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->flags & IGB_FLAG_HAS_MSIX) && 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]);
987 * igb_free_q_vector - Free memory allocated for specific interrupt vector
988 * @adapter: board private structure to initialize
989 * @v_idx: Index of vector to be freed
991 * This function frees the memory allocated to the q_vector.
993 static void igb_free_q_vector(struct igb_adapter *adapter, int v_idx)
995 struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
997 adapter->q_vector[v_idx] = NULL;
999 /* igb_get_stats64() might access the rings on this vector,
1000 * we must wait a grace period before freeing it.
1002 kfree_rcu(q_vector, rcu);
1006 * igb_reset_q_vector - Reset config for interrupt vector
1007 * @adapter: board private structure to initialize
1008 * @v_idx: Index of vector to be reset
1010 * If NAPI is enabled it will delete any references to the
1011 * NAPI struct. This is preparation for igb_free_q_vector.
1013 static void igb_reset_q_vector(struct igb_adapter *adapter, int v_idx)
1015 struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
1017 if (q_vector->tx.ring)
1018 adapter->tx_ring[q_vector->tx.ring->queue_index] = NULL;
1020 if (q_vector->rx.ring)
1021 adapter->tx_ring[q_vector->rx.ring->queue_index] = NULL;
1023 netif_napi_del(&q_vector->napi);
1027 static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
1029 int v_idx = adapter->num_q_vectors;
1031 if (adapter->flags & IGB_FLAG_HAS_MSIX)
1032 pci_disable_msix(adapter->pdev);
1033 else if (adapter->flags & IGB_FLAG_HAS_MSI)
1034 pci_disable_msi(adapter->pdev);
1037 igb_reset_q_vector(adapter, v_idx);
1041 * igb_free_q_vectors - Free memory allocated for interrupt vectors
1042 * @adapter: board private structure to initialize
1044 * This function frees the memory allocated to the q_vectors. In addition if
1045 * NAPI is enabled it will delete any references to the NAPI struct prior
1046 * to freeing the q_vector.
1048 static void igb_free_q_vectors(struct igb_adapter *adapter)
1050 int v_idx = adapter->num_q_vectors;
1052 adapter->num_tx_queues = 0;
1053 adapter->num_rx_queues = 0;
1054 adapter->num_q_vectors = 0;
1057 igb_reset_q_vector(adapter, v_idx);
1058 igb_free_q_vector(adapter, v_idx);
1063 * igb_clear_interrupt_scheme - reset the device to a state of no interrupts
1064 * @adapter: board private structure to initialize
1066 * This function resets the device so that it has 0 Rx queues, Tx queues, and
1067 * MSI-X interrupts allocated.
1069 static void igb_clear_interrupt_scheme(struct igb_adapter *adapter)
1071 igb_free_q_vectors(adapter);
1072 igb_reset_interrupt_capability(adapter);
1076 * igb_set_interrupt_capability - set MSI or MSI-X if supported
1077 * @adapter: board private structure to initialize
1078 * @msix: boolean value of MSIX capability
1080 * Attempt to configure interrupts using the best available
1081 * capabilities of the hardware and kernel.
1083 static void igb_set_interrupt_capability(struct igb_adapter *adapter, bool msix)
1090 adapter->flags |= IGB_FLAG_HAS_MSIX;
1092 /* Number of supported queues. */
1093 adapter->num_rx_queues = adapter->rss_queues;
1094 if (adapter->vfs_allocated_count)
1095 adapter->num_tx_queues = 1;
1097 adapter->num_tx_queues = adapter->rss_queues;
1099 /* start with one vector for every Rx queue */
1100 numvecs = adapter->num_rx_queues;
1102 /* if Tx handler is separate add 1 for every Tx queue */
1103 if (!(adapter->flags & IGB_FLAG_QUEUE_PAIRS))
1104 numvecs += adapter->num_tx_queues;
1106 /* store the number of vectors reserved for queues */
1107 adapter->num_q_vectors = numvecs;
1109 /* add 1 vector for link status interrupts */
1111 for (i = 0; i < numvecs; i++)
1112 adapter->msix_entries[i].entry = i;
1114 err = pci_enable_msix(adapter->pdev,
1115 adapter->msix_entries,
1120 igb_reset_interrupt_capability(adapter);
1122 /* If we can't do MSI-X, try MSI */
1124 #ifdef CONFIG_PCI_IOV
1125 /* disable SR-IOV for non MSI-X configurations */
1126 if (adapter->vf_data) {
1127 struct e1000_hw *hw = &adapter->hw;
1128 /* disable iov and allow time for transactions to clear */
1129 pci_disable_sriov(adapter->pdev);
1132 kfree(adapter->vf_data);
1133 adapter->vf_data = NULL;
1134 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
1137 dev_info(&adapter->pdev->dev, "IOV Disabled\n");
1140 adapter->vfs_allocated_count = 0;
1141 adapter->rss_queues = 1;
1142 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
1143 adapter->num_rx_queues = 1;
1144 adapter->num_tx_queues = 1;
1145 adapter->num_q_vectors = 1;
1146 if (!pci_enable_msi(adapter->pdev))
1147 adapter->flags |= IGB_FLAG_HAS_MSI;
1150 static void igb_add_ring(struct igb_ring *ring,
1151 struct igb_ring_container *head)
1158 * igb_alloc_q_vector - Allocate memory for a single interrupt vector
1159 * @adapter: board private structure to initialize
1160 * @v_count: q_vectors allocated on adapter, used for ring interleaving
1161 * @v_idx: index of vector in adapter struct
1162 * @txr_count: total number of Tx rings to allocate
1163 * @txr_idx: index of first Tx ring to allocate
1164 * @rxr_count: total number of Rx rings to allocate
1165 * @rxr_idx: index of first Rx ring to allocate
1167 * We allocate one q_vector. If allocation fails we return -ENOMEM.
1169 static int igb_alloc_q_vector(struct igb_adapter *adapter,
1170 int v_count, int v_idx,
1171 int txr_count, int txr_idx,
1172 int rxr_count, int rxr_idx)
1174 struct igb_q_vector *q_vector;
1175 struct igb_ring *ring;
1176 int ring_count, size;
1178 /* igb only supports 1 Tx and/or 1 Rx queue per vector */
1179 if (txr_count > 1 || rxr_count > 1)
1182 ring_count = txr_count + rxr_count;
1183 size = sizeof(struct igb_q_vector) +
1184 (sizeof(struct igb_ring) * ring_count);
1186 /* allocate q_vector and rings */
1187 q_vector = adapter->q_vector[v_idx];
1189 q_vector = kzalloc(size, GFP_KERNEL);
1193 /* initialize NAPI */
1194 netif_napi_add(adapter->netdev, &q_vector->napi,
1197 /* tie q_vector and adapter together */
1198 adapter->q_vector[v_idx] = q_vector;
1199 q_vector->adapter = adapter;
1201 /* initialize work limits */
1202 q_vector->tx.work_limit = adapter->tx_work_limit;
1204 /* initialize ITR configuration */
1205 q_vector->itr_register = adapter->hw.hw_addr + E1000_EITR(0);
1206 q_vector->itr_val = IGB_START_ITR;
1208 /* initialize pointer to rings */
1209 ring = q_vector->ring;
1213 /* rx or rx/tx vector */
1214 if (!adapter->rx_itr_setting || adapter->rx_itr_setting > 3)
1215 q_vector->itr_val = adapter->rx_itr_setting;
1217 /* tx only vector */
1218 if (!adapter->tx_itr_setting || adapter->tx_itr_setting > 3)
1219 q_vector->itr_val = adapter->tx_itr_setting;
1223 /* assign generic ring traits */
1224 ring->dev = &adapter->pdev->dev;
1225 ring->netdev = adapter->netdev;
1227 /* configure backlink on ring */
1228 ring->q_vector = q_vector;
1230 /* update q_vector Tx values */
1231 igb_add_ring(ring, &q_vector->tx);
1233 /* For 82575, context index must be unique per ring. */
1234 if (adapter->hw.mac.type == e1000_82575)
1235 set_bit(IGB_RING_FLAG_TX_CTX_IDX, &ring->flags);
1237 /* apply Tx specific ring traits */
1238 ring->count = adapter->tx_ring_count;
1239 ring->queue_index = txr_idx;
1241 u64_stats_init(&ring->tx_syncp);
1242 u64_stats_init(&ring->tx_syncp2);
1244 /* assign ring to adapter */
1245 adapter->tx_ring[txr_idx] = ring;
1247 /* push pointer to next ring */
1252 /* assign generic ring traits */
1253 ring->dev = &adapter->pdev->dev;
1254 ring->netdev = adapter->netdev;
1256 /* configure backlink on ring */
1257 ring->q_vector = q_vector;
1259 /* update q_vector Rx values */
1260 igb_add_ring(ring, &q_vector->rx);
1262 /* set flag indicating ring supports SCTP checksum offload */
1263 if (adapter->hw.mac.type >= e1000_82576)
1264 set_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags);
1267 * On i350, i354, i210, and i211, loopback VLAN packets
1268 * have the tag byte-swapped.
1270 if (adapter->hw.mac.type >= e1000_i350)
1271 set_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &ring->flags);
1273 /* apply Rx specific ring traits */
1274 ring->count = adapter->rx_ring_count;
1275 ring->queue_index = rxr_idx;
1277 u64_stats_init(&ring->rx_syncp);
1279 /* assign ring to adapter */
1280 adapter->rx_ring[rxr_idx] = ring;
1288 * igb_alloc_q_vectors - Allocate memory for interrupt vectors
1289 * @adapter: board private structure to initialize
1291 * We allocate one q_vector per queue interrupt. If allocation fails we
1294 static int igb_alloc_q_vectors(struct igb_adapter *adapter)
1296 int q_vectors = adapter->num_q_vectors;
1297 int rxr_remaining = adapter->num_rx_queues;
1298 int txr_remaining = adapter->num_tx_queues;
1299 int rxr_idx = 0, txr_idx = 0, v_idx = 0;
1302 if (q_vectors >= (rxr_remaining + txr_remaining)) {
1303 for (; rxr_remaining; v_idx++) {
1304 err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1310 /* update counts and index */
1316 for (; v_idx < q_vectors; v_idx++) {
1317 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_idx);
1318 int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_idx);
1319 err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1320 tqpv, txr_idx, rqpv, rxr_idx);
1325 /* update counts and index */
1326 rxr_remaining -= rqpv;
1327 txr_remaining -= tqpv;
1335 adapter->num_tx_queues = 0;
1336 adapter->num_rx_queues = 0;
1337 adapter->num_q_vectors = 0;
1340 igb_free_q_vector(adapter, v_idx);
1346 * igb_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
1347 * @adapter: board private structure to initialize
1348 * @msix: boolean value of MSIX capability
1350 * This function initializes the interrupts and allocates all of the queues.
1352 static int igb_init_interrupt_scheme(struct igb_adapter *adapter, bool msix)
1354 struct pci_dev *pdev = adapter->pdev;
1357 igb_set_interrupt_capability(adapter, msix);
1359 err = igb_alloc_q_vectors(adapter);
1361 dev_err(&pdev->dev, "Unable to allocate memory for vectors\n");
1362 goto err_alloc_q_vectors;
1365 igb_cache_ring_register(adapter);
1369 err_alloc_q_vectors:
1370 igb_reset_interrupt_capability(adapter);
1375 * igb_request_irq - initialize interrupts
1376 * @adapter: board private structure to initialize
1378 * Attempts to configure interrupts using the best available
1379 * capabilities of the hardware and kernel.
1381 static int igb_request_irq(struct igb_adapter *adapter)
1383 struct net_device *netdev = adapter->netdev;
1384 struct pci_dev *pdev = adapter->pdev;
1387 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1388 err = igb_request_msix(adapter);
1391 /* fall back to MSI */
1392 igb_free_all_tx_resources(adapter);
1393 igb_free_all_rx_resources(adapter);
1395 igb_clear_interrupt_scheme(adapter);
1396 err = igb_init_interrupt_scheme(adapter, false);
1400 igb_setup_all_tx_resources(adapter);
1401 igb_setup_all_rx_resources(adapter);
1402 igb_configure(adapter);
1405 igb_assign_vector(adapter->q_vector[0], 0);
1407 if (adapter->flags & IGB_FLAG_HAS_MSI) {
1408 err = request_irq(pdev->irq, igb_intr_msi, 0,
1409 netdev->name, adapter);
1413 /* fall back to legacy interrupts */
1414 igb_reset_interrupt_capability(adapter);
1415 adapter->flags &= ~IGB_FLAG_HAS_MSI;
1418 err = request_irq(pdev->irq, igb_intr, IRQF_SHARED,
1419 netdev->name, adapter);
1422 dev_err(&pdev->dev, "Error %d getting interrupt\n",
1429 static void igb_free_irq(struct igb_adapter *adapter)
1431 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1434 free_irq(adapter->msix_entries[vector++].vector, adapter);
1436 for (i = 0; i < adapter->num_q_vectors; i++)
1437 free_irq(adapter->msix_entries[vector++].vector,
1438 adapter->q_vector[i]);
1440 free_irq(adapter->pdev->irq, adapter);
1445 * igb_irq_disable - Mask off interrupt generation on the NIC
1446 * @adapter: board private structure
1448 static void igb_irq_disable(struct igb_adapter *adapter)
1450 struct e1000_hw *hw = &adapter->hw;
1452 /* we need to be careful when disabling interrupts. The VFs are also
1453 * mapped into these registers and so clearing the bits can cause
1454 * issues on the VF drivers so we only need to clear what we set
1456 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1457 u32 regval = rd32(E1000_EIAM);
1458 wr32(E1000_EIAM, regval & ~adapter->eims_enable_mask);
1459 wr32(E1000_EIMC, adapter->eims_enable_mask);
1460 regval = rd32(E1000_EIAC);
1461 wr32(E1000_EIAC, regval & ~adapter->eims_enable_mask);
1465 wr32(E1000_IMC, ~0);
1467 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1469 for (i = 0; i < adapter->num_q_vectors; i++)
1470 synchronize_irq(adapter->msix_entries[i].vector);
1472 synchronize_irq(adapter->pdev->irq);
1477 * igb_irq_enable - Enable default interrupt generation settings
1478 * @adapter: board private structure
1480 static void igb_irq_enable(struct igb_adapter *adapter)
1482 struct e1000_hw *hw = &adapter->hw;
1484 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1485 u32 ims = E1000_IMS_LSC | E1000_IMS_DOUTSYNC | E1000_IMS_DRSTA;
1486 u32 regval = rd32(E1000_EIAC);
1487 wr32(E1000_EIAC, regval | adapter->eims_enable_mask);
1488 regval = rd32(E1000_EIAM);
1489 wr32(E1000_EIAM, regval | adapter->eims_enable_mask);
1490 wr32(E1000_EIMS, adapter->eims_enable_mask);
1491 if (adapter->vfs_allocated_count) {
1492 wr32(E1000_MBVFIMR, 0xFF);
1493 ims |= E1000_IMS_VMMB;
1495 wr32(E1000_IMS, ims);
1497 wr32(E1000_IMS, IMS_ENABLE_MASK |
1499 wr32(E1000_IAM, IMS_ENABLE_MASK |
1504 static void igb_update_mng_vlan(struct igb_adapter *adapter)
1506 struct e1000_hw *hw = &adapter->hw;
1507 u16 vid = adapter->hw.mng_cookie.vlan_id;
1508 u16 old_vid = adapter->mng_vlan_id;
1510 if (hw->mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1511 /* add VID to filter table */
1512 igb_vfta_set(hw, vid, true);
1513 adapter->mng_vlan_id = vid;
1515 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1518 if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
1520 !test_bit(old_vid, adapter->active_vlans)) {
1521 /* remove VID from filter table */
1522 igb_vfta_set(hw, old_vid, false);
1527 * igb_release_hw_control - release control of the h/w to f/w
1528 * @adapter: address of board private structure
1530 * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
1531 * For ASF and Pass Through versions of f/w this means that the
1532 * driver is no longer loaded.
1534 static void igb_release_hw_control(struct igb_adapter *adapter)
1536 struct e1000_hw *hw = &adapter->hw;
1539 /* Let firmware take over control of h/w */
1540 ctrl_ext = rd32(E1000_CTRL_EXT);
1541 wr32(E1000_CTRL_EXT,
1542 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1546 * igb_get_hw_control - get control of the h/w from f/w
1547 * @adapter: address of board private structure
1549 * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
1550 * For ASF and Pass Through versions of f/w this means that
1551 * the driver is loaded.
1553 static void igb_get_hw_control(struct igb_adapter *adapter)
1555 struct e1000_hw *hw = &adapter->hw;
1558 /* Let firmware know the driver has taken over */
1559 ctrl_ext = rd32(E1000_CTRL_EXT);
1560 wr32(E1000_CTRL_EXT,
1561 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1565 * igb_configure - configure the hardware for RX and TX
1566 * @adapter: private board structure
1568 static void igb_configure(struct igb_adapter *adapter)
1570 struct net_device *netdev = adapter->netdev;
1573 igb_get_hw_control(adapter);
1574 igb_set_rx_mode(netdev);
1576 igb_restore_vlan(adapter);
1578 igb_setup_tctl(adapter);
1579 igb_setup_mrqc(adapter);
1580 igb_setup_rctl(adapter);
1582 igb_configure_tx(adapter);
1583 igb_configure_rx(adapter);
1585 igb_rx_fifo_flush_82575(&adapter->hw);
1587 /* call igb_desc_unused which always leaves
1588 * at least 1 descriptor unused to make sure
1589 * next_to_use != next_to_clean
1591 for (i = 0; i < adapter->num_rx_queues; i++) {
1592 struct igb_ring *ring = adapter->rx_ring[i];
1593 igb_alloc_rx_buffers(ring, igb_desc_unused(ring));
1598 * igb_power_up_link - Power up the phy/serdes link
1599 * @adapter: address of board private structure
1601 void igb_power_up_link(struct igb_adapter *adapter)
1603 igb_reset_phy(&adapter->hw);
1605 if (adapter->hw.phy.media_type == e1000_media_type_copper)
1606 igb_power_up_phy_copper(&adapter->hw);
1608 igb_power_up_serdes_link_82575(&adapter->hw);
1612 * igb_power_down_link - Power down the phy/serdes link
1613 * @adapter: address of board private structure
1615 static void igb_power_down_link(struct igb_adapter *adapter)
1617 if (adapter->hw.phy.media_type == e1000_media_type_copper)
1618 igb_power_down_phy_copper_82575(&adapter->hw);
1620 igb_shutdown_serdes_link_82575(&adapter->hw);
1624 * Detect and switch function for Media Auto Sense
1625 * @adapter: address of the board private structure
1627 static void igb_check_swap_media(struct igb_adapter *adapter)
1629 struct e1000_hw *hw = &adapter->hw;
1630 u32 ctrl_ext, connsw;
1631 bool swap_now = false;
1633 ctrl_ext = rd32(E1000_CTRL_EXT);
1634 connsw = rd32(E1000_CONNSW);
1636 /* need to live swap if current media is copper and we have fiber/serdes
1640 if ((hw->phy.media_type == e1000_media_type_copper) &&
1641 (!(connsw & E1000_CONNSW_AUTOSENSE_EN))) {
1643 } else if (!(connsw & E1000_CONNSW_SERDESD)) {
1644 /* copper signal takes time to appear */
1645 if (adapter->copper_tries < 4) {
1646 adapter->copper_tries++;
1647 connsw |= E1000_CONNSW_AUTOSENSE_CONF;
1648 wr32(E1000_CONNSW, connsw);
1651 adapter->copper_tries = 0;
1652 if ((connsw & E1000_CONNSW_PHYSD) &&
1653 (!(connsw & E1000_CONNSW_PHY_PDN))) {
1655 connsw &= ~E1000_CONNSW_AUTOSENSE_CONF;
1656 wr32(E1000_CONNSW, connsw);
1664 switch (hw->phy.media_type) {
1665 case e1000_media_type_copper:
1666 netdev_info(adapter->netdev,
1667 "MAS: changing media to fiber/serdes\n");
1669 E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
1670 adapter->flags |= IGB_FLAG_MEDIA_RESET;
1671 adapter->copper_tries = 0;
1673 case e1000_media_type_internal_serdes:
1674 case e1000_media_type_fiber:
1675 netdev_info(adapter->netdev,
1676 "MAS: changing media to copper\n");
1678 ~E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
1679 adapter->flags |= IGB_FLAG_MEDIA_RESET;
1682 /* shouldn't get here during regular operation */
1683 netdev_err(adapter->netdev,
1684 "AMS: Invalid media type found, returning\n");
1687 wr32(E1000_CTRL_EXT, ctrl_ext);
1691 * igb_up - Open the interface and prepare it to handle traffic
1692 * @adapter: board private structure
1694 int igb_up(struct igb_adapter *adapter)
1696 struct e1000_hw *hw = &adapter->hw;
1699 /* hardware has been reset, we need to reload some things */
1700 igb_configure(adapter);
1702 clear_bit(__IGB_DOWN, &adapter->state);
1704 for (i = 0; i < adapter->num_q_vectors; i++)
1705 napi_enable(&(adapter->q_vector[i]->napi));
1707 if (adapter->flags & IGB_FLAG_HAS_MSIX)
1708 igb_configure_msix(adapter);
1710 igb_assign_vector(adapter->q_vector[0], 0);
1712 /* Clear any pending interrupts. */
1714 igb_irq_enable(adapter);
1716 /* notify VFs that reset has been completed */
1717 if (adapter->vfs_allocated_count) {
1718 u32 reg_data = rd32(E1000_CTRL_EXT);
1719 reg_data |= E1000_CTRL_EXT_PFRSTD;
1720 wr32(E1000_CTRL_EXT, reg_data);
1723 netif_tx_start_all_queues(adapter->netdev);
1725 /* start the watchdog. */
1726 hw->mac.get_link_status = 1;
1727 schedule_work(&adapter->watchdog_task);
1732 void igb_down(struct igb_adapter *adapter)
1734 struct net_device *netdev = adapter->netdev;
1735 struct e1000_hw *hw = &adapter->hw;
1739 /* signal that we're down so the interrupt handler does not
1740 * reschedule our watchdog timer
1742 set_bit(__IGB_DOWN, &adapter->state);
1744 /* disable receives in the hardware */
1745 rctl = rd32(E1000_RCTL);
1746 wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
1747 /* flush and sleep below */
1749 netif_tx_stop_all_queues(netdev);
1751 /* disable transmits in the hardware */
1752 tctl = rd32(E1000_TCTL);
1753 tctl &= ~E1000_TCTL_EN;
1754 wr32(E1000_TCTL, tctl);
1755 /* flush both disables and wait for them to finish */
1759 igb_irq_disable(adapter);
1761 adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
1763 for (i = 0; i < adapter->num_q_vectors; i++) {
1764 napi_synchronize(&(adapter->q_vector[i]->napi));
1765 napi_disable(&(adapter->q_vector[i]->napi));
1769 del_timer_sync(&adapter->watchdog_timer);
1770 del_timer_sync(&adapter->phy_info_timer);
1772 netif_carrier_off(netdev);
1774 /* record the stats before reset*/
1775 spin_lock(&adapter->stats64_lock);
1776 igb_update_stats(adapter, &adapter->stats64);
1777 spin_unlock(&adapter->stats64_lock);
1779 adapter->link_speed = 0;
1780 adapter->link_duplex = 0;
1782 if (!pci_channel_offline(adapter->pdev))
1784 igb_clean_all_tx_rings(adapter);
1785 igb_clean_all_rx_rings(adapter);
1786 #ifdef CONFIG_IGB_DCA
1788 /* since we reset the hardware DCA settings were cleared */
1789 igb_setup_dca(adapter);
1793 void igb_reinit_locked(struct igb_adapter *adapter)
1795 WARN_ON(in_interrupt());
1796 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
1800 clear_bit(__IGB_RESETTING, &adapter->state);
1803 /** igb_enable_mas - Media Autosense re-enable after swap
1805 * @adapter: adapter struct
1807 static s32 igb_enable_mas(struct igb_adapter *adapter)
1809 struct e1000_hw *hw = &adapter->hw;
1813 connsw = rd32(E1000_CONNSW);
1814 if (!(hw->phy.media_type == e1000_media_type_copper))
1817 /* configure for SerDes media detect */
1818 if (!(connsw & E1000_CONNSW_SERDESD)) {
1819 connsw |= E1000_CONNSW_ENRGSRC;
1820 connsw |= E1000_CONNSW_AUTOSENSE_EN;
1821 wr32(E1000_CONNSW, connsw);
1823 } else if (connsw & E1000_CONNSW_SERDESD) {
1824 /* already SerDes, no need to enable anything */
1827 netdev_info(adapter->netdev,
1828 "MAS: Unable to configure feature, disabling..\n");
1829 adapter->flags &= ~IGB_FLAG_MAS_ENABLE;
1834 void igb_reset(struct igb_adapter *adapter)
1836 struct pci_dev *pdev = adapter->pdev;
1837 struct e1000_hw *hw = &adapter->hw;
1838 struct e1000_mac_info *mac = &hw->mac;
1839 struct e1000_fc_info *fc = &hw->fc;
1840 u32 pba = 0, tx_space, min_tx_space, min_rx_space, hwm;
1842 /* Repartition Pba for greater than 9k mtu
1843 * To take effect CTRL.RST is required.
1845 switch (mac->type) {
1849 pba = rd32(E1000_RXPBS);
1850 pba = igb_rxpbs_adjust_82580(pba);
1853 pba = rd32(E1000_RXPBS);
1854 pba &= E1000_RXPBS_SIZE_MASK_82576;
1860 pba = E1000_PBA_34K;
1864 if ((adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) &&
1865 (mac->type < e1000_82576)) {
1866 /* adjust PBA for jumbo frames */
1867 wr32(E1000_PBA, pba);
1869 /* To maintain wire speed transmits, the Tx FIFO should be
1870 * large enough to accommodate two full transmit packets,
1871 * rounded up to the next 1KB and expressed in KB. Likewise,
1872 * the Rx FIFO should be large enough to accommodate at least
1873 * one full receive packet and is similarly rounded up and
1876 pba = rd32(E1000_PBA);
1877 /* upper 16 bits has Tx packet buffer allocation size in KB */
1878 tx_space = pba >> 16;
1879 /* lower 16 bits has Rx packet buffer allocation size in KB */
1881 /* the Tx fifo also stores 16 bytes of information about the Tx
1882 * but don't include ethernet FCS because hardware appends it
1884 min_tx_space = (adapter->max_frame_size +
1885 sizeof(union e1000_adv_tx_desc) -
1887 min_tx_space = ALIGN(min_tx_space, 1024);
1888 min_tx_space >>= 10;
1889 /* software strips receive CRC, so leave room for it */
1890 min_rx_space = adapter->max_frame_size;
1891 min_rx_space = ALIGN(min_rx_space, 1024);
1892 min_rx_space >>= 10;
1894 /* If current Tx allocation is less than the min Tx FIFO size,
1895 * and the min Tx FIFO size is less than the current Rx FIFO
1896 * allocation, take space away from current Rx allocation
1898 if (tx_space < min_tx_space &&
1899 ((min_tx_space - tx_space) < pba)) {
1900 pba = pba - (min_tx_space - tx_space);
1902 /* if short on Rx space, Rx wins and must trump Tx
1905 if (pba < min_rx_space)
1908 wr32(E1000_PBA, pba);
1911 /* flow control settings */
1912 /* The high water mark must be low enough to fit one full frame
1913 * (or the size used for early receive) above it in the Rx FIFO.
1914 * Set it to the lower of:
1915 * - 90% of the Rx FIFO size, or
1916 * - the full Rx FIFO size minus one full frame
1918 hwm = min(((pba << 10) * 9 / 10),
1919 ((pba << 10) - 2 * adapter->max_frame_size));
1921 fc->high_water = hwm & 0xFFFFFFF0; /* 16-byte granularity */
1922 fc->low_water = fc->high_water - 16;
1923 fc->pause_time = 0xFFFF;
1925 fc->current_mode = fc->requested_mode;
1927 /* disable receive for all VFs and wait one second */
1928 if (adapter->vfs_allocated_count) {
1930 for (i = 0 ; i < adapter->vfs_allocated_count; i++)
1931 adapter->vf_data[i].flags &= IGB_VF_FLAG_PF_SET_MAC;
1933 /* ping all the active vfs to let them know we are going down */
1934 igb_ping_all_vfs(adapter);
1936 /* disable transmits and receives */
1937 wr32(E1000_VFRE, 0);
1938 wr32(E1000_VFTE, 0);
1941 /* Allow time for pending master requests to run */
1942 hw->mac.ops.reset_hw(hw);
1945 if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
1946 /* need to resetup here after media swap */
1947 adapter->ei.get_invariants(hw);
1948 adapter->flags &= ~IGB_FLAG_MEDIA_RESET;
1950 if (adapter->flags & IGB_FLAG_MAS_ENABLE) {
1951 if (igb_enable_mas(adapter))
1953 "Error enabling Media Auto Sense\n");
1955 if (hw->mac.ops.init_hw(hw))
1956 dev_err(&pdev->dev, "Hardware Error\n");
1958 /* Flow control settings reset on hardware reset, so guarantee flow
1959 * control is off when forcing speed.
1961 if (!hw->mac.autoneg)
1962 igb_force_mac_fc(hw);
1964 igb_init_dmac(adapter, pba);
1965 #ifdef CONFIG_IGB_HWMON
1966 /* Re-initialize the thermal sensor on i350 devices. */
1967 if (!test_bit(__IGB_DOWN, &adapter->state)) {
1968 if (mac->type == e1000_i350 && hw->bus.func == 0) {
1969 /* If present, re-initialize the external thermal sensor
1973 mac->ops.init_thermal_sensor_thresh(hw);
1977 if (!netif_running(adapter->netdev))
1978 igb_power_down_link(adapter);
1980 igb_update_mng_vlan(adapter);
1982 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
1983 wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
1985 /* Re-enable PTP, where applicable. */
1986 igb_ptp_reset(adapter);
1988 igb_get_phy_info(hw);
1991 static netdev_features_t igb_fix_features(struct net_device *netdev,
1992 netdev_features_t features)
1994 /* Since there is no support for separate Rx/Tx vlan accel
1995 * enable/disable make sure Tx flag is always in same state as Rx.
1997 if (features & NETIF_F_HW_VLAN_CTAG_RX)
1998 features |= NETIF_F_HW_VLAN_CTAG_TX;
2000 features &= ~NETIF_F_HW_VLAN_CTAG_TX;
2005 static int igb_set_features(struct net_device *netdev,
2006 netdev_features_t features)
2008 netdev_features_t changed = netdev->features ^ features;
2009 struct igb_adapter *adapter = netdev_priv(netdev);
2011 if (changed & NETIF_F_HW_VLAN_CTAG_RX)
2012 igb_vlan_mode(netdev, features);
2014 if (!(changed & NETIF_F_RXALL))
2017 netdev->features = features;
2019 if (netif_running(netdev))
2020 igb_reinit_locked(adapter);
2027 static const struct net_device_ops igb_netdev_ops = {
2028 .ndo_open = igb_open,
2029 .ndo_stop = igb_close,
2030 .ndo_start_xmit = igb_xmit_frame,
2031 .ndo_get_stats64 = igb_get_stats64,
2032 .ndo_set_rx_mode = igb_set_rx_mode,
2033 .ndo_set_mac_address = igb_set_mac,
2034 .ndo_change_mtu = igb_change_mtu,
2035 .ndo_do_ioctl = igb_ioctl,
2036 .ndo_tx_timeout = igb_tx_timeout,
2037 .ndo_validate_addr = eth_validate_addr,
2038 .ndo_vlan_rx_add_vid = igb_vlan_rx_add_vid,
2039 .ndo_vlan_rx_kill_vid = igb_vlan_rx_kill_vid,
2040 .ndo_set_vf_mac = igb_ndo_set_vf_mac,
2041 .ndo_set_vf_vlan = igb_ndo_set_vf_vlan,
2042 .ndo_set_vf_tx_rate = igb_ndo_set_vf_bw,
2043 .ndo_set_vf_spoofchk = igb_ndo_set_vf_spoofchk,
2044 .ndo_get_vf_config = igb_ndo_get_vf_config,
2045 #ifdef CONFIG_NET_POLL_CONTROLLER
2046 .ndo_poll_controller = igb_netpoll,
2048 .ndo_fix_features = igb_fix_features,
2049 .ndo_set_features = igb_set_features,
2053 * igb_set_fw_version - Configure version string for ethtool
2054 * @adapter: adapter struct
2056 void igb_set_fw_version(struct igb_adapter *adapter)
2058 struct e1000_hw *hw = &adapter->hw;
2059 struct e1000_fw_version fw;
2061 igb_get_fw_version(hw, &fw);
2063 switch (hw->mac.type) {
2066 if (!(igb_get_flash_presence_i210(hw))) {
2067 snprintf(adapter->fw_version,
2068 sizeof(adapter->fw_version),
2070 fw.invm_major, fw.invm_minor,
2076 /* if option is rom valid, display its version too */
2078 snprintf(adapter->fw_version,
2079 sizeof(adapter->fw_version),
2080 "%d.%d, 0x%08x, %d.%d.%d",
2081 fw.eep_major, fw.eep_minor, fw.etrack_id,
2082 fw.or_major, fw.or_build, fw.or_patch);
2084 } else if (fw.etrack_id != 0X0000) {
2085 snprintf(adapter->fw_version,
2086 sizeof(adapter->fw_version),
2088 fw.eep_major, fw.eep_minor, fw.etrack_id);
2090 snprintf(adapter->fw_version,
2091 sizeof(adapter->fw_version),
2093 fw.eep_major, fw.eep_minor, fw.eep_build);
2101 * igb_init_mas - init Media Autosense feature if enabled in the NVM
2103 * @adapter: adapter struct
2105 static void igb_init_mas(struct igb_adapter *adapter)
2107 struct e1000_hw *hw = &adapter->hw;
2110 hw->nvm.ops.read(hw, NVM_COMPAT, 1, &eeprom_data);
2111 switch (hw->bus.func) {
2113 if (eeprom_data & IGB_MAS_ENABLE_0) {
2114 adapter->flags |= IGB_FLAG_MAS_ENABLE;
2115 netdev_info(adapter->netdev,
2116 "MAS: Enabling Media Autosense for port %d\n",
2121 if (eeprom_data & IGB_MAS_ENABLE_1) {
2122 adapter->flags |= IGB_FLAG_MAS_ENABLE;
2123 netdev_info(adapter->netdev,
2124 "MAS: Enabling Media Autosense for port %d\n",
2129 if (eeprom_data & IGB_MAS_ENABLE_2) {
2130 adapter->flags |= IGB_FLAG_MAS_ENABLE;
2131 netdev_info(adapter->netdev,
2132 "MAS: Enabling Media Autosense for port %d\n",
2137 if (eeprom_data & IGB_MAS_ENABLE_3) {
2138 adapter->flags |= IGB_FLAG_MAS_ENABLE;
2139 netdev_info(adapter->netdev,
2140 "MAS: Enabling Media Autosense for port %d\n",
2145 /* Shouldn't get here */
2146 netdev_err(adapter->netdev,
2147 "MAS: Invalid port configuration, returning\n");
2153 * igb_init_i2c - Init I2C interface
2154 * @adapter: pointer to adapter structure
2156 static s32 igb_init_i2c(struct igb_adapter *adapter)
2158 s32 status = E1000_SUCCESS;
2160 /* I2C interface supported on i350 devices */
2161 if (adapter->hw.mac.type != e1000_i350)
2162 return E1000_SUCCESS;
2164 /* Initialize the i2c bus which is controlled by the registers.
2165 * This bus will use the i2c_algo_bit structue that implements
2166 * the protocol through toggling of the 4 bits in the register.
2168 adapter->i2c_adap.owner = THIS_MODULE;
2169 adapter->i2c_algo = igb_i2c_algo;
2170 adapter->i2c_algo.data = adapter;
2171 adapter->i2c_adap.algo_data = &adapter->i2c_algo;
2172 adapter->i2c_adap.dev.parent = &adapter->pdev->dev;
2173 strlcpy(adapter->i2c_adap.name, "igb BB",
2174 sizeof(adapter->i2c_adap.name));
2175 status = i2c_bit_add_bus(&adapter->i2c_adap);
2180 * igb_probe - Device Initialization Routine
2181 * @pdev: PCI device information struct
2182 * @ent: entry in igb_pci_tbl
2184 * Returns 0 on success, negative on failure
2186 * igb_probe initializes an adapter identified by a pci_dev structure.
2187 * The OS initialization, configuring of the adapter private structure,
2188 * and a hardware reset occur.
2190 static int igb_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2192 struct net_device *netdev;
2193 struct igb_adapter *adapter;
2194 struct e1000_hw *hw;
2195 u16 eeprom_data = 0;
2197 static int global_quad_port_a; /* global quad port a indication */
2198 const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
2199 int err, pci_using_dac;
2200 u8 part_str[E1000_PBANUM_LENGTH];
2202 /* Catch broken hardware that put the wrong VF device ID in
2203 * the PCIe SR-IOV capability.
2205 if (pdev->is_virtfn) {
2206 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
2207 pci_name(pdev), pdev->vendor, pdev->device);
2211 err = pci_enable_device_mem(pdev);
2216 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
2220 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
2223 "No usable DMA configuration, aborting\n");
2228 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
2234 pci_enable_pcie_error_reporting(pdev);
2236 pci_set_master(pdev);
2237 pci_save_state(pdev);
2240 netdev = alloc_etherdev_mq(sizeof(struct igb_adapter),
2243 goto err_alloc_etherdev;
2245 SET_NETDEV_DEV(netdev, &pdev->dev);
2247 pci_set_drvdata(pdev, netdev);
2248 adapter = netdev_priv(netdev);
2249 adapter->netdev = netdev;
2250 adapter->pdev = pdev;
2253 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
2256 hw->hw_addr = pci_iomap(pdev, 0, 0);
2260 netdev->netdev_ops = &igb_netdev_ops;
2261 igb_set_ethtool_ops(netdev);
2262 netdev->watchdog_timeo = 5 * HZ;
2264 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2266 netdev->mem_start = pci_resource_start(pdev, 0);
2267 netdev->mem_end = pci_resource_end(pdev, 0);
2269 /* PCI config space info */
2270 hw->vendor_id = pdev->vendor;
2271 hw->device_id = pdev->device;
2272 hw->revision_id = pdev->revision;
2273 hw->subsystem_vendor_id = pdev->subsystem_vendor;
2274 hw->subsystem_device_id = pdev->subsystem_device;
2276 /* Copy the default MAC, PHY and NVM function pointers */
2277 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
2278 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
2279 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
2280 /* Initialize skew-specific constants */
2281 err = ei->get_invariants(hw);
2285 /* setup the private structure */
2286 err = igb_sw_init(adapter);
2290 igb_get_bus_info_pcie(hw);
2292 hw->phy.autoneg_wait_to_complete = false;
2294 /* Copper options */
2295 if (hw->phy.media_type == e1000_media_type_copper) {
2296 hw->phy.mdix = AUTO_ALL_MODES;
2297 hw->phy.disable_polarity_correction = false;
2298 hw->phy.ms_type = e1000_ms_hw_default;
2301 if (igb_check_reset_block(hw))
2302 dev_info(&pdev->dev,
2303 "PHY reset is blocked due to SOL/IDER session.\n");
2305 /* features is initialized to 0 in allocation, it might have bits
2306 * set by igb_sw_init so we should use an or instead of an
2309 netdev->features |= NETIF_F_SG |
2316 NETIF_F_HW_VLAN_CTAG_RX |
2317 NETIF_F_HW_VLAN_CTAG_TX;
2319 /* copy netdev features into list of user selectable features */
2320 netdev->hw_features |= netdev->features;
2321 netdev->hw_features |= NETIF_F_RXALL;
2323 /* set this bit last since it cannot be part of hw_features */
2324 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
2326 netdev->vlan_features |= NETIF_F_TSO |
2332 netdev->priv_flags |= IFF_SUPP_NOFCS;
2334 if (pci_using_dac) {
2335 netdev->features |= NETIF_F_HIGHDMA;
2336 netdev->vlan_features |= NETIF_F_HIGHDMA;
2339 if (hw->mac.type >= e1000_82576) {
2340 netdev->hw_features |= NETIF_F_SCTP_CSUM;
2341 netdev->features |= NETIF_F_SCTP_CSUM;
2344 netdev->priv_flags |= IFF_UNICAST_FLT;
2346 adapter->en_mng_pt = igb_enable_mng_pass_thru(hw);
2348 /* before reading the NVM, reset the controller to put the device in a
2349 * known good starting state
2351 hw->mac.ops.reset_hw(hw);
2353 /* make sure the NVM is good , i211/i210 parts can have special NVM
2354 * that doesn't contain a checksum
2356 switch (hw->mac.type) {
2359 if (igb_get_flash_presence_i210(hw)) {
2360 if (hw->nvm.ops.validate(hw) < 0) {
2362 "The NVM Checksum Is Not Valid\n");
2369 if (hw->nvm.ops.validate(hw) < 0) {
2370 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
2377 /* copy the MAC address out of the NVM */
2378 if (hw->mac.ops.read_mac_addr(hw))
2379 dev_err(&pdev->dev, "NVM Read Error\n");
2381 memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
2383 if (!is_valid_ether_addr(netdev->dev_addr)) {
2384 dev_err(&pdev->dev, "Invalid MAC Address\n");
2389 /* get firmware version for ethtool -i */
2390 igb_set_fw_version(adapter);
2392 setup_timer(&adapter->watchdog_timer, igb_watchdog,
2393 (unsigned long) adapter);
2394 setup_timer(&adapter->phy_info_timer, igb_update_phy_info,
2395 (unsigned long) adapter);
2397 INIT_WORK(&adapter->reset_task, igb_reset_task);
2398 INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
2400 /* Initialize link properties that are user-changeable */
2401 adapter->fc_autoneg = true;
2402 hw->mac.autoneg = true;
2403 hw->phy.autoneg_advertised = 0x2f;
2405 hw->fc.requested_mode = e1000_fc_default;
2406 hw->fc.current_mode = e1000_fc_default;
2408 igb_validate_mdi_setting(hw);
2410 /* By default, support wake on port A */
2411 if (hw->bus.func == 0)
2412 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2414 /* Check the NVM for wake support on non-port A ports */
2415 if (hw->mac.type >= e1000_82580)
2416 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A +
2417 NVM_82580_LAN_FUNC_OFFSET(hw->bus.func), 1,
2419 else if (hw->bus.func == 1)
2420 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
2422 if (eeprom_data & IGB_EEPROM_APME)
2423 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2425 /* now that we have the eeprom settings, apply the special cases where
2426 * the eeprom may be wrong or the board simply won't support wake on
2427 * lan on a particular port
2429 switch (pdev->device) {
2430 case E1000_DEV_ID_82575GB_QUAD_COPPER:
2431 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2433 case E1000_DEV_ID_82575EB_FIBER_SERDES:
2434 case E1000_DEV_ID_82576_FIBER:
2435 case E1000_DEV_ID_82576_SERDES:
2436 /* Wake events only supported on port A for dual fiber
2437 * regardless of eeprom setting
2439 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
2440 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2442 case E1000_DEV_ID_82576_QUAD_COPPER:
2443 case E1000_DEV_ID_82576_QUAD_COPPER_ET2:
2444 /* if quad port adapter, disable WoL on all but port A */
2445 if (global_quad_port_a != 0)
2446 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2448 adapter->flags |= IGB_FLAG_QUAD_PORT_A;
2449 /* Reset for multiple quad port adapters */
2450 if (++global_quad_port_a == 4)
2451 global_quad_port_a = 0;
2454 /* If the device can't wake, don't set software support */
2455 if (!device_can_wakeup(&adapter->pdev->dev))
2456 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2459 /* initialize the wol settings based on the eeprom settings */
2460 if (adapter->flags & IGB_FLAG_WOL_SUPPORTED)
2461 adapter->wol |= E1000_WUFC_MAG;
2463 /* Some vendors want WoL disabled by default, but still supported */
2464 if ((hw->mac.type == e1000_i350) &&
2465 (pdev->subsystem_vendor == PCI_VENDOR_ID_HP)) {
2466 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2470 device_set_wakeup_enable(&adapter->pdev->dev,
2471 adapter->flags & IGB_FLAG_WOL_SUPPORTED);
2473 /* reset the hardware with the new settings */
2476 /* Init the I2C interface */
2477 err = igb_init_i2c(adapter);
2479 dev_err(&pdev->dev, "failed to init i2c interface\n");
2483 /* let the f/w know that the h/w is now under the control of the
2485 igb_get_hw_control(adapter);
2487 strcpy(netdev->name, "eth%d");
2488 err = register_netdev(netdev);
2492 /* carrier off reporting is important to ethtool even BEFORE open */
2493 netif_carrier_off(netdev);
2495 #ifdef CONFIG_IGB_DCA
2496 if (dca_add_requester(&pdev->dev) == 0) {
2497 adapter->flags |= IGB_FLAG_DCA_ENABLED;
2498 dev_info(&pdev->dev, "DCA enabled\n");
2499 igb_setup_dca(adapter);
2503 #ifdef CONFIG_IGB_HWMON
2504 /* Initialize the thermal sensor on i350 devices. */
2505 if (hw->mac.type == e1000_i350 && hw->bus.func == 0) {
2508 /* Read the NVM to determine if this i350 device supports an
2509 * external thermal sensor.
2511 hw->nvm.ops.read(hw, NVM_ETS_CFG, 1, &ets_word);
2512 if (ets_word != 0x0000 && ets_word != 0xFFFF)
2513 adapter->ets = true;
2515 adapter->ets = false;
2516 if (igb_sysfs_init(adapter))
2518 "failed to allocate sysfs resources\n");
2520 adapter->ets = false;
2523 /* Check if Media Autosense is enabled */
2525 if (hw->dev_spec._82575.mas_capable)
2526 igb_init_mas(adapter);
2528 /* do hw tstamp init after resetting */
2529 igb_ptp_init(adapter);
2531 dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
2532 /* print bus type/speed/width info, not applicable to i354 */
2533 if (hw->mac.type != e1000_i354) {
2534 dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
2536 ((hw->bus.speed == e1000_bus_speed_2500) ? "2.5Gb/s" :
2537 (hw->bus.speed == e1000_bus_speed_5000) ? "5.0Gb/s" :
2539 ((hw->bus.width == e1000_bus_width_pcie_x4) ?
2541 (hw->bus.width == e1000_bus_width_pcie_x2) ?
2543 (hw->bus.width == e1000_bus_width_pcie_x1) ?
2544 "Width x1" : "unknown"), netdev->dev_addr);
2547 if ((hw->mac.type >= e1000_i210 ||
2548 igb_get_flash_presence_i210(hw))) {
2549 ret_val = igb_read_part_string(hw, part_str,
2550 E1000_PBANUM_LENGTH);
2552 ret_val = -E1000_ERR_INVM_VALUE_NOT_FOUND;
2556 strcpy(part_str, "Unknown");
2557 dev_info(&pdev->dev, "%s: PBA No: %s\n", netdev->name, part_str);
2558 dev_info(&pdev->dev,
2559 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
2560 (adapter->flags & IGB_FLAG_HAS_MSIX) ? "MSI-X" :
2561 (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
2562 adapter->num_rx_queues, adapter->num_tx_queues);
2563 switch (hw->mac.type) {
2567 igb_set_eee_i350(hw);
2570 if (hw->phy.media_type == e1000_media_type_copper) {
2571 if ((rd32(E1000_CTRL_EXT) &
2572 E1000_CTRL_EXT_LINK_MODE_SGMII))
2573 igb_set_eee_i354(hw);
2580 pm_runtime_put_noidle(&pdev->dev);
2584 igb_release_hw_control(adapter);
2585 memset(&adapter->i2c_adap, 0, sizeof(adapter->i2c_adap));
2587 if (!igb_check_reset_block(hw))
2590 if (hw->flash_address)
2591 iounmap(hw->flash_address);
2593 igb_clear_interrupt_scheme(adapter);
2594 iounmap(hw->hw_addr);
2596 free_netdev(netdev);
2598 pci_release_selected_regions(pdev,
2599 pci_select_bars(pdev, IORESOURCE_MEM));
2602 pci_disable_device(pdev);
2606 #ifdef CONFIG_PCI_IOV
2607 static int igb_disable_sriov(struct pci_dev *pdev)
2609 struct net_device *netdev = pci_get_drvdata(pdev);
2610 struct igb_adapter *adapter = netdev_priv(netdev);
2611 struct e1000_hw *hw = &adapter->hw;
2613 /* reclaim resources allocated to VFs */
2614 if (adapter->vf_data) {
2615 /* disable iov and allow time for transactions to clear */
2616 if (pci_vfs_assigned(pdev)) {
2617 dev_warn(&pdev->dev,
2618 "Cannot deallocate SR-IOV virtual functions while they are assigned - VFs will not be deallocated\n");
2621 pci_disable_sriov(pdev);
2625 kfree(adapter->vf_data);
2626 adapter->vf_data = NULL;
2627 adapter->vfs_allocated_count = 0;
2628 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
2631 dev_info(&pdev->dev, "IOV Disabled\n");
2633 /* Re-enable DMA Coalescing flag since IOV is turned off */
2634 adapter->flags |= IGB_FLAG_DMAC;
2640 static int igb_enable_sriov(struct pci_dev *pdev, int num_vfs)
2642 struct net_device *netdev = pci_get_drvdata(pdev);
2643 struct igb_adapter *adapter = netdev_priv(netdev);
2644 int old_vfs = pci_num_vf(pdev);
2648 if (!(adapter->flags & IGB_FLAG_HAS_MSIX) || num_vfs > 7) {
2656 dev_info(&pdev->dev, "%d pre-allocated VFs found - override max_vfs setting of %d\n",
2658 adapter->vfs_allocated_count = old_vfs;
2660 adapter->vfs_allocated_count = num_vfs;
2662 adapter->vf_data = kcalloc(adapter->vfs_allocated_count,
2663 sizeof(struct vf_data_storage), GFP_KERNEL);
2665 /* if allocation failed then we do not support SR-IOV */
2666 if (!adapter->vf_data) {
2667 adapter->vfs_allocated_count = 0;
2669 "Unable to allocate memory for VF Data Storage\n");
2674 /* only call pci_enable_sriov() if no VFs are allocated already */
2676 err = pci_enable_sriov(pdev, adapter->vfs_allocated_count);
2680 dev_info(&pdev->dev, "%d VFs allocated\n",
2681 adapter->vfs_allocated_count);
2682 for (i = 0; i < adapter->vfs_allocated_count; i++)
2683 igb_vf_configure(adapter, i);
2685 /* DMA Coalescing is not supported in IOV mode. */
2686 adapter->flags &= ~IGB_FLAG_DMAC;
2690 kfree(adapter->vf_data);
2691 adapter->vf_data = NULL;
2692 adapter->vfs_allocated_count = 0;
2699 * igb_remove_i2c - Cleanup I2C interface
2700 * @adapter: pointer to adapter structure
2702 static void igb_remove_i2c(struct igb_adapter *adapter)
2704 /* free the adapter bus structure */
2705 i2c_del_adapter(&adapter->i2c_adap);
2709 * igb_remove - Device Removal Routine
2710 * @pdev: PCI device information struct
2712 * igb_remove is called by the PCI subsystem to alert the driver
2713 * that it should release a PCI device. The could be caused by a
2714 * Hot-Plug event, or because the driver is going to be removed from
2717 static void igb_remove(struct pci_dev *pdev)
2719 struct net_device *netdev = pci_get_drvdata(pdev);
2720 struct igb_adapter *adapter = netdev_priv(netdev);
2721 struct e1000_hw *hw = &adapter->hw;
2723 pm_runtime_get_noresume(&pdev->dev);
2724 #ifdef CONFIG_IGB_HWMON
2725 igb_sysfs_exit(adapter);
2727 igb_remove_i2c(adapter);
2728 igb_ptp_stop(adapter);
2729 /* The watchdog timer may be rescheduled, so explicitly
2730 * disable watchdog from being rescheduled.
2732 set_bit(__IGB_DOWN, &adapter->state);
2733 del_timer_sync(&adapter->watchdog_timer);
2734 del_timer_sync(&adapter->phy_info_timer);
2736 cancel_work_sync(&adapter->reset_task);
2737 cancel_work_sync(&adapter->watchdog_task);
2739 #ifdef CONFIG_IGB_DCA
2740 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
2741 dev_info(&pdev->dev, "DCA disabled\n");
2742 dca_remove_requester(&pdev->dev);
2743 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
2744 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
2748 /* Release control of h/w to f/w. If f/w is AMT enabled, this
2749 * would have already happened in close and is redundant.
2751 igb_release_hw_control(adapter);
2753 unregister_netdev(netdev);
2755 igb_clear_interrupt_scheme(adapter);
2757 #ifdef CONFIG_PCI_IOV
2758 igb_disable_sriov(pdev);
2761 iounmap(hw->hw_addr);
2762 if (hw->flash_address)
2763 iounmap(hw->flash_address);
2764 pci_release_selected_regions(pdev,
2765 pci_select_bars(pdev, IORESOURCE_MEM));
2767 kfree(adapter->shadow_vfta);
2768 free_netdev(netdev);
2770 pci_disable_pcie_error_reporting(pdev);
2772 pci_disable_device(pdev);
2776 * igb_probe_vfs - Initialize vf data storage and add VFs to pci config space
2777 * @adapter: board private structure to initialize
2779 * This function initializes the vf specific data storage and then attempts to
2780 * allocate the VFs. The reason for ordering it this way is because it is much
2781 * mor expensive time wise to disable SR-IOV than it is to allocate and free
2782 * the memory for the VFs.
2784 static void igb_probe_vfs(struct igb_adapter *adapter)
2786 #ifdef CONFIG_PCI_IOV
2787 struct pci_dev *pdev = adapter->pdev;
2788 struct e1000_hw *hw = &adapter->hw;
2790 /* Virtualization features not supported on i210 family. */
2791 if ((hw->mac.type == e1000_i210) || (hw->mac.type == e1000_i211))
2794 pci_sriov_set_totalvfs(pdev, 7);
2795 igb_pci_enable_sriov(pdev, max_vfs);
2797 #endif /* CONFIG_PCI_IOV */
2800 static void igb_init_queue_configuration(struct igb_adapter *adapter)
2802 struct e1000_hw *hw = &adapter->hw;
2805 /* Determine the maximum number of RSS queues supported. */
2806 switch (hw->mac.type) {
2808 max_rss_queues = IGB_MAX_RX_QUEUES_I211;
2812 max_rss_queues = IGB_MAX_RX_QUEUES_82575;
2815 /* I350 cannot do RSS and SR-IOV at the same time */
2816 if (!!adapter->vfs_allocated_count) {
2822 if (!!adapter->vfs_allocated_count) {
2830 max_rss_queues = IGB_MAX_RX_QUEUES;
2834 adapter->rss_queues = min_t(u32, max_rss_queues, num_online_cpus());
2836 /* Determine if we need to pair queues. */
2837 switch (hw->mac.type) {
2840 /* Device supports enough interrupts without queue pairing. */
2843 /* If VFs are going to be allocated with RSS queues then we
2844 * should pair the queues in order to conserve interrupts due
2845 * to limited supply.
2847 if ((adapter->rss_queues > 1) &&
2848 (adapter->vfs_allocated_count > 6))
2849 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
2856 /* If rss_queues > half of max_rss_queues, pair the queues in
2857 * order to conserve interrupts due to limited supply.
2859 if (adapter->rss_queues > (max_rss_queues / 2))
2860 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
2866 * igb_sw_init - Initialize general software structures (struct igb_adapter)
2867 * @adapter: board private structure to initialize
2869 * igb_sw_init initializes the Adapter private data structure.
2870 * Fields are initialized based on PCI device information and
2871 * OS network device settings (MTU size).
2873 static int igb_sw_init(struct igb_adapter *adapter)
2875 struct e1000_hw *hw = &adapter->hw;
2876 struct net_device *netdev = adapter->netdev;
2877 struct pci_dev *pdev = adapter->pdev;
2879 pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
2881 /* set default ring sizes */
2882 adapter->tx_ring_count = IGB_DEFAULT_TXD;
2883 adapter->rx_ring_count = IGB_DEFAULT_RXD;
2885 /* set default ITR values */
2886 adapter->rx_itr_setting = IGB_DEFAULT_ITR;
2887 adapter->tx_itr_setting = IGB_DEFAULT_ITR;
2889 /* set default work limits */
2890 adapter->tx_work_limit = IGB_DEFAULT_TX_WORK;
2892 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN +
2894 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2896 spin_lock_init(&adapter->stats64_lock);
2897 #ifdef CONFIG_PCI_IOV
2898 switch (hw->mac.type) {
2902 dev_warn(&pdev->dev,
2903 "Maximum of 7 VFs per PF, using max\n");
2904 max_vfs = adapter->vfs_allocated_count = 7;
2906 adapter->vfs_allocated_count = max_vfs;
2907 if (adapter->vfs_allocated_count)
2908 dev_warn(&pdev->dev,
2909 "Enabling SR-IOV VFs using the module parameter is deprecated - please use the pci sysfs interface.\n");
2914 #endif /* CONFIG_PCI_IOV */
2916 igb_init_queue_configuration(adapter);
2918 /* Setup and initialize a copy of the hw vlan table array */
2919 adapter->shadow_vfta = kcalloc(E1000_VLAN_FILTER_TBL_SIZE, sizeof(u32),
2922 /* This call may decrease the number of queues */
2923 if (igb_init_interrupt_scheme(adapter, true)) {
2924 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
2928 igb_probe_vfs(adapter);
2930 /* Explicitly disable IRQ since the NIC can be in any state. */
2931 igb_irq_disable(adapter);
2933 if (hw->mac.type >= e1000_i350)
2934 adapter->flags &= ~IGB_FLAG_DMAC;
2936 set_bit(__IGB_DOWN, &adapter->state);
2941 * igb_open - Called when a network interface is made active
2942 * @netdev: network interface device structure
2944 * Returns 0 on success, negative value on failure
2946 * The open entry point is called when a network interface is made
2947 * active by the system (IFF_UP). At this point all resources needed
2948 * for transmit and receive operations are allocated, the interrupt
2949 * handler is registered with the OS, the watchdog timer is started,
2950 * and the stack is notified that the interface is ready.
2952 static int __igb_open(struct net_device *netdev, bool resuming)
2954 struct igb_adapter *adapter = netdev_priv(netdev);
2955 struct e1000_hw *hw = &adapter->hw;
2956 struct pci_dev *pdev = adapter->pdev;
2960 /* disallow open during test */
2961 if (test_bit(__IGB_TESTING, &adapter->state)) {
2967 pm_runtime_get_sync(&pdev->dev);
2969 netif_carrier_off(netdev);
2971 /* allocate transmit descriptors */
2972 err = igb_setup_all_tx_resources(adapter);
2976 /* allocate receive descriptors */
2977 err = igb_setup_all_rx_resources(adapter);
2981 igb_power_up_link(adapter);
2983 /* before we allocate an interrupt, we must be ready to handle it.
2984 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
2985 * as soon as we call pci_request_irq, so we have to setup our
2986 * clean_rx handler before we do so.
2988 igb_configure(adapter);
2990 err = igb_request_irq(adapter);
2994 /* Notify the stack of the actual queue counts. */
2995 err = netif_set_real_num_tx_queues(adapter->netdev,
2996 adapter->num_tx_queues);
2998 goto err_set_queues;
3000 err = netif_set_real_num_rx_queues(adapter->netdev,
3001 adapter->num_rx_queues);
3003 goto err_set_queues;
3005 /* From here on the code is the same as igb_up() */
3006 clear_bit(__IGB_DOWN, &adapter->state);
3008 for (i = 0; i < adapter->num_q_vectors; i++)
3009 napi_enable(&(adapter->q_vector[i]->napi));
3011 /* Clear any pending interrupts. */
3014 igb_irq_enable(adapter);
3016 /* notify VFs that reset has been completed */
3017 if (adapter->vfs_allocated_count) {
3018 u32 reg_data = rd32(E1000_CTRL_EXT);
3019 reg_data |= E1000_CTRL_EXT_PFRSTD;
3020 wr32(E1000_CTRL_EXT, reg_data);
3023 netif_tx_start_all_queues(netdev);
3026 pm_runtime_put(&pdev->dev);
3028 /* start the watchdog. */
3029 hw->mac.get_link_status = 1;
3030 schedule_work(&adapter->watchdog_task);
3035 igb_free_irq(adapter);
3037 igb_release_hw_control(adapter);
3038 igb_power_down_link(adapter);
3039 igb_free_all_rx_resources(adapter);
3041 igb_free_all_tx_resources(adapter);
3045 pm_runtime_put(&pdev->dev);
3050 static int igb_open(struct net_device *netdev)
3052 return __igb_open(netdev, false);
3056 * igb_close - Disables a network interface
3057 * @netdev: network interface device structure
3059 * Returns 0, this is not allowed to fail
3061 * The close entry point is called when an interface is de-activated
3062 * by the OS. The hardware is still under the driver's control, but
3063 * needs to be disabled. A global MAC reset is issued to stop the
3064 * hardware, and all transmit and receive resources are freed.
3066 static int __igb_close(struct net_device *netdev, bool suspending)
3068 struct igb_adapter *adapter = netdev_priv(netdev);
3069 struct pci_dev *pdev = adapter->pdev;
3071 WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
3074 pm_runtime_get_sync(&pdev->dev);
3077 igb_free_irq(adapter);
3079 igb_free_all_tx_resources(adapter);
3080 igb_free_all_rx_resources(adapter);
3083 pm_runtime_put_sync(&pdev->dev);
3087 static int igb_close(struct net_device *netdev)
3089 return __igb_close(netdev, false);
3093 * igb_setup_tx_resources - allocate Tx resources (Descriptors)
3094 * @tx_ring: tx descriptor ring (for a specific queue) to setup
3096 * Return 0 on success, negative on failure
3098 int igb_setup_tx_resources(struct igb_ring *tx_ring)
3100 struct device *dev = tx_ring->dev;
3103 size = sizeof(struct igb_tx_buffer) * tx_ring->count;
3105 tx_ring->tx_buffer_info = vzalloc(size);
3106 if (!tx_ring->tx_buffer_info)
3109 /* round up to nearest 4K */
3110 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
3111 tx_ring->size = ALIGN(tx_ring->size, 4096);
3113 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
3114 &tx_ring->dma, GFP_KERNEL);
3118 tx_ring->next_to_use = 0;
3119 tx_ring->next_to_clean = 0;
3124 vfree(tx_ring->tx_buffer_info);
3125 tx_ring->tx_buffer_info = NULL;
3126 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
3131 * igb_setup_all_tx_resources - wrapper to allocate Tx resources
3132 * (Descriptors) for all queues
3133 * @adapter: board private structure
3135 * Return 0 on success, negative on failure
3137 static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
3139 struct pci_dev *pdev = adapter->pdev;
3142 for (i = 0; i < adapter->num_tx_queues; i++) {
3143 err = igb_setup_tx_resources(adapter->tx_ring[i]);
3146 "Allocation for Tx Queue %u failed\n", i);
3147 for (i--; i >= 0; i--)
3148 igb_free_tx_resources(adapter->tx_ring[i]);
3157 * igb_setup_tctl - configure the transmit control registers
3158 * @adapter: Board private structure
3160 void igb_setup_tctl(struct igb_adapter *adapter)
3162 struct e1000_hw *hw = &adapter->hw;
3165 /* disable queue 0 which is enabled by default on 82575 and 82576 */
3166 wr32(E1000_TXDCTL(0), 0);
3168 /* Program the Transmit Control Register */
3169 tctl = rd32(E1000_TCTL);
3170 tctl &= ~E1000_TCTL_CT;
3171 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
3172 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
3174 igb_config_collision_dist(hw);
3176 /* Enable transmits */
3177 tctl |= E1000_TCTL_EN;
3179 wr32(E1000_TCTL, tctl);
3183 * igb_configure_tx_ring - Configure transmit ring after Reset
3184 * @adapter: board private structure
3185 * @ring: tx ring to configure
3187 * Configure a transmit ring after a reset.
3189 void igb_configure_tx_ring(struct igb_adapter *adapter,
3190 struct igb_ring *ring)
3192 struct e1000_hw *hw = &adapter->hw;
3194 u64 tdba = ring->dma;
3195 int reg_idx = ring->reg_idx;
3197 /* disable the queue */
3198 wr32(E1000_TXDCTL(reg_idx), 0);
3202 wr32(E1000_TDLEN(reg_idx),
3203 ring->count * sizeof(union e1000_adv_tx_desc));
3204 wr32(E1000_TDBAL(reg_idx),
3205 tdba & 0x00000000ffffffffULL);
3206 wr32(E1000_TDBAH(reg_idx), tdba >> 32);
3208 ring->tail = hw->hw_addr + E1000_TDT(reg_idx);
3209 wr32(E1000_TDH(reg_idx), 0);
3210 writel(0, ring->tail);
3212 txdctl |= IGB_TX_PTHRESH;
3213 txdctl |= IGB_TX_HTHRESH << 8;
3214 txdctl |= IGB_TX_WTHRESH << 16;
3216 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
3217 wr32(E1000_TXDCTL(reg_idx), txdctl);
3221 * igb_configure_tx - Configure transmit Unit after Reset
3222 * @adapter: board private structure
3224 * Configure the Tx unit of the MAC after a reset.
3226 static void igb_configure_tx(struct igb_adapter *adapter)
3230 for (i = 0; i < adapter->num_tx_queues; i++)
3231 igb_configure_tx_ring(adapter, adapter->tx_ring[i]);
3235 * igb_setup_rx_resources - allocate Rx resources (Descriptors)
3236 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
3238 * Returns 0 on success, negative on failure
3240 int igb_setup_rx_resources(struct igb_ring *rx_ring)
3242 struct device *dev = rx_ring->dev;
3245 size = sizeof(struct igb_rx_buffer) * rx_ring->count;
3247 rx_ring->rx_buffer_info = vzalloc(size);
3248 if (!rx_ring->rx_buffer_info)
3251 /* Round up to nearest 4K */
3252 rx_ring->size = rx_ring->count * sizeof(union e1000_adv_rx_desc);
3253 rx_ring->size = ALIGN(rx_ring->size, 4096);
3255 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
3256 &rx_ring->dma, GFP_KERNEL);
3260 rx_ring->next_to_alloc = 0;
3261 rx_ring->next_to_clean = 0;
3262 rx_ring->next_to_use = 0;
3267 vfree(rx_ring->rx_buffer_info);
3268 rx_ring->rx_buffer_info = NULL;
3269 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
3274 * igb_setup_all_rx_resources - wrapper to allocate Rx resources
3275 * (Descriptors) for all queues
3276 * @adapter: board private structure
3278 * Return 0 on success, negative on failure
3280 static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
3282 struct pci_dev *pdev = adapter->pdev;
3285 for (i = 0; i < adapter->num_rx_queues; i++) {
3286 err = igb_setup_rx_resources(adapter->rx_ring[i]);
3289 "Allocation for Rx Queue %u failed\n", i);
3290 for (i--; i >= 0; i--)
3291 igb_free_rx_resources(adapter->rx_ring[i]);
3300 * igb_setup_mrqc - configure the multiple receive queue control registers
3301 * @adapter: Board private structure
3303 static void igb_setup_mrqc(struct igb_adapter *adapter)
3305 struct e1000_hw *hw = &adapter->hw;
3307 u32 j, num_rx_queues;
3308 static const u32 rsskey[10] = { 0xDA565A6D, 0xC20E5B25, 0x3D256741,
3309 0xB08FA343, 0xCB2BCAD0, 0xB4307BAE,
3310 0xA32DCB77, 0x0CF23080, 0x3BB7426A,
3313 /* Fill out hash function seeds */
3314 for (j = 0; j < 10; j++)
3315 wr32(E1000_RSSRK(j), rsskey[j]);
3317 num_rx_queues = adapter->rss_queues;
3319 switch (hw->mac.type) {
3321 /* 82576 supports 2 RSS queues for SR-IOV */
3322 if (adapter->vfs_allocated_count)
3329 if (adapter->rss_indir_tbl_init != num_rx_queues) {
3330 for (j = 0; j < IGB_RETA_SIZE; j++)
3331 adapter->rss_indir_tbl[j] = (j * num_rx_queues) / IGB_RETA_SIZE;
3332 adapter->rss_indir_tbl_init = num_rx_queues;
3334 igb_write_rss_indir_tbl(adapter);
3336 /* Disable raw packet checksumming so that RSS hash is placed in
3337 * descriptor on writeback. No need to enable TCP/UDP/IP checksum
3338 * offloads as they are enabled by default
3340 rxcsum = rd32(E1000_RXCSUM);
3341 rxcsum |= E1000_RXCSUM_PCSD;
3343 if (adapter->hw.mac.type >= e1000_82576)
3344 /* Enable Receive Checksum Offload for SCTP */
3345 rxcsum |= E1000_RXCSUM_CRCOFL;
3347 /* Don't need to set TUOFL or IPOFL, they default to 1 */
3348 wr32(E1000_RXCSUM, rxcsum);
3350 /* Generate RSS hash based on packet types, TCP/UDP
3351 * port numbers and/or IPv4/v6 src and dst addresses
3353 mrqc = E1000_MRQC_RSS_FIELD_IPV4 |
3354 E1000_MRQC_RSS_FIELD_IPV4_TCP |
3355 E1000_MRQC_RSS_FIELD_IPV6 |
3356 E1000_MRQC_RSS_FIELD_IPV6_TCP |
3357 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX;
3359 if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV4_UDP)
3360 mrqc |= E1000_MRQC_RSS_FIELD_IPV4_UDP;
3361 if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV6_UDP)
3362 mrqc |= E1000_MRQC_RSS_FIELD_IPV6_UDP;
3364 /* If VMDq is enabled then we set the appropriate mode for that, else
3365 * we default to RSS so that an RSS hash is calculated per packet even
3366 * if we are only using one queue
3368 if (adapter->vfs_allocated_count) {
3369 if (hw->mac.type > e1000_82575) {
3370 /* Set the default pool for the PF's first queue */
3371 u32 vtctl = rd32(E1000_VT_CTL);
3372 vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK |
3373 E1000_VT_CTL_DISABLE_DEF_POOL);
3374 vtctl |= adapter->vfs_allocated_count <<
3375 E1000_VT_CTL_DEFAULT_POOL_SHIFT;
3376 wr32(E1000_VT_CTL, vtctl);
3378 if (adapter->rss_queues > 1)
3379 mrqc |= E1000_MRQC_ENABLE_VMDQ_RSS_2Q;
3381 mrqc |= E1000_MRQC_ENABLE_VMDQ;
3383 if (hw->mac.type != e1000_i211)
3384 mrqc |= E1000_MRQC_ENABLE_RSS_4Q;
3386 igb_vmm_control(adapter);
3388 wr32(E1000_MRQC, mrqc);
3392 * igb_setup_rctl - configure the receive control registers
3393 * @adapter: Board private structure
3395 void igb_setup_rctl(struct igb_adapter *adapter)
3397 struct e1000_hw *hw = &adapter->hw;
3400 rctl = rd32(E1000_RCTL);
3402 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
3403 rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
3405 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
3406 (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
3408 /* enable stripping of CRC. It's unlikely this will break BMC
3409 * redirection as it did with e1000. Newer features require
3410 * that the HW strips the CRC.
3412 rctl |= E1000_RCTL_SECRC;
3414 /* disable store bad packets and clear size bits. */
3415 rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
3417 /* enable LPE to prevent packets larger than max_frame_size */
3418 rctl |= E1000_RCTL_LPE;
3420 /* disable queue 0 to prevent tail write w/o re-config */
3421 wr32(E1000_RXDCTL(0), 0);
3423 /* Attention!!! For SR-IOV PF driver operations you must enable
3424 * queue drop for all VF and PF queues to prevent head of line blocking
3425 * if an un-trusted VF does not provide descriptors to hardware.
3427 if (adapter->vfs_allocated_count) {
3428 /* set all queue drop enable bits */
3429 wr32(E1000_QDE, ALL_QUEUES);
3432 /* This is useful for sniffing bad packets. */
3433 if (adapter->netdev->features & NETIF_F_RXALL) {
3434 /* UPE and MPE will be handled by normal PROMISC logic
3435 * in e1000e_set_rx_mode
3437 rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
3438 E1000_RCTL_BAM | /* RX All Bcast Pkts */
3439 E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
3441 rctl &= ~(E1000_RCTL_VFE | /* Disable VLAN filter */
3442 E1000_RCTL_DPF | /* Allow filtered pause */
3443 E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
3444 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
3445 * and that breaks VLANs.
3449 wr32(E1000_RCTL, rctl);
3452 static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
3455 struct e1000_hw *hw = &adapter->hw;
3458 /* if it isn't the PF check to see if VFs are enabled and
3459 * increase the size to support vlan tags
3461 if (vfn < adapter->vfs_allocated_count &&
3462 adapter->vf_data[vfn].vlans_enabled)
3463 size += VLAN_TAG_SIZE;
3465 vmolr = rd32(E1000_VMOLR(vfn));
3466 vmolr &= ~E1000_VMOLR_RLPML_MASK;
3467 vmolr |= size | E1000_VMOLR_LPE;
3468 wr32(E1000_VMOLR(vfn), vmolr);
3474 * igb_rlpml_set - set maximum receive packet size
3475 * @adapter: board private structure
3477 * Configure maximum receivable packet size.
3479 static void igb_rlpml_set(struct igb_adapter *adapter)
3481 u32 max_frame_size = adapter->max_frame_size;
3482 struct e1000_hw *hw = &adapter->hw;
3483 u16 pf_id = adapter->vfs_allocated_count;
3486 igb_set_vf_rlpml(adapter, max_frame_size, pf_id);
3487 /* If we're in VMDQ or SR-IOV mode, then set global RLPML
3488 * to our max jumbo frame size, in case we need to enable
3489 * jumbo frames on one of the rings later.
3490 * This will not pass over-length frames into the default
3491 * queue because it's gated by the VMOLR.RLPML.
3493 max_frame_size = MAX_JUMBO_FRAME_SIZE;
3496 wr32(E1000_RLPML, max_frame_size);
3499 static inline void igb_set_vmolr(struct igb_adapter *adapter,
3502 struct e1000_hw *hw = &adapter->hw;
3505 /* This register exists only on 82576 and newer so if we are older then
3506 * we should exit and do nothing
3508 if (hw->mac.type < e1000_82576)
3511 vmolr = rd32(E1000_VMOLR(vfn));
3512 vmolr |= E1000_VMOLR_STRVLAN; /* Strip vlan tags */
3514 vmolr |= E1000_VMOLR_AUPE; /* Accept untagged packets */
3516 vmolr &= ~(E1000_VMOLR_AUPE); /* Tagged packets ONLY */
3518 /* clear all bits that might not be set */
3519 vmolr &= ~(E1000_VMOLR_BAM | E1000_VMOLR_RSSE);
3521 if (adapter->rss_queues > 1 && vfn == adapter->vfs_allocated_count)
3522 vmolr |= E1000_VMOLR_RSSE; /* enable RSS */
3523 /* for VMDq only allow the VFs and pool 0 to accept broadcast and
3526 if (vfn <= adapter->vfs_allocated_count)
3527 vmolr |= E1000_VMOLR_BAM; /* Accept broadcast */
3529 wr32(E1000_VMOLR(vfn), vmolr);
3533 * igb_configure_rx_ring - Configure a receive ring after Reset
3534 * @adapter: board private structure
3535 * @ring: receive ring to be configured
3537 * Configure the Rx unit of the MAC after a reset.
3539 void igb_configure_rx_ring(struct igb_adapter *adapter,
3540 struct igb_ring *ring)
3542 struct e1000_hw *hw = &adapter->hw;
3543 u64 rdba = ring->dma;
3544 int reg_idx = ring->reg_idx;
3545 u32 srrctl = 0, rxdctl = 0;
3547 /* disable the queue */
3548 wr32(E1000_RXDCTL(reg_idx), 0);
3550 /* Set DMA base address registers */
3551 wr32(E1000_RDBAL(reg_idx),
3552 rdba & 0x00000000ffffffffULL);
3553 wr32(E1000_RDBAH(reg_idx), rdba >> 32);
3554 wr32(E1000_RDLEN(reg_idx),
3555 ring->count * sizeof(union e1000_adv_rx_desc));
3557 /* initialize head and tail */
3558 ring->tail = hw->hw_addr + E1000_RDT(reg_idx);
3559 wr32(E1000_RDH(reg_idx), 0);
3560 writel(0, ring->tail);
3562 /* set descriptor configuration */
3563 srrctl = IGB_RX_HDR_LEN << E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
3564 srrctl |= IGB_RX_BUFSZ >> E1000_SRRCTL_BSIZEPKT_SHIFT;
3565 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
3566 if (hw->mac.type >= e1000_82580)
3567 srrctl |= E1000_SRRCTL_TIMESTAMP;
3568 /* Only set Drop Enable if we are supporting multiple queues */
3569 if (adapter->vfs_allocated_count || adapter->num_rx_queues > 1)
3570 srrctl |= E1000_SRRCTL_DROP_EN;
3572 wr32(E1000_SRRCTL(reg_idx), srrctl);
3574 /* set filtering for VMDQ pools */
3575 igb_set_vmolr(adapter, reg_idx & 0x7, true);
3577 rxdctl |= IGB_RX_PTHRESH;
3578 rxdctl |= IGB_RX_HTHRESH << 8;
3579 rxdctl |= IGB_RX_WTHRESH << 16;
3581 /* enable receive descriptor fetching */
3582 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
3583 wr32(E1000_RXDCTL(reg_idx), rxdctl);
3587 * igb_configure_rx - Configure receive Unit after Reset
3588 * @adapter: board private structure
3590 * Configure the Rx unit of the MAC after a reset.
3592 static void igb_configure_rx(struct igb_adapter *adapter)
3596 /* set UTA to appropriate mode */
3597 igb_set_uta(adapter);
3599 /* set the correct pool for the PF default MAC address in entry 0 */
3600 igb_rar_set_qsel(adapter, adapter->hw.mac.addr, 0,
3601 adapter->vfs_allocated_count);
3603 /* Setup the HW Rx Head and Tail Descriptor Pointers and
3604 * the Base and Length of the Rx Descriptor Ring
3606 for (i = 0; i < adapter->num_rx_queues; i++)
3607 igb_configure_rx_ring(adapter, adapter->rx_ring[i]);
3611 * igb_free_tx_resources - Free Tx Resources per Queue
3612 * @tx_ring: Tx descriptor ring for a specific queue
3614 * Free all transmit software resources
3616 void igb_free_tx_resources(struct igb_ring *tx_ring)
3618 igb_clean_tx_ring(tx_ring);
3620 vfree(tx_ring->tx_buffer_info);
3621 tx_ring->tx_buffer_info = NULL;
3623 /* if not set, then don't free */
3627 dma_free_coherent(tx_ring->dev, tx_ring->size,
3628 tx_ring->desc, tx_ring->dma);
3630 tx_ring->desc = NULL;
3634 * igb_free_all_tx_resources - Free Tx Resources for All Queues
3635 * @adapter: board private structure
3637 * Free all transmit software resources
3639 static void igb_free_all_tx_resources(struct igb_adapter *adapter)
3643 for (i = 0; i < adapter->num_tx_queues; i++)
3644 igb_free_tx_resources(adapter->tx_ring[i]);
3647 void igb_unmap_and_free_tx_resource(struct igb_ring *ring,
3648 struct igb_tx_buffer *tx_buffer)
3650 if (tx_buffer->skb) {
3651 dev_kfree_skb_any(tx_buffer->skb);
3652 if (dma_unmap_len(tx_buffer, len))
3653 dma_unmap_single(ring->dev,
3654 dma_unmap_addr(tx_buffer, dma),
3655 dma_unmap_len(tx_buffer, len),
3657 } else if (dma_unmap_len(tx_buffer, len)) {
3658 dma_unmap_page(ring->dev,
3659 dma_unmap_addr(tx_buffer, dma),
3660 dma_unmap_len(tx_buffer, len),
3663 tx_buffer->next_to_watch = NULL;
3664 tx_buffer->skb = NULL;
3665 dma_unmap_len_set(tx_buffer, len, 0);
3666 /* buffer_info must be completely set up in the transmit path */
3670 * igb_clean_tx_ring - Free Tx Buffers
3671 * @tx_ring: ring to be cleaned
3673 static void igb_clean_tx_ring(struct igb_ring *tx_ring)
3675 struct igb_tx_buffer *buffer_info;
3679 if (!tx_ring->tx_buffer_info)
3681 /* Free all the Tx ring sk_buffs */
3683 for (i = 0; i < tx_ring->count; i++) {
3684 buffer_info = &tx_ring->tx_buffer_info[i];
3685 igb_unmap_and_free_tx_resource(tx_ring, buffer_info);
3688 netdev_tx_reset_queue(txring_txq(tx_ring));
3690 size = sizeof(struct igb_tx_buffer) * tx_ring->count;
3691 memset(tx_ring->tx_buffer_info, 0, size);
3693 /* Zero out the descriptor ring */
3694 memset(tx_ring->desc, 0, tx_ring->size);
3696 tx_ring->next_to_use = 0;
3697 tx_ring->next_to_clean = 0;
3701 * igb_clean_all_tx_rings - Free Tx Buffers for all queues
3702 * @adapter: board private structure
3704 static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
3708 for (i = 0; i < adapter->num_tx_queues; i++)
3709 igb_clean_tx_ring(adapter->tx_ring[i]);
3713 * igb_free_rx_resources - Free Rx Resources
3714 * @rx_ring: ring to clean the resources from
3716 * Free all receive software resources
3718 void igb_free_rx_resources(struct igb_ring *rx_ring)
3720 igb_clean_rx_ring(rx_ring);
3722 vfree(rx_ring->rx_buffer_info);
3723 rx_ring->rx_buffer_info = NULL;
3725 /* if not set, then don't free */
3729 dma_free_coherent(rx_ring->dev, rx_ring->size,
3730 rx_ring->desc, rx_ring->dma);
3732 rx_ring->desc = NULL;
3736 * igb_free_all_rx_resources - Free Rx Resources for All Queues
3737 * @adapter: board private structure
3739 * Free all receive software resources
3741 static void igb_free_all_rx_resources(struct igb_adapter *adapter)
3745 for (i = 0; i < adapter->num_rx_queues; i++)
3746 igb_free_rx_resources(adapter->rx_ring[i]);
3750 * igb_clean_rx_ring - Free Rx Buffers per Queue
3751 * @rx_ring: ring to free buffers from
3753 static void igb_clean_rx_ring(struct igb_ring *rx_ring)
3759 dev_kfree_skb(rx_ring->skb);
3760 rx_ring->skb = NULL;
3762 if (!rx_ring->rx_buffer_info)
3765 /* Free all the Rx ring sk_buffs */
3766 for (i = 0; i < rx_ring->count; i++) {
3767 struct igb_rx_buffer *buffer_info = &rx_ring->rx_buffer_info[i];
3769 if (!buffer_info->page)
3772 dma_unmap_page(rx_ring->dev,
3776 __free_page(buffer_info->page);
3778 buffer_info->page = NULL;
3781 size = sizeof(struct igb_rx_buffer) * rx_ring->count;
3782 memset(rx_ring->rx_buffer_info, 0, size);
3784 /* Zero out the descriptor ring */
3785 memset(rx_ring->desc, 0, rx_ring->size);
3787 rx_ring->next_to_alloc = 0;
3788 rx_ring->next_to_clean = 0;
3789 rx_ring->next_to_use = 0;
3793 * igb_clean_all_rx_rings - Free Rx Buffers for all queues
3794 * @adapter: board private structure
3796 static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
3800 for (i = 0; i < adapter->num_rx_queues; i++)
3801 igb_clean_rx_ring(adapter->rx_ring[i]);
3805 * igb_set_mac - Change the Ethernet Address of the NIC
3806 * @netdev: network interface device structure
3807 * @p: pointer to an address structure
3809 * Returns 0 on success, negative on failure
3811 static int igb_set_mac(struct net_device *netdev, void *p)
3813 struct igb_adapter *adapter = netdev_priv(netdev);
3814 struct e1000_hw *hw = &adapter->hw;
3815 struct sockaddr *addr = p;
3817 if (!is_valid_ether_addr(addr->sa_data))
3818 return -EADDRNOTAVAIL;
3820 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3821 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
3823 /* set the correct pool for the new PF MAC address in entry 0 */
3824 igb_rar_set_qsel(adapter, hw->mac.addr, 0,
3825 adapter->vfs_allocated_count);
3831 * igb_write_mc_addr_list - write multicast addresses to MTA
3832 * @netdev: network interface device structure
3834 * Writes multicast address list to the MTA hash table.
3835 * Returns: -ENOMEM on failure
3836 * 0 on no addresses written
3837 * X on writing X addresses to MTA
3839 static int igb_write_mc_addr_list(struct net_device *netdev)
3841 struct igb_adapter *adapter = netdev_priv(netdev);
3842 struct e1000_hw *hw = &adapter->hw;
3843 struct netdev_hw_addr *ha;
3847 if (netdev_mc_empty(netdev)) {
3848 /* nothing to program, so clear mc list */
3849 igb_update_mc_addr_list(hw, NULL, 0);
3850 igb_restore_vf_multicasts(adapter);
3854 mta_list = kzalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
3858 /* The shared function expects a packed array of only addresses. */
3860 netdev_for_each_mc_addr(ha, netdev)
3861 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3863 igb_update_mc_addr_list(hw, mta_list, i);
3866 return netdev_mc_count(netdev);
3870 * igb_write_uc_addr_list - write unicast addresses to RAR table
3871 * @netdev: network interface device structure
3873 * Writes unicast address list to the RAR table.
3874 * Returns: -ENOMEM on failure/insufficient address space
3875 * 0 on no addresses written
3876 * X on writing X addresses to the RAR table
3878 static int igb_write_uc_addr_list(struct net_device *netdev)
3880 struct igb_adapter *adapter = netdev_priv(netdev);
3881 struct e1000_hw *hw = &adapter->hw;
3882 unsigned int vfn = adapter->vfs_allocated_count;
3883 unsigned int rar_entries = hw->mac.rar_entry_count - (vfn + 1);
3886 /* return ENOMEM indicating insufficient memory for addresses */
3887 if (netdev_uc_count(netdev) > rar_entries)
3890 if (!netdev_uc_empty(netdev) && rar_entries) {
3891 struct netdev_hw_addr *ha;
3893 netdev_for_each_uc_addr(ha, netdev) {
3896 igb_rar_set_qsel(adapter, ha->addr,
3902 /* write the addresses in reverse order to avoid write combining */
3903 for (; rar_entries > 0 ; rar_entries--) {
3904 wr32(E1000_RAH(rar_entries), 0);
3905 wr32(E1000_RAL(rar_entries), 0);
3913 * igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
3914 * @netdev: network interface device structure
3916 * The set_rx_mode entry point is called whenever the unicast or multicast
3917 * address lists or the network interface flags are updated. This routine is
3918 * responsible for configuring the hardware for proper unicast, multicast,
3919 * promiscuous mode, and all-multi behavior.
3921 static void igb_set_rx_mode(struct net_device *netdev)
3923 struct igb_adapter *adapter = netdev_priv(netdev);
3924 struct e1000_hw *hw = &adapter->hw;
3925 unsigned int vfn = adapter->vfs_allocated_count;
3926 u32 rctl, vmolr = 0;
3929 /* Check for Promiscuous and All Multicast modes */
3930 rctl = rd32(E1000_RCTL);
3932 /* clear the effected bits */
3933 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE | E1000_RCTL_VFE);
3935 if (netdev->flags & IFF_PROMISC) {
3936 /* retain VLAN HW filtering if in VT mode */
3937 if (adapter->vfs_allocated_count)
3938 rctl |= E1000_RCTL_VFE;
3939 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3940 vmolr |= (E1000_VMOLR_ROPE | E1000_VMOLR_MPME);
3942 if (netdev->flags & IFF_ALLMULTI) {
3943 rctl |= E1000_RCTL_MPE;
3944 vmolr |= E1000_VMOLR_MPME;
3946 /* Write addresses to the MTA, if the attempt fails
3947 * then we should just turn on promiscuous mode so
3948 * that we can at least receive multicast traffic
3950 count = igb_write_mc_addr_list(netdev);
3952 rctl |= E1000_RCTL_MPE;
3953 vmolr |= E1000_VMOLR_MPME;
3955 vmolr |= E1000_VMOLR_ROMPE;
3958 /* Write addresses to available RAR registers, if there is not
3959 * sufficient space to store all the addresses then enable
3960 * unicast promiscuous mode
3962 count = igb_write_uc_addr_list(netdev);
3964 rctl |= E1000_RCTL_UPE;
3965 vmolr |= E1000_VMOLR_ROPE;
3967 rctl |= E1000_RCTL_VFE;
3969 wr32(E1000_RCTL, rctl);
3971 /* In order to support SR-IOV and eventually VMDq it is necessary to set
3972 * the VMOLR to enable the appropriate modes. Without this workaround
3973 * we will have issues with VLAN tag stripping not being done for frames
3974 * that are only arriving because we are the default pool
3976 if ((hw->mac.type < e1000_82576) || (hw->mac.type > e1000_i350))
3979 vmolr |= rd32(E1000_VMOLR(vfn)) &
3980 ~(E1000_VMOLR_ROPE | E1000_VMOLR_MPME | E1000_VMOLR_ROMPE);
3981 wr32(E1000_VMOLR(vfn), vmolr);
3982 igb_restore_vf_multicasts(adapter);
3985 static void igb_check_wvbr(struct igb_adapter *adapter)
3987 struct e1000_hw *hw = &adapter->hw;
3990 switch (hw->mac.type) {
3993 if (!(wvbr = rd32(E1000_WVBR)))
4000 adapter->wvbr |= wvbr;
4003 #define IGB_STAGGERED_QUEUE_OFFSET 8
4005 static void igb_spoof_check(struct igb_adapter *adapter)
4012 for(j = 0; j < adapter->vfs_allocated_count; j++) {
4013 if (adapter->wvbr & (1 << j) ||
4014 adapter->wvbr & (1 << (j + IGB_STAGGERED_QUEUE_OFFSET))) {
4015 dev_warn(&adapter->pdev->dev,
4016 "Spoof event(s) detected on VF %d\n", j);
4019 (1 << (j + IGB_STAGGERED_QUEUE_OFFSET)));
4024 /* Need to wait a few seconds after link up to get diagnostic information from
4027 static void igb_update_phy_info(unsigned long data)
4029 struct igb_adapter *adapter = (struct igb_adapter *) data;
4030 igb_get_phy_info(&adapter->hw);
4034 * igb_has_link - check shared code for link and determine up/down
4035 * @adapter: pointer to driver private info
4037 bool igb_has_link(struct igb_adapter *adapter)
4039 struct e1000_hw *hw = &adapter->hw;
4040 bool link_active = false;
4042 /* get_link_status is set on LSC (link status) interrupt or
4043 * rx sequence error interrupt. get_link_status will stay
4044 * false until the e1000_check_for_link establishes link
4045 * for copper adapters ONLY
4047 switch (hw->phy.media_type) {
4048 case e1000_media_type_copper:
4049 if (!hw->mac.get_link_status)
4051 case e1000_media_type_internal_serdes:
4052 hw->mac.ops.check_for_link(hw);
4053 link_active = !hw->mac.get_link_status;
4056 case e1000_media_type_unknown:
4060 if (((hw->mac.type == e1000_i210) ||
4061 (hw->mac.type == e1000_i211)) &&
4062 (hw->phy.id == I210_I_PHY_ID)) {
4063 if (!netif_carrier_ok(adapter->netdev)) {
4064 adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
4065 } else if (!(adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)) {
4066 adapter->flags |= IGB_FLAG_NEED_LINK_UPDATE;
4067 adapter->link_check_timeout = jiffies;
4074 static bool igb_thermal_sensor_event(struct e1000_hw *hw, u32 event)
4077 u32 ctrl_ext, thstat;
4079 /* check for thermal sensor event on i350 copper only */
4080 if (hw->mac.type == e1000_i350) {
4081 thstat = rd32(E1000_THSTAT);
4082 ctrl_ext = rd32(E1000_CTRL_EXT);
4084 if ((hw->phy.media_type == e1000_media_type_copper) &&
4085 !(ctrl_ext & E1000_CTRL_EXT_LINK_MODE_SGMII))
4086 ret = !!(thstat & event);
4093 * igb_watchdog - Timer Call-back
4094 * @data: pointer to adapter cast into an unsigned long
4096 static void igb_watchdog(unsigned long data)
4098 struct igb_adapter *adapter = (struct igb_adapter *)data;
4099 /* Do the rest outside of interrupt context */
4100 schedule_work(&adapter->watchdog_task);
4103 static void igb_watchdog_task(struct work_struct *work)
4105 struct igb_adapter *adapter = container_of(work,
4108 struct e1000_hw *hw = &adapter->hw;
4109 struct e1000_phy_info *phy = &hw->phy;
4110 struct net_device *netdev = adapter->netdev;
4115 link = igb_has_link(adapter);
4117 if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE) {
4118 if (time_after(jiffies, (adapter->link_check_timeout + HZ)))
4119 adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
4124 /* Force link down if we have fiber to swap to */
4125 if (adapter->flags & IGB_FLAG_MAS_ENABLE) {
4126 if (hw->phy.media_type == e1000_media_type_copper) {
4127 connsw = rd32(E1000_CONNSW);
4128 if (!(connsw & E1000_CONNSW_AUTOSENSE_EN))
4133 /* Perform a reset if the media type changed. */
4134 if (hw->dev_spec._82575.media_changed) {
4135 hw->dev_spec._82575.media_changed = false;
4136 adapter->flags |= IGB_FLAG_MEDIA_RESET;
4139 /* Cancel scheduled suspend requests. */
4140 pm_runtime_resume(netdev->dev.parent);
4142 if (!netif_carrier_ok(netdev)) {
4144 hw->mac.ops.get_speed_and_duplex(hw,
4145 &adapter->link_speed,
4146 &adapter->link_duplex);
4148 ctrl = rd32(E1000_CTRL);
4149 /* Links status message must follow this format */
4150 printk(KERN_INFO "igb: %s NIC Link is Up %d Mbps %s "
4151 "Duplex, Flow Control: %s\n",
4153 adapter->link_speed,
4154 adapter->link_duplex == FULL_DUPLEX ?
4156 (ctrl & E1000_CTRL_TFCE) &&
4157 (ctrl & E1000_CTRL_RFCE) ? "RX/TX" :
4158 (ctrl & E1000_CTRL_RFCE) ? "RX" :
4159 (ctrl & E1000_CTRL_TFCE) ? "TX" : "None");
4161 /* check if SmartSpeed worked */
4162 igb_check_downshift(hw);
4163 if (phy->speed_downgraded)
4164 netdev_warn(netdev, "Link Speed was downgraded by SmartSpeed\n");
4166 /* check for thermal sensor event */
4167 if (igb_thermal_sensor_event(hw,
4168 E1000_THSTAT_LINK_THROTTLE)) {
4169 netdev_info(netdev, "The network adapter link "
4170 "speed was downshifted because it "
4174 /* adjust timeout factor according to speed/duplex */
4175 adapter->tx_timeout_factor = 1;
4176 switch (adapter->link_speed) {
4178 adapter->tx_timeout_factor = 14;
4181 /* maybe add some timeout factor ? */
4185 netif_carrier_on(netdev);
4187 igb_ping_all_vfs(adapter);
4188 igb_check_vf_rate_limit(adapter);
4190 /* link state has changed, schedule phy info update */
4191 if (!test_bit(__IGB_DOWN, &adapter->state))
4192 mod_timer(&adapter->phy_info_timer,
4193 round_jiffies(jiffies + 2 * HZ));
4196 if (netif_carrier_ok(netdev)) {
4197 adapter->link_speed = 0;
4198 adapter->link_duplex = 0;
4200 /* check for thermal sensor event */
4201 if (igb_thermal_sensor_event(hw,
4202 E1000_THSTAT_PWR_DOWN)) {
4203 netdev_err(netdev, "The network adapter was "
4204 "stopped because it overheated\n");
4207 /* Links status message must follow this format */
4208 printk(KERN_INFO "igb: %s NIC Link is Down\n",
4210 netif_carrier_off(netdev);
4212 igb_ping_all_vfs(adapter);
4214 /* link state has changed, schedule phy info update */
4215 if (!test_bit(__IGB_DOWN, &adapter->state))
4216 mod_timer(&adapter->phy_info_timer,
4217 round_jiffies(jiffies + 2 * HZ));
4219 /* link is down, time to check for alternate media */
4220 if (adapter->flags & IGB_FLAG_MAS_ENABLE) {
4221 igb_check_swap_media(adapter);
4222 if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
4223 schedule_work(&adapter->reset_task);
4224 /* return immediately */
4228 pm_schedule_suspend(netdev->dev.parent,
4231 /* also check for alternate media here */
4232 } else if (!netif_carrier_ok(netdev) &&
4233 (adapter->flags & IGB_FLAG_MAS_ENABLE)) {
4234 igb_check_swap_media(adapter);
4235 if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
4236 schedule_work(&adapter->reset_task);
4237 /* return immediately */
4243 spin_lock(&adapter->stats64_lock);
4244 igb_update_stats(adapter, &adapter->stats64);
4245 spin_unlock(&adapter->stats64_lock);
4247 for (i = 0; i < adapter->num_tx_queues; i++) {
4248 struct igb_ring *tx_ring = adapter->tx_ring[i];
4249 if (!netif_carrier_ok(netdev)) {
4250 /* We've lost link, so the controller stops DMA,
4251 * but we've got queued Tx work that's never going
4252 * to get done, so reset controller to flush Tx.
4253 * (Do the reset outside of interrupt context).
4255 if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) {
4256 adapter->tx_timeout_count++;
4257 schedule_work(&adapter->reset_task);
4258 /* return immediately since reset is imminent */
4263 /* Force detection of hung controller every watchdog period */
4264 set_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
4267 /* Cause software interrupt to ensure Rx ring is cleaned */
4268 if (adapter->flags & IGB_FLAG_HAS_MSIX) {
4270 for (i = 0; i < adapter->num_q_vectors; i++)
4271 eics |= adapter->q_vector[i]->eims_value;
4272 wr32(E1000_EICS, eics);
4274 wr32(E1000_ICS, E1000_ICS_RXDMT0);
4277 igb_spoof_check(adapter);
4278 igb_ptp_rx_hang(adapter);
4280 /* Reset the timer */
4281 if (!test_bit(__IGB_DOWN, &adapter->state)) {
4282 if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)
4283 mod_timer(&adapter->watchdog_timer,
4284 round_jiffies(jiffies + HZ));
4286 mod_timer(&adapter->watchdog_timer,
4287 round_jiffies(jiffies + 2 * HZ));
4291 enum latency_range {
4295 latency_invalid = 255
4299 * igb_update_ring_itr - update the dynamic ITR value based on packet size
4300 * @q_vector: pointer to q_vector
4302 * Stores a new ITR value based on strictly on packet size. This
4303 * algorithm is less sophisticated than that used in igb_update_itr,
4304 * due to the difficulty of synchronizing statistics across multiple
4305 * receive rings. The divisors and thresholds used by this function
4306 * were determined based on theoretical maximum wire speed and testing
4307 * data, in order to minimize response time while increasing bulk
4309 * This functionality is controlled by the InterruptThrottleRate module
4310 * parameter (see igb_param.c)
4311 * NOTE: This function is called only when operating in a multiqueue
4312 * receive environment.
4314 static void igb_update_ring_itr(struct igb_q_vector *q_vector)
4316 int new_val = q_vector->itr_val;
4317 int avg_wire_size = 0;
4318 struct igb_adapter *adapter = q_vector->adapter;
4319 unsigned int packets;
4321 /* For non-gigabit speeds, just fix the interrupt rate at 4000
4322 * ints/sec - ITR timer value of 120 ticks.
4324 if (adapter->link_speed != SPEED_1000) {
4325 new_val = IGB_4K_ITR;
4329 packets = q_vector->rx.total_packets;
4331 avg_wire_size = q_vector->rx.total_bytes / packets;
4333 packets = q_vector->tx.total_packets;
4335 avg_wire_size = max_t(u32, avg_wire_size,
4336 q_vector->tx.total_bytes / packets);
4338 /* if avg_wire_size isn't set no work was done */
4342 /* Add 24 bytes to size to account for CRC, preamble, and gap */
4343 avg_wire_size += 24;
4345 /* Don't starve jumbo frames */
4346 avg_wire_size = min(avg_wire_size, 3000);
4348 /* Give a little boost to mid-size frames */
4349 if ((avg_wire_size > 300) && (avg_wire_size < 1200))
4350 new_val = avg_wire_size / 3;
4352 new_val = avg_wire_size / 2;
4354 /* conservative mode (itr 3) eliminates the lowest_latency setting */
4355 if (new_val < IGB_20K_ITR &&
4356 ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
4357 (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
4358 new_val = IGB_20K_ITR;
4361 if (new_val != q_vector->itr_val) {
4362 q_vector->itr_val = new_val;
4363 q_vector->set_itr = 1;
4366 q_vector->rx.total_bytes = 0;
4367 q_vector->rx.total_packets = 0;
4368 q_vector->tx.total_bytes = 0;
4369 q_vector->tx.total_packets = 0;
4373 * igb_update_itr - update the dynamic ITR value based on statistics
4374 * @q_vector: pointer to q_vector
4375 * @ring_container: ring info to update the itr for
4377 * Stores a new ITR value based on packets and byte
4378 * counts during the last interrupt. The advantage of per interrupt
4379 * computation is faster updates and more accurate ITR for the current
4380 * traffic pattern. Constants in this function were computed
4381 * based on theoretical maximum wire speed and thresholds were set based
4382 * on testing data as well as attempting to minimize response time
4383 * while increasing bulk throughput.
4384 * this functionality is controlled by the InterruptThrottleRate module
4385 * parameter (see igb_param.c)
4386 * NOTE: These calculations are only valid when operating in a single-
4387 * queue environment.
4389 static void igb_update_itr(struct igb_q_vector *q_vector,
4390 struct igb_ring_container *ring_container)
4392 unsigned int packets = ring_container->total_packets;
4393 unsigned int bytes = ring_container->total_bytes;
4394 u8 itrval = ring_container->itr;
4396 /* no packets, exit with status unchanged */
4401 case lowest_latency:
4402 /* handle TSO and jumbo frames */
4403 if (bytes/packets > 8000)
4404 itrval = bulk_latency;
4405 else if ((packets < 5) && (bytes > 512))
4406 itrval = low_latency;
4408 case low_latency: /* 50 usec aka 20000 ints/s */
4409 if (bytes > 10000) {
4410 /* this if handles the TSO accounting */
4411 if (bytes/packets > 8000) {
4412 itrval = bulk_latency;
4413 } else if ((packets < 10) || ((bytes/packets) > 1200)) {
4414 itrval = bulk_latency;
4415 } else if ((packets > 35)) {
4416 itrval = lowest_latency;
4418 } else if (bytes/packets > 2000) {
4419 itrval = bulk_latency;
4420 } else if (packets <= 2 && bytes < 512) {
4421 itrval = lowest_latency;
4424 case bulk_latency: /* 250 usec aka 4000 ints/s */
4425 if (bytes > 25000) {
4427 itrval = low_latency;
4428 } else if (bytes < 1500) {
4429 itrval = low_latency;
4434 /* clear work counters since we have the values we need */
4435 ring_container->total_bytes = 0;
4436 ring_container->total_packets = 0;
4438 /* write updated itr to ring container */
4439 ring_container->itr = itrval;
4442 static void igb_set_itr(struct igb_q_vector *q_vector)
4444 struct igb_adapter *adapter = q_vector->adapter;
4445 u32 new_itr = q_vector->itr_val;
4448 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
4449 if (adapter->link_speed != SPEED_1000) {
4451 new_itr = IGB_4K_ITR;
4455 igb_update_itr(q_vector, &q_vector->tx);
4456 igb_update_itr(q_vector, &q_vector->rx);
4458 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
4460 /* conservative mode (itr 3) eliminates the lowest_latency setting */
4461 if (current_itr == lowest_latency &&
4462 ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
4463 (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
4464 current_itr = low_latency;
4466 switch (current_itr) {
4467 /* counts and packets in update_itr are dependent on these numbers */
4468 case lowest_latency:
4469 new_itr = IGB_70K_ITR; /* 70,000 ints/sec */
4472 new_itr = IGB_20K_ITR; /* 20,000 ints/sec */
4475 new_itr = IGB_4K_ITR; /* 4,000 ints/sec */
4482 if (new_itr != q_vector->itr_val) {
4483 /* this attempts to bias the interrupt rate towards Bulk
4484 * by adding intermediate steps when interrupt rate is
4487 new_itr = new_itr > q_vector->itr_val ?
4488 max((new_itr * q_vector->itr_val) /
4489 (new_itr + (q_vector->itr_val >> 2)),
4491 /* Don't write the value here; it resets the adapter's
4492 * internal timer, and causes us to delay far longer than
4493 * we should between interrupts. Instead, we write the ITR
4494 * value at the beginning of the next interrupt so the timing
4495 * ends up being correct.
4497 q_vector->itr_val = new_itr;
4498 q_vector->set_itr = 1;
4502 static void igb_tx_ctxtdesc(struct igb_ring *tx_ring, u32 vlan_macip_lens,
4503 u32 type_tucmd, u32 mss_l4len_idx)
4505 struct e1000_adv_tx_context_desc *context_desc;
4506 u16 i = tx_ring->next_to_use;
4508 context_desc = IGB_TX_CTXTDESC(tx_ring, i);
4511 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
4513 /* set bits to identify this as an advanced context descriptor */
4514 type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
4516 /* For 82575, context index must be unique per ring. */
4517 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
4518 mss_l4len_idx |= tx_ring->reg_idx << 4;
4520 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
4521 context_desc->seqnum_seed = 0;
4522 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd);
4523 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
4526 static int igb_tso(struct igb_ring *tx_ring,
4527 struct igb_tx_buffer *first,
4530 struct sk_buff *skb = first->skb;
4531 u32 vlan_macip_lens, type_tucmd;
4532 u32 mss_l4len_idx, l4len;
4534 if (skb->ip_summed != CHECKSUM_PARTIAL)
4537 if (!skb_is_gso(skb))
4540 if (skb_header_cloned(skb)) {
4541 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4546 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
4547 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
4549 if (first->protocol == __constant_htons(ETH_P_IP)) {
4550 struct iphdr *iph = ip_hdr(skb);
4553 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
4557 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
4558 first->tx_flags |= IGB_TX_FLAGS_TSO |
4561 } else if (skb_is_gso_v6(skb)) {
4562 ipv6_hdr(skb)->payload_len = 0;
4563 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4564 &ipv6_hdr(skb)->daddr,
4566 first->tx_flags |= IGB_TX_FLAGS_TSO |
4570 /* compute header lengths */
4571 l4len = tcp_hdrlen(skb);
4572 *hdr_len = skb_transport_offset(skb) + l4len;
4574 /* update gso size and bytecount with header size */
4575 first->gso_segs = skb_shinfo(skb)->gso_segs;
4576 first->bytecount += (first->gso_segs - 1) * *hdr_len;
4579 mss_l4len_idx = l4len << E1000_ADVTXD_L4LEN_SHIFT;
4580 mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT;
4582 /* VLAN MACLEN IPLEN */
4583 vlan_macip_lens = skb_network_header_len(skb);
4584 vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
4585 vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
4587 igb_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
4592 static void igb_tx_csum(struct igb_ring *tx_ring, struct igb_tx_buffer *first)
4594 struct sk_buff *skb = first->skb;
4595 u32 vlan_macip_lens = 0;
4596 u32 mss_l4len_idx = 0;
4599 if (skb->ip_summed != CHECKSUM_PARTIAL) {
4600 if (!(first->tx_flags & IGB_TX_FLAGS_VLAN))
4604 switch (first->protocol) {
4605 case __constant_htons(ETH_P_IP):
4606 vlan_macip_lens |= skb_network_header_len(skb);
4607 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
4608 l4_hdr = ip_hdr(skb)->protocol;
4610 case __constant_htons(ETH_P_IPV6):
4611 vlan_macip_lens |= skb_network_header_len(skb);
4612 l4_hdr = ipv6_hdr(skb)->nexthdr;
4615 if (unlikely(net_ratelimit())) {
4616 dev_warn(tx_ring->dev,
4617 "partial checksum but proto=%x!\n",
4625 type_tucmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
4626 mss_l4len_idx = tcp_hdrlen(skb) <<
4627 E1000_ADVTXD_L4LEN_SHIFT;
4630 type_tucmd |= E1000_ADVTXD_TUCMD_L4T_SCTP;
4631 mss_l4len_idx = sizeof(struct sctphdr) <<
4632 E1000_ADVTXD_L4LEN_SHIFT;
4635 mss_l4len_idx = sizeof(struct udphdr) <<
4636 E1000_ADVTXD_L4LEN_SHIFT;
4639 if (unlikely(net_ratelimit())) {
4640 dev_warn(tx_ring->dev,
4641 "partial checksum but l4 proto=%x!\n",
4647 /* update TX checksum flag */
4648 first->tx_flags |= IGB_TX_FLAGS_CSUM;
4651 vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
4652 vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
4654 igb_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
4657 #define IGB_SET_FLAG(_input, _flag, _result) \
4658 ((_flag <= _result) ? \
4659 ((u32)(_input & _flag) * (_result / _flag)) : \
4660 ((u32)(_input & _flag) / (_flag / _result)))
4662 static u32 igb_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
4664 /* set type for advanced descriptor with frame checksum insertion */
4665 u32 cmd_type = E1000_ADVTXD_DTYP_DATA |
4666 E1000_ADVTXD_DCMD_DEXT |
4667 E1000_ADVTXD_DCMD_IFCS;
4669 /* set HW vlan bit if vlan is present */
4670 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_VLAN,
4671 (E1000_ADVTXD_DCMD_VLE));
4673 /* set segmentation bits for TSO */
4674 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSO,
4675 (E1000_ADVTXD_DCMD_TSE));
4677 /* set timestamp bit if present */
4678 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSTAMP,
4679 (E1000_ADVTXD_MAC_TSTAMP));
4681 /* insert frame checksum */
4682 cmd_type ^= IGB_SET_FLAG(skb->no_fcs, 1, E1000_ADVTXD_DCMD_IFCS);
4687 static void igb_tx_olinfo_status(struct igb_ring *tx_ring,
4688 union e1000_adv_tx_desc *tx_desc,
4689 u32 tx_flags, unsigned int paylen)
4691 u32 olinfo_status = paylen << E1000_ADVTXD_PAYLEN_SHIFT;
4693 /* 82575 requires a unique index per ring */
4694 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
4695 olinfo_status |= tx_ring->reg_idx << 4;
4697 /* insert L4 checksum */
4698 olinfo_status |= IGB_SET_FLAG(tx_flags,
4700 (E1000_TXD_POPTS_TXSM << 8));
4702 /* insert IPv4 checksum */
4703 olinfo_status |= IGB_SET_FLAG(tx_flags,
4705 (E1000_TXD_POPTS_IXSM << 8));
4707 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
4710 static void igb_tx_map(struct igb_ring *tx_ring,
4711 struct igb_tx_buffer *first,
4714 struct sk_buff *skb = first->skb;
4715 struct igb_tx_buffer *tx_buffer;
4716 union e1000_adv_tx_desc *tx_desc;
4717 struct skb_frag_struct *frag;
4719 unsigned int data_len, size;
4720 u32 tx_flags = first->tx_flags;
4721 u32 cmd_type = igb_tx_cmd_type(skb, tx_flags);
4722 u16 i = tx_ring->next_to_use;
4724 tx_desc = IGB_TX_DESC(tx_ring, i);
4726 igb_tx_olinfo_status(tx_ring, tx_desc, tx_flags, skb->len - hdr_len);
4728 size = skb_headlen(skb);
4729 data_len = skb->data_len;
4731 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
4735 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
4736 if (dma_mapping_error(tx_ring->dev, dma))
4739 /* record length, and DMA address */
4740 dma_unmap_len_set(tx_buffer, len, size);
4741 dma_unmap_addr_set(tx_buffer, dma, dma);
4743 tx_desc->read.buffer_addr = cpu_to_le64(dma);
4745 while (unlikely(size > IGB_MAX_DATA_PER_TXD)) {
4746 tx_desc->read.cmd_type_len =
4747 cpu_to_le32(cmd_type ^ IGB_MAX_DATA_PER_TXD);
4751 if (i == tx_ring->count) {
4752 tx_desc = IGB_TX_DESC(tx_ring, 0);
4755 tx_desc->read.olinfo_status = 0;
4757 dma += IGB_MAX_DATA_PER_TXD;
4758 size -= IGB_MAX_DATA_PER_TXD;
4760 tx_desc->read.buffer_addr = cpu_to_le64(dma);
4763 if (likely(!data_len))
4766 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
4770 if (i == tx_ring->count) {
4771 tx_desc = IGB_TX_DESC(tx_ring, 0);
4774 tx_desc->read.olinfo_status = 0;
4776 size = skb_frag_size(frag);
4779 dma = skb_frag_dma_map(tx_ring->dev, frag, 0,
4780 size, DMA_TO_DEVICE);
4782 tx_buffer = &tx_ring->tx_buffer_info[i];
4785 /* write last descriptor with RS and EOP bits */
4786 cmd_type |= size | IGB_TXD_DCMD;
4787 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
4789 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
4791 /* set the timestamp */
4792 first->time_stamp = jiffies;
4794 /* Force memory writes to complete before letting h/w know there
4795 * are new descriptors to fetch. (Only applicable for weak-ordered
4796 * memory model archs, such as IA-64).
4798 * We also need this memory barrier to make certain all of the
4799 * status bits have been updated before next_to_watch is written.
4803 /* set next_to_watch value indicating a packet is present */
4804 first->next_to_watch = tx_desc;
4807 if (i == tx_ring->count)
4810 tx_ring->next_to_use = i;
4812 writel(i, tx_ring->tail);
4814 /* we need this if more than one processor can write to our tail
4815 * at a time, it synchronizes IO on IA64/Altix systems
4822 dev_err(tx_ring->dev, "TX DMA map failed\n");
4824 /* clear dma mappings for failed tx_buffer_info map */
4826 tx_buffer = &tx_ring->tx_buffer_info[i];
4827 igb_unmap_and_free_tx_resource(tx_ring, tx_buffer);
4828 if (tx_buffer == first)
4835 tx_ring->next_to_use = i;
4838 static int __igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
4840 struct net_device *netdev = tx_ring->netdev;
4842 netif_stop_subqueue(netdev, tx_ring->queue_index);
4844 /* Herbert's original patch had:
4845 * smp_mb__after_netif_stop_queue();
4846 * but since that doesn't exist yet, just open code it.
4850 /* We need to check again in a case another CPU has just
4851 * made room available.
4853 if (igb_desc_unused(tx_ring) < size)
4857 netif_wake_subqueue(netdev, tx_ring->queue_index);
4859 u64_stats_update_begin(&tx_ring->tx_syncp2);
4860 tx_ring->tx_stats.restart_queue2++;
4861 u64_stats_update_end(&tx_ring->tx_syncp2);
4866 static inline int igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
4868 if (igb_desc_unused(tx_ring) >= size)
4870 return __igb_maybe_stop_tx(tx_ring, size);
4873 netdev_tx_t igb_xmit_frame_ring(struct sk_buff *skb,
4874 struct igb_ring *tx_ring)
4876 struct igb_tx_buffer *first;
4879 u16 count = TXD_USE_COUNT(skb_headlen(skb));
4880 __be16 protocol = vlan_get_protocol(skb);
4883 /* need: 1 descriptor per page * PAGE_SIZE/IGB_MAX_DATA_PER_TXD,
4884 * + 1 desc for skb_headlen/IGB_MAX_DATA_PER_TXD,
4885 * + 2 desc gap to keep tail from touching head,
4886 * + 1 desc for context descriptor,
4887 * otherwise try next time
4889 if (NETDEV_FRAG_PAGE_MAX_SIZE > IGB_MAX_DATA_PER_TXD) {
4891 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
4892 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
4894 count += skb_shinfo(skb)->nr_frags;
4897 if (igb_maybe_stop_tx(tx_ring, count + 3)) {
4898 /* this is a hard error */
4899 return NETDEV_TX_BUSY;
4902 /* record the location of the first descriptor for this packet */
4903 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
4905 first->bytecount = skb->len;
4906 first->gso_segs = 1;
4908 skb_tx_timestamp(skb);
4910 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
4911 struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
4913 if (!(adapter->ptp_tx_skb)) {
4914 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
4915 tx_flags |= IGB_TX_FLAGS_TSTAMP;
4917 adapter->ptp_tx_skb = skb_get(skb);
4918 adapter->ptp_tx_start = jiffies;
4919 if (adapter->hw.mac.type == e1000_82576)
4920 schedule_work(&adapter->ptp_tx_work);
4924 if (vlan_tx_tag_present(skb)) {
4925 tx_flags |= IGB_TX_FLAGS_VLAN;
4926 tx_flags |= (vlan_tx_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
4929 /* record initial flags and protocol */
4930 first->tx_flags = tx_flags;
4931 first->protocol = protocol;
4933 tso = igb_tso(tx_ring, first, &hdr_len);
4937 igb_tx_csum(tx_ring, first);
4939 igb_tx_map(tx_ring, first, hdr_len);
4941 /* Make sure there is space in the ring for the next send. */
4942 igb_maybe_stop_tx(tx_ring, DESC_NEEDED);
4944 return NETDEV_TX_OK;
4947 igb_unmap_and_free_tx_resource(tx_ring, first);
4949 return NETDEV_TX_OK;
4952 static inline struct igb_ring *igb_tx_queue_mapping(struct igb_adapter *adapter,
4953 struct sk_buff *skb)
4955 unsigned int r_idx = skb->queue_mapping;
4957 if (r_idx >= adapter->num_tx_queues)
4958 r_idx = r_idx % adapter->num_tx_queues;
4960 return adapter->tx_ring[r_idx];
4963 static netdev_tx_t igb_xmit_frame(struct sk_buff *skb,
4964 struct net_device *netdev)
4966 struct igb_adapter *adapter = netdev_priv(netdev);
4968 if (test_bit(__IGB_DOWN, &adapter->state)) {
4969 dev_kfree_skb_any(skb);
4970 return NETDEV_TX_OK;
4973 if (skb->len <= 0) {
4974 dev_kfree_skb_any(skb);
4975 return NETDEV_TX_OK;
4978 /* The minimum packet size with TCTL.PSP set is 17 so pad the skb
4979 * in order to meet this minimum size requirement.
4981 if (unlikely(skb->len < 17)) {
4982 if (skb_pad(skb, 17 - skb->len))
4983 return NETDEV_TX_OK;
4985 skb_set_tail_pointer(skb, 17);
4988 return igb_xmit_frame_ring(skb, igb_tx_queue_mapping(adapter, skb));
4992 * igb_tx_timeout - Respond to a Tx Hang
4993 * @netdev: network interface device structure
4995 static void igb_tx_timeout(struct net_device *netdev)
4997 struct igb_adapter *adapter = netdev_priv(netdev);
4998 struct e1000_hw *hw = &adapter->hw;
5000 /* Do the reset outside of interrupt context */
5001 adapter->tx_timeout_count++;
5003 if (hw->mac.type >= e1000_82580)
5004 hw->dev_spec._82575.global_device_reset = true;
5006 schedule_work(&adapter->reset_task);
5008 (adapter->eims_enable_mask & ~adapter->eims_other));
5011 static void igb_reset_task(struct work_struct *work)
5013 struct igb_adapter *adapter;
5014 adapter = container_of(work, struct igb_adapter, reset_task);
5017 netdev_err(adapter->netdev, "Reset adapter\n");
5018 igb_reinit_locked(adapter);
5022 * igb_get_stats64 - Get System Network Statistics
5023 * @netdev: network interface device structure
5024 * @stats: rtnl_link_stats64 pointer
5026 static struct rtnl_link_stats64 *igb_get_stats64(struct net_device *netdev,
5027 struct rtnl_link_stats64 *stats)
5029 struct igb_adapter *adapter = netdev_priv(netdev);
5031 spin_lock(&adapter->stats64_lock);
5032 igb_update_stats(adapter, &adapter->stats64);
5033 memcpy(stats, &adapter->stats64, sizeof(*stats));
5034 spin_unlock(&adapter->stats64_lock);
5040 * igb_change_mtu - Change the Maximum Transfer Unit
5041 * @netdev: network interface device structure
5042 * @new_mtu: new value for maximum frame size
5044 * Returns 0 on success, negative on failure
5046 static int igb_change_mtu(struct net_device *netdev, int new_mtu)
5048 struct igb_adapter *adapter = netdev_priv(netdev);
5049 struct pci_dev *pdev = adapter->pdev;
5050 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
5052 if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
5053 dev_err(&pdev->dev, "Invalid MTU setting\n");
5057 #define MAX_STD_JUMBO_FRAME_SIZE 9238
5058 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
5059 dev_err(&pdev->dev, "MTU > 9216 not supported.\n");
5063 /* adjust max frame to be at least the size of a standard frame */
5064 if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
5065 max_frame = ETH_FRAME_LEN + ETH_FCS_LEN;
5067 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
5070 /* igb_down has a dependency on max_frame_size */
5071 adapter->max_frame_size = max_frame;
5073 if (netif_running(netdev))
5076 dev_info(&pdev->dev, "changing MTU from %d to %d\n",
5077 netdev->mtu, new_mtu);
5078 netdev->mtu = new_mtu;
5080 if (netif_running(netdev))
5085 clear_bit(__IGB_RESETTING, &adapter->state);
5091 * igb_update_stats - Update the board statistics counters
5092 * @adapter: board private structure
5094 void igb_update_stats(struct igb_adapter *adapter,
5095 struct rtnl_link_stats64 *net_stats)
5097 struct e1000_hw *hw = &adapter->hw;
5098 struct pci_dev *pdev = adapter->pdev;
5104 u64 _bytes, _packets;
5106 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
5108 /* Prevent stats update while adapter is being reset, or if the pci
5109 * connection is down.
5111 if (adapter->link_speed == 0)
5113 if (pci_channel_offline(pdev))
5120 for (i = 0; i < adapter->num_rx_queues; i++) {
5121 u32 rqdpc = rd32(E1000_RQDPC(i));
5122 struct igb_ring *ring = adapter->rx_ring[i];
5125 ring->rx_stats.drops += rqdpc;
5126 net_stats->rx_fifo_errors += rqdpc;
5130 start = u64_stats_fetch_begin_bh(&ring->rx_syncp);
5131 _bytes = ring->rx_stats.bytes;
5132 _packets = ring->rx_stats.packets;
5133 } while (u64_stats_fetch_retry_bh(&ring->rx_syncp, start));
5135 packets += _packets;
5138 net_stats->rx_bytes = bytes;
5139 net_stats->rx_packets = packets;
5143 for (i = 0; i < adapter->num_tx_queues; i++) {
5144 struct igb_ring *ring = adapter->tx_ring[i];
5146 start = u64_stats_fetch_begin_bh(&ring->tx_syncp);
5147 _bytes = ring->tx_stats.bytes;
5148 _packets = ring->tx_stats.packets;
5149 } while (u64_stats_fetch_retry_bh(&ring->tx_syncp, start));
5151 packets += _packets;
5153 net_stats->tx_bytes = bytes;
5154 net_stats->tx_packets = packets;
5157 /* read stats registers */
5158 adapter->stats.crcerrs += rd32(E1000_CRCERRS);
5159 adapter->stats.gprc += rd32(E1000_GPRC);
5160 adapter->stats.gorc += rd32(E1000_GORCL);
5161 rd32(E1000_GORCH); /* clear GORCL */
5162 adapter->stats.bprc += rd32(E1000_BPRC);
5163 adapter->stats.mprc += rd32(E1000_MPRC);
5164 adapter->stats.roc += rd32(E1000_ROC);
5166 adapter->stats.prc64 += rd32(E1000_PRC64);
5167 adapter->stats.prc127 += rd32(E1000_PRC127);
5168 adapter->stats.prc255 += rd32(E1000_PRC255);
5169 adapter->stats.prc511 += rd32(E1000_PRC511);
5170 adapter->stats.prc1023 += rd32(E1000_PRC1023);
5171 adapter->stats.prc1522 += rd32(E1000_PRC1522);
5172 adapter->stats.symerrs += rd32(E1000_SYMERRS);
5173 adapter->stats.sec += rd32(E1000_SEC);
5175 mpc = rd32(E1000_MPC);
5176 adapter->stats.mpc += mpc;
5177 net_stats->rx_fifo_errors += mpc;
5178 adapter->stats.scc += rd32(E1000_SCC);
5179 adapter->stats.ecol += rd32(E1000_ECOL);
5180 adapter->stats.mcc += rd32(E1000_MCC);
5181 adapter->stats.latecol += rd32(E1000_LATECOL);
5182 adapter->stats.dc += rd32(E1000_DC);
5183 adapter->stats.rlec += rd32(E1000_RLEC);
5184 adapter->stats.xonrxc += rd32(E1000_XONRXC);
5185 adapter->stats.xontxc += rd32(E1000_XONTXC);
5186 adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
5187 adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
5188 adapter->stats.fcruc += rd32(E1000_FCRUC);
5189 adapter->stats.gptc += rd32(E1000_GPTC);
5190 adapter->stats.gotc += rd32(E1000_GOTCL);
5191 rd32(E1000_GOTCH); /* clear GOTCL */
5192 adapter->stats.rnbc += rd32(E1000_RNBC);
5193 adapter->stats.ruc += rd32(E1000_RUC);
5194 adapter->stats.rfc += rd32(E1000_RFC);
5195 adapter->stats.rjc += rd32(E1000_RJC);
5196 adapter->stats.tor += rd32(E1000_TORH);
5197 adapter->stats.tot += rd32(E1000_TOTH);
5198 adapter->stats.tpr += rd32(E1000_TPR);
5200 adapter->stats.ptc64 += rd32(E1000_PTC64);
5201 adapter->stats.ptc127 += rd32(E1000_PTC127);
5202 adapter->stats.ptc255 += rd32(E1000_PTC255);
5203 adapter->stats.ptc511 += rd32(E1000_PTC511);
5204 adapter->stats.ptc1023 += rd32(E1000_PTC1023);
5205 adapter->stats.ptc1522 += rd32(E1000_PTC1522);
5207 adapter->stats.mptc += rd32(E1000_MPTC);
5208 adapter->stats.bptc += rd32(E1000_BPTC);
5210 adapter->stats.tpt += rd32(E1000_TPT);
5211 adapter->stats.colc += rd32(E1000_COLC);
5213 adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
5214 /* read internal phy specific stats */
5215 reg = rd32(E1000_CTRL_EXT);
5216 if (!(reg & E1000_CTRL_EXT_LINK_MODE_MASK)) {
5217 adapter->stats.rxerrc += rd32(E1000_RXERRC);
5219 /* this stat has invalid values on i210/i211 */
5220 if ((hw->mac.type != e1000_i210) &&
5221 (hw->mac.type != e1000_i211))
5222 adapter->stats.tncrs += rd32(E1000_TNCRS);
5225 adapter->stats.tsctc += rd32(E1000_TSCTC);
5226 adapter->stats.tsctfc += rd32(E1000_TSCTFC);
5228 adapter->stats.iac += rd32(E1000_IAC);
5229 adapter->stats.icrxoc += rd32(E1000_ICRXOC);
5230 adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
5231 adapter->stats.icrxatc += rd32(E1000_ICRXATC);
5232 adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
5233 adapter->stats.ictxatc += rd32(E1000_ICTXATC);
5234 adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
5235 adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
5236 adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
5238 /* Fill out the OS statistics structure */
5239 net_stats->multicast = adapter->stats.mprc;
5240 net_stats->collisions = adapter->stats.colc;
5244 /* RLEC on some newer hardware can be incorrect so build
5245 * our own version based on RUC and ROC
5247 net_stats->rx_errors = adapter->stats.rxerrc +
5248 adapter->stats.crcerrs + adapter->stats.algnerrc +
5249 adapter->stats.ruc + adapter->stats.roc +
5250 adapter->stats.cexterr;
5251 net_stats->rx_length_errors = adapter->stats.ruc +
5253 net_stats->rx_crc_errors = adapter->stats.crcerrs;
5254 net_stats->rx_frame_errors = adapter->stats.algnerrc;
5255 net_stats->rx_missed_errors = adapter->stats.mpc;
5258 net_stats->tx_errors = adapter->stats.ecol +
5259 adapter->stats.latecol;
5260 net_stats->tx_aborted_errors = adapter->stats.ecol;
5261 net_stats->tx_window_errors = adapter->stats.latecol;
5262 net_stats->tx_carrier_errors = adapter->stats.tncrs;
5264 /* Tx Dropped needs to be maintained elsewhere */
5267 if (hw->phy.media_type == e1000_media_type_copper) {
5268 if ((adapter->link_speed == SPEED_1000) &&
5269 (!igb_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
5270 phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
5271 adapter->phy_stats.idle_errors += phy_tmp;
5275 /* Management Stats */
5276 adapter->stats.mgptc += rd32(E1000_MGTPTC);
5277 adapter->stats.mgprc += rd32(E1000_MGTPRC);
5278 adapter->stats.mgpdc += rd32(E1000_MGTPDC);
5281 reg = rd32(E1000_MANC);
5282 if (reg & E1000_MANC_EN_BMC2OS) {
5283 adapter->stats.o2bgptc += rd32(E1000_O2BGPTC);
5284 adapter->stats.o2bspc += rd32(E1000_O2BSPC);
5285 adapter->stats.b2ospc += rd32(E1000_B2OSPC);
5286 adapter->stats.b2ogprc += rd32(E1000_B2OGPRC);
5290 static irqreturn_t igb_msix_other(int irq, void *data)
5292 struct igb_adapter *adapter = data;
5293 struct e1000_hw *hw = &adapter->hw;
5294 u32 icr = rd32(E1000_ICR);
5295 /* reading ICR causes bit 31 of EICR to be cleared */
5297 if (icr & E1000_ICR_DRSTA)
5298 schedule_work(&adapter->reset_task);
5300 if (icr & E1000_ICR_DOUTSYNC) {
5301 /* HW is reporting DMA is out of sync */
5302 adapter->stats.doosync++;
5303 /* The DMA Out of Sync is also indication of a spoof event
5304 * in IOV mode. Check the Wrong VM Behavior register to
5305 * see if it is really a spoof event.
5307 igb_check_wvbr(adapter);
5310 /* Check for a mailbox event */
5311 if (icr & E1000_ICR_VMMB)
5312 igb_msg_task(adapter);
5314 if (icr & E1000_ICR_LSC) {
5315 hw->mac.get_link_status = 1;
5316 /* guard against interrupt when we're going down */
5317 if (!test_bit(__IGB_DOWN, &adapter->state))
5318 mod_timer(&adapter->watchdog_timer, jiffies + 1);
5321 if (icr & E1000_ICR_TS) {
5322 u32 tsicr = rd32(E1000_TSICR);
5324 if (tsicr & E1000_TSICR_TXTS) {
5325 /* acknowledge the interrupt */
5326 wr32(E1000_TSICR, E1000_TSICR_TXTS);
5327 /* retrieve hardware timestamp */
5328 schedule_work(&adapter->ptp_tx_work);
5332 wr32(E1000_EIMS, adapter->eims_other);
5337 static void igb_write_itr(struct igb_q_vector *q_vector)
5339 struct igb_adapter *adapter = q_vector->adapter;
5340 u32 itr_val = q_vector->itr_val & 0x7FFC;
5342 if (!q_vector->set_itr)
5348 if (adapter->hw.mac.type == e1000_82575)
5349 itr_val |= itr_val << 16;
5351 itr_val |= E1000_EITR_CNT_IGNR;
5353 writel(itr_val, q_vector->itr_register);
5354 q_vector->set_itr = 0;
5357 static irqreturn_t igb_msix_ring(int irq, void *data)
5359 struct igb_q_vector *q_vector = data;
5361 /* Write the ITR value calculated from the previous interrupt. */
5362 igb_write_itr(q_vector);
5364 napi_schedule(&q_vector->napi);
5369 #ifdef CONFIG_IGB_DCA
5370 static void igb_update_tx_dca(struct igb_adapter *adapter,
5371 struct igb_ring *tx_ring,
5374 struct e1000_hw *hw = &adapter->hw;
5375 u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
5377 if (hw->mac.type != e1000_82575)
5378 txctrl <<= E1000_DCA_TXCTRL_CPUID_SHIFT;
5380 /* We can enable relaxed ordering for reads, but not writes when
5381 * DCA is enabled. This is due to a known issue in some chipsets
5382 * which will cause the DCA tag to be cleared.
5384 txctrl |= E1000_DCA_TXCTRL_DESC_RRO_EN |
5385 E1000_DCA_TXCTRL_DATA_RRO_EN |
5386 E1000_DCA_TXCTRL_DESC_DCA_EN;
5388 wr32(E1000_DCA_TXCTRL(tx_ring->reg_idx), txctrl);
5391 static void igb_update_rx_dca(struct igb_adapter *adapter,
5392 struct igb_ring *rx_ring,
5395 struct e1000_hw *hw = &adapter->hw;
5396 u32 rxctrl = dca3_get_tag(&adapter->pdev->dev, cpu);
5398 if (hw->mac.type != e1000_82575)
5399 rxctrl <<= E1000_DCA_RXCTRL_CPUID_SHIFT;
5401 /* We can enable relaxed ordering for reads, but not writes when
5402 * DCA is enabled. This is due to a known issue in some chipsets
5403 * which will cause the DCA tag to be cleared.
5405 rxctrl |= E1000_DCA_RXCTRL_DESC_RRO_EN |
5406 E1000_DCA_RXCTRL_DESC_DCA_EN;
5408 wr32(E1000_DCA_RXCTRL(rx_ring->reg_idx), rxctrl);
5411 static void igb_update_dca(struct igb_q_vector *q_vector)
5413 struct igb_adapter *adapter = q_vector->adapter;
5414 int cpu = get_cpu();
5416 if (q_vector->cpu == cpu)
5419 if (q_vector->tx.ring)
5420 igb_update_tx_dca(adapter, q_vector->tx.ring, cpu);
5422 if (q_vector->rx.ring)
5423 igb_update_rx_dca(adapter, q_vector->rx.ring, cpu);
5425 q_vector->cpu = cpu;
5430 static void igb_setup_dca(struct igb_adapter *adapter)
5432 struct e1000_hw *hw = &adapter->hw;
5435 if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
5438 /* Always use CB2 mode, difference is masked in the CB driver. */
5439 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
5441 for (i = 0; i < adapter->num_q_vectors; i++) {
5442 adapter->q_vector[i]->cpu = -1;
5443 igb_update_dca(adapter->q_vector[i]);
5447 static int __igb_notify_dca(struct device *dev, void *data)
5449 struct net_device *netdev = dev_get_drvdata(dev);
5450 struct igb_adapter *adapter = netdev_priv(netdev);
5451 struct pci_dev *pdev = adapter->pdev;
5452 struct e1000_hw *hw = &adapter->hw;
5453 unsigned long event = *(unsigned long *)data;
5456 case DCA_PROVIDER_ADD:
5457 /* if already enabled, don't do it again */
5458 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
5460 if (dca_add_requester(dev) == 0) {
5461 adapter->flags |= IGB_FLAG_DCA_ENABLED;
5462 dev_info(&pdev->dev, "DCA enabled\n");
5463 igb_setup_dca(adapter);
5466 /* Fall Through since DCA is disabled. */
5467 case DCA_PROVIDER_REMOVE:
5468 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
5469 /* without this a class_device is left
5470 * hanging around in the sysfs model
5472 dca_remove_requester(dev);
5473 dev_info(&pdev->dev, "DCA disabled\n");
5474 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
5475 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
5483 static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
5488 ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
5491 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
5493 #endif /* CONFIG_IGB_DCA */
5495 #ifdef CONFIG_PCI_IOV
5496 static int igb_vf_configure(struct igb_adapter *adapter, int vf)
5498 unsigned char mac_addr[ETH_ALEN];
5500 eth_zero_addr(mac_addr);
5501 igb_set_vf_mac(adapter, vf, mac_addr);
5503 /* By default spoof check is enabled for all VFs */
5504 adapter->vf_data[vf].spoofchk_enabled = true;
5510 static void igb_ping_all_vfs(struct igb_adapter *adapter)
5512 struct e1000_hw *hw = &adapter->hw;
5516 for (i = 0 ; i < adapter->vfs_allocated_count; i++) {
5517 ping = E1000_PF_CONTROL_MSG;
5518 if (adapter->vf_data[i].flags & IGB_VF_FLAG_CTS)
5519 ping |= E1000_VT_MSGTYPE_CTS;
5520 igb_write_mbx(hw, &ping, 1, i);
5524 static int igb_set_vf_promisc(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
5526 struct e1000_hw *hw = &adapter->hw;
5527 u32 vmolr = rd32(E1000_VMOLR(vf));
5528 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5530 vf_data->flags &= ~(IGB_VF_FLAG_UNI_PROMISC |
5531 IGB_VF_FLAG_MULTI_PROMISC);
5532 vmolr &= ~(E1000_VMOLR_ROPE | E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
5534 if (*msgbuf & E1000_VF_SET_PROMISC_MULTICAST) {
5535 vmolr |= E1000_VMOLR_MPME;
5536 vf_data->flags |= IGB_VF_FLAG_MULTI_PROMISC;
5537 *msgbuf &= ~E1000_VF_SET_PROMISC_MULTICAST;
5539 /* if we have hashes and we are clearing a multicast promisc
5540 * flag we need to write the hashes to the MTA as this step
5541 * was previously skipped
5543 if (vf_data->num_vf_mc_hashes > 30) {
5544 vmolr |= E1000_VMOLR_MPME;
5545 } else if (vf_data->num_vf_mc_hashes) {
5547 vmolr |= E1000_VMOLR_ROMPE;
5548 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
5549 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
5553 wr32(E1000_VMOLR(vf), vmolr);
5555 /* there are flags left unprocessed, likely not supported */
5556 if (*msgbuf & E1000_VT_MSGINFO_MASK)
5562 static int igb_set_vf_multicasts(struct igb_adapter *adapter,
5563 u32 *msgbuf, u32 vf)
5565 int n = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
5566 u16 *hash_list = (u16 *)&msgbuf[1];
5567 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5570 /* salt away the number of multicast addresses assigned
5571 * to this VF for later use to restore when the PF multi cast
5574 vf_data->num_vf_mc_hashes = n;
5576 /* only up to 30 hash values supported */
5580 /* store the hashes for later use */
5581 for (i = 0; i < n; i++)
5582 vf_data->vf_mc_hashes[i] = hash_list[i];
5584 /* Flush and reset the mta with the new values */
5585 igb_set_rx_mode(adapter->netdev);
5590 static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
5592 struct e1000_hw *hw = &adapter->hw;
5593 struct vf_data_storage *vf_data;
5596 for (i = 0; i < adapter->vfs_allocated_count; i++) {
5597 u32 vmolr = rd32(E1000_VMOLR(i));
5598 vmolr &= ~(E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
5600 vf_data = &adapter->vf_data[i];
5602 if ((vf_data->num_vf_mc_hashes > 30) ||
5603 (vf_data->flags & IGB_VF_FLAG_MULTI_PROMISC)) {
5604 vmolr |= E1000_VMOLR_MPME;
5605 } else if (vf_data->num_vf_mc_hashes) {
5606 vmolr |= E1000_VMOLR_ROMPE;
5607 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
5608 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
5610 wr32(E1000_VMOLR(i), vmolr);
5614 static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
5616 struct e1000_hw *hw = &adapter->hw;
5617 u32 pool_mask, reg, vid;
5620 pool_mask = 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
5622 /* Find the vlan filter for this id */
5623 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5624 reg = rd32(E1000_VLVF(i));
5626 /* remove the vf from the pool */
5629 /* if pool is empty then remove entry from vfta */
5630 if (!(reg & E1000_VLVF_POOLSEL_MASK) &&
5631 (reg & E1000_VLVF_VLANID_ENABLE)) {
5633 vid = reg & E1000_VLVF_VLANID_MASK;
5634 igb_vfta_set(hw, vid, false);
5637 wr32(E1000_VLVF(i), reg);
5640 adapter->vf_data[vf].vlans_enabled = 0;
5643 static s32 igb_vlvf_set(struct igb_adapter *adapter, u32 vid, bool add, u32 vf)
5645 struct e1000_hw *hw = &adapter->hw;
5648 /* The vlvf table only exists on 82576 hardware and newer */
5649 if (hw->mac.type < e1000_82576)
5652 /* we only need to do this if VMDq is enabled */
5653 if (!adapter->vfs_allocated_count)
5656 /* Find the vlan filter for this id */
5657 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5658 reg = rd32(E1000_VLVF(i));
5659 if ((reg & E1000_VLVF_VLANID_ENABLE) &&
5660 vid == (reg & E1000_VLVF_VLANID_MASK))
5665 if (i == E1000_VLVF_ARRAY_SIZE) {
5666 /* Did not find a matching VLAN ID entry that was
5667 * enabled. Search for a free filter entry, i.e.
5668 * one without the enable bit set
5670 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5671 reg = rd32(E1000_VLVF(i));
5672 if (!(reg & E1000_VLVF_VLANID_ENABLE))
5676 if (i < E1000_VLVF_ARRAY_SIZE) {
5677 /* Found an enabled/available entry */
5678 reg |= 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
5680 /* if !enabled we need to set this up in vfta */
5681 if (!(reg & E1000_VLVF_VLANID_ENABLE)) {
5682 /* add VID to filter table */
5683 igb_vfta_set(hw, vid, true);
5684 reg |= E1000_VLVF_VLANID_ENABLE;
5686 reg &= ~E1000_VLVF_VLANID_MASK;
5688 wr32(E1000_VLVF(i), reg);
5690 /* do not modify RLPML for PF devices */
5691 if (vf >= adapter->vfs_allocated_count)
5694 if (!adapter->vf_data[vf].vlans_enabled) {
5696 reg = rd32(E1000_VMOLR(vf));
5697 size = reg & E1000_VMOLR_RLPML_MASK;
5699 reg &= ~E1000_VMOLR_RLPML_MASK;
5701 wr32(E1000_VMOLR(vf), reg);
5704 adapter->vf_data[vf].vlans_enabled++;
5707 if (i < E1000_VLVF_ARRAY_SIZE) {
5708 /* remove vf from the pool */
5709 reg &= ~(1 << (E1000_VLVF_POOLSEL_SHIFT + vf));
5710 /* if pool is empty then remove entry from vfta */
5711 if (!(reg & E1000_VLVF_POOLSEL_MASK)) {
5713 igb_vfta_set(hw, vid, false);
5715 wr32(E1000_VLVF(i), reg);
5717 /* do not modify RLPML for PF devices */
5718 if (vf >= adapter->vfs_allocated_count)
5721 adapter->vf_data[vf].vlans_enabled--;
5722 if (!adapter->vf_data[vf].vlans_enabled) {
5724 reg = rd32(E1000_VMOLR(vf));
5725 size = reg & E1000_VMOLR_RLPML_MASK;
5727 reg &= ~E1000_VMOLR_RLPML_MASK;
5729 wr32(E1000_VMOLR(vf), reg);
5736 static void igb_set_vmvir(struct igb_adapter *adapter, u32 vid, u32 vf)
5738 struct e1000_hw *hw = &adapter->hw;
5741 wr32(E1000_VMVIR(vf), (vid | E1000_VMVIR_VLANA_DEFAULT));
5743 wr32(E1000_VMVIR(vf), 0);
5746 static int igb_ndo_set_vf_vlan(struct net_device *netdev,
5747 int vf, u16 vlan, u8 qos)
5750 struct igb_adapter *adapter = netdev_priv(netdev);
5752 if ((vf >= adapter->vfs_allocated_count) || (vlan > 4095) || (qos > 7))
5755 err = igb_vlvf_set(adapter, vlan, !!vlan, vf);
5758 igb_set_vmvir(adapter, vlan | (qos << VLAN_PRIO_SHIFT), vf);
5759 igb_set_vmolr(adapter, vf, !vlan);
5760 adapter->vf_data[vf].pf_vlan = vlan;
5761 adapter->vf_data[vf].pf_qos = qos;
5762 dev_info(&adapter->pdev->dev,
5763 "Setting VLAN %d, QOS 0x%x on VF %d\n", vlan, qos, vf);
5764 if (test_bit(__IGB_DOWN, &adapter->state)) {
5765 dev_warn(&adapter->pdev->dev,
5766 "The VF VLAN has been set, but the PF device is not up.\n");
5767 dev_warn(&adapter->pdev->dev,
5768 "Bring the PF device up before attempting to use the VF device.\n");
5771 igb_vlvf_set(adapter, adapter->vf_data[vf].pf_vlan,
5773 igb_set_vmvir(adapter, vlan, vf);
5774 igb_set_vmolr(adapter, vf, true);
5775 adapter->vf_data[vf].pf_vlan = 0;
5776 adapter->vf_data[vf].pf_qos = 0;
5782 static int igb_find_vlvf_entry(struct igb_adapter *adapter, int vid)
5784 struct e1000_hw *hw = &adapter->hw;
5788 /* Find the vlan filter for this id */
5789 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5790 reg = rd32(E1000_VLVF(i));
5791 if ((reg & E1000_VLVF_VLANID_ENABLE) &&
5792 vid == (reg & E1000_VLVF_VLANID_MASK))
5796 if (i >= E1000_VLVF_ARRAY_SIZE)
5802 static int igb_set_vf_vlan(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
5804 struct e1000_hw *hw = &adapter->hw;
5805 int add = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
5806 int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
5809 /* If in promiscuous mode we need to make sure the PF also has
5810 * the VLAN filter set.
5812 if (add && (adapter->netdev->flags & IFF_PROMISC))
5813 err = igb_vlvf_set(adapter, vid, add,
5814 adapter->vfs_allocated_count);
5818 err = igb_vlvf_set(adapter, vid, add, vf);
5823 /* Go through all the checks to see if the VLAN filter should
5824 * be wiped completely.
5826 if (!add && (adapter->netdev->flags & IFF_PROMISC)) {
5829 int regndx = igb_find_vlvf_entry(adapter, vid);
5832 /* See if any other pools are set for this VLAN filter
5833 * entry other than the PF.
5835 vlvf = bits = rd32(E1000_VLVF(regndx));
5836 bits &= 1 << (E1000_VLVF_POOLSEL_SHIFT +
5837 adapter->vfs_allocated_count);
5838 /* If the filter was removed then ensure PF pool bit
5839 * is cleared if the PF only added itself to the pool
5840 * because the PF is in promiscuous mode.
5842 if ((vlvf & VLAN_VID_MASK) == vid &&
5843 !test_bit(vid, adapter->active_vlans) &&
5845 igb_vlvf_set(adapter, vid, add,
5846 adapter->vfs_allocated_count);
5853 static inline void igb_vf_reset(struct igb_adapter *adapter, u32 vf)
5855 /* clear flags - except flag that indicates PF has set the MAC */
5856 adapter->vf_data[vf].flags &= IGB_VF_FLAG_PF_SET_MAC;
5857 adapter->vf_data[vf].last_nack = jiffies;
5859 /* reset offloads to defaults */
5860 igb_set_vmolr(adapter, vf, true);
5862 /* reset vlans for device */
5863 igb_clear_vf_vfta(adapter, vf);
5864 if (adapter->vf_data[vf].pf_vlan)
5865 igb_ndo_set_vf_vlan(adapter->netdev, vf,
5866 adapter->vf_data[vf].pf_vlan,
5867 adapter->vf_data[vf].pf_qos);
5869 igb_clear_vf_vfta(adapter, vf);
5871 /* reset multicast table array for vf */
5872 adapter->vf_data[vf].num_vf_mc_hashes = 0;
5874 /* Flush and reset the mta with the new values */
5875 igb_set_rx_mode(adapter->netdev);
5878 static void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
5880 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
5882 /* clear mac address as we were hotplug removed/added */
5883 if (!(adapter->vf_data[vf].flags & IGB_VF_FLAG_PF_SET_MAC))
5884 eth_zero_addr(vf_mac);
5886 /* process remaining reset events */
5887 igb_vf_reset(adapter, vf);
5890 static void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
5892 struct e1000_hw *hw = &adapter->hw;
5893 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
5894 int rar_entry = hw->mac.rar_entry_count - (vf + 1);
5896 u8 *addr = (u8 *)(&msgbuf[1]);
5898 /* process all the same items cleared in a function level reset */
5899 igb_vf_reset(adapter, vf);
5901 /* set vf mac address */
5902 igb_rar_set_qsel(adapter, vf_mac, rar_entry, vf);
5904 /* enable transmit and receive for vf */
5905 reg = rd32(E1000_VFTE);
5906 wr32(E1000_VFTE, reg | (1 << vf));
5907 reg = rd32(E1000_VFRE);
5908 wr32(E1000_VFRE, reg | (1 << vf));
5910 adapter->vf_data[vf].flags |= IGB_VF_FLAG_CTS;
5912 /* reply to reset with ack and vf mac address */
5913 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
5914 memcpy(addr, vf_mac, ETH_ALEN);
5915 igb_write_mbx(hw, msgbuf, 3, vf);
5918 static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf)
5920 /* The VF MAC Address is stored in a packed array of bytes
5921 * starting at the second 32 bit word of the msg array
5923 unsigned char *addr = (char *)&msg[1];
5926 if (is_valid_ether_addr(addr))
5927 err = igb_set_vf_mac(adapter, vf, addr);
5932 static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf)
5934 struct e1000_hw *hw = &adapter->hw;
5935 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5936 u32 msg = E1000_VT_MSGTYPE_NACK;
5938 /* if device isn't clear to send it shouldn't be reading either */
5939 if (!(vf_data->flags & IGB_VF_FLAG_CTS) &&
5940 time_after(jiffies, vf_data->last_nack + (2 * HZ))) {
5941 igb_write_mbx(hw, &msg, 1, vf);
5942 vf_data->last_nack = jiffies;
5946 static void igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
5948 struct pci_dev *pdev = adapter->pdev;
5949 u32 msgbuf[E1000_VFMAILBOX_SIZE];
5950 struct e1000_hw *hw = &adapter->hw;
5951 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5954 retval = igb_read_mbx(hw, msgbuf, E1000_VFMAILBOX_SIZE, vf);
5957 /* if receive failed revoke VF CTS stats and restart init */
5958 dev_err(&pdev->dev, "Error receiving message from VF\n");
5959 vf_data->flags &= ~IGB_VF_FLAG_CTS;
5960 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
5965 /* this is a message we already processed, do nothing */
5966 if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
5969 /* until the vf completes a reset it should not be
5970 * allowed to start any configuration.
5972 if (msgbuf[0] == E1000_VF_RESET) {
5973 igb_vf_reset_msg(adapter, vf);
5977 if (!(vf_data->flags & IGB_VF_FLAG_CTS)) {
5978 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
5984 switch ((msgbuf[0] & 0xFFFF)) {
5985 case E1000_VF_SET_MAC_ADDR:
5987 if (!(vf_data->flags & IGB_VF_FLAG_PF_SET_MAC))
5988 retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
5990 dev_warn(&pdev->dev,
5991 "VF %d attempted to override administratively set MAC address\nReload the VF driver to resume operations\n",
5994 case E1000_VF_SET_PROMISC:
5995 retval = igb_set_vf_promisc(adapter, msgbuf, vf);
5997 case E1000_VF_SET_MULTICAST:
5998 retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
6000 case E1000_VF_SET_LPE:
6001 retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf);
6003 case E1000_VF_SET_VLAN:
6005 if (vf_data->pf_vlan)
6006 dev_warn(&pdev->dev,
6007 "VF %d attempted to override administratively set VLAN tag\nReload the VF driver to resume operations\n",
6010 retval = igb_set_vf_vlan(adapter, msgbuf, vf);
6013 dev_err(&pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
6018 msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
6020 /* notify the VF of the results of what it sent us */
6022 msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
6024 msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
6026 igb_write_mbx(hw, msgbuf, 1, vf);
6029 static void igb_msg_task(struct igb_adapter *adapter)
6031 struct e1000_hw *hw = &adapter->hw;
6034 for (vf = 0; vf < adapter->vfs_allocated_count; vf++) {
6035 /* process any reset requests */
6036 if (!igb_check_for_rst(hw, vf))
6037 igb_vf_reset_event(adapter, vf);
6039 /* process any messages pending */
6040 if (!igb_check_for_msg(hw, vf))
6041 igb_rcv_msg_from_vf(adapter, vf);
6043 /* process any acks */
6044 if (!igb_check_for_ack(hw, vf))
6045 igb_rcv_ack_from_vf(adapter, vf);
6050 * igb_set_uta - Set unicast filter table address
6051 * @adapter: board private structure
6053 * The unicast table address is a register array of 32-bit registers.
6054 * The table is meant to be used in a way similar to how the MTA is used
6055 * however due to certain limitations in the hardware it is necessary to
6056 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
6057 * enable bit to allow vlan tag stripping when promiscuous mode is enabled
6059 static void igb_set_uta(struct igb_adapter *adapter)
6061 struct e1000_hw *hw = &adapter->hw;
6064 /* The UTA table only exists on 82576 hardware and newer */
6065 if (hw->mac.type < e1000_82576)
6068 /* we only need to do this if VMDq is enabled */
6069 if (!adapter->vfs_allocated_count)
6072 for (i = 0; i < hw->mac.uta_reg_count; i++)
6073 array_wr32(E1000_UTA, i, ~0);
6077 * igb_intr_msi - Interrupt Handler
6078 * @irq: interrupt number
6079 * @data: pointer to a network interface device structure
6081 static irqreturn_t igb_intr_msi(int irq, void *data)
6083 struct igb_adapter *adapter = data;
6084 struct igb_q_vector *q_vector = adapter->q_vector[0];
6085 struct e1000_hw *hw = &adapter->hw;
6086 /* read ICR disables interrupts using IAM */
6087 u32 icr = rd32(E1000_ICR);
6089 igb_write_itr(q_vector);
6091 if (icr & E1000_ICR_DRSTA)
6092 schedule_work(&adapter->reset_task);
6094 if (icr & E1000_ICR_DOUTSYNC) {
6095 /* HW is reporting DMA is out of sync */
6096 adapter->stats.doosync++;
6099 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
6100 hw->mac.get_link_status = 1;
6101 if (!test_bit(__IGB_DOWN, &adapter->state))
6102 mod_timer(&adapter->watchdog_timer, jiffies + 1);
6105 if (icr & E1000_ICR_TS) {
6106 u32 tsicr = rd32(E1000_TSICR);
6108 if (tsicr & E1000_TSICR_TXTS) {
6109 /* acknowledge the interrupt */
6110 wr32(E1000_TSICR, E1000_TSICR_TXTS);
6111 /* retrieve hardware timestamp */
6112 schedule_work(&adapter->ptp_tx_work);
6116 napi_schedule(&q_vector->napi);
6122 * igb_intr - Legacy Interrupt Handler
6123 * @irq: interrupt number
6124 * @data: pointer to a network interface device structure
6126 static irqreturn_t igb_intr(int irq, void *data)
6128 struct igb_adapter *adapter = data;
6129 struct igb_q_vector *q_vector = adapter->q_vector[0];
6130 struct e1000_hw *hw = &adapter->hw;
6131 /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked. No
6132 * need for the IMC write
6134 u32 icr = rd32(E1000_ICR);
6136 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
6137 * not set, then the adapter didn't send an interrupt
6139 if (!(icr & E1000_ICR_INT_ASSERTED))
6142 igb_write_itr(q_vector);
6144 if (icr & E1000_ICR_DRSTA)
6145 schedule_work(&adapter->reset_task);
6147 if (icr & E1000_ICR_DOUTSYNC) {
6148 /* HW is reporting DMA is out of sync */
6149 adapter->stats.doosync++;
6152 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
6153 hw->mac.get_link_status = 1;
6154 /* guard against interrupt when we're going down */
6155 if (!test_bit(__IGB_DOWN, &adapter->state))
6156 mod_timer(&adapter->watchdog_timer, jiffies + 1);
6159 if (icr & E1000_ICR_TS) {
6160 u32 tsicr = rd32(E1000_TSICR);
6162 if (tsicr & E1000_TSICR_TXTS) {
6163 /* acknowledge the interrupt */
6164 wr32(E1000_TSICR, E1000_TSICR_TXTS);
6165 /* retrieve hardware timestamp */
6166 schedule_work(&adapter->ptp_tx_work);
6170 napi_schedule(&q_vector->napi);
6175 static void igb_ring_irq_enable(struct igb_q_vector *q_vector)
6177 struct igb_adapter *adapter = q_vector->adapter;
6178 struct e1000_hw *hw = &adapter->hw;
6180 if ((q_vector->rx.ring && (adapter->rx_itr_setting & 3)) ||
6181 (!q_vector->rx.ring && (adapter->tx_itr_setting & 3))) {
6182 if ((adapter->num_q_vectors == 1) && !adapter->vf_data)
6183 igb_set_itr(q_vector);
6185 igb_update_ring_itr(q_vector);
6188 if (!test_bit(__IGB_DOWN, &adapter->state)) {
6189 if (adapter->flags & IGB_FLAG_HAS_MSIX)
6190 wr32(E1000_EIMS, q_vector->eims_value);
6192 igb_irq_enable(adapter);
6197 * igb_poll - NAPI Rx polling callback
6198 * @napi: napi polling structure
6199 * @budget: count of how many packets we should handle
6201 static int igb_poll(struct napi_struct *napi, int budget)
6203 struct igb_q_vector *q_vector = container_of(napi,
6204 struct igb_q_vector,
6206 bool clean_complete = true;
6208 #ifdef CONFIG_IGB_DCA
6209 if (q_vector->adapter->flags & IGB_FLAG_DCA_ENABLED)
6210 igb_update_dca(q_vector);
6212 if (q_vector->tx.ring)
6213 clean_complete = igb_clean_tx_irq(q_vector);
6215 if (q_vector->rx.ring)
6216 clean_complete &= igb_clean_rx_irq(q_vector, budget);
6218 /* If all work not completed, return budget and keep polling */
6219 if (!clean_complete)
6222 /* If not enough Rx work done, exit the polling mode */
6223 napi_complete(napi);
6224 igb_ring_irq_enable(q_vector);
6230 * igb_clean_tx_irq - Reclaim resources after transmit completes
6231 * @q_vector: pointer to q_vector containing needed info
6233 * returns true if ring is completely cleaned
6235 static bool igb_clean_tx_irq(struct igb_q_vector *q_vector)
6237 struct igb_adapter *adapter = q_vector->adapter;
6238 struct igb_ring *tx_ring = q_vector->tx.ring;
6239 struct igb_tx_buffer *tx_buffer;
6240 union e1000_adv_tx_desc *tx_desc;
6241 unsigned int total_bytes = 0, total_packets = 0;
6242 unsigned int budget = q_vector->tx.work_limit;
6243 unsigned int i = tx_ring->next_to_clean;
6245 if (test_bit(__IGB_DOWN, &adapter->state))
6248 tx_buffer = &tx_ring->tx_buffer_info[i];
6249 tx_desc = IGB_TX_DESC(tx_ring, i);
6250 i -= tx_ring->count;
6253 union e1000_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
6255 /* if next_to_watch is not set then there is no work pending */
6259 /* prevent any other reads prior to eop_desc */
6260 read_barrier_depends();
6262 /* if DD is not set pending work has not been completed */
6263 if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)))
6266 /* clear next_to_watch to prevent false hangs */
6267 tx_buffer->next_to_watch = NULL;
6269 /* update the statistics for this packet */
6270 total_bytes += tx_buffer->bytecount;
6271 total_packets += tx_buffer->gso_segs;
6274 dev_kfree_skb_any(tx_buffer->skb);
6276 /* unmap skb header data */
6277 dma_unmap_single(tx_ring->dev,
6278 dma_unmap_addr(tx_buffer, dma),
6279 dma_unmap_len(tx_buffer, len),
6282 /* clear tx_buffer data */
6283 tx_buffer->skb = NULL;
6284 dma_unmap_len_set(tx_buffer, len, 0);
6286 /* clear last DMA location and unmap remaining buffers */
6287 while (tx_desc != eop_desc) {
6292 i -= tx_ring->count;
6293 tx_buffer = tx_ring->tx_buffer_info;
6294 tx_desc = IGB_TX_DESC(tx_ring, 0);
6297 /* unmap any remaining paged data */
6298 if (dma_unmap_len(tx_buffer, len)) {
6299 dma_unmap_page(tx_ring->dev,
6300 dma_unmap_addr(tx_buffer, dma),
6301 dma_unmap_len(tx_buffer, len),
6303 dma_unmap_len_set(tx_buffer, len, 0);
6307 /* move us one more past the eop_desc for start of next pkt */
6312 i -= tx_ring->count;
6313 tx_buffer = tx_ring->tx_buffer_info;
6314 tx_desc = IGB_TX_DESC(tx_ring, 0);
6317 /* issue prefetch for next Tx descriptor */
6320 /* update budget accounting */
6322 } while (likely(budget));
6324 netdev_tx_completed_queue(txring_txq(tx_ring),
6325 total_packets, total_bytes);
6326 i += tx_ring->count;
6327 tx_ring->next_to_clean = i;
6328 u64_stats_update_begin(&tx_ring->tx_syncp);
6329 tx_ring->tx_stats.bytes += total_bytes;
6330 tx_ring->tx_stats.packets += total_packets;
6331 u64_stats_update_end(&tx_ring->tx_syncp);
6332 q_vector->tx.total_bytes += total_bytes;
6333 q_vector->tx.total_packets += total_packets;
6335 if (test_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags)) {
6336 struct e1000_hw *hw = &adapter->hw;
6338 /* Detect a transmit hang in hardware, this serializes the
6339 * check with the clearing of time_stamp and movement of i
6341 clear_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
6342 if (tx_buffer->next_to_watch &&
6343 time_after(jiffies, tx_buffer->time_stamp +
6344 (adapter->tx_timeout_factor * HZ)) &&
6345 !(rd32(E1000_STATUS) & E1000_STATUS_TXOFF)) {
6347 /* detected Tx unit hang */
6348 dev_err(tx_ring->dev,
6349 "Detected Tx Unit Hang\n"
6353 " next_to_use <%x>\n"
6354 " next_to_clean <%x>\n"
6355 "buffer_info[next_to_clean]\n"
6356 " time_stamp <%lx>\n"
6357 " next_to_watch <%p>\n"
6359 " desc.status <%x>\n",
6360 tx_ring->queue_index,
6361 rd32(E1000_TDH(tx_ring->reg_idx)),
6362 readl(tx_ring->tail),
6363 tx_ring->next_to_use,
6364 tx_ring->next_to_clean,
6365 tx_buffer->time_stamp,
6366 tx_buffer->next_to_watch,
6368 tx_buffer->next_to_watch->wb.status);
6369 netif_stop_subqueue(tx_ring->netdev,
6370 tx_ring->queue_index);
6372 /* we are about to reset, no point in enabling stuff */
6377 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
6378 if (unlikely(total_packets &&
6379 netif_carrier_ok(tx_ring->netdev) &&
6380 igb_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD)) {
6381 /* Make sure that anybody stopping the queue after this
6382 * sees the new next_to_clean.
6385 if (__netif_subqueue_stopped(tx_ring->netdev,
6386 tx_ring->queue_index) &&
6387 !(test_bit(__IGB_DOWN, &adapter->state))) {
6388 netif_wake_subqueue(tx_ring->netdev,
6389 tx_ring->queue_index);
6391 u64_stats_update_begin(&tx_ring->tx_syncp);
6392 tx_ring->tx_stats.restart_queue++;
6393 u64_stats_update_end(&tx_ring->tx_syncp);
6401 * igb_reuse_rx_page - page flip buffer and store it back on the ring
6402 * @rx_ring: rx descriptor ring to store buffers on
6403 * @old_buff: donor buffer to have page reused
6405 * Synchronizes page for reuse by the adapter
6407 static void igb_reuse_rx_page(struct igb_ring *rx_ring,
6408 struct igb_rx_buffer *old_buff)
6410 struct igb_rx_buffer *new_buff;
6411 u16 nta = rx_ring->next_to_alloc;
6413 new_buff = &rx_ring->rx_buffer_info[nta];
6415 /* update, and store next to alloc */
6417 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
6419 /* transfer page from old buffer to new buffer */
6420 memcpy(new_buff, old_buff, sizeof(struct igb_rx_buffer));
6422 /* sync the buffer for use by the device */
6423 dma_sync_single_range_for_device(rx_ring->dev, old_buff->dma,
6424 old_buff->page_offset,
6429 static bool igb_can_reuse_rx_page(struct igb_rx_buffer *rx_buffer,
6431 unsigned int truesize)
6433 /* avoid re-using remote pages */
6434 if (unlikely(page_to_nid(page) != numa_node_id()))
6437 #if (PAGE_SIZE < 8192)
6438 /* if we are only owner of page we can reuse it */
6439 if (unlikely(page_count(page) != 1))
6442 /* flip page offset to other buffer */
6443 rx_buffer->page_offset ^= IGB_RX_BUFSZ;
6445 /* since we are the only owner of the page and we need to
6446 * increment it, just set the value to 2 in order to avoid
6447 * an unnecessary locked operation
6449 atomic_set(&page->_count, 2);
6451 /* move offset up to the next cache line */
6452 rx_buffer->page_offset += truesize;
6454 if (rx_buffer->page_offset > (PAGE_SIZE - IGB_RX_BUFSZ))
6457 /* bump ref count on page before it is given to the stack */
6465 * igb_add_rx_frag - Add contents of Rx buffer to sk_buff
6466 * @rx_ring: rx descriptor ring to transact packets on
6467 * @rx_buffer: buffer containing page to add
6468 * @rx_desc: descriptor containing length of buffer written by hardware
6469 * @skb: sk_buff to place the data into
6471 * This function will add the data contained in rx_buffer->page to the skb.
6472 * This is done either through a direct copy if the data in the buffer is
6473 * less than the skb header size, otherwise it will just attach the page as
6474 * a frag to the skb.
6476 * The function will then update the page offset if necessary and return
6477 * true if the buffer can be reused by the adapter.
6479 static bool igb_add_rx_frag(struct igb_ring *rx_ring,
6480 struct igb_rx_buffer *rx_buffer,
6481 union e1000_adv_rx_desc *rx_desc,
6482 struct sk_buff *skb)
6484 struct page *page = rx_buffer->page;
6485 unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
6486 #if (PAGE_SIZE < 8192)
6487 unsigned int truesize = IGB_RX_BUFSZ;
6489 unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
6492 if ((size <= IGB_RX_HDR_LEN) && !skb_is_nonlinear(skb)) {
6493 unsigned char *va = page_address(page) + rx_buffer->page_offset;
6495 if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
6496 igb_ptp_rx_pktstamp(rx_ring->q_vector, va, skb);
6497 va += IGB_TS_HDR_LEN;
6498 size -= IGB_TS_HDR_LEN;
6501 memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
6503 /* we can reuse buffer as-is, just make sure it is local */
6504 if (likely(page_to_nid(page) == numa_node_id()))
6507 /* this page cannot be reused so discard it */
6512 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
6513 rx_buffer->page_offset, size, truesize);
6515 return igb_can_reuse_rx_page(rx_buffer, page, truesize);
6518 static struct sk_buff *igb_fetch_rx_buffer(struct igb_ring *rx_ring,
6519 union e1000_adv_rx_desc *rx_desc,
6520 struct sk_buff *skb)
6522 struct igb_rx_buffer *rx_buffer;
6525 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
6527 page = rx_buffer->page;
6531 void *page_addr = page_address(page) +
6532 rx_buffer->page_offset;
6534 /* prefetch first cache line of first page */
6535 prefetch(page_addr);
6536 #if L1_CACHE_BYTES < 128
6537 prefetch(page_addr + L1_CACHE_BYTES);
6540 /* allocate a skb to store the frags */
6541 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
6543 if (unlikely(!skb)) {
6544 rx_ring->rx_stats.alloc_failed++;
6548 /* we will be copying header into skb->data in
6549 * pskb_may_pull so it is in our interest to prefetch
6550 * it now to avoid a possible cache miss
6552 prefetchw(skb->data);
6555 /* we are reusing so sync this buffer for CPU use */
6556 dma_sync_single_range_for_cpu(rx_ring->dev,
6558 rx_buffer->page_offset,
6562 /* pull page into skb */
6563 if (igb_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
6564 /* hand second half of page back to the ring */
6565 igb_reuse_rx_page(rx_ring, rx_buffer);
6567 /* we are not reusing the buffer so unmap it */
6568 dma_unmap_page(rx_ring->dev, rx_buffer->dma,
6569 PAGE_SIZE, DMA_FROM_DEVICE);
6572 /* clear contents of rx_buffer */
6573 rx_buffer->page = NULL;
6578 static inline void igb_rx_checksum(struct igb_ring *ring,
6579 union e1000_adv_rx_desc *rx_desc,
6580 struct sk_buff *skb)
6582 skb_checksum_none_assert(skb);
6584 /* Ignore Checksum bit is set */
6585 if (igb_test_staterr(rx_desc, E1000_RXD_STAT_IXSM))
6588 /* Rx checksum disabled via ethtool */
6589 if (!(ring->netdev->features & NETIF_F_RXCSUM))
6592 /* TCP/UDP checksum error bit is set */
6593 if (igb_test_staterr(rx_desc,
6594 E1000_RXDEXT_STATERR_TCPE |
6595 E1000_RXDEXT_STATERR_IPE)) {
6596 /* work around errata with sctp packets where the TCPE aka
6597 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
6598 * packets, (aka let the stack check the crc32c)
6600 if (!((skb->len == 60) &&
6601 test_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags))) {
6602 u64_stats_update_begin(&ring->rx_syncp);
6603 ring->rx_stats.csum_err++;
6604 u64_stats_update_end(&ring->rx_syncp);
6606 /* let the stack verify checksum errors */
6609 /* It must be a TCP or UDP packet with a valid checksum */
6610 if (igb_test_staterr(rx_desc, E1000_RXD_STAT_TCPCS |
6611 E1000_RXD_STAT_UDPCS))
6612 skb->ip_summed = CHECKSUM_UNNECESSARY;
6614 dev_dbg(ring->dev, "cksum success: bits %08X\n",
6615 le32_to_cpu(rx_desc->wb.upper.status_error));
6618 static inline void igb_rx_hash(struct igb_ring *ring,
6619 union e1000_adv_rx_desc *rx_desc,
6620 struct sk_buff *skb)
6622 if (ring->netdev->features & NETIF_F_RXHASH)
6623 skb->rxhash = le32_to_cpu(rx_desc->wb.lower.hi_dword.rss);
6627 * igb_is_non_eop - process handling of non-EOP buffers
6628 * @rx_ring: Rx ring being processed
6629 * @rx_desc: Rx descriptor for current buffer
6630 * @skb: current socket buffer containing buffer in progress
6632 * This function updates next to clean. If the buffer is an EOP buffer
6633 * this function exits returning false, otherwise it will place the
6634 * sk_buff in the next buffer to be chained and return true indicating
6635 * that this is in fact a non-EOP buffer.
6637 static bool igb_is_non_eop(struct igb_ring *rx_ring,
6638 union e1000_adv_rx_desc *rx_desc)
6640 u32 ntc = rx_ring->next_to_clean + 1;
6642 /* fetch, update, and store next to clean */
6643 ntc = (ntc < rx_ring->count) ? ntc : 0;
6644 rx_ring->next_to_clean = ntc;
6646 prefetch(IGB_RX_DESC(rx_ring, ntc));
6648 if (likely(igb_test_staterr(rx_desc, E1000_RXD_STAT_EOP)))
6655 * igb_get_headlen - determine size of header for LRO/GRO
6656 * @data: pointer to the start of the headers
6657 * @max_len: total length of section to find headers in
6659 * This function is meant to determine the length of headers that will
6660 * be recognized by hardware for LRO, and GRO offloads. The main
6661 * motivation of doing this is to only perform one pull for IPv4 TCP
6662 * packets so that we can do basic things like calculating the gso_size
6663 * based on the average data per packet.
6665 static unsigned int igb_get_headlen(unsigned char *data,
6666 unsigned int max_len)
6669 unsigned char *network;
6672 struct vlan_hdr *vlan;
6675 struct ipv6hdr *ipv6;
6678 u8 nexthdr = 0; /* default to not TCP */
6681 /* this should never happen, but better safe than sorry */
6682 if (max_len < ETH_HLEN)
6685 /* initialize network frame pointer */
6688 /* set first protocol and move network header forward */
6689 protocol = hdr.eth->h_proto;
6690 hdr.network += ETH_HLEN;
6692 /* handle any vlan tag if present */
6693 if (protocol == __constant_htons(ETH_P_8021Q)) {
6694 if ((hdr.network - data) > (max_len - VLAN_HLEN))
6697 protocol = hdr.vlan->h_vlan_encapsulated_proto;
6698 hdr.network += VLAN_HLEN;
6701 /* handle L3 protocols */
6702 if (protocol == __constant_htons(ETH_P_IP)) {
6703 if ((hdr.network - data) > (max_len - sizeof(struct iphdr)))
6706 /* access ihl as a u8 to avoid unaligned access on ia64 */
6707 hlen = (hdr.network[0] & 0x0F) << 2;
6709 /* verify hlen meets minimum size requirements */
6710 if (hlen < sizeof(struct iphdr))
6711 return hdr.network - data;
6713 /* record next protocol if header is present */
6714 if (!(hdr.ipv4->frag_off & htons(IP_OFFSET)))
6715 nexthdr = hdr.ipv4->protocol;
6716 } else if (protocol == __constant_htons(ETH_P_IPV6)) {
6717 if ((hdr.network - data) > (max_len - sizeof(struct ipv6hdr)))
6720 /* record next protocol */
6721 nexthdr = hdr.ipv6->nexthdr;
6722 hlen = sizeof(struct ipv6hdr);
6724 return hdr.network - data;
6727 /* relocate pointer to start of L4 header */
6728 hdr.network += hlen;
6730 /* finally sort out TCP */
6731 if (nexthdr == IPPROTO_TCP) {
6732 if ((hdr.network - data) > (max_len - sizeof(struct tcphdr)))
6735 /* access doff as a u8 to avoid unaligned access on ia64 */
6736 hlen = (hdr.network[12] & 0xF0) >> 2;
6738 /* verify hlen meets minimum size requirements */
6739 if (hlen < sizeof(struct tcphdr))
6740 return hdr.network - data;
6742 hdr.network += hlen;
6743 } else if (nexthdr == IPPROTO_UDP) {
6744 if ((hdr.network - data) > (max_len - sizeof(struct udphdr)))
6747 hdr.network += sizeof(struct udphdr);
6750 /* If everything has gone correctly hdr.network should be the
6751 * data section of the packet and will be the end of the header.
6752 * If not then it probably represents the end of the last recognized
6755 if ((hdr.network - data) < max_len)
6756 return hdr.network - data;
6762 * igb_pull_tail - igb specific version of skb_pull_tail
6763 * @rx_ring: rx descriptor ring packet is being transacted on
6764 * @rx_desc: pointer to the EOP Rx descriptor
6765 * @skb: pointer to current skb being adjusted
6767 * This function is an igb specific version of __pskb_pull_tail. The
6768 * main difference between this version and the original function is that
6769 * this function can make several assumptions about the state of things
6770 * that allow for significant optimizations versus the standard function.
6771 * As a result we can do things like drop a frag and maintain an accurate
6772 * truesize for the skb.
6774 static void igb_pull_tail(struct igb_ring *rx_ring,
6775 union e1000_adv_rx_desc *rx_desc,
6776 struct sk_buff *skb)
6778 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
6780 unsigned int pull_len;
6782 /* it is valid to use page_address instead of kmap since we are
6783 * working with pages allocated out of the lomem pool per
6784 * alloc_page(GFP_ATOMIC)
6786 va = skb_frag_address(frag);
6788 if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
6789 /* retrieve timestamp from buffer */
6790 igb_ptp_rx_pktstamp(rx_ring->q_vector, va, skb);
6792 /* update pointers to remove timestamp header */
6793 skb_frag_size_sub(frag, IGB_TS_HDR_LEN);
6794 frag->page_offset += IGB_TS_HDR_LEN;
6795 skb->data_len -= IGB_TS_HDR_LEN;
6796 skb->len -= IGB_TS_HDR_LEN;
6798 /* move va to start of packet data */
6799 va += IGB_TS_HDR_LEN;
6802 /* we need the header to contain the greater of either ETH_HLEN or
6803 * 60 bytes if the skb->len is less than 60 for skb_pad.
6805 pull_len = igb_get_headlen(va, IGB_RX_HDR_LEN);
6807 /* align pull length to size of long to optimize memcpy performance */
6808 skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
6810 /* update all of the pointers */
6811 skb_frag_size_sub(frag, pull_len);
6812 frag->page_offset += pull_len;
6813 skb->data_len -= pull_len;
6814 skb->tail += pull_len;
6818 * igb_cleanup_headers - Correct corrupted or empty headers
6819 * @rx_ring: rx descriptor ring packet is being transacted on
6820 * @rx_desc: pointer to the EOP Rx descriptor
6821 * @skb: pointer to current skb being fixed
6823 * Address the case where we are pulling data in on pages only
6824 * and as such no data is present in the skb header.
6826 * In addition if skb is not at least 60 bytes we need to pad it so that
6827 * it is large enough to qualify as a valid Ethernet frame.
6829 * Returns true if an error was encountered and skb was freed.
6831 static bool igb_cleanup_headers(struct igb_ring *rx_ring,
6832 union e1000_adv_rx_desc *rx_desc,
6833 struct sk_buff *skb)
6835 if (unlikely((igb_test_staterr(rx_desc,
6836 E1000_RXDEXT_ERR_FRAME_ERR_MASK)))) {
6837 struct net_device *netdev = rx_ring->netdev;
6838 if (!(netdev->features & NETIF_F_RXALL)) {
6839 dev_kfree_skb_any(skb);
6844 /* place header in linear portion of buffer */
6845 if (skb_is_nonlinear(skb))
6846 igb_pull_tail(rx_ring, rx_desc, skb);
6848 /* if skb_pad returns an error the skb was freed */
6849 if (unlikely(skb->len < 60)) {
6850 int pad_len = 60 - skb->len;
6852 if (skb_pad(skb, pad_len))
6854 __skb_put(skb, pad_len);
6861 * igb_process_skb_fields - Populate skb header fields from Rx descriptor
6862 * @rx_ring: rx descriptor ring packet is being transacted on
6863 * @rx_desc: pointer to the EOP Rx descriptor
6864 * @skb: pointer to current skb being populated
6866 * This function checks the ring, descriptor, and packet information in
6867 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
6868 * other fields within the skb.
6870 static void igb_process_skb_fields(struct igb_ring *rx_ring,
6871 union e1000_adv_rx_desc *rx_desc,
6872 struct sk_buff *skb)
6874 struct net_device *dev = rx_ring->netdev;
6876 igb_rx_hash(rx_ring, rx_desc, skb);
6878 igb_rx_checksum(rx_ring, rx_desc, skb);
6880 igb_ptp_rx_hwtstamp(rx_ring, rx_desc, skb);
6882 if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
6883 igb_test_staterr(rx_desc, E1000_RXD_STAT_VP)) {
6885 if (igb_test_staterr(rx_desc, E1000_RXDEXT_STATERR_LB) &&
6886 test_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &rx_ring->flags))
6887 vid = be16_to_cpu(rx_desc->wb.upper.vlan);
6889 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
6891 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
6894 skb_record_rx_queue(skb, rx_ring->queue_index);
6896 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
6899 static bool igb_clean_rx_irq(struct igb_q_vector *q_vector, const int budget)
6901 struct igb_ring *rx_ring = q_vector->rx.ring;
6902 struct sk_buff *skb = rx_ring->skb;
6903 unsigned int total_bytes = 0, total_packets = 0;
6904 u16 cleaned_count = igb_desc_unused(rx_ring);
6907 union e1000_adv_rx_desc *rx_desc;
6909 /* return some buffers to hardware, one at a time is too slow */
6910 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
6911 igb_alloc_rx_buffers(rx_ring, cleaned_count);
6915 rx_desc = IGB_RX_DESC(rx_ring, rx_ring->next_to_clean);
6917 if (!igb_test_staterr(rx_desc, E1000_RXD_STAT_DD))
6920 /* This memory barrier is needed to keep us from reading
6921 * any other fields out of the rx_desc until we know the
6922 * RXD_STAT_DD bit is set
6926 /* retrieve a buffer from the ring */
6927 skb = igb_fetch_rx_buffer(rx_ring, rx_desc, skb);
6929 /* exit if we failed to retrieve a buffer */
6935 /* fetch next buffer in frame if non-eop */
6936 if (igb_is_non_eop(rx_ring, rx_desc))
6939 /* verify the packet layout is correct */
6940 if (igb_cleanup_headers(rx_ring, rx_desc, skb)) {
6945 /* probably a little skewed due to removing CRC */
6946 total_bytes += skb->len;
6948 /* populate checksum, timestamp, VLAN, and protocol */
6949 igb_process_skb_fields(rx_ring, rx_desc, skb);
6951 napi_gro_receive(&q_vector->napi, skb);
6953 /* reset skb pointer */
6956 /* update budget accounting */
6958 } while (likely(total_packets < budget));
6960 /* place incomplete frames back on ring for completion */
6963 u64_stats_update_begin(&rx_ring->rx_syncp);
6964 rx_ring->rx_stats.packets += total_packets;
6965 rx_ring->rx_stats.bytes += total_bytes;
6966 u64_stats_update_end(&rx_ring->rx_syncp);
6967 q_vector->rx.total_packets += total_packets;
6968 q_vector->rx.total_bytes += total_bytes;
6971 igb_alloc_rx_buffers(rx_ring, cleaned_count);
6973 return (total_packets < budget);
6976 static bool igb_alloc_mapped_page(struct igb_ring *rx_ring,
6977 struct igb_rx_buffer *bi)
6979 struct page *page = bi->page;
6982 /* since we are recycling buffers we should seldom need to alloc */
6986 /* alloc new page for storage */
6987 page = __skb_alloc_page(GFP_ATOMIC | __GFP_COLD, NULL);
6988 if (unlikely(!page)) {
6989 rx_ring->rx_stats.alloc_failed++;
6993 /* map page for use */
6994 dma = dma_map_page(rx_ring->dev, page, 0, PAGE_SIZE, DMA_FROM_DEVICE);
6996 /* if mapping failed free memory back to system since
6997 * there isn't much point in holding memory we can't use
6999 if (dma_mapping_error(rx_ring->dev, dma)) {
7002 rx_ring->rx_stats.alloc_failed++;
7008 bi->page_offset = 0;
7014 * igb_alloc_rx_buffers - Replace used receive buffers; packet split
7015 * @adapter: address of board private structure
7017 void igb_alloc_rx_buffers(struct igb_ring *rx_ring, u16 cleaned_count)
7019 union e1000_adv_rx_desc *rx_desc;
7020 struct igb_rx_buffer *bi;
7021 u16 i = rx_ring->next_to_use;
7027 rx_desc = IGB_RX_DESC(rx_ring, i);
7028 bi = &rx_ring->rx_buffer_info[i];
7029 i -= rx_ring->count;
7032 if (!igb_alloc_mapped_page(rx_ring, bi))
7035 /* Refresh the desc even if buffer_addrs didn't change
7036 * because each write-back erases this info.
7038 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
7044 rx_desc = IGB_RX_DESC(rx_ring, 0);
7045 bi = rx_ring->rx_buffer_info;
7046 i -= rx_ring->count;
7049 /* clear the hdr_addr for the next_to_use descriptor */
7050 rx_desc->read.hdr_addr = 0;
7053 } while (cleaned_count);
7055 i += rx_ring->count;
7057 if (rx_ring->next_to_use != i) {
7058 /* record the next descriptor to use */
7059 rx_ring->next_to_use = i;
7061 /* update next to alloc since we have filled the ring */
7062 rx_ring->next_to_alloc = i;
7064 /* Force memory writes to complete before letting h/w
7065 * know there are new descriptors to fetch. (Only
7066 * applicable for weak-ordered memory model archs,
7070 writel(i, rx_ring->tail);
7080 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
7082 struct igb_adapter *adapter = netdev_priv(netdev);
7083 struct mii_ioctl_data *data = if_mii(ifr);
7085 if (adapter->hw.phy.media_type != e1000_media_type_copper)
7090 data->phy_id = adapter->hw.phy.addr;
7093 if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
7110 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
7116 return igb_mii_ioctl(netdev, ifr, cmd);
7118 return igb_ptp_hwtstamp_ioctl(netdev, ifr, cmd);
7124 s32 igb_read_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
7126 struct igb_adapter *adapter = hw->back;
7128 if (pcie_capability_read_word(adapter->pdev, reg, value))
7129 return -E1000_ERR_CONFIG;
7134 s32 igb_write_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
7136 struct igb_adapter *adapter = hw->back;
7138 if (pcie_capability_write_word(adapter->pdev, reg, *value))
7139 return -E1000_ERR_CONFIG;
7144 static void igb_vlan_mode(struct net_device *netdev, netdev_features_t features)
7146 struct igb_adapter *adapter = netdev_priv(netdev);
7147 struct e1000_hw *hw = &adapter->hw;
7149 bool enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
7152 /* enable VLAN tag insert/strip */
7153 ctrl = rd32(E1000_CTRL);
7154 ctrl |= E1000_CTRL_VME;
7155 wr32(E1000_CTRL, ctrl);
7157 /* Disable CFI check */
7158 rctl = rd32(E1000_RCTL);
7159 rctl &= ~E1000_RCTL_CFIEN;
7160 wr32(E1000_RCTL, rctl);
7162 /* disable VLAN tag insert/strip */
7163 ctrl = rd32(E1000_CTRL);
7164 ctrl &= ~E1000_CTRL_VME;
7165 wr32(E1000_CTRL, ctrl);
7168 igb_rlpml_set(adapter);
7171 static int igb_vlan_rx_add_vid(struct net_device *netdev,
7172 __be16 proto, u16 vid)
7174 struct igb_adapter *adapter = netdev_priv(netdev);
7175 struct e1000_hw *hw = &adapter->hw;
7176 int pf_id = adapter->vfs_allocated_count;
7178 /* attempt to add filter to vlvf array */
7179 igb_vlvf_set(adapter, vid, true, pf_id);
7181 /* add the filter since PF can receive vlans w/o entry in vlvf */
7182 igb_vfta_set(hw, vid, true);
7184 set_bit(vid, adapter->active_vlans);
7189 static int igb_vlan_rx_kill_vid(struct net_device *netdev,
7190 __be16 proto, u16 vid)
7192 struct igb_adapter *adapter = netdev_priv(netdev);
7193 struct e1000_hw *hw = &adapter->hw;
7194 int pf_id = adapter->vfs_allocated_count;
7197 /* remove vlan from VLVF table array */
7198 err = igb_vlvf_set(adapter, vid, false, pf_id);
7200 /* if vid was not present in VLVF just remove it from table */
7202 igb_vfta_set(hw, vid, false);
7204 clear_bit(vid, adapter->active_vlans);
7209 static void igb_restore_vlan(struct igb_adapter *adapter)
7213 igb_vlan_mode(adapter->netdev, adapter->netdev->features);
7215 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
7216 igb_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
7219 int igb_set_spd_dplx(struct igb_adapter *adapter, u32 spd, u8 dplx)
7221 struct pci_dev *pdev = adapter->pdev;
7222 struct e1000_mac_info *mac = &adapter->hw.mac;
7226 /* Make sure dplx is at most 1 bit and lsb of speed is not set
7227 * for the switch() below to work
7229 if ((spd & 1) || (dplx & ~1))
7232 /* Fiber NIC's only allow 1000 gbps Full duplex
7233 * and 100Mbps Full duplex for 100baseFx sfp
7235 if (adapter->hw.phy.media_type == e1000_media_type_internal_serdes) {
7236 switch (spd + dplx) {
7237 case SPEED_10 + DUPLEX_HALF:
7238 case SPEED_10 + DUPLEX_FULL:
7239 case SPEED_100 + DUPLEX_HALF:
7246 switch (spd + dplx) {
7247 case SPEED_10 + DUPLEX_HALF:
7248 mac->forced_speed_duplex = ADVERTISE_10_HALF;
7250 case SPEED_10 + DUPLEX_FULL:
7251 mac->forced_speed_duplex = ADVERTISE_10_FULL;
7253 case SPEED_100 + DUPLEX_HALF:
7254 mac->forced_speed_duplex = ADVERTISE_100_HALF;
7256 case SPEED_100 + DUPLEX_FULL:
7257 mac->forced_speed_duplex = ADVERTISE_100_FULL;
7259 case SPEED_1000 + DUPLEX_FULL:
7261 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
7263 case SPEED_1000 + DUPLEX_HALF: /* not supported */
7268 /* clear MDI, MDI(-X) override is only allowed when autoneg enabled */
7269 adapter->hw.phy.mdix = AUTO_ALL_MODES;
7274 dev_err(&pdev->dev, "Unsupported Speed/Duplex configuration\n");
7278 static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake,
7281 struct net_device *netdev = pci_get_drvdata(pdev);
7282 struct igb_adapter *adapter = netdev_priv(netdev);
7283 struct e1000_hw *hw = &adapter->hw;
7284 u32 ctrl, rctl, status;
7285 u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
7290 netif_device_detach(netdev);
7292 if (netif_running(netdev))
7293 __igb_close(netdev, true);
7295 igb_clear_interrupt_scheme(adapter);
7298 retval = pci_save_state(pdev);
7303 status = rd32(E1000_STATUS);
7304 if (status & E1000_STATUS_LU)
7305 wufc &= ~E1000_WUFC_LNKC;
7308 igb_setup_rctl(adapter);
7309 igb_set_rx_mode(netdev);
7311 /* turn on all-multi mode if wake on multicast is enabled */
7312 if (wufc & E1000_WUFC_MC) {
7313 rctl = rd32(E1000_RCTL);
7314 rctl |= E1000_RCTL_MPE;
7315 wr32(E1000_RCTL, rctl);
7318 ctrl = rd32(E1000_CTRL);
7319 /* advertise wake from D3Cold */
7320 #define E1000_CTRL_ADVD3WUC 0x00100000
7321 /* phy power management enable */
7322 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
7323 ctrl |= E1000_CTRL_ADVD3WUC;
7324 wr32(E1000_CTRL, ctrl);
7326 /* Allow time for pending master requests to run */
7327 igb_disable_pcie_master(hw);
7329 wr32(E1000_WUC, E1000_WUC_PME_EN);
7330 wr32(E1000_WUFC, wufc);
7333 wr32(E1000_WUFC, 0);
7336 *enable_wake = wufc || adapter->en_mng_pt;
7338 igb_power_down_link(adapter);
7340 igb_power_up_link(adapter);
7342 /* Release control of h/w to f/w. If f/w is AMT enabled, this
7343 * would have already happened in close and is redundant.
7345 igb_release_hw_control(adapter);
7347 pci_disable_device(pdev);
7353 #ifdef CONFIG_PM_SLEEP
7354 static int igb_suspend(struct device *dev)
7358 struct pci_dev *pdev = to_pci_dev(dev);
7360 retval = __igb_shutdown(pdev, &wake, 0);
7365 pci_prepare_to_sleep(pdev);
7367 pci_wake_from_d3(pdev, false);
7368 pci_set_power_state(pdev, PCI_D3hot);
7373 #endif /* CONFIG_PM_SLEEP */
7375 static int igb_resume(struct device *dev)
7377 struct pci_dev *pdev = to_pci_dev(dev);
7378 struct net_device *netdev = pci_get_drvdata(pdev);
7379 struct igb_adapter *adapter = netdev_priv(netdev);
7380 struct e1000_hw *hw = &adapter->hw;
7383 pci_set_power_state(pdev, PCI_D0);
7384 pci_restore_state(pdev);
7385 pci_save_state(pdev);
7387 err = pci_enable_device_mem(pdev);
7390 "igb: Cannot enable PCI device from suspend\n");
7393 pci_set_master(pdev);
7395 pci_enable_wake(pdev, PCI_D3hot, 0);
7396 pci_enable_wake(pdev, PCI_D3cold, 0);
7398 if (igb_init_interrupt_scheme(adapter, true)) {
7399 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
7405 /* let the f/w know that the h/w is now under the control of the
7408 igb_get_hw_control(adapter);
7410 wr32(E1000_WUS, ~0);
7412 if (netdev->flags & IFF_UP) {
7414 err = __igb_open(netdev, true);
7420 netif_device_attach(netdev);
7424 #ifdef CONFIG_PM_RUNTIME
7425 static int igb_runtime_idle(struct device *dev)
7427 struct pci_dev *pdev = to_pci_dev(dev);
7428 struct net_device *netdev = pci_get_drvdata(pdev);
7429 struct igb_adapter *adapter = netdev_priv(netdev);
7431 if (!igb_has_link(adapter))
7432 pm_schedule_suspend(dev, MSEC_PER_SEC * 5);
7437 static int igb_runtime_suspend(struct device *dev)
7439 struct pci_dev *pdev = to_pci_dev(dev);
7443 retval = __igb_shutdown(pdev, &wake, 1);
7448 pci_prepare_to_sleep(pdev);
7450 pci_wake_from_d3(pdev, false);
7451 pci_set_power_state(pdev, PCI_D3hot);
7457 static int igb_runtime_resume(struct device *dev)
7459 return igb_resume(dev);
7461 #endif /* CONFIG_PM_RUNTIME */
7464 static void igb_shutdown(struct pci_dev *pdev)
7468 __igb_shutdown(pdev, &wake, 0);
7470 if (system_state == SYSTEM_POWER_OFF) {
7471 pci_wake_from_d3(pdev, wake);
7472 pci_set_power_state(pdev, PCI_D3hot);
7476 #ifdef CONFIG_PCI_IOV
7477 static int igb_sriov_reinit(struct pci_dev *dev)
7479 struct net_device *netdev = pci_get_drvdata(dev);
7480 struct igb_adapter *adapter = netdev_priv(netdev);
7481 struct pci_dev *pdev = adapter->pdev;
7485 if (netif_running(netdev))
7488 igb_clear_interrupt_scheme(adapter);
7490 igb_init_queue_configuration(adapter);
7492 if (igb_init_interrupt_scheme(adapter, true)) {
7493 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
7497 if (netif_running(netdev))
7505 static int igb_pci_disable_sriov(struct pci_dev *dev)
7507 int err = igb_disable_sriov(dev);
7510 err = igb_sriov_reinit(dev);
7515 static int igb_pci_enable_sriov(struct pci_dev *dev, int num_vfs)
7517 int err = igb_enable_sriov(dev, num_vfs);
7522 err = igb_sriov_reinit(dev);
7531 static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs)
7533 #ifdef CONFIG_PCI_IOV
7535 return igb_pci_disable_sriov(dev);
7537 return igb_pci_enable_sriov(dev, num_vfs);
7542 #ifdef CONFIG_NET_POLL_CONTROLLER
7543 /* Polling 'interrupt' - used by things like netconsole to send skbs
7544 * without having to re-enable interrupts. It's not called while
7545 * the interrupt routine is executing.
7547 static void igb_netpoll(struct net_device *netdev)
7549 struct igb_adapter *adapter = netdev_priv(netdev);
7550 struct e1000_hw *hw = &adapter->hw;
7551 struct igb_q_vector *q_vector;
7554 for (i = 0; i < adapter->num_q_vectors; i++) {
7555 q_vector = adapter->q_vector[i];
7556 if (adapter->flags & IGB_FLAG_HAS_MSIX)
7557 wr32(E1000_EIMC, q_vector->eims_value);
7559 igb_irq_disable(adapter);
7560 napi_schedule(&q_vector->napi);
7563 #endif /* CONFIG_NET_POLL_CONTROLLER */
7566 * igb_io_error_detected - called when PCI error is detected
7567 * @pdev: Pointer to PCI device
7568 * @state: The current pci connection state
7570 * This function is called after a PCI bus error affecting
7571 * this device has been detected.
7573 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
7574 pci_channel_state_t state)
7576 struct net_device *netdev = pci_get_drvdata(pdev);
7577 struct igb_adapter *adapter = netdev_priv(netdev);
7579 netif_device_detach(netdev);
7581 if (state == pci_channel_io_perm_failure)
7582 return PCI_ERS_RESULT_DISCONNECT;
7584 if (netif_running(netdev))
7586 pci_disable_device(pdev);
7588 /* Request a slot slot reset. */
7589 return PCI_ERS_RESULT_NEED_RESET;
7593 * igb_io_slot_reset - called after the pci bus has been reset.
7594 * @pdev: Pointer to PCI device
7596 * Restart the card from scratch, as if from a cold-boot. Implementation
7597 * resembles the first-half of the igb_resume routine.
7599 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
7601 struct net_device *netdev = pci_get_drvdata(pdev);
7602 struct igb_adapter *adapter = netdev_priv(netdev);
7603 struct e1000_hw *hw = &adapter->hw;
7604 pci_ers_result_t result;
7607 if (pci_enable_device_mem(pdev)) {
7609 "Cannot re-enable PCI device after reset.\n");
7610 result = PCI_ERS_RESULT_DISCONNECT;
7612 pci_set_master(pdev);
7613 pci_restore_state(pdev);
7614 pci_save_state(pdev);
7616 pci_enable_wake(pdev, PCI_D3hot, 0);
7617 pci_enable_wake(pdev, PCI_D3cold, 0);
7620 wr32(E1000_WUS, ~0);
7621 result = PCI_ERS_RESULT_RECOVERED;
7624 err = pci_cleanup_aer_uncorrect_error_status(pdev);
7627 "pci_cleanup_aer_uncorrect_error_status failed 0x%0x\n",
7629 /* non-fatal, continue */
7636 * igb_io_resume - called when traffic can start flowing again.
7637 * @pdev: Pointer to PCI device
7639 * This callback is called when the error recovery driver tells us that
7640 * its OK to resume normal operation. Implementation resembles the
7641 * second-half of the igb_resume routine.
7643 static void igb_io_resume(struct pci_dev *pdev)
7645 struct net_device *netdev = pci_get_drvdata(pdev);
7646 struct igb_adapter *adapter = netdev_priv(netdev);
7648 if (netif_running(netdev)) {
7649 if (igb_up(adapter)) {
7650 dev_err(&pdev->dev, "igb_up failed after reset\n");
7655 netif_device_attach(netdev);
7657 /* let the f/w know that the h/w is now under the control of the
7660 igb_get_hw_control(adapter);
7663 static void igb_rar_set_qsel(struct igb_adapter *adapter, u8 *addr, u32 index,
7666 u32 rar_low, rar_high;
7667 struct e1000_hw *hw = &adapter->hw;
7669 /* HW expects these in little endian so we reverse the byte order
7670 * from network order (big endian) to little endian
7672 rar_low = ((u32) addr[0] | ((u32) addr[1] << 8) |
7673 ((u32) addr[2] << 16) | ((u32) addr[3] << 24));
7674 rar_high = ((u32) addr[4] | ((u32) addr[5] << 8));
7676 /* Indicate to hardware the Address is Valid. */
7677 rar_high |= E1000_RAH_AV;
7679 if (hw->mac.type == e1000_82575)
7680 rar_high |= E1000_RAH_POOL_1 * qsel;
7682 rar_high |= E1000_RAH_POOL_1 << qsel;
7684 wr32(E1000_RAL(index), rar_low);
7686 wr32(E1000_RAH(index), rar_high);
7690 static int igb_set_vf_mac(struct igb_adapter *adapter,
7691 int vf, unsigned char *mac_addr)
7693 struct e1000_hw *hw = &adapter->hw;
7694 /* VF MAC addresses start at end of receive addresses and moves
7695 * towards the first, as a result a collision should not be possible
7697 int rar_entry = hw->mac.rar_entry_count - (vf + 1);
7699 memcpy(adapter->vf_data[vf].vf_mac_addresses, mac_addr, ETH_ALEN);
7701 igb_rar_set_qsel(adapter, mac_addr, rar_entry, vf);
7706 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
7708 struct igb_adapter *adapter = netdev_priv(netdev);
7709 if (!is_valid_ether_addr(mac) || (vf >= adapter->vfs_allocated_count))
7711 adapter->vf_data[vf].flags |= IGB_VF_FLAG_PF_SET_MAC;
7712 dev_info(&adapter->pdev->dev, "setting MAC %pM on VF %d\n", mac, vf);
7713 dev_info(&adapter->pdev->dev,
7714 "Reload the VF driver to make this change effective.");
7715 if (test_bit(__IGB_DOWN, &adapter->state)) {
7716 dev_warn(&adapter->pdev->dev,
7717 "The VF MAC address has been set, but the PF device is not up.\n");
7718 dev_warn(&adapter->pdev->dev,
7719 "Bring the PF device up before attempting to use the VF device.\n");
7721 return igb_set_vf_mac(adapter, vf, mac);
7724 static int igb_link_mbps(int internal_link_speed)
7726 switch (internal_link_speed) {
7736 static void igb_set_vf_rate_limit(struct e1000_hw *hw, int vf, int tx_rate,
7743 /* Calculate the rate factor values to set */
7744 rf_int = link_speed / tx_rate;
7745 rf_dec = (link_speed - (rf_int * tx_rate));
7746 rf_dec = (rf_dec * (1 << E1000_RTTBCNRC_RF_INT_SHIFT)) /
7749 bcnrc_val = E1000_RTTBCNRC_RS_ENA;
7750 bcnrc_val |= ((rf_int << E1000_RTTBCNRC_RF_INT_SHIFT) &
7751 E1000_RTTBCNRC_RF_INT_MASK);
7752 bcnrc_val |= (rf_dec & E1000_RTTBCNRC_RF_DEC_MASK);
7757 wr32(E1000_RTTDQSEL, vf); /* vf X uses queue X */
7758 /* Set global transmit compensation time to the MMW_SIZE in RTTBCNRM
7759 * register. MMW_SIZE=0x014 if 9728-byte jumbo is supported.
7761 wr32(E1000_RTTBCNRM, 0x14);
7762 wr32(E1000_RTTBCNRC, bcnrc_val);
7765 static void igb_check_vf_rate_limit(struct igb_adapter *adapter)
7767 int actual_link_speed, i;
7768 bool reset_rate = false;
7770 /* VF TX rate limit was not set or not supported */
7771 if ((adapter->vf_rate_link_speed == 0) ||
7772 (adapter->hw.mac.type != e1000_82576))
7775 actual_link_speed = igb_link_mbps(adapter->link_speed);
7776 if (actual_link_speed != adapter->vf_rate_link_speed) {
7778 adapter->vf_rate_link_speed = 0;
7779 dev_info(&adapter->pdev->dev,
7780 "Link speed has been changed. VF Transmit rate is disabled\n");
7783 for (i = 0; i < adapter->vfs_allocated_count; i++) {
7785 adapter->vf_data[i].tx_rate = 0;
7787 igb_set_vf_rate_limit(&adapter->hw, i,
7788 adapter->vf_data[i].tx_rate,
7793 static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf, int tx_rate)
7795 struct igb_adapter *adapter = netdev_priv(netdev);
7796 struct e1000_hw *hw = &adapter->hw;
7797 int actual_link_speed;
7799 if (hw->mac.type != e1000_82576)
7802 actual_link_speed = igb_link_mbps(adapter->link_speed);
7803 if ((vf >= adapter->vfs_allocated_count) ||
7804 (!(rd32(E1000_STATUS) & E1000_STATUS_LU)) ||
7805 (tx_rate < 0) || (tx_rate > actual_link_speed))
7808 adapter->vf_rate_link_speed = actual_link_speed;
7809 adapter->vf_data[vf].tx_rate = (u16)tx_rate;
7810 igb_set_vf_rate_limit(hw, vf, tx_rate, actual_link_speed);
7815 static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf,
7818 struct igb_adapter *adapter = netdev_priv(netdev);
7819 struct e1000_hw *hw = &adapter->hw;
7820 u32 reg_val, reg_offset;
7822 if (!adapter->vfs_allocated_count)
7825 if (vf >= adapter->vfs_allocated_count)
7828 reg_offset = (hw->mac.type == e1000_82576) ? E1000_DTXSWC : E1000_TXSWC;
7829 reg_val = rd32(reg_offset);
7831 reg_val |= ((1 << vf) |
7832 (1 << (vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT)));
7834 reg_val &= ~((1 << vf) |
7835 (1 << (vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT)));
7836 wr32(reg_offset, reg_val);
7838 adapter->vf_data[vf].spoofchk_enabled = setting;
7839 return E1000_SUCCESS;
7842 static int igb_ndo_get_vf_config(struct net_device *netdev,
7843 int vf, struct ifla_vf_info *ivi)
7845 struct igb_adapter *adapter = netdev_priv(netdev);
7846 if (vf >= adapter->vfs_allocated_count)
7849 memcpy(&ivi->mac, adapter->vf_data[vf].vf_mac_addresses, ETH_ALEN);
7850 ivi->tx_rate = adapter->vf_data[vf].tx_rate;
7851 ivi->vlan = adapter->vf_data[vf].pf_vlan;
7852 ivi->qos = adapter->vf_data[vf].pf_qos;
7853 ivi->spoofchk = adapter->vf_data[vf].spoofchk_enabled;
7857 static void igb_vmm_control(struct igb_adapter *adapter)
7859 struct e1000_hw *hw = &adapter->hw;
7862 switch (hw->mac.type) {
7868 /* replication is not supported for 82575 */
7871 /* notify HW that the MAC is adding vlan tags */
7872 reg = rd32(E1000_DTXCTL);
7873 reg |= E1000_DTXCTL_VLAN_ADDED;
7874 wr32(E1000_DTXCTL, reg);
7876 /* enable replication vlan tag stripping */
7877 reg = rd32(E1000_RPLOLR);
7878 reg |= E1000_RPLOLR_STRVLAN;
7879 wr32(E1000_RPLOLR, reg);
7881 /* none of the above registers are supported by i350 */
7885 if (adapter->vfs_allocated_count) {
7886 igb_vmdq_set_loopback_pf(hw, true);
7887 igb_vmdq_set_replication_pf(hw, true);
7888 igb_vmdq_set_anti_spoofing_pf(hw, true,
7889 adapter->vfs_allocated_count);
7891 igb_vmdq_set_loopback_pf(hw, false);
7892 igb_vmdq_set_replication_pf(hw, false);
7896 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba)
7898 struct e1000_hw *hw = &adapter->hw;
7902 if (hw->mac.type > e1000_82580) {
7903 if (adapter->flags & IGB_FLAG_DMAC) {
7906 /* force threshold to 0. */
7907 wr32(E1000_DMCTXTH, 0);
7909 /* DMA Coalescing high water mark needs to be greater
7910 * than the Rx threshold. Set hwm to PBA - max frame
7911 * size in 16B units, capping it at PBA - 6KB.
7913 hwm = 64 * pba - adapter->max_frame_size / 16;
7914 if (hwm < 64 * (pba - 6))
7915 hwm = 64 * (pba - 6);
7916 reg = rd32(E1000_FCRTC);
7917 reg &= ~E1000_FCRTC_RTH_COAL_MASK;
7918 reg |= ((hwm << E1000_FCRTC_RTH_COAL_SHIFT)
7919 & E1000_FCRTC_RTH_COAL_MASK);
7920 wr32(E1000_FCRTC, reg);
7922 /* Set the DMA Coalescing Rx threshold to PBA - 2 * max
7923 * frame size, capping it at PBA - 10KB.
7925 dmac_thr = pba - adapter->max_frame_size / 512;
7926 if (dmac_thr < pba - 10)
7927 dmac_thr = pba - 10;
7928 reg = rd32(E1000_DMACR);
7929 reg &= ~E1000_DMACR_DMACTHR_MASK;
7930 reg |= ((dmac_thr << E1000_DMACR_DMACTHR_SHIFT)
7931 & E1000_DMACR_DMACTHR_MASK);
7933 /* transition to L0x or L1 if available..*/
7934 reg |= (E1000_DMACR_DMAC_EN | E1000_DMACR_DMAC_LX_MASK);
7936 /* watchdog timer= +-1000 usec in 32usec intervals */
7939 /* Disable BMC-to-OS Watchdog Enable */
7940 if (hw->mac.type != e1000_i354)
7941 reg &= ~E1000_DMACR_DC_BMC2OSW_EN;
7943 wr32(E1000_DMACR, reg);
7945 /* no lower threshold to disable
7946 * coalescing(smart fifb)-UTRESH=0
7948 wr32(E1000_DMCRTRH, 0);
7950 reg = (IGB_DMCTLX_DCFLUSH_DIS | 0x4);
7952 wr32(E1000_DMCTLX, reg);
7954 /* free space in tx packet buffer to wake from
7957 wr32(E1000_DMCTXTH, (IGB_MIN_TXPBSIZE -
7958 (IGB_TX_BUF_4096 + adapter->max_frame_size)) >> 6);
7960 /* make low power state decision controlled
7963 reg = rd32(E1000_PCIEMISC);
7964 reg &= ~E1000_PCIEMISC_LX_DECISION;
7965 wr32(E1000_PCIEMISC, reg);
7966 } /* endif adapter->dmac is not disabled */
7967 } else if (hw->mac.type == e1000_82580) {
7968 u32 reg = rd32(E1000_PCIEMISC);
7969 wr32(E1000_PCIEMISC, reg & ~E1000_PCIEMISC_LX_DECISION);
7970 wr32(E1000_DMACR, 0);
7975 * igb_read_i2c_byte - Reads 8 bit word over I2C
7976 * @hw: pointer to hardware structure
7977 * @byte_offset: byte offset to read
7978 * @dev_addr: device address
7981 * Performs byte read operation over I2C interface at
7982 * a specified device address.
7984 s32 igb_read_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
7985 u8 dev_addr, u8 *data)
7987 struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
7988 struct i2c_client *this_client = adapter->i2c_client;
7993 return E1000_ERR_I2C;
7995 swfw_mask = E1000_SWFW_PHY0_SM;
7997 if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask)
7999 return E1000_ERR_SWFW_SYNC;
8001 status = i2c_smbus_read_byte_data(this_client, byte_offset);
8002 hw->mac.ops.release_swfw_sync(hw, swfw_mask);
8005 return E1000_ERR_I2C;
8008 return E1000_SUCCESS;
8013 * igb_write_i2c_byte - Writes 8 bit word over I2C
8014 * @hw: pointer to hardware structure
8015 * @byte_offset: byte offset to write
8016 * @dev_addr: device address
8017 * @data: value to write
8019 * Performs byte write operation over I2C interface at
8020 * a specified device address.
8022 s32 igb_write_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
8023 u8 dev_addr, u8 data)
8025 struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
8026 struct i2c_client *this_client = adapter->i2c_client;
8028 u16 swfw_mask = E1000_SWFW_PHY0_SM;
8031 return E1000_ERR_I2C;
8033 if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask) != E1000_SUCCESS)
8034 return E1000_ERR_SWFW_SYNC;
8035 status = i2c_smbus_write_byte_data(this_client, byte_offset, data);
8036 hw->mac.ops.release_swfw_sync(hw, swfw_mask);
8039 return E1000_ERR_I2C;
8041 return E1000_SUCCESS;
8045 int igb_reinit_queues(struct igb_adapter *adapter)
8047 struct net_device *netdev = adapter->netdev;
8048 struct pci_dev *pdev = adapter->pdev;
8051 if (netif_running(netdev))
8054 igb_reset_interrupt_capability(adapter);
8056 if (igb_init_interrupt_scheme(adapter, true)) {
8057 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
8061 if (netif_running(netdev))
8062 err = igb_open(netdev);