xref: /dpdk/drivers/net/e1000/igb_ethdev.c (revision cf435a07)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
5  *   All rights reserved.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
11  *     * Redistributions of source code must retain the above copyright
12  *       notice, this list of conditions and the following disclaimer.
13  *     * Redistributions in binary form must reproduce the above copyright
14  *       notice, this list of conditions and the following disclaimer in
15  *       the documentation and/or other materials provided with the
16  *       distribution.
17  *     * Neither the name of Intel Corporation nor the names of its
18  *       contributors may be used to endorse or promote products derived
19  *       from this software without specific prior written permission.
20  *
21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include <sys/queue.h>
35 #include <stdio.h>
36 #include <errno.h>
37 #include <stdint.h>
38 #include <stdarg.h>
39 
40 #include <rte_common.h>
41 #include <rte_interrupts.h>
42 #include <rte_byteorder.h>
43 #include <rte_log.h>
44 #include <rte_debug.h>
45 #include <rte_pci.h>
46 #include <rte_ether.h>
47 #include <rte_ethdev.h>
48 #include <rte_memory.h>
49 #include <rte_memzone.h>
50 #include <rte_eal.h>
51 #include <rte_atomic.h>
52 #include <rte_malloc.h>
53 #include <rte_dev.h>
54 
55 #include "e1000_logs.h"
56 #include "base/e1000_api.h"
57 #include "e1000_ethdev.h"
58 #include "igb_regs.h"
59 
60 /*
61  * Default values for port configuration
62  */
63 #define IGB_DEFAULT_RX_FREE_THRESH  32
64 
65 #define IGB_DEFAULT_RX_PTHRESH      ((hw->mac.type == e1000_i354) ? 12 : 8)
66 #define IGB_DEFAULT_RX_HTHRESH      8
67 #define IGB_DEFAULT_RX_WTHRESH      ((hw->mac.type == e1000_82576) ? 1 : 4)
68 
69 #define IGB_DEFAULT_TX_PTHRESH      ((hw->mac.type == e1000_i354) ? 20 : 8)
70 #define IGB_DEFAULT_TX_HTHRESH      1
71 #define IGB_DEFAULT_TX_WTHRESH      ((hw->mac.type == e1000_82576) ? 1 : 16)
72 
73 #define IGB_HKEY_MAX_INDEX 10
74 
75 /* Bit shift and mask */
76 #define IGB_4_BIT_WIDTH  (CHAR_BIT / 2)
77 #define IGB_4_BIT_MASK   RTE_LEN2MASK(IGB_4_BIT_WIDTH, uint8_t)
78 #define IGB_8_BIT_WIDTH  CHAR_BIT
79 #define IGB_8_BIT_MASK   UINT8_MAX
80 
81 /* Additional timesync values. */
82 #define E1000_CYCLECOUNTER_MASK      0xffffffffffffffffULL
83 #define E1000_ETQF_FILTER_1588       3
84 #define IGB_82576_TSYNC_SHIFT        16
85 #define E1000_INCPERIOD_82576        (1 << E1000_TIMINCA_16NS_SHIFT)
86 #define E1000_INCVALUE_82576         (16 << IGB_82576_TSYNC_SHIFT)
87 #define E1000_TSAUXC_DISABLE_SYSTIME 0x80000000
88 
89 #define E1000_VTIVAR_MISC                0x01740
90 #define E1000_VTIVAR_MISC_MASK           0xFF
91 #define E1000_VTIVAR_VALID               0x80
92 #define E1000_VTIVAR_MISC_MAILBOX        0
93 #define E1000_VTIVAR_MISC_INTR_MASK      0x3
94 
95 /* External VLAN Enable bit mask */
96 #define E1000_CTRL_EXT_EXT_VLAN      (1 << 26)
97 
98 /* External VLAN Ether Type bit mask and shift */
99 #define E1000_VET_VET_EXT            0xFFFF0000
100 #define E1000_VET_VET_EXT_SHIFT      16
101 
102 static int  eth_igb_configure(struct rte_eth_dev *dev);
103 static int  eth_igb_start(struct rte_eth_dev *dev);
104 static void eth_igb_stop(struct rte_eth_dev *dev);
105 static int  eth_igb_dev_set_link_up(struct rte_eth_dev *dev);
106 static int  eth_igb_dev_set_link_down(struct rte_eth_dev *dev);
107 static void eth_igb_close(struct rte_eth_dev *dev);
108 static void eth_igb_promiscuous_enable(struct rte_eth_dev *dev);
109 static void eth_igb_promiscuous_disable(struct rte_eth_dev *dev);
110 static void eth_igb_allmulticast_enable(struct rte_eth_dev *dev);
111 static void eth_igb_allmulticast_disable(struct rte_eth_dev *dev);
112 static int  eth_igb_link_update(struct rte_eth_dev *dev,
113 				int wait_to_complete);
114 static void eth_igb_stats_get(struct rte_eth_dev *dev,
115 				struct rte_eth_stats *rte_stats);
116 static int eth_igb_xstats_get(struct rte_eth_dev *dev,
117 			      struct rte_eth_xstat *xstats, unsigned n);
118 static int eth_igb_xstats_get_names(struct rte_eth_dev *dev,
119 				    struct rte_eth_xstat_name *xstats_names,
120 				    unsigned limit);
121 static void eth_igb_stats_reset(struct rte_eth_dev *dev);
122 static void eth_igb_xstats_reset(struct rte_eth_dev *dev);
123 static void eth_igb_infos_get(struct rte_eth_dev *dev,
124 			      struct rte_eth_dev_info *dev_info);
125 static const uint32_t *eth_igb_supported_ptypes_get(struct rte_eth_dev *dev);
126 static void eth_igbvf_infos_get(struct rte_eth_dev *dev,
127 				struct rte_eth_dev_info *dev_info);
128 static int  eth_igb_flow_ctrl_get(struct rte_eth_dev *dev,
129 				struct rte_eth_fc_conf *fc_conf);
130 static int  eth_igb_flow_ctrl_set(struct rte_eth_dev *dev,
131 				struct rte_eth_fc_conf *fc_conf);
132 static int eth_igb_lsc_interrupt_setup(struct rte_eth_dev *dev);
133 static int eth_igb_rxq_interrupt_setup(struct rte_eth_dev *dev);
134 static int eth_igb_interrupt_get_status(struct rte_eth_dev *dev);
135 static int eth_igb_interrupt_action(struct rte_eth_dev *dev);
136 static void eth_igb_interrupt_handler(struct rte_intr_handle *handle,
137 							void *param);
138 static int  igb_hardware_init(struct e1000_hw *hw);
139 static void igb_hw_control_acquire(struct e1000_hw *hw);
140 static void igb_hw_control_release(struct e1000_hw *hw);
141 static void igb_init_manageability(struct e1000_hw *hw);
142 static void igb_release_manageability(struct e1000_hw *hw);
143 
144 static int  eth_igb_mtu_set(struct rte_eth_dev *dev, uint16_t mtu);
145 
146 static int eth_igb_vlan_filter_set(struct rte_eth_dev *dev,
147 		uint16_t vlan_id, int on);
148 static int eth_igb_vlan_tpid_set(struct rte_eth_dev *dev,
149 				 enum rte_vlan_type vlan_type,
150 				 uint16_t tpid_id);
151 static void eth_igb_vlan_offload_set(struct rte_eth_dev *dev, int mask);
152 
153 static void igb_vlan_hw_filter_enable(struct rte_eth_dev *dev);
154 static void igb_vlan_hw_filter_disable(struct rte_eth_dev *dev);
155 static void igb_vlan_hw_strip_enable(struct rte_eth_dev *dev);
156 static void igb_vlan_hw_strip_disable(struct rte_eth_dev *dev);
157 static void igb_vlan_hw_extend_enable(struct rte_eth_dev *dev);
158 static void igb_vlan_hw_extend_disable(struct rte_eth_dev *dev);
159 
160 static int eth_igb_led_on(struct rte_eth_dev *dev);
161 static int eth_igb_led_off(struct rte_eth_dev *dev);
162 
163 static void igb_intr_disable(struct e1000_hw *hw);
164 static int  igb_get_rx_buffer_size(struct e1000_hw *hw);
165 static void eth_igb_rar_set(struct rte_eth_dev *dev,
166 		struct ether_addr *mac_addr,
167 		uint32_t index, uint32_t pool);
168 static void eth_igb_rar_clear(struct rte_eth_dev *dev, uint32_t index);
169 static void eth_igb_default_mac_addr_set(struct rte_eth_dev *dev,
170 		struct ether_addr *addr);
171 
172 static void igbvf_intr_disable(struct e1000_hw *hw);
173 static int igbvf_dev_configure(struct rte_eth_dev *dev);
174 static int igbvf_dev_start(struct rte_eth_dev *dev);
175 static void igbvf_dev_stop(struct rte_eth_dev *dev);
176 static void igbvf_dev_close(struct rte_eth_dev *dev);
177 static void igbvf_promiscuous_enable(struct rte_eth_dev *dev);
178 static void igbvf_promiscuous_disable(struct rte_eth_dev *dev);
179 static void igbvf_allmulticast_enable(struct rte_eth_dev *dev);
180 static void igbvf_allmulticast_disable(struct rte_eth_dev *dev);
181 static int eth_igbvf_link_update(struct e1000_hw *hw);
182 static void eth_igbvf_stats_get(struct rte_eth_dev *dev,
183 				struct rte_eth_stats *rte_stats);
184 static int eth_igbvf_xstats_get(struct rte_eth_dev *dev,
185 				struct rte_eth_xstat *xstats, unsigned n);
186 static int eth_igbvf_xstats_get_names(struct rte_eth_dev *dev,
187 				      struct rte_eth_xstat_name *xstats_names,
188 				      unsigned limit);
189 static void eth_igbvf_stats_reset(struct rte_eth_dev *dev);
190 static int igbvf_vlan_filter_set(struct rte_eth_dev *dev,
191 		uint16_t vlan_id, int on);
192 static int igbvf_set_vfta(struct e1000_hw *hw, uint16_t vid, bool on);
193 static void igbvf_set_vfta_all(struct rte_eth_dev *dev, bool on);
194 static void igbvf_default_mac_addr_set(struct rte_eth_dev *dev,
195 		struct ether_addr *addr);
196 static int igbvf_get_reg_length(struct rte_eth_dev *dev);
197 static int igbvf_get_regs(struct rte_eth_dev *dev,
198 		struct rte_dev_reg_info *regs);
199 
200 static int eth_igb_rss_reta_update(struct rte_eth_dev *dev,
201 				   struct rte_eth_rss_reta_entry64 *reta_conf,
202 				   uint16_t reta_size);
203 static int eth_igb_rss_reta_query(struct rte_eth_dev *dev,
204 				  struct rte_eth_rss_reta_entry64 *reta_conf,
205 				  uint16_t reta_size);
206 
207 static int eth_igb_syn_filter_set(struct rte_eth_dev *dev,
208 			struct rte_eth_syn_filter *filter,
209 			bool add);
210 static int eth_igb_syn_filter_get(struct rte_eth_dev *dev,
211 			struct rte_eth_syn_filter *filter);
212 static int eth_igb_syn_filter_handle(struct rte_eth_dev *dev,
213 			enum rte_filter_op filter_op,
214 			void *arg);
215 static int igb_add_2tuple_filter(struct rte_eth_dev *dev,
216 			struct rte_eth_ntuple_filter *ntuple_filter);
217 static int igb_remove_2tuple_filter(struct rte_eth_dev *dev,
218 			struct rte_eth_ntuple_filter *ntuple_filter);
219 static int eth_igb_add_del_flex_filter(struct rte_eth_dev *dev,
220 			struct rte_eth_flex_filter *filter,
221 			bool add);
222 static int eth_igb_get_flex_filter(struct rte_eth_dev *dev,
223 			struct rte_eth_flex_filter *filter);
224 static int eth_igb_flex_filter_handle(struct rte_eth_dev *dev,
225 			enum rte_filter_op filter_op,
226 			void *arg);
227 static int igb_add_5tuple_filter_82576(struct rte_eth_dev *dev,
228 			struct rte_eth_ntuple_filter *ntuple_filter);
229 static int igb_remove_5tuple_filter_82576(struct rte_eth_dev *dev,
230 			struct rte_eth_ntuple_filter *ntuple_filter);
231 static int igb_add_del_ntuple_filter(struct rte_eth_dev *dev,
232 			struct rte_eth_ntuple_filter *filter,
233 			bool add);
234 static int igb_get_ntuple_filter(struct rte_eth_dev *dev,
235 			struct rte_eth_ntuple_filter *filter);
236 static int igb_ntuple_filter_handle(struct rte_eth_dev *dev,
237 				enum rte_filter_op filter_op,
238 				void *arg);
239 static int igb_add_del_ethertype_filter(struct rte_eth_dev *dev,
240 			struct rte_eth_ethertype_filter *filter,
241 			bool add);
242 static int igb_ethertype_filter_handle(struct rte_eth_dev *dev,
243 				enum rte_filter_op filter_op,
244 				void *arg);
245 static int igb_get_ethertype_filter(struct rte_eth_dev *dev,
246 			struct rte_eth_ethertype_filter *filter);
247 static int eth_igb_filter_ctrl(struct rte_eth_dev *dev,
248 		     enum rte_filter_type filter_type,
249 		     enum rte_filter_op filter_op,
250 		     void *arg);
251 static int eth_igb_get_reg_length(struct rte_eth_dev *dev);
252 static int eth_igb_get_regs(struct rte_eth_dev *dev,
253 		struct rte_dev_reg_info *regs);
254 static int eth_igb_get_eeprom_length(struct rte_eth_dev *dev);
255 static int eth_igb_get_eeprom(struct rte_eth_dev *dev,
256 		struct rte_dev_eeprom_info *eeprom);
257 static int eth_igb_set_eeprom(struct rte_eth_dev *dev,
258 		struct rte_dev_eeprom_info *eeprom);
259 static int eth_igb_set_mc_addr_list(struct rte_eth_dev *dev,
260 				    struct ether_addr *mc_addr_set,
261 				    uint32_t nb_mc_addr);
262 static int igb_timesync_enable(struct rte_eth_dev *dev);
263 static int igb_timesync_disable(struct rte_eth_dev *dev);
264 static int igb_timesync_read_rx_timestamp(struct rte_eth_dev *dev,
265 					  struct timespec *timestamp,
266 					  uint32_t flags);
267 static int igb_timesync_read_tx_timestamp(struct rte_eth_dev *dev,
268 					  struct timespec *timestamp);
269 static int igb_timesync_adjust_time(struct rte_eth_dev *dev, int64_t delta);
270 static int igb_timesync_read_time(struct rte_eth_dev *dev,
271 				  struct timespec *timestamp);
272 static int igb_timesync_write_time(struct rte_eth_dev *dev,
273 				   const struct timespec *timestamp);
274 static int eth_igb_rx_queue_intr_enable(struct rte_eth_dev *dev,
275 					uint16_t queue_id);
276 static int eth_igb_rx_queue_intr_disable(struct rte_eth_dev *dev,
277 					 uint16_t queue_id);
278 static void eth_igb_assign_msix_vector(struct e1000_hw *hw, int8_t direction,
279 				       uint8_t queue, uint8_t msix_vector);
280 static void eth_igb_write_ivar(struct e1000_hw *hw, uint8_t msix_vector,
281 			       uint8_t index, uint8_t offset);
282 static void eth_igb_configure_msix_intr(struct rte_eth_dev *dev);
283 static void eth_igbvf_interrupt_handler(struct rte_intr_handle *handle,
284 					void *param);
285 static void igbvf_mbx_process(struct rte_eth_dev *dev);
286 
287 /*
288  * Define VF Stats MACRO for Non "cleared on read" register
289  */
290 #define UPDATE_VF_STAT(reg, last, cur)            \
291 {                                                 \
292 	u32 latest = E1000_READ_REG(hw, reg);     \
293 	cur += (latest - last) & UINT_MAX;        \
294 	last = latest;                            \
295 }
296 
297 #define IGB_FC_PAUSE_TIME 0x0680
298 #define IGB_LINK_UPDATE_CHECK_TIMEOUT  90  /* 9s */
299 #define IGB_LINK_UPDATE_CHECK_INTERVAL 100 /* ms */
300 
301 #define IGBVF_PMD_NAME "rte_igbvf_pmd"     /* PMD name */
302 
303 static enum e1000_fc_mode igb_fc_setting = e1000_fc_full;
304 
305 /*
306  * The set of PCI devices this driver supports
307  */
308 static const struct rte_pci_id pci_id_igb_map[] = {
309 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576) },
310 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576_FIBER) },
311 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576_SERDES) },
312 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576_QUAD_COPPER) },
313 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576_QUAD_COPPER_ET2) },
314 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576_NS) },
315 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576_NS_SERDES) },
316 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576_SERDES_QUAD) },
317 
318 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82575EB_COPPER) },
319 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82575EB_FIBER_SERDES) },
320 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82575GB_QUAD_COPPER) },
321 
322 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82580_COPPER) },
323 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82580_FIBER) },
324 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82580_SERDES) },
325 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82580_SGMII) },
326 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82580_COPPER_DUAL) },
327 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82580_QUAD_FIBER) },
328 
329 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I350_COPPER) },
330 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I350_FIBER) },
331 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I350_SERDES) },
332 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I350_SGMII) },
333 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I350_DA4) },
334 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I210_COPPER) },
335 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I210_COPPER_OEM1) },
336 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I210_COPPER_IT) },
337 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I210_FIBER) },
338 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I210_SERDES) },
339 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I210_SGMII) },
340 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I211_COPPER) },
341 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I354_BACKPLANE_1GBPS) },
342 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I354_SGMII) },
343 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I354_BACKPLANE_2_5GBPS) },
344 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_DH89XXCC_SGMII) },
345 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_DH89XXCC_SERDES) },
346 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_DH89XXCC_BACKPLANE) },
347 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_DH89XXCC_SFP) },
348 	{ .vendor_id = 0, /* sentinel */ },
349 };
350 
351 /*
352  * The set of PCI devices this driver supports (for 82576&I350 VF)
353  */
354 static const struct rte_pci_id pci_id_igbvf_map[] = {
355 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576_VF) },
356 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576_VF_HV) },
357 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I350_VF) },
358 	{ RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I350_VF_HV) },
359 	{ .vendor_id = 0, /* sentinel */ },
360 };
361 
362 static const struct rte_eth_desc_lim rx_desc_lim = {
363 	.nb_max = E1000_MAX_RING_DESC,
364 	.nb_min = E1000_MIN_RING_DESC,
365 	.nb_align = IGB_RXD_ALIGN,
366 };
367 
368 static const struct rte_eth_desc_lim tx_desc_lim = {
369 	.nb_max = E1000_MAX_RING_DESC,
370 	.nb_min = E1000_MIN_RING_DESC,
371 	.nb_align = IGB_RXD_ALIGN,
372 };
373 
374 static const struct eth_dev_ops eth_igb_ops = {
375 	.dev_configure        = eth_igb_configure,
376 	.dev_start            = eth_igb_start,
377 	.dev_stop             = eth_igb_stop,
378 	.dev_set_link_up      = eth_igb_dev_set_link_up,
379 	.dev_set_link_down    = eth_igb_dev_set_link_down,
380 	.dev_close            = eth_igb_close,
381 	.promiscuous_enable   = eth_igb_promiscuous_enable,
382 	.promiscuous_disable  = eth_igb_promiscuous_disable,
383 	.allmulticast_enable  = eth_igb_allmulticast_enable,
384 	.allmulticast_disable = eth_igb_allmulticast_disable,
385 	.link_update          = eth_igb_link_update,
386 	.stats_get            = eth_igb_stats_get,
387 	.xstats_get           = eth_igb_xstats_get,
388 	.xstats_get_names     = eth_igb_xstats_get_names,
389 	.stats_reset          = eth_igb_stats_reset,
390 	.xstats_reset         = eth_igb_xstats_reset,
391 	.dev_infos_get        = eth_igb_infos_get,
392 	.dev_supported_ptypes_get = eth_igb_supported_ptypes_get,
393 	.mtu_set              = eth_igb_mtu_set,
394 	.vlan_filter_set      = eth_igb_vlan_filter_set,
395 	.vlan_tpid_set        = eth_igb_vlan_tpid_set,
396 	.vlan_offload_set     = eth_igb_vlan_offload_set,
397 	.rx_queue_setup       = eth_igb_rx_queue_setup,
398 	.rx_queue_intr_enable = eth_igb_rx_queue_intr_enable,
399 	.rx_queue_intr_disable = eth_igb_rx_queue_intr_disable,
400 	.rx_queue_release     = eth_igb_rx_queue_release,
401 	.rx_queue_count       = eth_igb_rx_queue_count,
402 	.rx_descriptor_done   = eth_igb_rx_descriptor_done,
403 	.tx_queue_setup       = eth_igb_tx_queue_setup,
404 	.tx_queue_release     = eth_igb_tx_queue_release,
405 	.dev_led_on           = eth_igb_led_on,
406 	.dev_led_off          = eth_igb_led_off,
407 	.flow_ctrl_get        = eth_igb_flow_ctrl_get,
408 	.flow_ctrl_set        = eth_igb_flow_ctrl_set,
409 	.mac_addr_add         = eth_igb_rar_set,
410 	.mac_addr_remove      = eth_igb_rar_clear,
411 	.mac_addr_set         = eth_igb_default_mac_addr_set,
412 	.reta_update          = eth_igb_rss_reta_update,
413 	.reta_query           = eth_igb_rss_reta_query,
414 	.rss_hash_update      = eth_igb_rss_hash_update,
415 	.rss_hash_conf_get    = eth_igb_rss_hash_conf_get,
416 	.filter_ctrl          = eth_igb_filter_ctrl,
417 	.set_mc_addr_list     = eth_igb_set_mc_addr_list,
418 	.rxq_info_get         = igb_rxq_info_get,
419 	.txq_info_get         = igb_txq_info_get,
420 	.timesync_enable      = igb_timesync_enable,
421 	.timesync_disable     = igb_timesync_disable,
422 	.timesync_read_rx_timestamp = igb_timesync_read_rx_timestamp,
423 	.timesync_read_tx_timestamp = igb_timesync_read_tx_timestamp,
424 	.get_reg              = eth_igb_get_regs,
425 	.get_eeprom_length    = eth_igb_get_eeprom_length,
426 	.get_eeprom           = eth_igb_get_eeprom,
427 	.set_eeprom           = eth_igb_set_eeprom,
428 	.timesync_adjust_time = igb_timesync_adjust_time,
429 	.timesync_read_time   = igb_timesync_read_time,
430 	.timesync_write_time  = igb_timesync_write_time,
431 };
432 
433 /*
434  * dev_ops for virtual function, bare necessities for basic vf
435  * operation have been implemented
436  */
437 static const struct eth_dev_ops igbvf_eth_dev_ops = {
438 	.dev_configure        = igbvf_dev_configure,
439 	.dev_start            = igbvf_dev_start,
440 	.dev_stop             = igbvf_dev_stop,
441 	.dev_close            = igbvf_dev_close,
442 	.promiscuous_enable   = igbvf_promiscuous_enable,
443 	.promiscuous_disable  = igbvf_promiscuous_disable,
444 	.allmulticast_enable  = igbvf_allmulticast_enable,
445 	.allmulticast_disable = igbvf_allmulticast_disable,
446 	.link_update          = eth_igb_link_update,
447 	.stats_get            = eth_igbvf_stats_get,
448 	.xstats_get           = eth_igbvf_xstats_get,
449 	.xstats_get_names     = eth_igbvf_xstats_get_names,
450 	.stats_reset          = eth_igbvf_stats_reset,
451 	.xstats_reset         = eth_igbvf_stats_reset,
452 	.vlan_filter_set      = igbvf_vlan_filter_set,
453 	.dev_infos_get        = eth_igbvf_infos_get,
454 	.dev_supported_ptypes_get = eth_igb_supported_ptypes_get,
455 	.rx_queue_setup       = eth_igb_rx_queue_setup,
456 	.rx_queue_release     = eth_igb_rx_queue_release,
457 	.tx_queue_setup       = eth_igb_tx_queue_setup,
458 	.tx_queue_release     = eth_igb_tx_queue_release,
459 	.set_mc_addr_list     = eth_igb_set_mc_addr_list,
460 	.rxq_info_get         = igb_rxq_info_get,
461 	.txq_info_get         = igb_txq_info_get,
462 	.mac_addr_set         = igbvf_default_mac_addr_set,
463 	.get_reg              = igbvf_get_regs,
464 };
465 
466 /* store statistics names and its offset in stats structure */
467 struct rte_igb_xstats_name_off {
468 	char name[RTE_ETH_XSTATS_NAME_SIZE];
469 	unsigned offset;
470 };
471 
472 static const struct rte_igb_xstats_name_off rte_igb_stats_strings[] = {
473 	{"rx_crc_errors", offsetof(struct e1000_hw_stats, crcerrs)},
474 	{"rx_align_errors", offsetof(struct e1000_hw_stats, algnerrc)},
475 	{"rx_symbol_errors", offsetof(struct e1000_hw_stats, symerrs)},
476 	{"rx_missed_packets", offsetof(struct e1000_hw_stats, mpc)},
477 	{"tx_single_collision_packets", offsetof(struct e1000_hw_stats, scc)},
478 	{"tx_multiple_collision_packets", offsetof(struct e1000_hw_stats, mcc)},
479 	{"tx_excessive_collision_packets", offsetof(struct e1000_hw_stats,
480 		ecol)},
481 	{"tx_late_collisions", offsetof(struct e1000_hw_stats, latecol)},
482 	{"tx_total_collisions", offsetof(struct e1000_hw_stats, colc)},
483 	{"tx_deferred_packets", offsetof(struct e1000_hw_stats, dc)},
484 	{"tx_no_carrier_sense_packets", offsetof(struct e1000_hw_stats, tncrs)},
485 	{"rx_carrier_ext_errors", offsetof(struct e1000_hw_stats, cexterr)},
486 	{"rx_length_errors", offsetof(struct e1000_hw_stats, rlec)},
487 	{"rx_xon_packets", offsetof(struct e1000_hw_stats, xonrxc)},
488 	{"tx_xon_packets", offsetof(struct e1000_hw_stats, xontxc)},
489 	{"rx_xoff_packets", offsetof(struct e1000_hw_stats, xoffrxc)},
490 	{"tx_xoff_packets", offsetof(struct e1000_hw_stats, xofftxc)},
491 	{"rx_flow_control_unsupported_packets", offsetof(struct e1000_hw_stats,
492 		fcruc)},
493 	{"rx_size_64_packets", offsetof(struct e1000_hw_stats, prc64)},
494 	{"rx_size_65_to_127_packets", offsetof(struct e1000_hw_stats, prc127)},
495 	{"rx_size_128_to_255_packets", offsetof(struct e1000_hw_stats, prc255)},
496 	{"rx_size_256_to_511_packets", offsetof(struct e1000_hw_stats, prc511)},
497 	{"rx_size_512_to_1023_packets", offsetof(struct e1000_hw_stats,
498 		prc1023)},
499 	{"rx_size_1024_to_max_packets", offsetof(struct e1000_hw_stats,
500 		prc1522)},
501 	{"rx_broadcast_packets", offsetof(struct e1000_hw_stats, bprc)},
502 	{"rx_multicast_packets", offsetof(struct e1000_hw_stats, mprc)},
503 	{"rx_undersize_errors", offsetof(struct e1000_hw_stats, ruc)},
504 	{"rx_fragment_errors", offsetof(struct e1000_hw_stats, rfc)},
505 	{"rx_oversize_errors", offsetof(struct e1000_hw_stats, roc)},
506 	{"rx_jabber_errors", offsetof(struct e1000_hw_stats, rjc)},
507 	{"rx_management_packets", offsetof(struct e1000_hw_stats, mgprc)},
508 	{"rx_management_dropped", offsetof(struct e1000_hw_stats, mgpdc)},
509 	{"tx_management_packets", offsetof(struct e1000_hw_stats, mgptc)},
510 	{"rx_total_packets", offsetof(struct e1000_hw_stats, tpr)},
511 	{"tx_total_packets", offsetof(struct e1000_hw_stats, tpt)},
512 	{"rx_total_bytes", offsetof(struct e1000_hw_stats, tor)},
513 	{"tx_total_bytes", offsetof(struct e1000_hw_stats, tot)},
514 	{"tx_size_64_packets", offsetof(struct e1000_hw_stats, ptc64)},
515 	{"tx_size_65_to_127_packets", offsetof(struct e1000_hw_stats, ptc127)},
516 	{"tx_size_128_to_255_packets", offsetof(struct e1000_hw_stats, ptc255)},
517 	{"tx_size_256_to_511_packets", offsetof(struct e1000_hw_stats, ptc511)},
518 	{"tx_size_512_to_1023_packets", offsetof(struct e1000_hw_stats,
519 		ptc1023)},
520 	{"tx_size_1023_to_max_packets", offsetof(struct e1000_hw_stats,
521 		ptc1522)},
522 	{"tx_multicast_packets", offsetof(struct e1000_hw_stats, mptc)},
523 	{"tx_broadcast_packets", offsetof(struct e1000_hw_stats, bptc)},
524 	{"tx_tso_packets", offsetof(struct e1000_hw_stats, tsctc)},
525 	{"tx_tso_errors", offsetof(struct e1000_hw_stats, tsctfc)},
526 	{"rx_sent_to_host_packets", offsetof(struct e1000_hw_stats, rpthc)},
527 	{"tx_sent_by_host_packets", offsetof(struct e1000_hw_stats, hgptc)},
528 	{"rx_code_violation_packets", offsetof(struct e1000_hw_stats, scvpc)},
529 
530 	{"interrupt_assert_count", offsetof(struct e1000_hw_stats, iac)},
531 };
532 
533 #define IGB_NB_XSTATS (sizeof(rte_igb_stats_strings) / \
534 		sizeof(rte_igb_stats_strings[0]))
535 
536 static const struct rte_igb_xstats_name_off rte_igbvf_stats_strings[] = {
537 	{"rx_multicast_packets", offsetof(struct e1000_vf_stats, mprc)},
538 	{"rx_good_loopback_packets", offsetof(struct e1000_vf_stats, gprlbc)},
539 	{"tx_good_loopback_packets", offsetof(struct e1000_vf_stats, gptlbc)},
540 	{"rx_good_loopback_bytes", offsetof(struct e1000_vf_stats, gorlbc)},
541 	{"tx_good_loopback_bytes", offsetof(struct e1000_vf_stats, gotlbc)},
542 };
543 
544 #define IGBVF_NB_XSTATS (sizeof(rte_igbvf_stats_strings) / \
545 		sizeof(rte_igbvf_stats_strings[0]))
546 
547 /**
548  * Atomically reads the link status information from global
549  * structure rte_eth_dev.
550  *
551  * @param dev
552  *   - Pointer to the structure rte_eth_dev to read from.
553  *   - Pointer to the buffer to be saved with the link status.
554  *
555  * @return
556  *   - On success, zero.
557  *   - On failure, negative value.
558  */
559 static inline int
560 rte_igb_dev_atomic_read_link_status(struct rte_eth_dev *dev,
561 				struct rte_eth_link *link)
562 {
563 	struct rte_eth_link *dst = link;
564 	struct rte_eth_link *src = &(dev->data->dev_link);
565 
566 	if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
567 					*(uint64_t *)src) == 0)
568 		return -1;
569 
570 	return 0;
571 }
572 
573 /**
574  * Atomically writes the link status information into global
575  * structure rte_eth_dev.
576  *
577  * @param dev
578  *   - Pointer to the structure rte_eth_dev to read from.
579  *   - Pointer to the buffer to be saved with the link status.
580  *
581  * @return
582  *   - On success, zero.
583  *   - On failure, negative value.
584  */
585 static inline int
586 rte_igb_dev_atomic_write_link_status(struct rte_eth_dev *dev,
587 				struct rte_eth_link *link)
588 {
589 	struct rte_eth_link *dst = &(dev->data->dev_link);
590 	struct rte_eth_link *src = link;
591 
592 	if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
593 					*(uint64_t *)src) == 0)
594 		return -1;
595 
596 	return 0;
597 }
598 
599 static inline void
600 igb_intr_enable(struct rte_eth_dev *dev)
601 {
602 	struct e1000_interrupt *intr =
603 		E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
604 	struct e1000_hw *hw =
605 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
606 
607 	E1000_WRITE_REG(hw, E1000_IMS, intr->mask);
608 	E1000_WRITE_FLUSH(hw);
609 }
610 
611 static void
612 igb_intr_disable(struct e1000_hw *hw)
613 {
614 	E1000_WRITE_REG(hw, E1000_IMC, ~0);
615 	E1000_WRITE_FLUSH(hw);
616 }
617 
618 static inline void
619 igbvf_intr_enable(struct rte_eth_dev *dev)
620 {
621 	struct e1000_hw *hw =
622 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
623 
624 	/* only for mailbox */
625 	E1000_WRITE_REG(hw, E1000_EIAM, 1 << E1000_VTIVAR_MISC_MAILBOX);
626 	E1000_WRITE_REG(hw, E1000_EIAC, 1 << E1000_VTIVAR_MISC_MAILBOX);
627 	E1000_WRITE_REG(hw, E1000_EIMS, 1 << E1000_VTIVAR_MISC_MAILBOX);
628 	E1000_WRITE_FLUSH(hw);
629 }
630 
631 /* only for mailbox now. If RX/TX needed, should extend this function.  */
632 static void
633 igbvf_set_ivar_map(struct e1000_hw *hw, uint8_t msix_vector)
634 {
635 	uint32_t tmp = 0;
636 
637 	/* mailbox */
638 	tmp |= (msix_vector & E1000_VTIVAR_MISC_INTR_MASK);
639 	tmp |= E1000_VTIVAR_VALID;
640 	E1000_WRITE_REG(hw, E1000_VTIVAR_MISC, tmp);
641 }
642 
643 static void
644 eth_igbvf_configure_msix_intr(struct rte_eth_dev *dev)
645 {
646 	struct e1000_hw *hw =
647 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
648 
649 	/* Configure VF other cause ivar */
650 	igbvf_set_ivar_map(hw, E1000_VTIVAR_MISC_MAILBOX);
651 }
652 
653 static inline int32_t
654 igb_pf_reset_hw(struct e1000_hw *hw)
655 {
656 	uint32_t ctrl_ext;
657 	int32_t status;
658 
659 	status = e1000_reset_hw(hw);
660 
661 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
662 	/* Set PF Reset Done bit so PF/VF Mail Ops can work */
663 	ctrl_ext |= E1000_CTRL_EXT_PFRSTD;
664 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
665 	E1000_WRITE_FLUSH(hw);
666 
667 	return status;
668 }
669 
670 static void
671 igb_identify_hardware(struct rte_eth_dev *dev)
672 {
673 	struct e1000_hw *hw =
674 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
675 
676 	hw->vendor_id = dev->pci_dev->id.vendor_id;
677 	hw->device_id = dev->pci_dev->id.device_id;
678 	hw->subsystem_vendor_id = dev->pci_dev->id.subsystem_vendor_id;
679 	hw->subsystem_device_id = dev->pci_dev->id.subsystem_device_id;
680 
681 	e1000_set_mac_type(hw);
682 
683 	/* need to check if it is a vf device below */
684 }
685 
686 static int
687 igb_reset_swfw_lock(struct e1000_hw *hw)
688 {
689 	int ret_val;
690 
691 	/*
692 	 * Do mac ops initialization manually here, since we will need
693 	 * some function pointers set by this call.
694 	 */
695 	ret_val = e1000_init_mac_params(hw);
696 	if (ret_val)
697 		return ret_val;
698 
699 	/*
700 	 * SMBI lock should not fail in this early stage. If this is the case,
701 	 * it is due to an improper exit of the application.
702 	 * So force the release of the faulty lock.
703 	 */
704 	if (e1000_get_hw_semaphore_generic(hw) < 0) {
705 		PMD_DRV_LOG(DEBUG, "SMBI lock released");
706 	}
707 	e1000_put_hw_semaphore_generic(hw);
708 
709 	if (hw->mac.ops.acquire_swfw_sync != NULL) {
710 		uint16_t mask;
711 
712 		/*
713 		 * Phy lock should not fail in this early stage. If this is the case,
714 		 * it is due to an improper exit of the application.
715 		 * So force the release of the faulty lock.
716 		 */
717 		mask = E1000_SWFW_PHY0_SM << hw->bus.func;
718 		if (hw->bus.func > E1000_FUNC_1)
719 			mask <<= 2;
720 		if (hw->mac.ops.acquire_swfw_sync(hw, mask) < 0) {
721 			PMD_DRV_LOG(DEBUG, "SWFW phy%d lock released",
722 				    hw->bus.func);
723 		}
724 		hw->mac.ops.release_swfw_sync(hw, mask);
725 
726 		/*
727 		 * This one is more tricky since it is common to all ports; but
728 		 * swfw_sync retries last long enough (1s) to be almost sure that if
729 		 * lock can not be taken it is due to an improper lock of the
730 		 * semaphore.
731 		 */
732 		mask = E1000_SWFW_EEP_SM;
733 		if (hw->mac.ops.acquire_swfw_sync(hw, mask) < 0) {
734 			PMD_DRV_LOG(DEBUG, "SWFW common locks released");
735 		}
736 		hw->mac.ops.release_swfw_sync(hw, mask);
737 	}
738 
739 	return E1000_SUCCESS;
740 }
741 
742 static int
743 eth_igb_dev_init(struct rte_eth_dev *eth_dev)
744 {
745 	int error = 0;
746 	struct rte_pci_device *pci_dev;
747 	struct e1000_hw *hw =
748 		E1000_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
749 	struct e1000_vfta * shadow_vfta =
750 		E1000_DEV_PRIVATE_TO_VFTA(eth_dev->data->dev_private);
751 	struct e1000_filter_info *filter_info =
752 		E1000_DEV_PRIVATE_TO_FILTER_INFO(eth_dev->data->dev_private);
753 	struct e1000_adapter *adapter =
754 		E1000_DEV_PRIVATE(eth_dev->data->dev_private);
755 
756 	uint32_t ctrl_ext;
757 
758 	pci_dev = eth_dev->pci_dev;
759 
760 	eth_dev->dev_ops = &eth_igb_ops;
761 	eth_dev->rx_pkt_burst = &eth_igb_recv_pkts;
762 	eth_dev->tx_pkt_burst = &eth_igb_xmit_pkts;
763 
764 	/* for secondary processes, we don't initialise any further as primary
765 	 * has already done this work. Only check we don't need a different
766 	 * RX function */
767 	if (rte_eal_process_type() != RTE_PROC_PRIMARY){
768 		if (eth_dev->data->scattered_rx)
769 			eth_dev->rx_pkt_burst = &eth_igb_recv_scattered_pkts;
770 		return 0;
771 	}
772 
773 	rte_eth_copy_pci_info(eth_dev, pci_dev);
774 
775 	hw->hw_addr= (void *)pci_dev->mem_resource[0].addr;
776 
777 	igb_identify_hardware(eth_dev);
778 	if (e1000_setup_init_funcs(hw, FALSE) != E1000_SUCCESS) {
779 		error = -EIO;
780 		goto err_late;
781 	}
782 
783 	e1000_get_bus_info(hw);
784 
785 	/* Reset any pending lock */
786 	if (igb_reset_swfw_lock(hw) != E1000_SUCCESS) {
787 		error = -EIO;
788 		goto err_late;
789 	}
790 
791 	/* Finish initialization */
792 	if (e1000_setup_init_funcs(hw, TRUE) != E1000_SUCCESS) {
793 		error = -EIO;
794 		goto err_late;
795 	}
796 
797 	hw->mac.autoneg = 1;
798 	hw->phy.autoneg_wait_to_complete = 0;
799 	hw->phy.autoneg_advertised = E1000_ALL_SPEED_DUPLEX;
800 
801 	/* Copper options */
802 	if (hw->phy.media_type == e1000_media_type_copper) {
803 		hw->phy.mdix = 0; /* AUTO_ALL_MODES */
804 		hw->phy.disable_polarity_correction = 0;
805 		hw->phy.ms_type = e1000_ms_hw_default;
806 	}
807 
808 	/*
809 	 * Start from a known state, this is important in reading the nvm
810 	 * and mac from that.
811 	 */
812 	igb_pf_reset_hw(hw);
813 
814 	/* Make sure we have a good EEPROM before we read from it */
815 	if (e1000_validate_nvm_checksum(hw) < 0) {
816 		/*
817 		 * Some PCI-E parts fail the first check due to
818 		 * the link being in sleep state, call it again,
819 		 * if it fails a second time its a real issue.
820 		 */
821 		if (e1000_validate_nvm_checksum(hw) < 0) {
822 			PMD_INIT_LOG(ERR, "EEPROM checksum invalid");
823 			error = -EIO;
824 			goto err_late;
825 		}
826 	}
827 
828 	/* Read the permanent MAC address out of the EEPROM */
829 	if (e1000_read_mac_addr(hw) != 0) {
830 		PMD_INIT_LOG(ERR, "EEPROM error while reading MAC address");
831 		error = -EIO;
832 		goto err_late;
833 	}
834 
835 	/* Allocate memory for storing MAC addresses */
836 	eth_dev->data->mac_addrs = rte_zmalloc("e1000",
837 		ETHER_ADDR_LEN * hw->mac.rar_entry_count, 0);
838 	if (eth_dev->data->mac_addrs == NULL) {
839 		PMD_INIT_LOG(ERR, "Failed to allocate %d bytes needed to "
840 						"store MAC addresses",
841 				ETHER_ADDR_LEN * hw->mac.rar_entry_count);
842 		error = -ENOMEM;
843 		goto err_late;
844 	}
845 
846 	/* Copy the permanent MAC address */
847 	ether_addr_copy((struct ether_addr *)hw->mac.addr, &eth_dev->data->mac_addrs[0]);
848 
849 	/* initialize the vfta */
850 	memset(shadow_vfta, 0, sizeof(*shadow_vfta));
851 
852 	/* Now initialize the hardware */
853 	if (igb_hardware_init(hw) != 0) {
854 		PMD_INIT_LOG(ERR, "Hardware initialization failed");
855 		rte_free(eth_dev->data->mac_addrs);
856 		eth_dev->data->mac_addrs = NULL;
857 		error = -ENODEV;
858 		goto err_late;
859 	}
860 	hw->mac.get_link_status = 1;
861 	adapter->stopped = 0;
862 
863 	/* Indicate SOL/IDER usage */
864 	if (e1000_check_reset_block(hw) < 0) {
865 		PMD_INIT_LOG(ERR, "PHY reset is blocked due to"
866 					"SOL/IDER session");
867 	}
868 
869 	/* initialize PF if max_vfs not zero */
870 	igb_pf_host_init(eth_dev);
871 
872 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
873 	/* Set PF Reset Done bit so PF/VF Mail Ops can work */
874 	ctrl_ext |= E1000_CTRL_EXT_PFRSTD;
875 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
876 	E1000_WRITE_FLUSH(hw);
877 
878 	PMD_INIT_LOG(DEBUG, "port_id %d vendorID=0x%x deviceID=0x%x",
879 		     eth_dev->data->port_id, pci_dev->id.vendor_id,
880 		     pci_dev->id.device_id);
881 
882 	rte_intr_callback_register(&pci_dev->intr_handle,
883 				   eth_igb_interrupt_handler,
884 				   (void *)eth_dev);
885 
886 	/* enable uio/vfio intr/eventfd mapping */
887 	rte_intr_enable(&pci_dev->intr_handle);
888 
889 	/* enable support intr */
890 	igb_intr_enable(eth_dev);
891 
892 	TAILQ_INIT(&filter_info->flex_list);
893 	filter_info->flex_mask = 0;
894 	TAILQ_INIT(&filter_info->twotuple_list);
895 	filter_info->twotuple_mask = 0;
896 	TAILQ_INIT(&filter_info->fivetuple_list);
897 	filter_info->fivetuple_mask = 0;
898 
899 	return 0;
900 
901 err_late:
902 	igb_hw_control_release(hw);
903 
904 	return error;
905 }
906 
907 static int
908 eth_igb_dev_uninit(struct rte_eth_dev *eth_dev)
909 {
910 	struct rte_pci_device *pci_dev;
911 	struct e1000_hw *hw;
912 	struct e1000_adapter *adapter =
913 		E1000_DEV_PRIVATE(eth_dev->data->dev_private);
914 
915 	PMD_INIT_FUNC_TRACE();
916 
917 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
918 		return -EPERM;
919 
920 	hw = E1000_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
921 	pci_dev = eth_dev->pci_dev;
922 
923 	if (adapter->stopped == 0)
924 		eth_igb_close(eth_dev);
925 
926 	eth_dev->dev_ops = NULL;
927 	eth_dev->rx_pkt_burst = NULL;
928 	eth_dev->tx_pkt_burst = NULL;
929 
930 	/* Reset any pending lock */
931 	igb_reset_swfw_lock(hw);
932 
933 	rte_free(eth_dev->data->mac_addrs);
934 	eth_dev->data->mac_addrs = NULL;
935 
936 	/* uninitialize PF if max_vfs not zero */
937 	igb_pf_host_uninit(eth_dev);
938 
939 	/* disable uio intr before callback unregister */
940 	rte_intr_disable(&(pci_dev->intr_handle));
941 	rte_intr_callback_unregister(&(pci_dev->intr_handle),
942 		eth_igb_interrupt_handler, (void *)eth_dev);
943 
944 	return 0;
945 }
946 
947 /*
948  * Virtual Function device init
949  */
950 static int
951 eth_igbvf_dev_init(struct rte_eth_dev *eth_dev)
952 {
953 	struct rte_pci_device *pci_dev;
954 	struct e1000_adapter *adapter =
955 		E1000_DEV_PRIVATE(eth_dev->data->dev_private);
956 	struct e1000_hw *hw =
957 		E1000_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
958 	int diag;
959 	struct ether_addr *perm_addr = (struct ether_addr *)hw->mac.perm_addr;
960 
961 	PMD_INIT_FUNC_TRACE();
962 
963 	eth_dev->dev_ops = &igbvf_eth_dev_ops;
964 	eth_dev->rx_pkt_burst = &eth_igb_recv_pkts;
965 	eth_dev->tx_pkt_burst = &eth_igb_xmit_pkts;
966 
967 	/* for secondary processes, we don't initialise any further as primary
968 	 * has already done this work. Only check we don't need a different
969 	 * RX function */
970 	if (rte_eal_process_type() != RTE_PROC_PRIMARY){
971 		if (eth_dev->data->scattered_rx)
972 			eth_dev->rx_pkt_burst = &eth_igb_recv_scattered_pkts;
973 		return 0;
974 	}
975 
976 	pci_dev = eth_dev->pci_dev;
977 
978 	rte_eth_copy_pci_info(eth_dev, pci_dev);
979 
980 	hw->device_id = pci_dev->id.device_id;
981 	hw->vendor_id = pci_dev->id.vendor_id;
982 	hw->hw_addr = (void *)pci_dev->mem_resource[0].addr;
983 	adapter->stopped = 0;
984 
985 	/* Initialize the shared code (base driver) */
986 	diag = e1000_setup_init_funcs(hw, TRUE);
987 	if (diag != 0) {
988 		PMD_INIT_LOG(ERR, "Shared code init failed for igbvf: %d",
989 			diag);
990 		return -EIO;
991 	}
992 
993 	/* init_mailbox_params */
994 	hw->mbx.ops.init_params(hw);
995 
996 	/* Disable the interrupts for VF */
997 	igbvf_intr_disable(hw);
998 
999 	diag = hw->mac.ops.reset_hw(hw);
1000 
1001 	/* Allocate memory for storing MAC addresses */
1002 	eth_dev->data->mac_addrs = rte_zmalloc("igbvf", ETHER_ADDR_LEN *
1003 		hw->mac.rar_entry_count, 0);
1004 	if (eth_dev->data->mac_addrs == NULL) {
1005 		PMD_INIT_LOG(ERR,
1006 			"Failed to allocate %d bytes needed to store MAC "
1007 			"addresses",
1008 			ETHER_ADDR_LEN * hw->mac.rar_entry_count);
1009 		return -ENOMEM;
1010 	}
1011 
1012 	/* Generate a random MAC address, if none was assigned by PF. */
1013 	if (is_zero_ether_addr(perm_addr)) {
1014 		eth_random_addr(perm_addr->addr_bytes);
1015 		diag = e1000_rar_set(hw, perm_addr->addr_bytes, 0);
1016 		if (diag) {
1017 			rte_free(eth_dev->data->mac_addrs);
1018 			eth_dev->data->mac_addrs = NULL;
1019 			return diag;
1020 		}
1021 		PMD_INIT_LOG(INFO, "\tVF MAC address not assigned by Host PF");
1022 		PMD_INIT_LOG(INFO, "\tAssign randomly generated MAC address "
1023 			     "%02x:%02x:%02x:%02x:%02x:%02x",
1024 			     perm_addr->addr_bytes[0],
1025 			     perm_addr->addr_bytes[1],
1026 			     perm_addr->addr_bytes[2],
1027 			     perm_addr->addr_bytes[3],
1028 			     perm_addr->addr_bytes[4],
1029 			     perm_addr->addr_bytes[5]);
1030 	}
1031 
1032 	/* Copy the permanent MAC address */
1033 	ether_addr_copy((struct ether_addr *) hw->mac.perm_addr,
1034 			&eth_dev->data->mac_addrs[0]);
1035 
1036 	PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x "
1037 		     "mac.type=%s",
1038 		     eth_dev->data->port_id, pci_dev->id.vendor_id,
1039 		     pci_dev->id.device_id, "igb_mac_82576_vf");
1040 
1041 	rte_intr_callback_register(&pci_dev->intr_handle,
1042 				   eth_igbvf_interrupt_handler,
1043 				   (void *)eth_dev);
1044 
1045 	return 0;
1046 }
1047 
1048 static int
1049 eth_igbvf_dev_uninit(struct rte_eth_dev *eth_dev)
1050 {
1051 	struct e1000_adapter *adapter =
1052 		E1000_DEV_PRIVATE(eth_dev->data->dev_private);
1053 	struct rte_pci_device *pci_dev = eth_dev->pci_dev;
1054 
1055 	PMD_INIT_FUNC_TRACE();
1056 
1057 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1058 		return -EPERM;
1059 
1060 	if (adapter->stopped == 0)
1061 		igbvf_dev_close(eth_dev);
1062 
1063 	eth_dev->dev_ops = NULL;
1064 	eth_dev->rx_pkt_burst = NULL;
1065 	eth_dev->tx_pkt_burst = NULL;
1066 
1067 	rte_free(eth_dev->data->mac_addrs);
1068 	eth_dev->data->mac_addrs = NULL;
1069 
1070 	/* disable uio intr before callback unregister */
1071 	rte_intr_disable(&pci_dev->intr_handle);
1072 	rte_intr_callback_unregister(&pci_dev->intr_handle,
1073 				     eth_igbvf_interrupt_handler,
1074 				     (void *)eth_dev);
1075 
1076 	return 0;
1077 }
1078 
1079 static struct eth_driver rte_igb_pmd = {
1080 	.pci_drv = {
1081 		.id_table = pci_id_igb_map,
1082 		.drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC |
1083 			RTE_PCI_DRV_DETACHABLE,
1084 		.probe = rte_eth_dev_pci_probe,
1085 		.remove = rte_eth_dev_pci_remove,
1086 	},
1087 	.eth_dev_init = eth_igb_dev_init,
1088 	.eth_dev_uninit = eth_igb_dev_uninit,
1089 	.dev_private_size = sizeof(struct e1000_adapter),
1090 };
1091 
1092 /*
1093  * virtual function driver struct
1094  */
1095 static struct eth_driver rte_igbvf_pmd = {
1096 	.pci_drv = {
1097 		.id_table = pci_id_igbvf_map,
1098 		.drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_DETACHABLE,
1099 		.probe = rte_eth_dev_pci_probe,
1100 		.remove = rte_eth_dev_pci_remove,
1101 	},
1102 	.eth_dev_init = eth_igbvf_dev_init,
1103 	.eth_dev_uninit = eth_igbvf_dev_uninit,
1104 	.dev_private_size = sizeof(struct e1000_adapter),
1105 };
1106 
1107 static void
1108 igb_vmdq_vlan_hw_filter_enable(struct rte_eth_dev *dev)
1109 {
1110 	struct e1000_hw *hw =
1111 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1112 	/* RCTL: enable VLAN filter since VMDq always use VLAN filter */
1113 	uint32_t rctl = E1000_READ_REG(hw, E1000_RCTL);
1114 	rctl |= E1000_RCTL_VFE;
1115 	E1000_WRITE_REG(hw, E1000_RCTL, rctl);
1116 }
1117 
1118 static int
1119 igb_check_mq_mode(struct rte_eth_dev *dev)
1120 {
1121 	enum rte_eth_rx_mq_mode rx_mq_mode = dev->data->dev_conf.rxmode.mq_mode;
1122 	enum rte_eth_tx_mq_mode tx_mq_mode = dev->data->dev_conf.txmode.mq_mode;
1123 	uint16_t nb_rx_q = dev->data->nb_rx_queues;
1124 	uint16_t nb_tx_q = dev->data->nb_rx_queues;
1125 
1126 	if ((rx_mq_mode & ETH_MQ_RX_DCB_FLAG) ||
1127 	    tx_mq_mode == ETH_MQ_TX_DCB ||
1128 	    tx_mq_mode == ETH_MQ_TX_VMDQ_DCB) {
1129 		PMD_INIT_LOG(ERR, "DCB mode is not supported.");
1130 		return -EINVAL;
1131 	}
1132 	if (RTE_ETH_DEV_SRIOV(dev).active != 0) {
1133 		/* Check multi-queue mode.
1134 		 * To no break software we accept ETH_MQ_RX_NONE as this might
1135 		 * be used to turn off VLAN filter.
1136 		 */
1137 
1138 		if (rx_mq_mode == ETH_MQ_RX_NONE ||
1139 		    rx_mq_mode == ETH_MQ_RX_VMDQ_ONLY) {
1140 			dev->data->dev_conf.rxmode.mq_mode = ETH_MQ_RX_VMDQ_ONLY;
1141 			RTE_ETH_DEV_SRIOV(dev).nb_q_per_pool = 1;
1142 		} else {
1143 			/* Only support one queue on VFs.
1144 			 * RSS together with SRIOV is not supported.
1145 			 */
1146 			PMD_INIT_LOG(ERR, "SRIOV is active,"
1147 					" wrong mq_mode rx %d.",
1148 					rx_mq_mode);
1149 			return -EINVAL;
1150 		}
1151 		/* TX mode is not used here, so mode might be ignored.*/
1152 		if (tx_mq_mode != ETH_MQ_TX_VMDQ_ONLY) {
1153 			/* SRIOV only works in VMDq enable mode */
1154 			PMD_INIT_LOG(WARNING, "SRIOV is active,"
1155 					" TX mode %d is not supported. "
1156 					" Driver will behave as %d mode.",
1157 					tx_mq_mode, ETH_MQ_TX_VMDQ_ONLY);
1158 		}
1159 
1160 		/* check valid queue number */
1161 		if ((nb_rx_q > 1) || (nb_tx_q > 1)) {
1162 			PMD_INIT_LOG(ERR, "SRIOV is active,"
1163 					" only support one queue on VFs.");
1164 			return -EINVAL;
1165 		}
1166 	} else {
1167 		/* To no break software that set invalid mode, only display
1168 		 * warning if invalid mode is used.
1169 		 */
1170 		if (rx_mq_mode != ETH_MQ_RX_NONE &&
1171 		    rx_mq_mode != ETH_MQ_RX_VMDQ_ONLY &&
1172 		    rx_mq_mode != ETH_MQ_RX_RSS) {
1173 			/* RSS together with VMDq not supported*/
1174 			PMD_INIT_LOG(ERR, "RX mode %d is not supported.",
1175 				     rx_mq_mode);
1176 			return -EINVAL;
1177 		}
1178 
1179 		if (tx_mq_mode != ETH_MQ_TX_NONE &&
1180 		    tx_mq_mode != ETH_MQ_TX_VMDQ_ONLY) {
1181 			PMD_INIT_LOG(WARNING, "TX mode %d is not supported."
1182 					" Due to txmode is meaningless in this"
1183 					" driver, just ignore.",
1184 					tx_mq_mode);
1185 		}
1186 	}
1187 	return 0;
1188 }
1189 
1190 static int
1191 eth_igb_configure(struct rte_eth_dev *dev)
1192 {
1193 	struct e1000_interrupt *intr =
1194 		E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
1195 	int ret;
1196 
1197 	PMD_INIT_FUNC_TRACE();
1198 
1199 	/* multipe queue mode checking */
1200 	ret  = igb_check_mq_mode(dev);
1201 	if (ret != 0) {
1202 		PMD_DRV_LOG(ERR, "igb_check_mq_mode fails with %d.",
1203 			    ret);
1204 		return ret;
1205 	}
1206 
1207 	intr->flags |= E1000_FLAG_NEED_LINK_UPDATE;
1208 	PMD_INIT_FUNC_TRACE();
1209 
1210 	return 0;
1211 }
1212 
1213 static int
1214 eth_igb_start(struct rte_eth_dev *dev)
1215 {
1216 	struct e1000_hw *hw =
1217 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1218 	struct e1000_adapter *adapter =
1219 		E1000_DEV_PRIVATE(dev->data->dev_private);
1220 	struct rte_intr_handle *intr_handle = &dev->pci_dev->intr_handle;
1221 	int ret, mask;
1222 	uint32_t intr_vector = 0;
1223 	uint32_t ctrl_ext;
1224 	uint32_t *speeds;
1225 	int num_speeds;
1226 	bool autoneg;
1227 
1228 	PMD_INIT_FUNC_TRACE();
1229 
1230 	/* disable uio/vfio intr/eventfd mapping */
1231 	rte_intr_disable(intr_handle);
1232 
1233 	/* Power up the phy. Needed to make the link go Up */
1234 	eth_igb_dev_set_link_up(dev);
1235 
1236 	/*
1237 	 * Packet Buffer Allocation (PBA)
1238 	 * Writing PBA sets the receive portion of the buffer
1239 	 * the remainder is used for the transmit buffer.
1240 	 */
1241 	if (hw->mac.type == e1000_82575) {
1242 		uint32_t pba;
1243 
1244 		pba = E1000_PBA_32K; /* 32K for Rx, 16K for Tx */
1245 		E1000_WRITE_REG(hw, E1000_PBA, pba);
1246 	}
1247 
1248 	/* Put the address into the Receive Address Array */
1249 	e1000_rar_set(hw, hw->mac.addr, 0);
1250 
1251 	/* Initialize the hardware */
1252 	if (igb_hardware_init(hw)) {
1253 		PMD_INIT_LOG(ERR, "Unable to initialize the hardware");
1254 		return -EIO;
1255 	}
1256 	adapter->stopped = 0;
1257 
1258 	E1000_WRITE_REG(hw, E1000_VET, ETHER_TYPE_VLAN << 16 | ETHER_TYPE_VLAN);
1259 
1260 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
1261 	/* Set PF Reset Done bit so PF/VF Mail Ops can work */
1262 	ctrl_ext |= E1000_CTRL_EXT_PFRSTD;
1263 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
1264 	E1000_WRITE_FLUSH(hw);
1265 
1266 	/* configure PF module if SRIOV enabled */
1267 	igb_pf_host_configure(dev);
1268 
1269 	/* check and configure queue intr-vector mapping */
1270 	if ((rte_intr_cap_multiple(intr_handle) ||
1271 	     !RTE_ETH_DEV_SRIOV(dev).active) &&
1272 	    dev->data->dev_conf.intr_conf.rxq != 0) {
1273 		intr_vector = dev->data->nb_rx_queues;
1274 		if (rte_intr_efd_enable(intr_handle, intr_vector))
1275 			return -1;
1276 	}
1277 
1278 	if (rte_intr_dp_is_en(intr_handle) && !intr_handle->intr_vec) {
1279 		intr_handle->intr_vec =
1280 			rte_zmalloc("intr_vec",
1281 				    dev->data->nb_rx_queues * sizeof(int), 0);
1282 		if (intr_handle->intr_vec == NULL) {
1283 			PMD_INIT_LOG(ERR, "Failed to allocate %d rx_queues"
1284 				     " intr_vec\n", dev->data->nb_rx_queues);
1285 			return -ENOMEM;
1286 		}
1287 	}
1288 
1289 	/* confiugre msix for rx interrupt */
1290 	eth_igb_configure_msix_intr(dev);
1291 
1292 	/* Configure for OS presence */
1293 	igb_init_manageability(hw);
1294 
1295 	eth_igb_tx_init(dev);
1296 
1297 	/* This can fail when allocating mbufs for descriptor rings */
1298 	ret = eth_igb_rx_init(dev);
1299 	if (ret) {
1300 		PMD_INIT_LOG(ERR, "Unable to initialize RX hardware");
1301 		igb_dev_clear_queues(dev);
1302 		return ret;
1303 	}
1304 
1305 	e1000_clear_hw_cntrs_base_generic(hw);
1306 
1307 	/*
1308 	 * VLAN Offload Settings
1309 	 */
1310 	mask = ETH_VLAN_STRIP_MASK | ETH_VLAN_FILTER_MASK | \
1311 			ETH_VLAN_EXTEND_MASK;
1312 	eth_igb_vlan_offload_set(dev, mask);
1313 
1314 	if (dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_VMDQ_ONLY) {
1315 		/* Enable VLAN filter since VMDq always use VLAN filter */
1316 		igb_vmdq_vlan_hw_filter_enable(dev);
1317 	}
1318 
1319 	if ((hw->mac.type == e1000_82576) || (hw->mac.type == e1000_82580) ||
1320 		(hw->mac.type == e1000_i350) || (hw->mac.type == e1000_i210) ||
1321 		(hw->mac.type == e1000_i211)) {
1322 		/* Configure EITR with the maximum possible value (0xFFFF) */
1323 		E1000_WRITE_REG(hw, E1000_EITR(0), 0xFFFF);
1324 	}
1325 
1326 	/* Setup link speed and duplex */
1327 	speeds = &dev->data->dev_conf.link_speeds;
1328 	if (*speeds == ETH_LINK_SPEED_AUTONEG) {
1329 		hw->phy.autoneg_advertised = E1000_ALL_SPEED_DUPLEX;
1330 	} else {
1331 		num_speeds = 0;
1332 		autoneg = (*speeds & ETH_LINK_SPEED_FIXED) == 0;
1333 
1334 		/* Reset */
1335 		hw->phy.autoneg_advertised = 0;
1336 
1337 		if (*speeds & ~(ETH_LINK_SPEED_10M_HD | ETH_LINK_SPEED_10M |
1338 				ETH_LINK_SPEED_100M_HD | ETH_LINK_SPEED_100M |
1339 				ETH_LINK_SPEED_1G | ETH_LINK_SPEED_FIXED)) {
1340 			num_speeds = -1;
1341 			goto error_invalid_config;
1342 		}
1343 		if (*speeds & ETH_LINK_SPEED_10M_HD) {
1344 			hw->phy.autoneg_advertised |= ADVERTISE_10_HALF;
1345 			num_speeds++;
1346 		}
1347 		if (*speeds & ETH_LINK_SPEED_10M) {
1348 			hw->phy.autoneg_advertised |= ADVERTISE_10_FULL;
1349 			num_speeds++;
1350 		}
1351 		if (*speeds & ETH_LINK_SPEED_100M_HD) {
1352 			hw->phy.autoneg_advertised |= ADVERTISE_100_HALF;
1353 			num_speeds++;
1354 		}
1355 		if (*speeds & ETH_LINK_SPEED_100M) {
1356 			hw->phy.autoneg_advertised |= ADVERTISE_100_FULL;
1357 			num_speeds++;
1358 		}
1359 		if (*speeds & ETH_LINK_SPEED_1G) {
1360 			hw->phy.autoneg_advertised |= ADVERTISE_1000_FULL;
1361 			num_speeds++;
1362 		}
1363 		if (num_speeds == 0 || (!autoneg && (num_speeds > 1)))
1364 			goto error_invalid_config;
1365 	}
1366 
1367 	e1000_setup_link(hw);
1368 
1369 	if (rte_intr_allow_others(intr_handle)) {
1370 		/* check if lsc interrupt is enabled */
1371 		if (dev->data->dev_conf.intr_conf.lsc != 0)
1372 			eth_igb_lsc_interrupt_setup(dev);
1373 	} else {
1374 		rte_intr_callback_unregister(intr_handle,
1375 					     eth_igb_interrupt_handler,
1376 					     (void *)dev);
1377 		if (dev->data->dev_conf.intr_conf.lsc != 0)
1378 			PMD_INIT_LOG(INFO, "lsc won't enable because of"
1379 				     " no intr multiplex\n");
1380 	}
1381 
1382 	/* check if rxq interrupt is enabled */
1383 	if (dev->data->dev_conf.intr_conf.rxq != 0 &&
1384 	    rte_intr_dp_is_en(intr_handle))
1385 		eth_igb_rxq_interrupt_setup(dev);
1386 
1387 	/* enable uio/vfio intr/eventfd mapping */
1388 	rte_intr_enable(intr_handle);
1389 
1390 	/* resume enabled intr since hw reset */
1391 	igb_intr_enable(dev);
1392 
1393 	PMD_INIT_LOG(DEBUG, "<<");
1394 
1395 	return 0;
1396 
1397 error_invalid_config:
1398 	PMD_INIT_LOG(ERR, "Invalid advertised speeds (%u) for port %u",
1399 		     dev->data->dev_conf.link_speeds, dev->data->port_id);
1400 	igb_dev_clear_queues(dev);
1401 	return -EINVAL;
1402 }
1403 
1404 /*********************************************************************
1405  *
1406  *  This routine disables all traffic on the adapter by issuing a
1407  *  global reset on the MAC.
1408  *
1409  **********************************************************************/
1410 static void
1411 eth_igb_stop(struct rte_eth_dev *dev)
1412 {
1413 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1414 	struct e1000_filter_info *filter_info =
1415 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
1416 	struct rte_eth_link link;
1417 	struct e1000_flex_filter *p_flex;
1418 	struct e1000_5tuple_filter *p_5tuple, *p_5tuple_next;
1419 	struct e1000_2tuple_filter *p_2tuple, *p_2tuple_next;
1420 	struct rte_intr_handle *intr_handle = &dev->pci_dev->intr_handle;
1421 
1422 	igb_intr_disable(hw);
1423 
1424 	/* disable intr eventfd mapping */
1425 	rte_intr_disable(intr_handle);
1426 
1427 	igb_pf_reset_hw(hw);
1428 	E1000_WRITE_REG(hw, E1000_WUC, 0);
1429 
1430 	/* Set bit for Go Link disconnect */
1431 	if (hw->mac.type >= e1000_82580) {
1432 		uint32_t phpm_reg;
1433 
1434 		phpm_reg = E1000_READ_REG(hw, E1000_82580_PHY_POWER_MGMT);
1435 		phpm_reg |= E1000_82580_PM_GO_LINKD;
1436 		E1000_WRITE_REG(hw, E1000_82580_PHY_POWER_MGMT, phpm_reg);
1437 	}
1438 
1439 	/* Power down the phy. Needed to make the link go Down */
1440 	eth_igb_dev_set_link_down(dev);
1441 
1442 	igb_dev_clear_queues(dev);
1443 
1444 	/* clear the recorded link status */
1445 	memset(&link, 0, sizeof(link));
1446 	rte_igb_dev_atomic_write_link_status(dev, &link);
1447 
1448 	/* Remove all flex filters of the device */
1449 	while ((p_flex = TAILQ_FIRST(&filter_info->flex_list))) {
1450 		TAILQ_REMOVE(&filter_info->flex_list, p_flex, entries);
1451 		rte_free(p_flex);
1452 	}
1453 	filter_info->flex_mask = 0;
1454 
1455 	/* Remove all ntuple filters of the device */
1456 	for (p_5tuple = TAILQ_FIRST(&filter_info->fivetuple_list);
1457 	     p_5tuple != NULL; p_5tuple = p_5tuple_next) {
1458 		p_5tuple_next = TAILQ_NEXT(p_5tuple, entries);
1459 		TAILQ_REMOVE(&filter_info->fivetuple_list,
1460 			     p_5tuple, entries);
1461 		rte_free(p_5tuple);
1462 	}
1463 	filter_info->fivetuple_mask = 0;
1464 	for (p_2tuple = TAILQ_FIRST(&filter_info->twotuple_list);
1465 	     p_2tuple != NULL; p_2tuple = p_2tuple_next) {
1466 		p_2tuple_next = TAILQ_NEXT(p_2tuple, entries);
1467 		TAILQ_REMOVE(&filter_info->twotuple_list,
1468 			     p_2tuple, entries);
1469 		rte_free(p_2tuple);
1470 	}
1471 	filter_info->twotuple_mask = 0;
1472 
1473 	if (!rte_intr_allow_others(intr_handle))
1474 		/* resume to the default handler */
1475 		rte_intr_callback_register(intr_handle,
1476 					   eth_igb_interrupt_handler,
1477 					   (void *)dev);
1478 
1479 	/* Clean datapath event and queue/vec mapping */
1480 	rte_intr_efd_disable(intr_handle);
1481 	if (intr_handle->intr_vec != NULL) {
1482 		rte_free(intr_handle->intr_vec);
1483 		intr_handle->intr_vec = NULL;
1484 	}
1485 }
1486 
1487 static int
1488 eth_igb_dev_set_link_up(struct rte_eth_dev *dev)
1489 {
1490 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1491 
1492 	if (hw->phy.media_type == e1000_media_type_copper)
1493 		e1000_power_up_phy(hw);
1494 	else
1495 		e1000_power_up_fiber_serdes_link(hw);
1496 
1497 	return 0;
1498 }
1499 
1500 static int
1501 eth_igb_dev_set_link_down(struct rte_eth_dev *dev)
1502 {
1503 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1504 
1505 	if (hw->phy.media_type == e1000_media_type_copper)
1506 		e1000_power_down_phy(hw);
1507 	else
1508 		e1000_shutdown_fiber_serdes_link(hw);
1509 
1510 	return 0;
1511 }
1512 
1513 static void
1514 eth_igb_close(struct rte_eth_dev *dev)
1515 {
1516 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1517 	struct e1000_adapter *adapter =
1518 		E1000_DEV_PRIVATE(dev->data->dev_private);
1519 	struct rte_eth_link link;
1520 	struct rte_pci_device *pci_dev;
1521 
1522 	eth_igb_stop(dev);
1523 	adapter->stopped = 1;
1524 
1525 	e1000_phy_hw_reset(hw);
1526 	igb_release_manageability(hw);
1527 	igb_hw_control_release(hw);
1528 
1529 	/* Clear bit for Go Link disconnect */
1530 	if (hw->mac.type >= e1000_82580) {
1531 		uint32_t phpm_reg;
1532 
1533 		phpm_reg = E1000_READ_REG(hw, E1000_82580_PHY_POWER_MGMT);
1534 		phpm_reg &= ~E1000_82580_PM_GO_LINKD;
1535 		E1000_WRITE_REG(hw, E1000_82580_PHY_POWER_MGMT, phpm_reg);
1536 	}
1537 
1538 	igb_dev_free_queues(dev);
1539 
1540 	pci_dev = dev->pci_dev;
1541 	if (pci_dev->intr_handle.intr_vec) {
1542 		rte_free(pci_dev->intr_handle.intr_vec);
1543 		pci_dev->intr_handle.intr_vec = NULL;
1544 	}
1545 
1546 	memset(&link, 0, sizeof(link));
1547 	rte_igb_dev_atomic_write_link_status(dev, &link);
1548 }
1549 
1550 static int
1551 igb_get_rx_buffer_size(struct e1000_hw *hw)
1552 {
1553 	uint32_t rx_buf_size;
1554 	if (hw->mac.type == e1000_82576) {
1555 		rx_buf_size = (E1000_READ_REG(hw, E1000_RXPBS) & 0xffff) << 10;
1556 	} else if (hw->mac.type == e1000_82580 || hw->mac.type == e1000_i350) {
1557 		/* PBS needs to be translated according to a lookup table */
1558 		rx_buf_size = (E1000_READ_REG(hw, E1000_RXPBS) & 0xf);
1559 		rx_buf_size = (uint32_t) e1000_rxpbs_adjust_82580(rx_buf_size);
1560 		rx_buf_size = (rx_buf_size << 10);
1561 	} else if (hw->mac.type == e1000_i210 || hw->mac.type == e1000_i211) {
1562 		rx_buf_size = (E1000_READ_REG(hw, E1000_RXPBS) & 0x3f) << 10;
1563 	} else {
1564 		rx_buf_size = (E1000_READ_REG(hw, E1000_PBA) & 0xffff) << 10;
1565 	}
1566 
1567 	return rx_buf_size;
1568 }
1569 
1570 /*********************************************************************
1571  *
1572  *  Initialize the hardware
1573  *
1574  **********************************************************************/
1575 static int
1576 igb_hardware_init(struct e1000_hw *hw)
1577 {
1578 	uint32_t rx_buf_size;
1579 	int diag;
1580 
1581 	/* Let the firmware know the OS is in control */
1582 	igb_hw_control_acquire(hw);
1583 
1584 	/*
1585 	 * These parameters control the automatic generation (Tx) and
1586 	 * response (Rx) to Ethernet PAUSE frames.
1587 	 * - High water mark should allow for at least two standard size (1518)
1588 	 *   frames to be received after sending an XOFF.
1589 	 * - Low water mark works best when it is very near the high water mark.
1590 	 *   This allows the receiver to restart by sending XON when it has
1591 	 *   drained a bit. Here we use an arbitrary value of 1500 which will
1592 	 *   restart after one full frame is pulled from the buffer. There
1593 	 *   could be several smaller frames in the buffer and if so they will
1594 	 *   not trigger the XON until their total number reduces the buffer
1595 	 *   by 1500.
1596 	 * - The pause time is fairly large at 1000 x 512ns = 512 usec.
1597 	 */
1598 	rx_buf_size = igb_get_rx_buffer_size(hw);
1599 
1600 	hw->fc.high_water = rx_buf_size - (ETHER_MAX_LEN * 2);
1601 	hw->fc.low_water = hw->fc.high_water - 1500;
1602 	hw->fc.pause_time = IGB_FC_PAUSE_TIME;
1603 	hw->fc.send_xon = 1;
1604 
1605 	/* Set Flow control, use the tunable location if sane */
1606 	if ((igb_fc_setting != e1000_fc_none) && (igb_fc_setting < 4))
1607 		hw->fc.requested_mode = igb_fc_setting;
1608 	else
1609 		hw->fc.requested_mode = e1000_fc_none;
1610 
1611 	/* Issue a global reset */
1612 	igb_pf_reset_hw(hw);
1613 	E1000_WRITE_REG(hw, E1000_WUC, 0);
1614 
1615 	diag = e1000_init_hw(hw);
1616 	if (diag < 0)
1617 		return diag;
1618 
1619 	E1000_WRITE_REG(hw, E1000_VET, ETHER_TYPE_VLAN << 16 | ETHER_TYPE_VLAN);
1620 	e1000_get_phy_info(hw);
1621 	e1000_check_for_link(hw);
1622 
1623 	return 0;
1624 }
1625 
1626 /* This function is based on igb_update_stats_counters() in igb/if_igb.c */
1627 static void
1628 igb_read_stats_registers(struct e1000_hw *hw, struct e1000_hw_stats *stats)
1629 {
1630 	int pause_frames;
1631 
1632 	uint64_t old_gprc  = stats->gprc;
1633 	uint64_t old_gptc  = stats->gptc;
1634 	uint64_t old_tpr   = stats->tpr;
1635 	uint64_t old_tpt   = stats->tpt;
1636 	uint64_t old_rpthc = stats->rpthc;
1637 	uint64_t old_hgptc = stats->hgptc;
1638 
1639 	if(hw->phy.media_type == e1000_media_type_copper ||
1640 	    (E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU)) {
1641 		stats->symerrs +=
1642 		    E1000_READ_REG(hw,E1000_SYMERRS);
1643 		stats->sec += E1000_READ_REG(hw, E1000_SEC);
1644 	}
1645 
1646 	stats->crcerrs += E1000_READ_REG(hw, E1000_CRCERRS);
1647 	stats->mpc += E1000_READ_REG(hw, E1000_MPC);
1648 	stats->scc += E1000_READ_REG(hw, E1000_SCC);
1649 	stats->ecol += E1000_READ_REG(hw, E1000_ECOL);
1650 
1651 	stats->mcc += E1000_READ_REG(hw, E1000_MCC);
1652 	stats->latecol += E1000_READ_REG(hw, E1000_LATECOL);
1653 	stats->colc += E1000_READ_REG(hw, E1000_COLC);
1654 	stats->dc += E1000_READ_REG(hw, E1000_DC);
1655 	stats->rlec += E1000_READ_REG(hw, E1000_RLEC);
1656 	stats->xonrxc += E1000_READ_REG(hw, E1000_XONRXC);
1657 	stats->xontxc += E1000_READ_REG(hw, E1000_XONTXC);
1658 	/*
1659 	** For watchdog management we need to know if we have been
1660 	** paused during the last interval, so capture that here.
1661 	*/
1662 	pause_frames = E1000_READ_REG(hw, E1000_XOFFRXC);
1663 	stats->xoffrxc += pause_frames;
1664 	stats->xofftxc += E1000_READ_REG(hw, E1000_XOFFTXC);
1665 	stats->fcruc += E1000_READ_REG(hw, E1000_FCRUC);
1666 	stats->prc64 += E1000_READ_REG(hw, E1000_PRC64);
1667 	stats->prc127 += E1000_READ_REG(hw, E1000_PRC127);
1668 	stats->prc255 += E1000_READ_REG(hw, E1000_PRC255);
1669 	stats->prc511 += E1000_READ_REG(hw, E1000_PRC511);
1670 	stats->prc1023 += E1000_READ_REG(hw, E1000_PRC1023);
1671 	stats->prc1522 += E1000_READ_REG(hw, E1000_PRC1522);
1672 	stats->gprc += E1000_READ_REG(hw, E1000_GPRC);
1673 	stats->bprc += E1000_READ_REG(hw, E1000_BPRC);
1674 	stats->mprc += E1000_READ_REG(hw, E1000_MPRC);
1675 	stats->gptc += E1000_READ_REG(hw, E1000_GPTC);
1676 
1677 	/* For the 64-bit byte counters the low dword must be read first. */
1678 	/* Both registers clear on the read of the high dword */
1679 
1680 	/* Workaround CRC bytes included in size, take away 4 bytes/packet */
1681 	stats->gorc += E1000_READ_REG(hw, E1000_GORCL);
1682 	stats->gorc += ((uint64_t)E1000_READ_REG(hw, E1000_GORCH) << 32);
1683 	stats->gorc -= (stats->gprc - old_gprc) * ETHER_CRC_LEN;
1684 	stats->gotc += E1000_READ_REG(hw, E1000_GOTCL);
1685 	stats->gotc += ((uint64_t)E1000_READ_REG(hw, E1000_GOTCH) << 32);
1686 	stats->gotc -= (stats->gptc - old_gptc) * ETHER_CRC_LEN;
1687 
1688 	stats->rnbc += E1000_READ_REG(hw, E1000_RNBC);
1689 	stats->ruc += E1000_READ_REG(hw, E1000_RUC);
1690 	stats->rfc += E1000_READ_REG(hw, E1000_RFC);
1691 	stats->roc += E1000_READ_REG(hw, E1000_ROC);
1692 	stats->rjc += E1000_READ_REG(hw, E1000_RJC);
1693 
1694 	stats->tpr += E1000_READ_REG(hw, E1000_TPR);
1695 	stats->tpt += E1000_READ_REG(hw, E1000_TPT);
1696 
1697 	stats->tor += E1000_READ_REG(hw, E1000_TORL);
1698 	stats->tor += ((uint64_t)E1000_READ_REG(hw, E1000_TORH) << 32);
1699 	stats->tor -= (stats->tpr - old_tpr) * ETHER_CRC_LEN;
1700 	stats->tot += E1000_READ_REG(hw, E1000_TOTL);
1701 	stats->tot += ((uint64_t)E1000_READ_REG(hw, E1000_TOTH) << 32);
1702 	stats->tot -= (stats->tpt - old_tpt) * ETHER_CRC_LEN;
1703 
1704 	stats->ptc64 += E1000_READ_REG(hw, E1000_PTC64);
1705 	stats->ptc127 += E1000_READ_REG(hw, E1000_PTC127);
1706 	stats->ptc255 += E1000_READ_REG(hw, E1000_PTC255);
1707 	stats->ptc511 += E1000_READ_REG(hw, E1000_PTC511);
1708 	stats->ptc1023 += E1000_READ_REG(hw, E1000_PTC1023);
1709 	stats->ptc1522 += E1000_READ_REG(hw, E1000_PTC1522);
1710 	stats->mptc += E1000_READ_REG(hw, E1000_MPTC);
1711 	stats->bptc += E1000_READ_REG(hw, E1000_BPTC);
1712 
1713 	/* Interrupt Counts */
1714 
1715 	stats->iac += E1000_READ_REG(hw, E1000_IAC);
1716 	stats->icrxptc += E1000_READ_REG(hw, E1000_ICRXPTC);
1717 	stats->icrxatc += E1000_READ_REG(hw, E1000_ICRXATC);
1718 	stats->ictxptc += E1000_READ_REG(hw, E1000_ICTXPTC);
1719 	stats->ictxatc += E1000_READ_REG(hw, E1000_ICTXATC);
1720 	stats->ictxqec += E1000_READ_REG(hw, E1000_ICTXQEC);
1721 	stats->ictxqmtc += E1000_READ_REG(hw, E1000_ICTXQMTC);
1722 	stats->icrxdmtc += E1000_READ_REG(hw, E1000_ICRXDMTC);
1723 	stats->icrxoc += E1000_READ_REG(hw, E1000_ICRXOC);
1724 
1725 	/* Host to Card Statistics */
1726 
1727 	stats->cbtmpc += E1000_READ_REG(hw, E1000_CBTMPC);
1728 	stats->htdpmc += E1000_READ_REG(hw, E1000_HTDPMC);
1729 	stats->cbrdpc += E1000_READ_REG(hw, E1000_CBRDPC);
1730 	stats->cbrmpc += E1000_READ_REG(hw, E1000_CBRMPC);
1731 	stats->rpthc += E1000_READ_REG(hw, E1000_RPTHC);
1732 	stats->hgptc += E1000_READ_REG(hw, E1000_HGPTC);
1733 	stats->htcbdpc += E1000_READ_REG(hw, E1000_HTCBDPC);
1734 	stats->hgorc += E1000_READ_REG(hw, E1000_HGORCL);
1735 	stats->hgorc += ((uint64_t)E1000_READ_REG(hw, E1000_HGORCH) << 32);
1736 	stats->hgorc -= (stats->rpthc - old_rpthc) * ETHER_CRC_LEN;
1737 	stats->hgotc += E1000_READ_REG(hw, E1000_HGOTCL);
1738 	stats->hgotc += ((uint64_t)E1000_READ_REG(hw, E1000_HGOTCH) << 32);
1739 	stats->hgotc -= (stats->hgptc - old_hgptc) * ETHER_CRC_LEN;
1740 	stats->lenerrs += E1000_READ_REG(hw, E1000_LENERRS);
1741 	stats->scvpc += E1000_READ_REG(hw, E1000_SCVPC);
1742 	stats->hrmpc += E1000_READ_REG(hw, E1000_HRMPC);
1743 
1744 	stats->algnerrc += E1000_READ_REG(hw, E1000_ALGNERRC);
1745 	stats->rxerrc += E1000_READ_REG(hw, E1000_RXERRC);
1746 	stats->tncrs += E1000_READ_REG(hw, E1000_TNCRS);
1747 	stats->cexterr += E1000_READ_REG(hw, E1000_CEXTERR);
1748 	stats->tsctc += E1000_READ_REG(hw, E1000_TSCTC);
1749 	stats->tsctfc += E1000_READ_REG(hw, E1000_TSCTFC);
1750 }
1751 
1752 static void
1753 eth_igb_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *rte_stats)
1754 {
1755 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1756 	struct e1000_hw_stats *stats =
1757 			E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
1758 
1759 	igb_read_stats_registers(hw, stats);
1760 
1761 	if (rte_stats == NULL)
1762 		return;
1763 
1764 	/* Rx Errors */
1765 	rte_stats->imissed = stats->mpc;
1766 	rte_stats->ierrors = stats->crcerrs +
1767 	                     stats->rlec + stats->ruc + stats->roc +
1768 	                     stats->rxerrc + stats->algnerrc + stats->cexterr;
1769 
1770 	/* Tx Errors */
1771 	rte_stats->oerrors = stats->ecol + stats->latecol;
1772 
1773 	rte_stats->ipackets = stats->gprc;
1774 	rte_stats->opackets = stats->gptc;
1775 	rte_stats->ibytes   = stats->gorc;
1776 	rte_stats->obytes   = stats->gotc;
1777 }
1778 
1779 static void
1780 eth_igb_stats_reset(struct rte_eth_dev *dev)
1781 {
1782 	struct e1000_hw_stats *hw_stats =
1783 			E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
1784 
1785 	/* HW registers are cleared on read */
1786 	eth_igb_stats_get(dev, NULL);
1787 
1788 	/* Reset software totals */
1789 	memset(hw_stats, 0, sizeof(*hw_stats));
1790 }
1791 
1792 static void
1793 eth_igb_xstats_reset(struct rte_eth_dev *dev)
1794 {
1795 	struct e1000_hw_stats *stats =
1796 			E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
1797 
1798 	/* HW registers are cleared on read */
1799 	eth_igb_xstats_get(dev, NULL, IGB_NB_XSTATS);
1800 
1801 	/* Reset software totals */
1802 	memset(stats, 0, sizeof(*stats));
1803 }
1804 
1805 static int eth_igb_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
1806 	struct rte_eth_xstat_name *xstats_names,
1807 	__rte_unused unsigned limit)
1808 {
1809 	unsigned i;
1810 
1811 	if (xstats_names == NULL)
1812 		return IGB_NB_XSTATS;
1813 
1814 	/* Note: limit checked in rte_eth_xstats_names() */
1815 
1816 	for (i = 0; i < IGB_NB_XSTATS; i++) {
1817 		snprintf(xstats_names[i].name, sizeof(xstats_names[i].name),
1818 			 "%s", rte_igb_stats_strings[i].name);
1819 	}
1820 
1821 	return IGB_NB_XSTATS;
1822 }
1823 
1824 static int
1825 eth_igb_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1826 		   unsigned n)
1827 {
1828 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1829 	struct e1000_hw_stats *hw_stats =
1830 			E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
1831 	unsigned i;
1832 
1833 	if (n < IGB_NB_XSTATS)
1834 		return IGB_NB_XSTATS;
1835 
1836 	igb_read_stats_registers(hw, hw_stats);
1837 
1838 	/* If this is a reset xstats is NULL, and we have cleared the
1839 	 * registers by reading them.
1840 	 */
1841 	if (!xstats)
1842 		return 0;
1843 
1844 	/* Extended stats */
1845 	for (i = 0; i < IGB_NB_XSTATS; i++) {
1846 		xstats[i].id = i;
1847 		xstats[i].value = *(uint64_t *)(((char *)hw_stats) +
1848 			rte_igb_stats_strings[i].offset);
1849 	}
1850 
1851 	return IGB_NB_XSTATS;
1852 }
1853 
1854 static void
1855 igbvf_read_stats_registers(struct e1000_hw *hw, struct e1000_vf_stats *hw_stats)
1856 {
1857 	/* Good Rx packets, include VF loopback */
1858 	UPDATE_VF_STAT(E1000_VFGPRC,
1859 	    hw_stats->last_gprc, hw_stats->gprc);
1860 
1861 	/* Good Rx octets, include VF loopback */
1862 	UPDATE_VF_STAT(E1000_VFGORC,
1863 	    hw_stats->last_gorc, hw_stats->gorc);
1864 
1865 	/* Good Tx packets, include VF loopback */
1866 	UPDATE_VF_STAT(E1000_VFGPTC,
1867 	    hw_stats->last_gptc, hw_stats->gptc);
1868 
1869 	/* Good Tx octets, include VF loopback */
1870 	UPDATE_VF_STAT(E1000_VFGOTC,
1871 	    hw_stats->last_gotc, hw_stats->gotc);
1872 
1873 	/* Rx Multicst packets */
1874 	UPDATE_VF_STAT(E1000_VFMPRC,
1875 	    hw_stats->last_mprc, hw_stats->mprc);
1876 
1877 	/* Good Rx loopback packets */
1878 	UPDATE_VF_STAT(E1000_VFGPRLBC,
1879 	    hw_stats->last_gprlbc, hw_stats->gprlbc);
1880 
1881 	/* Good Rx loopback octets */
1882 	UPDATE_VF_STAT(E1000_VFGORLBC,
1883 	    hw_stats->last_gorlbc, hw_stats->gorlbc);
1884 
1885 	/* Good Tx loopback packets */
1886 	UPDATE_VF_STAT(E1000_VFGPTLBC,
1887 	    hw_stats->last_gptlbc, hw_stats->gptlbc);
1888 
1889 	/* Good Tx loopback octets */
1890 	UPDATE_VF_STAT(E1000_VFGOTLBC,
1891 	    hw_stats->last_gotlbc, hw_stats->gotlbc);
1892 }
1893 
1894 static int eth_igbvf_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
1895 				     struct rte_eth_xstat_name *xstats_names,
1896 				     __rte_unused unsigned limit)
1897 {
1898 	unsigned i;
1899 
1900 	if (xstats_names != NULL)
1901 		for (i = 0; i < IGBVF_NB_XSTATS; i++) {
1902 			snprintf(xstats_names[i].name,
1903 				sizeof(xstats_names[i].name), "%s",
1904 				rte_igbvf_stats_strings[i].name);
1905 		}
1906 	return IGBVF_NB_XSTATS;
1907 }
1908 
1909 static int
1910 eth_igbvf_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1911 		     unsigned n)
1912 {
1913 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1914 	struct e1000_vf_stats *hw_stats = (struct e1000_vf_stats *)
1915 			E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
1916 	unsigned i;
1917 
1918 	if (n < IGBVF_NB_XSTATS)
1919 		return IGBVF_NB_XSTATS;
1920 
1921 	igbvf_read_stats_registers(hw, hw_stats);
1922 
1923 	if (!xstats)
1924 		return 0;
1925 
1926 	for (i = 0; i < IGBVF_NB_XSTATS; i++) {
1927 		xstats[i].id = i;
1928 		xstats[i].value = *(uint64_t *)(((char *)hw_stats) +
1929 			rte_igbvf_stats_strings[i].offset);
1930 	}
1931 
1932 	return IGBVF_NB_XSTATS;
1933 }
1934 
1935 static void
1936 eth_igbvf_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *rte_stats)
1937 {
1938 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1939 	struct e1000_vf_stats *hw_stats = (struct e1000_vf_stats *)
1940 			  E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
1941 
1942 	igbvf_read_stats_registers(hw, hw_stats);
1943 
1944 	if (rte_stats == NULL)
1945 		return;
1946 
1947 	rte_stats->ipackets = hw_stats->gprc;
1948 	rte_stats->ibytes = hw_stats->gorc;
1949 	rte_stats->opackets = hw_stats->gptc;
1950 	rte_stats->obytes = hw_stats->gotc;
1951 }
1952 
1953 static void
1954 eth_igbvf_stats_reset(struct rte_eth_dev *dev)
1955 {
1956 	struct e1000_vf_stats *hw_stats = (struct e1000_vf_stats*)
1957 			E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
1958 
1959 	/* Sync HW register to the last stats */
1960 	eth_igbvf_stats_get(dev, NULL);
1961 
1962 	/* reset HW current stats*/
1963 	memset(&hw_stats->gprc, 0, sizeof(*hw_stats) -
1964 	       offsetof(struct e1000_vf_stats, gprc));
1965 }
1966 
1967 static void
1968 eth_igb_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1969 {
1970 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1971 
1972 	dev_info->min_rx_bufsize = 256; /* See BSIZE field of RCTL register. */
1973 	dev_info->max_rx_pktlen  = 0x3FFF; /* See RLPML register. */
1974 	dev_info->max_mac_addrs = hw->mac.rar_entry_count;
1975 	dev_info->rx_offload_capa =
1976 		DEV_RX_OFFLOAD_VLAN_STRIP |
1977 		DEV_RX_OFFLOAD_IPV4_CKSUM |
1978 		DEV_RX_OFFLOAD_UDP_CKSUM  |
1979 		DEV_RX_OFFLOAD_TCP_CKSUM;
1980 	dev_info->tx_offload_capa =
1981 		DEV_TX_OFFLOAD_VLAN_INSERT |
1982 		DEV_TX_OFFLOAD_IPV4_CKSUM  |
1983 		DEV_TX_OFFLOAD_UDP_CKSUM   |
1984 		DEV_TX_OFFLOAD_TCP_CKSUM   |
1985 		DEV_TX_OFFLOAD_SCTP_CKSUM  |
1986 		DEV_TX_OFFLOAD_TCP_TSO;
1987 
1988 	switch (hw->mac.type) {
1989 	case e1000_82575:
1990 		dev_info->max_rx_queues = 4;
1991 		dev_info->max_tx_queues = 4;
1992 		dev_info->max_vmdq_pools = 0;
1993 		break;
1994 
1995 	case e1000_82576:
1996 		dev_info->max_rx_queues = 16;
1997 		dev_info->max_tx_queues = 16;
1998 		dev_info->max_vmdq_pools = ETH_8_POOLS;
1999 		dev_info->vmdq_queue_num = 16;
2000 		break;
2001 
2002 	case e1000_82580:
2003 		dev_info->max_rx_queues = 8;
2004 		dev_info->max_tx_queues = 8;
2005 		dev_info->max_vmdq_pools = ETH_8_POOLS;
2006 		dev_info->vmdq_queue_num = 8;
2007 		break;
2008 
2009 	case e1000_i350:
2010 		dev_info->max_rx_queues = 8;
2011 		dev_info->max_tx_queues = 8;
2012 		dev_info->max_vmdq_pools = ETH_8_POOLS;
2013 		dev_info->vmdq_queue_num = 8;
2014 		break;
2015 
2016 	case e1000_i354:
2017 		dev_info->max_rx_queues = 8;
2018 		dev_info->max_tx_queues = 8;
2019 		break;
2020 
2021 	case e1000_i210:
2022 		dev_info->max_rx_queues = 4;
2023 		dev_info->max_tx_queues = 4;
2024 		dev_info->max_vmdq_pools = 0;
2025 		break;
2026 
2027 	case e1000_i211:
2028 		dev_info->max_rx_queues = 2;
2029 		dev_info->max_tx_queues = 2;
2030 		dev_info->max_vmdq_pools = 0;
2031 		break;
2032 
2033 	default:
2034 		/* Should not happen */
2035 		break;
2036 	}
2037 	dev_info->hash_key_size = IGB_HKEY_MAX_INDEX * sizeof(uint32_t);
2038 	dev_info->reta_size = ETH_RSS_RETA_SIZE_128;
2039 	dev_info->flow_type_rss_offloads = IGB_RSS_OFFLOAD_ALL;
2040 
2041 	dev_info->default_rxconf = (struct rte_eth_rxconf) {
2042 		.rx_thresh = {
2043 			.pthresh = IGB_DEFAULT_RX_PTHRESH,
2044 			.hthresh = IGB_DEFAULT_RX_HTHRESH,
2045 			.wthresh = IGB_DEFAULT_RX_WTHRESH,
2046 		},
2047 		.rx_free_thresh = IGB_DEFAULT_RX_FREE_THRESH,
2048 		.rx_drop_en = 0,
2049 	};
2050 
2051 	dev_info->default_txconf = (struct rte_eth_txconf) {
2052 		.tx_thresh = {
2053 			.pthresh = IGB_DEFAULT_TX_PTHRESH,
2054 			.hthresh = IGB_DEFAULT_TX_HTHRESH,
2055 			.wthresh = IGB_DEFAULT_TX_WTHRESH,
2056 		},
2057 		.txq_flags = 0,
2058 	};
2059 
2060 	dev_info->rx_desc_lim = rx_desc_lim;
2061 	dev_info->tx_desc_lim = tx_desc_lim;
2062 
2063 	dev_info->speed_capa = ETH_LINK_SPEED_10M_HD | ETH_LINK_SPEED_10M |
2064 			ETH_LINK_SPEED_100M_HD | ETH_LINK_SPEED_100M |
2065 			ETH_LINK_SPEED_1G;
2066 }
2067 
2068 static const uint32_t *
2069 eth_igb_supported_ptypes_get(struct rte_eth_dev *dev)
2070 {
2071 	static const uint32_t ptypes[] = {
2072 		/* refers to igb_rxd_pkt_info_to_pkt_type() */
2073 		RTE_PTYPE_L2_ETHER,
2074 		RTE_PTYPE_L3_IPV4,
2075 		RTE_PTYPE_L3_IPV4_EXT,
2076 		RTE_PTYPE_L3_IPV6,
2077 		RTE_PTYPE_L3_IPV6_EXT,
2078 		RTE_PTYPE_L4_TCP,
2079 		RTE_PTYPE_L4_UDP,
2080 		RTE_PTYPE_L4_SCTP,
2081 		RTE_PTYPE_TUNNEL_IP,
2082 		RTE_PTYPE_INNER_L3_IPV6,
2083 		RTE_PTYPE_INNER_L3_IPV6_EXT,
2084 		RTE_PTYPE_INNER_L4_TCP,
2085 		RTE_PTYPE_INNER_L4_UDP,
2086 		RTE_PTYPE_UNKNOWN
2087 	};
2088 
2089 	if (dev->rx_pkt_burst == eth_igb_recv_pkts ||
2090 	    dev->rx_pkt_burst == eth_igb_recv_scattered_pkts)
2091 		return ptypes;
2092 	return NULL;
2093 }
2094 
2095 static void
2096 eth_igbvf_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2097 {
2098 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2099 
2100 	dev_info->min_rx_bufsize = 256; /* See BSIZE field of RCTL register. */
2101 	dev_info->max_rx_pktlen  = 0x3FFF; /* See RLPML register. */
2102 	dev_info->max_mac_addrs = hw->mac.rar_entry_count;
2103 	dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP |
2104 				DEV_RX_OFFLOAD_IPV4_CKSUM |
2105 				DEV_RX_OFFLOAD_UDP_CKSUM  |
2106 				DEV_RX_OFFLOAD_TCP_CKSUM;
2107 	dev_info->tx_offload_capa = DEV_TX_OFFLOAD_VLAN_INSERT |
2108 				DEV_TX_OFFLOAD_IPV4_CKSUM  |
2109 				DEV_TX_OFFLOAD_UDP_CKSUM   |
2110 				DEV_TX_OFFLOAD_TCP_CKSUM   |
2111 				DEV_TX_OFFLOAD_SCTP_CKSUM  |
2112 				DEV_TX_OFFLOAD_TCP_TSO;
2113 	switch (hw->mac.type) {
2114 	case e1000_vfadapt:
2115 		dev_info->max_rx_queues = 2;
2116 		dev_info->max_tx_queues = 2;
2117 		break;
2118 	case e1000_vfadapt_i350:
2119 		dev_info->max_rx_queues = 1;
2120 		dev_info->max_tx_queues = 1;
2121 		break;
2122 	default:
2123 		/* Should not happen */
2124 		break;
2125 	}
2126 
2127 	dev_info->default_rxconf = (struct rte_eth_rxconf) {
2128 		.rx_thresh = {
2129 			.pthresh = IGB_DEFAULT_RX_PTHRESH,
2130 			.hthresh = IGB_DEFAULT_RX_HTHRESH,
2131 			.wthresh = IGB_DEFAULT_RX_WTHRESH,
2132 		},
2133 		.rx_free_thresh = IGB_DEFAULT_RX_FREE_THRESH,
2134 		.rx_drop_en = 0,
2135 	};
2136 
2137 	dev_info->default_txconf = (struct rte_eth_txconf) {
2138 		.tx_thresh = {
2139 			.pthresh = IGB_DEFAULT_TX_PTHRESH,
2140 			.hthresh = IGB_DEFAULT_TX_HTHRESH,
2141 			.wthresh = IGB_DEFAULT_TX_WTHRESH,
2142 		},
2143 		.txq_flags = 0,
2144 	};
2145 
2146 	dev_info->rx_desc_lim = rx_desc_lim;
2147 	dev_info->tx_desc_lim = tx_desc_lim;
2148 }
2149 
2150 /* return 0 means link status changed, -1 means not changed */
2151 static int
2152 eth_igb_link_update(struct rte_eth_dev *dev, int wait_to_complete)
2153 {
2154 	struct e1000_hw *hw =
2155 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2156 	struct rte_eth_link link, old;
2157 	int link_check, count;
2158 
2159 	link_check = 0;
2160 	hw->mac.get_link_status = 1;
2161 
2162 	/* possible wait-to-complete in up to 9 seconds */
2163 	for (count = 0; count < IGB_LINK_UPDATE_CHECK_TIMEOUT; count ++) {
2164 		/* Read the real link status */
2165 		switch (hw->phy.media_type) {
2166 		case e1000_media_type_copper:
2167 			/* Do the work to read phy */
2168 			e1000_check_for_link(hw);
2169 			link_check = !hw->mac.get_link_status;
2170 			break;
2171 
2172 		case e1000_media_type_fiber:
2173 			e1000_check_for_link(hw);
2174 			link_check = (E1000_READ_REG(hw, E1000_STATUS) &
2175 				      E1000_STATUS_LU);
2176 			break;
2177 
2178 		case e1000_media_type_internal_serdes:
2179 			e1000_check_for_link(hw);
2180 			link_check = hw->mac.serdes_has_link;
2181 			break;
2182 
2183 		/* VF device is type_unknown */
2184 		case e1000_media_type_unknown:
2185 			eth_igbvf_link_update(hw);
2186 			link_check = !hw->mac.get_link_status;
2187 			break;
2188 
2189 		default:
2190 			break;
2191 		}
2192 		if (link_check || wait_to_complete == 0)
2193 			break;
2194 		rte_delay_ms(IGB_LINK_UPDATE_CHECK_INTERVAL);
2195 	}
2196 	memset(&link, 0, sizeof(link));
2197 	rte_igb_dev_atomic_read_link_status(dev, &link);
2198 	old = link;
2199 
2200 	/* Now we check if a transition has happened */
2201 	if (link_check) {
2202 		uint16_t duplex, speed;
2203 		hw->mac.ops.get_link_up_info(hw, &speed, &duplex);
2204 		link.link_duplex = (duplex == FULL_DUPLEX) ?
2205 				ETH_LINK_FULL_DUPLEX :
2206 				ETH_LINK_HALF_DUPLEX;
2207 		link.link_speed = speed;
2208 		link.link_status = ETH_LINK_UP;
2209 		link.link_autoneg = !(dev->data->dev_conf.link_speeds &
2210 				ETH_LINK_SPEED_FIXED);
2211 	} else if (!link_check) {
2212 		link.link_speed = 0;
2213 		link.link_duplex = ETH_LINK_HALF_DUPLEX;
2214 		link.link_status = ETH_LINK_DOWN;
2215 		link.link_autoneg = ETH_LINK_SPEED_FIXED;
2216 	}
2217 	rte_igb_dev_atomic_write_link_status(dev, &link);
2218 
2219 	/* not changed */
2220 	if (old.link_status == link.link_status)
2221 		return -1;
2222 
2223 	/* changed */
2224 	return 0;
2225 }
2226 
2227 /*
2228  * igb_hw_control_acquire sets CTRL_EXT:DRV_LOAD bit.
2229  * For ASF and Pass Through versions of f/w this means
2230  * that the driver is loaded.
2231  */
2232 static void
2233 igb_hw_control_acquire(struct e1000_hw *hw)
2234 {
2235 	uint32_t ctrl_ext;
2236 
2237 	/* Let firmware know the driver has taken over */
2238 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
2239 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2240 }
2241 
2242 /*
2243  * igb_hw_control_release resets CTRL_EXT:DRV_LOAD bit.
2244  * For ASF and Pass Through versions of f/w this means that the
2245  * driver is no longer loaded.
2246  */
2247 static void
2248 igb_hw_control_release(struct e1000_hw *hw)
2249 {
2250 	uint32_t ctrl_ext;
2251 
2252 	/* Let firmware taken over control of h/w */
2253 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
2254 	E1000_WRITE_REG(hw, E1000_CTRL_EXT,
2255 			ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2256 }
2257 
2258 /*
2259  * Bit of a misnomer, what this really means is
2260  * to enable OS management of the system... aka
2261  * to disable special hardware management features.
2262  */
2263 static void
2264 igb_init_manageability(struct e1000_hw *hw)
2265 {
2266 	if (e1000_enable_mng_pass_thru(hw)) {
2267 		uint32_t manc2h = E1000_READ_REG(hw, E1000_MANC2H);
2268 		uint32_t manc = E1000_READ_REG(hw, E1000_MANC);
2269 
2270 		/* disable hardware interception of ARP */
2271 		manc &= ~(E1000_MANC_ARP_EN);
2272 
2273 		/* enable receiving management packets to the host */
2274 		manc |= E1000_MANC_EN_MNG2HOST;
2275 		manc2h |= 1 << 5;  /* Mng Port 623 */
2276 		manc2h |= 1 << 6;  /* Mng Port 664 */
2277 		E1000_WRITE_REG(hw, E1000_MANC2H, manc2h);
2278 		E1000_WRITE_REG(hw, E1000_MANC, manc);
2279 	}
2280 }
2281 
2282 static void
2283 igb_release_manageability(struct e1000_hw *hw)
2284 {
2285 	if (e1000_enable_mng_pass_thru(hw)) {
2286 		uint32_t manc = E1000_READ_REG(hw, E1000_MANC);
2287 
2288 		manc |= E1000_MANC_ARP_EN;
2289 		manc &= ~E1000_MANC_EN_MNG2HOST;
2290 
2291 		E1000_WRITE_REG(hw, E1000_MANC, manc);
2292 	}
2293 }
2294 
2295 static void
2296 eth_igb_promiscuous_enable(struct rte_eth_dev *dev)
2297 {
2298 	struct e1000_hw *hw =
2299 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2300 	uint32_t rctl;
2301 
2302 	rctl = E1000_READ_REG(hw, E1000_RCTL);
2303 	rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2304 	E1000_WRITE_REG(hw, E1000_RCTL, rctl);
2305 }
2306 
2307 static void
2308 eth_igb_promiscuous_disable(struct rte_eth_dev *dev)
2309 {
2310 	struct e1000_hw *hw =
2311 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2312 	uint32_t rctl;
2313 
2314 	rctl = E1000_READ_REG(hw, E1000_RCTL);
2315 	rctl &= (~E1000_RCTL_UPE);
2316 	if (dev->data->all_multicast == 1)
2317 		rctl |= E1000_RCTL_MPE;
2318 	else
2319 		rctl &= (~E1000_RCTL_MPE);
2320 	E1000_WRITE_REG(hw, E1000_RCTL, rctl);
2321 }
2322 
2323 static void
2324 eth_igb_allmulticast_enable(struct rte_eth_dev *dev)
2325 {
2326 	struct e1000_hw *hw =
2327 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2328 	uint32_t rctl;
2329 
2330 	rctl = E1000_READ_REG(hw, E1000_RCTL);
2331 	rctl |= E1000_RCTL_MPE;
2332 	E1000_WRITE_REG(hw, E1000_RCTL, rctl);
2333 }
2334 
2335 static void
2336 eth_igb_allmulticast_disable(struct rte_eth_dev *dev)
2337 {
2338 	struct e1000_hw *hw =
2339 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2340 	uint32_t rctl;
2341 
2342 	if (dev->data->promiscuous == 1)
2343 		return; /* must remain in all_multicast mode */
2344 	rctl = E1000_READ_REG(hw, E1000_RCTL);
2345 	rctl &= (~E1000_RCTL_MPE);
2346 	E1000_WRITE_REG(hw, E1000_RCTL, rctl);
2347 }
2348 
2349 static int
2350 eth_igb_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
2351 {
2352 	struct e1000_hw *hw =
2353 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2354 	struct e1000_vfta * shadow_vfta =
2355 		E1000_DEV_PRIVATE_TO_VFTA(dev->data->dev_private);
2356 	uint32_t vfta;
2357 	uint32_t vid_idx;
2358 	uint32_t vid_bit;
2359 
2360 	vid_idx = (uint32_t) ((vlan_id >> E1000_VFTA_ENTRY_SHIFT) &
2361 			      E1000_VFTA_ENTRY_MASK);
2362 	vid_bit = (uint32_t) (1 << (vlan_id & E1000_VFTA_ENTRY_BIT_SHIFT_MASK));
2363 	vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, vid_idx);
2364 	if (on)
2365 		vfta |= vid_bit;
2366 	else
2367 		vfta &= ~vid_bit;
2368 	E1000_WRITE_REG_ARRAY(hw, E1000_VFTA, vid_idx, vfta);
2369 
2370 	/* update local VFTA copy */
2371 	shadow_vfta->vfta[vid_idx] = vfta;
2372 
2373 	return 0;
2374 }
2375 
2376 static int
2377 eth_igb_vlan_tpid_set(struct rte_eth_dev *dev,
2378 		      enum rte_vlan_type vlan_type,
2379 		      uint16_t tpid)
2380 {
2381 	struct e1000_hw *hw =
2382 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2383 	uint32_t reg, qinq;
2384 
2385 	qinq = E1000_READ_REG(hw, E1000_CTRL_EXT);
2386 	qinq &= E1000_CTRL_EXT_EXT_VLAN;
2387 
2388 	/* only outer TPID of double VLAN can be configured*/
2389 	if (qinq && vlan_type == ETH_VLAN_TYPE_OUTER) {
2390 		reg = E1000_READ_REG(hw, E1000_VET);
2391 		reg = (reg & (~E1000_VET_VET_EXT)) |
2392 			((uint32_t)tpid << E1000_VET_VET_EXT_SHIFT);
2393 		E1000_WRITE_REG(hw, E1000_VET, reg);
2394 
2395 		return 0;
2396 	}
2397 
2398 	/* all other TPID values are read-only*/
2399 	PMD_DRV_LOG(ERR, "Not supported");
2400 
2401 	return -ENOTSUP;
2402 }
2403 
2404 static void
2405 igb_vlan_hw_filter_disable(struct rte_eth_dev *dev)
2406 {
2407 	struct e1000_hw *hw =
2408 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2409 	uint32_t reg;
2410 
2411 	/* Filter Table Disable */
2412 	reg = E1000_READ_REG(hw, E1000_RCTL);
2413 	reg &= ~E1000_RCTL_CFIEN;
2414 	reg &= ~E1000_RCTL_VFE;
2415 	E1000_WRITE_REG(hw, E1000_RCTL, reg);
2416 }
2417 
2418 static void
2419 igb_vlan_hw_filter_enable(struct rte_eth_dev *dev)
2420 {
2421 	struct e1000_hw *hw =
2422 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2423 	struct e1000_vfta * shadow_vfta =
2424 		E1000_DEV_PRIVATE_TO_VFTA(dev->data->dev_private);
2425 	uint32_t reg;
2426 	int i;
2427 
2428 	/* Filter Table Enable, CFI not used for packet acceptance */
2429 	reg = E1000_READ_REG(hw, E1000_RCTL);
2430 	reg &= ~E1000_RCTL_CFIEN;
2431 	reg |= E1000_RCTL_VFE;
2432 	E1000_WRITE_REG(hw, E1000_RCTL, reg);
2433 
2434 	/* restore VFTA table */
2435 	for (i = 0; i < IGB_VFTA_SIZE; i++)
2436 		E1000_WRITE_REG_ARRAY(hw, E1000_VFTA, i, shadow_vfta->vfta[i]);
2437 }
2438 
2439 static void
2440 igb_vlan_hw_strip_disable(struct rte_eth_dev *dev)
2441 {
2442 	struct e1000_hw *hw =
2443 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2444 	uint32_t reg;
2445 
2446 	/* VLAN Mode Disable */
2447 	reg = E1000_READ_REG(hw, E1000_CTRL);
2448 	reg &= ~E1000_CTRL_VME;
2449 	E1000_WRITE_REG(hw, E1000_CTRL, reg);
2450 }
2451 
2452 static void
2453 igb_vlan_hw_strip_enable(struct rte_eth_dev *dev)
2454 {
2455 	struct e1000_hw *hw =
2456 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2457 	uint32_t reg;
2458 
2459 	/* VLAN Mode Enable */
2460 	reg = E1000_READ_REG(hw, E1000_CTRL);
2461 	reg |= E1000_CTRL_VME;
2462 	E1000_WRITE_REG(hw, E1000_CTRL, reg);
2463 }
2464 
2465 static void
2466 igb_vlan_hw_extend_disable(struct rte_eth_dev *dev)
2467 {
2468 	struct e1000_hw *hw =
2469 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2470 	uint32_t reg;
2471 
2472 	/* CTRL_EXT: Extended VLAN */
2473 	reg = E1000_READ_REG(hw, E1000_CTRL_EXT);
2474 	reg &= ~E1000_CTRL_EXT_EXTEND_VLAN;
2475 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, reg);
2476 
2477 	/* Update maximum packet length */
2478 	if (dev->data->dev_conf.rxmode.jumbo_frame == 1)
2479 		E1000_WRITE_REG(hw, E1000_RLPML,
2480 			dev->data->dev_conf.rxmode.max_rx_pkt_len +
2481 						VLAN_TAG_SIZE);
2482 }
2483 
2484 static void
2485 igb_vlan_hw_extend_enable(struct rte_eth_dev *dev)
2486 {
2487 	struct e1000_hw *hw =
2488 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2489 	uint32_t reg;
2490 
2491 	/* CTRL_EXT: Extended VLAN */
2492 	reg = E1000_READ_REG(hw, E1000_CTRL_EXT);
2493 	reg |= E1000_CTRL_EXT_EXTEND_VLAN;
2494 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, reg);
2495 
2496 	/* Update maximum packet length */
2497 	if (dev->data->dev_conf.rxmode.jumbo_frame == 1)
2498 		E1000_WRITE_REG(hw, E1000_RLPML,
2499 			dev->data->dev_conf.rxmode.max_rx_pkt_len +
2500 						2 * VLAN_TAG_SIZE);
2501 }
2502 
2503 static void
2504 eth_igb_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2505 {
2506 	if(mask & ETH_VLAN_STRIP_MASK){
2507 		if (dev->data->dev_conf.rxmode.hw_vlan_strip)
2508 			igb_vlan_hw_strip_enable(dev);
2509 		else
2510 			igb_vlan_hw_strip_disable(dev);
2511 	}
2512 
2513 	if(mask & ETH_VLAN_FILTER_MASK){
2514 		if (dev->data->dev_conf.rxmode.hw_vlan_filter)
2515 			igb_vlan_hw_filter_enable(dev);
2516 		else
2517 			igb_vlan_hw_filter_disable(dev);
2518 	}
2519 
2520 	if(mask & ETH_VLAN_EXTEND_MASK){
2521 		if (dev->data->dev_conf.rxmode.hw_vlan_extend)
2522 			igb_vlan_hw_extend_enable(dev);
2523 		else
2524 			igb_vlan_hw_extend_disable(dev);
2525 	}
2526 }
2527 
2528 
2529 /**
2530  * It enables the interrupt mask and then enable the interrupt.
2531  *
2532  * @param dev
2533  *  Pointer to struct rte_eth_dev.
2534  *
2535  * @return
2536  *  - On success, zero.
2537  *  - On failure, a negative value.
2538  */
2539 static int
2540 eth_igb_lsc_interrupt_setup(struct rte_eth_dev *dev)
2541 {
2542 	struct e1000_interrupt *intr =
2543 		E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
2544 
2545 	intr->mask |= E1000_ICR_LSC;
2546 
2547 	return 0;
2548 }
2549 
2550 /* It clears the interrupt causes and enables the interrupt.
2551  * It will be called once only during nic initialized.
2552  *
2553  * @param dev
2554  *  Pointer to struct rte_eth_dev.
2555  *
2556  * @return
2557  *  - On success, zero.
2558  *  - On failure, a negative value.
2559  */
2560 static int eth_igb_rxq_interrupt_setup(struct rte_eth_dev *dev)
2561 {
2562 	uint32_t mask, regval;
2563 	struct e1000_hw *hw =
2564 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2565 	struct rte_eth_dev_info dev_info;
2566 
2567 	memset(&dev_info, 0, sizeof(dev_info));
2568 	eth_igb_infos_get(dev, &dev_info);
2569 
2570 	mask = 0xFFFFFFFF >> (32 - dev_info.max_rx_queues);
2571 	regval = E1000_READ_REG(hw, E1000_EIMS);
2572 	E1000_WRITE_REG(hw, E1000_EIMS, regval | mask);
2573 
2574 	return 0;
2575 }
2576 
2577 /*
2578  * It reads ICR and gets interrupt causes, check it and set a bit flag
2579  * to update link status.
2580  *
2581  * @param dev
2582  *  Pointer to struct rte_eth_dev.
2583  *
2584  * @return
2585  *  - On success, zero.
2586  *  - On failure, a negative value.
2587  */
2588 static int
2589 eth_igb_interrupt_get_status(struct rte_eth_dev *dev)
2590 {
2591 	uint32_t icr;
2592 	struct e1000_hw *hw =
2593 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2594 	struct e1000_interrupt *intr =
2595 		E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
2596 
2597 	igb_intr_disable(hw);
2598 
2599 	/* read-on-clear nic registers here */
2600 	icr = E1000_READ_REG(hw, E1000_ICR);
2601 
2602 	intr->flags = 0;
2603 	if (icr & E1000_ICR_LSC) {
2604 		intr->flags |= E1000_FLAG_NEED_LINK_UPDATE;
2605 	}
2606 
2607 	if (icr & E1000_ICR_VMMB)
2608 		intr->flags |= E1000_FLAG_MAILBOX;
2609 
2610 	return 0;
2611 }
2612 
2613 /*
2614  * It executes link_update after knowing an interrupt is prsent.
2615  *
2616  * @param dev
2617  *  Pointer to struct rte_eth_dev.
2618  *
2619  * @return
2620  *  - On success, zero.
2621  *  - On failure, a negative value.
2622  */
2623 static int
2624 eth_igb_interrupt_action(struct rte_eth_dev *dev)
2625 {
2626 	struct e1000_hw *hw =
2627 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2628 	struct e1000_interrupt *intr =
2629 		E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
2630 	uint32_t tctl, rctl;
2631 	struct rte_eth_link link;
2632 	int ret;
2633 
2634 	if (intr->flags & E1000_FLAG_MAILBOX) {
2635 		igb_pf_mbx_process(dev);
2636 		intr->flags &= ~E1000_FLAG_MAILBOX;
2637 	}
2638 
2639 	igb_intr_enable(dev);
2640 	rte_intr_enable(&(dev->pci_dev->intr_handle));
2641 
2642 	if (intr->flags & E1000_FLAG_NEED_LINK_UPDATE) {
2643 		intr->flags &= ~E1000_FLAG_NEED_LINK_UPDATE;
2644 
2645 		/* set get_link_status to check register later */
2646 		hw->mac.get_link_status = 1;
2647 		ret = eth_igb_link_update(dev, 0);
2648 
2649 		/* check if link has changed */
2650 		if (ret < 0)
2651 			return 0;
2652 
2653 		memset(&link, 0, sizeof(link));
2654 		rte_igb_dev_atomic_read_link_status(dev, &link);
2655 		if (link.link_status) {
2656 			PMD_INIT_LOG(INFO,
2657 				     " Port %d: Link Up - speed %u Mbps - %s",
2658 				     dev->data->port_id,
2659 				     (unsigned)link.link_speed,
2660 				     link.link_duplex == ETH_LINK_FULL_DUPLEX ?
2661 				     "full-duplex" : "half-duplex");
2662 		} else {
2663 			PMD_INIT_LOG(INFO, " Port %d: Link Down",
2664 				     dev->data->port_id);
2665 		}
2666 
2667 		PMD_INIT_LOG(DEBUG, "PCI Address: %04d:%02d:%02d:%d",
2668 			     dev->pci_dev->addr.domain,
2669 			     dev->pci_dev->addr.bus,
2670 			     dev->pci_dev->addr.devid,
2671 			     dev->pci_dev->addr.function);
2672 		tctl = E1000_READ_REG(hw, E1000_TCTL);
2673 		rctl = E1000_READ_REG(hw, E1000_RCTL);
2674 		if (link.link_status) {
2675 			/* enable Tx/Rx */
2676 			tctl |= E1000_TCTL_EN;
2677 			rctl |= E1000_RCTL_EN;
2678 		} else {
2679 			/* disable Tx/Rx */
2680 			tctl &= ~E1000_TCTL_EN;
2681 			rctl &= ~E1000_RCTL_EN;
2682 		}
2683 		E1000_WRITE_REG(hw, E1000_TCTL, tctl);
2684 		E1000_WRITE_REG(hw, E1000_RCTL, rctl);
2685 		E1000_WRITE_FLUSH(hw);
2686 		_rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC);
2687 	}
2688 
2689 	return 0;
2690 }
2691 
2692 /**
2693  * Interrupt handler which shall be registered at first.
2694  *
2695  * @param handle
2696  *  Pointer to interrupt handle.
2697  * @param param
2698  *  The address of parameter (struct rte_eth_dev *) regsitered before.
2699  *
2700  * @return
2701  *  void
2702  */
2703 static void
2704 eth_igb_interrupt_handler(__rte_unused struct rte_intr_handle *handle,
2705 							void *param)
2706 {
2707 	struct rte_eth_dev *dev = (struct rte_eth_dev *)param;
2708 
2709 	eth_igb_interrupt_get_status(dev);
2710 	eth_igb_interrupt_action(dev);
2711 }
2712 
2713 static int
2714 eth_igbvf_interrupt_get_status(struct rte_eth_dev *dev)
2715 {
2716 	uint32_t eicr;
2717 	struct e1000_hw *hw =
2718 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2719 	struct e1000_interrupt *intr =
2720 		E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
2721 
2722 	igbvf_intr_disable(hw);
2723 
2724 	/* read-on-clear nic registers here */
2725 	eicr = E1000_READ_REG(hw, E1000_EICR);
2726 	intr->flags = 0;
2727 
2728 	if (eicr == E1000_VTIVAR_MISC_MAILBOX)
2729 		intr->flags |= E1000_FLAG_MAILBOX;
2730 
2731 	return 0;
2732 }
2733 
2734 void igbvf_mbx_process(struct rte_eth_dev *dev)
2735 {
2736 	struct e1000_hw *hw =
2737 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2738 	struct e1000_mbx_info *mbx = &hw->mbx;
2739 	u32 in_msg = 0;
2740 
2741 	if (mbx->ops.read(hw, &in_msg, 1, 0))
2742 		return;
2743 
2744 	/* PF reset VF event */
2745 	if (in_msg == E1000_PF_CONTROL_MSG)
2746 		_rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_RESET);
2747 }
2748 
2749 static int
2750 eth_igbvf_interrupt_action(struct rte_eth_dev *dev)
2751 {
2752 	struct e1000_interrupt *intr =
2753 		E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
2754 
2755 	if (intr->flags & E1000_FLAG_MAILBOX) {
2756 		igbvf_mbx_process(dev);
2757 		intr->flags &= ~E1000_FLAG_MAILBOX;
2758 	}
2759 
2760 	igbvf_intr_enable(dev);
2761 	rte_intr_enable(&dev->pci_dev->intr_handle);
2762 
2763 	return 0;
2764 }
2765 
2766 static void
2767 eth_igbvf_interrupt_handler(__rte_unused struct rte_intr_handle *handle,
2768 			    void *param)
2769 {
2770 	struct rte_eth_dev *dev = (struct rte_eth_dev *)param;
2771 
2772 	eth_igbvf_interrupt_get_status(dev);
2773 	eth_igbvf_interrupt_action(dev);
2774 }
2775 
2776 static int
2777 eth_igb_led_on(struct rte_eth_dev *dev)
2778 {
2779 	struct e1000_hw *hw;
2780 
2781 	hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2782 	return e1000_led_on(hw) == E1000_SUCCESS ? 0 : -ENOTSUP;
2783 }
2784 
2785 static int
2786 eth_igb_led_off(struct rte_eth_dev *dev)
2787 {
2788 	struct e1000_hw *hw;
2789 
2790 	hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2791 	return e1000_led_off(hw) == E1000_SUCCESS ? 0 : -ENOTSUP;
2792 }
2793 
2794 static int
2795 eth_igb_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
2796 {
2797 	struct e1000_hw *hw;
2798 	uint32_t ctrl;
2799 	int tx_pause;
2800 	int rx_pause;
2801 
2802 	hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2803 	fc_conf->pause_time = hw->fc.pause_time;
2804 	fc_conf->high_water = hw->fc.high_water;
2805 	fc_conf->low_water = hw->fc.low_water;
2806 	fc_conf->send_xon = hw->fc.send_xon;
2807 	fc_conf->autoneg = hw->mac.autoneg;
2808 
2809 	/*
2810 	 * Return rx_pause and tx_pause status according to actual setting of
2811 	 * the TFCE and RFCE bits in the CTRL register.
2812 	 */
2813 	ctrl = E1000_READ_REG(hw, E1000_CTRL);
2814 	if (ctrl & E1000_CTRL_TFCE)
2815 		tx_pause = 1;
2816 	else
2817 		tx_pause = 0;
2818 
2819 	if (ctrl & E1000_CTRL_RFCE)
2820 		rx_pause = 1;
2821 	else
2822 		rx_pause = 0;
2823 
2824 	if (rx_pause && tx_pause)
2825 		fc_conf->mode = RTE_FC_FULL;
2826 	else if (rx_pause)
2827 		fc_conf->mode = RTE_FC_RX_PAUSE;
2828 	else if (tx_pause)
2829 		fc_conf->mode = RTE_FC_TX_PAUSE;
2830 	else
2831 		fc_conf->mode = RTE_FC_NONE;
2832 
2833 	return 0;
2834 }
2835 
2836 static int
2837 eth_igb_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
2838 {
2839 	struct e1000_hw *hw;
2840 	int err;
2841 	enum e1000_fc_mode rte_fcmode_2_e1000_fcmode[] = {
2842 		e1000_fc_none,
2843 		e1000_fc_rx_pause,
2844 		e1000_fc_tx_pause,
2845 		e1000_fc_full
2846 	};
2847 	uint32_t rx_buf_size;
2848 	uint32_t max_high_water;
2849 	uint32_t rctl;
2850 
2851 	hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2852 	if (fc_conf->autoneg != hw->mac.autoneg)
2853 		return -ENOTSUP;
2854 	rx_buf_size = igb_get_rx_buffer_size(hw);
2855 	PMD_INIT_LOG(DEBUG, "Rx packet buffer size = 0x%x", rx_buf_size);
2856 
2857 	/* At least reserve one Ethernet frame for watermark */
2858 	max_high_water = rx_buf_size - ETHER_MAX_LEN;
2859 	if ((fc_conf->high_water > max_high_water) ||
2860 	    (fc_conf->high_water < fc_conf->low_water)) {
2861 		PMD_INIT_LOG(ERR, "e1000 incorrect high/low water value");
2862 		PMD_INIT_LOG(ERR, "high water must <=  0x%x", max_high_water);
2863 		return -EINVAL;
2864 	}
2865 
2866 	hw->fc.requested_mode = rte_fcmode_2_e1000_fcmode[fc_conf->mode];
2867 	hw->fc.pause_time     = fc_conf->pause_time;
2868 	hw->fc.high_water     = fc_conf->high_water;
2869 	hw->fc.low_water      = fc_conf->low_water;
2870 	hw->fc.send_xon	      = fc_conf->send_xon;
2871 
2872 	err = e1000_setup_link_generic(hw);
2873 	if (err == E1000_SUCCESS) {
2874 
2875 		/* check if we want to forward MAC frames - driver doesn't have native
2876 		 * capability to do that, so we'll write the registers ourselves */
2877 
2878 		rctl = E1000_READ_REG(hw, E1000_RCTL);
2879 
2880 		/* set or clear MFLCN.PMCF bit depending on configuration */
2881 		if (fc_conf->mac_ctrl_frame_fwd != 0)
2882 			rctl |= E1000_RCTL_PMCF;
2883 		else
2884 			rctl &= ~E1000_RCTL_PMCF;
2885 
2886 		E1000_WRITE_REG(hw, E1000_RCTL, rctl);
2887 		E1000_WRITE_FLUSH(hw);
2888 
2889 		return 0;
2890 	}
2891 
2892 	PMD_INIT_LOG(ERR, "e1000_setup_link_generic = 0x%x", err);
2893 	return -EIO;
2894 }
2895 
2896 #define E1000_RAH_POOLSEL_SHIFT      (18)
2897 static void
2898 eth_igb_rar_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
2899 	        uint32_t index, __rte_unused uint32_t pool)
2900 {
2901 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2902 	uint32_t rah;
2903 
2904 	e1000_rar_set(hw, mac_addr->addr_bytes, index);
2905 	rah = E1000_READ_REG(hw, E1000_RAH(index));
2906 	rah |= (0x1 << (E1000_RAH_POOLSEL_SHIFT + pool));
2907 	E1000_WRITE_REG(hw, E1000_RAH(index), rah);
2908 }
2909 
2910 static void
2911 eth_igb_rar_clear(struct rte_eth_dev *dev, uint32_t index)
2912 {
2913 	uint8_t addr[ETHER_ADDR_LEN];
2914 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2915 
2916 	memset(addr, 0, sizeof(addr));
2917 
2918 	e1000_rar_set(hw, addr, index);
2919 }
2920 
2921 static void
2922 eth_igb_default_mac_addr_set(struct rte_eth_dev *dev,
2923 				struct ether_addr *addr)
2924 {
2925 	eth_igb_rar_clear(dev, 0);
2926 
2927 	eth_igb_rar_set(dev, (void *)addr, 0, 0);
2928 }
2929 /*
2930  * Virtual Function operations
2931  */
2932 static void
2933 igbvf_intr_disable(struct e1000_hw *hw)
2934 {
2935 	PMD_INIT_FUNC_TRACE();
2936 
2937 	/* Clear interrupt mask to stop from interrupts being generated */
2938 	E1000_WRITE_REG(hw, E1000_EIMC, 0xFFFF);
2939 
2940 	E1000_WRITE_FLUSH(hw);
2941 }
2942 
2943 static void
2944 igbvf_stop_adapter(struct rte_eth_dev *dev)
2945 {
2946 	u32 reg_val;
2947 	u16 i;
2948 	struct rte_eth_dev_info dev_info;
2949 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2950 
2951 	memset(&dev_info, 0, sizeof(dev_info));
2952 	eth_igbvf_infos_get(dev, &dev_info);
2953 
2954 	/* Clear interrupt mask to stop from interrupts being generated */
2955 	igbvf_intr_disable(hw);
2956 
2957 	/* Clear any pending interrupts, flush previous writes */
2958 	E1000_READ_REG(hw, E1000_EICR);
2959 
2960 	/* Disable the transmit unit.  Each queue must be disabled. */
2961 	for (i = 0; i < dev_info.max_tx_queues; i++)
2962 		E1000_WRITE_REG(hw, E1000_TXDCTL(i), E1000_TXDCTL_SWFLSH);
2963 
2964 	/* Disable the receive unit by stopping each queue */
2965 	for (i = 0; i < dev_info.max_rx_queues; i++) {
2966 		reg_val = E1000_READ_REG(hw, E1000_RXDCTL(i));
2967 		reg_val &= ~E1000_RXDCTL_QUEUE_ENABLE;
2968 		E1000_WRITE_REG(hw, E1000_RXDCTL(i), reg_val);
2969 		while (E1000_READ_REG(hw, E1000_RXDCTL(i)) & E1000_RXDCTL_QUEUE_ENABLE)
2970 			;
2971 	}
2972 
2973 	/* flush all queues disables */
2974 	E1000_WRITE_FLUSH(hw);
2975 	msec_delay(2);
2976 }
2977 
2978 static int eth_igbvf_link_update(struct e1000_hw *hw)
2979 {
2980 	struct e1000_mbx_info *mbx = &hw->mbx;
2981 	struct e1000_mac_info *mac = &hw->mac;
2982 	int ret_val = E1000_SUCCESS;
2983 
2984 	PMD_INIT_LOG(DEBUG, "e1000_check_for_link_vf");
2985 
2986 	/*
2987 	 * We only want to run this if there has been a rst asserted.
2988 	 * in this case that could mean a link change, device reset,
2989 	 * or a virtual function reset
2990 	 */
2991 
2992 	/* If we were hit with a reset or timeout drop the link */
2993 	if (!e1000_check_for_rst(hw, 0) || !mbx->timeout)
2994 		mac->get_link_status = TRUE;
2995 
2996 	if (!mac->get_link_status)
2997 		goto out;
2998 
2999 	/* if link status is down no point in checking to see if pf is up */
3000 	if (!(E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU))
3001 		goto out;
3002 
3003 	/* if we passed all the tests above then the link is up and we no
3004 	 * longer need to check for link */
3005 	mac->get_link_status = FALSE;
3006 
3007 out:
3008 	return ret_val;
3009 }
3010 
3011 
3012 static int
3013 igbvf_dev_configure(struct rte_eth_dev *dev)
3014 {
3015 	struct rte_eth_conf* conf = &dev->data->dev_conf;
3016 
3017 	PMD_INIT_LOG(DEBUG, "Configured Virtual Function port id: %d",
3018 		     dev->data->port_id);
3019 
3020 	/*
3021 	 * VF has no ability to enable/disable HW CRC
3022 	 * Keep the persistent behavior the same as Host PF
3023 	 */
3024 #ifndef RTE_LIBRTE_E1000_PF_DISABLE_STRIP_CRC
3025 	if (!conf->rxmode.hw_strip_crc) {
3026 		PMD_INIT_LOG(NOTICE, "VF can't disable HW CRC Strip");
3027 		conf->rxmode.hw_strip_crc = 1;
3028 	}
3029 #else
3030 	if (conf->rxmode.hw_strip_crc) {
3031 		PMD_INIT_LOG(NOTICE, "VF can't enable HW CRC Strip");
3032 		conf->rxmode.hw_strip_crc = 0;
3033 	}
3034 #endif
3035 
3036 	return 0;
3037 }
3038 
3039 static int
3040 igbvf_dev_start(struct rte_eth_dev *dev)
3041 {
3042 	struct e1000_hw *hw =
3043 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3044 	struct e1000_adapter *adapter =
3045 		E1000_DEV_PRIVATE(dev->data->dev_private);
3046 	int ret;
3047 	struct rte_intr_handle *intr_handle = &dev->pci_dev->intr_handle;
3048 	uint32_t intr_vector = 0;
3049 
3050 	PMD_INIT_FUNC_TRACE();
3051 
3052 	hw->mac.ops.reset_hw(hw);
3053 	adapter->stopped = 0;
3054 
3055 	/* Set all vfta */
3056 	igbvf_set_vfta_all(dev,1);
3057 
3058 	eth_igbvf_tx_init(dev);
3059 
3060 	/* This can fail when allocating mbufs for descriptor rings */
3061 	ret = eth_igbvf_rx_init(dev);
3062 	if (ret) {
3063 		PMD_INIT_LOG(ERR, "Unable to initialize RX hardware");
3064 		igb_dev_clear_queues(dev);
3065 		return ret;
3066 	}
3067 
3068 	/* check and configure queue intr-vector mapping */
3069 	if (dev->data->dev_conf.intr_conf.rxq != 0) {
3070 		intr_vector = dev->data->nb_rx_queues;
3071 		ret = rte_intr_efd_enable(intr_handle, intr_vector);
3072 		if (ret)
3073 			return ret;
3074 	}
3075 
3076 	if (rte_intr_dp_is_en(intr_handle) && !intr_handle->intr_vec) {
3077 		intr_handle->intr_vec =
3078 			rte_zmalloc("intr_vec",
3079 				    dev->data->nb_rx_queues * sizeof(int), 0);
3080 		if (!intr_handle->intr_vec) {
3081 			PMD_INIT_LOG(ERR, "Failed to allocate %d rx_queues"
3082 				     " intr_vec\n", dev->data->nb_rx_queues);
3083 			return -ENOMEM;
3084 		}
3085 	}
3086 
3087 	eth_igbvf_configure_msix_intr(dev);
3088 
3089 	/* enable uio/vfio intr/eventfd mapping */
3090 	rte_intr_enable(intr_handle);
3091 
3092 	/* resume enabled intr since hw reset */
3093 	igbvf_intr_enable(dev);
3094 
3095 	return 0;
3096 }
3097 
3098 static void
3099 igbvf_dev_stop(struct rte_eth_dev *dev)
3100 {
3101 	struct rte_intr_handle *intr_handle = &dev->pci_dev->intr_handle;
3102 
3103 	PMD_INIT_FUNC_TRACE();
3104 
3105 	igbvf_stop_adapter(dev);
3106 
3107 	/*
3108 	  * Clear what we set, but we still keep shadow_vfta to
3109 	  * restore after device starts
3110 	  */
3111 	igbvf_set_vfta_all(dev,0);
3112 
3113 	igb_dev_clear_queues(dev);
3114 
3115 	/* disable intr eventfd mapping */
3116 	rte_intr_disable(intr_handle);
3117 
3118 	/* Clean datapath event and queue/vec mapping */
3119 	rte_intr_efd_disable(intr_handle);
3120 	if (intr_handle->intr_vec) {
3121 		rte_free(intr_handle->intr_vec);
3122 		intr_handle->intr_vec = NULL;
3123 	}
3124 }
3125 
3126 static void
3127 igbvf_dev_close(struct rte_eth_dev *dev)
3128 {
3129 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3130 	struct e1000_adapter *adapter =
3131 		E1000_DEV_PRIVATE(dev->data->dev_private);
3132 	struct ether_addr addr;
3133 
3134 	PMD_INIT_FUNC_TRACE();
3135 
3136 	e1000_reset_hw(hw);
3137 
3138 	igbvf_dev_stop(dev);
3139 	adapter->stopped = 1;
3140 	igb_dev_free_queues(dev);
3141 
3142 	/**
3143 	 * reprogram the RAR with a zero mac address,
3144 	 * to ensure that the VF traffic goes to the PF
3145 	 * after stop, close and detach of the VF.
3146 	 **/
3147 
3148 	memset(&addr, 0, sizeof(addr));
3149 	igbvf_default_mac_addr_set(dev, &addr);
3150 }
3151 
3152 static void
3153 igbvf_promiscuous_enable(struct rte_eth_dev *dev)
3154 {
3155 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3156 
3157 	/* Set both unicast and multicast promisc */
3158 	e1000_promisc_set_vf(hw, e1000_promisc_enabled);
3159 }
3160 
3161 static void
3162 igbvf_promiscuous_disable(struct rte_eth_dev *dev)
3163 {
3164 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3165 
3166 	/* If in allmulticast mode leave multicast promisc */
3167 	if (dev->data->all_multicast == 1)
3168 		e1000_promisc_set_vf(hw, e1000_promisc_multicast);
3169 	else
3170 		e1000_promisc_set_vf(hw, e1000_promisc_disabled);
3171 }
3172 
3173 static void
3174 igbvf_allmulticast_enable(struct rte_eth_dev *dev)
3175 {
3176 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3177 
3178 	/* In promiscuous mode multicast promisc already set */
3179 	if (dev->data->promiscuous == 0)
3180 		e1000_promisc_set_vf(hw, e1000_promisc_multicast);
3181 }
3182 
3183 static void
3184 igbvf_allmulticast_disable(struct rte_eth_dev *dev)
3185 {
3186 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3187 
3188 	/* In promiscuous mode leave multicast promisc enabled */
3189 	if (dev->data->promiscuous == 0)
3190 		e1000_promisc_set_vf(hw, e1000_promisc_disabled);
3191 }
3192 
3193 static int igbvf_set_vfta(struct e1000_hw *hw, uint16_t vid, bool on)
3194 {
3195 	struct e1000_mbx_info *mbx = &hw->mbx;
3196 	uint32_t msgbuf[2];
3197 	s32 err;
3198 
3199 	/* After set vlan, vlan strip will also be enabled in igb driver*/
3200 	msgbuf[0] = E1000_VF_SET_VLAN;
3201 	msgbuf[1] = vid;
3202 	/* Setting the 8 bit field MSG INFO to TRUE indicates "add" */
3203 	if (on)
3204 		msgbuf[0] |= E1000_VF_SET_VLAN_ADD;
3205 
3206 	err = mbx->ops.write_posted(hw, msgbuf, 2, 0);
3207 	if (err)
3208 		goto mbx_err;
3209 
3210 	err = mbx->ops.read_posted(hw, msgbuf, 2, 0);
3211 	if (err)
3212 		goto mbx_err;
3213 
3214 	msgbuf[0] &= ~E1000_VT_MSGTYPE_CTS;
3215 	if (msgbuf[0] == (E1000_VF_SET_VLAN | E1000_VT_MSGTYPE_NACK))
3216 		err = -EINVAL;
3217 
3218 mbx_err:
3219 	return err;
3220 }
3221 
3222 static void igbvf_set_vfta_all(struct rte_eth_dev *dev, bool on)
3223 {
3224 	struct e1000_hw *hw =
3225 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3226 	struct e1000_vfta * shadow_vfta =
3227 		E1000_DEV_PRIVATE_TO_VFTA(dev->data->dev_private);
3228 	int i = 0, j = 0, vfta = 0, mask = 1;
3229 
3230 	for (i = 0; i < IGB_VFTA_SIZE; i++){
3231 		vfta = shadow_vfta->vfta[i];
3232 		if(vfta){
3233 			mask = 1;
3234 			for (j = 0; j < 32; j++){
3235 				if(vfta & mask)
3236 					igbvf_set_vfta(hw,
3237 						(uint16_t)((i<<5)+j), on);
3238 				mask<<=1;
3239 			}
3240 		}
3241 	}
3242 
3243 }
3244 
3245 static int
3246 igbvf_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
3247 {
3248 	struct e1000_hw *hw =
3249 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3250 	struct e1000_vfta * shadow_vfta =
3251 		E1000_DEV_PRIVATE_TO_VFTA(dev->data->dev_private);
3252 	uint32_t vid_idx = 0;
3253 	uint32_t vid_bit = 0;
3254 	int ret = 0;
3255 
3256 	PMD_INIT_FUNC_TRACE();
3257 
3258 	/*vind is not used in VF driver, set to 0, check ixgbe_set_vfta_vf*/
3259 	ret = igbvf_set_vfta(hw, vlan_id, !!on);
3260 	if(ret){
3261 		PMD_INIT_LOG(ERR, "Unable to set VF vlan");
3262 		return ret;
3263 	}
3264 	vid_idx = (uint32_t) ((vlan_id >> 5) & 0x7F);
3265 	vid_bit = (uint32_t) (1 << (vlan_id & 0x1F));
3266 
3267 	/*Save what we set and retore it after device reset*/
3268 	if (on)
3269 		shadow_vfta->vfta[vid_idx] |= vid_bit;
3270 	else
3271 		shadow_vfta->vfta[vid_idx] &= ~vid_bit;
3272 
3273 	return 0;
3274 }
3275 
3276 static void
3277 igbvf_default_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *addr)
3278 {
3279 	struct e1000_hw *hw =
3280 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3281 
3282 	/* index is not used by rar_set() */
3283 	hw->mac.ops.rar_set(hw, (void *)addr, 0);
3284 }
3285 
3286 
3287 static int
3288 eth_igb_rss_reta_update(struct rte_eth_dev *dev,
3289 			struct rte_eth_rss_reta_entry64 *reta_conf,
3290 			uint16_t reta_size)
3291 {
3292 	uint8_t i, j, mask;
3293 	uint32_t reta, r;
3294 	uint16_t idx, shift;
3295 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3296 
3297 	if (reta_size != ETH_RSS_RETA_SIZE_128) {
3298 		PMD_DRV_LOG(ERR, "The size of hash lookup table configured "
3299 			"(%d) doesn't match the number hardware can supported "
3300 			"(%d)\n", reta_size, ETH_RSS_RETA_SIZE_128);
3301 		return -EINVAL;
3302 	}
3303 
3304 	for (i = 0; i < reta_size; i += IGB_4_BIT_WIDTH) {
3305 		idx = i / RTE_RETA_GROUP_SIZE;
3306 		shift = i % RTE_RETA_GROUP_SIZE;
3307 		mask = (uint8_t)((reta_conf[idx].mask >> shift) &
3308 						IGB_4_BIT_MASK);
3309 		if (!mask)
3310 			continue;
3311 		if (mask == IGB_4_BIT_MASK)
3312 			r = 0;
3313 		else
3314 			r = E1000_READ_REG(hw, E1000_RETA(i >> 2));
3315 		for (j = 0, reta = 0; j < IGB_4_BIT_WIDTH; j++) {
3316 			if (mask & (0x1 << j))
3317 				reta |= reta_conf[idx].reta[shift + j] <<
3318 							(CHAR_BIT * j);
3319 			else
3320 				reta |= r & (IGB_8_BIT_MASK << (CHAR_BIT * j));
3321 		}
3322 		E1000_WRITE_REG(hw, E1000_RETA(i >> 2), reta);
3323 	}
3324 
3325 	return 0;
3326 }
3327 
3328 static int
3329 eth_igb_rss_reta_query(struct rte_eth_dev *dev,
3330 		       struct rte_eth_rss_reta_entry64 *reta_conf,
3331 		       uint16_t reta_size)
3332 {
3333 	uint8_t i, j, mask;
3334 	uint32_t reta;
3335 	uint16_t idx, shift;
3336 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3337 
3338 	if (reta_size != ETH_RSS_RETA_SIZE_128) {
3339 		PMD_DRV_LOG(ERR, "The size of hash lookup table configured "
3340 			"(%d) doesn't match the number hardware can supported "
3341 			"(%d)\n", reta_size, ETH_RSS_RETA_SIZE_128);
3342 		return -EINVAL;
3343 	}
3344 
3345 	for (i = 0; i < reta_size; i += IGB_4_BIT_WIDTH) {
3346 		idx = i / RTE_RETA_GROUP_SIZE;
3347 		shift = i % RTE_RETA_GROUP_SIZE;
3348 		mask = (uint8_t)((reta_conf[idx].mask >> shift) &
3349 						IGB_4_BIT_MASK);
3350 		if (!mask)
3351 			continue;
3352 		reta = E1000_READ_REG(hw, E1000_RETA(i >> 2));
3353 		for (j = 0; j < IGB_4_BIT_WIDTH; j++) {
3354 			if (mask & (0x1 << j))
3355 				reta_conf[idx].reta[shift + j] =
3356 					((reta >> (CHAR_BIT * j)) &
3357 						IGB_8_BIT_MASK);
3358 		}
3359 	}
3360 
3361 	return 0;
3362 }
3363 
3364 #define MAC_TYPE_FILTER_SUP(type)    do {\
3365 	if ((type) != e1000_82580 && (type) != e1000_i350 &&\
3366 		(type) != e1000_82576)\
3367 		return -ENOTSUP;\
3368 } while (0)
3369 
3370 static int
3371 eth_igb_syn_filter_set(struct rte_eth_dev *dev,
3372 			struct rte_eth_syn_filter *filter,
3373 			bool add)
3374 {
3375 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3376 	uint32_t synqf, rfctl;
3377 
3378 	if (filter->queue >= IGB_MAX_RX_QUEUE_NUM)
3379 		return -EINVAL;
3380 
3381 	synqf = E1000_READ_REG(hw, E1000_SYNQF(0));
3382 
3383 	if (add) {
3384 		if (synqf & E1000_SYN_FILTER_ENABLE)
3385 			return -EINVAL;
3386 
3387 		synqf = (uint32_t)(((filter->queue << E1000_SYN_FILTER_QUEUE_SHIFT) &
3388 			E1000_SYN_FILTER_QUEUE) | E1000_SYN_FILTER_ENABLE);
3389 
3390 		rfctl = E1000_READ_REG(hw, E1000_RFCTL);
3391 		if (filter->hig_pri)
3392 			rfctl |= E1000_RFCTL_SYNQFP;
3393 		else
3394 			rfctl &= ~E1000_RFCTL_SYNQFP;
3395 
3396 		E1000_WRITE_REG(hw, E1000_RFCTL, rfctl);
3397 	} else {
3398 		if (!(synqf & E1000_SYN_FILTER_ENABLE))
3399 			return -ENOENT;
3400 		synqf = 0;
3401 	}
3402 
3403 	E1000_WRITE_REG(hw, E1000_SYNQF(0), synqf);
3404 	E1000_WRITE_FLUSH(hw);
3405 	return 0;
3406 }
3407 
3408 static int
3409 eth_igb_syn_filter_get(struct rte_eth_dev *dev,
3410 			struct rte_eth_syn_filter *filter)
3411 {
3412 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3413 	uint32_t synqf, rfctl;
3414 
3415 	synqf = E1000_READ_REG(hw, E1000_SYNQF(0));
3416 	if (synqf & E1000_SYN_FILTER_ENABLE) {
3417 		rfctl = E1000_READ_REG(hw, E1000_RFCTL);
3418 		filter->hig_pri = (rfctl & E1000_RFCTL_SYNQFP) ? 1 : 0;
3419 		filter->queue = (uint8_t)((synqf & E1000_SYN_FILTER_QUEUE) >>
3420 				E1000_SYN_FILTER_QUEUE_SHIFT);
3421 		return 0;
3422 	}
3423 
3424 	return -ENOENT;
3425 }
3426 
3427 static int
3428 eth_igb_syn_filter_handle(struct rte_eth_dev *dev,
3429 			enum rte_filter_op filter_op,
3430 			void *arg)
3431 {
3432 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3433 	int ret;
3434 
3435 	MAC_TYPE_FILTER_SUP(hw->mac.type);
3436 
3437 	if (filter_op == RTE_ETH_FILTER_NOP)
3438 		return 0;
3439 
3440 	if (arg == NULL) {
3441 		PMD_DRV_LOG(ERR, "arg shouldn't be NULL for operation %u",
3442 			    filter_op);
3443 		return -EINVAL;
3444 	}
3445 
3446 	switch (filter_op) {
3447 	case RTE_ETH_FILTER_ADD:
3448 		ret = eth_igb_syn_filter_set(dev,
3449 				(struct rte_eth_syn_filter *)arg,
3450 				TRUE);
3451 		break;
3452 	case RTE_ETH_FILTER_DELETE:
3453 		ret = eth_igb_syn_filter_set(dev,
3454 				(struct rte_eth_syn_filter *)arg,
3455 				FALSE);
3456 		break;
3457 	case RTE_ETH_FILTER_GET:
3458 		ret = eth_igb_syn_filter_get(dev,
3459 				(struct rte_eth_syn_filter *)arg);
3460 		break;
3461 	default:
3462 		PMD_DRV_LOG(ERR, "unsupported operation %u\n", filter_op);
3463 		ret = -EINVAL;
3464 		break;
3465 	}
3466 
3467 	return ret;
3468 }
3469 
3470 #define MAC_TYPE_FILTER_SUP_EXT(type)    do {\
3471 	if ((type) != e1000_82580 && (type) != e1000_i350)\
3472 		return -ENOSYS; \
3473 } while (0)
3474 
3475 /* translate elements in struct rte_eth_ntuple_filter to struct e1000_2tuple_filter_info*/
3476 static inline int
3477 ntuple_filter_to_2tuple(struct rte_eth_ntuple_filter *filter,
3478 			struct e1000_2tuple_filter_info *filter_info)
3479 {
3480 	if (filter->queue >= IGB_MAX_RX_QUEUE_NUM)
3481 		return -EINVAL;
3482 	if (filter->priority > E1000_2TUPLE_MAX_PRI)
3483 		return -EINVAL;  /* filter index is out of range. */
3484 	if (filter->tcp_flags > TCP_FLAG_ALL)
3485 		return -EINVAL;  /* flags is invalid. */
3486 
3487 	switch (filter->dst_port_mask) {
3488 	case UINT16_MAX:
3489 		filter_info->dst_port_mask = 0;
3490 		filter_info->dst_port = filter->dst_port;
3491 		break;
3492 	case 0:
3493 		filter_info->dst_port_mask = 1;
3494 		break;
3495 	default:
3496 		PMD_DRV_LOG(ERR, "invalid dst_port mask.");
3497 		return -EINVAL;
3498 	}
3499 
3500 	switch (filter->proto_mask) {
3501 	case UINT8_MAX:
3502 		filter_info->proto_mask = 0;
3503 		filter_info->proto = filter->proto;
3504 		break;
3505 	case 0:
3506 		filter_info->proto_mask = 1;
3507 		break;
3508 	default:
3509 		PMD_DRV_LOG(ERR, "invalid protocol mask.");
3510 		return -EINVAL;
3511 	}
3512 
3513 	filter_info->priority = (uint8_t)filter->priority;
3514 	if (filter->flags & RTE_NTUPLE_FLAGS_TCP_FLAG)
3515 		filter_info->tcp_flags = filter->tcp_flags;
3516 	else
3517 		filter_info->tcp_flags = 0;
3518 
3519 	return 0;
3520 }
3521 
3522 static inline struct e1000_2tuple_filter *
3523 igb_2tuple_filter_lookup(struct e1000_2tuple_filter_list *filter_list,
3524 			struct e1000_2tuple_filter_info *key)
3525 {
3526 	struct e1000_2tuple_filter *it;
3527 
3528 	TAILQ_FOREACH(it, filter_list, entries) {
3529 		if (memcmp(key, &it->filter_info,
3530 			sizeof(struct e1000_2tuple_filter_info)) == 0) {
3531 			return it;
3532 		}
3533 	}
3534 	return NULL;
3535 }
3536 
3537 /*
3538  * igb_add_2tuple_filter - add a 2tuple filter
3539  *
3540  * @param
3541  * dev: Pointer to struct rte_eth_dev.
3542  * ntuple_filter: ponter to the filter that will be added.
3543  *
3544  * @return
3545  *    - On success, zero.
3546  *    - On failure, a negative value.
3547  */
3548 static int
3549 igb_add_2tuple_filter(struct rte_eth_dev *dev,
3550 			struct rte_eth_ntuple_filter *ntuple_filter)
3551 {
3552 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3553 	struct e1000_filter_info *filter_info =
3554 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
3555 	struct e1000_2tuple_filter *filter;
3556 	uint32_t ttqf = E1000_TTQF_DISABLE_MASK;
3557 	uint32_t imir, imir_ext = E1000_IMIREXT_SIZE_BP;
3558 	int i, ret;
3559 
3560 	filter = rte_zmalloc("e1000_2tuple_filter",
3561 			sizeof(struct e1000_2tuple_filter), 0);
3562 	if (filter == NULL)
3563 		return -ENOMEM;
3564 
3565 	ret = ntuple_filter_to_2tuple(ntuple_filter,
3566 				      &filter->filter_info);
3567 	if (ret < 0) {
3568 		rte_free(filter);
3569 		return ret;
3570 	}
3571 	if (igb_2tuple_filter_lookup(&filter_info->twotuple_list,
3572 					 &filter->filter_info) != NULL) {
3573 		PMD_DRV_LOG(ERR, "filter exists.");
3574 		rte_free(filter);
3575 		return -EEXIST;
3576 	}
3577 	filter->queue = ntuple_filter->queue;
3578 
3579 	/*
3580 	 * look for an unused 2tuple filter index,
3581 	 * and insert the filter to list.
3582 	 */
3583 	for (i = 0; i < E1000_MAX_TTQF_FILTERS; i++) {
3584 		if (!(filter_info->twotuple_mask & (1 << i))) {
3585 			filter_info->twotuple_mask |= 1 << i;
3586 			filter->index = i;
3587 			TAILQ_INSERT_TAIL(&filter_info->twotuple_list,
3588 					  filter,
3589 					  entries);
3590 			break;
3591 		}
3592 	}
3593 	if (i >= E1000_MAX_TTQF_FILTERS) {
3594 		PMD_DRV_LOG(ERR, "2tuple filters are full.");
3595 		rte_free(filter);
3596 		return -ENOSYS;
3597 	}
3598 
3599 	imir = (uint32_t)(filter->filter_info.dst_port & E1000_IMIR_DSTPORT);
3600 	if (filter->filter_info.dst_port_mask == 1) /* 1b means not compare. */
3601 		imir |= E1000_IMIR_PORT_BP;
3602 	else
3603 		imir &= ~E1000_IMIR_PORT_BP;
3604 
3605 	imir |= filter->filter_info.priority << E1000_IMIR_PRIORITY_SHIFT;
3606 
3607 	ttqf |= E1000_TTQF_QUEUE_ENABLE;
3608 	ttqf |= (uint32_t)(filter->queue << E1000_TTQF_QUEUE_SHIFT);
3609 	ttqf |= (uint32_t)(filter->filter_info.proto & E1000_TTQF_PROTOCOL_MASK);
3610 	if (filter->filter_info.proto_mask == 0)
3611 		ttqf &= ~E1000_TTQF_MASK_ENABLE;
3612 
3613 	/* tcp flags bits setting. */
3614 	if (filter->filter_info.tcp_flags & TCP_FLAG_ALL) {
3615 		if (filter->filter_info.tcp_flags & TCP_URG_FLAG)
3616 			imir_ext |= E1000_IMIREXT_CTRL_URG;
3617 		if (filter->filter_info.tcp_flags & TCP_ACK_FLAG)
3618 			imir_ext |= E1000_IMIREXT_CTRL_ACK;
3619 		if (filter->filter_info.tcp_flags & TCP_PSH_FLAG)
3620 			imir_ext |= E1000_IMIREXT_CTRL_PSH;
3621 		if (filter->filter_info.tcp_flags & TCP_RST_FLAG)
3622 			imir_ext |= E1000_IMIREXT_CTRL_RST;
3623 		if (filter->filter_info.tcp_flags & TCP_SYN_FLAG)
3624 			imir_ext |= E1000_IMIREXT_CTRL_SYN;
3625 		if (filter->filter_info.tcp_flags & TCP_FIN_FLAG)
3626 			imir_ext |= E1000_IMIREXT_CTRL_FIN;
3627 	} else
3628 		imir_ext |= E1000_IMIREXT_CTRL_BP;
3629 	E1000_WRITE_REG(hw, E1000_IMIR(i), imir);
3630 	E1000_WRITE_REG(hw, E1000_TTQF(i), ttqf);
3631 	E1000_WRITE_REG(hw, E1000_IMIREXT(i), imir_ext);
3632 	return 0;
3633 }
3634 
3635 /*
3636  * igb_remove_2tuple_filter - remove a 2tuple filter
3637  *
3638  * @param
3639  * dev: Pointer to struct rte_eth_dev.
3640  * ntuple_filter: ponter to the filter that will be removed.
3641  *
3642  * @return
3643  *    - On success, zero.
3644  *    - On failure, a negative value.
3645  */
3646 static int
3647 igb_remove_2tuple_filter(struct rte_eth_dev *dev,
3648 			struct rte_eth_ntuple_filter *ntuple_filter)
3649 {
3650 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3651 	struct e1000_filter_info *filter_info =
3652 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
3653 	struct e1000_2tuple_filter_info filter_2tuple;
3654 	struct e1000_2tuple_filter *filter;
3655 	int ret;
3656 
3657 	memset(&filter_2tuple, 0, sizeof(struct e1000_2tuple_filter_info));
3658 	ret = ntuple_filter_to_2tuple(ntuple_filter,
3659 				      &filter_2tuple);
3660 	if (ret < 0)
3661 		return ret;
3662 
3663 	filter = igb_2tuple_filter_lookup(&filter_info->twotuple_list,
3664 					 &filter_2tuple);
3665 	if (filter == NULL) {
3666 		PMD_DRV_LOG(ERR, "filter doesn't exist.");
3667 		return -ENOENT;
3668 	}
3669 
3670 	filter_info->twotuple_mask &= ~(1 << filter->index);
3671 	TAILQ_REMOVE(&filter_info->twotuple_list, filter, entries);
3672 	rte_free(filter);
3673 
3674 	E1000_WRITE_REG(hw, E1000_TTQF(filter->index), E1000_TTQF_DISABLE_MASK);
3675 	E1000_WRITE_REG(hw, E1000_IMIR(filter->index), 0);
3676 	E1000_WRITE_REG(hw, E1000_IMIREXT(filter->index), 0);
3677 	return 0;
3678 }
3679 
3680 static inline struct e1000_flex_filter *
3681 eth_igb_flex_filter_lookup(struct e1000_flex_filter_list *filter_list,
3682 			struct e1000_flex_filter_info *key)
3683 {
3684 	struct e1000_flex_filter *it;
3685 
3686 	TAILQ_FOREACH(it, filter_list, entries) {
3687 		if (memcmp(key, &it->filter_info,
3688 			sizeof(struct e1000_flex_filter_info)) == 0)
3689 			return it;
3690 	}
3691 
3692 	return NULL;
3693 }
3694 
3695 static int
3696 eth_igb_add_del_flex_filter(struct rte_eth_dev *dev,
3697 			struct rte_eth_flex_filter *filter,
3698 			bool add)
3699 {
3700 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3701 	struct e1000_filter_info *filter_info =
3702 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
3703 	struct e1000_flex_filter *flex_filter, *it;
3704 	uint32_t wufc, queueing, mask;
3705 	uint32_t reg_off;
3706 	uint8_t shift, i, j = 0;
3707 
3708 	flex_filter = rte_zmalloc("e1000_flex_filter",
3709 			sizeof(struct e1000_flex_filter), 0);
3710 	if (flex_filter == NULL)
3711 		return -ENOMEM;
3712 
3713 	flex_filter->filter_info.len = filter->len;
3714 	flex_filter->filter_info.priority = filter->priority;
3715 	memcpy(flex_filter->filter_info.dwords, filter->bytes, filter->len);
3716 	for (i = 0; i < RTE_ALIGN(filter->len, CHAR_BIT) / CHAR_BIT; i++) {
3717 		mask = 0;
3718 		/* reverse bits in flex filter's mask*/
3719 		for (shift = 0; shift < CHAR_BIT; shift++) {
3720 			if (filter->mask[i] & (0x01 << shift))
3721 				mask |= (0x80 >> shift);
3722 		}
3723 		flex_filter->filter_info.mask[i] = mask;
3724 	}
3725 
3726 	wufc = E1000_READ_REG(hw, E1000_WUFC);
3727 	if (flex_filter->index < E1000_MAX_FHFT)
3728 		reg_off = E1000_FHFT(flex_filter->index);
3729 	else
3730 		reg_off = E1000_FHFT_EXT(flex_filter->index - E1000_MAX_FHFT);
3731 
3732 	if (add) {
3733 		if (eth_igb_flex_filter_lookup(&filter_info->flex_list,
3734 				&flex_filter->filter_info) != NULL) {
3735 			PMD_DRV_LOG(ERR, "filter exists.");
3736 			rte_free(flex_filter);
3737 			return -EEXIST;
3738 		}
3739 		flex_filter->queue = filter->queue;
3740 		/*
3741 		 * look for an unused flex filter index
3742 		 * and insert the filter into the list.
3743 		 */
3744 		for (i = 0; i < E1000_MAX_FLEX_FILTERS; i++) {
3745 			if (!(filter_info->flex_mask & (1 << i))) {
3746 				filter_info->flex_mask |= 1 << i;
3747 				flex_filter->index = i;
3748 				TAILQ_INSERT_TAIL(&filter_info->flex_list,
3749 					flex_filter,
3750 					entries);
3751 				break;
3752 			}
3753 		}
3754 		if (i >= E1000_MAX_FLEX_FILTERS) {
3755 			PMD_DRV_LOG(ERR, "flex filters are full.");
3756 			rte_free(flex_filter);
3757 			return -ENOSYS;
3758 		}
3759 
3760 		E1000_WRITE_REG(hw, E1000_WUFC, wufc | E1000_WUFC_FLEX_HQ |
3761 				(E1000_WUFC_FLX0 << flex_filter->index));
3762 		queueing = filter->len |
3763 			(filter->queue << E1000_FHFT_QUEUEING_QUEUE_SHIFT) |
3764 			(filter->priority << E1000_FHFT_QUEUEING_PRIO_SHIFT);
3765 		E1000_WRITE_REG(hw, reg_off + E1000_FHFT_QUEUEING_OFFSET,
3766 				queueing);
3767 		for (i = 0; i < E1000_FLEX_FILTERS_MASK_SIZE; i++) {
3768 			E1000_WRITE_REG(hw, reg_off,
3769 					flex_filter->filter_info.dwords[j]);
3770 			reg_off += sizeof(uint32_t);
3771 			E1000_WRITE_REG(hw, reg_off,
3772 					flex_filter->filter_info.dwords[++j]);
3773 			reg_off += sizeof(uint32_t);
3774 			E1000_WRITE_REG(hw, reg_off,
3775 				(uint32_t)flex_filter->filter_info.mask[i]);
3776 			reg_off += sizeof(uint32_t) * 2;
3777 			++j;
3778 		}
3779 	} else {
3780 		it = eth_igb_flex_filter_lookup(&filter_info->flex_list,
3781 				&flex_filter->filter_info);
3782 		if (it == NULL) {
3783 			PMD_DRV_LOG(ERR, "filter doesn't exist.");
3784 			rte_free(flex_filter);
3785 			return -ENOENT;
3786 		}
3787 
3788 		for (i = 0; i < E1000_FHFT_SIZE_IN_DWD; i++)
3789 			E1000_WRITE_REG(hw, reg_off + i * sizeof(uint32_t), 0);
3790 		E1000_WRITE_REG(hw, E1000_WUFC, wufc &
3791 			(~(E1000_WUFC_FLX0 << it->index)));
3792 
3793 		filter_info->flex_mask &= ~(1 << it->index);
3794 		TAILQ_REMOVE(&filter_info->flex_list, it, entries);
3795 		rte_free(it);
3796 		rte_free(flex_filter);
3797 	}
3798 
3799 	return 0;
3800 }
3801 
3802 static int
3803 eth_igb_get_flex_filter(struct rte_eth_dev *dev,
3804 			struct rte_eth_flex_filter *filter)
3805 {
3806 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3807 	struct e1000_filter_info *filter_info =
3808 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
3809 	struct e1000_flex_filter flex_filter, *it;
3810 	uint32_t wufc, queueing, wufc_en = 0;
3811 
3812 	memset(&flex_filter, 0, sizeof(struct e1000_flex_filter));
3813 	flex_filter.filter_info.len = filter->len;
3814 	flex_filter.filter_info.priority = filter->priority;
3815 	memcpy(flex_filter.filter_info.dwords, filter->bytes, filter->len);
3816 	memcpy(flex_filter.filter_info.mask, filter->mask,
3817 			RTE_ALIGN(filter->len, sizeof(char)) / sizeof(char));
3818 
3819 	it = eth_igb_flex_filter_lookup(&filter_info->flex_list,
3820 				&flex_filter.filter_info);
3821 	if (it == NULL) {
3822 		PMD_DRV_LOG(ERR, "filter doesn't exist.");
3823 		return -ENOENT;
3824 	}
3825 
3826 	wufc = E1000_READ_REG(hw, E1000_WUFC);
3827 	wufc_en = E1000_WUFC_FLEX_HQ | (E1000_WUFC_FLX0 << it->index);
3828 
3829 	if ((wufc & wufc_en) == wufc_en) {
3830 		uint32_t reg_off = 0;
3831 		if (it->index < E1000_MAX_FHFT)
3832 			reg_off = E1000_FHFT(it->index);
3833 		else
3834 			reg_off = E1000_FHFT_EXT(it->index - E1000_MAX_FHFT);
3835 
3836 		queueing = E1000_READ_REG(hw,
3837 				reg_off + E1000_FHFT_QUEUEING_OFFSET);
3838 		filter->len = queueing & E1000_FHFT_QUEUEING_LEN;
3839 		filter->priority = (queueing & E1000_FHFT_QUEUEING_PRIO) >>
3840 			E1000_FHFT_QUEUEING_PRIO_SHIFT;
3841 		filter->queue = (queueing & E1000_FHFT_QUEUEING_QUEUE) >>
3842 			E1000_FHFT_QUEUEING_QUEUE_SHIFT;
3843 		return 0;
3844 	}
3845 	return -ENOENT;
3846 }
3847 
3848 static int
3849 eth_igb_flex_filter_handle(struct rte_eth_dev *dev,
3850 			enum rte_filter_op filter_op,
3851 			void *arg)
3852 {
3853 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3854 	struct rte_eth_flex_filter *filter;
3855 	int ret = 0;
3856 
3857 	MAC_TYPE_FILTER_SUP_EXT(hw->mac.type);
3858 
3859 	if (filter_op == RTE_ETH_FILTER_NOP)
3860 		return ret;
3861 
3862 	if (arg == NULL) {
3863 		PMD_DRV_LOG(ERR, "arg shouldn't be NULL for operation %u",
3864 			    filter_op);
3865 		return -EINVAL;
3866 	}
3867 
3868 	filter = (struct rte_eth_flex_filter *)arg;
3869 	if (filter->len == 0 || filter->len > E1000_MAX_FLEX_FILTER_LEN
3870 	    || filter->len % sizeof(uint64_t) != 0) {
3871 		PMD_DRV_LOG(ERR, "filter's length is out of range");
3872 		return -EINVAL;
3873 	}
3874 	if (filter->priority > E1000_MAX_FLEX_FILTER_PRI) {
3875 		PMD_DRV_LOG(ERR, "filter's priority is out of range");
3876 		return -EINVAL;
3877 	}
3878 
3879 	switch (filter_op) {
3880 	case RTE_ETH_FILTER_ADD:
3881 		ret = eth_igb_add_del_flex_filter(dev, filter, TRUE);
3882 		break;
3883 	case RTE_ETH_FILTER_DELETE:
3884 		ret = eth_igb_add_del_flex_filter(dev, filter, FALSE);
3885 		break;
3886 	case RTE_ETH_FILTER_GET:
3887 		ret = eth_igb_get_flex_filter(dev, filter);
3888 		break;
3889 	default:
3890 		PMD_DRV_LOG(ERR, "unsupported operation %u", filter_op);
3891 		ret = -EINVAL;
3892 		break;
3893 	}
3894 
3895 	return ret;
3896 }
3897 
3898 /* translate elements in struct rte_eth_ntuple_filter to struct e1000_5tuple_filter_info*/
3899 static inline int
3900 ntuple_filter_to_5tuple_82576(struct rte_eth_ntuple_filter *filter,
3901 			struct e1000_5tuple_filter_info *filter_info)
3902 {
3903 	if (filter->queue >= IGB_MAX_RX_QUEUE_NUM_82576)
3904 		return -EINVAL;
3905 	if (filter->priority > E1000_2TUPLE_MAX_PRI)
3906 		return -EINVAL;  /* filter index is out of range. */
3907 	if (filter->tcp_flags > TCP_FLAG_ALL)
3908 		return -EINVAL;  /* flags is invalid. */
3909 
3910 	switch (filter->dst_ip_mask) {
3911 	case UINT32_MAX:
3912 		filter_info->dst_ip_mask = 0;
3913 		filter_info->dst_ip = filter->dst_ip;
3914 		break;
3915 	case 0:
3916 		filter_info->dst_ip_mask = 1;
3917 		break;
3918 	default:
3919 		PMD_DRV_LOG(ERR, "invalid dst_ip mask.");
3920 		return -EINVAL;
3921 	}
3922 
3923 	switch (filter->src_ip_mask) {
3924 	case UINT32_MAX:
3925 		filter_info->src_ip_mask = 0;
3926 		filter_info->src_ip = filter->src_ip;
3927 		break;
3928 	case 0:
3929 		filter_info->src_ip_mask = 1;
3930 		break;
3931 	default:
3932 		PMD_DRV_LOG(ERR, "invalid src_ip mask.");
3933 		return -EINVAL;
3934 	}
3935 
3936 	switch (filter->dst_port_mask) {
3937 	case UINT16_MAX:
3938 		filter_info->dst_port_mask = 0;
3939 		filter_info->dst_port = filter->dst_port;
3940 		break;
3941 	case 0:
3942 		filter_info->dst_port_mask = 1;
3943 		break;
3944 	default:
3945 		PMD_DRV_LOG(ERR, "invalid dst_port mask.");
3946 		return -EINVAL;
3947 	}
3948 
3949 	switch (filter->src_port_mask) {
3950 	case UINT16_MAX:
3951 		filter_info->src_port_mask = 0;
3952 		filter_info->src_port = filter->src_port;
3953 		break;
3954 	case 0:
3955 		filter_info->src_port_mask = 1;
3956 		break;
3957 	default:
3958 		PMD_DRV_LOG(ERR, "invalid src_port mask.");
3959 		return -EINVAL;
3960 	}
3961 
3962 	switch (filter->proto_mask) {
3963 	case UINT8_MAX:
3964 		filter_info->proto_mask = 0;
3965 		filter_info->proto = filter->proto;
3966 		break;
3967 	case 0:
3968 		filter_info->proto_mask = 1;
3969 		break;
3970 	default:
3971 		PMD_DRV_LOG(ERR, "invalid protocol mask.");
3972 		return -EINVAL;
3973 	}
3974 
3975 	filter_info->priority = (uint8_t)filter->priority;
3976 	if (filter->flags & RTE_NTUPLE_FLAGS_TCP_FLAG)
3977 		filter_info->tcp_flags = filter->tcp_flags;
3978 	else
3979 		filter_info->tcp_flags = 0;
3980 
3981 	return 0;
3982 }
3983 
3984 static inline struct e1000_5tuple_filter *
3985 igb_5tuple_filter_lookup_82576(struct e1000_5tuple_filter_list *filter_list,
3986 			struct e1000_5tuple_filter_info *key)
3987 {
3988 	struct e1000_5tuple_filter *it;
3989 
3990 	TAILQ_FOREACH(it, filter_list, entries) {
3991 		if (memcmp(key, &it->filter_info,
3992 			sizeof(struct e1000_5tuple_filter_info)) == 0) {
3993 			return it;
3994 		}
3995 	}
3996 	return NULL;
3997 }
3998 
3999 /*
4000  * igb_add_5tuple_filter_82576 - add a 5tuple filter
4001  *
4002  * @param
4003  * dev: Pointer to struct rte_eth_dev.
4004  * ntuple_filter: ponter to the filter that will be added.
4005  *
4006  * @return
4007  *    - On success, zero.
4008  *    - On failure, a negative value.
4009  */
4010 static int
4011 igb_add_5tuple_filter_82576(struct rte_eth_dev *dev,
4012 			struct rte_eth_ntuple_filter *ntuple_filter)
4013 {
4014 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4015 	struct e1000_filter_info *filter_info =
4016 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
4017 	struct e1000_5tuple_filter *filter;
4018 	uint32_t ftqf = E1000_FTQF_VF_BP | E1000_FTQF_MASK;
4019 	uint32_t spqf, imir, imir_ext = E1000_IMIREXT_SIZE_BP;
4020 	uint8_t i;
4021 	int ret;
4022 
4023 	filter = rte_zmalloc("e1000_5tuple_filter",
4024 			sizeof(struct e1000_5tuple_filter), 0);
4025 	if (filter == NULL)
4026 		return -ENOMEM;
4027 
4028 	ret = ntuple_filter_to_5tuple_82576(ntuple_filter,
4029 					    &filter->filter_info);
4030 	if (ret < 0) {
4031 		rte_free(filter);
4032 		return ret;
4033 	}
4034 
4035 	if (igb_5tuple_filter_lookup_82576(&filter_info->fivetuple_list,
4036 					 &filter->filter_info) != NULL) {
4037 		PMD_DRV_LOG(ERR, "filter exists.");
4038 		rte_free(filter);
4039 		return -EEXIST;
4040 	}
4041 	filter->queue = ntuple_filter->queue;
4042 
4043 	/*
4044 	 * look for an unused 5tuple filter index,
4045 	 * and insert the filter to list.
4046 	 */
4047 	for (i = 0; i < E1000_MAX_FTQF_FILTERS; i++) {
4048 		if (!(filter_info->fivetuple_mask & (1 << i))) {
4049 			filter_info->fivetuple_mask |= 1 << i;
4050 			filter->index = i;
4051 			TAILQ_INSERT_TAIL(&filter_info->fivetuple_list,
4052 					  filter,
4053 					  entries);
4054 			break;
4055 		}
4056 	}
4057 	if (i >= E1000_MAX_FTQF_FILTERS) {
4058 		PMD_DRV_LOG(ERR, "5tuple filters are full.");
4059 		rte_free(filter);
4060 		return -ENOSYS;
4061 	}
4062 
4063 	ftqf |= filter->filter_info.proto & E1000_FTQF_PROTOCOL_MASK;
4064 	if (filter->filter_info.src_ip_mask == 0) /* 0b means compare. */
4065 		ftqf &= ~E1000_FTQF_MASK_SOURCE_ADDR_BP;
4066 	if (filter->filter_info.dst_ip_mask == 0)
4067 		ftqf &= ~E1000_FTQF_MASK_DEST_ADDR_BP;
4068 	if (filter->filter_info.src_port_mask == 0)
4069 		ftqf &= ~E1000_FTQF_MASK_SOURCE_PORT_BP;
4070 	if (filter->filter_info.proto_mask == 0)
4071 		ftqf &= ~E1000_FTQF_MASK_PROTO_BP;
4072 	ftqf |= (filter->queue << E1000_FTQF_QUEUE_SHIFT) &
4073 		E1000_FTQF_QUEUE_MASK;
4074 	ftqf |= E1000_FTQF_QUEUE_ENABLE;
4075 	E1000_WRITE_REG(hw, E1000_FTQF(i), ftqf);
4076 	E1000_WRITE_REG(hw, E1000_DAQF(i), filter->filter_info.dst_ip);
4077 	E1000_WRITE_REG(hw, E1000_SAQF(i), filter->filter_info.src_ip);
4078 
4079 	spqf = filter->filter_info.src_port & E1000_SPQF_SRCPORT;
4080 	E1000_WRITE_REG(hw, E1000_SPQF(i), spqf);
4081 
4082 	imir = (uint32_t)(filter->filter_info.dst_port & E1000_IMIR_DSTPORT);
4083 	if (filter->filter_info.dst_port_mask == 1) /* 1b means not compare. */
4084 		imir |= E1000_IMIR_PORT_BP;
4085 	else
4086 		imir &= ~E1000_IMIR_PORT_BP;
4087 	imir |= filter->filter_info.priority << E1000_IMIR_PRIORITY_SHIFT;
4088 
4089 	/* tcp flags bits setting. */
4090 	if (filter->filter_info.tcp_flags & TCP_FLAG_ALL) {
4091 		if (filter->filter_info.tcp_flags & TCP_URG_FLAG)
4092 			imir_ext |= E1000_IMIREXT_CTRL_URG;
4093 		if (filter->filter_info.tcp_flags & TCP_ACK_FLAG)
4094 			imir_ext |= E1000_IMIREXT_CTRL_ACK;
4095 		if (filter->filter_info.tcp_flags & TCP_PSH_FLAG)
4096 			imir_ext |= E1000_IMIREXT_CTRL_PSH;
4097 		if (filter->filter_info.tcp_flags & TCP_RST_FLAG)
4098 			imir_ext |= E1000_IMIREXT_CTRL_RST;
4099 		if (filter->filter_info.tcp_flags & TCP_SYN_FLAG)
4100 			imir_ext |= E1000_IMIREXT_CTRL_SYN;
4101 		if (filter->filter_info.tcp_flags & TCP_FIN_FLAG)
4102 			imir_ext |= E1000_IMIREXT_CTRL_FIN;
4103 	} else
4104 		imir_ext |= E1000_IMIREXT_CTRL_BP;
4105 	E1000_WRITE_REG(hw, E1000_IMIR(i), imir);
4106 	E1000_WRITE_REG(hw, E1000_IMIREXT(i), imir_ext);
4107 	return 0;
4108 }
4109 
4110 /*
4111  * igb_remove_5tuple_filter_82576 - remove a 5tuple filter
4112  *
4113  * @param
4114  * dev: Pointer to struct rte_eth_dev.
4115  * ntuple_filter: ponter to the filter that will be removed.
4116  *
4117  * @return
4118  *    - On success, zero.
4119  *    - On failure, a negative value.
4120  */
4121 static int
4122 igb_remove_5tuple_filter_82576(struct rte_eth_dev *dev,
4123 				struct rte_eth_ntuple_filter *ntuple_filter)
4124 {
4125 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4126 	struct e1000_filter_info *filter_info =
4127 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
4128 	struct e1000_5tuple_filter_info filter_5tuple;
4129 	struct e1000_5tuple_filter *filter;
4130 	int ret;
4131 
4132 	memset(&filter_5tuple, 0, sizeof(struct e1000_5tuple_filter_info));
4133 	ret = ntuple_filter_to_5tuple_82576(ntuple_filter,
4134 					    &filter_5tuple);
4135 	if (ret < 0)
4136 		return ret;
4137 
4138 	filter = igb_5tuple_filter_lookup_82576(&filter_info->fivetuple_list,
4139 					 &filter_5tuple);
4140 	if (filter == NULL) {
4141 		PMD_DRV_LOG(ERR, "filter doesn't exist.");
4142 		return -ENOENT;
4143 	}
4144 
4145 	filter_info->fivetuple_mask &= ~(1 << filter->index);
4146 	TAILQ_REMOVE(&filter_info->fivetuple_list, filter, entries);
4147 	rte_free(filter);
4148 
4149 	E1000_WRITE_REG(hw, E1000_FTQF(filter->index),
4150 			E1000_FTQF_VF_BP | E1000_FTQF_MASK);
4151 	E1000_WRITE_REG(hw, E1000_DAQF(filter->index), 0);
4152 	E1000_WRITE_REG(hw, E1000_SAQF(filter->index), 0);
4153 	E1000_WRITE_REG(hw, E1000_SPQF(filter->index), 0);
4154 	E1000_WRITE_REG(hw, E1000_IMIR(filter->index), 0);
4155 	E1000_WRITE_REG(hw, E1000_IMIREXT(filter->index), 0);
4156 	return 0;
4157 }
4158 
4159 static int
4160 eth_igb_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
4161 {
4162 	uint32_t rctl;
4163 	struct e1000_hw *hw;
4164 	struct rte_eth_dev_info dev_info;
4165 	uint32_t frame_size = mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN +
4166 				     VLAN_TAG_SIZE);
4167 
4168 	hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4169 
4170 #ifdef RTE_LIBRTE_82571_SUPPORT
4171 	/* XXX: not bigger than max_rx_pktlen */
4172 	if (hw->mac.type == e1000_82571)
4173 		return -ENOTSUP;
4174 #endif
4175 	eth_igb_infos_get(dev, &dev_info);
4176 
4177 	/* check that mtu is within the allowed range */
4178 	if ((mtu < ETHER_MIN_MTU) ||
4179 	    (frame_size > dev_info.max_rx_pktlen))
4180 		return -EINVAL;
4181 
4182 	/* refuse mtu that requires the support of scattered packets when this
4183 	 * feature has not been enabled before. */
4184 	if (!dev->data->scattered_rx &&
4185 	    frame_size > dev->data->min_rx_buf_size - RTE_PKTMBUF_HEADROOM)
4186 		return -EINVAL;
4187 
4188 	rctl = E1000_READ_REG(hw, E1000_RCTL);
4189 
4190 	/* switch to jumbo mode if needed */
4191 	if (frame_size > ETHER_MAX_LEN) {
4192 		dev->data->dev_conf.rxmode.jumbo_frame = 1;
4193 		rctl |= E1000_RCTL_LPE;
4194 	} else {
4195 		dev->data->dev_conf.rxmode.jumbo_frame = 0;
4196 		rctl &= ~E1000_RCTL_LPE;
4197 	}
4198 	E1000_WRITE_REG(hw, E1000_RCTL, rctl);
4199 
4200 	/* update max frame size */
4201 	dev->data->dev_conf.rxmode.max_rx_pkt_len = frame_size;
4202 
4203 	E1000_WRITE_REG(hw, E1000_RLPML,
4204 			dev->data->dev_conf.rxmode.max_rx_pkt_len);
4205 
4206 	return 0;
4207 }
4208 
4209 /*
4210  * igb_add_del_ntuple_filter - add or delete a ntuple filter
4211  *
4212  * @param
4213  * dev: Pointer to struct rte_eth_dev.
4214  * ntuple_filter: Pointer to struct rte_eth_ntuple_filter
4215  * add: if true, add filter, if false, remove filter
4216  *
4217  * @return
4218  *    - On success, zero.
4219  *    - On failure, a negative value.
4220  */
4221 static int
4222 igb_add_del_ntuple_filter(struct rte_eth_dev *dev,
4223 			struct rte_eth_ntuple_filter *ntuple_filter,
4224 			bool add)
4225 {
4226 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4227 	int ret;
4228 
4229 	switch (ntuple_filter->flags) {
4230 	case RTE_5TUPLE_FLAGS:
4231 	case (RTE_5TUPLE_FLAGS | RTE_NTUPLE_FLAGS_TCP_FLAG):
4232 		if (hw->mac.type != e1000_82576)
4233 			return -ENOTSUP;
4234 		if (add)
4235 			ret = igb_add_5tuple_filter_82576(dev,
4236 							  ntuple_filter);
4237 		else
4238 			ret = igb_remove_5tuple_filter_82576(dev,
4239 							     ntuple_filter);
4240 		break;
4241 	case RTE_2TUPLE_FLAGS:
4242 	case (RTE_2TUPLE_FLAGS | RTE_NTUPLE_FLAGS_TCP_FLAG):
4243 		if (hw->mac.type != e1000_82580 && hw->mac.type != e1000_i350)
4244 			return -ENOTSUP;
4245 		if (add)
4246 			ret = igb_add_2tuple_filter(dev, ntuple_filter);
4247 		else
4248 			ret = igb_remove_2tuple_filter(dev, ntuple_filter);
4249 		break;
4250 	default:
4251 		ret = -EINVAL;
4252 		break;
4253 	}
4254 
4255 	return ret;
4256 }
4257 
4258 /*
4259  * igb_get_ntuple_filter - get a ntuple filter
4260  *
4261  * @param
4262  * dev: Pointer to struct rte_eth_dev.
4263  * ntuple_filter: Pointer to struct rte_eth_ntuple_filter
4264  *
4265  * @return
4266  *    - On success, zero.
4267  *    - On failure, a negative value.
4268  */
4269 static int
4270 igb_get_ntuple_filter(struct rte_eth_dev *dev,
4271 			struct rte_eth_ntuple_filter *ntuple_filter)
4272 {
4273 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4274 	struct e1000_filter_info *filter_info =
4275 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
4276 	struct e1000_5tuple_filter_info filter_5tuple;
4277 	struct e1000_2tuple_filter_info filter_2tuple;
4278 	struct e1000_5tuple_filter *p_5tuple_filter;
4279 	struct e1000_2tuple_filter *p_2tuple_filter;
4280 	int ret;
4281 
4282 	switch (ntuple_filter->flags) {
4283 	case RTE_5TUPLE_FLAGS:
4284 	case (RTE_5TUPLE_FLAGS | RTE_NTUPLE_FLAGS_TCP_FLAG):
4285 		if (hw->mac.type != e1000_82576)
4286 			return -ENOTSUP;
4287 		memset(&filter_5tuple,
4288 			0,
4289 			sizeof(struct e1000_5tuple_filter_info));
4290 		ret = ntuple_filter_to_5tuple_82576(ntuple_filter,
4291 						    &filter_5tuple);
4292 		if (ret < 0)
4293 			return ret;
4294 		p_5tuple_filter = igb_5tuple_filter_lookup_82576(
4295 					&filter_info->fivetuple_list,
4296 					&filter_5tuple);
4297 		if (p_5tuple_filter == NULL) {
4298 			PMD_DRV_LOG(ERR, "filter doesn't exist.");
4299 			return -ENOENT;
4300 		}
4301 		ntuple_filter->queue = p_5tuple_filter->queue;
4302 		break;
4303 	case RTE_2TUPLE_FLAGS:
4304 	case (RTE_2TUPLE_FLAGS | RTE_NTUPLE_FLAGS_TCP_FLAG):
4305 		if (hw->mac.type != e1000_82580 && hw->mac.type != e1000_i350)
4306 			return -ENOTSUP;
4307 		memset(&filter_2tuple,
4308 			0,
4309 			sizeof(struct e1000_2tuple_filter_info));
4310 		ret = ntuple_filter_to_2tuple(ntuple_filter, &filter_2tuple);
4311 		if (ret < 0)
4312 			return ret;
4313 		p_2tuple_filter = igb_2tuple_filter_lookup(
4314 					&filter_info->twotuple_list,
4315 					&filter_2tuple);
4316 		if (p_2tuple_filter == NULL) {
4317 			PMD_DRV_LOG(ERR, "filter doesn't exist.");
4318 			return -ENOENT;
4319 		}
4320 		ntuple_filter->queue = p_2tuple_filter->queue;
4321 		break;
4322 	default:
4323 		ret = -EINVAL;
4324 		break;
4325 	}
4326 
4327 	return 0;
4328 }
4329 
4330 /*
4331  * igb_ntuple_filter_handle - Handle operations for ntuple filter.
4332  * @dev: pointer to rte_eth_dev structure
4333  * @filter_op:operation will be taken.
4334  * @arg: a pointer to specific structure corresponding to the filter_op
4335  */
4336 static int
4337 igb_ntuple_filter_handle(struct rte_eth_dev *dev,
4338 				enum rte_filter_op filter_op,
4339 				void *arg)
4340 {
4341 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4342 	int ret;
4343 
4344 	MAC_TYPE_FILTER_SUP(hw->mac.type);
4345 
4346 	if (filter_op == RTE_ETH_FILTER_NOP)
4347 		return 0;
4348 
4349 	if (arg == NULL) {
4350 		PMD_DRV_LOG(ERR, "arg shouldn't be NULL for operation %u.",
4351 			    filter_op);
4352 		return -EINVAL;
4353 	}
4354 
4355 	switch (filter_op) {
4356 	case RTE_ETH_FILTER_ADD:
4357 		ret = igb_add_del_ntuple_filter(dev,
4358 			(struct rte_eth_ntuple_filter *)arg,
4359 			TRUE);
4360 		break;
4361 	case RTE_ETH_FILTER_DELETE:
4362 		ret = igb_add_del_ntuple_filter(dev,
4363 			(struct rte_eth_ntuple_filter *)arg,
4364 			FALSE);
4365 		break;
4366 	case RTE_ETH_FILTER_GET:
4367 		ret = igb_get_ntuple_filter(dev,
4368 			(struct rte_eth_ntuple_filter *)arg);
4369 		break;
4370 	default:
4371 		PMD_DRV_LOG(ERR, "unsupported operation %u.", filter_op);
4372 		ret = -EINVAL;
4373 		break;
4374 	}
4375 	return ret;
4376 }
4377 
4378 static inline int
4379 igb_ethertype_filter_lookup(struct e1000_filter_info *filter_info,
4380 			uint16_t ethertype)
4381 {
4382 	int i;
4383 
4384 	for (i = 0; i < E1000_MAX_ETQF_FILTERS; i++) {
4385 		if (filter_info->ethertype_filters[i] == ethertype &&
4386 		    (filter_info->ethertype_mask & (1 << i)))
4387 			return i;
4388 	}
4389 	return -1;
4390 }
4391 
4392 static inline int
4393 igb_ethertype_filter_insert(struct e1000_filter_info *filter_info,
4394 			uint16_t ethertype)
4395 {
4396 	int i;
4397 
4398 	for (i = 0; i < E1000_MAX_ETQF_FILTERS; i++) {
4399 		if (!(filter_info->ethertype_mask & (1 << i))) {
4400 			filter_info->ethertype_mask |= 1 << i;
4401 			filter_info->ethertype_filters[i] = ethertype;
4402 			return i;
4403 		}
4404 	}
4405 	return -1;
4406 }
4407 
4408 static inline int
4409 igb_ethertype_filter_remove(struct e1000_filter_info *filter_info,
4410 			uint8_t idx)
4411 {
4412 	if (idx >= E1000_MAX_ETQF_FILTERS)
4413 		return -1;
4414 	filter_info->ethertype_mask &= ~(1 << idx);
4415 	filter_info->ethertype_filters[idx] = 0;
4416 	return idx;
4417 }
4418 
4419 
4420 static int
4421 igb_add_del_ethertype_filter(struct rte_eth_dev *dev,
4422 			struct rte_eth_ethertype_filter *filter,
4423 			bool add)
4424 {
4425 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4426 	struct e1000_filter_info *filter_info =
4427 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
4428 	uint32_t etqf = 0;
4429 	int ret;
4430 
4431 	if (filter->ether_type == ETHER_TYPE_IPv4 ||
4432 		filter->ether_type == ETHER_TYPE_IPv6) {
4433 		PMD_DRV_LOG(ERR, "unsupported ether_type(0x%04x) in"
4434 			" ethertype filter.", filter->ether_type);
4435 		return -EINVAL;
4436 	}
4437 
4438 	if (filter->flags & RTE_ETHTYPE_FLAGS_MAC) {
4439 		PMD_DRV_LOG(ERR, "mac compare is unsupported.");
4440 		return -EINVAL;
4441 	}
4442 	if (filter->flags & RTE_ETHTYPE_FLAGS_DROP) {
4443 		PMD_DRV_LOG(ERR, "drop option is unsupported.");
4444 		return -EINVAL;
4445 	}
4446 
4447 	ret = igb_ethertype_filter_lookup(filter_info, filter->ether_type);
4448 	if (ret >= 0 && add) {
4449 		PMD_DRV_LOG(ERR, "ethertype (0x%04x) filter exists.",
4450 			    filter->ether_type);
4451 		return -EEXIST;
4452 	}
4453 	if (ret < 0 && !add) {
4454 		PMD_DRV_LOG(ERR, "ethertype (0x%04x) filter doesn't exist.",
4455 			    filter->ether_type);
4456 		return -ENOENT;
4457 	}
4458 
4459 	if (add) {
4460 		ret = igb_ethertype_filter_insert(filter_info,
4461 			filter->ether_type);
4462 		if (ret < 0) {
4463 			PMD_DRV_LOG(ERR, "ethertype filters are full.");
4464 			return -ENOSYS;
4465 		}
4466 
4467 		etqf |= E1000_ETQF_FILTER_ENABLE | E1000_ETQF_QUEUE_ENABLE;
4468 		etqf |= (uint32_t)(filter->ether_type & E1000_ETQF_ETHERTYPE);
4469 		etqf |= filter->queue << E1000_ETQF_QUEUE_SHIFT;
4470 	} else {
4471 		ret = igb_ethertype_filter_remove(filter_info, (uint8_t)ret);
4472 		if (ret < 0)
4473 			return -ENOSYS;
4474 	}
4475 	E1000_WRITE_REG(hw, E1000_ETQF(ret), etqf);
4476 	E1000_WRITE_FLUSH(hw);
4477 
4478 	return 0;
4479 }
4480 
4481 static int
4482 igb_get_ethertype_filter(struct rte_eth_dev *dev,
4483 			struct rte_eth_ethertype_filter *filter)
4484 {
4485 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4486 	struct e1000_filter_info *filter_info =
4487 		E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
4488 	uint32_t etqf;
4489 	int ret;
4490 
4491 	ret = igb_ethertype_filter_lookup(filter_info, filter->ether_type);
4492 	if (ret < 0) {
4493 		PMD_DRV_LOG(ERR, "ethertype (0x%04x) filter doesn't exist.",
4494 			    filter->ether_type);
4495 		return -ENOENT;
4496 	}
4497 
4498 	etqf = E1000_READ_REG(hw, E1000_ETQF(ret));
4499 	if (etqf & E1000_ETQF_FILTER_ENABLE) {
4500 		filter->ether_type = etqf & E1000_ETQF_ETHERTYPE;
4501 		filter->flags = 0;
4502 		filter->queue = (etqf & E1000_ETQF_QUEUE) >>
4503 				E1000_ETQF_QUEUE_SHIFT;
4504 		return 0;
4505 	}
4506 
4507 	return -ENOENT;
4508 }
4509 
4510 /*
4511  * igb_ethertype_filter_handle - Handle operations for ethertype filter.
4512  * @dev: pointer to rte_eth_dev structure
4513  * @filter_op:operation will be taken.
4514  * @arg: a pointer to specific structure corresponding to the filter_op
4515  */
4516 static int
4517 igb_ethertype_filter_handle(struct rte_eth_dev *dev,
4518 				enum rte_filter_op filter_op,
4519 				void *arg)
4520 {
4521 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4522 	int ret;
4523 
4524 	MAC_TYPE_FILTER_SUP(hw->mac.type);
4525 
4526 	if (filter_op == RTE_ETH_FILTER_NOP)
4527 		return 0;
4528 
4529 	if (arg == NULL) {
4530 		PMD_DRV_LOG(ERR, "arg shouldn't be NULL for operation %u.",
4531 			    filter_op);
4532 		return -EINVAL;
4533 	}
4534 
4535 	switch (filter_op) {
4536 	case RTE_ETH_FILTER_ADD:
4537 		ret = igb_add_del_ethertype_filter(dev,
4538 			(struct rte_eth_ethertype_filter *)arg,
4539 			TRUE);
4540 		break;
4541 	case RTE_ETH_FILTER_DELETE:
4542 		ret = igb_add_del_ethertype_filter(dev,
4543 			(struct rte_eth_ethertype_filter *)arg,
4544 			FALSE);
4545 		break;
4546 	case RTE_ETH_FILTER_GET:
4547 		ret = igb_get_ethertype_filter(dev,
4548 			(struct rte_eth_ethertype_filter *)arg);
4549 		break;
4550 	default:
4551 		PMD_DRV_LOG(ERR, "unsupported operation %u.", filter_op);
4552 		ret = -EINVAL;
4553 		break;
4554 	}
4555 	return ret;
4556 }
4557 
4558 static int
4559 eth_igb_filter_ctrl(struct rte_eth_dev *dev,
4560 		     enum rte_filter_type filter_type,
4561 		     enum rte_filter_op filter_op,
4562 		     void *arg)
4563 {
4564 	int ret = -EINVAL;
4565 
4566 	switch (filter_type) {
4567 	case RTE_ETH_FILTER_NTUPLE:
4568 		ret = igb_ntuple_filter_handle(dev, filter_op, arg);
4569 		break;
4570 	case RTE_ETH_FILTER_ETHERTYPE:
4571 		ret = igb_ethertype_filter_handle(dev, filter_op, arg);
4572 		break;
4573 	case RTE_ETH_FILTER_SYN:
4574 		ret = eth_igb_syn_filter_handle(dev, filter_op, arg);
4575 		break;
4576 	case RTE_ETH_FILTER_FLEXIBLE:
4577 		ret = eth_igb_flex_filter_handle(dev, filter_op, arg);
4578 		break;
4579 	default:
4580 		PMD_DRV_LOG(WARNING, "Filter type (%d) not supported",
4581 							filter_type);
4582 		break;
4583 	}
4584 
4585 	return ret;
4586 }
4587 
4588 static int
4589 eth_igb_set_mc_addr_list(struct rte_eth_dev *dev,
4590 			 struct ether_addr *mc_addr_set,
4591 			 uint32_t nb_mc_addr)
4592 {
4593 	struct e1000_hw *hw;
4594 
4595 	hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4596 	e1000_update_mc_addr_list(hw, (u8 *)mc_addr_set, nb_mc_addr);
4597 	return 0;
4598 }
4599 
4600 static uint64_t
4601 igb_read_systime_cyclecounter(struct rte_eth_dev *dev)
4602 {
4603 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4604 	uint64_t systime_cycles;
4605 
4606 	switch (hw->mac.type) {
4607 	case e1000_i210:
4608 	case e1000_i211:
4609 		/*
4610 		 * Need to read System Time Residue Register to be able
4611 		 * to read the other two registers.
4612 		 */
4613 		E1000_READ_REG(hw, E1000_SYSTIMR);
4614 		/* SYSTIMEL stores ns and SYSTIMEH stores seconds. */
4615 		systime_cycles = (uint64_t)E1000_READ_REG(hw, E1000_SYSTIML);
4616 		systime_cycles += (uint64_t)E1000_READ_REG(hw, E1000_SYSTIMH)
4617 				* NSEC_PER_SEC;
4618 		break;
4619 	case e1000_82580:
4620 	case e1000_i350:
4621 	case e1000_i354:
4622 		/*
4623 		 * Need to read System Time Residue Register to be able
4624 		 * to read the other two registers.
4625 		 */
4626 		E1000_READ_REG(hw, E1000_SYSTIMR);
4627 		systime_cycles = (uint64_t)E1000_READ_REG(hw, E1000_SYSTIML);
4628 		/* Only the 8 LSB are valid. */
4629 		systime_cycles |= (uint64_t)(E1000_READ_REG(hw, E1000_SYSTIMH)
4630 				& 0xff) << 32;
4631 		break;
4632 	default:
4633 		systime_cycles = (uint64_t)E1000_READ_REG(hw, E1000_SYSTIML);
4634 		systime_cycles |= (uint64_t)E1000_READ_REG(hw, E1000_SYSTIMH)
4635 				<< 32;
4636 		break;
4637 	}
4638 
4639 	return systime_cycles;
4640 }
4641 
4642 static uint64_t
4643 igb_read_rx_tstamp_cyclecounter(struct rte_eth_dev *dev)
4644 {
4645 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4646 	uint64_t rx_tstamp_cycles;
4647 
4648 	switch (hw->mac.type) {
4649 	case e1000_i210:
4650 	case e1000_i211:
4651 		/* RXSTMPL stores ns and RXSTMPH stores seconds. */
4652 		rx_tstamp_cycles = (uint64_t)E1000_READ_REG(hw, E1000_RXSTMPL);
4653 		rx_tstamp_cycles += (uint64_t)E1000_READ_REG(hw, E1000_RXSTMPH)
4654 				* NSEC_PER_SEC;
4655 		break;
4656 	case e1000_82580:
4657 	case e1000_i350:
4658 	case e1000_i354:
4659 		rx_tstamp_cycles = (uint64_t)E1000_READ_REG(hw, E1000_RXSTMPL);
4660 		/* Only the 8 LSB are valid. */
4661 		rx_tstamp_cycles |= (uint64_t)(E1000_READ_REG(hw, E1000_RXSTMPH)
4662 				& 0xff) << 32;
4663 		break;
4664 	default:
4665 		rx_tstamp_cycles = (uint64_t)E1000_READ_REG(hw, E1000_RXSTMPL);
4666 		rx_tstamp_cycles |= (uint64_t)E1000_READ_REG(hw, E1000_RXSTMPH)
4667 				<< 32;
4668 		break;
4669 	}
4670 
4671 	return rx_tstamp_cycles;
4672 }
4673 
4674 static uint64_t
4675 igb_read_tx_tstamp_cyclecounter(struct rte_eth_dev *dev)
4676 {
4677 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4678 	uint64_t tx_tstamp_cycles;
4679 
4680 	switch (hw->mac.type) {
4681 	case e1000_i210:
4682 	case e1000_i211:
4683 		/* RXSTMPL stores ns and RXSTMPH stores seconds. */
4684 		tx_tstamp_cycles = (uint64_t)E1000_READ_REG(hw, E1000_TXSTMPL);
4685 		tx_tstamp_cycles += (uint64_t)E1000_READ_REG(hw, E1000_TXSTMPH)
4686 				* NSEC_PER_SEC;
4687 		break;
4688 	case e1000_82580:
4689 	case e1000_i350:
4690 	case e1000_i354:
4691 		tx_tstamp_cycles = (uint64_t)E1000_READ_REG(hw, E1000_TXSTMPL);
4692 		/* Only the 8 LSB are valid. */
4693 		tx_tstamp_cycles |= (uint64_t)(E1000_READ_REG(hw, E1000_TXSTMPH)
4694 				& 0xff) << 32;
4695 		break;
4696 	default:
4697 		tx_tstamp_cycles = (uint64_t)E1000_READ_REG(hw, E1000_TXSTMPL);
4698 		tx_tstamp_cycles |= (uint64_t)E1000_READ_REG(hw, E1000_TXSTMPH)
4699 				<< 32;
4700 		break;
4701 	}
4702 
4703 	return tx_tstamp_cycles;
4704 }
4705 
4706 static void
4707 igb_start_timecounters(struct rte_eth_dev *dev)
4708 {
4709 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4710 	struct e1000_adapter *adapter =
4711 		(struct e1000_adapter *)dev->data->dev_private;
4712 	uint32_t incval = 1;
4713 	uint32_t shift = 0;
4714 	uint64_t mask = E1000_CYCLECOUNTER_MASK;
4715 
4716 	switch (hw->mac.type) {
4717 	case e1000_82580:
4718 	case e1000_i350:
4719 	case e1000_i354:
4720 		/* 32 LSB bits + 8 MSB bits = 40 bits */
4721 		mask = (1ULL << 40) - 1;
4722 		/* fall-through */
4723 	case e1000_i210:
4724 	case e1000_i211:
4725 		/*
4726 		 * Start incrementing the register
4727 		 * used to timestamp PTP packets.
4728 		 */
4729 		E1000_WRITE_REG(hw, E1000_TIMINCA, incval);
4730 		break;
4731 	case e1000_82576:
4732 		incval = E1000_INCVALUE_82576;
4733 		shift = IGB_82576_TSYNC_SHIFT;
4734 		E1000_WRITE_REG(hw, E1000_TIMINCA,
4735 				E1000_INCPERIOD_82576 | incval);
4736 		break;
4737 	default:
4738 		/* Not supported */
4739 		return;
4740 	}
4741 
4742 	memset(&adapter->systime_tc, 0, sizeof(struct rte_timecounter));
4743 	memset(&adapter->rx_tstamp_tc, 0, sizeof(struct rte_timecounter));
4744 	memset(&adapter->tx_tstamp_tc, 0, sizeof(struct rte_timecounter));
4745 
4746 	adapter->systime_tc.cc_mask = mask;
4747 	adapter->systime_tc.cc_shift = shift;
4748 	adapter->systime_tc.nsec_mask = (1ULL << shift) - 1;
4749 
4750 	adapter->rx_tstamp_tc.cc_mask = mask;
4751 	adapter->rx_tstamp_tc.cc_shift = shift;
4752 	adapter->rx_tstamp_tc.nsec_mask = (1ULL << shift) - 1;
4753 
4754 	adapter->tx_tstamp_tc.cc_mask = mask;
4755 	adapter->tx_tstamp_tc.cc_shift = shift;
4756 	adapter->tx_tstamp_tc.nsec_mask = (1ULL << shift) - 1;
4757 }
4758 
4759 static int
4760 igb_timesync_adjust_time(struct rte_eth_dev *dev, int64_t delta)
4761 {
4762 	struct e1000_adapter *adapter =
4763 			(struct e1000_adapter *)dev->data->dev_private;
4764 
4765 	adapter->systime_tc.nsec += delta;
4766 	adapter->rx_tstamp_tc.nsec += delta;
4767 	adapter->tx_tstamp_tc.nsec += delta;
4768 
4769 	return 0;
4770 }
4771 
4772 static int
4773 igb_timesync_write_time(struct rte_eth_dev *dev, const struct timespec *ts)
4774 {
4775 	uint64_t ns;
4776 	struct e1000_adapter *adapter =
4777 			(struct e1000_adapter *)dev->data->dev_private;
4778 
4779 	ns = rte_timespec_to_ns(ts);
4780 
4781 	/* Set the timecounters to a new value. */
4782 	adapter->systime_tc.nsec = ns;
4783 	adapter->rx_tstamp_tc.nsec = ns;
4784 	adapter->tx_tstamp_tc.nsec = ns;
4785 
4786 	return 0;
4787 }
4788 
4789 static int
4790 igb_timesync_read_time(struct rte_eth_dev *dev, struct timespec *ts)
4791 {
4792 	uint64_t ns, systime_cycles;
4793 	struct e1000_adapter *adapter =
4794 			(struct e1000_adapter *)dev->data->dev_private;
4795 
4796 	systime_cycles = igb_read_systime_cyclecounter(dev);
4797 	ns = rte_timecounter_update(&adapter->systime_tc, systime_cycles);
4798 	*ts = rte_ns_to_timespec(ns);
4799 
4800 	return 0;
4801 }
4802 
4803 static int
4804 igb_timesync_enable(struct rte_eth_dev *dev)
4805 {
4806 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4807 	uint32_t tsync_ctl;
4808 	uint32_t tsauxc;
4809 
4810 	/* Stop the timesync system time. */
4811 	E1000_WRITE_REG(hw, E1000_TIMINCA, 0x0);
4812 	/* Reset the timesync system time value. */
4813 	switch (hw->mac.type) {
4814 	case e1000_82580:
4815 	case e1000_i350:
4816 	case e1000_i354:
4817 	case e1000_i210:
4818 	case e1000_i211:
4819 		E1000_WRITE_REG(hw, E1000_SYSTIMR, 0x0);
4820 		/* fall-through */
4821 	case e1000_82576:
4822 		E1000_WRITE_REG(hw, E1000_SYSTIML, 0x0);
4823 		E1000_WRITE_REG(hw, E1000_SYSTIMH, 0x0);
4824 		break;
4825 	default:
4826 		/* Not supported. */
4827 		return -ENOTSUP;
4828 	}
4829 
4830 	/* Enable system time for it isn't on by default. */
4831 	tsauxc = E1000_READ_REG(hw, E1000_TSAUXC);
4832 	tsauxc &= ~E1000_TSAUXC_DISABLE_SYSTIME;
4833 	E1000_WRITE_REG(hw, E1000_TSAUXC, tsauxc);
4834 
4835 	igb_start_timecounters(dev);
4836 
4837 	/* Enable L2 filtering of IEEE1588/802.1AS Ethernet frame types. */
4838 	E1000_WRITE_REG(hw, E1000_ETQF(E1000_ETQF_FILTER_1588),
4839 			(ETHER_TYPE_1588 |
4840 			 E1000_ETQF_FILTER_ENABLE |
4841 			 E1000_ETQF_1588));
4842 
4843 	/* Enable timestamping of received PTP packets. */
4844 	tsync_ctl = E1000_READ_REG(hw, E1000_TSYNCRXCTL);
4845 	tsync_ctl |= E1000_TSYNCRXCTL_ENABLED;
4846 	E1000_WRITE_REG(hw, E1000_TSYNCRXCTL, tsync_ctl);
4847 
4848 	/* Enable Timestamping of transmitted PTP packets. */
4849 	tsync_ctl = E1000_READ_REG(hw, E1000_TSYNCTXCTL);
4850 	tsync_ctl |= E1000_TSYNCTXCTL_ENABLED;
4851 	E1000_WRITE_REG(hw, E1000_TSYNCTXCTL, tsync_ctl);
4852 
4853 	return 0;
4854 }
4855 
4856 static int
4857 igb_timesync_disable(struct rte_eth_dev *dev)
4858 {
4859 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4860 	uint32_t tsync_ctl;
4861 
4862 	/* Disable timestamping of transmitted PTP packets. */
4863 	tsync_ctl = E1000_READ_REG(hw, E1000_TSYNCTXCTL);
4864 	tsync_ctl &= ~E1000_TSYNCTXCTL_ENABLED;
4865 	E1000_WRITE_REG(hw, E1000_TSYNCTXCTL, tsync_ctl);
4866 
4867 	/* Disable timestamping of received PTP packets. */
4868 	tsync_ctl = E1000_READ_REG(hw, E1000_TSYNCRXCTL);
4869 	tsync_ctl &= ~E1000_TSYNCRXCTL_ENABLED;
4870 	E1000_WRITE_REG(hw, E1000_TSYNCRXCTL, tsync_ctl);
4871 
4872 	/* Disable L2 filtering of IEEE1588/802.1AS Ethernet frame types. */
4873 	E1000_WRITE_REG(hw, E1000_ETQF(E1000_ETQF_FILTER_1588), 0);
4874 
4875 	/* Stop incrementating the System Time registers. */
4876 	E1000_WRITE_REG(hw, E1000_TIMINCA, 0);
4877 
4878 	return 0;
4879 }
4880 
4881 static int
4882 igb_timesync_read_rx_timestamp(struct rte_eth_dev *dev,
4883 			       struct timespec *timestamp,
4884 			       uint32_t flags __rte_unused)
4885 {
4886 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4887 	struct e1000_adapter *adapter =
4888 			(struct e1000_adapter *)dev->data->dev_private;
4889 	uint32_t tsync_rxctl;
4890 	uint64_t rx_tstamp_cycles;
4891 	uint64_t ns;
4892 
4893 	tsync_rxctl = E1000_READ_REG(hw, E1000_TSYNCRXCTL);
4894 	if ((tsync_rxctl & E1000_TSYNCRXCTL_VALID) == 0)
4895 		return -EINVAL;
4896 
4897 	rx_tstamp_cycles = igb_read_rx_tstamp_cyclecounter(dev);
4898 	ns = rte_timecounter_update(&adapter->rx_tstamp_tc, rx_tstamp_cycles);
4899 	*timestamp = rte_ns_to_timespec(ns);
4900 
4901 	return  0;
4902 }
4903 
4904 static int
4905 igb_timesync_read_tx_timestamp(struct rte_eth_dev *dev,
4906 			       struct timespec *timestamp)
4907 {
4908 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4909 	struct e1000_adapter *adapter =
4910 			(struct e1000_adapter *)dev->data->dev_private;
4911 	uint32_t tsync_txctl;
4912 	uint64_t tx_tstamp_cycles;
4913 	uint64_t ns;
4914 
4915 	tsync_txctl = E1000_READ_REG(hw, E1000_TSYNCTXCTL);
4916 	if ((tsync_txctl & E1000_TSYNCTXCTL_VALID) == 0)
4917 		return -EINVAL;
4918 
4919 	tx_tstamp_cycles = igb_read_tx_tstamp_cyclecounter(dev);
4920 	ns = rte_timecounter_update(&adapter->tx_tstamp_tc, tx_tstamp_cycles);
4921 	*timestamp = rte_ns_to_timespec(ns);
4922 
4923 	return  0;
4924 }
4925 
4926 static int
4927 eth_igb_get_reg_length(struct rte_eth_dev *dev __rte_unused)
4928 {
4929 	int count = 0;
4930 	int g_ind = 0;
4931 	const struct reg_info *reg_group;
4932 
4933 	while ((reg_group = igb_regs[g_ind++]))
4934 		count += igb_reg_group_count(reg_group);
4935 
4936 	return count;
4937 }
4938 
4939 static int
4940 igbvf_get_reg_length(struct rte_eth_dev *dev __rte_unused)
4941 {
4942 	int count = 0;
4943 	int g_ind = 0;
4944 	const struct reg_info *reg_group;
4945 
4946 	while ((reg_group = igbvf_regs[g_ind++]))
4947 		count += igb_reg_group_count(reg_group);
4948 
4949 	return count;
4950 }
4951 
4952 static int
4953 eth_igb_get_regs(struct rte_eth_dev *dev,
4954 	struct rte_dev_reg_info *regs)
4955 {
4956 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4957 	uint32_t *data = regs->data;
4958 	int g_ind = 0;
4959 	int count = 0;
4960 	const struct reg_info *reg_group;
4961 
4962 	if (data == NULL) {
4963 		regs->length = eth_igb_get_reg_length(dev);
4964 		regs->width = sizeof(uint32_t);
4965 		return 0;
4966 	}
4967 
4968 	/* Support only full register dump */
4969 	if ((regs->length == 0) ||
4970 	    (regs->length == (uint32_t)eth_igb_get_reg_length(dev))) {
4971 		regs->version = hw->mac.type << 24 | hw->revision_id << 16 |
4972 			hw->device_id;
4973 		while ((reg_group = igb_regs[g_ind++]))
4974 			count += igb_read_regs_group(dev, &data[count],
4975 							reg_group);
4976 		return 0;
4977 	}
4978 
4979 	return -ENOTSUP;
4980 }
4981 
4982 static int
4983 igbvf_get_regs(struct rte_eth_dev *dev,
4984 	struct rte_dev_reg_info *regs)
4985 {
4986 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4987 	uint32_t *data = regs->data;
4988 	int g_ind = 0;
4989 	int count = 0;
4990 	const struct reg_info *reg_group;
4991 
4992 	if (data == NULL) {
4993 		regs->length = igbvf_get_reg_length(dev);
4994 		regs->width = sizeof(uint32_t);
4995 		return 0;
4996 	}
4997 
4998 	/* Support only full register dump */
4999 	if ((regs->length == 0) ||
5000 	    (regs->length == (uint32_t)igbvf_get_reg_length(dev))) {
5001 		regs->version = hw->mac.type << 24 | hw->revision_id << 16 |
5002 			hw->device_id;
5003 		while ((reg_group = igbvf_regs[g_ind++]))
5004 			count += igb_read_regs_group(dev, &data[count],
5005 							reg_group);
5006 		return 0;
5007 	}
5008 
5009 	return -ENOTSUP;
5010 }
5011 
5012 static int
5013 eth_igb_get_eeprom_length(struct rte_eth_dev *dev)
5014 {
5015 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5016 
5017 	/* Return unit is byte count */
5018 	return hw->nvm.word_size * 2;
5019 }
5020 
5021 static int
5022 eth_igb_get_eeprom(struct rte_eth_dev *dev,
5023 	struct rte_dev_eeprom_info *in_eeprom)
5024 {
5025 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5026 	struct e1000_nvm_info *nvm = &hw->nvm;
5027 	uint16_t *data = in_eeprom->data;
5028 	int first, length;
5029 
5030 	first = in_eeprom->offset >> 1;
5031 	length = in_eeprom->length >> 1;
5032 	if ((first >= hw->nvm.word_size) ||
5033 	    ((first + length) >= hw->nvm.word_size))
5034 		return -EINVAL;
5035 
5036 	in_eeprom->magic = hw->vendor_id |
5037 		((uint32_t)hw->device_id << 16);
5038 
5039 	if ((nvm->ops.read) == NULL)
5040 		return -ENOTSUP;
5041 
5042 	return nvm->ops.read(hw, first, length, data);
5043 }
5044 
5045 static int
5046 eth_igb_set_eeprom(struct rte_eth_dev *dev,
5047 	struct rte_dev_eeprom_info *in_eeprom)
5048 {
5049 	struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5050 	struct e1000_nvm_info *nvm = &hw->nvm;
5051 	uint16_t *data = in_eeprom->data;
5052 	int first, length;
5053 
5054 	first = in_eeprom->offset >> 1;
5055 	length = in_eeprom->length >> 1;
5056 	if ((first >= hw->nvm.word_size) ||
5057 	    ((first + length) >= hw->nvm.word_size))
5058 		return -EINVAL;
5059 
5060 	in_eeprom->magic = (uint32_t)hw->vendor_id |
5061 		((uint32_t)hw->device_id << 16);
5062 
5063 	if ((nvm->ops.write) == NULL)
5064 		return -ENOTSUP;
5065 	return nvm->ops.write(hw,  first, length, data);
5066 }
5067 
5068 static int
5069 eth_igb_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
5070 {
5071 	struct e1000_hw *hw =
5072 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5073 	uint32_t mask = 1 << queue_id;
5074 
5075 	E1000_WRITE_REG(hw, E1000_EIMC, mask);
5076 	E1000_WRITE_FLUSH(hw);
5077 
5078 	return 0;
5079 }
5080 
5081 static int
5082 eth_igb_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
5083 {
5084 	struct e1000_hw *hw =
5085 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5086 	uint32_t mask = 1 << queue_id;
5087 	uint32_t regval;
5088 
5089 	regval = E1000_READ_REG(hw, E1000_EIMS);
5090 	E1000_WRITE_REG(hw, E1000_EIMS, regval | mask);
5091 	E1000_WRITE_FLUSH(hw);
5092 
5093 	rte_intr_enable(&dev->pci_dev->intr_handle);
5094 
5095 	return 0;
5096 }
5097 
5098 static void
5099 eth_igb_write_ivar(struct e1000_hw *hw, uint8_t  msix_vector,
5100 		   uint8_t index, uint8_t offset)
5101 {
5102 	uint32_t val = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
5103 
5104 	/* clear bits */
5105 	val &= ~((uint32_t)0xFF << offset);
5106 
5107 	/* write vector and valid bit */
5108 	val |= (msix_vector | E1000_IVAR_VALID) << offset;
5109 
5110 	E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, val);
5111 }
5112 
5113 static void
5114 eth_igb_assign_msix_vector(struct e1000_hw *hw, int8_t direction,
5115 			   uint8_t queue, uint8_t msix_vector)
5116 {
5117 	uint32_t tmp = 0;
5118 
5119 	if (hw->mac.type == e1000_82575) {
5120 		if (direction == 0)
5121 			tmp = E1000_EICR_RX_QUEUE0 << queue;
5122 		else if (direction == 1)
5123 			tmp = E1000_EICR_TX_QUEUE0 << queue;
5124 		E1000_WRITE_REG(hw, E1000_MSIXBM(msix_vector), tmp);
5125 	} else if (hw->mac.type == e1000_82576) {
5126 		if ((direction == 0) || (direction == 1))
5127 			eth_igb_write_ivar(hw, msix_vector, queue & 0x7,
5128 					   ((queue & 0x8) << 1) +
5129 					   8 * direction);
5130 	} else if ((hw->mac.type == e1000_82580) ||
5131 			(hw->mac.type == e1000_i350) ||
5132 			(hw->mac.type == e1000_i354) ||
5133 			(hw->mac.type == e1000_i210) ||
5134 			(hw->mac.type == e1000_i211)) {
5135 		if ((direction == 0) || (direction == 1))
5136 			eth_igb_write_ivar(hw, msix_vector,
5137 					   queue >> 1,
5138 					   ((queue & 0x1) << 4) +
5139 					   8 * direction);
5140 	}
5141 }
5142 
5143 /* Sets up the hardware to generate MSI-X interrupts properly
5144  * @hw
5145  *  board private structure
5146  */
5147 static void
5148 eth_igb_configure_msix_intr(struct rte_eth_dev *dev)
5149 {
5150 	int queue_id;
5151 	uint32_t tmpval, regval, intr_mask;
5152 	struct e1000_hw *hw =
5153 		E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5154 	uint32_t vec = E1000_MISC_VEC_ID;
5155 	uint32_t base = E1000_MISC_VEC_ID;
5156 	uint32_t misc_shift = 0;
5157 
5158 	struct rte_intr_handle *intr_handle = &dev->pci_dev->intr_handle;
5159 
5160 	/* won't configure msix register if no mapping is done
5161 	 * between intr vector and event fd
5162 	 */
5163 	if (!rte_intr_dp_is_en(intr_handle))
5164 		return;
5165 
5166 	if (rte_intr_allow_others(intr_handle)) {
5167 		vec = base = E1000_RX_VEC_START;
5168 		misc_shift = 1;
5169 	}
5170 
5171 	/* set interrupt vector for other causes */
5172 	if (hw->mac.type == e1000_82575) {
5173 		tmpval = E1000_READ_REG(hw, E1000_CTRL_EXT);
5174 		/* enable MSI-X PBA support */
5175 		tmpval |= E1000_CTRL_EXT_PBA_CLR;
5176 
5177 		/* Auto-Mask interrupts upon ICR read */
5178 		tmpval |= E1000_CTRL_EXT_EIAME;
5179 		tmpval |= E1000_CTRL_EXT_IRCA;
5180 
5181 		E1000_WRITE_REG(hw, E1000_CTRL_EXT, tmpval);
5182 
5183 		/* enable msix_other interrupt */
5184 		E1000_WRITE_REG_ARRAY(hw, E1000_MSIXBM(0), 0, E1000_EIMS_OTHER);
5185 		regval = E1000_READ_REG(hw, E1000_EIAC);
5186 		E1000_WRITE_REG(hw, E1000_EIAC, regval | E1000_EIMS_OTHER);
5187 		regval = E1000_READ_REG(hw, E1000_EIAM);
5188 		E1000_WRITE_REG(hw, E1000_EIMS, regval | E1000_EIMS_OTHER);
5189 	} else if ((hw->mac.type == e1000_82576) ||
5190 			(hw->mac.type == e1000_82580) ||
5191 			(hw->mac.type == e1000_i350) ||
5192 			(hw->mac.type == e1000_i354) ||
5193 			(hw->mac.type == e1000_i210) ||
5194 			(hw->mac.type == e1000_i211)) {
5195 		/* turn on MSI-X capability first */
5196 		E1000_WRITE_REG(hw, E1000_GPIE, E1000_GPIE_MSIX_MODE |
5197 					E1000_GPIE_PBA | E1000_GPIE_EIAME |
5198 					E1000_GPIE_NSICR);
5199 		intr_mask = RTE_LEN2MASK(intr_handle->nb_efd, uint32_t) <<
5200 			misc_shift;
5201 		regval = E1000_READ_REG(hw, E1000_EIAC);
5202 		E1000_WRITE_REG(hw, E1000_EIAC, regval | intr_mask);
5203 
5204 		/* enable msix_other interrupt */
5205 		regval = E1000_READ_REG(hw, E1000_EIMS);
5206 		E1000_WRITE_REG(hw, E1000_EIMS, regval | intr_mask);
5207 		tmpval = (dev->data->nb_rx_queues | E1000_IVAR_VALID) << 8;
5208 		E1000_WRITE_REG(hw, E1000_IVAR_MISC, tmpval);
5209 	}
5210 
5211 	/* use EIAM to auto-mask when MSI-X interrupt
5212 	 * is asserted, this saves a register write for every interrupt
5213 	 */
5214 	intr_mask = RTE_LEN2MASK(intr_handle->nb_efd, uint32_t) <<
5215 		misc_shift;
5216 	regval = E1000_READ_REG(hw, E1000_EIAM);
5217 	E1000_WRITE_REG(hw, E1000_EIAM, regval | intr_mask);
5218 
5219 	for (queue_id = 0; queue_id < dev->data->nb_rx_queues; queue_id++) {
5220 		eth_igb_assign_msix_vector(hw, 0, queue_id, vec);
5221 		intr_handle->intr_vec[queue_id] = vec;
5222 		if (vec < base + intr_handle->nb_efd - 1)
5223 			vec++;
5224 	}
5225 
5226 	E1000_WRITE_FLUSH(hw);
5227 }
5228 
5229 DRIVER_REGISTER_PCI(net_e1000_igb, rte_igb_pmd.pci_drv);
5230 DRIVER_REGISTER_PCI_TABLE(net_e1000_igb, pci_id_igb_map);
5231 DRIVER_REGISTER_PCI(net_e1000_igb_vf, rte_igbvf_pmd.pci_drv);
5232 DRIVER_REGISTER_PCI_TABLE(net_e1000_igb_vf, pci_id_igbvf_map);
5233