1 /*******************************************************************************
3 Intel(R) Gigabit Ethernet Linux driver
4 Copyright(c) 2007-2013 Intel Corporation.
6 This program is free software; you can redistribute it and/or modify it
7 under the terms and conditions of the GNU General Public License,
8 version 2, as published by the Free Software Foundation.
10 This program is distributed in the hope it will be useful, but WITHOUT
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
23 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26 *******************************************************************************/
28 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
30 #include <linux/module.h>
31 #include <linux/types.h>
32 #include <linux/init.h>
33 #include <linux/bitops.h>
34 #include <linux/vmalloc.h>
35 #include <linux/pagemap.h>
36 #include <linux/netdevice.h>
37 #include <linux/ipv6.h>
38 #include <linux/slab.h>
39 #include <net/checksum.h>
40 #include <net/ip6_checksum.h>
41 #include <linux/net_tstamp.h>
42 #include <linux/mii.h>
43 #include <linux/ethtool.h>
45 #include <linux/if_vlan.h>
46 #include <linux/pci.h>
47 #include <linux/pci-aspm.h>
48 #include <linux/delay.h>
49 #include <linux/interrupt.h>
51 #include <linux/tcp.h>
52 #include <linux/sctp.h>
53 #include <linux/if_ether.h>
54 #include <linux/aer.h>
55 #include <linux/prefetch.h>
56 #include <linux/pm_runtime.h>
58 #include <linux/dca.h>
60 #include <linux/i2c.h>
66 #define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) "." \
67 __stringify(BUILD) "-k"
68 char igb_driver_name[] = "igb";
69 char igb_driver_version[] = DRV_VERSION;
70 static const char igb_driver_string[] =
71 "Intel(R) Gigabit Ethernet Network Driver";
72 static const char igb_copyright[] =
73 "Copyright (c) 2007-2013 Intel Corporation.";
75 static const struct e1000_info *igb_info_tbl[] = {
76 [board_82575] = &e1000_82575_info,
79 static DEFINE_PCI_DEVICE_TABLE(igb_pci_tbl) = {
80 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_BACKPLANE_1GBPS) },
81 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_SGMII) },
82 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_BACKPLANE_2_5GBPS) },
83 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I211_COPPER), board_82575 },
84 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER), board_82575 },
85 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_FIBER), board_82575 },
86 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES), board_82575 },
87 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SGMII), board_82575 },
88 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER_FLASHLESS), board_82575 },
89 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES_FLASHLESS), board_82575 },
90 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_COPPER), board_82575 },
91 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_FIBER), board_82575 },
92 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SERDES), board_82575 },
93 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SGMII), board_82575 },
94 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER), board_82575 },
95 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_FIBER), board_82575 },
96 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_QUAD_FIBER), board_82575 },
97 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SERDES), board_82575 },
98 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SGMII), board_82575 },
99 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER_DUAL), board_82575 },
100 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SGMII), board_82575 },
101 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SERDES), board_82575 },
102 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_BACKPLANE), board_82575 },
103 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SFP), board_82575 },
104 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576), board_82575 },
105 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS), board_82575 },
106 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS_SERDES), board_82575 },
107 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_FIBER), board_82575 },
108 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES), board_82575 },
109 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES_QUAD), board_82575 },
110 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER_ET2), board_82575 },
111 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER), board_82575 },
112 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
113 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
114 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
115 /* required last entry */
119 MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
121 void igb_reset(struct igb_adapter *);
122 static int igb_setup_all_tx_resources(struct igb_adapter *);
123 static int igb_setup_all_rx_resources(struct igb_adapter *);
124 static void igb_free_all_tx_resources(struct igb_adapter *);
125 static void igb_free_all_rx_resources(struct igb_adapter *);
126 static void igb_setup_mrqc(struct igb_adapter *);
127 static int igb_probe(struct pci_dev *, const struct pci_device_id *);
128 static void igb_remove(struct pci_dev *pdev);
129 static int igb_sw_init(struct igb_adapter *);
130 static int igb_open(struct net_device *);
131 static int igb_close(struct net_device *);
132 static void igb_configure(struct igb_adapter *);
133 static void igb_configure_tx(struct igb_adapter *);
134 static void igb_configure_rx(struct igb_adapter *);
135 static void igb_clean_all_tx_rings(struct igb_adapter *);
136 static void igb_clean_all_rx_rings(struct igb_adapter *);
137 static void igb_clean_tx_ring(struct igb_ring *);
138 static void igb_clean_rx_ring(struct igb_ring *);
139 static void igb_set_rx_mode(struct net_device *);
140 static void igb_update_phy_info(unsigned long);
141 static void igb_watchdog(unsigned long);
142 static void igb_watchdog_task(struct work_struct *);
143 static netdev_tx_t igb_xmit_frame(struct sk_buff *skb, struct net_device *);
144 static struct rtnl_link_stats64 *igb_get_stats64(struct net_device *dev,
145 struct rtnl_link_stats64 *stats);
146 static int igb_change_mtu(struct net_device *, int);
147 static int igb_set_mac(struct net_device *, void *);
148 static void igb_set_uta(struct igb_adapter *adapter);
149 static irqreturn_t igb_intr(int irq, void *);
150 static irqreturn_t igb_intr_msi(int irq, void *);
151 static irqreturn_t igb_msix_other(int irq, void *);
152 static irqreturn_t igb_msix_ring(int irq, void *);
153 #ifdef CONFIG_IGB_DCA
154 static void igb_update_dca(struct igb_q_vector *);
155 static void igb_setup_dca(struct igb_adapter *);
156 #endif /* CONFIG_IGB_DCA */
157 static int igb_poll(struct napi_struct *, int);
158 static bool igb_clean_tx_irq(struct igb_q_vector *);
159 static bool igb_clean_rx_irq(struct igb_q_vector *, int);
160 static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
161 static void igb_tx_timeout(struct net_device *);
162 static void igb_reset_task(struct work_struct *);
163 static void igb_vlan_mode(struct net_device *netdev, netdev_features_t features);
164 static int igb_vlan_rx_add_vid(struct net_device *, __be16, u16);
165 static int igb_vlan_rx_kill_vid(struct net_device *, __be16, u16);
166 static void igb_restore_vlan(struct igb_adapter *);
167 static void igb_rar_set_qsel(struct igb_adapter *, u8 *, u32 , u8);
168 static void igb_ping_all_vfs(struct igb_adapter *);
169 static void igb_msg_task(struct igb_adapter *);
170 static void igb_vmm_control(struct igb_adapter *);
171 static int igb_set_vf_mac(struct igb_adapter *, int, unsigned char *);
172 static void igb_restore_vf_multicasts(struct igb_adapter *adapter);
173 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac);
174 static int igb_ndo_set_vf_vlan(struct net_device *netdev,
175 int vf, u16 vlan, u8 qos);
176 static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf, int tx_rate);
177 static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf,
179 static int igb_ndo_get_vf_config(struct net_device *netdev, int vf,
180 struct ifla_vf_info *ivi);
181 static void igb_check_vf_rate_limit(struct igb_adapter *);
183 #ifdef CONFIG_PCI_IOV
184 static int igb_vf_configure(struct igb_adapter *adapter, int vf);
188 #ifdef CONFIG_PM_SLEEP
189 static int igb_suspend(struct device *);
191 static int igb_resume(struct device *);
192 #ifdef CONFIG_PM_RUNTIME
193 static int igb_runtime_suspend(struct device *dev);
194 static int igb_runtime_resume(struct device *dev);
195 static int igb_runtime_idle(struct device *dev);
197 static const struct dev_pm_ops igb_pm_ops = {
198 SET_SYSTEM_SLEEP_PM_OPS(igb_suspend, igb_resume)
199 SET_RUNTIME_PM_OPS(igb_runtime_suspend, igb_runtime_resume,
203 static void igb_shutdown(struct pci_dev *);
204 static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs);
205 #ifdef CONFIG_IGB_DCA
206 static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
207 static struct notifier_block dca_notifier = {
208 .notifier_call = igb_notify_dca,
213 #ifdef CONFIG_NET_POLL_CONTROLLER
214 /* for netdump / net console */
215 static void igb_netpoll(struct net_device *);
217 #ifdef CONFIG_PCI_IOV
218 static unsigned int max_vfs = 0;
219 module_param(max_vfs, uint, 0);
220 MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate "
221 "per physical function");
222 #endif /* CONFIG_PCI_IOV */
224 static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
225 pci_channel_state_t);
226 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
227 static void igb_io_resume(struct pci_dev *);
229 static const struct pci_error_handlers igb_err_handler = {
230 .error_detected = igb_io_error_detected,
231 .slot_reset = igb_io_slot_reset,
232 .resume = igb_io_resume,
235 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba);
237 static struct pci_driver igb_driver = {
238 .name = igb_driver_name,
239 .id_table = igb_pci_tbl,
241 .remove = igb_remove,
243 .driver.pm = &igb_pm_ops,
245 .shutdown = igb_shutdown,
246 .sriov_configure = igb_pci_sriov_configure,
247 .err_handler = &igb_err_handler
250 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
251 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
252 MODULE_LICENSE("GPL");
253 MODULE_VERSION(DRV_VERSION);
255 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
256 static int debug = -1;
257 module_param(debug, int, 0);
258 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
260 struct igb_reg_info {
265 static const struct igb_reg_info igb_reg_info_tbl[] = {
267 /* General Registers */
268 {E1000_CTRL, "CTRL"},
269 {E1000_STATUS, "STATUS"},
270 {E1000_CTRL_EXT, "CTRL_EXT"},
272 /* Interrupt Registers */
276 {E1000_RCTL, "RCTL"},
277 {E1000_RDLEN(0), "RDLEN"},
278 {E1000_RDH(0), "RDH"},
279 {E1000_RDT(0), "RDT"},
280 {E1000_RXDCTL(0), "RXDCTL"},
281 {E1000_RDBAL(0), "RDBAL"},
282 {E1000_RDBAH(0), "RDBAH"},
285 {E1000_TCTL, "TCTL"},
286 {E1000_TDBAL(0), "TDBAL"},
287 {E1000_TDBAH(0), "TDBAH"},
288 {E1000_TDLEN(0), "TDLEN"},
289 {E1000_TDH(0), "TDH"},
290 {E1000_TDT(0), "TDT"},
291 {E1000_TXDCTL(0), "TXDCTL"},
292 {E1000_TDFH, "TDFH"},
293 {E1000_TDFT, "TDFT"},
294 {E1000_TDFHS, "TDFHS"},
295 {E1000_TDFPC, "TDFPC"},
297 /* List Terminator */
301 /* igb_regdump - register printout routine */
302 static void igb_regdump(struct e1000_hw *hw, struct igb_reg_info *reginfo)
308 switch (reginfo->ofs) {
310 for (n = 0; n < 4; n++)
311 regs[n] = rd32(E1000_RDLEN(n));
314 for (n = 0; n < 4; n++)
315 regs[n] = rd32(E1000_RDH(n));
318 for (n = 0; n < 4; n++)
319 regs[n] = rd32(E1000_RDT(n));
321 case E1000_RXDCTL(0):
322 for (n = 0; n < 4; n++)
323 regs[n] = rd32(E1000_RXDCTL(n));
326 for (n = 0; n < 4; n++)
327 regs[n] = rd32(E1000_RDBAL(n));
330 for (n = 0; n < 4; n++)
331 regs[n] = rd32(E1000_RDBAH(n));
334 for (n = 0; n < 4; n++)
335 regs[n] = rd32(E1000_RDBAL(n));
338 for (n = 0; n < 4; n++)
339 regs[n] = rd32(E1000_TDBAH(n));
342 for (n = 0; n < 4; n++)
343 regs[n] = rd32(E1000_TDLEN(n));
346 for (n = 0; n < 4; n++)
347 regs[n] = rd32(E1000_TDH(n));
350 for (n = 0; n < 4; n++)
351 regs[n] = rd32(E1000_TDT(n));
353 case E1000_TXDCTL(0):
354 for (n = 0; n < 4; n++)
355 regs[n] = rd32(E1000_TXDCTL(n));
358 pr_info("%-15s %08x\n", reginfo->name, rd32(reginfo->ofs));
362 snprintf(rname, 16, "%s%s", reginfo->name, "[0-3]");
363 pr_info("%-15s %08x %08x %08x %08x\n", rname, regs[0], regs[1],
367 /* igb_dump - Print registers, Tx-rings and Rx-rings */
368 static void igb_dump(struct igb_adapter *adapter)
370 struct net_device *netdev = adapter->netdev;
371 struct e1000_hw *hw = &adapter->hw;
372 struct igb_reg_info *reginfo;
373 struct igb_ring *tx_ring;
374 union e1000_adv_tx_desc *tx_desc;
375 struct my_u0 { u64 a; u64 b; } *u0;
376 struct igb_ring *rx_ring;
377 union e1000_adv_rx_desc *rx_desc;
381 if (!netif_msg_hw(adapter))
384 /* Print netdevice Info */
386 dev_info(&adapter->pdev->dev, "Net device Info\n");
387 pr_info("Device Name state trans_start "
389 pr_info("%-15s %016lX %016lX %016lX\n", netdev->name,
390 netdev->state, netdev->trans_start, netdev->last_rx);
393 /* Print Registers */
394 dev_info(&adapter->pdev->dev, "Register Dump\n");
395 pr_info(" Register Name Value\n");
396 for (reginfo = (struct igb_reg_info *)igb_reg_info_tbl;
397 reginfo->name; reginfo++) {
398 igb_regdump(hw, reginfo);
401 /* Print TX Ring Summary */
402 if (!netdev || !netif_running(netdev))
405 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
406 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
407 for (n = 0; n < adapter->num_tx_queues; n++) {
408 struct igb_tx_buffer *buffer_info;
409 tx_ring = adapter->tx_ring[n];
410 buffer_info = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
411 pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
412 n, tx_ring->next_to_use, tx_ring->next_to_clean,
413 (u64)dma_unmap_addr(buffer_info, dma),
414 dma_unmap_len(buffer_info, len),
415 buffer_info->next_to_watch,
416 (u64)buffer_info->time_stamp);
420 if (!netif_msg_tx_done(adapter))
421 goto rx_ring_summary;
423 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
425 /* Transmit Descriptor Formats
427 * Advanced Transmit Descriptor
428 * +--------------------------------------------------------------+
429 * 0 | Buffer Address [63:0] |
430 * +--------------------------------------------------------------+
431 * 8 | PAYLEN | PORTS |CC|IDX | STA | DCMD |DTYP|MAC|RSV| DTALEN |
432 * +--------------------------------------------------------------+
433 * 63 46 45 40 39 38 36 35 32 31 24 15 0
436 for (n = 0; n < adapter->num_tx_queues; n++) {
437 tx_ring = adapter->tx_ring[n];
438 pr_info("------------------------------------\n");
439 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
440 pr_info("------------------------------------\n");
441 pr_info("T [desc] [address 63:0 ] [PlPOCIStDDM Ln] "
442 "[bi->dma ] leng ntw timestamp "
445 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
446 const char *next_desc;
447 struct igb_tx_buffer *buffer_info;
448 tx_desc = IGB_TX_DESC(tx_ring, i);
449 buffer_info = &tx_ring->tx_buffer_info[i];
450 u0 = (struct my_u0 *)tx_desc;
451 if (i == tx_ring->next_to_use &&
452 i == tx_ring->next_to_clean)
453 next_desc = " NTC/U";
454 else if (i == tx_ring->next_to_use)
456 else if (i == tx_ring->next_to_clean)
461 pr_info("T [0x%03X] %016llX %016llX %016llX"
462 " %04X %p %016llX %p%s\n", i,
465 (u64)dma_unmap_addr(buffer_info, dma),
466 dma_unmap_len(buffer_info, len),
467 buffer_info->next_to_watch,
468 (u64)buffer_info->time_stamp,
469 buffer_info->skb, next_desc);
471 if (netif_msg_pktdata(adapter) && buffer_info->skb)
472 print_hex_dump(KERN_INFO, "",
474 16, 1, buffer_info->skb->data,
475 dma_unmap_len(buffer_info, len),
480 /* Print RX Rings Summary */
482 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
483 pr_info("Queue [NTU] [NTC]\n");
484 for (n = 0; n < adapter->num_rx_queues; n++) {
485 rx_ring = adapter->rx_ring[n];
486 pr_info(" %5d %5X %5X\n",
487 n, rx_ring->next_to_use, rx_ring->next_to_clean);
491 if (!netif_msg_rx_status(adapter))
494 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
496 /* Advanced Receive Descriptor (Read) Format
498 * +-----------------------------------------------------+
499 * 0 | Packet Buffer Address [63:1] |A0/NSE|
500 * +----------------------------------------------+------+
501 * 8 | Header Buffer Address [63:1] | DD |
502 * +-----------------------------------------------------+
505 * Advanced Receive Descriptor (Write-Back) Format
507 * 63 48 47 32 31 30 21 20 17 16 4 3 0
508 * +------------------------------------------------------+
509 * 0 | Packet IP |SPH| HDR_LEN | RSV|Packet| RSS |
510 * | Checksum Ident | | | | Type | Type |
511 * +------------------------------------------------------+
512 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
513 * +------------------------------------------------------+
514 * 63 48 47 32 31 20 19 0
517 for (n = 0; n < adapter->num_rx_queues; n++) {
518 rx_ring = adapter->rx_ring[n];
519 pr_info("------------------------------------\n");
520 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
521 pr_info("------------------------------------\n");
522 pr_info("R [desc] [ PktBuf A0] [ HeadBuf DD] "
523 "[bi->dma ] [bi->skb] <-- Adv Rx Read format\n");
524 pr_info("RWB[desc] [PcsmIpSHl PtRs] [vl er S cks ln] -----"
525 "----------- [bi->skb] <-- Adv Rx Write-Back format\n");
527 for (i = 0; i < rx_ring->count; i++) {
528 const char *next_desc;
529 struct igb_rx_buffer *buffer_info;
530 buffer_info = &rx_ring->rx_buffer_info[i];
531 rx_desc = IGB_RX_DESC(rx_ring, i);
532 u0 = (struct my_u0 *)rx_desc;
533 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
535 if (i == rx_ring->next_to_use)
537 else if (i == rx_ring->next_to_clean)
542 if (staterr & E1000_RXD_STAT_DD) {
543 /* Descriptor Done */
544 pr_info("%s[0x%03X] %016llX %016llX ---------------- %s\n",
550 pr_info("%s[0x%03X] %016llX %016llX %016llX %s\n",
554 (u64)buffer_info->dma,
557 if (netif_msg_pktdata(adapter) &&
558 buffer_info->dma && buffer_info->page) {
559 print_hex_dump(KERN_INFO, "",
562 page_address(buffer_info->page) +
563 buffer_info->page_offset,
575 * igb_get_i2c_data - Reads the I2C SDA data bit
576 * @hw: pointer to hardware structure
577 * @i2cctl: Current value of I2CCTL register
579 * Returns the I2C data bit value
581 static int igb_get_i2c_data(void *data)
583 struct igb_adapter *adapter = (struct igb_adapter *)data;
584 struct e1000_hw *hw = &adapter->hw;
585 s32 i2cctl = rd32(E1000_I2CPARAMS);
587 return ((i2cctl & E1000_I2C_DATA_IN) != 0);
591 * igb_set_i2c_data - Sets the I2C data bit
592 * @data: pointer to hardware structure
593 * @state: I2C data value (0 or 1) to set
595 * Sets the I2C data bit
597 static void igb_set_i2c_data(void *data, int state)
599 struct igb_adapter *adapter = (struct igb_adapter *)data;
600 struct e1000_hw *hw = &adapter->hw;
601 s32 i2cctl = rd32(E1000_I2CPARAMS);
604 i2cctl |= E1000_I2C_DATA_OUT;
606 i2cctl &= ~E1000_I2C_DATA_OUT;
608 i2cctl &= ~E1000_I2C_DATA_OE_N;
609 i2cctl |= E1000_I2C_CLK_OE_N;
610 wr32(E1000_I2CPARAMS, i2cctl);
616 * igb_set_i2c_clk - Sets the I2C SCL clock
617 * @data: pointer to hardware structure
618 * @state: state to set clock
620 * Sets the I2C clock line to state
622 static void igb_set_i2c_clk(void *data, int state)
624 struct igb_adapter *adapter = (struct igb_adapter *)data;
625 struct e1000_hw *hw = &adapter->hw;
626 s32 i2cctl = rd32(E1000_I2CPARAMS);
629 i2cctl |= E1000_I2C_CLK_OUT;
630 i2cctl &= ~E1000_I2C_CLK_OE_N;
632 i2cctl &= ~E1000_I2C_CLK_OUT;
633 i2cctl &= ~E1000_I2C_CLK_OE_N;
635 wr32(E1000_I2CPARAMS, i2cctl);
640 * igb_get_i2c_clk - Gets the I2C SCL clock state
641 * @data: pointer to hardware structure
643 * Gets the I2C clock state
645 static int igb_get_i2c_clk(void *data)
647 struct igb_adapter *adapter = (struct igb_adapter *)data;
648 struct e1000_hw *hw = &adapter->hw;
649 s32 i2cctl = rd32(E1000_I2CPARAMS);
651 return ((i2cctl & E1000_I2C_CLK_IN) != 0);
654 static const struct i2c_algo_bit_data igb_i2c_algo = {
655 .setsda = igb_set_i2c_data,
656 .setscl = igb_set_i2c_clk,
657 .getsda = igb_get_i2c_data,
658 .getscl = igb_get_i2c_clk,
664 * igb_get_hw_dev - return device
665 * @hw: pointer to hardware structure
667 * used by hardware layer to print debugging information
669 struct net_device *igb_get_hw_dev(struct e1000_hw *hw)
671 struct igb_adapter *adapter = hw->back;
672 return adapter->netdev;
676 * igb_init_module - Driver Registration Routine
678 * igb_init_module is the first routine called when the driver is
679 * loaded. All it does is register with the PCI subsystem.
681 static int __init igb_init_module(void)
684 pr_info("%s - version %s\n",
685 igb_driver_string, igb_driver_version);
687 pr_info("%s\n", igb_copyright);
689 #ifdef CONFIG_IGB_DCA
690 dca_register_notify(&dca_notifier);
692 ret = pci_register_driver(&igb_driver);
696 module_init(igb_init_module);
699 * igb_exit_module - Driver Exit Cleanup Routine
701 * igb_exit_module is called just before the driver is removed
704 static void __exit igb_exit_module(void)
706 #ifdef CONFIG_IGB_DCA
707 dca_unregister_notify(&dca_notifier);
709 pci_unregister_driver(&igb_driver);
712 module_exit(igb_exit_module);
714 #define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
716 * igb_cache_ring_register - Descriptor ring to register mapping
717 * @adapter: board private structure to initialize
719 * Once we know the feature-set enabled for the device, we'll cache
720 * the register offset the descriptor ring is assigned to.
722 static void igb_cache_ring_register(struct igb_adapter *adapter)
725 u32 rbase_offset = adapter->vfs_allocated_count;
727 switch (adapter->hw.mac.type) {
729 /* The queues are allocated for virtualization such that VF 0
730 * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
731 * In order to avoid collision we start at the first free queue
732 * and continue consuming queues in the same sequence
734 if (adapter->vfs_allocated_count) {
735 for (; i < adapter->rss_queues; i++)
736 adapter->rx_ring[i]->reg_idx = rbase_offset +
746 for (; i < adapter->num_rx_queues; i++)
747 adapter->rx_ring[i]->reg_idx = rbase_offset + i;
748 for (; j < adapter->num_tx_queues; j++)
749 adapter->tx_ring[j]->reg_idx = rbase_offset + j;
755 * igb_write_ivar - configure ivar for given MSI-X vector
756 * @hw: pointer to the HW structure
757 * @msix_vector: vector number we are allocating to a given ring
758 * @index: row index of IVAR register to write within IVAR table
759 * @offset: column offset of in IVAR, should be multiple of 8
761 * This function is intended to handle the writing of the IVAR register
762 * for adapters 82576 and newer. The IVAR table consists of 2 columns,
763 * each containing an cause allocation for an Rx and Tx ring, and a
764 * variable number of rows depending on the number of queues supported.
766 static void igb_write_ivar(struct e1000_hw *hw, int msix_vector,
767 int index, int offset)
769 u32 ivar = array_rd32(E1000_IVAR0, index);
771 /* clear any bits that are currently set */
772 ivar &= ~((u32)0xFF << offset);
774 /* write vector and valid bit */
775 ivar |= (msix_vector | E1000_IVAR_VALID) << offset;
777 array_wr32(E1000_IVAR0, index, ivar);
780 #define IGB_N0_QUEUE -1
781 static void igb_assign_vector(struct igb_q_vector *q_vector, int msix_vector)
783 struct igb_adapter *adapter = q_vector->adapter;
784 struct e1000_hw *hw = &adapter->hw;
785 int rx_queue = IGB_N0_QUEUE;
786 int tx_queue = IGB_N0_QUEUE;
789 if (q_vector->rx.ring)
790 rx_queue = q_vector->rx.ring->reg_idx;
791 if (q_vector->tx.ring)
792 tx_queue = q_vector->tx.ring->reg_idx;
794 switch (hw->mac.type) {
796 /* The 82575 assigns vectors using a bitmask, which matches the
797 * bitmask for the EICR/EIMS/EIMC registers. To assign one
798 * or more queues to a vector, we write the appropriate bits
799 * into the MSIXBM register for that vector.
801 if (rx_queue > IGB_N0_QUEUE)
802 msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
803 if (tx_queue > IGB_N0_QUEUE)
804 msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
805 if (!adapter->msix_entries && msix_vector == 0)
806 msixbm |= E1000_EIMS_OTHER;
807 array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
808 q_vector->eims_value = msixbm;
811 /* 82576 uses a table that essentially consists of 2 columns
812 * with 8 rows. The ordering is column-major so we use the
813 * lower 3 bits as the row index, and the 4th bit as the
816 if (rx_queue > IGB_N0_QUEUE)
817 igb_write_ivar(hw, msix_vector,
819 (rx_queue & 0x8) << 1);
820 if (tx_queue > IGB_N0_QUEUE)
821 igb_write_ivar(hw, msix_vector,
823 ((tx_queue & 0x8) << 1) + 8);
824 q_vector->eims_value = 1 << msix_vector;
831 /* On 82580 and newer adapters the scheme is similar to 82576
832 * however instead of ordering column-major we have things
833 * ordered row-major. So we traverse the table by using
834 * bit 0 as the column offset, and the remaining bits as the
837 if (rx_queue > IGB_N0_QUEUE)
838 igb_write_ivar(hw, msix_vector,
840 (rx_queue & 0x1) << 4);
841 if (tx_queue > IGB_N0_QUEUE)
842 igb_write_ivar(hw, msix_vector,
844 ((tx_queue & 0x1) << 4) + 8);
845 q_vector->eims_value = 1 << msix_vector;
852 /* add q_vector eims value to global eims_enable_mask */
853 adapter->eims_enable_mask |= q_vector->eims_value;
855 /* configure q_vector to set itr on first interrupt */
856 q_vector->set_itr = 1;
860 * igb_configure_msix - Configure MSI-X hardware
861 * @adapter: board private structure to initialize
863 * igb_configure_msix sets up the hardware to properly
864 * generate MSI-X interrupts.
866 static void igb_configure_msix(struct igb_adapter *adapter)
870 struct e1000_hw *hw = &adapter->hw;
872 adapter->eims_enable_mask = 0;
874 /* set vector for other causes, i.e. link changes */
875 switch (hw->mac.type) {
877 tmp = rd32(E1000_CTRL_EXT);
878 /* enable MSI-X PBA support*/
879 tmp |= E1000_CTRL_EXT_PBA_CLR;
881 /* Auto-Mask interrupts upon ICR read. */
882 tmp |= E1000_CTRL_EXT_EIAME;
883 tmp |= E1000_CTRL_EXT_IRCA;
885 wr32(E1000_CTRL_EXT, tmp);
887 /* enable msix_other interrupt */
888 array_wr32(E1000_MSIXBM(0), vector++, E1000_EIMS_OTHER);
889 adapter->eims_other = E1000_EIMS_OTHER;
899 /* Turn on MSI-X capability first, or our settings
900 * won't stick. And it will take days to debug.
902 wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
903 E1000_GPIE_PBA | E1000_GPIE_EIAME |
906 /* enable msix_other interrupt */
907 adapter->eims_other = 1 << vector;
908 tmp = (vector++ | E1000_IVAR_VALID) << 8;
910 wr32(E1000_IVAR_MISC, tmp);
913 /* do nothing, since nothing else supports MSI-X */
915 } /* switch (hw->mac.type) */
917 adapter->eims_enable_mask |= adapter->eims_other;
919 for (i = 0; i < adapter->num_q_vectors; i++)
920 igb_assign_vector(adapter->q_vector[i], vector++);
926 * igb_request_msix - Initialize MSI-X interrupts
927 * @adapter: board private structure to initialize
929 * igb_request_msix allocates MSI-X vectors and requests interrupts from the
932 static int igb_request_msix(struct igb_adapter *adapter)
934 struct net_device *netdev = adapter->netdev;
935 struct e1000_hw *hw = &adapter->hw;
936 int i, err = 0, vector = 0, free_vector = 0;
938 err = request_irq(adapter->msix_entries[vector].vector,
939 igb_msix_other, 0, netdev->name, adapter);
943 for (i = 0; i < adapter->num_q_vectors; i++) {
944 struct igb_q_vector *q_vector = adapter->q_vector[i];
948 q_vector->itr_register = hw->hw_addr + E1000_EITR(vector);
950 if (q_vector->rx.ring && q_vector->tx.ring)
951 sprintf(q_vector->name, "%s-TxRx-%u", netdev->name,
952 q_vector->rx.ring->queue_index);
953 else if (q_vector->tx.ring)
954 sprintf(q_vector->name, "%s-tx-%u", netdev->name,
955 q_vector->tx.ring->queue_index);
956 else if (q_vector->rx.ring)
957 sprintf(q_vector->name, "%s-rx-%u", netdev->name,
958 q_vector->rx.ring->queue_index);
960 sprintf(q_vector->name, "%s-unused", netdev->name);
962 err = request_irq(adapter->msix_entries[vector].vector,
963 igb_msix_ring, 0, q_vector->name,
969 igb_configure_msix(adapter);
973 /* free already assigned IRQs */
974 free_irq(adapter->msix_entries[free_vector++].vector, adapter);
977 for (i = 0; i < vector; i++) {
978 free_irq(adapter->msix_entries[free_vector++].vector,
979 adapter->q_vector[i]);
985 static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
987 if (adapter->msix_entries) {
988 pci_disable_msix(adapter->pdev);
989 kfree(adapter->msix_entries);
990 adapter->msix_entries = NULL;
991 } else if (adapter->flags & IGB_FLAG_HAS_MSI) {
992 pci_disable_msi(adapter->pdev);
997 * igb_free_q_vector - Free memory allocated for specific interrupt vector
998 * @adapter: board private structure to initialize
999 * @v_idx: Index of vector to be freed
1001 * This function frees the memory allocated to the q_vector. In addition if
1002 * NAPI is enabled it will delete any references to the NAPI struct prior
1003 * to freeing the q_vector.
1005 static void igb_free_q_vector(struct igb_adapter *adapter, int v_idx)
1007 struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
1009 if (q_vector->tx.ring)
1010 adapter->tx_ring[q_vector->tx.ring->queue_index] = NULL;
1012 if (q_vector->rx.ring)
1013 adapter->tx_ring[q_vector->rx.ring->queue_index] = NULL;
1015 adapter->q_vector[v_idx] = NULL;
1016 netif_napi_del(&q_vector->napi);
1018 /* igb_get_stats64() might access the rings on this vector,
1019 * we must wait a grace period before freeing it.
1021 kfree_rcu(q_vector, rcu);
1025 * igb_free_q_vectors - Free memory allocated for interrupt vectors
1026 * @adapter: board private structure to initialize
1028 * This function frees the memory allocated to the q_vectors. In addition if
1029 * NAPI is enabled it will delete any references to the NAPI struct prior
1030 * to freeing the q_vector.
1032 static void igb_free_q_vectors(struct igb_adapter *adapter)
1034 int v_idx = adapter->num_q_vectors;
1036 adapter->num_tx_queues = 0;
1037 adapter->num_rx_queues = 0;
1038 adapter->num_q_vectors = 0;
1041 igb_free_q_vector(adapter, v_idx);
1045 * igb_clear_interrupt_scheme - reset the device to a state of no interrupts
1046 * @adapter: board private structure to initialize
1048 * This function resets the device so that it has 0 Rx queues, Tx queues, and
1049 * MSI-X interrupts allocated.
1051 static void igb_clear_interrupt_scheme(struct igb_adapter *adapter)
1053 igb_free_q_vectors(adapter);
1054 igb_reset_interrupt_capability(adapter);
1058 * igb_set_interrupt_capability - set MSI or MSI-X if supported
1059 * @adapter: board private structure to initialize
1060 * @msix: boolean value of MSIX capability
1062 * Attempt to configure interrupts using the best available
1063 * capabilities of the hardware and kernel.
1065 static void igb_set_interrupt_capability(struct igb_adapter *adapter, bool msix)
1073 /* Number of supported queues. */
1074 adapter->num_rx_queues = adapter->rss_queues;
1075 if (adapter->vfs_allocated_count)
1076 adapter->num_tx_queues = 1;
1078 adapter->num_tx_queues = adapter->rss_queues;
1080 /* start with one vector for every Rx queue */
1081 numvecs = adapter->num_rx_queues;
1083 /* if Tx handler is separate add 1 for every Tx queue */
1084 if (!(adapter->flags & IGB_FLAG_QUEUE_PAIRS))
1085 numvecs += adapter->num_tx_queues;
1087 /* store the number of vectors reserved for queues */
1088 adapter->num_q_vectors = numvecs;
1090 /* add 1 vector for link status interrupts */
1092 adapter->msix_entries = kcalloc(numvecs, sizeof(struct msix_entry),
1095 if (!adapter->msix_entries)
1098 for (i = 0; i < numvecs; i++)
1099 adapter->msix_entries[i].entry = i;
1101 err = pci_enable_msix(adapter->pdev,
1102 adapter->msix_entries,
1107 igb_reset_interrupt_capability(adapter);
1109 /* If we can't do MSI-X, try MSI */
1111 #ifdef CONFIG_PCI_IOV
1112 /* disable SR-IOV for non MSI-X configurations */
1113 if (adapter->vf_data) {
1114 struct e1000_hw *hw = &adapter->hw;
1115 /* disable iov and allow time for transactions to clear */
1116 pci_disable_sriov(adapter->pdev);
1119 kfree(adapter->vf_data);
1120 adapter->vf_data = NULL;
1121 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
1124 dev_info(&adapter->pdev->dev, "IOV Disabled\n");
1127 adapter->vfs_allocated_count = 0;
1128 adapter->rss_queues = 1;
1129 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
1130 adapter->num_rx_queues = 1;
1131 adapter->num_tx_queues = 1;
1132 adapter->num_q_vectors = 1;
1133 if (!pci_enable_msi(adapter->pdev))
1134 adapter->flags |= IGB_FLAG_HAS_MSI;
1137 static void igb_add_ring(struct igb_ring *ring,
1138 struct igb_ring_container *head)
1145 * igb_alloc_q_vector - Allocate memory for a single interrupt vector
1146 * @adapter: board private structure to initialize
1147 * @v_count: q_vectors allocated on adapter, used for ring interleaving
1148 * @v_idx: index of vector in adapter struct
1149 * @txr_count: total number of Tx rings to allocate
1150 * @txr_idx: index of first Tx ring to allocate
1151 * @rxr_count: total number of Rx rings to allocate
1152 * @rxr_idx: index of first Rx ring to allocate
1154 * We allocate one q_vector. If allocation fails we return -ENOMEM.
1156 static int igb_alloc_q_vector(struct igb_adapter *adapter,
1157 int v_count, int v_idx,
1158 int txr_count, int txr_idx,
1159 int rxr_count, int rxr_idx)
1161 struct igb_q_vector *q_vector;
1162 struct igb_ring *ring;
1163 int ring_count, size;
1165 /* igb only supports 1 Tx and/or 1 Rx queue per vector */
1166 if (txr_count > 1 || rxr_count > 1)
1169 ring_count = txr_count + rxr_count;
1170 size = sizeof(struct igb_q_vector) +
1171 (sizeof(struct igb_ring) * ring_count);
1173 /* allocate q_vector and rings */
1174 q_vector = kzalloc(size, GFP_KERNEL);
1178 /* initialize NAPI */
1179 netif_napi_add(adapter->netdev, &q_vector->napi,
1182 /* tie q_vector and adapter together */
1183 adapter->q_vector[v_idx] = q_vector;
1184 q_vector->adapter = adapter;
1186 /* initialize work limits */
1187 q_vector->tx.work_limit = adapter->tx_work_limit;
1189 /* initialize ITR configuration */
1190 q_vector->itr_register = adapter->hw.hw_addr + E1000_EITR(0);
1191 q_vector->itr_val = IGB_START_ITR;
1193 /* initialize pointer to rings */
1194 ring = q_vector->ring;
1198 /* rx or rx/tx vector */
1199 if (!adapter->rx_itr_setting || adapter->rx_itr_setting > 3)
1200 q_vector->itr_val = adapter->rx_itr_setting;
1202 /* tx only vector */
1203 if (!adapter->tx_itr_setting || adapter->tx_itr_setting > 3)
1204 q_vector->itr_val = adapter->tx_itr_setting;
1208 /* assign generic ring traits */
1209 ring->dev = &adapter->pdev->dev;
1210 ring->netdev = adapter->netdev;
1212 /* configure backlink on ring */
1213 ring->q_vector = q_vector;
1215 /* update q_vector Tx values */
1216 igb_add_ring(ring, &q_vector->tx);
1218 /* For 82575, context index must be unique per ring. */
1219 if (adapter->hw.mac.type == e1000_82575)
1220 set_bit(IGB_RING_FLAG_TX_CTX_IDX, &ring->flags);
1222 /* apply Tx specific ring traits */
1223 ring->count = adapter->tx_ring_count;
1224 ring->queue_index = txr_idx;
1226 u64_stats_init(&ring->tx_syncp);
1227 u64_stats_init(&ring->tx_syncp2);
1229 /* assign ring to adapter */
1230 adapter->tx_ring[txr_idx] = ring;
1232 /* push pointer to next ring */
1237 /* assign generic ring traits */
1238 ring->dev = &adapter->pdev->dev;
1239 ring->netdev = adapter->netdev;
1241 /* configure backlink on ring */
1242 ring->q_vector = q_vector;
1244 /* update q_vector Rx values */
1245 igb_add_ring(ring, &q_vector->rx);
1247 /* set flag indicating ring supports SCTP checksum offload */
1248 if (adapter->hw.mac.type >= e1000_82576)
1249 set_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags);
1252 * On i350, i354, i210, and i211, loopback VLAN packets
1253 * have the tag byte-swapped.
1255 if (adapter->hw.mac.type >= e1000_i350)
1256 set_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &ring->flags);
1258 /* apply Rx specific ring traits */
1259 ring->count = adapter->rx_ring_count;
1260 ring->queue_index = rxr_idx;
1262 u64_stats_init(&ring->rx_syncp);
1264 /* assign ring to adapter */
1265 adapter->rx_ring[rxr_idx] = ring;
1273 * igb_alloc_q_vectors - Allocate memory for interrupt vectors
1274 * @adapter: board private structure to initialize
1276 * We allocate one q_vector per queue interrupt. If allocation fails we
1279 static int igb_alloc_q_vectors(struct igb_adapter *adapter)
1281 int q_vectors = adapter->num_q_vectors;
1282 int rxr_remaining = adapter->num_rx_queues;
1283 int txr_remaining = adapter->num_tx_queues;
1284 int rxr_idx = 0, txr_idx = 0, v_idx = 0;
1287 if (q_vectors >= (rxr_remaining + txr_remaining)) {
1288 for (; rxr_remaining; v_idx++) {
1289 err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1295 /* update counts and index */
1301 for (; v_idx < q_vectors; v_idx++) {
1302 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_idx);
1303 int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_idx);
1304 err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1305 tqpv, txr_idx, rqpv, rxr_idx);
1310 /* update counts and index */
1311 rxr_remaining -= rqpv;
1312 txr_remaining -= tqpv;
1320 adapter->num_tx_queues = 0;
1321 adapter->num_rx_queues = 0;
1322 adapter->num_q_vectors = 0;
1325 igb_free_q_vector(adapter, v_idx);
1331 * igb_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
1332 * @adapter: board private structure to initialize
1333 * @msix: boolean value of MSIX capability
1335 * This function initializes the interrupts and allocates all of the queues.
1337 static int igb_init_interrupt_scheme(struct igb_adapter *adapter, bool msix)
1339 struct pci_dev *pdev = adapter->pdev;
1342 igb_set_interrupt_capability(adapter, msix);
1344 err = igb_alloc_q_vectors(adapter);
1346 dev_err(&pdev->dev, "Unable to allocate memory for vectors\n");
1347 goto err_alloc_q_vectors;
1350 igb_cache_ring_register(adapter);
1354 err_alloc_q_vectors:
1355 igb_reset_interrupt_capability(adapter);
1360 * igb_request_irq - initialize interrupts
1361 * @adapter: board private structure to initialize
1363 * Attempts to configure interrupts using the best available
1364 * capabilities of the hardware and kernel.
1366 static int igb_request_irq(struct igb_adapter *adapter)
1368 struct net_device *netdev = adapter->netdev;
1369 struct pci_dev *pdev = adapter->pdev;
1372 if (adapter->msix_entries) {
1373 err = igb_request_msix(adapter);
1376 /* fall back to MSI */
1377 igb_free_all_tx_resources(adapter);
1378 igb_free_all_rx_resources(adapter);
1380 igb_clear_interrupt_scheme(adapter);
1381 err = igb_init_interrupt_scheme(adapter, false);
1385 igb_setup_all_tx_resources(adapter);
1386 igb_setup_all_rx_resources(adapter);
1387 igb_configure(adapter);
1390 igb_assign_vector(adapter->q_vector[0], 0);
1392 if (adapter->flags & IGB_FLAG_HAS_MSI) {
1393 err = request_irq(pdev->irq, igb_intr_msi, 0,
1394 netdev->name, adapter);
1398 /* fall back to legacy interrupts */
1399 igb_reset_interrupt_capability(adapter);
1400 adapter->flags &= ~IGB_FLAG_HAS_MSI;
1403 err = request_irq(pdev->irq, igb_intr, IRQF_SHARED,
1404 netdev->name, adapter);
1407 dev_err(&pdev->dev, "Error %d getting interrupt\n",
1414 static void igb_free_irq(struct igb_adapter *adapter)
1416 if (adapter->msix_entries) {
1419 free_irq(adapter->msix_entries[vector++].vector, adapter);
1421 for (i = 0; i < adapter->num_q_vectors; i++)
1422 free_irq(adapter->msix_entries[vector++].vector,
1423 adapter->q_vector[i]);
1425 free_irq(adapter->pdev->irq, adapter);
1430 * igb_irq_disable - Mask off interrupt generation on the NIC
1431 * @adapter: board private structure
1433 static void igb_irq_disable(struct igb_adapter *adapter)
1435 struct e1000_hw *hw = &adapter->hw;
1437 /* we need to be careful when disabling interrupts. The VFs are also
1438 * mapped into these registers and so clearing the bits can cause
1439 * issues on the VF drivers so we only need to clear what we set
1441 if (adapter->msix_entries) {
1442 u32 regval = rd32(E1000_EIAM);
1443 wr32(E1000_EIAM, regval & ~adapter->eims_enable_mask);
1444 wr32(E1000_EIMC, adapter->eims_enable_mask);
1445 regval = rd32(E1000_EIAC);
1446 wr32(E1000_EIAC, regval & ~adapter->eims_enable_mask);
1450 wr32(E1000_IMC, ~0);
1452 if (adapter->msix_entries) {
1454 for (i = 0; i < adapter->num_q_vectors; i++)
1455 synchronize_irq(adapter->msix_entries[i].vector);
1457 synchronize_irq(adapter->pdev->irq);
1462 * igb_irq_enable - Enable default interrupt generation settings
1463 * @adapter: board private structure
1465 static void igb_irq_enable(struct igb_adapter *adapter)
1467 struct e1000_hw *hw = &adapter->hw;
1469 if (adapter->msix_entries) {
1470 u32 ims = E1000_IMS_LSC | E1000_IMS_DOUTSYNC | E1000_IMS_DRSTA;
1471 u32 regval = rd32(E1000_EIAC);
1472 wr32(E1000_EIAC, regval | adapter->eims_enable_mask);
1473 regval = rd32(E1000_EIAM);
1474 wr32(E1000_EIAM, regval | adapter->eims_enable_mask);
1475 wr32(E1000_EIMS, adapter->eims_enable_mask);
1476 if (adapter->vfs_allocated_count) {
1477 wr32(E1000_MBVFIMR, 0xFF);
1478 ims |= E1000_IMS_VMMB;
1480 wr32(E1000_IMS, ims);
1482 wr32(E1000_IMS, IMS_ENABLE_MASK |
1484 wr32(E1000_IAM, IMS_ENABLE_MASK |
1489 static void igb_update_mng_vlan(struct igb_adapter *adapter)
1491 struct e1000_hw *hw = &adapter->hw;
1492 u16 vid = adapter->hw.mng_cookie.vlan_id;
1493 u16 old_vid = adapter->mng_vlan_id;
1495 if (hw->mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1496 /* add VID to filter table */
1497 igb_vfta_set(hw, vid, true);
1498 adapter->mng_vlan_id = vid;
1500 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1503 if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
1505 !test_bit(old_vid, adapter->active_vlans)) {
1506 /* remove VID from filter table */
1507 igb_vfta_set(hw, old_vid, false);
1512 * igb_release_hw_control - release control of the h/w to f/w
1513 * @adapter: address of board private structure
1515 * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
1516 * For ASF and Pass Through versions of f/w this means that the
1517 * driver is no longer loaded.
1519 static void igb_release_hw_control(struct igb_adapter *adapter)
1521 struct e1000_hw *hw = &adapter->hw;
1524 /* Let firmware take over control of h/w */
1525 ctrl_ext = rd32(E1000_CTRL_EXT);
1526 wr32(E1000_CTRL_EXT,
1527 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1531 * igb_get_hw_control - get control of the h/w from f/w
1532 * @adapter: address of board private structure
1534 * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
1535 * For ASF and Pass Through versions of f/w this means that
1536 * the driver is loaded.
1538 static void igb_get_hw_control(struct igb_adapter *adapter)
1540 struct e1000_hw *hw = &adapter->hw;
1543 /* Let firmware know the driver has taken over */
1544 ctrl_ext = rd32(E1000_CTRL_EXT);
1545 wr32(E1000_CTRL_EXT,
1546 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1550 * igb_configure - configure the hardware for RX and TX
1551 * @adapter: private board structure
1553 static void igb_configure(struct igb_adapter *adapter)
1555 struct net_device *netdev = adapter->netdev;
1558 igb_get_hw_control(adapter);
1559 igb_set_rx_mode(netdev);
1561 igb_restore_vlan(adapter);
1563 igb_setup_tctl(adapter);
1564 igb_setup_mrqc(adapter);
1565 igb_setup_rctl(adapter);
1567 igb_configure_tx(adapter);
1568 igb_configure_rx(adapter);
1570 igb_rx_fifo_flush_82575(&adapter->hw);
1572 /* call igb_desc_unused which always leaves
1573 * at least 1 descriptor unused to make sure
1574 * next_to_use != next_to_clean
1576 for (i = 0; i < adapter->num_rx_queues; i++) {
1577 struct igb_ring *ring = adapter->rx_ring[i];
1578 igb_alloc_rx_buffers(ring, igb_desc_unused(ring));
1583 * igb_power_up_link - Power up the phy/serdes link
1584 * @adapter: address of board private structure
1586 void igb_power_up_link(struct igb_adapter *adapter)
1588 igb_reset_phy(&adapter->hw);
1590 if (adapter->hw.phy.media_type == e1000_media_type_copper)
1591 igb_power_up_phy_copper(&adapter->hw);
1593 igb_power_up_serdes_link_82575(&adapter->hw);
1597 * igb_power_down_link - Power down the phy/serdes link
1598 * @adapter: address of board private structure
1600 static void igb_power_down_link(struct igb_adapter *adapter)
1602 if (adapter->hw.phy.media_type == e1000_media_type_copper)
1603 igb_power_down_phy_copper_82575(&adapter->hw);
1605 igb_shutdown_serdes_link_82575(&adapter->hw);
1609 * igb_up - Open the interface and prepare it to handle traffic
1610 * @adapter: board private structure
1612 int igb_up(struct igb_adapter *adapter)
1614 struct e1000_hw *hw = &adapter->hw;
1617 /* hardware has been reset, we need to reload some things */
1618 igb_configure(adapter);
1620 clear_bit(__IGB_DOWN, &adapter->state);
1622 for (i = 0; i < adapter->num_q_vectors; i++)
1623 napi_enable(&(adapter->q_vector[i]->napi));
1625 if (adapter->msix_entries)
1626 igb_configure_msix(adapter);
1628 igb_assign_vector(adapter->q_vector[0], 0);
1630 /* Clear any pending interrupts. */
1632 igb_irq_enable(adapter);
1634 /* notify VFs that reset has been completed */
1635 if (adapter->vfs_allocated_count) {
1636 u32 reg_data = rd32(E1000_CTRL_EXT);
1637 reg_data |= E1000_CTRL_EXT_PFRSTD;
1638 wr32(E1000_CTRL_EXT, reg_data);
1641 netif_tx_start_all_queues(adapter->netdev);
1643 /* start the watchdog. */
1644 hw->mac.get_link_status = 1;
1645 schedule_work(&adapter->watchdog_task);
1650 void igb_down(struct igb_adapter *adapter)
1652 struct net_device *netdev = adapter->netdev;
1653 struct e1000_hw *hw = &adapter->hw;
1657 /* signal that we're down so the interrupt handler does not
1658 * reschedule our watchdog timer
1660 set_bit(__IGB_DOWN, &adapter->state);
1662 /* disable receives in the hardware */
1663 rctl = rd32(E1000_RCTL);
1664 wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
1665 /* flush and sleep below */
1667 netif_tx_stop_all_queues(netdev);
1669 /* disable transmits in the hardware */
1670 tctl = rd32(E1000_TCTL);
1671 tctl &= ~E1000_TCTL_EN;
1672 wr32(E1000_TCTL, tctl);
1673 /* flush both disables and wait for them to finish */
1677 igb_irq_disable(adapter);
1679 adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
1681 for (i = 0; i < adapter->num_q_vectors; i++) {
1682 napi_synchronize(&(adapter->q_vector[i]->napi));
1683 napi_disable(&(adapter->q_vector[i]->napi));
1687 del_timer_sync(&adapter->watchdog_timer);
1688 del_timer_sync(&adapter->phy_info_timer);
1690 netif_carrier_off(netdev);
1692 /* record the stats before reset*/
1693 spin_lock(&adapter->stats64_lock);
1694 igb_update_stats(adapter, &adapter->stats64);
1695 spin_unlock(&adapter->stats64_lock);
1697 adapter->link_speed = 0;
1698 adapter->link_duplex = 0;
1700 if (!pci_channel_offline(adapter->pdev))
1702 igb_clean_all_tx_rings(adapter);
1703 igb_clean_all_rx_rings(adapter);
1704 #ifdef CONFIG_IGB_DCA
1706 /* since we reset the hardware DCA settings were cleared */
1707 igb_setup_dca(adapter);
1711 void igb_reinit_locked(struct igb_adapter *adapter)
1713 WARN_ON(in_interrupt());
1714 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
1718 clear_bit(__IGB_RESETTING, &adapter->state);
1721 void igb_reset(struct igb_adapter *adapter)
1723 struct pci_dev *pdev = adapter->pdev;
1724 struct e1000_hw *hw = &adapter->hw;
1725 struct e1000_mac_info *mac = &hw->mac;
1726 struct e1000_fc_info *fc = &hw->fc;
1727 u32 pba = 0, tx_space, min_tx_space, min_rx_space, hwm;
1729 /* Repartition Pba for greater than 9k mtu
1730 * To take effect CTRL.RST is required.
1732 switch (mac->type) {
1736 pba = rd32(E1000_RXPBS);
1737 pba = igb_rxpbs_adjust_82580(pba);
1740 pba = rd32(E1000_RXPBS);
1741 pba &= E1000_RXPBS_SIZE_MASK_82576;
1747 pba = E1000_PBA_34K;
1751 if ((adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) &&
1752 (mac->type < e1000_82576)) {
1753 /* adjust PBA for jumbo frames */
1754 wr32(E1000_PBA, pba);
1756 /* To maintain wire speed transmits, the Tx FIFO should be
1757 * large enough to accommodate two full transmit packets,
1758 * rounded up to the next 1KB and expressed in KB. Likewise,
1759 * the Rx FIFO should be large enough to accommodate at least
1760 * one full receive packet and is similarly rounded up and
1763 pba = rd32(E1000_PBA);
1764 /* upper 16 bits has Tx packet buffer allocation size in KB */
1765 tx_space = pba >> 16;
1766 /* lower 16 bits has Rx packet buffer allocation size in KB */
1768 /* the Tx fifo also stores 16 bytes of information about the Tx
1769 * but don't include ethernet FCS because hardware appends it
1771 min_tx_space = (adapter->max_frame_size +
1772 sizeof(union e1000_adv_tx_desc) -
1774 min_tx_space = ALIGN(min_tx_space, 1024);
1775 min_tx_space >>= 10;
1776 /* software strips receive CRC, so leave room for it */
1777 min_rx_space = adapter->max_frame_size;
1778 min_rx_space = ALIGN(min_rx_space, 1024);
1779 min_rx_space >>= 10;
1781 /* If current Tx allocation is less than the min Tx FIFO size,
1782 * and the min Tx FIFO size is less than the current Rx FIFO
1783 * allocation, take space away from current Rx allocation
1785 if (tx_space < min_tx_space &&
1786 ((min_tx_space - tx_space) < pba)) {
1787 pba = pba - (min_tx_space - tx_space);
1789 /* if short on Rx space, Rx wins and must trump Tx
1792 if (pba < min_rx_space)
1795 wr32(E1000_PBA, pba);
1798 /* flow control settings */
1799 /* The high water mark must be low enough to fit one full frame
1800 * (or the size used for early receive) above it in the Rx FIFO.
1801 * Set it to the lower of:
1802 * - 90% of the Rx FIFO size, or
1803 * - the full Rx FIFO size minus one full frame
1805 hwm = min(((pba << 10) * 9 / 10),
1806 ((pba << 10) - 2 * adapter->max_frame_size));
1808 fc->high_water = hwm & 0xFFFFFFF0; /* 16-byte granularity */
1809 fc->low_water = fc->high_water - 16;
1810 fc->pause_time = 0xFFFF;
1812 fc->current_mode = fc->requested_mode;
1814 /* disable receive for all VFs and wait one second */
1815 if (adapter->vfs_allocated_count) {
1817 for (i = 0 ; i < adapter->vfs_allocated_count; i++)
1818 adapter->vf_data[i].flags &= IGB_VF_FLAG_PF_SET_MAC;
1820 /* ping all the active vfs to let them know we are going down */
1821 igb_ping_all_vfs(adapter);
1823 /* disable transmits and receives */
1824 wr32(E1000_VFRE, 0);
1825 wr32(E1000_VFTE, 0);
1828 /* Allow time for pending master requests to run */
1829 hw->mac.ops.reset_hw(hw);
1832 if (hw->mac.ops.init_hw(hw))
1833 dev_err(&pdev->dev, "Hardware Error\n");
1835 /* Flow control settings reset on hardware reset, so guarantee flow
1836 * control is off when forcing speed.
1838 if (!hw->mac.autoneg)
1839 igb_force_mac_fc(hw);
1841 igb_init_dmac(adapter, pba);
1842 #ifdef CONFIG_IGB_HWMON
1843 /* Re-initialize the thermal sensor on i350 devices. */
1844 if (!test_bit(__IGB_DOWN, &adapter->state)) {
1845 if (mac->type == e1000_i350 && hw->bus.func == 0) {
1846 /* If present, re-initialize the external thermal sensor
1850 mac->ops.init_thermal_sensor_thresh(hw);
1854 if (!netif_running(adapter->netdev))
1855 igb_power_down_link(adapter);
1857 igb_update_mng_vlan(adapter);
1859 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
1860 wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
1862 /* Re-enable PTP, where applicable. */
1863 igb_ptp_reset(adapter);
1865 igb_get_phy_info(hw);
1868 static netdev_features_t igb_fix_features(struct net_device *netdev,
1869 netdev_features_t features)
1871 /* Since there is no support for separate Rx/Tx vlan accel
1872 * enable/disable make sure Tx flag is always in same state as Rx.
1874 if (features & NETIF_F_HW_VLAN_CTAG_RX)
1875 features |= NETIF_F_HW_VLAN_CTAG_TX;
1877 features &= ~NETIF_F_HW_VLAN_CTAG_TX;
1882 static int igb_set_features(struct net_device *netdev,
1883 netdev_features_t features)
1885 netdev_features_t changed = netdev->features ^ features;
1886 struct igb_adapter *adapter = netdev_priv(netdev);
1888 if (changed & NETIF_F_HW_VLAN_CTAG_RX)
1889 igb_vlan_mode(netdev, features);
1891 if (!(changed & NETIF_F_RXALL))
1894 netdev->features = features;
1896 if (netif_running(netdev))
1897 igb_reinit_locked(adapter);
1904 static const struct net_device_ops igb_netdev_ops = {
1905 .ndo_open = igb_open,
1906 .ndo_stop = igb_close,
1907 .ndo_start_xmit = igb_xmit_frame,
1908 .ndo_get_stats64 = igb_get_stats64,
1909 .ndo_set_rx_mode = igb_set_rx_mode,
1910 .ndo_set_mac_address = igb_set_mac,
1911 .ndo_change_mtu = igb_change_mtu,
1912 .ndo_do_ioctl = igb_ioctl,
1913 .ndo_tx_timeout = igb_tx_timeout,
1914 .ndo_validate_addr = eth_validate_addr,
1915 .ndo_vlan_rx_add_vid = igb_vlan_rx_add_vid,
1916 .ndo_vlan_rx_kill_vid = igb_vlan_rx_kill_vid,
1917 .ndo_set_vf_mac = igb_ndo_set_vf_mac,
1918 .ndo_set_vf_vlan = igb_ndo_set_vf_vlan,
1919 .ndo_set_vf_tx_rate = igb_ndo_set_vf_bw,
1920 .ndo_set_vf_spoofchk = igb_ndo_set_vf_spoofchk,
1921 .ndo_get_vf_config = igb_ndo_get_vf_config,
1922 #ifdef CONFIG_NET_POLL_CONTROLLER
1923 .ndo_poll_controller = igb_netpoll,
1925 .ndo_fix_features = igb_fix_features,
1926 .ndo_set_features = igb_set_features,
1930 * igb_set_fw_version - Configure version string for ethtool
1931 * @adapter: adapter struct
1933 void igb_set_fw_version(struct igb_adapter *adapter)
1935 struct e1000_hw *hw = &adapter->hw;
1936 struct e1000_fw_version fw;
1938 igb_get_fw_version(hw, &fw);
1940 switch (hw->mac.type) {
1943 if (!(igb_get_flash_presence_i210(hw))) {
1944 snprintf(adapter->fw_version,
1945 sizeof(adapter->fw_version),
1947 fw.invm_major, fw.invm_minor,
1953 /* if option is rom valid, display its version too */
1955 snprintf(adapter->fw_version,
1956 sizeof(adapter->fw_version),
1957 "%d.%d, 0x%08x, %d.%d.%d",
1958 fw.eep_major, fw.eep_minor, fw.etrack_id,
1959 fw.or_major, fw.or_build, fw.or_patch);
1961 } else if (fw.etrack_id != 0X0000) {
1962 snprintf(adapter->fw_version,
1963 sizeof(adapter->fw_version),
1965 fw.eep_major, fw.eep_minor, fw.etrack_id);
1967 snprintf(adapter->fw_version,
1968 sizeof(adapter->fw_version),
1970 fw.eep_major, fw.eep_minor, fw.eep_build);
1978 * igb_init_i2c - Init I2C interface
1979 * @adapter: pointer to adapter structure
1981 static s32 igb_init_i2c(struct igb_adapter *adapter)
1983 s32 status = E1000_SUCCESS;
1985 /* I2C interface supported on i350 devices */
1986 if (adapter->hw.mac.type != e1000_i350)
1987 return E1000_SUCCESS;
1989 /* Initialize the i2c bus which is controlled by the registers.
1990 * This bus will use the i2c_algo_bit structue that implements
1991 * the protocol through toggling of the 4 bits in the register.
1993 adapter->i2c_adap.owner = THIS_MODULE;
1994 adapter->i2c_algo = igb_i2c_algo;
1995 adapter->i2c_algo.data = adapter;
1996 adapter->i2c_adap.algo_data = &adapter->i2c_algo;
1997 adapter->i2c_adap.dev.parent = &adapter->pdev->dev;
1998 strlcpy(adapter->i2c_adap.name, "igb BB",
1999 sizeof(adapter->i2c_adap.name));
2000 status = i2c_bit_add_bus(&adapter->i2c_adap);
2005 * igb_probe - Device Initialization Routine
2006 * @pdev: PCI device information struct
2007 * @ent: entry in igb_pci_tbl
2009 * Returns 0 on success, negative on failure
2011 * igb_probe initializes an adapter identified by a pci_dev structure.
2012 * The OS initialization, configuring of the adapter private structure,
2013 * and a hardware reset occur.
2015 static int igb_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2017 struct net_device *netdev;
2018 struct igb_adapter *adapter;
2019 struct e1000_hw *hw;
2020 u16 eeprom_data = 0;
2022 static int global_quad_port_a; /* global quad port a indication */
2023 const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
2024 unsigned long mmio_start, mmio_len;
2025 int err, pci_using_dac;
2026 u8 part_str[E1000_PBANUM_LENGTH];
2028 /* Catch broken hardware that put the wrong VF device ID in
2029 * the PCIe SR-IOV capability.
2031 if (pdev->is_virtfn) {
2032 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
2033 pci_name(pdev), pdev->vendor, pdev->device);
2037 err = pci_enable_device_mem(pdev);
2042 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
2044 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
2048 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
2050 err = dma_set_coherent_mask(&pdev->dev,
2054 "No usable DMA configuration, aborting\n");
2060 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
2066 pci_enable_pcie_error_reporting(pdev);
2068 pci_set_master(pdev);
2069 pci_save_state(pdev);
2072 netdev = alloc_etherdev_mq(sizeof(struct igb_adapter),
2075 goto err_alloc_etherdev;
2077 SET_NETDEV_DEV(netdev, &pdev->dev);
2079 pci_set_drvdata(pdev, netdev);
2080 adapter = netdev_priv(netdev);
2081 adapter->netdev = netdev;
2082 adapter->pdev = pdev;
2085 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
2087 mmio_start = pci_resource_start(pdev, 0);
2088 mmio_len = pci_resource_len(pdev, 0);
2091 hw->hw_addr = ioremap(mmio_start, mmio_len);
2095 netdev->netdev_ops = &igb_netdev_ops;
2096 igb_set_ethtool_ops(netdev);
2097 netdev->watchdog_timeo = 5 * HZ;
2099 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2101 netdev->mem_start = mmio_start;
2102 netdev->mem_end = mmio_start + mmio_len;
2104 /* PCI config space info */
2105 hw->vendor_id = pdev->vendor;
2106 hw->device_id = pdev->device;
2107 hw->revision_id = pdev->revision;
2108 hw->subsystem_vendor_id = pdev->subsystem_vendor;
2109 hw->subsystem_device_id = pdev->subsystem_device;
2111 /* Copy the default MAC, PHY and NVM function pointers */
2112 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
2113 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
2114 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
2115 /* Initialize skew-specific constants */
2116 err = ei->get_invariants(hw);
2120 /* setup the private structure */
2121 err = igb_sw_init(adapter);
2125 igb_get_bus_info_pcie(hw);
2127 hw->phy.autoneg_wait_to_complete = false;
2129 /* Copper options */
2130 if (hw->phy.media_type == e1000_media_type_copper) {
2131 hw->phy.mdix = AUTO_ALL_MODES;
2132 hw->phy.disable_polarity_correction = false;
2133 hw->phy.ms_type = e1000_ms_hw_default;
2136 if (igb_check_reset_block(hw))
2137 dev_info(&pdev->dev,
2138 "PHY reset is blocked due to SOL/IDER session.\n");
2140 /* features is initialized to 0 in allocation, it might have bits
2141 * set by igb_sw_init so we should use an or instead of an
2144 netdev->features |= NETIF_F_SG |
2151 NETIF_F_HW_VLAN_CTAG_RX |
2152 NETIF_F_HW_VLAN_CTAG_TX;
2154 /* copy netdev features into list of user selectable features */
2155 netdev->hw_features |= netdev->features;
2156 netdev->hw_features |= NETIF_F_RXALL;
2158 /* set this bit last since it cannot be part of hw_features */
2159 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
2161 netdev->vlan_features |= NETIF_F_TSO |
2167 netdev->priv_flags |= IFF_SUPP_NOFCS;
2169 if (pci_using_dac) {
2170 netdev->features |= NETIF_F_HIGHDMA;
2171 netdev->vlan_features |= NETIF_F_HIGHDMA;
2174 if (hw->mac.type >= e1000_82576) {
2175 netdev->hw_features |= NETIF_F_SCTP_CSUM;
2176 netdev->features |= NETIF_F_SCTP_CSUM;
2179 netdev->priv_flags |= IFF_UNICAST_FLT;
2181 adapter->en_mng_pt = igb_enable_mng_pass_thru(hw);
2183 /* before reading the NVM, reset the controller to put the device in a
2184 * known good starting state
2186 hw->mac.ops.reset_hw(hw);
2188 /* make sure the NVM is good , i211/i210 parts can have special NVM
2189 * that doesn't contain a checksum
2191 switch (hw->mac.type) {
2194 if (igb_get_flash_presence_i210(hw)) {
2195 if (hw->nvm.ops.validate(hw) < 0) {
2197 "The NVM Checksum Is Not Valid\n");
2204 if (hw->nvm.ops.validate(hw) < 0) {
2205 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
2212 /* copy the MAC address out of the NVM */
2213 if (hw->mac.ops.read_mac_addr(hw))
2214 dev_err(&pdev->dev, "NVM Read Error\n");
2216 memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
2218 if (!is_valid_ether_addr(netdev->dev_addr)) {
2219 dev_err(&pdev->dev, "Invalid MAC Address\n");
2224 /* get firmware version for ethtool -i */
2225 igb_set_fw_version(adapter);
2227 setup_timer(&adapter->watchdog_timer, igb_watchdog,
2228 (unsigned long) adapter);
2229 setup_timer(&adapter->phy_info_timer, igb_update_phy_info,
2230 (unsigned long) adapter);
2232 INIT_WORK(&adapter->reset_task, igb_reset_task);
2233 INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
2235 /* Initialize link properties that are user-changeable */
2236 adapter->fc_autoneg = true;
2237 hw->mac.autoneg = true;
2238 hw->phy.autoneg_advertised = 0x2f;
2240 hw->fc.requested_mode = e1000_fc_default;
2241 hw->fc.current_mode = e1000_fc_default;
2243 igb_validate_mdi_setting(hw);
2245 /* By default, support wake on port A */
2246 if (hw->bus.func == 0)
2247 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2249 /* Check the NVM for wake support on non-port A ports */
2250 if (hw->mac.type >= e1000_82580)
2251 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A +
2252 NVM_82580_LAN_FUNC_OFFSET(hw->bus.func), 1,
2254 else if (hw->bus.func == 1)
2255 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
2257 if (eeprom_data & IGB_EEPROM_APME)
2258 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2260 /* now that we have the eeprom settings, apply the special cases where
2261 * the eeprom may be wrong or the board simply won't support wake on
2262 * lan on a particular port
2264 switch (pdev->device) {
2265 case E1000_DEV_ID_82575GB_QUAD_COPPER:
2266 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2268 case E1000_DEV_ID_82575EB_FIBER_SERDES:
2269 case E1000_DEV_ID_82576_FIBER:
2270 case E1000_DEV_ID_82576_SERDES:
2271 /* Wake events only supported on port A for dual fiber
2272 * regardless of eeprom setting
2274 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
2275 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2277 case E1000_DEV_ID_82576_QUAD_COPPER:
2278 case E1000_DEV_ID_82576_QUAD_COPPER_ET2:
2279 /* if quad port adapter, disable WoL on all but port A */
2280 if (global_quad_port_a != 0)
2281 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2283 adapter->flags |= IGB_FLAG_QUAD_PORT_A;
2284 /* Reset for multiple quad port adapters */
2285 if (++global_quad_port_a == 4)
2286 global_quad_port_a = 0;
2289 /* If the device can't wake, don't set software support */
2290 if (!device_can_wakeup(&adapter->pdev->dev))
2291 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2294 /* initialize the wol settings based on the eeprom settings */
2295 if (adapter->flags & IGB_FLAG_WOL_SUPPORTED)
2296 adapter->wol |= E1000_WUFC_MAG;
2298 /* Some vendors want WoL disabled by default, but still supported */
2299 if ((hw->mac.type == e1000_i350) &&
2300 (pdev->subsystem_vendor == PCI_VENDOR_ID_HP)) {
2301 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2305 device_set_wakeup_enable(&adapter->pdev->dev,
2306 adapter->flags & IGB_FLAG_WOL_SUPPORTED);
2308 /* reset the hardware with the new settings */
2311 /* Init the I2C interface */
2312 err = igb_init_i2c(adapter);
2314 dev_err(&pdev->dev, "failed to init i2c interface\n");
2318 /* let the f/w know that the h/w is now under the control of the
2320 igb_get_hw_control(adapter);
2322 strcpy(netdev->name, "eth%d");
2323 err = register_netdev(netdev);
2327 /* carrier off reporting is important to ethtool even BEFORE open */
2328 netif_carrier_off(netdev);
2330 #ifdef CONFIG_IGB_DCA
2331 if (dca_add_requester(&pdev->dev) == 0) {
2332 adapter->flags |= IGB_FLAG_DCA_ENABLED;
2333 dev_info(&pdev->dev, "DCA enabled\n");
2334 igb_setup_dca(adapter);
2338 #ifdef CONFIG_IGB_HWMON
2339 /* Initialize the thermal sensor on i350 devices. */
2340 if (hw->mac.type == e1000_i350 && hw->bus.func == 0) {
2343 /* Read the NVM to determine if this i350 device supports an
2344 * external thermal sensor.
2346 hw->nvm.ops.read(hw, NVM_ETS_CFG, 1, &ets_word);
2347 if (ets_word != 0x0000 && ets_word != 0xFFFF)
2348 adapter->ets = true;
2350 adapter->ets = false;
2351 if (igb_sysfs_init(adapter))
2353 "failed to allocate sysfs resources\n");
2355 adapter->ets = false;
2358 /* do hw tstamp init after resetting */
2359 igb_ptp_init(adapter);
2361 dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
2362 /* print bus type/speed/width info, not applicable to i354 */
2363 if (hw->mac.type != e1000_i354) {
2364 dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
2366 ((hw->bus.speed == e1000_bus_speed_2500) ? "2.5Gb/s" :
2367 (hw->bus.speed == e1000_bus_speed_5000) ? "5.0Gb/s" :
2369 ((hw->bus.width == e1000_bus_width_pcie_x4) ?
2371 (hw->bus.width == e1000_bus_width_pcie_x2) ?
2373 (hw->bus.width == e1000_bus_width_pcie_x1) ?
2374 "Width x1" : "unknown"), netdev->dev_addr);
2377 if ((hw->mac.type >= e1000_i210 ||
2378 igb_get_flash_presence_i210(hw))) {
2379 ret_val = igb_read_part_string(hw, part_str,
2380 E1000_PBANUM_LENGTH);
2382 ret_val = -E1000_ERR_INVM_VALUE_NOT_FOUND;
2386 strcpy(part_str, "Unknown");
2387 dev_info(&pdev->dev, "%s: PBA No: %s\n", netdev->name, part_str);
2388 dev_info(&pdev->dev,
2389 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
2390 adapter->msix_entries ? "MSI-X" :
2391 (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
2392 adapter->num_rx_queues, adapter->num_tx_queues);
2393 switch (hw->mac.type) {
2397 igb_set_eee_i350(hw);
2400 if (hw->phy.media_type == e1000_media_type_copper) {
2401 if ((rd32(E1000_CTRL_EXT) &
2402 E1000_CTRL_EXT_LINK_MODE_SGMII))
2403 igb_set_eee_i354(hw);
2410 pm_runtime_put_noidle(&pdev->dev);
2414 igb_release_hw_control(adapter);
2415 memset(&adapter->i2c_adap, 0, sizeof(adapter->i2c_adap));
2417 if (!igb_check_reset_block(hw))
2420 if (hw->flash_address)
2421 iounmap(hw->flash_address);
2423 igb_clear_interrupt_scheme(adapter);
2424 iounmap(hw->hw_addr);
2426 free_netdev(netdev);
2428 pci_release_selected_regions(pdev,
2429 pci_select_bars(pdev, IORESOURCE_MEM));
2432 pci_disable_device(pdev);
2436 #ifdef CONFIG_PCI_IOV
2437 static int igb_disable_sriov(struct pci_dev *pdev)
2439 struct net_device *netdev = pci_get_drvdata(pdev);
2440 struct igb_adapter *adapter = netdev_priv(netdev);
2441 struct e1000_hw *hw = &adapter->hw;
2443 /* reclaim resources allocated to VFs */
2444 if (adapter->vf_data) {
2445 /* disable iov and allow time for transactions to clear */
2446 if (pci_vfs_assigned(pdev)) {
2447 dev_warn(&pdev->dev,
2448 "Cannot deallocate SR-IOV virtual functions while they are assigned - VFs will not be deallocated\n");
2451 pci_disable_sriov(pdev);
2455 kfree(adapter->vf_data);
2456 adapter->vf_data = NULL;
2457 adapter->vfs_allocated_count = 0;
2458 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
2461 dev_info(&pdev->dev, "IOV Disabled\n");
2463 /* Re-enable DMA Coalescing flag since IOV is turned off */
2464 adapter->flags |= IGB_FLAG_DMAC;
2470 static int igb_enable_sriov(struct pci_dev *pdev, int num_vfs)
2472 struct net_device *netdev = pci_get_drvdata(pdev);
2473 struct igb_adapter *adapter = netdev_priv(netdev);
2474 int old_vfs = pci_num_vf(pdev);
2478 if (!adapter->msix_entries) {
2485 else if (old_vfs && old_vfs == num_vfs)
2487 else if (old_vfs && old_vfs != num_vfs)
2488 err = igb_disable_sriov(pdev);
2498 adapter->vfs_allocated_count = num_vfs;
2500 adapter->vf_data = kcalloc(adapter->vfs_allocated_count,
2501 sizeof(struct vf_data_storage), GFP_KERNEL);
2503 /* if allocation failed then we do not support SR-IOV */
2504 if (!adapter->vf_data) {
2505 adapter->vfs_allocated_count = 0;
2507 "Unable to allocate memory for VF Data Storage\n");
2512 err = pci_enable_sriov(pdev, adapter->vfs_allocated_count);
2516 dev_info(&pdev->dev, "%d VFs allocated\n",
2517 adapter->vfs_allocated_count);
2518 for (i = 0; i < adapter->vfs_allocated_count; i++)
2519 igb_vf_configure(adapter, i);
2521 /* DMA Coalescing is not supported in IOV mode. */
2522 adapter->flags &= ~IGB_FLAG_DMAC;
2526 kfree(adapter->vf_data);
2527 adapter->vf_data = NULL;
2528 adapter->vfs_allocated_count = 0;
2535 * igb_remove_i2c - Cleanup I2C interface
2536 * @adapter: pointer to adapter structure
2538 static void igb_remove_i2c(struct igb_adapter *adapter)
2540 /* free the adapter bus structure */
2541 i2c_del_adapter(&adapter->i2c_adap);
2545 * igb_remove - Device Removal Routine
2546 * @pdev: PCI device information struct
2548 * igb_remove is called by the PCI subsystem to alert the driver
2549 * that it should release a PCI device. The could be caused by a
2550 * Hot-Plug event, or because the driver is going to be removed from
2553 static void igb_remove(struct pci_dev *pdev)
2555 struct net_device *netdev = pci_get_drvdata(pdev);
2556 struct igb_adapter *adapter = netdev_priv(netdev);
2557 struct e1000_hw *hw = &adapter->hw;
2559 pm_runtime_get_noresume(&pdev->dev);
2560 #ifdef CONFIG_IGB_HWMON
2561 igb_sysfs_exit(adapter);
2563 igb_remove_i2c(adapter);
2564 igb_ptp_stop(adapter);
2565 /* The watchdog timer may be rescheduled, so explicitly
2566 * disable watchdog from being rescheduled.
2568 set_bit(__IGB_DOWN, &adapter->state);
2569 del_timer_sync(&adapter->watchdog_timer);
2570 del_timer_sync(&adapter->phy_info_timer);
2572 cancel_work_sync(&adapter->reset_task);
2573 cancel_work_sync(&adapter->watchdog_task);
2575 #ifdef CONFIG_IGB_DCA
2576 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
2577 dev_info(&pdev->dev, "DCA disabled\n");
2578 dca_remove_requester(&pdev->dev);
2579 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
2580 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
2584 /* Release control of h/w to f/w. If f/w is AMT enabled, this
2585 * would have already happened in close and is redundant.
2587 igb_release_hw_control(adapter);
2589 unregister_netdev(netdev);
2591 igb_clear_interrupt_scheme(adapter);
2593 #ifdef CONFIG_PCI_IOV
2594 igb_disable_sriov(pdev);
2597 iounmap(hw->hw_addr);
2598 if (hw->flash_address)
2599 iounmap(hw->flash_address);
2600 pci_release_selected_regions(pdev,
2601 pci_select_bars(pdev, IORESOURCE_MEM));
2603 kfree(adapter->shadow_vfta);
2604 free_netdev(netdev);
2606 pci_disable_pcie_error_reporting(pdev);
2608 pci_disable_device(pdev);
2612 * igb_probe_vfs - Initialize vf data storage and add VFs to pci config space
2613 * @adapter: board private structure to initialize
2615 * This function initializes the vf specific data storage and then attempts to
2616 * allocate the VFs. The reason for ordering it this way is because it is much
2617 * mor expensive time wise to disable SR-IOV than it is to allocate and free
2618 * the memory for the VFs.
2620 static void igb_probe_vfs(struct igb_adapter *adapter)
2622 #ifdef CONFIG_PCI_IOV
2623 struct pci_dev *pdev = adapter->pdev;
2624 struct e1000_hw *hw = &adapter->hw;
2626 /* Virtualization features not supported on i210 family. */
2627 if ((hw->mac.type == e1000_i210) || (hw->mac.type == e1000_i211))
2630 pci_sriov_set_totalvfs(pdev, 7);
2631 igb_enable_sriov(pdev, max_vfs);
2633 #endif /* CONFIG_PCI_IOV */
2636 static void igb_init_queue_configuration(struct igb_adapter *adapter)
2638 struct e1000_hw *hw = &adapter->hw;
2641 /* Determine the maximum number of RSS queues supported. */
2642 switch (hw->mac.type) {
2644 max_rss_queues = IGB_MAX_RX_QUEUES_I211;
2648 max_rss_queues = IGB_MAX_RX_QUEUES_82575;
2651 /* I350 cannot do RSS and SR-IOV at the same time */
2652 if (!!adapter->vfs_allocated_count) {
2658 if (!!adapter->vfs_allocated_count) {
2666 max_rss_queues = IGB_MAX_RX_QUEUES;
2670 adapter->rss_queues = min_t(u32, max_rss_queues, num_online_cpus());
2672 /* Determine if we need to pair queues. */
2673 switch (hw->mac.type) {
2676 /* Device supports enough interrupts without queue pairing. */
2679 /* If VFs are going to be allocated with RSS queues then we
2680 * should pair the queues in order to conserve interrupts due
2681 * to limited supply.
2683 if ((adapter->rss_queues > 1) &&
2684 (adapter->vfs_allocated_count > 6))
2685 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
2692 /* If rss_queues > half of max_rss_queues, pair the queues in
2693 * order to conserve interrupts due to limited supply.
2695 if (adapter->rss_queues > (max_rss_queues / 2))
2696 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
2702 * igb_sw_init - Initialize general software structures (struct igb_adapter)
2703 * @adapter: board private structure to initialize
2705 * igb_sw_init initializes the Adapter private data structure.
2706 * Fields are initialized based on PCI device information and
2707 * OS network device settings (MTU size).
2709 static int igb_sw_init(struct igb_adapter *adapter)
2711 struct e1000_hw *hw = &adapter->hw;
2712 struct net_device *netdev = adapter->netdev;
2713 struct pci_dev *pdev = adapter->pdev;
2715 pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
2717 /* set default ring sizes */
2718 adapter->tx_ring_count = IGB_DEFAULT_TXD;
2719 adapter->rx_ring_count = IGB_DEFAULT_RXD;
2721 /* set default ITR values */
2722 adapter->rx_itr_setting = IGB_DEFAULT_ITR;
2723 adapter->tx_itr_setting = IGB_DEFAULT_ITR;
2725 /* set default work limits */
2726 adapter->tx_work_limit = IGB_DEFAULT_TX_WORK;
2728 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN +
2730 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2732 spin_lock_init(&adapter->stats64_lock);
2733 #ifdef CONFIG_PCI_IOV
2734 switch (hw->mac.type) {
2738 dev_warn(&pdev->dev,
2739 "Maximum of 7 VFs per PF, using max\n");
2740 max_vfs = adapter->vfs_allocated_count = 7;
2742 adapter->vfs_allocated_count = max_vfs;
2743 if (adapter->vfs_allocated_count)
2744 dev_warn(&pdev->dev,
2745 "Enabling SR-IOV VFs using the module parameter is deprecated - please use the pci sysfs interface.\n");
2750 #endif /* CONFIG_PCI_IOV */
2752 igb_init_queue_configuration(adapter);
2754 /* Setup and initialize a copy of the hw vlan table array */
2755 adapter->shadow_vfta = kcalloc(E1000_VLAN_FILTER_TBL_SIZE, sizeof(u32),
2758 /* This call may decrease the number of queues */
2759 if (igb_init_interrupt_scheme(adapter, true)) {
2760 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
2764 igb_probe_vfs(adapter);
2766 /* Explicitly disable IRQ since the NIC can be in any state. */
2767 igb_irq_disable(adapter);
2769 if (hw->mac.type >= e1000_i350)
2770 adapter->flags &= ~IGB_FLAG_DMAC;
2772 set_bit(__IGB_DOWN, &adapter->state);
2777 * igb_open - Called when a network interface is made active
2778 * @netdev: network interface device structure
2780 * Returns 0 on success, negative value on failure
2782 * The open entry point is called when a network interface is made
2783 * active by the system (IFF_UP). At this point all resources needed
2784 * for transmit and receive operations are allocated, the interrupt
2785 * handler is registered with the OS, the watchdog timer is started,
2786 * and the stack is notified that the interface is ready.
2788 static int __igb_open(struct net_device *netdev, bool resuming)
2790 struct igb_adapter *adapter = netdev_priv(netdev);
2791 struct e1000_hw *hw = &adapter->hw;
2792 struct pci_dev *pdev = adapter->pdev;
2796 /* disallow open during test */
2797 if (test_bit(__IGB_TESTING, &adapter->state)) {
2803 pm_runtime_get_sync(&pdev->dev);
2805 netif_carrier_off(netdev);
2807 /* allocate transmit descriptors */
2808 err = igb_setup_all_tx_resources(adapter);
2812 /* allocate receive descriptors */
2813 err = igb_setup_all_rx_resources(adapter);
2817 igb_power_up_link(adapter);
2819 /* before we allocate an interrupt, we must be ready to handle it.
2820 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
2821 * as soon as we call pci_request_irq, so we have to setup our
2822 * clean_rx handler before we do so.
2824 igb_configure(adapter);
2826 err = igb_request_irq(adapter);
2830 /* Notify the stack of the actual queue counts. */
2831 err = netif_set_real_num_tx_queues(adapter->netdev,
2832 adapter->num_tx_queues);
2834 goto err_set_queues;
2836 err = netif_set_real_num_rx_queues(adapter->netdev,
2837 adapter->num_rx_queues);
2839 goto err_set_queues;
2841 /* From here on the code is the same as igb_up() */
2842 clear_bit(__IGB_DOWN, &adapter->state);
2844 for (i = 0; i < adapter->num_q_vectors; i++)
2845 napi_enable(&(adapter->q_vector[i]->napi));
2847 /* Clear any pending interrupts. */
2850 igb_irq_enable(adapter);
2852 /* notify VFs that reset has been completed */
2853 if (adapter->vfs_allocated_count) {
2854 u32 reg_data = rd32(E1000_CTRL_EXT);
2855 reg_data |= E1000_CTRL_EXT_PFRSTD;
2856 wr32(E1000_CTRL_EXT, reg_data);
2859 netif_tx_start_all_queues(netdev);
2862 pm_runtime_put(&pdev->dev);
2864 /* start the watchdog. */
2865 hw->mac.get_link_status = 1;
2866 schedule_work(&adapter->watchdog_task);
2871 igb_free_irq(adapter);
2873 igb_release_hw_control(adapter);
2874 igb_power_down_link(adapter);
2875 igb_free_all_rx_resources(adapter);
2877 igb_free_all_tx_resources(adapter);
2881 pm_runtime_put(&pdev->dev);
2886 static int igb_open(struct net_device *netdev)
2888 return __igb_open(netdev, false);
2892 * igb_close - Disables a network interface
2893 * @netdev: network interface device structure
2895 * Returns 0, this is not allowed to fail
2897 * The close entry point is called when an interface is de-activated
2898 * by the OS. The hardware is still under the driver's control, but
2899 * needs to be disabled. A global MAC reset is issued to stop the
2900 * hardware, and all transmit and receive resources are freed.
2902 static int __igb_close(struct net_device *netdev, bool suspending)
2904 struct igb_adapter *adapter = netdev_priv(netdev);
2905 struct pci_dev *pdev = adapter->pdev;
2907 WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
2910 pm_runtime_get_sync(&pdev->dev);
2913 igb_free_irq(adapter);
2915 igb_free_all_tx_resources(adapter);
2916 igb_free_all_rx_resources(adapter);
2919 pm_runtime_put_sync(&pdev->dev);
2923 static int igb_close(struct net_device *netdev)
2925 return __igb_close(netdev, false);
2929 * igb_setup_tx_resources - allocate Tx resources (Descriptors)
2930 * @tx_ring: tx descriptor ring (for a specific queue) to setup
2932 * Return 0 on success, negative on failure
2934 int igb_setup_tx_resources(struct igb_ring *tx_ring)
2936 struct device *dev = tx_ring->dev;
2939 size = sizeof(struct igb_tx_buffer) * tx_ring->count;
2941 tx_ring->tx_buffer_info = vzalloc(size);
2942 if (!tx_ring->tx_buffer_info)
2945 /* round up to nearest 4K */
2946 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
2947 tx_ring->size = ALIGN(tx_ring->size, 4096);
2949 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
2950 &tx_ring->dma, GFP_KERNEL);
2954 tx_ring->next_to_use = 0;
2955 tx_ring->next_to_clean = 0;
2960 vfree(tx_ring->tx_buffer_info);
2961 tx_ring->tx_buffer_info = NULL;
2962 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
2967 * igb_setup_all_tx_resources - wrapper to allocate Tx resources
2968 * (Descriptors) for all queues
2969 * @adapter: board private structure
2971 * Return 0 on success, negative on failure
2973 static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
2975 struct pci_dev *pdev = adapter->pdev;
2978 for (i = 0; i < adapter->num_tx_queues; i++) {
2979 err = igb_setup_tx_resources(adapter->tx_ring[i]);
2982 "Allocation for Tx Queue %u failed\n", i);
2983 for (i--; i >= 0; i--)
2984 igb_free_tx_resources(adapter->tx_ring[i]);
2993 * igb_setup_tctl - configure the transmit control registers
2994 * @adapter: Board private structure
2996 void igb_setup_tctl(struct igb_adapter *adapter)
2998 struct e1000_hw *hw = &adapter->hw;
3001 /* disable queue 0 which is enabled by default on 82575 and 82576 */
3002 wr32(E1000_TXDCTL(0), 0);
3004 /* Program the Transmit Control Register */
3005 tctl = rd32(E1000_TCTL);
3006 tctl &= ~E1000_TCTL_CT;
3007 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
3008 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
3010 igb_config_collision_dist(hw);
3012 /* Enable transmits */
3013 tctl |= E1000_TCTL_EN;
3015 wr32(E1000_TCTL, tctl);
3019 * igb_configure_tx_ring - Configure transmit ring after Reset
3020 * @adapter: board private structure
3021 * @ring: tx ring to configure
3023 * Configure a transmit ring after a reset.
3025 void igb_configure_tx_ring(struct igb_adapter *adapter,
3026 struct igb_ring *ring)
3028 struct e1000_hw *hw = &adapter->hw;
3030 u64 tdba = ring->dma;
3031 int reg_idx = ring->reg_idx;
3033 /* disable the queue */
3034 wr32(E1000_TXDCTL(reg_idx), 0);
3038 wr32(E1000_TDLEN(reg_idx),
3039 ring->count * sizeof(union e1000_adv_tx_desc));
3040 wr32(E1000_TDBAL(reg_idx),
3041 tdba & 0x00000000ffffffffULL);
3042 wr32(E1000_TDBAH(reg_idx), tdba >> 32);
3044 ring->tail = hw->hw_addr + E1000_TDT(reg_idx);
3045 wr32(E1000_TDH(reg_idx), 0);
3046 writel(0, ring->tail);
3048 txdctl |= IGB_TX_PTHRESH;
3049 txdctl |= IGB_TX_HTHRESH << 8;
3050 txdctl |= IGB_TX_WTHRESH << 16;
3052 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
3053 wr32(E1000_TXDCTL(reg_idx), txdctl);
3057 * igb_configure_tx - Configure transmit Unit after Reset
3058 * @adapter: board private structure
3060 * Configure the Tx unit of the MAC after a reset.
3062 static void igb_configure_tx(struct igb_adapter *adapter)
3066 for (i = 0; i < adapter->num_tx_queues; i++)
3067 igb_configure_tx_ring(adapter, adapter->tx_ring[i]);
3071 * igb_setup_rx_resources - allocate Rx resources (Descriptors)
3072 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
3074 * Returns 0 on success, negative on failure
3076 int igb_setup_rx_resources(struct igb_ring *rx_ring)
3078 struct device *dev = rx_ring->dev;
3081 size = sizeof(struct igb_rx_buffer) * rx_ring->count;
3083 rx_ring->rx_buffer_info = vzalloc(size);
3084 if (!rx_ring->rx_buffer_info)
3087 /* Round up to nearest 4K */
3088 rx_ring->size = rx_ring->count * sizeof(union e1000_adv_rx_desc);
3089 rx_ring->size = ALIGN(rx_ring->size, 4096);
3091 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
3092 &rx_ring->dma, GFP_KERNEL);
3096 rx_ring->next_to_alloc = 0;
3097 rx_ring->next_to_clean = 0;
3098 rx_ring->next_to_use = 0;
3103 vfree(rx_ring->rx_buffer_info);
3104 rx_ring->rx_buffer_info = NULL;
3105 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
3110 * igb_setup_all_rx_resources - wrapper to allocate Rx resources
3111 * (Descriptors) for all queues
3112 * @adapter: board private structure
3114 * Return 0 on success, negative on failure
3116 static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
3118 struct pci_dev *pdev = adapter->pdev;
3121 for (i = 0; i < adapter->num_rx_queues; i++) {
3122 err = igb_setup_rx_resources(adapter->rx_ring[i]);
3125 "Allocation for Rx Queue %u failed\n", i);
3126 for (i--; i >= 0; i--)
3127 igb_free_rx_resources(adapter->rx_ring[i]);
3136 * igb_setup_mrqc - configure the multiple receive queue control registers
3137 * @adapter: Board private structure
3139 static void igb_setup_mrqc(struct igb_adapter *adapter)
3141 struct e1000_hw *hw = &adapter->hw;
3143 u32 j, num_rx_queues;
3144 static const u32 rsskey[10] = { 0xDA565A6D, 0xC20E5B25, 0x3D256741,
3145 0xB08FA343, 0xCB2BCAD0, 0xB4307BAE,
3146 0xA32DCB77, 0x0CF23080, 0x3BB7426A,
3149 /* Fill out hash function seeds */
3150 for (j = 0; j < 10; j++)
3151 wr32(E1000_RSSRK(j), rsskey[j]);
3153 num_rx_queues = adapter->rss_queues;
3155 switch (hw->mac.type) {
3157 /* 82576 supports 2 RSS queues for SR-IOV */
3158 if (adapter->vfs_allocated_count)
3165 if (adapter->rss_indir_tbl_init != num_rx_queues) {
3166 for (j = 0; j < IGB_RETA_SIZE; j++)
3167 adapter->rss_indir_tbl[j] = (j * num_rx_queues) / IGB_RETA_SIZE;
3168 adapter->rss_indir_tbl_init = num_rx_queues;
3170 igb_write_rss_indir_tbl(adapter);
3172 /* Disable raw packet checksumming so that RSS hash is placed in
3173 * descriptor on writeback. No need to enable TCP/UDP/IP checksum
3174 * offloads as they are enabled by default
3176 rxcsum = rd32(E1000_RXCSUM);
3177 rxcsum |= E1000_RXCSUM_PCSD;
3179 if (adapter->hw.mac.type >= e1000_82576)
3180 /* Enable Receive Checksum Offload for SCTP */
3181 rxcsum |= E1000_RXCSUM_CRCOFL;
3183 /* Don't need to set TUOFL or IPOFL, they default to 1 */
3184 wr32(E1000_RXCSUM, rxcsum);
3186 /* Generate RSS hash based on packet types, TCP/UDP
3187 * port numbers and/or IPv4/v6 src and dst addresses
3189 mrqc = E1000_MRQC_RSS_FIELD_IPV4 |
3190 E1000_MRQC_RSS_FIELD_IPV4_TCP |
3191 E1000_MRQC_RSS_FIELD_IPV6 |
3192 E1000_MRQC_RSS_FIELD_IPV6_TCP |
3193 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX;
3195 if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV4_UDP)
3196 mrqc |= E1000_MRQC_RSS_FIELD_IPV4_UDP;
3197 if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV6_UDP)
3198 mrqc |= E1000_MRQC_RSS_FIELD_IPV6_UDP;
3200 /* If VMDq is enabled then we set the appropriate mode for that, else
3201 * we default to RSS so that an RSS hash is calculated per packet even
3202 * if we are only using one queue
3204 if (adapter->vfs_allocated_count) {
3205 if (hw->mac.type > e1000_82575) {
3206 /* Set the default pool for the PF's first queue */
3207 u32 vtctl = rd32(E1000_VT_CTL);
3208 vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK |
3209 E1000_VT_CTL_DISABLE_DEF_POOL);
3210 vtctl |= adapter->vfs_allocated_count <<
3211 E1000_VT_CTL_DEFAULT_POOL_SHIFT;
3212 wr32(E1000_VT_CTL, vtctl);
3214 if (adapter->rss_queues > 1)
3215 mrqc |= E1000_MRQC_ENABLE_VMDQ_RSS_2Q;
3217 mrqc |= E1000_MRQC_ENABLE_VMDQ;
3219 if (hw->mac.type != e1000_i211)
3220 mrqc |= E1000_MRQC_ENABLE_RSS_4Q;
3222 igb_vmm_control(adapter);
3224 wr32(E1000_MRQC, mrqc);
3228 * igb_setup_rctl - configure the receive control registers
3229 * @adapter: Board private structure
3231 void igb_setup_rctl(struct igb_adapter *adapter)
3233 struct e1000_hw *hw = &adapter->hw;
3236 rctl = rd32(E1000_RCTL);
3238 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
3239 rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
3241 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
3242 (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
3244 /* enable stripping of CRC. It's unlikely this will break BMC
3245 * redirection as it did with e1000. Newer features require
3246 * that the HW strips the CRC.
3248 rctl |= E1000_RCTL_SECRC;
3250 /* disable store bad packets and clear size bits. */
3251 rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
3253 /* enable LPE to prevent packets larger than max_frame_size */
3254 rctl |= E1000_RCTL_LPE;
3256 /* disable queue 0 to prevent tail write w/o re-config */
3257 wr32(E1000_RXDCTL(0), 0);
3259 /* Attention!!! For SR-IOV PF driver operations you must enable
3260 * queue drop for all VF and PF queues to prevent head of line blocking
3261 * if an un-trusted VF does not provide descriptors to hardware.
3263 if (adapter->vfs_allocated_count) {
3264 /* set all queue drop enable bits */
3265 wr32(E1000_QDE, ALL_QUEUES);
3268 /* This is useful for sniffing bad packets. */
3269 if (adapter->netdev->features & NETIF_F_RXALL) {
3270 /* UPE and MPE will be handled by normal PROMISC logic
3271 * in e1000e_set_rx_mode
3273 rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
3274 E1000_RCTL_BAM | /* RX All Bcast Pkts */
3275 E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
3277 rctl &= ~(E1000_RCTL_VFE | /* Disable VLAN filter */
3278 E1000_RCTL_DPF | /* Allow filtered pause */
3279 E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
3280 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
3281 * and that breaks VLANs.
3285 wr32(E1000_RCTL, rctl);
3288 static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
3291 struct e1000_hw *hw = &adapter->hw;
3294 /* if it isn't the PF check to see if VFs are enabled and
3295 * increase the size to support vlan tags
3297 if (vfn < adapter->vfs_allocated_count &&
3298 adapter->vf_data[vfn].vlans_enabled)
3299 size += VLAN_TAG_SIZE;
3301 vmolr = rd32(E1000_VMOLR(vfn));
3302 vmolr &= ~E1000_VMOLR_RLPML_MASK;
3303 vmolr |= size | E1000_VMOLR_LPE;
3304 wr32(E1000_VMOLR(vfn), vmolr);
3310 * igb_rlpml_set - set maximum receive packet size
3311 * @adapter: board private structure
3313 * Configure maximum receivable packet size.
3315 static void igb_rlpml_set(struct igb_adapter *adapter)
3317 u32 max_frame_size = adapter->max_frame_size;
3318 struct e1000_hw *hw = &adapter->hw;
3319 u16 pf_id = adapter->vfs_allocated_count;
3322 igb_set_vf_rlpml(adapter, max_frame_size, pf_id);
3323 /* If we're in VMDQ or SR-IOV mode, then set global RLPML
3324 * to our max jumbo frame size, in case we need to enable
3325 * jumbo frames on one of the rings later.
3326 * This will not pass over-length frames into the default
3327 * queue because it's gated by the VMOLR.RLPML.
3329 max_frame_size = MAX_JUMBO_FRAME_SIZE;
3332 wr32(E1000_RLPML, max_frame_size);
3335 static inline void igb_set_vmolr(struct igb_adapter *adapter,
3338 struct e1000_hw *hw = &adapter->hw;
3341 /* This register exists only on 82576 and newer so if we are older then
3342 * we should exit and do nothing
3344 if (hw->mac.type < e1000_82576)
3347 vmolr = rd32(E1000_VMOLR(vfn));
3348 vmolr |= E1000_VMOLR_STRVLAN; /* Strip vlan tags */
3350 vmolr |= E1000_VMOLR_AUPE; /* Accept untagged packets */
3352 vmolr &= ~(E1000_VMOLR_AUPE); /* Tagged packets ONLY */
3354 /* clear all bits that might not be set */
3355 vmolr &= ~(E1000_VMOLR_BAM | E1000_VMOLR_RSSE);
3357 if (adapter->rss_queues > 1 && vfn == adapter->vfs_allocated_count)
3358 vmolr |= E1000_VMOLR_RSSE; /* enable RSS */
3359 /* for VMDq only allow the VFs and pool 0 to accept broadcast and
3362 if (vfn <= adapter->vfs_allocated_count)
3363 vmolr |= E1000_VMOLR_BAM; /* Accept broadcast */
3365 wr32(E1000_VMOLR(vfn), vmolr);
3369 * igb_configure_rx_ring - Configure a receive ring after Reset
3370 * @adapter: board private structure
3371 * @ring: receive ring to be configured
3373 * Configure the Rx unit of the MAC after a reset.
3375 void igb_configure_rx_ring(struct igb_adapter *adapter,
3376 struct igb_ring *ring)
3378 struct e1000_hw *hw = &adapter->hw;
3379 u64 rdba = ring->dma;
3380 int reg_idx = ring->reg_idx;
3381 u32 srrctl = 0, rxdctl = 0;
3383 /* disable the queue */
3384 wr32(E1000_RXDCTL(reg_idx), 0);
3386 /* Set DMA base address registers */
3387 wr32(E1000_RDBAL(reg_idx),
3388 rdba & 0x00000000ffffffffULL);
3389 wr32(E1000_RDBAH(reg_idx), rdba >> 32);
3390 wr32(E1000_RDLEN(reg_idx),
3391 ring->count * sizeof(union e1000_adv_rx_desc));
3393 /* initialize head and tail */
3394 ring->tail = hw->hw_addr + E1000_RDT(reg_idx);
3395 wr32(E1000_RDH(reg_idx), 0);
3396 writel(0, ring->tail);
3398 /* set descriptor configuration */
3399 srrctl = IGB_RX_HDR_LEN << E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
3400 srrctl |= IGB_RX_BUFSZ >> E1000_SRRCTL_BSIZEPKT_SHIFT;
3401 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
3402 if (hw->mac.type >= e1000_82580)
3403 srrctl |= E1000_SRRCTL_TIMESTAMP;
3404 /* Only set Drop Enable if we are supporting multiple queues */
3405 if (adapter->vfs_allocated_count || adapter->num_rx_queues > 1)
3406 srrctl |= E1000_SRRCTL_DROP_EN;
3408 wr32(E1000_SRRCTL(reg_idx), srrctl);
3410 /* set filtering for VMDQ pools */
3411 igb_set_vmolr(adapter, reg_idx & 0x7, true);
3413 rxdctl |= IGB_RX_PTHRESH;
3414 rxdctl |= IGB_RX_HTHRESH << 8;
3415 rxdctl |= IGB_RX_WTHRESH << 16;
3417 /* enable receive descriptor fetching */
3418 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
3419 wr32(E1000_RXDCTL(reg_idx), rxdctl);
3423 * igb_configure_rx - Configure receive Unit after Reset
3424 * @adapter: board private structure
3426 * Configure the Rx unit of the MAC after a reset.
3428 static void igb_configure_rx(struct igb_adapter *adapter)
3432 /* set UTA to appropriate mode */
3433 igb_set_uta(adapter);
3435 /* set the correct pool for the PF default MAC address in entry 0 */
3436 igb_rar_set_qsel(adapter, adapter->hw.mac.addr, 0,
3437 adapter->vfs_allocated_count);
3439 /* Setup the HW Rx Head and Tail Descriptor Pointers and
3440 * the Base and Length of the Rx Descriptor Ring
3442 for (i = 0; i < adapter->num_rx_queues; i++)
3443 igb_configure_rx_ring(adapter, adapter->rx_ring[i]);
3447 * igb_free_tx_resources - Free Tx Resources per Queue
3448 * @tx_ring: Tx descriptor ring for a specific queue
3450 * Free all transmit software resources
3452 void igb_free_tx_resources(struct igb_ring *tx_ring)
3454 igb_clean_tx_ring(tx_ring);
3456 vfree(tx_ring->tx_buffer_info);
3457 tx_ring->tx_buffer_info = NULL;
3459 /* if not set, then don't free */
3463 dma_free_coherent(tx_ring->dev, tx_ring->size,
3464 tx_ring->desc, tx_ring->dma);
3466 tx_ring->desc = NULL;
3470 * igb_free_all_tx_resources - Free Tx Resources for All Queues
3471 * @adapter: board private structure
3473 * Free all transmit software resources
3475 static void igb_free_all_tx_resources(struct igb_adapter *adapter)
3479 for (i = 0; i < adapter->num_tx_queues; i++)
3480 igb_free_tx_resources(adapter->tx_ring[i]);
3483 void igb_unmap_and_free_tx_resource(struct igb_ring *ring,
3484 struct igb_tx_buffer *tx_buffer)
3486 if (tx_buffer->skb) {
3487 dev_kfree_skb_any(tx_buffer->skb);
3488 if (dma_unmap_len(tx_buffer, len))
3489 dma_unmap_single(ring->dev,
3490 dma_unmap_addr(tx_buffer, dma),
3491 dma_unmap_len(tx_buffer, len),
3493 } else if (dma_unmap_len(tx_buffer, len)) {
3494 dma_unmap_page(ring->dev,
3495 dma_unmap_addr(tx_buffer, dma),
3496 dma_unmap_len(tx_buffer, len),
3499 tx_buffer->next_to_watch = NULL;
3500 tx_buffer->skb = NULL;
3501 dma_unmap_len_set(tx_buffer, len, 0);
3502 /* buffer_info must be completely set up in the transmit path */
3506 * igb_clean_tx_ring - Free Tx Buffers
3507 * @tx_ring: ring to be cleaned
3509 static void igb_clean_tx_ring(struct igb_ring *tx_ring)
3511 struct igb_tx_buffer *buffer_info;
3515 if (!tx_ring->tx_buffer_info)
3517 /* Free all the Tx ring sk_buffs */
3519 for (i = 0; i < tx_ring->count; i++) {
3520 buffer_info = &tx_ring->tx_buffer_info[i];
3521 igb_unmap_and_free_tx_resource(tx_ring, buffer_info);
3524 netdev_tx_reset_queue(txring_txq(tx_ring));
3526 size = sizeof(struct igb_tx_buffer) * tx_ring->count;
3527 memset(tx_ring->tx_buffer_info, 0, size);
3529 /* Zero out the descriptor ring */
3530 memset(tx_ring->desc, 0, tx_ring->size);
3532 tx_ring->next_to_use = 0;
3533 tx_ring->next_to_clean = 0;
3537 * igb_clean_all_tx_rings - Free Tx Buffers for all queues
3538 * @adapter: board private structure
3540 static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
3544 for (i = 0; i < adapter->num_tx_queues; i++)
3545 igb_clean_tx_ring(adapter->tx_ring[i]);
3549 * igb_free_rx_resources - Free Rx Resources
3550 * @rx_ring: ring to clean the resources from
3552 * Free all receive software resources
3554 void igb_free_rx_resources(struct igb_ring *rx_ring)
3556 igb_clean_rx_ring(rx_ring);
3558 vfree(rx_ring->rx_buffer_info);
3559 rx_ring->rx_buffer_info = NULL;
3561 /* if not set, then don't free */
3565 dma_free_coherent(rx_ring->dev, rx_ring->size,
3566 rx_ring->desc, rx_ring->dma);
3568 rx_ring->desc = NULL;
3572 * igb_free_all_rx_resources - Free Rx Resources for All Queues
3573 * @adapter: board private structure
3575 * Free all receive software resources
3577 static void igb_free_all_rx_resources(struct igb_adapter *adapter)
3581 for (i = 0; i < adapter->num_rx_queues; i++)
3582 igb_free_rx_resources(adapter->rx_ring[i]);
3586 * igb_clean_rx_ring - Free Rx Buffers per Queue
3587 * @rx_ring: ring to free buffers from
3589 static void igb_clean_rx_ring(struct igb_ring *rx_ring)
3595 dev_kfree_skb(rx_ring->skb);
3596 rx_ring->skb = NULL;
3598 if (!rx_ring->rx_buffer_info)
3601 /* Free all the Rx ring sk_buffs */
3602 for (i = 0; i < rx_ring->count; i++) {
3603 struct igb_rx_buffer *buffer_info = &rx_ring->rx_buffer_info[i];
3605 if (!buffer_info->page)
3608 dma_unmap_page(rx_ring->dev,
3612 __free_page(buffer_info->page);
3614 buffer_info->page = NULL;
3617 size = sizeof(struct igb_rx_buffer) * rx_ring->count;
3618 memset(rx_ring->rx_buffer_info, 0, size);
3620 /* Zero out the descriptor ring */
3621 memset(rx_ring->desc, 0, rx_ring->size);
3623 rx_ring->next_to_alloc = 0;
3624 rx_ring->next_to_clean = 0;
3625 rx_ring->next_to_use = 0;
3629 * igb_clean_all_rx_rings - Free Rx Buffers for all queues
3630 * @adapter: board private structure
3632 static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
3636 for (i = 0; i < adapter->num_rx_queues; i++)
3637 igb_clean_rx_ring(adapter->rx_ring[i]);
3641 * igb_set_mac - Change the Ethernet Address of the NIC
3642 * @netdev: network interface device structure
3643 * @p: pointer to an address structure
3645 * Returns 0 on success, negative on failure
3647 static int igb_set_mac(struct net_device *netdev, void *p)
3649 struct igb_adapter *adapter = netdev_priv(netdev);
3650 struct e1000_hw *hw = &adapter->hw;
3651 struct sockaddr *addr = p;
3653 if (!is_valid_ether_addr(addr->sa_data))
3654 return -EADDRNOTAVAIL;
3656 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3657 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
3659 /* set the correct pool for the new PF MAC address in entry 0 */
3660 igb_rar_set_qsel(adapter, hw->mac.addr, 0,
3661 adapter->vfs_allocated_count);
3667 * igb_write_mc_addr_list - write multicast addresses to MTA
3668 * @netdev: network interface device structure
3670 * Writes multicast address list to the MTA hash table.
3671 * Returns: -ENOMEM on failure
3672 * 0 on no addresses written
3673 * X on writing X addresses to MTA
3675 static int igb_write_mc_addr_list(struct net_device *netdev)
3677 struct igb_adapter *adapter = netdev_priv(netdev);
3678 struct e1000_hw *hw = &adapter->hw;
3679 struct netdev_hw_addr *ha;
3683 if (netdev_mc_empty(netdev)) {
3684 /* nothing to program, so clear mc list */
3685 igb_update_mc_addr_list(hw, NULL, 0);
3686 igb_restore_vf_multicasts(adapter);
3690 mta_list = kzalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
3694 /* The shared function expects a packed array of only addresses. */
3696 netdev_for_each_mc_addr(ha, netdev)
3697 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3699 igb_update_mc_addr_list(hw, mta_list, i);
3702 return netdev_mc_count(netdev);
3706 * igb_write_uc_addr_list - write unicast addresses to RAR table
3707 * @netdev: network interface device structure
3709 * Writes unicast address list to the RAR table.
3710 * Returns: -ENOMEM on failure/insufficient address space
3711 * 0 on no addresses written
3712 * X on writing X addresses to the RAR table
3714 static int igb_write_uc_addr_list(struct net_device *netdev)
3716 struct igb_adapter *adapter = netdev_priv(netdev);
3717 struct e1000_hw *hw = &adapter->hw;
3718 unsigned int vfn = adapter->vfs_allocated_count;
3719 unsigned int rar_entries = hw->mac.rar_entry_count - (vfn + 1);
3722 /* return ENOMEM indicating insufficient memory for addresses */
3723 if (netdev_uc_count(netdev) > rar_entries)
3726 if (!netdev_uc_empty(netdev) && rar_entries) {
3727 struct netdev_hw_addr *ha;
3729 netdev_for_each_uc_addr(ha, netdev) {
3732 igb_rar_set_qsel(adapter, ha->addr,
3738 /* write the addresses in reverse order to avoid write combining */
3739 for (; rar_entries > 0 ; rar_entries--) {
3740 wr32(E1000_RAH(rar_entries), 0);
3741 wr32(E1000_RAL(rar_entries), 0);
3749 * igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
3750 * @netdev: network interface device structure
3752 * The set_rx_mode entry point is called whenever the unicast or multicast
3753 * address lists or the network interface flags are updated. This routine is
3754 * responsible for configuring the hardware for proper unicast, multicast,
3755 * promiscuous mode, and all-multi behavior.
3757 static void igb_set_rx_mode(struct net_device *netdev)
3759 struct igb_adapter *adapter = netdev_priv(netdev);
3760 struct e1000_hw *hw = &adapter->hw;
3761 unsigned int vfn = adapter->vfs_allocated_count;
3762 u32 rctl, vmolr = 0;
3765 /* Check for Promiscuous and All Multicast modes */
3766 rctl = rd32(E1000_RCTL);
3768 /* clear the effected bits */
3769 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE | E1000_RCTL_VFE);
3771 if (netdev->flags & IFF_PROMISC) {
3772 /* retain VLAN HW filtering if in VT mode */
3773 if (adapter->vfs_allocated_count)
3774 rctl |= E1000_RCTL_VFE;
3775 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3776 vmolr |= (E1000_VMOLR_ROPE | E1000_VMOLR_MPME);
3778 if (netdev->flags & IFF_ALLMULTI) {
3779 rctl |= E1000_RCTL_MPE;
3780 vmolr |= E1000_VMOLR_MPME;
3782 /* Write addresses to the MTA, if the attempt fails
3783 * then we should just turn on promiscuous mode so
3784 * that we can at least receive multicast traffic
3786 count = igb_write_mc_addr_list(netdev);
3788 rctl |= E1000_RCTL_MPE;
3789 vmolr |= E1000_VMOLR_MPME;
3791 vmolr |= E1000_VMOLR_ROMPE;
3794 /* Write addresses to available RAR registers, if there is not
3795 * sufficient space to store all the addresses then enable
3796 * unicast promiscuous mode
3798 count = igb_write_uc_addr_list(netdev);
3800 rctl |= E1000_RCTL_UPE;
3801 vmolr |= E1000_VMOLR_ROPE;
3803 rctl |= E1000_RCTL_VFE;
3805 wr32(E1000_RCTL, rctl);
3807 /* In order to support SR-IOV and eventually VMDq it is necessary to set
3808 * the VMOLR to enable the appropriate modes. Without this workaround
3809 * we will have issues with VLAN tag stripping not being done for frames
3810 * that are only arriving because we are the default pool
3812 if ((hw->mac.type < e1000_82576) || (hw->mac.type > e1000_i350))
3815 vmolr |= rd32(E1000_VMOLR(vfn)) &
3816 ~(E1000_VMOLR_ROPE | E1000_VMOLR_MPME | E1000_VMOLR_ROMPE);
3817 wr32(E1000_VMOLR(vfn), vmolr);
3818 igb_restore_vf_multicasts(adapter);
3821 static void igb_check_wvbr(struct igb_adapter *adapter)
3823 struct e1000_hw *hw = &adapter->hw;
3826 switch (hw->mac.type) {
3829 if (!(wvbr = rd32(E1000_WVBR)))
3836 adapter->wvbr |= wvbr;
3839 #define IGB_STAGGERED_QUEUE_OFFSET 8
3841 static void igb_spoof_check(struct igb_adapter *adapter)
3848 for(j = 0; j < adapter->vfs_allocated_count; j++) {
3849 if (adapter->wvbr & (1 << j) ||
3850 adapter->wvbr & (1 << (j + IGB_STAGGERED_QUEUE_OFFSET))) {
3851 dev_warn(&adapter->pdev->dev,
3852 "Spoof event(s) detected on VF %d\n", j);
3855 (1 << (j + IGB_STAGGERED_QUEUE_OFFSET)));
3860 /* Need to wait a few seconds after link up to get diagnostic information from
3863 static void igb_update_phy_info(unsigned long data)
3865 struct igb_adapter *adapter = (struct igb_adapter *) data;
3866 igb_get_phy_info(&adapter->hw);
3870 * igb_has_link - check shared code for link and determine up/down
3871 * @adapter: pointer to driver private info
3873 bool igb_has_link(struct igb_adapter *adapter)
3875 struct e1000_hw *hw = &adapter->hw;
3876 bool link_active = false;
3878 /* get_link_status is set on LSC (link status) interrupt or
3879 * rx sequence error interrupt. get_link_status will stay
3880 * false until the e1000_check_for_link establishes link
3881 * for copper adapters ONLY
3883 switch (hw->phy.media_type) {
3884 case e1000_media_type_copper:
3885 if (!hw->mac.get_link_status)
3887 case e1000_media_type_internal_serdes:
3888 hw->mac.ops.check_for_link(hw);
3889 link_active = !hw->mac.get_link_status;
3892 case e1000_media_type_unknown:
3896 if (((hw->mac.type == e1000_i210) ||
3897 (hw->mac.type == e1000_i211)) &&
3898 (hw->phy.id == I210_I_PHY_ID)) {
3899 if (!netif_carrier_ok(adapter->netdev)) {
3900 adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
3901 } else if (!(adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)) {
3902 adapter->flags |= IGB_FLAG_NEED_LINK_UPDATE;
3903 adapter->link_check_timeout = jiffies;
3910 static bool igb_thermal_sensor_event(struct e1000_hw *hw, u32 event)
3913 u32 ctrl_ext, thstat;
3915 /* check for thermal sensor event on i350 copper only */
3916 if (hw->mac.type == e1000_i350) {
3917 thstat = rd32(E1000_THSTAT);
3918 ctrl_ext = rd32(E1000_CTRL_EXT);
3920 if ((hw->phy.media_type == e1000_media_type_copper) &&
3921 !(ctrl_ext & E1000_CTRL_EXT_LINK_MODE_SGMII))
3922 ret = !!(thstat & event);
3929 * igb_watchdog - Timer Call-back
3930 * @data: pointer to adapter cast into an unsigned long
3932 static void igb_watchdog(unsigned long data)
3934 struct igb_adapter *adapter = (struct igb_adapter *)data;
3935 /* Do the rest outside of interrupt context */
3936 schedule_work(&adapter->watchdog_task);
3939 static void igb_watchdog_task(struct work_struct *work)
3941 struct igb_adapter *adapter = container_of(work,
3944 struct e1000_hw *hw = &adapter->hw;
3945 struct e1000_phy_info *phy = &hw->phy;
3946 struct net_device *netdev = adapter->netdev;
3950 link = igb_has_link(adapter);
3952 if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE) {
3953 if (time_after(jiffies, (adapter->link_check_timeout + HZ)))
3954 adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
3960 /* Cancel scheduled suspend requests. */
3961 pm_runtime_resume(netdev->dev.parent);
3963 if (!netif_carrier_ok(netdev)) {
3965 hw->mac.ops.get_speed_and_duplex(hw,
3966 &adapter->link_speed,
3967 &adapter->link_duplex);
3969 ctrl = rd32(E1000_CTRL);
3970 /* Links status message must follow this format */
3971 printk(KERN_INFO "igb: %s NIC Link is Up %d Mbps %s "
3972 "Duplex, Flow Control: %s\n",
3974 adapter->link_speed,
3975 adapter->link_duplex == FULL_DUPLEX ?
3977 (ctrl & E1000_CTRL_TFCE) &&
3978 (ctrl & E1000_CTRL_RFCE) ? "RX/TX" :
3979 (ctrl & E1000_CTRL_RFCE) ? "RX" :
3980 (ctrl & E1000_CTRL_TFCE) ? "TX" : "None");
3982 /* check if SmartSpeed worked */
3983 igb_check_downshift(hw);
3984 if (phy->speed_downgraded)
3985 netdev_warn(netdev, "Link Speed was downgraded by SmartSpeed\n");
3987 /* check for thermal sensor event */
3988 if (igb_thermal_sensor_event(hw,
3989 E1000_THSTAT_LINK_THROTTLE)) {
3990 netdev_info(netdev, "The network adapter link "
3991 "speed was downshifted because it "
3995 /* adjust timeout factor according to speed/duplex */
3996 adapter->tx_timeout_factor = 1;
3997 switch (adapter->link_speed) {
3999 adapter->tx_timeout_factor = 14;
4002 /* maybe add some timeout factor ? */
4006 netif_carrier_on(netdev);
4008 igb_ping_all_vfs(adapter);
4009 igb_check_vf_rate_limit(adapter);
4011 /* link state has changed, schedule phy info update */
4012 if (!test_bit(__IGB_DOWN, &adapter->state))
4013 mod_timer(&adapter->phy_info_timer,
4014 round_jiffies(jiffies + 2 * HZ));
4017 if (netif_carrier_ok(netdev)) {
4018 adapter->link_speed = 0;
4019 adapter->link_duplex = 0;
4021 /* check for thermal sensor event */
4022 if (igb_thermal_sensor_event(hw,
4023 E1000_THSTAT_PWR_DOWN)) {
4024 netdev_err(netdev, "The network adapter was "
4025 "stopped because it overheated\n");
4028 /* Links status message must follow this format */
4029 printk(KERN_INFO "igb: %s NIC Link is Down\n",
4031 netif_carrier_off(netdev);
4033 igb_ping_all_vfs(adapter);
4035 /* link state has changed, schedule phy info update */
4036 if (!test_bit(__IGB_DOWN, &adapter->state))
4037 mod_timer(&adapter->phy_info_timer,
4038 round_jiffies(jiffies + 2 * HZ));
4040 pm_schedule_suspend(netdev->dev.parent,
4045 spin_lock(&adapter->stats64_lock);
4046 igb_update_stats(adapter, &adapter->stats64);
4047 spin_unlock(&adapter->stats64_lock);
4049 for (i = 0; i < adapter->num_tx_queues; i++) {
4050 struct igb_ring *tx_ring = adapter->tx_ring[i];
4051 if (!netif_carrier_ok(netdev)) {
4052 /* We've lost link, so the controller stops DMA,
4053 * but we've got queued Tx work that's never going
4054 * to get done, so reset controller to flush Tx.
4055 * (Do the reset outside of interrupt context).
4057 if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) {
4058 adapter->tx_timeout_count++;
4059 schedule_work(&adapter->reset_task);
4060 /* return immediately since reset is imminent */
4065 /* Force detection of hung controller every watchdog period */
4066 set_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
4069 /* Cause software interrupt to ensure Rx ring is cleaned */
4070 if (adapter->msix_entries) {
4072 for (i = 0; i < adapter->num_q_vectors; i++)
4073 eics |= adapter->q_vector[i]->eims_value;
4074 wr32(E1000_EICS, eics);
4076 wr32(E1000_ICS, E1000_ICS_RXDMT0);
4079 igb_spoof_check(adapter);
4080 igb_ptp_rx_hang(adapter);
4082 /* Reset the timer */
4083 if (!test_bit(__IGB_DOWN, &adapter->state)) {
4084 if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)
4085 mod_timer(&adapter->watchdog_timer,
4086 round_jiffies(jiffies + HZ));
4088 mod_timer(&adapter->watchdog_timer,
4089 round_jiffies(jiffies + 2 * HZ));
4093 enum latency_range {
4097 latency_invalid = 255
4101 * igb_update_ring_itr - update the dynamic ITR value based on packet size
4102 * @q_vector: pointer to q_vector
4104 * Stores a new ITR value based on strictly on packet size. This
4105 * algorithm is less sophisticated than that used in igb_update_itr,
4106 * due to the difficulty of synchronizing statistics across multiple
4107 * receive rings. The divisors and thresholds used by this function
4108 * were determined based on theoretical maximum wire speed and testing
4109 * data, in order to minimize response time while increasing bulk
4111 * This functionality is controlled by the InterruptThrottleRate module
4112 * parameter (see igb_param.c)
4113 * NOTE: This function is called only when operating in a multiqueue
4114 * receive environment.
4116 static void igb_update_ring_itr(struct igb_q_vector *q_vector)
4118 int new_val = q_vector->itr_val;
4119 int avg_wire_size = 0;
4120 struct igb_adapter *adapter = q_vector->adapter;
4121 unsigned int packets;
4123 /* For non-gigabit speeds, just fix the interrupt rate at 4000
4124 * ints/sec - ITR timer value of 120 ticks.
4126 if (adapter->link_speed != SPEED_1000) {
4127 new_val = IGB_4K_ITR;
4131 packets = q_vector->rx.total_packets;
4133 avg_wire_size = q_vector->rx.total_bytes / packets;
4135 packets = q_vector->tx.total_packets;
4137 avg_wire_size = max_t(u32, avg_wire_size,
4138 q_vector->tx.total_bytes / packets);
4140 /* if avg_wire_size isn't set no work was done */
4144 /* Add 24 bytes to size to account for CRC, preamble, and gap */
4145 avg_wire_size += 24;
4147 /* Don't starve jumbo frames */
4148 avg_wire_size = min(avg_wire_size, 3000);
4150 /* Give a little boost to mid-size frames */
4151 if ((avg_wire_size > 300) && (avg_wire_size < 1200))
4152 new_val = avg_wire_size / 3;
4154 new_val = avg_wire_size / 2;
4156 /* conservative mode (itr 3) eliminates the lowest_latency setting */
4157 if (new_val < IGB_20K_ITR &&
4158 ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
4159 (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
4160 new_val = IGB_20K_ITR;
4163 if (new_val != q_vector->itr_val) {
4164 q_vector->itr_val = new_val;
4165 q_vector->set_itr = 1;
4168 q_vector->rx.total_bytes = 0;
4169 q_vector->rx.total_packets = 0;
4170 q_vector->tx.total_bytes = 0;
4171 q_vector->tx.total_packets = 0;
4175 * igb_update_itr - update the dynamic ITR value based on statistics
4176 * @q_vector: pointer to q_vector
4177 * @ring_container: ring info to update the itr for
4179 * Stores a new ITR value based on packets and byte
4180 * counts during the last interrupt. The advantage of per interrupt
4181 * computation is faster updates and more accurate ITR for the current
4182 * traffic pattern. Constants in this function were computed
4183 * based on theoretical maximum wire speed and thresholds were set based
4184 * on testing data as well as attempting to minimize response time
4185 * while increasing bulk throughput.
4186 * this functionality is controlled by the InterruptThrottleRate module
4187 * parameter (see igb_param.c)
4188 * NOTE: These calculations are only valid when operating in a single-
4189 * queue environment.
4191 static void igb_update_itr(struct igb_q_vector *q_vector,
4192 struct igb_ring_container *ring_container)
4194 unsigned int packets = ring_container->total_packets;
4195 unsigned int bytes = ring_container->total_bytes;
4196 u8 itrval = ring_container->itr;
4198 /* no packets, exit with status unchanged */
4203 case lowest_latency:
4204 /* handle TSO and jumbo frames */
4205 if (bytes/packets > 8000)
4206 itrval = bulk_latency;
4207 else if ((packets < 5) && (bytes > 512))
4208 itrval = low_latency;
4210 case low_latency: /* 50 usec aka 20000 ints/s */
4211 if (bytes > 10000) {
4212 /* this if handles the TSO accounting */
4213 if (bytes/packets > 8000) {
4214 itrval = bulk_latency;
4215 } else if ((packets < 10) || ((bytes/packets) > 1200)) {
4216 itrval = bulk_latency;
4217 } else if ((packets > 35)) {
4218 itrval = lowest_latency;
4220 } else if (bytes/packets > 2000) {
4221 itrval = bulk_latency;
4222 } else if (packets <= 2 && bytes < 512) {
4223 itrval = lowest_latency;
4226 case bulk_latency: /* 250 usec aka 4000 ints/s */
4227 if (bytes > 25000) {
4229 itrval = low_latency;
4230 } else if (bytes < 1500) {
4231 itrval = low_latency;
4236 /* clear work counters since we have the values we need */
4237 ring_container->total_bytes = 0;
4238 ring_container->total_packets = 0;
4240 /* write updated itr to ring container */
4241 ring_container->itr = itrval;
4244 static void igb_set_itr(struct igb_q_vector *q_vector)
4246 struct igb_adapter *adapter = q_vector->adapter;
4247 u32 new_itr = q_vector->itr_val;
4250 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
4251 if (adapter->link_speed != SPEED_1000) {
4253 new_itr = IGB_4K_ITR;
4257 igb_update_itr(q_vector, &q_vector->tx);
4258 igb_update_itr(q_vector, &q_vector->rx);
4260 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
4262 /* conservative mode (itr 3) eliminates the lowest_latency setting */
4263 if (current_itr == lowest_latency &&
4264 ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
4265 (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
4266 current_itr = low_latency;
4268 switch (current_itr) {
4269 /* counts and packets in update_itr are dependent on these numbers */
4270 case lowest_latency:
4271 new_itr = IGB_70K_ITR; /* 70,000 ints/sec */
4274 new_itr = IGB_20K_ITR; /* 20,000 ints/sec */
4277 new_itr = IGB_4K_ITR; /* 4,000 ints/sec */
4284 if (new_itr != q_vector->itr_val) {
4285 /* this attempts to bias the interrupt rate towards Bulk
4286 * by adding intermediate steps when interrupt rate is
4289 new_itr = new_itr > q_vector->itr_val ?
4290 max((new_itr * q_vector->itr_val) /
4291 (new_itr + (q_vector->itr_val >> 2)),
4293 /* Don't write the value here; it resets the adapter's
4294 * internal timer, and causes us to delay far longer than
4295 * we should between interrupts. Instead, we write the ITR
4296 * value at the beginning of the next interrupt so the timing
4297 * ends up being correct.
4299 q_vector->itr_val = new_itr;
4300 q_vector->set_itr = 1;
4304 static void igb_tx_ctxtdesc(struct igb_ring *tx_ring, u32 vlan_macip_lens,
4305 u32 type_tucmd, u32 mss_l4len_idx)
4307 struct e1000_adv_tx_context_desc *context_desc;
4308 u16 i = tx_ring->next_to_use;
4310 context_desc = IGB_TX_CTXTDESC(tx_ring, i);
4313 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
4315 /* set bits to identify this as an advanced context descriptor */
4316 type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
4318 /* For 82575, context index must be unique per ring. */
4319 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
4320 mss_l4len_idx |= tx_ring->reg_idx << 4;
4322 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
4323 context_desc->seqnum_seed = 0;
4324 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd);
4325 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
4328 static int igb_tso(struct igb_ring *tx_ring,
4329 struct igb_tx_buffer *first,
4332 struct sk_buff *skb = first->skb;
4333 u32 vlan_macip_lens, type_tucmd;
4334 u32 mss_l4len_idx, l4len;
4336 if (skb->ip_summed != CHECKSUM_PARTIAL)
4339 if (!skb_is_gso(skb))
4342 if (skb_header_cloned(skb)) {
4343 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4348 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
4349 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
4351 if (first->protocol == __constant_htons(ETH_P_IP)) {
4352 struct iphdr *iph = ip_hdr(skb);
4355 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
4359 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
4360 first->tx_flags |= IGB_TX_FLAGS_TSO |
4363 } else if (skb_is_gso_v6(skb)) {
4364 ipv6_hdr(skb)->payload_len = 0;
4365 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4366 &ipv6_hdr(skb)->daddr,
4368 first->tx_flags |= IGB_TX_FLAGS_TSO |
4372 /* compute header lengths */
4373 l4len = tcp_hdrlen(skb);
4374 *hdr_len = skb_transport_offset(skb) + l4len;
4376 /* update gso size and bytecount with header size */
4377 first->gso_segs = skb_shinfo(skb)->gso_segs;
4378 first->bytecount += (first->gso_segs - 1) * *hdr_len;
4381 mss_l4len_idx = l4len << E1000_ADVTXD_L4LEN_SHIFT;
4382 mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT;
4384 /* VLAN MACLEN IPLEN */
4385 vlan_macip_lens = skb_network_header_len(skb);
4386 vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
4387 vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
4389 igb_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
4394 static void igb_tx_csum(struct igb_ring *tx_ring, struct igb_tx_buffer *first)
4396 struct sk_buff *skb = first->skb;
4397 u32 vlan_macip_lens = 0;
4398 u32 mss_l4len_idx = 0;
4401 if (skb->ip_summed != CHECKSUM_PARTIAL) {
4402 if (!(first->tx_flags & IGB_TX_FLAGS_VLAN))
4406 switch (first->protocol) {
4407 case __constant_htons(ETH_P_IP):
4408 vlan_macip_lens |= skb_network_header_len(skb);
4409 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
4410 l4_hdr = ip_hdr(skb)->protocol;
4412 case __constant_htons(ETH_P_IPV6):
4413 vlan_macip_lens |= skb_network_header_len(skb);
4414 l4_hdr = ipv6_hdr(skb)->nexthdr;
4417 if (unlikely(net_ratelimit())) {
4418 dev_warn(tx_ring->dev,
4419 "partial checksum but proto=%x!\n",
4427 type_tucmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
4428 mss_l4len_idx = tcp_hdrlen(skb) <<
4429 E1000_ADVTXD_L4LEN_SHIFT;
4432 type_tucmd |= E1000_ADVTXD_TUCMD_L4T_SCTP;
4433 mss_l4len_idx = sizeof(struct sctphdr) <<
4434 E1000_ADVTXD_L4LEN_SHIFT;
4437 mss_l4len_idx = sizeof(struct udphdr) <<
4438 E1000_ADVTXD_L4LEN_SHIFT;
4441 if (unlikely(net_ratelimit())) {
4442 dev_warn(tx_ring->dev,
4443 "partial checksum but l4 proto=%x!\n",
4449 /* update TX checksum flag */
4450 first->tx_flags |= IGB_TX_FLAGS_CSUM;
4453 vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
4454 vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
4456 igb_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
4459 #define IGB_SET_FLAG(_input, _flag, _result) \
4460 ((_flag <= _result) ? \
4461 ((u32)(_input & _flag) * (_result / _flag)) : \
4462 ((u32)(_input & _flag) / (_flag / _result)))
4464 static u32 igb_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
4466 /* set type for advanced descriptor with frame checksum insertion */
4467 u32 cmd_type = E1000_ADVTXD_DTYP_DATA |
4468 E1000_ADVTXD_DCMD_DEXT |
4469 E1000_ADVTXD_DCMD_IFCS;
4471 /* set HW vlan bit if vlan is present */
4472 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_VLAN,
4473 (E1000_ADVTXD_DCMD_VLE));
4475 /* set segmentation bits for TSO */
4476 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSO,
4477 (E1000_ADVTXD_DCMD_TSE));
4479 /* set timestamp bit if present */
4480 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSTAMP,
4481 (E1000_ADVTXD_MAC_TSTAMP));
4483 /* insert frame checksum */
4484 cmd_type ^= IGB_SET_FLAG(skb->no_fcs, 1, E1000_ADVTXD_DCMD_IFCS);
4489 static void igb_tx_olinfo_status(struct igb_ring *tx_ring,
4490 union e1000_adv_tx_desc *tx_desc,
4491 u32 tx_flags, unsigned int paylen)
4493 u32 olinfo_status = paylen << E1000_ADVTXD_PAYLEN_SHIFT;
4495 /* 82575 requires a unique index per ring */
4496 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
4497 olinfo_status |= tx_ring->reg_idx << 4;
4499 /* insert L4 checksum */
4500 olinfo_status |= IGB_SET_FLAG(tx_flags,
4502 (E1000_TXD_POPTS_TXSM << 8));
4504 /* insert IPv4 checksum */
4505 olinfo_status |= IGB_SET_FLAG(tx_flags,
4507 (E1000_TXD_POPTS_IXSM << 8));
4509 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
4512 static void igb_tx_map(struct igb_ring *tx_ring,
4513 struct igb_tx_buffer *first,
4516 struct sk_buff *skb = first->skb;
4517 struct igb_tx_buffer *tx_buffer;
4518 union e1000_adv_tx_desc *tx_desc;
4519 struct skb_frag_struct *frag;
4521 unsigned int data_len, size;
4522 u32 tx_flags = first->tx_flags;
4523 u32 cmd_type = igb_tx_cmd_type(skb, tx_flags);
4524 u16 i = tx_ring->next_to_use;
4526 tx_desc = IGB_TX_DESC(tx_ring, i);
4528 igb_tx_olinfo_status(tx_ring, tx_desc, tx_flags, skb->len - hdr_len);
4530 size = skb_headlen(skb);
4531 data_len = skb->data_len;
4533 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
4537 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
4538 if (dma_mapping_error(tx_ring->dev, dma))
4541 /* record length, and DMA address */
4542 dma_unmap_len_set(tx_buffer, len, size);
4543 dma_unmap_addr_set(tx_buffer, dma, dma);
4545 tx_desc->read.buffer_addr = cpu_to_le64(dma);
4547 while (unlikely(size > IGB_MAX_DATA_PER_TXD)) {
4548 tx_desc->read.cmd_type_len =
4549 cpu_to_le32(cmd_type ^ IGB_MAX_DATA_PER_TXD);
4553 if (i == tx_ring->count) {
4554 tx_desc = IGB_TX_DESC(tx_ring, 0);
4557 tx_desc->read.olinfo_status = 0;
4559 dma += IGB_MAX_DATA_PER_TXD;
4560 size -= IGB_MAX_DATA_PER_TXD;
4562 tx_desc->read.buffer_addr = cpu_to_le64(dma);
4565 if (likely(!data_len))
4568 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
4572 if (i == tx_ring->count) {
4573 tx_desc = IGB_TX_DESC(tx_ring, 0);
4576 tx_desc->read.olinfo_status = 0;
4578 size = skb_frag_size(frag);
4581 dma = skb_frag_dma_map(tx_ring->dev, frag, 0,
4582 size, DMA_TO_DEVICE);
4584 tx_buffer = &tx_ring->tx_buffer_info[i];
4587 /* write last descriptor with RS and EOP bits */
4588 cmd_type |= size | IGB_TXD_DCMD;
4589 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
4591 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
4593 /* set the timestamp */
4594 first->time_stamp = jiffies;
4596 /* Force memory writes to complete before letting h/w know there
4597 * are new descriptors to fetch. (Only applicable for weak-ordered
4598 * memory model archs, such as IA-64).
4600 * We also need this memory barrier to make certain all of the
4601 * status bits have been updated before next_to_watch is written.
4605 /* set next_to_watch value indicating a packet is present */
4606 first->next_to_watch = tx_desc;
4609 if (i == tx_ring->count)
4612 tx_ring->next_to_use = i;
4614 writel(i, tx_ring->tail);
4616 /* we need this if more than one processor can write to our tail
4617 * at a time, it synchronizes IO on IA64/Altix systems
4624 dev_err(tx_ring->dev, "TX DMA map failed\n");
4626 /* clear dma mappings for failed tx_buffer_info map */
4628 tx_buffer = &tx_ring->tx_buffer_info[i];
4629 igb_unmap_and_free_tx_resource(tx_ring, tx_buffer);
4630 if (tx_buffer == first)
4637 tx_ring->next_to_use = i;
4640 static int __igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
4642 struct net_device *netdev = tx_ring->netdev;
4644 netif_stop_subqueue(netdev, tx_ring->queue_index);
4646 /* Herbert's original patch had:
4647 * smp_mb__after_netif_stop_queue();
4648 * but since that doesn't exist yet, just open code it.
4652 /* We need to check again in a case another CPU has just
4653 * made room available.
4655 if (igb_desc_unused(tx_ring) < size)
4659 netif_wake_subqueue(netdev, tx_ring->queue_index);
4661 u64_stats_update_begin(&tx_ring->tx_syncp2);
4662 tx_ring->tx_stats.restart_queue2++;
4663 u64_stats_update_end(&tx_ring->tx_syncp2);
4668 static inline int igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
4670 if (igb_desc_unused(tx_ring) >= size)
4672 return __igb_maybe_stop_tx(tx_ring, size);
4675 netdev_tx_t igb_xmit_frame_ring(struct sk_buff *skb,
4676 struct igb_ring *tx_ring)
4678 struct igb_tx_buffer *first;
4681 u16 count = TXD_USE_COUNT(skb_headlen(skb));
4682 __be16 protocol = vlan_get_protocol(skb);
4685 /* need: 1 descriptor per page * PAGE_SIZE/IGB_MAX_DATA_PER_TXD,
4686 * + 1 desc for skb_headlen/IGB_MAX_DATA_PER_TXD,
4687 * + 2 desc gap to keep tail from touching head,
4688 * + 1 desc for context descriptor,
4689 * otherwise try next time
4691 if (NETDEV_FRAG_PAGE_MAX_SIZE > IGB_MAX_DATA_PER_TXD) {
4693 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
4694 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
4696 count += skb_shinfo(skb)->nr_frags;
4699 if (igb_maybe_stop_tx(tx_ring, count + 3)) {
4700 /* this is a hard error */
4701 return NETDEV_TX_BUSY;
4704 /* record the location of the first descriptor for this packet */
4705 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
4707 first->bytecount = skb->len;
4708 first->gso_segs = 1;
4710 skb_tx_timestamp(skb);
4712 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
4713 struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
4715 if (!(adapter->ptp_tx_skb)) {
4716 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
4717 tx_flags |= IGB_TX_FLAGS_TSTAMP;
4719 adapter->ptp_tx_skb = skb_get(skb);
4720 adapter->ptp_tx_start = jiffies;
4721 if (adapter->hw.mac.type == e1000_82576)
4722 schedule_work(&adapter->ptp_tx_work);
4726 if (vlan_tx_tag_present(skb)) {
4727 tx_flags |= IGB_TX_FLAGS_VLAN;
4728 tx_flags |= (vlan_tx_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
4731 /* record initial flags and protocol */
4732 first->tx_flags = tx_flags;
4733 first->protocol = protocol;
4735 tso = igb_tso(tx_ring, first, &hdr_len);
4739 igb_tx_csum(tx_ring, first);
4741 igb_tx_map(tx_ring, first, hdr_len);
4743 /* Make sure there is space in the ring for the next send. */
4744 igb_maybe_stop_tx(tx_ring, DESC_NEEDED);
4746 return NETDEV_TX_OK;
4749 igb_unmap_and_free_tx_resource(tx_ring, first);
4751 return NETDEV_TX_OK;
4754 static inline struct igb_ring *igb_tx_queue_mapping(struct igb_adapter *adapter,
4755 struct sk_buff *skb)
4757 unsigned int r_idx = skb->queue_mapping;
4759 if (r_idx >= adapter->num_tx_queues)
4760 r_idx = r_idx % adapter->num_tx_queues;
4762 return adapter->tx_ring[r_idx];
4765 static netdev_tx_t igb_xmit_frame(struct sk_buff *skb,
4766 struct net_device *netdev)
4768 struct igb_adapter *adapter = netdev_priv(netdev);
4770 if (test_bit(__IGB_DOWN, &adapter->state)) {
4771 dev_kfree_skb_any(skb);
4772 return NETDEV_TX_OK;
4775 if (skb->len <= 0) {
4776 dev_kfree_skb_any(skb);
4777 return NETDEV_TX_OK;
4780 /* The minimum packet size with TCTL.PSP set is 17 so pad the skb
4781 * in order to meet this minimum size requirement.
4783 if (unlikely(skb->len < 17)) {
4784 if (skb_pad(skb, 17 - skb->len))
4785 return NETDEV_TX_OK;
4787 skb_set_tail_pointer(skb, 17);
4790 return igb_xmit_frame_ring(skb, igb_tx_queue_mapping(adapter, skb));
4794 * igb_tx_timeout - Respond to a Tx Hang
4795 * @netdev: network interface device structure
4797 static void igb_tx_timeout(struct net_device *netdev)
4799 struct igb_adapter *adapter = netdev_priv(netdev);
4800 struct e1000_hw *hw = &adapter->hw;
4802 /* Do the reset outside of interrupt context */
4803 adapter->tx_timeout_count++;
4805 if (hw->mac.type >= e1000_82580)
4806 hw->dev_spec._82575.global_device_reset = true;
4808 schedule_work(&adapter->reset_task);
4810 (adapter->eims_enable_mask & ~adapter->eims_other));
4813 static void igb_reset_task(struct work_struct *work)
4815 struct igb_adapter *adapter;
4816 adapter = container_of(work, struct igb_adapter, reset_task);
4819 netdev_err(adapter->netdev, "Reset adapter\n");
4820 igb_reinit_locked(adapter);
4824 * igb_get_stats64 - Get System Network Statistics
4825 * @netdev: network interface device structure
4826 * @stats: rtnl_link_stats64 pointer
4828 static struct rtnl_link_stats64 *igb_get_stats64(struct net_device *netdev,
4829 struct rtnl_link_stats64 *stats)
4831 struct igb_adapter *adapter = netdev_priv(netdev);
4833 spin_lock(&adapter->stats64_lock);
4834 igb_update_stats(adapter, &adapter->stats64);
4835 memcpy(stats, &adapter->stats64, sizeof(*stats));
4836 spin_unlock(&adapter->stats64_lock);
4842 * igb_change_mtu - Change the Maximum Transfer Unit
4843 * @netdev: network interface device structure
4844 * @new_mtu: new value for maximum frame size
4846 * Returns 0 on success, negative on failure
4848 static int igb_change_mtu(struct net_device *netdev, int new_mtu)
4850 struct igb_adapter *adapter = netdev_priv(netdev);
4851 struct pci_dev *pdev = adapter->pdev;
4852 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
4854 if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
4855 dev_err(&pdev->dev, "Invalid MTU setting\n");
4859 #define MAX_STD_JUMBO_FRAME_SIZE 9238
4860 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
4861 dev_err(&pdev->dev, "MTU > 9216 not supported.\n");
4865 /* adjust max frame to be at least the size of a standard frame */
4866 if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
4867 max_frame = ETH_FRAME_LEN + ETH_FCS_LEN;
4869 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
4872 /* igb_down has a dependency on max_frame_size */
4873 adapter->max_frame_size = max_frame;
4875 if (netif_running(netdev))
4878 dev_info(&pdev->dev, "changing MTU from %d to %d\n",
4879 netdev->mtu, new_mtu);
4880 netdev->mtu = new_mtu;
4882 if (netif_running(netdev))
4887 clear_bit(__IGB_RESETTING, &adapter->state);
4893 * igb_update_stats - Update the board statistics counters
4894 * @adapter: board private structure
4896 void igb_update_stats(struct igb_adapter *adapter,
4897 struct rtnl_link_stats64 *net_stats)
4899 struct e1000_hw *hw = &adapter->hw;
4900 struct pci_dev *pdev = adapter->pdev;
4906 u64 _bytes, _packets;
4908 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
4910 /* Prevent stats update while adapter is being reset, or if the pci
4911 * connection is down.
4913 if (adapter->link_speed == 0)
4915 if (pci_channel_offline(pdev))
4922 for (i = 0; i < adapter->num_rx_queues; i++) {
4923 u32 rqdpc = rd32(E1000_RQDPC(i));
4924 struct igb_ring *ring = adapter->rx_ring[i];
4927 ring->rx_stats.drops += rqdpc;
4928 net_stats->rx_fifo_errors += rqdpc;
4932 start = u64_stats_fetch_begin_bh(&ring->rx_syncp);
4933 _bytes = ring->rx_stats.bytes;
4934 _packets = ring->rx_stats.packets;
4935 } while (u64_stats_fetch_retry_bh(&ring->rx_syncp, start));
4937 packets += _packets;
4940 net_stats->rx_bytes = bytes;
4941 net_stats->rx_packets = packets;
4945 for (i = 0; i < adapter->num_tx_queues; i++) {
4946 struct igb_ring *ring = adapter->tx_ring[i];
4948 start = u64_stats_fetch_begin_bh(&ring->tx_syncp);
4949 _bytes = ring->tx_stats.bytes;
4950 _packets = ring->tx_stats.packets;
4951 } while (u64_stats_fetch_retry_bh(&ring->tx_syncp, start));
4953 packets += _packets;
4955 net_stats->tx_bytes = bytes;
4956 net_stats->tx_packets = packets;
4959 /* read stats registers */
4960 adapter->stats.crcerrs += rd32(E1000_CRCERRS);
4961 adapter->stats.gprc += rd32(E1000_GPRC);
4962 adapter->stats.gorc += rd32(E1000_GORCL);
4963 rd32(E1000_GORCH); /* clear GORCL */
4964 adapter->stats.bprc += rd32(E1000_BPRC);
4965 adapter->stats.mprc += rd32(E1000_MPRC);
4966 adapter->stats.roc += rd32(E1000_ROC);
4968 adapter->stats.prc64 += rd32(E1000_PRC64);
4969 adapter->stats.prc127 += rd32(E1000_PRC127);
4970 adapter->stats.prc255 += rd32(E1000_PRC255);
4971 adapter->stats.prc511 += rd32(E1000_PRC511);
4972 adapter->stats.prc1023 += rd32(E1000_PRC1023);
4973 adapter->stats.prc1522 += rd32(E1000_PRC1522);
4974 adapter->stats.symerrs += rd32(E1000_SYMERRS);
4975 adapter->stats.sec += rd32(E1000_SEC);
4977 mpc = rd32(E1000_MPC);
4978 adapter->stats.mpc += mpc;
4979 net_stats->rx_fifo_errors += mpc;
4980 adapter->stats.scc += rd32(E1000_SCC);
4981 adapter->stats.ecol += rd32(E1000_ECOL);
4982 adapter->stats.mcc += rd32(E1000_MCC);
4983 adapter->stats.latecol += rd32(E1000_LATECOL);
4984 adapter->stats.dc += rd32(E1000_DC);
4985 adapter->stats.rlec += rd32(E1000_RLEC);
4986 adapter->stats.xonrxc += rd32(E1000_XONRXC);
4987 adapter->stats.xontxc += rd32(E1000_XONTXC);
4988 adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
4989 adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
4990 adapter->stats.fcruc += rd32(E1000_FCRUC);
4991 adapter->stats.gptc += rd32(E1000_GPTC);
4992 adapter->stats.gotc += rd32(E1000_GOTCL);
4993 rd32(E1000_GOTCH); /* clear GOTCL */
4994 adapter->stats.rnbc += rd32(E1000_RNBC);
4995 adapter->stats.ruc += rd32(E1000_RUC);
4996 adapter->stats.rfc += rd32(E1000_RFC);
4997 adapter->stats.rjc += rd32(E1000_RJC);
4998 adapter->stats.tor += rd32(E1000_TORH);
4999 adapter->stats.tot += rd32(E1000_TOTH);
5000 adapter->stats.tpr += rd32(E1000_TPR);
5002 adapter->stats.ptc64 += rd32(E1000_PTC64);
5003 adapter->stats.ptc127 += rd32(E1000_PTC127);
5004 adapter->stats.ptc255 += rd32(E1000_PTC255);
5005 adapter->stats.ptc511 += rd32(E1000_PTC511);
5006 adapter->stats.ptc1023 += rd32(E1000_PTC1023);
5007 adapter->stats.ptc1522 += rd32(E1000_PTC1522);
5009 adapter->stats.mptc += rd32(E1000_MPTC);
5010 adapter->stats.bptc += rd32(E1000_BPTC);
5012 adapter->stats.tpt += rd32(E1000_TPT);
5013 adapter->stats.colc += rd32(E1000_COLC);
5015 adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
5016 /* read internal phy specific stats */
5017 reg = rd32(E1000_CTRL_EXT);
5018 if (!(reg & E1000_CTRL_EXT_LINK_MODE_MASK)) {
5019 adapter->stats.rxerrc += rd32(E1000_RXERRC);
5021 /* this stat has invalid values on i210/i211 */
5022 if ((hw->mac.type != e1000_i210) &&
5023 (hw->mac.type != e1000_i211))
5024 adapter->stats.tncrs += rd32(E1000_TNCRS);
5027 adapter->stats.tsctc += rd32(E1000_TSCTC);
5028 adapter->stats.tsctfc += rd32(E1000_TSCTFC);
5030 adapter->stats.iac += rd32(E1000_IAC);
5031 adapter->stats.icrxoc += rd32(E1000_ICRXOC);
5032 adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
5033 adapter->stats.icrxatc += rd32(E1000_ICRXATC);
5034 adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
5035 adapter->stats.ictxatc += rd32(E1000_ICTXATC);
5036 adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
5037 adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
5038 adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
5040 /* Fill out the OS statistics structure */
5041 net_stats->multicast = adapter->stats.mprc;
5042 net_stats->collisions = adapter->stats.colc;
5046 /* RLEC on some newer hardware can be incorrect so build
5047 * our own version based on RUC and ROC
5049 net_stats->rx_errors = adapter->stats.rxerrc +
5050 adapter->stats.crcerrs + adapter->stats.algnerrc +
5051 adapter->stats.ruc + adapter->stats.roc +
5052 adapter->stats.cexterr;
5053 net_stats->rx_length_errors = adapter->stats.ruc +
5055 net_stats->rx_crc_errors = adapter->stats.crcerrs;
5056 net_stats->rx_frame_errors = adapter->stats.algnerrc;
5057 net_stats->rx_missed_errors = adapter->stats.mpc;
5060 net_stats->tx_errors = adapter->stats.ecol +
5061 adapter->stats.latecol;
5062 net_stats->tx_aborted_errors = adapter->stats.ecol;
5063 net_stats->tx_window_errors = adapter->stats.latecol;
5064 net_stats->tx_carrier_errors = adapter->stats.tncrs;
5066 /* Tx Dropped needs to be maintained elsewhere */
5069 if (hw->phy.media_type == e1000_media_type_copper) {
5070 if ((adapter->link_speed == SPEED_1000) &&
5071 (!igb_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
5072 phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
5073 adapter->phy_stats.idle_errors += phy_tmp;
5077 /* Management Stats */
5078 adapter->stats.mgptc += rd32(E1000_MGTPTC);
5079 adapter->stats.mgprc += rd32(E1000_MGTPRC);
5080 adapter->stats.mgpdc += rd32(E1000_MGTPDC);
5083 reg = rd32(E1000_MANC);
5084 if (reg & E1000_MANC_EN_BMC2OS) {
5085 adapter->stats.o2bgptc += rd32(E1000_O2BGPTC);
5086 adapter->stats.o2bspc += rd32(E1000_O2BSPC);
5087 adapter->stats.b2ospc += rd32(E1000_B2OSPC);
5088 adapter->stats.b2ogprc += rd32(E1000_B2OGPRC);
5092 static irqreturn_t igb_msix_other(int irq, void *data)
5094 struct igb_adapter *adapter = data;
5095 struct e1000_hw *hw = &adapter->hw;
5096 u32 icr = rd32(E1000_ICR);
5097 /* reading ICR causes bit 31 of EICR to be cleared */
5099 if (icr & E1000_ICR_DRSTA)
5100 schedule_work(&adapter->reset_task);
5102 if (icr & E1000_ICR_DOUTSYNC) {
5103 /* HW is reporting DMA is out of sync */
5104 adapter->stats.doosync++;
5105 /* The DMA Out of Sync is also indication of a spoof event
5106 * in IOV mode. Check the Wrong VM Behavior register to
5107 * see if it is really a spoof event.
5109 igb_check_wvbr(adapter);
5112 /* Check for a mailbox event */
5113 if (icr & E1000_ICR_VMMB)
5114 igb_msg_task(adapter);
5116 if (icr & E1000_ICR_LSC) {
5117 hw->mac.get_link_status = 1;
5118 /* guard against interrupt when we're going down */
5119 if (!test_bit(__IGB_DOWN, &adapter->state))
5120 mod_timer(&adapter->watchdog_timer, jiffies + 1);
5123 if (icr & E1000_ICR_TS) {
5124 u32 tsicr = rd32(E1000_TSICR);
5126 if (tsicr & E1000_TSICR_TXTS) {
5127 /* acknowledge the interrupt */
5128 wr32(E1000_TSICR, E1000_TSICR_TXTS);
5129 /* retrieve hardware timestamp */
5130 schedule_work(&adapter->ptp_tx_work);
5134 wr32(E1000_EIMS, adapter->eims_other);
5139 static void igb_write_itr(struct igb_q_vector *q_vector)
5141 struct igb_adapter *adapter = q_vector->adapter;
5142 u32 itr_val = q_vector->itr_val & 0x7FFC;
5144 if (!q_vector->set_itr)
5150 if (adapter->hw.mac.type == e1000_82575)
5151 itr_val |= itr_val << 16;
5153 itr_val |= E1000_EITR_CNT_IGNR;
5155 writel(itr_val, q_vector->itr_register);
5156 q_vector->set_itr = 0;
5159 static irqreturn_t igb_msix_ring(int irq, void *data)
5161 struct igb_q_vector *q_vector = data;
5163 /* Write the ITR value calculated from the previous interrupt. */
5164 igb_write_itr(q_vector);
5166 napi_schedule(&q_vector->napi);
5171 #ifdef CONFIG_IGB_DCA
5172 static void igb_update_tx_dca(struct igb_adapter *adapter,
5173 struct igb_ring *tx_ring,
5176 struct e1000_hw *hw = &adapter->hw;
5177 u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
5179 if (hw->mac.type != e1000_82575)
5180 txctrl <<= E1000_DCA_TXCTRL_CPUID_SHIFT;
5182 /* We can enable relaxed ordering for reads, but not writes when
5183 * DCA is enabled. This is due to a known issue in some chipsets
5184 * which will cause the DCA tag to be cleared.
5186 txctrl |= E1000_DCA_TXCTRL_DESC_RRO_EN |
5187 E1000_DCA_TXCTRL_DATA_RRO_EN |
5188 E1000_DCA_TXCTRL_DESC_DCA_EN;
5190 wr32(E1000_DCA_TXCTRL(tx_ring->reg_idx), txctrl);
5193 static void igb_update_rx_dca(struct igb_adapter *adapter,
5194 struct igb_ring *rx_ring,
5197 struct e1000_hw *hw = &adapter->hw;
5198 u32 rxctrl = dca3_get_tag(&adapter->pdev->dev, cpu);
5200 if (hw->mac.type != e1000_82575)
5201 rxctrl <<= E1000_DCA_RXCTRL_CPUID_SHIFT;
5203 /* We can enable relaxed ordering for reads, but not writes when
5204 * DCA is enabled. This is due to a known issue in some chipsets
5205 * which will cause the DCA tag to be cleared.
5207 rxctrl |= E1000_DCA_RXCTRL_DESC_RRO_EN |
5208 E1000_DCA_RXCTRL_DESC_DCA_EN;
5210 wr32(E1000_DCA_RXCTRL(rx_ring->reg_idx), rxctrl);
5213 static void igb_update_dca(struct igb_q_vector *q_vector)
5215 struct igb_adapter *adapter = q_vector->adapter;
5216 int cpu = get_cpu();
5218 if (q_vector->cpu == cpu)
5221 if (q_vector->tx.ring)
5222 igb_update_tx_dca(adapter, q_vector->tx.ring, cpu);
5224 if (q_vector->rx.ring)
5225 igb_update_rx_dca(adapter, q_vector->rx.ring, cpu);
5227 q_vector->cpu = cpu;
5232 static void igb_setup_dca(struct igb_adapter *adapter)
5234 struct e1000_hw *hw = &adapter->hw;
5237 if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
5240 /* Always use CB2 mode, difference is masked in the CB driver. */
5241 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
5243 for (i = 0; i < adapter->num_q_vectors; i++) {
5244 adapter->q_vector[i]->cpu = -1;
5245 igb_update_dca(adapter->q_vector[i]);
5249 static int __igb_notify_dca(struct device *dev, void *data)
5251 struct net_device *netdev = dev_get_drvdata(dev);
5252 struct igb_adapter *adapter = netdev_priv(netdev);
5253 struct pci_dev *pdev = adapter->pdev;
5254 struct e1000_hw *hw = &adapter->hw;
5255 unsigned long event = *(unsigned long *)data;
5258 case DCA_PROVIDER_ADD:
5259 /* if already enabled, don't do it again */
5260 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
5262 if (dca_add_requester(dev) == 0) {
5263 adapter->flags |= IGB_FLAG_DCA_ENABLED;
5264 dev_info(&pdev->dev, "DCA enabled\n");
5265 igb_setup_dca(adapter);
5268 /* Fall Through since DCA is disabled. */
5269 case DCA_PROVIDER_REMOVE:
5270 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
5271 /* without this a class_device is left
5272 * hanging around in the sysfs model
5274 dca_remove_requester(dev);
5275 dev_info(&pdev->dev, "DCA disabled\n");
5276 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
5277 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
5285 static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
5290 ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
5293 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
5295 #endif /* CONFIG_IGB_DCA */
5297 #ifdef CONFIG_PCI_IOV
5298 static int igb_vf_configure(struct igb_adapter *adapter, int vf)
5300 unsigned char mac_addr[ETH_ALEN];
5302 eth_zero_addr(mac_addr);
5303 igb_set_vf_mac(adapter, vf, mac_addr);
5305 /* By default spoof check is enabled for all VFs */
5306 adapter->vf_data[vf].spoofchk_enabled = true;
5312 static void igb_ping_all_vfs(struct igb_adapter *adapter)
5314 struct e1000_hw *hw = &adapter->hw;
5318 for (i = 0 ; i < adapter->vfs_allocated_count; i++) {
5319 ping = E1000_PF_CONTROL_MSG;
5320 if (adapter->vf_data[i].flags & IGB_VF_FLAG_CTS)
5321 ping |= E1000_VT_MSGTYPE_CTS;
5322 igb_write_mbx(hw, &ping, 1, i);
5326 static int igb_set_vf_promisc(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
5328 struct e1000_hw *hw = &adapter->hw;
5329 u32 vmolr = rd32(E1000_VMOLR(vf));
5330 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5332 vf_data->flags &= ~(IGB_VF_FLAG_UNI_PROMISC |
5333 IGB_VF_FLAG_MULTI_PROMISC);
5334 vmolr &= ~(E1000_VMOLR_ROPE | E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
5336 if (*msgbuf & E1000_VF_SET_PROMISC_MULTICAST) {
5337 vmolr |= E1000_VMOLR_MPME;
5338 vf_data->flags |= IGB_VF_FLAG_MULTI_PROMISC;
5339 *msgbuf &= ~E1000_VF_SET_PROMISC_MULTICAST;
5341 /* if we have hashes and we are clearing a multicast promisc
5342 * flag we need to write the hashes to the MTA as this step
5343 * was previously skipped
5345 if (vf_data->num_vf_mc_hashes > 30) {
5346 vmolr |= E1000_VMOLR_MPME;
5347 } else if (vf_data->num_vf_mc_hashes) {
5349 vmolr |= E1000_VMOLR_ROMPE;
5350 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
5351 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
5355 wr32(E1000_VMOLR(vf), vmolr);
5357 /* there are flags left unprocessed, likely not supported */
5358 if (*msgbuf & E1000_VT_MSGINFO_MASK)
5364 static int igb_set_vf_multicasts(struct igb_adapter *adapter,
5365 u32 *msgbuf, u32 vf)
5367 int n = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
5368 u16 *hash_list = (u16 *)&msgbuf[1];
5369 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5372 /* salt away the number of multicast addresses assigned
5373 * to this VF for later use to restore when the PF multi cast
5376 vf_data->num_vf_mc_hashes = n;
5378 /* only up to 30 hash values supported */
5382 /* store the hashes for later use */
5383 for (i = 0; i < n; i++)
5384 vf_data->vf_mc_hashes[i] = hash_list[i];
5386 /* Flush and reset the mta with the new values */
5387 igb_set_rx_mode(adapter->netdev);
5392 static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
5394 struct e1000_hw *hw = &adapter->hw;
5395 struct vf_data_storage *vf_data;
5398 for (i = 0; i < adapter->vfs_allocated_count; i++) {
5399 u32 vmolr = rd32(E1000_VMOLR(i));
5400 vmolr &= ~(E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
5402 vf_data = &adapter->vf_data[i];
5404 if ((vf_data->num_vf_mc_hashes > 30) ||
5405 (vf_data->flags & IGB_VF_FLAG_MULTI_PROMISC)) {
5406 vmolr |= E1000_VMOLR_MPME;
5407 } else if (vf_data->num_vf_mc_hashes) {
5408 vmolr |= E1000_VMOLR_ROMPE;
5409 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
5410 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
5412 wr32(E1000_VMOLR(i), vmolr);
5416 static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
5418 struct e1000_hw *hw = &adapter->hw;
5419 u32 pool_mask, reg, vid;
5422 pool_mask = 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
5424 /* Find the vlan filter for this id */
5425 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5426 reg = rd32(E1000_VLVF(i));
5428 /* remove the vf from the pool */
5431 /* if pool is empty then remove entry from vfta */
5432 if (!(reg & E1000_VLVF_POOLSEL_MASK) &&
5433 (reg & E1000_VLVF_VLANID_ENABLE)) {
5435 vid = reg & E1000_VLVF_VLANID_MASK;
5436 igb_vfta_set(hw, vid, false);
5439 wr32(E1000_VLVF(i), reg);
5442 adapter->vf_data[vf].vlans_enabled = 0;
5445 static s32 igb_vlvf_set(struct igb_adapter *adapter, u32 vid, bool add, u32 vf)
5447 struct e1000_hw *hw = &adapter->hw;
5450 /* The vlvf table only exists on 82576 hardware and newer */
5451 if (hw->mac.type < e1000_82576)
5454 /* we only need to do this if VMDq is enabled */
5455 if (!adapter->vfs_allocated_count)
5458 /* Find the vlan filter for this id */
5459 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5460 reg = rd32(E1000_VLVF(i));
5461 if ((reg & E1000_VLVF_VLANID_ENABLE) &&
5462 vid == (reg & E1000_VLVF_VLANID_MASK))
5467 if (i == E1000_VLVF_ARRAY_SIZE) {
5468 /* Did not find a matching VLAN ID entry that was
5469 * enabled. Search for a free filter entry, i.e.
5470 * one without the enable bit set
5472 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5473 reg = rd32(E1000_VLVF(i));
5474 if (!(reg & E1000_VLVF_VLANID_ENABLE))
5478 if (i < E1000_VLVF_ARRAY_SIZE) {
5479 /* Found an enabled/available entry */
5480 reg |= 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
5482 /* if !enabled we need to set this up in vfta */
5483 if (!(reg & E1000_VLVF_VLANID_ENABLE)) {
5484 /* add VID to filter table */
5485 igb_vfta_set(hw, vid, true);
5486 reg |= E1000_VLVF_VLANID_ENABLE;
5488 reg &= ~E1000_VLVF_VLANID_MASK;
5490 wr32(E1000_VLVF(i), reg);
5492 /* do not modify RLPML for PF devices */
5493 if (vf >= adapter->vfs_allocated_count)
5496 if (!adapter->vf_data[vf].vlans_enabled) {
5498 reg = rd32(E1000_VMOLR(vf));
5499 size = reg & E1000_VMOLR_RLPML_MASK;
5501 reg &= ~E1000_VMOLR_RLPML_MASK;
5503 wr32(E1000_VMOLR(vf), reg);
5506 adapter->vf_data[vf].vlans_enabled++;
5509 if (i < E1000_VLVF_ARRAY_SIZE) {
5510 /* remove vf from the pool */
5511 reg &= ~(1 << (E1000_VLVF_POOLSEL_SHIFT + vf));
5512 /* if pool is empty then remove entry from vfta */
5513 if (!(reg & E1000_VLVF_POOLSEL_MASK)) {
5515 igb_vfta_set(hw, vid, false);
5517 wr32(E1000_VLVF(i), reg);
5519 /* do not modify RLPML for PF devices */
5520 if (vf >= adapter->vfs_allocated_count)
5523 adapter->vf_data[vf].vlans_enabled--;
5524 if (!adapter->vf_data[vf].vlans_enabled) {
5526 reg = rd32(E1000_VMOLR(vf));
5527 size = reg & E1000_VMOLR_RLPML_MASK;
5529 reg &= ~E1000_VMOLR_RLPML_MASK;
5531 wr32(E1000_VMOLR(vf), reg);
5538 static void igb_set_vmvir(struct igb_adapter *adapter, u32 vid, u32 vf)
5540 struct e1000_hw *hw = &adapter->hw;
5543 wr32(E1000_VMVIR(vf), (vid | E1000_VMVIR_VLANA_DEFAULT));
5545 wr32(E1000_VMVIR(vf), 0);
5548 static int igb_ndo_set_vf_vlan(struct net_device *netdev,
5549 int vf, u16 vlan, u8 qos)
5552 struct igb_adapter *adapter = netdev_priv(netdev);
5554 if ((vf >= adapter->vfs_allocated_count) || (vlan > 4095) || (qos > 7))
5557 err = igb_vlvf_set(adapter, vlan, !!vlan, vf);
5560 igb_set_vmvir(adapter, vlan | (qos << VLAN_PRIO_SHIFT), vf);
5561 igb_set_vmolr(adapter, vf, !vlan);
5562 adapter->vf_data[vf].pf_vlan = vlan;
5563 adapter->vf_data[vf].pf_qos = qos;
5564 dev_info(&adapter->pdev->dev,
5565 "Setting VLAN %d, QOS 0x%x on VF %d\n", vlan, qos, vf);
5566 if (test_bit(__IGB_DOWN, &adapter->state)) {
5567 dev_warn(&adapter->pdev->dev,
5568 "The VF VLAN has been set, but the PF device is not up.\n");
5569 dev_warn(&adapter->pdev->dev,
5570 "Bring the PF device up before attempting to use the VF device.\n");
5573 igb_vlvf_set(adapter, adapter->vf_data[vf].pf_vlan,
5575 igb_set_vmvir(adapter, vlan, vf);
5576 igb_set_vmolr(adapter, vf, true);
5577 adapter->vf_data[vf].pf_vlan = 0;
5578 adapter->vf_data[vf].pf_qos = 0;
5584 static int igb_find_vlvf_entry(struct igb_adapter *adapter, int vid)
5586 struct e1000_hw *hw = &adapter->hw;
5590 /* Find the vlan filter for this id */
5591 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5592 reg = rd32(E1000_VLVF(i));
5593 if ((reg & E1000_VLVF_VLANID_ENABLE) &&
5594 vid == (reg & E1000_VLVF_VLANID_MASK))
5598 if (i >= E1000_VLVF_ARRAY_SIZE)
5604 static int igb_set_vf_vlan(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
5606 struct e1000_hw *hw = &adapter->hw;
5607 int add = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
5608 int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
5611 /* If in promiscuous mode we need to make sure the PF also has
5612 * the VLAN filter set.
5614 if (add && (adapter->netdev->flags & IFF_PROMISC))
5615 err = igb_vlvf_set(adapter, vid, add,
5616 adapter->vfs_allocated_count);
5620 err = igb_vlvf_set(adapter, vid, add, vf);
5625 /* Go through all the checks to see if the VLAN filter should
5626 * be wiped completely.
5628 if (!add && (adapter->netdev->flags & IFF_PROMISC)) {
5631 int regndx = igb_find_vlvf_entry(adapter, vid);
5634 /* See if any other pools are set for this VLAN filter
5635 * entry other than the PF.
5637 vlvf = bits = rd32(E1000_VLVF(regndx));
5638 bits &= 1 << (E1000_VLVF_POOLSEL_SHIFT +
5639 adapter->vfs_allocated_count);
5640 /* If the filter was removed then ensure PF pool bit
5641 * is cleared if the PF only added itself to the pool
5642 * because the PF is in promiscuous mode.
5644 if ((vlvf & VLAN_VID_MASK) == vid &&
5645 !test_bit(vid, adapter->active_vlans) &&
5647 igb_vlvf_set(adapter, vid, add,
5648 adapter->vfs_allocated_count);
5655 static inline void igb_vf_reset(struct igb_adapter *adapter, u32 vf)
5657 /* clear flags - except flag that indicates PF has set the MAC */
5658 adapter->vf_data[vf].flags &= IGB_VF_FLAG_PF_SET_MAC;
5659 adapter->vf_data[vf].last_nack = jiffies;
5661 /* reset offloads to defaults */
5662 igb_set_vmolr(adapter, vf, true);
5664 /* reset vlans for device */
5665 igb_clear_vf_vfta(adapter, vf);
5666 if (adapter->vf_data[vf].pf_vlan)
5667 igb_ndo_set_vf_vlan(adapter->netdev, vf,
5668 adapter->vf_data[vf].pf_vlan,
5669 adapter->vf_data[vf].pf_qos);
5671 igb_clear_vf_vfta(adapter, vf);
5673 /* reset multicast table array for vf */
5674 adapter->vf_data[vf].num_vf_mc_hashes = 0;
5676 /* Flush and reset the mta with the new values */
5677 igb_set_rx_mode(adapter->netdev);
5680 static void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
5682 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
5684 /* clear mac address as we were hotplug removed/added */
5685 if (!(adapter->vf_data[vf].flags & IGB_VF_FLAG_PF_SET_MAC))
5686 eth_zero_addr(vf_mac);
5688 /* process remaining reset events */
5689 igb_vf_reset(adapter, vf);
5692 static void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
5694 struct e1000_hw *hw = &adapter->hw;
5695 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
5696 int rar_entry = hw->mac.rar_entry_count - (vf + 1);
5698 u8 *addr = (u8 *)(&msgbuf[1]);
5700 /* process all the same items cleared in a function level reset */
5701 igb_vf_reset(adapter, vf);
5703 /* set vf mac address */
5704 igb_rar_set_qsel(adapter, vf_mac, rar_entry, vf);
5706 /* enable transmit and receive for vf */
5707 reg = rd32(E1000_VFTE);
5708 wr32(E1000_VFTE, reg | (1 << vf));
5709 reg = rd32(E1000_VFRE);
5710 wr32(E1000_VFRE, reg | (1 << vf));
5712 adapter->vf_data[vf].flags |= IGB_VF_FLAG_CTS;
5714 /* reply to reset with ack and vf mac address */
5715 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
5716 memcpy(addr, vf_mac, 6);
5717 igb_write_mbx(hw, msgbuf, 3, vf);
5720 static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf)
5722 /* The VF MAC Address is stored in a packed array of bytes
5723 * starting at the second 32 bit word of the msg array
5725 unsigned char *addr = (char *)&msg[1];
5728 if (is_valid_ether_addr(addr))
5729 err = igb_set_vf_mac(adapter, vf, addr);
5734 static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf)
5736 struct e1000_hw *hw = &adapter->hw;
5737 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5738 u32 msg = E1000_VT_MSGTYPE_NACK;
5740 /* if device isn't clear to send it shouldn't be reading either */
5741 if (!(vf_data->flags & IGB_VF_FLAG_CTS) &&
5742 time_after(jiffies, vf_data->last_nack + (2 * HZ))) {
5743 igb_write_mbx(hw, &msg, 1, vf);
5744 vf_data->last_nack = jiffies;
5748 static void igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
5750 struct pci_dev *pdev = adapter->pdev;
5751 u32 msgbuf[E1000_VFMAILBOX_SIZE];
5752 struct e1000_hw *hw = &adapter->hw;
5753 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5756 retval = igb_read_mbx(hw, msgbuf, E1000_VFMAILBOX_SIZE, vf);
5759 /* if receive failed revoke VF CTS stats and restart init */
5760 dev_err(&pdev->dev, "Error receiving message from VF\n");
5761 vf_data->flags &= ~IGB_VF_FLAG_CTS;
5762 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
5767 /* this is a message we already processed, do nothing */
5768 if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
5771 /* until the vf completes a reset it should not be
5772 * allowed to start any configuration.
5774 if (msgbuf[0] == E1000_VF_RESET) {
5775 igb_vf_reset_msg(adapter, vf);
5779 if (!(vf_data->flags & IGB_VF_FLAG_CTS)) {
5780 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
5786 switch ((msgbuf[0] & 0xFFFF)) {
5787 case E1000_VF_SET_MAC_ADDR:
5789 if (!(vf_data->flags & IGB_VF_FLAG_PF_SET_MAC))
5790 retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
5792 dev_warn(&pdev->dev,
5793 "VF %d attempted to override administratively set MAC address\nReload the VF driver to resume operations\n",
5796 case E1000_VF_SET_PROMISC:
5797 retval = igb_set_vf_promisc(adapter, msgbuf, vf);
5799 case E1000_VF_SET_MULTICAST:
5800 retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
5802 case E1000_VF_SET_LPE:
5803 retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf);
5805 case E1000_VF_SET_VLAN:
5807 if (vf_data->pf_vlan)
5808 dev_warn(&pdev->dev,
5809 "VF %d attempted to override administratively set VLAN tag\nReload the VF driver to resume operations\n",
5812 retval = igb_set_vf_vlan(adapter, msgbuf, vf);
5815 dev_err(&pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
5820 msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
5822 /* notify the VF of the results of what it sent us */
5824 msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
5826 msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
5828 igb_write_mbx(hw, msgbuf, 1, vf);
5831 static void igb_msg_task(struct igb_adapter *adapter)
5833 struct e1000_hw *hw = &adapter->hw;
5836 for (vf = 0; vf < adapter->vfs_allocated_count; vf++) {
5837 /* process any reset requests */
5838 if (!igb_check_for_rst(hw, vf))
5839 igb_vf_reset_event(adapter, vf);
5841 /* process any messages pending */
5842 if (!igb_check_for_msg(hw, vf))
5843 igb_rcv_msg_from_vf(adapter, vf);
5845 /* process any acks */
5846 if (!igb_check_for_ack(hw, vf))
5847 igb_rcv_ack_from_vf(adapter, vf);
5852 * igb_set_uta - Set unicast filter table address
5853 * @adapter: board private structure
5855 * The unicast table address is a register array of 32-bit registers.
5856 * The table is meant to be used in a way similar to how the MTA is used
5857 * however due to certain limitations in the hardware it is necessary to
5858 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
5859 * enable bit to allow vlan tag stripping when promiscuous mode is enabled
5861 static void igb_set_uta(struct igb_adapter *adapter)
5863 struct e1000_hw *hw = &adapter->hw;
5866 /* The UTA table only exists on 82576 hardware and newer */
5867 if (hw->mac.type < e1000_82576)
5870 /* we only need to do this if VMDq is enabled */
5871 if (!adapter->vfs_allocated_count)
5874 for (i = 0; i < hw->mac.uta_reg_count; i++)
5875 array_wr32(E1000_UTA, i, ~0);
5879 * igb_intr_msi - Interrupt Handler
5880 * @irq: interrupt number
5881 * @data: pointer to a network interface device structure
5883 static irqreturn_t igb_intr_msi(int irq, void *data)
5885 struct igb_adapter *adapter = data;
5886 struct igb_q_vector *q_vector = adapter->q_vector[0];
5887 struct e1000_hw *hw = &adapter->hw;
5888 /* read ICR disables interrupts using IAM */
5889 u32 icr = rd32(E1000_ICR);
5891 igb_write_itr(q_vector);
5893 if (icr & E1000_ICR_DRSTA)
5894 schedule_work(&adapter->reset_task);
5896 if (icr & E1000_ICR_DOUTSYNC) {
5897 /* HW is reporting DMA is out of sync */
5898 adapter->stats.doosync++;
5901 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
5902 hw->mac.get_link_status = 1;
5903 if (!test_bit(__IGB_DOWN, &adapter->state))
5904 mod_timer(&adapter->watchdog_timer, jiffies + 1);
5907 if (icr & E1000_ICR_TS) {
5908 u32 tsicr = rd32(E1000_TSICR);
5910 if (tsicr & E1000_TSICR_TXTS) {
5911 /* acknowledge the interrupt */
5912 wr32(E1000_TSICR, E1000_TSICR_TXTS);
5913 /* retrieve hardware timestamp */
5914 schedule_work(&adapter->ptp_tx_work);
5918 napi_schedule(&q_vector->napi);
5924 * igb_intr - Legacy Interrupt Handler
5925 * @irq: interrupt number
5926 * @data: pointer to a network interface device structure
5928 static irqreturn_t igb_intr(int irq, void *data)
5930 struct igb_adapter *adapter = data;
5931 struct igb_q_vector *q_vector = adapter->q_vector[0];
5932 struct e1000_hw *hw = &adapter->hw;
5933 /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked. No
5934 * need for the IMC write
5936 u32 icr = rd32(E1000_ICR);
5938 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
5939 * not set, then the adapter didn't send an interrupt
5941 if (!(icr & E1000_ICR_INT_ASSERTED))
5944 igb_write_itr(q_vector);
5946 if (icr & E1000_ICR_DRSTA)
5947 schedule_work(&adapter->reset_task);
5949 if (icr & E1000_ICR_DOUTSYNC) {
5950 /* HW is reporting DMA is out of sync */
5951 adapter->stats.doosync++;
5954 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
5955 hw->mac.get_link_status = 1;
5956 /* guard against interrupt when we're going down */
5957 if (!test_bit(__IGB_DOWN, &adapter->state))
5958 mod_timer(&adapter->watchdog_timer, jiffies + 1);
5961 if (icr & E1000_ICR_TS) {
5962 u32 tsicr = rd32(E1000_TSICR);
5964 if (tsicr & E1000_TSICR_TXTS) {
5965 /* acknowledge the interrupt */
5966 wr32(E1000_TSICR, E1000_TSICR_TXTS);
5967 /* retrieve hardware timestamp */
5968 schedule_work(&adapter->ptp_tx_work);
5972 napi_schedule(&q_vector->napi);
5977 static void igb_ring_irq_enable(struct igb_q_vector *q_vector)
5979 struct igb_adapter *adapter = q_vector->adapter;
5980 struct e1000_hw *hw = &adapter->hw;
5982 if ((q_vector->rx.ring && (adapter->rx_itr_setting & 3)) ||
5983 (!q_vector->rx.ring && (adapter->tx_itr_setting & 3))) {
5984 if ((adapter->num_q_vectors == 1) && !adapter->vf_data)
5985 igb_set_itr(q_vector);
5987 igb_update_ring_itr(q_vector);
5990 if (!test_bit(__IGB_DOWN, &adapter->state)) {
5991 if (adapter->msix_entries)
5992 wr32(E1000_EIMS, q_vector->eims_value);
5994 igb_irq_enable(adapter);
5999 * igb_poll - NAPI Rx polling callback
6000 * @napi: napi polling structure
6001 * @budget: count of how many packets we should handle
6003 static int igb_poll(struct napi_struct *napi, int budget)
6005 struct igb_q_vector *q_vector = container_of(napi,
6006 struct igb_q_vector,
6008 bool clean_complete = true;
6010 #ifdef CONFIG_IGB_DCA
6011 if (q_vector->adapter->flags & IGB_FLAG_DCA_ENABLED)
6012 igb_update_dca(q_vector);
6014 if (q_vector->tx.ring)
6015 clean_complete = igb_clean_tx_irq(q_vector);
6017 if (q_vector->rx.ring)
6018 clean_complete &= igb_clean_rx_irq(q_vector, budget);
6020 /* If all work not completed, return budget and keep polling */
6021 if (!clean_complete)
6024 /* If not enough Rx work done, exit the polling mode */
6025 napi_complete(napi);
6026 igb_ring_irq_enable(q_vector);
6032 * igb_clean_tx_irq - Reclaim resources after transmit completes
6033 * @q_vector: pointer to q_vector containing needed info
6035 * returns true if ring is completely cleaned
6037 static bool igb_clean_tx_irq(struct igb_q_vector *q_vector)
6039 struct igb_adapter *adapter = q_vector->adapter;
6040 struct igb_ring *tx_ring = q_vector->tx.ring;
6041 struct igb_tx_buffer *tx_buffer;
6042 union e1000_adv_tx_desc *tx_desc;
6043 unsigned int total_bytes = 0, total_packets = 0;
6044 unsigned int budget = q_vector->tx.work_limit;
6045 unsigned int i = tx_ring->next_to_clean;
6047 if (test_bit(__IGB_DOWN, &adapter->state))
6050 tx_buffer = &tx_ring->tx_buffer_info[i];
6051 tx_desc = IGB_TX_DESC(tx_ring, i);
6052 i -= tx_ring->count;
6055 union e1000_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
6057 /* if next_to_watch is not set then there is no work pending */
6061 /* prevent any other reads prior to eop_desc */
6062 read_barrier_depends();
6064 /* if DD is not set pending work has not been completed */
6065 if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)))
6068 /* clear next_to_watch to prevent false hangs */
6069 tx_buffer->next_to_watch = NULL;
6071 /* update the statistics for this packet */
6072 total_bytes += tx_buffer->bytecount;
6073 total_packets += tx_buffer->gso_segs;
6076 dev_kfree_skb_any(tx_buffer->skb);
6078 /* unmap skb header data */
6079 dma_unmap_single(tx_ring->dev,
6080 dma_unmap_addr(tx_buffer, dma),
6081 dma_unmap_len(tx_buffer, len),
6084 /* clear tx_buffer data */
6085 tx_buffer->skb = NULL;
6086 dma_unmap_len_set(tx_buffer, len, 0);
6088 /* clear last DMA location and unmap remaining buffers */
6089 while (tx_desc != eop_desc) {
6094 i -= tx_ring->count;
6095 tx_buffer = tx_ring->tx_buffer_info;
6096 tx_desc = IGB_TX_DESC(tx_ring, 0);
6099 /* unmap any remaining paged data */
6100 if (dma_unmap_len(tx_buffer, len)) {
6101 dma_unmap_page(tx_ring->dev,
6102 dma_unmap_addr(tx_buffer, dma),
6103 dma_unmap_len(tx_buffer, len),
6105 dma_unmap_len_set(tx_buffer, len, 0);
6109 /* move us one more past the eop_desc for start of next pkt */
6114 i -= tx_ring->count;
6115 tx_buffer = tx_ring->tx_buffer_info;
6116 tx_desc = IGB_TX_DESC(tx_ring, 0);
6119 /* issue prefetch for next Tx descriptor */
6122 /* update budget accounting */
6124 } while (likely(budget));
6126 netdev_tx_completed_queue(txring_txq(tx_ring),
6127 total_packets, total_bytes);
6128 i += tx_ring->count;
6129 tx_ring->next_to_clean = i;
6130 u64_stats_update_begin(&tx_ring->tx_syncp);
6131 tx_ring->tx_stats.bytes += total_bytes;
6132 tx_ring->tx_stats.packets += total_packets;
6133 u64_stats_update_end(&tx_ring->tx_syncp);
6134 q_vector->tx.total_bytes += total_bytes;
6135 q_vector->tx.total_packets += total_packets;
6137 if (test_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags)) {
6138 struct e1000_hw *hw = &adapter->hw;
6140 /* Detect a transmit hang in hardware, this serializes the
6141 * check with the clearing of time_stamp and movement of i
6143 clear_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
6144 if (tx_buffer->next_to_watch &&
6145 time_after(jiffies, tx_buffer->time_stamp +
6146 (adapter->tx_timeout_factor * HZ)) &&
6147 !(rd32(E1000_STATUS) & E1000_STATUS_TXOFF)) {
6149 /* detected Tx unit hang */
6150 dev_err(tx_ring->dev,
6151 "Detected Tx Unit Hang\n"
6155 " next_to_use <%x>\n"
6156 " next_to_clean <%x>\n"
6157 "buffer_info[next_to_clean]\n"
6158 " time_stamp <%lx>\n"
6159 " next_to_watch <%p>\n"
6161 " desc.status <%x>\n",
6162 tx_ring->queue_index,
6163 rd32(E1000_TDH(tx_ring->reg_idx)),
6164 readl(tx_ring->tail),
6165 tx_ring->next_to_use,
6166 tx_ring->next_to_clean,
6167 tx_buffer->time_stamp,
6168 tx_buffer->next_to_watch,
6170 tx_buffer->next_to_watch->wb.status);
6171 netif_stop_subqueue(tx_ring->netdev,
6172 tx_ring->queue_index);
6174 /* we are about to reset, no point in enabling stuff */
6179 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
6180 if (unlikely(total_packets &&
6181 netif_carrier_ok(tx_ring->netdev) &&
6182 igb_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD)) {
6183 /* Make sure that anybody stopping the queue after this
6184 * sees the new next_to_clean.
6187 if (__netif_subqueue_stopped(tx_ring->netdev,
6188 tx_ring->queue_index) &&
6189 !(test_bit(__IGB_DOWN, &adapter->state))) {
6190 netif_wake_subqueue(tx_ring->netdev,
6191 tx_ring->queue_index);
6193 u64_stats_update_begin(&tx_ring->tx_syncp);
6194 tx_ring->tx_stats.restart_queue++;
6195 u64_stats_update_end(&tx_ring->tx_syncp);
6203 * igb_reuse_rx_page - page flip buffer and store it back on the ring
6204 * @rx_ring: rx descriptor ring to store buffers on
6205 * @old_buff: donor buffer to have page reused
6207 * Synchronizes page for reuse by the adapter
6209 static void igb_reuse_rx_page(struct igb_ring *rx_ring,
6210 struct igb_rx_buffer *old_buff)
6212 struct igb_rx_buffer *new_buff;
6213 u16 nta = rx_ring->next_to_alloc;
6215 new_buff = &rx_ring->rx_buffer_info[nta];
6217 /* update, and store next to alloc */
6219 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
6221 /* transfer page from old buffer to new buffer */
6222 memcpy(new_buff, old_buff, sizeof(struct igb_rx_buffer));
6224 /* sync the buffer for use by the device */
6225 dma_sync_single_range_for_device(rx_ring->dev, old_buff->dma,
6226 old_buff->page_offset,
6231 static bool igb_can_reuse_rx_page(struct igb_rx_buffer *rx_buffer,
6233 unsigned int truesize)
6235 /* avoid re-using remote pages */
6236 if (unlikely(page_to_nid(page) != numa_node_id()))
6239 #if (PAGE_SIZE < 8192)
6240 /* if we are only owner of page we can reuse it */
6241 if (unlikely(page_count(page) != 1))
6244 /* flip page offset to other buffer */
6245 rx_buffer->page_offset ^= IGB_RX_BUFSZ;
6247 /* since we are the only owner of the page and we need to
6248 * increment it, just set the value to 2 in order to avoid
6249 * an unnecessary locked operation
6251 atomic_set(&page->_count, 2);
6253 /* move offset up to the next cache line */
6254 rx_buffer->page_offset += truesize;
6256 if (rx_buffer->page_offset > (PAGE_SIZE - IGB_RX_BUFSZ))
6259 /* bump ref count on page before it is given to the stack */
6267 * igb_add_rx_frag - Add contents of Rx buffer to sk_buff
6268 * @rx_ring: rx descriptor ring to transact packets on
6269 * @rx_buffer: buffer containing page to add
6270 * @rx_desc: descriptor containing length of buffer written by hardware
6271 * @skb: sk_buff to place the data into
6273 * This function will add the data contained in rx_buffer->page to the skb.
6274 * This is done either through a direct copy if the data in the buffer is
6275 * less than the skb header size, otherwise it will just attach the page as
6276 * a frag to the skb.
6278 * The function will then update the page offset if necessary and return
6279 * true if the buffer can be reused by the adapter.
6281 static bool igb_add_rx_frag(struct igb_ring *rx_ring,
6282 struct igb_rx_buffer *rx_buffer,
6283 union e1000_adv_rx_desc *rx_desc,
6284 struct sk_buff *skb)
6286 struct page *page = rx_buffer->page;
6287 unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
6288 #if (PAGE_SIZE < 8192)
6289 unsigned int truesize = IGB_RX_BUFSZ;
6291 unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
6294 if ((size <= IGB_RX_HDR_LEN) && !skb_is_nonlinear(skb)) {
6295 unsigned char *va = page_address(page) + rx_buffer->page_offset;
6297 if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
6298 igb_ptp_rx_pktstamp(rx_ring->q_vector, va, skb);
6299 va += IGB_TS_HDR_LEN;
6300 size -= IGB_TS_HDR_LEN;
6303 memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
6305 /* we can reuse buffer as-is, just make sure it is local */
6306 if (likely(page_to_nid(page) == numa_node_id()))
6309 /* this page cannot be reused so discard it */
6314 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
6315 rx_buffer->page_offset, size, truesize);
6317 return igb_can_reuse_rx_page(rx_buffer, page, truesize);
6320 static struct sk_buff *igb_fetch_rx_buffer(struct igb_ring *rx_ring,
6321 union e1000_adv_rx_desc *rx_desc,
6322 struct sk_buff *skb)
6324 struct igb_rx_buffer *rx_buffer;
6327 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
6329 page = rx_buffer->page;
6333 void *page_addr = page_address(page) +
6334 rx_buffer->page_offset;
6336 /* prefetch first cache line of first page */
6337 prefetch(page_addr);
6338 #if L1_CACHE_BYTES < 128
6339 prefetch(page_addr + L1_CACHE_BYTES);
6342 /* allocate a skb to store the frags */
6343 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
6345 if (unlikely(!skb)) {
6346 rx_ring->rx_stats.alloc_failed++;
6350 /* we will be copying header into skb->data in
6351 * pskb_may_pull so it is in our interest to prefetch
6352 * it now to avoid a possible cache miss
6354 prefetchw(skb->data);
6357 /* we are reusing so sync this buffer for CPU use */
6358 dma_sync_single_range_for_cpu(rx_ring->dev,
6360 rx_buffer->page_offset,
6364 /* pull page into skb */
6365 if (igb_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
6366 /* hand second half of page back to the ring */
6367 igb_reuse_rx_page(rx_ring, rx_buffer);
6369 /* we are not reusing the buffer so unmap it */
6370 dma_unmap_page(rx_ring->dev, rx_buffer->dma,
6371 PAGE_SIZE, DMA_FROM_DEVICE);
6374 /* clear contents of rx_buffer */
6375 rx_buffer->page = NULL;
6380 static inline void igb_rx_checksum(struct igb_ring *ring,
6381 union e1000_adv_rx_desc *rx_desc,
6382 struct sk_buff *skb)
6384 skb_checksum_none_assert(skb);
6386 /* Ignore Checksum bit is set */
6387 if (igb_test_staterr(rx_desc, E1000_RXD_STAT_IXSM))
6390 /* Rx checksum disabled via ethtool */
6391 if (!(ring->netdev->features & NETIF_F_RXCSUM))
6394 /* TCP/UDP checksum error bit is set */
6395 if (igb_test_staterr(rx_desc,
6396 E1000_RXDEXT_STATERR_TCPE |
6397 E1000_RXDEXT_STATERR_IPE)) {
6398 /* work around errata with sctp packets where the TCPE aka
6399 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
6400 * packets, (aka let the stack check the crc32c)
6402 if (!((skb->len == 60) &&
6403 test_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags))) {
6404 u64_stats_update_begin(&ring->rx_syncp);
6405 ring->rx_stats.csum_err++;
6406 u64_stats_update_end(&ring->rx_syncp);
6408 /* let the stack verify checksum errors */
6411 /* It must be a TCP or UDP packet with a valid checksum */
6412 if (igb_test_staterr(rx_desc, E1000_RXD_STAT_TCPCS |
6413 E1000_RXD_STAT_UDPCS))
6414 skb->ip_summed = CHECKSUM_UNNECESSARY;
6416 dev_dbg(ring->dev, "cksum success: bits %08X\n",
6417 le32_to_cpu(rx_desc->wb.upper.status_error));
6420 static inline void igb_rx_hash(struct igb_ring *ring,
6421 union e1000_adv_rx_desc *rx_desc,
6422 struct sk_buff *skb)
6424 if (ring->netdev->features & NETIF_F_RXHASH)
6425 skb->rxhash = le32_to_cpu(rx_desc->wb.lower.hi_dword.rss);
6429 * igb_is_non_eop - process handling of non-EOP buffers
6430 * @rx_ring: Rx ring being processed
6431 * @rx_desc: Rx descriptor for current buffer
6432 * @skb: current socket buffer containing buffer in progress
6434 * This function updates next to clean. If the buffer is an EOP buffer
6435 * this function exits returning false, otherwise it will place the
6436 * sk_buff in the next buffer to be chained and return true indicating
6437 * that this is in fact a non-EOP buffer.
6439 static bool igb_is_non_eop(struct igb_ring *rx_ring,
6440 union e1000_adv_rx_desc *rx_desc)
6442 u32 ntc = rx_ring->next_to_clean + 1;
6444 /* fetch, update, and store next to clean */
6445 ntc = (ntc < rx_ring->count) ? ntc : 0;
6446 rx_ring->next_to_clean = ntc;
6448 prefetch(IGB_RX_DESC(rx_ring, ntc));
6450 if (likely(igb_test_staterr(rx_desc, E1000_RXD_STAT_EOP)))
6457 * igb_get_headlen - determine size of header for LRO/GRO
6458 * @data: pointer to the start of the headers
6459 * @max_len: total length of section to find headers in
6461 * This function is meant to determine the length of headers that will
6462 * be recognized by hardware for LRO, and GRO offloads. The main
6463 * motivation of doing this is to only perform one pull for IPv4 TCP
6464 * packets so that we can do basic things like calculating the gso_size
6465 * based on the average data per packet.
6467 static unsigned int igb_get_headlen(unsigned char *data,
6468 unsigned int max_len)
6471 unsigned char *network;
6474 struct vlan_hdr *vlan;
6477 struct ipv6hdr *ipv6;
6480 u8 nexthdr = 0; /* default to not TCP */
6483 /* this should never happen, but better safe than sorry */
6484 if (max_len < ETH_HLEN)
6487 /* initialize network frame pointer */
6490 /* set first protocol and move network header forward */
6491 protocol = hdr.eth->h_proto;
6492 hdr.network += ETH_HLEN;
6494 /* handle any vlan tag if present */
6495 if (protocol == __constant_htons(ETH_P_8021Q)) {
6496 if ((hdr.network - data) > (max_len - VLAN_HLEN))
6499 protocol = hdr.vlan->h_vlan_encapsulated_proto;
6500 hdr.network += VLAN_HLEN;
6503 /* handle L3 protocols */
6504 if (protocol == __constant_htons(ETH_P_IP)) {
6505 if ((hdr.network - data) > (max_len - sizeof(struct iphdr)))
6508 /* access ihl as a u8 to avoid unaligned access on ia64 */
6509 hlen = (hdr.network[0] & 0x0F) << 2;
6511 /* verify hlen meets minimum size requirements */
6512 if (hlen < sizeof(struct iphdr))
6513 return hdr.network - data;
6515 /* record next protocol if header is present */
6516 if (!(hdr.ipv4->frag_off & htons(IP_OFFSET)))
6517 nexthdr = hdr.ipv4->protocol;
6518 } else if (protocol == __constant_htons(ETH_P_IPV6)) {
6519 if ((hdr.network - data) > (max_len - sizeof(struct ipv6hdr)))
6522 /* record next protocol */
6523 nexthdr = hdr.ipv6->nexthdr;
6524 hlen = sizeof(struct ipv6hdr);
6526 return hdr.network - data;
6529 /* relocate pointer to start of L4 header */
6530 hdr.network += hlen;
6532 /* finally sort out TCP */
6533 if (nexthdr == IPPROTO_TCP) {
6534 if ((hdr.network - data) > (max_len - sizeof(struct tcphdr)))
6537 /* access doff as a u8 to avoid unaligned access on ia64 */
6538 hlen = (hdr.network[12] & 0xF0) >> 2;
6540 /* verify hlen meets minimum size requirements */
6541 if (hlen < sizeof(struct tcphdr))
6542 return hdr.network - data;
6544 hdr.network += hlen;
6545 } else if (nexthdr == IPPROTO_UDP) {
6546 if ((hdr.network - data) > (max_len - sizeof(struct udphdr)))
6549 hdr.network += sizeof(struct udphdr);
6552 /* If everything has gone correctly hdr.network should be the
6553 * data section of the packet and will be the end of the header.
6554 * If not then it probably represents the end of the last recognized
6557 if ((hdr.network - data) < max_len)
6558 return hdr.network - data;
6564 * igb_pull_tail - igb specific version of skb_pull_tail
6565 * @rx_ring: rx descriptor ring packet is being transacted on
6566 * @rx_desc: pointer to the EOP Rx descriptor
6567 * @skb: pointer to current skb being adjusted
6569 * This function is an igb specific version of __pskb_pull_tail. The
6570 * main difference between this version and the original function is that
6571 * this function can make several assumptions about the state of things
6572 * that allow for significant optimizations versus the standard function.
6573 * As a result we can do things like drop a frag and maintain an accurate
6574 * truesize for the skb.
6576 static void igb_pull_tail(struct igb_ring *rx_ring,
6577 union e1000_adv_rx_desc *rx_desc,
6578 struct sk_buff *skb)
6580 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
6582 unsigned int pull_len;
6584 /* it is valid to use page_address instead of kmap since we are
6585 * working with pages allocated out of the lomem pool per
6586 * alloc_page(GFP_ATOMIC)
6588 va = skb_frag_address(frag);
6590 if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
6591 /* retrieve timestamp from buffer */
6592 igb_ptp_rx_pktstamp(rx_ring->q_vector, va, skb);
6594 /* update pointers to remove timestamp header */
6595 skb_frag_size_sub(frag, IGB_TS_HDR_LEN);
6596 frag->page_offset += IGB_TS_HDR_LEN;
6597 skb->data_len -= IGB_TS_HDR_LEN;
6598 skb->len -= IGB_TS_HDR_LEN;
6600 /* move va to start of packet data */
6601 va += IGB_TS_HDR_LEN;
6604 /* we need the header to contain the greater of either ETH_HLEN or
6605 * 60 bytes if the skb->len is less than 60 for skb_pad.
6607 pull_len = igb_get_headlen(va, IGB_RX_HDR_LEN);
6609 /* align pull length to size of long to optimize memcpy performance */
6610 skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
6612 /* update all of the pointers */
6613 skb_frag_size_sub(frag, pull_len);
6614 frag->page_offset += pull_len;
6615 skb->data_len -= pull_len;
6616 skb->tail += pull_len;
6620 * igb_cleanup_headers - Correct corrupted or empty headers
6621 * @rx_ring: rx descriptor ring packet is being transacted on
6622 * @rx_desc: pointer to the EOP Rx descriptor
6623 * @skb: pointer to current skb being fixed
6625 * Address the case where we are pulling data in on pages only
6626 * and as such no data is present in the skb header.
6628 * In addition if skb is not at least 60 bytes we need to pad it so that
6629 * it is large enough to qualify as a valid Ethernet frame.
6631 * Returns true if an error was encountered and skb was freed.
6633 static bool igb_cleanup_headers(struct igb_ring *rx_ring,
6634 union e1000_adv_rx_desc *rx_desc,
6635 struct sk_buff *skb)
6637 if (unlikely((igb_test_staterr(rx_desc,
6638 E1000_RXDEXT_ERR_FRAME_ERR_MASK)))) {
6639 struct net_device *netdev = rx_ring->netdev;
6640 if (!(netdev->features & NETIF_F_RXALL)) {
6641 dev_kfree_skb_any(skb);
6646 /* place header in linear portion of buffer */
6647 if (skb_is_nonlinear(skb))
6648 igb_pull_tail(rx_ring, rx_desc, skb);
6650 /* if skb_pad returns an error the skb was freed */
6651 if (unlikely(skb->len < 60)) {
6652 int pad_len = 60 - skb->len;
6654 if (skb_pad(skb, pad_len))
6656 __skb_put(skb, pad_len);
6663 * igb_process_skb_fields - Populate skb header fields from Rx descriptor
6664 * @rx_ring: rx descriptor ring packet is being transacted on
6665 * @rx_desc: pointer to the EOP Rx descriptor
6666 * @skb: pointer to current skb being populated
6668 * This function checks the ring, descriptor, and packet information in
6669 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
6670 * other fields within the skb.
6672 static void igb_process_skb_fields(struct igb_ring *rx_ring,
6673 union e1000_adv_rx_desc *rx_desc,
6674 struct sk_buff *skb)
6676 struct net_device *dev = rx_ring->netdev;
6678 igb_rx_hash(rx_ring, rx_desc, skb);
6680 igb_rx_checksum(rx_ring, rx_desc, skb);
6682 igb_ptp_rx_hwtstamp(rx_ring, rx_desc, skb);
6684 if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
6685 igb_test_staterr(rx_desc, E1000_RXD_STAT_VP)) {
6687 if (igb_test_staterr(rx_desc, E1000_RXDEXT_STATERR_LB) &&
6688 test_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &rx_ring->flags))
6689 vid = be16_to_cpu(rx_desc->wb.upper.vlan);
6691 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
6693 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
6696 skb_record_rx_queue(skb, rx_ring->queue_index);
6698 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
6701 static bool igb_clean_rx_irq(struct igb_q_vector *q_vector, const int budget)
6703 struct igb_ring *rx_ring = q_vector->rx.ring;
6704 struct sk_buff *skb = rx_ring->skb;
6705 unsigned int total_bytes = 0, total_packets = 0;
6706 u16 cleaned_count = igb_desc_unused(rx_ring);
6709 union e1000_adv_rx_desc *rx_desc;
6711 /* return some buffers to hardware, one at a time is too slow */
6712 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
6713 igb_alloc_rx_buffers(rx_ring, cleaned_count);
6717 rx_desc = IGB_RX_DESC(rx_ring, rx_ring->next_to_clean);
6719 if (!igb_test_staterr(rx_desc, E1000_RXD_STAT_DD))
6722 /* This memory barrier is needed to keep us from reading
6723 * any other fields out of the rx_desc until we know the
6724 * RXD_STAT_DD bit is set
6728 /* retrieve a buffer from the ring */
6729 skb = igb_fetch_rx_buffer(rx_ring, rx_desc, skb);
6731 /* exit if we failed to retrieve a buffer */
6737 /* fetch next buffer in frame if non-eop */
6738 if (igb_is_non_eop(rx_ring, rx_desc))
6741 /* verify the packet layout is correct */
6742 if (igb_cleanup_headers(rx_ring, rx_desc, skb)) {
6747 /* probably a little skewed due to removing CRC */
6748 total_bytes += skb->len;
6750 /* populate checksum, timestamp, VLAN, and protocol */
6751 igb_process_skb_fields(rx_ring, rx_desc, skb);
6753 napi_gro_receive(&q_vector->napi, skb);
6755 /* reset skb pointer */
6758 /* update budget accounting */
6760 } while (likely(total_packets < budget));
6762 /* place incomplete frames back on ring for completion */
6765 u64_stats_update_begin(&rx_ring->rx_syncp);
6766 rx_ring->rx_stats.packets += total_packets;
6767 rx_ring->rx_stats.bytes += total_bytes;
6768 u64_stats_update_end(&rx_ring->rx_syncp);
6769 q_vector->rx.total_packets += total_packets;
6770 q_vector->rx.total_bytes += total_bytes;
6773 igb_alloc_rx_buffers(rx_ring, cleaned_count);
6775 return (total_packets < budget);
6778 static bool igb_alloc_mapped_page(struct igb_ring *rx_ring,
6779 struct igb_rx_buffer *bi)
6781 struct page *page = bi->page;
6784 /* since we are recycling buffers we should seldom need to alloc */
6788 /* alloc new page for storage */
6789 page = __skb_alloc_page(GFP_ATOMIC | __GFP_COLD, NULL);
6790 if (unlikely(!page)) {
6791 rx_ring->rx_stats.alloc_failed++;
6795 /* map page for use */
6796 dma = dma_map_page(rx_ring->dev, page, 0, PAGE_SIZE, DMA_FROM_DEVICE);
6798 /* if mapping failed free memory back to system since
6799 * there isn't much point in holding memory we can't use
6801 if (dma_mapping_error(rx_ring->dev, dma)) {
6804 rx_ring->rx_stats.alloc_failed++;
6810 bi->page_offset = 0;
6816 * igb_alloc_rx_buffers - Replace used receive buffers; packet split
6817 * @adapter: address of board private structure
6819 void igb_alloc_rx_buffers(struct igb_ring *rx_ring, u16 cleaned_count)
6821 union e1000_adv_rx_desc *rx_desc;
6822 struct igb_rx_buffer *bi;
6823 u16 i = rx_ring->next_to_use;
6829 rx_desc = IGB_RX_DESC(rx_ring, i);
6830 bi = &rx_ring->rx_buffer_info[i];
6831 i -= rx_ring->count;
6834 if (!igb_alloc_mapped_page(rx_ring, bi))
6837 /* Refresh the desc even if buffer_addrs didn't change
6838 * because each write-back erases this info.
6840 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
6846 rx_desc = IGB_RX_DESC(rx_ring, 0);
6847 bi = rx_ring->rx_buffer_info;
6848 i -= rx_ring->count;
6851 /* clear the hdr_addr for the next_to_use descriptor */
6852 rx_desc->read.hdr_addr = 0;
6855 } while (cleaned_count);
6857 i += rx_ring->count;
6859 if (rx_ring->next_to_use != i) {
6860 /* record the next descriptor to use */
6861 rx_ring->next_to_use = i;
6863 /* update next to alloc since we have filled the ring */
6864 rx_ring->next_to_alloc = i;
6866 /* Force memory writes to complete before letting h/w
6867 * know there are new descriptors to fetch. (Only
6868 * applicable for weak-ordered memory model archs,
6872 writel(i, rx_ring->tail);
6882 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
6884 struct igb_adapter *adapter = netdev_priv(netdev);
6885 struct mii_ioctl_data *data = if_mii(ifr);
6887 if (adapter->hw.phy.media_type != e1000_media_type_copper)
6892 data->phy_id = adapter->hw.phy.addr;
6895 if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
6912 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
6918 return igb_mii_ioctl(netdev, ifr, cmd);
6920 return igb_ptp_hwtstamp_ioctl(netdev, ifr, cmd);
6926 s32 igb_read_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
6928 struct igb_adapter *adapter = hw->back;
6930 if (pcie_capability_read_word(adapter->pdev, reg, value))
6931 return -E1000_ERR_CONFIG;
6936 s32 igb_write_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
6938 struct igb_adapter *adapter = hw->back;
6940 if (pcie_capability_write_word(adapter->pdev, reg, *value))
6941 return -E1000_ERR_CONFIG;
6946 static void igb_vlan_mode(struct net_device *netdev, netdev_features_t features)
6948 struct igb_adapter *adapter = netdev_priv(netdev);
6949 struct e1000_hw *hw = &adapter->hw;
6951 bool enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
6954 /* enable VLAN tag insert/strip */
6955 ctrl = rd32(E1000_CTRL);
6956 ctrl |= E1000_CTRL_VME;
6957 wr32(E1000_CTRL, ctrl);
6959 /* Disable CFI check */
6960 rctl = rd32(E1000_RCTL);
6961 rctl &= ~E1000_RCTL_CFIEN;
6962 wr32(E1000_RCTL, rctl);
6964 /* disable VLAN tag insert/strip */
6965 ctrl = rd32(E1000_CTRL);
6966 ctrl &= ~E1000_CTRL_VME;
6967 wr32(E1000_CTRL, ctrl);
6970 igb_rlpml_set(adapter);
6973 static int igb_vlan_rx_add_vid(struct net_device *netdev,
6974 __be16 proto, u16 vid)
6976 struct igb_adapter *adapter = netdev_priv(netdev);
6977 struct e1000_hw *hw = &adapter->hw;
6978 int pf_id = adapter->vfs_allocated_count;
6980 /* attempt to add filter to vlvf array */
6981 igb_vlvf_set(adapter, vid, true, pf_id);
6983 /* add the filter since PF can receive vlans w/o entry in vlvf */
6984 igb_vfta_set(hw, vid, true);
6986 set_bit(vid, adapter->active_vlans);
6991 static int igb_vlan_rx_kill_vid(struct net_device *netdev,
6992 __be16 proto, u16 vid)
6994 struct igb_adapter *adapter = netdev_priv(netdev);
6995 struct e1000_hw *hw = &adapter->hw;
6996 int pf_id = adapter->vfs_allocated_count;
6999 /* remove vlan from VLVF table array */
7000 err = igb_vlvf_set(adapter, vid, false, pf_id);
7002 /* if vid was not present in VLVF just remove it from table */
7004 igb_vfta_set(hw, vid, false);
7006 clear_bit(vid, adapter->active_vlans);
7011 static void igb_restore_vlan(struct igb_adapter *adapter)
7015 igb_vlan_mode(adapter->netdev, adapter->netdev->features);
7017 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
7018 igb_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
7021 int igb_set_spd_dplx(struct igb_adapter *adapter, u32 spd, u8 dplx)
7023 struct pci_dev *pdev = adapter->pdev;
7024 struct e1000_mac_info *mac = &adapter->hw.mac;
7028 /* Make sure dplx is at most 1 bit and lsb of speed is not set
7029 * for the switch() below to work
7031 if ((spd & 1) || (dplx & ~1))
7034 /* Fiber NIC's only allow 1000 gbps Full duplex
7035 * and 100Mbps Full duplex for 100baseFx sfp
7037 if (adapter->hw.phy.media_type == e1000_media_type_internal_serdes) {
7038 switch (spd + dplx) {
7039 case SPEED_10 + DUPLEX_HALF:
7040 case SPEED_10 + DUPLEX_FULL:
7041 case SPEED_100 + DUPLEX_HALF:
7048 switch (spd + dplx) {
7049 case SPEED_10 + DUPLEX_HALF:
7050 mac->forced_speed_duplex = ADVERTISE_10_HALF;
7052 case SPEED_10 + DUPLEX_FULL:
7053 mac->forced_speed_duplex = ADVERTISE_10_FULL;
7055 case SPEED_100 + DUPLEX_HALF:
7056 mac->forced_speed_duplex = ADVERTISE_100_HALF;
7058 case SPEED_100 + DUPLEX_FULL:
7059 mac->forced_speed_duplex = ADVERTISE_100_FULL;
7061 case SPEED_1000 + DUPLEX_FULL:
7063 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
7065 case SPEED_1000 + DUPLEX_HALF: /* not supported */
7070 /* clear MDI, MDI(-X) override is only allowed when autoneg enabled */
7071 adapter->hw.phy.mdix = AUTO_ALL_MODES;
7076 dev_err(&pdev->dev, "Unsupported Speed/Duplex configuration\n");
7080 static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake,
7083 struct net_device *netdev = pci_get_drvdata(pdev);
7084 struct igb_adapter *adapter = netdev_priv(netdev);
7085 struct e1000_hw *hw = &adapter->hw;
7086 u32 ctrl, rctl, status;
7087 u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
7092 netif_device_detach(netdev);
7094 if (netif_running(netdev))
7095 __igb_close(netdev, true);
7097 igb_clear_interrupt_scheme(adapter);
7100 retval = pci_save_state(pdev);
7105 status = rd32(E1000_STATUS);
7106 if (status & E1000_STATUS_LU)
7107 wufc &= ~E1000_WUFC_LNKC;
7110 igb_setup_rctl(adapter);
7111 igb_set_rx_mode(netdev);
7113 /* turn on all-multi mode if wake on multicast is enabled */
7114 if (wufc & E1000_WUFC_MC) {
7115 rctl = rd32(E1000_RCTL);
7116 rctl |= E1000_RCTL_MPE;
7117 wr32(E1000_RCTL, rctl);
7120 ctrl = rd32(E1000_CTRL);
7121 /* advertise wake from D3Cold */
7122 #define E1000_CTRL_ADVD3WUC 0x00100000
7123 /* phy power management enable */
7124 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
7125 ctrl |= E1000_CTRL_ADVD3WUC;
7126 wr32(E1000_CTRL, ctrl);
7128 /* Allow time for pending master requests to run */
7129 igb_disable_pcie_master(hw);
7131 wr32(E1000_WUC, E1000_WUC_PME_EN);
7132 wr32(E1000_WUFC, wufc);
7135 wr32(E1000_WUFC, 0);
7138 *enable_wake = wufc || adapter->en_mng_pt;
7140 igb_power_down_link(adapter);
7142 igb_power_up_link(adapter);
7144 /* Release control of h/w to f/w. If f/w is AMT enabled, this
7145 * would have already happened in close and is redundant.
7147 igb_release_hw_control(adapter);
7149 pci_disable_device(pdev);
7155 #ifdef CONFIG_PM_SLEEP
7156 static int igb_suspend(struct device *dev)
7160 struct pci_dev *pdev = to_pci_dev(dev);
7162 retval = __igb_shutdown(pdev, &wake, 0);
7167 pci_prepare_to_sleep(pdev);
7169 pci_wake_from_d3(pdev, false);
7170 pci_set_power_state(pdev, PCI_D3hot);
7175 #endif /* CONFIG_PM_SLEEP */
7177 static int igb_resume(struct device *dev)
7179 struct pci_dev *pdev = to_pci_dev(dev);
7180 struct net_device *netdev = pci_get_drvdata(pdev);
7181 struct igb_adapter *adapter = netdev_priv(netdev);
7182 struct e1000_hw *hw = &adapter->hw;
7185 pci_set_power_state(pdev, PCI_D0);
7186 pci_restore_state(pdev);
7187 pci_save_state(pdev);
7189 err = pci_enable_device_mem(pdev);
7192 "igb: Cannot enable PCI device from suspend\n");
7195 pci_set_master(pdev);
7197 pci_enable_wake(pdev, PCI_D3hot, 0);
7198 pci_enable_wake(pdev, PCI_D3cold, 0);
7200 if (igb_init_interrupt_scheme(adapter, true)) {
7201 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
7207 /* let the f/w know that the h/w is now under the control of the
7210 igb_get_hw_control(adapter);
7212 wr32(E1000_WUS, ~0);
7214 if (netdev->flags & IFF_UP) {
7216 err = __igb_open(netdev, true);
7222 netif_device_attach(netdev);
7226 #ifdef CONFIG_PM_RUNTIME
7227 static int igb_runtime_idle(struct device *dev)
7229 struct pci_dev *pdev = to_pci_dev(dev);
7230 struct net_device *netdev = pci_get_drvdata(pdev);
7231 struct igb_adapter *adapter = netdev_priv(netdev);
7233 if (!igb_has_link(adapter))
7234 pm_schedule_suspend(dev, MSEC_PER_SEC * 5);
7239 static int igb_runtime_suspend(struct device *dev)
7241 struct pci_dev *pdev = to_pci_dev(dev);
7245 retval = __igb_shutdown(pdev, &wake, 1);
7250 pci_prepare_to_sleep(pdev);
7252 pci_wake_from_d3(pdev, false);
7253 pci_set_power_state(pdev, PCI_D3hot);
7259 static int igb_runtime_resume(struct device *dev)
7261 return igb_resume(dev);
7263 #endif /* CONFIG_PM_RUNTIME */
7266 static void igb_shutdown(struct pci_dev *pdev)
7270 __igb_shutdown(pdev, &wake, 0);
7272 if (system_state == SYSTEM_POWER_OFF) {
7273 pci_wake_from_d3(pdev, wake);
7274 pci_set_power_state(pdev, PCI_D3hot);
7278 #ifdef CONFIG_PCI_IOV
7279 static int igb_sriov_reinit(struct pci_dev *dev)
7281 struct net_device *netdev = pci_get_drvdata(dev);
7282 struct igb_adapter *adapter = netdev_priv(netdev);
7283 struct pci_dev *pdev = adapter->pdev;
7287 if (netif_running(netdev))
7290 igb_clear_interrupt_scheme(adapter);
7292 igb_init_queue_configuration(adapter);
7294 if (igb_init_interrupt_scheme(adapter, true)) {
7295 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
7299 if (netif_running(netdev))
7307 static int igb_pci_disable_sriov(struct pci_dev *dev)
7309 int err = igb_disable_sriov(dev);
7312 err = igb_sriov_reinit(dev);
7317 static int igb_pci_enable_sriov(struct pci_dev *dev, int num_vfs)
7319 int err = igb_enable_sriov(dev, num_vfs);
7324 err = igb_sriov_reinit(dev);
7333 static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs)
7335 #ifdef CONFIG_PCI_IOV
7337 return igb_pci_disable_sriov(dev);
7339 return igb_pci_enable_sriov(dev, num_vfs);
7344 #ifdef CONFIG_NET_POLL_CONTROLLER
7345 /* Polling 'interrupt' - used by things like netconsole to send skbs
7346 * without having to re-enable interrupts. It's not called while
7347 * the interrupt routine is executing.
7349 static void igb_netpoll(struct net_device *netdev)
7351 struct igb_adapter *adapter = netdev_priv(netdev);
7352 struct e1000_hw *hw = &adapter->hw;
7353 struct igb_q_vector *q_vector;
7356 for (i = 0; i < adapter->num_q_vectors; i++) {
7357 q_vector = adapter->q_vector[i];
7358 if (adapter->msix_entries)
7359 wr32(E1000_EIMC, q_vector->eims_value);
7361 igb_irq_disable(adapter);
7362 napi_schedule(&q_vector->napi);
7365 #endif /* CONFIG_NET_POLL_CONTROLLER */
7368 * igb_io_error_detected - called when PCI error is detected
7369 * @pdev: Pointer to PCI device
7370 * @state: The current pci connection state
7372 * This function is called after a PCI bus error affecting
7373 * this device has been detected.
7375 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
7376 pci_channel_state_t state)
7378 struct net_device *netdev = pci_get_drvdata(pdev);
7379 struct igb_adapter *adapter = netdev_priv(netdev);
7381 netif_device_detach(netdev);
7383 if (state == pci_channel_io_perm_failure)
7384 return PCI_ERS_RESULT_DISCONNECT;
7386 if (netif_running(netdev))
7388 pci_disable_device(pdev);
7390 /* Request a slot slot reset. */
7391 return PCI_ERS_RESULT_NEED_RESET;
7395 * igb_io_slot_reset - called after the pci bus has been reset.
7396 * @pdev: Pointer to PCI device
7398 * Restart the card from scratch, as if from a cold-boot. Implementation
7399 * resembles the first-half of the igb_resume routine.
7401 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
7403 struct net_device *netdev = pci_get_drvdata(pdev);
7404 struct igb_adapter *adapter = netdev_priv(netdev);
7405 struct e1000_hw *hw = &adapter->hw;
7406 pci_ers_result_t result;
7409 if (pci_enable_device_mem(pdev)) {
7411 "Cannot re-enable PCI device after reset.\n");
7412 result = PCI_ERS_RESULT_DISCONNECT;
7414 pci_set_master(pdev);
7415 pci_restore_state(pdev);
7416 pci_save_state(pdev);
7418 pci_enable_wake(pdev, PCI_D3hot, 0);
7419 pci_enable_wake(pdev, PCI_D3cold, 0);
7422 wr32(E1000_WUS, ~0);
7423 result = PCI_ERS_RESULT_RECOVERED;
7426 err = pci_cleanup_aer_uncorrect_error_status(pdev);
7429 "pci_cleanup_aer_uncorrect_error_status failed 0x%0x\n",
7431 /* non-fatal, continue */
7438 * igb_io_resume - called when traffic can start flowing again.
7439 * @pdev: Pointer to PCI device
7441 * This callback is called when the error recovery driver tells us that
7442 * its OK to resume normal operation. Implementation resembles the
7443 * second-half of the igb_resume routine.
7445 static void igb_io_resume(struct pci_dev *pdev)
7447 struct net_device *netdev = pci_get_drvdata(pdev);
7448 struct igb_adapter *adapter = netdev_priv(netdev);
7450 if (netif_running(netdev)) {
7451 if (igb_up(adapter)) {
7452 dev_err(&pdev->dev, "igb_up failed after reset\n");
7457 netif_device_attach(netdev);
7459 /* let the f/w know that the h/w is now under the control of the
7462 igb_get_hw_control(adapter);
7465 static void igb_rar_set_qsel(struct igb_adapter *adapter, u8 *addr, u32 index,
7468 u32 rar_low, rar_high;
7469 struct e1000_hw *hw = &adapter->hw;
7471 /* HW expects these in little endian so we reverse the byte order
7472 * from network order (big endian) to little endian
7474 rar_low = ((u32) addr[0] | ((u32) addr[1] << 8) |
7475 ((u32) addr[2] << 16) | ((u32) addr[3] << 24));
7476 rar_high = ((u32) addr[4] | ((u32) addr[5] << 8));
7478 /* Indicate to hardware the Address is Valid. */
7479 rar_high |= E1000_RAH_AV;
7481 if (hw->mac.type == e1000_82575)
7482 rar_high |= E1000_RAH_POOL_1 * qsel;
7484 rar_high |= E1000_RAH_POOL_1 << qsel;
7486 wr32(E1000_RAL(index), rar_low);
7488 wr32(E1000_RAH(index), rar_high);
7492 static int igb_set_vf_mac(struct igb_adapter *adapter,
7493 int vf, unsigned char *mac_addr)
7495 struct e1000_hw *hw = &adapter->hw;
7496 /* VF MAC addresses start at end of receive addresses and moves
7497 * towards the first, as a result a collision should not be possible
7499 int rar_entry = hw->mac.rar_entry_count - (vf + 1);
7501 memcpy(adapter->vf_data[vf].vf_mac_addresses, mac_addr, ETH_ALEN);
7503 igb_rar_set_qsel(adapter, mac_addr, rar_entry, vf);
7508 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
7510 struct igb_adapter *adapter = netdev_priv(netdev);
7511 if (!is_valid_ether_addr(mac) || (vf >= adapter->vfs_allocated_count))
7513 adapter->vf_data[vf].flags |= IGB_VF_FLAG_PF_SET_MAC;
7514 dev_info(&adapter->pdev->dev, "setting MAC %pM on VF %d\n", mac, vf);
7515 dev_info(&adapter->pdev->dev,
7516 "Reload the VF driver to make this change effective.");
7517 if (test_bit(__IGB_DOWN, &adapter->state)) {
7518 dev_warn(&adapter->pdev->dev,
7519 "The VF MAC address has been set, but the PF device is not up.\n");
7520 dev_warn(&adapter->pdev->dev,
7521 "Bring the PF device up before attempting to use the VF device.\n");
7523 return igb_set_vf_mac(adapter, vf, mac);
7526 static int igb_link_mbps(int internal_link_speed)
7528 switch (internal_link_speed) {
7538 static void igb_set_vf_rate_limit(struct e1000_hw *hw, int vf, int tx_rate,
7545 /* Calculate the rate factor values to set */
7546 rf_int = link_speed / tx_rate;
7547 rf_dec = (link_speed - (rf_int * tx_rate));
7548 rf_dec = (rf_dec * (1 << E1000_RTTBCNRC_RF_INT_SHIFT)) /
7551 bcnrc_val = E1000_RTTBCNRC_RS_ENA;
7552 bcnrc_val |= ((rf_int << E1000_RTTBCNRC_RF_INT_SHIFT) &
7553 E1000_RTTBCNRC_RF_INT_MASK);
7554 bcnrc_val |= (rf_dec & E1000_RTTBCNRC_RF_DEC_MASK);
7559 wr32(E1000_RTTDQSEL, vf); /* vf X uses queue X */
7560 /* Set global transmit compensation time to the MMW_SIZE in RTTBCNRM
7561 * register. MMW_SIZE=0x014 if 9728-byte jumbo is supported.
7563 wr32(E1000_RTTBCNRM, 0x14);
7564 wr32(E1000_RTTBCNRC, bcnrc_val);
7567 static void igb_check_vf_rate_limit(struct igb_adapter *adapter)
7569 int actual_link_speed, i;
7570 bool reset_rate = false;
7572 /* VF TX rate limit was not set or not supported */
7573 if ((adapter->vf_rate_link_speed == 0) ||
7574 (adapter->hw.mac.type != e1000_82576))
7577 actual_link_speed = igb_link_mbps(adapter->link_speed);
7578 if (actual_link_speed != adapter->vf_rate_link_speed) {
7580 adapter->vf_rate_link_speed = 0;
7581 dev_info(&adapter->pdev->dev,
7582 "Link speed has been changed. VF Transmit rate is disabled\n");
7585 for (i = 0; i < adapter->vfs_allocated_count; i++) {
7587 adapter->vf_data[i].tx_rate = 0;
7589 igb_set_vf_rate_limit(&adapter->hw, i,
7590 adapter->vf_data[i].tx_rate,
7595 static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf, int tx_rate)
7597 struct igb_adapter *adapter = netdev_priv(netdev);
7598 struct e1000_hw *hw = &adapter->hw;
7599 int actual_link_speed;
7601 if (hw->mac.type != e1000_82576)
7604 actual_link_speed = igb_link_mbps(adapter->link_speed);
7605 if ((vf >= adapter->vfs_allocated_count) ||
7606 (!(rd32(E1000_STATUS) & E1000_STATUS_LU)) ||
7607 (tx_rate < 0) || (tx_rate > actual_link_speed))
7610 adapter->vf_rate_link_speed = actual_link_speed;
7611 adapter->vf_data[vf].tx_rate = (u16)tx_rate;
7612 igb_set_vf_rate_limit(hw, vf, tx_rate, actual_link_speed);
7617 static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf,
7620 struct igb_adapter *adapter = netdev_priv(netdev);
7621 struct e1000_hw *hw = &adapter->hw;
7622 u32 reg_val, reg_offset;
7624 if (!adapter->vfs_allocated_count)
7627 if (vf >= adapter->vfs_allocated_count)
7630 reg_offset = (hw->mac.type == e1000_82576) ? E1000_DTXSWC : E1000_TXSWC;
7631 reg_val = rd32(reg_offset);
7633 reg_val |= ((1 << vf) |
7634 (1 << (vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT)));
7636 reg_val &= ~((1 << vf) |
7637 (1 << (vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT)));
7638 wr32(reg_offset, reg_val);
7640 adapter->vf_data[vf].spoofchk_enabled = setting;
7641 return E1000_SUCCESS;
7644 static int igb_ndo_get_vf_config(struct net_device *netdev,
7645 int vf, struct ifla_vf_info *ivi)
7647 struct igb_adapter *adapter = netdev_priv(netdev);
7648 if (vf >= adapter->vfs_allocated_count)
7651 memcpy(&ivi->mac, adapter->vf_data[vf].vf_mac_addresses, ETH_ALEN);
7652 ivi->tx_rate = adapter->vf_data[vf].tx_rate;
7653 ivi->vlan = adapter->vf_data[vf].pf_vlan;
7654 ivi->qos = adapter->vf_data[vf].pf_qos;
7655 ivi->spoofchk = adapter->vf_data[vf].spoofchk_enabled;
7659 static void igb_vmm_control(struct igb_adapter *adapter)
7661 struct e1000_hw *hw = &adapter->hw;
7664 switch (hw->mac.type) {
7670 /* replication is not supported for 82575 */
7673 /* notify HW that the MAC is adding vlan tags */
7674 reg = rd32(E1000_DTXCTL);
7675 reg |= E1000_DTXCTL_VLAN_ADDED;
7676 wr32(E1000_DTXCTL, reg);
7678 /* enable replication vlan tag stripping */
7679 reg = rd32(E1000_RPLOLR);
7680 reg |= E1000_RPLOLR_STRVLAN;
7681 wr32(E1000_RPLOLR, reg);
7683 /* none of the above registers are supported by i350 */
7687 if (adapter->vfs_allocated_count) {
7688 igb_vmdq_set_loopback_pf(hw, true);
7689 igb_vmdq_set_replication_pf(hw, true);
7690 igb_vmdq_set_anti_spoofing_pf(hw, true,
7691 adapter->vfs_allocated_count);
7693 igb_vmdq_set_loopback_pf(hw, false);
7694 igb_vmdq_set_replication_pf(hw, false);
7698 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba)
7700 struct e1000_hw *hw = &adapter->hw;
7704 if (hw->mac.type > e1000_82580) {
7705 if (adapter->flags & IGB_FLAG_DMAC) {
7708 /* force threshold to 0. */
7709 wr32(E1000_DMCTXTH, 0);
7711 /* DMA Coalescing high water mark needs to be greater
7712 * than the Rx threshold. Set hwm to PBA - max frame
7713 * size in 16B units, capping it at PBA - 6KB.
7715 hwm = 64 * pba - adapter->max_frame_size / 16;
7716 if (hwm < 64 * (pba - 6))
7717 hwm = 64 * (pba - 6);
7718 reg = rd32(E1000_FCRTC);
7719 reg &= ~E1000_FCRTC_RTH_COAL_MASK;
7720 reg |= ((hwm << E1000_FCRTC_RTH_COAL_SHIFT)
7721 & E1000_FCRTC_RTH_COAL_MASK);
7722 wr32(E1000_FCRTC, reg);
7724 /* Set the DMA Coalescing Rx threshold to PBA - 2 * max
7725 * frame size, capping it at PBA - 10KB.
7727 dmac_thr = pba - adapter->max_frame_size / 512;
7728 if (dmac_thr < pba - 10)
7729 dmac_thr = pba - 10;
7730 reg = rd32(E1000_DMACR);
7731 reg &= ~E1000_DMACR_DMACTHR_MASK;
7732 reg |= ((dmac_thr << E1000_DMACR_DMACTHR_SHIFT)
7733 & E1000_DMACR_DMACTHR_MASK);
7735 /* transition to L0x or L1 if available..*/
7736 reg |= (E1000_DMACR_DMAC_EN | E1000_DMACR_DMAC_LX_MASK);
7738 /* watchdog timer= +-1000 usec in 32usec intervals */
7741 /* Disable BMC-to-OS Watchdog Enable */
7742 if (hw->mac.type != e1000_i354)
7743 reg &= ~E1000_DMACR_DC_BMC2OSW_EN;
7745 wr32(E1000_DMACR, reg);
7747 /* no lower threshold to disable
7748 * coalescing(smart fifb)-UTRESH=0
7750 wr32(E1000_DMCRTRH, 0);
7752 reg = (IGB_DMCTLX_DCFLUSH_DIS | 0x4);
7754 wr32(E1000_DMCTLX, reg);
7756 /* free space in tx packet buffer to wake from
7759 wr32(E1000_DMCTXTH, (IGB_MIN_TXPBSIZE -
7760 (IGB_TX_BUF_4096 + adapter->max_frame_size)) >> 6);
7762 /* make low power state decision controlled
7765 reg = rd32(E1000_PCIEMISC);
7766 reg &= ~E1000_PCIEMISC_LX_DECISION;
7767 wr32(E1000_PCIEMISC, reg);
7768 } /* endif adapter->dmac is not disabled */
7769 } else if (hw->mac.type == e1000_82580) {
7770 u32 reg = rd32(E1000_PCIEMISC);
7771 wr32(E1000_PCIEMISC, reg & ~E1000_PCIEMISC_LX_DECISION);
7772 wr32(E1000_DMACR, 0);
7777 * igb_read_i2c_byte - Reads 8 bit word over I2C
7778 * @hw: pointer to hardware structure
7779 * @byte_offset: byte offset to read
7780 * @dev_addr: device address
7783 * Performs byte read operation over I2C interface at
7784 * a specified device address.
7786 s32 igb_read_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
7787 u8 dev_addr, u8 *data)
7789 struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
7790 struct i2c_client *this_client = adapter->i2c_client;
7795 return E1000_ERR_I2C;
7797 swfw_mask = E1000_SWFW_PHY0_SM;
7799 if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask)
7801 return E1000_ERR_SWFW_SYNC;
7803 status = i2c_smbus_read_byte_data(this_client, byte_offset);
7804 hw->mac.ops.release_swfw_sync(hw, swfw_mask);
7807 return E1000_ERR_I2C;
7810 return E1000_SUCCESS;
7815 * igb_write_i2c_byte - Writes 8 bit word over I2C
7816 * @hw: pointer to hardware structure
7817 * @byte_offset: byte offset to write
7818 * @dev_addr: device address
7819 * @data: value to write
7821 * Performs byte write operation over I2C interface at
7822 * a specified device address.
7824 s32 igb_write_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
7825 u8 dev_addr, u8 data)
7827 struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
7828 struct i2c_client *this_client = adapter->i2c_client;
7830 u16 swfw_mask = E1000_SWFW_PHY0_SM;
7833 return E1000_ERR_I2C;
7835 if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask) != E1000_SUCCESS)
7836 return E1000_ERR_SWFW_SYNC;
7837 status = i2c_smbus_write_byte_data(this_client, byte_offset, data);
7838 hw->mac.ops.release_swfw_sync(hw, swfw_mask);
7841 return E1000_ERR_I2C;
7843 return E1000_SUCCESS;