ethdev: add return value to stats get dev op
[dpdk.git] / drivers / net / virtio / virtio_ethdev.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 <stdint.h>
35 #include <string.h>
36 #include <stdio.h>
37 #include <errno.h>
38 #include <unistd.h>
39
40 #include <rte_ethdev.h>
41 #include <rte_ethdev_pci.h>
42 #include <rte_memcpy.h>
43 #include <rte_string_fns.h>
44 #include <rte_memzone.h>
45 #include <rte_malloc.h>
46 #include <rte_atomic.h>
47 #include <rte_branch_prediction.h>
48 #include <rte_pci.h>
49 #include <rte_ether.h>
50 #include <rte_common.h>
51 #include <rte_errno.h>
52 #include <rte_cpuflags.h>
53
54 #include <rte_memory.h>
55 #include <rte_eal.h>
56 #include <rte_dev.h>
57
58 #include "virtio_ethdev.h"
59 #include "virtio_pci.h"
60 #include "virtio_logs.h"
61 #include "virtqueue.h"
62 #include "virtio_rxtx.h"
63
64 static int eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev);
65 static int  virtio_dev_configure(struct rte_eth_dev *dev);
66 static int  virtio_dev_start(struct rte_eth_dev *dev);
67 static void virtio_dev_stop(struct rte_eth_dev *dev);
68 static void virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
69 static void virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
70 static void virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
71 static void virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
72 static void virtio_dev_info_get(struct rte_eth_dev *dev,
73                                 struct rte_eth_dev_info *dev_info);
74 static int virtio_dev_link_update(struct rte_eth_dev *dev,
75         int wait_to_complete);
76
77 static void virtio_set_hwaddr(struct virtio_hw *hw);
78 static void virtio_get_hwaddr(struct virtio_hw *hw);
79
80 static int virtio_dev_stats_get(struct rte_eth_dev *dev,
81                                  struct rte_eth_stats *stats);
82 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
83                                  struct rte_eth_xstat *xstats, unsigned n);
84 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
85                                        struct rte_eth_xstat_name *xstats_names,
86                                        unsigned limit);
87 static void virtio_dev_stats_reset(struct rte_eth_dev *dev);
88 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
89 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
90                                 uint16_t vlan_id, int on);
91 static int virtio_mac_addr_add(struct rte_eth_dev *dev,
92                                 struct ether_addr *mac_addr,
93                                 uint32_t index, uint32_t vmdq);
94 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
95 static void virtio_mac_addr_set(struct rte_eth_dev *dev,
96                                 struct ether_addr *mac_addr);
97
98 static int virtio_dev_queue_stats_mapping_set(
99         struct rte_eth_dev *eth_dev,
100         uint16_t queue_id,
101         uint8_t stat_idx,
102         uint8_t is_rx);
103
104 /*
105  * The set of PCI devices this driver supports
106  */
107 static const struct rte_pci_id pci_id_virtio_map[] = {
108         { RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_LEGACY_DEVICEID_NET) },
109         { RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_MODERN_DEVICEID_NET) },
110         { .vendor_id = 0, /* sentinel */ },
111 };
112
113 struct rte_virtio_xstats_name_off {
114         char name[RTE_ETH_XSTATS_NAME_SIZE];
115         unsigned offset;
116 };
117
118 /* [rt]x_qX_ is prepended to the name string here */
119 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
120         {"good_packets",           offsetof(struct virtnet_rx, stats.packets)},
121         {"good_bytes",             offsetof(struct virtnet_rx, stats.bytes)},
122         {"errors",                 offsetof(struct virtnet_rx, stats.errors)},
123         {"multicast_packets",      offsetof(struct virtnet_rx, stats.multicast)},
124         {"broadcast_packets",      offsetof(struct virtnet_rx, stats.broadcast)},
125         {"undersize_packets",      offsetof(struct virtnet_rx, stats.size_bins[0])},
126         {"size_64_packets",        offsetof(struct virtnet_rx, stats.size_bins[1])},
127         {"size_65_127_packets",    offsetof(struct virtnet_rx, stats.size_bins[2])},
128         {"size_128_255_packets",   offsetof(struct virtnet_rx, stats.size_bins[3])},
129         {"size_256_511_packets",   offsetof(struct virtnet_rx, stats.size_bins[4])},
130         {"size_512_1023_packets",  offsetof(struct virtnet_rx, stats.size_bins[5])},
131         {"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
132         {"size_1519_max_packets",  offsetof(struct virtnet_rx, stats.size_bins[7])},
133 };
134
135 /* [rt]x_qX_ is prepended to the name string here */
136 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
137         {"good_packets",           offsetof(struct virtnet_tx, stats.packets)},
138         {"good_bytes",             offsetof(struct virtnet_tx, stats.bytes)},
139         {"errors",                 offsetof(struct virtnet_tx, stats.errors)},
140         {"multicast_packets",      offsetof(struct virtnet_tx, stats.multicast)},
141         {"broadcast_packets",      offsetof(struct virtnet_tx, stats.broadcast)},
142         {"undersize_packets",      offsetof(struct virtnet_tx, stats.size_bins[0])},
143         {"size_64_packets",        offsetof(struct virtnet_tx, stats.size_bins[1])},
144         {"size_65_127_packets",    offsetof(struct virtnet_tx, stats.size_bins[2])},
145         {"size_128_255_packets",   offsetof(struct virtnet_tx, stats.size_bins[3])},
146         {"size_256_511_packets",   offsetof(struct virtnet_tx, stats.size_bins[4])},
147         {"size_512_1023_packets",  offsetof(struct virtnet_tx, stats.size_bins[5])},
148         {"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
149         {"size_1519_max_packets",  offsetof(struct virtnet_tx, stats.size_bins[7])},
150 };
151
152 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
153                             sizeof(rte_virtio_rxq_stat_strings[0]))
154 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
155                             sizeof(rte_virtio_txq_stat_strings[0]))
156
157 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
158
159 static int
160 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
161                 int *dlen, int pkt_num)
162 {
163         uint32_t head, i;
164         int k, sum = 0;
165         virtio_net_ctrl_ack status = ~0;
166         struct virtio_pmd_ctrl *result;
167         struct virtqueue *vq;
168
169         ctrl->status = status;
170
171         if (!cvq || !cvq->vq) {
172                 PMD_INIT_LOG(ERR, "Control queue is not supported.");
173                 return -1;
174         }
175         vq = cvq->vq;
176         head = vq->vq_desc_head_idx;
177
178         PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
179                 "vq->hw->cvq = %p vq = %p",
180                 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
181
182         if ((vq->vq_free_cnt < ((uint32_t)pkt_num + 2)) || (pkt_num < 1))
183                 return -1;
184
185         memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
186                 sizeof(struct virtio_pmd_ctrl));
187
188         /*
189          * Format is enforced in qemu code:
190          * One TX packet for header;
191          * At least one TX packet per argument;
192          * One RX packet for ACK.
193          */
194         vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
195         vq->vq_ring.desc[head].addr = cvq->virtio_net_hdr_mem;
196         vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
197         vq->vq_free_cnt--;
198         i = vq->vq_ring.desc[head].next;
199
200         for (k = 0; k < pkt_num; k++) {
201                 vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
202                 vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
203                         + sizeof(struct virtio_net_ctrl_hdr)
204                         + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
205                 vq->vq_ring.desc[i].len = dlen[k];
206                 sum += dlen[k];
207                 vq->vq_free_cnt--;
208                 i = vq->vq_ring.desc[i].next;
209         }
210
211         vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
212         vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
213                         + sizeof(struct virtio_net_ctrl_hdr);
214         vq->vq_ring.desc[i].len = sizeof(ctrl->status);
215         vq->vq_free_cnt--;
216
217         vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
218
219         vq_update_avail_ring(vq, head);
220         vq_update_avail_idx(vq);
221
222         PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
223
224         virtqueue_notify(vq);
225
226         rte_rmb();
227         while (VIRTQUEUE_NUSED(vq) == 0) {
228                 rte_rmb();
229                 usleep(100);
230         }
231
232         while (VIRTQUEUE_NUSED(vq)) {
233                 uint32_t idx, desc_idx, used_idx;
234                 struct vring_used_elem *uep;
235
236                 used_idx = (uint32_t)(vq->vq_used_cons_idx
237                                 & (vq->vq_nentries - 1));
238                 uep = &vq->vq_ring.used->ring[used_idx];
239                 idx = (uint32_t) uep->id;
240                 desc_idx = idx;
241
242                 while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
243                         desc_idx = vq->vq_ring.desc[desc_idx].next;
244                         vq->vq_free_cnt++;
245                 }
246
247                 vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
248                 vq->vq_desc_head_idx = idx;
249
250                 vq->vq_used_cons_idx++;
251                 vq->vq_free_cnt++;
252         }
253
254         PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
255                         vq->vq_free_cnt, vq->vq_desc_head_idx);
256
257         result = cvq->virtio_net_hdr_mz->addr;
258
259         return result->status;
260 }
261
262 static int
263 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
264 {
265         struct virtio_hw *hw = dev->data->dev_private;
266         struct virtio_pmd_ctrl ctrl;
267         int dlen[1];
268         int ret;
269
270         ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
271         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
272         memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
273
274         dlen[0] = sizeof(uint16_t);
275
276         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
277         if (ret) {
278                 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
279                           "failed, this is too late now...");
280                 return -EINVAL;
281         }
282
283         return 0;
284 }
285
286 static void
287 virtio_dev_queue_release(void *queue __rte_unused)
288 {
289         /* do nothing */
290 }
291
292 static int
293 virtio_get_queue_type(struct virtio_hw *hw, uint16_t vtpci_queue_idx)
294 {
295         if (vtpci_queue_idx == hw->max_queue_pairs * 2)
296                 return VTNET_CQ;
297         else if (vtpci_queue_idx % 2 == 0)
298                 return VTNET_RQ;
299         else
300                 return VTNET_TQ;
301 }
302
303 static uint16_t
304 virtio_get_nr_vq(struct virtio_hw *hw)
305 {
306         uint16_t nr_vq = hw->max_queue_pairs * 2;
307
308         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
309                 nr_vq += 1;
310
311         return nr_vq;
312 }
313
314 static void
315 virtio_init_vring(struct virtqueue *vq)
316 {
317         int size = vq->vq_nentries;
318         struct vring *vr = &vq->vq_ring;
319         uint8_t *ring_mem = vq->vq_ring_virt_mem;
320
321         PMD_INIT_FUNC_TRACE();
322
323         /*
324          * Reinitialise since virtio port might have been stopped and restarted
325          */
326         memset(ring_mem, 0, vq->vq_ring_size);
327         vring_init(vr, size, ring_mem, VIRTIO_PCI_VRING_ALIGN);
328         vq->vq_used_cons_idx = 0;
329         vq->vq_desc_head_idx = 0;
330         vq->vq_avail_idx = 0;
331         vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
332         vq->vq_free_cnt = vq->vq_nentries;
333         memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
334
335         vring_desc_init(vr->desc, size);
336
337         /*
338          * Disable device(host) interrupting guest
339          */
340         virtqueue_disable_intr(vq);
341 }
342
343 static int
344 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
345 {
346         char vq_name[VIRTQUEUE_MAX_NAME_SZ];
347         char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
348         const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
349         unsigned int vq_size, size;
350         struct virtio_hw *hw = dev->data->dev_private;
351         struct virtnet_rx *rxvq = NULL;
352         struct virtnet_tx *txvq = NULL;
353         struct virtnet_ctl *cvq = NULL;
354         struct virtqueue *vq;
355         size_t sz_hdr_mz = 0;
356         void *sw_ring = NULL;
357         int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
358         int ret;
359
360         PMD_INIT_LOG(DEBUG, "setting up queue: %u", vtpci_queue_idx);
361
362         /*
363          * Read the virtqueue size from the Queue Size field
364          * Always power of 2 and if 0 virtqueue does not exist
365          */
366         vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
367         PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
368         if (vq_size == 0) {
369                 PMD_INIT_LOG(ERR, "virtqueue does not exist");
370                 return -EINVAL;
371         }
372
373         if (!rte_is_power_of_2(vq_size)) {
374                 PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
375                 return -EINVAL;
376         }
377
378         snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
379                  dev->data->port_id, vtpci_queue_idx);
380
381         size = RTE_ALIGN_CEIL(sizeof(*vq) +
382                                 vq_size * sizeof(struct vq_desc_extra),
383                                 RTE_CACHE_LINE_SIZE);
384         if (queue_type == VTNET_TQ) {
385                 /*
386                  * For each xmit packet, allocate a virtio_net_hdr
387                  * and indirect ring elements
388                  */
389                 sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
390         } else if (queue_type == VTNET_CQ) {
391                 /* Allocate a page for control vq command, data and status */
392                 sz_hdr_mz = PAGE_SIZE;
393         }
394
395         vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
396                                 SOCKET_ID_ANY);
397         if (vq == NULL) {
398                 PMD_INIT_LOG(ERR, "can not allocate vq");
399                 return -ENOMEM;
400         }
401         hw->vqs[vtpci_queue_idx] = vq;
402
403         vq->hw = hw;
404         vq->vq_queue_index = vtpci_queue_idx;
405         vq->vq_nentries = vq_size;
406
407         /*
408          * Reserve a memzone for vring elements
409          */
410         size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
411         vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
412         PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
413                      size, vq->vq_ring_size);
414
415         mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
416                                          SOCKET_ID_ANY,
417                                          0, VIRTIO_PCI_VRING_ALIGN);
418         if (mz == NULL) {
419                 if (rte_errno == EEXIST)
420                         mz = rte_memzone_lookup(vq_name);
421                 if (mz == NULL) {
422                         ret = -ENOMEM;
423                         goto fail_q_alloc;
424                 }
425         }
426
427         memset(mz->addr, 0, mz->len);
428
429         vq->vq_ring_mem = mz->phys_addr;
430         vq->vq_ring_virt_mem = mz->addr;
431         PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%" PRIx64,
432                      (uint64_t)mz->phys_addr);
433         PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
434                      (uint64_t)(uintptr_t)mz->addr);
435
436         virtio_init_vring(vq);
437
438         if (sz_hdr_mz) {
439                 snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
440                          dev->data->port_id, vtpci_queue_idx);
441                 hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
442                                                      SOCKET_ID_ANY, 0,
443                                                      RTE_CACHE_LINE_SIZE);
444                 if (hdr_mz == NULL) {
445                         if (rte_errno == EEXIST)
446                                 hdr_mz = rte_memzone_lookup(vq_hdr_name);
447                         if (hdr_mz == NULL) {
448                                 ret = -ENOMEM;
449                                 goto fail_q_alloc;
450                         }
451                 }
452         }
453
454         if (queue_type == VTNET_RQ) {
455                 size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
456                                sizeof(vq->sw_ring[0]);
457
458                 sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
459                                 RTE_CACHE_LINE_SIZE, SOCKET_ID_ANY);
460                 if (!sw_ring) {
461                         PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
462                         ret = -ENOMEM;
463                         goto fail_q_alloc;
464                 }
465
466                 vq->sw_ring = sw_ring;
467                 rxvq = &vq->rxq;
468                 rxvq->vq = vq;
469                 rxvq->port_id = dev->data->port_id;
470                 rxvq->mz = mz;
471         } else if (queue_type == VTNET_TQ) {
472                 txvq = &vq->txq;
473                 txvq->vq = vq;
474                 txvq->port_id = dev->data->port_id;
475                 txvq->mz = mz;
476                 txvq->virtio_net_hdr_mz = hdr_mz;
477                 txvq->virtio_net_hdr_mem = hdr_mz->phys_addr;
478         } else if (queue_type == VTNET_CQ) {
479                 cvq = &vq->cq;
480                 cvq->vq = vq;
481                 cvq->mz = mz;
482                 cvq->virtio_net_hdr_mz = hdr_mz;
483                 cvq->virtio_net_hdr_mem = hdr_mz->phys_addr;
484                 memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
485
486                 hw->cvq = cvq;
487         }
488
489         /* For virtio_user case (that is when hw->dev is NULL), we use
490          * virtual address. And we need properly set _offset_, please see
491          * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
492          */
493         if (!hw->virtio_user_dev)
494                 vq->offset = offsetof(struct rte_mbuf, buf_physaddr);
495         else {
496                 vq->vq_ring_mem = (uintptr_t)mz->addr;
497                 vq->offset = offsetof(struct rte_mbuf, buf_addr);
498                 if (queue_type == VTNET_TQ)
499                         txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
500                 else if (queue_type == VTNET_CQ)
501                         cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
502         }
503
504         if (queue_type == VTNET_TQ) {
505                 struct virtio_tx_region *txr;
506                 unsigned int i;
507
508                 txr = hdr_mz->addr;
509                 memset(txr, 0, vq_size * sizeof(*txr));
510                 for (i = 0; i < vq_size; i++) {
511                         struct vring_desc *start_dp = txr[i].tx_indir;
512
513                         vring_desc_init(start_dp, RTE_DIM(txr[i].tx_indir));
514
515                         /* first indirect descriptor is always the tx header */
516                         start_dp->addr = txvq->virtio_net_hdr_mem
517                                 + i * sizeof(*txr)
518                                 + offsetof(struct virtio_tx_region, tx_hdr);
519
520                         start_dp->len = hw->vtnet_hdr_size;
521                         start_dp->flags = VRING_DESC_F_NEXT;
522                 }
523         }
524
525         if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
526                 PMD_INIT_LOG(ERR, "setup_queue failed");
527                 return -EINVAL;
528         }
529
530         return 0;
531
532 fail_q_alloc:
533         rte_free(sw_ring);
534         rte_memzone_free(hdr_mz);
535         rte_memzone_free(mz);
536         rte_free(vq);
537
538         return ret;
539 }
540
541 static void
542 virtio_free_queues(struct virtio_hw *hw)
543 {
544         uint16_t nr_vq = virtio_get_nr_vq(hw);
545         struct virtqueue *vq;
546         int queue_type;
547         uint16_t i;
548
549         if (hw->vqs == NULL)
550                 return;
551
552         for (i = 0; i < nr_vq; i++) {
553                 vq = hw->vqs[i];
554                 if (!vq)
555                         continue;
556
557                 queue_type = virtio_get_queue_type(hw, i);
558                 if (queue_type == VTNET_RQ) {
559                         rte_free(vq->sw_ring);
560                         rte_memzone_free(vq->rxq.mz);
561                 } else if (queue_type == VTNET_TQ) {
562                         rte_memzone_free(vq->txq.mz);
563                         rte_memzone_free(vq->txq.virtio_net_hdr_mz);
564                 } else {
565                         rte_memzone_free(vq->cq.mz);
566                         rte_memzone_free(vq->cq.virtio_net_hdr_mz);
567                 }
568
569                 rte_free(vq);
570                 hw->vqs[i] = NULL;
571         }
572
573         rte_free(hw->vqs);
574         hw->vqs = NULL;
575 }
576
577 static int
578 virtio_alloc_queues(struct rte_eth_dev *dev)
579 {
580         struct virtio_hw *hw = dev->data->dev_private;
581         uint16_t nr_vq = virtio_get_nr_vq(hw);
582         uint16_t i;
583         int ret;
584
585         hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
586         if (!hw->vqs) {
587                 PMD_INIT_LOG(ERR, "failed to allocate vqs");
588                 return -ENOMEM;
589         }
590
591         for (i = 0; i < nr_vq; i++) {
592                 ret = virtio_init_queue(dev, i);
593                 if (ret < 0) {
594                         virtio_free_queues(hw);
595                         return ret;
596                 }
597         }
598
599         return 0;
600 }
601
602 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
603
604 static void
605 virtio_dev_close(struct rte_eth_dev *dev)
606 {
607         struct virtio_hw *hw = dev->data->dev_private;
608         struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
609
610         PMD_INIT_LOG(DEBUG, "virtio_dev_close");
611
612         /* reset the NIC */
613         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
614                 VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
615         if (intr_conf->rxq)
616                 virtio_queues_unbind_intr(dev);
617
618         if (intr_conf->lsc || intr_conf->rxq) {
619                 rte_intr_disable(dev->intr_handle);
620                 rte_intr_efd_disable(dev->intr_handle);
621                 rte_free(dev->intr_handle->intr_vec);
622                 dev->intr_handle->intr_vec = NULL;
623         }
624
625         vtpci_reset(hw);
626         virtio_dev_free_mbufs(dev);
627         virtio_free_queues(hw);
628 }
629
630 static void
631 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
632 {
633         struct virtio_hw *hw = dev->data->dev_private;
634         struct virtio_pmd_ctrl ctrl;
635         int dlen[1];
636         int ret;
637
638         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
639                 PMD_INIT_LOG(INFO, "host does not support rx control");
640                 return;
641         }
642
643         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
644         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
645         ctrl.data[0] = 1;
646         dlen[0] = 1;
647
648         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
649         if (ret)
650                 PMD_INIT_LOG(ERR, "Failed to enable promisc");
651 }
652
653 static void
654 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
655 {
656         struct virtio_hw *hw = dev->data->dev_private;
657         struct virtio_pmd_ctrl ctrl;
658         int dlen[1];
659         int ret;
660
661         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
662                 PMD_INIT_LOG(INFO, "host does not support rx control");
663                 return;
664         }
665
666         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
667         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
668         ctrl.data[0] = 0;
669         dlen[0] = 1;
670
671         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
672         if (ret)
673                 PMD_INIT_LOG(ERR, "Failed to disable promisc");
674 }
675
676 static void
677 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
678 {
679         struct virtio_hw *hw = dev->data->dev_private;
680         struct virtio_pmd_ctrl ctrl;
681         int dlen[1];
682         int ret;
683
684         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
685                 PMD_INIT_LOG(INFO, "host does not support rx control");
686                 return;
687         }
688
689         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
690         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
691         ctrl.data[0] = 1;
692         dlen[0] = 1;
693
694         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
695         if (ret)
696                 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
697 }
698
699 static void
700 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
701 {
702         struct virtio_hw *hw = dev->data->dev_private;
703         struct virtio_pmd_ctrl ctrl;
704         int dlen[1];
705         int ret;
706
707         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
708                 PMD_INIT_LOG(INFO, "host does not support rx control");
709                 return;
710         }
711
712         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
713         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
714         ctrl.data[0] = 0;
715         dlen[0] = 1;
716
717         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
718         if (ret)
719                 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
720 }
721
722 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
723 static int
724 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
725 {
726         struct virtio_hw *hw = dev->data->dev_private;
727         uint32_t ether_hdr_len = ETHER_HDR_LEN + VLAN_TAG_LEN +
728                                  hw->vtnet_hdr_size;
729         uint32_t frame_size = mtu + ether_hdr_len;
730         uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
731
732         max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
733
734         if (mtu < ETHER_MIN_MTU || frame_size > max_frame_size) {
735                 PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
736                         ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
737                 return -EINVAL;
738         }
739         return 0;
740 }
741
742 static int
743 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
744 {
745         struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
746         struct virtqueue *vq = rxvq->vq;
747
748         virtqueue_enable_intr(vq);
749         return 0;
750 }
751
752 static int
753 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
754 {
755         struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
756         struct virtqueue *vq = rxvq->vq;
757
758         virtqueue_disable_intr(vq);
759         return 0;
760 }
761
762 /*
763  * dev_ops for virtio, bare necessities for basic operation
764  */
765 static const struct eth_dev_ops virtio_eth_dev_ops = {
766         .dev_configure           = virtio_dev_configure,
767         .dev_start               = virtio_dev_start,
768         .dev_stop                = virtio_dev_stop,
769         .dev_close               = virtio_dev_close,
770         .promiscuous_enable      = virtio_dev_promiscuous_enable,
771         .promiscuous_disable     = virtio_dev_promiscuous_disable,
772         .allmulticast_enable     = virtio_dev_allmulticast_enable,
773         .allmulticast_disable    = virtio_dev_allmulticast_disable,
774         .mtu_set                 = virtio_mtu_set,
775         .dev_infos_get           = virtio_dev_info_get,
776         .stats_get               = virtio_dev_stats_get,
777         .xstats_get              = virtio_dev_xstats_get,
778         .xstats_get_names        = virtio_dev_xstats_get_names,
779         .stats_reset             = virtio_dev_stats_reset,
780         .xstats_reset            = virtio_dev_stats_reset,
781         .link_update             = virtio_dev_link_update,
782         .rx_queue_setup          = virtio_dev_rx_queue_setup,
783         .rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
784         .rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
785         .rx_queue_release        = virtio_dev_queue_release,
786         .rx_descriptor_done      = virtio_dev_rx_queue_done,
787         .tx_queue_setup          = virtio_dev_tx_queue_setup,
788         .tx_queue_release        = virtio_dev_queue_release,
789         /* collect stats per queue */
790         .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
791         .vlan_filter_set         = virtio_vlan_filter_set,
792         .mac_addr_add            = virtio_mac_addr_add,
793         .mac_addr_remove         = virtio_mac_addr_remove,
794         .mac_addr_set            = virtio_mac_addr_set,
795 };
796
797 static inline int
798 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
799                                 struct rte_eth_link *link)
800 {
801         struct rte_eth_link *dst = link;
802         struct rte_eth_link *src = &(dev->data->dev_link);
803
804         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
805                         *(uint64_t *)src) == 0)
806                 return -1;
807
808         return 0;
809 }
810
811 /**
812  * Atomically writes the link status information into global
813  * structure rte_eth_dev.
814  *
815  * @param dev
816  *   - Pointer to the structure rte_eth_dev to read from.
817  *   - Pointer to the buffer to be saved with the link status.
818  *
819  * @return
820  *   - On success, zero.
821  *   - On failure, negative value.
822  */
823 static inline int
824 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
825                 struct rte_eth_link *link)
826 {
827         struct rte_eth_link *dst = &(dev->data->dev_link);
828         struct rte_eth_link *src = link;
829
830         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
831                                         *(uint64_t *)src) == 0)
832                 return -1;
833
834         return 0;
835 }
836
837 static void
838 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
839 {
840         unsigned i;
841
842         for (i = 0; i < dev->data->nb_tx_queues; i++) {
843                 const struct virtnet_tx *txvq = dev->data->tx_queues[i];
844                 if (txvq == NULL)
845                         continue;
846
847                 stats->opackets += txvq->stats.packets;
848                 stats->obytes += txvq->stats.bytes;
849                 stats->oerrors += txvq->stats.errors;
850
851                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
852                         stats->q_opackets[i] = txvq->stats.packets;
853                         stats->q_obytes[i] = txvq->stats.bytes;
854                 }
855         }
856
857         for (i = 0; i < dev->data->nb_rx_queues; i++) {
858                 const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
859                 if (rxvq == NULL)
860                         continue;
861
862                 stats->ipackets += rxvq->stats.packets;
863                 stats->ibytes += rxvq->stats.bytes;
864                 stats->ierrors += rxvq->stats.errors;
865
866                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
867                         stats->q_ipackets[i] = rxvq->stats.packets;
868                         stats->q_ibytes[i] = rxvq->stats.bytes;
869                 }
870         }
871
872         stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
873 }
874
875 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
876                                        struct rte_eth_xstat_name *xstats_names,
877                                        __rte_unused unsigned limit)
878 {
879         unsigned i;
880         unsigned count = 0;
881         unsigned t;
882
883         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
884                 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
885
886         if (xstats_names != NULL) {
887                 /* Note: limit checked in rte_eth_xstats_names() */
888
889                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
890                         struct virtqueue *rxvq = dev->data->rx_queues[i];
891                         if (rxvq == NULL)
892                                 continue;
893                         for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
894                                 snprintf(xstats_names[count].name,
895                                         sizeof(xstats_names[count].name),
896                                         "rx_q%u_%s", i,
897                                         rte_virtio_rxq_stat_strings[t].name);
898                                 count++;
899                         }
900                 }
901
902                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
903                         struct virtqueue *txvq = dev->data->tx_queues[i];
904                         if (txvq == NULL)
905                                 continue;
906                         for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
907                                 snprintf(xstats_names[count].name,
908                                         sizeof(xstats_names[count].name),
909                                         "tx_q%u_%s", i,
910                                         rte_virtio_txq_stat_strings[t].name);
911                                 count++;
912                         }
913                 }
914                 return count;
915         }
916         return nstats;
917 }
918
919 static int
920 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
921                       unsigned n)
922 {
923         unsigned i;
924         unsigned count = 0;
925
926         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
927                 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
928
929         if (n < nstats)
930                 return nstats;
931
932         for (i = 0; i < dev->data->nb_rx_queues; i++) {
933                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
934
935                 if (rxvq == NULL)
936                         continue;
937
938                 unsigned t;
939
940                 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
941                         xstats[count].value = *(uint64_t *)(((char *)rxvq) +
942                                 rte_virtio_rxq_stat_strings[t].offset);
943                         xstats[count].id = count;
944                         count++;
945                 }
946         }
947
948         for (i = 0; i < dev->data->nb_tx_queues; i++) {
949                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
950
951                 if (txvq == NULL)
952                         continue;
953
954                 unsigned t;
955
956                 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
957                         xstats[count].value = *(uint64_t *)(((char *)txvq) +
958                                 rte_virtio_txq_stat_strings[t].offset);
959                         xstats[count].id = count;
960                         count++;
961                 }
962         }
963
964         return count;
965 }
966
967 static int
968 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
969 {
970         virtio_update_stats(dev, stats);
971
972         return 0;
973 }
974
975 static void
976 virtio_dev_stats_reset(struct rte_eth_dev *dev)
977 {
978         unsigned int i;
979
980         for (i = 0; i < dev->data->nb_tx_queues; i++) {
981                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
982                 if (txvq == NULL)
983                         continue;
984
985                 txvq->stats.packets = 0;
986                 txvq->stats.bytes = 0;
987                 txvq->stats.errors = 0;
988                 txvq->stats.multicast = 0;
989                 txvq->stats.broadcast = 0;
990                 memset(txvq->stats.size_bins, 0,
991                        sizeof(txvq->stats.size_bins[0]) * 8);
992         }
993
994         for (i = 0; i < dev->data->nb_rx_queues; i++) {
995                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
996                 if (rxvq == NULL)
997                         continue;
998
999                 rxvq->stats.packets = 0;
1000                 rxvq->stats.bytes = 0;
1001                 rxvq->stats.errors = 0;
1002                 rxvq->stats.multicast = 0;
1003                 rxvq->stats.broadcast = 0;
1004                 memset(rxvq->stats.size_bins, 0,
1005                        sizeof(rxvq->stats.size_bins[0]) * 8);
1006         }
1007 }
1008
1009 static void
1010 virtio_set_hwaddr(struct virtio_hw *hw)
1011 {
1012         vtpci_write_dev_config(hw,
1013                         offsetof(struct virtio_net_config, mac),
1014                         &hw->mac_addr, ETHER_ADDR_LEN);
1015 }
1016
1017 static void
1018 virtio_get_hwaddr(struct virtio_hw *hw)
1019 {
1020         if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1021                 vtpci_read_dev_config(hw,
1022                         offsetof(struct virtio_net_config, mac),
1023                         &hw->mac_addr, ETHER_ADDR_LEN);
1024         } else {
1025                 eth_random_addr(&hw->mac_addr[0]);
1026                 virtio_set_hwaddr(hw);
1027         }
1028 }
1029
1030 static int
1031 virtio_mac_table_set(struct virtio_hw *hw,
1032                      const struct virtio_net_ctrl_mac *uc,
1033                      const struct virtio_net_ctrl_mac *mc)
1034 {
1035         struct virtio_pmd_ctrl ctrl;
1036         int err, len[2];
1037
1038         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1039                 PMD_DRV_LOG(INFO, "host does not support mac table");
1040                 return -1;
1041         }
1042
1043         ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1044         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1045
1046         len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
1047         memcpy(ctrl.data, uc, len[0]);
1048
1049         len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
1050         memcpy(ctrl.data + len[0], mc, len[1]);
1051
1052         err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1053         if (err != 0)
1054                 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1055         return err;
1056 }
1057
1058 static int
1059 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
1060                     uint32_t index, uint32_t vmdq __rte_unused)
1061 {
1062         struct virtio_hw *hw = dev->data->dev_private;
1063         const struct ether_addr *addrs = dev->data->mac_addrs;
1064         unsigned int i;
1065         struct virtio_net_ctrl_mac *uc, *mc;
1066
1067         if (index >= VIRTIO_MAX_MAC_ADDRS) {
1068                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1069                 return -EINVAL;
1070         }
1071
1072         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1073         uc->entries = 0;
1074         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1075         mc->entries = 0;
1076
1077         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1078                 const struct ether_addr *addr
1079                         = (i == index) ? mac_addr : addrs + i;
1080                 struct virtio_net_ctrl_mac *tbl
1081                         = is_multicast_ether_addr(addr) ? mc : uc;
1082
1083                 memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
1084         }
1085
1086         return virtio_mac_table_set(hw, uc, mc);
1087 }
1088
1089 static void
1090 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1091 {
1092         struct virtio_hw *hw = dev->data->dev_private;
1093         struct ether_addr *addrs = dev->data->mac_addrs;
1094         struct virtio_net_ctrl_mac *uc, *mc;
1095         unsigned int i;
1096
1097         if (index >= VIRTIO_MAX_MAC_ADDRS) {
1098                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1099                 return;
1100         }
1101
1102         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1103         uc->entries = 0;
1104         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1105         mc->entries = 0;
1106
1107         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1108                 struct virtio_net_ctrl_mac *tbl;
1109
1110                 if (i == index || is_zero_ether_addr(addrs + i))
1111                         continue;
1112
1113                 tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
1114                 memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
1115         }
1116
1117         virtio_mac_table_set(hw, uc, mc);
1118 }
1119
1120 static void
1121 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1122 {
1123         struct virtio_hw *hw = dev->data->dev_private;
1124
1125         memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
1126
1127         /* Use atomic update if available */
1128         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1129                 struct virtio_pmd_ctrl ctrl;
1130                 int len = ETHER_ADDR_LEN;
1131
1132                 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1133                 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1134
1135                 memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
1136                 virtio_send_command(hw->cvq, &ctrl, &len, 1);
1137         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1138                 virtio_set_hwaddr(hw);
1139 }
1140
1141 static int
1142 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1143 {
1144         struct virtio_hw *hw = dev->data->dev_private;
1145         struct virtio_pmd_ctrl ctrl;
1146         int len;
1147
1148         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1149                 return -ENOTSUP;
1150
1151         ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1152         ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1153         memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1154         len = sizeof(vlan_id);
1155
1156         return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1157 }
1158
1159 static int
1160 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1161 {
1162         uint64_t host_features;
1163
1164         /* Prepare guest_features: feature that driver wants to support */
1165         PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1166                 req_features);
1167
1168         /* Read device(host) feature bits */
1169         host_features = VTPCI_OPS(hw)->get_features(hw);
1170         PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1171                 host_features);
1172
1173         /* If supported, ensure MTU value is valid before acknowledging it. */
1174         if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1175                 struct virtio_net_config config;
1176
1177                 vtpci_read_dev_config(hw,
1178                         offsetof(struct virtio_net_config, mtu),
1179                         &config.mtu, sizeof(config.mtu));
1180
1181                 if (config.mtu < ETHER_MIN_MTU)
1182                         req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1183         }
1184
1185         /*
1186          * Negotiate features: Subset of device feature bits are written back
1187          * guest feature bits.
1188          */
1189         hw->guest_features = req_features;
1190         hw->guest_features = vtpci_negotiate_features(hw, host_features);
1191         PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1192                 hw->guest_features);
1193
1194         if (hw->modern) {
1195                 if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1196                         PMD_INIT_LOG(ERR,
1197                                 "VIRTIO_F_VERSION_1 features is not enabled.");
1198                         return -1;
1199                 }
1200                 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1201                 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1202                         PMD_INIT_LOG(ERR,
1203                                 "failed to set FEATURES_OK status!");
1204                         return -1;
1205                 }
1206         }
1207
1208         hw->req_guest_features = req_features;
1209
1210         return 0;
1211 }
1212
1213 /*
1214  * Process Virtio Config changed interrupt and call the callback
1215  * if link state changed.
1216  */
1217 void
1218 virtio_interrupt_handler(void *param)
1219 {
1220         struct rte_eth_dev *dev = param;
1221         struct virtio_hw *hw = dev->data->dev_private;
1222         uint8_t isr;
1223
1224         /* Read interrupt status which clears interrupt */
1225         isr = vtpci_isr(hw);
1226         PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1227
1228         if (rte_intr_enable(dev->intr_handle) < 0)
1229                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1230
1231         if (isr & VIRTIO_PCI_ISR_CONFIG) {
1232                 if (virtio_dev_link_update(dev, 0) == 0)
1233                         _rte_eth_dev_callback_process(dev,
1234                                                       RTE_ETH_EVENT_INTR_LSC,
1235                                                       NULL, NULL);
1236         }
1237
1238 }
1239
1240 /* set rx and tx handlers according to what is supported */
1241 static void
1242 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1243 {
1244         struct virtio_hw *hw = eth_dev->data->dev_private;
1245
1246         if (hw->use_simple_rx) {
1247                 PMD_INIT_LOG(INFO, "virtio: using simple Rx path on port %u",
1248                         eth_dev->data->port_id);
1249                 eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1250         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1251                 PMD_INIT_LOG(INFO,
1252                         "virtio: using mergeable buffer Rx path on port %u",
1253                         eth_dev->data->port_id);
1254                 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1255         } else {
1256                 PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1257                         eth_dev->data->port_id);
1258                 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1259         }
1260
1261         if (hw->use_simple_tx) {
1262                 PMD_INIT_LOG(INFO, "virtio: using simple Tx path on port %u",
1263                         eth_dev->data->port_id);
1264                 eth_dev->tx_pkt_burst = virtio_xmit_pkts_simple;
1265         } else {
1266                 PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1267                         eth_dev->data->port_id);
1268                 eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1269         }
1270 }
1271
1272 /* Only support 1:1 queue/interrupt mapping so far.
1273  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1274  * interrupt vectors (<N+1).
1275  */
1276 static int
1277 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1278 {
1279         uint32_t i;
1280         struct virtio_hw *hw = dev->data->dev_private;
1281
1282         PMD_INIT_LOG(INFO, "queue/interrupt binding");
1283         for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1284                 dev->intr_handle->intr_vec[i] = i + 1;
1285                 if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1286                                                  VIRTIO_MSI_NO_VECTOR) {
1287                         PMD_DRV_LOG(ERR, "failed to set queue vector");
1288                         return -EBUSY;
1289                 }
1290         }
1291
1292         return 0;
1293 }
1294
1295 static void
1296 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1297 {
1298         uint32_t i;
1299         struct virtio_hw *hw = dev->data->dev_private;
1300
1301         PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1302         for (i = 0; i < dev->data->nb_rx_queues; ++i)
1303                 VTPCI_OPS(hw)->set_queue_irq(hw,
1304                                              hw->vqs[i * VTNET_CQ],
1305                                              VIRTIO_MSI_NO_VECTOR);
1306 }
1307
1308 static int
1309 virtio_configure_intr(struct rte_eth_dev *dev)
1310 {
1311         struct virtio_hw *hw = dev->data->dev_private;
1312
1313         if (!rte_intr_cap_multiple(dev->intr_handle)) {
1314                 PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1315                 return -ENOTSUP;
1316         }
1317
1318         if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1319                 PMD_INIT_LOG(ERR, "Fail to create eventfd");
1320                 return -1;
1321         }
1322
1323         if (!dev->intr_handle->intr_vec) {
1324                 dev->intr_handle->intr_vec =
1325                         rte_zmalloc("intr_vec",
1326                                     hw->max_queue_pairs * sizeof(int), 0);
1327                 if (!dev->intr_handle->intr_vec) {
1328                         PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1329                                      hw->max_queue_pairs);
1330                         return -ENOMEM;
1331                 }
1332         }
1333
1334         /* Re-register callback to update max_intr */
1335         rte_intr_callback_unregister(dev->intr_handle,
1336                                      virtio_interrupt_handler,
1337                                      dev);
1338         rte_intr_callback_register(dev->intr_handle,
1339                                    virtio_interrupt_handler,
1340                                    dev);
1341
1342         /* DO NOT try to remove this! This function will enable msix, or QEMU
1343          * will encounter SIGSEGV when DRIVER_OK is sent.
1344          * And for legacy devices, this should be done before queue/vec binding
1345          * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1346          * (22) will be ignored.
1347          */
1348         if (rte_intr_enable(dev->intr_handle) < 0) {
1349                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1350                 return -1;
1351         }
1352
1353         if (virtio_queues_bind_intr(dev) < 0) {
1354                 PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1355                 return -1;
1356         }
1357
1358         return 0;
1359 }
1360
1361 /* reset device and renegotiate features if needed */
1362 static int
1363 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1364 {
1365         struct virtio_hw *hw = eth_dev->data->dev_private;
1366         struct virtio_net_config *config;
1367         struct virtio_net_config local_config;
1368         struct rte_pci_device *pci_dev = NULL;
1369         int ret;
1370
1371         /* Reset the device although not necessary at startup */
1372         vtpci_reset(hw);
1373
1374         /* Tell the host we've noticed this device. */
1375         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1376
1377         /* Tell the host we've known how to drive the device. */
1378         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1379         if (virtio_negotiate_features(hw, req_features) < 0)
1380                 return -1;
1381
1382         if (!hw->virtio_user_dev) {
1383                 pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1384                 rte_eth_copy_pci_info(eth_dev, pci_dev);
1385         }
1386
1387         eth_dev->data->dev_flags = RTE_ETH_DEV_DETACHABLE;
1388         /* If host does not support both status and MSI-X then disable LSC */
1389         if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) && hw->use_msix)
1390                 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1391         else
1392                 eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1393
1394         /* Setting up rx_header size for the device */
1395         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1396             vtpci_with_feature(hw, VIRTIO_F_VERSION_1))
1397                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1398         else
1399                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1400
1401         /* Copy the permanent MAC address to: virtio_hw */
1402         virtio_get_hwaddr(hw);
1403         ether_addr_copy((struct ether_addr *) hw->mac_addr,
1404                         &eth_dev->data->mac_addrs[0]);
1405         PMD_INIT_LOG(DEBUG,
1406                      "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1407                      hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1408                      hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1409
1410         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1411                 config = &local_config;
1412
1413                 vtpci_read_dev_config(hw,
1414                         offsetof(struct virtio_net_config, mac),
1415                         &config->mac, sizeof(config->mac));
1416
1417                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1418                         vtpci_read_dev_config(hw,
1419                                 offsetof(struct virtio_net_config, status),
1420                                 &config->status, sizeof(config->status));
1421                 } else {
1422                         PMD_INIT_LOG(DEBUG,
1423                                      "VIRTIO_NET_F_STATUS is not supported");
1424                         config->status = 0;
1425                 }
1426
1427                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1428                         vtpci_read_dev_config(hw,
1429                                 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1430                                 &config->max_virtqueue_pairs,
1431                                 sizeof(config->max_virtqueue_pairs));
1432                 } else {
1433                         PMD_INIT_LOG(DEBUG,
1434                                      "VIRTIO_NET_F_MQ is not supported");
1435                         config->max_virtqueue_pairs = 1;
1436                 }
1437
1438                 hw->max_queue_pairs = config->max_virtqueue_pairs;
1439
1440                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1441                         vtpci_read_dev_config(hw,
1442                                 offsetof(struct virtio_net_config, mtu),
1443                                 &config->mtu,
1444                                 sizeof(config->mtu));
1445
1446                         /*
1447                          * MTU value has already been checked at negotiation
1448                          * time, but check again in case it has changed since
1449                          * then, which should not happen.
1450                          */
1451                         if (config->mtu < ETHER_MIN_MTU) {
1452                                 PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1453                                                 config->mtu);
1454                                 return -1;
1455                         }
1456
1457                         hw->max_mtu = config->mtu;
1458                         /* Set initial MTU to maximum one supported by vhost */
1459                         eth_dev->data->mtu = config->mtu;
1460
1461                 } else {
1462                         hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - ETHER_HDR_LEN -
1463                                 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1464                 }
1465
1466                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1467                                 config->max_virtqueue_pairs);
1468                 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1469                 PMD_INIT_LOG(DEBUG,
1470                                 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1471                                 config->mac[0], config->mac[1],
1472                                 config->mac[2], config->mac[3],
1473                                 config->mac[4], config->mac[5]);
1474         } else {
1475                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1476                 hw->max_queue_pairs = 1;
1477         }
1478
1479         ret = virtio_alloc_queues(eth_dev);
1480         if (ret < 0)
1481                 return ret;
1482
1483         if (eth_dev->data->dev_conf.intr_conf.rxq) {
1484                 if (virtio_configure_intr(eth_dev) < 0) {
1485                         PMD_INIT_LOG(ERR, "failed to configure interrupt");
1486                         return -1;
1487                 }
1488         }
1489
1490         vtpci_reinit_complete(hw);
1491
1492         if (pci_dev)
1493                 PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1494                         eth_dev->data->port_id, pci_dev->id.vendor_id,
1495                         pci_dev->id.device_id);
1496
1497         return 0;
1498 }
1499
1500 /*
1501  * Remap the PCI device again (IO port map for legacy device and
1502  * memory map for modern device), so that the secondary process
1503  * could have the PCI initiated correctly.
1504  */
1505 static int
1506 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1507 {
1508         if (hw->modern) {
1509                 /*
1510                  * We don't have to re-parse the PCI config space, since
1511                  * rte_pci_map_device() makes sure the mapped address
1512                  * in secondary process would equal to the one mapped in
1513                  * the primary process: error will be returned if that
1514                  * requirement is not met.
1515                  *
1516                  * That said, we could simply reuse all cap pointers
1517                  * (such as dev_cfg, common_cfg, etc.) parsed from the
1518                  * primary process, which is stored in shared memory.
1519                  */
1520                 if (rte_pci_map_device(pci_dev)) {
1521                         PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1522                         return -1;
1523                 }
1524         } else {
1525                 if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1526                         return -1;
1527         }
1528
1529         return 0;
1530 }
1531
1532 static void
1533 virtio_set_vtpci_ops(struct virtio_hw *hw)
1534 {
1535 #ifdef RTE_VIRTIO_USER
1536         if (hw->virtio_user_dev)
1537                 VTPCI_OPS(hw) = &virtio_user_ops;
1538         else
1539 #endif
1540         if (hw->modern)
1541                 VTPCI_OPS(hw) = &modern_ops;
1542         else
1543                 VTPCI_OPS(hw) = &legacy_ops;
1544 }
1545
1546 /*
1547  * This function is based on probe() function in virtio_pci.c
1548  * It returns 0 on success.
1549  */
1550 int
1551 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1552 {
1553         struct virtio_hw *hw = eth_dev->data->dev_private;
1554         int ret;
1555
1556         RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1557
1558         eth_dev->dev_ops = &virtio_eth_dev_ops;
1559
1560         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1561                 if (!hw->virtio_user_dev) {
1562                         ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1563                         if (ret)
1564                                 return ret;
1565                 }
1566
1567                 virtio_set_vtpci_ops(hw);
1568                 set_rxtx_funcs(eth_dev);
1569
1570                 return 0;
1571         }
1572
1573         /* Allocate memory for storing MAC addresses */
1574         eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
1575         if (eth_dev->data->mac_addrs == NULL) {
1576                 PMD_INIT_LOG(ERR,
1577                         "Failed to allocate %d bytes needed to store MAC addresses",
1578                         VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
1579                 return -ENOMEM;
1580         }
1581
1582         hw->port_id = eth_dev->data->port_id;
1583         /* For virtio_user case the hw->virtio_user_dev is populated by
1584          * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1585          */
1586         if (!hw->virtio_user_dev) {
1587                 ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1588                 if (ret)
1589                         return ret;
1590         }
1591
1592         /* reset device and negotiate default features */
1593         ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1594         if (ret < 0)
1595                 return ret;
1596
1597         /* Setup interrupt callback  */
1598         if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1599                 rte_intr_callback_register(eth_dev->intr_handle,
1600                         virtio_interrupt_handler, eth_dev);
1601
1602         return 0;
1603 }
1604
1605 static int
1606 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1607 {
1608         PMD_INIT_FUNC_TRACE();
1609
1610         if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1611                 return -EPERM;
1612
1613         virtio_dev_stop(eth_dev);
1614         virtio_dev_close(eth_dev);
1615
1616         eth_dev->dev_ops = NULL;
1617         eth_dev->tx_pkt_burst = NULL;
1618         eth_dev->rx_pkt_burst = NULL;
1619
1620         rte_free(eth_dev->data->mac_addrs);
1621         eth_dev->data->mac_addrs = NULL;
1622
1623         /* reset interrupt callback  */
1624         if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1625                 rte_intr_callback_unregister(eth_dev->intr_handle,
1626                                                 virtio_interrupt_handler,
1627                                                 eth_dev);
1628         if (eth_dev->device)
1629                 rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1630
1631         PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1632
1633         return 0;
1634 }
1635
1636 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1637         struct rte_pci_device *pci_dev)
1638 {
1639         return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
1640                 eth_virtio_dev_init);
1641 }
1642
1643 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
1644 {
1645         return rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
1646 }
1647
1648 static struct rte_pci_driver rte_virtio_pmd = {
1649         .driver = {
1650                 .name = "net_virtio",
1651         },
1652         .id_table = pci_id_virtio_map,
1653         .drv_flags = 0,
1654         .probe = eth_virtio_pci_probe,
1655         .remove = eth_virtio_pci_remove,
1656 };
1657
1658 RTE_INIT(rte_virtio_pmd_init);
1659 static void
1660 rte_virtio_pmd_init(void)
1661 {
1662         if (rte_eal_iopl_init() != 0) {
1663                 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1664                 return;
1665         }
1666
1667         rte_pci_register(&rte_virtio_pmd);
1668 }
1669
1670 /*
1671  * Configure virtio device
1672  * It returns 0 on success.
1673  */
1674 static int
1675 virtio_dev_configure(struct rte_eth_dev *dev)
1676 {
1677         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1678         struct virtio_hw *hw = dev->data->dev_private;
1679         uint64_t req_features;
1680         int ret;
1681
1682         PMD_INIT_LOG(DEBUG, "configure");
1683         req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
1684
1685         if (dev->data->dev_conf.intr_conf.rxq) {
1686                 ret = virtio_init_device(dev, hw->req_guest_features);
1687                 if (ret < 0)
1688                         return ret;
1689         }
1690
1691         /* The name hw_ip_checksum is a bit confusing since it can be
1692          * set by the application to request L3 and/or L4 checksums. In
1693          * case of virtio, only L4 checksum is supported.
1694          */
1695         if (rxmode->hw_ip_checksum)
1696                 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
1697
1698         if (rxmode->enable_lro)
1699                 req_features |=
1700                         (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
1701                         (1ULL << VIRTIO_NET_F_GUEST_TSO6);
1702
1703         /* if request features changed, reinit the device */
1704         if (req_features != hw->req_guest_features) {
1705                 ret = virtio_init_device(dev, req_features);
1706                 if (ret < 0)
1707                         return ret;
1708         }
1709
1710         if (rxmode->hw_ip_checksum &&
1711                 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
1712                 PMD_DRV_LOG(ERR,
1713                         "rx checksum not available on this host");
1714                 return -ENOTSUP;
1715         }
1716
1717         if (rxmode->enable_lro &&
1718                 (!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
1719                         !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4))) {
1720                 PMD_DRV_LOG(ERR,
1721                         "Large Receive Offload not available on this host");
1722                 return -ENOTSUP;
1723         }
1724
1725         /* start control queue */
1726         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
1727                 virtio_dev_cq_start(dev);
1728
1729         hw->vlan_strip = rxmode->hw_vlan_strip;
1730
1731         if (rxmode->hw_vlan_filter
1732             && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1733                 PMD_DRV_LOG(ERR,
1734                             "vlan filtering not available on this host");
1735                 return -ENOTSUP;
1736         }
1737
1738         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1739                 /* Enable vector (0) for Link State Intrerrupt */
1740                 if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
1741                                 VIRTIO_MSI_NO_VECTOR) {
1742                         PMD_DRV_LOG(ERR, "failed to set config vector");
1743                         return -EBUSY;
1744                 }
1745
1746         hw->use_simple_rx = 1;
1747         hw->use_simple_tx = 1;
1748
1749 #if defined RTE_ARCH_ARM64 || defined CONFIG_RTE_ARCH_ARM
1750         if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
1751                 hw->use_simple_rx = 0;
1752                 hw->use_simple_tx = 0;
1753         }
1754 #endif
1755         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1756                 hw->use_simple_rx = 0;
1757                 hw->use_simple_tx = 0;
1758         }
1759
1760         if (rxmode->hw_ip_checksum)
1761                 hw->use_simple_rx = 0;
1762
1763         return 0;
1764 }
1765
1766
1767 static int
1768 virtio_dev_start(struct rte_eth_dev *dev)
1769 {
1770         uint16_t nb_queues, i;
1771         struct virtnet_rx *rxvq;
1772         struct virtnet_tx *txvq __rte_unused;
1773         struct virtio_hw *hw = dev->data->dev_private;
1774         int ret;
1775
1776         /* Finish the initialization of the queues */
1777         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1778                 ret = virtio_dev_rx_queue_setup_finish(dev, i);
1779                 if (ret < 0)
1780                         return ret;
1781         }
1782         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1783                 ret = virtio_dev_tx_queue_setup_finish(dev, i);
1784                 if (ret < 0)
1785                         return ret;
1786         }
1787
1788         /* check if lsc interrupt feature is enabled */
1789         if (dev->data->dev_conf.intr_conf.lsc) {
1790                 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1791                         PMD_DRV_LOG(ERR, "link status not supported by host");
1792                         return -ENOTSUP;
1793                 }
1794         }
1795
1796         /* Enable uio/vfio intr/eventfd mapping: althrough we already did that
1797          * in device configure, but it could be unmapped  when device is
1798          * stopped.
1799          */
1800         if (dev->data->dev_conf.intr_conf.lsc ||
1801             dev->data->dev_conf.intr_conf.rxq) {
1802                 rte_intr_disable(dev->intr_handle);
1803
1804                 if (rte_intr_enable(dev->intr_handle) < 0) {
1805                         PMD_DRV_LOG(ERR, "interrupt enable failed");
1806                         return -EIO;
1807                 }
1808         }
1809
1810         /*Notify the backend
1811          *Otherwise the tap backend might already stop its queue due to fullness.
1812          *vhost backend will have no chance to be waked up
1813          */
1814         nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
1815         if (hw->max_queue_pairs > 1) {
1816                 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1817                         return -EINVAL;
1818         }
1819
1820         PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1821
1822         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1823                 rxvq = dev->data->rx_queues[i];
1824                 virtqueue_notify(rxvq->vq);
1825         }
1826
1827         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1828                 txvq = dev->data->tx_queues[i];
1829                 virtqueue_notify(txvq->vq);
1830         }
1831
1832         PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1833
1834         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1835                 rxvq = dev->data->rx_queues[i];
1836                 VIRTQUEUE_DUMP(rxvq->vq);
1837         }
1838
1839         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1840                 txvq = dev->data->tx_queues[i];
1841                 VIRTQUEUE_DUMP(txvq->vq);
1842         }
1843
1844         set_rxtx_funcs(dev);
1845         hw->started = 1;
1846
1847         /* Initialize Link state */
1848         virtio_dev_link_update(dev, 0);
1849
1850         return 0;
1851 }
1852
1853 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1854 {
1855         struct rte_mbuf *buf;
1856         int i, mbuf_num = 0;
1857
1858         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1859                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1860
1861                 PMD_INIT_LOG(DEBUG,
1862                              "Before freeing rxq[%d] used and unused buf", i);
1863                 VIRTQUEUE_DUMP(rxvq->vq);
1864
1865                 PMD_INIT_LOG(DEBUG, "rx_queues[%d]=%p", i, rxvq);
1866                 while ((buf = virtqueue_detatch_unused(rxvq->vq)) != NULL) {
1867                         rte_pktmbuf_free(buf);
1868                         mbuf_num++;
1869                 }
1870
1871                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1872                 PMD_INIT_LOG(DEBUG,
1873                              "After freeing rxq[%d] used and unused buf", i);
1874                 VIRTQUEUE_DUMP(rxvq->vq);
1875         }
1876
1877         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1878                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1879
1880                 PMD_INIT_LOG(DEBUG,
1881                              "Before freeing txq[%d] used and unused bufs",
1882                              i);
1883                 VIRTQUEUE_DUMP(txvq->vq);
1884
1885                 mbuf_num = 0;
1886                 while ((buf = virtqueue_detatch_unused(txvq->vq)) != NULL) {
1887                         rte_pktmbuf_free(buf);
1888                         mbuf_num++;
1889                 }
1890
1891                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1892                 PMD_INIT_LOG(DEBUG,
1893                              "After freeing txq[%d] used and unused buf", i);
1894                 VIRTQUEUE_DUMP(txvq->vq);
1895         }
1896 }
1897
1898 /*
1899  * Stop device: disable interrupt and mark link down
1900  */
1901 static void
1902 virtio_dev_stop(struct rte_eth_dev *dev)
1903 {
1904         struct virtio_hw *hw = dev->data->dev_private;
1905         struct rte_eth_link link;
1906         struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
1907
1908         PMD_INIT_LOG(DEBUG, "stop");
1909
1910         if (intr_conf->lsc || intr_conf->rxq)
1911                 rte_intr_disable(dev->intr_handle);
1912
1913         hw->started = 0;
1914         memset(&link, 0, sizeof(link));
1915         virtio_dev_atomic_write_link_status(dev, &link);
1916 }
1917
1918 static int
1919 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1920 {
1921         struct rte_eth_link link, old;
1922         uint16_t status;
1923         struct virtio_hw *hw = dev->data->dev_private;
1924         memset(&link, 0, sizeof(link));
1925         virtio_dev_atomic_read_link_status(dev, &link);
1926         old = link;
1927         link.link_duplex = ETH_LINK_FULL_DUPLEX;
1928         link.link_speed  = SPEED_10G;
1929
1930         if (hw->started == 0) {
1931                 link.link_status = ETH_LINK_DOWN;
1932         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1933                 PMD_INIT_LOG(DEBUG, "Get link status from hw");
1934                 vtpci_read_dev_config(hw,
1935                                 offsetof(struct virtio_net_config, status),
1936                                 &status, sizeof(status));
1937                 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1938                         link.link_status = ETH_LINK_DOWN;
1939                         PMD_INIT_LOG(DEBUG, "Port %d is down",
1940                                      dev->data->port_id);
1941                 } else {
1942                         link.link_status = ETH_LINK_UP;
1943                         PMD_INIT_LOG(DEBUG, "Port %d is up",
1944                                      dev->data->port_id);
1945                 }
1946         } else {
1947                 link.link_status = ETH_LINK_UP;
1948         }
1949         virtio_dev_atomic_write_link_status(dev, &link);
1950
1951         return (old.link_status == link.link_status) ? -1 : 0;
1952 }
1953
1954 static void
1955 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1956 {
1957         uint64_t tso_mask, host_features;
1958         struct virtio_hw *hw = dev->data->dev_private;
1959
1960         dev_info->speed_capa = ETH_LINK_SPEED_10G; /* fake value */
1961
1962         dev_info->pci_dev = dev->device ? RTE_ETH_DEV_TO_PCI(dev) : NULL;
1963         dev_info->max_rx_queues =
1964                 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
1965         dev_info->max_tx_queues =
1966                 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
1967         dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1968         dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1969         dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1970         dev_info->default_txconf = (struct rte_eth_txconf) {
1971                 .txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
1972         };
1973
1974         host_features = VTPCI_OPS(hw)->get_features(hw);
1975         dev_info->rx_offload_capa = 0;
1976         if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
1977                 dev_info->rx_offload_capa |=
1978                         DEV_RX_OFFLOAD_TCP_CKSUM |
1979                         DEV_RX_OFFLOAD_UDP_CKSUM;
1980         }
1981         tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
1982                 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
1983         if ((host_features & tso_mask) == tso_mask)
1984                 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
1985
1986         dev_info->tx_offload_capa = 0;
1987         if (hw->guest_features & (1ULL << VIRTIO_NET_F_CSUM)) {
1988                 dev_info->tx_offload_capa |=
1989                         DEV_TX_OFFLOAD_UDP_CKSUM |
1990                         DEV_TX_OFFLOAD_TCP_CKSUM;
1991         }
1992         tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
1993                 (1ULL << VIRTIO_NET_F_HOST_TSO6);
1994         if ((hw->guest_features & tso_mask) == tso_mask)
1995                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
1996 }
1997
1998 /*
1999  * It enables testpmd to collect per queue stats.
2000  */
2001 static int
2002 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2003 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2004 __rte_unused uint8_t is_rx)
2005 {
2006         return 0;
2007 }
2008
2009 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
2010 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
2011 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");