193d967c7a9ddbef2161ead0968e64beb68ce124
[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_eal.h>
55 #include <rte_alarm.h>
56 #include <rte_ether.h>
57 #include <rte_ethdev.h>
58 #include <rte_ethdev_pci.h>
59 #include <rte_malloc.h>
60 #include <rte_dev.h>
61
62 #include "i40e_logs.h"
63 #include "base/i40e_prototype.h"
64 #include "base/i40e_adminq_cmd.h"
65 #include "base/i40e_type.h"
66
67 #include "i40e_rxtx.h"
68 #include "i40e_ethdev.h"
69 #include "i40e_pf.h"
70 #define I40EVF_VSI_DEFAULT_MSIX_INTR     1
71 #define I40EVF_VSI_DEFAULT_MSIX_INTR_LNX 0
72
73 /* busy wait delay in msec */
74 #define I40EVF_BUSY_WAIT_DELAY 10
75 #define I40EVF_BUSY_WAIT_COUNT 50
76 #define MAX_RESET_WAIT_CNT     20
77
78 struct i40evf_arq_msg_info {
79         enum virtchnl_ops ops;
80         enum i40e_status_code result;
81         uint16_t buf_len;
82         uint16_t msg_len;
83         uint8_t *msg;
84 };
85
86 struct vf_cmd_info {
87         enum virtchnl_ops ops;
88         uint8_t *in_args;
89         uint32_t in_args_size;
90         uint8_t *out_buffer;
91         /* Input & output type. pass in buffer size and pass out
92          * actual return result
93          */
94         uint32_t out_size;
95 };
96
97 enum i40evf_aq_result {
98         I40EVF_MSG_ERR = -1, /* Meet error when accessing admin queue */
99         I40EVF_MSG_NON,      /* Read nothing from admin queue */
100         I40EVF_MSG_SYS,      /* Read system msg from admin queue */
101         I40EVF_MSG_CMD,      /* Read async command result */
102 };
103
104 static int i40evf_dev_configure(struct rte_eth_dev *dev);
105 static int i40evf_dev_start(struct rte_eth_dev *dev);
106 static void i40evf_dev_stop(struct rte_eth_dev *dev);
107 static void i40evf_dev_info_get(struct rte_eth_dev *dev,
108                                 struct rte_eth_dev_info *dev_info);
109 static int i40evf_dev_link_update(struct rte_eth_dev *dev,
110                                   int wait_to_complete);
111 static int i40evf_dev_stats_get(struct rte_eth_dev *dev,
112                                 struct rte_eth_stats *stats);
113 static int i40evf_dev_xstats_get(struct rte_eth_dev *dev,
114                                  struct rte_eth_xstat *xstats, unsigned n);
115 static int i40evf_dev_xstats_get_names(struct rte_eth_dev *dev,
116                                        struct rte_eth_xstat_name *xstats_names,
117                                        unsigned limit);
118 static void i40evf_dev_xstats_reset(struct rte_eth_dev *dev);
119 static int i40evf_vlan_filter_set(struct rte_eth_dev *dev,
120                                   uint16_t vlan_id, int on);
121 static int i40evf_vlan_offload_set(struct rte_eth_dev *dev, int mask);
122 static void i40evf_dev_close(struct rte_eth_dev *dev);
123 static int  i40evf_dev_reset(struct rte_eth_dev *dev);
124 static void i40evf_dev_promiscuous_enable(struct rte_eth_dev *dev);
125 static void i40evf_dev_promiscuous_disable(struct rte_eth_dev *dev);
126 static void i40evf_dev_allmulticast_enable(struct rte_eth_dev *dev);
127 static void i40evf_dev_allmulticast_disable(struct rte_eth_dev *dev);
128 static int i40evf_init_vlan(struct rte_eth_dev *dev);
129 static int i40evf_dev_rx_queue_start(struct rte_eth_dev *dev,
130                                      uint16_t rx_queue_id);
131 static int i40evf_dev_rx_queue_stop(struct rte_eth_dev *dev,
132                                     uint16_t rx_queue_id);
133 static int i40evf_dev_tx_queue_start(struct rte_eth_dev *dev,
134                                      uint16_t tx_queue_id);
135 static int i40evf_dev_tx_queue_stop(struct rte_eth_dev *dev,
136                                     uint16_t tx_queue_id);
137 static int i40evf_add_mac_addr(struct rte_eth_dev *dev,
138                                struct ether_addr *addr,
139                                uint32_t index,
140                                uint32_t pool);
141 static void i40evf_del_mac_addr(struct rte_eth_dev *dev, uint32_t index);
142 static int i40evf_dev_rss_reta_update(struct rte_eth_dev *dev,
143                         struct rte_eth_rss_reta_entry64 *reta_conf,
144                         uint16_t reta_size);
145 static int i40evf_dev_rss_reta_query(struct rte_eth_dev *dev,
146                         struct rte_eth_rss_reta_entry64 *reta_conf,
147                         uint16_t reta_size);
148 static int i40evf_config_rss(struct i40e_vf *vf);
149 static int i40evf_dev_rss_hash_update(struct rte_eth_dev *dev,
150                                       struct rte_eth_rss_conf *rss_conf);
151 static int i40evf_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
152                                         struct rte_eth_rss_conf *rss_conf);
153 static int i40evf_dev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu);
154 static void i40evf_set_default_mac_addr(struct rte_eth_dev *dev,
155                                         struct ether_addr *mac_addr);
156 static int
157 i40evf_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id);
158 static int
159 i40evf_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id);
160 static void i40evf_handle_pf_event(struct rte_eth_dev *dev,
161                                    uint8_t *msg,
162                                    uint16_t msglen);
163
164 /* Default hash key buffer for RSS */
165 static uint32_t rss_key_default[I40E_VFQF_HKEY_MAX_INDEX + 1];
166
167 struct rte_i40evf_xstats_name_off {
168         char name[RTE_ETH_XSTATS_NAME_SIZE];
169         unsigned offset;
170 };
171
172 static const struct rte_i40evf_xstats_name_off rte_i40evf_stats_strings[] = {
173         {"rx_bytes", offsetof(struct i40e_eth_stats, rx_bytes)},
174         {"rx_unicast_packets", offsetof(struct i40e_eth_stats, rx_unicast)},
175         {"rx_multicast_packets", offsetof(struct i40e_eth_stats, rx_multicast)},
176         {"rx_broadcast_packets", offsetof(struct i40e_eth_stats, rx_broadcast)},
177         {"rx_dropped_packets", offsetof(struct i40e_eth_stats, rx_discards)},
178         {"rx_unknown_protocol_packets", offsetof(struct i40e_eth_stats,
179                 rx_unknown_protocol)},
180         {"tx_bytes", offsetof(struct i40e_eth_stats, tx_bytes)},
181         {"tx_unicast_packets", offsetof(struct i40e_eth_stats, tx_unicast)},
182         {"tx_multicast_packets", offsetof(struct i40e_eth_stats, tx_multicast)},
183         {"tx_broadcast_packets", offsetof(struct i40e_eth_stats, tx_broadcast)},
184         {"tx_dropped_packets", offsetof(struct i40e_eth_stats, tx_discards)},
185         {"tx_error_packets", offsetof(struct i40e_eth_stats, tx_errors)},
186 };
187
188 #define I40EVF_NB_XSTATS (sizeof(rte_i40evf_stats_strings) / \
189                 sizeof(rte_i40evf_stats_strings[0]))
190
191 static const struct eth_dev_ops i40evf_eth_dev_ops = {
192         .dev_configure        = i40evf_dev_configure,
193         .dev_start            = i40evf_dev_start,
194         .dev_stop             = i40evf_dev_stop,
195         .promiscuous_enable   = i40evf_dev_promiscuous_enable,
196         .promiscuous_disable  = i40evf_dev_promiscuous_disable,
197         .allmulticast_enable  = i40evf_dev_allmulticast_enable,
198         .allmulticast_disable = i40evf_dev_allmulticast_disable,
199         .link_update          = i40evf_dev_link_update,
200         .stats_get            = i40evf_dev_stats_get,
201         .stats_reset          = i40evf_dev_xstats_reset,
202         .xstats_get           = i40evf_dev_xstats_get,
203         .xstats_get_names     = i40evf_dev_xstats_get_names,
204         .xstats_reset         = i40evf_dev_xstats_reset,
205         .dev_close            = i40evf_dev_close,
206         .dev_reset            = i40evf_dev_reset,
207         .dev_infos_get        = i40evf_dev_info_get,
208         .dev_supported_ptypes_get = i40e_dev_supported_ptypes_get,
209         .vlan_filter_set      = i40evf_vlan_filter_set,
210         .vlan_offload_set     = i40evf_vlan_offload_set,
211         .rx_queue_start       = i40evf_dev_rx_queue_start,
212         .rx_queue_stop        = i40evf_dev_rx_queue_stop,
213         .tx_queue_start       = i40evf_dev_tx_queue_start,
214         .tx_queue_stop        = i40evf_dev_tx_queue_stop,
215         .rx_queue_setup       = i40e_dev_rx_queue_setup,
216         .rx_queue_release     = i40e_dev_rx_queue_release,
217         .rx_queue_intr_enable = i40evf_dev_rx_queue_intr_enable,
218         .rx_queue_intr_disable = i40evf_dev_rx_queue_intr_disable,
219         .rx_descriptor_done   = i40e_dev_rx_descriptor_done,
220         .rx_descriptor_status = i40e_dev_rx_descriptor_status,
221         .tx_descriptor_status = i40e_dev_tx_descriptor_status,
222         .tx_queue_setup       = i40e_dev_tx_queue_setup,
223         .tx_queue_release     = i40e_dev_tx_queue_release,
224         .rx_queue_count       = i40e_dev_rx_queue_count,
225         .rxq_info_get         = i40e_rxq_info_get,
226         .txq_info_get         = i40e_txq_info_get,
227         .mac_addr_add         = i40evf_add_mac_addr,
228         .mac_addr_remove      = i40evf_del_mac_addr,
229         .reta_update          = i40evf_dev_rss_reta_update,
230         .reta_query           = i40evf_dev_rss_reta_query,
231         .rss_hash_update      = i40evf_dev_rss_hash_update,
232         .rss_hash_conf_get    = i40evf_dev_rss_hash_conf_get,
233         .mtu_set              = i40evf_dev_mtu_set,
234         .mac_addr_set         = i40evf_set_default_mac_addr,
235 };
236
237 /*
238  * Read data in admin queue to get msg from pf driver
239  */
240 static enum i40evf_aq_result
241 i40evf_read_pfmsg(struct rte_eth_dev *dev, struct i40evf_arq_msg_info *data)
242 {
243         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
244         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
245         struct i40e_arq_event_info event;
246         enum virtchnl_ops opcode;
247         enum i40e_status_code retval;
248         int ret;
249         enum i40evf_aq_result result = I40EVF_MSG_NON;
250
251         event.buf_len = data->buf_len;
252         event.msg_buf = data->msg;
253         ret = i40e_clean_arq_element(hw, &event, NULL);
254         /* Can't read any msg from adminQ */
255         if (ret) {
256                 if (ret != I40E_ERR_ADMIN_QUEUE_NO_WORK)
257                         result = I40EVF_MSG_ERR;
258                 return result;
259         }
260
261         opcode = (enum virtchnl_ops)rte_le_to_cpu_32(event.desc.cookie_high);
262         retval = (enum i40e_status_code)rte_le_to_cpu_32(event.desc.cookie_low);
263         /* pf sys event */
264         if (opcode == VIRTCHNL_OP_EVENT) {
265                 struct virtchnl_pf_event *vpe =
266                         (struct virtchnl_pf_event *)event.msg_buf;
267
268                 result = I40EVF_MSG_SYS;
269                 switch (vpe->event) {
270                 case VIRTCHNL_EVENT_LINK_CHANGE:
271                         vf->link_up =
272                                 vpe->event_data.link_event.link_status;
273                         vf->link_speed =
274                                 vpe->event_data.link_event.link_speed;
275                         vf->pend_msg |= PFMSG_LINK_CHANGE;
276                         PMD_DRV_LOG(INFO, "Link status update:%s",
277                                     vf->link_up ? "up" : "down");
278                         break;
279                 case VIRTCHNL_EVENT_RESET_IMPENDING:
280                         vf->vf_reset = true;
281                         vf->pend_msg |= PFMSG_RESET_IMPENDING;
282                         PMD_DRV_LOG(INFO, "vf is reseting");
283                         break;
284                 case VIRTCHNL_EVENT_PF_DRIVER_CLOSE:
285                         vf->dev_closed = true;
286                         vf->pend_msg |= PFMSG_DRIVER_CLOSE;
287                         PMD_DRV_LOG(INFO, "PF driver closed");
288                         break;
289                 default:
290                         PMD_DRV_LOG(ERR, "%s: Unknown event %d from pf",
291                                     __func__, vpe->event);
292                 }
293         } else {
294                 /* async reply msg on command issued by vf previously */
295                 result = I40EVF_MSG_CMD;
296                 /* Actual data length read from PF */
297                 data->msg_len = event.msg_len;
298         }
299
300         data->result = retval;
301         data->ops = opcode;
302
303         return result;
304 }
305
306 /**
307  * clear current command. Only call in case execute
308  * _atomic_set_cmd successfully.
309  */
310 static inline void
311 _clear_cmd(struct i40e_vf *vf)
312 {
313         rte_wmb();
314         vf->pend_cmd = VIRTCHNL_OP_UNKNOWN;
315 }
316
317 /*
318  * Check there is pending cmd in execution. If none, set new command.
319  */
320 static inline int
321 _atomic_set_cmd(struct i40e_vf *vf, enum virtchnl_ops ops)
322 {
323         int ret = rte_atomic32_cmpset(&vf->pend_cmd,
324                         VIRTCHNL_OP_UNKNOWN, ops);
325
326         if (!ret)
327                 PMD_DRV_LOG(ERR, "There is incomplete cmd %d", vf->pend_cmd);
328
329         return !ret;
330 }
331
332 #define MAX_TRY_TIMES 200
333 #define ASQ_DELAY_MS  10
334
335 static int
336 i40evf_execute_vf_cmd(struct rte_eth_dev *dev, struct vf_cmd_info *args)
337 {
338         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
339         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
340         struct i40evf_arq_msg_info info;
341         enum i40evf_aq_result ret;
342         int err, i = 0;
343
344         if (_atomic_set_cmd(vf, args->ops))
345                 return -1;
346
347         info.msg = args->out_buffer;
348         info.buf_len = args->out_size;
349         info.ops = VIRTCHNL_OP_UNKNOWN;
350         info.result = I40E_SUCCESS;
351
352         err = i40e_aq_send_msg_to_pf(hw, args->ops, I40E_SUCCESS,
353                      args->in_args, args->in_args_size, NULL);
354         if (err) {
355                 PMD_DRV_LOG(ERR, "fail to send cmd %d", args->ops);
356                 _clear_cmd(vf);
357                 return err;
358         }
359
360         switch (args->ops) {
361         case VIRTCHNL_OP_RESET_VF:
362                 /*no need to process in this function */
363                 err = 0;
364                 break;
365         case VIRTCHNL_OP_VERSION:
366         case VIRTCHNL_OP_GET_VF_RESOURCES:
367                 /* for init adminq commands, need to poll the response */
368                 err = -1;
369                 do {
370                         ret = i40evf_read_pfmsg(dev, &info);
371                         vf->cmd_retval = info.result;
372                         if (ret == I40EVF_MSG_CMD) {
373                                 err = 0;
374                                 break;
375                         } else if (ret == I40EVF_MSG_ERR)
376                                 break;
377                         rte_delay_ms(ASQ_DELAY_MS);
378                         /* If don't read msg or read sys event, continue */
379                 } while (i++ < MAX_TRY_TIMES);
380                 _clear_cmd(vf);
381                 break;
382
383         default:
384                 /* for other adminq in running time, waiting the cmd done flag */
385                 err = -1;
386                 do {
387                         if (vf->pend_cmd == VIRTCHNL_OP_UNKNOWN) {
388                                 err = 0;
389                                 break;
390                         }
391                         rte_delay_ms(ASQ_DELAY_MS);
392                         /* If don't read msg or read sys event, continue */
393                 } while (i++ < MAX_TRY_TIMES);
394                 /* If there's no response is received, clear command */
395                 if (i >= MAX_TRY_TIMES) {
396                         PMD_DRV_LOG(WARNING, "No response for %d", args->ops);
397                         _clear_cmd(vf);
398                 }
399                 break;
400         }
401
402         return err | vf->cmd_retval;
403 }
404
405 /*
406  * Check API version with sync wait until version read or fail from admin queue
407  */
408 static int
409 i40evf_check_api_version(struct rte_eth_dev *dev)
410 {
411         struct virtchnl_version_info version, *pver;
412         int err;
413         struct vf_cmd_info args;
414         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
415
416         version.major = VIRTCHNL_VERSION_MAJOR;
417         version.minor = VIRTCHNL_VERSION_MINOR;
418
419         args.ops = VIRTCHNL_OP_VERSION;
420         args.in_args = (uint8_t *)&version;
421         args.in_args_size = sizeof(version);
422         args.out_buffer = vf->aq_resp;
423         args.out_size = I40E_AQ_BUF_SZ;
424
425         err = i40evf_execute_vf_cmd(dev, &args);
426         if (err) {
427                 PMD_INIT_LOG(ERR, "fail to execute command OP_VERSION");
428                 return err;
429         }
430
431         pver = (struct virtchnl_version_info *)args.out_buffer;
432         vf->version_major = pver->major;
433         vf->version_minor = pver->minor;
434         if ((vf->version_major == VIRTCHNL_VERSION_MAJOR) &&
435                 (vf->version_minor <= VIRTCHNL_VERSION_MINOR))
436                 PMD_DRV_LOG(INFO, "Peer is Linux PF host");
437         else {
438                 PMD_INIT_LOG(ERR, "PF/VF API version mismatch:(%u.%u)-(%u.%u)",
439                                         vf->version_major, vf->version_minor,
440                                                 VIRTCHNL_VERSION_MAJOR,
441                                                 VIRTCHNL_VERSION_MINOR);
442                 return -1;
443         }
444
445         return 0;
446 }
447
448 static int
449 i40evf_get_vf_resource(struct rte_eth_dev *dev)
450 {
451         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
452         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
453         int err;
454         struct vf_cmd_info args;
455         uint32_t caps, len;
456
457         args.ops = VIRTCHNL_OP_GET_VF_RESOURCES;
458         args.out_buffer = vf->aq_resp;
459         args.out_size = I40E_AQ_BUF_SZ;
460         if (PF_IS_V11(vf)) {
461                 caps = VIRTCHNL_VF_OFFLOAD_L2 |
462                        VIRTCHNL_VF_OFFLOAD_RSS_AQ |
463                        VIRTCHNL_VF_OFFLOAD_RSS_REG |
464                        VIRTCHNL_VF_OFFLOAD_VLAN |
465                        VIRTCHNL_VF_OFFLOAD_RX_POLLING;
466                 args.in_args = (uint8_t *)&caps;
467                 args.in_args_size = sizeof(caps);
468         } else {
469                 args.in_args = NULL;
470                 args.in_args_size = 0;
471         }
472         err = i40evf_execute_vf_cmd(dev, &args);
473
474         if (err) {
475                 PMD_DRV_LOG(ERR, "fail to execute command OP_GET_VF_RESOURCE");
476                 return err;
477         }
478
479         len =  sizeof(struct virtchnl_vf_resource) +
480                 I40E_MAX_VF_VSI * sizeof(struct virtchnl_vsi_resource);
481
482         rte_memcpy(vf->vf_res, args.out_buffer,
483                         RTE_MIN(args.out_size, len));
484         i40e_vf_parse_hw_config(hw, vf->vf_res);
485
486         return 0;
487 }
488
489 static int
490 i40evf_config_promisc(struct rte_eth_dev *dev,
491                       bool enable_unicast,
492                       bool enable_multicast)
493 {
494         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
495         int err;
496         struct vf_cmd_info args;
497         struct virtchnl_promisc_info promisc;
498
499         promisc.flags = 0;
500         promisc.vsi_id = vf->vsi_res->vsi_id;
501
502         if (enable_unicast)
503                 promisc.flags |= FLAG_VF_UNICAST_PROMISC;
504
505         if (enable_multicast)
506                 promisc.flags |= FLAG_VF_MULTICAST_PROMISC;
507
508         args.ops = VIRTCHNL_OP_CONFIG_PROMISCUOUS_MODE;
509         args.in_args = (uint8_t *)&promisc;
510         args.in_args_size = sizeof(promisc);
511         args.out_buffer = vf->aq_resp;
512         args.out_size = I40E_AQ_BUF_SZ;
513
514         err = i40evf_execute_vf_cmd(dev, &args);
515
516         if (err)
517                 PMD_DRV_LOG(ERR, "fail to execute command "
518                             "CONFIG_PROMISCUOUS_MODE");
519         return err;
520 }
521
522 static int
523 i40evf_enable_vlan_strip(struct rte_eth_dev *dev)
524 {
525         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
526         struct vf_cmd_info args;
527         int ret;
528
529         memset(&args, 0, sizeof(args));
530         args.ops = VIRTCHNL_OP_ENABLE_VLAN_STRIPPING;
531         args.in_args = NULL;
532         args.in_args_size = 0;
533         args.out_buffer = vf->aq_resp;
534         args.out_size = I40E_AQ_BUF_SZ;
535         ret = i40evf_execute_vf_cmd(dev, &args);
536         if (ret)
537                 PMD_DRV_LOG(ERR, "Failed to execute command of "
538                             "VIRTCHNL_OP_ENABLE_VLAN_STRIPPING");
539
540         return ret;
541 }
542
543 static int
544 i40evf_disable_vlan_strip(struct rte_eth_dev *dev)
545 {
546         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
547         struct vf_cmd_info args;
548         int ret;
549
550         memset(&args, 0, sizeof(args));
551         args.ops = VIRTCHNL_OP_DISABLE_VLAN_STRIPPING;
552         args.in_args = NULL;
553         args.in_args_size = 0;
554         args.out_buffer = vf->aq_resp;
555         args.out_size = I40E_AQ_BUF_SZ;
556         ret = i40evf_execute_vf_cmd(dev, &args);
557         if (ret)
558                 PMD_DRV_LOG(ERR, "Failed to execute command of "
559                             "VIRTCHNL_OP_DISABLE_VLAN_STRIPPING");
560
561         return ret;
562 }
563
564 static void
565 i40evf_fill_virtchnl_vsi_txq_info(struct virtchnl_txq_info *txq_info,
566                                   uint16_t vsi_id,
567                                   uint16_t queue_id,
568                                   uint16_t nb_txq,
569                                   struct i40e_tx_queue *txq)
570 {
571         txq_info->vsi_id = vsi_id;
572         txq_info->queue_id = queue_id;
573         if (queue_id < nb_txq) {
574                 txq_info->ring_len = txq->nb_tx_desc;
575                 txq_info->dma_ring_addr = txq->tx_ring_phys_addr;
576         }
577 }
578
579 static void
580 i40evf_fill_virtchnl_vsi_rxq_info(struct virtchnl_rxq_info *rxq_info,
581                                   uint16_t vsi_id,
582                                   uint16_t queue_id,
583                                   uint16_t nb_rxq,
584                                   uint32_t max_pkt_size,
585                                   struct i40e_rx_queue *rxq)
586 {
587         rxq_info->vsi_id = vsi_id;
588         rxq_info->queue_id = queue_id;
589         rxq_info->max_pkt_size = max_pkt_size;
590         if (queue_id < nb_rxq) {
591                 rxq_info->ring_len = rxq->nb_rx_desc;
592                 rxq_info->dma_ring_addr = rxq->rx_ring_phys_addr;
593                 rxq_info->databuffer_size =
594                         (rte_pktmbuf_data_room_size(rxq->mp) -
595                                 RTE_PKTMBUF_HEADROOM);
596         }
597 }
598
599 static int
600 i40evf_configure_vsi_queues(struct rte_eth_dev *dev)
601 {
602         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
603         struct i40e_rx_queue **rxq =
604                 (struct i40e_rx_queue **)dev->data->rx_queues;
605         struct i40e_tx_queue **txq =
606                 (struct i40e_tx_queue **)dev->data->tx_queues;
607         struct virtchnl_vsi_queue_config_info *vc_vqci;
608         struct virtchnl_queue_pair_info *vc_qpi;
609         struct vf_cmd_info args;
610         uint16_t i, nb_qp = vf->num_queue_pairs;
611         const uint32_t size =
612                 I40E_VIRTCHNL_CONFIG_VSI_QUEUES_SIZE(vc_vqci, nb_qp);
613         uint8_t buff[size];
614         int ret;
615
616         memset(buff, 0, sizeof(buff));
617         vc_vqci = (struct virtchnl_vsi_queue_config_info *)buff;
618         vc_vqci->vsi_id = vf->vsi_res->vsi_id;
619         vc_vqci->num_queue_pairs = nb_qp;
620
621         for (i = 0, vc_qpi = vc_vqci->qpair; i < nb_qp; i++, vc_qpi++) {
622                 i40evf_fill_virtchnl_vsi_txq_info(&vc_qpi->txq,
623                         vc_vqci->vsi_id, i, dev->data->nb_tx_queues, txq[i]);
624                 i40evf_fill_virtchnl_vsi_rxq_info(&vc_qpi->rxq,
625                         vc_vqci->vsi_id, i, dev->data->nb_rx_queues,
626                                         vf->max_pkt_len, rxq[i]);
627         }
628         memset(&args, 0, sizeof(args));
629         args.ops = VIRTCHNL_OP_CONFIG_VSI_QUEUES;
630         args.in_args = (uint8_t *)vc_vqci;
631         args.in_args_size = size;
632         args.out_buffer = vf->aq_resp;
633         args.out_size = I40E_AQ_BUF_SZ;
634         ret = i40evf_execute_vf_cmd(dev, &args);
635         if (ret)
636                 PMD_DRV_LOG(ERR, "Failed to execute command of "
637                         "VIRTCHNL_OP_CONFIG_VSI_QUEUES");
638
639         return ret;
640 }
641
642 static int
643 i40evf_config_irq_map(struct rte_eth_dev *dev)
644 {
645         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
646         struct vf_cmd_info args;
647         uint8_t cmd_buffer[sizeof(struct virtchnl_irq_map_info) + \
648                 sizeof(struct virtchnl_vector_map)];
649         struct virtchnl_irq_map_info *map_info;
650         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
651         struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
652         uint32_t vector_id;
653         int i, err;
654
655         if (rte_intr_allow_others(intr_handle))
656                 vector_id = I40EVF_VSI_DEFAULT_MSIX_INTR_LNX;
657         else
658                 vector_id = I40E_MISC_VEC_ID;
659
660         map_info = (struct virtchnl_irq_map_info *)cmd_buffer;
661         map_info->num_vectors = 1;
662         map_info->vecmap[0].rxitr_idx = I40E_ITR_INDEX_DEFAULT;
663         map_info->vecmap[0].vsi_id = vf->vsi_res->vsi_id;
664         /* Alway use default dynamic MSIX interrupt */
665         map_info->vecmap[0].vector_id = vector_id;
666         /* Don't map any tx queue */
667         map_info->vecmap[0].txq_map = 0;
668         map_info->vecmap[0].rxq_map = 0;
669         for (i = 0; i < dev->data->nb_rx_queues; i++) {
670                 map_info->vecmap[0].rxq_map |= 1 << i;
671                 if (rte_intr_dp_is_en(intr_handle))
672                         intr_handle->intr_vec[i] = vector_id;
673         }
674
675         args.ops = VIRTCHNL_OP_CONFIG_IRQ_MAP;
676         args.in_args = (u8 *)cmd_buffer;
677         args.in_args_size = sizeof(cmd_buffer);
678         args.out_buffer = vf->aq_resp;
679         args.out_size = I40E_AQ_BUF_SZ;
680         err = i40evf_execute_vf_cmd(dev, &args);
681         if (err)
682                 PMD_DRV_LOG(ERR, "fail to execute command OP_ENABLE_QUEUES");
683
684         return err;
685 }
686
687 static int
688 i40evf_switch_queue(struct rte_eth_dev *dev, bool isrx, uint16_t qid,
689                                 bool on)
690 {
691         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
692         struct virtchnl_queue_select queue_select;
693         int err;
694         struct vf_cmd_info args;
695         memset(&queue_select, 0, sizeof(queue_select));
696         queue_select.vsi_id = vf->vsi_res->vsi_id;
697
698         if (isrx)
699                 queue_select.rx_queues |= 1 << qid;
700         else
701                 queue_select.tx_queues |= 1 << qid;
702
703         if (on)
704                 args.ops = VIRTCHNL_OP_ENABLE_QUEUES;
705         else
706                 args.ops = VIRTCHNL_OP_DISABLE_QUEUES;
707         args.in_args = (u8 *)&queue_select;
708         args.in_args_size = sizeof(queue_select);
709         args.out_buffer = vf->aq_resp;
710         args.out_size = I40E_AQ_BUF_SZ;
711         err = i40evf_execute_vf_cmd(dev, &args);
712         if (err)
713                 PMD_DRV_LOG(ERR, "fail to switch %s %u %s",
714                             isrx ? "RX" : "TX", qid, on ? "on" : "off");
715
716         return err;
717 }
718
719 static int
720 i40evf_start_queues(struct rte_eth_dev *dev)
721 {
722         struct rte_eth_dev_data *dev_data = dev->data;
723         int i;
724         struct i40e_rx_queue *rxq;
725         struct i40e_tx_queue *txq;
726
727         for (i = 0; i < dev->data->nb_rx_queues; i++) {
728                 rxq = dev_data->rx_queues[i];
729                 if (rxq->rx_deferred_start)
730                         continue;
731                 if (i40evf_dev_rx_queue_start(dev, i) != 0) {
732                         PMD_DRV_LOG(ERR, "Fail to start queue %u", i);
733                         return -1;
734                 }
735         }
736
737         for (i = 0; i < dev->data->nb_tx_queues; i++) {
738                 txq = dev_data->tx_queues[i];
739                 if (txq->tx_deferred_start)
740                         continue;
741                 if (i40evf_dev_tx_queue_start(dev, i) != 0) {
742                         PMD_DRV_LOG(ERR, "Fail to start queue %u", i);
743                         return -1;
744                 }
745         }
746
747         return 0;
748 }
749
750 static int
751 i40evf_stop_queues(struct rte_eth_dev *dev)
752 {
753         int i;
754
755         /* Stop TX queues first */
756         for (i = 0; i < dev->data->nb_tx_queues; i++) {
757                 if (i40evf_dev_tx_queue_stop(dev, i) != 0) {
758                         PMD_DRV_LOG(ERR, "Fail to stop queue %u", i);
759                         return -1;
760                 }
761         }
762
763         /* Then stop RX queues */
764         for (i = 0; i < dev->data->nb_rx_queues; i++) {
765                 if (i40evf_dev_rx_queue_stop(dev, i) != 0) {
766                         PMD_DRV_LOG(ERR, "Fail to stop queue %u", i);
767                         return -1;
768                 }
769         }
770
771         return 0;
772 }
773
774 static int
775 i40evf_add_mac_addr(struct rte_eth_dev *dev,
776                     struct ether_addr *addr,
777                     __rte_unused uint32_t index,
778                     __rte_unused uint32_t pool)
779 {
780         struct virtchnl_ether_addr_list *list;
781         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
782         uint8_t cmd_buffer[sizeof(struct virtchnl_ether_addr_list) + \
783                         sizeof(struct virtchnl_ether_addr)];
784         int err;
785         struct vf_cmd_info args;
786
787         if (is_zero_ether_addr(addr)) {
788                 PMD_DRV_LOG(ERR, "Invalid mac:%x:%x:%x:%x:%x:%x",
789                             addr->addr_bytes[0], addr->addr_bytes[1],
790                             addr->addr_bytes[2], addr->addr_bytes[3],
791                             addr->addr_bytes[4], addr->addr_bytes[5]);
792                 return I40E_ERR_INVALID_MAC_ADDR;
793         }
794
795         list = (struct virtchnl_ether_addr_list *)cmd_buffer;
796         list->vsi_id = vf->vsi_res->vsi_id;
797         list->num_elements = 1;
798         rte_memcpy(list->list[0].addr, addr->addr_bytes,
799                                         sizeof(addr->addr_bytes));
800
801         args.ops = VIRTCHNL_OP_ADD_ETH_ADDR;
802         args.in_args = cmd_buffer;
803         args.in_args_size = sizeof(cmd_buffer);
804         args.out_buffer = vf->aq_resp;
805         args.out_size = I40E_AQ_BUF_SZ;
806         err = i40evf_execute_vf_cmd(dev, &args);
807         if (err)
808                 PMD_DRV_LOG(ERR, "fail to execute command "
809                             "OP_ADD_ETHER_ADDRESS");
810         else
811                 vf->vsi.mac_num++;
812
813         return err;
814 }
815
816 static void
817 i40evf_del_mac_addr_by_addr(struct rte_eth_dev *dev,
818                             struct ether_addr *addr)
819 {
820         struct virtchnl_ether_addr_list *list;
821         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
822         uint8_t cmd_buffer[sizeof(struct virtchnl_ether_addr_list) + \
823                         sizeof(struct virtchnl_ether_addr)];
824         int err;
825         struct vf_cmd_info args;
826
827         if (i40e_validate_mac_addr(addr->addr_bytes) != I40E_SUCCESS) {
828                 PMD_DRV_LOG(ERR, "Invalid mac:%x-%x-%x-%x-%x-%x",
829                             addr->addr_bytes[0], addr->addr_bytes[1],
830                             addr->addr_bytes[2], addr->addr_bytes[3],
831                             addr->addr_bytes[4], addr->addr_bytes[5]);
832                 return;
833         }
834
835         list = (struct virtchnl_ether_addr_list *)cmd_buffer;
836         list->vsi_id = vf->vsi_res->vsi_id;
837         list->num_elements = 1;
838         rte_memcpy(list->list[0].addr, addr->addr_bytes,
839                         sizeof(addr->addr_bytes));
840
841         args.ops = VIRTCHNL_OP_DEL_ETH_ADDR;
842         args.in_args = cmd_buffer;
843         args.in_args_size = sizeof(cmd_buffer);
844         args.out_buffer = vf->aq_resp;
845         args.out_size = I40E_AQ_BUF_SZ;
846         err = i40evf_execute_vf_cmd(dev, &args);
847         if (err)
848                 PMD_DRV_LOG(ERR, "fail to execute command "
849                             "OP_DEL_ETHER_ADDRESS");
850         else
851                 vf->vsi.mac_num--;
852         return;
853 }
854
855 static void
856 i40evf_del_mac_addr(struct rte_eth_dev *dev, uint32_t index)
857 {
858         struct rte_eth_dev_data *data = dev->data;
859         struct ether_addr *addr;
860
861         addr = &data->mac_addrs[index];
862
863         i40evf_del_mac_addr_by_addr(dev, addr);
864 }
865
866 static int
867 i40evf_query_stats(struct rte_eth_dev *dev, struct i40e_eth_stats **pstats)
868 {
869         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
870         struct virtchnl_queue_select q_stats;
871         int err;
872         struct vf_cmd_info args;
873
874         memset(&q_stats, 0, sizeof(q_stats));
875         q_stats.vsi_id = vf->vsi_res->vsi_id;
876         args.ops = VIRTCHNL_OP_GET_STATS;
877         args.in_args = (u8 *)&q_stats;
878         args.in_args_size = sizeof(q_stats);
879         args.out_buffer = vf->aq_resp;
880         args.out_size = I40E_AQ_BUF_SZ;
881
882         err = i40evf_execute_vf_cmd(dev, &args);
883         if (err) {
884                 PMD_DRV_LOG(ERR, "fail to execute command OP_GET_STATS");
885                 *pstats = NULL;
886                 return err;
887         }
888         *pstats = (struct i40e_eth_stats *)args.out_buffer;
889         return 0;
890 }
891
892 static void
893 i40evf_stat_update_48(uint64_t *offset,
894                    uint64_t *stat)
895 {
896         if (*stat >= *offset)
897                 *stat = *stat - *offset;
898         else
899                 *stat = (uint64_t)((*stat +
900                         ((uint64_t)1 << I40E_48_BIT_WIDTH)) - *offset);
901
902         *stat &= I40E_48_BIT_MASK;
903 }
904
905 static void
906 i40evf_stat_update_32(uint64_t *offset,
907                    uint64_t *stat)
908 {
909         if (*stat >= *offset)
910                 *stat = (uint64_t)(*stat - *offset);
911         else
912                 *stat = (uint64_t)((*stat +
913                         ((uint64_t)1 << I40E_32_BIT_WIDTH)) - *offset);
914 }
915
916 static void
917 i40evf_update_stats(struct i40e_vsi *vsi,
918                                         struct i40e_eth_stats *nes)
919 {
920         struct i40e_eth_stats *oes = &vsi->eth_stats_offset;
921
922         i40evf_stat_update_48(&oes->rx_bytes,
923                             &nes->rx_bytes);
924         i40evf_stat_update_48(&oes->rx_unicast,
925                             &nes->rx_unicast);
926         i40evf_stat_update_48(&oes->rx_multicast,
927                             &nes->rx_multicast);
928         i40evf_stat_update_48(&oes->rx_broadcast,
929                             &nes->rx_broadcast);
930         i40evf_stat_update_32(&oes->rx_discards,
931                                 &nes->rx_discards);
932         i40evf_stat_update_32(&oes->rx_unknown_protocol,
933                             &nes->rx_unknown_protocol);
934         i40evf_stat_update_48(&oes->tx_bytes,
935                             &nes->tx_bytes);
936         i40evf_stat_update_48(&oes->tx_unicast,
937                             &nes->tx_unicast);
938         i40evf_stat_update_48(&oes->tx_multicast,
939                             &nes->tx_multicast);
940         i40evf_stat_update_48(&oes->tx_broadcast,
941                             &nes->tx_broadcast);
942         i40evf_stat_update_32(&oes->tx_errors, &nes->tx_errors);
943         i40evf_stat_update_32(&oes->tx_discards, &nes->tx_discards);
944 }
945
946 static void
947 i40evf_dev_xstats_reset(struct rte_eth_dev *dev)
948 {
949         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
950         struct i40e_eth_stats *pstats = NULL;
951
952         /* read stat values to clear hardware registers */
953         i40evf_query_stats(dev, &pstats);
954
955         /* set stats offset base on current values */
956         vf->vsi.eth_stats_offset = *pstats;
957 }
958
959 static int i40evf_dev_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
960                                       struct rte_eth_xstat_name *xstats_names,
961                                       __rte_unused unsigned limit)
962 {
963         unsigned i;
964
965         if (xstats_names != NULL)
966                 for (i = 0; i < I40EVF_NB_XSTATS; i++) {
967                         snprintf(xstats_names[i].name,
968                                 sizeof(xstats_names[i].name),
969                                 "%s", rte_i40evf_stats_strings[i].name);
970                 }
971         return I40EVF_NB_XSTATS;
972 }
973
974 static int i40evf_dev_xstats_get(struct rte_eth_dev *dev,
975                                  struct rte_eth_xstat *xstats, unsigned n)
976 {
977         int ret;
978         unsigned i;
979         struct i40e_eth_stats *pstats = NULL;
980         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
981         struct i40e_vsi *vsi = &vf->vsi;
982
983         if (n < I40EVF_NB_XSTATS)
984                 return I40EVF_NB_XSTATS;
985
986         ret = i40evf_query_stats(dev, &pstats);
987         if (ret != 0)
988                 return 0;
989
990         if (!xstats)
991                 return 0;
992
993         i40evf_update_stats(vsi, pstats);
994
995         /* loop over xstats array and values from pstats */
996         for (i = 0; i < I40EVF_NB_XSTATS; i++) {
997                 xstats[i].id = i;
998                 xstats[i].value = *(uint64_t *)(((char *)pstats) +
999                         rte_i40evf_stats_strings[i].offset);
1000         }
1001
1002         return I40EVF_NB_XSTATS;
1003 }
1004
1005 static int
1006 i40evf_add_vlan(struct rte_eth_dev *dev, uint16_t vlanid)
1007 {
1008         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
1009         struct virtchnl_vlan_filter_list *vlan_list;
1010         uint8_t cmd_buffer[sizeof(struct virtchnl_vlan_filter_list) +
1011                                                         sizeof(uint16_t)];
1012         int err;
1013         struct vf_cmd_info args;
1014
1015         vlan_list = (struct virtchnl_vlan_filter_list *)cmd_buffer;
1016         vlan_list->vsi_id = vf->vsi_res->vsi_id;
1017         vlan_list->num_elements = 1;
1018         vlan_list->vlan_id[0] = vlanid;
1019
1020         args.ops = VIRTCHNL_OP_ADD_VLAN;
1021         args.in_args = (u8 *)&cmd_buffer;
1022         args.in_args_size = sizeof(cmd_buffer);
1023         args.out_buffer = vf->aq_resp;
1024         args.out_size = I40E_AQ_BUF_SZ;
1025         err = i40evf_execute_vf_cmd(dev, &args);
1026         if (err)
1027                 PMD_DRV_LOG(ERR, "fail to execute command OP_ADD_VLAN");
1028
1029         return err;
1030 }
1031
1032 static int
1033 i40evf_del_vlan(struct rte_eth_dev *dev, uint16_t vlanid)
1034 {
1035         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
1036         struct virtchnl_vlan_filter_list *vlan_list;
1037         uint8_t cmd_buffer[sizeof(struct virtchnl_vlan_filter_list) +
1038                                                         sizeof(uint16_t)];
1039         int err;
1040         struct vf_cmd_info args;
1041
1042         vlan_list = (struct virtchnl_vlan_filter_list *)cmd_buffer;
1043         vlan_list->vsi_id = vf->vsi_res->vsi_id;
1044         vlan_list->num_elements = 1;
1045         vlan_list->vlan_id[0] = vlanid;
1046
1047         args.ops = VIRTCHNL_OP_DEL_VLAN;
1048         args.in_args = (u8 *)&cmd_buffer;
1049         args.in_args_size = sizeof(cmd_buffer);
1050         args.out_buffer = vf->aq_resp;
1051         args.out_size = I40E_AQ_BUF_SZ;
1052         err = i40evf_execute_vf_cmd(dev, &args);
1053         if (err)
1054                 PMD_DRV_LOG(ERR, "fail to execute command OP_DEL_VLAN");
1055
1056         return err;
1057 }
1058
1059 static const struct rte_pci_id pci_id_i40evf_map[] = {
1060         { RTE_PCI_DEVICE(I40E_INTEL_VENDOR_ID, I40E_DEV_ID_VF) },
1061         { RTE_PCI_DEVICE(I40E_INTEL_VENDOR_ID, I40E_DEV_ID_VF_HV) },
1062         { RTE_PCI_DEVICE(I40E_INTEL_VENDOR_ID, I40E_DEV_ID_X722_A0_VF) },
1063         { RTE_PCI_DEVICE(I40E_INTEL_VENDOR_ID, I40E_DEV_ID_X722_VF) },
1064         { .vendor_id = 0, /* sentinel */ },
1065 };
1066
1067 static inline int
1068 i40evf_dev_atomic_write_link_status(struct rte_eth_dev *dev,
1069                                     struct rte_eth_link *link)
1070 {
1071         struct rte_eth_link *dst = &(dev->data->dev_link);
1072         struct rte_eth_link *src = link;
1073
1074         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
1075                                         *(uint64_t *)src) == 0)
1076                 return -1;
1077
1078         return 0;
1079 }
1080
1081 /* Disable IRQ0 */
1082 static inline void
1083 i40evf_disable_irq0(struct i40e_hw *hw)
1084 {
1085         /* Disable all interrupt types */
1086         I40E_WRITE_REG(hw, I40E_VFINT_ICR0_ENA1, 0);
1087         I40E_WRITE_REG(hw, I40E_VFINT_DYN_CTL01,
1088                        I40E_VFINT_DYN_CTL01_ITR_INDX_MASK);
1089         I40EVF_WRITE_FLUSH(hw);
1090 }
1091
1092 /* Enable IRQ0 */
1093 static inline void
1094 i40evf_enable_irq0(struct i40e_hw *hw)
1095 {
1096         /* Enable admin queue interrupt trigger */
1097         uint32_t val;
1098
1099         i40evf_disable_irq0(hw);
1100         val = I40E_READ_REG(hw, I40E_VFINT_ICR0_ENA1);
1101         val |= I40E_VFINT_ICR0_ENA1_ADMINQ_MASK |
1102                 I40E_VFINT_ICR0_ENA1_LINK_STAT_CHANGE_MASK;
1103         I40E_WRITE_REG(hw, I40E_VFINT_ICR0_ENA1, val);
1104
1105         I40E_WRITE_REG(hw, I40E_VFINT_DYN_CTL01,
1106                 I40E_VFINT_DYN_CTL01_INTENA_MASK |
1107                 I40E_VFINT_DYN_CTL01_CLEARPBA_MASK |
1108                 I40E_VFINT_DYN_CTL01_ITR_INDX_MASK);
1109
1110         I40EVF_WRITE_FLUSH(hw);
1111 }
1112
1113 static int
1114 i40evf_check_vf_reset_done(struct i40e_hw *hw)
1115 {
1116         int i, reset;
1117
1118         for (i = 0; i < MAX_RESET_WAIT_CNT; i++) {
1119                 reset = I40E_READ_REG(hw, I40E_VFGEN_RSTAT) &
1120                         I40E_VFGEN_RSTAT_VFR_STATE_MASK;
1121                 reset = reset >> I40E_VFGEN_RSTAT_VFR_STATE_SHIFT;
1122                 if (reset == VIRTCHNL_VFR_VFACTIVE ||
1123                     reset == VIRTCHNL_VFR_COMPLETED)
1124                         break;
1125                 rte_delay_ms(50);
1126         }
1127
1128         if (i >= MAX_RESET_WAIT_CNT)
1129                 return -1;
1130
1131         return 0;
1132 }
1133 static int
1134 i40evf_reset_vf(struct i40e_hw *hw)
1135 {
1136         int ret;
1137
1138         if (i40e_vf_reset(hw) != I40E_SUCCESS) {
1139                 PMD_INIT_LOG(ERR, "Reset VF NIC failed");
1140                 return -1;
1141         }
1142         /**
1143           * After issuing vf reset command to pf, pf won't necessarily
1144           * reset vf, it depends on what state it exactly is. If it's not
1145           * initialized yet, it won't have vf reset since it's in a certain
1146           * state. If not, it will try to reset. Even vf is reset, pf will
1147           * set I40E_VFGEN_RSTAT to COMPLETE first, then wait 10ms and set
1148           * it to ACTIVE. In this duration, vf may not catch the moment that
1149           * COMPLETE is set. So, for vf, we'll try to wait a long time.
1150           */
1151         rte_delay_ms(200);
1152
1153         ret = i40evf_check_vf_reset_done(hw);
1154         if (ret) {
1155                 PMD_INIT_LOG(ERR, "VF is still resetting");
1156                 return ret;
1157         }
1158
1159         return 0;
1160 }
1161
1162 static int
1163 i40evf_init_vf(struct rte_eth_dev *dev)
1164 {
1165         int i, err, bufsz;
1166         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1167         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
1168         uint16_t interval =
1169                 i40e_calc_itr_interval(I40E_QUEUE_ITR_INTERVAL_MAX);
1170
1171         vf->adapter = I40E_DEV_PRIVATE_TO_ADAPTER(dev->data->dev_private);
1172         vf->dev_data = dev->data;
1173         err = i40e_set_mac_type(hw);
1174         if (err) {
1175                 PMD_INIT_LOG(ERR, "set_mac_type failed: %d", err);
1176                 goto err;
1177         }
1178
1179         err = i40evf_check_vf_reset_done(hw);
1180         if (err)
1181                 goto err;
1182
1183         i40e_init_adminq_parameter(hw);
1184         err = i40e_init_adminq(hw);
1185         if (err) {
1186                 PMD_INIT_LOG(ERR, "init_adminq failed: %d", err);
1187                 goto err;
1188         }
1189
1190         /* Reset VF and wait until it's complete */
1191         if (i40evf_reset_vf(hw)) {
1192                 PMD_INIT_LOG(ERR, "reset NIC failed");
1193                 goto err_aq;
1194         }
1195
1196         /* VF reset, shutdown admin queue and initialize again */
1197         if (i40e_shutdown_adminq(hw) != I40E_SUCCESS) {
1198                 PMD_INIT_LOG(ERR, "i40e_shutdown_adminq failed");
1199                 goto err;
1200         }
1201
1202         i40e_init_adminq_parameter(hw);
1203         if (i40e_init_adminq(hw) != I40E_SUCCESS) {
1204                 PMD_INIT_LOG(ERR, "init_adminq failed");
1205                 goto err;
1206         }
1207
1208         vf->aq_resp = rte_zmalloc("vf_aq_resp", I40E_AQ_BUF_SZ, 0);
1209         if (!vf->aq_resp) {
1210                 PMD_INIT_LOG(ERR, "unable to allocate vf_aq_resp memory");
1211                 goto err_aq;
1212         }
1213         if (i40evf_check_api_version(dev) != 0) {
1214                 PMD_INIT_LOG(ERR, "check_api version failed");
1215                 goto err_api;
1216         }
1217         bufsz = sizeof(struct virtchnl_vf_resource) +
1218                 (I40E_MAX_VF_VSI * sizeof(struct virtchnl_vsi_resource));
1219         vf->vf_res = rte_zmalloc("vf_res", bufsz, 0);
1220         if (!vf->vf_res) {
1221                 PMD_INIT_LOG(ERR, "unable to allocate vf_res memory");
1222                 goto err_api;
1223         }
1224
1225         if (i40evf_get_vf_resource(dev) != 0) {
1226                 PMD_INIT_LOG(ERR, "i40evf_get_vf_config failed");
1227                 goto err_alloc;
1228         }
1229
1230         /* got VF config message back from PF, now we can parse it */
1231         for (i = 0; i < vf->vf_res->num_vsis; i++) {
1232                 if (vf->vf_res->vsi_res[i].vsi_type == VIRTCHNL_VSI_SRIOV)
1233                         vf->vsi_res = &vf->vf_res->vsi_res[i];
1234         }
1235
1236         if (!vf->vsi_res) {
1237                 PMD_INIT_LOG(ERR, "no LAN VSI found");
1238                 goto err_alloc;
1239         }
1240
1241         if (hw->mac.type == I40E_MAC_X722_VF)
1242                 vf->flags = I40E_FLAG_RSS_AQ_CAPABLE;
1243         vf->vsi.vsi_id = vf->vsi_res->vsi_id;
1244
1245         switch (vf->vsi_res->vsi_type) {
1246         case VIRTCHNL_VSI_SRIOV:
1247                 vf->vsi.type = I40E_VSI_SRIOV;
1248                 break;
1249         default:
1250                 vf->vsi.type = I40E_VSI_TYPE_UNKNOWN;
1251                 break;
1252         }
1253         vf->vsi.nb_qps = vf->vsi_res->num_queue_pairs;
1254         vf->vsi.adapter = I40E_DEV_PRIVATE_TO_ADAPTER(dev->data->dev_private);
1255
1256         /* Store the MAC address configured by host, or generate random one */
1257         if (is_valid_assigned_ether_addr((struct ether_addr *)hw->mac.addr))
1258                 vf->flags |= I40E_FLAG_VF_MAC_BY_PF;
1259         else
1260                 eth_random_addr(hw->mac.addr); /* Generate a random one */
1261
1262         I40E_WRITE_REG(hw, I40E_VFINT_DYN_CTL01,
1263                        (I40E_ITR_INDEX_DEFAULT <<
1264                         I40E_VFINT_DYN_CTL0_ITR_INDX_SHIFT) |
1265                        (interval <<
1266                         I40E_VFINT_DYN_CTL0_INTERVAL_SHIFT));
1267         I40EVF_WRITE_FLUSH(hw);
1268
1269         return 0;
1270
1271 err_alloc:
1272         rte_free(vf->vf_res);
1273         vf->vsi_res = NULL;
1274 err_api:
1275         rte_free(vf->aq_resp);
1276 err_aq:
1277         i40e_shutdown_adminq(hw); /* ignore error */
1278 err:
1279         return -1;
1280 }
1281
1282 static int
1283 i40evf_uninit_vf(struct rte_eth_dev *dev)
1284 {
1285         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
1286         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1287
1288         PMD_INIT_FUNC_TRACE();
1289
1290         if (hw->adapter_stopped == 0)
1291                 i40evf_dev_close(dev);
1292         rte_free(vf->vf_res);
1293         vf->vf_res = NULL;
1294         rte_free(vf->aq_resp);
1295         vf->aq_resp = NULL;
1296
1297         return 0;
1298 }
1299
1300 static void
1301 i40evf_handle_pf_event(struct rte_eth_dev *dev, uint8_t *msg,
1302                 __rte_unused uint16_t msglen)
1303 {
1304         struct virtchnl_pf_event *pf_msg =
1305                         (struct virtchnl_pf_event *)msg;
1306         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
1307
1308         switch (pf_msg->event) {
1309         case VIRTCHNL_EVENT_RESET_IMPENDING:
1310                 PMD_DRV_LOG(DEBUG, "VIRTCHNL_EVENT_RESET_IMPENDING event");
1311                 _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_RESET,
1312                                               NULL, NULL);
1313                 break;
1314         case VIRTCHNL_EVENT_LINK_CHANGE:
1315                 PMD_DRV_LOG(DEBUG, "VIRTCHNL_EVENT_LINK_CHANGE event");
1316                 vf->link_up = pf_msg->event_data.link_event.link_status;
1317                 vf->link_speed = pf_msg->event_data.link_event.link_speed;
1318                 break;
1319         case VIRTCHNL_EVENT_PF_DRIVER_CLOSE:
1320                 PMD_DRV_LOG(DEBUG, "VIRTCHNL_EVENT_PF_DRIVER_CLOSE event");
1321                 break;
1322         default:
1323                 PMD_DRV_LOG(ERR, " unknown event received %u", pf_msg->event);
1324                 break;
1325         }
1326 }
1327
1328 static void
1329 i40evf_handle_aq_msg(struct rte_eth_dev *dev)
1330 {
1331         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1332         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
1333         struct i40e_arq_event_info info;
1334         uint16_t pending, aq_opc;
1335         enum virtchnl_ops msg_opc;
1336         enum i40e_status_code msg_ret;
1337         int ret;
1338
1339         info.buf_len = I40E_AQ_BUF_SZ;
1340         if (!vf->aq_resp) {
1341                 PMD_DRV_LOG(ERR, "Buffer for adminq resp should not be NULL");
1342                 return;
1343         }
1344         info.msg_buf = vf->aq_resp;
1345
1346         pending = 1;
1347         while (pending) {
1348                 ret = i40e_clean_arq_element(hw, &info, &pending);
1349
1350                 if (ret != I40E_SUCCESS) {
1351                         PMD_DRV_LOG(INFO, "Failed to read msg from AdminQ,"
1352                                     "ret: %d", ret);
1353                         break;
1354                 }
1355                 aq_opc = rte_le_to_cpu_16(info.desc.opcode);
1356                 /* For the message sent from pf to vf, opcode is stored in
1357                  * cookie_high of struct i40e_aq_desc, while return error code
1358                  * are stored in cookie_low, Which is done by
1359                  * i40e_aq_send_msg_to_vf in PF driver.*/
1360                 msg_opc = (enum virtchnl_ops)rte_le_to_cpu_32(
1361                                                   info.desc.cookie_high);
1362                 msg_ret = (enum i40e_status_code)rte_le_to_cpu_32(
1363                                                   info.desc.cookie_low);
1364                 switch (aq_opc) {
1365                 case i40e_aqc_opc_send_msg_to_vf:
1366                         if (msg_opc == VIRTCHNL_OP_EVENT)
1367                                 /* process event*/
1368                                 i40evf_handle_pf_event(dev, info.msg_buf,
1369                                                        info.msg_len);
1370                         else {
1371                                 /* read message and it's expected one */
1372                                 if (msg_opc == vf->pend_cmd) {
1373                                         vf->cmd_retval = msg_ret;
1374                                         /* prevent compiler reordering */
1375                                         rte_compiler_barrier();
1376                                         _clear_cmd(vf);
1377                                 } else
1378                                         PMD_DRV_LOG(ERR, "command mismatch,"
1379                                                 "expect %u, get %u",
1380                                                 vf->pend_cmd, msg_opc);
1381                                 PMD_DRV_LOG(DEBUG, "adminq response is received,"
1382                                              " opcode = %d", msg_opc);
1383                         }
1384                         break;
1385                 default:
1386                         PMD_DRV_LOG(ERR, "Request %u is not supported yet",
1387                                     aq_opc);
1388                         break;
1389                 }
1390         }
1391 }
1392
1393 /**
1394  * Interrupt handler triggered by NIC  for handling
1395  * specific interrupt. Only adminq interrupt is processed in VF.
1396  *
1397  * @param handle
1398  *  Pointer to interrupt handle.
1399  * @param param
1400  *  The address of parameter (struct rte_eth_dev *) regsitered before.
1401  *
1402  * @return
1403  *  void
1404  */
1405 static void
1406 i40evf_dev_interrupt_handler(void *param)
1407 {
1408         struct rte_eth_dev *dev = (struct rte_eth_dev *)param;
1409         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1410         uint32_t icr0;
1411
1412         i40evf_disable_irq0(hw);
1413
1414         /* read out interrupt causes */
1415         icr0 = I40E_READ_REG(hw, I40E_VFINT_ICR01);
1416
1417         /* No interrupt event indicated */
1418         if (!(icr0 & I40E_VFINT_ICR01_INTEVENT_MASK)) {
1419                 PMD_DRV_LOG(DEBUG, "No interrupt event, nothing to do");
1420                 goto done;
1421         }
1422
1423         if (icr0 & I40E_VFINT_ICR01_ADMINQ_MASK) {
1424                 PMD_DRV_LOG(DEBUG, "ICR01_ADMINQ is reported");
1425                 i40evf_handle_aq_msg(dev);
1426         }
1427
1428         /* Link Status Change interrupt */
1429         if (icr0 & I40E_VFINT_ICR01_LINK_STAT_CHANGE_MASK)
1430                 PMD_DRV_LOG(DEBUG, "LINK_STAT_CHANGE is reported,"
1431                                    " do nothing");
1432
1433 done:
1434         i40evf_enable_irq0(hw);
1435 }
1436
1437 static int
1438 i40evf_dev_init(struct rte_eth_dev *eth_dev)
1439 {
1440         struct i40e_hw *hw
1441                 = I40E_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
1442         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1443
1444         PMD_INIT_FUNC_TRACE();
1445
1446         /* assign ops func pointer */
1447         eth_dev->dev_ops = &i40evf_eth_dev_ops;
1448         eth_dev->rx_pkt_burst = &i40e_recv_pkts;
1449         eth_dev->tx_pkt_burst = &i40e_xmit_pkts;
1450
1451         /*
1452          * For secondary processes, we don't initialise any further as primary
1453          * has already done this work.
1454          */
1455         if (rte_eal_process_type() != RTE_PROC_PRIMARY){
1456                 i40e_set_rx_function(eth_dev);
1457                 i40e_set_tx_function(eth_dev);
1458                 return 0;
1459         }
1460         i40e_set_default_ptype_table(eth_dev);
1461         i40e_set_default_pctype_table(eth_dev);
1462         rte_eth_copy_pci_info(eth_dev, pci_dev);
1463
1464         hw->vendor_id = pci_dev->id.vendor_id;
1465         hw->device_id = pci_dev->id.device_id;
1466         hw->subsystem_vendor_id = pci_dev->id.subsystem_vendor_id;
1467         hw->subsystem_device_id = pci_dev->id.subsystem_device_id;
1468         hw->bus.device = pci_dev->addr.devid;
1469         hw->bus.func = pci_dev->addr.function;
1470         hw->hw_addr = (void *)pci_dev->mem_resource[0].addr;
1471         hw->adapter_stopped = 0;
1472
1473         if(i40evf_init_vf(eth_dev) != 0) {
1474                 PMD_INIT_LOG(ERR, "Init vf failed");
1475                 return -1;
1476         }
1477
1478         /* register callback func to eal lib */
1479         rte_intr_callback_register(&pci_dev->intr_handle,
1480                 i40evf_dev_interrupt_handler, (void *)eth_dev);
1481
1482         /* enable uio intr after callback register */
1483         rte_intr_enable(&pci_dev->intr_handle);
1484
1485         /* configure and enable device interrupt */
1486         i40evf_enable_irq0(hw);
1487
1488         /* copy mac addr */
1489         eth_dev->data->mac_addrs = rte_zmalloc("i40evf_mac",
1490                                         ETHER_ADDR_LEN * I40E_NUM_MACADDR_MAX,
1491                                         0);
1492         if (eth_dev->data->mac_addrs == NULL) {
1493                 PMD_INIT_LOG(ERR, "Failed to allocate %d bytes needed to"
1494                                 " store MAC addresses",
1495                                 ETHER_ADDR_LEN * I40E_NUM_MACADDR_MAX);
1496                 return -ENOMEM;
1497         }
1498         ether_addr_copy((struct ether_addr *)hw->mac.addr,
1499                         &eth_dev->data->mac_addrs[0]);
1500
1501         return 0;
1502 }
1503
1504 static int
1505 i40evf_dev_uninit(struct rte_eth_dev *eth_dev)
1506 {
1507         PMD_INIT_FUNC_TRACE();
1508
1509         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1510                 return -EPERM;
1511
1512         eth_dev->dev_ops = NULL;
1513         eth_dev->rx_pkt_burst = NULL;
1514         eth_dev->tx_pkt_burst = NULL;
1515
1516         if (i40evf_uninit_vf(eth_dev) != 0) {
1517                 PMD_INIT_LOG(ERR, "i40evf_uninit_vf failed");
1518                 return -1;
1519         }
1520
1521         rte_free(eth_dev->data->mac_addrs);
1522         eth_dev->data->mac_addrs = NULL;
1523
1524         return 0;
1525 }
1526
1527 static int eth_i40evf_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1528         struct rte_pci_device *pci_dev)
1529 {
1530         return rte_eth_dev_pci_generic_probe(pci_dev,
1531                 sizeof(struct i40e_adapter), i40evf_dev_init);
1532 }
1533
1534 static int eth_i40evf_pci_remove(struct rte_pci_device *pci_dev)
1535 {
1536         return rte_eth_dev_pci_generic_remove(pci_dev, i40evf_dev_uninit);
1537 }
1538
1539 /*
1540  * virtual function driver struct
1541  */
1542 static struct rte_pci_driver rte_i40evf_pmd = {
1543         .id_table = pci_id_i40evf_map,
1544         .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_IOVA_AS_VA,
1545         .probe = eth_i40evf_pci_probe,
1546         .remove = eth_i40evf_pci_remove,
1547 };
1548
1549 RTE_PMD_REGISTER_PCI(net_i40e_vf, rte_i40evf_pmd);
1550 RTE_PMD_REGISTER_PCI_TABLE(net_i40e_vf, pci_id_i40evf_map);
1551 RTE_PMD_REGISTER_KMOD_DEP(net_i40e_vf, "* igb_uio | vfio-pci");
1552
1553 static int
1554 i40evf_dev_configure(struct rte_eth_dev *dev)
1555 {
1556         struct i40e_adapter *ad =
1557                 I40E_DEV_PRIVATE_TO_ADAPTER(dev->data->dev_private);
1558         struct rte_eth_conf *conf = &dev->data->dev_conf;
1559         struct i40e_vf *vf;
1560
1561         /* Initialize to TRUE. If any of Rx queues doesn't meet the bulk
1562          * allocation or vector Rx preconditions we will reset it.
1563          */
1564         ad->rx_bulk_alloc_allowed = true;
1565         ad->rx_vec_allowed = true;
1566         ad->tx_simple_allowed = true;
1567         ad->tx_vec_allowed = true;
1568
1569         /* For non-DPDK PF drivers, VF has no ability to disable HW
1570          * CRC strip, and is implicitly enabled by the PF.
1571          */
1572         if (!conf->rxmode.hw_strip_crc) {
1573                 vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
1574                 if ((vf->version_major == VIRTCHNL_VERSION_MAJOR) &&
1575                     (vf->version_minor <= VIRTCHNL_VERSION_MINOR)) {
1576                         /* Peer is running non-DPDK PF driver. */
1577                         PMD_INIT_LOG(ERR, "VF can't disable HW CRC Strip");
1578                         return -EINVAL;
1579                 }
1580         }
1581
1582         return i40evf_init_vlan(dev);
1583 }
1584
1585 static int
1586 i40evf_init_vlan(struct rte_eth_dev *dev)
1587 {
1588         /* Apply vlan offload setting */
1589         return i40evf_vlan_offload_set(dev, ETH_VLAN_STRIP_MASK);
1590 }
1591
1592 static int
1593 i40evf_vlan_offload_set(struct rte_eth_dev *dev, int mask)
1594 {
1595         struct rte_eth_conf *dev_conf = &dev->data->dev_conf;
1596
1597         /* Vlan stripping setting */
1598         if (mask & ETH_VLAN_STRIP_MASK) {
1599                 /* Enable or disable VLAN stripping */
1600                 if (dev_conf->rxmode.hw_vlan_strip)
1601                         i40evf_enable_vlan_strip(dev);
1602                 else
1603                         i40evf_disable_vlan_strip(dev);
1604         }
1605
1606         return 0;
1607 }
1608
1609 static int
1610 i40evf_dev_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1611 {
1612         struct i40e_rx_queue *rxq;
1613         int err = 0;
1614         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1615
1616         PMD_INIT_FUNC_TRACE();
1617
1618         if (rx_queue_id < dev->data->nb_rx_queues) {
1619                 rxq = dev->data->rx_queues[rx_queue_id];
1620
1621                 err = i40e_alloc_rx_queue_mbufs(rxq);
1622                 if (err) {
1623                         PMD_DRV_LOG(ERR, "Failed to allocate RX queue mbuf");
1624                         return err;
1625                 }
1626
1627                 rte_wmb();
1628
1629                 /* Init the RX tail register. */
1630                 I40E_PCI_REG_WRITE(rxq->qrx_tail, rxq->nb_rx_desc - 1);
1631                 I40EVF_WRITE_FLUSH(hw);
1632
1633                 /* Ready to switch the queue on */
1634                 err = i40evf_switch_queue(dev, TRUE, rx_queue_id, TRUE);
1635
1636                 if (err)
1637                         PMD_DRV_LOG(ERR, "Failed to switch RX queue %u on",
1638                                     rx_queue_id);
1639                 else
1640                         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1641         }
1642
1643         return err;
1644 }
1645
1646 static int
1647 i40evf_dev_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1648 {
1649         struct i40e_rx_queue *rxq;
1650         int err;
1651
1652         if (rx_queue_id < dev->data->nb_rx_queues) {
1653                 rxq = dev->data->rx_queues[rx_queue_id];
1654
1655                 err = i40evf_switch_queue(dev, TRUE, rx_queue_id, FALSE);
1656
1657                 if (err) {
1658                         PMD_DRV_LOG(ERR, "Failed to switch RX queue %u off",
1659                                     rx_queue_id);
1660                         return err;
1661                 }
1662
1663                 i40e_rx_queue_release_mbufs(rxq);
1664                 i40e_reset_rx_queue(rxq);
1665                 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1666         }
1667
1668         return 0;
1669 }
1670
1671 static int
1672 i40evf_dev_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1673 {
1674         int err = 0;
1675
1676         PMD_INIT_FUNC_TRACE();
1677
1678         if (tx_queue_id < dev->data->nb_tx_queues) {
1679
1680                 /* Ready to switch the queue on */
1681                 err = i40evf_switch_queue(dev, FALSE, tx_queue_id, TRUE);
1682
1683                 if (err)
1684                         PMD_DRV_LOG(ERR, "Failed to switch TX queue %u on",
1685                                     tx_queue_id);
1686                 else
1687                         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1688         }
1689
1690         return err;
1691 }
1692
1693 static int
1694 i40evf_dev_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1695 {
1696         struct i40e_tx_queue *txq;
1697         int err;
1698
1699         if (tx_queue_id < dev->data->nb_tx_queues) {
1700                 txq = dev->data->tx_queues[tx_queue_id];
1701
1702                 err = i40evf_switch_queue(dev, FALSE, tx_queue_id, FALSE);
1703
1704                 if (err) {
1705                         PMD_DRV_LOG(ERR, "Failed to switch TX queue %u off",
1706                                     tx_queue_id);
1707                         return err;
1708                 }
1709
1710                 i40e_tx_queue_release_mbufs(txq);
1711                 i40e_reset_tx_queue(txq);
1712                 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1713         }
1714
1715         return 0;
1716 }
1717
1718 static int
1719 i40evf_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1720 {
1721         int ret;
1722
1723         if (on)
1724                 ret = i40evf_add_vlan(dev, vlan_id);
1725         else
1726                 ret = i40evf_del_vlan(dev,vlan_id);
1727
1728         return ret;
1729 }
1730
1731 static int
1732 i40evf_rxq_init(struct rte_eth_dev *dev, struct i40e_rx_queue *rxq)
1733 {
1734         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1735         struct rte_eth_dev_data *dev_data = dev->data;
1736         struct rte_pktmbuf_pool_private *mbp_priv;
1737         uint16_t buf_size, len;
1738
1739         rxq->qrx_tail = hw->hw_addr + I40E_QRX_TAIL1(rxq->queue_id);
1740         I40E_PCI_REG_WRITE(rxq->qrx_tail, rxq->nb_rx_desc - 1);
1741         I40EVF_WRITE_FLUSH(hw);
1742
1743         /* Calculate the maximum packet length allowed */
1744         mbp_priv = rte_mempool_get_priv(rxq->mp);
1745         buf_size = (uint16_t)(mbp_priv->mbuf_data_room_size -
1746                                         RTE_PKTMBUF_HEADROOM);
1747         rxq->hs_mode = i40e_header_split_none;
1748         rxq->rx_hdr_len = 0;
1749         rxq->rx_buf_len = RTE_ALIGN(buf_size, (1 << I40E_RXQ_CTX_DBUFF_SHIFT));
1750         len = rxq->rx_buf_len * I40E_MAX_CHAINED_RX_BUFFERS;
1751         rxq->max_pkt_len = RTE_MIN(len,
1752                 dev_data->dev_conf.rxmode.max_rx_pkt_len);
1753
1754         /**
1755          * Check if the jumbo frame and maximum packet length are set correctly
1756          */
1757         if (dev_data->dev_conf.rxmode.jumbo_frame == 1) {
1758                 if (rxq->max_pkt_len <= ETHER_MAX_LEN ||
1759                     rxq->max_pkt_len > I40E_FRAME_SIZE_MAX) {
1760                         PMD_DRV_LOG(ERR, "maximum packet length must be "
1761                                 "larger than %u and smaller than %u, as jumbo "
1762                                 "frame is enabled", (uint32_t)ETHER_MAX_LEN,
1763                                         (uint32_t)I40E_FRAME_SIZE_MAX);
1764                         return I40E_ERR_CONFIG;
1765                 }
1766         } else {
1767                 if (rxq->max_pkt_len < ETHER_MIN_LEN ||
1768                     rxq->max_pkt_len > ETHER_MAX_LEN) {
1769                         PMD_DRV_LOG(ERR, "maximum packet length must be "
1770                                 "larger than %u and smaller than %u, as jumbo "
1771                                 "frame is disabled", (uint32_t)ETHER_MIN_LEN,
1772                                                 (uint32_t)ETHER_MAX_LEN);
1773                         return I40E_ERR_CONFIG;
1774                 }
1775         }
1776
1777         if (dev_data->dev_conf.rxmode.enable_scatter ||
1778             (rxq->max_pkt_len + 2 * I40E_VLAN_TAG_SIZE) > buf_size) {
1779                 dev_data->scattered_rx = 1;
1780         }
1781
1782         return 0;
1783 }
1784
1785 static int
1786 i40evf_rx_init(struct rte_eth_dev *dev)
1787 {
1788         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
1789         uint16_t i;
1790         int ret = I40E_SUCCESS;
1791         struct i40e_rx_queue **rxq =
1792                 (struct i40e_rx_queue **)dev->data->rx_queues;
1793
1794         i40evf_config_rss(vf);
1795         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1796                 if (!rxq[i] || !rxq[i]->q_set)
1797                         continue;
1798                 ret = i40evf_rxq_init(dev, rxq[i]);
1799                 if (ret != I40E_SUCCESS)
1800                         break;
1801         }
1802         if (ret == I40E_SUCCESS)
1803                 i40e_set_rx_function(dev);
1804
1805         return ret;
1806 }
1807
1808 static void
1809 i40evf_tx_init(struct rte_eth_dev *dev)
1810 {
1811         uint16_t i;
1812         struct i40e_tx_queue **txq =
1813                 (struct i40e_tx_queue **)dev->data->tx_queues;
1814         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1815
1816         for (i = 0; i < dev->data->nb_tx_queues; i++)
1817                 txq[i]->qtx_tail = hw->hw_addr + I40E_QTX_TAIL1(i);
1818
1819         i40e_set_tx_function(dev);
1820 }
1821
1822 static inline void
1823 i40evf_enable_queues_intr(struct rte_eth_dev *dev)
1824 {
1825         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1826         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1827         struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
1828
1829         if (!rte_intr_allow_others(intr_handle)) {
1830                 I40E_WRITE_REG(hw,
1831                                I40E_VFINT_DYN_CTL01,
1832                                I40E_VFINT_DYN_CTL01_INTENA_MASK |
1833                                I40E_VFINT_DYN_CTL01_CLEARPBA_MASK |
1834                                I40E_VFINT_DYN_CTL01_ITR_INDX_MASK);
1835                 I40EVF_WRITE_FLUSH(hw);
1836                 return;
1837         }
1838
1839         I40EVF_WRITE_FLUSH(hw);
1840 }
1841
1842 static inline void
1843 i40evf_disable_queues_intr(struct rte_eth_dev *dev)
1844 {
1845         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1846         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1847         struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
1848
1849         if (!rte_intr_allow_others(intr_handle)) {
1850                 I40E_WRITE_REG(hw, I40E_VFINT_DYN_CTL01,
1851                                I40E_VFINT_DYN_CTL01_ITR_INDX_MASK);
1852                 I40EVF_WRITE_FLUSH(hw);
1853                 return;
1854         }
1855
1856         I40EVF_WRITE_FLUSH(hw);
1857 }
1858
1859 static int
1860 i40evf_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
1861 {
1862         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1863         struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
1864         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1865         uint16_t interval =
1866                 i40e_calc_itr_interval(RTE_LIBRTE_I40E_ITR_INTERVAL);
1867         uint16_t msix_intr;
1868
1869         msix_intr = intr_handle->intr_vec[queue_id];
1870         if (msix_intr == I40E_MISC_VEC_ID)
1871                 I40E_WRITE_REG(hw, I40E_VFINT_DYN_CTL01,
1872                                I40E_VFINT_DYN_CTL01_INTENA_MASK |
1873                                I40E_VFINT_DYN_CTL01_CLEARPBA_MASK |
1874                                (0 << I40E_VFINT_DYN_CTL01_ITR_INDX_SHIFT) |
1875                                (interval <<
1876                                 I40E_VFINT_DYN_CTL01_INTERVAL_SHIFT));
1877         else
1878                 I40E_WRITE_REG(hw,
1879                                I40E_VFINT_DYN_CTLN1(msix_intr -
1880                                                     I40E_RX_VEC_START),
1881                                I40E_VFINT_DYN_CTLN1_INTENA_MASK |
1882                                I40E_VFINT_DYN_CTLN1_CLEARPBA_MASK |
1883                                (0 << I40E_VFINT_DYN_CTLN1_ITR_INDX_SHIFT) |
1884                                (interval <<
1885                                 I40E_VFINT_DYN_CTLN1_INTERVAL_SHIFT));
1886
1887         I40EVF_WRITE_FLUSH(hw);
1888
1889         rte_intr_enable(&pci_dev->intr_handle);
1890
1891         return 0;
1892 }
1893
1894 static int
1895 i40evf_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
1896 {
1897         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1898         struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
1899         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1900         uint16_t msix_intr;
1901
1902         msix_intr = intr_handle->intr_vec[queue_id];
1903         if (msix_intr == I40E_MISC_VEC_ID)
1904                 I40E_WRITE_REG(hw, I40E_VFINT_DYN_CTL01, 0);
1905         else
1906                 I40E_WRITE_REG(hw,
1907                                I40E_VFINT_DYN_CTLN1(msix_intr -
1908                                                     I40E_RX_VEC_START),
1909                                0);
1910
1911         I40EVF_WRITE_FLUSH(hw);
1912
1913         return 0;
1914 }
1915
1916 static void
1917 i40evf_add_del_all_mac_addr(struct rte_eth_dev *dev, bool add)
1918 {
1919         struct virtchnl_ether_addr_list *list;
1920         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
1921         int err, i, j;
1922         int next_begin = 0;
1923         int begin = 0;
1924         uint32_t len;
1925         struct ether_addr *addr;
1926         struct vf_cmd_info args;
1927
1928         do {
1929                 j = 0;
1930                 len = sizeof(struct virtchnl_ether_addr_list);
1931                 for (i = begin; i < I40E_NUM_MACADDR_MAX; i++, next_begin++) {
1932                         if (is_zero_ether_addr(&dev->data->mac_addrs[i]))
1933                                 continue;
1934                         len += sizeof(struct virtchnl_ether_addr);
1935                         if (len >= I40E_AQ_BUF_SZ) {
1936                                 next_begin = i + 1;
1937                                 break;
1938                         }
1939                 }
1940
1941                 list = rte_zmalloc("i40evf_del_mac_buffer", len, 0);
1942                 if (!list) {
1943                         PMD_DRV_LOG(ERR, "fail to allocate memory");
1944                         return;
1945                 }
1946
1947                 for (i = begin; i < next_begin; i++) {
1948                         addr = &dev->data->mac_addrs[i];
1949                         if (is_zero_ether_addr(addr))
1950                                 continue;
1951                         rte_memcpy(list->list[j].addr, addr->addr_bytes,
1952                                          sizeof(addr->addr_bytes));
1953                         PMD_DRV_LOG(DEBUG, "add/rm mac:%x:%x:%x:%x:%x:%x",
1954                                     addr->addr_bytes[0], addr->addr_bytes[1],
1955                                     addr->addr_bytes[2], addr->addr_bytes[3],
1956                                     addr->addr_bytes[4], addr->addr_bytes[5]);
1957                         j++;
1958                 }
1959                 list->vsi_id = vf->vsi_res->vsi_id;
1960                 list->num_elements = j;
1961                 args.ops = add ? VIRTCHNL_OP_ADD_ETH_ADDR :
1962                            VIRTCHNL_OP_DEL_ETH_ADDR;
1963                 args.in_args = (uint8_t *)list;
1964                 args.in_args_size = len;
1965                 args.out_buffer = vf->aq_resp;
1966                 args.out_size = I40E_AQ_BUF_SZ;
1967                 err = i40evf_execute_vf_cmd(dev, &args);
1968                 if (err) {
1969                         PMD_DRV_LOG(ERR, "fail to execute command %s",
1970                                     add ? "OP_ADD_ETHER_ADDRESS" :
1971                                     "OP_DEL_ETHER_ADDRESS");
1972                 } else {
1973                         if (add)
1974                                 vf->vsi.mac_num++;
1975                         else
1976                                 vf->vsi.mac_num--;
1977                 }
1978                 rte_free(list);
1979                 begin = next_begin;
1980         } while (begin < I40E_NUM_MACADDR_MAX);
1981 }
1982
1983 static int
1984 i40evf_dev_start(struct rte_eth_dev *dev)
1985 {
1986         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
1987         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1988         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1989         struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
1990         uint32_t intr_vector = 0;
1991
1992         PMD_INIT_FUNC_TRACE();
1993
1994         hw->adapter_stopped = 0;
1995
1996         vf->max_pkt_len = dev->data->dev_conf.rxmode.max_rx_pkt_len;
1997         vf->num_queue_pairs = RTE_MAX(dev->data->nb_rx_queues,
1998                                         dev->data->nb_tx_queues);
1999
2000         /* check and configure queue intr-vector mapping */
2001         if (dev->data->dev_conf.intr_conf.rxq != 0) {
2002                 intr_vector = dev->data->nb_rx_queues;
2003                 if (rte_intr_efd_enable(intr_handle, intr_vector))
2004                         return -1;
2005         }
2006
2007         if (rte_intr_dp_is_en(intr_handle) && !intr_handle->intr_vec) {
2008                 intr_handle->intr_vec =
2009                         rte_zmalloc("intr_vec",
2010                                     dev->data->nb_rx_queues * sizeof(int), 0);
2011                 if (!intr_handle->intr_vec) {
2012                         PMD_INIT_LOG(ERR, "Failed to allocate %d rx_queues"
2013                                      " intr_vec", dev->data->nb_rx_queues);
2014                         return -ENOMEM;
2015                 }
2016         }
2017
2018         if (i40evf_rx_init(dev) != 0){
2019                 PMD_DRV_LOG(ERR, "failed to do RX init");
2020                 return -1;
2021         }
2022
2023         i40evf_tx_init(dev);
2024
2025         if (i40evf_configure_vsi_queues(dev) != 0) {
2026                 PMD_DRV_LOG(ERR, "configure queues failed");
2027                 goto err_queue;
2028         }
2029         if (i40evf_config_irq_map(dev)) {
2030                 PMD_DRV_LOG(ERR, "config_irq_map failed");
2031                 goto err_queue;
2032         }
2033
2034         /* Set all mac addrs */
2035         i40evf_add_del_all_mac_addr(dev, TRUE);
2036
2037         if (i40evf_start_queues(dev) != 0) {
2038                 PMD_DRV_LOG(ERR, "enable queues failed");
2039                 goto err_mac;
2040         }
2041
2042         /* When a VF port is bound to VFIO-PCI, only miscellaneous interrupt
2043          * is mapped to VFIO vector 0 in i40evf_dev_init( ).
2044          * If previous VFIO interrupt mapping set in i40evf_dev_init( ) is
2045          * not cleared, it will fail when rte_intr_enable( ) tries to map Rx
2046          * queue interrupt to other VFIO vectors.
2047          * So clear uio/vfio intr/evevnfd first to avoid failure.
2048          */
2049         if (dev->data->dev_conf.intr_conf.rxq != 0) {
2050                 rte_intr_disable(intr_handle);
2051                 rte_intr_enable(intr_handle);
2052         }
2053
2054         i40evf_enable_queues_intr(dev);
2055
2056         return 0;
2057
2058 err_mac:
2059         i40evf_add_del_all_mac_addr(dev, FALSE);
2060 err_queue:
2061         return -1;
2062 }
2063
2064 static void
2065 i40evf_dev_stop(struct rte_eth_dev *dev)
2066 {
2067         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2068         struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
2069         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2070
2071         PMD_INIT_FUNC_TRACE();
2072
2073         if (hw->adapter_stopped == 1)
2074                 return;
2075         i40evf_stop_queues(dev);
2076         i40evf_disable_queues_intr(dev);
2077         i40e_dev_clear_queues(dev);
2078
2079         /* Clean datapath event and queue/vec mapping */
2080         rte_intr_efd_disable(intr_handle);
2081         if (intr_handle->intr_vec) {
2082                 rte_free(intr_handle->intr_vec);
2083                 intr_handle->intr_vec = NULL;
2084         }
2085         /* remove all mac addrs */
2086         i40evf_add_del_all_mac_addr(dev, FALSE);
2087         hw->adapter_stopped = 1;
2088
2089 }
2090
2091 static int
2092 i40evf_dev_link_update(struct rte_eth_dev *dev,
2093                        __rte_unused int wait_to_complete)
2094 {
2095         struct rte_eth_link new_link;
2096         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2097         /*
2098          * DPDK pf host provide interfacet to acquire link status
2099          * while Linux driver does not
2100          */
2101
2102         /* Linux driver PF host */
2103         switch (vf->link_speed) {
2104         case I40E_LINK_SPEED_100MB:
2105                 new_link.link_speed = ETH_SPEED_NUM_100M;
2106                 break;
2107         case I40E_LINK_SPEED_1GB:
2108                 new_link.link_speed = ETH_SPEED_NUM_1G;
2109                 break;
2110         case I40E_LINK_SPEED_10GB:
2111                 new_link.link_speed = ETH_SPEED_NUM_10G;
2112                 break;
2113         case I40E_LINK_SPEED_20GB:
2114                 new_link.link_speed = ETH_SPEED_NUM_20G;
2115                 break;
2116         case I40E_LINK_SPEED_25GB:
2117                 new_link.link_speed = ETH_SPEED_NUM_25G;
2118                 break;
2119         case I40E_LINK_SPEED_40GB:
2120                 new_link.link_speed = ETH_SPEED_NUM_40G;
2121                 break;
2122         default:
2123                 new_link.link_speed = ETH_SPEED_NUM_100M;
2124                 break;
2125         }
2126         /* full duplex only */
2127         new_link.link_duplex = ETH_LINK_FULL_DUPLEX;
2128         new_link.link_status = vf->link_up ? ETH_LINK_UP :
2129                                              ETH_LINK_DOWN;
2130         new_link.link_autoneg =
2131                 dev->data->dev_conf.link_speeds & ETH_LINK_SPEED_FIXED;
2132
2133         i40evf_dev_atomic_write_link_status(dev, &new_link);
2134
2135         return 0;
2136 }
2137
2138 static void
2139 i40evf_dev_promiscuous_enable(struct rte_eth_dev *dev)
2140 {
2141         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2142         int ret;
2143
2144         /* If enabled, just return */
2145         if (vf->promisc_unicast_enabled)
2146                 return;
2147
2148         ret = i40evf_config_promisc(dev, 1, vf->promisc_multicast_enabled);
2149         if (ret == 0)
2150                 vf->promisc_unicast_enabled = TRUE;
2151 }
2152
2153 static void
2154 i40evf_dev_promiscuous_disable(struct rte_eth_dev *dev)
2155 {
2156         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2157         int ret;
2158
2159         /* If disabled, just return */
2160         if (!vf->promisc_unicast_enabled)
2161                 return;
2162
2163         ret = i40evf_config_promisc(dev, 0, vf->promisc_multicast_enabled);
2164         if (ret == 0)
2165                 vf->promisc_unicast_enabled = FALSE;
2166 }
2167
2168 static void
2169 i40evf_dev_allmulticast_enable(struct rte_eth_dev *dev)
2170 {
2171         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2172         int ret;
2173
2174         /* If enabled, just return */
2175         if (vf->promisc_multicast_enabled)
2176                 return;
2177
2178         ret = i40evf_config_promisc(dev, vf->promisc_unicast_enabled, 1);
2179         if (ret == 0)
2180                 vf->promisc_multicast_enabled = TRUE;
2181 }
2182
2183 static void
2184 i40evf_dev_allmulticast_disable(struct rte_eth_dev *dev)
2185 {
2186         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2187         int ret;
2188
2189         /* If enabled, just return */
2190         if (!vf->promisc_multicast_enabled)
2191                 return;
2192
2193         ret = i40evf_config_promisc(dev, vf->promisc_unicast_enabled, 0);
2194         if (ret == 0)
2195                 vf->promisc_multicast_enabled = FALSE;
2196 }
2197
2198 static void
2199 i40evf_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2200 {
2201         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2202
2203         memset(dev_info, 0, sizeof(*dev_info));
2204         dev_info->pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2205         dev_info->max_rx_queues = vf->vsi_res->num_queue_pairs;
2206         dev_info->max_tx_queues = vf->vsi_res->num_queue_pairs;
2207         dev_info->min_rx_bufsize = I40E_BUF_SIZE_MIN;
2208         dev_info->max_rx_pktlen = I40E_FRAME_SIZE_MAX;
2209         dev_info->hash_key_size = (I40E_VFQF_HKEY_MAX_INDEX + 1) * sizeof(uint32_t);
2210         dev_info->reta_size = ETH_RSS_RETA_SIZE_64;
2211         dev_info->flow_type_rss_offloads = vf->adapter->flow_types_mask;
2212         dev_info->max_mac_addrs = I40E_NUM_MACADDR_MAX;
2213         dev_info->rx_offload_capa =
2214                 DEV_RX_OFFLOAD_VLAN_STRIP |
2215                 DEV_RX_OFFLOAD_QINQ_STRIP |
2216                 DEV_RX_OFFLOAD_IPV4_CKSUM |
2217                 DEV_RX_OFFLOAD_UDP_CKSUM |
2218                 DEV_RX_OFFLOAD_TCP_CKSUM;
2219         dev_info->tx_offload_capa =
2220                 DEV_TX_OFFLOAD_VLAN_INSERT |
2221                 DEV_TX_OFFLOAD_QINQ_INSERT |
2222                 DEV_TX_OFFLOAD_IPV4_CKSUM |
2223                 DEV_TX_OFFLOAD_UDP_CKSUM |
2224                 DEV_TX_OFFLOAD_TCP_CKSUM |
2225                 DEV_TX_OFFLOAD_SCTP_CKSUM;
2226
2227         dev_info->default_rxconf = (struct rte_eth_rxconf) {
2228                 .rx_thresh = {
2229                         .pthresh = I40E_DEFAULT_RX_PTHRESH,
2230                         .hthresh = I40E_DEFAULT_RX_HTHRESH,
2231                         .wthresh = I40E_DEFAULT_RX_WTHRESH,
2232                 },
2233                 .rx_free_thresh = I40E_DEFAULT_RX_FREE_THRESH,
2234                 .rx_drop_en = 0,
2235         };
2236
2237         dev_info->default_txconf = (struct rte_eth_txconf) {
2238                 .tx_thresh = {
2239                         .pthresh = I40E_DEFAULT_TX_PTHRESH,
2240                         .hthresh = I40E_DEFAULT_TX_HTHRESH,
2241                         .wthresh = I40E_DEFAULT_TX_WTHRESH,
2242                 },
2243                 .tx_free_thresh = I40E_DEFAULT_TX_FREE_THRESH,
2244                 .tx_rs_thresh = I40E_DEFAULT_TX_RSBIT_THRESH,
2245                 .txq_flags = ETH_TXQ_FLAGS_NOMULTSEGS |
2246                                 ETH_TXQ_FLAGS_NOOFFLOADS,
2247         };
2248
2249         dev_info->rx_desc_lim = (struct rte_eth_desc_lim) {
2250                 .nb_max = I40E_MAX_RING_DESC,
2251                 .nb_min = I40E_MIN_RING_DESC,
2252                 .nb_align = I40E_ALIGN_RING_DESC,
2253         };
2254
2255         dev_info->tx_desc_lim = (struct rte_eth_desc_lim) {
2256                 .nb_max = I40E_MAX_RING_DESC,
2257                 .nb_min = I40E_MIN_RING_DESC,
2258                 .nb_align = I40E_ALIGN_RING_DESC,
2259         };
2260 }
2261
2262 static int
2263 i40evf_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
2264 {
2265         int ret;
2266         struct i40e_eth_stats *pstats = NULL;
2267         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2268         struct i40e_vsi *vsi = &vf->vsi;
2269
2270         ret = i40evf_query_stats(dev, &pstats);
2271         if (ret == 0) {
2272                 i40evf_update_stats(vsi, pstats);
2273
2274                 stats->ipackets = pstats->rx_unicast + pstats->rx_multicast +
2275                                                 pstats->rx_broadcast;
2276                 stats->opackets = pstats->tx_broadcast + pstats->tx_multicast +
2277                                                 pstats->tx_unicast;
2278                 stats->imissed = pstats->rx_discards;
2279                 stats->oerrors = pstats->tx_errors + pstats->tx_discards;
2280                 stats->ibytes = pstats->rx_bytes;
2281                 stats->obytes = pstats->tx_bytes;
2282         } else {
2283                 PMD_DRV_LOG(ERR, "Get statistics failed");
2284         }
2285         return ret;
2286 }
2287
2288 static void
2289 i40evf_dev_close(struct rte_eth_dev *dev)
2290 {
2291         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2292         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2293         struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
2294
2295         i40evf_dev_stop(dev);
2296         i40e_dev_free_queues(dev);
2297         i40evf_reset_vf(hw);
2298         i40e_shutdown_adminq(hw);
2299         /* disable uio intr before callback unregister */
2300         rte_intr_disable(intr_handle);
2301
2302         /* unregister callback func from eal lib */
2303         rte_intr_callback_unregister(intr_handle,
2304                                      i40evf_dev_interrupt_handler, dev);
2305         i40evf_disable_irq0(hw);
2306 }
2307
2308 /*
2309  * Reset VF device only to re-initialize resources in PMD layer
2310  */
2311 static int
2312 i40evf_dev_reset(struct rte_eth_dev *dev)
2313 {
2314         int ret;
2315
2316         ret = i40evf_dev_uninit(dev);
2317         if (ret)
2318                 return ret;
2319
2320         ret = i40evf_dev_init(dev);
2321
2322         return ret;
2323 }
2324
2325 static int
2326 i40evf_get_rss_lut(struct i40e_vsi *vsi, uint8_t *lut, uint16_t lut_size)
2327 {
2328         struct i40e_vf *vf = I40E_VSI_TO_VF(vsi);
2329         struct i40e_hw *hw = I40E_VSI_TO_HW(vsi);
2330         int ret;
2331
2332         if (!lut)
2333                 return -EINVAL;
2334
2335         if (vf->flags & I40E_FLAG_RSS_AQ_CAPABLE) {
2336                 ret = i40e_aq_get_rss_lut(hw, vsi->vsi_id, FALSE,
2337                                           lut, lut_size);
2338                 if (ret) {
2339                         PMD_DRV_LOG(ERR, "Failed to get RSS lookup table");
2340                         return ret;
2341                 }
2342         } else {
2343                 uint32_t *lut_dw = (uint32_t *)lut;
2344                 uint16_t i, lut_size_dw = lut_size / 4;
2345
2346                 for (i = 0; i < lut_size_dw; i++)
2347                         lut_dw[i] = I40E_READ_REG(hw, I40E_VFQF_HLUT(i));
2348         }
2349
2350         return 0;
2351 }
2352
2353 static int
2354 i40evf_set_rss_lut(struct i40e_vsi *vsi, uint8_t *lut, uint16_t lut_size)
2355 {
2356         struct i40e_vf *vf;
2357         struct i40e_hw *hw;
2358         int ret;
2359
2360         if (!vsi || !lut)
2361                 return -EINVAL;
2362
2363         vf = I40E_VSI_TO_VF(vsi);
2364         hw = I40E_VSI_TO_HW(vsi);
2365
2366         if (vf->flags & I40E_FLAG_RSS_AQ_CAPABLE) {
2367                 ret = i40e_aq_set_rss_lut(hw, vsi->vsi_id, FALSE,
2368                                           lut, lut_size);
2369                 if (ret) {
2370                         PMD_DRV_LOG(ERR, "Failed to set RSS lookup table");
2371                         return ret;
2372                 }
2373         } else {
2374                 uint32_t *lut_dw = (uint32_t *)lut;
2375                 uint16_t i, lut_size_dw = lut_size / 4;
2376
2377                 for (i = 0; i < lut_size_dw; i++)
2378                         I40E_WRITE_REG(hw, I40E_VFQF_HLUT(i), lut_dw[i]);
2379                 I40EVF_WRITE_FLUSH(hw);
2380         }
2381
2382         return 0;
2383 }
2384
2385 static int
2386 i40evf_dev_rss_reta_update(struct rte_eth_dev *dev,
2387                            struct rte_eth_rss_reta_entry64 *reta_conf,
2388                            uint16_t reta_size)
2389 {
2390         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2391         uint8_t *lut;
2392         uint16_t i, idx, shift;
2393         int ret;
2394
2395         if (reta_size != ETH_RSS_RETA_SIZE_64) {
2396                 PMD_DRV_LOG(ERR, "The size of hash lookup table configured "
2397                         "(%d) doesn't match the number of hardware can "
2398                         "support (%d)", reta_size, ETH_RSS_RETA_SIZE_64);
2399                 return -EINVAL;
2400         }
2401
2402         lut = rte_zmalloc("i40e_rss_lut", reta_size, 0);
2403         if (!lut) {
2404                 PMD_DRV_LOG(ERR, "No memory can be allocated");
2405                 return -ENOMEM;
2406         }
2407         ret = i40evf_get_rss_lut(&vf->vsi, lut, reta_size);
2408         if (ret)
2409                 goto out;
2410         for (i = 0; i < reta_size; i++) {
2411                 idx = i / RTE_RETA_GROUP_SIZE;
2412                 shift = i % RTE_RETA_GROUP_SIZE;
2413                 if (reta_conf[idx].mask & (1ULL << shift))
2414                         lut[i] = reta_conf[idx].reta[shift];
2415         }
2416         ret = i40evf_set_rss_lut(&vf->vsi, lut, reta_size);
2417
2418 out:
2419         rte_free(lut);
2420
2421         return ret;
2422 }
2423
2424 static int
2425 i40evf_dev_rss_reta_query(struct rte_eth_dev *dev,
2426                           struct rte_eth_rss_reta_entry64 *reta_conf,
2427                           uint16_t reta_size)
2428 {
2429         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2430         uint16_t i, idx, shift;
2431         uint8_t *lut;
2432         int ret;
2433
2434         if (reta_size != ETH_RSS_RETA_SIZE_64) {
2435                 PMD_DRV_LOG(ERR, "The size of hash lookup table configured "
2436                         "(%d) doesn't match the number of hardware can "
2437                         "support (%d)", reta_size, ETH_RSS_RETA_SIZE_64);
2438                 return -EINVAL;
2439         }
2440
2441         lut = rte_zmalloc("i40e_rss_lut", reta_size, 0);
2442         if (!lut) {
2443                 PMD_DRV_LOG(ERR, "No memory can be allocated");
2444                 return -ENOMEM;
2445         }
2446
2447         ret = i40evf_get_rss_lut(&vf->vsi, lut, reta_size);
2448         if (ret)
2449                 goto out;
2450         for (i = 0; i < reta_size; i++) {
2451                 idx = i / RTE_RETA_GROUP_SIZE;
2452                 shift = i % RTE_RETA_GROUP_SIZE;
2453                 if (reta_conf[idx].mask & (1ULL << shift))
2454                         reta_conf[idx].reta[shift] = lut[i];
2455         }
2456
2457 out:
2458         rte_free(lut);
2459
2460         return ret;
2461 }
2462
2463 static int
2464 i40evf_set_rss_key(struct i40e_vsi *vsi, uint8_t *key, uint8_t key_len)
2465 {
2466         struct i40e_vf *vf = I40E_VSI_TO_VF(vsi);
2467         struct i40e_hw *hw = I40E_VSI_TO_HW(vsi);
2468         int ret = 0;
2469
2470         if (!key || key_len == 0) {
2471                 PMD_DRV_LOG(DEBUG, "No key to be configured");
2472                 return 0;
2473         } else if (key_len != (I40E_VFQF_HKEY_MAX_INDEX + 1) *
2474                 sizeof(uint32_t)) {
2475                 PMD_DRV_LOG(ERR, "Invalid key length %u", key_len);
2476                 return -EINVAL;
2477         }
2478
2479         if (vf->flags & I40E_FLAG_RSS_AQ_CAPABLE) {
2480                 struct i40e_aqc_get_set_rss_key_data *key_dw =
2481                         (struct i40e_aqc_get_set_rss_key_data *)key;
2482
2483                 ret = i40e_aq_set_rss_key(hw, vsi->vsi_id, key_dw);
2484                 if (ret)
2485                         PMD_INIT_LOG(ERR, "Failed to configure RSS key "
2486                                      "via AQ");
2487         } else {
2488                 uint32_t *hash_key = (uint32_t *)key;
2489                 uint16_t i;
2490
2491                 for (i = 0; i <= I40E_VFQF_HKEY_MAX_INDEX; i++)
2492                         i40e_write_rx_ctl(hw, I40E_VFQF_HKEY(i), hash_key[i]);
2493                 I40EVF_WRITE_FLUSH(hw);
2494         }
2495
2496         return ret;
2497 }
2498
2499 static int
2500 i40evf_get_rss_key(struct i40e_vsi *vsi, uint8_t *key, uint8_t *key_len)
2501 {
2502         struct i40e_vf *vf = I40E_VSI_TO_VF(vsi);
2503         struct i40e_hw *hw = I40E_VSI_TO_HW(vsi);
2504         int ret;
2505
2506         if (!key || !key_len)
2507                 return -EINVAL;
2508
2509         if (vf->flags & I40E_FLAG_RSS_AQ_CAPABLE) {
2510                 ret = i40e_aq_get_rss_key(hw, vsi->vsi_id,
2511                         (struct i40e_aqc_get_set_rss_key_data *)key);
2512                 if (ret) {
2513                         PMD_INIT_LOG(ERR, "Failed to get RSS key via AQ");
2514                         return ret;
2515                 }
2516         } else {
2517                 uint32_t *key_dw = (uint32_t *)key;
2518                 uint16_t i;
2519
2520                 for (i = 0; i <= I40E_VFQF_HKEY_MAX_INDEX; i++)
2521                         key_dw[i] = i40e_read_rx_ctl(hw, I40E_VFQF_HKEY(i));
2522         }
2523         *key_len = (I40E_VFQF_HKEY_MAX_INDEX + 1) * sizeof(uint32_t);
2524
2525         return 0;
2526 }
2527
2528 static int
2529 i40evf_hw_rss_hash_set(struct i40e_vf *vf, struct rte_eth_rss_conf *rss_conf)
2530 {
2531         struct i40e_hw *hw = I40E_VF_TO_HW(vf);
2532         uint64_t hena;
2533         int ret;
2534
2535         ret = i40evf_set_rss_key(&vf->vsi, rss_conf->rss_key,
2536                                  rss_conf->rss_key_len);
2537         if (ret)
2538                 return ret;
2539
2540         hena = i40e_config_hena(vf->adapter, rss_conf->rss_hf);
2541         i40e_write_rx_ctl(hw, I40E_VFQF_HENA(0), (uint32_t)hena);
2542         i40e_write_rx_ctl(hw, I40E_VFQF_HENA(1), (uint32_t)(hena >> 32));
2543         I40EVF_WRITE_FLUSH(hw);
2544
2545         return 0;
2546 }
2547
2548 static void
2549 i40evf_disable_rss(struct i40e_vf *vf)
2550 {
2551         struct i40e_hw *hw = I40E_VF_TO_HW(vf);
2552
2553         i40e_write_rx_ctl(hw, I40E_VFQF_HENA(0), 0);
2554         i40e_write_rx_ctl(hw, I40E_VFQF_HENA(1), 0);
2555         I40EVF_WRITE_FLUSH(hw);
2556 }
2557
2558 static int
2559 i40evf_config_rss(struct i40e_vf *vf)
2560 {
2561         struct i40e_hw *hw = I40E_VF_TO_HW(vf);
2562         struct rte_eth_rss_conf rss_conf;
2563         uint32_t i, j, lut = 0, nb_q = (I40E_VFQF_HLUT_MAX_INDEX + 1) * 4;
2564         uint16_t num;
2565
2566         if (vf->dev_data->dev_conf.rxmode.mq_mode != ETH_MQ_RX_RSS) {
2567                 i40evf_disable_rss(vf);
2568                 PMD_DRV_LOG(DEBUG, "RSS not configured");
2569                 return 0;
2570         }
2571
2572         num = RTE_MIN(vf->dev_data->nb_rx_queues, I40E_MAX_QP_NUM_PER_VF);
2573         /* Fill out the look up table */
2574         for (i = 0, j = 0; i < nb_q; i++, j++) {
2575                 if (j >= num)
2576                         j = 0;
2577                 lut = (lut << 8) | j;
2578                 if ((i & 3) == 3)
2579                         I40E_WRITE_REG(hw, I40E_VFQF_HLUT(i >> 2), lut);
2580         }
2581
2582         rss_conf = vf->dev_data->dev_conf.rx_adv_conf.rss_conf;
2583         if ((rss_conf.rss_hf & vf->adapter->flow_types_mask) == 0) {
2584                 i40evf_disable_rss(vf);
2585                 PMD_DRV_LOG(DEBUG, "No hash flag is set");
2586                 return 0;
2587         }
2588
2589         if (rss_conf.rss_key == NULL || rss_conf.rss_key_len <
2590                 (I40E_VFQF_HKEY_MAX_INDEX + 1) * sizeof(uint32_t)) {
2591                 /* Calculate the default hash key */
2592                 for (i = 0; i <= I40E_VFQF_HKEY_MAX_INDEX; i++)
2593                         rss_key_default[i] = (uint32_t)rte_rand();
2594                 rss_conf.rss_key = (uint8_t *)rss_key_default;
2595                 rss_conf.rss_key_len = (I40E_VFQF_HKEY_MAX_INDEX + 1) *
2596                         sizeof(uint32_t);
2597         }
2598
2599         return i40evf_hw_rss_hash_set(vf, &rss_conf);
2600 }
2601
2602 static int
2603 i40evf_dev_rss_hash_update(struct rte_eth_dev *dev,
2604                            struct rte_eth_rss_conf *rss_conf)
2605 {
2606         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2607         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2608         uint64_t rss_hf = rss_conf->rss_hf & vf->adapter->flow_types_mask;
2609         uint64_t hena;
2610
2611         hena = (uint64_t)i40e_read_rx_ctl(hw, I40E_VFQF_HENA(0));
2612         hena |= ((uint64_t)i40e_read_rx_ctl(hw, I40E_VFQF_HENA(1))) << 32;
2613
2614         if (!(hena & vf->adapter->pctypes_mask)) { /* RSS disabled */
2615                 if (rss_hf != 0) /* Enable RSS */
2616                         return -EINVAL;
2617                 return 0;
2618         }
2619
2620         /* RSS enabled */
2621         if (rss_hf == 0) /* Disable RSS */
2622                 return -EINVAL;
2623
2624         return i40evf_hw_rss_hash_set(vf, rss_conf);
2625 }
2626
2627 static int
2628 i40evf_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
2629                              struct rte_eth_rss_conf *rss_conf)
2630 {
2631         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2632         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2633         uint64_t hena;
2634
2635         i40evf_get_rss_key(&vf->vsi, rss_conf->rss_key,
2636                            &rss_conf->rss_key_len);
2637
2638         hena = (uint64_t)i40e_read_rx_ctl(hw, I40E_VFQF_HENA(0));
2639         hena |= ((uint64_t)i40e_read_rx_ctl(hw, I40E_VFQF_HENA(1))) << 32;
2640         rss_conf->rss_hf = i40e_parse_hena(vf->adapter, hena);
2641
2642         return 0;
2643 }
2644
2645 static int
2646 i40evf_dev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
2647 {
2648         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2649         struct rte_eth_dev_data *dev_data = vf->dev_data;
2650         uint32_t frame_size = mtu + I40E_ETH_OVERHEAD;
2651         int ret = 0;
2652
2653         /* check if mtu is within the allowed range */
2654         if ((mtu < ETHER_MIN_MTU) || (frame_size > I40E_FRAME_SIZE_MAX))
2655                 return -EINVAL;
2656
2657         /* mtu setting is forbidden if port is start */
2658         if (dev_data->dev_started) {
2659                 PMD_DRV_LOG(ERR, "port %d must be stopped before configuration",
2660                             dev_data->port_id);
2661                 return -EBUSY;
2662         }
2663
2664         if (frame_size > ETHER_MAX_LEN)
2665                 dev_data->dev_conf.rxmode.jumbo_frame = 1;
2666         else
2667                 dev_data->dev_conf.rxmode.jumbo_frame = 0;
2668
2669         dev_data->dev_conf.rxmode.max_rx_pkt_len = frame_size;
2670
2671         return ret;
2672 }
2673
2674 static void
2675 i40evf_set_default_mac_addr(struct rte_eth_dev *dev,
2676                             struct ether_addr *mac_addr)
2677 {
2678         struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2679
2680         if (!is_valid_assigned_ether_addr(mac_addr)) {
2681                 PMD_DRV_LOG(ERR, "Tried to set invalid MAC address.");
2682                 return;
2683         }
2684
2685         if (is_same_ether_addr(mac_addr, dev->data->mac_addrs))
2686                 return;
2687
2688         if (vf->flags & I40E_FLAG_VF_MAC_BY_PF)
2689                 return;
2690
2691         i40evf_del_mac_addr_by_addr(dev, dev->data->mac_addrs);
2692
2693         i40evf_add_mac_addr(dev, mac_addr, 0, 0);
2694 }