1 /*******************************************************************************
3 Intel(R) Gigabit Ethernet Linux driver
4 Copyright(c) 2007-2013 Intel Corporation.
6 This program is free software; you can redistribute it and/or modify it
7 under the terms and conditions of the GNU General Public License,
8 version 2, as published by the Free Software Foundation.
10 This program is distributed in the hope it will be useful, but WITHOUT
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
23 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26 *******************************************************************************/
28 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
30 #include <linux/module.h>
31 #include <linux/types.h>
32 #include <linux/init.h>
33 #include <linux/bitops.h>
34 #include <linux/vmalloc.h>
35 #include <linux/pagemap.h>
36 #include <linux/netdevice.h>
37 #include <linux/ipv6.h>
38 #include <linux/slab.h>
39 #include <net/checksum.h>
40 #include <net/ip6_checksum.h>
41 #include <linux/net_tstamp.h>
42 #include <linux/mii.h>
43 #include <linux/ethtool.h>
45 #include <linux/if_vlan.h>
46 #include <linux/pci.h>
47 #include <linux/pci-aspm.h>
48 #include <linux/delay.h>
49 #include <linux/interrupt.h>
51 #include <linux/tcp.h>
52 #include <linux/sctp.h>
53 #include <linux/if_ether.h>
54 #include <linux/aer.h>
55 #include <linux/prefetch.h>
56 #include <linux/pm_runtime.h>
58 #include <linux/dca.h>
60 #include <linux/i2c.h>
66 #define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) "." \
67 __stringify(BUILD) "-k"
68 char igb_driver_name[] = "igb";
69 char igb_driver_version[] = DRV_VERSION;
70 static const char igb_driver_string[] =
71 "Intel(R) Gigabit Ethernet Network Driver";
72 static const char igb_copyright[] =
73 "Copyright (c) 2007-2013 Intel Corporation.";
75 static const struct e1000_info *igb_info_tbl[] = {
76 [board_82575] = &e1000_82575_info,
79 static DEFINE_PCI_DEVICE_TABLE(igb_pci_tbl) = {
80 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_BACKPLANE_1GBPS) },
81 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_SGMII) },
82 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_BACKPLANE_2_5GBPS) },
83 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I211_COPPER), board_82575 },
84 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER), board_82575 },
85 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_FIBER), board_82575 },
86 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES), board_82575 },
87 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SGMII), board_82575 },
88 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER_FLASHLESS), board_82575 },
89 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES_FLASHLESS), board_82575 },
90 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_COPPER), board_82575 },
91 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_FIBER), board_82575 },
92 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SERDES), board_82575 },
93 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SGMII), board_82575 },
94 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER), board_82575 },
95 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_FIBER), board_82575 },
96 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_QUAD_FIBER), board_82575 },
97 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SERDES), board_82575 },
98 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SGMII), board_82575 },
99 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER_DUAL), board_82575 },
100 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SGMII), board_82575 },
101 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SERDES), board_82575 },
102 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_BACKPLANE), board_82575 },
103 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SFP), board_82575 },
104 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576), board_82575 },
105 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS), board_82575 },
106 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS_SERDES), board_82575 },
107 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_FIBER), board_82575 },
108 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES), board_82575 },
109 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES_QUAD), board_82575 },
110 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER_ET2), board_82575 },
111 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER), board_82575 },
112 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
113 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
114 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
115 /* required last entry */
119 MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
121 void igb_reset(struct igb_adapter *);
122 static int igb_setup_all_tx_resources(struct igb_adapter *);
123 static int igb_setup_all_rx_resources(struct igb_adapter *);
124 static void igb_free_all_tx_resources(struct igb_adapter *);
125 static void igb_free_all_rx_resources(struct igb_adapter *);
126 static void igb_setup_mrqc(struct igb_adapter *);
127 static int igb_probe(struct pci_dev *, const struct pci_device_id *);
128 static void igb_remove(struct pci_dev *pdev);
129 static int igb_sw_init(struct igb_adapter *);
130 static int igb_open(struct net_device *);
131 static int igb_close(struct net_device *);
132 static void igb_configure(struct igb_adapter *);
133 static void igb_configure_tx(struct igb_adapter *);
134 static void igb_configure_rx(struct igb_adapter *);
135 static void igb_clean_all_tx_rings(struct igb_adapter *);
136 static void igb_clean_all_rx_rings(struct igb_adapter *);
137 static void igb_clean_tx_ring(struct igb_ring *);
138 static void igb_clean_rx_ring(struct igb_ring *);
139 static void igb_set_rx_mode(struct net_device *);
140 static void igb_update_phy_info(unsigned long);
141 static void igb_watchdog(unsigned long);
142 static void igb_watchdog_task(struct work_struct *);
143 static netdev_tx_t igb_xmit_frame(struct sk_buff *skb, struct net_device *);
144 static struct rtnl_link_stats64 *igb_get_stats64(struct net_device *dev,
145 struct rtnl_link_stats64 *stats);
146 static int igb_change_mtu(struct net_device *, int);
147 static int igb_set_mac(struct net_device *, void *);
148 static void igb_set_uta(struct igb_adapter *adapter);
149 static irqreturn_t igb_intr(int irq, void *);
150 static irqreturn_t igb_intr_msi(int irq, void *);
151 static irqreturn_t igb_msix_other(int irq, void *);
152 static irqreturn_t igb_msix_ring(int irq, void *);
153 #ifdef CONFIG_IGB_DCA
154 static void igb_update_dca(struct igb_q_vector *);
155 static void igb_setup_dca(struct igb_adapter *);
156 #endif /* CONFIG_IGB_DCA */
157 static int igb_poll(struct napi_struct *, int);
158 static bool igb_clean_tx_irq(struct igb_q_vector *);
159 static bool igb_clean_rx_irq(struct igb_q_vector *, int);
160 static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
161 static void igb_tx_timeout(struct net_device *);
162 static void igb_reset_task(struct work_struct *);
163 static void igb_vlan_mode(struct net_device *netdev, netdev_features_t features);
164 static int igb_vlan_rx_add_vid(struct net_device *, __be16, u16);
165 static int igb_vlan_rx_kill_vid(struct net_device *, __be16, u16);
166 static void igb_restore_vlan(struct igb_adapter *);
167 static void igb_rar_set_qsel(struct igb_adapter *, u8 *, u32 , u8);
168 static void igb_ping_all_vfs(struct igb_adapter *);
169 static void igb_msg_task(struct igb_adapter *);
170 static void igb_vmm_control(struct igb_adapter *);
171 static int igb_set_vf_mac(struct igb_adapter *, int, unsigned char *);
172 static void igb_restore_vf_multicasts(struct igb_adapter *adapter);
173 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac);
174 static int igb_ndo_set_vf_vlan(struct net_device *netdev,
175 int vf, u16 vlan, u8 qos);
176 static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf, int tx_rate);
177 static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf,
179 static int igb_ndo_get_vf_config(struct net_device *netdev, int vf,
180 struct ifla_vf_info *ivi);
181 static void igb_check_vf_rate_limit(struct igb_adapter *);
183 #ifdef CONFIG_PCI_IOV
184 static int igb_vf_configure(struct igb_adapter *adapter, int vf);
185 static int igb_pci_enable_sriov(struct pci_dev *dev, int num_vfs);
189 #ifdef CONFIG_PM_SLEEP
190 static int igb_suspend(struct device *);
192 static int igb_resume(struct device *);
193 #ifdef CONFIG_PM_RUNTIME
194 static int igb_runtime_suspend(struct device *dev);
195 static int igb_runtime_resume(struct device *dev);
196 static int igb_runtime_idle(struct device *dev);
198 static const struct dev_pm_ops igb_pm_ops = {
199 SET_SYSTEM_SLEEP_PM_OPS(igb_suspend, igb_resume)
200 SET_RUNTIME_PM_OPS(igb_runtime_suspend, igb_runtime_resume,
204 static void igb_shutdown(struct pci_dev *);
205 static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs);
206 #ifdef CONFIG_IGB_DCA
207 static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
208 static struct notifier_block dca_notifier = {
209 .notifier_call = igb_notify_dca,
214 #ifdef CONFIG_NET_POLL_CONTROLLER
215 /* for netdump / net console */
216 static void igb_netpoll(struct net_device *);
218 #ifdef CONFIG_PCI_IOV
219 static unsigned int max_vfs = 0;
220 module_param(max_vfs, uint, 0);
221 MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate "
222 "per physical function");
223 #endif /* CONFIG_PCI_IOV */
225 static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
226 pci_channel_state_t);
227 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
228 static void igb_io_resume(struct pci_dev *);
230 static const struct pci_error_handlers igb_err_handler = {
231 .error_detected = igb_io_error_detected,
232 .slot_reset = igb_io_slot_reset,
233 .resume = igb_io_resume,
236 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba);
238 static struct pci_driver igb_driver = {
239 .name = igb_driver_name,
240 .id_table = igb_pci_tbl,
242 .remove = igb_remove,
244 .driver.pm = &igb_pm_ops,
246 .shutdown = igb_shutdown,
247 .sriov_configure = igb_pci_sriov_configure,
248 .err_handler = &igb_err_handler
251 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
252 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
253 MODULE_LICENSE("GPL");
254 MODULE_VERSION(DRV_VERSION);
256 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
257 static int debug = -1;
258 module_param(debug, int, 0);
259 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
261 struct igb_reg_info {
266 static const struct igb_reg_info igb_reg_info_tbl[] = {
268 /* General Registers */
269 {E1000_CTRL, "CTRL"},
270 {E1000_STATUS, "STATUS"},
271 {E1000_CTRL_EXT, "CTRL_EXT"},
273 /* Interrupt Registers */
277 {E1000_RCTL, "RCTL"},
278 {E1000_RDLEN(0), "RDLEN"},
279 {E1000_RDH(0), "RDH"},
280 {E1000_RDT(0), "RDT"},
281 {E1000_RXDCTL(0), "RXDCTL"},
282 {E1000_RDBAL(0), "RDBAL"},
283 {E1000_RDBAH(0), "RDBAH"},
286 {E1000_TCTL, "TCTL"},
287 {E1000_TDBAL(0), "TDBAL"},
288 {E1000_TDBAH(0), "TDBAH"},
289 {E1000_TDLEN(0), "TDLEN"},
290 {E1000_TDH(0), "TDH"},
291 {E1000_TDT(0), "TDT"},
292 {E1000_TXDCTL(0), "TXDCTL"},
293 {E1000_TDFH, "TDFH"},
294 {E1000_TDFT, "TDFT"},
295 {E1000_TDFHS, "TDFHS"},
296 {E1000_TDFPC, "TDFPC"},
298 /* List Terminator */
302 /* igb_regdump - register printout routine */
303 static void igb_regdump(struct e1000_hw *hw, struct igb_reg_info *reginfo)
309 switch (reginfo->ofs) {
311 for (n = 0; n < 4; n++)
312 regs[n] = rd32(E1000_RDLEN(n));
315 for (n = 0; n < 4; n++)
316 regs[n] = rd32(E1000_RDH(n));
319 for (n = 0; n < 4; n++)
320 regs[n] = rd32(E1000_RDT(n));
322 case E1000_RXDCTL(0):
323 for (n = 0; n < 4; n++)
324 regs[n] = rd32(E1000_RXDCTL(n));
327 for (n = 0; n < 4; n++)
328 regs[n] = rd32(E1000_RDBAL(n));
331 for (n = 0; n < 4; n++)
332 regs[n] = rd32(E1000_RDBAH(n));
335 for (n = 0; n < 4; n++)
336 regs[n] = rd32(E1000_RDBAL(n));
339 for (n = 0; n < 4; n++)
340 regs[n] = rd32(E1000_TDBAH(n));
343 for (n = 0; n < 4; n++)
344 regs[n] = rd32(E1000_TDLEN(n));
347 for (n = 0; n < 4; n++)
348 regs[n] = rd32(E1000_TDH(n));
351 for (n = 0; n < 4; n++)
352 regs[n] = rd32(E1000_TDT(n));
354 case E1000_TXDCTL(0):
355 for (n = 0; n < 4; n++)
356 regs[n] = rd32(E1000_TXDCTL(n));
359 pr_info("%-15s %08x\n", reginfo->name, rd32(reginfo->ofs));
363 snprintf(rname, 16, "%s%s", reginfo->name, "[0-3]");
364 pr_info("%-15s %08x %08x %08x %08x\n", rname, regs[0], regs[1],
368 /* igb_dump - Print registers, Tx-rings and Rx-rings */
369 static void igb_dump(struct igb_adapter *adapter)
371 struct net_device *netdev = adapter->netdev;
372 struct e1000_hw *hw = &adapter->hw;
373 struct igb_reg_info *reginfo;
374 struct igb_ring *tx_ring;
375 union e1000_adv_tx_desc *tx_desc;
376 struct my_u0 { u64 a; u64 b; } *u0;
377 struct igb_ring *rx_ring;
378 union e1000_adv_rx_desc *rx_desc;
382 if (!netif_msg_hw(adapter))
385 /* Print netdevice Info */
387 dev_info(&adapter->pdev->dev, "Net device Info\n");
388 pr_info("Device Name state trans_start "
390 pr_info("%-15s %016lX %016lX %016lX\n", netdev->name,
391 netdev->state, netdev->trans_start, netdev->last_rx);
394 /* Print Registers */
395 dev_info(&adapter->pdev->dev, "Register Dump\n");
396 pr_info(" Register Name Value\n");
397 for (reginfo = (struct igb_reg_info *)igb_reg_info_tbl;
398 reginfo->name; reginfo++) {
399 igb_regdump(hw, reginfo);
402 /* Print TX Ring Summary */
403 if (!netdev || !netif_running(netdev))
406 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
407 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
408 for (n = 0; n < adapter->num_tx_queues; n++) {
409 struct igb_tx_buffer *buffer_info;
410 tx_ring = adapter->tx_ring[n];
411 buffer_info = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
412 pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
413 n, tx_ring->next_to_use, tx_ring->next_to_clean,
414 (u64)dma_unmap_addr(buffer_info, dma),
415 dma_unmap_len(buffer_info, len),
416 buffer_info->next_to_watch,
417 (u64)buffer_info->time_stamp);
421 if (!netif_msg_tx_done(adapter))
422 goto rx_ring_summary;
424 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
426 /* Transmit Descriptor Formats
428 * Advanced Transmit Descriptor
429 * +--------------------------------------------------------------+
430 * 0 | Buffer Address [63:0] |
431 * +--------------------------------------------------------------+
432 * 8 | PAYLEN | PORTS |CC|IDX | STA | DCMD |DTYP|MAC|RSV| DTALEN |
433 * +--------------------------------------------------------------+
434 * 63 46 45 40 39 38 36 35 32 31 24 15 0
437 for (n = 0; n < adapter->num_tx_queues; n++) {
438 tx_ring = adapter->tx_ring[n];
439 pr_info("------------------------------------\n");
440 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
441 pr_info("------------------------------------\n");
442 pr_info("T [desc] [address 63:0 ] [PlPOCIStDDM Ln] "
443 "[bi->dma ] leng ntw timestamp "
446 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
447 const char *next_desc;
448 struct igb_tx_buffer *buffer_info;
449 tx_desc = IGB_TX_DESC(tx_ring, i);
450 buffer_info = &tx_ring->tx_buffer_info[i];
451 u0 = (struct my_u0 *)tx_desc;
452 if (i == tx_ring->next_to_use &&
453 i == tx_ring->next_to_clean)
454 next_desc = " NTC/U";
455 else if (i == tx_ring->next_to_use)
457 else if (i == tx_ring->next_to_clean)
462 pr_info("T [0x%03X] %016llX %016llX %016llX"
463 " %04X %p %016llX %p%s\n", i,
466 (u64)dma_unmap_addr(buffer_info, dma),
467 dma_unmap_len(buffer_info, len),
468 buffer_info->next_to_watch,
469 (u64)buffer_info->time_stamp,
470 buffer_info->skb, next_desc);
472 if (netif_msg_pktdata(adapter) && buffer_info->skb)
473 print_hex_dump(KERN_INFO, "",
475 16, 1, buffer_info->skb->data,
476 dma_unmap_len(buffer_info, len),
481 /* Print RX Rings Summary */
483 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
484 pr_info("Queue [NTU] [NTC]\n");
485 for (n = 0; n < adapter->num_rx_queues; n++) {
486 rx_ring = adapter->rx_ring[n];
487 pr_info(" %5d %5X %5X\n",
488 n, rx_ring->next_to_use, rx_ring->next_to_clean);
492 if (!netif_msg_rx_status(adapter))
495 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
497 /* Advanced Receive Descriptor (Read) Format
499 * +-----------------------------------------------------+
500 * 0 | Packet Buffer Address [63:1] |A0/NSE|
501 * +----------------------------------------------+------+
502 * 8 | Header Buffer Address [63:1] | DD |
503 * +-----------------------------------------------------+
506 * Advanced Receive Descriptor (Write-Back) Format
508 * 63 48 47 32 31 30 21 20 17 16 4 3 0
509 * +------------------------------------------------------+
510 * 0 | Packet IP |SPH| HDR_LEN | RSV|Packet| RSS |
511 * | Checksum Ident | | | | Type | Type |
512 * +------------------------------------------------------+
513 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
514 * +------------------------------------------------------+
515 * 63 48 47 32 31 20 19 0
518 for (n = 0; n < adapter->num_rx_queues; n++) {
519 rx_ring = adapter->rx_ring[n];
520 pr_info("------------------------------------\n");
521 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
522 pr_info("------------------------------------\n");
523 pr_info("R [desc] [ PktBuf A0] [ HeadBuf DD] "
524 "[bi->dma ] [bi->skb] <-- Adv Rx Read format\n");
525 pr_info("RWB[desc] [PcsmIpSHl PtRs] [vl er S cks ln] -----"
526 "----------- [bi->skb] <-- Adv Rx Write-Back format\n");
528 for (i = 0; i < rx_ring->count; i++) {
529 const char *next_desc;
530 struct igb_rx_buffer *buffer_info;
531 buffer_info = &rx_ring->rx_buffer_info[i];
532 rx_desc = IGB_RX_DESC(rx_ring, i);
533 u0 = (struct my_u0 *)rx_desc;
534 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
536 if (i == rx_ring->next_to_use)
538 else if (i == rx_ring->next_to_clean)
543 if (staterr & E1000_RXD_STAT_DD) {
544 /* Descriptor Done */
545 pr_info("%s[0x%03X] %016llX %016llX ---------------- %s\n",
551 pr_info("%s[0x%03X] %016llX %016llX %016llX %s\n",
555 (u64)buffer_info->dma,
558 if (netif_msg_pktdata(adapter) &&
559 buffer_info->dma && buffer_info->page) {
560 print_hex_dump(KERN_INFO, "",
563 page_address(buffer_info->page) +
564 buffer_info->page_offset,
576 * igb_get_i2c_data - Reads the I2C SDA data bit
577 * @hw: pointer to hardware structure
578 * @i2cctl: Current value of I2CCTL register
580 * Returns the I2C data bit value
582 static int igb_get_i2c_data(void *data)
584 struct igb_adapter *adapter = (struct igb_adapter *)data;
585 struct e1000_hw *hw = &adapter->hw;
586 s32 i2cctl = rd32(E1000_I2CPARAMS);
588 return ((i2cctl & E1000_I2C_DATA_IN) != 0);
592 * igb_set_i2c_data - Sets the I2C data bit
593 * @data: pointer to hardware structure
594 * @state: I2C data value (0 or 1) to set
596 * Sets the I2C data bit
598 static void igb_set_i2c_data(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_DATA_OUT;
607 i2cctl &= ~E1000_I2C_DATA_OUT;
609 i2cctl &= ~E1000_I2C_DATA_OE_N;
610 i2cctl |= E1000_I2C_CLK_OE_N;
611 wr32(E1000_I2CPARAMS, i2cctl);
617 * igb_set_i2c_clk - Sets the I2C SCL clock
618 * @data: pointer to hardware structure
619 * @state: state to set clock
621 * Sets the I2C clock line to state
623 static void igb_set_i2c_clk(void *data, int state)
625 struct igb_adapter *adapter = (struct igb_adapter *)data;
626 struct e1000_hw *hw = &adapter->hw;
627 s32 i2cctl = rd32(E1000_I2CPARAMS);
630 i2cctl |= E1000_I2C_CLK_OUT;
631 i2cctl &= ~E1000_I2C_CLK_OE_N;
633 i2cctl &= ~E1000_I2C_CLK_OUT;
634 i2cctl &= ~E1000_I2C_CLK_OE_N;
636 wr32(E1000_I2CPARAMS, i2cctl);
641 * igb_get_i2c_clk - Gets the I2C SCL clock state
642 * @data: pointer to hardware structure
644 * Gets the I2C clock state
646 static int igb_get_i2c_clk(void *data)
648 struct igb_adapter *adapter = (struct igb_adapter *)data;
649 struct e1000_hw *hw = &adapter->hw;
650 s32 i2cctl = rd32(E1000_I2CPARAMS);
652 return ((i2cctl & E1000_I2C_CLK_IN) != 0);
655 static const struct i2c_algo_bit_data igb_i2c_algo = {
656 .setsda = igb_set_i2c_data,
657 .setscl = igb_set_i2c_clk,
658 .getsda = igb_get_i2c_data,
659 .getscl = igb_get_i2c_clk,
665 * igb_get_hw_dev - return device
666 * @hw: pointer to hardware structure
668 * used by hardware layer to print debugging information
670 struct net_device *igb_get_hw_dev(struct e1000_hw *hw)
672 struct igb_adapter *adapter = hw->back;
673 return adapter->netdev;
677 * igb_init_module - Driver Registration Routine
679 * igb_init_module is the first routine called when the driver is
680 * loaded. All it does is register with the PCI subsystem.
682 static int __init igb_init_module(void)
685 pr_info("%s - version %s\n",
686 igb_driver_string, igb_driver_version);
688 pr_info("%s\n", igb_copyright);
690 #ifdef CONFIG_IGB_DCA
691 dca_register_notify(&dca_notifier);
693 ret = pci_register_driver(&igb_driver);
697 module_init(igb_init_module);
700 * igb_exit_module - Driver Exit Cleanup Routine
702 * igb_exit_module is called just before the driver is removed
705 static void __exit igb_exit_module(void)
707 #ifdef CONFIG_IGB_DCA
708 dca_unregister_notify(&dca_notifier);
710 pci_unregister_driver(&igb_driver);
713 module_exit(igb_exit_module);
715 #define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
717 * igb_cache_ring_register - Descriptor ring to register mapping
718 * @adapter: board private structure to initialize
720 * Once we know the feature-set enabled for the device, we'll cache
721 * the register offset the descriptor ring is assigned to.
723 static void igb_cache_ring_register(struct igb_adapter *adapter)
726 u32 rbase_offset = adapter->vfs_allocated_count;
728 switch (adapter->hw.mac.type) {
730 /* The queues are allocated for virtualization such that VF 0
731 * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
732 * In order to avoid collision we start at the first free queue
733 * and continue consuming queues in the same sequence
735 if (adapter->vfs_allocated_count) {
736 for (; i < adapter->rss_queues; i++)
737 adapter->rx_ring[i]->reg_idx = rbase_offset +
747 for (; i < adapter->num_rx_queues; i++)
748 adapter->rx_ring[i]->reg_idx = rbase_offset + i;
749 for (; j < adapter->num_tx_queues; j++)
750 adapter->tx_ring[j]->reg_idx = rbase_offset + j;
756 * igb_write_ivar - configure ivar for given MSI-X vector
757 * @hw: pointer to the HW structure
758 * @msix_vector: vector number we are allocating to a given ring
759 * @index: row index of IVAR register to write within IVAR table
760 * @offset: column offset of in IVAR, should be multiple of 8
762 * This function is intended to handle the writing of the IVAR register
763 * for adapters 82576 and newer. The IVAR table consists of 2 columns,
764 * each containing an cause allocation for an Rx and Tx ring, and a
765 * variable number of rows depending on the number of queues supported.
767 static void igb_write_ivar(struct e1000_hw *hw, int msix_vector,
768 int index, int offset)
770 u32 ivar = array_rd32(E1000_IVAR0, index);
772 /* clear any bits that are currently set */
773 ivar &= ~((u32)0xFF << offset);
775 /* write vector and valid bit */
776 ivar |= (msix_vector | E1000_IVAR_VALID) << offset;
778 array_wr32(E1000_IVAR0, index, ivar);
781 #define IGB_N0_QUEUE -1
782 static void igb_assign_vector(struct igb_q_vector *q_vector, int msix_vector)
784 struct igb_adapter *adapter = q_vector->adapter;
785 struct e1000_hw *hw = &adapter->hw;
786 int rx_queue = IGB_N0_QUEUE;
787 int tx_queue = IGB_N0_QUEUE;
790 if (q_vector->rx.ring)
791 rx_queue = q_vector->rx.ring->reg_idx;
792 if (q_vector->tx.ring)
793 tx_queue = q_vector->tx.ring->reg_idx;
795 switch (hw->mac.type) {
797 /* The 82575 assigns vectors using a bitmask, which matches the
798 * bitmask for the EICR/EIMS/EIMC registers. To assign one
799 * or more queues to a vector, we write the appropriate bits
800 * into the MSIXBM register for that vector.
802 if (rx_queue > IGB_N0_QUEUE)
803 msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
804 if (tx_queue > IGB_N0_QUEUE)
805 msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
806 if (!adapter->msix_entries && msix_vector == 0)
807 msixbm |= E1000_EIMS_OTHER;
808 array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
809 q_vector->eims_value = msixbm;
812 /* 82576 uses a table that essentially consists of 2 columns
813 * with 8 rows. The ordering is column-major so we use the
814 * lower 3 bits as the row index, and the 4th bit as the
817 if (rx_queue > IGB_N0_QUEUE)
818 igb_write_ivar(hw, msix_vector,
820 (rx_queue & 0x8) << 1);
821 if (tx_queue > IGB_N0_QUEUE)
822 igb_write_ivar(hw, msix_vector,
824 ((tx_queue & 0x8) << 1) + 8);
825 q_vector->eims_value = 1 << msix_vector;
832 /* On 82580 and newer adapters the scheme is similar to 82576
833 * however instead of ordering column-major we have things
834 * ordered row-major. So we traverse the table by using
835 * bit 0 as the column offset, and the remaining bits as the
838 if (rx_queue > IGB_N0_QUEUE)
839 igb_write_ivar(hw, msix_vector,
841 (rx_queue & 0x1) << 4);
842 if (tx_queue > IGB_N0_QUEUE)
843 igb_write_ivar(hw, msix_vector,
845 ((tx_queue & 0x1) << 4) + 8);
846 q_vector->eims_value = 1 << msix_vector;
853 /* add q_vector eims value to global eims_enable_mask */
854 adapter->eims_enable_mask |= q_vector->eims_value;
856 /* configure q_vector to set itr on first interrupt */
857 q_vector->set_itr = 1;
861 * igb_configure_msix - Configure MSI-X hardware
862 * @adapter: board private structure to initialize
864 * igb_configure_msix sets up the hardware to properly
865 * generate MSI-X interrupts.
867 static void igb_configure_msix(struct igb_adapter *adapter)
871 struct e1000_hw *hw = &adapter->hw;
873 adapter->eims_enable_mask = 0;
875 /* set vector for other causes, i.e. link changes */
876 switch (hw->mac.type) {
878 tmp = rd32(E1000_CTRL_EXT);
879 /* enable MSI-X PBA support*/
880 tmp |= E1000_CTRL_EXT_PBA_CLR;
882 /* Auto-Mask interrupts upon ICR read. */
883 tmp |= E1000_CTRL_EXT_EIAME;
884 tmp |= E1000_CTRL_EXT_IRCA;
886 wr32(E1000_CTRL_EXT, tmp);
888 /* enable msix_other interrupt */
889 array_wr32(E1000_MSIXBM(0), vector++, E1000_EIMS_OTHER);
890 adapter->eims_other = E1000_EIMS_OTHER;
900 /* Turn on MSI-X capability first, or our settings
901 * won't stick. And it will take days to debug.
903 wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
904 E1000_GPIE_PBA | E1000_GPIE_EIAME |
907 /* enable msix_other interrupt */
908 adapter->eims_other = 1 << vector;
909 tmp = (vector++ | E1000_IVAR_VALID) << 8;
911 wr32(E1000_IVAR_MISC, tmp);
914 /* do nothing, since nothing else supports MSI-X */
916 } /* switch (hw->mac.type) */
918 adapter->eims_enable_mask |= adapter->eims_other;
920 for (i = 0; i < adapter->num_q_vectors; i++)
921 igb_assign_vector(adapter->q_vector[i], vector++);
927 * igb_request_msix - Initialize MSI-X interrupts
928 * @adapter: board private structure to initialize
930 * igb_request_msix allocates MSI-X vectors and requests interrupts from the
933 static int igb_request_msix(struct igb_adapter *adapter)
935 struct net_device *netdev = adapter->netdev;
936 struct e1000_hw *hw = &adapter->hw;
937 int i, err = 0, vector = 0, free_vector = 0;
939 err = request_irq(adapter->msix_entries[vector].vector,
940 igb_msix_other, 0, netdev->name, adapter);
944 for (i = 0; i < adapter->num_q_vectors; i++) {
945 struct igb_q_vector *q_vector = adapter->q_vector[i];
949 q_vector->itr_register = hw->hw_addr + E1000_EITR(vector);
951 if (q_vector->rx.ring && q_vector->tx.ring)
952 sprintf(q_vector->name, "%s-TxRx-%u", netdev->name,
953 q_vector->rx.ring->queue_index);
954 else if (q_vector->tx.ring)
955 sprintf(q_vector->name, "%s-tx-%u", netdev->name,
956 q_vector->tx.ring->queue_index);
957 else if (q_vector->rx.ring)
958 sprintf(q_vector->name, "%s-rx-%u", netdev->name,
959 q_vector->rx.ring->queue_index);
961 sprintf(q_vector->name, "%s-unused", netdev->name);
963 err = request_irq(adapter->msix_entries[vector].vector,
964 igb_msix_ring, 0, q_vector->name,
970 igb_configure_msix(adapter);
974 /* free already assigned IRQs */
975 free_irq(adapter->msix_entries[free_vector++].vector, adapter);
978 for (i = 0; i < vector; i++) {
979 free_irq(adapter->msix_entries[free_vector++].vector,
980 adapter->q_vector[i]);
986 static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
988 if (adapter->msix_entries) {
989 pci_disable_msix(adapter->pdev);
990 kfree(adapter->msix_entries);
991 adapter->msix_entries = NULL;
992 } else if (adapter->flags & IGB_FLAG_HAS_MSI) {
993 pci_disable_msi(adapter->pdev);
998 * igb_free_q_vector - Free memory allocated for specific interrupt vector
999 * @adapter: board private structure to initialize
1000 * @v_idx: Index of vector to be freed
1002 * This function frees the memory allocated to the q_vector. In addition if
1003 * NAPI is enabled it will delete any references to the NAPI struct prior
1004 * to freeing the q_vector.
1006 static void igb_free_q_vector(struct igb_adapter *adapter, int v_idx)
1008 struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
1010 if (q_vector->tx.ring)
1011 adapter->tx_ring[q_vector->tx.ring->queue_index] = NULL;
1013 if (q_vector->rx.ring)
1014 adapter->tx_ring[q_vector->rx.ring->queue_index] = NULL;
1016 adapter->q_vector[v_idx] = NULL;
1017 netif_napi_del(&q_vector->napi);
1019 /* igb_get_stats64() might access the rings on this vector,
1020 * we must wait a grace period before freeing it.
1022 kfree_rcu(q_vector, rcu);
1026 * igb_free_q_vectors - Free memory allocated for interrupt vectors
1027 * @adapter: board private structure to initialize
1029 * This function frees the memory allocated to the q_vectors. In addition if
1030 * NAPI is enabled it will delete any references to the NAPI struct prior
1031 * to freeing the q_vector.
1033 static void igb_free_q_vectors(struct igb_adapter *adapter)
1035 int v_idx = adapter->num_q_vectors;
1037 adapter->num_tx_queues = 0;
1038 adapter->num_rx_queues = 0;
1039 adapter->num_q_vectors = 0;
1042 igb_free_q_vector(adapter, v_idx);
1046 * igb_clear_interrupt_scheme - reset the device to a state of no interrupts
1047 * @adapter: board private structure to initialize
1049 * This function resets the device so that it has 0 Rx queues, Tx queues, and
1050 * MSI-X interrupts allocated.
1052 static void igb_clear_interrupt_scheme(struct igb_adapter *adapter)
1054 igb_free_q_vectors(adapter);
1055 igb_reset_interrupt_capability(adapter);
1059 * igb_set_interrupt_capability - set MSI or MSI-X if supported
1060 * @adapter: board private structure to initialize
1061 * @msix: boolean value of MSIX capability
1063 * Attempt to configure interrupts using the best available
1064 * capabilities of the hardware and kernel.
1066 static void igb_set_interrupt_capability(struct igb_adapter *adapter, bool msix)
1074 /* Number of supported queues. */
1075 adapter->num_rx_queues = adapter->rss_queues;
1076 if (adapter->vfs_allocated_count)
1077 adapter->num_tx_queues = 1;
1079 adapter->num_tx_queues = adapter->rss_queues;
1081 /* start with one vector for every Rx queue */
1082 numvecs = adapter->num_rx_queues;
1084 /* if Tx handler is separate add 1 for every Tx queue */
1085 if (!(adapter->flags & IGB_FLAG_QUEUE_PAIRS))
1086 numvecs += adapter->num_tx_queues;
1088 /* store the number of vectors reserved for queues */
1089 adapter->num_q_vectors = numvecs;
1091 /* add 1 vector for link status interrupts */
1093 adapter->msix_entries = kcalloc(numvecs, sizeof(struct msix_entry),
1096 if (!adapter->msix_entries)
1099 for (i = 0; i < numvecs; i++)
1100 adapter->msix_entries[i].entry = i;
1102 err = pci_enable_msix(adapter->pdev,
1103 adapter->msix_entries,
1108 igb_reset_interrupt_capability(adapter);
1110 /* If we can't do MSI-X, try MSI */
1112 #ifdef CONFIG_PCI_IOV
1113 /* disable SR-IOV for non MSI-X configurations */
1114 if (adapter->vf_data) {
1115 struct e1000_hw *hw = &adapter->hw;
1116 /* disable iov and allow time for transactions to clear */
1117 pci_disable_sriov(adapter->pdev);
1120 kfree(adapter->vf_data);
1121 adapter->vf_data = NULL;
1122 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
1125 dev_info(&adapter->pdev->dev, "IOV Disabled\n");
1128 adapter->vfs_allocated_count = 0;
1129 adapter->rss_queues = 1;
1130 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
1131 adapter->num_rx_queues = 1;
1132 adapter->num_tx_queues = 1;
1133 adapter->num_q_vectors = 1;
1134 if (!pci_enable_msi(adapter->pdev))
1135 adapter->flags |= IGB_FLAG_HAS_MSI;
1138 static void igb_add_ring(struct igb_ring *ring,
1139 struct igb_ring_container *head)
1146 * igb_alloc_q_vector - Allocate memory for a single interrupt vector
1147 * @adapter: board private structure to initialize
1148 * @v_count: q_vectors allocated on adapter, used for ring interleaving
1149 * @v_idx: index of vector in adapter struct
1150 * @txr_count: total number of Tx rings to allocate
1151 * @txr_idx: index of first Tx ring to allocate
1152 * @rxr_count: total number of Rx rings to allocate
1153 * @rxr_idx: index of first Rx ring to allocate
1155 * We allocate one q_vector. If allocation fails we return -ENOMEM.
1157 static int igb_alloc_q_vector(struct igb_adapter *adapter,
1158 int v_count, int v_idx,
1159 int txr_count, int txr_idx,
1160 int rxr_count, int rxr_idx)
1162 struct igb_q_vector *q_vector;
1163 struct igb_ring *ring;
1164 int ring_count, size;
1166 /* igb only supports 1 Tx and/or 1 Rx queue per vector */
1167 if (txr_count > 1 || rxr_count > 1)
1170 ring_count = txr_count + rxr_count;
1171 size = sizeof(struct igb_q_vector) +
1172 (sizeof(struct igb_ring) * ring_count);
1174 /* allocate q_vector and rings */
1175 q_vector = kzalloc(size, GFP_KERNEL);
1179 /* initialize NAPI */
1180 netif_napi_add(adapter->netdev, &q_vector->napi,
1183 /* tie q_vector and adapter together */
1184 adapter->q_vector[v_idx] = q_vector;
1185 q_vector->adapter = adapter;
1187 /* initialize work limits */
1188 q_vector->tx.work_limit = adapter->tx_work_limit;
1190 /* initialize ITR configuration */
1191 q_vector->itr_register = adapter->hw.hw_addr + E1000_EITR(0);
1192 q_vector->itr_val = IGB_START_ITR;
1194 /* initialize pointer to rings */
1195 ring = q_vector->ring;
1199 /* rx or rx/tx vector */
1200 if (!adapter->rx_itr_setting || adapter->rx_itr_setting > 3)
1201 q_vector->itr_val = adapter->rx_itr_setting;
1203 /* tx only vector */
1204 if (!adapter->tx_itr_setting || adapter->tx_itr_setting > 3)
1205 q_vector->itr_val = adapter->tx_itr_setting;
1209 /* assign generic ring traits */
1210 ring->dev = &adapter->pdev->dev;
1211 ring->netdev = adapter->netdev;
1213 /* configure backlink on ring */
1214 ring->q_vector = q_vector;
1216 /* update q_vector Tx values */
1217 igb_add_ring(ring, &q_vector->tx);
1219 /* For 82575, context index must be unique per ring. */
1220 if (adapter->hw.mac.type == e1000_82575)
1221 set_bit(IGB_RING_FLAG_TX_CTX_IDX, &ring->flags);
1223 /* apply Tx specific ring traits */
1224 ring->count = adapter->tx_ring_count;
1225 ring->queue_index = txr_idx;
1227 u64_stats_init(&ring->tx_syncp);
1228 u64_stats_init(&ring->tx_syncp2);
1230 /* assign ring to adapter */
1231 adapter->tx_ring[txr_idx] = ring;
1233 /* push pointer to next ring */
1238 /* assign generic ring traits */
1239 ring->dev = &adapter->pdev->dev;
1240 ring->netdev = adapter->netdev;
1242 /* configure backlink on ring */
1243 ring->q_vector = q_vector;
1245 /* update q_vector Rx values */
1246 igb_add_ring(ring, &q_vector->rx);
1248 /* set flag indicating ring supports SCTP checksum offload */
1249 if (adapter->hw.mac.type >= e1000_82576)
1250 set_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags);
1253 * On i350, i354, i210, and i211, loopback VLAN packets
1254 * have the tag byte-swapped.
1256 if (adapter->hw.mac.type >= e1000_i350)
1257 set_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &ring->flags);
1259 /* apply Rx specific ring traits */
1260 ring->count = adapter->rx_ring_count;
1261 ring->queue_index = rxr_idx;
1263 u64_stats_init(&ring->rx_syncp);
1265 /* assign ring to adapter */
1266 adapter->rx_ring[rxr_idx] = ring;
1274 * igb_alloc_q_vectors - Allocate memory for interrupt vectors
1275 * @adapter: board private structure to initialize
1277 * We allocate one q_vector per queue interrupt. If allocation fails we
1280 static int igb_alloc_q_vectors(struct igb_adapter *adapter)
1282 int q_vectors = adapter->num_q_vectors;
1283 int rxr_remaining = adapter->num_rx_queues;
1284 int txr_remaining = adapter->num_tx_queues;
1285 int rxr_idx = 0, txr_idx = 0, v_idx = 0;
1288 if (q_vectors >= (rxr_remaining + txr_remaining)) {
1289 for (; rxr_remaining; v_idx++) {
1290 err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1296 /* update counts and index */
1302 for (; v_idx < q_vectors; v_idx++) {
1303 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_idx);
1304 int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_idx);
1305 err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1306 tqpv, txr_idx, rqpv, rxr_idx);
1311 /* update counts and index */
1312 rxr_remaining -= rqpv;
1313 txr_remaining -= tqpv;
1321 adapter->num_tx_queues = 0;
1322 adapter->num_rx_queues = 0;
1323 adapter->num_q_vectors = 0;
1326 igb_free_q_vector(adapter, v_idx);
1332 * igb_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
1333 * @adapter: board private structure to initialize
1334 * @msix: boolean value of MSIX capability
1336 * This function initializes the interrupts and allocates all of the queues.
1338 static int igb_init_interrupt_scheme(struct igb_adapter *adapter, bool msix)
1340 struct pci_dev *pdev = adapter->pdev;
1343 igb_set_interrupt_capability(adapter, msix);
1345 err = igb_alloc_q_vectors(adapter);
1347 dev_err(&pdev->dev, "Unable to allocate memory for vectors\n");
1348 goto err_alloc_q_vectors;
1351 igb_cache_ring_register(adapter);
1355 err_alloc_q_vectors:
1356 igb_reset_interrupt_capability(adapter);
1361 * igb_request_irq - initialize interrupts
1362 * @adapter: board private structure to initialize
1364 * Attempts to configure interrupts using the best available
1365 * capabilities of the hardware and kernel.
1367 static int igb_request_irq(struct igb_adapter *adapter)
1369 struct net_device *netdev = adapter->netdev;
1370 struct pci_dev *pdev = adapter->pdev;
1373 if (adapter->msix_entries) {
1374 err = igb_request_msix(adapter);
1377 /* fall back to MSI */
1378 igb_free_all_tx_resources(adapter);
1379 igb_free_all_rx_resources(adapter);
1381 igb_clear_interrupt_scheme(adapter);
1382 err = igb_init_interrupt_scheme(adapter, false);
1386 igb_setup_all_tx_resources(adapter);
1387 igb_setup_all_rx_resources(adapter);
1388 igb_configure(adapter);
1391 igb_assign_vector(adapter->q_vector[0], 0);
1393 if (adapter->flags & IGB_FLAG_HAS_MSI) {
1394 err = request_irq(pdev->irq, igb_intr_msi, 0,
1395 netdev->name, adapter);
1399 /* fall back to legacy interrupts */
1400 igb_reset_interrupt_capability(adapter);
1401 adapter->flags &= ~IGB_FLAG_HAS_MSI;
1404 err = request_irq(pdev->irq, igb_intr, IRQF_SHARED,
1405 netdev->name, adapter);
1408 dev_err(&pdev->dev, "Error %d getting interrupt\n",
1415 static void igb_free_irq(struct igb_adapter *adapter)
1417 if (adapter->msix_entries) {
1420 free_irq(adapter->msix_entries[vector++].vector, adapter);
1422 for (i = 0; i < adapter->num_q_vectors; i++)
1423 free_irq(adapter->msix_entries[vector++].vector,
1424 adapter->q_vector[i]);
1426 free_irq(adapter->pdev->irq, adapter);
1431 * igb_irq_disable - Mask off interrupt generation on the NIC
1432 * @adapter: board private structure
1434 static void igb_irq_disable(struct igb_adapter *adapter)
1436 struct e1000_hw *hw = &adapter->hw;
1438 /* we need to be careful when disabling interrupts. The VFs are also
1439 * mapped into these registers and so clearing the bits can cause
1440 * issues on the VF drivers so we only need to clear what we set
1442 if (adapter->msix_entries) {
1443 u32 regval = rd32(E1000_EIAM);
1444 wr32(E1000_EIAM, regval & ~adapter->eims_enable_mask);
1445 wr32(E1000_EIMC, adapter->eims_enable_mask);
1446 regval = rd32(E1000_EIAC);
1447 wr32(E1000_EIAC, regval & ~adapter->eims_enable_mask);
1451 wr32(E1000_IMC, ~0);
1453 if (adapter->msix_entries) {
1455 for (i = 0; i < adapter->num_q_vectors; i++)
1456 synchronize_irq(adapter->msix_entries[i].vector);
1458 synchronize_irq(adapter->pdev->irq);
1463 * igb_irq_enable - Enable default interrupt generation settings
1464 * @adapter: board private structure
1466 static void igb_irq_enable(struct igb_adapter *adapter)
1468 struct e1000_hw *hw = &adapter->hw;
1470 if (adapter->msix_entries) {
1471 u32 ims = E1000_IMS_LSC | E1000_IMS_DOUTSYNC | E1000_IMS_DRSTA;
1472 u32 regval = rd32(E1000_EIAC);
1473 wr32(E1000_EIAC, regval | adapter->eims_enable_mask);
1474 regval = rd32(E1000_EIAM);
1475 wr32(E1000_EIAM, regval | adapter->eims_enable_mask);
1476 wr32(E1000_EIMS, adapter->eims_enable_mask);
1477 if (adapter->vfs_allocated_count) {
1478 wr32(E1000_MBVFIMR, 0xFF);
1479 ims |= E1000_IMS_VMMB;
1481 wr32(E1000_IMS, ims);
1483 wr32(E1000_IMS, IMS_ENABLE_MASK |
1485 wr32(E1000_IAM, IMS_ENABLE_MASK |
1490 static void igb_update_mng_vlan(struct igb_adapter *adapter)
1492 struct e1000_hw *hw = &adapter->hw;
1493 u16 vid = adapter->hw.mng_cookie.vlan_id;
1494 u16 old_vid = adapter->mng_vlan_id;
1496 if (hw->mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1497 /* add VID to filter table */
1498 igb_vfta_set(hw, vid, true);
1499 adapter->mng_vlan_id = vid;
1501 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1504 if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
1506 !test_bit(old_vid, adapter->active_vlans)) {
1507 /* remove VID from filter table */
1508 igb_vfta_set(hw, old_vid, false);
1513 * igb_release_hw_control - release control of the h/w to f/w
1514 * @adapter: address of board private structure
1516 * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
1517 * For ASF and Pass Through versions of f/w this means that the
1518 * driver is no longer loaded.
1520 static void igb_release_hw_control(struct igb_adapter *adapter)
1522 struct e1000_hw *hw = &adapter->hw;
1525 /* Let firmware take over control of h/w */
1526 ctrl_ext = rd32(E1000_CTRL_EXT);
1527 wr32(E1000_CTRL_EXT,
1528 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1532 * igb_get_hw_control - get control of the h/w from f/w
1533 * @adapter: address of board private structure
1535 * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
1536 * For ASF and Pass Through versions of f/w this means that
1537 * the driver is loaded.
1539 static void igb_get_hw_control(struct igb_adapter *adapter)
1541 struct e1000_hw *hw = &adapter->hw;
1544 /* Let firmware know the driver has taken over */
1545 ctrl_ext = rd32(E1000_CTRL_EXT);
1546 wr32(E1000_CTRL_EXT,
1547 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1551 * igb_configure - configure the hardware for RX and TX
1552 * @adapter: private board structure
1554 static void igb_configure(struct igb_adapter *adapter)
1556 struct net_device *netdev = adapter->netdev;
1559 igb_get_hw_control(adapter);
1560 igb_set_rx_mode(netdev);
1562 igb_restore_vlan(adapter);
1564 igb_setup_tctl(adapter);
1565 igb_setup_mrqc(adapter);
1566 igb_setup_rctl(adapter);
1568 igb_configure_tx(adapter);
1569 igb_configure_rx(adapter);
1571 igb_rx_fifo_flush_82575(&adapter->hw);
1573 /* call igb_desc_unused which always leaves
1574 * at least 1 descriptor unused to make sure
1575 * next_to_use != next_to_clean
1577 for (i = 0; i < adapter->num_rx_queues; i++) {
1578 struct igb_ring *ring = adapter->rx_ring[i];
1579 igb_alloc_rx_buffers(ring, igb_desc_unused(ring));
1584 * igb_power_up_link - Power up the phy/serdes link
1585 * @adapter: address of board private structure
1587 void igb_power_up_link(struct igb_adapter *adapter)
1589 igb_reset_phy(&adapter->hw);
1591 if (adapter->hw.phy.media_type == e1000_media_type_copper)
1592 igb_power_up_phy_copper(&adapter->hw);
1594 igb_power_up_serdes_link_82575(&adapter->hw);
1598 * igb_power_down_link - Power down the phy/serdes link
1599 * @adapter: address of board private structure
1601 static void igb_power_down_link(struct igb_adapter *adapter)
1603 if (adapter->hw.phy.media_type == e1000_media_type_copper)
1604 igb_power_down_phy_copper_82575(&adapter->hw);
1606 igb_shutdown_serdes_link_82575(&adapter->hw);
1610 * Detect and switch function for Media Auto Sense
1611 * @adapter: address of the board private structure
1613 static void igb_check_swap_media(struct igb_adapter *adapter)
1615 struct e1000_hw *hw = &adapter->hw;
1616 u32 ctrl_ext, connsw;
1617 bool swap_now = false;
1619 ctrl_ext = rd32(E1000_CTRL_EXT);
1620 connsw = rd32(E1000_CONNSW);
1622 /* need to live swap if current media is copper and we have fiber/serdes
1626 if ((hw->phy.media_type == e1000_media_type_copper) &&
1627 (!(connsw & E1000_CONNSW_AUTOSENSE_EN))) {
1629 } else if (!(connsw & E1000_CONNSW_SERDESD)) {
1630 /* copper signal takes time to appear */
1631 if (adapter->copper_tries < 4) {
1632 adapter->copper_tries++;
1633 connsw |= E1000_CONNSW_AUTOSENSE_CONF;
1634 wr32(E1000_CONNSW, connsw);
1637 adapter->copper_tries = 0;
1638 if ((connsw & E1000_CONNSW_PHYSD) &&
1639 (!(connsw & E1000_CONNSW_PHY_PDN))) {
1641 connsw &= ~E1000_CONNSW_AUTOSENSE_CONF;
1642 wr32(E1000_CONNSW, connsw);
1650 switch (hw->phy.media_type) {
1651 case e1000_media_type_copper:
1652 netdev_info(adapter->netdev,
1653 "MAS: changing media to fiber/serdes\n");
1655 E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
1656 adapter->flags |= IGB_FLAG_MEDIA_RESET;
1657 adapter->copper_tries = 0;
1659 case e1000_media_type_internal_serdes:
1660 case e1000_media_type_fiber:
1661 netdev_info(adapter->netdev,
1662 "MAS: changing media to copper\n");
1664 ~E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
1665 adapter->flags |= IGB_FLAG_MEDIA_RESET;
1668 /* shouldn't get here during regular operation */
1669 netdev_err(adapter->netdev,
1670 "AMS: Invalid media type found, returning\n");
1673 wr32(E1000_CTRL_EXT, ctrl_ext);
1677 * igb_up - Open the interface and prepare it to handle traffic
1678 * @adapter: board private structure
1680 int igb_up(struct igb_adapter *adapter)
1682 struct e1000_hw *hw = &adapter->hw;
1685 /* hardware has been reset, we need to reload some things */
1686 igb_configure(adapter);
1688 clear_bit(__IGB_DOWN, &adapter->state);
1690 for (i = 0; i < adapter->num_q_vectors; i++)
1691 napi_enable(&(adapter->q_vector[i]->napi));
1693 if (adapter->msix_entries)
1694 igb_configure_msix(adapter);
1696 igb_assign_vector(adapter->q_vector[0], 0);
1698 /* Clear any pending interrupts. */
1700 igb_irq_enable(adapter);
1702 /* notify VFs that reset has been completed */
1703 if (adapter->vfs_allocated_count) {
1704 u32 reg_data = rd32(E1000_CTRL_EXT);
1705 reg_data |= E1000_CTRL_EXT_PFRSTD;
1706 wr32(E1000_CTRL_EXT, reg_data);
1709 netif_tx_start_all_queues(adapter->netdev);
1711 /* start the watchdog. */
1712 hw->mac.get_link_status = 1;
1713 schedule_work(&adapter->watchdog_task);
1718 void igb_down(struct igb_adapter *adapter)
1720 struct net_device *netdev = adapter->netdev;
1721 struct e1000_hw *hw = &adapter->hw;
1725 /* signal that we're down so the interrupt handler does not
1726 * reschedule our watchdog timer
1728 set_bit(__IGB_DOWN, &adapter->state);
1730 /* disable receives in the hardware */
1731 rctl = rd32(E1000_RCTL);
1732 wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
1733 /* flush and sleep below */
1735 netif_tx_stop_all_queues(netdev);
1737 /* disable transmits in the hardware */
1738 tctl = rd32(E1000_TCTL);
1739 tctl &= ~E1000_TCTL_EN;
1740 wr32(E1000_TCTL, tctl);
1741 /* flush both disables and wait for them to finish */
1745 igb_irq_disable(adapter);
1747 adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
1749 for (i = 0; i < adapter->num_q_vectors; i++) {
1750 napi_synchronize(&(adapter->q_vector[i]->napi));
1751 napi_disable(&(adapter->q_vector[i]->napi));
1755 del_timer_sync(&adapter->watchdog_timer);
1756 del_timer_sync(&adapter->phy_info_timer);
1758 netif_carrier_off(netdev);
1760 /* record the stats before reset*/
1761 spin_lock(&adapter->stats64_lock);
1762 igb_update_stats(adapter, &adapter->stats64);
1763 spin_unlock(&adapter->stats64_lock);
1765 adapter->link_speed = 0;
1766 adapter->link_duplex = 0;
1768 if (!pci_channel_offline(adapter->pdev))
1770 igb_clean_all_tx_rings(adapter);
1771 igb_clean_all_rx_rings(adapter);
1772 #ifdef CONFIG_IGB_DCA
1774 /* since we reset the hardware DCA settings were cleared */
1775 igb_setup_dca(adapter);
1779 void igb_reinit_locked(struct igb_adapter *adapter)
1781 WARN_ON(in_interrupt());
1782 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
1786 clear_bit(__IGB_RESETTING, &adapter->state);
1789 /** igb_enable_mas - Media Autosense re-enable after swap
1791 * @adapter: adapter struct
1793 static s32 igb_enable_mas(struct igb_adapter *adapter)
1795 struct e1000_hw *hw = &adapter->hw;
1799 connsw = rd32(E1000_CONNSW);
1800 if (!(hw->phy.media_type == e1000_media_type_copper))
1803 /* configure for SerDes media detect */
1804 if (!(connsw & E1000_CONNSW_SERDESD)) {
1805 connsw |= E1000_CONNSW_ENRGSRC;
1806 connsw |= E1000_CONNSW_AUTOSENSE_EN;
1807 wr32(E1000_CONNSW, connsw);
1809 } else if (connsw & E1000_CONNSW_SERDESD) {
1810 /* already SerDes, no need to enable anything */
1813 netdev_info(adapter->netdev,
1814 "MAS: Unable to configure feature, disabling..\n");
1815 adapter->flags &= ~IGB_FLAG_MAS_ENABLE;
1820 void igb_reset(struct igb_adapter *adapter)
1822 struct pci_dev *pdev = adapter->pdev;
1823 struct e1000_hw *hw = &adapter->hw;
1824 struct e1000_mac_info *mac = &hw->mac;
1825 struct e1000_fc_info *fc = &hw->fc;
1826 u32 pba = 0, tx_space, min_tx_space, min_rx_space, hwm;
1828 /* Repartition Pba for greater than 9k mtu
1829 * To take effect CTRL.RST is required.
1831 switch (mac->type) {
1835 pba = rd32(E1000_RXPBS);
1836 pba = igb_rxpbs_adjust_82580(pba);
1839 pba = rd32(E1000_RXPBS);
1840 pba &= E1000_RXPBS_SIZE_MASK_82576;
1846 pba = E1000_PBA_34K;
1850 if ((adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) &&
1851 (mac->type < e1000_82576)) {
1852 /* adjust PBA for jumbo frames */
1853 wr32(E1000_PBA, pba);
1855 /* To maintain wire speed transmits, the Tx FIFO should be
1856 * large enough to accommodate two full transmit packets,
1857 * rounded up to the next 1KB and expressed in KB. Likewise,
1858 * the Rx FIFO should be large enough to accommodate at least
1859 * one full receive packet and is similarly rounded up and
1862 pba = rd32(E1000_PBA);
1863 /* upper 16 bits has Tx packet buffer allocation size in KB */
1864 tx_space = pba >> 16;
1865 /* lower 16 bits has Rx packet buffer allocation size in KB */
1867 /* the Tx fifo also stores 16 bytes of information about the Tx
1868 * but don't include ethernet FCS because hardware appends it
1870 min_tx_space = (adapter->max_frame_size +
1871 sizeof(union e1000_adv_tx_desc) -
1873 min_tx_space = ALIGN(min_tx_space, 1024);
1874 min_tx_space >>= 10;
1875 /* software strips receive CRC, so leave room for it */
1876 min_rx_space = adapter->max_frame_size;
1877 min_rx_space = ALIGN(min_rx_space, 1024);
1878 min_rx_space >>= 10;
1880 /* If current Tx allocation is less than the min Tx FIFO size,
1881 * and the min Tx FIFO size is less than the current Rx FIFO
1882 * allocation, take space away from current Rx allocation
1884 if (tx_space < min_tx_space &&
1885 ((min_tx_space - tx_space) < pba)) {
1886 pba = pba - (min_tx_space - tx_space);
1888 /* if short on Rx space, Rx wins and must trump Tx
1891 if (pba < min_rx_space)
1894 wr32(E1000_PBA, pba);
1897 /* flow control settings */
1898 /* The high water mark must be low enough to fit one full frame
1899 * (or the size used for early receive) above it in the Rx FIFO.
1900 * Set it to the lower of:
1901 * - 90% of the Rx FIFO size, or
1902 * - the full Rx FIFO size minus one full frame
1904 hwm = min(((pba << 10) * 9 / 10),
1905 ((pba << 10) - 2 * adapter->max_frame_size));
1907 fc->high_water = hwm & 0xFFFFFFF0; /* 16-byte granularity */
1908 fc->low_water = fc->high_water - 16;
1909 fc->pause_time = 0xFFFF;
1911 fc->current_mode = fc->requested_mode;
1913 /* disable receive for all VFs and wait one second */
1914 if (adapter->vfs_allocated_count) {
1916 for (i = 0 ; i < adapter->vfs_allocated_count; i++)
1917 adapter->vf_data[i].flags &= IGB_VF_FLAG_PF_SET_MAC;
1919 /* ping all the active vfs to let them know we are going down */
1920 igb_ping_all_vfs(adapter);
1922 /* disable transmits and receives */
1923 wr32(E1000_VFRE, 0);
1924 wr32(E1000_VFTE, 0);
1927 /* Allow time for pending master requests to run */
1928 hw->mac.ops.reset_hw(hw);
1931 if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
1932 /* need to resetup here after media swap */
1933 adapter->ei.get_invariants(hw);
1934 adapter->flags &= ~IGB_FLAG_MEDIA_RESET;
1936 if (adapter->flags & IGB_FLAG_MAS_ENABLE) {
1937 if (igb_enable_mas(adapter))
1939 "Error enabling Media Auto Sense\n");
1941 if (hw->mac.ops.init_hw(hw))
1942 dev_err(&pdev->dev, "Hardware Error\n");
1944 /* Flow control settings reset on hardware reset, so guarantee flow
1945 * control is off when forcing speed.
1947 if (!hw->mac.autoneg)
1948 igb_force_mac_fc(hw);
1950 igb_init_dmac(adapter, pba);
1951 #ifdef CONFIG_IGB_HWMON
1952 /* Re-initialize the thermal sensor on i350 devices. */
1953 if (!test_bit(__IGB_DOWN, &adapter->state)) {
1954 if (mac->type == e1000_i350 && hw->bus.func == 0) {
1955 /* If present, re-initialize the external thermal sensor
1959 mac->ops.init_thermal_sensor_thresh(hw);
1963 if (!netif_running(adapter->netdev))
1964 igb_power_down_link(adapter);
1966 igb_update_mng_vlan(adapter);
1968 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
1969 wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
1971 /* Re-enable PTP, where applicable. */
1972 igb_ptp_reset(adapter);
1974 igb_get_phy_info(hw);
1977 static netdev_features_t igb_fix_features(struct net_device *netdev,
1978 netdev_features_t features)
1980 /* Since there is no support for separate Rx/Tx vlan accel
1981 * enable/disable make sure Tx flag is always in same state as Rx.
1983 if (features & NETIF_F_HW_VLAN_CTAG_RX)
1984 features |= NETIF_F_HW_VLAN_CTAG_TX;
1986 features &= ~NETIF_F_HW_VLAN_CTAG_TX;
1991 static int igb_set_features(struct net_device *netdev,
1992 netdev_features_t features)
1994 netdev_features_t changed = netdev->features ^ features;
1995 struct igb_adapter *adapter = netdev_priv(netdev);
1997 if (changed & NETIF_F_HW_VLAN_CTAG_RX)
1998 igb_vlan_mode(netdev, features);
2000 if (!(changed & NETIF_F_RXALL))
2003 netdev->features = features;
2005 if (netif_running(netdev))
2006 igb_reinit_locked(adapter);
2013 static const struct net_device_ops igb_netdev_ops = {
2014 .ndo_open = igb_open,
2015 .ndo_stop = igb_close,
2016 .ndo_start_xmit = igb_xmit_frame,
2017 .ndo_get_stats64 = igb_get_stats64,
2018 .ndo_set_rx_mode = igb_set_rx_mode,
2019 .ndo_set_mac_address = igb_set_mac,
2020 .ndo_change_mtu = igb_change_mtu,
2021 .ndo_do_ioctl = igb_ioctl,
2022 .ndo_tx_timeout = igb_tx_timeout,
2023 .ndo_validate_addr = eth_validate_addr,
2024 .ndo_vlan_rx_add_vid = igb_vlan_rx_add_vid,
2025 .ndo_vlan_rx_kill_vid = igb_vlan_rx_kill_vid,
2026 .ndo_set_vf_mac = igb_ndo_set_vf_mac,
2027 .ndo_set_vf_vlan = igb_ndo_set_vf_vlan,
2028 .ndo_set_vf_tx_rate = igb_ndo_set_vf_bw,
2029 .ndo_set_vf_spoofchk = igb_ndo_set_vf_spoofchk,
2030 .ndo_get_vf_config = igb_ndo_get_vf_config,
2031 #ifdef CONFIG_NET_POLL_CONTROLLER
2032 .ndo_poll_controller = igb_netpoll,
2034 .ndo_fix_features = igb_fix_features,
2035 .ndo_set_features = igb_set_features,
2039 * igb_set_fw_version - Configure version string for ethtool
2040 * @adapter: adapter struct
2042 void igb_set_fw_version(struct igb_adapter *adapter)
2044 struct e1000_hw *hw = &adapter->hw;
2045 struct e1000_fw_version fw;
2047 igb_get_fw_version(hw, &fw);
2049 switch (hw->mac.type) {
2052 if (!(igb_get_flash_presence_i210(hw))) {
2053 snprintf(adapter->fw_version,
2054 sizeof(adapter->fw_version),
2056 fw.invm_major, fw.invm_minor,
2062 /* if option is rom valid, display its version too */
2064 snprintf(adapter->fw_version,
2065 sizeof(adapter->fw_version),
2066 "%d.%d, 0x%08x, %d.%d.%d",
2067 fw.eep_major, fw.eep_minor, fw.etrack_id,
2068 fw.or_major, fw.or_build, fw.or_patch);
2070 } else if (fw.etrack_id != 0X0000) {
2071 snprintf(adapter->fw_version,
2072 sizeof(adapter->fw_version),
2074 fw.eep_major, fw.eep_minor, fw.etrack_id);
2076 snprintf(adapter->fw_version,
2077 sizeof(adapter->fw_version),
2079 fw.eep_major, fw.eep_minor, fw.eep_build);
2087 * igb_init_mas - init Media Autosense feature if enabled in the NVM
2089 * @adapter: adapter struct
2091 static void igb_init_mas(struct igb_adapter *adapter)
2093 struct e1000_hw *hw = &adapter->hw;
2096 hw->nvm.ops.read(hw, NVM_COMPAT, 1, &eeprom_data);
2097 switch (hw->bus.func) {
2099 if (eeprom_data & IGB_MAS_ENABLE_0) {
2100 adapter->flags |= IGB_FLAG_MAS_ENABLE;
2101 netdev_info(adapter->netdev,
2102 "MAS: Enabling Media Autosense for port %d\n",
2107 if (eeprom_data & IGB_MAS_ENABLE_1) {
2108 adapter->flags |= IGB_FLAG_MAS_ENABLE;
2109 netdev_info(adapter->netdev,
2110 "MAS: Enabling Media Autosense for port %d\n",
2115 if (eeprom_data & IGB_MAS_ENABLE_2) {
2116 adapter->flags |= IGB_FLAG_MAS_ENABLE;
2117 netdev_info(adapter->netdev,
2118 "MAS: Enabling Media Autosense for port %d\n",
2123 if (eeprom_data & IGB_MAS_ENABLE_3) {
2124 adapter->flags |= IGB_FLAG_MAS_ENABLE;
2125 netdev_info(adapter->netdev,
2126 "MAS: Enabling Media Autosense for port %d\n",
2131 /* Shouldn't get here */
2132 netdev_err(adapter->netdev,
2133 "MAS: Invalid port configuration, returning\n");
2139 * igb_init_i2c - Init I2C interface
2140 * @adapter: pointer to adapter structure
2142 static s32 igb_init_i2c(struct igb_adapter *adapter)
2144 s32 status = E1000_SUCCESS;
2146 /* I2C interface supported on i350 devices */
2147 if (adapter->hw.mac.type != e1000_i350)
2148 return E1000_SUCCESS;
2150 /* Initialize the i2c bus which is controlled by the registers.
2151 * This bus will use the i2c_algo_bit structue that implements
2152 * the protocol through toggling of the 4 bits in the register.
2154 adapter->i2c_adap.owner = THIS_MODULE;
2155 adapter->i2c_algo = igb_i2c_algo;
2156 adapter->i2c_algo.data = adapter;
2157 adapter->i2c_adap.algo_data = &adapter->i2c_algo;
2158 adapter->i2c_adap.dev.parent = &adapter->pdev->dev;
2159 strlcpy(adapter->i2c_adap.name, "igb BB",
2160 sizeof(adapter->i2c_adap.name));
2161 status = i2c_bit_add_bus(&adapter->i2c_adap);
2166 * igb_probe - Device Initialization Routine
2167 * @pdev: PCI device information struct
2168 * @ent: entry in igb_pci_tbl
2170 * Returns 0 on success, negative on failure
2172 * igb_probe initializes an adapter identified by a pci_dev structure.
2173 * The OS initialization, configuring of the adapter private structure,
2174 * and a hardware reset occur.
2176 static int igb_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2178 struct net_device *netdev;
2179 struct igb_adapter *adapter;
2180 struct e1000_hw *hw;
2181 u16 eeprom_data = 0;
2183 static int global_quad_port_a; /* global quad port a indication */
2184 const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
2185 int err, pci_using_dac;
2186 u8 part_str[E1000_PBANUM_LENGTH];
2188 /* Catch broken hardware that put the wrong VF device ID in
2189 * the PCIe SR-IOV capability.
2191 if (pdev->is_virtfn) {
2192 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
2193 pci_name(pdev), pdev->vendor, pdev->device);
2197 err = pci_enable_device_mem(pdev);
2202 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
2206 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
2209 "No usable DMA configuration, aborting\n");
2214 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
2220 pci_enable_pcie_error_reporting(pdev);
2222 pci_set_master(pdev);
2223 pci_save_state(pdev);
2226 netdev = alloc_etherdev_mq(sizeof(struct igb_adapter),
2229 goto err_alloc_etherdev;
2231 SET_NETDEV_DEV(netdev, &pdev->dev);
2233 pci_set_drvdata(pdev, netdev);
2234 adapter = netdev_priv(netdev);
2235 adapter->netdev = netdev;
2236 adapter->pdev = pdev;
2239 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
2242 hw->hw_addr = pci_iomap(pdev, 0, 0);
2246 netdev->netdev_ops = &igb_netdev_ops;
2247 igb_set_ethtool_ops(netdev);
2248 netdev->watchdog_timeo = 5 * HZ;
2250 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2252 netdev->mem_start = pci_resource_start(pdev, 0);
2253 netdev->mem_end = pci_resource_end(pdev, 0);
2255 /* PCI config space info */
2256 hw->vendor_id = pdev->vendor;
2257 hw->device_id = pdev->device;
2258 hw->revision_id = pdev->revision;
2259 hw->subsystem_vendor_id = pdev->subsystem_vendor;
2260 hw->subsystem_device_id = pdev->subsystem_device;
2262 /* Copy the default MAC, PHY and NVM function pointers */
2263 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
2264 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
2265 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
2266 /* Initialize skew-specific constants */
2267 err = ei->get_invariants(hw);
2271 /* setup the private structure */
2272 err = igb_sw_init(adapter);
2276 igb_get_bus_info_pcie(hw);
2278 hw->phy.autoneg_wait_to_complete = false;
2280 /* Copper options */
2281 if (hw->phy.media_type == e1000_media_type_copper) {
2282 hw->phy.mdix = AUTO_ALL_MODES;
2283 hw->phy.disable_polarity_correction = false;
2284 hw->phy.ms_type = e1000_ms_hw_default;
2287 if (igb_check_reset_block(hw))
2288 dev_info(&pdev->dev,
2289 "PHY reset is blocked due to SOL/IDER session.\n");
2291 /* features is initialized to 0 in allocation, it might have bits
2292 * set by igb_sw_init so we should use an or instead of an
2295 netdev->features |= NETIF_F_SG |
2302 NETIF_F_HW_VLAN_CTAG_RX |
2303 NETIF_F_HW_VLAN_CTAG_TX;
2305 /* copy netdev features into list of user selectable features */
2306 netdev->hw_features |= netdev->features;
2307 netdev->hw_features |= NETIF_F_RXALL;
2309 /* set this bit last since it cannot be part of hw_features */
2310 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
2312 netdev->vlan_features |= NETIF_F_TSO |
2318 netdev->priv_flags |= IFF_SUPP_NOFCS;
2320 if (pci_using_dac) {
2321 netdev->features |= NETIF_F_HIGHDMA;
2322 netdev->vlan_features |= NETIF_F_HIGHDMA;
2325 if (hw->mac.type >= e1000_82576) {
2326 netdev->hw_features |= NETIF_F_SCTP_CSUM;
2327 netdev->features |= NETIF_F_SCTP_CSUM;
2330 netdev->priv_flags |= IFF_UNICAST_FLT;
2332 adapter->en_mng_pt = igb_enable_mng_pass_thru(hw);
2334 /* before reading the NVM, reset the controller to put the device in a
2335 * known good starting state
2337 hw->mac.ops.reset_hw(hw);
2339 /* make sure the NVM is good , i211/i210 parts can have special NVM
2340 * that doesn't contain a checksum
2342 switch (hw->mac.type) {
2345 if (igb_get_flash_presence_i210(hw)) {
2346 if (hw->nvm.ops.validate(hw) < 0) {
2348 "The NVM Checksum Is Not Valid\n");
2355 if (hw->nvm.ops.validate(hw) < 0) {
2356 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
2363 /* copy the MAC address out of the NVM */
2364 if (hw->mac.ops.read_mac_addr(hw))
2365 dev_err(&pdev->dev, "NVM Read Error\n");
2367 memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
2369 if (!is_valid_ether_addr(netdev->dev_addr)) {
2370 dev_err(&pdev->dev, "Invalid MAC Address\n");
2375 /* get firmware version for ethtool -i */
2376 igb_set_fw_version(adapter);
2378 setup_timer(&adapter->watchdog_timer, igb_watchdog,
2379 (unsigned long) adapter);
2380 setup_timer(&adapter->phy_info_timer, igb_update_phy_info,
2381 (unsigned long) adapter);
2383 INIT_WORK(&adapter->reset_task, igb_reset_task);
2384 INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
2386 /* Initialize link properties that are user-changeable */
2387 adapter->fc_autoneg = true;
2388 hw->mac.autoneg = true;
2389 hw->phy.autoneg_advertised = 0x2f;
2391 hw->fc.requested_mode = e1000_fc_default;
2392 hw->fc.current_mode = e1000_fc_default;
2394 igb_validate_mdi_setting(hw);
2396 /* By default, support wake on port A */
2397 if (hw->bus.func == 0)
2398 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2400 /* Check the NVM for wake support on non-port A ports */
2401 if (hw->mac.type >= e1000_82580)
2402 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A +
2403 NVM_82580_LAN_FUNC_OFFSET(hw->bus.func), 1,
2405 else if (hw->bus.func == 1)
2406 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
2408 if (eeprom_data & IGB_EEPROM_APME)
2409 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2411 /* now that we have the eeprom settings, apply the special cases where
2412 * the eeprom may be wrong or the board simply won't support wake on
2413 * lan on a particular port
2415 switch (pdev->device) {
2416 case E1000_DEV_ID_82575GB_QUAD_COPPER:
2417 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2419 case E1000_DEV_ID_82575EB_FIBER_SERDES:
2420 case E1000_DEV_ID_82576_FIBER:
2421 case E1000_DEV_ID_82576_SERDES:
2422 /* Wake events only supported on port A for dual fiber
2423 * regardless of eeprom setting
2425 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
2426 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2428 case E1000_DEV_ID_82576_QUAD_COPPER:
2429 case E1000_DEV_ID_82576_QUAD_COPPER_ET2:
2430 /* if quad port adapter, disable WoL on all but port A */
2431 if (global_quad_port_a != 0)
2432 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2434 adapter->flags |= IGB_FLAG_QUAD_PORT_A;
2435 /* Reset for multiple quad port adapters */
2436 if (++global_quad_port_a == 4)
2437 global_quad_port_a = 0;
2440 /* If the device can't wake, don't set software support */
2441 if (!device_can_wakeup(&adapter->pdev->dev))
2442 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2445 /* initialize the wol settings based on the eeprom settings */
2446 if (adapter->flags & IGB_FLAG_WOL_SUPPORTED)
2447 adapter->wol |= E1000_WUFC_MAG;
2449 /* Some vendors want WoL disabled by default, but still supported */
2450 if ((hw->mac.type == e1000_i350) &&
2451 (pdev->subsystem_vendor == PCI_VENDOR_ID_HP)) {
2452 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2456 device_set_wakeup_enable(&adapter->pdev->dev,
2457 adapter->flags & IGB_FLAG_WOL_SUPPORTED);
2459 /* reset the hardware with the new settings */
2462 /* Init the I2C interface */
2463 err = igb_init_i2c(adapter);
2465 dev_err(&pdev->dev, "failed to init i2c interface\n");
2469 /* let the f/w know that the h/w is now under the control of the
2471 igb_get_hw_control(adapter);
2473 strcpy(netdev->name, "eth%d");
2474 err = register_netdev(netdev);
2478 /* carrier off reporting is important to ethtool even BEFORE open */
2479 netif_carrier_off(netdev);
2481 #ifdef CONFIG_IGB_DCA
2482 if (dca_add_requester(&pdev->dev) == 0) {
2483 adapter->flags |= IGB_FLAG_DCA_ENABLED;
2484 dev_info(&pdev->dev, "DCA enabled\n");
2485 igb_setup_dca(adapter);
2489 #ifdef CONFIG_IGB_HWMON
2490 /* Initialize the thermal sensor on i350 devices. */
2491 if (hw->mac.type == e1000_i350 && hw->bus.func == 0) {
2494 /* Read the NVM to determine if this i350 device supports an
2495 * external thermal sensor.
2497 hw->nvm.ops.read(hw, NVM_ETS_CFG, 1, &ets_word);
2498 if (ets_word != 0x0000 && ets_word != 0xFFFF)
2499 adapter->ets = true;
2501 adapter->ets = false;
2502 if (igb_sysfs_init(adapter))
2504 "failed to allocate sysfs resources\n");
2506 adapter->ets = false;
2509 /* Check if Media Autosense is enabled */
2511 if (hw->dev_spec._82575.mas_capable)
2512 igb_init_mas(adapter);
2514 /* do hw tstamp init after resetting */
2515 igb_ptp_init(adapter);
2517 dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
2518 /* print bus type/speed/width info, not applicable to i354 */
2519 if (hw->mac.type != e1000_i354) {
2520 dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
2522 ((hw->bus.speed == e1000_bus_speed_2500) ? "2.5Gb/s" :
2523 (hw->bus.speed == e1000_bus_speed_5000) ? "5.0Gb/s" :
2525 ((hw->bus.width == e1000_bus_width_pcie_x4) ?
2527 (hw->bus.width == e1000_bus_width_pcie_x2) ?
2529 (hw->bus.width == e1000_bus_width_pcie_x1) ?
2530 "Width x1" : "unknown"), netdev->dev_addr);
2533 if ((hw->mac.type >= e1000_i210 ||
2534 igb_get_flash_presence_i210(hw))) {
2535 ret_val = igb_read_part_string(hw, part_str,
2536 E1000_PBANUM_LENGTH);
2538 ret_val = -E1000_ERR_INVM_VALUE_NOT_FOUND;
2542 strcpy(part_str, "Unknown");
2543 dev_info(&pdev->dev, "%s: PBA No: %s\n", netdev->name, part_str);
2544 dev_info(&pdev->dev,
2545 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
2546 adapter->msix_entries ? "MSI-X" :
2547 (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
2548 adapter->num_rx_queues, adapter->num_tx_queues);
2549 switch (hw->mac.type) {
2553 igb_set_eee_i350(hw);
2556 if (hw->phy.media_type == e1000_media_type_copper) {
2557 if ((rd32(E1000_CTRL_EXT) &
2558 E1000_CTRL_EXT_LINK_MODE_SGMII))
2559 igb_set_eee_i354(hw);
2566 pm_runtime_put_noidle(&pdev->dev);
2570 igb_release_hw_control(adapter);
2571 memset(&adapter->i2c_adap, 0, sizeof(adapter->i2c_adap));
2573 if (!igb_check_reset_block(hw))
2576 if (hw->flash_address)
2577 iounmap(hw->flash_address);
2579 igb_clear_interrupt_scheme(adapter);
2580 iounmap(hw->hw_addr);
2582 free_netdev(netdev);
2584 pci_release_selected_regions(pdev,
2585 pci_select_bars(pdev, IORESOURCE_MEM));
2588 pci_disable_device(pdev);
2592 #ifdef CONFIG_PCI_IOV
2593 static int igb_disable_sriov(struct pci_dev *pdev)
2595 struct net_device *netdev = pci_get_drvdata(pdev);
2596 struct igb_adapter *adapter = netdev_priv(netdev);
2597 struct e1000_hw *hw = &adapter->hw;
2599 /* reclaim resources allocated to VFs */
2600 if (adapter->vf_data) {
2601 /* disable iov and allow time for transactions to clear */
2602 if (pci_vfs_assigned(pdev)) {
2603 dev_warn(&pdev->dev,
2604 "Cannot deallocate SR-IOV virtual functions while they are assigned - VFs will not be deallocated\n");
2607 pci_disable_sriov(pdev);
2611 kfree(adapter->vf_data);
2612 adapter->vf_data = NULL;
2613 adapter->vfs_allocated_count = 0;
2614 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
2617 dev_info(&pdev->dev, "IOV Disabled\n");
2619 /* Re-enable DMA Coalescing flag since IOV is turned off */
2620 adapter->flags |= IGB_FLAG_DMAC;
2626 static int igb_enable_sriov(struct pci_dev *pdev, int num_vfs)
2628 struct net_device *netdev = pci_get_drvdata(pdev);
2629 struct igb_adapter *adapter = netdev_priv(netdev);
2630 int old_vfs = pci_num_vf(pdev);
2634 if (!adapter->msix_entries || num_vfs > 7) {
2642 dev_info(&pdev->dev, "%d pre-allocated VFs found - override max_vfs setting of %d\n",
2644 adapter->vfs_allocated_count = old_vfs;
2646 adapter->vfs_allocated_count = num_vfs;
2648 adapter->vf_data = kcalloc(adapter->vfs_allocated_count,
2649 sizeof(struct vf_data_storage), GFP_KERNEL);
2651 /* if allocation failed then we do not support SR-IOV */
2652 if (!adapter->vf_data) {
2653 adapter->vfs_allocated_count = 0;
2655 "Unable to allocate memory for VF Data Storage\n");
2660 /* only call pci_enable_sriov() if no VFs are allocated already */
2662 err = pci_enable_sriov(pdev, adapter->vfs_allocated_count);
2666 dev_info(&pdev->dev, "%d VFs allocated\n",
2667 adapter->vfs_allocated_count);
2668 for (i = 0; i < adapter->vfs_allocated_count; i++)
2669 igb_vf_configure(adapter, i);
2671 /* DMA Coalescing is not supported in IOV mode. */
2672 adapter->flags &= ~IGB_FLAG_DMAC;
2676 kfree(adapter->vf_data);
2677 adapter->vf_data = NULL;
2678 adapter->vfs_allocated_count = 0;
2685 * igb_remove_i2c - Cleanup I2C interface
2686 * @adapter: pointer to adapter structure
2688 static void igb_remove_i2c(struct igb_adapter *adapter)
2690 /* free the adapter bus structure */
2691 i2c_del_adapter(&adapter->i2c_adap);
2695 * igb_remove - Device Removal Routine
2696 * @pdev: PCI device information struct
2698 * igb_remove is called by the PCI subsystem to alert the driver
2699 * that it should release a PCI device. The could be caused by a
2700 * Hot-Plug event, or because the driver is going to be removed from
2703 static void igb_remove(struct pci_dev *pdev)
2705 struct net_device *netdev = pci_get_drvdata(pdev);
2706 struct igb_adapter *adapter = netdev_priv(netdev);
2707 struct e1000_hw *hw = &adapter->hw;
2709 pm_runtime_get_noresume(&pdev->dev);
2710 #ifdef CONFIG_IGB_HWMON
2711 igb_sysfs_exit(adapter);
2713 igb_remove_i2c(adapter);
2714 igb_ptp_stop(adapter);
2715 /* The watchdog timer may be rescheduled, so explicitly
2716 * disable watchdog from being rescheduled.
2718 set_bit(__IGB_DOWN, &adapter->state);
2719 del_timer_sync(&adapter->watchdog_timer);
2720 del_timer_sync(&adapter->phy_info_timer);
2722 cancel_work_sync(&adapter->reset_task);
2723 cancel_work_sync(&adapter->watchdog_task);
2725 #ifdef CONFIG_IGB_DCA
2726 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
2727 dev_info(&pdev->dev, "DCA disabled\n");
2728 dca_remove_requester(&pdev->dev);
2729 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
2730 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
2734 /* Release control of h/w to f/w. If f/w is AMT enabled, this
2735 * would have already happened in close and is redundant.
2737 igb_release_hw_control(adapter);
2739 unregister_netdev(netdev);
2741 igb_clear_interrupt_scheme(adapter);
2743 #ifdef CONFIG_PCI_IOV
2744 igb_disable_sriov(pdev);
2747 iounmap(hw->hw_addr);
2748 if (hw->flash_address)
2749 iounmap(hw->flash_address);
2750 pci_release_selected_regions(pdev,
2751 pci_select_bars(pdev, IORESOURCE_MEM));
2753 kfree(adapter->shadow_vfta);
2754 free_netdev(netdev);
2756 pci_disable_pcie_error_reporting(pdev);
2758 pci_disable_device(pdev);
2762 * igb_probe_vfs - Initialize vf data storage and add VFs to pci config space
2763 * @adapter: board private structure to initialize
2765 * This function initializes the vf specific data storage and then attempts to
2766 * allocate the VFs. The reason for ordering it this way is because it is much
2767 * mor expensive time wise to disable SR-IOV than it is to allocate and free
2768 * the memory for the VFs.
2770 static void igb_probe_vfs(struct igb_adapter *adapter)
2772 #ifdef CONFIG_PCI_IOV
2773 struct pci_dev *pdev = adapter->pdev;
2774 struct e1000_hw *hw = &adapter->hw;
2776 /* Virtualization features not supported on i210 family. */
2777 if ((hw->mac.type == e1000_i210) || (hw->mac.type == e1000_i211))
2780 pci_sriov_set_totalvfs(pdev, 7);
2781 igb_pci_enable_sriov(pdev, max_vfs);
2783 #endif /* CONFIG_PCI_IOV */
2786 static void igb_init_queue_configuration(struct igb_adapter *adapter)
2788 struct e1000_hw *hw = &adapter->hw;
2791 /* Determine the maximum number of RSS queues supported. */
2792 switch (hw->mac.type) {
2794 max_rss_queues = IGB_MAX_RX_QUEUES_I211;
2798 max_rss_queues = IGB_MAX_RX_QUEUES_82575;
2801 /* I350 cannot do RSS and SR-IOV at the same time */
2802 if (!!adapter->vfs_allocated_count) {
2808 if (!!adapter->vfs_allocated_count) {
2816 max_rss_queues = IGB_MAX_RX_QUEUES;
2820 adapter->rss_queues = min_t(u32, max_rss_queues, num_online_cpus());
2822 /* Determine if we need to pair queues. */
2823 switch (hw->mac.type) {
2826 /* Device supports enough interrupts without queue pairing. */
2829 /* If VFs are going to be allocated with RSS queues then we
2830 * should pair the queues in order to conserve interrupts due
2831 * to limited supply.
2833 if ((adapter->rss_queues > 1) &&
2834 (adapter->vfs_allocated_count > 6))
2835 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
2842 /* If rss_queues > half of max_rss_queues, pair the queues in
2843 * order to conserve interrupts due to limited supply.
2845 if (adapter->rss_queues > (max_rss_queues / 2))
2846 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
2852 * igb_sw_init - Initialize general software structures (struct igb_adapter)
2853 * @adapter: board private structure to initialize
2855 * igb_sw_init initializes the Adapter private data structure.
2856 * Fields are initialized based on PCI device information and
2857 * OS network device settings (MTU size).
2859 static int igb_sw_init(struct igb_adapter *adapter)
2861 struct e1000_hw *hw = &adapter->hw;
2862 struct net_device *netdev = adapter->netdev;
2863 struct pci_dev *pdev = adapter->pdev;
2865 pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
2867 /* set default ring sizes */
2868 adapter->tx_ring_count = IGB_DEFAULT_TXD;
2869 adapter->rx_ring_count = IGB_DEFAULT_RXD;
2871 /* set default ITR values */
2872 adapter->rx_itr_setting = IGB_DEFAULT_ITR;
2873 adapter->tx_itr_setting = IGB_DEFAULT_ITR;
2875 /* set default work limits */
2876 adapter->tx_work_limit = IGB_DEFAULT_TX_WORK;
2878 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN +
2880 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2882 spin_lock_init(&adapter->stats64_lock);
2883 #ifdef CONFIG_PCI_IOV
2884 switch (hw->mac.type) {
2888 dev_warn(&pdev->dev,
2889 "Maximum of 7 VFs per PF, using max\n");
2890 max_vfs = adapter->vfs_allocated_count = 7;
2892 adapter->vfs_allocated_count = max_vfs;
2893 if (adapter->vfs_allocated_count)
2894 dev_warn(&pdev->dev,
2895 "Enabling SR-IOV VFs using the module parameter is deprecated - please use the pci sysfs interface.\n");
2900 #endif /* CONFIG_PCI_IOV */
2902 igb_init_queue_configuration(adapter);
2904 /* Setup and initialize a copy of the hw vlan table array */
2905 adapter->shadow_vfta = kcalloc(E1000_VLAN_FILTER_TBL_SIZE, sizeof(u32),
2908 /* This call may decrease the number of queues */
2909 if (igb_init_interrupt_scheme(adapter, true)) {
2910 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
2914 igb_probe_vfs(adapter);
2916 /* Explicitly disable IRQ since the NIC can be in any state. */
2917 igb_irq_disable(adapter);
2919 if (hw->mac.type >= e1000_i350)
2920 adapter->flags &= ~IGB_FLAG_DMAC;
2922 set_bit(__IGB_DOWN, &adapter->state);
2927 * igb_open - Called when a network interface is made active
2928 * @netdev: network interface device structure
2930 * Returns 0 on success, negative value on failure
2932 * The open entry point is called when a network interface is made
2933 * active by the system (IFF_UP). At this point all resources needed
2934 * for transmit and receive operations are allocated, the interrupt
2935 * handler is registered with the OS, the watchdog timer is started,
2936 * and the stack is notified that the interface is ready.
2938 static int __igb_open(struct net_device *netdev, bool resuming)
2940 struct igb_adapter *adapter = netdev_priv(netdev);
2941 struct e1000_hw *hw = &adapter->hw;
2942 struct pci_dev *pdev = adapter->pdev;
2946 /* disallow open during test */
2947 if (test_bit(__IGB_TESTING, &adapter->state)) {
2953 pm_runtime_get_sync(&pdev->dev);
2955 netif_carrier_off(netdev);
2957 /* allocate transmit descriptors */
2958 err = igb_setup_all_tx_resources(adapter);
2962 /* allocate receive descriptors */
2963 err = igb_setup_all_rx_resources(adapter);
2967 igb_power_up_link(adapter);
2969 /* before we allocate an interrupt, we must be ready to handle it.
2970 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
2971 * as soon as we call pci_request_irq, so we have to setup our
2972 * clean_rx handler before we do so.
2974 igb_configure(adapter);
2976 err = igb_request_irq(adapter);
2980 /* Notify the stack of the actual queue counts. */
2981 err = netif_set_real_num_tx_queues(adapter->netdev,
2982 adapter->num_tx_queues);
2984 goto err_set_queues;
2986 err = netif_set_real_num_rx_queues(adapter->netdev,
2987 adapter->num_rx_queues);
2989 goto err_set_queues;
2991 /* From here on the code is the same as igb_up() */
2992 clear_bit(__IGB_DOWN, &adapter->state);
2994 for (i = 0; i < adapter->num_q_vectors; i++)
2995 napi_enable(&(adapter->q_vector[i]->napi));
2997 /* Clear any pending interrupts. */
3000 igb_irq_enable(adapter);
3002 /* notify VFs that reset has been completed */
3003 if (adapter->vfs_allocated_count) {
3004 u32 reg_data = rd32(E1000_CTRL_EXT);
3005 reg_data |= E1000_CTRL_EXT_PFRSTD;
3006 wr32(E1000_CTRL_EXT, reg_data);
3009 netif_tx_start_all_queues(netdev);
3012 pm_runtime_put(&pdev->dev);
3014 /* start the watchdog. */
3015 hw->mac.get_link_status = 1;
3016 schedule_work(&adapter->watchdog_task);
3021 igb_free_irq(adapter);
3023 igb_release_hw_control(adapter);
3024 igb_power_down_link(adapter);
3025 igb_free_all_rx_resources(adapter);
3027 igb_free_all_tx_resources(adapter);
3031 pm_runtime_put(&pdev->dev);
3036 static int igb_open(struct net_device *netdev)
3038 return __igb_open(netdev, false);
3042 * igb_close - Disables a network interface
3043 * @netdev: network interface device structure
3045 * Returns 0, this is not allowed to fail
3047 * The close entry point is called when an interface is de-activated
3048 * by the OS. The hardware is still under the driver's control, but
3049 * needs to be disabled. A global MAC reset is issued to stop the
3050 * hardware, and all transmit and receive resources are freed.
3052 static int __igb_close(struct net_device *netdev, bool suspending)
3054 struct igb_adapter *adapter = netdev_priv(netdev);
3055 struct pci_dev *pdev = adapter->pdev;
3057 WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
3060 pm_runtime_get_sync(&pdev->dev);
3063 igb_free_irq(adapter);
3065 igb_free_all_tx_resources(adapter);
3066 igb_free_all_rx_resources(adapter);
3069 pm_runtime_put_sync(&pdev->dev);
3073 static int igb_close(struct net_device *netdev)
3075 return __igb_close(netdev, false);
3079 * igb_setup_tx_resources - allocate Tx resources (Descriptors)
3080 * @tx_ring: tx descriptor ring (for a specific queue) to setup
3082 * Return 0 on success, negative on failure
3084 int igb_setup_tx_resources(struct igb_ring *tx_ring)
3086 struct device *dev = tx_ring->dev;
3089 size = sizeof(struct igb_tx_buffer) * tx_ring->count;
3091 tx_ring->tx_buffer_info = vzalloc(size);
3092 if (!tx_ring->tx_buffer_info)
3095 /* round up to nearest 4K */
3096 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
3097 tx_ring->size = ALIGN(tx_ring->size, 4096);
3099 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
3100 &tx_ring->dma, GFP_KERNEL);
3104 tx_ring->next_to_use = 0;
3105 tx_ring->next_to_clean = 0;
3110 vfree(tx_ring->tx_buffer_info);
3111 tx_ring->tx_buffer_info = NULL;
3112 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
3117 * igb_setup_all_tx_resources - wrapper to allocate Tx resources
3118 * (Descriptors) for all queues
3119 * @adapter: board private structure
3121 * Return 0 on success, negative on failure
3123 static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
3125 struct pci_dev *pdev = adapter->pdev;
3128 for (i = 0; i < adapter->num_tx_queues; i++) {
3129 err = igb_setup_tx_resources(adapter->tx_ring[i]);
3132 "Allocation for Tx Queue %u failed\n", i);
3133 for (i--; i >= 0; i--)
3134 igb_free_tx_resources(adapter->tx_ring[i]);
3143 * igb_setup_tctl - configure the transmit control registers
3144 * @adapter: Board private structure
3146 void igb_setup_tctl(struct igb_adapter *adapter)
3148 struct e1000_hw *hw = &adapter->hw;
3151 /* disable queue 0 which is enabled by default on 82575 and 82576 */
3152 wr32(E1000_TXDCTL(0), 0);
3154 /* Program the Transmit Control Register */
3155 tctl = rd32(E1000_TCTL);
3156 tctl &= ~E1000_TCTL_CT;
3157 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
3158 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
3160 igb_config_collision_dist(hw);
3162 /* Enable transmits */
3163 tctl |= E1000_TCTL_EN;
3165 wr32(E1000_TCTL, tctl);
3169 * igb_configure_tx_ring - Configure transmit ring after Reset
3170 * @adapter: board private structure
3171 * @ring: tx ring to configure
3173 * Configure a transmit ring after a reset.
3175 void igb_configure_tx_ring(struct igb_adapter *adapter,
3176 struct igb_ring *ring)
3178 struct e1000_hw *hw = &adapter->hw;
3180 u64 tdba = ring->dma;
3181 int reg_idx = ring->reg_idx;
3183 /* disable the queue */
3184 wr32(E1000_TXDCTL(reg_idx), 0);
3188 wr32(E1000_TDLEN(reg_idx),
3189 ring->count * sizeof(union e1000_adv_tx_desc));
3190 wr32(E1000_TDBAL(reg_idx),
3191 tdba & 0x00000000ffffffffULL);
3192 wr32(E1000_TDBAH(reg_idx), tdba >> 32);
3194 ring->tail = hw->hw_addr + E1000_TDT(reg_idx);
3195 wr32(E1000_TDH(reg_idx), 0);
3196 writel(0, ring->tail);
3198 txdctl |= IGB_TX_PTHRESH;
3199 txdctl |= IGB_TX_HTHRESH << 8;
3200 txdctl |= IGB_TX_WTHRESH << 16;
3202 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
3203 wr32(E1000_TXDCTL(reg_idx), txdctl);
3207 * igb_configure_tx - Configure transmit Unit after Reset
3208 * @adapter: board private structure
3210 * Configure the Tx unit of the MAC after a reset.
3212 static void igb_configure_tx(struct igb_adapter *adapter)
3216 for (i = 0; i < adapter->num_tx_queues; i++)
3217 igb_configure_tx_ring(adapter, adapter->tx_ring[i]);
3221 * igb_setup_rx_resources - allocate Rx resources (Descriptors)
3222 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
3224 * Returns 0 on success, negative on failure
3226 int igb_setup_rx_resources(struct igb_ring *rx_ring)
3228 struct device *dev = rx_ring->dev;
3231 size = sizeof(struct igb_rx_buffer) * rx_ring->count;
3233 rx_ring->rx_buffer_info = vzalloc(size);
3234 if (!rx_ring->rx_buffer_info)
3237 /* Round up to nearest 4K */
3238 rx_ring->size = rx_ring->count * sizeof(union e1000_adv_rx_desc);
3239 rx_ring->size = ALIGN(rx_ring->size, 4096);
3241 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
3242 &rx_ring->dma, GFP_KERNEL);
3246 rx_ring->next_to_alloc = 0;
3247 rx_ring->next_to_clean = 0;
3248 rx_ring->next_to_use = 0;
3253 vfree(rx_ring->rx_buffer_info);
3254 rx_ring->rx_buffer_info = NULL;
3255 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
3260 * igb_setup_all_rx_resources - wrapper to allocate Rx resources
3261 * (Descriptors) for all queues
3262 * @adapter: board private structure
3264 * Return 0 on success, negative on failure
3266 static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
3268 struct pci_dev *pdev = adapter->pdev;
3271 for (i = 0; i < adapter->num_rx_queues; i++) {
3272 err = igb_setup_rx_resources(adapter->rx_ring[i]);
3275 "Allocation for Rx Queue %u failed\n", i);
3276 for (i--; i >= 0; i--)
3277 igb_free_rx_resources(adapter->rx_ring[i]);
3286 * igb_setup_mrqc - configure the multiple receive queue control registers
3287 * @adapter: Board private structure
3289 static void igb_setup_mrqc(struct igb_adapter *adapter)
3291 struct e1000_hw *hw = &adapter->hw;
3293 u32 j, num_rx_queues;
3294 static const u32 rsskey[10] = { 0xDA565A6D, 0xC20E5B25, 0x3D256741,
3295 0xB08FA343, 0xCB2BCAD0, 0xB4307BAE,
3296 0xA32DCB77, 0x0CF23080, 0x3BB7426A,
3299 /* Fill out hash function seeds */
3300 for (j = 0; j < 10; j++)
3301 wr32(E1000_RSSRK(j), rsskey[j]);
3303 num_rx_queues = adapter->rss_queues;
3305 switch (hw->mac.type) {
3307 /* 82576 supports 2 RSS queues for SR-IOV */
3308 if (adapter->vfs_allocated_count)
3315 if (adapter->rss_indir_tbl_init != num_rx_queues) {
3316 for (j = 0; j < IGB_RETA_SIZE; j++)
3317 adapter->rss_indir_tbl[j] = (j * num_rx_queues) / IGB_RETA_SIZE;
3318 adapter->rss_indir_tbl_init = num_rx_queues;
3320 igb_write_rss_indir_tbl(adapter);
3322 /* Disable raw packet checksumming so that RSS hash is placed in
3323 * descriptor on writeback. No need to enable TCP/UDP/IP checksum
3324 * offloads as they are enabled by default
3326 rxcsum = rd32(E1000_RXCSUM);
3327 rxcsum |= E1000_RXCSUM_PCSD;
3329 if (adapter->hw.mac.type >= e1000_82576)
3330 /* Enable Receive Checksum Offload for SCTP */
3331 rxcsum |= E1000_RXCSUM_CRCOFL;
3333 /* Don't need to set TUOFL or IPOFL, they default to 1 */
3334 wr32(E1000_RXCSUM, rxcsum);
3336 /* Generate RSS hash based on packet types, TCP/UDP
3337 * port numbers and/or IPv4/v6 src and dst addresses
3339 mrqc = E1000_MRQC_RSS_FIELD_IPV4 |
3340 E1000_MRQC_RSS_FIELD_IPV4_TCP |
3341 E1000_MRQC_RSS_FIELD_IPV6 |
3342 E1000_MRQC_RSS_FIELD_IPV6_TCP |
3343 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX;
3345 if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV4_UDP)
3346 mrqc |= E1000_MRQC_RSS_FIELD_IPV4_UDP;
3347 if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV6_UDP)
3348 mrqc |= E1000_MRQC_RSS_FIELD_IPV6_UDP;
3350 /* If VMDq is enabled then we set the appropriate mode for that, else
3351 * we default to RSS so that an RSS hash is calculated per packet even
3352 * if we are only using one queue
3354 if (adapter->vfs_allocated_count) {
3355 if (hw->mac.type > e1000_82575) {
3356 /* Set the default pool for the PF's first queue */
3357 u32 vtctl = rd32(E1000_VT_CTL);
3358 vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK |
3359 E1000_VT_CTL_DISABLE_DEF_POOL);
3360 vtctl |= adapter->vfs_allocated_count <<
3361 E1000_VT_CTL_DEFAULT_POOL_SHIFT;
3362 wr32(E1000_VT_CTL, vtctl);
3364 if (adapter->rss_queues > 1)
3365 mrqc |= E1000_MRQC_ENABLE_VMDQ_RSS_2Q;
3367 mrqc |= E1000_MRQC_ENABLE_VMDQ;
3369 if (hw->mac.type != e1000_i211)
3370 mrqc |= E1000_MRQC_ENABLE_RSS_4Q;
3372 igb_vmm_control(adapter);
3374 wr32(E1000_MRQC, mrqc);
3378 * igb_setup_rctl - configure the receive control registers
3379 * @adapter: Board private structure
3381 void igb_setup_rctl(struct igb_adapter *adapter)
3383 struct e1000_hw *hw = &adapter->hw;
3386 rctl = rd32(E1000_RCTL);
3388 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
3389 rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
3391 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
3392 (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
3394 /* enable stripping of CRC. It's unlikely this will break BMC
3395 * redirection as it did with e1000. Newer features require
3396 * that the HW strips the CRC.
3398 rctl |= E1000_RCTL_SECRC;
3400 /* disable store bad packets and clear size bits. */
3401 rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
3403 /* enable LPE to prevent packets larger than max_frame_size */
3404 rctl |= E1000_RCTL_LPE;
3406 /* disable queue 0 to prevent tail write w/o re-config */
3407 wr32(E1000_RXDCTL(0), 0);
3409 /* Attention!!! For SR-IOV PF driver operations you must enable
3410 * queue drop for all VF and PF queues to prevent head of line blocking
3411 * if an un-trusted VF does not provide descriptors to hardware.
3413 if (adapter->vfs_allocated_count) {
3414 /* set all queue drop enable bits */
3415 wr32(E1000_QDE, ALL_QUEUES);
3418 /* This is useful for sniffing bad packets. */
3419 if (adapter->netdev->features & NETIF_F_RXALL) {
3420 /* UPE and MPE will be handled by normal PROMISC logic
3421 * in e1000e_set_rx_mode
3423 rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
3424 E1000_RCTL_BAM | /* RX All Bcast Pkts */
3425 E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
3427 rctl &= ~(E1000_RCTL_VFE | /* Disable VLAN filter */
3428 E1000_RCTL_DPF | /* Allow filtered pause */
3429 E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
3430 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
3431 * and that breaks VLANs.
3435 wr32(E1000_RCTL, rctl);
3438 static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
3441 struct e1000_hw *hw = &adapter->hw;
3444 /* if it isn't the PF check to see if VFs are enabled and
3445 * increase the size to support vlan tags
3447 if (vfn < adapter->vfs_allocated_count &&
3448 adapter->vf_data[vfn].vlans_enabled)
3449 size += VLAN_TAG_SIZE;
3451 vmolr = rd32(E1000_VMOLR(vfn));
3452 vmolr &= ~E1000_VMOLR_RLPML_MASK;
3453 vmolr |= size | E1000_VMOLR_LPE;
3454 wr32(E1000_VMOLR(vfn), vmolr);
3460 * igb_rlpml_set - set maximum receive packet size
3461 * @adapter: board private structure
3463 * Configure maximum receivable packet size.
3465 static void igb_rlpml_set(struct igb_adapter *adapter)
3467 u32 max_frame_size = adapter->max_frame_size;
3468 struct e1000_hw *hw = &adapter->hw;
3469 u16 pf_id = adapter->vfs_allocated_count;
3472 igb_set_vf_rlpml(adapter, max_frame_size, pf_id);
3473 /* If we're in VMDQ or SR-IOV mode, then set global RLPML
3474 * to our max jumbo frame size, in case we need to enable
3475 * jumbo frames on one of the rings later.
3476 * This will not pass over-length frames into the default
3477 * queue because it's gated by the VMOLR.RLPML.
3479 max_frame_size = MAX_JUMBO_FRAME_SIZE;
3482 wr32(E1000_RLPML, max_frame_size);
3485 static inline void igb_set_vmolr(struct igb_adapter *adapter,
3488 struct e1000_hw *hw = &adapter->hw;
3491 /* This register exists only on 82576 and newer so if we are older then
3492 * we should exit and do nothing
3494 if (hw->mac.type < e1000_82576)
3497 vmolr = rd32(E1000_VMOLR(vfn));
3498 vmolr |= E1000_VMOLR_STRVLAN; /* Strip vlan tags */
3500 vmolr |= E1000_VMOLR_AUPE; /* Accept untagged packets */
3502 vmolr &= ~(E1000_VMOLR_AUPE); /* Tagged packets ONLY */
3504 /* clear all bits that might not be set */
3505 vmolr &= ~(E1000_VMOLR_BAM | E1000_VMOLR_RSSE);
3507 if (adapter->rss_queues > 1 && vfn == adapter->vfs_allocated_count)
3508 vmolr |= E1000_VMOLR_RSSE; /* enable RSS */
3509 /* for VMDq only allow the VFs and pool 0 to accept broadcast and
3512 if (vfn <= adapter->vfs_allocated_count)
3513 vmolr |= E1000_VMOLR_BAM; /* Accept broadcast */
3515 wr32(E1000_VMOLR(vfn), vmolr);
3519 * igb_configure_rx_ring - Configure a receive ring after Reset
3520 * @adapter: board private structure
3521 * @ring: receive ring to be configured
3523 * Configure the Rx unit of the MAC after a reset.
3525 void igb_configure_rx_ring(struct igb_adapter *adapter,
3526 struct igb_ring *ring)
3528 struct e1000_hw *hw = &adapter->hw;
3529 u64 rdba = ring->dma;
3530 int reg_idx = ring->reg_idx;
3531 u32 srrctl = 0, rxdctl = 0;
3533 /* disable the queue */
3534 wr32(E1000_RXDCTL(reg_idx), 0);
3536 /* Set DMA base address registers */
3537 wr32(E1000_RDBAL(reg_idx),
3538 rdba & 0x00000000ffffffffULL);
3539 wr32(E1000_RDBAH(reg_idx), rdba >> 32);
3540 wr32(E1000_RDLEN(reg_idx),
3541 ring->count * sizeof(union e1000_adv_rx_desc));
3543 /* initialize head and tail */
3544 ring->tail = hw->hw_addr + E1000_RDT(reg_idx);
3545 wr32(E1000_RDH(reg_idx), 0);
3546 writel(0, ring->tail);
3548 /* set descriptor configuration */
3549 srrctl = IGB_RX_HDR_LEN << E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
3550 srrctl |= IGB_RX_BUFSZ >> E1000_SRRCTL_BSIZEPKT_SHIFT;
3551 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
3552 if (hw->mac.type >= e1000_82580)
3553 srrctl |= E1000_SRRCTL_TIMESTAMP;
3554 /* Only set Drop Enable if we are supporting multiple queues */
3555 if (adapter->vfs_allocated_count || adapter->num_rx_queues > 1)
3556 srrctl |= E1000_SRRCTL_DROP_EN;
3558 wr32(E1000_SRRCTL(reg_idx), srrctl);
3560 /* set filtering for VMDQ pools */
3561 igb_set_vmolr(adapter, reg_idx & 0x7, true);
3563 rxdctl |= IGB_RX_PTHRESH;
3564 rxdctl |= IGB_RX_HTHRESH << 8;
3565 rxdctl |= IGB_RX_WTHRESH << 16;
3567 /* enable receive descriptor fetching */
3568 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
3569 wr32(E1000_RXDCTL(reg_idx), rxdctl);
3573 * igb_configure_rx - Configure receive Unit after Reset
3574 * @adapter: board private structure
3576 * Configure the Rx unit of the MAC after a reset.
3578 static void igb_configure_rx(struct igb_adapter *adapter)
3582 /* set UTA to appropriate mode */
3583 igb_set_uta(adapter);
3585 /* set the correct pool for the PF default MAC address in entry 0 */
3586 igb_rar_set_qsel(adapter, adapter->hw.mac.addr, 0,
3587 adapter->vfs_allocated_count);
3589 /* Setup the HW Rx Head and Tail Descriptor Pointers and
3590 * the Base and Length of the Rx Descriptor Ring
3592 for (i = 0; i < adapter->num_rx_queues; i++)
3593 igb_configure_rx_ring(adapter, adapter->rx_ring[i]);
3597 * igb_free_tx_resources - Free Tx Resources per Queue
3598 * @tx_ring: Tx descriptor ring for a specific queue
3600 * Free all transmit software resources
3602 void igb_free_tx_resources(struct igb_ring *tx_ring)
3604 igb_clean_tx_ring(tx_ring);
3606 vfree(tx_ring->tx_buffer_info);
3607 tx_ring->tx_buffer_info = NULL;
3609 /* if not set, then don't free */
3613 dma_free_coherent(tx_ring->dev, tx_ring->size,
3614 tx_ring->desc, tx_ring->dma);
3616 tx_ring->desc = NULL;
3620 * igb_free_all_tx_resources - Free Tx Resources for All Queues
3621 * @adapter: board private structure
3623 * Free all transmit software resources
3625 static void igb_free_all_tx_resources(struct igb_adapter *adapter)
3629 for (i = 0; i < adapter->num_tx_queues; i++)
3630 igb_free_tx_resources(adapter->tx_ring[i]);
3633 void igb_unmap_and_free_tx_resource(struct igb_ring *ring,
3634 struct igb_tx_buffer *tx_buffer)
3636 if (tx_buffer->skb) {
3637 dev_kfree_skb_any(tx_buffer->skb);
3638 if (dma_unmap_len(tx_buffer, len))
3639 dma_unmap_single(ring->dev,
3640 dma_unmap_addr(tx_buffer, dma),
3641 dma_unmap_len(tx_buffer, len),
3643 } else if (dma_unmap_len(tx_buffer, len)) {
3644 dma_unmap_page(ring->dev,
3645 dma_unmap_addr(tx_buffer, dma),
3646 dma_unmap_len(tx_buffer, len),
3649 tx_buffer->next_to_watch = NULL;
3650 tx_buffer->skb = NULL;
3651 dma_unmap_len_set(tx_buffer, len, 0);
3652 /* buffer_info must be completely set up in the transmit path */
3656 * igb_clean_tx_ring - Free Tx Buffers
3657 * @tx_ring: ring to be cleaned
3659 static void igb_clean_tx_ring(struct igb_ring *tx_ring)
3661 struct igb_tx_buffer *buffer_info;
3665 if (!tx_ring->tx_buffer_info)
3667 /* Free all the Tx ring sk_buffs */
3669 for (i = 0; i < tx_ring->count; i++) {
3670 buffer_info = &tx_ring->tx_buffer_info[i];
3671 igb_unmap_and_free_tx_resource(tx_ring, buffer_info);
3674 netdev_tx_reset_queue(txring_txq(tx_ring));
3676 size = sizeof(struct igb_tx_buffer) * tx_ring->count;
3677 memset(tx_ring->tx_buffer_info, 0, size);
3679 /* Zero out the descriptor ring */
3680 memset(tx_ring->desc, 0, tx_ring->size);
3682 tx_ring->next_to_use = 0;
3683 tx_ring->next_to_clean = 0;
3687 * igb_clean_all_tx_rings - Free Tx Buffers for all queues
3688 * @adapter: board private structure
3690 static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
3694 for (i = 0; i < adapter->num_tx_queues; i++)
3695 igb_clean_tx_ring(adapter->tx_ring[i]);
3699 * igb_free_rx_resources - Free Rx Resources
3700 * @rx_ring: ring to clean the resources from
3702 * Free all receive software resources
3704 void igb_free_rx_resources(struct igb_ring *rx_ring)
3706 igb_clean_rx_ring(rx_ring);
3708 vfree(rx_ring->rx_buffer_info);
3709 rx_ring->rx_buffer_info = NULL;
3711 /* if not set, then don't free */
3715 dma_free_coherent(rx_ring->dev, rx_ring->size,
3716 rx_ring->desc, rx_ring->dma);
3718 rx_ring->desc = NULL;
3722 * igb_free_all_rx_resources - Free Rx Resources for All Queues
3723 * @adapter: board private structure
3725 * Free all receive software resources
3727 static void igb_free_all_rx_resources(struct igb_adapter *adapter)
3731 for (i = 0; i < adapter->num_rx_queues; i++)
3732 igb_free_rx_resources(adapter->rx_ring[i]);
3736 * igb_clean_rx_ring - Free Rx Buffers per Queue
3737 * @rx_ring: ring to free buffers from
3739 static void igb_clean_rx_ring(struct igb_ring *rx_ring)
3745 dev_kfree_skb(rx_ring->skb);
3746 rx_ring->skb = NULL;
3748 if (!rx_ring->rx_buffer_info)
3751 /* Free all the Rx ring sk_buffs */
3752 for (i = 0; i < rx_ring->count; i++) {
3753 struct igb_rx_buffer *buffer_info = &rx_ring->rx_buffer_info[i];
3755 if (!buffer_info->page)
3758 dma_unmap_page(rx_ring->dev,
3762 __free_page(buffer_info->page);
3764 buffer_info->page = NULL;
3767 size = sizeof(struct igb_rx_buffer) * rx_ring->count;
3768 memset(rx_ring->rx_buffer_info, 0, size);
3770 /* Zero out the descriptor ring */
3771 memset(rx_ring->desc, 0, rx_ring->size);
3773 rx_ring->next_to_alloc = 0;
3774 rx_ring->next_to_clean = 0;
3775 rx_ring->next_to_use = 0;
3779 * igb_clean_all_rx_rings - Free Rx Buffers for all queues
3780 * @adapter: board private structure
3782 static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
3786 for (i = 0; i < adapter->num_rx_queues; i++)
3787 igb_clean_rx_ring(adapter->rx_ring[i]);
3791 * igb_set_mac - Change the Ethernet Address of the NIC
3792 * @netdev: network interface device structure
3793 * @p: pointer to an address structure
3795 * Returns 0 on success, negative on failure
3797 static int igb_set_mac(struct net_device *netdev, void *p)
3799 struct igb_adapter *adapter = netdev_priv(netdev);
3800 struct e1000_hw *hw = &adapter->hw;
3801 struct sockaddr *addr = p;
3803 if (!is_valid_ether_addr(addr->sa_data))
3804 return -EADDRNOTAVAIL;
3806 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3807 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
3809 /* set the correct pool for the new PF MAC address in entry 0 */
3810 igb_rar_set_qsel(adapter, hw->mac.addr, 0,
3811 adapter->vfs_allocated_count);
3817 * igb_write_mc_addr_list - write multicast addresses to MTA
3818 * @netdev: network interface device structure
3820 * Writes multicast address list to the MTA hash table.
3821 * Returns: -ENOMEM on failure
3822 * 0 on no addresses written
3823 * X on writing X addresses to MTA
3825 static int igb_write_mc_addr_list(struct net_device *netdev)
3827 struct igb_adapter *adapter = netdev_priv(netdev);
3828 struct e1000_hw *hw = &adapter->hw;
3829 struct netdev_hw_addr *ha;
3833 if (netdev_mc_empty(netdev)) {
3834 /* nothing to program, so clear mc list */
3835 igb_update_mc_addr_list(hw, NULL, 0);
3836 igb_restore_vf_multicasts(adapter);
3840 mta_list = kzalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
3844 /* The shared function expects a packed array of only addresses. */
3846 netdev_for_each_mc_addr(ha, netdev)
3847 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3849 igb_update_mc_addr_list(hw, mta_list, i);
3852 return netdev_mc_count(netdev);
3856 * igb_write_uc_addr_list - write unicast addresses to RAR table
3857 * @netdev: network interface device structure
3859 * Writes unicast address list to the RAR table.
3860 * Returns: -ENOMEM on failure/insufficient address space
3861 * 0 on no addresses written
3862 * X on writing X addresses to the RAR table
3864 static int igb_write_uc_addr_list(struct net_device *netdev)
3866 struct igb_adapter *adapter = netdev_priv(netdev);
3867 struct e1000_hw *hw = &adapter->hw;
3868 unsigned int vfn = adapter->vfs_allocated_count;
3869 unsigned int rar_entries = hw->mac.rar_entry_count - (vfn + 1);
3872 /* return ENOMEM indicating insufficient memory for addresses */
3873 if (netdev_uc_count(netdev) > rar_entries)
3876 if (!netdev_uc_empty(netdev) && rar_entries) {
3877 struct netdev_hw_addr *ha;
3879 netdev_for_each_uc_addr(ha, netdev) {
3882 igb_rar_set_qsel(adapter, ha->addr,
3888 /* write the addresses in reverse order to avoid write combining */
3889 for (; rar_entries > 0 ; rar_entries--) {
3890 wr32(E1000_RAH(rar_entries), 0);
3891 wr32(E1000_RAL(rar_entries), 0);
3899 * igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
3900 * @netdev: network interface device structure
3902 * The set_rx_mode entry point is called whenever the unicast or multicast
3903 * address lists or the network interface flags are updated. This routine is
3904 * responsible for configuring the hardware for proper unicast, multicast,
3905 * promiscuous mode, and all-multi behavior.
3907 static void igb_set_rx_mode(struct net_device *netdev)
3909 struct igb_adapter *adapter = netdev_priv(netdev);
3910 struct e1000_hw *hw = &adapter->hw;
3911 unsigned int vfn = adapter->vfs_allocated_count;
3912 u32 rctl, vmolr = 0;
3915 /* Check for Promiscuous and All Multicast modes */
3916 rctl = rd32(E1000_RCTL);
3918 /* clear the effected bits */
3919 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE | E1000_RCTL_VFE);
3921 if (netdev->flags & IFF_PROMISC) {
3922 /* retain VLAN HW filtering if in VT mode */
3923 if (adapter->vfs_allocated_count)
3924 rctl |= E1000_RCTL_VFE;
3925 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3926 vmolr |= (E1000_VMOLR_ROPE | E1000_VMOLR_MPME);
3928 if (netdev->flags & IFF_ALLMULTI) {
3929 rctl |= E1000_RCTL_MPE;
3930 vmolr |= E1000_VMOLR_MPME;
3932 /* Write addresses to the MTA, if the attempt fails
3933 * then we should just turn on promiscuous mode so
3934 * that we can at least receive multicast traffic
3936 count = igb_write_mc_addr_list(netdev);
3938 rctl |= E1000_RCTL_MPE;
3939 vmolr |= E1000_VMOLR_MPME;
3941 vmolr |= E1000_VMOLR_ROMPE;
3944 /* Write addresses to available RAR registers, if there is not
3945 * sufficient space to store all the addresses then enable
3946 * unicast promiscuous mode
3948 count = igb_write_uc_addr_list(netdev);
3950 rctl |= E1000_RCTL_UPE;
3951 vmolr |= E1000_VMOLR_ROPE;
3953 rctl |= E1000_RCTL_VFE;
3955 wr32(E1000_RCTL, rctl);
3957 /* In order to support SR-IOV and eventually VMDq it is necessary to set
3958 * the VMOLR to enable the appropriate modes. Without this workaround
3959 * we will have issues with VLAN tag stripping not being done for frames
3960 * that are only arriving because we are the default pool
3962 if ((hw->mac.type < e1000_82576) || (hw->mac.type > e1000_i350))
3965 vmolr |= rd32(E1000_VMOLR(vfn)) &
3966 ~(E1000_VMOLR_ROPE | E1000_VMOLR_MPME | E1000_VMOLR_ROMPE);
3967 wr32(E1000_VMOLR(vfn), vmolr);
3968 igb_restore_vf_multicasts(adapter);
3971 static void igb_check_wvbr(struct igb_adapter *adapter)
3973 struct e1000_hw *hw = &adapter->hw;
3976 switch (hw->mac.type) {
3979 if (!(wvbr = rd32(E1000_WVBR)))
3986 adapter->wvbr |= wvbr;
3989 #define IGB_STAGGERED_QUEUE_OFFSET 8
3991 static void igb_spoof_check(struct igb_adapter *adapter)
3998 for(j = 0; j < adapter->vfs_allocated_count; j++) {
3999 if (adapter->wvbr & (1 << j) ||
4000 adapter->wvbr & (1 << (j + IGB_STAGGERED_QUEUE_OFFSET))) {
4001 dev_warn(&adapter->pdev->dev,
4002 "Spoof event(s) detected on VF %d\n", j);
4005 (1 << (j + IGB_STAGGERED_QUEUE_OFFSET)));
4010 /* Need to wait a few seconds after link up to get diagnostic information from
4013 static void igb_update_phy_info(unsigned long data)
4015 struct igb_adapter *adapter = (struct igb_adapter *) data;
4016 igb_get_phy_info(&adapter->hw);
4020 * igb_has_link - check shared code for link and determine up/down
4021 * @adapter: pointer to driver private info
4023 bool igb_has_link(struct igb_adapter *adapter)
4025 struct e1000_hw *hw = &adapter->hw;
4026 bool link_active = false;
4028 /* get_link_status is set on LSC (link status) interrupt or
4029 * rx sequence error interrupt. get_link_status will stay
4030 * false until the e1000_check_for_link establishes link
4031 * for copper adapters ONLY
4033 switch (hw->phy.media_type) {
4034 case e1000_media_type_copper:
4035 if (!hw->mac.get_link_status)
4037 case e1000_media_type_internal_serdes:
4038 hw->mac.ops.check_for_link(hw);
4039 link_active = !hw->mac.get_link_status;
4042 case e1000_media_type_unknown:
4046 if (((hw->mac.type == e1000_i210) ||
4047 (hw->mac.type == e1000_i211)) &&
4048 (hw->phy.id == I210_I_PHY_ID)) {
4049 if (!netif_carrier_ok(adapter->netdev)) {
4050 adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
4051 } else if (!(adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)) {
4052 adapter->flags |= IGB_FLAG_NEED_LINK_UPDATE;
4053 adapter->link_check_timeout = jiffies;
4060 static bool igb_thermal_sensor_event(struct e1000_hw *hw, u32 event)
4063 u32 ctrl_ext, thstat;
4065 /* check for thermal sensor event on i350 copper only */
4066 if (hw->mac.type == e1000_i350) {
4067 thstat = rd32(E1000_THSTAT);
4068 ctrl_ext = rd32(E1000_CTRL_EXT);
4070 if ((hw->phy.media_type == e1000_media_type_copper) &&
4071 !(ctrl_ext & E1000_CTRL_EXT_LINK_MODE_SGMII))
4072 ret = !!(thstat & event);
4079 * igb_watchdog - Timer Call-back
4080 * @data: pointer to adapter cast into an unsigned long
4082 static void igb_watchdog(unsigned long data)
4084 struct igb_adapter *adapter = (struct igb_adapter *)data;
4085 /* Do the rest outside of interrupt context */
4086 schedule_work(&adapter->watchdog_task);
4089 static void igb_watchdog_task(struct work_struct *work)
4091 struct igb_adapter *adapter = container_of(work,
4094 struct e1000_hw *hw = &adapter->hw;
4095 struct e1000_phy_info *phy = &hw->phy;
4096 struct net_device *netdev = adapter->netdev;
4101 link = igb_has_link(adapter);
4103 if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE) {
4104 if (time_after(jiffies, (adapter->link_check_timeout + HZ)))
4105 adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
4110 /* Force link down if we have fiber to swap to */
4111 if (adapter->flags & IGB_FLAG_MAS_ENABLE) {
4112 if (hw->phy.media_type == e1000_media_type_copper) {
4113 connsw = rd32(E1000_CONNSW);
4114 if (!(connsw & E1000_CONNSW_AUTOSENSE_EN))
4119 /* Perform a reset if the media type changed. */
4120 if (hw->dev_spec._82575.media_changed) {
4121 hw->dev_spec._82575.media_changed = false;
4122 adapter->flags |= IGB_FLAG_MEDIA_RESET;
4125 /* Cancel scheduled suspend requests. */
4126 pm_runtime_resume(netdev->dev.parent);
4128 if (!netif_carrier_ok(netdev)) {
4130 hw->mac.ops.get_speed_and_duplex(hw,
4131 &adapter->link_speed,
4132 &adapter->link_duplex);
4134 ctrl = rd32(E1000_CTRL);
4135 /* Links status message must follow this format */
4136 printk(KERN_INFO "igb: %s NIC Link is Up %d Mbps %s "
4137 "Duplex, Flow Control: %s\n",
4139 adapter->link_speed,
4140 adapter->link_duplex == FULL_DUPLEX ?
4142 (ctrl & E1000_CTRL_TFCE) &&
4143 (ctrl & E1000_CTRL_RFCE) ? "RX/TX" :
4144 (ctrl & E1000_CTRL_RFCE) ? "RX" :
4145 (ctrl & E1000_CTRL_TFCE) ? "TX" : "None");
4147 /* check if SmartSpeed worked */
4148 igb_check_downshift(hw);
4149 if (phy->speed_downgraded)
4150 netdev_warn(netdev, "Link Speed was downgraded by SmartSpeed\n");
4152 /* check for thermal sensor event */
4153 if (igb_thermal_sensor_event(hw,
4154 E1000_THSTAT_LINK_THROTTLE)) {
4155 netdev_info(netdev, "The network adapter link "
4156 "speed was downshifted because it "
4160 /* adjust timeout factor according to speed/duplex */
4161 adapter->tx_timeout_factor = 1;
4162 switch (adapter->link_speed) {
4164 adapter->tx_timeout_factor = 14;
4167 /* maybe add some timeout factor ? */
4171 netif_carrier_on(netdev);
4173 igb_ping_all_vfs(adapter);
4174 igb_check_vf_rate_limit(adapter);
4176 /* link state has changed, schedule phy info update */
4177 if (!test_bit(__IGB_DOWN, &adapter->state))
4178 mod_timer(&adapter->phy_info_timer,
4179 round_jiffies(jiffies + 2 * HZ));
4182 if (netif_carrier_ok(netdev)) {
4183 adapter->link_speed = 0;
4184 adapter->link_duplex = 0;
4186 /* check for thermal sensor event */
4187 if (igb_thermal_sensor_event(hw,
4188 E1000_THSTAT_PWR_DOWN)) {
4189 netdev_err(netdev, "The network adapter was "
4190 "stopped because it overheated\n");
4193 /* Links status message must follow this format */
4194 printk(KERN_INFO "igb: %s NIC Link is Down\n",
4196 netif_carrier_off(netdev);
4198 igb_ping_all_vfs(adapter);
4200 /* link state has changed, schedule phy info update */
4201 if (!test_bit(__IGB_DOWN, &adapter->state))
4202 mod_timer(&adapter->phy_info_timer,
4203 round_jiffies(jiffies + 2 * HZ));
4205 /* link is down, time to check for alternate media */
4206 if (adapter->flags & IGB_FLAG_MAS_ENABLE) {
4207 igb_check_swap_media(adapter);
4208 if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
4209 schedule_work(&adapter->reset_task);
4210 /* return immediately */
4214 pm_schedule_suspend(netdev->dev.parent,
4217 /* also check for alternate media here */
4218 } else if (!netif_carrier_ok(netdev) &&
4219 (adapter->flags & IGB_FLAG_MAS_ENABLE)) {
4220 igb_check_swap_media(adapter);
4221 if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
4222 schedule_work(&adapter->reset_task);
4223 /* return immediately */
4229 spin_lock(&adapter->stats64_lock);
4230 igb_update_stats(adapter, &adapter->stats64);
4231 spin_unlock(&adapter->stats64_lock);
4233 for (i = 0; i < adapter->num_tx_queues; i++) {
4234 struct igb_ring *tx_ring = adapter->tx_ring[i];
4235 if (!netif_carrier_ok(netdev)) {
4236 /* We've lost link, so the controller stops DMA,
4237 * but we've got queued Tx work that's never going
4238 * to get done, so reset controller to flush Tx.
4239 * (Do the reset outside of interrupt context).
4241 if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) {
4242 adapter->tx_timeout_count++;
4243 schedule_work(&adapter->reset_task);
4244 /* return immediately since reset is imminent */
4249 /* Force detection of hung controller every watchdog period */
4250 set_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
4253 /* Cause software interrupt to ensure Rx ring is cleaned */
4254 if (adapter->msix_entries) {
4256 for (i = 0; i < adapter->num_q_vectors; i++)
4257 eics |= adapter->q_vector[i]->eims_value;
4258 wr32(E1000_EICS, eics);
4260 wr32(E1000_ICS, E1000_ICS_RXDMT0);
4263 igb_spoof_check(adapter);
4264 igb_ptp_rx_hang(adapter);
4266 /* Reset the timer */
4267 if (!test_bit(__IGB_DOWN, &adapter->state)) {
4268 if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)
4269 mod_timer(&adapter->watchdog_timer,
4270 round_jiffies(jiffies + HZ));
4272 mod_timer(&adapter->watchdog_timer,
4273 round_jiffies(jiffies + 2 * HZ));
4277 enum latency_range {
4281 latency_invalid = 255
4285 * igb_update_ring_itr - update the dynamic ITR value based on packet size
4286 * @q_vector: pointer to q_vector
4288 * Stores a new ITR value based on strictly on packet size. This
4289 * algorithm is less sophisticated than that used in igb_update_itr,
4290 * due to the difficulty of synchronizing statistics across multiple
4291 * receive rings. The divisors and thresholds used by this function
4292 * were determined based on theoretical maximum wire speed and testing
4293 * data, in order to minimize response time while increasing bulk
4295 * This functionality is controlled by the InterruptThrottleRate module
4296 * parameter (see igb_param.c)
4297 * NOTE: This function is called only when operating in a multiqueue
4298 * receive environment.
4300 static void igb_update_ring_itr(struct igb_q_vector *q_vector)
4302 int new_val = q_vector->itr_val;
4303 int avg_wire_size = 0;
4304 struct igb_adapter *adapter = q_vector->adapter;
4305 unsigned int packets;
4307 /* For non-gigabit speeds, just fix the interrupt rate at 4000
4308 * ints/sec - ITR timer value of 120 ticks.
4310 if (adapter->link_speed != SPEED_1000) {
4311 new_val = IGB_4K_ITR;
4315 packets = q_vector->rx.total_packets;
4317 avg_wire_size = q_vector->rx.total_bytes / packets;
4319 packets = q_vector->tx.total_packets;
4321 avg_wire_size = max_t(u32, avg_wire_size,
4322 q_vector->tx.total_bytes / packets);
4324 /* if avg_wire_size isn't set no work was done */
4328 /* Add 24 bytes to size to account for CRC, preamble, and gap */
4329 avg_wire_size += 24;
4331 /* Don't starve jumbo frames */
4332 avg_wire_size = min(avg_wire_size, 3000);
4334 /* Give a little boost to mid-size frames */
4335 if ((avg_wire_size > 300) && (avg_wire_size < 1200))
4336 new_val = avg_wire_size / 3;
4338 new_val = avg_wire_size / 2;
4340 /* conservative mode (itr 3) eliminates the lowest_latency setting */
4341 if (new_val < IGB_20K_ITR &&
4342 ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
4343 (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
4344 new_val = IGB_20K_ITR;
4347 if (new_val != q_vector->itr_val) {
4348 q_vector->itr_val = new_val;
4349 q_vector->set_itr = 1;
4352 q_vector->rx.total_bytes = 0;
4353 q_vector->rx.total_packets = 0;
4354 q_vector->tx.total_bytes = 0;
4355 q_vector->tx.total_packets = 0;
4359 * igb_update_itr - update the dynamic ITR value based on statistics
4360 * @q_vector: pointer to q_vector
4361 * @ring_container: ring info to update the itr for
4363 * Stores a new ITR value based on packets and byte
4364 * counts during the last interrupt. The advantage of per interrupt
4365 * computation is faster updates and more accurate ITR for the current
4366 * traffic pattern. Constants in this function were computed
4367 * based on theoretical maximum wire speed and thresholds were set based
4368 * on testing data as well as attempting to minimize response time
4369 * while increasing bulk throughput.
4370 * this functionality is controlled by the InterruptThrottleRate module
4371 * parameter (see igb_param.c)
4372 * NOTE: These calculations are only valid when operating in a single-
4373 * queue environment.
4375 static void igb_update_itr(struct igb_q_vector *q_vector,
4376 struct igb_ring_container *ring_container)
4378 unsigned int packets = ring_container->total_packets;
4379 unsigned int bytes = ring_container->total_bytes;
4380 u8 itrval = ring_container->itr;
4382 /* no packets, exit with status unchanged */
4387 case lowest_latency:
4388 /* handle TSO and jumbo frames */
4389 if (bytes/packets > 8000)
4390 itrval = bulk_latency;
4391 else if ((packets < 5) && (bytes > 512))
4392 itrval = low_latency;
4394 case low_latency: /* 50 usec aka 20000 ints/s */
4395 if (bytes > 10000) {
4396 /* this if handles the TSO accounting */
4397 if (bytes/packets > 8000) {
4398 itrval = bulk_latency;
4399 } else if ((packets < 10) || ((bytes/packets) > 1200)) {
4400 itrval = bulk_latency;
4401 } else if ((packets > 35)) {
4402 itrval = lowest_latency;
4404 } else if (bytes/packets > 2000) {
4405 itrval = bulk_latency;
4406 } else if (packets <= 2 && bytes < 512) {
4407 itrval = lowest_latency;
4410 case bulk_latency: /* 250 usec aka 4000 ints/s */
4411 if (bytes > 25000) {
4413 itrval = low_latency;
4414 } else if (bytes < 1500) {
4415 itrval = low_latency;
4420 /* clear work counters since we have the values we need */
4421 ring_container->total_bytes = 0;
4422 ring_container->total_packets = 0;
4424 /* write updated itr to ring container */
4425 ring_container->itr = itrval;
4428 static void igb_set_itr(struct igb_q_vector *q_vector)
4430 struct igb_adapter *adapter = q_vector->adapter;
4431 u32 new_itr = q_vector->itr_val;
4434 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
4435 if (adapter->link_speed != SPEED_1000) {
4437 new_itr = IGB_4K_ITR;
4441 igb_update_itr(q_vector, &q_vector->tx);
4442 igb_update_itr(q_vector, &q_vector->rx);
4444 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
4446 /* conservative mode (itr 3) eliminates the lowest_latency setting */
4447 if (current_itr == lowest_latency &&
4448 ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
4449 (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
4450 current_itr = low_latency;
4452 switch (current_itr) {
4453 /* counts and packets in update_itr are dependent on these numbers */
4454 case lowest_latency:
4455 new_itr = IGB_70K_ITR; /* 70,000 ints/sec */
4458 new_itr = IGB_20K_ITR; /* 20,000 ints/sec */
4461 new_itr = IGB_4K_ITR; /* 4,000 ints/sec */
4468 if (new_itr != q_vector->itr_val) {
4469 /* this attempts to bias the interrupt rate towards Bulk
4470 * by adding intermediate steps when interrupt rate is
4473 new_itr = new_itr > q_vector->itr_val ?
4474 max((new_itr * q_vector->itr_val) /
4475 (new_itr + (q_vector->itr_val >> 2)),
4477 /* Don't write the value here; it resets the adapter's
4478 * internal timer, and causes us to delay far longer than
4479 * we should between interrupts. Instead, we write the ITR
4480 * value at the beginning of the next interrupt so the timing
4481 * ends up being correct.
4483 q_vector->itr_val = new_itr;
4484 q_vector->set_itr = 1;
4488 static void igb_tx_ctxtdesc(struct igb_ring *tx_ring, u32 vlan_macip_lens,
4489 u32 type_tucmd, u32 mss_l4len_idx)
4491 struct e1000_adv_tx_context_desc *context_desc;
4492 u16 i = tx_ring->next_to_use;
4494 context_desc = IGB_TX_CTXTDESC(tx_ring, i);
4497 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
4499 /* set bits to identify this as an advanced context descriptor */
4500 type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
4502 /* For 82575, context index must be unique per ring. */
4503 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
4504 mss_l4len_idx |= tx_ring->reg_idx << 4;
4506 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
4507 context_desc->seqnum_seed = 0;
4508 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd);
4509 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
4512 static int igb_tso(struct igb_ring *tx_ring,
4513 struct igb_tx_buffer *first,
4516 struct sk_buff *skb = first->skb;
4517 u32 vlan_macip_lens, type_tucmd;
4518 u32 mss_l4len_idx, l4len;
4520 if (skb->ip_summed != CHECKSUM_PARTIAL)
4523 if (!skb_is_gso(skb))
4526 if (skb_header_cloned(skb)) {
4527 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4532 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
4533 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
4535 if (first->protocol == __constant_htons(ETH_P_IP)) {
4536 struct iphdr *iph = ip_hdr(skb);
4539 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
4543 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
4544 first->tx_flags |= IGB_TX_FLAGS_TSO |
4547 } else if (skb_is_gso_v6(skb)) {
4548 ipv6_hdr(skb)->payload_len = 0;
4549 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4550 &ipv6_hdr(skb)->daddr,
4552 first->tx_flags |= IGB_TX_FLAGS_TSO |
4556 /* compute header lengths */
4557 l4len = tcp_hdrlen(skb);
4558 *hdr_len = skb_transport_offset(skb) + l4len;
4560 /* update gso size and bytecount with header size */
4561 first->gso_segs = skb_shinfo(skb)->gso_segs;
4562 first->bytecount += (first->gso_segs - 1) * *hdr_len;
4565 mss_l4len_idx = l4len << E1000_ADVTXD_L4LEN_SHIFT;
4566 mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT;
4568 /* VLAN MACLEN IPLEN */
4569 vlan_macip_lens = skb_network_header_len(skb);
4570 vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
4571 vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
4573 igb_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
4578 static void igb_tx_csum(struct igb_ring *tx_ring, struct igb_tx_buffer *first)
4580 struct sk_buff *skb = first->skb;
4581 u32 vlan_macip_lens = 0;
4582 u32 mss_l4len_idx = 0;
4585 if (skb->ip_summed != CHECKSUM_PARTIAL) {
4586 if (!(first->tx_flags & IGB_TX_FLAGS_VLAN))
4590 switch (first->protocol) {
4591 case __constant_htons(ETH_P_IP):
4592 vlan_macip_lens |= skb_network_header_len(skb);
4593 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
4594 l4_hdr = ip_hdr(skb)->protocol;
4596 case __constant_htons(ETH_P_IPV6):
4597 vlan_macip_lens |= skb_network_header_len(skb);
4598 l4_hdr = ipv6_hdr(skb)->nexthdr;
4601 if (unlikely(net_ratelimit())) {
4602 dev_warn(tx_ring->dev,
4603 "partial checksum but proto=%x!\n",
4611 type_tucmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
4612 mss_l4len_idx = tcp_hdrlen(skb) <<
4613 E1000_ADVTXD_L4LEN_SHIFT;
4616 type_tucmd |= E1000_ADVTXD_TUCMD_L4T_SCTP;
4617 mss_l4len_idx = sizeof(struct sctphdr) <<
4618 E1000_ADVTXD_L4LEN_SHIFT;
4621 mss_l4len_idx = sizeof(struct udphdr) <<
4622 E1000_ADVTXD_L4LEN_SHIFT;
4625 if (unlikely(net_ratelimit())) {
4626 dev_warn(tx_ring->dev,
4627 "partial checksum but l4 proto=%x!\n",
4633 /* update TX checksum flag */
4634 first->tx_flags |= IGB_TX_FLAGS_CSUM;
4637 vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
4638 vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
4640 igb_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
4643 #define IGB_SET_FLAG(_input, _flag, _result) \
4644 ((_flag <= _result) ? \
4645 ((u32)(_input & _flag) * (_result / _flag)) : \
4646 ((u32)(_input & _flag) / (_flag / _result)))
4648 static u32 igb_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
4650 /* set type for advanced descriptor with frame checksum insertion */
4651 u32 cmd_type = E1000_ADVTXD_DTYP_DATA |
4652 E1000_ADVTXD_DCMD_DEXT |
4653 E1000_ADVTXD_DCMD_IFCS;
4655 /* set HW vlan bit if vlan is present */
4656 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_VLAN,
4657 (E1000_ADVTXD_DCMD_VLE));
4659 /* set segmentation bits for TSO */
4660 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSO,
4661 (E1000_ADVTXD_DCMD_TSE));
4663 /* set timestamp bit if present */
4664 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSTAMP,
4665 (E1000_ADVTXD_MAC_TSTAMP));
4667 /* insert frame checksum */
4668 cmd_type ^= IGB_SET_FLAG(skb->no_fcs, 1, E1000_ADVTXD_DCMD_IFCS);
4673 static void igb_tx_olinfo_status(struct igb_ring *tx_ring,
4674 union e1000_adv_tx_desc *tx_desc,
4675 u32 tx_flags, unsigned int paylen)
4677 u32 olinfo_status = paylen << E1000_ADVTXD_PAYLEN_SHIFT;
4679 /* 82575 requires a unique index per ring */
4680 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
4681 olinfo_status |= tx_ring->reg_idx << 4;
4683 /* insert L4 checksum */
4684 olinfo_status |= IGB_SET_FLAG(tx_flags,
4686 (E1000_TXD_POPTS_TXSM << 8));
4688 /* insert IPv4 checksum */
4689 olinfo_status |= IGB_SET_FLAG(tx_flags,
4691 (E1000_TXD_POPTS_IXSM << 8));
4693 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
4696 static void igb_tx_map(struct igb_ring *tx_ring,
4697 struct igb_tx_buffer *first,
4700 struct sk_buff *skb = first->skb;
4701 struct igb_tx_buffer *tx_buffer;
4702 union e1000_adv_tx_desc *tx_desc;
4703 struct skb_frag_struct *frag;
4705 unsigned int data_len, size;
4706 u32 tx_flags = first->tx_flags;
4707 u32 cmd_type = igb_tx_cmd_type(skb, tx_flags);
4708 u16 i = tx_ring->next_to_use;
4710 tx_desc = IGB_TX_DESC(tx_ring, i);
4712 igb_tx_olinfo_status(tx_ring, tx_desc, tx_flags, skb->len - hdr_len);
4714 size = skb_headlen(skb);
4715 data_len = skb->data_len;
4717 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
4721 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
4722 if (dma_mapping_error(tx_ring->dev, dma))
4725 /* record length, and DMA address */
4726 dma_unmap_len_set(tx_buffer, len, size);
4727 dma_unmap_addr_set(tx_buffer, dma, dma);
4729 tx_desc->read.buffer_addr = cpu_to_le64(dma);
4731 while (unlikely(size > IGB_MAX_DATA_PER_TXD)) {
4732 tx_desc->read.cmd_type_len =
4733 cpu_to_le32(cmd_type ^ IGB_MAX_DATA_PER_TXD);
4737 if (i == tx_ring->count) {
4738 tx_desc = IGB_TX_DESC(tx_ring, 0);
4741 tx_desc->read.olinfo_status = 0;
4743 dma += IGB_MAX_DATA_PER_TXD;
4744 size -= IGB_MAX_DATA_PER_TXD;
4746 tx_desc->read.buffer_addr = cpu_to_le64(dma);
4749 if (likely(!data_len))
4752 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
4756 if (i == tx_ring->count) {
4757 tx_desc = IGB_TX_DESC(tx_ring, 0);
4760 tx_desc->read.olinfo_status = 0;
4762 size = skb_frag_size(frag);
4765 dma = skb_frag_dma_map(tx_ring->dev, frag, 0,
4766 size, DMA_TO_DEVICE);
4768 tx_buffer = &tx_ring->tx_buffer_info[i];
4771 /* write last descriptor with RS and EOP bits */
4772 cmd_type |= size | IGB_TXD_DCMD;
4773 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
4775 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
4777 /* set the timestamp */
4778 first->time_stamp = jiffies;
4780 /* Force memory writes to complete before letting h/w know there
4781 * are new descriptors to fetch. (Only applicable for weak-ordered
4782 * memory model archs, such as IA-64).
4784 * We also need this memory barrier to make certain all of the
4785 * status bits have been updated before next_to_watch is written.
4789 /* set next_to_watch value indicating a packet is present */
4790 first->next_to_watch = tx_desc;
4793 if (i == tx_ring->count)
4796 tx_ring->next_to_use = i;
4798 writel(i, tx_ring->tail);
4800 /* we need this if more than one processor can write to our tail
4801 * at a time, it synchronizes IO on IA64/Altix systems
4808 dev_err(tx_ring->dev, "TX DMA map failed\n");
4810 /* clear dma mappings for failed tx_buffer_info map */
4812 tx_buffer = &tx_ring->tx_buffer_info[i];
4813 igb_unmap_and_free_tx_resource(tx_ring, tx_buffer);
4814 if (tx_buffer == first)
4821 tx_ring->next_to_use = i;
4824 static int __igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
4826 struct net_device *netdev = tx_ring->netdev;
4828 netif_stop_subqueue(netdev, tx_ring->queue_index);
4830 /* Herbert's original patch had:
4831 * smp_mb__after_netif_stop_queue();
4832 * but since that doesn't exist yet, just open code it.
4836 /* We need to check again in a case another CPU has just
4837 * made room available.
4839 if (igb_desc_unused(tx_ring) < size)
4843 netif_wake_subqueue(netdev, tx_ring->queue_index);
4845 u64_stats_update_begin(&tx_ring->tx_syncp2);
4846 tx_ring->tx_stats.restart_queue2++;
4847 u64_stats_update_end(&tx_ring->tx_syncp2);
4852 static inline int igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
4854 if (igb_desc_unused(tx_ring) >= size)
4856 return __igb_maybe_stop_tx(tx_ring, size);
4859 netdev_tx_t igb_xmit_frame_ring(struct sk_buff *skb,
4860 struct igb_ring *tx_ring)
4862 struct igb_tx_buffer *first;
4865 u16 count = TXD_USE_COUNT(skb_headlen(skb));
4866 __be16 protocol = vlan_get_protocol(skb);
4869 /* need: 1 descriptor per page * PAGE_SIZE/IGB_MAX_DATA_PER_TXD,
4870 * + 1 desc for skb_headlen/IGB_MAX_DATA_PER_TXD,
4871 * + 2 desc gap to keep tail from touching head,
4872 * + 1 desc for context descriptor,
4873 * otherwise try next time
4875 if (NETDEV_FRAG_PAGE_MAX_SIZE > IGB_MAX_DATA_PER_TXD) {
4877 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
4878 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
4880 count += skb_shinfo(skb)->nr_frags;
4883 if (igb_maybe_stop_tx(tx_ring, count + 3)) {
4884 /* this is a hard error */
4885 return NETDEV_TX_BUSY;
4888 /* record the location of the first descriptor for this packet */
4889 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
4891 first->bytecount = skb->len;
4892 first->gso_segs = 1;
4894 skb_tx_timestamp(skb);
4896 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
4897 struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
4899 if (!(adapter->ptp_tx_skb)) {
4900 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
4901 tx_flags |= IGB_TX_FLAGS_TSTAMP;
4903 adapter->ptp_tx_skb = skb_get(skb);
4904 adapter->ptp_tx_start = jiffies;
4905 if (adapter->hw.mac.type == e1000_82576)
4906 schedule_work(&adapter->ptp_tx_work);
4910 if (vlan_tx_tag_present(skb)) {
4911 tx_flags |= IGB_TX_FLAGS_VLAN;
4912 tx_flags |= (vlan_tx_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
4915 /* record initial flags and protocol */
4916 first->tx_flags = tx_flags;
4917 first->protocol = protocol;
4919 tso = igb_tso(tx_ring, first, &hdr_len);
4923 igb_tx_csum(tx_ring, first);
4925 igb_tx_map(tx_ring, first, hdr_len);
4927 /* Make sure there is space in the ring for the next send. */
4928 igb_maybe_stop_tx(tx_ring, DESC_NEEDED);
4930 return NETDEV_TX_OK;
4933 igb_unmap_and_free_tx_resource(tx_ring, first);
4935 return NETDEV_TX_OK;
4938 static inline struct igb_ring *igb_tx_queue_mapping(struct igb_adapter *adapter,
4939 struct sk_buff *skb)
4941 unsigned int r_idx = skb->queue_mapping;
4943 if (r_idx >= adapter->num_tx_queues)
4944 r_idx = r_idx % adapter->num_tx_queues;
4946 return adapter->tx_ring[r_idx];
4949 static netdev_tx_t igb_xmit_frame(struct sk_buff *skb,
4950 struct net_device *netdev)
4952 struct igb_adapter *adapter = netdev_priv(netdev);
4954 if (test_bit(__IGB_DOWN, &adapter->state)) {
4955 dev_kfree_skb_any(skb);
4956 return NETDEV_TX_OK;
4959 if (skb->len <= 0) {
4960 dev_kfree_skb_any(skb);
4961 return NETDEV_TX_OK;
4964 /* The minimum packet size with TCTL.PSP set is 17 so pad the skb
4965 * in order to meet this minimum size requirement.
4967 if (unlikely(skb->len < 17)) {
4968 if (skb_pad(skb, 17 - skb->len))
4969 return NETDEV_TX_OK;
4971 skb_set_tail_pointer(skb, 17);
4974 return igb_xmit_frame_ring(skb, igb_tx_queue_mapping(adapter, skb));
4978 * igb_tx_timeout - Respond to a Tx Hang
4979 * @netdev: network interface device structure
4981 static void igb_tx_timeout(struct net_device *netdev)
4983 struct igb_adapter *adapter = netdev_priv(netdev);
4984 struct e1000_hw *hw = &adapter->hw;
4986 /* Do the reset outside of interrupt context */
4987 adapter->tx_timeout_count++;
4989 if (hw->mac.type >= e1000_82580)
4990 hw->dev_spec._82575.global_device_reset = true;
4992 schedule_work(&adapter->reset_task);
4994 (adapter->eims_enable_mask & ~adapter->eims_other));
4997 static void igb_reset_task(struct work_struct *work)
4999 struct igb_adapter *adapter;
5000 adapter = container_of(work, struct igb_adapter, reset_task);
5003 netdev_err(adapter->netdev, "Reset adapter\n");
5004 igb_reinit_locked(adapter);
5008 * igb_get_stats64 - Get System Network Statistics
5009 * @netdev: network interface device structure
5010 * @stats: rtnl_link_stats64 pointer
5012 static struct rtnl_link_stats64 *igb_get_stats64(struct net_device *netdev,
5013 struct rtnl_link_stats64 *stats)
5015 struct igb_adapter *adapter = netdev_priv(netdev);
5017 spin_lock(&adapter->stats64_lock);
5018 igb_update_stats(adapter, &adapter->stats64);
5019 memcpy(stats, &adapter->stats64, sizeof(*stats));
5020 spin_unlock(&adapter->stats64_lock);
5026 * igb_change_mtu - Change the Maximum Transfer Unit
5027 * @netdev: network interface device structure
5028 * @new_mtu: new value for maximum frame size
5030 * Returns 0 on success, negative on failure
5032 static int igb_change_mtu(struct net_device *netdev, int new_mtu)
5034 struct igb_adapter *adapter = netdev_priv(netdev);
5035 struct pci_dev *pdev = adapter->pdev;
5036 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
5038 if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
5039 dev_err(&pdev->dev, "Invalid MTU setting\n");
5043 #define MAX_STD_JUMBO_FRAME_SIZE 9238
5044 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
5045 dev_err(&pdev->dev, "MTU > 9216 not supported.\n");
5049 /* adjust max frame to be at least the size of a standard frame */
5050 if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
5051 max_frame = ETH_FRAME_LEN + ETH_FCS_LEN;
5053 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
5056 /* igb_down has a dependency on max_frame_size */
5057 adapter->max_frame_size = max_frame;
5059 if (netif_running(netdev))
5062 dev_info(&pdev->dev, "changing MTU from %d to %d\n",
5063 netdev->mtu, new_mtu);
5064 netdev->mtu = new_mtu;
5066 if (netif_running(netdev))
5071 clear_bit(__IGB_RESETTING, &adapter->state);
5077 * igb_update_stats - Update the board statistics counters
5078 * @adapter: board private structure
5080 void igb_update_stats(struct igb_adapter *adapter,
5081 struct rtnl_link_stats64 *net_stats)
5083 struct e1000_hw *hw = &adapter->hw;
5084 struct pci_dev *pdev = adapter->pdev;
5090 u64 _bytes, _packets;
5092 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
5094 /* Prevent stats update while adapter is being reset, or if the pci
5095 * connection is down.
5097 if (adapter->link_speed == 0)
5099 if (pci_channel_offline(pdev))
5106 for (i = 0; i < adapter->num_rx_queues; i++) {
5107 u32 rqdpc = rd32(E1000_RQDPC(i));
5108 struct igb_ring *ring = adapter->rx_ring[i];
5111 ring->rx_stats.drops += rqdpc;
5112 net_stats->rx_fifo_errors += rqdpc;
5116 start = u64_stats_fetch_begin_bh(&ring->rx_syncp);
5117 _bytes = ring->rx_stats.bytes;
5118 _packets = ring->rx_stats.packets;
5119 } while (u64_stats_fetch_retry_bh(&ring->rx_syncp, start));
5121 packets += _packets;
5124 net_stats->rx_bytes = bytes;
5125 net_stats->rx_packets = packets;
5129 for (i = 0; i < adapter->num_tx_queues; i++) {
5130 struct igb_ring *ring = adapter->tx_ring[i];
5132 start = u64_stats_fetch_begin_bh(&ring->tx_syncp);
5133 _bytes = ring->tx_stats.bytes;
5134 _packets = ring->tx_stats.packets;
5135 } while (u64_stats_fetch_retry_bh(&ring->tx_syncp, start));
5137 packets += _packets;
5139 net_stats->tx_bytes = bytes;
5140 net_stats->tx_packets = packets;
5143 /* read stats registers */
5144 adapter->stats.crcerrs += rd32(E1000_CRCERRS);
5145 adapter->stats.gprc += rd32(E1000_GPRC);
5146 adapter->stats.gorc += rd32(E1000_GORCL);
5147 rd32(E1000_GORCH); /* clear GORCL */
5148 adapter->stats.bprc += rd32(E1000_BPRC);
5149 adapter->stats.mprc += rd32(E1000_MPRC);
5150 adapter->stats.roc += rd32(E1000_ROC);
5152 adapter->stats.prc64 += rd32(E1000_PRC64);
5153 adapter->stats.prc127 += rd32(E1000_PRC127);
5154 adapter->stats.prc255 += rd32(E1000_PRC255);
5155 adapter->stats.prc511 += rd32(E1000_PRC511);
5156 adapter->stats.prc1023 += rd32(E1000_PRC1023);
5157 adapter->stats.prc1522 += rd32(E1000_PRC1522);
5158 adapter->stats.symerrs += rd32(E1000_SYMERRS);
5159 adapter->stats.sec += rd32(E1000_SEC);
5161 mpc = rd32(E1000_MPC);
5162 adapter->stats.mpc += mpc;
5163 net_stats->rx_fifo_errors += mpc;
5164 adapter->stats.scc += rd32(E1000_SCC);
5165 adapter->stats.ecol += rd32(E1000_ECOL);
5166 adapter->stats.mcc += rd32(E1000_MCC);
5167 adapter->stats.latecol += rd32(E1000_LATECOL);
5168 adapter->stats.dc += rd32(E1000_DC);
5169 adapter->stats.rlec += rd32(E1000_RLEC);
5170 adapter->stats.xonrxc += rd32(E1000_XONRXC);
5171 adapter->stats.xontxc += rd32(E1000_XONTXC);
5172 adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
5173 adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
5174 adapter->stats.fcruc += rd32(E1000_FCRUC);
5175 adapter->stats.gptc += rd32(E1000_GPTC);
5176 adapter->stats.gotc += rd32(E1000_GOTCL);
5177 rd32(E1000_GOTCH); /* clear GOTCL */
5178 adapter->stats.rnbc += rd32(E1000_RNBC);
5179 adapter->stats.ruc += rd32(E1000_RUC);
5180 adapter->stats.rfc += rd32(E1000_RFC);
5181 adapter->stats.rjc += rd32(E1000_RJC);
5182 adapter->stats.tor += rd32(E1000_TORH);
5183 adapter->stats.tot += rd32(E1000_TOTH);
5184 adapter->stats.tpr += rd32(E1000_TPR);
5186 adapter->stats.ptc64 += rd32(E1000_PTC64);
5187 adapter->stats.ptc127 += rd32(E1000_PTC127);
5188 adapter->stats.ptc255 += rd32(E1000_PTC255);
5189 adapter->stats.ptc511 += rd32(E1000_PTC511);
5190 adapter->stats.ptc1023 += rd32(E1000_PTC1023);
5191 adapter->stats.ptc1522 += rd32(E1000_PTC1522);
5193 adapter->stats.mptc += rd32(E1000_MPTC);
5194 adapter->stats.bptc += rd32(E1000_BPTC);
5196 adapter->stats.tpt += rd32(E1000_TPT);
5197 adapter->stats.colc += rd32(E1000_COLC);
5199 adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
5200 /* read internal phy specific stats */
5201 reg = rd32(E1000_CTRL_EXT);
5202 if (!(reg & E1000_CTRL_EXT_LINK_MODE_MASK)) {
5203 adapter->stats.rxerrc += rd32(E1000_RXERRC);
5205 /* this stat has invalid values on i210/i211 */
5206 if ((hw->mac.type != e1000_i210) &&
5207 (hw->mac.type != e1000_i211))
5208 adapter->stats.tncrs += rd32(E1000_TNCRS);
5211 adapter->stats.tsctc += rd32(E1000_TSCTC);
5212 adapter->stats.tsctfc += rd32(E1000_TSCTFC);
5214 adapter->stats.iac += rd32(E1000_IAC);
5215 adapter->stats.icrxoc += rd32(E1000_ICRXOC);
5216 adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
5217 adapter->stats.icrxatc += rd32(E1000_ICRXATC);
5218 adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
5219 adapter->stats.ictxatc += rd32(E1000_ICTXATC);
5220 adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
5221 adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
5222 adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
5224 /* Fill out the OS statistics structure */
5225 net_stats->multicast = adapter->stats.mprc;
5226 net_stats->collisions = adapter->stats.colc;
5230 /* RLEC on some newer hardware can be incorrect so build
5231 * our own version based on RUC and ROC
5233 net_stats->rx_errors = adapter->stats.rxerrc +
5234 adapter->stats.crcerrs + adapter->stats.algnerrc +
5235 adapter->stats.ruc + adapter->stats.roc +
5236 adapter->stats.cexterr;
5237 net_stats->rx_length_errors = adapter->stats.ruc +
5239 net_stats->rx_crc_errors = adapter->stats.crcerrs;
5240 net_stats->rx_frame_errors = adapter->stats.algnerrc;
5241 net_stats->rx_missed_errors = adapter->stats.mpc;
5244 net_stats->tx_errors = adapter->stats.ecol +
5245 adapter->stats.latecol;
5246 net_stats->tx_aborted_errors = adapter->stats.ecol;
5247 net_stats->tx_window_errors = adapter->stats.latecol;
5248 net_stats->tx_carrier_errors = adapter->stats.tncrs;
5250 /* Tx Dropped needs to be maintained elsewhere */
5253 if (hw->phy.media_type == e1000_media_type_copper) {
5254 if ((adapter->link_speed == SPEED_1000) &&
5255 (!igb_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
5256 phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
5257 adapter->phy_stats.idle_errors += phy_tmp;
5261 /* Management Stats */
5262 adapter->stats.mgptc += rd32(E1000_MGTPTC);
5263 adapter->stats.mgprc += rd32(E1000_MGTPRC);
5264 adapter->stats.mgpdc += rd32(E1000_MGTPDC);
5267 reg = rd32(E1000_MANC);
5268 if (reg & E1000_MANC_EN_BMC2OS) {
5269 adapter->stats.o2bgptc += rd32(E1000_O2BGPTC);
5270 adapter->stats.o2bspc += rd32(E1000_O2BSPC);
5271 adapter->stats.b2ospc += rd32(E1000_B2OSPC);
5272 adapter->stats.b2ogprc += rd32(E1000_B2OGPRC);
5276 static irqreturn_t igb_msix_other(int irq, void *data)
5278 struct igb_adapter *adapter = data;
5279 struct e1000_hw *hw = &adapter->hw;
5280 u32 icr = rd32(E1000_ICR);
5281 /* reading ICR causes bit 31 of EICR to be cleared */
5283 if (icr & E1000_ICR_DRSTA)
5284 schedule_work(&adapter->reset_task);
5286 if (icr & E1000_ICR_DOUTSYNC) {
5287 /* HW is reporting DMA is out of sync */
5288 adapter->stats.doosync++;
5289 /* The DMA Out of Sync is also indication of a spoof event
5290 * in IOV mode. Check the Wrong VM Behavior register to
5291 * see if it is really a spoof event.
5293 igb_check_wvbr(adapter);
5296 /* Check for a mailbox event */
5297 if (icr & E1000_ICR_VMMB)
5298 igb_msg_task(adapter);
5300 if (icr & E1000_ICR_LSC) {
5301 hw->mac.get_link_status = 1;
5302 /* guard against interrupt when we're going down */
5303 if (!test_bit(__IGB_DOWN, &adapter->state))
5304 mod_timer(&adapter->watchdog_timer, jiffies + 1);
5307 if (icr & E1000_ICR_TS) {
5308 u32 tsicr = rd32(E1000_TSICR);
5310 if (tsicr & E1000_TSICR_TXTS) {
5311 /* acknowledge the interrupt */
5312 wr32(E1000_TSICR, E1000_TSICR_TXTS);
5313 /* retrieve hardware timestamp */
5314 schedule_work(&adapter->ptp_tx_work);
5318 wr32(E1000_EIMS, adapter->eims_other);
5323 static void igb_write_itr(struct igb_q_vector *q_vector)
5325 struct igb_adapter *adapter = q_vector->adapter;
5326 u32 itr_val = q_vector->itr_val & 0x7FFC;
5328 if (!q_vector->set_itr)
5334 if (adapter->hw.mac.type == e1000_82575)
5335 itr_val |= itr_val << 16;
5337 itr_val |= E1000_EITR_CNT_IGNR;
5339 writel(itr_val, q_vector->itr_register);
5340 q_vector->set_itr = 0;
5343 static irqreturn_t igb_msix_ring(int irq, void *data)
5345 struct igb_q_vector *q_vector = data;
5347 /* Write the ITR value calculated from the previous interrupt. */
5348 igb_write_itr(q_vector);
5350 napi_schedule(&q_vector->napi);
5355 #ifdef CONFIG_IGB_DCA
5356 static void igb_update_tx_dca(struct igb_adapter *adapter,
5357 struct igb_ring *tx_ring,
5360 struct e1000_hw *hw = &adapter->hw;
5361 u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
5363 if (hw->mac.type != e1000_82575)
5364 txctrl <<= E1000_DCA_TXCTRL_CPUID_SHIFT;
5366 /* We can enable relaxed ordering for reads, but not writes when
5367 * DCA is enabled. This is due to a known issue in some chipsets
5368 * which will cause the DCA tag to be cleared.
5370 txctrl |= E1000_DCA_TXCTRL_DESC_RRO_EN |
5371 E1000_DCA_TXCTRL_DATA_RRO_EN |
5372 E1000_DCA_TXCTRL_DESC_DCA_EN;
5374 wr32(E1000_DCA_TXCTRL(tx_ring->reg_idx), txctrl);
5377 static void igb_update_rx_dca(struct igb_adapter *adapter,
5378 struct igb_ring *rx_ring,
5381 struct e1000_hw *hw = &adapter->hw;
5382 u32 rxctrl = dca3_get_tag(&adapter->pdev->dev, cpu);
5384 if (hw->mac.type != e1000_82575)
5385 rxctrl <<= E1000_DCA_RXCTRL_CPUID_SHIFT;
5387 /* We can enable relaxed ordering for reads, but not writes when
5388 * DCA is enabled. This is due to a known issue in some chipsets
5389 * which will cause the DCA tag to be cleared.
5391 rxctrl |= E1000_DCA_RXCTRL_DESC_RRO_EN |
5392 E1000_DCA_RXCTRL_DESC_DCA_EN;
5394 wr32(E1000_DCA_RXCTRL(rx_ring->reg_idx), rxctrl);
5397 static void igb_update_dca(struct igb_q_vector *q_vector)
5399 struct igb_adapter *adapter = q_vector->adapter;
5400 int cpu = get_cpu();
5402 if (q_vector->cpu == cpu)
5405 if (q_vector->tx.ring)
5406 igb_update_tx_dca(adapter, q_vector->tx.ring, cpu);
5408 if (q_vector->rx.ring)
5409 igb_update_rx_dca(adapter, q_vector->rx.ring, cpu);
5411 q_vector->cpu = cpu;
5416 static void igb_setup_dca(struct igb_adapter *adapter)
5418 struct e1000_hw *hw = &adapter->hw;
5421 if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
5424 /* Always use CB2 mode, difference is masked in the CB driver. */
5425 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
5427 for (i = 0; i < adapter->num_q_vectors; i++) {
5428 adapter->q_vector[i]->cpu = -1;
5429 igb_update_dca(adapter->q_vector[i]);
5433 static int __igb_notify_dca(struct device *dev, void *data)
5435 struct net_device *netdev = dev_get_drvdata(dev);
5436 struct igb_adapter *adapter = netdev_priv(netdev);
5437 struct pci_dev *pdev = adapter->pdev;
5438 struct e1000_hw *hw = &adapter->hw;
5439 unsigned long event = *(unsigned long *)data;
5442 case DCA_PROVIDER_ADD:
5443 /* if already enabled, don't do it again */
5444 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
5446 if (dca_add_requester(dev) == 0) {
5447 adapter->flags |= IGB_FLAG_DCA_ENABLED;
5448 dev_info(&pdev->dev, "DCA enabled\n");
5449 igb_setup_dca(adapter);
5452 /* Fall Through since DCA is disabled. */
5453 case DCA_PROVIDER_REMOVE:
5454 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
5455 /* without this a class_device is left
5456 * hanging around in the sysfs model
5458 dca_remove_requester(dev);
5459 dev_info(&pdev->dev, "DCA disabled\n");
5460 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
5461 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
5469 static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
5474 ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
5477 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
5479 #endif /* CONFIG_IGB_DCA */
5481 #ifdef CONFIG_PCI_IOV
5482 static int igb_vf_configure(struct igb_adapter *adapter, int vf)
5484 unsigned char mac_addr[ETH_ALEN];
5486 eth_zero_addr(mac_addr);
5487 igb_set_vf_mac(adapter, vf, mac_addr);
5489 /* By default spoof check is enabled for all VFs */
5490 adapter->vf_data[vf].spoofchk_enabled = true;
5496 static void igb_ping_all_vfs(struct igb_adapter *adapter)
5498 struct e1000_hw *hw = &adapter->hw;
5502 for (i = 0 ; i < adapter->vfs_allocated_count; i++) {
5503 ping = E1000_PF_CONTROL_MSG;
5504 if (adapter->vf_data[i].flags & IGB_VF_FLAG_CTS)
5505 ping |= E1000_VT_MSGTYPE_CTS;
5506 igb_write_mbx(hw, &ping, 1, i);
5510 static int igb_set_vf_promisc(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
5512 struct e1000_hw *hw = &adapter->hw;
5513 u32 vmolr = rd32(E1000_VMOLR(vf));
5514 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5516 vf_data->flags &= ~(IGB_VF_FLAG_UNI_PROMISC |
5517 IGB_VF_FLAG_MULTI_PROMISC);
5518 vmolr &= ~(E1000_VMOLR_ROPE | E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
5520 if (*msgbuf & E1000_VF_SET_PROMISC_MULTICAST) {
5521 vmolr |= E1000_VMOLR_MPME;
5522 vf_data->flags |= IGB_VF_FLAG_MULTI_PROMISC;
5523 *msgbuf &= ~E1000_VF_SET_PROMISC_MULTICAST;
5525 /* if we have hashes and we are clearing a multicast promisc
5526 * flag we need to write the hashes to the MTA as this step
5527 * was previously skipped
5529 if (vf_data->num_vf_mc_hashes > 30) {
5530 vmolr |= E1000_VMOLR_MPME;
5531 } else if (vf_data->num_vf_mc_hashes) {
5533 vmolr |= E1000_VMOLR_ROMPE;
5534 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
5535 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
5539 wr32(E1000_VMOLR(vf), vmolr);
5541 /* there are flags left unprocessed, likely not supported */
5542 if (*msgbuf & E1000_VT_MSGINFO_MASK)
5548 static int igb_set_vf_multicasts(struct igb_adapter *adapter,
5549 u32 *msgbuf, u32 vf)
5551 int n = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
5552 u16 *hash_list = (u16 *)&msgbuf[1];
5553 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5556 /* salt away the number of multicast addresses assigned
5557 * to this VF for later use to restore when the PF multi cast
5560 vf_data->num_vf_mc_hashes = n;
5562 /* only up to 30 hash values supported */
5566 /* store the hashes for later use */
5567 for (i = 0; i < n; i++)
5568 vf_data->vf_mc_hashes[i] = hash_list[i];
5570 /* Flush and reset the mta with the new values */
5571 igb_set_rx_mode(adapter->netdev);
5576 static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
5578 struct e1000_hw *hw = &adapter->hw;
5579 struct vf_data_storage *vf_data;
5582 for (i = 0; i < adapter->vfs_allocated_count; i++) {
5583 u32 vmolr = rd32(E1000_VMOLR(i));
5584 vmolr &= ~(E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
5586 vf_data = &adapter->vf_data[i];
5588 if ((vf_data->num_vf_mc_hashes > 30) ||
5589 (vf_data->flags & IGB_VF_FLAG_MULTI_PROMISC)) {
5590 vmolr |= E1000_VMOLR_MPME;
5591 } else if (vf_data->num_vf_mc_hashes) {
5592 vmolr |= E1000_VMOLR_ROMPE;
5593 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
5594 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
5596 wr32(E1000_VMOLR(i), vmolr);
5600 static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
5602 struct e1000_hw *hw = &adapter->hw;
5603 u32 pool_mask, reg, vid;
5606 pool_mask = 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
5608 /* Find the vlan filter for this id */
5609 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5610 reg = rd32(E1000_VLVF(i));
5612 /* remove the vf from the pool */
5615 /* if pool is empty then remove entry from vfta */
5616 if (!(reg & E1000_VLVF_POOLSEL_MASK) &&
5617 (reg & E1000_VLVF_VLANID_ENABLE)) {
5619 vid = reg & E1000_VLVF_VLANID_MASK;
5620 igb_vfta_set(hw, vid, false);
5623 wr32(E1000_VLVF(i), reg);
5626 adapter->vf_data[vf].vlans_enabled = 0;
5629 static s32 igb_vlvf_set(struct igb_adapter *adapter, u32 vid, bool add, u32 vf)
5631 struct e1000_hw *hw = &adapter->hw;
5634 /* The vlvf table only exists on 82576 hardware and newer */
5635 if (hw->mac.type < e1000_82576)
5638 /* we only need to do this if VMDq is enabled */
5639 if (!adapter->vfs_allocated_count)
5642 /* Find the vlan filter for this id */
5643 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5644 reg = rd32(E1000_VLVF(i));
5645 if ((reg & E1000_VLVF_VLANID_ENABLE) &&
5646 vid == (reg & E1000_VLVF_VLANID_MASK))
5651 if (i == E1000_VLVF_ARRAY_SIZE) {
5652 /* Did not find a matching VLAN ID entry that was
5653 * enabled. Search for a free filter entry, i.e.
5654 * one without the enable bit set
5656 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5657 reg = rd32(E1000_VLVF(i));
5658 if (!(reg & E1000_VLVF_VLANID_ENABLE))
5662 if (i < E1000_VLVF_ARRAY_SIZE) {
5663 /* Found an enabled/available entry */
5664 reg |= 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
5666 /* if !enabled we need to set this up in vfta */
5667 if (!(reg & E1000_VLVF_VLANID_ENABLE)) {
5668 /* add VID to filter table */
5669 igb_vfta_set(hw, vid, true);
5670 reg |= E1000_VLVF_VLANID_ENABLE;
5672 reg &= ~E1000_VLVF_VLANID_MASK;
5674 wr32(E1000_VLVF(i), reg);
5676 /* do not modify RLPML for PF devices */
5677 if (vf >= adapter->vfs_allocated_count)
5680 if (!adapter->vf_data[vf].vlans_enabled) {
5682 reg = rd32(E1000_VMOLR(vf));
5683 size = reg & E1000_VMOLR_RLPML_MASK;
5685 reg &= ~E1000_VMOLR_RLPML_MASK;
5687 wr32(E1000_VMOLR(vf), reg);
5690 adapter->vf_data[vf].vlans_enabled++;
5693 if (i < E1000_VLVF_ARRAY_SIZE) {
5694 /* remove vf from the pool */
5695 reg &= ~(1 << (E1000_VLVF_POOLSEL_SHIFT + vf));
5696 /* if pool is empty then remove entry from vfta */
5697 if (!(reg & E1000_VLVF_POOLSEL_MASK)) {
5699 igb_vfta_set(hw, vid, false);
5701 wr32(E1000_VLVF(i), reg);
5703 /* do not modify RLPML for PF devices */
5704 if (vf >= adapter->vfs_allocated_count)
5707 adapter->vf_data[vf].vlans_enabled--;
5708 if (!adapter->vf_data[vf].vlans_enabled) {
5710 reg = rd32(E1000_VMOLR(vf));
5711 size = reg & E1000_VMOLR_RLPML_MASK;
5713 reg &= ~E1000_VMOLR_RLPML_MASK;
5715 wr32(E1000_VMOLR(vf), reg);
5722 static void igb_set_vmvir(struct igb_adapter *adapter, u32 vid, u32 vf)
5724 struct e1000_hw *hw = &adapter->hw;
5727 wr32(E1000_VMVIR(vf), (vid | E1000_VMVIR_VLANA_DEFAULT));
5729 wr32(E1000_VMVIR(vf), 0);
5732 static int igb_ndo_set_vf_vlan(struct net_device *netdev,
5733 int vf, u16 vlan, u8 qos)
5736 struct igb_adapter *adapter = netdev_priv(netdev);
5738 if ((vf >= adapter->vfs_allocated_count) || (vlan > 4095) || (qos > 7))
5741 err = igb_vlvf_set(adapter, vlan, !!vlan, vf);
5744 igb_set_vmvir(adapter, vlan | (qos << VLAN_PRIO_SHIFT), vf);
5745 igb_set_vmolr(adapter, vf, !vlan);
5746 adapter->vf_data[vf].pf_vlan = vlan;
5747 adapter->vf_data[vf].pf_qos = qos;
5748 dev_info(&adapter->pdev->dev,
5749 "Setting VLAN %d, QOS 0x%x on VF %d\n", vlan, qos, vf);
5750 if (test_bit(__IGB_DOWN, &adapter->state)) {
5751 dev_warn(&adapter->pdev->dev,
5752 "The VF VLAN has been set, but the PF device is not up.\n");
5753 dev_warn(&adapter->pdev->dev,
5754 "Bring the PF device up before attempting to use the VF device.\n");
5757 igb_vlvf_set(adapter, adapter->vf_data[vf].pf_vlan,
5759 igb_set_vmvir(adapter, vlan, vf);
5760 igb_set_vmolr(adapter, vf, true);
5761 adapter->vf_data[vf].pf_vlan = 0;
5762 adapter->vf_data[vf].pf_qos = 0;
5768 static int igb_find_vlvf_entry(struct igb_adapter *adapter, int vid)
5770 struct e1000_hw *hw = &adapter->hw;
5774 /* Find the vlan filter for this id */
5775 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5776 reg = rd32(E1000_VLVF(i));
5777 if ((reg & E1000_VLVF_VLANID_ENABLE) &&
5778 vid == (reg & E1000_VLVF_VLANID_MASK))
5782 if (i >= E1000_VLVF_ARRAY_SIZE)
5788 static int igb_set_vf_vlan(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
5790 struct e1000_hw *hw = &adapter->hw;
5791 int add = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
5792 int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
5795 /* If in promiscuous mode we need to make sure the PF also has
5796 * the VLAN filter set.
5798 if (add && (adapter->netdev->flags & IFF_PROMISC))
5799 err = igb_vlvf_set(adapter, vid, add,
5800 adapter->vfs_allocated_count);
5804 err = igb_vlvf_set(adapter, vid, add, vf);
5809 /* Go through all the checks to see if the VLAN filter should
5810 * be wiped completely.
5812 if (!add && (adapter->netdev->flags & IFF_PROMISC)) {
5815 int regndx = igb_find_vlvf_entry(adapter, vid);
5818 /* See if any other pools are set for this VLAN filter
5819 * entry other than the PF.
5821 vlvf = bits = rd32(E1000_VLVF(regndx));
5822 bits &= 1 << (E1000_VLVF_POOLSEL_SHIFT +
5823 adapter->vfs_allocated_count);
5824 /* If the filter was removed then ensure PF pool bit
5825 * is cleared if the PF only added itself to the pool
5826 * because the PF is in promiscuous mode.
5828 if ((vlvf & VLAN_VID_MASK) == vid &&
5829 !test_bit(vid, adapter->active_vlans) &&
5831 igb_vlvf_set(adapter, vid, add,
5832 adapter->vfs_allocated_count);
5839 static inline void igb_vf_reset(struct igb_adapter *adapter, u32 vf)
5841 /* clear flags - except flag that indicates PF has set the MAC */
5842 adapter->vf_data[vf].flags &= IGB_VF_FLAG_PF_SET_MAC;
5843 adapter->vf_data[vf].last_nack = jiffies;
5845 /* reset offloads to defaults */
5846 igb_set_vmolr(adapter, vf, true);
5848 /* reset vlans for device */
5849 igb_clear_vf_vfta(adapter, vf);
5850 if (adapter->vf_data[vf].pf_vlan)
5851 igb_ndo_set_vf_vlan(adapter->netdev, vf,
5852 adapter->vf_data[vf].pf_vlan,
5853 adapter->vf_data[vf].pf_qos);
5855 igb_clear_vf_vfta(adapter, vf);
5857 /* reset multicast table array for vf */
5858 adapter->vf_data[vf].num_vf_mc_hashes = 0;
5860 /* Flush and reset the mta with the new values */
5861 igb_set_rx_mode(adapter->netdev);
5864 static void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
5866 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
5868 /* clear mac address as we were hotplug removed/added */
5869 if (!(adapter->vf_data[vf].flags & IGB_VF_FLAG_PF_SET_MAC))
5870 eth_zero_addr(vf_mac);
5872 /* process remaining reset events */
5873 igb_vf_reset(adapter, vf);
5876 static void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
5878 struct e1000_hw *hw = &adapter->hw;
5879 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
5880 int rar_entry = hw->mac.rar_entry_count - (vf + 1);
5882 u8 *addr = (u8 *)(&msgbuf[1]);
5884 /* process all the same items cleared in a function level reset */
5885 igb_vf_reset(adapter, vf);
5887 /* set vf mac address */
5888 igb_rar_set_qsel(adapter, vf_mac, rar_entry, vf);
5890 /* enable transmit and receive for vf */
5891 reg = rd32(E1000_VFTE);
5892 wr32(E1000_VFTE, reg | (1 << vf));
5893 reg = rd32(E1000_VFRE);
5894 wr32(E1000_VFRE, reg | (1 << vf));
5896 adapter->vf_data[vf].flags |= IGB_VF_FLAG_CTS;
5898 /* reply to reset with ack and vf mac address */
5899 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
5900 memcpy(addr, vf_mac, ETH_ALEN);
5901 igb_write_mbx(hw, msgbuf, 3, vf);
5904 static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf)
5906 /* The VF MAC Address is stored in a packed array of bytes
5907 * starting at the second 32 bit word of the msg array
5909 unsigned char *addr = (char *)&msg[1];
5912 if (is_valid_ether_addr(addr))
5913 err = igb_set_vf_mac(adapter, vf, addr);
5918 static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf)
5920 struct e1000_hw *hw = &adapter->hw;
5921 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5922 u32 msg = E1000_VT_MSGTYPE_NACK;
5924 /* if device isn't clear to send it shouldn't be reading either */
5925 if (!(vf_data->flags & IGB_VF_FLAG_CTS) &&
5926 time_after(jiffies, vf_data->last_nack + (2 * HZ))) {
5927 igb_write_mbx(hw, &msg, 1, vf);
5928 vf_data->last_nack = jiffies;
5932 static void igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
5934 struct pci_dev *pdev = adapter->pdev;
5935 u32 msgbuf[E1000_VFMAILBOX_SIZE];
5936 struct e1000_hw *hw = &adapter->hw;
5937 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5940 retval = igb_read_mbx(hw, msgbuf, E1000_VFMAILBOX_SIZE, vf);
5943 /* if receive failed revoke VF CTS stats and restart init */
5944 dev_err(&pdev->dev, "Error receiving message from VF\n");
5945 vf_data->flags &= ~IGB_VF_FLAG_CTS;
5946 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
5951 /* this is a message we already processed, do nothing */
5952 if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
5955 /* until the vf completes a reset it should not be
5956 * allowed to start any configuration.
5958 if (msgbuf[0] == E1000_VF_RESET) {
5959 igb_vf_reset_msg(adapter, vf);
5963 if (!(vf_data->flags & IGB_VF_FLAG_CTS)) {
5964 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
5970 switch ((msgbuf[0] & 0xFFFF)) {
5971 case E1000_VF_SET_MAC_ADDR:
5973 if (!(vf_data->flags & IGB_VF_FLAG_PF_SET_MAC))
5974 retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
5976 dev_warn(&pdev->dev,
5977 "VF %d attempted to override administratively set MAC address\nReload the VF driver to resume operations\n",
5980 case E1000_VF_SET_PROMISC:
5981 retval = igb_set_vf_promisc(adapter, msgbuf, vf);
5983 case E1000_VF_SET_MULTICAST:
5984 retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
5986 case E1000_VF_SET_LPE:
5987 retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf);
5989 case E1000_VF_SET_VLAN:
5991 if (vf_data->pf_vlan)
5992 dev_warn(&pdev->dev,
5993 "VF %d attempted to override administratively set VLAN tag\nReload the VF driver to resume operations\n",
5996 retval = igb_set_vf_vlan(adapter, msgbuf, vf);
5999 dev_err(&pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
6004 msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
6006 /* notify the VF of the results of what it sent us */
6008 msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
6010 msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
6012 igb_write_mbx(hw, msgbuf, 1, vf);
6015 static void igb_msg_task(struct igb_adapter *adapter)
6017 struct e1000_hw *hw = &adapter->hw;
6020 for (vf = 0; vf < adapter->vfs_allocated_count; vf++) {
6021 /* process any reset requests */
6022 if (!igb_check_for_rst(hw, vf))
6023 igb_vf_reset_event(adapter, vf);
6025 /* process any messages pending */
6026 if (!igb_check_for_msg(hw, vf))
6027 igb_rcv_msg_from_vf(adapter, vf);
6029 /* process any acks */
6030 if (!igb_check_for_ack(hw, vf))
6031 igb_rcv_ack_from_vf(adapter, vf);
6036 * igb_set_uta - Set unicast filter table address
6037 * @adapter: board private structure
6039 * The unicast table address is a register array of 32-bit registers.
6040 * The table is meant to be used in a way similar to how the MTA is used
6041 * however due to certain limitations in the hardware it is necessary to
6042 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
6043 * enable bit to allow vlan tag stripping when promiscuous mode is enabled
6045 static void igb_set_uta(struct igb_adapter *adapter)
6047 struct e1000_hw *hw = &adapter->hw;
6050 /* The UTA table only exists on 82576 hardware and newer */
6051 if (hw->mac.type < e1000_82576)
6054 /* we only need to do this if VMDq is enabled */
6055 if (!adapter->vfs_allocated_count)
6058 for (i = 0; i < hw->mac.uta_reg_count; i++)
6059 array_wr32(E1000_UTA, i, ~0);
6063 * igb_intr_msi - Interrupt Handler
6064 * @irq: interrupt number
6065 * @data: pointer to a network interface device structure
6067 static irqreturn_t igb_intr_msi(int irq, void *data)
6069 struct igb_adapter *adapter = data;
6070 struct igb_q_vector *q_vector = adapter->q_vector[0];
6071 struct e1000_hw *hw = &adapter->hw;
6072 /* read ICR disables interrupts using IAM */
6073 u32 icr = rd32(E1000_ICR);
6075 igb_write_itr(q_vector);
6077 if (icr & E1000_ICR_DRSTA)
6078 schedule_work(&adapter->reset_task);
6080 if (icr & E1000_ICR_DOUTSYNC) {
6081 /* HW is reporting DMA is out of sync */
6082 adapter->stats.doosync++;
6085 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
6086 hw->mac.get_link_status = 1;
6087 if (!test_bit(__IGB_DOWN, &adapter->state))
6088 mod_timer(&adapter->watchdog_timer, jiffies + 1);
6091 if (icr & E1000_ICR_TS) {
6092 u32 tsicr = rd32(E1000_TSICR);
6094 if (tsicr & E1000_TSICR_TXTS) {
6095 /* acknowledge the interrupt */
6096 wr32(E1000_TSICR, E1000_TSICR_TXTS);
6097 /* retrieve hardware timestamp */
6098 schedule_work(&adapter->ptp_tx_work);
6102 napi_schedule(&q_vector->napi);
6108 * igb_intr - Legacy Interrupt Handler
6109 * @irq: interrupt number
6110 * @data: pointer to a network interface device structure
6112 static irqreturn_t igb_intr(int irq, void *data)
6114 struct igb_adapter *adapter = data;
6115 struct igb_q_vector *q_vector = adapter->q_vector[0];
6116 struct e1000_hw *hw = &adapter->hw;
6117 /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked. No
6118 * need for the IMC write
6120 u32 icr = rd32(E1000_ICR);
6122 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
6123 * not set, then the adapter didn't send an interrupt
6125 if (!(icr & E1000_ICR_INT_ASSERTED))
6128 igb_write_itr(q_vector);
6130 if (icr & E1000_ICR_DRSTA)
6131 schedule_work(&adapter->reset_task);
6133 if (icr & E1000_ICR_DOUTSYNC) {
6134 /* HW is reporting DMA is out of sync */
6135 adapter->stats.doosync++;
6138 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
6139 hw->mac.get_link_status = 1;
6140 /* guard against interrupt when we're going down */
6141 if (!test_bit(__IGB_DOWN, &adapter->state))
6142 mod_timer(&adapter->watchdog_timer, jiffies + 1);
6145 if (icr & E1000_ICR_TS) {
6146 u32 tsicr = rd32(E1000_TSICR);
6148 if (tsicr & E1000_TSICR_TXTS) {
6149 /* acknowledge the interrupt */
6150 wr32(E1000_TSICR, E1000_TSICR_TXTS);
6151 /* retrieve hardware timestamp */
6152 schedule_work(&adapter->ptp_tx_work);
6156 napi_schedule(&q_vector->napi);
6161 static void igb_ring_irq_enable(struct igb_q_vector *q_vector)
6163 struct igb_adapter *adapter = q_vector->adapter;
6164 struct e1000_hw *hw = &adapter->hw;
6166 if ((q_vector->rx.ring && (adapter->rx_itr_setting & 3)) ||
6167 (!q_vector->rx.ring && (adapter->tx_itr_setting & 3))) {
6168 if ((adapter->num_q_vectors == 1) && !adapter->vf_data)
6169 igb_set_itr(q_vector);
6171 igb_update_ring_itr(q_vector);
6174 if (!test_bit(__IGB_DOWN, &adapter->state)) {
6175 if (adapter->msix_entries)
6176 wr32(E1000_EIMS, q_vector->eims_value);
6178 igb_irq_enable(adapter);
6183 * igb_poll - NAPI Rx polling callback
6184 * @napi: napi polling structure
6185 * @budget: count of how many packets we should handle
6187 static int igb_poll(struct napi_struct *napi, int budget)
6189 struct igb_q_vector *q_vector = container_of(napi,
6190 struct igb_q_vector,
6192 bool clean_complete = true;
6194 #ifdef CONFIG_IGB_DCA
6195 if (q_vector->adapter->flags & IGB_FLAG_DCA_ENABLED)
6196 igb_update_dca(q_vector);
6198 if (q_vector->tx.ring)
6199 clean_complete = igb_clean_tx_irq(q_vector);
6201 if (q_vector->rx.ring)
6202 clean_complete &= igb_clean_rx_irq(q_vector, budget);
6204 /* If all work not completed, return budget and keep polling */
6205 if (!clean_complete)
6208 /* If not enough Rx work done, exit the polling mode */
6209 napi_complete(napi);
6210 igb_ring_irq_enable(q_vector);
6216 * igb_clean_tx_irq - Reclaim resources after transmit completes
6217 * @q_vector: pointer to q_vector containing needed info
6219 * returns true if ring is completely cleaned
6221 static bool igb_clean_tx_irq(struct igb_q_vector *q_vector)
6223 struct igb_adapter *adapter = q_vector->adapter;
6224 struct igb_ring *tx_ring = q_vector->tx.ring;
6225 struct igb_tx_buffer *tx_buffer;
6226 union e1000_adv_tx_desc *tx_desc;
6227 unsigned int total_bytes = 0, total_packets = 0;
6228 unsigned int budget = q_vector->tx.work_limit;
6229 unsigned int i = tx_ring->next_to_clean;
6231 if (test_bit(__IGB_DOWN, &adapter->state))
6234 tx_buffer = &tx_ring->tx_buffer_info[i];
6235 tx_desc = IGB_TX_DESC(tx_ring, i);
6236 i -= tx_ring->count;
6239 union e1000_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
6241 /* if next_to_watch is not set then there is no work pending */
6245 /* prevent any other reads prior to eop_desc */
6246 read_barrier_depends();
6248 /* if DD is not set pending work has not been completed */
6249 if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)))
6252 /* clear next_to_watch to prevent false hangs */
6253 tx_buffer->next_to_watch = NULL;
6255 /* update the statistics for this packet */
6256 total_bytes += tx_buffer->bytecount;
6257 total_packets += tx_buffer->gso_segs;
6260 dev_kfree_skb_any(tx_buffer->skb);
6262 /* unmap skb header data */
6263 dma_unmap_single(tx_ring->dev,
6264 dma_unmap_addr(tx_buffer, dma),
6265 dma_unmap_len(tx_buffer, len),
6268 /* clear tx_buffer data */
6269 tx_buffer->skb = NULL;
6270 dma_unmap_len_set(tx_buffer, len, 0);
6272 /* clear last DMA location and unmap remaining buffers */
6273 while (tx_desc != eop_desc) {
6278 i -= tx_ring->count;
6279 tx_buffer = tx_ring->tx_buffer_info;
6280 tx_desc = IGB_TX_DESC(tx_ring, 0);
6283 /* unmap any remaining paged data */
6284 if (dma_unmap_len(tx_buffer, len)) {
6285 dma_unmap_page(tx_ring->dev,
6286 dma_unmap_addr(tx_buffer, dma),
6287 dma_unmap_len(tx_buffer, len),
6289 dma_unmap_len_set(tx_buffer, len, 0);
6293 /* move us one more past the eop_desc for start of next pkt */
6298 i -= tx_ring->count;
6299 tx_buffer = tx_ring->tx_buffer_info;
6300 tx_desc = IGB_TX_DESC(tx_ring, 0);
6303 /* issue prefetch for next Tx descriptor */
6306 /* update budget accounting */
6308 } while (likely(budget));
6310 netdev_tx_completed_queue(txring_txq(tx_ring),
6311 total_packets, total_bytes);
6312 i += tx_ring->count;
6313 tx_ring->next_to_clean = i;
6314 u64_stats_update_begin(&tx_ring->tx_syncp);
6315 tx_ring->tx_stats.bytes += total_bytes;
6316 tx_ring->tx_stats.packets += total_packets;
6317 u64_stats_update_end(&tx_ring->tx_syncp);
6318 q_vector->tx.total_bytes += total_bytes;
6319 q_vector->tx.total_packets += total_packets;
6321 if (test_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags)) {
6322 struct e1000_hw *hw = &adapter->hw;
6324 /* Detect a transmit hang in hardware, this serializes the
6325 * check with the clearing of time_stamp and movement of i
6327 clear_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
6328 if (tx_buffer->next_to_watch &&
6329 time_after(jiffies, tx_buffer->time_stamp +
6330 (adapter->tx_timeout_factor * HZ)) &&
6331 !(rd32(E1000_STATUS) & E1000_STATUS_TXOFF)) {
6333 /* detected Tx unit hang */
6334 dev_err(tx_ring->dev,
6335 "Detected Tx Unit Hang\n"
6339 " next_to_use <%x>\n"
6340 " next_to_clean <%x>\n"
6341 "buffer_info[next_to_clean]\n"
6342 " time_stamp <%lx>\n"
6343 " next_to_watch <%p>\n"
6345 " desc.status <%x>\n",
6346 tx_ring->queue_index,
6347 rd32(E1000_TDH(tx_ring->reg_idx)),
6348 readl(tx_ring->tail),
6349 tx_ring->next_to_use,
6350 tx_ring->next_to_clean,
6351 tx_buffer->time_stamp,
6352 tx_buffer->next_to_watch,
6354 tx_buffer->next_to_watch->wb.status);
6355 netif_stop_subqueue(tx_ring->netdev,
6356 tx_ring->queue_index);
6358 /* we are about to reset, no point in enabling stuff */
6363 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
6364 if (unlikely(total_packets &&
6365 netif_carrier_ok(tx_ring->netdev) &&
6366 igb_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD)) {
6367 /* Make sure that anybody stopping the queue after this
6368 * sees the new next_to_clean.
6371 if (__netif_subqueue_stopped(tx_ring->netdev,
6372 tx_ring->queue_index) &&
6373 !(test_bit(__IGB_DOWN, &adapter->state))) {
6374 netif_wake_subqueue(tx_ring->netdev,
6375 tx_ring->queue_index);
6377 u64_stats_update_begin(&tx_ring->tx_syncp);
6378 tx_ring->tx_stats.restart_queue++;
6379 u64_stats_update_end(&tx_ring->tx_syncp);
6387 * igb_reuse_rx_page - page flip buffer and store it back on the ring
6388 * @rx_ring: rx descriptor ring to store buffers on
6389 * @old_buff: donor buffer to have page reused
6391 * Synchronizes page for reuse by the adapter
6393 static void igb_reuse_rx_page(struct igb_ring *rx_ring,
6394 struct igb_rx_buffer *old_buff)
6396 struct igb_rx_buffer *new_buff;
6397 u16 nta = rx_ring->next_to_alloc;
6399 new_buff = &rx_ring->rx_buffer_info[nta];
6401 /* update, and store next to alloc */
6403 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
6405 /* transfer page from old buffer to new buffer */
6406 memcpy(new_buff, old_buff, sizeof(struct igb_rx_buffer));
6408 /* sync the buffer for use by the device */
6409 dma_sync_single_range_for_device(rx_ring->dev, old_buff->dma,
6410 old_buff->page_offset,
6415 static bool igb_can_reuse_rx_page(struct igb_rx_buffer *rx_buffer,
6417 unsigned int truesize)
6419 /* avoid re-using remote pages */
6420 if (unlikely(page_to_nid(page) != numa_node_id()))
6423 #if (PAGE_SIZE < 8192)
6424 /* if we are only owner of page we can reuse it */
6425 if (unlikely(page_count(page) != 1))
6428 /* flip page offset to other buffer */
6429 rx_buffer->page_offset ^= IGB_RX_BUFSZ;
6431 /* since we are the only owner of the page and we need to
6432 * increment it, just set the value to 2 in order to avoid
6433 * an unnecessary locked operation
6435 atomic_set(&page->_count, 2);
6437 /* move offset up to the next cache line */
6438 rx_buffer->page_offset += truesize;
6440 if (rx_buffer->page_offset > (PAGE_SIZE - IGB_RX_BUFSZ))
6443 /* bump ref count on page before it is given to the stack */
6451 * igb_add_rx_frag - Add contents of Rx buffer to sk_buff
6452 * @rx_ring: rx descriptor ring to transact packets on
6453 * @rx_buffer: buffer containing page to add
6454 * @rx_desc: descriptor containing length of buffer written by hardware
6455 * @skb: sk_buff to place the data into
6457 * This function will add the data contained in rx_buffer->page to the skb.
6458 * This is done either through a direct copy if the data in the buffer is
6459 * less than the skb header size, otherwise it will just attach the page as
6460 * a frag to the skb.
6462 * The function will then update the page offset if necessary and return
6463 * true if the buffer can be reused by the adapter.
6465 static bool igb_add_rx_frag(struct igb_ring *rx_ring,
6466 struct igb_rx_buffer *rx_buffer,
6467 union e1000_adv_rx_desc *rx_desc,
6468 struct sk_buff *skb)
6470 struct page *page = rx_buffer->page;
6471 unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
6472 #if (PAGE_SIZE < 8192)
6473 unsigned int truesize = IGB_RX_BUFSZ;
6475 unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
6478 if ((size <= IGB_RX_HDR_LEN) && !skb_is_nonlinear(skb)) {
6479 unsigned char *va = page_address(page) + rx_buffer->page_offset;
6481 if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
6482 igb_ptp_rx_pktstamp(rx_ring->q_vector, va, skb);
6483 va += IGB_TS_HDR_LEN;
6484 size -= IGB_TS_HDR_LEN;
6487 memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
6489 /* we can reuse buffer as-is, just make sure it is local */
6490 if (likely(page_to_nid(page) == numa_node_id()))
6493 /* this page cannot be reused so discard it */
6498 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
6499 rx_buffer->page_offset, size, truesize);
6501 return igb_can_reuse_rx_page(rx_buffer, page, truesize);
6504 static struct sk_buff *igb_fetch_rx_buffer(struct igb_ring *rx_ring,
6505 union e1000_adv_rx_desc *rx_desc,
6506 struct sk_buff *skb)
6508 struct igb_rx_buffer *rx_buffer;
6511 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
6513 page = rx_buffer->page;
6517 void *page_addr = page_address(page) +
6518 rx_buffer->page_offset;
6520 /* prefetch first cache line of first page */
6521 prefetch(page_addr);
6522 #if L1_CACHE_BYTES < 128
6523 prefetch(page_addr + L1_CACHE_BYTES);
6526 /* allocate a skb to store the frags */
6527 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
6529 if (unlikely(!skb)) {
6530 rx_ring->rx_stats.alloc_failed++;
6534 /* we will be copying header into skb->data in
6535 * pskb_may_pull so it is in our interest to prefetch
6536 * it now to avoid a possible cache miss
6538 prefetchw(skb->data);
6541 /* we are reusing so sync this buffer for CPU use */
6542 dma_sync_single_range_for_cpu(rx_ring->dev,
6544 rx_buffer->page_offset,
6548 /* pull page into skb */
6549 if (igb_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
6550 /* hand second half of page back to the ring */
6551 igb_reuse_rx_page(rx_ring, rx_buffer);
6553 /* we are not reusing the buffer so unmap it */
6554 dma_unmap_page(rx_ring->dev, rx_buffer->dma,
6555 PAGE_SIZE, DMA_FROM_DEVICE);
6558 /* clear contents of rx_buffer */
6559 rx_buffer->page = NULL;
6564 static inline void igb_rx_checksum(struct igb_ring *ring,
6565 union e1000_adv_rx_desc *rx_desc,
6566 struct sk_buff *skb)
6568 skb_checksum_none_assert(skb);
6570 /* Ignore Checksum bit is set */
6571 if (igb_test_staterr(rx_desc, E1000_RXD_STAT_IXSM))
6574 /* Rx checksum disabled via ethtool */
6575 if (!(ring->netdev->features & NETIF_F_RXCSUM))
6578 /* TCP/UDP checksum error bit is set */
6579 if (igb_test_staterr(rx_desc,
6580 E1000_RXDEXT_STATERR_TCPE |
6581 E1000_RXDEXT_STATERR_IPE)) {
6582 /* work around errata with sctp packets where the TCPE aka
6583 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
6584 * packets, (aka let the stack check the crc32c)
6586 if (!((skb->len == 60) &&
6587 test_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags))) {
6588 u64_stats_update_begin(&ring->rx_syncp);
6589 ring->rx_stats.csum_err++;
6590 u64_stats_update_end(&ring->rx_syncp);
6592 /* let the stack verify checksum errors */
6595 /* It must be a TCP or UDP packet with a valid checksum */
6596 if (igb_test_staterr(rx_desc, E1000_RXD_STAT_TCPCS |
6597 E1000_RXD_STAT_UDPCS))
6598 skb->ip_summed = CHECKSUM_UNNECESSARY;
6600 dev_dbg(ring->dev, "cksum success: bits %08X\n",
6601 le32_to_cpu(rx_desc->wb.upper.status_error));
6604 static inline void igb_rx_hash(struct igb_ring *ring,
6605 union e1000_adv_rx_desc *rx_desc,
6606 struct sk_buff *skb)
6608 if (ring->netdev->features & NETIF_F_RXHASH)
6609 skb->rxhash = le32_to_cpu(rx_desc->wb.lower.hi_dword.rss);
6613 * igb_is_non_eop - process handling of non-EOP buffers
6614 * @rx_ring: Rx ring being processed
6615 * @rx_desc: Rx descriptor for current buffer
6616 * @skb: current socket buffer containing buffer in progress
6618 * This function updates next to clean. If the buffer is an EOP buffer
6619 * this function exits returning false, otherwise it will place the
6620 * sk_buff in the next buffer to be chained and return true indicating
6621 * that this is in fact a non-EOP buffer.
6623 static bool igb_is_non_eop(struct igb_ring *rx_ring,
6624 union e1000_adv_rx_desc *rx_desc)
6626 u32 ntc = rx_ring->next_to_clean + 1;
6628 /* fetch, update, and store next to clean */
6629 ntc = (ntc < rx_ring->count) ? ntc : 0;
6630 rx_ring->next_to_clean = ntc;
6632 prefetch(IGB_RX_DESC(rx_ring, ntc));
6634 if (likely(igb_test_staterr(rx_desc, E1000_RXD_STAT_EOP)))
6641 * igb_get_headlen - determine size of header for LRO/GRO
6642 * @data: pointer to the start of the headers
6643 * @max_len: total length of section to find headers in
6645 * This function is meant to determine the length of headers that will
6646 * be recognized by hardware for LRO, and GRO offloads. The main
6647 * motivation of doing this is to only perform one pull for IPv4 TCP
6648 * packets so that we can do basic things like calculating the gso_size
6649 * based on the average data per packet.
6651 static unsigned int igb_get_headlen(unsigned char *data,
6652 unsigned int max_len)
6655 unsigned char *network;
6658 struct vlan_hdr *vlan;
6661 struct ipv6hdr *ipv6;
6664 u8 nexthdr = 0; /* default to not TCP */
6667 /* this should never happen, but better safe than sorry */
6668 if (max_len < ETH_HLEN)
6671 /* initialize network frame pointer */
6674 /* set first protocol and move network header forward */
6675 protocol = hdr.eth->h_proto;
6676 hdr.network += ETH_HLEN;
6678 /* handle any vlan tag if present */
6679 if (protocol == __constant_htons(ETH_P_8021Q)) {
6680 if ((hdr.network - data) > (max_len - VLAN_HLEN))
6683 protocol = hdr.vlan->h_vlan_encapsulated_proto;
6684 hdr.network += VLAN_HLEN;
6687 /* handle L3 protocols */
6688 if (protocol == __constant_htons(ETH_P_IP)) {
6689 if ((hdr.network - data) > (max_len - sizeof(struct iphdr)))
6692 /* access ihl as a u8 to avoid unaligned access on ia64 */
6693 hlen = (hdr.network[0] & 0x0F) << 2;
6695 /* verify hlen meets minimum size requirements */
6696 if (hlen < sizeof(struct iphdr))
6697 return hdr.network - data;
6699 /* record next protocol if header is present */
6700 if (!(hdr.ipv4->frag_off & htons(IP_OFFSET)))
6701 nexthdr = hdr.ipv4->protocol;
6702 } else if (protocol == __constant_htons(ETH_P_IPV6)) {
6703 if ((hdr.network - data) > (max_len - sizeof(struct ipv6hdr)))
6706 /* record next protocol */
6707 nexthdr = hdr.ipv6->nexthdr;
6708 hlen = sizeof(struct ipv6hdr);
6710 return hdr.network - data;
6713 /* relocate pointer to start of L4 header */
6714 hdr.network += hlen;
6716 /* finally sort out TCP */
6717 if (nexthdr == IPPROTO_TCP) {
6718 if ((hdr.network - data) > (max_len - sizeof(struct tcphdr)))
6721 /* access doff as a u8 to avoid unaligned access on ia64 */
6722 hlen = (hdr.network[12] & 0xF0) >> 2;
6724 /* verify hlen meets minimum size requirements */
6725 if (hlen < sizeof(struct tcphdr))
6726 return hdr.network - data;
6728 hdr.network += hlen;
6729 } else if (nexthdr == IPPROTO_UDP) {
6730 if ((hdr.network - data) > (max_len - sizeof(struct udphdr)))
6733 hdr.network += sizeof(struct udphdr);
6736 /* If everything has gone correctly hdr.network should be the
6737 * data section of the packet and will be the end of the header.
6738 * If not then it probably represents the end of the last recognized
6741 if ((hdr.network - data) < max_len)
6742 return hdr.network - data;
6748 * igb_pull_tail - igb specific version of skb_pull_tail
6749 * @rx_ring: rx descriptor ring packet is being transacted on
6750 * @rx_desc: pointer to the EOP Rx descriptor
6751 * @skb: pointer to current skb being adjusted
6753 * This function is an igb specific version of __pskb_pull_tail. The
6754 * main difference between this version and the original function is that
6755 * this function can make several assumptions about the state of things
6756 * that allow for significant optimizations versus the standard function.
6757 * As a result we can do things like drop a frag and maintain an accurate
6758 * truesize for the skb.
6760 static void igb_pull_tail(struct igb_ring *rx_ring,
6761 union e1000_adv_rx_desc *rx_desc,
6762 struct sk_buff *skb)
6764 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
6766 unsigned int pull_len;
6768 /* it is valid to use page_address instead of kmap since we are
6769 * working with pages allocated out of the lomem pool per
6770 * alloc_page(GFP_ATOMIC)
6772 va = skb_frag_address(frag);
6774 if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
6775 /* retrieve timestamp from buffer */
6776 igb_ptp_rx_pktstamp(rx_ring->q_vector, va, skb);
6778 /* update pointers to remove timestamp header */
6779 skb_frag_size_sub(frag, IGB_TS_HDR_LEN);
6780 frag->page_offset += IGB_TS_HDR_LEN;
6781 skb->data_len -= IGB_TS_HDR_LEN;
6782 skb->len -= IGB_TS_HDR_LEN;
6784 /* move va to start of packet data */
6785 va += IGB_TS_HDR_LEN;
6788 /* we need the header to contain the greater of either ETH_HLEN or
6789 * 60 bytes if the skb->len is less than 60 for skb_pad.
6791 pull_len = igb_get_headlen(va, IGB_RX_HDR_LEN);
6793 /* align pull length to size of long to optimize memcpy performance */
6794 skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
6796 /* update all of the pointers */
6797 skb_frag_size_sub(frag, pull_len);
6798 frag->page_offset += pull_len;
6799 skb->data_len -= pull_len;
6800 skb->tail += pull_len;
6804 * igb_cleanup_headers - Correct corrupted or empty headers
6805 * @rx_ring: rx descriptor ring packet is being transacted on
6806 * @rx_desc: pointer to the EOP Rx descriptor
6807 * @skb: pointer to current skb being fixed
6809 * Address the case where we are pulling data in on pages only
6810 * and as such no data is present in the skb header.
6812 * In addition if skb is not at least 60 bytes we need to pad it so that
6813 * it is large enough to qualify as a valid Ethernet frame.
6815 * Returns true if an error was encountered and skb was freed.
6817 static bool igb_cleanup_headers(struct igb_ring *rx_ring,
6818 union e1000_adv_rx_desc *rx_desc,
6819 struct sk_buff *skb)
6821 if (unlikely((igb_test_staterr(rx_desc,
6822 E1000_RXDEXT_ERR_FRAME_ERR_MASK)))) {
6823 struct net_device *netdev = rx_ring->netdev;
6824 if (!(netdev->features & NETIF_F_RXALL)) {
6825 dev_kfree_skb_any(skb);
6830 /* place header in linear portion of buffer */
6831 if (skb_is_nonlinear(skb))
6832 igb_pull_tail(rx_ring, rx_desc, skb);
6834 /* if skb_pad returns an error the skb was freed */
6835 if (unlikely(skb->len < 60)) {
6836 int pad_len = 60 - skb->len;
6838 if (skb_pad(skb, pad_len))
6840 __skb_put(skb, pad_len);
6847 * igb_process_skb_fields - Populate skb header fields from Rx descriptor
6848 * @rx_ring: rx descriptor ring packet is being transacted on
6849 * @rx_desc: pointer to the EOP Rx descriptor
6850 * @skb: pointer to current skb being populated
6852 * This function checks the ring, descriptor, and packet information in
6853 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
6854 * other fields within the skb.
6856 static void igb_process_skb_fields(struct igb_ring *rx_ring,
6857 union e1000_adv_rx_desc *rx_desc,
6858 struct sk_buff *skb)
6860 struct net_device *dev = rx_ring->netdev;
6862 igb_rx_hash(rx_ring, rx_desc, skb);
6864 igb_rx_checksum(rx_ring, rx_desc, skb);
6866 igb_ptp_rx_hwtstamp(rx_ring, rx_desc, skb);
6868 if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
6869 igb_test_staterr(rx_desc, E1000_RXD_STAT_VP)) {
6871 if (igb_test_staterr(rx_desc, E1000_RXDEXT_STATERR_LB) &&
6872 test_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &rx_ring->flags))
6873 vid = be16_to_cpu(rx_desc->wb.upper.vlan);
6875 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
6877 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
6880 skb_record_rx_queue(skb, rx_ring->queue_index);
6882 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
6885 static bool igb_clean_rx_irq(struct igb_q_vector *q_vector, const int budget)
6887 struct igb_ring *rx_ring = q_vector->rx.ring;
6888 struct sk_buff *skb = rx_ring->skb;
6889 unsigned int total_bytes = 0, total_packets = 0;
6890 u16 cleaned_count = igb_desc_unused(rx_ring);
6893 union e1000_adv_rx_desc *rx_desc;
6895 /* return some buffers to hardware, one at a time is too slow */
6896 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
6897 igb_alloc_rx_buffers(rx_ring, cleaned_count);
6901 rx_desc = IGB_RX_DESC(rx_ring, rx_ring->next_to_clean);
6903 if (!igb_test_staterr(rx_desc, E1000_RXD_STAT_DD))
6906 /* This memory barrier is needed to keep us from reading
6907 * any other fields out of the rx_desc until we know the
6908 * RXD_STAT_DD bit is set
6912 /* retrieve a buffer from the ring */
6913 skb = igb_fetch_rx_buffer(rx_ring, rx_desc, skb);
6915 /* exit if we failed to retrieve a buffer */
6921 /* fetch next buffer in frame if non-eop */
6922 if (igb_is_non_eop(rx_ring, rx_desc))
6925 /* verify the packet layout is correct */
6926 if (igb_cleanup_headers(rx_ring, rx_desc, skb)) {
6931 /* probably a little skewed due to removing CRC */
6932 total_bytes += skb->len;
6934 /* populate checksum, timestamp, VLAN, and protocol */
6935 igb_process_skb_fields(rx_ring, rx_desc, skb);
6937 napi_gro_receive(&q_vector->napi, skb);
6939 /* reset skb pointer */
6942 /* update budget accounting */
6944 } while (likely(total_packets < budget));
6946 /* place incomplete frames back on ring for completion */
6949 u64_stats_update_begin(&rx_ring->rx_syncp);
6950 rx_ring->rx_stats.packets += total_packets;
6951 rx_ring->rx_stats.bytes += total_bytes;
6952 u64_stats_update_end(&rx_ring->rx_syncp);
6953 q_vector->rx.total_packets += total_packets;
6954 q_vector->rx.total_bytes += total_bytes;
6957 igb_alloc_rx_buffers(rx_ring, cleaned_count);
6959 return (total_packets < budget);
6962 static bool igb_alloc_mapped_page(struct igb_ring *rx_ring,
6963 struct igb_rx_buffer *bi)
6965 struct page *page = bi->page;
6968 /* since we are recycling buffers we should seldom need to alloc */
6972 /* alloc new page for storage */
6973 page = __skb_alloc_page(GFP_ATOMIC | __GFP_COLD, NULL);
6974 if (unlikely(!page)) {
6975 rx_ring->rx_stats.alloc_failed++;
6979 /* map page for use */
6980 dma = dma_map_page(rx_ring->dev, page, 0, PAGE_SIZE, DMA_FROM_DEVICE);
6982 /* if mapping failed free memory back to system since
6983 * there isn't much point in holding memory we can't use
6985 if (dma_mapping_error(rx_ring->dev, dma)) {
6988 rx_ring->rx_stats.alloc_failed++;
6994 bi->page_offset = 0;
7000 * igb_alloc_rx_buffers - Replace used receive buffers; packet split
7001 * @adapter: address of board private structure
7003 void igb_alloc_rx_buffers(struct igb_ring *rx_ring, u16 cleaned_count)
7005 union e1000_adv_rx_desc *rx_desc;
7006 struct igb_rx_buffer *bi;
7007 u16 i = rx_ring->next_to_use;
7013 rx_desc = IGB_RX_DESC(rx_ring, i);
7014 bi = &rx_ring->rx_buffer_info[i];
7015 i -= rx_ring->count;
7018 if (!igb_alloc_mapped_page(rx_ring, bi))
7021 /* Refresh the desc even if buffer_addrs didn't change
7022 * because each write-back erases this info.
7024 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
7030 rx_desc = IGB_RX_DESC(rx_ring, 0);
7031 bi = rx_ring->rx_buffer_info;
7032 i -= rx_ring->count;
7035 /* clear the hdr_addr for the next_to_use descriptor */
7036 rx_desc->read.hdr_addr = 0;
7039 } while (cleaned_count);
7041 i += rx_ring->count;
7043 if (rx_ring->next_to_use != i) {
7044 /* record the next descriptor to use */
7045 rx_ring->next_to_use = i;
7047 /* update next to alloc since we have filled the ring */
7048 rx_ring->next_to_alloc = i;
7050 /* Force memory writes to complete before letting h/w
7051 * know there are new descriptors to fetch. (Only
7052 * applicable for weak-ordered memory model archs,
7056 writel(i, rx_ring->tail);
7066 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
7068 struct igb_adapter *adapter = netdev_priv(netdev);
7069 struct mii_ioctl_data *data = if_mii(ifr);
7071 if (adapter->hw.phy.media_type != e1000_media_type_copper)
7076 data->phy_id = adapter->hw.phy.addr;
7079 if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
7096 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
7102 return igb_mii_ioctl(netdev, ifr, cmd);
7104 return igb_ptp_hwtstamp_ioctl(netdev, ifr, cmd);
7110 s32 igb_read_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
7112 struct igb_adapter *adapter = hw->back;
7114 if (pcie_capability_read_word(adapter->pdev, reg, value))
7115 return -E1000_ERR_CONFIG;
7120 s32 igb_write_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
7122 struct igb_adapter *adapter = hw->back;
7124 if (pcie_capability_write_word(adapter->pdev, reg, *value))
7125 return -E1000_ERR_CONFIG;
7130 static void igb_vlan_mode(struct net_device *netdev, netdev_features_t features)
7132 struct igb_adapter *adapter = netdev_priv(netdev);
7133 struct e1000_hw *hw = &adapter->hw;
7135 bool enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
7138 /* enable VLAN tag insert/strip */
7139 ctrl = rd32(E1000_CTRL);
7140 ctrl |= E1000_CTRL_VME;
7141 wr32(E1000_CTRL, ctrl);
7143 /* Disable CFI check */
7144 rctl = rd32(E1000_RCTL);
7145 rctl &= ~E1000_RCTL_CFIEN;
7146 wr32(E1000_RCTL, rctl);
7148 /* disable VLAN tag insert/strip */
7149 ctrl = rd32(E1000_CTRL);
7150 ctrl &= ~E1000_CTRL_VME;
7151 wr32(E1000_CTRL, ctrl);
7154 igb_rlpml_set(adapter);
7157 static int igb_vlan_rx_add_vid(struct net_device *netdev,
7158 __be16 proto, u16 vid)
7160 struct igb_adapter *adapter = netdev_priv(netdev);
7161 struct e1000_hw *hw = &adapter->hw;
7162 int pf_id = adapter->vfs_allocated_count;
7164 /* attempt to add filter to vlvf array */
7165 igb_vlvf_set(adapter, vid, true, pf_id);
7167 /* add the filter since PF can receive vlans w/o entry in vlvf */
7168 igb_vfta_set(hw, vid, true);
7170 set_bit(vid, adapter->active_vlans);
7175 static int igb_vlan_rx_kill_vid(struct net_device *netdev,
7176 __be16 proto, u16 vid)
7178 struct igb_adapter *adapter = netdev_priv(netdev);
7179 struct e1000_hw *hw = &adapter->hw;
7180 int pf_id = adapter->vfs_allocated_count;
7183 /* remove vlan from VLVF table array */
7184 err = igb_vlvf_set(adapter, vid, false, pf_id);
7186 /* if vid was not present in VLVF just remove it from table */
7188 igb_vfta_set(hw, vid, false);
7190 clear_bit(vid, adapter->active_vlans);
7195 static void igb_restore_vlan(struct igb_adapter *adapter)
7199 igb_vlan_mode(adapter->netdev, adapter->netdev->features);
7201 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
7202 igb_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
7205 int igb_set_spd_dplx(struct igb_adapter *adapter, u32 spd, u8 dplx)
7207 struct pci_dev *pdev = adapter->pdev;
7208 struct e1000_mac_info *mac = &adapter->hw.mac;
7212 /* Make sure dplx is at most 1 bit and lsb of speed is not set
7213 * for the switch() below to work
7215 if ((spd & 1) || (dplx & ~1))
7218 /* Fiber NIC's only allow 1000 gbps Full duplex
7219 * and 100Mbps Full duplex for 100baseFx sfp
7221 if (adapter->hw.phy.media_type == e1000_media_type_internal_serdes) {
7222 switch (spd + dplx) {
7223 case SPEED_10 + DUPLEX_HALF:
7224 case SPEED_10 + DUPLEX_FULL:
7225 case SPEED_100 + DUPLEX_HALF:
7232 switch (spd + dplx) {
7233 case SPEED_10 + DUPLEX_HALF:
7234 mac->forced_speed_duplex = ADVERTISE_10_HALF;
7236 case SPEED_10 + DUPLEX_FULL:
7237 mac->forced_speed_duplex = ADVERTISE_10_FULL;
7239 case SPEED_100 + DUPLEX_HALF:
7240 mac->forced_speed_duplex = ADVERTISE_100_HALF;
7242 case SPEED_100 + DUPLEX_FULL:
7243 mac->forced_speed_duplex = ADVERTISE_100_FULL;
7245 case SPEED_1000 + DUPLEX_FULL:
7247 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
7249 case SPEED_1000 + DUPLEX_HALF: /* not supported */
7254 /* clear MDI, MDI(-X) override is only allowed when autoneg enabled */
7255 adapter->hw.phy.mdix = AUTO_ALL_MODES;
7260 dev_err(&pdev->dev, "Unsupported Speed/Duplex configuration\n");
7264 static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake,
7267 struct net_device *netdev = pci_get_drvdata(pdev);
7268 struct igb_adapter *adapter = netdev_priv(netdev);
7269 struct e1000_hw *hw = &adapter->hw;
7270 u32 ctrl, rctl, status;
7271 u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
7276 netif_device_detach(netdev);
7278 if (netif_running(netdev))
7279 __igb_close(netdev, true);
7281 igb_clear_interrupt_scheme(adapter);
7284 retval = pci_save_state(pdev);
7289 status = rd32(E1000_STATUS);
7290 if (status & E1000_STATUS_LU)
7291 wufc &= ~E1000_WUFC_LNKC;
7294 igb_setup_rctl(adapter);
7295 igb_set_rx_mode(netdev);
7297 /* turn on all-multi mode if wake on multicast is enabled */
7298 if (wufc & E1000_WUFC_MC) {
7299 rctl = rd32(E1000_RCTL);
7300 rctl |= E1000_RCTL_MPE;
7301 wr32(E1000_RCTL, rctl);
7304 ctrl = rd32(E1000_CTRL);
7305 /* advertise wake from D3Cold */
7306 #define E1000_CTRL_ADVD3WUC 0x00100000
7307 /* phy power management enable */
7308 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
7309 ctrl |= E1000_CTRL_ADVD3WUC;
7310 wr32(E1000_CTRL, ctrl);
7312 /* Allow time for pending master requests to run */
7313 igb_disable_pcie_master(hw);
7315 wr32(E1000_WUC, E1000_WUC_PME_EN);
7316 wr32(E1000_WUFC, wufc);
7319 wr32(E1000_WUFC, 0);
7322 *enable_wake = wufc || adapter->en_mng_pt;
7324 igb_power_down_link(adapter);
7326 igb_power_up_link(adapter);
7328 /* Release control of h/w to f/w. If f/w is AMT enabled, this
7329 * would have already happened in close and is redundant.
7331 igb_release_hw_control(adapter);
7333 pci_disable_device(pdev);
7339 #ifdef CONFIG_PM_SLEEP
7340 static int igb_suspend(struct device *dev)
7344 struct pci_dev *pdev = to_pci_dev(dev);
7346 retval = __igb_shutdown(pdev, &wake, 0);
7351 pci_prepare_to_sleep(pdev);
7353 pci_wake_from_d3(pdev, false);
7354 pci_set_power_state(pdev, PCI_D3hot);
7359 #endif /* CONFIG_PM_SLEEP */
7361 static int igb_resume(struct device *dev)
7363 struct pci_dev *pdev = to_pci_dev(dev);
7364 struct net_device *netdev = pci_get_drvdata(pdev);
7365 struct igb_adapter *adapter = netdev_priv(netdev);
7366 struct e1000_hw *hw = &adapter->hw;
7369 pci_set_power_state(pdev, PCI_D0);
7370 pci_restore_state(pdev);
7371 pci_save_state(pdev);
7373 err = pci_enable_device_mem(pdev);
7376 "igb: Cannot enable PCI device from suspend\n");
7379 pci_set_master(pdev);
7381 pci_enable_wake(pdev, PCI_D3hot, 0);
7382 pci_enable_wake(pdev, PCI_D3cold, 0);
7384 if (igb_init_interrupt_scheme(adapter, true)) {
7385 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
7391 /* let the f/w know that the h/w is now under the control of the
7394 igb_get_hw_control(adapter);
7396 wr32(E1000_WUS, ~0);
7398 if (netdev->flags & IFF_UP) {
7400 err = __igb_open(netdev, true);
7406 netif_device_attach(netdev);
7410 #ifdef CONFIG_PM_RUNTIME
7411 static int igb_runtime_idle(struct device *dev)
7413 struct pci_dev *pdev = to_pci_dev(dev);
7414 struct net_device *netdev = pci_get_drvdata(pdev);
7415 struct igb_adapter *adapter = netdev_priv(netdev);
7417 if (!igb_has_link(adapter))
7418 pm_schedule_suspend(dev, MSEC_PER_SEC * 5);
7423 static int igb_runtime_suspend(struct device *dev)
7425 struct pci_dev *pdev = to_pci_dev(dev);
7429 retval = __igb_shutdown(pdev, &wake, 1);
7434 pci_prepare_to_sleep(pdev);
7436 pci_wake_from_d3(pdev, false);
7437 pci_set_power_state(pdev, PCI_D3hot);
7443 static int igb_runtime_resume(struct device *dev)
7445 return igb_resume(dev);
7447 #endif /* CONFIG_PM_RUNTIME */
7450 static void igb_shutdown(struct pci_dev *pdev)
7454 __igb_shutdown(pdev, &wake, 0);
7456 if (system_state == SYSTEM_POWER_OFF) {
7457 pci_wake_from_d3(pdev, wake);
7458 pci_set_power_state(pdev, PCI_D3hot);
7462 #ifdef CONFIG_PCI_IOV
7463 static int igb_sriov_reinit(struct pci_dev *dev)
7465 struct net_device *netdev = pci_get_drvdata(dev);
7466 struct igb_adapter *adapter = netdev_priv(netdev);
7467 struct pci_dev *pdev = adapter->pdev;
7471 if (netif_running(netdev))
7474 igb_clear_interrupt_scheme(adapter);
7476 igb_init_queue_configuration(adapter);
7478 if (igb_init_interrupt_scheme(adapter, true)) {
7479 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
7483 if (netif_running(netdev))
7491 static int igb_pci_disable_sriov(struct pci_dev *dev)
7493 int err = igb_disable_sriov(dev);
7496 err = igb_sriov_reinit(dev);
7501 static int igb_pci_enable_sriov(struct pci_dev *dev, int num_vfs)
7503 int err = igb_enable_sriov(dev, num_vfs);
7508 err = igb_sriov_reinit(dev);
7517 static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs)
7519 #ifdef CONFIG_PCI_IOV
7521 return igb_pci_disable_sriov(dev);
7523 return igb_pci_enable_sriov(dev, num_vfs);
7528 #ifdef CONFIG_NET_POLL_CONTROLLER
7529 /* Polling 'interrupt' - used by things like netconsole to send skbs
7530 * without having to re-enable interrupts. It's not called while
7531 * the interrupt routine is executing.
7533 static void igb_netpoll(struct net_device *netdev)
7535 struct igb_adapter *adapter = netdev_priv(netdev);
7536 struct e1000_hw *hw = &adapter->hw;
7537 struct igb_q_vector *q_vector;
7540 for (i = 0; i < adapter->num_q_vectors; i++) {
7541 q_vector = adapter->q_vector[i];
7542 if (adapter->msix_entries)
7543 wr32(E1000_EIMC, q_vector->eims_value);
7545 igb_irq_disable(adapter);
7546 napi_schedule(&q_vector->napi);
7549 #endif /* CONFIG_NET_POLL_CONTROLLER */
7552 * igb_io_error_detected - called when PCI error is detected
7553 * @pdev: Pointer to PCI device
7554 * @state: The current pci connection state
7556 * This function is called after a PCI bus error affecting
7557 * this device has been detected.
7559 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
7560 pci_channel_state_t state)
7562 struct net_device *netdev = pci_get_drvdata(pdev);
7563 struct igb_adapter *adapter = netdev_priv(netdev);
7565 netif_device_detach(netdev);
7567 if (state == pci_channel_io_perm_failure)
7568 return PCI_ERS_RESULT_DISCONNECT;
7570 if (netif_running(netdev))
7572 pci_disable_device(pdev);
7574 /* Request a slot slot reset. */
7575 return PCI_ERS_RESULT_NEED_RESET;
7579 * igb_io_slot_reset - called after the pci bus has been reset.
7580 * @pdev: Pointer to PCI device
7582 * Restart the card from scratch, as if from a cold-boot. Implementation
7583 * resembles the first-half of the igb_resume routine.
7585 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
7587 struct net_device *netdev = pci_get_drvdata(pdev);
7588 struct igb_adapter *adapter = netdev_priv(netdev);
7589 struct e1000_hw *hw = &adapter->hw;
7590 pci_ers_result_t result;
7593 if (pci_enable_device_mem(pdev)) {
7595 "Cannot re-enable PCI device after reset.\n");
7596 result = PCI_ERS_RESULT_DISCONNECT;
7598 pci_set_master(pdev);
7599 pci_restore_state(pdev);
7600 pci_save_state(pdev);
7602 pci_enable_wake(pdev, PCI_D3hot, 0);
7603 pci_enable_wake(pdev, PCI_D3cold, 0);
7606 wr32(E1000_WUS, ~0);
7607 result = PCI_ERS_RESULT_RECOVERED;
7610 err = pci_cleanup_aer_uncorrect_error_status(pdev);
7613 "pci_cleanup_aer_uncorrect_error_status failed 0x%0x\n",
7615 /* non-fatal, continue */
7622 * igb_io_resume - called when traffic can start flowing again.
7623 * @pdev: Pointer to PCI device
7625 * This callback is called when the error recovery driver tells us that
7626 * its OK to resume normal operation. Implementation resembles the
7627 * second-half of the igb_resume routine.
7629 static void igb_io_resume(struct pci_dev *pdev)
7631 struct net_device *netdev = pci_get_drvdata(pdev);
7632 struct igb_adapter *adapter = netdev_priv(netdev);
7634 if (netif_running(netdev)) {
7635 if (igb_up(adapter)) {
7636 dev_err(&pdev->dev, "igb_up failed after reset\n");
7641 netif_device_attach(netdev);
7643 /* let the f/w know that the h/w is now under the control of the
7646 igb_get_hw_control(adapter);
7649 static void igb_rar_set_qsel(struct igb_adapter *adapter, u8 *addr, u32 index,
7652 u32 rar_low, rar_high;
7653 struct e1000_hw *hw = &adapter->hw;
7655 /* HW expects these in little endian so we reverse the byte order
7656 * from network order (big endian) to little endian
7658 rar_low = ((u32) addr[0] | ((u32) addr[1] << 8) |
7659 ((u32) addr[2] << 16) | ((u32) addr[3] << 24));
7660 rar_high = ((u32) addr[4] | ((u32) addr[5] << 8));
7662 /* Indicate to hardware the Address is Valid. */
7663 rar_high |= E1000_RAH_AV;
7665 if (hw->mac.type == e1000_82575)
7666 rar_high |= E1000_RAH_POOL_1 * qsel;
7668 rar_high |= E1000_RAH_POOL_1 << qsel;
7670 wr32(E1000_RAL(index), rar_low);
7672 wr32(E1000_RAH(index), rar_high);
7676 static int igb_set_vf_mac(struct igb_adapter *adapter,
7677 int vf, unsigned char *mac_addr)
7679 struct e1000_hw *hw = &adapter->hw;
7680 /* VF MAC addresses start at end of receive addresses and moves
7681 * towards the first, as a result a collision should not be possible
7683 int rar_entry = hw->mac.rar_entry_count - (vf + 1);
7685 memcpy(adapter->vf_data[vf].vf_mac_addresses, mac_addr, ETH_ALEN);
7687 igb_rar_set_qsel(adapter, mac_addr, rar_entry, vf);
7692 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
7694 struct igb_adapter *adapter = netdev_priv(netdev);
7695 if (!is_valid_ether_addr(mac) || (vf >= adapter->vfs_allocated_count))
7697 adapter->vf_data[vf].flags |= IGB_VF_FLAG_PF_SET_MAC;
7698 dev_info(&adapter->pdev->dev, "setting MAC %pM on VF %d\n", mac, vf);
7699 dev_info(&adapter->pdev->dev,
7700 "Reload the VF driver to make this change effective.");
7701 if (test_bit(__IGB_DOWN, &adapter->state)) {
7702 dev_warn(&adapter->pdev->dev,
7703 "The VF MAC address has been set, but the PF device is not up.\n");
7704 dev_warn(&adapter->pdev->dev,
7705 "Bring the PF device up before attempting to use the VF device.\n");
7707 return igb_set_vf_mac(adapter, vf, mac);
7710 static int igb_link_mbps(int internal_link_speed)
7712 switch (internal_link_speed) {
7722 static void igb_set_vf_rate_limit(struct e1000_hw *hw, int vf, int tx_rate,
7729 /* Calculate the rate factor values to set */
7730 rf_int = link_speed / tx_rate;
7731 rf_dec = (link_speed - (rf_int * tx_rate));
7732 rf_dec = (rf_dec * (1 << E1000_RTTBCNRC_RF_INT_SHIFT)) /
7735 bcnrc_val = E1000_RTTBCNRC_RS_ENA;
7736 bcnrc_val |= ((rf_int << E1000_RTTBCNRC_RF_INT_SHIFT) &
7737 E1000_RTTBCNRC_RF_INT_MASK);
7738 bcnrc_val |= (rf_dec & E1000_RTTBCNRC_RF_DEC_MASK);
7743 wr32(E1000_RTTDQSEL, vf); /* vf X uses queue X */
7744 /* Set global transmit compensation time to the MMW_SIZE in RTTBCNRM
7745 * register. MMW_SIZE=0x014 if 9728-byte jumbo is supported.
7747 wr32(E1000_RTTBCNRM, 0x14);
7748 wr32(E1000_RTTBCNRC, bcnrc_val);
7751 static void igb_check_vf_rate_limit(struct igb_adapter *adapter)
7753 int actual_link_speed, i;
7754 bool reset_rate = false;
7756 /* VF TX rate limit was not set or not supported */
7757 if ((adapter->vf_rate_link_speed == 0) ||
7758 (adapter->hw.mac.type != e1000_82576))
7761 actual_link_speed = igb_link_mbps(adapter->link_speed);
7762 if (actual_link_speed != adapter->vf_rate_link_speed) {
7764 adapter->vf_rate_link_speed = 0;
7765 dev_info(&adapter->pdev->dev,
7766 "Link speed has been changed. VF Transmit rate is disabled\n");
7769 for (i = 0; i < adapter->vfs_allocated_count; i++) {
7771 adapter->vf_data[i].tx_rate = 0;
7773 igb_set_vf_rate_limit(&adapter->hw, i,
7774 adapter->vf_data[i].tx_rate,
7779 static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf, int tx_rate)
7781 struct igb_adapter *adapter = netdev_priv(netdev);
7782 struct e1000_hw *hw = &adapter->hw;
7783 int actual_link_speed;
7785 if (hw->mac.type != e1000_82576)
7788 actual_link_speed = igb_link_mbps(adapter->link_speed);
7789 if ((vf >= adapter->vfs_allocated_count) ||
7790 (!(rd32(E1000_STATUS) & E1000_STATUS_LU)) ||
7791 (tx_rate < 0) || (tx_rate > actual_link_speed))
7794 adapter->vf_rate_link_speed = actual_link_speed;
7795 adapter->vf_data[vf].tx_rate = (u16)tx_rate;
7796 igb_set_vf_rate_limit(hw, vf, tx_rate, actual_link_speed);
7801 static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf,
7804 struct igb_adapter *adapter = netdev_priv(netdev);
7805 struct e1000_hw *hw = &adapter->hw;
7806 u32 reg_val, reg_offset;
7808 if (!adapter->vfs_allocated_count)
7811 if (vf >= adapter->vfs_allocated_count)
7814 reg_offset = (hw->mac.type == e1000_82576) ? E1000_DTXSWC : E1000_TXSWC;
7815 reg_val = rd32(reg_offset);
7817 reg_val |= ((1 << vf) |
7818 (1 << (vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT)));
7820 reg_val &= ~((1 << vf) |
7821 (1 << (vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT)));
7822 wr32(reg_offset, reg_val);
7824 adapter->vf_data[vf].spoofchk_enabled = setting;
7825 return E1000_SUCCESS;
7828 static int igb_ndo_get_vf_config(struct net_device *netdev,
7829 int vf, struct ifla_vf_info *ivi)
7831 struct igb_adapter *adapter = netdev_priv(netdev);
7832 if (vf >= adapter->vfs_allocated_count)
7835 memcpy(&ivi->mac, adapter->vf_data[vf].vf_mac_addresses, ETH_ALEN);
7836 ivi->tx_rate = adapter->vf_data[vf].tx_rate;
7837 ivi->vlan = adapter->vf_data[vf].pf_vlan;
7838 ivi->qos = adapter->vf_data[vf].pf_qos;
7839 ivi->spoofchk = adapter->vf_data[vf].spoofchk_enabled;
7843 static void igb_vmm_control(struct igb_adapter *adapter)
7845 struct e1000_hw *hw = &adapter->hw;
7848 switch (hw->mac.type) {
7854 /* replication is not supported for 82575 */
7857 /* notify HW that the MAC is adding vlan tags */
7858 reg = rd32(E1000_DTXCTL);
7859 reg |= E1000_DTXCTL_VLAN_ADDED;
7860 wr32(E1000_DTXCTL, reg);
7862 /* enable replication vlan tag stripping */
7863 reg = rd32(E1000_RPLOLR);
7864 reg |= E1000_RPLOLR_STRVLAN;
7865 wr32(E1000_RPLOLR, reg);
7867 /* none of the above registers are supported by i350 */
7871 if (adapter->vfs_allocated_count) {
7872 igb_vmdq_set_loopback_pf(hw, true);
7873 igb_vmdq_set_replication_pf(hw, true);
7874 igb_vmdq_set_anti_spoofing_pf(hw, true,
7875 adapter->vfs_allocated_count);
7877 igb_vmdq_set_loopback_pf(hw, false);
7878 igb_vmdq_set_replication_pf(hw, false);
7882 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba)
7884 struct e1000_hw *hw = &adapter->hw;
7888 if (hw->mac.type > e1000_82580) {
7889 if (adapter->flags & IGB_FLAG_DMAC) {
7892 /* force threshold to 0. */
7893 wr32(E1000_DMCTXTH, 0);
7895 /* DMA Coalescing high water mark needs to be greater
7896 * than the Rx threshold. Set hwm to PBA - max frame
7897 * size in 16B units, capping it at PBA - 6KB.
7899 hwm = 64 * pba - adapter->max_frame_size / 16;
7900 if (hwm < 64 * (pba - 6))
7901 hwm = 64 * (pba - 6);
7902 reg = rd32(E1000_FCRTC);
7903 reg &= ~E1000_FCRTC_RTH_COAL_MASK;
7904 reg |= ((hwm << E1000_FCRTC_RTH_COAL_SHIFT)
7905 & E1000_FCRTC_RTH_COAL_MASK);
7906 wr32(E1000_FCRTC, reg);
7908 /* Set the DMA Coalescing Rx threshold to PBA - 2 * max
7909 * frame size, capping it at PBA - 10KB.
7911 dmac_thr = pba - adapter->max_frame_size / 512;
7912 if (dmac_thr < pba - 10)
7913 dmac_thr = pba - 10;
7914 reg = rd32(E1000_DMACR);
7915 reg &= ~E1000_DMACR_DMACTHR_MASK;
7916 reg |= ((dmac_thr << E1000_DMACR_DMACTHR_SHIFT)
7917 & E1000_DMACR_DMACTHR_MASK);
7919 /* transition to L0x or L1 if available..*/
7920 reg |= (E1000_DMACR_DMAC_EN | E1000_DMACR_DMAC_LX_MASK);
7922 /* watchdog timer= +-1000 usec in 32usec intervals */
7925 /* Disable BMC-to-OS Watchdog Enable */
7926 if (hw->mac.type != e1000_i354)
7927 reg &= ~E1000_DMACR_DC_BMC2OSW_EN;
7929 wr32(E1000_DMACR, reg);
7931 /* no lower threshold to disable
7932 * coalescing(smart fifb)-UTRESH=0
7934 wr32(E1000_DMCRTRH, 0);
7936 reg = (IGB_DMCTLX_DCFLUSH_DIS | 0x4);
7938 wr32(E1000_DMCTLX, reg);
7940 /* free space in tx packet buffer to wake from
7943 wr32(E1000_DMCTXTH, (IGB_MIN_TXPBSIZE -
7944 (IGB_TX_BUF_4096 + adapter->max_frame_size)) >> 6);
7946 /* make low power state decision controlled
7949 reg = rd32(E1000_PCIEMISC);
7950 reg &= ~E1000_PCIEMISC_LX_DECISION;
7951 wr32(E1000_PCIEMISC, reg);
7952 } /* endif adapter->dmac is not disabled */
7953 } else if (hw->mac.type == e1000_82580) {
7954 u32 reg = rd32(E1000_PCIEMISC);
7955 wr32(E1000_PCIEMISC, reg & ~E1000_PCIEMISC_LX_DECISION);
7956 wr32(E1000_DMACR, 0);
7961 * igb_read_i2c_byte - Reads 8 bit word over I2C
7962 * @hw: pointer to hardware structure
7963 * @byte_offset: byte offset to read
7964 * @dev_addr: device address
7967 * Performs byte read operation over I2C interface at
7968 * a specified device address.
7970 s32 igb_read_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
7971 u8 dev_addr, u8 *data)
7973 struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
7974 struct i2c_client *this_client = adapter->i2c_client;
7979 return E1000_ERR_I2C;
7981 swfw_mask = E1000_SWFW_PHY0_SM;
7983 if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask)
7985 return E1000_ERR_SWFW_SYNC;
7987 status = i2c_smbus_read_byte_data(this_client, byte_offset);
7988 hw->mac.ops.release_swfw_sync(hw, swfw_mask);
7991 return E1000_ERR_I2C;
7994 return E1000_SUCCESS;
7999 * igb_write_i2c_byte - Writes 8 bit word over I2C
8000 * @hw: pointer to hardware structure
8001 * @byte_offset: byte offset to write
8002 * @dev_addr: device address
8003 * @data: value to write
8005 * Performs byte write operation over I2C interface at
8006 * a specified device address.
8008 s32 igb_write_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
8009 u8 dev_addr, u8 data)
8011 struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
8012 struct i2c_client *this_client = adapter->i2c_client;
8014 u16 swfw_mask = E1000_SWFW_PHY0_SM;
8017 return E1000_ERR_I2C;
8019 if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask) != E1000_SUCCESS)
8020 return E1000_ERR_SWFW_SYNC;
8021 status = i2c_smbus_write_byte_data(this_client, byte_offset, data);
8022 hw->mac.ops.release_swfw_sync(hw, swfw_mask);
8025 return E1000_ERR_I2C;
8027 return E1000_SUCCESS;
8031 int igb_reinit_queues(struct igb_adapter *adapter)
8033 struct net_device *netdev = adapter->netdev;
8034 struct pci_dev *pdev = adapter->pdev;
8037 if (netif_running(netdev))
8040 igb_clear_interrupt_scheme(adapter);
8042 if (igb_init_interrupt_scheme(adapter, true)) {
8043 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
8047 if (netif_running(netdev))
8048 err = igb_open(netdev);