xref: /f-stack/dpdk/drivers/net/e1000/igb_ethdev.c (revision fa64a7ff)
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_common.h>
12 #include <rte_interrupts.h>
13 #include <rte_byteorder.h>
14 #include <rte_log.h>
15 #include <rte_debug.h>
16 #include <rte_pci.h>
17 #include <rte_bus_pci.h>
18 #include <rte_ether.h>
19 #include <rte_ethdev_driver.h>
20 #include <rte_ethdev_pci.h>
21 #include <rte_memory.h>
22 #include <rte_eal.h>
23 #include <rte_malloc.h>
24 #include <rte_dev.h>
25 
26 #include "e1000_logs.h"
27 #include "base/e1000_api.h"
28 #include "e1000_ethdev.h"
29 #include "igb_regs.h"
30 
31 /*
32  * Default values for port configuration
33  */
34 #define IGB_DEFAULT_RX_FREE_THRESH  32
35 
36 #define IGB_DEFAULT_RX_PTHRESH      ((hw->mac.type == e1000_i354) ? 12 : 8)
37 #define IGB_DEFAULT_RX_HTHRESH      8
38 #define IGB_DEFAULT_RX_WTHRESH      ((hw->mac.type == e1000_82576) ? 1 : 4)
39 
40 #define IGB_DEFAULT_TX_PTHRESH      ((hw->mac.type == e1000_i354) ? 20 : 8)
41 #define IGB_DEFAULT_TX_HTHRESH      1
42 #define IGB_DEFAULT_TX_WTHRESH      ((hw->mac.type == e1000_82576) ? 1 : 16)
43 
44 /* Bit shift and mask */
45 #define IGB_4_BIT_WIDTH  (CHAR_BIT / 2)
46 #define IGB_4_BIT_MASK   RTE_LEN2MASK(IGB_4_BIT_WIDTH, uint8_t)
47 #define IGB_8_BIT_WIDTH  CHAR_BIT
48 #define IGB_8_BIT_MASK   UINT8_MAX
49 
50 /* Additional timesync values. */
51 #define E1000_CYCLECOUNTER_MASK      0xffffffffffffffffULL
52 #define E1000_ETQF_FILTER_1588       3
53 #define IGB_82576_TSYNC_SHIFT        16
54 #define E1000_INCPERIOD_82576        (1 << E1000_TIMINCA_16NS_SHIFT)
55 #define E1000_INCVALUE_82576         (16 << IGB_82576_TSYNC_SHIFT)
56 #define E1000_TSAUXC_DISABLE_SYSTIME 0x80000000
57 
58 #define E1000_VTIVAR_MISC                0x01740
59 #define E1000_VTIVAR_MISC_MASK           0xFF
60 #define E1000_VTIVAR_VALID               0x80
61 #define E1000_VTIVAR_MISC_MAILBOX        0
62 #define E1000_VTIVAR_MISC_INTR_MASK      0x3
63 
64 /* External VLAN Enable bit mask */
65 #define E1000_CTRL_EXT_EXT_VLAN      (1 << 26)
66 
67 /* External VLAN Ether Type bit mask and shift */
68 #define E1000_VET_VET_EXT            0xFFFF0000
69 #define E1000_VET_VET_EXT_SHIFT      16
70 
71 /* MSI-X other interrupt vector */
72 #define IGB_MSIX_OTHER_INTR_VEC      0
73 
74 static int  eth_igb_configure(struct rte_eth_dev *dev);
75 static int  eth_igb_start(struct rte_eth_dev *dev);
76 static void eth_igb_stop(struct rte_eth_dev *dev);
77 static int  eth_igb_dev_set_link_up(struct rte_eth_dev *dev);
78 static int  eth_igb_dev_set_link_down(struct rte_eth_dev *dev);
79 static void eth_igb_close(struct rte_eth_dev *dev);
80 static int eth_igb_reset(struct rte_eth_dev *dev);
81 static void eth_igb_promiscuous_enable(struct rte_eth_dev *dev);
82 static void eth_igb_promiscuous_disable(struct rte_eth_dev *dev);
83 static void eth_igb_allmulticast_enable(struct rte_eth_dev *dev);
84 static void eth_igb_allmulticast_disable(struct rte_eth_dev *dev);
85 static int  eth_igb_link_update(struct rte_eth_dev *dev,
86 				int wait_to_complete);
87 static int eth_igb_stats_get(struct rte_eth_dev *dev,
88 				struct rte_eth_stats *rte_stats);
89 static int eth_igb_xstats_get(struct rte_eth_dev *dev,
90 			      struct rte_eth_xstat *xstats, unsigned n);
91 static int eth_igb_xstats_get_by_id(struct rte_eth_dev *dev,
92 		const uint64_t *ids,
93 		uint64_t *values, unsigned int n);
94 static int eth_igb_xstats_get_names(struct rte_eth_dev *dev,
95 				    struct rte_eth_xstat_name *xstats_names,
96 				    unsigned int size);
97 static int eth_igb_xstats_get_names_by_id(struct rte_eth_dev *dev,
98 		struct rte_eth_xstat_name *xstats_names, const uint64_t *ids,
99 		unsigned int limit);
100 static void eth_igb_stats_reset(struct rte_eth_dev *dev);
101 static void eth_igb_xstats_reset(struct rte_eth_dev *dev);
102 static int eth_igb_fw_version_get(struct rte_eth_dev *dev,
103 				   char *fw_version, size_t fw_size);
104 static void eth_igb_infos_get(struct rte_eth_dev *dev,
105 			      struct rte_eth_dev_info *dev_info);
106 static const uint32_t *eth_igb_supported_ptypes_get(struct rte_eth_dev *dev);
107 static void eth_igbvf_infos_get(struct rte_eth_dev *dev,
108 				struct rte_eth_dev_info *dev_info);
109 static int  eth_igb_flow_ctrl_get(struct rte_eth_dev *dev,
110 				struct rte_eth_fc_conf *fc_conf);
111 static int  eth_igb_flow_ctrl_set(struct rte_eth_dev *dev,
112 				struct rte_eth_fc_conf *fc_conf);
113 static int eth_igb_lsc_interrupt_setup(struct rte_eth_dev *dev, uint8_t on);
114 static int eth_igb_rxq_interrupt_setup(struct rte_eth_dev *dev);
115 static int eth_igb_interrupt_get_status(struct rte_eth_dev *dev);
116 static int eth_igb_interrupt_action(struct rte_eth_dev *dev,
117 				    struct rte_intr_handle *handle);
118 static void eth_igb_interrupt_handler(void *param);
119 static int  igb_hardware_init(struct e1000_hw *hw);
120 static void igb_hw_control_acquire(struct e1000_hw *hw);
121 static void igb_hw_control_release(struct e1000_hw *hw);
122 static void igb_init_manageability(struct e1000_hw *hw);
123 static void igb_release_manageability(struct e1000_hw *hw);
124 
125 static int  eth_igb_mtu_set(struct rte_eth_dev *dev, uint16_t mtu);
126 
127 static int eth_igb_vlan_filter_set(struct rte_eth_dev *dev,
128 		uint16_t vlan_id, int on);
129 static int eth_igb_vlan_tpid_set(struct rte_eth_dev *dev,
130 				 enum rte_vlan_type vlan_type,
131 				 uint16_t tpid_id);
132 static int eth_igb_vlan_offload_set(struct rte_eth_dev *dev, int mask);
133 
134 static void igb_vlan_hw_filter_enable(struct rte_eth_dev *dev);
135 static void igb_vlan_hw_filter_disable(struct rte_eth_dev *dev);
136 static void igb_vlan_hw_strip_enable(struct rte_eth_dev *dev);
137 static void igb_vlan_hw_strip_disable(struct rte_eth_dev *dev);
138 static void igb_vlan_hw_extend_enable(struct rte_eth_dev *dev);
139 static void igb_vlan_hw_extend_disable(struct rte_eth_dev *dev);
140 
141 static int eth_igb_led_on(struct rte_eth_dev *dev);
142 static int eth_igb_led_off(struct rte_eth_dev *dev);
143 
144 static void igb_intr_disable(struct rte_eth_dev *dev);
145 static int  igb_get_rx_buffer_size(struct e1000_hw *hw);
146 static int eth_igb_rar_set(struct rte_eth_dev *dev,
147 			   struct ether_addr *mac_addr,
148 			   uint32_t index, uint32_t pool);
149 static void eth_igb_rar_clear(struct rte_eth_dev *dev, uint32_t index);
150 static int eth_igb_default_mac_addr_set(struct rte_eth_dev *dev,
151 		struct ether_addr *addr);
152 
153 static void igbvf_intr_disable(struct e1000_hw *hw);
154 static int igbvf_dev_configure(struct rte_eth_dev *dev);
155 static int igbvf_dev_start(struct rte_eth_dev *dev);
156 static void igbvf_dev_stop(struct rte_eth_dev *dev);
157 static void igbvf_dev_close(struct rte_eth_dev *dev);
158 static void igbvf_promiscuous_enable(struct rte_eth_dev *dev);
159 static void igbvf_promiscuous_disable(struct rte_eth_dev *dev);
160 static void igbvf_allmulticast_enable(struct rte_eth_dev *dev);
161 static void igbvf_allmulticast_disable(struct rte_eth_dev *dev);
162 static int eth_igbvf_link_update(struct e1000_hw *hw);
163 static int eth_igbvf_stats_get(struct rte_eth_dev *dev,
164 				struct rte_eth_stats *rte_stats);
165 static int eth_igbvf_xstats_get(struct rte_eth_dev *dev,
166 				struct rte_eth_xstat *xstats, unsigned n);
167 static int eth_igbvf_xstats_get_names(struct rte_eth_dev *dev,
168 				      struct rte_eth_xstat_name *xstats_names,
169 				      unsigned limit);
170 static void eth_igbvf_stats_reset(struct rte_eth_dev *dev);
171 static int igbvf_vlan_filter_set(struct rte_eth_dev *dev,
172 		uint16_t vlan_id, int on);
173 static int igbvf_set_vfta(struct e1000_hw *hw, uint16_t vid, bool on);
174 static void igbvf_set_vfta_all(struct rte_eth_dev *dev, bool on);
175 static int igbvf_default_mac_addr_set(struct rte_eth_dev *dev,
176 		struct ether_addr *addr);
177 static int igbvf_get_reg_length(struct rte_eth_dev *dev);
178 static int igbvf_get_regs(struct rte_eth_dev *dev,
179 		struct rte_dev_reg_info *regs);
180 
181 static int eth_igb_rss_reta_update(struct rte_eth_dev *dev,
182 				   struct rte_eth_rss_reta_entry64 *reta_conf,
183 				   uint16_t reta_size);
184 static int eth_igb_rss_reta_query(struct rte_eth_dev *dev,
185 				  struct rte_eth_rss_reta_entry64 *reta_conf,
186 				  uint16_t reta_size);
187 
188 static int eth_igb_syn_filter_get(struct rte_eth_dev *dev,
189 			struct rte_eth_syn_filter *filter);
190 static int eth_igb_syn_filter_handle(struct rte_eth_dev *dev,
191 			enum rte_filter_op filter_op,
192 			void *arg);
193 static int igb_add_2tuple_filter(struct rte_eth_dev *dev,
194 			struct rte_eth_ntuple_filter *ntuple_filter);
195 static int igb_remove_2tuple_filter(struct rte_eth_dev *dev,
196 			struct rte_eth_ntuple_filter *ntuple_filter);
197 static int eth_igb_get_flex_filter(struct rte_eth_dev *dev,
198 			struct rte_eth_flex_filter *filter);
199 static int eth_igb_flex_filter_handle(struct rte_eth_dev *dev,
200 			enum rte_filter_op filter_op,
201 			void *arg);
202 static int igb_add_5tuple_filter_82576(struct rte_eth_dev *dev,
203 			struct rte_eth_ntuple_filter *ntuple_filter);
204 static int igb_remove_5tuple_filter_82576(struct rte_eth_dev *dev,
205 			struct rte_eth_ntuple_filter *ntuple_filter);
206 static int igb_get_ntuple_filter(struct rte_eth_dev *dev,
207 			struct rte_eth_ntuple_filter *filter);
208 static int igb_ntuple_filter_handle(struct rte_eth_dev *dev,
209 				enum rte_filter_op filter_op,
210 				void *arg);
211 static int igb_ethertype_filter_handle(struct rte_eth_dev *dev,
212 				enum rte_filter_op filter_op,
213 				void *arg);
214 static int igb_get_ethertype_filter(struct rte_eth_dev *dev,
215 			struct rte_eth_ethertype_filter *filter);
216 static int eth_igb_filter_ctrl(struct rte_eth_dev *dev,
217 		     enum rte_filter_type filter_type,
218 		     enum rte_filter_op filter_op,
219 		     void *arg);
220 static int eth_igb_get_reg_length(struct rte_eth_dev *dev);
221 static int eth_igb_get_regs(struct rte_eth_dev *dev,
222 		struct rte_dev_reg_info *regs);
223 static int eth_igb_get_eeprom_length(struct rte_eth_dev *dev);
224 static int eth_igb_get_eeprom(struct rte_eth_dev *dev,
225 		struct rte_dev_eeprom_info *eeprom);
226 static int eth_igb_set_eeprom(struct rte_eth_dev *dev,
227 		struct rte_dev_eeprom_info *eeprom);
228 static int eth_igb_get_module_info(struct rte_eth_dev *dev,
229 				   struct rte_eth_dev_module_info *modinfo);
230 static int eth_igb_get_module_eeprom(struct rte_eth_dev *dev,
231 				     struct rte_dev_eeprom_info *info);
232 static int eth_igb_set_mc_addr_list(struct rte_eth_dev *dev,
233 				    struct ether_addr *mc_addr_set,
234 				    uint32_t nb_mc_addr);
235 static int igb_timesync_enable(struct rte_eth_dev *dev);
236 static int igb_timesync_disable(struct rte_eth_dev *dev);
237 static int igb_timesync_read_rx_timestamp(struct rte_eth_dev *dev,
238 					  struct timespec *timestamp,
239 					  uint32_t flags);
240 static int igb_timesync_read_tx_timestamp(struct rte_eth_dev *dev,
241 					  struct timespec *timestamp);
242 static int igb_timesync_adjust_time(struct rte_eth_dev *dev, int64_t delta);
243 static int igb_timesync_read_time(struct rte_eth_dev *dev,
244 				  struct timespec *timestamp);
245 static int igb_timesync_write_time(struct rte_eth_dev *dev,
246 				   const struct timespec *timestamp);
247 static int eth_igb_rx_queue_intr_enable(struct rte_eth_dev *dev,
248 					uint16_t queue_id);
249 static int eth_igb_rx_queue_intr_disable(struct rte_eth_dev *dev,
250 					 uint16_t queue_id);
251 static void eth_igb_assign_msix_vector(struct e1000_hw *hw, int8_t direction,
252 				       uint8_t queue, uint8_t msix_vector);
253 static void eth_igb_write_ivar(struct e1000_hw *hw, uint8_t msix_vector,
254 			       uint8_t index, uint8_t offset);
255 static void eth_igb_configure_msix_intr(struct rte_eth_dev *dev);
256 static void eth_igbvf_interrupt_handler(void *param);
257 static void igbvf_mbx_process(struct rte_eth_dev *dev);
258 static int igb_filter_restore(struct rte_eth_dev *dev);
259 
260 /*
261  * Define VF Stats MACRO for Non "cleared on read" register
262  */
263 #define UPDATE_VF_STAT(reg, last, cur)            \
264 {                                                 \
265 	u32 latest = E1000_READ_REG(hw, reg);     \
266 	cur += (latest - last) & UINT_MAX;        \
267 	last = latest;                            \
268 }
269 
270 #define IGB_FC_PAUSE_TIME 0x0680
271 #define IGB_LINK_UPDATE_CHECK_TIMEOUT  90  /* 9s */
272 #define IGB_LINK_UPDATE_CHECK_INTERVAL 100 /* ms */
273 
274 #define IGBVF_PMD_NAME "rte_igbvf_pmd"     /* PMD name */
275 
276 static enum e1000_fc_mode igb_fc_setting = e1000_fc_full;
277 
278 /*
279  * The set of PCI devices this driver supports
280  */
281 static const struct rte_pci_id pci_id_igb_map[] = {
282 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576) },
283 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576_FIBER) },
284 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576_SERDES) },
285 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576_QUAD_COPPER) },
286 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576_QUAD_COPPER_ET2) },
287 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576_NS) },
288 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576_NS_SERDES) },
289 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576_SERDES_QUAD) },
290 
291 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82575EB_COPPER) },
292 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82575EB_FIBER_SERDES) },
293 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82575GB_QUAD_COPPER) },
294 
295 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82580_COPPER) },
296 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82580_FIBER) },
297 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82580_SERDES) },
298 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82580_SGMII) },
299 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82580_COPPER_DUAL) },
300 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82580_QUAD_FIBER) },
301 
302 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I350_COPPER) },
303 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I350_FIBER) },
304 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I350_SERDES) },
305 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I350_SGMII) },
306 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I350_DA4) },
307 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I210_COPPER) },
308 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I210_COPPER_OEM1) },
309 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I210_COPPER_IT) },
310 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I210_FIBER) },
311 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I210_SERDES) },
312 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I210_SGMII) },
313 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I210_COPPER_FLASHLESS) },
314 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I210_SERDES_FLASHLESS) },
315 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I211_COPPER) },
316 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I354_BACKPLANE_1GBPS) },
317 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I354_SGMII) },
318 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I354_BACKPLANE_2_5GBPS) },
319 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_DH89XXCC_SGMII) },
320 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_DH89XXCC_SERDES) },
321 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_DH89XXCC_BACKPLANE) },
322 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_DH89XXCC_SFP) },
323 	{ .vendor_id = 0, /* sentinel */ },
324 };
325 
326 /*
327  * The set of PCI devices this driver supports (for 82576&I350 VF)
328  */
329 static const struct rte_pci_id pci_id_igbvf_map[] = {
330 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576_VF) },
331 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576_VF_HV) },
332 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I350_VF) },
333 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I350_VF_HV) },
334 	{ .vendor_id = 0, /* sentinel */ },
335 };
336 
337 static const struct rte_eth_desc_lim rx_desc_lim = {
338 	.nb_max = E1000_MAX_RING_DESC,
339 	.nb_min = E1000_MIN_RING_DESC,
340 	.nb_align = IGB_RXD_ALIGN,
341 };
342 
343 static const struct rte_eth_desc_lim tx_desc_lim = {
344 	.nb_max = E1000_MAX_RING_DESC,
345 	.nb_min = E1000_MIN_RING_DESC,
346 	.nb_align = IGB_RXD_ALIGN,
347 	.nb_seg_max = IGB_TX_MAX_SEG,
348 	.nb_mtu_seg_max = IGB_TX_MAX_MTU_SEG,
349 };
350 
351 static const struct eth_dev_ops eth_igb_ops = {
352 	.dev_configure        = eth_igb_configure,
353 	.dev_start            = eth_igb_start,
354 	.dev_stop             = eth_igb_stop,
355 	.dev_set_link_up      = eth_igb_dev_set_link_up,
356 	.dev_set_link_down    = eth_igb_dev_set_link_down,
357 	.dev_close            = eth_igb_close,
358 	.dev_reset            = eth_igb_reset,
359 	.promiscuous_enable   = eth_igb_promiscuous_enable,
360 	.promiscuous_disable  = eth_igb_promiscuous_disable,
361 	.allmulticast_enable  = eth_igb_allmulticast_enable,
362 	.allmulticast_disable = eth_igb_allmulticast_disable,
363 	.link_update          = eth_igb_link_update,
364 	.stats_get            = eth_igb_stats_get,
365 	.xstats_get           = eth_igb_xstats_get,
366 	.xstats_get_by_id     = eth_igb_xstats_get_by_id,
367 	.xstats_get_names_by_id = eth_igb_xstats_get_names_by_id,
368 	.xstats_get_names     = eth_igb_xstats_get_names,
369 	.stats_reset          = eth_igb_stats_reset,
370 	.xstats_reset         = eth_igb_xstats_reset,
371 	.fw_version_get       = eth_igb_fw_version_get,
372 	.dev_infos_get        = eth_igb_infos_get,
373 	.dev_supported_ptypes_get = eth_igb_supported_ptypes_get,
374 	.mtu_set              = eth_igb_mtu_set,
375 	.vlan_filter_set      = eth_igb_vlan_filter_set,
376 	.vlan_tpid_set        = eth_igb_vlan_tpid_set,
377 	.vlan_offload_set     = eth_igb_vlan_offload_set,
378 	.rx_queue_setup       = eth_igb_rx_queue_setup,
379 	.rx_queue_intr_enable = eth_igb_rx_queue_intr_enable,
380 	.rx_queue_intr_disable = eth_igb_rx_queue_intr_disable,
381 	.rx_queue_release     = eth_igb_rx_queue_release,
382 	.rx_queue_count       = eth_igb_rx_queue_count,
383 	.rx_descriptor_done   = eth_igb_rx_descriptor_done,
384 	.rx_descriptor_status = eth_igb_rx_descriptor_status,
385 	.tx_descriptor_status = eth_igb_tx_descriptor_status,
386 	.tx_queue_setup       = eth_igb_tx_queue_setup,
387 	.tx_queue_release     = eth_igb_tx_queue_release,
388 	.tx_done_cleanup      = eth_igb_tx_done_cleanup,
389 	.dev_led_on           = eth_igb_led_on,
390 	.dev_led_off          = eth_igb_led_off,
391 	.flow_ctrl_get        = eth_igb_flow_ctrl_get,
392 	.flow_ctrl_set        = eth_igb_flow_ctrl_set,
393 	.mac_addr_add         = eth_igb_rar_set,
394 	.mac_addr_remove      = eth_igb_rar_clear,
395 	.mac_addr_set         = eth_igb_default_mac_addr_set,
396 	.reta_update          = eth_igb_rss_reta_update,
397 	.reta_query           = eth_igb_rss_reta_query,
398 	.rss_hash_update      = eth_igb_rss_hash_update,
399 	.rss_hash_conf_get    = eth_igb_rss_hash_conf_get,
400 	.filter_ctrl          = eth_igb_filter_ctrl,
401 	.set_mc_addr_list     = eth_igb_set_mc_addr_list,
402 	.rxq_info_get         = igb_rxq_info_get,
403 	.txq_info_get         = igb_txq_info_get,
404 	.timesync_enable      = igb_timesync_enable,
405 	.timesync_disable     = igb_timesync_disable,
406 	.timesync_read_rx_timestamp = igb_timesync_read_rx_timestamp,
407 	.timesync_read_tx_timestamp = igb_timesync_read_tx_timestamp,
408 	.get_reg              = eth_igb_get_regs,
409 	.get_eeprom_length    = eth_igb_get_eeprom_length,
410 	.get_eeprom           = eth_igb_get_eeprom,
411 	.set_eeprom           = eth_igb_set_eeprom,
412 	.get_module_info      = eth_igb_get_module_info,
413 	.get_module_eeprom    = eth_igb_get_module_eeprom,
414 	.timesync_adjust_time = igb_timesync_adjust_time,
415 	.timesync_read_time   = igb_timesync_read_time,
416 	.timesync_write_time  = igb_timesync_write_time,
417 };
418 
419 /*
420  * dev_ops for virtual function, bare necessities for basic vf
421  * operation have been implemented
422  */
423 static const struct eth_dev_ops igbvf_eth_dev_ops = {
424 	.dev_configure        = igbvf_dev_configure,
425 	.dev_start            = igbvf_dev_start,
426 	.dev_stop             = igbvf_dev_stop,
427 	.dev_close            = igbvf_dev_close,
428 	.promiscuous_enable   = igbvf_promiscuous_enable,
429 	.promiscuous_disable  = igbvf_promiscuous_disable,
430 	.allmulticast_enable  = igbvf_allmulticast_enable,
431 	.allmulticast_disable = igbvf_allmulticast_disable,
432 	.link_update          = eth_igb_link_update,
433 	.stats_get            = eth_igbvf_stats_get,
434 	.xstats_get           = eth_igbvf_xstats_get,
435 	.xstats_get_names     = eth_igbvf_xstats_get_names,
436 	.stats_reset          = eth_igbvf_stats_reset,
437 	.xstats_reset         = eth_igbvf_stats_reset,
438 	.vlan_filter_set      = igbvf_vlan_filter_set,
439 	.dev_infos_get        = eth_igbvf_infos_get,
440 	.dev_supported_ptypes_get = eth_igb_supported_ptypes_get,
441 	.rx_queue_setup       = eth_igb_rx_queue_setup,
442 	.rx_queue_release     = eth_igb_rx_queue_release,
443 	.rx_descriptor_done   = eth_igb_rx_descriptor_done,
444 	.rx_descriptor_status = eth_igb_rx_descriptor_status,
445 	.tx_descriptor_status = eth_igb_tx_descriptor_status,
446 	.tx_queue_setup       = eth_igb_tx_queue_setup,
447 	.tx_queue_release     = eth_igb_tx_queue_release,
448 	.set_mc_addr_list     = eth_igb_set_mc_addr_list,
449 	.rxq_info_get         = igb_rxq_info_get,
450 	.txq_info_get         = igb_txq_info_get,
451 	.mac_addr_set         = igbvf_default_mac_addr_set,
452 	.get_reg              = igbvf_get_regs,
453 };
454 
455 /* store statistics names and its offset in stats structure */
456 struct rte_igb_xstats_name_off {
457 	char name[RTE_ETH_XSTATS_NAME_SIZE];
458 	unsigned offset;
459 };
460 
461 static const struct rte_igb_xstats_name_off rte_igb_stats_strings[] = {
462 	{"rx_crc_errors", offsetof(struct e1000_hw_stats, crcerrs)},
463 	{"rx_align_errors", offsetof(struct e1000_hw_stats, algnerrc)},
464 	{"rx_symbol_errors", offsetof(struct e1000_hw_stats, symerrs)},
465 	{"rx_missed_packets", offsetof(struct e1000_hw_stats, mpc)},
466 	{"tx_single_collision_packets", offsetof(struct e1000_hw_stats, scc)},
467 	{"tx_multiple_collision_packets", offsetof(struct e1000_hw_stats, mcc)},
468 	{"tx_excessive_collision_packets", offsetof(struct e1000_hw_stats,
469 		ecol)},
470 	{"tx_late_collisions", offsetof(struct e1000_hw_stats, latecol)},
471 	{"tx_total_collisions", offsetof(struct e1000_hw_stats, colc)},
472 	{"tx_deferred_packets", offsetof(struct e1000_hw_stats, dc)},
473 	{"tx_no_carrier_sense_packets", offsetof(struct e1000_hw_stats, tncrs)},
474 	{"rx_carrier_ext_errors", offsetof(struct e1000_hw_stats, cexterr)},
475 	{"rx_length_errors", offsetof(struct e1000_hw_stats, rlec)},
476 	{"rx_xon_packets", offsetof(struct e1000_hw_stats, xonrxc)},
477 	{"tx_xon_packets", offsetof(struct e1000_hw_stats, xontxc)},
478 	{"rx_xoff_packets", offsetof(struct e1000_hw_stats, xoffrxc)},
479 	{"tx_xoff_packets", offsetof(struct e1000_hw_stats, xofftxc)},
480 	{"rx_flow_control_unsupported_packets", offsetof(struct e1000_hw_stats,
481 		fcruc)},
482 	{"rx_size_64_packets", offsetof(struct e1000_hw_stats, prc64)},
483 	{"rx_size_65_to_127_packets", offsetof(struct e1000_hw_stats, prc127)},
484 	{"rx_size_128_to_255_packets", offsetof(struct e1000_hw_stats, prc255)},
485 	{"rx_size_256_to_511_packets", offsetof(struct e1000_hw_stats, prc511)},
486 	{"rx_size_512_to_1023_packets", offsetof(struct e1000_hw_stats,
487 		prc1023)},
488 	{"rx_size_1024_to_max_packets", offsetof(struct e1000_hw_stats,
489 		prc1522)},
490 	{"rx_broadcast_packets", offsetof(struct e1000_hw_stats, bprc)},
491 	{"rx_multicast_packets", offsetof(struct e1000_hw_stats, mprc)},
492 	{"rx_undersize_errors", offsetof(struct e1000_hw_stats, ruc)},
493 	{"rx_fragment_errors", offsetof(struct e1000_hw_stats, rfc)},
494 	{"rx_oversize_errors", offsetof(struct e1000_hw_stats, roc)},
495 	{"rx_jabber_errors", offsetof(struct e1000_hw_stats, rjc)},
496 	{"rx_management_packets", offsetof(struct e1000_hw_stats, mgprc)},
497 	{"rx_management_dropped", offsetof(struct e1000_hw_stats, mgpdc)},
498 	{"tx_management_packets", offsetof(struct e1000_hw_stats, mgptc)},
499 	{"rx_total_packets", offsetof(struct e1000_hw_stats, tpr)},
500 	{"tx_total_packets", offsetof(struct e1000_hw_stats, tpt)},
501 	{"rx_total_bytes", offsetof(struct e1000_hw_stats, tor)},
502 	{"tx_total_bytes", offsetof(struct e1000_hw_stats, tot)},
503 	{"tx_size_64_packets", offsetof(struct e1000_hw_stats, ptc64)},
504 	{"tx_size_65_to_127_packets", offsetof(struct e1000_hw_stats, ptc127)},
505 	{"tx_size_128_to_255_packets", offsetof(struct e1000_hw_stats, ptc255)},
506 	{"tx_size_256_to_511_packets", offsetof(struct e1000_hw_stats, ptc511)},
507 	{"tx_size_512_to_1023_packets", offsetof(struct e1000_hw_stats,
508 		ptc1023)},
509 	{"tx_size_1023_to_max_packets", offsetof(struct e1000_hw_stats,
510 		ptc1522)},
511 	{"tx_multicast_packets", offsetof(struct e1000_hw_stats, mptc)},
512 	{"tx_broadcast_packets", offsetof(struct e1000_hw_stats, bptc)},
513 	{"tx_tso_packets", offsetof(struct e1000_hw_stats, tsctc)},
514 	{"tx_tso_errors", offsetof(struct e1000_hw_stats, tsctfc)},
515 	{"rx_sent_to_host_packets", offsetof(struct e1000_hw_stats, rpthc)},
516 	{"tx_sent_by_host_packets", offsetof(struct e1000_hw_stats, hgptc)},
517 	{"rx_code_violation_packets", offsetof(struct e1000_hw_stats, scvpc)},
518 
519 	{"interrupt_assert_count", offsetof(struct e1000_hw_stats, iac)},
520 };
521 
522 #define IGB_NB_XSTATS (sizeof(rte_igb_stats_strings) / \
523 		sizeof(rte_igb_stats_strings[0]))
524 
525 static const struct rte_igb_xstats_name_off rte_igbvf_stats_strings[] = {
526 	{"rx_multicast_packets", offsetof(struct e1000_vf_stats, mprc)},
527 	{"rx_good_loopback_packets", offsetof(struct e1000_vf_stats, gprlbc)},
528 	{"tx_good_loopback_packets", offsetof(struct e1000_vf_stats, gptlbc)},
529 	{"rx_good_loopback_bytes", offsetof(struct e1000_vf_stats, gorlbc)},
530 	{"tx_good_loopback_bytes", offsetof(struct e1000_vf_stats, gotlbc)},
531 };
532 
533 #define IGBVF_NB_XSTATS (sizeof(rte_igbvf_stats_strings) / \
534 		sizeof(rte_igbvf_stats_strings[0]))
535 
536 
537 static inline void
538 igb_intr_enable(struct rte_eth_dev *dev)
539 {
540 	struct e1000_interrupt *intr =
541 		E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
542 	struct e1000_hw *hw =
543 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
544 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
545 	struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
546 
547 	if (rte_intr_allow_others(intr_handle) &&
548 		dev->data->dev_conf.intr_conf.lsc != 0) {
549 		E1000_WRITE_REG(hw, E1000_EIMS, 1 << IGB_MSIX_OTHER_INTR_VEC);
550 	}
551 
552 	E1000_WRITE_REG(hw, E1000_IMS, intr->mask);
553 	E1000_WRITE_FLUSH(hw);
554 }
555 
556 static void
557 igb_intr_disable(struct rte_eth_dev *dev)
558 {
559 	struct e1000_hw *hw =
560 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
561 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
562 	struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
563 
564 	if (rte_intr_allow_others(intr_handle) &&
565 		dev->data->dev_conf.intr_conf.lsc != 0) {
566 		E1000_WRITE_REG(hw, E1000_EIMC, 1 << IGB_MSIX_OTHER_INTR_VEC);
567 	}
568 
569 	E1000_WRITE_REG(hw, E1000_IMC, ~0);
570 	E1000_WRITE_FLUSH(hw);
571 }
572 
573 static inline void
574 igbvf_intr_enable(struct rte_eth_dev *dev)
575 {
576 	struct e1000_hw *hw =
577 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
578 
579 	/* only for mailbox */
580 	E1000_WRITE_REG(hw, E1000_EIAM, 1 << E1000_VTIVAR_MISC_MAILBOX);
581 	E1000_WRITE_REG(hw, E1000_EIAC, 1 << E1000_VTIVAR_MISC_MAILBOX);
582 	E1000_WRITE_REG(hw, E1000_EIMS, 1 << E1000_VTIVAR_MISC_MAILBOX);
583 	E1000_WRITE_FLUSH(hw);
584 }
585 
586 /* only for mailbox now. If RX/TX needed, should extend this function.  */
587 static void
588 igbvf_set_ivar_map(struct e1000_hw *hw, uint8_t msix_vector)
589 {
590 	uint32_t tmp = 0;
591 
592 	/* mailbox */
593 	tmp |= (msix_vector & E1000_VTIVAR_MISC_INTR_MASK);
594 	tmp |= E1000_VTIVAR_VALID;
595 	E1000_WRITE_REG(hw, E1000_VTIVAR_MISC, tmp);
596 }
597 
598 static void
599 eth_igbvf_configure_msix_intr(struct rte_eth_dev *dev)
600 {
601 	struct e1000_hw *hw =
602 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
603 
604 	/* Configure VF other cause ivar */
605 	igbvf_set_ivar_map(hw, E1000_VTIVAR_MISC_MAILBOX);
606 }
607 
608 static inline int32_t
609 igb_pf_reset_hw(struct e1000_hw *hw)
610 {
611 	uint32_t ctrl_ext;
612 	int32_t status;
613 
614 	status = e1000_reset_hw(hw);
615 
616 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
617 	/* Set PF Reset Done bit so PF/VF Mail Ops can work */
618 	ctrl_ext |= E1000_CTRL_EXT_PFRSTD;
619 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
620 	E1000_WRITE_FLUSH(hw);
621 
622 	return status;
623 }
624 
625 static void
626 igb_identify_hardware(struct rte_eth_dev *dev, struct rte_pci_device *pci_dev)
627 {
628 	struct e1000_hw *hw =
629 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
630 
631 
632 	hw->vendor_id = pci_dev->id.vendor_id;
633 	hw->device_id = pci_dev->id.device_id;
634 	hw->subsystem_vendor_id = pci_dev->id.subsystem_vendor_id;
635 	hw->subsystem_device_id = pci_dev->id.subsystem_device_id;
636 
637 	e1000_set_mac_type(hw);
638 
639 	/* need to check if it is a vf device below */
640 }
641 
642 static int
643 igb_reset_swfw_lock(struct e1000_hw *hw)
644 {
645 	int ret_val;
646 
647 	/*
648 	 * Do mac ops initialization manually here, since we will need
649 	 * some function pointers set by this call.
650 	 */
651 	ret_val = e1000_init_mac_params(hw);
652 	if (ret_val)
653 		return ret_val;
654 
655 	/*
656 	 * SMBI lock should not fail in this early stage. If this is the case,
657 	 * it is due to an improper exit of the application.
658 	 * So force the release of the faulty lock.
659 	 */
660 	if (e1000_get_hw_semaphore_generic(hw) < 0) {
661 		PMD_DRV_LOG(DEBUG, "SMBI lock released");
662 	}
663 	e1000_put_hw_semaphore_generic(hw);
664 
665 	if (hw->mac.ops.acquire_swfw_sync != NULL) {
666 		uint16_t mask;
667 
668 		/*
669 		 * Phy lock should not fail in this early stage. If this is the case,
670 		 * it is due to an improper exit of the application.
671 		 * So force the release of the faulty lock.
672 		 */
673 		mask = E1000_SWFW_PHY0_SM << hw->bus.func;
674 		if (hw->bus.func > E1000_FUNC_1)
675 			mask <<= 2;
676 		if (hw->mac.ops.acquire_swfw_sync(hw, mask) < 0) {
677 			PMD_DRV_LOG(DEBUG, "SWFW phy%d lock released",
678 				    hw->bus.func);
679 		}
680 		hw->mac.ops.release_swfw_sync(hw, mask);
681 
682 		/*
683 		 * This one is more tricky since it is common to all ports; but
684 		 * swfw_sync retries last long enough (1s) to be almost sure that if
685 		 * lock can not be taken it is due to an improper lock of the
686 		 * semaphore.
687 		 */
688 		mask = E1000_SWFW_EEP_SM;
689 		if (hw->mac.ops.acquire_swfw_sync(hw, mask) < 0) {
690 			PMD_DRV_LOG(DEBUG, "SWFW common locks released");
691 		}
692 		hw->mac.ops.release_swfw_sync(hw, mask);
693 	}
694 
695 	return E1000_SUCCESS;
696 }
697 
698 /* Remove all ntuple filters of the device */
699 static int igb_ntuple_filter_uninit(struct rte_eth_dev *eth_dev)
700 {
701 	struct e1000_filter_info *filter_info =
702 		E1000_DEV_PRIVATE_TO_FILTER_INFO(eth_dev->data->dev_private);
703 	struct e1000_5tuple_filter *p_5tuple;
704 	struct e1000_2tuple_filter *p_2tuple;
705 
706 	while ((p_5tuple = TAILQ_FIRST(&filter_info->fivetuple_list))) {
707 		TAILQ_REMOVE(&filter_info->fivetuple_list,
708 			p_5tuple, entries);
709 			rte_free(p_5tuple);
710 	}
711 	filter_info->fivetuple_mask = 0;
712 	while ((p_2tuple = TAILQ_FIRST(&filter_info->twotuple_list))) {
713 		TAILQ_REMOVE(&filter_info->twotuple_list,
714 			p_2tuple, entries);
715 			rte_free(p_2tuple);
716 	}
717 	filter_info->twotuple_mask = 0;
718 
719 	return 0;
720 }
721 
722 /* Remove all flex filters of the device */
723 static int igb_flex_filter_uninit(struct rte_eth_dev *eth_dev)
724 {
725 	struct e1000_filter_info *filter_info =
726 		E1000_DEV_PRIVATE_TO_FILTER_INFO(eth_dev->data->dev_private);
727 	struct e1000_flex_filter *p_flex;
728 
729 	while ((p_flex = TAILQ_FIRST(&filter_info->flex_list))) {
730 		TAILQ_REMOVE(&filter_info->flex_list, p_flex, entries);
731 		rte_free(p_flex);
732 	}
733 	filter_info->flex_mask = 0;
734 
735 	return 0;
736 }
737 
738 static int
739 eth_igb_dev_init(struct rte_eth_dev *eth_dev)
740 {
741 	int error = 0;
742 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
743 	struct e1000_hw *hw =
744 		E1000_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
745 	struct e1000_vfta * shadow_vfta =
746 		E1000_DEV_PRIVATE_TO_VFTA(eth_dev->data->dev_private);
747 	struct e1000_filter_info *filter_info =
748 		E1000_DEV_PRIVATE_TO_FILTER_INFO(eth_dev->data->dev_private);
749 	struct e1000_adapter *adapter =
750 		E1000_DEV_PRIVATE(eth_dev->data->dev_private);
751 
752 	uint32_t ctrl_ext;
753 
754 	eth_dev->dev_ops = &eth_igb_ops;
755 	eth_dev->rx_pkt_burst = &eth_igb_recv_pkts;
756 	eth_dev->tx_pkt_burst = &eth_igb_xmit_pkts;
757 	eth_dev->tx_pkt_prepare = &eth_igb_prep_pkts;
758 
759 	/* for secondary processes, we don't initialise any further as primary
760 	 * has already done this work. Only check we don't need a different
761 	 * RX function */
762 	if (rte_eal_process_type() != RTE_PROC_PRIMARY){
763 		if (eth_dev->data->scattered_rx)
764 			eth_dev->rx_pkt_burst = &eth_igb_recv_scattered_pkts;
765 		return 0;
766 	}
767 
768 	rte_eth_copy_pci_info(eth_dev, pci_dev);
769 
770 	hw->hw_addr= (void *)pci_dev->mem_resource[0].addr;
771 
772 	igb_identify_hardware(eth_dev, pci_dev);
773 	if (e1000_setup_init_funcs(hw, FALSE) != E1000_SUCCESS) {
774 		error = -EIO;
775 		goto err_late;
776 	}
777 
778 	e1000_get_bus_info(hw);
779 
780 	/* Reset any pending lock */
781 	if (igb_reset_swfw_lock(hw) != E1000_SUCCESS) {
782 		error = -EIO;
783 		goto err_late;
784 	}
785 
786 	/* Finish initialization */
787 	if (e1000_setup_init_funcs(hw, TRUE) != E1000_SUCCESS) {
788 		error = -EIO;
789 		goto err_late;
790 	}
791 
792 	hw->mac.autoneg = 1;
793 	hw->phy.autoneg_wait_to_complete = 0;
794 	hw->phy.autoneg_advertised = E1000_ALL_SPEED_DUPLEX;
795 
796 	/* Copper options */
797 	if (hw->phy.media_type == e1000_media_type_copper) {
798 		hw->phy.mdix = 0; /* AUTO_ALL_MODES */
799 		hw->phy.disable_polarity_correction = 0;
800 		hw->phy.ms_type = e1000_ms_hw_default;
801 	}
802 
803 	/*
804 	 * Start from a known state, this is important in reading the nvm
805 	 * and mac from that.
806 	 */
807 	igb_pf_reset_hw(hw);
808 
809 	/* Make sure we have a good EEPROM before we read from it */
810 	if (e1000_validate_nvm_checksum(hw) < 0) {
811 		/*
812 		 * Some PCI-E parts fail the first check due to
813 		 * the link being in sleep state, call it again,
814 		 * if it fails a second time its a real issue.
815 		 */
816 		if (e1000_validate_nvm_checksum(hw) < 0) {
817 			PMD_INIT_LOG(ERR, "EEPROM checksum invalid");
818 			error = -EIO;
819 			goto err_late;
820 		}
821 	}
822 
823 	/* Read the permanent MAC address out of the EEPROM */
824 	if (e1000_read_mac_addr(hw) != 0) {
825 		PMD_INIT_LOG(ERR, "EEPROM error while reading MAC address");
826 		error = -EIO;
827 		goto err_late;
828 	}
829 
830 	/* Allocate memory for storing MAC addresses */
831 	eth_dev->data->mac_addrs = rte_zmalloc("e1000",
832 		ETHER_ADDR_LEN * hw->mac.rar_entry_count, 0);
833 	if (eth_dev->data->mac_addrs == NULL) {
834 		PMD_INIT_LOG(ERR, "Failed to allocate %d bytes needed to "
835 						"store MAC addresses",
836 				ETHER_ADDR_LEN * hw->mac.rar_entry_count);
837 		error = -ENOMEM;
838 		goto err_late;
839 	}
840 
841 	/* Copy the permanent MAC address */
842 	ether_addr_copy((struct ether_addr *)hw->mac.addr, &eth_dev->data->mac_addrs[0]);
843 
844 	/* initialize the vfta */
845 	memset(shadow_vfta, 0, sizeof(*shadow_vfta));
846 
847 	/* Now initialize the hardware */
848 	if (igb_hardware_init(hw) != 0) {
849 		PMD_INIT_LOG(ERR, "Hardware initialization failed");
850 		rte_free(eth_dev->data->mac_addrs);
851 		eth_dev->data->mac_addrs = NULL;
852 		error = -ENODEV;
853 		goto err_late;
854 	}
855 	hw->mac.get_link_status = 1;
856 	adapter->stopped = 0;
857 
858 	/* Indicate SOL/IDER usage */
859 	if (e1000_check_reset_block(hw) < 0) {
860 		PMD_INIT_LOG(ERR, "PHY reset is blocked due to"
861 					"SOL/IDER session");
862 	}
863 
864 	/* initialize PF if max_vfs not zero */
865 	igb_pf_host_init(eth_dev);
866 
867 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
868 	/* Set PF Reset Done bit so PF/VF Mail Ops can work */
869 	ctrl_ext |= E1000_CTRL_EXT_PFRSTD;
870 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
871 	E1000_WRITE_FLUSH(hw);
872 
873 	PMD_INIT_LOG(DEBUG, "port_id %d vendorID=0x%x deviceID=0x%x",
874 		     eth_dev->data->port_id, pci_dev->id.vendor_id,
875 		     pci_dev->id.device_id);
876 
877 	rte_intr_callback_register(&pci_dev->intr_handle,
878 				   eth_igb_interrupt_handler,
879 				   (void *)eth_dev);
880 
881 	/* enable uio/vfio intr/eventfd mapping */
882 	rte_intr_enable(&pci_dev->intr_handle);
883 
884 	/* enable support intr */
885 	igb_intr_enable(eth_dev);
886 
887 	/* initialize filter info */
888 	memset(filter_info, 0,
889 	       sizeof(struct e1000_filter_info));
890 
891 	TAILQ_INIT(&filter_info->flex_list);
892 	TAILQ_INIT(&filter_info->twotuple_list);
893 	TAILQ_INIT(&filter_info->fivetuple_list);
894 
895 	TAILQ_INIT(&igb_filter_ntuple_list);
896 	TAILQ_INIT(&igb_filter_ethertype_list);
897 	TAILQ_INIT(&igb_filter_syn_list);
898 	TAILQ_INIT(&igb_filter_flex_list);
899 	TAILQ_INIT(&igb_filter_rss_list);
900 	TAILQ_INIT(&igb_flow_list);
901 
902 	return 0;
903 
904 err_late:
905 	igb_hw_control_release(hw);
906 
907 	return error;
908 }
909 
910 static int
911 eth_igb_dev_uninit(struct rte_eth_dev *eth_dev)
912 {
913 	struct rte_pci_device *pci_dev;
914 	struct rte_intr_handle *intr_handle;
915 	struct e1000_hw *hw;
916 	struct e1000_adapter *adapter =
917 		E1000_DEV_PRIVATE(eth_dev->data->dev_private);
918 	struct e1000_filter_info *filter_info =
919 		E1000_DEV_PRIVATE_TO_FILTER_INFO(eth_dev->data->dev_private);
920 
921 	PMD_INIT_FUNC_TRACE();
922 
923 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
924 		return -EPERM;
925 
926 	hw = E1000_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
927 	pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
928 	intr_handle = &pci_dev->intr_handle;
929 
930 	if (adapter->stopped == 0)
931 		eth_igb_close(eth_dev);
932 
933 	eth_dev->dev_ops = NULL;
934 	eth_dev->rx_pkt_burst = NULL;
935 	eth_dev->tx_pkt_burst = NULL;
936 
937 	/* Reset any pending lock */
938 	igb_reset_swfw_lock(hw);
939 
940 	/* uninitialize PF if max_vfs not zero */
941 	igb_pf_host_uninit(eth_dev);
942 
943 	/* disable uio intr before callback unregister */
944 	rte_intr_disable(intr_handle);
945 	rte_intr_callback_unregister(intr_handle,
946 				     eth_igb_interrupt_handler, eth_dev);
947 
948 	/* clear the SYN filter info */
949 	filter_info->syn_info = 0;
950 
951 	/* clear the ethertype filters info */
952 	filter_info->ethertype_mask = 0;
953 	memset(filter_info->ethertype_filters, 0,
954 		E1000_MAX_ETQF_FILTERS * sizeof(struct igb_ethertype_filter));
955 
956 	/* clear the rss filter info */
957 	memset(&filter_info->rss_info, 0,
958 		sizeof(struct igb_rte_flow_rss_conf));
959 
960 	/* remove all ntuple filters of the device */
961 	igb_ntuple_filter_uninit(eth_dev);
962 
963 	/* remove all flex filters of the device */
964 	igb_flex_filter_uninit(eth_dev);
965 
966 	/* clear all the filters list */
967 	igb_filterlist_flush(eth_dev);
968 
969 	return 0;
970 }
971 
972 /*
973  * Virtual Function device init
974  */
975 static int
976 eth_igbvf_dev_init(struct rte_eth_dev *eth_dev)
977 {
978 	struct rte_pci_device *pci_dev;
979 	struct rte_intr_handle *intr_handle;
980 	struct e1000_adapter *adapter =
981 		E1000_DEV_PRIVATE(eth_dev->data->dev_private);
982 	struct e1000_hw *hw =
983 		E1000_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
984 	int diag;
985 	struct ether_addr *perm_addr = (struct ether_addr *)hw->mac.perm_addr;
986 
987 	PMD_INIT_FUNC_TRACE();
988 
989 	eth_dev->dev_ops = &igbvf_eth_dev_ops;
990 	eth_dev->rx_pkt_burst = &eth_igb_recv_pkts;
991 	eth_dev->tx_pkt_burst = &eth_igb_xmit_pkts;
992 	eth_dev->tx_pkt_prepare = &eth_igb_prep_pkts;
993 
994 	/* for secondary processes, we don't initialise any further as primary
995 	 * has already done this work. Only check we don't need a different
996 	 * RX function */
997 	if (rte_eal_process_type() != RTE_PROC_PRIMARY){
998 		if (eth_dev->data->scattered_rx)
999 			eth_dev->rx_pkt_burst = &eth_igb_recv_scattered_pkts;
1000 		return 0;
1001 	}
1002 
1003 	pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1004 	rte_eth_copy_pci_info(eth_dev, pci_dev);
1005 
1006 	hw->device_id = pci_dev->id.device_id;
1007 	hw->vendor_id = pci_dev->id.vendor_id;
1008 	hw->hw_addr = (void *)pci_dev->mem_resource[0].addr;
1009 	adapter->stopped = 0;
1010 
1011 	/* Initialize the shared code (base driver) */
1012 	diag = e1000_setup_init_funcs(hw, TRUE);
1013 	if (diag != 0) {
1014 		PMD_INIT_LOG(ERR, "Shared code init failed for igbvf: %d",
1015 			diag);
1016 		return -EIO;
1017 	}
1018 
1019 	/* init_mailbox_params */
1020 	hw->mbx.ops.init_params(hw);
1021 
1022 	/* Disable the interrupts for VF */
1023 	igbvf_intr_disable(hw);
1024 
1025 	diag = hw->mac.ops.reset_hw(hw);
1026 
1027 	/* Allocate memory for storing MAC addresses */
1028 	eth_dev->data->mac_addrs = rte_zmalloc("igbvf", ETHER_ADDR_LEN *
1029 		hw->mac.rar_entry_count, 0);
1030 	if (eth_dev->data->mac_addrs == NULL) {
1031 		PMD_INIT_LOG(ERR,
1032 			"Failed to allocate %d bytes needed to store MAC "
1033 			"addresses",
1034 			ETHER_ADDR_LEN * hw->mac.rar_entry_count);
1035 		return -ENOMEM;
1036 	}
1037 
1038 	/* Generate a random MAC address, if none was assigned by PF. */
1039 	if (is_zero_ether_addr(perm_addr)) {
1040 		eth_random_addr(perm_addr->addr_bytes);
1041 		PMD_INIT_LOG(INFO, "\tVF MAC address not assigned by Host PF");
1042 		PMD_INIT_LOG(INFO, "\tAssign randomly generated MAC address "
1043 			     "%02x:%02x:%02x:%02x:%02x:%02x",
1044 			     perm_addr->addr_bytes[0],
1045 			     perm_addr->addr_bytes[1],
1046 			     perm_addr->addr_bytes[2],
1047 			     perm_addr->addr_bytes[3],
1048 			     perm_addr->addr_bytes[4],
1049 			     perm_addr->addr_bytes[5]);
1050 	}
1051 
1052 	diag = e1000_rar_set(hw, perm_addr->addr_bytes, 0);
1053 	if (diag) {
1054 		rte_free(eth_dev->data->mac_addrs);
1055 		eth_dev->data->mac_addrs = NULL;
1056 		return diag;
1057 	}
1058 	/* Copy the permanent MAC address */
1059 	ether_addr_copy((struct ether_addr *) hw->mac.perm_addr,
1060 			&eth_dev->data->mac_addrs[0]);
1061 
1062 	PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x "
1063 		     "mac.type=%s",
1064 		     eth_dev->data->port_id, pci_dev->id.vendor_id,
1065 		     pci_dev->id.device_id, "igb_mac_82576_vf");
1066 
1067 	intr_handle = &pci_dev->intr_handle;
1068 	rte_intr_callback_register(intr_handle,
1069 				   eth_igbvf_interrupt_handler, eth_dev);
1070 
1071 	return 0;
1072 }
1073 
1074 static int
1075 eth_igbvf_dev_uninit(struct rte_eth_dev *eth_dev)
1076 {
1077 	struct e1000_adapter *adapter =
1078 		E1000_DEV_PRIVATE(eth_dev->data->dev_private);
1079 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1080 
1081 	PMD_INIT_FUNC_TRACE();
1082 
1083 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1084 		return -EPERM;
1085 
1086 	if (adapter->stopped == 0)
1087 		igbvf_dev_close(eth_dev);
1088 
1089 	eth_dev->dev_ops = NULL;
1090 	eth_dev->rx_pkt_burst = NULL;
1091 	eth_dev->tx_pkt_burst = NULL;
1092 
1093 	/* disable uio intr before callback unregister */
1094 	rte_intr_disable(&pci_dev->intr_handle);
1095 	rte_intr_callback_unregister(&pci_dev->intr_handle,
1096 				     eth_igbvf_interrupt_handler,
1097 				     (void *)eth_dev);
1098 
1099 	return 0;
1100 }
1101 
1102 static int eth_igb_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1103 	struct rte_pci_device *pci_dev)
1104 {
1105 	return rte_eth_dev_pci_generic_probe(pci_dev,
1106 		sizeof(struct e1000_adapter), eth_igb_dev_init);
1107 }
1108 
1109 static int eth_igb_pci_remove(struct rte_pci_device *pci_dev)
1110 {
1111 	return rte_eth_dev_pci_generic_remove(pci_dev, eth_igb_dev_uninit);
1112 }
1113 
1114 static struct rte_pci_driver rte_igb_pmd = {
1115 	.id_table = pci_id_igb_map,
1116 	.drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC |
1117 		     RTE_PCI_DRV_IOVA_AS_VA,
1118 	.probe = eth_igb_pci_probe,
1119 	.remove = eth_igb_pci_remove,
1120 };
1121 
1122 
1123 static int eth_igbvf_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1124 	struct rte_pci_device *pci_dev)
1125 {
1126 	return rte_eth_dev_pci_generic_probe(pci_dev,
1127 		sizeof(struct e1000_adapter), eth_igbvf_dev_init);
1128 }
1129 
1130 static int eth_igbvf_pci_remove(struct rte_pci_device *pci_dev)
1131 {
1132 	return rte_eth_dev_pci_generic_remove(pci_dev, eth_igbvf_dev_uninit);
1133 }
1134 
1135 /*
1136  * virtual function driver struct
1137  */
1138 static struct rte_pci_driver rte_igbvf_pmd = {
1139 	.id_table = pci_id_igbvf_map,
1140 	.drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_IOVA_AS_VA,
1141 	.probe = eth_igbvf_pci_probe,
1142 	.remove = eth_igbvf_pci_remove,
1143 };
1144 
1145 static void
1146 igb_vmdq_vlan_hw_filter_enable(struct rte_eth_dev *dev)
1147 {
1148 	struct e1000_hw *hw =
1149 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1150 	/* RCTL: enable VLAN filter since VMDq always use VLAN filter */
1151 	uint32_t rctl = E1000_READ_REG(hw, E1000_RCTL);
1152 	rctl |= E1000_RCTL_VFE;
1153 	E1000_WRITE_REG(hw, E1000_RCTL, rctl);
1154 }
1155 
1156 static int
1157 igb_check_mq_mode(struct rte_eth_dev *dev)
1158 {
1159 	enum rte_eth_rx_mq_mode rx_mq_mode = dev->data->dev_conf.rxmode.mq_mode;
1160 	enum rte_eth_tx_mq_mode tx_mq_mode = dev->data->dev_conf.txmode.mq_mode;
1161 	uint16_t nb_rx_q = dev->data->nb_rx_queues;
1162 	uint16_t nb_tx_q = dev->data->nb_tx_queues;
1163 
1164 	if ((rx_mq_mode & ETH_MQ_RX_DCB_FLAG) ||
1165 	    tx_mq_mode == ETH_MQ_TX_DCB ||
1166 	    tx_mq_mode == ETH_MQ_TX_VMDQ_DCB) {
1167 		PMD_INIT_LOG(ERR, "DCB mode is not supported.");
1168 		return -EINVAL;
1169 	}
1170 	if (RTE_ETH_DEV_SRIOV(dev).active != 0) {
1171 		/* Check multi-queue mode.
1172 		 * To no break software we accept ETH_MQ_RX_NONE as this might
1173 		 * be used to turn off VLAN filter.
1174 		 */
1175 
1176 		if (rx_mq_mode == ETH_MQ_RX_NONE ||
1177 		    rx_mq_mode == ETH_MQ_RX_VMDQ_ONLY) {
1178 			dev->data->dev_conf.rxmode.mq_mode = ETH_MQ_RX_VMDQ_ONLY;
1179 			RTE_ETH_DEV_SRIOV(dev).nb_q_per_pool = 1;
1180 		} else {
1181 			/* Only support one queue on VFs.
1182 			 * RSS together with SRIOV is not supported.
1183 			 */
1184 			PMD_INIT_LOG(ERR, "SRIOV is active,"
1185 					" wrong mq_mode rx %d.",
1186 					rx_mq_mode);
1187 			return -EINVAL;
1188 		}
1189 		/* TX mode is not used here, so mode might be ignored.*/
1190 		if (tx_mq_mode != ETH_MQ_TX_VMDQ_ONLY) {
1191 			/* SRIOV only works in VMDq enable mode */
1192 			PMD_INIT_LOG(WARNING, "SRIOV is active,"
1193 					" TX mode %d is not supported. "
1194 					" Driver will behave as %d mode.",
1195 					tx_mq_mode, ETH_MQ_TX_VMDQ_ONLY);
1196 		}
1197 
1198 		/* check valid queue number */
1199 		if ((nb_rx_q > 1) || (nb_tx_q > 1)) {
1200 			PMD_INIT_LOG(ERR, "SRIOV is active,"
1201 					" only support one queue on VFs.");
1202 			return -EINVAL;
1203 		}
1204 	} else {
1205 		/* To no break software that set invalid mode, only display
1206 		 * warning if invalid mode is used.
1207 		 */
1208 		if (rx_mq_mode != ETH_MQ_RX_NONE &&
1209 		    rx_mq_mode != ETH_MQ_RX_VMDQ_ONLY &&
1210 		    rx_mq_mode != ETH_MQ_RX_RSS) {
1211 			/* RSS together with VMDq not supported*/
1212 			PMD_INIT_LOG(ERR, "RX mode %d is not supported.",
1213 				     rx_mq_mode);
1214 			return -EINVAL;
1215 		}
1216 
1217 		if (tx_mq_mode != ETH_MQ_TX_NONE &&
1218 		    tx_mq_mode != ETH_MQ_TX_VMDQ_ONLY) {
1219 			PMD_INIT_LOG(WARNING, "TX mode %d is not supported."
1220 					" Due to txmode is meaningless in this"
1221 					" driver, just ignore.",
1222 					tx_mq_mode);
1223 		}
1224 	}
1225 	return 0;
1226 }
1227 
1228 static int
1229 eth_igb_configure(struct rte_eth_dev *dev)
1230 {
1231 	struct e1000_interrupt *intr =
1232 		E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
1233 	int ret;
1234 
1235 	PMD_INIT_FUNC_TRACE();
1236 
1237 	/* multipe queue mode checking */
1238 	ret  = igb_check_mq_mode(dev);
1239 	if (ret != 0) {
1240 		PMD_DRV_LOG(ERR, "igb_check_mq_mode fails with %d.",
1241 			    ret);
1242 		return ret;
1243 	}
1244 
1245 	intr->flags |= E1000_FLAG_NEED_LINK_UPDATE;
1246 	PMD_INIT_FUNC_TRACE();
1247 
1248 	return 0;
1249 }
1250 
1251 static void
1252 eth_igb_rxtx_control(struct rte_eth_dev *dev,
1253 		     bool enable)
1254 {
1255 	struct e1000_hw *hw =
1256 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1257 	uint32_t tctl, rctl;
1258 
1259 	tctl = E1000_READ_REG(hw, E1000_TCTL);
1260 	rctl = E1000_READ_REG(hw, E1000_RCTL);
1261 
1262 	if (enable) {
1263 		/* enable Tx/Rx */
1264 		tctl |= E1000_TCTL_EN;
1265 		rctl |= E1000_RCTL_EN;
1266 	} else {
1267 		/* disable Tx/Rx */
1268 		tctl &= ~E1000_TCTL_EN;
1269 		rctl &= ~E1000_RCTL_EN;
1270 	}
1271 	E1000_WRITE_REG(hw, E1000_TCTL, tctl);
1272 	E1000_WRITE_REG(hw, E1000_RCTL, rctl);
1273 	E1000_WRITE_FLUSH(hw);
1274 }
1275 
1276 static int
1277 eth_igb_start(struct rte_eth_dev *dev)
1278 {
1279 	struct e1000_hw *hw =
1280 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1281 	struct e1000_adapter *adapter =
1282 		E1000_DEV_PRIVATE(dev->data->dev_private);
1283 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1284 	struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
1285 	int ret, mask;
1286 	uint32_t intr_vector = 0;
1287 	uint32_t ctrl_ext;
1288 	uint32_t *speeds;
1289 	int num_speeds;
1290 	bool autoneg;
1291 
1292 	PMD_INIT_FUNC_TRACE();
1293 
1294 	/* disable uio/vfio intr/eventfd mapping */
1295 	rte_intr_disable(intr_handle);
1296 
1297 	/* Power up the phy. Needed to make the link go Up */
1298 	eth_igb_dev_set_link_up(dev);
1299 
1300 	/*
1301 	 * Packet Buffer Allocation (PBA)
1302 	 * Writing PBA sets the receive portion of the buffer
1303 	 * the remainder is used for the transmit buffer.
1304 	 */
1305 	if (hw->mac.type == e1000_82575) {
1306 		uint32_t pba;
1307 
1308 		pba = E1000_PBA_32K; /* 32K for Rx, 16K for Tx */
1309 		E1000_WRITE_REG(hw, E1000_PBA, pba);
1310 	}
1311 
1312 	/* Put the address into the Receive Address Array */
1313 	e1000_rar_set(hw, hw->mac.addr, 0);
1314 
1315 	/* Initialize the hardware */
1316 	if (igb_hardware_init(hw)) {
1317 		PMD_INIT_LOG(ERR, "Unable to initialize the hardware");
1318 		return -EIO;
1319 	}
1320 	adapter->stopped = 0;
1321 
1322 	E1000_WRITE_REG(hw, E1000_VET, ETHER_TYPE_VLAN << 16 | ETHER_TYPE_VLAN);
1323 
1324 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
1325 	/* Set PF Reset Done bit so PF/VF Mail Ops can work */
1326 	ctrl_ext |= E1000_CTRL_EXT_PFRSTD;
1327 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
1328 	E1000_WRITE_FLUSH(hw);
1329 
1330 	/* configure PF module if SRIOV enabled */
1331 	igb_pf_host_configure(dev);
1332 
1333 	/* check and configure queue intr-vector mapping */
1334 	if ((rte_intr_cap_multiple(intr_handle) ||
1335 	     !RTE_ETH_DEV_SRIOV(dev).active) &&
1336 	    dev->data->dev_conf.intr_conf.rxq != 0) {
1337 		intr_vector = dev->data->nb_rx_queues;
1338 		if (rte_intr_efd_enable(intr_handle, intr_vector))
1339 			return -1;
1340 	}
1341 
1342 	if (rte_intr_dp_is_en(intr_handle) && !intr_handle->intr_vec) {
1343 		intr_handle->intr_vec =
1344 			rte_zmalloc("intr_vec",
1345 				    dev->data->nb_rx_queues * sizeof(int), 0);
1346 		if (intr_handle->intr_vec == NULL) {
1347 			PMD_INIT_LOG(ERR, "Failed to allocate %d rx_queues"
1348 				     " intr_vec", dev->data->nb_rx_queues);
1349 			return -ENOMEM;
1350 		}
1351 	}
1352 
1353 	/* confiugre msix for rx interrupt */
1354 	eth_igb_configure_msix_intr(dev);
1355 
1356 	/* Configure for OS presence */
1357 	igb_init_manageability(hw);
1358 
1359 	eth_igb_tx_init(dev);
1360 
1361 	/* This can fail when allocating mbufs for descriptor rings */
1362 	ret = eth_igb_rx_init(dev);
1363 	if (ret) {
1364 		PMD_INIT_LOG(ERR, "Unable to initialize RX hardware");
1365 		igb_dev_clear_queues(dev);
1366 		return ret;
1367 	}
1368 
1369 	e1000_clear_hw_cntrs_base_generic(hw);
1370 
1371 	/*
1372 	 * VLAN Offload Settings
1373 	 */
1374 	mask = ETH_VLAN_STRIP_MASK | ETH_VLAN_FILTER_MASK | \
1375 			ETH_VLAN_EXTEND_MASK;
1376 	ret = eth_igb_vlan_offload_set(dev, mask);
1377 	if (ret) {
1378 		PMD_INIT_LOG(ERR, "Unable to set vlan offload");
1379 		igb_dev_clear_queues(dev);
1380 		return ret;
1381 	}
1382 
1383 	if (dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_VMDQ_ONLY) {
1384 		/* Enable VLAN filter since VMDq always use VLAN filter */
1385 		igb_vmdq_vlan_hw_filter_enable(dev);
1386 	}
1387 
1388 	if ((hw->mac.type == e1000_82576) || (hw->mac.type == e1000_82580) ||
1389 		(hw->mac.type == e1000_i350) || (hw->mac.type == e1000_i210) ||
1390 		(hw->mac.type == e1000_i211)) {
1391 		/* Configure EITR with the maximum possible value (0xFFFF) */
1392 		E1000_WRITE_REG(hw, E1000_EITR(0), 0xFFFF);
1393 	}
1394 
1395 	/* Setup link speed and duplex */
1396 	speeds = &dev->data->dev_conf.link_speeds;
1397 	if (*speeds == ETH_LINK_SPEED_AUTONEG) {
1398 		hw->phy.autoneg_advertised = E1000_ALL_SPEED_DUPLEX;
1399 		hw->mac.autoneg = 1;
1400 	} else {
1401 		num_speeds = 0;
1402 		autoneg = (*speeds & ETH_LINK_SPEED_FIXED) == 0;
1403 
1404 		/* Reset */
1405 		hw->phy.autoneg_advertised = 0;
1406 
1407 		if (*speeds & ~(ETH_LINK_SPEED_10M_HD | ETH_LINK_SPEED_10M |
1408 				ETH_LINK_SPEED_100M_HD | ETH_LINK_SPEED_100M |
1409 				ETH_LINK_SPEED_1G | ETH_LINK_SPEED_FIXED)) {
1410 			num_speeds = -1;
1411 			goto error_invalid_config;
1412 		}
1413 		if (*speeds & ETH_LINK_SPEED_10M_HD) {
1414 			hw->phy.autoneg_advertised |= ADVERTISE_10_HALF;
1415 			num_speeds++;
1416 		}
1417 		if (*speeds & ETH_LINK_SPEED_10M) {
1418 			hw->phy.autoneg_advertised |= ADVERTISE_10_FULL;
1419 			num_speeds++;
1420 		}
1421 		if (*speeds & ETH_LINK_SPEED_100M_HD) {
1422 			hw->phy.autoneg_advertised |= ADVERTISE_100_HALF;
1423 			num_speeds++;
1424 		}
1425 		if (*speeds & ETH_LINK_SPEED_100M) {
1426 			hw->phy.autoneg_advertised |= ADVERTISE_100_FULL;
1427 			num_speeds++;
1428 		}
1429 		if (*speeds & ETH_LINK_SPEED_1G) {
1430 			hw->phy.autoneg_advertised |= ADVERTISE_1000_FULL;
1431 			num_speeds++;
1432 		}
1433 		if (num_speeds == 0 || (!autoneg && (num_speeds > 1)))
1434 			goto error_invalid_config;
1435 
1436 		/* Set/reset the mac.autoneg based on the link speed,
1437 		 * fixed or not
1438 		 */
1439 		if (!autoneg) {
1440 			hw->mac.autoneg = 0;
1441 			hw->mac.forced_speed_duplex =
1442 					hw->phy.autoneg_advertised;
1443 		} else {
1444 			hw->mac.autoneg = 1;
1445 		}
1446 	}
1447 
1448 	e1000_setup_link(hw);
1449 
1450 	if (rte_intr_allow_others(intr_handle)) {
1451 		/* check if lsc interrupt is enabled */
1452 		if (dev->data->dev_conf.intr_conf.lsc != 0)
1453 			eth_igb_lsc_interrupt_setup(dev, TRUE);
1454 		else
1455 			eth_igb_lsc_interrupt_setup(dev, FALSE);
1456 	} else {
1457 		rte_intr_callback_unregister(intr_handle,
1458 					     eth_igb_interrupt_handler,
1459 					     (void *)dev);
1460 		if (dev->data->dev_conf.intr_conf.lsc != 0)
1461 			PMD_INIT_LOG(INFO, "lsc won't enable because of"
1462 				     " no intr multiplex");
1463 	}
1464 
1465 	/* check if rxq interrupt is enabled */
1466 	if (dev->data->dev_conf.intr_conf.rxq != 0 &&
1467 	    rte_intr_dp_is_en(intr_handle))
1468 		eth_igb_rxq_interrupt_setup(dev);
1469 
1470 	/* enable uio/vfio intr/eventfd mapping */
1471 	rte_intr_enable(intr_handle);
1472 
1473 	/* resume enabled intr since hw reset */
1474 	igb_intr_enable(dev);
1475 
1476 	/* restore all types filter */
1477 	igb_filter_restore(dev);
1478 
1479 	eth_igb_rxtx_control(dev, true);
1480 	eth_igb_link_update(dev, 0);
1481 
1482 	PMD_INIT_LOG(DEBUG, "<<");
1483 
1484 	return 0;
1485 
1486 error_invalid_config:
1487 	PMD_INIT_LOG(ERR, "Invalid advertised speeds (%u) for port %u",
1488 		     dev->data->dev_conf.link_speeds, dev->data->port_id);
1489 	igb_dev_clear_queues(dev);
1490 	return -EINVAL;
1491 }
1492 
1493 /*********************************************************************
1494  *
1495  *  This routine disables all traffic on the adapter by issuing a
1496  *  global reset on the MAC.
1497  *
1498  **********************************************************************/
1499 static void
1500 eth_igb_stop(struct rte_eth_dev *dev)
1501 {
1502 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1503 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1504 	struct rte_eth_link link;
1505 	struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
1506 
1507 	eth_igb_rxtx_control(dev, false);
1508 
1509 	igb_intr_disable(dev);
1510 
1511 	/* disable intr eventfd mapping */
1512 	rte_intr_disable(intr_handle);
1513 
1514 	igb_pf_reset_hw(hw);
1515 	E1000_WRITE_REG(hw, E1000_WUC, 0);
1516 
1517 	/* Set bit for Go Link disconnect */
1518 	if (hw->mac.type >= e1000_82580) {
1519 		uint32_t phpm_reg;
1520 
1521 		phpm_reg = E1000_READ_REG(hw, E1000_82580_PHY_POWER_MGMT);
1522 		phpm_reg |= E1000_82580_PM_GO_LINKD;
1523 		E1000_WRITE_REG(hw, E1000_82580_PHY_POWER_MGMT, phpm_reg);
1524 	}
1525 
1526 	/* Power down the phy. Needed to make the link go Down */
1527 	eth_igb_dev_set_link_down(dev);
1528 
1529 	igb_dev_clear_queues(dev);
1530 
1531 	/* clear the recorded link status */
1532 	memset(&link, 0, sizeof(link));
1533 	rte_eth_linkstatus_set(dev, &link);
1534 
1535 	if (!rte_intr_allow_others(intr_handle))
1536 		/* resume to the default handler */
1537 		rte_intr_callback_register(intr_handle,
1538 					   eth_igb_interrupt_handler,
1539 					   (void *)dev);
1540 
1541 	/* Clean datapath event and queue/vec mapping */
1542 	rte_intr_efd_disable(intr_handle);
1543 	if (intr_handle->intr_vec != NULL) {
1544 		rte_free(intr_handle->intr_vec);
1545 		intr_handle->intr_vec = NULL;
1546 	}
1547 }
1548 
1549 static int
1550 eth_igb_dev_set_link_up(struct rte_eth_dev *dev)
1551 {
1552 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1553 
1554 	if (hw->phy.media_type == e1000_media_type_copper)
1555 		e1000_power_up_phy(hw);
1556 	else
1557 		e1000_power_up_fiber_serdes_link(hw);
1558 
1559 	return 0;
1560 }
1561 
1562 static int
1563 eth_igb_dev_set_link_down(struct rte_eth_dev *dev)
1564 {
1565 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1566 
1567 	if (hw->phy.media_type == e1000_media_type_copper)
1568 		e1000_power_down_phy(hw);
1569 	else
1570 		e1000_shutdown_fiber_serdes_link(hw);
1571 
1572 	return 0;
1573 }
1574 
1575 static void
1576 eth_igb_close(struct rte_eth_dev *dev)
1577 {
1578 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1579 	struct e1000_adapter *adapter =
1580 		E1000_DEV_PRIVATE(dev->data->dev_private);
1581 	struct rte_eth_link link;
1582 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1583 	struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
1584 
1585 	eth_igb_stop(dev);
1586 	adapter->stopped = 1;
1587 
1588 	e1000_phy_hw_reset(hw);
1589 	igb_release_manageability(hw);
1590 	igb_hw_control_release(hw);
1591 
1592 	/* Clear bit for Go Link disconnect */
1593 	if (hw->mac.type >= e1000_82580) {
1594 		uint32_t phpm_reg;
1595 
1596 		phpm_reg = E1000_READ_REG(hw, E1000_82580_PHY_POWER_MGMT);
1597 		phpm_reg &= ~E1000_82580_PM_GO_LINKD;
1598 		E1000_WRITE_REG(hw, E1000_82580_PHY_POWER_MGMT, phpm_reg);
1599 	}
1600 
1601 	igb_dev_free_queues(dev);
1602 
1603 	if (intr_handle->intr_vec) {
1604 		rte_free(intr_handle->intr_vec);
1605 		intr_handle->intr_vec = NULL;
1606 	}
1607 
1608 	memset(&link, 0, sizeof(link));
1609 	rte_eth_linkstatus_set(dev, &link);
1610 }
1611 
1612 /*
1613  * Reset PF device.
1614  */
1615 static int
1616 eth_igb_reset(struct rte_eth_dev *dev)
1617 {
1618 	int ret;
1619 
1620 	/* When a DPDK PMD PF begin to reset PF port, it should notify all
1621 	 * its VF to make them align with it. The detailed notification
1622 	 * mechanism is PMD specific and is currently not implemented.
1623 	 * To avoid unexpected behavior in VF, currently reset of PF with
1624 	 * SR-IOV activation is not supported. It might be supported later.
1625 	 */
1626 	if (dev->data->sriov.active)
1627 		return -ENOTSUP;
1628 
1629 	ret = eth_igb_dev_uninit(dev);
1630 	if (ret)
1631 		return ret;
1632 
1633 	ret = eth_igb_dev_init(dev);
1634 
1635 	return ret;
1636 }
1637 
1638 
1639 static int
1640 igb_get_rx_buffer_size(struct e1000_hw *hw)
1641 {
1642 	uint32_t rx_buf_size;
1643 	if (hw->mac.type == e1000_82576) {
1644 		rx_buf_size = (E1000_READ_REG(hw, E1000_RXPBS) & 0xffff) << 10;
1645 	} else if (hw->mac.type == e1000_82580 || hw->mac.type == e1000_i350) {
1646 		/* PBS needs to be translated according to a lookup table */
1647 		rx_buf_size = (E1000_READ_REG(hw, E1000_RXPBS) & 0xf);
1648 		rx_buf_size = (uint32_t) e1000_rxpbs_adjust_82580(rx_buf_size);
1649 		rx_buf_size = (rx_buf_size << 10);
1650 	} else if (hw->mac.type == e1000_i210 || hw->mac.type == e1000_i211) {
1651 		rx_buf_size = (E1000_READ_REG(hw, E1000_RXPBS) & 0x3f) << 10;
1652 	} else {
1653 		rx_buf_size = (E1000_READ_REG(hw, E1000_PBA) & 0xffff) << 10;
1654 	}
1655 
1656 	return rx_buf_size;
1657 }
1658 
1659 /*********************************************************************
1660  *
1661  *  Initialize the hardware
1662  *
1663  **********************************************************************/
1664 static int
1665 igb_hardware_init(struct e1000_hw *hw)
1666 {
1667 	uint32_t rx_buf_size;
1668 	int diag;
1669 
1670 	/* Let the firmware know the OS is in control */
1671 	igb_hw_control_acquire(hw);
1672 
1673 	/*
1674 	 * These parameters control the automatic generation (Tx) and
1675 	 * response (Rx) to Ethernet PAUSE frames.
1676 	 * - High water mark should allow for at least two standard size (1518)
1677 	 *   frames to be received after sending an XOFF.
1678 	 * - Low water mark works best when it is very near the high water mark.
1679 	 *   This allows the receiver to restart by sending XON when it has
1680 	 *   drained a bit. Here we use an arbitrary value of 1500 which will
1681 	 *   restart after one full frame is pulled from the buffer. There
1682 	 *   could be several smaller frames in the buffer and if so they will
1683 	 *   not trigger the XON until their total number reduces the buffer
1684 	 *   by 1500.
1685 	 * - The pause time is fairly large at 1000 x 512ns = 512 usec.
1686 	 */
1687 	rx_buf_size = igb_get_rx_buffer_size(hw);
1688 
1689 	hw->fc.high_water = rx_buf_size - (ETHER_MAX_LEN * 2);
1690 	hw->fc.low_water = hw->fc.high_water - 1500;
1691 	hw->fc.pause_time = IGB_FC_PAUSE_TIME;
1692 	hw->fc.send_xon = 1;
1693 
1694 	/* Set Flow control, use the tunable location if sane */
1695 	if ((igb_fc_setting != e1000_fc_none) && (igb_fc_setting < 4))
1696 		hw->fc.requested_mode = igb_fc_setting;
1697 	else
1698 		hw->fc.requested_mode = e1000_fc_none;
1699 
1700 	/* Issue a global reset */
1701 	igb_pf_reset_hw(hw);
1702 	E1000_WRITE_REG(hw, E1000_WUC, 0);
1703 
1704 	diag = e1000_init_hw(hw);
1705 	if (diag < 0)
1706 		return diag;
1707 
1708 	E1000_WRITE_REG(hw, E1000_VET, ETHER_TYPE_VLAN << 16 | ETHER_TYPE_VLAN);
1709 	e1000_get_phy_info(hw);
1710 	e1000_check_for_link(hw);
1711 
1712 	return 0;
1713 }
1714 
1715 /* This function is based on igb_update_stats_counters() in igb/if_igb.c */
1716 static void
1717 igb_read_stats_registers(struct e1000_hw *hw, struct e1000_hw_stats *stats)
1718 {
1719 	int pause_frames;
1720 
1721 	uint64_t old_gprc  = stats->gprc;
1722 	uint64_t old_gptc  = stats->gptc;
1723 	uint64_t old_tpr   = stats->tpr;
1724 	uint64_t old_tpt   = stats->tpt;
1725 	uint64_t old_rpthc = stats->rpthc;
1726 	uint64_t old_hgptc = stats->hgptc;
1727 
1728 	if(hw->phy.media_type == e1000_media_type_copper ||
1729 	    (E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU)) {
1730 		stats->symerrs +=
1731 		    E1000_READ_REG(hw,E1000_SYMERRS);
1732 		stats->sec += E1000_READ_REG(hw, E1000_SEC);
1733 	}
1734 
1735 	stats->crcerrs += E1000_READ_REG(hw, E1000_CRCERRS);
1736 	stats->mpc += E1000_READ_REG(hw, E1000_MPC);
1737 	stats->scc += E1000_READ_REG(hw, E1000_SCC);
1738 	stats->ecol += E1000_READ_REG(hw, E1000_ECOL);
1739 
1740 	stats->mcc += E1000_READ_REG(hw, E1000_MCC);
1741 	stats->latecol += E1000_READ_REG(hw, E1000_LATECOL);
1742 	stats->colc += E1000_READ_REG(hw, E1000_COLC);
1743 	stats->dc += E1000_READ_REG(hw, E1000_DC);
1744 	stats->rlec += E1000_READ_REG(hw, E1000_RLEC);
1745 	stats->xonrxc += E1000_READ_REG(hw, E1000_XONRXC);
1746 	stats->xontxc += E1000_READ_REG(hw, E1000_XONTXC);
1747 	/*
1748 	** For watchdog management we need to know if we have been
1749 	** paused during the last interval, so capture that here.
1750 	*/
1751 	pause_frames = E1000_READ_REG(hw, E1000_XOFFRXC);
1752 	stats->xoffrxc += pause_frames;
1753 	stats->xofftxc += E1000_READ_REG(hw, E1000_XOFFTXC);
1754 	stats->fcruc += E1000_READ_REG(hw, E1000_FCRUC);
1755 	stats->prc64 += E1000_READ_REG(hw, E1000_PRC64);
1756 	stats->prc127 += E1000_READ_REG(hw, E1000_PRC127);
1757 	stats->prc255 += E1000_READ_REG(hw, E1000_PRC255);
1758 	stats->prc511 += E1000_READ_REG(hw, E1000_PRC511);
1759 	stats->prc1023 += E1000_READ_REG(hw, E1000_PRC1023);
1760 	stats->prc1522 += E1000_READ_REG(hw, E1000_PRC1522);
1761 	stats->gprc += E1000_READ_REG(hw, E1000_GPRC);
1762 	stats->bprc += E1000_READ_REG(hw, E1000_BPRC);
1763 	stats->mprc += E1000_READ_REG(hw, E1000_MPRC);
1764 	stats->gptc += E1000_READ_REG(hw, E1000_GPTC);
1765 
1766 	/* For the 64-bit byte counters the low dword must be read first. */
1767 	/* Both registers clear on the read of the high dword */
1768 
1769 	/* Workaround CRC bytes included in size, take away 4 bytes/packet */
1770 	stats->gorc += E1000_READ_REG(hw, E1000_GORCL);
1771 	stats->gorc += ((uint64_t)E1000_READ_REG(hw, E1000_GORCH) << 32);
1772 	stats->gorc -= (stats->gprc - old_gprc) * ETHER_CRC_LEN;
1773 	stats->gotc += E1000_READ_REG(hw, E1000_GOTCL);
1774 	stats->gotc += ((uint64_t)E1000_READ_REG(hw, E1000_GOTCH) << 32);
1775 	stats->gotc -= (stats->gptc - old_gptc) * ETHER_CRC_LEN;
1776 
1777 	stats->rnbc += E1000_READ_REG(hw, E1000_RNBC);
1778 	stats->ruc += E1000_READ_REG(hw, E1000_RUC);
1779 	stats->rfc += E1000_READ_REG(hw, E1000_RFC);
1780 	stats->roc += E1000_READ_REG(hw, E1000_ROC);
1781 	stats->rjc += E1000_READ_REG(hw, E1000_RJC);
1782 
1783 	stats->tpr += E1000_READ_REG(hw, E1000_TPR);
1784 	stats->tpt += E1000_READ_REG(hw, E1000_TPT);
1785 
1786 	stats->tor += E1000_READ_REG(hw, E1000_TORL);
1787 	stats->tor += ((uint64_t)E1000_READ_REG(hw, E1000_TORH) << 32);
1788 	stats->tor -= (stats->tpr - old_tpr) * ETHER_CRC_LEN;
1789 	stats->tot += E1000_READ_REG(hw, E1000_TOTL);
1790 	stats->tot += ((uint64_t)E1000_READ_REG(hw, E1000_TOTH) << 32);
1791 	stats->tot -= (stats->tpt - old_tpt) * ETHER_CRC_LEN;
1792 
1793 	stats->ptc64 += E1000_READ_REG(hw, E1000_PTC64);
1794 	stats->ptc127 += E1000_READ_REG(hw, E1000_PTC127);
1795 	stats->ptc255 += E1000_READ_REG(hw, E1000_PTC255);
1796 	stats->ptc511 += E1000_READ_REG(hw, E1000_PTC511);
1797 	stats->ptc1023 += E1000_READ_REG(hw, E1000_PTC1023);
1798 	stats->ptc1522 += E1000_READ_REG(hw, E1000_PTC1522);
1799 	stats->mptc += E1000_READ_REG(hw, E1000_MPTC);
1800 	stats->bptc += E1000_READ_REG(hw, E1000_BPTC);
1801 
1802 	/* Interrupt Counts */
1803 
1804 	stats->iac += E1000_READ_REG(hw, E1000_IAC);
1805 	stats->icrxptc += E1000_READ_REG(hw, E1000_ICRXPTC);
1806 	stats->icrxatc += E1000_READ_REG(hw, E1000_ICRXATC);
1807 	stats->ictxptc += E1000_READ_REG(hw, E1000_ICTXPTC);
1808 	stats->ictxatc += E1000_READ_REG(hw, E1000_ICTXATC);
1809 	stats->ictxqec += E1000_READ_REG(hw, E1000_ICTXQEC);
1810 	stats->ictxqmtc += E1000_READ_REG(hw, E1000_ICTXQMTC);
1811 	stats->icrxdmtc += E1000_READ_REG(hw, E1000_ICRXDMTC);
1812 	stats->icrxoc += E1000_READ_REG(hw, E1000_ICRXOC);
1813 
1814 	/* Host to Card Statistics */
1815 
1816 	stats->cbtmpc += E1000_READ_REG(hw, E1000_CBTMPC);
1817 	stats->htdpmc += E1000_READ_REG(hw, E1000_HTDPMC);
1818 	stats->cbrdpc += E1000_READ_REG(hw, E1000_CBRDPC);
1819 	stats->cbrmpc += E1000_READ_REG(hw, E1000_CBRMPC);
1820 	stats->rpthc += E1000_READ_REG(hw, E1000_RPTHC);
1821 	stats->hgptc += E1000_READ_REG(hw, E1000_HGPTC);
1822 	stats->htcbdpc += E1000_READ_REG(hw, E1000_HTCBDPC);
1823 	stats->hgorc += E1000_READ_REG(hw, E1000_HGORCL);
1824 	stats->hgorc += ((uint64_t)E1000_READ_REG(hw, E1000_HGORCH) << 32);
1825 	stats->hgorc -= (stats->rpthc - old_rpthc) * ETHER_CRC_LEN;
1826 	stats->hgotc += E1000_READ_REG(hw, E1000_HGOTCL);
1827 	stats->hgotc += ((uint64_t)E1000_READ_REG(hw, E1000_HGOTCH) << 32);
1828 	stats->hgotc -= (stats->hgptc - old_hgptc) * ETHER_CRC_LEN;
1829 	stats->lenerrs += E1000_READ_REG(hw, E1000_LENERRS);
1830 	stats->scvpc += E1000_READ_REG(hw, E1000_SCVPC);
1831 	stats->hrmpc += E1000_READ_REG(hw, E1000_HRMPC);
1832 
1833 	stats->algnerrc += E1000_READ_REG(hw, E1000_ALGNERRC);
1834 	stats->rxerrc += E1000_READ_REG(hw, E1000_RXERRC);
1835 	stats->tncrs += E1000_READ_REG(hw, E1000_TNCRS);
1836 	stats->cexterr += E1000_READ_REG(hw, E1000_CEXTERR);
1837 	stats->tsctc += E1000_READ_REG(hw, E1000_TSCTC);
1838 	stats->tsctfc += E1000_READ_REG(hw, E1000_TSCTFC);
1839 }
1840 
1841 static int
1842 eth_igb_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *rte_stats)
1843 {
1844 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1845 	struct e1000_hw_stats *stats =
1846 			E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
1847 
1848 	igb_read_stats_registers(hw, stats);
1849 
1850 	if (rte_stats == NULL)
1851 		return -EINVAL;
1852 
1853 	/* Rx Errors */
1854 	rte_stats->imissed = stats->mpc;
1855 	rte_stats->ierrors = stats->crcerrs +
1856 	                     stats->rlec + stats->ruc + stats->roc +
1857 	                     stats->rxerrc + stats->algnerrc + stats->cexterr;
1858 
1859 	/* Tx Errors */
1860 	rte_stats->oerrors = stats->ecol + stats->latecol;
1861 
1862 	rte_stats->ipackets = stats->gprc;
1863 	rte_stats->opackets = stats->gptc;
1864 	rte_stats->ibytes   = stats->gorc;
1865 	rte_stats->obytes   = stats->gotc;
1866 	return 0;
1867 }
1868 
1869 static void
1870 eth_igb_stats_reset(struct rte_eth_dev *dev)
1871 {
1872 	struct e1000_hw_stats *hw_stats =
1873 			E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
1874 
1875 	/* HW registers are cleared on read */
1876 	eth_igb_stats_get(dev, NULL);
1877 
1878 	/* Reset software totals */
1879 	memset(hw_stats, 0, sizeof(*hw_stats));
1880 }
1881 
1882 static void
1883 eth_igb_xstats_reset(struct rte_eth_dev *dev)
1884 {
1885 	struct e1000_hw_stats *stats =
1886 			E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
1887 
1888 	/* HW registers are cleared on read */
1889 	eth_igb_xstats_get(dev, NULL, IGB_NB_XSTATS);
1890 
1891 	/* Reset software totals */
1892 	memset(stats, 0, sizeof(*stats));
1893 }
1894 
1895 static int eth_igb_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
1896 	struct rte_eth_xstat_name *xstats_names,
1897 	__rte_unused unsigned int size)
1898 {
1899 	unsigned i;
1900 
1901 	if (xstats_names == NULL)
1902 		return IGB_NB_XSTATS;
1903 
1904 	/* Note: limit checked in rte_eth_xstats_names() */
1905 
1906 	for (i = 0; i < IGB_NB_XSTATS; i++) {
1907 		snprintf(xstats_names[i].name, sizeof(xstats_names[i].name),
1908 			 "%s", rte_igb_stats_strings[i].name);
1909 	}
1910 
1911 	return IGB_NB_XSTATS;
1912 }
1913 
1914 static int eth_igb_xstats_get_names_by_id(struct rte_eth_dev *dev,
1915 		struct rte_eth_xstat_name *xstats_names, const uint64_t *ids,
1916 		unsigned int limit)
1917 {
1918 	unsigned int i;
1919 
1920 	if (!ids) {
1921 		if (xstats_names == NULL)
1922 			return IGB_NB_XSTATS;
1923 
1924 		for (i = 0; i < IGB_NB_XSTATS; i++)
1925 			snprintf(xstats_names[i].name,
1926 					sizeof(xstats_names[i].name),
1927 					"%s", rte_igb_stats_strings[i].name);
1928 
1929 		return IGB_NB_XSTATS;
1930 
1931 	} else {
1932 		struct rte_eth_xstat_name xstats_names_copy[IGB_NB_XSTATS];
1933 
1934 		eth_igb_xstats_get_names_by_id(dev, xstats_names_copy, NULL,
1935 				IGB_NB_XSTATS);
1936 
1937 		for (i = 0; i < limit; i++) {
1938 			if (ids[i] >= IGB_NB_XSTATS) {
1939 				PMD_INIT_LOG(ERR, "id value isn't valid");
1940 				return -1;
1941 			}
1942 			strcpy(xstats_names[i].name,
1943 					xstats_names_copy[ids[i]].name);
1944 		}
1945 		return limit;
1946 	}
1947 }
1948 
1949 static int
1950 eth_igb_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1951 		   unsigned n)
1952 {
1953 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1954 	struct e1000_hw_stats *hw_stats =
1955 			E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
1956 	unsigned i;
1957 
1958 	if (n < IGB_NB_XSTATS)
1959 		return IGB_NB_XSTATS;
1960 
1961 	igb_read_stats_registers(hw, hw_stats);
1962 
1963 	/* If this is a reset xstats is NULL, and we have cleared the
1964 	 * registers by reading them.
1965 	 */
1966 	if (!xstats)
1967 		return 0;
1968 
1969 	/* Extended stats */
1970 	for (i = 0; i < IGB_NB_XSTATS; i++) {
1971 		xstats[i].id = i;
1972 		xstats[i].value = *(uint64_t *)(((char *)hw_stats) +
1973 			rte_igb_stats_strings[i].offset);
1974 	}
1975 
1976 	return IGB_NB_XSTATS;
1977 }
1978 
1979 static int
1980 eth_igb_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
1981 		uint64_t *values, unsigned int n)
1982 {
1983 	unsigned int i;
1984 
1985 	if (!ids) {
1986 		struct e1000_hw *hw =
1987 			E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1988 		struct e1000_hw_stats *hw_stats =
1989 			E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
1990 
1991 		if (n < IGB_NB_XSTATS)
1992 			return IGB_NB_XSTATS;
1993 
1994 		igb_read_stats_registers(hw, hw_stats);
1995 
1996 		/* If this is a reset xstats is NULL, and we have cleared the
1997 		 * registers by reading them.
1998 		 */
1999 		if (!values)
2000 			return 0;
2001 
2002 		/* Extended stats */
2003 		for (i = 0; i < IGB_NB_XSTATS; i++)
2004 			values[i] = *(uint64_t *)(((char *)hw_stats) +
2005 					rte_igb_stats_strings[i].offset);
2006 
2007 		return IGB_NB_XSTATS;
2008 
2009 	} else {
2010 		uint64_t values_copy[IGB_NB_XSTATS];
2011 
2012 		eth_igb_xstats_get_by_id(dev, NULL, values_copy,
2013 				IGB_NB_XSTATS);
2014 
2015 		for (i = 0; i < n; i++) {
2016 			if (ids[i] >= IGB_NB_XSTATS) {
2017 				PMD_INIT_LOG(ERR, "id value isn't valid");
2018 				return -1;
2019 			}
2020 			values[i] = values_copy[ids[i]];
2021 		}
2022 		return n;
2023 	}
2024 }
2025 
2026 static void
2027 igbvf_read_stats_registers(struct e1000_hw *hw, struct e1000_vf_stats *hw_stats)
2028 {
2029 	/* Good Rx packets, include VF loopback */
2030 	UPDATE_VF_STAT(E1000_VFGPRC,
2031 	    hw_stats->last_gprc, hw_stats->gprc);
2032 
2033 	/* Good Rx octets, include VF loopback */
2034 	UPDATE_VF_STAT(E1000_VFGORC,
2035 	    hw_stats->last_gorc, hw_stats->gorc);
2036 
2037 	/* Good Tx packets, include VF loopback */
2038 	UPDATE_VF_STAT(E1000_VFGPTC,
2039 	    hw_stats->last_gptc, hw_stats->gptc);
2040 
2041 	/* Good Tx octets, include VF loopback */
2042 	UPDATE_VF_STAT(E1000_VFGOTC,
2043 	    hw_stats->last_gotc, hw_stats->gotc);
2044 
2045 	/* Rx Multicst packets */
2046 	UPDATE_VF_STAT(E1000_VFMPRC,
2047 	    hw_stats->last_mprc, hw_stats->mprc);
2048 
2049 	/* Good Rx loopback packets */
2050 	UPDATE_VF_STAT(E1000_VFGPRLBC,
2051 	    hw_stats->last_gprlbc, hw_stats->gprlbc);
2052 
2053 	/* Good Rx loopback octets */
2054 	UPDATE_VF_STAT(E1000_VFGORLBC,
2055 	    hw_stats->last_gorlbc, hw_stats->gorlbc);
2056 
2057 	/* Good Tx loopback packets */
2058 	UPDATE_VF_STAT(E1000_VFGPTLBC,
2059 	    hw_stats->last_gptlbc, hw_stats->gptlbc);
2060 
2061 	/* Good Tx loopback octets */
2062 	UPDATE_VF_STAT(E1000_VFGOTLBC,
2063 	    hw_stats->last_gotlbc, hw_stats->gotlbc);
2064 }
2065 
2066 static int eth_igbvf_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
2067 				     struct rte_eth_xstat_name *xstats_names,
2068 				     __rte_unused unsigned limit)
2069 {
2070 	unsigned i;
2071 
2072 	if (xstats_names != NULL)
2073 		for (i = 0; i < IGBVF_NB_XSTATS; i++) {
2074 			snprintf(xstats_names[i].name,
2075 				sizeof(xstats_names[i].name), "%s",
2076 				rte_igbvf_stats_strings[i].name);
2077 		}
2078 	return IGBVF_NB_XSTATS;
2079 }
2080 
2081 static int
2082 eth_igbvf_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
2083 		     unsigned n)
2084 {
2085 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2086 	struct e1000_vf_stats *hw_stats = (struct e1000_vf_stats *)
2087 			E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
2088 	unsigned i;
2089 
2090 	if (n < IGBVF_NB_XSTATS)
2091 		return IGBVF_NB_XSTATS;
2092 
2093 	igbvf_read_stats_registers(hw, hw_stats);
2094 
2095 	if (!xstats)
2096 		return 0;
2097 
2098 	for (i = 0; i < IGBVF_NB_XSTATS; i++) {
2099 		xstats[i].id = i;
2100 		xstats[i].value = *(uint64_t *)(((char *)hw_stats) +
2101 			rte_igbvf_stats_strings[i].offset);
2102 	}
2103 
2104 	return IGBVF_NB_XSTATS;
2105 }
2106 
2107 static int
2108 eth_igbvf_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *rte_stats)
2109 {
2110 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2111 	struct e1000_vf_stats *hw_stats = (struct e1000_vf_stats *)
2112 			  E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
2113 
2114 	igbvf_read_stats_registers(hw, hw_stats);
2115 
2116 	if (rte_stats == NULL)
2117 		return -EINVAL;
2118 
2119 	rte_stats->ipackets = hw_stats->gprc;
2120 	rte_stats->ibytes = hw_stats->gorc;
2121 	rte_stats->opackets = hw_stats->gptc;
2122 	rte_stats->obytes = hw_stats->gotc;
2123 	return 0;
2124 }
2125 
2126 static void
2127 eth_igbvf_stats_reset(struct rte_eth_dev *dev)
2128 {
2129 	struct e1000_vf_stats *hw_stats = (struct e1000_vf_stats*)
2130 			E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
2131 
2132 	/* Sync HW register to the last stats */
2133 	eth_igbvf_stats_get(dev, NULL);
2134 
2135 	/* reset HW current stats*/
2136 	memset(&hw_stats->gprc, 0, sizeof(*hw_stats) -
2137 	       offsetof(struct e1000_vf_stats, gprc));
2138 }
2139 
2140 static int
2141 eth_igb_fw_version_get(struct rte_eth_dev *dev, char *fw_version,
2142 		       size_t fw_size)
2143 {
2144 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2145 	struct e1000_fw_version fw;
2146 	int ret;
2147 
2148 	e1000_get_fw_version(hw, &fw);
2149 
2150 	switch (hw->mac.type) {
2151 	case e1000_i210:
2152 	case e1000_i211:
2153 		if (!(e1000_get_flash_presence_i210(hw))) {
2154 			ret = snprintf(fw_version, fw_size,
2155 				 "%2d.%2d-%d",
2156 				 fw.invm_major, fw.invm_minor,
2157 				 fw.invm_img_type);
2158 			break;
2159 		}
2160 		/* fall through */
2161 	default:
2162 		/* if option rom is valid, display its version too */
2163 		if (fw.or_valid) {
2164 			ret = snprintf(fw_version, fw_size,
2165 				 "%d.%d, 0x%08x, %d.%d.%d",
2166 				 fw.eep_major, fw.eep_minor, fw.etrack_id,
2167 				 fw.or_major, fw.or_build, fw.or_patch);
2168 		/* no option rom */
2169 		} else {
2170 			if (fw.etrack_id != 0X0000) {
2171 				ret = snprintf(fw_version, fw_size,
2172 					 "%d.%d, 0x%08x",
2173 					 fw.eep_major, fw.eep_minor,
2174 					 fw.etrack_id);
2175 			} else {
2176 				ret = snprintf(fw_version, fw_size,
2177 					 "%d.%d.%d",
2178 					 fw.eep_major, fw.eep_minor,
2179 					 fw.eep_build);
2180 			}
2181 		}
2182 		break;
2183 	}
2184 
2185 	ret += 1; /* add the size of '\0' */
2186 	if (fw_size < (u32)ret)
2187 		return ret;
2188 	else
2189 		return 0;
2190 }
2191 
2192 static void
2193 eth_igb_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2194 {
2195 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2196 
2197 	dev_info->min_rx_bufsize = 256; /* See BSIZE field of RCTL register. */
2198 	dev_info->max_rx_pktlen  = 0x3FFF; /* See RLPML register. */
2199 	dev_info->max_mac_addrs = hw->mac.rar_entry_count;
2200 	dev_info->rx_queue_offload_capa = igb_get_rx_queue_offloads_capa(dev);
2201 	dev_info->rx_offload_capa = igb_get_rx_port_offloads_capa(dev) |
2202 				    dev_info->rx_queue_offload_capa;
2203 	dev_info->tx_queue_offload_capa = igb_get_tx_queue_offloads_capa(dev);
2204 	dev_info->tx_offload_capa = igb_get_tx_port_offloads_capa(dev) |
2205 				    dev_info->tx_queue_offload_capa;
2206 
2207 	switch (hw->mac.type) {
2208 	case e1000_82575:
2209 		dev_info->max_rx_queues = 4;
2210 		dev_info->max_tx_queues = 4;
2211 		dev_info->max_vmdq_pools = 0;
2212 		break;
2213 
2214 	case e1000_82576:
2215 		dev_info->max_rx_queues = 16;
2216 		dev_info->max_tx_queues = 16;
2217 		dev_info->max_vmdq_pools = ETH_8_POOLS;
2218 		dev_info->vmdq_queue_num = 16;
2219 		break;
2220 
2221 	case e1000_82580:
2222 		dev_info->max_rx_queues = 8;
2223 		dev_info->max_tx_queues = 8;
2224 		dev_info->max_vmdq_pools = ETH_8_POOLS;
2225 		dev_info->vmdq_queue_num = 8;
2226 		break;
2227 
2228 	case e1000_i350:
2229 		dev_info->max_rx_queues = 8;
2230 		dev_info->max_tx_queues = 8;
2231 		dev_info->max_vmdq_pools = ETH_8_POOLS;
2232 		dev_info->vmdq_queue_num = 8;
2233 		break;
2234 
2235 	case e1000_i354:
2236 		dev_info->max_rx_queues = 8;
2237 		dev_info->max_tx_queues = 8;
2238 		break;
2239 
2240 	case e1000_i210:
2241 		dev_info->max_rx_queues = 4;
2242 		dev_info->max_tx_queues = 4;
2243 		dev_info->max_vmdq_pools = 0;
2244 		break;
2245 
2246 	case e1000_i211:
2247 		dev_info->max_rx_queues = 2;
2248 		dev_info->max_tx_queues = 2;
2249 		dev_info->max_vmdq_pools = 0;
2250 		break;
2251 
2252 	default:
2253 		/* Should not happen */
2254 		break;
2255 	}
2256 	dev_info->hash_key_size = IGB_HKEY_MAX_INDEX * sizeof(uint32_t);
2257 	dev_info->reta_size = ETH_RSS_RETA_SIZE_128;
2258 	dev_info->flow_type_rss_offloads = IGB_RSS_OFFLOAD_ALL;
2259 
2260 	dev_info->default_rxconf = (struct rte_eth_rxconf) {
2261 		.rx_thresh = {
2262 			.pthresh = IGB_DEFAULT_RX_PTHRESH,
2263 			.hthresh = IGB_DEFAULT_RX_HTHRESH,
2264 			.wthresh = IGB_DEFAULT_RX_WTHRESH,
2265 		},
2266 		.rx_free_thresh = IGB_DEFAULT_RX_FREE_THRESH,
2267 		.rx_drop_en = 0,
2268 		.offloads = 0,
2269 	};
2270 
2271 	dev_info->default_txconf = (struct rte_eth_txconf) {
2272 		.tx_thresh = {
2273 			.pthresh = IGB_DEFAULT_TX_PTHRESH,
2274 			.hthresh = IGB_DEFAULT_TX_HTHRESH,
2275 			.wthresh = IGB_DEFAULT_TX_WTHRESH,
2276 		},
2277 		.offloads = 0,
2278 	};
2279 
2280 	dev_info->rx_desc_lim = rx_desc_lim;
2281 	dev_info->tx_desc_lim = tx_desc_lim;
2282 
2283 	dev_info->speed_capa = ETH_LINK_SPEED_10M_HD | ETH_LINK_SPEED_10M |
2284 			ETH_LINK_SPEED_100M_HD | ETH_LINK_SPEED_100M |
2285 			ETH_LINK_SPEED_1G;
2286 }
2287 
2288 static const uint32_t *
2289 eth_igb_supported_ptypes_get(struct rte_eth_dev *dev)
2290 {
2291 	static const uint32_t ptypes[] = {
2292 		/* refers to igb_rxd_pkt_info_to_pkt_type() */
2293 		RTE_PTYPE_L2_ETHER,
2294 		RTE_PTYPE_L3_IPV4,
2295 		RTE_PTYPE_L3_IPV4_EXT,
2296 		RTE_PTYPE_L3_IPV6,
2297 		RTE_PTYPE_L3_IPV6_EXT,
2298 		RTE_PTYPE_L4_TCP,
2299 		RTE_PTYPE_L4_UDP,
2300 		RTE_PTYPE_L4_SCTP,
2301 		RTE_PTYPE_TUNNEL_IP,
2302 		RTE_PTYPE_INNER_L3_IPV6,
2303 		RTE_PTYPE_INNER_L3_IPV6_EXT,
2304 		RTE_PTYPE_INNER_L4_TCP,
2305 		RTE_PTYPE_INNER_L4_UDP,
2306 		RTE_PTYPE_UNKNOWN
2307 	};
2308 
2309 	if (dev->rx_pkt_burst == eth_igb_recv_pkts ||
2310 	    dev->rx_pkt_burst == eth_igb_recv_scattered_pkts)
2311 		return ptypes;
2312 	return NULL;
2313 }
2314 
2315 static void
2316 eth_igbvf_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2317 {
2318 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2319 
2320 	dev_info->min_rx_bufsize = 256; /* See BSIZE field of RCTL register. */
2321 	dev_info->max_rx_pktlen  = 0x3FFF; /* See RLPML register. */
2322 	dev_info->max_mac_addrs = hw->mac.rar_entry_count;
2323 	dev_info->tx_offload_capa = DEV_TX_OFFLOAD_VLAN_INSERT |
2324 				DEV_TX_OFFLOAD_IPV4_CKSUM  |
2325 				DEV_TX_OFFLOAD_UDP_CKSUM   |
2326 				DEV_TX_OFFLOAD_TCP_CKSUM   |
2327 				DEV_TX_OFFLOAD_SCTP_CKSUM  |
2328 				DEV_TX_OFFLOAD_TCP_TSO;
2329 	switch (hw->mac.type) {
2330 	case e1000_vfadapt:
2331 		dev_info->max_rx_queues = 2;
2332 		dev_info->max_tx_queues = 2;
2333 		break;
2334 	case e1000_vfadapt_i350:
2335 		dev_info->max_rx_queues = 1;
2336 		dev_info->max_tx_queues = 1;
2337 		break;
2338 	default:
2339 		/* Should not happen */
2340 		break;
2341 	}
2342 
2343 	dev_info->rx_queue_offload_capa = igb_get_rx_queue_offloads_capa(dev);
2344 	dev_info->rx_offload_capa = igb_get_rx_port_offloads_capa(dev) |
2345 				    dev_info->rx_queue_offload_capa;
2346 	dev_info->tx_queue_offload_capa = igb_get_tx_queue_offloads_capa(dev);
2347 	dev_info->tx_offload_capa = igb_get_tx_port_offloads_capa(dev) |
2348 				    dev_info->tx_queue_offload_capa;
2349 
2350 	dev_info->default_rxconf = (struct rte_eth_rxconf) {
2351 		.rx_thresh = {
2352 			.pthresh = IGB_DEFAULT_RX_PTHRESH,
2353 			.hthresh = IGB_DEFAULT_RX_HTHRESH,
2354 			.wthresh = IGB_DEFAULT_RX_WTHRESH,
2355 		},
2356 		.rx_free_thresh = IGB_DEFAULT_RX_FREE_THRESH,
2357 		.rx_drop_en = 0,
2358 		.offloads = 0,
2359 	};
2360 
2361 	dev_info->default_txconf = (struct rte_eth_txconf) {
2362 		.tx_thresh = {
2363 			.pthresh = IGB_DEFAULT_TX_PTHRESH,
2364 			.hthresh = IGB_DEFAULT_TX_HTHRESH,
2365 			.wthresh = IGB_DEFAULT_TX_WTHRESH,
2366 		},
2367 		.offloads = 0,
2368 	};
2369 
2370 	dev_info->rx_desc_lim = rx_desc_lim;
2371 	dev_info->tx_desc_lim = tx_desc_lim;
2372 }
2373 
2374 /* return 0 means link status changed, -1 means not changed */
2375 static int
2376 eth_igb_link_update(struct rte_eth_dev *dev, int wait_to_complete)
2377 {
2378 	struct e1000_hw *hw =
2379 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2380 	struct rte_eth_link link;
2381 	int link_check, count;
2382 
2383 	link_check = 0;
2384 	hw->mac.get_link_status = 1;
2385 
2386 	/* possible wait-to-complete in up to 9 seconds */
2387 	for (count = 0; count < IGB_LINK_UPDATE_CHECK_TIMEOUT; count ++) {
2388 		/* Read the real link status */
2389 		switch (hw->phy.media_type) {
2390 		case e1000_media_type_copper:
2391 			/* Do the work to read phy */
2392 			e1000_check_for_link(hw);
2393 			link_check = !hw->mac.get_link_status;
2394 			break;
2395 
2396 		case e1000_media_type_fiber:
2397 			e1000_check_for_link(hw);
2398 			link_check = (E1000_READ_REG(hw, E1000_STATUS) &
2399 				      E1000_STATUS_LU);
2400 			break;
2401 
2402 		case e1000_media_type_internal_serdes:
2403 			e1000_check_for_link(hw);
2404 			link_check = hw->mac.serdes_has_link;
2405 			break;
2406 
2407 		/* VF device is type_unknown */
2408 		case e1000_media_type_unknown:
2409 			eth_igbvf_link_update(hw);
2410 			link_check = !hw->mac.get_link_status;
2411 			break;
2412 
2413 		default:
2414 			break;
2415 		}
2416 		if (link_check || wait_to_complete == 0)
2417 			break;
2418 		rte_delay_ms(IGB_LINK_UPDATE_CHECK_INTERVAL);
2419 	}
2420 	memset(&link, 0, sizeof(link));
2421 
2422 	/* Now we check if a transition has happened */
2423 	if (link_check) {
2424 		uint16_t duplex, speed;
2425 		hw->mac.ops.get_link_up_info(hw, &speed, &duplex);
2426 		link.link_duplex = (duplex == FULL_DUPLEX) ?
2427 				ETH_LINK_FULL_DUPLEX :
2428 				ETH_LINK_HALF_DUPLEX;
2429 		link.link_speed = speed;
2430 		link.link_status = ETH_LINK_UP;
2431 		link.link_autoneg = !(dev->data->dev_conf.link_speeds &
2432 				ETH_LINK_SPEED_FIXED);
2433 	} else if (!link_check) {
2434 		link.link_speed = 0;
2435 		link.link_duplex = ETH_LINK_HALF_DUPLEX;
2436 		link.link_status = ETH_LINK_DOWN;
2437 		link.link_autoneg = ETH_LINK_FIXED;
2438 	}
2439 
2440 	return rte_eth_linkstatus_set(dev, &link);
2441 }
2442 
2443 /*
2444  * igb_hw_control_acquire sets CTRL_EXT:DRV_LOAD bit.
2445  * For ASF and Pass Through versions of f/w this means
2446  * that the driver is loaded.
2447  */
2448 static void
2449 igb_hw_control_acquire(struct e1000_hw *hw)
2450 {
2451 	uint32_t ctrl_ext;
2452 
2453 	/* Let firmware know the driver has taken over */
2454 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
2455 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2456 }
2457 
2458 /*
2459  * igb_hw_control_release resets CTRL_EXT:DRV_LOAD bit.
2460  * For ASF and Pass Through versions of f/w this means that the
2461  * driver is no longer loaded.
2462  */
2463 static void
2464 igb_hw_control_release(struct e1000_hw *hw)
2465 {
2466 	uint32_t ctrl_ext;
2467 
2468 	/* Let firmware taken over control of h/w */
2469 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
2470 	E1000_WRITE_REG(hw, E1000_CTRL_EXT,
2471 			ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2472 }
2473 
2474 /*
2475  * Bit of a misnomer, what this really means is
2476  * to enable OS management of the system... aka
2477  * to disable special hardware management features.
2478  */
2479 static void
2480 igb_init_manageability(struct e1000_hw *hw)
2481 {
2482 	if (e1000_enable_mng_pass_thru(hw)) {
2483 		uint32_t manc2h = E1000_READ_REG(hw, E1000_MANC2H);
2484 		uint32_t manc = E1000_READ_REG(hw, E1000_MANC);
2485 
2486 		/* disable hardware interception of ARP */
2487 		manc &= ~(E1000_MANC_ARP_EN);
2488 
2489 		/* enable receiving management packets to the host */
2490 		manc |= E1000_MANC_EN_MNG2HOST;
2491 		manc2h |= 1 << 5;  /* Mng Port 623 */
2492 		manc2h |= 1 << 6;  /* Mng Port 664 */
2493 		E1000_WRITE_REG(hw, E1000_MANC2H, manc2h);
2494 		E1000_WRITE_REG(hw, E1000_MANC, manc);
2495 	}
2496 }
2497 
2498 static void
2499 igb_release_manageability(struct e1000_hw *hw)
2500 {
2501 	if (e1000_enable_mng_pass_thru(hw)) {
2502 		uint32_t manc = E1000_READ_REG(hw, E1000_MANC);
2503 
2504 		manc |= E1000_MANC_ARP_EN;
2505 		manc &= ~E1000_MANC_EN_MNG2HOST;
2506 
2507 		E1000_WRITE_REG(hw, E1000_MANC, manc);
2508 	}
2509 }
2510 
2511 static void
2512 eth_igb_promiscuous_enable(struct rte_eth_dev *dev)
2513 {
2514 	struct e1000_hw *hw =
2515 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2516 	uint32_t rctl;
2517 
2518 	rctl = E1000_READ_REG(hw, E1000_RCTL);
2519 	rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2520 	E1000_WRITE_REG(hw, E1000_RCTL, rctl);
2521 }
2522 
2523 static void
2524 eth_igb_promiscuous_disable(struct rte_eth_dev *dev)
2525 {
2526 	struct e1000_hw *hw =
2527 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2528 	uint32_t rctl;
2529 
2530 	rctl = E1000_READ_REG(hw, E1000_RCTL);
2531 	rctl &= (~E1000_RCTL_UPE);
2532 	if (dev->data->all_multicast == 1)
2533 		rctl |= E1000_RCTL_MPE;
2534 	else
2535 		rctl &= (~E1000_RCTL_MPE);
2536 	E1000_WRITE_REG(hw, E1000_RCTL, rctl);
2537 }
2538 
2539 static void
2540 eth_igb_allmulticast_enable(struct rte_eth_dev *dev)
2541 {
2542 	struct e1000_hw *hw =
2543 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2544 	uint32_t rctl;
2545 
2546 	rctl = E1000_READ_REG(hw, E1000_RCTL);
2547 	rctl |= E1000_RCTL_MPE;
2548 	E1000_WRITE_REG(hw, E1000_RCTL, rctl);
2549 }
2550 
2551 static void
2552 eth_igb_allmulticast_disable(struct rte_eth_dev *dev)
2553 {
2554 	struct e1000_hw *hw =
2555 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2556 	uint32_t rctl;
2557 
2558 	if (dev->data->promiscuous == 1)
2559 		return; /* must remain in all_multicast mode */
2560 	rctl = E1000_READ_REG(hw, E1000_RCTL);
2561 	rctl &= (~E1000_RCTL_MPE);
2562 	E1000_WRITE_REG(hw, E1000_RCTL, rctl);
2563 }
2564 
2565 static int
2566 eth_igb_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
2567 {
2568 	struct e1000_hw *hw =
2569 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2570 	struct e1000_vfta * shadow_vfta =
2571 		E1000_DEV_PRIVATE_TO_VFTA(dev->data->dev_private);
2572 	uint32_t vfta;
2573 	uint32_t vid_idx;
2574 	uint32_t vid_bit;
2575 
2576 	vid_idx = (uint32_t) ((vlan_id >> E1000_VFTA_ENTRY_SHIFT) &
2577 			      E1000_VFTA_ENTRY_MASK);
2578 	vid_bit = (uint32_t) (1 << (vlan_id & E1000_VFTA_ENTRY_BIT_SHIFT_MASK));
2579 	vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, vid_idx);
2580 	if (on)
2581 		vfta |= vid_bit;
2582 	else
2583 		vfta &= ~vid_bit;
2584 	E1000_WRITE_REG_ARRAY(hw, E1000_VFTA, vid_idx, vfta);
2585 
2586 	/* update local VFTA copy */
2587 	shadow_vfta->vfta[vid_idx] = vfta;
2588 
2589 	return 0;
2590 }
2591 
2592 static int
2593 eth_igb_vlan_tpid_set(struct rte_eth_dev *dev,
2594 		      enum rte_vlan_type vlan_type,
2595 		      uint16_t tpid)
2596 {
2597 	struct e1000_hw *hw =
2598 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2599 	uint32_t reg, qinq;
2600 
2601 	qinq = E1000_READ_REG(hw, E1000_CTRL_EXT);
2602 	qinq &= E1000_CTRL_EXT_EXT_VLAN;
2603 
2604 	/* only outer TPID of double VLAN can be configured*/
2605 	if (qinq && vlan_type == ETH_VLAN_TYPE_OUTER) {
2606 		reg = E1000_READ_REG(hw, E1000_VET);
2607 		reg = (reg & (~E1000_VET_VET_EXT)) |
2608 			((uint32_t)tpid << E1000_VET_VET_EXT_SHIFT);
2609 		E1000_WRITE_REG(hw, E1000_VET, reg);
2610 
2611 		return 0;
2612 	}
2613 
2614 	/* all other TPID values are read-only*/
2615 	PMD_DRV_LOG(ERR, "Not supported");
2616 
2617 	return -ENOTSUP;
2618 }
2619 
2620 static void
2621 igb_vlan_hw_filter_disable(struct rte_eth_dev *dev)
2622 {
2623 	struct e1000_hw *hw =
2624 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2625 	uint32_t reg;
2626 
2627 	/* Filter Table Disable */
2628 	reg = E1000_READ_REG(hw, E1000_RCTL);
2629 	reg &= ~E1000_RCTL_CFIEN;
2630 	reg &= ~E1000_RCTL_VFE;
2631 	E1000_WRITE_REG(hw, E1000_RCTL, reg);
2632 }
2633 
2634 static void
2635 igb_vlan_hw_filter_enable(struct rte_eth_dev *dev)
2636 {
2637 	struct e1000_hw *hw =
2638 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2639 	struct e1000_vfta * shadow_vfta =
2640 		E1000_DEV_PRIVATE_TO_VFTA(dev->data->dev_private);
2641 	uint32_t reg;
2642 	int i;
2643 
2644 	/* Filter Table Enable, CFI not used for packet acceptance */
2645 	reg = E1000_READ_REG(hw, E1000_RCTL);
2646 	reg &= ~E1000_RCTL_CFIEN;
2647 	reg |= E1000_RCTL_VFE;
2648 	E1000_WRITE_REG(hw, E1000_RCTL, reg);
2649 
2650 	/* restore VFTA table */
2651 	for (i = 0; i < IGB_VFTA_SIZE; i++)
2652 		E1000_WRITE_REG_ARRAY(hw, E1000_VFTA, i, shadow_vfta->vfta[i]);
2653 }
2654 
2655 static void
2656 igb_vlan_hw_strip_disable(struct rte_eth_dev *dev)
2657 {
2658 	struct e1000_hw *hw =
2659 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2660 	uint32_t reg;
2661 
2662 	/* VLAN Mode Disable */
2663 	reg = E1000_READ_REG(hw, E1000_CTRL);
2664 	reg &= ~E1000_CTRL_VME;
2665 	E1000_WRITE_REG(hw, E1000_CTRL, reg);
2666 }
2667 
2668 static void
2669 igb_vlan_hw_strip_enable(struct rte_eth_dev *dev)
2670 {
2671 	struct e1000_hw *hw =
2672 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2673 	uint32_t reg;
2674 
2675 	/* VLAN Mode Enable */
2676 	reg = E1000_READ_REG(hw, E1000_CTRL);
2677 	reg |= E1000_CTRL_VME;
2678 	E1000_WRITE_REG(hw, E1000_CTRL, reg);
2679 }
2680 
2681 static void
2682 igb_vlan_hw_extend_disable(struct rte_eth_dev *dev)
2683 {
2684 	struct e1000_hw *hw =
2685 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2686 	uint32_t reg;
2687 
2688 	/* CTRL_EXT: Extended VLAN */
2689 	reg = E1000_READ_REG(hw, E1000_CTRL_EXT);
2690 	reg &= ~E1000_CTRL_EXT_EXTEND_VLAN;
2691 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, reg);
2692 
2693 	/* Update maximum packet length */
2694 	if (dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_JUMBO_FRAME)
2695 		E1000_WRITE_REG(hw, E1000_RLPML,
2696 			dev->data->dev_conf.rxmode.max_rx_pkt_len +
2697 						VLAN_TAG_SIZE);
2698 }
2699 
2700 static void
2701 igb_vlan_hw_extend_enable(struct rte_eth_dev *dev)
2702 {
2703 	struct e1000_hw *hw =
2704 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2705 	uint32_t reg;
2706 
2707 	/* CTRL_EXT: Extended VLAN */
2708 	reg = E1000_READ_REG(hw, E1000_CTRL_EXT);
2709 	reg |= E1000_CTRL_EXT_EXTEND_VLAN;
2710 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, reg);
2711 
2712 	/* Update maximum packet length */
2713 	if (dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_JUMBO_FRAME)
2714 		E1000_WRITE_REG(hw, E1000_RLPML,
2715 			dev->data->dev_conf.rxmode.max_rx_pkt_len +
2716 						2 * VLAN_TAG_SIZE);
2717 }
2718 
2719 static int
2720 eth_igb_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2721 {
2722 	struct rte_eth_rxmode *rxmode;
2723 
2724 	rxmode = &dev->data->dev_conf.rxmode;
2725 	if(mask & ETH_VLAN_STRIP_MASK){
2726 		if (rxmode->offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2727 			igb_vlan_hw_strip_enable(dev);
2728 		else
2729 			igb_vlan_hw_strip_disable(dev);
2730 	}
2731 
2732 	if(mask & ETH_VLAN_FILTER_MASK){
2733 		if (rxmode->offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
2734 			igb_vlan_hw_filter_enable(dev);
2735 		else
2736 			igb_vlan_hw_filter_disable(dev);
2737 	}
2738 
2739 	if(mask & ETH_VLAN_EXTEND_MASK){
2740 		if (rxmode->offloads & DEV_RX_OFFLOAD_VLAN_EXTEND)
2741 			igb_vlan_hw_extend_enable(dev);
2742 		else
2743 			igb_vlan_hw_extend_disable(dev);
2744 	}
2745 
2746 	return 0;
2747 }
2748 
2749 
2750 /**
2751  * It enables the interrupt mask and then enable the interrupt.
2752  *
2753  * @param dev
2754  *  Pointer to struct rte_eth_dev.
2755  * @param on
2756  *  Enable or Disable
2757  *
2758  * @return
2759  *  - On success, zero.
2760  *  - On failure, a negative value.
2761  */
2762 static int
2763 eth_igb_lsc_interrupt_setup(struct rte_eth_dev *dev, uint8_t on)
2764 {
2765 	struct e1000_interrupt *intr =
2766 		E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
2767 
2768 	if (on)
2769 		intr->mask |= E1000_ICR_LSC;
2770 	else
2771 		intr->mask &= ~E1000_ICR_LSC;
2772 
2773 	return 0;
2774 }
2775 
2776 /* It clears the interrupt causes and enables the interrupt.
2777  * It will be called once only during nic initialized.
2778  *
2779  * @param dev
2780  *  Pointer to struct rte_eth_dev.
2781  *
2782  * @return
2783  *  - On success, zero.
2784  *  - On failure, a negative value.
2785  */
2786 static int eth_igb_rxq_interrupt_setup(struct rte_eth_dev *dev)
2787 {
2788 	uint32_t mask, regval;
2789 	struct e1000_hw *hw =
2790 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2791 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2792 	struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
2793 	int misc_shift = rte_intr_allow_others(intr_handle) ? 1 : 0;
2794 	struct rte_eth_dev_info dev_info;
2795 
2796 	memset(&dev_info, 0, sizeof(dev_info));
2797 	eth_igb_infos_get(dev, &dev_info);
2798 
2799 	mask = (0xFFFFFFFF >> (32 - dev_info.max_rx_queues)) << misc_shift;
2800 	regval = E1000_READ_REG(hw, E1000_EIMS);
2801 	E1000_WRITE_REG(hw, E1000_EIMS, regval | mask);
2802 
2803 	return 0;
2804 }
2805 
2806 /*
2807  * It reads ICR and gets interrupt causes, check it and set a bit flag
2808  * to update link status.
2809  *
2810  * @param dev
2811  *  Pointer to struct rte_eth_dev.
2812  *
2813  * @return
2814  *  - On success, zero.
2815  *  - On failure, a negative value.
2816  */
2817 static int
2818 eth_igb_interrupt_get_status(struct rte_eth_dev *dev)
2819 {
2820 	uint32_t icr;
2821 	struct e1000_hw *hw =
2822 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2823 	struct e1000_interrupt *intr =
2824 		E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
2825 
2826 	igb_intr_disable(dev);
2827 
2828 	/* read-on-clear nic registers here */
2829 	icr = E1000_READ_REG(hw, E1000_ICR);
2830 
2831 	intr->flags = 0;
2832 	if (icr & E1000_ICR_LSC) {
2833 		intr->flags |= E1000_FLAG_NEED_LINK_UPDATE;
2834 	}
2835 
2836 	if (icr & E1000_ICR_VMMB)
2837 		intr->flags |= E1000_FLAG_MAILBOX;
2838 
2839 	return 0;
2840 }
2841 
2842 /*
2843  * It executes link_update after knowing an interrupt is prsent.
2844  *
2845  * @param dev
2846  *  Pointer to struct rte_eth_dev.
2847  *
2848  * @return
2849  *  - On success, zero.
2850  *  - On failure, a negative value.
2851  */
2852 static int
2853 eth_igb_interrupt_action(struct rte_eth_dev *dev,
2854 			 struct rte_intr_handle *intr_handle)
2855 {
2856 	struct e1000_hw *hw =
2857 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2858 	struct e1000_interrupt *intr =
2859 		E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
2860 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2861 	struct rte_eth_link link;
2862 	int ret;
2863 
2864 	if (intr->flags & E1000_FLAG_MAILBOX) {
2865 		igb_pf_mbx_process(dev);
2866 		intr->flags &= ~E1000_FLAG_MAILBOX;
2867 	}
2868 
2869 	igb_intr_enable(dev);
2870 	rte_intr_enable(intr_handle);
2871 
2872 	if (intr->flags & E1000_FLAG_NEED_LINK_UPDATE) {
2873 		intr->flags &= ~E1000_FLAG_NEED_LINK_UPDATE;
2874 
2875 		/* set get_link_status to check register later */
2876 		hw->mac.get_link_status = 1;
2877 		ret = eth_igb_link_update(dev, 0);
2878 
2879 		/* check if link has changed */
2880 		if (ret < 0)
2881 			return 0;
2882 
2883 		rte_eth_linkstatus_get(dev, &link);
2884 		if (link.link_status) {
2885 			PMD_INIT_LOG(INFO,
2886 				     " Port %d: Link Up - speed %u Mbps - %s",
2887 				     dev->data->port_id,
2888 				     (unsigned)link.link_speed,
2889 				     link.link_duplex == ETH_LINK_FULL_DUPLEX ?
2890 				     "full-duplex" : "half-duplex");
2891 		} else {
2892 			PMD_INIT_LOG(INFO, " Port %d: Link Down",
2893 				     dev->data->port_id);
2894 		}
2895 
2896 		PMD_INIT_LOG(DEBUG, "PCI Address: %04d:%02d:%02d:%d",
2897 			     pci_dev->addr.domain,
2898 			     pci_dev->addr.bus,
2899 			     pci_dev->addr.devid,
2900 			     pci_dev->addr.function);
2901 		_rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC,
2902 					      NULL);
2903 	}
2904 
2905 	return 0;
2906 }
2907 
2908 /**
2909  * Interrupt handler which shall be registered at first.
2910  *
2911  * @param handle
2912  *  Pointer to interrupt handle.
2913  * @param param
2914  *  The address of parameter (struct rte_eth_dev *) regsitered before.
2915  *
2916  * @return
2917  *  void
2918  */
2919 static void
2920 eth_igb_interrupt_handler(void *param)
2921 {
2922 	struct rte_eth_dev *dev = (struct rte_eth_dev *)param;
2923 
2924 	eth_igb_interrupt_get_status(dev);
2925 	eth_igb_interrupt_action(dev, dev->intr_handle);
2926 }
2927 
2928 static int
2929 eth_igbvf_interrupt_get_status(struct rte_eth_dev *dev)
2930 {
2931 	uint32_t eicr;
2932 	struct e1000_hw *hw =
2933 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2934 	struct e1000_interrupt *intr =
2935 		E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
2936 
2937 	igbvf_intr_disable(hw);
2938 
2939 	/* read-on-clear nic registers here */
2940 	eicr = E1000_READ_REG(hw, E1000_EICR);
2941 	intr->flags = 0;
2942 
2943 	if (eicr == E1000_VTIVAR_MISC_MAILBOX)
2944 		intr->flags |= E1000_FLAG_MAILBOX;
2945 
2946 	return 0;
2947 }
2948 
2949 void igbvf_mbx_process(struct rte_eth_dev *dev)
2950 {
2951 	struct e1000_hw *hw =
2952 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2953 	struct e1000_mbx_info *mbx = &hw->mbx;
2954 	u32 in_msg = 0;
2955 
2956 	/* peek the message first */
2957 	in_msg = E1000_READ_REG(hw, E1000_VMBMEM(0));
2958 
2959 	/* PF reset VF event */
2960 	if (in_msg == E1000_PF_CONTROL_MSG) {
2961 		/* dummy mbx read to ack pf */
2962 		if (mbx->ops.read(hw, &in_msg, 1, 0))
2963 			return;
2964 		_rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_RESET,
2965 					      NULL);
2966 	}
2967 }
2968 
2969 static int
2970 eth_igbvf_interrupt_action(struct rte_eth_dev *dev, struct rte_intr_handle *intr_handle)
2971 {
2972 	struct e1000_interrupt *intr =
2973 		E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
2974 
2975 	if (intr->flags & E1000_FLAG_MAILBOX) {
2976 		igbvf_mbx_process(dev);
2977 		intr->flags &= ~E1000_FLAG_MAILBOX;
2978 	}
2979 
2980 	igbvf_intr_enable(dev);
2981 	rte_intr_enable(intr_handle);
2982 
2983 	return 0;
2984 }
2985 
2986 static void
2987 eth_igbvf_interrupt_handler(void *param)
2988 {
2989 	struct rte_eth_dev *dev = (struct rte_eth_dev *)param;
2990 
2991 	eth_igbvf_interrupt_get_status(dev);
2992 	eth_igbvf_interrupt_action(dev, dev->intr_handle);
2993 }
2994 
2995 static int
2996 eth_igb_led_on(struct rte_eth_dev *dev)
2997 {
2998 	struct e1000_hw *hw;
2999 
3000 	hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3001 	return e1000_led_on(hw) == E1000_SUCCESS ? 0 : -ENOTSUP;
3002 }
3003 
3004 static int
3005 eth_igb_led_off(struct rte_eth_dev *dev)
3006 {
3007 	struct e1000_hw *hw;
3008 
3009 	hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3010 	return e1000_led_off(hw) == E1000_SUCCESS ? 0 : -ENOTSUP;
3011 }
3012 
3013 static int
3014 eth_igb_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
3015 {
3016 	struct e1000_hw *hw;
3017 	uint32_t ctrl;
3018 	int tx_pause;
3019 	int rx_pause;
3020 
3021 	hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3022 	fc_conf->pause_time = hw->fc.pause_time;
3023 	fc_conf->high_water = hw->fc.high_water;
3024 	fc_conf->low_water = hw->fc.low_water;
3025 	fc_conf->send_xon = hw->fc.send_xon;
3026 	fc_conf->autoneg = hw->mac.autoneg;
3027 
3028 	/*
3029 	 * Return rx_pause and tx_pause status according to actual setting of
3030 	 * the TFCE and RFCE bits in the CTRL register.
3031 	 */
3032 	ctrl = E1000_READ_REG(hw, E1000_CTRL);
3033 	if (ctrl & E1000_CTRL_TFCE)
3034 		tx_pause = 1;
3035 	else
3036 		tx_pause = 0;
3037 
3038 	if (ctrl & E1000_CTRL_RFCE)
3039 		rx_pause = 1;
3040 	else
3041 		rx_pause = 0;
3042 
3043 	if (rx_pause && tx_pause)
3044 		fc_conf->mode = RTE_FC_FULL;
3045 	else if (rx_pause)
3046 		fc_conf->mode = RTE_FC_RX_PAUSE;
3047 	else if (tx_pause)
3048 		fc_conf->mode = RTE_FC_TX_PAUSE;
3049 	else
3050 		fc_conf->mode = RTE_FC_NONE;
3051 
3052 	return 0;
3053 }
3054 
3055 static int
3056 eth_igb_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
3057 {
3058 	struct e1000_hw *hw;
3059 	int err;
3060 	enum e1000_fc_mode rte_fcmode_2_e1000_fcmode[] = {
3061 		e1000_fc_none,
3062 		e1000_fc_rx_pause,
3063 		e1000_fc_tx_pause,
3064 		e1000_fc_full
3065 	};
3066 	uint32_t rx_buf_size;
3067 	uint32_t max_high_water;
3068 	uint32_t rctl;
3069 
3070 	hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3071 	if (fc_conf->autoneg != hw->mac.autoneg)
3072 		return -ENOTSUP;
3073 	rx_buf_size = igb_get_rx_buffer_size(hw);
3074 	PMD_INIT_LOG(DEBUG, "Rx packet buffer size = 0x%x", rx_buf_size);
3075 
3076 	/* At least reserve one Ethernet frame for watermark */
3077 	max_high_water = rx_buf_size - ETHER_MAX_LEN;
3078 	if ((fc_conf->high_water > max_high_water) ||
3079 	    (fc_conf->high_water < fc_conf->low_water)) {
3080 		PMD_INIT_LOG(ERR, "e1000 incorrect high/low water value");
3081 		PMD_INIT_LOG(ERR, "high water must <=  0x%x", max_high_water);
3082 		return -EINVAL;
3083 	}
3084 
3085 	hw->fc.requested_mode = rte_fcmode_2_e1000_fcmode[fc_conf->mode];
3086 	hw->fc.pause_time     = fc_conf->pause_time;
3087 	hw->fc.high_water     = fc_conf->high_water;
3088 	hw->fc.low_water      = fc_conf->low_water;
3089 	hw->fc.send_xon	      = fc_conf->send_xon;
3090 
3091 	err = e1000_setup_link_generic(hw);
3092 	if (err == E1000_SUCCESS) {
3093 
3094 		/* check if we want to forward MAC frames - driver doesn't have native
3095 		 * capability to do that, so we'll write the registers ourselves */
3096 
3097 		rctl = E1000_READ_REG(hw, E1000_RCTL);
3098 
3099 		/* set or clear MFLCN.PMCF bit depending on configuration */
3100 		if (fc_conf->mac_ctrl_frame_fwd != 0)
3101 			rctl |= E1000_RCTL_PMCF;
3102 		else
3103 			rctl &= ~E1000_RCTL_PMCF;
3104 
3105 		E1000_WRITE_REG(hw, E1000_RCTL, rctl);
3106 		E1000_WRITE_FLUSH(hw);
3107 
3108 		return 0;
3109 	}
3110 
3111 	PMD_INIT_LOG(ERR, "e1000_setup_link_generic = 0x%x", err);
3112 	return -EIO;
3113 }
3114 
3115 #define E1000_RAH_POOLSEL_SHIFT      (18)
3116 static int
3117 eth_igb_rar_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
3118 		uint32_t index, uint32_t pool)
3119 {
3120 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3121 	uint32_t rah;
3122 
3123 	e1000_rar_set(hw, mac_addr->addr_bytes, index);
3124 	rah = E1000_READ_REG(hw, E1000_RAH(index));
3125 	rah |= (0x1 << (E1000_RAH_POOLSEL_SHIFT + pool));
3126 	E1000_WRITE_REG(hw, E1000_RAH(index), rah);
3127 	return 0;
3128 }
3129 
3130 static void
3131 eth_igb_rar_clear(struct rte_eth_dev *dev, uint32_t index)
3132 {
3133 	uint8_t addr[ETHER_ADDR_LEN];
3134 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3135 
3136 	memset(addr, 0, sizeof(addr));
3137 
3138 	e1000_rar_set(hw, addr, index);
3139 }
3140 
3141 static int
3142 eth_igb_default_mac_addr_set(struct rte_eth_dev *dev,
3143 				struct ether_addr *addr)
3144 {
3145 	eth_igb_rar_clear(dev, 0);
3146 	eth_igb_rar_set(dev, (void *)addr, 0, 0);
3147 
3148 	return 0;
3149 }
3150 /*
3151  * Virtual Function operations
3152  */
3153 static void
3154 igbvf_intr_disable(struct e1000_hw *hw)
3155 {
3156 	PMD_INIT_FUNC_TRACE();
3157 
3158 	/* Clear interrupt mask to stop from interrupts being generated */
3159 	E1000_WRITE_REG(hw, E1000_EIMC, 0xFFFF);
3160 
3161 	E1000_WRITE_FLUSH(hw);
3162 }
3163 
3164 static void
3165 igbvf_stop_adapter(struct rte_eth_dev *dev)
3166 {
3167 	u32 reg_val;
3168 	u16 i;
3169 	struct rte_eth_dev_info dev_info;
3170 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3171 
3172 	memset(&dev_info, 0, sizeof(dev_info));
3173 	eth_igbvf_infos_get(dev, &dev_info);
3174 
3175 	/* Clear interrupt mask to stop from interrupts being generated */
3176 	igbvf_intr_disable(hw);
3177 
3178 	/* Clear any pending interrupts, flush previous writes */
3179 	E1000_READ_REG(hw, E1000_EICR);
3180 
3181 	/* Disable the transmit unit.  Each queue must be disabled. */
3182 	for (i = 0; i < dev_info.max_tx_queues; i++)
3183 		E1000_WRITE_REG(hw, E1000_TXDCTL(i), E1000_TXDCTL_SWFLSH);
3184 
3185 	/* Disable the receive unit by stopping each queue */
3186 	for (i = 0; i < dev_info.max_rx_queues; i++) {
3187 		reg_val = E1000_READ_REG(hw, E1000_RXDCTL(i));
3188 		reg_val &= ~E1000_RXDCTL_QUEUE_ENABLE;
3189 		E1000_WRITE_REG(hw, E1000_RXDCTL(i), reg_val);
3190 		while (E1000_READ_REG(hw, E1000_RXDCTL(i)) & E1000_RXDCTL_QUEUE_ENABLE)
3191 			;
3192 	}
3193 
3194 	/* flush all queues disables */
3195 	E1000_WRITE_FLUSH(hw);
3196 	msec_delay(2);
3197 }
3198 
3199 static int eth_igbvf_link_update(struct e1000_hw *hw)
3200 {
3201 	struct e1000_mbx_info *mbx = &hw->mbx;
3202 	struct e1000_mac_info *mac = &hw->mac;
3203 	int ret_val = E1000_SUCCESS;
3204 
3205 	PMD_INIT_LOG(DEBUG, "e1000_check_for_link_vf");
3206 
3207 	/*
3208 	 * We only want to run this if there has been a rst asserted.
3209 	 * in this case that could mean a link change, device reset,
3210 	 * or a virtual function reset
3211 	 */
3212 
3213 	/* If we were hit with a reset or timeout drop the link */
3214 	if (!e1000_check_for_rst(hw, 0) || !mbx->timeout)
3215 		mac->get_link_status = TRUE;
3216 
3217 	if (!mac->get_link_status)
3218 		goto out;
3219 
3220 	/* if link status is down no point in checking to see if pf is up */
3221 	if (!(E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU))
3222 		goto out;
3223 
3224 	/* if we passed all the tests above then the link is up and we no
3225 	 * longer need to check for link */
3226 	mac->get_link_status = FALSE;
3227 
3228 out:
3229 	return ret_val;
3230 }
3231 
3232 
3233 static int
3234 igbvf_dev_configure(struct rte_eth_dev *dev)
3235 {
3236 	struct rte_eth_conf* conf = &dev->data->dev_conf;
3237 
3238 	PMD_INIT_LOG(DEBUG, "Configured Virtual Function port id: %d",
3239 		     dev->data->port_id);
3240 
3241 	/*
3242 	 * VF has no ability to enable/disable HW CRC
3243 	 * Keep the persistent behavior the same as Host PF
3244 	 */
3245 #ifndef RTE_LIBRTE_E1000_PF_DISABLE_STRIP_CRC
3246 	if (conf->rxmode.offloads & DEV_RX_OFFLOAD_KEEP_CRC) {
3247 		PMD_INIT_LOG(NOTICE, "VF can't disable HW CRC Strip");
3248 		conf->rxmode.offloads &= ~DEV_RX_OFFLOAD_KEEP_CRC;
3249 	}
3250 #else
3251 	if (!(conf->rxmode.offloads & DEV_RX_OFFLOAD_KEEP_CRC)) {
3252 		PMD_INIT_LOG(NOTICE, "VF can't enable HW CRC Strip");
3253 		conf->rxmode.offloads |= DEV_RX_OFFLOAD_KEEP_CRC;
3254 	}
3255 #endif
3256 
3257 	return 0;
3258 }
3259 
3260 static int
3261 igbvf_dev_start(struct rte_eth_dev *dev)
3262 {
3263 	struct e1000_hw *hw =
3264 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3265 	struct e1000_adapter *adapter =
3266 		E1000_DEV_PRIVATE(dev->data->dev_private);
3267 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
3268 	struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
3269 	int ret;
3270 	uint32_t intr_vector = 0;
3271 
3272 	PMD_INIT_FUNC_TRACE();
3273 
3274 	hw->mac.ops.reset_hw(hw);
3275 	adapter->stopped = 0;
3276 
3277 	/* Set all vfta */
3278 	igbvf_set_vfta_all(dev,1);
3279 
3280 	eth_igbvf_tx_init(dev);
3281 
3282 	/* This can fail when allocating mbufs for descriptor rings */
3283 	ret = eth_igbvf_rx_init(dev);
3284 	if (ret) {
3285 		PMD_INIT_LOG(ERR, "Unable to initialize RX hardware");
3286 		igb_dev_clear_queues(dev);
3287 		return ret;
3288 	}
3289 
3290 	/* check and configure queue intr-vector mapping */
3291 	if (rte_intr_cap_multiple(intr_handle) &&
3292 	    dev->data->dev_conf.intr_conf.rxq) {
3293 		intr_vector = dev->data->nb_rx_queues;
3294 		ret = rte_intr_efd_enable(intr_handle, intr_vector);
3295 		if (ret)
3296 			return ret;
3297 	}
3298 
3299 	if (rte_intr_dp_is_en(intr_handle) && !intr_handle->intr_vec) {
3300 		intr_handle->intr_vec =
3301 			rte_zmalloc("intr_vec",
3302 				    dev->data->nb_rx_queues * sizeof(int), 0);
3303 		if (!intr_handle->intr_vec) {
3304 			PMD_INIT_LOG(ERR, "Failed to allocate %d rx_queues"
3305 				     " intr_vec", dev->data->nb_rx_queues);
3306 			return -ENOMEM;
3307 		}
3308 	}
3309 
3310 	eth_igbvf_configure_msix_intr(dev);
3311 
3312 	/* enable uio/vfio intr/eventfd mapping */
3313 	rte_intr_enable(intr_handle);
3314 
3315 	/* resume enabled intr since hw reset */
3316 	igbvf_intr_enable(dev);
3317 
3318 	return 0;
3319 }
3320 
3321 static void
3322 igbvf_dev_stop(struct rte_eth_dev *dev)
3323 {
3324 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
3325 	struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
3326 
3327 	PMD_INIT_FUNC_TRACE();
3328 
3329 	igbvf_stop_adapter(dev);
3330 
3331 	/*
3332 	  * Clear what we set, but we still keep shadow_vfta to
3333 	  * restore after device starts
3334 	  */
3335 	igbvf_set_vfta_all(dev,0);
3336 
3337 	igb_dev_clear_queues(dev);
3338 
3339 	/* disable intr eventfd mapping */
3340 	rte_intr_disable(intr_handle);
3341 
3342 	/* Clean datapath event and queue/vec mapping */
3343 	rte_intr_efd_disable(intr_handle);
3344 	if (intr_handle->intr_vec) {
3345 		rte_free(intr_handle->intr_vec);
3346 		intr_handle->intr_vec = NULL;
3347 	}
3348 }
3349 
3350 static void
3351 igbvf_dev_close(struct rte_eth_dev *dev)
3352 {
3353 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3354 	struct e1000_adapter *adapter =
3355 		E1000_DEV_PRIVATE(dev->data->dev_private);
3356 	struct ether_addr addr;
3357 
3358 	PMD_INIT_FUNC_TRACE();
3359 
3360 	e1000_reset_hw(hw);
3361 
3362 	igbvf_dev_stop(dev);
3363 	adapter->stopped = 1;
3364 	igb_dev_free_queues(dev);
3365 
3366 	/**
3367 	 * reprogram the RAR with a zero mac address,
3368 	 * to ensure that the VF traffic goes to the PF
3369 	 * after stop, close and detach of the VF.
3370 	 **/
3371 
3372 	memset(&addr, 0, sizeof(addr));
3373 	igbvf_default_mac_addr_set(dev, &addr);
3374 }
3375 
3376 static void
3377 igbvf_promiscuous_enable(struct rte_eth_dev *dev)
3378 {
3379 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3380 
3381 	/* Set both unicast and multicast promisc */
3382 	e1000_promisc_set_vf(hw, e1000_promisc_enabled);
3383 }
3384 
3385 static void
3386 igbvf_promiscuous_disable(struct rte_eth_dev *dev)
3387 {
3388 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3389 
3390 	/* If in allmulticast mode leave multicast promisc */
3391 	if (dev->data->all_multicast == 1)
3392 		e1000_promisc_set_vf(hw, e1000_promisc_multicast);
3393 	else
3394 		e1000_promisc_set_vf(hw, e1000_promisc_disabled);
3395 }
3396 
3397 static void
3398 igbvf_allmulticast_enable(struct rte_eth_dev *dev)
3399 {
3400 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3401 
3402 	/* In promiscuous mode multicast promisc already set */
3403 	if (dev->data->promiscuous == 0)
3404 		e1000_promisc_set_vf(hw, e1000_promisc_multicast);
3405 }
3406 
3407 static void
3408 igbvf_allmulticast_disable(struct rte_eth_dev *dev)
3409 {
3410 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3411 
3412 	/* In promiscuous mode leave multicast promisc enabled */
3413 	if (dev->data->promiscuous == 0)
3414 		e1000_promisc_set_vf(hw, e1000_promisc_disabled);
3415 }
3416 
3417 static int igbvf_set_vfta(struct e1000_hw *hw, uint16_t vid, bool on)
3418 {
3419 	struct e1000_mbx_info *mbx = &hw->mbx;
3420 	uint32_t msgbuf[2];
3421 	s32 err;
3422 
3423 	/* After set vlan, vlan strip will also be enabled in igb driver*/
3424 	msgbuf[0] = E1000_VF_SET_VLAN;
3425 	msgbuf[1] = vid;
3426 	/* Setting the 8 bit field MSG INFO to TRUE indicates "add" */
3427 	if (on)
3428 		msgbuf[0] |= E1000_VF_SET_VLAN_ADD;
3429 
3430 	err = mbx->ops.write_posted(hw, msgbuf, 2, 0);
3431 	if (err)
3432 		goto mbx_err;
3433 
3434 	err = mbx->ops.read_posted(hw, msgbuf, 2, 0);
3435 	if (err)
3436 		goto mbx_err;
3437 
3438 	msgbuf[0] &= ~E1000_VT_MSGTYPE_CTS;
3439 	if (msgbuf[0] == (E1000_VF_SET_VLAN | E1000_VT_MSGTYPE_NACK))
3440 		err = -EINVAL;
3441 
3442 mbx_err:
3443 	return err;
3444 }
3445 
3446 static void igbvf_set_vfta_all(struct rte_eth_dev *dev, bool on)
3447 {
3448 	struct e1000_hw *hw =
3449 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3450 	struct e1000_vfta * shadow_vfta =
3451 		E1000_DEV_PRIVATE_TO_VFTA(dev->data->dev_private);
3452 	int i = 0, j = 0, vfta = 0, mask = 1;
3453 
3454 	for (i = 0; i < IGB_VFTA_SIZE; i++){
3455 		vfta = shadow_vfta->vfta[i];
3456 		if(vfta){
3457 			mask = 1;
3458 			for (j = 0; j < 32; j++){
3459 				if(vfta & mask)
3460 					igbvf_set_vfta(hw,
3461 						(uint16_t)((i<<5)+j), on);
3462 				mask<<=1;
3463 			}
3464 		}
3465 	}
3466 
3467 }
3468 
3469 static int
3470 igbvf_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
3471 {
3472 	struct e1000_hw *hw =
3473 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3474 	struct e1000_vfta * shadow_vfta =
3475 		E1000_DEV_PRIVATE_TO_VFTA(dev->data->dev_private);
3476 	uint32_t vid_idx = 0;
3477 	uint32_t vid_bit = 0;
3478 	int ret = 0;
3479 
3480 	PMD_INIT_FUNC_TRACE();
3481 
3482 	/*vind is not used in VF driver, set to 0, check ixgbe_set_vfta_vf*/
3483 	ret = igbvf_set_vfta(hw, vlan_id, !!on);
3484 	if(ret){
3485 		PMD_INIT_LOG(ERR, "Unable to set VF vlan");
3486 		return ret;
3487 	}
3488 	vid_idx = (uint32_t) ((vlan_id >> 5) & 0x7F);
3489 	vid_bit = (uint32_t) (1 << (vlan_id & 0x1F));
3490 
3491 	/*Save what we set and retore it after device reset*/
3492 	if (on)
3493 		shadow_vfta->vfta[vid_idx] |= vid_bit;
3494 	else
3495 		shadow_vfta->vfta[vid_idx] &= ~vid_bit;
3496 
3497 	return 0;
3498 }
3499 
3500 static int
3501 igbvf_default_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *addr)
3502 {
3503 	struct e1000_hw *hw =
3504 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3505 
3506 	/* index is not used by rar_set() */
3507 	hw->mac.ops.rar_set(hw, (void *)addr, 0);
3508 	return 0;
3509 }
3510 
3511 
3512 static int
3513 eth_igb_rss_reta_update(struct rte_eth_dev *dev,
3514 			struct rte_eth_rss_reta_entry64 *reta_conf,
3515 			uint16_t reta_size)
3516 {
3517 	uint8_t i, j, mask;
3518 	uint32_t reta, r;
3519 	uint16_t idx, shift;
3520 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3521 
3522 	if (reta_size != ETH_RSS_RETA_SIZE_128) {
3523 		PMD_DRV_LOG(ERR, "The size of hash lookup table configured "
3524 			"(%d) doesn't match the number hardware can supported "
3525 			"(%d)", reta_size, ETH_RSS_RETA_SIZE_128);
3526 		return -EINVAL;
3527 	}
3528 
3529 	for (i = 0; i < reta_size; i += IGB_4_BIT_WIDTH) {
3530 		idx = i / RTE_RETA_GROUP_SIZE;
3531 		shift = i % RTE_RETA_GROUP_SIZE;
3532 		mask = (uint8_t)((reta_conf[idx].mask >> shift) &
3533 						IGB_4_BIT_MASK);
3534 		if (!mask)
3535 			continue;
3536 		if (mask == IGB_4_BIT_MASK)
3537 			r = 0;
3538 		else
3539 			r = E1000_READ_REG(hw, E1000_RETA(i >> 2));
3540 		for (j = 0, reta = 0; j < IGB_4_BIT_WIDTH; j++) {
3541 			if (mask & (0x1 << j))
3542 				reta |= reta_conf[idx].reta[shift + j] <<
3543 							(CHAR_BIT * j);
3544 			else
3545 				reta |= r & (IGB_8_BIT_MASK << (CHAR_BIT * j));
3546 		}
3547 		E1000_WRITE_REG(hw, E1000_RETA(i >> 2), reta);
3548 	}
3549 
3550 	return 0;
3551 }
3552 
3553 static int
3554 eth_igb_rss_reta_query(struct rte_eth_dev *dev,
3555 		       struct rte_eth_rss_reta_entry64 *reta_conf,
3556 		       uint16_t reta_size)
3557 {
3558 	uint8_t i, j, mask;
3559 	uint32_t reta;
3560 	uint16_t idx, shift;
3561 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3562 
3563 	if (reta_size != ETH_RSS_RETA_SIZE_128) {
3564 		PMD_DRV_LOG(ERR, "The size of hash lookup table configured "
3565 			"(%d) doesn't match the number hardware can supported "
3566 			"(%d)", reta_size, ETH_RSS_RETA_SIZE_128);
3567 		return -EINVAL;
3568 	}
3569 
3570 	for (i = 0; i < reta_size; i += IGB_4_BIT_WIDTH) {
3571 		idx = i / RTE_RETA_GROUP_SIZE;
3572 		shift = i % RTE_RETA_GROUP_SIZE;
3573 		mask = (uint8_t)((reta_conf[idx].mask >> shift) &
3574 						IGB_4_BIT_MASK);
3575 		if (!mask)
3576 			continue;
3577 		reta = E1000_READ_REG(hw, E1000_RETA(i >> 2));
3578 		for (j = 0; j < IGB_4_BIT_WIDTH; j++) {
3579 			if (mask & (0x1 << j))
3580 				reta_conf[idx].reta[shift + j] =
3581 					((reta >> (CHAR_BIT * j)) &
3582 						IGB_8_BIT_MASK);
3583 		}
3584 	}
3585 
3586 	return 0;
3587 }
3588 
3589 int
3590 eth_igb_syn_filter_set(struct rte_eth_dev *dev,
3591 			struct rte_eth_syn_filter *filter,
3592 			bool add)
3593 {
3594 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3595 	struct e1000_filter_info *filter_info =
3596 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
3597 	uint32_t synqf, rfctl;
3598 
3599 	if (filter->queue >= IGB_MAX_RX_QUEUE_NUM)
3600 		return -EINVAL;
3601 
3602 	synqf = E1000_READ_REG(hw, E1000_SYNQF(0));
3603 
3604 	if (add) {
3605 		if (synqf & E1000_SYN_FILTER_ENABLE)
3606 			return -EINVAL;
3607 
3608 		synqf = (uint32_t)(((filter->queue << E1000_SYN_FILTER_QUEUE_SHIFT) &
3609 			E1000_SYN_FILTER_QUEUE) | E1000_SYN_FILTER_ENABLE);
3610 
3611 		rfctl = E1000_READ_REG(hw, E1000_RFCTL);
3612 		if (filter->hig_pri)
3613 			rfctl |= E1000_RFCTL_SYNQFP;
3614 		else
3615 			rfctl &= ~E1000_RFCTL_SYNQFP;
3616 
3617 		E1000_WRITE_REG(hw, E1000_RFCTL, rfctl);
3618 	} else {
3619 		if (!(synqf & E1000_SYN_FILTER_ENABLE))
3620 			return -ENOENT;
3621 		synqf = 0;
3622 	}
3623 
3624 	filter_info->syn_info = synqf;
3625 	E1000_WRITE_REG(hw, E1000_SYNQF(0), synqf);
3626 	E1000_WRITE_FLUSH(hw);
3627 	return 0;
3628 }
3629 
3630 static int
3631 eth_igb_syn_filter_get(struct rte_eth_dev *dev,
3632 			struct rte_eth_syn_filter *filter)
3633 {
3634 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3635 	uint32_t synqf, rfctl;
3636 
3637 	synqf = E1000_READ_REG(hw, E1000_SYNQF(0));
3638 	if (synqf & E1000_SYN_FILTER_ENABLE) {
3639 		rfctl = E1000_READ_REG(hw, E1000_RFCTL);
3640 		filter->hig_pri = (rfctl & E1000_RFCTL_SYNQFP) ? 1 : 0;
3641 		filter->queue = (uint8_t)((synqf & E1000_SYN_FILTER_QUEUE) >>
3642 				E1000_SYN_FILTER_QUEUE_SHIFT);
3643 		return 0;
3644 	}
3645 
3646 	return -ENOENT;
3647 }
3648 
3649 static int
3650 eth_igb_syn_filter_handle(struct rte_eth_dev *dev,
3651 			enum rte_filter_op filter_op,
3652 			void *arg)
3653 {
3654 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3655 	int ret;
3656 
3657 	MAC_TYPE_FILTER_SUP(hw->mac.type);
3658 
3659 	if (filter_op == RTE_ETH_FILTER_NOP)
3660 		return 0;
3661 
3662 	if (arg == NULL) {
3663 		PMD_DRV_LOG(ERR, "arg shouldn't be NULL for operation %u",
3664 			    filter_op);
3665 		return -EINVAL;
3666 	}
3667 
3668 	switch (filter_op) {
3669 	case RTE_ETH_FILTER_ADD:
3670 		ret = eth_igb_syn_filter_set(dev,
3671 				(struct rte_eth_syn_filter *)arg,
3672 				TRUE);
3673 		break;
3674 	case RTE_ETH_FILTER_DELETE:
3675 		ret = eth_igb_syn_filter_set(dev,
3676 				(struct rte_eth_syn_filter *)arg,
3677 				FALSE);
3678 		break;
3679 	case RTE_ETH_FILTER_GET:
3680 		ret = eth_igb_syn_filter_get(dev,
3681 				(struct rte_eth_syn_filter *)arg);
3682 		break;
3683 	default:
3684 		PMD_DRV_LOG(ERR, "unsupported operation %u", filter_op);
3685 		ret = -EINVAL;
3686 		break;
3687 	}
3688 
3689 	return ret;
3690 }
3691 
3692 /* translate elements in struct rte_eth_ntuple_filter to struct e1000_2tuple_filter_info*/
3693 static inline int
3694 ntuple_filter_to_2tuple(struct rte_eth_ntuple_filter *filter,
3695 			struct e1000_2tuple_filter_info *filter_info)
3696 {
3697 	if (filter->queue >= IGB_MAX_RX_QUEUE_NUM)
3698 		return -EINVAL;
3699 	if (filter->priority > E1000_2TUPLE_MAX_PRI)
3700 		return -EINVAL;  /* filter index is out of range. */
3701 	if (filter->tcp_flags > TCP_FLAG_ALL)
3702 		return -EINVAL;  /* flags is invalid. */
3703 
3704 	switch (filter->dst_port_mask) {
3705 	case UINT16_MAX:
3706 		filter_info->dst_port_mask = 0;
3707 		filter_info->dst_port = filter->dst_port;
3708 		break;
3709 	case 0:
3710 		filter_info->dst_port_mask = 1;
3711 		break;
3712 	default:
3713 		PMD_DRV_LOG(ERR, "invalid dst_port mask.");
3714 		return -EINVAL;
3715 	}
3716 
3717 	switch (filter->proto_mask) {
3718 	case UINT8_MAX:
3719 		filter_info->proto_mask = 0;
3720 		filter_info->proto = filter->proto;
3721 		break;
3722 	case 0:
3723 		filter_info->proto_mask = 1;
3724 		break;
3725 	default:
3726 		PMD_DRV_LOG(ERR, "invalid protocol mask.");
3727 		return -EINVAL;
3728 	}
3729 
3730 	filter_info->priority = (uint8_t)filter->priority;
3731 	if (filter->flags & RTE_NTUPLE_FLAGS_TCP_FLAG)
3732 		filter_info->tcp_flags = filter->tcp_flags;
3733 	else
3734 		filter_info->tcp_flags = 0;
3735 
3736 	return 0;
3737 }
3738 
3739 static inline struct e1000_2tuple_filter *
3740 igb_2tuple_filter_lookup(struct e1000_2tuple_filter_list *filter_list,
3741 			struct e1000_2tuple_filter_info *key)
3742 {
3743 	struct e1000_2tuple_filter *it;
3744 
3745 	TAILQ_FOREACH(it, filter_list, entries) {
3746 		if (memcmp(key, &it->filter_info,
3747 			sizeof(struct e1000_2tuple_filter_info)) == 0) {
3748 			return it;
3749 		}
3750 	}
3751 	return NULL;
3752 }
3753 
3754 /* inject a igb 2tuple filter to HW */
3755 static inline void
3756 igb_inject_2uple_filter(struct rte_eth_dev *dev,
3757 			   struct e1000_2tuple_filter *filter)
3758 {
3759 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3760 	uint32_t ttqf = E1000_TTQF_DISABLE_MASK;
3761 	uint32_t imir, imir_ext = E1000_IMIREXT_SIZE_BP;
3762 	int i;
3763 
3764 	i = filter->index;
3765 	imir = (uint32_t)(filter->filter_info.dst_port & E1000_IMIR_DSTPORT);
3766 	if (filter->filter_info.dst_port_mask == 1) /* 1b means not compare. */
3767 		imir |= E1000_IMIR_PORT_BP;
3768 	else
3769 		imir &= ~E1000_IMIR_PORT_BP;
3770 
3771 	imir |= filter->filter_info.priority << E1000_IMIR_PRIORITY_SHIFT;
3772 
3773 	ttqf |= E1000_TTQF_QUEUE_ENABLE;
3774 	ttqf |= (uint32_t)(filter->queue << E1000_TTQF_QUEUE_SHIFT);
3775 	ttqf |= (uint32_t)(filter->filter_info.proto &
3776 						E1000_TTQF_PROTOCOL_MASK);
3777 	if (filter->filter_info.proto_mask == 0)
3778 		ttqf &= ~E1000_TTQF_MASK_ENABLE;
3779 
3780 	/* tcp flags bits setting. */
3781 	if (filter->filter_info.tcp_flags & TCP_FLAG_ALL) {
3782 		if (filter->filter_info.tcp_flags & TCP_URG_FLAG)
3783 			imir_ext |= E1000_IMIREXT_CTRL_URG;
3784 		if (filter->filter_info.tcp_flags & TCP_ACK_FLAG)
3785 			imir_ext |= E1000_IMIREXT_CTRL_ACK;
3786 		if (filter->filter_info.tcp_flags & TCP_PSH_FLAG)
3787 			imir_ext |= E1000_IMIREXT_CTRL_PSH;
3788 		if (filter->filter_info.tcp_flags & TCP_RST_FLAG)
3789 			imir_ext |= E1000_IMIREXT_CTRL_RST;
3790 		if (filter->filter_info.tcp_flags & TCP_SYN_FLAG)
3791 			imir_ext |= E1000_IMIREXT_CTRL_SYN;
3792 		if (filter->filter_info.tcp_flags & TCP_FIN_FLAG)
3793 			imir_ext |= E1000_IMIREXT_CTRL_FIN;
3794 	} else {
3795 		imir_ext |= E1000_IMIREXT_CTRL_BP;
3796 	}
3797 	E1000_WRITE_REG(hw, E1000_IMIR(i), imir);
3798 	E1000_WRITE_REG(hw, E1000_TTQF(i), ttqf);
3799 	E1000_WRITE_REG(hw, E1000_IMIREXT(i), imir_ext);
3800 }
3801 
3802 /*
3803  * igb_add_2tuple_filter - add a 2tuple filter
3804  *
3805  * @param
3806  * dev: Pointer to struct rte_eth_dev.
3807  * ntuple_filter: ponter to the filter that will be added.
3808  *
3809  * @return
3810  *    - On success, zero.
3811  *    - On failure, a negative value.
3812  */
3813 static int
3814 igb_add_2tuple_filter(struct rte_eth_dev *dev,
3815 			struct rte_eth_ntuple_filter *ntuple_filter)
3816 {
3817 	struct e1000_filter_info *filter_info =
3818 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
3819 	struct e1000_2tuple_filter *filter;
3820 	int i, ret;
3821 
3822 	filter = rte_zmalloc("e1000_2tuple_filter",
3823 			sizeof(struct e1000_2tuple_filter), 0);
3824 	if (filter == NULL)
3825 		return -ENOMEM;
3826 
3827 	ret = ntuple_filter_to_2tuple(ntuple_filter,
3828 				      &filter->filter_info);
3829 	if (ret < 0) {
3830 		rte_free(filter);
3831 		return ret;
3832 	}
3833 	if (igb_2tuple_filter_lookup(&filter_info->twotuple_list,
3834 					 &filter->filter_info) != NULL) {
3835 		PMD_DRV_LOG(ERR, "filter exists.");
3836 		rte_free(filter);
3837 		return -EEXIST;
3838 	}
3839 	filter->queue = ntuple_filter->queue;
3840 
3841 	/*
3842 	 * look for an unused 2tuple filter index,
3843 	 * and insert the filter to list.
3844 	 */
3845 	for (i = 0; i < E1000_MAX_TTQF_FILTERS; i++) {
3846 		if (!(filter_info->twotuple_mask & (1 << i))) {
3847 			filter_info->twotuple_mask |= 1 << i;
3848 			filter->index = i;
3849 			TAILQ_INSERT_TAIL(&filter_info->twotuple_list,
3850 					  filter,
3851 					  entries);
3852 			break;
3853 		}
3854 	}
3855 	if (i >= E1000_MAX_TTQF_FILTERS) {
3856 		PMD_DRV_LOG(ERR, "2tuple filters are full.");
3857 		rte_free(filter);
3858 		return -ENOSYS;
3859 	}
3860 
3861 	igb_inject_2uple_filter(dev, filter);
3862 	return 0;
3863 }
3864 
3865 int
3866 igb_delete_2tuple_filter(struct rte_eth_dev *dev,
3867 			struct e1000_2tuple_filter *filter)
3868 {
3869 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3870 	struct e1000_filter_info *filter_info =
3871 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
3872 
3873 	filter_info->twotuple_mask &= ~(1 << filter->index);
3874 	TAILQ_REMOVE(&filter_info->twotuple_list, filter, entries);
3875 	rte_free(filter);
3876 
3877 	E1000_WRITE_REG(hw, E1000_TTQF(filter->index), E1000_TTQF_DISABLE_MASK);
3878 	E1000_WRITE_REG(hw, E1000_IMIR(filter->index), 0);
3879 	E1000_WRITE_REG(hw, E1000_IMIREXT(filter->index), 0);
3880 	return 0;
3881 }
3882 
3883 /*
3884  * igb_remove_2tuple_filter - remove a 2tuple filter
3885  *
3886  * @param
3887  * dev: Pointer to struct rte_eth_dev.
3888  * ntuple_filter: ponter to the filter that will be removed.
3889  *
3890  * @return
3891  *    - On success, zero.
3892  *    - On failure, a negative value.
3893  */
3894 static int
3895 igb_remove_2tuple_filter(struct rte_eth_dev *dev,
3896 			struct rte_eth_ntuple_filter *ntuple_filter)
3897 {
3898 	struct e1000_filter_info *filter_info =
3899 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
3900 	struct e1000_2tuple_filter_info filter_2tuple;
3901 	struct e1000_2tuple_filter *filter;
3902 	int ret;
3903 
3904 	memset(&filter_2tuple, 0, sizeof(struct e1000_2tuple_filter_info));
3905 	ret = ntuple_filter_to_2tuple(ntuple_filter,
3906 				      &filter_2tuple);
3907 	if (ret < 0)
3908 		return ret;
3909 
3910 	filter = igb_2tuple_filter_lookup(&filter_info->twotuple_list,
3911 					 &filter_2tuple);
3912 	if (filter == NULL) {
3913 		PMD_DRV_LOG(ERR, "filter doesn't exist.");
3914 		return -ENOENT;
3915 	}
3916 
3917 	igb_delete_2tuple_filter(dev, filter);
3918 
3919 	return 0;
3920 }
3921 
3922 /* inject a igb flex filter to HW */
3923 static inline void
3924 igb_inject_flex_filter(struct rte_eth_dev *dev,
3925 			   struct e1000_flex_filter *filter)
3926 {
3927 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3928 	uint32_t wufc, queueing;
3929 	uint32_t reg_off;
3930 	uint8_t i, j = 0;
3931 
3932 	wufc = E1000_READ_REG(hw, E1000_WUFC);
3933 	if (filter->index < E1000_MAX_FHFT)
3934 		reg_off = E1000_FHFT(filter->index);
3935 	else
3936 		reg_off = E1000_FHFT_EXT(filter->index - E1000_MAX_FHFT);
3937 
3938 	E1000_WRITE_REG(hw, E1000_WUFC, wufc | E1000_WUFC_FLEX_HQ |
3939 			(E1000_WUFC_FLX0 << filter->index));
3940 	queueing = filter->filter_info.len |
3941 		(filter->queue << E1000_FHFT_QUEUEING_QUEUE_SHIFT) |
3942 		(filter->filter_info.priority <<
3943 			E1000_FHFT_QUEUEING_PRIO_SHIFT);
3944 	E1000_WRITE_REG(hw, reg_off + E1000_FHFT_QUEUEING_OFFSET,
3945 			queueing);
3946 
3947 	for (i = 0; i < E1000_FLEX_FILTERS_MASK_SIZE; i++) {
3948 		E1000_WRITE_REG(hw, reg_off,
3949 				filter->filter_info.dwords[j]);
3950 		reg_off += sizeof(uint32_t);
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 			(uint32_t)filter->filter_info.mask[i]);
3956 		reg_off += sizeof(uint32_t) * 2;
3957 		++j;
3958 	}
3959 }
3960 
3961 static inline struct e1000_flex_filter *
3962 eth_igb_flex_filter_lookup(struct e1000_flex_filter_list *filter_list,
3963 			struct e1000_flex_filter_info *key)
3964 {
3965 	struct e1000_flex_filter *it;
3966 
3967 	TAILQ_FOREACH(it, filter_list, entries) {
3968 		if (memcmp(key, &it->filter_info,
3969 			sizeof(struct e1000_flex_filter_info)) == 0)
3970 			return it;
3971 	}
3972 
3973 	return NULL;
3974 }
3975 
3976 /* remove a flex byte filter
3977  * @param
3978  * dev: Pointer to struct rte_eth_dev.
3979  * filter: the pointer of the filter will be removed.
3980  */
3981 void
3982 igb_remove_flex_filter(struct rte_eth_dev *dev,
3983 			struct e1000_flex_filter *filter)
3984 {
3985 	struct e1000_filter_info *filter_info =
3986 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
3987 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3988 	uint32_t wufc, i;
3989 	uint32_t reg_off;
3990 
3991 	wufc = E1000_READ_REG(hw, E1000_WUFC);
3992 	if (filter->index < E1000_MAX_FHFT)
3993 		reg_off = E1000_FHFT(filter->index);
3994 	else
3995 		reg_off = E1000_FHFT_EXT(filter->index - E1000_MAX_FHFT);
3996 
3997 	for (i = 0; i < E1000_FHFT_SIZE_IN_DWD; i++)
3998 		E1000_WRITE_REG(hw, reg_off + i * sizeof(uint32_t), 0);
3999 
4000 	E1000_WRITE_REG(hw, E1000_WUFC, wufc &
4001 		(~(E1000_WUFC_FLX0 << filter->index)));
4002 
4003 	filter_info->flex_mask &= ~(1 << filter->index);
4004 	TAILQ_REMOVE(&filter_info->flex_list, filter, entries);
4005 	rte_free(filter);
4006 }
4007 
4008 int
4009 eth_igb_add_del_flex_filter(struct rte_eth_dev *dev,
4010 			struct rte_eth_flex_filter *filter,
4011 			bool add)
4012 {
4013 	struct e1000_filter_info *filter_info =
4014 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
4015 	struct e1000_flex_filter *flex_filter, *it;
4016 	uint32_t mask;
4017 	uint8_t shift, i;
4018 
4019 	flex_filter = rte_zmalloc("e1000_flex_filter",
4020 			sizeof(struct e1000_flex_filter), 0);
4021 	if (flex_filter == NULL)
4022 		return -ENOMEM;
4023 
4024 	flex_filter->filter_info.len = filter->len;
4025 	flex_filter->filter_info.priority = filter->priority;
4026 	memcpy(flex_filter->filter_info.dwords, filter->bytes, filter->len);
4027 	for (i = 0; i < RTE_ALIGN(filter->len, CHAR_BIT) / CHAR_BIT; i++) {
4028 		mask = 0;
4029 		/* reverse bits in flex filter's mask*/
4030 		for (shift = 0; shift < CHAR_BIT; shift++) {
4031 			if (filter->mask[i] & (0x01 << shift))
4032 				mask |= (0x80 >> shift);
4033 		}
4034 		flex_filter->filter_info.mask[i] = mask;
4035 	}
4036 
4037 	it = eth_igb_flex_filter_lookup(&filter_info->flex_list,
4038 				&flex_filter->filter_info);
4039 	if (it == NULL && !add) {
4040 		PMD_DRV_LOG(ERR, "filter doesn't exist.");
4041 		rte_free(flex_filter);
4042 		return -ENOENT;
4043 	}
4044 	if (it != NULL && add) {
4045 		PMD_DRV_LOG(ERR, "filter exists.");
4046 		rte_free(flex_filter);
4047 		return -EEXIST;
4048 	}
4049 
4050 	if (add) {
4051 		flex_filter->queue = filter->queue;
4052 		/*
4053 		 * look for an unused flex filter index
4054 		 * and insert the filter into the list.
4055 		 */
4056 		for (i = 0; i < E1000_MAX_FLEX_FILTERS; i++) {
4057 			if (!(filter_info->flex_mask & (1 << i))) {
4058 				filter_info->flex_mask |= 1 << i;
4059 				flex_filter->index = i;
4060 				TAILQ_INSERT_TAIL(&filter_info->flex_list,
4061 					flex_filter,
4062 					entries);
4063 				break;
4064 			}
4065 		}
4066 		if (i >= E1000_MAX_FLEX_FILTERS) {
4067 			PMD_DRV_LOG(ERR, "flex filters are full.");
4068 			rte_free(flex_filter);
4069 			return -ENOSYS;
4070 		}
4071 
4072 		igb_inject_flex_filter(dev, flex_filter);
4073 
4074 	} else {
4075 		igb_remove_flex_filter(dev, it);
4076 		rte_free(flex_filter);
4077 	}
4078 
4079 	return 0;
4080 }
4081 
4082 static int
4083 eth_igb_get_flex_filter(struct rte_eth_dev *dev,
4084 			struct rte_eth_flex_filter *filter)
4085 {
4086 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4087 	struct e1000_filter_info *filter_info =
4088 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
4089 	struct e1000_flex_filter flex_filter, *it;
4090 	uint32_t wufc, queueing, wufc_en = 0;
4091 
4092 	memset(&flex_filter, 0, sizeof(struct e1000_flex_filter));
4093 	flex_filter.filter_info.len = filter->len;
4094 	flex_filter.filter_info.priority = filter->priority;
4095 	memcpy(flex_filter.filter_info.dwords, filter->bytes, filter->len);
4096 	memcpy(flex_filter.filter_info.mask, filter->mask,
4097 			RTE_ALIGN(filter->len, CHAR_BIT) / CHAR_BIT);
4098 
4099 	it = eth_igb_flex_filter_lookup(&filter_info->flex_list,
4100 				&flex_filter.filter_info);
4101 	if (it == NULL) {
4102 		PMD_DRV_LOG(ERR, "filter doesn't exist.");
4103 		return -ENOENT;
4104 	}
4105 
4106 	wufc = E1000_READ_REG(hw, E1000_WUFC);
4107 	wufc_en = E1000_WUFC_FLEX_HQ | (E1000_WUFC_FLX0 << it->index);
4108 
4109 	if ((wufc & wufc_en) == wufc_en) {
4110 		uint32_t reg_off = 0;
4111 		if (it->index < E1000_MAX_FHFT)
4112 			reg_off = E1000_FHFT(it->index);
4113 		else
4114 			reg_off = E1000_FHFT_EXT(it->index - E1000_MAX_FHFT);
4115 
4116 		queueing = E1000_READ_REG(hw,
4117 				reg_off + E1000_FHFT_QUEUEING_OFFSET);
4118 		filter->len = queueing & E1000_FHFT_QUEUEING_LEN;
4119 		filter->priority = (queueing & E1000_FHFT_QUEUEING_PRIO) >>
4120 			E1000_FHFT_QUEUEING_PRIO_SHIFT;
4121 		filter->queue = (queueing & E1000_FHFT_QUEUEING_QUEUE) >>
4122 			E1000_FHFT_QUEUEING_QUEUE_SHIFT;
4123 		return 0;
4124 	}
4125 	return -ENOENT;
4126 }
4127 
4128 static int
4129 eth_igb_flex_filter_handle(struct rte_eth_dev *dev,
4130 			enum rte_filter_op filter_op,
4131 			void *arg)
4132 {
4133 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4134 	struct rte_eth_flex_filter *filter;
4135 	int ret = 0;
4136 
4137 	MAC_TYPE_FILTER_SUP_EXT(hw->mac.type);
4138 
4139 	if (filter_op == RTE_ETH_FILTER_NOP)
4140 		return ret;
4141 
4142 	if (arg == NULL) {
4143 		PMD_DRV_LOG(ERR, "arg shouldn't be NULL for operation %u",
4144 			    filter_op);
4145 		return -EINVAL;
4146 	}
4147 
4148 	filter = (struct rte_eth_flex_filter *)arg;
4149 	if (filter->len == 0 || filter->len > E1000_MAX_FLEX_FILTER_LEN
4150 	    || filter->len % sizeof(uint64_t) != 0) {
4151 		PMD_DRV_LOG(ERR, "filter's length is out of range");
4152 		return -EINVAL;
4153 	}
4154 	if (filter->priority > E1000_MAX_FLEX_FILTER_PRI) {
4155 		PMD_DRV_LOG(ERR, "filter's priority is out of range");
4156 		return -EINVAL;
4157 	}
4158 
4159 	switch (filter_op) {
4160 	case RTE_ETH_FILTER_ADD:
4161 		ret = eth_igb_add_del_flex_filter(dev, filter, TRUE);
4162 		break;
4163 	case RTE_ETH_FILTER_DELETE:
4164 		ret = eth_igb_add_del_flex_filter(dev, filter, FALSE);
4165 		break;
4166 	case RTE_ETH_FILTER_GET:
4167 		ret = eth_igb_get_flex_filter(dev, filter);
4168 		break;
4169 	default:
4170 		PMD_DRV_LOG(ERR, "unsupported operation %u", filter_op);
4171 		ret = -EINVAL;
4172 		break;
4173 	}
4174 
4175 	return ret;
4176 }
4177 
4178 /* translate elements in struct rte_eth_ntuple_filter to struct e1000_5tuple_filter_info*/
4179 static inline int
4180 ntuple_filter_to_5tuple_82576(struct rte_eth_ntuple_filter *filter,
4181 			struct e1000_5tuple_filter_info *filter_info)
4182 {
4183 	if (filter->queue >= IGB_MAX_RX_QUEUE_NUM_82576)
4184 		return -EINVAL;
4185 	if (filter->priority > E1000_2TUPLE_MAX_PRI)
4186 		return -EINVAL;  /* filter index is out of range. */
4187 	if (filter->tcp_flags > TCP_FLAG_ALL)
4188 		return -EINVAL;  /* flags is invalid. */
4189 
4190 	switch (filter->dst_ip_mask) {
4191 	case UINT32_MAX:
4192 		filter_info->dst_ip_mask = 0;
4193 		filter_info->dst_ip = filter->dst_ip;
4194 		break;
4195 	case 0:
4196 		filter_info->dst_ip_mask = 1;
4197 		break;
4198 	default:
4199 		PMD_DRV_LOG(ERR, "invalid dst_ip mask.");
4200 		return -EINVAL;
4201 	}
4202 
4203 	switch (filter->src_ip_mask) {
4204 	case UINT32_MAX:
4205 		filter_info->src_ip_mask = 0;
4206 		filter_info->src_ip = filter->src_ip;
4207 		break;
4208 	case 0:
4209 		filter_info->src_ip_mask = 1;
4210 		break;
4211 	default:
4212 		PMD_DRV_LOG(ERR, "invalid src_ip mask.");
4213 		return -EINVAL;
4214 	}
4215 
4216 	switch (filter->dst_port_mask) {
4217 	case UINT16_MAX:
4218 		filter_info->dst_port_mask = 0;
4219 		filter_info->dst_port = filter->dst_port;
4220 		break;
4221 	case 0:
4222 		filter_info->dst_port_mask = 1;
4223 		break;
4224 	default:
4225 		PMD_DRV_LOG(ERR, "invalid dst_port mask.");
4226 		return -EINVAL;
4227 	}
4228 
4229 	switch (filter->src_port_mask) {
4230 	case UINT16_MAX:
4231 		filter_info->src_port_mask = 0;
4232 		filter_info->src_port = filter->src_port;
4233 		break;
4234 	case 0:
4235 		filter_info->src_port_mask = 1;
4236 		break;
4237 	default:
4238 		PMD_DRV_LOG(ERR, "invalid src_port mask.");
4239 		return -EINVAL;
4240 	}
4241 
4242 	switch (filter->proto_mask) {
4243 	case UINT8_MAX:
4244 		filter_info->proto_mask = 0;
4245 		filter_info->proto = filter->proto;
4246 		break;
4247 	case 0:
4248 		filter_info->proto_mask = 1;
4249 		break;
4250 	default:
4251 		PMD_DRV_LOG(ERR, "invalid protocol mask.");
4252 		return -EINVAL;
4253 	}
4254 
4255 	filter_info->priority = (uint8_t)filter->priority;
4256 	if (filter->flags & RTE_NTUPLE_FLAGS_TCP_FLAG)
4257 		filter_info->tcp_flags = filter->tcp_flags;
4258 	else
4259 		filter_info->tcp_flags = 0;
4260 
4261 	return 0;
4262 }
4263 
4264 static inline struct e1000_5tuple_filter *
4265 igb_5tuple_filter_lookup_82576(struct e1000_5tuple_filter_list *filter_list,
4266 			struct e1000_5tuple_filter_info *key)
4267 {
4268 	struct e1000_5tuple_filter *it;
4269 
4270 	TAILQ_FOREACH(it, filter_list, entries) {
4271 		if (memcmp(key, &it->filter_info,
4272 			sizeof(struct e1000_5tuple_filter_info)) == 0) {
4273 			return it;
4274 		}
4275 	}
4276 	return NULL;
4277 }
4278 
4279 /* inject a igb 5-tuple filter to HW */
4280 static inline void
4281 igb_inject_5tuple_filter_82576(struct rte_eth_dev *dev,
4282 			   struct e1000_5tuple_filter *filter)
4283 {
4284 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4285 	uint32_t ftqf = E1000_FTQF_VF_BP | E1000_FTQF_MASK;
4286 	uint32_t spqf, imir, imir_ext = E1000_IMIREXT_SIZE_BP;
4287 	uint8_t i;
4288 
4289 	i = filter->index;
4290 	ftqf |= filter->filter_info.proto & E1000_FTQF_PROTOCOL_MASK;
4291 	if (filter->filter_info.src_ip_mask == 0) /* 0b means compare. */
4292 		ftqf &= ~E1000_FTQF_MASK_SOURCE_ADDR_BP;
4293 	if (filter->filter_info.dst_ip_mask == 0)
4294 		ftqf &= ~E1000_FTQF_MASK_DEST_ADDR_BP;
4295 	if (filter->filter_info.src_port_mask == 0)
4296 		ftqf &= ~E1000_FTQF_MASK_SOURCE_PORT_BP;
4297 	if (filter->filter_info.proto_mask == 0)
4298 		ftqf &= ~E1000_FTQF_MASK_PROTO_BP;
4299 	ftqf |= (filter->queue << E1000_FTQF_QUEUE_SHIFT) &
4300 		E1000_FTQF_QUEUE_MASK;
4301 	ftqf |= E1000_FTQF_QUEUE_ENABLE;
4302 	E1000_WRITE_REG(hw, E1000_FTQF(i), ftqf);
4303 	E1000_WRITE_REG(hw, E1000_DAQF(i), filter->filter_info.dst_ip);
4304 	E1000_WRITE_REG(hw, E1000_SAQF(i), filter->filter_info.src_ip);
4305 
4306 	spqf = filter->filter_info.src_port & E1000_SPQF_SRCPORT;
4307 	E1000_WRITE_REG(hw, E1000_SPQF(i), spqf);
4308 
4309 	imir = (uint32_t)(filter->filter_info.dst_port & E1000_IMIR_DSTPORT);
4310 	if (filter->filter_info.dst_port_mask == 1) /* 1b means not compare. */
4311 		imir |= E1000_IMIR_PORT_BP;
4312 	else
4313 		imir &= ~E1000_IMIR_PORT_BP;
4314 	imir |= filter->filter_info.priority << E1000_IMIR_PRIORITY_SHIFT;
4315 
4316 	/* tcp flags bits setting. */
4317 	if (filter->filter_info.tcp_flags & TCP_FLAG_ALL) {
4318 		if (filter->filter_info.tcp_flags & TCP_URG_FLAG)
4319 			imir_ext |= E1000_IMIREXT_CTRL_URG;
4320 		if (filter->filter_info.tcp_flags & TCP_ACK_FLAG)
4321 			imir_ext |= E1000_IMIREXT_CTRL_ACK;
4322 		if (filter->filter_info.tcp_flags & TCP_PSH_FLAG)
4323 			imir_ext |= E1000_IMIREXT_CTRL_PSH;
4324 		if (filter->filter_info.tcp_flags & TCP_RST_FLAG)
4325 			imir_ext |= E1000_IMIREXT_CTRL_RST;
4326 		if (filter->filter_info.tcp_flags & TCP_SYN_FLAG)
4327 			imir_ext |= E1000_IMIREXT_CTRL_SYN;
4328 		if (filter->filter_info.tcp_flags & TCP_FIN_FLAG)
4329 			imir_ext |= E1000_IMIREXT_CTRL_FIN;
4330 	} else {
4331 		imir_ext |= E1000_IMIREXT_CTRL_BP;
4332 	}
4333 	E1000_WRITE_REG(hw, E1000_IMIR(i), imir);
4334 	E1000_WRITE_REG(hw, E1000_IMIREXT(i), imir_ext);
4335 }
4336 
4337 /*
4338  * igb_add_5tuple_filter_82576 - add a 5tuple filter
4339  *
4340  * @param
4341  * dev: Pointer to struct rte_eth_dev.
4342  * ntuple_filter: ponter to the filter that will be added.
4343  *
4344  * @return
4345  *    - On success, zero.
4346  *    - On failure, a negative value.
4347  */
4348 static int
4349 igb_add_5tuple_filter_82576(struct rte_eth_dev *dev,
4350 			struct rte_eth_ntuple_filter *ntuple_filter)
4351 {
4352 	struct e1000_filter_info *filter_info =
4353 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
4354 	struct e1000_5tuple_filter *filter;
4355 	uint8_t i;
4356 	int ret;
4357 
4358 	filter = rte_zmalloc("e1000_5tuple_filter",
4359 			sizeof(struct e1000_5tuple_filter), 0);
4360 	if (filter == NULL)
4361 		return -ENOMEM;
4362 
4363 	ret = ntuple_filter_to_5tuple_82576(ntuple_filter,
4364 					    &filter->filter_info);
4365 	if (ret < 0) {
4366 		rte_free(filter);
4367 		return ret;
4368 	}
4369 
4370 	if (igb_5tuple_filter_lookup_82576(&filter_info->fivetuple_list,
4371 					 &filter->filter_info) != NULL) {
4372 		PMD_DRV_LOG(ERR, "filter exists.");
4373 		rte_free(filter);
4374 		return -EEXIST;
4375 	}
4376 	filter->queue = ntuple_filter->queue;
4377 
4378 	/*
4379 	 * look for an unused 5tuple filter index,
4380 	 * and insert the filter to list.
4381 	 */
4382 	for (i = 0; i < E1000_MAX_FTQF_FILTERS; i++) {
4383 		if (!(filter_info->fivetuple_mask & (1 << i))) {
4384 			filter_info->fivetuple_mask |= 1 << i;
4385 			filter->index = i;
4386 			TAILQ_INSERT_TAIL(&filter_info->fivetuple_list,
4387 					  filter,
4388 					  entries);
4389 			break;
4390 		}
4391 	}
4392 	if (i >= E1000_MAX_FTQF_FILTERS) {
4393 		PMD_DRV_LOG(ERR, "5tuple filters are full.");
4394 		rte_free(filter);
4395 		return -ENOSYS;
4396 	}
4397 
4398 	igb_inject_5tuple_filter_82576(dev, filter);
4399 	return 0;
4400 }
4401 
4402 int
4403 igb_delete_5tuple_filter_82576(struct rte_eth_dev *dev,
4404 				struct e1000_5tuple_filter *filter)
4405 {
4406 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4407 	struct e1000_filter_info *filter_info =
4408 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
4409 
4410 	filter_info->fivetuple_mask &= ~(1 << filter->index);
4411 	TAILQ_REMOVE(&filter_info->fivetuple_list, filter, entries);
4412 	rte_free(filter);
4413 
4414 	E1000_WRITE_REG(hw, E1000_FTQF(filter->index),
4415 			E1000_FTQF_VF_BP | E1000_FTQF_MASK);
4416 	E1000_WRITE_REG(hw, E1000_DAQF(filter->index), 0);
4417 	E1000_WRITE_REG(hw, E1000_SAQF(filter->index), 0);
4418 	E1000_WRITE_REG(hw, E1000_SPQF(filter->index), 0);
4419 	E1000_WRITE_REG(hw, E1000_IMIR(filter->index), 0);
4420 	E1000_WRITE_REG(hw, E1000_IMIREXT(filter->index), 0);
4421 	return 0;
4422 }
4423 
4424 /*
4425  * igb_remove_5tuple_filter_82576 - remove a 5tuple filter
4426  *
4427  * @param
4428  * dev: Pointer to struct rte_eth_dev.
4429  * ntuple_filter: ponter to the filter that will be removed.
4430  *
4431  * @return
4432  *    - On success, zero.
4433  *    - On failure, a negative value.
4434  */
4435 static int
4436 igb_remove_5tuple_filter_82576(struct rte_eth_dev *dev,
4437 				struct rte_eth_ntuple_filter *ntuple_filter)
4438 {
4439 	struct e1000_filter_info *filter_info =
4440 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
4441 	struct e1000_5tuple_filter_info filter_5tuple;
4442 	struct e1000_5tuple_filter *filter;
4443 	int ret;
4444 
4445 	memset(&filter_5tuple, 0, sizeof(struct e1000_5tuple_filter_info));
4446 	ret = ntuple_filter_to_5tuple_82576(ntuple_filter,
4447 					    &filter_5tuple);
4448 	if (ret < 0)
4449 		return ret;
4450 
4451 	filter = igb_5tuple_filter_lookup_82576(&filter_info->fivetuple_list,
4452 					 &filter_5tuple);
4453 	if (filter == NULL) {
4454 		PMD_DRV_LOG(ERR, "filter doesn't exist.");
4455 		return -ENOENT;
4456 	}
4457 
4458 	igb_delete_5tuple_filter_82576(dev, filter);
4459 
4460 	return 0;
4461 }
4462 
4463 static int
4464 eth_igb_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
4465 {
4466 	uint32_t rctl;
4467 	struct e1000_hw *hw;
4468 	struct rte_eth_dev_info dev_info;
4469 	uint32_t frame_size = mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN +
4470 				     VLAN_TAG_SIZE);
4471 
4472 	hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4473 
4474 #ifdef RTE_LIBRTE_82571_SUPPORT
4475 	/* XXX: not bigger than max_rx_pktlen */
4476 	if (hw->mac.type == e1000_82571)
4477 		return -ENOTSUP;
4478 #endif
4479 	eth_igb_infos_get(dev, &dev_info);
4480 
4481 	/* check that mtu is within the allowed range */
4482 	if ((mtu < ETHER_MIN_MTU) ||
4483 	    (frame_size > dev_info.max_rx_pktlen))
4484 		return -EINVAL;
4485 
4486 	/* refuse mtu that requires the support of scattered packets when this
4487 	 * feature has not been enabled before. */
4488 	if (!dev->data->scattered_rx &&
4489 	    frame_size > dev->data->min_rx_buf_size - RTE_PKTMBUF_HEADROOM)
4490 		return -EINVAL;
4491 
4492 	rctl = E1000_READ_REG(hw, E1000_RCTL);
4493 
4494 	/* switch to jumbo mode if needed */
4495 	if (frame_size > ETHER_MAX_LEN) {
4496 		dev->data->dev_conf.rxmode.offloads |=
4497 			DEV_RX_OFFLOAD_JUMBO_FRAME;
4498 		rctl |= E1000_RCTL_LPE;
4499 	} else {
4500 		dev->data->dev_conf.rxmode.offloads &=
4501 			~DEV_RX_OFFLOAD_JUMBO_FRAME;
4502 		rctl &= ~E1000_RCTL_LPE;
4503 	}
4504 	E1000_WRITE_REG(hw, E1000_RCTL, rctl);
4505 
4506 	/* update max frame size */
4507 	dev->data->dev_conf.rxmode.max_rx_pkt_len = frame_size;
4508 
4509 	E1000_WRITE_REG(hw, E1000_RLPML,
4510 			dev->data->dev_conf.rxmode.max_rx_pkt_len);
4511 
4512 	return 0;
4513 }
4514 
4515 /*
4516  * igb_add_del_ntuple_filter - add or delete a ntuple filter
4517  *
4518  * @param
4519  * dev: Pointer to struct rte_eth_dev.
4520  * ntuple_filter: Pointer to struct rte_eth_ntuple_filter
4521  * add: if true, add filter, if false, remove filter
4522  *
4523  * @return
4524  *    - On success, zero.
4525  *    - On failure, a negative value.
4526  */
4527 int
4528 igb_add_del_ntuple_filter(struct rte_eth_dev *dev,
4529 			struct rte_eth_ntuple_filter *ntuple_filter,
4530 			bool add)
4531 {
4532 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4533 	int ret;
4534 
4535 	switch (ntuple_filter->flags) {
4536 	case RTE_5TUPLE_FLAGS:
4537 	case (RTE_5TUPLE_FLAGS | RTE_NTUPLE_FLAGS_TCP_FLAG):
4538 		if (hw->mac.type != e1000_82576)
4539 			return -ENOTSUP;
4540 		if (add)
4541 			ret = igb_add_5tuple_filter_82576(dev,
4542 							  ntuple_filter);
4543 		else
4544 			ret = igb_remove_5tuple_filter_82576(dev,
4545 							     ntuple_filter);
4546 		break;
4547 	case RTE_2TUPLE_FLAGS:
4548 	case (RTE_2TUPLE_FLAGS | RTE_NTUPLE_FLAGS_TCP_FLAG):
4549 		if (hw->mac.type != e1000_82580 && hw->mac.type != e1000_i350 &&
4550 			hw->mac.type != e1000_i210 &&
4551 			hw->mac.type != e1000_i211)
4552 			return -ENOTSUP;
4553 		if (add)
4554 			ret = igb_add_2tuple_filter(dev, ntuple_filter);
4555 		else
4556 			ret = igb_remove_2tuple_filter(dev, ntuple_filter);
4557 		break;
4558 	default:
4559 		ret = -EINVAL;
4560 		break;
4561 	}
4562 
4563 	return ret;
4564 }
4565 
4566 /*
4567  * igb_get_ntuple_filter - get a ntuple filter
4568  *
4569  * @param
4570  * dev: Pointer to struct rte_eth_dev.
4571  * ntuple_filter: Pointer to struct rte_eth_ntuple_filter
4572  *
4573  * @return
4574  *    - On success, zero.
4575  *    - On failure, a negative value.
4576  */
4577 static int
4578 igb_get_ntuple_filter(struct rte_eth_dev *dev,
4579 			struct rte_eth_ntuple_filter *ntuple_filter)
4580 {
4581 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4582 	struct e1000_filter_info *filter_info =
4583 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
4584 	struct e1000_5tuple_filter_info filter_5tuple;
4585 	struct e1000_2tuple_filter_info filter_2tuple;
4586 	struct e1000_5tuple_filter *p_5tuple_filter;
4587 	struct e1000_2tuple_filter *p_2tuple_filter;
4588 	int ret;
4589 
4590 	switch (ntuple_filter->flags) {
4591 	case RTE_5TUPLE_FLAGS:
4592 	case (RTE_5TUPLE_FLAGS | RTE_NTUPLE_FLAGS_TCP_FLAG):
4593 		if (hw->mac.type != e1000_82576)
4594 			return -ENOTSUP;
4595 		memset(&filter_5tuple,
4596 			0,
4597 			sizeof(struct e1000_5tuple_filter_info));
4598 		ret = ntuple_filter_to_5tuple_82576(ntuple_filter,
4599 						    &filter_5tuple);
4600 		if (ret < 0)
4601 			return ret;
4602 		p_5tuple_filter = igb_5tuple_filter_lookup_82576(
4603 					&filter_info->fivetuple_list,
4604 					&filter_5tuple);
4605 		if (p_5tuple_filter == NULL) {
4606 			PMD_DRV_LOG(ERR, "filter doesn't exist.");
4607 			return -ENOENT;
4608 		}
4609 		ntuple_filter->queue = p_5tuple_filter->queue;
4610 		break;
4611 	case RTE_2TUPLE_FLAGS:
4612 	case (RTE_2TUPLE_FLAGS | RTE_NTUPLE_FLAGS_TCP_FLAG):
4613 		if (hw->mac.type != e1000_82580 && hw->mac.type != e1000_i350)
4614 			return -ENOTSUP;
4615 		memset(&filter_2tuple,
4616 			0,
4617 			sizeof(struct e1000_2tuple_filter_info));
4618 		ret = ntuple_filter_to_2tuple(ntuple_filter, &filter_2tuple);
4619 		if (ret < 0)
4620 			return ret;
4621 		p_2tuple_filter = igb_2tuple_filter_lookup(
4622 					&filter_info->twotuple_list,
4623 					&filter_2tuple);
4624 		if (p_2tuple_filter == NULL) {
4625 			PMD_DRV_LOG(ERR, "filter doesn't exist.");
4626 			return -ENOENT;
4627 		}
4628 		ntuple_filter->queue = p_2tuple_filter->queue;
4629 		break;
4630 	default:
4631 		ret = -EINVAL;
4632 		break;
4633 	}
4634 
4635 	return 0;
4636 }
4637 
4638 /*
4639  * igb_ntuple_filter_handle - Handle operations for ntuple filter.
4640  * @dev: pointer to rte_eth_dev structure
4641  * @filter_op:operation will be taken.
4642  * @arg: a pointer to specific structure corresponding to the filter_op
4643  */
4644 static int
4645 igb_ntuple_filter_handle(struct rte_eth_dev *dev,
4646 				enum rte_filter_op filter_op,
4647 				void *arg)
4648 {
4649 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4650 	int ret;
4651 
4652 	MAC_TYPE_FILTER_SUP(hw->mac.type);
4653 
4654 	if (filter_op == RTE_ETH_FILTER_NOP)
4655 		return 0;
4656 
4657 	if (arg == NULL) {
4658 		PMD_DRV_LOG(ERR, "arg shouldn't be NULL for operation %u.",
4659 			    filter_op);
4660 		return -EINVAL;
4661 	}
4662 
4663 	switch (filter_op) {
4664 	case RTE_ETH_FILTER_ADD:
4665 		ret = igb_add_del_ntuple_filter(dev,
4666 			(struct rte_eth_ntuple_filter *)arg,
4667 			TRUE);
4668 		break;
4669 	case RTE_ETH_FILTER_DELETE:
4670 		ret = igb_add_del_ntuple_filter(dev,
4671 			(struct rte_eth_ntuple_filter *)arg,
4672 			FALSE);
4673 		break;
4674 	case RTE_ETH_FILTER_GET:
4675 		ret = igb_get_ntuple_filter(dev,
4676 			(struct rte_eth_ntuple_filter *)arg);
4677 		break;
4678 	default:
4679 		PMD_DRV_LOG(ERR, "unsupported operation %u.", filter_op);
4680 		ret = -EINVAL;
4681 		break;
4682 	}
4683 	return ret;
4684 }
4685 
4686 static inline int
4687 igb_ethertype_filter_lookup(struct e1000_filter_info *filter_info,
4688 			uint16_t ethertype)
4689 {
4690 	int i;
4691 
4692 	for (i = 0; i < E1000_MAX_ETQF_FILTERS; i++) {
4693 		if (filter_info->ethertype_filters[i].ethertype == ethertype &&
4694 		    (filter_info->ethertype_mask & (1 << i)))
4695 			return i;
4696 	}
4697 	return -1;
4698 }
4699 
4700 static inline int
4701 igb_ethertype_filter_insert(struct e1000_filter_info *filter_info,
4702 			uint16_t ethertype, uint32_t etqf)
4703 {
4704 	int i;
4705 
4706 	for (i = 0; i < E1000_MAX_ETQF_FILTERS; i++) {
4707 		if (!(filter_info->ethertype_mask & (1 << i))) {
4708 			filter_info->ethertype_mask |= 1 << i;
4709 			filter_info->ethertype_filters[i].ethertype = ethertype;
4710 			filter_info->ethertype_filters[i].etqf = etqf;
4711 			return i;
4712 		}
4713 	}
4714 	return -1;
4715 }
4716 
4717 int
4718 igb_ethertype_filter_remove(struct e1000_filter_info *filter_info,
4719 			uint8_t idx)
4720 {
4721 	if (idx >= E1000_MAX_ETQF_FILTERS)
4722 		return -1;
4723 	filter_info->ethertype_mask &= ~(1 << idx);
4724 	filter_info->ethertype_filters[idx].ethertype = 0;
4725 	filter_info->ethertype_filters[idx].etqf = 0;
4726 	return idx;
4727 }
4728 
4729 
4730 int
4731 igb_add_del_ethertype_filter(struct rte_eth_dev *dev,
4732 			struct rte_eth_ethertype_filter *filter,
4733 			bool add)
4734 {
4735 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4736 	struct e1000_filter_info *filter_info =
4737 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
4738 	uint32_t etqf = 0;
4739 	int ret;
4740 
4741 	if (filter->ether_type == ETHER_TYPE_IPv4 ||
4742 		filter->ether_type == ETHER_TYPE_IPv6) {
4743 		PMD_DRV_LOG(ERR, "unsupported ether_type(0x%04x) in"
4744 			" ethertype filter.", filter->ether_type);
4745 		return -EINVAL;
4746 	}
4747 
4748 	if (filter->flags & RTE_ETHTYPE_FLAGS_MAC) {
4749 		PMD_DRV_LOG(ERR, "mac compare is unsupported.");
4750 		return -EINVAL;
4751 	}
4752 	if (filter->flags & RTE_ETHTYPE_FLAGS_DROP) {
4753 		PMD_DRV_LOG(ERR, "drop option is unsupported.");
4754 		return -EINVAL;
4755 	}
4756 
4757 	ret = igb_ethertype_filter_lookup(filter_info, filter->ether_type);
4758 	if (ret >= 0 && add) {
4759 		PMD_DRV_LOG(ERR, "ethertype (0x%04x) filter exists.",
4760 			    filter->ether_type);
4761 		return -EEXIST;
4762 	}
4763 	if (ret < 0 && !add) {
4764 		PMD_DRV_LOG(ERR, "ethertype (0x%04x) filter doesn't exist.",
4765 			    filter->ether_type);
4766 		return -ENOENT;
4767 	}
4768 
4769 	if (add) {
4770 		etqf |= E1000_ETQF_FILTER_ENABLE | E1000_ETQF_QUEUE_ENABLE;
4771 		etqf |= (uint32_t)(filter->ether_type & E1000_ETQF_ETHERTYPE);
4772 		etqf |= filter->queue << E1000_ETQF_QUEUE_SHIFT;
4773 		ret = igb_ethertype_filter_insert(filter_info,
4774 				filter->ether_type, etqf);
4775 		if (ret < 0) {
4776 			PMD_DRV_LOG(ERR, "ethertype filters are full.");
4777 			return -ENOSYS;
4778 		}
4779 	} else {
4780 		ret = igb_ethertype_filter_remove(filter_info, (uint8_t)ret);
4781 		if (ret < 0)
4782 			return -ENOSYS;
4783 	}
4784 	E1000_WRITE_REG(hw, E1000_ETQF(ret), etqf);
4785 	E1000_WRITE_FLUSH(hw);
4786 
4787 	return 0;
4788 }
4789 
4790 static int
4791 igb_get_ethertype_filter(struct rte_eth_dev *dev,
4792 			struct rte_eth_ethertype_filter *filter)
4793 {
4794 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4795 	struct e1000_filter_info *filter_info =
4796 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
4797 	uint32_t etqf;
4798 	int ret;
4799 
4800 	ret = igb_ethertype_filter_lookup(filter_info, filter->ether_type);
4801 	if (ret < 0) {
4802 		PMD_DRV_LOG(ERR, "ethertype (0x%04x) filter doesn't exist.",
4803 			    filter->ether_type);
4804 		return -ENOENT;
4805 	}
4806 
4807 	etqf = E1000_READ_REG(hw, E1000_ETQF(ret));
4808 	if (etqf & E1000_ETQF_FILTER_ENABLE) {
4809 		filter->ether_type = etqf & E1000_ETQF_ETHERTYPE;
4810 		filter->flags = 0;
4811 		filter->queue = (etqf & E1000_ETQF_QUEUE) >>
4812 				E1000_ETQF_QUEUE_SHIFT;
4813 		return 0;
4814 	}
4815 
4816 	return -ENOENT;
4817 }
4818 
4819 /*
4820  * igb_ethertype_filter_handle - Handle operations for ethertype filter.
4821  * @dev: pointer to rte_eth_dev structure
4822  * @filter_op:operation will be taken.
4823  * @arg: a pointer to specific structure corresponding to the filter_op
4824  */
4825 static int
4826 igb_ethertype_filter_handle(struct rte_eth_dev *dev,
4827 				enum rte_filter_op filter_op,
4828 				void *arg)
4829 {
4830 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4831 	int ret;
4832 
4833 	MAC_TYPE_FILTER_SUP(hw->mac.type);
4834 
4835 	if (filter_op == RTE_ETH_FILTER_NOP)
4836 		return 0;
4837 
4838 	if (arg == NULL) {
4839 		PMD_DRV_LOG(ERR, "arg shouldn't be NULL for operation %u.",
4840 			    filter_op);
4841 		return -EINVAL;
4842 	}
4843 
4844 	switch (filter_op) {
4845 	case RTE_ETH_FILTER_ADD:
4846 		ret = igb_add_del_ethertype_filter(dev,
4847 			(struct rte_eth_ethertype_filter *)arg,
4848 			TRUE);
4849 		break;
4850 	case RTE_ETH_FILTER_DELETE:
4851 		ret = igb_add_del_ethertype_filter(dev,
4852 			(struct rte_eth_ethertype_filter *)arg,
4853 			FALSE);
4854 		break;
4855 	case RTE_ETH_FILTER_GET:
4856 		ret = igb_get_ethertype_filter(dev,
4857 			(struct rte_eth_ethertype_filter *)arg);
4858 		break;
4859 	default:
4860 		PMD_DRV_LOG(ERR, "unsupported operation %u.", filter_op);
4861 		ret = -EINVAL;
4862 		break;
4863 	}
4864 	return ret;
4865 }
4866 
4867 static int
4868 eth_igb_filter_ctrl(struct rte_eth_dev *dev,
4869 		     enum rte_filter_type filter_type,
4870 		     enum rte_filter_op filter_op,
4871 		     void *arg)
4872 {
4873 	int ret = 0;
4874 
4875 	switch (filter_type) {
4876 	case RTE_ETH_FILTER_NTUPLE:
4877 		ret = igb_ntuple_filter_handle(dev, filter_op, arg);
4878 		break;
4879 	case RTE_ETH_FILTER_ETHERTYPE:
4880 		ret = igb_ethertype_filter_handle(dev, filter_op, arg);
4881 		break;
4882 	case RTE_ETH_FILTER_SYN:
4883 		ret = eth_igb_syn_filter_handle(dev, filter_op, arg);
4884 		break;
4885 	case RTE_ETH_FILTER_FLEXIBLE:
4886 		ret = eth_igb_flex_filter_handle(dev, filter_op, arg);
4887 		break;
4888 	case RTE_ETH_FILTER_GENERIC:
4889 		if (filter_op != RTE_ETH_FILTER_GET)
4890 			return -EINVAL;
4891 		*(const void **)arg = &igb_flow_ops;
4892 		break;
4893 	default:
4894 		PMD_DRV_LOG(WARNING, "Filter type (%d) not supported",
4895 							filter_type);
4896 		break;
4897 	}
4898 
4899 	return ret;
4900 }
4901 
4902 static int
4903 eth_igb_set_mc_addr_list(struct rte_eth_dev *dev,
4904 			 struct ether_addr *mc_addr_set,
4905 			 uint32_t nb_mc_addr)
4906 {
4907 	struct e1000_hw *hw;
4908 
4909 	hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4910 	e1000_update_mc_addr_list(hw, (u8 *)mc_addr_set, nb_mc_addr);
4911 	return 0;
4912 }
4913 
4914 static uint64_t
4915 igb_read_systime_cyclecounter(struct rte_eth_dev *dev)
4916 {
4917 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4918 	uint64_t systime_cycles;
4919 
4920 	switch (hw->mac.type) {
4921 	case e1000_i210:
4922 	case e1000_i211:
4923 		/*
4924 		 * Need to read System Time Residue Register to be able
4925 		 * to read the other two registers.
4926 		 */
4927 		E1000_READ_REG(hw, E1000_SYSTIMR);
4928 		/* SYSTIMEL stores ns and SYSTIMEH stores seconds. */
4929 		systime_cycles = (uint64_t)E1000_READ_REG(hw, E1000_SYSTIML);
4930 		systime_cycles += (uint64_t)E1000_READ_REG(hw, E1000_SYSTIMH)
4931 				* NSEC_PER_SEC;
4932 		break;
4933 	case e1000_82580:
4934 	case e1000_i350:
4935 	case e1000_i354:
4936 		/*
4937 		 * Need to read System Time Residue Register to be able
4938 		 * to read the other two registers.
4939 		 */
4940 		E1000_READ_REG(hw, E1000_SYSTIMR);
4941 		systime_cycles = (uint64_t)E1000_READ_REG(hw, E1000_SYSTIML);
4942 		/* Only the 8 LSB are valid. */
4943 		systime_cycles |= (uint64_t)(E1000_READ_REG(hw, E1000_SYSTIMH)
4944 				& 0xff) << 32;
4945 		break;
4946 	default:
4947 		systime_cycles = (uint64_t)E1000_READ_REG(hw, E1000_SYSTIML);
4948 		systime_cycles |= (uint64_t)E1000_READ_REG(hw, E1000_SYSTIMH)
4949 				<< 32;
4950 		break;
4951 	}
4952 
4953 	return systime_cycles;
4954 }
4955 
4956 static uint64_t
4957 igb_read_rx_tstamp_cyclecounter(struct rte_eth_dev *dev)
4958 {
4959 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4960 	uint64_t rx_tstamp_cycles;
4961 
4962 	switch (hw->mac.type) {
4963 	case e1000_i210:
4964 	case e1000_i211:
4965 		/* RXSTMPL stores ns and RXSTMPH stores seconds. */
4966 		rx_tstamp_cycles = (uint64_t)E1000_READ_REG(hw, E1000_RXSTMPL);
4967 		rx_tstamp_cycles += (uint64_t)E1000_READ_REG(hw, E1000_RXSTMPH)
4968 				* NSEC_PER_SEC;
4969 		break;
4970 	case e1000_82580:
4971 	case e1000_i350:
4972 	case e1000_i354:
4973 		rx_tstamp_cycles = (uint64_t)E1000_READ_REG(hw, E1000_RXSTMPL);
4974 		/* Only the 8 LSB are valid. */
4975 		rx_tstamp_cycles |= (uint64_t)(E1000_READ_REG(hw, E1000_RXSTMPH)
4976 				& 0xff) << 32;
4977 		break;
4978 	default:
4979 		rx_tstamp_cycles = (uint64_t)E1000_READ_REG(hw, E1000_RXSTMPL);
4980 		rx_tstamp_cycles |= (uint64_t)E1000_READ_REG(hw, E1000_RXSTMPH)
4981 				<< 32;
4982 		break;
4983 	}
4984 
4985 	return rx_tstamp_cycles;
4986 }
4987 
4988 static uint64_t
4989 igb_read_tx_tstamp_cyclecounter(struct rte_eth_dev *dev)
4990 {
4991 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4992 	uint64_t tx_tstamp_cycles;
4993 
4994 	switch (hw->mac.type) {
4995 	case e1000_i210:
4996 	case e1000_i211:
4997 		/* RXSTMPL stores ns and RXSTMPH stores seconds. */
4998 		tx_tstamp_cycles = (uint64_t)E1000_READ_REG(hw, E1000_TXSTMPL);
4999 		tx_tstamp_cycles += (uint64_t)E1000_READ_REG(hw, E1000_TXSTMPH)
5000 				* NSEC_PER_SEC;
5001 		break;
5002 	case e1000_82580:
5003 	case e1000_i350:
5004 	case e1000_i354:
5005 		tx_tstamp_cycles = (uint64_t)E1000_READ_REG(hw, E1000_TXSTMPL);
5006 		/* Only the 8 LSB are valid. */
5007 		tx_tstamp_cycles |= (uint64_t)(E1000_READ_REG(hw, E1000_TXSTMPH)
5008 				& 0xff) << 32;
5009 		break;
5010 	default:
5011 		tx_tstamp_cycles = (uint64_t)E1000_READ_REG(hw, E1000_TXSTMPL);
5012 		tx_tstamp_cycles |= (uint64_t)E1000_READ_REG(hw, E1000_TXSTMPH)
5013 				<< 32;
5014 		break;
5015 	}
5016 
5017 	return tx_tstamp_cycles;
5018 }
5019 
5020 static void
5021 igb_start_timecounters(struct rte_eth_dev *dev)
5022 {
5023 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5024 	struct e1000_adapter *adapter = dev->data->dev_private;
5025 	uint32_t incval = 1;
5026 	uint32_t shift = 0;
5027 	uint64_t mask = E1000_CYCLECOUNTER_MASK;
5028 
5029 	switch (hw->mac.type) {
5030 	case e1000_82580:
5031 	case e1000_i350:
5032 	case e1000_i354:
5033 		/* 32 LSB bits + 8 MSB bits = 40 bits */
5034 		mask = (1ULL << 40) - 1;
5035 		/* fall-through */
5036 	case e1000_i210:
5037 	case e1000_i211:
5038 		/*
5039 		 * Start incrementing the register
5040 		 * used to timestamp PTP packets.
5041 		 */
5042 		E1000_WRITE_REG(hw, E1000_TIMINCA, incval);
5043 		break;
5044 	case e1000_82576:
5045 		incval = E1000_INCVALUE_82576;
5046 		shift = IGB_82576_TSYNC_SHIFT;
5047 		E1000_WRITE_REG(hw, E1000_TIMINCA,
5048 				E1000_INCPERIOD_82576 | incval);
5049 		break;
5050 	default:
5051 		/* Not supported */
5052 		return;
5053 	}
5054 
5055 	memset(&adapter->systime_tc, 0, sizeof(struct rte_timecounter));
5056 	memset(&adapter->rx_tstamp_tc, 0, sizeof(struct rte_timecounter));
5057 	memset(&adapter->tx_tstamp_tc, 0, sizeof(struct rte_timecounter));
5058 
5059 	adapter->systime_tc.cc_mask = mask;
5060 	adapter->systime_tc.cc_shift = shift;
5061 	adapter->systime_tc.nsec_mask = (1ULL << shift) - 1;
5062 
5063 	adapter->rx_tstamp_tc.cc_mask = mask;
5064 	adapter->rx_tstamp_tc.cc_shift = shift;
5065 	adapter->rx_tstamp_tc.nsec_mask = (1ULL << shift) - 1;
5066 
5067 	adapter->tx_tstamp_tc.cc_mask = mask;
5068 	adapter->tx_tstamp_tc.cc_shift = shift;
5069 	adapter->tx_tstamp_tc.nsec_mask = (1ULL << shift) - 1;
5070 }
5071 
5072 static int
5073 igb_timesync_adjust_time(struct rte_eth_dev *dev, int64_t delta)
5074 {
5075 	struct e1000_adapter *adapter = dev->data->dev_private;
5076 
5077 	adapter->systime_tc.nsec += delta;
5078 	adapter->rx_tstamp_tc.nsec += delta;
5079 	adapter->tx_tstamp_tc.nsec += delta;
5080 
5081 	return 0;
5082 }
5083 
5084 static int
5085 igb_timesync_write_time(struct rte_eth_dev *dev, const struct timespec *ts)
5086 {
5087 	uint64_t ns;
5088 	struct e1000_adapter *adapter = dev->data->dev_private;
5089 
5090 	ns = rte_timespec_to_ns(ts);
5091 
5092 	/* Set the timecounters to a new value. */
5093 	adapter->systime_tc.nsec = ns;
5094 	adapter->rx_tstamp_tc.nsec = ns;
5095 	adapter->tx_tstamp_tc.nsec = ns;
5096 
5097 	return 0;
5098 }
5099 
5100 static int
5101 igb_timesync_read_time(struct rte_eth_dev *dev, struct timespec *ts)
5102 {
5103 	uint64_t ns, systime_cycles;
5104 	struct e1000_adapter *adapter = dev->data->dev_private;
5105 
5106 	systime_cycles = igb_read_systime_cyclecounter(dev);
5107 	ns = rte_timecounter_update(&adapter->systime_tc, systime_cycles);
5108 	*ts = rte_ns_to_timespec(ns);
5109 
5110 	return 0;
5111 }
5112 
5113 static int
5114 igb_timesync_enable(struct rte_eth_dev *dev)
5115 {
5116 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5117 	uint32_t tsync_ctl;
5118 	uint32_t tsauxc;
5119 
5120 	/* Stop the timesync system time. */
5121 	E1000_WRITE_REG(hw, E1000_TIMINCA, 0x0);
5122 	/* Reset the timesync system time value. */
5123 	switch (hw->mac.type) {
5124 	case e1000_82580:
5125 	case e1000_i350:
5126 	case e1000_i354:
5127 	case e1000_i210:
5128 	case e1000_i211:
5129 		E1000_WRITE_REG(hw, E1000_SYSTIMR, 0x0);
5130 		/* fall-through */
5131 	case e1000_82576:
5132 		E1000_WRITE_REG(hw, E1000_SYSTIML, 0x0);
5133 		E1000_WRITE_REG(hw, E1000_SYSTIMH, 0x0);
5134 		break;
5135 	default:
5136 		/* Not supported. */
5137 		return -ENOTSUP;
5138 	}
5139 
5140 	/* Enable system time for it isn't on by default. */
5141 	tsauxc = E1000_READ_REG(hw, E1000_TSAUXC);
5142 	tsauxc &= ~E1000_TSAUXC_DISABLE_SYSTIME;
5143 	E1000_WRITE_REG(hw, E1000_TSAUXC, tsauxc);
5144 
5145 	igb_start_timecounters(dev);
5146 
5147 	/* Enable L2 filtering of IEEE1588/802.1AS Ethernet frame types. */
5148 	E1000_WRITE_REG(hw, E1000_ETQF(E1000_ETQF_FILTER_1588),
5149 			(ETHER_TYPE_1588 |
5150 			 E1000_ETQF_FILTER_ENABLE |
5151 			 E1000_ETQF_1588));
5152 
5153 	/* Enable timestamping of received PTP packets. */
5154 	tsync_ctl = E1000_READ_REG(hw, E1000_TSYNCRXCTL);
5155 	tsync_ctl |= E1000_TSYNCRXCTL_ENABLED;
5156 	E1000_WRITE_REG(hw, E1000_TSYNCRXCTL, tsync_ctl);
5157 
5158 	/* Enable Timestamping of transmitted PTP packets. */
5159 	tsync_ctl = E1000_READ_REG(hw, E1000_TSYNCTXCTL);
5160 	tsync_ctl |= E1000_TSYNCTXCTL_ENABLED;
5161 	E1000_WRITE_REG(hw, E1000_TSYNCTXCTL, tsync_ctl);
5162 
5163 	return 0;
5164 }
5165 
5166 static int
5167 igb_timesync_disable(struct rte_eth_dev *dev)
5168 {
5169 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5170 	uint32_t tsync_ctl;
5171 
5172 	/* Disable timestamping of transmitted PTP packets. */
5173 	tsync_ctl = E1000_READ_REG(hw, E1000_TSYNCTXCTL);
5174 	tsync_ctl &= ~E1000_TSYNCTXCTL_ENABLED;
5175 	E1000_WRITE_REG(hw, E1000_TSYNCTXCTL, tsync_ctl);
5176 
5177 	/* Disable timestamping of received PTP packets. */
5178 	tsync_ctl = E1000_READ_REG(hw, E1000_TSYNCRXCTL);
5179 	tsync_ctl &= ~E1000_TSYNCRXCTL_ENABLED;
5180 	E1000_WRITE_REG(hw, E1000_TSYNCRXCTL, tsync_ctl);
5181 
5182 	/* Disable L2 filtering of IEEE1588/802.1AS Ethernet frame types. */
5183 	E1000_WRITE_REG(hw, E1000_ETQF(E1000_ETQF_FILTER_1588), 0);
5184 
5185 	/* Stop incrementating the System Time registers. */
5186 	E1000_WRITE_REG(hw, E1000_TIMINCA, 0);
5187 
5188 	return 0;
5189 }
5190 
5191 static int
5192 igb_timesync_read_rx_timestamp(struct rte_eth_dev *dev,
5193 			       struct timespec *timestamp,
5194 			       uint32_t flags __rte_unused)
5195 {
5196 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5197 	struct e1000_adapter *adapter = dev->data->dev_private;
5198 	uint32_t tsync_rxctl;
5199 	uint64_t rx_tstamp_cycles;
5200 	uint64_t ns;
5201 
5202 	tsync_rxctl = E1000_READ_REG(hw, E1000_TSYNCRXCTL);
5203 	if ((tsync_rxctl & E1000_TSYNCRXCTL_VALID) == 0)
5204 		return -EINVAL;
5205 
5206 	rx_tstamp_cycles = igb_read_rx_tstamp_cyclecounter(dev);
5207 	ns = rte_timecounter_update(&adapter->rx_tstamp_tc, rx_tstamp_cycles);
5208 	*timestamp = rte_ns_to_timespec(ns);
5209 
5210 	return  0;
5211 }
5212 
5213 static int
5214 igb_timesync_read_tx_timestamp(struct rte_eth_dev *dev,
5215 			       struct timespec *timestamp)
5216 {
5217 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5218 	struct e1000_adapter *adapter = dev->data->dev_private;
5219 	uint32_t tsync_txctl;
5220 	uint64_t tx_tstamp_cycles;
5221 	uint64_t ns;
5222 
5223 	tsync_txctl = E1000_READ_REG(hw, E1000_TSYNCTXCTL);
5224 	if ((tsync_txctl & E1000_TSYNCTXCTL_VALID) == 0)
5225 		return -EINVAL;
5226 
5227 	tx_tstamp_cycles = igb_read_tx_tstamp_cyclecounter(dev);
5228 	ns = rte_timecounter_update(&adapter->tx_tstamp_tc, tx_tstamp_cycles);
5229 	*timestamp = rte_ns_to_timespec(ns);
5230 
5231 	return  0;
5232 }
5233 
5234 static int
5235 eth_igb_get_reg_length(struct rte_eth_dev *dev __rte_unused)
5236 {
5237 	int count = 0;
5238 	int g_ind = 0;
5239 	const struct reg_info *reg_group;
5240 
5241 	while ((reg_group = igb_regs[g_ind++]))
5242 		count += igb_reg_group_count(reg_group);
5243 
5244 	return count;
5245 }
5246 
5247 static int
5248 igbvf_get_reg_length(struct rte_eth_dev *dev __rte_unused)
5249 {
5250 	int count = 0;
5251 	int g_ind = 0;
5252 	const struct reg_info *reg_group;
5253 
5254 	while ((reg_group = igbvf_regs[g_ind++]))
5255 		count += igb_reg_group_count(reg_group);
5256 
5257 	return count;
5258 }
5259 
5260 static int
5261 eth_igb_get_regs(struct rte_eth_dev *dev,
5262 	struct rte_dev_reg_info *regs)
5263 {
5264 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5265 	uint32_t *data = regs->data;
5266 	int g_ind = 0;
5267 	int count = 0;
5268 	const struct reg_info *reg_group;
5269 
5270 	if (data == NULL) {
5271 		regs->length = eth_igb_get_reg_length(dev);
5272 		regs->width = sizeof(uint32_t);
5273 		return 0;
5274 	}
5275 
5276 	/* Support only full register dump */
5277 	if ((regs->length == 0) ||
5278 	    (regs->length == (uint32_t)eth_igb_get_reg_length(dev))) {
5279 		regs->version = hw->mac.type << 24 | hw->revision_id << 16 |
5280 			hw->device_id;
5281 		while ((reg_group = igb_regs[g_ind++]))
5282 			count += igb_read_regs_group(dev, &data[count],
5283 							reg_group);
5284 		return 0;
5285 	}
5286 
5287 	return -ENOTSUP;
5288 }
5289 
5290 static int
5291 igbvf_get_regs(struct rte_eth_dev *dev,
5292 	struct rte_dev_reg_info *regs)
5293 {
5294 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5295 	uint32_t *data = regs->data;
5296 	int g_ind = 0;
5297 	int count = 0;
5298 	const struct reg_info *reg_group;
5299 
5300 	if (data == NULL) {
5301 		regs->length = igbvf_get_reg_length(dev);
5302 		regs->width = sizeof(uint32_t);
5303 		return 0;
5304 	}
5305 
5306 	/* Support only full register dump */
5307 	if ((regs->length == 0) ||
5308 	    (regs->length == (uint32_t)igbvf_get_reg_length(dev))) {
5309 		regs->version = hw->mac.type << 24 | hw->revision_id << 16 |
5310 			hw->device_id;
5311 		while ((reg_group = igbvf_regs[g_ind++]))
5312 			count += igb_read_regs_group(dev, &data[count],
5313 							reg_group);
5314 		return 0;
5315 	}
5316 
5317 	return -ENOTSUP;
5318 }
5319 
5320 static int
5321 eth_igb_get_eeprom_length(struct rte_eth_dev *dev)
5322 {
5323 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5324 
5325 	/* Return unit is byte count */
5326 	return hw->nvm.word_size * 2;
5327 }
5328 
5329 static int
5330 eth_igb_get_eeprom(struct rte_eth_dev *dev,
5331 	struct rte_dev_eeprom_info *in_eeprom)
5332 {
5333 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5334 	struct e1000_nvm_info *nvm = &hw->nvm;
5335 	uint16_t *data = in_eeprom->data;
5336 	int first, length;
5337 
5338 	first = in_eeprom->offset >> 1;
5339 	length = in_eeprom->length >> 1;
5340 	if ((first >= hw->nvm.word_size) ||
5341 	    ((first + length) >= hw->nvm.word_size))
5342 		return -EINVAL;
5343 
5344 	in_eeprom->magic = hw->vendor_id |
5345 		((uint32_t)hw->device_id << 16);
5346 
5347 	if ((nvm->ops.read) == NULL)
5348 		return -ENOTSUP;
5349 
5350 	return nvm->ops.read(hw, first, length, data);
5351 }
5352 
5353 static int
5354 eth_igb_set_eeprom(struct rte_eth_dev *dev,
5355 	struct rte_dev_eeprom_info *in_eeprom)
5356 {
5357 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5358 	struct e1000_nvm_info *nvm = &hw->nvm;
5359 	uint16_t *data = in_eeprom->data;
5360 	int first, length;
5361 
5362 	first = in_eeprom->offset >> 1;
5363 	length = in_eeprom->length >> 1;
5364 	if ((first >= hw->nvm.word_size) ||
5365 	    ((first + length) >= hw->nvm.word_size))
5366 		return -EINVAL;
5367 
5368 	in_eeprom->magic = (uint32_t)hw->vendor_id |
5369 		((uint32_t)hw->device_id << 16);
5370 
5371 	if ((nvm->ops.write) == NULL)
5372 		return -ENOTSUP;
5373 	return nvm->ops.write(hw,  first, length, data);
5374 }
5375 
5376 static int
5377 eth_igb_get_module_info(struct rte_eth_dev *dev,
5378 			struct rte_eth_dev_module_info *modinfo)
5379 {
5380 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5381 
5382 	uint32_t status = 0;
5383 	uint16_t sff8472_rev, addr_mode;
5384 	bool page_swap = false;
5385 
5386 	if (hw->phy.media_type == e1000_media_type_copper ||
5387 	    hw->phy.media_type == e1000_media_type_unknown)
5388 		return -EOPNOTSUPP;
5389 
5390 	/* Check whether we support SFF-8472 or not */
5391 	status = e1000_read_phy_reg_i2c(hw, IGB_SFF_8472_COMP, &sff8472_rev);
5392 	if (status)
5393 		return -EIO;
5394 
5395 	/* addressing mode is not supported */
5396 	status = e1000_read_phy_reg_i2c(hw, IGB_SFF_8472_SWAP, &addr_mode);
5397 	if (status)
5398 		return -EIO;
5399 
5400 	/* addressing mode is not supported */
5401 	if ((addr_mode & 0xFF) & IGB_SFF_ADDRESSING_MODE) {
5402 		PMD_DRV_LOG(ERR,
5403 			    "Address change required to access page 0xA2, "
5404 			    "but not supported. Please report the module "
5405 			    "type to the driver maintainers.\n");
5406 		page_swap = true;
5407 	}
5408 
5409 	if ((sff8472_rev & 0xFF) == IGB_SFF_8472_UNSUP || page_swap) {
5410 		/* We have an SFP, but it does not support SFF-8472 */
5411 		modinfo->type = RTE_ETH_MODULE_SFF_8079;
5412 		modinfo->eeprom_len = RTE_ETH_MODULE_SFF_8079_LEN;
5413 	} else {
5414 		/* We have an SFP which supports a revision of SFF-8472 */
5415 		modinfo->type = RTE_ETH_MODULE_SFF_8472;
5416 		modinfo->eeprom_len = RTE_ETH_MODULE_SFF_8472_LEN;
5417 	}
5418 
5419 	return 0;
5420 }
5421 
5422 static int
5423 eth_igb_get_module_eeprom(struct rte_eth_dev *dev,
5424 			  struct rte_dev_eeprom_info *info)
5425 {
5426 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5427 
5428 	uint32_t status = 0;
5429 	uint16_t dataword[RTE_ETH_MODULE_SFF_8472_LEN / 2 + 1];
5430 	u16 first_word, last_word;
5431 	int i = 0;
5432 
5433 	if (info->length == 0)
5434 		return -EINVAL;
5435 
5436 	first_word = info->offset >> 1;
5437 	last_word = (info->offset + info->length - 1) >> 1;
5438 
5439 	/* Read EEPROM block, SFF-8079/SFF-8472, word at a time */
5440 	for (i = 0; i < last_word - first_word + 1; i++) {
5441 		status = e1000_read_phy_reg_i2c(hw, (first_word + i) * 2,
5442 						&dataword[i]);
5443 		if (status) {
5444 			/* Error occurred while reading module */
5445 			return -EIO;
5446 		}
5447 
5448 		dataword[i] = rte_be_to_cpu_16(dataword[i]);
5449 	}
5450 
5451 	memcpy(info->data, (u8 *)dataword + (info->offset & 1), info->length);
5452 
5453 	return 0;
5454 }
5455 
5456 static int
5457 eth_igb_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
5458 {
5459 	struct e1000_hw *hw =
5460 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5461 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
5462 	struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
5463 	uint32_t vec = E1000_MISC_VEC_ID;
5464 
5465 	if (rte_intr_allow_others(intr_handle))
5466 		vec = E1000_RX_VEC_START;
5467 
5468 	uint32_t mask = 1 << (queue_id + vec);
5469 
5470 	E1000_WRITE_REG(hw, E1000_EIMC, mask);
5471 	E1000_WRITE_FLUSH(hw);
5472 
5473 	return 0;
5474 }
5475 
5476 static int
5477 eth_igb_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
5478 {
5479 	struct e1000_hw *hw =
5480 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5481 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
5482 	struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
5483 	uint32_t vec = E1000_MISC_VEC_ID;
5484 
5485 	if (rte_intr_allow_others(intr_handle))
5486 		vec = E1000_RX_VEC_START;
5487 
5488 	uint32_t mask = 1 << (queue_id + vec);
5489 	uint32_t regval;
5490 
5491 	regval = E1000_READ_REG(hw, E1000_EIMS);
5492 	E1000_WRITE_REG(hw, E1000_EIMS, regval | mask);
5493 	E1000_WRITE_FLUSH(hw);
5494 
5495 	rte_intr_enable(intr_handle);
5496 
5497 	return 0;
5498 }
5499 
5500 static void
5501 eth_igb_write_ivar(struct e1000_hw *hw, uint8_t  msix_vector,
5502 		   uint8_t index, uint8_t offset)
5503 {
5504 	uint32_t val = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
5505 
5506 	/* clear bits */
5507 	val &= ~((uint32_t)0xFF << offset);
5508 
5509 	/* write vector and valid bit */
5510 	val |= (msix_vector | E1000_IVAR_VALID) << offset;
5511 
5512 	E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, val);
5513 }
5514 
5515 static void
5516 eth_igb_assign_msix_vector(struct e1000_hw *hw, int8_t direction,
5517 			   uint8_t queue, uint8_t msix_vector)
5518 {
5519 	uint32_t tmp = 0;
5520 
5521 	if (hw->mac.type == e1000_82575) {
5522 		if (direction == 0)
5523 			tmp = E1000_EICR_RX_QUEUE0 << queue;
5524 		else if (direction == 1)
5525 			tmp = E1000_EICR_TX_QUEUE0 << queue;
5526 		E1000_WRITE_REG(hw, E1000_MSIXBM(msix_vector), tmp);
5527 	} else if (hw->mac.type == e1000_82576) {
5528 		if ((direction == 0) || (direction == 1))
5529 			eth_igb_write_ivar(hw, msix_vector, queue & 0x7,
5530 					   ((queue & 0x8) << 1) +
5531 					   8 * direction);
5532 	} else if ((hw->mac.type == e1000_82580) ||
5533 			(hw->mac.type == e1000_i350) ||
5534 			(hw->mac.type == e1000_i354) ||
5535 			(hw->mac.type == e1000_i210) ||
5536 			(hw->mac.type == e1000_i211)) {
5537 		if ((direction == 0) || (direction == 1))
5538 			eth_igb_write_ivar(hw, msix_vector,
5539 					   queue >> 1,
5540 					   ((queue & 0x1) << 4) +
5541 					   8 * direction);
5542 	}
5543 }
5544 
5545 /* Sets up the hardware to generate MSI-X interrupts properly
5546  * @hw
5547  *  board private structure
5548  */
5549 static void
5550 eth_igb_configure_msix_intr(struct rte_eth_dev *dev)
5551 {
5552 	int queue_id;
5553 	uint32_t tmpval, regval, intr_mask;
5554 	struct e1000_hw *hw =
5555 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5556 	uint32_t vec = E1000_MISC_VEC_ID;
5557 	uint32_t base = E1000_MISC_VEC_ID;
5558 	uint32_t misc_shift = 0;
5559 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
5560 	struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
5561 
5562 	/* won't configure msix register if no mapping is done
5563 	 * between intr vector and event fd
5564 	 */
5565 	if (!rte_intr_dp_is_en(intr_handle))
5566 		return;
5567 
5568 	if (rte_intr_allow_others(intr_handle)) {
5569 		vec = base = E1000_RX_VEC_START;
5570 		misc_shift = 1;
5571 	}
5572 
5573 	/* set interrupt vector for other causes */
5574 	if (hw->mac.type == e1000_82575) {
5575 		tmpval = E1000_READ_REG(hw, E1000_CTRL_EXT);
5576 		/* enable MSI-X PBA support */
5577 		tmpval |= E1000_CTRL_EXT_PBA_CLR;
5578 
5579 		/* Auto-Mask interrupts upon ICR read */
5580 		tmpval |= E1000_CTRL_EXT_EIAME;
5581 		tmpval |= E1000_CTRL_EXT_IRCA;
5582 
5583 		E1000_WRITE_REG(hw, E1000_CTRL_EXT, tmpval);
5584 
5585 		/* enable msix_other interrupt */
5586 		E1000_WRITE_REG_ARRAY(hw, E1000_MSIXBM(0), 0, E1000_EIMS_OTHER);
5587 		regval = E1000_READ_REG(hw, E1000_EIAC);
5588 		E1000_WRITE_REG(hw, E1000_EIAC, regval | E1000_EIMS_OTHER);
5589 		regval = E1000_READ_REG(hw, E1000_EIAM);
5590 		E1000_WRITE_REG(hw, E1000_EIMS, regval | E1000_EIMS_OTHER);
5591 	} else if ((hw->mac.type == e1000_82576) ||
5592 			(hw->mac.type == e1000_82580) ||
5593 			(hw->mac.type == e1000_i350) ||
5594 			(hw->mac.type == e1000_i354) ||
5595 			(hw->mac.type == e1000_i210) ||
5596 			(hw->mac.type == e1000_i211)) {
5597 		/* turn on MSI-X capability first */
5598 		E1000_WRITE_REG(hw, E1000_GPIE, E1000_GPIE_MSIX_MODE |
5599 					E1000_GPIE_PBA | E1000_GPIE_EIAME |
5600 					E1000_GPIE_NSICR);
5601 		intr_mask = RTE_LEN2MASK(intr_handle->nb_efd, uint32_t) <<
5602 			misc_shift;
5603 
5604 		if (dev->data->dev_conf.intr_conf.lsc != 0)
5605 			intr_mask |= (1 << IGB_MSIX_OTHER_INTR_VEC);
5606 
5607 		regval = E1000_READ_REG(hw, E1000_EIAC);
5608 		E1000_WRITE_REG(hw, E1000_EIAC, regval | intr_mask);
5609 
5610 		/* enable msix_other interrupt */
5611 		regval = E1000_READ_REG(hw, E1000_EIMS);
5612 		E1000_WRITE_REG(hw, E1000_EIMS, regval | intr_mask);
5613 		tmpval = (IGB_MSIX_OTHER_INTR_VEC | E1000_IVAR_VALID) << 8;
5614 		E1000_WRITE_REG(hw, E1000_IVAR_MISC, tmpval);
5615 	}
5616 
5617 	/* use EIAM to auto-mask when MSI-X interrupt
5618 	 * is asserted, this saves a register write for every interrupt
5619 	 */
5620 	intr_mask = RTE_LEN2MASK(intr_handle->nb_efd, uint32_t) <<
5621 		misc_shift;
5622 
5623 	if (dev->data->dev_conf.intr_conf.lsc != 0)
5624 		intr_mask |= (1 << IGB_MSIX_OTHER_INTR_VEC);
5625 
5626 	regval = E1000_READ_REG(hw, E1000_EIAM);
5627 	E1000_WRITE_REG(hw, E1000_EIAM, regval | intr_mask);
5628 
5629 	for (queue_id = 0; queue_id < dev->data->nb_rx_queues; queue_id++) {
5630 		eth_igb_assign_msix_vector(hw, 0, queue_id, vec);
5631 		intr_handle->intr_vec[queue_id] = vec;
5632 		if (vec < base + intr_handle->nb_efd - 1)
5633 			vec++;
5634 	}
5635 
5636 	E1000_WRITE_FLUSH(hw);
5637 }
5638 
5639 /* restore n-tuple filter */
5640 static inline void
5641 igb_ntuple_filter_restore(struct rte_eth_dev *dev)
5642 {
5643 	struct e1000_filter_info *filter_info =
5644 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
5645 	struct e1000_5tuple_filter *p_5tuple;
5646 	struct e1000_2tuple_filter *p_2tuple;
5647 
5648 	TAILQ_FOREACH(p_5tuple, &filter_info->fivetuple_list, entries) {
5649 		igb_inject_5tuple_filter_82576(dev, p_5tuple);
5650 	}
5651 
5652 	TAILQ_FOREACH(p_2tuple, &filter_info->twotuple_list, entries) {
5653 		igb_inject_2uple_filter(dev, p_2tuple);
5654 	}
5655 }
5656 
5657 /* restore SYN filter */
5658 static inline void
5659 igb_syn_filter_restore(struct rte_eth_dev *dev)
5660 {
5661 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5662 	struct e1000_filter_info *filter_info =
5663 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
5664 	uint32_t synqf;
5665 
5666 	synqf = filter_info->syn_info;
5667 
5668 	if (synqf & E1000_SYN_FILTER_ENABLE) {
5669 		E1000_WRITE_REG(hw, E1000_SYNQF(0), synqf);
5670 		E1000_WRITE_FLUSH(hw);
5671 	}
5672 }
5673 
5674 /* restore ethernet type filter */
5675 static inline void
5676 igb_ethertype_filter_restore(struct rte_eth_dev *dev)
5677 {
5678 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5679 	struct e1000_filter_info *filter_info =
5680 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
5681 	int i;
5682 
5683 	for (i = 0; i < E1000_MAX_ETQF_FILTERS; i++) {
5684 		if (filter_info->ethertype_mask & (1 << i)) {
5685 			E1000_WRITE_REG(hw, E1000_ETQF(i),
5686 				filter_info->ethertype_filters[i].etqf);
5687 			E1000_WRITE_FLUSH(hw);
5688 		}
5689 	}
5690 }
5691 
5692 /* restore flex byte filter */
5693 static inline void
5694 igb_flex_filter_restore(struct rte_eth_dev *dev)
5695 {
5696 	struct e1000_filter_info *filter_info =
5697 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
5698 	struct e1000_flex_filter *flex_filter;
5699 
5700 	TAILQ_FOREACH(flex_filter, &filter_info->flex_list, entries) {
5701 		igb_inject_flex_filter(dev, flex_filter);
5702 	}
5703 }
5704 
5705 /* restore rss filter */
5706 static inline void
5707 igb_rss_filter_restore(struct rte_eth_dev *dev)
5708 {
5709 	struct e1000_filter_info *filter_info =
5710 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
5711 
5712 	if (filter_info->rss_info.conf.queue_num)
5713 		igb_config_rss_filter(dev, &filter_info->rss_info, TRUE);
5714 }
5715 
5716 /* restore all types filter */
5717 static int
5718 igb_filter_restore(struct rte_eth_dev *dev)
5719 {
5720 	igb_ntuple_filter_restore(dev);
5721 	igb_ethertype_filter_restore(dev);
5722 	igb_syn_filter_restore(dev);
5723 	igb_flex_filter_restore(dev);
5724 	igb_rss_filter_restore(dev);
5725 
5726 	return 0;
5727 }
5728 
5729 RTE_PMD_REGISTER_PCI(net_e1000_igb, rte_igb_pmd);
5730 RTE_PMD_REGISTER_PCI_TABLE(net_e1000_igb, pci_id_igb_map);
5731 RTE_PMD_REGISTER_KMOD_DEP(net_e1000_igb, "* igb_uio | uio_pci_generic | vfio-pci");
5732 RTE_PMD_REGISTER_PCI(net_e1000_igb_vf, rte_igbvf_pmd);
5733 RTE_PMD_REGISTER_PCI_TABLE(net_e1000_igb_vf, pci_id_igbvf_map);
5734 RTE_PMD_REGISTER_KMOD_DEP(net_e1000_igb_vf, "* igb_uio | vfio-pci");
5735 
5736 /* see e1000_logs.c */
5737 RTE_INIT(e1000_init_log)
5738 {
5739 	e1000_igb_init_log();
5740 }
5741