pci: introduce library and driver
[dpdk.git] / drivers / net / i40e / i40e_ethdev_vf.c
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
5  *   All rights reserved.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
11  *     * Redistributions of source code must retain the above copyright
12  *       notice, this list of conditions and the following disclaimer.
13  *     * Redistributions in binary form must reproduce the above copyright
14  *       notice, this list of conditions and the following disclaimer in
15  *       the documentation and/or other materials provided with the
16  *       distribution.
17  *     * Neither the name of Intel Corporation nor the names of its
18  *       contributors may be used to endorse or promote products derived
19  *       from this software without specific prior written permission.
20  *
21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33
34 #include <sys/queue.h>
35 #include <stdio.h>
36 #include <errno.h>
37 #include <stdint.h>
38 #include <string.h>
39 #include <unistd.h>
40 #include <stdarg.h>
41 #include <inttypes.h>
42 #include <rte_byteorder.h>
43 #include <rte_common.h>
44 #include <rte_cycles.h>
45
46 #include <rte_interrupts.h>
47 #include <rte_log.h>
48 #include <rte_debug.h>
49 #include <rte_pci.h>
50 #include <rte_bus_pci.h>
51 #include <rte_atomic.h>
52 #include <rte_branch_prediction.h>
53 #include <rte_memory.h>
54 #include <rte_memzone.h>
55 #include <rte_eal.h>
56 #include <rte_alarm.h>
57 #include <rte_ether.h>
58 #include <rte_ethdev.h>
59 #include <rte_ethdev_pci.h>
60 #include <rte_malloc.h>
61 #include <rte_dev.h>
62
63 #include "i40e_logs.h"
64 #include "base/i40e_prototype.h"
65 #include "base/i40e_adminq_cmd.h"
66 #include "base/i40e_type.h"
67
68 #include "i40e_rxtx.h"
69 #include "i40e_ethdev.h"
70 #include "i40e_pf.h"
71 #define I40EVF_VSI_DEFAULT_MSIX_INTR     1
72 #define I40EVF_VSI_DEFAULT_MSIX_INTR_LNX 0
73
74 /* busy wait delay in msec */
75 #define I40EVF_BUSY_WAIT_DELAY 10
76 #define I40EVF_BUSY_WAIT_COUNT 50
77 #define MAX_RESET_WAIT_CNT     20
78
79 struct i40evf_arq_msg_info {
80         enum virtchnl_ops ops;
81         enum i40e_status_code result;
82         uint16_t buf_len;
83         uint16_t msg_len;
84         uint8_t *msg;
85 };
86
87 struct vf_cmd_info {
88         enum virtchnl_ops ops;
89         uint8_t *in_args;
90         uint32_t in_args_size;
91         uint8_t *out_buffer;
92         /* Input & output type. pass in buffer size and pass out
93          * actual return result
94          */
95         uint32_t out_size;
96 };
97
98 enum i40evf_aq_result {
99         I40EVF_MSG_ERR = -1, /* Meet error when accessing admin queue */
100         I40EVF_MSG_NON,      /* Read nothing from admin queue */
101         I40EVF_MSG_SYS,      /* Read system msg from admin queue */
102         I40EVF_MSG_CMD,      /* Read async command result */
103 };
104
105 static int i40evf_dev_configure(struct rte_eth_dev *dev);
106 static int i40evf_dev_start(struct rte_eth_dev *dev);
107 static void i40evf_dev_stop(struct rte_eth_dev *dev);
108 static void i40evf_dev_info_get(struct rte_eth_dev *dev,
109                                 struct rte_eth_dev_info *dev_info);
110 static int i40evf_dev_link_update(struct rte_eth_dev *dev,
111                                   int wait_to_complete);
112 static int i40evf_dev_stats_get(struct rte_eth_dev *dev,
113                                 struct rte_eth_stats *stats);
114 static int i40evf_dev_xstats_get(struct rte_eth_dev *dev,
115                                  struct rte_eth_xstat *xstats, unsigned n);
116 static int i40evf_dev_xstats_get_names(struct rte_eth_dev *dev,
117                                        struct rte_eth_xstat_name *xstats_names,
118                                        unsigned limit);
119 static void i40evf_dev_xstats_reset(struct rte_eth_dev *dev);
120 static int i40evf_vlan_filter_set(struct rte_eth_dev *dev,
121                                   uint16_t vlan_id, int on);
122 static int i40evf_vlan_offload_set(struct rte_eth_dev *dev, int mask);
123 static void i40evf_dev_close(struct rte_eth_dev *dev);
124 static int  i40evf_dev_reset(struct rte_eth_dev *dev);
125 static void i40evf_dev_promiscuous_enable(struct rte_eth_dev *dev);
126 static void i40evf_dev_promiscuous_disable(struct rte_eth_dev *dev);
127 static void i40evf_dev_allmulticast_enable(struct rte_eth_dev *dev);
128 static void i40evf_dev_allmulticast_disable(struct rte_eth_dev *dev);
129 static int i40evf_init_vlan(struct rte_eth_dev *dev);
130 static int i40evf_dev_rx_queue_start(struct rte_eth_dev *dev,
131                                      uint16_t rx_queue_id);
132 static int i40evf_dev_rx_queue_stop(struct rte_eth_dev *dev,
133                                     uint16_t rx_queue_id);
134 static int i40evf_dev_tx_queue_start(struct rte_eth_dev *dev,
135                                      uint16_t tx_queue_id);
136 static int i40evf_dev_tx_queue_stop(struct rte_eth_dev *dev,
137                                     uint16_t tx_queue_id);
138 static int i40evf_add_mac_addr(struct rte_eth_dev *dev,
139                                struct ether_addr *addr,
140                                uint32_t index,
141                                uint32_t pool);
142 static void i40evf_del_mac_addr(struct rte_eth_dev *dev, uint32_t index);
143 static int i40evf_dev_rss_reta_update(struct rte_eth_dev *dev,
144                         struct rte_eth_rss_reta_entry64 *reta_conf,
145                         uint16_t reta_size);
146 static int i40evf_dev_rss_reta_query(struct rte_eth_dev *dev,
147                         struct rte_eth_rss_reta_entry64 *reta_conf,
148                         uint16_t reta_size);
149 static int i40evf_config_rss(struct i40e_vf *vf);
150 static int i40evf_dev_rss_hash_update(struct rte_eth_dev *dev,
151                                       struct rte_eth_rss_conf *rss_conf);
152 static int i40evf_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
153                                         struct rte_eth_rss_conf *rss_conf);
154 static int i40evf_dev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu);
155 static void i40evf_set_default_mac_addr(struct rte_eth_dev *dev,
156                                         struct ether_addr *mac_addr);
157 static int
158 i40evf_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id);
159 static int
160 i40evf_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id);
161 static void i40evf_handle_pf_event(struct rte_eth_dev *dev,
162                                    uint8_t *msg,
163                                    uint16_t msglen);
164
165 /* Default hash key buffer for RSS */
166 static uint32_t rss_key_default[I40E_VFQF_HKEY_MAX_INDEX + 1];
167
168 struct rte_i40evf_xstats_name_off {
169         char name[RTE_ETH_XSTATS_NAME_SIZE];
170         unsigned offset;
171 };
172
173 static const struct rte_i40evf_xstats_name_off rte_i40evf_stats_strings[] = {
174         {"rx_bytes", offsetof(struct i40e_eth_stats, rx_bytes)},
175         {"rx_unicast_packets", offsetof(struct i40e_eth_stats, rx_unicast)},
176         {"rx_multicast_packets", offsetof(struct i40e_eth_stats, rx_multicast)},
177         {"rx_broadcast_packets", offsetof(struct i40e_eth_stats, rx_broadcast)},
178         {"rx_dropped_packets", offsetof(struct i40e_eth_stats, rx_discards)},
179         {"rx_unknown_protocol_packets", offsetof(struct i40e_eth_stats,
180                 rx_unknown_protocol)},
181         {"tx_bytes", offsetof(struct i40e_eth_stats, tx_bytes)},
182         {"tx_unicast_packets", offsetof(struct i40e_eth_stats, tx_unicast)},
183         {"tx_multicast_packets", offsetof(struct i40e_eth_stats, tx_multicast)},
184         {"tx_broadcast_packets", offsetof(struct i40e_eth_stats, tx_broadcast)},
185         {"tx_dropped_packets", offsetof(struct i40e_eth_stats, tx_discards)},
186         {"tx_error_packets", offsetof(struct i40e_eth_stats, tx_errors)},
187 };
188
189 #define I40EVF_NB_XSTATS (sizeof(rte_i40evf_stats_strings) / \
190                 sizeof(rte_i40evf_stats_strings[0]))
191
192 static const struct eth_dev_ops i40evf_eth_dev_ops = {
193         .dev_configure        = i40evf_dev_configure,
194         .dev_start            = i40evf_dev_start,
195         .dev_stop             = i40evf_dev_stop,
196         .promiscuous_enable   = i40evf_dev_promiscuous_enable,
197         .promiscuous_disable  = i40evf_dev_promiscuous_disable,
198         .allmulticast_enable  = i40evf_dev_allmulticast_enable,
199         .allmulticast_disable = i40evf_dev_allmulticast_disable,
200         .link_update          = i40evf_dev_link_update,
201         .stats_get            = i40evf_dev_stats_get,
202         .stats_reset          = i40evf_dev_xstats_reset,
203         .xstats_get           = i40evf_dev_xstats_get,
204         .xstats_get_names     = i40evf_dev_xstats_get_names,
205         .xstats_reset         = i40evf_dev_xstats_reset,
206         .dev_close            = i40evf_dev_close,
207         .dev_reset            = i40evf_dev_reset,
208         .dev_infos_get        = i40evf_dev_info_get,
209         .dev_supported_ptypes_get = i40e_dev_supported_ptypes_get,
210         .vlan_filter_set      = i40evf_vlan_filter_set,
211         .vlan_offload_set     = i40evf_vlan_offload_set,
212         .rx_queue_start       = i40evf_dev_rx_queue_start,
213         .rx_queue_stop        = i40evf_dev_rx_queue_stop,
214         .tx_queue_start       = i40evf_dev_tx_queue_start,
215         .tx_queue_stop        = i40evf_dev_tx_queue_stop,
216         .rx_queue_setup       = i40e_dev_rx_queue_setup,
217         .rx_queue_release     = i40e_dev_rx_queue_release,
218         .rx_queue_intr_enable = i40evf_dev_rx_queue_intr_enable,
219         .rx_queue_intr_disable = i40evf_dev_rx_queue_intr_disable,
220         .rx_descriptor_done   = i40e_dev_rx_descriptor_done,
221         .rx_descriptor_status = i40e_dev_rx_descriptor_status,
222         .tx_descriptor_status = i40e_dev_tx_descriptor_status,
223         .tx_queue_setup       = i40e_dev_tx_queue_setup,
224         .tx_queue_release     = i40e_dev_tx_queue_release,
225         .rx_queue_count       = i40e_dev_rx_queue_count,
226         .rxq_info_get         = i40e_rxq_info_get,
227         .txq_info_get         = i40e_txq_info_get,
228         .mac_addr_add         = i40evf_add_mac_addr,
229         .mac_addr_remove      = i40evf_del_mac_addr,
230         .reta_update          = i40evf_dev_rss_reta_update,
231         .reta_query           = i40evf_dev_rss_reta_query,
232         .rss_hash_update      = i40evf_dev_rss_hash_update,
233         .rss_hash_conf_get    = i40evf_dev_rss_hash_conf_get,
234         .mtu_set              = i40evf_dev_mtu_set,
235         .mac_addr_set         = i40evf_set_default_mac_addr,
236 };
237
238 /*
239  * Read data in admin queue to get msg from pf driver
240  */
241 static enum i40evf_aq_result
242 i40evf_read_pfmsg(struct rte_eth_dev *dev, struct i40evf_arq_msg_info *data)
243 {
244         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
245         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
246         struct i40e_arq_event_info event;
247         enum virtchnl_ops opcode;
248         enum i40e_status_code retval;
249         int ret;
250         enum i40evf_aq_result result = I40EVF_MSG_NON;
251
252         event.buf_len = data->buf_len;
253         event.msg_buf = data->msg;
254         ret = i40e_clean_arq_element(hw, &event, NULL);
255         /* Can't read any msg from adminQ */
256         if (ret) {
257                 if (ret != I40E_ERR_ADMIN_QUEUE_NO_WORK)
258                         result = I40EVF_MSG_ERR;
259                 return result;
260         }
261
262         opcode = (enum virtchnl_ops)rte_le_to_cpu_32(event.desc.cookie_high);
263         retval = (enum i40e_status_code)rte_le_to_cpu_32(event.desc.cookie_low);
264         /* pf sys event */
265         if (opcode == VIRTCHNL_OP_EVENT) {
266                 struct virtchnl_pf_event *vpe =
267                         (struct virtchnl_pf_event *)event.msg_buf;
268
269                 result = I40EVF_MSG_SYS;
270                 switch (vpe->event) {
271                 case VIRTCHNL_EVENT_LINK_CHANGE:
272                         vf->link_up =
273                                 vpe->event_data.link_event.link_status;
274                         vf->link_speed =
275                                 vpe->event_data.link_event.link_speed;
276                         vf->pend_msg |= PFMSG_LINK_CHANGE;
277                         PMD_DRV_LOG(INFO, "Link status update:%s",
278                                     vf->link_up ? "up" : "down");
279                         break;
280                 case VIRTCHNL_EVENT_RESET_IMPENDING:
281                         vf->vf_reset = true;
282                         vf->pend_msg |= PFMSG_RESET_IMPENDING;
283                         PMD_DRV_LOG(INFO, "vf is reseting");
284                         break;
285                 case VIRTCHNL_EVENT_PF_DRIVER_CLOSE:
286                         vf->dev_closed = true;
287                         vf->pend_msg |= PFMSG_DRIVER_CLOSE;
288                         PMD_DRV_LOG(INFO, "PF driver closed");
289                         break;
290                 default:
291                         PMD_DRV_LOG(ERR, "%s: Unknown event %d from pf",
292                                     __func__, vpe->event);
293                 }
294         } else {
295                 /* async reply msg on command issued by vf previously */
296                 result = I40EVF_MSG_CMD;
297                 /* Actual data length read from PF */
298                 data->msg_len = event.msg_len;
299         }
300
301         data->result = retval;
302         data->ops = opcode;
303
304         return result;
305 }
306
307 /**
308  * clear current command. Only call in case execute
309  * _atomic_set_cmd successfully.
310  */
311 static inline void
312 _clear_cmd(struct i40e_vf *vf)
313 {
314         rte_wmb();
315         vf->pend_cmd = VIRTCHNL_OP_UNKNOWN;
316 }
317
318 /*
319  * Check there is pending cmd in execution. If none, set new command.
320  */
321 static inline int
322 _atomic_set_cmd(struct i40e_vf *vf, enum virtchnl_ops ops)
323 {
324         int ret = rte_atomic32_cmpset(&vf->pend_cmd,
325                         VIRTCHNL_OP_UNKNOWN, ops);
326
327         if (!ret)
328                 PMD_DRV_LOG(ERR, "There is incomplete cmd %d", vf->pend_cmd);
329
330         return !ret;
331 }
332
333 #define MAX_TRY_TIMES 200
334 #define ASQ_DELAY_MS  10
335
336 static int
337 i40evf_execute_vf_cmd(struct rte_eth_dev *dev, struct vf_cmd_info *args)
338 {
339         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
340         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
341         struct i40evf_arq_msg_info info;
342         enum i40evf_aq_result ret;
343         int err, i = 0;
344
345         if (_atomic_set_cmd(vf, args->ops))
346                 return -1;
347
348         info.msg = args->out_buffer;
349         info.buf_len = args->out_size;
350         info.ops = VIRTCHNL_OP_UNKNOWN;
351         info.result = I40E_SUCCESS;
352
353         err = i40e_aq_send_msg_to_pf(hw, args->ops, I40E_SUCCESS,
354                      args->in_args, args->in_args_size, NULL);
355         if (err) {
356                 PMD_DRV_LOG(ERR, "fail to send cmd %d", args->ops);
357                 _clear_cmd(vf);
358                 return err;
359         }
360
361         switch (args->ops) {
362         case VIRTCHNL_OP_RESET_VF:
363                 /*no need to process in this function */
364                 err = 0;
365                 break;
366         case VIRTCHNL_OP_VERSION:
367         case VIRTCHNL_OP_GET_VF_RESOURCES:
368                 /* for init adminq commands, need to poll the response */
369                 err = -1;
370                 do {
371                         ret = i40evf_read_pfmsg(dev, &info);
372                         vf->cmd_retval = info.result;
373                         if (ret == I40EVF_MSG_CMD) {
374                                 err = 0;
375                                 break;
376                         } else if (ret == I40EVF_MSG_ERR)
377                                 break;
378                         rte_delay_ms(ASQ_DELAY_MS);
379                         /* If don't read msg or read sys event, continue */
380                 } while (i++ < MAX_TRY_TIMES);
381                 _clear_cmd(vf);
382                 break;
383
384         default:
385                 /* for other adminq in running time, waiting the cmd done flag */
386                 err = -1;
387                 do {
388                         if (vf->pend_cmd == VIRTCHNL_OP_UNKNOWN) {
389                                 err = 0;
390                                 break;
391                         }
392                         rte_delay_ms(ASQ_DELAY_MS);
393                         /* If don't read msg or read sys event, continue */
394                 } while (i++ < MAX_TRY_TIMES);
395                 /* If there's no response is received, clear command */
396                 if (i >= MAX_TRY_TIMES) {
397                         PMD_DRV_LOG(WARNING, "No response for %d", args->ops);
398                         _clear_cmd(vf);
399                 }
400                 break;
401         }
402
403         return err | vf->cmd_retval;
404 }
405
406 /*
407  * Check API version with sync wait until version read or fail from admin queue
408  */
409 static int
410 i40evf_check_api_version(struct rte_eth_dev *dev)
411 {
412         struct virtchnl_version_info version, *pver;
413         int err;
414         struct vf_cmd_info args;
415         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
416
417         version.major = VIRTCHNL_VERSION_MAJOR;
418         version.minor = VIRTCHNL_VERSION_MINOR;
419
420         args.ops = VIRTCHNL_OP_VERSION;
421         args.in_args = (uint8_t *)&version;
422         args.in_args_size = sizeof(version);
423         args.out_buffer = vf->aq_resp;
424         args.out_size = I40E_AQ_BUF_SZ;
425
426         err = i40evf_execute_vf_cmd(dev, &args);
427         if (err) {
428                 PMD_INIT_LOG(ERR, "fail to execute command OP_VERSION");
429                 return err;
430         }
431
432         pver = (struct virtchnl_version_info *)args.out_buffer;
433         vf->version_major = pver->major;
434         vf->version_minor = pver->minor;
435         if ((vf->version_major == VIRTCHNL_VERSION_MAJOR) &&
436                 (vf->version_minor <= VIRTCHNL_VERSION_MINOR))
437                 PMD_DRV_LOG(INFO, "Peer is Linux PF host");
438         else {
439                 PMD_INIT_LOG(ERR, "PF/VF API version mismatch:(%u.%u)-(%u.%u)",
440                                         vf->version_major, vf->version_minor,
441                                                 VIRTCHNL_VERSION_MAJOR,
442                                                 VIRTCHNL_VERSION_MINOR);
443                 return -1;
444         }
445
446         return 0;
447 }
448
449 static int
450 i40evf_get_vf_resource(struct rte_eth_dev *dev)
451 {
452         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
453         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
454         int err;
455         struct vf_cmd_info args;
456         uint32_t caps, len;
457
458         args.ops = VIRTCHNL_OP_GET_VF_RESOURCES;
459         args.out_buffer = vf->aq_resp;
460         args.out_size = I40E_AQ_BUF_SZ;
461         if (PF_IS_V11(vf)) {
462                 caps = VIRTCHNL_VF_OFFLOAD_L2 |
463                        VIRTCHNL_VF_OFFLOAD_RSS_AQ |
464                        VIRTCHNL_VF_OFFLOAD_RSS_REG |
465                        VIRTCHNL_VF_OFFLOAD_VLAN |
466                        VIRTCHNL_VF_OFFLOAD_RX_POLLING;
467                 args.in_args = (uint8_t *)&caps;
468                 args.in_args_size = sizeof(caps);
469         } else {
470                 args.in_args = NULL;
471                 args.in_args_size = 0;
472         }
473         err = i40evf_execute_vf_cmd(dev, &args);
474
475         if (err) {
476                 PMD_DRV_LOG(ERR, "fail to execute command OP_GET_VF_RESOURCE");
477                 return err;
478         }
479
480         len =  sizeof(struct virtchnl_vf_resource) +
481                 I40E_MAX_VF_VSI * sizeof(struct virtchnl_vsi_resource);
482
483         rte_memcpy(vf->vf_res, args.out_buffer,
484                         RTE_MIN(args.out_size, len));
485         i40e_vf_parse_hw_config(hw, vf->vf_res);
486
487         return 0;
488 }
489
490 static int
491 i40evf_config_promisc(struct rte_eth_dev *dev,
492                       bool enable_unicast,
493                       bool enable_multicast)
494 {
495         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
496         int err;
497         struct vf_cmd_info args;
498         struct virtchnl_promisc_info promisc;
499
500         promisc.flags = 0;
501         promisc.vsi_id = vf->vsi_res->vsi_id;
502
503         if (enable_unicast)
504                 promisc.flags |= FLAG_VF_UNICAST_PROMISC;
505
506         if (enable_multicast)
507                 promisc.flags |= FLAG_VF_MULTICAST_PROMISC;
508
509         args.ops = VIRTCHNL_OP_CONFIG_PROMISCUOUS_MODE;
510         args.in_args = (uint8_t *)&promisc;
511         args.in_args_size = sizeof(promisc);
512         args.out_buffer = vf->aq_resp;
513         args.out_size = I40E_AQ_BUF_SZ;
514
515         err = i40evf_execute_vf_cmd(dev, &args);
516
517         if (err)
518                 PMD_DRV_LOG(ERR, "fail to execute command "
519                             "CONFIG_PROMISCUOUS_MODE");
520         return err;
521 }
522
523 static int
524 i40evf_enable_vlan_strip(struct rte_eth_dev *dev)
525 {
526         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
527         struct vf_cmd_info args;
528         int ret;
529
530         memset(&args, 0, sizeof(args));
531         args.ops = VIRTCHNL_OP_ENABLE_VLAN_STRIPPING;
532         args.in_args = NULL;
533         args.in_args_size = 0;
534         args.out_buffer = vf->aq_resp;
535         args.out_size = I40E_AQ_BUF_SZ;
536         ret = i40evf_execute_vf_cmd(dev, &args);
537         if (ret)
538                 PMD_DRV_LOG(ERR, "Failed to execute command of "
539                             "VIRTCHNL_OP_ENABLE_VLAN_STRIPPING");
540
541         return ret;
542 }
543
544 static int
545 i40evf_disable_vlan_strip(struct rte_eth_dev *dev)
546 {
547         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
548         struct vf_cmd_info args;
549         int ret;
550
551         memset(&args, 0, sizeof(args));
552         args.ops = VIRTCHNL_OP_DISABLE_VLAN_STRIPPING;
553         args.in_args = NULL;
554         args.in_args_size = 0;
555         args.out_buffer = vf->aq_resp;
556         args.out_size = I40E_AQ_BUF_SZ;
557         ret = i40evf_execute_vf_cmd(dev, &args);
558         if (ret)
559                 PMD_DRV_LOG(ERR, "Failed to execute command of "
560                             "VIRTCHNL_OP_DISABLE_VLAN_STRIPPING");
561
562         return ret;
563 }
564
565 static void
566 i40evf_fill_virtchnl_vsi_txq_info(struct virtchnl_txq_info *txq_info,
567                                   uint16_t vsi_id,
568                                   uint16_t queue_id,
569                                   uint16_t nb_txq,
570                                   struct i40e_tx_queue *txq)
571 {
572         txq_info->vsi_id = vsi_id;
573         txq_info->queue_id = queue_id;
574         if (queue_id < nb_txq) {
575                 txq_info->ring_len = txq->nb_tx_desc;
576                 txq_info->dma_ring_addr = txq->tx_ring_phys_addr;
577         }
578 }
579
580 static void
581 i40evf_fill_virtchnl_vsi_rxq_info(struct virtchnl_rxq_info *rxq_info,
582                                   uint16_t vsi_id,
583                                   uint16_t queue_id,
584                                   uint16_t nb_rxq,
585                                   uint32_t max_pkt_size,
586                                   struct i40e_rx_queue *rxq)
587 {
588         rxq_info->vsi_id = vsi_id;
589         rxq_info->queue_id = queue_id;
590         rxq_info->max_pkt_size = max_pkt_size;
591         if (queue_id < nb_rxq) {
592                 rxq_info->ring_len = rxq->nb_rx_desc;
593                 rxq_info->dma_ring_addr = rxq->rx_ring_phys_addr;
594                 rxq_info->databuffer_size =
595                         (rte_pktmbuf_data_room_size(rxq->mp) -
596                                 RTE_PKTMBUF_HEADROOM);
597         }
598 }
599
600 static int
601 i40evf_configure_vsi_queues(struct rte_eth_dev *dev)
602 {
603         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
604         struct i40e_rx_queue **rxq =
605                 (struct i40e_rx_queue **)dev->data->rx_queues;
606         struct i40e_tx_queue **txq =
607                 (struct i40e_tx_queue **)dev->data->tx_queues;
608         struct virtchnl_vsi_queue_config_info *vc_vqci;
609         struct virtchnl_queue_pair_info *vc_qpi;
610         struct vf_cmd_info args;
611         uint16_t i, nb_qp = vf->num_queue_pairs;
612         const uint32_t size =
613                 I40E_VIRTCHNL_CONFIG_VSI_QUEUES_SIZE(vc_vqci, nb_qp);
614         uint8_t buff[size];
615         int ret;
616
617         memset(buff, 0, sizeof(buff));
618         vc_vqci = (struct virtchnl_vsi_queue_config_info *)buff;
619         vc_vqci->vsi_id = vf->vsi_res->vsi_id;
620         vc_vqci->num_queue_pairs = nb_qp;
621
622         for (i = 0, vc_qpi = vc_vqci->qpair; i < nb_qp; i++, vc_qpi++) {
623                 i40evf_fill_virtchnl_vsi_txq_info(&vc_qpi->txq,
624                         vc_vqci->vsi_id, i, dev->data->nb_tx_queues, txq[i]);
625                 i40evf_fill_virtchnl_vsi_rxq_info(&vc_qpi->rxq,
626                         vc_vqci->vsi_id, i, dev->data->nb_rx_queues,
627                                         vf->max_pkt_len, rxq[i]);
628         }
629         memset(&args, 0, sizeof(args));
630         args.ops = VIRTCHNL_OP_CONFIG_VSI_QUEUES;
631         args.in_args = (uint8_t *)vc_vqci;
632         args.in_args_size = size;
633         args.out_buffer = vf->aq_resp;
634         args.out_size = I40E_AQ_BUF_SZ;
635         ret = i40evf_execute_vf_cmd(dev, &args);
636         if (ret)
637                 PMD_DRV_LOG(ERR, "Failed to execute command of "
638                         "VIRTCHNL_OP_CONFIG_VSI_QUEUES");
639
640         return ret;
641 }
642
643 static int
644 i40evf_config_irq_map(struct rte_eth_dev *dev)
645 {
646         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
647         struct vf_cmd_info args;
648         uint8_t cmd_buffer[sizeof(struct virtchnl_irq_map_info) + \
649                 sizeof(struct virtchnl_vector_map)];
650         struct virtchnl_irq_map_info *map_info;
651         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
652         struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
653         uint32_t vector_id;
654         int i, err;
655
656         if (rte_intr_allow_others(intr_handle))
657                 vector_id = I40EVF_VSI_DEFAULT_MSIX_INTR_LNX;
658         else
659                 vector_id = I40E_MISC_VEC_ID;
660
661         map_info = (struct virtchnl_irq_map_info *)cmd_buffer;
662         map_info->num_vectors = 1;
663         map_info->vecmap[0].rxitr_idx = I40E_ITR_INDEX_DEFAULT;
664         map_info->vecmap[0].vsi_id = vf->vsi_res->vsi_id;
665         /* Alway use default dynamic MSIX interrupt */
666         map_info->vecmap[0].vector_id = vector_id;
667         /* Don't map any tx queue */
668         map_info->vecmap[0].txq_map = 0;
669         map_info->vecmap[0].rxq_map = 0;
670         for (i = 0; i < dev->data->nb_rx_queues; i++) {
671                 map_info->vecmap[0].rxq_map |= 1 << i;
672                 if (rte_intr_dp_is_en(intr_handle))
673                         intr_handle->intr_vec[i] = vector_id;
674         }
675
676         args.ops = VIRTCHNL_OP_CONFIG_IRQ_MAP;
677         args.in_args = (u8 *)cmd_buffer;
678         args.in_args_size = sizeof(cmd_buffer);
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 execute command OP_ENABLE_QUEUES");
684
685         return err;
686 }
687
688 static int
689 i40evf_switch_queue(struct rte_eth_dev *dev, bool isrx, uint16_t qid,
690                                 bool on)
691 {
692         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
693         struct virtchnl_queue_select queue_select;
694         int err;
695         struct vf_cmd_info args;
696         memset(&queue_select, 0, sizeof(queue_select));
697         queue_select.vsi_id = vf->vsi_res->vsi_id;
698
699         if (isrx)
700                 queue_select.rx_queues |= 1 << qid;
701         else
702                 queue_select.tx_queues |= 1 << qid;
703
704         if (on)
705                 args.ops = VIRTCHNL_OP_ENABLE_QUEUES;
706         else
707                 args.ops = VIRTCHNL_OP_DISABLE_QUEUES;
708         args.in_args = (u8 *)&queue_select;
709         args.in_args_size = sizeof(queue_select);
710         args.out_buffer = vf->aq_resp;
711         args.out_size = I40E_AQ_BUF_SZ;
712         err = i40evf_execute_vf_cmd(dev, &args);
713         if (err)
714                 PMD_DRV_LOG(ERR, "fail to switch %s %u %s",
715                             isrx ? "RX" : "TX", qid, on ? "on" : "off");
716
717         return err;
718 }
719
720 static int
721 i40evf_start_queues(struct rte_eth_dev *dev)
722 {
723         struct rte_eth_dev_data *dev_data = dev->data;
724         int i;
725         struct i40e_rx_queue *rxq;
726         struct i40e_tx_queue *txq;
727
728         for (i = 0; i < dev->data->nb_rx_queues; i++) {
729                 rxq = dev_data->rx_queues[i];
730                 if (rxq->rx_deferred_start)
731                         continue;
732                 if (i40evf_dev_rx_queue_start(dev, i) != 0) {
733                         PMD_DRV_LOG(ERR, "Fail to start queue %u", i);
734                         return -1;
735                 }
736         }
737
738         for (i = 0; i < dev->data->nb_tx_queues; i++) {
739                 txq = dev_data->tx_queues[i];
740                 if (txq->tx_deferred_start)
741                         continue;
742                 if (i40evf_dev_tx_queue_start(dev, i) != 0) {
743                         PMD_DRV_LOG(ERR, "Fail to start queue %u", i);
744                         return -1;
745                 }
746         }
747
748         return 0;
749 }
750
751 static int
752 i40evf_stop_queues(struct rte_eth_dev *dev)
753 {
754         int i;
755
756         /* Stop TX queues first */
757         for (i = 0; i < dev->data->nb_tx_queues; i++) {
758                 if (i40evf_dev_tx_queue_stop(dev, i) != 0) {
759                         PMD_DRV_LOG(ERR, "Fail to stop queue %u", i);
760                         return -1;
761                 }
762         }
763
764         /* Then stop RX queues */
765         for (i = 0; i < dev->data->nb_rx_queues; i++) {
766                 if (i40evf_dev_rx_queue_stop(dev, i) != 0) {
767                         PMD_DRV_LOG(ERR, "Fail to stop queue %u", i);
768                         return -1;
769                 }
770         }
771
772         return 0;
773 }
774
775 static int
776 i40evf_add_mac_addr(struct rte_eth_dev *dev,
777                     struct ether_addr *addr,
778                     __rte_unused uint32_t index,
779                     __rte_unused uint32_t pool)
780 {
781         struct virtchnl_ether_addr_list *list;
782         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
783         uint8_t cmd_buffer[sizeof(struct virtchnl_ether_addr_list) + \
784                         sizeof(struct virtchnl_ether_addr)];
785         int err;
786         struct vf_cmd_info args;
787
788         if (is_zero_ether_addr(addr)) {
789                 PMD_DRV_LOG(ERR, "Invalid mac:%x:%x:%x:%x:%x:%x",
790                             addr->addr_bytes[0], addr->addr_bytes[1],
791                             addr->addr_bytes[2], addr->addr_bytes[3],
792                             addr->addr_bytes[4], addr->addr_bytes[5]);
793                 return I40E_ERR_INVALID_MAC_ADDR;
794         }
795
796         list = (struct virtchnl_ether_addr_list *)cmd_buffer;
797         list->vsi_id = vf->vsi_res->vsi_id;
798         list->num_elements = 1;
799         rte_memcpy(list->list[0].addr, addr->addr_bytes,
800                                         sizeof(addr->addr_bytes));
801
802         args.ops = VIRTCHNL_OP_ADD_ETH_ADDR;
803         args.in_args = cmd_buffer;
804         args.in_args_size = sizeof(cmd_buffer);
805         args.out_buffer = vf->aq_resp;
806         args.out_size = I40E_AQ_BUF_SZ;
807         err = i40evf_execute_vf_cmd(dev, &args);
808         if (err)
809                 PMD_DRV_LOG(ERR, "fail to execute command "
810                             "OP_ADD_ETHER_ADDRESS");
811         else
812                 vf->vsi.mac_num++;
813
814         return err;
815 }
816
817 static void
818 i40evf_del_mac_addr_by_addr(struct rte_eth_dev *dev,
819                             struct ether_addr *addr)
820 {
821         struct virtchnl_ether_addr_list *list;
822         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
823         uint8_t cmd_buffer[sizeof(struct virtchnl_ether_addr_list) + \
824                         sizeof(struct virtchnl_ether_addr)];
825         int err;
826         struct vf_cmd_info args;
827
828         if (i40e_validate_mac_addr(addr->addr_bytes) != I40E_SUCCESS) {
829                 PMD_DRV_LOG(ERR, "Invalid mac:%x-%x-%x-%x-%x-%x",
830                             addr->addr_bytes[0], addr->addr_bytes[1],
831                             addr->addr_bytes[2], addr->addr_bytes[3],
832                             addr->addr_bytes[4], addr->addr_bytes[5]);
833                 return;
834         }
835
836         list = (struct virtchnl_ether_addr_list *)cmd_buffer;
837         list->vsi_id = vf->vsi_res->vsi_id;
838         list->num_elements = 1;
839         rte_memcpy(list->list[0].addr, addr->addr_bytes,
840                         sizeof(addr->addr_bytes));
841
842         args.ops = VIRTCHNL_OP_DEL_ETH_ADDR;
843         args.in_args = cmd_buffer;
844         args.in_args_size = sizeof(cmd_buffer);
845         args.out_buffer = vf->aq_resp;
846         args.out_size = I40E_AQ_BUF_SZ;
847         err = i40evf_execute_vf_cmd(dev, &args);
848         if (err)
849                 PMD_DRV_LOG(ERR, "fail to execute command "
850                             "OP_DEL_ETHER_ADDRESS");
851         else
852                 vf->vsi.mac_num--;
853         return;
854 }
855
856 static void
857 i40evf_del_mac_addr(struct rte_eth_dev *dev, uint32_t index)
858 {
859         struct rte_eth_dev_data *data = dev->data;
860         struct ether_addr *addr;
861
862         addr = &data->mac_addrs[index];
863
864         i40evf_del_mac_addr_by_addr(dev, addr);
865 }
866
867 static int
868 i40evf_query_stats(struct rte_eth_dev *dev, struct i40e_eth_stats **pstats)
869 {
870         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
871         struct virtchnl_queue_select q_stats;
872         int err;
873         struct vf_cmd_info args;
874
875         memset(&q_stats, 0, sizeof(q_stats));
876         q_stats.vsi_id = vf->vsi_res->vsi_id;
877         args.ops = VIRTCHNL_OP_GET_STATS;
878         args.in_args = (u8 *)&q_stats;
879         args.in_args_size = sizeof(q_stats);
880         args.out_buffer = vf->aq_resp;
881         args.out_size = I40E_AQ_BUF_SZ;
882
883         err = i40evf_execute_vf_cmd(dev, &args);
884         if (err) {
885                 PMD_DRV_LOG(ERR, "fail to execute command OP_GET_STATS");
886                 *pstats = NULL;
887                 return err;
888         }
889         *pstats = (struct i40e_eth_stats *)args.out_buffer;
890         return 0;
891 }
892
893 static void
894 i40evf_stat_update_48(uint64_t *offset,
895                    uint64_t *stat)
896 {
897         if (*stat >= *offset)
898                 *stat = *stat - *offset;
899         else
900                 *stat = (uint64_t)((*stat +
901                         ((uint64_t)1 << I40E_48_BIT_WIDTH)) - *offset);
902
903         *stat &= I40E_48_BIT_MASK;
904 }
905
906 static void
907 i40evf_stat_update_32(uint64_t *offset,
908                    uint64_t *stat)
909 {
910         if (*stat >= *offset)
911                 *stat = (uint64_t)(*stat - *offset);
912         else
913                 *stat = (uint64_t)((*stat +
914                         ((uint64_t)1 << I40E_32_BIT_WIDTH)) - *offset);
915 }
916
917 static void
918 i40evf_update_stats(struct i40e_vsi *vsi,
919                                         struct i40e_eth_stats *nes)
920 {
921         struct i40e_eth_stats *oes = &vsi->eth_stats_offset;
922
923         i40evf_stat_update_48(&oes->rx_bytes,
924                             &nes->rx_bytes);
925         i40evf_stat_update_48(&oes->rx_unicast,
926                             &nes->rx_unicast);
927         i40evf_stat_update_48(&oes->rx_multicast,
928                             &nes->rx_multicast);
929         i40evf_stat_update_48(&oes->rx_broadcast,
930                             &nes->rx_broadcast);
931         i40evf_stat_update_32(&oes->rx_discards,
932                                 &nes->rx_discards);
933         i40evf_stat_update_32(&oes->rx_unknown_protocol,
934                             &nes->rx_unknown_protocol);
935         i40evf_stat_update_48(&oes->tx_bytes,
936                             &nes->tx_bytes);
937         i40evf_stat_update_48(&oes->tx_unicast,
938                             &nes->tx_unicast);
939         i40evf_stat_update_48(&oes->tx_multicast,
940                             &nes->tx_multicast);
941         i40evf_stat_update_48(&oes->tx_broadcast,
942                             &nes->tx_broadcast);
943         i40evf_stat_update_32(&oes->tx_errors, &nes->tx_errors);
944         i40evf_stat_update_32(&oes->tx_discards, &nes->tx_discards);
945 }
946
947 static void
948 i40evf_dev_xstats_reset(struct rte_eth_dev *dev)
949 {
950         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
951         struct i40e_eth_stats *pstats = NULL;
952
953         /* read stat values to clear hardware registers */
954         i40evf_query_stats(dev, &pstats);
955
956         /* set stats offset base on current values */
957         vf->vsi.eth_stats_offset = *pstats;
958 }
959
960 static int i40evf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
961                                       struct rte_eth_xstat_name *xstats_names,
962                                       __rte_unused unsigned limit)
963 {
964         unsigned i;
965
966         if (xstats_names != NULL)
967                 for (i = 0; i < I40EVF_NB_XSTATS; i++) {
968                         snprintf(xstats_names[i].name,
969                                 sizeof(xstats_names[i].name),
970                                 "%s", rte_i40evf_stats_strings[i].name);
971                 }
972         return I40EVF_NB_XSTATS;
973 }
974
975 static int i40evf_dev_xstats_get(struct rte_eth_dev *dev,
976                                  struct rte_eth_xstat *xstats, unsigned n)
977 {
978         int ret;
979         unsigned i;
980         struct i40e_eth_stats *pstats = NULL;
981         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
982         struct i40e_vsi *vsi = &vf->vsi;
983
984         if (n < I40EVF_NB_XSTATS)
985                 return I40EVF_NB_XSTATS;
986
987         ret = i40evf_query_stats(dev, &pstats);
988         if (ret != 0)
989                 return 0;
990
991         if (!xstats)
992                 return 0;
993
994         i40evf_update_stats(vsi, pstats);
995
996         /* loop over xstats array and values from pstats */
997         for (i = 0; i < I40EVF_NB_XSTATS; i++) {
998                 xstats[i].id = i;
999                 xstats[i].value = *(uint64_t *)(((char *)pstats) +
1000                         rte_i40evf_stats_strings[i].offset);
1001         }
1002
1003         return I40EVF_NB_XSTATS;
1004 }
1005
1006 static int
1007 i40evf_add_vlan(struct rte_eth_dev *dev, uint16_t vlanid)
1008 {
1009         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
1010         struct virtchnl_vlan_filter_list *vlan_list;
1011         uint8_t cmd_buffer[sizeof(struct virtchnl_vlan_filter_list) +
1012                                                         sizeof(uint16_t)];
1013         int err;
1014         struct vf_cmd_info args;
1015
1016         vlan_list = (struct virtchnl_vlan_filter_list *)cmd_buffer;
1017         vlan_list->vsi_id = vf->vsi_res->vsi_id;
1018         vlan_list->num_elements = 1;
1019         vlan_list->vlan_id[0] = vlanid;
1020
1021         args.ops = VIRTCHNL_OP_ADD_VLAN;
1022         args.in_args = (u8 *)&cmd_buffer;
1023         args.in_args_size = sizeof(cmd_buffer);
1024         args.out_buffer = vf->aq_resp;
1025         args.out_size = I40E_AQ_BUF_SZ;
1026         err = i40evf_execute_vf_cmd(dev, &args);
1027         if (err)
1028                 PMD_DRV_LOG(ERR, "fail to execute command OP_ADD_VLAN");
1029
1030         return err;
1031 }
1032
1033 static int
1034 i40evf_del_vlan(struct rte_eth_dev *dev, uint16_t vlanid)
1035 {
1036         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
1037         struct virtchnl_vlan_filter_list *vlan_list;
1038         uint8_t cmd_buffer[sizeof(struct virtchnl_vlan_filter_list) +
1039                                                         sizeof(uint16_t)];
1040         int err;
1041         struct vf_cmd_info args;
1042
1043         vlan_list = (struct virtchnl_vlan_filter_list *)cmd_buffer;
1044         vlan_list->vsi_id = vf->vsi_res->vsi_id;
1045         vlan_list->num_elements = 1;
1046         vlan_list->vlan_id[0] = vlanid;
1047
1048         args.ops = VIRTCHNL_OP_DEL_VLAN;
1049         args.in_args = (u8 *)&cmd_buffer;
1050         args.in_args_size = sizeof(cmd_buffer);
1051         args.out_buffer = vf->aq_resp;
1052         args.out_size = I40E_AQ_BUF_SZ;
1053         err = i40evf_execute_vf_cmd(dev, &args);
1054         if (err)
1055                 PMD_DRV_LOG(ERR, "fail to execute command OP_DEL_VLAN");
1056
1057         return err;
1058 }
1059
1060 static const struct rte_pci_id pci_id_i40evf_map[] = {
1061         { RTE_PCI_DEVICE(I40E_INTEL_VENDOR_ID, I40E_DEV_ID_VF) },
1062         { RTE_PCI_DEVICE(I40E_INTEL_VENDOR_ID, I40E_DEV_ID_VF_HV) },
1063         { RTE_PCI_DEVICE(I40E_INTEL_VENDOR_ID, I40E_DEV_ID_X722_A0_VF) },
1064         { RTE_PCI_DEVICE(I40E_INTEL_VENDOR_ID, I40E_DEV_ID_X722_VF) },
1065         { .vendor_id = 0, /* sentinel */ },
1066 };
1067
1068 static inline int
1069 i40evf_dev_atomic_write_link_status(struct rte_eth_dev *dev,
1070                                     struct rte_eth_link *link)
1071 {
1072         struct rte_eth_link *dst = &(dev->data->dev_link);
1073         struct rte_eth_link *src = link;
1074
1075         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
1076                                         *(uint64_t *)src) == 0)
1077                 return -1;
1078
1079         return 0;
1080 }
1081
1082 /* Disable IRQ0 */
1083 static inline void
1084 i40evf_disable_irq0(struct i40e_hw *hw)
1085 {
1086         /* Disable all interrupt types */
1087         I40E_WRITE_REG(hw, I40E_VFINT_ICR0_ENA1, 0);
1088         I40E_WRITE_REG(hw, I40E_VFINT_DYN_CTL01,
1089                        I40E_VFINT_DYN_CTL01_ITR_INDX_MASK);
1090         I40EVF_WRITE_FLUSH(hw);
1091 }
1092
1093 /* Enable IRQ0 */
1094 static inline void
1095 i40evf_enable_irq0(struct i40e_hw *hw)
1096 {
1097         /* Enable admin queue interrupt trigger */
1098         uint32_t val;
1099
1100         i40evf_disable_irq0(hw);
1101         val = I40E_READ_REG(hw, I40E_VFINT_ICR0_ENA1);
1102         val |= I40E_VFINT_ICR0_ENA1_ADMINQ_MASK |
1103                 I40E_VFINT_ICR0_ENA1_LINK_STAT_CHANGE_MASK;
1104         I40E_WRITE_REG(hw, I40E_VFINT_ICR0_ENA1, val);
1105
1106         I40E_WRITE_REG(hw, I40E_VFINT_DYN_CTL01,
1107                 I40E_VFINT_DYN_CTL01_INTENA_MASK |
1108                 I40E_VFINT_DYN_CTL01_CLEARPBA_MASK |
1109                 I40E_VFINT_DYN_CTL01_ITR_INDX_MASK);
1110
1111         I40EVF_WRITE_FLUSH(hw);
1112 }
1113
1114 static int
1115 i40evf_check_vf_reset_done(struct i40e_hw *hw)
1116 {
1117         int i, reset;
1118
1119         for (i = 0; i < MAX_RESET_WAIT_CNT; i++) {
1120                 reset = I40E_READ_REG(hw, I40E_VFGEN_RSTAT) &
1121                         I40E_VFGEN_RSTAT_VFR_STATE_MASK;
1122                 reset = reset >> I40E_VFGEN_RSTAT_VFR_STATE_SHIFT;
1123                 if (reset == VIRTCHNL_VFR_VFACTIVE ||
1124                     reset == VIRTCHNL_VFR_COMPLETED)
1125                         break;
1126                 rte_delay_ms(50);
1127         }
1128
1129         if (i >= MAX_RESET_WAIT_CNT)
1130                 return -1;
1131
1132         return 0;
1133 }
1134 static int
1135 i40evf_reset_vf(struct i40e_hw *hw)
1136 {
1137         int ret;
1138
1139         if (i40e_vf_reset(hw) != I40E_SUCCESS) {
1140                 PMD_INIT_LOG(ERR, "Reset VF NIC failed");
1141                 return -1;
1142         }
1143         /**
1144           * After issuing vf reset command to pf, pf won't necessarily
1145           * reset vf, it depends on what state it exactly is. If it's not
1146           * initialized yet, it won't have vf reset since it's in a certain
1147           * state. If not, it will try to reset. Even vf is reset, pf will
1148           * set I40E_VFGEN_RSTAT to COMPLETE first, then wait 10ms and set
1149           * it to ACTIVE. In this duration, vf may not catch the moment that
1150           * COMPLETE is set. So, for vf, we'll try to wait a long time.
1151           */
1152         rte_delay_ms(200);
1153
1154         ret = i40evf_check_vf_reset_done(hw);
1155         if (ret) {
1156                 PMD_INIT_LOG(ERR, "VF is still resetting");
1157                 return ret;
1158         }
1159
1160         return 0;
1161 }
1162
1163 static int
1164 i40evf_init_vf(struct rte_eth_dev *dev)
1165 {
1166         int i, err, bufsz;
1167         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1168         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
1169         uint16_t interval =
1170                 i40e_calc_itr_interval(I40E_QUEUE_ITR_INTERVAL_MAX);
1171
1172         vf->adapter = I40E_DEV_PRIVATE_TO_ADAPTER(dev->data->dev_private);
1173         vf->dev_data = dev->data;
1174         err = i40e_set_mac_type(hw);
1175         if (err) {
1176                 PMD_INIT_LOG(ERR, "set_mac_type failed: %d", err);
1177                 goto err;
1178         }
1179
1180         err = i40evf_check_vf_reset_done(hw);
1181         if (err)
1182                 goto err;
1183
1184         i40e_init_adminq_parameter(hw);
1185         err = i40e_init_adminq(hw);
1186         if (err) {
1187                 PMD_INIT_LOG(ERR, "init_adminq failed: %d", err);
1188                 goto err;
1189         }
1190
1191         /* Reset VF and wait until it's complete */
1192         if (i40evf_reset_vf(hw)) {
1193                 PMD_INIT_LOG(ERR, "reset NIC failed");
1194                 goto err_aq;
1195         }
1196
1197         /* VF reset, shutdown admin queue and initialize again */
1198         if (i40e_shutdown_adminq(hw) != I40E_SUCCESS) {
1199                 PMD_INIT_LOG(ERR, "i40e_shutdown_adminq failed");
1200                 goto err;
1201         }
1202
1203         i40e_init_adminq_parameter(hw);
1204         if (i40e_init_adminq(hw) != I40E_SUCCESS) {
1205                 PMD_INIT_LOG(ERR, "init_adminq failed");
1206                 goto err;
1207         }
1208
1209         vf->aq_resp = rte_zmalloc("vf_aq_resp", I40E_AQ_BUF_SZ, 0);
1210         if (!vf->aq_resp) {
1211                 PMD_INIT_LOG(ERR, "unable to allocate vf_aq_resp memory");
1212                 goto err_aq;
1213         }
1214         if (i40evf_check_api_version(dev) != 0) {
1215                 PMD_INIT_LOG(ERR, "check_api version failed");
1216                 goto err_api;
1217         }
1218         bufsz = sizeof(struct virtchnl_vf_resource) +
1219                 (I40E_MAX_VF_VSI * sizeof(struct virtchnl_vsi_resource));
1220         vf->vf_res = rte_zmalloc("vf_res", bufsz, 0);
1221         if (!vf->vf_res) {
1222                 PMD_INIT_LOG(ERR, "unable to allocate vf_res memory");
1223                 goto err_api;
1224         }
1225
1226         if (i40evf_get_vf_resource(dev) != 0) {
1227                 PMD_INIT_LOG(ERR, "i40evf_get_vf_config failed");
1228                 goto err_alloc;
1229         }
1230
1231         /* got VF config message back from PF, now we can parse it */
1232         for (i = 0; i < vf->vf_res->num_vsis; i++) {
1233                 if (vf->vf_res->vsi_res[i].vsi_type == VIRTCHNL_VSI_SRIOV)
1234                         vf->vsi_res = &vf->vf_res->vsi_res[i];
1235         }
1236
1237         if (!vf->vsi_res) {
1238                 PMD_INIT_LOG(ERR, "no LAN VSI found");
1239                 goto err_alloc;
1240         }
1241
1242         if (hw->mac.type == I40E_MAC_X722_VF)
1243                 vf->flags = I40E_FLAG_RSS_AQ_CAPABLE;
1244         vf->vsi.vsi_id = vf->vsi_res->vsi_id;
1245
1246         switch (vf->vsi_res->vsi_type) {
1247         case VIRTCHNL_VSI_SRIOV:
1248                 vf->vsi.type = I40E_VSI_SRIOV;
1249                 break;
1250         default:
1251                 vf->vsi.type = I40E_VSI_TYPE_UNKNOWN;
1252                 break;
1253         }
1254         vf->vsi.nb_qps = vf->vsi_res->num_queue_pairs;
1255         vf->vsi.adapter = I40E_DEV_PRIVATE_TO_ADAPTER(dev->data->dev_private);
1256
1257         /* Store the MAC address configured by host, or generate random one */
1258         if (is_valid_assigned_ether_addr((struct ether_addr *)hw->mac.addr))
1259                 vf->flags |= I40E_FLAG_VF_MAC_BY_PF;
1260         else
1261                 eth_random_addr(hw->mac.addr); /* Generate a random one */
1262
1263         I40E_WRITE_REG(hw, I40E_VFINT_DYN_CTL01,
1264                        (I40E_ITR_INDEX_DEFAULT <<
1265                         I40E_VFINT_DYN_CTL0_ITR_INDX_SHIFT) |
1266                        (interval <<
1267                         I40E_VFINT_DYN_CTL0_INTERVAL_SHIFT));
1268         I40EVF_WRITE_FLUSH(hw);
1269
1270         return 0;
1271
1272 err_alloc:
1273         rte_free(vf->vf_res);
1274         vf->vsi_res = NULL;
1275 err_api:
1276         rte_free(vf->aq_resp);
1277 err_aq:
1278         i40e_shutdown_adminq(hw); /* ignore error */
1279 err:
1280         return -1;
1281 }
1282
1283 static int
1284 i40evf_uninit_vf(struct rte_eth_dev *dev)
1285 {
1286         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
1287         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1288
1289         PMD_INIT_FUNC_TRACE();
1290
1291         if (hw->adapter_stopped == 0)
1292                 i40evf_dev_close(dev);
1293         rte_free(vf->vf_res);
1294         vf->vf_res = NULL;
1295         rte_free(vf->aq_resp);
1296         vf->aq_resp = NULL;
1297
1298         return 0;
1299 }
1300
1301 static void
1302 i40evf_handle_pf_event(struct rte_eth_dev *dev, uint8_t *msg,
1303                 __rte_unused uint16_t msglen)
1304 {
1305         struct virtchnl_pf_event *pf_msg =
1306                         (struct virtchnl_pf_event *)msg;
1307         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
1308
1309         switch (pf_msg->event) {
1310         case VIRTCHNL_EVENT_RESET_IMPENDING:
1311                 PMD_DRV_LOG(DEBUG, "VIRTCHNL_EVENT_RESET_IMPENDING event");
1312                 _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_RESET,
1313                                               NULL, NULL);
1314                 break;
1315         case VIRTCHNL_EVENT_LINK_CHANGE:
1316                 PMD_DRV_LOG(DEBUG, "VIRTCHNL_EVENT_LINK_CHANGE event");
1317                 vf->link_up = pf_msg->event_data.link_event.link_status;
1318                 vf->link_speed = pf_msg->event_data.link_event.link_speed;
1319                 break;
1320         case VIRTCHNL_EVENT_PF_DRIVER_CLOSE:
1321                 PMD_DRV_LOG(DEBUG, "VIRTCHNL_EVENT_PF_DRIVER_CLOSE event");
1322                 break;
1323         default:
1324                 PMD_DRV_LOG(ERR, " unknown event received %u", pf_msg->event);
1325                 break;
1326         }
1327 }
1328
1329 static void
1330 i40evf_handle_aq_msg(struct rte_eth_dev *dev)
1331 {
1332         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1333         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
1334         struct i40e_arq_event_info info;
1335         uint16_t pending, aq_opc;
1336         enum virtchnl_ops msg_opc;
1337         enum i40e_status_code msg_ret;
1338         int ret;
1339
1340         info.buf_len = I40E_AQ_BUF_SZ;
1341         if (!vf->aq_resp) {
1342                 PMD_DRV_LOG(ERR, "Buffer for adminq resp should not be NULL");
1343                 return;
1344         }
1345         info.msg_buf = vf->aq_resp;
1346
1347         pending = 1;
1348         while (pending) {
1349                 ret = i40e_clean_arq_element(hw, &info, &pending);
1350
1351                 if (ret != I40E_SUCCESS) {
1352                         PMD_DRV_LOG(INFO, "Failed to read msg from AdminQ,"
1353                                     "ret: %d", ret);
1354                         break;
1355                 }
1356                 aq_opc = rte_le_to_cpu_16(info.desc.opcode);
1357                 /* For the message sent from pf to vf, opcode is stored in
1358                  * cookie_high of struct i40e_aq_desc, while return error code
1359                  * are stored in cookie_low, Which is done by
1360                  * i40e_aq_send_msg_to_vf in PF driver.*/
1361                 msg_opc = (enum virtchnl_ops)rte_le_to_cpu_32(
1362                                                   info.desc.cookie_high);
1363                 msg_ret = (enum i40e_status_code)rte_le_to_cpu_32(
1364                                                   info.desc.cookie_low);
1365                 switch (aq_opc) {
1366                 case i40e_aqc_opc_send_msg_to_vf:
1367                         if (msg_opc == VIRTCHNL_OP_EVENT)
1368                                 /* process event*/
1369                                 i40evf_handle_pf_event(dev, info.msg_buf,
1370                                                        info.msg_len);
1371                         else {
1372                                 /* read message and it's expected one */
1373                                 if (msg_opc == vf->pend_cmd) {
1374                                         vf->cmd_retval = msg_ret;
1375                                         /* prevent compiler reordering */
1376                                         rte_compiler_barrier();
1377                                         _clear_cmd(vf);
1378                                 } else
1379                                         PMD_DRV_LOG(ERR, "command mismatch,"
1380                                                 "expect %u, get %u",
1381                                                 vf->pend_cmd, msg_opc);
1382                                 PMD_DRV_LOG(DEBUG, "adminq response is received,"
1383                                              " opcode = %d", msg_opc);
1384                         }
1385                         break;
1386                 default:
1387                         PMD_DRV_LOG(ERR, "Request %u is not supported yet",
1388                                     aq_opc);
1389                         break;
1390                 }
1391         }
1392 }
1393
1394 /**
1395  * Interrupt handler triggered by NIC  for handling
1396  * specific interrupt. Only adminq interrupt is processed in VF.
1397  *
1398  * @param handle
1399  *  Pointer to interrupt handle.
1400  * @param param
1401  *  The address of parameter (struct rte_eth_dev *) regsitered before.
1402  *
1403  * @return
1404  *  void
1405  */
1406 static void
1407 i40evf_dev_interrupt_handler(void *param)
1408 {
1409         struct rte_eth_dev *dev = (struct rte_eth_dev *)param;
1410         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1411         uint32_t icr0;
1412
1413         i40evf_disable_irq0(hw);
1414
1415         /* read out interrupt causes */
1416         icr0 = I40E_READ_REG(hw, I40E_VFINT_ICR01);
1417
1418         /* No interrupt event indicated */
1419         if (!(icr0 & I40E_VFINT_ICR01_INTEVENT_MASK)) {
1420                 PMD_DRV_LOG(DEBUG, "No interrupt event, nothing to do");
1421                 goto done;
1422         }
1423
1424         if (icr0 & I40E_VFINT_ICR01_ADMINQ_MASK) {
1425                 PMD_DRV_LOG(DEBUG, "ICR01_ADMINQ is reported");
1426                 i40evf_handle_aq_msg(dev);
1427         }
1428
1429         /* Link Status Change interrupt */
1430         if (icr0 & I40E_VFINT_ICR01_LINK_STAT_CHANGE_MASK)
1431                 PMD_DRV_LOG(DEBUG, "LINK_STAT_CHANGE is reported,"
1432                                    " do nothing");
1433
1434 done:
1435         i40evf_enable_irq0(hw);
1436         rte_intr_enable(dev->intr_handle);
1437 }
1438
1439 static int
1440 i40evf_dev_init(struct rte_eth_dev *eth_dev)
1441 {
1442         struct i40e_hw *hw
1443                 = I40E_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
1444         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1445
1446         PMD_INIT_FUNC_TRACE();
1447
1448         /* assign ops func pointer */
1449         eth_dev->dev_ops = &i40evf_eth_dev_ops;
1450         eth_dev->rx_pkt_burst = &i40e_recv_pkts;
1451         eth_dev->tx_pkt_burst = &i40e_xmit_pkts;
1452
1453         /*
1454          * For secondary processes, we don't initialise any further as primary
1455          * has already done this work.
1456          */
1457         if (rte_eal_process_type() != RTE_PROC_PRIMARY){
1458                 i40e_set_rx_function(eth_dev);
1459                 i40e_set_tx_function(eth_dev);
1460                 return 0;
1461         }
1462         i40e_set_default_ptype_table(eth_dev);
1463         i40e_set_default_pctype_table(eth_dev);
1464         rte_eth_copy_pci_info(eth_dev, pci_dev);
1465
1466         hw->vendor_id = pci_dev->id.vendor_id;
1467         hw->device_id = pci_dev->id.device_id;
1468         hw->subsystem_vendor_id = pci_dev->id.subsystem_vendor_id;
1469         hw->subsystem_device_id = pci_dev->id.subsystem_device_id;
1470         hw->bus.device = pci_dev->addr.devid;
1471         hw->bus.func = pci_dev->addr.function;
1472         hw->hw_addr = (void *)pci_dev->mem_resource[0].addr;
1473         hw->adapter_stopped = 0;
1474
1475         if(i40evf_init_vf(eth_dev) != 0) {
1476                 PMD_INIT_LOG(ERR, "Init vf failed");
1477                 return -1;
1478         }
1479
1480         /* register callback func to eal lib */
1481         rte_intr_callback_register(&pci_dev->intr_handle,
1482                 i40evf_dev_interrupt_handler, (void *)eth_dev);
1483
1484         /* enable uio intr after callback register */
1485         rte_intr_enable(&pci_dev->intr_handle);
1486
1487         /* configure and enable device interrupt */
1488         i40evf_enable_irq0(hw);
1489
1490         /* copy mac addr */
1491         eth_dev->data->mac_addrs = rte_zmalloc("i40evf_mac",
1492                                         ETHER_ADDR_LEN * I40E_NUM_MACADDR_MAX,
1493                                         0);
1494         if (eth_dev->data->mac_addrs == NULL) {
1495                 PMD_INIT_LOG(ERR, "Failed to allocate %d bytes needed to"
1496                                 " store MAC addresses",
1497                                 ETHER_ADDR_LEN * I40E_NUM_MACADDR_MAX);
1498                 return -ENOMEM;
1499         }
1500         ether_addr_copy((struct ether_addr *)hw->mac.addr,
1501                         &eth_dev->data->mac_addrs[0]);
1502
1503         return 0;
1504 }
1505
1506 static int
1507 i40evf_dev_uninit(struct rte_eth_dev *eth_dev)
1508 {
1509         PMD_INIT_FUNC_TRACE();
1510
1511         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1512                 return -EPERM;
1513
1514         eth_dev->dev_ops = NULL;
1515         eth_dev->rx_pkt_burst = NULL;
1516         eth_dev->tx_pkt_burst = NULL;
1517
1518         if (i40evf_uninit_vf(eth_dev) != 0) {
1519                 PMD_INIT_LOG(ERR, "i40evf_uninit_vf failed");
1520                 return -1;
1521         }
1522
1523         rte_free(eth_dev->data->mac_addrs);
1524         eth_dev->data->mac_addrs = NULL;
1525
1526         return 0;
1527 }
1528
1529 static int eth_i40evf_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1530         struct rte_pci_device *pci_dev)
1531 {
1532         return rte_eth_dev_pci_generic_probe(pci_dev,
1533                 sizeof(struct i40e_adapter), i40evf_dev_init);
1534 }
1535
1536 static int eth_i40evf_pci_remove(struct rte_pci_device *pci_dev)
1537 {
1538         return rte_eth_dev_pci_generic_remove(pci_dev, i40evf_dev_uninit);
1539 }
1540
1541 /*
1542  * virtual function driver struct
1543  */
1544 static struct rte_pci_driver rte_i40evf_pmd = {
1545         .id_table = pci_id_i40evf_map,
1546         .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_IOVA_AS_VA,
1547         .probe = eth_i40evf_pci_probe,
1548         .remove = eth_i40evf_pci_remove,
1549 };
1550
1551 RTE_PMD_REGISTER_PCI(net_i40e_vf, rte_i40evf_pmd);
1552 RTE_PMD_REGISTER_PCI_TABLE(net_i40e_vf, pci_id_i40evf_map);
1553 RTE_PMD_REGISTER_KMOD_DEP(net_i40e_vf, "* igb_uio | vfio-pci");
1554
1555 static int
1556 i40evf_dev_configure(struct rte_eth_dev *dev)
1557 {
1558         struct i40e_adapter *ad =
1559                 I40E_DEV_PRIVATE_TO_ADAPTER(dev->data->dev_private);
1560         struct rte_eth_conf *conf = &dev->data->dev_conf;
1561         struct i40e_vf *vf;
1562
1563         /* Initialize to TRUE. If any of Rx queues doesn't meet the bulk
1564          * allocation or vector Rx preconditions we will reset it.
1565          */
1566         ad->rx_bulk_alloc_allowed = true;
1567         ad->rx_vec_allowed = true;
1568         ad->tx_simple_allowed = true;
1569         ad->tx_vec_allowed = true;
1570
1571         /* For non-DPDK PF drivers, VF has no ability to disable HW
1572          * CRC strip, and is implicitly enabled by the PF.
1573          */
1574         if (!conf->rxmode.hw_strip_crc) {
1575                 vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
1576                 if ((vf->version_major == VIRTCHNL_VERSION_MAJOR) &&
1577                     (vf->version_minor <= VIRTCHNL_VERSION_MINOR)) {
1578                         /* Peer is running non-DPDK PF driver. */
1579                         PMD_INIT_LOG(ERR, "VF can't disable HW CRC Strip");
1580                         return -EINVAL;
1581                 }
1582         }
1583
1584         return i40evf_init_vlan(dev);
1585 }
1586
1587 static int
1588 i40evf_init_vlan(struct rte_eth_dev *dev)
1589 {
1590         /* Apply vlan offload setting */
1591         return i40evf_vlan_offload_set(dev, ETH_VLAN_STRIP_MASK);
1592 }
1593
1594 static int
1595 i40evf_vlan_offload_set(struct rte_eth_dev *dev, int mask)
1596 {
1597         struct rte_eth_conf *dev_conf = &dev->data->dev_conf;
1598
1599         /* Vlan stripping setting */
1600         if (mask & ETH_VLAN_STRIP_MASK) {
1601                 /* Enable or disable VLAN stripping */
1602                 if (dev_conf->rxmode.hw_vlan_strip)
1603                         i40evf_enable_vlan_strip(dev);
1604                 else
1605                         i40evf_disable_vlan_strip(dev);
1606         }
1607
1608         return 0;
1609 }
1610
1611 static int
1612 i40evf_dev_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1613 {
1614         struct i40e_rx_queue *rxq;
1615         int err = 0;
1616         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1617
1618         PMD_INIT_FUNC_TRACE();
1619
1620         if (rx_queue_id < dev->data->nb_rx_queues) {
1621                 rxq = dev->data->rx_queues[rx_queue_id];
1622
1623                 err = i40e_alloc_rx_queue_mbufs(rxq);
1624                 if (err) {
1625                         PMD_DRV_LOG(ERR, "Failed to allocate RX queue mbuf");
1626                         return err;
1627                 }
1628
1629                 rte_wmb();
1630
1631                 /* Init the RX tail register. */
1632                 I40E_PCI_REG_WRITE(rxq->qrx_tail, rxq->nb_rx_desc - 1);
1633                 I40EVF_WRITE_FLUSH(hw);
1634
1635                 /* Ready to switch the queue on */
1636                 err = i40evf_switch_queue(dev, TRUE, rx_queue_id, TRUE);
1637
1638                 if (err)
1639                         PMD_DRV_LOG(ERR, "Failed to switch RX queue %u on",
1640                                     rx_queue_id);
1641                 else
1642                         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1643         }
1644
1645         return err;
1646 }
1647
1648 static int
1649 i40evf_dev_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1650 {
1651         struct i40e_rx_queue *rxq;
1652         int err;
1653
1654         if (rx_queue_id < dev->data->nb_rx_queues) {
1655                 rxq = dev->data->rx_queues[rx_queue_id];
1656
1657                 err = i40evf_switch_queue(dev, TRUE, rx_queue_id, FALSE);
1658
1659                 if (err) {
1660                         PMD_DRV_LOG(ERR, "Failed to switch RX queue %u off",
1661                                     rx_queue_id);
1662                         return err;
1663                 }
1664
1665                 i40e_rx_queue_release_mbufs(rxq);
1666                 i40e_reset_rx_queue(rxq);
1667                 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1668         }
1669
1670         return 0;
1671 }
1672
1673 static int
1674 i40evf_dev_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1675 {
1676         int err = 0;
1677
1678         PMD_INIT_FUNC_TRACE();
1679
1680         if (tx_queue_id < dev->data->nb_tx_queues) {
1681
1682                 /* Ready to switch the queue on */
1683                 err = i40evf_switch_queue(dev, FALSE, tx_queue_id, TRUE);
1684
1685                 if (err)
1686                         PMD_DRV_LOG(ERR, "Failed to switch TX queue %u on",
1687                                     tx_queue_id);
1688                 else
1689                         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1690         }
1691
1692         return err;
1693 }
1694
1695 static int
1696 i40evf_dev_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1697 {
1698         struct i40e_tx_queue *txq;
1699         int err;
1700
1701         if (tx_queue_id < dev->data->nb_tx_queues) {
1702                 txq = dev->data->tx_queues[tx_queue_id];
1703
1704                 err = i40evf_switch_queue(dev, FALSE, tx_queue_id, FALSE);
1705
1706                 if (err) {
1707                         PMD_DRV_LOG(ERR, "Failed to switch TX queue %u off",
1708                                     tx_queue_id);
1709                         return err;
1710                 }
1711
1712                 i40e_tx_queue_release_mbufs(txq);
1713                 i40e_reset_tx_queue(txq);
1714                 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1715         }
1716
1717         return 0;
1718 }
1719
1720 static int
1721 i40evf_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1722 {
1723         int ret;
1724
1725         if (on)
1726                 ret = i40evf_add_vlan(dev, vlan_id);
1727         else
1728                 ret = i40evf_del_vlan(dev,vlan_id);
1729
1730         return ret;
1731 }
1732
1733 static int
1734 i40evf_rxq_init(struct rte_eth_dev *dev, struct i40e_rx_queue *rxq)
1735 {
1736         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1737         struct rte_eth_dev_data *dev_data = dev->data;
1738         struct rte_pktmbuf_pool_private *mbp_priv;
1739         uint16_t buf_size, len;
1740
1741         rxq->qrx_tail = hw->hw_addr + I40E_QRX_TAIL1(rxq->queue_id);
1742         I40E_PCI_REG_WRITE(rxq->qrx_tail, rxq->nb_rx_desc - 1);
1743         I40EVF_WRITE_FLUSH(hw);
1744
1745         /* Calculate the maximum packet length allowed */
1746         mbp_priv = rte_mempool_get_priv(rxq->mp);
1747         buf_size = (uint16_t)(mbp_priv->mbuf_data_room_size -
1748                                         RTE_PKTMBUF_HEADROOM);
1749         rxq->hs_mode = i40e_header_split_none;
1750         rxq->rx_hdr_len = 0;
1751         rxq->rx_buf_len = RTE_ALIGN(buf_size, (1 << I40E_RXQ_CTX_DBUFF_SHIFT));
1752         len = rxq->rx_buf_len * I40E_MAX_CHAINED_RX_BUFFERS;
1753         rxq->max_pkt_len = RTE_MIN(len,
1754                 dev_data->dev_conf.rxmode.max_rx_pkt_len);
1755
1756         /**
1757          * Check if the jumbo frame and maximum packet length are set correctly
1758          */
1759         if (dev_data->dev_conf.rxmode.jumbo_frame == 1) {
1760                 if (rxq->max_pkt_len <= ETHER_MAX_LEN ||
1761                     rxq->max_pkt_len > I40E_FRAME_SIZE_MAX) {
1762                         PMD_DRV_LOG(ERR, "maximum packet length must be "
1763                                 "larger than %u and smaller than %u, as jumbo "
1764                                 "frame is enabled", (uint32_t)ETHER_MAX_LEN,
1765                                         (uint32_t)I40E_FRAME_SIZE_MAX);
1766                         return I40E_ERR_CONFIG;
1767                 }
1768         } else {
1769                 if (rxq->max_pkt_len < ETHER_MIN_LEN ||
1770                     rxq->max_pkt_len > ETHER_MAX_LEN) {
1771                         PMD_DRV_LOG(ERR, "maximum packet length must be "
1772                                 "larger than %u and smaller than %u, as jumbo "
1773                                 "frame is disabled", (uint32_t)ETHER_MIN_LEN,
1774                                                 (uint32_t)ETHER_MAX_LEN);
1775                         return I40E_ERR_CONFIG;
1776                 }
1777         }
1778
1779         if (dev_data->dev_conf.rxmode.enable_scatter ||
1780             (rxq->max_pkt_len + 2 * I40E_VLAN_TAG_SIZE) > buf_size) {
1781                 dev_data->scattered_rx = 1;
1782         }
1783
1784         return 0;
1785 }
1786
1787 static int
1788 i40evf_rx_init(struct rte_eth_dev *dev)
1789 {
1790         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
1791         uint16_t i;
1792         int ret = I40E_SUCCESS;
1793         struct i40e_rx_queue **rxq =
1794                 (struct i40e_rx_queue **)dev->data->rx_queues;
1795
1796         i40evf_config_rss(vf);
1797         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1798                 if (!rxq[i] || !rxq[i]->q_set)
1799                         continue;
1800                 ret = i40evf_rxq_init(dev, rxq[i]);
1801                 if (ret != I40E_SUCCESS)
1802                         break;
1803         }
1804         if (ret == I40E_SUCCESS)
1805                 i40e_set_rx_function(dev);
1806
1807         return ret;
1808 }
1809
1810 static void
1811 i40evf_tx_init(struct rte_eth_dev *dev)
1812 {
1813         uint16_t i;
1814         struct i40e_tx_queue **txq =
1815                 (struct i40e_tx_queue **)dev->data->tx_queues;
1816         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1817
1818         for (i = 0; i < dev->data->nb_tx_queues; i++)
1819                 txq[i]->qtx_tail = hw->hw_addr + I40E_QTX_TAIL1(i);
1820
1821         i40e_set_tx_function(dev);
1822 }
1823
1824 static inline void
1825 i40evf_enable_queues_intr(struct rte_eth_dev *dev)
1826 {
1827         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1828         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1829         struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
1830
1831         if (!rte_intr_allow_others(intr_handle)) {
1832                 I40E_WRITE_REG(hw,
1833                                I40E_VFINT_DYN_CTL01,
1834                                I40E_VFINT_DYN_CTL01_INTENA_MASK |
1835                                I40E_VFINT_DYN_CTL01_CLEARPBA_MASK |
1836                                I40E_VFINT_DYN_CTL01_ITR_INDX_MASK);
1837                 I40EVF_WRITE_FLUSH(hw);
1838                 return;
1839         }
1840
1841         I40EVF_WRITE_FLUSH(hw);
1842 }
1843
1844 static inline void
1845 i40evf_disable_queues_intr(struct rte_eth_dev *dev)
1846 {
1847         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1848         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1849         struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
1850
1851         if (!rte_intr_allow_others(intr_handle)) {
1852                 I40E_WRITE_REG(hw, I40E_VFINT_DYN_CTL01,
1853                                I40E_VFINT_DYN_CTL01_ITR_INDX_MASK);
1854                 I40EVF_WRITE_FLUSH(hw);
1855                 return;
1856         }
1857
1858         I40EVF_WRITE_FLUSH(hw);
1859 }
1860
1861 static int
1862 i40evf_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
1863 {
1864         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1865         struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
1866         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1867         uint16_t interval =
1868                 i40e_calc_itr_interval(RTE_LIBRTE_I40E_ITR_INTERVAL);
1869         uint16_t msix_intr;
1870
1871         msix_intr = intr_handle->intr_vec[queue_id];
1872         if (msix_intr == I40E_MISC_VEC_ID)
1873                 I40E_WRITE_REG(hw, I40E_VFINT_DYN_CTL01,
1874                                I40E_VFINT_DYN_CTL01_INTENA_MASK |
1875                                I40E_VFINT_DYN_CTL01_CLEARPBA_MASK |
1876                                (0 << I40E_VFINT_DYN_CTL01_ITR_INDX_SHIFT) |
1877                                (interval <<
1878                                 I40E_VFINT_DYN_CTL01_INTERVAL_SHIFT));
1879         else
1880                 I40E_WRITE_REG(hw,
1881                                I40E_VFINT_DYN_CTLN1(msix_intr -
1882                                                     I40E_RX_VEC_START),
1883                                I40E_VFINT_DYN_CTLN1_INTENA_MASK |
1884                                I40E_VFINT_DYN_CTLN1_CLEARPBA_MASK |
1885                                (0 << I40E_VFINT_DYN_CTLN1_ITR_INDX_SHIFT) |
1886                                (interval <<
1887                                 I40E_VFINT_DYN_CTLN1_INTERVAL_SHIFT));
1888
1889         I40EVF_WRITE_FLUSH(hw);
1890
1891         rte_intr_enable(&pci_dev->intr_handle);
1892
1893         return 0;
1894 }
1895
1896 static int
1897 i40evf_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
1898 {
1899         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1900         struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
1901         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1902         uint16_t msix_intr;
1903
1904         msix_intr = intr_handle->intr_vec[queue_id];
1905         if (msix_intr == I40E_MISC_VEC_ID)
1906                 I40E_WRITE_REG(hw, I40E_VFINT_DYN_CTL01, 0);
1907         else
1908                 I40E_WRITE_REG(hw,
1909                                I40E_VFINT_DYN_CTLN1(msix_intr -
1910                                                     I40E_RX_VEC_START),
1911                                0);
1912
1913         I40EVF_WRITE_FLUSH(hw);
1914
1915         return 0;
1916 }
1917
1918 static void
1919 i40evf_add_del_all_mac_addr(struct rte_eth_dev *dev, bool add)
1920 {
1921         struct virtchnl_ether_addr_list *list;
1922         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
1923         int err, i, j;
1924         int next_begin = 0;
1925         int begin = 0;
1926         uint32_t len;
1927         struct ether_addr *addr;
1928         struct vf_cmd_info args;
1929
1930         do {
1931                 j = 0;
1932                 len = sizeof(struct virtchnl_ether_addr_list);
1933                 for (i = begin; i < I40E_NUM_MACADDR_MAX; i++, next_begin++) {
1934                         if (is_zero_ether_addr(&dev->data->mac_addrs[i]))
1935                                 continue;
1936                         len += sizeof(struct virtchnl_ether_addr);
1937                         if (len >= I40E_AQ_BUF_SZ) {
1938                                 next_begin = i + 1;
1939                                 break;
1940                         }
1941                 }
1942
1943                 list = rte_zmalloc("i40evf_del_mac_buffer", len, 0);
1944                 if (!list) {
1945                         PMD_DRV_LOG(ERR, "fail to allocate memory");
1946                         return;
1947                 }
1948
1949                 for (i = begin; i < next_begin; i++) {
1950                         addr = &dev->data->mac_addrs[i];
1951                         if (is_zero_ether_addr(addr))
1952                                 continue;
1953                         rte_memcpy(list->list[j].addr, addr->addr_bytes,
1954                                          sizeof(addr->addr_bytes));
1955                         PMD_DRV_LOG(DEBUG, "add/rm mac:%x:%x:%x:%x:%x:%x",
1956                                     addr->addr_bytes[0], addr->addr_bytes[1],
1957                                     addr->addr_bytes[2], addr->addr_bytes[3],
1958                                     addr->addr_bytes[4], addr->addr_bytes[5]);
1959                         j++;
1960                 }
1961                 list->vsi_id = vf->vsi_res->vsi_id;
1962                 list->num_elements = j;
1963                 args.ops = add ? VIRTCHNL_OP_ADD_ETH_ADDR :
1964                            VIRTCHNL_OP_DEL_ETH_ADDR;
1965                 args.in_args = (uint8_t *)list;
1966                 args.in_args_size = len;
1967                 args.out_buffer = vf->aq_resp;
1968                 args.out_size = I40E_AQ_BUF_SZ;
1969                 err = i40evf_execute_vf_cmd(dev, &args);
1970                 if (err) {
1971                         PMD_DRV_LOG(ERR, "fail to execute command %s",
1972                                     add ? "OP_ADD_ETHER_ADDRESS" :
1973                                     "OP_DEL_ETHER_ADDRESS");
1974                 } else {
1975                         if (add)
1976                                 vf->vsi.mac_num++;
1977                         else
1978                                 vf->vsi.mac_num--;
1979                 }
1980                 rte_free(list);
1981                 begin = next_begin;
1982         } while (begin < I40E_NUM_MACADDR_MAX);
1983 }
1984
1985 static int
1986 i40evf_dev_start(struct rte_eth_dev *dev)
1987 {
1988         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
1989         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1990         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1991         struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
1992         uint32_t intr_vector = 0;
1993
1994         PMD_INIT_FUNC_TRACE();
1995
1996         hw->adapter_stopped = 0;
1997
1998         vf->max_pkt_len = dev->data->dev_conf.rxmode.max_rx_pkt_len;
1999         vf->num_queue_pairs = RTE_MAX(dev->data->nb_rx_queues,
2000                                         dev->data->nb_tx_queues);
2001
2002         /* check and configure queue intr-vector mapping */
2003         if (dev->data->dev_conf.intr_conf.rxq != 0) {
2004                 intr_vector = dev->data->nb_rx_queues;
2005                 if (rte_intr_efd_enable(intr_handle, intr_vector))
2006                         return -1;
2007         }
2008
2009         if (rte_intr_dp_is_en(intr_handle) && !intr_handle->intr_vec) {
2010                 intr_handle->intr_vec =
2011                         rte_zmalloc("intr_vec",
2012                                     dev->data->nb_rx_queues * sizeof(int), 0);
2013                 if (!intr_handle->intr_vec) {
2014                         PMD_INIT_LOG(ERR, "Failed to allocate %d rx_queues"
2015                                      " intr_vec", dev->data->nb_rx_queues);
2016                         return -ENOMEM;
2017                 }
2018         }
2019
2020         if (i40evf_rx_init(dev) != 0){
2021                 PMD_DRV_LOG(ERR, "failed to do RX init");
2022                 return -1;
2023         }
2024
2025         i40evf_tx_init(dev);
2026
2027         if (i40evf_configure_vsi_queues(dev) != 0) {
2028                 PMD_DRV_LOG(ERR, "configure queues failed");
2029                 goto err_queue;
2030         }
2031         if (i40evf_config_irq_map(dev)) {
2032                 PMD_DRV_LOG(ERR, "config_irq_map failed");
2033                 goto err_queue;
2034         }
2035
2036         /* Set all mac addrs */
2037         i40evf_add_del_all_mac_addr(dev, TRUE);
2038
2039         if (i40evf_start_queues(dev) != 0) {
2040                 PMD_DRV_LOG(ERR, "enable queues failed");
2041                 goto err_mac;
2042         }
2043
2044         i40evf_enable_queues_intr(dev);
2045         return 0;
2046
2047 err_mac:
2048         i40evf_add_del_all_mac_addr(dev, FALSE);
2049 err_queue:
2050         return -1;
2051 }
2052
2053 static void
2054 i40evf_dev_stop(struct rte_eth_dev *dev)
2055 {
2056         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2057         struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
2058         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2059
2060         PMD_INIT_FUNC_TRACE();
2061
2062         if (hw->adapter_stopped == 1)
2063                 return;
2064         i40evf_stop_queues(dev);
2065         i40evf_disable_queues_intr(dev);
2066         i40e_dev_clear_queues(dev);
2067
2068         /* Clean datapath event and queue/vec mapping */
2069         rte_intr_efd_disable(intr_handle);
2070         if (intr_handle->intr_vec) {
2071                 rte_free(intr_handle->intr_vec);
2072                 intr_handle->intr_vec = NULL;
2073         }
2074         /* remove all mac addrs */
2075         i40evf_add_del_all_mac_addr(dev, FALSE);
2076         hw->adapter_stopped = 1;
2077
2078 }
2079
2080 static int
2081 i40evf_dev_link_update(struct rte_eth_dev *dev,
2082                        __rte_unused int wait_to_complete)
2083 {
2084         struct rte_eth_link new_link;
2085         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2086         /*
2087          * DPDK pf host provide interfacet to acquire link status
2088          * while Linux driver does not
2089          */
2090
2091         /* Linux driver PF host */
2092         switch (vf->link_speed) {
2093         case I40E_LINK_SPEED_100MB:
2094                 new_link.link_speed = ETH_SPEED_NUM_100M;
2095                 break;
2096         case I40E_LINK_SPEED_1GB:
2097                 new_link.link_speed = ETH_SPEED_NUM_1G;
2098                 break;
2099         case I40E_LINK_SPEED_10GB:
2100                 new_link.link_speed = ETH_SPEED_NUM_10G;
2101                 break;
2102         case I40E_LINK_SPEED_20GB:
2103                 new_link.link_speed = ETH_SPEED_NUM_20G;
2104                 break;
2105         case I40E_LINK_SPEED_25GB:
2106                 new_link.link_speed = ETH_SPEED_NUM_25G;
2107                 break;
2108         case I40E_LINK_SPEED_40GB:
2109                 new_link.link_speed = ETH_SPEED_NUM_40G;
2110                 break;
2111         default:
2112                 new_link.link_speed = ETH_SPEED_NUM_100M;
2113                 break;
2114         }
2115         /* full duplex only */
2116         new_link.link_duplex = ETH_LINK_FULL_DUPLEX;
2117         new_link.link_status = vf->link_up ? ETH_LINK_UP :
2118                                              ETH_LINK_DOWN;
2119         new_link.link_autoneg =
2120                 dev->data->dev_conf.link_speeds & ETH_LINK_SPEED_FIXED;
2121
2122         i40evf_dev_atomic_write_link_status(dev, &new_link);
2123
2124         return 0;
2125 }
2126
2127 static void
2128 i40evf_dev_promiscuous_enable(struct rte_eth_dev *dev)
2129 {
2130         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2131         int ret;
2132
2133         /* If enabled, just return */
2134         if (vf->promisc_unicast_enabled)
2135                 return;
2136
2137         ret = i40evf_config_promisc(dev, 1, vf->promisc_multicast_enabled);
2138         if (ret == 0)
2139                 vf->promisc_unicast_enabled = TRUE;
2140 }
2141
2142 static void
2143 i40evf_dev_promiscuous_disable(struct rte_eth_dev *dev)
2144 {
2145         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2146         int ret;
2147
2148         /* If disabled, just return */
2149         if (!vf->promisc_unicast_enabled)
2150                 return;
2151
2152         ret = i40evf_config_promisc(dev, 0, vf->promisc_multicast_enabled);
2153         if (ret == 0)
2154                 vf->promisc_unicast_enabled = FALSE;
2155 }
2156
2157 static void
2158 i40evf_dev_allmulticast_enable(struct rte_eth_dev *dev)
2159 {
2160         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2161         int ret;
2162
2163         /* If enabled, just return */
2164         if (vf->promisc_multicast_enabled)
2165                 return;
2166
2167         ret = i40evf_config_promisc(dev, vf->promisc_unicast_enabled, 1);
2168         if (ret == 0)
2169                 vf->promisc_multicast_enabled = TRUE;
2170 }
2171
2172 static void
2173 i40evf_dev_allmulticast_disable(struct rte_eth_dev *dev)
2174 {
2175         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2176         int ret;
2177
2178         /* If enabled, just return */
2179         if (!vf->promisc_multicast_enabled)
2180                 return;
2181
2182         ret = i40evf_config_promisc(dev, vf->promisc_unicast_enabled, 0);
2183         if (ret == 0)
2184                 vf->promisc_multicast_enabled = FALSE;
2185 }
2186
2187 static void
2188 i40evf_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2189 {
2190         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2191
2192         memset(dev_info, 0, sizeof(*dev_info));
2193         dev_info->pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2194         dev_info->max_rx_queues = vf->vsi_res->num_queue_pairs;
2195         dev_info->max_tx_queues = vf->vsi_res->num_queue_pairs;
2196         dev_info->min_rx_bufsize = I40E_BUF_SIZE_MIN;
2197         dev_info->max_rx_pktlen = I40E_FRAME_SIZE_MAX;
2198         dev_info->hash_key_size = (I40E_VFQF_HKEY_MAX_INDEX + 1) * sizeof(uint32_t);
2199         dev_info->reta_size = ETH_RSS_RETA_SIZE_64;
2200         dev_info->flow_type_rss_offloads = vf->adapter->flow_types_mask;
2201         dev_info->max_mac_addrs = I40E_NUM_MACADDR_MAX;
2202         dev_info->rx_offload_capa =
2203                 DEV_RX_OFFLOAD_VLAN_STRIP |
2204                 DEV_RX_OFFLOAD_QINQ_STRIP |
2205                 DEV_RX_OFFLOAD_IPV4_CKSUM |
2206                 DEV_RX_OFFLOAD_UDP_CKSUM |
2207                 DEV_RX_OFFLOAD_TCP_CKSUM;
2208         dev_info->tx_offload_capa =
2209                 DEV_TX_OFFLOAD_VLAN_INSERT |
2210                 DEV_TX_OFFLOAD_QINQ_INSERT |
2211                 DEV_TX_OFFLOAD_IPV4_CKSUM |
2212                 DEV_TX_OFFLOAD_UDP_CKSUM |
2213                 DEV_TX_OFFLOAD_TCP_CKSUM |
2214                 DEV_TX_OFFLOAD_SCTP_CKSUM;
2215
2216         dev_info->default_rxconf = (struct rte_eth_rxconf) {
2217                 .rx_thresh = {
2218                         .pthresh = I40E_DEFAULT_RX_PTHRESH,
2219                         .hthresh = I40E_DEFAULT_RX_HTHRESH,
2220                         .wthresh = I40E_DEFAULT_RX_WTHRESH,
2221                 },
2222                 .rx_free_thresh = I40E_DEFAULT_RX_FREE_THRESH,
2223                 .rx_drop_en = 0,
2224         };
2225
2226         dev_info->default_txconf = (struct rte_eth_txconf) {
2227                 .tx_thresh = {
2228                         .pthresh = I40E_DEFAULT_TX_PTHRESH,
2229                         .hthresh = I40E_DEFAULT_TX_HTHRESH,
2230                         .wthresh = I40E_DEFAULT_TX_WTHRESH,
2231                 },
2232                 .tx_free_thresh = I40E_DEFAULT_TX_FREE_THRESH,
2233                 .tx_rs_thresh = I40E_DEFAULT_TX_RSBIT_THRESH,
2234                 .txq_flags = ETH_TXQ_FLAGS_NOMULTSEGS |
2235                                 ETH_TXQ_FLAGS_NOOFFLOADS,
2236         };
2237
2238         dev_info->rx_desc_lim = (struct rte_eth_desc_lim) {
2239                 .nb_max = I40E_MAX_RING_DESC,
2240                 .nb_min = I40E_MIN_RING_DESC,
2241                 .nb_align = I40E_ALIGN_RING_DESC,
2242         };
2243
2244         dev_info->tx_desc_lim = (struct rte_eth_desc_lim) {
2245                 .nb_max = I40E_MAX_RING_DESC,
2246                 .nb_min = I40E_MIN_RING_DESC,
2247                 .nb_align = I40E_ALIGN_RING_DESC,
2248         };
2249 }
2250
2251 static int
2252 i40evf_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
2253 {
2254         int ret;
2255         struct i40e_eth_stats *pstats = NULL;
2256         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2257         struct i40e_vsi *vsi = &vf->vsi;
2258
2259         ret = i40evf_query_stats(dev, &pstats);
2260         if (ret == 0) {
2261                 i40evf_update_stats(vsi, pstats);
2262
2263                 stats->ipackets = pstats->rx_unicast + pstats->rx_multicast +
2264                                                 pstats->rx_broadcast;
2265                 stats->opackets = pstats->tx_broadcast + pstats->tx_multicast +
2266                                                 pstats->tx_unicast;
2267                 stats->imissed = pstats->rx_discards;
2268                 stats->oerrors = pstats->tx_errors + pstats->tx_discards;
2269                 stats->ibytes = pstats->rx_bytes;
2270                 stats->obytes = pstats->tx_bytes;
2271         } else {
2272                 PMD_DRV_LOG(ERR, "Get statistics failed");
2273         }
2274         return ret;
2275 }
2276
2277 static void
2278 i40evf_dev_close(struct rte_eth_dev *dev)
2279 {
2280         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2281         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2282         struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
2283
2284         i40evf_dev_stop(dev);
2285         i40e_dev_free_queues(dev);
2286         i40evf_reset_vf(hw);
2287         i40e_shutdown_adminq(hw);
2288         /* disable uio intr before callback unregister */
2289         rte_intr_disable(intr_handle);
2290
2291         /* unregister callback func from eal lib */
2292         rte_intr_callback_unregister(intr_handle,
2293                                      i40evf_dev_interrupt_handler, dev);
2294         i40evf_disable_irq0(hw);
2295 }
2296
2297 /*
2298  * Reset VF device only to re-initialize resources in PMD layer
2299  */
2300 static int
2301 i40evf_dev_reset(struct rte_eth_dev *dev)
2302 {
2303         int ret;
2304
2305         ret = i40evf_dev_uninit(dev);
2306         if (ret)
2307                 return ret;
2308
2309         ret = i40evf_dev_init(dev);
2310
2311         return ret;
2312 }
2313
2314 static int
2315 i40evf_get_rss_lut(struct i40e_vsi *vsi, uint8_t *lut, uint16_t lut_size)
2316 {
2317         struct i40e_vf *vf = I40E_VSI_TO_VF(vsi);
2318         struct i40e_hw *hw = I40E_VSI_TO_HW(vsi);
2319         int ret;
2320
2321         if (!lut)
2322                 return -EINVAL;
2323
2324         if (vf->flags & I40E_FLAG_RSS_AQ_CAPABLE) {
2325                 ret = i40e_aq_get_rss_lut(hw, vsi->vsi_id, FALSE,
2326                                           lut, lut_size);
2327                 if (ret) {
2328                         PMD_DRV_LOG(ERR, "Failed to get RSS lookup table");
2329                         return ret;
2330                 }
2331         } else {
2332                 uint32_t *lut_dw = (uint32_t *)lut;
2333                 uint16_t i, lut_size_dw = lut_size / 4;
2334
2335                 for (i = 0; i < lut_size_dw; i++)
2336                         lut_dw[i] = I40E_READ_REG(hw, I40E_VFQF_HLUT(i));
2337         }
2338
2339         return 0;
2340 }
2341
2342 static int
2343 i40evf_set_rss_lut(struct i40e_vsi *vsi, uint8_t *lut, uint16_t lut_size)
2344 {
2345         struct i40e_vf *vf;
2346         struct i40e_hw *hw;
2347         int ret;
2348
2349         if (!vsi || !lut)
2350                 return -EINVAL;
2351
2352         vf = I40E_VSI_TO_VF(vsi);
2353         hw = I40E_VSI_TO_HW(vsi);
2354
2355         if (vf->flags & I40E_FLAG_RSS_AQ_CAPABLE) {
2356                 ret = i40e_aq_set_rss_lut(hw, vsi->vsi_id, FALSE,
2357                                           lut, lut_size);
2358                 if (ret) {
2359                         PMD_DRV_LOG(ERR, "Failed to set RSS lookup table");
2360                         return ret;
2361                 }
2362         } else {
2363                 uint32_t *lut_dw = (uint32_t *)lut;
2364                 uint16_t i, lut_size_dw = lut_size / 4;
2365
2366                 for (i = 0; i < lut_size_dw; i++)
2367                         I40E_WRITE_REG(hw, I40E_VFQF_HLUT(i), lut_dw[i]);
2368                 I40EVF_WRITE_FLUSH(hw);
2369         }
2370
2371         return 0;
2372 }
2373
2374 static int
2375 i40evf_dev_rss_reta_update(struct rte_eth_dev *dev,
2376                            struct rte_eth_rss_reta_entry64 *reta_conf,
2377                            uint16_t reta_size)
2378 {
2379         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2380         uint8_t *lut;
2381         uint16_t i, idx, shift;
2382         int ret;
2383
2384         if (reta_size != ETH_RSS_RETA_SIZE_64) {
2385                 PMD_DRV_LOG(ERR, "The size of hash lookup table configured "
2386                         "(%d) doesn't match the number of hardware can "
2387                         "support (%d)", reta_size, ETH_RSS_RETA_SIZE_64);
2388                 return -EINVAL;
2389         }
2390
2391         lut = rte_zmalloc("i40e_rss_lut", reta_size, 0);
2392         if (!lut) {
2393                 PMD_DRV_LOG(ERR, "No memory can be allocated");
2394                 return -ENOMEM;
2395         }
2396         ret = i40evf_get_rss_lut(&vf->vsi, lut, reta_size);
2397         if (ret)
2398                 goto out;
2399         for (i = 0; i < reta_size; i++) {
2400                 idx = i / RTE_RETA_GROUP_SIZE;
2401                 shift = i % RTE_RETA_GROUP_SIZE;
2402                 if (reta_conf[idx].mask & (1ULL << shift))
2403                         lut[i] = reta_conf[idx].reta[shift];
2404         }
2405         ret = i40evf_set_rss_lut(&vf->vsi, lut, reta_size);
2406
2407 out:
2408         rte_free(lut);
2409
2410         return ret;
2411 }
2412
2413 static int
2414 i40evf_dev_rss_reta_query(struct rte_eth_dev *dev,
2415                           struct rte_eth_rss_reta_entry64 *reta_conf,
2416                           uint16_t reta_size)
2417 {
2418         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2419         uint16_t i, idx, shift;
2420         uint8_t *lut;
2421         int ret;
2422
2423         if (reta_size != ETH_RSS_RETA_SIZE_64) {
2424                 PMD_DRV_LOG(ERR, "The size of hash lookup table configured "
2425                         "(%d) doesn't match the number of hardware can "
2426                         "support (%d)", reta_size, ETH_RSS_RETA_SIZE_64);
2427                 return -EINVAL;
2428         }
2429
2430         lut = rte_zmalloc("i40e_rss_lut", reta_size, 0);
2431         if (!lut) {
2432                 PMD_DRV_LOG(ERR, "No memory can be allocated");
2433                 return -ENOMEM;
2434         }
2435
2436         ret = i40evf_get_rss_lut(&vf->vsi, lut, reta_size);
2437         if (ret)
2438                 goto out;
2439         for (i = 0; i < reta_size; i++) {
2440                 idx = i / RTE_RETA_GROUP_SIZE;
2441                 shift = i % RTE_RETA_GROUP_SIZE;
2442                 if (reta_conf[idx].mask & (1ULL << shift))
2443                         reta_conf[idx].reta[shift] = lut[i];
2444         }
2445
2446 out:
2447         rte_free(lut);
2448
2449         return ret;
2450 }
2451
2452 static int
2453 i40evf_set_rss_key(struct i40e_vsi *vsi, uint8_t *key, uint8_t key_len)
2454 {
2455         struct i40e_vf *vf = I40E_VSI_TO_VF(vsi);
2456         struct i40e_hw *hw = I40E_VSI_TO_HW(vsi);
2457         int ret = 0;
2458
2459         if (!key || key_len == 0) {
2460                 PMD_DRV_LOG(DEBUG, "No key to be configured");
2461                 return 0;
2462         } else if (key_len != (I40E_VFQF_HKEY_MAX_INDEX + 1) *
2463                 sizeof(uint32_t)) {
2464                 PMD_DRV_LOG(ERR, "Invalid key length %u", key_len);
2465                 return -EINVAL;
2466         }
2467
2468         if (vf->flags & I40E_FLAG_RSS_AQ_CAPABLE) {
2469                 struct i40e_aqc_get_set_rss_key_data *key_dw =
2470                         (struct i40e_aqc_get_set_rss_key_data *)key;
2471
2472                 ret = i40e_aq_set_rss_key(hw, vsi->vsi_id, key_dw);
2473                 if (ret)
2474                         PMD_INIT_LOG(ERR, "Failed to configure RSS key "
2475                                      "via AQ");
2476         } else {
2477                 uint32_t *hash_key = (uint32_t *)key;
2478                 uint16_t i;
2479
2480                 for (i = 0; i <= I40E_VFQF_HKEY_MAX_INDEX; i++)
2481                         i40e_write_rx_ctl(hw, I40E_VFQF_HKEY(i), hash_key[i]);
2482                 I40EVF_WRITE_FLUSH(hw);
2483         }
2484
2485         return ret;
2486 }
2487
2488 static int
2489 i40evf_get_rss_key(struct i40e_vsi *vsi, uint8_t *key, uint8_t *key_len)
2490 {
2491         struct i40e_vf *vf = I40E_VSI_TO_VF(vsi);
2492         struct i40e_hw *hw = I40E_VSI_TO_HW(vsi);
2493         int ret;
2494
2495         if (!key || !key_len)
2496                 return -EINVAL;
2497
2498         if (vf->flags & I40E_FLAG_RSS_AQ_CAPABLE) {
2499                 ret = i40e_aq_get_rss_key(hw, vsi->vsi_id,
2500                         (struct i40e_aqc_get_set_rss_key_data *)key);
2501                 if (ret) {
2502                         PMD_INIT_LOG(ERR, "Failed to get RSS key via AQ");
2503                         return ret;
2504                 }
2505         } else {
2506                 uint32_t *key_dw = (uint32_t *)key;
2507                 uint16_t i;
2508
2509                 for (i = 0; i <= I40E_VFQF_HKEY_MAX_INDEX; i++)
2510                         key_dw[i] = i40e_read_rx_ctl(hw, I40E_VFQF_HKEY(i));
2511         }
2512         *key_len = (I40E_VFQF_HKEY_MAX_INDEX + 1) * sizeof(uint32_t);
2513
2514         return 0;
2515 }
2516
2517 static int
2518 i40evf_hw_rss_hash_set(struct i40e_vf *vf, struct rte_eth_rss_conf *rss_conf)
2519 {
2520         struct i40e_hw *hw = I40E_VF_TO_HW(vf);
2521         uint64_t hena;
2522         int ret;
2523
2524         ret = i40evf_set_rss_key(&vf->vsi, rss_conf->rss_key,
2525                                  rss_conf->rss_key_len);
2526         if (ret)
2527                 return ret;
2528
2529         hena = i40e_config_hena(vf->adapter, rss_conf->rss_hf);
2530         i40e_write_rx_ctl(hw, I40E_VFQF_HENA(0), (uint32_t)hena);
2531         i40e_write_rx_ctl(hw, I40E_VFQF_HENA(1), (uint32_t)(hena >> 32));
2532         I40EVF_WRITE_FLUSH(hw);
2533
2534         return 0;
2535 }
2536
2537 static void
2538 i40evf_disable_rss(struct i40e_vf *vf)
2539 {
2540         struct i40e_hw *hw = I40E_VF_TO_HW(vf);
2541
2542         i40e_write_rx_ctl(hw, I40E_VFQF_HENA(0), 0);
2543         i40e_write_rx_ctl(hw, I40E_VFQF_HENA(1), 0);
2544         I40EVF_WRITE_FLUSH(hw);
2545 }
2546
2547 static int
2548 i40evf_config_rss(struct i40e_vf *vf)
2549 {
2550         struct i40e_hw *hw = I40E_VF_TO_HW(vf);
2551         struct rte_eth_rss_conf rss_conf;
2552         uint32_t i, j, lut = 0, nb_q = (I40E_VFQF_HLUT_MAX_INDEX + 1) * 4;
2553         uint16_t num;
2554
2555         if (vf->dev_data->dev_conf.rxmode.mq_mode != ETH_MQ_RX_RSS) {
2556                 i40evf_disable_rss(vf);
2557                 PMD_DRV_LOG(DEBUG, "RSS not configured");
2558                 return 0;
2559         }
2560
2561         num = RTE_MIN(vf->dev_data->nb_rx_queues, I40E_MAX_QP_NUM_PER_VF);
2562         /* Fill out the look up table */
2563         for (i = 0, j = 0; i < nb_q; i++, j++) {
2564                 if (j >= num)
2565                         j = 0;
2566                 lut = (lut << 8) | j;
2567                 if ((i & 3) == 3)
2568                         I40E_WRITE_REG(hw, I40E_VFQF_HLUT(i >> 2), lut);
2569         }
2570
2571         rss_conf = vf->dev_data->dev_conf.rx_adv_conf.rss_conf;
2572         if ((rss_conf.rss_hf & vf->adapter->flow_types_mask) == 0) {
2573                 i40evf_disable_rss(vf);
2574                 PMD_DRV_LOG(DEBUG, "No hash flag is set");
2575                 return 0;
2576         }
2577
2578         if (rss_conf.rss_key == NULL || rss_conf.rss_key_len <
2579                 (I40E_VFQF_HKEY_MAX_INDEX + 1) * sizeof(uint32_t)) {
2580                 /* Calculate the default hash key */
2581                 for (i = 0; i <= I40E_VFQF_HKEY_MAX_INDEX; i++)
2582                         rss_key_default[i] = (uint32_t)rte_rand();
2583                 rss_conf.rss_key = (uint8_t *)rss_key_default;
2584                 rss_conf.rss_key_len = (I40E_VFQF_HKEY_MAX_INDEX + 1) *
2585                         sizeof(uint32_t);
2586         }
2587
2588         return i40evf_hw_rss_hash_set(vf, &rss_conf);
2589 }
2590
2591 static int
2592 i40evf_dev_rss_hash_update(struct rte_eth_dev *dev,
2593                            struct rte_eth_rss_conf *rss_conf)
2594 {
2595         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2596         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2597         uint64_t rss_hf = rss_conf->rss_hf & vf->adapter->flow_types_mask;
2598         uint64_t hena;
2599
2600         hena = (uint64_t)i40e_read_rx_ctl(hw, I40E_VFQF_HENA(0));
2601         hena |= ((uint64_t)i40e_read_rx_ctl(hw, I40E_VFQF_HENA(1))) << 32;
2602
2603         if (!(hena & vf->adapter->pctypes_mask)) { /* RSS disabled */
2604                 if (rss_hf != 0) /* Enable RSS */
2605                         return -EINVAL;
2606                 return 0;
2607         }
2608
2609         /* RSS enabled */
2610         if (rss_hf == 0) /* Disable RSS */
2611                 return -EINVAL;
2612
2613         return i40evf_hw_rss_hash_set(vf, rss_conf);
2614 }
2615
2616 static int
2617 i40evf_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
2618                              struct rte_eth_rss_conf *rss_conf)
2619 {
2620         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2621         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2622         uint64_t hena;
2623
2624         i40evf_get_rss_key(&vf->vsi, rss_conf->rss_key,
2625                            &rss_conf->rss_key_len);
2626
2627         hena = (uint64_t)i40e_read_rx_ctl(hw, I40E_VFQF_HENA(0));
2628         hena |= ((uint64_t)i40e_read_rx_ctl(hw, I40E_VFQF_HENA(1))) << 32;
2629         rss_conf->rss_hf = i40e_parse_hena(vf->adapter, hena);
2630
2631         return 0;
2632 }
2633
2634 static int
2635 i40evf_dev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
2636 {
2637         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2638         struct rte_eth_dev_data *dev_data = vf->dev_data;
2639         uint32_t frame_size = mtu + I40E_ETH_OVERHEAD;
2640         int ret = 0;
2641
2642         /* check if mtu is within the allowed range */
2643         if ((mtu < ETHER_MIN_MTU) || (frame_size > I40E_FRAME_SIZE_MAX))
2644                 return -EINVAL;
2645
2646         /* mtu setting is forbidden if port is start */
2647         if (dev_data->dev_started) {
2648                 PMD_DRV_LOG(ERR, "port %d must be stopped before configuration",
2649                             dev_data->port_id);
2650                 return -EBUSY;
2651         }
2652
2653         if (frame_size > ETHER_MAX_LEN)
2654                 dev_data->dev_conf.rxmode.jumbo_frame = 1;
2655         else
2656                 dev_data->dev_conf.rxmode.jumbo_frame = 0;
2657
2658         dev_data->dev_conf.rxmode.max_rx_pkt_len = frame_size;
2659
2660         return ret;
2661 }
2662
2663 static void
2664 i40evf_set_default_mac_addr(struct rte_eth_dev *dev,
2665                             struct ether_addr *mac_addr)
2666 {
2667         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2668
2669         if (!is_valid_assigned_ether_addr(mac_addr)) {
2670                 PMD_DRV_LOG(ERR, "Tried to set invalid MAC address.");
2671                 return;
2672         }
2673
2674         if (is_same_ether_addr(mac_addr, dev->data->mac_addrs))
2675                 return;
2676
2677         if (vf->flags & I40E_FLAG_VF_MAC_BY_PF)
2678                 return;
2679
2680         i40evf_del_mac_addr_by_addr(dev, dev->data->mac_addrs);
2681
2682         i40evf_add_mac_addr(dev, mac_addr, 0, 0);
2683 }