netdev);
9714 if (vf >= adapter->vfs_allocated_count)
9717 /* Setting the VF MAC to 0 reverts the IGB_VF_FLAG_PF_SET_MAC
9718 * flag and allows to overwrite the MAC via VF netdev. This
9719 * is necessary to allow libvirt a way to restore the original
9720 * MAC after unbinding vfio-pci and reloading igbvf after shutting
9723 if (is_zero_ether_addr(mac)) {
9724 adapter->vf_data[vf].flags &= ~IGB_VF_FLAG_PF_SET_MAC;
9725 dev_info(&adapter->pdev->dev,
9726 "remove administratively set MAC on VF %d\n",
9728 } else if (is_valid_ether_addr(mac)) {
9729 adapter->vf_data[vf].flags |= IGB_VF_FLAG_PF_SET_MAC;
9730 dev_info(&adapter->pdev->dev, "setting MAC %pM on VF %d\n",
9732 dev_info(&adapter->pdev->dev,
9733 "Reload the VF driver to make this change effective.");
9734 /* Generate additional warning if PF is down */
9735 if (test_bit(__IGB_DOWN, &adapter->state)) {
9736 dev_warn(&adapter->pdev->dev,
9737 "The VF MAC address has been set, but the PF device is not up.\n");
9738 dev_warn(&adapter->pdev->dev,
9739 "Bring the PF device up before attempting to use the VF device.\n");
9744 return igb_set_vf_mac(adapter, vf, mac);
9747 static int igb_link_mbps(int internal_link_speed)
9749 switch (internal_link_speed) {
9759 static void igb_set_vf_rate_limit(struct e1000_hw *hw, int vf, int tx_rate,
9766 /* Calculate the rate factor values to set */
9767 rf_int = link_speed / tx_rate;
9768 rf_dec = (link_speed - (rf_int * tx_rate));
9769 rf_dec = (rf_dec * BIT(E1000_RTTBCNRC_RF_INT_SHIFT)) /
9772 bcnrc_val = E1000_RTTBCNRC_RS_ENA;
9773 bcnrc_val |= ((rf_int << E1000_RTTBCNRC_RF_INT_SHIFT) &
9774 E1000_RTTBCNRC_RF_INT_MASK);
9775 bcnrc_val |= (rf_dec & E1000_RTTBCNRC_RF_DEC_MASK);
9780 wr32(E1000_RTTDQSEL, vf); /* vf X uses queue X */
9781 /* Set global transmit compensation time to the MMW_SIZE in RTTBCNRM
9782 * register. MMW_SIZE=0x014 if 9728-byte jumbo is supported.
9784 wr32(E1000_RTTBCNRM, 0x14);
9785 wr32(E1000_RTTBCNRC, bcnrc_val);
9788 static void igb_check_vf_rate_limit(struct igb_adapter *adapter)
9790 int actual_link_speed, i;
9791 bool reset_rate = false;
9793 /* VF TX rate limit was not set or not supported */
9794 if ((adapter->vf_rate_link_speed == 0) ||
9795 (adapter->hw.mac.type != e1000_82576))
9798 actual_link_speed = igb_link_mbps(adapter->link_speed);
9799 if (actual_link_speed != adapter->vf_rate_link_speed) {
9801 adapter->vf_rate_link_speed = 0;
9802 dev_info(&adapter->pdev->dev,
9803 "Link speed has been changed. VF Transmit rate is disabled\n");
9806 for (i = 0; i < adapter->vfs_allocated_count; i++) {
9808 adapter->vf_data[i].tx_rate = 0;
9810 igb_set_vf_rate_limit(&adapter->hw, i,
9811 adapter->vf_data[i].tx_rate,
9816 static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf,
9817 int min_tx_rate, int max_tx_rate)
9819 struct igb_adapter *adapter = netdev_priv(netdev);
9820 struct e1000_hw *hw = &adapter->hw;
9821 int actual_link_speed;
9823 if (hw->mac.type != e1000_82576)
9829 actual_link_speed = igb_link_mbps(adapter->link_speed);
9830 if ((vf >= adapter->vfs_allocated_count) ||
9831 (!(rd32(E1000_STATUS) & E1000_STATUS_LU)) ||
9832 (max_tx_rate < 0) ||
9833 (max_tx_rate > actual_link_speed))
9836 adapter->vf_rate_link_speed = actual_link_speed;
9837 adapter->vf_data[vf].tx_rate = (u16)max_tx_rate;
9838 igb_set_vf_rate_limit(hw, vf, max_tx_rate, actual_link_speed);
9843 static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf,
9846 struct igb_adapter *adapter = netdev_priv(netdev);
9847 struct e1000_hw *hw = &adapter->hw;
9848 u32 reg_val, reg_offset;
9850 if (!adapter->vfs_allocated_count)
9853 if (vf >= adapter->vfs_allocated_count)
9856 reg_offset = (hw->mac.type == e1000_82576) ? E1000_DTXSWC : E1000_TXSWC;
9857 reg_val = rd32(reg_offset);
9859 reg_val |= (BIT(vf) |
9860 BIT(vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT));
9862 reg_val &= ~(BIT(vf) |
9863 BIT(vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT));
9864 wr32(reg_offset, reg_val);
9866 adapter->vf_data[vf].spoofchk_enabled = setting;
9870 static int igb_ndo_set_vf_trust(struct net_device *netdev, int vf, bool setting)
9872 struct igb_adapter *adapter = netdev_priv(netdev);
9874 if (vf >= adapter->vfs_allocated_count)
9876 if (adapter->vf_data[vf].trusted == setting)
9879 adapter->vf_data[vf].trusted = setting;
9881 dev_info(&adapter->pdev->dev, "VF %u is %strusted\n",
9882 vf, setting ? "" : "not ");
9886 static int igb_ndo_get_vf_config(struct net_device *netdev,
9887 int vf, struct ifla_vf_info *ivi)
9889 struct igb_adapter *adapter = netdev_priv(netdev);
9890 if (vf >= adapter->vfs_allocated_count)
9893 memcpy(&ivi->mac, adapter->vf_data[vf].vf_mac_addresses, ETH_ALEN);
9894 ivi->max_tx_rate = adapter->vf_data[vf].tx_rate;
9895 ivi->min_tx_rate = 0;
9896 ivi->vlan = adapter->vf_data[vf].pf_vlan;
9897 ivi->qos = adapter->vf_data[vf].pf_qos;
9898 ivi->spoofchk = adapter->vf_data[vf].spoofchk_enabled;
9899 ivi->trusted = adapter->vf_data[vf].trusted;
9903 static void igb_vmm_control(struct igb_adapter *adapter)
9905 struct e1000_hw *hw = &adapter->hw;
9908 switch (hw->mac.type) {
9914 /* replication is not supported for 82575 */
9917 /* notify HW that the MAC is adding vlan tags */
9918 reg = rd32(E1000_DTXCTL);
9919 reg |= E1000_DTXCTL_VLAN_ADDED;
9920 wr32(E1000_DTXCTL, reg);
9923 /* enable replication vlan tag stripping */
9924 reg = rd32(E1000_RPLOLR);
9925 reg |= E1000_RPLOLR_STRVLAN;
9926 wr32(E1000_RPLOLR, reg);
9929 /* none of the above registers are supported by i350 */
9933 if (adapter->vfs_allocated_count) {
9934 igb_vmdq_set_loopback_pf(hw, true);
9935 igb_vmdq_set_replication_pf(hw, true);
9936 igb_vmdq_set_anti_spoofing_pf(hw, true,
9937 adapter->vfs_allocated_count);
9939 igb_vmdq_set_loopback_pf(hw, false);
9940 igb_vmdq_set_replication_pf(hw, false);
9944 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba)
9946 struct e1000_hw *hw = &adapter->hw;
9951 if (hw->mac.type > e1000_82580) {
9952 if (adapter->flags & IGB_FLAG_DMAC) {
9953 /* force threshold to 0. */
9954 wr32(E1000_DMCTXTH, 0);
9956 /* DMA Coalescing high water mark needs to be greater
9957 * than the Rx threshold. Set hwm to PBA - max frame
9958 * size in 16B units, capping it at PBA - 6KB.
9960 hwm = 64 * (pba - 6);
9961 reg = rd32(E1000_FCRTC);
9962 reg &= ~E1000_FCRTC_RTH_COAL_MASK;
9963 reg |= ((hwm << E1000_FCRTC_RTH_COAL_SHIFT)
9964 & E1000_FCRTC_RTH_COAL_MASK);
9965 wr32(E1000_FCRTC, reg);
9967 /* Set the DMA Coalescing Rx threshold to PBA - 2 * max
9968 * frame size, capping it at PBA - 10KB.
9970 dmac_thr = pba - 10;
9971 reg = rd32(E1000_DMACR);
9972 reg &= ~E1000_DMACR_DMACTHR_MASK;
9973 reg |= ((dmac_thr << E1000_DMACR_DMACTHR_SHIFT)
9974 & E1000_DMACR_DMACTHR_MASK);
9976 /* transition to L0x or L1 if available..*/
9977 reg |= (E1000_DMACR_DMAC_EN | E1000_DMACR_DMAC_LX_MASK);
9979 /* watchdog timer= +-1000 usec in 32usec intervals */
9982 /* Disable BMC-to-OS Watchdog Enable */
9983 if (hw->mac.type != e1000_i354)
9984 reg &= ~E1000_DMACR_DC_BMC2OSW_EN;
9985 wr32(E1000_DMACR, reg);
9987 /* no lower threshold to disable
9988 * coalescing(smart fifb)-UTRESH=0
9990 wr32(E1000_DMCRTRH, 0);
9992 reg = (IGB_DMCTLX_DCFLUSH_DIS | 0x4);
9994 wr32(E1000_DMCTLX, reg);
9996 /* free space in tx packet buffer to wake from
9999 wr32(E1000_DMCTXTH, (IGB_MIN_TXPBSIZE -
10000 (IGB_TX_BUF_4096 + adapter->max_frame_size)) >> 6);
10003 if (hw->mac.type >= e1000_i210 ||
10004 (adapter->flags & IGB_FLAG_DMAC)) {
10005 reg = rd32(E1000_PCIEMISC);
10006 reg |= E1000_PCIEMISC_LX_DECISION;
10007 wr32(E1000_PCIEMISC, reg);
10008 } /* endif adapter->dmac is not disabled */
10009 } else if (hw->mac.type == e1000_82580) {
10010 u32 reg = rd32(E1000_PCIEMISC);
10012 wr32(E1000_PCIEMISC, reg & ~E1000_PCIEMISC_LX_DECISION);
10013 wr32(E1000_DMACR, 0);
10018 * igb_read_i2c_byte - Reads 8 bit word over I2C
10019 * @hw: pointer to hardware structure
10020 * @byte_offset: byte offset to read
10021 * @dev_addr: device address
10022 * @data: value read
10024 * Performs byte read operation over I2C interface at
10025 * a specified device address.
10027 s32 igb_read_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
10028 u8 dev_addr, u8 *data)
10030 struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
10031 struct i2c_client *this_client = adapter->i2c_client;
10036 return E1000_ERR_I2C;
10038 swfw_mask = E1000_SWFW_PHY0_SM;
10040 if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask))
10041 return E1000_ERR_SWFW_SYNC;
10043 status = i2c_smbus_read_byte_data(this_client, byte_offset);
10044 hw->mac.ops.release_swfw_sync(hw, swfw_mask);
10047 return E1000_ERR_I2C;
10055 * igb_write_i2c_byte - Writes 8 bit word over I2C
10056 * @hw: pointer to hardware structure
10057 * @byte_offset: byte offset to write
10058 * @dev_addr: device address
10059 * @data: value to write
10061 * Performs byte write operation over I2C interface at
10062 * a specified device address.
10064 s32 igb_write_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;
10070 u16 swfw_mask = E1000_SWFW_PHY0_SM;
10073 return E1000_ERR_I2C;
10075 if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask))
10076 return E1000_ERR_SWFW_SYNC;
10077 status = i2c_smbus_write_byte_data(this_client, byte_offset, data);
10078 hw->mac.ops.release_swfw_sync(hw, swfw_mask);
10081 return E1000_ERR_I2C;
10087 int igb_reinit_queues(struct igb_adapter *adapter)
10089 struct net_device *netdev = adapter->netdev;
10090 struct pci_dev *pdev = adapter->pdev;
10093 if (netif_running(netdev))
10096 igb_reset_interrupt_capability(adapter);
10098 if (igb_init_interrupt_scheme(adapter, true)) {
10099 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
10103 if (netif_running(netdev))
10104 err = igb_open(netdev);
10109 static void igb_nfc_filter_exit(struct igb_adapter *adapter)
10111 struct igb_nfc_filter *rule;
10113 spin_lock(&adapter->nfc_lock);
10115 hlist_for_each_entry(rule, &adapter->nfc_filter_list, nfc_node)
10116 igb_erase_filter(adapter, rule);
10118 hlist_for_each_entry(rule, &adapter->cls_flower_list, nfc_node)
10119 igb_erase_filter(adapter, rule);
10121 spin_unlock(&adapter->nfc_lock);
10124 static void igb_nfc_filter_restore(struct igb_adapter *adapter)
10126 struct igb_nfc_filter *rule;
10128 spin_lock(&adapter->nfc_lock);
10130 hlist_for_each_entry(rule, &adapter->nfc_filter_list, nfc_node)
10131 igb_add_filter(adapter, rule);
10133 spin_unlock(&adapter->nfc_lock);