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_I211_COPPER), board_82575 },
81 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER), board_82575 },
82 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_FIBER), board_82575 },
83 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES), board_82575 },
84 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SGMII), board_82575 },
85 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_COPPER), board_82575 },
86 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_FIBER), board_82575 },
87 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SERDES), board_82575 },
88 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SGMII), board_82575 },
89 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER), board_82575 },
90 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_FIBER), board_82575 },
91 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_QUAD_FIBER), board_82575 },
92 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SERDES), board_82575 },
93 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SGMII), board_82575 },
94 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER_DUAL), board_82575 },
95 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SGMII), board_82575 },
96 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SERDES), board_82575 },
97 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_BACKPLANE), board_82575 },
98 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SFP), board_82575 },
99 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576), board_82575 },
100 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS), board_82575 },
101 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS_SERDES), board_82575 },
102 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_FIBER), board_82575 },
103 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES), board_82575 },
104 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES_QUAD), board_82575 },
105 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER_ET2), board_82575 },
106 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER), board_82575 },
107 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
108 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
109 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
110 /* required last entry */
114 MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
116 void igb_reset(struct igb_adapter *);
117 static int igb_setup_all_tx_resources(struct igb_adapter *);
118 static int igb_setup_all_rx_resources(struct igb_adapter *);
119 static void igb_free_all_tx_resources(struct igb_adapter *);
120 static void igb_free_all_rx_resources(struct igb_adapter *);
121 static void igb_setup_mrqc(struct igb_adapter *);
122 static int igb_probe(struct pci_dev *, const struct pci_device_id *);
123 static void igb_remove(struct pci_dev *pdev);
124 static int igb_sw_init(struct igb_adapter *);
125 static int igb_open(struct net_device *);
126 static int igb_close(struct net_device *);
127 static void igb_configure(struct igb_adapter *);
128 static void igb_configure_tx(struct igb_adapter *);
129 static void igb_configure_rx(struct igb_adapter *);
130 static void igb_clean_all_tx_rings(struct igb_adapter *);
131 static void igb_clean_all_rx_rings(struct igb_adapter *);
132 static void igb_clean_tx_ring(struct igb_ring *);
133 static void igb_clean_rx_ring(struct igb_ring *);
134 static void igb_set_rx_mode(struct net_device *);
135 static void igb_update_phy_info(unsigned long);
136 static void igb_watchdog(unsigned long);
137 static void igb_watchdog_task(struct work_struct *);
138 static netdev_tx_t igb_xmit_frame(struct sk_buff *skb, struct net_device *);
139 static struct rtnl_link_stats64 *igb_get_stats64(struct net_device *dev,
140 struct rtnl_link_stats64 *stats);
141 static int igb_change_mtu(struct net_device *, int);
142 static int igb_set_mac(struct net_device *, void *);
143 static void igb_set_uta(struct igb_adapter *adapter);
144 static irqreturn_t igb_intr(int irq, void *);
145 static irqreturn_t igb_intr_msi(int irq, void *);
146 static irqreturn_t igb_msix_other(int irq, void *);
147 static irqreturn_t igb_msix_ring(int irq, void *);
148 #ifdef CONFIG_IGB_DCA
149 static void igb_update_dca(struct igb_q_vector *);
150 static void igb_setup_dca(struct igb_adapter *);
151 #endif /* CONFIG_IGB_DCA */
152 static int igb_poll(struct napi_struct *, int);
153 static bool igb_clean_tx_irq(struct igb_q_vector *);
154 static bool igb_clean_rx_irq(struct igb_q_vector *, int);
155 static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
156 static void igb_tx_timeout(struct net_device *);
157 static void igb_reset_task(struct work_struct *);
158 static void igb_vlan_mode(struct net_device *netdev, netdev_features_t features);
159 static int igb_vlan_rx_add_vid(struct net_device *, u16);
160 static int igb_vlan_rx_kill_vid(struct net_device *, u16);
161 static void igb_restore_vlan(struct igb_adapter *);
162 static void igb_rar_set_qsel(struct igb_adapter *, u8 *, u32 , u8);
163 static void igb_ping_all_vfs(struct igb_adapter *);
164 static void igb_msg_task(struct igb_adapter *);
165 static void igb_vmm_control(struct igb_adapter *);
166 static int igb_set_vf_mac(struct igb_adapter *, int, unsigned char *);
167 static void igb_restore_vf_multicasts(struct igb_adapter *adapter);
168 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac);
169 static int igb_ndo_set_vf_vlan(struct net_device *netdev,
170 int vf, u16 vlan, u8 qos);
171 static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf, int tx_rate);
172 static int igb_ndo_get_vf_config(struct net_device *netdev, int vf,
173 struct ifla_vf_info *ivi);
174 static void igb_check_vf_rate_limit(struct igb_adapter *);
176 #ifdef CONFIG_PCI_IOV
177 static int igb_vf_configure(struct igb_adapter *adapter, int vf);
178 static bool igb_vfs_are_assigned(struct igb_adapter *adapter);
182 #ifdef CONFIG_PM_SLEEP
183 static int igb_suspend(struct device *);
185 static int igb_resume(struct device *);
186 #ifdef CONFIG_PM_RUNTIME
187 static int igb_runtime_suspend(struct device *dev);
188 static int igb_runtime_resume(struct device *dev);
189 static int igb_runtime_idle(struct device *dev);
191 static const struct dev_pm_ops igb_pm_ops = {
192 SET_SYSTEM_SLEEP_PM_OPS(igb_suspend, igb_resume)
193 SET_RUNTIME_PM_OPS(igb_runtime_suspend, igb_runtime_resume,
197 static void igb_shutdown(struct pci_dev *);
198 static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs);
199 #ifdef CONFIG_IGB_DCA
200 static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
201 static struct notifier_block dca_notifier = {
202 .notifier_call = igb_notify_dca,
207 #ifdef CONFIG_NET_POLL_CONTROLLER
208 /* for netdump / net console */
209 static void igb_netpoll(struct net_device *);
211 #ifdef CONFIG_PCI_IOV
212 static unsigned int max_vfs = 0;
213 module_param(max_vfs, uint, 0);
214 MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate "
215 "per physical function");
216 #endif /* CONFIG_PCI_IOV */
218 static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
219 pci_channel_state_t);
220 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
221 static void igb_io_resume(struct pci_dev *);
223 static const struct pci_error_handlers igb_err_handler = {
224 .error_detected = igb_io_error_detected,
225 .slot_reset = igb_io_slot_reset,
226 .resume = igb_io_resume,
229 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba);
231 static struct pci_driver igb_driver = {
232 .name = igb_driver_name,
233 .id_table = igb_pci_tbl,
235 .remove = igb_remove,
237 .driver.pm = &igb_pm_ops,
239 .shutdown = igb_shutdown,
240 .sriov_configure = igb_pci_sriov_configure,
241 .err_handler = &igb_err_handler
244 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
245 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
246 MODULE_LICENSE("GPL");
247 MODULE_VERSION(DRV_VERSION);
249 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
250 static int debug = -1;
251 module_param(debug, int, 0);
252 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
254 struct igb_reg_info {
259 static const struct igb_reg_info igb_reg_info_tbl[] = {
261 /* General Registers */
262 {E1000_CTRL, "CTRL"},
263 {E1000_STATUS, "STATUS"},
264 {E1000_CTRL_EXT, "CTRL_EXT"},
266 /* Interrupt Registers */
270 {E1000_RCTL, "RCTL"},
271 {E1000_RDLEN(0), "RDLEN"},
272 {E1000_RDH(0), "RDH"},
273 {E1000_RDT(0), "RDT"},
274 {E1000_RXDCTL(0), "RXDCTL"},
275 {E1000_RDBAL(0), "RDBAL"},
276 {E1000_RDBAH(0), "RDBAH"},
279 {E1000_TCTL, "TCTL"},
280 {E1000_TDBAL(0), "TDBAL"},
281 {E1000_TDBAH(0), "TDBAH"},
282 {E1000_TDLEN(0), "TDLEN"},
283 {E1000_TDH(0), "TDH"},
284 {E1000_TDT(0), "TDT"},
285 {E1000_TXDCTL(0), "TXDCTL"},
286 {E1000_TDFH, "TDFH"},
287 {E1000_TDFT, "TDFT"},
288 {E1000_TDFHS, "TDFHS"},
289 {E1000_TDFPC, "TDFPC"},
291 /* List Terminator */
296 * igb_regdump - register printout routine
298 static void igb_regdump(struct e1000_hw *hw, struct igb_reg_info *reginfo)
304 switch (reginfo->ofs) {
306 for (n = 0; n < 4; n++)
307 regs[n] = rd32(E1000_RDLEN(n));
310 for (n = 0; n < 4; n++)
311 regs[n] = rd32(E1000_RDH(n));
314 for (n = 0; n < 4; n++)
315 regs[n] = rd32(E1000_RDT(n));
317 case E1000_RXDCTL(0):
318 for (n = 0; n < 4; n++)
319 regs[n] = rd32(E1000_RXDCTL(n));
322 for (n = 0; n < 4; n++)
323 regs[n] = rd32(E1000_RDBAL(n));
326 for (n = 0; n < 4; n++)
327 regs[n] = rd32(E1000_RDBAH(n));
330 for (n = 0; n < 4; n++)
331 regs[n] = rd32(E1000_RDBAL(n));
334 for (n = 0; n < 4; n++)
335 regs[n] = rd32(E1000_TDBAH(n));
338 for (n = 0; n < 4; n++)
339 regs[n] = rd32(E1000_TDLEN(n));
342 for (n = 0; n < 4; n++)
343 regs[n] = rd32(E1000_TDH(n));
346 for (n = 0; n < 4; n++)
347 regs[n] = rd32(E1000_TDT(n));
349 case E1000_TXDCTL(0):
350 for (n = 0; n < 4; n++)
351 regs[n] = rd32(E1000_TXDCTL(n));
354 pr_info("%-15s %08x\n", reginfo->name, rd32(reginfo->ofs));
358 snprintf(rname, 16, "%s%s", reginfo->name, "[0-3]");
359 pr_info("%-15s %08x %08x %08x %08x\n", rname, regs[0], regs[1],
364 * igb_dump - Print registers, tx-rings and rx-rings
366 static void igb_dump(struct igb_adapter *adapter)
368 struct net_device *netdev = adapter->netdev;
369 struct e1000_hw *hw = &adapter->hw;
370 struct igb_reg_info *reginfo;
371 struct igb_ring *tx_ring;
372 union e1000_adv_tx_desc *tx_desc;
373 struct my_u0 { u64 a; u64 b; } *u0;
374 struct igb_ring *rx_ring;
375 union e1000_adv_rx_desc *rx_desc;
379 if (!netif_msg_hw(adapter))
382 /* Print netdevice Info */
384 dev_info(&adapter->pdev->dev, "Net device Info\n");
385 pr_info("Device Name state trans_start "
387 pr_info("%-15s %016lX %016lX %016lX\n", netdev->name,
388 netdev->state, netdev->trans_start, netdev->last_rx);
391 /* Print Registers */
392 dev_info(&adapter->pdev->dev, "Register Dump\n");
393 pr_info(" Register Name Value\n");
394 for (reginfo = (struct igb_reg_info *)igb_reg_info_tbl;
395 reginfo->name; reginfo++) {
396 igb_regdump(hw, reginfo);
399 /* Print TX Ring Summary */
400 if (!netdev || !netif_running(netdev))
403 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
404 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
405 for (n = 0; n < adapter->num_tx_queues; n++) {
406 struct igb_tx_buffer *buffer_info;
407 tx_ring = adapter->tx_ring[n];
408 buffer_info = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
409 pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
410 n, tx_ring->next_to_use, tx_ring->next_to_clean,
411 (u64)dma_unmap_addr(buffer_info, dma),
412 dma_unmap_len(buffer_info, len),
413 buffer_info->next_to_watch,
414 (u64)buffer_info->time_stamp);
418 if (!netif_msg_tx_done(adapter))
419 goto rx_ring_summary;
421 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
423 /* Transmit Descriptor Formats
425 * Advanced Transmit Descriptor
426 * +--------------------------------------------------------------+
427 * 0 | Buffer Address [63:0] |
428 * +--------------------------------------------------------------+
429 * 8 | PAYLEN | PORTS |CC|IDX | STA | DCMD |DTYP|MAC|RSV| DTALEN |
430 * +--------------------------------------------------------------+
431 * 63 46 45 40 39 38 36 35 32 31 24 15 0
434 for (n = 0; n < adapter->num_tx_queues; n++) {
435 tx_ring = adapter->tx_ring[n];
436 pr_info("------------------------------------\n");
437 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
438 pr_info("------------------------------------\n");
439 pr_info("T [desc] [address 63:0 ] [PlPOCIStDDM Ln] "
440 "[bi->dma ] leng ntw timestamp "
443 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
444 const char *next_desc;
445 struct igb_tx_buffer *buffer_info;
446 tx_desc = IGB_TX_DESC(tx_ring, i);
447 buffer_info = &tx_ring->tx_buffer_info[i];
448 u0 = (struct my_u0 *)tx_desc;
449 if (i == tx_ring->next_to_use &&
450 i == tx_ring->next_to_clean)
451 next_desc = " NTC/U";
452 else if (i == tx_ring->next_to_use)
454 else if (i == tx_ring->next_to_clean)
459 pr_info("T [0x%03X] %016llX %016llX %016llX"
460 " %04X %p %016llX %p%s\n", i,
463 (u64)dma_unmap_addr(buffer_info, dma),
464 dma_unmap_len(buffer_info, len),
465 buffer_info->next_to_watch,
466 (u64)buffer_info->time_stamp,
467 buffer_info->skb, next_desc);
469 if (netif_msg_pktdata(adapter) && buffer_info->skb)
470 print_hex_dump(KERN_INFO, "",
472 16, 1, buffer_info->skb->data,
473 dma_unmap_len(buffer_info, len),
478 /* Print RX Rings Summary */
480 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
481 pr_info("Queue [NTU] [NTC]\n");
482 for (n = 0; n < adapter->num_rx_queues; n++) {
483 rx_ring = adapter->rx_ring[n];
484 pr_info(" %5d %5X %5X\n",
485 n, rx_ring->next_to_use, rx_ring->next_to_clean);
489 if (!netif_msg_rx_status(adapter))
492 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
494 /* Advanced Receive Descriptor (Read) Format
496 * +-----------------------------------------------------+
497 * 0 | Packet Buffer Address [63:1] |A0/NSE|
498 * +----------------------------------------------+------+
499 * 8 | Header Buffer Address [63:1] | DD |
500 * +-----------------------------------------------------+
503 * Advanced Receive Descriptor (Write-Back) Format
505 * 63 48 47 32 31 30 21 20 17 16 4 3 0
506 * +------------------------------------------------------+
507 * 0 | Packet IP |SPH| HDR_LEN | RSV|Packet| RSS |
508 * | Checksum Ident | | | | Type | Type |
509 * +------------------------------------------------------+
510 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
511 * +------------------------------------------------------+
512 * 63 48 47 32 31 20 19 0
515 for (n = 0; n < adapter->num_rx_queues; n++) {
516 rx_ring = adapter->rx_ring[n];
517 pr_info("------------------------------------\n");
518 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
519 pr_info("------------------------------------\n");
520 pr_info("R [desc] [ PktBuf A0] [ HeadBuf DD] "
521 "[bi->dma ] [bi->skb] <-- Adv Rx Read format\n");
522 pr_info("RWB[desc] [PcsmIpSHl PtRs] [vl er S cks ln] -----"
523 "----------- [bi->skb] <-- Adv Rx Write-Back format\n");
525 for (i = 0; i < rx_ring->count; i++) {
526 const char *next_desc;
527 struct igb_rx_buffer *buffer_info;
528 buffer_info = &rx_ring->rx_buffer_info[i];
529 rx_desc = IGB_RX_DESC(rx_ring, i);
530 u0 = (struct my_u0 *)rx_desc;
531 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
533 if (i == rx_ring->next_to_use)
535 else if (i == rx_ring->next_to_clean)
540 if (staterr & E1000_RXD_STAT_DD) {
541 /* Descriptor Done */
542 pr_info("%s[0x%03X] %016llX %016llX ---------------- %s\n",
548 pr_info("%s[0x%03X] %016llX %016llX %016llX %s\n",
552 (u64)buffer_info->dma,
555 if (netif_msg_pktdata(adapter) &&
556 buffer_info->dma && buffer_info->page) {
557 print_hex_dump(KERN_INFO, "",
560 page_address(buffer_info->page) +
561 buffer_info->page_offset,
572 /* igb_get_i2c_data - Reads the I2C SDA data bit
573 * @hw: pointer to hardware structure
574 * @i2cctl: Current value of I2CCTL register
576 * Returns the I2C data bit value
578 static int igb_get_i2c_data(void *data)
580 struct igb_adapter *adapter = (struct igb_adapter *)data;
581 struct e1000_hw *hw = &adapter->hw;
582 s32 i2cctl = rd32(E1000_I2CPARAMS);
584 return ((i2cctl & E1000_I2C_DATA_IN) != 0);
587 /* igb_set_i2c_data - Sets the I2C data bit
588 * @data: pointer to hardware structure
589 * @state: I2C data value (0 or 1) to set
591 * Sets the I2C data bit
593 static void igb_set_i2c_data(void *data, int state)
595 struct igb_adapter *adapter = (struct igb_adapter *)data;
596 struct e1000_hw *hw = &adapter->hw;
597 s32 i2cctl = rd32(E1000_I2CPARAMS);
600 i2cctl |= E1000_I2C_DATA_OUT;
602 i2cctl &= ~E1000_I2C_DATA_OUT;
604 i2cctl &= ~E1000_I2C_DATA_OE_N;
605 i2cctl |= E1000_I2C_CLK_OE_N;
606 wr32(E1000_I2CPARAMS, i2cctl);
611 /* igb_set_i2c_clk - Sets the I2C SCL clock
612 * @data: pointer to hardware structure
613 * @state: state to set clock
615 * Sets the I2C clock line to state
617 static void igb_set_i2c_clk(void *data, int state)
619 struct igb_adapter *adapter = (struct igb_adapter *)data;
620 struct e1000_hw *hw = &adapter->hw;
621 s32 i2cctl = rd32(E1000_I2CPARAMS);
624 i2cctl |= E1000_I2C_CLK_OUT;
625 i2cctl &= ~E1000_I2C_CLK_OE_N;
627 i2cctl &= ~E1000_I2C_CLK_OUT;
628 i2cctl &= ~E1000_I2C_CLK_OE_N;
630 wr32(E1000_I2CPARAMS, i2cctl);
634 /* igb_get_i2c_clk - Gets the I2C SCL clock state
635 * @data: pointer to hardware structure
637 * Gets the I2C clock state
639 static int igb_get_i2c_clk(void *data)
641 struct igb_adapter *adapter = (struct igb_adapter *)data;
642 struct e1000_hw *hw = &adapter->hw;
643 s32 i2cctl = rd32(E1000_I2CPARAMS);
645 return ((i2cctl & E1000_I2C_CLK_IN) != 0);
648 static const struct i2c_algo_bit_data igb_i2c_algo = {
649 .setsda = igb_set_i2c_data,
650 .setscl = igb_set_i2c_clk,
651 .getsda = igb_get_i2c_data,
652 .getscl = igb_get_i2c_clk,
658 * igb_get_hw_dev - return device
659 * used by hardware layer to print debugging information
661 struct net_device *igb_get_hw_dev(struct e1000_hw *hw)
663 struct igb_adapter *adapter = hw->back;
664 return adapter->netdev;
668 * igb_init_module - Driver Registration Routine
670 * igb_init_module is the first routine called when the driver is
671 * loaded. All it does is register with the PCI subsystem.
673 static int __init igb_init_module(void)
676 pr_info("%s - version %s\n",
677 igb_driver_string, igb_driver_version);
679 pr_info("%s\n", igb_copyright);
681 #ifdef CONFIG_IGB_DCA
682 dca_register_notify(&dca_notifier);
684 ret = pci_register_driver(&igb_driver);
688 module_init(igb_init_module);
691 * igb_exit_module - Driver Exit Cleanup Routine
693 * igb_exit_module is called just before the driver is removed
696 static void __exit igb_exit_module(void)
698 #ifdef CONFIG_IGB_DCA
699 dca_unregister_notify(&dca_notifier);
701 pci_unregister_driver(&igb_driver);
704 module_exit(igb_exit_module);
706 #define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
708 * igb_cache_ring_register - Descriptor ring to register mapping
709 * @adapter: board private structure to initialize
711 * Once we know the feature-set enabled for the device, we'll cache
712 * the register offset the descriptor ring is assigned to.
714 static void igb_cache_ring_register(struct igb_adapter *adapter)
717 u32 rbase_offset = adapter->vfs_allocated_count;
719 switch (adapter->hw.mac.type) {
721 /* The queues are allocated for virtualization such that VF 0
722 * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
723 * In order to avoid collision we start at the first free queue
724 * and continue consuming queues in the same sequence
726 if (adapter->vfs_allocated_count) {
727 for (; i < adapter->rss_queues; i++)
728 adapter->rx_ring[i]->reg_idx = rbase_offset +
737 for (; i < adapter->num_rx_queues; i++)
738 adapter->rx_ring[i]->reg_idx = rbase_offset + i;
739 for (; j < adapter->num_tx_queues; j++)
740 adapter->tx_ring[j]->reg_idx = rbase_offset + j;
746 * igb_write_ivar - configure ivar for given MSI-X vector
747 * @hw: pointer to the HW structure
748 * @msix_vector: vector number we are allocating to a given ring
749 * @index: row index of IVAR register to write within IVAR table
750 * @offset: column offset of in IVAR, should be multiple of 8
752 * This function is intended to handle the writing of the IVAR register
753 * for adapters 82576 and newer. The IVAR table consists of 2 columns,
754 * each containing an cause allocation for an Rx and Tx ring, and a
755 * variable number of rows depending on the number of queues supported.
757 static void igb_write_ivar(struct e1000_hw *hw, int msix_vector,
758 int index, int offset)
760 u32 ivar = array_rd32(E1000_IVAR0, index);
762 /* clear any bits that are currently set */
763 ivar &= ~((u32)0xFF << offset);
765 /* write vector and valid bit */
766 ivar |= (msix_vector | E1000_IVAR_VALID) << offset;
768 array_wr32(E1000_IVAR0, index, ivar);
771 #define IGB_N0_QUEUE -1
772 static void igb_assign_vector(struct igb_q_vector *q_vector, int msix_vector)
774 struct igb_adapter *adapter = q_vector->adapter;
775 struct e1000_hw *hw = &adapter->hw;
776 int rx_queue = IGB_N0_QUEUE;
777 int tx_queue = IGB_N0_QUEUE;
780 if (q_vector->rx.ring)
781 rx_queue = q_vector->rx.ring->reg_idx;
782 if (q_vector->tx.ring)
783 tx_queue = q_vector->tx.ring->reg_idx;
785 switch (hw->mac.type) {
787 /* The 82575 assigns vectors using a bitmask, which matches the
788 bitmask for the EICR/EIMS/EIMC registers. To assign one
789 or more queues to a vector, we write the appropriate bits
790 into the MSIXBM register for that vector. */
791 if (rx_queue > IGB_N0_QUEUE)
792 msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
793 if (tx_queue > IGB_N0_QUEUE)
794 msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
795 if (!adapter->msix_entries && msix_vector == 0)
796 msixbm |= E1000_EIMS_OTHER;
797 array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
798 q_vector->eims_value = msixbm;
802 * 82576 uses a table that essentially consists of 2 columns
803 * with 8 rows. The ordering is column-major so we use the
804 * lower 3 bits as the row index, and the 4th bit as the
807 if (rx_queue > IGB_N0_QUEUE)
808 igb_write_ivar(hw, msix_vector,
810 (rx_queue & 0x8) << 1);
811 if (tx_queue > IGB_N0_QUEUE)
812 igb_write_ivar(hw, msix_vector,
814 ((tx_queue & 0x8) << 1) + 8);
815 q_vector->eims_value = 1 << msix_vector;
822 * On 82580 and newer adapters the scheme is similar to 82576
823 * however instead of ordering column-major we have things
824 * ordered row-major. So we traverse the table by using
825 * bit 0 as the column offset, and the remaining bits as the
828 if (rx_queue > IGB_N0_QUEUE)
829 igb_write_ivar(hw, msix_vector,
831 (rx_queue & 0x1) << 4);
832 if (tx_queue > IGB_N0_QUEUE)
833 igb_write_ivar(hw, msix_vector,
835 ((tx_queue & 0x1) << 4) + 8);
836 q_vector->eims_value = 1 << msix_vector;
843 /* add q_vector eims value to global eims_enable_mask */
844 adapter->eims_enable_mask |= q_vector->eims_value;
846 /* configure q_vector to set itr on first interrupt */
847 q_vector->set_itr = 1;
851 * igb_configure_msix - Configure MSI-X hardware
853 * igb_configure_msix sets up the hardware to properly
854 * generate MSI-X interrupts.
856 static void igb_configure_msix(struct igb_adapter *adapter)
860 struct e1000_hw *hw = &adapter->hw;
862 adapter->eims_enable_mask = 0;
864 /* set vector for other causes, i.e. link changes */
865 switch (hw->mac.type) {
867 tmp = rd32(E1000_CTRL_EXT);
868 /* enable MSI-X PBA support*/
869 tmp |= E1000_CTRL_EXT_PBA_CLR;
871 /* Auto-Mask interrupts upon ICR read. */
872 tmp |= E1000_CTRL_EXT_EIAME;
873 tmp |= E1000_CTRL_EXT_IRCA;
875 wr32(E1000_CTRL_EXT, tmp);
877 /* enable msix_other interrupt */
878 array_wr32(E1000_MSIXBM(0), vector++,
880 adapter->eims_other = E1000_EIMS_OTHER;
889 /* Turn on MSI-X capability first, or our settings
890 * won't stick. And it will take days to debug. */
891 wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
892 E1000_GPIE_PBA | E1000_GPIE_EIAME |
895 /* enable msix_other interrupt */
896 adapter->eims_other = 1 << vector;
897 tmp = (vector++ | E1000_IVAR_VALID) << 8;
899 wr32(E1000_IVAR_MISC, tmp);
902 /* do nothing, since nothing else supports MSI-X */
904 } /* switch (hw->mac.type) */
906 adapter->eims_enable_mask |= adapter->eims_other;
908 for (i = 0; i < adapter->num_q_vectors; i++)
909 igb_assign_vector(adapter->q_vector[i], vector++);
915 * igb_request_msix - Initialize MSI-X interrupts
917 * igb_request_msix allocates MSI-X vectors and requests interrupts from the
920 static int igb_request_msix(struct igb_adapter *adapter)
922 struct net_device *netdev = adapter->netdev;
923 struct e1000_hw *hw = &adapter->hw;
924 int i, err = 0, vector = 0, free_vector = 0;
926 err = request_irq(adapter->msix_entries[vector].vector,
927 igb_msix_other, 0, netdev->name, adapter);
931 for (i = 0; i < adapter->num_q_vectors; i++) {
932 struct igb_q_vector *q_vector = adapter->q_vector[i];
936 q_vector->itr_register = hw->hw_addr + E1000_EITR(vector);
938 if (q_vector->rx.ring && q_vector->tx.ring)
939 sprintf(q_vector->name, "%s-TxRx-%u", netdev->name,
940 q_vector->rx.ring->queue_index);
941 else if (q_vector->tx.ring)
942 sprintf(q_vector->name, "%s-tx-%u", netdev->name,
943 q_vector->tx.ring->queue_index);
944 else if (q_vector->rx.ring)
945 sprintf(q_vector->name, "%s-rx-%u", netdev->name,
946 q_vector->rx.ring->queue_index);
948 sprintf(q_vector->name, "%s-unused", netdev->name);
950 err = request_irq(adapter->msix_entries[vector].vector,
951 igb_msix_ring, 0, q_vector->name,
957 igb_configure_msix(adapter);
961 /* free already assigned IRQs */
962 free_irq(adapter->msix_entries[free_vector++].vector, adapter);
965 for (i = 0; i < vector; i++) {
966 free_irq(adapter->msix_entries[free_vector++].vector,
967 adapter->q_vector[i]);
973 static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
975 if (adapter->msix_entries) {
976 pci_disable_msix(adapter->pdev);
977 kfree(adapter->msix_entries);
978 adapter->msix_entries = NULL;
979 } else if (adapter->flags & IGB_FLAG_HAS_MSI) {
980 pci_disable_msi(adapter->pdev);
985 * igb_free_q_vector - Free memory allocated for specific interrupt vector
986 * @adapter: board private structure to initialize
987 * @v_idx: Index of vector to be freed
989 * This function frees the memory allocated to the q_vector. In addition if
990 * NAPI is enabled it will delete any references to the NAPI struct prior
991 * to freeing 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 if (q_vector->tx.ring)
998 adapter->tx_ring[q_vector->tx.ring->queue_index] = NULL;
1000 if (q_vector->rx.ring)
1001 adapter->tx_ring[q_vector->rx.ring->queue_index] = NULL;
1003 adapter->q_vector[v_idx] = NULL;
1004 netif_napi_del(&q_vector->napi);
1007 * ixgbe_get_stats64() might access the rings on this vector,
1008 * we must wait a grace period before freeing it.
1010 kfree_rcu(q_vector, rcu);
1014 * igb_free_q_vectors - Free memory allocated for interrupt vectors
1015 * @adapter: board private structure to initialize
1017 * This function frees the memory allocated to the q_vectors. In addition if
1018 * NAPI is enabled it will delete any references to the NAPI struct prior
1019 * to freeing the q_vector.
1021 static void igb_free_q_vectors(struct igb_adapter *adapter)
1023 int v_idx = adapter->num_q_vectors;
1025 adapter->num_tx_queues = 0;
1026 adapter->num_rx_queues = 0;
1027 adapter->num_q_vectors = 0;
1030 igb_free_q_vector(adapter, v_idx);
1034 * igb_clear_interrupt_scheme - reset the device to a state of no interrupts
1036 * This function resets the device so that it has 0 rx queues, tx queues, and
1037 * MSI-X interrupts allocated.
1039 static void igb_clear_interrupt_scheme(struct igb_adapter *adapter)
1041 igb_free_q_vectors(adapter);
1042 igb_reset_interrupt_capability(adapter);
1046 * igb_set_interrupt_capability - set MSI or MSI-X if supported
1048 * Attempt to configure interrupts using the best available
1049 * capabilities of the hardware and kernel.
1051 static void igb_set_interrupt_capability(struct igb_adapter *adapter, bool msix)
1059 /* Number of supported queues. */
1060 adapter->num_rx_queues = adapter->rss_queues;
1061 if (adapter->vfs_allocated_count)
1062 adapter->num_tx_queues = 1;
1064 adapter->num_tx_queues = adapter->rss_queues;
1066 /* start with one vector for every rx queue */
1067 numvecs = adapter->num_rx_queues;
1069 /* if tx handler is separate add 1 for every tx queue */
1070 if (!(adapter->flags & IGB_FLAG_QUEUE_PAIRS))
1071 numvecs += adapter->num_tx_queues;
1073 /* store the number of vectors reserved for queues */
1074 adapter->num_q_vectors = numvecs;
1076 /* add 1 vector for link status interrupts */
1078 adapter->msix_entries = kcalloc(numvecs, sizeof(struct msix_entry),
1081 if (!adapter->msix_entries)
1084 for (i = 0; i < numvecs; i++)
1085 adapter->msix_entries[i].entry = i;
1087 err = pci_enable_msix(adapter->pdev,
1088 adapter->msix_entries,
1093 igb_reset_interrupt_capability(adapter);
1095 /* If we can't do MSI-X, try MSI */
1097 #ifdef CONFIG_PCI_IOV
1098 /* disable SR-IOV for non MSI-X configurations */
1099 if (adapter->vf_data) {
1100 struct e1000_hw *hw = &adapter->hw;
1101 /* disable iov and allow time for transactions to clear */
1102 pci_disable_sriov(adapter->pdev);
1105 kfree(adapter->vf_data);
1106 adapter->vf_data = NULL;
1107 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
1110 dev_info(&adapter->pdev->dev, "IOV Disabled\n");
1113 adapter->vfs_allocated_count = 0;
1114 adapter->rss_queues = 1;
1115 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
1116 adapter->num_rx_queues = 1;
1117 adapter->num_tx_queues = 1;
1118 adapter->num_q_vectors = 1;
1119 if (!pci_enable_msi(adapter->pdev))
1120 adapter->flags |= IGB_FLAG_HAS_MSI;
1123 static void igb_add_ring(struct igb_ring *ring,
1124 struct igb_ring_container *head)
1131 * igb_alloc_q_vector - Allocate memory for a single interrupt vector
1132 * @adapter: board private structure to initialize
1133 * @v_count: q_vectors allocated on adapter, used for ring interleaving
1134 * @v_idx: index of vector in adapter struct
1135 * @txr_count: total number of Tx rings to allocate
1136 * @txr_idx: index of first Tx ring to allocate
1137 * @rxr_count: total number of Rx rings to allocate
1138 * @rxr_idx: index of first Rx ring to allocate
1140 * We allocate one q_vector. If allocation fails we return -ENOMEM.
1142 static int igb_alloc_q_vector(struct igb_adapter *adapter,
1143 int v_count, int v_idx,
1144 int txr_count, int txr_idx,
1145 int rxr_count, int rxr_idx)
1147 struct igb_q_vector *q_vector;
1148 struct igb_ring *ring;
1149 int ring_count, size;
1151 /* igb only supports 1 Tx and/or 1 Rx queue per vector */
1152 if (txr_count > 1 || rxr_count > 1)
1155 ring_count = txr_count + rxr_count;
1156 size = sizeof(struct igb_q_vector) +
1157 (sizeof(struct igb_ring) * ring_count);
1159 /* allocate q_vector and rings */
1160 q_vector = kzalloc(size, GFP_KERNEL);
1164 /* initialize NAPI */
1165 netif_napi_add(adapter->netdev, &q_vector->napi,
1168 /* tie q_vector and adapter together */
1169 adapter->q_vector[v_idx] = q_vector;
1170 q_vector->adapter = adapter;
1172 /* initialize work limits */
1173 q_vector->tx.work_limit = adapter->tx_work_limit;
1175 /* initialize ITR configuration */
1176 q_vector->itr_register = adapter->hw.hw_addr + E1000_EITR(0);
1177 q_vector->itr_val = IGB_START_ITR;
1179 /* initialize pointer to rings */
1180 ring = q_vector->ring;
1183 /* assign generic ring traits */
1184 ring->dev = &adapter->pdev->dev;
1185 ring->netdev = adapter->netdev;
1187 /* configure backlink on ring */
1188 ring->q_vector = q_vector;
1190 /* update q_vector Tx values */
1191 igb_add_ring(ring, &q_vector->tx);
1193 /* For 82575, context index must be unique per ring. */
1194 if (adapter->hw.mac.type == e1000_82575)
1195 set_bit(IGB_RING_FLAG_TX_CTX_IDX, &ring->flags);
1197 /* apply Tx specific ring traits */
1198 ring->count = adapter->tx_ring_count;
1199 ring->queue_index = txr_idx;
1201 /* assign ring to adapter */
1202 adapter->tx_ring[txr_idx] = ring;
1204 /* push pointer to next ring */
1209 /* assign generic ring traits */
1210 ring->dev = &adapter->pdev->dev;
1211 ring->netdev = adapter->netdev;
1213 /* configure backlink on ring */
1214 ring->q_vector = q_vector;
1216 /* update q_vector Rx values */
1217 igb_add_ring(ring, &q_vector->rx);
1219 /* set flag indicating ring supports SCTP checksum offload */
1220 if (adapter->hw.mac.type >= e1000_82576)
1221 set_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags);
1224 * On i350, i210, and i211, loopback VLAN packets
1225 * have the tag byte-swapped.
1227 if (adapter->hw.mac.type >= e1000_i350)
1228 set_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &ring->flags);
1230 /* apply Rx specific ring traits */
1231 ring->count = adapter->rx_ring_count;
1232 ring->queue_index = rxr_idx;
1234 /* assign ring to adapter */
1235 adapter->rx_ring[rxr_idx] = ring;
1243 * igb_alloc_q_vectors - Allocate memory for interrupt vectors
1244 * @adapter: board private structure to initialize
1246 * We allocate one q_vector per queue interrupt. If allocation fails we
1249 static int igb_alloc_q_vectors(struct igb_adapter *adapter)
1251 int q_vectors = adapter->num_q_vectors;
1252 int rxr_remaining = adapter->num_rx_queues;
1253 int txr_remaining = adapter->num_tx_queues;
1254 int rxr_idx = 0, txr_idx = 0, v_idx = 0;
1257 if (q_vectors >= (rxr_remaining + txr_remaining)) {
1258 for (; rxr_remaining; v_idx++) {
1259 err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1265 /* update counts and index */
1271 for (; v_idx < q_vectors; v_idx++) {
1272 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_idx);
1273 int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_idx);
1274 err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1275 tqpv, txr_idx, rqpv, rxr_idx);
1280 /* update counts and index */
1281 rxr_remaining -= rqpv;
1282 txr_remaining -= tqpv;
1290 adapter->num_tx_queues = 0;
1291 adapter->num_rx_queues = 0;
1292 adapter->num_q_vectors = 0;
1295 igb_free_q_vector(adapter, v_idx);
1301 * igb_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
1303 * This function initializes the interrupts and allocates all of the queues.
1305 static int igb_init_interrupt_scheme(struct igb_adapter *adapter, bool msix)
1307 struct pci_dev *pdev = adapter->pdev;
1310 igb_set_interrupt_capability(adapter, msix);
1312 err = igb_alloc_q_vectors(adapter);
1314 dev_err(&pdev->dev, "Unable to allocate memory for vectors\n");
1315 goto err_alloc_q_vectors;
1318 igb_cache_ring_register(adapter);
1322 err_alloc_q_vectors:
1323 igb_reset_interrupt_capability(adapter);
1328 * igb_request_irq - initialize interrupts
1330 * Attempts to configure interrupts using the best available
1331 * capabilities of the hardware and kernel.
1333 static int igb_request_irq(struct igb_adapter *adapter)
1335 struct net_device *netdev = adapter->netdev;
1336 struct pci_dev *pdev = adapter->pdev;
1339 if (adapter->msix_entries) {
1340 err = igb_request_msix(adapter);
1343 /* fall back to MSI */
1344 igb_free_all_tx_resources(adapter);
1345 igb_free_all_rx_resources(adapter);
1347 igb_clear_interrupt_scheme(adapter);
1348 err = igb_init_interrupt_scheme(adapter, false);
1352 igb_setup_all_tx_resources(adapter);
1353 igb_setup_all_rx_resources(adapter);
1354 igb_configure(adapter);
1357 igb_assign_vector(adapter->q_vector[0], 0);
1359 if (adapter->flags & IGB_FLAG_HAS_MSI) {
1360 err = request_irq(pdev->irq, igb_intr_msi, 0,
1361 netdev->name, adapter);
1365 /* fall back to legacy interrupts */
1366 igb_reset_interrupt_capability(adapter);
1367 adapter->flags &= ~IGB_FLAG_HAS_MSI;
1370 err = request_irq(pdev->irq, igb_intr, IRQF_SHARED,
1371 netdev->name, adapter);
1374 dev_err(&pdev->dev, "Error %d getting interrupt\n",
1381 static void igb_free_irq(struct igb_adapter *adapter)
1383 if (adapter->msix_entries) {
1386 free_irq(adapter->msix_entries[vector++].vector, adapter);
1388 for (i = 0; i < adapter->num_q_vectors; i++)
1389 free_irq(adapter->msix_entries[vector++].vector,
1390 adapter->q_vector[i]);
1392 free_irq(adapter->pdev->irq, adapter);
1397 * igb_irq_disable - Mask off interrupt generation on the NIC
1398 * @adapter: board private structure
1400 static void igb_irq_disable(struct igb_adapter *adapter)
1402 struct e1000_hw *hw = &adapter->hw;
1405 * we need to be careful when disabling interrupts. The VFs are also
1406 * mapped into these registers and so clearing the bits can cause
1407 * issues on the VF drivers so we only need to clear what we set
1409 if (adapter->msix_entries) {
1410 u32 regval = rd32(E1000_EIAM);
1411 wr32(E1000_EIAM, regval & ~adapter->eims_enable_mask);
1412 wr32(E1000_EIMC, adapter->eims_enable_mask);
1413 regval = rd32(E1000_EIAC);
1414 wr32(E1000_EIAC, regval & ~adapter->eims_enable_mask);
1418 wr32(E1000_IMC, ~0);
1420 if (adapter->msix_entries) {
1422 for (i = 0; i < adapter->num_q_vectors; i++)
1423 synchronize_irq(adapter->msix_entries[i].vector);
1425 synchronize_irq(adapter->pdev->irq);
1430 * igb_irq_enable - Enable default interrupt generation settings
1431 * @adapter: board private structure
1433 static void igb_irq_enable(struct igb_adapter *adapter)
1435 struct e1000_hw *hw = &adapter->hw;
1437 if (adapter->msix_entries) {
1438 u32 ims = E1000_IMS_LSC | E1000_IMS_DOUTSYNC | E1000_IMS_DRSTA;
1439 u32 regval = rd32(E1000_EIAC);
1440 wr32(E1000_EIAC, regval | adapter->eims_enable_mask);
1441 regval = rd32(E1000_EIAM);
1442 wr32(E1000_EIAM, regval | adapter->eims_enable_mask);
1443 wr32(E1000_EIMS, adapter->eims_enable_mask);
1444 if (adapter->vfs_allocated_count) {
1445 wr32(E1000_MBVFIMR, 0xFF);
1446 ims |= E1000_IMS_VMMB;
1448 wr32(E1000_IMS, ims);
1450 wr32(E1000_IMS, IMS_ENABLE_MASK |
1452 wr32(E1000_IAM, IMS_ENABLE_MASK |
1457 static void igb_update_mng_vlan(struct igb_adapter *adapter)
1459 struct e1000_hw *hw = &adapter->hw;
1460 u16 vid = adapter->hw.mng_cookie.vlan_id;
1461 u16 old_vid = adapter->mng_vlan_id;
1463 if (hw->mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1464 /* add VID to filter table */
1465 igb_vfta_set(hw, vid, true);
1466 adapter->mng_vlan_id = vid;
1468 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1471 if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
1473 !test_bit(old_vid, adapter->active_vlans)) {
1474 /* remove VID from filter table */
1475 igb_vfta_set(hw, old_vid, false);
1480 * igb_release_hw_control - release control of the h/w to f/w
1481 * @adapter: address of board private structure
1483 * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
1484 * For ASF and Pass Through versions of f/w this means that the
1485 * driver is no longer loaded.
1488 static void igb_release_hw_control(struct igb_adapter *adapter)
1490 struct e1000_hw *hw = &adapter->hw;
1493 /* Let firmware take over control of h/w */
1494 ctrl_ext = rd32(E1000_CTRL_EXT);
1495 wr32(E1000_CTRL_EXT,
1496 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1500 * igb_get_hw_control - get control of the h/w from f/w
1501 * @adapter: address of board private structure
1503 * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
1504 * For ASF and Pass Through versions of f/w this means that
1505 * the driver is loaded.
1508 static void igb_get_hw_control(struct igb_adapter *adapter)
1510 struct e1000_hw *hw = &adapter->hw;
1513 /* Let firmware know the driver has taken over */
1514 ctrl_ext = rd32(E1000_CTRL_EXT);
1515 wr32(E1000_CTRL_EXT,
1516 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1520 * igb_configure - configure the hardware for RX and TX
1521 * @adapter: private board structure
1523 static void igb_configure(struct igb_adapter *adapter)
1525 struct net_device *netdev = adapter->netdev;
1528 igb_get_hw_control(adapter);
1529 igb_set_rx_mode(netdev);
1531 igb_restore_vlan(adapter);
1533 igb_setup_tctl(adapter);
1534 igb_setup_mrqc(adapter);
1535 igb_setup_rctl(adapter);
1537 igb_configure_tx(adapter);
1538 igb_configure_rx(adapter);
1540 igb_rx_fifo_flush_82575(&adapter->hw);
1542 /* call igb_desc_unused which always leaves
1543 * at least 1 descriptor unused to make sure
1544 * next_to_use != next_to_clean */
1545 for (i = 0; i < adapter->num_rx_queues; i++) {
1546 struct igb_ring *ring = adapter->rx_ring[i];
1547 igb_alloc_rx_buffers(ring, igb_desc_unused(ring));
1552 * igb_power_up_link - Power up the phy/serdes link
1553 * @adapter: address of board private structure
1555 void igb_power_up_link(struct igb_adapter *adapter)
1557 igb_reset_phy(&adapter->hw);
1559 if (adapter->hw.phy.media_type == e1000_media_type_copper)
1560 igb_power_up_phy_copper(&adapter->hw);
1562 igb_power_up_serdes_link_82575(&adapter->hw);
1566 * igb_power_down_link - Power down the phy/serdes link
1567 * @adapter: address of board private structure
1569 static void igb_power_down_link(struct igb_adapter *adapter)
1571 if (adapter->hw.phy.media_type == e1000_media_type_copper)
1572 igb_power_down_phy_copper_82575(&adapter->hw);
1574 igb_shutdown_serdes_link_82575(&adapter->hw);
1578 * igb_up - Open the interface and prepare it to handle traffic
1579 * @adapter: board private structure
1581 int igb_up(struct igb_adapter *adapter)
1583 struct e1000_hw *hw = &adapter->hw;
1586 /* hardware has been reset, we need to reload some things */
1587 igb_configure(adapter);
1589 clear_bit(__IGB_DOWN, &adapter->state);
1591 for (i = 0; i < adapter->num_q_vectors; i++)
1592 napi_enable(&(adapter->q_vector[i]->napi));
1594 if (adapter->msix_entries)
1595 igb_configure_msix(adapter);
1597 igb_assign_vector(adapter->q_vector[0], 0);
1599 /* Clear any pending interrupts. */
1601 igb_irq_enable(adapter);
1603 /* notify VFs that reset has been completed */
1604 if (adapter->vfs_allocated_count) {
1605 u32 reg_data = rd32(E1000_CTRL_EXT);
1606 reg_data |= E1000_CTRL_EXT_PFRSTD;
1607 wr32(E1000_CTRL_EXT, reg_data);
1610 netif_tx_start_all_queues(adapter->netdev);
1612 /* start the watchdog. */
1613 hw->mac.get_link_status = 1;
1614 schedule_work(&adapter->watchdog_task);
1619 void igb_down(struct igb_adapter *adapter)
1621 struct net_device *netdev = adapter->netdev;
1622 struct e1000_hw *hw = &adapter->hw;
1626 /* signal that we're down so the interrupt handler does not
1627 * reschedule our watchdog timer */
1628 set_bit(__IGB_DOWN, &adapter->state);
1630 /* disable receives in the hardware */
1631 rctl = rd32(E1000_RCTL);
1632 wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
1633 /* flush and sleep below */
1635 netif_tx_stop_all_queues(netdev);
1637 /* disable transmits in the hardware */
1638 tctl = rd32(E1000_TCTL);
1639 tctl &= ~E1000_TCTL_EN;
1640 wr32(E1000_TCTL, tctl);
1641 /* flush both disables and wait for them to finish */
1645 for (i = 0; i < adapter->num_q_vectors; i++)
1646 napi_disable(&(adapter->q_vector[i]->napi));
1648 igb_irq_disable(adapter);
1650 del_timer_sync(&adapter->watchdog_timer);
1651 del_timer_sync(&adapter->phy_info_timer);
1653 netif_carrier_off(netdev);
1655 /* record the stats before reset*/
1656 spin_lock(&adapter->stats64_lock);
1657 igb_update_stats(adapter, &adapter->stats64);
1658 spin_unlock(&adapter->stats64_lock);
1660 adapter->link_speed = 0;
1661 adapter->link_duplex = 0;
1663 if (!pci_channel_offline(adapter->pdev))
1665 igb_clean_all_tx_rings(adapter);
1666 igb_clean_all_rx_rings(adapter);
1667 #ifdef CONFIG_IGB_DCA
1669 /* since we reset the hardware DCA settings were cleared */
1670 igb_setup_dca(adapter);
1674 void igb_reinit_locked(struct igb_adapter *adapter)
1676 WARN_ON(in_interrupt());
1677 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
1681 clear_bit(__IGB_RESETTING, &adapter->state);
1684 void igb_reset(struct igb_adapter *adapter)
1686 struct pci_dev *pdev = adapter->pdev;
1687 struct e1000_hw *hw = &adapter->hw;
1688 struct e1000_mac_info *mac = &hw->mac;
1689 struct e1000_fc_info *fc = &hw->fc;
1690 u32 pba = 0, tx_space, min_tx_space, min_rx_space, hwm;
1692 /* Repartition Pba for greater than 9k mtu
1693 * To take effect CTRL.RST is required.
1695 switch (mac->type) {
1698 pba = rd32(E1000_RXPBS);
1699 pba = igb_rxpbs_adjust_82580(pba);
1702 pba = rd32(E1000_RXPBS);
1703 pba &= E1000_RXPBS_SIZE_MASK_82576;
1709 pba = E1000_PBA_34K;
1713 if ((adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) &&
1714 (mac->type < e1000_82576)) {
1715 /* adjust PBA for jumbo frames */
1716 wr32(E1000_PBA, pba);
1718 /* To maintain wire speed transmits, the Tx FIFO should be
1719 * large enough to accommodate two full transmit packets,
1720 * rounded up to the next 1KB and expressed in KB. Likewise,
1721 * the Rx FIFO should be large enough to accommodate at least
1722 * one full receive packet and is similarly rounded up and
1723 * expressed in KB. */
1724 pba = rd32(E1000_PBA);
1725 /* upper 16 bits has Tx packet buffer allocation size in KB */
1726 tx_space = pba >> 16;
1727 /* lower 16 bits has Rx packet buffer allocation size in KB */
1729 /* the tx fifo also stores 16 bytes of information about the tx
1730 * but don't include ethernet FCS because hardware appends it */
1731 min_tx_space = (adapter->max_frame_size +
1732 sizeof(union e1000_adv_tx_desc) -
1734 min_tx_space = ALIGN(min_tx_space, 1024);
1735 min_tx_space >>= 10;
1736 /* software strips receive CRC, so leave room for it */
1737 min_rx_space = adapter->max_frame_size;
1738 min_rx_space = ALIGN(min_rx_space, 1024);
1739 min_rx_space >>= 10;
1741 /* If current Tx allocation is less than the min Tx FIFO size,
1742 * and the min Tx FIFO size is less than the current Rx FIFO
1743 * allocation, take space away from current Rx allocation */
1744 if (tx_space < min_tx_space &&
1745 ((min_tx_space - tx_space) < pba)) {
1746 pba = pba - (min_tx_space - tx_space);
1748 /* if short on rx space, rx wins and must trump tx
1750 if (pba < min_rx_space)
1753 wr32(E1000_PBA, pba);
1756 /* flow control settings */
1757 /* The high water mark must be low enough to fit one full frame
1758 * (or the size used for early receive) above it in the Rx FIFO.
1759 * Set it to the lower of:
1760 * - 90% of the Rx FIFO size, or
1761 * - the full Rx FIFO size minus one full frame */
1762 hwm = min(((pba << 10) * 9 / 10),
1763 ((pba << 10) - 2 * adapter->max_frame_size));
1765 fc->high_water = hwm & 0xFFFFFFF0; /* 16-byte granularity */
1766 fc->low_water = fc->high_water - 16;
1767 fc->pause_time = 0xFFFF;
1769 fc->current_mode = fc->requested_mode;
1771 /* disable receive for all VFs and wait one second */
1772 if (adapter->vfs_allocated_count) {
1774 for (i = 0 ; i < adapter->vfs_allocated_count; i++)
1775 adapter->vf_data[i].flags &= IGB_VF_FLAG_PF_SET_MAC;
1777 /* ping all the active vfs to let them know we are going down */
1778 igb_ping_all_vfs(adapter);
1780 /* disable transmits and receives */
1781 wr32(E1000_VFRE, 0);
1782 wr32(E1000_VFTE, 0);
1785 /* Allow time for pending master requests to run */
1786 hw->mac.ops.reset_hw(hw);
1789 if (hw->mac.ops.init_hw(hw))
1790 dev_err(&pdev->dev, "Hardware Error\n");
1793 * Flow control settings reset on hardware reset, so guarantee flow
1794 * control is off when forcing speed.
1796 if (!hw->mac.autoneg)
1797 igb_force_mac_fc(hw);
1799 igb_init_dmac(adapter, pba);
1800 #ifdef CONFIG_IGB_HWMON
1801 /* Re-initialize the thermal sensor on i350 devices. */
1802 if (!test_bit(__IGB_DOWN, &adapter->state)) {
1803 if (mac->type == e1000_i350 && hw->bus.func == 0) {
1804 /* If present, re-initialize the external thermal sensor
1808 mac->ops.init_thermal_sensor_thresh(hw);
1812 if (!netif_running(adapter->netdev))
1813 igb_power_down_link(adapter);
1815 igb_update_mng_vlan(adapter);
1817 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
1818 wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
1820 /* Re-enable PTP, where applicable. */
1821 igb_ptp_reset(adapter);
1823 igb_get_phy_info(hw);
1826 static netdev_features_t igb_fix_features(struct net_device *netdev,
1827 netdev_features_t features)
1830 * Since there is no support for separate rx/tx vlan accel
1831 * enable/disable make sure tx flag is always in same state as rx.
1833 if (features & NETIF_F_HW_VLAN_RX)
1834 features |= NETIF_F_HW_VLAN_TX;
1836 features &= ~NETIF_F_HW_VLAN_TX;
1841 static int igb_set_features(struct net_device *netdev,
1842 netdev_features_t features)
1844 netdev_features_t changed = netdev->features ^ features;
1845 struct igb_adapter *adapter = netdev_priv(netdev);
1847 if (changed & NETIF_F_HW_VLAN_RX)
1848 igb_vlan_mode(netdev, features);
1850 if (!(changed & NETIF_F_RXALL))
1853 netdev->features = features;
1855 if (netif_running(netdev))
1856 igb_reinit_locked(adapter);
1863 static const struct net_device_ops igb_netdev_ops = {
1864 .ndo_open = igb_open,
1865 .ndo_stop = igb_close,
1866 .ndo_start_xmit = igb_xmit_frame,
1867 .ndo_get_stats64 = igb_get_stats64,
1868 .ndo_set_rx_mode = igb_set_rx_mode,
1869 .ndo_set_mac_address = igb_set_mac,
1870 .ndo_change_mtu = igb_change_mtu,
1871 .ndo_do_ioctl = igb_ioctl,
1872 .ndo_tx_timeout = igb_tx_timeout,
1873 .ndo_validate_addr = eth_validate_addr,
1874 .ndo_vlan_rx_add_vid = igb_vlan_rx_add_vid,
1875 .ndo_vlan_rx_kill_vid = igb_vlan_rx_kill_vid,
1876 .ndo_set_vf_mac = igb_ndo_set_vf_mac,
1877 .ndo_set_vf_vlan = igb_ndo_set_vf_vlan,
1878 .ndo_set_vf_tx_rate = igb_ndo_set_vf_bw,
1879 .ndo_get_vf_config = igb_ndo_get_vf_config,
1880 #ifdef CONFIG_NET_POLL_CONTROLLER
1881 .ndo_poll_controller = igb_netpoll,
1883 .ndo_fix_features = igb_fix_features,
1884 .ndo_set_features = igb_set_features,
1888 * igb_set_fw_version - Configure version string for ethtool
1889 * @adapter: adapter struct
1892 void igb_set_fw_version(struct igb_adapter *adapter)
1894 struct e1000_hw *hw = &adapter->hw;
1895 struct e1000_fw_version fw;
1897 igb_get_fw_version(hw, &fw);
1899 switch (hw->mac.type) {
1901 snprintf(adapter->fw_version, sizeof(adapter->fw_version),
1903 fw.invm_major, fw.invm_minor, fw.invm_img_type);
1907 /* if option is rom valid, display its version too */
1909 snprintf(adapter->fw_version,
1910 sizeof(adapter->fw_version),
1911 "%d.%d, 0x%08x, %d.%d.%d",
1912 fw.eep_major, fw.eep_minor, fw.etrack_id,
1913 fw.or_major, fw.or_build, fw.or_patch);
1916 snprintf(adapter->fw_version,
1917 sizeof(adapter->fw_version),
1919 fw.eep_major, fw.eep_minor, fw.etrack_id);
1926 static const struct i2c_board_info i350_sensor_info = {
1927 I2C_BOARD_INFO("i350bb", 0Xf8),
1930 /* igb_init_i2c - Init I2C interface
1931 * @adapter: pointer to adapter structure
1934 static s32 igb_init_i2c(struct igb_adapter *adapter)
1936 s32 status = E1000_SUCCESS;
1938 /* I2C interface supported on i350 devices */
1939 if (adapter->hw.mac.type != e1000_i350)
1940 return E1000_SUCCESS;
1942 /* Initialize the i2c bus which is controlled by the registers.
1943 * This bus will use the i2c_algo_bit structue that implements
1944 * the protocol through toggling of the 4 bits in the register.
1946 adapter->i2c_adap.owner = THIS_MODULE;
1947 adapter->i2c_algo = igb_i2c_algo;
1948 adapter->i2c_algo.data = adapter;
1949 adapter->i2c_adap.algo_data = &adapter->i2c_algo;
1950 adapter->i2c_adap.dev.parent = &adapter->pdev->dev;
1951 strlcpy(adapter->i2c_adap.name, "igb BB",
1952 sizeof(adapter->i2c_adap.name));
1953 status = i2c_bit_add_bus(&adapter->i2c_adap);
1958 * igb_probe - Device Initialization Routine
1959 * @pdev: PCI device information struct
1960 * @ent: entry in igb_pci_tbl
1962 * Returns 0 on success, negative on failure
1964 * igb_probe initializes an adapter identified by a pci_dev structure.
1965 * The OS initialization, configuring of the adapter private structure,
1966 * and a hardware reset occur.
1968 static int igb_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
1970 struct net_device *netdev;
1971 struct igb_adapter *adapter;
1972 struct e1000_hw *hw;
1973 u16 eeprom_data = 0;
1975 static int global_quad_port_a; /* global quad port a indication */
1976 const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
1977 unsigned long mmio_start, mmio_len;
1978 int err, pci_using_dac;
1979 u8 part_str[E1000_PBANUM_LENGTH];
1981 /* Catch broken hardware that put the wrong VF device ID in
1982 * the PCIe SR-IOV capability.
1984 if (pdev->is_virtfn) {
1985 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
1986 pci_name(pdev), pdev->vendor, pdev->device);
1990 err = pci_enable_device_mem(pdev);
1995 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
1997 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
2001 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
2003 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(32));
2005 dev_err(&pdev->dev, "No usable DMA "
2006 "configuration, aborting\n");
2012 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
2018 pci_enable_pcie_error_reporting(pdev);
2020 pci_set_master(pdev);
2021 pci_save_state(pdev);
2024 netdev = alloc_etherdev_mq(sizeof(struct igb_adapter),
2027 goto err_alloc_etherdev;
2029 SET_NETDEV_DEV(netdev, &pdev->dev);
2031 pci_set_drvdata(pdev, netdev);
2032 adapter = netdev_priv(netdev);
2033 adapter->netdev = netdev;
2034 adapter->pdev = pdev;
2037 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
2039 mmio_start = pci_resource_start(pdev, 0);
2040 mmio_len = pci_resource_len(pdev, 0);
2043 hw->hw_addr = ioremap(mmio_start, mmio_len);
2047 netdev->netdev_ops = &igb_netdev_ops;
2048 igb_set_ethtool_ops(netdev);
2049 netdev->watchdog_timeo = 5 * HZ;
2051 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2053 netdev->mem_start = mmio_start;
2054 netdev->mem_end = mmio_start + mmio_len;
2056 /* PCI config space info */
2057 hw->vendor_id = pdev->vendor;
2058 hw->device_id = pdev->device;
2059 hw->revision_id = pdev->revision;
2060 hw->subsystem_vendor_id = pdev->subsystem_vendor;
2061 hw->subsystem_device_id = pdev->subsystem_device;
2063 /* Copy the default MAC, PHY and NVM function pointers */
2064 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
2065 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
2066 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
2067 /* Initialize skew-specific constants */
2068 err = ei->get_invariants(hw);
2072 /* setup the private structure */
2073 err = igb_sw_init(adapter);
2077 igb_get_bus_info_pcie(hw);
2079 hw->phy.autoneg_wait_to_complete = false;
2081 /* Copper options */
2082 if (hw->phy.media_type == e1000_media_type_copper) {
2083 hw->phy.mdix = AUTO_ALL_MODES;
2084 hw->phy.disable_polarity_correction = false;
2085 hw->phy.ms_type = e1000_ms_hw_default;
2088 if (igb_check_reset_block(hw))
2089 dev_info(&pdev->dev,
2090 "PHY reset is blocked due to SOL/IDER session.\n");
2093 * features is initialized to 0 in allocation, it might have bits
2094 * set by igb_sw_init so we should use an or instead of an
2097 netdev->features |= NETIF_F_SG |
2104 NETIF_F_HW_VLAN_RX |
2107 /* copy netdev features into list of user selectable features */
2108 netdev->hw_features |= netdev->features;
2109 netdev->hw_features |= NETIF_F_RXALL;
2111 /* set this bit last since it cannot be part of hw_features */
2112 netdev->features |= NETIF_F_HW_VLAN_FILTER;
2114 netdev->vlan_features |= NETIF_F_TSO |
2120 netdev->priv_flags |= IFF_SUPP_NOFCS;
2122 if (pci_using_dac) {
2123 netdev->features |= NETIF_F_HIGHDMA;
2124 netdev->vlan_features |= NETIF_F_HIGHDMA;
2127 if (hw->mac.type >= e1000_82576) {
2128 netdev->hw_features |= NETIF_F_SCTP_CSUM;
2129 netdev->features |= NETIF_F_SCTP_CSUM;
2132 netdev->priv_flags |= IFF_UNICAST_FLT;
2134 adapter->en_mng_pt = igb_enable_mng_pass_thru(hw);
2136 /* before reading the NVM, reset the controller to put the device in a
2137 * known good starting state */
2138 hw->mac.ops.reset_hw(hw);
2141 * make sure the NVM is good , i211 parts have special NVM that
2142 * doesn't contain a checksum
2144 if (hw->mac.type != e1000_i211) {
2145 if (hw->nvm.ops.validate(hw) < 0) {
2146 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
2152 /* copy the MAC address out of the NVM */
2153 if (hw->mac.ops.read_mac_addr(hw))
2154 dev_err(&pdev->dev, "NVM Read Error\n");
2156 memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
2158 if (!is_valid_ether_addr(netdev->dev_addr)) {
2159 dev_err(&pdev->dev, "Invalid MAC Address\n");
2164 /* get firmware version for ethtool -i */
2165 igb_set_fw_version(adapter);
2167 setup_timer(&adapter->watchdog_timer, igb_watchdog,
2168 (unsigned long) adapter);
2169 setup_timer(&adapter->phy_info_timer, igb_update_phy_info,
2170 (unsigned long) adapter);
2172 INIT_WORK(&adapter->reset_task, igb_reset_task);
2173 INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
2175 /* Initialize link properties that are user-changeable */
2176 adapter->fc_autoneg = true;
2177 hw->mac.autoneg = true;
2178 hw->phy.autoneg_advertised = 0x2f;
2180 hw->fc.requested_mode = e1000_fc_default;
2181 hw->fc.current_mode = e1000_fc_default;
2183 igb_validate_mdi_setting(hw);
2185 /* By default, support wake on port A */
2186 if (hw->bus.func == 0)
2187 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2189 /* Check the NVM for wake support on non-port A ports */
2190 if (hw->mac.type >= e1000_82580)
2191 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A +
2192 NVM_82580_LAN_FUNC_OFFSET(hw->bus.func), 1,
2194 else if (hw->bus.func == 1)
2195 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
2197 if (eeprom_data & IGB_EEPROM_APME)
2198 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2200 /* now that we have the eeprom settings, apply the special cases where
2201 * the eeprom may be wrong or the board simply won't support wake on
2202 * lan on a particular port */
2203 switch (pdev->device) {
2204 case E1000_DEV_ID_82575GB_QUAD_COPPER:
2205 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2207 case E1000_DEV_ID_82575EB_FIBER_SERDES:
2208 case E1000_DEV_ID_82576_FIBER:
2209 case E1000_DEV_ID_82576_SERDES:
2210 /* Wake events only supported on port A for dual fiber
2211 * regardless of eeprom setting */
2212 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
2213 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2215 case E1000_DEV_ID_82576_QUAD_COPPER:
2216 case E1000_DEV_ID_82576_QUAD_COPPER_ET2:
2217 /* if quad port adapter, disable WoL on all but port A */
2218 if (global_quad_port_a != 0)
2219 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2221 adapter->flags |= IGB_FLAG_QUAD_PORT_A;
2222 /* Reset for multiple quad port adapters */
2223 if (++global_quad_port_a == 4)
2224 global_quad_port_a = 0;
2227 /* If the device can't wake, don't set software support */
2228 if (!device_can_wakeup(&adapter->pdev->dev))
2229 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2232 /* initialize the wol settings based on the eeprom settings */
2233 if (adapter->flags & IGB_FLAG_WOL_SUPPORTED)
2234 adapter->wol |= E1000_WUFC_MAG;
2236 /* Some vendors want WoL disabled by default, but still supported */
2237 if ((hw->mac.type == e1000_i350) &&
2238 (pdev->subsystem_vendor == PCI_VENDOR_ID_HP)) {
2239 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2243 device_set_wakeup_enable(&adapter->pdev->dev,
2244 adapter->flags & IGB_FLAG_WOL_SUPPORTED);
2246 /* reset the hardware with the new settings */
2249 /* Init the I2C interface */
2250 err = igb_init_i2c(adapter);
2252 dev_err(&pdev->dev, "failed to init i2c interface\n");
2256 /* let the f/w know that the h/w is now under the control of the
2258 igb_get_hw_control(adapter);
2260 strcpy(netdev->name, "eth%d");
2261 err = register_netdev(netdev);
2265 /* carrier off reporting is important to ethtool even BEFORE open */
2266 netif_carrier_off(netdev);
2268 #ifdef CONFIG_IGB_DCA
2269 if (dca_add_requester(&pdev->dev) == 0) {
2270 adapter->flags |= IGB_FLAG_DCA_ENABLED;
2271 dev_info(&pdev->dev, "DCA enabled\n");
2272 igb_setup_dca(adapter);
2276 #ifdef CONFIG_IGB_HWMON
2277 /* Initialize the thermal sensor on i350 devices. */
2278 if (hw->mac.type == e1000_i350 && hw->bus.func == 0) {
2282 * Read the NVM to determine if this i350 device supports an
2283 * external thermal sensor.
2285 hw->nvm.ops.read(hw, NVM_ETS_CFG, 1, &ets_word);
2286 if (ets_word != 0x0000 && ets_word != 0xFFFF)
2287 adapter->ets = true;
2289 adapter->ets = false;
2290 if (igb_sysfs_init(adapter))
2292 "failed to allocate sysfs resources\n");
2294 adapter->ets = false;
2297 /* do hw tstamp init after resetting */
2298 igb_ptp_init(adapter);
2300 dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
2301 /* print bus type/speed/width info */
2302 dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
2304 ((hw->bus.speed == e1000_bus_speed_2500) ? "2.5Gb/s" :
2305 (hw->bus.speed == e1000_bus_speed_5000) ? "5.0Gb/s" :
2307 ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
2308 (hw->bus.width == e1000_bus_width_pcie_x2) ? "Width x2" :
2309 (hw->bus.width == e1000_bus_width_pcie_x1) ? "Width x1" :
2313 ret_val = igb_read_part_string(hw, part_str, E1000_PBANUM_LENGTH);
2315 strcpy(part_str, "Unknown");
2316 dev_info(&pdev->dev, "%s: PBA No: %s\n", netdev->name, part_str);
2317 dev_info(&pdev->dev,
2318 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
2319 adapter->msix_entries ? "MSI-X" :
2320 (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
2321 adapter->num_rx_queues, adapter->num_tx_queues);
2322 switch (hw->mac.type) {
2326 igb_set_eee_i350(hw);
2332 pm_runtime_put_noidle(&pdev->dev);
2336 igb_release_hw_control(adapter);
2337 memset(&adapter->i2c_adap, 0, sizeof(adapter->i2c_adap));
2339 if (!igb_check_reset_block(hw))
2342 if (hw->flash_address)
2343 iounmap(hw->flash_address);
2345 igb_clear_interrupt_scheme(adapter);
2346 iounmap(hw->hw_addr);
2348 free_netdev(netdev);
2350 pci_release_selected_regions(pdev,
2351 pci_select_bars(pdev, IORESOURCE_MEM));
2354 pci_disable_device(pdev);
2358 #ifdef CONFIG_PCI_IOV
2359 static int igb_disable_sriov(struct pci_dev *pdev)
2361 struct net_device *netdev = pci_get_drvdata(pdev);
2362 struct igb_adapter *adapter = netdev_priv(netdev);
2363 struct e1000_hw *hw = &adapter->hw;
2365 /* reclaim resources allocated to VFs */
2366 if (adapter->vf_data) {
2367 /* disable iov and allow time for transactions to clear */
2368 if (igb_vfs_are_assigned(adapter)) {
2369 dev_warn(&pdev->dev,
2370 "Cannot deallocate SR-IOV virtual functions while they are assigned - VFs will not be deallocated\n");
2373 pci_disable_sriov(pdev);
2377 kfree(adapter->vf_data);
2378 adapter->vf_data = NULL;
2379 adapter->vfs_allocated_count = 0;
2380 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
2383 dev_info(&pdev->dev, "IOV Disabled\n");
2385 /* Re-enable DMA Coalescing flag since IOV is turned off */
2386 adapter->flags |= IGB_FLAG_DMAC;
2392 static int igb_enable_sriov(struct pci_dev *pdev, int num_vfs)
2394 struct net_device *netdev = pci_get_drvdata(pdev);
2395 struct igb_adapter *adapter = netdev_priv(netdev);
2396 int old_vfs = pci_num_vf(pdev);
2402 else if (old_vfs && old_vfs == num_vfs)
2404 else if (old_vfs && old_vfs != num_vfs)
2405 err = igb_disable_sriov(pdev);
2415 adapter->vfs_allocated_count = num_vfs;
2417 adapter->vf_data = kcalloc(adapter->vfs_allocated_count,
2418 sizeof(struct vf_data_storage), GFP_KERNEL);
2420 /* if allocation failed then we do not support SR-IOV */
2421 if (!adapter->vf_data) {
2422 adapter->vfs_allocated_count = 0;
2424 "Unable to allocate memory for VF Data Storage\n");
2429 err = pci_enable_sriov(pdev, adapter->vfs_allocated_count);
2433 dev_info(&pdev->dev, "%d VFs allocated\n",
2434 adapter->vfs_allocated_count);
2435 for (i = 0; i < adapter->vfs_allocated_count; i++)
2436 igb_vf_configure(adapter, i);
2438 /* DMA Coalescing is not supported in IOV mode. */
2439 adapter->flags &= ~IGB_FLAG_DMAC;
2443 kfree(adapter->vf_data);
2444 adapter->vf_data = NULL;
2445 adapter->vfs_allocated_count = 0;
2452 * igb_remove_i2c - Cleanup I2C interface
2453 * @adapter: pointer to adapter structure
2456 static void igb_remove_i2c(struct igb_adapter *adapter)
2459 /* free the adapter bus structure */
2460 i2c_del_adapter(&adapter->i2c_adap);
2464 * igb_remove - Device Removal Routine
2465 * @pdev: PCI device information struct
2467 * igb_remove is called by the PCI subsystem to alert the driver
2468 * that it should release a PCI device. The could be caused by a
2469 * Hot-Plug event, or because the driver is going to be removed from
2472 static void igb_remove(struct pci_dev *pdev)
2474 struct net_device *netdev = pci_get_drvdata(pdev);
2475 struct igb_adapter *adapter = netdev_priv(netdev);
2476 struct e1000_hw *hw = &adapter->hw;
2478 pm_runtime_get_noresume(&pdev->dev);
2479 #ifdef CONFIG_IGB_HWMON
2480 igb_sysfs_exit(adapter);
2482 igb_remove_i2c(adapter);
2483 igb_ptp_stop(adapter);
2485 * The watchdog timer may be rescheduled, so explicitly
2486 * disable watchdog from being rescheduled.
2488 set_bit(__IGB_DOWN, &adapter->state);
2489 del_timer_sync(&adapter->watchdog_timer);
2490 del_timer_sync(&adapter->phy_info_timer);
2492 cancel_work_sync(&adapter->reset_task);
2493 cancel_work_sync(&adapter->watchdog_task);
2495 #ifdef CONFIG_IGB_DCA
2496 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
2497 dev_info(&pdev->dev, "DCA disabled\n");
2498 dca_remove_requester(&pdev->dev);
2499 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
2500 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
2504 /* Release control of h/w to f/w. If f/w is AMT enabled, this
2505 * would have already happened in close and is redundant. */
2506 igb_release_hw_control(adapter);
2508 unregister_netdev(netdev);
2510 igb_clear_interrupt_scheme(adapter);
2512 #ifdef CONFIG_PCI_IOV
2513 igb_disable_sriov(pdev);
2516 iounmap(hw->hw_addr);
2517 if (hw->flash_address)
2518 iounmap(hw->flash_address);
2519 pci_release_selected_regions(pdev,
2520 pci_select_bars(pdev, IORESOURCE_MEM));
2522 kfree(adapter->shadow_vfta);
2523 free_netdev(netdev);
2525 pci_disable_pcie_error_reporting(pdev);
2527 pci_disable_device(pdev);
2531 * igb_probe_vfs - Initialize vf data storage and add VFs to pci config space
2532 * @adapter: board private structure to initialize
2534 * This function initializes the vf specific data storage and then attempts to
2535 * allocate the VFs. The reason for ordering it this way is because it is much
2536 * mor expensive time wise to disable SR-IOV than it is to allocate and free
2537 * the memory for the VFs.
2539 static void igb_probe_vfs(struct igb_adapter *adapter)
2541 #ifdef CONFIG_PCI_IOV
2542 struct pci_dev *pdev = adapter->pdev;
2543 struct e1000_hw *hw = &adapter->hw;
2545 /* Virtualization features not supported on i210 family. */
2546 if ((hw->mac.type == e1000_i210) || (hw->mac.type == e1000_i211))
2549 igb_enable_sriov(pdev, max_vfs);
2550 pci_sriov_set_totalvfs(pdev, 7);
2552 #endif /* CONFIG_PCI_IOV */
2555 static void igb_init_queue_configuration(struct igb_adapter *adapter)
2557 struct e1000_hw *hw = &adapter->hw;
2560 /* Determine the maximum number of RSS queues supported. */
2561 switch (hw->mac.type) {
2563 max_rss_queues = IGB_MAX_RX_QUEUES_I211;
2567 max_rss_queues = IGB_MAX_RX_QUEUES_82575;
2570 /* I350 cannot do RSS and SR-IOV at the same time */
2571 if (!!adapter->vfs_allocated_count) {
2577 if (!!adapter->vfs_allocated_count) {
2584 max_rss_queues = IGB_MAX_RX_QUEUES;
2588 adapter->rss_queues = min_t(u32, max_rss_queues, num_online_cpus());
2590 /* Determine if we need to pair queues. */
2591 switch (hw->mac.type) {
2594 /* Device supports enough interrupts without queue pairing. */
2598 * If VFs are going to be allocated with RSS queues then we
2599 * should pair the queues in order to conserve interrupts due
2600 * to limited supply.
2602 if ((adapter->rss_queues > 1) &&
2603 (adapter->vfs_allocated_count > 6))
2604 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
2611 * If rss_queues > half of max_rss_queues, pair the queues in
2612 * order to conserve interrupts due to limited supply.
2614 if (adapter->rss_queues > (max_rss_queues / 2))
2615 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
2621 * igb_sw_init - Initialize general software structures (struct igb_adapter)
2622 * @adapter: board private structure to initialize
2624 * igb_sw_init initializes the Adapter private data structure.
2625 * Fields are initialized based on PCI device information and
2626 * OS network device settings (MTU size).
2628 static int igb_sw_init(struct igb_adapter *adapter)
2630 struct e1000_hw *hw = &adapter->hw;
2631 struct net_device *netdev = adapter->netdev;
2632 struct pci_dev *pdev = adapter->pdev;
2634 pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
2636 /* set default ring sizes */
2637 adapter->tx_ring_count = IGB_DEFAULT_TXD;
2638 adapter->rx_ring_count = IGB_DEFAULT_RXD;
2640 /* set default ITR values */
2641 adapter->rx_itr_setting = IGB_DEFAULT_ITR;
2642 adapter->tx_itr_setting = IGB_DEFAULT_ITR;
2644 /* set default work limits */
2645 adapter->tx_work_limit = IGB_DEFAULT_TX_WORK;
2647 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN +
2649 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2651 spin_lock_init(&adapter->stats64_lock);
2652 #ifdef CONFIG_PCI_IOV
2653 switch (hw->mac.type) {
2657 dev_warn(&pdev->dev,
2658 "Maximum of 7 VFs per PF, using max\n");
2659 adapter->vfs_allocated_count = 7;
2661 adapter->vfs_allocated_count = max_vfs;
2662 if (adapter->vfs_allocated_count)
2663 dev_warn(&pdev->dev,
2664 "Enabling SR-IOV VFs using the module parameter is deprecated - please use the pci sysfs interface.\n");
2669 #endif /* CONFIG_PCI_IOV */
2671 igb_init_queue_configuration(adapter);
2673 /* Setup and initialize a copy of the hw vlan table array */
2674 adapter->shadow_vfta = kcalloc(E1000_VLAN_FILTER_TBL_SIZE, sizeof(u32),
2677 /* This call may decrease the number of queues */
2678 if (igb_init_interrupt_scheme(adapter, true)) {
2679 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
2683 igb_probe_vfs(adapter);
2685 /* Explicitly disable IRQ since the NIC can be in any state. */
2686 igb_irq_disable(adapter);
2688 if (hw->mac.type >= e1000_i350)
2689 adapter->flags &= ~IGB_FLAG_DMAC;
2691 set_bit(__IGB_DOWN, &adapter->state);
2696 * igb_open - Called when a network interface is made active
2697 * @netdev: network interface device structure
2699 * Returns 0 on success, negative value on failure
2701 * The open entry point is called when a network interface is made
2702 * active by the system (IFF_UP). At this point all resources needed
2703 * for transmit and receive operations are allocated, the interrupt
2704 * handler is registered with the OS, the watchdog timer is started,
2705 * and the stack is notified that the interface is ready.
2707 static int __igb_open(struct net_device *netdev, bool resuming)
2709 struct igb_adapter *adapter = netdev_priv(netdev);
2710 struct e1000_hw *hw = &adapter->hw;
2711 struct pci_dev *pdev = adapter->pdev;
2715 /* disallow open during test */
2716 if (test_bit(__IGB_TESTING, &adapter->state)) {
2722 pm_runtime_get_sync(&pdev->dev);
2724 netif_carrier_off(netdev);
2726 /* allocate transmit descriptors */
2727 err = igb_setup_all_tx_resources(adapter);
2731 /* allocate receive descriptors */
2732 err = igb_setup_all_rx_resources(adapter);
2736 igb_power_up_link(adapter);
2738 /* before we allocate an interrupt, we must be ready to handle it.
2739 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
2740 * as soon as we call pci_request_irq, so we have to setup our
2741 * clean_rx handler before we do so. */
2742 igb_configure(adapter);
2744 err = igb_request_irq(adapter);
2748 /* Notify the stack of the actual queue counts. */
2749 err = netif_set_real_num_tx_queues(adapter->netdev,
2750 adapter->num_tx_queues);
2752 goto err_set_queues;
2754 err = netif_set_real_num_rx_queues(adapter->netdev,
2755 adapter->num_rx_queues);
2757 goto err_set_queues;
2759 /* From here on the code is the same as igb_up() */
2760 clear_bit(__IGB_DOWN, &adapter->state);
2762 for (i = 0; i < adapter->num_q_vectors; i++)
2763 napi_enable(&(adapter->q_vector[i]->napi));
2765 /* Clear any pending interrupts. */
2768 igb_irq_enable(adapter);
2770 /* notify VFs that reset has been completed */
2771 if (adapter->vfs_allocated_count) {
2772 u32 reg_data = rd32(E1000_CTRL_EXT);
2773 reg_data |= E1000_CTRL_EXT_PFRSTD;
2774 wr32(E1000_CTRL_EXT, reg_data);
2777 netif_tx_start_all_queues(netdev);
2780 pm_runtime_put(&pdev->dev);
2782 /* start the watchdog. */
2783 hw->mac.get_link_status = 1;
2784 schedule_work(&adapter->watchdog_task);
2789 igb_free_irq(adapter);
2791 igb_release_hw_control(adapter);
2792 igb_power_down_link(adapter);
2793 igb_free_all_rx_resources(adapter);
2795 igb_free_all_tx_resources(adapter);
2799 pm_runtime_put(&pdev->dev);
2804 static int igb_open(struct net_device *netdev)
2806 return __igb_open(netdev, false);
2810 * igb_close - Disables a network interface
2811 * @netdev: network interface device structure
2813 * Returns 0, this is not allowed to fail
2815 * The close entry point is called when an interface is de-activated
2816 * by the OS. The hardware is still under the driver's control, but
2817 * needs to be disabled. A global MAC reset is issued to stop the
2818 * hardware, and all transmit and receive resources are freed.
2820 static int __igb_close(struct net_device *netdev, bool suspending)
2822 struct igb_adapter *adapter = netdev_priv(netdev);
2823 struct pci_dev *pdev = adapter->pdev;
2825 WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
2828 pm_runtime_get_sync(&pdev->dev);
2831 igb_free_irq(adapter);
2833 igb_free_all_tx_resources(adapter);
2834 igb_free_all_rx_resources(adapter);
2837 pm_runtime_put_sync(&pdev->dev);
2841 static int igb_close(struct net_device *netdev)
2843 return __igb_close(netdev, false);
2847 * igb_setup_tx_resources - allocate Tx resources (Descriptors)
2848 * @tx_ring: tx descriptor ring (for a specific queue) to setup
2850 * Return 0 on success, negative on failure
2852 int igb_setup_tx_resources(struct igb_ring *tx_ring)
2854 struct device *dev = tx_ring->dev;
2857 size = sizeof(struct igb_tx_buffer) * tx_ring->count;
2859 tx_ring->tx_buffer_info = vzalloc(size);
2860 if (!tx_ring->tx_buffer_info)
2863 /* round up to nearest 4K */
2864 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
2865 tx_ring->size = ALIGN(tx_ring->size, 4096);
2867 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
2868 &tx_ring->dma, GFP_KERNEL);
2872 tx_ring->next_to_use = 0;
2873 tx_ring->next_to_clean = 0;
2878 vfree(tx_ring->tx_buffer_info);
2879 tx_ring->tx_buffer_info = NULL;
2880 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
2885 * igb_setup_all_tx_resources - wrapper to allocate Tx resources
2886 * (Descriptors) for all queues
2887 * @adapter: board private structure
2889 * Return 0 on success, negative on failure
2891 static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
2893 struct pci_dev *pdev = adapter->pdev;
2896 for (i = 0; i < adapter->num_tx_queues; i++) {
2897 err = igb_setup_tx_resources(adapter->tx_ring[i]);
2900 "Allocation for Tx Queue %u failed\n", i);
2901 for (i--; i >= 0; i--)
2902 igb_free_tx_resources(adapter->tx_ring[i]);
2911 * igb_setup_tctl - configure the transmit control registers
2912 * @adapter: Board private structure
2914 void igb_setup_tctl(struct igb_adapter *adapter)
2916 struct e1000_hw *hw = &adapter->hw;
2919 /* disable queue 0 which is enabled by default on 82575 and 82576 */
2920 wr32(E1000_TXDCTL(0), 0);
2922 /* Program the Transmit Control Register */
2923 tctl = rd32(E1000_TCTL);
2924 tctl &= ~E1000_TCTL_CT;
2925 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2926 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2928 igb_config_collision_dist(hw);
2930 /* Enable transmits */
2931 tctl |= E1000_TCTL_EN;
2933 wr32(E1000_TCTL, tctl);
2937 * igb_configure_tx_ring - Configure transmit ring after Reset
2938 * @adapter: board private structure
2939 * @ring: tx ring to configure
2941 * Configure a transmit ring after a reset.
2943 void igb_configure_tx_ring(struct igb_adapter *adapter,
2944 struct igb_ring *ring)
2946 struct e1000_hw *hw = &adapter->hw;
2948 u64 tdba = ring->dma;
2949 int reg_idx = ring->reg_idx;
2951 /* disable the queue */
2952 wr32(E1000_TXDCTL(reg_idx), 0);
2956 wr32(E1000_TDLEN(reg_idx),
2957 ring->count * sizeof(union e1000_adv_tx_desc));
2958 wr32(E1000_TDBAL(reg_idx),
2959 tdba & 0x00000000ffffffffULL);
2960 wr32(E1000_TDBAH(reg_idx), tdba >> 32);
2962 ring->tail = hw->hw_addr + E1000_TDT(reg_idx);
2963 wr32(E1000_TDH(reg_idx), 0);
2964 writel(0, ring->tail);
2966 txdctl |= IGB_TX_PTHRESH;
2967 txdctl |= IGB_TX_HTHRESH << 8;
2968 txdctl |= IGB_TX_WTHRESH << 16;
2970 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
2971 wr32(E1000_TXDCTL(reg_idx), txdctl);
2975 * igb_configure_tx - Configure transmit Unit after Reset
2976 * @adapter: board private structure
2978 * Configure the Tx unit of the MAC after a reset.
2980 static void igb_configure_tx(struct igb_adapter *adapter)
2984 for (i = 0; i < adapter->num_tx_queues; i++)
2985 igb_configure_tx_ring(adapter, adapter->tx_ring[i]);
2989 * igb_setup_rx_resources - allocate Rx resources (Descriptors)
2990 * @rx_ring: rx descriptor ring (for a specific queue) to setup
2992 * Returns 0 on success, negative on failure
2994 int igb_setup_rx_resources(struct igb_ring *rx_ring)
2996 struct device *dev = rx_ring->dev;
2999 size = sizeof(struct igb_rx_buffer) * rx_ring->count;
3001 rx_ring->rx_buffer_info = vzalloc(size);
3002 if (!rx_ring->rx_buffer_info)
3005 /* Round up to nearest 4K */
3006 rx_ring->size = rx_ring->count * sizeof(union e1000_adv_rx_desc);
3007 rx_ring->size = ALIGN(rx_ring->size, 4096);
3009 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
3010 &rx_ring->dma, GFP_KERNEL);
3014 rx_ring->next_to_alloc = 0;
3015 rx_ring->next_to_clean = 0;
3016 rx_ring->next_to_use = 0;
3021 vfree(rx_ring->rx_buffer_info);
3022 rx_ring->rx_buffer_info = NULL;
3023 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
3028 * igb_setup_all_rx_resources - wrapper to allocate Rx resources
3029 * (Descriptors) for all queues
3030 * @adapter: board private structure
3032 * Return 0 on success, negative on failure
3034 static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
3036 struct pci_dev *pdev = adapter->pdev;
3039 for (i = 0; i < adapter->num_rx_queues; i++) {
3040 err = igb_setup_rx_resources(adapter->rx_ring[i]);
3043 "Allocation for Rx Queue %u failed\n", i);
3044 for (i--; i >= 0; i--)
3045 igb_free_rx_resources(adapter->rx_ring[i]);
3054 * igb_setup_mrqc - configure the multiple receive queue control registers
3055 * @adapter: Board private structure
3057 static void igb_setup_mrqc(struct igb_adapter *adapter)
3059 struct e1000_hw *hw = &adapter->hw;
3061 u32 j, num_rx_queues, shift = 0;
3062 static const u32 rsskey[10] = { 0xDA565A6D, 0xC20E5B25, 0x3D256741,
3063 0xB08FA343, 0xCB2BCAD0, 0xB4307BAE,
3064 0xA32DCB77, 0x0CF23080, 0x3BB7426A,
3067 /* Fill out hash function seeds */
3068 for (j = 0; j < 10; j++)
3069 wr32(E1000_RSSRK(j), rsskey[j]);
3071 num_rx_queues = adapter->rss_queues;
3073 switch (hw->mac.type) {
3078 /* 82576 supports 2 RSS queues for SR-IOV */
3079 if (adapter->vfs_allocated_count) {
3089 * Populate the indirection table 4 entries at a time. To do this
3090 * we are generating the results for n and n+2 and then interleaving
3091 * those with the results with n+1 and n+3.
3093 for (j = 0; j < 32; j++) {
3094 /* first pass generates n and n+2 */
3095 u32 base = ((j * 0x00040004) + 0x00020000) * num_rx_queues;
3096 u32 reta = (base & 0x07800780) >> (7 - shift);
3098 /* second pass generates n+1 and n+3 */
3099 base += 0x00010001 * num_rx_queues;
3100 reta |= (base & 0x07800780) << (1 + shift);
3102 wr32(E1000_RETA(j), reta);
3106 * Disable raw packet checksumming so that RSS hash is placed in
3107 * descriptor on writeback. No need to enable TCP/UDP/IP checksum
3108 * offloads as they are enabled by default
3110 rxcsum = rd32(E1000_RXCSUM);
3111 rxcsum |= E1000_RXCSUM_PCSD;
3113 if (adapter->hw.mac.type >= e1000_82576)
3114 /* Enable Receive Checksum Offload for SCTP */
3115 rxcsum |= E1000_RXCSUM_CRCOFL;
3117 /* Don't need to set TUOFL or IPOFL, they default to 1 */
3118 wr32(E1000_RXCSUM, rxcsum);
3120 /* Generate RSS hash based on packet types, TCP/UDP
3121 * port numbers and/or IPv4/v6 src and dst addresses
3123 mrqc = E1000_MRQC_RSS_FIELD_IPV4 |
3124 E1000_MRQC_RSS_FIELD_IPV4_TCP |
3125 E1000_MRQC_RSS_FIELD_IPV6 |
3126 E1000_MRQC_RSS_FIELD_IPV6_TCP |
3127 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX;
3129 if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV4_UDP)
3130 mrqc |= E1000_MRQC_RSS_FIELD_IPV4_UDP;
3131 if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV6_UDP)
3132 mrqc |= E1000_MRQC_RSS_FIELD_IPV6_UDP;
3134 /* If VMDq is enabled then we set the appropriate mode for that, else
3135 * we default to RSS so that an RSS hash is calculated per packet even
3136 * if we are only using one queue */
3137 if (adapter->vfs_allocated_count) {
3138 if (hw->mac.type > e1000_82575) {
3139 /* Set the default pool for the PF's first queue */
3140 u32 vtctl = rd32(E1000_VT_CTL);
3141 vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK |
3142 E1000_VT_CTL_DISABLE_DEF_POOL);
3143 vtctl |= adapter->vfs_allocated_count <<
3144 E1000_VT_CTL_DEFAULT_POOL_SHIFT;
3145 wr32(E1000_VT_CTL, vtctl);
3147 if (adapter->rss_queues > 1)
3148 mrqc |= E1000_MRQC_ENABLE_VMDQ_RSS_2Q;
3150 mrqc |= E1000_MRQC_ENABLE_VMDQ;
3152 if (hw->mac.type != e1000_i211)
3153 mrqc |= E1000_MRQC_ENABLE_RSS_4Q;
3155 igb_vmm_control(adapter);
3157 wr32(E1000_MRQC, mrqc);
3161 * igb_setup_rctl - configure the receive control registers
3162 * @adapter: Board private structure
3164 void igb_setup_rctl(struct igb_adapter *adapter)
3166 struct e1000_hw *hw = &adapter->hw;
3169 rctl = rd32(E1000_RCTL);
3171 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
3172 rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
3174 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
3175 (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
3178 * enable stripping of CRC. It's unlikely this will break BMC
3179 * redirection as it did with e1000. Newer features require
3180 * that the HW strips the CRC.
3182 rctl |= E1000_RCTL_SECRC;
3184 /* disable store bad packets and clear size bits. */
3185 rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
3187 /* enable LPE to prevent packets larger than max_frame_size */
3188 rctl |= E1000_RCTL_LPE;
3190 /* disable queue 0 to prevent tail write w/o re-config */
3191 wr32(E1000_RXDCTL(0), 0);
3193 /* Attention!!! For SR-IOV PF driver operations you must enable
3194 * queue drop for all VF and PF queues to prevent head of line blocking
3195 * if an un-trusted VF does not provide descriptors to hardware.
3197 if (adapter->vfs_allocated_count) {
3198 /* set all queue drop enable bits */
3199 wr32(E1000_QDE, ALL_QUEUES);
3202 /* This is useful for sniffing bad packets. */
3203 if (adapter->netdev->features & NETIF_F_RXALL) {
3204 /* UPE and MPE will be handled by normal PROMISC logic
3205 * in e1000e_set_rx_mode */
3206 rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
3207 E1000_RCTL_BAM | /* RX All Bcast Pkts */
3208 E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
3210 rctl &= ~(E1000_RCTL_VFE | /* Disable VLAN filter */
3211 E1000_RCTL_DPF | /* Allow filtered pause */
3212 E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
3213 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
3214 * and that breaks VLANs.
3218 wr32(E1000_RCTL, rctl);
3221 static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
3224 struct e1000_hw *hw = &adapter->hw;
3227 /* if it isn't the PF check to see if VFs are enabled and
3228 * increase the size to support vlan tags */
3229 if (vfn < adapter->vfs_allocated_count &&
3230 adapter->vf_data[vfn].vlans_enabled)
3231 size += VLAN_TAG_SIZE;
3233 vmolr = rd32(E1000_VMOLR(vfn));
3234 vmolr &= ~E1000_VMOLR_RLPML_MASK;
3235 vmolr |= size | E1000_VMOLR_LPE;
3236 wr32(E1000_VMOLR(vfn), vmolr);
3242 * igb_rlpml_set - set maximum receive packet size
3243 * @adapter: board private structure
3245 * Configure maximum receivable packet size.
3247 static void igb_rlpml_set(struct igb_adapter *adapter)
3249 u32 max_frame_size = adapter->max_frame_size;
3250 struct e1000_hw *hw = &adapter->hw;
3251 u16 pf_id = adapter->vfs_allocated_count;
3254 igb_set_vf_rlpml(adapter, max_frame_size, pf_id);
3256 * If we're in VMDQ or SR-IOV mode, then set global RLPML
3257 * to our max jumbo frame size, in case we need to enable
3258 * jumbo frames on one of the rings later.
3259 * This will not pass over-length frames into the default
3260 * queue because it's gated by the VMOLR.RLPML.
3262 max_frame_size = MAX_JUMBO_FRAME_SIZE;
3265 wr32(E1000_RLPML, max_frame_size);
3268 static inline void igb_set_vmolr(struct igb_adapter *adapter,
3271 struct e1000_hw *hw = &adapter->hw;
3275 * This register exists only on 82576 and newer so if we are older then
3276 * we should exit and do nothing
3278 if (hw->mac.type < e1000_82576)
3281 vmolr = rd32(E1000_VMOLR(vfn));
3282 vmolr |= E1000_VMOLR_STRVLAN; /* Strip vlan tags */
3284 vmolr |= E1000_VMOLR_AUPE; /* Accept untagged packets */
3286 vmolr &= ~(E1000_VMOLR_AUPE); /* Tagged packets ONLY */
3288 /* clear all bits that might not be set */
3289 vmolr &= ~(E1000_VMOLR_BAM | E1000_VMOLR_RSSE);
3291 if (adapter->rss_queues > 1 && vfn == adapter->vfs_allocated_count)
3292 vmolr |= E1000_VMOLR_RSSE; /* enable RSS */
3294 * for VMDq only allow the VFs and pool 0 to accept broadcast and
3297 if (vfn <= adapter->vfs_allocated_count)
3298 vmolr |= E1000_VMOLR_BAM; /* Accept broadcast */
3300 wr32(E1000_VMOLR(vfn), vmolr);
3304 * igb_configure_rx_ring - Configure a receive ring after Reset
3305 * @adapter: board private structure
3306 * @ring: receive ring to be configured
3308 * Configure the Rx unit of the MAC after a reset.
3310 void igb_configure_rx_ring(struct igb_adapter *adapter,
3311 struct igb_ring *ring)
3313 struct e1000_hw *hw = &adapter->hw;
3314 u64 rdba = ring->dma;
3315 int reg_idx = ring->reg_idx;
3316 u32 srrctl = 0, rxdctl = 0;
3318 /* disable the queue */
3319 wr32(E1000_RXDCTL(reg_idx), 0);
3321 /* Set DMA base address registers */
3322 wr32(E1000_RDBAL(reg_idx),
3323 rdba & 0x00000000ffffffffULL);
3324 wr32(E1000_RDBAH(reg_idx), rdba >> 32);
3325 wr32(E1000_RDLEN(reg_idx),
3326 ring->count * sizeof(union e1000_adv_rx_desc));
3328 /* initialize head and tail */
3329 ring->tail = hw->hw_addr + E1000_RDT(reg_idx);
3330 wr32(E1000_RDH(reg_idx), 0);
3331 writel(0, ring->tail);
3333 /* set descriptor configuration */
3334 srrctl = IGB_RX_HDR_LEN << E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
3335 srrctl |= IGB_RX_BUFSZ >> E1000_SRRCTL_BSIZEPKT_SHIFT;
3336 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
3337 if (hw->mac.type >= e1000_82580)
3338 srrctl |= E1000_SRRCTL_TIMESTAMP;
3339 /* Only set Drop Enable if we are supporting multiple queues */
3340 if (adapter->vfs_allocated_count || adapter->num_rx_queues > 1)
3341 srrctl |= E1000_SRRCTL_DROP_EN;
3343 wr32(E1000_SRRCTL(reg_idx), srrctl);
3345 /* set filtering for VMDQ pools */
3346 igb_set_vmolr(adapter, reg_idx & 0x7, true);
3348 rxdctl |= IGB_RX_PTHRESH;
3349 rxdctl |= IGB_RX_HTHRESH << 8;
3350 rxdctl |= IGB_RX_WTHRESH << 16;
3352 /* enable receive descriptor fetching */
3353 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
3354 wr32(E1000_RXDCTL(reg_idx), rxdctl);
3357 static void igb_set_rx_buffer_len(struct igb_adapter *adapter,
3358 struct igb_ring *rx_ring)
3360 #define IGB_MAX_BUILD_SKB_SIZE \
3361 (SKB_WITH_OVERHEAD(IGB_RX_BUFSZ) - \
3362 (NET_SKB_PAD + NET_IP_ALIGN + IGB_TS_HDR_LEN))
3364 /* set build_skb flag */
3365 if (adapter->max_frame_size <= IGB_MAX_BUILD_SKB_SIZE)
3366 set_ring_build_skb_enabled(rx_ring);
3368 clear_ring_build_skb_enabled(rx_ring);
3372 * igb_configure_rx - Configure receive Unit after Reset
3373 * @adapter: board private structure
3375 * Configure the Rx unit of the MAC after a reset.
3377 static void igb_configure_rx(struct igb_adapter *adapter)
3381 /* set UTA to appropriate mode */
3382 igb_set_uta(adapter);
3384 /* set the correct pool for the PF default MAC address in entry 0 */
3385 igb_rar_set_qsel(adapter, adapter->hw.mac.addr, 0,
3386 adapter->vfs_allocated_count);
3388 /* Setup the HW Rx Head and Tail Descriptor Pointers and
3389 * the Base and Length of the Rx Descriptor Ring */
3390 for (i = 0; i < adapter->num_rx_queues; i++) {
3391 struct igb_ring *rx_ring = adapter->rx_ring[i];
3392 igb_set_rx_buffer_len(adapter, rx_ring);
3393 igb_configure_rx_ring(adapter, rx_ring);
3398 * igb_free_tx_resources - Free Tx Resources per Queue
3399 * @tx_ring: Tx descriptor ring for a specific queue
3401 * Free all transmit software resources
3403 void igb_free_tx_resources(struct igb_ring *tx_ring)
3405 igb_clean_tx_ring(tx_ring);
3407 vfree(tx_ring->tx_buffer_info);
3408 tx_ring->tx_buffer_info = NULL;
3410 /* if not set, then don't free */
3414 dma_free_coherent(tx_ring->dev, tx_ring->size,
3415 tx_ring->desc, tx_ring->dma);
3417 tx_ring->desc = NULL;
3421 * igb_free_all_tx_resources - Free Tx Resources for All Queues
3422 * @adapter: board private structure
3424 * Free all transmit software resources
3426 static void igb_free_all_tx_resources(struct igb_adapter *adapter)
3430 for (i = 0; i < adapter->num_tx_queues; i++)
3431 igb_free_tx_resources(adapter->tx_ring[i]);
3434 void igb_unmap_and_free_tx_resource(struct igb_ring *ring,
3435 struct igb_tx_buffer *tx_buffer)
3437 if (tx_buffer->skb) {
3438 dev_kfree_skb_any(tx_buffer->skb);
3439 if (dma_unmap_len(tx_buffer, len))
3440 dma_unmap_single(ring->dev,
3441 dma_unmap_addr(tx_buffer, dma),
3442 dma_unmap_len(tx_buffer, len),
3444 } else if (dma_unmap_len(tx_buffer, len)) {
3445 dma_unmap_page(ring->dev,
3446 dma_unmap_addr(tx_buffer, dma),
3447 dma_unmap_len(tx_buffer, len),
3450 tx_buffer->next_to_watch = NULL;
3451 tx_buffer->skb = NULL;
3452 dma_unmap_len_set(tx_buffer, len, 0);
3453 /* buffer_info must be completely set up in the transmit path */
3457 * igb_clean_tx_ring - Free Tx Buffers
3458 * @tx_ring: ring to be cleaned
3460 static void igb_clean_tx_ring(struct igb_ring *tx_ring)
3462 struct igb_tx_buffer *buffer_info;
3466 if (!tx_ring->tx_buffer_info)
3468 /* Free all the Tx ring sk_buffs */
3470 for (i = 0; i < tx_ring->count; i++) {
3471 buffer_info = &tx_ring->tx_buffer_info[i];
3472 igb_unmap_and_free_tx_resource(tx_ring, buffer_info);
3475 netdev_tx_reset_queue(txring_txq(tx_ring));
3477 size = sizeof(struct igb_tx_buffer) * tx_ring->count;
3478 memset(tx_ring->tx_buffer_info, 0, size);
3480 /* Zero out the descriptor ring */
3481 memset(tx_ring->desc, 0, tx_ring->size);
3483 tx_ring->next_to_use = 0;
3484 tx_ring->next_to_clean = 0;
3488 * igb_clean_all_tx_rings - Free Tx Buffers for all queues
3489 * @adapter: board private structure
3491 static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
3495 for (i = 0; i < adapter->num_tx_queues; i++)
3496 igb_clean_tx_ring(adapter->tx_ring[i]);
3500 * igb_free_rx_resources - Free Rx Resources
3501 * @rx_ring: ring to clean the resources from
3503 * Free all receive software resources
3505 void igb_free_rx_resources(struct igb_ring *rx_ring)
3507 igb_clean_rx_ring(rx_ring);
3509 vfree(rx_ring->rx_buffer_info);
3510 rx_ring->rx_buffer_info = NULL;
3512 /* if not set, then don't free */
3516 dma_free_coherent(rx_ring->dev, rx_ring->size,
3517 rx_ring->desc, rx_ring->dma);
3519 rx_ring->desc = NULL;
3523 * igb_free_all_rx_resources - Free Rx Resources for All Queues
3524 * @adapter: board private structure
3526 * Free all receive software resources
3528 static void igb_free_all_rx_resources(struct igb_adapter *adapter)
3532 for (i = 0; i < adapter->num_rx_queues; i++)
3533 igb_free_rx_resources(adapter->rx_ring[i]);
3537 * igb_clean_rx_ring - Free Rx Buffers per Queue
3538 * @rx_ring: ring to free buffers from
3540 static void igb_clean_rx_ring(struct igb_ring *rx_ring)
3546 dev_kfree_skb(rx_ring->skb);
3547 rx_ring->skb = NULL;
3549 if (!rx_ring->rx_buffer_info)
3552 /* Free all the Rx ring sk_buffs */
3553 for (i = 0; i < rx_ring->count; i++) {
3554 struct igb_rx_buffer *buffer_info = &rx_ring->rx_buffer_info[i];
3556 if (!buffer_info->page)
3559 dma_unmap_page(rx_ring->dev,
3563 __free_page(buffer_info->page);
3565 buffer_info->page = NULL;
3568 size = sizeof(struct igb_rx_buffer) * rx_ring->count;
3569 memset(rx_ring->rx_buffer_info, 0, size);
3571 /* Zero out the descriptor ring */
3572 memset(rx_ring->desc, 0, rx_ring->size);
3574 rx_ring->next_to_alloc = 0;
3575 rx_ring->next_to_clean = 0;
3576 rx_ring->next_to_use = 0;
3580 * igb_clean_all_rx_rings - Free Rx Buffers for all queues
3581 * @adapter: board private structure
3583 static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
3587 for (i = 0; i < adapter->num_rx_queues; i++)
3588 igb_clean_rx_ring(adapter->rx_ring[i]);
3592 * igb_set_mac - Change the Ethernet Address of the NIC
3593 * @netdev: network interface device structure
3594 * @p: pointer to an address structure
3596 * Returns 0 on success, negative on failure
3598 static int igb_set_mac(struct net_device *netdev, void *p)
3600 struct igb_adapter *adapter = netdev_priv(netdev);
3601 struct e1000_hw *hw = &adapter->hw;
3602 struct sockaddr *addr = p;
3604 if (!is_valid_ether_addr(addr->sa_data))
3605 return -EADDRNOTAVAIL;
3607 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3608 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
3610 /* set the correct pool for the new PF MAC address in entry 0 */
3611 igb_rar_set_qsel(adapter, hw->mac.addr, 0,
3612 adapter->vfs_allocated_count);
3618 * igb_write_mc_addr_list - write multicast addresses to MTA
3619 * @netdev: network interface device structure
3621 * Writes multicast address list to the MTA hash table.
3622 * Returns: -ENOMEM on failure
3623 * 0 on no addresses written
3624 * X on writing X addresses to MTA
3626 static int igb_write_mc_addr_list(struct net_device *netdev)
3628 struct igb_adapter *adapter = netdev_priv(netdev);
3629 struct e1000_hw *hw = &adapter->hw;
3630 struct netdev_hw_addr *ha;
3634 if (netdev_mc_empty(netdev)) {
3635 /* nothing to program, so clear mc list */
3636 igb_update_mc_addr_list(hw, NULL, 0);
3637 igb_restore_vf_multicasts(adapter);
3641 mta_list = kzalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
3645 /* The shared function expects a packed array of only addresses. */
3647 netdev_for_each_mc_addr(ha, netdev)
3648 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3650 igb_update_mc_addr_list(hw, mta_list, i);
3653 return netdev_mc_count(netdev);
3657 * igb_write_uc_addr_list - write unicast addresses to RAR table
3658 * @netdev: network interface device structure
3660 * Writes unicast address list to the RAR table.
3661 * Returns: -ENOMEM on failure/insufficient address space
3662 * 0 on no addresses written
3663 * X on writing X addresses to the RAR table
3665 static int igb_write_uc_addr_list(struct net_device *netdev)
3667 struct igb_adapter *adapter = netdev_priv(netdev);
3668 struct e1000_hw *hw = &adapter->hw;
3669 unsigned int vfn = adapter->vfs_allocated_count;
3670 unsigned int rar_entries = hw->mac.rar_entry_count - (vfn + 1);
3673 /* return ENOMEM indicating insufficient memory for addresses */
3674 if (netdev_uc_count(netdev) > rar_entries)
3677 if (!netdev_uc_empty(netdev) && rar_entries) {
3678 struct netdev_hw_addr *ha;
3680 netdev_for_each_uc_addr(ha, netdev) {
3683 igb_rar_set_qsel(adapter, ha->addr,
3689 /* write the addresses in reverse order to avoid write combining */
3690 for (; rar_entries > 0 ; rar_entries--) {
3691 wr32(E1000_RAH(rar_entries), 0);
3692 wr32(E1000_RAL(rar_entries), 0);
3700 * igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
3701 * @netdev: network interface device structure
3703 * The set_rx_mode entry point is called whenever the unicast or multicast
3704 * address lists or the network interface flags are updated. This routine is
3705 * responsible for configuring the hardware for proper unicast, multicast,
3706 * promiscuous mode, and all-multi behavior.
3708 static void igb_set_rx_mode(struct net_device *netdev)
3710 struct igb_adapter *adapter = netdev_priv(netdev);
3711 struct e1000_hw *hw = &adapter->hw;
3712 unsigned int vfn = adapter->vfs_allocated_count;
3713 u32 rctl, vmolr = 0;
3716 /* Check for Promiscuous and All Multicast modes */
3717 rctl = rd32(E1000_RCTL);
3719 /* clear the effected bits */
3720 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE | E1000_RCTL_VFE);
3722 if (netdev->flags & IFF_PROMISC) {
3723 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3724 vmolr |= (E1000_VMOLR_ROPE | E1000_VMOLR_MPME);
3726 if (netdev->flags & IFF_ALLMULTI) {
3727 rctl |= E1000_RCTL_MPE;
3728 vmolr |= E1000_VMOLR_MPME;
3731 * Write addresses to the MTA, if the attempt fails
3732 * then we should just turn on promiscuous mode so
3733 * that we can at least receive multicast traffic
3735 count = igb_write_mc_addr_list(netdev);
3737 rctl |= E1000_RCTL_MPE;
3738 vmolr |= E1000_VMOLR_MPME;
3740 vmolr |= E1000_VMOLR_ROMPE;
3744 * Write addresses to available RAR registers, if there is not
3745 * sufficient space to store all the addresses then enable
3746 * unicast promiscuous mode
3748 count = igb_write_uc_addr_list(netdev);
3750 rctl |= E1000_RCTL_UPE;
3751 vmolr |= E1000_VMOLR_ROPE;
3753 rctl |= E1000_RCTL_VFE;
3755 wr32(E1000_RCTL, rctl);
3758 * In order to support SR-IOV and eventually VMDq it is necessary to set
3759 * the VMOLR to enable the appropriate modes. Without this workaround
3760 * we will have issues with VLAN tag stripping not being done for frames
3761 * that are only arriving because we are the default pool
3763 if ((hw->mac.type < e1000_82576) || (hw->mac.type > e1000_i350))
3766 vmolr |= rd32(E1000_VMOLR(vfn)) &
3767 ~(E1000_VMOLR_ROPE | E1000_VMOLR_MPME | E1000_VMOLR_ROMPE);
3768 wr32(E1000_VMOLR(vfn), vmolr);
3769 igb_restore_vf_multicasts(adapter);
3772 static void igb_check_wvbr(struct igb_adapter *adapter)
3774 struct e1000_hw *hw = &adapter->hw;
3777 switch (hw->mac.type) {
3780 if (!(wvbr = rd32(E1000_WVBR)))
3787 adapter->wvbr |= wvbr;
3790 #define IGB_STAGGERED_QUEUE_OFFSET 8
3792 static void igb_spoof_check(struct igb_adapter *adapter)
3799 for(j = 0; j < adapter->vfs_allocated_count; j++) {
3800 if (adapter->wvbr & (1 << j) ||
3801 adapter->wvbr & (1 << (j + IGB_STAGGERED_QUEUE_OFFSET))) {
3802 dev_warn(&adapter->pdev->dev,
3803 "Spoof event(s) detected on VF %d\n", j);
3806 (1 << (j + IGB_STAGGERED_QUEUE_OFFSET)));
3811 /* Need to wait a few seconds after link up to get diagnostic information from
3813 static void igb_update_phy_info(unsigned long data)
3815 struct igb_adapter *adapter = (struct igb_adapter *) data;
3816 igb_get_phy_info(&adapter->hw);
3820 * igb_has_link - check shared code for link and determine up/down
3821 * @adapter: pointer to driver private info
3823 bool igb_has_link(struct igb_adapter *adapter)
3825 struct e1000_hw *hw = &adapter->hw;
3826 bool link_active = false;
3829 /* get_link_status is set on LSC (link status) interrupt or
3830 * rx sequence error interrupt. get_link_status will stay
3831 * false until the e1000_check_for_link establishes link
3832 * for copper adapters ONLY
3834 switch (hw->phy.media_type) {
3835 case e1000_media_type_copper:
3836 if (hw->mac.get_link_status) {
3837 ret_val = hw->mac.ops.check_for_link(hw);
3838 link_active = !hw->mac.get_link_status;
3843 case e1000_media_type_internal_serdes:
3844 ret_val = hw->mac.ops.check_for_link(hw);
3845 link_active = hw->mac.serdes_has_link;
3848 case e1000_media_type_unknown:
3855 static bool igb_thermal_sensor_event(struct e1000_hw *hw, u32 event)
3858 u32 ctrl_ext, thstat;
3860 /* check for thermal sensor event on i350 copper only */
3861 if (hw->mac.type == e1000_i350) {
3862 thstat = rd32(E1000_THSTAT);
3863 ctrl_ext = rd32(E1000_CTRL_EXT);
3865 if ((hw->phy.media_type == e1000_media_type_copper) &&
3866 !(ctrl_ext & E1000_CTRL_EXT_LINK_MODE_SGMII)) {
3867 ret = !!(thstat & event);
3875 * igb_watchdog - Timer Call-back
3876 * @data: pointer to adapter cast into an unsigned long
3878 static void igb_watchdog(unsigned long data)
3880 struct igb_adapter *adapter = (struct igb_adapter *)data;
3881 /* Do the rest outside of interrupt context */
3882 schedule_work(&adapter->watchdog_task);
3885 static void igb_watchdog_task(struct work_struct *work)
3887 struct igb_adapter *adapter = container_of(work,
3890 struct e1000_hw *hw = &adapter->hw;
3891 struct net_device *netdev = adapter->netdev;
3895 link = igb_has_link(adapter);
3897 /* Cancel scheduled suspend requests. */
3898 pm_runtime_resume(netdev->dev.parent);
3900 if (!netif_carrier_ok(netdev)) {
3902 hw->mac.ops.get_speed_and_duplex(hw,
3903 &adapter->link_speed,
3904 &adapter->link_duplex);
3906 ctrl = rd32(E1000_CTRL);
3907 /* Links status message must follow this format */
3908 printk(KERN_INFO "igb: %s NIC Link is Up %d Mbps %s "
3909 "Duplex, Flow Control: %s\n",
3911 adapter->link_speed,
3912 adapter->link_duplex == FULL_DUPLEX ?
3914 (ctrl & E1000_CTRL_TFCE) &&
3915 (ctrl & E1000_CTRL_RFCE) ? "RX/TX" :
3916 (ctrl & E1000_CTRL_RFCE) ? "RX" :
3917 (ctrl & E1000_CTRL_TFCE) ? "TX" : "None");
3919 /* check for thermal sensor event */
3920 if (igb_thermal_sensor_event(hw,
3921 E1000_THSTAT_LINK_THROTTLE)) {
3922 netdev_info(netdev, "The network adapter link "
3923 "speed was downshifted because it "
3927 /* adjust timeout factor according to speed/duplex */
3928 adapter->tx_timeout_factor = 1;
3929 switch (adapter->link_speed) {
3931 adapter->tx_timeout_factor = 14;
3934 /* maybe add some timeout factor ? */
3938 netif_carrier_on(netdev);
3940 igb_ping_all_vfs(adapter);
3941 igb_check_vf_rate_limit(adapter);
3943 /* link state has changed, schedule phy info update */
3944 if (!test_bit(__IGB_DOWN, &adapter->state))
3945 mod_timer(&adapter->phy_info_timer,
3946 round_jiffies(jiffies + 2 * HZ));
3949 if (netif_carrier_ok(netdev)) {
3950 adapter->link_speed = 0;
3951 adapter->link_duplex = 0;
3953 /* check for thermal sensor event */
3954 if (igb_thermal_sensor_event(hw,
3955 E1000_THSTAT_PWR_DOWN)) {
3956 netdev_err(netdev, "The network adapter was "
3957 "stopped because it overheated\n");
3960 /* Links status message must follow this format */
3961 printk(KERN_INFO "igb: %s NIC Link is Down\n",
3963 netif_carrier_off(netdev);
3965 igb_ping_all_vfs(adapter);
3967 /* link state has changed, schedule phy info update */
3968 if (!test_bit(__IGB_DOWN, &adapter->state))
3969 mod_timer(&adapter->phy_info_timer,
3970 round_jiffies(jiffies + 2 * HZ));
3972 pm_schedule_suspend(netdev->dev.parent,
3977 spin_lock(&adapter->stats64_lock);
3978 igb_update_stats(adapter, &adapter->stats64);
3979 spin_unlock(&adapter->stats64_lock);
3981 for (i = 0; i < adapter->num_tx_queues; i++) {
3982 struct igb_ring *tx_ring = adapter->tx_ring[i];
3983 if (!netif_carrier_ok(netdev)) {
3984 /* We've lost link, so the controller stops DMA,
3985 * but we've got queued Tx work that's never going
3986 * to get done, so reset controller to flush Tx.
3987 * (Do the reset outside of interrupt context). */
3988 if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) {
3989 adapter->tx_timeout_count++;
3990 schedule_work(&adapter->reset_task);
3991 /* return immediately since reset is imminent */
3996 /* Force detection of hung controller every watchdog period */
3997 set_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
4000 /* Cause software interrupt to ensure rx ring is cleaned */
4001 if (adapter->msix_entries) {
4003 for (i = 0; i < adapter->num_q_vectors; i++)
4004 eics |= adapter->q_vector[i]->eims_value;
4005 wr32(E1000_EICS, eics);
4007 wr32(E1000_ICS, E1000_ICS_RXDMT0);
4010 igb_spoof_check(adapter);
4011 igb_ptp_rx_hang(adapter);
4013 /* Reset the timer */
4014 if (!test_bit(__IGB_DOWN, &adapter->state))
4015 mod_timer(&adapter->watchdog_timer,
4016 round_jiffies(jiffies + 2 * HZ));
4019 enum latency_range {
4023 latency_invalid = 255
4027 * igb_update_ring_itr - update the dynamic ITR value based on packet size
4029 * Stores a new ITR value based on strictly on packet size. This
4030 * algorithm is less sophisticated than that used in igb_update_itr,
4031 * due to the difficulty of synchronizing statistics across multiple
4032 * receive rings. The divisors and thresholds used by this function
4033 * were determined based on theoretical maximum wire speed and testing
4034 * data, in order to minimize response time while increasing bulk
4036 * This functionality is controlled by the InterruptThrottleRate module
4037 * parameter (see igb_param.c)
4038 * NOTE: This function is called only when operating in a multiqueue
4039 * receive environment.
4040 * @q_vector: pointer to q_vector
4042 static void igb_update_ring_itr(struct igb_q_vector *q_vector)
4044 int new_val = q_vector->itr_val;
4045 int avg_wire_size = 0;
4046 struct igb_adapter *adapter = q_vector->adapter;
4047 unsigned int packets;
4049 /* For non-gigabit speeds, just fix the interrupt rate at 4000
4050 * ints/sec - ITR timer value of 120 ticks.
4052 if (adapter->link_speed != SPEED_1000) {
4053 new_val = IGB_4K_ITR;
4057 packets = q_vector->rx.total_packets;
4059 avg_wire_size = q_vector->rx.total_bytes / packets;
4061 packets = q_vector->tx.total_packets;
4063 avg_wire_size = max_t(u32, avg_wire_size,
4064 q_vector->tx.total_bytes / packets);
4066 /* if avg_wire_size isn't set no work was done */
4070 /* Add 24 bytes to size to account for CRC, preamble, and gap */
4071 avg_wire_size += 24;
4073 /* Don't starve jumbo frames */
4074 avg_wire_size = min(avg_wire_size, 3000);
4076 /* Give a little boost to mid-size frames */
4077 if ((avg_wire_size > 300) && (avg_wire_size < 1200))
4078 new_val = avg_wire_size / 3;
4080 new_val = avg_wire_size / 2;
4082 /* conservative mode (itr 3) eliminates the lowest_latency setting */
4083 if (new_val < IGB_20K_ITR &&
4084 ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
4085 (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
4086 new_val = IGB_20K_ITR;
4089 if (new_val != q_vector->itr_val) {
4090 q_vector->itr_val = new_val;
4091 q_vector->set_itr = 1;
4094 q_vector->rx.total_bytes = 0;
4095 q_vector->rx.total_packets = 0;
4096 q_vector->tx.total_bytes = 0;
4097 q_vector->tx.total_packets = 0;
4101 * igb_update_itr - update the dynamic ITR value based on statistics
4102 * Stores a new ITR value based on packets and byte
4103 * counts during the last interrupt. The advantage of per interrupt
4104 * computation is faster updates and more accurate ITR for the current
4105 * traffic pattern. Constants in this function were computed
4106 * based on theoretical maximum wire speed and thresholds were set based
4107 * on testing data as well as attempting to minimize response time
4108 * while increasing bulk throughput.
4109 * this functionality is controlled by the InterruptThrottleRate module
4110 * parameter (see igb_param.c)
4111 * NOTE: These calculations are only valid when operating in a single-
4112 * queue environment.
4113 * @q_vector: pointer to q_vector
4114 * @ring_container: ring info to update the itr for
4116 static void igb_update_itr(struct igb_q_vector *q_vector,
4117 struct igb_ring_container *ring_container)
4119 unsigned int packets = ring_container->total_packets;
4120 unsigned int bytes = ring_container->total_bytes;
4121 u8 itrval = ring_container->itr;
4123 /* no packets, exit with status unchanged */
4128 case lowest_latency:
4129 /* handle TSO and jumbo frames */
4130 if (bytes/packets > 8000)
4131 itrval = bulk_latency;
4132 else if ((packets < 5) && (bytes > 512))
4133 itrval = low_latency;
4135 case low_latency: /* 50 usec aka 20000 ints/s */
4136 if (bytes > 10000) {
4137 /* this if handles the TSO accounting */
4138 if (bytes/packets > 8000) {
4139 itrval = bulk_latency;
4140 } else if ((packets < 10) || ((bytes/packets) > 1200)) {
4141 itrval = bulk_latency;
4142 } else if ((packets > 35)) {
4143 itrval = lowest_latency;
4145 } else if (bytes/packets > 2000) {
4146 itrval = bulk_latency;
4147 } else if (packets <= 2 && bytes < 512) {
4148 itrval = lowest_latency;
4151 case bulk_latency: /* 250 usec aka 4000 ints/s */
4152 if (bytes > 25000) {
4154 itrval = low_latency;
4155 } else if (bytes < 1500) {
4156 itrval = low_latency;
4161 /* clear work counters since we have the values we need */
4162 ring_container->total_bytes = 0;
4163 ring_container->total_packets = 0;
4165 /* write updated itr to ring container */
4166 ring_container->itr = itrval;
4169 static void igb_set_itr(struct igb_q_vector *q_vector)
4171 struct igb_adapter *adapter = q_vector->adapter;
4172 u32 new_itr = q_vector->itr_val;
4175 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
4176 if (adapter->link_speed != SPEED_1000) {
4178 new_itr = IGB_4K_ITR;
4182 igb_update_itr(q_vector, &q_vector->tx);
4183 igb_update_itr(q_vector, &q_vector->rx);
4185 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
4187 /* conservative mode (itr 3) eliminates the lowest_latency setting */
4188 if (current_itr == lowest_latency &&
4189 ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
4190 (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
4191 current_itr = low_latency;
4193 switch (current_itr) {
4194 /* counts and packets in update_itr are dependent on these numbers */
4195 case lowest_latency:
4196 new_itr = IGB_70K_ITR; /* 70,000 ints/sec */
4199 new_itr = IGB_20K_ITR; /* 20,000 ints/sec */
4202 new_itr = IGB_4K_ITR; /* 4,000 ints/sec */
4209 if (new_itr != q_vector->itr_val) {
4210 /* this attempts to bias the interrupt rate towards Bulk
4211 * by adding intermediate steps when interrupt rate is
4213 new_itr = new_itr > q_vector->itr_val ?
4214 max((new_itr * q_vector->itr_val) /
4215 (new_itr + (q_vector->itr_val >> 2)),
4218 /* Don't write the value here; it resets the adapter's
4219 * internal timer, and causes us to delay far longer than
4220 * we should between interrupts. Instead, we write the ITR
4221 * value at the beginning of the next interrupt so the timing
4222 * ends up being correct.
4224 q_vector->itr_val = new_itr;
4225 q_vector->set_itr = 1;
4229 static void igb_tx_ctxtdesc(struct igb_ring *tx_ring, u32 vlan_macip_lens,
4230 u32 type_tucmd, u32 mss_l4len_idx)
4232 struct e1000_adv_tx_context_desc *context_desc;
4233 u16 i = tx_ring->next_to_use;
4235 context_desc = IGB_TX_CTXTDESC(tx_ring, i);
4238 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
4240 /* set bits to identify this as an advanced context descriptor */
4241 type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
4243 /* For 82575, context index must be unique per ring. */
4244 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
4245 mss_l4len_idx |= tx_ring->reg_idx << 4;
4247 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
4248 context_desc->seqnum_seed = 0;
4249 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd);
4250 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
4253 static int igb_tso(struct igb_ring *tx_ring,
4254 struct igb_tx_buffer *first,
4257 struct sk_buff *skb = first->skb;
4258 u32 vlan_macip_lens, type_tucmd;
4259 u32 mss_l4len_idx, l4len;
4261 if (skb->ip_summed != CHECKSUM_PARTIAL)
4264 if (!skb_is_gso(skb))
4267 if (skb_header_cloned(skb)) {
4268 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4273 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
4274 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
4276 if (first->protocol == __constant_htons(ETH_P_IP)) {
4277 struct iphdr *iph = ip_hdr(skb);
4280 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
4284 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
4285 first->tx_flags |= IGB_TX_FLAGS_TSO |
4288 } else if (skb_is_gso_v6(skb)) {
4289 ipv6_hdr(skb)->payload_len = 0;
4290 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4291 &ipv6_hdr(skb)->daddr,
4293 first->tx_flags |= IGB_TX_FLAGS_TSO |
4297 /* compute header lengths */
4298 l4len = tcp_hdrlen(skb);
4299 *hdr_len = skb_transport_offset(skb) + l4len;
4301 /* update gso size and bytecount with header size */
4302 first->gso_segs = skb_shinfo(skb)->gso_segs;
4303 first->bytecount += (first->gso_segs - 1) * *hdr_len;
4306 mss_l4len_idx = l4len << E1000_ADVTXD_L4LEN_SHIFT;
4307 mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT;
4309 /* VLAN MACLEN IPLEN */
4310 vlan_macip_lens = skb_network_header_len(skb);
4311 vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
4312 vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
4314 igb_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
4319 static void igb_tx_csum(struct igb_ring *tx_ring, struct igb_tx_buffer *first)
4321 struct sk_buff *skb = first->skb;
4322 u32 vlan_macip_lens = 0;
4323 u32 mss_l4len_idx = 0;
4326 if (skb->ip_summed != CHECKSUM_PARTIAL) {
4327 if (!(first->tx_flags & IGB_TX_FLAGS_VLAN))
4331 switch (first->protocol) {
4332 case __constant_htons(ETH_P_IP):
4333 vlan_macip_lens |= skb_network_header_len(skb);
4334 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
4335 l4_hdr = ip_hdr(skb)->protocol;
4337 case __constant_htons(ETH_P_IPV6):
4338 vlan_macip_lens |= skb_network_header_len(skb);
4339 l4_hdr = ipv6_hdr(skb)->nexthdr;
4342 if (unlikely(net_ratelimit())) {
4343 dev_warn(tx_ring->dev,
4344 "partial checksum but proto=%x!\n",
4352 type_tucmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
4353 mss_l4len_idx = tcp_hdrlen(skb) <<
4354 E1000_ADVTXD_L4LEN_SHIFT;
4357 type_tucmd |= E1000_ADVTXD_TUCMD_L4T_SCTP;
4358 mss_l4len_idx = sizeof(struct sctphdr) <<
4359 E1000_ADVTXD_L4LEN_SHIFT;
4362 mss_l4len_idx = sizeof(struct udphdr) <<
4363 E1000_ADVTXD_L4LEN_SHIFT;
4366 if (unlikely(net_ratelimit())) {
4367 dev_warn(tx_ring->dev,
4368 "partial checksum but l4 proto=%x!\n",
4374 /* update TX checksum flag */
4375 first->tx_flags |= IGB_TX_FLAGS_CSUM;
4378 vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
4379 vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
4381 igb_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
4384 #define IGB_SET_FLAG(_input, _flag, _result) \
4385 ((_flag <= _result) ? \
4386 ((u32)(_input & _flag) * (_result / _flag)) : \
4387 ((u32)(_input & _flag) / (_flag / _result)))
4389 static u32 igb_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
4391 /* set type for advanced descriptor with frame checksum insertion */
4392 u32 cmd_type = E1000_ADVTXD_DTYP_DATA |
4393 E1000_ADVTXD_DCMD_DEXT |
4394 E1000_ADVTXD_DCMD_IFCS;
4396 /* set HW vlan bit if vlan is present */
4397 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_VLAN,
4398 (E1000_ADVTXD_DCMD_VLE));
4400 /* set segmentation bits for TSO */
4401 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSO,
4402 (E1000_ADVTXD_DCMD_TSE));
4404 /* set timestamp bit if present */
4405 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSTAMP,
4406 (E1000_ADVTXD_MAC_TSTAMP));
4408 /* insert frame checksum */
4409 cmd_type ^= IGB_SET_FLAG(skb->no_fcs, 1, E1000_ADVTXD_DCMD_IFCS);
4414 static void igb_tx_olinfo_status(struct igb_ring *tx_ring,
4415 union e1000_adv_tx_desc *tx_desc,
4416 u32 tx_flags, unsigned int paylen)
4418 u32 olinfo_status = paylen << E1000_ADVTXD_PAYLEN_SHIFT;
4420 /* 82575 requires a unique index per ring */
4421 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
4422 olinfo_status |= tx_ring->reg_idx << 4;
4424 /* insert L4 checksum */
4425 olinfo_status |= IGB_SET_FLAG(tx_flags,
4427 (E1000_TXD_POPTS_TXSM << 8));
4429 /* insert IPv4 checksum */
4430 olinfo_status |= IGB_SET_FLAG(tx_flags,
4432 (E1000_TXD_POPTS_IXSM << 8));
4434 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
4437 static void igb_tx_map(struct igb_ring *tx_ring,
4438 struct igb_tx_buffer *first,
4441 struct sk_buff *skb = first->skb;
4442 struct igb_tx_buffer *tx_buffer;
4443 union e1000_adv_tx_desc *tx_desc;
4444 struct skb_frag_struct *frag;
4446 unsigned int data_len, size;
4447 u32 tx_flags = first->tx_flags;
4448 u32 cmd_type = igb_tx_cmd_type(skb, tx_flags);
4449 u16 i = tx_ring->next_to_use;
4451 tx_desc = IGB_TX_DESC(tx_ring, i);
4453 igb_tx_olinfo_status(tx_ring, tx_desc, tx_flags, skb->len - hdr_len);
4455 size = skb_headlen(skb);
4456 data_len = skb->data_len;
4458 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
4462 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
4463 if (dma_mapping_error(tx_ring->dev, dma))
4466 /* record length, and DMA address */
4467 dma_unmap_len_set(tx_buffer, len, size);
4468 dma_unmap_addr_set(tx_buffer, dma, dma);
4470 tx_desc->read.buffer_addr = cpu_to_le64(dma);
4472 while (unlikely(size > IGB_MAX_DATA_PER_TXD)) {
4473 tx_desc->read.cmd_type_len =
4474 cpu_to_le32(cmd_type ^ IGB_MAX_DATA_PER_TXD);
4478 if (i == tx_ring->count) {
4479 tx_desc = IGB_TX_DESC(tx_ring, 0);
4482 tx_desc->read.olinfo_status = 0;
4484 dma += IGB_MAX_DATA_PER_TXD;
4485 size -= IGB_MAX_DATA_PER_TXD;
4487 tx_desc->read.buffer_addr = cpu_to_le64(dma);
4490 if (likely(!data_len))
4493 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
4497 if (i == tx_ring->count) {
4498 tx_desc = IGB_TX_DESC(tx_ring, 0);
4501 tx_desc->read.olinfo_status = 0;
4503 size = skb_frag_size(frag);
4506 dma = skb_frag_dma_map(tx_ring->dev, frag, 0,
4507 size, DMA_TO_DEVICE);
4509 tx_buffer = &tx_ring->tx_buffer_info[i];
4512 /* write last descriptor with RS and EOP bits */
4513 cmd_type |= size | IGB_TXD_DCMD;
4514 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
4516 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
4518 /* set the timestamp */
4519 first->time_stamp = jiffies;
4522 * Force memory writes to complete before letting h/w know there
4523 * are new descriptors to fetch. (Only applicable for weak-ordered
4524 * memory model archs, such as IA-64).
4526 * We also need this memory barrier to make certain all of the
4527 * status bits have been updated before next_to_watch is written.
4531 /* set next_to_watch value indicating a packet is present */
4532 first->next_to_watch = tx_desc;
4535 if (i == tx_ring->count)
4538 tx_ring->next_to_use = i;
4540 writel(i, tx_ring->tail);
4542 /* we need this if more than one processor can write to our tail
4543 * at a time, it syncronizes IO on IA64/Altix systems */
4549 dev_err(tx_ring->dev, "TX DMA map failed\n");
4551 /* clear dma mappings for failed tx_buffer_info map */
4553 tx_buffer = &tx_ring->tx_buffer_info[i];
4554 igb_unmap_and_free_tx_resource(tx_ring, tx_buffer);
4555 if (tx_buffer == first)
4562 tx_ring->next_to_use = i;
4565 static int __igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
4567 struct net_device *netdev = tx_ring->netdev;
4569 netif_stop_subqueue(netdev, tx_ring->queue_index);
4571 /* Herbert's original patch had:
4572 * smp_mb__after_netif_stop_queue();
4573 * but since that doesn't exist yet, just open code it. */
4576 /* We need to check again in a case another CPU has just
4577 * made room available. */
4578 if (igb_desc_unused(tx_ring) < size)
4582 netif_wake_subqueue(netdev, tx_ring->queue_index);
4584 u64_stats_update_begin(&tx_ring->tx_syncp2);
4585 tx_ring->tx_stats.restart_queue2++;
4586 u64_stats_update_end(&tx_ring->tx_syncp2);
4591 static inline int igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
4593 if (igb_desc_unused(tx_ring) >= size)
4595 return __igb_maybe_stop_tx(tx_ring, size);
4598 netdev_tx_t igb_xmit_frame_ring(struct sk_buff *skb,
4599 struct igb_ring *tx_ring)
4601 struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
4602 struct igb_tx_buffer *first;
4605 u16 count = TXD_USE_COUNT(skb_headlen(skb));
4606 __be16 protocol = vlan_get_protocol(skb);
4609 /* need: 1 descriptor per page * PAGE_SIZE/IGB_MAX_DATA_PER_TXD,
4610 * + 1 desc for skb_headlen/IGB_MAX_DATA_PER_TXD,
4611 * + 2 desc gap to keep tail from touching head,
4612 * + 1 desc for context descriptor,
4613 * otherwise try next time
4615 if (NETDEV_FRAG_PAGE_MAX_SIZE > IGB_MAX_DATA_PER_TXD) {
4617 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
4618 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
4620 count += skb_shinfo(skb)->nr_frags;
4623 if (igb_maybe_stop_tx(tx_ring, count + 3)) {
4624 /* this is a hard error */
4625 return NETDEV_TX_BUSY;
4628 /* record the location of the first descriptor for this packet */
4629 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
4631 first->bytecount = skb->len;
4632 first->gso_segs = 1;
4634 skb_tx_timestamp(skb);
4636 if (unlikely((skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
4637 !(adapter->ptp_tx_skb))) {
4638 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
4639 tx_flags |= IGB_TX_FLAGS_TSTAMP;
4641 adapter->ptp_tx_skb = skb_get(skb);
4642 adapter->ptp_tx_start = jiffies;
4643 if (adapter->hw.mac.type == e1000_82576)
4644 schedule_work(&adapter->ptp_tx_work);
4647 if (vlan_tx_tag_present(skb)) {
4648 tx_flags |= IGB_TX_FLAGS_VLAN;
4649 tx_flags |= (vlan_tx_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
4652 /* record initial flags and protocol */
4653 first->tx_flags = tx_flags;
4654 first->protocol = protocol;
4656 tso = igb_tso(tx_ring, first, &hdr_len);
4660 igb_tx_csum(tx_ring, first);
4662 igb_tx_map(tx_ring, first, hdr_len);
4664 /* Make sure there is space in the ring for the next send. */
4665 igb_maybe_stop_tx(tx_ring, DESC_NEEDED);
4667 return NETDEV_TX_OK;
4670 igb_unmap_and_free_tx_resource(tx_ring, first);
4672 return NETDEV_TX_OK;
4675 static inline struct igb_ring *igb_tx_queue_mapping(struct igb_adapter *adapter,
4676 struct sk_buff *skb)
4678 unsigned int r_idx = skb->queue_mapping;
4680 if (r_idx >= adapter->num_tx_queues)
4681 r_idx = r_idx % adapter->num_tx_queues;
4683 return adapter->tx_ring[r_idx];
4686 static netdev_tx_t igb_xmit_frame(struct sk_buff *skb,
4687 struct net_device *netdev)
4689 struct igb_adapter *adapter = netdev_priv(netdev);
4691 if (test_bit(__IGB_DOWN, &adapter->state)) {
4692 dev_kfree_skb_any(skb);
4693 return NETDEV_TX_OK;
4696 if (skb->len <= 0) {
4697 dev_kfree_skb_any(skb);
4698 return NETDEV_TX_OK;
4702 * The minimum packet size with TCTL.PSP set is 17 so pad the skb
4703 * in order to meet this minimum size requirement.
4705 if (unlikely(skb->len < 17)) {
4706 if (skb_pad(skb, 17 - skb->len))
4707 return NETDEV_TX_OK;
4709 skb_set_tail_pointer(skb, 17);
4712 return igb_xmit_frame_ring(skb, igb_tx_queue_mapping(adapter, skb));
4716 * igb_tx_timeout - Respond to a Tx Hang
4717 * @netdev: network interface device structure
4719 static void igb_tx_timeout(struct net_device *netdev)
4721 struct igb_adapter *adapter = netdev_priv(netdev);
4722 struct e1000_hw *hw = &adapter->hw;
4724 /* Do the reset outside of interrupt context */
4725 adapter->tx_timeout_count++;
4727 if (hw->mac.type >= e1000_82580)
4728 hw->dev_spec._82575.global_device_reset = true;
4730 schedule_work(&adapter->reset_task);
4732 (adapter->eims_enable_mask & ~adapter->eims_other));
4735 static void igb_reset_task(struct work_struct *work)
4737 struct igb_adapter *adapter;
4738 adapter = container_of(work, struct igb_adapter, reset_task);
4741 netdev_err(adapter->netdev, "Reset adapter\n");
4742 igb_reinit_locked(adapter);
4746 * igb_get_stats64 - Get System Network Statistics
4747 * @netdev: network interface device structure
4748 * @stats: rtnl_link_stats64 pointer
4751 static struct rtnl_link_stats64 *igb_get_stats64(struct net_device *netdev,
4752 struct rtnl_link_stats64 *stats)
4754 struct igb_adapter *adapter = netdev_priv(netdev);
4756 spin_lock(&adapter->stats64_lock);
4757 igb_update_stats(adapter, &adapter->stats64);
4758 memcpy(stats, &adapter->stats64, sizeof(*stats));
4759 spin_unlock(&adapter->stats64_lock);
4765 * igb_change_mtu - Change the Maximum Transfer Unit
4766 * @netdev: network interface device structure
4767 * @new_mtu: new value for maximum frame size
4769 * Returns 0 on success, negative on failure
4771 static int igb_change_mtu(struct net_device *netdev, int new_mtu)
4773 struct igb_adapter *adapter = netdev_priv(netdev);
4774 struct pci_dev *pdev = adapter->pdev;
4775 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
4777 if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
4778 dev_err(&pdev->dev, "Invalid MTU setting\n");
4782 #define MAX_STD_JUMBO_FRAME_SIZE 9238
4783 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
4784 dev_err(&pdev->dev, "MTU > 9216 not supported.\n");
4788 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
4791 /* igb_down has a dependency on max_frame_size */
4792 adapter->max_frame_size = max_frame;
4794 if (netif_running(netdev))
4797 dev_info(&pdev->dev, "changing MTU from %d to %d\n",
4798 netdev->mtu, new_mtu);
4799 netdev->mtu = new_mtu;
4801 if (netif_running(netdev))
4806 clear_bit(__IGB_RESETTING, &adapter->state);
4812 * igb_update_stats - Update the board statistics counters
4813 * @adapter: board private structure
4816 void igb_update_stats(struct igb_adapter *adapter,
4817 struct rtnl_link_stats64 *net_stats)
4819 struct e1000_hw *hw = &adapter->hw;
4820 struct pci_dev *pdev = adapter->pdev;
4826 u64 _bytes, _packets;
4828 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
4831 * Prevent stats update while adapter is being reset, or if the pci
4832 * connection is down.
4834 if (adapter->link_speed == 0)
4836 if (pci_channel_offline(pdev))
4841 for (i = 0; i < adapter->num_rx_queues; i++) {
4842 u32 rqdpc = rd32(E1000_RQDPC(i));
4843 struct igb_ring *ring = adapter->rx_ring[i];
4846 ring->rx_stats.drops += rqdpc;
4847 net_stats->rx_fifo_errors += rqdpc;
4851 start = u64_stats_fetch_begin_bh(&ring->rx_syncp);
4852 _bytes = ring->rx_stats.bytes;
4853 _packets = ring->rx_stats.packets;
4854 } while (u64_stats_fetch_retry_bh(&ring->rx_syncp, start));
4856 packets += _packets;
4859 net_stats->rx_bytes = bytes;
4860 net_stats->rx_packets = packets;
4864 for (i = 0; i < adapter->num_tx_queues; i++) {
4865 struct igb_ring *ring = adapter->tx_ring[i];
4867 start = u64_stats_fetch_begin_bh(&ring->tx_syncp);
4868 _bytes = ring->tx_stats.bytes;
4869 _packets = ring->tx_stats.packets;
4870 } while (u64_stats_fetch_retry_bh(&ring->tx_syncp, start));
4872 packets += _packets;
4874 net_stats->tx_bytes = bytes;
4875 net_stats->tx_packets = packets;
4877 /* read stats registers */
4878 adapter->stats.crcerrs += rd32(E1000_CRCERRS);
4879 adapter->stats.gprc += rd32(E1000_GPRC);
4880 adapter->stats.gorc += rd32(E1000_GORCL);
4881 rd32(E1000_GORCH); /* clear GORCL */
4882 adapter->stats.bprc += rd32(E1000_BPRC);
4883 adapter->stats.mprc += rd32(E1000_MPRC);
4884 adapter->stats.roc += rd32(E1000_ROC);
4886 adapter->stats.prc64 += rd32(E1000_PRC64);
4887 adapter->stats.prc127 += rd32(E1000_PRC127);
4888 adapter->stats.prc255 += rd32(E1000_PRC255);
4889 adapter->stats.prc511 += rd32(E1000_PRC511);
4890 adapter->stats.prc1023 += rd32(E1000_PRC1023);
4891 adapter->stats.prc1522 += rd32(E1000_PRC1522);
4892 adapter->stats.symerrs += rd32(E1000_SYMERRS);
4893 adapter->stats.sec += rd32(E1000_SEC);
4895 mpc = rd32(E1000_MPC);
4896 adapter->stats.mpc += mpc;
4897 net_stats->rx_fifo_errors += mpc;
4898 adapter->stats.scc += rd32(E1000_SCC);
4899 adapter->stats.ecol += rd32(E1000_ECOL);
4900 adapter->stats.mcc += rd32(E1000_MCC);
4901 adapter->stats.latecol += rd32(E1000_LATECOL);
4902 adapter->stats.dc += rd32(E1000_DC);
4903 adapter->stats.rlec += rd32(E1000_RLEC);
4904 adapter->stats.xonrxc += rd32(E1000_XONRXC);
4905 adapter->stats.xontxc += rd32(E1000_XONTXC);
4906 adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
4907 adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
4908 adapter->stats.fcruc += rd32(E1000_FCRUC);
4909 adapter->stats.gptc += rd32(E1000_GPTC);
4910 adapter->stats.gotc += rd32(E1000_GOTCL);
4911 rd32(E1000_GOTCH); /* clear GOTCL */
4912 adapter->stats.rnbc += rd32(E1000_RNBC);
4913 adapter->stats.ruc += rd32(E1000_RUC);
4914 adapter->stats.rfc += rd32(E1000_RFC);
4915 adapter->stats.rjc += rd32(E1000_RJC);
4916 adapter->stats.tor += rd32(E1000_TORH);
4917 adapter->stats.tot += rd32(E1000_TOTH);
4918 adapter->stats.tpr += rd32(E1000_TPR);
4920 adapter->stats.ptc64 += rd32(E1000_PTC64);
4921 adapter->stats.ptc127 += rd32(E1000_PTC127);
4922 adapter->stats.ptc255 += rd32(E1000_PTC255);
4923 adapter->stats.ptc511 += rd32(E1000_PTC511);
4924 adapter->stats.ptc1023 += rd32(E1000_PTC1023);
4925 adapter->stats.ptc1522 += rd32(E1000_PTC1522);
4927 adapter->stats.mptc += rd32(E1000_MPTC);
4928 adapter->stats.bptc += rd32(E1000_BPTC);
4930 adapter->stats.tpt += rd32(E1000_TPT);
4931 adapter->stats.colc += rd32(E1000_COLC);
4933 adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
4934 /* read internal phy specific stats */
4935 reg = rd32(E1000_CTRL_EXT);
4936 if (!(reg & E1000_CTRL_EXT_LINK_MODE_MASK)) {
4937 adapter->stats.rxerrc += rd32(E1000_RXERRC);
4939 /* this stat has invalid values on i210/i211 */
4940 if ((hw->mac.type != e1000_i210) &&
4941 (hw->mac.type != e1000_i211))
4942 adapter->stats.tncrs += rd32(E1000_TNCRS);
4945 adapter->stats.tsctc += rd32(E1000_TSCTC);
4946 adapter->stats.tsctfc += rd32(E1000_TSCTFC);
4948 adapter->stats.iac += rd32(E1000_IAC);
4949 adapter->stats.icrxoc += rd32(E1000_ICRXOC);
4950 adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
4951 adapter->stats.icrxatc += rd32(E1000_ICRXATC);
4952 adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
4953 adapter->stats.ictxatc += rd32(E1000_ICTXATC);
4954 adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
4955 adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
4956 adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
4958 /* Fill out the OS statistics structure */
4959 net_stats->multicast = adapter->stats.mprc;
4960 net_stats->collisions = adapter->stats.colc;
4964 /* RLEC on some newer hardware can be incorrect so build
4965 * our own version based on RUC and ROC */
4966 net_stats->rx_errors = adapter->stats.rxerrc +
4967 adapter->stats.crcerrs + adapter->stats.algnerrc +
4968 adapter->stats.ruc + adapter->stats.roc +
4969 adapter->stats.cexterr;
4970 net_stats->rx_length_errors = adapter->stats.ruc +
4972 net_stats->rx_crc_errors = adapter->stats.crcerrs;
4973 net_stats->rx_frame_errors = adapter->stats.algnerrc;
4974 net_stats->rx_missed_errors = adapter->stats.mpc;
4977 net_stats->tx_errors = adapter->stats.ecol +
4978 adapter->stats.latecol;
4979 net_stats->tx_aborted_errors = adapter->stats.ecol;
4980 net_stats->tx_window_errors = adapter->stats.latecol;
4981 net_stats->tx_carrier_errors = adapter->stats.tncrs;
4983 /* Tx Dropped needs to be maintained elsewhere */
4986 if (hw->phy.media_type == e1000_media_type_copper) {
4987 if ((adapter->link_speed == SPEED_1000) &&
4988 (!igb_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
4989 phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
4990 adapter->phy_stats.idle_errors += phy_tmp;
4994 /* Management Stats */
4995 adapter->stats.mgptc += rd32(E1000_MGTPTC);
4996 adapter->stats.mgprc += rd32(E1000_MGTPRC);
4997 adapter->stats.mgpdc += rd32(E1000_MGTPDC);
5000 reg = rd32(E1000_MANC);
5001 if (reg & E1000_MANC_EN_BMC2OS) {
5002 adapter->stats.o2bgptc += rd32(E1000_O2BGPTC);
5003 adapter->stats.o2bspc += rd32(E1000_O2BSPC);
5004 adapter->stats.b2ospc += rd32(E1000_B2OSPC);
5005 adapter->stats.b2ogprc += rd32(E1000_B2OGPRC);
5009 static irqreturn_t igb_msix_other(int irq, void *data)
5011 struct igb_adapter *adapter = data;
5012 struct e1000_hw *hw = &adapter->hw;
5013 u32 icr = rd32(E1000_ICR);
5014 /* reading ICR causes bit 31 of EICR to be cleared */
5016 if (icr & E1000_ICR_DRSTA)
5017 schedule_work(&adapter->reset_task);
5019 if (icr & E1000_ICR_DOUTSYNC) {
5020 /* HW is reporting DMA is out of sync */
5021 adapter->stats.doosync++;
5022 /* The DMA Out of Sync is also indication of a spoof event
5023 * in IOV mode. Check the Wrong VM Behavior register to
5024 * see if it is really a spoof event. */
5025 igb_check_wvbr(adapter);
5028 /* Check for a mailbox event */
5029 if (icr & E1000_ICR_VMMB)
5030 igb_msg_task(adapter);
5032 if (icr & E1000_ICR_LSC) {
5033 hw->mac.get_link_status = 1;
5034 /* guard against interrupt when we're going down */
5035 if (!test_bit(__IGB_DOWN, &adapter->state))
5036 mod_timer(&adapter->watchdog_timer, jiffies + 1);
5039 if (icr & E1000_ICR_TS) {
5040 u32 tsicr = rd32(E1000_TSICR);
5042 if (tsicr & E1000_TSICR_TXTS) {
5043 /* acknowledge the interrupt */
5044 wr32(E1000_TSICR, E1000_TSICR_TXTS);
5045 /* retrieve hardware timestamp */
5046 schedule_work(&adapter->ptp_tx_work);
5050 wr32(E1000_EIMS, adapter->eims_other);
5055 static void igb_write_itr(struct igb_q_vector *q_vector)
5057 struct igb_adapter *adapter = q_vector->adapter;
5058 u32 itr_val = q_vector->itr_val & 0x7FFC;
5060 if (!q_vector->set_itr)
5066 if (adapter->hw.mac.type == e1000_82575)
5067 itr_val |= itr_val << 16;
5069 itr_val |= E1000_EITR_CNT_IGNR;
5071 writel(itr_val, q_vector->itr_register);
5072 q_vector->set_itr = 0;
5075 static irqreturn_t igb_msix_ring(int irq, void *data)
5077 struct igb_q_vector *q_vector = data;
5079 /* Write the ITR value calculated from the previous interrupt. */
5080 igb_write_itr(q_vector);
5082 napi_schedule(&q_vector->napi);
5087 #ifdef CONFIG_IGB_DCA
5088 static void igb_update_tx_dca(struct igb_adapter *adapter,
5089 struct igb_ring *tx_ring,
5092 struct e1000_hw *hw = &adapter->hw;
5093 u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
5095 if (hw->mac.type != e1000_82575)
5096 txctrl <<= E1000_DCA_TXCTRL_CPUID_SHIFT;
5099 * We can enable relaxed ordering for reads, but not writes when
5100 * DCA is enabled. This is due to a known issue in some chipsets
5101 * which will cause the DCA tag to be cleared.
5103 txctrl |= E1000_DCA_TXCTRL_DESC_RRO_EN |
5104 E1000_DCA_TXCTRL_DATA_RRO_EN |
5105 E1000_DCA_TXCTRL_DESC_DCA_EN;
5107 wr32(E1000_DCA_TXCTRL(tx_ring->reg_idx), txctrl);
5110 static void igb_update_rx_dca(struct igb_adapter *adapter,
5111 struct igb_ring *rx_ring,
5114 struct e1000_hw *hw = &adapter->hw;
5115 u32 rxctrl = dca3_get_tag(&adapter->pdev->dev, cpu);
5117 if (hw->mac.type != e1000_82575)
5118 rxctrl <<= E1000_DCA_RXCTRL_CPUID_SHIFT;
5121 * We can enable relaxed ordering for reads, but not writes when
5122 * DCA is enabled. This is due to a known issue in some chipsets
5123 * which will cause the DCA tag to be cleared.
5125 rxctrl |= E1000_DCA_RXCTRL_DESC_RRO_EN |
5126 E1000_DCA_RXCTRL_DESC_DCA_EN;
5128 wr32(E1000_DCA_RXCTRL(rx_ring->reg_idx), rxctrl);
5131 static void igb_update_dca(struct igb_q_vector *q_vector)
5133 struct igb_adapter *adapter = q_vector->adapter;
5134 int cpu = get_cpu();
5136 if (q_vector->cpu == cpu)
5139 if (q_vector->tx.ring)
5140 igb_update_tx_dca(adapter, q_vector->tx.ring, cpu);
5142 if (q_vector->rx.ring)
5143 igb_update_rx_dca(adapter, q_vector->rx.ring, cpu);
5145 q_vector->cpu = cpu;
5150 static void igb_setup_dca(struct igb_adapter *adapter)
5152 struct e1000_hw *hw = &adapter->hw;
5155 if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
5158 /* Always use CB2 mode, difference is masked in the CB driver. */
5159 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
5161 for (i = 0; i < adapter->num_q_vectors; i++) {
5162 adapter->q_vector[i]->cpu = -1;
5163 igb_update_dca(adapter->q_vector[i]);
5167 static int __igb_notify_dca(struct device *dev, void *data)
5169 struct net_device *netdev = dev_get_drvdata(dev);
5170 struct igb_adapter *adapter = netdev_priv(netdev);
5171 struct pci_dev *pdev = adapter->pdev;
5172 struct e1000_hw *hw = &adapter->hw;
5173 unsigned long event = *(unsigned long *)data;
5176 case DCA_PROVIDER_ADD:
5177 /* if already enabled, don't do it again */
5178 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
5180 if (dca_add_requester(dev) == 0) {
5181 adapter->flags |= IGB_FLAG_DCA_ENABLED;
5182 dev_info(&pdev->dev, "DCA enabled\n");
5183 igb_setup_dca(adapter);
5186 /* Fall Through since DCA is disabled. */
5187 case DCA_PROVIDER_REMOVE:
5188 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
5189 /* without this a class_device is left
5190 * hanging around in the sysfs model */
5191 dca_remove_requester(dev);
5192 dev_info(&pdev->dev, "DCA disabled\n");
5193 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
5194 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
5202 static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
5207 ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
5210 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
5212 #endif /* CONFIG_IGB_DCA */
5214 #ifdef CONFIG_PCI_IOV
5215 static int igb_vf_configure(struct igb_adapter *adapter, int vf)
5217 unsigned char mac_addr[ETH_ALEN];
5219 eth_zero_addr(mac_addr);
5220 igb_set_vf_mac(adapter, vf, mac_addr);
5225 static bool igb_vfs_are_assigned(struct igb_adapter *adapter)
5227 struct pci_dev *pdev = adapter->pdev;
5228 struct pci_dev *vfdev;
5231 switch (adapter->hw.mac.type) {
5233 dev_id = IGB_82576_VF_DEV_ID;
5236 dev_id = IGB_I350_VF_DEV_ID;
5242 /* loop through all the VFs to see if we own any that are assigned */
5243 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, dev_id, NULL);
5245 /* if we don't own it we don't care */
5246 if (vfdev->is_virtfn && vfdev->physfn == pdev) {
5247 /* if it is assigned we cannot release it */
5248 if (vfdev->dev_flags & PCI_DEV_FLAGS_ASSIGNED)
5252 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, dev_id, vfdev);
5259 static void igb_ping_all_vfs(struct igb_adapter *adapter)
5261 struct e1000_hw *hw = &adapter->hw;
5265 for (i = 0 ; i < adapter->vfs_allocated_count; i++) {
5266 ping = E1000_PF_CONTROL_MSG;
5267 if (adapter->vf_data[i].flags & IGB_VF_FLAG_CTS)
5268 ping |= E1000_VT_MSGTYPE_CTS;
5269 igb_write_mbx(hw, &ping, 1, i);
5273 static int igb_set_vf_promisc(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
5275 struct e1000_hw *hw = &adapter->hw;
5276 u32 vmolr = rd32(E1000_VMOLR(vf));
5277 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5279 vf_data->flags &= ~(IGB_VF_FLAG_UNI_PROMISC |
5280 IGB_VF_FLAG_MULTI_PROMISC);
5281 vmolr &= ~(E1000_VMOLR_ROPE | E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
5283 if (*msgbuf & E1000_VF_SET_PROMISC_MULTICAST) {
5284 vmolr |= E1000_VMOLR_MPME;
5285 vf_data->flags |= IGB_VF_FLAG_MULTI_PROMISC;
5286 *msgbuf &= ~E1000_VF_SET_PROMISC_MULTICAST;
5289 * if we have hashes and we are clearing a multicast promisc
5290 * flag we need to write the hashes to the MTA as this step
5291 * was previously skipped
5293 if (vf_data->num_vf_mc_hashes > 30) {
5294 vmolr |= E1000_VMOLR_MPME;
5295 } else if (vf_data->num_vf_mc_hashes) {
5297 vmolr |= E1000_VMOLR_ROMPE;
5298 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
5299 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
5303 wr32(E1000_VMOLR(vf), vmolr);
5305 /* there are flags left unprocessed, likely not supported */
5306 if (*msgbuf & E1000_VT_MSGINFO_MASK)
5313 static int igb_set_vf_multicasts(struct igb_adapter *adapter,
5314 u32 *msgbuf, u32 vf)
5316 int n = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
5317 u16 *hash_list = (u16 *)&msgbuf[1];
5318 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5321 /* salt away the number of multicast addresses assigned
5322 * to this VF for later use to restore when the PF multi cast
5325 vf_data->num_vf_mc_hashes = n;
5327 /* only up to 30 hash values supported */
5331 /* store the hashes for later use */
5332 for (i = 0; i < n; i++)
5333 vf_data->vf_mc_hashes[i] = hash_list[i];
5335 /* Flush and reset the mta with the new values */
5336 igb_set_rx_mode(adapter->netdev);
5341 static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
5343 struct e1000_hw *hw = &adapter->hw;
5344 struct vf_data_storage *vf_data;
5347 for (i = 0; i < adapter->vfs_allocated_count; i++) {
5348 u32 vmolr = rd32(E1000_VMOLR(i));
5349 vmolr &= ~(E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
5351 vf_data = &adapter->vf_data[i];
5353 if ((vf_data->num_vf_mc_hashes > 30) ||
5354 (vf_data->flags & IGB_VF_FLAG_MULTI_PROMISC)) {
5355 vmolr |= E1000_VMOLR_MPME;
5356 } else if (vf_data->num_vf_mc_hashes) {
5357 vmolr |= E1000_VMOLR_ROMPE;
5358 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
5359 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
5361 wr32(E1000_VMOLR(i), vmolr);
5365 static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
5367 struct e1000_hw *hw = &adapter->hw;
5368 u32 pool_mask, reg, vid;
5371 pool_mask = 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
5373 /* Find the vlan filter for this id */
5374 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5375 reg = rd32(E1000_VLVF(i));
5377 /* remove the vf from the pool */
5380 /* if pool is empty then remove entry from vfta */
5381 if (!(reg & E1000_VLVF_POOLSEL_MASK) &&
5382 (reg & E1000_VLVF_VLANID_ENABLE)) {
5384 vid = reg & E1000_VLVF_VLANID_MASK;
5385 igb_vfta_set(hw, vid, false);
5388 wr32(E1000_VLVF(i), reg);
5391 adapter->vf_data[vf].vlans_enabled = 0;
5394 static s32 igb_vlvf_set(struct igb_adapter *adapter, u32 vid, bool add, u32 vf)
5396 struct e1000_hw *hw = &adapter->hw;
5399 /* The vlvf table only exists on 82576 hardware and newer */
5400 if (hw->mac.type < e1000_82576)
5403 /* we only need to do this if VMDq is enabled */
5404 if (!adapter->vfs_allocated_count)
5407 /* Find the vlan filter for this id */
5408 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5409 reg = rd32(E1000_VLVF(i));
5410 if ((reg & E1000_VLVF_VLANID_ENABLE) &&
5411 vid == (reg & E1000_VLVF_VLANID_MASK))
5416 if (i == E1000_VLVF_ARRAY_SIZE) {
5417 /* Did not find a matching VLAN ID entry that was
5418 * enabled. Search for a free filter entry, i.e.
5419 * one without the enable bit set
5421 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5422 reg = rd32(E1000_VLVF(i));
5423 if (!(reg & E1000_VLVF_VLANID_ENABLE))
5427 if (i < E1000_VLVF_ARRAY_SIZE) {
5428 /* Found an enabled/available entry */
5429 reg |= 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
5431 /* if !enabled we need to set this up in vfta */
5432 if (!(reg & E1000_VLVF_VLANID_ENABLE)) {
5433 /* add VID to filter table */
5434 igb_vfta_set(hw, vid, true);
5435 reg |= E1000_VLVF_VLANID_ENABLE;
5437 reg &= ~E1000_VLVF_VLANID_MASK;
5439 wr32(E1000_VLVF(i), reg);
5441 /* do not modify RLPML for PF devices */
5442 if (vf >= adapter->vfs_allocated_count)
5445 if (!adapter->vf_data[vf].vlans_enabled) {
5447 reg = rd32(E1000_VMOLR(vf));
5448 size = reg & E1000_VMOLR_RLPML_MASK;
5450 reg &= ~E1000_VMOLR_RLPML_MASK;
5452 wr32(E1000_VMOLR(vf), reg);
5455 adapter->vf_data[vf].vlans_enabled++;
5458 if (i < E1000_VLVF_ARRAY_SIZE) {
5459 /* remove vf from the pool */
5460 reg &= ~(1 << (E1000_VLVF_POOLSEL_SHIFT + vf));
5461 /* if pool is empty then remove entry from vfta */
5462 if (!(reg & E1000_VLVF_POOLSEL_MASK)) {
5464 igb_vfta_set(hw, vid, false);
5466 wr32(E1000_VLVF(i), reg);
5468 /* do not modify RLPML for PF devices */
5469 if (vf >= adapter->vfs_allocated_count)
5472 adapter->vf_data[vf].vlans_enabled--;
5473 if (!adapter->vf_data[vf].vlans_enabled) {
5475 reg = rd32(E1000_VMOLR(vf));
5476 size = reg & E1000_VMOLR_RLPML_MASK;
5478 reg &= ~E1000_VMOLR_RLPML_MASK;
5480 wr32(E1000_VMOLR(vf), reg);
5487 static void igb_set_vmvir(struct igb_adapter *adapter, u32 vid, u32 vf)
5489 struct e1000_hw *hw = &adapter->hw;
5492 wr32(E1000_VMVIR(vf), (vid | E1000_VMVIR_VLANA_DEFAULT));
5494 wr32(E1000_VMVIR(vf), 0);
5497 static int igb_ndo_set_vf_vlan(struct net_device *netdev,
5498 int vf, u16 vlan, u8 qos)
5501 struct igb_adapter *adapter = netdev_priv(netdev);
5503 if ((vf >= adapter->vfs_allocated_count) || (vlan > 4095) || (qos > 7))
5506 err = igb_vlvf_set(adapter, vlan, !!vlan, vf);
5509 igb_set_vmvir(adapter, vlan | (qos << VLAN_PRIO_SHIFT), vf);
5510 igb_set_vmolr(adapter, vf, !vlan);
5511 adapter->vf_data[vf].pf_vlan = vlan;
5512 adapter->vf_data[vf].pf_qos = qos;
5513 dev_info(&adapter->pdev->dev,
5514 "Setting VLAN %d, QOS 0x%x on VF %d\n", vlan, qos, vf);
5515 if (test_bit(__IGB_DOWN, &adapter->state)) {
5516 dev_warn(&adapter->pdev->dev,
5517 "The VF VLAN has been set,"
5518 " but the PF device is not up.\n");
5519 dev_warn(&adapter->pdev->dev,
5520 "Bring the PF device up before"
5521 " attempting to use the VF device.\n");
5524 igb_vlvf_set(adapter, adapter->vf_data[vf].pf_vlan,
5526 igb_set_vmvir(adapter, vlan, vf);
5527 igb_set_vmolr(adapter, vf, true);
5528 adapter->vf_data[vf].pf_vlan = 0;
5529 adapter->vf_data[vf].pf_qos = 0;
5535 static int igb_set_vf_vlan(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
5537 int add = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
5538 int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
5540 return igb_vlvf_set(adapter, vid, add, vf);
5543 static inline void igb_vf_reset(struct igb_adapter *adapter, u32 vf)
5545 /* clear flags - except flag that indicates PF has set the MAC */
5546 adapter->vf_data[vf].flags &= IGB_VF_FLAG_PF_SET_MAC;
5547 adapter->vf_data[vf].last_nack = jiffies;
5549 /* reset offloads to defaults */
5550 igb_set_vmolr(adapter, vf, true);
5552 /* reset vlans for device */
5553 igb_clear_vf_vfta(adapter, vf);
5554 if (adapter->vf_data[vf].pf_vlan)
5555 igb_ndo_set_vf_vlan(adapter->netdev, vf,
5556 adapter->vf_data[vf].pf_vlan,
5557 adapter->vf_data[vf].pf_qos);
5559 igb_clear_vf_vfta(adapter, vf);
5561 /* reset multicast table array for vf */
5562 adapter->vf_data[vf].num_vf_mc_hashes = 0;
5564 /* Flush and reset the mta with the new values */
5565 igb_set_rx_mode(adapter->netdev);
5568 static void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
5570 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
5572 /* clear mac address as we were hotplug removed/added */
5573 if (!(adapter->vf_data[vf].flags & IGB_VF_FLAG_PF_SET_MAC))
5574 eth_zero_addr(vf_mac);
5576 /* process remaining reset events */
5577 igb_vf_reset(adapter, vf);
5580 static void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
5582 struct e1000_hw *hw = &adapter->hw;
5583 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
5584 int rar_entry = hw->mac.rar_entry_count - (vf + 1);
5586 u8 *addr = (u8 *)(&msgbuf[1]);
5588 /* process all the same items cleared in a function level reset */
5589 igb_vf_reset(adapter, vf);
5591 /* set vf mac address */
5592 igb_rar_set_qsel(adapter, vf_mac, rar_entry, vf);
5594 /* enable transmit and receive for vf */
5595 reg = rd32(E1000_VFTE);
5596 wr32(E1000_VFTE, reg | (1 << vf));
5597 reg = rd32(E1000_VFRE);
5598 wr32(E1000_VFRE, reg | (1 << vf));
5600 adapter->vf_data[vf].flags |= IGB_VF_FLAG_CTS;
5602 /* reply to reset with ack and vf mac address */
5603 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
5604 memcpy(addr, vf_mac, 6);
5605 igb_write_mbx(hw, msgbuf, 3, vf);
5608 static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf)
5611 * The VF MAC Address is stored in a packed array of bytes
5612 * starting at the second 32 bit word of the msg array
5614 unsigned char *addr = (char *)&msg[1];
5617 if (is_valid_ether_addr(addr))
5618 err = igb_set_vf_mac(adapter, vf, addr);
5623 static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf)
5625 struct e1000_hw *hw = &adapter->hw;
5626 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5627 u32 msg = E1000_VT_MSGTYPE_NACK;
5629 /* if device isn't clear to send it shouldn't be reading either */
5630 if (!(vf_data->flags & IGB_VF_FLAG_CTS) &&
5631 time_after(jiffies, vf_data->last_nack + (2 * HZ))) {
5632 igb_write_mbx(hw, &msg, 1, vf);
5633 vf_data->last_nack = jiffies;
5637 static void igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
5639 struct pci_dev *pdev = adapter->pdev;
5640 u32 msgbuf[E1000_VFMAILBOX_SIZE];
5641 struct e1000_hw *hw = &adapter->hw;
5642 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5645 retval = igb_read_mbx(hw, msgbuf, E1000_VFMAILBOX_SIZE, vf);
5648 /* if receive failed revoke VF CTS stats and restart init */
5649 dev_err(&pdev->dev, "Error receiving message from VF\n");
5650 vf_data->flags &= ~IGB_VF_FLAG_CTS;
5651 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
5656 /* this is a message we already processed, do nothing */
5657 if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
5661 * until the vf completes a reset it should not be
5662 * allowed to start any configuration.
5665 if (msgbuf[0] == E1000_VF_RESET) {
5666 igb_vf_reset_msg(adapter, vf);
5670 if (!(vf_data->flags & IGB_VF_FLAG_CTS)) {
5671 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
5677 switch ((msgbuf[0] & 0xFFFF)) {
5678 case E1000_VF_SET_MAC_ADDR:
5680 if (!(vf_data->flags & IGB_VF_FLAG_PF_SET_MAC))
5681 retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
5683 dev_warn(&pdev->dev,
5684 "VF %d attempted to override administratively "
5685 "set MAC address\nReload the VF driver to "
5686 "resume operations\n", vf);
5688 case E1000_VF_SET_PROMISC:
5689 retval = igb_set_vf_promisc(adapter, msgbuf, vf);
5691 case E1000_VF_SET_MULTICAST:
5692 retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
5694 case E1000_VF_SET_LPE:
5695 retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf);
5697 case E1000_VF_SET_VLAN:
5699 if (vf_data->pf_vlan)
5700 dev_warn(&pdev->dev,
5701 "VF %d attempted to override administratively "
5702 "set VLAN tag\nReload the VF driver to "
5703 "resume operations\n", vf);
5705 retval = igb_set_vf_vlan(adapter, msgbuf, vf);
5708 dev_err(&pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
5713 msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
5715 /* notify the VF of the results of what it sent us */
5717 msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
5719 msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
5721 igb_write_mbx(hw, msgbuf, 1, vf);
5724 static void igb_msg_task(struct igb_adapter *adapter)
5726 struct e1000_hw *hw = &adapter->hw;
5729 for (vf = 0; vf < adapter->vfs_allocated_count; vf++) {
5730 /* process any reset requests */
5731 if (!igb_check_for_rst(hw, vf))
5732 igb_vf_reset_event(adapter, vf);
5734 /* process any messages pending */
5735 if (!igb_check_for_msg(hw, vf))
5736 igb_rcv_msg_from_vf(adapter, vf);
5738 /* process any acks */
5739 if (!igb_check_for_ack(hw, vf))
5740 igb_rcv_ack_from_vf(adapter, vf);
5745 * igb_set_uta - Set unicast filter table address
5746 * @adapter: board private structure
5748 * The unicast table address is a register array of 32-bit registers.
5749 * The table is meant to be used in a way similar to how the MTA is used
5750 * however due to certain limitations in the hardware it is necessary to
5751 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
5752 * enable bit to allow vlan tag stripping when promiscuous mode is enabled
5754 static void igb_set_uta(struct igb_adapter *adapter)
5756 struct e1000_hw *hw = &adapter->hw;
5759 /* The UTA table only exists on 82576 hardware and newer */
5760 if (hw->mac.type < e1000_82576)
5763 /* we only need to do this if VMDq is enabled */
5764 if (!adapter->vfs_allocated_count)
5767 for (i = 0; i < hw->mac.uta_reg_count; i++)
5768 array_wr32(E1000_UTA, i, ~0);
5772 * igb_intr_msi - Interrupt Handler
5773 * @irq: interrupt number
5774 * @data: pointer to a network interface device structure
5776 static irqreturn_t igb_intr_msi(int irq, void *data)
5778 struct igb_adapter *adapter = data;
5779 struct igb_q_vector *q_vector = adapter->q_vector[0];
5780 struct e1000_hw *hw = &adapter->hw;
5781 /* read ICR disables interrupts using IAM */
5782 u32 icr = rd32(E1000_ICR);
5784 igb_write_itr(q_vector);
5786 if (icr & E1000_ICR_DRSTA)
5787 schedule_work(&adapter->reset_task);
5789 if (icr & E1000_ICR_DOUTSYNC) {
5790 /* HW is reporting DMA is out of sync */
5791 adapter->stats.doosync++;
5794 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
5795 hw->mac.get_link_status = 1;
5796 if (!test_bit(__IGB_DOWN, &adapter->state))
5797 mod_timer(&adapter->watchdog_timer, jiffies + 1);
5800 if (icr & E1000_ICR_TS) {
5801 u32 tsicr = rd32(E1000_TSICR);
5803 if (tsicr & E1000_TSICR_TXTS) {
5804 /* acknowledge the interrupt */
5805 wr32(E1000_TSICR, E1000_TSICR_TXTS);
5806 /* retrieve hardware timestamp */
5807 schedule_work(&adapter->ptp_tx_work);
5811 napi_schedule(&q_vector->napi);
5817 * igb_intr - Legacy Interrupt Handler
5818 * @irq: interrupt number
5819 * @data: pointer to a network interface device structure
5821 static irqreturn_t igb_intr(int irq, void *data)
5823 struct igb_adapter *adapter = data;
5824 struct igb_q_vector *q_vector = adapter->q_vector[0];
5825 struct e1000_hw *hw = &adapter->hw;
5826 /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked. No
5827 * need for the IMC write */
5828 u32 icr = rd32(E1000_ICR);
5830 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
5831 * not set, then the adapter didn't send an interrupt */
5832 if (!(icr & E1000_ICR_INT_ASSERTED))
5835 igb_write_itr(q_vector);
5837 if (icr & E1000_ICR_DRSTA)
5838 schedule_work(&adapter->reset_task);
5840 if (icr & E1000_ICR_DOUTSYNC) {
5841 /* HW is reporting DMA is out of sync */
5842 adapter->stats.doosync++;
5845 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
5846 hw->mac.get_link_status = 1;
5847 /* guard against interrupt when we're going down */
5848 if (!test_bit(__IGB_DOWN, &adapter->state))
5849 mod_timer(&adapter->watchdog_timer, jiffies + 1);
5852 if (icr & E1000_ICR_TS) {
5853 u32 tsicr = rd32(E1000_TSICR);
5855 if (tsicr & E1000_TSICR_TXTS) {
5856 /* acknowledge the interrupt */
5857 wr32(E1000_TSICR, E1000_TSICR_TXTS);
5858 /* retrieve hardware timestamp */
5859 schedule_work(&adapter->ptp_tx_work);
5863 napi_schedule(&q_vector->napi);
5868 static void igb_ring_irq_enable(struct igb_q_vector *q_vector)
5870 struct igb_adapter *adapter = q_vector->adapter;
5871 struct e1000_hw *hw = &adapter->hw;
5873 if ((q_vector->rx.ring && (adapter->rx_itr_setting & 3)) ||
5874 (!q_vector->rx.ring && (adapter->tx_itr_setting & 3))) {
5875 if ((adapter->num_q_vectors == 1) && !adapter->vf_data)
5876 igb_set_itr(q_vector);
5878 igb_update_ring_itr(q_vector);
5881 if (!test_bit(__IGB_DOWN, &adapter->state)) {
5882 if (adapter->msix_entries)
5883 wr32(E1000_EIMS, q_vector->eims_value);
5885 igb_irq_enable(adapter);
5890 * igb_poll - NAPI Rx polling callback
5891 * @napi: napi polling structure
5892 * @budget: count of how many packets we should handle
5894 static int igb_poll(struct napi_struct *napi, int budget)
5896 struct igb_q_vector *q_vector = container_of(napi,
5897 struct igb_q_vector,
5899 bool clean_complete = true;
5901 #ifdef CONFIG_IGB_DCA
5902 if (q_vector->adapter->flags & IGB_FLAG_DCA_ENABLED)
5903 igb_update_dca(q_vector);
5905 if (q_vector->tx.ring)
5906 clean_complete = igb_clean_tx_irq(q_vector);
5908 if (q_vector->rx.ring)
5909 clean_complete &= igb_clean_rx_irq(q_vector, budget);
5911 /* If all work not completed, return budget and keep polling */
5912 if (!clean_complete)
5915 /* If not enough Rx work done, exit the polling mode */
5916 napi_complete(napi);
5917 igb_ring_irq_enable(q_vector);
5923 * igb_clean_tx_irq - Reclaim resources after transmit completes
5924 * @q_vector: pointer to q_vector containing needed info
5926 * returns true if ring is completely cleaned
5928 static bool igb_clean_tx_irq(struct igb_q_vector *q_vector)
5930 struct igb_adapter *adapter = q_vector->adapter;
5931 struct igb_ring *tx_ring = q_vector->tx.ring;
5932 struct igb_tx_buffer *tx_buffer;
5933 union e1000_adv_tx_desc *tx_desc;
5934 unsigned int total_bytes = 0, total_packets = 0;
5935 unsigned int budget = q_vector->tx.work_limit;
5936 unsigned int i = tx_ring->next_to_clean;
5938 if (test_bit(__IGB_DOWN, &adapter->state))
5941 tx_buffer = &tx_ring->tx_buffer_info[i];
5942 tx_desc = IGB_TX_DESC(tx_ring, i);
5943 i -= tx_ring->count;
5946 union e1000_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
5948 /* if next_to_watch is not set then there is no work pending */
5952 /* prevent any other reads prior to eop_desc */
5953 read_barrier_depends();
5955 /* if DD is not set pending work has not been completed */
5956 if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)))
5959 /* clear next_to_watch to prevent false hangs */
5960 tx_buffer->next_to_watch = NULL;
5962 /* update the statistics for this packet */
5963 total_bytes += tx_buffer->bytecount;
5964 total_packets += tx_buffer->gso_segs;
5967 dev_kfree_skb_any(tx_buffer->skb);
5969 /* unmap skb header data */
5970 dma_unmap_single(tx_ring->dev,
5971 dma_unmap_addr(tx_buffer, dma),
5972 dma_unmap_len(tx_buffer, len),
5975 /* clear tx_buffer data */
5976 tx_buffer->skb = NULL;
5977 dma_unmap_len_set(tx_buffer, len, 0);
5979 /* clear last DMA location and unmap remaining buffers */
5980 while (tx_desc != eop_desc) {
5985 i -= tx_ring->count;
5986 tx_buffer = tx_ring->tx_buffer_info;
5987 tx_desc = IGB_TX_DESC(tx_ring, 0);
5990 /* unmap any remaining paged data */
5991 if (dma_unmap_len(tx_buffer, len)) {
5992 dma_unmap_page(tx_ring->dev,
5993 dma_unmap_addr(tx_buffer, dma),
5994 dma_unmap_len(tx_buffer, len),
5996 dma_unmap_len_set(tx_buffer, len, 0);
6000 /* move us one more past the eop_desc for start of next pkt */
6005 i -= tx_ring->count;
6006 tx_buffer = tx_ring->tx_buffer_info;
6007 tx_desc = IGB_TX_DESC(tx_ring, 0);
6010 /* issue prefetch for next Tx descriptor */
6013 /* update budget accounting */
6015 } while (likely(budget));
6017 netdev_tx_completed_queue(txring_txq(tx_ring),
6018 total_packets, total_bytes);
6019 i += tx_ring->count;
6020 tx_ring->next_to_clean = i;
6021 u64_stats_update_begin(&tx_ring->tx_syncp);
6022 tx_ring->tx_stats.bytes += total_bytes;
6023 tx_ring->tx_stats.packets += total_packets;
6024 u64_stats_update_end(&tx_ring->tx_syncp);
6025 q_vector->tx.total_bytes += total_bytes;
6026 q_vector->tx.total_packets += total_packets;
6028 if (test_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags)) {
6029 struct e1000_hw *hw = &adapter->hw;
6031 /* Detect a transmit hang in hardware, this serializes the
6032 * check with the clearing of time_stamp and movement of i */
6033 clear_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
6034 if (tx_buffer->next_to_watch &&
6035 time_after(jiffies, tx_buffer->time_stamp +
6036 (adapter->tx_timeout_factor * HZ)) &&
6037 !(rd32(E1000_STATUS) & E1000_STATUS_TXOFF)) {
6039 /* detected Tx unit hang */
6040 dev_err(tx_ring->dev,
6041 "Detected Tx Unit Hang\n"
6045 " next_to_use <%x>\n"
6046 " next_to_clean <%x>\n"
6047 "buffer_info[next_to_clean]\n"
6048 " time_stamp <%lx>\n"
6049 " next_to_watch <%p>\n"
6051 " desc.status <%x>\n",
6052 tx_ring->queue_index,
6053 rd32(E1000_TDH(tx_ring->reg_idx)),
6054 readl(tx_ring->tail),
6055 tx_ring->next_to_use,
6056 tx_ring->next_to_clean,
6057 tx_buffer->time_stamp,
6058 tx_buffer->next_to_watch,
6060 tx_buffer->next_to_watch->wb.status);
6061 netif_stop_subqueue(tx_ring->netdev,
6062 tx_ring->queue_index);
6064 /* we are about to reset, no point in enabling stuff */
6069 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
6070 if (unlikely(total_packets &&
6071 netif_carrier_ok(tx_ring->netdev) &&
6072 igb_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD)) {
6073 /* Make sure that anybody stopping the queue after this
6074 * sees the new next_to_clean.
6077 if (__netif_subqueue_stopped(tx_ring->netdev,
6078 tx_ring->queue_index) &&
6079 !(test_bit(__IGB_DOWN, &adapter->state))) {
6080 netif_wake_subqueue(tx_ring->netdev,
6081 tx_ring->queue_index);
6083 u64_stats_update_begin(&tx_ring->tx_syncp);
6084 tx_ring->tx_stats.restart_queue++;
6085 u64_stats_update_end(&tx_ring->tx_syncp);
6093 * igb_reuse_rx_page - page flip buffer and store it back on the ring
6094 * @rx_ring: rx descriptor ring to store buffers on
6095 * @old_buff: donor buffer to have page reused
6097 * Synchronizes page for reuse by the adapter
6099 static void igb_reuse_rx_page(struct igb_ring *rx_ring,
6100 struct igb_rx_buffer *old_buff)
6102 struct igb_rx_buffer *new_buff;
6103 u16 nta = rx_ring->next_to_alloc;
6105 new_buff = &rx_ring->rx_buffer_info[nta];
6107 /* update, and store next to alloc */
6109 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
6111 /* transfer page from old buffer to new buffer */
6112 memcpy(new_buff, old_buff, sizeof(struct igb_rx_buffer));
6114 /* sync the buffer for use by the device */
6115 dma_sync_single_range_for_device(rx_ring->dev, old_buff->dma,
6116 old_buff->page_offset,
6121 static bool igb_can_reuse_rx_page(struct igb_rx_buffer *rx_buffer,
6123 unsigned int truesize)
6125 /* avoid re-using remote pages */
6126 if (unlikely(page_to_nid(page) != numa_node_id()))
6129 #if (PAGE_SIZE < 8192)
6130 /* if we are only owner of page we can reuse it */
6131 if (unlikely(page_count(page) != 1))
6134 /* flip page offset to other buffer */
6135 rx_buffer->page_offset ^= IGB_RX_BUFSZ;
6137 /* since we are the only owner of the page and we need to
6138 * increment it, just set the value to 2 in order to avoid
6139 * an unnecessary locked operation
6141 atomic_set(&page->_count, 2);
6143 /* move offset up to the next cache line */
6144 rx_buffer->page_offset += truesize;
6146 if (rx_buffer->page_offset > (PAGE_SIZE - IGB_RX_BUFSZ))
6149 /* bump ref count on page before it is given to the stack */
6157 * igb_add_rx_frag - Add contents of Rx buffer to sk_buff
6158 * @rx_ring: rx descriptor ring to transact packets on
6159 * @rx_buffer: buffer containing page to add
6160 * @rx_desc: descriptor containing length of buffer written by hardware
6161 * @skb: sk_buff to place the data into
6163 * This function will add the data contained in rx_buffer->page to the skb.
6164 * This is done either through a direct copy if the data in the buffer is
6165 * less than the skb header size, otherwise it will just attach the page as
6166 * a frag to the skb.
6168 * The function will then update the page offset if necessary and return
6169 * true if the buffer can be reused by the adapter.
6171 static bool igb_add_rx_frag(struct igb_ring *rx_ring,
6172 struct igb_rx_buffer *rx_buffer,
6173 union e1000_adv_rx_desc *rx_desc,
6174 struct sk_buff *skb)
6176 struct page *page = rx_buffer->page;
6177 unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
6178 #if (PAGE_SIZE < 8192)
6179 unsigned int truesize = IGB_RX_BUFSZ;
6181 unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
6184 if ((size <= IGB_RX_HDR_LEN) && !skb_is_nonlinear(skb)) {
6185 unsigned char *va = page_address(page) + rx_buffer->page_offset;
6187 if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
6188 igb_ptp_rx_pktstamp(rx_ring->q_vector, va, skb);
6189 va += IGB_TS_HDR_LEN;
6190 size -= IGB_TS_HDR_LEN;
6193 memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
6195 /* we can reuse buffer as-is, just make sure it is local */
6196 if (likely(page_to_nid(page) == numa_node_id()))
6199 /* this page cannot be reused so discard it */
6204 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
6205 rx_buffer->page_offset, size, truesize);
6207 return igb_can_reuse_rx_page(rx_buffer, page, truesize);
6210 static struct sk_buff *igb_build_rx_buffer(struct igb_ring *rx_ring,
6211 union e1000_adv_rx_desc *rx_desc)
6213 struct igb_rx_buffer *rx_buffer;
6214 struct sk_buff *skb;
6217 unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
6218 #if (PAGE_SIZE < 8192)
6219 unsigned int truesize = IGB_RX_BUFSZ;
6221 unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
6222 SKB_DATA_ALIGN(NET_SKB_PAD +
6227 /* If we spanned a buffer we have a huge mess so test for it */
6228 BUG_ON(unlikely(!igb_test_staterr(rx_desc, E1000_RXD_STAT_EOP)));
6230 /* Guarantee this function can be used by verifying buffer sizes */
6231 BUILD_BUG_ON(SKB_WITH_OVERHEAD(IGB_RX_BUFSZ) < (NET_SKB_PAD +
6237 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
6238 page = rx_buffer->page;
6241 page_addr = page_address(page) + rx_buffer->page_offset;
6243 /* prefetch first cache line of first page */
6244 prefetch(page_addr + NET_SKB_PAD + NET_IP_ALIGN);
6245 #if L1_CACHE_BYTES < 128
6246 prefetch(page_addr + L1_CACHE_BYTES + NET_SKB_PAD + NET_IP_ALIGN);
6249 /* build an skb to around the page buffer */
6250 skb = build_skb(page_addr, truesize);
6251 if (unlikely(!skb)) {
6252 rx_ring->rx_stats.alloc_failed++;
6256 /* we are reusing so sync this buffer for CPU use */
6257 dma_sync_single_range_for_cpu(rx_ring->dev,
6259 rx_buffer->page_offset,
6263 /* update pointers within the skb to store the data */
6264 skb_reserve(skb, NET_IP_ALIGN + NET_SKB_PAD);
6265 __skb_put(skb, size);
6267 /* pull timestamp out of packet data */
6268 if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
6269 igb_ptp_rx_pktstamp(rx_ring->q_vector, skb->data, skb);
6270 __skb_pull(skb, IGB_TS_HDR_LEN);
6273 if (igb_can_reuse_rx_page(rx_buffer, page, truesize)) {
6274 /* hand second half of page back to the ring */
6275 igb_reuse_rx_page(rx_ring, rx_buffer);
6277 /* we are not reusing the buffer so unmap it */
6278 dma_unmap_page(rx_ring->dev, rx_buffer->dma,
6279 PAGE_SIZE, DMA_FROM_DEVICE);
6282 /* clear contents of buffer_info */
6284 rx_buffer->page = NULL;
6289 static struct sk_buff *igb_fetch_rx_buffer(struct igb_ring *rx_ring,
6290 union e1000_adv_rx_desc *rx_desc,
6291 struct sk_buff *skb)
6293 struct igb_rx_buffer *rx_buffer;
6296 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
6298 page = rx_buffer->page;
6302 void *page_addr = page_address(page) +
6303 rx_buffer->page_offset;
6305 /* prefetch first cache line of first page */
6306 prefetch(page_addr);
6307 #if L1_CACHE_BYTES < 128
6308 prefetch(page_addr + L1_CACHE_BYTES);
6311 /* allocate a skb to store the frags */
6312 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
6314 if (unlikely(!skb)) {
6315 rx_ring->rx_stats.alloc_failed++;
6320 * we will be copying header into skb->data in
6321 * pskb_may_pull so it is in our interest to prefetch
6322 * it now to avoid a possible cache miss
6324 prefetchw(skb->data);
6327 /* we are reusing so sync this buffer for CPU use */
6328 dma_sync_single_range_for_cpu(rx_ring->dev,
6330 rx_buffer->page_offset,
6334 /* pull page into skb */
6335 if (igb_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
6336 /* hand second half of page back to the ring */
6337 igb_reuse_rx_page(rx_ring, rx_buffer);
6339 /* we are not reusing the buffer so unmap it */
6340 dma_unmap_page(rx_ring->dev, rx_buffer->dma,
6341 PAGE_SIZE, DMA_FROM_DEVICE);
6344 /* clear contents of rx_buffer */
6345 rx_buffer->page = NULL;
6350 static inline void igb_rx_checksum(struct igb_ring *ring,
6351 union e1000_adv_rx_desc *rx_desc,
6352 struct sk_buff *skb)
6354 skb_checksum_none_assert(skb);
6356 /* Ignore Checksum bit is set */
6357 if (igb_test_staterr(rx_desc, E1000_RXD_STAT_IXSM))
6360 /* Rx checksum disabled via ethtool */
6361 if (!(ring->netdev->features & NETIF_F_RXCSUM))
6364 /* TCP/UDP checksum error bit is set */
6365 if (igb_test_staterr(rx_desc,
6366 E1000_RXDEXT_STATERR_TCPE |
6367 E1000_RXDEXT_STATERR_IPE)) {
6369 * work around errata with sctp packets where the TCPE aka
6370 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
6371 * packets, (aka let the stack check the crc32c)
6373 if (!((skb->len == 60) &&
6374 test_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags))) {
6375 u64_stats_update_begin(&ring->rx_syncp);
6376 ring->rx_stats.csum_err++;
6377 u64_stats_update_end(&ring->rx_syncp);
6379 /* let the stack verify checksum errors */
6382 /* It must be a TCP or UDP packet with a valid checksum */
6383 if (igb_test_staterr(rx_desc, E1000_RXD_STAT_TCPCS |
6384 E1000_RXD_STAT_UDPCS))
6385 skb->ip_summed = CHECKSUM_UNNECESSARY;
6387 dev_dbg(ring->dev, "cksum success: bits %08X\n",
6388 le32_to_cpu(rx_desc->wb.upper.status_error));
6391 static inline void igb_rx_hash(struct igb_ring *ring,
6392 union e1000_adv_rx_desc *rx_desc,
6393 struct sk_buff *skb)
6395 if (ring->netdev->features & NETIF_F_RXHASH)
6396 skb->rxhash = le32_to_cpu(rx_desc->wb.lower.hi_dword.rss);
6400 * igb_is_non_eop - process handling of non-EOP buffers
6401 * @rx_ring: Rx ring being processed
6402 * @rx_desc: Rx descriptor for current buffer
6403 * @skb: current socket buffer containing buffer in progress
6405 * This function updates next to clean. If the buffer is an EOP buffer
6406 * this function exits returning false, otherwise it will place the
6407 * sk_buff in the next buffer to be chained and return true indicating
6408 * that this is in fact a non-EOP buffer.
6410 static bool igb_is_non_eop(struct igb_ring *rx_ring,
6411 union e1000_adv_rx_desc *rx_desc)
6413 u32 ntc = rx_ring->next_to_clean + 1;
6415 /* fetch, update, and store next to clean */
6416 ntc = (ntc < rx_ring->count) ? ntc : 0;
6417 rx_ring->next_to_clean = ntc;
6419 prefetch(IGB_RX_DESC(rx_ring, ntc));
6421 if (likely(igb_test_staterr(rx_desc, E1000_RXD_STAT_EOP)))
6428 * igb_get_headlen - determine size of header for LRO/GRO
6429 * @data: pointer to the start of the headers
6430 * @max_len: total length of section to find headers in
6432 * This function is meant to determine the length of headers that will
6433 * be recognized by hardware for LRO, and GRO offloads. The main
6434 * motivation of doing this is to only perform one pull for IPv4 TCP
6435 * packets so that we can do basic things like calculating the gso_size
6436 * based on the average data per packet.
6438 static unsigned int igb_get_headlen(unsigned char *data,
6439 unsigned int max_len)
6442 unsigned char *network;
6445 struct vlan_hdr *vlan;
6448 struct ipv6hdr *ipv6;
6451 u8 nexthdr = 0; /* default to not TCP */
6454 /* this should never happen, but better safe than sorry */
6455 if (max_len < ETH_HLEN)
6458 /* initialize network frame pointer */
6461 /* set first protocol and move network header forward */
6462 protocol = hdr.eth->h_proto;
6463 hdr.network += ETH_HLEN;
6465 /* handle any vlan tag if present */
6466 if (protocol == __constant_htons(ETH_P_8021Q)) {
6467 if ((hdr.network - data) > (max_len - VLAN_HLEN))
6470 protocol = hdr.vlan->h_vlan_encapsulated_proto;
6471 hdr.network += VLAN_HLEN;
6474 /* handle L3 protocols */
6475 if (protocol == __constant_htons(ETH_P_IP)) {
6476 if ((hdr.network - data) > (max_len - sizeof(struct iphdr)))
6479 /* access ihl as a u8 to avoid unaligned access on ia64 */
6480 hlen = (hdr.network[0] & 0x0F) << 2;
6482 /* verify hlen meets minimum size requirements */
6483 if (hlen < sizeof(struct iphdr))
6484 return hdr.network - data;
6486 /* record next protocol if header is present */
6487 if (!hdr.ipv4->frag_off)
6488 nexthdr = hdr.ipv4->protocol;
6489 } else if (protocol == __constant_htons(ETH_P_IPV6)) {
6490 if ((hdr.network - data) > (max_len - sizeof(struct ipv6hdr)))
6493 /* record next protocol */
6494 nexthdr = hdr.ipv6->nexthdr;
6495 hlen = sizeof(struct ipv6hdr);
6497 return hdr.network - data;
6500 /* relocate pointer to start of L4 header */
6501 hdr.network += hlen;
6503 /* finally sort out TCP */
6504 if (nexthdr == IPPROTO_TCP) {
6505 if ((hdr.network - data) > (max_len - sizeof(struct tcphdr)))
6508 /* access doff as a u8 to avoid unaligned access on ia64 */
6509 hlen = (hdr.network[12] & 0xF0) >> 2;
6511 /* verify hlen meets minimum size requirements */
6512 if (hlen < sizeof(struct tcphdr))
6513 return hdr.network - data;
6515 hdr.network += hlen;
6516 } else if (nexthdr == IPPROTO_UDP) {
6517 if ((hdr.network - data) > (max_len - sizeof(struct udphdr)))
6520 hdr.network += sizeof(struct udphdr);
6524 * If everything has gone correctly hdr.network should be the
6525 * data section of the packet and will be the end of the header.
6526 * If not then it probably represents the end of the last recognized
6529 if ((hdr.network - data) < max_len)
6530 return hdr.network - data;
6536 * igb_pull_tail - igb specific version of skb_pull_tail
6537 * @rx_ring: rx descriptor ring packet is being transacted on
6538 * @rx_desc: pointer to the EOP Rx descriptor
6539 * @skb: pointer to current skb being adjusted
6541 * This function is an igb specific version of __pskb_pull_tail. The
6542 * main difference between this version and the original function is that
6543 * this function can make several assumptions about the state of things
6544 * that allow for significant optimizations versus the standard function.
6545 * As a result we can do things like drop a frag and maintain an accurate
6546 * truesize for the skb.
6548 static void igb_pull_tail(struct igb_ring *rx_ring,
6549 union e1000_adv_rx_desc *rx_desc,
6550 struct sk_buff *skb)
6552 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
6554 unsigned int pull_len;
6557 * it is valid to use page_address instead of kmap since we are
6558 * working with pages allocated out of the lomem pool per
6559 * alloc_page(GFP_ATOMIC)
6561 va = skb_frag_address(frag);
6563 if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
6564 /* retrieve timestamp from buffer */
6565 igb_ptp_rx_pktstamp(rx_ring->q_vector, va, skb);
6567 /* update pointers to remove timestamp header */
6568 skb_frag_size_sub(frag, IGB_TS_HDR_LEN);
6569 frag->page_offset += IGB_TS_HDR_LEN;
6570 skb->data_len -= IGB_TS_HDR_LEN;
6571 skb->len -= IGB_TS_HDR_LEN;
6573 /* move va to start of packet data */
6574 va += IGB_TS_HDR_LEN;
6578 * we need the header to contain the greater of either ETH_HLEN or
6579 * 60 bytes if the skb->len is less than 60 for skb_pad.
6581 pull_len = igb_get_headlen(va, IGB_RX_HDR_LEN);
6583 /* align pull length to size of long to optimize memcpy performance */
6584 skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
6586 /* update all of the pointers */
6587 skb_frag_size_sub(frag, pull_len);
6588 frag->page_offset += pull_len;
6589 skb->data_len -= pull_len;
6590 skb->tail += pull_len;
6594 * igb_cleanup_headers - Correct corrupted or empty headers
6595 * @rx_ring: rx descriptor ring packet is being transacted on
6596 * @rx_desc: pointer to the EOP Rx descriptor
6597 * @skb: pointer to current skb being fixed
6599 * Address the case where we are pulling data in on pages only
6600 * and as such no data is present in the skb header.
6602 * In addition if skb is not at least 60 bytes we need to pad it so that
6603 * it is large enough to qualify as a valid Ethernet frame.
6605 * Returns true if an error was encountered and skb was freed.
6607 static bool igb_cleanup_headers(struct igb_ring *rx_ring,
6608 union e1000_adv_rx_desc *rx_desc,
6609 struct sk_buff *skb)
6612 if (unlikely((igb_test_staterr(rx_desc,
6613 E1000_RXDEXT_ERR_FRAME_ERR_MASK)))) {
6614 struct net_device *netdev = rx_ring->netdev;
6615 if (!(netdev->features & NETIF_F_RXALL)) {
6616 dev_kfree_skb_any(skb);
6621 /* place header in linear portion of buffer */
6622 if (skb_is_nonlinear(skb))
6623 igb_pull_tail(rx_ring, rx_desc, skb);
6625 /* if skb_pad returns an error the skb was freed */
6626 if (unlikely(skb->len < 60)) {
6627 int pad_len = 60 - skb->len;
6629 if (skb_pad(skb, pad_len))
6631 __skb_put(skb, pad_len);
6638 * igb_process_skb_fields - Populate skb header fields from Rx descriptor
6639 * @rx_ring: rx descriptor ring packet is being transacted on
6640 * @rx_desc: pointer to the EOP Rx descriptor
6641 * @skb: pointer to current skb being populated
6643 * This function checks the ring, descriptor, and packet information in
6644 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
6645 * other fields within the skb.
6647 static void igb_process_skb_fields(struct igb_ring *rx_ring,
6648 union e1000_adv_rx_desc *rx_desc,
6649 struct sk_buff *skb)
6651 struct net_device *dev = rx_ring->netdev;
6653 igb_rx_hash(rx_ring, rx_desc, skb);
6655 igb_rx_checksum(rx_ring, rx_desc, skb);
6657 igb_ptp_rx_hwtstamp(rx_ring->q_vector, rx_desc, skb);
6659 if ((dev->features & NETIF_F_HW_VLAN_RX) &&
6660 igb_test_staterr(rx_desc, E1000_RXD_STAT_VP)) {
6662 if (igb_test_staterr(rx_desc, E1000_RXDEXT_STATERR_LB) &&
6663 test_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &rx_ring->flags))
6664 vid = be16_to_cpu(rx_desc->wb.upper.vlan);
6666 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
6668 __vlan_hwaccel_put_tag(skb, vid);
6671 skb_record_rx_queue(skb, rx_ring->queue_index);
6673 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
6676 static bool igb_clean_rx_irq(struct igb_q_vector *q_vector, const int budget)
6678 struct igb_ring *rx_ring = q_vector->rx.ring;
6679 struct sk_buff *skb = rx_ring->skb;
6680 unsigned int total_bytes = 0, total_packets = 0;
6681 u16 cleaned_count = igb_desc_unused(rx_ring);
6684 union e1000_adv_rx_desc *rx_desc;
6686 /* return some buffers to hardware, one at a time is too slow */
6687 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
6688 igb_alloc_rx_buffers(rx_ring, cleaned_count);
6692 rx_desc = IGB_RX_DESC(rx_ring, rx_ring->next_to_clean);
6694 if (!igb_test_staterr(rx_desc, E1000_RXD_STAT_DD))
6697 /* This memory barrier is needed to keep us from reading
6698 * any other fields out of the rx_desc until we know the
6699 * RXD_STAT_DD bit is set
6703 /* retrieve a buffer from the ring */
6704 if (ring_uses_build_skb(rx_ring))
6705 skb = igb_build_rx_buffer(rx_ring, rx_desc);
6707 skb = igb_fetch_rx_buffer(rx_ring, rx_desc, skb);
6709 /* exit if we failed to retrieve a buffer */
6715 /* fetch next buffer in frame if non-eop */
6716 if (igb_is_non_eop(rx_ring, rx_desc))
6719 /* verify the packet layout is correct */
6720 if (igb_cleanup_headers(rx_ring, rx_desc, skb)) {
6725 /* probably a little skewed due to removing CRC */
6726 total_bytes += skb->len;
6728 /* populate checksum, timestamp, VLAN, and protocol */
6729 igb_process_skb_fields(rx_ring, rx_desc, skb);
6731 napi_gro_receive(&q_vector->napi, skb);
6733 /* reset skb pointer */
6736 /* update budget accounting */
6738 } while (likely(total_packets < budget));
6740 /* place incomplete frames back on ring for completion */
6743 u64_stats_update_begin(&rx_ring->rx_syncp);
6744 rx_ring->rx_stats.packets += total_packets;
6745 rx_ring->rx_stats.bytes += total_bytes;
6746 u64_stats_update_end(&rx_ring->rx_syncp);
6747 q_vector->rx.total_packets += total_packets;
6748 q_vector->rx.total_bytes += total_bytes;
6751 igb_alloc_rx_buffers(rx_ring, cleaned_count);
6753 return (total_packets < budget);
6756 static bool igb_alloc_mapped_page(struct igb_ring *rx_ring,
6757 struct igb_rx_buffer *bi)
6759 struct page *page = bi->page;
6762 /* since we are recycling buffers we should seldom need to alloc */
6766 /* alloc new page for storage */
6767 page = __skb_alloc_page(GFP_ATOMIC | __GFP_COLD, NULL);
6768 if (unlikely(!page)) {
6769 rx_ring->rx_stats.alloc_failed++;
6773 /* map page for use */
6774 dma = dma_map_page(rx_ring->dev, page, 0, PAGE_SIZE, DMA_FROM_DEVICE);
6777 * if mapping failed free memory back to system since
6778 * there isn't much point in holding memory we can't use
6780 if (dma_mapping_error(rx_ring->dev, dma)) {
6783 rx_ring->rx_stats.alloc_failed++;
6789 bi->page_offset = 0;
6794 static inline unsigned int igb_rx_offset(struct igb_ring *rx_ring)
6796 if (ring_uses_build_skb(rx_ring))
6797 return NET_SKB_PAD + NET_IP_ALIGN;
6803 * igb_alloc_rx_buffers - Replace used receive buffers; packet split
6804 * @adapter: address of board private structure
6806 void igb_alloc_rx_buffers(struct igb_ring *rx_ring, u16 cleaned_count)
6808 union e1000_adv_rx_desc *rx_desc;
6809 struct igb_rx_buffer *bi;
6810 u16 i = rx_ring->next_to_use;
6816 rx_desc = IGB_RX_DESC(rx_ring, i);
6817 bi = &rx_ring->rx_buffer_info[i];
6818 i -= rx_ring->count;
6821 if (!igb_alloc_mapped_page(rx_ring, bi))
6825 * Refresh the desc even if buffer_addrs didn't change
6826 * because each write-back erases this info.
6828 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma +
6830 igb_rx_offset(rx_ring));
6836 rx_desc = IGB_RX_DESC(rx_ring, 0);
6837 bi = rx_ring->rx_buffer_info;
6838 i -= rx_ring->count;
6841 /* clear the hdr_addr for the next_to_use descriptor */
6842 rx_desc->read.hdr_addr = 0;
6845 } while (cleaned_count);
6847 i += rx_ring->count;
6849 if (rx_ring->next_to_use != i) {
6850 /* record the next descriptor to use */
6851 rx_ring->next_to_use = i;
6853 /* update next to alloc since we have filled the ring */
6854 rx_ring->next_to_alloc = i;
6857 * Force memory writes to complete before letting h/w
6858 * know there are new descriptors to fetch. (Only
6859 * applicable for weak-ordered memory model archs,
6863 writel(i, rx_ring->tail);
6873 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
6875 struct igb_adapter *adapter = netdev_priv(netdev);
6876 struct mii_ioctl_data *data = if_mii(ifr);
6878 if (adapter->hw.phy.media_type != e1000_media_type_copper)
6883 data->phy_id = adapter->hw.phy.addr;
6886 if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
6903 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
6909 return igb_mii_ioctl(netdev, ifr, cmd);
6911 return igb_ptp_hwtstamp_ioctl(netdev, ifr, cmd);
6917 s32 igb_read_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
6919 struct igb_adapter *adapter = hw->back;
6921 if (pcie_capability_read_word(adapter->pdev, reg, value))
6922 return -E1000_ERR_CONFIG;
6927 s32 igb_write_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
6929 struct igb_adapter *adapter = hw->back;
6931 if (pcie_capability_write_word(adapter->pdev, reg, *value))
6932 return -E1000_ERR_CONFIG;
6937 static void igb_vlan_mode(struct net_device *netdev, netdev_features_t features)
6939 struct igb_adapter *adapter = netdev_priv(netdev);
6940 struct e1000_hw *hw = &adapter->hw;
6942 bool enable = !!(features & NETIF_F_HW_VLAN_RX);
6945 /* enable VLAN tag insert/strip */
6946 ctrl = rd32(E1000_CTRL);
6947 ctrl |= E1000_CTRL_VME;
6948 wr32(E1000_CTRL, ctrl);
6950 /* Disable CFI check */
6951 rctl = rd32(E1000_RCTL);
6952 rctl &= ~E1000_RCTL_CFIEN;
6953 wr32(E1000_RCTL, rctl);
6955 /* disable VLAN tag insert/strip */
6956 ctrl = rd32(E1000_CTRL);
6957 ctrl &= ~E1000_CTRL_VME;
6958 wr32(E1000_CTRL, ctrl);
6961 igb_rlpml_set(adapter);
6964 static int igb_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
6966 struct igb_adapter *adapter = netdev_priv(netdev);
6967 struct e1000_hw *hw = &adapter->hw;
6968 int pf_id = adapter->vfs_allocated_count;
6970 /* attempt to add filter to vlvf array */
6971 igb_vlvf_set(adapter, vid, true, pf_id);
6973 /* add the filter since PF can receive vlans w/o entry in vlvf */
6974 igb_vfta_set(hw, vid, true);
6976 set_bit(vid, adapter->active_vlans);
6981 static int igb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
6983 struct igb_adapter *adapter = netdev_priv(netdev);
6984 struct e1000_hw *hw = &adapter->hw;
6985 int pf_id = adapter->vfs_allocated_count;
6988 /* remove vlan from VLVF table array */
6989 err = igb_vlvf_set(adapter, vid, false, pf_id);
6991 /* if vid was not present in VLVF just remove it from table */
6993 igb_vfta_set(hw, vid, false);
6995 clear_bit(vid, adapter->active_vlans);
7000 static void igb_restore_vlan(struct igb_adapter *adapter)
7004 igb_vlan_mode(adapter->netdev, adapter->netdev->features);
7006 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
7007 igb_vlan_rx_add_vid(adapter->netdev, vid);
7010 int igb_set_spd_dplx(struct igb_adapter *adapter, u32 spd, u8 dplx)
7012 struct pci_dev *pdev = adapter->pdev;
7013 struct e1000_mac_info *mac = &adapter->hw.mac;
7017 /* Make sure dplx is at most 1 bit and lsb of speed is not set
7018 * for the switch() below to work */
7019 if ((spd & 1) || (dplx & ~1))
7022 /* Fiber NIC's only allow 1000 Gbps Full duplex */
7023 if ((adapter->hw.phy.media_type == e1000_media_type_internal_serdes) &&
7024 spd != SPEED_1000 &&
7025 dplx != DUPLEX_FULL)
7028 switch (spd + dplx) {
7029 case SPEED_10 + DUPLEX_HALF:
7030 mac->forced_speed_duplex = ADVERTISE_10_HALF;
7032 case SPEED_10 + DUPLEX_FULL:
7033 mac->forced_speed_duplex = ADVERTISE_10_FULL;
7035 case SPEED_100 + DUPLEX_HALF:
7036 mac->forced_speed_duplex = ADVERTISE_100_HALF;
7038 case SPEED_100 + DUPLEX_FULL:
7039 mac->forced_speed_duplex = ADVERTISE_100_FULL;
7041 case SPEED_1000 + DUPLEX_FULL:
7043 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
7045 case SPEED_1000 + DUPLEX_HALF: /* not supported */
7050 /* clear MDI, MDI(-X) override is only allowed when autoneg enabled */
7051 adapter->hw.phy.mdix = AUTO_ALL_MODES;
7056 dev_err(&pdev->dev, "Unsupported Speed/Duplex configuration\n");
7060 static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake,
7063 struct net_device *netdev = pci_get_drvdata(pdev);
7064 struct igb_adapter *adapter = netdev_priv(netdev);
7065 struct e1000_hw *hw = &adapter->hw;
7066 u32 ctrl, rctl, status;
7067 u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
7072 netif_device_detach(netdev);
7074 if (netif_running(netdev))
7075 __igb_close(netdev, true);
7077 igb_clear_interrupt_scheme(adapter);
7080 retval = pci_save_state(pdev);
7085 status = rd32(E1000_STATUS);
7086 if (status & E1000_STATUS_LU)
7087 wufc &= ~E1000_WUFC_LNKC;
7090 igb_setup_rctl(adapter);
7091 igb_set_rx_mode(netdev);
7093 /* turn on all-multi mode if wake on multicast is enabled */
7094 if (wufc & E1000_WUFC_MC) {
7095 rctl = rd32(E1000_RCTL);
7096 rctl |= E1000_RCTL_MPE;
7097 wr32(E1000_RCTL, rctl);
7100 ctrl = rd32(E1000_CTRL);
7101 /* advertise wake from D3Cold */
7102 #define E1000_CTRL_ADVD3WUC 0x00100000
7103 /* phy power management enable */
7104 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
7105 ctrl |= E1000_CTRL_ADVD3WUC;
7106 wr32(E1000_CTRL, ctrl);
7108 /* Allow time for pending master requests to run */
7109 igb_disable_pcie_master(hw);
7111 wr32(E1000_WUC, E1000_WUC_PME_EN);
7112 wr32(E1000_WUFC, wufc);
7115 wr32(E1000_WUFC, 0);
7118 *enable_wake = wufc || adapter->en_mng_pt;
7120 igb_power_down_link(adapter);
7122 igb_power_up_link(adapter);
7124 /* Release control of h/w to f/w. If f/w is AMT enabled, this
7125 * would have already happened in close and is redundant. */
7126 igb_release_hw_control(adapter);
7128 pci_disable_device(pdev);
7134 #ifdef CONFIG_PM_SLEEP
7135 static int igb_suspend(struct device *dev)
7139 struct pci_dev *pdev = to_pci_dev(dev);
7141 retval = __igb_shutdown(pdev, &wake, 0);
7146 pci_prepare_to_sleep(pdev);
7148 pci_wake_from_d3(pdev, false);
7149 pci_set_power_state(pdev, PCI_D3hot);
7154 #endif /* CONFIG_PM_SLEEP */
7156 static int igb_resume(struct device *dev)
7158 struct pci_dev *pdev = to_pci_dev(dev);
7159 struct net_device *netdev = pci_get_drvdata(pdev);
7160 struct igb_adapter *adapter = netdev_priv(netdev);
7161 struct e1000_hw *hw = &adapter->hw;
7164 pci_set_power_state(pdev, PCI_D0);
7165 pci_restore_state(pdev);
7166 pci_save_state(pdev);
7168 err = pci_enable_device_mem(pdev);
7171 "igb: Cannot enable PCI device from suspend\n");
7174 pci_set_master(pdev);
7176 pci_enable_wake(pdev, PCI_D3hot, 0);
7177 pci_enable_wake(pdev, PCI_D3cold, 0);
7179 if (igb_init_interrupt_scheme(adapter, true)) {
7180 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
7186 /* let the f/w know that the h/w is now under the control of the
7188 igb_get_hw_control(adapter);
7190 wr32(E1000_WUS, ~0);
7192 if (netdev->flags & IFF_UP) {
7194 err = __igb_open(netdev, true);
7200 netif_device_attach(netdev);
7204 #ifdef CONFIG_PM_RUNTIME
7205 static int igb_runtime_idle(struct device *dev)
7207 struct pci_dev *pdev = to_pci_dev(dev);
7208 struct net_device *netdev = pci_get_drvdata(pdev);
7209 struct igb_adapter *adapter = netdev_priv(netdev);
7211 if (!igb_has_link(adapter))
7212 pm_schedule_suspend(dev, MSEC_PER_SEC * 5);
7217 static int igb_runtime_suspend(struct device *dev)
7219 struct pci_dev *pdev = to_pci_dev(dev);
7223 retval = __igb_shutdown(pdev, &wake, 1);
7228 pci_prepare_to_sleep(pdev);
7230 pci_wake_from_d3(pdev, false);
7231 pci_set_power_state(pdev, PCI_D3hot);
7237 static int igb_runtime_resume(struct device *dev)
7239 return igb_resume(dev);
7241 #endif /* CONFIG_PM_RUNTIME */
7244 static void igb_shutdown(struct pci_dev *pdev)
7248 __igb_shutdown(pdev, &wake, 0);
7250 if (system_state == SYSTEM_POWER_OFF) {
7251 pci_wake_from_d3(pdev, wake);
7252 pci_set_power_state(pdev, PCI_D3hot);
7256 #ifdef CONFIG_PCI_IOV
7257 static int igb_sriov_reinit(struct pci_dev *dev)
7259 struct net_device *netdev = pci_get_drvdata(dev);
7260 struct igb_adapter *adapter = netdev_priv(netdev);
7261 struct pci_dev *pdev = adapter->pdev;
7265 if (netif_running(netdev))
7268 igb_clear_interrupt_scheme(adapter);
7270 igb_init_queue_configuration(adapter);
7272 if (igb_init_interrupt_scheme(adapter, true)) {
7273 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
7277 if (netif_running(netdev))
7285 static int igb_pci_disable_sriov(struct pci_dev *dev)
7287 int err = igb_disable_sriov(dev);
7290 err = igb_sriov_reinit(dev);
7295 static int igb_pci_enable_sriov(struct pci_dev *dev, int num_vfs)
7297 int err = igb_enable_sriov(dev, num_vfs);
7302 err = igb_sriov_reinit(dev);
7311 static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs)
7313 #ifdef CONFIG_PCI_IOV
7315 return igb_pci_disable_sriov(dev);
7317 return igb_pci_enable_sriov(dev, num_vfs);
7322 #ifdef CONFIG_NET_POLL_CONTROLLER
7324 * Polling 'interrupt' - used by things like netconsole to send skbs
7325 * without having to re-enable interrupts. It's not called while
7326 * the interrupt routine is executing.
7328 static void igb_netpoll(struct net_device *netdev)
7330 struct igb_adapter *adapter = netdev_priv(netdev);
7331 struct e1000_hw *hw = &adapter->hw;
7332 struct igb_q_vector *q_vector;
7335 for (i = 0; i < adapter->num_q_vectors; i++) {
7336 q_vector = adapter->q_vector[i];
7337 if (adapter->msix_entries)
7338 wr32(E1000_EIMC, q_vector->eims_value);
7340 igb_irq_disable(adapter);
7341 napi_schedule(&q_vector->napi);
7344 #endif /* CONFIG_NET_POLL_CONTROLLER */
7347 * igb_io_error_detected - called when PCI error is detected
7348 * @pdev: Pointer to PCI device
7349 * @state: The current pci connection state
7351 * This function is called after a PCI bus error affecting
7352 * this device has been detected.
7354 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
7355 pci_channel_state_t state)
7357 struct net_device *netdev = pci_get_drvdata(pdev);
7358 struct igb_adapter *adapter = netdev_priv(netdev);
7360 netif_device_detach(netdev);
7362 if (state == pci_channel_io_perm_failure)
7363 return PCI_ERS_RESULT_DISCONNECT;
7365 if (netif_running(netdev))
7367 pci_disable_device(pdev);
7369 /* Request a slot slot reset. */
7370 return PCI_ERS_RESULT_NEED_RESET;
7374 * igb_io_slot_reset - called after the pci bus has been reset.
7375 * @pdev: Pointer to PCI device
7377 * Restart the card from scratch, as if from a cold-boot. Implementation
7378 * resembles the first-half of the igb_resume routine.
7380 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
7382 struct net_device *netdev = pci_get_drvdata(pdev);
7383 struct igb_adapter *adapter = netdev_priv(netdev);
7384 struct e1000_hw *hw = &adapter->hw;
7385 pci_ers_result_t result;
7388 if (pci_enable_device_mem(pdev)) {
7390 "Cannot re-enable PCI device after reset.\n");
7391 result = PCI_ERS_RESULT_DISCONNECT;
7393 pci_set_master(pdev);
7394 pci_restore_state(pdev);
7395 pci_save_state(pdev);
7397 pci_enable_wake(pdev, PCI_D3hot, 0);
7398 pci_enable_wake(pdev, PCI_D3cold, 0);
7401 wr32(E1000_WUS, ~0);
7402 result = PCI_ERS_RESULT_RECOVERED;
7405 err = pci_cleanup_aer_uncorrect_error_status(pdev);
7407 dev_err(&pdev->dev, "pci_cleanup_aer_uncorrect_error_status "
7408 "failed 0x%0x\n", err);
7409 /* non-fatal, continue */
7416 * igb_io_resume - called when traffic can start flowing again.
7417 * @pdev: Pointer to PCI device
7419 * This callback is called when the error recovery driver tells us that
7420 * its OK to resume normal operation. Implementation resembles the
7421 * second-half of the igb_resume routine.
7423 static void igb_io_resume(struct pci_dev *pdev)
7425 struct net_device *netdev = pci_get_drvdata(pdev);
7426 struct igb_adapter *adapter = netdev_priv(netdev);
7428 if (netif_running(netdev)) {
7429 if (igb_up(adapter)) {
7430 dev_err(&pdev->dev, "igb_up failed after reset\n");
7435 netif_device_attach(netdev);
7437 /* let the f/w know that the h/w is now under the control of the
7439 igb_get_hw_control(adapter);
7442 static void igb_rar_set_qsel(struct igb_adapter *adapter, u8 *addr, u32 index,
7445 u32 rar_low, rar_high;
7446 struct e1000_hw *hw = &adapter->hw;
7448 /* HW expects these in little endian so we reverse the byte order
7449 * from network order (big endian) to little endian
7451 rar_low = ((u32) addr[0] | ((u32) addr[1] << 8) |
7452 ((u32) addr[2] << 16) | ((u32) addr[3] << 24));
7453 rar_high = ((u32) addr[4] | ((u32) addr[5] << 8));
7455 /* Indicate to hardware the Address is Valid. */
7456 rar_high |= E1000_RAH_AV;
7458 if (hw->mac.type == e1000_82575)
7459 rar_high |= E1000_RAH_POOL_1 * qsel;
7461 rar_high |= E1000_RAH_POOL_1 << qsel;
7463 wr32(E1000_RAL(index), rar_low);
7465 wr32(E1000_RAH(index), rar_high);
7469 static int igb_set_vf_mac(struct igb_adapter *adapter,
7470 int vf, unsigned char *mac_addr)
7472 struct e1000_hw *hw = &adapter->hw;
7473 /* VF MAC addresses start at end of receive addresses and moves
7474 * torwards the first, as a result a collision should not be possible */
7475 int rar_entry = hw->mac.rar_entry_count - (vf + 1);
7477 memcpy(adapter->vf_data[vf].vf_mac_addresses, mac_addr, ETH_ALEN);
7479 igb_rar_set_qsel(adapter, mac_addr, rar_entry, vf);
7484 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
7486 struct igb_adapter *adapter = netdev_priv(netdev);
7487 if (!is_valid_ether_addr(mac) || (vf >= adapter->vfs_allocated_count))
7489 adapter->vf_data[vf].flags |= IGB_VF_FLAG_PF_SET_MAC;
7490 dev_info(&adapter->pdev->dev, "setting MAC %pM on VF %d\n", mac, vf);
7491 dev_info(&adapter->pdev->dev, "Reload the VF driver to make this"
7492 " change effective.");
7493 if (test_bit(__IGB_DOWN, &adapter->state)) {
7494 dev_warn(&adapter->pdev->dev, "The VF MAC address has been set,"
7495 " but the PF device is not up.\n");
7496 dev_warn(&adapter->pdev->dev, "Bring the PF device up before"
7497 " attempting to use the VF device.\n");
7499 return igb_set_vf_mac(adapter, vf, mac);
7502 static int igb_link_mbps(int internal_link_speed)
7504 switch (internal_link_speed) {
7514 static void igb_set_vf_rate_limit(struct e1000_hw *hw, int vf, int tx_rate,
7521 /* Calculate the rate factor values to set */
7522 rf_int = link_speed / tx_rate;
7523 rf_dec = (link_speed - (rf_int * tx_rate));
7524 rf_dec = (rf_dec * (1<<E1000_RTTBCNRC_RF_INT_SHIFT)) / tx_rate;
7526 bcnrc_val = E1000_RTTBCNRC_RS_ENA;
7527 bcnrc_val |= ((rf_int<<E1000_RTTBCNRC_RF_INT_SHIFT) &
7528 E1000_RTTBCNRC_RF_INT_MASK);
7529 bcnrc_val |= (rf_dec & E1000_RTTBCNRC_RF_DEC_MASK);
7534 wr32(E1000_RTTDQSEL, vf); /* vf X uses queue X */
7536 * Set global transmit compensation time to the MMW_SIZE in RTTBCNRM
7537 * register. MMW_SIZE=0x014 if 9728-byte jumbo is supported.
7539 wr32(E1000_RTTBCNRM, 0x14);
7540 wr32(E1000_RTTBCNRC, bcnrc_val);
7543 static void igb_check_vf_rate_limit(struct igb_adapter *adapter)
7545 int actual_link_speed, i;
7546 bool reset_rate = false;
7548 /* VF TX rate limit was not set or not supported */
7549 if ((adapter->vf_rate_link_speed == 0) ||
7550 (adapter->hw.mac.type != e1000_82576))
7553 actual_link_speed = igb_link_mbps(adapter->link_speed);
7554 if (actual_link_speed != adapter->vf_rate_link_speed) {
7556 adapter->vf_rate_link_speed = 0;
7557 dev_info(&adapter->pdev->dev,
7558 "Link speed has been changed. VF Transmit "
7559 "rate is disabled\n");
7562 for (i = 0; i < adapter->vfs_allocated_count; i++) {
7564 adapter->vf_data[i].tx_rate = 0;
7566 igb_set_vf_rate_limit(&adapter->hw, i,
7567 adapter->vf_data[i].tx_rate,
7572 static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf, int tx_rate)
7574 struct igb_adapter *adapter = netdev_priv(netdev);
7575 struct e1000_hw *hw = &adapter->hw;
7576 int actual_link_speed;
7578 if (hw->mac.type != e1000_82576)
7581 actual_link_speed = igb_link_mbps(adapter->link_speed);
7582 if ((vf >= adapter->vfs_allocated_count) ||
7583 (!(rd32(E1000_STATUS) & E1000_STATUS_LU)) ||
7584 (tx_rate < 0) || (tx_rate > actual_link_speed))
7587 adapter->vf_rate_link_speed = actual_link_speed;
7588 adapter->vf_data[vf].tx_rate = (u16)tx_rate;
7589 igb_set_vf_rate_limit(hw, vf, tx_rate, actual_link_speed);
7594 static int igb_ndo_get_vf_config(struct net_device *netdev,
7595 int vf, struct ifla_vf_info *ivi)
7597 struct igb_adapter *adapter = netdev_priv(netdev);
7598 if (vf >= adapter->vfs_allocated_count)
7601 memcpy(&ivi->mac, adapter->vf_data[vf].vf_mac_addresses, ETH_ALEN);
7602 ivi->tx_rate = adapter->vf_data[vf].tx_rate;
7603 ivi->vlan = adapter->vf_data[vf].pf_vlan;
7604 ivi->qos = adapter->vf_data[vf].pf_qos;
7608 static void igb_vmm_control(struct igb_adapter *adapter)
7610 struct e1000_hw *hw = &adapter->hw;
7613 switch (hw->mac.type) {
7618 /* replication is not supported for 82575 */
7621 /* notify HW that the MAC is adding vlan tags */
7622 reg = rd32(E1000_DTXCTL);
7623 reg |= E1000_DTXCTL_VLAN_ADDED;
7624 wr32(E1000_DTXCTL, reg);
7626 /* enable replication vlan tag stripping */
7627 reg = rd32(E1000_RPLOLR);
7628 reg |= E1000_RPLOLR_STRVLAN;
7629 wr32(E1000_RPLOLR, reg);
7631 /* none of the above registers are supported by i350 */
7635 if (adapter->vfs_allocated_count) {
7636 igb_vmdq_set_loopback_pf(hw, true);
7637 igb_vmdq_set_replication_pf(hw, true);
7638 igb_vmdq_set_anti_spoofing_pf(hw, true,
7639 adapter->vfs_allocated_count);
7641 igb_vmdq_set_loopback_pf(hw, false);
7642 igb_vmdq_set_replication_pf(hw, false);
7646 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba)
7648 struct e1000_hw *hw = &adapter->hw;
7652 if (hw->mac.type > e1000_82580) {
7653 if (adapter->flags & IGB_FLAG_DMAC) {
7656 /* force threshold to 0. */
7657 wr32(E1000_DMCTXTH, 0);
7660 * DMA Coalescing high water mark needs to be greater
7661 * than the Rx threshold. Set hwm to PBA - max frame
7662 * size in 16B units, capping it at PBA - 6KB.
7664 hwm = 64 * pba - adapter->max_frame_size / 16;
7665 if (hwm < 64 * (pba - 6))
7666 hwm = 64 * (pba - 6);
7667 reg = rd32(E1000_FCRTC);
7668 reg &= ~E1000_FCRTC_RTH_COAL_MASK;
7669 reg |= ((hwm << E1000_FCRTC_RTH_COAL_SHIFT)
7670 & E1000_FCRTC_RTH_COAL_MASK);
7671 wr32(E1000_FCRTC, reg);
7674 * Set the DMA Coalescing Rx threshold to PBA - 2 * max
7675 * frame size, capping it at PBA - 10KB.
7677 dmac_thr = pba - adapter->max_frame_size / 512;
7678 if (dmac_thr < pba - 10)
7679 dmac_thr = pba - 10;
7680 reg = rd32(E1000_DMACR);
7681 reg &= ~E1000_DMACR_DMACTHR_MASK;
7682 reg |= ((dmac_thr << E1000_DMACR_DMACTHR_SHIFT)
7683 & E1000_DMACR_DMACTHR_MASK);
7685 /* transition to L0x or L1 if available..*/
7686 reg |= (E1000_DMACR_DMAC_EN | E1000_DMACR_DMAC_LX_MASK);
7688 /* watchdog timer= +-1000 usec in 32usec intervals */
7691 /* Disable BMC-to-OS Watchdog Enable */
7692 reg &= ~E1000_DMACR_DC_BMC2OSW_EN;
7693 wr32(E1000_DMACR, reg);
7696 * no lower threshold to disable
7697 * coalescing(smart fifb)-UTRESH=0
7699 wr32(E1000_DMCRTRH, 0);
7701 reg = (IGB_DMCTLX_DCFLUSH_DIS | 0x4);
7703 wr32(E1000_DMCTLX, reg);
7706 * free space in tx packet buffer to wake from
7709 wr32(E1000_DMCTXTH, (IGB_MIN_TXPBSIZE -
7710 (IGB_TX_BUF_4096 + adapter->max_frame_size)) >> 6);
7713 * make low power state decision controlled
7716 reg = rd32(E1000_PCIEMISC);
7717 reg &= ~E1000_PCIEMISC_LX_DECISION;
7718 wr32(E1000_PCIEMISC, reg);
7719 } /* endif adapter->dmac is not disabled */
7720 } else if (hw->mac.type == e1000_82580) {
7721 u32 reg = rd32(E1000_PCIEMISC);
7722 wr32(E1000_PCIEMISC, reg & ~E1000_PCIEMISC_LX_DECISION);
7723 wr32(E1000_DMACR, 0);
7727 static DEFINE_SPINLOCK(i2c_clients_lock);
7729 /* igb_get_i2c_client - returns matching client
7730 * in adapters's client list.
7731 * @adapter: adapter struct
7732 * @dev_addr: device address of i2c needed.
7734 static struct i2c_client *
7735 igb_get_i2c_client(struct igb_adapter *adapter, u8 dev_addr)
7738 struct igb_i2c_client_list *client_list;
7739 struct i2c_client *client = NULL;
7740 struct i2c_board_info client_info = {
7741 I2C_BOARD_INFO("igb", 0x00),
7744 spin_lock_irqsave(&i2c_clients_lock, flags);
7745 client_list = adapter->i2c_clients;
7747 /* See if we already have an i2c_client */
7748 while (client_list) {
7749 if (client_list->client->addr == (dev_addr >> 1)) {
7750 client = client_list->client;
7753 client_list = client_list->next;
7757 /* no client_list found, create a new one */
7758 client_list = kzalloc(sizeof(*client_list), GFP_ATOMIC);
7759 if (client_list == NULL)
7762 /* dev_addr passed to us is left-shifted by 1 bit
7763 * i2c_new_device call expects it to be flush to the right.
7765 client_info.addr = dev_addr >> 1;
7766 client_info.platform_data = adapter;
7767 client_list->client = i2c_new_device(&adapter->i2c_adap, &client_info);
7768 if (client_list->client == NULL) {
7769 dev_info(&adapter->pdev->dev,
7770 "Failed to create new i2c device..\n");
7774 /* insert new client at head of list */
7775 client_list->next = adapter->i2c_clients;
7776 adapter->i2c_clients = client_list;
7778 client = client_list->client;
7784 spin_unlock_irqrestore(&i2c_clients_lock, flags);
7788 /* igb_read_i2c_byte - Reads 8 bit word over I2C
7789 * @hw: pointer to hardware structure
7790 * @byte_offset: byte offset to read
7791 * @dev_addr: device address
7794 * Performs byte read operation over I2C interface at
7795 * a specified device address.
7797 s32 igb_read_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
7798 u8 dev_addr, u8 *data)
7800 struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
7801 struct i2c_client *this_client = igb_get_i2c_client(adapter, dev_addr);
7806 return E1000_ERR_I2C;
7808 swfw_mask = E1000_SWFW_PHY0_SM;
7810 if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask)
7812 return E1000_ERR_SWFW_SYNC;
7814 status = i2c_smbus_read_byte_data(this_client, byte_offset);
7815 hw->mac.ops.release_swfw_sync(hw, swfw_mask);
7818 return E1000_ERR_I2C;
7821 return E1000_SUCCESS;
7825 /* igb_write_i2c_byte - Writes 8 bit word over I2C
7826 * @hw: pointer to hardware structure
7827 * @byte_offset: byte offset to write
7828 * @dev_addr: device address
7829 * @data: value to write
7831 * Performs byte write operation over I2C interface at
7832 * a specified device address.
7834 s32 igb_write_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
7835 u8 dev_addr, u8 data)
7837 struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
7838 struct i2c_client *this_client = igb_get_i2c_client(adapter, dev_addr);
7840 u16 swfw_mask = E1000_SWFW_PHY0_SM;
7843 return E1000_ERR_I2C;
7845 if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask) != E1000_SUCCESS)
7846 return E1000_ERR_SWFW_SYNC;
7847 status = i2c_smbus_write_byte_data(this_client, byte_offset, data);
7848 hw->mac.ops.release_swfw_sync(hw, swfw_mask);
7851 return E1000_ERR_I2C;
7853 return E1000_SUCCESS;