xref: /dpdk/drivers/net/e1000/igb_ethdev.c (revision ba57777d)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2016 Intel Corporation
3  */
4 
5 #include <sys/queue.h>
6 #include <stdio.h>
7 #include <errno.h>
8 #include <stdint.h>
9 #include <stdarg.h>
10 
11 #include <rte_string_fns.h>
12 #include <rte_common.h>
13 #include <rte_interrupts.h>
14 #include <rte_byteorder.h>
15 #include <rte_log.h>
16 #include <rte_debug.h>
17 #include <rte_pci.h>
18 #include <rte_bus_pci.h>
19 #include <rte_ether.h>
20 #include <ethdev_driver.h>
21 #include <ethdev_pci.h>
22 #include <rte_memory.h>
23 #include <rte_eal.h>
24 #include <rte_malloc.h>
25 #include <rte_dev.h>
26 
27 #include "e1000_logs.h"
28 #include "base/e1000_api.h"
29 #include "e1000_ethdev.h"
30 #include "igb_regs.h"
31 
32 /*
33  * Default values for port configuration
34  */
35 #define IGB_DEFAULT_RX_FREE_THRESH  32
36 
37 #define IGB_DEFAULT_RX_PTHRESH      ((hw->mac.type == e1000_i354) ? 12 : 8)
38 #define IGB_DEFAULT_RX_HTHRESH      8
39 #define IGB_DEFAULT_RX_WTHRESH      ((hw->mac.type == e1000_82576) ? 1 : 4)
40 
41 #define IGB_DEFAULT_TX_PTHRESH      ((hw->mac.type == e1000_i354) ? 20 : 8)
42 #define IGB_DEFAULT_TX_HTHRESH      1
43 #define IGB_DEFAULT_TX_WTHRESH      ((hw->mac.type == e1000_82576) ? 1 : 16)
44 
45 /* Bit shift and mask */
46 #define IGB_4_BIT_WIDTH  (CHAR_BIT / 2)
47 #define IGB_4_BIT_MASK   RTE_LEN2MASK(IGB_4_BIT_WIDTH, uint8_t)
48 #define IGB_8_BIT_WIDTH  CHAR_BIT
49 #define IGB_8_BIT_MASK   UINT8_MAX
50 
51 /* Additional timesync values. */
52 #define E1000_CYCLECOUNTER_MASK      0xffffffffffffffffULL
53 #define E1000_ETQF_FILTER_1588       3
54 #define IGB_82576_TSYNC_SHIFT        16
55 #define E1000_INCPERIOD_82576        (1 << E1000_TIMINCA_16NS_SHIFT)
56 #define E1000_INCVALUE_82576         (16 << IGB_82576_TSYNC_SHIFT)
57 #define E1000_TSAUXC_DISABLE_SYSTIME 0x80000000
58 
59 #define E1000_VTIVAR_MISC                0x01740
60 #define E1000_VTIVAR_MISC_MASK           0xFF
61 #define E1000_VTIVAR_VALID               0x80
62 #define E1000_VTIVAR_MISC_MAILBOX        0
63 #define E1000_VTIVAR_MISC_INTR_MASK      0x3
64 
65 /* External VLAN Enable bit mask */
66 #define E1000_CTRL_EXT_EXT_VLAN      (1 << 26)
67 
68 /* External VLAN Ether Type bit mask and shift */
69 #define E1000_VET_VET_EXT            0xFFFF0000
70 #define E1000_VET_VET_EXT_SHIFT      16
71 
72 /* MSI-X other interrupt vector */
73 #define IGB_MSIX_OTHER_INTR_VEC      0
74 
75 static int  eth_igb_configure(struct rte_eth_dev *dev);
76 static int  eth_igb_start(struct rte_eth_dev *dev);
77 static int  eth_igb_stop(struct rte_eth_dev *dev);
78 static int  eth_igb_dev_set_link_up(struct rte_eth_dev *dev);
79 static int  eth_igb_dev_set_link_down(struct rte_eth_dev *dev);
80 static int eth_igb_close(struct rte_eth_dev *dev);
81 static int eth_igb_reset(struct rte_eth_dev *dev);
82 static int  eth_igb_promiscuous_enable(struct rte_eth_dev *dev);
83 static int  eth_igb_promiscuous_disable(struct rte_eth_dev *dev);
84 static int  eth_igb_allmulticast_enable(struct rte_eth_dev *dev);
85 static int  eth_igb_allmulticast_disable(struct rte_eth_dev *dev);
86 static int  eth_igb_link_update(struct rte_eth_dev *dev,
87 				int wait_to_complete);
88 static int eth_igb_stats_get(struct rte_eth_dev *dev,
89 				struct rte_eth_stats *rte_stats);
90 static int eth_igb_xstats_get(struct rte_eth_dev *dev,
91 			      struct rte_eth_xstat *xstats, unsigned n);
92 static int eth_igb_xstats_get_by_id(struct rte_eth_dev *dev,
93 		const uint64_t *ids,
94 		uint64_t *values, unsigned int n);
95 static int eth_igb_xstats_get_names(struct rte_eth_dev *dev,
96 				    struct rte_eth_xstat_name *xstats_names,
97 				    unsigned int size);
98 static int eth_igb_xstats_get_names_by_id(struct rte_eth_dev *dev,
99 		struct rte_eth_xstat_name *xstats_names, const uint64_t *ids,
100 		unsigned int limit);
101 static int eth_igb_stats_reset(struct rte_eth_dev *dev);
102 static int eth_igb_xstats_reset(struct rte_eth_dev *dev);
103 static int eth_igb_fw_version_get(struct rte_eth_dev *dev,
104 				   char *fw_version, size_t fw_size);
105 static int eth_igb_infos_get(struct rte_eth_dev *dev,
106 			      struct rte_eth_dev_info *dev_info);
107 static const uint32_t *eth_igb_supported_ptypes_get(struct rte_eth_dev *dev);
108 static int eth_igbvf_infos_get(struct rte_eth_dev *dev,
109 				struct rte_eth_dev_info *dev_info);
110 static int  eth_igb_flow_ctrl_get(struct rte_eth_dev *dev,
111 				struct rte_eth_fc_conf *fc_conf);
112 static int  eth_igb_flow_ctrl_set(struct rte_eth_dev *dev,
113 				struct rte_eth_fc_conf *fc_conf);
114 static int eth_igb_lsc_interrupt_setup(struct rte_eth_dev *dev, uint8_t on);
115 static int eth_igb_rxq_interrupt_setup(struct rte_eth_dev *dev);
116 static int eth_igb_interrupt_get_status(struct rte_eth_dev *dev);
117 static int eth_igb_interrupt_action(struct rte_eth_dev *dev,
118 				    struct rte_intr_handle *handle);
119 static void eth_igb_interrupt_handler(void *param);
120 static int  igb_hardware_init(struct e1000_hw *hw);
121 static void igb_hw_control_acquire(struct e1000_hw *hw);
122 static void igb_hw_control_release(struct e1000_hw *hw);
123 static void igb_init_manageability(struct e1000_hw *hw);
124 static void igb_release_manageability(struct e1000_hw *hw);
125 
126 static int  eth_igb_mtu_set(struct rte_eth_dev *dev, uint16_t mtu);
127 
128 static int eth_igb_vlan_filter_set(struct rte_eth_dev *dev,
129 		uint16_t vlan_id, int on);
130 static int eth_igb_vlan_tpid_set(struct rte_eth_dev *dev,
131 				 enum rte_vlan_type vlan_type,
132 				 uint16_t tpid_id);
133 static int eth_igb_vlan_offload_set(struct rte_eth_dev *dev, int mask);
134 
135 static void igb_vlan_hw_filter_enable(struct rte_eth_dev *dev);
136 static void igb_vlan_hw_filter_disable(struct rte_eth_dev *dev);
137 static void igb_vlan_hw_strip_enable(struct rte_eth_dev *dev);
138 static void igb_vlan_hw_strip_disable(struct rte_eth_dev *dev);
139 static void igb_vlan_hw_extend_enable(struct rte_eth_dev *dev);
140 static void igb_vlan_hw_extend_disable(struct rte_eth_dev *dev);
141 
142 static int eth_igb_led_on(struct rte_eth_dev *dev);
143 static int eth_igb_led_off(struct rte_eth_dev *dev);
144 
145 static void igb_intr_disable(struct rte_eth_dev *dev);
146 static int  igb_get_rx_buffer_size(struct e1000_hw *hw);
147 static int eth_igb_rar_set(struct rte_eth_dev *dev,
148 			   struct rte_ether_addr *mac_addr,
149 			   uint32_t index, uint32_t pool);
150 static void eth_igb_rar_clear(struct rte_eth_dev *dev, uint32_t index);
151 static int eth_igb_default_mac_addr_set(struct rte_eth_dev *dev,
152 		struct rte_ether_addr *addr);
153 
154 static void igbvf_intr_disable(struct e1000_hw *hw);
155 static int igbvf_dev_configure(struct rte_eth_dev *dev);
156 static int igbvf_dev_start(struct rte_eth_dev *dev);
157 static int igbvf_dev_stop(struct rte_eth_dev *dev);
158 static int igbvf_dev_close(struct rte_eth_dev *dev);
159 static int igbvf_promiscuous_enable(struct rte_eth_dev *dev);
160 static int igbvf_promiscuous_disable(struct rte_eth_dev *dev);
161 static int igbvf_allmulticast_enable(struct rte_eth_dev *dev);
162 static int igbvf_allmulticast_disable(struct rte_eth_dev *dev);
163 static int eth_igbvf_link_update(struct e1000_hw *hw);
164 static int eth_igbvf_stats_get(struct rte_eth_dev *dev,
165 				struct rte_eth_stats *rte_stats);
166 static int eth_igbvf_xstats_get(struct rte_eth_dev *dev,
167 				struct rte_eth_xstat *xstats, unsigned n);
168 static int eth_igbvf_xstats_get_names(struct rte_eth_dev *dev,
169 				      struct rte_eth_xstat_name *xstats_names,
170 				      unsigned limit);
171 static int eth_igbvf_stats_reset(struct rte_eth_dev *dev);
172 static int igbvf_vlan_filter_set(struct rte_eth_dev *dev,
173 		uint16_t vlan_id, int on);
174 static int igbvf_set_vfta(struct e1000_hw *hw, uint16_t vid, bool on);
175 static void igbvf_set_vfta_all(struct rte_eth_dev *dev, bool on);
176 static int igbvf_default_mac_addr_set(struct rte_eth_dev *dev,
177 		struct rte_ether_addr *addr);
178 static int igbvf_get_reg_length(struct rte_eth_dev *dev);
179 static int igbvf_get_regs(struct rte_eth_dev *dev,
180 		struct rte_dev_reg_info *regs);
181 
182 static int eth_igb_rss_reta_update(struct rte_eth_dev *dev,
183 				   struct rte_eth_rss_reta_entry64 *reta_conf,
184 				   uint16_t reta_size);
185 static int eth_igb_rss_reta_query(struct rte_eth_dev *dev,
186 				  struct rte_eth_rss_reta_entry64 *reta_conf,
187 				  uint16_t reta_size);
188 
189 static int igb_add_2tuple_filter(struct rte_eth_dev *dev,
190 			struct rte_eth_ntuple_filter *ntuple_filter);
191 static int igb_remove_2tuple_filter(struct rte_eth_dev *dev,
192 			struct rte_eth_ntuple_filter *ntuple_filter);
193 static int igb_add_5tuple_filter_82576(struct rte_eth_dev *dev,
194 			struct rte_eth_ntuple_filter *ntuple_filter);
195 static int igb_remove_5tuple_filter_82576(struct rte_eth_dev *dev,
196 			struct rte_eth_ntuple_filter *ntuple_filter);
197 static int eth_igb_flow_ops_get(struct rte_eth_dev *dev,
198 				const struct rte_flow_ops **ops);
199 static int eth_igb_get_reg_length(struct rte_eth_dev *dev);
200 static int eth_igb_get_regs(struct rte_eth_dev *dev,
201 		struct rte_dev_reg_info *regs);
202 static int eth_igb_get_eeprom_length(struct rte_eth_dev *dev);
203 static int eth_igb_get_eeprom(struct rte_eth_dev *dev,
204 		struct rte_dev_eeprom_info *eeprom);
205 static int eth_igb_set_eeprom(struct rte_eth_dev *dev,
206 		struct rte_dev_eeprom_info *eeprom);
207 static int eth_igb_get_module_info(struct rte_eth_dev *dev,
208 				   struct rte_eth_dev_module_info *modinfo);
209 static int eth_igb_get_module_eeprom(struct rte_eth_dev *dev,
210 				     struct rte_dev_eeprom_info *info);
211 static int eth_igb_set_mc_addr_list(struct rte_eth_dev *dev,
212 				    struct rte_ether_addr *mc_addr_set,
213 				    uint32_t nb_mc_addr);
214 static int igb_timesync_enable(struct rte_eth_dev *dev);
215 static int igb_timesync_disable(struct rte_eth_dev *dev);
216 static int igb_timesync_read_rx_timestamp(struct rte_eth_dev *dev,
217 					  struct timespec *timestamp,
218 					  uint32_t flags);
219 static int igb_timesync_read_tx_timestamp(struct rte_eth_dev *dev,
220 					  struct timespec *timestamp);
221 static int igb_timesync_adjust_time(struct rte_eth_dev *dev, int64_t delta);
222 static int igb_timesync_read_time(struct rte_eth_dev *dev,
223 				  struct timespec *timestamp);
224 static int igb_timesync_write_time(struct rte_eth_dev *dev,
225 				   const struct timespec *timestamp);
226 static int eth_igb_rx_queue_intr_enable(struct rte_eth_dev *dev,
227 					uint16_t queue_id);
228 static int eth_igb_rx_queue_intr_disable(struct rte_eth_dev *dev,
229 					 uint16_t queue_id);
230 static void eth_igb_assign_msix_vector(struct e1000_hw *hw, int8_t direction,
231 				       uint8_t queue, uint8_t msix_vector);
232 static void eth_igb_write_ivar(struct e1000_hw *hw, uint8_t msix_vector,
233 			       uint8_t index, uint8_t offset);
234 static void eth_igb_configure_msix_intr(struct rte_eth_dev *dev);
235 static void eth_igbvf_interrupt_handler(void *param);
236 static void igbvf_mbx_process(struct rte_eth_dev *dev);
237 static int igb_filter_restore(struct rte_eth_dev *dev);
238 
239 /*
240  * Define VF Stats MACRO for Non "cleared on read" register
241  */
242 #define UPDATE_VF_STAT(reg, last, cur)            \
243 {                                                 \
244 	u32 latest = E1000_READ_REG(hw, reg);     \
245 	cur += (latest - last) & UINT_MAX;        \
246 	last = latest;                            \
247 }
248 
249 #define IGB_FC_PAUSE_TIME 0x0680
250 #define IGB_LINK_UPDATE_CHECK_TIMEOUT  90  /* 9s */
251 #define IGB_LINK_UPDATE_CHECK_INTERVAL 100 /* ms */
252 
253 #define IGBVF_PMD_NAME "rte_igbvf_pmd"     /* PMD name */
254 
255 static enum e1000_fc_mode igb_fc_setting = e1000_fc_full;
256 
257 /*
258  * The set of PCI devices this driver supports
259  */
260 static const struct rte_pci_id pci_id_igb_map[] = {
261 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576) },
262 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576_FIBER) },
263 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576_SERDES) },
264 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576_QUAD_COPPER) },
265 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576_QUAD_COPPER_ET2) },
266 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576_NS) },
267 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576_NS_SERDES) },
268 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576_SERDES_QUAD) },
269 
270 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82575EB_COPPER) },
271 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82575EB_FIBER_SERDES) },
272 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82575GB_QUAD_COPPER) },
273 
274 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82580_COPPER) },
275 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82580_FIBER) },
276 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82580_SERDES) },
277 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82580_SGMII) },
278 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82580_COPPER_DUAL) },
279 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82580_QUAD_FIBER) },
280 
281 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I350_COPPER) },
282 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I350_FIBER) },
283 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I350_SERDES) },
284 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I350_SGMII) },
285 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I350_DA4) },
286 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I210_COPPER) },
287 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I210_COPPER_OEM1) },
288 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I210_COPPER_IT) },
289 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I210_FIBER) },
290 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I210_SERDES) },
291 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I210_SGMII) },
292 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I210_COPPER_FLASHLESS) },
293 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I210_SERDES_FLASHLESS) },
294 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I211_COPPER) },
295 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I354_BACKPLANE_1GBPS) },
296 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I354_SGMII) },
297 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I354_BACKPLANE_2_5GBPS) },
298 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_DH89XXCC_SGMII) },
299 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_DH89XXCC_SERDES) },
300 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_DH89XXCC_BACKPLANE) },
301 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_DH89XXCC_SFP) },
302 	{ .vendor_id = 0, /* sentinel */ },
303 };
304 
305 /*
306  * The set of PCI devices this driver supports (for 82576&I350 VF)
307  */
308 static const struct rte_pci_id pci_id_igbvf_map[] = {
309 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576_VF) },
310 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576_VF_HV) },
311 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I350_VF) },
312 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I350_VF_HV) },
313 	{ .vendor_id = 0, /* sentinel */ },
314 };
315 
316 static const struct rte_eth_desc_lim rx_desc_lim = {
317 	.nb_max = E1000_MAX_RING_DESC,
318 	.nb_min = E1000_MIN_RING_DESC,
319 	.nb_align = IGB_RXD_ALIGN,
320 };
321 
322 static const struct rte_eth_desc_lim tx_desc_lim = {
323 	.nb_max = E1000_MAX_RING_DESC,
324 	.nb_min = E1000_MIN_RING_DESC,
325 	.nb_align = IGB_RXD_ALIGN,
326 	.nb_seg_max = IGB_TX_MAX_SEG,
327 	.nb_mtu_seg_max = IGB_TX_MAX_MTU_SEG,
328 };
329 
330 static const struct eth_dev_ops eth_igb_ops = {
331 	.dev_configure        = eth_igb_configure,
332 	.dev_start            = eth_igb_start,
333 	.dev_stop             = eth_igb_stop,
334 	.dev_set_link_up      = eth_igb_dev_set_link_up,
335 	.dev_set_link_down    = eth_igb_dev_set_link_down,
336 	.dev_close            = eth_igb_close,
337 	.dev_reset            = eth_igb_reset,
338 	.promiscuous_enable   = eth_igb_promiscuous_enable,
339 	.promiscuous_disable  = eth_igb_promiscuous_disable,
340 	.allmulticast_enable  = eth_igb_allmulticast_enable,
341 	.allmulticast_disable = eth_igb_allmulticast_disable,
342 	.link_update          = eth_igb_link_update,
343 	.stats_get            = eth_igb_stats_get,
344 	.xstats_get           = eth_igb_xstats_get,
345 	.xstats_get_by_id     = eth_igb_xstats_get_by_id,
346 	.xstats_get_names_by_id = eth_igb_xstats_get_names_by_id,
347 	.xstats_get_names     = eth_igb_xstats_get_names,
348 	.stats_reset          = eth_igb_stats_reset,
349 	.xstats_reset         = eth_igb_xstats_reset,
350 	.fw_version_get       = eth_igb_fw_version_get,
351 	.dev_infos_get        = eth_igb_infos_get,
352 	.dev_supported_ptypes_get = eth_igb_supported_ptypes_get,
353 	.mtu_set              = eth_igb_mtu_set,
354 	.vlan_filter_set      = eth_igb_vlan_filter_set,
355 	.vlan_tpid_set        = eth_igb_vlan_tpid_set,
356 	.vlan_offload_set     = eth_igb_vlan_offload_set,
357 	.rx_queue_setup       = eth_igb_rx_queue_setup,
358 	.rx_queue_intr_enable = eth_igb_rx_queue_intr_enable,
359 	.rx_queue_intr_disable = eth_igb_rx_queue_intr_disable,
360 	.rx_queue_release     = eth_igb_rx_queue_release,
361 	.tx_queue_setup       = eth_igb_tx_queue_setup,
362 	.tx_queue_release     = eth_igb_tx_queue_release,
363 	.tx_done_cleanup      = eth_igb_tx_done_cleanup,
364 	.dev_led_on           = eth_igb_led_on,
365 	.dev_led_off          = eth_igb_led_off,
366 	.flow_ctrl_get        = eth_igb_flow_ctrl_get,
367 	.flow_ctrl_set        = eth_igb_flow_ctrl_set,
368 	.mac_addr_add         = eth_igb_rar_set,
369 	.mac_addr_remove      = eth_igb_rar_clear,
370 	.mac_addr_set         = eth_igb_default_mac_addr_set,
371 	.reta_update          = eth_igb_rss_reta_update,
372 	.reta_query           = eth_igb_rss_reta_query,
373 	.rss_hash_update      = eth_igb_rss_hash_update,
374 	.rss_hash_conf_get    = eth_igb_rss_hash_conf_get,
375 	.flow_ops_get         = eth_igb_flow_ops_get,
376 	.set_mc_addr_list     = eth_igb_set_mc_addr_list,
377 	.rxq_info_get         = igb_rxq_info_get,
378 	.txq_info_get         = igb_txq_info_get,
379 	.timesync_enable      = igb_timesync_enable,
380 	.timesync_disable     = igb_timesync_disable,
381 	.timesync_read_rx_timestamp = igb_timesync_read_rx_timestamp,
382 	.timesync_read_tx_timestamp = igb_timesync_read_tx_timestamp,
383 	.get_reg              = eth_igb_get_regs,
384 	.get_eeprom_length    = eth_igb_get_eeprom_length,
385 	.get_eeprom           = eth_igb_get_eeprom,
386 	.set_eeprom           = eth_igb_set_eeprom,
387 	.get_module_info      = eth_igb_get_module_info,
388 	.get_module_eeprom    = eth_igb_get_module_eeprom,
389 	.timesync_adjust_time = igb_timesync_adjust_time,
390 	.timesync_read_time   = igb_timesync_read_time,
391 	.timesync_write_time  = igb_timesync_write_time,
392 };
393 
394 /*
395  * dev_ops for virtual function, bare necessities for basic vf
396  * operation have been implemented
397  */
398 static const struct eth_dev_ops igbvf_eth_dev_ops = {
399 	.dev_configure        = igbvf_dev_configure,
400 	.dev_start            = igbvf_dev_start,
401 	.dev_stop             = igbvf_dev_stop,
402 	.dev_close            = igbvf_dev_close,
403 	.promiscuous_enable   = igbvf_promiscuous_enable,
404 	.promiscuous_disable  = igbvf_promiscuous_disable,
405 	.allmulticast_enable  = igbvf_allmulticast_enable,
406 	.allmulticast_disable = igbvf_allmulticast_disable,
407 	.link_update          = eth_igb_link_update,
408 	.stats_get            = eth_igbvf_stats_get,
409 	.xstats_get           = eth_igbvf_xstats_get,
410 	.xstats_get_names     = eth_igbvf_xstats_get_names,
411 	.stats_reset          = eth_igbvf_stats_reset,
412 	.xstats_reset         = eth_igbvf_stats_reset,
413 	.vlan_filter_set      = igbvf_vlan_filter_set,
414 	.dev_infos_get        = eth_igbvf_infos_get,
415 	.dev_supported_ptypes_get = eth_igb_supported_ptypes_get,
416 	.rx_queue_setup       = eth_igb_rx_queue_setup,
417 	.rx_queue_release     = eth_igb_rx_queue_release,
418 	.tx_queue_setup       = eth_igb_tx_queue_setup,
419 	.tx_queue_release     = eth_igb_tx_queue_release,
420 	.tx_done_cleanup      = eth_igb_tx_done_cleanup,
421 	.set_mc_addr_list     = eth_igb_set_mc_addr_list,
422 	.rxq_info_get         = igb_rxq_info_get,
423 	.txq_info_get         = igb_txq_info_get,
424 	.mac_addr_set         = igbvf_default_mac_addr_set,
425 	.get_reg              = igbvf_get_regs,
426 };
427 
428 /* store statistics names and its offset in stats structure */
429 struct rte_igb_xstats_name_off {
430 	char name[RTE_ETH_XSTATS_NAME_SIZE];
431 	unsigned offset;
432 };
433 
434 static const struct rte_igb_xstats_name_off rte_igb_stats_strings[] = {
435 	{"rx_crc_errors", offsetof(struct e1000_hw_stats, crcerrs)},
436 	{"rx_align_errors", offsetof(struct e1000_hw_stats, algnerrc)},
437 	{"rx_symbol_errors", offsetof(struct e1000_hw_stats, symerrs)},
438 	{"rx_missed_packets", offsetof(struct e1000_hw_stats, mpc)},
439 	{"tx_single_collision_packets", offsetof(struct e1000_hw_stats, scc)},
440 	{"tx_multiple_collision_packets", offsetof(struct e1000_hw_stats, mcc)},
441 	{"tx_excessive_collision_packets", offsetof(struct e1000_hw_stats,
442 		ecol)},
443 	{"tx_late_collisions", offsetof(struct e1000_hw_stats, latecol)},
444 	{"tx_total_collisions", offsetof(struct e1000_hw_stats, colc)},
445 	{"tx_deferred_packets", offsetof(struct e1000_hw_stats, dc)},
446 	{"tx_no_carrier_sense_packets", offsetof(struct e1000_hw_stats, tncrs)},
447 	{"rx_carrier_ext_errors", offsetof(struct e1000_hw_stats, cexterr)},
448 	{"rx_length_errors", offsetof(struct e1000_hw_stats, rlec)},
449 	{"rx_xon_packets", offsetof(struct e1000_hw_stats, xonrxc)},
450 	{"tx_xon_packets", offsetof(struct e1000_hw_stats, xontxc)},
451 	{"rx_xoff_packets", offsetof(struct e1000_hw_stats, xoffrxc)},
452 	{"tx_xoff_packets", offsetof(struct e1000_hw_stats, xofftxc)},
453 	{"rx_flow_control_unsupported_packets", offsetof(struct e1000_hw_stats,
454 		fcruc)},
455 	{"rx_size_64_packets", offsetof(struct e1000_hw_stats, prc64)},
456 	{"rx_size_65_to_127_packets", offsetof(struct e1000_hw_stats, prc127)},
457 	{"rx_size_128_to_255_packets", offsetof(struct e1000_hw_stats, prc255)},
458 	{"rx_size_256_to_511_packets", offsetof(struct e1000_hw_stats, prc511)},
459 	{"rx_size_512_to_1023_packets", offsetof(struct e1000_hw_stats,
460 		prc1023)},
461 	{"rx_size_1024_to_max_packets", offsetof(struct e1000_hw_stats,
462 		prc1522)},
463 	{"rx_broadcast_packets", offsetof(struct e1000_hw_stats, bprc)},
464 	{"rx_multicast_packets", offsetof(struct e1000_hw_stats, mprc)},
465 	{"rx_undersize_errors", offsetof(struct e1000_hw_stats, ruc)},
466 	{"rx_fragment_errors", offsetof(struct e1000_hw_stats, rfc)},
467 	{"rx_oversize_errors", offsetof(struct e1000_hw_stats, roc)},
468 	{"rx_jabber_errors", offsetof(struct e1000_hw_stats, rjc)},
469 	{"rx_management_packets", offsetof(struct e1000_hw_stats, mgprc)},
470 	{"rx_management_dropped", offsetof(struct e1000_hw_stats, mgpdc)},
471 	{"tx_management_packets", offsetof(struct e1000_hw_stats, mgptc)},
472 	{"rx_total_packets", offsetof(struct e1000_hw_stats, tpr)},
473 	{"tx_total_packets", offsetof(struct e1000_hw_stats, tpt)},
474 	{"rx_total_bytes", offsetof(struct e1000_hw_stats, tor)},
475 	{"tx_total_bytes", offsetof(struct e1000_hw_stats, tot)},
476 	{"tx_size_64_packets", offsetof(struct e1000_hw_stats, ptc64)},
477 	{"tx_size_65_to_127_packets", offsetof(struct e1000_hw_stats, ptc127)},
478 	{"tx_size_128_to_255_packets", offsetof(struct e1000_hw_stats, ptc255)},
479 	{"tx_size_256_to_511_packets", offsetof(struct e1000_hw_stats, ptc511)},
480 	{"tx_size_512_to_1023_packets", offsetof(struct e1000_hw_stats,
481 		ptc1023)},
482 	{"tx_size_1023_to_max_packets", offsetof(struct e1000_hw_stats,
483 		ptc1522)},
484 	{"tx_multicast_packets", offsetof(struct e1000_hw_stats, mptc)},
485 	{"tx_broadcast_packets", offsetof(struct e1000_hw_stats, bptc)},
486 	{"tx_tso_packets", offsetof(struct e1000_hw_stats, tsctc)},
487 	{"tx_tso_errors", offsetof(struct e1000_hw_stats, tsctfc)},
488 	{"rx_sent_to_host_packets", offsetof(struct e1000_hw_stats, rpthc)},
489 	{"tx_sent_by_host_packets", offsetof(struct e1000_hw_stats, hgptc)},
490 	{"rx_code_violation_packets", offsetof(struct e1000_hw_stats, scvpc)},
491 
492 	{"interrupt_assert_count", offsetof(struct e1000_hw_stats, iac)},
493 };
494 
495 #define IGB_NB_XSTATS (sizeof(rte_igb_stats_strings) / \
496 		sizeof(rte_igb_stats_strings[0]))
497 
498 static const struct rte_igb_xstats_name_off rte_igbvf_stats_strings[] = {
499 	{"rx_multicast_packets", offsetof(struct e1000_vf_stats, mprc)},
500 	{"rx_good_loopback_packets", offsetof(struct e1000_vf_stats, gprlbc)},
501 	{"tx_good_loopback_packets", offsetof(struct e1000_vf_stats, gptlbc)},
502 	{"rx_good_loopback_bytes", offsetof(struct e1000_vf_stats, gorlbc)},
503 	{"tx_good_loopback_bytes", offsetof(struct e1000_vf_stats, gotlbc)},
504 };
505 
506 #define IGBVF_NB_XSTATS (sizeof(rte_igbvf_stats_strings) / \
507 		sizeof(rte_igbvf_stats_strings[0]))
508 
509 
510 static inline void
511 igb_intr_enable(struct rte_eth_dev *dev)
512 {
513 	struct e1000_interrupt *intr =
514 		E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
515 	struct e1000_hw *hw =
516 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
517 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
518 	struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
519 
520 	if (rte_intr_allow_others(intr_handle) &&
521 		dev->data->dev_conf.intr_conf.lsc != 0) {
522 		E1000_WRITE_REG(hw, E1000_EIMS, 1 << IGB_MSIX_OTHER_INTR_VEC);
523 	}
524 
525 	E1000_WRITE_REG(hw, E1000_IMS, intr->mask);
526 	E1000_WRITE_FLUSH(hw);
527 }
528 
529 static void
530 igb_intr_disable(struct rte_eth_dev *dev)
531 {
532 	struct e1000_hw *hw =
533 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
534 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
535 	struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
536 
537 	if (rte_intr_allow_others(intr_handle) &&
538 		dev->data->dev_conf.intr_conf.lsc != 0) {
539 		E1000_WRITE_REG(hw, E1000_EIMC, 1 << IGB_MSIX_OTHER_INTR_VEC);
540 	}
541 
542 	E1000_WRITE_REG(hw, E1000_IMC, ~0);
543 	E1000_WRITE_FLUSH(hw);
544 }
545 
546 static inline void
547 igbvf_intr_enable(struct rte_eth_dev *dev)
548 {
549 	struct e1000_hw *hw =
550 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
551 
552 	/* only for mailbox */
553 	E1000_WRITE_REG(hw, E1000_EIAM, 1 << E1000_VTIVAR_MISC_MAILBOX);
554 	E1000_WRITE_REG(hw, E1000_EIAC, 1 << E1000_VTIVAR_MISC_MAILBOX);
555 	E1000_WRITE_REG(hw, E1000_EIMS, 1 << E1000_VTIVAR_MISC_MAILBOX);
556 	E1000_WRITE_FLUSH(hw);
557 }
558 
559 /* only for mailbox now. If RX/TX needed, should extend this function.  */
560 static void
561 igbvf_set_ivar_map(struct e1000_hw *hw, uint8_t msix_vector)
562 {
563 	uint32_t tmp = 0;
564 
565 	/* mailbox */
566 	tmp |= (msix_vector & E1000_VTIVAR_MISC_INTR_MASK);
567 	tmp |= E1000_VTIVAR_VALID;
568 	E1000_WRITE_REG(hw, E1000_VTIVAR_MISC, tmp);
569 }
570 
571 static void
572 eth_igbvf_configure_msix_intr(struct rte_eth_dev *dev)
573 {
574 	struct e1000_hw *hw =
575 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
576 
577 	/* Configure VF other cause ivar */
578 	igbvf_set_ivar_map(hw, E1000_VTIVAR_MISC_MAILBOX);
579 }
580 
581 static inline int32_t
582 igb_pf_reset_hw(struct e1000_hw *hw)
583 {
584 	uint32_t ctrl_ext;
585 	int32_t status;
586 
587 	status = e1000_reset_hw(hw);
588 
589 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
590 	/* Set PF Reset Done bit so PF/VF Mail Ops can work */
591 	ctrl_ext |= E1000_CTRL_EXT_PFRSTD;
592 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
593 	E1000_WRITE_FLUSH(hw);
594 
595 	return status;
596 }
597 
598 static void
599 igb_identify_hardware(struct rte_eth_dev *dev, struct rte_pci_device *pci_dev)
600 {
601 	struct e1000_hw *hw =
602 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
603 
604 
605 	hw->vendor_id = pci_dev->id.vendor_id;
606 	hw->device_id = pci_dev->id.device_id;
607 	hw->subsystem_vendor_id = pci_dev->id.subsystem_vendor_id;
608 	hw->subsystem_device_id = pci_dev->id.subsystem_device_id;
609 
610 	e1000_set_mac_type(hw);
611 
612 	/* need to check if it is a vf device below */
613 }
614 
615 static int
616 igb_reset_swfw_lock(struct e1000_hw *hw)
617 {
618 	int ret_val;
619 
620 	/*
621 	 * Do mac ops initialization manually here, since we will need
622 	 * some function pointers set by this call.
623 	 */
624 	ret_val = e1000_init_mac_params(hw);
625 	if (ret_val)
626 		return ret_val;
627 
628 	/*
629 	 * SMBI lock should not fail in this early stage. If this is the case,
630 	 * it is due to an improper exit of the application.
631 	 * So force the release of the faulty lock.
632 	 */
633 	if (e1000_get_hw_semaphore_generic(hw) < 0) {
634 		PMD_DRV_LOG(DEBUG, "SMBI lock released");
635 	}
636 	e1000_put_hw_semaphore_generic(hw);
637 
638 	if (hw->mac.ops.acquire_swfw_sync != NULL) {
639 		uint16_t mask;
640 
641 		/*
642 		 * Phy lock should not fail in this early stage. If this is the case,
643 		 * it is due to an improper exit of the application.
644 		 * So force the release of the faulty lock.
645 		 */
646 		mask = E1000_SWFW_PHY0_SM << hw->bus.func;
647 		if (hw->bus.func > E1000_FUNC_1)
648 			mask <<= 2;
649 		if (hw->mac.ops.acquire_swfw_sync(hw, mask) < 0) {
650 			PMD_DRV_LOG(DEBUG, "SWFW phy%d lock released",
651 				    hw->bus.func);
652 		}
653 		hw->mac.ops.release_swfw_sync(hw, mask);
654 
655 		/*
656 		 * This one is more tricky since it is common to all ports; but
657 		 * swfw_sync retries last long enough (1s) to be almost sure that if
658 		 * lock can not be taken it is due to an improper lock of the
659 		 * semaphore.
660 		 */
661 		mask = E1000_SWFW_EEP_SM;
662 		if (hw->mac.ops.acquire_swfw_sync(hw, mask) < 0) {
663 			PMD_DRV_LOG(DEBUG, "SWFW common locks released");
664 		}
665 		hw->mac.ops.release_swfw_sync(hw, mask);
666 	}
667 
668 	return E1000_SUCCESS;
669 }
670 
671 /* Remove all ntuple filters of the device */
672 static int igb_ntuple_filter_uninit(struct rte_eth_dev *eth_dev)
673 {
674 	struct e1000_filter_info *filter_info =
675 		E1000_DEV_PRIVATE_TO_FILTER_INFO(eth_dev->data->dev_private);
676 	struct e1000_5tuple_filter *p_5tuple;
677 	struct e1000_2tuple_filter *p_2tuple;
678 
679 	while ((p_5tuple = TAILQ_FIRST(&filter_info->fivetuple_list))) {
680 		TAILQ_REMOVE(&filter_info->fivetuple_list,
681 			p_5tuple, entries);
682 			rte_free(p_5tuple);
683 	}
684 	filter_info->fivetuple_mask = 0;
685 	while ((p_2tuple = TAILQ_FIRST(&filter_info->twotuple_list))) {
686 		TAILQ_REMOVE(&filter_info->twotuple_list,
687 			p_2tuple, entries);
688 			rte_free(p_2tuple);
689 	}
690 	filter_info->twotuple_mask = 0;
691 
692 	return 0;
693 }
694 
695 /* Remove all flex filters of the device */
696 static int igb_flex_filter_uninit(struct rte_eth_dev *eth_dev)
697 {
698 	struct e1000_filter_info *filter_info =
699 		E1000_DEV_PRIVATE_TO_FILTER_INFO(eth_dev->data->dev_private);
700 	struct e1000_flex_filter *p_flex;
701 
702 	while ((p_flex = TAILQ_FIRST(&filter_info->flex_list))) {
703 		TAILQ_REMOVE(&filter_info->flex_list, p_flex, entries);
704 		rte_free(p_flex);
705 	}
706 	filter_info->flex_mask = 0;
707 
708 	return 0;
709 }
710 
711 static int
712 eth_igb_dev_init(struct rte_eth_dev *eth_dev)
713 {
714 	int error = 0;
715 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
716 	struct e1000_hw *hw =
717 		E1000_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
718 	struct e1000_vfta * shadow_vfta =
719 		E1000_DEV_PRIVATE_TO_VFTA(eth_dev->data->dev_private);
720 	struct e1000_filter_info *filter_info =
721 		E1000_DEV_PRIVATE_TO_FILTER_INFO(eth_dev->data->dev_private);
722 	struct e1000_adapter *adapter =
723 		E1000_DEV_PRIVATE(eth_dev->data->dev_private);
724 
725 	uint32_t ctrl_ext;
726 
727 	eth_dev->dev_ops = &eth_igb_ops;
728 	eth_dev->rx_queue_count = eth_igb_rx_queue_count;
729 	eth_dev->rx_descriptor_done   = eth_igb_rx_descriptor_done;
730 	eth_dev->rx_descriptor_status = eth_igb_rx_descriptor_status;
731 	eth_dev->tx_descriptor_status = eth_igb_tx_descriptor_status;
732 	eth_dev->rx_pkt_burst = &eth_igb_recv_pkts;
733 	eth_dev->tx_pkt_burst = &eth_igb_xmit_pkts;
734 	eth_dev->tx_pkt_prepare = &eth_igb_prep_pkts;
735 
736 	/* for secondary processes, we don't initialise any further as primary
737 	 * has already done this work. Only check we don't need a different
738 	 * RX function */
739 	if (rte_eal_process_type() != RTE_PROC_PRIMARY){
740 		if (eth_dev->data->scattered_rx)
741 			eth_dev->rx_pkt_burst = &eth_igb_recv_scattered_pkts;
742 		return 0;
743 	}
744 
745 	rte_eth_copy_pci_info(eth_dev, pci_dev);
746 	eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
747 
748 	hw->hw_addr= (void *)pci_dev->mem_resource[0].addr;
749 
750 	igb_identify_hardware(eth_dev, pci_dev);
751 	if (e1000_setup_init_funcs(hw, FALSE) != E1000_SUCCESS) {
752 		error = -EIO;
753 		goto err_late;
754 	}
755 
756 	e1000_get_bus_info(hw);
757 
758 	/* Reset any pending lock */
759 	if (igb_reset_swfw_lock(hw) != E1000_SUCCESS) {
760 		error = -EIO;
761 		goto err_late;
762 	}
763 
764 	/* Finish initialization */
765 	if (e1000_setup_init_funcs(hw, TRUE) != E1000_SUCCESS) {
766 		error = -EIO;
767 		goto err_late;
768 	}
769 
770 	hw->mac.autoneg = 1;
771 	hw->phy.autoneg_wait_to_complete = 0;
772 	hw->phy.autoneg_advertised = E1000_ALL_SPEED_DUPLEX;
773 
774 	/* Copper options */
775 	if (hw->phy.media_type == e1000_media_type_copper) {
776 		hw->phy.mdix = 0; /* AUTO_ALL_MODES */
777 		hw->phy.disable_polarity_correction = 0;
778 		hw->phy.ms_type = e1000_ms_hw_default;
779 	}
780 
781 	/*
782 	 * Start from a known state, this is important in reading the nvm
783 	 * and mac from that.
784 	 */
785 	igb_pf_reset_hw(hw);
786 
787 	/* Make sure we have a good EEPROM before we read from it */
788 	if (e1000_validate_nvm_checksum(hw) < 0) {
789 		/*
790 		 * Some PCI-E parts fail the first check due to
791 		 * the link being in sleep state, call it again,
792 		 * if it fails a second time its a real issue.
793 		 */
794 		if (e1000_validate_nvm_checksum(hw) < 0) {
795 			PMD_INIT_LOG(ERR, "EEPROM checksum invalid");
796 			error = -EIO;
797 			goto err_late;
798 		}
799 	}
800 
801 	/* Read the permanent MAC address out of the EEPROM */
802 	if (e1000_read_mac_addr(hw) != 0) {
803 		PMD_INIT_LOG(ERR, "EEPROM error while reading MAC address");
804 		error = -EIO;
805 		goto err_late;
806 	}
807 
808 	/* Allocate memory for storing MAC addresses */
809 	eth_dev->data->mac_addrs = rte_zmalloc("e1000",
810 		RTE_ETHER_ADDR_LEN * hw->mac.rar_entry_count, 0);
811 	if (eth_dev->data->mac_addrs == NULL) {
812 		PMD_INIT_LOG(ERR, "Failed to allocate %d bytes needed to "
813 						"store MAC addresses",
814 				RTE_ETHER_ADDR_LEN * hw->mac.rar_entry_count);
815 		error = -ENOMEM;
816 		goto err_late;
817 	}
818 
819 	/* Copy the permanent MAC address */
820 	rte_ether_addr_copy((struct rte_ether_addr *)hw->mac.addr,
821 			&eth_dev->data->mac_addrs[0]);
822 
823 	/* initialize the vfta */
824 	memset(shadow_vfta, 0, sizeof(*shadow_vfta));
825 
826 	/* Now initialize the hardware */
827 	if (igb_hardware_init(hw) != 0) {
828 		PMD_INIT_LOG(ERR, "Hardware initialization failed");
829 		rte_free(eth_dev->data->mac_addrs);
830 		eth_dev->data->mac_addrs = NULL;
831 		error = -ENODEV;
832 		goto err_late;
833 	}
834 	hw->mac.get_link_status = 1;
835 	adapter->stopped = 0;
836 
837 	/* Indicate SOL/IDER usage */
838 	if (e1000_check_reset_block(hw) < 0) {
839 		PMD_INIT_LOG(ERR, "PHY reset is blocked due to"
840 					"SOL/IDER session");
841 	}
842 
843 	/* initialize PF if max_vfs not zero */
844 	igb_pf_host_init(eth_dev);
845 
846 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
847 	/* Set PF Reset Done bit so PF/VF Mail Ops can work */
848 	ctrl_ext |= E1000_CTRL_EXT_PFRSTD;
849 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
850 	E1000_WRITE_FLUSH(hw);
851 
852 	PMD_INIT_LOG(DEBUG, "port_id %d vendorID=0x%x deviceID=0x%x",
853 		     eth_dev->data->port_id, pci_dev->id.vendor_id,
854 		     pci_dev->id.device_id);
855 
856 	rte_intr_callback_register(&pci_dev->intr_handle,
857 				   eth_igb_interrupt_handler,
858 				   (void *)eth_dev);
859 
860 	/* enable uio/vfio intr/eventfd mapping */
861 	rte_intr_enable(&pci_dev->intr_handle);
862 
863 	/* enable support intr */
864 	igb_intr_enable(eth_dev);
865 
866 	eth_igb_dev_set_link_down(eth_dev);
867 
868 	/* initialize filter info */
869 	memset(filter_info, 0,
870 	       sizeof(struct e1000_filter_info));
871 
872 	TAILQ_INIT(&filter_info->flex_list);
873 	TAILQ_INIT(&filter_info->twotuple_list);
874 	TAILQ_INIT(&filter_info->fivetuple_list);
875 
876 	TAILQ_INIT(&igb_filter_ntuple_list);
877 	TAILQ_INIT(&igb_filter_ethertype_list);
878 	TAILQ_INIT(&igb_filter_syn_list);
879 	TAILQ_INIT(&igb_filter_flex_list);
880 	TAILQ_INIT(&igb_filter_rss_list);
881 	TAILQ_INIT(&igb_flow_list);
882 
883 	return 0;
884 
885 err_late:
886 	igb_hw_control_release(hw);
887 
888 	return error;
889 }
890 
891 static int
892 eth_igb_dev_uninit(struct rte_eth_dev *eth_dev)
893 {
894 	PMD_INIT_FUNC_TRACE();
895 
896 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
897 		return 0;
898 
899 	eth_igb_close(eth_dev);
900 
901 	return 0;
902 }
903 
904 /*
905  * Virtual Function device init
906  */
907 static int
908 eth_igbvf_dev_init(struct rte_eth_dev *eth_dev)
909 {
910 	struct rte_pci_device *pci_dev;
911 	struct rte_intr_handle *intr_handle;
912 	struct e1000_adapter *adapter =
913 		E1000_DEV_PRIVATE(eth_dev->data->dev_private);
914 	struct e1000_hw *hw =
915 		E1000_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
916 	int diag;
917 	struct rte_ether_addr *perm_addr =
918 		(struct rte_ether_addr *)hw->mac.perm_addr;
919 
920 	PMD_INIT_FUNC_TRACE();
921 
922 	eth_dev->dev_ops = &igbvf_eth_dev_ops;
923 	eth_dev->rx_descriptor_done   = eth_igb_rx_descriptor_done;
924 	eth_dev->rx_descriptor_status = eth_igb_rx_descriptor_status;
925 	eth_dev->tx_descriptor_status = eth_igb_tx_descriptor_status;
926 	eth_dev->rx_pkt_burst = &eth_igb_recv_pkts;
927 	eth_dev->tx_pkt_burst = &eth_igb_xmit_pkts;
928 	eth_dev->tx_pkt_prepare = &eth_igb_prep_pkts;
929 
930 	/* for secondary processes, we don't initialise any further as primary
931 	 * has already done this work. Only check we don't need a different
932 	 * RX function */
933 	if (rte_eal_process_type() != RTE_PROC_PRIMARY){
934 		if (eth_dev->data->scattered_rx)
935 			eth_dev->rx_pkt_burst = &eth_igb_recv_scattered_pkts;
936 		return 0;
937 	}
938 
939 	pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
940 	rte_eth_copy_pci_info(eth_dev, pci_dev);
941 	eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
942 
943 	hw->device_id = pci_dev->id.device_id;
944 	hw->vendor_id = pci_dev->id.vendor_id;
945 	hw->hw_addr = (void *)pci_dev->mem_resource[0].addr;
946 	adapter->stopped = 0;
947 
948 	/* Initialize the shared code (base driver) */
949 	diag = e1000_setup_init_funcs(hw, TRUE);
950 	if (diag != 0) {
951 		PMD_INIT_LOG(ERR, "Shared code init failed for igbvf: %d",
952 			diag);
953 		return -EIO;
954 	}
955 
956 	/* init_mailbox_params */
957 	hw->mbx.ops.init_params(hw);
958 
959 	/* Disable the interrupts for VF */
960 	igbvf_intr_disable(hw);
961 
962 	diag = hw->mac.ops.reset_hw(hw);
963 
964 	/* Allocate memory for storing MAC addresses */
965 	eth_dev->data->mac_addrs = rte_zmalloc("igbvf", RTE_ETHER_ADDR_LEN *
966 		hw->mac.rar_entry_count, 0);
967 	if (eth_dev->data->mac_addrs == NULL) {
968 		PMD_INIT_LOG(ERR,
969 			"Failed to allocate %d bytes needed to store MAC "
970 			"addresses",
971 			RTE_ETHER_ADDR_LEN * hw->mac.rar_entry_count);
972 		return -ENOMEM;
973 	}
974 
975 	/* Generate a random MAC address, if none was assigned by PF. */
976 	if (rte_is_zero_ether_addr(perm_addr)) {
977 		rte_eth_random_addr(perm_addr->addr_bytes);
978 		PMD_INIT_LOG(INFO, "\tVF MAC address not assigned by Host PF");
979 		PMD_INIT_LOG(INFO, "\tAssign randomly generated MAC address "
980 			     "%02x:%02x:%02x:%02x:%02x:%02x",
981 			     perm_addr->addr_bytes[0],
982 			     perm_addr->addr_bytes[1],
983 			     perm_addr->addr_bytes[2],
984 			     perm_addr->addr_bytes[3],
985 			     perm_addr->addr_bytes[4],
986 			     perm_addr->addr_bytes[5]);
987 	}
988 
989 	diag = e1000_rar_set(hw, perm_addr->addr_bytes, 0);
990 	if (diag) {
991 		rte_free(eth_dev->data->mac_addrs);
992 		eth_dev->data->mac_addrs = NULL;
993 		return diag;
994 	}
995 	/* Copy the permanent MAC address */
996 	rte_ether_addr_copy((struct rte_ether_addr *)hw->mac.perm_addr,
997 			&eth_dev->data->mac_addrs[0]);
998 
999 	PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x "
1000 		     "mac.type=%s",
1001 		     eth_dev->data->port_id, pci_dev->id.vendor_id,
1002 		     pci_dev->id.device_id, "igb_mac_82576_vf");
1003 
1004 	intr_handle = &pci_dev->intr_handle;
1005 	rte_intr_callback_register(intr_handle,
1006 				   eth_igbvf_interrupt_handler, eth_dev);
1007 
1008 	return 0;
1009 }
1010 
1011 static int
1012 eth_igbvf_dev_uninit(struct rte_eth_dev *eth_dev)
1013 {
1014 	PMD_INIT_FUNC_TRACE();
1015 
1016 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1017 		return 0;
1018 
1019 	igbvf_dev_close(eth_dev);
1020 
1021 	return 0;
1022 }
1023 
1024 static int eth_igb_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1025 	struct rte_pci_device *pci_dev)
1026 {
1027 	return rte_eth_dev_pci_generic_probe(pci_dev,
1028 		sizeof(struct e1000_adapter), eth_igb_dev_init);
1029 }
1030 
1031 static int eth_igb_pci_remove(struct rte_pci_device *pci_dev)
1032 {
1033 	return rte_eth_dev_pci_generic_remove(pci_dev, eth_igb_dev_uninit);
1034 }
1035 
1036 static struct rte_pci_driver rte_igb_pmd = {
1037 	.id_table = pci_id_igb_map,
1038 	.drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC,
1039 	.probe = eth_igb_pci_probe,
1040 	.remove = eth_igb_pci_remove,
1041 };
1042 
1043 
1044 static int eth_igbvf_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1045 	struct rte_pci_device *pci_dev)
1046 {
1047 	return rte_eth_dev_pci_generic_probe(pci_dev,
1048 		sizeof(struct e1000_adapter), eth_igbvf_dev_init);
1049 }
1050 
1051 static int eth_igbvf_pci_remove(struct rte_pci_device *pci_dev)
1052 {
1053 	return rte_eth_dev_pci_generic_remove(pci_dev, eth_igbvf_dev_uninit);
1054 }
1055 
1056 /*
1057  * virtual function driver struct
1058  */
1059 static struct rte_pci_driver rte_igbvf_pmd = {
1060 	.id_table = pci_id_igbvf_map,
1061 	.drv_flags = RTE_PCI_DRV_NEED_MAPPING,
1062 	.probe = eth_igbvf_pci_probe,
1063 	.remove = eth_igbvf_pci_remove,
1064 };
1065 
1066 static void
1067 igb_vmdq_vlan_hw_filter_enable(struct rte_eth_dev *dev)
1068 {
1069 	struct e1000_hw *hw =
1070 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1071 	/* RCTL: enable VLAN filter since VMDq always use VLAN filter */
1072 	uint32_t rctl = E1000_READ_REG(hw, E1000_RCTL);
1073 	rctl |= E1000_RCTL_VFE;
1074 	E1000_WRITE_REG(hw, E1000_RCTL, rctl);
1075 }
1076 
1077 static int
1078 igb_check_mq_mode(struct rte_eth_dev *dev)
1079 {
1080 	enum rte_eth_rx_mq_mode rx_mq_mode = dev->data->dev_conf.rxmode.mq_mode;
1081 	enum rte_eth_tx_mq_mode tx_mq_mode = dev->data->dev_conf.txmode.mq_mode;
1082 	uint16_t nb_rx_q = dev->data->nb_rx_queues;
1083 	uint16_t nb_tx_q = dev->data->nb_tx_queues;
1084 
1085 	if ((rx_mq_mode & ETH_MQ_RX_DCB_FLAG) ||
1086 	    tx_mq_mode == ETH_MQ_TX_DCB ||
1087 	    tx_mq_mode == ETH_MQ_TX_VMDQ_DCB) {
1088 		PMD_INIT_LOG(ERR, "DCB mode is not supported.");
1089 		return -EINVAL;
1090 	}
1091 	if (RTE_ETH_DEV_SRIOV(dev).active != 0) {
1092 		/* Check multi-queue mode.
1093 		 * To no break software we accept ETH_MQ_RX_NONE as this might
1094 		 * be used to turn off VLAN filter.
1095 		 */
1096 
1097 		if (rx_mq_mode == ETH_MQ_RX_NONE ||
1098 		    rx_mq_mode == ETH_MQ_RX_VMDQ_ONLY) {
1099 			dev->data->dev_conf.rxmode.mq_mode = ETH_MQ_RX_VMDQ_ONLY;
1100 			RTE_ETH_DEV_SRIOV(dev).nb_q_per_pool = 1;
1101 		} else {
1102 			/* Only support one queue on VFs.
1103 			 * RSS together with SRIOV is not supported.
1104 			 */
1105 			PMD_INIT_LOG(ERR, "SRIOV is active,"
1106 					" wrong mq_mode rx %d.",
1107 					rx_mq_mode);
1108 			return -EINVAL;
1109 		}
1110 		/* TX mode is not used here, so mode might be ignored.*/
1111 		if (tx_mq_mode != ETH_MQ_TX_VMDQ_ONLY) {
1112 			/* SRIOV only works in VMDq enable mode */
1113 			PMD_INIT_LOG(WARNING, "SRIOV is active,"
1114 					" TX mode %d is not supported. "
1115 					" Driver will behave as %d mode.",
1116 					tx_mq_mode, ETH_MQ_TX_VMDQ_ONLY);
1117 		}
1118 
1119 		/* check valid queue number */
1120 		if ((nb_rx_q > 1) || (nb_tx_q > 1)) {
1121 			PMD_INIT_LOG(ERR, "SRIOV is active,"
1122 					" only support one queue on VFs.");
1123 			return -EINVAL;
1124 		}
1125 	} else {
1126 		/* To no break software that set invalid mode, only display
1127 		 * warning if invalid mode is used.
1128 		 */
1129 		if (rx_mq_mode != ETH_MQ_RX_NONE &&
1130 		    rx_mq_mode != ETH_MQ_RX_VMDQ_ONLY &&
1131 		    rx_mq_mode != ETH_MQ_RX_RSS) {
1132 			/* RSS together with VMDq not supported*/
1133 			PMD_INIT_LOG(ERR, "RX mode %d is not supported.",
1134 				     rx_mq_mode);
1135 			return -EINVAL;
1136 		}
1137 
1138 		if (tx_mq_mode != ETH_MQ_TX_NONE &&
1139 		    tx_mq_mode != ETH_MQ_TX_VMDQ_ONLY) {
1140 			PMD_INIT_LOG(WARNING, "TX mode %d is not supported."
1141 					" Due to txmode is meaningless in this"
1142 					" driver, just ignore.",
1143 					tx_mq_mode);
1144 		}
1145 	}
1146 	return 0;
1147 }
1148 
1149 static int
1150 eth_igb_configure(struct rte_eth_dev *dev)
1151 {
1152 	struct e1000_interrupt *intr =
1153 		E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
1154 	int ret;
1155 
1156 	PMD_INIT_FUNC_TRACE();
1157 
1158 	if (dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG)
1159 		dev->data->dev_conf.rxmode.offloads |= DEV_RX_OFFLOAD_RSS_HASH;
1160 
1161 	/* multipe queue mode checking */
1162 	ret  = igb_check_mq_mode(dev);
1163 	if (ret != 0) {
1164 		PMD_DRV_LOG(ERR, "igb_check_mq_mode fails with %d.",
1165 			    ret);
1166 		return ret;
1167 	}
1168 
1169 	intr->flags |= E1000_FLAG_NEED_LINK_UPDATE;
1170 	PMD_INIT_FUNC_TRACE();
1171 
1172 	return 0;
1173 }
1174 
1175 static void
1176 eth_igb_rxtx_control(struct rte_eth_dev *dev,
1177 		     bool enable)
1178 {
1179 	struct e1000_hw *hw =
1180 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1181 	uint32_t tctl, rctl;
1182 
1183 	tctl = E1000_READ_REG(hw, E1000_TCTL);
1184 	rctl = E1000_READ_REG(hw, E1000_RCTL);
1185 
1186 	if (enable) {
1187 		/* enable Tx/Rx */
1188 		tctl |= E1000_TCTL_EN;
1189 		rctl |= E1000_RCTL_EN;
1190 	} else {
1191 		/* disable Tx/Rx */
1192 		tctl &= ~E1000_TCTL_EN;
1193 		rctl &= ~E1000_RCTL_EN;
1194 	}
1195 	E1000_WRITE_REG(hw, E1000_TCTL, tctl);
1196 	E1000_WRITE_REG(hw, E1000_RCTL, rctl);
1197 	E1000_WRITE_FLUSH(hw);
1198 }
1199 
1200 static int
1201 eth_igb_start(struct rte_eth_dev *dev)
1202 {
1203 	struct e1000_hw *hw =
1204 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1205 	struct e1000_adapter *adapter =
1206 		E1000_DEV_PRIVATE(dev->data->dev_private);
1207 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1208 	struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
1209 	int ret, mask;
1210 	uint32_t intr_vector = 0;
1211 	uint32_t ctrl_ext;
1212 	uint32_t *speeds;
1213 	int num_speeds;
1214 	bool autoneg;
1215 
1216 	PMD_INIT_FUNC_TRACE();
1217 
1218 	/* disable uio/vfio intr/eventfd mapping */
1219 	rte_intr_disable(intr_handle);
1220 
1221 	/* Power up the phy. Needed to make the link go Up */
1222 	eth_igb_dev_set_link_up(dev);
1223 
1224 	/*
1225 	 * Packet Buffer Allocation (PBA)
1226 	 * Writing PBA sets the receive portion of the buffer
1227 	 * the remainder is used for the transmit buffer.
1228 	 */
1229 	if (hw->mac.type == e1000_82575) {
1230 		uint32_t pba;
1231 
1232 		pba = E1000_PBA_32K; /* 32K for Rx, 16K for Tx */
1233 		E1000_WRITE_REG(hw, E1000_PBA, pba);
1234 	}
1235 
1236 	/* Put the address into the Receive Address Array */
1237 	e1000_rar_set(hw, hw->mac.addr, 0);
1238 
1239 	/* Initialize the hardware */
1240 	if (igb_hardware_init(hw)) {
1241 		PMD_INIT_LOG(ERR, "Unable to initialize the hardware");
1242 		return -EIO;
1243 	}
1244 	adapter->stopped = 0;
1245 
1246 	E1000_WRITE_REG(hw, E1000_VET,
1247 			RTE_ETHER_TYPE_VLAN << 16 | RTE_ETHER_TYPE_VLAN);
1248 
1249 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
1250 	/* Set PF Reset Done bit so PF/VF Mail Ops can work */
1251 	ctrl_ext |= E1000_CTRL_EXT_PFRSTD;
1252 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
1253 	E1000_WRITE_FLUSH(hw);
1254 
1255 	/* configure PF module if SRIOV enabled */
1256 	igb_pf_host_configure(dev);
1257 
1258 	/* check and configure queue intr-vector mapping */
1259 	if ((rte_intr_cap_multiple(intr_handle) ||
1260 	     !RTE_ETH_DEV_SRIOV(dev).active) &&
1261 	    dev->data->dev_conf.intr_conf.rxq != 0) {
1262 		intr_vector = dev->data->nb_rx_queues;
1263 		if (rte_intr_efd_enable(intr_handle, intr_vector))
1264 			return -1;
1265 	}
1266 
1267 	if (rte_intr_dp_is_en(intr_handle) && !intr_handle->intr_vec) {
1268 		intr_handle->intr_vec =
1269 			rte_zmalloc("intr_vec",
1270 				    dev->data->nb_rx_queues * sizeof(int), 0);
1271 		if (intr_handle->intr_vec == NULL) {
1272 			PMD_INIT_LOG(ERR, "Failed to allocate %d rx_queues"
1273 				     " intr_vec", dev->data->nb_rx_queues);
1274 			return -ENOMEM;
1275 		}
1276 	}
1277 
1278 	/* confiugre msix for rx interrupt */
1279 	eth_igb_configure_msix_intr(dev);
1280 
1281 	/* Configure for OS presence */
1282 	igb_init_manageability(hw);
1283 
1284 	eth_igb_tx_init(dev);
1285 
1286 	/* This can fail when allocating mbufs for descriptor rings */
1287 	ret = eth_igb_rx_init(dev);
1288 	if (ret) {
1289 		PMD_INIT_LOG(ERR, "Unable to initialize RX hardware");
1290 		igb_dev_clear_queues(dev);
1291 		return ret;
1292 	}
1293 
1294 	e1000_clear_hw_cntrs_base_generic(hw);
1295 
1296 	/*
1297 	 * VLAN Offload Settings
1298 	 */
1299 	mask = ETH_VLAN_STRIP_MASK | ETH_VLAN_FILTER_MASK | \
1300 			ETH_VLAN_EXTEND_MASK;
1301 	ret = eth_igb_vlan_offload_set(dev, mask);
1302 	if (ret) {
1303 		PMD_INIT_LOG(ERR, "Unable to set vlan offload");
1304 		igb_dev_clear_queues(dev);
1305 		return ret;
1306 	}
1307 
1308 	if (dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_VMDQ_ONLY) {
1309 		/* Enable VLAN filter since VMDq always use VLAN filter */
1310 		igb_vmdq_vlan_hw_filter_enable(dev);
1311 	}
1312 
1313 	if ((hw->mac.type == e1000_82576) || (hw->mac.type == e1000_82580) ||
1314 		(hw->mac.type == e1000_i350) || (hw->mac.type == e1000_i210) ||
1315 		(hw->mac.type == e1000_i211)) {
1316 		/* Configure EITR with the maximum possible value (0xFFFF) */
1317 		E1000_WRITE_REG(hw, E1000_EITR(0), 0xFFFF);
1318 	}
1319 
1320 	/* Setup link speed and duplex */
1321 	speeds = &dev->data->dev_conf.link_speeds;
1322 	if (*speeds == ETH_LINK_SPEED_AUTONEG) {
1323 		hw->phy.autoneg_advertised = E1000_ALL_SPEED_DUPLEX;
1324 		hw->mac.autoneg = 1;
1325 	} else {
1326 		num_speeds = 0;
1327 		autoneg = (*speeds & ETH_LINK_SPEED_FIXED) == 0;
1328 
1329 		/* Reset */
1330 		hw->phy.autoneg_advertised = 0;
1331 
1332 		if (*speeds & ~(ETH_LINK_SPEED_10M_HD | ETH_LINK_SPEED_10M |
1333 				ETH_LINK_SPEED_100M_HD | ETH_LINK_SPEED_100M |
1334 				ETH_LINK_SPEED_1G | ETH_LINK_SPEED_FIXED)) {
1335 			num_speeds = -1;
1336 			goto error_invalid_config;
1337 		}
1338 		if (*speeds & ETH_LINK_SPEED_10M_HD) {
1339 			hw->phy.autoneg_advertised |= ADVERTISE_10_HALF;
1340 			num_speeds++;
1341 		}
1342 		if (*speeds & ETH_LINK_SPEED_10M) {
1343 			hw->phy.autoneg_advertised |= ADVERTISE_10_FULL;
1344 			num_speeds++;
1345 		}
1346 		if (*speeds & ETH_LINK_SPEED_100M_HD) {
1347 			hw->phy.autoneg_advertised |= ADVERTISE_100_HALF;
1348 			num_speeds++;
1349 		}
1350 		if (*speeds & ETH_LINK_SPEED_100M) {
1351 			hw->phy.autoneg_advertised |= ADVERTISE_100_FULL;
1352 			num_speeds++;
1353 		}
1354 		if (*speeds & ETH_LINK_SPEED_1G) {
1355 			hw->phy.autoneg_advertised |= ADVERTISE_1000_FULL;
1356 			num_speeds++;
1357 		}
1358 		if (num_speeds == 0 || (!autoneg && (num_speeds > 1)))
1359 			goto error_invalid_config;
1360 
1361 		/* Set/reset the mac.autoneg based on the link speed,
1362 		 * fixed or not
1363 		 */
1364 		if (!autoneg) {
1365 			hw->mac.autoneg = 0;
1366 			hw->mac.forced_speed_duplex =
1367 					hw->phy.autoneg_advertised;
1368 		} else {
1369 			hw->mac.autoneg = 1;
1370 		}
1371 	}
1372 
1373 	e1000_setup_link(hw);
1374 
1375 	if (rte_intr_allow_others(intr_handle)) {
1376 		/* check if lsc interrupt is enabled */
1377 		if (dev->data->dev_conf.intr_conf.lsc != 0)
1378 			eth_igb_lsc_interrupt_setup(dev, TRUE);
1379 		else
1380 			eth_igb_lsc_interrupt_setup(dev, FALSE);
1381 	} else {
1382 		rte_intr_callback_unregister(intr_handle,
1383 					     eth_igb_interrupt_handler,
1384 					     (void *)dev);
1385 		if (dev->data->dev_conf.intr_conf.lsc != 0)
1386 			PMD_INIT_LOG(INFO, "lsc won't enable because of"
1387 				     " no intr multiplex");
1388 	}
1389 
1390 	/* check if rxq interrupt is enabled */
1391 	if (dev->data->dev_conf.intr_conf.rxq != 0 &&
1392 	    rte_intr_dp_is_en(intr_handle))
1393 		eth_igb_rxq_interrupt_setup(dev);
1394 
1395 	/* enable uio/vfio intr/eventfd mapping */
1396 	rte_intr_enable(intr_handle);
1397 
1398 	/* resume enabled intr since hw reset */
1399 	igb_intr_enable(dev);
1400 
1401 	/* restore all types filter */
1402 	igb_filter_restore(dev);
1403 
1404 	eth_igb_rxtx_control(dev, true);
1405 	eth_igb_link_update(dev, 0);
1406 
1407 	PMD_INIT_LOG(DEBUG, "<<");
1408 
1409 	return 0;
1410 
1411 error_invalid_config:
1412 	PMD_INIT_LOG(ERR, "Invalid advertised speeds (%u) for port %u",
1413 		     dev->data->dev_conf.link_speeds, dev->data->port_id);
1414 	igb_dev_clear_queues(dev);
1415 	return -EINVAL;
1416 }
1417 
1418 /*********************************************************************
1419  *
1420  *  This routine disables all traffic on the adapter by issuing a
1421  *  global reset on the MAC.
1422  *
1423  **********************************************************************/
1424 static int
1425 eth_igb_stop(struct rte_eth_dev *dev)
1426 {
1427 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1428 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1429 	struct rte_eth_link link;
1430 	struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
1431 	struct e1000_adapter *adapter =
1432 		E1000_DEV_PRIVATE(dev->data->dev_private);
1433 
1434 	if (adapter->stopped)
1435 		return 0;
1436 
1437 	eth_igb_rxtx_control(dev, false);
1438 
1439 	igb_intr_disable(dev);
1440 
1441 	/* disable intr eventfd mapping */
1442 	rte_intr_disable(intr_handle);
1443 
1444 	igb_pf_reset_hw(hw);
1445 	E1000_WRITE_REG(hw, E1000_WUC, 0);
1446 
1447 	/* Set bit for Go Link disconnect if PHY reset is not blocked */
1448 	if (hw->mac.type >= e1000_82580 &&
1449 	    (e1000_check_reset_block(hw) != E1000_BLK_PHY_RESET)) {
1450 		uint32_t phpm_reg;
1451 
1452 		phpm_reg = E1000_READ_REG(hw, E1000_82580_PHY_POWER_MGMT);
1453 		phpm_reg |= E1000_82580_PM_GO_LINKD;
1454 		E1000_WRITE_REG(hw, E1000_82580_PHY_POWER_MGMT, phpm_reg);
1455 	}
1456 
1457 	/* Power down the phy. Needed to make the link go Down */
1458 	eth_igb_dev_set_link_down(dev);
1459 
1460 	igb_dev_clear_queues(dev);
1461 
1462 	/* clear the recorded link status */
1463 	memset(&link, 0, sizeof(link));
1464 	rte_eth_linkstatus_set(dev, &link);
1465 
1466 	if (!rte_intr_allow_others(intr_handle))
1467 		/* resume to the default handler */
1468 		rte_intr_callback_register(intr_handle,
1469 					   eth_igb_interrupt_handler,
1470 					   (void *)dev);
1471 
1472 	/* Clean datapath event and queue/vec mapping */
1473 	rte_intr_efd_disable(intr_handle);
1474 	if (intr_handle->intr_vec != NULL) {
1475 		rte_free(intr_handle->intr_vec);
1476 		intr_handle->intr_vec = NULL;
1477 	}
1478 
1479 	adapter->stopped = true;
1480 	dev->data->dev_started = 0;
1481 
1482 	return 0;
1483 }
1484 
1485 static int
1486 eth_igb_dev_set_link_up(struct rte_eth_dev *dev)
1487 {
1488 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1489 
1490 	if (hw->phy.media_type == e1000_media_type_copper)
1491 		e1000_power_up_phy(hw);
1492 	else
1493 		e1000_power_up_fiber_serdes_link(hw);
1494 
1495 	return 0;
1496 }
1497 
1498 static int
1499 eth_igb_dev_set_link_down(struct rte_eth_dev *dev)
1500 {
1501 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1502 
1503 	if (hw->phy.media_type == e1000_media_type_copper)
1504 		e1000_power_down_phy(hw);
1505 	else
1506 		e1000_shutdown_fiber_serdes_link(hw);
1507 
1508 	return 0;
1509 }
1510 
1511 static int
1512 eth_igb_close(struct rte_eth_dev *dev)
1513 {
1514 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1515 	struct rte_eth_link link;
1516 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1517 	struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
1518 	struct e1000_filter_info *filter_info =
1519 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
1520 	int ret;
1521 
1522 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1523 		return 0;
1524 
1525 	ret = eth_igb_stop(dev);
1526 
1527 	e1000_phy_hw_reset(hw);
1528 	igb_release_manageability(hw);
1529 	igb_hw_control_release(hw);
1530 
1531 	/* Clear bit for Go Link disconnect if PHY reset is not blocked */
1532 	if (hw->mac.type >= e1000_82580 &&
1533 	    (e1000_check_reset_block(hw) != E1000_BLK_PHY_RESET)) {
1534 		uint32_t phpm_reg;
1535 
1536 		phpm_reg = E1000_READ_REG(hw, E1000_82580_PHY_POWER_MGMT);
1537 		phpm_reg &= ~E1000_82580_PM_GO_LINKD;
1538 		E1000_WRITE_REG(hw, E1000_82580_PHY_POWER_MGMT, phpm_reg);
1539 	}
1540 
1541 	igb_dev_free_queues(dev);
1542 
1543 	if (intr_handle->intr_vec) {
1544 		rte_free(intr_handle->intr_vec);
1545 		intr_handle->intr_vec = NULL;
1546 	}
1547 
1548 	memset(&link, 0, sizeof(link));
1549 	rte_eth_linkstatus_set(dev, &link);
1550 
1551 	/* Reset any pending lock */
1552 	igb_reset_swfw_lock(hw);
1553 
1554 	/* uninitialize PF if max_vfs not zero */
1555 	igb_pf_host_uninit(dev);
1556 
1557 	rte_intr_callback_unregister(intr_handle,
1558 				     eth_igb_interrupt_handler, dev);
1559 
1560 	/* clear the SYN filter info */
1561 	filter_info->syn_info = 0;
1562 
1563 	/* clear the ethertype filters info */
1564 	filter_info->ethertype_mask = 0;
1565 	memset(filter_info->ethertype_filters, 0,
1566 		E1000_MAX_ETQF_FILTERS * sizeof(struct igb_ethertype_filter));
1567 
1568 	/* clear the rss filter info */
1569 	memset(&filter_info->rss_info, 0,
1570 		sizeof(struct igb_rte_flow_rss_conf));
1571 
1572 	/* remove all ntuple filters of the device */
1573 	igb_ntuple_filter_uninit(dev);
1574 
1575 	/* remove all flex filters of the device */
1576 	igb_flex_filter_uninit(dev);
1577 
1578 	/* clear all the filters list */
1579 	igb_filterlist_flush(dev);
1580 
1581 	return ret;
1582 }
1583 
1584 /*
1585  * Reset PF device.
1586  */
1587 static int
1588 eth_igb_reset(struct rte_eth_dev *dev)
1589 {
1590 	int ret;
1591 
1592 	/* When a DPDK PMD PF begin to reset PF port, it should notify all
1593 	 * its VF to make them align with it. The detailed notification
1594 	 * mechanism is PMD specific and is currently not implemented.
1595 	 * To avoid unexpected behavior in VF, currently reset of PF with
1596 	 * SR-IOV activation is not supported. It might be supported later.
1597 	 */
1598 	if (dev->data->sriov.active)
1599 		return -ENOTSUP;
1600 
1601 	ret = eth_igb_dev_uninit(dev);
1602 	if (ret)
1603 		return ret;
1604 
1605 	ret = eth_igb_dev_init(dev);
1606 
1607 	return ret;
1608 }
1609 
1610 
1611 static int
1612 igb_get_rx_buffer_size(struct e1000_hw *hw)
1613 {
1614 	uint32_t rx_buf_size;
1615 	if (hw->mac.type == e1000_82576) {
1616 		rx_buf_size = (E1000_READ_REG(hw, E1000_RXPBS) & 0xffff) << 10;
1617 	} else if (hw->mac.type == e1000_82580 || hw->mac.type == e1000_i350) {
1618 		/* PBS needs to be translated according to a lookup table */
1619 		rx_buf_size = (E1000_READ_REG(hw, E1000_RXPBS) & 0xf);
1620 		rx_buf_size = (uint32_t) e1000_rxpbs_adjust_82580(rx_buf_size);
1621 		rx_buf_size = (rx_buf_size << 10);
1622 	} else if (hw->mac.type == e1000_i210 || hw->mac.type == e1000_i211) {
1623 		rx_buf_size = (E1000_READ_REG(hw, E1000_RXPBS) & 0x3f) << 10;
1624 	} else {
1625 		rx_buf_size = (E1000_READ_REG(hw, E1000_PBA) & 0xffff) << 10;
1626 	}
1627 
1628 	return rx_buf_size;
1629 }
1630 
1631 /*********************************************************************
1632  *
1633  *  Initialize the hardware
1634  *
1635  **********************************************************************/
1636 static int
1637 igb_hardware_init(struct e1000_hw *hw)
1638 {
1639 	uint32_t rx_buf_size;
1640 	int diag;
1641 
1642 	/* Let the firmware know the OS is in control */
1643 	igb_hw_control_acquire(hw);
1644 
1645 	/*
1646 	 * These parameters control the automatic generation (Tx) and
1647 	 * response (Rx) to Ethernet PAUSE frames.
1648 	 * - High water mark should allow for at least two standard size (1518)
1649 	 *   frames to be received after sending an XOFF.
1650 	 * - Low water mark works best when it is very near the high water mark.
1651 	 *   This allows the receiver to restart by sending XON when it has
1652 	 *   drained a bit. Here we use an arbitrary value of 1500 which will
1653 	 *   restart after one full frame is pulled from the buffer. There
1654 	 *   could be several smaller frames in the buffer and if so they will
1655 	 *   not trigger the XON until their total number reduces the buffer
1656 	 *   by 1500.
1657 	 * - The pause time is fairly large at 1000 x 512ns = 512 usec.
1658 	 */
1659 	rx_buf_size = igb_get_rx_buffer_size(hw);
1660 
1661 	hw->fc.high_water = rx_buf_size - (RTE_ETHER_MAX_LEN * 2);
1662 	hw->fc.low_water = hw->fc.high_water - 1500;
1663 	hw->fc.pause_time = IGB_FC_PAUSE_TIME;
1664 	hw->fc.send_xon = 1;
1665 
1666 	/* Set Flow control, use the tunable location if sane */
1667 	if ((igb_fc_setting != e1000_fc_none) && (igb_fc_setting < 4))
1668 		hw->fc.requested_mode = igb_fc_setting;
1669 	else
1670 		hw->fc.requested_mode = e1000_fc_none;
1671 
1672 	/* Issue a global reset */
1673 	igb_pf_reset_hw(hw);
1674 	E1000_WRITE_REG(hw, E1000_WUC, 0);
1675 
1676 	diag = e1000_init_hw(hw);
1677 	if (diag < 0)
1678 		return diag;
1679 
1680 	E1000_WRITE_REG(hw, E1000_VET,
1681 			RTE_ETHER_TYPE_VLAN << 16 | RTE_ETHER_TYPE_VLAN);
1682 	e1000_get_phy_info(hw);
1683 	e1000_check_for_link(hw);
1684 
1685 	return 0;
1686 }
1687 
1688 /* This function is based on igb_update_stats_counters() in igb/if_igb.c */
1689 static void
1690 igb_read_stats_registers(struct e1000_hw *hw, struct e1000_hw_stats *stats)
1691 {
1692 	int pause_frames;
1693 
1694 	uint64_t old_gprc  = stats->gprc;
1695 	uint64_t old_gptc  = stats->gptc;
1696 	uint64_t old_tpr   = stats->tpr;
1697 	uint64_t old_tpt   = stats->tpt;
1698 	uint64_t old_rpthc = stats->rpthc;
1699 	uint64_t old_hgptc = stats->hgptc;
1700 
1701 	if(hw->phy.media_type == e1000_media_type_copper ||
1702 	    (E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU)) {
1703 		stats->symerrs +=
1704 		    E1000_READ_REG(hw,E1000_SYMERRS);
1705 		stats->sec += E1000_READ_REG(hw, E1000_SEC);
1706 	}
1707 
1708 	stats->crcerrs += E1000_READ_REG(hw, E1000_CRCERRS);
1709 	stats->mpc += E1000_READ_REG(hw, E1000_MPC);
1710 	stats->scc += E1000_READ_REG(hw, E1000_SCC);
1711 	stats->ecol += E1000_READ_REG(hw, E1000_ECOL);
1712 
1713 	stats->mcc += E1000_READ_REG(hw, E1000_MCC);
1714 	stats->latecol += E1000_READ_REG(hw, E1000_LATECOL);
1715 	stats->colc += E1000_READ_REG(hw, E1000_COLC);
1716 	stats->dc += E1000_READ_REG(hw, E1000_DC);
1717 	stats->rlec += E1000_READ_REG(hw, E1000_RLEC);
1718 	stats->xonrxc += E1000_READ_REG(hw, E1000_XONRXC);
1719 	stats->xontxc += E1000_READ_REG(hw, E1000_XONTXC);
1720 	/*
1721 	** For watchdog management we need to know if we have been
1722 	** paused during the last interval, so capture that here.
1723 	*/
1724 	pause_frames = E1000_READ_REG(hw, E1000_XOFFRXC);
1725 	stats->xoffrxc += pause_frames;
1726 	stats->xofftxc += E1000_READ_REG(hw, E1000_XOFFTXC);
1727 	stats->fcruc += E1000_READ_REG(hw, E1000_FCRUC);
1728 	stats->prc64 += E1000_READ_REG(hw, E1000_PRC64);
1729 	stats->prc127 += E1000_READ_REG(hw, E1000_PRC127);
1730 	stats->prc255 += E1000_READ_REG(hw, E1000_PRC255);
1731 	stats->prc511 += E1000_READ_REG(hw, E1000_PRC511);
1732 	stats->prc1023 += E1000_READ_REG(hw, E1000_PRC1023);
1733 	stats->prc1522 += E1000_READ_REG(hw, E1000_PRC1522);
1734 	stats->gprc += E1000_READ_REG(hw, E1000_GPRC);
1735 	stats->bprc += E1000_READ_REG(hw, E1000_BPRC);
1736 	stats->mprc += E1000_READ_REG(hw, E1000_MPRC);
1737 	stats->gptc += E1000_READ_REG(hw, E1000_GPTC);
1738 
1739 	/* For the 64-bit byte counters the low dword must be read first. */
1740 	/* Both registers clear on the read of the high dword */
1741 
1742 	/* Workaround CRC bytes included in size, take away 4 bytes/packet */
1743 	stats->gorc += E1000_READ_REG(hw, E1000_GORCL);
1744 	stats->gorc += ((uint64_t)E1000_READ_REG(hw, E1000_GORCH) << 32);
1745 	stats->gorc -= (stats->gprc - old_gprc) * RTE_ETHER_CRC_LEN;
1746 	stats->gotc += E1000_READ_REG(hw, E1000_GOTCL);
1747 	stats->gotc += ((uint64_t)E1000_READ_REG(hw, E1000_GOTCH) << 32);
1748 	stats->gotc -= (stats->gptc - old_gptc) * RTE_ETHER_CRC_LEN;
1749 
1750 	stats->rnbc += E1000_READ_REG(hw, E1000_RNBC);
1751 	stats->ruc += E1000_READ_REG(hw, E1000_RUC);
1752 	stats->rfc += E1000_READ_REG(hw, E1000_RFC);
1753 	stats->roc += E1000_READ_REG(hw, E1000_ROC);
1754 	stats->rjc += E1000_READ_REG(hw, E1000_RJC);
1755 
1756 	stats->tpr += E1000_READ_REG(hw, E1000_TPR);
1757 	stats->tpt += E1000_READ_REG(hw, E1000_TPT);
1758 
1759 	stats->tor += E1000_READ_REG(hw, E1000_TORL);
1760 	stats->tor += ((uint64_t)E1000_READ_REG(hw, E1000_TORH) << 32);
1761 	stats->tor -= (stats->tpr - old_tpr) * RTE_ETHER_CRC_LEN;
1762 	stats->tot += E1000_READ_REG(hw, E1000_TOTL);
1763 	stats->tot += ((uint64_t)E1000_READ_REG(hw, E1000_TOTH) << 32);
1764 	stats->tot -= (stats->tpt - old_tpt) * RTE_ETHER_CRC_LEN;
1765 
1766 	stats->ptc64 += E1000_READ_REG(hw, E1000_PTC64);
1767 	stats->ptc127 += E1000_READ_REG(hw, E1000_PTC127);
1768 	stats->ptc255 += E1000_READ_REG(hw, E1000_PTC255);
1769 	stats->ptc511 += E1000_READ_REG(hw, E1000_PTC511);
1770 	stats->ptc1023 += E1000_READ_REG(hw, E1000_PTC1023);
1771 	stats->ptc1522 += E1000_READ_REG(hw, E1000_PTC1522);
1772 	stats->mptc += E1000_READ_REG(hw, E1000_MPTC);
1773 	stats->bptc += E1000_READ_REG(hw, E1000_BPTC);
1774 
1775 	/* Interrupt Counts */
1776 
1777 	stats->iac += E1000_READ_REG(hw, E1000_IAC);
1778 	stats->icrxptc += E1000_READ_REG(hw, E1000_ICRXPTC);
1779 	stats->icrxatc += E1000_READ_REG(hw, E1000_ICRXATC);
1780 	stats->ictxptc += E1000_READ_REG(hw, E1000_ICTXPTC);
1781 	stats->ictxatc += E1000_READ_REG(hw, E1000_ICTXATC);
1782 	stats->ictxqec += E1000_READ_REG(hw, E1000_ICTXQEC);
1783 	stats->ictxqmtc += E1000_READ_REG(hw, E1000_ICTXQMTC);
1784 	stats->icrxdmtc += E1000_READ_REG(hw, E1000_ICRXDMTC);
1785 	stats->icrxoc += E1000_READ_REG(hw, E1000_ICRXOC);
1786 
1787 	/* Host to Card Statistics */
1788 
1789 	stats->cbtmpc += E1000_READ_REG(hw, E1000_CBTMPC);
1790 	stats->htdpmc += E1000_READ_REG(hw, E1000_HTDPMC);
1791 	stats->cbrdpc += E1000_READ_REG(hw, E1000_CBRDPC);
1792 	stats->cbrmpc += E1000_READ_REG(hw, E1000_CBRMPC);
1793 	stats->rpthc += E1000_READ_REG(hw, E1000_RPTHC);
1794 	stats->hgptc += E1000_READ_REG(hw, E1000_HGPTC);
1795 	stats->htcbdpc += E1000_READ_REG(hw, E1000_HTCBDPC);
1796 	stats->hgorc += E1000_READ_REG(hw, E1000_HGORCL);
1797 	stats->hgorc += ((uint64_t)E1000_READ_REG(hw, E1000_HGORCH) << 32);
1798 	stats->hgorc -= (stats->rpthc - old_rpthc) * RTE_ETHER_CRC_LEN;
1799 	stats->hgotc += E1000_READ_REG(hw, E1000_HGOTCL);
1800 	stats->hgotc += ((uint64_t)E1000_READ_REG(hw, E1000_HGOTCH) << 32);
1801 	stats->hgotc -= (stats->hgptc - old_hgptc) * RTE_ETHER_CRC_LEN;
1802 	stats->lenerrs += E1000_READ_REG(hw, E1000_LENERRS);
1803 	stats->scvpc += E1000_READ_REG(hw, E1000_SCVPC);
1804 	stats->hrmpc += E1000_READ_REG(hw, E1000_HRMPC);
1805 
1806 	stats->algnerrc += E1000_READ_REG(hw, E1000_ALGNERRC);
1807 	stats->rxerrc += E1000_READ_REG(hw, E1000_RXERRC);
1808 	stats->tncrs += E1000_READ_REG(hw, E1000_TNCRS);
1809 	stats->cexterr += E1000_READ_REG(hw, E1000_CEXTERR);
1810 	stats->tsctc += E1000_READ_REG(hw, E1000_TSCTC);
1811 	stats->tsctfc += E1000_READ_REG(hw, E1000_TSCTFC);
1812 }
1813 
1814 static int
1815 eth_igb_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *rte_stats)
1816 {
1817 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1818 	struct e1000_hw_stats *stats =
1819 			E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
1820 
1821 	igb_read_stats_registers(hw, stats);
1822 
1823 	if (rte_stats == NULL)
1824 		return -EINVAL;
1825 
1826 	/* Rx Errors */
1827 	rte_stats->imissed = stats->mpc;
1828 	rte_stats->ierrors = stats->crcerrs + stats->rlec +
1829 	                     stats->rxerrc + stats->algnerrc + stats->cexterr;
1830 
1831 	/* Tx Errors */
1832 	rte_stats->oerrors = stats->ecol + stats->latecol;
1833 
1834 	rte_stats->ipackets = stats->gprc;
1835 	rte_stats->opackets = stats->gptc;
1836 	rte_stats->ibytes   = stats->gorc;
1837 	rte_stats->obytes   = stats->gotc;
1838 	return 0;
1839 }
1840 
1841 static int
1842 eth_igb_stats_reset(struct rte_eth_dev *dev)
1843 {
1844 	struct e1000_hw_stats *hw_stats =
1845 			E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
1846 
1847 	/* HW registers are cleared on read */
1848 	eth_igb_stats_get(dev, NULL);
1849 
1850 	/* Reset software totals */
1851 	memset(hw_stats, 0, sizeof(*hw_stats));
1852 
1853 	return 0;
1854 }
1855 
1856 static int
1857 eth_igb_xstats_reset(struct rte_eth_dev *dev)
1858 {
1859 	struct e1000_hw_stats *stats =
1860 			E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
1861 
1862 	/* HW registers are cleared on read */
1863 	eth_igb_xstats_get(dev, NULL, IGB_NB_XSTATS);
1864 
1865 	/* Reset software totals */
1866 	memset(stats, 0, sizeof(*stats));
1867 
1868 	return 0;
1869 }
1870 
1871 static int eth_igb_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
1872 	struct rte_eth_xstat_name *xstats_names,
1873 	__rte_unused unsigned int size)
1874 {
1875 	unsigned i;
1876 
1877 	if (xstats_names == NULL)
1878 		return IGB_NB_XSTATS;
1879 
1880 	/* Note: limit checked in rte_eth_xstats_names() */
1881 
1882 	for (i = 0; i < IGB_NB_XSTATS; i++) {
1883 		strlcpy(xstats_names[i].name, rte_igb_stats_strings[i].name,
1884 			sizeof(xstats_names[i].name));
1885 	}
1886 
1887 	return IGB_NB_XSTATS;
1888 }
1889 
1890 static int eth_igb_xstats_get_names_by_id(struct rte_eth_dev *dev,
1891 		struct rte_eth_xstat_name *xstats_names, const uint64_t *ids,
1892 		unsigned int limit)
1893 {
1894 	unsigned int i;
1895 
1896 	if (!ids) {
1897 		if (xstats_names == NULL)
1898 			return IGB_NB_XSTATS;
1899 
1900 		for (i = 0; i < IGB_NB_XSTATS; i++)
1901 			strlcpy(xstats_names[i].name,
1902 				rte_igb_stats_strings[i].name,
1903 				sizeof(xstats_names[i].name));
1904 
1905 		return IGB_NB_XSTATS;
1906 
1907 	} else {
1908 		struct rte_eth_xstat_name xstats_names_copy[IGB_NB_XSTATS];
1909 
1910 		eth_igb_xstats_get_names_by_id(dev, xstats_names_copy, NULL,
1911 				IGB_NB_XSTATS);
1912 
1913 		for (i = 0; i < limit; i++) {
1914 			if (ids[i] >= IGB_NB_XSTATS) {
1915 				PMD_INIT_LOG(ERR, "id value isn't valid");
1916 				return -1;
1917 			}
1918 			strcpy(xstats_names[i].name,
1919 					xstats_names_copy[ids[i]].name);
1920 		}
1921 		return limit;
1922 	}
1923 }
1924 
1925 static int
1926 eth_igb_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1927 		   unsigned n)
1928 {
1929 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1930 	struct e1000_hw_stats *hw_stats =
1931 			E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
1932 	unsigned i;
1933 
1934 	if (n < IGB_NB_XSTATS)
1935 		return IGB_NB_XSTATS;
1936 
1937 	igb_read_stats_registers(hw, hw_stats);
1938 
1939 	/* If this is a reset xstats is NULL, and we have cleared the
1940 	 * registers by reading them.
1941 	 */
1942 	if (!xstats)
1943 		return 0;
1944 
1945 	/* Extended stats */
1946 	for (i = 0; i < IGB_NB_XSTATS; i++) {
1947 		xstats[i].id = i;
1948 		xstats[i].value = *(uint64_t *)(((char *)hw_stats) +
1949 			rte_igb_stats_strings[i].offset);
1950 	}
1951 
1952 	return IGB_NB_XSTATS;
1953 }
1954 
1955 static int
1956 eth_igb_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
1957 		uint64_t *values, unsigned int n)
1958 {
1959 	unsigned int i;
1960 
1961 	if (!ids) {
1962 		struct e1000_hw *hw =
1963 			E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1964 		struct e1000_hw_stats *hw_stats =
1965 			E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
1966 
1967 		if (n < IGB_NB_XSTATS)
1968 			return IGB_NB_XSTATS;
1969 
1970 		igb_read_stats_registers(hw, hw_stats);
1971 
1972 		/* If this is a reset xstats is NULL, and we have cleared the
1973 		 * registers by reading them.
1974 		 */
1975 		if (!values)
1976 			return 0;
1977 
1978 		/* Extended stats */
1979 		for (i = 0; i < IGB_NB_XSTATS; i++)
1980 			values[i] = *(uint64_t *)(((char *)hw_stats) +
1981 					rte_igb_stats_strings[i].offset);
1982 
1983 		return IGB_NB_XSTATS;
1984 
1985 	} else {
1986 		uint64_t values_copy[IGB_NB_XSTATS];
1987 
1988 		eth_igb_xstats_get_by_id(dev, NULL, values_copy,
1989 				IGB_NB_XSTATS);
1990 
1991 		for (i = 0; i < n; i++) {
1992 			if (ids[i] >= IGB_NB_XSTATS) {
1993 				PMD_INIT_LOG(ERR, "id value isn't valid");
1994 				return -1;
1995 			}
1996 			values[i] = values_copy[ids[i]];
1997 		}
1998 		return n;
1999 	}
2000 }
2001 
2002 static void
2003 igbvf_read_stats_registers(struct e1000_hw *hw, struct e1000_vf_stats *hw_stats)
2004 {
2005 	/* Good Rx packets, include VF loopback */
2006 	UPDATE_VF_STAT(E1000_VFGPRC,
2007 	    hw_stats->last_gprc, hw_stats->gprc);
2008 
2009 	/* Good Rx octets, include VF loopback */
2010 	UPDATE_VF_STAT(E1000_VFGORC,
2011 	    hw_stats->last_gorc, hw_stats->gorc);
2012 
2013 	/* Good Tx packets, include VF loopback */
2014 	UPDATE_VF_STAT(E1000_VFGPTC,
2015 	    hw_stats->last_gptc, hw_stats->gptc);
2016 
2017 	/* Good Tx octets, include VF loopback */
2018 	UPDATE_VF_STAT(E1000_VFGOTC,
2019 	    hw_stats->last_gotc, hw_stats->gotc);
2020 
2021 	/* Rx Multicst packets */
2022 	UPDATE_VF_STAT(E1000_VFMPRC,
2023 	    hw_stats->last_mprc, hw_stats->mprc);
2024 
2025 	/* Good Rx loopback packets */
2026 	UPDATE_VF_STAT(E1000_VFGPRLBC,
2027 	    hw_stats->last_gprlbc, hw_stats->gprlbc);
2028 
2029 	/* Good Rx loopback octets */
2030 	UPDATE_VF_STAT(E1000_VFGORLBC,
2031 	    hw_stats->last_gorlbc, hw_stats->gorlbc);
2032 
2033 	/* Good Tx loopback packets */
2034 	UPDATE_VF_STAT(E1000_VFGPTLBC,
2035 	    hw_stats->last_gptlbc, hw_stats->gptlbc);
2036 
2037 	/* Good Tx loopback octets */
2038 	UPDATE_VF_STAT(E1000_VFGOTLBC,
2039 	    hw_stats->last_gotlbc, hw_stats->gotlbc);
2040 }
2041 
2042 static int eth_igbvf_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
2043 				     struct rte_eth_xstat_name *xstats_names,
2044 				     __rte_unused unsigned limit)
2045 {
2046 	unsigned i;
2047 
2048 	if (xstats_names != NULL)
2049 		for (i = 0; i < IGBVF_NB_XSTATS; i++) {
2050 			strlcpy(xstats_names[i].name,
2051 				rte_igbvf_stats_strings[i].name,
2052 				sizeof(xstats_names[i].name));
2053 		}
2054 	return IGBVF_NB_XSTATS;
2055 }
2056 
2057 static int
2058 eth_igbvf_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
2059 		     unsigned n)
2060 {
2061 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2062 	struct e1000_vf_stats *hw_stats = (struct e1000_vf_stats *)
2063 			E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
2064 	unsigned i;
2065 
2066 	if (n < IGBVF_NB_XSTATS)
2067 		return IGBVF_NB_XSTATS;
2068 
2069 	igbvf_read_stats_registers(hw, hw_stats);
2070 
2071 	if (!xstats)
2072 		return 0;
2073 
2074 	for (i = 0; i < IGBVF_NB_XSTATS; i++) {
2075 		xstats[i].id = i;
2076 		xstats[i].value = *(uint64_t *)(((char *)hw_stats) +
2077 			rte_igbvf_stats_strings[i].offset);
2078 	}
2079 
2080 	return IGBVF_NB_XSTATS;
2081 }
2082 
2083 static int
2084 eth_igbvf_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *rte_stats)
2085 {
2086 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2087 	struct e1000_vf_stats *hw_stats = (struct e1000_vf_stats *)
2088 			  E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
2089 
2090 	igbvf_read_stats_registers(hw, hw_stats);
2091 
2092 	if (rte_stats == NULL)
2093 		return -EINVAL;
2094 
2095 	rte_stats->ipackets = hw_stats->gprc;
2096 	rte_stats->ibytes = hw_stats->gorc;
2097 	rte_stats->opackets = hw_stats->gptc;
2098 	rte_stats->obytes = hw_stats->gotc;
2099 	return 0;
2100 }
2101 
2102 static int
2103 eth_igbvf_stats_reset(struct rte_eth_dev *dev)
2104 {
2105 	struct e1000_vf_stats *hw_stats = (struct e1000_vf_stats*)
2106 			E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
2107 
2108 	/* Sync HW register to the last stats */
2109 	eth_igbvf_stats_get(dev, NULL);
2110 
2111 	/* reset HW current stats*/
2112 	memset(&hw_stats->gprc, 0, sizeof(*hw_stats) -
2113 	       offsetof(struct e1000_vf_stats, gprc));
2114 
2115 	return 0;
2116 }
2117 
2118 static int
2119 eth_igb_fw_version_get(struct rte_eth_dev *dev, char *fw_version,
2120 		       size_t fw_size)
2121 {
2122 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2123 	struct e1000_fw_version fw;
2124 	int ret;
2125 
2126 	e1000_get_fw_version(hw, &fw);
2127 
2128 	switch (hw->mac.type) {
2129 	case e1000_i210:
2130 	case e1000_i211:
2131 		if (!(e1000_get_flash_presence_i210(hw))) {
2132 			ret = snprintf(fw_version, fw_size,
2133 				 "%2d.%2d-%d",
2134 				 fw.invm_major, fw.invm_minor,
2135 				 fw.invm_img_type);
2136 			break;
2137 		}
2138 		/* fall through */
2139 	default:
2140 		/* if option rom is valid, display its version too */
2141 		if (fw.or_valid) {
2142 			ret = snprintf(fw_version, fw_size,
2143 				 "%d.%d, 0x%08x, %d.%d.%d",
2144 				 fw.eep_major, fw.eep_minor, fw.etrack_id,
2145 				 fw.or_major, fw.or_build, fw.or_patch);
2146 		/* no option rom */
2147 		} else {
2148 			if (fw.etrack_id != 0X0000) {
2149 				ret = snprintf(fw_version, fw_size,
2150 					 "%d.%d, 0x%08x",
2151 					 fw.eep_major, fw.eep_minor,
2152 					 fw.etrack_id);
2153 			} else {
2154 				ret = snprintf(fw_version, fw_size,
2155 					 "%d.%d.%d",
2156 					 fw.eep_major, fw.eep_minor,
2157 					 fw.eep_build);
2158 			}
2159 		}
2160 		break;
2161 	}
2162 	if (ret < 0)
2163 		return -EINVAL;
2164 
2165 	ret += 1; /* add the size of '\0' */
2166 	if (fw_size < (size_t)ret)
2167 		return ret;
2168 	else
2169 		return 0;
2170 }
2171 
2172 static int
2173 eth_igb_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2174 {
2175 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2176 
2177 	dev_info->min_rx_bufsize = 256; /* See BSIZE field of RCTL register. */
2178 	dev_info->max_rx_pktlen  = 0x3FFF; /* See RLPML register. */
2179 	dev_info->max_mac_addrs = hw->mac.rar_entry_count;
2180 	dev_info->rx_queue_offload_capa = igb_get_rx_queue_offloads_capa(dev);
2181 	dev_info->rx_offload_capa = igb_get_rx_port_offloads_capa(dev) |
2182 				    dev_info->rx_queue_offload_capa;
2183 	dev_info->tx_queue_offload_capa = igb_get_tx_queue_offloads_capa(dev);
2184 	dev_info->tx_offload_capa = igb_get_tx_port_offloads_capa(dev) |
2185 				    dev_info->tx_queue_offload_capa;
2186 
2187 	switch (hw->mac.type) {
2188 	case e1000_82575:
2189 		dev_info->max_rx_queues = 4;
2190 		dev_info->max_tx_queues = 4;
2191 		dev_info->max_vmdq_pools = 0;
2192 		break;
2193 
2194 	case e1000_82576:
2195 		dev_info->max_rx_queues = 16;
2196 		dev_info->max_tx_queues = 16;
2197 		dev_info->max_vmdq_pools = ETH_8_POOLS;
2198 		dev_info->vmdq_queue_num = 16;
2199 		break;
2200 
2201 	case e1000_82580:
2202 		dev_info->max_rx_queues = 8;
2203 		dev_info->max_tx_queues = 8;
2204 		dev_info->max_vmdq_pools = ETH_8_POOLS;
2205 		dev_info->vmdq_queue_num = 8;
2206 		break;
2207 
2208 	case e1000_i350:
2209 		dev_info->max_rx_queues = 8;
2210 		dev_info->max_tx_queues = 8;
2211 		dev_info->max_vmdq_pools = ETH_8_POOLS;
2212 		dev_info->vmdq_queue_num = 8;
2213 		break;
2214 
2215 	case e1000_i354:
2216 		dev_info->max_rx_queues = 8;
2217 		dev_info->max_tx_queues = 8;
2218 		break;
2219 
2220 	case e1000_i210:
2221 		dev_info->max_rx_queues = 4;
2222 		dev_info->max_tx_queues = 4;
2223 		dev_info->max_vmdq_pools = 0;
2224 		break;
2225 
2226 	case e1000_i211:
2227 		dev_info->max_rx_queues = 2;
2228 		dev_info->max_tx_queues = 2;
2229 		dev_info->max_vmdq_pools = 0;
2230 		break;
2231 
2232 	default:
2233 		/* Should not happen */
2234 		return -EINVAL;
2235 	}
2236 	dev_info->hash_key_size = IGB_HKEY_MAX_INDEX * sizeof(uint32_t);
2237 	dev_info->reta_size = ETH_RSS_RETA_SIZE_128;
2238 	dev_info->flow_type_rss_offloads = IGB_RSS_OFFLOAD_ALL;
2239 
2240 	dev_info->default_rxconf = (struct rte_eth_rxconf) {
2241 		.rx_thresh = {
2242 			.pthresh = IGB_DEFAULT_RX_PTHRESH,
2243 			.hthresh = IGB_DEFAULT_RX_HTHRESH,
2244 			.wthresh = IGB_DEFAULT_RX_WTHRESH,
2245 		},
2246 		.rx_free_thresh = IGB_DEFAULT_RX_FREE_THRESH,
2247 		.rx_drop_en = 0,
2248 		.offloads = 0,
2249 	};
2250 
2251 	dev_info->default_txconf = (struct rte_eth_txconf) {
2252 		.tx_thresh = {
2253 			.pthresh = IGB_DEFAULT_TX_PTHRESH,
2254 			.hthresh = IGB_DEFAULT_TX_HTHRESH,
2255 			.wthresh = IGB_DEFAULT_TX_WTHRESH,
2256 		},
2257 		.offloads = 0,
2258 	};
2259 
2260 	dev_info->rx_desc_lim = rx_desc_lim;
2261 	dev_info->tx_desc_lim = tx_desc_lim;
2262 
2263 	dev_info->speed_capa = ETH_LINK_SPEED_10M_HD | ETH_LINK_SPEED_10M |
2264 			ETH_LINK_SPEED_100M_HD | ETH_LINK_SPEED_100M |
2265 			ETH_LINK_SPEED_1G;
2266 
2267 	dev_info->max_mtu = dev_info->max_rx_pktlen - E1000_ETH_OVERHEAD;
2268 	dev_info->min_mtu = RTE_ETHER_MIN_MTU;
2269 
2270 	return 0;
2271 }
2272 
2273 static const uint32_t *
2274 eth_igb_supported_ptypes_get(struct rte_eth_dev *dev)
2275 {
2276 	static const uint32_t ptypes[] = {
2277 		/* refers to igb_rxd_pkt_info_to_pkt_type() */
2278 		RTE_PTYPE_L2_ETHER,
2279 		RTE_PTYPE_L3_IPV4,
2280 		RTE_PTYPE_L3_IPV4_EXT,
2281 		RTE_PTYPE_L3_IPV6,
2282 		RTE_PTYPE_L3_IPV6_EXT,
2283 		RTE_PTYPE_L4_TCP,
2284 		RTE_PTYPE_L4_UDP,
2285 		RTE_PTYPE_L4_SCTP,
2286 		RTE_PTYPE_TUNNEL_IP,
2287 		RTE_PTYPE_INNER_L3_IPV6,
2288 		RTE_PTYPE_INNER_L3_IPV6_EXT,
2289 		RTE_PTYPE_INNER_L4_TCP,
2290 		RTE_PTYPE_INNER_L4_UDP,
2291 		RTE_PTYPE_UNKNOWN
2292 	};
2293 
2294 	if (dev->rx_pkt_burst == eth_igb_recv_pkts ||
2295 	    dev->rx_pkt_burst == eth_igb_recv_scattered_pkts)
2296 		return ptypes;
2297 	return NULL;
2298 }
2299 
2300 static int
2301 eth_igbvf_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2302 {
2303 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2304 
2305 	dev_info->min_rx_bufsize = 256; /* See BSIZE field of RCTL register. */
2306 	dev_info->max_rx_pktlen  = 0x3FFF; /* See RLPML register. */
2307 	dev_info->max_mac_addrs = hw->mac.rar_entry_count;
2308 	dev_info->tx_offload_capa = DEV_TX_OFFLOAD_VLAN_INSERT |
2309 				DEV_TX_OFFLOAD_IPV4_CKSUM  |
2310 				DEV_TX_OFFLOAD_UDP_CKSUM   |
2311 				DEV_TX_OFFLOAD_TCP_CKSUM   |
2312 				DEV_TX_OFFLOAD_SCTP_CKSUM  |
2313 				DEV_TX_OFFLOAD_TCP_TSO;
2314 	switch (hw->mac.type) {
2315 	case e1000_vfadapt:
2316 		dev_info->max_rx_queues = 2;
2317 		dev_info->max_tx_queues = 2;
2318 		break;
2319 	case e1000_vfadapt_i350:
2320 		dev_info->max_rx_queues = 1;
2321 		dev_info->max_tx_queues = 1;
2322 		break;
2323 	default:
2324 		/* Should not happen */
2325 		return -EINVAL;
2326 	}
2327 
2328 	dev_info->rx_queue_offload_capa = igb_get_rx_queue_offloads_capa(dev);
2329 	dev_info->rx_offload_capa = igb_get_rx_port_offloads_capa(dev) |
2330 				    dev_info->rx_queue_offload_capa;
2331 	dev_info->tx_queue_offload_capa = igb_get_tx_queue_offloads_capa(dev);
2332 	dev_info->tx_offload_capa = igb_get_tx_port_offloads_capa(dev) |
2333 				    dev_info->tx_queue_offload_capa;
2334 
2335 	dev_info->default_rxconf = (struct rte_eth_rxconf) {
2336 		.rx_thresh = {
2337 			.pthresh = IGB_DEFAULT_RX_PTHRESH,
2338 			.hthresh = IGB_DEFAULT_RX_HTHRESH,
2339 			.wthresh = IGB_DEFAULT_RX_WTHRESH,
2340 		},
2341 		.rx_free_thresh = IGB_DEFAULT_RX_FREE_THRESH,
2342 		.rx_drop_en = 0,
2343 		.offloads = 0,
2344 	};
2345 
2346 	dev_info->default_txconf = (struct rte_eth_txconf) {
2347 		.tx_thresh = {
2348 			.pthresh = IGB_DEFAULT_TX_PTHRESH,
2349 			.hthresh = IGB_DEFAULT_TX_HTHRESH,
2350 			.wthresh = IGB_DEFAULT_TX_WTHRESH,
2351 		},
2352 		.offloads = 0,
2353 	};
2354 
2355 	dev_info->rx_desc_lim = rx_desc_lim;
2356 	dev_info->tx_desc_lim = tx_desc_lim;
2357 
2358 	return 0;
2359 }
2360 
2361 /* return 0 means link status changed, -1 means not changed */
2362 static int
2363 eth_igb_link_update(struct rte_eth_dev *dev, int wait_to_complete)
2364 {
2365 	struct e1000_hw *hw =
2366 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2367 	struct rte_eth_link link;
2368 	int link_check, count;
2369 
2370 	link_check = 0;
2371 	hw->mac.get_link_status = 1;
2372 
2373 	/* possible wait-to-complete in up to 9 seconds */
2374 	for (count = 0; count < IGB_LINK_UPDATE_CHECK_TIMEOUT; count ++) {
2375 		/* Read the real link status */
2376 		switch (hw->phy.media_type) {
2377 		case e1000_media_type_copper:
2378 			/* Do the work to read phy */
2379 			e1000_check_for_link(hw);
2380 			link_check = !hw->mac.get_link_status;
2381 			break;
2382 
2383 		case e1000_media_type_fiber:
2384 			e1000_check_for_link(hw);
2385 			link_check = (E1000_READ_REG(hw, E1000_STATUS) &
2386 				      E1000_STATUS_LU);
2387 			break;
2388 
2389 		case e1000_media_type_internal_serdes:
2390 			e1000_check_for_link(hw);
2391 			link_check = hw->mac.serdes_has_link;
2392 			break;
2393 
2394 		/* VF device is type_unknown */
2395 		case e1000_media_type_unknown:
2396 			eth_igbvf_link_update(hw);
2397 			link_check = !hw->mac.get_link_status;
2398 			break;
2399 
2400 		default:
2401 			break;
2402 		}
2403 		if (link_check || wait_to_complete == 0)
2404 			break;
2405 		rte_delay_ms(IGB_LINK_UPDATE_CHECK_INTERVAL);
2406 	}
2407 	memset(&link, 0, sizeof(link));
2408 
2409 	/* Now we check if a transition has happened */
2410 	if (link_check) {
2411 		uint16_t duplex, speed;
2412 		hw->mac.ops.get_link_up_info(hw, &speed, &duplex);
2413 		link.link_duplex = (duplex == FULL_DUPLEX) ?
2414 				ETH_LINK_FULL_DUPLEX :
2415 				ETH_LINK_HALF_DUPLEX;
2416 		link.link_speed = speed;
2417 		link.link_status = ETH_LINK_UP;
2418 		link.link_autoneg = !(dev->data->dev_conf.link_speeds &
2419 				ETH_LINK_SPEED_FIXED);
2420 	} else if (!link_check) {
2421 		link.link_speed = 0;
2422 		link.link_duplex = ETH_LINK_HALF_DUPLEX;
2423 		link.link_status = ETH_LINK_DOWN;
2424 		link.link_autoneg = ETH_LINK_FIXED;
2425 	}
2426 
2427 	return rte_eth_linkstatus_set(dev, &link);
2428 }
2429 
2430 /*
2431  * igb_hw_control_acquire sets CTRL_EXT:DRV_LOAD bit.
2432  * For ASF and Pass Through versions of f/w this means
2433  * that the driver is loaded.
2434  */
2435 static void
2436 igb_hw_control_acquire(struct e1000_hw *hw)
2437 {
2438 	uint32_t ctrl_ext;
2439 
2440 	/* Let firmware know the driver has taken over */
2441 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
2442 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2443 }
2444 
2445 /*
2446  * igb_hw_control_release resets CTRL_EXT:DRV_LOAD bit.
2447  * For ASF and Pass Through versions of f/w this means that the
2448  * driver is no longer loaded.
2449  */
2450 static void
2451 igb_hw_control_release(struct e1000_hw *hw)
2452 {
2453 	uint32_t ctrl_ext;
2454 
2455 	/* Let firmware taken over control of h/w */
2456 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
2457 	E1000_WRITE_REG(hw, E1000_CTRL_EXT,
2458 			ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2459 }
2460 
2461 /*
2462  * Bit of a misnomer, what this really means is
2463  * to enable OS management of the system... aka
2464  * to disable special hardware management features.
2465  */
2466 static void
2467 igb_init_manageability(struct e1000_hw *hw)
2468 {
2469 	if (e1000_enable_mng_pass_thru(hw)) {
2470 		uint32_t manc2h = E1000_READ_REG(hw, E1000_MANC2H);
2471 		uint32_t manc = E1000_READ_REG(hw, E1000_MANC);
2472 
2473 		/* disable hardware interception of ARP */
2474 		manc &= ~(E1000_MANC_ARP_EN);
2475 
2476 		/* enable receiving management packets to the host */
2477 		manc |= E1000_MANC_EN_MNG2HOST;
2478 		manc2h |= 1 << 5;  /* Mng Port 623 */
2479 		manc2h |= 1 << 6;  /* Mng Port 664 */
2480 		E1000_WRITE_REG(hw, E1000_MANC2H, manc2h);
2481 		E1000_WRITE_REG(hw, E1000_MANC, manc);
2482 	}
2483 }
2484 
2485 static void
2486 igb_release_manageability(struct e1000_hw *hw)
2487 {
2488 	if (e1000_enable_mng_pass_thru(hw)) {
2489 		uint32_t manc = E1000_READ_REG(hw, E1000_MANC);
2490 
2491 		manc |= E1000_MANC_ARP_EN;
2492 		manc &= ~E1000_MANC_EN_MNG2HOST;
2493 
2494 		E1000_WRITE_REG(hw, E1000_MANC, manc);
2495 	}
2496 }
2497 
2498 static int
2499 eth_igb_promiscuous_enable(struct rte_eth_dev *dev)
2500 {
2501 	struct e1000_hw *hw =
2502 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2503 	uint32_t rctl;
2504 
2505 	rctl = E1000_READ_REG(hw, E1000_RCTL);
2506 	rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2507 	E1000_WRITE_REG(hw, E1000_RCTL, rctl);
2508 
2509 	return 0;
2510 }
2511 
2512 static int
2513 eth_igb_promiscuous_disable(struct rte_eth_dev *dev)
2514 {
2515 	struct e1000_hw *hw =
2516 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2517 	uint32_t rctl;
2518 
2519 	rctl = E1000_READ_REG(hw, E1000_RCTL);
2520 	rctl &= (~E1000_RCTL_UPE);
2521 	if (dev->data->all_multicast == 1)
2522 		rctl |= E1000_RCTL_MPE;
2523 	else
2524 		rctl &= (~E1000_RCTL_MPE);
2525 	E1000_WRITE_REG(hw, E1000_RCTL, rctl);
2526 
2527 	return 0;
2528 }
2529 
2530 static int
2531 eth_igb_allmulticast_enable(struct rte_eth_dev *dev)
2532 {
2533 	struct e1000_hw *hw =
2534 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2535 	uint32_t rctl;
2536 
2537 	rctl = E1000_READ_REG(hw, E1000_RCTL);
2538 	rctl |= E1000_RCTL_MPE;
2539 	E1000_WRITE_REG(hw, E1000_RCTL, rctl);
2540 
2541 	return 0;
2542 }
2543 
2544 static int
2545 eth_igb_allmulticast_disable(struct rte_eth_dev *dev)
2546 {
2547 	struct e1000_hw *hw =
2548 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2549 	uint32_t rctl;
2550 
2551 	if (dev->data->promiscuous == 1)
2552 		return 0; /* must remain in all_multicast mode */
2553 	rctl = E1000_READ_REG(hw, E1000_RCTL);
2554 	rctl &= (~E1000_RCTL_MPE);
2555 	E1000_WRITE_REG(hw, E1000_RCTL, rctl);
2556 
2557 	return 0;
2558 }
2559 
2560 static int
2561 eth_igb_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
2562 {
2563 	struct e1000_hw *hw =
2564 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2565 	struct e1000_vfta * shadow_vfta =
2566 		E1000_DEV_PRIVATE_TO_VFTA(dev->data->dev_private);
2567 	uint32_t vfta;
2568 	uint32_t vid_idx;
2569 	uint32_t vid_bit;
2570 
2571 	vid_idx = (uint32_t) ((vlan_id >> E1000_VFTA_ENTRY_SHIFT) &
2572 			      E1000_VFTA_ENTRY_MASK);
2573 	vid_bit = (uint32_t) (1 << (vlan_id & E1000_VFTA_ENTRY_BIT_SHIFT_MASK));
2574 	vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, vid_idx);
2575 	if (on)
2576 		vfta |= vid_bit;
2577 	else
2578 		vfta &= ~vid_bit;
2579 	E1000_WRITE_REG_ARRAY(hw, E1000_VFTA, vid_idx, vfta);
2580 
2581 	/* update local VFTA copy */
2582 	shadow_vfta->vfta[vid_idx] = vfta;
2583 
2584 	return 0;
2585 }
2586 
2587 static int
2588 eth_igb_vlan_tpid_set(struct rte_eth_dev *dev,
2589 		      enum rte_vlan_type vlan_type,
2590 		      uint16_t tpid)
2591 {
2592 	struct e1000_hw *hw =
2593 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2594 	uint32_t reg, qinq;
2595 
2596 	qinq = E1000_READ_REG(hw, E1000_CTRL_EXT);
2597 	qinq &= E1000_CTRL_EXT_EXT_VLAN;
2598 
2599 	/* only outer TPID of double VLAN can be configured*/
2600 	if (qinq && vlan_type == ETH_VLAN_TYPE_OUTER) {
2601 		reg = E1000_READ_REG(hw, E1000_VET);
2602 		reg = (reg & (~E1000_VET_VET_EXT)) |
2603 			((uint32_t)tpid << E1000_VET_VET_EXT_SHIFT);
2604 		E1000_WRITE_REG(hw, E1000_VET, reg);
2605 
2606 		return 0;
2607 	}
2608 
2609 	/* all other TPID values are read-only*/
2610 	PMD_DRV_LOG(ERR, "Not supported");
2611 
2612 	return -ENOTSUP;
2613 }
2614 
2615 static void
2616 igb_vlan_hw_filter_disable(struct rte_eth_dev *dev)
2617 {
2618 	struct e1000_hw *hw =
2619 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2620 	uint32_t reg;
2621 
2622 	/* Filter Table Disable */
2623 	reg = E1000_READ_REG(hw, E1000_RCTL);
2624 	reg &= ~E1000_RCTL_CFIEN;
2625 	reg &= ~E1000_RCTL_VFE;
2626 	E1000_WRITE_REG(hw, E1000_RCTL, reg);
2627 }
2628 
2629 static void
2630 igb_vlan_hw_filter_enable(struct rte_eth_dev *dev)
2631 {
2632 	struct e1000_hw *hw =
2633 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2634 	struct e1000_vfta * shadow_vfta =
2635 		E1000_DEV_PRIVATE_TO_VFTA(dev->data->dev_private);
2636 	uint32_t reg;
2637 	int i;
2638 
2639 	/* Filter Table Enable, CFI not used for packet acceptance */
2640 	reg = E1000_READ_REG(hw, E1000_RCTL);
2641 	reg &= ~E1000_RCTL_CFIEN;
2642 	reg |= E1000_RCTL_VFE;
2643 	E1000_WRITE_REG(hw, E1000_RCTL, reg);
2644 
2645 	/* restore VFTA table */
2646 	for (i = 0; i < IGB_VFTA_SIZE; i++)
2647 		E1000_WRITE_REG_ARRAY(hw, E1000_VFTA, i, shadow_vfta->vfta[i]);
2648 }
2649 
2650 static void
2651 igb_vlan_hw_strip_disable(struct rte_eth_dev *dev)
2652 {
2653 	struct e1000_hw *hw =
2654 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2655 	uint32_t reg;
2656 
2657 	/* VLAN Mode Disable */
2658 	reg = E1000_READ_REG(hw, E1000_CTRL);
2659 	reg &= ~E1000_CTRL_VME;
2660 	E1000_WRITE_REG(hw, E1000_CTRL, reg);
2661 }
2662 
2663 static void
2664 igb_vlan_hw_strip_enable(struct rte_eth_dev *dev)
2665 {
2666 	struct e1000_hw *hw =
2667 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2668 	uint32_t reg;
2669 
2670 	/* VLAN Mode Enable */
2671 	reg = E1000_READ_REG(hw, E1000_CTRL);
2672 	reg |= E1000_CTRL_VME;
2673 	E1000_WRITE_REG(hw, E1000_CTRL, reg);
2674 }
2675 
2676 static void
2677 igb_vlan_hw_extend_disable(struct rte_eth_dev *dev)
2678 {
2679 	struct e1000_hw *hw =
2680 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2681 	uint32_t reg;
2682 
2683 	/* CTRL_EXT: Extended VLAN */
2684 	reg = E1000_READ_REG(hw, E1000_CTRL_EXT);
2685 	reg &= ~E1000_CTRL_EXT_EXTEND_VLAN;
2686 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, reg);
2687 
2688 	/* Update maximum packet length */
2689 	if (dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_JUMBO_FRAME)
2690 		E1000_WRITE_REG(hw, E1000_RLPML,
2691 				dev->data->dev_conf.rxmode.max_rx_pkt_len);
2692 }
2693 
2694 static void
2695 igb_vlan_hw_extend_enable(struct rte_eth_dev *dev)
2696 {
2697 	struct e1000_hw *hw =
2698 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2699 	uint32_t reg;
2700 
2701 	/* CTRL_EXT: Extended VLAN */
2702 	reg = E1000_READ_REG(hw, E1000_CTRL_EXT);
2703 	reg |= E1000_CTRL_EXT_EXTEND_VLAN;
2704 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, reg);
2705 
2706 	/* Update maximum packet length */
2707 	if (dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_JUMBO_FRAME)
2708 		E1000_WRITE_REG(hw, E1000_RLPML,
2709 			dev->data->dev_conf.rxmode.max_rx_pkt_len +
2710 						VLAN_TAG_SIZE);
2711 }
2712 
2713 static int
2714 eth_igb_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2715 {
2716 	struct rte_eth_rxmode *rxmode;
2717 
2718 	rxmode = &dev->data->dev_conf.rxmode;
2719 	if(mask & ETH_VLAN_STRIP_MASK){
2720 		if (rxmode->offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2721 			igb_vlan_hw_strip_enable(dev);
2722 		else
2723 			igb_vlan_hw_strip_disable(dev);
2724 	}
2725 
2726 	if(mask & ETH_VLAN_FILTER_MASK){
2727 		if (rxmode->offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
2728 			igb_vlan_hw_filter_enable(dev);
2729 		else
2730 			igb_vlan_hw_filter_disable(dev);
2731 	}
2732 
2733 	if(mask & ETH_VLAN_EXTEND_MASK){
2734 		if (rxmode->offloads & DEV_RX_OFFLOAD_VLAN_EXTEND)
2735 			igb_vlan_hw_extend_enable(dev);
2736 		else
2737 			igb_vlan_hw_extend_disable(dev);
2738 	}
2739 
2740 	return 0;
2741 }
2742 
2743 
2744 /**
2745  * It enables the interrupt mask and then enable the interrupt.
2746  *
2747  * @param dev
2748  *  Pointer to struct rte_eth_dev.
2749  * @param on
2750  *  Enable or Disable
2751  *
2752  * @return
2753  *  - On success, zero.
2754  *  - On failure, a negative value.
2755  */
2756 static int
2757 eth_igb_lsc_interrupt_setup(struct rte_eth_dev *dev, uint8_t on)
2758 {
2759 	struct e1000_interrupt *intr =
2760 		E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
2761 
2762 	if (on)
2763 		intr->mask |= E1000_ICR_LSC;
2764 	else
2765 		intr->mask &= ~E1000_ICR_LSC;
2766 
2767 	return 0;
2768 }
2769 
2770 /* It clears the interrupt causes and enables the interrupt.
2771  * It will be called once only during nic initialized.
2772  *
2773  * @param dev
2774  *  Pointer to struct rte_eth_dev.
2775  *
2776  * @return
2777  *  - On success, zero.
2778  *  - On failure, a negative value.
2779  */
2780 static int eth_igb_rxq_interrupt_setup(struct rte_eth_dev *dev)
2781 {
2782 	uint32_t mask, regval;
2783 	int ret;
2784 	struct e1000_hw *hw =
2785 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2786 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2787 	struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
2788 	int misc_shift = rte_intr_allow_others(intr_handle) ? 1 : 0;
2789 	struct rte_eth_dev_info dev_info;
2790 
2791 	memset(&dev_info, 0, sizeof(dev_info));
2792 	ret = eth_igb_infos_get(dev, &dev_info);
2793 	if (ret != 0)
2794 		return ret;
2795 
2796 	mask = (0xFFFFFFFF >> (32 - dev_info.max_rx_queues)) << misc_shift;
2797 	regval = E1000_READ_REG(hw, E1000_EIMS);
2798 	E1000_WRITE_REG(hw, E1000_EIMS, regval | mask);
2799 
2800 	return 0;
2801 }
2802 
2803 /*
2804  * It reads ICR and gets interrupt causes, check it and set a bit flag
2805  * to update link status.
2806  *
2807  * @param dev
2808  *  Pointer to struct rte_eth_dev.
2809  *
2810  * @return
2811  *  - On success, zero.
2812  *  - On failure, a negative value.
2813  */
2814 static int
2815 eth_igb_interrupt_get_status(struct rte_eth_dev *dev)
2816 {
2817 	uint32_t icr;
2818 	struct e1000_hw *hw =
2819 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2820 	struct e1000_interrupt *intr =
2821 		E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
2822 
2823 	igb_intr_disable(dev);
2824 
2825 	/* read-on-clear nic registers here */
2826 	icr = E1000_READ_REG(hw, E1000_ICR);
2827 
2828 	intr->flags = 0;
2829 	if (icr & E1000_ICR_LSC) {
2830 		intr->flags |= E1000_FLAG_NEED_LINK_UPDATE;
2831 	}
2832 
2833 	if (icr & E1000_ICR_VMMB)
2834 		intr->flags |= E1000_FLAG_MAILBOX;
2835 
2836 	return 0;
2837 }
2838 
2839 /*
2840  * It executes link_update after knowing an interrupt is prsent.
2841  *
2842  * @param dev
2843  *  Pointer to struct rte_eth_dev.
2844  *
2845  * @return
2846  *  - On success, zero.
2847  *  - On failure, a negative value.
2848  */
2849 static int
2850 eth_igb_interrupt_action(struct rte_eth_dev *dev,
2851 			 struct rte_intr_handle *intr_handle)
2852 {
2853 	struct e1000_hw *hw =
2854 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2855 	struct e1000_interrupt *intr =
2856 		E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
2857 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2858 	struct rte_eth_link link;
2859 	int ret;
2860 
2861 	if (intr->flags & E1000_FLAG_MAILBOX) {
2862 		igb_pf_mbx_process(dev);
2863 		intr->flags &= ~E1000_FLAG_MAILBOX;
2864 	}
2865 
2866 	igb_intr_enable(dev);
2867 	rte_intr_ack(intr_handle);
2868 
2869 	if (intr->flags & E1000_FLAG_NEED_LINK_UPDATE) {
2870 		intr->flags &= ~E1000_FLAG_NEED_LINK_UPDATE;
2871 
2872 		/* set get_link_status to check register later */
2873 		hw->mac.get_link_status = 1;
2874 		ret = eth_igb_link_update(dev, 0);
2875 
2876 		/* check if link has changed */
2877 		if (ret < 0)
2878 			return 0;
2879 
2880 		rte_eth_linkstatus_get(dev, &link);
2881 		if (link.link_status) {
2882 			PMD_INIT_LOG(INFO,
2883 				     " Port %d: Link Up - speed %u Mbps - %s",
2884 				     dev->data->port_id,
2885 				     (unsigned)link.link_speed,
2886 				     link.link_duplex == ETH_LINK_FULL_DUPLEX ?
2887 				     "full-duplex" : "half-duplex");
2888 		} else {
2889 			PMD_INIT_LOG(INFO, " Port %d: Link Down",
2890 				     dev->data->port_id);
2891 		}
2892 
2893 		PMD_INIT_LOG(DEBUG, "PCI Address: " PCI_PRI_FMT,
2894 			     pci_dev->addr.domain,
2895 			     pci_dev->addr.bus,
2896 			     pci_dev->addr.devid,
2897 			     pci_dev->addr.function);
2898 		rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC, NULL);
2899 	}
2900 
2901 	return 0;
2902 }
2903 
2904 /**
2905  * Interrupt handler which shall be registered at first.
2906  *
2907  * @param handle
2908  *  Pointer to interrupt handle.
2909  * @param param
2910  *  The address of parameter (struct rte_eth_dev *) regsitered before.
2911  *
2912  * @return
2913  *  void
2914  */
2915 static void
2916 eth_igb_interrupt_handler(void *param)
2917 {
2918 	struct rte_eth_dev *dev = (struct rte_eth_dev *)param;
2919 
2920 	eth_igb_interrupt_get_status(dev);
2921 	eth_igb_interrupt_action(dev, dev->intr_handle);
2922 }
2923 
2924 static int
2925 eth_igbvf_interrupt_get_status(struct rte_eth_dev *dev)
2926 {
2927 	uint32_t eicr;
2928 	struct e1000_hw *hw =
2929 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2930 	struct e1000_interrupt *intr =
2931 		E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
2932 
2933 	igbvf_intr_disable(hw);
2934 
2935 	/* read-on-clear nic registers here */
2936 	eicr = E1000_READ_REG(hw, E1000_EICR);
2937 	intr->flags = 0;
2938 
2939 	if (eicr == E1000_VTIVAR_MISC_MAILBOX)
2940 		intr->flags |= E1000_FLAG_MAILBOX;
2941 
2942 	return 0;
2943 }
2944 
2945 void igbvf_mbx_process(struct rte_eth_dev *dev)
2946 {
2947 	struct e1000_hw *hw =
2948 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2949 	struct e1000_mbx_info *mbx = &hw->mbx;
2950 	u32 in_msg = 0;
2951 
2952 	/* peek the message first */
2953 	in_msg = E1000_READ_REG(hw, E1000_VMBMEM(0));
2954 
2955 	/* PF reset VF event */
2956 	if (in_msg == E1000_PF_CONTROL_MSG) {
2957 		/* dummy mbx read to ack pf */
2958 		if (mbx->ops.read(hw, &in_msg, 1, 0))
2959 			return;
2960 		rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_RESET,
2961 					     NULL);
2962 	}
2963 }
2964 
2965 static int
2966 eth_igbvf_interrupt_action(struct rte_eth_dev *dev, struct rte_intr_handle *intr_handle)
2967 {
2968 	struct e1000_interrupt *intr =
2969 		E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
2970 
2971 	if (intr->flags & E1000_FLAG_MAILBOX) {
2972 		igbvf_mbx_process(dev);
2973 		intr->flags &= ~E1000_FLAG_MAILBOX;
2974 	}
2975 
2976 	igbvf_intr_enable(dev);
2977 	rte_intr_ack(intr_handle);
2978 
2979 	return 0;
2980 }
2981 
2982 static void
2983 eth_igbvf_interrupt_handler(void *param)
2984 {
2985 	struct rte_eth_dev *dev = (struct rte_eth_dev *)param;
2986 
2987 	eth_igbvf_interrupt_get_status(dev);
2988 	eth_igbvf_interrupt_action(dev, dev->intr_handle);
2989 }
2990 
2991 static int
2992 eth_igb_led_on(struct rte_eth_dev *dev)
2993 {
2994 	struct e1000_hw *hw;
2995 
2996 	hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2997 	return e1000_led_on(hw) == E1000_SUCCESS ? 0 : -ENOTSUP;
2998 }
2999 
3000 static int
3001 eth_igb_led_off(struct rte_eth_dev *dev)
3002 {
3003 	struct e1000_hw *hw;
3004 
3005 	hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3006 	return e1000_led_off(hw) == E1000_SUCCESS ? 0 : -ENOTSUP;
3007 }
3008 
3009 static int
3010 eth_igb_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
3011 {
3012 	struct e1000_hw *hw;
3013 	uint32_t ctrl;
3014 	int tx_pause;
3015 	int rx_pause;
3016 
3017 	hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3018 	fc_conf->pause_time = hw->fc.pause_time;
3019 	fc_conf->high_water = hw->fc.high_water;
3020 	fc_conf->low_water = hw->fc.low_water;
3021 	fc_conf->send_xon = hw->fc.send_xon;
3022 	fc_conf->autoneg = hw->mac.autoneg;
3023 
3024 	/*
3025 	 * Return rx_pause and tx_pause status according to actual setting of
3026 	 * the TFCE and RFCE bits in the CTRL register.
3027 	 */
3028 	ctrl = E1000_READ_REG(hw, E1000_CTRL);
3029 	if (ctrl & E1000_CTRL_TFCE)
3030 		tx_pause = 1;
3031 	else
3032 		tx_pause = 0;
3033 
3034 	if (ctrl & E1000_CTRL_RFCE)
3035 		rx_pause = 1;
3036 	else
3037 		rx_pause = 0;
3038 
3039 	if (rx_pause && tx_pause)
3040 		fc_conf->mode = RTE_FC_FULL;
3041 	else if (rx_pause)
3042 		fc_conf->mode = RTE_FC_RX_PAUSE;
3043 	else if (tx_pause)
3044 		fc_conf->mode = RTE_FC_TX_PAUSE;
3045 	else
3046 		fc_conf->mode = RTE_FC_NONE;
3047 
3048 	return 0;
3049 }
3050 
3051 static int
3052 eth_igb_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
3053 {
3054 	struct e1000_hw *hw;
3055 	int err;
3056 	enum e1000_fc_mode rte_fcmode_2_e1000_fcmode[] = {
3057 		e1000_fc_none,
3058 		e1000_fc_rx_pause,
3059 		e1000_fc_tx_pause,
3060 		e1000_fc_full
3061 	};
3062 	uint32_t rx_buf_size;
3063 	uint32_t max_high_water;
3064 	uint32_t rctl;
3065 	uint32_t ctrl;
3066 
3067 	hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3068 	if (fc_conf->autoneg != hw->mac.autoneg)
3069 		return -ENOTSUP;
3070 	rx_buf_size = igb_get_rx_buffer_size(hw);
3071 	PMD_INIT_LOG(DEBUG, "Rx packet buffer size = 0x%x", rx_buf_size);
3072 
3073 	/* At least reserve one Ethernet frame for watermark */
3074 	max_high_water = rx_buf_size - RTE_ETHER_MAX_LEN;
3075 	if ((fc_conf->high_water > max_high_water) ||
3076 	    (fc_conf->high_water < fc_conf->low_water)) {
3077 		PMD_INIT_LOG(ERR, "e1000 incorrect high/low water value");
3078 		PMD_INIT_LOG(ERR, "high water must <=  0x%x", max_high_water);
3079 		return -EINVAL;
3080 	}
3081 
3082 	hw->fc.requested_mode = rte_fcmode_2_e1000_fcmode[fc_conf->mode];
3083 	hw->fc.pause_time     = fc_conf->pause_time;
3084 	hw->fc.high_water     = fc_conf->high_water;
3085 	hw->fc.low_water      = fc_conf->low_water;
3086 	hw->fc.send_xon	      = fc_conf->send_xon;
3087 
3088 	err = e1000_setup_link_generic(hw);
3089 	if (err == E1000_SUCCESS) {
3090 
3091 		/* check if we want to forward MAC frames - driver doesn't have native
3092 		 * capability to do that, so we'll write the registers ourselves */
3093 
3094 		rctl = E1000_READ_REG(hw, E1000_RCTL);
3095 
3096 		/* set or clear MFLCN.PMCF bit depending on configuration */
3097 		if (fc_conf->mac_ctrl_frame_fwd != 0)
3098 			rctl |= E1000_RCTL_PMCF;
3099 		else
3100 			rctl &= ~E1000_RCTL_PMCF;
3101 
3102 		E1000_WRITE_REG(hw, E1000_RCTL, rctl);
3103 
3104 		/*
3105 		 * check if we want to change flow control mode - driver doesn't have native
3106 		 * capability to do that, so we'll write the registers ourselves
3107 		 */
3108 		ctrl = E1000_READ_REG(hw, E1000_CTRL);
3109 
3110 		/*
3111 		 * set or clear E1000_CTRL_RFCE and E1000_CTRL_TFCE bits depending
3112 		 * on configuration
3113 		 */
3114 		switch (fc_conf->mode) {
3115 		case RTE_FC_NONE:
3116 			ctrl &= ~E1000_CTRL_RFCE & ~E1000_CTRL_TFCE;
3117 			break;
3118 		case RTE_FC_RX_PAUSE:
3119 			ctrl |= E1000_CTRL_RFCE;
3120 			ctrl &= ~E1000_CTRL_TFCE;
3121 			break;
3122 		case RTE_FC_TX_PAUSE:
3123 			ctrl |= E1000_CTRL_TFCE;
3124 			ctrl &= ~E1000_CTRL_RFCE;
3125 			break;
3126 		case RTE_FC_FULL:
3127 			ctrl |= E1000_CTRL_RFCE | E1000_CTRL_TFCE;
3128 			break;
3129 		default:
3130 			PMD_INIT_LOG(ERR, "invalid flow control mode");
3131 			return -EINVAL;
3132 		}
3133 
3134 		E1000_WRITE_REG(hw, E1000_CTRL, ctrl);
3135 
3136 		E1000_WRITE_FLUSH(hw);
3137 
3138 		return 0;
3139 	}
3140 
3141 	PMD_INIT_LOG(ERR, "e1000_setup_link_generic = 0x%x", err);
3142 	return -EIO;
3143 }
3144 
3145 #define E1000_RAH_POOLSEL_SHIFT      (18)
3146 static int
3147 eth_igb_rar_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
3148 		uint32_t index, uint32_t pool)
3149 {
3150 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3151 	uint32_t rah;
3152 
3153 	e1000_rar_set(hw, mac_addr->addr_bytes, index);
3154 	rah = E1000_READ_REG(hw, E1000_RAH(index));
3155 	rah |= (0x1 << (E1000_RAH_POOLSEL_SHIFT + pool));
3156 	E1000_WRITE_REG(hw, E1000_RAH(index), rah);
3157 	return 0;
3158 }
3159 
3160 static void
3161 eth_igb_rar_clear(struct rte_eth_dev *dev, uint32_t index)
3162 {
3163 	uint8_t addr[RTE_ETHER_ADDR_LEN];
3164 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3165 
3166 	memset(addr, 0, sizeof(addr));
3167 
3168 	e1000_rar_set(hw, addr, index);
3169 }
3170 
3171 static int
3172 eth_igb_default_mac_addr_set(struct rte_eth_dev *dev,
3173 				struct rte_ether_addr *addr)
3174 {
3175 	eth_igb_rar_clear(dev, 0);
3176 	eth_igb_rar_set(dev, (void *)addr, 0, 0);
3177 
3178 	return 0;
3179 }
3180 /*
3181  * Virtual Function operations
3182  */
3183 static void
3184 igbvf_intr_disable(struct e1000_hw *hw)
3185 {
3186 	PMD_INIT_FUNC_TRACE();
3187 
3188 	/* Clear interrupt mask to stop from interrupts being generated */
3189 	E1000_WRITE_REG(hw, E1000_EIMC, 0xFFFF);
3190 
3191 	E1000_WRITE_FLUSH(hw);
3192 }
3193 
3194 static void
3195 igbvf_stop_adapter(struct rte_eth_dev *dev)
3196 {
3197 	u32 reg_val;
3198 	u16 i;
3199 	struct rte_eth_dev_info dev_info;
3200 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3201 	int ret;
3202 
3203 	memset(&dev_info, 0, sizeof(dev_info));
3204 	ret = eth_igbvf_infos_get(dev, &dev_info);
3205 	if (ret != 0)
3206 		return;
3207 
3208 	/* Clear interrupt mask to stop from interrupts being generated */
3209 	igbvf_intr_disable(hw);
3210 
3211 	/* Clear any pending interrupts, flush previous writes */
3212 	E1000_READ_REG(hw, E1000_EICR);
3213 
3214 	/* Disable the transmit unit.  Each queue must be disabled. */
3215 	for (i = 0; i < dev_info.max_tx_queues; i++)
3216 		E1000_WRITE_REG(hw, E1000_TXDCTL(i), E1000_TXDCTL_SWFLSH);
3217 
3218 	/* Disable the receive unit by stopping each queue */
3219 	for (i = 0; i < dev_info.max_rx_queues; i++) {
3220 		reg_val = E1000_READ_REG(hw, E1000_RXDCTL(i));
3221 		reg_val &= ~E1000_RXDCTL_QUEUE_ENABLE;
3222 		E1000_WRITE_REG(hw, E1000_RXDCTL(i), reg_val);
3223 		while (E1000_READ_REG(hw, E1000_RXDCTL(i)) & E1000_RXDCTL_QUEUE_ENABLE)
3224 			;
3225 	}
3226 
3227 	/* flush all queues disables */
3228 	E1000_WRITE_FLUSH(hw);
3229 	msec_delay(2);
3230 }
3231 
3232 static int eth_igbvf_link_update(struct e1000_hw *hw)
3233 {
3234 	struct e1000_mbx_info *mbx = &hw->mbx;
3235 	struct e1000_mac_info *mac = &hw->mac;
3236 	int ret_val = E1000_SUCCESS;
3237 
3238 	PMD_INIT_LOG(DEBUG, "e1000_check_for_link_vf");
3239 
3240 	/*
3241 	 * We only want to run this if there has been a rst asserted.
3242 	 * in this case that could mean a link change, device reset,
3243 	 * or a virtual function reset
3244 	 */
3245 
3246 	/* If we were hit with a reset or timeout drop the link */
3247 	if (!e1000_check_for_rst(hw, 0) || !mbx->timeout)
3248 		mac->get_link_status = TRUE;
3249 
3250 	if (!mac->get_link_status)
3251 		goto out;
3252 
3253 	/* if link status is down no point in checking to see if pf is up */
3254 	if (!(E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU))
3255 		goto out;
3256 
3257 	/* if we passed all the tests above then the link is up and we no
3258 	 * longer need to check for link */
3259 	mac->get_link_status = FALSE;
3260 
3261 out:
3262 	return ret_val;
3263 }
3264 
3265 
3266 static int
3267 igbvf_dev_configure(struct rte_eth_dev *dev)
3268 {
3269 	struct rte_eth_conf* conf = &dev->data->dev_conf;
3270 
3271 	PMD_INIT_LOG(DEBUG, "Configured Virtual Function port id: %d",
3272 		     dev->data->port_id);
3273 
3274 	if (dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG)
3275 		dev->data->dev_conf.rxmode.offloads |= DEV_RX_OFFLOAD_RSS_HASH;
3276 
3277 	/*
3278 	 * VF has no ability to enable/disable HW CRC
3279 	 * Keep the persistent behavior the same as Host PF
3280 	 */
3281 #ifndef RTE_LIBRTE_E1000_PF_DISABLE_STRIP_CRC
3282 	if (conf->rxmode.offloads & DEV_RX_OFFLOAD_KEEP_CRC) {
3283 		PMD_INIT_LOG(NOTICE, "VF can't disable HW CRC Strip");
3284 		conf->rxmode.offloads &= ~DEV_RX_OFFLOAD_KEEP_CRC;
3285 	}
3286 #else
3287 	if (!(conf->rxmode.offloads & DEV_RX_OFFLOAD_KEEP_CRC)) {
3288 		PMD_INIT_LOG(NOTICE, "VF can't enable HW CRC Strip");
3289 		conf->rxmode.offloads |= DEV_RX_OFFLOAD_KEEP_CRC;
3290 	}
3291 #endif
3292 
3293 	return 0;
3294 }
3295 
3296 static int
3297 igbvf_dev_start(struct rte_eth_dev *dev)
3298 {
3299 	struct e1000_hw *hw =
3300 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3301 	struct e1000_adapter *adapter =
3302 		E1000_DEV_PRIVATE(dev->data->dev_private);
3303 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
3304 	struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
3305 	int ret;
3306 	uint32_t intr_vector = 0;
3307 
3308 	PMD_INIT_FUNC_TRACE();
3309 
3310 	hw->mac.ops.reset_hw(hw);
3311 	adapter->stopped = 0;
3312 
3313 	/* Set all vfta */
3314 	igbvf_set_vfta_all(dev,1);
3315 
3316 	eth_igbvf_tx_init(dev);
3317 
3318 	/* This can fail when allocating mbufs for descriptor rings */
3319 	ret = eth_igbvf_rx_init(dev);
3320 	if (ret) {
3321 		PMD_INIT_LOG(ERR, "Unable to initialize RX hardware");
3322 		igb_dev_clear_queues(dev);
3323 		return ret;
3324 	}
3325 
3326 	/* check and configure queue intr-vector mapping */
3327 	if (rte_intr_cap_multiple(intr_handle) &&
3328 	    dev->data->dev_conf.intr_conf.rxq) {
3329 		intr_vector = dev->data->nb_rx_queues;
3330 		ret = rte_intr_efd_enable(intr_handle, intr_vector);
3331 		if (ret)
3332 			return ret;
3333 	}
3334 
3335 	if (rte_intr_dp_is_en(intr_handle) && !intr_handle->intr_vec) {
3336 		intr_handle->intr_vec =
3337 			rte_zmalloc("intr_vec",
3338 				    dev->data->nb_rx_queues * sizeof(int), 0);
3339 		if (!intr_handle->intr_vec) {
3340 			PMD_INIT_LOG(ERR, "Failed to allocate %d rx_queues"
3341 				     " intr_vec", dev->data->nb_rx_queues);
3342 			return -ENOMEM;
3343 		}
3344 	}
3345 
3346 	eth_igbvf_configure_msix_intr(dev);
3347 
3348 	/* enable uio/vfio intr/eventfd mapping */
3349 	rte_intr_enable(intr_handle);
3350 
3351 	/* resume enabled intr since hw reset */
3352 	igbvf_intr_enable(dev);
3353 
3354 	return 0;
3355 }
3356 
3357 static int
3358 igbvf_dev_stop(struct rte_eth_dev *dev)
3359 {
3360 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
3361 	struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
3362 	struct e1000_adapter *adapter =
3363 		E1000_DEV_PRIVATE(dev->data->dev_private);
3364 
3365 	if (adapter->stopped)
3366 		return 0;
3367 
3368 	PMD_INIT_FUNC_TRACE();
3369 
3370 	igbvf_stop_adapter(dev);
3371 
3372 	/*
3373 	  * Clear what we set, but we still keep shadow_vfta to
3374 	  * restore after device starts
3375 	  */
3376 	igbvf_set_vfta_all(dev,0);
3377 
3378 	igb_dev_clear_queues(dev);
3379 
3380 	/* disable intr eventfd mapping */
3381 	rte_intr_disable(intr_handle);
3382 
3383 	/* Clean datapath event and queue/vec mapping */
3384 	rte_intr_efd_disable(intr_handle);
3385 	if (intr_handle->intr_vec) {
3386 		rte_free(intr_handle->intr_vec);
3387 		intr_handle->intr_vec = NULL;
3388 	}
3389 
3390 	adapter->stopped = true;
3391 	dev->data->dev_started = 0;
3392 
3393 	return 0;
3394 }
3395 
3396 static int
3397 igbvf_dev_close(struct rte_eth_dev *dev)
3398 {
3399 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3400 	struct rte_ether_addr addr;
3401 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
3402 	int ret;
3403 
3404 	PMD_INIT_FUNC_TRACE();
3405 
3406 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
3407 		return 0;
3408 
3409 	e1000_reset_hw(hw);
3410 
3411 	ret = igbvf_dev_stop(dev);
3412 	if (ret != 0)
3413 		return ret;
3414 
3415 	igb_dev_free_queues(dev);
3416 
3417 	/**
3418 	 * reprogram the RAR with a zero mac address,
3419 	 * to ensure that the VF traffic goes to the PF
3420 	 * after stop, close and detach of the VF.
3421 	 **/
3422 
3423 	memset(&addr, 0, sizeof(addr));
3424 	igbvf_default_mac_addr_set(dev, &addr);
3425 
3426 	rte_intr_callback_unregister(&pci_dev->intr_handle,
3427 				     eth_igbvf_interrupt_handler,
3428 				     (void *)dev);
3429 
3430 	return 0;
3431 }
3432 
3433 static int
3434 igbvf_promiscuous_enable(struct rte_eth_dev *dev)
3435 {
3436 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3437 
3438 	/* Set both unicast and multicast promisc */
3439 	e1000_promisc_set_vf(hw, e1000_promisc_enabled);
3440 
3441 	return 0;
3442 }
3443 
3444 static int
3445 igbvf_promiscuous_disable(struct rte_eth_dev *dev)
3446 {
3447 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3448 
3449 	/* If in allmulticast mode leave multicast promisc */
3450 	if (dev->data->all_multicast == 1)
3451 		e1000_promisc_set_vf(hw, e1000_promisc_multicast);
3452 	else
3453 		e1000_promisc_set_vf(hw, e1000_promisc_disabled);
3454 
3455 	return 0;
3456 }
3457 
3458 static int
3459 igbvf_allmulticast_enable(struct rte_eth_dev *dev)
3460 {
3461 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3462 
3463 	/* In promiscuous mode multicast promisc already set */
3464 	if (dev->data->promiscuous == 0)
3465 		e1000_promisc_set_vf(hw, e1000_promisc_multicast);
3466 
3467 	return 0;
3468 }
3469 
3470 static int
3471 igbvf_allmulticast_disable(struct rte_eth_dev *dev)
3472 {
3473 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3474 
3475 	/* In promiscuous mode leave multicast promisc enabled */
3476 	if (dev->data->promiscuous == 0)
3477 		e1000_promisc_set_vf(hw, e1000_promisc_disabled);
3478 
3479 	return 0;
3480 }
3481 
3482 static int igbvf_set_vfta(struct e1000_hw *hw, uint16_t vid, bool on)
3483 {
3484 	struct e1000_mbx_info *mbx = &hw->mbx;
3485 	uint32_t msgbuf[2];
3486 	s32 err;
3487 
3488 	/* After set vlan, vlan strip will also be enabled in igb driver*/
3489 	msgbuf[0] = E1000_VF_SET_VLAN;
3490 	msgbuf[1] = vid;
3491 	/* Setting the 8 bit field MSG INFO to TRUE indicates "add" */
3492 	if (on)
3493 		msgbuf[0] |= E1000_VF_SET_VLAN_ADD;
3494 
3495 	err = mbx->ops.write_posted(hw, msgbuf, 2, 0);
3496 	if (err)
3497 		goto mbx_err;
3498 
3499 	err = mbx->ops.read_posted(hw, msgbuf, 2, 0);
3500 	if (err)
3501 		goto mbx_err;
3502 
3503 	msgbuf[0] &= ~E1000_VT_MSGTYPE_CTS;
3504 	if (msgbuf[0] == (E1000_VF_SET_VLAN | E1000_VT_MSGTYPE_NACK))
3505 		err = -EINVAL;
3506 
3507 mbx_err:
3508 	return err;
3509 }
3510 
3511 static void igbvf_set_vfta_all(struct rte_eth_dev *dev, bool on)
3512 {
3513 	struct e1000_hw *hw =
3514 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3515 	struct e1000_vfta * shadow_vfta =
3516 		E1000_DEV_PRIVATE_TO_VFTA(dev->data->dev_private);
3517 	int i = 0, j = 0, vfta = 0, mask = 1;
3518 
3519 	for (i = 0; i < IGB_VFTA_SIZE; i++){
3520 		vfta = shadow_vfta->vfta[i];
3521 		if(vfta){
3522 			mask = 1;
3523 			for (j = 0; j < 32; j++){
3524 				if(vfta & mask)
3525 					igbvf_set_vfta(hw,
3526 						(uint16_t)((i<<5)+j), on);
3527 				mask<<=1;
3528 			}
3529 		}
3530 	}
3531 
3532 }
3533 
3534 static int
3535 igbvf_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
3536 {
3537 	struct e1000_hw *hw =
3538 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3539 	struct e1000_vfta * shadow_vfta =
3540 		E1000_DEV_PRIVATE_TO_VFTA(dev->data->dev_private);
3541 	uint32_t vid_idx = 0;
3542 	uint32_t vid_bit = 0;
3543 	int ret = 0;
3544 
3545 	PMD_INIT_FUNC_TRACE();
3546 
3547 	/*vind is not used in VF driver, set to 0, check ixgbe_set_vfta_vf*/
3548 	ret = igbvf_set_vfta(hw, vlan_id, !!on);
3549 	if(ret){
3550 		PMD_INIT_LOG(ERR, "Unable to set VF vlan");
3551 		return ret;
3552 	}
3553 	vid_idx = (uint32_t) ((vlan_id >> 5) & 0x7F);
3554 	vid_bit = (uint32_t) (1 << (vlan_id & 0x1F));
3555 
3556 	/*Save what we set and retore it after device reset*/
3557 	if (on)
3558 		shadow_vfta->vfta[vid_idx] |= vid_bit;
3559 	else
3560 		shadow_vfta->vfta[vid_idx] &= ~vid_bit;
3561 
3562 	return 0;
3563 }
3564 
3565 static int
3566 igbvf_default_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *addr)
3567 {
3568 	struct e1000_hw *hw =
3569 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3570 
3571 	/* index is not used by rar_set() */
3572 	hw->mac.ops.rar_set(hw, (void *)addr, 0);
3573 	return 0;
3574 }
3575 
3576 
3577 static int
3578 eth_igb_rss_reta_update(struct rte_eth_dev *dev,
3579 			struct rte_eth_rss_reta_entry64 *reta_conf,
3580 			uint16_t reta_size)
3581 {
3582 	uint8_t i, j, mask;
3583 	uint32_t reta, r;
3584 	uint16_t idx, shift;
3585 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3586 
3587 	if (reta_size != ETH_RSS_RETA_SIZE_128) {
3588 		PMD_DRV_LOG(ERR, "The size of hash lookup table configured "
3589 			"(%d) doesn't match the number hardware can supported "
3590 			"(%d)", reta_size, ETH_RSS_RETA_SIZE_128);
3591 		return -EINVAL;
3592 	}
3593 
3594 	for (i = 0; i < reta_size; i += IGB_4_BIT_WIDTH) {
3595 		idx = i / RTE_RETA_GROUP_SIZE;
3596 		shift = i % RTE_RETA_GROUP_SIZE;
3597 		mask = (uint8_t)((reta_conf[idx].mask >> shift) &
3598 						IGB_4_BIT_MASK);
3599 		if (!mask)
3600 			continue;
3601 		if (mask == IGB_4_BIT_MASK)
3602 			r = 0;
3603 		else
3604 			r = E1000_READ_REG(hw, E1000_RETA(i >> 2));
3605 		for (j = 0, reta = 0; j < IGB_4_BIT_WIDTH; j++) {
3606 			if (mask & (0x1 << j))
3607 				reta |= reta_conf[idx].reta[shift + j] <<
3608 							(CHAR_BIT * j);
3609 			else
3610 				reta |= r & (IGB_8_BIT_MASK << (CHAR_BIT * j));
3611 		}
3612 		E1000_WRITE_REG(hw, E1000_RETA(i >> 2), reta);
3613 	}
3614 
3615 	return 0;
3616 }
3617 
3618 static int
3619 eth_igb_rss_reta_query(struct rte_eth_dev *dev,
3620 		       struct rte_eth_rss_reta_entry64 *reta_conf,
3621 		       uint16_t reta_size)
3622 {
3623 	uint8_t i, j, mask;
3624 	uint32_t reta;
3625 	uint16_t idx, shift;
3626 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3627 
3628 	if (reta_size != ETH_RSS_RETA_SIZE_128) {
3629 		PMD_DRV_LOG(ERR, "The size of hash lookup table configured "
3630 			"(%d) doesn't match the number hardware can supported "
3631 			"(%d)", reta_size, ETH_RSS_RETA_SIZE_128);
3632 		return -EINVAL;
3633 	}
3634 
3635 	for (i = 0; i < reta_size; i += IGB_4_BIT_WIDTH) {
3636 		idx = i / RTE_RETA_GROUP_SIZE;
3637 		shift = i % RTE_RETA_GROUP_SIZE;
3638 		mask = (uint8_t)((reta_conf[idx].mask >> shift) &
3639 						IGB_4_BIT_MASK);
3640 		if (!mask)
3641 			continue;
3642 		reta = E1000_READ_REG(hw, E1000_RETA(i >> 2));
3643 		for (j = 0; j < IGB_4_BIT_WIDTH; j++) {
3644 			if (mask & (0x1 << j))
3645 				reta_conf[idx].reta[shift + j] =
3646 					((reta >> (CHAR_BIT * j)) &
3647 						IGB_8_BIT_MASK);
3648 		}
3649 	}
3650 
3651 	return 0;
3652 }
3653 
3654 int
3655 eth_igb_syn_filter_set(struct rte_eth_dev *dev,
3656 			struct rte_eth_syn_filter *filter,
3657 			bool add)
3658 {
3659 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3660 	struct e1000_filter_info *filter_info =
3661 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
3662 	uint32_t synqf, rfctl;
3663 
3664 	if (filter->queue >= IGB_MAX_RX_QUEUE_NUM)
3665 		return -EINVAL;
3666 
3667 	synqf = E1000_READ_REG(hw, E1000_SYNQF(0));
3668 
3669 	if (add) {
3670 		if (synqf & E1000_SYN_FILTER_ENABLE)
3671 			return -EINVAL;
3672 
3673 		synqf = (uint32_t)(((filter->queue << E1000_SYN_FILTER_QUEUE_SHIFT) &
3674 			E1000_SYN_FILTER_QUEUE) | E1000_SYN_FILTER_ENABLE);
3675 
3676 		rfctl = E1000_READ_REG(hw, E1000_RFCTL);
3677 		if (filter->hig_pri)
3678 			rfctl |= E1000_RFCTL_SYNQFP;
3679 		else
3680 			rfctl &= ~E1000_RFCTL_SYNQFP;
3681 
3682 		E1000_WRITE_REG(hw, E1000_RFCTL, rfctl);
3683 	} else {
3684 		if (!(synqf & E1000_SYN_FILTER_ENABLE))
3685 			return -ENOENT;
3686 		synqf = 0;
3687 	}
3688 
3689 	filter_info->syn_info = synqf;
3690 	E1000_WRITE_REG(hw, E1000_SYNQF(0), synqf);
3691 	E1000_WRITE_FLUSH(hw);
3692 	return 0;
3693 }
3694 
3695 /* translate elements in struct rte_eth_ntuple_filter to struct e1000_2tuple_filter_info*/
3696 static inline int
3697 ntuple_filter_to_2tuple(struct rte_eth_ntuple_filter *filter,
3698 			struct e1000_2tuple_filter_info *filter_info)
3699 {
3700 	if (filter->queue >= IGB_MAX_RX_QUEUE_NUM)
3701 		return -EINVAL;
3702 	if (filter->priority > E1000_2TUPLE_MAX_PRI)
3703 		return -EINVAL;  /* filter index is out of range. */
3704 	if (filter->tcp_flags > RTE_NTUPLE_TCP_FLAGS_MASK)
3705 		return -EINVAL;  /* flags is invalid. */
3706 
3707 	switch (filter->dst_port_mask) {
3708 	case UINT16_MAX:
3709 		filter_info->dst_port_mask = 0;
3710 		filter_info->dst_port = filter->dst_port;
3711 		break;
3712 	case 0:
3713 		filter_info->dst_port_mask = 1;
3714 		break;
3715 	default:
3716 		PMD_DRV_LOG(ERR, "invalid dst_port mask.");
3717 		return -EINVAL;
3718 	}
3719 
3720 	switch (filter->proto_mask) {
3721 	case UINT8_MAX:
3722 		filter_info->proto_mask = 0;
3723 		filter_info->proto = filter->proto;
3724 		break;
3725 	case 0:
3726 		filter_info->proto_mask = 1;
3727 		break;
3728 	default:
3729 		PMD_DRV_LOG(ERR, "invalid protocol mask.");
3730 		return -EINVAL;
3731 	}
3732 
3733 	filter_info->priority = (uint8_t)filter->priority;
3734 	if (filter->flags & RTE_NTUPLE_FLAGS_TCP_FLAG)
3735 		filter_info->tcp_flags = filter->tcp_flags;
3736 	else
3737 		filter_info->tcp_flags = 0;
3738 
3739 	return 0;
3740 }
3741 
3742 static inline struct e1000_2tuple_filter *
3743 igb_2tuple_filter_lookup(struct e1000_2tuple_filter_list *filter_list,
3744 			struct e1000_2tuple_filter_info *key)
3745 {
3746 	struct e1000_2tuple_filter *it;
3747 
3748 	TAILQ_FOREACH(it, filter_list, entries) {
3749 		if (memcmp(key, &it->filter_info,
3750 			sizeof(struct e1000_2tuple_filter_info)) == 0) {
3751 			return it;
3752 		}
3753 	}
3754 	return NULL;
3755 }
3756 
3757 /* inject a igb 2tuple filter to HW */
3758 static inline void
3759 igb_inject_2uple_filter(struct rte_eth_dev *dev,
3760 			   struct e1000_2tuple_filter *filter)
3761 {
3762 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3763 	uint32_t ttqf = E1000_TTQF_DISABLE_MASK;
3764 	uint32_t imir, imir_ext = E1000_IMIREXT_SIZE_BP;
3765 	int i;
3766 
3767 	i = filter->index;
3768 	imir = (uint32_t)(filter->filter_info.dst_port & E1000_IMIR_DSTPORT);
3769 	if (filter->filter_info.dst_port_mask == 1) /* 1b means not compare. */
3770 		imir |= E1000_IMIR_PORT_BP;
3771 	else
3772 		imir &= ~E1000_IMIR_PORT_BP;
3773 
3774 	imir |= filter->filter_info.priority << E1000_IMIR_PRIORITY_SHIFT;
3775 
3776 	ttqf |= E1000_TTQF_QUEUE_ENABLE;
3777 	ttqf |= (uint32_t)(filter->queue << E1000_TTQF_QUEUE_SHIFT);
3778 	ttqf |= (uint32_t)(filter->filter_info.proto &
3779 						E1000_TTQF_PROTOCOL_MASK);
3780 	if (filter->filter_info.proto_mask == 0)
3781 		ttqf &= ~E1000_TTQF_MASK_ENABLE;
3782 
3783 	/* tcp flags bits setting. */
3784 	if (filter->filter_info.tcp_flags & RTE_NTUPLE_TCP_FLAGS_MASK) {
3785 		if (filter->filter_info.tcp_flags & RTE_TCP_URG_FLAG)
3786 			imir_ext |= E1000_IMIREXT_CTRL_URG;
3787 		if (filter->filter_info.tcp_flags & RTE_TCP_ACK_FLAG)
3788 			imir_ext |= E1000_IMIREXT_CTRL_ACK;
3789 		if (filter->filter_info.tcp_flags & RTE_TCP_PSH_FLAG)
3790 			imir_ext |= E1000_IMIREXT_CTRL_PSH;
3791 		if (filter->filter_info.tcp_flags & RTE_TCP_RST_FLAG)
3792 			imir_ext |= E1000_IMIREXT_CTRL_RST;
3793 		if (filter->filter_info.tcp_flags & RTE_TCP_SYN_FLAG)
3794 			imir_ext |= E1000_IMIREXT_CTRL_SYN;
3795 		if (filter->filter_info.tcp_flags & RTE_TCP_FIN_FLAG)
3796 			imir_ext |= E1000_IMIREXT_CTRL_FIN;
3797 	} else {
3798 		imir_ext |= E1000_IMIREXT_CTRL_BP;
3799 	}
3800 	E1000_WRITE_REG(hw, E1000_IMIR(i), imir);
3801 	E1000_WRITE_REG(hw, E1000_TTQF(i), ttqf);
3802 	E1000_WRITE_REG(hw, E1000_IMIREXT(i), imir_ext);
3803 }
3804 
3805 /*
3806  * igb_add_2tuple_filter - add a 2tuple filter
3807  *
3808  * @param
3809  * dev: Pointer to struct rte_eth_dev.
3810  * ntuple_filter: ponter to the filter that will be added.
3811  *
3812  * @return
3813  *    - On success, zero.
3814  *    - On failure, a negative value.
3815  */
3816 static int
3817 igb_add_2tuple_filter(struct rte_eth_dev *dev,
3818 			struct rte_eth_ntuple_filter *ntuple_filter)
3819 {
3820 	struct e1000_filter_info *filter_info =
3821 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
3822 	struct e1000_2tuple_filter *filter;
3823 	int i, ret;
3824 
3825 	filter = rte_zmalloc("e1000_2tuple_filter",
3826 			sizeof(struct e1000_2tuple_filter), 0);
3827 	if (filter == NULL)
3828 		return -ENOMEM;
3829 
3830 	ret = ntuple_filter_to_2tuple(ntuple_filter,
3831 				      &filter->filter_info);
3832 	if (ret < 0) {
3833 		rte_free(filter);
3834 		return ret;
3835 	}
3836 	if (igb_2tuple_filter_lookup(&filter_info->twotuple_list,
3837 					 &filter->filter_info) != NULL) {
3838 		PMD_DRV_LOG(ERR, "filter exists.");
3839 		rte_free(filter);
3840 		return -EEXIST;
3841 	}
3842 	filter->queue = ntuple_filter->queue;
3843 
3844 	/*
3845 	 * look for an unused 2tuple filter index,
3846 	 * and insert the filter to list.
3847 	 */
3848 	for (i = 0; i < E1000_MAX_TTQF_FILTERS; i++) {
3849 		if (!(filter_info->twotuple_mask & (1 << i))) {
3850 			filter_info->twotuple_mask |= 1 << i;
3851 			filter->index = i;
3852 			TAILQ_INSERT_TAIL(&filter_info->twotuple_list,
3853 					  filter,
3854 					  entries);
3855 			break;
3856 		}
3857 	}
3858 	if (i >= E1000_MAX_TTQF_FILTERS) {
3859 		PMD_DRV_LOG(ERR, "2tuple filters are full.");
3860 		rte_free(filter);
3861 		return -ENOSYS;
3862 	}
3863 
3864 	igb_inject_2uple_filter(dev, filter);
3865 	return 0;
3866 }
3867 
3868 int
3869 igb_delete_2tuple_filter(struct rte_eth_dev *dev,
3870 			struct e1000_2tuple_filter *filter)
3871 {
3872 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3873 	struct e1000_filter_info *filter_info =
3874 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
3875 
3876 	filter_info->twotuple_mask &= ~(1 << filter->index);
3877 	TAILQ_REMOVE(&filter_info->twotuple_list, filter, entries);
3878 	rte_free(filter);
3879 
3880 	E1000_WRITE_REG(hw, E1000_TTQF(filter->index), E1000_TTQF_DISABLE_MASK);
3881 	E1000_WRITE_REG(hw, E1000_IMIR(filter->index), 0);
3882 	E1000_WRITE_REG(hw, E1000_IMIREXT(filter->index), 0);
3883 	return 0;
3884 }
3885 
3886 /*
3887  * igb_remove_2tuple_filter - remove a 2tuple filter
3888  *
3889  * @param
3890  * dev: Pointer to struct rte_eth_dev.
3891  * ntuple_filter: ponter to the filter that will be removed.
3892  *
3893  * @return
3894  *    - On success, zero.
3895  *    - On failure, a negative value.
3896  */
3897 static int
3898 igb_remove_2tuple_filter(struct rte_eth_dev *dev,
3899 			struct rte_eth_ntuple_filter *ntuple_filter)
3900 {
3901 	struct e1000_filter_info *filter_info =
3902 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
3903 	struct e1000_2tuple_filter_info filter_2tuple;
3904 	struct e1000_2tuple_filter *filter;
3905 	int ret;
3906 
3907 	memset(&filter_2tuple, 0, sizeof(struct e1000_2tuple_filter_info));
3908 	ret = ntuple_filter_to_2tuple(ntuple_filter,
3909 				      &filter_2tuple);
3910 	if (ret < 0)
3911 		return ret;
3912 
3913 	filter = igb_2tuple_filter_lookup(&filter_info->twotuple_list,
3914 					 &filter_2tuple);
3915 	if (filter == NULL) {
3916 		PMD_DRV_LOG(ERR, "filter doesn't exist.");
3917 		return -ENOENT;
3918 	}
3919 
3920 	igb_delete_2tuple_filter(dev, filter);
3921 
3922 	return 0;
3923 }
3924 
3925 /* inject a igb flex filter to HW */
3926 static inline void
3927 igb_inject_flex_filter(struct rte_eth_dev *dev,
3928 			   struct e1000_flex_filter *filter)
3929 {
3930 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3931 	uint32_t wufc, queueing;
3932 	uint32_t reg_off;
3933 	uint8_t i, j = 0;
3934 
3935 	wufc = E1000_READ_REG(hw, E1000_WUFC);
3936 	if (filter->index < E1000_MAX_FHFT)
3937 		reg_off = E1000_FHFT(filter->index);
3938 	else
3939 		reg_off = E1000_FHFT_EXT(filter->index - E1000_MAX_FHFT);
3940 
3941 	E1000_WRITE_REG(hw, E1000_WUFC, wufc | E1000_WUFC_FLEX_HQ |
3942 			(E1000_WUFC_FLX0 << filter->index));
3943 	queueing = filter->filter_info.len |
3944 		(filter->queue << E1000_FHFT_QUEUEING_QUEUE_SHIFT) |
3945 		(filter->filter_info.priority <<
3946 			E1000_FHFT_QUEUEING_PRIO_SHIFT);
3947 	E1000_WRITE_REG(hw, reg_off + E1000_FHFT_QUEUEING_OFFSET,
3948 			queueing);
3949 
3950 	for (i = 0; i < E1000_FLEX_FILTERS_MASK_SIZE; i++) {
3951 		E1000_WRITE_REG(hw, reg_off,
3952 				filter->filter_info.dwords[j]);
3953 		reg_off += sizeof(uint32_t);
3954 		E1000_WRITE_REG(hw, reg_off,
3955 				filter->filter_info.dwords[++j]);
3956 		reg_off += sizeof(uint32_t);
3957 		E1000_WRITE_REG(hw, reg_off,
3958 			(uint32_t)filter->filter_info.mask[i]);
3959 		reg_off += sizeof(uint32_t) * 2;
3960 		++j;
3961 	}
3962 }
3963 
3964 static inline struct e1000_flex_filter *
3965 eth_igb_flex_filter_lookup(struct e1000_flex_filter_list *filter_list,
3966 			struct e1000_flex_filter_info *key)
3967 {
3968 	struct e1000_flex_filter *it;
3969 
3970 	TAILQ_FOREACH(it, filter_list, entries) {
3971 		if (memcmp(key, &it->filter_info,
3972 			sizeof(struct e1000_flex_filter_info)) == 0)
3973 			return it;
3974 	}
3975 
3976 	return NULL;
3977 }
3978 
3979 /* remove a flex byte filter
3980  * @param
3981  * dev: Pointer to struct rte_eth_dev.
3982  * filter: the pointer of the filter will be removed.
3983  */
3984 void
3985 igb_remove_flex_filter(struct rte_eth_dev *dev,
3986 			struct e1000_flex_filter *filter)
3987 {
3988 	struct e1000_filter_info *filter_info =
3989 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
3990 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3991 	uint32_t wufc, i;
3992 	uint32_t reg_off;
3993 
3994 	wufc = E1000_READ_REG(hw, E1000_WUFC);
3995 	if (filter->index < E1000_MAX_FHFT)
3996 		reg_off = E1000_FHFT(filter->index);
3997 	else
3998 		reg_off = E1000_FHFT_EXT(filter->index - E1000_MAX_FHFT);
3999 
4000 	for (i = 0; i < E1000_FHFT_SIZE_IN_DWD; i++)
4001 		E1000_WRITE_REG(hw, reg_off + i * sizeof(uint32_t), 0);
4002 
4003 	E1000_WRITE_REG(hw, E1000_WUFC, wufc &
4004 		(~(E1000_WUFC_FLX0 << filter->index)));
4005 
4006 	filter_info->flex_mask &= ~(1 << filter->index);
4007 	TAILQ_REMOVE(&filter_info->flex_list, filter, entries);
4008 	rte_free(filter);
4009 }
4010 
4011 int
4012 eth_igb_add_del_flex_filter(struct rte_eth_dev *dev,
4013 			struct igb_flex_filter *filter,
4014 			bool add)
4015 {
4016 	struct e1000_filter_info *filter_info =
4017 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
4018 	struct e1000_flex_filter *flex_filter, *it;
4019 	uint32_t mask;
4020 	uint8_t shift, i;
4021 
4022 	flex_filter = rte_zmalloc("e1000_flex_filter",
4023 			sizeof(struct e1000_flex_filter), 0);
4024 	if (flex_filter == NULL)
4025 		return -ENOMEM;
4026 
4027 	flex_filter->filter_info.len = filter->len;
4028 	flex_filter->filter_info.priority = filter->priority;
4029 	memcpy(flex_filter->filter_info.dwords, filter->bytes, filter->len);
4030 	for (i = 0; i < RTE_ALIGN(filter->len, CHAR_BIT) / CHAR_BIT; i++) {
4031 		mask = 0;
4032 		/* reverse bits in flex filter's mask*/
4033 		for (shift = 0; shift < CHAR_BIT; shift++) {
4034 			if (filter->mask[i] & (0x01 << shift))
4035 				mask |= (0x80 >> shift);
4036 		}
4037 		flex_filter->filter_info.mask[i] = mask;
4038 	}
4039 
4040 	it = eth_igb_flex_filter_lookup(&filter_info->flex_list,
4041 				&flex_filter->filter_info);
4042 	if (it == NULL && !add) {
4043 		PMD_DRV_LOG(ERR, "filter doesn't exist.");
4044 		rte_free(flex_filter);
4045 		return -ENOENT;
4046 	}
4047 	if (it != NULL && add) {
4048 		PMD_DRV_LOG(ERR, "filter exists.");
4049 		rte_free(flex_filter);
4050 		return -EEXIST;
4051 	}
4052 
4053 	if (add) {
4054 		flex_filter->queue = filter->queue;
4055 		/*
4056 		 * look for an unused flex filter index
4057 		 * and insert the filter into the list.
4058 		 */
4059 		for (i = 0; i < E1000_MAX_FLEX_FILTERS; i++) {
4060 			if (!(filter_info->flex_mask & (1 << i))) {
4061 				filter_info->flex_mask |= 1 << i;
4062 				flex_filter->index = i;
4063 				TAILQ_INSERT_TAIL(&filter_info->flex_list,
4064 					flex_filter,
4065 					entries);
4066 				break;
4067 			}
4068 		}
4069 		if (i >= E1000_MAX_FLEX_FILTERS) {
4070 			PMD_DRV_LOG(ERR, "flex filters are full.");
4071 			rte_free(flex_filter);
4072 			return -ENOSYS;
4073 		}
4074 
4075 		igb_inject_flex_filter(dev, flex_filter);
4076 
4077 	} else {
4078 		igb_remove_flex_filter(dev, it);
4079 		rte_free(flex_filter);
4080 	}
4081 
4082 	return 0;
4083 }
4084 
4085 /* translate elements in struct rte_eth_ntuple_filter to struct e1000_5tuple_filter_info*/
4086 static inline int
4087 ntuple_filter_to_5tuple_82576(struct rte_eth_ntuple_filter *filter,
4088 			struct e1000_5tuple_filter_info *filter_info)
4089 {
4090 	if (filter->queue >= IGB_MAX_RX_QUEUE_NUM_82576)
4091 		return -EINVAL;
4092 	if (filter->priority > E1000_2TUPLE_MAX_PRI)
4093 		return -EINVAL;  /* filter index is out of range. */
4094 	if (filter->tcp_flags > RTE_NTUPLE_TCP_FLAGS_MASK)
4095 		return -EINVAL;  /* flags is invalid. */
4096 
4097 	switch (filter->dst_ip_mask) {
4098 	case UINT32_MAX:
4099 		filter_info->dst_ip_mask = 0;
4100 		filter_info->dst_ip = filter->dst_ip;
4101 		break;
4102 	case 0:
4103 		filter_info->dst_ip_mask = 1;
4104 		break;
4105 	default:
4106 		PMD_DRV_LOG(ERR, "invalid dst_ip mask.");
4107 		return -EINVAL;
4108 	}
4109 
4110 	switch (filter->src_ip_mask) {
4111 	case UINT32_MAX:
4112 		filter_info->src_ip_mask = 0;
4113 		filter_info->src_ip = filter->src_ip;
4114 		break;
4115 	case 0:
4116 		filter_info->src_ip_mask = 1;
4117 		break;
4118 	default:
4119 		PMD_DRV_LOG(ERR, "invalid src_ip mask.");
4120 		return -EINVAL;
4121 	}
4122 
4123 	switch (filter->dst_port_mask) {
4124 	case UINT16_MAX:
4125 		filter_info->dst_port_mask = 0;
4126 		filter_info->dst_port = filter->dst_port;
4127 		break;
4128 	case 0:
4129 		filter_info->dst_port_mask = 1;
4130 		break;
4131 	default:
4132 		PMD_DRV_LOG(ERR, "invalid dst_port mask.");
4133 		return -EINVAL;
4134 	}
4135 
4136 	switch (filter->src_port_mask) {
4137 	case UINT16_MAX:
4138 		filter_info->src_port_mask = 0;
4139 		filter_info->src_port = filter->src_port;
4140 		break;
4141 	case 0:
4142 		filter_info->src_port_mask = 1;
4143 		break;
4144 	default:
4145 		PMD_DRV_LOG(ERR, "invalid src_port mask.");
4146 		return -EINVAL;
4147 	}
4148 
4149 	switch (filter->proto_mask) {
4150 	case UINT8_MAX:
4151 		filter_info->proto_mask = 0;
4152 		filter_info->proto = filter->proto;
4153 		break;
4154 	case 0:
4155 		filter_info->proto_mask = 1;
4156 		break;
4157 	default:
4158 		PMD_DRV_LOG(ERR, "invalid protocol mask.");
4159 		return -EINVAL;
4160 	}
4161 
4162 	filter_info->priority = (uint8_t)filter->priority;
4163 	if (filter->flags & RTE_NTUPLE_FLAGS_TCP_FLAG)
4164 		filter_info->tcp_flags = filter->tcp_flags;
4165 	else
4166 		filter_info->tcp_flags = 0;
4167 
4168 	return 0;
4169 }
4170 
4171 static inline struct e1000_5tuple_filter *
4172 igb_5tuple_filter_lookup_82576(struct e1000_5tuple_filter_list *filter_list,
4173 			struct e1000_5tuple_filter_info *key)
4174 {
4175 	struct e1000_5tuple_filter *it;
4176 
4177 	TAILQ_FOREACH(it, filter_list, entries) {
4178 		if (memcmp(key, &it->filter_info,
4179 			sizeof(struct e1000_5tuple_filter_info)) == 0) {
4180 			return it;
4181 		}
4182 	}
4183 	return NULL;
4184 }
4185 
4186 /* inject a igb 5-tuple filter to HW */
4187 static inline void
4188 igb_inject_5tuple_filter_82576(struct rte_eth_dev *dev,
4189 			   struct e1000_5tuple_filter *filter)
4190 {
4191 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4192 	uint32_t ftqf = E1000_FTQF_VF_BP | E1000_FTQF_MASK;
4193 	uint32_t spqf, imir, imir_ext = E1000_IMIREXT_SIZE_BP;
4194 	uint8_t i;
4195 
4196 	i = filter->index;
4197 	ftqf |= filter->filter_info.proto & E1000_FTQF_PROTOCOL_MASK;
4198 	if (filter->filter_info.src_ip_mask == 0) /* 0b means compare. */
4199 		ftqf &= ~E1000_FTQF_MASK_SOURCE_ADDR_BP;
4200 	if (filter->filter_info.dst_ip_mask == 0)
4201 		ftqf &= ~E1000_FTQF_MASK_DEST_ADDR_BP;
4202 	if (filter->filter_info.src_port_mask == 0)
4203 		ftqf &= ~E1000_FTQF_MASK_SOURCE_PORT_BP;
4204 	if (filter->filter_info.proto_mask == 0)
4205 		ftqf &= ~E1000_FTQF_MASK_PROTO_BP;
4206 	ftqf |= (filter->queue << E1000_FTQF_QUEUE_SHIFT) &
4207 		E1000_FTQF_QUEUE_MASK;
4208 	ftqf |= E1000_FTQF_QUEUE_ENABLE;
4209 	E1000_WRITE_REG(hw, E1000_FTQF(i), ftqf);
4210 	E1000_WRITE_REG(hw, E1000_DAQF(i), filter->filter_info.dst_ip);
4211 	E1000_WRITE_REG(hw, E1000_SAQF(i), filter->filter_info.src_ip);
4212 
4213 	spqf = filter->filter_info.src_port & E1000_SPQF_SRCPORT;
4214 	E1000_WRITE_REG(hw, E1000_SPQF(i), spqf);
4215 
4216 	imir = (uint32_t)(filter->filter_info.dst_port & E1000_IMIR_DSTPORT);
4217 	if (filter->filter_info.dst_port_mask == 1) /* 1b means not compare. */
4218 		imir |= E1000_IMIR_PORT_BP;
4219 	else
4220 		imir &= ~E1000_IMIR_PORT_BP;
4221 	imir |= filter->filter_info.priority << E1000_IMIR_PRIORITY_SHIFT;
4222 
4223 	/* tcp flags bits setting. */
4224 	if (filter->filter_info.tcp_flags & RTE_NTUPLE_TCP_FLAGS_MASK) {
4225 		if (filter->filter_info.tcp_flags & RTE_TCP_URG_FLAG)
4226 			imir_ext |= E1000_IMIREXT_CTRL_URG;
4227 		if (filter->filter_info.tcp_flags & RTE_TCP_ACK_FLAG)
4228 			imir_ext |= E1000_IMIREXT_CTRL_ACK;
4229 		if (filter->filter_info.tcp_flags & RTE_TCP_PSH_FLAG)
4230 			imir_ext |= E1000_IMIREXT_CTRL_PSH;
4231 		if (filter->filter_info.tcp_flags & RTE_TCP_RST_FLAG)
4232 			imir_ext |= E1000_IMIREXT_CTRL_RST;
4233 		if (filter->filter_info.tcp_flags & RTE_TCP_SYN_FLAG)
4234 			imir_ext |= E1000_IMIREXT_CTRL_SYN;
4235 		if (filter->filter_info.tcp_flags & RTE_TCP_FIN_FLAG)
4236 			imir_ext |= E1000_IMIREXT_CTRL_FIN;
4237 	} else {
4238 		imir_ext |= E1000_IMIREXT_CTRL_BP;
4239 	}
4240 	E1000_WRITE_REG(hw, E1000_IMIR(i), imir);
4241 	E1000_WRITE_REG(hw, E1000_IMIREXT(i), imir_ext);
4242 }
4243 
4244 /*
4245  * igb_add_5tuple_filter_82576 - add a 5tuple filter
4246  *
4247  * @param
4248  * dev: Pointer to struct rte_eth_dev.
4249  * ntuple_filter: ponter to the filter that will be added.
4250  *
4251  * @return
4252  *    - On success, zero.
4253  *    - On failure, a negative value.
4254  */
4255 static int
4256 igb_add_5tuple_filter_82576(struct rte_eth_dev *dev,
4257 			struct rte_eth_ntuple_filter *ntuple_filter)
4258 {
4259 	struct e1000_filter_info *filter_info =
4260 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
4261 	struct e1000_5tuple_filter *filter;
4262 	uint8_t i;
4263 	int ret;
4264 
4265 	filter = rte_zmalloc("e1000_5tuple_filter",
4266 			sizeof(struct e1000_5tuple_filter), 0);
4267 	if (filter == NULL)
4268 		return -ENOMEM;
4269 
4270 	ret = ntuple_filter_to_5tuple_82576(ntuple_filter,
4271 					    &filter->filter_info);
4272 	if (ret < 0) {
4273 		rte_free(filter);
4274 		return ret;
4275 	}
4276 
4277 	if (igb_5tuple_filter_lookup_82576(&filter_info->fivetuple_list,
4278 					 &filter->filter_info) != NULL) {
4279 		PMD_DRV_LOG(ERR, "filter exists.");
4280 		rte_free(filter);
4281 		return -EEXIST;
4282 	}
4283 	filter->queue = ntuple_filter->queue;
4284 
4285 	/*
4286 	 * look for an unused 5tuple filter index,
4287 	 * and insert the filter to list.
4288 	 */
4289 	for (i = 0; i < E1000_MAX_FTQF_FILTERS; i++) {
4290 		if (!(filter_info->fivetuple_mask & (1 << i))) {
4291 			filter_info->fivetuple_mask |= 1 << i;
4292 			filter->index = i;
4293 			TAILQ_INSERT_TAIL(&filter_info->fivetuple_list,
4294 					  filter,
4295 					  entries);
4296 			break;
4297 		}
4298 	}
4299 	if (i >= E1000_MAX_FTQF_FILTERS) {
4300 		PMD_DRV_LOG(ERR, "5tuple filters are full.");
4301 		rte_free(filter);
4302 		return -ENOSYS;
4303 	}
4304 
4305 	igb_inject_5tuple_filter_82576(dev, filter);
4306 	return 0;
4307 }
4308 
4309 int
4310 igb_delete_5tuple_filter_82576(struct rte_eth_dev *dev,
4311 				struct e1000_5tuple_filter *filter)
4312 {
4313 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4314 	struct e1000_filter_info *filter_info =
4315 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
4316 
4317 	filter_info->fivetuple_mask &= ~(1 << filter->index);
4318 	TAILQ_REMOVE(&filter_info->fivetuple_list, filter, entries);
4319 	rte_free(filter);
4320 
4321 	E1000_WRITE_REG(hw, E1000_FTQF(filter->index),
4322 			E1000_FTQF_VF_BP | E1000_FTQF_MASK);
4323 	E1000_WRITE_REG(hw, E1000_DAQF(filter->index), 0);
4324 	E1000_WRITE_REG(hw, E1000_SAQF(filter->index), 0);
4325 	E1000_WRITE_REG(hw, E1000_SPQF(filter->index), 0);
4326 	E1000_WRITE_REG(hw, E1000_IMIR(filter->index), 0);
4327 	E1000_WRITE_REG(hw, E1000_IMIREXT(filter->index), 0);
4328 	return 0;
4329 }
4330 
4331 /*
4332  * igb_remove_5tuple_filter_82576 - remove a 5tuple filter
4333  *
4334  * @param
4335  * dev: Pointer to struct rte_eth_dev.
4336  * ntuple_filter: ponter to the filter that will be removed.
4337  *
4338  * @return
4339  *    - On success, zero.
4340  *    - On failure, a negative value.
4341  */
4342 static int
4343 igb_remove_5tuple_filter_82576(struct rte_eth_dev *dev,
4344 				struct rte_eth_ntuple_filter *ntuple_filter)
4345 {
4346 	struct e1000_filter_info *filter_info =
4347 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
4348 	struct e1000_5tuple_filter_info filter_5tuple;
4349 	struct e1000_5tuple_filter *filter;
4350 	int ret;
4351 
4352 	memset(&filter_5tuple, 0, sizeof(struct e1000_5tuple_filter_info));
4353 	ret = ntuple_filter_to_5tuple_82576(ntuple_filter,
4354 					    &filter_5tuple);
4355 	if (ret < 0)
4356 		return ret;
4357 
4358 	filter = igb_5tuple_filter_lookup_82576(&filter_info->fivetuple_list,
4359 					 &filter_5tuple);
4360 	if (filter == NULL) {
4361 		PMD_DRV_LOG(ERR, "filter doesn't exist.");
4362 		return -ENOENT;
4363 	}
4364 
4365 	igb_delete_5tuple_filter_82576(dev, filter);
4366 
4367 	return 0;
4368 }
4369 
4370 static int
4371 eth_igb_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
4372 {
4373 	uint32_t rctl;
4374 	struct e1000_hw *hw;
4375 	struct rte_eth_dev_info dev_info;
4376 	uint32_t frame_size = mtu + E1000_ETH_OVERHEAD;
4377 	int ret;
4378 
4379 	hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4380 
4381 #ifdef RTE_LIBRTE_82571_SUPPORT
4382 	/* XXX: not bigger than max_rx_pktlen */
4383 	if (hw->mac.type == e1000_82571)
4384 		return -ENOTSUP;
4385 #endif
4386 	ret = eth_igb_infos_get(dev, &dev_info);
4387 	if (ret != 0)
4388 		return ret;
4389 
4390 	/* check that mtu is within the allowed range */
4391 	if (mtu < RTE_ETHER_MIN_MTU ||
4392 			frame_size > dev_info.max_rx_pktlen)
4393 		return -EINVAL;
4394 
4395 	/*
4396 	 * If device is started, refuse mtu that requires the support of
4397 	 * scattered packets when this feature has not been enabled before.
4398 	 */
4399 	if (dev->data->dev_started && !dev->data->scattered_rx &&
4400 	    frame_size > dev->data->min_rx_buf_size - RTE_PKTMBUF_HEADROOM) {
4401 		PMD_INIT_LOG(ERR, "Stop port first.");
4402 		return -EINVAL;
4403 	}
4404 
4405 	rctl = E1000_READ_REG(hw, E1000_RCTL);
4406 
4407 	/* switch to jumbo mode if needed */
4408 	if (frame_size > E1000_ETH_MAX_LEN) {
4409 		dev->data->dev_conf.rxmode.offloads |=
4410 			DEV_RX_OFFLOAD_JUMBO_FRAME;
4411 		rctl |= E1000_RCTL_LPE;
4412 	} else {
4413 		dev->data->dev_conf.rxmode.offloads &=
4414 			~DEV_RX_OFFLOAD_JUMBO_FRAME;
4415 		rctl &= ~E1000_RCTL_LPE;
4416 	}
4417 	E1000_WRITE_REG(hw, E1000_RCTL, rctl);
4418 
4419 	/* update max frame size */
4420 	dev->data->dev_conf.rxmode.max_rx_pkt_len = frame_size;
4421 
4422 	E1000_WRITE_REG(hw, E1000_RLPML,
4423 			dev->data->dev_conf.rxmode.max_rx_pkt_len);
4424 
4425 	return 0;
4426 }
4427 
4428 /*
4429  * igb_add_del_ntuple_filter - add or delete a ntuple filter
4430  *
4431  * @param
4432  * dev: Pointer to struct rte_eth_dev.
4433  * ntuple_filter: Pointer to struct rte_eth_ntuple_filter
4434  * add: if true, add filter, if false, remove filter
4435  *
4436  * @return
4437  *    - On success, zero.
4438  *    - On failure, a negative value.
4439  */
4440 int
4441 igb_add_del_ntuple_filter(struct rte_eth_dev *dev,
4442 			struct rte_eth_ntuple_filter *ntuple_filter,
4443 			bool add)
4444 {
4445 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4446 	int ret;
4447 
4448 	switch (ntuple_filter->flags) {
4449 	case RTE_5TUPLE_FLAGS:
4450 	case (RTE_5TUPLE_FLAGS | RTE_NTUPLE_FLAGS_TCP_FLAG):
4451 		if (hw->mac.type != e1000_82576)
4452 			return -ENOTSUP;
4453 		if (add)
4454 			ret = igb_add_5tuple_filter_82576(dev,
4455 							  ntuple_filter);
4456 		else
4457 			ret = igb_remove_5tuple_filter_82576(dev,
4458 							     ntuple_filter);
4459 		break;
4460 	case RTE_2TUPLE_FLAGS:
4461 	case (RTE_2TUPLE_FLAGS | RTE_NTUPLE_FLAGS_TCP_FLAG):
4462 		if (hw->mac.type != e1000_82580 && hw->mac.type != e1000_i350 &&
4463 			hw->mac.type != e1000_i210 &&
4464 			hw->mac.type != e1000_i211)
4465 			return -ENOTSUP;
4466 		if (add)
4467 			ret = igb_add_2tuple_filter(dev, ntuple_filter);
4468 		else
4469 			ret = igb_remove_2tuple_filter(dev, ntuple_filter);
4470 		break;
4471 	default:
4472 		ret = -EINVAL;
4473 		break;
4474 	}
4475 
4476 	return ret;
4477 }
4478 
4479 static inline int
4480 igb_ethertype_filter_lookup(struct e1000_filter_info *filter_info,
4481 			uint16_t ethertype)
4482 {
4483 	int i;
4484 
4485 	for (i = 0; i < E1000_MAX_ETQF_FILTERS; i++) {
4486 		if (filter_info->ethertype_filters[i].ethertype == ethertype &&
4487 		    (filter_info->ethertype_mask & (1 << i)))
4488 			return i;
4489 	}
4490 	return -1;
4491 }
4492 
4493 static inline int
4494 igb_ethertype_filter_insert(struct e1000_filter_info *filter_info,
4495 			uint16_t ethertype, uint32_t etqf)
4496 {
4497 	int i;
4498 
4499 	for (i = 0; i < E1000_MAX_ETQF_FILTERS; i++) {
4500 		if (!(filter_info->ethertype_mask & (1 << i))) {
4501 			filter_info->ethertype_mask |= 1 << i;
4502 			filter_info->ethertype_filters[i].ethertype = ethertype;
4503 			filter_info->ethertype_filters[i].etqf = etqf;
4504 			return i;
4505 		}
4506 	}
4507 	return -1;
4508 }
4509 
4510 int
4511 igb_ethertype_filter_remove(struct e1000_filter_info *filter_info,
4512 			uint8_t idx)
4513 {
4514 	if (idx >= E1000_MAX_ETQF_FILTERS)
4515 		return -1;
4516 	filter_info->ethertype_mask &= ~(1 << idx);
4517 	filter_info->ethertype_filters[idx].ethertype = 0;
4518 	filter_info->ethertype_filters[idx].etqf = 0;
4519 	return idx;
4520 }
4521 
4522 
4523 int
4524 igb_add_del_ethertype_filter(struct rte_eth_dev *dev,
4525 			struct rte_eth_ethertype_filter *filter,
4526 			bool add)
4527 {
4528 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4529 	struct e1000_filter_info *filter_info =
4530 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
4531 	uint32_t etqf = 0;
4532 	int ret;
4533 
4534 	if (filter->ether_type == RTE_ETHER_TYPE_IPV4 ||
4535 		filter->ether_type == RTE_ETHER_TYPE_IPV6) {
4536 		PMD_DRV_LOG(ERR, "unsupported ether_type(0x%04x) in"
4537 			" ethertype filter.", filter->ether_type);
4538 		return -EINVAL;
4539 	}
4540 
4541 	if (filter->flags & RTE_ETHTYPE_FLAGS_MAC) {
4542 		PMD_DRV_LOG(ERR, "mac compare is unsupported.");
4543 		return -EINVAL;
4544 	}
4545 	if (filter->flags & RTE_ETHTYPE_FLAGS_DROP) {
4546 		PMD_DRV_LOG(ERR, "drop option is unsupported.");
4547 		return -EINVAL;
4548 	}
4549 
4550 	ret = igb_ethertype_filter_lookup(filter_info, filter->ether_type);
4551 	if (ret >= 0 && add) {
4552 		PMD_DRV_LOG(ERR, "ethertype (0x%04x) filter exists.",
4553 			    filter->ether_type);
4554 		return -EEXIST;
4555 	}
4556 	if (ret < 0 && !add) {
4557 		PMD_DRV_LOG(ERR, "ethertype (0x%04x) filter doesn't exist.",
4558 			    filter->ether_type);
4559 		return -ENOENT;
4560 	}
4561 
4562 	if (add) {
4563 		etqf |= E1000_ETQF_FILTER_ENABLE | E1000_ETQF_QUEUE_ENABLE;
4564 		etqf |= (uint32_t)(filter->ether_type & E1000_ETQF_ETHERTYPE);
4565 		etqf |= filter->queue << E1000_ETQF_QUEUE_SHIFT;
4566 		ret = igb_ethertype_filter_insert(filter_info,
4567 				filter->ether_type, etqf);
4568 		if (ret < 0) {
4569 			PMD_DRV_LOG(ERR, "ethertype filters are full.");
4570 			return -ENOSYS;
4571 		}
4572 	} else {
4573 		ret = igb_ethertype_filter_remove(filter_info, (uint8_t)ret);
4574 		if (ret < 0)
4575 			return -ENOSYS;
4576 	}
4577 	E1000_WRITE_REG(hw, E1000_ETQF(ret), etqf);
4578 	E1000_WRITE_FLUSH(hw);
4579 
4580 	return 0;
4581 }
4582 
4583 static int
4584 eth_igb_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
4585 		     const struct rte_flow_ops **ops)
4586 {
4587 	*ops = &igb_flow_ops;
4588 	return 0;
4589 }
4590 
4591 static int
4592 eth_igb_set_mc_addr_list(struct rte_eth_dev *dev,
4593 			 struct rte_ether_addr *mc_addr_set,
4594 			 uint32_t nb_mc_addr)
4595 {
4596 	struct e1000_hw *hw;
4597 
4598 	hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4599 	e1000_update_mc_addr_list(hw, (u8 *)mc_addr_set, nb_mc_addr);
4600 	return 0;
4601 }
4602 
4603 static uint64_t
4604 igb_read_systime_cyclecounter(struct rte_eth_dev *dev)
4605 {
4606 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4607 	uint64_t systime_cycles;
4608 
4609 	switch (hw->mac.type) {
4610 	case e1000_i210:
4611 	case e1000_i211:
4612 		/*
4613 		 * Need to read System Time Residue Register to be able
4614 		 * to read the other two registers.
4615 		 */
4616 		E1000_READ_REG(hw, E1000_SYSTIMR);
4617 		/* SYSTIMEL stores ns and SYSTIMEH stores seconds. */
4618 		systime_cycles = (uint64_t)E1000_READ_REG(hw, E1000_SYSTIML);
4619 		systime_cycles += (uint64_t)E1000_READ_REG(hw, E1000_SYSTIMH)
4620 				* NSEC_PER_SEC;
4621 		break;
4622 	case e1000_82580:
4623 	case e1000_i350:
4624 	case e1000_i354:
4625 		/*
4626 		 * Need to read System Time Residue Register to be able
4627 		 * to read the other two registers.
4628 		 */
4629 		E1000_READ_REG(hw, E1000_SYSTIMR);
4630 		systime_cycles = (uint64_t)E1000_READ_REG(hw, E1000_SYSTIML);
4631 		/* Only the 8 LSB are valid. */
4632 		systime_cycles |= (uint64_t)(E1000_READ_REG(hw, E1000_SYSTIMH)
4633 				& 0xff) << 32;
4634 		break;
4635 	default:
4636 		systime_cycles = (uint64_t)E1000_READ_REG(hw, E1000_SYSTIML);
4637 		systime_cycles |= (uint64_t)E1000_READ_REG(hw, E1000_SYSTIMH)
4638 				<< 32;
4639 		break;
4640 	}
4641 
4642 	return systime_cycles;
4643 }
4644 
4645 static uint64_t
4646 igb_read_rx_tstamp_cyclecounter(struct rte_eth_dev *dev)
4647 {
4648 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4649 	uint64_t rx_tstamp_cycles;
4650 
4651 	switch (hw->mac.type) {
4652 	case e1000_i210:
4653 	case e1000_i211:
4654 		/* RXSTMPL stores ns and RXSTMPH stores seconds. */
4655 		rx_tstamp_cycles = (uint64_t)E1000_READ_REG(hw, E1000_RXSTMPL);
4656 		rx_tstamp_cycles += (uint64_t)E1000_READ_REG(hw, E1000_RXSTMPH)
4657 				* NSEC_PER_SEC;
4658 		break;
4659 	case e1000_82580:
4660 	case e1000_i350:
4661 	case e1000_i354:
4662 		rx_tstamp_cycles = (uint64_t)E1000_READ_REG(hw, E1000_RXSTMPL);
4663 		/* Only the 8 LSB are valid. */
4664 		rx_tstamp_cycles |= (uint64_t)(E1000_READ_REG(hw, E1000_RXSTMPH)
4665 				& 0xff) << 32;
4666 		break;
4667 	default:
4668 		rx_tstamp_cycles = (uint64_t)E1000_READ_REG(hw, E1000_RXSTMPL);
4669 		rx_tstamp_cycles |= (uint64_t)E1000_READ_REG(hw, E1000_RXSTMPH)
4670 				<< 32;
4671 		break;
4672 	}
4673 
4674 	return rx_tstamp_cycles;
4675 }
4676 
4677 static uint64_t
4678 igb_read_tx_tstamp_cyclecounter(struct rte_eth_dev *dev)
4679 {
4680 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4681 	uint64_t tx_tstamp_cycles;
4682 
4683 	switch (hw->mac.type) {
4684 	case e1000_i210:
4685 	case e1000_i211:
4686 		/* RXSTMPL stores ns and RXSTMPH stores seconds. */
4687 		tx_tstamp_cycles = (uint64_t)E1000_READ_REG(hw, E1000_TXSTMPL);
4688 		tx_tstamp_cycles += (uint64_t)E1000_READ_REG(hw, E1000_TXSTMPH)
4689 				* NSEC_PER_SEC;
4690 		break;
4691 	case e1000_82580:
4692 	case e1000_i350:
4693 	case e1000_i354:
4694 		tx_tstamp_cycles = (uint64_t)E1000_READ_REG(hw, E1000_TXSTMPL);
4695 		/* Only the 8 LSB are valid. */
4696 		tx_tstamp_cycles |= (uint64_t)(E1000_READ_REG(hw, E1000_TXSTMPH)
4697 				& 0xff) << 32;
4698 		break;
4699 	default:
4700 		tx_tstamp_cycles = (uint64_t)E1000_READ_REG(hw, E1000_TXSTMPL);
4701 		tx_tstamp_cycles |= (uint64_t)E1000_READ_REG(hw, E1000_TXSTMPH)
4702 				<< 32;
4703 		break;
4704 	}
4705 
4706 	return tx_tstamp_cycles;
4707 }
4708 
4709 static void
4710 igb_start_timecounters(struct rte_eth_dev *dev)
4711 {
4712 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4713 	struct e1000_adapter *adapter = dev->data->dev_private;
4714 	uint32_t incval = 1;
4715 	uint32_t shift = 0;
4716 	uint64_t mask = E1000_CYCLECOUNTER_MASK;
4717 
4718 	switch (hw->mac.type) {
4719 	case e1000_82580:
4720 	case e1000_i350:
4721 	case e1000_i354:
4722 		/* 32 LSB bits + 8 MSB bits = 40 bits */
4723 		mask = (1ULL << 40) - 1;
4724 		/* fall-through */
4725 	case e1000_i210:
4726 	case e1000_i211:
4727 		/*
4728 		 * Start incrementing the register
4729 		 * used to timestamp PTP packets.
4730 		 */
4731 		E1000_WRITE_REG(hw, E1000_TIMINCA, incval);
4732 		break;
4733 	case e1000_82576:
4734 		incval = E1000_INCVALUE_82576;
4735 		shift = IGB_82576_TSYNC_SHIFT;
4736 		E1000_WRITE_REG(hw, E1000_TIMINCA,
4737 				E1000_INCPERIOD_82576 | incval);
4738 		break;
4739 	default:
4740 		/* Not supported */
4741 		return;
4742 	}
4743 
4744 	memset(&adapter->systime_tc, 0, sizeof(struct rte_timecounter));
4745 	memset(&adapter->rx_tstamp_tc, 0, sizeof(struct rte_timecounter));
4746 	memset(&adapter->tx_tstamp_tc, 0, sizeof(struct rte_timecounter));
4747 
4748 	adapter->systime_tc.cc_mask = mask;
4749 	adapter->systime_tc.cc_shift = shift;
4750 	adapter->systime_tc.nsec_mask = (1ULL << shift) - 1;
4751 
4752 	adapter->rx_tstamp_tc.cc_mask = mask;
4753 	adapter->rx_tstamp_tc.cc_shift = shift;
4754 	adapter->rx_tstamp_tc.nsec_mask = (1ULL << shift) - 1;
4755 
4756 	adapter->tx_tstamp_tc.cc_mask = mask;
4757 	adapter->tx_tstamp_tc.cc_shift = shift;
4758 	adapter->tx_tstamp_tc.nsec_mask = (1ULL << shift) - 1;
4759 }
4760 
4761 static int
4762 igb_timesync_adjust_time(struct rte_eth_dev *dev, int64_t delta)
4763 {
4764 	struct e1000_adapter *adapter = dev->data->dev_private;
4765 
4766 	adapter->systime_tc.nsec += delta;
4767 	adapter->rx_tstamp_tc.nsec += delta;
4768 	adapter->tx_tstamp_tc.nsec += delta;
4769 
4770 	return 0;
4771 }
4772 
4773 static int
4774 igb_timesync_write_time(struct rte_eth_dev *dev, const struct timespec *ts)
4775 {
4776 	uint64_t ns;
4777 	struct e1000_adapter *adapter = dev->data->dev_private;
4778 
4779 	ns = rte_timespec_to_ns(ts);
4780 
4781 	/* Set the timecounters to a new value. */
4782 	adapter->systime_tc.nsec = ns;
4783 	adapter->rx_tstamp_tc.nsec = ns;
4784 	adapter->tx_tstamp_tc.nsec = ns;
4785 
4786 	return 0;
4787 }
4788 
4789 static int
4790 igb_timesync_read_time(struct rte_eth_dev *dev, struct timespec *ts)
4791 {
4792 	uint64_t ns, systime_cycles;
4793 	struct e1000_adapter *adapter = dev->data->dev_private;
4794 
4795 	systime_cycles = igb_read_systime_cyclecounter(dev);
4796 	ns = rte_timecounter_update(&adapter->systime_tc, systime_cycles);
4797 	*ts = rte_ns_to_timespec(ns);
4798 
4799 	return 0;
4800 }
4801 
4802 static int
4803 igb_timesync_enable(struct rte_eth_dev *dev)
4804 {
4805 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4806 	uint32_t tsync_ctl;
4807 	uint32_t tsauxc;
4808 
4809 	/* Stop the timesync system time. */
4810 	E1000_WRITE_REG(hw, E1000_TIMINCA, 0x0);
4811 	/* Reset the timesync system time value. */
4812 	switch (hw->mac.type) {
4813 	case e1000_82580:
4814 	case e1000_i350:
4815 	case e1000_i354:
4816 	case e1000_i210:
4817 	case e1000_i211:
4818 		E1000_WRITE_REG(hw, E1000_SYSTIMR, 0x0);
4819 		/* fall-through */
4820 	case e1000_82576:
4821 		E1000_WRITE_REG(hw, E1000_SYSTIML, 0x0);
4822 		E1000_WRITE_REG(hw, E1000_SYSTIMH, 0x0);
4823 		break;
4824 	default:
4825 		/* Not supported. */
4826 		return -ENOTSUP;
4827 	}
4828 
4829 	/* Enable system time for it isn't on by default. */
4830 	tsauxc = E1000_READ_REG(hw, E1000_TSAUXC);
4831 	tsauxc &= ~E1000_TSAUXC_DISABLE_SYSTIME;
4832 	E1000_WRITE_REG(hw, E1000_TSAUXC, tsauxc);
4833 
4834 	igb_start_timecounters(dev);
4835 
4836 	/* Enable L2 filtering of IEEE1588/802.1AS Ethernet frame types. */
4837 	E1000_WRITE_REG(hw, E1000_ETQF(E1000_ETQF_FILTER_1588),
4838 			(RTE_ETHER_TYPE_1588 |
4839 			 E1000_ETQF_FILTER_ENABLE |
4840 			 E1000_ETQF_1588));
4841 
4842 	/* Enable timestamping of received PTP packets. */
4843 	tsync_ctl = E1000_READ_REG(hw, E1000_TSYNCRXCTL);
4844 	tsync_ctl |= E1000_TSYNCRXCTL_ENABLED;
4845 	E1000_WRITE_REG(hw, E1000_TSYNCRXCTL, tsync_ctl);
4846 
4847 	/* Enable Timestamping of transmitted PTP packets. */
4848 	tsync_ctl = E1000_READ_REG(hw, E1000_TSYNCTXCTL);
4849 	tsync_ctl |= E1000_TSYNCTXCTL_ENABLED;
4850 	E1000_WRITE_REG(hw, E1000_TSYNCTXCTL, tsync_ctl);
4851 
4852 	return 0;
4853 }
4854 
4855 static int
4856 igb_timesync_disable(struct rte_eth_dev *dev)
4857 {
4858 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4859 	uint32_t tsync_ctl;
4860 
4861 	/* Disable timestamping of transmitted PTP packets. */
4862 	tsync_ctl = E1000_READ_REG(hw, E1000_TSYNCTXCTL);
4863 	tsync_ctl &= ~E1000_TSYNCTXCTL_ENABLED;
4864 	E1000_WRITE_REG(hw, E1000_TSYNCTXCTL, tsync_ctl);
4865 
4866 	/* Disable timestamping of received PTP packets. */
4867 	tsync_ctl = E1000_READ_REG(hw, E1000_TSYNCRXCTL);
4868 	tsync_ctl &= ~E1000_TSYNCRXCTL_ENABLED;
4869 	E1000_WRITE_REG(hw, E1000_TSYNCRXCTL, tsync_ctl);
4870 
4871 	/* Disable L2 filtering of IEEE1588/802.1AS Ethernet frame types. */
4872 	E1000_WRITE_REG(hw, E1000_ETQF(E1000_ETQF_FILTER_1588), 0);
4873 
4874 	/* Stop incrementating the System Time registers. */
4875 	E1000_WRITE_REG(hw, E1000_TIMINCA, 0);
4876 
4877 	return 0;
4878 }
4879 
4880 static int
4881 igb_timesync_read_rx_timestamp(struct rte_eth_dev *dev,
4882 			       struct timespec *timestamp,
4883 			       uint32_t flags __rte_unused)
4884 {
4885 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4886 	struct e1000_adapter *adapter = dev->data->dev_private;
4887 	uint32_t tsync_rxctl;
4888 	uint64_t rx_tstamp_cycles;
4889 	uint64_t ns;
4890 
4891 	tsync_rxctl = E1000_READ_REG(hw, E1000_TSYNCRXCTL);
4892 	if ((tsync_rxctl & E1000_TSYNCRXCTL_VALID) == 0)
4893 		return -EINVAL;
4894 
4895 	rx_tstamp_cycles = igb_read_rx_tstamp_cyclecounter(dev);
4896 	ns = rte_timecounter_update(&adapter->rx_tstamp_tc, rx_tstamp_cycles);
4897 	*timestamp = rte_ns_to_timespec(ns);
4898 
4899 	return  0;
4900 }
4901 
4902 static int
4903 igb_timesync_read_tx_timestamp(struct rte_eth_dev *dev,
4904 			       struct timespec *timestamp)
4905 {
4906 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4907 	struct e1000_adapter *adapter = dev->data->dev_private;
4908 	uint32_t tsync_txctl;
4909 	uint64_t tx_tstamp_cycles;
4910 	uint64_t ns;
4911 
4912 	tsync_txctl = E1000_READ_REG(hw, E1000_TSYNCTXCTL);
4913 	if ((tsync_txctl & E1000_TSYNCTXCTL_VALID) == 0)
4914 		return -EINVAL;
4915 
4916 	tx_tstamp_cycles = igb_read_tx_tstamp_cyclecounter(dev);
4917 	ns = rte_timecounter_update(&adapter->tx_tstamp_tc, tx_tstamp_cycles);
4918 	*timestamp = rte_ns_to_timespec(ns);
4919 
4920 	return  0;
4921 }
4922 
4923 static int
4924 eth_igb_get_reg_length(struct rte_eth_dev *dev __rte_unused)
4925 {
4926 	int count = 0;
4927 	int g_ind = 0;
4928 	const struct reg_info *reg_group;
4929 
4930 	while ((reg_group = igb_regs[g_ind++]))
4931 		count += igb_reg_group_count(reg_group);
4932 
4933 	return count;
4934 }
4935 
4936 static int
4937 igbvf_get_reg_length(struct rte_eth_dev *dev __rte_unused)
4938 {
4939 	int count = 0;
4940 	int g_ind = 0;
4941 	const struct reg_info *reg_group;
4942 
4943 	while ((reg_group = igbvf_regs[g_ind++]))
4944 		count += igb_reg_group_count(reg_group);
4945 
4946 	return count;
4947 }
4948 
4949 static int
4950 eth_igb_get_regs(struct rte_eth_dev *dev,
4951 	struct rte_dev_reg_info *regs)
4952 {
4953 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4954 	uint32_t *data = regs->data;
4955 	int g_ind = 0;
4956 	int count = 0;
4957 	const struct reg_info *reg_group;
4958 
4959 	if (data == NULL) {
4960 		regs->length = eth_igb_get_reg_length(dev);
4961 		regs->width = sizeof(uint32_t);
4962 		return 0;
4963 	}
4964 
4965 	/* Support only full register dump */
4966 	if ((regs->length == 0) ||
4967 	    (regs->length == (uint32_t)eth_igb_get_reg_length(dev))) {
4968 		regs->version = hw->mac.type << 24 | hw->revision_id << 16 |
4969 			hw->device_id;
4970 		while ((reg_group = igb_regs[g_ind++]))
4971 			count += igb_read_regs_group(dev, &data[count],
4972 							reg_group);
4973 		return 0;
4974 	}
4975 
4976 	return -ENOTSUP;
4977 }
4978 
4979 static int
4980 igbvf_get_regs(struct rte_eth_dev *dev,
4981 	struct rte_dev_reg_info *regs)
4982 {
4983 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4984 	uint32_t *data = regs->data;
4985 	int g_ind = 0;
4986 	int count = 0;
4987 	const struct reg_info *reg_group;
4988 
4989 	if (data == NULL) {
4990 		regs->length = igbvf_get_reg_length(dev);
4991 		regs->width = sizeof(uint32_t);
4992 		return 0;
4993 	}
4994 
4995 	/* Support only full register dump */
4996 	if ((regs->length == 0) ||
4997 	    (regs->length == (uint32_t)igbvf_get_reg_length(dev))) {
4998 		regs->version = hw->mac.type << 24 | hw->revision_id << 16 |
4999 			hw->device_id;
5000 		while ((reg_group = igbvf_regs[g_ind++]))
5001 			count += igb_read_regs_group(dev, &data[count],
5002 							reg_group);
5003 		return 0;
5004 	}
5005 
5006 	return -ENOTSUP;
5007 }
5008 
5009 static int
5010 eth_igb_get_eeprom_length(struct rte_eth_dev *dev)
5011 {
5012 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5013 
5014 	/* Return unit is byte count */
5015 	return hw->nvm.word_size * 2;
5016 }
5017 
5018 static int
5019 eth_igb_get_eeprom(struct rte_eth_dev *dev,
5020 	struct rte_dev_eeprom_info *in_eeprom)
5021 {
5022 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5023 	struct e1000_nvm_info *nvm = &hw->nvm;
5024 	uint16_t *data = in_eeprom->data;
5025 	int first, length;
5026 
5027 	first = in_eeprom->offset >> 1;
5028 	length = in_eeprom->length >> 1;
5029 	if ((first >= hw->nvm.word_size) ||
5030 	    ((first + length) >= hw->nvm.word_size))
5031 		return -EINVAL;
5032 
5033 	in_eeprom->magic = hw->vendor_id |
5034 		((uint32_t)hw->device_id << 16);
5035 
5036 	if ((nvm->ops.read) == NULL)
5037 		return -ENOTSUP;
5038 
5039 	return nvm->ops.read(hw, first, length, data);
5040 }
5041 
5042 static int
5043 eth_igb_set_eeprom(struct rte_eth_dev *dev,
5044 	struct rte_dev_eeprom_info *in_eeprom)
5045 {
5046 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5047 	struct e1000_nvm_info *nvm = &hw->nvm;
5048 	uint16_t *data = in_eeprom->data;
5049 	int first, length;
5050 
5051 	first = in_eeprom->offset >> 1;
5052 	length = in_eeprom->length >> 1;
5053 	if ((first >= hw->nvm.word_size) ||
5054 	    ((first + length) >= hw->nvm.word_size))
5055 		return -EINVAL;
5056 
5057 	in_eeprom->magic = (uint32_t)hw->vendor_id |
5058 		((uint32_t)hw->device_id << 16);
5059 
5060 	if ((nvm->ops.write) == NULL)
5061 		return -ENOTSUP;
5062 	return nvm->ops.write(hw,  first, length, data);
5063 }
5064 
5065 static int
5066 eth_igb_get_module_info(struct rte_eth_dev *dev,
5067 			struct rte_eth_dev_module_info *modinfo)
5068 {
5069 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5070 
5071 	uint32_t status = 0;
5072 	uint16_t sff8472_rev, addr_mode;
5073 	bool page_swap = false;
5074 
5075 	if (hw->phy.media_type == e1000_media_type_copper ||
5076 	    hw->phy.media_type == e1000_media_type_unknown)
5077 		return -EOPNOTSUPP;
5078 
5079 	/* Check whether we support SFF-8472 or not */
5080 	status = e1000_read_phy_reg_i2c(hw, IGB_SFF_8472_COMP, &sff8472_rev);
5081 	if (status)
5082 		return -EIO;
5083 
5084 	/* addressing mode is not supported */
5085 	status = e1000_read_phy_reg_i2c(hw, IGB_SFF_8472_SWAP, &addr_mode);
5086 	if (status)
5087 		return -EIO;
5088 
5089 	/* addressing mode is not supported */
5090 	if ((addr_mode & 0xFF) & IGB_SFF_ADDRESSING_MODE) {
5091 		PMD_DRV_LOG(ERR,
5092 			    "Address change required to access page 0xA2, "
5093 			    "but not supported. Please report the module "
5094 			    "type to the driver maintainers.\n");
5095 		page_swap = true;
5096 	}
5097 
5098 	if ((sff8472_rev & 0xFF) == IGB_SFF_8472_UNSUP || page_swap) {
5099 		/* We have an SFP, but it does not support SFF-8472 */
5100 		modinfo->type = RTE_ETH_MODULE_SFF_8079;
5101 		modinfo->eeprom_len = RTE_ETH_MODULE_SFF_8079_LEN;
5102 	} else {
5103 		/* We have an SFP which supports a revision of SFF-8472 */
5104 		modinfo->type = RTE_ETH_MODULE_SFF_8472;
5105 		modinfo->eeprom_len = RTE_ETH_MODULE_SFF_8472_LEN;
5106 	}
5107 
5108 	return 0;
5109 }
5110 
5111 static int
5112 eth_igb_get_module_eeprom(struct rte_eth_dev *dev,
5113 			  struct rte_dev_eeprom_info *info)
5114 {
5115 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5116 
5117 	uint32_t status = 0;
5118 	uint16_t dataword[RTE_ETH_MODULE_SFF_8472_LEN / 2 + 1];
5119 	u16 first_word, last_word;
5120 	int i = 0;
5121 
5122 	first_word = info->offset >> 1;
5123 	last_word = (info->offset + info->length - 1) >> 1;
5124 
5125 	/* Read EEPROM block, SFF-8079/SFF-8472, word at a time */
5126 	for (i = 0; i < last_word - first_word + 1; i++) {
5127 		status = e1000_read_phy_reg_i2c(hw, (first_word + i) * 2,
5128 						&dataword[i]);
5129 		if (status) {
5130 			/* Error occurred while reading module */
5131 			return -EIO;
5132 		}
5133 
5134 		dataword[i] = rte_be_to_cpu_16(dataword[i]);
5135 	}
5136 
5137 	memcpy(info->data, (u8 *)dataword + (info->offset & 1), info->length);
5138 
5139 	return 0;
5140 }
5141 
5142 static int
5143 eth_igb_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
5144 {
5145 	struct e1000_hw *hw =
5146 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5147 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
5148 	struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
5149 	uint32_t vec = E1000_MISC_VEC_ID;
5150 
5151 	if (rte_intr_allow_others(intr_handle))
5152 		vec = E1000_RX_VEC_START;
5153 
5154 	uint32_t mask = 1 << (queue_id + vec);
5155 
5156 	E1000_WRITE_REG(hw, E1000_EIMC, mask);
5157 	E1000_WRITE_FLUSH(hw);
5158 
5159 	return 0;
5160 }
5161 
5162 static int
5163 eth_igb_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
5164 {
5165 	struct e1000_hw *hw =
5166 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5167 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
5168 	struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
5169 	uint32_t vec = E1000_MISC_VEC_ID;
5170 
5171 	if (rte_intr_allow_others(intr_handle))
5172 		vec = E1000_RX_VEC_START;
5173 
5174 	uint32_t mask = 1 << (queue_id + vec);
5175 	uint32_t regval;
5176 
5177 	regval = E1000_READ_REG(hw, E1000_EIMS);
5178 	E1000_WRITE_REG(hw, E1000_EIMS, regval | mask);
5179 	E1000_WRITE_FLUSH(hw);
5180 
5181 	rte_intr_ack(intr_handle);
5182 
5183 	return 0;
5184 }
5185 
5186 static void
5187 eth_igb_write_ivar(struct e1000_hw *hw, uint8_t  msix_vector,
5188 		   uint8_t index, uint8_t offset)
5189 {
5190 	uint32_t val = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
5191 
5192 	/* clear bits */
5193 	val &= ~((uint32_t)0xFF << offset);
5194 
5195 	/* write vector and valid bit */
5196 	val |= (msix_vector | E1000_IVAR_VALID) << offset;
5197 
5198 	E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, val);
5199 }
5200 
5201 static void
5202 eth_igb_assign_msix_vector(struct e1000_hw *hw, int8_t direction,
5203 			   uint8_t queue, uint8_t msix_vector)
5204 {
5205 	uint32_t tmp = 0;
5206 
5207 	if (hw->mac.type == e1000_82575) {
5208 		if (direction == 0)
5209 			tmp = E1000_EICR_RX_QUEUE0 << queue;
5210 		else if (direction == 1)
5211 			tmp = E1000_EICR_TX_QUEUE0 << queue;
5212 		E1000_WRITE_REG(hw, E1000_MSIXBM(msix_vector), tmp);
5213 	} else if (hw->mac.type == e1000_82576) {
5214 		if ((direction == 0) || (direction == 1))
5215 			eth_igb_write_ivar(hw, msix_vector, queue & 0x7,
5216 					   ((queue & 0x8) << 1) +
5217 					   8 * direction);
5218 	} else if ((hw->mac.type == e1000_82580) ||
5219 			(hw->mac.type == e1000_i350) ||
5220 			(hw->mac.type == e1000_i354) ||
5221 			(hw->mac.type == e1000_i210) ||
5222 			(hw->mac.type == e1000_i211)) {
5223 		if ((direction == 0) || (direction == 1))
5224 			eth_igb_write_ivar(hw, msix_vector,
5225 					   queue >> 1,
5226 					   ((queue & 0x1) << 4) +
5227 					   8 * direction);
5228 	}
5229 }
5230 
5231 /* Sets up the hardware to generate MSI-X interrupts properly
5232  * @hw
5233  *  board private structure
5234  */
5235 static void
5236 eth_igb_configure_msix_intr(struct rte_eth_dev *dev)
5237 {
5238 	int queue_id;
5239 	uint32_t tmpval, regval, intr_mask;
5240 	struct e1000_hw *hw =
5241 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5242 	uint32_t vec = E1000_MISC_VEC_ID;
5243 	uint32_t base = E1000_MISC_VEC_ID;
5244 	uint32_t misc_shift = 0;
5245 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
5246 	struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
5247 
5248 	/* won't configure msix register if no mapping is done
5249 	 * between intr vector and event fd
5250 	 */
5251 	if (!rte_intr_dp_is_en(intr_handle))
5252 		return;
5253 
5254 	if (rte_intr_allow_others(intr_handle)) {
5255 		vec = base = E1000_RX_VEC_START;
5256 		misc_shift = 1;
5257 	}
5258 
5259 	/* set interrupt vector for other causes */
5260 	if (hw->mac.type == e1000_82575) {
5261 		tmpval = E1000_READ_REG(hw, E1000_CTRL_EXT);
5262 		/* enable MSI-X PBA support */
5263 		tmpval |= E1000_CTRL_EXT_PBA_CLR;
5264 
5265 		/* Auto-Mask interrupts upon ICR read */
5266 		tmpval |= E1000_CTRL_EXT_EIAME;
5267 		tmpval |= E1000_CTRL_EXT_IRCA;
5268 
5269 		E1000_WRITE_REG(hw, E1000_CTRL_EXT, tmpval);
5270 
5271 		/* enable msix_other interrupt */
5272 		E1000_WRITE_REG_ARRAY(hw, E1000_MSIXBM(0), 0, E1000_EIMS_OTHER);
5273 		regval = E1000_READ_REG(hw, E1000_EIAC);
5274 		E1000_WRITE_REG(hw, E1000_EIAC, regval | E1000_EIMS_OTHER);
5275 		regval = E1000_READ_REG(hw, E1000_EIAM);
5276 		E1000_WRITE_REG(hw, E1000_EIMS, regval | E1000_EIMS_OTHER);
5277 	} else if ((hw->mac.type == e1000_82576) ||
5278 			(hw->mac.type == e1000_82580) ||
5279 			(hw->mac.type == e1000_i350) ||
5280 			(hw->mac.type == e1000_i354) ||
5281 			(hw->mac.type == e1000_i210) ||
5282 			(hw->mac.type == e1000_i211)) {
5283 		/* turn on MSI-X capability first */
5284 		E1000_WRITE_REG(hw, E1000_GPIE, E1000_GPIE_MSIX_MODE |
5285 					E1000_GPIE_PBA | E1000_GPIE_EIAME |
5286 					E1000_GPIE_NSICR);
5287 		intr_mask = RTE_LEN2MASK(intr_handle->nb_efd, uint32_t) <<
5288 			misc_shift;
5289 
5290 		if (dev->data->dev_conf.intr_conf.lsc != 0)
5291 			intr_mask |= (1 << IGB_MSIX_OTHER_INTR_VEC);
5292 
5293 		regval = E1000_READ_REG(hw, E1000_EIAC);
5294 		E1000_WRITE_REG(hw, E1000_EIAC, regval | intr_mask);
5295 
5296 		/* enable msix_other interrupt */
5297 		regval = E1000_READ_REG(hw, E1000_EIMS);
5298 		E1000_WRITE_REG(hw, E1000_EIMS, regval | intr_mask);
5299 		tmpval = (IGB_MSIX_OTHER_INTR_VEC | E1000_IVAR_VALID) << 8;
5300 		E1000_WRITE_REG(hw, E1000_IVAR_MISC, tmpval);
5301 	}
5302 
5303 	/* use EIAM to auto-mask when MSI-X interrupt
5304 	 * is asserted, this saves a register write for every interrupt
5305 	 */
5306 	intr_mask = RTE_LEN2MASK(intr_handle->nb_efd, uint32_t) <<
5307 		misc_shift;
5308 
5309 	if (dev->data->dev_conf.intr_conf.lsc != 0)
5310 		intr_mask |= (1 << IGB_MSIX_OTHER_INTR_VEC);
5311 
5312 	regval = E1000_READ_REG(hw, E1000_EIAM);
5313 	E1000_WRITE_REG(hw, E1000_EIAM, regval | intr_mask);
5314 
5315 	for (queue_id = 0; queue_id < dev->data->nb_rx_queues; queue_id++) {
5316 		eth_igb_assign_msix_vector(hw, 0, queue_id, vec);
5317 		intr_handle->intr_vec[queue_id] = vec;
5318 		if (vec < base + intr_handle->nb_efd - 1)
5319 			vec++;
5320 	}
5321 
5322 	E1000_WRITE_FLUSH(hw);
5323 }
5324 
5325 /* restore n-tuple filter */
5326 static inline void
5327 igb_ntuple_filter_restore(struct rte_eth_dev *dev)
5328 {
5329 	struct e1000_filter_info *filter_info =
5330 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
5331 	struct e1000_5tuple_filter *p_5tuple;
5332 	struct e1000_2tuple_filter *p_2tuple;
5333 
5334 	TAILQ_FOREACH(p_5tuple, &filter_info->fivetuple_list, entries) {
5335 		igb_inject_5tuple_filter_82576(dev, p_5tuple);
5336 	}
5337 
5338 	TAILQ_FOREACH(p_2tuple, &filter_info->twotuple_list, entries) {
5339 		igb_inject_2uple_filter(dev, p_2tuple);
5340 	}
5341 }
5342 
5343 /* restore SYN filter */
5344 static inline void
5345 igb_syn_filter_restore(struct rte_eth_dev *dev)
5346 {
5347 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5348 	struct e1000_filter_info *filter_info =
5349 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
5350 	uint32_t synqf;
5351 
5352 	synqf = filter_info->syn_info;
5353 
5354 	if (synqf & E1000_SYN_FILTER_ENABLE) {
5355 		E1000_WRITE_REG(hw, E1000_SYNQF(0), synqf);
5356 		E1000_WRITE_FLUSH(hw);
5357 	}
5358 }
5359 
5360 /* restore ethernet type filter */
5361 static inline void
5362 igb_ethertype_filter_restore(struct rte_eth_dev *dev)
5363 {
5364 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5365 	struct e1000_filter_info *filter_info =
5366 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
5367 	int i;
5368 
5369 	for (i = 0; i < E1000_MAX_ETQF_FILTERS; i++) {
5370 		if (filter_info->ethertype_mask & (1 << i)) {
5371 			E1000_WRITE_REG(hw, E1000_ETQF(i),
5372 				filter_info->ethertype_filters[i].etqf);
5373 			E1000_WRITE_FLUSH(hw);
5374 		}
5375 	}
5376 }
5377 
5378 /* restore flex byte filter */
5379 static inline void
5380 igb_flex_filter_restore(struct rte_eth_dev *dev)
5381 {
5382 	struct e1000_filter_info *filter_info =
5383 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
5384 	struct e1000_flex_filter *flex_filter;
5385 
5386 	TAILQ_FOREACH(flex_filter, &filter_info->flex_list, entries) {
5387 		igb_inject_flex_filter(dev, flex_filter);
5388 	}
5389 }
5390 
5391 /* restore rss filter */
5392 static inline void
5393 igb_rss_filter_restore(struct rte_eth_dev *dev)
5394 {
5395 	struct e1000_filter_info *filter_info =
5396 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
5397 
5398 	if (filter_info->rss_info.conf.queue_num)
5399 		igb_config_rss_filter(dev, &filter_info->rss_info, TRUE);
5400 }
5401 
5402 /* restore all types filter */
5403 static int
5404 igb_filter_restore(struct rte_eth_dev *dev)
5405 {
5406 	igb_ntuple_filter_restore(dev);
5407 	igb_ethertype_filter_restore(dev);
5408 	igb_syn_filter_restore(dev);
5409 	igb_flex_filter_restore(dev);
5410 	igb_rss_filter_restore(dev);
5411 
5412 	return 0;
5413 }
5414 
5415 RTE_PMD_REGISTER_PCI(net_e1000_igb, rte_igb_pmd);
5416 RTE_PMD_REGISTER_PCI_TABLE(net_e1000_igb, pci_id_igb_map);
5417 RTE_PMD_REGISTER_KMOD_DEP(net_e1000_igb, "* igb_uio | uio_pci_generic | vfio-pci");
5418 RTE_PMD_REGISTER_PCI(net_e1000_igb_vf, rte_igbvf_pmd);
5419 RTE_PMD_REGISTER_PCI_TABLE(net_e1000_igb_vf, pci_id_igbvf_map);
5420 RTE_PMD_REGISTER_KMOD_DEP(net_e1000_igb_vf, "* igb_uio | vfio-pci");
5421