net/i40e: fix setting of MAC address on i40evf
[dpdk.git] / drivers / net / i40e / i40e_ethdev_vf.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 <string.h>
10 #include <unistd.h>
11 #include <stdarg.h>
12 #include <inttypes.h>
13 #include <rte_byteorder.h>
14 #include <rte_common.h>
15 #include <rte_cycles.h>
16
17 #include <rte_interrupts.h>
18 #include <rte_log.h>
19 #include <rte_debug.h>
20 #include <rte_pci.h>
21 #include <rte_bus_pci.h>
22 #include <rte_atomic.h>
23 #include <rte_branch_prediction.h>
24 #include <rte_memory.h>
25 #include <rte_eal.h>
26 #include <rte_alarm.h>
27 #include <rte_ether.h>
28 #include <rte_ethdev.h>
29 #include <rte_ethdev_pci.h>
30 #include <rte_malloc.h>
31 #include <rte_dev.h>
32
33 #include "i40e_logs.h"
34 #include "base/i40e_prototype.h"
35 #include "base/i40e_adminq_cmd.h"
36 #include "base/i40e_type.h"
37
38 #include "i40e_rxtx.h"
39 #include "i40e_ethdev.h"
40 #include "i40e_pf.h"
41
42 /* busy wait delay in msec */
43 #define I40EVF_BUSY_WAIT_DELAY 10
44 #define I40EVF_BUSY_WAIT_COUNT 50
45 #define MAX_RESET_WAIT_CNT     20
46
47 struct i40evf_arq_msg_info {
48         enum virtchnl_ops ops;
49         enum i40e_status_code result;
50         uint16_t buf_len;
51         uint16_t msg_len;
52         uint8_t *msg;
53 };
54
55 struct vf_cmd_info {
56         enum virtchnl_ops ops;
57         uint8_t *in_args;
58         uint32_t in_args_size;
59         uint8_t *out_buffer;
60         /* Input & output type. pass in buffer size and pass out
61          * actual return result
62          */
63         uint32_t out_size;
64 };
65
66 enum i40evf_aq_result {
67         I40EVF_MSG_ERR = -1, /* Meet error when accessing admin queue */
68         I40EVF_MSG_NON,      /* Read nothing from admin queue */
69         I40EVF_MSG_SYS,      /* Read system msg from admin queue */
70         I40EVF_MSG_CMD,      /* Read async command result */
71 };
72
73 static int i40evf_dev_configure(struct rte_eth_dev *dev);
74 static int i40evf_dev_start(struct rte_eth_dev *dev);
75 static void i40evf_dev_stop(struct rte_eth_dev *dev);
76 static void i40evf_dev_info_get(struct rte_eth_dev *dev,
77                                 struct rte_eth_dev_info *dev_info);
78 static int i40evf_dev_link_update(struct rte_eth_dev *dev,
79                                   int wait_to_complete);
80 static int i40evf_dev_stats_get(struct rte_eth_dev *dev,
81                                 struct rte_eth_stats *stats);
82 static int i40evf_dev_xstats_get(struct rte_eth_dev *dev,
83                                  struct rte_eth_xstat *xstats, unsigned n);
84 static int i40evf_dev_xstats_get_names(struct rte_eth_dev *dev,
85                                        struct rte_eth_xstat_name *xstats_names,
86                                        unsigned limit);
87 static void i40evf_dev_xstats_reset(struct rte_eth_dev *dev);
88 static int i40evf_vlan_filter_set(struct rte_eth_dev *dev,
89                                   uint16_t vlan_id, int on);
90 static int i40evf_vlan_offload_set(struct rte_eth_dev *dev, int mask);
91 static void i40evf_dev_close(struct rte_eth_dev *dev);
92 static int  i40evf_dev_reset(struct rte_eth_dev *dev);
93 static void i40evf_dev_promiscuous_enable(struct rte_eth_dev *dev);
94 static void i40evf_dev_promiscuous_disable(struct rte_eth_dev *dev);
95 static void i40evf_dev_allmulticast_enable(struct rte_eth_dev *dev);
96 static void i40evf_dev_allmulticast_disable(struct rte_eth_dev *dev);
97 static int i40evf_init_vlan(struct rte_eth_dev *dev);
98 static int i40evf_dev_rx_queue_start(struct rte_eth_dev *dev,
99                                      uint16_t rx_queue_id);
100 static int i40evf_dev_rx_queue_stop(struct rte_eth_dev *dev,
101                                     uint16_t rx_queue_id);
102 static int i40evf_dev_tx_queue_start(struct rte_eth_dev *dev,
103                                      uint16_t tx_queue_id);
104 static int i40evf_dev_tx_queue_stop(struct rte_eth_dev *dev,
105                                     uint16_t tx_queue_id);
106 static int i40evf_add_mac_addr(struct rte_eth_dev *dev,
107                                struct ether_addr *addr,
108                                uint32_t index,
109                                uint32_t pool);
110 static void i40evf_del_mac_addr(struct rte_eth_dev *dev, uint32_t index);
111 static int i40evf_dev_rss_reta_update(struct rte_eth_dev *dev,
112                         struct rte_eth_rss_reta_entry64 *reta_conf,
113                         uint16_t reta_size);
114 static int i40evf_dev_rss_reta_query(struct rte_eth_dev *dev,
115                         struct rte_eth_rss_reta_entry64 *reta_conf,
116                         uint16_t reta_size);
117 static int i40evf_config_rss(struct i40e_vf *vf);
118 static int i40evf_dev_rss_hash_update(struct rte_eth_dev *dev,
119                                       struct rte_eth_rss_conf *rss_conf);
120 static int i40evf_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
121                                         struct rte_eth_rss_conf *rss_conf);
122 static int i40evf_dev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu);
123 static void i40evf_set_default_mac_addr(struct rte_eth_dev *dev,
124                                         struct ether_addr *mac_addr);
125 static int
126 i40evf_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id);
127 static int
128 i40evf_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id);
129 static void i40evf_handle_pf_event(struct rte_eth_dev *dev,
130                                    uint8_t *msg,
131                                    uint16_t msglen);
132
133 /* Default hash key buffer for RSS */
134 static uint32_t rss_key_default[I40E_VFQF_HKEY_MAX_INDEX + 1];
135
136 struct rte_i40evf_xstats_name_off {
137         char name[RTE_ETH_XSTATS_NAME_SIZE];
138         unsigned offset;
139 };
140
141 static const struct rte_i40evf_xstats_name_off rte_i40evf_stats_strings[] = {
142         {"rx_bytes", offsetof(struct i40e_eth_stats, rx_bytes)},
143         {"rx_unicast_packets", offsetof(struct i40e_eth_stats, rx_unicast)},
144         {"rx_multicast_packets", offsetof(struct i40e_eth_stats, rx_multicast)},
145         {"rx_broadcast_packets", offsetof(struct i40e_eth_stats, rx_broadcast)},
146         {"rx_dropped_packets", offsetof(struct i40e_eth_stats, rx_discards)},
147         {"rx_unknown_protocol_packets", offsetof(struct i40e_eth_stats,
148                 rx_unknown_protocol)},
149         {"tx_bytes", offsetof(struct i40e_eth_stats, tx_bytes)},
150         {"tx_unicast_packets", offsetof(struct i40e_eth_stats, tx_unicast)},
151         {"tx_multicast_packets", offsetof(struct i40e_eth_stats, tx_multicast)},
152         {"tx_broadcast_packets", offsetof(struct i40e_eth_stats, tx_broadcast)},
153         {"tx_dropped_packets", offsetof(struct i40e_eth_stats, tx_discards)},
154         {"tx_error_packets", offsetof(struct i40e_eth_stats, tx_errors)},
155 };
156
157 #define I40EVF_NB_XSTATS (sizeof(rte_i40evf_stats_strings) / \
158                 sizeof(rte_i40evf_stats_strings[0]))
159
160 static const struct eth_dev_ops i40evf_eth_dev_ops = {
161         .dev_configure        = i40evf_dev_configure,
162         .dev_start            = i40evf_dev_start,
163         .dev_stop             = i40evf_dev_stop,
164         .promiscuous_enable   = i40evf_dev_promiscuous_enable,
165         .promiscuous_disable  = i40evf_dev_promiscuous_disable,
166         .allmulticast_enable  = i40evf_dev_allmulticast_enable,
167         .allmulticast_disable = i40evf_dev_allmulticast_disable,
168         .link_update          = i40evf_dev_link_update,
169         .stats_get            = i40evf_dev_stats_get,
170         .stats_reset          = i40evf_dev_xstats_reset,
171         .xstats_get           = i40evf_dev_xstats_get,
172         .xstats_get_names     = i40evf_dev_xstats_get_names,
173         .xstats_reset         = i40evf_dev_xstats_reset,
174         .dev_close            = i40evf_dev_close,
175         .dev_reset            = i40evf_dev_reset,
176         .dev_infos_get        = i40evf_dev_info_get,
177         .dev_supported_ptypes_get = i40e_dev_supported_ptypes_get,
178         .vlan_filter_set      = i40evf_vlan_filter_set,
179         .vlan_offload_set     = i40evf_vlan_offload_set,
180         .rx_queue_start       = i40evf_dev_rx_queue_start,
181         .rx_queue_stop        = i40evf_dev_rx_queue_stop,
182         .tx_queue_start       = i40evf_dev_tx_queue_start,
183         .tx_queue_stop        = i40evf_dev_tx_queue_stop,
184         .rx_queue_setup       = i40e_dev_rx_queue_setup,
185         .rx_queue_release     = i40e_dev_rx_queue_release,
186         .rx_queue_intr_enable = i40evf_dev_rx_queue_intr_enable,
187         .rx_queue_intr_disable = i40evf_dev_rx_queue_intr_disable,
188         .rx_descriptor_done   = i40e_dev_rx_descriptor_done,
189         .rx_descriptor_status = i40e_dev_rx_descriptor_status,
190         .tx_descriptor_status = i40e_dev_tx_descriptor_status,
191         .tx_queue_setup       = i40e_dev_tx_queue_setup,
192         .tx_queue_release     = i40e_dev_tx_queue_release,
193         .rx_queue_count       = i40e_dev_rx_queue_count,
194         .rxq_info_get         = i40e_rxq_info_get,
195         .txq_info_get         = i40e_txq_info_get,
196         .mac_addr_add         = i40evf_add_mac_addr,
197         .mac_addr_remove      = i40evf_del_mac_addr,
198         .reta_update          = i40evf_dev_rss_reta_update,
199         .reta_query           = i40evf_dev_rss_reta_query,
200         .rss_hash_update      = i40evf_dev_rss_hash_update,
201         .rss_hash_conf_get    = i40evf_dev_rss_hash_conf_get,
202         .mtu_set              = i40evf_dev_mtu_set,
203         .mac_addr_set         = i40evf_set_default_mac_addr,
204 };
205
206 /*
207  * Read data in admin queue to get msg from pf driver
208  */
209 static enum i40evf_aq_result
210 i40evf_read_pfmsg(struct rte_eth_dev *dev, struct i40evf_arq_msg_info *data)
211 {
212         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
213         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
214         struct i40e_arq_event_info event;
215         enum virtchnl_ops opcode;
216         enum i40e_status_code retval;
217         int ret;
218         enum i40evf_aq_result result = I40EVF_MSG_NON;
219
220         event.buf_len = data->buf_len;
221         event.msg_buf = data->msg;
222         ret = i40e_clean_arq_element(hw, &event, NULL);
223         /* Can't read any msg from adminQ */
224         if (ret) {
225                 if (ret != I40E_ERR_ADMIN_QUEUE_NO_WORK)
226                         result = I40EVF_MSG_ERR;
227                 return result;
228         }
229
230         opcode = (enum virtchnl_ops)rte_le_to_cpu_32(event.desc.cookie_high);
231         retval = (enum i40e_status_code)rte_le_to_cpu_32(event.desc.cookie_low);
232         /* pf sys event */
233         if (opcode == VIRTCHNL_OP_EVENT) {
234                 struct virtchnl_pf_event *vpe =
235                         (struct virtchnl_pf_event *)event.msg_buf;
236
237                 result = I40EVF_MSG_SYS;
238                 switch (vpe->event) {
239                 case VIRTCHNL_EVENT_LINK_CHANGE:
240                         vf->link_up =
241                                 vpe->event_data.link_event.link_status;
242                         vf->link_speed =
243                                 vpe->event_data.link_event.link_speed;
244                         vf->pend_msg |= PFMSG_LINK_CHANGE;
245                         PMD_DRV_LOG(INFO, "Link status update:%s",
246                                     vf->link_up ? "up" : "down");
247                         break;
248                 case VIRTCHNL_EVENT_RESET_IMPENDING:
249                         vf->vf_reset = true;
250                         vf->pend_msg |= PFMSG_RESET_IMPENDING;
251                         PMD_DRV_LOG(INFO, "vf is reseting");
252                         break;
253                 case VIRTCHNL_EVENT_PF_DRIVER_CLOSE:
254                         vf->dev_closed = true;
255                         vf->pend_msg |= PFMSG_DRIVER_CLOSE;
256                         PMD_DRV_LOG(INFO, "PF driver closed");
257                         break;
258                 default:
259                         PMD_DRV_LOG(ERR, "%s: Unknown event %d from pf",
260                                     __func__, vpe->event);
261                 }
262         } else {
263                 /* async reply msg on command issued by vf previously */
264                 result = I40EVF_MSG_CMD;
265                 /* Actual data length read from PF */
266                 data->msg_len = event.msg_len;
267         }
268
269         data->result = retval;
270         data->ops = opcode;
271
272         return result;
273 }
274
275 /**
276  * clear current command. Only call in case execute
277  * _atomic_set_cmd successfully.
278  */
279 static inline void
280 _clear_cmd(struct i40e_vf *vf)
281 {
282         rte_wmb();
283         vf->pend_cmd = VIRTCHNL_OP_UNKNOWN;
284 }
285
286 /*
287  * Check there is pending cmd in execution. If none, set new command.
288  */
289 static inline int
290 _atomic_set_cmd(struct i40e_vf *vf, enum virtchnl_ops ops)
291 {
292         int ret = rte_atomic32_cmpset(&vf->pend_cmd,
293                         VIRTCHNL_OP_UNKNOWN, ops);
294
295         if (!ret)
296                 PMD_DRV_LOG(ERR, "There is incomplete cmd %d", vf->pend_cmd);
297
298         return !ret;
299 }
300
301 #define MAX_TRY_TIMES 200
302 #define ASQ_DELAY_MS  10
303
304 static int
305 i40evf_execute_vf_cmd(struct rte_eth_dev *dev, struct vf_cmd_info *args)
306 {
307         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
308         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
309         struct i40evf_arq_msg_info info;
310         enum i40evf_aq_result ret;
311         int err, i = 0;
312
313         if (_atomic_set_cmd(vf, args->ops))
314                 return -1;
315
316         info.msg = args->out_buffer;
317         info.buf_len = args->out_size;
318         info.ops = VIRTCHNL_OP_UNKNOWN;
319         info.result = I40E_SUCCESS;
320
321         err = i40e_aq_send_msg_to_pf(hw, args->ops, I40E_SUCCESS,
322                      args->in_args, args->in_args_size, NULL);
323         if (err) {
324                 PMD_DRV_LOG(ERR, "fail to send cmd %d", args->ops);
325                 _clear_cmd(vf);
326                 return err;
327         }
328
329         switch (args->ops) {
330         case VIRTCHNL_OP_RESET_VF:
331                 /*no need to process in this function */
332                 err = 0;
333                 break;
334         case VIRTCHNL_OP_VERSION:
335         case VIRTCHNL_OP_GET_VF_RESOURCES:
336                 /* for init adminq commands, need to poll the response */
337                 err = -1;
338                 do {
339                         ret = i40evf_read_pfmsg(dev, &info);
340                         vf->cmd_retval = info.result;
341                         if (ret == I40EVF_MSG_CMD) {
342                                 err = 0;
343                                 break;
344                         } else if (ret == I40EVF_MSG_ERR)
345                                 break;
346                         rte_delay_ms(ASQ_DELAY_MS);
347                         /* If don't read msg or read sys event, continue */
348                 } while (i++ < MAX_TRY_TIMES);
349                 _clear_cmd(vf);
350                 break;
351
352         default:
353                 /* for other adminq in running time, waiting the cmd done flag */
354                 err = -1;
355                 do {
356                         if (vf->pend_cmd == VIRTCHNL_OP_UNKNOWN) {
357                                 err = 0;
358                                 break;
359                         }
360                         rte_delay_ms(ASQ_DELAY_MS);
361                         /* If don't read msg or read sys event, continue */
362                 } while (i++ < MAX_TRY_TIMES);
363                 /* If there's no response is received, clear command */
364                 if (i >= MAX_TRY_TIMES) {
365                         PMD_DRV_LOG(WARNING, "No response for %d", args->ops);
366                         _clear_cmd(vf);
367                 }
368                 break;
369         }
370
371         return err | vf->cmd_retval;
372 }
373
374 /*
375  * Check API version with sync wait until version read or fail from admin queue
376  */
377 static int
378 i40evf_check_api_version(struct rte_eth_dev *dev)
379 {
380         struct virtchnl_version_info version, *pver;
381         int err;
382         struct vf_cmd_info args;
383         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
384
385         version.major = VIRTCHNL_VERSION_MAJOR;
386         version.minor = VIRTCHNL_VERSION_MINOR;
387
388         args.ops = VIRTCHNL_OP_VERSION;
389         args.in_args = (uint8_t *)&version;
390         args.in_args_size = sizeof(version);
391         args.out_buffer = vf->aq_resp;
392         args.out_size = I40E_AQ_BUF_SZ;
393
394         err = i40evf_execute_vf_cmd(dev, &args);
395         if (err) {
396                 PMD_INIT_LOG(ERR, "fail to execute command OP_VERSION");
397                 return err;
398         }
399
400         pver = (struct virtchnl_version_info *)args.out_buffer;
401         vf->version_major = pver->major;
402         vf->version_minor = pver->minor;
403         if ((vf->version_major == VIRTCHNL_VERSION_MAJOR) &&
404                 (vf->version_minor <= VIRTCHNL_VERSION_MINOR))
405                 PMD_DRV_LOG(INFO, "Peer is Linux PF host");
406         else {
407                 PMD_INIT_LOG(ERR, "PF/VF API version mismatch:(%u.%u)-(%u.%u)",
408                                         vf->version_major, vf->version_minor,
409                                                 VIRTCHNL_VERSION_MAJOR,
410                                                 VIRTCHNL_VERSION_MINOR);
411                 return -1;
412         }
413
414         return 0;
415 }
416
417 static int
418 i40evf_get_vf_resource(struct rte_eth_dev *dev)
419 {
420         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
421         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
422         int err;
423         struct vf_cmd_info args;
424         uint32_t caps, len;
425
426         args.ops = VIRTCHNL_OP_GET_VF_RESOURCES;
427         args.out_buffer = vf->aq_resp;
428         args.out_size = I40E_AQ_BUF_SZ;
429         if (PF_IS_V11(vf)) {
430                 caps = VIRTCHNL_VF_OFFLOAD_L2 |
431                        VIRTCHNL_VF_OFFLOAD_RSS_AQ |
432                        VIRTCHNL_VF_OFFLOAD_RSS_REG |
433                        VIRTCHNL_VF_OFFLOAD_VLAN |
434                        VIRTCHNL_VF_OFFLOAD_RX_POLLING;
435                 args.in_args = (uint8_t *)&caps;
436                 args.in_args_size = sizeof(caps);
437         } else {
438                 args.in_args = NULL;
439                 args.in_args_size = 0;
440         }
441         err = i40evf_execute_vf_cmd(dev, &args);
442
443         if (err) {
444                 PMD_DRV_LOG(ERR, "fail to execute command OP_GET_VF_RESOURCE");
445                 return err;
446         }
447
448         len =  sizeof(struct virtchnl_vf_resource) +
449                 I40E_MAX_VF_VSI * sizeof(struct virtchnl_vsi_resource);
450
451         rte_memcpy(vf->vf_res, args.out_buffer,
452                         RTE_MIN(args.out_size, len));
453         i40e_vf_parse_hw_config(hw, vf->vf_res);
454
455         return 0;
456 }
457
458 static int
459 i40evf_config_promisc(struct rte_eth_dev *dev,
460                       bool enable_unicast,
461                       bool enable_multicast)
462 {
463         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
464         int err;
465         struct vf_cmd_info args;
466         struct virtchnl_promisc_info promisc;
467
468         promisc.flags = 0;
469         promisc.vsi_id = vf->vsi_res->vsi_id;
470
471         if (enable_unicast)
472                 promisc.flags |= FLAG_VF_UNICAST_PROMISC;
473
474         if (enable_multicast)
475                 promisc.flags |= FLAG_VF_MULTICAST_PROMISC;
476
477         args.ops = VIRTCHNL_OP_CONFIG_PROMISCUOUS_MODE;
478         args.in_args = (uint8_t *)&promisc;
479         args.in_args_size = sizeof(promisc);
480         args.out_buffer = vf->aq_resp;
481         args.out_size = I40E_AQ_BUF_SZ;
482
483         err = i40evf_execute_vf_cmd(dev, &args);
484
485         if (err)
486                 PMD_DRV_LOG(ERR, "fail to execute command "
487                             "CONFIG_PROMISCUOUS_MODE");
488         return err;
489 }
490
491 static int
492 i40evf_enable_vlan_strip(struct rte_eth_dev *dev)
493 {
494         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
495         struct vf_cmd_info args;
496         int ret;
497
498         memset(&args, 0, sizeof(args));
499         args.ops = VIRTCHNL_OP_ENABLE_VLAN_STRIPPING;
500         args.in_args = NULL;
501         args.in_args_size = 0;
502         args.out_buffer = vf->aq_resp;
503         args.out_size = I40E_AQ_BUF_SZ;
504         ret = i40evf_execute_vf_cmd(dev, &args);
505         if (ret)
506                 PMD_DRV_LOG(ERR, "Failed to execute command of "
507                             "VIRTCHNL_OP_ENABLE_VLAN_STRIPPING");
508
509         return ret;
510 }
511
512 static int
513 i40evf_disable_vlan_strip(struct rte_eth_dev *dev)
514 {
515         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
516         struct vf_cmd_info args;
517         int ret;
518
519         memset(&args, 0, sizeof(args));
520         args.ops = VIRTCHNL_OP_DISABLE_VLAN_STRIPPING;
521         args.in_args = NULL;
522         args.in_args_size = 0;
523         args.out_buffer = vf->aq_resp;
524         args.out_size = I40E_AQ_BUF_SZ;
525         ret = i40evf_execute_vf_cmd(dev, &args);
526         if (ret)
527                 PMD_DRV_LOG(ERR, "Failed to execute command of "
528                             "VIRTCHNL_OP_DISABLE_VLAN_STRIPPING");
529
530         return ret;
531 }
532
533 static void
534 i40evf_fill_virtchnl_vsi_txq_info(struct virtchnl_txq_info *txq_info,
535                                   uint16_t vsi_id,
536                                   uint16_t queue_id,
537                                   uint16_t nb_txq,
538                                   struct i40e_tx_queue *txq)
539 {
540         txq_info->vsi_id = vsi_id;
541         txq_info->queue_id = queue_id;
542         if (queue_id < nb_txq) {
543                 txq_info->ring_len = txq->nb_tx_desc;
544                 txq_info->dma_ring_addr = txq->tx_ring_phys_addr;
545         }
546 }
547
548 static void
549 i40evf_fill_virtchnl_vsi_rxq_info(struct virtchnl_rxq_info *rxq_info,
550                                   uint16_t vsi_id,
551                                   uint16_t queue_id,
552                                   uint16_t nb_rxq,
553                                   uint32_t max_pkt_size,
554                                   struct i40e_rx_queue *rxq)
555 {
556         rxq_info->vsi_id = vsi_id;
557         rxq_info->queue_id = queue_id;
558         rxq_info->max_pkt_size = max_pkt_size;
559         if (queue_id < nb_rxq) {
560                 rxq_info->ring_len = rxq->nb_rx_desc;
561                 rxq_info->dma_ring_addr = rxq->rx_ring_phys_addr;
562                 rxq_info->databuffer_size =
563                         (rte_pktmbuf_data_room_size(rxq->mp) -
564                                 RTE_PKTMBUF_HEADROOM);
565         }
566 }
567
568 static int
569 i40evf_configure_vsi_queues(struct rte_eth_dev *dev)
570 {
571         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
572         struct i40e_rx_queue **rxq =
573                 (struct i40e_rx_queue **)dev->data->rx_queues;
574         struct i40e_tx_queue **txq =
575                 (struct i40e_tx_queue **)dev->data->tx_queues;
576         struct virtchnl_vsi_queue_config_info *vc_vqci;
577         struct virtchnl_queue_pair_info *vc_qpi;
578         struct vf_cmd_info args;
579         uint16_t i, nb_qp = vf->num_queue_pairs;
580         const uint32_t size =
581                 I40E_VIRTCHNL_CONFIG_VSI_QUEUES_SIZE(vc_vqci, nb_qp);
582         uint8_t buff[size];
583         int ret;
584
585         memset(buff, 0, sizeof(buff));
586         vc_vqci = (struct virtchnl_vsi_queue_config_info *)buff;
587         vc_vqci->vsi_id = vf->vsi_res->vsi_id;
588         vc_vqci->num_queue_pairs = nb_qp;
589
590         for (i = 0, vc_qpi = vc_vqci->qpair; i < nb_qp; i++, vc_qpi++) {
591                 i40evf_fill_virtchnl_vsi_txq_info(&vc_qpi->txq,
592                         vc_vqci->vsi_id, i, dev->data->nb_tx_queues, txq[i]);
593                 i40evf_fill_virtchnl_vsi_rxq_info(&vc_qpi->rxq,
594                         vc_vqci->vsi_id, i, dev->data->nb_rx_queues,
595                                         vf->max_pkt_len, rxq[i]);
596         }
597         memset(&args, 0, sizeof(args));
598         args.ops = VIRTCHNL_OP_CONFIG_VSI_QUEUES;
599         args.in_args = (uint8_t *)vc_vqci;
600         args.in_args_size = size;
601         args.out_buffer = vf->aq_resp;
602         args.out_size = I40E_AQ_BUF_SZ;
603         ret = i40evf_execute_vf_cmd(dev, &args);
604         if (ret)
605                 PMD_DRV_LOG(ERR, "Failed to execute command of "
606                         "VIRTCHNL_OP_CONFIG_VSI_QUEUES");
607
608         return ret;
609 }
610
611 static int
612 i40evf_config_irq_map(struct rte_eth_dev *dev)
613 {
614         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
615         struct vf_cmd_info args;
616         uint8_t cmd_buffer[sizeof(struct virtchnl_irq_map_info) + \
617                 sizeof(struct virtchnl_vector_map)];
618         struct virtchnl_irq_map_info *map_info;
619         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
620         struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
621         uint32_t vector_id;
622         int i, err;
623
624         if (dev->data->dev_conf.intr_conf.rxq != 0 &&
625             rte_intr_allow_others(intr_handle))
626                 vector_id = I40E_RX_VEC_START;
627         else
628                 vector_id = I40E_MISC_VEC_ID;
629
630         map_info = (struct virtchnl_irq_map_info *)cmd_buffer;
631         map_info->num_vectors = 1;
632         map_info->vecmap[0].rxitr_idx = I40E_ITR_INDEX_DEFAULT;
633         map_info->vecmap[0].vsi_id = vf->vsi_res->vsi_id;
634         /* Alway use default dynamic MSIX interrupt */
635         map_info->vecmap[0].vector_id = vector_id;
636         /* Don't map any tx queue */
637         map_info->vecmap[0].txq_map = 0;
638         map_info->vecmap[0].rxq_map = 0;
639         for (i = 0; i < dev->data->nb_rx_queues; i++) {
640                 map_info->vecmap[0].rxq_map |= 1 << i;
641                 if (rte_intr_dp_is_en(intr_handle))
642                         intr_handle->intr_vec[i] = vector_id;
643         }
644
645         args.ops = VIRTCHNL_OP_CONFIG_IRQ_MAP;
646         args.in_args = (u8 *)cmd_buffer;
647         args.in_args_size = sizeof(cmd_buffer);
648         args.out_buffer = vf->aq_resp;
649         args.out_size = I40E_AQ_BUF_SZ;
650         err = i40evf_execute_vf_cmd(dev, &args);
651         if (err)
652                 PMD_DRV_LOG(ERR, "fail to execute command OP_ENABLE_QUEUES");
653
654         return err;
655 }
656
657 static int
658 i40evf_switch_queue(struct rte_eth_dev *dev, bool isrx, uint16_t qid,
659                                 bool on)
660 {
661         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
662         struct virtchnl_queue_select queue_select;
663         int err;
664         struct vf_cmd_info args;
665         memset(&queue_select, 0, sizeof(queue_select));
666         queue_select.vsi_id = vf->vsi_res->vsi_id;
667
668         if (isrx)
669                 queue_select.rx_queues |= 1 << qid;
670         else
671                 queue_select.tx_queues |= 1 << qid;
672
673         if (on)
674                 args.ops = VIRTCHNL_OP_ENABLE_QUEUES;
675         else
676                 args.ops = VIRTCHNL_OP_DISABLE_QUEUES;
677         args.in_args = (u8 *)&queue_select;
678         args.in_args_size = sizeof(queue_select);
679         args.out_buffer = vf->aq_resp;
680         args.out_size = I40E_AQ_BUF_SZ;
681         err = i40evf_execute_vf_cmd(dev, &args);
682         if (err)
683                 PMD_DRV_LOG(ERR, "fail to switch %s %u %s",
684                             isrx ? "RX" : "TX", qid, on ? "on" : "off");
685
686         return err;
687 }
688
689 static int
690 i40evf_start_queues(struct rte_eth_dev *dev)
691 {
692         struct rte_eth_dev_data *dev_data = dev->data;
693         int i;
694         struct i40e_rx_queue *rxq;
695         struct i40e_tx_queue *txq;
696
697         for (i = 0; i < dev->data->nb_rx_queues; i++) {
698                 rxq = dev_data->rx_queues[i];
699                 if (rxq->rx_deferred_start)
700                         continue;
701                 if (i40evf_dev_rx_queue_start(dev, i) != 0) {
702                         PMD_DRV_LOG(ERR, "Fail to start queue %u", i);
703                         return -1;
704                 }
705         }
706
707         for (i = 0; i < dev->data->nb_tx_queues; i++) {
708                 txq = dev_data->tx_queues[i];
709                 if (txq->tx_deferred_start)
710                         continue;
711                 if (i40evf_dev_tx_queue_start(dev, i) != 0) {
712                         PMD_DRV_LOG(ERR, "Fail to start queue %u", i);
713                         return -1;
714                 }
715         }
716
717         return 0;
718 }
719
720 static int
721 i40evf_stop_queues(struct rte_eth_dev *dev)
722 {
723         int i;
724
725         /* Stop TX queues first */
726         for (i = 0; i < dev->data->nb_tx_queues; i++) {
727                 if (i40evf_dev_tx_queue_stop(dev, i) != 0) {
728                         PMD_DRV_LOG(ERR, "Fail to stop queue %u", i);
729                         return -1;
730                 }
731         }
732
733         /* Then stop RX queues */
734         for (i = 0; i < dev->data->nb_rx_queues; i++) {
735                 if (i40evf_dev_rx_queue_stop(dev, i) != 0) {
736                         PMD_DRV_LOG(ERR, "Fail to stop queue %u", i);
737                         return -1;
738                 }
739         }
740
741         return 0;
742 }
743
744 static int
745 i40evf_add_mac_addr(struct rte_eth_dev *dev,
746                     struct ether_addr *addr,
747                     __rte_unused uint32_t index,
748                     __rte_unused uint32_t pool)
749 {
750         struct virtchnl_ether_addr_list *list;
751         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
752         uint8_t cmd_buffer[sizeof(struct virtchnl_ether_addr_list) + \
753                         sizeof(struct virtchnl_ether_addr)];
754         int err;
755         struct vf_cmd_info args;
756
757         if (is_zero_ether_addr(addr)) {
758                 PMD_DRV_LOG(ERR, "Invalid mac:%x:%x:%x:%x:%x:%x",
759                             addr->addr_bytes[0], addr->addr_bytes[1],
760                             addr->addr_bytes[2], addr->addr_bytes[3],
761                             addr->addr_bytes[4], addr->addr_bytes[5]);
762                 return I40E_ERR_INVALID_MAC_ADDR;
763         }
764
765         list = (struct virtchnl_ether_addr_list *)cmd_buffer;
766         list->vsi_id = vf->vsi_res->vsi_id;
767         list->num_elements = 1;
768         rte_memcpy(list->list[0].addr, addr->addr_bytes,
769                                         sizeof(addr->addr_bytes));
770
771         args.ops = VIRTCHNL_OP_ADD_ETH_ADDR;
772         args.in_args = cmd_buffer;
773         args.in_args_size = sizeof(cmd_buffer);
774         args.out_buffer = vf->aq_resp;
775         args.out_size = I40E_AQ_BUF_SZ;
776         err = i40evf_execute_vf_cmd(dev, &args);
777         if (err)
778                 PMD_DRV_LOG(ERR, "fail to execute command "
779                             "OP_ADD_ETHER_ADDRESS");
780         else
781                 vf->vsi.mac_num++;
782
783         return err;
784 }
785
786 static void
787 i40evf_del_mac_addr_by_addr(struct rte_eth_dev *dev,
788                             struct ether_addr *addr)
789 {
790         struct virtchnl_ether_addr_list *list;
791         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
792         uint8_t cmd_buffer[sizeof(struct virtchnl_ether_addr_list) + \
793                         sizeof(struct virtchnl_ether_addr)];
794         int err;
795         struct vf_cmd_info args;
796
797         if (i40e_validate_mac_addr(addr->addr_bytes) != I40E_SUCCESS) {
798                 PMD_DRV_LOG(ERR, "Invalid mac:%x-%x-%x-%x-%x-%x",
799                             addr->addr_bytes[0], addr->addr_bytes[1],
800                             addr->addr_bytes[2], addr->addr_bytes[3],
801                             addr->addr_bytes[4], addr->addr_bytes[5]);
802                 return;
803         }
804
805         list = (struct virtchnl_ether_addr_list *)cmd_buffer;
806         list->vsi_id = vf->vsi_res->vsi_id;
807         list->num_elements = 1;
808         rte_memcpy(list->list[0].addr, addr->addr_bytes,
809                         sizeof(addr->addr_bytes));
810
811         args.ops = VIRTCHNL_OP_DEL_ETH_ADDR;
812         args.in_args = cmd_buffer;
813         args.in_args_size = sizeof(cmd_buffer);
814         args.out_buffer = vf->aq_resp;
815         args.out_size = I40E_AQ_BUF_SZ;
816         err = i40evf_execute_vf_cmd(dev, &args);
817         if (err)
818                 PMD_DRV_LOG(ERR, "fail to execute command "
819                             "OP_DEL_ETHER_ADDRESS");
820         else
821                 vf->vsi.mac_num--;
822         return;
823 }
824
825 static void
826 i40evf_del_mac_addr(struct rte_eth_dev *dev, uint32_t index)
827 {
828         struct rte_eth_dev_data *data = dev->data;
829         struct ether_addr *addr;
830
831         addr = &data->mac_addrs[index];
832
833         i40evf_del_mac_addr_by_addr(dev, addr);
834 }
835
836 static int
837 i40evf_query_stats(struct rte_eth_dev *dev, struct i40e_eth_stats **pstats)
838 {
839         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
840         struct virtchnl_queue_select q_stats;
841         int err;
842         struct vf_cmd_info args;
843
844         memset(&q_stats, 0, sizeof(q_stats));
845         q_stats.vsi_id = vf->vsi_res->vsi_id;
846         args.ops = VIRTCHNL_OP_GET_STATS;
847         args.in_args = (u8 *)&q_stats;
848         args.in_args_size = sizeof(q_stats);
849         args.out_buffer = vf->aq_resp;
850         args.out_size = I40E_AQ_BUF_SZ;
851
852         err = i40evf_execute_vf_cmd(dev, &args);
853         if (err) {
854                 PMD_DRV_LOG(ERR, "fail to execute command OP_GET_STATS");
855                 *pstats = NULL;
856                 return err;
857         }
858         *pstats = (struct i40e_eth_stats *)args.out_buffer;
859         return 0;
860 }
861
862 static void
863 i40evf_stat_update_48(uint64_t *offset,
864                    uint64_t *stat)
865 {
866         if (*stat >= *offset)
867                 *stat = *stat - *offset;
868         else
869                 *stat = (uint64_t)((*stat +
870                         ((uint64_t)1 << I40E_48_BIT_WIDTH)) - *offset);
871
872         *stat &= I40E_48_BIT_MASK;
873 }
874
875 static void
876 i40evf_stat_update_32(uint64_t *offset,
877                    uint64_t *stat)
878 {
879         if (*stat >= *offset)
880                 *stat = (uint64_t)(*stat - *offset);
881         else
882                 *stat = (uint64_t)((*stat +
883                         ((uint64_t)1 << I40E_32_BIT_WIDTH)) - *offset);
884 }
885
886 static void
887 i40evf_update_stats(struct i40e_vsi *vsi,
888                                         struct i40e_eth_stats *nes)
889 {
890         struct i40e_eth_stats *oes = &vsi->eth_stats_offset;
891
892         i40evf_stat_update_48(&oes->rx_bytes,
893                             &nes->rx_bytes);
894         i40evf_stat_update_48(&oes->rx_unicast,
895                             &nes->rx_unicast);
896         i40evf_stat_update_48(&oes->rx_multicast,
897                             &nes->rx_multicast);
898         i40evf_stat_update_48(&oes->rx_broadcast,
899                             &nes->rx_broadcast);
900         i40evf_stat_update_32(&oes->rx_discards,
901                                 &nes->rx_discards);
902         i40evf_stat_update_32(&oes->rx_unknown_protocol,
903                             &nes->rx_unknown_protocol);
904         i40evf_stat_update_48(&oes->tx_bytes,
905                             &nes->tx_bytes);
906         i40evf_stat_update_48(&oes->tx_unicast,
907                             &nes->tx_unicast);
908         i40evf_stat_update_48(&oes->tx_multicast,
909                             &nes->tx_multicast);
910         i40evf_stat_update_48(&oes->tx_broadcast,
911                             &nes->tx_broadcast);
912         i40evf_stat_update_32(&oes->tx_errors, &nes->tx_errors);
913         i40evf_stat_update_32(&oes->tx_discards, &nes->tx_discards);
914 }
915
916 static void
917 i40evf_dev_xstats_reset(struct rte_eth_dev *dev)
918 {
919         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
920         struct i40e_eth_stats *pstats = NULL;
921
922         /* read stat values to clear hardware registers */
923         i40evf_query_stats(dev, &pstats);
924
925         /* set stats offset base on current values */
926         vf->vsi.eth_stats_offset = *pstats;
927 }
928
929 static int i40evf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
930                                       struct rte_eth_xstat_name *xstats_names,
931                                       __rte_unused unsigned limit)
932 {
933         unsigned i;
934
935         if (xstats_names != NULL)
936                 for (i = 0; i < I40EVF_NB_XSTATS; i++) {
937                         snprintf(xstats_names[i].name,
938                                 sizeof(xstats_names[i].name),
939                                 "%s", rte_i40evf_stats_strings[i].name);
940                 }
941         return I40EVF_NB_XSTATS;
942 }
943
944 static int i40evf_dev_xstats_get(struct rte_eth_dev *dev,
945                                  struct rte_eth_xstat *xstats, unsigned n)
946 {
947         int ret;
948         unsigned i;
949         struct i40e_eth_stats *pstats = NULL;
950         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
951         struct i40e_vsi *vsi = &vf->vsi;
952
953         if (n < I40EVF_NB_XSTATS)
954                 return I40EVF_NB_XSTATS;
955
956         ret = i40evf_query_stats(dev, &pstats);
957         if (ret != 0)
958                 return 0;
959
960         if (!xstats)
961                 return 0;
962
963         i40evf_update_stats(vsi, pstats);
964
965         /* loop over xstats array and values from pstats */
966         for (i = 0; i < I40EVF_NB_XSTATS; i++) {
967                 xstats[i].id = i;
968                 xstats[i].value = *(uint64_t *)(((char *)pstats) +
969                         rte_i40evf_stats_strings[i].offset);
970         }
971
972         return I40EVF_NB_XSTATS;
973 }
974
975 static int
976 i40evf_add_vlan(struct rte_eth_dev *dev, uint16_t vlanid)
977 {
978         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
979         struct virtchnl_vlan_filter_list *vlan_list;
980         uint8_t cmd_buffer[sizeof(struct virtchnl_vlan_filter_list) +
981                                                         sizeof(uint16_t)];
982         int err;
983         struct vf_cmd_info args;
984
985         vlan_list = (struct virtchnl_vlan_filter_list *)cmd_buffer;
986         vlan_list->vsi_id = vf->vsi_res->vsi_id;
987         vlan_list->num_elements = 1;
988         vlan_list->vlan_id[0] = vlanid;
989
990         args.ops = VIRTCHNL_OP_ADD_VLAN;
991         args.in_args = (u8 *)&cmd_buffer;
992         args.in_args_size = sizeof(cmd_buffer);
993         args.out_buffer = vf->aq_resp;
994         args.out_size = I40E_AQ_BUF_SZ;
995         err = i40evf_execute_vf_cmd(dev, &args);
996         if (err)
997                 PMD_DRV_LOG(ERR, "fail to execute command OP_ADD_VLAN");
998
999         return err;
1000 }
1001
1002 static int
1003 i40evf_del_vlan(struct rte_eth_dev *dev, uint16_t vlanid)
1004 {
1005         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
1006         struct virtchnl_vlan_filter_list *vlan_list;
1007         uint8_t cmd_buffer[sizeof(struct virtchnl_vlan_filter_list) +
1008                                                         sizeof(uint16_t)];
1009         int err;
1010         struct vf_cmd_info args;
1011
1012         vlan_list = (struct virtchnl_vlan_filter_list *)cmd_buffer;
1013         vlan_list->vsi_id = vf->vsi_res->vsi_id;
1014         vlan_list->num_elements = 1;
1015         vlan_list->vlan_id[0] = vlanid;
1016
1017         args.ops = VIRTCHNL_OP_DEL_VLAN;
1018         args.in_args = (u8 *)&cmd_buffer;
1019         args.in_args_size = sizeof(cmd_buffer);
1020         args.out_buffer = vf->aq_resp;
1021         args.out_size = I40E_AQ_BUF_SZ;
1022         err = i40evf_execute_vf_cmd(dev, &args);
1023         if (err)
1024                 PMD_DRV_LOG(ERR, "fail to execute command OP_DEL_VLAN");
1025
1026         return err;
1027 }
1028
1029 static const struct rte_pci_id pci_id_i40evf_map[] = {
1030         { RTE_PCI_DEVICE(I40E_INTEL_VENDOR_ID, I40E_DEV_ID_VF) },
1031         { RTE_PCI_DEVICE(I40E_INTEL_VENDOR_ID, I40E_DEV_ID_VF_HV) },
1032         { RTE_PCI_DEVICE(I40E_INTEL_VENDOR_ID, I40E_DEV_ID_X722_A0_VF) },
1033         { RTE_PCI_DEVICE(I40E_INTEL_VENDOR_ID, I40E_DEV_ID_X722_VF) },
1034         { .vendor_id = 0, /* sentinel */ },
1035 };
1036
1037 static inline int
1038 i40evf_dev_atomic_write_link_status(struct rte_eth_dev *dev,
1039                                     struct rte_eth_link *link)
1040 {
1041         struct rte_eth_link *dst = &(dev->data->dev_link);
1042         struct rte_eth_link *src = link;
1043
1044         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
1045                                         *(uint64_t *)src) == 0)
1046                 return -1;
1047
1048         return 0;
1049 }
1050
1051 /* Disable IRQ0 */
1052 static inline void
1053 i40evf_disable_irq0(struct i40e_hw *hw)
1054 {
1055         /* Disable all interrupt types */
1056         I40E_WRITE_REG(hw, I40E_VFINT_ICR0_ENA1, 0);
1057         I40E_WRITE_REG(hw, I40E_VFINT_DYN_CTL01,
1058                        I40E_VFINT_DYN_CTL01_ITR_INDX_MASK);
1059         I40EVF_WRITE_FLUSH(hw);
1060 }
1061
1062 /* Enable IRQ0 */
1063 static inline void
1064 i40evf_enable_irq0(struct i40e_hw *hw)
1065 {
1066         /* Enable admin queue interrupt trigger */
1067         uint32_t val;
1068
1069         i40evf_disable_irq0(hw);
1070         val = I40E_READ_REG(hw, I40E_VFINT_ICR0_ENA1);
1071         val |= I40E_VFINT_ICR0_ENA1_ADMINQ_MASK |
1072                 I40E_VFINT_ICR0_ENA1_LINK_STAT_CHANGE_MASK;
1073         I40E_WRITE_REG(hw, I40E_VFINT_ICR0_ENA1, val);
1074
1075         I40E_WRITE_REG(hw, I40E_VFINT_DYN_CTL01,
1076                 I40E_VFINT_DYN_CTL01_INTENA_MASK |
1077                 I40E_VFINT_DYN_CTL01_CLEARPBA_MASK |
1078                 I40E_VFINT_DYN_CTL01_ITR_INDX_MASK);
1079
1080         I40EVF_WRITE_FLUSH(hw);
1081 }
1082
1083 static int
1084 i40evf_check_vf_reset_done(struct i40e_hw *hw)
1085 {
1086         int i, reset;
1087
1088         for (i = 0; i < MAX_RESET_WAIT_CNT; i++) {
1089                 reset = I40E_READ_REG(hw, I40E_VFGEN_RSTAT) &
1090                         I40E_VFGEN_RSTAT_VFR_STATE_MASK;
1091                 reset = reset >> I40E_VFGEN_RSTAT_VFR_STATE_SHIFT;
1092                 if (reset == VIRTCHNL_VFR_VFACTIVE ||
1093                     reset == VIRTCHNL_VFR_COMPLETED)
1094                         break;
1095                 rte_delay_ms(50);
1096         }
1097
1098         if (i >= MAX_RESET_WAIT_CNT)
1099                 return -1;
1100
1101         return 0;
1102 }
1103 static int
1104 i40evf_reset_vf(struct i40e_hw *hw)
1105 {
1106         int ret;
1107
1108         if (i40e_vf_reset(hw) != I40E_SUCCESS) {
1109                 PMD_INIT_LOG(ERR, "Reset VF NIC failed");
1110                 return -1;
1111         }
1112         /**
1113           * After issuing vf reset command to pf, pf won't necessarily
1114           * reset vf, it depends on what state it exactly is. If it's not
1115           * initialized yet, it won't have vf reset since it's in a certain
1116           * state. If not, it will try to reset. Even vf is reset, pf will
1117           * set I40E_VFGEN_RSTAT to COMPLETE first, then wait 10ms and set
1118           * it to ACTIVE. In this duration, vf may not catch the moment that
1119           * COMPLETE is set. So, for vf, we'll try to wait a long time.
1120           */
1121         rte_delay_ms(200);
1122
1123         ret = i40evf_check_vf_reset_done(hw);
1124         if (ret) {
1125                 PMD_INIT_LOG(ERR, "VF is still resetting");
1126                 return ret;
1127         }
1128
1129         return 0;
1130 }
1131
1132 static int
1133 i40evf_init_vf(struct rte_eth_dev *dev)
1134 {
1135         int i, err, bufsz;
1136         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1137         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
1138         uint16_t interval =
1139                 i40e_calc_itr_interval(I40E_QUEUE_ITR_INTERVAL_MAX);
1140
1141         vf->adapter = I40E_DEV_PRIVATE_TO_ADAPTER(dev->data->dev_private);
1142         vf->dev_data = dev->data;
1143         err = i40e_set_mac_type(hw);
1144         if (err) {
1145                 PMD_INIT_LOG(ERR, "set_mac_type failed: %d", err);
1146                 goto err;
1147         }
1148
1149         err = i40evf_check_vf_reset_done(hw);
1150         if (err)
1151                 goto err;
1152
1153         i40e_init_adminq_parameter(hw);
1154         err = i40e_init_adminq(hw);
1155         if (err) {
1156                 PMD_INIT_LOG(ERR, "init_adminq failed: %d", err);
1157                 goto err;
1158         }
1159
1160         /* Reset VF and wait until it's complete */
1161         if (i40evf_reset_vf(hw)) {
1162                 PMD_INIT_LOG(ERR, "reset NIC failed");
1163                 goto err_aq;
1164         }
1165
1166         /* VF reset, shutdown admin queue and initialize again */
1167         if (i40e_shutdown_adminq(hw) != I40E_SUCCESS) {
1168                 PMD_INIT_LOG(ERR, "i40e_shutdown_adminq failed");
1169                 goto err;
1170         }
1171
1172         i40e_init_adminq_parameter(hw);
1173         if (i40e_init_adminq(hw) != I40E_SUCCESS) {
1174                 PMD_INIT_LOG(ERR, "init_adminq failed");
1175                 goto err;
1176         }
1177
1178         vf->aq_resp = rte_zmalloc("vf_aq_resp", I40E_AQ_BUF_SZ, 0);
1179         if (!vf->aq_resp) {
1180                 PMD_INIT_LOG(ERR, "unable to allocate vf_aq_resp memory");
1181                 goto err_aq;
1182         }
1183         if (i40evf_check_api_version(dev) != 0) {
1184                 PMD_INIT_LOG(ERR, "check_api version failed");
1185                 goto err_api;
1186         }
1187         bufsz = sizeof(struct virtchnl_vf_resource) +
1188                 (I40E_MAX_VF_VSI * sizeof(struct virtchnl_vsi_resource));
1189         vf->vf_res = rte_zmalloc("vf_res", bufsz, 0);
1190         if (!vf->vf_res) {
1191                 PMD_INIT_LOG(ERR, "unable to allocate vf_res memory");
1192                 goto err_api;
1193         }
1194
1195         if (i40evf_get_vf_resource(dev) != 0) {
1196                 PMD_INIT_LOG(ERR, "i40evf_get_vf_config failed");
1197                 goto err_alloc;
1198         }
1199
1200         /* got VF config message back from PF, now we can parse it */
1201         for (i = 0; i < vf->vf_res->num_vsis; i++) {
1202                 if (vf->vf_res->vsi_res[i].vsi_type == VIRTCHNL_VSI_SRIOV)
1203                         vf->vsi_res = &vf->vf_res->vsi_res[i];
1204         }
1205
1206         if (!vf->vsi_res) {
1207                 PMD_INIT_LOG(ERR, "no LAN VSI found");
1208                 goto err_alloc;
1209         }
1210
1211         if (hw->mac.type == I40E_MAC_X722_VF)
1212                 vf->flags = I40E_FLAG_RSS_AQ_CAPABLE;
1213         vf->vsi.vsi_id = vf->vsi_res->vsi_id;
1214
1215         switch (vf->vsi_res->vsi_type) {
1216         case VIRTCHNL_VSI_SRIOV:
1217                 vf->vsi.type = I40E_VSI_SRIOV;
1218                 break;
1219         default:
1220                 vf->vsi.type = I40E_VSI_TYPE_UNKNOWN;
1221                 break;
1222         }
1223         vf->vsi.nb_qps = vf->vsi_res->num_queue_pairs;
1224         vf->vsi.adapter = I40E_DEV_PRIVATE_TO_ADAPTER(dev->data->dev_private);
1225
1226         /* Store the MAC address configured by host, or generate random one */
1227         if (is_valid_assigned_ether_addr((struct ether_addr *)hw->mac.addr))
1228                 vf->flags |= I40E_FLAG_VF_MAC_BY_PF;
1229         else
1230                 eth_random_addr(hw->mac.addr); /* Generate a random one */
1231
1232         I40E_WRITE_REG(hw, I40E_VFINT_DYN_CTL01,
1233                        (I40E_ITR_INDEX_DEFAULT <<
1234                         I40E_VFINT_DYN_CTL0_ITR_INDX_SHIFT) |
1235                        (interval <<
1236                         I40E_VFINT_DYN_CTL0_INTERVAL_SHIFT));
1237         I40EVF_WRITE_FLUSH(hw);
1238
1239         return 0;
1240
1241 err_alloc:
1242         rte_free(vf->vf_res);
1243         vf->vsi_res = NULL;
1244 err_api:
1245         rte_free(vf->aq_resp);
1246 err_aq:
1247         i40e_shutdown_adminq(hw); /* ignore error */
1248 err:
1249         return -1;
1250 }
1251
1252 static int
1253 i40evf_uninit_vf(struct rte_eth_dev *dev)
1254 {
1255         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
1256         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1257
1258         PMD_INIT_FUNC_TRACE();
1259
1260         if (hw->adapter_stopped == 0)
1261                 i40evf_dev_close(dev);
1262         rte_free(vf->vf_res);
1263         vf->vf_res = NULL;
1264         rte_free(vf->aq_resp);
1265         vf->aq_resp = NULL;
1266
1267         return 0;
1268 }
1269
1270 static void
1271 i40evf_handle_pf_event(struct rte_eth_dev *dev, uint8_t *msg,
1272                 __rte_unused uint16_t msglen)
1273 {
1274         struct virtchnl_pf_event *pf_msg =
1275                         (struct virtchnl_pf_event *)msg;
1276         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
1277
1278         switch (pf_msg->event) {
1279         case VIRTCHNL_EVENT_RESET_IMPENDING:
1280                 PMD_DRV_LOG(DEBUG, "VIRTCHNL_EVENT_RESET_IMPENDING event");
1281                 _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_RESET,
1282                                               NULL, NULL);
1283                 break;
1284         case VIRTCHNL_EVENT_LINK_CHANGE:
1285                 PMD_DRV_LOG(DEBUG, "VIRTCHNL_EVENT_LINK_CHANGE event");
1286                 vf->link_up = pf_msg->event_data.link_event.link_status;
1287                 vf->link_speed = pf_msg->event_data.link_event.link_speed;
1288                 break;
1289         case VIRTCHNL_EVENT_PF_DRIVER_CLOSE:
1290                 PMD_DRV_LOG(DEBUG, "VIRTCHNL_EVENT_PF_DRIVER_CLOSE event");
1291                 break;
1292         default:
1293                 PMD_DRV_LOG(ERR, " unknown event received %u", pf_msg->event);
1294                 break;
1295         }
1296 }
1297
1298 static void
1299 i40evf_handle_aq_msg(struct rte_eth_dev *dev)
1300 {
1301         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1302         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
1303         struct i40e_arq_event_info info;
1304         uint16_t pending, aq_opc;
1305         enum virtchnl_ops msg_opc;
1306         enum i40e_status_code msg_ret;
1307         int ret;
1308
1309         info.buf_len = I40E_AQ_BUF_SZ;
1310         if (!vf->aq_resp) {
1311                 PMD_DRV_LOG(ERR, "Buffer for adminq resp should not be NULL");
1312                 return;
1313         }
1314         info.msg_buf = vf->aq_resp;
1315
1316         pending = 1;
1317         while (pending) {
1318                 ret = i40e_clean_arq_element(hw, &info, &pending);
1319
1320                 if (ret != I40E_SUCCESS) {
1321                         PMD_DRV_LOG(INFO, "Failed to read msg from AdminQ,"
1322                                     "ret: %d", ret);
1323                         break;
1324                 }
1325                 aq_opc = rte_le_to_cpu_16(info.desc.opcode);
1326                 /* For the message sent from pf to vf, opcode is stored in
1327                  * cookie_high of struct i40e_aq_desc, while return error code
1328                  * are stored in cookie_low, Which is done by
1329                  * i40e_aq_send_msg_to_vf in PF driver.*/
1330                 msg_opc = (enum virtchnl_ops)rte_le_to_cpu_32(
1331                                                   info.desc.cookie_high);
1332                 msg_ret = (enum i40e_status_code)rte_le_to_cpu_32(
1333                                                   info.desc.cookie_low);
1334                 switch (aq_opc) {
1335                 case i40e_aqc_opc_send_msg_to_vf:
1336                         if (msg_opc == VIRTCHNL_OP_EVENT)
1337                                 /* process event*/
1338                                 i40evf_handle_pf_event(dev, info.msg_buf,
1339                                                        info.msg_len);
1340                         else {
1341                                 /* read message and it's expected one */
1342                                 if (msg_opc == vf->pend_cmd) {
1343                                         vf->cmd_retval = msg_ret;
1344                                         /* prevent compiler reordering */
1345                                         rte_compiler_barrier();
1346                                         _clear_cmd(vf);
1347                                 } else
1348                                         PMD_DRV_LOG(ERR, "command mismatch,"
1349                                                 "expect %u, get %u",
1350                                                 vf->pend_cmd, msg_opc);
1351                                 PMD_DRV_LOG(DEBUG, "adminq response is received,"
1352                                              " opcode = %d", msg_opc);
1353                         }
1354                         break;
1355                 default:
1356                         PMD_DRV_LOG(ERR, "Request %u is not supported yet",
1357                                     aq_opc);
1358                         break;
1359                 }
1360         }
1361 }
1362
1363 /**
1364  * Interrupt handler triggered by NIC  for handling
1365  * specific interrupt. Only adminq interrupt is processed in VF.
1366  *
1367  * @param handle
1368  *  Pointer to interrupt handle.
1369  * @param param
1370  *  The address of parameter (struct rte_eth_dev *) regsitered before.
1371  *
1372  * @return
1373  *  void
1374  */
1375 static void
1376 i40evf_dev_interrupt_handler(void *param)
1377 {
1378         struct rte_eth_dev *dev = (struct rte_eth_dev *)param;
1379         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1380         uint32_t icr0;
1381
1382         i40evf_disable_irq0(hw);
1383
1384         /* read out interrupt causes */
1385         icr0 = I40E_READ_REG(hw, I40E_VFINT_ICR01);
1386
1387         /* No interrupt event indicated */
1388         if (!(icr0 & I40E_VFINT_ICR01_INTEVENT_MASK)) {
1389                 PMD_DRV_LOG(DEBUG, "No interrupt event, nothing to do");
1390                 goto done;
1391         }
1392
1393         if (icr0 & I40E_VFINT_ICR01_ADMINQ_MASK) {
1394                 PMD_DRV_LOG(DEBUG, "ICR01_ADMINQ is reported");
1395                 i40evf_handle_aq_msg(dev);
1396         }
1397
1398         /* Link Status Change interrupt */
1399         if (icr0 & I40E_VFINT_ICR01_LINK_STAT_CHANGE_MASK)
1400                 PMD_DRV_LOG(DEBUG, "LINK_STAT_CHANGE is reported,"
1401                                    " do nothing");
1402
1403 done:
1404         i40evf_enable_irq0(hw);
1405 }
1406
1407 static int
1408 i40evf_dev_init(struct rte_eth_dev *eth_dev)
1409 {
1410         struct i40e_hw *hw
1411                 = I40E_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
1412         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1413
1414         PMD_INIT_FUNC_TRACE();
1415
1416         /* assign ops func pointer */
1417         eth_dev->dev_ops = &i40evf_eth_dev_ops;
1418         eth_dev->rx_pkt_burst = &i40e_recv_pkts;
1419         eth_dev->tx_pkt_burst = &i40e_xmit_pkts;
1420
1421         /*
1422          * For secondary processes, we don't initialise any further as primary
1423          * has already done this work.
1424          */
1425         if (rte_eal_process_type() != RTE_PROC_PRIMARY){
1426                 i40e_set_rx_function(eth_dev);
1427                 i40e_set_tx_function(eth_dev);
1428                 return 0;
1429         }
1430         i40e_set_default_ptype_table(eth_dev);
1431         i40e_set_default_pctype_table(eth_dev);
1432         rte_eth_copy_pci_info(eth_dev, pci_dev);
1433
1434         hw->vendor_id = pci_dev->id.vendor_id;
1435         hw->device_id = pci_dev->id.device_id;
1436         hw->subsystem_vendor_id = pci_dev->id.subsystem_vendor_id;
1437         hw->subsystem_device_id = pci_dev->id.subsystem_device_id;
1438         hw->bus.device = pci_dev->addr.devid;
1439         hw->bus.func = pci_dev->addr.function;
1440         hw->hw_addr = (void *)pci_dev->mem_resource[0].addr;
1441         hw->adapter_stopped = 0;
1442
1443         if(i40evf_init_vf(eth_dev) != 0) {
1444                 PMD_INIT_LOG(ERR, "Init vf failed");
1445                 return -1;
1446         }
1447
1448         /* register callback func to eal lib */
1449         rte_intr_callback_register(&pci_dev->intr_handle,
1450                 i40evf_dev_interrupt_handler, (void *)eth_dev);
1451
1452         /* enable uio intr after callback register */
1453         rte_intr_enable(&pci_dev->intr_handle);
1454
1455         /* configure and enable device interrupt */
1456         i40evf_enable_irq0(hw);
1457
1458         /* copy mac addr */
1459         eth_dev->data->mac_addrs = rte_zmalloc("i40evf_mac",
1460                                         ETHER_ADDR_LEN * I40E_NUM_MACADDR_MAX,
1461                                         0);
1462         if (eth_dev->data->mac_addrs == NULL) {
1463                 PMD_INIT_LOG(ERR, "Failed to allocate %d bytes needed to"
1464                                 " store MAC addresses",
1465                                 ETHER_ADDR_LEN * I40E_NUM_MACADDR_MAX);
1466                 return -ENOMEM;
1467         }
1468         ether_addr_copy((struct ether_addr *)hw->mac.addr,
1469                         &eth_dev->data->mac_addrs[0]);
1470
1471         return 0;
1472 }
1473
1474 static int
1475 i40evf_dev_uninit(struct rte_eth_dev *eth_dev)
1476 {
1477         PMD_INIT_FUNC_TRACE();
1478
1479         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1480                 return -EPERM;
1481
1482         eth_dev->dev_ops = NULL;
1483         eth_dev->rx_pkt_burst = NULL;
1484         eth_dev->tx_pkt_burst = NULL;
1485
1486         if (i40evf_uninit_vf(eth_dev) != 0) {
1487                 PMD_INIT_LOG(ERR, "i40evf_uninit_vf failed");
1488                 return -1;
1489         }
1490
1491         rte_free(eth_dev->data->mac_addrs);
1492         eth_dev->data->mac_addrs = NULL;
1493
1494         return 0;
1495 }
1496
1497 static int eth_i40evf_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1498         struct rte_pci_device *pci_dev)
1499 {
1500         return rte_eth_dev_pci_generic_probe(pci_dev,
1501                 sizeof(struct i40e_adapter), i40evf_dev_init);
1502 }
1503
1504 static int eth_i40evf_pci_remove(struct rte_pci_device *pci_dev)
1505 {
1506         return rte_eth_dev_pci_generic_remove(pci_dev, i40evf_dev_uninit);
1507 }
1508
1509 /*
1510  * virtual function driver struct
1511  */
1512 static struct rte_pci_driver rte_i40evf_pmd = {
1513         .id_table = pci_id_i40evf_map,
1514         .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_IOVA_AS_VA,
1515         .probe = eth_i40evf_pci_probe,
1516         .remove = eth_i40evf_pci_remove,
1517 };
1518
1519 RTE_PMD_REGISTER_PCI(net_i40e_vf, rte_i40evf_pmd);
1520 RTE_PMD_REGISTER_PCI_TABLE(net_i40e_vf, pci_id_i40evf_map);
1521 RTE_PMD_REGISTER_KMOD_DEP(net_i40e_vf, "* igb_uio | vfio-pci");
1522
1523 static int
1524 i40evf_dev_configure(struct rte_eth_dev *dev)
1525 {
1526         struct i40e_adapter *ad =
1527                 I40E_DEV_PRIVATE_TO_ADAPTER(dev->data->dev_private);
1528         struct rte_eth_conf *conf = &dev->data->dev_conf;
1529         struct i40e_vf *vf;
1530
1531         /* Initialize to TRUE. If any of Rx queues doesn't meet the bulk
1532          * allocation or vector Rx preconditions we will reset it.
1533          */
1534         ad->rx_bulk_alloc_allowed = true;
1535         ad->rx_vec_allowed = true;
1536         ad->tx_simple_allowed = true;
1537         ad->tx_vec_allowed = true;
1538
1539         /* For non-DPDK PF drivers, VF has no ability to disable HW
1540          * CRC strip, and is implicitly enabled by the PF.
1541          */
1542         if (!conf->rxmode.hw_strip_crc) {
1543                 vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
1544                 if ((vf->version_major == VIRTCHNL_VERSION_MAJOR) &&
1545                     (vf->version_minor <= VIRTCHNL_VERSION_MINOR)) {
1546                         /* Peer is running non-DPDK PF driver. */
1547                         PMD_INIT_LOG(ERR, "VF can't disable HW CRC Strip");
1548                         return -EINVAL;
1549                 }
1550         }
1551
1552         return i40evf_init_vlan(dev);
1553 }
1554
1555 static int
1556 i40evf_init_vlan(struct rte_eth_dev *dev)
1557 {
1558         /* Apply vlan offload setting */
1559         i40evf_vlan_offload_set(dev, ETH_VLAN_STRIP_MASK);
1560
1561         return 0;
1562 }
1563
1564 static int
1565 i40evf_vlan_offload_set(struct rte_eth_dev *dev, int mask)
1566 {
1567         struct rte_eth_conf *dev_conf = &dev->data->dev_conf;
1568         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
1569
1570         if (!(vf->vf_res->vf_offload_flags & VIRTCHNL_VF_OFFLOAD_VLAN))
1571                 return -ENOTSUP;
1572
1573         /* Vlan stripping setting */
1574         if (mask & ETH_VLAN_STRIP_MASK) {
1575                 /* Enable or disable VLAN stripping */
1576                 if (dev_conf->rxmode.hw_vlan_strip)
1577                         i40evf_enable_vlan_strip(dev);
1578                 else
1579                         i40evf_disable_vlan_strip(dev);
1580         }
1581
1582         return 0;
1583 }
1584
1585 static int
1586 i40evf_dev_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1587 {
1588         struct i40e_rx_queue *rxq;
1589         int err = 0;
1590         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1591
1592         PMD_INIT_FUNC_TRACE();
1593
1594         if (rx_queue_id < dev->data->nb_rx_queues) {
1595                 rxq = dev->data->rx_queues[rx_queue_id];
1596
1597                 err = i40e_alloc_rx_queue_mbufs(rxq);
1598                 if (err) {
1599                         PMD_DRV_LOG(ERR, "Failed to allocate RX queue mbuf");
1600                         return err;
1601                 }
1602
1603                 rte_wmb();
1604
1605                 /* Init the RX tail register. */
1606                 I40E_PCI_REG_WRITE(rxq->qrx_tail, rxq->nb_rx_desc - 1);
1607                 I40EVF_WRITE_FLUSH(hw);
1608
1609                 /* Ready to switch the queue on */
1610                 err = i40evf_switch_queue(dev, TRUE, rx_queue_id, TRUE);
1611
1612                 if (err)
1613                         PMD_DRV_LOG(ERR, "Failed to switch RX queue %u on",
1614                                     rx_queue_id);
1615                 else
1616                         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1617         }
1618
1619         return err;
1620 }
1621
1622 static int
1623 i40evf_dev_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1624 {
1625         struct i40e_rx_queue *rxq;
1626         int err;
1627
1628         if (rx_queue_id < dev->data->nb_rx_queues) {
1629                 rxq = dev->data->rx_queues[rx_queue_id];
1630
1631                 err = i40evf_switch_queue(dev, TRUE, rx_queue_id, FALSE);
1632
1633                 if (err) {
1634                         PMD_DRV_LOG(ERR, "Failed to switch RX queue %u off",
1635                                     rx_queue_id);
1636                         return err;
1637                 }
1638
1639                 i40e_rx_queue_release_mbufs(rxq);
1640                 i40e_reset_rx_queue(rxq);
1641                 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1642         }
1643
1644         return 0;
1645 }
1646
1647 static int
1648 i40evf_dev_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1649 {
1650         int err = 0;
1651
1652         PMD_INIT_FUNC_TRACE();
1653
1654         if (tx_queue_id < dev->data->nb_tx_queues) {
1655
1656                 /* Ready to switch the queue on */
1657                 err = i40evf_switch_queue(dev, FALSE, tx_queue_id, TRUE);
1658
1659                 if (err)
1660                         PMD_DRV_LOG(ERR, "Failed to switch TX queue %u on",
1661                                     tx_queue_id);
1662                 else
1663                         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1664         }
1665
1666         return err;
1667 }
1668
1669 static int
1670 i40evf_dev_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1671 {
1672         struct i40e_tx_queue *txq;
1673         int err;
1674
1675         if (tx_queue_id < dev->data->nb_tx_queues) {
1676                 txq = dev->data->tx_queues[tx_queue_id];
1677
1678                 err = i40evf_switch_queue(dev, FALSE, tx_queue_id, FALSE);
1679
1680                 if (err) {
1681                         PMD_DRV_LOG(ERR, "Failed to switch TX queue %u off",
1682                                     tx_queue_id);
1683                         return err;
1684                 }
1685
1686                 i40e_tx_queue_release_mbufs(txq);
1687                 i40e_reset_tx_queue(txq);
1688                 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1689         }
1690
1691         return 0;
1692 }
1693
1694 static int
1695 i40evf_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1696 {
1697         int ret;
1698
1699         if (on)
1700                 ret = i40evf_add_vlan(dev, vlan_id);
1701         else
1702                 ret = i40evf_del_vlan(dev,vlan_id);
1703
1704         return ret;
1705 }
1706
1707 static int
1708 i40evf_rxq_init(struct rte_eth_dev *dev, struct i40e_rx_queue *rxq)
1709 {
1710         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1711         struct rte_eth_dev_data *dev_data = dev->data;
1712         struct rte_pktmbuf_pool_private *mbp_priv;
1713         uint16_t buf_size, len;
1714
1715         rxq->qrx_tail = hw->hw_addr + I40E_QRX_TAIL1(rxq->queue_id);
1716         I40E_PCI_REG_WRITE(rxq->qrx_tail, rxq->nb_rx_desc - 1);
1717         I40EVF_WRITE_FLUSH(hw);
1718
1719         /* Calculate the maximum packet length allowed */
1720         mbp_priv = rte_mempool_get_priv(rxq->mp);
1721         buf_size = (uint16_t)(mbp_priv->mbuf_data_room_size -
1722                                         RTE_PKTMBUF_HEADROOM);
1723         rxq->hs_mode = i40e_header_split_none;
1724         rxq->rx_hdr_len = 0;
1725         rxq->rx_buf_len = RTE_ALIGN(buf_size, (1 << I40E_RXQ_CTX_DBUFF_SHIFT));
1726         len = rxq->rx_buf_len * I40E_MAX_CHAINED_RX_BUFFERS;
1727         rxq->max_pkt_len = RTE_MIN(len,
1728                 dev_data->dev_conf.rxmode.max_rx_pkt_len);
1729
1730         /**
1731          * Check if the jumbo frame and maximum packet length are set correctly
1732          */
1733         if (dev_data->dev_conf.rxmode.jumbo_frame == 1) {
1734                 if (rxq->max_pkt_len <= ETHER_MAX_LEN ||
1735                     rxq->max_pkt_len > I40E_FRAME_SIZE_MAX) {
1736                         PMD_DRV_LOG(ERR, "maximum packet length must be "
1737                                 "larger than %u and smaller than %u, as jumbo "
1738                                 "frame is enabled", (uint32_t)ETHER_MAX_LEN,
1739                                         (uint32_t)I40E_FRAME_SIZE_MAX);
1740                         return I40E_ERR_CONFIG;
1741                 }
1742         } else {
1743                 if (rxq->max_pkt_len < ETHER_MIN_LEN ||
1744                     rxq->max_pkt_len > ETHER_MAX_LEN) {
1745                         PMD_DRV_LOG(ERR, "maximum packet length must be "
1746                                 "larger than %u and smaller than %u, as jumbo "
1747                                 "frame is disabled", (uint32_t)ETHER_MIN_LEN,
1748                                                 (uint32_t)ETHER_MAX_LEN);
1749                         return I40E_ERR_CONFIG;
1750                 }
1751         }
1752
1753         if (dev_data->dev_conf.rxmode.enable_scatter ||
1754             (rxq->max_pkt_len + 2 * I40E_VLAN_TAG_SIZE) > buf_size) {
1755                 dev_data->scattered_rx = 1;
1756         }
1757
1758         return 0;
1759 }
1760
1761 static int
1762 i40evf_rx_init(struct rte_eth_dev *dev)
1763 {
1764         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
1765         uint16_t i;
1766         int ret = I40E_SUCCESS;
1767         struct i40e_rx_queue **rxq =
1768                 (struct i40e_rx_queue **)dev->data->rx_queues;
1769
1770         i40evf_config_rss(vf);
1771         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1772                 if (!rxq[i] || !rxq[i]->q_set)
1773                         continue;
1774                 ret = i40evf_rxq_init(dev, rxq[i]);
1775                 if (ret != I40E_SUCCESS)
1776                         break;
1777         }
1778         if (ret == I40E_SUCCESS)
1779                 i40e_set_rx_function(dev);
1780
1781         return ret;
1782 }
1783
1784 static void
1785 i40evf_tx_init(struct rte_eth_dev *dev)
1786 {
1787         uint16_t i;
1788         struct i40e_tx_queue **txq =
1789                 (struct i40e_tx_queue **)dev->data->tx_queues;
1790         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1791
1792         for (i = 0; i < dev->data->nb_tx_queues; i++)
1793                 txq[i]->qtx_tail = hw->hw_addr + I40E_QTX_TAIL1(i);
1794
1795         i40e_set_tx_function(dev);
1796 }
1797
1798 static inline void
1799 i40evf_enable_queues_intr(struct rte_eth_dev *dev)
1800 {
1801         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1802         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1803         struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
1804
1805         if (!rte_intr_allow_others(intr_handle)) {
1806                 I40E_WRITE_REG(hw,
1807                                I40E_VFINT_DYN_CTL01,
1808                                I40E_VFINT_DYN_CTL01_INTENA_MASK |
1809                                I40E_VFINT_DYN_CTL01_CLEARPBA_MASK |
1810                                I40E_VFINT_DYN_CTL01_ITR_INDX_MASK);
1811                 I40EVF_WRITE_FLUSH(hw);
1812                 return;
1813         }
1814
1815         I40EVF_WRITE_FLUSH(hw);
1816 }
1817
1818 static inline void
1819 i40evf_disable_queues_intr(struct rte_eth_dev *dev)
1820 {
1821         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1822         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1823         struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
1824
1825         if (!rte_intr_allow_others(intr_handle)) {
1826                 I40E_WRITE_REG(hw, I40E_VFINT_DYN_CTL01,
1827                                I40E_VFINT_DYN_CTL01_ITR_INDX_MASK);
1828                 I40EVF_WRITE_FLUSH(hw);
1829                 return;
1830         }
1831
1832         I40EVF_WRITE_FLUSH(hw);
1833 }
1834
1835 static int
1836 i40evf_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
1837 {
1838         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1839         struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
1840         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1841         uint16_t interval =
1842                 i40e_calc_itr_interval(RTE_LIBRTE_I40E_ITR_INTERVAL);
1843         uint16_t msix_intr;
1844
1845         msix_intr = intr_handle->intr_vec[queue_id];
1846         if (msix_intr == I40E_MISC_VEC_ID)
1847                 I40E_WRITE_REG(hw, I40E_VFINT_DYN_CTL01,
1848                                I40E_VFINT_DYN_CTL01_INTENA_MASK |
1849                                I40E_VFINT_DYN_CTL01_CLEARPBA_MASK |
1850                                (0 << I40E_VFINT_DYN_CTL01_ITR_INDX_SHIFT) |
1851                                (interval <<
1852                                 I40E_VFINT_DYN_CTL01_INTERVAL_SHIFT));
1853         else
1854                 I40E_WRITE_REG(hw,
1855                                I40E_VFINT_DYN_CTLN1(msix_intr -
1856                                                     I40E_RX_VEC_START),
1857                                I40E_VFINT_DYN_CTLN1_INTENA_MASK |
1858                                I40E_VFINT_DYN_CTLN1_CLEARPBA_MASK |
1859                                (0 << I40E_VFINT_DYN_CTLN1_ITR_INDX_SHIFT) |
1860                                (interval <<
1861                                 I40E_VFINT_DYN_CTLN1_INTERVAL_SHIFT));
1862
1863         I40EVF_WRITE_FLUSH(hw);
1864
1865         rte_intr_enable(&pci_dev->intr_handle);
1866
1867         return 0;
1868 }
1869
1870 static int
1871 i40evf_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
1872 {
1873         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1874         struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
1875         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1876         uint16_t msix_intr;
1877
1878         msix_intr = intr_handle->intr_vec[queue_id];
1879         if (msix_intr == I40E_MISC_VEC_ID)
1880                 I40E_WRITE_REG(hw, I40E_VFINT_DYN_CTL01, 0);
1881         else
1882                 I40E_WRITE_REG(hw,
1883                                I40E_VFINT_DYN_CTLN1(msix_intr -
1884                                                     I40E_RX_VEC_START),
1885                                0);
1886
1887         I40EVF_WRITE_FLUSH(hw);
1888
1889         return 0;
1890 }
1891
1892 static void
1893 i40evf_add_del_all_mac_addr(struct rte_eth_dev *dev, bool add)
1894 {
1895         struct virtchnl_ether_addr_list *list;
1896         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
1897         int err, i, j;
1898         int next_begin = 0;
1899         int begin = 0;
1900         uint32_t len;
1901         struct ether_addr *addr;
1902         struct vf_cmd_info args;
1903
1904         do {
1905                 j = 0;
1906                 len = sizeof(struct virtchnl_ether_addr_list);
1907                 for (i = begin; i < I40E_NUM_MACADDR_MAX; i++, next_begin++) {
1908                         if (is_zero_ether_addr(&dev->data->mac_addrs[i]))
1909                                 continue;
1910                         len += sizeof(struct virtchnl_ether_addr);
1911                         if (len >= I40E_AQ_BUF_SZ) {
1912                                 next_begin = i + 1;
1913                                 break;
1914                         }
1915                 }
1916
1917                 list = rte_zmalloc("i40evf_del_mac_buffer", len, 0);
1918                 if (!list) {
1919                         PMD_DRV_LOG(ERR, "fail to allocate memory");
1920                         return;
1921                 }
1922
1923                 for (i = begin; i < next_begin; i++) {
1924                         addr = &dev->data->mac_addrs[i];
1925                         if (is_zero_ether_addr(addr))
1926                                 continue;
1927                         rte_memcpy(list->list[j].addr, addr->addr_bytes,
1928                                          sizeof(addr->addr_bytes));
1929                         PMD_DRV_LOG(DEBUG, "add/rm mac:%x:%x:%x:%x:%x:%x",
1930                                     addr->addr_bytes[0], addr->addr_bytes[1],
1931                                     addr->addr_bytes[2], addr->addr_bytes[3],
1932                                     addr->addr_bytes[4], addr->addr_bytes[5]);
1933                         j++;
1934                 }
1935                 list->vsi_id = vf->vsi_res->vsi_id;
1936                 list->num_elements = j;
1937                 args.ops = add ? VIRTCHNL_OP_ADD_ETH_ADDR :
1938                            VIRTCHNL_OP_DEL_ETH_ADDR;
1939                 args.in_args = (uint8_t *)list;
1940                 args.in_args_size = len;
1941                 args.out_buffer = vf->aq_resp;
1942                 args.out_size = I40E_AQ_BUF_SZ;
1943                 err = i40evf_execute_vf_cmd(dev, &args);
1944                 if (err) {
1945                         PMD_DRV_LOG(ERR, "fail to execute command %s",
1946                                     add ? "OP_ADD_ETHER_ADDRESS" :
1947                                     "OP_DEL_ETHER_ADDRESS");
1948                 } else {
1949                         if (add)
1950                                 vf->vsi.mac_num++;
1951                         else
1952                                 vf->vsi.mac_num--;
1953                 }
1954                 rte_free(list);
1955                 begin = next_begin;
1956         } while (begin < I40E_NUM_MACADDR_MAX);
1957 }
1958
1959 static int
1960 i40evf_dev_start(struct rte_eth_dev *dev)
1961 {
1962         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
1963         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1964         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1965         struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
1966         uint32_t intr_vector = 0;
1967
1968         PMD_INIT_FUNC_TRACE();
1969
1970         hw->adapter_stopped = 0;
1971
1972         vf->max_pkt_len = dev->data->dev_conf.rxmode.max_rx_pkt_len;
1973         vf->num_queue_pairs = RTE_MAX(dev->data->nb_rx_queues,
1974                                         dev->data->nb_tx_queues);
1975
1976         /* check and configure queue intr-vector mapping */
1977         if (dev->data->dev_conf.intr_conf.rxq != 0) {
1978                 intr_vector = dev->data->nb_rx_queues;
1979                 if (rte_intr_efd_enable(intr_handle, intr_vector))
1980                         return -1;
1981         }
1982
1983         if (rte_intr_dp_is_en(intr_handle) && !intr_handle->intr_vec) {
1984                 intr_handle->intr_vec =
1985                         rte_zmalloc("intr_vec",
1986                                     dev->data->nb_rx_queues * sizeof(int), 0);
1987                 if (!intr_handle->intr_vec) {
1988                         PMD_INIT_LOG(ERR, "Failed to allocate %d rx_queues"
1989                                      " intr_vec", dev->data->nb_rx_queues);
1990                         return -ENOMEM;
1991                 }
1992         }
1993
1994         if (i40evf_rx_init(dev) != 0){
1995                 PMD_DRV_LOG(ERR, "failed to do RX init");
1996                 return -1;
1997         }
1998
1999         i40evf_tx_init(dev);
2000
2001         if (i40evf_configure_vsi_queues(dev) != 0) {
2002                 PMD_DRV_LOG(ERR, "configure queues failed");
2003                 goto err_queue;
2004         }
2005         if (i40evf_config_irq_map(dev)) {
2006                 PMD_DRV_LOG(ERR, "config_irq_map failed");
2007                 goto err_queue;
2008         }
2009
2010         /* Set all mac addrs */
2011         i40evf_add_del_all_mac_addr(dev, TRUE);
2012
2013         if (i40evf_start_queues(dev) != 0) {
2014                 PMD_DRV_LOG(ERR, "enable queues failed");
2015                 goto err_mac;
2016         }
2017
2018         /* When a VF port is bound to VFIO-PCI, only miscellaneous interrupt
2019          * is mapped to VFIO vector 0 in i40evf_dev_init( ).
2020          * If previous VFIO interrupt mapping set in i40evf_dev_init( ) is
2021          * not cleared, it will fail when rte_intr_enable( ) tries to map Rx
2022          * queue interrupt to other VFIO vectors.
2023          * So clear uio/vfio intr/evevnfd first to avoid failure.
2024          */
2025         if (dev->data->dev_conf.intr_conf.rxq != 0) {
2026                 rte_intr_disable(intr_handle);
2027                 rte_intr_enable(intr_handle);
2028         }
2029
2030         i40evf_enable_queues_intr(dev);
2031
2032         return 0;
2033
2034 err_mac:
2035         i40evf_add_del_all_mac_addr(dev, FALSE);
2036 err_queue:
2037         return -1;
2038 }
2039
2040 static void
2041 i40evf_dev_stop(struct rte_eth_dev *dev)
2042 {
2043         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2044         struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
2045         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2046
2047         PMD_INIT_FUNC_TRACE();
2048
2049         if (hw->adapter_stopped == 1)
2050                 return;
2051         i40evf_stop_queues(dev);
2052         i40evf_disable_queues_intr(dev);
2053         i40e_dev_clear_queues(dev);
2054
2055         /* Clean datapath event and queue/vec mapping */
2056         rte_intr_efd_disable(intr_handle);
2057         if (intr_handle->intr_vec) {
2058                 rte_free(intr_handle->intr_vec);
2059                 intr_handle->intr_vec = NULL;
2060         }
2061         /* remove all mac addrs */
2062         i40evf_add_del_all_mac_addr(dev, FALSE);
2063         hw->adapter_stopped = 1;
2064
2065 }
2066
2067 static int
2068 i40evf_dev_link_update(struct rte_eth_dev *dev,
2069                        __rte_unused int wait_to_complete)
2070 {
2071         struct rte_eth_link new_link;
2072         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2073         /*
2074          * DPDK pf host provide interfacet to acquire link status
2075          * while Linux driver does not
2076          */
2077
2078         /* Linux driver PF host */
2079         switch (vf->link_speed) {
2080         case I40E_LINK_SPEED_100MB:
2081                 new_link.link_speed = ETH_SPEED_NUM_100M;
2082                 break;
2083         case I40E_LINK_SPEED_1GB:
2084                 new_link.link_speed = ETH_SPEED_NUM_1G;
2085                 break;
2086         case I40E_LINK_SPEED_10GB:
2087                 new_link.link_speed = ETH_SPEED_NUM_10G;
2088                 break;
2089         case I40E_LINK_SPEED_20GB:
2090                 new_link.link_speed = ETH_SPEED_NUM_20G;
2091                 break;
2092         case I40E_LINK_SPEED_25GB:
2093                 new_link.link_speed = ETH_SPEED_NUM_25G;
2094                 break;
2095         case I40E_LINK_SPEED_40GB:
2096                 new_link.link_speed = ETH_SPEED_NUM_40G;
2097                 break;
2098         default:
2099                 new_link.link_speed = ETH_SPEED_NUM_100M;
2100                 break;
2101         }
2102         /* full duplex only */
2103         new_link.link_duplex = ETH_LINK_FULL_DUPLEX;
2104         new_link.link_status = vf->link_up ? ETH_LINK_UP :
2105                                              ETH_LINK_DOWN;
2106         new_link.link_autoneg =
2107                 dev->data->dev_conf.link_speeds & ETH_LINK_SPEED_FIXED;
2108
2109         i40evf_dev_atomic_write_link_status(dev, &new_link);
2110
2111         return 0;
2112 }
2113
2114 static void
2115 i40evf_dev_promiscuous_enable(struct rte_eth_dev *dev)
2116 {
2117         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2118         int ret;
2119
2120         /* If enabled, just return */
2121         if (vf->promisc_unicast_enabled)
2122                 return;
2123
2124         ret = i40evf_config_promisc(dev, 1, vf->promisc_multicast_enabled);
2125         if (ret == 0)
2126                 vf->promisc_unicast_enabled = TRUE;
2127 }
2128
2129 static void
2130 i40evf_dev_promiscuous_disable(struct rte_eth_dev *dev)
2131 {
2132         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2133         int ret;
2134
2135         /* If disabled, just return */
2136         if (!vf->promisc_unicast_enabled)
2137                 return;
2138
2139         ret = i40evf_config_promisc(dev, 0, vf->promisc_multicast_enabled);
2140         if (ret == 0)
2141                 vf->promisc_unicast_enabled = FALSE;
2142 }
2143
2144 static void
2145 i40evf_dev_allmulticast_enable(struct rte_eth_dev *dev)
2146 {
2147         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2148         int ret;
2149
2150         /* If enabled, just return */
2151         if (vf->promisc_multicast_enabled)
2152                 return;
2153
2154         ret = i40evf_config_promisc(dev, vf->promisc_unicast_enabled, 1);
2155         if (ret == 0)
2156                 vf->promisc_multicast_enabled = TRUE;
2157 }
2158
2159 static void
2160 i40evf_dev_allmulticast_disable(struct rte_eth_dev *dev)
2161 {
2162         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2163         int ret;
2164
2165         /* If enabled, just return */
2166         if (!vf->promisc_multicast_enabled)
2167                 return;
2168
2169         ret = i40evf_config_promisc(dev, vf->promisc_unicast_enabled, 0);
2170         if (ret == 0)
2171                 vf->promisc_multicast_enabled = FALSE;
2172 }
2173
2174 static void
2175 i40evf_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2176 {
2177         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2178
2179         memset(dev_info, 0, sizeof(*dev_info));
2180         dev_info->pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2181         dev_info->max_rx_queues = vf->vsi_res->num_queue_pairs;
2182         dev_info->max_tx_queues = vf->vsi_res->num_queue_pairs;
2183         dev_info->min_rx_bufsize = I40E_BUF_SIZE_MIN;
2184         dev_info->max_rx_pktlen = I40E_FRAME_SIZE_MAX;
2185         dev_info->hash_key_size = (I40E_VFQF_HKEY_MAX_INDEX + 1) * sizeof(uint32_t);
2186         dev_info->reta_size = ETH_RSS_RETA_SIZE_64;
2187         dev_info->flow_type_rss_offloads = vf->adapter->flow_types_mask;
2188         dev_info->max_mac_addrs = I40E_NUM_MACADDR_MAX;
2189         dev_info->rx_offload_capa =
2190                 DEV_RX_OFFLOAD_VLAN_STRIP |
2191                 DEV_RX_OFFLOAD_QINQ_STRIP |
2192                 DEV_RX_OFFLOAD_IPV4_CKSUM |
2193                 DEV_RX_OFFLOAD_UDP_CKSUM |
2194                 DEV_RX_OFFLOAD_TCP_CKSUM;
2195         dev_info->tx_offload_capa =
2196                 DEV_TX_OFFLOAD_VLAN_INSERT |
2197                 DEV_TX_OFFLOAD_QINQ_INSERT |
2198                 DEV_TX_OFFLOAD_IPV4_CKSUM |
2199                 DEV_TX_OFFLOAD_UDP_CKSUM |
2200                 DEV_TX_OFFLOAD_TCP_CKSUM |
2201                 DEV_TX_OFFLOAD_SCTP_CKSUM |
2202                 DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM |
2203                 DEV_TX_OFFLOAD_TCP_TSO |
2204                 DEV_TX_OFFLOAD_VXLAN_TNL_TSO |
2205                 DEV_TX_OFFLOAD_GRE_TNL_TSO |
2206                 DEV_TX_OFFLOAD_IPIP_TNL_TSO |
2207                 DEV_TX_OFFLOAD_GENEVE_TNL_TSO;
2208
2209         dev_info->default_rxconf = (struct rte_eth_rxconf) {
2210                 .rx_thresh = {
2211                         .pthresh = I40E_DEFAULT_RX_PTHRESH,
2212                         .hthresh = I40E_DEFAULT_RX_HTHRESH,
2213                         .wthresh = I40E_DEFAULT_RX_WTHRESH,
2214                 },
2215                 .rx_free_thresh = I40E_DEFAULT_RX_FREE_THRESH,
2216                 .rx_drop_en = 0,
2217         };
2218
2219         dev_info->default_txconf = (struct rte_eth_txconf) {
2220                 .tx_thresh = {
2221                         .pthresh = I40E_DEFAULT_TX_PTHRESH,
2222                         .hthresh = I40E_DEFAULT_TX_HTHRESH,
2223                         .wthresh = I40E_DEFAULT_TX_WTHRESH,
2224                 },
2225                 .tx_free_thresh = I40E_DEFAULT_TX_FREE_THRESH,
2226                 .tx_rs_thresh = I40E_DEFAULT_TX_RSBIT_THRESH,
2227                 .txq_flags = ETH_TXQ_FLAGS_NOMULTSEGS |
2228                                 ETH_TXQ_FLAGS_NOOFFLOADS,
2229         };
2230
2231         dev_info->rx_desc_lim = (struct rte_eth_desc_lim) {
2232                 .nb_max = I40E_MAX_RING_DESC,
2233                 .nb_min = I40E_MIN_RING_DESC,
2234                 .nb_align = I40E_ALIGN_RING_DESC,
2235         };
2236
2237         dev_info->tx_desc_lim = (struct rte_eth_desc_lim) {
2238                 .nb_max = I40E_MAX_RING_DESC,
2239                 .nb_min = I40E_MIN_RING_DESC,
2240                 .nb_align = I40E_ALIGN_RING_DESC,
2241         };
2242 }
2243
2244 static int
2245 i40evf_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
2246 {
2247         int ret;
2248         struct i40e_eth_stats *pstats = NULL;
2249         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2250         struct i40e_vsi *vsi = &vf->vsi;
2251
2252         ret = i40evf_query_stats(dev, &pstats);
2253         if (ret == 0) {
2254                 i40evf_update_stats(vsi, pstats);
2255
2256                 stats->ipackets = pstats->rx_unicast + pstats->rx_multicast +
2257                                                 pstats->rx_broadcast;
2258                 stats->opackets = pstats->tx_broadcast + pstats->tx_multicast +
2259                                                 pstats->tx_unicast;
2260                 stats->imissed = pstats->rx_discards;
2261                 stats->oerrors = pstats->tx_errors + pstats->tx_discards;
2262                 stats->ibytes = pstats->rx_bytes;
2263                 stats->obytes = pstats->tx_bytes;
2264         } else {
2265                 PMD_DRV_LOG(ERR, "Get statistics failed");
2266         }
2267         return ret;
2268 }
2269
2270 static void
2271 i40evf_dev_close(struct rte_eth_dev *dev)
2272 {
2273         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2274         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2275         struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
2276
2277         i40evf_dev_stop(dev);
2278         i40e_dev_free_queues(dev);
2279         i40evf_reset_vf(hw);
2280         i40e_shutdown_adminq(hw);
2281         /* disable uio intr before callback unregister */
2282         rte_intr_disable(intr_handle);
2283
2284         /* unregister callback func from eal lib */
2285         rte_intr_callback_unregister(intr_handle,
2286                                      i40evf_dev_interrupt_handler, dev);
2287         i40evf_disable_irq0(hw);
2288 }
2289
2290 /*
2291  * Reset VF device only to re-initialize resources in PMD layer
2292  */
2293 static int
2294 i40evf_dev_reset(struct rte_eth_dev *dev)
2295 {
2296         int ret;
2297
2298         ret = i40evf_dev_uninit(dev);
2299         if (ret)
2300                 return ret;
2301
2302         ret = i40evf_dev_init(dev);
2303
2304         return ret;
2305 }
2306
2307 static int
2308 i40evf_get_rss_lut(struct i40e_vsi *vsi, uint8_t *lut, uint16_t lut_size)
2309 {
2310         struct i40e_vf *vf = I40E_VSI_TO_VF(vsi);
2311         struct i40e_hw *hw = I40E_VSI_TO_HW(vsi);
2312         int ret;
2313
2314         if (!lut)
2315                 return -EINVAL;
2316
2317         if (vf->flags & I40E_FLAG_RSS_AQ_CAPABLE) {
2318                 ret = i40e_aq_get_rss_lut(hw, vsi->vsi_id, FALSE,
2319                                           lut, lut_size);
2320                 if (ret) {
2321                         PMD_DRV_LOG(ERR, "Failed to get RSS lookup table");
2322                         return ret;
2323                 }
2324         } else {
2325                 uint32_t *lut_dw = (uint32_t *)lut;
2326                 uint16_t i, lut_size_dw = lut_size / 4;
2327
2328                 for (i = 0; i < lut_size_dw; i++)
2329                         lut_dw[i] = I40E_READ_REG(hw, I40E_VFQF_HLUT(i));
2330         }
2331
2332         return 0;
2333 }
2334
2335 static int
2336 i40evf_set_rss_lut(struct i40e_vsi *vsi, uint8_t *lut, uint16_t lut_size)
2337 {
2338         struct i40e_vf *vf;
2339         struct i40e_hw *hw;
2340         int ret;
2341
2342         if (!vsi || !lut)
2343                 return -EINVAL;
2344
2345         vf = I40E_VSI_TO_VF(vsi);
2346         hw = I40E_VSI_TO_HW(vsi);
2347
2348         if (vf->flags & I40E_FLAG_RSS_AQ_CAPABLE) {
2349                 ret = i40e_aq_set_rss_lut(hw, vsi->vsi_id, FALSE,
2350                                           lut, lut_size);
2351                 if (ret) {
2352                         PMD_DRV_LOG(ERR, "Failed to set RSS lookup table");
2353                         return ret;
2354                 }
2355         } else {
2356                 uint32_t *lut_dw = (uint32_t *)lut;
2357                 uint16_t i, lut_size_dw = lut_size / 4;
2358
2359                 for (i = 0; i < lut_size_dw; i++)
2360                         I40E_WRITE_REG(hw, I40E_VFQF_HLUT(i), lut_dw[i]);
2361                 I40EVF_WRITE_FLUSH(hw);
2362         }
2363
2364         return 0;
2365 }
2366
2367 static int
2368 i40evf_dev_rss_reta_update(struct rte_eth_dev *dev,
2369                            struct rte_eth_rss_reta_entry64 *reta_conf,
2370                            uint16_t reta_size)
2371 {
2372         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2373         uint8_t *lut;
2374         uint16_t i, idx, shift;
2375         int ret;
2376
2377         if (reta_size != ETH_RSS_RETA_SIZE_64) {
2378                 PMD_DRV_LOG(ERR, "The size of hash lookup table configured "
2379                         "(%d) doesn't match the number of hardware can "
2380                         "support (%d)", reta_size, ETH_RSS_RETA_SIZE_64);
2381                 return -EINVAL;
2382         }
2383
2384         lut = rte_zmalloc("i40e_rss_lut", reta_size, 0);
2385         if (!lut) {
2386                 PMD_DRV_LOG(ERR, "No memory can be allocated");
2387                 return -ENOMEM;
2388         }
2389         ret = i40evf_get_rss_lut(&vf->vsi, lut, reta_size);
2390         if (ret)
2391                 goto out;
2392         for (i = 0; i < reta_size; i++) {
2393                 idx = i / RTE_RETA_GROUP_SIZE;
2394                 shift = i % RTE_RETA_GROUP_SIZE;
2395                 if (reta_conf[idx].mask & (1ULL << shift))
2396                         lut[i] = reta_conf[idx].reta[shift];
2397         }
2398         ret = i40evf_set_rss_lut(&vf->vsi, lut, reta_size);
2399
2400 out:
2401         rte_free(lut);
2402
2403         return ret;
2404 }
2405
2406 static int
2407 i40evf_dev_rss_reta_query(struct rte_eth_dev *dev,
2408                           struct rte_eth_rss_reta_entry64 *reta_conf,
2409                           uint16_t reta_size)
2410 {
2411         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2412         uint16_t i, idx, shift;
2413         uint8_t *lut;
2414         int ret;
2415
2416         if (reta_size != ETH_RSS_RETA_SIZE_64) {
2417                 PMD_DRV_LOG(ERR, "The size of hash lookup table configured "
2418                         "(%d) doesn't match the number of hardware can "
2419                         "support (%d)", reta_size, ETH_RSS_RETA_SIZE_64);
2420                 return -EINVAL;
2421         }
2422
2423         lut = rte_zmalloc("i40e_rss_lut", reta_size, 0);
2424         if (!lut) {
2425                 PMD_DRV_LOG(ERR, "No memory can be allocated");
2426                 return -ENOMEM;
2427         }
2428
2429         ret = i40evf_get_rss_lut(&vf->vsi, lut, reta_size);
2430         if (ret)
2431                 goto out;
2432         for (i = 0; i < reta_size; i++) {
2433                 idx = i / RTE_RETA_GROUP_SIZE;
2434                 shift = i % RTE_RETA_GROUP_SIZE;
2435                 if (reta_conf[idx].mask & (1ULL << shift))
2436                         reta_conf[idx].reta[shift] = lut[i];
2437         }
2438
2439 out:
2440         rte_free(lut);
2441
2442         return ret;
2443 }
2444
2445 static int
2446 i40evf_set_rss_key(struct i40e_vsi *vsi, uint8_t *key, uint8_t key_len)
2447 {
2448         struct i40e_vf *vf = I40E_VSI_TO_VF(vsi);
2449         struct i40e_hw *hw = I40E_VSI_TO_HW(vsi);
2450         int ret = 0;
2451
2452         if (!key || key_len == 0) {
2453                 PMD_DRV_LOG(DEBUG, "No key to be configured");
2454                 return 0;
2455         } else if (key_len != (I40E_VFQF_HKEY_MAX_INDEX + 1) *
2456                 sizeof(uint32_t)) {
2457                 PMD_DRV_LOG(ERR, "Invalid key length %u", key_len);
2458                 return -EINVAL;
2459         }
2460
2461         if (vf->flags & I40E_FLAG_RSS_AQ_CAPABLE) {
2462                 struct i40e_aqc_get_set_rss_key_data *key_dw =
2463                         (struct i40e_aqc_get_set_rss_key_data *)key;
2464
2465                 ret = i40e_aq_set_rss_key(hw, vsi->vsi_id, key_dw);
2466                 if (ret)
2467                         PMD_INIT_LOG(ERR, "Failed to configure RSS key "
2468                                      "via AQ");
2469         } else {
2470                 uint32_t *hash_key = (uint32_t *)key;
2471                 uint16_t i;
2472
2473                 for (i = 0; i <= I40E_VFQF_HKEY_MAX_INDEX; i++)
2474                         i40e_write_rx_ctl(hw, I40E_VFQF_HKEY(i), hash_key[i]);
2475                 I40EVF_WRITE_FLUSH(hw);
2476         }
2477
2478         return ret;
2479 }
2480
2481 static int
2482 i40evf_get_rss_key(struct i40e_vsi *vsi, uint8_t *key, uint8_t *key_len)
2483 {
2484         struct i40e_vf *vf = I40E_VSI_TO_VF(vsi);
2485         struct i40e_hw *hw = I40E_VSI_TO_HW(vsi);
2486         int ret;
2487
2488         if (!key || !key_len)
2489                 return -EINVAL;
2490
2491         if (vf->flags & I40E_FLAG_RSS_AQ_CAPABLE) {
2492                 ret = i40e_aq_get_rss_key(hw, vsi->vsi_id,
2493                         (struct i40e_aqc_get_set_rss_key_data *)key);
2494                 if (ret) {
2495                         PMD_INIT_LOG(ERR, "Failed to get RSS key via AQ");
2496                         return ret;
2497                 }
2498         } else {
2499                 uint32_t *key_dw = (uint32_t *)key;
2500                 uint16_t i;
2501
2502                 for (i = 0; i <= I40E_VFQF_HKEY_MAX_INDEX; i++)
2503                         key_dw[i] = i40e_read_rx_ctl(hw, I40E_VFQF_HKEY(i));
2504         }
2505         *key_len = (I40E_VFQF_HKEY_MAX_INDEX + 1) * sizeof(uint32_t);
2506
2507         return 0;
2508 }
2509
2510 static int
2511 i40evf_hw_rss_hash_set(struct i40e_vf *vf, struct rte_eth_rss_conf *rss_conf)
2512 {
2513         struct i40e_hw *hw = I40E_VF_TO_HW(vf);
2514         uint64_t hena;
2515         int ret;
2516
2517         ret = i40evf_set_rss_key(&vf->vsi, rss_conf->rss_key,
2518                                  rss_conf->rss_key_len);
2519         if (ret)
2520                 return ret;
2521
2522         hena = i40e_config_hena(vf->adapter, rss_conf->rss_hf);
2523         i40e_write_rx_ctl(hw, I40E_VFQF_HENA(0), (uint32_t)hena);
2524         i40e_write_rx_ctl(hw, I40E_VFQF_HENA(1), (uint32_t)(hena >> 32));
2525         I40EVF_WRITE_FLUSH(hw);
2526
2527         return 0;
2528 }
2529
2530 static void
2531 i40evf_disable_rss(struct i40e_vf *vf)
2532 {
2533         struct i40e_hw *hw = I40E_VF_TO_HW(vf);
2534
2535         i40e_write_rx_ctl(hw, I40E_VFQF_HENA(0), 0);
2536         i40e_write_rx_ctl(hw, I40E_VFQF_HENA(1), 0);
2537         I40EVF_WRITE_FLUSH(hw);
2538 }
2539
2540 static int
2541 i40evf_config_rss(struct i40e_vf *vf)
2542 {
2543         struct i40e_hw *hw = I40E_VF_TO_HW(vf);
2544         struct rte_eth_rss_conf rss_conf;
2545         uint32_t i, j, lut = 0, nb_q = (I40E_VFQF_HLUT_MAX_INDEX + 1) * 4;
2546         uint16_t num;
2547
2548         if (vf->dev_data->dev_conf.rxmode.mq_mode != ETH_MQ_RX_RSS) {
2549                 i40evf_disable_rss(vf);
2550                 PMD_DRV_LOG(DEBUG, "RSS not configured");
2551                 return 0;
2552         }
2553
2554         num = RTE_MIN(vf->dev_data->nb_rx_queues, I40E_MAX_QP_NUM_PER_VF);
2555         /* Fill out the look up table */
2556         for (i = 0, j = 0; i < nb_q; i++, j++) {
2557                 if (j >= num)
2558                         j = 0;
2559                 lut = (lut << 8) | j;
2560                 if ((i & 3) == 3)
2561                         I40E_WRITE_REG(hw, I40E_VFQF_HLUT(i >> 2), lut);
2562         }
2563
2564         rss_conf = vf->dev_data->dev_conf.rx_adv_conf.rss_conf;
2565         if ((rss_conf.rss_hf & vf->adapter->flow_types_mask) == 0) {
2566                 i40evf_disable_rss(vf);
2567                 PMD_DRV_LOG(DEBUG, "No hash flag is set");
2568                 return 0;
2569         }
2570
2571         if (rss_conf.rss_key == NULL || rss_conf.rss_key_len <
2572                 (I40E_VFQF_HKEY_MAX_INDEX + 1) * sizeof(uint32_t)) {
2573                 /* Calculate the default hash key */
2574                 for (i = 0; i <= I40E_VFQF_HKEY_MAX_INDEX; i++)
2575                         rss_key_default[i] = (uint32_t)rte_rand();
2576                 rss_conf.rss_key = (uint8_t *)rss_key_default;
2577                 rss_conf.rss_key_len = (I40E_VFQF_HKEY_MAX_INDEX + 1) *
2578                         sizeof(uint32_t);
2579         }
2580
2581         return i40evf_hw_rss_hash_set(vf, &rss_conf);
2582 }
2583
2584 static int
2585 i40evf_dev_rss_hash_update(struct rte_eth_dev *dev,
2586                            struct rte_eth_rss_conf *rss_conf)
2587 {
2588         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2589         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2590         uint64_t rss_hf = rss_conf->rss_hf & vf->adapter->flow_types_mask;
2591         uint64_t hena;
2592
2593         hena = (uint64_t)i40e_read_rx_ctl(hw, I40E_VFQF_HENA(0));
2594         hena |= ((uint64_t)i40e_read_rx_ctl(hw, I40E_VFQF_HENA(1))) << 32;
2595
2596         if (!(hena & vf->adapter->pctypes_mask)) { /* RSS disabled */
2597                 if (rss_hf != 0) /* Enable RSS */
2598                         return -EINVAL;
2599                 return 0;
2600         }
2601
2602         /* RSS enabled */
2603         if (rss_hf == 0) /* Disable RSS */
2604                 return -EINVAL;
2605
2606         return i40evf_hw_rss_hash_set(vf, rss_conf);
2607 }
2608
2609 static int
2610 i40evf_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
2611                              struct rte_eth_rss_conf *rss_conf)
2612 {
2613         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2614         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2615         uint64_t hena;
2616
2617         i40evf_get_rss_key(&vf->vsi, rss_conf->rss_key,
2618                            &rss_conf->rss_key_len);
2619
2620         hena = (uint64_t)i40e_read_rx_ctl(hw, I40E_VFQF_HENA(0));
2621         hena |= ((uint64_t)i40e_read_rx_ctl(hw, I40E_VFQF_HENA(1))) << 32;
2622         rss_conf->rss_hf = i40e_parse_hena(vf->adapter, hena);
2623
2624         return 0;
2625 }
2626
2627 static int
2628 i40evf_dev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
2629 {
2630         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2631         struct rte_eth_dev_data *dev_data = vf->dev_data;
2632         uint32_t frame_size = mtu + I40E_ETH_OVERHEAD;
2633         int ret = 0;
2634
2635         /* check if mtu is within the allowed range */
2636         if ((mtu < ETHER_MIN_MTU) || (frame_size > I40E_FRAME_SIZE_MAX))
2637                 return -EINVAL;
2638
2639         /* mtu setting is forbidden if port is start */
2640         if (dev_data->dev_started) {
2641                 PMD_DRV_LOG(ERR, "port %d must be stopped before configuration",
2642                             dev_data->port_id);
2643                 return -EBUSY;
2644         }
2645
2646         if (frame_size > ETHER_MAX_LEN)
2647                 dev_data->dev_conf.rxmode.jumbo_frame = 1;
2648         else
2649                 dev_data->dev_conf.rxmode.jumbo_frame = 0;
2650
2651         dev_data->dev_conf.rxmode.max_rx_pkt_len = frame_size;
2652
2653         return ret;
2654 }
2655
2656 static void
2657 i40evf_set_default_mac_addr(struct rte_eth_dev *dev,
2658                             struct ether_addr *mac_addr)
2659 {
2660         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2661
2662         if (!is_valid_assigned_ether_addr(mac_addr)) {
2663                 PMD_DRV_LOG(ERR, "Tried to set invalid MAC address.");
2664                 return;
2665         }
2666
2667         if (vf->flags & I40E_FLAG_VF_MAC_BY_PF)
2668                 return;
2669
2670         i40evf_del_mac_addr_by_addr(dev, dev->data->mac_addrs);
2671
2672         i40evf_add_mac_addr(dev, mac_addr, 0, 0);
2673 }