net/ngbe: support jumbo frame
[dpdk.git] / drivers / net / e1000 / igb_ethdev.c
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2016 Intel Corporation
3  */
4
5 #include <sys/queue.h>
6 #include <stdio.h>
7 #include <errno.h>
8 #include <stdint.h>
9 #include <stdarg.h>
10
11 #include <rte_string_fns.h>
12 #include <rte_common.h>
13 #include <rte_interrupts.h>
14 #include <rte_byteorder.h>
15 #include <rte_log.h>
16 #include <rte_debug.h>
17 #include <rte_pci.h>
18 #include <rte_bus_pci.h>
19 #include <rte_ether.h>
20 #include <ethdev_driver.h>
21 #include <ethdev_pci.h>
22 #include <rte_memory.h>
23 #include <rte_eal.h>
24 #include <rte_malloc.h>
25 #include <rte_dev.h>
26
27 #include "e1000_logs.h"
28 #include "base/e1000_api.h"
29 #include "e1000_ethdev.h"
30 #include "igb_regs.h"
31
32 /*
33  * Default values for port configuration
34  */
35 #define IGB_DEFAULT_RX_FREE_THRESH  32
36
37 #define IGB_DEFAULT_RX_PTHRESH      ((hw->mac.type == e1000_i354) ? 12 : 8)
38 #define IGB_DEFAULT_RX_HTHRESH      8
39 #define IGB_DEFAULT_RX_WTHRESH      ((hw->mac.type == e1000_82576) ? 1 : 4)
40
41 #define IGB_DEFAULT_TX_PTHRESH      ((hw->mac.type == e1000_i354) ? 20 : 8)
42 #define IGB_DEFAULT_TX_HTHRESH      1
43 #define IGB_DEFAULT_TX_WTHRESH      ((hw->mac.type == e1000_82576) ? 1 : 16)
44
45 /* Bit shift and mask */
46 #define IGB_4_BIT_WIDTH  (CHAR_BIT / 2)
47 #define IGB_4_BIT_MASK   RTE_LEN2MASK(IGB_4_BIT_WIDTH, uint8_t)
48 #define IGB_8_BIT_WIDTH  CHAR_BIT
49 #define IGB_8_BIT_MASK   UINT8_MAX
50
51 /* Additional timesync values. */
52 #define E1000_CYCLECOUNTER_MASK      0xffffffffffffffffULL
53 #define E1000_ETQF_FILTER_1588       3
54 #define IGB_82576_TSYNC_SHIFT        16
55 #define E1000_INCPERIOD_82576        (1 << E1000_TIMINCA_16NS_SHIFT)
56 #define E1000_INCVALUE_82576         (16 << IGB_82576_TSYNC_SHIFT)
57 #define E1000_TSAUXC_DISABLE_SYSTIME 0x80000000
58
59 #define E1000_VTIVAR_MISC                0x01740
60 #define E1000_VTIVAR_MISC_MASK           0xFF
61 #define E1000_VTIVAR_VALID               0x80
62 #define E1000_VTIVAR_MISC_MAILBOX        0
63 #define E1000_VTIVAR_MISC_INTR_MASK      0x3
64
65 /* External VLAN Enable bit mask */
66 #define E1000_CTRL_EXT_EXT_VLAN      (1 << 26)
67
68 /* External VLAN Ether Type bit mask and shift */
69 #define E1000_VET_VET_EXT            0xFFFF0000
70 #define E1000_VET_VET_EXT_SHIFT      16
71
72 /* MSI-X other interrupt vector */
73 #define IGB_MSIX_OTHER_INTR_VEC      0
74
75 static int  eth_igb_configure(struct rte_eth_dev *dev);
76 static int  eth_igb_start(struct rte_eth_dev *dev);
77 static int  eth_igb_stop(struct rte_eth_dev *dev);
78 static int  eth_igb_dev_set_link_up(struct rte_eth_dev *dev);
79 static int  eth_igb_dev_set_link_down(struct rte_eth_dev *dev);
80 static int eth_igb_close(struct rte_eth_dev *dev);
81 static int eth_igb_reset(struct rte_eth_dev *dev);
82 static int  eth_igb_promiscuous_enable(struct rte_eth_dev *dev);
83 static int  eth_igb_promiscuous_disable(struct rte_eth_dev *dev);
84 static int  eth_igb_allmulticast_enable(struct rte_eth_dev *dev);
85 static int  eth_igb_allmulticast_disable(struct rte_eth_dev *dev);
86 static int  eth_igb_link_update(struct rte_eth_dev *dev,
87                                 int wait_to_complete);
88 static int eth_igb_stats_get(struct rte_eth_dev *dev,
89                                 struct rte_eth_stats *rte_stats);
90 static int eth_igb_xstats_get(struct rte_eth_dev *dev,
91                               struct rte_eth_xstat *xstats, unsigned n);
92 static int eth_igb_xstats_get_by_id(struct rte_eth_dev *dev,
93                 const uint64_t *ids,
94                 uint64_t *values, unsigned int n);
95 static int eth_igb_xstats_get_names(struct rte_eth_dev *dev,
96                                     struct rte_eth_xstat_name *xstats_names,
97                                     unsigned int size);
98 static int eth_igb_xstats_get_names_by_id(struct rte_eth_dev *dev,
99                 const uint64_t *ids, struct rte_eth_xstat_name *xstats_names,
100                 unsigned int limit);
101 static int eth_igb_stats_reset(struct rte_eth_dev *dev);
102 static int eth_igb_xstats_reset(struct rte_eth_dev *dev);
103 static int eth_igb_fw_version_get(struct rte_eth_dev *dev,
104                                    char *fw_version, size_t fw_size);
105 static int eth_igb_infos_get(struct rte_eth_dev *dev,
106                               struct rte_eth_dev_info *dev_info);
107 static const uint32_t *eth_igb_supported_ptypes_get(struct rte_eth_dev *dev);
108 static int eth_igbvf_infos_get(struct rte_eth_dev *dev,
109                                 struct rte_eth_dev_info *dev_info);
110 static int  eth_igb_flow_ctrl_get(struct rte_eth_dev *dev,
111                                 struct rte_eth_fc_conf *fc_conf);
112 static int  eth_igb_flow_ctrl_set(struct rte_eth_dev *dev,
113                                 struct rte_eth_fc_conf *fc_conf);
114 static int eth_igb_lsc_interrupt_setup(struct rte_eth_dev *dev, uint8_t on);
115 static int eth_igb_rxq_interrupt_setup(struct rte_eth_dev *dev);
116 static int eth_igb_interrupt_get_status(struct rte_eth_dev *dev);
117 static int eth_igb_interrupt_action(struct rte_eth_dev *dev,
118                                     struct rte_intr_handle *handle);
119 static void eth_igb_interrupt_handler(void *param);
120 static int  igb_hardware_init(struct e1000_hw *hw);
121 static void igb_hw_control_acquire(struct e1000_hw *hw);
122 static void igb_hw_control_release(struct e1000_hw *hw);
123 static void igb_init_manageability(struct e1000_hw *hw);
124 static void igb_release_manageability(struct e1000_hw *hw);
125
126 static int  eth_igb_mtu_set(struct rte_eth_dev *dev, uint16_t mtu);
127
128 static int eth_igb_vlan_filter_set(struct rte_eth_dev *dev,
129                 uint16_t vlan_id, int on);
130 static int eth_igb_vlan_tpid_set(struct rte_eth_dev *dev,
131                                  enum rte_vlan_type vlan_type,
132                                  uint16_t tpid_id);
133 static int eth_igb_vlan_offload_set(struct rte_eth_dev *dev, int mask);
134
135 static void igb_vlan_hw_filter_enable(struct rte_eth_dev *dev);
136 static void igb_vlan_hw_filter_disable(struct rte_eth_dev *dev);
137 static void igb_vlan_hw_strip_enable(struct rte_eth_dev *dev);
138 static void igb_vlan_hw_strip_disable(struct rte_eth_dev *dev);
139 static void igb_vlan_hw_extend_enable(struct rte_eth_dev *dev);
140 static void igb_vlan_hw_extend_disable(struct rte_eth_dev *dev);
141
142 static int eth_igb_led_on(struct rte_eth_dev *dev);
143 static int eth_igb_led_off(struct rte_eth_dev *dev);
144
145 static void igb_intr_disable(struct rte_eth_dev *dev);
146 static int  igb_get_rx_buffer_size(struct e1000_hw *hw);
147 static int eth_igb_rar_set(struct rte_eth_dev *dev,
148                            struct rte_ether_addr *mac_addr,
149                            uint32_t index, uint32_t pool);
150 static void eth_igb_rar_clear(struct rte_eth_dev *dev, uint32_t index);
151 static int eth_igb_default_mac_addr_set(struct rte_eth_dev *dev,
152                 struct rte_ether_addr *addr);
153
154 static void igbvf_intr_disable(struct e1000_hw *hw);
155 static int igbvf_dev_configure(struct rte_eth_dev *dev);
156 static int igbvf_dev_start(struct rte_eth_dev *dev);
157 static int igbvf_dev_stop(struct rte_eth_dev *dev);
158 static int igbvf_dev_close(struct rte_eth_dev *dev);
159 static int igbvf_promiscuous_enable(struct rte_eth_dev *dev);
160 static int igbvf_promiscuous_disable(struct rte_eth_dev *dev);
161 static int igbvf_allmulticast_enable(struct rte_eth_dev *dev);
162 static int igbvf_allmulticast_disable(struct rte_eth_dev *dev);
163 static int eth_igbvf_link_update(struct e1000_hw *hw);
164 static int eth_igbvf_stats_get(struct rte_eth_dev *dev,
165                                 struct rte_eth_stats *rte_stats);
166 static int eth_igbvf_xstats_get(struct rte_eth_dev *dev,
167                                 struct rte_eth_xstat *xstats, unsigned n);
168 static int eth_igbvf_xstats_get_names(struct rte_eth_dev *dev,
169                                       struct rte_eth_xstat_name *xstats_names,
170                                       unsigned limit);
171 static int eth_igbvf_stats_reset(struct rte_eth_dev *dev);
172 static int igbvf_vlan_filter_set(struct rte_eth_dev *dev,
173                 uint16_t vlan_id, int on);
174 static int igbvf_set_vfta(struct e1000_hw *hw, uint16_t vid, bool on);
175 static void igbvf_set_vfta_all(struct rte_eth_dev *dev, bool on);
176 static int igbvf_default_mac_addr_set(struct rte_eth_dev *dev,
177                 struct rte_ether_addr *addr);
178 static int igbvf_get_reg_length(struct rte_eth_dev *dev);
179 static int igbvf_get_regs(struct rte_eth_dev *dev,
180                 struct rte_dev_reg_info *regs);
181
182 static int eth_igb_rss_reta_update(struct rte_eth_dev *dev,
183                                    struct rte_eth_rss_reta_entry64 *reta_conf,
184                                    uint16_t reta_size);
185 static int eth_igb_rss_reta_query(struct rte_eth_dev *dev,
186                                   struct rte_eth_rss_reta_entry64 *reta_conf,
187                                   uint16_t reta_size);
188
189 static int igb_add_2tuple_filter(struct rte_eth_dev *dev,
190                         struct rte_eth_ntuple_filter *ntuple_filter);
191 static int igb_remove_2tuple_filter(struct rte_eth_dev *dev,
192                         struct rte_eth_ntuple_filter *ntuple_filter);
193 static int igb_add_5tuple_filter_82576(struct rte_eth_dev *dev,
194                         struct rte_eth_ntuple_filter *ntuple_filter);
195 static int igb_remove_5tuple_filter_82576(struct rte_eth_dev *dev,
196                         struct rte_eth_ntuple_filter *ntuple_filter);
197 static int eth_igb_flow_ops_get(struct rte_eth_dev *dev,
198                                 const struct rte_flow_ops **ops);
199 static int eth_igb_get_reg_length(struct rte_eth_dev *dev);
200 static int eth_igb_get_regs(struct rte_eth_dev *dev,
201                 struct rte_dev_reg_info *regs);
202 static int eth_igb_get_eeprom_length(struct rte_eth_dev *dev);
203 static int eth_igb_get_eeprom(struct rte_eth_dev *dev,
204                 struct rte_dev_eeprom_info *eeprom);
205 static int eth_igb_set_eeprom(struct rte_eth_dev *dev,
206                 struct rte_dev_eeprom_info *eeprom);
207 static int eth_igb_get_module_info(struct rte_eth_dev *dev,
208                                    struct rte_eth_dev_module_info *modinfo);
209 static int eth_igb_get_module_eeprom(struct rte_eth_dev *dev,
210                                      struct rte_dev_eeprom_info *info);
211 static int eth_igb_set_mc_addr_list(struct rte_eth_dev *dev,
212                                     struct rte_ether_addr *mc_addr_set,
213                                     uint32_t nb_mc_addr);
214 static int igb_timesync_enable(struct rte_eth_dev *dev);
215 static int igb_timesync_disable(struct rte_eth_dev *dev);
216 static int igb_timesync_read_rx_timestamp(struct rte_eth_dev *dev,
217                                           struct timespec *timestamp,
218                                           uint32_t flags);
219 static int igb_timesync_read_tx_timestamp(struct rte_eth_dev *dev,
220                                           struct timespec *timestamp);
221 static int igb_timesync_adjust_time(struct rte_eth_dev *dev, int64_t delta);
222 static int igb_timesync_read_time(struct rte_eth_dev *dev,
223                                   struct timespec *timestamp);
224 static int igb_timesync_write_time(struct rte_eth_dev *dev,
225                                    const struct timespec *timestamp);
226 static int eth_igb_rx_queue_intr_enable(struct rte_eth_dev *dev,
227                                         uint16_t queue_id);
228 static int eth_igb_rx_queue_intr_disable(struct rte_eth_dev *dev,
229                                          uint16_t queue_id);
230 static void eth_igb_assign_msix_vector(struct e1000_hw *hw, int8_t direction,
231                                        uint8_t queue, uint8_t msix_vector);
232 static void eth_igb_write_ivar(struct e1000_hw *hw, uint8_t msix_vector,
233                                uint8_t index, uint8_t offset);
234 static void eth_igb_configure_msix_intr(struct rte_eth_dev *dev);
235 static void eth_igbvf_interrupt_handler(void *param);
236 static void igbvf_mbx_process(struct rte_eth_dev *dev);
237 static int igb_filter_restore(struct rte_eth_dev *dev);
238
239 /*
240  * Define VF Stats MACRO for Non "cleared on read" register
241  */
242 #define UPDATE_VF_STAT(reg, last, cur)            \
243 {                                                 \
244         u32 latest = E1000_READ_REG(hw, reg);     \
245         cur += (latest - last) & UINT_MAX;        \
246         last = latest;                            \
247 }
248
249 #define IGB_FC_PAUSE_TIME 0x0680
250 #define IGB_LINK_UPDATE_CHECK_TIMEOUT  90  /* 9s */
251 #define IGB_LINK_UPDATE_CHECK_INTERVAL 100 /* ms */
252
253 #define IGBVF_PMD_NAME "rte_igbvf_pmd"     /* PMD name */
254
255 static enum e1000_fc_mode igb_fc_setting = e1000_fc_full;
256
257 /*
258  * The set of PCI devices this driver supports
259  */
260 static const struct rte_pci_id pci_id_igb_map[] = {
261         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576) },
262         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576_FIBER) },
263         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576_SERDES) },
264         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576_QUAD_COPPER) },
265         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576_QUAD_COPPER_ET2) },
266         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576_NS) },
267         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576_NS_SERDES) },
268         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576_SERDES_QUAD) },
269
270         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82575EB_COPPER) },
271         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82575EB_FIBER_SERDES) },
272         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82575GB_QUAD_COPPER) },
273
274         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82580_COPPER) },
275         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82580_FIBER) },
276         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82580_SERDES) },
277         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82580_SGMII) },
278         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82580_COPPER_DUAL) },
279         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82580_QUAD_FIBER) },
280
281         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I350_COPPER) },
282         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I350_FIBER) },
283         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I350_SERDES) },
284         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I350_SGMII) },
285         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I350_DA4) },
286         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I210_COPPER) },
287         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I210_COPPER_OEM1) },
288         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I210_COPPER_IT) },
289         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I210_FIBER) },
290         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I210_SERDES) },
291         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I210_SGMII) },
292         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I210_COPPER_FLASHLESS) },
293         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I210_SERDES_FLASHLESS) },
294         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I211_COPPER) },
295         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I354_BACKPLANE_1GBPS) },
296         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I354_SGMII) },
297         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I354_BACKPLANE_2_5GBPS) },
298         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_DH89XXCC_SGMII) },
299         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_DH89XXCC_SERDES) },
300         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_DH89XXCC_BACKPLANE) },
301         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_DH89XXCC_SFP) },
302         { .vendor_id = 0, /* sentinel */ },
303 };
304
305 /*
306  * The set of PCI devices this driver supports (for 82576&I350 VF)
307  */
308 static const struct rte_pci_id pci_id_igbvf_map[] = {
309         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576_VF) },
310         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82576_VF_HV) },
311         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I350_VF) },
312         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_I350_VF_HV) },
313         { .vendor_id = 0, /* sentinel */ },
314 };
315
316 static const struct rte_eth_desc_lim rx_desc_lim = {
317         .nb_max = E1000_MAX_RING_DESC,
318         .nb_min = E1000_MIN_RING_DESC,
319         .nb_align = IGB_RXD_ALIGN,
320 };
321
322 static const struct rte_eth_desc_lim tx_desc_lim = {
323         .nb_max = E1000_MAX_RING_DESC,
324         .nb_min = E1000_MIN_RING_DESC,
325         .nb_align = IGB_RXD_ALIGN,
326         .nb_seg_max = IGB_TX_MAX_SEG,
327         .nb_mtu_seg_max = IGB_TX_MAX_MTU_SEG,
328 };
329
330 static const struct eth_dev_ops eth_igb_ops = {
331         .dev_configure        = eth_igb_configure,
332         .dev_start            = eth_igb_start,
333         .dev_stop             = eth_igb_stop,
334         .dev_set_link_up      = eth_igb_dev_set_link_up,
335         .dev_set_link_down    = eth_igb_dev_set_link_down,
336         .dev_close            = eth_igb_close,
337         .dev_reset            = eth_igb_reset,
338         .promiscuous_enable   = eth_igb_promiscuous_enable,
339         .promiscuous_disable  = eth_igb_promiscuous_disable,
340         .allmulticast_enable  = eth_igb_allmulticast_enable,
341         .allmulticast_disable = eth_igb_allmulticast_disable,
342         .link_update          = eth_igb_link_update,
343         .stats_get            = eth_igb_stats_get,
344         .xstats_get           = eth_igb_xstats_get,
345         .xstats_get_by_id     = eth_igb_xstats_get_by_id,
346         .xstats_get_names_by_id = eth_igb_xstats_get_names_by_id,
347         .xstats_get_names     = eth_igb_xstats_get_names,
348         .stats_reset          = eth_igb_stats_reset,
349         .xstats_reset         = eth_igb_xstats_reset,
350         .fw_version_get       = eth_igb_fw_version_get,
351         .dev_infos_get        = eth_igb_infos_get,
352         .dev_supported_ptypes_get = eth_igb_supported_ptypes_get,
353         .mtu_set              = eth_igb_mtu_set,
354         .vlan_filter_set      = eth_igb_vlan_filter_set,
355         .vlan_tpid_set        = eth_igb_vlan_tpid_set,
356         .vlan_offload_set     = eth_igb_vlan_offload_set,
357         .rx_queue_setup       = eth_igb_rx_queue_setup,
358         .rx_queue_intr_enable = eth_igb_rx_queue_intr_enable,
359         .rx_queue_intr_disable = eth_igb_rx_queue_intr_disable,
360         .rx_queue_release     = eth_igb_rx_queue_release,
361         .tx_queue_setup       = eth_igb_tx_queue_setup,
362         .tx_queue_release     = eth_igb_tx_queue_release,
363         .tx_done_cleanup      = eth_igb_tx_done_cleanup,
364         .dev_led_on           = eth_igb_led_on,
365         .dev_led_off          = eth_igb_led_off,
366         .flow_ctrl_get        = eth_igb_flow_ctrl_get,
367         .flow_ctrl_set        = eth_igb_flow_ctrl_set,
368         .mac_addr_add         = eth_igb_rar_set,
369         .mac_addr_remove      = eth_igb_rar_clear,
370         .mac_addr_set         = eth_igb_default_mac_addr_set,
371         .reta_update          = eth_igb_rss_reta_update,
372         .reta_query           = eth_igb_rss_reta_query,
373         .rss_hash_update      = eth_igb_rss_hash_update,
374         .rss_hash_conf_get    = eth_igb_rss_hash_conf_get,
375         .flow_ops_get         = eth_igb_flow_ops_get,
376         .set_mc_addr_list     = eth_igb_set_mc_addr_list,
377         .rxq_info_get         = igb_rxq_info_get,
378         .txq_info_get         = igb_txq_info_get,
379         .timesync_enable      = igb_timesync_enable,
380         .timesync_disable     = igb_timesync_disable,
381         .timesync_read_rx_timestamp = igb_timesync_read_rx_timestamp,
382         .timesync_read_tx_timestamp = igb_timesync_read_tx_timestamp,
383         .get_reg              = eth_igb_get_regs,
384         .get_eeprom_length    = eth_igb_get_eeprom_length,
385         .get_eeprom           = eth_igb_get_eeprom,
386         .set_eeprom           = eth_igb_set_eeprom,
387         .get_module_info      = eth_igb_get_module_info,
388         .get_module_eeprom    = eth_igb_get_module_eeprom,
389         .timesync_adjust_time = igb_timesync_adjust_time,
390         .timesync_read_time   = igb_timesync_read_time,
391         .timesync_write_time  = igb_timesync_write_time,
392 };
393
394 /*
395  * dev_ops for virtual function, bare necessities for basic vf
396  * operation have been implemented
397  */
398 static const struct eth_dev_ops igbvf_eth_dev_ops = {
399         .dev_configure        = igbvf_dev_configure,
400         .dev_start            = igbvf_dev_start,
401         .dev_stop             = igbvf_dev_stop,
402         .dev_close            = igbvf_dev_close,
403         .promiscuous_enable   = igbvf_promiscuous_enable,
404         .promiscuous_disable  = igbvf_promiscuous_disable,
405         .allmulticast_enable  = igbvf_allmulticast_enable,
406         .allmulticast_disable = igbvf_allmulticast_disable,
407         .link_update          = eth_igb_link_update,
408         .stats_get            = eth_igbvf_stats_get,
409         .xstats_get           = eth_igbvf_xstats_get,
410         .xstats_get_names     = eth_igbvf_xstats_get_names,
411         .stats_reset          = eth_igbvf_stats_reset,
412         .xstats_reset         = eth_igbvf_stats_reset,
413         .vlan_filter_set      = igbvf_vlan_filter_set,
414         .dev_infos_get        = eth_igbvf_infos_get,
415         .dev_supported_ptypes_get = eth_igb_supported_ptypes_get,
416         .rx_queue_setup       = eth_igb_rx_queue_setup,
417         .rx_queue_release     = eth_igb_rx_queue_release,
418         .tx_queue_setup       = eth_igb_tx_queue_setup,
419         .tx_queue_release     = eth_igb_tx_queue_release,
420         .tx_done_cleanup      = eth_igb_tx_done_cleanup,
421         .set_mc_addr_list     = eth_igb_set_mc_addr_list,
422         .rxq_info_get         = igb_rxq_info_get,
423         .txq_info_get         = igb_txq_info_get,
424         .mac_addr_set         = igbvf_default_mac_addr_set,
425         .get_reg              = igbvf_get_regs,
426 };
427
428 /* store statistics names and its offset in stats structure */
429 struct rte_igb_xstats_name_off {
430         char name[RTE_ETH_XSTATS_NAME_SIZE];
431         unsigned offset;
432 };
433
434 static const struct rte_igb_xstats_name_off rte_igb_stats_strings[] = {
435         {"rx_crc_errors", offsetof(struct e1000_hw_stats, crcerrs)},
436         {"rx_align_errors", offsetof(struct e1000_hw_stats, algnerrc)},
437         {"rx_symbol_errors", offsetof(struct e1000_hw_stats, symerrs)},
438         {"rx_missed_packets", offsetof(struct e1000_hw_stats, mpc)},
439         {"tx_single_collision_packets", offsetof(struct e1000_hw_stats, scc)},
440         {"tx_multiple_collision_packets", offsetof(struct e1000_hw_stats, mcc)},
441         {"tx_excessive_collision_packets", offsetof(struct e1000_hw_stats,
442                 ecol)},
443         {"tx_late_collisions", offsetof(struct e1000_hw_stats, latecol)},
444         {"tx_total_collisions", offsetof(struct e1000_hw_stats, colc)},
445         {"tx_deferred_packets", offsetof(struct e1000_hw_stats, dc)},
446         {"tx_no_carrier_sense_packets", offsetof(struct e1000_hw_stats, tncrs)},
447         {"rx_carrier_ext_errors", offsetof(struct e1000_hw_stats, cexterr)},
448         {"rx_length_errors", offsetof(struct e1000_hw_stats, rlec)},
449         {"rx_xon_packets", offsetof(struct e1000_hw_stats, xonrxc)},
450         {"tx_xon_packets", offsetof(struct e1000_hw_stats, xontxc)},
451         {"rx_xoff_packets", offsetof(struct e1000_hw_stats, xoffrxc)},
452         {"tx_xoff_packets", offsetof(struct e1000_hw_stats, xofftxc)},
453         {"rx_flow_control_unsupported_packets", offsetof(struct e1000_hw_stats,
454                 fcruc)},
455         {"rx_size_64_packets", offsetof(struct e1000_hw_stats, prc64)},
456         {"rx_size_65_to_127_packets", offsetof(struct e1000_hw_stats, prc127)},
457         {"rx_size_128_to_255_packets", offsetof(struct e1000_hw_stats, prc255)},
458         {"rx_size_256_to_511_packets", offsetof(struct e1000_hw_stats, prc511)},
459         {"rx_size_512_to_1023_packets", offsetof(struct e1000_hw_stats,
460                 prc1023)},
461         {"rx_size_1024_to_max_packets", offsetof(struct e1000_hw_stats,
462                 prc1522)},
463         {"rx_broadcast_packets", offsetof(struct e1000_hw_stats, bprc)},
464         {"rx_multicast_packets", offsetof(struct e1000_hw_stats, mprc)},
465         {"rx_undersize_errors", offsetof(struct e1000_hw_stats, ruc)},
466         {"rx_fragment_errors", offsetof(struct e1000_hw_stats, rfc)},
467         {"rx_oversize_errors", offsetof(struct e1000_hw_stats, roc)},
468         {"rx_jabber_errors", offsetof(struct e1000_hw_stats, rjc)},
469         {"rx_management_packets", offsetof(struct e1000_hw_stats, mgprc)},
470         {"rx_management_dropped", offsetof(struct e1000_hw_stats, mgpdc)},
471         {"tx_management_packets", offsetof(struct e1000_hw_stats, mgptc)},
472         {"rx_total_packets", offsetof(struct e1000_hw_stats, tpr)},
473         {"tx_total_packets", offsetof(struct e1000_hw_stats, tpt)},
474         {"rx_total_bytes", offsetof(struct e1000_hw_stats, tor)},
475         {"tx_total_bytes", offsetof(struct e1000_hw_stats, tot)},
476         {"tx_size_64_packets", offsetof(struct e1000_hw_stats, ptc64)},
477         {"tx_size_65_to_127_packets", offsetof(struct e1000_hw_stats, ptc127)},
478         {"tx_size_128_to_255_packets", offsetof(struct e1000_hw_stats, ptc255)},
479         {"tx_size_256_to_511_packets", offsetof(struct e1000_hw_stats, ptc511)},
480         {"tx_size_512_to_1023_packets", offsetof(struct e1000_hw_stats,
481                 ptc1023)},
482         {"tx_size_1023_to_max_packets", offsetof(struct e1000_hw_stats,
483                 ptc1522)},
484         {"tx_multicast_packets", offsetof(struct e1000_hw_stats, mptc)},
485         {"tx_broadcast_packets", offsetof(struct e1000_hw_stats, bptc)},
486         {"tx_tso_packets", offsetof(struct e1000_hw_stats, tsctc)},
487         {"tx_tso_errors", offsetof(struct e1000_hw_stats, tsctfc)},
488         {"rx_sent_to_host_packets", offsetof(struct e1000_hw_stats, rpthc)},
489         {"tx_sent_by_host_packets", offsetof(struct e1000_hw_stats, hgptc)},
490         {"rx_code_violation_packets", offsetof(struct e1000_hw_stats, scvpc)},
491
492         {"interrupt_assert_count", offsetof(struct e1000_hw_stats, iac)},
493 };
494
495 #define IGB_NB_XSTATS (sizeof(rte_igb_stats_strings) / \
496                 sizeof(rte_igb_stats_strings[0]))
497
498 static const struct rte_igb_xstats_name_off rte_igbvf_stats_strings[] = {
499         {"rx_multicast_packets", offsetof(struct e1000_vf_stats, mprc)},
500         {"rx_good_loopback_packets", offsetof(struct e1000_vf_stats, gprlbc)},
501         {"tx_good_loopback_packets", offsetof(struct e1000_vf_stats, gptlbc)},
502         {"rx_good_loopback_bytes", offsetof(struct e1000_vf_stats, gorlbc)},
503         {"tx_good_loopback_bytes", offsetof(struct e1000_vf_stats, gotlbc)},
504 };
505
506 #define IGBVF_NB_XSTATS (sizeof(rte_igbvf_stats_strings) / \
507                 sizeof(rte_igbvf_stats_strings[0]))
508
509
510 static inline void
511 igb_intr_enable(struct rte_eth_dev *dev)
512 {
513         struct e1000_interrupt *intr =
514                 E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
515         struct e1000_hw *hw =
516                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
517         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
518         struct rte_intr_handle *intr_handle = pci_dev->intr_handle;
519
520         if (rte_intr_allow_others(intr_handle) &&
521                 dev->data->dev_conf.intr_conf.lsc != 0) {
522                 E1000_WRITE_REG(hw, E1000_EIMS, 1 << IGB_MSIX_OTHER_INTR_VEC);
523         }
524
525         E1000_WRITE_REG(hw, E1000_IMS, intr->mask);
526         E1000_WRITE_FLUSH(hw);
527 }
528
529 static void
530 igb_intr_disable(struct rte_eth_dev *dev)
531 {
532         struct e1000_hw *hw =
533                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
534         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
535         struct rte_intr_handle *intr_handle = pci_dev->intr_handle;
536
537         if (rte_intr_allow_others(intr_handle) &&
538                 dev->data->dev_conf.intr_conf.lsc != 0) {
539                 E1000_WRITE_REG(hw, E1000_EIMC, 1 << IGB_MSIX_OTHER_INTR_VEC);
540         }
541
542         E1000_WRITE_REG(hw, E1000_IMC, ~0);
543         E1000_WRITE_FLUSH(hw);
544 }
545
546 static inline void
547 igbvf_intr_enable(struct rte_eth_dev *dev)
548 {
549         struct e1000_hw *hw =
550                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
551
552         /* only for mailbox */
553         E1000_WRITE_REG(hw, E1000_EIAM, 1 << E1000_VTIVAR_MISC_MAILBOX);
554         E1000_WRITE_REG(hw, E1000_EIAC, 1 << E1000_VTIVAR_MISC_MAILBOX);
555         E1000_WRITE_REG(hw, E1000_EIMS, 1 << E1000_VTIVAR_MISC_MAILBOX);
556         E1000_WRITE_FLUSH(hw);
557 }
558
559 /* only for mailbox now. If RX/TX needed, should extend this function.  */
560 static void
561 igbvf_set_ivar_map(struct e1000_hw *hw, uint8_t msix_vector)
562 {
563         uint32_t tmp = 0;
564
565         /* mailbox */
566         tmp |= (msix_vector & E1000_VTIVAR_MISC_INTR_MASK);
567         tmp |= E1000_VTIVAR_VALID;
568         E1000_WRITE_REG(hw, E1000_VTIVAR_MISC, tmp);
569 }
570
571 static void
572 eth_igbvf_configure_msix_intr(struct rte_eth_dev *dev)
573 {
574         struct e1000_hw *hw =
575                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
576
577         /* Configure VF other cause ivar */
578         igbvf_set_ivar_map(hw, E1000_VTIVAR_MISC_MAILBOX);
579 }
580
581 static inline int32_t
582 igb_pf_reset_hw(struct e1000_hw *hw)
583 {
584         uint32_t ctrl_ext;
585         int32_t status;
586
587         status = e1000_reset_hw(hw);
588
589         ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
590         /* Set PF Reset Done bit so PF/VF Mail Ops can work */
591         ctrl_ext |= E1000_CTRL_EXT_PFRSTD;
592         E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
593         E1000_WRITE_FLUSH(hw);
594
595         return status;
596 }
597
598 static void
599 igb_identify_hardware(struct rte_eth_dev *dev, struct rte_pci_device *pci_dev)
600 {
601         struct e1000_hw *hw =
602                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
603
604
605         hw->vendor_id = pci_dev->id.vendor_id;
606         hw->device_id = pci_dev->id.device_id;
607         hw->subsystem_vendor_id = pci_dev->id.subsystem_vendor_id;
608         hw->subsystem_device_id = pci_dev->id.subsystem_device_id;
609
610         e1000_set_mac_type(hw);
611
612         /* need to check if it is a vf device below */
613 }
614
615 static int
616 igb_reset_swfw_lock(struct e1000_hw *hw)
617 {
618         int ret_val;
619
620         /*
621          * Do mac ops initialization manually here, since we will need
622          * some function pointers set by this call.
623          */
624         ret_val = e1000_init_mac_params(hw);
625         if (ret_val)
626                 return ret_val;
627
628         /*
629          * SMBI lock should not fail in this early stage. If this is the case,
630          * it is due to an improper exit of the application.
631          * So force the release of the faulty lock.
632          */
633         if (e1000_get_hw_semaphore_generic(hw) < 0) {
634                 PMD_DRV_LOG(DEBUG, "SMBI lock released");
635         }
636         e1000_put_hw_semaphore_generic(hw);
637
638         if (hw->mac.ops.acquire_swfw_sync != NULL) {
639                 uint16_t mask;
640
641                 /*
642                  * Phy lock should not fail in this early stage. If this is the case,
643                  * it is due to an improper exit of the application.
644                  * So force the release of the faulty lock.
645                  */
646                 mask = E1000_SWFW_PHY0_SM << hw->bus.func;
647                 if (hw->bus.func > E1000_FUNC_1)
648                         mask <<= 2;
649                 if (hw->mac.ops.acquire_swfw_sync(hw, mask) < 0) {
650                         PMD_DRV_LOG(DEBUG, "SWFW phy%d lock released",
651                                     hw->bus.func);
652                 }
653                 hw->mac.ops.release_swfw_sync(hw, mask);
654
655                 /*
656                  * This one is more tricky since it is common to all ports; but
657                  * swfw_sync retries last long enough (1s) to be almost sure that if
658                  * lock can not be taken it is due to an improper lock of the
659                  * semaphore.
660                  */
661                 mask = E1000_SWFW_EEP_SM;
662                 if (hw->mac.ops.acquire_swfw_sync(hw, mask) < 0) {
663                         PMD_DRV_LOG(DEBUG, "SWFW common locks released");
664                 }
665                 hw->mac.ops.release_swfw_sync(hw, mask);
666         }
667
668         return E1000_SUCCESS;
669 }
670
671 /* Remove all ntuple filters of the device */
672 static int igb_ntuple_filter_uninit(struct rte_eth_dev *eth_dev)
673 {
674         struct e1000_filter_info *filter_info =
675                 E1000_DEV_PRIVATE_TO_FILTER_INFO(eth_dev->data->dev_private);
676         struct e1000_5tuple_filter *p_5tuple;
677         struct e1000_2tuple_filter *p_2tuple;
678
679         while ((p_5tuple = TAILQ_FIRST(&filter_info->fivetuple_list))) {
680                 TAILQ_REMOVE(&filter_info->fivetuple_list,
681                         p_5tuple, entries);
682                         rte_free(p_5tuple);
683         }
684         filter_info->fivetuple_mask = 0;
685         while ((p_2tuple = TAILQ_FIRST(&filter_info->twotuple_list))) {
686                 TAILQ_REMOVE(&filter_info->twotuple_list,
687                         p_2tuple, entries);
688                         rte_free(p_2tuple);
689         }
690         filter_info->twotuple_mask = 0;
691
692         return 0;
693 }
694
695 /* Remove all flex filters of the device */
696 static int igb_flex_filter_uninit(struct rte_eth_dev *eth_dev)
697 {
698         struct e1000_filter_info *filter_info =
699                 E1000_DEV_PRIVATE_TO_FILTER_INFO(eth_dev->data->dev_private);
700         struct e1000_flex_filter *p_flex;
701
702         while ((p_flex = TAILQ_FIRST(&filter_info->flex_list))) {
703                 TAILQ_REMOVE(&filter_info->flex_list, p_flex, entries);
704                 rte_free(p_flex);
705         }
706         filter_info->flex_mask = 0;
707
708         return 0;
709 }
710
711 static int
712 eth_igb_dev_init(struct rte_eth_dev *eth_dev)
713 {
714         int error = 0;
715         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
716         struct e1000_hw *hw =
717                 E1000_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
718         struct e1000_vfta * shadow_vfta =
719                 E1000_DEV_PRIVATE_TO_VFTA(eth_dev->data->dev_private);
720         struct e1000_filter_info *filter_info =
721                 E1000_DEV_PRIVATE_TO_FILTER_INFO(eth_dev->data->dev_private);
722         struct e1000_adapter *adapter =
723                 E1000_DEV_PRIVATE(eth_dev->data->dev_private);
724
725         uint32_t ctrl_ext;
726
727         eth_dev->dev_ops = &eth_igb_ops;
728         eth_dev->rx_queue_count = eth_igb_rx_queue_count;
729         eth_dev->rx_descriptor_status = eth_igb_rx_descriptor_status;
730         eth_dev->tx_descriptor_status = eth_igb_tx_descriptor_status;
731         eth_dev->rx_pkt_burst = &eth_igb_recv_pkts;
732         eth_dev->tx_pkt_burst = &eth_igb_xmit_pkts;
733         eth_dev->tx_pkt_prepare = &eth_igb_prep_pkts;
734
735         /* for secondary processes, we don't initialise any further as primary
736          * has already done this work. Only check we don't need a different
737          * RX function */
738         if (rte_eal_process_type() != RTE_PROC_PRIMARY){
739                 if (eth_dev->data->scattered_rx)
740                         eth_dev->rx_pkt_burst = &eth_igb_recv_scattered_pkts;
741                 return 0;
742         }
743
744         rte_eth_copy_pci_info(eth_dev, pci_dev);
745
746         hw->hw_addr= (void *)pci_dev->mem_resource[0].addr;
747
748         igb_identify_hardware(eth_dev, pci_dev);
749         if (e1000_setup_init_funcs(hw, FALSE) != E1000_SUCCESS) {
750                 error = -EIO;
751                 goto err_late;
752         }
753
754         e1000_get_bus_info(hw);
755
756         /* Reset any pending lock */
757         if (igb_reset_swfw_lock(hw) != E1000_SUCCESS) {
758                 error = -EIO;
759                 goto err_late;
760         }
761
762         /* Finish initialization */
763         if (e1000_setup_init_funcs(hw, TRUE) != E1000_SUCCESS) {
764                 error = -EIO;
765                 goto err_late;
766         }
767
768         hw->mac.autoneg = 1;
769         hw->phy.autoneg_wait_to_complete = 0;
770         hw->phy.autoneg_advertised = E1000_ALL_SPEED_DUPLEX;
771
772         /* Copper options */
773         if (hw->phy.media_type == e1000_media_type_copper) {
774                 hw->phy.mdix = 0; /* AUTO_ALL_MODES */
775                 hw->phy.disable_polarity_correction = 0;
776                 hw->phy.ms_type = e1000_ms_hw_default;
777         }
778
779         /*
780          * Start from a known state, this is important in reading the nvm
781          * and mac from that.
782          */
783         igb_pf_reset_hw(hw);
784
785         /* Make sure we have a good EEPROM before we read from it */
786         if (e1000_validate_nvm_checksum(hw) < 0) {
787                 /*
788                  * Some PCI-E parts fail the first check due to
789                  * the link being in sleep state, call it again,
790                  * if it fails a second time its a real issue.
791                  */
792                 if (e1000_validate_nvm_checksum(hw) < 0) {
793                         PMD_INIT_LOG(ERR, "EEPROM checksum invalid");
794                         error = -EIO;
795                         goto err_late;
796                 }
797         }
798
799         /* Read the permanent MAC address out of the EEPROM */
800         if (e1000_read_mac_addr(hw) != 0) {
801                 PMD_INIT_LOG(ERR, "EEPROM error while reading MAC address");
802                 error = -EIO;
803                 goto err_late;
804         }
805
806         /* Allocate memory for storing MAC addresses */
807         eth_dev->data->mac_addrs = rte_zmalloc("e1000",
808                 RTE_ETHER_ADDR_LEN * hw->mac.rar_entry_count, 0);
809         if (eth_dev->data->mac_addrs == NULL) {
810                 PMD_INIT_LOG(ERR, "Failed to allocate %d bytes needed to "
811                                                 "store MAC addresses",
812                                 RTE_ETHER_ADDR_LEN * hw->mac.rar_entry_count);
813                 error = -ENOMEM;
814                 goto err_late;
815         }
816
817         /* Copy the permanent MAC address */
818         rte_ether_addr_copy((struct rte_ether_addr *)hw->mac.addr,
819                         &eth_dev->data->mac_addrs[0]);
820
821         /* initialize the vfta */
822         memset(shadow_vfta, 0, sizeof(*shadow_vfta));
823
824         /* Now initialize the hardware */
825         if (igb_hardware_init(hw) != 0) {
826                 PMD_INIT_LOG(ERR, "Hardware initialization failed");
827                 rte_free(eth_dev->data->mac_addrs);
828                 eth_dev->data->mac_addrs = NULL;
829                 error = -ENODEV;
830                 goto err_late;
831         }
832         hw->mac.get_link_status = 1;
833         adapter->stopped = 0;
834
835         /* Indicate SOL/IDER usage */
836         if (e1000_check_reset_block(hw) < 0) {
837                 PMD_INIT_LOG(ERR, "PHY reset is blocked due to"
838                                         "SOL/IDER session");
839         }
840
841         /* initialize PF if max_vfs not zero */
842         igb_pf_host_init(eth_dev);
843
844         ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
845         /* Set PF Reset Done bit so PF/VF Mail Ops can work */
846         ctrl_ext |= E1000_CTRL_EXT_PFRSTD;
847         E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
848         E1000_WRITE_FLUSH(hw);
849
850         PMD_INIT_LOG(DEBUG, "port_id %d vendorID=0x%x deviceID=0x%x",
851                      eth_dev->data->port_id, pci_dev->id.vendor_id,
852                      pci_dev->id.device_id);
853
854         rte_intr_callback_register(pci_dev->intr_handle,
855                                    eth_igb_interrupt_handler,
856                                    (void *)eth_dev);
857
858         /* enable uio/vfio intr/eventfd mapping */
859         rte_intr_enable(pci_dev->intr_handle);
860
861         /* enable support intr */
862         igb_intr_enable(eth_dev);
863
864         eth_igb_dev_set_link_down(eth_dev);
865
866         /* initialize filter info */
867         memset(filter_info, 0,
868                sizeof(struct e1000_filter_info));
869
870         TAILQ_INIT(&filter_info->flex_list);
871         TAILQ_INIT(&filter_info->twotuple_list);
872         TAILQ_INIT(&filter_info->fivetuple_list);
873
874         TAILQ_INIT(&igb_filter_ntuple_list);
875         TAILQ_INIT(&igb_filter_ethertype_list);
876         TAILQ_INIT(&igb_filter_syn_list);
877         TAILQ_INIT(&igb_filter_flex_list);
878         TAILQ_INIT(&igb_filter_rss_list);
879         TAILQ_INIT(&igb_flow_list);
880
881         return 0;
882
883 err_late:
884         igb_hw_control_release(hw);
885
886         return error;
887 }
888
889 static int
890 eth_igb_dev_uninit(struct rte_eth_dev *eth_dev)
891 {
892         PMD_INIT_FUNC_TRACE();
893
894         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
895                 return 0;
896
897         eth_igb_close(eth_dev);
898
899         return 0;
900 }
901
902 /*
903  * Virtual Function device init
904  */
905 static int
906 eth_igbvf_dev_init(struct rte_eth_dev *eth_dev)
907 {
908         struct rte_pci_device *pci_dev;
909         struct rte_intr_handle *intr_handle;
910         struct e1000_adapter *adapter =
911                 E1000_DEV_PRIVATE(eth_dev->data->dev_private);
912         struct e1000_hw *hw =
913                 E1000_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
914         int diag;
915         struct rte_ether_addr *perm_addr =
916                 (struct rte_ether_addr *)hw->mac.perm_addr;
917
918         PMD_INIT_FUNC_TRACE();
919
920         eth_dev->dev_ops = &igbvf_eth_dev_ops;
921         eth_dev->rx_descriptor_status = eth_igb_rx_descriptor_status;
922         eth_dev->tx_descriptor_status = eth_igb_tx_descriptor_status;
923         eth_dev->rx_pkt_burst = &eth_igb_recv_pkts;
924         eth_dev->tx_pkt_burst = &eth_igb_xmit_pkts;
925         eth_dev->tx_pkt_prepare = &eth_igb_prep_pkts;
926
927         /* for secondary processes, we don't initialise any further as primary
928          * has already done this work. Only check we don't need a different
929          * RX function */
930         if (rte_eal_process_type() != RTE_PROC_PRIMARY){
931                 if (eth_dev->data->scattered_rx)
932                         eth_dev->rx_pkt_burst = &eth_igb_recv_scattered_pkts;
933                 return 0;
934         }
935
936         pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
937         rte_eth_copy_pci_info(eth_dev, pci_dev);
938
939         hw->device_id = pci_dev->id.device_id;
940         hw->vendor_id = pci_dev->id.vendor_id;
941         hw->hw_addr = (void *)pci_dev->mem_resource[0].addr;
942         adapter->stopped = 0;
943
944         /* Initialize the shared code (base driver) */
945         diag = e1000_setup_init_funcs(hw, TRUE);
946         if (diag != 0) {
947                 PMD_INIT_LOG(ERR, "Shared code init failed for igbvf: %d",
948                         diag);
949                 return -EIO;
950         }
951
952         /* init_mailbox_params */
953         hw->mbx.ops.init_params(hw);
954
955         /* Disable the interrupts for VF */
956         igbvf_intr_disable(hw);
957
958         diag = hw->mac.ops.reset_hw(hw);
959
960         /* Allocate memory for storing MAC addresses */
961         eth_dev->data->mac_addrs = rte_zmalloc("igbvf", RTE_ETHER_ADDR_LEN *
962                 hw->mac.rar_entry_count, 0);
963         if (eth_dev->data->mac_addrs == NULL) {
964                 PMD_INIT_LOG(ERR,
965                         "Failed to allocate %d bytes needed to store MAC "
966                         "addresses",
967                         RTE_ETHER_ADDR_LEN * hw->mac.rar_entry_count);
968                 return -ENOMEM;
969         }
970
971         /* Generate a random MAC address, if none was assigned by PF. */
972         if (rte_is_zero_ether_addr(perm_addr)) {
973                 rte_eth_random_addr(perm_addr->addr_bytes);
974                 PMD_INIT_LOG(INFO, "\tVF MAC address not assigned by Host PF");
975                 PMD_INIT_LOG(INFO, "\tAssign randomly generated MAC address "
976                              RTE_ETHER_ADDR_PRT_FMT,
977                              RTE_ETHER_ADDR_BYTES(perm_addr));
978         }
979
980         diag = e1000_rar_set(hw, perm_addr->addr_bytes, 0);
981         if (diag) {
982                 rte_free(eth_dev->data->mac_addrs);
983                 eth_dev->data->mac_addrs = NULL;
984                 return diag;
985         }
986         /* Copy the permanent MAC address */
987         rte_ether_addr_copy((struct rte_ether_addr *)hw->mac.perm_addr,
988                         &eth_dev->data->mac_addrs[0]);
989
990         PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x "
991                      "mac.type=%s",
992                      eth_dev->data->port_id, pci_dev->id.vendor_id,
993                      pci_dev->id.device_id, "igb_mac_82576_vf");
994
995         intr_handle = pci_dev->intr_handle;
996         rte_intr_callback_register(intr_handle,
997                                    eth_igbvf_interrupt_handler, eth_dev);
998
999         return 0;
1000 }
1001
1002 static int
1003 eth_igbvf_dev_uninit(struct rte_eth_dev *eth_dev)
1004 {
1005         PMD_INIT_FUNC_TRACE();
1006
1007         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1008                 return 0;
1009
1010         igbvf_dev_close(eth_dev);
1011
1012         return 0;
1013 }
1014
1015 static int eth_igb_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1016         struct rte_pci_device *pci_dev)
1017 {
1018         return rte_eth_dev_pci_generic_probe(pci_dev,
1019                 sizeof(struct e1000_adapter), eth_igb_dev_init);
1020 }
1021
1022 static int eth_igb_pci_remove(struct rte_pci_device *pci_dev)
1023 {
1024         return rte_eth_dev_pci_generic_remove(pci_dev, eth_igb_dev_uninit);
1025 }
1026
1027 static struct rte_pci_driver rte_igb_pmd = {
1028         .id_table = pci_id_igb_map,
1029         .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC,
1030         .probe = eth_igb_pci_probe,
1031         .remove = eth_igb_pci_remove,
1032 };
1033
1034
1035 static int eth_igbvf_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1036         struct rte_pci_device *pci_dev)
1037 {
1038         return rte_eth_dev_pci_generic_probe(pci_dev,
1039                 sizeof(struct e1000_adapter), eth_igbvf_dev_init);
1040 }
1041
1042 static int eth_igbvf_pci_remove(struct rte_pci_device *pci_dev)
1043 {
1044         return rte_eth_dev_pci_generic_remove(pci_dev, eth_igbvf_dev_uninit);
1045 }
1046
1047 /*
1048  * virtual function driver struct
1049  */
1050 static struct rte_pci_driver rte_igbvf_pmd = {
1051         .id_table = pci_id_igbvf_map,
1052         .drv_flags = RTE_PCI_DRV_NEED_MAPPING,
1053         .probe = eth_igbvf_pci_probe,
1054         .remove = eth_igbvf_pci_remove,
1055 };
1056
1057 static void
1058 igb_vmdq_vlan_hw_filter_enable(struct rte_eth_dev *dev)
1059 {
1060         struct e1000_hw *hw =
1061                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1062         /* RCTL: enable VLAN filter since VMDq always use VLAN filter */
1063         uint32_t rctl = E1000_READ_REG(hw, E1000_RCTL);
1064         rctl |= E1000_RCTL_VFE;
1065         E1000_WRITE_REG(hw, E1000_RCTL, rctl);
1066 }
1067
1068 static int
1069 igb_check_mq_mode(struct rte_eth_dev *dev)
1070 {
1071         enum rte_eth_rx_mq_mode rx_mq_mode = dev->data->dev_conf.rxmode.mq_mode;
1072         enum rte_eth_tx_mq_mode tx_mq_mode = dev->data->dev_conf.txmode.mq_mode;
1073         uint16_t nb_rx_q = dev->data->nb_rx_queues;
1074         uint16_t nb_tx_q = dev->data->nb_tx_queues;
1075
1076         if ((rx_mq_mode & RTE_ETH_MQ_RX_DCB_FLAG) ||
1077             tx_mq_mode == RTE_ETH_MQ_TX_DCB ||
1078             tx_mq_mode == RTE_ETH_MQ_TX_VMDQ_DCB) {
1079                 PMD_INIT_LOG(ERR, "DCB mode is not supported.");
1080                 return -EINVAL;
1081         }
1082         if (RTE_ETH_DEV_SRIOV(dev).active != 0) {
1083                 /* Check multi-queue mode.
1084                  * To no break software we accept RTE_ETH_MQ_RX_NONE as this might
1085                  * be used to turn off VLAN filter.
1086                  */
1087
1088                 if (rx_mq_mode == RTE_ETH_MQ_RX_NONE ||
1089                     rx_mq_mode == RTE_ETH_MQ_RX_VMDQ_ONLY) {
1090                         dev->data->dev_conf.rxmode.mq_mode = RTE_ETH_MQ_RX_VMDQ_ONLY;
1091                         RTE_ETH_DEV_SRIOV(dev).nb_q_per_pool = 1;
1092                 } else {
1093                         /* Only support one queue on VFs.
1094                          * RSS together with SRIOV is not supported.
1095                          */
1096                         PMD_INIT_LOG(ERR, "SRIOV is active,"
1097                                         " wrong mq_mode rx %d.",
1098                                         rx_mq_mode);
1099                         return -EINVAL;
1100                 }
1101                 /* TX mode is not used here, so mode might be ignored.*/
1102                 if (tx_mq_mode != RTE_ETH_MQ_TX_VMDQ_ONLY) {
1103                         /* SRIOV only works in VMDq enable mode */
1104                         PMD_INIT_LOG(WARNING, "SRIOV is active,"
1105                                         " TX mode %d is not supported. "
1106                                         " Driver will behave as %d mode.",
1107                                         tx_mq_mode, RTE_ETH_MQ_TX_VMDQ_ONLY);
1108                 }
1109
1110                 /* check valid queue number */
1111                 if ((nb_rx_q > 1) || (nb_tx_q > 1)) {
1112                         PMD_INIT_LOG(ERR, "SRIOV is active,"
1113                                         " only support one queue on VFs.");
1114                         return -EINVAL;
1115                 }
1116         } else {
1117                 /* To no break software that set invalid mode, only display
1118                  * warning if invalid mode is used.
1119                  */
1120                 if (rx_mq_mode != RTE_ETH_MQ_RX_NONE &&
1121                     rx_mq_mode != RTE_ETH_MQ_RX_VMDQ_ONLY &&
1122                     rx_mq_mode != RTE_ETH_MQ_RX_RSS) {
1123                         /* RSS together with VMDq not supported*/
1124                         PMD_INIT_LOG(ERR, "RX mode %d is not supported.",
1125                                      rx_mq_mode);
1126                         return -EINVAL;
1127                 }
1128
1129                 if (tx_mq_mode != RTE_ETH_MQ_TX_NONE &&
1130                     tx_mq_mode != RTE_ETH_MQ_TX_VMDQ_ONLY) {
1131                         PMD_INIT_LOG(WARNING, "TX mode %d is not supported."
1132                                         " Due to txmode is meaningless in this"
1133                                         " driver, just ignore.",
1134                                         tx_mq_mode);
1135                 }
1136         }
1137         return 0;
1138 }
1139
1140 static int
1141 eth_igb_configure(struct rte_eth_dev *dev)
1142 {
1143         struct e1000_interrupt *intr =
1144                 E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
1145         int ret;
1146
1147         PMD_INIT_FUNC_TRACE();
1148
1149         if (dev->data->dev_conf.rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG)
1150                 dev->data->dev_conf.rxmode.offloads |= RTE_ETH_RX_OFFLOAD_RSS_HASH;
1151
1152         /* multipe queue mode checking */
1153         ret  = igb_check_mq_mode(dev);
1154         if (ret != 0) {
1155                 PMD_DRV_LOG(ERR, "igb_check_mq_mode fails with %d.",
1156                             ret);
1157                 return ret;
1158         }
1159
1160         intr->flags |= E1000_FLAG_NEED_LINK_UPDATE;
1161         PMD_INIT_FUNC_TRACE();
1162
1163         return 0;
1164 }
1165
1166 static void
1167 eth_igb_rxtx_control(struct rte_eth_dev *dev,
1168                      bool enable)
1169 {
1170         struct e1000_hw *hw =
1171                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1172         uint32_t tctl, rctl;
1173
1174         tctl = E1000_READ_REG(hw, E1000_TCTL);
1175         rctl = E1000_READ_REG(hw, E1000_RCTL);
1176
1177         if (enable) {
1178                 /* enable Tx/Rx */
1179                 tctl |= E1000_TCTL_EN;
1180                 rctl |= E1000_RCTL_EN;
1181         } else {
1182                 /* disable Tx/Rx */
1183                 tctl &= ~E1000_TCTL_EN;
1184                 rctl &= ~E1000_RCTL_EN;
1185         }
1186         E1000_WRITE_REG(hw, E1000_TCTL, tctl);
1187         E1000_WRITE_REG(hw, E1000_RCTL, rctl);
1188         E1000_WRITE_FLUSH(hw);
1189 }
1190
1191 static int
1192 eth_igb_start(struct rte_eth_dev *dev)
1193 {
1194         struct e1000_hw *hw =
1195                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1196         struct e1000_adapter *adapter =
1197                 E1000_DEV_PRIVATE(dev->data->dev_private);
1198         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1199         struct rte_intr_handle *intr_handle = pci_dev->intr_handle;
1200         int ret, mask;
1201         uint32_t intr_vector = 0;
1202         uint32_t ctrl_ext;
1203         uint32_t *speeds;
1204         int num_speeds;
1205         bool autoneg;
1206
1207         PMD_INIT_FUNC_TRACE();
1208
1209         /* disable uio/vfio intr/eventfd mapping */
1210         rte_intr_disable(intr_handle);
1211
1212         /* Power up the phy. Needed to make the link go Up */
1213         eth_igb_dev_set_link_up(dev);
1214
1215         /*
1216          * Packet Buffer Allocation (PBA)
1217          * Writing PBA sets the receive portion of the buffer
1218          * the remainder is used for the transmit buffer.
1219          */
1220         if (hw->mac.type == e1000_82575) {
1221                 uint32_t pba;
1222
1223                 pba = E1000_PBA_32K; /* 32K for Rx, 16K for Tx */
1224                 E1000_WRITE_REG(hw, E1000_PBA, pba);
1225         }
1226
1227         /* Put the address into the Receive Address Array */
1228         e1000_rar_set(hw, hw->mac.addr, 0);
1229
1230         /* Initialize the hardware */
1231         if (igb_hardware_init(hw)) {
1232                 PMD_INIT_LOG(ERR, "Unable to initialize the hardware");
1233                 return -EIO;
1234         }
1235         adapter->stopped = 0;
1236
1237         E1000_WRITE_REG(hw, E1000_VET,
1238                         RTE_ETHER_TYPE_VLAN << 16 | RTE_ETHER_TYPE_VLAN);
1239
1240         ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
1241         /* Set PF Reset Done bit so PF/VF Mail Ops can work */
1242         ctrl_ext |= E1000_CTRL_EXT_PFRSTD;
1243         E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
1244         E1000_WRITE_FLUSH(hw);
1245
1246         /* configure PF module if SRIOV enabled */
1247         igb_pf_host_configure(dev);
1248
1249         /* check and configure queue intr-vector mapping */
1250         if ((rte_intr_cap_multiple(intr_handle) ||
1251              !RTE_ETH_DEV_SRIOV(dev).active) &&
1252             dev->data->dev_conf.intr_conf.rxq != 0) {
1253                 intr_vector = dev->data->nb_rx_queues;
1254                 if (rte_intr_efd_enable(intr_handle, intr_vector))
1255                         return -1;
1256         }
1257
1258         /* Allocate the vector list */
1259         if (rte_intr_dp_is_en(intr_handle)) {
1260                 if (rte_intr_vec_list_alloc(intr_handle, "intr_vec",
1261                                                    dev->data->nb_rx_queues)) {
1262                         PMD_INIT_LOG(ERR, "Failed to allocate %d rx_queues"
1263                                      " intr_vec", dev->data->nb_rx_queues);
1264                         return -ENOMEM;
1265                 }
1266         }
1267
1268         /* confiugre msix for rx interrupt */
1269         eth_igb_configure_msix_intr(dev);
1270
1271         /* Configure for OS presence */
1272         igb_init_manageability(hw);
1273
1274         eth_igb_tx_init(dev);
1275
1276         /* This can fail when allocating mbufs for descriptor rings */
1277         ret = eth_igb_rx_init(dev);
1278         if (ret) {
1279                 PMD_INIT_LOG(ERR, "Unable to initialize RX hardware");
1280                 igb_dev_clear_queues(dev);
1281                 return ret;
1282         }
1283
1284         e1000_clear_hw_cntrs_base_generic(hw);
1285
1286         /*
1287          * VLAN Offload Settings
1288          */
1289         mask = RTE_ETH_VLAN_STRIP_MASK | RTE_ETH_VLAN_FILTER_MASK |
1290                         RTE_ETH_VLAN_EXTEND_MASK;
1291         ret = eth_igb_vlan_offload_set(dev, mask);
1292         if (ret) {
1293                 PMD_INIT_LOG(ERR, "Unable to set vlan offload");
1294                 igb_dev_clear_queues(dev);
1295                 return ret;
1296         }
1297
1298         if (dev->data->dev_conf.rxmode.mq_mode == RTE_ETH_MQ_RX_VMDQ_ONLY) {
1299                 /* Enable VLAN filter since VMDq always use VLAN filter */
1300                 igb_vmdq_vlan_hw_filter_enable(dev);
1301         }
1302
1303         if ((hw->mac.type == e1000_82576) || (hw->mac.type == e1000_82580) ||
1304                 (hw->mac.type == e1000_i350) || (hw->mac.type == e1000_i210) ||
1305                 (hw->mac.type == e1000_i211)) {
1306                 /* Configure EITR with the maximum possible value (0xFFFF) */
1307                 E1000_WRITE_REG(hw, E1000_EITR(0), 0xFFFF);
1308         }
1309
1310         /* Setup link speed and duplex */
1311         speeds = &dev->data->dev_conf.link_speeds;
1312         if (*speeds == RTE_ETH_LINK_SPEED_AUTONEG) {
1313                 hw->phy.autoneg_advertised = E1000_ALL_SPEED_DUPLEX;
1314                 hw->mac.autoneg = 1;
1315         } else {
1316                 num_speeds = 0;
1317                 autoneg = (*speeds & RTE_ETH_LINK_SPEED_FIXED) == 0;
1318
1319                 /* Reset */
1320                 hw->phy.autoneg_advertised = 0;
1321
1322                 if (*speeds & ~(RTE_ETH_LINK_SPEED_10M_HD | RTE_ETH_LINK_SPEED_10M |
1323                                 RTE_ETH_LINK_SPEED_100M_HD | RTE_ETH_LINK_SPEED_100M |
1324                                 RTE_ETH_LINK_SPEED_1G | RTE_ETH_LINK_SPEED_FIXED)) {
1325                         num_speeds = -1;
1326                         goto error_invalid_config;
1327                 }
1328                 if (*speeds & RTE_ETH_LINK_SPEED_10M_HD) {
1329                         hw->phy.autoneg_advertised |= ADVERTISE_10_HALF;
1330                         num_speeds++;
1331                 }
1332                 if (*speeds & RTE_ETH_LINK_SPEED_10M) {
1333                         hw->phy.autoneg_advertised |= ADVERTISE_10_FULL;
1334                         num_speeds++;
1335                 }
1336                 if (*speeds & RTE_ETH_LINK_SPEED_100M_HD) {
1337                         hw->phy.autoneg_advertised |= ADVERTISE_100_HALF;
1338                         num_speeds++;
1339                 }
1340                 if (*speeds & RTE_ETH_LINK_SPEED_100M) {
1341                         hw->phy.autoneg_advertised |= ADVERTISE_100_FULL;
1342                         num_speeds++;
1343                 }
1344                 if (*speeds & RTE_ETH_LINK_SPEED_1G) {
1345                         hw->phy.autoneg_advertised |= ADVERTISE_1000_FULL;
1346                         num_speeds++;
1347                 }
1348                 if (num_speeds == 0 || (!autoneg && (num_speeds > 1)))
1349                         goto error_invalid_config;
1350
1351                 /* Set/reset the mac.autoneg based on the link speed,
1352                  * fixed or not
1353                  */
1354                 if (!autoneg) {
1355                         hw->mac.autoneg = 0;
1356                         hw->mac.forced_speed_duplex =
1357                                         hw->phy.autoneg_advertised;
1358                 } else {
1359                         hw->mac.autoneg = 1;
1360                 }
1361         }
1362
1363         e1000_setup_link(hw);
1364
1365         if (rte_intr_allow_others(intr_handle)) {
1366                 /* check if lsc interrupt is enabled */
1367                 if (dev->data->dev_conf.intr_conf.lsc != 0)
1368                         eth_igb_lsc_interrupt_setup(dev, TRUE);
1369                 else
1370                         eth_igb_lsc_interrupt_setup(dev, FALSE);
1371         } else {
1372                 rte_intr_callback_unregister(intr_handle,
1373                                              eth_igb_interrupt_handler,
1374                                              (void *)dev);
1375                 if (dev->data->dev_conf.intr_conf.lsc != 0)
1376                         PMD_INIT_LOG(INFO, "lsc won't enable because of"
1377                                      " no intr multiplex");
1378         }
1379
1380         /* check if rxq interrupt is enabled */
1381         if (dev->data->dev_conf.intr_conf.rxq != 0 &&
1382             rte_intr_dp_is_en(intr_handle))
1383                 eth_igb_rxq_interrupt_setup(dev);
1384
1385         /* enable uio/vfio intr/eventfd mapping */
1386         rte_intr_enable(intr_handle);
1387
1388         /* resume enabled intr since hw reset */
1389         igb_intr_enable(dev);
1390
1391         /* restore all types filter */
1392         igb_filter_restore(dev);
1393
1394         eth_igb_rxtx_control(dev, true);
1395         eth_igb_link_update(dev, 0);
1396
1397         PMD_INIT_LOG(DEBUG, "<<");
1398
1399         return 0;
1400
1401 error_invalid_config:
1402         PMD_INIT_LOG(ERR, "Invalid advertised speeds (%u) for port %u",
1403                      dev->data->dev_conf.link_speeds, dev->data->port_id);
1404         igb_dev_clear_queues(dev);
1405         return -EINVAL;
1406 }
1407
1408 /*********************************************************************
1409  *
1410  *  This routine disables all traffic on the adapter by issuing a
1411  *  global reset on the MAC.
1412  *
1413  **********************************************************************/
1414 static int
1415 eth_igb_stop(struct rte_eth_dev *dev)
1416 {
1417         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1418         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1419         struct rte_eth_link link;
1420         struct rte_intr_handle *intr_handle = pci_dev->intr_handle;
1421         struct e1000_adapter *adapter =
1422                 E1000_DEV_PRIVATE(dev->data->dev_private);
1423
1424         if (adapter->stopped)
1425                 return 0;
1426
1427         eth_igb_rxtx_control(dev, false);
1428
1429         igb_intr_disable(dev);
1430
1431         /* disable intr eventfd mapping */
1432         rte_intr_disable(intr_handle);
1433
1434         igb_pf_reset_hw(hw);
1435         E1000_WRITE_REG(hw, E1000_WUC, 0);
1436
1437         /* Set bit for Go Link disconnect if PHY reset is not blocked */
1438         if (hw->mac.type >= e1000_82580 &&
1439             (e1000_check_reset_block(hw) != E1000_BLK_PHY_RESET)) {
1440                 uint32_t phpm_reg;
1441
1442                 phpm_reg = E1000_READ_REG(hw, E1000_82580_PHY_POWER_MGMT);
1443                 phpm_reg |= E1000_82580_PM_GO_LINKD;
1444                 E1000_WRITE_REG(hw, E1000_82580_PHY_POWER_MGMT, phpm_reg);
1445         }
1446
1447         /* Power down the phy. Needed to make the link go Down */
1448         eth_igb_dev_set_link_down(dev);
1449
1450         igb_dev_clear_queues(dev);
1451
1452         /* clear the recorded link status */
1453         memset(&link, 0, sizeof(link));
1454         rte_eth_linkstatus_set(dev, &link);
1455
1456         if (!rte_intr_allow_others(intr_handle))
1457                 /* resume to the default handler */
1458                 rte_intr_callback_register(intr_handle,
1459                                            eth_igb_interrupt_handler,
1460                                            (void *)dev);
1461
1462         /* Clean datapath event and queue/vec mapping */
1463         rte_intr_efd_disable(intr_handle);
1464         rte_intr_vec_list_free(intr_handle);
1465
1466         adapter->stopped = true;
1467         dev->data->dev_started = 0;
1468
1469         return 0;
1470 }
1471
1472 static int
1473 eth_igb_dev_set_link_up(struct rte_eth_dev *dev)
1474 {
1475         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1476
1477         if (hw->phy.media_type == e1000_media_type_copper)
1478                 e1000_power_up_phy(hw);
1479         else
1480                 e1000_power_up_fiber_serdes_link(hw);
1481
1482         return 0;
1483 }
1484
1485 static int
1486 eth_igb_dev_set_link_down(struct rte_eth_dev *dev)
1487 {
1488         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1489
1490         if (hw->phy.media_type == e1000_media_type_copper)
1491                 e1000_power_down_phy(hw);
1492         else
1493                 e1000_shutdown_fiber_serdes_link(hw);
1494
1495         return 0;
1496 }
1497
1498 static int
1499 eth_igb_close(struct rte_eth_dev *dev)
1500 {
1501         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1502         struct rte_eth_link link;
1503         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1504         struct rte_intr_handle *intr_handle = pci_dev->intr_handle;
1505         struct e1000_filter_info *filter_info =
1506                 E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
1507         int ret;
1508
1509         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1510                 return 0;
1511
1512         ret = eth_igb_stop(dev);
1513
1514         e1000_phy_hw_reset(hw);
1515         igb_release_manageability(hw);
1516         igb_hw_control_release(hw);
1517
1518         /* Clear bit for Go Link disconnect if PHY reset is not blocked */
1519         if (hw->mac.type >= e1000_82580 &&
1520             (e1000_check_reset_block(hw) != E1000_BLK_PHY_RESET)) {
1521                 uint32_t phpm_reg;
1522
1523                 phpm_reg = E1000_READ_REG(hw, E1000_82580_PHY_POWER_MGMT);
1524                 phpm_reg &= ~E1000_82580_PM_GO_LINKD;
1525                 E1000_WRITE_REG(hw, E1000_82580_PHY_POWER_MGMT, phpm_reg);
1526         }
1527
1528         igb_dev_free_queues(dev);
1529
1530         /* Cleanup vector list */
1531         rte_intr_vec_list_free(intr_handle);
1532
1533         memset(&link, 0, sizeof(link));
1534         rte_eth_linkstatus_set(dev, &link);
1535
1536         /* Reset any pending lock */
1537         igb_reset_swfw_lock(hw);
1538
1539         /* uninitialize PF if max_vfs not zero */
1540         igb_pf_host_uninit(dev);
1541
1542         rte_intr_callback_unregister(intr_handle,
1543                                      eth_igb_interrupt_handler, dev);
1544
1545         /* clear the SYN filter info */
1546         filter_info->syn_info = 0;
1547
1548         /* clear the ethertype filters info */
1549         filter_info->ethertype_mask = 0;
1550         memset(filter_info->ethertype_filters, 0,
1551                 E1000_MAX_ETQF_FILTERS * sizeof(struct igb_ethertype_filter));
1552
1553         /* clear the rss filter info */
1554         memset(&filter_info->rss_info, 0,
1555                 sizeof(struct igb_rte_flow_rss_conf));
1556
1557         /* remove all ntuple filters of the device */
1558         igb_ntuple_filter_uninit(dev);
1559
1560         /* remove all flex filters of the device */
1561         igb_flex_filter_uninit(dev);
1562
1563         /* clear all the filters list */
1564         igb_filterlist_flush(dev);
1565
1566         return ret;
1567 }
1568
1569 /*
1570  * Reset PF device.
1571  */
1572 static int
1573 eth_igb_reset(struct rte_eth_dev *dev)
1574 {
1575         int ret;
1576
1577         /* When a DPDK PMD PF begin to reset PF port, it should notify all
1578          * its VF to make them align with it. The detailed notification
1579          * mechanism is PMD specific and is currently not implemented.
1580          * To avoid unexpected behavior in VF, currently reset of PF with
1581          * SR-IOV activation is not supported. It might be supported later.
1582          */
1583         if (dev->data->sriov.active)
1584                 return -ENOTSUP;
1585
1586         ret = eth_igb_dev_uninit(dev);
1587         if (ret)
1588                 return ret;
1589
1590         ret = eth_igb_dev_init(dev);
1591
1592         return ret;
1593 }
1594
1595
1596 static int
1597 igb_get_rx_buffer_size(struct e1000_hw *hw)
1598 {
1599         uint32_t rx_buf_size;
1600         if (hw->mac.type == e1000_82576) {
1601                 rx_buf_size = (E1000_READ_REG(hw, E1000_RXPBS) & 0xffff) << 10;
1602         } else if (hw->mac.type == e1000_82580 || hw->mac.type == e1000_i350) {
1603                 /* PBS needs to be translated according to a lookup table */
1604                 rx_buf_size = (E1000_READ_REG(hw, E1000_RXPBS) & 0xf);
1605                 rx_buf_size = (uint32_t) e1000_rxpbs_adjust_82580(rx_buf_size);
1606                 rx_buf_size = (rx_buf_size << 10);
1607         } else if (hw->mac.type == e1000_i210 || hw->mac.type == e1000_i211) {
1608                 rx_buf_size = (E1000_READ_REG(hw, E1000_RXPBS) & 0x3f) << 10;
1609         } else {
1610                 rx_buf_size = (E1000_READ_REG(hw, E1000_PBA) & 0xffff) << 10;
1611         }
1612
1613         return rx_buf_size;
1614 }
1615
1616 /*********************************************************************
1617  *
1618  *  Initialize the hardware
1619  *
1620  **********************************************************************/
1621 static int
1622 igb_hardware_init(struct e1000_hw *hw)
1623 {
1624         uint32_t rx_buf_size;
1625         int diag;
1626
1627         /* Let the firmware know the OS is in control */
1628         igb_hw_control_acquire(hw);
1629
1630         /*
1631          * These parameters control the automatic generation (Tx) and
1632          * response (Rx) to Ethernet PAUSE frames.
1633          * - High water mark should allow for at least two standard size (1518)
1634          *   frames to be received after sending an XOFF.
1635          * - Low water mark works best when it is very near the high water mark.
1636          *   This allows the receiver to restart by sending XON when it has
1637          *   drained a bit. Here we use an arbitrary value of 1500 which will
1638          *   restart after one full frame is pulled from the buffer. There
1639          *   could be several smaller frames in the buffer and if so they will
1640          *   not trigger the XON until their total number reduces the buffer
1641          *   by 1500.
1642          * - The pause time is fairly large at 1000 x 512ns = 512 usec.
1643          */
1644         rx_buf_size = igb_get_rx_buffer_size(hw);
1645
1646         hw->fc.high_water = rx_buf_size - (RTE_ETHER_MAX_LEN * 2);
1647         hw->fc.low_water = hw->fc.high_water - 1500;
1648         hw->fc.pause_time = IGB_FC_PAUSE_TIME;
1649         hw->fc.send_xon = 1;
1650
1651         /* Set Flow control, use the tunable location if sane */
1652         if ((igb_fc_setting != e1000_fc_none) && (igb_fc_setting < 4))
1653                 hw->fc.requested_mode = igb_fc_setting;
1654         else
1655                 hw->fc.requested_mode = e1000_fc_none;
1656
1657         /* Issue a global reset */
1658         igb_pf_reset_hw(hw);
1659         E1000_WRITE_REG(hw, E1000_WUC, 0);
1660
1661         diag = e1000_init_hw(hw);
1662         if (diag < 0)
1663                 return diag;
1664
1665         E1000_WRITE_REG(hw, E1000_VET,
1666                         RTE_ETHER_TYPE_VLAN << 16 | RTE_ETHER_TYPE_VLAN);
1667         e1000_get_phy_info(hw);
1668         e1000_check_for_link(hw);
1669
1670         return 0;
1671 }
1672
1673 /* This function is based on igb_update_stats_counters() in igb/if_igb.c */
1674 static void
1675 igb_read_stats_registers(struct e1000_hw *hw, struct e1000_hw_stats *stats)
1676 {
1677         int pause_frames;
1678
1679         uint64_t old_gprc  = stats->gprc;
1680         uint64_t old_gptc  = stats->gptc;
1681         uint64_t old_tpr   = stats->tpr;
1682         uint64_t old_tpt   = stats->tpt;
1683         uint64_t old_rpthc = stats->rpthc;
1684         uint64_t old_hgptc = stats->hgptc;
1685
1686         if(hw->phy.media_type == e1000_media_type_copper ||
1687             (E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU)) {
1688                 stats->symerrs +=
1689                     E1000_READ_REG(hw,E1000_SYMERRS);
1690                 stats->sec += E1000_READ_REG(hw, E1000_SEC);
1691         }
1692
1693         stats->crcerrs += E1000_READ_REG(hw, E1000_CRCERRS);
1694         stats->mpc += E1000_READ_REG(hw, E1000_MPC);
1695         stats->scc += E1000_READ_REG(hw, E1000_SCC);
1696         stats->ecol += E1000_READ_REG(hw, E1000_ECOL);
1697
1698         stats->mcc += E1000_READ_REG(hw, E1000_MCC);
1699         stats->latecol += E1000_READ_REG(hw, E1000_LATECOL);
1700         stats->colc += E1000_READ_REG(hw, E1000_COLC);
1701         stats->dc += E1000_READ_REG(hw, E1000_DC);
1702         stats->rlec += E1000_READ_REG(hw, E1000_RLEC);
1703         stats->xonrxc += E1000_READ_REG(hw, E1000_XONRXC);
1704         stats->xontxc += E1000_READ_REG(hw, E1000_XONTXC);
1705         /*
1706         ** For watchdog management we need to know if we have been
1707         ** paused during the last interval, so capture that here.
1708         */
1709         pause_frames = E1000_READ_REG(hw, E1000_XOFFRXC);
1710         stats->xoffrxc += pause_frames;
1711         stats->xofftxc += E1000_READ_REG(hw, E1000_XOFFTXC);
1712         stats->fcruc += E1000_READ_REG(hw, E1000_FCRUC);
1713         stats->prc64 += E1000_READ_REG(hw, E1000_PRC64);
1714         stats->prc127 += E1000_READ_REG(hw, E1000_PRC127);
1715         stats->prc255 += E1000_READ_REG(hw, E1000_PRC255);
1716         stats->prc511 += E1000_READ_REG(hw, E1000_PRC511);
1717         stats->prc1023 += E1000_READ_REG(hw, E1000_PRC1023);
1718         stats->prc1522 += E1000_READ_REG(hw, E1000_PRC1522);
1719         stats->gprc += E1000_READ_REG(hw, E1000_GPRC);
1720         stats->bprc += E1000_READ_REG(hw, E1000_BPRC);
1721         stats->mprc += E1000_READ_REG(hw, E1000_MPRC);
1722         stats->gptc += E1000_READ_REG(hw, E1000_GPTC);
1723
1724         /* For the 64-bit byte counters the low dword must be read first. */
1725         /* Both registers clear on the read of the high dword */
1726
1727         /* Workaround CRC bytes included in size, take away 4 bytes/packet */
1728         stats->gorc += E1000_READ_REG(hw, E1000_GORCL);
1729         stats->gorc += ((uint64_t)E1000_READ_REG(hw, E1000_GORCH) << 32);
1730         stats->gorc -= (stats->gprc - old_gprc) * RTE_ETHER_CRC_LEN;
1731         stats->gotc += E1000_READ_REG(hw, E1000_GOTCL);
1732         stats->gotc += ((uint64_t)E1000_READ_REG(hw, E1000_GOTCH) << 32);
1733         stats->gotc -= (stats->gptc - old_gptc) * RTE_ETHER_CRC_LEN;
1734
1735         stats->rnbc += E1000_READ_REG(hw, E1000_RNBC);
1736         stats->ruc += E1000_READ_REG(hw, E1000_RUC);
1737         stats->rfc += E1000_READ_REG(hw, E1000_RFC);
1738         stats->roc += E1000_READ_REG(hw, E1000_ROC);
1739         stats->rjc += E1000_READ_REG(hw, E1000_RJC);
1740
1741         stats->tpr += E1000_READ_REG(hw, E1000_TPR);
1742         stats->tpt += E1000_READ_REG(hw, E1000_TPT);
1743
1744         stats->tor += E1000_READ_REG(hw, E1000_TORL);
1745         stats->tor += ((uint64_t)E1000_READ_REG(hw, E1000_TORH) << 32);
1746         stats->tor -= (stats->tpr - old_tpr) * RTE_ETHER_CRC_LEN;
1747         stats->tot += E1000_READ_REG(hw, E1000_TOTL);
1748         stats->tot += ((uint64_t)E1000_READ_REG(hw, E1000_TOTH) << 32);
1749         stats->tot -= (stats->tpt - old_tpt) * RTE_ETHER_CRC_LEN;
1750
1751         stats->ptc64 += E1000_READ_REG(hw, E1000_PTC64);
1752         stats->ptc127 += E1000_READ_REG(hw, E1000_PTC127);
1753         stats->ptc255 += E1000_READ_REG(hw, E1000_PTC255);
1754         stats->ptc511 += E1000_READ_REG(hw, E1000_PTC511);
1755         stats->ptc1023 += E1000_READ_REG(hw, E1000_PTC1023);
1756         stats->ptc1522 += E1000_READ_REG(hw, E1000_PTC1522);
1757         stats->mptc += E1000_READ_REG(hw, E1000_MPTC);
1758         stats->bptc += E1000_READ_REG(hw, E1000_BPTC);
1759
1760         /* Interrupt Counts */
1761
1762         stats->iac += E1000_READ_REG(hw, E1000_IAC);
1763         stats->icrxptc += E1000_READ_REG(hw, E1000_ICRXPTC);
1764         stats->icrxatc += E1000_READ_REG(hw, E1000_ICRXATC);
1765         stats->ictxptc += E1000_READ_REG(hw, E1000_ICTXPTC);
1766         stats->ictxatc += E1000_READ_REG(hw, E1000_ICTXATC);
1767         stats->ictxqec += E1000_READ_REG(hw, E1000_ICTXQEC);
1768         stats->ictxqmtc += E1000_READ_REG(hw, E1000_ICTXQMTC);
1769         stats->icrxdmtc += E1000_READ_REG(hw, E1000_ICRXDMTC);
1770         stats->icrxoc += E1000_READ_REG(hw, E1000_ICRXOC);
1771
1772         /* Host to Card Statistics */
1773
1774         stats->cbtmpc += E1000_READ_REG(hw, E1000_CBTMPC);
1775         stats->htdpmc += E1000_READ_REG(hw, E1000_HTDPMC);
1776         stats->cbrdpc += E1000_READ_REG(hw, E1000_CBRDPC);
1777         stats->cbrmpc += E1000_READ_REG(hw, E1000_CBRMPC);
1778         stats->rpthc += E1000_READ_REG(hw, E1000_RPTHC);
1779         stats->hgptc += E1000_READ_REG(hw, E1000_HGPTC);
1780         stats->htcbdpc += E1000_READ_REG(hw, E1000_HTCBDPC);
1781         stats->hgorc += E1000_READ_REG(hw, E1000_HGORCL);
1782         stats->hgorc += ((uint64_t)E1000_READ_REG(hw, E1000_HGORCH) << 32);
1783         stats->hgorc -= (stats->rpthc - old_rpthc) * RTE_ETHER_CRC_LEN;
1784         stats->hgotc += E1000_READ_REG(hw, E1000_HGOTCL);
1785         stats->hgotc += ((uint64_t)E1000_READ_REG(hw, E1000_HGOTCH) << 32);
1786         stats->hgotc -= (stats->hgptc - old_hgptc) * RTE_ETHER_CRC_LEN;
1787         stats->lenerrs += E1000_READ_REG(hw, E1000_LENERRS);
1788         stats->scvpc += E1000_READ_REG(hw, E1000_SCVPC);
1789         stats->hrmpc += E1000_READ_REG(hw, E1000_HRMPC);
1790
1791         stats->algnerrc += E1000_READ_REG(hw, E1000_ALGNERRC);
1792         stats->rxerrc += E1000_READ_REG(hw, E1000_RXERRC);
1793         stats->tncrs += E1000_READ_REG(hw, E1000_TNCRS);
1794         stats->cexterr += E1000_READ_REG(hw, E1000_CEXTERR);
1795         stats->tsctc += E1000_READ_REG(hw, E1000_TSCTC);
1796         stats->tsctfc += E1000_READ_REG(hw, E1000_TSCTFC);
1797 }
1798
1799 static int
1800 eth_igb_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *rte_stats)
1801 {
1802         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1803         struct e1000_hw_stats *stats =
1804                         E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
1805
1806         igb_read_stats_registers(hw, stats);
1807
1808         if (rte_stats == NULL)
1809                 return -EINVAL;
1810
1811         /* Rx Errors */
1812         rte_stats->imissed = stats->mpc;
1813         rte_stats->ierrors = stats->crcerrs + stats->rlec +
1814                              stats->rxerrc + stats->algnerrc + stats->cexterr;
1815
1816         /* Tx Errors */
1817         rte_stats->oerrors = stats->ecol + stats->latecol;
1818
1819         rte_stats->ipackets = stats->gprc;
1820         rte_stats->opackets = stats->gptc;
1821         rte_stats->ibytes   = stats->gorc;
1822         rte_stats->obytes   = stats->gotc;
1823         return 0;
1824 }
1825
1826 static int
1827 eth_igb_stats_reset(struct rte_eth_dev *dev)
1828 {
1829         struct e1000_hw_stats *hw_stats =
1830                         E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
1831
1832         /* HW registers are cleared on read */
1833         eth_igb_stats_get(dev, NULL);
1834
1835         /* Reset software totals */
1836         memset(hw_stats, 0, sizeof(*hw_stats));
1837
1838         return 0;
1839 }
1840
1841 static int
1842 eth_igb_xstats_reset(struct rte_eth_dev *dev)
1843 {
1844         struct e1000_hw_stats *stats =
1845                         E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
1846
1847         /* HW registers are cleared on read */
1848         eth_igb_xstats_get(dev, NULL, IGB_NB_XSTATS);
1849
1850         /* Reset software totals */
1851         memset(stats, 0, sizeof(*stats));
1852
1853         return 0;
1854 }
1855
1856 static int eth_igb_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
1857         struct rte_eth_xstat_name *xstats_names,
1858         __rte_unused unsigned int size)
1859 {
1860         unsigned i;
1861
1862         if (xstats_names == NULL)
1863                 return IGB_NB_XSTATS;
1864
1865         /* Note: limit checked in rte_eth_xstats_names() */
1866
1867         for (i = 0; i < IGB_NB_XSTATS; i++) {
1868                 strlcpy(xstats_names[i].name, rte_igb_stats_strings[i].name,
1869                         sizeof(xstats_names[i].name));
1870         }
1871
1872         return IGB_NB_XSTATS;
1873 }
1874
1875 static int eth_igb_xstats_get_names_by_id(struct rte_eth_dev *dev,
1876                 const uint64_t *ids, struct rte_eth_xstat_name *xstats_names,
1877                 unsigned int limit)
1878 {
1879         unsigned int i;
1880
1881         if (!ids) {
1882                 if (xstats_names == NULL)
1883                         return IGB_NB_XSTATS;
1884
1885                 for (i = 0; i < IGB_NB_XSTATS; i++)
1886                         strlcpy(xstats_names[i].name,
1887                                 rte_igb_stats_strings[i].name,
1888                                 sizeof(xstats_names[i].name));
1889
1890                 return IGB_NB_XSTATS;
1891
1892         } else {
1893                 struct rte_eth_xstat_name xstats_names_copy[IGB_NB_XSTATS];
1894
1895                 eth_igb_xstats_get_names_by_id(dev, NULL, xstats_names_copy,
1896                                 IGB_NB_XSTATS);
1897
1898                 for (i = 0; i < limit; i++) {
1899                         if (ids[i] >= IGB_NB_XSTATS) {
1900                                 PMD_INIT_LOG(ERR, "id value isn't valid");
1901                                 return -1;
1902                         }
1903                         strcpy(xstats_names[i].name,
1904                                         xstats_names_copy[ids[i]].name);
1905                 }
1906                 return limit;
1907         }
1908 }
1909
1910 static int
1911 eth_igb_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1912                    unsigned n)
1913 {
1914         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1915         struct e1000_hw_stats *hw_stats =
1916                         E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
1917         unsigned i;
1918
1919         if (n < IGB_NB_XSTATS)
1920                 return IGB_NB_XSTATS;
1921
1922         igb_read_stats_registers(hw, hw_stats);
1923
1924         /* If this is a reset xstats is NULL, and we have cleared the
1925          * registers by reading them.
1926          */
1927         if (!xstats)
1928                 return 0;
1929
1930         /* Extended stats */
1931         for (i = 0; i < IGB_NB_XSTATS; i++) {
1932                 xstats[i].id = i;
1933                 xstats[i].value = *(uint64_t *)(((char *)hw_stats) +
1934                         rte_igb_stats_strings[i].offset);
1935         }
1936
1937         return IGB_NB_XSTATS;
1938 }
1939
1940 static int
1941 eth_igb_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
1942                 uint64_t *values, unsigned int n)
1943 {
1944         unsigned int i;
1945
1946         if (!ids) {
1947                 struct e1000_hw *hw =
1948                         E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1949                 struct e1000_hw_stats *hw_stats =
1950                         E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
1951
1952                 if (n < IGB_NB_XSTATS)
1953                         return IGB_NB_XSTATS;
1954
1955                 igb_read_stats_registers(hw, hw_stats);
1956
1957                 /* If this is a reset xstats is NULL, and we have cleared the
1958                  * registers by reading them.
1959                  */
1960                 if (!values)
1961                         return 0;
1962
1963                 /* Extended stats */
1964                 for (i = 0; i < IGB_NB_XSTATS; i++)
1965                         values[i] = *(uint64_t *)(((char *)hw_stats) +
1966                                         rte_igb_stats_strings[i].offset);
1967
1968                 return IGB_NB_XSTATS;
1969
1970         } else {
1971                 uint64_t values_copy[IGB_NB_XSTATS];
1972
1973                 eth_igb_xstats_get_by_id(dev, NULL, values_copy,
1974                                 IGB_NB_XSTATS);
1975
1976                 for (i = 0; i < n; i++) {
1977                         if (ids[i] >= IGB_NB_XSTATS) {
1978                                 PMD_INIT_LOG(ERR, "id value isn't valid");
1979                                 return -1;
1980                         }
1981                         values[i] = values_copy[ids[i]];
1982                 }
1983                 return n;
1984         }
1985 }
1986
1987 static void
1988 igbvf_read_stats_registers(struct e1000_hw *hw, struct e1000_vf_stats *hw_stats)
1989 {
1990         /* Good Rx packets, include VF loopback */
1991         UPDATE_VF_STAT(E1000_VFGPRC,
1992             hw_stats->last_gprc, hw_stats->gprc);
1993
1994         /* Good Rx octets, include VF loopback */
1995         UPDATE_VF_STAT(E1000_VFGORC,
1996             hw_stats->last_gorc, hw_stats->gorc);
1997
1998         /* Good Tx packets, include VF loopback */
1999         UPDATE_VF_STAT(E1000_VFGPTC,
2000             hw_stats->last_gptc, hw_stats->gptc);
2001
2002         /* Good Tx octets, include VF loopback */
2003         UPDATE_VF_STAT(E1000_VFGOTC,
2004             hw_stats->last_gotc, hw_stats->gotc);
2005
2006         /* Rx Multicst packets */
2007         UPDATE_VF_STAT(E1000_VFMPRC,
2008             hw_stats->last_mprc, hw_stats->mprc);
2009
2010         /* Good Rx loopback packets */
2011         UPDATE_VF_STAT(E1000_VFGPRLBC,
2012             hw_stats->last_gprlbc, hw_stats->gprlbc);
2013
2014         /* Good Rx loopback octets */
2015         UPDATE_VF_STAT(E1000_VFGORLBC,
2016             hw_stats->last_gorlbc, hw_stats->gorlbc);
2017
2018         /* Good Tx loopback packets */
2019         UPDATE_VF_STAT(E1000_VFGPTLBC,
2020             hw_stats->last_gptlbc, hw_stats->gptlbc);
2021
2022         /* Good Tx loopback octets */
2023         UPDATE_VF_STAT(E1000_VFGOTLBC,
2024             hw_stats->last_gotlbc, hw_stats->gotlbc);
2025 }
2026
2027 static int eth_igbvf_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
2028                                      struct rte_eth_xstat_name *xstats_names,
2029                                      __rte_unused unsigned limit)
2030 {
2031         unsigned i;
2032
2033         if (xstats_names != NULL)
2034                 for (i = 0; i < IGBVF_NB_XSTATS; i++) {
2035                         strlcpy(xstats_names[i].name,
2036                                 rte_igbvf_stats_strings[i].name,
2037                                 sizeof(xstats_names[i].name));
2038                 }
2039         return IGBVF_NB_XSTATS;
2040 }
2041
2042 static int
2043 eth_igbvf_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
2044                      unsigned n)
2045 {
2046         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2047         struct e1000_vf_stats *hw_stats = (struct e1000_vf_stats *)
2048                         E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
2049         unsigned i;
2050
2051         if (n < IGBVF_NB_XSTATS)
2052                 return IGBVF_NB_XSTATS;
2053
2054         igbvf_read_stats_registers(hw, hw_stats);
2055
2056         if (!xstats)
2057                 return 0;
2058
2059         for (i = 0; i < IGBVF_NB_XSTATS; i++) {
2060                 xstats[i].id = i;
2061                 xstats[i].value = *(uint64_t *)(((char *)hw_stats) +
2062                         rte_igbvf_stats_strings[i].offset);
2063         }
2064
2065         return IGBVF_NB_XSTATS;
2066 }
2067
2068 static int
2069 eth_igbvf_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *rte_stats)
2070 {
2071         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2072         struct e1000_vf_stats *hw_stats = (struct e1000_vf_stats *)
2073                           E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
2074
2075         igbvf_read_stats_registers(hw, hw_stats);
2076
2077         if (rte_stats == NULL)
2078                 return -EINVAL;
2079
2080         rte_stats->ipackets = hw_stats->gprc;
2081         rte_stats->ibytes = hw_stats->gorc;
2082         rte_stats->opackets = hw_stats->gptc;
2083         rte_stats->obytes = hw_stats->gotc;
2084         return 0;
2085 }
2086
2087 static int
2088 eth_igbvf_stats_reset(struct rte_eth_dev *dev)
2089 {
2090         struct e1000_vf_stats *hw_stats = (struct e1000_vf_stats*)
2091                         E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
2092
2093         /* Sync HW register to the last stats */
2094         eth_igbvf_stats_get(dev, NULL);
2095
2096         /* reset HW current stats*/
2097         memset(&hw_stats->gprc, 0, sizeof(*hw_stats) -
2098                offsetof(struct e1000_vf_stats, gprc));
2099
2100         return 0;
2101 }
2102
2103 static int
2104 eth_igb_fw_version_get(struct rte_eth_dev *dev, char *fw_version,
2105                        size_t fw_size)
2106 {
2107         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2108         struct e1000_fw_version fw;
2109         int ret;
2110
2111         e1000_get_fw_version(hw, &fw);
2112
2113         switch (hw->mac.type) {
2114         case e1000_i210:
2115         case e1000_i211:
2116                 if (!(e1000_get_flash_presence_i210(hw))) {
2117                         ret = snprintf(fw_version, fw_size,
2118                                  "%2d.%2d-%d",
2119                                  fw.invm_major, fw.invm_minor,
2120                                  fw.invm_img_type);
2121                         break;
2122                 }
2123                 /* fall through */
2124         default:
2125                 /* if option rom is valid, display its version too */
2126                 if (fw.or_valid) {
2127                         ret = snprintf(fw_version, fw_size,
2128                                  "%d.%d, 0x%08x, %d.%d.%d",
2129                                  fw.eep_major, fw.eep_minor, fw.etrack_id,
2130                                  fw.or_major, fw.or_build, fw.or_patch);
2131                 /* no option rom */
2132                 } else {
2133                         if (fw.etrack_id != 0X0000) {
2134                                 ret = snprintf(fw_version, fw_size,
2135                                          "%d.%d, 0x%08x",
2136                                          fw.eep_major, fw.eep_minor,
2137                                          fw.etrack_id);
2138                         } else {
2139                                 ret = snprintf(fw_version, fw_size,
2140                                          "%d.%d.%d",
2141                                          fw.eep_major, fw.eep_minor,
2142                                          fw.eep_build);
2143                         }
2144                 }
2145                 break;
2146         }
2147         if (ret < 0)
2148                 return -EINVAL;
2149
2150         ret += 1; /* add the size of '\0' */
2151         if (fw_size < (size_t)ret)
2152                 return ret;
2153         else
2154                 return 0;
2155 }
2156
2157 static int
2158 eth_igb_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2159 {
2160         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2161
2162         dev_info->min_rx_bufsize = 256; /* See BSIZE field of RCTL register. */
2163         dev_info->max_rx_pktlen  = 0x3FFF; /* See RLPML register. */
2164         dev_info->max_mac_addrs = hw->mac.rar_entry_count;
2165         dev_info->rx_queue_offload_capa = igb_get_rx_queue_offloads_capa(dev);
2166         dev_info->rx_offload_capa = igb_get_rx_port_offloads_capa(dev) |
2167                                     dev_info->rx_queue_offload_capa;
2168         dev_info->tx_queue_offload_capa = igb_get_tx_queue_offloads_capa(dev);
2169         dev_info->tx_offload_capa = igb_get_tx_port_offloads_capa(dev) |
2170                                     dev_info->tx_queue_offload_capa;
2171
2172         switch (hw->mac.type) {
2173         case e1000_82575:
2174                 dev_info->max_rx_queues = 4;
2175                 dev_info->max_tx_queues = 4;
2176                 dev_info->max_vmdq_pools = 0;
2177                 break;
2178
2179         case e1000_82576:
2180                 dev_info->max_rx_queues = 16;
2181                 dev_info->max_tx_queues = 16;
2182                 dev_info->max_vmdq_pools = RTE_ETH_8_POOLS;
2183                 dev_info->vmdq_queue_num = 16;
2184                 break;
2185
2186         case e1000_82580:
2187                 dev_info->max_rx_queues = 8;
2188                 dev_info->max_tx_queues = 8;
2189                 dev_info->max_vmdq_pools = RTE_ETH_8_POOLS;
2190                 dev_info->vmdq_queue_num = 8;
2191                 break;
2192
2193         case e1000_i350:
2194                 dev_info->max_rx_queues = 8;
2195                 dev_info->max_tx_queues = 8;
2196                 dev_info->max_vmdq_pools = RTE_ETH_8_POOLS;
2197                 dev_info->vmdq_queue_num = 8;
2198                 break;
2199
2200         case e1000_i354:
2201                 dev_info->max_rx_queues = 8;
2202                 dev_info->max_tx_queues = 8;
2203                 break;
2204
2205         case e1000_i210:
2206                 dev_info->max_rx_queues = 4;
2207                 dev_info->max_tx_queues = 4;
2208                 dev_info->max_vmdq_pools = 0;
2209                 break;
2210
2211         case e1000_i211:
2212                 dev_info->max_rx_queues = 2;
2213                 dev_info->max_tx_queues = 2;
2214                 dev_info->max_vmdq_pools = 0;
2215                 break;
2216
2217         default:
2218                 /* Should not happen */
2219                 return -EINVAL;
2220         }
2221         dev_info->hash_key_size = IGB_HKEY_MAX_INDEX * sizeof(uint32_t);
2222         dev_info->reta_size = RTE_ETH_RSS_RETA_SIZE_128;
2223         dev_info->flow_type_rss_offloads = IGB_RSS_OFFLOAD_ALL;
2224
2225         dev_info->default_rxconf = (struct rte_eth_rxconf) {
2226                 .rx_thresh = {
2227                         .pthresh = IGB_DEFAULT_RX_PTHRESH,
2228                         .hthresh = IGB_DEFAULT_RX_HTHRESH,
2229                         .wthresh = IGB_DEFAULT_RX_WTHRESH,
2230                 },
2231                 .rx_free_thresh = IGB_DEFAULT_RX_FREE_THRESH,
2232                 .rx_drop_en = 0,
2233                 .offloads = 0,
2234         };
2235
2236         dev_info->default_txconf = (struct rte_eth_txconf) {
2237                 .tx_thresh = {
2238                         .pthresh = IGB_DEFAULT_TX_PTHRESH,
2239                         .hthresh = IGB_DEFAULT_TX_HTHRESH,
2240                         .wthresh = IGB_DEFAULT_TX_WTHRESH,
2241                 },
2242                 .offloads = 0,
2243         };
2244
2245         dev_info->rx_desc_lim = rx_desc_lim;
2246         dev_info->tx_desc_lim = tx_desc_lim;
2247
2248         dev_info->speed_capa = RTE_ETH_LINK_SPEED_10M_HD | RTE_ETH_LINK_SPEED_10M |
2249                         RTE_ETH_LINK_SPEED_100M_HD | RTE_ETH_LINK_SPEED_100M |
2250                         RTE_ETH_LINK_SPEED_1G;
2251
2252         dev_info->max_mtu = dev_info->max_rx_pktlen - E1000_ETH_OVERHEAD;
2253         dev_info->min_mtu = RTE_ETHER_MIN_MTU;
2254
2255         return 0;
2256 }
2257
2258 static const uint32_t *
2259 eth_igb_supported_ptypes_get(struct rte_eth_dev *dev)
2260 {
2261         static const uint32_t ptypes[] = {
2262                 /* refers to igb_rxd_pkt_info_to_pkt_type() */
2263                 RTE_PTYPE_L2_ETHER,
2264                 RTE_PTYPE_L3_IPV4,
2265                 RTE_PTYPE_L3_IPV4_EXT,
2266                 RTE_PTYPE_L3_IPV6,
2267                 RTE_PTYPE_L3_IPV6_EXT,
2268                 RTE_PTYPE_L4_TCP,
2269                 RTE_PTYPE_L4_UDP,
2270                 RTE_PTYPE_L4_SCTP,
2271                 RTE_PTYPE_TUNNEL_IP,
2272                 RTE_PTYPE_INNER_L3_IPV6,
2273                 RTE_PTYPE_INNER_L3_IPV6_EXT,
2274                 RTE_PTYPE_INNER_L4_TCP,
2275                 RTE_PTYPE_INNER_L4_UDP,
2276                 RTE_PTYPE_UNKNOWN
2277         };
2278
2279         if (dev->rx_pkt_burst == eth_igb_recv_pkts ||
2280             dev->rx_pkt_burst == eth_igb_recv_scattered_pkts)
2281                 return ptypes;
2282         return NULL;
2283 }
2284
2285 static int
2286 eth_igbvf_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2287 {
2288         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2289
2290         dev_info->min_rx_bufsize = 256; /* See BSIZE field of RCTL register. */
2291         dev_info->max_rx_pktlen  = 0x3FFF; /* See RLPML register. */
2292         dev_info->max_mac_addrs = hw->mac.rar_entry_count;
2293         dev_info->tx_offload_capa = RTE_ETH_TX_OFFLOAD_VLAN_INSERT |
2294                                 RTE_ETH_TX_OFFLOAD_IPV4_CKSUM  |
2295                                 RTE_ETH_TX_OFFLOAD_UDP_CKSUM   |
2296                                 RTE_ETH_TX_OFFLOAD_TCP_CKSUM   |
2297                                 RTE_ETH_TX_OFFLOAD_SCTP_CKSUM  |
2298                                 RTE_ETH_TX_OFFLOAD_TCP_TSO;
2299         switch (hw->mac.type) {
2300         case e1000_vfadapt:
2301                 dev_info->max_rx_queues = 2;
2302                 dev_info->max_tx_queues = 2;
2303                 break;
2304         case e1000_vfadapt_i350:
2305                 dev_info->max_rx_queues = 1;
2306                 dev_info->max_tx_queues = 1;
2307                 break;
2308         default:
2309                 /* Should not happen */
2310                 return -EINVAL;
2311         }
2312
2313         dev_info->rx_queue_offload_capa = igb_get_rx_queue_offloads_capa(dev);
2314         dev_info->rx_offload_capa = igb_get_rx_port_offloads_capa(dev) |
2315                                     dev_info->rx_queue_offload_capa;
2316         dev_info->tx_queue_offload_capa = igb_get_tx_queue_offloads_capa(dev);
2317         dev_info->tx_offload_capa = igb_get_tx_port_offloads_capa(dev) |
2318                                     dev_info->tx_queue_offload_capa;
2319
2320         dev_info->default_rxconf = (struct rte_eth_rxconf) {
2321                 .rx_thresh = {
2322                         .pthresh = IGB_DEFAULT_RX_PTHRESH,
2323                         .hthresh = IGB_DEFAULT_RX_HTHRESH,
2324                         .wthresh = IGB_DEFAULT_RX_WTHRESH,
2325                 },
2326                 .rx_free_thresh = IGB_DEFAULT_RX_FREE_THRESH,
2327                 .rx_drop_en = 0,
2328                 .offloads = 0,
2329         };
2330
2331         dev_info->default_txconf = (struct rte_eth_txconf) {
2332                 .tx_thresh = {
2333                         .pthresh = IGB_DEFAULT_TX_PTHRESH,
2334                         .hthresh = IGB_DEFAULT_TX_HTHRESH,
2335                         .wthresh = IGB_DEFAULT_TX_WTHRESH,
2336                 },
2337                 .offloads = 0,
2338         };
2339
2340         dev_info->rx_desc_lim = rx_desc_lim;
2341         dev_info->tx_desc_lim = tx_desc_lim;
2342
2343         return 0;
2344 }
2345
2346 /* return 0 means link status changed, -1 means not changed */
2347 static int
2348 eth_igb_link_update(struct rte_eth_dev *dev, int wait_to_complete)
2349 {
2350         struct e1000_hw *hw =
2351                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2352         struct rte_eth_link link;
2353         int link_check, count;
2354
2355         link_check = 0;
2356         hw->mac.get_link_status = 1;
2357
2358         /* possible wait-to-complete in up to 9 seconds */
2359         for (count = 0; count < IGB_LINK_UPDATE_CHECK_TIMEOUT; count ++) {
2360                 /* Read the real link status */
2361                 switch (hw->phy.media_type) {
2362                 case e1000_media_type_copper:
2363                         /* Do the work to read phy */
2364                         e1000_check_for_link(hw);
2365                         link_check = !hw->mac.get_link_status;
2366                         break;
2367
2368                 case e1000_media_type_fiber:
2369                         e1000_check_for_link(hw);
2370                         link_check = (E1000_READ_REG(hw, E1000_STATUS) &
2371                                       E1000_STATUS_LU);
2372                         break;
2373
2374                 case e1000_media_type_internal_serdes:
2375                         e1000_check_for_link(hw);
2376                         link_check = hw->mac.serdes_has_link;
2377                         break;
2378
2379                 /* VF device is type_unknown */
2380                 case e1000_media_type_unknown:
2381                         eth_igbvf_link_update(hw);
2382                         link_check = !hw->mac.get_link_status;
2383                         break;
2384
2385                 default:
2386                         break;
2387                 }
2388                 if (link_check || wait_to_complete == 0)
2389                         break;
2390                 rte_delay_ms(IGB_LINK_UPDATE_CHECK_INTERVAL);
2391         }
2392         memset(&link, 0, sizeof(link));
2393
2394         /* Now we check if a transition has happened */
2395         if (link_check) {
2396                 uint16_t duplex, speed;
2397                 hw->mac.ops.get_link_up_info(hw, &speed, &duplex);
2398                 link.link_duplex = (duplex == FULL_DUPLEX) ?
2399                                 RTE_ETH_LINK_FULL_DUPLEX :
2400                                 RTE_ETH_LINK_HALF_DUPLEX;
2401                 link.link_speed = speed;
2402                 link.link_status = RTE_ETH_LINK_UP;
2403                 link.link_autoneg = !(dev->data->dev_conf.link_speeds &
2404                                 RTE_ETH_LINK_SPEED_FIXED);
2405         } else if (!link_check) {
2406                 link.link_speed = 0;
2407                 link.link_duplex = RTE_ETH_LINK_HALF_DUPLEX;
2408                 link.link_status = RTE_ETH_LINK_DOWN;
2409                 link.link_autoneg = RTE_ETH_LINK_FIXED;
2410         }
2411
2412         return rte_eth_linkstatus_set(dev, &link);
2413 }
2414
2415 /*
2416  * igb_hw_control_acquire sets CTRL_EXT:DRV_LOAD bit.
2417  * For ASF and Pass Through versions of f/w this means
2418  * that the driver is loaded.
2419  */
2420 static void
2421 igb_hw_control_acquire(struct e1000_hw *hw)
2422 {
2423         uint32_t ctrl_ext;
2424
2425         /* Let firmware know the driver has taken over */
2426         ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
2427         E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2428 }
2429
2430 /*
2431  * igb_hw_control_release resets CTRL_EXT:DRV_LOAD bit.
2432  * For ASF and Pass Through versions of f/w this means that the
2433  * driver is no longer loaded.
2434  */
2435 static void
2436 igb_hw_control_release(struct e1000_hw *hw)
2437 {
2438         uint32_t ctrl_ext;
2439
2440         /* Let firmware taken over control of h/w */
2441         ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
2442         E1000_WRITE_REG(hw, E1000_CTRL_EXT,
2443                         ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2444 }
2445
2446 /*
2447  * Bit of a misnomer, what this really means is
2448  * to enable OS management of the system... aka
2449  * to disable special hardware management features.
2450  */
2451 static void
2452 igb_init_manageability(struct e1000_hw *hw)
2453 {
2454         if (e1000_enable_mng_pass_thru(hw)) {
2455                 uint32_t manc2h = E1000_READ_REG(hw, E1000_MANC2H);
2456                 uint32_t manc = E1000_READ_REG(hw, E1000_MANC);
2457
2458                 /* disable hardware interception of ARP */
2459                 manc &= ~(E1000_MANC_ARP_EN);
2460
2461                 /* enable receiving management packets to the host */
2462                 manc |= E1000_MANC_EN_MNG2HOST;
2463                 manc2h |= 1 << 5;  /* Mng Port 623 */
2464                 manc2h |= 1 << 6;  /* Mng Port 664 */
2465                 E1000_WRITE_REG(hw, E1000_MANC2H, manc2h);
2466                 E1000_WRITE_REG(hw, E1000_MANC, manc);
2467         }
2468 }
2469
2470 static void
2471 igb_release_manageability(struct e1000_hw *hw)
2472 {
2473         if (e1000_enable_mng_pass_thru(hw)) {
2474                 uint32_t manc = E1000_READ_REG(hw, E1000_MANC);
2475
2476                 manc |= E1000_MANC_ARP_EN;
2477                 manc &= ~E1000_MANC_EN_MNG2HOST;
2478
2479                 E1000_WRITE_REG(hw, E1000_MANC, manc);
2480         }
2481 }
2482
2483 static int
2484 eth_igb_promiscuous_enable(struct rte_eth_dev *dev)
2485 {
2486         struct e1000_hw *hw =
2487                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2488         uint32_t rctl;
2489
2490         rctl = E1000_READ_REG(hw, E1000_RCTL);
2491         rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2492         E1000_WRITE_REG(hw, E1000_RCTL, rctl);
2493
2494         return 0;
2495 }
2496
2497 static int
2498 eth_igb_promiscuous_disable(struct rte_eth_dev *dev)
2499 {
2500         struct e1000_hw *hw =
2501                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2502         uint32_t rctl;
2503
2504         rctl = E1000_READ_REG(hw, E1000_RCTL);
2505         rctl &= (~E1000_RCTL_UPE);
2506         if (dev->data->all_multicast == 1)
2507                 rctl |= E1000_RCTL_MPE;
2508         else
2509                 rctl &= (~E1000_RCTL_MPE);
2510         E1000_WRITE_REG(hw, E1000_RCTL, rctl);
2511
2512         return 0;
2513 }
2514
2515 static int
2516 eth_igb_allmulticast_enable(struct rte_eth_dev *dev)
2517 {
2518         struct e1000_hw *hw =
2519                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2520         uint32_t rctl;
2521
2522         rctl = E1000_READ_REG(hw, E1000_RCTL);
2523         rctl |= E1000_RCTL_MPE;
2524         E1000_WRITE_REG(hw, E1000_RCTL, rctl);
2525
2526         return 0;
2527 }
2528
2529 static int
2530 eth_igb_allmulticast_disable(struct rte_eth_dev *dev)
2531 {
2532         struct e1000_hw *hw =
2533                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2534         uint32_t rctl;
2535
2536         if (dev->data->promiscuous == 1)
2537                 return 0; /* must remain in all_multicast mode */
2538         rctl = E1000_READ_REG(hw, E1000_RCTL);
2539         rctl &= (~E1000_RCTL_MPE);
2540         E1000_WRITE_REG(hw, E1000_RCTL, rctl);
2541
2542         return 0;
2543 }
2544
2545 static int
2546 eth_igb_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
2547 {
2548         struct e1000_hw *hw =
2549                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2550         struct e1000_vfta * shadow_vfta =
2551                 E1000_DEV_PRIVATE_TO_VFTA(dev->data->dev_private);
2552         uint32_t vfta;
2553         uint32_t vid_idx;
2554         uint32_t vid_bit;
2555
2556         vid_idx = (uint32_t) ((vlan_id >> E1000_VFTA_ENTRY_SHIFT) &
2557                               E1000_VFTA_ENTRY_MASK);
2558         vid_bit = (uint32_t) (1 << (vlan_id & E1000_VFTA_ENTRY_BIT_SHIFT_MASK));
2559         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, vid_idx);
2560         if (on)
2561                 vfta |= vid_bit;
2562         else
2563                 vfta &= ~vid_bit;
2564         E1000_WRITE_REG_ARRAY(hw, E1000_VFTA, vid_idx, vfta);
2565
2566         /* update local VFTA copy */
2567         shadow_vfta->vfta[vid_idx] = vfta;
2568
2569         return 0;
2570 }
2571
2572 static int
2573 eth_igb_vlan_tpid_set(struct rte_eth_dev *dev,
2574                       enum rte_vlan_type vlan_type,
2575                       uint16_t tpid)
2576 {
2577         struct e1000_hw *hw =
2578                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2579         uint32_t reg, qinq;
2580
2581         qinq = E1000_READ_REG(hw, E1000_CTRL_EXT);
2582         qinq &= E1000_CTRL_EXT_EXT_VLAN;
2583
2584         /* only outer TPID of double VLAN can be configured*/
2585         if (qinq && vlan_type == RTE_ETH_VLAN_TYPE_OUTER) {
2586                 reg = E1000_READ_REG(hw, E1000_VET);
2587                 reg = (reg & (~E1000_VET_VET_EXT)) |
2588                         ((uint32_t)tpid << E1000_VET_VET_EXT_SHIFT);
2589                 E1000_WRITE_REG(hw, E1000_VET, reg);
2590
2591                 return 0;
2592         }
2593
2594         /* all other TPID values are read-only*/
2595         PMD_DRV_LOG(ERR, "Not supported");
2596
2597         return -ENOTSUP;
2598 }
2599
2600 static void
2601 igb_vlan_hw_filter_disable(struct rte_eth_dev *dev)
2602 {
2603         struct e1000_hw *hw =
2604                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2605         uint32_t reg;
2606
2607         /* Filter Table Disable */
2608         reg = E1000_READ_REG(hw, E1000_RCTL);
2609         reg &= ~E1000_RCTL_CFIEN;
2610         reg &= ~E1000_RCTL_VFE;
2611         E1000_WRITE_REG(hw, E1000_RCTL, reg);
2612 }
2613
2614 static void
2615 igb_vlan_hw_filter_enable(struct rte_eth_dev *dev)
2616 {
2617         struct e1000_hw *hw =
2618                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2619         struct e1000_vfta * shadow_vfta =
2620                 E1000_DEV_PRIVATE_TO_VFTA(dev->data->dev_private);
2621         uint32_t reg;
2622         int i;
2623
2624         /* Filter Table Enable, CFI not used for packet acceptance */
2625         reg = E1000_READ_REG(hw, E1000_RCTL);
2626         reg &= ~E1000_RCTL_CFIEN;
2627         reg |= E1000_RCTL_VFE;
2628         E1000_WRITE_REG(hw, E1000_RCTL, reg);
2629
2630         /* restore VFTA table */
2631         for (i = 0; i < IGB_VFTA_SIZE; i++)
2632                 E1000_WRITE_REG_ARRAY(hw, E1000_VFTA, i, shadow_vfta->vfta[i]);
2633 }
2634
2635 static void
2636 igb_vlan_hw_strip_disable(struct rte_eth_dev *dev)
2637 {
2638         struct e1000_hw *hw =
2639                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2640         uint32_t reg;
2641
2642         /* VLAN Mode Disable */
2643         reg = E1000_READ_REG(hw, E1000_CTRL);
2644         reg &= ~E1000_CTRL_VME;
2645         E1000_WRITE_REG(hw, E1000_CTRL, reg);
2646 }
2647
2648 static void
2649 igb_vlan_hw_strip_enable(struct rte_eth_dev *dev)
2650 {
2651         struct e1000_hw *hw =
2652                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2653         uint32_t reg;
2654
2655         /* VLAN Mode Enable */
2656         reg = E1000_READ_REG(hw, E1000_CTRL);
2657         reg |= E1000_CTRL_VME;
2658         E1000_WRITE_REG(hw, E1000_CTRL, reg);
2659 }
2660
2661 static void
2662 igb_vlan_hw_extend_disable(struct rte_eth_dev *dev)
2663 {
2664         struct e1000_hw *hw =
2665                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2666         uint32_t reg;
2667
2668         /* CTRL_EXT: Extended VLAN */
2669         reg = E1000_READ_REG(hw, E1000_CTRL_EXT);
2670         reg &= ~E1000_CTRL_EXT_EXTEND_VLAN;
2671         E1000_WRITE_REG(hw, E1000_CTRL_EXT, reg);
2672
2673         /* Update maximum packet length */
2674         E1000_WRITE_REG(hw, E1000_RLPML, dev->data->mtu + E1000_ETH_OVERHEAD);
2675 }
2676
2677 static void
2678 igb_vlan_hw_extend_enable(struct rte_eth_dev *dev)
2679 {
2680         struct e1000_hw *hw =
2681                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2682         uint32_t reg;
2683
2684         /* CTRL_EXT: Extended VLAN */
2685         reg = E1000_READ_REG(hw, E1000_CTRL_EXT);
2686         reg |= E1000_CTRL_EXT_EXTEND_VLAN;
2687         E1000_WRITE_REG(hw, E1000_CTRL_EXT, reg);
2688
2689         /* Update maximum packet length */
2690         E1000_WRITE_REG(hw, E1000_RLPML,
2691                 dev->data->mtu + E1000_ETH_OVERHEAD + VLAN_TAG_SIZE);
2692 }
2693
2694 static int
2695 eth_igb_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2696 {
2697         struct rte_eth_rxmode *rxmode;
2698
2699         rxmode = &dev->data->dev_conf.rxmode;
2700         if (mask & RTE_ETH_VLAN_STRIP_MASK) {
2701                 if (rxmode->offloads & RTE_ETH_RX_OFFLOAD_VLAN_STRIP)
2702                         igb_vlan_hw_strip_enable(dev);
2703                 else
2704                         igb_vlan_hw_strip_disable(dev);
2705         }
2706
2707         if (mask & RTE_ETH_VLAN_FILTER_MASK) {
2708                 if (rxmode->offloads & RTE_ETH_RX_OFFLOAD_VLAN_FILTER)
2709                         igb_vlan_hw_filter_enable(dev);
2710                 else
2711                         igb_vlan_hw_filter_disable(dev);
2712         }
2713
2714         if (mask & RTE_ETH_VLAN_EXTEND_MASK) {
2715                 if (rxmode->offloads & RTE_ETH_RX_OFFLOAD_VLAN_EXTEND)
2716                         igb_vlan_hw_extend_enable(dev);
2717                 else
2718                         igb_vlan_hw_extend_disable(dev);
2719         }
2720
2721         return 0;
2722 }
2723
2724
2725 /**
2726  * It enables the interrupt mask and then enable the interrupt.
2727  *
2728  * @param dev
2729  *  Pointer to struct rte_eth_dev.
2730  * @param on
2731  *  Enable or Disable
2732  *
2733  * @return
2734  *  - On success, zero.
2735  *  - On failure, a negative value.
2736  */
2737 static int
2738 eth_igb_lsc_interrupt_setup(struct rte_eth_dev *dev, uint8_t on)
2739 {
2740         struct e1000_interrupt *intr =
2741                 E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
2742
2743         if (on)
2744                 intr->mask |= E1000_ICR_LSC;
2745         else
2746                 intr->mask &= ~E1000_ICR_LSC;
2747
2748         return 0;
2749 }
2750
2751 /* It clears the interrupt causes and enables the interrupt.
2752  * It will be called once only during nic initialized.
2753  *
2754  * @param dev
2755  *  Pointer to struct rte_eth_dev.
2756  *
2757  * @return
2758  *  - On success, zero.
2759  *  - On failure, a negative value.
2760  */
2761 static int eth_igb_rxq_interrupt_setup(struct rte_eth_dev *dev)
2762 {
2763         uint32_t mask, regval;
2764         int ret;
2765         struct e1000_hw *hw =
2766                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2767         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2768         struct rte_intr_handle *intr_handle = pci_dev->intr_handle;
2769         int misc_shift = rte_intr_allow_others(intr_handle) ? 1 : 0;
2770         struct rte_eth_dev_info dev_info;
2771
2772         memset(&dev_info, 0, sizeof(dev_info));
2773         ret = eth_igb_infos_get(dev, &dev_info);
2774         if (ret != 0)
2775                 return ret;
2776
2777         mask = (0xFFFFFFFF >> (32 - dev_info.max_rx_queues)) << misc_shift;
2778         regval = E1000_READ_REG(hw, E1000_EIMS);
2779         E1000_WRITE_REG(hw, E1000_EIMS, regval | mask);
2780
2781         return 0;
2782 }
2783
2784 /*
2785  * It reads ICR and gets interrupt causes, check it and set a bit flag
2786  * to update link status.
2787  *
2788  * @param dev
2789  *  Pointer to struct rte_eth_dev.
2790  *
2791  * @return
2792  *  - On success, zero.
2793  *  - On failure, a negative value.
2794  */
2795 static int
2796 eth_igb_interrupt_get_status(struct rte_eth_dev *dev)
2797 {
2798         uint32_t icr;
2799         struct e1000_hw *hw =
2800                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2801         struct e1000_interrupt *intr =
2802                 E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
2803
2804         igb_intr_disable(dev);
2805
2806         /* read-on-clear nic registers here */
2807         icr = E1000_READ_REG(hw, E1000_ICR);
2808
2809         intr->flags = 0;
2810         if (icr & E1000_ICR_LSC) {
2811                 intr->flags |= E1000_FLAG_NEED_LINK_UPDATE;
2812         }
2813
2814         if (icr & E1000_ICR_VMMB)
2815                 intr->flags |= E1000_FLAG_MAILBOX;
2816
2817         return 0;
2818 }
2819
2820 /*
2821  * It executes link_update after knowing an interrupt is prsent.
2822  *
2823  * @param dev
2824  *  Pointer to struct rte_eth_dev.
2825  *
2826  * @return
2827  *  - On success, zero.
2828  *  - On failure, a negative value.
2829  */
2830 static int
2831 eth_igb_interrupt_action(struct rte_eth_dev *dev,
2832                          struct rte_intr_handle *intr_handle)
2833 {
2834         struct e1000_hw *hw =
2835                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2836         struct e1000_interrupt *intr =
2837                 E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
2838         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2839         struct rte_eth_link link;
2840         int ret;
2841
2842         if (intr->flags & E1000_FLAG_MAILBOX) {
2843                 igb_pf_mbx_process(dev);
2844                 intr->flags &= ~E1000_FLAG_MAILBOX;
2845         }
2846
2847         igb_intr_enable(dev);
2848         rte_intr_ack(intr_handle);
2849
2850         if (intr->flags & E1000_FLAG_NEED_LINK_UPDATE) {
2851                 intr->flags &= ~E1000_FLAG_NEED_LINK_UPDATE;
2852
2853                 /* set get_link_status to check register later */
2854                 hw->mac.get_link_status = 1;
2855                 ret = eth_igb_link_update(dev, 0);
2856
2857                 /* check if link has changed */
2858                 if (ret < 0)
2859                         return 0;
2860
2861                 rte_eth_linkstatus_get(dev, &link);
2862                 if (link.link_status) {
2863                         PMD_INIT_LOG(INFO,
2864                                      " Port %d: Link Up - speed %u Mbps - %s",
2865                                      dev->data->port_id,
2866                                      (unsigned)link.link_speed,
2867                                      link.link_duplex == RTE_ETH_LINK_FULL_DUPLEX ?
2868                                      "full-duplex" : "half-duplex");
2869                 } else {
2870                         PMD_INIT_LOG(INFO, " Port %d: Link Down",
2871                                      dev->data->port_id);
2872                 }
2873
2874                 PMD_INIT_LOG(DEBUG, "PCI Address: " PCI_PRI_FMT,
2875                              pci_dev->addr.domain,
2876                              pci_dev->addr.bus,
2877                              pci_dev->addr.devid,
2878                              pci_dev->addr.function);
2879                 rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC, NULL);
2880         }
2881
2882         return 0;
2883 }
2884
2885 /**
2886  * Interrupt handler which shall be registered at first.
2887  *
2888  * @param handle
2889  *  Pointer to interrupt handle.
2890  * @param param
2891  *  The address of parameter (struct rte_eth_dev *) regsitered before.
2892  *
2893  * @return
2894  *  void
2895  */
2896 static void
2897 eth_igb_interrupt_handler(void *param)
2898 {
2899         struct rte_eth_dev *dev = (struct rte_eth_dev *)param;
2900
2901         eth_igb_interrupt_get_status(dev);
2902         eth_igb_interrupt_action(dev, dev->intr_handle);
2903 }
2904
2905 static int
2906 eth_igbvf_interrupt_get_status(struct rte_eth_dev *dev)
2907 {
2908         uint32_t eicr;
2909         struct e1000_hw *hw =
2910                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2911         struct e1000_interrupt *intr =
2912                 E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
2913
2914         igbvf_intr_disable(hw);
2915
2916         /* read-on-clear nic registers here */
2917         eicr = E1000_READ_REG(hw, E1000_EICR);
2918         intr->flags = 0;
2919
2920         if (eicr == E1000_VTIVAR_MISC_MAILBOX)
2921                 intr->flags |= E1000_FLAG_MAILBOX;
2922
2923         return 0;
2924 }
2925
2926 void igbvf_mbx_process(struct rte_eth_dev *dev)
2927 {
2928         struct e1000_hw *hw =
2929                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2930         struct e1000_mbx_info *mbx = &hw->mbx;
2931         u32 in_msg = 0;
2932
2933         /* peek the message first */
2934         in_msg = E1000_READ_REG(hw, E1000_VMBMEM(0));
2935
2936         /* PF reset VF event */
2937         if (in_msg == E1000_PF_CONTROL_MSG) {
2938                 /* dummy mbx read to ack pf */
2939                 if (mbx->ops.read(hw, &in_msg, 1, 0))
2940                         return;
2941                 rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_RESET,
2942                                              NULL);
2943         }
2944 }
2945
2946 static int
2947 eth_igbvf_interrupt_action(struct rte_eth_dev *dev, struct rte_intr_handle *intr_handle)
2948 {
2949         struct e1000_interrupt *intr =
2950                 E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
2951
2952         if (intr->flags & E1000_FLAG_MAILBOX) {
2953                 igbvf_mbx_process(dev);
2954                 intr->flags &= ~E1000_FLAG_MAILBOX;
2955         }
2956
2957         igbvf_intr_enable(dev);
2958         rte_intr_ack(intr_handle);
2959
2960         return 0;
2961 }
2962
2963 static void
2964 eth_igbvf_interrupt_handler(void *param)
2965 {
2966         struct rte_eth_dev *dev = (struct rte_eth_dev *)param;
2967
2968         eth_igbvf_interrupt_get_status(dev);
2969         eth_igbvf_interrupt_action(dev, dev->intr_handle);
2970 }
2971
2972 static int
2973 eth_igb_led_on(struct rte_eth_dev *dev)
2974 {
2975         struct e1000_hw *hw;
2976
2977         hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2978         return e1000_led_on(hw) == E1000_SUCCESS ? 0 : -ENOTSUP;
2979 }
2980
2981 static int
2982 eth_igb_led_off(struct rte_eth_dev *dev)
2983 {
2984         struct e1000_hw *hw;
2985
2986         hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2987         return e1000_led_off(hw) == E1000_SUCCESS ? 0 : -ENOTSUP;
2988 }
2989
2990 static int
2991 eth_igb_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
2992 {
2993         struct e1000_hw *hw;
2994         uint32_t ctrl;
2995         int tx_pause;
2996         int rx_pause;
2997
2998         hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2999         fc_conf->pause_time = hw->fc.pause_time;
3000         fc_conf->high_water = hw->fc.high_water;
3001         fc_conf->low_water = hw->fc.low_water;
3002         fc_conf->send_xon = hw->fc.send_xon;
3003         fc_conf->autoneg = hw->mac.autoneg;
3004
3005         /*
3006          * Return rx_pause and tx_pause status according to actual setting of
3007          * the TFCE and RFCE bits in the CTRL register.
3008          */
3009         ctrl = E1000_READ_REG(hw, E1000_CTRL);
3010         if (ctrl & E1000_CTRL_TFCE)
3011                 tx_pause = 1;
3012         else
3013                 tx_pause = 0;
3014
3015         if (ctrl & E1000_CTRL_RFCE)
3016                 rx_pause = 1;
3017         else
3018                 rx_pause = 0;
3019
3020         if (rx_pause && tx_pause)
3021                 fc_conf->mode = RTE_ETH_FC_FULL;
3022         else if (rx_pause)
3023                 fc_conf->mode = RTE_ETH_FC_RX_PAUSE;
3024         else if (tx_pause)
3025                 fc_conf->mode = RTE_ETH_FC_TX_PAUSE;
3026         else
3027                 fc_conf->mode = RTE_ETH_FC_NONE;
3028
3029         return 0;
3030 }
3031
3032 static int
3033 eth_igb_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
3034 {
3035         struct e1000_hw *hw;
3036         int err;
3037         enum e1000_fc_mode rte_fcmode_2_e1000_fcmode[] = {
3038                 e1000_fc_none,
3039                 e1000_fc_rx_pause,
3040                 e1000_fc_tx_pause,
3041                 e1000_fc_full
3042         };
3043         uint32_t rx_buf_size;
3044         uint32_t max_high_water;
3045         uint32_t rctl;
3046         uint32_t ctrl;
3047
3048         hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3049         if (fc_conf->autoneg != hw->mac.autoneg)
3050                 return -ENOTSUP;
3051         rx_buf_size = igb_get_rx_buffer_size(hw);
3052         PMD_INIT_LOG(DEBUG, "Rx packet buffer size = 0x%x", rx_buf_size);
3053
3054         /* At least reserve one Ethernet frame for watermark */
3055         max_high_water = rx_buf_size - RTE_ETHER_MAX_LEN;
3056         if ((fc_conf->high_water > max_high_water) ||
3057             (fc_conf->high_water < fc_conf->low_water)) {
3058                 PMD_INIT_LOG(ERR, "e1000 incorrect high/low water value");
3059                 PMD_INIT_LOG(ERR, "high water must <=  0x%x", max_high_water);
3060                 return -EINVAL;
3061         }
3062
3063         hw->fc.requested_mode = rte_fcmode_2_e1000_fcmode[fc_conf->mode];
3064         hw->fc.pause_time     = fc_conf->pause_time;
3065         hw->fc.high_water     = fc_conf->high_water;
3066         hw->fc.low_water      = fc_conf->low_water;
3067         hw->fc.send_xon       = fc_conf->send_xon;
3068
3069         err = e1000_setup_link_generic(hw);
3070         if (err == E1000_SUCCESS) {
3071
3072                 /* check if we want to forward MAC frames - driver doesn't have native
3073                  * capability to do that, so we'll write the registers ourselves */
3074
3075                 rctl = E1000_READ_REG(hw, E1000_RCTL);
3076
3077                 /* set or clear MFLCN.PMCF bit depending on configuration */
3078                 if (fc_conf->mac_ctrl_frame_fwd != 0)
3079                         rctl |= E1000_RCTL_PMCF;
3080                 else
3081                         rctl &= ~E1000_RCTL_PMCF;
3082
3083                 E1000_WRITE_REG(hw, E1000_RCTL, rctl);
3084
3085                 /*
3086                  * check if we want to change flow control mode - driver doesn't have native
3087                  * capability to do that, so we'll write the registers ourselves
3088                  */
3089                 ctrl = E1000_READ_REG(hw, E1000_CTRL);
3090
3091                 /*
3092                  * set or clear E1000_CTRL_RFCE and E1000_CTRL_TFCE bits depending
3093                  * on configuration
3094                  */
3095                 switch (fc_conf->mode) {
3096                 case RTE_ETH_FC_NONE:
3097                         ctrl &= ~E1000_CTRL_RFCE & ~E1000_CTRL_TFCE;
3098                         break;
3099                 case RTE_ETH_FC_RX_PAUSE:
3100                         ctrl |= E1000_CTRL_RFCE;
3101                         ctrl &= ~E1000_CTRL_TFCE;
3102                         break;
3103                 case RTE_ETH_FC_TX_PAUSE:
3104                         ctrl |= E1000_CTRL_TFCE;
3105                         ctrl &= ~E1000_CTRL_RFCE;
3106                         break;
3107                 case RTE_ETH_FC_FULL:
3108                         ctrl |= E1000_CTRL_RFCE | E1000_CTRL_TFCE;
3109                         break;
3110                 default:
3111                         PMD_INIT_LOG(ERR, "invalid flow control mode");
3112                         return -EINVAL;
3113                 }
3114
3115                 E1000_WRITE_REG(hw, E1000_CTRL, ctrl);
3116
3117                 E1000_WRITE_FLUSH(hw);
3118
3119                 return 0;
3120         }
3121
3122         PMD_INIT_LOG(ERR, "e1000_setup_link_generic = 0x%x", err);
3123         return -EIO;
3124 }
3125
3126 #define E1000_RAH_POOLSEL_SHIFT      (18)
3127 static int
3128 eth_igb_rar_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
3129                 uint32_t index, uint32_t pool)
3130 {
3131         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3132         uint32_t rah;
3133
3134         e1000_rar_set(hw, mac_addr->addr_bytes, index);
3135         rah = E1000_READ_REG(hw, E1000_RAH(index));
3136         rah |= (0x1 << (E1000_RAH_POOLSEL_SHIFT + pool));
3137         E1000_WRITE_REG(hw, E1000_RAH(index), rah);
3138         return 0;
3139 }
3140
3141 static void
3142 eth_igb_rar_clear(struct rte_eth_dev *dev, uint32_t index)
3143 {
3144         uint8_t addr[RTE_ETHER_ADDR_LEN];
3145         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3146
3147         memset(addr, 0, sizeof(addr));
3148
3149         e1000_rar_set(hw, addr, index);
3150 }
3151
3152 static int
3153 eth_igb_default_mac_addr_set(struct rte_eth_dev *dev,
3154                                 struct rte_ether_addr *addr)
3155 {
3156         eth_igb_rar_clear(dev, 0);
3157         eth_igb_rar_set(dev, (void *)addr, 0, 0);
3158
3159         return 0;
3160 }
3161 /*
3162  * Virtual Function operations
3163  */
3164 static void
3165 igbvf_intr_disable(struct e1000_hw *hw)
3166 {
3167         PMD_INIT_FUNC_TRACE();
3168
3169         /* Clear interrupt mask to stop from interrupts being generated */
3170         E1000_WRITE_REG(hw, E1000_EIMC, 0xFFFF);
3171
3172         E1000_WRITE_FLUSH(hw);
3173 }
3174
3175 static void
3176 igbvf_stop_adapter(struct rte_eth_dev *dev)
3177 {
3178         u32 reg_val;
3179         u16 i;
3180         struct rte_eth_dev_info dev_info;
3181         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3182         int ret;
3183
3184         memset(&dev_info, 0, sizeof(dev_info));
3185         ret = eth_igbvf_infos_get(dev, &dev_info);
3186         if (ret != 0)
3187                 return;
3188
3189         /* Clear interrupt mask to stop from interrupts being generated */
3190         igbvf_intr_disable(hw);
3191
3192         /* Clear any pending interrupts, flush previous writes */
3193         E1000_READ_REG(hw, E1000_EICR);
3194
3195         /* Disable the transmit unit.  Each queue must be disabled. */
3196         for (i = 0; i < dev_info.max_tx_queues; i++)
3197                 E1000_WRITE_REG(hw, E1000_TXDCTL(i), E1000_TXDCTL_SWFLSH);
3198
3199         /* Disable the receive unit by stopping each queue */
3200         for (i = 0; i < dev_info.max_rx_queues; i++) {
3201                 reg_val = E1000_READ_REG(hw, E1000_RXDCTL(i));
3202                 reg_val &= ~E1000_RXDCTL_QUEUE_ENABLE;
3203                 E1000_WRITE_REG(hw, E1000_RXDCTL(i), reg_val);
3204                 while (E1000_READ_REG(hw, E1000_RXDCTL(i)) & E1000_RXDCTL_QUEUE_ENABLE)
3205                         ;
3206         }
3207
3208         /* flush all queues disables */
3209         E1000_WRITE_FLUSH(hw);
3210         msec_delay(2);
3211 }
3212
3213 static int eth_igbvf_link_update(struct e1000_hw *hw)
3214 {
3215         struct e1000_mbx_info *mbx = &hw->mbx;
3216         struct e1000_mac_info *mac = &hw->mac;
3217         int ret_val = E1000_SUCCESS;
3218
3219         PMD_INIT_LOG(DEBUG, "e1000_check_for_link_vf");
3220
3221         /*
3222          * We only want to run this if there has been a rst asserted.
3223          * in this case that could mean a link change, device reset,
3224          * or a virtual function reset
3225          */
3226
3227         /* If we were hit with a reset or timeout drop the link */
3228         if (!e1000_check_for_rst(hw, 0) || !mbx->timeout)
3229                 mac->get_link_status = TRUE;
3230
3231         if (!mac->get_link_status)
3232                 goto out;
3233
3234         /* if link status is down no point in checking to see if pf is up */
3235         if (!(E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU))
3236                 goto out;
3237
3238         /* if we passed all the tests above then the link is up and we no
3239          * longer need to check for link */
3240         mac->get_link_status = FALSE;
3241
3242 out:
3243         return ret_val;
3244 }
3245
3246
3247 static int
3248 igbvf_dev_configure(struct rte_eth_dev *dev)
3249 {
3250         struct rte_eth_conf* conf = &dev->data->dev_conf;
3251
3252         PMD_INIT_LOG(DEBUG, "Configured Virtual Function port id: %d",
3253                      dev->data->port_id);
3254
3255         if (dev->data->dev_conf.rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG)
3256                 dev->data->dev_conf.rxmode.offloads |= RTE_ETH_RX_OFFLOAD_RSS_HASH;
3257
3258         /*
3259          * VF has no ability to enable/disable HW CRC
3260          * Keep the persistent behavior the same as Host PF
3261          */
3262 #ifndef RTE_LIBRTE_E1000_PF_DISABLE_STRIP_CRC
3263         if (conf->rxmode.offloads & RTE_ETH_RX_OFFLOAD_KEEP_CRC) {
3264                 PMD_INIT_LOG(NOTICE, "VF can't disable HW CRC Strip");
3265                 conf->rxmode.offloads &= ~RTE_ETH_RX_OFFLOAD_KEEP_CRC;
3266         }
3267 #else
3268         if (!(conf->rxmode.offloads & RTE_ETH_RX_OFFLOAD_KEEP_CRC)) {
3269                 PMD_INIT_LOG(NOTICE, "VF can't enable HW CRC Strip");
3270                 conf->rxmode.offloads |= RTE_ETH_RX_OFFLOAD_KEEP_CRC;
3271         }
3272 #endif
3273
3274         return 0;
3275 }
3276
3277 static int
3278 igbvf_dev_start(struct rte_eth_dev *dev)
3279 {
3280         struct e1000_hw *hw =
3281                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3282         struct e1000_adapter *adapter =
3283                 E1000_DEV_PRIVATE(dev->data->dev_private);
3284         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
3285         struct rte_intr_handle *intr_handle = pci_dev->intr_handle;
3286         int ret;
3287         uint32_t intr_vector = 0;
3288
3289         PMD_INIT_FUNC_TRACE();
3290
3291         hw->mac.ops.reset_hw(hw);
3292         adapter->stopped = 0;
3293
3294         /* Set all vfta */
3295         igbvf_set_vfta_all(dev,1);
3296
3297         eth_igbvf_tx_init(dev);
3298
3299         /* This can fail when allocating mbufs for descriptor rings */
3300         ret = eth_igbvf_rx_init(dev);
3301         if (ret) {
3302                 PMD_INIT_LOG(ERR, "Unable to initialize RX hardware");
3303                 igb_dev_clear_queues(dev);
3304                 return ret;
3305         }
3306
3307         /* check and configure queue intr-vector mapping */
3308         if (rte_intr_cap_multiple(intr_handle) &&
3309             dev->data->dev_conf.intr_conf.rxq) {
3310                 intr_vector = dev->data->nb_rx_queues;
3311                 ret = rte_intr_efd_enable(intr_handle, intr_vector);
3312                 if (ret)
3313                         return ret;
3314         }
3315
3316         /* Allocate the vector list */
3317         if (rte_intr_dp_is_en(intr_handle)) {
3318                 if (rte_intr_vec_list_alloc(intr_handle, "intr_vec",
3319                                                    dev->data->nb_rx_queues)) {
3320                         PMD_INIT_LOG(ERR, "Failed to allocate %d rx_queues"
3321                                      " intr_vec", dev->data->nb_rx_queues);
3322                         return -ENOMEM;
3323                 }
3324         }
3325
3326         eth_igbvf_configure_msix_intr(dev);
3327
3328         /* enable uio/vfio intr/eventfd mapping */
3329         rte_intr_enable(intr_handle);
3330
3331         /* resume enabled intr since hw reset */
3332         igbvf_intr_enable(dev);
3333
3334         return 0;
3335 }
3336
3337 static int
3338 igbvf_dev_stop(struct rte_eth_dev *dev)
3339 {
3340         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
3341         struct rte_intr_handle *intr_handle = pci_dev->intr_handle;
3342         struct e1000_adapter *adapter =
3343                 E1000_DEV_PRIVATE(dev->data->dev_private);
3344
3345         if (adapter->stopped)
3346                 return 0;
3347
3348         PMD_INIT_FUNC_TRACE();
3349
3350         igbvf_stop_adapter(dev);
3351
3352         /*
3353           * Clear what we set, but we still keep shadow_vfta to
3354           * restore after device starts
3355           */
3356         igbvf_set_vfta_all(dev,0);
3357
3358         igb_dev_clear_queues(dev);
3359
3360         /* disable intr eventfd mapping */
3361         rte_intr_disable(intr_handle);
3362
3363         /* Clean datapath event and queue/vec mapping */
3364         rte_intr_efd_disable(intr_handle);
3365
3366         /* Clean vector list */
3367         rte_intr_vec_list_free(intr_handle);
3368
3369         adapter->stopped = true;
3370         dev->data->dev_started = 0;
3371
3372         return 0;
3373 }
3374
3375 static int
3376 igbvf_dev_close(struct rte_eth_dev *dev)
3377 {
3378         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3379         struct rte_ether_addr addr;
3380         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
3381         int ret;
3382
3383         PMD_INIT_FUNC_TRACE();
3384
3385         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
3386                 return 0;
3387
3388         e1000_reset_hw(hw);
3389
3390         ret = igbvf_dev_stop(dev);
3391         if (ret != 0)
3392                 return ret;
3393
3394         igb_dev_free_queues(dev);
3395
3396         /**
3397          * reprogram the RAR with a zero mac address,
3398          * to ensure that the VF traffic goes to the PF
3399          * after stop, close and detach of the VF.
3400          **/
3401
3402         memset(&addr, 0, sizeof(addr));
3403         igbvf_default_mac_addr_set(dev, &addr);
3404
3405         rte_intr_callback_unregister(pci_dev->intr_handle,
3406                                      eth_igbvf_interrupt_handler,
3407                                      (void *)dev);
3408
3409         return 0;
3410 }
3411
3412 static int
3413 igbvf_promiscuous_enable(struct rte_eth_dev *dev)
3414 {
3415         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3416
3417         /* Set both unicast and multicast promisc */
3418         e1000_promisc_set_vf(hw, e1000_promisc_enabled);
3419
3420         return 0;
3421 }
3422
3423 static int
3424 igbvf_promiscuous_disable(struct rte_eth_dev *dev)
3425 {
3426         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3427
3428         /* If in allmulticast mode leave multicast promisc */
3429         if (dev->data->all_multicast == 1)
3430                 e1000_promisc_set_vf(hw, e1000_promisc_multicast);
3431         else
3432                 e1000_promisc_set_vf(hw, e1000_promisc_disabled);
3433
3434         return 0;
3435 }
3436
3437 static int
3438 igbvf_allmulticast_enable(struct rte_eth_dev *dev)
3439 {
3440         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3441
3442         /* In promiscuous mode multicast promisc already set */
3443         if (dev->data->promiscuous == 0)
3444                 e1000_promisc_set_vf(hw, e1000_promisc_multicast);
3445
3446         return 0;
3447 }
3448
3449 static int
3450 igbvf_allmulticast_disable(struct rte_eth_dev *dev)
3451 {
3452         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3453
3454         /* In promiscuous mode leave multicast promisc enabled */
3455         if (dev->data->promiscuous == 0)
3456                 e1000_promisc_set_vf(hw, e1000_promisc_disabled);
3457
3458         return 0;
3459 }
3460
3461 static int igbvf_set_vfta(struct e1000_hw *hw, uint16_t vid, bool on)
3462 {
3463         struct e1000_mbx_info *mbx = &hw->mbx;
3464         uint32_t msgbuf[2];
3465         s32 err;
3466
3467         /* After set vlan, vlan strip will also be enabled in igb driver*/
3468         msgbuf[0] = E1000_VF_SET_VLAN;
3469         msgbuf[1] = vid;
3470         /* Setting the 8 bit field MSG INFO to TRUE indicates "add" */
3471         if (on)
3472                 msgbuf[0] |= E1000_VF_SET_VLAN_ADD;
3473
3474         err = mbx->ops.write_posted(hw, msgbuf, 2, 0);
3475         if (err)
3476                 goto mbx_err;
3477
3478         err = mbx->ops.read_posted(hw, msgbuf, 2, 0);
3479         if (err)
3480                 goto mbx_err;
3481
3482         msgbuf[0] &= ~E1000_VT_MSGTYPE_CTS;
3483         if (msgbuf[0] == (E1000_VF_SET_VLAN | E1000_VT_MSGTYPE_NACK))
3484                 err = -EINVAL;
3485
3486 mbx_err:
3487         return err;
3488 }
3489
3490 static void igbvf_set_vfta_all(struct rte_eth_dev *dev, bool on)
3491 {
3492         struct e1000_hw *hw =
3493                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3494         struct e1000_vfta * shadow_vfta =
3495                 E1000_DEV_PRIVATE_TO_VFTA(dev->data->dev_private);
3496         int i = 0, j = 0, vfta = 0, mask = 1;
3497
3498         for (i = 0; i < IGB_VFTA_SIZE; i++){
3499                 vfta = shadow_vfta->vfta[i];
3500                 if(vfta){
3501                         mask = 1;
3502                         for (j = 0; j < 32; j++){
3503                                 if(vfta & mask)
3504                                         igbvf_set_vfta(hw,
3505                                                 (uint16_t)((i<<5)+j), on);
3506                                 mask<<=1;
3507                         }
3508                 }
3509         }
3510
3511 }
3512
3513 static int
3514 igbvf_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
3515 {
3516         struct e1000_hw *hw =
3517                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3518         struct e1000_vfta * shadow_vfta =
3519                 E1000_DEV_PRIVATE_TO_VFTA(dev->data->dev_private);
3520         uint32_t vid_idx = 0;
3521         uint32_t vid_bit = 0;
3522         int ret = 0;
3523
3524         PMD_INIT_FUNC_TRACE();
3525
3526         /*vind is not used in VF driver, set to 0, check ixgbe_set_vfta_vf*/
3527         ret = igbvf_set_vfta(hw, vlan_id, !!on);
3528         if(ret){
3529                 PMD_INIT_LOG(ERR, "Unable to set VF vlan");
3530                 return ret;
3531         }
3532         vid_idx = (uint32_t) ((vlan_id >> 5) & 0x7F);
3533         vid_bit = (uint32_t) (1 << (vlan_id & 0x1F));
3534
3535         /*Save what we set and retore it after device reset*/
3536         if (on)
3537                 shadow_vfta->vfta[vid_idx] |= vid_bit;
3538         else
3539                 shadow_vfta->vfta[vid_idx] &= ~vid_bit;
3540
3541         return 0;
3542 }
3543
3544 static int
3545 igbvf_default_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *addr)
3546 {
3547         struct e1000_hw *hw =
3548                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3549
3550         /* index is not used by rar_set() */
3551         hw->mac.ops.rar_set(hw, (void *)addr, 0);
3552         return 0;
3553 }
3554
3555
3556 static int
3557 eth_igb_rss_reta_update(struct rte_eth_dev *dev,
3558                         struct rte_eth_rss_reta_entry64 *reta_conf,
3559                         uint16_t reta_size)
3560 {
3561         uint8_t i, j, mask;
3562         uint32_t reta, r;
3563         uint16_t idx, shift;
3564         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3565
3566         if (reta_size != RTE_ETH_RSS_RETA_SIZE_128) {
3567                 PMD_DRV_LOG(ERR, "The size of hash lookup table configured "
3568                         "(%d) doesn't match the number hardware can supported "
3569                         "(%d)", reta_size, RTE_ETH_RSS_RETA_SIZE_128);
3570                 return -EINVAL;
3571         }
3572
3573         for (i = 0; i < reta_size; i += IGB_4_BIT_WIDTH) {
3574                 idx = i / RTE_ETH_RETA_GROUP_SIZE;
3575                 shift = i % RTE_ETH_RETA_GROUP_SIZE;
3576                 mask = (uint8_t)((reta_conf[idx].mask >> shift) &
3577                                                 IGB_4_BIT_MASK);
3578                 if (!mask)
3579                         continue;
3580                 if (mask == IGB_4_BIT_MASK)
3581                         r = 0;
3582                 else
3583                         r = E1000_READ_REG(hw, E1000_RETA(i >> 2));
3584                 for (j = 0, reta = 0; j < IGB_4_BIT_WIDTH; j++) {
3585                         if (mask & (0x1 << j))
3586                                 reta |= reta_conf[idx].reta[shift + j] <<
3587                                                         (CHAR_BIT * j);
3588                         else
3589                                 reta |= r & (IGB_8_BIT_MASK << (CHAR_BIT * j));
3590                 }
3591                 E1000_WRITE_REG(hw, E1000_RETA(i >> 2), reta);
3592         }
3593
3594         return 0;
3595 }
3596
3597 static int
3598 eth_igb_rss_reta_query(struct rte_eth_dev *dev,
3599                        struct rte_eth_rss_reta_entry64 *reta_conf,
3600                        uint16_t reta_size)
3601 {
3602         uint8_t i, j, mask;
3603         uint32_t reta;
3604         uint16_t idx, shift;
3605         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3606
3607         if (reta_size != RTE_ETH_RSS_RETA_SIZE_128) {
3608                 PMD_DRV_LOG(ERR, "The size of hash lookup table configured "
3609                         "(%d) doesn't match the number hardware can supported "
3610                         "(%d)", reta_size, RTE_ETH_RSS_RETA_SIZE_128);
3611                 return -EINVAL;
3612         }
3613
3614         for (i = 0; i < reta_size; i += IGB_4_BIT_WIDTH) {
3615                 idx = i / RTE_ETH_RETA_GROUP_SIZE;
3616                 shift = i % RTE_ETH_RETA_GROUP_SIZE;
3617                 mask = (uint8_t)((reta_conf[idx].mask >> shift) &
3618                                                 IGB_4_BIT_MASK);
3619                 if (!mask)
3620                         continue;
3621                 reta = E1000_READ_REG(hw, E1000_RETA(i >> 2));
3622                 for (j = 0; j < IGB_4_BIT_WIDTH; j++) {
3623                         if (mask & (0x1 << j))
3624                                 reta_conf[idx].reta[shift + j] =
3625                                         ((reta >> (CHAR_BIT * j)) &
3626                                                 IGB_8_BIT_MASK);
3627                 }
3628         }
3629
3630         return 0;
3631 }
3632
3633 int
3634 eth_igb_syn_filter_set(struct rte_eth_dev *dev,
3635                         struct rte_eth_syn_filter *filter,
3636                         bool add)
3637 {
3638         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3639         struct e1000_filter_info *filter_info =
3640                 E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
3641         uint32_t synqf, rfctl;
3642
3643         if (filter->queue >= IGB_MAX_RX_QUEUE_NUM)
3644                 return -EINVAL;
3645
3646         synqf = E1000_READ_REG(hw, E1000_SYNQF(0));
3647
3648         if (add) {
3649                 if (synqf & E1000_SYN_FILTER_ENABLE)
3650                         return -EINVAL;
3651
3652                 synqf = (uint32_t)(((filter->queue << E1000_SYN_FILTER_QUEUE_SHIFT) &
3653                         E1000_SYN_FILTER_QUEUE) | E1000_SYN_FILTER_ENABLE);
3654
3655                 rfctl = E1000_READ_REG(hw, E1000_RFCTL);
3656                 if (filter->hig_pri)
3657                         rfctl |= E1000_RFCTL_SYNQFP;
3658                 else
3659                         rfctl &= ~E1000_RFCTL_SYNQFP;
3660
3661                 E1000_WRITE_REG(hw, E1000_RFCTL, rfctl);
3662         } else {
3663                 if (!(synqf & E1000_SYN_FILTER_ENABLE))
3664                         return -ENOENT;
3665                 synqf = 0;
3666         }
3667
3668         filter_info->syn_info = synqf;
3669         E1000_WRITE_REG(hw, E1000_SYNQF(0), synqf);
3670         E1000_WRITE_FLUSH(hw);
3671         return 0;
3672 }
3673
3674 /* translate elements in struct rte_eth_ntuple_filter to struct e1000_2tuple_filter_info*/
3675 static inline int
3676 ntuple_filter_to_2tuple(struct rte_eth_ntuple_filter *filter,
3677                         struct e1000_2tuple_filter_info *filter_info)
3678 {
3679         if (filter->queue >= IGB_MAX_RX_QUEUE_NUM)
3680                 return -EINVAL;
3681         if (filter->priority > E1000_2TUPLE_MAX_PRI)
3682                 return -EINVAL;  /* filter index is out of range. */
3683         if (filter->tcp_flags > RTE_NTUPLE_TCP_FLAGS_MASK)
3684                 return -EINVAL;  /* flags is invalid. */
3685
3686         switch (filter->dst_port_mask) {
3687         case UINT16_MAX:
3688                 filter_info->dst_port_mask = 0;
3689                 filter_info->dst_port = filter->dst_port;
3690                 break;
3691         case 0:
3692                 filter_info->dst_port_mask = 1;
3693                 break;
3694         default:
3695                 PMD_DRV_LOG(ERR, "invalid dst_port mask.");
3696                 return -EINVAL;
3697         }
3698
3699         switch (filter->proto_mask) {
3700         case UINT8_MAX:
3701                 filter_info->proto_mask = 0;
3702                 filter_info->proto = filter->proto;
3703                 break;
3704         case 0:
3705                 filter_info->proto_mask = 1;
3706                 break;
3707         default:
3708                 PMD_DRV_LOG(ERR, "invalid protocol mask.");
3709                 return -EINVAL;
3710         }
3711
3712         filter_info->priority = (uint8_t)filter->priority;
3713         if (filter->flags & RTE_NTUPLE_FLAGS_TCP_FLAG)
3714                 filter_info->tcp_flags = filter->tcp_flags;
3715         else
3716                 filter_info->tcp_flags = 0;
3717
3718         return 0;
3719 }
3720
3721 static inline struct e1000_2tuple_filter *
3722 igb_2tuple_filter_lookup(struct e1000_2tuple_filter_list *filter_list,
3723                         struct e1000_2tuple_filter_info *key)
3724 {
3725         struct e1000_2tuple_filter *it;
3726
3727         TAILQ_FOREACH(it, filter_list, entries) {
3728                 if (memcmp(key, &it->filter_info,
3729                         sizeof(struct e1000_2tuple_filter_info)) == 0) {
3730                         return it;
3731                 }
3732         }
3733         return NULL;
3734 }
3735
3736 /* inject a igb 2tuple filter to HW */
3737 static inline void
3738 igb_inject_2uple_filter(struct rte_eth_dev *dev,
3739                            struct e1000_2tuple_filter *filter)
3740 {
3741         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3742         uint32_t ttqf = E1000_TTQF_DISABLE_MASK;
3743         uint32_t imir, imir_ext = E1000_IMIREXT_SIZE_BP;
3744         int i;
3745
3746         i = filter->index;
3747         imir = (uint32_t)(filter->filter_info.dst_port & E1000_IMIR_DSTPORT);
3748         if (filter->filter_info.dst_port_mask == 1) /* 1b means not compare. */
3749                 imir |= E1000_IMIR_PORT_BP;
3750         else
3751                 imir &= ~E1000_IMIR_PORT_BP;
3752
3753         imir |= filter->filter_info.priority << E1000_IMIR_PRIORITY_SHIFT;
3754
3755         ttqf |= E1000_TTQF_QUEUE_ENABLE;
3756         ttqf |= (uint32_t)(filter->queue << E1000_TTQF_QUEUE_SHIFT);
3757         ttqf |= (uint32_t)(filter->filter_info.proto &
3758                                                 E1000_TTQF_PROTOCOL_MASK);
3759         if (filter->filter_info.proto_mask == 0)
3760                 ttqf &= ~E1000_TTQF_MASK_ENABLE;
3761
3762         /* tcp flags bits setting. */
3763         if (filter->filter_info.tcp_flags & RTE_NTUPLE_TCP_FLAGS_MASK) {
3764                 if (filter->filter_info.tcp_flags & RTE_TCP_URG_FLAG)
3765                         imir_ext |= E1000_IMIREXT_CTRL_URG;
3766                 if (filter->filter_info.tcp_flags & RTE_TCP_ACK_FLAG)
3767                         imir_ext |= E1000_IMIREXT_CTRL_ACK;
3768                 if (filter->filter_info.tcp_flags & RTE_TCP_PSH_FLAG)
3769                         imir_ext |= E1000_IMIREXT_CTRL_PSH;
3770                 if (filter->filter_info.tcp_flags & RTE_TCP_RST_FLAG)
3771                         imir_ext |= E1000_IMIREXT_CTRL_RST;
3772                 if (filter->filter_info.tcp_flags & RTE_TCP_SYN_FLAG)
3773                         imir_ext |= E1000_IMIREXT_CTRL_SYN;
3774                 if (filter->filter_info.tcp_flags & RTE_TCP_FIN_FLAG)
3775                         imir_ext |= E1000_IMIREXT_CTRL_FIN;
3776         } else {
3777                 imir_ext |= E1000_IMIREXT_CTRL_BP;
3778         }
3779         E1000_WRITE_REG(hw, E1000_IMIR(i), imir);
3780         E1000_WRITE_REG(hw, E1000_TTQF(i), ttqf);
3781         E1000_WRITE_REG(hw, E1000_IMIREXT(i), imir_ext);
3782 }
3783
3784 /*
3785  * igb_add_2tuple_filter - add a 2tuple filter
3786  *
3787  * @param
3788  * dev: Pointer to struct rte_eth_dev.
3789  * ntuple_filter: ponter to the filter that will be added.
3790  *
3791  * @return
3792  *    - On success, zero.
3793  *    - On failure, a negative value.
3794  */
3795 static int
3796 igb_add_2tuple_filter(struct rte_eth_dev *dev,
3797                         struct rte_eth_ntuple_filter *ntuple_filter)
3798 {
3799         struct e1000_filter_info *filter_info =
3800                 E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
3801         struct e1000_2tuple_filter *filter;
3802         int i, ret;
3803
3804         filter = rte_zmalloc("e1000_2tuple_filter",
3805                         sizeof(struct e1000_2tuple_filter), 0);
3806         if (filter == NULL)
3807                 return -ENOMEM;
3808
3809         ret = ntuple_filter_to_2tuple(ntuple_filter,
3810                                       &filter->filter_info);
3811         if (ret < 0) {
3812                 rte_free(filter);
3813                 return ret;
3814         }
3815         if (igb_2tuple_filter_lookup(&filter_info->twotuple_list,
3816                                          &filter->filter_info) != NULL) {
3817                 PMD_DRV_LOG(ERR, "filter exists.");
3818                 rte_free(filter);
3819                 return -EEXIST;
3820         }
3821         filter->queue = ntuple_filter->queue;
3822
3823         /*
3824          * look for an unused 2tuple filter index,
3825          * and insert the filter to list.
3826          */
3827         for (i = 0; i < E1000_MAX_TTQF_FILTERS; i++) {
3828                 if (!(filter_info->twotuple_mask & (1 << i))) {
3829                         filter_info->twotuple_mask |= 1 << i;
3830                         filter->index = i;
3831                         TAILQ_INSERT_TAIL(&filter_info->twotuple_list,
3832                                           filter,
3833                                           entries);
3834                         break;
3835                 }
3836         }
3837         if (i >= E1000_MAX_TTQF_FILTERS) {
3838                 PMD_DRV_LOG(ERR, "2tuple filters are full.");
3839                 rte_free(filter);
3840                 return -ENOSYS;
3841         }
3842
3843         igb_inject_2uple_filter(dev, filter);
3844         return 0;
3845 }
3846
3847 int
3848 igb_delete_2tuple_filter(struct rte_eth_dev *dev,
3849                         struct e1000_2tuple_filter *filter)
3850 {
3851         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3852         struct e1000_filter_info *filter_info =
3853                 E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
3854
3855         filter_info->twotuple_mask &= ~(1 << filter->index);
3856         TAILQ_REMOVE(&filter_info->twotuple_list, filter, entries);
3857         rte_free(filter);
3858
3859         E1000_WRITE_REG(hw, E1000_TTQF(filter->index), E1000_TTQF_DISABLE_MASK);
3860         E1000_WRITE_REG(hw, E1000_IMIR(filter->index), 0);
3861         E1000_WRITE_REG(hw, E1000_IMIREXT(filter->index), 0);
3862         return 0;
3863 }
3864
3865 /*
3866  * igb_remove_2tuple_filter - remove a 2tuple filter
3867  *
3868  * @param
3869  * dev: Pointer to struct rte_eth_dev.
3870  * ntuple_filter: ponter to the filter that will be removed.
3871  *
3872  * @return
3873  *    - On success, zero.
3874  *    - On failure, a negative value.
3875  */
3876 static int
3877 igb_remove_2tuple_filter(struct rte_eth_dev *dev,
3878                         struct rte_eth_ntuple_filter *ntuple_filter)
3879 {
3880         struct e1000_filter_info *filter_info =
3881                 E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
3882         struct e1000_2tuple_filter_info filter_2tuple;
3883         struct e1000_2tuple_filter *filter;
3884         int ret;
3885
3886         memset(&filter_2tuple, 0, sizeof(struct e1000_2tuple_filter_info));
3887         ret = ntuple_filter_to_2tuple(ntuple_filter,
3888                                       &filter_2tuple);
3889         if (ret < 0)
3890                 return ret;
3891
3892         filter = igb_2tuple_filter_lookup(&filter_info->twotuple_list,
3893                                          &filter_2tuple);
3894         if (filter == NULL) {
3895                 PMD_DRV_LOG(ERR, "filter doesn't exist.");
3896                 return -ENOENT;
3897         }
3898
3899         igb_delete_2tuple_filter(dev, filter);
3900
3901         return 0;
3902 }
3903
3904 /* inject a igb flex filter to HW */
3905 static inline void
3906 igb_inject_flex_filter(struct rte_eth_dev *dev,
3907                            struct e1000_flex_filter *filter)
3908 {
3909         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3910         uint32_t wufc, queueing;
3911         uint32_t reg_off;
3912         uint8_t i, j = 0;
3913
3914         wufc = E1000_READ_REG(hw, E1000_WUFC);
3915         if (filter->index < E1000_MAX_FHFT)
3916                 reg_off = E1000_FHFT(filter->index);
3917         else
3918                 reg_off = E1000_FHFT_EXT(filter->index - E1000_MAX_FHFT);
3919
3920         E1000_WRITE_REG(hw, E1000_WUFC, wufc | E1000_WUFC_FLEX_HQ |
3921                         (E1000_WUFC_FLX0 << filter->index));
3922         queueing = filter->filter_info.len |
3923                 (filter->queue << E1000_FHFT_QUEUEING_QUEUE_SHIFT) |
3924                 (filter->filter_info.priority <<
3925                         E1000_FHFT_QUEUEING_PRIO_SHIFT);
3926         E1000_WRITE_REG(hw, reg_off + E1000_FHFT_QUEUEING_OFFSET,
3927                         queueing);
3928
3929         for (i = 0; i < E1000_FLEX_FILTERS_MASK_SIZE; i++) {
3930                 E1000_WRITE_REG(hw, reg_off,
3931                                 filter->filter_info.dwords[j]);
3932                 reg_off += sizeof(uint32_t);
3933                 E1000_WRITE_REG(hw, reg_off,
3934                                 filter->filter_info.dwords[++j]);
3935                 reg_off += sizeof(uint32_t);
3936                 E1000_WRITE_REG(hw, reg_off,
3937                         (uint32_t)filter->filter_info.mask[i]);
3938                 reg_off += sizeof(uint32_t) * 2;
3939                 ++j;
3940         }
3941 }
3942
3943 static inline struct e1000_flex_filter *
3944 eth_igb_flex_filter_lookup(struct e1000_flex_filter_list *filter_list,
3945                         struct e1000_flex_filter_info *key)
3946 {
3947         struct e1000_flex_filter *it;
3948
3949         TAILQ_FOREACH(it, filter_list, entries) {
3950                 if (memcmp(key, &it->filter_info,
3951                         sizeof(struct e1000_flex_filter_info)) == 0)
3952                         return it;
3953         }
3954
3955         return NULL;
3956 }
3957
3958 /* remove a flex byte filter
3959  * @param
3960  * dev: Pointer to struct rte_eth_dev.
3961  * filter: the pointer of the filter will be removed.
3962  */
3963 void
3964 igb_remove_flex_filter(struct rte_eth_dev *dev,
3965                         struct e1000_flex_filter *filter)
3966 {
3967         struct e1000_filter_info *filter_info =
3968                 E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
3969         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3970         uint32_t wufc, i;
3971         uint32_t reg_off;
3972
3973         wufc = E1000_READ_REG(hw, E1000_WUFC);
3974         if (filter->index < E1000_MAX_FHFT)
3975                 reg_off = E1000_FHFT(filter->index);
3976         else
3977                 reg_off = E1000_FHFT_EXT(filter->index - E1000_MAX_FHFT);
3978
3979         for (i = 0; i < E1000_FHFT_SIZE_IN_DWD; i++)
3980                 E1000_WRITE_REG(hw, reg_off + i * sizeof(uint32_t), 0);
3981
3982         E1000_WRITE_REG(hw, E1000_WUFC, wufc &
3983                 (~(E1000_WUFC_FLX0 << filter->index)));
3984
3985         filter_info->flex_mask &= ~(1 << filter->index);
3986         TAILQ_REMOVE(&filter_info->flex_list, filter, entries);
3987         rte_free(filter);
3988 }
3989
3990 int
3991 eth_igb_add_del_flex_filter(struct rte_eth_dev *dev,
3992                         struct igb_flex_filter *filter,
3993                         bool add)
3994 {
3995         struct e1000_filter_info *filter_info =
3996                 E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
3997         struct e1000_flex_filter *flex_filter, *it;
3998         uint32_t mask;
3999         uint8_t shift, i;
4000
4001         flex_filter = rte_zmalloc("e1000_flex_filter",
4002                         sizeof(struct e1000_flex_filter), 0);
4003         if (flex_filter == NULL)
4004                 return -ENOMEM;
4005
4006         flex_filter->filter_info.len = filter->len;
4007         flex_filter->filter_info.priority = filter->priority;
4008         memcpy(flex_filter->filter_info.dwords, filter->bytes, filter->len);
4009         for (i = 0; i < RTE_ALIGN(filter->len, CHAR_BIT) / CHAR_BIT; i++) {
4010                 mask = 0;
4011                 /* reverse bits in flex filter's mask*/
4012                 for (shift = 0; shift < CHAR_BIT; shift++) {
4013                         if (filter->mask[i] & (0x01 << shift))
4014                                 mask |= (0x80 >> shift);
4015                 }
4016                 flex_filter->filter_info.mask[i] = mask;
4017         }
4018
4019         it = eth_igb_flex_filter_lookup(&filter_info->flex_list,
4020                                 &flex_filter->filter_info);
4021         if (it == NULL && !add) {
4022                 PMD_DRV_LOG(ERR, "filter doesn't exist.");
4023                 rte_free(flex_filter);
4024                 return -ENOENT;
4025         }
4026         if (it != NULL && add) {
4027                 PMD_DRV_LOG(ERR, "filter exists.");
4028                 rte_free(flex_filter);
4029                 return -EEXIST;
4030         }
4031
4032         if (add) {
4033                 flex_filter->queue = filter->queue;
4034                 /*
4035                  * look for an unused flex filter index
4036                  * and insert the filter into the list.
4037                  */
4038                 for (i = 0; i < E1000_MAX_FLEX_FILTERS; i++) {
4039                         if (!(filter_info->flex_mask & (1 << i))) {
4040                                 filter_info->flex_mask |= 1 << i;
4041                                 flex_filter->index = i;
4042                                 TAILQ_INSERT_TAIL(&filter_info->flex_list,
4043                                         flex_filter,
4044                                         entries);
4045                                 break;
4046                         }
4047                 }
4048                 if (i >= E1000_MAX_FLEX_FILTERS) {
4049                         PMD_DRV_LOG(ERR, "flex filters are full.");
4050                         rte_free(flex_filter);
4051                         return -ENOSYS;
4052                 }
4053
4054                 igb_inject_flex_filter(dev, flex_filter);
4055
4056         } else {
4057                 igb_remove_flex_filter(dev, it);
4058                 rte_free(flex_filter);
4059         }
4060
4061         return 0;
4062 }
4063
4064 /* translate elements in struct rte_eth_ntuple_filter to struct e1000_5tuple_filter_info*/
4065 static inline int
4066 ntuple_filter_to_5tuple_82576(struct rte_eth_ntuple_filter *filter,
4067                         struct e1000_5tuple_filter_info *filter_info)
4068 {
4069         if (filter->queue >= IGB_MAX_RX_QUEUE_NUM_82576)
4070                 return -EINVAL;
4071         if (filter->priority > E1000_2TUPLE_MAX_PRI)
4072                 return -EINVAL;  /* filter index is out of range. */
4073         if (filter->tcp_flags > RTE_NTUPLE_TCP_FLAGS_MASK)
4074                 return -EINVAL;  /* flags is invalid. */
4075
4076         switch (filter->dst_ip_mask) {
4077         case UINT32_MAX:
4078                 filter_info->dst_ip_mask = 0;
4079                 filter_info->dst_ip = filter->dst_ip;
4080                 break;
4081         case 0:
4082                 filter_info->dst_ip_mask = 1;
4083                 break;
4084         default:
4085                 PMD_DRV_LOG(ERR, "invalid dst_ip mask.");
4086                 return -EINVAL;
4087         }
4088
4089         switch (filter->src_ip_mask) {
4090         case UINT32_MAX:
4091                 filter_info->src_ip_mask = 0;
4092                 filter_info->src_ip = filter->src_ip;
4093                 break;
4094         case 0:
4095                 filter_info->src_ip_mask = 1;
4096                 break;
4097         default:
4098                 PMD_DRV_LOG(ERR, "invalid src_ip mask.");
4099                 return -EINVAL;
4100         }
4101
4102         switch (filter->dst_port_mask) {
4103         case UINT16_MAX:
4104                 filter_info->dst_port_mask = 0;
4105                 filter_info->dst_port = filter->dst_port;
4106                 break;
4107         case 0:
4108                 filter_info->dst_port_mask = 1;
4109                 break;
4110         default:
4111                 PMD_DRV_LOG(ERR, "invalid dst_port mask.");
4112                 return -EINVAL;
4113         }
4114
4115         switch (filter->src_port_mask) {
4116         case UINT16_MAX:
4117                 filter_info->src_port_mask = 0;
4118                 filter_info->src_port = filter->src_port;
4119                 break;
4120         case 0:
4121                 filter_info->src_port_mask = 1;
4122                 break;
4123         default:
4124                 PMD_DRV_LOG(ERR, "invalid src_port mask.");
4125                 return -EINVAL;
4126         }
4127
4128         switch (filter->proto_mask) {
4129         case UINT8_MAX:
4130                 filter_info->proto_mask = 0;
4131                 filter_info->proto = filter->proto;
4132                 break;
4133         case 0:
4134                 filter_info->proto_mask = 1;
4135                 break;
4136         default:
4137                 PMD_DRV_LOG(ERR, "invalid protocol mask.");
4138                 return -EINVAL;
4139         }
4140
4141         filter_info->priority = (uint8_t)filter->priority;
4142         if (filter->flags & RTE_NTUPLE_FLAGS_TCP_FLAG)
4143                 filter_info->tcp_flags = filter->tcp_flags;
4144         else
4145                 filter_info->tcp_flags = 0;
4146
4147         return 0;
4148 }
4149
4150 static inline struct e1000_5tuple_filter *
4151 igb_5tuple_filter_lookup_82576(struct e1000_5tuple_filter_list *filter_list,
4152                         struct e1000_5tuple_filter_info *key)
4153 {
4154         struct e1000_5tuple_filter *it;
4155
4156         TAILQ_FOREACH(it, filter_list, entries) {
4157                 if (memcmp(key, &it->filter_info,
4158                         sizeof(struct e1000_5tuple_filter_info)) == 0) {
4159                         return it;
4160                 }
4161         }
4162         return NULL;
4163 }
4164
4165 /* inject a igb 5-tuple filter to HW */
4166 static inline void
4167 igb_inject_5tuple_filter_82576(struct rte_eth_dev *dev,
4168                            struct e1000_5tuple_filter *filter)
4169 {
4170         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4171         uint32_t ftqf = E1000_FTQF_VF_BP | E1000_FTQF_MASK;
4172         uint32_t spqf, imir, imir_ext = E1000_IMIREXT_SIZE_BP;
4173         uint8_t i;
4174
4175         i = filter->index;
4176         ftqf |= filter->filter_info.proto & E1000_FTQF_PROTOCOL_MASK;
4177         if (filter->filter_info.src_ip_mask == 0) /* 0b means compare. */
4178                 ftqf &= ~E1000_FTQF_MASK_SOURCE_ADDR_BP;
4179         if (filter->filter_info.dst_ip_mask == 0)
4180                 ftqf &= ~E1000_FTQF_MASK_DEST_ADDR_BP;
4181         if (filter->filter_info.src_port_mask == 0)
4182                 ftqf &= ~E1000_FTQF_MASK_SOURCE_PORT_BP;
4183         if (filter->filter_info.proto_mask == 0)
4184                 ftqf &= ~E1000_FTQF_MASK_PROTO_BP;
4185         ftqf |= (filter->queue << E1000_FTQF_QUEUE_SHIFT) &
4186                 E1000_FTQF_QUEUE_MASK;
4187         ftqf |= E1000_FTQF_QUEUE_ENABLE;
4188         E1000_WRITE_REG(hw, E1000_FTQF(i), ftqf);
4189         E1000_WRITE_REG(hw, E1000_DAQF(i), filter->filter_info.dst_ip);
4190         E1000_WRITE_REG(hw, E1000_SAQF(i), filter->filter_info.src_ip);
4191
4192         spqf = filter->filter_info.src_port & E1000_SPQF_SRCPORT;
4193         E1000_WRITE_REG(hw, E1000_SPQF(i), spqf);
4194
4195         imir = (uint32_t)(filter->filter_info.dst_port & E1000_IMIR_DSTPORT);
4196         if (filter->filter_info.dst_port_mask == 1) /* 1b means not compare. */
4197                 imir |= E1000_IMIR_PORT_BP;
4198         else
4199                 imir &= ~E1000_IMIR_PORT_BP;
4200         imir |= filter->filter_info.priority << E1000_IMIR_PRIORITY_SHIFT;
4201
4202         /* tcp flags bits setting. */
4203         if (filter->filter_info.tcp_flags & RTE_NTUPLE_TCP_FLAGS_MASK) {
4204                 if (filter->filter_info.tcp_flags & RTE_TCP_URG_FLAG)
4205                         imir_ext |= E1000_IMIREXT_CTRL_URG;
4206                 if (filter->filter_info.tcp_flags & RTE_TCP_ACK_FLAG)
4207                         imir_ext |= E1000_IMIREXT_CTRL_ACK;
4208                 if (filter->filter_info.tcp_flags & RTE_TCP_PSH_FLAG)
4209                         imir_ext |= E1000_IMIREXT_CTRL_PSH;
4210                 if (filter->filter_info.tcp_flags & RTE_TCP_RST_FLAG)
4211                         imir_ext |= E1000_IMIREXT_CTRL_RST;
4212                 if (filter->filter_info.tcp_flags & RTE_TCP_SYN_FLAG)
4213                         imir_ext |= E1000_IMIREXT_CTRL_SYN;
4214                 if (filter->filter_info.tcp_flags & RTE_TCP_FIN_FLAG)
4215                         imir_ext |= E1000_IMIREXT_CTRL_FIN;
4216         } else {
4217                 imir_ext |= E1000_IMIREXT_CTRL_BP;
4218         }
4219         E1000_WRITE_REG(hw, E1000_IMIR(i), imir);
4220         E1000_WRITE_REG(hw, E1000_IMIREXT(i), imir_ext);
4221 }
4222
4223 /*
4224  * igb_add_5tuple_filter_82576 - add a 5tuple filter
4225  *
4226  * @param
4227  * dev: Pointer to struct rte_eth_dev.
4228  * ntuple_filter: ponter to the filter that will be added.
4229  *
4230  * @return
4231  *    - On success, zero.
4232  *    - On failure, a negative value.
4233  */
4234 static int
4235 igb_add_5tuple_filter_82576(struct rte_eth_dev *dev,
4236                         struct rte_eth_ntuple_filter *ntuple_filter)
4237 {
4238         struct e1000_filter_info *filter_info =
4239                 E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
4240         struct e1000_5tuple_filter *filter;
4241         uint8_t i;
4242         int ret;
4243
4244         filter = rte_zmalloc("e1000_5tuple_filter",
4245                         sizeof(struct e1000_5tuple_filter), 0);
4246         if (filter == NULL)
4247                 return -ENOMEM;
4248
4249         ret = ntuple_filter_to_5tuple_82576(ntuple_filter,
4250                                             &filter->filter_info);
4251         if (ret < 0) {
4252                 rte_free(filter);
4253                 return ret;
4254         }
4255
4256         if (igb_5tuple_filter_lookup_82576(&filter_info->fivetuple_list,
4257                                          &filter->filter_info) != NULL) {
4258                 PMD_DRV_LOG(ERR, "filter exists.");
4259                 rte_free(filter);
4260                 return -EEXIST;
4261         }
4262         filter->queue = ntuple_filter->queue;
4263
4264         /*
4265          * look for an unused 5tuple filter index,
4266          * and insert the filter to list.
4267          */
4268         for (i = 0; i < E1000_MAX_FTQF_FILTERS; i++) {
4269                 if (!(filter_info->fivetuple_mask & (1 << i))) {
4270                         filter_info->fivetuple_mask |= 1 << i;
4271                         filter->index = i;
4272                         TAILQ_INSERT_TAIL(&filter_info->fivetuple_list,
4273                                           filter,
4274                                           entries);
4275                         break;
4276                 }
4277         }
4278         if (i >= E1000_MAX_FTQF_FILTERS) {
4279                 PMD_DRV_LOG(ERR, "5tuple filters are full.");
4280                 rte_free(filter);
4281                 return -ENOSYS;
4282         }
4283
4284         igb_inject_5tuple_filter_82576(dev, filter);
4285         return 0;
4286 }
4287
4288 int
4289 igb_delete_5tuple_filter_82576(struct rte_eth_dev *dev,
4290                                 struct e1000_5tuple_filter *filter)
4291 {
4292         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4293         struct e1000_filter_info *filter_info =
4294                 E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
4295
4296         filter_info->fivetuple_mask &= ~(1 << filter->index);
4297         TAILQ_REMOVE(&filter_info->fivetuple_list, filter, entries);
4298         rte_free(filter);
4299
4300         E1000_WRITE_REG(hw, E1000_FTQF(filter->index),
4301                         E1000_FTQF_VF_BP | E1000_FTQF_MASK);
4302         E1000_WRITE_REG(hw, E1000_DAQF(filter->index), 0);
4303         E1000_WRITE_REG(hw, E1000_SAQF(filter->index), 0);
4304         E1000_WRITE_REG(hw, E1000_SPQF(filter->index), 0);
4305         E1000_WRITE_REG(hw, E1000_IMIR(filter->index), 0);
4306         E1000_WRITE_REG(hw, E1000_IMIREXT(filter->index), 0);
4307         return 0;
4308 }
4309
4310 /*
4311  * igb_remove_5tuple_filter_82576 - remove a 5tuple filter
4312  *
4313  * @param
4314  * dev: Pointer to struct rte_eth_dev.
4315  * ntuple_filter: ponter to the filter that will be removed.
4316  *
4317  * @return
4318  *    - On success, zero.
4319  *    - On failure, a negative value.
4320  */
4321 static int
4322 igb_remove_5tuple_filter_82576(struct rte_eth_dev *dev,
4323                                 struct rte_eth_ntuple_filter *ntuple_filter)
4324 {
4325         struct e1000_filter_info *filter_info =
4326                 E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
4327         struct e1000_5tuple_filter_info filter_5tuple;
4328         struct e1000_5tuple_filter *filter;
4329         int ret;
4330
4331         memset(&filter_5tuple, 0, sizeof(struct e1000_5tuple_filter_info));
4332         ret = ntuple_filter_to_5tuple_82576(ntuple_filter,
4333                                             &filter_5tuple);
4334         if (ret < 0)
4335                 return ret;
4336
4337         filter = igb_5tuple_filter_lookup_82576(&filter_info->fivetuple_list,
4338                                          &filter_5tuple);
4339         if (filter == NULL) {
4340                 PMD_DRV_LOG(ERR, "filter doesn't exist.");
4341                 return -ENOENT;
4342         }
4343
4344         igb_delete_5tuple_filter_82576(dev, filter);
4345
4346         return 0;
4347 }
4348
4349 static int
4350 eth_igb_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
4351 {
4352         uint32_t rctl;
4353         struct e1000_hw *hw;
4354         uint32_t frame_size = mtu + E1000_ETH_OVERHEAD;
4355
4356         hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4357
4358 #ifdef RTE_LIBRTE_82571_SUPPORT
4359         /* XXX: not bigger than max_rx_pktlen */
4360         if (hw->mac.type == e1000_82571)
4361                 return -ENOTSUP;
4362 #endif
4363         /*
4364          * If device is started, refuse mtu that requires the support of
4365          * scattered packets when this feature has not been enabled before.
4366          */
4367         if (dev->data->dev_started && !dev->data->scattered_rx &&
4368             frame_size > dev->data->min_rx_buf_size - RTE_PKTMBUF_HEADROOM) {
4369                 PMD_INIT_LOG(ERR, "Stop port first.");
4370                 return -EINVAL;
4371         }
4372
4373         rctl = E1000_READ_REG(hw, E1000_RCTL);
4374
4375         /* switch to jumbo mode if needed */
4376         if (mtu > RTE_ETHER_MTU)
4377                 rctl |= E1000_RCTL_LPE;
4378         else
4379                 rctl &= ~E1000_RCTL_LPE;
4380         E1000_WRITE_REG(hw, E1000_RCTL, rctl);
4381
4382         E1000_WRITE_REG(hw, E1000_RLPML, frame_size);
4383
4384         return 0;
4385 }
4386
4387 /*
4388  * igb_add_del_ntuple_filter - add or delete a ntuple filter
4389  *
4390  * @param
4391  * dev: Pointer to struct rte_eth_dev.
4392  * ntuple_filter: Pointer to struct rte_eth_ntuple_filter
4393  * add: if true, add filter, if false, remove filter
4394  *
4395  * @return
4396  *    - On success, zero.
4397  *    - On failure, a negative value.
4398  */
4399 int
4400 igb_add_del_ntuple_filter(struct rte_eth_dev *dev,
4401                         struct rte_eth_ntuple_filter *ntuple_filter,
4402                         bool add)
4403 {
4404         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4405         int ret;
4406
4407         switch (ntuple_filter->flags) {
4408         case RTE_5TUPLE_FLAGS:
4409         case (RTE_5TUPLE_FLAGS | RTE_NTUPLE_FLAGS_TCP_FLAG):
4410                 if (hw->mac.type != e1000_82576)
4411                         return -ENOTSUP;
4412                 if (add)
4413                         ret = igb_add_5tuple_filter_82576(dev,
4414                                                           ntuple_filter);
4415                 else
4416                         ret = igb_remove_5tuple_filter_82576(dev,
4417                                                              ntuple_filter);
4418                 break;
4419         case RTE_2TUPLE_FLAGS:
4420         case (RTE_2TUPLE_FLAGS | RTE_NTUPLE_FLAGS_TCP_FLAG):
4421                 if (hw->mac.type != e1000_82580 && hw->mac.type != e1000_i350 &&
4422                         hw->mac.type != e1000_i210 &&
4423                         hw->mac.type != e1000_i211)
4424                         return -ENOTSUP;
4425                 if (add)
4426                         ret = igb_add_2tuple_filter(dev, ntuple_filter);
4427                 else
4428                         ret = igb_remove_2tuple_filter(dev, ntuple_filter);
4429                 break;
4430         default:
4431                 ret = -EINVAL;
4432                 break;
4433         }
4434
4435         return ret;
4436 }
4437
4438 static inline int
4439 igb_ethertype_filter_lookup(struct e1000_filter_info *filter_info,
4440                         uint16_t ethertype)
4441 {
4442         int i;
4443
4444         for (i = 0; i < E1000_MAX_ETQF_FILTERS; i++) {
4445                 if (filter_info->ethertype_filters[i].ethertype == ethertype &&
4446                     (filter_info->ethertype_mask & (1 << i)))
4447                         return i;
4448         }
4449         return -1;
4450 }
4451
4452 static inline int
4453 igb_ethertype_filter_insert(struct e1000_filter_info *filter_info,
4454                         uint16_t ethertype, uint32_t etqf)
4455 {
4456         int i;
4457
4458         for (i = 0; i < E1000_MAX_ETQF_FILTERS; i++) {
4459                 if (!(filter_info->ethertype_mask & (1 << i))) {
4460                         filter_info->ethertype_mask |= 1 << i;
4461                         filter_info->ethertype_filters[i].ethertype = ethertype;
4462                         filter_info->ethertype_filters[i].etqf = etqf;
4463                         return i;
4464                 }
4465         }
4466         return -1;
4467 }
4468
4469 int
4470 igb_ethertype_filter_remove(struct e1000_filter_info *filter_info,
4471                         uint8_t idx)
4472 {
4473         if (idx >= E1000_MAX_ETQF_FILTERS)
4474                 return -1;
4475         filter_info->ethertype_mask &= ~(1 << idx);
4476         filter_info->ethertype_filters[idx].ethertype = 0;
4477         filter_info->ethertype_filters[idx].etqf = 0;
4478         return idx;
4479 }
4480
4481
4482 int
4483 igb_add_del_ethertype_filter(struct rte_eth_dev *dev,
4484                         struct rte_eth_ethertype_filter *filter,
4485                         bool add)
4486 {
4487         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4488         struct e1000_filter_info *filter_info =
4489                 E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
4490         uint32_t etqf = 0;
4491         int ret;
4492
4493         if (filter->ether_type == RTE_ETHER_TYPE_IPV4 ||
4494                 filter->ether_type == RTE_ETHER_TYPE_IPV6) {
4495                 PMD_DRV_LOG(ERR, "unsupported ether_type(0x%04x) in"
4496                         " ethertype filter.", filter->ether_type);
4497                 return -EINVAL;
4498         }
4499
4500         if (filter->flags & RTE_ETHTYPE_FLAGS_MAC) {
4501                 PMD_DRV_LOG(ERR, "mac compare is unsupported.");
4502                 return -EINVAL;
4503         }
4504         if (filter->flags & RTE_ETHTYPE_FLAGS_DROP) {
4505                 PMD_DRV_LOG(ERR, "drop option is unsupported.");
4506                 return -EINVAL;
4507         }
4508
4509         ret = igb_ethertype_filter_lookup(filter_info, filter->ether_type);
4510         if (ret >= 0 && add) {
4511                 PMD_DRV_LOG(ERR, "ethertype (0x%04x) filter exists.",
4512                             filter->ether_type);
4513                 return -EEXIST;
4514         }
4515         if (ret < 0 && !add) {
4516                 PMD_DRV_LOG(ERR, "ethertype (0x%04x) filter doesn't exist.",
4517                             filter->ether_type);
4518                 return -ENOENT;
4519         }
4520
4521         if (add) {
4522                 etqf |= E1000_ETQF_FILTER_ENABLE | E1000_ETQF_QUEUE_ENABLE;
4523                 etqf |= (uint32_t)(filter->ether_type & E1000_ETQF_ETHERTYPE);
4524                 etqf |= filter->queue << E1000_ETQF_QUEUE_SHIFT;
4525                 ret = igb_ethertype_filter_insert(filter_info,
4526                                 filter->ether_type, etqf);
4527                 if (ret < 0) {
4528                         PMD_DRV_LOG(ERR, "ethertype filters are full.");
4529                         return -ENOSYS;
4530                 }
4531         } else {
4532                 ret = igb_ethertype_filter_remove(filter_info, (uint8_t)ret);
4533                 if (ret < 0)
4534                         return -ENOSYS;
4535         }
4536         E1000_WRITE_REG(hw, E1000_ETQF(ret), etqf);
4537         E1000_WRITE_FLUSH(hw);
4538
4539         return 0;
4540 }
4541
4542 static int
4543 eth_igb_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
4544                      const struct rte_flow_ops **ops)
4545 {
4546         *ops = &igb_flow_ops;
4547         return 0;
4548 }
4549
4550 static int
4551 eth_igb_set_mc_addr_list(struct rte_eth_dev *dev,
4552                          struct rte_ether_addr *mc_addr_set,
4553                          uint32_t nb_mc_addr)
4554 {
4555         struct e1000_hw *hw;
4556
4557         hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4558         e1000_update_mc_addr_list(hw, (u8 *)mc_addr_set, nb_mc_addr);
4559         return 0;
4560 }
4561
4562 static uint64_t
4563 igb_read_systime_cyclecounter(struct rte_eth_dev *dev)
4564 {
4565         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4566         uint64_t systime_cycles;
4567
4568         switch (hw->mac.type) {
4569         case e1000_i210:
4570         case e1000_i211:
4571                 /*
4572                  * Need to read System Time Residue Register to be able
4573                  * to read the other two registers.
4574                  */
4575                 E1000_READ_REG(hw, E1000_SYSTIMR);
4576                 /* SYSTIMEL stores ns and SYSTIMEH stores seconds. */
4577                 systime_cycles = (uint64_t)E1000_READ_REG(hw, E1000_SYSTIML);
4578                 systime_cycles += (uint64_t)E1000_READ_REG(hw, E1000_SYSTIMH)
4579                                 * NSEC_PER_SEC;
4580                 break;
4581         case e1000_82580:
4582         case e1000_i350:
4583         case e1000_i354:
4584                 /*
4585                  * Need to read System Time Residue Register to be able
4586                  * to read the other two registers.
4587                  */
4588                 E1000_READ_REG(hw, E1000_SYSTIMR);
4589                 systime_cycles = (uint64_t)E1000_READ_REG(hw, E1000_SYSTIML);
4590                 /* Only the 8 LSB are valid. */
4591                 systime_cycles |= (uint64_t)(E1000_READ_REG(hw, E1000_SYSTIMH)
4592                                 & 0xff) << 32;
4593                 break;
4594         default:
4595                 systime_cycles = (uint64_t)E1000_READ_REG(hw, E1000_SYSTIML);
4596                 systime_cycles |= (uint64_t)E1000_READ_REG(hw, E1000_SYSTIMH)
4597                                 << 32;
4598                 break;
4599         }
4600
4601         return systime_cycles;
4602 }
4603
4604 static uint64_t
4605 igb_read_rx_tstamp_cyclecounter(struct rte_eth_dev *dev)
4606 {
4607         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4608         uint64_t rx_tstamp_cycles;
4609
4610         switch (hw->mac.type) {
4611         case e1000_i210:
4612         case e1000_i211:
4613                 /* RXSTMPL stores ns and RXSTMPH stores seconds. */
4614                 rx_tstamp_cycles = (uint64_t)E1000_READ_REG(hw, E1000_RXSTMPL);
4615                 rx_tstamp_cycles += (uint64_t)E1000_READ_REG(hw, E1000_RXSTMPH)
4616                                 * NSEC_PER_SEC;
4617                 break;
4618         case e1000_82580:
4619         case e1000_i350:
4620         case e1000_i354:
4621                 rx_tstamp_cycles = (uint64_t)E1000_READ_REG(hw, E1000_RXSTMPL);
4622                 /* Only the 8 LSB are valid. */
4623                 rx_tstamp_cycles |= (uint64_t)(E1000_READ_REG(hw, E1000_RXSTMPH)
4624                                 & 0xff) << 32;
4625                 break;
4626         default:
4627                 rx_tstamp_cycles = (uint64_t)E1000_READ_REG(hw, E1000_RXSTMPL);
4628                 rx_tstamp_cycles |= (uint64_t)E1000_READ_REG(hw, E1000_RXSTMPH)
4629                                 << 32;
4630                 break;
4631         }
4632
4633         return rx_tstamp_cycles;
4634 }
4635
4636 static uint64_t
4637 igb_read_tx_tstamp_cyclecounter(struct rte_eth_dev *dev)
4638 {
4639         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4640         uint64_t tx_tstamp_cycles;
4641
4642         switch (hw->mac.type) {
4643         case e1000_i210:
4644         case e1000_i211:
4645                 /* RXSTMPL stores ns and RXSTMPH stores seconds. */
4646                 tx_tstamp_cycles = (uint64_t)E1000_READ_REG(hw, E1000_TXSTMPL);
4647                 tx_tstamp_cycles += (uint64_t)E1000_READ_REG(hw, E1000_TXSTMPH)
4648                                 * NSEC_PER_SEC;
4649                 break;
4650         case e1000_82580:
4651         case e1000_i350:
4652         case e1000_i354:
4653                 tx_tstamp_cycles = (uint64_t)E1000_READ_REG(hw, E1000_TXSTMPL);
4654                 /* Only the 8 LSB are valid. */
4655                 tx_tstamp_cycles |= (uint64_t)(E1000_READ_REG(hw, E1000_TXSTMPH)
4656                                 & 0xff) << 32;
4657                 break;
4658         default:
4659                 tx_tstamp_cycles = (uint64_t)E1000_READ_REG(hw, E1000_TXSTMPL);
4660                 tx_tstamp_cycles |= (uint64_t)E1000_READ_REG(hw, E1000_TXSTMPH)
4661                                 << 32;
4662                 break;
4663         }
4664
4665         return tx_tstamp_cycles;
4666 }
4667
4668 static void
4669 igb_start_timecounters(struct rte_eth_dev *dev)
4670 {
4671         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4672         struct e1000_adapter *adapter = dev->data->dev_private;
4673         uint32_t incval = 1;
4674         uint32_t shift = 0;
4675         uint64_t mask = E1000_CYCLECOUNTER_MASK;
4676
4677         switch (hw->mac.type) {
4678         case e1000_82580:
4679         case e1000_i350:
4680         case e1000_i354:
4681                 /* 32 LSB bits + 8 MSB bits = 40 bits */
4682                 mask = (1ULL << 40) - 1;
4683                 /* fall-through */
4684         case e1000_i210:
4685         case e1000_i211:
4686                 /*
4687                  * Start incrementing the register
4688                  * used to timestamp PTP packets.
4689                  */
4690                 E1000_WRITE_REG(hw, E1000_TIMINCA, incval);
4691                 break;
4692         case e1000_82576:
4693                 incval = E1000_INCVALUE_82576;
4694                 shift = IGB_82576_TSYNC_SHIFT;
4695                 E1000_WRITE_REG(hw, E1000_TIMINCA,
4696                                 E1000_INCPERIOD_82576 | incval);
4697                 break;
4698         default:
4699                 /* Not supported */
4700                 return;
4701         }
4702
4703         memset(&adapter->systime_tc, 0, sizeof(struct rte_timecounter));
4704         memset(&adapter->rx_tstamp_tc, 0, sizeof(struct rte_timecounter));
4705         memset(&adapter->tx_tstamp_tc, 0, sizeof(struct rte_timecounter));
4706
4707         adapter->systime_tc.cc_mask = mask;
4708         adapter->systime_tc.cc_shift = shift;
4709         adapter->systime_tc.nsec_mask = (1ULL << shift) - 1;
4710
4711         adapter->rx_tstamp_tc.cc_mask = mask;
4712         adapter->rx_tstamp_tc.cc_shift = shift;
4713         adapter->rx_tstamp_tc.nsec_mask = (1ULL << shift) - 1;
4714
4715         adapter->tx_tstamp_tc.cc_mask = mask;
4716         adapter->tx_tstamp_tc.cc_shift = shift;
4717         adapter->tx_tstamp_tc.nsec_mask = (1ULL << shift) - 1;
4718 }
4719
4720 static int
4721 igb_timesync_adjust_time(struct rte_eth_dev *dev, int64_t delta)
4722 {
4723         struct e1000_adapter *adapter = dev->data->dev_private;
4724
4725         adapter->systime_tc.nsec += delta;
4726         adapter->rx_tstamp_tc.nsec += delta;
4727         adapter->tx_tstamp_tc.nsec += delta;
4728
4729         return 0;
4730 }
4731
4732 static int
4733 igb_timesync_write_time(struct rte_eth_dev *dev, const struct timespec *ts)
4734 {
4735         uint64_t ns;
4736         struct e1000_adapter *adapter = dev->data->dev_private;
4737
4738         ns = rte_timespec_to_ns(ts);
4739
4740         /* Set the timecounters to a new value. */
4741         adapter->systime_tc.nsec = ns;
4742         adapter->rx_tstamp_tc.nsec = ns;
4743         adapter->tx_tstamp_tc.nsec = ns;
4744
4745         return 0;
4746 }
4747
4748 static int
4749 igb_timesync_read_time(struct rte_eth_dev *dev, struct timespec *ts)
4750 {
4751         uint64_t ns, systime_cycles;
4752         struct e1000_adapter *adapter = dev->data->dev_private;
4753
4754         systime_cycles = igb_read_systime_cyclecounter(dev);
4755         ns = rte_timecounter_update(&adapter->systime_tc, systime_cycles);
4756         *ts = rte_ns_to_timespec(ns);
4757
4758         return 0;
4759 }
4760
4761 static int
4762 igb_timesync_enable(struct rte_eth_dev *dev)
4763 {
4764         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4765         uint32_t tsync_ctl;
4766         uint32_t tsauxc;
4767
4768         /* Stop the timesync system time. */
4769         E1000_WRITE_REG(hw, E1000_TIMINCA, 0x0);
4770         /* Reset the timesync system time value. */
4771         switch (hw->mac.type) {
4772         case e1000_82580:
4773         case e1000_i350:
4774         case e1000_i354:
4775         case e1000_i210:
4776         case e1000_i211:
4777                 E1000_WRITE_REG(hw, E1000_SYSTIMR, 0x0);
4778                 /* fall-through */
4779         case e1000_82576:
4780                 E1000_WRITE_REG(hw, E1000_SYSTIML, 0x0);
4781                 E1000_WRITE_REG(hw, E1000_SYSTIMH, 0x0);
4782                 break;
4783         default:
4784                 /* Not supported. */
4785                 return -ENOTSUP;
4786         }
4787
4788         /* Enable system time for it isn't on by default. */
4789         tsauxc = E1000_READ_REG(hw, E1000_TSAUXC);
4790         tsauxc &= ~E1000_TSAUXC_DISABLE_SYSTIME;
4791         E1000_WRITE_REG(hw, E1000_TSAUXC, tsauxc);
4792
4793         igb_start_timecounters(dev);
4794
4795         /* Enable L2 filtering of IEEE1588/802.1AS Ethernet frame types. */
4796         E1000_WRITE_REG(hw, E1000_ETQF(E1000_ETQF_FILTER_1588),
4797                         (RTE_ETHER_TYPE_1588 |
4798                          E1000_ETQF_FILTER_ENABLE |
4799                          E1000_ETQF_1588));
4800
4801         /* Enable timestamping of received PTP packets. */
4802         tsync_ctl = E1000_READ_REG(hw, E1000_TSYNCRXCTL);
4803         tsync_ctl |= E1000_TSYNCRXCTL_ENABLED;
4804         E1000_WRITE_REG(hw, E1000_TSYNCRXCTL, tsync_ctl);
4805
4806         /* Enable Timestamping of transmitted PTP packets. */
4807         tsync_ctl = E1000_READ_REG(hw, E1000_TSYNCTXCTL);
4808         tsync_ctl |= E1000_TSYNCTXCTL_ENABLED;
4809         E1000_WRITE_REG(hw, E1000_TSYNCTXCTL, tsync_ctl);
4810
4811         return 0;
4812 }
4813
4814 static int
4815 igb_timesync_disable(struct rte_eth_dev *dev)
4816 {
4817         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4818         uint32_t tsync_ctl;
4819
4820         /* Disable timestamping of transmitted PTP packets. */
4821         tsync_ctl = E1000_READ_REG(hw, E1000_TSYNCTXCTL);
4822         tsync_ctl &= ~E1000_TSYNCTXCTL_ENABLED;
4823         E1000_WRITE_REG(hw, E1000_TSYNCTXCTL, tsync_ctl);
4824
4825         /* Disable timestamping of received PTP packets. */
4826         tsync_ctl = E1000_READ_REG(hw, E1000_TSYNCRXCTL);
4827         tsync_ctl &= ~E1000_TSYNCRXCTL_ENABLED;
4828         E1000_WRITE_REG(hw, E1000_TSYNCRXCTL, tsync_ctl);
4829
4830         /* Disable L2 filtering of IEEE1588/802.1AS Ethernet frame types. */
4831         E1000_WRITE_REG(hw, E1000_ETQF(E1000_ETQF_FILTER_1588), 0);
4832
4833         /* Stop incrementating the System Time registers. */
4834         E1000_WRITE_REG(hw, E1000_TIMINCA, 0);
4835
4836         return 0;
4837 }
4838
4839 static int
4840 igb_timesync_read_rx_timestamp(struct rte_eth_dev *dev,
4841                                struct timespec *timestamp,
4842                                uint32_t flags __rte_unused)
4843 {
4844         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4845         struct e1000_adapter *adapter = dev->data->dev_private;
4846         uint32_t tsync_rxctl;
4847         uint64_t rx_tstamp_cycles;
4848         uint64_t ns;
4849
4850         tsync_rxctl = E1000_READ_REG(hw, E1000_TSYNCRXCTL);
4851         if ((tsync_rxctl & E1000_TSYNCRXCTL_VALID) == 0)
4852                 return -EINVAL;
4853
4854         rx_tstamp_cycles = igb_read_rx_tstamp_cyclecounter(dev);
4855         ns = rte_timecounter_update(&adapter->rx_tstamp_tc, rx_tstamp_cycles);
4856         *timestamp = rte_ns_to_timespec(ns);
4857
4858         return  0;
4859 }
4860
4861 static int
4862 igb_timesync_read_tx_timestamp(struct rte_eth_dev *dev,
4863                                struct timespec *timestamp)
4864 {
4865         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4866         struct e1000_adapter *adapter = dev->data->dev_private;
4867         uint32_t tsync_txctl;
4868         uint64_t tx_tstamp_cycles;
4869         uint64_t ns;
4870
4871         tsync_txctl = E1000_READ_REG(hw, E1000_TSYNCTXCTL);
4872         if ((tsync_txctl & E1000_TSYNCTXCTL_VALID) == 0)
4873                 return -EINVAL;
4874
4875         tx_tstamp_cycles = igb_read_tx_tstamp_cyclecounter(dev);
4876         ns = rte_timecounter_update(&adapter->tx_tstamp_tc, tx_tstamp_cycles);
4877         *timestamp = rte_ns_to_timespec(ns);
4878
4879         return  0;
4880 }
4881
4882 static int
4883 eth_igb_get_reg_length(struct rte_eth_dev *dev __rte_unused)
4884 {
4885         int count = 0;
4886         int g_ind = 0;
4887         const struct reg_info *reg_group;
4888
4889         while ((reg_group = igb_regs[g_ind++]))
4890                 count += igb_reg_group_count(reg_group);
4891
4892         return count;
4893 }
4894
4895 static int
4896 igbvf_get_reg_length(struct rte_eth_dev *dev __rte_unused)
4897 {
4898         int count = 0;
4899         int g_ind = 0;
4900         const struct reg_info *reg_group;
4901
4902         while ((reg_group = igbvf_regs[g_ind++]))
4903                 count += igb_reg_group_count(reg_group);
4904
4905         return count;
4906 }
4907
4908 static int
4909 eth_igb_get_regs(struct rte_eth_dev *dev,
4910         struct rte_dev_reg_info *regs)
4911 {
4912         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4913         uint32_t *data = regs->data;
4914         int g_ind = 0;
4915         int count = 0;
4916         const struct reg_info *reg_group;
4917
4918         if (data == NULL) {
4919                 regs->length = eth_igb_get_reg_length(dev);
4920                 regs->width = sizeof(uint32_t);
4921                 return 0;
4922         }
4923
4924         /* Support only full register dump */
4925         if ((regs->length == 0) ||
4926             (regs->length == (uint32_t)eth_igb_get_reg_length(dev))) {
4927                 regs->version = hw->mac.type << 24 | hw->revision_id << 16 |
4928                         hw->device_id;
4929                 while ((reg_group = igb_regs[g_ind++]))
4930                         count += igb_read_regs_group(dev, &data[count],
4931                                                         reg_group);
4932                 return 0;
4933         }
4934
4935         return -ENOTSUP;
4936 }
4937
4938 static int
4939 igbvf_get_regs(struct rte_eth_dev *dev,
4940         struct rte_dev_reg_info *regs)
4941 {
4942         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4943         uint32_t *data = regs->data;
4944         int g_ind = 0;
4945         int count = 0;
4946         const struct reg_info *reg_group;
4947
4948         if (data == NULL) {
4949                 regs->length = igbvf_get_reg_length(dev);
4950                 regs->width = sizeof(uint32_t);
4951                 return 0;
4952         }
4953
4954         /* Support only full register dump */
4955         if ((regs->length == 0) ||
4956             (regs->length == (uint32_t)igbvf_get_reg_length(dev))) {
4957                 regs->version = hw->mac.type << 24 | hw->revision_id << 16 |
4958                         hw->device_id;
4959                 while ((reg_group = igbvf_regs[g_ind++]))
4960                         count += igb_read_regs_group(dev, &data[count],
4961                                                         reg_group);
4962                 return 0;
4963         }
4964
4965         return -ENOTSUP;
4966 }
4967
4968 static int
4969 eth_igb_get_eeprom_length(struct rte_eth_dev *dev)
4970 {
4971         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4972
4973         /* Return unit is byte count */
4974         return hw->nvm.word_size * 2;
4975 }
4976
4977 static int
4978 eth_igb_get_eeprom(struct rte_eth_dev *dev,
4979         struct rte_dev_eeprom_info *in_eeprom)
4980 {
4981         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4982         struct e1000_nvm_info *nvm = &hw->nvm;
4983         uint16_t *data = in_eeprom->data;
4984         int first, length;
4985
4986         first = in_eeprom->offset >> 1;
4987         length = in_eeprom->length >> 1;
4988         if ((first >= hw->nvm.word_size) ||
4989             ((first + length) >= hw->nvm.word_size))
4990                 return -EINVAL;
4991
4992         in_eeprom->magic = hw->vendor_id |
4993                 ((uint32_t)hw->device_id << 16);
4994
4995         if ((nvm->ops.read) == NULL)
4996                 return -ENOTSUP;
4997
4998         return nvm->ops.read(hw, first, length, data);
4999 }
5000
5001 static int
5002 eth_igb_set_eeprom(struct rte_eth_dev *dev,
5003         struct rte_dev_eeprom_info *in_eeprom)
5004 {
5005         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5006         struct e1000_nvm_info *nvm = &hw->nvm;
5007         uint16_t *data = in_eeprom->data;
5008         int first, length;
5009
5010         first = in_eeprom->offset >> 1;
5011         length = in_eeprom->length >> 1;
5012         if ((first >= hw->nvm.word_size) ||
5013             ((first + length) >= hw->nvm.word_size))
5014                 return -EINVAL;
5015
5016         in_eeprom->magic = (uint32_t)hw->vendor_id |
5017                 ((uint32_t)hw->device_id << 16);
5018
5019         if ((nvm->ops.write) == NULL)
5020                 return -ENOTSUP;
5021         return nvm->ops.write(hw,  first, length, data);
5022 }
5023
5024 static int
5025 eth_igb_get_module_info(struct rte_eth_dev *dev,
5026                         struct rte_eth_dev_module_info *modinfo)
5027 {
5028         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5029
5030         uint32_t status = 0;
5031         uint16_t sff8472_rev, addr_mode;
5032         bool page_swap = false;
5033
5034         if (hw->phy.media_type == e1000_media_type_copper ||
5035             hw->phy.media_type == e1000_media_type_unknown)
5036                 return -EOPNOTSUPP;
5037
5038         /* Check whether we support SFF-8472 or not */
5039         status = e1000_read_phy_reg_i2c(hw, IGB_SFF_8472_COMP, &sff8472_rev);
5040         if (status)
5041                 return -EIO;
5042
5043         /* addressing mode is not supported */
5044         status = e1000_read_phy_reg_i2c(hw, IGB_SFF_8472_SWAP, &addr_mode);
5045         if (status)
5046                 return -EIO;
5047
5048         /* addressing mode is not supported */
5049         if ((addr_mode & 0xFF) & IGB_SFF_ADDRESSING_MODE) {
5050                 PMD_DRV_LOG(ERR,
5051                             "Address change required to access page 0xA2, "
5052                             "but not supported. Please report the module "
5053                             "type to the driver maintainers.\n");
5054                 page_swap = true;
5055         }
5056
5057         if ((sff8472_rev & 0xFF) == IGB_SFF_8472_UNSUP || page_swap) {
5058                 /* We have an SFP, but it does not support SFF-8472 */
5059                 modinfo->type = RTE_ETH_MODULE_SFF_8079;
5060                 modinfo->eeprom_len = RTE_ETH_MODULE_SFF_8079_LEN;
5061         } else {
5062                 /* We have an SFP which supports a revision of SFF-8472 */
5063                 modinfo->type = RTE_ETH_MODULE_SFF_8472;
5064                 modinfo->eeprom_len = RTE_ETH_MODULE_SFF_8472_LEN;
5065         }
5066
5067         return 0;
5068 }
5069
5070 static int
5071 eth_igb_get_module_eeprom(struct rte_eth_dev *dev,
5072                           struct rte_dev_eeprom_info *info)
5073 {
5074         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5075
5076         uint32_t status = 0;
5077         uint16_t dataword[RTE_ETH_MODULE_SFF_8472_LEN / 2 + 1];
5078         u16 first_word, last_word;
5079         int i = 0;
5080
5081         first_word = info->offset >> 1;
5082         last_word = (info->offset + info->length - 1) >> 1;
5083
5084         /* Read EEPROM block, SFF-8079/SFF-8472, word at a time */
5085         for (i = 0; i < last_word - first_word + 1; i++) {
5086                 status = e1000_read_phy_reg_i2c(hw, (first_word + i) * 2,
5087                                                 &dataword[i]);
5088                 if (status) {
5089                         /* Error occurred while reading module */
5090                         return -EIO;
5091                 }
5092
5093                 dataword[i] = rte_be_to_cpu_16(dataword[i]);
5094         }
5095
5096         memcpy(info->data, (u8 *)dataword + (info->offset & 1), info->length);
5097
5098         return 0;
5099 }
5100
5101 static int
5102 eth_igb_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
5103 {
5104         struct e1000_hw *hw =
5105                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5106         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
5107         struct rte_intr_handle *intr_handle = pci_dev->intr_handle;
5108         uint32_t vec = E1000_MISC_VEC_ID;
5109
5110         if (rte_intr_allow_others(intr_handle))
5111                 vec = E1000_RX_VEC_START;
5112
5113         uint32_t mask = 1 << (queue_id + vec);
5114
5115         E1000_WRITE_REG(hw, E1000_EIMC, mask);
5116         E1000_WRITE_FLUSH(hw);
5117
5118         return 0;
5119 }
5120
5121 static int
5122 eth_igb_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
5123 {
5124         struct e1000_hw *hw =
5125                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5126         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
5127         struct rte_intr_handle *intr_handle = pci_dev->intr_handle;
5128         uint32_t vec = E1000_MISC_VEC_ID;
5129
5130         if (rte_intr_allow_others(intr_handle))
5131                 vec = E1000_RX_VEC_START;
5132
5133         uint32_t mask = 1 << (queue_id + vec);
5134         uint32_t regval;
5135
5136         regval = E1000_READ_REG(hw, E1000_EIMS);
5137         E1000_WRITE_REG(hw, E1000_EIMS, regval | mask);
5138         E1000_WRITE_FLUSH(hw);
5139
5140         rte_intr_ack(intr_handle);
5141
5142         return 0;
5143 }
5144
5145 static void
5146 eth_igb_write_ivar(struct e1000_hw *hw, uint8_t  msix_vector,
5147                    uint8_t index, uint8_t offset)
5148 {
5149         uint32_t val = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
5150
5151         /* clear bits */
5152         val &= ~((uint32_t)0xFF << offset);
5153
5154         /* write vector and valid bit */
5155         val |= (msix_vector | E1000_IVAR_VALID) << offset;
5156
5157         E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, val);
5158 }
5159
5160 static void
5161 eth_igb_assign_msix_vector(struct e1000_hw *hw, int8_t direction,
5162                            uint8_t queue, uint8_t msix_vector)
5163 {
5164         uint32_t tmp = 0;
5165
5166         if (hw->mac.type == e1000_82575) {
5167                 if (direction == 0)
5168                         tmp = E1000_EICR_RX_QUEUE0 << queue;
5169                 else if (direction == 1)
5170                         tmp = E1000_EICR_TX_QUEUE0 << queue;
5171                 E1000_WRITE_REG(hw, E1000_MSIXBM(msix_vector), tmp);
5172         } else if (hw->mac.type == e1000_82576) {
5173                 if ((direction == 0) || (direction == 1))
5174                         eth_igb_write_ivar(hw, msix_vector, queue & 0x7,
5175                                            ((queue & 0x8) << 1) +
5176                                            8 * direction);
5177         } else if ((hw->mac.type == e1000_82580) ||
5178                         (hw->mac.type == e1000_i350) ||
5179                         (hw->mac.type == e1000_i354) ||
5180                         (hw->mac.type == e1000_i210) ||
5181                         (hw->mac.type == e1000_i211)) {
5182                 if ((direction == 0) || (direction == 1))
5183                         eth_igb_write_ivar(hw, msix_vector,
5184                                            queue >> 1,
5185                                            ((queue & 0x1) << 4) +
5186                                            8 * direction);
5187         }
5188 }
5189
5190 /* Sets up the hardware to generate MSI-X interrupts properly
5191  * @hw
5192  *  board private structure
5193  */
5194 static void
5195 eth_igb_configure_msix_intr(struct rte_eth_dev *dev)
5196 {
5197         int queue_id;
5198         uint32_t tmpval, regval, intr_mask;
5199         struct e1000_hw *hw =
5200                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5201         uint32_t vec = E1000_MISC_VEC_ID;
5202         uint32_t base = E1000_MISC_VEC_ID;
5203         uint32_t misc_shift = 0;
5204         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
5205         struct rte_intr_handle *intr_handle = pci_dev->intr_handle;
5206
5207         /* won't configure msix register if no mapping is done
5208          * between intr vector and event fd
5209          */
5210         if (!rte_intr_dp_is_en(intr_handle))
5211                 return;
5212
5213         if (rte_intr_allow_others(intr_handle)) {
5214                 vec = base = E1000_RX_VEC_START;
5215                 misc_shift = 1;
5216         }
5217
5218         /* set interrupt vector for other causes */
5219         if (hw->mac.type == e1000_82575) {
5220                 tmpval = E1000_READ_REG(hw, E1000_CTRL_EXT);
5221                 /* enable MSI-X PBA support */
5222                 tmpval |= E1000_CTRL_EXT_PBA_CLR;
5223
5224                 /* Auto-Mask interrupts upon ICR read */
5225                 tmpval |= E1000_CTRL_EXT_EIAME;
5226                 tmpval |= E1000_CTRL_EXT_IRCA;
5227
5228                 E1000_WRITE_REG(hw, E1000_CTRL_EXT, tmpval);
5229
5230                 /* enable msix_other interrupt */
5231                 E1000_WRITE_REG_ARRAY(hw, E1000_MSIXBM(0), 0, E1000_EIMS_OTHER);
5232                 regval = E1000_READ_REG(hw, E1000_EIAC);
5233                 E1000_WRITE_REG(hw, E1000_EIAC, regval | E1000_EIMS_OTHER);
5234                 regval = E1000_READ_REG(hw, E1000_EIAM);
5235                 E1000_WRITE_REG(hw, E1000_EIMS, regval | E1000_EIMS_OTHER);
5236         } else if ((hw->mac.type == e1000_82576) ||
5237                         (hw->mac.type == e1000_82580) ||
5238                         (hw->mac.type == e1000_i350) ||
5239                         (hw->mac.type == e1000_i354) ||
5240                         (hw->mac.type == e1000_i210) ||
5241                         (hw->mac.type == e1000_i211)) {
5242                 /* turn on MSI-X capability first */
5243                 E1000_WRITE_REG(hw, E1000_GPIE, E1000_GPIE_MSIX_MODE |
5244                                         E1000_GPIE_PBA | E1000_GPIE_EIAME |
5245                                         E1000_GPIE_NSICR);
5246                 intr_mask =
5247                         RTE_LEN2MASK(rte_intr_nb_efd_get(intr_handle),
5248                                      uint32_t) << misc_shift;
5249
5250                 if (dev->data->dev_conf.intr_conf.lsc != 0)
5251                         intr_mask |= (1 << IGB_MSIX_OTHER_INTR_VEC);
5252
5253                 regval = E1000_READ_REG(hw, E1000_EIAC);
5254                 E1000_WRITE_REG(hw, E1000_EIAC, regval | intr_mask);
5255
5256                 /* enable msix_other interrupt */
5257                 regval = E1000_READ_REG(hw, E1000_EIMS);
5258                 E1000_WRITE_REG(hw, E1000_EIMS, regval | intr_mask);
5259                 tmpval = (IGB_MSIX_OTHER_INTR_VEC | E1000_IVAR_VALID) << 8;
5260                 E1000_WRITE_REG(hw, E1000_IVAR_MISC, tmpval);
5261         }
5262
5263         /* use EIAM to auto-mask when MSI-X interrupt
5264          * is asserted, this saves a register write for every interrupt
5265          */
5266         intr_mask = RTE_LEN2MASK(rte_intr_nb_efd_get(intr_handle),
5267                                  uint32_t) << misc_shift;
5268
5269         if (dev->data->dev_conf.intr_conf.lsc != 0)
5270                 intr_mask |= (1 << IGB_MSIX_OTHER_INTR_VEC);
5271
5272         regval = E1000_READ_REG(hw, E1000_EIAM);
5273         E1000_WRITE_REG(hw, E1000_EIAM, regval | intr_mask);
5274
5275         for (queue_id = 0; queue_id < dev->data->nb_rx_queues; queue_id++) {
5276                 eth_igb_assign_msix_vector(hw, 0, queue_id, vec);
5277                 rte_intr_vec_list_index_set(intr_handle, queue_id, vec);
5278                 if (vec < base + rte_intr_nb_efd_get(intr_handle) - 1)
5279                         vec++;
5280         }
5281
5282         E1000_WRITE_FLUSH(hw);
5283 }
5284
5285 /* restore n-tuple filter */
5286 static inline void
5287 igb_ntuple_filter_restore(struct rte_eth_dev *dev)
5288 {
5289         struct e1000_filter_info *filter_info =
5290                 E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
5291         struct e1000_5tuple_filter *p_5tuple;
5292         struct e1000_2tuple_filter *p_2tuple;
5293
5294         TAILQ_FOREACH(p_5tuple, &filter_info->fivetuple_list, entries) {
5295                 igb_inject_5tuple_filter_82576(dev, p_5tuple);
5296         }
5297
5298         TAILQ_FOREACH(p_2tuple, &filter_info->twotuple_list, entries) {
5299                 igb_inject_2uple_filter(dev, p_2tuple);
5300         }
5301 }
5302
5303 /* restore SYN filter */
5304 static inline void
5305 igb_syn_filter_restore(struct rte_eth_dev *dev)
5306 {
5307         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5308         struct e1000_filter_info *filter_info =
5309                 E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
5310         uint32_t synqf;
5311
5312         synqf = filter_info->syn_info;
5313
5314         if (synqf & E1000_SYN_FILTER_ENABLE) {
5315                 E1000_WRITE_REG(hw, E1000_SYNQF(0), synqf);
5316                 E1000_WRITE_FLUSH(hw);
5317         }
5318 }
5319
5320 /* restore ethernet type filter */
5321 static inline void
5322 igb_ethertype_filter_restore(struct rte_eth_dev *dev)
5323 {
5324         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5325         struct e1000_filter_info *filter_info =
5326                 E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
5327         int i;
5328
5329         for (i = 0; i < E1000_MAX_ETQF_FILTERS; i++) {
5330                 if (filter_info->ethertype_mask & (1 << i)) {
5331                         E1000_WRITE_REG(hw, E1000_ETQF(i),
5332                                 filter_info->ethertype_filters[i].etqf);
5333                         E1000_WRITE_FLUSH(hw);
5334                 }
5335         }
5336 }
5337
5338 /* restore flex byte filter */
5339 static inline void
5340 igb_flex_filter_restore(struct rte_eth_dev *dev)
5341 {
5342         struct e1000_filter_info *filter_info =
5343                 E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
5344         struct e1000_flex_filter *flex_filter;
5345
5346         TAILQ_FOREACH(flex_filter, &filter_info->flex_list, entries) {
5347                 igb_inject_flex_filter(dev, flex_filter);
5348         }
5349 }
5350
5351 /* restore rss filter */
5352 static inline void
5353 igb_rss_filter_restore(struct rte_eth_dev *dev)
5354 {
5355         struct e1000_filter_info *filter_info =
5356                 E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
5357
5358         if (filter_info->rss_info.conf.queue_num)
5359                 igb_config_rss_filter(dev, &filter_info->rss_info, TRUE);
5360 }
5361
5362 /* restore all types filter */
5363 static int
5364 igb_filter_restore(struct rte_eth_dev *dev)
5365 {
5366         igb_ntuple_filter_restore(dev);
5367         igb_ethertype_filter_restore(dev);
5368         igb_syn_filter_restore(dev);
5369         igb_flex_filter_restore(dev);
5370         igb_rss_filter_restore(dev);
5371
5372         return 0;
5373 }
5374
5375 RTE_PMD_REGISTER_PCI(net_e1000_igb, rte_igb_pmd);
5376 RTE_PMD_REGISTER_PCI_TABLE(net_e1000_igb, pci_id_igb_map);
5377 RTE_PMD_REGISTER_KMOD_DEP(net_e1000_igb, "* igb_uio | uio_pci_generic | vfio-pci");
5378 RTE_PMD_REGISTER_PCI(net_e1000_igb_vf, rte_igbvf_pmd);
5379 RTE_PMD_REGISTER_PCI_TABLE(net_e1000_igb_vf, pci_id_igbvf_map);
5380 RTE_PMD_REGISTER_KMOD_DEP(net_e1000_igb_vf, "* igb_uio | vfio-pci");