1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright(c) 1999 - 2018 Intel Corporation. */
4 #include <linux/types.h>
5 #include <linux/module.h>
7 #include <linux/netdevice.h>
8 #include <linux/vmalloc.h>
9 #include <linux/string.h>
11 #include <linux/interrupt.h>
13 #include <linux/tcp.h>
14 #include <linux/sctp.h>
15 #include <linux/pkt_sched.h>
16 #include <linux/ipv6.h>
17 #include <linux/slab.h>
18 #include <net/checksum.h>
19 #include <net/ip6_checksum.h>
20 #include <linux/etherdevice.h>
21 #include <linux/ethtool.h>
23 #include <linux/if_vlan.h>
24 #include <linux/if_macvlan.h>
25 #include <linux/if_bridge.h>
26 #include <linux/prefetch.h>
27 #include <linux/bpf.h>
28 #include <linux/bpf_trace.h>
29 #include <linux/atomic.h>
30 #include <linux/numa.h>
31 #include <generated/utsrelease.h>
32 #include <scsi/fc/fc_fcoe.h>
33 #include <net/udp_tunnel.h>
34 #include <net/pkt_cls.h>
35 #include <net/tc_act/tc_gact.h>
36 #include <net/tc_act/tc_mirred.h>
37 #include <net/vxlan.h>
39 #include <net/xdp_sock_drv.h>
43 #include "ixgbe_common.h"
44 #include "ixgbe_dcb_82599.h"
45 #include "ixgbe_phy.h"
46 #include "ixgbe_sriov.h"
47 #include "ixgbe_model.h"
48 #include "ixgbe_txrx_common.h"
50 char ixgbe_driver_name
[] = "ixgbe";
51 static const char ixgbe_driver_string
[] =
52 "Intel(R) 10 Gigabit PCI Express Network Driver";
54 char ixgbe_default_device_descr
[] =
55 "Intel(R) 10 Gigabit Network Connection";
57 static char ixgbe_default_device_descr
[] =
58 "Intel(R) 10 Gigabit Network Connection";
60 static const char ixgbe_copyright
[] =
61 "Copyright (c) 1999-2016 Intel Corporation.";
63 static const char ixgbe_overheat_msg
[] = "Network adapter has been stopped because it has over heated. Restart the computer. If the problem persists, power off the system and replace the adapter";
65 static const struct ixgbe_info
*ixgbe_info_tbl
[] = {
66 [board_82598
] = &ixgbe_82598_info
,
67 [board_82599
] = &ixgbe_82599_info
,
68 [board_X540
] = &ixgbe_X540_info
,
69 [board_X550
] = &ixgbe_X550_info
,
70 [board_X550EM_x
] = &ixgbe_X550EM_x_info
,
71 [board_x550em_x_fw
] = &ixgbe_x550em_x_fw_info
,
72 [board_x550em_a
] = &ixgbe_x550em_a_info
,
73 [board_x550em_a_fw
] = &ixgbe_x550em_a_fw_info
,
76 /* ixgbe_pci_tbl - PCI Device ID Table
78 * Wildcard entries (PCI_ANY_ID) should come last
79 * Last entry must be all 0s
81 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
82 * Class, Class Mask, private data (not used) }
84 static const struct pci_device_id ixgbe_pci_tbl
[] = {
85 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598
), board_82598
},
86 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598AF_DUAL_PORT
), board_82598
},
87 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598AF_SINGLE_PORT
), board_82598
},
88 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598AT
), board_82598
},
89 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598AT2
), board_82598
},
90 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598EB_CX4
), board_82598
},
91 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598_CX4_DUAL_PORT
), board_82598
},
92 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598_DA_DUAL_PORT
), board_82598
},
93 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM
), board_82598
},
94 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598EB_XF_LR
), board_82598
},
95 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598EB_SFP_LOM
), board_82598
},
96 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598_BX
), board_82598
},
97 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_KX4
), board_82599
},
98 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_XAUI_LOM
), board_82599
},
99 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_KR
), board_82599
},
100 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_SFP
), board_82599
},
101 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_SFP_EM
), board_82599
},
102 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_KX4_MEZZ
), board_82599
},
103 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_CX4
), board_82599
},
104 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_BACKPLANE_FCOE
), board_82599
},
105 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_SFP_FCOE
), board_82599
},
106 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_T3_LOM
), board_82599
},
107 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_COMBO_BACKPLANE
), board_82599
},
108 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_X540T
), board_X540
},
109 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_SFP_SF2
), board_82599
},
110 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_LS
), board_82599
},
111 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_QSFP_SF_QP
), board_82599
},
112 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599EN_SFP
), board_82599
},
113 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_SFP_SF_QP
), board_82599
},
114 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_X540T1
), board_X540
},
115 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_X550T
), board_X550
},
116 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_X550T1
), board_X550
},
117 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_X550EM_X_KX4
), board_X550EM_x
},
118 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_X550EM_X_XFI
), board_X550EM_x
},
119 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_X550EM_X_KR
), board_X550EM_x
},
120 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_X550EM_X_10G_T
), board_X550EM_x
},
121 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_X550EM_X_SFP
), board_X550EM_x
},
122 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_X550EM_X_1G_T
), board_x550em_x_fw
},
123 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_X550EM_A_KR
), board_x550em_a
},
124 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_X550EM_A_KR_L
), board_x550em_a
},
125 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_X550EM_A_SFP_N
), board_x550em_a
},
126 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_X550EM_A_SGMII
), board_x550em_a
},
127 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_X550EM_A_SGMII_L
), board_x550em_a
},
128 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_X550EM_A_10G_T
), board_x550em_a
},
129 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_X550EM_A_SFP
), board_x550em_a
},
130 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_X550EM_A_1G_T
), board_x550em_a_fw
},
131 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_X550EM_A_1G_T_L
), board_x550em_a_fw
},
132 /* required last entry */
135 MODULE_DEVICE_TABLE(pci
, ixgbe_pci_tbl
);
137 #ifdef CONFIG_IXGBE_DCA
138 static int ixgbe_notify_dca(struct notifier_block
*, unsigned long event
,
140 static struct notifier_block dca_notifier
= {
141 .notifier_call
= ixgbe_notify_dca
,
147 #ifdef CONFIG_PCI_IOV
148 static unsigned int max_vfs
;
149 module_param(max_vfs
, uint
, 0);
150 MODULE_PARM_DESC(max_vfs
,
151 "Maximum number of virtual functions to allocate per physical function - default is zero and maximum value is 63. (Deprecated)");
152 #endif /* CONFIG_PCI_IOV */
154 static unsigned int allow_unsupported_sfp
;
155 module_param(allow_unsupported_sfp
, uint
, 0);
156 MODULE_PARM_DESC(allow_unsupported_sfp
,
157 "Allow unsupported and untested SFP+ modules on 82599-based adapters");
159 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
160 static int debug
= -1;
161 module_param(debug
, int, 0);
162 MODULE_PARM_DESC(debug
, "Debug level (0=none,...,16=all)");
164 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
165 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
166 MODULE_LICENSE("GPL v2");
168 static struct workqueue_struct
*ixgbe_wq
;
170 static bool ixgbe_check_cfg_remove(struct ixgbe_hw
*hw
, struct pci_dev
*pdev
);
171 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter
*);
173 static const struct net_device_ops ixgbe_netdev_ops
;
175 static bool netif_is_ixgbe(struct net_device
*dev
)
177 return dev
&& (dev
->netdev_ops
== &ixgbe_netdev_ops
);
180 static int ixgbe_read_pci_cfg_word_parent(struct ixgbe_adapter
*adapter
,
183 struct pci_dev
*parent_dev
;
184 struct pci_bus
*parent_bus
;
186 parent_bus
= adapter
->pdev
->bus
->parent
;
190 parent_dev
= parent_bus
->self
;
194 if (!pci_is_pcie(parent_dev
))
197 pcie_capability_read_word(parent_dev
, reg
, value
);
198 if (*value
== IXGBE_FAILED_READ_CFG_WORD
&&
199 ixgbe_check_cfg_remove(&adapter
->hw
, parent_dev
))
204 static s32
ixgbe_get_parent_bus_info(struct ixgbe_adapter
*adapter
)
206 struct ixgbe_hw
*hw
= &adapter
->hw
;
210 hw
->bus
.type
= ixgbe_bus_type_pci_express
;
212 /* Get the negotiated link width and speed from PCI config space of the
213 * parent, as this device is behind a switch
215 err
= ixgbe_read_pci_cfg_word_parent(adapter
, 18, &link_status
);
217 /* assume caller will handle error case */
221 hw
->bus
.width
= ixgbe_convert_bus_width(link_status
);
222 hw
->bus
.speed
= ixgbe_convert_bus_speed(link_status
);
228 * ixgbe_check_from_parent - Determine whether PCIe info should come from parent
229 * @hw: hw specific details
231 * This function is used by probe to determine whether a device's PCI-Express
232 * bandwidth details should be gathered from the parent bus instead of from the
233 * device. Used to ensure that various locations all have the correct device ID
236 static inline bool ixgbe_pcie_from_parent(struct ixgbe_hw
*hw
)
238 switch (hw
->device_id
) {
239 case IXGBE_DEV_ID_82599_SFP_SF_QP
:
240 case IXGBE_DEV_ID_82599_QSFP_SF_QP
:
247 static void ixgbe_check_minimum_link(struct ixgbe_adapter
*adapter
,
250 struct ixgbe_hw
*hw
= &adapter
->hw
;
251 struct pci_dev
*pdev
;
253 /* Some devices are not connected over PCIe and thus do not negotiate
254 * speed. These devices do not have valid bus info, and thus any report
255 * we generate may not be correct.
257 if (hw
->bus
.type
== ixgbe_bus_type_internal
)
260 /* determine whether to use the parent device */
261 if (ixgbe_pcie_from_parent(&adapter
->hw
))
262 pdev
= adapter
->pdev
->bus
->parent
->self
;
264 pdev
= adapter
->pdev
;
266 pcie_print_link_status(pdev
);
269 static void ixgbe_service_event_schedule(struct ixgbe_adapter
*adapter
)
271 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
) &&
272 !test_bit(__IXGBE_REMOVING
, &adapter
->state
) &&
273 !test_and_set_bit(__IXGBE_SERVICE_SCHED
, &adapter
->state
))
274 queue_work(ixgbe_wq
, &adapter
->service_task
);
277 static void ixgbe_remove_adapter(struct ixgbe_hw
*hw
)
279 struct ixgbe_adapter
*adapter
= hw
->back
;
284 e_dev_err("Adapter removed\n");
285 if (test_bit(__IXGBE_SERVICE_INITED
, &adapter
->state
))
286 ixgbe_service_event_schedule(adapter
);
289 static u32
ixgbe_check_remove(struct ixgbe_hw
*hw
, u32 reg
)
291 u8 __iomem
*reg_addr
;
295 reg_addr
= READ_ONCE(hw
->hw_addr
);
296 if (ixgbe_removed(reg_addr
))
297 return IXGBE_FAILED_READ_REG
;
299 /* Register read of 0xFFFFFFF can indicate the adapter has been removed,
300 * so perform several status register reads to determine if the adapter
303 for (i
= 0; i
< IXGBE_FAILED_READ_RETRIES
; i
++) {
304 value
= readl(reg_addr
+ IXGBE_STATUS
);
305 if (value
!= IXGBE_FAILED_READ_REG
)
310 if (value
== IXGBE_FAILED_READ_REG
)
311 ixgbe_remove_adapter(hw
);
313 value
= readl(reg_addr
+ reg
);
318 * ixgbe_read_reg - Read from device register
319 * @hw: hw specific details
320 * @reg: offset of register to read
322 * Returns : value read or IXGBE_FAILED_READ_REG if removed
324 * This function is used to read device registers. It checks for device
325 * removal by confirming any read that returns all ones by checking the
326 * status register value for all ones. This function avoids reading from
327 * the hardware if a removal was previously detected in which case it
328 * returns IXGBE_FAILED_READ_REG (all ones).
330 u32
ixgbe_read_reg(struct ixgbe_hw
*hw
, u32 reg
)
332 u8 __iomem
*reg_addr
= READ_ONCE(hw
->hw_addr
);
335 if (ixgbe_removed(reg_addr
))
336 return IXGBE_FAILED_READ_REG
;
337 if (unlikely(hw
->phy
.nw_mng_if_sel
&
338 IXGBE_NW_MNG_IF_SEL_SGMII_ENABLE
)) {
339 struct ixgbe_adapter
*adapter
;
342 for (i
= 0; i
< 200; ++i
) {
343 value
= readl(reg_addr
+ IXGBE_MAC_SGMII_BUSY
);
345 goto writes_completed
;
346 if (value
== IXGBE_FAILED_READ_REG
) {
347 ixgbe_remove_adapter(hw
);
348 return IXGBE_FAILED_READ_REG
;
354 e_warn(hw
, "register writes incomplete %08x\n", value
);
358 value
= readl(reg_addr
+ reg
);
359 if (unlikely(value
== IXGBE_FAILED_READ_REG
))
360 value
= ixgbe_check_remove(hw
, reg
);
364 static bool ixgbe_check_cfg_remove(struct ixgbe_hw
*hw
, struct pci_dev
*pdev
)
368 pci_read_config_word(pdev
, PCI_VENDOR_ID
, &value
);
369 if (value
== IXGBE_FAILED_READ_CFG_WORD
) {
370 ixgbe_remove_adapter(hw
);
376 u16
ixgbe_read_pci_cfg_word(struct ixgbe_hw
*hw
, u32 reg
)
378 struct ixgbe_adapter
*adapter
= hw
->back
;
381 if (ixgbe_removed(hw
->hw_addr
))
382 return IXGBE_FAILED_READ_CFG_WORD
;
383 pci_read_config_word(adapter
->pdev
, reg
, &value
);
384 if (value
== IXGBE_FAILED_READ_CFG_WORD
&&
385 ixgbe_check_cfg_remove(hw
, adapter
->pdev
))
386 return IXGBE_FAILED_READ_CFG_WORD
;
390 #ifdef CONFIG_PCI_IOV
391 static u32
ixgbe_read_pci_cfg_dword(struct ixgbe_hw
*hw
, u32 reg
)
393 struct ixgbe_adapter
*adapter
= hw
->back
;
396 if (ixgbe_removed(hw
->hw_addr
))
397 return IXGBE_FAILED_READ_CFG_DWORD
;
398 pci_read_config_dword(adapter
->pdev
, reg
, &value
);
399 if (value
== IXGBE_FAILED_READ_CFG_DWORD
&&
400 ixgbe_check_cfg_remove(hw
, adapter
->pdev
))
401 return IXGBE_FAILED_READ_CFG_DWORD
;
404 #endif /* CONFIG_PCI_IOV */
406 void ixgbe_write_pci_cfg_word(struct ixgbe_hw
*hw
, u32 reg
, u16 value
)
408 struct ixgbe_adapter
*adapter
= hw
->back
;
410 if (ixgbe_removed(hw
->hw_addr
))
412 pci_write_config_word(adapter
->pdev
, reg
, value
);
415 static void ixgbe_service_event_complete(struct ixgbe_adapter
*adapter
)
417 BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED
, &adapter
->state
));
419 /* flush memory to make sure state is correct before next watchdog */
420 smp_mb__before_atomic();
421 clear_bit(__IXGBE_SERVICE_SCHED
, &adapter
->state
);
424 struct ixgbe_reg_info
{
429 static const struct ixgbe_reg_info ixgbe_reg_info_tbl
[] = {
431 /* General Registers */
432 {IXGBE_CTRL
, "CTRL"},
433 {IXGBE_STATUS
, "STATUS"},
434 {IXGBE_CTRL_EXT
, "CTRL_EXT"},
436 /* Interrupt Registers */
437 {IXGBE_EICR
, "EICR"},
440 {IXGBE_SRRCTL(0), "SRRCTL"},
441 {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
442 {IXGBE_RDLEN(0), "RDLEN"},
443 {IXGBE_RDH(0), "RDH"},
444 {IXGBE_RDT(0), "RDT"},
445 {IXGBE_RXDCTL(0), "RXDCTL"},
446 {IXGBE_RDBAL(0), "RDBAL"},
447 {IXGBE_RDBAH(0), "RDBAH"},
450 {IXGBE_TDBAL(0), "TDBAL"},
451 {IXGBE_TDBAH(0), "TDBAH"},
452 {IXGBE_TDLEN(0), "TDLEN"},
453 {IXGBE_TDH(0), "TDH"},
454 {IXGBE_TDT(0), "TDT"},
455 {IXGBE_TXDCTL(0), "TXDCTL"},
457 /* List Terminator */
463 * ixgbe_regdump - register printout routine
465 static void ixgbe_regdump(struct ixgbe_hw
*hw
, struct ixgbe_reg_info
*reginfo
)
471 switch (reginfo
->ofs
) {
472 case IXGBE_SRRCTL(0):
473 for (i
= 0; i
< 64; i
++)
474 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_SRRCTL(i
));
476 case IXGBE_DCA_RXCTRL(0):
477 for (i
= 0; i
< 64; i
++)
478 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_DCA_RXCTRL(i
));
481 for (i
= 0; i
< 64; i
++)
482 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_RDLEN(i
));
485 for (i
= 0; i
< 64; i
++)
486 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_RDH(i
));
489 for (i
= 0; i
< 64; i
++)
490 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_RDT(i
));
492 case IXGBE_RXDCTL(0):
493 for (i
= 0; i
< 64; i
++)
494 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_RXDCTL(i
));
497 for (i
= 0; i
< 64; i
++)
498 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_RDBAL(i
));
501 for (i
= 0; i
< 64; i
++)
502 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_RDBAH(i
));
505 for (i
= 0; i
< 64; i
++)
506 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_TDBAL(i
));
509 for (i
= 0; i
< 64; i
++)
510 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_TDBAH(i
));
513 for (i
= 0; i
< 64; i
++)
514 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_TDLEN(i
));
517 for (i
= 0; i
< 64; i
++)
518 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_TDH(i
));
521 for (i
= 0; i
< 64; i
++)
522 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_TDT(i
));
524 case IXGBE_TXDCTL(0):
525 for (i
= 0; i
< 64; i
++)
526 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_TXDCTL(i
));
529 pr_info("%-15s %08x\n",
530 reginfo
->name
, IXGBE_READ_REG(hw
, reginfo
->ofs
));
540 snprintf(rname
, 16, "%s[%d-%d]", reginfo
->name
, i
, i
+ 7);
541 for (j
= 0; j
< 8; j
++)
542 p
+= sprintf(p
, " %08x", regs
[i
++]);
543 pr_err("%-15s%s\n", rname
, buf
);
548 static void ixgbe_print_buffer(struct ixgbe_ring
*ring
, int n
)
550 struct ixgbe_tx_buffer
*tx_buffer
;
552 tx_buffer
= &ring
->tx_buffer_info
[ring
->next_to_clean
];
553 pr_info(" %5d %5X %5X %016llX %08X %p %016llX\n",
554 n
, ring
->next_to_use
, ring
->next_to_clean
,
555 (u64
)dma_unmap_addr(tx_buffer
, dma
),
556 dma_unmap_len(tx_buffer
, len
),
557 tx_buffer
->next_to_watch
,
558 (u64
)tx_buffer
->time_stamp
);
562 * ixgbe_dump - Print registers, tx-rings and rx-rings
564 static void ixgbe_dump(struct ixgbe_adapter
*adapter
)
566 struct net_device
*netdev
= adapter
->netdev
;
567 struct ixgbe_hw
*hw
= &adapter
->hw
;
568 struct ixgbe_reg_info
*reginfo
;
570 struct ixgbe_ring
*ring
;
571 struct ixgbe_tx_buffer
*tx_buffer
;
572 union ixgbe_adv_tx_desc
*tx_desc
;
573 struct my_u0
{ u64 a
; u64 b
; } *u0
;
574 struct ixgbe_ring
*rx_ring
;
575 union ixgbe_adv_rx_desc
*rx_desc
;
576 struct ixgbe_rx_buffer
*rx_buffer_info
;
579 if (!netif_msg_hw(adapter
))
582 /* Print netdevice Info */
584 dev_info(&adapter
->pdev
->dev
, "Net device Info\n");
585 pr_info("Device Name state "
587 pr_info("%-15s %016lX %016lX\n",
590 dev_trans_start(netdev
));
593 /* Print Registers */
594 dev_info(&adapter
->pdev
->dev
, "Register Dump\n");
595 pr_info(" Register Name Value\n");
596 for (reginfo
= (struct ixgbe_reg_info
*)ixgbe_reg_info_tbl
;
597 reginfo
->name
; reginfo
++) {
598 ixgbe_regdump(hw
, reginfo
);
601 /* Print TX Ring Summary */
602 if (!netdev
|| !netif_running(netdev
))
605 dev_info(&adapter
->pdev
->dev
, "TX Rings Summary\n");
606 pr_info(" %s %s %s %s\n",
607 "Queue [NTU] [NTC] [bi(ntc)->dma ]",
608 "leng", "ntw", "timestamp");
609 for (n
= 0; n
< adapter
->num_tx_queues
; n
++) {
610 ring
= adapter
->tx_ring
[n
];
611 ixgbe_print_buffer(ring
, n
);
614 for (n
= 0; n
< adapter
->num_xdp_queues
; n
++) {
615 ring
= adapter
->xdp_ring
[n
];
616 ixgbe_print_buffer(ring
, n
);
620 if (!netif_msg_tx_done(adapter
))
621 goto rx_ring_summary
;
623 dev_info(&adapter
->pdev
->dev
, "TX Rings Dump\n");
625 /* Transmit Descriptor Formats
627 * 82598 Advanced Transmit Descriptor
628 * +--------------------------------------------------------------+
629 * 0 | Buffer Address [63:0] |
630 * +--------------------------------------------------------------+
631 * 8 | PAYLEN | POPTS | IDX | STA | DCMD |DTYP | RSV | DTALEN |
632 * +--------------------------------------------------------------+
633 * 63 46 45 40 39 36 35 32 31 24 23 20 19 0
635 * 82598 Advanced Transmit Descriptor (Write-Back Format)
636 * +--------------------------------------------------------------+
638 * +--------------------------------------------------------------+
639 * 8 | RSV | STA | NXTSEQ |
640 * +--------------------------------------------------------------+
643 * 82599+ Advanced Transmit Descriptor
644 * +--------------------------------------------------------------+
645 * 0 | Buffer Address [63:0] |
646 * +--------------------------------------------------------------+
647 * 8 |PAYLEN |POPTS|CC|IDX |STA |DCMD |DTYP |MAC |RSV |DTALEN |
648 * +--------------------------------------------------------------+
649 * 63 46 45 40 39 38 36 35 32 31 24 23 20 19 18 17 16 15 0
651 * 82599+ Advanced Transmit Descriptor (Write-Back Format)
652 * +--------------------------------------------------------------+
654 * +--------------------------------------------------------------+
655 * 8 | RSV | STA | RSV |
656 * +--------------------------------------------------------------+
660 for (n
= 0; n
< adapter
->num_tx_queues
; n
++) {
661 ring
= adapter
->tx_ring
[n
];
662 pr_info("------------------------------------\n");
663 pr_info("TX QUEUE INDEX = %d\n", ring
->queue_index
);
664 pr_info("------------------------------------\n");
665 pr_info("%s%s %s %s %s %s\n",
666 "T [desc] [address 63:0 ] ",
667 "[PlPOIdStDDt Ln] [bi->dma ] ",
668 "leng", "ntw", "timestamp", "bi->skb");
670 for (i
= 0; ring
->desc
&& (i
< ring
->count
); i
++) {
671 tx_desc
= IXGBE_TX_DESC(ring
, i
);
672 tx_buffer
= &ring
->tx_buffer_info
[i
];
673 u0
= (struct my_u0
*)tx_desc
;
674 if (dma_unmap_len(tx_buffer
, len
) > 0) {
675 const char *ring_desc
;
677 if (i
== ring
->next_to_use
&&
678 i
== ring
->next_to_clean
)
679 ring_desc
= " NTC/U";
680 else if (i
== ring
->next_to_use
)
682 else if (i
== ring
->next_to_clean
)
686 pr_info("T [0x%03X] %016llX %016llX %016llX %08X %p %016llX %p%s",
688 le64_to_cpu((__force __le64
)u0
->a
),
689 le64_to_cpu((__force __le64
)u0
->b
),
690 (u64
)dma_unmap_addr(tx_buffer
, dma
),
691 dma_unmap_len(tx_buffer
, len
),
692 tx_buffer
->next_to_watch
,
693 (u64
)tx_buffer
->time_stamp
,
697 if (netif_msg_pktdata(adapter
) &&
699 print_hex_dump(KERN_INFO
, "",
700 DUMP_PREFIX_ADDRESS
, 16, 1,
701 tx_buffer
->skb
->data
,
702 dma_unmap_len(tx_buffer
, len
),
708 /* Print RX Rings Summary */
710 dev_info(&adapter
->pdev
->dev
, "RX Rings Summary\n");
711 pr_info("Queue [NTU] [NTC]\n");
712 for (n
= 0; n
< adapter
->num_rx_queues
; n
++) {
713 rx_ring
= adapter
->rx_ring
[n
];
714 pr_info("%5d %5X %5X\n",
715 n
, rx_ring
->next_to_use
, rx_ring
->next_to_clean
);
719 if (!netif_msg_rx_status(adapter
))
722 dev_info(&adapter
->pdev
->dev
, "RX Rings Dump\n");
724 /* Receive Descriptor Formats
726 * 82598 Advanced Receive Descriptor (Read) Format
728 * +-----------------------------------------------------+
729 * 0 | Packet Buffer Address [63:1] |A0/NSE|
730 * +----------------------------------------------+------+
731 * 8 | Header Buffer Address [63:1] | DD |
732 * +-----------------------------------------------------+
735 * 82598 Advanced Receive Descriptor (Write-Back) Format
737 * 63 48 47 32 31 30 21 20 16 15 4 3 0
738 * +------------------------------------------------------+
739 * 0 | RSS Hash / |SPH| HDR_LEN | RSV |Packet| RSS |
740 * | Packet | IP | | | | Type | Type |
741 * | Checksum | Ident | | | | | |
742 * +------------------------------------------------------+
743 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
744 * +------------------------------------------------------+
745 * 63 48 47 32 31 20 19 0
747 * 82599+ Advanced Receive Descriptor (Read) Format
749 * +-----------------------------------------------------+
750 * 0 | Packet Buffer Address [63:1] |A0/NSE|
751 * +----------------------------------------------+------+
752 * 8 | Header Buffer Address [63:1] | DD |
753 * +-----------------------------------------------------+
756 * 82599+ Advanced Receive Descriptor (Write-Back) Format
758 * 63 48 47 32 31 30 21 20 17 16 4 3 0
759 * +------------------------------------------------------+
760 * 0 |RSS / Frag Checksum|SPH| HDR_LEN |RSC- |Packet| RSS |
761 * |/ RTT / PCoE_PARAM | | | CNT | Type | Type |
762 * |/ Flow Dir Flt ID | | | | | |
763 * +------------------------------------------------------+
764 * 8 | VLAN Tag | Length |Extended Error| Xtnd Status/NEXTP |
765 * +------------------------------------------------------+
766 * 63 48 47 32 31 20 19 0
769 for (n
= 0; n
< adapter
->num_rx_queues
; n
++) {
770 rx_ring
= adapter
->rx_ring
[n
];
771 pr_info("------------------------------------\n");
772 pr_info("RX QUEUE INDEX = %d\n", rx_ring
->queue_index
);
773 pr_info("------------------------------------\n");
775 "R [desc] [ PktBuf A0] ",
776 "[ HeadBuf DD] [bi->dma ] [bi->skb ] ",
777 "<-- Adv Rx Read format");
779 "RWB[desc] [PcsmIpSHl PtRs] ",
780 "[vl er S cks ln] ---------------- [bi->skb ] ",
781 "<-- Adv Rx Write-Back format");
783 for (i
= 0; i
< rx_ring
->count
; i
++) {
784 const char *ring_desc
;
786 if (i
== rx_ring
->next_to_use
)
788 else if (i
== rx_ring
->next_to_clean
)
793 rx_buffer_info
= &rx_ring
->rx_buffer_info
[i
];
794 rx_desc
= IXGBE_RX_DESC(rx_ring
, i
);
795 u0
= (struct my_u0
*)rx_desc
;
796 if (rx_desc
->wb
.upper
.length
) {
797 /* Descriptor Done */
798 pr_info("RWB[0x%03X] %016llX %016llX ---------------- %p%s\n",
800 le64_to_cpu((__force __le64
)u0
->a
),
801 le64_to_cpu((__force __le64
)u0
->b
),
805 pr_info("R [0x%03X] %016llX %016llX %016llX %p%s\n",
807 le64_to_cpu((__force __le64
)u0
->a
),
808 le64_to_cpu((__force __le64
)u0
->b
),
809 (u64
)rx_buffer_info
->dma
,
813 if (netif_msg_pktdata(adapter
) &&
814 rx_buffer_info
->dma
) {
815 print_hex_dump(KERN_INFO
, "",
816 DUMP_PREFIX_ADDRESS
, 16, 1,
817 page_address(rx_buffer_info
->page
) +
818 rx_buffer_info
->page_offset
,
819 ixgbe_rx_bufsz(rx_ring
), true);
826 static void ixgbe_release_hw_control(struct ixgbe_adapter
*adapter
)
830 /* Let firmware take over control of h/w */
831 ctrl_ext
= IXGBE_READ_REG(&adapter
->hw
, IXGBE_CTRL_EXT
);
832 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_CTRL_EXT
,
833 ctrl_ext
& ~IXGBE_CTRL_EXT_DRV_LOAD
);
836 static void ixgbe_get_hw_control(struct ixgbe_adapter
*adapter
)
840 /* Let firmware know the driver has taken over */
841 ctrl_ext
= IXGBE_READ_REG(&adapter
->hw
, IXGBE_CTRL_EXT
);
842 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_CTRL_EXT
,
843 ctrl_ext
| IXGBE_CTRL_EXT_DRV_LOAD
);
847 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
848 * @adapter: pointer to adapter struct
849 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
850 * @queue: queue to map the corresponding interrupt to
851 * @msix_vector: the vector to map to the corresponding queue
854 static void ixgbe_set_ivar(struct ixgbe_adapter
*adapter
, s8 direction
,
855 u8 queue
, u8 msix_vector
)
858 struct ixgbe_hw
*hw
= &adapter
->hw
;
859 switch (hw
->mac
.type
) {
860 case ixgbe_mac_82598EB
:
861 msix_vector
|= IXGBE_IVAR_ALLOC_VAL
;
864 index
= (((direction
* 64) + queue
) >> 2) & 0x1F;
865 ivar
= IXGBE_READ_REG(hw
, IXGBE_IVAR(index
));
866 ivar
&= ~(0xFF << (8 * (queue
& 0x3)));
867 ivar
|= (msix_vector
<< (8 * (queue
& 0x3)));
868 IXGBE_WRITE_REG(hw
, IXGBE_IVAR(index
), ivar
);
870 case ixgbe_mac_82599EB
:
873 case ixgbe_mac_X550EM_x
:
874 case ixgbe_mac_x550em_a
:
875 if (direction
== -1) {
877 msix_vector
|= IXGBE_IVAR_ALLOC_VAL
;
878 index
= ((queue
& 1) * 8);
879 ivar
= IXGBE_READ_REG(&adapter
->hw
, IXGBE_IVAR_MISC
);
880 ivar
&= ~(0xFF << index
);
881 ivar
|= (msix_vector
<< index
);
882 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_IVAR_MISC
, ivar
);
885 /* tx or rx causes */
886 msix_vector
|= IXGBE_IVAR_ALLOC_VAL
;
887 index
= ((16 * (queue
& 1)) + (8 * direction
));
888 ivar
= IXGBE_READ_REG(hw
, IXGBE_IVAR(queue
>> 1));
889 ivar
&= ~(0xFF << index
);
890 ivar
|= (msix_vector
<< index
);
891 IXGBE_WRITE_REG(hw
, IXGBE_IVAR(queue
>> 1), ivar
);
899 void ixgbe_irq_rearm_queues(struct ixgbe_adapter
*adapter
,
904 switch (adapter
->hw
.mac
.type
) {
905 case ixgbe_mac_82598EB
:
906 mask
= (IXGBE_EIMS_RTX_QUEUE
& qmask
);
907 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EICS
, mask
);
909 case ixgbe_mac_82599EB
:
912 case ixgbe_mac_X550EM_x
:
913 case ixgbe_mac_x550em_a
:
914 mask
= (qmask
& 0xFFFFFFFF);
915 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EICS_EX(0), mask
);
916 mask
= (qmask
>> 32);
917 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EICS_EX(1), mask
);
924 static void ixgbe_update_xoff_rx_lfc(struct ixgbe_adapter
*adapter
)
926 struct ixgbe_hw
*hw
= &adapter
->hw
;
927 struct ixgbe_hw_stats
*hwstats
= &adapter
->stats
;
931 if ((hw
->fc
.current_mode
!= ixgbe_fc_full
) &&
932 (hw
->fc
.current_mode
!= ixgbe_fc_rx_pause
))
935 switch (hw
->mac
.type
) {
936 case ixgbe_mac_82598EB
:
937 data
= IXGBE_READ_REG(hw
, IXGBE_LXOFFRXC
);
940 data
= IXGBE_READ_REG(hw
, IXGBE_LXOFFRXCNT
);
942 hwstats
->lxoffrxc
+= data
;
944 /* refill credits (no tx hang) if we received xoff */
948 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
949 clear_bit(__IXGBE_HANG_CHECK_ARMED
,
950 &adapter
->tx_ring
[i
]->state
);
952 for (i
= 0; i
< adapter
->num_xdp_queues
; i
++)
953 clear_bit(__IXGBE_HANG_CHECK_ARMED
,
954 &adapter
->xdp_ring
[i
]->state
);
957 static void ixgbe_update_xoff_received(struct ixgbe_adapter
*adapter
)
959 struct ixgbe_hw
*hw
= &adapter
->hw
;
960 struct ixgbe_hw_stats
*hwstats
= &adapter
->stats
;
964 bool pfc_en
= adapter
->dcb_cfg
.pfc_mode_enable
;
966 if (adapter
->ixgbe_ieee_pfc
)
967 pfc_en
|= !!(adapter
->ixgbe_ieee_pfc
->pfc_en
);
969 if (!(adapter
->flags
& IXGBE_FLAG_DCB_ENABLED
) || !pfc_en
) {
970 ixgbe_update_xoff_rx_lfc(adapter
);
974 /* update stats for each tc, only valid with PFC enabled */
975 for (i
= 0; i
< MAX_TX_PACKET_BUFFERS
; i
++) {
978 switch (hw
->mac
.type
) {
979 case ixgbe_mac_82598EB
:
980 pxoffrxc
= IXGBE_READ_REG(hw
, IXGBE_PXOFFRXC(i
));
983 pxoffrxc
= IXGBE_READ_REG(hw
, IXGBE_PXOFFRXCNT(i
));
985 hwstats
->pxoffrxc
[i
] += pxoffrxc
;
986 /* Get the TC for given UP */
987 tc
= netdev_get_prio_tc_map(adapter
->netdev
, i
);
988 xoff
[tc
] += pxoffrxc
;
991 /* disarm tx queues that have received xoff frames */
992 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
993 struct ixgbe_ring
*tx_ring
= adapter
->tx_ring
[i
];
995 tc
= tx_ring
->dcb_tc
;
997 clear_bit(__IXGBE_HANG_CHECK_ARMED
, &tx_ring
->state
);
1000 for (i
= 0; i
< adapter
->num_xdp_queues
; i
++) {
1001 struct ixgbe_ring
*xdp_ring
= adapter
->xdp_ring
[i
];
1003 tc
= xdp_ring
->dcb_tc
;
1005 clear_bit(__IXGBE_HANG_CHECK_ARMED
, &xdp_ring
->state
);
1009 static u64
ixgbe_get_tx_completed(struct ixgbe_ring
*ring
)
1011 return ring
->stats
.packets
;
1014 static u64
ixgbe_get_tx_pending(struct ixgbe_ring
*ring
)
1016 unsigned int head
, tail
;
1018 head
= ring
->next_to_clean
;
1019 tail
= ring
->next_to_use
;
1021 return ((head
<= tail
) ? tail
: tail
+ ring
->count
) - head
;
1024 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring
*tx_ring
)
1026 u32 tx_done
= ixgbe_get_tx_completed(tx_ring
);
1027 u32 tx_done_old
= tx_ring
->tx_stats
.tx_done_old
;
1028 u32 tx_pending
= ixgbe_get_tx_pending(tx_ring
);
1030 clear_check_for_tx_hang(tx_ring
);
1033 * Check for a hung queue, but be thorough. This verifies
1034 * that a transmit has been completed since the previous
1035 * check AND there is at least one packet pending. The
1036 * ARMED bit is set to indicate a potential hang. The
1037 * bit is cleared if a pause frame is received to remove
1038 * false hang detection due to PFC or 802.3x frames. By
1039 * requiring this to fail twice we avoid races with
1040 * pfc clearing the ARMED bit and conditions where we
1041 * run the check_tx_hang logic with a transmit completion
1042 * pending but without time to complete it yet.
1044 if (tx_done_old
== tx_done
&& tx_pending
)
1045 /* make sure it is true for two checks in a row */
1046 return test_and_set_bit(__IXGBE_HANG_CHECK_ARMED
,
1048 /* update completed stats and continue */
1049 tx_ring
->tx_stats
.tx_done_old
= tx_done
;
1050 /* reset the countdown */
1051 clear_bit(__IXGBE_HANG_CHECK_ARMED
, &tx_ring
->state
);
1057 * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
1058 * @adapter: driver private struct
1060 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter
*adapter
)
1063 /* Do the reset outside of interrupt context */
1064 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
)) {
1065 set_bit(__IXGBE_RESET_REQUESTED
, &adapter
->state
);
1066 e_warn(drv
, "initiating reset due to tx timeout\n");
1067 ixgbe_service_event_schedule(adapter
);
1072 * ixgbe_tx_maxrate - callback to set the maximum per-queue bitrate
1073 * @netdev: network interface device structure
1074 * @queue_index: Tx queue to set
1075 * @maxrate: desired maximum transmit bitrate
1077 static int ixgbe_tx_maxrate(struct net_device
*netdev
,
1078 int queue_index
, u32 maxrate
)
1080 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
1081 struct ixgbe_hw
*hw
= &adapter
->hw
;
1082 u32 bcnrc_val
= ixgbe_link_mbps(adapter
);
1087 /* Calculate the rate factor values to set */
1088 bcnrc_val
<<= IXGBE_RTTBCNRC_RF_INT_SHIFT
;
1089 bcnrc_val
/= maxrate
;
1091 /* clear everything but the rate factor */
1092 bcnrc_val
&= IXGBE_RTTBCNRC_RF_INT_MASK
|
1093 IXGBE_RTTBCNRC_RF_DEC_MASK
;
1095 /* enable the rate scheduler */
1096 bcnrc_val
|= IXGBE_RTTBCNRC_RS_ENA
;
1098 IXGBE_WRITE_REG(hw
, IXGBE_RTTDQSEL
, queue_index
);
1099 IXGBE_WRITE_REG(hw
, IXGBE_RTTBCNRC
, bcnrc_val
);
1105 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
1106 * @q_vector: structure containing interrupt and ring information
1107 * @tx_ring: tx ring to clean
1108 * @napi_budget: Used to determine if we are in netpoll
1110 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector
*q_vector
,
1111 struct ixgbe_ring
*tx_ring
, int napi_budget
)
1113 struct ixgbe_adapter
*adapter
= q_vector
->adapter
;
1114 struct ixgbe_tx_buffer
*tx_buffer
;
1115 union ixgbe_adv_tx_desc
*tx_desc
;
1116 unsigned int total_bytes
= 0, total_packets
= 0, total_ipsec
= 0;
1117 unsigned int budget
= q_vector
->tx
.work_limit
;
1118 unsigned int i
= tx_ring
->next_to_clean
;
1120 if (test_bit(__IXGBE_DOWN
, &adapter
->state
))
1123 tx_buffer
= &tx_ring
->tx_buffer_info
[i
];
1124 tx_desc
= IXGBE_TX_DESC(tx_ring
, i
);
1125 i
-= tx_ring
->count
;
1128 union ixgbe_adv_tx_desc
*eop_desc
= tx_buffer
->next_to_watch
;
1130 /* if next_to_watch is not set then there is no work pending */
1134 /* prevent any other reads prior to eop_desc */
1137 /* if DD is not set pending work has not been completed */
1138 if (!(eop_desc
->wb
.status
& cpu_to_le32(IXGBE_TXD_STAT_DD
)))
1141 /* clear next_to_watch to prevent false hangs */
1142 tx_buffer
->next_to_watch
= NULL
;
1144 /* update the statistics for this packet */
1145 total_bytes
+= tx_buffer
->bytecount
;
1146 total_packets
+= tx_buffer
->gso_segs
;
1147 if (tx_buffer
->tx_flags
& IXGBE_TX_FLAGS_IPSEC
)
1151 if (ring_is_xdp(tx_ring
))
1152 xdp_return_frame(tx_buffer
->xdpf
);
1154 napi_consume_skb(tx_buffer
->skb
, napi_budget
);
1156 /* unmap skb header data */
1157 dma_unmap_single(tx_ring
->dev
,
1158 dma_unmap_addr(tx_buffer
, dma
),
1159 dma_unmap_len(tx_buffer
, len
),
1162 /* clear tx_buffer data */
1163 dma_unmap_len_set(tx_buffer
, len
, 0);
1165 /* unmap remaining buffers */
1166 while (tx_desc
!= eop_desc
) {
1171 i
-= tx_ring
->count
;
1172 tx_buffer
= tx_ring
->tx_buffer_info
;
1173 tx_desc
= IXGBE_TX_DESC(tx_ring
, 0);
1176 /* unmap any remaining paged data */
1177 if (dma_unmap_len(tx_buffer
, len
)) {
1178 dma_unmap_page(tx_ring
->dev
,
1179 dma_unmap_addr(tx_buffer
, dma
),
1180 dma_unmap_len(tx_buffer
, len
),
1182 dma_unmap_len_set(tx_buffer
, len
, 0);
1186 /* move us one more past the eop_desc for start of next pkt */
1191 i
-= tx_ring
->count
;
1192 tx_buffer
= tx_ring
->tx_buffer_info
;
1193 tx_desc
= IXGBE_TX_DESC(tx_ring
, 0);
1196 /* issue prefetch for next Tx descriptor */
1199 /* update budget accounting */
1201 } while (likely(budget
));
1203 i
+= tx_ring
->count
;
1204 tx_ring
->next_to_clean
= i
;
1205 u64_stats_update_begin(&tx_ring
->syncp
);
1206 tx_ring
->stats
.bytes
+= total_bytes
;
1207 tx_ring
->stats
.packets
+= total_packets
;
1208 u64_stats_update_end(&tx_ring
->syncp
);
1209 q_vector
->tx
.total_bytes
+= total_bytes
;
1210 q_vector
->tx
.total_packets
+= total_packets
;
1211 adapter
->tx_ipsec
+= total_ipsec
;
1213 if (check_for_tx_hang(tx_ring
) && ixgbe_check_tx_hang(tx_ring
)) {
1214 /* schedule immediate reset if we believe we hung */
1215 struct ixgbe_hw
*hw
= &adapter
->hw
;
1216 e_err(drv
, "Detected Tx Unit Hang %s\n"
1218 " TDH, TDT <%x>, <%x>\n"
1219 " next_to_use <%x>\n"
1220 " next_to_clean <%x>\n"
1221 "tx_buffer_info[next_to_clean]\n"
1222 " time_stamp <%lx>\n"
1224 ring_is_xdp(tx_ring
) ? "(XDP)" : "",
1225 tx_ring
->queue_index
,
1226 IXGBE_READ_REG(hw
, IXGBE_TDH(tx_ring
->reg_idx
)),
1227 IXGBE_READ_REG(hw
, IXGBE_TDT(tx_ring
->reg_idx
)),
1228 tx_ring
->next_to_use
, i
,
1229 tx_ring
->tx_buffer_info
[i
].time_stamp
, jiffies
);
1231 if (!ring_is_xdp(tx_ring
))
1232 netif_stop_subqueue(tx_ring
->netdev
,
1233 tx_ring
->queue_index
);
1236 "tx hang %d detected on queue %d, resetting adapter\n",
1237 adapter
->tx_timeout_count
+ 1, tx_ring
->queue_index
);
1239 /* schedule immediate reset if we believe we hung */
1240 ixgbe_tx_timeout_reset(adapter
);
1242 /* the adapter is about to reset, no point in enabling stuff */
1246 if (ring_is_xdp(tx_ring
))
1249 netdev_tx_completed_queue(txring_txq(tx_ring
),
1250 total_packets
, total_bytes
);
1252 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
1253 if (unlikely(total_packets
&& netif_carrier_ok(tx_ring
->netdev
) &&
1254 (ixgbe_desc_unused(tx_ring
) >= TX_WAKE_THRESHOLD
))) {
1255 /* Make sure that anybody stopping the queue after this
1256 * sees the new next_to_clean.
1259 if (__netif_subqueue_stopped(tx_ring
->netdev
,
1260 tx_ring
->queue_index
)
1261 && !test_bit(__IXGBE_DOWN
, &adapter
->state
)) {
1262 netif_wake_subqueue(tx_ring
->netdev
,
1263 tx_ring
->queue_index
);
1264 ++tx_ring
->tx_stats
.restart_queue
;
1271 #ifdef CONFIG_IXGBE_DCA
1272 static void ixgbe_update_tx_dca(struct ixgbe_adapter
*adapter
,
1273 struct ixgbe_ring
*tx_ring
,
1276 struct ixgbe_hw
*hw
= &adapter
->hw
;
1280 if (adapter
->flags
& IXGBE_FLAG_DCA_ENABLED
)
1281 txctrl
= dca3_get_tag(tx_ring
->dev
, cpu
);
1283 switch (hw
->mac
.type
) {
1284 case ixgbe_mac_82598EB
:
1285 reg_offset
= IXGBE_DCA_TXCTRL(tx_ring
->reg_idx
);
1287 case ixgbe_mac_82599EB
:
1288 case ixgbe_mac_X540
:
1289 reg_offset
= IXGBE_DCA_TXCTRL_82599(tx_ring
->reg_idx
);
1290 txctrl
<<= IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599
;
1293 /* for unknown hardware do not write register */
1298 * We can enable relaxed ordering for reads, but not writes when
1299 * DCA is enabled. This is due to a known issue in some chipsets
1300 * which will cause the DCA tag to be cleared.
1302 txctrl
|= IXGBE_DCA_TXCTRL_DESC_RRO_EN
|
1303 IXGBE_DCA_TXCTRL_DATA_RRO_EN
|
1304 IXGBE_DCA_TXCTRL_DESC_DCA_EN
;
1306 IXGBE_WRITE_REG(hw
, reg_offset
, txctrl
);
1309 static void ixgbe_update_rx_dca(struct ixgbe_adapter
*adapter
,
1310 struct ixgbe_ring
*rx_ring
,
1313 struct ixgbe_hw
*hw
= &adapter
->hw
;
1315 u8 reg_idx
= rx_ring
->reg_idx
;
1317 if (adapter
->flags
& IXGBE_FLAG_DCA_ENABLED
)
1318 rxctrl
= dca3_get_tag(rx_ring
->dev
, cpu
);
1320 switch (hw
->mac
.type
) {
1321 case ixgbe_mac_82599EB
:
1322 case ixgbe_mac_X540
:
1323 rxctrl
<<= IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599
;
1330 * We can enable relaxed ordering for reads, but not writes when
1331 * DCA is enabled. This is due to a known issue in some chipsets
1332 * which will cause the DCA tag to be cleared.
1334 rxctrl
|= IXGBE_DCA_RXCTRL_DESC_RRO_EN
|
1335 IXGBE_DCA_RXCTRL_DATA_DCA_EN
|
1336 IXGBE_DCA_RXCTRL_DESC_DCA_EN
;
1338 IXGBE_WRITE_REG(hw
, IXGBE_DCA_RXCTRL(reg_idx
), rxctrl
);
1341 static void ixgbe_update_dca(struct ixgbe_q_vector
*q_vector
)
1343 struct ixgbe_adapter
*adapter
= q_vector
->adapter
;
1344 struct ixgbe_ring
*ring
;
1345 int cpu
= get_cpu();
1347 if (q_vector
->cpu
== cpu
)
1350 ixgbe_for_each_ring(ring
, q_vector
->tx
)
1351 ixgbe_update_tx_dca(adapter
, ring
, cpu
);
1353 ixgbe_for_each_ring(ring
, q_vector
->rx
)
1354 ixgbe_update_rx_dca(adapter
, ring
, cpu
);
1356 q_vector
->cpu
= cpu
;
1361 static void ixgbe_setup_dca(struct ixgbe_adapter
*adapter
)
1365 /* always use CB2 mode, difference is masked in the CB driver */
1366 if (adapter
->flags
& IXGBE_FLAG_DCA_ENABLED
)
1367 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_DCA_CTRL
,
1368 IXGBE_DCA_CTRL_DCA_MODE_CB2
);
1370 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_DCA_CTRL
,
1371 IXGBE_DCA_CTRL_DCA_DISABLE
);
1373 for (i
= 0; i
< adapter
->num_q_vectors
; i
++) {
1374 adapter
->q_vector
[i
]->cpu
= -1;
1375 ixgbe_update_dca(adapter
->q_vector
[i
]);
1379 static int __ixgbe_notify_dca(struct device
*dev
, void *data
)
1381 struct ixgbe_adapter
*adapter
= dev_get_drvdata(dev
);
1382 unsigned long event
= *(unsigned long *)data
;
1384 if (!(adapter
->flags
& IXGBE_FLAG_DCA_CAPABLE
))
1388 case DCA_PROVIDER_ADD
:
1389 /* if we're already enabled, don't do it again */
1390 if (adapter
->flags
& IXGBE_FLAG_DCA_ENABLED
)
1392 if (dca_add_requester(dev
) == 0) {
1393 adapter
->flags
|= IXGBE_FLAG_DCA_ENABLED
;
1394 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_DCA_CTRL
,
1395 IXGBE_DCA_CTRL_DCA_MODE_CB2
);
1398 fallthrough
; /* DCA is disabled. */
1399 case DCA_PROVIDER_REMOVE
:
1400 if (adapter
->flags
& IXGBE_FLAG_DCA_ENABLED
) {
1401 dca_remove_requester(dev
);
1402 adapter
->flags
&= ~IXGBE_FLAG_DCA_ENABLED
;
1403 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_DCA_CTRL
,
1404 IXGBE_DCA_CTRL_DCA_DISABLE
);
1412 #endif /* CONFIG_IXGBE_DCA */
1414 #define IXGBE_RSS_L4_TYPES_MASK \
1415 ((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
1416 (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
1417 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
1418 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
1420 static inline void ixgbe_rx_hash(struct ixgbe_ring
*ring
,
1421 union ixgbe_adv_rx_desc
*rx_desc
,
1422 struct sk_buff
*skb
)
1426 if (!(ring
->netdev
->features
& NETIF_F_RXHASH
))
1429 rss_type
= le16_to_cpu(rx_desc
->wb
.lower
.lo_dword
.hs_rss
.pkt_info
) &
1430 IXGBE_RXDADV_RSSTYPE_MASK
;
1435 skb_set_hash(skb
, le32_to_cpu(rx_desc
->wb
.lower
.hi_dword
.rss
),
1436 (IXGBE_RSS_L4_TYPES_MASK
& (1ul << rss_type
)) ?
1437 PKT_HASH_TYPE_L4
: PKT_HASH_TYPE_L3
);
1442 * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1443 * @ring: structure containing ring specific data
1444 * @rx_desc: advanced rx descriptor
1446 * Returns : true if it is FCoE pkt
1448 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_ring
*ring
,
1449 union ixgbe_adv_rx_desc
*rx_desc
)
1451 __le16 pkt_info
= rx_desc
->wb
.lower
.lo_dword
.hs_rss
.pkt_info
;
1453 return test_bit(__IXGBE_RX_FCOE
, &ring
->state
) &&
1454 ((pkt_info
& cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK
)) ==
1455 (cpu_to_le16(IXGBE_ETQF_FILTER_FCOE
<<
1456 IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT
)));
1459 #endif /* IXGBE_FCOE */
1461 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1462 * @ring: structure containing ring specific data
1463 * @rx_desc: current Rx descriptor being processed
1464 * @skb: skb currently being received and modified
1466 static inline void ixgbe_rx_checksum(struct ixgbe_ring
*ring
,
1467 union ixgbe_adv_rx_desc
*rx_desc
,
1468 struct sk_buff
*skb
)
1470 __le16 pkt_info
= rx_desc
->wb
.lower
.lo_dword
.hs_rss
.pkt_info
;
1471 bool encap_pkt
= false;
1473 skb_checksum_none_assert(skb
);
1475 /* Rx csum disabled */
1476 if (!(ring
->netdev
->features
& NETIF_F_RXCSUM
))
1479 /* check for VXLAN and Geneve packets */
1480 if (pkt_info
& cpu_to_le16(IXGBE_RXDADV_PKTTYPE_VXLAN
)) {
1482 skb
->encapsulation
= 1;
1485 /* if IP and error */
1486 if (ixgbe_test_staterr(rx_desc
, IXGBE_RXD_STAT_IPCS
) &&
1487 ixgbe_test_staterr(rx_desc
, IXGBE_RXDADV_ERR_IPE
)) {
1488 ring
->rx_stats
.csum_err
++;
1492 if (!ixgbe_test_staterr(rx_desc
, IXGBE_RXD_STAT_L4CS
))
1495 if (ixgbe_test_staterr(rx_desc
, IXGBE_RXDADV_ERR_TCPE
)) {
1497 * 82599 errata, UDP frames with a 0 checksum can be marked as
1500 if ((pkt_info
& cpu_to_le16(IXGBE_RXDADV_PKTTYPE_UDP
)) &&
1501 test_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR
, &ring
->state
))
1504 ring
->rx_stats
.csum_err
++;
1508 /* It must be a TCP or UDP packet with a valid checksum */
1509 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
1511 if (!ixgbe_test_staterr(rx_desc
, IXGBE_RXD_STAT_OUTERIPCS
))
1514 if (ixgbe_test_staterr(rx_desc
, IXGBE_RXDADV_ERR_OUTERIPER
)) {
1515 skb
->ip_summed
= CHECKSUM_NONE
;
1518 /* If we checked the outer header let the stack know */
1519 skb
->csum_level
= 1;
1523 static inline unsigned int ixgbe_rx_offset(struct ixgbe_ring
*rx_ring
)
1525 return ring_uses_build_skb(rx_ring
) ? IXGBE_SKB_PAD
: 0;
1528 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring
*rx_ring
,
1529 struct ixgbe_rx_buffer
*bi
)
1531 struct page
*page
= bi
->page
;
1534 /* since we are recycling buffers we should seldom need to alloc */
1538 /* alloc new page for storage */
1539 page
= dev_alloc_pages(ixgbe_rx_pg_order(rx_ring
));
1540 if (unlikely(!page
)) {
1541 rx_ring
->rx_stats
.alloc_rx_page_failed
++;
1545 /* map page for use */
1546 dma
= dma_map_page_attrs(rx_ring
->dev
, page
, 0,
1547 ixgbe_rx_pg_size(rx_ring
),
1552 * if mapping failed free memory back to system since
1553 * there isn't much point in holding memory we can't use
1555 if (dma_mapping_error(rx_ring
->dev
, dma
)) {
1556 __free_pages(page
, ixgbe_rx_pg_order(rx_ring
));
1558 rx_ring
->rx_stats
.alloc_rx_page_failed
++;
1564 bi
->page_offset
= ixgbe_rx_offset(rx_ring
);
1565 page_ref_add(page
, USHRT_MAX
- 1);
1566 bi
->pagecnt_bias
= USHRT_MAX
;
1567 rx_ring
->rx_stats
.alloc_rx_page
++;
1573 * ixgbe_alloc_rx_buffers - Replace used receive buffers
1574 * @rx_ring: ring to place buffers on
1575 * @cleaned_count: number of buffers to replace
1577 void ixgbe_alloc_rx_buffers(struct ixgbe_ring
*rx_ring
, u16 cleaned_count
)
1579 union ixgbe_adv_rx_desc
*rx_desc
;
1580 struct ixgbe_rx_buffer
*bi
;
1581 u16 i
= rx_ring
->next_to_use
;
1588 rx_desc
= IXGBE_RX_DESC(rx_ring
, i
);
1589 bi
= &rx_ring
->rx_buffer_info
[i
];
1590 i
-= rx_ring
->count
;
1592 bufsz
= ixgbe_rx_bufsz(rx_ring
);
1595 if (!ixgbe_alloc_mapped_page(rx_ring
, bi
))
1598 /* sync the buffer for use by the device */
1599 dma_sync_single_range_for_device(rx_ring
->dev
, bi
->dma
,
1600 bi
->page_offset
, bufsz
,
1604 * Refresh the desc even if buffer_addrs didn't change
1605 * because each write-back erases this info.
1607 rx_desc
->read
.pkt_addr
= cpu_to_le64(bi
->dma
+ bi
->page_offset
);
1613 rx_desc
= IXGBE_RX_DESC(rx_ring
, 0);
1614 bi
= rx_ring
->rx_buffer_info
;
1615 i
-= rx_ring
->count
;
1618 /* clear the length for the next_to_use descriptor */
1619 rx_desc
->wb
.upper
.length
= 0;
1622 } while (cleaned_count
);
1624 i
+= rx_ring
->count
;
1626 if (rx_ring
->next_to_use
!= i
) {
1627 rx_ring
->next_to_use
= i
;
1629 /* update next to alloc since we have filled the ring */
1630 rx_ring
->next_to_alloc
= i
;
1632 /* Force memory writes to complete before letting h/w
1633 * know there are new descriptors to fetch. (Only
1634 * applicable for weak-ordered memory model archs,
1638 writel(i
, rx_ring
->tail
);
1642 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring
*ring
,
1643 struct sk_buff
*skb
)
1645 u16 hdr_len
= skb_headlen(skb
);
1647 /* set gso_size to avoid messing up TCP MSS */
1648 skb_shinfo(skb
)->gso_size
= DIV_ROUND_UP((skb
->len
- hdr_len
),
1649 IXGBE_CB(skb
)->append_cnt
);
1650 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV4
;
1653 static void ixgbe_update_rsc_stats(struct ixgbe_ring
*rx_ring
,
1654 struct sk_buff
*skb
)
1656 /* if append_cnt is 0 then frame is not RSC */
1657 if (!IXGBE_CB(skb
)->append_cnt
)
1660 rx_ring
->rx_stats
.rsc_count
+= IXGBE_CB(skb
)->append_cnt
;
1661 rx_ring
->rx_stats
.rsc_flush
++;
1663 ixgbe_set_rsc_gso_size(rx_ring
, skb
);
1665 /* gso_size is computed using append_cnt so always clear it last */
1666 IXGBE_CB(skb
)->append_cnt
= 0;
1670 * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor
1671 * @rx_ring: rx descriptor ring packet is being transacted on
1672 * @rx_desc: pointer to the EOP Rx descriptor
1673 * @skb: pointer to current skb being populated
1675 * This function checks the ring, descriptor, and packet information in
1676 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1677 * other fields within the skb.
1679 void ixgbe_process_skb_fields(struct ixgbe_ring
*rx_ring
,
1680 union ixgbe_adv_rx_desc
*rx_desc
,
1681 struct sk_buff
*skb
)
1683 struct net_device
*dev
= rx_ring
->netdev
;
1684 u32 flags
= rx_ring
->q_vector
->adapter
->flags
;
1686 ixgbe_update_rsc_stats(rx_ring
, skb
);
1688 ixgbe_rx_hash(rx_ring
, rx_desc
, skb
);
1690 ixgbe_rx_checksum(rx_ring
, rx_desc
, skb
);
1692 if (unlikely(flags
& IXGBE_FLAG_RX_HWTSTAMP_ENABLED
))
1693 ixgbe_ptp_rx_hwtstamp(rx_ring
, rx_desc
, skb
);
1695 if ((dev
->features
& NETIF_F_HW_VLAN_CTAG_RX
) &&
1696 ixgbe_test_staterr(rx_desc
, IXGBE_RXD_STAT_VP
)) {
1697 u16 vid
= le16_to_cpu(rx_desc
->wb
.upper
.vlan
);
1698 __vlan_hwaccel_put_tag(skb
, htons(ETH_P_8021Q
), vid
);
1701 if (ixgbe_test_staterr(rx_desc
, IXGBE_RXDADV_STAT_SECP
))
1702 ixgbe_ipsec_rx(rx_ring
, rx_desc
, skb
);
1704 /* record Rx queue, or update MACVLAN statistics */
1705 if (netif_is_ixgbe(dev
))
1706 skb_record_rx_queue(skb
, rx_ring
->queue_index
);
1708 macvlan_count_rx(netdev_priv(dev
), skb
->len
+ ETH_HLEN
, true,
1711 skb
->protocol
= eth_type_trans(skb
, dev
);
1714 void ixgbe_rx_skb(struct ixgbe_q_vector
*q_vector
,
1715 struct sk_buff
*skb
)
1717 napi_gro_receive(&q_vector
->napi
, skb
);
1721 * ixgbe_is_non_eop - process handling of non-EOP buffers
1722 * @rx_ring: Rx ring being processed
1723 * @rx_desc: Rx descriptor for current buffer
1724 * @skb: Current socket buffer containing buffer in progress
1726 * This function updates next to clean. If the buffer is an EOP buffer
1727 * this function exits returning false, otherwise it will place the
1728 * sk_buff in the next buffer to be chained and return true indicating
1729 * that this is in fact a non-EOP buffer.
1731 static bool ixgbe_is_non_eop(struct ixgbe_ring
*rx_ring
,
1732 union ixgbe_adv_rx_desc
*rx_desc
,
1733 struct sk_buff
*skb
)
1735 u32 ntc
= rx_ring
->next_to_clean
+ 1;
1737 /* fetch, update, and store next to clean */
1738 ntc
= (ntc
< rx_ring
->count
) ? ntc
: 0;
1739 rx_ring
->next_to_clean
= ntc
;
1741 prefetch(IXGBE_RX_DESC(rx_ring
, ntc
));
1743 /* update RSC append count if present */
1744 if (ring_is_rsc_enabled(rx_ring
)) {
1745 __le32 rsc_enabled
= rx_desc
->wb
.lower
.lo_dword
.data
&
1746 cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK
);
1748 if (unlikely(rsc_enabled
)) {
1749 u32 rsc_cnt
= le32_to_cpu(rsc_enabled
);
1751 rsc_cnt
>>= IXGBE_RXDADV_RSCCNT_SHIFT
;
1752 IXGBE_CB(skb
)->append_cnt
+= rsc_cnt
- 1;
1754 /* update ntc based on RSC value */
1755 ntc
= le32_to_cpu(rx_desc
->wb
.upper
.status_error
);
1756 ntc
&= IXGBE_RXDADV_NEXTP_MASK
;
1757 ntc
>>= IXGBE_RXDADV_NEXTP_SHIFT
;
1761 /* if we are the last buffer then there is nothing else to do */
1762 if (likely(ixgbe_test_staterr(rx_desc
, IXGBE_RXD_STAT_EOP
)))
1765 /* place skb in next buffer to be received */
1766 rx_ring
->rx_buffer_info
[ntc
].skb
= skb
;
1767 rx_ring
->rx_stats
.non_eop_descs
++;
1773 * ixgbe_pull_tail - ixgbe specific version of skb_pull_tail
1774 * @rx_ring: rx descriptor ring packet is being transacted on
1775 * @skb: pointer to current skb being adjusted
1777 * This function is an ixgbe specific version of __pskb_pull_tail. The
1778 * main difference between this version and the original function is that
1779 * this function can make several assumptions about the state of things
1780 * that allow for significant optimizations versus the standard function.
1781 * As a result we can do things like drop a frag and maintain an accurate
1782 * truesize for the skb.
1784 static void ixgbe_pull_tail(struct ixgbe_ring
*rx_ring
,
1785 struct sk_buff
*skb
)
1787 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[0];
1789 unsigned int pull_len
;
1792 * it is valid to use page_address instead of kmap since we are
1793 * working with pages allocated out of the lomem pool per
1794 * alloc_page(GFP_ATOMIC)
1796 va
= skb_frag_address(frag
);
1799 * we need the header to contain the greater of either ETH_HLEN or
1800 * 60 bytes if the skb->len is less than 60 for skb_pad.
1802 pull_len
= eth_get_headlen(skb
->dev
, va
, IXGBE_RX_HDR_SIZE
);
1804 /* align pull length to size of long to optimize memcpy performance */
1805 skb_copy_to_linear_data(skb
, va
, ALIGN(pull_len
, sizeof(long)));
1807 /* update all of the pointers */
1808 skb_frag_size_sub(frag
, pull_len
);
1809 skb_frag_off_add(frag
, pull_len
);
1810 skb
->data_len
-= pull_len
;
1811 skb
->tail
+= pull_len
;
1815 * ixgbe_dma_sync_frag - perform DMA sync for first frag of SKB
1816 * @rx_ring: rx descriptor ring packet is being transacted on
1817 * @skb: pointer to current skb being updated
1819 * This function provides a basic DMA sync up for the first fragment of an
1820 * skb. The reason for doing this is that the first fragment cannot be
1821 * unmapped until we have reached the end of packet descriptor for a buffer
1824 static void ixgbe_dma_sync_frag(struct ixgbe_ring
*rx_ring
,
1825 struct sk_buff
*skb
)
1827 if (ring_uses_build_skb(rx_ring
)) {
1828 unsigned long offset
= (unsigned long)(skb
->data
) & ~PAGE_MASK
;
1830 dma_sync_single_range_for_cpu(rx_ring
->dev
,
1836 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[0];
1838 dma_sync_single_range_for_cpu(rx_ring
->dev
,
1841 skb_frag_size(frag
),
1845 /* If the page was released, just unmap it. */
1846 if (unlikely(IXGBE_CB(skb
)->page_released
)) {
1847 dma_unmap_page_attrs(rx_ring
->dev
, IXGBE_CB(skb
)->dma
,
1848 ixgbe_rx_pg_size(rx_ring
),
1855 * ixgbe_cleanup_headers - Correct corrupted or empty headers
1856 * @rx_ring: rx descriptor ring packet is being transacted on
1857 * @rx_desc: pointer to the EOP Rx descriptor
1858 * @skb: pointer to current skb being fixed
1860 * Check if the skb is valid in the XDP case it will be an error pointer.
1861 * Return true in this case to abort processing and advance to next
1864 * Check for corrupted packet headers caused by senders on the local L2
1865 * embedded NIC switch not setting up their Tx Descriptors right. These
1866 * should be very rare.
1868 * Also address the case where we are pulling data in on pages only
1869 * and as such no data is present in the skb header.
1871 * In addition if skb is not at least 60 bytes we need to pad it so that
1872 * it is large enough to qualify as a valid Ethernet frame.
1874 * Returns true if an error was encountered and skb was freed.
1876 bool ixgbe_cleanup_headers(struct ixgbe_ring
*rx_ring
,
1877 union ixgbe_adv_rx_desc
*rx_desc
,
1878 struct sk_buff
*skb
)
1880 struct net_device
*netdev
= rx_ring
->netdev
;
1882 /* XDP packets use error pointer so abort at this point */
1886 /* Verify netdev is present, and that packet does not have any
1887 * errors that would be unacceptable to the netdev.
1890 (unlikely(ixgbe_test_staterr(rx_desc
,
1891 IXGBE_RXDADV_ERR_FRAME_ERR_MASK
) &&
1892 !(netdev
->features
& NETIF_F_RXALL
)))) {
1893 dev_kfree_skb_any(skb
);
1897 /* place header in linear portion of buffer */
1898 if (!skb_headlen(skb
))
1899 ixgbe_pull_tail(rx_ring
, skb
);
1902 /* do not attempt to pad FCoE Frames as this will disrupt DDP */
1903 if (ixgbe_rx_is_fcoe(rx_ring
, rx_desc
))
1907 /* if eth_skb_pad returns an error the skb was freed */
1908 if (eth_skb_pad(skb
))
1915 * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1916 * @rx_ring: rx descriptor ring to store buffers on
1917 * @old_buff: donor buffer to have page reused
1919 * Synchronizes page for reuse by the adapter
1921 static void ixgbe_reuse_rx_page(struct ixgbe_ring
*rx_ring
,
1922 struct ixgbe_rx_buffer
*old_buff
)
1924 struct ixgbe_rx_buffer
*new_buff
;
1925 u16 nta
= rx_ring
->next_to_alloc
;
1927 new_buff
= &rx_ring
->rx_buffer_info
[nta
];
1929 /* update, and store next to alloc */
1931 rx_ring
->next_to_alloc
= (nta
< rx_ring
->count
) ? nta
: 0;
1933 /* Transfer page from old buffer to new buffer.
1934 * Move each member individually to avoid possible store
1935 * forwarding stalls and unnecessary copy of skb.
1937 new_buff
->dma
= old_buff
->dma
;
1938 new_buff
->page
= old_buff
->page
;
1939 new_buff
->page_offset
= old_buff
->page_offset
;
1940 new_buff
->pagecnt_bias
= old_buff
->pagecnt_bias
;
1943 static inline bool ixgbe_page_is_reserved(struct page
*page
)
1945 return (page_to_nid(page
) != numa_mem_id()) || page_is_pfmemalloc(page
);
1948 static bool ixgbe_can_reuse_rx_page(struct ixgbe_rx_buffer
*rx_buffer
)
1950 unsigned int pagecnt_bias
= rx_buffer
->pagecnt_bias
;
1951 struct page
*page
= rx_buffer
->page
;
1953 /* avoid re-using remote pages */
1954 if (unlikely(ixgbe_page_is_reserved(page
)))
1957 #if (PAGE_SIZE < 8192)
1958 /* if we are only owner of page we can reuse it */
1959 if (unlikely((page_ref_count(page
) - pagecnt_bias
) > 1))
1962 /* The last offset is a bit aggressive in that we assume the
1963 * worst case of FCoE being enabled and using a 3K buffer.
1964 * However this should have minimal impact as the 1K extra is
1965 * still less than one buffer in size.
1967 #define IXGBE_LAST_OFFSET \
1968 (SKB_WITH_OVERHEAD(PAGE_SIZE) - IXGBE_RXBUFFER_3K)
1969 if (rx_buffer
->page_offset
> IXGBE_LAST_OFFSET
)
1973 /* If we have drained the page fragment pool we need to update
1974 * the pagecnt_bias and page count so that we fully restock the
1975 * number of references the driver holds.
1977 if (unlikely(pagecnt_bias
== 1)) {
1978 page_ref_add(page
, USHRT_MAX
- 1);
1979 rx_buffer
->pagecnt_bias
= USHRT_MAX
;
1986 * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1987 * @rx_ring: rx descriptor ring to transact packets on
1988 * @rx_buffer: buffer containing page to add
1989 * @skb: sk_buff to place the data into
1990 * @size: size of data in rx_buffer
1992 * This function will add the data contained in rx_buffer->page to the skb.
1993 * This is done either through a direct copy if the data in the buffer is
1994 * less than the skb header size, otherwise it will just attach the page as
1995 * a frag to the skb.
1997 * The function will then update the page offset if necessary and return
1998 * true if the buffer can be reused by the adapter.
2000 static void ixgbe_add_rx_frag(struct ixgbe_ring
*rx_ring
,
2001 struct ixgbe_rx_buffer
*rx_buffer
,
2002 struct sk_buff
*skb
,
2005 #if (PAGE_SIZE < 8192)
2006 unsigned int truesize
= ixgbe_rx_pg_size(rx_ring
) / 2;
2008 unsigned int truesize
= ring_uses_build_skb(rx_ring
) ?
2009 SKB_DATA_ALIGN(IXGBE_SKB_PAD
+ size
) :
2010 SKB_DATA_ALIGN(size
);
2012 skb_add_rx_frag(skb
, skb_shinfo(skb
)->nr_frags
, rx_buffer
->page
,
2013 rx_buffer
->page_offset
, size
, truesize
);
2014 #if (PAGE_SIZE < 8192)
2015 rx_buffer
->page_offset
^= truesize
;
2017 rx_buffer
->page_offset
+= truesize
;
2021 static struct ixgbe_rx_buffer
*ixgbe_get_rx_buffer(struct ixgbe_ring
*rx_ring
,
2022 union ixgbe_adv_rx_desc
*rx_desc
,
2023 struct sk_buff
**skb
,
2024 const unsigned int size
)
2026 struct ixgbe_rx_buffer
*rx_buffer
;
2028 rx_buffer
= &rx_ring
->rx_buffer_info
[rx_ring
->next_to_clean
];
2029 prefetchw(rx_buffer
->page
);
2030 *skb
= rx_buffer
->skb
;
2032 /* Delay unmapping of the first packet. It carries the header
2033 * information, HW may still access the header after the writeback.
2034 * Only unmap it when EOP is reached
2036 if (!ixgbe_test_staterr(rx_desc
, IXGBE_RXD_STAT_EOP
)) {
2041 ixgbe_dma_sync_frag(rx_ring
, *skb
);
2044 /* we are reusing so sync this buffer for CPU use */
2045 dma_sync_single_range_for_cpu(rx_ring
->dev
,
2047 rx_buffer
->page_offset
,
2051 rx_buffer
->pagecnt_bias
--;
2056 static void ixgbe_put_rx_buffer(struct ixgbe_ring
*rx_ring
,
2057 struct ixgbe_rx_buffer
*rx_buffer
,
2058 struct sk_buff
*skb
)
2060 if (ixgbe_can_reuse_rx_page(rx_buffer
)) {
2061 /* hand second half of page back to the ring */
2062 ixgbe_reuse_rx_page(rx_ring
, rx_buffer
);
2064 if (!IS_ERR(skb
) && IXGBE_CB(skb
)->dma
== rx_buffer
->dma
) {
2065 /* the page has been released from the ring */
2066 IXGBE_CB(skb
)->page_released
= true;
2068 /* we are not reusing the buffer so unmap it */
2069 dma_unmap_page_attrs(rx_ring
->dev
, rx_buffer
->dma
,
2070 ixgbe_rx_pg_size(rx_ring
),
2074 __page_frag_cache_drain(rx_buffer
->page
,
2075 rx_buffer
->pagecnt_bias
);
2078 /* clear contents of rx_buffer */
2079 rx_buffer
->page
= NULL
;
2080 rx_buffer
->skb
= NULL
;
2083 static struct sk_buff
*ixgbe_construct_skb(struct ixgbe_ring
*rx_ring
,
2084 struct ixgbe_rx_buffer
*rx_buffer
,
2085 struct xdp_buff
*xdp
,
2086 union ixgbe_adv_rx_desc
*rx_desc
)
2088 unsigned int size
= xdp
->data_end
- xdp
->data
;
2089 #if (PAGE_SIZE < 8192)
2090 unsigned int truesize
= ixgbe_rx_pg_size(rx_ring
) / 2;
2092 unsigned int truesize
= SKB_DATA_ALIGN(xdp
->data_end
-
2093 xdp
->data_hard_start
);
2095 struct sk_buff
*skb
;
2097 /* prefetch first cache line of first page */
2098 prefetch(xdp
->data
);
2099 #if L1_CACHE_BYTES < 128
2100 prefetch(xdp
->data
+ L1_CACHE_BYTES
);
2102 /* Note, we get here by enabling legacy-rx via:
2104 * ethtool --set-priv-flags <dev> legacy-rx on
2106 * In this mode, we currently get 0 extra XDP headroom as
2107 * opposed to having legacy-rx off, where we process XDP
2108 * packets going to stack via ixgbe_build_skb(). The latter
2109 * provides us currently with 192 bytes of headroom.
2111 * For ixgbe_construct_skb() mode it means that the
2112 * xdp->data_meta will always point to xdp->data, since
2113 * the helper cannot expand the head. Should this ever
2114 * change in future for legacy-rx mode on, then lets also
2115 * add xdp->data_meta handling here.
2118 /* allocate a skb to store the frags */
2119 skb
= napi_alloc_skb(&rx_ring
->q_vector
->napi
, IXGBE_RX_HDR_SIZE
);
2123 if (size
> IXGBE_RX_HDR_SIZE
) {
2124 if (!ixgbe_test_staterr(rx_desc
, IXGBE_RXD_STAT_EOP
))
2125 IXGBE_CB(skb
)->dma
= rx_buffer
->dma
;
2127 skb_add_rx_frag(skb
, 0, rx_buffer
->page
,
2128 xdp
->data
- page_address(rx_buffer
->page
),
2130 #if (PAGE_SIZE < 8192)
2131 rx_buffer
->page_offset
^= truesize
;
2133 rx_buffer
->page_offset
+= truesize
;
2136 memcpy(__skb_put(skb
, size
),
2137 xdp
->data
, ALIGN(size
, sizeof(long)));
2138 rx_buffer
->pagecnt_bias
++;
2144 static struct sk_buff
*ixgbe_build_skb(struct ixgbe_ring
*rx_ring
,
2145 struct ixgbe_rx_buffer
*rx_buffer
,
2146 struct xdp_buff
*xdp
,
2147 union ixgbe_adv_rx_desc
*rx_desc
)
2149 unsigned int metasize
= xdp
->data
- xdp
->data_meta
;
2150 #if (PAGE_SIZE < 8192)
2151 unsigned int truesize
= ixgbe_rx_pg_size(rx_ring
) / 2;
2153 unsigned int truesize
= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
)) +
2154 SKB_DATA_ALIGN(xdp
->data_end
-
2155 xdp
->data_hard_start
);
2157 struct sk_buff
*skb
;
2159 /* Prefetch first cache line of first page. If xdp->data_meta
2160 * is unused, this points extactly as xdp->data, otherwise we
2161 * likely have a consumer accessing first few bytes of meta
2162 * data, and then actual data.
2164 prefetch(xdp
->data_meta
);
2165 #if L1_CACHE_BYTES < 128
2166 prefetch(xdp
->data_meta
+ L1_CACHE_BYTES
);
2169 /* build an skb to around the page buffer */
2170 skb
= build_skb(xdp
->data_hard_start
, truesize
);
2174 /* update pointers within the skb to store the data */
2175 skb_reserve(skb
, xdp
->data
- xdp
->data_hard_start
);
2176 __skb_put(skb
, xdp
->data_end
- xdp
->data
);
2178 skb_metadata_set(skb
, metasize
);
2180 /* record DMA address if this is the start of a chain of buffers */
2181 if (!ixgbe_test_staterr(rx_desc
, IXGBE_RXD_STAT_EOP
))
2182 IXGBE_CB(skb
)->dma
= rx_buffer
->dma
;
2184 /* update buffer offset */
2185 #if (PAGE_SIZE < 8192)
2186 rx_buffer
->page_offset
^= truesize
;
2188 rx_buffer
->page_offset
+= truesize
;
2194 static struct sk_buff
*ixgbe_run_xdp(struct ixgbe_adapter
*adapter
,
2195 struct ixgbe_ring
*rx_ring
,
2196 struct xdp_buff
*xdp
)
2198 int err
, result
= IXGBE_XDP_PASS
;
2199 struct bpf_prog
*xdp_prog
;
2200 struct xdp_frame
*xdpf
;
2204 xdp_prog
= READ_ONCE(rx_ring
->xdp_prog
);
2209 prefetchw(xdp
->data_hard_start
); /* xdp_frame write */
2211 act
= bpf_prog_run_xdp(xdp_prog
, xdp
);
2216 xdpf
= xdp_convert_buff_to_frame(xdp
);
2217 if (unlikely(!xdpf
)) {
2218 result
= IXGBE_XDP_CONSUMED
;
2221 result
= ixgbe_xmit_xdp_ring(adapter
, xdpf
);
2224 err
= xdp_do_redirect(adapter
->netdev
, xdp
, xdp_prog
);
2226 result
= IXGBE_XDP_REDIR
;
2228 result
= IXGBE_XDP_CONSUMED
;
2231 bpf_warn_invalid_xdp_action(act
);
2234 trace_xdp_exception(rx_ring
->netdev
, xdp_prog
, act
);
2235 fallthrough
; /* handle aborts by dropping packet */
2237 result
= IXGBE_XDP_CONSUMED
;
2242 return ERR_PTR(-result
);
2245 static unsigned int ixgbe_rx_frame_truesize(struct ixgbe_ring
*rx_ring
,
2248 unsigned int truesize
;
2250 #if (PAGE_SIZE < 8192)
2251 truesize
= ixgbe_rx_pg_size(rx_ring
) / 2; /* Must be power-of-2 */
2253 truesize
= ring_uses_build_skb(rx_ring
) ?
2254 SKB_DATA_ALIGN(IXGBE_SKB_PAD
+ size
) +
2255 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
)) :
2256 SKB_DATA_ALIGN(size
);
2261 static void ixgbe_rx_buffer_flip(struct ixgbe_ring
*rx_ring
,
2262 struct ixgbe_rx_buffer
*rx_buffer
,
2265 unsigned int truesize
= ixgbe_rx_frame_truesize(rx_ring
, size
);
2266 #if (PAGE_SIZE < 8192)
2267 rx_buffer
->page_offset
^= truesize
;
2269 rx_buffer
->page_offset
+= truesize
;
2274 * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
2275 * @q_vector: structure containing interrupt and ring information
2276 * @rx_ring: rx descriptor ring to transact packets on
2277 * @budget: Total limit on number of packets to process
2279 * This function provides a "bounce buffer" approach to Rx interrupt
2280 * processing. The advantage to this is that on systems that have
2281 * expensive overhead for IOMMU access this provides a means of avoiding
2282 * it by maintaining the mapping of the page to the syste.
2284 * Returns amount of work completed
2286 static int ixgbe_clean_rx_irq(struct ixgbe_q_vector
*q_vector
,
2287 struct ixgbe_ring
*rx_ring
,
2290 unsigned int total_rx_bytes
= 0, total_rx_packets
= 0;
2291 struct ixgbe_adapter
*adapter
= q_vector
->adapter
;
2294 unsigned int mss
= 0;
2295 #endif /* IXGBE_FCOE */
2296 u16 cleaned_count
= ixgbe_desc_unused(rx_ring
);
2297 unsigned int xdp_xmit
= 0;
2298 struct xdp_buff xdp
;
2300 xdp
.rxq
= &rx_ring
->xdp_rxq
;
2302 /* Frame size depend on rx_ring setup when PAGE_SIZE=4K */
2303 #if (PAGE_SIZE < 8192)
2304 xdp
.frame_sz
= ixgbe_rx_frame_truesize(rx_ring
, 0);
2307 while (likely(total_rx_packets
< budget
)) {
2308 union ixgbe_adv_rx_desc
*rx_desc
;
2309 struct ixgbe_rx_buffer
*rx_buffer
;
2310 struct sk_buff
*skb
;
2313 /* return some buffers to hardware, one at a time is too slow */
2314 if (cleaned_count
>= IXGBE_RX_BUFFER_WRITE
) {
2315 ixgbe_alloc_rx_buffers(rx_ring
, cleaned_count
);
2319 rx_desc
= IXGBE_RX_DESC(rx_ring
, rx_ring
->next_to_clean
);
2320 size
= le16_to_cpu(rx_desc
->wb
.upper
.length
);
2324 /* This memory barrier is needed to keep us from reading
2325 * any other fields out of the rx_desc until we know the
2326 * descriptor has been written back
2330 rx_buffer
= ixgbe_get_rx_buffer(rx_ring
, rx_desc
, &skb
, size
);
2332 /* retrieve a buffer from the ring */
2334 xdp
.data
= page_address(rx_buffer
->page
) +
2335 rx_buffer
->page_offset
;
2336 xdp
.data_meta
= xdp
.data
;
2337 xdp
.data_hard_start
= xdp
.data
-
2338 ixgbe_rx_offset(rx_ring
);
2339 xdp
.data_end
= xdp
.data
+ size
;
2340 #if (PAGE_SIZE > 4096)
2341 /* At larger PAGE_SIZE, frame_sz depend on len size */
2342 xdp
.frame_sz
= ixgbe_rx_frame_truesize(rx_ring
, size
);
2344 skb
= ixgbe_run_xdp(adapter
, rx_ring
, &xdp
);
2348 unsigned int xdp_res
= -PTR_ERR(skb
);
2350 if (xdp_res
& (IXGBE_XDP_TX
| IXGBE_XDP_REDIR
)) {
2351 xdp_xmit
|= xdp_res
;
2352 ixgbe_rx_buffer_flip(rx_ring
, rx_buffer
, size
);
2354 rx_buffer
->pagecnt_bias
++;
2357 total_rx_bytes
+= size
;
2359 ixgbe_add_rx_frag(rx_ring
, rx_buffer
, skb
, size
);
2360 } else if (ring_uses_build_skb(rx_ring
)) {
2361 skb
= ixgbe_build_skb(rx_ring
, rx_buffer
,
2364 skb
= ixgbe_construct_skb(rx_ring
, rx_buffer
,
2368 /* exit if we failed to retrieve a buffer */
2370 rx_ring
->rx_stats
.alloc_rx_buff_failed
++;
2371 rx_buffer
->pagecnt_bias
++;
2375 ixgbe_put_rx_buffer(rx_ring
, rx_buffer
, skb
);
2378 /* place incomplete frames back on ring for completion */
2379 if (ixgbe_is_non_eop(rx_ring
, rx_desc
, skb
))
2382 /* verify the packet layout is correct */
2383 if (ixgbe_cleanup_headers(rx_ring
, rx_desc
, skb
))
2386 /* probably a little skewed due to removing CRC */
2387 total_rx_bytes
+= skb
->len
;
2389 /* populate checksum, timestamp, VLAN, and protocol */
2390 ixgbe_process_skb_fields(rx_ring
, rx_desc
, skb
);
2393 /* if ddp, not passing to ULD unless for FCP_RSP or error */
2394 if (ixgbe_rx_is_fcoe(rx_ring
, rx_desc
)) {
2395 ddp_bytes
= ixgbe_fcoe_ddp(adapter
, rx_desc
, skb
);
2396 /* include DDPed FCoE data */
2397 if (ddp_bytes
> 0) {
2399 mss
= rx_ring
->netdev
->mtu
-
2400 sizeof(struct fcoe_hdr
) -
2401 sizeof(struct fc_frame_header
) -
2402 sizeof(struct fcoe_crc_eof
);
2406 total_rx_bytes
+= ddp_bytes
;
2407 total_rx_packets
+= DIV_ROUND_UP(ddp_bytes
,
2411 dev_kfree_skb_any(skb
);
2416 #endif /* IXGBE_FCOE */
2417 ixgbe_rx_skb(q_vector
, skb
);
2419 /* update budget accounting */
2423 if (xdp_xmit
& IXGBE_XDP_REDIR
)
2426 if (xdp_xmit
& IXGBE_XDP_TX
) {
2427 struct ixgbe_ring
*ring
= adapter
->xdp_ring
[smp_processor_id()];
2429 /* Force memory writes to complete before letting h/w
2430 * know there are new descriptors to fetch.
2433 writel(ring
->next_to_use
, ring
->tail
);
2436 u64_stats_update_begin(&rx_ring
->syncp
);
2437 rx_ring
->stats
.packets
+= total_rx_packets
;
2438 rx_ring
->stats
.bytes
+= total_rx_bytes
;
2439 u64_stats_update_end(&rx_ring
->syncp
);
2440 q_vector
->rx
.total_packets
+= total_rx_packets
;
2441 q_vector
->rx
.total_bytes
+= total_rx_bytes
;
2443 return total_rx_packets
;
2447 * ixgbe_configure_msix - Configure MSI-X hardware
2448 * @adapter: board private structure
2450 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
2453 static void ixgbe_configure_msix(struct ixgbe_adapter
*adapter
)
2455 struct ixgbe_q_vector
*q_vector
;
2459 /* Populate MSIX to EITR Select */
2460 if (adapter
->num_vfs
> 32) {
2461 u32 eitrsel
= BIT(adapter
->num_vfs
- 32) - 1;
2462 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EITRSEL
, eitrsel
);
2466 * Populate the IVAR table and set the ITR values to the
2467 * corresponding register.
2469 for (v_idx
= 0; v_idx
< adapter
->num_q_vectors
; v_idx
++) {
2470 struct ixgbe_ring
*ring
;
2471 q_vector
= adapter
->q_vector
[v_idx
];
2473 ixgbe_for_each_ring(ring
, q_vector
->rx
)
2474 ixgbe_set_ivar(adapter
, 0, ring
->reg_idx
, v_idx
);
2476 ixgbe_for_each_ring(ring
, q_vector
->tx
)
2477 ixgbe_set_ivar(adapter
, 1, ring
->reg_idx
, v_idx
);
2479 ixgbe_write_eitr(q_vector
);
2482 switch (adapter
->hw
.mac
.type
) {
2483 case ixgbe_mac_82598EB
:
2484 ixgbe_set_ivar(adapter
, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX
,
2487 case ixgbe_mac_82599EB
:
2488 case ixgbe_mac_X540
:
2489 case ixgbe_mac_X550
:
2490 case ixgbe_mac_X550EM_x
:
2491 case ixgbe_mac_x550em_a
:
2492 ixgbe_set_ivar(adapter
, -1, 1, v_idx
);
2497 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EITR(v_idx
), 1950);
2499 /* set up to autoclear timer, and the vectors */
2500 mask
= IXGBE_EIMS_ENABLE_MASK
;
2501 mask
&= ~(IXGBE_EIMS_OTHER
|
2502 IXGBE_EIMS_MAILBOX
|
2505 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EIAC
, mask
);
2509 * ixgbe_update_itr - update the dynamic ITR value based on statistics
2510 * @q_vector: structure containing interrupt and ring information
2511 * @ring_container: structure containing ring performance data
2513 * Stores a new ITR value based on packets and byte
2514 * counts during the last interrupt. The advantage of per interrupt
2515 * computation is faster updates and more accurate ITR for the current
2516 * traffic pattern. Constants in this function were computed
2517 * based on theoretical maximum wire speed and thresholds were set based
2518 * on testing data as well as attempting to minimize response time
2519 * while increasing bulk throughput.
2521 static void ixgbe_update_itr(struct ixgbe_q_vector
*q_vector
,
2522 struct ixgbe_ring_container
*ring_container
)
2524 unsigned int itr
= IXGBE_ITR_ADAPTIVE_MIN_USECS
|
2525 IXGBE_ITR_ADAPTIVE_LATENCY
;
2526 unsigned int avg_wire_size
, packets
, bytes
;
2527 unsigned long next_update
= jiffies
;
2529 /* If we don't have any rings just leave ourselves set for maximum
2530 * possible latency so we take ourselves out of the equation.
2532 if (!ring_container
->ring
)
2535 /* If we didn't update within up to 1 - 2 jiffies we can assume
2536 * that either packets are coming in so slow there hasn't been
2537 * any work, or that there is so much work that NAPI is dealing
2538 * with interrupt moderation and we don't need to do anything.
2540 if (time_after(next_update
, ring_container
->next_update
))
2543 packets
= ring_container
->total_packets
;
2545 /* We have no packets to actually measure against. This means
2546 * either one of the other queues on this vector is active or
2547 * we are a Tx queue doing TSO with too high of an interrupt rate.
2549 * When this occurs just tick up our delay by the minimum value
2550 * and hope that this extra delay will prevent us from being called
2551 * without any work on our queue.
2554 itr
= (q_vector
->itr
>> 2) + IXGBE_ITR_ADAPTIVE_MIN_INC
;
2555 if (itr
> IXGBE_ITR_ADAPTIVE_MAX_USECS
)
2556 itr
= IXGBE_ITR_ADAPTIVE_MAX_USECS
;
2557 itr
+= ring_container
->itr
& IXGBE_ITR_ADAPTIVE_LATENCY
;
2561 bytes
= ring_container
->total_bytes
;
2563 /* If packets are less than 4 or bytes are less than 9000 assume
2564 * insufficient data to use bulk rate limiting approach. We are
2565 * likely latency driven.
2567 if (packets
< 4 && bytes
< 9000) {
2568 itr
= IXGBE_ITR_ADAPTIVE_LATENCY
;
2569 goto adjust_by_size
;
2572 /* Between 4 and 48 we can assume that our current interrupt delay
2573 * is only slightly too low. As such we should increase it by a small
2577 itr
= (q_vector
->itr
>> 2) + IXGBE_ITR_ADAPTIVE_MIN_INC
;
2578 if (itr
> IXGBE_ITR_ADAPTIVE_MAX_USECS
)
2579 itr
= IXGBE_ITR_ADAPTIVE_MAX_USECS
;
2583 /* Between 48 and 96 is our "goldilocks" zone where we are working
2584 * out "just right". Just report that our current ITR is good for us.
2587 itr
= q_vector
->itr
>> 2;
2591 /* If packet count is 96 or greater we are likely looking at a slight
2592 * overrun of the delay we want. Try halving our delay to see if that
2593 * will cut the number of packets in half per interrupt.
2595 if (packets
< 256) {
2596 itr
= q_vector
->itr
>> 3;
2597 if (itr
< IXGBE_ITR_ADAPTIVE_MIN_USECS
)
2598 itr
= IXGBE_ITR_ADAPTIVE_MIN_USECS
;
2602 /* The paths below assume we are dealing with a bulk ITR since number
2603 * of packets is 256 or greater. We are just going to have to compute
2604 * a value and try to bring the count under control, though for smaller
2605 * packet sizes there isn't much we can do as NAPI polling will likely
2606 * be kicking in sooner rather than later.
2608 itr
= IXGBE_ITR_ADAPTIVE_BULK
;
2611 /* If packet counts are 256 or greater we can assume we have a gross
2612 * overestimation of what the rate should be. Instead of trying to fine
2613 * tune it just use the formula below to try and dial in an exact value
2614 * give the current packet size of the frame.
2616 avg_wire_size
= bytes
/ packets
;
2618 /* The following is a crude approximation of:
2619 * wmem_default / (size + overhead) = desired_pkts_per_int
2620 * rate / bits_per_byte / (size + ethernet overhead) = pkt_rate
2621 * (desired_pkt_rate / pkt_rate) * usecs_per_sec = ITR value
2623 * Assuming wmem_default is 212992 and overhead is 640 bytes per
2624 * packet, (256 skb, 64 headroom, 320 shared info), we can reduce the
2627 * (170 * (size + 24)) / (size + 640) = ITR
2629 * We first do some math on the packet size and then finally bitshift
2630 * by 8 after rounding up. We also have to account for PCIe link speed
2631 * difference as ITR scales based on this.
2633 if (avg_wire_size
<= 60) {
2634 /* Start at 50k ints/sec */
2635 avg_wire_size
= 5120;
2636 } else if (avg_wire_size
<= 316) {
2637 /* 50K ints/sec to 16K ints/sec */
2638 avg_wire_size
*= 40;
2639 avg_wire_size
+= 2720;
2640 } else if (avg_wire_size
<= 1084) {
2641 /* 16K ints/sec to 9.2K ints/sec */
2642 avg_wire_size
*= 15;
2643 avg_wire_size
+= 11452;
2644 } else if (avg_wire_size
< 1968) {
2645 /* 9.2K ints/sec to 8K ints/sec */
2647 avg_wire_size
+= 22420;
2649 /* plateau at a limit of 8K ints/sec */
2650 avg_wire_size
= 32256;
2653 /* If we are in low latency mode half our delay which doubles the rate
2654 * to somewhere between 100K to 16K ints/sec
2656 if (itr
& IXGBE_ITR_ADAPTIVE_LATENCY
)
2657 avg_wire_size
>>= 1;
2659 /* Resultant value is 256 times larger than it needs to be. This
2660 * gives us room to adjust the value as needed to either increase
2661 * or decrease the value based on link speeds of 10G, 2.5G, 1G, etc.
2663 * Use addition as we have already recorded the new latency flag
2664 * for the ITR value.
2666 switch (q_vector
->adapter
->link_speed
) {
2667 case IXGBE_LINK_SPEED_10GB_FULL
:
2668 case IXGBE_LINK_SPEED_100_FULL
:
2670 itr
+= DIV_ROUND_UP(avg_wire_size
,
2671 IXGBE_ITR_ADAPTIVE_MIN_INC
* 256) *
2672 IXGBE_ITR_ADAPTIVE_MIN_INC
;
2674 case IXGBE_LINK_SPEED_2_5GB_FULL
:
2675 case IXGBE_LINK_SPEED_1GB_FULL
:
2676 case IXGBE_LINK_SPEED_10_FULL
:
2677 if (avg_wire_size
> 8064)
2678 avg_wire_size
= 8064;
2679 itr
+= DIV_ROUND_UP(avg_wire_size
,
2680 IXGBE_ITR_ADAPTIVE_MIN_INC
* 64) *
2681 IXGBE_ITR_ADAPTIVE_MIN_INC
;
2686 /* write back value */
2687 ring_container
->itr
= itr
;
2689 /* next update should occur within next jiffy */
2690 ring_container
->next_update
= next_update
+ 1;
2692 ring_container
->total_bytes
= 0;
2693 ring_container
->total_packets
= 0;
2697 * ixgbe_write_eitr - write EITR register in hardware specific way
2698 * @q_vector: structure containing interrupt and ring information
2700 * This function is made to be called by ethtool and by the driver
2701 * when it needs to update EITR registers at runtime. Hardware
2702 * specific quirks/differences are taken care of here.
2704 void ixgbe_write_eitr(struct ixgbe_q_vector
*q_vector
)
2706 struct ixgbe_adapter
*adapter
= q_vector
->adapter
;
2707 struct ixgbe_hw
*hw
= &adapter
->hw
;
2708 int v_idx
= q_vector
->v_idx
;
2709 u32 itr_reg
= q_vector
->itr
& IXGBE_MAX_EITR
;
2711 switch (adapter
->hw
.mac
.type
) {
2712 case ixgbe_mac_82598EB
:
2713 /* must write high and low 16 bits to reset counter */
2714 itr_reg
|= (itr_reg
<< 16);
2716 case ixgbe_mac_82599EB
:
2717 case ixgbe_mac_X540
:
2718 case ixgbe_mac_X550
:
2719 case ixgbe_mac_X550EM_x
:
2720 case ixgbe_mac_x550em_a
:
2722 * set the WDIS bit to not clear the timer bits and cause an
2723 * immediate assertion of the interrupt
2725 itr_reg
|= IXGBE_EITR_CNT_WDIS
;
2730 IXGBE_WRITE_REG(hw
, IXGBE_EITR(v_idx
), itr_reg
);
2733 static void ixgbe_set_itr(struct ixgbe_q_vector
*q_vector
)
2737 ixgbe_update_itr(q_vector
, &q_vector
->tx
);
2738 ixgbe_update_itr(q_vector
, &q_vector
->rx
);
2740 /* use the smallest value of new ITR delay calculations */
2741 new_itr
= min(q_vector
->rx
.itr
, q_vector
->tx
.itr
);
2743 /* Clear latency flag if set, shift into correct position */
2744 new_itr
&= ~IXGBE_ITR_ADAPTIVE_LATENCY
;
2747 if (new_itr
!= q_vector
->itr
) {
2748 /* save the algorithm value here */
2749 q_vector
->itr
= new_itr
;
2751 ixgbe_write_eitr(q_vector
);
2756 * ixgbe_check_overtemp_subtask - check for over temperature
2757 * @adapter: pointer to adapter
2759 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter
*adapter
)
2761 struct ixgbe_hw
*hw
= &adapter
->hw
;
2762 u32 eicr
= adapter
->interrupt_event
;
2765 if (test_bit(__IXGBE_DOWN
, &adapter
->state
))
2768 if (!(adapter
->flags2
& IXGBE_FLAG2_TEMP_SENSOR_EVENT
))
2771 adapter
->flags2
&= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT
;
2773 switch (hw
->device_id
) {
2774 case IXGBE_DEV_ID_82599_T3_LOM
:
2776 * Since the warning interrupt is for both ports
2777 * we don't have to check if:
2778 * - This interrupt wasn't for our port.
2779 * - We may have missed the interrupt so always have to
2780 * check if we got a LSC
2782 if (!(eicr
& IXGBE_EICR_GPI_SDP0_8259X
) &&
2783 !(eicr
& IXGBE_EICR_LSC
))
2786 if (!(eicr
& IXGBE_EICR_LSC
) && hw
->mac
.ops
.check_link
) {
2788 bool link_up
= false;
2790 hw
->mac
.ops
.check_link(hw
, &speed
, &link_up
, false);
2796 /* Check if this is not due to overtemp */
2797 if (hw
->phy
.ops
.check_overtemp(hw
) != IXGBE_ERR_OVERTEMP
)
2801 case IXGBE_DEV_ID_X550EM_A_1G_T
:
2802 case IXGBE_DEV_ID_X550EM_A_1G_T_L
:
2803 rc
= hw
->phy
.ops
.check_overtemp(hw
);
2804 if (rc
!= IXGBE_ERR_OVERTEMP
)
2808 if (adapter
->hw
.mac
.type
>= ixgbe_mac_X540
)
2810 if (!(eicr
& IXGBE_EICR_GPI_SDP0(hw
)))
2814 e_crit(drv
, "%s\n", ixgbe_overheat_msg
);
2816 adapter
->interrupt_event
= 0;
2819 static void ixgbe_check_fan_failure(struct ixgbe_adapter
*adapter
, u32 eicr
)
2821 struct ixgbe_hw
*hw
= &adapter
->hw
;
2823 if ((adapter
->flags
& IXGBE_FLAG_FAN_FAIL_CAPABLE
) &&
2824 (eicr
& IXGBE_EICR_GPI_SDP1(hw
))) {
2825 e_crit(probe
, "Fan has stopped, replace the adapter\n");
2826 /* write to clear the interrupt */
2827 IXGBE_WRITE_REG(hw
, IXGBE_EICR
, IXGBE_EICR_GPI_SDP1(hw
));
2831 static void ixgbe_check_overtemp_event(struct ixgbe_adapter
*adapter
, u32 eicr
)
2833 struct ixgbe_hw
*hw
= &adapter
->hw
;
2835 if (!(adapter
->flags2
& IXGBE_FLAG2_TEMP_SENSOR_CAPABLE
))
2838 switch (adapter
->hw
.mac
.type
) {
2839 case ixgbe_mac_82599EB
:
2841 * Need to check link state so complete overtemp check
2844 if (((eicr
& IXGBE_EICR_GPI_SDP0(hw
)) ||
2845 (eicr
& IXGBE_EICR_LSC
)) &&
2846 (!test_bit(__IXGBE_DOWN
, &adapter
->state
))) {
2847 adapter
->interrupt_event
= eicr
;
2848 adapter
->flags2
|= IXGBE_FLAG2_TEMP_SENSOR_EVENT
;
2849 ixgbe_service_event_schedule(adapter
);
2853 case ixgbe_mac_x550em_a
:
2854 if (eicr
& IXGBE_EICR_GPI_SDP0_X550EM_a
) {
2855 adapter
->interrupt_event
= eicr
;
2856 adapter
->flags2
|= IXGBE_FLAG2_TEMP_SENSOR_EVENT
;
2857 ixgbe_service_event_schedule(adapter
);
2858 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EIMC
,
2859 IXGBE_EICR_GPI_SDP0_X550EM_a
);
2860 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EICR
,
2861 IXGBE_EICR_GPI_SDP0_X550EM_a
);
2864 case ixgbe_mac_X550
:
2865 case ixgbe_mac_X540
:
2866 if (!(eicr
& IXGBE_EICR_TS
))
2873 e_crit(drv
, "%s\n", ixgbe_overheat_msg
);
2876 static inline bool ixgbe_is_sfp(struct ixgbe_hw
*hw
)
2878 switch (hw
->mac
.type
) {
2879 case ixgbe_mac_82598EB
:
2880 if (hw
->phy
.type
== ixgbe_phy_nl
)
2883 case ixgbe_mac_82599EB
:
2884 case ixgbe_mac_X550EM_x
:
2885 case ixgbe_mac_x550em_a
:
2886 switch (hw
->mac
.ops
.get_media_type(hw
)) {
2887 case ixgbe_media_type_fiber
:
2888 case ixgbe_media_type_fiber_qsfp
:
2898 static void ixgbe_check_sfp_event(struct ixgbe_adapter
*adapter
, u32 eicr
)
2900 struct ixgbe_hw
*hw
= &adapter
->hw
;
2901 u32 eicr_mask
= IXGBE_EICR_GPI_SDP2(hw
);
2903 if (!ixgbe_is_sfp(hw
))
2906 /* Later MAC's use different SDP */
2907 if (hw
->mac
.type
>= ixgbe_mac_X540
)
2908 eicr_mask
= IXGBE_EICR_GPI_SDP0_X540
;
2910 if (eicr
& eicr_mask
) {
2911 /* Clear the interrupt */
2912 IXGBE_WRITE_REG(hw
, IXGBE_EICR
, eicr_mask
);
2913 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
)) {
2914 adapter
->flags2
|= IXGBE_FLAG2_SFP_NEEDS_RESET
;
2915 adapter
->sfp_poll_time
= 0;
2916 ixgbe_service_event_schedule(adapter
);
2920 if (adapter
->hw
.mac
.type
== ixgbe_mac_82599EB
&&
2921 (eicr
& IXGBE_EICR_GPI_SDP1(hw
))) {
2922 /* Clear the interrupt */
2923 IXGBE_WRITE_REG(hw
, IXGBE_EICR
, IXGBE_EICR_GPI_SDP1(hw
));
2924 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
)) {
2925 adapter
->flags
|= IXGBE_FLAG_NEED_LINK_CONFIG
;
2926 ixgbe_service_event_schedule(adapter
);
2931 static void ixgbe_check_lsc(struct ixgbe_adapter
*adapter
)
2933 struct ixgbe_hw
*hw
= &adapter
->hw
;
2936 adapter
->flags
|= IXGBE_FLAG_NEED_LINK_UPDATE
;
2937 adapter
->link_check_timeout
= jiffies
;
2938 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
)) {
2939 IXGBE_WRITE_REG(hw
, IXGBE_EIMC
, IXGBE_EIMC_LSC
);
2940 IXGBE_WRITE_FLUSH(hw
);
2941 ixgbe_service_event_schedule(adapter
);
2945 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter
*adapter
,
2949 struct ixgbe_hw
*hw
= &adapter
->hw
;
2951 switch (hw
->mac
.type
) {
2952 case ixgbe_mac_82598EB
:
2953 mask
= (IXGBE_EIMS_RTX_QUEUE
& qmask
);
2954 IXGBE_WRITE_REG(hw
, IXGBE_EIMS
, mask
);
2956 case ixgbe_mac_82599EB
:
2957 case ixgbe_mac_X540
:
2958 case ixgbe_mac_X550
:
2959 case ixgbe_mac_X550EM_x
:
2960 case ixgbe_mac_x550em_a
:
2961 mask
= (qmask
& 0xFFFFFFFF);
2963 IXGBE_WRITE_REG(hw
, IXGBE_EIMS_EX(0), mask
);
2964 mask
= (qmask
>> 32);
2966 IXGBE_WRITE_REG(hw
, IXGBE_EIMS_EX(1), mask
);
2971 /* skip the flush */
2975 * ixgbe_irq_enable - Enable default interrupt generation settings
2976 * @adapter: board private structure
2977 * @queues: enable irqs for queues
2978 * @flush: flush register write
2980 static inline void ixgbe_irq_enable(struct ixgbe_adapter
*adapter
, bool queues
,
2983 struct ixgbe_hw
*hw
= &adapter
->hw
;
2984 u32 mask
= (IXGBE_EIMS_ENABLE_MASK
& ~IXGBE_EIMS_RTX_QUEUE
);
2986 /* don't reenable LSC while waiting for link */
2987 if (adapter
->flags
& IXGBE_FLAG_NEED_LINK_UPDATE
)
2988 mask
&= ~IXGBE_EIMS_LSC
;
2990 if (adapter
->flags2
& IXGBE_FLAG2_TEMP_SENSOR_CAPABLE
)
2991 switch (adapter
->hw
.mac
.type
) {
2992 case ixgbe_mac_82599EB
:
2993 mask
|= IXGBE_EIMS_GPI_SDP0(hw
);
2995 case ixgbe_mac_X540
:
2996 case ixgbe_mac_X550
:
2997 case ixgbe_mac_X550EM_x
:
2998 case ixgbe_mac_x550em_a
:
2999 mask
|= IXGBE_EIMS_TS
;
3004 if (adapter
->flags
& IXGBE_FLAG_FAN_FAIL_CAPABLE
)
3005 mask
|= IXGBE_EIMS_GPI_SDP1(hw
);
3006 switch (adapter
->hw
.mac
.type
) {
3007 case ixgbe_mac_82599EB
:
3008 mask
|= IXGBE_EIMS_GPI_SDP1(hw
);
3009 mask
|= IXGBE_EIMS_GPI_SDP2(hw
);
3011 case ixgbe_mac_X540
:
3012 case ixgbe_mac_X550
:
3013 case ixgbe_mac_X550EM_x
:
3014 case ixgbe_mac_x550em_a
:
3015 if (adapter
->hw
.device_id
== IXGBE_DEV_ID_X550EM_X_SFP
||
3016 adapter
->hw
.device_id
== IXGBE_DEV_ID_X550EM_A_SFP
||
3017 adapter
->hw
.device_id
== IXGBE_DEV_ID_X550EM_A_SFP_N
)
3018 mask
|= IXGBE_EIMS_GPI_SDP0(&adapter
->hw
);
3019 if (adapter
->hw
.phy
.type
== ixgbe_phy_x550em_ext_t
)
3020 mask
|= IXGBE_EICR_GPI_SDP0_X540
;
3021 mask
|= IXGBE_EIMS_ECC
;
3022 mask
|= IXGBE_EIMS_MAILBOX
;
3028 if ((adapter
->flags
& IXGBE_FLAG_FDIR_HASH_CAPABLE
) &&
3029 !(adapter
->flags2
& IXGBE_FLAG2_FDIR_REQUIRES_REINIT
))
3030 mask
|= IXGBE_EIMS_FLOW_DIR
;
3032 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EIMS
, mask
);
3034 ixgbe_irq_enable_queues(adapter
, ~0);
3036 IXGBE_WRITE_FLUSH(&adapter
->hw
);
3039 static irqreturn_t
ixgbe_msix_other(int irq
, void *data
)
3041 struct ixgbe_adapter
*adapter
= data
;
3042 struct ixgbe_hw
*hw
= &adapter
->hw
;
3046 * Workaround for Silicon errata. Use clear-by-write instead
3047 * of clear-by-read. Reading with EICS will return the
3048 * interrupt causes without clearing, which later be done
3049 * with the write to EICR.
3051 eicr
= IXGBE_READ_REG(hw
, IXGBE_EICS
);
3053 /* The lower 16bits of the EICR register are for the queue interrupts
3054 * which should be masked here in order to not accidentally clear them if
3055 * the bits are high when ixgbe_msix_other is called. There is a race
3056 * condition otherwise which results in possible performance loss
3057 * especially if the ixgbe_msix_other interrupt is triggering
3058 * consistently (as it would when PPS is turned on for the X540 device)
3062 IXGBE_WRITE_REG(hw
, IXGBE_EICR
, eicr
);
3064 if (eicr
& IXGBE_EICR_LSC
)
3065 ixgbe_check_lsc(adapter
);
3067 if (eicr
& IXGBE_EICR_MAILBOX
)
3068 ixgbe_msg_task(adapter
);
3070 switch (hw
->mac
.type
) {
3071 case ixgbe_mac_82599EB
:
3072 case ixgbe_mac_X540
:
3073 case ixgbe_mac_X550
:
3074 case ixgbe_mac_X550EM_x
:
3075 case ixgbe_mac_x550em_a
:
3076 if (hw
->phy
.type
== ixgbe_phy_x550em_ext_t
&&
3077 (eicr
& IXGBE_EICR_GPI_SDP0_X540
)) {
3078 adapter
->flags2
|= IXGBE_FLAG2_PHY_INTERRUPT
;
3079 ixgbe_service_event_schedule(adapter
);
3080 IXGBE_WRITE_REG(hw
, IXGBE_EICR
,
3081 IXGBE_EICR_GPI_SDP0_X540
);
3083 if (eicr
& IXGBE_EICR_ECC
) {
3084 e_info(link
, "Received ECC Err, initiating reset\n");
3085 set_bit(__IXGBE_RESET_REQUESTED
, &adapter
->state
);
3086 ixgbe_service_event_schedule(adapter
);
3087 IXGBE_WRITE_REG(hw
, IXGBE_EICR
, IXGBE_EICR_ECC
);
3089 /* Handle Flow Director Full threshold interrupt */
3090 if (eicr
& IXGBE_EICR_FLOW_DIR
) {
3091 int reinit_count
= 0;
3093 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
3094 struct ixgbe_ring
*ring
= adapter
->tx_ring
[i
];
3095 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE
,
3100 /* no more flow director interrupts until after init */
3101 IXGBE_WRITE_REG(hw
, IXGBE_EIMC
, IXGBE_EIMC_FLOW_DIR
);
3102 adapter
->flags2
|= IXGBE_FLAG2_FDIR_REQUIRES_REINIT
;
3103 ixgbe_service_event_schedule(adapter
);
3106 ixgbe_check_sfp_event(adapter
, eicr
);
3107 ixgbe_check_overtemp_event(adapter
, eicr
);
3113 ixgbe_check_fan_failure(adapter
, eicr
);
3115 if (unlikely(eicr
& IXGBE_EICR_TIMESYNC
))
3116 ixgbe_ptp_check_pps_event(adapter
);
3118 /* re-enable the original interrupt state, no lsc, no queues */
3119 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
))
3120 ixgbe_irq_enable(adapter
, false, false);
3125 static irqreturn_t
ixgbe_msix_clean_rings(int irq
, void *data
)
3127 struct ixgbe_q_vector
*q_vector
= data
;
3129 /* EIAM disabled interrupts (on this vector) for us */
3131 if (q_vector
->rx
.ring
|| q_vector
->tx
.ring
)
3132 napi_schedule_irqoff(&q_vector
->napi
);
3138 * ixgbe_poll - NAPI Rx polling callback
3139 * @napi: structure for representing this polling device
3140 * @budget: how many packets driver is allowed to clean
3142 * This function is used for legacy and MSI, NAPI mode
3144 int ixgbe_poll(struct napi_struct
*napi
, int budget
)
3146 struct ixgbe_q_vector
*q_vector
=
3147 container_of(napi
, struct ixgbe_q_vector
, napi
);
3148 struct ixgbe_adapter
*adapter
= q_vector
->adapter
;
3149 struct ixgbe_ring
*ring
;
3150 int per_ring_budget
, work_done
= 0;
3151 bool clean_complete
= true;
3153 #ifdef CONFIG_IXGBE_DCA
3154 if (adapter
->flags
& IXGBE_FLAG_DCA_ENABLED
)
3155 ixgbe_update_dca(q_vector
);
3158 ixgbe_for_each_ring(ring
, q_vector
->tx
) {
3159 bool wd
= ring
->xsk_umem
?
3160 ixgbe_clean_xdp_tx_irq(q_vector
, ring
, budget
) :
3161 ixgbe_clean_tx_irq(q_vector
, ring
, budget
);
3164 clean_complete
= false;
3167 /* Exit if we are called by netpoll */
3171 /* attempt to distribute budget to each queue fairly, but don't allow
3172 * the budget to go below 1 because we'll exit polling */
3173 if (q_vector
->rx
.count
> 1)
3174 per_ring_budget
= max(budget
/q_vector
->rx
.count
, 1);
3176 per_ring_budget
= budget
;
3178 ixgbe_for_each_ring(ring
, q_vector
->rx
) {
3179 int cleaned
= ring
->xsk_umem
?
3180 ixgbe_clean_rx_irq_zc(q_vector
, ring
,
3182 ixgbe_clean_rx_irq(q_vector
, ring
,
3185 work_done
+= cleaned
;
3186 if (cleaned
>= per_ring_budget
)
3187 clean_complete
= false;
3190 /* If all work not completed, return budget and keep polling */
3191 if (!clean_complete
)
3194 /* all work done, exit the polling mode */
3195 if (likely(napi_complete_done(napi
, work_done
))) {
3196 if (adapter
->rx_itr_setting
& 1)
3197 ixgbe_set_itr(q_vector
);
3198 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
))
3199 ixgbe_irq_enable_queues(adapter
,
3200 BIT_ULL(q_vector
->v_idx
));
3203 return min(work_done
, budget
- 1);
3207 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
3208 * @adapter: board private structure
3210 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
3211 * interrupts from the kernel.
3213 static int ixgbe_request_msix_irqs(struct ixgbe_adapter
*adapter
)
3215 struct net_device
*netdev
= adapter
->netdev
;
3216 unsigned int ri
= 0, ti
= 0;
3219 for (vector
= 0; vector
< adapter
->num_q_vectors
; vector
++) {
3220 struct ixgbe_q_vector
*q_vector
= adapter
->q_vector
[vector
];
3221 struct msix_entry
*entry
= &adapter
->msix_entries
[vector
];
3223 if (q_vector
->tx
.ring
&& q_vector
->rx
.ring
) {
3224 snprintf(q_vector
->name
, sizeof(q_vector
->name
),
3225 "%s-TxRx-%u", netdev
->name
, ri
++);
3227 } else if (q_vector
->rx
.ring
) {
3228 snprintf(q_vector
->name
, sizeof(q_vector
->name
),
3229 "%s-rx-%u", netdev
->name
, ri
++);
3230 } else if (q_vector
->tx
.ring
) {
3231 snprintf(q_vector
->name
, sizeof(q_vector
->name
),
3232 "%s-tx-%u", netdev
->name
, ti
++);
3234 /* skip this unused q_vector */
3237 err
= request_irq(entry
->vector
, &ixgbe_msix_clean_rings
, 0,
3238 q_vector
->name
, q_vector
);
3240 e_err(probe
, "request_irq failed for MSIX interrupt "
3241 "Error: %d\n", err
);
3242 goto free_queue_irqs
;
3244 /* If Flow Director is enabled, set interrupt affinity */
3245 if (adapter
->flags
& IXGBE_FLAG_FDIR_HASH_CAPABLE
) {
3246 /* assign the mask for this irq */
3247 irq_set_affinity_hint(entry
->vector
,
3248 &q_vector
->affinity_mask
);
3252 err
= request_irq(adapter
->msix_entries
[vector
].vector
,
3253 ixgbe_msix_other
, 0, netdev
->name
, adapter
);
3255 e_err(probe
, "request_irq for msix_other failed: %d\n", err
);
3256 goto free_queue_irqs
;
3264 irq_set_affinity_hint(adapter
->msix_entries
[vector
].vector
,
3266 free_irq(adapter
->msix_entries
[vector
].vector
,
3267 adapter
->q_vector
[vector
]);
3269 adapter
->flags
&= ~IXGBE_FLAG_MSIX_ENABLED
;
3270 pci_disable_msix(adapter
->pdev
);
3271 kfree(adapter
->msix_entries
);
3272 adapter
->msix_entries
= NULL
;
3277 * ixgbe_intr - legacy mode Interrupt Handler
3278 * @irq: interrupt number
3279 * @data: pointer to a network interface device structure
3281 static irqreturn_t
ixgbe_intr(int irq
, void *data
)
3283 struct ixgbe_adapter
*adapter
= data
;
3284 struct ixgbe_hw
*hw
= &adapter
->hw
;
3285 struct ixgbe_q_vector
*q_vector
= adapter
->q_vector
[0];
3289 * Workaround for silicon errata #26 on 82598. Mask the interrupt
3290 * before the read of EICR.
3292 IXGBE_WRITE_REG(hw
, IXGBE_EIMC
, IXGBE_IRQ_CLEAR_MASK
);
3294 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
3295 * therefore no explicit interrupt disable is necessary */
3296 eicr
= IXGBE_READ_REG(hw
, IXGBE_EICR
);
3299 * shared interrupt alert!
3300 * make sure interrupts are enabled because the read will
3301 * have disabled interrupts due to EIAM
3302 * finish the workaround of silicon errata on 82598. Unmask
3303 * the interrupt that we masked before the EICR read.
3305 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
))
3306 ixgbe_irq_enable(adapter
, true, true);
3307 return IRQ_NONE
; /* Not our interrupt */
3310 if (eicr
& IXGBE_EICR_LSC
)
3311 ixgbe_check_lsc(adapter
);
3313 switch (hw
->mac
.type
) {
3314 case ixgbe_mac_82599EB
:
3315 ixgbe_check_sfp_event(adapter
, eicr
);
3317 case ixgbe_mac_X540
:
3318 case ixgbe_mac_X550
:
3319 case ixgbe_mac_X550EM_x
:
3320 case ixgbe_mac_x550em_a
:
3321 if (eicr
& IXGBE_EICR_ECC
) {
3322 e_info(link
, "Received ECC Err, initiating reset\n");
3323 set_bit(__IXGBE_RESET_REQUESTED
, &adapter
->state
);
3324 ixgbe_service_event_schedule(adapter
);
3325 IXGBE_WRITE_REG(hw
, IXGBE_EICR
, IXGBE_EICR_ECC
);
3327 ixgbe_check_overtemp_event(adapter
, eicr
);
3333 ixgbe_check_fan_failure(adapter
, eicr
);
3334 if (unlikely(eicr
& IXGBE_EICR_TIMESYNC
))
3335 ixgbe_ptp_check_pps_event(adapter
);
3337 /* would disable interrupts here but EIAM disabled it */
3338 napi_schedule_irqoff(&q_vector
->napi
);
3341 * re-enable link(maybe) and non-queue interrupts, no flush.
3342 * ixgbe_poll will re-enable the queue interrupts
3344 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
))
3345 ixgbe_irq_enable(adapter
, false, false);
3351 * ixgbe_request_irq - initialize interrupts
3352 * @adapter: board private structure
3354 * Attempts to configure interrupts using the best available
3355 * capabilities of the hardware and kernel.
3357 static int ixgbe_request_irq(struct ixgbe_adapter
*adapter
)
3359 struct net_device
*netdev
= adapter
->netdev
;
3362 if (adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
)
3363 err
= ixgbe_request_msix_irqs(adapter
);
3364 else if (adapter
->flags
& IXGBE_FLAG_MSI_ENABLED
)
3365 err
= request_irq(adapter
->pdev
->irq
, ixgbe_intr
, 0,
3366 netdev
->name
, adapter
);
3368 err
= request_irq(adapter
->pdev
->irq
, ixgbe_intr
, IRQF_SHARED
,
3369 netdev
->name
, adapter
);
3372 e_err(probe
, "request_irq failed, Error %d\n", err
);
3377 static void ixgbe_free_irq(struct ixgbe_adapter
*adapter
)
3381 if (!(adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
)) {
3382 free_irq(adapter
->pdev
->irq
, adapter
);
3386 if (!adapter
->msix_entries
)
3389 for (vector
= 0; vector
< adapter
->num_q_vectors
; vector
++) {
3390 struct ixgbe_q_vector
*q_vector
= adapter
->q_vector
[vector
];
3391 struct msix_entry
*entry
= &adapter
->msix_entries
[vector
];
3393 /* free only the irqs that were actually requested */
3394 if (!q_vector
->rx
.ring
&& !q_vector
->tx
.ring
)
3397 /* clear the affinity_mask in the IRQ descriptor */
3398 irq_set_affinity_hint(entry
->vector
, NULL
);
3400 free_irq(entry
->vector
, q_vector
);
3403 free_irq(adapter
->msix_entries
[vector
].vector
, adapter
);
3407 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
3408 * @adapter: board private structure
3410 static inline void ixgbe_irq_disable(struct ixgbe_adapter
*adapter
)
3412 switch (adapter
->hw
.mac
.type
) {
3413 case ixgbe_mac_82598EB
:
3414 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EIMC
, ~0);
3416 case ixgbe_mac_82599EB
:
3417 case ixgbe_mac_X540
:
3418 case ixgbe_mac_X550
:
3419 case ixgbe_mac_X550EM_x
:
3420 case ixgbe_mac_x550em_a
:
3421 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EIMC
, 0xFFFF0000);
3422 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EIMC_EX(0), ~0);
3423 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EIMC_EX(1), ~0);
3428 IXGBE_WRITE_FLUSH(&adapter
->hw
);
3429 if (adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
) {
3432 for (vector
= 0; vector
< adapter
->num_q_vectors
; vector
++)
3433 synchronize_irq(adapter
->msix_entries
[vector
].vector
);
3435 synchronize_irq(adapter
->msix_entries
[vector
++].vector
);
3437 synchronize_irq(adapter
->pdev
->irq
);
3442 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
3443 * @adapter: board private structure
3446 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter
*adapter
)
3448 struct ixgbe_q_vector
*q_vector
= adapter
->q_vector
[0];
3450 ixgbe_write_eitr(q_vector
);
3452 ixgbe_set_ivar(adapter
, 0, 0, 0);
3453 ixgbe_set_ivar(adapter
, 1, 0, 0);
3455 e_info(hw
, "Legacy interrupt IVAR setup done\n");
3459 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
3460 * @adapter: board private structure
3461 * @ring: structure containing ring specific data
3463 * Configure the Tx descriptor ring after a reset.
3465 void ixgbe_configure_tx_ring(struct ixgbe_adapter
*adapter
,
3466 struct ixgbe_ring
*ring
)
3468 struct ixgbe_hw
*hw
= &adapter
->hw
;
3469 u64 tdba
= ring
->dma
;
3471 u32 txdctl
= IXGBE_TXDCTL_ENABLE
;
3472 u8 reg_idx
= ring
->reg_idx
;
3474 ring
->xsk_umem
= NULL
;
3475 if (ring_is_xdp(ring
))
3476 ring
->xsk_umem
= ixgbe_xsk_umem(adapter
, ring
);
3478 /* disable queue to avoid issues while updating state */
3479 IXGBE_WRITE_REG(hw
, IXGBE_TXDCTL(reg_idx
), 0);
3480 IXGBE_WRITE_FLUSH(hw
);
3482 IXGBE_WRITE_REG(hw
, IXGBE_TDBAL(reg_idx
),
3483 (tdba
& DMA_BIT_MASK(32)));
3484 IXGBE_WRITE_REG(hw
, IXGBE_TDBAH(reg_idx
), (tdba
>> 32));
3485 IXGBE_WRITE_REG(hw
, IXGBE_TDLEN(reg_idx
),
3486 ring
->count
* sizeof(union ixgbe_adv_tx_desc
));
3487 IXGBE_WRITE_REG(hw
, IXGBE_TDH(reg_idx
), 0);
3488 IXGBE_WRITE_REG(hw
, IXGBE_TDT(reg_idx
), 0);
3489 ring
->tail
= adapter
->io_addr
+ IXGBE_TDT(reg_idx
);
3492 * set WTHRESH to encourage burst writeback, it should not be set
3493 * higher than 1 when:
3494 * - ITR is 0 as it could cause false TX hangs
3495 * - ITR is set to > 100k int/sec and BQL is enabled
3497 * In order to avoid issues WTHRESH + PTHRESH should always be equal
3498 * to or less than the number of on chip descriptors, which is
3501 if (!ring
->q_vector
|| (ring
->q_vector
->itr
< IXGBE_100K_ITR
))
3502 txdctl
|= 1u << 16; /* WTHRESH = 1 */
3504 txdctl
|= 8u << 16; /* WTHRESH = 8 */
3507 * Setting PTHRESH to 32 both improves performance
3508 * and avoids a TX hang with DFP enabled
3510 txdctl
|= (1u << 8) | /* HTHRESH = 1 */
3511 32; /* PTHRESH = 32 */
3513 /* reinitialize flowdirector state */
3514 if (adapter
->flags
& IXGBE_FLAG_FDIR_HASH_CAPABLE
) {
3515 ring
->atr_sample_rate
= adapter
->atr_sample_rate
;
3516 ring
->atr_count
= 0;
3517 set_bit(__IXGBE_TX_FDIR_INIT_DONE
, &ring
->state
);
3519 ring
->atr_sample_rate
= 0;
3522 /* initialize XPS */
3523 if (!test_and_set_bit(__IXGBE_TX_XPS_INIT_DONE
, &ring
->state
)) {
3524 struct ixgbe_q_vector
*q_vector
= ring
->q_vector
;
3527 netif_set_xps_queue(ring
->netdev
,
3528 &q_vector
->affinity_mask
,
3532 clear_bit(__IXGBE_HANG_CHECK_ARMED
, &ring
->state
);
3534 /* reinitialize tx_buffer_info */
3535 memset(ring
->tx_buffer_info
, 0,
3536 sizeof(struct ixgbe_tx_buffer
) * ring
->count
);
3539 IXGBE_WRITE_REG(hw
, IXGBE_TXDCTL(reg_idx
), txdctl
);
3541 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3542 if (hw
->mac
.type
== ixgbe_mac_82598EB
&&
3543 !(IXGBE_READ_REG(hw
, IXGBE_LINKS
) & IXGBE_LINKS_UP
))
3546 /* poll to verify queue is enabled */
3548 usleep_range(1000, 2000);
3549 txdctl
= IXGBE_READ_REG(hw
, IXGBE_TXDCTL(reg_idx
));
3550 } while (--wait_loop
&& !(txdctl
& IXGBE_TXDCTL_ENABLE
));
3552 hw_dbg(hw
, "Could not enable Tx Queue %d\n", reg_idx
);
3555 static void ixgbe_setup_mtqc(struct ixgbe_adapter
*adapter
)
3557 struct ixgbe_hw
*hw
= &adapter
->hw
;
3559 u8 tcs
= adapter
->hw_tcs
;
3561 if (hw
->mac
.type
== ixgbe_mac_82598EB
)
3564 /* disable the arbiter while setting MTQC */
3565 rttdcs
= IXGBE_READ_REG(hw
, IXGBE_RTTDCS
);
3566 rttdcs
|= IXGBE_RTTDCS_ARBDIS
;
3567 IXGBE_WRITE_REG(hw
, IXGBE_RTTDCS
, rttdcs
);
3569 /* set transmit pool layout */
3570 if (adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
) {
3571 mtqc
= IXGBE_MTQC_VT_ENA
;
3573 mtqc
|= IXGBE_MTQC_RT_ENA
| IXGBE_MTQC_8TC_8TQ
;
3575 mtqc
|= IXGBE_MTQC_RT_ENA
| IXGBE_MTQC_4TC_4TQ
;
3576 else if (adapter
->ring_feature
[RING_F_VMDQ
].mask
==
3577 IXGBE_82599_VMDQ_4Q_MASK
)
3578 mtqc
|= IXGBE_MTQC_32VF
;
3580 mtqc
|= IXGBE_MTQC_64VF
;
3583 mtqc
= IXGBE_MTQC_RT_ENA
| IXGBE_MTQC_8TC_8TQ
;
3584 } else if (tcs
> 1) {
3585 mtqc
= IXGBE_MTQC_RT_ENA
| IXGBE_MTQC_4TC_4TQ
;
3587 u8 max_txq
= adapter
->num_tx_queues
+
3588 adapter
->num_xdp_queues
;
3590 mtqc
= IXGBE_MTQC_RT_ENA
| IXGBE_MTQC_4TC_4TQ
;
3592 mtqc
= IXGBE_MTQC_64Q_1PB
;
3596 IXGBE_WRITE_REG(hw
, IXGBE_MTQC
, mtqc
);
3598 /* Enable Security TX Buffer IFG for multiple pb */
3600 u32 sectx
= IXGBE_READ_REG(hw
, IXGBE_SECTXMINIFG
);
3601 sectx
|= IXGBE_SECTX_DCB
;
3602 IXGBE_WRITE_REG(hw
, IXGBE_SECTXMINIFG
, sectx
);
3605 /* re-enable the arbiter */
3606 rttdcs
&= ~IXGBE_RTTDCS_ARBDIS
;
3607 IXGBE_WRITE_REG(hw
, IXGBE_RTTDCS
, rttdcs
);
3611 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
3612 * @adapter: board private structure
3614 * Configure the Tx unit of the MAC after a reset.
3616 static void ixgbe_configure_tx(struct ixgbe_adapter
*adapter
)
3618 struct ixgbe_hw
*hw
= &adapter
->hw
;
3622 ixgbe_setup_mtqc(adapter
);
3624 if (hw
->mac
.type
!= ixgbe_mac_82598EB
) {
3625 /* DMATXCTL.EN must be before Tx queues are enabled */
3626 dmatxctl
= IXGBE_READ_REG(hw
, IXGBE_DMATXCTL
);
3627 dmatxctl
|= IXGBE_DMATXCTL_TE
;
3628 IXGBE_WRITE_REG(hw
, IXGBE_DMATXCTL
, dmatxctl
);
3631 /* Setup the HW Tx Head and Tail descriptor pointers */
3632 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
3633 ixgbe_configure_tx_ring(adapter
, adapter
->tx_ring
[i
]);
3634 for (i
= 0; i
< adapter
->num_xdp_queues
; i
++)
3635 ixgbe_configure_tx_ring(adapter
, adapter
->xdp_ring
[i
]);
3638 static void ixgbe_enable_rx_drop(struct ixgbe_adapter
*adapter
,
3639 struct ixgbe_ring
*ring
)
3641 struct ixgbe_hw
*hw
= &adapter
->hw
;
3642 u8 reg_idx
= ring
->reg_idx
;
3643 u32 srrctl
= IXGBE_READ_REG(hw
, IXGBE_SRRCTL(reg_idx
));
3645 srrctl
|= IXGBE_SRRCTL_DROP_EN
;
3647 IXGBE_WRITE_REG(hw
, IXGBE_SRRCTL(reg_idx
), srrctl
);
3650 static void ixgbe_disable_rx_drop(struct ixgbe_adapter
*adapter
,
3651 struct ixgbe_ring
*ring
)
3653 struct ixgbe_hw
*hw
= &adapter
->hw
;
3654 u8 reg_idx
= ring
->reg_idx
;
3655 u32 srrctl
= IXGBE_READ_REG(hw
, IXGBE_SRRCTL(reg_idx
));
3657 srrctl
&= ~IXGBE_SRRCTL_DROP_EN
;
3659 IXGBE_WRITE_REG(hw
, IXGBE_SRRCTL(reg_idx
), srrctl
);
3662 #ifdef CONFIG_IXGBE_DCB
3663 void ixgbe_set_rx_drop_en(struct ixgbe_adapter
*adapter
)
3665 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter
*adapter
)
3669 bool pfc_en
= adapter
->dcb_cfg
.pfc_mode_enable
;
3671 if (adapter
->ixgbe_ieee_pfc
)
3672 pfc_en
|= !!(adapter
->ixgbe_ieee_pfc
->pfc_en
);
3675 * We should set the drop enable bit if:
3678 * Number of Rx queues > 1 and flow control is disabled
3680 * This allows us to avoid head of line blocking for security
3681 * and performance reasons.
3683 if (adapter
->num_vfs
|| (adapter
->num_rx_queues
> 1 &&
3684 !(adapter
->hw
.fc
.current_mode
& ixgbe_fc_tx_pause
) && !pfc_en
)) {
3685 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
3686 ixgbe_enable_rx_drop(adapter
, adapter
->rx_ring
[i
]);
3688 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
3689 ixgbe_disable_rx_drop(adapter
, adapter
->rx_ring
[i
]);
3693 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
3695 static void ixgbe_configure_srrctl(struct ixgbe_adapter
*adapter
,
3696 struct ixgbe_ring
*rx_ring
)
3698 struct ixgbe_hw
*hw
= &adapter
->hw
;
3700 u8 reg_idx
= rx_ring
->reg_idx
;
3702 if (hw
->mac
.type
== ixgbe_mac_82598EB
) {
3703 u16 mask
= adapter
->ring_feature
[RING_F_RSS
].mask
;
3706 * if VMDq is not active we must program one srrctl register
3707 * per RSS queue since we have enabled RDRXCTL.MVMEN
3712 /* configure header buffer length, needed for RSC */
3713 srrctl
= IXGBE_RX_HDR_SIZE
<< IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT
;
3715 /* configure the packet buffer length */
3716 if (rx_ring
->xsk_umem
) {
3717 u32 xsk_buf_len
= xsk_umem_get_rx_frame_size(rx_ring
->xsk_umem
);
3719 /* If the MAC support setting RXDCTL.RLPML, the
3720 * SRRCTL[n].BSIZEPKT is set to PAGE_SIZE and
3721 * RXDCTL.RLPML is set to the actual UMEM buffer
3722 * size. If not, then we are stuck with a 1k buffer
3723 * size resolution. In this case frames larger than
3724 * the UMEM buffer size viewed in a 1k resolution will
3727 if (hw
->mac
.type
!= ixgbe_mac_82599EB
)
3728 srrctl
|= PAGE_SIZE
>> IXGBE_SRRCTL_BSIZEPKT_SHIFT
;
3730 srrctl
|= xsk_buf_len
>> IXGBE_SRRCTL_BSIZEPKT_SHIFT
;
3731 } else if (test_bit(__IXGBE_RX_3K_BUFFER
, &rx_ring
->state
)) {
3732 srrctl
|= IXGBE_RXBUFFER_3K
>> IXGBE_SRRCTL_BSIZEPKT_SHIFT
;
3734 srrctl
|= IXGBE_RXBUFFER_2K
>> IXGBE_SRRCTL_BSIZEPKT_SHIFT
;
3737 /* configure descriptor type */
3738 srrctl
|= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF
;
3740 IXGBE_WRITE_REG(hw
, IXGBE_SRRCTL(reg_idx
), srrctl
);
3744 * ixgbe_rss_indir_tbl_entries - Return RSS indirection table entries
3745 * @adapter: device handle
3747 * - 82598/82599/X540: 128
3748 * - X550(non-SRIOV mode): 512
3749 * - X550(SRIOV mode): 64
3751 u32
ixgbe_rss_indir_tbl_entries(struct ixgbe_adapter
*adapter
)
3753 if (adapter
->hw
.mac
.type
< ixgbe_mac_X550
)
3755 else if (adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
)
3762 * ixgbe_store_key - Write the RSS key to HW
3763 * @adapter: device handle
3765 * Write the RSS key stored in adapter.rss_key to HW.
3767 void ixgbe_store_key(struct ixgbe_adapter
*adapter
)
3769 struct ixgbe_hw
*hw
= &adapter
->hw
;
3772 for (i
= 0; i
< 10; i
++)
3773 IXGBE_WRITE_REG(hw
, IXGBE_RSSRK(i
), adapter
->rss_key
[i
]);
3777 * ixgbe_init_rss_key - Initialize adapter RSS key
3778 * @adapter: device handle
3780 * Allocates and initializes the RSS key if it is not allocated.
3782 static inline int ixgbe_init_rss_key(struct ixgbe_adapter
*adapter
)
3786 if (!adapter
->rss_key
) {
3787 rss_key
= kzalloc(IXGBE_RSS_KEY_SIZE
, GFP_KERNEL
);
3788 if (unlikely(!rss_key
))
3791 netdev_rss_key_fill(rss_key
, IXGBE_RSS_KEY_SIZE
);
3792 adapter
->rss_key
= rss_key
;
3799 * ixgbe_store_reta - Write the RETA table to HW
3800 * @adapter: device handle
3802 * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3804 void ixgbe_store_reta(struct ixgbe_adapter
*adapter
)
3806 u32 i
, reta_entries
= ixgbe_rss_indir_tbl_entries(adapter
);
3807 struct ixgbe_hw
*hw
= &adapter
->hw
;
3810 u8
*indir_tbl
= adapter
->rss_indir_tbl
;
3812 /* Fill out the redirection table as follows:
3813 * - 82598: 8 bit wide entries containing pair of 4 bit RSS
3815 * - 82599/X540: 8 bit wide entries containing 4 bit RSS index
3816 * - X550: 8 bit wide entries containing 6 bit RSS index
3818 if (adapter
->hw
.mac
.type
== ixgbe_mac_82598EB
)
3819 indices_multi
= 0x11;
3821 indices_multi
= 0x1;
3823 /* Write redirection table to HW */
3824 for (i
= 0; i
< reta_entries
; i
++) {
3825 reta
|= indices_multi
* indir_tbl
[i
] << (i
& 0x3) * 8;
3828 IXGBE_WRITE_REG(hw
, IXGBE_RETA(i
>> 2), reta
);
3830 IXGBE_WRITE_REG(hw
, IXGBE_ERETA((i
>> 2) - 32),
3838 * ixgbe_store_vfreta - Write the RETA table to HW (x550 devices in SRIOV mode)
3839 * @adapter: device handle
3841 * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3843 static void ixgbe_store_vfreta(struct ixgbe_adapter
*adapter
)
3845 u32 i
, reta_entries
= ixgbe_rss_indir_tbl_entries(adapter
);
3846 struct ixgbe_hw
*hw
= &adapter
->hw
;
3849 /* Write redirection table to HW */
3850 for (i
= 0; i
< reta_entries
; i
++) {
3851 u16 pool
= adapter
->num_rx_pools
;
3853 vfreta
|= (u32
)adapter
->rss_indir_tbl
[i
] << (i
& 0x3) * 8;
3859 IXGBE_PFVFRETA(i
>> 2, VMDQ_P(pool
)),
3865 static void ixgbe_setup_reta(struct ixgbe_adapter
*adapter
)
3868 u32 reta_entries
= ixgbe_rss_indir_tbl_entries(adapter
);
3869 u16 rss_i
= adapter
->ring_feature
[RING_F_RSS
].indices
;
3871 /* Program table for at least 4 queues w/ SR-IOV so that VFs can
3872 * make full use of any rings they may have. We will use the
3873 * PSRTYPE register to control how many rings we use within the PF.
3875 if ((adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
) && (rss_i
< 4))
3878 /* Fill out hash function seeds */
3879 ixgbe_store_key(adapter
);
3881 /* Fill out redirection table */
3882 memset(adapter
->rss_indir_tbl
, 0, sizeof(adapter
->rss_indir_tbl
));
3884 for (i
= 0, j
= 0; i
< reta_entries
; i
++, j
++) {
3888 adapter
->rss_indir_tbl
[i
] = j
;
3891 ixgbe_store_reta(adapter
);
3894 static void ixgbe_setup_vfreta(struct ixgbe_adapter
*adapter
)
3896 struct ixgbe_hw
*hw
= &adapter
->hw
;
3897 u16 rss_i
= adapter
->ring_feature
[RING_F_RSS
].indices
;
3900 /* Fill out hash function seeds */
3901 for (i
= 0; i
< 10; i
++) {
3902 u16 pool
= adapter
->num_rx_pools
;
3906 IXGBE_PFVFRSSRK(i
, VMDQ_P(pool
)),
3907 *(adapter
->rss_key
+ i
));
3910 /* Fill out the redirection table */
3911 for (i
= 0, j
= 0; i
< 64; i
++, j
++) {
3915 adapter
->rss_indir_tbl
[i
] = j
;
3918 ixgbe_store_vfreta(adapter
);
3921 static void ixgbe_setup_mrqc(struct ixgbe_adapter
*adapter
)
3923 struct ixgbe_hw
*hw
= &adapter
->hw
;
3924 u32 mrqc
= 0, rss_field
= 0, vfmrqc
= 0;
3927 /* Disable indicating checksum in descriptor, enables RSS hash */
3928 rxcsum
= IXGBE_READ_REG(hw
, IXGBE_RXCSUM
);
3929 rxcsum
|= IXGBE_RXCSUM_PCSD
;
3930 IXGBE_WRITE_REG(hw
, IXGBE_RXCSUM
, rxcsum
);
3932 if (adapter
->hw
.mac
.type
== ixgbe_mac_82598EB
) {
3933 if (adapter
->ring_feature
[RING_F_RSS
].mask
)
3934 mrqc
= IXGBE_MRQC_RSSEN
;
3936 u8 tcs
= adapter
->hw_tcs
;
3938 if (adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
) {
3940 mrqc
= IXGBE_MRQC_VMDQRT8TCEN
; /* 8 TCs */
3942 mrqc
= IXGBE_MRQC_VMDQRT4TCEN
; /* 4 TCs */
3943 else if (adapter
->ring_feature
[RING_F_VMDQ
].mask
==
3944 IXGBE_82599_VMDQ_4Q_MASK
)
3945 mrqc
= IXGBE_MRQC_VMDQRSS32EN
;
3947 mrqc
= IXGBE_MRQC_VMDQRSS64EN
;
3949 /* Enable L3/L4 for Tx Switched packets only for X550,
3950 * older devices do not support this feature
3952 if (hw
->mac
.type
>= ixgbe_mac_X550
)
3953 mrqc
|= IXGBE_MRQC_L3L4TXSWEN
;
3956 mrqc
= IXGBE_MRQC_RTRSS8TCEN
;
3958 mrqc
= IXGBE_MRQC_RTRSS4TCEN
;
3960 mrqc
= IXGBE_MRQC_RSSEN
;
3964 /* Perform hash on these packet types */
3965 rss_field
|= IXGBE_MRQC_RSS_FIELD_IPV4
|
3966 IXGBE_MRQC_RSS_FIELD_IPV4_TCP
|
3967 IXGBE_MRQC_RSS_FIELD_IPV6
|
3968 IXGBE_MRQC_RSS_FIELD_IPV6_TCP
;
3970 if (adapter
->flags2
& IXGBE_FLAG2_RSS_FIELD_IPV4_UDP
)
3971 rss_field
|= IXGBE_MRQC_RSS_FIELD_IPV4_UDP
;
3972 if (adapter
->flags2
& IXGBE_FLAG2_RSS_FIELD_IPV6_UDP
)
3973 rss_field
|= IXGBE_MRQC_RSS_FIELD_IPV6_UDP
;
3975 if ((hw
->mac
.type
>= ixgbe_mac_X550
) &&
3976 (adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
)) {
3977 u16 pool
= adapter
->num_rx_pools
;
3979 /* Enable VF RSS mode */
3980 mrqc
|= IXGBE_MRQC_MULTIPLE_RSS
;
3981 IXGBE_WRITE_REG(hw
, IXGBE_MRQC
, mrqc
);
3983 /* Setup RSS through the VF registers */
3984 ixgbe_setup_vfreta(adapter
);
3985 vfmrqc
= IXGBE_MRQC_RSSEN
;
3986 vfmrqc
|= rss_field
;
3990 IXGBE_PFVFMRQC(VMDQ_P(pool
)),
3993 ixgbe_setup_reta(adapter
);
3995 IXGBE_WRITE_REG(hw
, IXGBE_MRQC
, mrqc
);
4000 * ixgbe_configure_rscctl - enable RSC for the indicated ring
4001 * @adapter: address of board private structure
4002 * @ring: structure containing ring specific data
4004 static void ixgbe_configure_rscctl(struct ixgbe_adapter
*adapter
,
4005 struct ixgbe_ring
*ring
)
4007 struct ixgbe_hw
*hw
= &adapter
->hw
;
4009 u8 reg_idx
= ring
->reg_idx
;
4011 if (!ring_is_rsc_enabled(ring
))
4014 rscctrl
= IXGBE_READ_REG(hw
, IXGBE_RSCCTL(reg_idx
));
4015 rscctrl
|= IXGBE_RSCCTL_RSCEN
;
4017 * we must limit the number of descriptors so that the
4018 * total size of max desc * buf_len is not greater
4021 rscctrl
|= IXGBE_RSCCTL_MAXDESC_16
;
4022 IXGBE_WRITE_REG(hw
, IXGBE_RSCCTL(reg_idx
), rscctrl
);
4025 #define IXGBE_MAX_RX_DESC_POLL 10
4026 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter
*adapter
,
4027 struct ixgbe_ring
*ring
)
4029 struct ixgbe_hw
*hw
= &adapter
->hw
;
4030 int wait_loop
= IXGBE_MAX_RX_DESC_POLL
;
4032 u8 reg_idx
= ring
->reg_idx
;
4034 if (ixgbe_removed(hw
->hw_addr
))
4036 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
4037 if (hw
->mac
.type
== ixgbe_mac_82598EB
&&
4038 !(IXGBE_READ_REG(hw
, IXGBE_LINKS
) & IXGBE_LINKS_UP
))
4042 usleep_range(1000, 2000);
4043 rxdctl
= IXGBE_READ_REG(hw
, IXGBE_RXDCTL(reg_idx
));
4044 } while (--wait_loop
&& !(rxdctl
& IXGBE_RXDCTL_ENABLE
));
4047 e_err(drv
, "RXDCTL.ENABLE on Rx queue %d not set within "
4048 "the polling period\n", reg_idx
);
4052 void ixgbe_configure_rx_ring(struct ixgbe_adapter
*adapter
,
4053 struct ixgbe_ring
*ring
)
4055 struct ixgbe_hw
*hw
= &adapter
->hw
;
4056 union ixgbe_adv_rx_desc
*rx_desc
;
4057 u64 rdba
= ring
->dma
;
4059 u8 reg_idx
= ring
->reg_idx
;
4061 xdp_rxq_info_unreg_mem_model(&ring
->xdp_rxq
);
4062 ring
->xsk_umem
= ixgbe_xsk_umem(adapter
, ring
);
4063 if (ring
->xsk_umem
) {
4064 WARN_ON(xdp_rxq_info_reg_mem_model(&ring
->xdp_rxq
,
4065 MEM_TYPE_XSK_BUFF_POOL
,
4067 xsk_buff_set_rxq_info(ring
->xsk_umem
, &ring
->xdp_rxq
);
4069 WARN_ON(xdp_rxq_info_reg_mem_model(&ring
->xdp_rxq
,
4070 MEM_TYPE_PAGE_SHARED
, NULL
));
4073 /* disable queue to avoid use of these values while updating state */
4074 rxdctl
= IXGBE_READ_REG(hw
, IXGBE_RXDCTL(reg_idx
));
4075 rxdctl
&= ~IXGBE_RXDCTL_ENABLE
;
4077 /* write value back with RXDCTL.ENABLE bit cleared */
4078 IXGBE_WRITE_REG(hw
, IXGBE_RXDCTL(reg_idx
), rxdctl
);
4079 IXGBE_WRITE_FLUSH(hw
);
4081 IXGBE_WRITE_REG(hw
, IXGBE_RDBAL(reg_idx
), (rdba
& DMA_BIT_MASK(32)));
4082 IXGBE_WRITE_REG(hw
, IXGBE_RDBAH(reg_idx
), (rdba
>> 32));
4083 IXGBE_WRITE_REG(hw
, IXGBE_RDLEN(reg_idx
),
4084 ring
->count
* sizeof(union ixgbe_adv_rx_desc
));
4085 /* Force flushing of IXGBE_RDLEN to prevent MDD */
4086 IXGBE_WRITE_FLUSH(hw
);
4088 IXGBE_WRITE_REG(hw
, IXGBE_RDH(reg_idx
), 0);
4089 IXGBE_WRITE_REG(hw
, IXGBE_RDT(reg_idx
), 0);
4090 ring
->tail
= adapter
->io_addr
+ IXGBE_RDT(reg_idx
);
4092 ixgbe_configure_srrctl(adapter
, ring
);
4093 ixgbe_configure_rscctl(adapter
, ring
);
4095 if (hw
->mac
.type
== ixgbe_mac_82598EB
) {
4097 * enable cache line friendly hardware writes:
4098 * PTHRESH=32 descriptors (half the internal cache),
4099 * this also removes ugly rx_no_buffer_count increment
4100 * HTHRESH=4 descriptors (to minimize latency on fetch)
4101 * WTHRESH=8 burst writeback up to two cache lines
4103 rxdctl
&= ~0x3FFFFF;
4105 #if (PAGE_SIZE < 8192)
4106 /* RXDCTL.RLPML does not work on 82599 */
4107 } else if (hw
->mac
.type
!= ixgbe_mac_82599EB
) {
4108 rxdctl
&= ~(IXGBE_RXDCTL_RLPMLMASK
|
4109 IXGBE_RXDCTL_RLPML_EN
);
4111 /* Limit the maximum frame size so we don't overrun the skb.
4112 * This can happen in SRIOV mode when the MTU of the VF is
4113 * higher than the MTU of the PF.
4115 if (ring_uses_build_skb(ring
) &&
4116 !test_bit(__IXGBE_RX_3K_BUFFER
, &ring
->state
))
4117 rxdctl
|= IXGBE_MAX_2K_FRAME_BUILD_SKB
|
4118 IXGBE_RXDCTL_RLPML_EN
;
4122 if (ring
->xsk_umem
&& hw
->mac
.type
!= ixgbe_mac_82599EB
) {
4123 u32 xsk_buf_len
= xsk_umem_get_rx_frame_size(ring
->xsk_umem
);
4125 rxdctl
&= ~(IXGBE_RXDCTL_RLPMLMASK
|
4126 IXGBE_RXDCTL_RLPML_EN
);
4127 rxdctl
|= xsk_buf_len
| IXGBE_RXDCTL_RLPML_EN
;
4129 ring
->rx_buf_len
= xsk_buf_len
;
4132 /* initialize rx_buffer_info */
4133 memset(ring
->rx_buffer_info
, 0,
4134 sizeof(struct ixgbe_rx_buffer
) * ring
->count
);
4136 /* initialize Rx descriptor 0 */
4137 rx_desc
= IXGBE_RX_DESC(ring
, 0);
4138 rx_desc
->wb
.upper
.length
= 0;
4140 /* enable receive descriptor ring */
4141 rxdctl
|= IXGBE_RXDCTL_ENABLE
;
4142 IXGBE_WRITE_REG(hw
, IXGBE_RXDCTL(reg_idx
), rxdctl
);
4144 ixgbe_rx_desc_queue_enable(adapter
, ring
);
4146 ixgbe_alloc_rx_buffers_zc(ring
, ixgbe_desc_unused(ring
));
4148 ixgbe_alloc_rx_buffers(ring
, ixgbe_desc_unused(ring
));
4151 static void ixgbe_setup_psrtype(struct ixgbe_adapter
*adapter
)
4153 struct ixgbe_hw
*hw
= &adapter
->hw
;
4154 int rss_i
= adapter
->ring_feature
[RING_F_RSS
].indices
;
4155 u16 pool
= adapter
->num_rx_pools
;
4157 /* PSRTYPE must be initialized in non 82598 adapters */
4158 u32 psrtype
= IXGBE_PSRTYPE_TCPHDR
|
4159 IXGBE_PSRTYPE_UDPHDR
|
4160 IXGBE_PSRTYPE_IPV4HDR
|
4161 IXGBE_PSRTYPE_L2HDR
|
4162 IXGBE_PSRTYPE_IPV6HDR
;
4164 if (hw
->mac
.type
== ixgbe_mac_82598EB
)
4168 psrtype
|= 2u << 29;
4170 psrtype
|= 1u << 29;
4173 IXGBE_WRITE_REG(hw
, IXGBE_PSRTYPE(VMDQ_P(pool
)), psrtype
);
4176 static void ixgbe_configure_virtualization(struct ixgbe_adapter
*adapter
)
4178 struct ixgbe_hw
*hw
= &adapter
->hw
;
4179 u16 pool
= adapter
->num_rx_pools
;
4180 u32 reg_offset
, vf_shift
, vmolr
;
4181 u32 gcr_ext
, vmdctl
;
4184 if (!(adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
))
4187 vmdctl
= IXGBE_READ_REG(hw
, IXGBE_VT_CTL
);
4188 vmdctl
|= IXGBE_VMD_CTL_VMDQ_EN
;
4189 vmdctl
&= ~IXGBE_VT_CTL_POOL_MASK
;
4190 vmdctl
|= VMDQ_P(0) << IXGBE_VT_CTL_POOL_SHIFT
;
4191 vmdctl
|= IXGBE_VT_CTL_REPLEN
;
4192 IXGBE_WRITE_REG(hw
, IXGBE_VT_CTL
, vmdctl
);
4194 /* accept untagged packets until a vlan tag is
4195 * specifically set for the VMDQ queue/pool
4197 vmolr
= IXGBE_VMOLR_AUPE
;
4199 IXGBE_WRITE_REG(hw
, IXGBE_VMOLR(VMDQ_P(pool
)), vmolr
);
4201 vf_shift
= VMDQ_P(0) % 32;
4202 reg_offset
= (VMDQ_P(0) >= 32) ? 1 : 0;
4204 /* Enable only the PF's pool for Tx/Rx */
4205 IXGBE_WRITE_REG(hw
, IXGBE_VFRE(reg_offset
), GENMASK(31, vf_shift
));
4206 IXGBE_WRITE_REG(hw
, IXGBE_VFRE(reg_offset
^ 1), reg_offset
- 1);
4207 IXGBE_WRITE_REG(hw
, IXGBE_VFTE(reg_offset
), GENMASK(31, vf_shift
));
4208 IXGBE_WRITE_REG(hw
, IXGBE_VFTE(reg_offset
^ 1), reg_offset
- 1);
4209 if (adapter
->bridge_mode
== BRIDGE_MODE_VEB
)
4210 IXGBE_WRITE_REG(hw
, IXGBE_PFDTXGSWC
, IXGBE_PFDTXGSWC_VT_LBEN
);
4212 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
4213 hw
->mac
.ops
.set_vmdq(hw
, 0, VMDQ_P(0));
4215 /* clear VLAN promisc flag so VFTA will be updated if necessary */
4216 adapter
->flags2
&= ~IXGBE_FLAG2_VLAN_PROMISC
;
4219 * Set up VF register offsets for selected VT Mode,
4220 * i.e. 32 or 64 VFs for SR-IOV
4222 switch (adapter
->ring_feature
[RING_F_VMDQ
].mask
) {
4223 case IXGBE_82599_VMDQ_8Q_MASK
:
4224 gcr_ext
= IXGBE_GCR_EXT_VT_MODE_16
;
4226 case IXGBE_82599_VMDQ_4Q_MASK
:
4227 gcr_ext
= IXGBE_GCR_EXT_VT_MODE_32
;
4230 gcr_ext
= IXGBE_GCR_EXT_VT_MODE_64
;
4234 IXGBE_WRITE_REG(hw
, IXGBE_GCR_EXT
, gcr_ext
);
4236 for (i
= 0; i
< adapter
->num_vfs
; i
++) {
4237 /* configure spoof checking */
4238 ixgbe_ndo_set_vf_spoofchk(adapter
->netdev
, i
,
4239 adapter
->vfinfo
[i
].spoofchk_enabled
);
4241 /* Enable/Disable RSS query feature */
4242 ixgbe_ndo_set_vf_rss_query_en(adapter
->netdev
, i
,
4243 adapter
->vfinfo
[i
].rss_query_enabled
);
4247 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter
*adapter
)
4249 struct ixgbe_hw
*hw
= &adapter
->hw
;
4250 struct net_device
*netdev
= adapter
->netdev
;
4251 int max_frame
= netdev
->mtu
+ ETH_HLEN
+ ETH_FCS_LEN
;
4252 struct ixgbe_ring
*rx_ring
;
4257 /* adjust max frame to be able to do baby jumbo for FCoE */
4258 if ((adapter
->flags
& IXGBE_FLAG_FCOE_ENABLED
) &&
4259 (max_frame
< IXGBE_FCOE_JUMBO_FRAME_SIZE
))
4260 max_frame
= IXGBE_FCOE_JUMBO_FRAME_SIZE
;
4262 #endif /* IXGBE_FCOE */
4264 /* adjust max frame to be at least the size of a standard frame */
4265 if (max_frame
< (ETH_FRAME_LEN
+ ETH_FCS_LEN
))
4266 max_frame
= (ETH_FRAME_LEN
+ ETH_FCS_LEN
);
4268 mhadd
= IXGBE_READ_REG(hw
, IXGBE_MHADD
);
4269 if (max_frame
!= (mhadd
>> IXGBE_MHADD_MFS_SHIFT
)) {
4270 mhadd
&= ~IXGBE_MHADD_MFS_MASK
;
4271 mhadd
|= max_frame
<< IXGBE_MHADD_MFS_SHIFT
;
4273 IXGBE_WRITE_REG(hw
, IXGBE_MHADD
, mhadd
);
4276 hlreg0
= IXGBE_READ_REG(hw
, IXGBE_HLREG0
);
4277 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
4278 hlreg0
|= IXGBE_HLREG0_JUMBOEN
;
4279 IXGBE_WRITE_REG(hw
, IXGBE_HLREG0
, hlreg0
);
4282 * Setup the HW Rx Head and Tail Descriptor Pointers and
4283 * the Base and Length of the Rx Descriptor Ring
4285 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
4286 rx_ring
= adapter
->rx_ring
[i
];
4288 clear_ring_rsc_enabled(rx_ring
);
4289 clear_bit(__IXGBE_RX_3K_BUFFER
, &rx_ring
->state
);
4290 clear_bit(__IXGBE_RX_BUILD_SKB_ENABLED
, &rx_ring
->state
);
4292 if (adapter
->flags2
& IXGBE_FLAG2_RSC_ENABLED
)
4293 set_ring_rsc_enabled(rx_ring
);
4295 if (test_bit(__IXGBE_RX_FCOE
, &rx_ring
->state
))
4296 set_bit(__IXGBE_RX_3K_BUFFER
, &rx_ring
->state
);
4298 if (adapter
->flags2
& IXGBE_FLAG2_RX_LEGACY
)
4301 set_bit(__IXGBE_RX_BUILD_SKB_ENABLED
, &rx_ring
->state
);
4303 #if (PAGE_SIZE < 8192)
4304 if (adapter
->flags2
& IXGBE_FLAG2_RSC_ENABLED
)
4305 set_bit(__IXGBE_RX_3K_BUFFER
, &rx_ring
->state
);
4307 if (IXGBE_2K_TOO_SMALL_WITH_PADDING
||
4308 (max_frame
> (ETH_FRAME_LEN
+ ETH_FCS_LEN
)))
4309 set_bit(__IXGBE_RX_3K_BUFFER
, &rx_ring
->state
);
4314 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter
*adapter
)
4316 struct ixgbe_hw
*hw
= &adapter
->hw
;
4317 u32 rdrxctl
= IXGBE_READ_REG(hw
, IXGBE_RDRXCTL
);
4319 switch (hw
->mac
.type
) {
4320 case ixgbe_mac_82598EB
:
4322 * For VMDq support of different descriptor types or
4323 * buffer sizes through the use of multiple SRRCTL
4324 * registers, RDRXCTL.MVMEN must be set to 1
4326 * also, the manual doesn't mention it clearly but DCA hints
4327 * will only use queue 0's tags unless this bit is set. Side
4328 * effects of setting this bit are only that SRRCTL must be
4329 * fully programmed [0..15]
4331 rdrxctl
|= IXGBE_RDRXCTL_MVMEN
;
4333 case ixgbe_mac_X550
:
4334 case ixgbe_mac_X550EM_x
:
4335 case ixgbe_mac_x550em_a
:
4336 if (adapter
->num_vfs
)
4337 rdrxctl
|= IXGBE_RDRXCTL_PSP
;
4339 case ixgbe_mac_82599EB
:
4340 case ixgbe_mac_X540
:
4341 /* Disable RSC for ACK packets */
4342 IXGBE_WRITE_REG(hw
, IXGBE_RSCDBU
,
4343 (IXGBE_RSCDBU_RSCACKDIS
| IXGBE_READ_REG(hw
, IXGBE_RSCDBU
)));
4344 rdrxctl
&= ~IXGBE_RDRXCTL_RSCFRSTSIZE
;
4345 /* hardware requires some bits to be set by default */
4346 rdrxctl
|= (IXGBE_RDRXCTL_RSCACKC
| IXGBE_RDRXCTL_FCOE_WRFIX
);
4347 rdrxctl
|= IXGBE_RDRXCTL_CRCSTRIP
;
4350 /* We should do nothing since we don't know this hardware */
4354 IXGBE_WRITE_REG(hw
, IXGBE_RDRXCTL
, rdrxctl
);
4358 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
4359 * @adapter: board private structure
4361 * Configure the Rx unit of the MAC after a reset.
4363 static void ixgbe_configure_rx(struct ixgbe_adapter
*adapter
)
4365 struct ixgbe_hw
*hw
= &adapter
->hw
;
4369 /* disable receives while setting up the descriptors */
4370 hw
->mac
.ops
.disable_rx(hw
);
4372 ixgbe_setup_psrtype(adapter
);
4373 ixgbe_setup_rdrxctl(adapter
);
4376 rfctl
= IXGBE_READ_REG(hw
, IXGBE_RFCTL
);
4377 rfctl
&= ~IXGBE_RFCTL_RSC_DIS
;
4378 if (!(adapter
->flags2
& IXGBE_FLAG2_RSC_ENABLED
))
4379 rfctl
|= IXGBE_RFCTL_RSC_DIS
;
4381 /* disable NFS filtering */
4382 rfctl
|= (IXGBE_RFCTL_NFSW_DIS
| IXGBE_RFCTL_NFSR_DIS
);
4383 IXGBE_WRITE_REG(hw
, IXGBE_RFCTL
, rfctl
);
4385 /* Program registers for the distribution of queues */
4386 ixgbe_setup_mrqc(adapter
);
4388 /* set_rx_buffer_len must be called before ring initialization */
4389 ixgbe_set_rx_buffer_len(adapter
);
4392 * Setup the HW Rx Head and Tail Descriptor Pointers and
4393 * the Base and Length of the Rx Descriptor Ring
4395 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
4396 ixgbe_configure_rx_ring(adapter
, adapter
->rx_ring
[i
]);
4398 rxctrl
= IXGBE_READ_REG(hw
, IXGBE_RXCTRL
);
4399 /* disable drop enable for 82598 parts */
4400 if (hw
->mac
.type
== ixgbe_mac_82598EB
)
4401 rxctrl
|= IXGBE_RXCTRL_DMBYPS
;
4403 /* enable all receives */
4404 rxctrl
|= IXGBE_RXCTRL_RXEN
;
4405 hw
->mac
.ops
.enable_rx_dma(hw
, rxctrl
);
4408 static int ixgbe_vlan_rx_add_vid(struct net_device
*netdev
,
4409 __be16 proto
, u16 vid
)
4411 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
4412 struct ixgbe_hw
*hw
= &adapter
->hw
;
4414 /* add VID to filter table */
4415 if (!vid
|| !(adapter
->flags2
& IXGBE_FLAG2_VLAN_PROMISC
))
4416 hw
->mac
.ops
.set_vfta(&adapter
->hw
, vid
, VMDQ_P(0), true, !!vid
);
4418 set_bit(vid
, adapter
->active_vlans
);
4423 static int ixgbe_find_vlvf_entry(struct ixgbe_hw
*hw
, u32 vlan
)
4428 /* short cut the special case */
4432 /* Search for the vlan id in the VLVF entries */
4433 for (idx
= IXGBE_VLVF_ENTRIES
; --idx
;) {
4434 vlvf
= IXGBE_READ_REG(hw
, IXGBE_VLVF(idx
));
4435 if ((vlvf
& VLAN_VID_MASK
) == vlan
)
4442 void ixgbe_update_pf_promisc_vlvf(struct ixgbe_adapter
*adapter
, u32 vid
)
4444 struct ixgbe_hw
*hw
= &adapter
->hw
;
4448 idx
= ixgbe_find_vlvf_entry(hw
, vid
);
4452 /* See if any other pools are set for this VLAN filter
4453 * entry other than the PF.
4455 word
= idx
* 2 + (VMDQ_P(0) / 32);
4456 bits
= ~BIT(VMDQ_P(0) % 32);
4457 bits
&= IXGBE_READ_REG(hw
, IXGBE_VLVFB(word
));
4459 /* Disable the filter so this falls into the default pool. */
4460 if (!bits
&& !IXGBE_READ_REG(hw
, IXGBE_VLVFB(word
^ 1))) {
4461 if (!(adapter
->flags2
& IXGBE_FLAG2_VLAN_PROMISC
))
4462 IXGBE_WRITE_REG(hw
, IXGBE_VLVFB(word
), 0);
4463 IXGBE_WRITE_REG(hw
, IXGBE_VLVF(idx
), 0);
4467 static int ixgbe_vlan_rx_kill_vid(struct net_device
*netdev
,
4468 __be16 proto
, u16 vid
)
4470 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
4471 struct ixgbe_hw
*hw
= &adapter
->hw
;
4473 /* remove VID from filter table */
4474 if (vid
&& !(adapter
->flags2
& IXGBE_FLAG2_VLAN_PROMISC
))
4475 hw
->mac
.ops
.set_vfta(hw
, vid
, VMDQ_P(0), false, true);
4477 clear_bit(vid
, adapter
->active_vlans
);
4483 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
4484 * @adapter: driver data
4486 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter
*adapter
)
4488 struct ixgbe_hw
*hw
= &adapter
->hw
;
4492 switch (hw
->mac
.type
) {
4493 case ixgbe_mac_82598EB
:
4494 vlnctrl
= IXGBE_READ_REG(hw
, IXGBE_VLNCTRL
);
4495 vlnctrl
&= ~IXGBE_VLNCTRL_VME
;
4496 IXGBE_WRITE_REG(hw
, IXGBE_VLNCTRL
, vlnctrl
);
4498 case ixgbe_mac_82599EB
:
4499 case ixgbe_mac_X540
:
4500 case ixgbe_mac_X550
:
4501 case ixgbe_mac_X550EM_x
:
4502 case ixgbe_mac_x550em_a
:
4503 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
4504 struct ixgbe_ring
*ring
= adapter
->rx_ring
[i
];
4506 if (!netif_is_ixgbe(ring
->netdev
))
4510 vlnctrl
= IXGBE_READ_REG(hw
, IXGBE_RXDCTL(j
));
4511 vlnctrl
&= ~IXGBE_RXDCTL_VME
;
4512 IXGBE_WRITE_REG(hw
, IXGBE_RXDCTL(j
), vlnctrl
);
4521 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
4522 * @adapter: driver data
4524 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter
*adapter
)
4526 struct ixgbe_hw
*hw
= &adapter
->hw
;
4530 switch (hw
->mac
.type
) {
4531 case ixgbe_mac_82598EB
:
4532 vlnctrl
= IXGBE_READ_REG(hw
, IXGBE_VLNCTRL
);
4533 vlnctrl
|= IXGBE_VLNCTRL_VME
;
4534 IXGBE_WRITE_REG(hw
, IXGBE_VLNCTRL
, vlnctrl
);
4536 case ixgbe_mac_82599EB
:
4537 case ixgbe_mac_X540
:
4538 case ixgbe_mac_X550
:
4539 case ixgbe_mac_X550EM_x
:
4540 case ixgbe_mac_x550em_a
:
4541 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
4542 struct ixgbe_ring
*ring
= adapter
->rx_ring
[i
];
4544 if (!netif_is_ixgbe(ring
->netdev
))
4548 vlnctrl
= IXGBE_READ_REG(hw
, IXGBE_RXDCTL(j
));
4549 vlnctrl
|= IXGBE_RXDCTL_VME
;
4550 IXGBE_WRITE_REG(hw
, IXGBE_RXDCTL(j
), vlnctrl
);
4558 static void ixgbe_vlan_promisc_enable(struct ixgbe_adapter
*adapter
)
4560 struct ixgbe_hw
*hw
= &adapter
->hw
;
4563 vlnctrl
= IXGBE_READ_REG(hw
, IXGBE_VLNCTRL
);
4565 if (adapter
->flags
& IXGBE_FLAG_VMDQ_ENABLED
) {
4566 /* For VMDq and SR-IOV we must leave VLAN filtering enabled */
4567 vlnctrl
|= IXGBE_VLNCTRL_VFE
;
4568 IXGBE_WRITE_REG(hw
, IXGBE_VLNCTRL
, vlnctrl
);
4570 vlnctrl
&= ~IXGBE_VLNCTRL_VFE
;
4571 IXGBE_WRITE_REG(hw
, IXGBE_VLNCTRL
, vlnctrl
);
4575 /* Nothing to do for 82598 */
4576 if (hw
->mac
.type
== ixgbe_mac_82598EB
)
4579 /* We are already in VLAN promisc, nothing to do */
4580 if (adapter
->flags2
& IXGBE_FLAG2_VLAN_PROMISC
)
4583 /* Set flag so we don't redo unnecessary work */
4584 adapter
->flags2
|= IXGBE_FLAG2_VLAN_PROMISC
;
4586 /* Add PF to all active pools */
4587 for (i
= IXGBE_VLVF_ENTRIES
; --i
;) {
4588 u32 reg_offset
= IXGBE_VLVFB(i
* 2 + VMDQ_P(0) / 32);
4589 u32 vlvfb
= IXGBE_READ_REG(hw
, reg_offset
);
4591 vlvfb
|= BIT(VMDQ_P(0) % 32);
4592 IXGBE_WRITE_REG(hw
, reg_offset
, vlvfb
);
4595 /* Set all bits in the VLAN filter table array */
4596 for (i
= hw
->mac
.vft_size
; i
--;)
4597 IXGBE_WRITE_REG(hw
, IXGBE_VFTA(i
), ~0U);
4600 #define VFTA_BLOCK_SIZE 8
4601 static void ixgbe_scrub_vfta(struct ixgbe_adapter
*adapter
, u32 vfta_offset
)
4603 struct ixgbe_hw
*hw
= &adapter
->hw
;
4604 u32 vfta
[VFTA_BLOCK_SIZE
] = { 0 };
4605 u32 vid_start
= vfta_offset
* 32;
4606 u32 vid_end
= vid_start
+ (VFTA_BLOCK_SIZE
* 32);
4607 u32 i
, vid
, word
, bits
;
4609 for (i
= IXGBE_VLVF_ENTRIES
; --i
;) {
4610 u32 vlvf
= IXGBE_READ_REG(hw
, IXGBE_VLVF(i
));
4612 /* pull VLAN ID from VLVF */
4613 vid
= vlvf
& VLAN_VID_MASK
;
4615 /* only concern outselves with a certain range */
4616 if (vid
< vid_start
|| vid
>= vid_end
)
4620 /* record VLAN ID in VFTA */
4621 vfta
[(vid
- vid_start
) / 32] |= BIT(vid
% 32);
4623 /* if PF is part of this then continue */
4624 if (test_bit(vid
, adapter
->active_vlans
))
4628 /* remove PF from the pool */
4629 word
= i
* 2 + VMDQ_P(0) / 32;
4630 bits
= ~BIT(VMDQ_P(0) % 32);
4631 bits
&= IXGBE_READ_REG(hw
, IXGBE_VLVFB(word
));
4632 IXGBE_WRITE_REG(hw
, IXGBE_VLVFB(word
), bits
);
4635 /* extract values from active_vlans and write back to VFTA */
4636 for (i
= VFTA_BLOCK_SIZE
; i
--;) {
4637 vid
= (vfta_offset
+ i
) * 32;
4638 word
= vid
/ BITS_PER_LONG
;
4639 bits
= vid
% BITS_PER_LONG
;
4641 vfta
[i
] |= adapter
->active_vlans
[word
] >> bits
;
4643 IXGBE_WRITE_REG(hw
, IXGBE_VFTA(vfta_offset
+ i
), vfta
[i
]);
4647 static void ixgbe_vlan_promisc_disable(struct ixgbe_adapter
*adapter
)
4649 struct ixgbe_hw
*hw
= &adapter
->hw
;
4652 /* Set VLAN filtering to enabled */
4653 vlnctrl
= IXGBE_READ_REG(hw
, IXGBE_VLNCTRL
);
4654 vlnctrl
|= IXGBE_VLNCTRL_VFE
;
4655 IXGBE_WRITE_REG(hw
, IXGBE_VLNCTRL
, vlnctrl
);
4657 if (!(adapter
->flags
& IXGBE_FLAG_VMDQ_ENABLED
) ||
4658 hw
->mac
.type
== ixgbe_mac_82598EB
)
4661 /* We are not in VLAN promisc, nothing to do */
4662 if (!(adapter
->flags2
& IXGBE_FLAG2_VLAN_PROMISC
))
4665 /* Set flag so we don't redo unnecessary work */
4666 adapter
->flags2
&= ~IXGBE_FLAG2_VLAN_PROMISC
;
4668 for (i
= 0; i
< hw
->mac
.vft_size
; i
+= VFTA_BLOCK_SIZE
)
4669 ixgbe_scrub_vfta(adapter
, i
);
4672 static void ixgbe_restore_vlan(struct ixgbe_adapter
*adapter
)
4676 ixgbe_vlan_rx_add_vid(adapter
->netdev
, htons(ETH_P_8021Q
), 0);
4678 for_each_set_bit_from(vid
, adapter
->active_vlans
, VLAN_N_VID
)
4679 ixgbe_vlan_rx_add_vid(adapter
->netdev
, htons(ETH_P_8021Q
), vid
);
4683 * ixgbe_write_mc_addr_list - write multicast addresses to MTA
4684 * @netdev: network interface device structure
4686 * Writes multicast address list to the MTA hash table.
4687 * Returns: -ENOMEM on failure
4688 * 0 on no addresses written
4689 * X on writing X addresses to MTA
4691 static int ixgbe_write_mc_addr_list(struct net_device
*netdev
)
4693 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
4694 struct ixgbe_hw
*hw
= &adapter
->hw
;
4696 if (!netif_running(netdev
))
4699 if (hw
->mac
.ops
.update_mc_addr_list
)
4700 hw
->mac
.ops
.update_mc_addr_list(hw
, netdev
);
4704 #ifdef CONFIG_PCI_IOV
4705 ixgbe_restore_vf_multicasts(adapter
);
4708 return netdev_mc_count(netdev
);
4711 #ifdef CONFIG_PCI_IOV
4712 void ixgbe_full_sync_mac_table(struct ixgbe_adapter
*adapter
)
4714 struct ixgbe_mac_addr
*mac_table
= &adapter
->mac_table
[0];
4715 struct ixgbe_hw
*hw
= &adapter
->hw
;
4718 for (i
= 0; i
< hw
->mac
.num_rar_entries
; i
++, mac_table
++) {
4719 mac_table
->state
&= ~IXGBE_MAC_STATE_MODIFIED
;
4721 if (mac_table
->state
& IXGBE_MAC_STATE_IN_USE
)
4722 hw
->mac
.ops
.set_rar(hw
, i
,
4727 hw
->mac
.ops
.clear_rar(hw
, i
);
4732 static void ixgbe_sync_mac_table(struct ixgbe_adapter
*adapter
)
4734 struct ixgbe_mac_addr
*mac_table
= &adapter
->mac_table
[0];
4735 struct ixgbe_hw
*hw
= &adapter
->hw
;
4738 for (i
= 0; i
< hw
->mac
.num_rar_entries
; i
++, mac_table
++) {
4739 if (!(mac_table
->state
& IXGBE_MAC_STATE_MODIFIED
))
4742 mac_table
->state
&= ~IXGBE_MAC_STATE_MODIFIED
;
4744 if (mac_table
->state
& IXGBE_MAC_STATE_IN_USE
)
4745 hw
->mac
.ops
.set_rar(hw
, i
,
4750 hw
->mac
.ops
.clear_rar(hw
, i
);
4754 static void ixgbe_flush_sw_mac_table(struct ixgbe_adapter
*adapter
)
4756 struct ixgbe_mac_addr
*mac_table
= &adapter
->mac_table
[0];
4757 struct ixgbe_hw
*hw
= &adapter
->hw
;
4760 for (i
= 0; i
< hw
->mac
.num_rar_entries
; i
++, mac_table
++) {
4761 mac_table
->state
|= IXGBE_MAC_STATE_MODIFIED
;
4762 mac_table
->state
&= ~IXGBE_MAC_STATE_IN_USE
;
4765 ixgbe_sync_mac_table(adapter
);
4768 static int ixgbe_available_rars(struct ixgbe_adapter
*adapter
, u16 pool
)
4770 struct ixgbe_mac_addr
*mac_table
= &adapter
->mac_table
[0];
4771 struct ixgbe_hw
*hw
= &adapter
->hw
;
4774 for (i
= 0; i
< hw
->mac
.num_rar_entries
; i
++, mac_table
++) {
4775 /* do not count default RAR as available */
4776 if (mac_table
->state
& IXGBE_MAC_STATE_DEFAULT
)
4779 /* only count unused and addresses that belong to us */
4780 if (mac_table
->state
& IXGBE_MAC_STATE_IN_USE
) {
4781 if (mac_table
->pool
!= pool
)
4791 /* this function destroys the first RAR entry */
4792 static void ixgbe_mac_set_default_filter(struct ixgbe_adapter
*adapter
)
4794 struct ixgbe_mac_addr
*mac_table
= &adapter
->mac_table
[0];
4795 struct ixgbe_hw
*hw
= &adapter
->hw
;
4797 memcpy(&mac_table
->addr
, hw
->mac
.addr
, ETH_ALEN
);
4798 mac_table
->pool
= VMDQ_P(0);
4800 mac_table
->state
= IXGBE_MAC_STATE_DEFAULT
| IXGBE_MAC_STATE_IN_USE
;
4802 hw
->mac
.ops
.set_rar(hw
, 0, mac_table
->addr
, mac_table
->pool
,
4806 int ixgbe_add_mac_filter(struct ixgbe_adapter
*adapter
,
4807 const u8
*addr
, u16 pool
)
4809 struct ixgbe_mac_addr
*mac_table
= &adapter
->mac_table
[0];
4810 struct ixgbe_hw
*hw
= &adapter
->hw
;
4813 if (is_zero_ether_addr(addr
))
4816 for (i
= 0; i
< hw
->mac
.num_rar_entries
; i
++, mac_table
++) {
4817 if (mac_table
->state
& IXGBE_MAC_STATE_IN_USE
)
4820 ether_addr_copy(mac_table
->addr
, addr
);
4821 mac_table
->pool
= pool
;
4823 mac_table
->state
|= IXGBE_MAC_STATE_MODIFIED
|
4824 IXGBE_MAC_STATE_IN_USE
;
4826 ixgbe_sync_mac_table(adapter
);
4834 int ixgbe_del_mac_filter(struct ixgbe_adapter
*adapter
,
4835 const u8
*addr
, u16 pool
)
4837 struct ixgbe_mac_addr
*mac_table
= &adapter
->mac_table
[0];
4838 struct ixgbe_hw
*hw
= &adapter
->hw
;
4841 if (is_zero_ether_addr(addr
))
4844 /* search table for addr, if found clear IN_USE flag and sync */
4845 for (i
= 0; i
< hw
->mac
.num_rar_entries
; i
++, mac_table
++) {
4846 /* we can only delete an entry if it is in use */
4847 if (!(mac_table
->state
& IXGBE_MAC_STATE_IN_USE
))
4849 /* we only care about entries that belong to the given pool */
4850 if (mac_table
->pool
!= pool
)
4852 /* we only care about a specific MAC address */
4853 if (!ether_addr_equal(addr
, mac_table
->addr
))
4856 mac_table
->state
|= IXGBE_MAC_STATE_MODIFIED
;
4857 mac_table
->state
&= ~IXGBE_MAC_STATE_IN_USE
;
4859 ixgbe_sync_mac_table(adapter
);
4867 static int ixgbe_uc_sync(struct net_device
*netdev
, const unsigned char *addr
)
4869 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
4872 ret
= ixgbe_add_mac_filter(adapter
, addr
, VMDQ_P(0));
4874 return min_t(int, ret
, 0);
4877 static int ixgbe_uc_unsync(struct net_device
*netdev
, const unsigned char *addr
)
4879 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
4881 ixgbe_del_mac_filter(adapter
, addr
, VMDQ_P(0));
4887 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
4888 * @netdev: network interface device structure
4890 * The set_rx_method entry point is called whenever the unicast/multicast
4891 * address list or the network interface flags are updated. This routine is
4892 * responsible for configuring the hardware for proper unicast, multicast and
4895 void ixgbe_set_rx_mode(struct net_device
*netdev
)
4897 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
4898 struct ixgbe_hw
*hw
= &adapter
->hw
;
4899 u32 fctrl
, vmolr
= IXGBE_VMOLR_BAM
| IXGBE_VMOLR_AUPE
;
4900 netdev_features_t features
= netdev
->features
;
4903 /* Check for Promiscuous and All Multicast modes */
4904 fctrl
= IXGBE_READ_REG(hw
, IXGBE_FCTRL
);
4906 /* set all bits that we expect to always be set */
4907 fctrl
&= ~IXGBE_FCTRL_SBP
; /* disable store-bad-packets */
4908 fctrl
|= IXGBE_FCTRL_BAM
;
4909 fctrl
|= IXGBE_FCTRL_DPF
; /* discard pause frames when FC enabled */
4910 fctrl
|= IXGBE_FCTRL_PMCF
;
4912 /* clear the bits we are changing the status of */
4913 fctrl
&= ~(IXGBE_FCTRL_UPE
| IXGBE_FCTRL_MPE
);
4914 if (netdev
->flags
& IFF_PROMISC
) {
4915 hw
->addr_ctrl
.user_set_promisc
= true;
4916 fctrl
|= (IXGBE_FCTRL_UPE
| IXGBE_FCTRL_MPE
);
4917 vmolr
|= IXGBE_VMOLR_MPE
;
4918 features
&= ~NETIF_F_HW_VLAN_CTAG_FILTER
;
4920 if (netdev
->flags
& IFF_ALLMULTI
) {
4921 fctrl
|= IXGBE_FCTRL_MPE
;
4922 vmolr
|= IXGBE_VMOLR_MPE
;
4924 hw
->addr_ctrl
.user_set_promisc
= false;
4928 * Write addresses to available RAR registers, if there is not
4929 * sufficient space to store all the addresses then enable
4930 * unicast promiscuous mode
4932 if (__dev_uc_sync(netdev
, ixgbe_uc_sync
, ixgbe_uc_unsync
)) {
4933 fctrl
|= IXGBE_FCTRL_UPE
;
4934 vmolr
|= IXGBE_VMOLR_ROPE
;
4937 /* Write addresses to the MTA, if the attempt fails
4938 * then we should just turn on promiscuous mode so
4939 * that we can at least receive multicast traffic
4941 count
= ixgbe_write_mc_addr_list(netdev
);
4943 fctrl
|= IXGBE_FCTRL_MPE
;
4944 vmolr
|= IXGBE_VMOLR_MPE
;
4946 vmolr
|= IXGBE_VMOLR_ROMPE
;
4949 if (hw
->mac
.type
!= ixgbe_mac_82598EB
) {
4950 vmolr
|= IXGBE_READ_REG(hw
, IXGBE_VMOLR(VMDQ_P(0))) &
4951 ~(IXGBE_VMOLR_MPE
| IXGBE_VMOLR_ROMPE
|
4953 IXGBE_WRITE_REG(hw
, IXGBE_VMOLR(VMDQ_P(0)), vmolr
);
4956 /* This is useful for sniffing bad packets. */
4957 if (features
& NETIF_F_RXALL
) {
4958 /* UPE and MPE will be handled by normal PROMISC logic
4959 * in e1000e_set_rx_mode */
4960 fctrl
|= (IXGBE_FCTRL_SBP
| /* Receive bad packets */
4961 IXGBE_FCTRL_BAM
| /* RX All Bcast Pkts */
4962 IXGBE_FCTRL_PMCF
); /* RX All MAC Ctrl Pkts */
4964 fctrl
&= ~(IXGBE_FCTRL_DPF
);
4965 /* NOTE: VLAN filtering is disabled by setting PROMISC */
4968 IXGBE_WRITE_REG(hw
, IXGBE_FCTRL
, fctrl
);
4970 if (features
& NETIF_F_HW_VLAN_CTAG_RX
)
4971 ixgbe_vlan_strip_enable(adapter
);
4973 ixgbe_vlan_strip_disable(adapter
);
4975 if (features
& NETIF_F_HW_VLAN_CTAG_FILTER
)
4976 ixgbe_vlan_promisc_disable(adapter
);
4978 ixgbe_vlan_promisc_enable(adapter
);
4981 static void ixgbe_napi_enable_all(struct ixgbe_adapter
*adapter
)
4985 for (q_idx
= 0; q_idx
< adapter
->num_q_vectors
; q_idx
++)
4986 napi_enable(&adapter
->q_vector
[q_idx
]->napi
);
4989 static void ixgbe_napi_disable_all(struct ixgbe_adapter
*adapter
)
4993 for (q_idx
= 0; q_idx
< adapter
->num_q_vectors
; q_idx
++)
4994 napi_disable(&adapter
->q_vector
[q_idx
]->napi
);
4997 static void ixgbe_clear_udp_tunnel_port(struct ixgbe_adapter
*adapter
, u32 mask
)
4999 struct ixgbe_hw
*hw
= &adapter
->hw
;
5002 if (!(adapter
->flags
& (IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE
|
5003 IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE
)))
5006 vxlanctrl
= IXGBE_READ_REG(hw
, IXGBE_VXLANCTRL
) & ~mask
;
5007 IXGBE_WRITE_REG(hw
, IXGBE_VXLANCTRL
, vxlanctrl
);
5009 if (mask
& IXGBE_VXLANCTRL_VXLAN_UDPPORT_MASK
)
5010 adapter
->vxlan_port
= 0;
5012 if (mask
& IXGBE_VXLANCTRL_GENEVE_UDPPORT_MASK
)
5013 adapter
->geneve_port
= 0;
5016 #ifdef CONFIG_IXGBE_DCB
5018 * ixgbe_configure_dcb - Configure DCB hardware
5019 * @adapter: ixgbe adapter struct
5021 * This is called by the driver on open to configure the DCB hardware.
5022 * This is also called by the gennetlink interface when reconfiguring
5025 static void ixgbe_configure_dcb(struct ixgbe_adapter
*adapter
)
5027 struct ixgbe_hw
*hw
= &adapter
->hw
;
5028 int max_frame
= adapter
->netdev
->mtu
+ ETH_HLEN
+ ETH_FCS_LEN
;
5030 if (!(adapter
->flags
& IXGBE_FLAG_DCB_ENABLED
)) {
5031 if (hw
->mac
.type
== ixgbe_mac_82598EB
)
5032 netif_set_gso_max_size(adapter
->netdev
, 65536);
5036 if (hw
->mac
.type
== ixgbe_mac_82598EB
)
5037 netif_set_gso_max_size(adapter
->netdev
, 32768);
5040 if (adapter
->netdev
->features
& NETIF_F_FCOE_MTU
)
5041 max_frame
= max(max_frame
, IXGBE_FCOE_JUMBO_FRAME_SIZE
);
5044 /* reconfigure the hardware */
5045 if (adapter
->dcbx_cap
& DCB_CAP_DCBX_VER_CEE
) {
5046 ixgbe_dcb_calculate_tc_credits(hw
, &adapter
->dcb_cfg
, max_frame
,
5048 ixgbe_dcb_calculate_tc_credits(hw
, &adapter
->dcb_cfg
, max_frame
,
5050 ixgbe_dcb_hw_config(hw
, &adapter
->dcb_cfg
);
5051 } else if (adapter
->ixgbe_ieee_ets
&& adapter
->ixgbe_ieee_pfc
) {
5052 ixgbe_dcb_hw_ets(&adapter
->hw
,
5053 adapter
->ixgbe_ieee_ets
,
5055 ixgbe_dcb_hw_pfc_config(&adapter
->hw
,
5056 adapter
->ixgbe_ieee_pfc
->pfc_en
,
5057 adapter
->ixgbe_ieee_ets
->prio_tc
);
5060 /* Enable RSS Hash per TC */
5061 if (hw
->mac
.type
!= ixgbe_mac_82598EB
) {
5063 u16 rss_i
= adapter
->ring_feature
[RING_F_RSS
].indices
- 1;
5070 /* write msb to all 8 TCs in one write */
5071 IXGBE_WRITE_REG(hw
, IXGBE_RQTC
, msb
* 0x11111111);
5076 /* Additional bittime to account for IXGBE framing */
5077 #define IXGBE_ETH_FRAMING 20
5080 * ixgbe_hpbthresh - calculate high water mark for flow control
5082 * @adapter: board private structure to calculate for
5083 * @pb: packet buffer to calculate
5085 static int ixgbe_hpbthresh(struct ixgbe_adapter
*adapter
, int pb
)
5087 struct ixgbe_hw
*hw
= &adapter
->hw
;
5088 struct net_device
*dev
= adapter
->netdev
;
5089 int link
, tc
, kb
, marker
;
5092 /* Calculate max LAN frame size */
5093 tc
= link
= dev
->mtu
+ ETH_HLEN
+ ETH_FCS_LEN
+ IXGBE_ETH_FRAMING
;
5096 /* FCoE traffic class uses FCOE jumbo frames */
5097 if ((dev
->features
& NETIF_F_FCOE_MTU
) &&
5098 (tc
< IXGBE_FCOE_JUMBO_FRAME_SIZE
) &&
5099 (pb
== ixgbe_fcoe_get_tc(adapter
)))
5100 tc
= IXGBE_FCOE_JUMBO_FRAME_SIZE
;
5103 /* Calculate delay value for device */
5104 switch (hw
->mac
.type
) {
5105 case ixgbe_mac_X540
:
5106 case ixgbe_mac_X550
:
5107 case ixgbe_mac_X550EM_x
:
5108 case ixgbe_mac_x550em_a
:
5109 dv_id
= IXGBE_DV_X540(link
, tc
);
5112 dv_id
= IXGBE_DV(link
, tc
);
5116 /* Loopback switch introduces additional latency */
5117 if (adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
)
5118 dv_id
+= IXGBE_B2BT(tc
);
5120 /* Delay value is calculated in bit times convert to KB */
5121 kb
= IXGBE_BT2KB(dv_id
);
5122 rx_pba
= IXGBE_READ_REG(hw
, IXGBE_RXPBSIZE(pb
)) >> 10;
5124 marker
= rx_pba
- kb
;
5126 /* It is possible that the packet buffer is not large enough
5127 * to provide required headroom. In this case throw an error
5128 * to user and a do the best we can.
5131 e_warn(drv
, "Packet Buffer(%i) can not provide enough"
5132 "headroom to support flow control."
5133 "Decrease MTU or number of traffic classes\n", pb
);
5141 * ixgbe_lpbthresh - calculate low water mark for for flow control
5143 * @adapter: board private structure to calculate for
5144 * @pb: packet buffer to calculate
5146 static int ixgbe_lpbthresh(struct ixgbe_adapter
*adapter
, int pb
)
5148 struct ixgbe_hw
*hw
= &adapter
->hw
;
5149 struct net_device
*dev
= adapter
->netdev
;
5153 /* Calculate max LAN frame size */
5154 tc
= dev
->mtu
+ ETH_HLEN
+ ETH_FCS_LEN
;
5157 /* FCoE traffic class uses FCOE jumbo frames */
5158 if ((dev
->features
& NETIF_F_FCOE_MTU
) &&
5159 (tc
< IXGBE_FCOE_JUMBO_FRAME_SIZE
) &&
5160 (pb
== netdev_get_prio_tc_map(dev
, adapter
->fcoe
.up
)))
5161 tc
= IXGBE_FCOE_JUMBO_FRAME_SIZE
;
5164 /* Calculate delay value for device */
5165 switch (hw
->mac
.type
) {
5166 case ixgbe_mac_X540
:
5167 case ixgbe_mac_X550
:
5168 case ixgbe_mac_X550EM_x
:
5169 case ixgbe_mac_x550em_a
:
5170 dv_id
= IXGBE_LOW_DV_X540(tc
);
5173 dv_id
= IXGBE_LOW_DV(tc
);
5177 /* Delay value is calculated in bit times convert to KB */
5178 return IXGBE_BT2KB(dv_id
);
5182 * ixgbe_pbthresh_setup - calculate and setup high low water marks
5184 static void ixgbe_pbthresh_setup(struct ixgbe_adapter
*adapter
)
5186 struct ixgbe_hw
*hw
= &adapter
->hw
;
5187 int num_tc
= adapter
->hw_tcs
;
5193 for (i
= 0; i
< num_tc
; i
++) {
5194 hw
->fc
.high_water
[i
] = ixgbe_hpbthresh(adapter
, i
);
5195 hw
->fc
.low_water
[i
] = ixgbe_lpbthresh(adapter
, i
);
5197 /* Low water marks must not be larger than high water marks */
5198 if (hw
->fc
.low_water
[i
] > hw
->fc
.high_water
[i
])
5199 hw
->fc
.low_water
[i
] = 0;
5202 for (; i
< MAX_TRAFFIC_CLASS
; i
++)
5203 hw
->fc
.high_water
[i
] = 0;
5206 static void ixgbe_configure_pb(struct ixgbe_adapter
*adapter
)
5208 struct ixgbe_hw
*hw
= &adapter
->hw
;
5210 u8 tc
= adapter
->hw_tcs
;
5212 if (adapter
->flags
& IXGBE_FLAG_FDIR_HASH_CAPABLE
||
5213 adapter
->flags
& IXGBE_FLAG_FDIR_PERFECT_CAPABLE
)
5214 hdrm
= 32 << adapter
->fdir_pballoc
;
5218 hw
->mac
.ops
.set_rxpba(hw
, tc
, hdrm
, PBA_STRATEGY_EQUAL
);
5219 ixgbe_pbthresh_setup(adapter
);
5222 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter
*adapter
)
5224 struct ixgbe_hw
*hw
= &adapter
->hw
;
5225 struct hlist_node
*node2
;
5226 struct ixgbe_fdir_filter
*filter
;
5229 spin_lock(&adapter
->fdir_perfect_lock
);
5231 if (!hlist_empty(&adapter
->fdir_filter_list
))
5232 ixgbe_fdir_set_input_mask_82599(hw
, &adapter
->fdir_mask
);
5234 hlist_for_each_entry_safe(filter
, node2
,
5235 &adapter
->fdir_filter_list
, fdir_node
) {
5236 if (filter
->action
== IXGBE_FDIR_DROP_QUEUE
) {
5237 queue
= IXGBE_FDIR_DROP_QUEUE
;
5239 u32 ring
= ethtool_get_flow_spec_ring(filter
->action
);
5240 u8 vf
= ethtool_get_flow_spec_ring_vf(filter
->action
);
5242 if (!vf
&& (ring
>= adapter
->num_rx_queues
)) {
5243 e_err(drv
, "FDIR restore failed without VF, ring: %u\n",
5247 ((vf
> adapter
->num_vfs
) ||
5248 ring
>= adapter
->num_rx_queues_per_pool
)) {
5249 e_err(drv
, "FDIR restore failed with VF, vf: %hhu, ring: %u\n",
5254 /* Map the ring onto the absolute queue index */
5256 queue
= adapter
->rx_ring
[ring
]->reg_idx
;
5259 adapter
->num_rx_queues_per_pool
) + ring
;
5262 ixgbe_fdir_write_perfect_filter_82599(hw
,
5263 &filter
->filter
, filter
->sw_idx
, queue
);
5266 spin_unlock(&adapter
->fdir_perfect_lock
);
5270 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
5271 * @rx_ring: ring to free buffers from
5273 static void ixgbe_clean_rx_ring(struct ixgbe_ring
*rx_ring
)
5275 u16 i
= rx_ring
->next_to_clean
;
5276 struct ixgbe_rx_buffer
*rx_buffer
= &rx_ring
->rx_buffer_info
[i
];
5278 if (rx_ring
->xsk_umem
) {
5279 ixgbe_xsk_clean_rx_ring(rx_ring
);
5283 /* Free all the Rx ring sk_buffs */
5284 while (i
!= rx_ring
->next_to_alloc
) {
5285 if (rx_buffer
->skb
) {
5286 struct sk_buff
*skb
= rx_buffer
->skb
;
5287 if (IXGBE_CB(skb
)->page_released
)
5288 dma_unmap_page_attrs(rx_ring
->dev
,
5290 ixgbe_rx_pg_size(rx_ring
),
5296 /* Invalidate cache lines that may have been written to by
5297 * device so that we avoid corrupting memory.
5299 dma_sync_single_range_for_cpu(rx_ring
->dev
,
5301 rx_buffer
->page_offset
,
5302 ixgbe_rx_bufsz(rx_ring
),
5305 /* free resources associated with mapping */
5306 dma_unmap_page_attrs(rx_ring
->dev
, rx_buffer
->dma
,
5307 ixgbe_rx_pg_size(rx_ring
),
5310 __page_frag_cache_drain(rx_buffer
->page
,
5311 rx_buffer
->pagecnt_bias
);
5315 if (i
== rx_ring
->count
) {
5317 rx_buffer
= rx_ring
->rx_buffer_info
;
5322 rx_ring
->next_to_alloc
= 0;
5323 rx_ring
->next_to_clean
= 0;
5324 rx_ring
->next_to_use
= 0;
5327 static int ixgbe_fwd_ring_up(struct ixgbe_adapter
*adapter
,
5328 struct ixgbe_fwd_adapter
*accel
)
5330 u16 rss_i
= adapter
->ring_feature
[RING_F_RSS
].indices
;
5331 int num_tc
= netdev_get_num_tc(adapter
->netdev
);
5332 struct net_device
*vdev
= accel
->netdev
;
5335 baseq
= accel
->pool
* adapter
->num_rx_queues_per_pool
;
5336 netdev_dbg(vdev
, "pool %i:%i queues %i:%i\n",
5337 accel
->pool
, adapter
->num_rx_pools
,
5338 baseq
, baseq
+ adapter
->num_rx_queues_per_pool
);
5340 accel
->rx_base_queue
= baseq
;
5341 accel
->tx_base_queue
= baseq
;
5343 /* record configuration for macvlan interface in vdev */
5344 for (i
= 0; i
< num_tc
; i
++)
5345 netdev_bind_sb_channel_queue(adapter
->netdev
, vdev
,
5346 i
, rss_i
, baseq
+ (rss_i
* i
));
5348 for (i
= 0; i
< adapter
->num_rx_queues_per_pool
; i
++)
5349 adapter
->rx_ring
[baseq
+ i
]->netdev
= vdev
;
5351 /* Guarantee all rings are updated before we update the
5352 * MAC address filter.
5356 /* ixgbe_add_mac_filter will return an index if it succeeds, so we
5357 * need to only treat it as an error value if it is negative.
5359 err
= ixgbe_add_mac_filter(adapter
, vdev
->dev_addr
,
5360 VMDQ_P(accel
->pool
));
5364 /* if we cannot add the MAC rule then disable the offload */
5365 macvlan_release_l2fw_offload(vdev
);
5367 for (i
= 0; i
< adapter
->num_rx_queues_per_pool
; i
++)
5368 adapter
->rx_ring
[baseq
+ i
]->netdev
= NULL
;
5370 netdev_err(vdev
, "L2FW offload disabled due to L2 filter error\n");
5372 /* unbind the queues and drop the subordinate channel config */
5373 netdev_unbind_sb_channel(adapter
->netdev
, vdev
);
5374 netdev_set_sb_channel(vdev
, 0);
5376 clear_bit(accel
->pool
, adapter
->fwd_bitmask
);
5382 static int ixgbe_macvlan_up(struct net_device
*vdev
, void *data
)
5384 struct ixgbe_adapter
*adapter
= data
;
5385 struct ixgbe_fwd_adapter
*accel
;
5387 if (!netif_is_macvlan(vdev
))
5390 accel
= macvlan_accel_priv(vdev
);
5394 ixgbe_fwd_ring_up(adapter
, accel
);
5399 static void ixgbe_configure_dfwd(struct ixgbe_adapter
*adapter
)
5401 netdev_walk_all_upper_dev_rcu(adapter
->netdev
,
5402 ixgbe_macvlan_up
, adapter
);
5405 static void ixgbe_configure(struct ixgbe_adapter
*adapter
)
5407 struct ixgbe_hw
*hw
= &adapter
->hw
;
5409 ixgbe_configure_pb(adapter
);
5410 #ifdef CONFIG_IXGBE_DCB
5411 ixgbe_configure_dcb(adapter
);
5414 * We must restore virtualization before VLANs or else
5415 * the VLVF registers will not be populated
5417 ixgbe_configure_virtualization(adapter
);
5419 ixgbe_set_rx_mode(adapter
->netdev
);
5420 ixgbe_restore_vlan(adapter
);
5421 ixgbe_ipsec_restore(adapter
);
5423 switch (hw
->mac
.type
) {
5424 case ixgbe_mac_82599EB
:
5425 case ixgbe_mac_X540
:
5426 hw
->mac
.ops
.disable_rx_buff(hw
);
5432 if (adapter
->flags
& IXGBE_FLAG_FDIR_HASH_CAPABLE
) {
5433 ixgbe_init_fdir_signature_82599(&adapter
->hw
,
5434 adapter
->fdir_pballoc
);
5435 } else if (adapter
->flags
& IXGBE_FLAG_FDIR_PERFECT_CAPABLE
) {
5436 ixgbe_init_fdir_perfect_82599(&adapter
->hw
,
5437 adapter
->fdir_pballoc
);
5438 ixgbe_fdir_filter_restore(adapter
);
5441 switch (hw
->mac
.type
) {
5442 case ixgbe_mac_82599EB
:
5443 case ixgbe_mac_X540
:
5444 hw
->mac
.ops
.enable_rx_buff(hw
);
5450 #ifdef CONFIG_IXGBE_DCA
5452 if (adapter
->flags
& IXGBE_FLAG_DCA_CAPABLE
)
5453 ixgbe_setup_dca(adapter
);
5454 #endif /* CONFIG_IXGBE_DCA */
5457 /* configure FCoE L2 filters, redirection table, and Rx control */
5458 ixgbe_configure_fcoe(adapter
);
5460 #endif /* IXGBE_FCOE */
5461 ixgbe_configure_tx(adapter
);
5462 ixgbe_configure_rx(adapter
);
5463 ixgbe_configure_dfwd(adapter
);
5467 * ixgbe_sfp_link_config - set up SFP+ link
5468 * @adapter: pointer to private adapter struct
5470 static void ixgbe_sfp_link_config(struct ixgbe_adapter
*adapter
)
5473 * We are assuming the worst case scenario here, and that
5474 * is that an SFP was inserted/removed after the reset
5475 * but before SFP detection was enabled. As such the best
5476 * solution is to just start searching as soon as we start
5478 if (adapter
->hw
.mac
.type
== ixgbe_mac_82598EB
)
5479 adapter
->flags2
|= IXGBE_FLAG2_SEARCH_FOR_SFP
;
5481 adapter
->flags2
|= IXGBE_FLAG2_SFP_NEEDS_RESET
;
5482 adapter
->sfp_poll_time
= 0;
5486 * ixgbe_non_sfp_link_config - set up non-SFP+ link
5487 * @hw: pointer to private hardware struct
5489 * Returns 0 on success, negative on failure
5491 static int ixgbe_non_sfp_link_config(struct ixgbe_hw
*hw
)
5494 bool autoneg
, link_up
= false;
5495 int ret
= IXGBE_ERR_LINK_SETUP
;
5497 if (hw
->mac
.ops
.check_link
)
5498 ret
= hw
->mac
.ops
.check_link(hw
, &speed
, &link_up
, false);
5503 speed
= hw
->phy
.autoneg_advertised
;
5504 if ((!speed
) && (hw
->mac
.ops
.get_link_capabilities
))
5505 ret
= hw
->mac
.ops
.get_link_capabilities(hw
, &speed
,
5510 if (hw
->mac
.ops
.setup_link
)
5511 ret
= hw
->mac
.ops
.setup_link(hw
, speed
, link_up
);
5516 static void ixgbe_setup_gpie(struct ixgbe_adapter
*adapter
)
5518 struct ixgbe_hw
*hw
= &adapter
->hw
;
5521 if (adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
) {
5522 gpie
= IXGBE_GPIE_MSIX_MODE
| IXGBE_GPIE_PBA_SUPPORT
|
5524 gpie
|= IXGBE_GPIE_EIAME
;
5526 * use EIAM to auto-mask when MSI-X interrupt is asserted
5527 * this saves a register write for every interrupt
5529 switch (hw
->mac
.type
) {
5530 case ixgbe_mac_82598EB
:
5531 IXGBE_WRITE_REG(hw
, IXGBE_EIAM
, IXGBE_EICS_RTX_QUEUE
);
5533 case ixgbe_mac_82599EB
:
5534 case ixgbe_mac_X540
:
5535 case ixgbe_mac_X550
:
5536 case ixgbe_mac_X550EM_x
:
5537 case ixgbe_mac_x550em_a
:
5539 IXGBE_WRITE_REG(hw
, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
5540 IXGBE_WRITE_REG(hw
, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
5544 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
5545 * specifically only auto mask tx and rx interrupts */
5546 IXGBE_WRITE_REG(hw
, IXGBE_EIAM
, IXGBE_EICS_RTX_QUEUE
);
5549 /* XXX: to interrupt immediately for EICS writes, enable this */
5550 /* gpie |= IXGBE_GPIE_EIMEN; */
5552 if (adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
) {
5553 gpie
&= ~IXGBE_GPIE_VTMODE_MASK
;
5555 switch (adapter
->ring_feature
[RING_F_VMDQ
].mask
) {
5556 case IXGBE_82599_VMDQ_8Q_MASK
:
5557 gpie
|= IXGBE_GPIE_VTMODE_16
;
5559 case IXGBE_82599_VMDQ_4Q_MASK
:
5560 gpie
|= IXGBE_GPIE_VTMODE_32
;
5563 gpie
|= IXGBE_GPIE_VTMODE_64
;
5568 /* Enable Thermal over heat sensor interrupt */
5569 if (adapter
->flags2
& IXGBE_FLAG2_TEMP_SENSOR_CAPABLE
) {
5570 switch (adapter
->hw
.mac
.type
) {
5571 case ixgbe_mac_82599EB
:
5572 gpie
|= IXGBE_SDP0_GPIEN_8259X
;
5579 /* Enable fan failure interrupt */
5580 if (adapter
->flags
& IXGBE_FLAG_FAN_FAIL_CAPABLE
)
5581 gpie
|= IXGBE_SDP1_GPIEN(hw
);
5583 switch (hw
->mac
.type
) {
5584 case ixgbe_mac_82599EB
:
5585 gpie
|= IXGBE_SDP1_GPIEN_8259X
| IXGBE_SDP2_GPIEN_8259X
;
5587 case ixgbe_mac_X550EM_x
:
5588 case ixgbe_mac_x550em_a
:
5589 gpie
|= IXGBE_SDP0_GPIEN_X540
;
5595 IXGBE_WRITE_REG(hw
, IXGBE_GPIE
, gpie
);
5598 static void ixgbe_up_complete(struct ixgbe_adapter
*adapter
)
5600 struct ixgbe_hw
*hw
= &adapter
->hw
;
5604 ixgbe_get_hw_control(adapter
);
5605 ixgbe_setup_gpie(adapter
);
5607 if (adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
)
5608 ixgbe_configure_msix(adapter
);
5610 ixgbe_configure_msi_and_legacy(adapter
);
5612 /* enable the optics for 82599 SFP+ fiber */
5613 if (hw
->mac
.ops
.enable_tx_laser
)
5614 hw
->mac
.ops
.enable_tx_laser(hw
);
5616 if (hw
->phy
.ops
.set_phy_power
)
5617 hw
->phy
.ops
.set_phy_power(hw
, true);
5619 smp_mb__before_atomic();
5620 clear_bit(__IXGBE_DOWN
, &adapter
->state
);
5621 ixgbe_napi_enable_all(adapter
);
5623 if (ixgbe_is_sfp(hw
)) {
5624 ixgbe_sfp_link_config(adapter
);
5626 err
= ixgbe_non_sfp_link_config(hw
);
5628 e_err(probe
, "link_config FAILED %d\n", err
);
5631 /* clear any pending interrupts, may auto mask */
5632 IXGBE_READ_REG(hw
, IXGBE_EICR
);
5633 ixgbe_irq_enable(adapter
, true, true);
5636 * If this adapter has a fan, check to see if we had a failure
5637 * before we enabled the interrupt.
5639 if (adapter
->flags
& IXGBE_FLAG_FAN_FAIL_CAPABLE
) {
5640 u32 esdp
= IXGBE_READ_REG(hw
, IXGBE_ESDP
);
5641 if (esdp
& IXGBE_ESDP_SDP1
)
5642 e_crit(drv
, "Fan has stopped, replace the adapter\n");
5645 /* bring the link up in the watchdog, this could race with our first
5646 * link up interrupt but shouldn't be a problem */
5647 adapter
->flags
|= IXGBE_FLAG_NEED_LINK_UPDATE
;
5648 adapter
->link_check_timeout
= jiffies
;
5649 mod_timer(&adapter
->service_timer
, jiffies
);
5651 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
5652 ctrl_ext
= IXGBE_READ_REG(hw
, IXGBE_CTRL_EXT
);
5653 ctrl_ext
|= IXGBE_CTRL_EXT_PFRSTD
;
5654 IXGBE_WRITE_REG(hw
, IXGBE_CTRL_EXT
, ctrl_ext
);
5657 void ixgbe_reinit_locked(struct ixgbe_adapter
*adapter
)
5659 WARN_ON(in_interrupt());
5660 /* put off any impending NetWatchDogTimeout */
5661 netif_trans_update(adapter
->netdev
);
5663 while (test_and_set_bit(__IXGBE_RESETTING
, &adapter
->state
))
5664 usleep_range(1000, 2000);
5665 if (adapter
->hw
.phy
.type
== ixgbe_phy_fw
)
5666 ixgbe_watchdog_link_is_down(adapter
);
5667 ixgbe_down(adapter
);
5669 * If SR-IOV enabled then wait a bit before bringing the adapter
5670 * back up to give the VFs time to respond to the reset. The
5671 * two second wait is based upon the watchdog timer cycle in
5674 if (adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
)
5677 clear_bit(__IXGBE_RESETTING
, &adapter
->state
);
5680 void ixgbe_up(struct ixgbe_adapter
*adapter
)
5682 /* hardware has been reset, we need to reload some things */
5683 ixgbe_configure(adapter
);
5685 ixgbe_up_complete(adapter
);
5688 static unsigned long ixgbe_get_completion_timeout(struct ixgbe_adapter
*adapter
)
5692 pcie_capability_read_word(adapter
->pdev
, PCI_EXP_DEVCTL2
, &devctl2
);
5694 switch (devctl2
& IXGBE_PCIDEVCTRL2_TIMEO_MASK
) {
5695 case IXGBE_PCIDEVCTRL2_17_34s
:
5696 case IXGBE_PCIDEVCTRL2_4_8s
:
5697 /* For now we cap the upper limit on delay to 2 seconds
5698 * as we end up going up to 34 seconds of delay in worst
5699 * case timeout value.
5701 case IXGBE_PCIDEVCTRL2_1_2s
:
5702 return 2000000ul; /* 2.0 s */
5703 case IXGBE_PCIDEVCTRL2_260_520ms
:
5704 return 520000ul; /* 520 ms */
5705 case IXGBE_PCIDEVCTRL2_65_130ms
:
5706 return 130000ul; /* 130 ms */
5707 case IXGBE_PCIDEVCTRL2_16_32ms
:
5708 return 32000ul; /* 32 ms */
5709 case IXGBE_PCIDEVCTRL2_1_2ms
:
5710 return 2000ul; /* 2 ms */
5711 case IXGBE_PCIDEVCTRL2_50_100us
:
5712 return 100ul; /* 100 us */
5713 case IXGBE_PCIDEVCTRL2_16_32ms_def
:
5714 return 32000ul; /* 32 ms */
5719 /* We shouldn't need to hit this path, but just in case default as
5720 * though completion timeout is not supported and support 32ms.
5725 void ixgbe_disable_rx(struct ixgbe_adapter
*adapter
)
5727 unsigned long wait_delay
, delay_interval
;
5728 struct ixgbe_hw
*hw
= &adapter
->hw
;
5732 /* disable receives */
5733 hw
->mac
.ops
.disable_rx(hw
);
5735 if (ixgbe_removed(hw
->hw_addr
))
5738 /* disable all enabled Rx queues */
5739 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
5740 struct ixgbe_ring
*ring
= adapter
->rx_ring
[i
];
5741 u8 reg_idx
= ring
->reg_idx
;
5743 rxdctl
= IXGBE_READ_REG(hw
, IXGBE_RXDCTL(reg_idx
));
5744 rxdctl
&= ~IXGBE_RXDCTL_ENABLE
;
5745 rxdctl
|= IXGBE_RXDCTL_SWFLSH
;
5747 /* write value back with RXDCTL.ENABLE bit cleared */
5748 IXGBE_WRITE_REG(hw
, IXGBE_RXDCTL(reg_idx
), rxdctl
);
5751 /* RXDCTL.EN may not change on 82598 if link is down, so skip it */
5752 if (hw
->mac
.type
== ixgbe_mac_82598EB
&&
5753 !(IXGBE_READ_REG(hw
, IXGBE_LINKS
) & IXGBE_LINKS_UP
))
5756 /* Determine our minimum delay interval. We will increase this value
5757 * with each subsequent test. This way if the device returns quickly
5758 * we should spend as little time as possible waiting, however as
5759 * the time increases we will wait for larger periods of time.
5761 * The trick here is that we increase the interval using the
5762 * following pattern: 1x 3x 5x 7x 9x 11x 13x 15x 17x 19x. The result
5763 * of that wait is that it totals up to 100x whatever interval we
5764 * choose. Since our minimum wait is 100us we can just divide the
5765 * total timeout by 100 to get our minimum delay interval.
5767 delay_interval
= ixgbe_get_completion_timeout(adapter
) / 100;
5769 wait_loop
= IXGBE_MAX_RX_DESC_POLL
;
5770 wait_delay
= delay_interval
;
5772 while (wait_loop
--) {
5773 usleep_range(wait_delay
, wait_delay
+ 10);
5774 wait_delay
+= delay_interval
* 2;
5777 /* OR together the reading of all the active RXDCTL registers,
5778 * and then test the result. We need the disable to complete
5779 * before we start freeing the memory and invalidating the
5782 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
5783 struct ixgbe_ring
*ring
= adapter
->rx_ring
[i
];
5784 u8 reg_idx
= ring
->reg_idx
;
5786 rxdctl
|= IXGBE_READ_REG(hw
, IXGBE_RXDCTL(reg_idx
));
5789 if (!(rxdctl
& IXGBE_RXDCTL_ENABLE
))
5794 "RXDCTL.ENABLE for one or more queues not cleared within the polling period\n");
5797 void ixgbe_disable_tx(struct ixgbe_adapter
*adapter
)
5799 unsigned long wait_delay
, delay_interval
;
5800 struct ixgbe_hw
*hw
= &adapter
->hw
;
5804 if (ixgbe_removed(hw
->hw_addr
))
5807 /* disable all enabled Tx queues */
5808 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
5809 struct ixgbe_ring
*ring
= adapter
->tx_ring
[i
];
5810 u8 reg_idx
= ring
->reg_idx
;
5812 IXGBE_WRITE_REG(hw
, IXGBE_TXDCTL(reg_idx
), IXGBE_TXDCTL_SWFLSH
);
5815 /* disable all enabled XDP Tx queues */
5816 for (i
= 0; i
< adapter
->num_xdp_queues
; i
++) {
5817 struct ixgbe_ring
*ring
= adapter
->xdp_ring
[i
];
5818 u8 reg_idx
= ring
->reg_idx
;
5820 IXGBE_WRITE_REG(hw
, IXGBE_TXDCTL(reg_idx
), IXGBE_TXDCTL_SWFLSH
);
5823 /* If the link is not up there shouldn't be much in the way of
5824 * pending transactions. Those that are left will be flushed out
5825 * when the reset logic goes through the flush sequence to clean out
5826 * the pending Tx transactions.
5828 if (!(IXGBE_READ_REG(hw
, IXGBE_LINKS
) & IXGBE_LINKS_UP
))
5829 goto dma_engine_disable
;
5831 /* Determine our minimum delay interval. We will increase this value
5832 * with each subsequent test. This way if the device returns quickly
5833 * we should spend as little time as possible waiting, however as
5834 * the time increases we will wait for larger periods of time.
5836 * The trick here is that we increase the interval using the
5837 * following pattern: 1x 3x 5x 7x 9x 11x 13x 15x 17x 19x. The result
5838 * of that wait is that it totals up to 100x whatever interval we
5839 * choose. Since our minimum wait is 100us we can just divide the
5840 * total timeout by 100 to get our minimum delay interval.
5842 delay_interval
= ixgbe_get_completion_timeout(adapter
) / 100;
5844 wait_loop
= IXGBE_MAX_RX_DESC_POLL
;
5845 wait_delay
= delay_interval
;
5847 while (wait_loop
--) {
5848 usleep_range(wait_delay
, wait_delay
+ 10);
5849 wait_delay
+= delay_interval
* 2;
5852 /* OR together the reading of all the active TXDCTL registers,
5853 * and then test the result. We need the disable to complete
5854 * before we start freeing the memory and invalidating the
5857 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
5858 struct ixgbe_ring
*ring
= adapter
->tx_ring
[i
];
5859 u8 reg_idx
= ring
->reg_idx
;
5861 txdctl
|= IXGBE_READ_REG(hw
, IXGBE_TXDCTL(reg_idx
));
5863 for (i
= 0; i
< adapter
->num_xdp_queues
; i
++) {
5864 struct ixgbe_ring
*ring
= adapter
->xdp_ring
[i
];
5865 u8 reg_idx
= ring
->reg_idx
;
5867 txdctl
|= IXGBE_READ_REG(hw
, IXGBE_TXDCTL(reg_idx
));
5870 if (!(txdctl
& IXGBE_TXDCTL_ENABLE
))
5871 goto dma_engine_disable
;
5875 "TXDCTL.ENABLE for one or more queues not cleared within the polling period\n");
5878 /* Disable the Tx DMA engine on 82599 and later MAC */
5879 switch (hw
->mac
.type
) {
5880 case ixgbe_mac_82599EB
:
5881 case ixgbe_mac_X540
:
5882 case ixgbe_mac_X550
:
5883 case ixgbe_mac_X550EM_x
:
5884 case ixgbe_mac_x550em_a
:
5885 IXGBE_WRITE_REG(hw
, IXGBE_DMATXCTL
,
5886 (IXGBE_READ_REG(hw
, IXGBE_DMATXCTL
) &
5887 ~IXGBE_DMATXCTL_TE
));
5894 void ixgbe_reset(struct ixgbe_adapter
*adapter
)
5896 struct ixgbe_hw
*hw
= &adapter
->hw
;
5897 struct net_device
*netdev
= adapter
->netdev
;
5900 if (ixgbe_removed(hw
->hw_addr
))
5902 /* lock SFP init bit to prevent race conditions with the watchdog */
5903 while (test_and_set_bit(__IXGBE_IN_SFP_INIT
, &adapter
->state
))
5904 usleep_range(1000, 2000);
5906 /* clear all SFP and link config related flags while holding SFP_INIT */
5907 adapter
->flags2
&= ~(IXGBE_FLAG2_SEARCH_FOR_SFP
|
5908 IXGBE_FLAG2_SFP_NEEDS_RESET
);
5909 adapter
->flags
&= ~IXGBE_FLAG_NEED_LINK_CONFIG
;
5911 err
= hw
->mac
.ops
.init_hw(hw
);
5914 case IXGBE_ERR_SFP_NOT_PRESENT
:
5915 case IXGBE_ERR_SFP_NOT_SUPPORTED
:
5917 case IXGBE_ERR_MASTER_REQUESTS_PENDING
:
5918 e_dev_err("master disable timed out\n");
5920 case IXGBE_ERR_EEPROM_VERSION
:
5921 /* We are running on a pre-production device, log a warning */
5922 e_dev_warn("This device is a pre-production adapter/LOM. "
5923 "Please be aware there may be issues associated with "
5924 "your hardware. If you are experiencing problems "
5925 "please contact your Intel or hardware "
5926 "representative who provided you with this "
5930 e_dev_err("Hardware Error: %d\n", err
);
5933 clear_bit(__IXGBE_IN_SFP_INIT
, &adapter
->state
);
5935 /* flush entries out of MAC table */
5936 ixgbe_flush_sw_mac_table(adapter
);
5937 __dev_uc_unsync(netdev
, NULL
);
5939 /* do not flush user set addresses */
5940 ixgbe_mac_set_default_filter(adapter
);
5942 /* update SAN MAC vmdq pool selection */
5943 if (hw
->mac
.san_mac_rar_index
)
5944 hw
->mac
.ops
.set_vmdq_san_mac(hw
, VMDQ_P(0));
5946 if (test_bit(__IXGBE_PTP_RUNNING
, &adapter
->state
))
5947 ixgbe_ptp_reset(adapter
);
5949 if (hw
->phy
.ops
.set_phy_power
) {
5950 if (!netif_running(adapter
->netdev
) && !adapter
->wol
)
5951 hw
->phy
.ops
.set_phy_power(hw
, false);
5953 hw
->phy
.ops
.set_phy_power(hw
, true);
5958 * ixgbe_clean_tx_ring - Free Tx Buffers
5959 * @tx_ring: ring to be cleaned
5961 static void ixgbe_clean_tx_ring(struct ixgbe_ring
*tx_ring
)
5963 u16 i
= tx_ring
->next_to_clean
;
5964 struct ixgbe_tx_buffer
*tx_buffer
= &tx_ring
->tx_buffer_info
[i
];
5966 if (tx_ring
->xsk_umem
) {
5967 ixgbe_xsk_clean_tx_ring(tx_ring
);
5971 while (i
!= tx_ring
->next_to_use
) {
5972 union ixgbe_adv_tx_desc
*eop_desc
, *tx_desc
;
5974 /* Free all the Tx ring sk_buffs */
5975 if (ring_is_xdp(tx_ring
))
5976 xdp_return_frame(tx_buffer
->xdpf
);
5978 dev_kfree_skb_any(tx_buffer
->skb
);
5980 /* unmap skb header data */
5981 dma_unmap_single(tx_ring
->dev
,
5982 dma_unmap_addr(tx_buffer
, dma
),
5983 dma_unmap_len(tx_buffer
, len
),
5986 /* check for eop_desc to determine the end of the packet */
5987 eop_desc
= tx_buffer
->next_to_watch
;
5988 tx_desc
= IXGBE_TX_DESC(tx_ring
, i
);
5990 /* unmap remaining buffers */
5991 while (tx_desc
!= eop_desc
) {
5995 if (unlikely(i
== tx_ring
->count
)) {
5997 tx_buffer
= tx_ring
->tx_buffer_info
;
5998 tx_desc
= IXGBE_TX_DESC(tx_ring
, 0);
6001 /* unmap any remaining paged data */
6002 if (dma_unmap_len(tx_buffer
, len
))
6003 dma_unmap_page(tx_ring
->dev
,
6004 dma_unmap_addr(tx_buffer
, dma
),
6005 dma_unmap_len(tx_buffer
, len
),
6009 /* move us one more past the eop_desc for start of next pkt */
6012 if (unlikely(i
== tx_ring
->count
)) {
6014 tx_buffer
= tx_ring
->tx_buffer_info
;
6018 /* reset BQL for queue */
6019 if (!ring_is_xdp(tx_ring
))
6020 netdev_tx_reset_queue(txring_txq(tx_ring
));
6023 /* reset next_to_use and next_to_clean */
6024 tx_ring
->next_to_use
= 0;
6025 tx_ring
->next_to_clean
= 0;
6029 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
6030 * @adapter: board private structure
6032 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter
*adapter
)
6036 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
6037 ixgbe_clean_rx_ring(adapter
->rx_ring
[i
]);
6041 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
6042 * @adapter: board private structure
6044 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter
*adapter
)
6048 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
6049 ixgbe_clean_tx_ring(adapter
->tx_ring
[i
]);
6050 for (i
= 0; i
< adapter
->num_xdp_queues
; i
++)
6051 ixgbe_clean_tx_ring(adapter
->xdp_ring
[i
]);
6054 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter
*adapter
)
6056 struct hlist_node
*node2
;
6057 struct ixgbe_fdir_filter
*filter
;
6059 spin_lock(&adapter
->fdir_perfect_lock
);
6061 hlist_for_each_entry_safe(filter
, node2
,
6062 &adapter
->fdir_filter_list
, fdir_node
) {
6063 hlist_del(&filter
->fdir_node
);
6066 adapter
->fdir_filter_count
= 0;
6068 spin_unlock(&adapter
->fdir_perfect_lock
);
6071 void ixgbe_down(struct ixgbe_adapter
*adapter
)
6073 struct net_device
*netdev
= adapter
->netdev
;
6074 struct ixgbe_hw
*hw
= &adapter
->hw
;
6077 /* signal that we are down to the interrupt handler */
6078 if (test_and_set_bit(__IXGBE_DOWN
, &adapter
->state
))
6079 return; /* do nothing if already down */
6081 /* Shut off incoming Tx traffic */
6082 netif_tx_stop_all_queues(netdev
);
6084 /* call carrier off first to avoid false dev_watchdog timeouts */
6085 netif_carrier_off(netdev
);
6086 netif_tx_disable(netdev
);
6089 ixgbe_disable_rx(adapter
);
6091 /* synchronize_rcu() needed for pending XDP buffers to drain */
6092 if (adapter
->xdp_ring
[0])
6095 ixgbe_irq_disable(adapter
);
6097 ixgbe_napi_disable_all(adapter
);
6099 clear_bit(__IXGBE_RESET_REQUESTED
, &adapter
->state
);
6100 adapter
->flags2
&= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT
;
6101 adapter
->flags
&= ~IXGBE_FLAG_NEED_LINK_UPDATE
;
6103 del_timer_sync(&adapter
->service_timer
);
6105 if (adapter
->num_vfs
) {
6106 /* Clear EITR Select mapping */
6107 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EITRSEL
, 0);
6109 /* Mark all the VFs as inactive */
6110 for (i
= 0 ; i
< adapter
->num_vfs
; i
++)
6111 adapter
->vfinfo
[i
].clear_to_send
= false;
6113 /* ping all the active vfs to let them know we are going down */
6114 ixgbe_ping_all_vfs(adapter
);
6116 /* Disable all VFTE/VFRE TX/RX */
6117 ixgbe_disable_tx_rx(adapter
);
6120 /* disable transmits in the hardware now that interrupts are off */
6121 ixgbe_disable_tx(adapter
);
6123 if (!pci_channel_offline(adapter
->pdev
))
6124 ixgbe_reset(adapter
);
6126 /* power down the optics for 82599 SFP+ fiber */
6127 if (hw
->mac
.ops
.disable_tx_laser
)
6128 hw
->mac
.ops
.disable_tx_laser(hw
);
6130 ixgbe_clean_all_tx_rings(adapter
);
6131 ixgbe_clean_all_rx_rings(adapter
);
6135 * ixgbe_eee_capable - helper function to determine EEE support on X550
6136 * @adapter: board private structure
6138 static void ixgbe_set_eee_capable(struct ixgbe_adapter
*adapter
)
6140 struct ixgbe_hw
*hw
= &adapter
->hw
;
6142 switch (hw
->device_id
) {
6143 case IXGBE_DEV_ID_X550EM_A_1G_T
:
6144 case IXGBE_DEV_ID_X550EM_A_1G_T_L
:
6145 if (!hw
->phy
.eee_speeds_supported
)
6147 adapter
->flags2
|= IXGBE_FLAG2_EEE_CAPABLE
;
6148 if (!hw
->phy
.eee_speeds_advertised
)
6150 adapter
->flags2
|= IXGBE_FLAG2_EEE_ENABLED
;
6153 adapter
->flags2
&= ~IXGBE_FLAG2_EEE_CAPABLE
;
6154 adapter
->flags2
&= ~IXGBE_FLAG2_EEE_ENABLED
;
6160 * ixgbe_tx_timeout - Respond to a Tx Hang
6161 * @netdev: network interface device structure
6163 static void ixgbe_tx_timeout(struct net_device
*netdev
, unsigned int txqueue
)
6165 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
6167 /* Do the reset outside of interrupt context */
6168 ixgbe_tx_timeout_reset(adapter
);
6171 #ifdef CONFIG_IXGBE_DCB
6172 static void ixgbe_init_dcb(struct ixgbe_adapter
*adapter
)
6174 struct ixgbe_hw
*hw
= &adapter
->hw
;
6175 struct tc_configuration
*tc
;
6178 switch (hw
->mac
.type
) {
6179 case ixgbe_mac_82598EB
:
6180 case ixgbe_mac_82599EB
:
6181 adapter
->dcb_cfg
.num_tcs
.pg_tcs
= MAX_TRAFFIC_CLASS
;
6182 adapter
->dcb_cfg
.num_tcs
.pfc_tcs
= MAX_TRAFFIC_CLASS
;
6184 case ixgbe_mac_X540
:
6185 case ixgbe_mac_X550
:
6186 adapter
->dcb_cfg
.num_tcs
.pg_tcs
= X540_TRAFFIC_CLASS
;
6187 adapter
->dcb_cfg
.num_tcs
.pfc_tcs
= X540_TRAFFIC_CLASS
;
6189 case ixgbe_mac_X550EM_x
:
6190 case ixgbe_mac_x550em_a
:
6192 adapter
->dcb_cfg
.num_tcs
.pg_tcs
= DEF_TRAFFIC_CLASS
;
6193 adapter
->dcb_cfg
.num_tcs
.pfc_tcs
= DEF_TRAFFIC_CLASS
;
6197 /* Configure DCB traffic classes */
6198 for (j
= 0; j
< MAX_TRAFFIC_CLASS
; j
++) {
6199 tc
= &adapter
->dcb_cfg
.tc_config
[j
];
6200 tc
->path
[DCB_TX_CONFIG
].bwg_id
= 0;
6201 tc
->path
[DCB_TX_CONFIG
].bwg_percent
= 12 + (j
& 1);
6202 tc
->path
[DCB_RX_CONFIG
].bwg_id
= 0;
6203 tc
->path
[DCB_RX_CONFIG
].bwg_percent
= 12 + (j
& 1);
6204 tc
->dcb_pfc
= pfc_disabled
;
6207 /* Initialize default user to priority mapping, UPx->TC0 */
6208 tc
= &adapter
->dcb_cfg
.tc_config
[0];
6209 tc
->path
[DCB_TX_CONFIG
].up_to_tc_bitmap
= 0xFF;
6210 tc
->path
[DCB_RX_CONFIG
].up_to_tc_bitmap
= 0xFF;
6212 adapter
->dcb_cfg
.bw_percentage
[DCB_TX_CONFIG
][0] = 100;
6213 adapter
->dcb_cfg
.bw_percentage
[DCB_RX_CONFIG
][0] = 100;
6214 adapter
->dcb_cfg
.pfc_mode_enable
= false;
6215 adapter
->dcb_set_bitmap
= 0x00;
6216 if (adapter
->flags
& IXGBE_FLAG_DCB_CAPABLE
)
6217 adapter
->dcbx_cap
= DCB_CAP_DCBX_HOST
| DCB_CAP_DCBX_VER_CEE
;
6218 memcpy(&adapter
->temp_dcb_cfg
, &adapter
->dcb_cfg
,
6219 sizeof(adapter
->temp_dcb_cfg
));
6224 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
6225 * @adapter: board private structure to initialize
6226 * @ii: pointer to ixgbe_info for device
6228 * ixgbe_sw_init initializes the Adapter private data structure.
6229 * Fields are initialized based on PCI device information and
6230 * OS network device settings (MTU size).
6232 static int ixgbe_sw_init(struct ixgbe_adapter
*adapter
,
6233 const struct ixgbe_info
*ii
)
6235 struct ixgbe_hw
*hw
= &adapter
->hw
;
6236 struct pci_dev
*pdev
= adapter
->pdev
;
6237 unsigned int rss
, fdir
;
6241 /* PCI config space info */
6243 hw
->vendor_id
= pdev
->vendor
;
6244 hw
->device_id
= pdev
->device
;
6245 hw
->revision_id
= pdev
->revision
;
6246 hw
->subsystem_vendor_id
= pdev
->subsystem_vendor
;
6247 hw
->subsystem_device_id
= pdev
->subsystem_device
;
6249 /* get_invariants needs the device IDs */
6250 ii
->get_invariants(hw
);
6252 /* Set common capability flags and settings */
6253 rss
= min_t(int, ixgbe_max_rss_indices(adapter
), num_online_cpus());
6254 adapter
->ring_feature
[RING_F_RSS
].limit
= rss
;
6255 adapter
->flags2
|= IXGBE_FLAG2_RSC_CAPABLE
;
6256 adapter
->max_q_vectors
= MAX_Q_VECTORS_82599
;
6257 adapter
->atr_sample_rate
= 20;
6258 fdir
= min_t(int, IXGBE_MAX_FDIR_INDICES
, num_online_cpus());
6259 adapter
->ring_feature
[RING_F_FDIR
].limit
= fdir
;
6260 adapter
->fdir_pballoc
= IXGBE_FDIR_PBALLOC_64K
;
6261 adapter
->ring_feature
[RING_F_VMDQ
].limit
= 1;
6262 #ifdef CONFIG_IXGBE_DCA
6263 adapter
->flags
|= IXGBE_FLAG_DCA_CAPABLE
;
6265 #ifdef CONFIG_IXGBE_DCB
6266 adapter
->flags
|= IXGBE_FLAG_DCB_CAPABLE
;
6267 adapter
->flags
&= ~IXGBE_FLAG_DCB_ENABLED
;
6270 adapter
->flags
|= IXGBE_FLAG_FCOE_CAPABLE
;
6271 adapter
->flags
&= ~IXGBE_FLAG_FCOE_ENABLED
;
6272 #ifdef CONFIG_IXGBE_DCB
6273 /* Default traffic class to use for FCoE */
6274 adapter
->fcoe
.up
= IXGBE_FCOE_DEFTC
;
6275 #endif /* CONFIG_IXGBE_DCB */
6276 #endif /* IXGBE_FCOE */
6278 /* initialize static ixgbe jump table entries */
6279 adapter
->jump_tables
[0] = kzalloc(sizeof(*adapter
->jump_tables
[0]),
6281 if (!adapter
->jump_tables
[0])
6283 adapter
->jump_tables
[0]->mat
= ixgbe_ipv4_fields
;
6285 for (i
= 1; i
< IXGBE_MAX_LINK_HANDLE
; i
++)
6286 adapter
->jump_tables
[i
] = NULL
;
6288 adapter
->mac_table
= kcalloc(hw
->mac
.num_rar_entries
,
6289 sizeof(struct ixgbe_mac_addr
),
6291 if (!adapter
->mac_table
)
6294 if (ixgbe_init_rss_key(adapter
))
6297 adapter
->af_xdp_zc_qps
= bitmap_zalloc(MAX_XDP_QUEUES
, GFP_KERNEL
);
6298 if (!adapter
->af_xdp_zc_qps
)
6301 /* Set MAC specific capability flags and exceptions */
6302 switch (hw
->mac
.type
) {
6303 case ixgbe_mac_82598EB
:
6304 adapter
->flags2
&= ~IXGBE_FLAG2_RSC_CAPABLE
;
6306 if (hw
->device_id
== IXGBE_DEV_ID_82598AT
)
6307 adapter
->flags
|= IXGBE_FLAG_FAN_FAIL_CAPABLE
;
6309 adapter
->max_q_vectors
= MAX_Q_VECTORS_82598
;
6310 adapter
->ring_feature
[RING_F_FDIR
].limit
= 0;
6311 adapter
->atr_sample_rate
= 0;
6312 adapter
->fdir_pballoc
= 0;
6314 adapter
->flags
&= ~IXGBE_FLAG_FCOE_CAPABLE
;
6315 adapter
->flags
&= ~IXGBE_FLAG_FCOE_ENABLED
;
6316 #ifdef CONFIG_IXGBE_DCB
6317 adapter
->fcoe
.up
= 0;
6318 #endif /* IXGBE_DCB */
6319 #endif /* IXGBE_FCOE */
6321 case ixgbe_mac_82599EB
:
6322 if (hw
->device_id
== IXGBE_DEV_ID_82599_T3_LOM
)
6323 adapter
->flags2
|= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE
;
6325 case ixgbe_mac_X540
:
6326 fwsm
= IXGBE_READ_REG(hw
, IXGBE_FWSM(hw
));
6327 if (fwsm
& IXGBE_FWSM_TS_ENABLED
)
6328 adapter
->flags2
|= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE
;
6330 case ixgbe_mac_x550em_a
:
6331 adapter
->flags
|= IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE
;
6332 switch (hw
->device_id
) {
6333 case IXGBE_DEV_ID_X550EM_A_1G_T
:
6334 case IXGBE_DEV_ID_X550EM_A_1G_T_L
:
6335 adapter
->flags2
|= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE
;
6341 case ixgbe_mac_X550EM_x
:
6342 #ifdef CONFIG_IXGBE_DCB
6343 adapter
->flags
&= ~IXGBE_FLAG_DCB_CAPABLE
;
6346 adapter
->flags
&= ~IXGBE_FLAG_FCOE_CAPABLE
;
6347 #ifdef CONFIG_IXGBE_DCB
6348 adapter
->fcoe
.up
= 0;
6349 #endif /* IXGBE_DCB */
6350 #endif /* IXGBE_FCOE */
6352 case ixgbe_mac_X550
:
6353 if (hw
->mac
.type
== ixgbe_mac_X550
)
6354 adapter
->flags2
|= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE
;
6355 #ifdef CONFIG_IXGBE_DCA
6356 adapter
->flags
&= ~IXGBE_FLAG_DCA_CAPABLE
;
6358 adapter
->flags
|= IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE
;
6365 /* FCoE support exists, always init the FCoE lock */
6366 spin_lock_init(&adapter
->fcoe
.lock
);
6369 /* n-tuple support exists, always init our spinlock */
6370 spin_lock_init(&adapter
->fdir_perfect_lock
);
6372 #ifdef CONFIG_IXGBE_DCB
6373 ixgbe_init_dcb(adapter
);
6375 ixgbe_init_ipsec_offload(adapter
);
6377 /* default flow control settings */
6378 hw
->fc
.requested_mode
= ixgbe_fc_full
;
6379 hw
->fc
.current_mode
= ixgbe_fc_full
; /* init for ethtool output */
6380 ixgbe_pbthresh_setup(adapter
);
6381 hw
->fc
.pause_time
= IXGBE_DEFAULT_FCPAUSE
;
6382 hw
->fc
.send_xon
= true;
6383 hw
->fc
.disable_fc_autoneg
= ixgbe_device_supports_autoneg_fc(hw
);
6385 #ifdef CONFIG_PCI_IOV
6387 e_dev_warn("Enabling SR-IOV VFs using the max_vfs module parameter is deprecated - please use the pci sysfs interface instead.\n");
6389 /* assign number of SR-IOV VFs */
6390 if (hw
->mac
.type
!= ixgbe_mac_82598EB
) {
6391 if (max_vfs
> IXGBE_MAX_VFS_DRV_LIMIT
) {
6393 e_dev_warn("max_vfs parameter out of range. Not assigning any SR-IOV VFs\n");
6396 #endif /* CONFIG_PCI_IOV */
6398 /* enable itr by default in dynamic mode */
6399 adapter
->rx_itr_setting
= 1;
6400 adapter
->tx_itr_setting
= 1;
6402 /* set default ring sizes */
6403 adapter
->tx_ring_count
= IXGBE_DEFAULT_TXD
;
6404 adapter
->rx_ring_count
= IXGBE_DEFAULT_RXD
;
6406 /* set default work limits */
6407 adapter
->tx_work_limit
= IXGBE_DEFAULT_TX_WORK
;
6409 /* initialize eeprom parameters */
6410 if (ixgbe_init_eeprom_params_generic(hw
)) {
6411 e_dev_err("EEPROM initialization failed\n");
6415 /* PF holds first pool slot */
6416 set_bit(0, adapter
->fwd_bitmask
);
6417 set_bit(__IXGBE_DOWN
, &adapter
->state
);
6423 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
6424 * @tx_ring: tx descriptor ring (for a specific queue) to setup
6426 * Return 0 on success, negative on failure
6428 int ixgbe_setup_tx_resources(struct ixgbe_ring
*tx_ring
)
6430 struct device
*dev
= tx_ring
->dev
;
6431 int orig_node
= dev_to_node(dev
);
6432 int ring_node
= NUMA_NO_NODE
;
6435 size
= sizeof(struct ixgbe_tx_buffer
) * tx_ring
->count
;
6437 if (tx_ring
->q_vector
)
6438 ring_node
= tx_ring
->q_vector
->numa_node
;
6440 tx_ring
->tx_buffer_info
= vmalloc_node(size
, ring_node
);
6441 if (!tx_ring
->tx_buffer_info
)
6442 tx_ring
->tx_buffer_info
= vmalloc(size
);
6443 if (!tx_ring
->tx_buffer_info
)
6446 /* round up to nearest 4K */
6447 tx_ring
->size
= tx_ring
->count
* sizeof(union ixgbe_adv_tx_desc
);
6448 tx_ring
->size
= ALIGN(tx_ring
->size
, 4096);
6450 set_dev_node(dev
, ring_node
);
6451 tx_ring
->desc
= dma_alloc_coherent(dev
,
6455 set_dev_node(dev
, orig_node
);
6457 tx_ring
->desc
= dma_alloc_coherent(dev
, tx_ring
->size
,
6458 &tx_ring
->dma
, GFP_KERNEL
);
6462 tx_ring
->next_to_use
= 0;
6463 tx_ring
->next_to_clean
= 0;
6467 vfree(tx_ring
->tx_buffer_info
);
6468 tx_ring
->tx_buffer_info
= NULL
;
6469 dev_err(dev
, "Unable to allocate memory for the Tx descriptor ring\n");
6474 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
6475 * @adapter: board private structure
6477 * If this function returns with an error, then it's possible one or
6478 * more of the rings is populated (while the rest are not). It is the
6479 * callers duty to clean those orphaned rings.
6481 * Return 0 on success, negative on failure
6483 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter
*adapter
)
6485 int i
, j
= 0, err
= 0;
6487 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
6488 err
= ixgbe_setup_tx_resources(adapter
->tx_ring
[i
]);
6492 e_err(probe
, "Allocation for Tx Queue %u failed\n", i
);
6495 for (j
= 0; j
< adapter
->num_xdp_queues
; j
++) {
6496 err
= ixgbe_setup_tx_resources(adapter
->xdp_ring
[j
]);
6500 e_err(probe
, "Allocation for Tx Queue %u failed\n", j
);
6506 /* rewind the index freeing the rings as we go */
6508 ixgbe_free_tx_resources(adapter
->xdp_ring
[j
]);
6510 ixgbe_free_tx_resources(adapter
->tx_ring
[i
]);
6515 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
6516 * @adapter: pointer to ixgbe_adapter
6517 * @rx_ring: rx descriptor ring (for a specific queue) to setup
6519 * Returns 0 on success, negative on failure
6521 int ixgbe_setup_rx_resources(struct ixgbe_adapter
*adapter
,
6522 struct ixgbe_ring
*rx_ring
)
6524 struct device
*dev
= rx_ring
->dev
;
6525 int orig_node
= dev_to_node(dev
);
6526 int ring_node
= NUMA_NO_NODE
;
6529 size
= sizeof(struct ixgbe_rx_buffer
) * rx_ring
->count
;
6531 if (rx_ring
->q_vector
)
6532 ring_node
= rx_ring
->q_vector
->numa_node
;
6534 rx_ring
->rx_buffer_info
= vmalloc_node(size
, ring_node
);
6535 if (!rx_ring
->rx_buffer_info
)
6536 rx_ring
->rx_buffer_info
= vmalloc(size
);
6537 if (!rx_ring
->rx_buffer_info
)
6540 /* Round up to nearest 4K */
6541 rx_ring
->size
= rx_ring
->count
* sizeof(union ixgbe_adv_rx_desc
);
6542 rx_ring
->size
= ALIGN(rx_ring
->size
, 4096);
6544 set_dev_node(dev
, ring_node
);
6545 rx_ring
->desc
= dma_alloc_coherent(dev
,
6549 set_dev_node(dev
, orig_node
);
6551 rx_ring
->desc
= dma_alloc_coherent(dev
, rx_ring
->size
,
6552 &rx_ring
->dma
, GFP_KERNEL
);
6556 rx_ring
->next_to_clean
= 0;
6557 rx_ring
->next_to_use
= 0;
6559 /* XDP RX-queue info */
6560 if (xdp_rxq_info_reg(&rx_ring
->xdp_rxq
, adapter
->netdev
,
6561 rx_ring
->queue_index
) < 0)
6564 rx_ring
->xdp_prog
= adapter
->xdp_prog
;
6568 vfree(rx_ring
->rx_buffer_info
);
6569 rx_ring
->rx_buffer_info
= NULL
;
6570 dev_err(dev
, "Unable to allocate memory for the Rx descriptor ring\n");
6575 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
6576 * @adapter: board private structure
6578 * If this function returns with an error, then it's possible one or
6579 * more of the rings is populated (while the rest are not). It is the
6580 * callers duty to clean those orphaned rings.
6582 * Return 0 on success, negative on failure
6584 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter
*adapter
)
6588 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
6589 err
= ixgbe_setup_rx_resources(adapter
, adapter
->rx_ring
[i
]);
6593 e_err(probe
, "Allocation for Rx Queue %u failed\n", i
);
6598 err
= ixgbe_setup_fcoe_ddp_resources(adapter
);
6603 /* rewind the index freeing the rings as we go */
6605 ixgbe_free_rx_resources(adapter
->rx_ring
[i
]);
6610 * ixgbe_free_tx_resources - Free Tx Resources per Queue
6611 * @tx_ring: Tx descriptor ring for a specific queue
6613 * Free all transmit software resources
6615 void ixgbe_free_tx_resources(struct ixgbe_ring
*tx_ring
)
6617 ixgbe_clean_tx_ring(tx_ring
);
6619 vfree(tx_ring
->tx_buffer_info
);
6620 tx_ring
->tx_buffer_info
= NULL
;
6622 /* if not set, then don't free */
6626 dma_free_coherent(tx_ring
->dev
, tx_ring
->size
,
6627 tx_ring
->desc
, tx_ring
->dma
);
6629 tx_ring
->desc
= NULL
;
6633 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
6634 * @adapter: board private structure
6636 * Free all transmit software resources
6638 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter
*adapter
)
6642 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
6643 if (adapter
->tx_ring
[i
]->desc
)
6644 ixgbe_free_tx_resources(adapter
->tx_ring
[i
]);
6645 for (i
= 0; i
< adapter
->num_xdp_queues
; i
++)
6646 if (adapter
->xdp_ring
[i
]->desc
)
6647 ixgbe_free_tx_resources(adapter
->xdp_ring
[i
]);
6651 * ixgbe_free_rx_resources - Free Rx Resources
6652 * @rx_ring: ring to clean the resources from
6654 * Free all receive software resources
6656 void ixgbe_free_rx_resources(struct ixgbe_ring
*rx_ring
)
6658 ixgbe_clean_rx_ring(rx_ring
);
6660 rx_ring
->xdp_prog
= NULL
;
6661 xdp_rxq_info_unreg(&rx_ring
->xdp_rxq
);
6662 vfree(rx_ring
->rx_buffer_info
);
6663 rx_ring
->rx_buffer_info
= NULL
;
6665 /* if not set, then don't free */
6669 dma_free_coherent(rx_ring
->dev
, rx_ring
->size
,
6670 rx_ring
->desc
, rx_ring
->dma
);
6672 rx_ring
->desc
= NULL
;
6676 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
6677 * @adapter: board private structure
6679 * Free all receive software resources
6681 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter
*adapter
)
6686 ixgbe_free_fcoe_ddp_resources(adapter
);
6689 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
6690 if (adapter
->rx_ring
[i
]->desc
)
6691 ixgbe_free_rx_resources(adapter
->rx_ring
[i
]);
6695 * ixgbe_change_mtu - Change the Maximum Transfer Unit
6696 * @netdev: network interface device structure
6697 * @new_mtu: new value for maximum frame size
6699 * Returns 0 on success, negative on failure
6701 static int ixgbe_change_mtu(struct net_device
*netdev
, int new_mtu
)
6703 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
6705 if (adapter
->xdp_prog
) {
6706 int new_frame_size
= new_mtu
+ ETH_HLEN
+ ETH_FCS_LEN
+
6710 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
6711 struct ixgbe_ring
*ring
= adapter
->rx_ring
[i
];
6713 if (new_frame_size
> ixgbe_rx_bufsz(ring
)) {
6714 e_warn(probe
, "Requested MTU size is not supported with XDP\n");
6721 * For 82599EB we cannot allow legacy VFs to enable their receive
6722 * paths when MTU greater than 1500 is configured. So display a
6723 * warning that legacy VFs will be disabled.
6725 if ((adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
) &&
6726 (adapter
->hw
.mac
.type
== ixgbe_mac_82599EB
) &&
6727 (new_mtu
> ETH_DATA_LEN
))
6728 e_warn(probe
, "Setting MTU > 1500 will disable legacy VFs\n");
6730 netdev_dbg(netdev
, "changing MTU from %d to %d\n",
6731 netdev
->mtu
, new_mtu
);
6733 /* must set new MTU before calling down or up */
6734 netdev
->mtu
= new_mtu
;
6736 if (netif_running(netdev
))
6737 ixgbe_reinit_locked(adapter
);
6743 * ixgbe_open - Called when a network interface is made active
6744 * @netdev: network interface device structure
6746 * Returns 0 on success, negative value on failure
6748 * The open entry point is called when a network interface is made
6749 * active by the system (IFF_UP). At this point all resources needed
6750 * for transmit and receive operations are allocated, the interrupt
6751 * handler is registered with the OS, the watchdog timer is started,
6752 * and the stack is notified that the interface is ready.
6754 int ixgbe_open(struct net_device
*netdev
)
6756 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
6757 struct ixgbe_hw
*hw
= &adapter
->hw
;
6760 /* disallow open during test */
6761 if (test_bit(__IXGBE_TESTING
, &adapter
->state
))
6764 netif_carrier_off(netdev
);
6766 /* allocate transmit descriptors */
6767 err
= ixgbe_setup_all_tx_resources(adapter
);
6771 /* allocate receive descriptors */
6772 err
= ixgbe_setup_all_rx_resources(adapter
);
6776 ixgbe_configure(adapter
);
6778 err
= ixgbe_request_irq(adapter
);
6782 /* Notify the stack of the actual queue counts. */
6783 queues
= adapter
->num_tx_queues
;
6784 err
= netif_set_real_num_tx_queues(netdev
, queues
);
6786 goto err_set_queues
;
6788 queues
= adapter
->num_rx_queues
;
6789 err
= netif_set_real_num_rx_queues(netdev
, queues
);
6791 goto err_set_queues
;
6793 ixgbe_ptp_init(adapter
);
6795 ixgbe_up_complete(adapter
);
6797 ixgbe_clear_udp_tunnel_port(adapter
, IXGBE_VXLANCTRL_ALL_UDPPORT_MASK
);
6798 udp_tunnel_get_rx_info(netdev
);
6803 ixgbe_free_irq(adapter
);
6805 ixgbe_free_all_rx_resources(adapter
);
6806 if (hw
->phy
.ops
.set_phy_power
&& !adapter
->wol
)
6807 hw
->phy
.ops
.set_phy_power(&adapter
->hw
, false);
6809 ixgbe_free_all_tx_resources(adapter
);
6811 ixgbe_reset(adapter
);
6816 static void ixgbe_close_suspend(struct ixgbe_adapter
*adapter
)
6818 ixgbe_ptp_suspend(adapter
);
6820 if (adapter
->hw
.phy
.ops
.enter_lplu
) {
6821 adapter
->hw
.phy
.reset_disable
= true;
6822 ixgbe_down(adapter
);
6823 adapter
->hw
.phy
.ops
.enter_lplu(&adapter
->hw
);
6824 adapter
->hw
.phy
.reset_disable
= false;
6826 ixgbe_down(adapter
);
6829 ixgbe_free_irq(adapter
);
6831 ixgbe_free_all_tx_resources(adapter
);
6832 ixgbe_free_all_rx_resources(adapter
);
6836 * ixgbe_close - Disables a network interface
6837 * @netdev: network interface device structure
6839 * Returns 0, this is not allowed to fail
6841 * The close entry point is called when an interface is de-activated
6842 * by the OS. The hardware is still under the drivers control, but
6843 * needs to be disabled. A global MAC reset is issued to stop the
6844 * hardware, and all transmit and receive resources are freed.
6846 int ixgbe_close(struct net_device
*netdev
)
6848 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
6850 ixgbe_ptp_stop(adapter
);
6852 if (netif_device_present(netdev
))
6853 ixgbe_close_suspend(adapter
);
6855 ixgbe_fdir_filter_exit(adapter
);
6857 ixgbe_release_hw_control(adapter
);
6863 static int ixgbe_resume(struct pci_dev
*pdev
)
6865 struct ixgbe_adapter
*adapter
= pci_get_drvdata(pdev
);
6866 struct net_device
*netdev
= adapter
->netdev
;
6869 adapter
->hw
.hw_addr
= adapter
->io_addr
;
6870 pci_set_power_state(pdev
, PCI_D0
);
6871 pci_restore_state(pdev
);
6873 * pci_restore_state clears dev->state_saved so call
6874 * pci_save_state to restore it.
6876 pci_save_state(pdev
);
6878 err
= pci_enable_device_mem(pdev
);
6880 e_dev_err("Cannot enable PCI device from suspend\n");
6883 smp_mb__before_atomic();
6884 clear_bit(__IXGBE_DISABLED
, &adapter
->state
);
6885 pci_set_master(pdev
);
6887 pci_wake_from_d3(pdev
, false);
6889 ixgbe_reset(adapter
);
6891 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_WUS
, ~0);
6894 err
= ixgbe_init_interrupt_scheme(adapter
);
6895 if (!err
&& netif_running(netdev
))
6896 err
= ixgbe_open(netdev
);
6900 netif_device_attach(netdev
);
6905 #endif /* CONFIG_PM */
6907 static int __ixgbe_shutdown(struct pci_dev
*pdev
, bool *enable_wake
)
6909 struct ixgbe_adapter
*adapter
= pci_get_drvdata(pdev
);
6910 struct net_device
*netdev
= adapter
->netdev
;
6911 struct ixgbe_hw
*hw
= &adapter
->hw
;
6913 u32 wufc
= adapter
->wol
;
6919 netif_device_detach(netdev
);
6921 if (netif_running(netdev
))
6922 ixgbe_close_suspend(adapter
);
6924 ixgbe_clear_interrupt_scheme(adapter
);
6928 retval
= pci_save_state(pdev
);
6933 if (hw
->mac
.ops
.stop_link_on_d3
)
6934 hw
->mac
.ops
.stop_link_on_d3(hw
);
6939 ixgbe_set_rx_mode(netdev
);
6941 /* enable the optics for 82599 SFP+ fiber as we can WoL */
6942 if (hw
->mac
.ops
.enable_tx_laser
)
6943 hw
->mac
.ops
.enable_tx_laser(hw
);
6945 /* enable the reception of multicast packets */
6946 fctrl
= IXGBE_READ_REG(hw
, IXGBE_FCTRL
);
6947 fctrl
|= IXGBE_FCTRL_MPE
;
6948 IXGBE_WRITE_REG(hw
, IXGBE_FCTRL
, fctrl
);
6950 ctrl
= IXGBE_READ_REG(hw
, IXGBE_CTRL
);
6951 ctrl
|= IXGBE_CTRL_GIO_DIS
;
6952 IXGBE_WRITE_REG(hw
, IXGBE_CTRL
, ctrl
);
6954 IXGBE_WRITE_REG(hw
, IXGBE_WUFC
, wufc
);
6956 IXGBE_WRITE_REG(hw
, IXGBE_WUC
, 0);
6957 IXGBE_WRITE_REG(hw
, IXGBE_WUFC
, 0);
6960 switch (hw
->mac
.type
) {
6961 case ixgbe_mac_82598EB
:
6962 pci_wake_from_d3(pdev
, false);
6964 case ixgbe_mac_82599EB
:
6965 case ixgbe_mac_X540
:
6966 case ixgbe_mac_X550
:
6967 case ixgbe_mac_X550EM_x
:
6968 case ixgbe_mac_x550em_a
:
6969 pci_wake_from_d3(pdev
, !!wufc
);
6975 *enable_wake
= !!wufc
;
6976 if (hw
->phy
.ops
.set_phy_power
&& !*enable_wake
)
6977 hw
->phy
.ops
.set_phy_power(hw
, false);
6979 ixgbe_release_hw_control(adapter
);
6981 if (!test_and_set_bit(__IXGBE_DISABLED
, &adapter
->state
))
6982 pci_disable_device(pdev
);
6988 static int ixgbe_suspend(struct pci_dev
*pdev
, pm_message_t state
)
6993 retval
= __ixgbe_shutdown(pdev
, &wake
);
6998 pci_prepare_to_sleep(pdev
);
7000 pci_wake_from_d3(pdev
, false);
7001 pci_set_power_state(pdev
, PCI_D3hot
);
7006 #endif /* CONFIG_PM */
7008 static void ixgbe_shutdown(struct pci_dev
*pdev
)
7012 __ixgbe_shutdown(pdev
, &wake
);
7014 if (system_state
== SYSTEM_POWER_OFF
) {
7015 pci_wake_from_d3(pdev
, wake
);
7016 pci_set_power_state(pdev
, PCI_D3hot
);
7021 * ixgbe_update_stats - Update the board statistics counters.
7022 * @adapter: board private structure
7024 void ixgbe_update_stats(struct ixgbe_adapter
*adapter
)
7026 struct net_device
*netdev
= adapter
->netdev
;
7027 struct ixgbe_hw
*hw
= &adapter
->hw
;
7028 struct ixgbe_hw_stats
*hwstats
= &adapter
->stats
;
7030 u32 i
, missed_rx
= 0, mpc
, bprc
, lxon
, lxoff
, xon_off_tot
;
7031 u64 non_eop_descs
= 0, restart_queue
= 0, tx_busy
= 0;
7032 u64 alloc_rx_page_failed
= 0, alloc_rx_buff_failed
= 0;
7033 u64 alloc_rx_page
= 0;
7034 u64 bytes
= 0, packets
= 0, hw_csum_rx_error
= 0;
7036 if (test_bit(__IXGBE_DOWN
, &adapter
->state
) ||
7037 test_bit(__IXGBE_RESETTING
, &adapter
->state
))
7040 if (adapter
->flags2
& IXGBE_FLAG2_RSC_ENABLED
) {
7043 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
7044 rsc_count
+= adapter
->rx_ring
[i
]->rx_stats
.rsc_count
;
7045 rsc_flush
+= adapter
->rx_ring
[i
]->rx_stats
.rsc_flush
;
7047 adapter
->rsc_total_count
= rsc_count
;
7048 adapter
->rsc_total_flush
= rsc_flush
;
7051 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
7052 struct ixgbe_ring
*rx_ring
= READ_ONCE(adapter
->rx_ring
[i
]);
7056 non_eop_descs
+= rx_ring
->rx_stats
.non_eop_descs
;
7057 alloc_rx_page
+= rx_ring
->rx_stats
.alloc_rx_page
;
7058 alloc_rx_page_failed
+= rx_ring
->rx_stats
.alloc_rx_page_failed
;
7059 alloc_rx_buff_failed
+= rx_ring
->rx_stats
.alloc_rx_buff_failed
;
7060 hw_csum_rx_error
+= rx_ring
->rx_stats
.csum_err
;
7061 bytes
+= rx_ring
->stats
.bytes
;
7062 packets
+= rx_ring
->stats
.packets
;
7064 adapter
->non_eop_descs
= non_eop_descs
;
7065 adapter
->alloc_rx_page
= alloc_rx_page
;
7066 adapter
->alloc_rx_page_failed
= alloc_rx_page_failed
;
7067 adapter
->alloc_rx_buff_failed
= alloc_rx_buff_failed
;
7068 adapter
->hw_csum_rx_error
= hw_csum_rx_error
;
7069 netdev
->stats
.rx_bytes
= bytes
;
7070 netdev
->stats
.rx_packets
= packets
;
7074 /* gather some stats to the adapter struct that are per queue */
7075 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
7076 struct ixgbe_ring
*tx_ring
= READ_ONCE(adapter
->tx_ring
[i
]);
7080 restart_queue
+= tx_ring
->tx_stats
.restart_queue
;
7081 tx_busy
+= tx_ring
->tx_stats
.tx_busy
;
7082 bytes
+= tx_ring
->stats
.bytes
;
7083 packets
+= tx_ring
->stats
.packets
;
7085 for (i
= 0; i
< adapter
->num_xdp_queues
; i
++) {
7086 struct ixgbe_ring
*xdp_ring
= READ_ONCE(adapter
->xdp_ring
[i
]);
7090 restart_queue
+= xdp_ring
->tx_stats
.restart_queue
;
7091 tx_busy
+= xdp_ring
->tx_stats
.tx_busy
;
7092 bytes
+= xdp_ring
->stats
.bytes
;
7093 packets
+= xdp_ring
->stats
.packets
;
7095 adapter
->restart_queue
= restart_queue
;
7096 adapter
->tx_busy
= tx_busy
;
7097 netdev
->stats
.tx_bytes
= bytes
;
7098 netdev
->stats
.tx_packets
= packets
;
7100 hwstats
->crcerrs
+= IXGBE_READ_REG(hw
, IXGBE_CRCERRS
);
7102 /* 8 register reads */
7103 for (i
= 0; i
< 8; i
++) {
7104 /* for packet buffers not used, the register should read 0 */
7105 mpc
= IXGBE_READ_REG(hw
, IXGBE_MPC(i
));
7107 hwstats
->mpc
[i
] += mpc
;
7108 total_mpc
+= hwstats
->mpc
[i
];
7109 hwstats
->pxontxc
[i
] += IXGBE_READ_REG(hw
, IXGBE_PXONTXC(i
));
7110 hwstats
->pxofftxc
[i
] += IXGBE_READ_REG(hw
, IXGBE_PXOFFTXC(i
));
7111 switch (hw
->mac
.type
) {
7112 case ixgbe_mac_82598EB
:
7113 hwstats
->rnbc
[i
] += IXGBE_READ_REG(hw
, IXGBE_RNBC(i
));
7114 hwstats
->qbtc
[i
] += IXGBE_READ_REG(hw
, IXGBE_QBTC(i
));
7115 hwstats
->qbrc
[i
] += IXGBE_READ_REG(hw
, IXGBE_QBRC(i
));
7116 hwstats
->pxonrxc
[i
] +=
7117 IXGBE_READ_REG(hw
, IXGBE_PXONRXC(i
));
7119 case ixgbe_mac_82599EB
:
7120 case ixgbe_mac_X540
:
7121 case ixgbe_mac_X550
:
7122 case ixgbe_mac_X550EM_x
:
7123 case ixgbe_mac_x550em_a
:
7124 hwstats
->pxonrxc
[i
] +=
7125 IXGBE_READ_REG(hw
, IXGBE_PXONRXCNT(i
));
7132 /*16 register reads */
7133 for (i
= 0; i
< 16; i
++) {
7134 hwstats
->qptc
[i
] += IXGBE_READ_REG(hw
, IXGBE_QPTC(i
));
7135 hwstats
->qprc
[i
] += IXGBE_READ_REG(hw
, IXGBE_QPRC(i
));
7136 if ((hw
->mac
.type
== ixgbe_mac_82599EB
) ||
7137 (hw
->mac
.type
== ixgbe_mac_X540
) ||
7138 (hw
->mac
.type
== ixgbe_mac_X550
) ||
7139 (hw
->mac
.type
== ixgbe_mac_X550EM_x
) ||
7140 (hw
->mac
.type
== ixgbe_mac_x550em_a
)) {
7141 hwstats
->qbtc
[i
] += IXGBE_READ_REG(hw
, IXGBE_QBTC_L(i
));
7142 IXGBE_READ_REG(hw
, IXGBE_QBTC_H(i
)); /* to clear */
7143 hwstats
->qbrc
[i
] += IXGBE_READ_REG(hw
, IXGBE_QBRC_L(i
));
7144 IXGBE_READ_REG(hw
, IXGBE_QBRC_H(i
)); /* to clear */
7148 hwstats
->gprc
+= IXGBE_READ_REG(hw
, IXGBE_GPRC
);
7149 /* work around hardware counting issue */
7150 hwstats
->gprc
-= missed_rx
;
7152 ixgbe_update_xoff_received(adapter
);
7154 /* 82598 hardware only has a 32 bit counter in the high register */
7155 switch (hw
->mac
.type
) {
7156 case ixgbe_mac_82598EB
:
7157 hwstats
->lxonrxc
+= IXGBE_READ_REG(hw
, IXGBE_LXONRXC
);
7158 hwstats
->gorc
+= IXGBE_READ_REG(hw
, IXGBE_GORCH
);
7159 hwstats
->gotc
+= IXGBE_READ_REG(hw
, IXGBE_GOTCH
);
7160 hwstats
->tor
+= IXGBE_READ_REG(hw
, IXGBE_TORH
);
7162 case ixgbe_mac_X540
:
7163 case ixgbe_mac_X550
:
7164 case ixgbe_mac_X550EM_x
:
7165 case ixgbe_mac_x550em_a
:
7166 /* OS2BMC stats are X540 and later */
7167 hwstats
->o2bgptc
+= IXGBE_READ_REG(hw
, IXGBE_O2BGPTC
);
7168 hwstats
->o2bspc
+= IXGBE_READ_REG(hw
, IXGBE_O2BSPC
);
7169 hwstats
->b2ospc
+= IXGBE_READ_REG(hw
, IXGBE_B2OSPC
);
7170 hwstats
->b2ogprc
+= IXGBE_READ_REG(hw
, IXGBE_B2OGPRC
);
7172 case ixgbe_mac_82599EB
:
7173 for (i
= 0; i
< 16; i
++)
7174 adapter
->hw_rx_no_dma_resources
+=
7175 IXGBE_READ_REG(hw
, IXGBE_QPRDC(i
));
7176 hwstats
->gorc
+= IXGBE_READ_REG(hw
, IXGBE_GORCL
);
7177 IXGBE_READ_REG(hw
, IXGBE_GORCH
); /* to clear */
7178 hwstats
->gotc
+= IXGBE_READ_REG(hw
, IXGBE_GOTCL
);
7179 IXGBE_READ_REG(hw
, IXGBE_GOTCH
); /* to clear */
7180 hwstats
->tor
+= IXGBE_READ_REG(hw
, IXGBE_TORL
);
7181 IXGBE_READ_REG(hw
, IXGBE_TORH
); /* to clear */
7182 hwstats
->lxonrxc
+= IXGBE_READ_REG(hw
, IXGBE_LXONRXCNT
);
7183 hwstats
->fdirmatch
+= IXGBE_READ_REG(hw
, IXGBE_FDIRMATCH
);
7184 hwstats
->fdirmiss
+= IXGBE_READ_REG(hw
, IXGBE_FDIRMISS
);
7186 hwstats
->fccrc
+= IXGBE_READ_REG(hw
, IXGBE_FCCRC
);
7187 hwstats
->fcoerpdc
+= IXGBE_READ_REG(hw
, IXGBE_FCOERPDC
);
7188 hwstats
->fcoeprc
+= IXGBE_READ_REG(hw
, IXGBE_FCOEPRC
);
7189 hwstats
->fcoeptc
+= IXGBE_READ_REG(hw
, IXGBE_FCOEPTC
);
7190 hwstats
->fcoedwrc
+= IXGBE_READ_REG(hw
, IXGBE_FCOEDWRC
);
7191 hwstats
->fcoedwtc
+= IXGBE_READ_REG(hw
, IXGBE_FCOEDWTC
);
7192 /* Add up per cpu counters for total ddp aloc fail */
7193 if (adapter
->fcoe
.ddp_pool
) {
7194 struct ixgbe_fcoe
*fcoe
= &adapter
->fcoe
;
7195 struct ixgbe_fcoe_ddp_pool
*ddp_pool
;
7197 u64 noddp
= 0, noddp_ext_buff
= 0;
7198 for_each_possible_cpu(cpu
) {
7199 ddp_pool
= per_cpu_ptr(fcoe
->ddp_pool
, cpu
);
7200 noddp
+= ddp_pool
->noddp
;
7201 noddp_ext_buff
+= ddp_pool
->noddp_ext_buff
;
7203 hwstats
->fcoe_noddp
= noddp
;
7204 hwstats
->fcoe_noddp_ext_buff
= noddp_ext_buff
;
7206 #endif /* IXGBE_FCOE */
7211 bprc
= IXGBE_READ_REG(hw
, IXGBE_BPRC
);
7212 hwstats
->bprc
+= bprc
;
7213 hwstats
->mprc
+= IXGBE_READ_REG(hw
, IXGBE_MPRC
);
7214 if (hw
->mac
.type
== ixgbe_mac_82598EB
)
7215 hwstats
->mprc
-= bprc
;
7216 hwstats
->roc
+= IXGBE_READ_REG(hw
, IXGBE_ROC
);
7217 hwstats
->prc64
+= IXGBE_READ_REG(hw
, IXGBE_PRC64
);
7218 hwstats
->prc127
+= IXGBE_READ_REG(hw
, IXGBE_PRC127
);
7219 hwstats
->prc255
+= IXGBE_READ_REG(hw
, IXGBE_PRC255
);
7220 hwstats
->prc511
+= IXGBE_READ_REG(hw
, IXGBE_PRC511
);
7221 hwstats
->prc1023
+= IXGBE_READ_REG(hw
, IXGBE_PRC1023
);
7222 hwstats
->prc1522
+= IXGBE_READ_REG(hw
, IXGBE_PRC1522
);
7223 hwstats
->rlec
+= IXGBE_READ_REG(hw
, IXGBE_RLEC
);
7224 lxon
= IXGBE_READ_REG(hw
, IXGBE_LXONTXC
);
7225 hwstats
->lxontxc
+= lxon
;
7226 lxoff
= IXGBE_READ_REG(hw
, IXGBE_LXOFFTXC
);
7227 hwstats
->lxofftxc
+= lxoff
;
7228 hwstats
->gptc
+= IXGBE_READ_REG(hw
, IXGBE_GPTC
);
7229 hwstats
->mptc
+= IXGBE_READ_REG(hw
, IXGBE_MPTC
);
7231 * 82598 errata - tx of flow control packets is included in tx counters
7233 xon_off_tot
= lxon
+ lxoff
;
7234 hwstats
->gptc
-= xon_off_tot
;
7235 hwstats
->mptc
-= xon_off_tot
;
7236 hwstats
->gotc
-= (xon_off_tot
* (ETH_ZLEN
+ ETH_FCS_LEN
));
7237 hwstats
->ruc
+= IXGBE_READ_REG(hw
, IXGBE_RUC
);
7238 hwstats
->rfc
+= IXGBE_READ_REG(hw
, IXGBE_RFC
);
7239 hwstats
->rjc
+= IXGBE_READ_REG(hw
, IXGBE_RJC
);
7240 hwstats
->tpr
+= IXGBE_READ_REG(hw
, IXGBE_TPR
);
7241 hwstats
->ptc64
+= IXGBE_READ_REG(hw
, IXGBE_PTC64
);
7242 hwstats
->ptc64
-= xon_off_tot
;
7243 hwstats
->ptc127
+= IXGBE_READ_REG(hw
, IXGBE_PTC127
);
7244 hwstats
->ptc255
+= IXGBE_READ_REG(hw
, IXGBE_PTC255
);
7245 hwstats
->ptc511
+= IXGBE_READ_REG(hw
, IXGBE_PTC511
);
7246 hwstats
->ptc1023
+= IXGBE_READ_REG(hw
, IXGBE_PTC1023
);
7247 hwstats
->ptc1522
+= IXGBE_READ_REG(hw
, IXGBE_PTC1522
);
7248 hwstats
->bptc
+= IXGBE_READ_REG(hw
, IXGBE_BPTC
);
7250 /* Fill out the OS statistics structure */
7251 netdev
->stats
.multicast
= hwstats
->mprc
;
7254 netdev
->stats
.rx_errors
= hwstats
->crcerrs
+ hwstats
->rlec
;
7255 netdev
->stats
.rx_dropped
= 0;
7256 netdev
->stats
.rx_length_errors
= hwstats
->rlec
;
7257 netdev
->stats
.rx_crc_errors
= hwstats
->crcerrs
;
7258 netdev
->stats
.rx_missed_errors
= total_mpc
;
7262 * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
7263 * @adapter: pointer to the device adapter structure
7265 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter
*adapter
)
7267 struct ixgbe_hw
*hw
= &adapter
->hw
;
7270 if (!(adapter
->flags2
& IXGBE_FLAG2_FDIR_REQUIRES_REINIT
))
7273 adapter
->flags2
&= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT
;
7275 /* if interface is down do nothing */
7276 if (test_bit(__IXGBE_DOWN
, &adapter
->state
))
7279 /* do nothing if we are not using signature filters */
7280 if (!(adapter
->flags
& IXGBE_FLAG_FDIR_HASH_CAPABLE
))
7283 adapter
->fdir_overflow
++;
7285 if (ixgbe_reinit_fdir_tables_82599(hw
) == 0) {
7286 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
7287 set_bit(__IXGBE_TX_FDIR_INIT_DONE
,
7288 &(adapter
->tx_ring
[i
]->state
));
7289 for (i
= 0; i
< adapter
->num_xdp_queues
; i
++)
7290 set_bit(__IXGBE_TX_FDIR_INIT_DONE
,
7291 &adapter
->xdp_ring
[i
]->state
);
7292 /* re-enable flow director interrupts */
7293 IXGBE_WRITE_REG(hw
, IXGBE_EIMS
, IXGBE_EIMS_FLOW_DIR
);
7295 e_err(probe
, "failed to finish FDIR re-initialization, "
7296 "ignored adding FDIR ATR filters\n");
7301 * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
7302 * @adapter: pointer to the device adapter structure
7304 * This function serves two purposes. First it strobes the interrupt lines
7305 * in order to make certain interrupts are occurring. Secondly it sets the
7306 * bits needed to check for TX hangs. As a result we should immediately
7307 * determine if a hang has occurred.
7309 static void ixgbe_check_hang_subtask(struct ixgbe_adapter
*adapter
)
7311 struct ixgbe_hw
*hw
= &adapter
->hw
;
7315 /* If we're down, removing or resetting, just bail */
7316 if (test_bit(__IXGBE_DOWN
, &adapter
->state
) ||
7317 test_bit(__IXGBE_REMOVING
, &adapter
->state
) ||
7318 test_bit(__IXGBE_RESETTING
, &adapter
->state
))
7321 /* Force detection of hung controller */
7322 if (netif_carrier_ok(adapter
->netdev
)) {
7323 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
7324 set_check_for_tx_hang(adapter
->tx_ring
[i
]);
7325 for (i
= 0; i
< adapter
->num_xdp_queues
; i
++)
7326 set_check_for_tx_hang(adapter
->xdp_ring
[i
]);
7329 if (!(adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
)) {
7331 * for legacy and MSI interrupts don't set any bits
7332 * that are enabled for EIAM, because this operation
7333 * would set *both* EIMS and EICS for any bit in EIAM
7335 IXGBE_WRITE_REG(hw
, IXGBE_EICS
,
7336 (IXGBE_EICS_TCP_TIMER
| IXGBE_EICS_OTHER
));
7338 /* get one bit for every active tx/rx interrupt vector */
7339 for (i
= 0; i
< adapter
->num_q_vectors
; i
++) {
7340 struct ixgbe_q_vector
*qv
= adapter
->q_vector
[i
];
7341 if (qv
->rx
.ring
|| qv
->tx
.ring
)
7346 /* Cause software interrupt to ensure rings are cleaned */
7347 ixgbe_irq_rearm_queues(adapter
, eics
);
7351 * ixgbe_watchdog_update_link - update the link status
7352 * @adapter: pointer to the device adapter structure
7354 static void ixgbe_watchdog_update_link(struct ixgbe_adapter
*adapter
)
7356 struct ixgbe_hw
*hw
= &adapter
->hw
;
7357 u32 link_speed
= adapter
->link_speed
;
7358 bool link_up
= adapter
->link_up
;
7359 bool pfc_en
= adapter
->dcb_cfg
.pfc_mode_enable
;
7361 if (!(adapter
->flags
& IXGBE_FLAG_NEED_LINK_UPDATE
))
7364 if (hw
->mac
.ops
.check_link
) {
7365 hw
->mac
.ops
.check_link(hw
, &link_speed
, &link_up
, false);
7367 /* always assume link is up, if no check link function */
7368 link_speed
= IXGBE_LINK_SPEED_10GB_FULL
;
7372 if (adapter
->ixgbe_ieee_pfc
)
7373 pfc_en
|= !!(adapter
->ixgbe_ieee_pfc
->pfc_en
);
7375 if (link_up
&& !((adapter
->flags
& IXGBE_FLAG_DCB_ENABLED
) && pfc_en
)) {
7376 hw
->mac
.ops
.fc_enable(hw
);
7377 ixgbe_set_rx_drop_en(adapter
);
7381 time_after(jiffies
, (adapter
->link_check_timeout
+
7382 IXGBE_TRY_LINK_TIMEOUT
))) {
7383 adapter
->flags
&= ~IXGBE_FLAG_NEED_LINK_UPDATE
;
7384 IXGBE_WRITE_REG(hw
, IXGBE_EIMS
, IXGBE_EIMC_LSC
);
7385 IXGBE_WRITE_FLUSH(hw
);
7388 adapter
->link_up
= link_up
;
7389 adapter
->link_speed
= link_speed
;
7392 static void ixgbe_update_default_up(struct ixgbe_adapter
*adapter
)
7394 #ifdef CONFIG_IXGBE_DCB
7395 struct net_device
*netdev
= adapter
->netdev
;
7396 struct dcb_app app
= {
7397 .selector
= IEEE_8021QAZ_APP_SEL_ETHERTYPE
,
7402 if (adapter
->dcbx_cap
& DCB_CAP_DCBX_VER_IEEE
)
7403 up
= dcb_ieee_getapp_mask(netdev
, &app
);
7405 adapter
->default_up
= (up
> 1) ? (ffs(up
) - 1) : 0;
7410 * ixgbe_watchdog_link_is_up - update netif_carrier status and
7411 * print link up message
7412 * @adapter: pointer to the device adapter structure
7414 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter
*adapter
)
7416 struct net_device
*netdev
= adapter
->netdev
;
7417 struct ixgbe_hw
*hw
= &adapter
->hw
;
7418 u32 link_speed
= adapter
->link_speed
;
7419 const char *speed_str
;
7420 bool flow_rx
, flow_tx
;
7422 /* only continue if link was previously down */
7423 if (netif_carrier_ok(netdev
))
7426 adapter
->flags2
&= ~IXGBE_FLAG2_SEARCH_FOR_SFP
;
7428 switch (hw
->mac
.type
) {
7429 case ixgbe_mac_82598EB
: {
7430 u32 frctl
= IXGBE_READ_REG(hw
, IXGBE_FCTRL
);
7431 u32 rmcs
= IXGBE_READ_REG(hw
, IXGBE_RMCS
);
7432 flow_rx
= !!(frctl
& IXGBE_FCTRL_RFCE
);
7433 flow_tx
= !!(rmcs
& IXGBE_RMCS_TFCE_802_3X
);
7436 case ixgbe_mac_X540
:
7437 case ixgbe_mac_X550
:
7438 case ixgbe_mac_X550EM_x
:
7439 case ixgbe_mac_x550em_a
:
7440 case ixgbe_mac_82599EB
: {
7441 u32 mflcn
= IXGBE_READ_REG(hw
, IXGBE_MFLCN
);
7442 u32 fccfg
= IXGBE_READ_REG(hw
, IXGBE_FCCFG
);
7443 flow_rx
= !!(mflcn
& IXGBE_MFLCN_RFCE
);
7444 flow_tx
= !!(fccfg
& IXGBE_FCCFG_TFCE_802_3X
);
7453 adapter
->last_rx_ptp_check
= jiffies
;
7455 if (test_bit(__IXGBE_PTP_RUNNING
, &adapter
->state
))
7456 ixgbe_ptp_start_cyclecounter(adapter
);
7458 switch (link_speed
) {
7459 case IXGBE_LINK_SPEED_10GB_FULL
:
7460 speed_str
= "10 Gbps";
7462 case IXGBE_LINK_SPEED_5GB_FULL
:
7463 speed_str
= "5 Gbps";
7465 case IXGBE_LINK_SPEED_2_5GB_FULL
:
7466 speed_str
= "2.5 Gbps";
7468 case IXGBE_LINK_SPEED_1GB_FULL
:
7469 speed_str
= "1 Gbps";
7471 case IXGBE_LINK_SPEED_100_FULL
:
7472 speed_str
= "100 Mbps";
7474 case IXGBE_LINK_SPEED_10_FULL
:
7475 speed_str
= "10 Mbps";
7478 speed_str
= "unknown speed";
7481 e_info(drv
, "NIC Link is Up %s, Flow Control: %s\n", speed_str
,
7482 ((flow_rx
&& flow_tx
) ? "RX/TX" :
7484 (flow_tx
? "TX" : "None"))));
7486 netif_carrier_on(netdev
);
7487 ixgbe_check_vf_rate_limit(adapter
);
7489 /* enable transmits */
7490 netif_tx_wake_all_queues(adapter
->netdev
);
7492 /* update the default user priority for VFs */
7493 ixgbe_update_default_up(adapter
);
7495 /* ping all the active vfs to let them know link has changed */
7496 ixgbe_ping_all_vfs(adapter
);
7500 * ixgbe_watchdog_link_is_down - update netif_carrier status and
7501 * print link down message
7502 * @adapter: pointer to the adapter structure
7504 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter
*adapter
)
7506 struct net_device
*netdev
= adapter
->netdev
;
7507 struct ixgbe_hw
*hw
= &adapter
->hw
;
7509 adapter
->link_up
= false;
7510 adapter
->link_speed
= 0;
7512 /* only continue if link was up previously */
7513 if (!netif_carrier_ok(netdev
))
7516 /* poll for SFP+ cable when link is down */
7517 if (ixgbe_is_sfp(hw
) && hw
->mac
.type
== ixgbe_mac_82598EB
)
7518 adapter
->flags2
|= IXGBE_FLAG2_SEARCH_FOR_SFP
;
7520 if (test_bit(__IXGBE_PTP_RUNNING
, &adapter
->state
))
7521 ixgbe_ptp_start_cyclecounter(adapter
);
7523 e_info(drv
, "NIC Link is Down\n");
7524 netif_carrier_off(netdev
);
7526 /* ping all the active vfs to let them know link has changed */
7527 ixgbe_ping_all_vfs(adapter
);
7530 static bool ixgbe_ring_tx_pending(struct ixgbe_adapter
*adapter
)
7534 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
7535 struct ixgbe_ring
*tx_ring
= adapter
->tx_ring
[i
];
7537 if (tx_ring
->next_to_use
!= tx_ring
->next_to_clean
)
7541 for (i
= 0; i
< adapter
->num_xdp_queues
; i
++) {
7542 struct ixgbe_ring
*ring
= adapter
->xdp_ring
[i
];
7544 if (ring
->next_to_use
!= ring
->next_to_clean
)
7551 static bool ixgbe_vf_tx_pending(struct ixgbe_adapter
*adapter
)
7553 struct ixgbe_hw
*hw
= &adapter
->hw
;
7554 struct ixgbe_ring_feature
*vmdq
= &adapter
->ring_feature
[RING_F_VMDQ
];
7555 u32 q_per_pool
= __ALIGN_MASK(1, ~vmdq
->mask
);
7559 if (!adapter
->num_vfs
)
7562 /* resetting the PF is only needed for MAC before X550 */
7563 if (hw
->mac
.type
>= ixgbe_mac_X550
)
7566 for (i
= 0; i
< adapter
->num_vfs
; i
++) {
7567 for (j
= 0; j
< q_per_pool
; j
++) {
7570 h
= IXGBE_READ_REG(hw
, IXGBE_PVFTDHN(q_per_pool
, i
, j
));
7571 t
= IXGBE_READ_REG(hw
, IXGBE_PVFTDTN(q_per_pool
, i
, j
));
7582 * ixgbe_watchdog_flush_tx - flush queues on link down
7583 * @adapter: pointer to the device adapter structure
7585 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter
*adapter
)
7587 if (!netif_carrier_ok(adapter
->netdev
)) {
7588 if (ixgbe_ring_tx_pending(adapter
) ||
7589 ixgbe_vf_tx_pending(adapter
)) {
7590 /* We've lost link, so the controller stops DMA,
7591 * but we've got queued Tx work that's never going
7592 * to get done, so reset controller to flush Tx.
7593 * (Do the reset outside of interrupt context).
7595 e_warn(drv
, "initiating reset to clear Tx work after link loss\n");
7596 set_bit(__IXGBE_RESET_REQUESTED
, &adapter
->state
);
7601 #ifdef CONFIG_PCI_IOV
7602 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter
*adapter
)
7604 struct ixgbe_hw
*hw
= &adapter
->hw
;
7605 struct pci_dev
*pdev
= adapter
->pdev
;
7609 if (!(netif_carrier_ok(adapter
->netdev
)))
7612 gpc
= IXGBE_READ_REG(hw
, IXGBE_TXDGPC
);
7613 if (gpc
) /* If incrementing then no need for the check below */
7615 /* Check to see if a bad DMA write target from an errant or
7616 * malicious VF has caused a PCIe error. If so then we can
7617 * issue a VFLR to the offending VF(s) and then resume without
7618 * requesting a full slot reset.
7624 /* check status reg for all VFs owned by this PF */
7625 for (vf
= 0; vf
< adapter
->num_vfs
; ++vf
) {
7626 struct pci_dev
*vfdev
= adapter
->vfinfo
[vf
].vfdev
;
7631 pci_read_config_word(vfdev
, PCI_STATUS
, &status_reg
);
7632 if (status_reg
!= IXGBE_FAILED_READ_CFG_WORD
&&
7633 status_reg
& PCI_STATUS_REC_MASTER_ABORT
)
7638 static void ixgbe_spoof_check(struct ixgbe_adapter
*adapter
)
7642 /* Do not perform spoof check for 82598 or if not in IOV mode */
7643 if (adapter
->hw
.mac
.type
== ixgbe_mac_82598EB
||
7644 adapter
->num_vfs
== 0)
7647 ssvpc
= IXGBE_READ_REG(&adapter
->hw
, IXGBE_SSVPC
);
7650 * ssvpc register is cleared on read, if zero then no
7651 * spoofed packets in the last interval.
7656 e_warn(drv
, "%u Spoofed packets detected\n", ssvpc
);
7659 static void ixgbe_spoof_check(struct ixgbe_adapter __always_unused
*adapter
)
7664 ixgbe_check_for_bad_vf(struct ixgbe_adapter __always_unused
*adapter
)
7667 #endif /* CONFIG_PCI_IOV */
7671 * ixgbe_watchdog_subtask - check and bring link up
7672 * @adapter: pointer to the device adapter structure
7674 static void ixgbe_watchdog_subtask(struct ixgbe_adapter
*adapter
)
7676 /* if interface is down, removing or resetting, do nothing */
7677 if (test_bit(__IXGBE_DOWN
, &adapter
->state
) ||
7678 test_bit(__IXGBE_REMOVING
, &adapter
->state
) ||
7679 test_bit(__IXGBE_RESETTING
, &adapter
->state
))
7682 ixgbe_watchdog_update_link(adapter
);
7684 if (adapter
->link_up
)
7685 ixgbe_watchdog_link_is_up(adapter
);
7687 ixgbe_watchdog_link_is_down(adapter
);
7689 ixgbe_check_for_bad_vf(adapter
);
7690 ixgbe_spoof_check(adapter
);
7691 ixgbe_update_stats(adapter
);
7693 ixgbe_watchdog_flush_tx(adapter
);
7697 * ixgbe_sfp_detection_subtask - poll for SFP+ cable
7698 * @adapter: the ixgbe adapter structure
7700 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter
*adapter
)
7702 struct ixgbe_hw
*hw
= &adapter
->hw
;
7705 /* not searching for SFP so there is nothing to do here */
7706 if (!(adapter
->flags2
& IXGBE_FLAG2_SEARCH_FOR_SFP
) &&
7707 !(adapter
->flags2
& IXGBE_FLAG2_SFP_NEEDS_RESET
))
7710 if (adapter
->sfp_poll_time
&&
7711 time_after(adapter
->sfp_poll_time
, jiffies
))
7712 return; /* If not yet time to poll for SFP */
7714 /* someone else is in init, wait until next service event */
7715 if (test_and_set_bit(__IXGBE_IN_SFP_INIT
, &adapter
->state
))
7718 adapter
->sfp_poll_time
= jiffies
+ IXGBE_SFP_POLL_JIFFIES
- 1;
7720 err
= hw
->phy
.ops
.identify_sfp(hw
);
7721 if (err
== IXGBE_ERR_SFP_NOT_SUPPORTED
)
7724 if (err
== IXGBE_ERR_SFP_NOT_PRESENT
) {
7725 /* If no cable is present, then we need to reset
7726 * the next time we find a good cable. */
7727 adapter
->flags2
|= IXGBE_FLAG2_SFP_NEEDS_RESET
;
7734 /* exit if reset not needed */
7735 if (!(adapter
->flags2
& IXGBE_FLAG2_SFP_NEEDS_RESET
))
7738 adapter
->flags2
&= ~IXGBE_FLAG2_SFP_NEEDS_RESET
;
7741 * A module may be identified correctly, but the EEPROM may not have
7742 * support for that module. setup_sfp() will fail in that case, so
7743 * we should not allow that module to load.
7745 if (hw
->mac
.type
== ixgbe_mac_82598EB
)
7746 err
= hw
->phy
.ops
.reset(hw
);
7748 err
= hw
->mac
.ops
.setup_sfp(hw
);
7750 if (err
== IXGBE_ERR_SFP_NOT_SUPPORTED
)
7753 adapter
->flags
|= IXGBE_FLAG_NEED_LINK_CONFIG
;
7754 e_info(probe
, "detected SFP+: %d\n", hw
->phy
.sfp_type
);
7757 clear_bit(__IXGBE_IN_SFP_INIT
, &adapter
->state
);
7759 if ((err
== IXGBE_ERR_SFP_NOT_SUPPORTED
) &&
7760 (adapter
->netdev
->reg_state
== NETREG_REGISTERED
)) {
7761 e_dev_err("failed to initialize because an unsupported "
7762 "SFP+ module type was detected.\n");
7763 e_dev_err("Reload the driver after installing a "
7764 "supported module.\n");
7765 unregister_netdev(adapter
->netdev
);
7770 * ixgbe_sfp_link_config_subtask - set up link SFP after module install
7771 * @adapter: the ixgbe adapter structure
7773 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter
*adapter
)
7775 struct ixgbe_hw
*hw
= &adapter
->hw
;
7778 bool autoneg
= false;
7780 if (!(adapter
->flags
& IXGBE_FLAG_NEED_LINK_CONFIG
))
7783 /* someone else is in init, wait until next service event */
7784 if (test_and_set_bit(__IXGBE_IN_SFP_INIT
, &adapter
->state
))
7787 adapter
->flags
&= ~IXGBE_FLAG_NEED_LINK_CONFIG
;
7789 hw
->mac
.ops
.get_link_capabilities(hw
, &cap_speed
, &autoneg
);
7791 /* advertise highest capable link speed */
7792 if (!autoneg
&& (cap_speed
& IXGBE_LINK_SPEED_10GB_FULL
))
7793 speed
= IXGBE_LINK_SPEED_10GB_FULL
;
7795 speed
= cap_speed
& (IXGBE_LINK_SPEED_10GB_FULL
|
7796 IXGBE_LINK_SPEED_1GB_FULL
);
7798 if (hw
->mac
.ops
.setup_link
)
7799 hw
->mac
.ops
.setup_link(hw
, speed
, true);
7801 adapter
->flags
|= IXGBE_FLAG_NEED_LINK_UPDATE
;
7802 adapter
->link_check_timeout
= jiffies
;
7803 clear_bit(__IXGBE_IN_SFP_INIT
, &adapter
->state
);
7807 * ixgbe_service_timer - Timer Call-back
7808 * @t: pointer to timer_list structure
7810 static void ixgbe_service_timer(struct timer_list
*t
)
7812 struct ixgbe_adapter
*adapter
= from_timer(adapter
, t
, service_timer
);
7813 unsigned long next_event_offset
;
7815 /* poll faster when waiting for link */
7816 if (adapter
->flags
& IXGBE_FLAG_NEED_LINK_UPDATE
)
7817 next_event_offset
= HZ
/ 10;
7819 next_event_offset
= HZ
* 2;
7821 /* Reset the timer */
7822 mod_timer(&adapter
->service_timer
, next_event_offset
+ jiffies
);
7824 ixgbe_service_event_schedule(adapter
);
7827 static void ixgbe_phy_interrupt_subtask(struct ixgbe_adapter
*adapter
)
7829 struct ixgbe_hw
*hw
= &adapter
->hw
;
7832 if (!(adapter
->flags2
& IXGBE_FLAG2_PHY_INTERRUPT
))
7835 adapter
->flags2
&= ~IXGBE_FLAG2_PHY_INTERRUPT
;
7837 if (!hw
->phy
.ops
.handle_lasi
)
7840 status
= hw
->phy
.ops
.handle_lasi(&adapter
->hw
);
7841 if (status
!= IXGBE_ERR_OVERTEMP
)
7844 e_crit(drv
, "%s\n", ixgbe_overheat_msg
);
7847 static void ixgbe_reset_subtask(struct ixgbe_adapter
*adapter
)
7849 if (!test_and_clear_bit(__IXGBE_RESET_REQUESTED
, &adapter
->state
))
7853 /* If we're already down, removing or resetting, just bail */
7854 if (test_bit(__IXGBE_DOWN
, &adapter
->state
) ||
7855 test_bit(__IXGBE_REMOVING
, &adapter
->state
) ||
7856 test_bit(__IXGBE_RESETTING
, &adapter
->state
)) {
7861 ixgbe_dump(adapter
);
7862 netdev_err(adapter
->netdev
, "Reset adapter\n");
7863 adapter
->tx_timeout_count
++;
7865 ixgbe_reinit_locked(adapter
);
7870 * ixgbe_check_fw_error - Check firmware for errors
7871 * @adapter: the adapter private structure
7873 * Check firmware errors in register FWSM
7875 static bool ixgbe_check_fw_error(struct ixgbe_adapter
*adapter
)
7877 struct ixgbe_hw
*hw
= &adapter
->hw
;
7880 /* read fwsm.ext_err_ind register and log errors */
7881 fwsm
= IXGBE_READ_REG(hw
, IXGBE_FWSM(hw
));
7883 if (fwsm
& IXGBE_FWSM_EXT_ERR_IND_MASK
||
7884 !(fwsm
& IXGBE_FWSM_FW_VAL_BIT
))
7885 e_dev_warn("Warning firmware error detected FWSM: 0x%08X\n",
7888 if (hw
->mac
.ops
.fw_recovery_mode
&& hw
->mac
.ops
.fw_recovery_mode(hw
)) {
7889 e_dev_err("Firmware recovery mode detected. Limiting functionality. Refer to the Intel(R) Ethernet Adapters and Devices User Guide for details on firmware recovery mode.\n");
7897 * ixgbe_service_task - manages and runs subtasks
7898 * @work: pointer to work_struct containing our data
7900 static void ixgbe_service_task(struct work_struct
*work
)
7902 struct ixgbe_adapter
*adapter
= container_of(work
,
7903 struct ixgbe_adapter
,
7905 if (ixgbe_removed(adapter
->hw
.hw_addr
)) {
7906 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
)) {
7908 ixgbe_down(adapter
);
7911 ixgbe_service_event_complete(adapter
);
7914 if (ixgbe_check_fw_error(adapter
)) {
7915 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
))
7916 unregister_netdev(adapter
->netdev
);
7917 ixgbe_service_event_complete(adapter
);
7920 if (adapter
->flags2
& IXGBE_FLAG2_UDP_TUN_REREG_NEEDED
) {
7922 adapter
->flags2
&= ~IXGBE_FLAG2_UDP_TUN_REREG_NEEDED
;
7923 udp_tunnel_get_rx_info(adapter
->netdev
);
7926 ixgbe_reset_subtask(adapter
);
7927 ixgbe_phy_interrupt_subtask(adapter
);
7928 ixgbe_sfp_detection_subtask(adapter
);
7929 ixgbe_sfp_link_config_subtask(adapter
);
7930 ixgbe_check_overtemp_subtask(adapter
);
7931 ixgbe_watchdog_subtask(adapter
);
7932 ixgbe_fdir_reinit_subtask(adapter
);
7933 ixgbe_check_hang_subtask(adapter
);
7935 if (test_bit(__IXGBE_PTP_RUNNING
, &adapter
->state
)) {
7936 ixgbe_ptp_overflow_check(adapter
);
7937 if (adapter
->flags
& IXGBE_FLAG_RX_HWTSTAMP_IN_REGISTER
)
7938 ixgbe_ptp_rx_hang(adapter
);
7939 ixgbe_ptp_tx_hang(adapter
);
7942 ixgbe_service_event_complete(adapter
);
7945 static int ixgbe_tso(struct ixgbe_ring
*tx_ring
,
7946 struct ixgbe_tx_buffer
*first
,
7948 struct ixgbe_ipsec_tx_data
*itd
)
7950 u32 vlan_macip_lens
, type_tucmd
, mss_l4len_idx
;
7951 struct sk_buff
*skb
= first
->skb
;
7962 u32 paylen
, l4_offset
;
7963 u32 fceof_saidx
= 0;
7966 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
)
7969 if (!skb_is_gso(skb
))
7972 err
= skb_cow_head(skb
, 0);
7976 if (eth_p_mpls(first
->protocol
))
7977 ip
.hdr
= skb_inner_network_header(skb
);
7979 ip
.hdr
= skb_network_header(skb
);
7980 l4
.hdr
= skb_checksum_start(skb
);
7982 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
7983 type_tucmd
= (skb_shinfo(skb
)->gso_type
& SKB_GSO_UDP_L4
) ?
7984 IXGBE_ADVTXD_TUCMD_L4T_UDP
: IXGBE_ADVTXD_TUCMD_L4T_TCP
;
7986 /* initialize outer IP header fields */
7987 if (ip
.v4
->version
== 4) {
7988 unsigned char *csum_start
= skb_checksum_start(skb
);
7989 unsigned char *trans_start
= ip
.hdr
+ (ip
.v4
->ihl
* 4);
7990 int len
= csum_start
- trans_start
;
7992 /* IP header will have to cancel out any data that
7993 * is not a part of the outer IP header, so set to
7994 * a reverse csum if needed, else init check to 0.
7996 ip
.v4
->check
= (skb_shinfo(skb
)->gso_type
& SKB_GSO_PARTIAL
) ?
7997 csum_fold(csum_partial(trans_start
,
7999 type_tucmd
|= IXGBE_ADVTXD_TUCMD_IPV4
;
8002 first
->tx_flags
|= IXGBE_TX_FLAGS_TSO
|
8003 IXGBE_TX_FLAGS_CSUM
|
8004 IXGBE_TX_FLAGS_IPV4
;
8006 ip
.v6
->payload_len
= 0;
8007 first
->tx_flags
|= IXGBE_TX_FLAGS_TSO
|
8008 IXGBE_TX_FLAGS_CSUM
;
8011 /* determine offset of inner transport header */
8012 l4_offset
= l4
.hdr
- skb
->data
;
8014 /* remove payload length from inner checksum */
8015 paylen
= skb
->len
- l4_offset
;
8017 if (type_tucmd
& IXGBE_ADVTXD_TUCMD_L4T_TCP
) {
8018 /* compute length of segmentation header */
8019 *hdr_len
= (l4
.tcp
->doff
* 4) + l4_offset
;
8020 csum_replace_by_diff(&l4
.tcp
->check
,
8021 (__force __wsum
)htonl(paylen
));
8023 /* compute length of segmentation header */
8024 *hdr_len
= sizeof(*l4
.udp
) + l4_offset
;
8025 csum_replace_by_diff(&l4
.udp
->check
,
8026 (__force __wsum
)htonl(paylen
));
8029 /* update gso size and bytecount with header size */
8030 first
->gso_segs
= skb_shinfo(skb
)->gso_segs
;
8031 first
->bytecount
+= (first
->gso_segs
- 1) * *hdr_len
;
8033 /* mss_l4len_id: use 0 as index for TSO */
8034 mss_l4len_idx
= (*hdr_len
- l4_offset
) << IXGBE_ADVTXD_L4LEN_SHIFT
;
8035 mss_l4len_idx
|= skb_shinfo(skb
)->gso_size
<< IXGBE_ADVTXD_MSS_SHIFT
;
8037 fceof_saidx
|= itd
->sa_idx
;
8038 type_tucmd
|= itd
->flags
| itd
->trailer_len
;
8040 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
8041 vlan_macip_lens
= l4
.hdr
- ip
.hdr
;
8042 vlan_macip_lens
|= (ip
.hdr
- skb
->data
) << IXGBE_ADVTXD_MACLEN_SHIFT
;
8043 vlan_macip_lens
|= first
->tx_flags
& IXGBE_TX_FLAGS_VLAN_MASK
;
8045 ixgbe_tx_ctxtdesc(tx_ring
, vlan_macip_lens
, fceof_saidx
, type_tucmd
,
8051 static inline bool ixgbe_ipv6_csum_is_sctp(struct sk_buff
*skb
)
8053 unsigned int offset
= 0;
8055 ipv6_find_hdr(skb
, &offset
, IPPROTO_SCTP
, NULL
, NULL
);
8057 return offset
== skb_checksum_start_offset(skb
);
8060 static void ixgbe_tx_csum(struct ixgbe_ring
*tx_ring
,
8061 struct ixgbe_tx_buffer
*first
,
8062 struct ixgbe_ipsec_tx_data
*itd
)
8064 struct sk_buff
*skb
= first
->skb
;
8065 u32 vlan_macip_lens
= 0;
8066 u32 fceof_saidx
= 0;
8069 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
) {
8071 if (!(first
->tx_flags
& (IXGBE_TX_FLAGS_HW_VLAN
|
8072 IXGBE_TX_FLAGS_CC
)))
8077 switch (skb
->csum_offset
) {
8078 case offsetof(struct tcphdr
, check
):
8079 type_tucmd
= IXGBE_ADVTXD_TUCMD_L4T_TCP
;
8081 case offsetof(struct udphdr
, check
):
8083 case offsetof(struct sctphdr
, checksum
):
8084 /* validate that this is actually an SCTP request */
8085 if (((first
->protocol
== htons(ETH_P_IP
)) &&
8086 (ip_hdr(skb
)->protocol
== IPPROTO_SCTP
)) ||
8087 ((first
->protocol
== htons(ETH_P_IPV6
)) &&
8088 ixgbe_ipv6_csum_is_sctp(skb
))) {
8089 type_tucmd
= IXGBE_ADVTXD_TUCMD_L4T_SCTP
;
8094 skb_checksum_help(skb
);
8098 /* update TX checksum flag */
8099 first
->tx_flags
|= IXGBE_TX_FLAGS_CSUM
;
8100 vlan_macip_lens
= skb_checksum_start_offset(skb
) -
8101 skb_network_offset(skb
);
8103 /* vlan_macip_lens: MACLEN, VLAN tag */
8104 vlan_macip_lens
|= skb_network_offset(skb
) << IXGBE_ADVTXD_MACLEN_SHIFT
;
8105 vlan_macip_lens
|= first
->tx_flags
& IXGBE_TX_FLAGS_VLAN_MASK
;
8107 fceof_saidx
|= itd
->sa_idx
;
8108 type_tucmd
|= itd
->flags
| itd
->trailer_len
;
8110 ixgbe_tx_ctxtdesc(tx_ring
, vlan_macip_lens
, fceof_saidx
, type_tucmd
, 0);
8113 #define IXGBE_SET_FLAG(_input, _flag, _result) \
8114 ((_flag <= _result) ? \
8115 ((u32)(_input & _flag) * (_result / _flag)) : \
8116 ((u32)(_input & _flag) / (_flag / _result)))
8118 static u32
ixgbe_tx_cmd_type(struct sk_buff
*skb
, u32 tx_flags
)
8120 /* set type for advanced descriptor with frame checksum insertion */
8121 u32 cmd_type
= IXGBE_ADVTXD_DTYP_DATA
|
8122 IXGBE_ADVTXD_DCMD_DEXT
|
8123 IXGBE_ADVTXD_DCMD_IFCS
;
8125 /* set HW vlan bit if vlan is present */
8126 cmd_type
|= IXGBE_SET_FLAG(tx_flags
, IXGBE_TX_FLAGS_HW_VLAN
,
8127 IXGBE_ADVTXD_DCMD_VLE
);
8129 /* set segmentation enable bits for TSO/FSO */
8130 cmd_type
|= IXGBE_SET_FLAG(tx_flags
, IXGBE_TX_FLAGS_TSO
,
8131 IXGBE_ADVTXD_DCMD_TSE
);
8133 /* set timestamp bit if present */
8134 cmd_type
|= IXGBE_SET_FLAG(tx_flags
, IXGBE_TX_FLAGS_TSTAMP
,
8135 IXGBE_ADVTXD_MAC_TSTAMP
);
8137 /* insert frame checksum */
8138 cmd_type
^= IXGBE_SET_FLAG(skb
->no_fcs
, 1, IXGBE_ADVTXD_DCMD_IFCS
);
8143 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc
*tx_desc
,
8144 u32 tx_flags
, unsigned int paylen
)
8146 u32 olinfo_status
= paylen
<< IXGBE_ADVTXD_PAYLEN_SHIFT
;
8148 /* enable L4 checksum for TSO and TX checksum offload */
8149 olinfo_status
|= IXGBE_SET_FLAG(tx_flags
,
8150 IXGBE_TX_FLAGS_CSUM
,
8151 IXGBE_ADVTXD_POPTS_TXSM
);
8153 /* enable IPv4 checksum for TSO */
8154 olinfo_status
|= IXGBE_SET_FLAG(tx_flags
,
8155 IXGBE_TX_FLAGS_IPV4
,
8156 IXGBE_ADVTXD_POPTS_IXSM
);
8159 olinfo_status
|= IXGBE_SET_FLAG(tx_flags
,
8160 IXGBE_TX_FLAGS_IPSEC
,
8161 IXGBE_ADVTXD_POPTS_IPSEC
);
8164 * Check Context must be set if Tx switch is enabled, which it
8165 * always is for case where virtual functions are running
8167 olinfo_status
|= IXGBE_SET_FLAG(tx_flags
,
8171 tx_desc
->read
.olinfo_status
= cpu_to_le32(olinfo_status
);
8174 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring
*tx_ring
, u16 size
)
8176 netif_stop_subqueue(tx_ring
->netdev
, tx_ring
->queue_index
);
8178 /* Herbert's original patch had:
8179 * smp_mb__after_netif_stop_queue();
8180 * but since that doesn't exist yet, just open code it.
8184 /* We need to check again in a case another CPU has just
8185 * made room available.
8187 if (likely(ixgbe_desc_unused(tx_ring
) < size
))
8190 /* A reprieve! - use start_queue because it doesn't call schedule */
8191 netif_start_subqueue(tx_ring
->netdev
, tx_ring
->queue_index
);
8192 ++tx_ring
->tx_stats
.restart_queue
;
8196 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring
*tx_ring
, u16 size
)
8198 if (likely(ixgbe_desc_unused(tx_ring
) >= size
))
8201 return __ixgbe_maybe_stop_tx(tx_ring
, size
);
8204 static int ixgbe_tx_map(struct ixgbe_ring
*tx_ring
,
8205 struct ixgbe_tx_buffer
*first
,
8208 struct sk_buff
*skb
= first
->skb
;
8209 struct ixgbe_tx_buffer
*tx_buffer
;
8210 union ixgbe_adv_tx_desc
*tx_desc
;
8213 unsigned int data_len
, size
;
8214 u32 tx_flags
= first
->tx_flags
;
8215 u32 cmd_type
= ixgbe_tx_cmd_type(skb
, tx_flags
);
8216 u16 i
= tx_ring
->next_to_use
;
8218 tx_desc
= IXGBE_TX_DESC(tx_ring
, i
);
8220 ixgbe_tx_olinfo_status(tx_desc
, tx_flags
, skb
->len
- hdr_len
);
8222 size
= skb_headlen(skb
);
8223 data_len
= skb
->data_len
;
8226 if (tx_flags
& IXGBE_TX_FLAGS_FCOE
) {
8227 if (data_len
< sizeof(struct fcoe_crc_eof
)) {
8228 size
-= sizeof(struct fcoe_crc_eof
) - data_len
;
8231 data_len
-= sizeof(struct fcoe_crc_eof
);
8236 dma
= dma_map_single(tx_ring
->dev
, skb
->data
, size
, DMA_TO_DEVICE
);
8240 for (frag
= &skb_shinfo(skb
)->frags
[0];; frag
++) {
8241 if (dma_mapping_error(tx_ring
->dev
, dma
))
8244 /* record length, and DMA address */
8245 dma_unmap_len_set(tx_buffer
, len
, size
);
8246 dma_unmap_addr_set(tx_buffer
, dma
, dma
);
8248 tx_desc
->read
.buffer_addr
= cpu_to_le64(dma
);
8250 while (unlikely(size
> IXGBE_MAX_DATA_PER_TXD
)) {
8251 tx_desc
->read
.cmd_type_len
=
8252 cpu_to_le32(cmd_type
^ IXGBE_MAX_DATA_PER_TXD
);
8256 if (i
== tx_ring
->count
) {
8257 tx_desc
= IXGBE_TX_DESC(tx_ring
, 0);
8260 tx_desc
->read
.olinfo_status
= 0;
8262 dma
+= IXGBE_MAX_DATA_PER_TXD
;
8263 size
-= IXGBE_MAX_DATA_PER_TXD
;
8265 tx_desc
->read
.buffer_addr
= cpu_to_le64(dma
);
8268 if (likely(!data_len
))
8271 tx_desc
->read
.cmd_type_len
= cpu_to_le32(cmd_type
^ size
);
8275 if (i
== tx_ring
->count
) {
8276 tx_desc
= IXGBE_TX_DESC(tx_ring
, 0);
8279 tx_desc
->read
.olinfo_status
= 0;
8282 size
= min_t(unsigned int, data_len
, skb_frag_size(frag
));
8284 size
= skb_frag_size(frag
);
8288 dma
= skb_frag_dma_map(tx_ring
->dev
, frag
, 0, size
,
8291 tx_buffer
= &tx_ring
->tx_buffer_info
[i
];
8294 /* write last descriptor with RS and EOP bits */
8295 cmd_type
|= size
| IXGBE_TXD_CMD
;
8296 tx_desc
->read
.cmd_type_len
= cpu_to_le32(cmd_type
);
8298 netdev_tx_sent_queue(txring_txq(tx_ring
), first
->bytecount
);
8300 /* set the timestamp */
8301 first
->time_stamp
= jiffies
;
8303 skb_tx_timestamp(skb
);
8306 * Force memory writes to complete before letting h/w know there
8307 * are new descriptors to fetch. (Only applicable for weak-ordered
8308 * memory model archs, such as IA-64).
8310 * We also need this memory barrier to make certain all of the
8311 * status bits have been updated before next_to_watch is written.
8315 /* set next_to_watch value indicating a packet is present */
8316 first
->next_to_watch
= tx_desc
;
8319 if (i
== tx_ring
->count
)
8322 tx_ring
->next_to_use
= i
;
8324 ixgbe_maybe_stop_tx(tx_ring
, DESC_NEEDED
);
8326 if (netif_xmit_stopped(txring_txq(tx_ring
)) || !netdev_xmit_more()) {
8327 writel(i
, tx_ring
->tail
);
8332 dev_err(tx_ring
->dev
, "TX DMA map failed\n");
8334 /* clear dma mappings for failed tx_buffer_info map */
8336 tx_buffer
= &tx_ring
->tx_buffer_info
[i
];
8337 if (dma_unmap_len(tx_buffer
, len
))
8338 dma_unmap_page(tx_ring
->dev
,
8339 dma_unmap_addr(tx_buffer
, dma
),
8340 dma_unmap_len(tx_buffer
, len
),
8342 dma_unmap_len_set(tx_buffer
, len
, 0);
8343 if (tx_buffer
== first
)
8346 i
+= tx_ring
->count
;
8350 dev_kfree_skb_any(first
->skb
);
8353 tx_ring
->next_to_use
= i
;
8358 static void ixgbe_atr(struct ixgbe_ring
*ring
,
8359 struct ixgbe_tx_buffer
*first
)
8361 struct ixgbe_q_vector
*q_vector
= ring
->q_vector
;
8362 union ixgbe_atr_hash_dword input
= { .dword
= 0 };
8363 union ixgbe_atr_hash_dword common
= { .dword
= 0 };
8365 unsigned char *network
;
8367 struct ipv6hdr
*ipv6
;
8371 struct sk_buff
*skb
;
8375 /* if ring doesn't have a interrupt vector, cannot perform ATR */
8379 /* do nothing if sampling is disabled */
8380 if (!ring
->atr_sample_rate
)
8385 /* currently only IPv4/IPv6 with TCP is supported */
8386 if ((first
->protocol
!= htons(ETH_P_IP
)) &&
8387 (first
->protocol
!= htons(ETH_P_IPV6
)))
8390 /* snag network header to get L4 type and address */
8392 hdr
.network
= skb_network_header(skb
);
8393 if (unlikely(hdr
.network
<= skb
->data
))
8395 if (skb
->encapsulation
&&
8396 first
->protocol
== htons(ETH_P_IP
) &&
8397 hdr
.ipv4
->protocol
== IPPROTO_UDP
) {
8398 struct ixgbe_adapter
*adapter
= q_vector
->adapter
;
8400 if (unlikely(skb_tail_pointer(skb
) < hdr
.network
+
8404 /* verify the port is recognized as VXLAN */
8405 if (adapter
->vxlan_port
&&
8406 udp_hdr(skb
)->dest
== adapter
->vxlan_port
)
8407 hdr
.network
= skb_inner_network_header(skb
);
8409 if (adapter
->geneve_port
&&
8410 udp_hdr(skb
)->dest
== adapter
->geneve_port
)
8411 hdr
.network
= skb_inner_network_header(skb
);
8414 /* Make sure we have at least [minimum IPv4 header + TCP]
8415 * or [IPv6 header] bytes
8417 if (unlikely(skb_tail_pointer(skb
) < hdr
.network
+ 40))
8420 /* Currently only IPv4/IPv6 with TCP is supported */
8421 switch (hdr
.ipv4
->version
) {
8423 /* access ihl as u8 to avoid unaligned access on ia64 */
8424 hlen
= (hdr
.network
[0] & 0x0F) << 2;
8425 l4_proto
= hdr
.ipv4
->protocol
;
8428 hlen
= hdr
.network
- skb
->data
;
8429 l4_proto
= ipv6_find_hdr(skb
, &hlen
, IPPROTO_TCP
, NULL
, NULL
);
8430 hlen
-= hdr
.network
- skb
->data
;
8436 if (l4_proto
!= IPPROTO_TCP
)
8439 if (unlikely(skb_tail_pointer(skb
) < hdr
.network
+
8440 hlen
+ sizeof(struct tcphdr
)))
8443 th
= (struct tcphdr
*)(hdr
.network
+ hlen
);
8445 /* skip this packet since the socket is closing */
8449 /* sample on all syn packets or once every atr sample count */
8450 if (!th
->syn
&& (ring
->atr_count
< ring
->atr_sample_rate
))
8453 /* reset sample count */
8454 ring
->atr_count
= 0;
8456 vlan_id
= htons(first
->tx_flags
>> IXGBE_TX_FLAGS_VLAN_SHIFT
);
8459 * src and dst are inverted, think how the receiver sees them
8461 * The input is broken into two sections, a non-compressed section
8462 * containing vm_pool, vlan_id, and flow_type. The rest of the data
8463 * is XORed together and stored in the compressed dword.
8465 input
.formatted
.vlan_id
= vlan_id
;
8468 * since src port and flex bytes occupy the same word XOR them together
8469 * and write the value to source port portion of compressed dword
8471 if (first
->tx_flags
& (IXGBE_TX_FLAGS_SW_VLAN
| IXGBE_TX_FLAGS_HW_VLAN
))
8472 common
.port
.src
^= th
->dest
^ htons(ETH_P_8021Q
);
8474 common
.port
.src
^= th
->dest
^ first
->protocol
;
8475 common
.port
.dst
^= th
->source
;
8477 switch (hdr
.ipv4
->version
) {
8479 input
.formatted
.flow_type
= IXGBE_ATR_FLOW_TYPE_TCPV4
;
8480 common
.ip
^= hdr
.ipv4
->saddr
^ hdr
.ipv4
->daddr
;
8483 input
.formatted
.flow_type
= IXGBE_ATR_FLOW_TYPE_TCPV6
;
8484 common
.ip
^= hdr
.ipv6
->saddr
.s6_addr32
[0] ^
8485 hdr
.ipv6
->saddr
.s6_addr32
[1] ^
8486 hdr
.ipv6
->saddr
.s6_addr32
[2] ^
8487 hdr
.ipv6
->saddr
.s6_addr32
[3] ^
8488 hdr
.ipv6
->daddr
.s6_addr32
[0] ^
8489 hdr
.ipv6
->daddr
.s6_addr32
[1] ^
8490 hdr
.ipv6
->daddr
.s6_addr32
[2] ^
8491 hdr
.ipv6
->daddr
.s6_addr32
[3];
8497 if (hdr
.network
!= skb_network_header(skb
))
8498 input
.formatted
.flow_type
|= IXGBE_ATR_L4TYPE_TUNNEL_MASK
;
8500 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
8501 ixgbe_fdir_add_signature_filter_82599(&q_vector
->adapter
->hw
,
8502 input
, common
, ring
->queue_index
);
8506 static u16
ixgbe_select_queue(struct net_device
*dev
, struct sk_buff
*skb
,
8507 struct net_device
*sb_dev
)
8509 struct ixgbe_adapter
*adapter
;
8510 struct ixgbe_ring_feature
*f
;
8514 u8 tc
= netdev_get_prio_tc_map(dev
, skb
->priority
);
8515 struct net_device
*vdev
= sb_dev
;
8517 txq
= vdev
->tc_to_txq
[tc
].offset
;
8518 txq
+= reciprocal_scale(skb_get_hash(skb
),
8519 vdev
->tc_to_txq
[tc
].count
);
8525 * only execute the code below if protocol is FCoE
8526 * or FIP and we have FCoE enabled on the adapter
8528 switch (vlan_get_protocol(skb
)) {
8529 case htons(ETH_P_FCOE
):
8530 case htons(ETH_P_FIP
):
8531 adapter
= netdev_priv(dev
);
8533 if (!sb_dev
&& (adapter
->flags
& IXGBE_FLAG_FCOE_ENABLED
))
8537 return netdev_pick_tx(dev
, skb
, sb_dev
);
8540 f
= &adapter
->ring_feature
[RING_F_FCOE
];
8542 txq
= skb_rx_queue_recorded(skb
) ? skb_get_rx_queue(skb
) :
8545 while (txq
>= f
->indices
)
8548 return txq
+ f
->offset
;
8552 int ixgbe_xmit_xdp_ring(struct ixgbe_adapter
*adapter
,
8553 struct xdp_frame
*xdpf
)
8555 struct ixgbe_ring
*ring
= adapter
->xdp_ring
[smp_processor_id()];
8556 struct ixgbe_tx_buffer
*tx_buffer
;
8557 union ixgbe_adv_tx_desc
*tx_desc
;
8564 if (unlikely(!ixgbe_desc_unused(ring
)))
8565 return IXGBE_XDP_CONSUMED
;
8567 dma
= dma_map_single(ring
->dev
, xdpf
->data
, len
, DMA_TO_DEVICE
);
8568 if (dma_mapping_error(ring
->dev
, dma
))
8569 return IXGBE_XDP_CONSUMED
;
8571 /* record the location of the first descriptor for this packet */
8572 tx_buffer
= &ring
->tx_buffer_info
[ring
->next_to_use
];
8573 tx_buffer
->bytecount
= len
;
8574 tx_buffer
->gso_segs
= 1;
8575 tx_buffer
->protocol
= 0;
8577 i
= ring
->next_to_use
;
8578 tx_desc
= IXGBE_TX_DESC(ring
, i
);
8580 dma_unmap_len_set(tx_buffer
, len
, len
);
8581 dma_unmap_addr_set(tx_buffer
, dma
, dma
);
8582 tx_buffer
->xdpf
= xdpf
;
8584 tx_desc
->read
.buffer_addr
= cpu_to_le64(dma
);
8586 /* put descriptor type bits */
8587 cmd_type
= IXGBE_ADVTXD_DTYP_DATA
|
8588 IXGBE_ADVTXD_DCMD_DEXT
|
8589 IXGBE_ADVTXD_DCMD_IFCS
;
8590 cmd_type
|= len
| IXGBE_TXD_CMD
;
8591 tx_desc
->read
.cmd_type_len
= cpu_to_le32(cmd_type
);
8592 tx_desc
->read
.olinfo_status
=
8593 cpu_to_le32(len
<< IXGBE_ADVTXD_PAYLEN_SHIFT
);
8595 /* Avoid any potential race with xdp_xmit and cleanup */
8598 /* set next_to_watch value indicating a packet is present */
8600 if (i
== ring
->count
)
8603 tx_buffer
->next_to_watch
= tx_desc
;
8604 ring
->next_to_use
= i
;
8606 return IXGBE_XDP_TX
;
8609 netdev_tx_t
ixgbe_xmit_frame_ring(struct sk_buff
*skb
,
8610 struct ixgbe_adapter
*adapter
,
8611 struct ixgbe_ring
*tx_ring
)
8613 struct ixgbe_tx_buffer
*first
;
8617 u16 count
= TXD_USE_COUNT(skb_headlen(skb
));
8618 struct ixgbe_ipsec_tx_data ipsec_tx
= { 0 };
8619 __be16 protocol
= skb
->protocol
;
8623 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
8624 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
8625 * + 2 desc gap to keep tail from touching head,
8626 * + 1 desc for context descriptor,
8627 * otherwise try next time
8629 for (f
= 0; f
< skb_shinfo(skb
)->nr_frags
; f
++)
8630 count
+= TXD_USE_COUNT(skb_frag_size(
8631 &skb_shinfo(skb
)->frags
[f
]));
8633 if (ixgbe_maybe_stop_tx(tx_ring
, count
+ 3)) {
8634 tx_ring
->tx_stats
.tx_busy
++;
8635 return NETDEV_TX_BUSY
;
8638 /* record the location of the first descriptor for this packet */
8639 first
= &tx_ring
->tx_buffer_info
[tx_ring
->next_to_use
];
8641 first
->bytecount
= skb
->len
;
8642 first
->gso_segs
= 1;
8644 /* if we have a HW VLAN tag being added default to the HW one */
8645 if (skb_vlan_tag_present(skb
)) {
8646 tx_flags
|= skb_vlan_tag_get(skb
) << IXGBE_TX_FLAGS_VLAN_SHIFT
;
8647 tx_flags
|= IXGBE_TX_FLAGS_HW_VLAN
;
8648 /* else if it is a SW VLAN check the next protocol and store the tag */
8649 } else if (protocol
== htons(ETH_P_8021Q
)) {
8650 struct vlan_hdr
*vhdr
, _vhdr
;
8651 vhdr
= skb_header_pointer(skb
, ETH_HLEN
, sizeof(_vhdr
), &_vhdr
);
8655 tx_flags
|= ntohs(vhdr
->h_vlan_TCI
) <<
8656 IXGBE_TX_FLAGS_VLAN_SHIFT
;
8657 tx_flags
|= IXGBE_TX_FLAGS_SW_VLAN
;
8659 protocol
= vlan_get_protocol(skb
);
8661 if (unlikely(skb_shinfo(skb
)->tx_flags
& SKBTX_HW_TSTAMP
) &&
8662 adapter
->ptp_clock
) {
8663 if (adapter
->tstamp_config
.tx_type
== HWTSTAMP_TX_ON
&&
8664 !test_and_set_bit_lock(__IXGBE_PTP_TX_IN_PROGRESS
,
8666 skb_shinfo(skb
)->tx_flags
|= SKBTX_IN_PROGRESS
;
8667 tx_flags
|= IXGBE_TX_FLAGS_TSTAMP
;
8669 /* schedule check for Tx timestamp */
8670 adapter
->ptp_tx_skb
= skb_get(skb
);
8671 adapter
->ptp_tx_start
= jiffies
;
8672 schedule_work(&adapter
->ptp_tx_work
);
8674 adapter
->tx_hwtstamp_skipped
++;
8678 #ifdef CONFIG_PCI_IOV
8680 * Use the l2switch_enable flag - would be false if the DMA
8681 * Tx switch had been disabled.
8683 if (adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
)
8684 tx_flags
|= IXGBE_TX_FLAGS_CC
;
8687 /* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
8688 if ((adapter
->flags
& IXGBE_FLAG_DCB_ENABLED
) &&
8689 ((tx_flags
& (IXGBE_TX_FLAGS_HW_VLAN
| IXGBE_TX_FLAGS_SW_VLAN
)) ||
8690 (skb
->priority
!= TC_PRIO_CONTROL
))) {
8691 tx_flags
&= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK
;
8692 tx_flags
|= (skb
->priority
& 0x7) <<
8693 IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT
;
8694 if (tx_flags
& IXGBE_TX_FLAGS_SW_VLAN
) {
8695 struct vlan_ethhdr
*vhdr
;
8697 if (skb_cow_head(skb
, 0))
8699 vhdr
= (struct vlan_ethhdr
*)skb
->data
;
8700 vhdr
->h_vlan_TCI
= htons(tx_flags
>>
8701 IXGBE_TX_FLAGS_VLAN_SHIFT
);
8703 tx_flags
|= IXGBE_TX_FLAGS_HW_VLAN
;
8707 /* record initial flags and protocol */
8708 first
->tx_flags
= tx_flags
;
8709 first
->protocol
= protocol
;
8712 /* setup tx offload for FCoE */
8713 if ((protocol
== htons(ETH_P_FCOE
)) &&
8714 (tx_ring
->netdev
->features
& (NETIF_F_FSO
| NETIF_F_FCOE_CRC
))) {
8715 tso
= ixgbe_fso(tx_ring
, first
, &hdr_len
);
8722 #endif /* IXGBE_FCOE */
8724 #ifdef CONFIG_IXGBE_IPSEC
8725 if (xfrm_offload(skb
) &&
8726 !ixgbe_ipsec_tx(tx_ring
, first
, &ipsec_tx
))
8729 tso
= ixgbe_tso(tx_ring
, first
, &hdr_len
, &ipsec_tx
);
8733 ixgbe_tx_csum(tx_ring
, first
, &ipsec_tx
);
8735 /* add the ATR filter if ATR is on */
8736 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE
, &tx_ring
->state
))
8737 ixgbe_atr(tx_ring
, first
);
8741 #endif /* IXGBE_FCOE */
8742 if (ixgbe_tx_map(tx_ring
, first
, hdr_len
))
8743 goto cleanup_tx_timestamp
;
8745 return NETDEV_TX_OK
;
8748 dev_kfree_skb_any(first
->skb
);
8750 cleanup_tx_timestamp
:
8751 if (unlikely(tx_flags
& IXGBE_TX_FLAGS_TSTAMP
)) {
8752 dev_kfree_skb_any(adapter
->ptp_tx_skb
);
8753 adapter
->ptp_tx_skb
= NULL
;
8754 cancel_work_sync(&adapter
->ptp_tx_work
);
8755 clear_bit_unlock(__IXGBE_PTP_TX_IN_PROGRESS
, &adapter
->state
);
8758 return NETDEV_TX_OK
;
8761 static netdev_tx_t
__ixgbe_xmit_frame(struct sk_buff
*skb
,
8762 struct net_device
*netdev
,
8763 struct ixgbe_ring
*ring
)
8765 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
8766 struct ixgbe_ring
*tx_ring
;
8769 * The minimum packet size for olinfo paylen is 17 so pad the skb
8770 * in order to meet this minimum size requirement.
8772 if (skb_put_padto(skb
, 17))
8773 return NETDEV_TX_OK
;
8775 tx_ring
= ring
? ring
: adapter
->tx_ring
[skb_get_queue_mapping(skb
)];
8776 if (unlikely(test_bit(__IXGBE_TX_DISABLED
, &tx_ring
->state
)))
8777 return NETDEV_TX_BUSY
;
8779 return ixgbe_xmit_frame_ring(skb
, adapter
, tx_ring
);
8782 static netdev_tx_t
ixgbe_xmit_frame(struct sk_buff
*skb
,
8783 struct net_device
*netdev
)
8785 return __ixgbe_xmit_frame(skb
, netdev
, NULL
);
8789 * ixgbe_set_mac - Change the Ethernet Address of the NIC
8790 * @netdev: network interface device structure
8791 * @p: pointer to an address structure
8793 * Returns 0 on success, negative on failure
8795 static int ixgbe_set_mac(struct net_device
*netdev
, void *p
)
8797 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
8798 struct ixgbe_hw
*hw
= &adapter
->hw
;
8799 struct sockaddr
*addr
= p
;
8801 if (!is_valid_ether_addr(addr
->sa_data
))
8802 return -EADDRNOTAVAIL
;
8804 memcpy(netdev
->dev_addr
, addr
->sa_data
, netdev
->addr_len
);
8805 memcpy(hw
->mac
.addr
, addr
->sa_data
, netdev
->addr_len
);
8807 ixgbe_mac_set_default_filter(adapter
);
8813 ixgbe_mdio_read(struct net_device
*netdev
, int prtad
, int devad
, u16 addr
)
8815 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
8816 struct ixgbe_hw
*hw
= &adapter
->hw
;
8820 if (adapter
->mii_bus
) {
8823 if (devad
!= MDIO_DEVAD_NONE
)
8824 regnum
|= (devad
<< 16) | MII_ADDR_C45
;
8826 return mdiobus_read(adapter
->mii_bus
, prtad
, regnum
);
8829 if (prtad
!= hw
->phy
.mdio
.prtad
)
8831 rc
= hw
->phy
.ops
.read_reg(hw
, addr
, devad
, &value
);
8837 static int ixgbe_mdio_write(struct net_device
*netdev
, int prtad
, int devad
,
8838 u16 addr
, u16 value
)
8840 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
8841 struct ixgbe_hw
*hw
= &adapter
->hw
;
8843 if (adapter
->mii_bus
) {
8846 if (devad
!= MDIO_DEVAD_NONE
)
8847 regnum
|= (devad
<< 16) | MII_ADDR_C45
;
8849 return mdiobus_write(adapter
->mii_bus
, prtad
, regnum
, value
);
8852 if (prtad
!= hw
->phy
.mdio
.prtad
)
8854 return hw
->phy
.ops
.write_reg(hw
, addr
, devad
, value
);
8857 static int ixgbe_ioctl(struct net_device
*netdev
, struct ifreq
*req
, int cmd
)
8859 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
8863 return ixgbe_ptp_set_ts_config(adapter
, req
);
8865 return ixgbe_ptp_get_ts_config(adapter
, req
);
8867 if (!adapter
->hw
.phy
.ops
.read_reg
)
8871 return mdio_mii_ioctl(&adapter
->hw
.phy
.mdio
, if_mii(req
), cmd
);
8876 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
8878 * @dev: network interface device structure
8880 * Returns non-zero on failure
8882 static int ixgbe_add_sanmac_netdev(struct net_device
*dev
)
8885 struct ixgbe_adapter
*adapter
= netdev_priv(dev
);
8886 struct ixgbe_hw
*hw
= &adapter
->hw
;
8888 if (is_valid_ether_addr(hw
->mac
.san_addr
)) {
8890 err
= dev_addr_add(dev
, hw
->mac
.san_addr
, NETDEV_HW_ADDR_T_SAN
);
8893 /* update SAN MAC vmdq pool selection */
8894 hw
->mac
.ops
.set_vmdq_san_mac(hw
, VMDQ_P(0));
8900 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
8902 * @dev: network interface device structure
8904 * Returns non-zero on failure
8906 static int ixgbe_del_sanmac_netdev(struct net_device
*dev
)
8909 struct ixgbe_adapter
*adapter
= netdev_priv(dev
);
8910 struct ixgbe_mac_info
*mac
= &adapter
->hw
.mac
;
8912 if (is_valid_ether_addr(mac
->san_addr
)) {
8914 err
= dev_addr_del(dev
, mac
->san_addr
, NETDEV_HW_ADDR_T_SAN
);
8920 static void ixgbe_get_ring_stats64(struct rtnl_link_stats64
*stats
,
8921 struct ixgbe_ring
*ring
)
8928 start
= u64_stats_fetch_begin_irq(&ring
->syncp
);
8929 packets
= ring
->stats
.packets
;
8930 bytes
= ring
->stats
.bytes
;
8931 } while (u64_stats_fetch_retry_irq(&ring
->syncp
, start
));
8932 stats
->tx_packets
+= packets
;
8933 stats
->tx_bytes
+= bytes
;
8937 static void ixgbe_get_stats64(struct net_device
*netdev
,
8938 struct rtnl_link_stats64
*stats
)
8940 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
8944 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
8945 struct ixgbe_ring
*ring
= READ_ONCE(adapter
->rx_ring
[i
]);
8951 start
= u64_stats_fetch_begin_irq(&ring
->syncp
);
8952 packets
= ring
->stats
.packets
;
8953 bytes
= ring
->stats
.bytes
;
8954 } while (u64_stats_fetch_retry_irq(&ring
->syncp
, start
));
8955 stats
->rx_packets
+= packets
;
8956 stats
->rx_bytes
+= bytes
;
8960 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
8961 struct ixgbe_ring
*ring
= READ_ONCE(adapter
->tx_ring
[i
]);
8963 ixgbe_get_ring_stats64(stats
, ring
);
8965 for (i
= 0; i
< adapter
->num_xdp_queues
; i
++) {
8966 struct ixgbe_ring
*ring
= READ_ONCE(adapter
->xdp_ring
[i
]);
8968 ixgbe_get_ring_stats64(stats
, ring
);
8972 /* following stats updated by ixgbe_watchdog_task() */
8973 stats
->multicast
= netdev
->stats
.multicast
;
8974 stats
->rx_errors
= netdev
->stats
.rx_errors
;
8975 stats
->rx_length_errors
= netdev
->stats
.rx_length_errors
;
8976 stats
->rx_crc_errors
= netdev
->stats
.rx_crc_errors
;
8977 stats
->rx_missed_errors
= netdev
->stats
.rx_missed_errors
;
8980 #ifdef CONFIG_IXGBE_DCB
8982 * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
8983 * @adapter: pointer to ixgbe_adapter
8984 * @tc: number of traffic classes currently enabled
8986 * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
8987 * 802.1Q priority maps to a packet buffer that exists.
8989 static void ixgbe_validate_rtr(struct ixgbe_adapter
*adapter
, u8 tc
)
8991 struct ixgbe_hw
*hw
= &adapter
->hw
;
8995 /* 82598 have a static priority to TC mapping that can not
8996 * be changed so no validation is needed.
8998 if (hw
->mac
.type
== ixgbe_mac_82598EB
)
9001 reg
= IXGBE_READ_REG(hw
, IXGBE_RTRUP2TC
);
9004 for (i
= 0; i
< MAX_TRAFFIC_CLASS
; i
++) {
9005 u8 up2tc
= reg
>> (i
* IXGBE_RTRUP2TC_UP_SHIFT
);
9007 /* If up2tc is out of bounds default to zero */
9009 reg
&= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT
);
9013 IXGBE_WRITE_REG(hw
, IXGBE_RTRUP2TC
, reg
);
9019 * ixgbe_set_prio_tc_map - Configure netdev prio tc map
9020 * @adapter: Pointer to adapter struct
9022 * Populate the netdev user priority to tc map
9024 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter
*adapter
)
9026 struct net_device
*dev
= adapter
->netdev
;
9027 struct ixgbe_dcb_config
*dcb_cfg
= &adapter
->dcb_cfg
;
9028 struct ieee_ets
*ets
= adapter
->ixgbe_ieee_ets
;
9031 for (prio
= 0; prio
< MAX_USER_PRIORITY
; prio
++) {
9034 if (adapter
->dcbx_cap
& DCB_CAP_DCBX_VER_CEE
)
9035 tc
= ixgbe_dcb_get_tc_from_up(dcb_cfg
, 0, prio
);
9037 tc
= ets
->prio_tc
[prio
];
9039 netdev_set_prio_tc_map(dev
, prio
, tc
);
9043 #endif /* CONFIG_IXGBE_DCB */
9044 static int ixgbe_reassign_macvlan_pool(struct net_device
*vdev
, void *data
)
9046 struct ixgbe_adapter
*adapter
= data
;
9047 struct ixgbe_fwd_adapter
*accel
;
9050 /* we only care about macvlans... */
9051 if (!netif_is_macvlan(vdev
))
9054 /* that have hardware offload enabled... */
9055 accel
= macvlan_accel_priv(vdev
);
9059 /* If we can relocate to a different bit do so */
9060 pool
= find_first_zero_bit(adapter
->fwd_bitmask
, adapter
->num_rx_pools
);
9061 if (pool
< adapter
->num_rx_pools
) {
9062 set_bit(pool
, adapter
->fwd_bitmask
);
9067 /* if we cannot find a free pool then disable the offload */
9068 netdev_err(vdev
, "L2FW offload disabled due to lack of queue resources\n");
9069 macvlan_release_l2fw_offload(vdev
);
9071 /* unbind the queues and drop the subordinate channel config */
9072 netdev_unbind_sb_channel(adapter
->netdev
, vdev
);
9073 netdev_set_sb_channel(vdev
, 0);
9080 static void ixgbe_defrag_macvlan_pools(struct net_device
*dev
)
9082 struct ixgbe_adapter
*adapter
= netdev_priv(dev
);
9084 /* flush any stale bits out of the fwd bitmask */
9085 bitmap_clear(adapter
->fwd_bitmask
, 1, 63);
9087 /* walk through upper devices reassigning pools */
9088 netdev_walk_all_upper_dev_rcu(dev
, ixgbe_reassign_macvlan_pool
,
9093 * ixgbe_setup_tc - configure net_device for multiple traffic classes
9095 * @dev: net device to configure
9096 * @tc: number of traffic classes to enable
9098 int ixgbe_setup_tc(struct net_device
*dev
, u8 tc
)
9100 struct ixgbe_adapter
*adapter
= netdev_priv(dev
);
9101 struct ixgbe_hw
*hw
= &adapter
->hw
;
9103 /* Hardware supports up to 8 traffic classes */
9104 if (tc
> adapter
->dcb_cfg
.num_tcs
.pg_tcs
)
9107 if (hw
->mac
.type
== ixgbe_mac_82598EB
&& tc
&& tc
< MAX_TRAFFIC_CLASS
)
9110 /* Hardware has to reinitialize queues and interrupts to
9111 * match packet buffer alignment. Unfortunately, the
9112 * hardware is not flexible enough to do this dynamically.
9114 if (netif_running(dev
))
9117 ixgbe_reset(adapter
);
9119 ixgbe_clear_interrupt_scheme(adapter
);
9121 #ifdef CONFIG_IXGBE_DCB
9123 if (adapter
->xdp_prog
) {
9124 e_warn(probe
, "DCB is not supported with XDP\n");
9126 ixgbe_init_interrupt_scheme(adapter
);
9127 if (netif_running(dev
))
9132 netdev_set_num_tc(dev
, tc
);
9133 ixgbe_set_prio_tc_map(adapter
);
9135 adapter
->hw_tcs
= tc
;
9136 adapter
->flags
|= IXGBE_FLAG_DCB_ENABLED
;
9138 if (adapter
->hw
.mac
.type
== ixgbe_mac_82598EB
) {
9139 adapter
->last_lfc_mode
= adapter
->hw
.fc
.requested_mode
;
9140 adapter
->hw
.fc
.requested_mode
= ixgbe_fc_none
;
9143 netdev_reset_tc(dev
);
9145 if (adapter
->hw
.mac
.type
== ixgbe_mac_82598EB
)
9146 adapter
->hw
.fc
.requested_mode
= adapter
->last_lfc_mode
;
9148 adapter
->flags
&= ~IXGBE_FLAG_DCB_ENABLED
;
9149 adapter
->hw_tcs
= tc
;
9151 adapter
->temp_dcb_cfg
.pfc_mode_enable
= false;
9152 adapter
->dcb_cfg
.pfc_mode_enable
= false;
9155 ixgbe_validate_rtr(adapter
, tc
);
9157 #endif /* CONFIG_IXGBE_DCB */
9158 ixgbe_init_interrupt_scheme(adapter
);
9160 ixgbe_defrag_macvlan_pools(dev
);
9162 if (netif_running(dev
))
9163 return ixgbe_open(dev
);
9168 static int ixgbe_delete_clsu32(struct ixgbe_adapter
*adapter
,
9169 struct tc_cls_u32_offload
*cls
)
9171 u32 hdl
= cls
->knode
.handle
;
9172 u32 uhtid
= TC_U32_USERHTID(cls
->knode
.handle
);
9173 u32 loc
= cls
->knode
.handle
& 0xfffff;
9175 struct ixgbe_jump_table
*jump
= NULL
;
9177 if (loc
> IXGBE_MAX_HW_ENTRIES
)
9180 if ((uhtid
!= 0x800) && (uhtid
>= IXGBE_MAX_LINK_HANDLE
))
9183 /* Clear this filter in the link data it is associated with */
9184 if (uhtid
!= 0x800) {
9185 jump
= adapter
->jump_tables
[uhtid
];
9188 if (!test_bit(loc
- 1, jump
->child_loc_map
))
9190 clear_bit(loc
- 1, jump
->child_loc_map
);
9193 /* Check if the filter being deleted is a link */
9194 for (i
= 1; i
< IXGBE_MAX_LINK_HANDLE
; i
++) {
9195 jump
= adapter
->jump_tables
[i
];
9196 if (jump
&& jump
->link_hdl
== hdl
) {
9197 /* Delete filters in the hardware in the child hash
9198 * table associated with this link
9200 for (j
= 0; j
< IXGBE_MAX_HW_ENTRIES
; j
++) {
9201 if (!test_bit(j
, jump
->child_loc_map
))
9203 spin_lock(&adapter
->fdir_perfect_lock
);
9204 err
= ixgbe_update_ethtool_fdir_entry(adapter
,
9207 spin_unlock(&adapter
->fdir_perfect_lock
);
9208 clear_bit(j
, jump
->child_loc_map
);
9210 /* Remove resources for this link */
9214 adapter
->jump_tables
[i
] = NULL
;
9219 spin_lock(&adapter
->fdir_perfect_lock
);
9220 err
= ixgbe_update_ethtool_fdir_entry(adapter
, NULL
, loc
);
9221 spin_unlock(&adapter
->fdir_perfect_lock
);
9225 static int ixgbe_configure_clsu32_add_hnode(struct ixgbe_adapter
*adapter
,
9226 struct tc_cls_u32_offload
*cls
)
9228 u32 uhtid
= TC_U32_USERHTID(cls
->hnode
.handle
);
9230 if (uhtid
>= IXGBE_MAX_LINK_HANDLE
)
9233 /* This ixgbe devices do not support hash tables at the moment
9234 * so abort when given hash tables.
9236 if (cls
->hnode
.divisor
> 0)
9239 set_bit(uhtid
- 1, &adapter
->tables
);
9243 static int ixgbe_configure_clsu32_del_hnode(struct ixgbe_adapter
*adapter
,
9244 struct tc_cls_u32_offload
*cls
)
9246 u32 uhtid
= TC_U32_USERHTID(cls
->hnode
.handle
);
9248 if (uhtid
>= IXGBE_MAX_LINK_HANDLE
)
9251 clear_bit(uhtid
- 1, &adapter
->tables
);
9255 #ifdef CONFIG_NET_CLS_ACT
9256 struct upper_walk_data
{
9257 struct ixgbe_adapter
*adapter
;
9263 static int get_macvlan_queue(struct net_device
*upper
, void *_data
)
9265 if (netif_is_macvlan(upper
)) {
9266 struct ixgbe_fwd_adapter
*vadapter
= macvlan_accel_priv(upper
);
9267 struct upper_walk_data
*data
= _data
;
9268 struct ixgbe_adapter
*adapter
= data
->adapter
;
9269 int ifindex
= data
->ifindex
;
9271 if (vadapter
&& upper
->ifindex
== ifindex
) {
9272 data
->queue
= adapter
->rx_ring
[vadapter
->rx_base_queue
]->reg_idx
;
9273 data
->action
= data
->queue
;
9281 static int handle_redirect_action(struct ixgbe_adapter
*adapter
, int ifindex
,
9282 u8
*queue
, u64
*action
)
9284 struct ixgbe_ring_feature
*vmdq
= &adapter
->ring_feature
[RING_F_VMDQ
];
9285 unsigned int num_vfs
= adapter
->num_vfs
, vf
;
9286 struct upper_walk_data data
;
9287 struct net_device
*upper
;
9289 /* redirect to a SRIOV VF */
9290 for (vf
= 0; vf
< num_vfs
; ++vf
) {
9291 upper
= pci_get_drvdata(adapter
->vfinfo
[vf
].vfdev
);
9292 if (upper
->ifindex
== ifindex
) {
9293 *queue
= vf
* __ALIGN_MASK(1, ~vmdq
->mask
);
9295 *action
<<= ETHTOOL_RX_FLOW_SPEC_RING_VF_OFF
;
9300 /* redirect to a offloaded macvlan netdev */
9301 data
.adapter
= adapter
;
9302 data
.ifindex
= ifindex
;
9305 if (netdev_walk_all_upper_dev_rcu(adapter
->netdev
,
9306 get_macvlan_queue
, &data
)) {
9307 *action
= data
.action
;
9308 *queue
= data
.queue
;
9316 static int parse_tc_actions(struct ixgbe_adapter
*adapter
,
9317 struct tcf_exts
*exts
, u64
*action
, u8
*queue
)
9319 const struct tc_action
*a
;
9322 if (!tcf_exts_has_actions(exts
))
9325 tcf_exts_for_each_action(i
, a
, exts
) {
9327 if (is_tcf_gact_shot(a
)) {
9328 *action
= IXGBE_FDIR_DROP_QUEUE
;
9329 *queue
= IXGBE_FDIR_DROP_QUEUE
;
9333 /* Redirect to a VF or a offloaded macvlan */
9334 if (is_tcf_mirred_egress_redirect(a
)) {
9335 struct net_device
*dev
= tcf_mirred_dev(a
);
9339 return handle_redirect_action(adapter
, dev
->ifindex
,
9349 static int parse_tc_actions(struct ixgbe_adapter
*adapter
,
9350 struct tcf_exts
*exts
, u64
*action
, u8
*queue
)
9354 #endif /* CONFIG_NET_CLS_ACT */
9356 static int ixgbe_clsu32_build_input(struct ixgbe_fdir_filter
*input
,
9357 union ixgbe_atr_input
*mask
,
9358 struct tc_cls_u32_offload
*cls
,
9359 struct ixgbe_mat_field
*field_ptr
,
9360 struct ixgbe_nexthdr
*nexthdr
)
9364 bool found_entry
= false, found_jump_field
= false;
9366 for (i
= 0; i
< cls
->knode
.sel
->nkeys
; i
++) {
9367 off
= cls
->knode
.sel
->keys
[i
].off
;
9368 val
= cls
->knode
.sel
->keys
[i
].val
;
9369 m
= cls
->knode
.sel
->keys
[i
].mask
;
9371 for (j
= 0; field_ptr
[j
].val
; j
++) {
9372 if (field_ptr
[j
].off
== off
) {
9373 field_ptr
[j
].val(input
, mask
, (__force u32
)val
,
9375 input
->filter
.formatted
.flow_type
|=
9382 if (nexthdr
->off
== cls
->knode
.sel
->keys
[i
].off
&&
9384 (__force u32
)cls
->knode
.sel
->keys
[i
].val
&&
9386 (__force u32
)cls
->knode
.sel
->keys
[i
].mask
)
9387 found_jump_field
= true;
9393 if (nexthdr
&& !found_jump_field
)
9399 mask
->formatted
.flow_type
= IXGBE_ATR_L4TYPE_IPV6_MASK
|
9400 IXGBE_ATR_L4TYPE_MASK
;
9402 if (input
->filter
.formatted
.flow_type
== IXGBE_ATR_FLOW_TYPE_IPV4
)
9403 mask
->formatted
.flow_type
&= IXGBE_ATR_L4TYPE_IPV6_MASK
;
9408 static int ixgbe_configure_clsu32(struct ixgbe_adapter
*adapter
,
9409 struct tc_cls_u32_offload
*cls
)
9411 __be16 protocol
= cls
->common
.protocol
;
9412 u32 loc
= cls
->knode
.handle
& 0xfffff;
9413 struct ixgbe_hw
*hw
= &adapter
->hw
;
9414 struct ixgbe_mat_field
*field_ptr
;
9415 struct ixgbe_fdir_filter
*input
= NULL
;
9416 union ixgbe_atr_input
*mask
= NULL
;
9417 struct ixgbe_jump_table
*jump
= NULL
;
9418 int i
, err
= -EINVAL
;
9420 u32 uhtid
, link_uhtid
;
9422 uhtid
= TC_U32_USERHTID(cls
->knode
.handle
);
9423 link_uhtid
= TC_U32_USERHTID(cls
->knode
.link_handle
);
9425 /* At the moment cls_u32 jumps to network layer and skips past
9426 * L2 headers. The canonical method to match L2 frames is to use
9427 * negative values. However this is error prone at best but really
9428 * just broken because there is no way to "know" what sort of hdr
9429 * is in front of the network layer. Fix cls_u32 to support L2
9430 * headers when needed.
9432 if (protocol
!= htons(ETH_P_IP
))
9435 if (loc
>= ((1024 << adapter
->fdir_pballoc
) - 2)) {
9436 e_err(drv
, "Location out of range\n");
9440 /* cls u32 is a graph starting at root node 0x800. The driver tracks
9441 * links and also the fields used to advance the parser across each
9442 * link (e.g. nexthdr/eat parameters from 'tc'). This way we can map
9443 * the u32 graph onto the hardware parse graph denoted in ixgbe_model.h
9444 * To add support for new nodes update ixgbe_model.h parse structures
9445 * this function _should_ be generic try not to hardcode values here.
9447 if (uhtid
== 0x800) {
9448 field_ptr
= (adapter
->jump_tables
[0])->mat
;
9450 if (uhtid
>= IXGBE_MAX_LINK_HANDLE
)
9452 if (!adapter
->jump_tables
[uhtid
])
9454 field_ptr
= (adapter
->jump_tables
[uhtid
])->mat
;
9460 /* At this point we know the field_ptr is valid and need to either
9461 * build cls_u32 link or attach filter. Because adding a link to
9462 * a handle that does not exist is invalid and the same for adding
9463 * rules to handles that don't exist.
9467 struct ixgbe_nexthdr
*nexthdr
= ixgbe_ipv4_jumps
;
9469 if (link_uhtid
>= IXGBE_MAX_LINK_HANDLE
)
9472 if (!test_bit(link_uhtid
- 1, &adapter
->tables
))
9475 /* Multiple filters as links to the same hash table are not
9476 * supported. To add a new filter with the same next header
9477 * but different match/jump conditions, create a new hash table
9480 if (adapter
->jump_tables
[link_uhtid
] &&
9481 (adapter
->jump_tables
[link_uhtid
])->link_hdl
) {
9482 e_err(drv
, "Link filter exists for link: %x\n",
9487 for (i
= 0; nexthdr
[i
].jump
; i
++) {
9488 if (nexthdr
[i
].o
!= cls
->knode
.sel
->offoff
||
9489 nexthdr
[i
].s
!= cls
->knode
.sel
->offshift
||
9491 (__force u32
)cls
->knode
.sel
->offmask
)
9494 jump
= kzalloc(sizeof(*jump
), GFP_KERNEL
);
9497 input
= kzalloc(sizeof(*input
), GFP_KERNEL
);
9502 mask
= kzalloc(sizeof(*mask
), GFP_KERNEL
);
9507 jump
->input
= input
;
9509 jump
->link_hdl
= cls
->knode
.handle
;
9511 err
= ixgbe_clsu32_build_input(input
, mask
, cls
,
9512 field_ptr
, &nexthdr
[i
]);
9514 jump
->mat
= nexthdr
[i
].jump
;
9515 adapter
->jump_tables
[link_uhtid
] = jump
;
9526 input
= kzalloc(sizeof(*input
), GFP_KERNEL
);
9529 mask
= kzalloc(sizeof(*mask
), GFP_KERNEL
);
9535 if ((uhtid
!= 0x800) && (adapter
->jump_tables
[uhtid
])) {
9536 if ((adapter
->jump_tables
[uhtid
])->input
)
9537 memcpy(input
, (adapter
->jump_tables
[uhtid
])->input
,
9539 if ((adapter
->jump_tables
[uhtid
])->mask
)
9540 memcpy(mask
, (adapter
->jump_tables
[uhtid
])->mask
,
9543 /* Lookup in all child hash tables if this location is already
9544 * filled with a filter
9546 for (i
= 1; i
< IXGBE_MAX_LINK_HANDLE
; i
++) {
9547 struct ixgbe_jump_table
*link
= adapter
->jump_tables
[i
];
9549 if (link
&& (test_bit(loc
- 1, link
->child_loc_map
))) {
9550 e_err(drv
, "Filter exists in location: %x\n",
9557 err
= ixgbe_clsu32_build_input(input
, mask
, cls
, field_ptr
, NULL
);
9561 err
= parse_tc_actions(adapter
, cls
->knode
.exts
, &input
->action
,
9566 input
->sw_idx
= loc
;
9568 spin_lock(&adapter
->fdir_perfect_lock
);
9570 if (hlist_empty(&adapter
->fdir_filter_list
)) {
9571 memcpy(&adapter
->fdir_mask
, mask
, sizeof(*mask
));
9572 err
= ixgbe_fdir_set_input_mask_82599(hw
, mask
);
9574 goto err_out_w_lock
;
9575 } else if (memcmp(&adapter
->fdir_mask
, mask
, sizeof(*mask
))) {
9577 goto err_out_w_lock
;
9580 ixgbe_atr_compute_perfect_hash_82599(&input
->filter
, mask
);
9581 err
= ixgbe_fdir_write_perfect_filter_82599(hw
, &input
->filter
,
9582 input
->sw_idx
, queue
);
9584 ixgbe_update_ethtool_fdir_entry(adapter
, input
, input
->sw_idx
);
9585 spin_unlock(&adapter
->fdir_perfect_lock
);
9587 if ((uhtid
!= 0x800) && (adapter
->jump_tables
[uhtid
]))
9588 set_bit(loc
- 1, (adapter
->jump_tables
[uhtid
])->child_loc_map
);
9593 spin_unlock(&adapter
->fdir_perfect_lock
);
9603 static int ixgbe_setup_tc_cls_u32(struct ixgbe_adapter
*adapter
,
9604 struct tc_cls_u32_offload
*cls_u32
)
9606 switch (cls_u32
->command
) {
9607 case TC_CLSU32_NEW_KNODE
:
9608 case TC_CLSU32_REPLACE_KNODE
:
9609 return ixgbe_configure_clsu32(adapter
, cls_u32
);
9610 case TC_CLSU32_DELETE_KNODE
:
9611 return ixgbe_delete_clsu32(adapter
, cls_u32
);
9612 case TC_CLSU32_NEW_HNODE
:
9613 case TC_CLSU32_REPLACE_HNODE
:
9614 return ixgbe_configure_clsu32_add_hnode(adapter
, cls_u32
);
9615 case TC_CLSU32_DELETE_HNODE
:
9616 return ixgbe_configure_clsu32_del_hnode(adapter
, cls_u32
);
9622 static int ixgbe_setup_tc_block_cb(enum tc_setup_type type
, void *type_data
,
9625 struct ixgbe_adapter
*adapter
= cb_priv
;
9627 if (!tc_cls_can_offload_and_chain0(adapter
->netdev
, type_data
))
9631 case TC_SETUP_CLSU32
:
9632 return ixgbe_setup_tc_cls_u32(adapter
, type_data
);
9638 static int ixgbe_setup_tc_mqprio(struct net_device
*dev
,
9639 struct tc_mqprio_qopt
*mqprio
)
9641 mqprio
->hw
= TC_MQPRIO_HW_OFFLOAD_TCS
;
9642 return ixgbe_setup_tc(dev
, mqprio
->num_tc
);
9645 static LIST_HEAD(ixgbe_block_cb_list
);
9647 static int __ixgbe_setup_tc(struct net_device
*dev
, enum tc_setup_type type
,
9650 struct ixgbe_adapter
*adapter
= netdev_priv(dev
);
9653 case TC_SETUP_BLOCK
:
9654 return flow_block_cb_setup_simple(type_data
,
9655 &ixgbe_block_cb_list
,
9656 ixgbe_setup_tc_block_cb
,
9657 adapter
, adapter
, true);
9658 case TC_SETUP_QDISC_MQPRIO
:
9659 return ixgbe_setup_tc_mqprio(dev
, type_data
);
9665 #ifdef CONFIG_PCI_IOV
9666 void ixgbe_sriov_reinit(struct ixgbe_adapter
*adapter
)
9668 struct net_device
*netdev
= adapter
->netdev
;
9671 ixgbe_setup_tc(netdev
, adapter
->hw_tcs
);
9676 void ixgbe_do_reset(struct net_device
*netdev
)
9678 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
9680 if (netif_running(netdev
))
9681 ixgbe_reinit_locked(adapter
);
9683 ixgbe_reset(adapter
);
9686 static netdev_features_t
ixgbe_fix_features(struct net_device
*netdev
,
9687 netdev_features_t features
)
9689 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
9691 /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
9692 if (!(features
& NETIF_F_RXCSUM
))
9693 features
&= ~NETIF_F_LRO
;
9695 /* Turn off LRO if not RSC capable */
9696 if (!(adapter
->flags2
& IXGBE_FLAG2_RSC_CAPABLE
))
9697 features
&= ~NETIF_F_LRO
;
9699 if (adapter
->xdp_prog
&& (features
& NETIF_F_LRO
)) {
9700 e_dev_err("LRO is not supported with XDP\n");
9701 features
&= ~NETIF_F_LRO
;
9707 static void ixgbe_reset_l2fw_offload(struct ixgbe_adapter
*adapter
)
9709 int rss
= min_t(int, ixgbe_max_rss_indices(adapter
),
9712 /* go back to full RSS if we're not running SR-IOV */
9713 if (!adapter
->ring_feature
[RING_F_VMDQ
].offset
)
9714 adapter
->flags
&= ~(IXGBE_FLAG_VMDQ_ENABLED
|
9715 IXGBE_FLAG_SRIOV_ENABLED
);
9717 adapter
->ring_feature
[RING_F_RSS
].limit
= rss
;
9718 adapter
->ring_feature
[RING_F_VMDQ
].limit
= 1;
9720 ixgbe_setup_tc(adapter
->netdev
, adapter
->hw_tcs
);
9723 static int ixgbe_set_features(struct net_device
*netdev
,
9724 netdev_features_t features
)
9726 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
9727 netdev_features_t changed
= netdev
->features
^ features
;
9728 bool need_reset
= false;
9730 /* Make sure RSC matches LRO, reset if change */
9731 if (!(features
& NETIF_F_LRO
)) {
9732 if (adapter
->flags2
& IXGBE_FLAG2_RSC_ENABLED
)
9734 adapter
->flags2
&= ~IXGBE_FLAG2_RSC_ENABLED
;
9735 } else if ((adapter
->flags2
& IXGBE_FLAG2_RSC_CAPABLE
) &&
9736 !(adapter
->flags2
& IXGBE_FLAG2_RSC_ENABLED
)) {
9737 if (adapter
->rx_itr_setting
== 1 ||
9738 adapter
->rx_itr_setting
> IXGBE_MIN_RSC_ITR
) {
9739 adapter
->flags2
|= IXGBE_FLAG2_RSC_ENABLED
;
9741 } else if ((changed
^ features
) & NETIF_F_LRO
) {
9742 e_info(probe
, "rx-usecs set too low, "
9748 * Check if Flow Director n-tuple support or hw_tc support was
9749 * enabled or disabled. If the state changed, we need to reset.
9751 if ((features
& NETIF_F_NTUPLE
) || (features
& NETIF_F_HW_TC
)) {
9752 /* turn off ATR, enable perfect filters and reset */
9753 if (!(adapter
->flags
& IXGBE_FLAG_FDIR_PERFECT_CAPABLE
))
9756 adapter
->flags
&= ~IXGBE_FLAG_FDIR_HASH_CAPABLE
;
9757 adapter
->flags
|= IXGBE_FLAG_FDIR_PERFECT_CAPABLE
;
9759 /* turn off perfect filters, enable ATR and reset */
9760 if (adapter
->flags
& IXGBE_FLAG_FDIR_PERFECT_CAPABLE
)
9763 adapter
->flags
&= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE
;
9765 /* We cannot enable ATR if SR-IOV is enabled */
9766 if (adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
||
9767 /* We cannot enable ATR if we have 2 or more tcs */
9768 (adapter
->hw_tcs
> 1) ||
9769 /* We cannot enable ATR if RSS is disabled */
9770 (adapter
->ring_feature
[RING_F_RSS
].limit
<= 1) ||
9771 /* A sample rate of 0 indicates ATR disabled */
9772 (!adapter
->atr_sample_rate
))
9773 ; /* do nothing not supported */
9774 else /* otherwise supported and set the flag */
9775 adapter
->flags
|= IXGBE_FLAG_FDIR_HASH_CAPABLE
;
9778 if (changed
& NETIF_F_RXALL
)
9781 netdev
->features
= features
;
9783 if ((adapter
->flags
& IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE
)) {
9784 if (features
& NETIF_F_RXCSUM
) {
9785 adapter
->flags2
|= IXGBE_FLAG2_UDP_TUN_REREG_NEEDED
;
9787 u32 port_mask
= IXGBE_VXLANCTRL_VXLAN_UDPPORT_MASK
;
9789 ixgbe_clear_udp_tunnel_port(adapter
, port_mask
);
9793 if ((adapter
->flags
& IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE
)) {
9794 if (features
& NETIF_F_RXCSUM
) {
9795 adapter
->flags2
|= IXGBE_FLAG2_UDP_TUN_REREG_NEEDED
;
9797 u32 port_mask
= IXGBE_VXLANCTRL_GENEVE_UDPPORT_MASK
;
9799 ixgbe_clear_udp_tunnel_port(adapter
, port_mask
);
9803 if ((changed
& NETIF_F_HW_L2FW_DOFFLOAD
) && adapter
->num_rx_pools
> 1)
9804 ixgbe_reset_l2fw_offload(adapter
);
9805 else if (need_reset
)
9806 ixgbe_do_reset(netdev
);
9807 else if (changed
& (NETIF_F_HW_VLAN_CTAG_RX
|
9808 NETIF_F_HW_VLAN_CTAG_FILTER
))
9809 ixgbe_set_rx_mode(netdev
);
9815 * ixgbe_add_udp_tunnel_port - Get notifications about adding UDP tunnel ports
9816 * @dev: The port's netdev
9817 * @ti: Tunnel endpoint information
9819 static void ixgbe_add_udp_tunnel_port(struct net_device
*dev
,
9820 struct udp_tunnel_info
*ti
)
9822 struct ixgbe_adapter
*adapter
= netdev_priv(dev
);
9823 struct ixgbe_hw
*hw
= &adapter
->hw
;
9824 __be16 port
= ti
->port
;
9828 if (ti
->sa_family
!= AF_INET
)
9832 case UDP_TUNNEL_TYPE_VXLAN
:
9833 if (!(adapter
->flags
& IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE
))
9836 if (adapter
->vxlan_port
== port
)
9839 if (adapter
->vxlan_port
) {
9841 "VXLAN port %d set, not adding port %d\n",
9842 ntohs(adapter
->vxlan_port
),
9847 adapter
->vxlan_port
= port
;
9849 case UDP_TUNNEL_TYPE_GENEVE
:
9850 if (!(adapter
->flags
& IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE
))
9853 if (adapter
->geneve_port
== port
)
9856 if (adapter
->geneve_port
) {
9858 "GENEVE port %d set, not adding port %d\n",
9859 ntohs(adapter
->geneve_port
),
9864 port_shift
= IXGBE_VXLANCTRL_GENEVE_UDPPORT_SHIFT
;
9865 adapter
->geneve_port
= port
;
9871 reg
= IXGBE_READ_REG(hw
, IXGBE_VXLANCTRL
) | ntohs(port
) << port_shift
;
9872 IXGBE_WRITE_REG(hw
, IXGBE_VXLANCTRL
, reg
);
9876 * ixgbe_del_udp_tunnel_port - Get notifications about removing UDP tunnel ports
9877 * @dev: The port's netdev
9878 * @ti: Tunnel endpoint information
9880 static void ixgbe_del_udp_tunnel_port(struct net_device
*dev
,
9881 struct udp_tunnel_info
*ti
)
9883 struct ixgbe_adapter
*adapter
= netdev_priv(dev
);
9886 if (ti
->type
!= UDP_TUNNEL_TYPE_VXLAN
&&
9887 ti
->type
!= UDP_TUNNEL_TYPE_GENEVE
)
9890 if (ti
->sa_family
!= AF_INET
)
9894 case UDP_TUNNEL_TYPE_VXLAN
:
9895 if (!(adapter
->flags
& IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE
))
9898 if (adapter
->vxlan_port
!= ti
->port
) {
9899 netdev_info(dev
, "VXLAN port %d not found\n",
9904 port_mask
= IXGBE_VXLANCTRL_VXLAN_UDPPORT_MASK
;
9906 case UDP_TUNNEL_TYPE_GENEVE
:
9907 if (!(adapter
->flags
& IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE
))
9910 if (adapter
->geneve_port
!= ti
->port
) {
9911 netdev_info(dev
, "GENEVE port %d not found\n",
9916 port_mask
= IXGBE_VXLANCTRL_GENEVE_UDPPORT_MASK
;
9922 ixgbe_clear_udp_tunnel_port(adapter
, port_mask
);
9923 adapter
->flags2
|= IXGBE_FLAG2_UDP_TUN_REREG_NEEDED
;
9926 static int ixgbe_ndo_fdb_add(struct ndmsg
*ndm
, struct nlattr
*tb
[],
9927 struct net_device
*dev
,
9928 const unsigned char *addr
, u16 vid
,
9930 struct netlink_ext_ack
*extack
)
9932 /* guarantee we can provide a unique filter for the unicast address */
9933 if (is_unicast_ether_addr(addr
) || is_link_local_ether_addr(addr
)) {
9934 struct ixgbe_adapter
*adapter
= netdev_priv(dev
);
9935 u16 pool
= VMDQ_P(0);
9937 if (netdev_uc_count(dev
) >= ixgbe_available_rars(adapter
, pool
))
9941 return ndo_dflt_fdb_add(ndm
, tb
, dev
, addr
, vid
, flags
);
9945 * ixgbe_configure_bridge_mode - set various bridge modes
9946 * @adapter: the private structure
9947 * @mode: requested bridge mode
9949 * Configure some settings require for various bridge modes.
9951 static int ixgbe_configure_bridge_mode(struct ixgbe_adapter
*adapter
,
9954 struct ixgbe_hw
*hw
= &adapter
->hw
;
9955 unsigned int p
, num_pools
;
9959 case BRIDGE_MODE_VEPA
:
9960 /* disable Tx loopback, rely on switch hairpin mode */
9961 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_PFDTXGSWC
, 0);
9963 /* must enable Rx switching replication to allow multicast
9964 * packet reception on all VFs, and to enable source address
9967 vmdctl
= IXGBE_READ_REG(hw
, IXGBE_VMD_CTL
);
9968 vmdctl
|= IXGBE_VT_CTL_REPLEN
;
9969 IXGBE_WRITE_REG(hw
, IXGBE_VMD_CTL
, vmdctl
);
9971 /* enable Rx source address pruning. Note, this requires
9972 * replication to be enabled or else it does nothing.
9974 num_pools
= adapter
->num_vfs
+ adapter
->num_rx_pools
;
9975 for (p
= 0; p
< num_pools
; p
++) {
9976 if (hw
->mac
.ops
.set_source_address_pruning
)
9977 hw
->mac
.ops
.set_source_address_pruning(hw
,
9982 case BRIDGE_MODE_VEB
:
9983 /* enable Tx loopback for internal VF/PF communication */
9984 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_PFDTXGSWC
,
9985 IXGBE_PFDTXGSWC_VT_LBEN
);
9987 /* disable Rx switching replication unless we have SR-IOV
9990 vmdctl
= IXGBE_READ_REG(hw
, IXGBE_VMD_CTL
);
9991 if (!adapter
->num_vfs
)
9992 vmdctl
&= ~IXGBE_VT_CTL_REPLEN
;
9993 IXGBE_WRITE_REG(hw
, IXGBE_VMD_CTL
, vmdctl
);
9995 /* disable Rx source address pruning, since we don't expect to
9996 * be receiving external loopback of our transmitted frames.
9998 num_pools
= adapter
->num_vfs
+ adapter
->num_rx_pools
;
9999 for (p
= 0; p
< num_pools
; p
++) {
10000 if (hw
->mac
.ops
.set_source_address_pruning
)
10001 hw
->mac
.ops
.set_source_address_pruning(hw
,
10010 adapter
->bridge_mode
= mode
;
10012 e_info(drv
, "enabling bridge mode: %s\n",
10013 mode
== BRIDGE_MODE_VEPA
? "VEPA" : "VEB");
10018 static int ixgbe_ndo_bridge_setlink(struct net_device
*dev
,
10019 struct nlmsghdr
*nlh
, u16 flags
,
10020 struct netlink_ext_ack
*extack
)
10022 struct ixgbe_adapter
*adapter
= netdev_priv(dev
);
10023 struct nlattr
*attr
, *br_spec
;
10026 if (!(adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
))
10027 return -EOPNOTSUPP
;
10029 br_spec
= nlmsg_find_attr(nlh
, sizeof(struct ifinfomsg
), IFLA_AF_SPEC
);
10033 nla_for_each_nested(attr
, br_spec
, rem
) {
10037 if (nla_type(attr
) != IFLA_BRIDGE_MODE
)
10040 if (nla_len(attr
) < sizeof(mode
))
10043 mode
= nla_get_u16(attr
);
10044 status
= ixgbe_configure_bridge_mode(adapter
, mode
);
10054 static int ixgbe_ndo_bridge_getlink(struct sk_buff
*skb
, u32 pid
, u32 seq
,
10055 struct net_device
*dev
,
10056 u32 filter_mask
, int nlflags
)
10058 struct ixgbe_adapter
*adapter
= netdev_priv(dev
);
10060 if (!(adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
))
10063 return ndo_dflt_bridge_getlink(skb
, pid
, seq
, dev
,
10064 adapter
->bridge_mode
, 0, 0, nlflags
,
10065 filter_mask
, NULL
);
10068 static void *ixgbe_fwd_add(struct net_device
*pdev
, struct net_device
*vdev
)
10070 struct ixgbe_adapter
*adapter
= netdev_priv(pdev
);
10071 struct ixgbe_fwd_adapter
*accel
;
10072 int tcs
= adapter
->hw_tcs
? : 1;
10075 if (adapter
->xdp_prog
) {
10076 e_warn(probe
, "L2FW offload is not supported with XDP\n");
10077 return ERR_PTR(-EINVAL
);
10080 /* The hardware supported by ixgbe only filters on the destination MAC
10081 * address. In order to avoid issues we only support offloading modes
10082 * where the hardware can actually provide the functionality.
10084 if (!macvlan_supports_dest_filter(vdev
))
10085 return ERR_PTR(-EMEDIUMTYPE
);
10087 /* We need to lock down the macvlan to be a single queue device so that
10088 * we can reuse the tc_to_txq field in the macvlan netdev to represent
10089 * the queue mapping to our netdev.
10091 if (netif_is_multiqueue(vdev
))
10092 return ERR_PTR(-ERANGE
);
10094 pool
= find_first_zero_bit(adapter
->fwd_bitmask
, adapter
->num_rx_pools
);
10095 if (pool
== adapter
->num_rx_pools
) {
10096 u16 used_pools
= adapter
->num_vfs
+ adapter
->num_rx_pools
;
10097 u16 reserved_pools
;
10099 if (((adapter
->flags
& IXGBE_FLAG_DCB_ENABLED
) &&
10100 adapter
->num_rx_pools
>= (MAX_TX_QUEUES
/ tcs
)) ||
10101 adapter
->num_rx_pools
> IXGBE_MAX_MACVLANS
)
10102 return ERR_PTR(-EBUSY
);
10104 /* Hardware has a limited number of available pools. Each VF,
10105 * and the PF require a pool. Check to ensure we don't
10106 * attempt to use more then the available number of pools.
10108 if (used_pools
>= IXGBE_MAX_VF_FUNCTIONS
)
10109 return ERR_PTR(-EBUSY
);
10111 /* Enable VMDq flag so device will be set in VM mode */
10112 adapter
->flags
|= IXGBE_FLAG_VMDQ_ENABLED
|
10113 IXGBE_FLAG_SRIOV_ENABLED
;
10115 /* Try to reserve as many queues per pool as possible,
10116 * we start with the configurations that support 4 queues
10117 * per pools, followed by 2, and then by just 1 per pool.
10119 if (used_pools
< 32 && adapter
->num_rx_pools
< 16)
10120 reserved_pools
= min_t(u16
,
10122 16 - adapter
->num_rx_pools
);
10123 else if (adapter
->num_rx_pools
< 32)
10124 reserved_pools
= min_t(u16
,
10126 32 - adapter
->num_rx_pools
);
10128 reserved_pools
= 64 - used_pools
;
10131 if (!reserved_pools
)
10132 return ERR_PTR(-EBUSY
);
10134 adapter
->ring_feature
[RING_F_VMDQ
].limit
+= reserved_pools
;
10136 /* Force reinit of ring allocation with VMDQ enabled */
10137 err
= ixgbe_setup_tc(pdev
, adapter
->hw_tcs
);
10139 return ERR_PTR(err
);
10141 if (pool
>= adapter
->num_rx_pools
)
10142 return ERR_PTR(-ENOMEM
);
10145 accel
= kzalloc(sizeof(*accel
), GFP_KERNEL
);
10147 return ERR_PTR(-ENOMEM
);
10149 set_bit(pool
, adapter
->fwd_bitmask
);
10150 netdev_set_sb_channel(vdev
, pool
);
10151 accel
->pool
= pool
;
10152 accel
->netdev
= vdev
;
10154 if (!netif_running(pdev
))
10157 err
= ixgbe_fwd_ring_up(adapter
, accel
);
10159 return ERR_PTR(err
);
10164 static void ixgbe_fwd_del(struct net_device
*pdev
, void *priv
)
10166 struct ixgbe_fwd_adapter
*accel
= priv
;
10167 struct ixgbe_adapter
*adapter
= netdev_priv(pdev
);
10168 unsigned int rxbase
= accel
->rx_base_queue
;
10171 /* delete unicast filter associated with offloaded interface */
10172 ixgbe_del_mac_filter(adapter
, accel
->netdev
->dev_addr
,
10173 VMDQ_P(accel
->pool
));
10175 /* Allow remaining Rx packets to get flushed out of the
10176 * Rx FIFO before we drop the netdev for the ring.
10178 usleep_range(10000, 20000);
10180 for (i
= 0; i
< adapter
->num_rx_queues_per_pool
; i
++) {
10181 struct ixgbe_ring
*ring
= adapter
->rx_ring
[rxbase
+ i
];
10182 struct ixgbe_q_vector
*qv
= ring
->q_vector
;
10184 /* Make sure we aren't processing any packets and clear
10185 * netdev to shut down the ring.
10187 if (netif_running(adapter
->netdev
))
10188 napi_synchronize(&qv
->napi
);
10189 ring
->netdev
= NULL
;
10192 /* unbind the queues and drop the subordinate channel config */
10193 netdev_unbind_sb_channel(pdev
, accel
->netdev
);
10194 netdev_set_sb_channel(accel
->netdev
, 0);
10196 clear_bit(accel
->pool
, adapter
->fwd_bitmask
);
10200 #define IXGBE_MAX_MAC_HDR_LEN 127
10201 #define IXGBE_MAX_NETWORK_HDR_LEN 511
10203 static netdev_features_t
10204 ixgbe_features_check(struct sk_buff
*skb
, struct net_device
*dev
,
10205 netdev_features_t features
)
10207 unsigned int network_hdr_len
, mac_hdr_len
;
10209 /* Make certain the headers can be described by a context descriptor */
10210 mac_hdr_len
= skb_network_header(skb
) - skb
->data
;
10211 if (unlikely(mac_hdr_len
> IXGBE_MAX_MAC_HDR_LEN
))
10212 return features
& ~(NETIF_F_HW_CSUM
|
10214 NETIF_F_GSO_UDP_L4
|
10215 NETIF_F_HW_VLAN_CTAG_TX
|
10219 network_hdr_len
= skb_checksum_start(skb
) - skb_network_header(skb
);
10220 if (unlikely(network_hdr_len
> IXGBE_MAX_NETWORK_HDR_LEN
))
10221 return features
& ~(NETIF_F_HW_CSUM
|
10223 NETIF_F_GSO_UDP_L4
|
10227 /* We can only support IPV4 TSO in tunnels if we can mangle the
10228 * inner IP ID field, so strip TSO if MANGLEID is not supported.
10229 * IPsec offoad sets skb->encapsulation but still can handle
10230 * the TSO, so it's the exception.
10232 if (skb
->encapsulation
&& !(features
& NETIF_F_TSO_MANGLEID
)) {
10233 #ifdef CONFIG_IXGBE_IPSEC
10234 if (!secpath_exists(skb
))
10236 features
&= ~NETIF_F_TSO
;
10242 static int ixgbe_xdp_setup(struct net_device
*dev
, struct bpf_prog
*prog
)
10244 int i
, frame_size
= dev
->mtu
+ ETH_HLEN
+ ETH_FCS_LEN
+ VLAN_HLEN
;
10245 struct ixgbe_adapter
*adapter
= netdev_priv(dev
);
10246 struct bpf_prog
*old_prog
;
10249 if (adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
)
10252 if (adapter
->flags
& IXGBE_FLAG_DCB_ENABLED
)
10255 /* verify ixgbe ring attributes are sufficient for XDP */
10256 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
10257 struct ixgbe_ring
*ring
= adapter
->rx_ring
[i
];
10259 if (ring_is_rsc_enabled(ring
))
10262 if (frame_size
> ixgbe_rx_bufsz(ring
))
10266 if (nr_cpu_ids
> MAX_XDP_QUEUES
)
10269 old_prog
= xchg(&adapter
->xdp_prog
, prog
);
10270 need_reset
= (!!prog
!= !!old_prog
);
10272 /* If transitioning XDP modes reconfigure rings */
10277 /* Wait until ndo_xsk_wakeup completes. */
10279 err
= ixgbe_setup_tc(dev
, adapter
->hw_tcs
);
10282 rcu_assign_pointer(adapter
->xdp_prog
, old_prog
);
10286 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
10287 (void)xchg(&adapter
->rx_ring
[i
]->xdp_prog
,
10288 adapter
->xdp_prog
);
10292 bpf_prog_put(old_prog
);
10294 /* Kick start the NAPI context if there is an AF_XDP socket open
10295 * on that queue id. This so that receiving will start.
10297 if (need_reset
&& prog
)
10298 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
10299 if (adapter
->xdp_ring
[i
]->xsk_umem
)
10300 (void)ixgbe_xsk_wakeup(adapter
->netdev
, i
,
10306 static int ixgbe_xdp(struct net_device
*dev
, struct netdev_bpf
*xdp
)
10308 struct ixgbe_adapter
*adapter
= netdev_priv(dev
);
10310 switch (xdp
->command
) {
10311 case XDP_SETUP_PROG
:
10312 return ixgbe_xdp_setup(dev
, xdp
->prog
);
10313 case XDP_QUERY_PROG
:
10314 xdp
->prog_id
= adapter
->xdp_prog
?
10315 adapter
->xdp_prog
->aux
->id
: 0;
10317 case XDP_SETUP_XSK_UMEM
:
10318 return ixgbe_xsk_umem_setup(adapter
, xdp
->xsk
.umem
,
10319 xdp
->xsk
.queue_id
);
10326 void ixgbe_xdp_ring_update_tail(struct ixgbe_ring
*ring
)
10328 /* Force memory writes to complete before letting h/w know there
10329 * are new descriptors to fetch.
10332 writel(ring
->next_to_use
, ring
->tail
);
10335 static int ixgbe_xdp_xmit(struct net_device
*dev
, int n
,
10336 struct xdp_frame
**frames
, u32 flags
)
10338 struct ixgbe_adapter
*adapter
= netdev_priv(dev
);
10339 struct ixgbe_ring
*ring
;
10343 if (unlikely(test_bit(__IXGBE_DOWN
, &adapter
->state
)))
10346 if (unlikely(flags
& ~XDP_XMIT_FLAGS_MASK
))
10349 /* During program transitions its possible adapter->xdp_prog is assigned
10350 * but ring has not been configured yet. In this case simply abort xmit.
10352 ring
= adapter
->xdp_prog
? adapter
->xdp_ring
[smp_processor_id()] : NULL
;
10353 if (unlikely(!ring
))
10356 if (unlikely(test_bit(__IXGBE_TX_DISABLED
, &ring
->state
)))
10359 for (i
= 0; i
< n
; i
++) {
10360 struct xdp_frame
*xdpf
= frames
[i
];
10363 err
= ixgbe_xmit_xdp_ring(adapter
, xdpf
);
10364 if (err
!= IXGBE_XDP_TX
) {
10365 xdp_return_frame_rx_napi(xdpf
);
10370 if (unlikely(flags
& XDP_XMIT_FLUSH
))
10371 ixgbe_xdp_ring_update_tail(ring
);
10376 static const struct net_device_ops ixgbe_netdev_ops
= {
10377 .ndo_open
= ixgbe_open
,
10378 .ndo_stop
= ixgbe_close
,
10379 .ndo_start_xmit
= ixgbe_xmit_frame
,
10380 .ndo_set_rx_mode
= ixgbe_set_rx_mode
,
10381 .ndo_validate_addr
= eth_validate_addr
,
10382 .ndo_set_mac_address
= ixgbe_set_mac
,
10383 .ndo_change_mtu
= ixgbe_change_mtu
,
10384 .ndo_tx_timeout
= ixgbe_tx_timeout
,
10385 .ndo_set_tx_maxrate
= ixgbe_tx_maxrate
,
10386 .ndo_vlan_rx_add_vid
= ixgbe_vlan_rx_add_vid
,
10387 .ndo_vlan_rx_kill_vid
= ixgbe_vlan_rx_kill_vid
,
10388 .ndo_do_ioctl
= ixgbe_ioctl
,
10389 .ndo_set_vf_mac
= ixgbe_ndo_set_vf_mac
,
10390 .ndo_set_vf_vlan
= ixgbe_ndo_set_vf_vlan
,
10391 .ndo_set_vf_rate
= ixgbe_ndo_set_vf_bw
,
10392 .ndo_set_vf_spoofchk
= ixgbe_ndo_set_vf_spoofchk
,
10393 .ndo_set_vf_rss_query_en
= ixgbe_ndo_set_vf_rss_query_en
,
10394 .ndo_set_vf_trust
= ixgbe_ndo_set_vf_trust
,
10395 .ndo_get_vf_config
= ixgbe_ndo_get_vf_config
,
10396 .ndo_get_stats64
= ixgbe_get_stats64
,
10397 .ndo_setup_tc
= __ixgbe_setup_tc
,
10399 .ndo_select_queue
= ixgbe_select_queue
,
10400 .ndo_fcoe_ddp_setup
= ixgbe_fcoe_ddp_get
,
10401 .ndo_fcoe_ddp_target
= ixgbe_fcoe_ddp_target
,
10402 .ndo_fcoe_ddp_done
= ixgbe_fcoe_ddp_put
,
10403 .ndo_fcoe_enable
= ixgbe_fcoe_enable
,
10404 .ndo_fcoe_disable
= ixgbe_fcoe_disable
,
10405 .ndo_fcoe_get_wwn
= ixgbe_fcoe_get_wwn
,
10406 .ndo_fcoe_get_hbainfo
= ixgbe_fcoe_get_hbainfo
,
10407 #endif /* IXGBE_FCOE */
10408 .ndo_set_features
= ixgbe_set_features
,
10409 .ndo_fix_features
= ixgbe_fix_features
,
10410 .ndo_fdb_add
= ixgbe_ndo_fdb_add
,
10411 .ndo_bridge_setlink
= ixgbe_ndo_bridge_setlink
,
10412 .ndo_bridge_getlink
= ixgbe_ndo_bridge_getlink
,
10413 .ndo_dfwd_add_station
= ixgbe_fwd_add
,
10414 .ndo_dfwd_del_station
= ixgbe_fwd_del
,
10415 .ndo_udp_tunnel_add
= ixgbe_add_udp_tunnel_port
,
10416 .ndo_udp_tunnel_del
= ixgbe_del_udp_tunnel_port
,
10417 .ndo_features_check
= ixgbe_features_check
,
10418 .ndo_bpf
= ixgbe_xdp
,
10419 .ndo_xdp_xmit
= ixgbe_xdp_xmit
,
10420 .ndo_xsk_wakeup
= ixgbe_xsk_wakeup
,
10423 static void ixgbe_disable_txr_hw(struct ixgbe_adapter
*adapter
,
10424 struct ixgbe_ring
*tx_ring
)
10426 unsigned long wait_delay
, delay_interval
;
10427 struct ixgbe_hw
*hw
= &adapter
->hw
;
10428 u8 reg_idx
= tx_ring
->reg_idx
;
10432 IXGBE_WRITE_REG(hw
, IXGBE_TXDCTL(reg_idx
), IXGBE_TXDCTL_SWFLSH
);
10434 /* delay mechanism from ixgbe_disable_tx */
10435 delay_interval
= ixgbe_get_completion_timeout(adapter
) / 100;
10437 wait_loop
= IXGBE_MAX_RX_DESC_POLL
;
10438 wait_delay
= delay_interval
;
10440 while (wait_loop
--) {
10441 usleep_range(wait_delay
, wait_delay
+ 10);
10442 wait_delay
+= delay_interval
* 2;
10443 txdctl
= IXGBE_READ_REG(hw
, IXGBE_TXDCTL(reg_idx
));
10445 if (!(txdctl
& IXGBE_TXDCTL_ENABLE
))
10449 e_err(drv
, "TXDCTL.ENABLE not cleared within the polling period\n");
10452 static void ixgbe_disable_txr(struct ixgbe_adapter
*adapter
,
10453 struct ixgbe_ring
*tx_ring
)
10455 set_bit(__IXGBE_TX_DISABLED
, &tx_ring
->state
);
10456 ixgbe_disable_txr_hw(adapter
, tx_ring
);
10459 static void ixgbe_disable_rxr_hw(struct ixgbe_adapter
*adapter
,
10460 struct ixgbe_ring
*rx_ring
)
10462 unsigned long wait_delay
, delay_interval
;
10463 struct ixgbe_hw
*hw
= &adapter
->hw
;
10464 u8 reg_idx
= rx_ring
->reg_idx
;
10468 rxdctl
= IXGBE_READ_REG(hw
, IXGBE_RXDCTL(reg_idx
));
10469 rxdctl
&= ~IXGBE_RXDCTL_ENABLE
;
10470 rxdctl
|= IXGBE_RXDCTL_SWFLSH
;
10472 /* write value back with RXDCTL.ENABLE bit cleared */
10473 IXGBE_WRITE_REG(hw
, IXGBE_RXDCTL(reg_idx
), rxdctl
);
10475 /* RXDCTL.EN may not change on 82598 if link is down, so skip it */
10476 if (hw
->mac
.type
== ixgbe_mac_82598EB
&&
10477 !(IXGBE_READ_REG(hw
, IXGBE_LINKS
) & IXGBE_LINKS_UP
))
10480 /* delay mechanism from ixgbe_disable_rx */
10481 delay_interval
= ixgbe_get_completion_timeout(adapter
) / 100;
10483 wait_loop
= IXGBE_MAX_RX_DESC_POLL
;
10484 wait_delay
= delay_interval
;
10486 while (wait_loop
--) {
10487 usleep_range(wait_delay
, wait_delay
+ 10);
10488 wait_delay
+= delay_interval
* 2;
10489 rxdctl
= IXGBE_READ_REG(hw
, IXGBE_RXDCTL(reg_idx
));
10491 if (!(rxdctl
& IXGBE_RXDCTL_ENABLE
))
10495 e_err(drv
, "RXDCTL.ENABLE not cleared within the polling period\n");
10498 static void ixgbe_reset_txr_stats(struct ixgbe_ring
*tx_ring
)
10500 memset(&tx_ring
->stats
, 0, sizeof(tx_ring
->stats
));
10501 memset(&tx_ring
->tx_stats
, 0, sizeof(tx_ring
->tx_stats
));
10504 static void ixgbe_reset_rxr_stats(struct ixgbe_ring
*rx_ring
)
10506 memset(&rx_ring
->stats
, 0, sizeof(rx_ring
->stats
));
10507 memset(&rx_ring
->rx_stats
, 0, sizeof(rx_ring
->rx_stats
));
10511 * ixgbe_txrx_ring_disable - Disable Rx/Tx/XDP Tx rings
10512 * @adapter: adapter structure
10513 * @ring: ring index
10515 * This function disables a certain Rx/Tx/XDP Tx ring. The function
10516 * assumes that the netdev is running.
10518 void ixgbe_txrx_ring_disable(struct ixgbe_adapter
*adapter
, int ring
)
10520 struct ixgbe_ring
*rx_ring
, *tx_ring
, *xdp_ring
;
10522 rx_ring
= adapter
->rx_ring
[ring
];
10523 tx_ring
= adapter
->tx_ring
[ring
];
10524 xdp_ring
= adapter
->xdp_ring
[ring
];
10526 ixgbe_disable_txr(adapter
, tx_ring
);
10528 ixgbe_disable_txr(adapter
, xdp_ring
);
10529 ixgbe_disable_rxr_hw(adapter
, rx_ring
);
10534 /* Rx/Tx/XDP Tx share the same napi context. */
10535 napi_disable(&rx_ring
->q_vector
->napi
);
10537 ixgbe_clean_tx_ring(tx_ring
);
10539 ixgbe_clean_tx_ring(xdp_ring
);
10540 ixgbe_clean_rx_ring(rx_ring
);
10542 ixgbe_reset_txr_stats(tx_ring
);
10544 ixgbe_reset_txr_stats(xdp_ring
);
10545 ixgbe_reset_rxr_stats(rx_ring
);
10549 * ixgbe_txrx_ring_enable - Enable Rx/Tx/XDP Tx rings
10550 * @adapter: adapter structure
10551 * @ring: ring index
10553 * This function enables a certain Rx/Tx/XDP Tx ring. The function
10554 * assumes that the netdev is running.
10556 void ixgbe_txrx_ring_enable(struct ixgbe_adapter
*adapter
, int ring
)
10558 struct ixgbe_ring
*rx_ring
, *tx_ring
, *xdp_ring
;
10560 rx_ring
= adapter
->rx_ring
[ring
];
10561 tx_ring
= adapter
->tx_ring
[ring
];
10562 xdp_ring
= adapter
->xdp_ring
[ring
];
10564 /* Rx/Tx/XDP Tx share the same napi context. */
10565 napi_enable(&rx_ring
->q_vector
->napi
);
10567 ixgbe_configure_tx_ring(adapter
, tx_ring
);
10569 ixgbe_configure_tx_ring(adapter
, xdp_ring
);
10570 ixgbe_configure_rx_ring(adapter
, rx_ring
);
10572 clear_bit(__IXGBE_TX_DISABLED
, &tx_ring
->state
);
10574 clear_bit(__IXGBE_TX_DISABLED
, &xdp_ring
->state
);
10578 * ixgbe_enumerate_functions - Get the number of ports this device has
10579 * @adapter: adapter structure
10581 * This function enumerates the phsyical functions co-located on a single slot,
10582 * in order to determine how many ports a device has. This is most useful in
10583 * determining the required GT/s of PCIe bandwidth necessary for optimal
10586 static inline int ixgbe_enumerate_functions(struct ixgbe_adapter
*adapter
)
10588 struct pci_dev
*entry
, *pdev
= adapter
->pdev
;
10591 /* Some cards can not use the generic count PCIe functions method,
10592 * because they are behind a parent switch, so we hardcode these with
10593 * the correct number of functions.
10595 if (ixgbe_pcie_from_parent(&adapter
->hw
))
10598 list_for_each_entry(entry
, &adapter
->pdev
->bus
->devices
, bus_list
) {
10599 /* don't count virtual functions */
10600 if (entry
->is_virtfn
)
10603 /* When the devices on the bus don't all match our device ID,
10604 * we can't reliably determine the correct number of
10605 * functions. This can occur if a function has been direct
10606 * attached to a virtual machine using VT-d, for example. In
10607 * this case, simply return -1 to indicate this.
10609 if ((entry
->vendor
!= pdev
->vendor
) ||
10610 (entry
->device
!= pdev
->device
))
10620 * ixgbe_wol_supported - Check whether device supports WoL
10621 * @adapter: the adapter private structure
10622 * @device_id: the device ID
10623 * @subdevice_id: the subsystem device ID
10625 * This function is used by probe and ethtool to determine
10626 * which devices have WoL support
10629 bool ixgbe_wol_supported(struct ixgbe_adapter
*adapter
, u16 device_id
,
10632 struct ixgbe_hw
*hw
= &adapter
->hw
;
10633 u16 wol_cap
= adapter
->eeprom_cap
& IXGBE_DEVICE_CAPS_WOL_MASK
;
10635 /* WOL not supported on 82598 */
10636 if (hw
->mac
.type
== ixgbe_mac_82598EB
)
10639 /* check eeprom to see if WOL is enabled for X540 and newer */
10640 if (hw
->mac
.type
>= ixgbe_mac_X540
) {
10641 if ((wol_cap
== IXGBE_DEVICE_CAPS_WOL_PORT0_1
) ||
10642 ((wol_cap
== IXGBE_DEVICE_CAPS_WOL_PORT0
) &&
10643 (hw
->bus
.func
== 0)))
10647 /* WOL is determined based on device IDs for 82599 MACs */
10648 switch (device_id
) {
10649 case IXGBE_DEV_ID_82599_SFP
:
10650 /* Only these subdevices could supports WOL */
10651 switch (subdevice_id
) {
10652 case IXGBE_SUBDEV_ID_82599_560FLR
:
10653 case IXGBE_SUBDEV_ID_82599_LOM_SNAP6
:
10654 case IXGBE_SUBDEV_ID_82599_SFP_WOL0
:
10655 case IXGBE_SUBDEV_ID_82599_SFP_2OCP
:
10656 /* only support first port */
10657 if (hw
->bus
.func
!= 0)
10660 case IXGBE_SUBDEV_ID_82599_SP_560FLR
:
10661 case IXGBE_SUBDEV_ID_82599_SFP
:
10662 case IXGBE_SUBDEV_ID_82599_RNDC
:
10663 case IXGBE_SUBDEV_ID_82599_ECNA_DP
:
10664 case IXGBE_SUBDEV_ID_82599_SFP_1OCP
:
10665 case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM1
:
10666 case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM2
:
10670 case IXGBE_DEV_ID_82599EN_SFP
:
10671 /* Only these subdevices support WOL */
10672 switch (subdevice_id
) {
10673 case IXGBE_SUBDEV_ID_82599EN_SFP_OCP1
:
10677 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE
:
10678 /* All except this subdevice support WOL */
10679 if (subdevice_id
!= IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ
)
10682 case IXGBE_DEV_ID_82599_KX4
:
10692 * ixgbe_set_fw_version - Set FW version
10693 * @adapter: the adapter private structure
10695 * This function is used by probe and ethtool to determine the FW version to
10696 * format to display. The FW version is taken from the EEPROM/NVM.
10698 static void ixgbe_set_fw_version(struct ixgbe_adapter
*adapter
)
10700 struct ixgbe_hw
*hw
= &adapter
->hw
;
10701 struct ixgbe_nvm_version nvm_ver
;
10703 ixgbe_get_oem_prod_version(hw
, &nvm_ver
);
10704 if (nvm_ver
.oem_valid
) {
10705 snprintf(adapter
->eeprom_id
, sizeof(adapter
->eeprom_id
),
10706 "%x.%x.%x", nvm_ver
.oem_major
, nvm_ver
.oem_minor
,
10707 nvm_ver
.oem_release
);
10711 ixgbe_get_etk_id(hw
, &nvm_ver
);
10712 ixgbe_get_orom_version(hw
, &nvm_ver
);
10714 if (nvm_ver
.or_valid
) {
10715 snprintf(adapter
->eeprom_id
, sizeof(adapter
->eeprom_id
),
10716 "0x%08x, %d.%d.%d", nvm_ver
.etk_id
, nvm_ver
.or_major
,
10717 nvm_ver
.or_build
, nvm_ver
.or_patch
);
10721 /* Set ETrack ID format */
10722 snprintf(adapter
->eeprom_id
, sizeof(adapter
->eeprom_id
),
10723 "0x%08x", nvm_ver
.etk_id
);
10727 * ixgbe_probe - Device Initialization Routine
10728 * @pdev: PCI device information struct
10729 * @ent: entry in ixgbe_pci_tbl
10731 * Returns 0 on success, negative on failure
10733 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
10734 * The OS initialization, configuring of the adapter private structure,
10735 * and a hardware reset occur.
10737 static int ixgbe_probe(struct pci_dev
*pdev
, const struct pci_device_id
*ent
)
10739 struct net_device
*netdev
;
10740 struct ixgbe_adapter
*adapter
= NULL
;
10741 struct ixgbe_hw
*hw
;
10742 const struct ixgbe_info
*ii
= ixgbe_info_tbl
[ent
->driver_data
];
10743 int i
, err
, pci_using_dac
, expected_gts
;
10744 unsigned int indices
= MAX_TX_QUEUES
;
10745 u8 part_str
[IXGBE_PBANUM_LENGTH
];
10746 bool disable_dev
= false;
10752 /* Catch broken hardware that put the wrong VF device ID in
10753 * the PCIe SR-IOV capability.
10755 if (pdev
->is_virtfn
) {
10756 WARN(1, KERN_ERR
"%s (%hx:%hx) should not be a VF!\n",
10757 pci_name(pdev
), pdev
->vendor
, pdev
->device
);
10761 err
= pci_enable_device_mem(pdev
);
10765 if (!dma_set_mask_and_coherent(&pdev
->dev
, DMA_BIT_MASK(64))) {
10768 err
= dma_set_mask_and_coherent(&pdev
->dev
, DMA_BIT_MASK(32));
10770 dev_err(&pdev
->dev
,
10771 "No usable DMA configuration, aborting\n");
10777 err
= pci_request_mem_regions(pdev
, ixgbe_driver_name
);
10779 dev_err(&pdev
->dev
,
10780 "pci_request_selected_regions failed 0x%x\n", err
);
10784 pci_enable_pcie_error_reporting(pdev
);
10786 pci_set_master(pdev
);
10787 pci_save_state(pdev
);
10789 if (ii
->mac
== ixgbe_mac_82598EB
) {
10790 #ifdef CONFIG_IXGBE_DCB
10791 /* 8 TC w/ 4 queues per TC */
10792 indices
= 4 * MAX_TRAFFIC_CLASS
;
10794 indices
= IXGBE_MAX_RSS_INDICES
;
10798 netdev
= alloc_etherdev_mq(sizeof(struct ixgbe_adapter
), indices
);
10801 goto err_alloc_etherdev
;
10804 SET_NETDEV_DEV(netdev
, &pdev
->dev
);
10806 adapter
= netdev_priv(netdev
);
10808 adapter
->netdev
= netdev
;
10809 adapter
->pdev
= pdev
;
10811 hw
->back
= adapter
;
10812 adapter
->msg_enable
= netif_msg_init(debug
, DEFAULT_MSG_ENABLE
);
10814 hw
->hw_addr
= ioremap(pci_resource_start(pdev
, 0),
10815 pci_resource_len(pdev
, 0));
10816 adapter
->io_addr
= hw
->hw_addr
;
10817 if (!hw
->hw_addr
) {
10822 netdev
->netdev_ops
= &ixgbe_netdev_ops
;
10823 ixgbe_set_ethtool_ops(netdev
);
10824 netdev
->watchdog_timeo
= 5 * HZ
;
10825 strlcpy(netdev
->name
, pci_name(pdev
), sizeof(netdev
->name
));
10828 hw
->mac
.ops
= *ii
->mac_ops
;
10829 hw
->mac
.type
= ii
->mac
;
10830 hw
->mvals
= ii
->mvals
;
10832 hw
->link
.ops
= *ii
->link_ops
;
10835 hw
->eeprom
.ops
= *ii
->eeprom_ops
;
10836 eec
= IXGBE_READ_REG(hw
, IXGBE_EEC(hw
));
10837 if (ixgbe_removed(hw
->hw_addr
)) {
10841 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
10842 if (!(eec
& BIT(8)))
10843 hw
->eeprom
.ops
.read
= &ixgbe_read_eeprom_bit_bang_generic
;
10846 hw
->phy
.ops
= *ii
->phy_ops
;
10847 hw
->phy
.sfp_type
= ixgbe_sfp_type_unknown
;
10848 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
10849 hw
->phy
.mdio
.prtad
= MDIO_PRTAD_NONE
;
10850 hw
->phy
.mdio
.mmds
= 0;
10851 hw
->phy
.mdio
.mode_support
= MDIO_SUPPORTS_C45
| MDIO_EMULATE_C22
;
10852 hw
->phy
.mdio
.dev
= netdev
;
10853 hw
->phy
.mdio
.mdio_read
= ixgbe_mdio_read
;
10854 hw
->phy
.mdio
.mdio_write
= ixgbe_mdio_write
;
10856 /* setup the private structure */
10857 err
= ixgbe_sw_init(adapter
, ii
);
10861 /* Make sure the SWFW semaphore is in a valid state */
10862 if (hw
->mac
.ops
.init_swfw_sync
)
10863 hw
->mac
.ops
.init_swfw_sync(hw
);
10865 /* Make it possible the adapter to be woken up via WOL */
10866 switch (adapter
->hw
.mac
.type
) {
10867 case ixgbe_mac_82599EB
:
10868 case ixgbe_mac_X540
:
10869 case ixgbe_mac_X550
:
10870 case ixgbe_mac_X550EM_x
:
10871 case ixgbe_mac_x550em_a
:
10872 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_WUS
, ~0);
10879 * If there is a fan on this device and it has failed log the
10882 if (adapter
->flags
& IXGBE_FLAG_FAN_FAIL_CAPABLE
) {
10883 u32 esdp
= IXGBE_READ_REG(hw
, IXGBE_ESDP
);
10884 if (esdp
& IXGBE_ESDP_SDP1
)
10885 e_crit(probe
, "Fan has stopped, replace the adapter\n");
10888 if (allow_unsupported_sfp
)
10889 hw
->allow_unsupported_sfp
= allow_unsupported_sfp
;
10891 /* reset_hw fills in the perm_addr as well */
10892 hw
->phy
.reset_if_overtemp
= true;
10893 err
= hw
->mac
.ops
.reset_hw(hw
);
10894 hw
->phy
.reset_if_overtemp
= false;
10895 ixgbe_set_eee_capable(adapter
);
10896 if (err
== IXGBE_ERR_SFP_NOT_PRESENT
) {
10898 } else if (err
== IXGBE_ERR_SFP_NOT_SUPPORTED
) {
10899 e_dev_err("failed to load because an unsupported SFP+ or QSFP module type was detected.\n");
10900 e_dev_err("Reload the driver after installing a supported module.\n");
10903 e_dev_err("HW Init failed: %d\n", err
);
10907 #ifdef CONFIG_PCI_IOV
10908 /* SR-IOV not supported on the 82598 */
10909 if (adapter
->hw
.mac
.type
== ixgbe_mac_82598EB
)
10912 ixgbe_init_mbx_params_pf(hw
);
10913 hw
->mbx
.ops
= ii
->mbx_ops
;
10914 pci_sriov_set_totalvfs(pdev
, IXGBE_MAX_VFS_DRV_LIMIT
);
10915 ixgbe_enable_sriov(adapter
, max_vfs
);
10919 netdev
->features
= NETIF_F_SG
|
10926 #define IXGBE_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
10927 NETIF_F_GSO_GRE_CSUM | \
10928 NETIF_F_GSO_IPXIP4 | \
10929 NETIF_F_GSO_IPXIP6 | \
10930 NETIF_F_GSO_UDP_TUNNEL | \
10931 NETIF_F_GSO_UDP_TUNNEL_CSUM)
10933 netdev
->gso_partial_features
= IXGBE_GSO_PARTIAL_FEATURES
;
10934 netdev
->features
|= NETIF_F_GSO_PARTIAL
|
10935 IXGBE_GSO_PARTIAL_FEATURES
;
10937 if (hw
->mac
.type
>= ixgbe_mac_82599EB
)
10938 netdev
->features
|= NETIF_F_SCTP_CRC
| NETIF_F_GSO_UDP_L4
;
10940 #ifdef CONFIG_IXGBE_IPSEC
10941 #define IXGBE_ESP_FEATURES (NETIF_F_HW_ESP | \
10942 NETIF_F_HW_ESP_TX_CSUM | \
10945 if (adapter
->ipsec
)
10946 netdev
->features
|= IXGBE_ESP_FEATURES
;
10948 /* copy netdev features into list of user selectable features */
10949 netdev
->hw_features
|= netdev
->features
|
10950 NETIF_F_HW_VLAN_CTAG_FILTER
|
10951 NETIF_F_HW_VLAN_CTAG_RX
|
10952 NETIF_F_HW_VLAN_CTAG_TX
|
10954 NETIF_F_HW_L2FW_DOFFLOAD
;
10956 if (hw
->mac
.type
>= ixgbe_mac_82599EB
)
10957 netdev
->hw_features
|= NETIF_F_NTUPLE
|
10961 netdev
->features
|= NETIF_F_HIGHDMA
;
10963 netdev
->vlan_features
|= netdev
->features
| NETIF_F_TSO_MANGLEID
;
10964 netdev
->hw_enc_features
|= netdev
->vlan_features
;
10965 netdev
->mpls_features
|= NETIF_F_SG
|
10969 netdev
->mpls_features
|= IXGBE_GSO_PARTIAL_FEATURES
;
10971 /* set this bit last since it cannot be part of vlan_features */
10972 netdev
->features
|= NETIF_F_HW_VLAN_CTAG_FILTER
|
10973 NETIF_F_HW_VLAN_CTAG_RX
|
10974 NETIF_F_HW_VLAN_CTAG_TX
;
10976 netdev
->priv_flags
|= IFF_UNICAST_FLT
;
10977 netdev
->priv_flags
|= IFF_SUPP_NOFCS
;
10979 /* MTU range: 68 - 9710 */
10980 netdev
->min_mtu
= ETH_MIN_MTU
;
10981 netdev
->max_mtu
= IXGBE_MAX_JUMBO_FRAME_SIZE
- (ETH_HLEN
+ ETH_FCS_LEN
);
10983 #ifdef CONFIG_IXGBE_DCB
10984 if (adapter
->flags
& IXGBE_FLAG_DCB_CAPABLE
)
10985 netdev
->dcbnl_ops
= &ixgbe_dcbnl_ops
;
10989 if (adapter
->flags
& IXGBE_FLAG_FCOE_CAPABLE
) {
10990 unsigned int fcoe_l
;
10992 if (hw
->mac
.ops
.get_device_caps
) {
10993 hw
->mac
.ops
.get_device_caps(hw
, &device_caps
);
10994 if (device_caps
& IXGBE_DEVICE_CAPS_FCOE_OFFLOADS
)
10995 adapter
->flags
&= ~IXGBE_FLAG_FCOE_CAPABLE
;
10999 fcoe_l
= min_t(int, IXGBE_FCRETA_SIZE
, num_online_cpus());
11000 adapter
->ring_feature
[RING_F_FCOE
].limit
= fcoe_l
;
11002 netdev
->features
|= NETIF_F_FSO
|
11005 netdev
->vlan_features
|= NETIF_F_FSO
|
11009 #endif /* IXGBE_FCOE */
11010 if (adapter
->flags2
& IXGBE_FLAG2_RSC_CAPABLE
)
11011 netdev
->hw_features
|= NETIF_F_LRO
;
11012 if (adapter
->flags2
& IXGBE_FLAG2_RSC_ENABLED
)
11013 netdev
->features
|= NETIF_F_LRO
;
11015 if (ixgbe_check_fw_error(adapter
)) {
11020 /* make sure the EEPROM is good */
11021 if (hw
->eeprom
.ops
.validate_checksum(hw
, NULL
) < 0) {
11022 e_dev_err("The EEPROM Checksum Is Not Valid\n");
11027 eth_platform_get_mac_address(&adapter
->pdev
->dev
,
11028 adapter
->hw
.mac
.perm_addr
);
11030 memcpy(netdev
->dev_addr
, hw
->mac
.perm_addr
, netdev
->addr_len
);
11032 if (!is_valid_ether_addr(netdev
->dev_addr
)) {
11033 e_dev_err("invalid MAC address\n");
11038 /* Set hw->mac.addr to permanent MAC address */
11039 ether_addr_copy(hw
->mac
.addr
, hw
->mac
.perm_addr
);
11040 ixgbe_mac_set_default_filter(adapter
);
11042 timer_setup(&adapter
->service_timer
, ixgbe_service_timer
, 0);
11044 if (ixgbe_removed(hw
->hw_addr
)) {
11048 INIT_WORK(&adapter
->service_task
, ixgbe_service_task
);
11049 set_bit(__IXGBE_SERVICE_INITED
, &adapter
->state
);
11050 clear_bit(__IXGBE_SERVICE_SCHED
, &adapter
->state
);
11052 err
= ixgbe_init_interrupt_scheme(adapter
);
11056 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
11057 u64_stats_init(&adapter
->rx_ring
[i
]->syncp
);
11058 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
11059 u64_stats_init(&adapter
->tx_ring
[i
]->syncp
);
11060 for (i
= 0; i
< adapter
->num_xdp_queues
; i
++)
11061 u64_stats_init(&adapter
->xdp_ring
[i
]->syncp
);
11063 /* WOL not supported for all devices */
11065 hw
->eeprom
.ops
.read(hw
, 0x2c, &adapter
->eeprom_cap
);
11066 hw
->wol_enabled
= ixgbe_wol_supported(adapter
, pdev
->device
,
11067 pdev
->subsystem_device
);
11068 if (hw
->wol_enabled
)
11069 adapter
->wol
= IXGBE_WUFC_MAG
;
11071 device_set_wakeup_enable(&adapter
->pdev
->dev
, adapter
->wol
);
11073 /* save off EEPROM version number */
11074 ixgbe_set_fw_version(adapter
);
11076 /* pick up the PCI bus settings for reporting later */
11077 if (ixgbe_pcie_from_parent(hw
))
11078 ixgbe_get_parent_bus_info(adapter
);
11080 hw
->mac
.ops
.get_bus_info(hw
);
11082 /* calculate the expected PCIe bandwidth required for optimal
11083 * performance. Note that some older parts will never have enough
11084 * bandwidth due to being older generation PCIe parts. We clamp these
11085 * parts to ensure no warning is displayed if it can't be fixed.
11087 switch (hw
->mac
.type
) {
11088 case ixgbe_mac_82598EB
:
11089 expected_gts
= min(ixgbe_enumerate_functions(adapter
) * 10, 16);
11092 expected_gts
= ixgbe_enumerate_functions(adapter
) * 10;
11096 /* don't check link if we failed to enumerate functions */
11097 if (expected_gts
> 0)
11098 ixgbe_check_minimum_link(adapter
, expected_gts
);
11100 err
= ixgbe_read_pba_string_generic(hw
, part_str
, sizeof(part_str
));
11102 strlcpy(part_str
, "Unknown", sizeof(part_str
));
11103 if (ixgbe_is_sfp(hw
) && hw
->phy
.sfp_type
!= ixgbe_sfp_type_not_present
)
11104 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
11105 hw
->mac
.type
, hw
->phy
.type
, hw
->phy
.sfp_type
,
11108 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
11109 hw
->mac
.type
, hw
->phy
.type
, part_str
);
11111 e_dev_info("%pM\n", netdev
->dev_addr
);
11113 /* reset the hardware with the new settings */
11114 err
= hw
->mac
.ops
.start_hw(hw
);
11115 if (err
== IXGBE_ERR_EEPROM_VERSION
) {
11116 /* We are running on a pre-production device, log a warning */
11117 e_dev_warn("This device is a pre-production adapter/LOM. "
11118 "Please be aware there may be issues associated "
11119 "with your hardware. If you are experiencing "
11120 "problems please contact your Intel or hardware "
11121 "representative who provided you with this "
11124 strcpy(netdev
->name
, "eth%d");
11125 pci_set_drvdata(pdev
, adapter
);
11126 err
= register_netdev(netdev
);
11131 /* power down the optics for 82599 SFP+ fiber */
11132 if (hw
->mac
.ops
.disable_tx_laser
)
11133 hw
->mac
.ops
.disable_tx_laser(hw
);
11135 /* carrier off reporting is important to ethtool even BEFORE open */
11136 netif_carrier_off(netdev
);
11138 #ifdef CONFIG_IXGBE_DCA
11139 if (dca_add_requester(&pdev
->dev
) == 0) {
11140 adapter
->flags
|= IXGBE_FLAG_DCA_ENABLED
;
11141 ixgbe_setup_dca(adapter
);
11144 if (adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
) {
11145 e_info(probe
, "IOV is enabled with %d VFs\n", adapter
->num_vfs
);
11146 for (i
= 0; i
< adapter
->num_vfs
; i
++)
11147 ixgbe_vf_configuration(pdev
, (i
| 0x10000000));
11150 /* firmware requires driver version to be 0xFFFFFFFF
11151 * since os does not support feature
11153 if (hw
->mac
.ops
.set_fw_drv_ver
)
11154 hw
->mac
.ops
.set_fw_drv_ver(hw
, 0xFF, 0xFF, 0xFF, 0xFF,
11155 sizeof(UTS_RELEASE
) - 1,
11158 /* add san mac addr to netdev */
11159 ixgbe_add_sanmac_netdev(netdev
);
11161 e_dev_info("%s\n", ixgbe_default_device_descr
);
11163 #ifdef CONFIG_IXGBE_HWMON
11164 if (ixgbe_sysfs_init(adapter
))
11165 e_err(probe
, "failed to allocate sysfs resources\n");
11166 #endif /* CONFIG_IXGBE_HWMON */
11168 ixgbe_dbg_adapter_init(adapter
);
11170 /* setup link for SFP devices with MNG FW, else wait for IXGBE_UP */
11171 if (ixgbe_mng_enabled(hw
) && ixgbe_is_sfp(hw
) && hw
->mac
.ops
.setup_link
)
11172 hw
->mac
.ops
.setup_link(hw
,
11173 IXGBE_LINK_SPEED_10GB_FULL
| IXGBE_LINK_SPEED_1GB_FULL
,
11176 err
= ixgbe_mii_bus_init(hw
);
11183 unregister_netdev(netdev
);
11185 ixgbe_release_hw_control(adapter
);
11186 ixgbe_clear_interrupt_scheme(adapter
);
11188 ixgbe_disable_sriov(adapter
);
11189 adapter
->flags2
&= ~IXGBE_FLAG2_SEARCH_FOR_SFP
;
11190 iounmap(adapter
->io_addr
);
11191 kfree(adapter
->jump_tables
[0]);
11192 kfree(adapter
->mac_table
);
11193 kfree(adapter
->rss_key
);
11194 bitmap_free(adapter
->af_xdp_zc_qps
);
11196 disable_dev
= !test_and_set_bit(__IXGBE_DISABLED
, &adapter
->state
);
11197 free_netdev(netdev
);
11198 err_alloc_etherdev
:
11199 pci_release_mem_regions(pdev
);
11202 if (!adapter
|| disable_dev
)
11203 pci_disable_device(pdev
);
11208 * ixgbe_remove - Device Removal Routine
11209 * @pdev: PCI device information struct
11211 * ixgbe_remove is called by the PCI subsystem to alert the driver
11212 * that it should release a PCI device. The could be caused by a
11213 * Hot-Plug event, or because the driver is going to be removed from
11216 static void ixgbe_remove(struct pci_dev
*pdev
)
11218 struct ixgbe_adapter
*adapter
= pci_get_drvdata(pdev
);
11219 struct net_device
*netdev
;
11223 /* if !adapter then we already cleaned up in probe */
11227 netdev
= adapter
->netdev
;
11228 ixgbe_dbg_adapter_exit(adapter
);
11230 set_bit(__IXGBE_REMOVING
, &adapter
->state
);
11231 cancel_work_sync(&adapter
->service_task
);
11233 if (adapter
->mii_bus
)
11234 mdiobus_unregister(adapter
->mii_bus
);
11236 #ifdef CONFIG_IXGBE_DCA
11237 if (adapter
->flags
& IXGBE_FLAG_DCA_ENABLED
) {
11238 adapter
->flags
&= ~IXGBE_FLAG_DCA_ENABLED
;
11239 dca_remove_requester(&pdev
->dev
);
11240 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_DCA_CTRL
,
11241 IXGBE_DCA_CTRL_DCA_DISABLE
);
11245 #ifdef CONFIG_IXGBE_HWMON
11246 ixgbe_sysfs_exit(adapter
);
11247 #endif /* CONFIG_IXGBE_HWMON */
11249 /* remove the added san mac */
11250 ixgbe_del_sanmac_netdev(netdev
);
11252 #ifdef CONFIG_PCI_IOV
11253 ixgbe_disable_sriov(adapter
);
11255 if (netdev
->reg_state
== NETREG_REGISTERED
)
11256 unregister_netdev(netdev
);
11258 ixgbe_stop_ipsec_offload(adapter
);
11259 ixgbe_clear_interrupt_scheme(adapter
);
11261 ixgbe_release_hw_control(adapter
);
11264 kfree(adapter
->ixgbe_ieee_pfc
);
11265 kfree(adapter
->ixgbe_ieee_ets
);
11268 iounmap(adapter
->io_addr
);
11269 pci_release_mem_regions(pdev
);
11271 e_dev_info("complete\n");
11273 for (i
= 0; i
< IXGBE_MAX_LINK_HANDLE
; i
++) {
11274 if (adapter
->jump_tables
[i
]) {
11275 kfree(adapter
->jump_tables
[i
]->input
);
11276 kfree(adapter
->jump_tables
[i
]->mask
);
11278 kfree(adapter
->jump_tables
[i
]);
11281 kfree(adapter
->mac_table
);
11282 kfree(adapter
->rss_key
);
11283 bitmap_free(adapter
->af_xdp_zc_qps
);
11284 disable_dev
= !test_and_set_bit(__IXGBE_DISABLED
, &adapter
->state
);
11285 free_netdev(netdev
);
11287 pci_disable_pcie_error_reporting(pdev
);
11290 pci_disable_device(pdev
);
11294 * ixgbe_io_error_detected - called when PCI error is detected
11295 * @pdev: Pointer to PCI device
11296 * @state: The current pci connection state
11298 * This function is called after a PCI bus error affecting
11299 * this device has been detected.
11301 static pci_ers_result_t
ixgbe_io_error_detected(struct pci_dev
*pdev
,
11302 pci_channel_state_t state
)
11304 struct ixgbe_adapter
*adapter
= pci_get_drvdata(pdev
);
11305 struct net_device
*netdev
= adapter
->netdev
;
11307 #ifdef CONFIG_PCI_IOV
11308 struct ixgbe_hw
*hw
= &adapter
->hw
;
11309 struct pci_dev
*bdev
, *vfdev
;
11310 u32 dw0
, dw1
, dw2
, dw3
;
11312 u16 req_id
, pf_func
;
11314 if (adapter
->hw
.mac
.type
== ixgbe_mac_82598EB
||
11315 adapter
->num_vfs
== 0)
11316 goto skip_bad_vf_detection
;
11318 bdev
= pdev
->bus
->self
;
11319 while (bdev
&& (pci_pcie_type(bdev
) != PCI_EXP_TYPE_ROOT_PORT
))
11320 bdev
= bdev
->bus
->self
;
11323 goto skip_bad_vf_detection
;
11325 pos
= pci_find_ext_capability(bdev
, PCI_EXT_CAP_ID_ERR
);
11327 goto skip_bad_vf_detection
;
11329 dw0
= ixgbe_read_pci_cfg_dword(hw
, pos
+ PCI_ERR_HEADER_LOG
);
11330 dw1
= ixgbe_read_pci_cfg_dword(hw
, pos
+ PCI_ERR_HEADER_LOG
+ 4);
11331 dw2
= ixgbe_read_pci_cfg_dword(hw
, pos
+ PCI_ERR_HEADER_LOG
+ 8);
11332 dw3
= ixgbe_read_pci_cfg_dword(hw
, pos
+ PCI_ERR_HEADER_LOG
+ 12);
11333 if (ixgbe_removed(hw
->hw_addr
))
11334 goto skip_bad_vf_detection
;
11336 req_id
= dw1
>> 16;
11337 /* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
11338 if (!(req_id
& 0x0080))
11339 goto skip_bad_vf_detection
;
11341 pf_func
= req_id
& 0x01;
11342 if ((pf_func
& 1) == (pdev
->devfn
& 1)) {
11343 unsigned int device_id
;
11345 vf
= (req_id
& 0x7F) >> 1;
11346 e_dev_err("VF %d has caused a PCIe error\n", vf
);
11347 e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
11348 "%8.8x\tdw3: %8.8x\n",
11349 dw0
, dw1
, dw2
, dw3
);
11350 switch (adapter
->hw
.mac
.type
) {
11351 case ixgbe_mac_82599EB
:
11352 device_id
= IXGBE_82599_VF_DEVICE_ID
;
11354 case ixgbe_mac_X540
:
11355 device_id
= IXGBE_X540_VF_DEVICE_ID
;
11357 case ixgbe_mac_X550
:
11358 device_id
= IXGBE_DEV_ID_X550_VF
;
11360 case ixgbe_mac_X550EM_x
:
11361 device_id
= IXGBE_DEV_ID_X550EM_X_VF
;
11363 case ixgbe_mac_x550em_a
:
11364 device_id
= IXGBE_DEV_ID_X550EM_A_VF
;
11371 /* Find the pci device of the offending VF */
11372 vfdev
= pci_get_device(PCI_VENDOR_ID_INTEL
, device_id
, NULL
);
11374 if (vfdev
->devfn
== (req_id
& 0xFF))
11376 vfdev
= pci_get_device(PCI_VENDOR_ID_INTEL
,
11380 * There's a slim chance the VF could have been hot plugged,
11381 * so if it is no longer present we don't need to issue the
11382 * VFLR. Just clean up the AER in that case.
11386 /* Free device reference count */
11387 pci_dev_put(vfdev
);
11392 * Even though the error may have occurred on the other port
11393 * we still need to increment the vf error reference count for
11394 * both ports because the I/O resume function will be called
11395 * for both of them.
11397 adapter
->vferr_refcount
++;
11399 return PCI_ERS_RESULT_RECOVERED
;
11401 skip_bad_vf_detection
:
11402 #endif /* CONFIG_PCI_IOV */
11403 if (!test_bit(__IXGBE_SERVICE_INITED
, &adapter
->state
))
11404 return PCI_ERS_RESULT_DISCONNECT
;
11406 if (!netif_device_present(netdev
))
11407 return PCI_ERS_RESULT_DISCONNECT
;
11410 netif_device_detach(netdev
);
11412 if (netif_running(netdev
))
11413 ixgbe_close_suspend(adapter
);
11415 if (state
== pci_channel_io_perm_failure
) {
11417 return PCI_ERS_RESULT_DISCONNECT
;
11420 if (!test_and_set_bit(__IXGBE_DISABLED
, &adapter
->state
))
11421 pci_disable_device(pdev
);
11424 /* Request a slot reset. */
11425 return PCI_ERS_RESULT_NEED_RESET
;
11429 * ixgbe_io_slot_reset - called after the pci bus has been reset.
11430 * @pdev: Pointer to PCI device
11432 * Restart the card from scratch, as if from a cold-boot.
11434 static pci_ers_result_t
ixgbe_io_slot_reset(struct pci_dev
*pdev
)
11436 struct ixgbe_adapter
*adapter
= pci_get_drvdata(pdev
);
11437 pci_ers_result_t result
;
11439 if (pci_enable_device_mem(pdev
)) {
11440 e_err(probe
, "Cannot re-enable PCI device after reset.\n");
11441 result
= PCI_ERS_RESULT_DISCONNECT
;
11443 smp_mb__before_atomic();
11444 clear_bit(__IXGBE_DISABLED
, &adapter
->state
);
11445 adapter
->hw
.hw_addr
= adapter
->io_addr
;
11446 pci_set_master(pdev
);
11447 pci_restore_state(pdev
);
11448 pci_save_state(pdev
);
11450 pci_wake_from_d3(pdev
, false);
11452 ixgbe_reset(adapter
);
11453 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_WUS
, ~0);
11454 result
= PCI_ERS_RESULT_RECOVERED
;
11461 * ixgbe_io_resume - called when traffic can start flowing again.
11462 * @pdev: Pointer to PCI device
11464 * This callback is called when the error recovery driver tells us that
11465 * its OK to resume normal operation.
11467 static void ixgbe_io_resume(struct pci_dev
*pdev
)
11469 struct ixgbe_adapter
*adapter
= pci_get_drvdata(pdev
);
11470 struct net_device
*netdev
= adapter
->netdev
;
11472 #ifdef CONFIG_PCI_IOV
11473 if (adapter
->vferr_refcount
) {
11474 e_info(drv
, "Resuming after VF err\n");
11475 adapter
->vferr_refcount
--;
11481 if (netif_running(netdev
))
11482 ixgbe_open(netdev
);
11484 netif_device_attach(netdev
);
11488 static const struct pci_error_handlers ixgbe_err_handler
= {
11489 .error_detected
= ixgbe_io_error_detected
,
11490 .slot_reset
= ixgbe_io_slot_reset
,
11491 .resume
= ixgbe_io_resume
,
11494 static struct pci_driver ixgbe_driver
= {
11495 .name
= ixgbe_driver_name
,
11496 .id_table
= ixgbe_pci_tbl
,
11497 .probe
= ixgbe_probe
,
11498 .remove
= ixgbe_remove
,
11500 .suspend
= ixgbe_suspend
,
11501 .resume
= ixgbe_resume
,
11503 .shutdown
= ixgbe_shutdown
,
11504 .sriov_configure
= ixgbe_pci_sriov_configure
,
11505 .err_handler
= &ixgbe_err_handler
11509 * ixgbe_init_module - Driver Registration Routine
11511 * ixgbe_init_module is the first routine called when the driver is
11512 * loaded. All it does is register with the PCI subsystem.
11514 static int __init
ixgbe_init_module(void)
11517 pr_info("%s\n", ixgbe_driver_string
);
11518 pr_info("%s\n", ixgbe_copyright
);
11520 ixgbe_wq
= create_singlethread_workqueue(ixgbe_driver_name
);
11522 pr_err("%s: Failed to create workqueue\n", ixgbe_driver_name
);
11528 ret
= pci_register_driver(&ixgbe_driver
);
11530 destroy_workqueue(ixgbe_wq
);
11535 #ifdef CONFIG_IXGBE_DCA
11536 dca_register_notify(&dca_notifier
);
11542 module_init(ixgbe_init_module
);
11545 * ixgbe_exit_module - Driver Exit Cleanup Routine
11547 * ixgbe_exit_module is called just before the driver is removed
11550 static void __exit
ixgbe_exit_module(void)
11552 #ifdef CONFIG_IXGBE_DCA
11553 dca_unregister_notify(&dca_notifier
);
11555 pci_unregister_driver(&ixgbe_driver
);
11559 destroy_workqueue(ixgbe_wq
);
11564 #ifdef CONFIG_IXGBE_DCA
11565 static int ixgbe_notify_dca(struct notifier_block
*nb
, unsigned long event
,
11570 ret_val
= driver_for_each_device(&ixgbe_driver
.driver
, NULL
, &event
,
11571 __ixgbe_notify_dca
);
11573 return ret_val
? NOTIFY_BAD
: NOTIFY_DONE
;
11576 #endif /* CONFIG_IXGBE_DCA */
11578 module_exit(ixgbe_exit_module
);