a006d4982d7fce31016769832034016d8bdef5f8
[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 void 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         memcpy(&result, cvq->virtio_net_hdr_mz->addr,
258                         sizeof(struct virtio_pmd_ctrl));
259
260         return result.status;
261 }
262
263 static int
264 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
265 {
266         struct virtio_hw *hw = dev->data->dev_private;
267         struct virtio_pmd_ctrl ctrl;
268         int dlen[1];
269         int ret;
270
271         ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
272         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
273         memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
274
275         dlen[0] = sizeof(uint16_t);
276
277         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
278         if (ret) {
279                 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
280                           "failed, this is too late now...");
281                 return -EINVAL;
282         }
283
284         return 0;
285 }
286
287 static void
288 virtio_dev_queue_release(void *queue __rte_unused)
289 {
290         /* do nothing */
291 }
292
293 static int
294 virtio_get_queue_type(struct virtio_hw *hw, uint16_t vtpci_queue_idx)
295 {
296         if (vtpci_queue_idx == hw->max_queue_pairs * 2)
297                 return VTNET_CQ;
298         else if (vtpci_queue_idx % 2 == 0)
299                 return VTNET_RQ;
300         else
301                 return VTNET_TQ;
302 }
303
304 static uint16_t
305 virtio_get_nr_vq(struct virtio_hw *hw)
306 {
307         uint16_t nr_vq = hw->max_queue_pairs * 2;
308
309         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
310                 nr_vq += 1;
311
312         return nr_vq;
313 }
314
315 static void
316 virtio_init_vring(struct virtqueue *vq)
317 {
318         int size = vq->vq_nentries;
319         struct vring *vr = &vq->vq_ring;
320         uint8_t *ring_mem = vq->vq_ring_virt_mem;
321
322         PMD_INIT_FUNC_TRACE();
323
324         /*
325          * Reinitialise since virtio port might have been stopped and restarted
326          */
327         memset(ring_mem, 0, vq->vq_ring_size);
328         vring_init(vr, size, ring_mem, VIRTIO_PCI_VRING_ALIGN);
329         vq->vq_used_cons_idx = 0;
330         vq->vq_desc_head_idx = 0;
331         vq->vq_avail_idx = 0;
332         vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
333         vq->vq_free_cnt = vq->vq_nentries;
334         memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
335
336         vring_desc_init(vr->desc, size);
337
338         /*
339          * Disable device(host) interrupting guest
340          */
341         virtqueue_disable_intr(vq);
342 }
343
344 static int
345 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
346 {
347         char vq_name[VIRTQUEUE_MAX_NAME_SZ];
348         char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
349         const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
350         unsigned int vq_size, size;
351         struct virtio_hw *hw = dev->data->dev_private;
352         struct virtnet_rx *rxvq = NULL;
353         struct virtnet_tx *txvq = NULL;
354         struct virtnet_ctl *cvq = NULL;
355         struct virtqueue *vq;
356         size_t sz_hdr_mz = 0;
357         void *sw_ring = NULL;
358         int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
359         int ret;
360
361         PMD_INIT_LOG(DEBUG, "setting up queue: %u", vtpci_queue_idx);
362
363         /*
364          * Read the virtqueue size from the Queue Size field
365          * Always power of 2 and if 0 virtqueue does not exist
366          */
367         vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
368         PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
369         if (vq_size == 0) {
370                 PMD_INIT_LOG(ERR, "virtqueue does not exist");
371                 return -EINVAL;
372         }
373
374         if (!rte_is_power_of_2(vq_size)) {
375                 PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
376                 return -EINVAL;
377         }
378
379         snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
380                  dev->data->port_id, vtpci_queue_idx);
381
382         size = RTE_ALIGN_CEIL(sizeof(*vq) +
383                                 vq_size * sizeof(struct vq_desc_extra),
384                                 RTE_CACHE_LINE_SIZE);
385         if (queue_type == VTNET_TQ) {
386                 /*
387                  * For each xmit packet, allocate a virtio_net_hdr
388                  * and indirect ring elements
389                  */
390                 sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
391         } else if (queue_type == VTNET_CQ) {
392                 /* Allocate a page for control vq command, data and status */
393                 sz_hdr_mz = PAGE_SIZE;
394         }
395
396         vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
397                                 SOCKET_ID_ANY);
398         if (vq == NULL) {
399                 PMD_INIT_LOG(ERR, "can not allocate vq");
400                 return -ENOMEM;
401         }
402         hw->vqs[vtpci_queue_idx] = vq;
403
404         vq->hw = hw;
405         vq->vq_queue_index = vtpci_queue_idx;
406         vq->vq_nentries = vq_size;
407
408         /*
409          * Reserve a memzone for vring elements
410          */
411         size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
412         vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
413         PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
414                      size, vq->vq_ring_size);
415
416         mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
417                                          SOCKET_ID_ANY,
418                                          0, VIRTIO_PCI_VRING_ALIGN);
419         if (mz == NULL) {
420                 if (rte_errno == EEXIST)
421                         mz = rte_memzone_lookup(vq_name);
422                 if (mz == NULL) {
423                         ret = -ENOMEM;
424                         goto fail_q_alloc;
425                 }
426         }
427
428         memset(mz->addr, 0, mz->len);
429
430         vq->vq_ring_mem = mz->phys_addr;
431         vq->vq_ring_virt_mem = mz->addr;
432         PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%" PRIx64,
433                      (uint64_t)mz->phys_addr);
434         PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
435                      (uint64_t)(uintptr_t)mz->addr);
436
437         virtio_init_vring(vq);
438
439         if (sz_hdr_mz) {
440                 snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
441                          dev->data->port_id, vtpci_queue_idx);
442                 hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
443                                                      SOCKET_ID_ANY, 0,
444                                                      RTE_CACHE_LINE_SIZE);
445                 if (hdr_mz == NULL) {
446                         if (rte_errno == EEXIST)
447                                 hdr_mz = rte_memzone_lookup(vq_hdr_name);
448                         if (hdr_mz == NULL) {
449                                 ret = -ENOMEM;
450                                 goto fail_q_alloc;
451                         }
452                 }
453         }
454
455         if (queue_type == VTNET_RQ) {
456                 size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
457                                sizeof(vq->sw_ring[0]);
458
459                 sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
460                                 RTE_CACHE_LINE_SIZE, SOCKET_ID_ANY);
461                 if (!sw_ring) {
462                         PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
463                         ret = -ENOMEM;
464                         goto fail_q_alloc;
465                 }
466
467                 vq->sw_ring = sw_ring;
468                 rxvq = &vq->rxq;
469                 rxvq->vq = vq;
470                 rxvq->port_id = dev->data->port_id;
471                 rxvq->mz = mz;
472         } else if (queue_type == VTNET_TQ) {
473                 txvq = &vq->txq;
474                 txvq->vq = vq;
475                 txvq->port_id = dev->data->port_id;
476                 txvq->mz = mz;
477                 txvq->virtio_net_hdr_mz = hdr_mz;
478                 txvq->virtio_net_hdr_mem = hdr_mz->phys_addr;
479         } else if (queue_type == VTNET_CQ) {
480                 cvq = &vq->cq;
481                 cvq->vq = vq;
482                 cvq->mz = mz;
483                 cvq->virtio_net_hdr_mz = hdr_mz;
484                 cvq->virtio_net_hdr_mem = hdr_mz->phys_addr;
485                 memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
486
487                 hw->cvq = cvq;
488         }
489
490         /* For virtio_user case (that is when hw->dev is NULL), we use
491          * virtual address. And we need properly set _offset_, please see
492          * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
493          */
494         if (!hw->virtio_user_dev)
495                 vq->offset = offsetof(struct rte_mbuf, buf_physaddr);
496         else {
497                 vq->vq_ring_mem = (uintptr_t)mz->addr;
498                 vq->offset = offsetof(struct rte_mbuf, buf_addr);
499                 if (queue_type == VTNET_TQ)
500                         txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
501                 else if (queue_type == VTNET_CQ)
502                         cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
503         }
504
505         if (queue_type == VTNET_TQ) {
506                 struct virtio_tx_region *txr;
507                 unsigned int i;
508
509                 txr = hdr_mz->addr;
510                 memset(txr, 0, vq_size * sizeof(*txr));
511                 for (i = 0; i < vq_size; i++) {
512                         struct vring_desc *start_dp = txr[i].tx_indir;
513
514                         vring_desc_init(start_dp, RTE_DIM(txr[i].tx_indir));
515
516                         /* first indirect descriptor is always the tx header */
517                         start_dp->addr = txvq->virtio_net_hdr_mem
518                                 + i * sizeof(*txr)
519                                 + offsetof(struct virtio_tx_region, tx_hdr);
520
521                         start_dp->len = hw->vtnet_hdr_size;
522                         start_dp->flags = VRING_DESC_F_NEXT;
523                 }
524         }
525
526         if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
527                 PMD_INIT_LOG(ERR, "setup_queue failed");
528                 return -EINVAL;
529         }
530
531         return 0;
532
533 fail_q_alloc:
534         rte_free(sw_ring);
535         rte_memzone_free(hdr_mz);
536         rte_memzone_free(mz);
537         rte_free(vq);
538
539         return ret;
540 }
541
542 static void
543 virtio_free_queues(struct virtio_hw *hw)
544 {
545         uint16_t nr_vq = virtio_get_nr_vq(hw);
546         struct virtqueue *vq;
547         int queue_type;
548         uint16_t i;
549
550         if (hw->vqs == NULL)
551                 return;
552
553         for (i = 0; i < nr_vq; i++) {
554                 vq = hw->vqs[i];
555                 if (!vq)
556                         continue;
557
558                 queue_type = virtio_get_queue_type(hw, i);
559                 if (queue_type == VTNET_RQ) {
560                         rte_free(vq->sw_ring);
561                         rte_memzone_free(vq->rxq.mz);
562                 } else if (queue_type == VTNET_TQ) {
563                         rte_memzone_free(vq->txq.mz);
564                         rte_memzone_free(vq->txq.virtio_net_hdr_mz);
565                 } else {
566                         rte_memzone_free(vq->cq.mz);
567                         rte_memzone_free(vq->cq.virtio_net_hdr_mz);
568                 }
569
570                 rte_free(vq);
571                 hw->vqs[i] = NULL;
572         }
573
574         rte_free(hw->vqs);
575         hw->vqs = NULL;
576 }
577
578 static int
579 virtio_alloc_queues(struct rte_eth_dev *dev)
580 {
581         struct virtio_hw *hw = dev->data->dev_private;
582         uint16_t nr_vq = virtio_get_nr_vq(hw);
583         uint16_t i;
584         int ret;
585
586         hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
587         if (!hw->vqs) {
588                 PMD_INIT_LOG(ERR, "failed to allocate vqs");
589                 return -ENOMEM;
590         }
591
592         for (i = 0; i < nr_vq; i++) {
593                 ret = virtio_init_queue(dev, i);
594                 if (ret < 0) {
595                         virtio_free_queues(hw);
596                         return ret;
597                 }
598         }
599
600         return 0;
601 }
602
603 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
604
605 static void
606 virtio_dev_close(struct rte_eth_dev *dev)
607 {
608         struct virtio_hw *hw = dev->data->dev_private;
609         struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
610
611         PMD_INIT_LOG(DEBUG, "virtio_dev_close");
612
613         /* reset the NIC */
614         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
615                 VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
616         if (intr_conf->rxq)
617                 virtio_queues_unbind_intr(dev);
618
619         if (intr_conf->lsc || intr_conf->rxq) {
620                 rte_intr_disable(dev->intr_handle);
621                 rte_intr_efd_disable(dev->intr_handle);
622                 rte_free(dev->intr_handle->intr_vec);
623                 dev->intr_handle->intr_vec = NULL;
624         }
625
626         vtpci_reset(hw);
627         virtio_dev_free_mbufs(dev);
628         virtio_free_queues(hw);
629 }
630
631 static void
632 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
633 {
634         struct virtio_hw *hw = dev->data->dev_private;
635         struct virtio_pmd_ctrl ctrl;
636         int dlen[1];
637         int ret;
638
639         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
640                 PMD_INIT_LOG(INFO, "host does not support rx control");
641                 return;
642         }
643
644         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
645         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
646         ctrl.data[0] = 1;
647         dlen[0] = 1;
648
649         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
650         if (ret)
651                 PMD_INIT_LOG(ERR, "Failed to enable promisc");
652 }
653
654 static void
655 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
656 {
657         struct virtio_hw *hw = dev->data->dev_private;
658         struct virtio_pmd_ctrl ctrl;
659         int dlen[1];
660         int ret;
661
662         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
663                 PMD_INIT_LOG(INFO, "host does not support rx control");
664                 return;
665         }
666
667         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
668         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
669         ctrl.data[0] = 0;
670         dlen[0] = 1;
671
672         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
673         if (ret)
674                 PMD_INIT_LOG(ERR, "Failed to disable promisc");
675 }
676
677 static void
678 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
679 {
680         struct virtio_hw *hw = dev->data->dev_private;
681         struct virtio_pmd_ctrl ctrl;
682         int dlen[1];
683         int ret;
684
685         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
686                 PMD_INIT_LOG(INFO, "host does not support rx control");
687                 return;
688         }
689
690         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
691         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
692         ctrl.data[0] = 1;
693         dlen[0] = 1;
694
695         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
696         if (ret)
697                 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
698 }
699
700 static void
701 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
702 {
703         struct virtio_hw *hw = dev->data->dev_private;
704         struct virtio_pmd_ctrl ctrl;
705         int dlen[1];
706         int ret;
707
708         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
709                 PMD_INIT_LOG(INFO, "host does not support rx control");
710                 return;
711         }
712
713         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
714         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
715         ctrl.data[0] = 0;
716         dlen[0] = 1;
717
718         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
719         if (ret)
720                 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
721 }
722
723 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
724 static int
725 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
726 {
727         struct virtio_hw *hw = dev->data->dev_private;
728         uint32_t ether_hdr_len = ETHER_HDR_LEN + VLAN_TAG_LEN +
729                                  hw->vtnet_hdr_size;
730         uint32_t frame_size = mtu + ether_hdr_len;
731         uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
732
733         max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
734
735         if (mtu < ETHER_MIN_MTU || frame_size > max_frame_size) {
736                 PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
737                         ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
738                 return -EINVAL;
739         }
740         return 0;
741 }
742
743 static int
744 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
745 {
746         struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
747         struct virtqueue *vq = rxvq->vq;
748
749         virtqueue_enable_intr(vq);
750         return 0;
751 }
752
753 static int
754 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
755 {
756         struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
757         struct virtqueue *vq = rxvq->vq;
758
759         virtqueue_disable_intr(vq);
760         return 0;
761 }
762
763 /*
764  * dev_ops for virtio, bare necessities for basic operation
765  */
766 static const struct eth_dev_ops virtio_eth_dev_ops = {
767         .dev_configure           = virtio_dev_configure,
768         .dev_start               = virtio_dev_start,
769         .dev_stop                = virtio_dev_stop,
770         .dev_close               = virtio_dev_close,
771         .promiscuous_enable      = virtio_dev_promiscuous_enable,
772         .promiscuous_disable     = virtio_dev_promiscuous_disable,
773         .allmulticast_enable     = virtio_dev_allmulticast_enable,
774         .allmulticast_disable    = virtio_dev_allmulticast_disable,
775         .mtu_set                 = virtio_mtu_set,
776         .dev_infos_get           = virtio_dev_info_get,
777         .stats_get               = virtio_dev_stats_get,
778         .xstats_get              = virtio_dev_xstats_get,
779         .xstats_get_names        = virtio_dev_xstats_get_names,
780         .stats_reset             = virtio_dev_stats_reset,
781         .xstats_reset            = virtio_dev_stats_reset,
782         .link_update             = virtio_dev_link_update,
783         .rx_queue_setup          = virtio_dev_rx_queue_setup,
784         .rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
785         .rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
786         .rx_queue_release        = virtio_dev_queue_release,
787         .rx_descriptor_done      = virtio_dev_rx_queue_done,
788         .tx_queue_setup          = virtio_dev_tx_queue_setup,
789         .tx_queue_release        = virtio_dev_queue_release,
790         /* collect stats per queue */
791         .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
792         .vlan_filter_set         = virtio_vlan_filter_set,
793         .mac_addr_add            = virtio_mac_addr_add,
794         .mac_addr_remove         = virtio_mac_addr_remove,
795         .mac_addr_set            = virtio_mac_addr_set,
796 };
797
798 static inline int
799 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
800                                 struct rte_eth_link *link)
801 {
802         struct rte_eth_link *dst = link;
803         struct rte_eth_link *src = &(dev->data->dev_link);
804
805         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
806                         *(uint64_t *)src) == 0)
807                 return -1;
808
809         return 0;
810 }
811
812 /**
813  * Atomically writes the link status information into global
814  * structure rte_eth_dev.
815  *
816  * @param dev
817  *   - Pointer to the structure rte_eth_dev to read from.
818  *   - Pointer to the buffer to be saved with the link status.
819  *
820  * @return
821  *   - On success, zero.
822  *   - On failure, negative value.
823  */
824 static inline int
825 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
826                 struct rte_eth_link *link)
827 {
828         struct rte_eth_link *dst = &(dev->data->dev_link);
829         struct rte_eth_link *src = link;
830
831         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
832                                         *(uint64_t *)src) == 0)
833                 return -1;
834
835         return 0;
836 }
837
838 static void
839 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
840 {
841         unsigned i;
842
843         for (i = 0; i < dev->data->nb_tx_queues; i++) {
844                 const struct virtnet_tx *txvq = dev->data->tx_queues[i];
845                 if (txvq == NULL)
846                         continue;
847
848                 stats->opackets += txvq->stats.packets;
849                 stats->obytes += txvq->stats.bytes;
850                 stats->oerrors += txvq->stats.errors;
851
852                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
853                         stats->q_opackets[i] = txvq->stats.packets;
854                         stats->q_obytes[i] = txvq->stats.bytes;
855                 }
856         }
857
858         for (i = 0; i < dev->data->nb_rx_queues; i++) {
859                 const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
860                 if (rxvq == NULL)
861                         continue;
862
863                 stats->ipackets += rxvq->stats.packets;
864                 stats->ibytes += rxvq->stats.bytes;
865                 stats->ierrors += rxvq->stats.errors;
866
867                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
868                         stats->q_ipackets[i] = rxvq->stats.packets;
869                         stats->q_ibytes[i] = rxvq->stats.bytes;
870                 }
871         }
872
873         stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
874 }
875
876 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
877                                        struct rte_eth_xstat_name *xstats_names,
878                                        __rte_unused unsigned limit)
879 {
880         unsigned i;
881         unsigned count = 0;
882         unsigned t;
883
884         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
885                 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
886
887         if (xstats_names != NULL) {
888                 /* Note: limit checked in rte_eth_xstats_names() */
889
890                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
891                         struct virtqueue *rxvq = dev->data->rx_queues[i];
892                         if (rxvq == NULL)
893                                 continue;
894                         for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
895                                 snprintf(xstats_names[count].name,
896                                         sizeof(xstats_names[count].name),
897                                         "rx_q%u_%s", i,
898                                         rte_virtio_rxq_stat_strings[t].name);
899                                 count++;
900                         }
901                 }
902
903                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
904                         struct virtqueue *txvq = dev->data->tx_queues[i];
905                         if (txvq == NULL)
906                                 continue;
907                         for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
908                                 snprintf(xstats_names[count].name,
909                                         sizeof(xstats_names[count].name),
910                                         "tx_q%u_%s", i,
911                                         rte_virtio_txq_stat_strings[t].name);
912                                 count++;
913                         }
914                 }
915                 return count;
916         }
917         return nstats;
918 }
919
920 static int
921 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
922                       unsigned n)
923 {
924         unsigned i;
925         unsigned count = 0;
926
927         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
928                 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
929
930         if (n < nstats)
931                 return nstats;
932
933         for (i = 0; i < dev->data->nb_rx_queues; i++) {
934                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
935
936                 if (rxvq == NULL)
937                         continue;
938
939                 unsigned t;
940
941                 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
942                         xstats[count].value = *(uint64_t *)(((char *)rxvq) +
943                                 rte_virtio_rxq_stat_strings[t].offset);
944                         xstats[count].id = count;
945                         count++;
946                 }
947         }
948
949         for (i = 0; i < dev->data->nb_tx_queues; i++) {
950                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
951
952                 if (txvq == NULL)
953                         continue;
954
955                 unsigned t;
956
957                 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
958                         xstats[count].value = *(uint64_t *)(((char *)txvq) +
959                                 rte_virtio_txq_stat_strings[t].offset);
960                         xstats[count].id = count;
961                         count++;
962                 }
963         }
964
965         return count;
966 }
967
968 static void
969 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
970 {
971         virtio_update_stats(dev, stats);
972 }
973
974 static void
975 virtio_dev_stats_reset(struct rte_eth_dev *dev)
976 {
977         unsigned int i;
978
979         for (i = 0; i < dev->data->nb_tx_queues; i++) {
980                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
981                 if (txvq == NULL)
982                         continue;
983
984                 txvq->stats.packets = 0;
985                 txvq->stats.bytes = 0;
986                 txvq->stats.errors = 0;
987                 txvq->stats.multicast = 0;
988                 txvq->stats.broadcast = 0;
989                 memset(txvq->stats.size_bins, 0,
990                        sizeof(txvq->stats.size_bins[0]) * 8);
991         }
992
993         for (i = 0; i < dev->data->nb_rx_queues; i++) {
994                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
995                 if (rxvq == NULL)
996                         continue;
997
998                 rxvq->stats.packets = 0;
999                 rxvq->stats.bytes = 0;
1000                 rxvq->stats.errors = 0;
1001                 rxvq->stats.multicast = 0;
1002                 rxvq->stats.broadcast = 0;
1003                 memset(rxvq->stats.size_bins, 0,
1004                        sizeof(rxvq->stats.size_bins[0]) * 8);
1005         }
1006 }
1007
1008 static void
1009 virtio_set_hwaddr(struct virtio_hw *hw)
1010 {
1011         vtpci_write_dev_config(hw,
1012                         offsetof(struct virtio_net_config, mac),
1013                         &hw->mac_addr, ETHER_ADDR_LEN);
1014 }
1015
1016 static void
1017 virtio_get_hwaddr(struct virtio_hw *hw)
1018 {
1019         if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1020                 vtpci_read_dev_config(hw,
1021                         offsetof(struct virtio_net_config, mac),
1022                         &hw->mac_addr, ETHER_ADDR_LEN);
1023         } else {
1024                 eth_random_addr(&hw->mac_addr[0]);
1025                 virtio_set_hwaddr(hw);
1026         }
1027 }
1028
1029 static int
1030 virtio_mac_table_set(struct virtio_hw *hw,
1031                      const struct virtio_net_ctrl_mac *uc,
1032                      const struct virtio_net_ctrl_mac *mc)
1033 {
1034         struct virtio_pmd_ctrl ctrl;
1035         int err, len[2];
1036
1037         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1038                 PMD_DRV_LOG(INFO, "host does not support mac table");
1039                 return -1;
1040         }
1041
1042         ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1043         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1044
1045         len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
1046         memcpy(ctrl.data, uc, len[0]);
1047
1048         len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
1049         memcpy(ctrl.data + len[0], mc, len[1]);
1050
1051         err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1052         if (err != 0)
1053                 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1054         return err;
1055 }
1056
1057 static int
1058 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
1059                     uint32_t index, uint32_t vmdq __rte_unused)
1060 {
1061         struct virtio_hw *hw = dev->data->dev_private;
1062         const struct ether_addr *addrs = dev->data->mac_addrs;
1063         unsigned int i;
1064         struct virtio_net_ctrl_mac *uc, *mc;
1065
1066         if (index >= VIRTIO_MAX_MAC_ADDRS) {
1067                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1068                 return -EINVAL;
1069         }
1070
1071         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1072         uc->entries = 0;
1073         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1074         mc->entries = 0;
1075
1076         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1077                 const struct ether_addr *addr
1078                         = (i == index) ? mac_addr : addrs + i;
1079                 struct virtio_net_ctrl_mac *tbl
1080                         = is_multicast_ether_addr(addr) ? mc : uc;
1081
1082                 memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
1083         }
1084
1085         return virtio_mac_table_set(hw, uc, mc);
1086 }
1087
1088 static void
1089 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1090 {
1091         struct virtio_hw *hw = dev->data->dev_private;
1092         struct ether_addr *addrs = dev->data->mac_addrs;
1093         struct virtio_net_ctrl_mac *uc, *mc;
1094         unsigned int i;
1095
1096         if (index >= VIRTIO_MAX_MAC_ADDRS) {
1097                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1098                 return;
1099         }
1100
1101         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1102         uc->entries = 0;
1103         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1104         mc->entries = 0;
1105
1106         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1107                 struct virtio_net_ctrl_mac *tbl;
1108
1109                 if (i == index || is_zero_ether_addr(addrs + i))
1110                         continue;
1111
1112                 tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
1113                 memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
1114         }
1115
1116         virtio_mac_table_set(hw, uc, mc);
1117 }
1118
1119 static void
1120 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1121 {
1122         struct virtio_hw *hw = dev->data->dev_private;
1123
1124         memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
1125
1126         /* Use atomic update if available */
1127         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1128                 struct virtio_pmd_ctrl ctrl;
1129                 int len = ETHER_ADDR_LEN;
1130
1131                 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1132                 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1133
1134                 memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
1135                 virtio_send_command(hw->cvq, &ctrl, &len, 1);
1136         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1137                 virtio_set_hwaddr(hw);
1138 }
1139
1140 static int
1141 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1142 {
1143         struct virtio_hw *hw = dev->data->dev_private;
1144         struct virtio_pmd_ctrl ctrl;
1145         int len;
1146
1147         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1148                 return -ENOTSUP;
1149
1150         ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1151         ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1152         memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1153         len = sizeof(vlan_id);
1154
1155         return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1156 }
1157
1158 static int
1159 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1160 {
1161         uint64_t host_features;
1162
1163         /* Prepare guest_features: feature that driver wants to support */
1164         PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1165                 req_features);
1166
1167         /* Read device(host) feature bits */
1168         host_features = VTPCI_OPS(hw)->get_features(hw);
1169         PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1170                 host_features);
1171
1172         /* If supported, ensure MTU value is valid before acknowledging it. */
1173         if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1174                 struct virtio_net_config config;
1175
1176                 vtpci_read_dev_config(hw,
1177                         offsetof(struct virtio_net_config, mtu),
1178                         &config.mtu, sizeof(config.mtu));
1179
1180                 if (config.mtu < ETHER_MIN_MTU)
1181                         req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1182         }
1183
1184         /*
1185          * Negotiate features: Subset of device feature bits are written back
1186          * guest feature bits.
1187          */
1188         hw->guest_features = req_features;
1189         hw->guest_features = vtpci_negotiate_features(hw, host_features);
1190         PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1191                 hw->guest_features);
1192
1193         if (hw->modern) {
1194                 if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1195                         PMD_INIT_LOG(ERR,
1196                                 "VIRTIO_F_VERSION_1 features is not enabled.");
1197                         return -1;
1198                 }
1199                 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1200                 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1201                         PMD_INIT_LOG(ERR,
1202                                 "failed to set FEATURES_OK status!");
1203                         return -1;
1204                 }
1205         }
1206
1207         hw->req_guest_features = req_features;
1208
1209         return 0;
1210 }
1211
1212 /*
1213  * Process Virtio Config changed interrupt and call the callback
1214  * if link state changed.
1215  */
1216 void
1217 virtio_interrupt_handler(void *param)
1218 {
1219         struct rte_eth_dev *dev = param;
1220         struct virtio_hw *hw = dev->data->dev_private;
1221         uint8_t isr;
1222
1223         /* Read interrupt status which clears interrupt */
1224         isr = vtpci_isr(hw);
1225         PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1226
1227         if (rte_intr_enable(dev->intr_handle) < 0)
1228                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1229
1230         if (isr & VIRTIO_PCI_ISR_CONFIG) {
1231                 if (virtio_dev_link_update(dev, 0) == 0)
1232                         _rte_eth_dev_callback_process(dev,
1233                                                       RTE_ETH_EVENT_INTR_LSC,
1234                                                       NULL, NULL);
1235         }
1236
1237 }
1238
1239 /* set rx and tx handlers according to what is supported */
1240 static void
1241 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1242 {
1243         struct virtio_hw *hw = eth_dev->data->dev_private;
1244
1245         if (hw->use_simple_rx) {
1246                 PMD_INIT_LOG(INFO, "virtio: using simple Rx path on port %u",
1247                         eth_dev->data->port_id);
1248                 eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1249         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1250                 PMD_INIT_LOG(INFO,
1251                         "virtio: using mergeable buffer Rx path on port %u",
1252                         eth_dev->data->port_id);
1253                 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1254         } else {
1255                 PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1256                         eth_dev->data->port_id);
1257                 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1258         }
1259
1260         if (hw->use_simple_tx) {
1261                 PMD_INIT_LOG(INFO, "virtio: using simple Tx path on port %u",
1262                         eth_dev->data->port_id);
1263                 eth_dev->tx_pkt_burst = virtio_xmit_pkts_simple;
1264         } else {
1265                 PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1266                         eth_dev->data->port_id);
1267                 eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1268         }
1269 }
1270
1271 /* Only support 1:1 queue/interrupt mapping so far.
1272  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1273  * interrupt vectors (<N+1).
1274  */
1275 static int
1276 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1277 {
1278         uint32_t i;
1279         struct virtio_hw *hw = dev->data->dev_private;
1280
1281         PMD_INIT_LOG(INFO, "queue/interrupt binding");
1282         for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1283                 dev->intr_handle->intr_vec[i] = i + 1;
1284                 if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1285                                                  VIRTIO_MSI_NO_VECTOR) {
1286                         PMD_DRV_LOG(ERR, "failed to set queue vector");
1287                         return -EBUSY;
1288                 }
1289         }
1290
1291         return 0;
1292 }
1293
1294 static void
1295 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1296 {
1297         uint32_t i;
1298         struct virtio_hw *hw = dev->data->dev_private;
1299
1300         PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1301         for (i = 0; i < dev->data->nb_rx_queues; ++i)
1302                 VTPCI_OPS(hw)->set_queue_irq(hw,
1303                                              hw->vqs[i * VTNET_CQ],
1304                                              VIRTIO_MSI_NO_VECTOR);
1305 }
1306
1307 static int
1308 virtio_configure_intr(struct rte_eth_dev *dev)
1309 {
1310         struct virtio_hw *hw = dev->data->dev_private;
1311
1312         if (!rte_intr_cap_multiple(dev->intr_handle)) {
1313                 PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1314                 return -ENOTSUP;
1315         }
1316
1317         if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1318                 PMD_INIT_LOG(ERR, "Fail to create eventfd");
1319                 return -1;
1320         }
1321
1322         if (!dev->intr_handle->intr_vec) {
1323                 dev->intr_handle->intr_vec =
1324                         rte_zmalloc("intr_vec",
1325                                     hw->max_queue_pairs * sizeof(int), 0);
1326                 if (!dev->intr_handle->intr_vec) {
1327                         PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1328                                      hw->max_queue_pairs);
1329                         return -ENOMEM;
1330                 }
1331         }
1332
1333         /* Re-register callback to update max_intr */
1334         rte_intr_callback_unregister(dev->intr_handle,
1335                                      virtio_interrupt_handler,
1336                                      dev);
1337         rte_intr_callback_register(dev->intr_handle,
1338                                    virtio_interrupt_handler,
1339                                    dev);
1340
1341         /* DO NOT try to remove this! This function will enable msix, or QEMU
1342          * will encounter SIGSEGV when DRIVER_OK is sent.
1343          * And for legacy devices, this should be done before queue/vec binding
1344          * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1345          * (22) will be ignored.
1346          */
1347         if (rte_intr_enable(dev->intr_handle) < 0) {
1348                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1349                 return -1;
1350         }
1351
1352         if (virtio_queues_bind_intr(dev) < 0) {
1353                 PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1354                 return -1;
1355         }
1356
1357         return 0;
1358 }
1359
1360 /* reset device and renegotiate features if needed */
1361 static int
1362 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1363 {
1364         struct virtio_hw *hw = eth_dev->data->dev_private;
1365         struct virtio_net_config *config;
1366         struct virtio_net_config local_config;
1367         struct rte_pci_device *pci_dev = NULL;
1368         int ret;
1369
1370         /* Reset the device although not necessary at startup */
1371         vtpci_reset(hw);
1372
1373         /* Tell the host we've noticed this device. */
1374         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1375
1376         /* Tell the host we've known how to drive the device. */
1377         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1378         if (virtio_negotiate_features(hw, req_features) < 0)
1379                 return -1;
1380
1381         if (!hw->virtio_user_dev) {
1382                 pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1383                 rte_eth_copy_pci_info(eth_dev, pci_dev);
1384         }
1385
1386         eth_dev->data->dev_flags = RTE_ETH_DEV_DETACHABLE;
1387         /* If host does not support both status and MSI-X then disable LSC */
1388         if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) && hw->use_msix)
1389                 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1390         else
1391                 eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1392
1393         /* Setting up rx_header size for the device */
1394         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1395             vtpci_with_feature(hw, VIRTIO_F_VERSION_1))
1396                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1397         else
1398                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1399
1400         /* Copy the permanent MAC address to: virtio_hw */
1401         virtio_get_hwaddr(hw);
1402         ether_addr_copy((struct ether_addr *) hw->mac_addr,
1403                         &eth_dev->data->mac_addrs[0]);
1404         PMD_INIT_LOG(DEBUG,
1405                      "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1406                      hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1407                      hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1408
1409         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1410                 config = &local_config;
1411
1412                 vtpci_read_dev_config(hw,
1413                         offsetof(struct virtio_net_config, mac),
1414                         &config->mac, sizeof(config->mac));
1415
1416                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1417                         vtpci_read_dev_config(hw,
1418                                 offsetof(struct virtio_net_config, status),
1419                                 &config->status, sizeof(config->status));
1420                 } else {
1421                         PMD_INIT_LOG(DEBUG,
1422                                      "VIRTIO_NET_F_STATUS is not supported");
1423                         config->status = 0;
1424                 }
1425
1426                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1427                         vtpci_read_dev_config(hw,
1428                                 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1429                                 &config->max_virtqueue_pairs,
1430                                 sizeof(config->max_virtqueue_pairs));
1431                 } else {
1432                         PMD_INIT_LOG(DEBUG,
1433                                      "VIRTIO_NET_F_MQ is not supported");
1434                         config->max_virtqueue_pairs = 1;
1435                 }
1436
1437                 hw->max_queue_pairs = config->max_virtqueue_pairs;
1438
1439                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1440                         vtpci_read_dev_config(hw,
1441                                 offsetof(struct virtio_net_config, mtu),
1442                                 &config->mtu,
1443                                 sizeof(config->mtu));
1444
1445                         /*
1446                          * MTU value has already been checked at negotiation
1447                          * time, but check again in case it has changed since
1448                          * then, which should not happen.
1449                          */
1450                         if (config->mtu < ETHER_MIN_MTU) {
1451                                 PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1452                                                 config->mtu);
1453                                 return -1;
1454                         }
1455
1456                         hw->max_mtu = config->mtu;
1457                         /* Set initial MTU to maximum one supported by vhost */
1458                         eth_dev->data->mtu = config->mtu;
1459
1460                 } else {
1461                         hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - ETHER_HDR_LEN -
1462                                 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1463                 }
1464
1465                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1466                                 config->max_virtqueue_pairs);
1467                 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1468                 PMD_INIT_LOG(DEBUG,
1469                                 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1470                                 config->mac[0], config->mac[1],
1471                                 config->mac[2], config->mac[3],
1472                                 config->mac[4], config->mac[5]);
1473         } else {
1474                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1475                 hw->max_queue_pairs = 1;
1476         }
1477
1478         ret = virtio_alloc_queues(eth_dev);
1479         if (ret < 0)
1480                 return ret;
1481
1482         if (eth_dev->data->dev_conf.intr_conf.rxq) {
1483                 if (virtio_configure_intr(eth_dev) < 0) {
1484                         PMD_INIT_LOG(ERR, "failed to configure interrupt");
1485                         return -1;
1486                 }
1487         }
1488
1489         vtpci_reinit_complete(hw);
1490
1491         if (pci_dev)
1492                 PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1493                         eth_dev->data->port_id, pci_dev->id.vendor_id,
1494                         pci_dev->id.device_id);
1495
1496         return 0;
1497 }
1498
1499 /*
1500  * Remap the PCI device again (IO port map for legacy device and
1501  * memory map for modern device), so that the secondary process
1502  * could have the PCI initiated correctly.
1503  */
1504 static int
1505 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1506 {
1507         if (hw->modern) {
1508                 /*
1509                  * We don't have to re-parse the PCI config space, since
1510                  * rte_pci_map_device() makes sure the mapped address
1511                  * in secondary process would equal to the one mapped in
1512                  * the primary process: error will be returned if that
1513                  * requirement is not met.
1514                  *
1515                  * That said, we could simply reuse all cap pointers
1516                  * (such as dev_cfg, common_cfg, etc.) parsed from the
1517                  * primary process, which is stored in shared memory.
1518                  */
1519                 if (rte_pci_map_device(pci_dev)) {
1520                         PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1521                         return -1;
1522                 }
1523         } else {
1524                 if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1525                         return -1;
1526         }
1527
1528         return 0;
1529 }
1530
1531 static void
1532 virtio_set_vtpci_ops(struct virtio_hw *hw)
1533 {
1534 #ifdef RTE_VIRTIO_USER
1535         if (hw->virtio_user_dev)
1536                 VTPCI_OPS(hw) = &virtio_user_ops;
1537         else
1538 #endif
1539         if (hw->modern)
1540                 VTPCI_OPS(hw) = &modern_ops;
1541         else
1542                 VTPCI_OPS(hw) = &legacy_ops;
1543 }
1544
1545 /*
1546  * This function is based on probe() function in virtio_pci.c
1547  * It returns 0 on success.
1548  */
1549 int
1550 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1551 {
1552         struct virtio_hw *hw = eth_dev->data->dev_private;
1553         int ret;
1554
1555         RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1556
1557         eth_dev->dev_ops = &virtio_eth_dev_ops;
1558
1559         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1560                 if (!hw->virtio_user_dev) {
1561                         ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1562                         if (ret)
1563                                 return ret;
1564                 }
1565
1566                 virtio_set_vtpci_ops(hw);
1567                 set_rxtx_funcs(eth_dev);
1568
1569                 return 0;
1570         }
1571
1572         /* Allocate memory for storing MAC addresses */
1573         eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
1574         if (eth_dev->data->mac_addrs == NULL) {
1575                 PMD_INIT_LOG(ERR,
1576                         "Failed to allocate %d bytes needed to store MAC addresses",
1577                         VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
1578                 return -ENOMEM;
1579         }
1580
1581         hw->port_id = eth_dev->data->port_id;
1582         /* For virtio_user case the hw->virtio_user_dev is populated by
1583          * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1584          */
1585         if (!hw->virtio_user_dev) {
1586                 ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1587                 if (ret)
1588                         return ret;
1589         }
1590
1591         /* reset device and negotiate default features */
1592         ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1593         if (ret < 0)
1594                 return ret;
1595
1596         /* Setup interrupt callback  */
1597         if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1598                 rte_intr_callback_register(eth_dev->intr_handle,
1599                         virtio_interrupt_handler, eth_dev);
1600
1601         return 0;
1602 }
1603
1604 static int
1605 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1606 {
1607         PMD_INIT_FUNC_TRACE();
1608
1609         if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1610                 return -EPERM;
1611
1612         virtio_dev_stop(eth_dev);
1613         virtio_dev_close(eth_dev);
1614
1615         eth_dev->dev_ops = NULL;
1616         eth_dev->tx_pkt_burst = NULL;
1617         eth_dev->rx_pkt_burst = NULL;
1618
1619         rte_free(eth_dev->data->mac_addrs);
1620         eth_dev->data->mac_addrs = NULL;
1621
1622         /* reset interrupt callback  */
1623         if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1624                 rte_intr_callback_unregister(eth_dev->intr_handle,
1625                                                 virtio_interrupt_handler,
1626                                                 eth_dev);
1627         if (eth_dev->device)
1628                 rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1629
1630         PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1631
1632         return 0;
1633 }
1634
1635 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1636         struct rte_pci_device *pci_dev)
1637 {
1638         return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
1639                 eth_virtio_dev_init);
1640 }
1641
1642 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
1643 {
1644         return rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
1645 }
1646
1647 static struct rte_pci_driver rte_virtio_pmd = {
1648         .driver = {
1649                 .name = "net_virtio",
1650         },
1651         .id_table = pci_id_virtio_map,
1652         .drv_flags = 0,
1653         .probe = eth_virtio_pci_probe,
1654         .remove = eth_virtio_pci_remove,
1655 };
1656
1657 RTE_INIT(rte_virtio_pmd_init);
1658 static void
1659 rte_virtio_pmd_init(void)
1660 {
1661         if (rte_eal_iopl_init() != 0) {
1662                 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1663                 return;
1664         }
1665
1666         rte_pci_register(&rte_virtio_pmd);
1667 }
1668
1669 /*
1670  * Configure virtio device
1671  * It returns 0 on success.
1672  */
1673 static int
1674 virtio_dev_configure(struct rte_eth_dev *dev)
1675 {
1676         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1677         struct virtio_hw *hw = dev->data->dev_private;
1678         uint64_t req_features;
1679         int ret;
1680
1681         PMD_INIT_LOG(DEBUG, "configure");
1682         req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
1683
1684         if (dev->data->dev_conf.intr_conf.rxq) {
1685                 ret = virtio_init_device(dev, hw->req_guest_features);
1686                 if (ret < 0)
1687                         return ret;
1688         }
1689
1690         /* The name hw_ip_checksum is a bit confusing since it can be
1691          * set by the application to request L3 and/or L4 checksums. In
1692          * case of virtio, only L4 checksum is supported.
1693          */
1694         if (rxmode->hw_ip_checksum)
1695                 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
1696
1697         if (rxmode->enable_lro)
1698                 req_features |=
1699                         (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
1700                         (1ULL << VIRTIO_NET_F_GUEST_TSO6);
1701
1702         /* if request features changed, reinit the device */
1703         if (req_features != hw->req_guest_features) {
1704                 ret = virtio_init_device(dev, req_features);
1705                 if (ret < 0)
1706                         return ret;
1707         }
1708
1709         if (rxmode->hw_ip_checksum &&
1710                 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
1711                 PMD_DRV_LOG(ERR,
1712                         "rx checksum not available on this host");
1713                 return -ENOTSUP;
1714         }
1715
1716         if (rxmode->enable_lro &&
1717                 (!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
1718                         !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4))) {
1719                 PMD_DRV_LOG(ERR,
1720                         "Large Receive Offload not available on this host");
1721                 return -ENOTSUP;
1722         }
1723
1724         /* start control queue */
1725         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
1726                 virtio_dev_cq_start(dev);
1727
1728         hw->vlan_strip = rxmode->hw_vlan_strip;
1729
1730         if (rxmode->hw_vlan_filter
1731             && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1732                 PMD_DRV_LOG(ERR,
1733                             "vlan filtering not available on this host");
1734                 return -ENOTSUP;
1735         }
1736
1737         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1738                 /* Enable vector (0) for Link State Intrerrupt */
1739                 if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
1740                                 VIRTIO_MSI_NO_VECTOR) {
1741                         PMD_DRV_LOG(ERR, "failed to set config vector");
1742                         return -EBUSY;
1743                 }
1744
1745         hw->use_simple_rx = 1;
1746         hw->use_simple_tx = 1;
1747
1748 #if defined RTE_ARCH_ARM64 || defined CONFIG_RTE_ARCH_ARM
1749         if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
1750                 hw->use_simple_rx = 0;
1751                 hw->use_simple_tx = 0;
1752         }
1753 #endif
1754         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1755                 hw->use_simple_rx = 0;
1756                 hw->use_simple_tx = 0;
1757         }
1758
1759         if (rxmode->hw_ip_checksum)
1760                 hw->use_simple_rx = 0;
1761
1762         return 0;
1763 }
1764
1765
1766 static int
1767 virtio_dev_start(struct rte_eth_dev *dev)
1768 {
1769         uint16_t nb_queues, i;
1770         struct virtnet_rx *rxvq;
1771         struct virtnet_tx *txvq __rte_unused;
1772         struct virtio_hw *hw = dev->data->dev_private;
1773         int ret;
1774
1775         /* Finish the initialization of the queues */
1776         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1777                 ret = virtio_dev_rx_queue_setup_finish(dev, i);
1778                 if (ret < 0)
1779                         return ret;
1780         }
1781         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1782                 ret = virtio_dev_tx_queue_setup_finish(dev, i);
1783                 if (ret < 0)
1784                         return ret;
1785         }
1786
1787         /* check if lsc interrupt feature is enabled */
1788         if (dev->data->dev_conf.intr_conf.lsc) {
1789                 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1790                         PMD_DRV_LOG(ERR, "link status not supported by host");
1791                         return -ENOTSUP;
1792                 }
1793         }
1794
1795         /* Enable uio/vfio intr/eventfd mapping: althrough we already did that
1796          * in device configure, but it could be unmapped  when device is
1797          * stopped.
1798          */
1799         if (dev->data->dev_conf.intr_conf.lsc ||
1800             dev->data->dev_conf.intr_conf.rxq) {
1801                 rte_intr_disable(dev->intr_handle);
1802
1803                 if (rte_intr_enable(dev->intr_handle) < 0) {
1804                         PMD_DRV_LOG(ERR, "interrupt enable failed");
1805                         return -EIO;
1806                 }
1807         }
1808
1809         /*Notify the backend
1810          *Otherwise the tap backend might already stop its queue due to fullness.
1811          *vhost backend will have no chance to be waked up
1812          */
1813         nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
1814         if (hw->max_queue_pairs > 1) {
1815                 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1816                         return -EINVAL;
1817         }
1818
1819         PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1820
1821         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1822                 rxvq = dev->data->rx_queues[i];
1823                 virtqueue_notify(rxvq->vq);
1824         }
1825
1826         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1827                 txvq = dev->data->tx_queues[i];
1828                 virtqueue_notify(txvq->vq);
1829         }
1830
1831         PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1832
1833         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1834                 rxvq = dev->data->rx_queues[i];
1835                 VIRTQUEUE_DUMP(rxvq->vq);
1836         }
1837
1838         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1839                 txvq = dev->data->tx_queues[i];
1840                 VIRTQUEUE_DUMP(txvq->vq);
1841         }
1842
1843         set_rxtx_funcs(dev);
1844         hw->started = 1;
1845
1846         /* Initialize Link state */
1847         virtio_dev_link_update(dev, 0);
1848
1849         return 0;
1850 }
1851
1852 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1853 {
1854         struct rte_mbuf *buf;
1855         int i, mbuf_num = 0;
1856
1857         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1858                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1859
1860                 PMD_INIT_LOG(DEBUG,
1861                              "Before freeing rxq[%d] used and unused buf", i);
1862                 VIRTQUEUE_DUMP(rxvq->vq);
1863
1864                 PMD_INIT_LOG(DEBUG, "rx_queues[%d]=%p", i, rxvq);
1865                 while ((buf = virtqueue_detatch_unused(rxvq->vq)) != NULL) {
1866                         rte_pktmbuf_free(buf);
1867                         mbuf_num++;
1868                 }
1869
1870                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1871                 PMD_INIT_LOG(DEBUG,
1872                              "After freeing rxq[%d] used and unused buf", i);
1873                 VIRTQUEUE_DUMP(rxvq->vq);
1874         }
1875
1876         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1877                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1878
1879                 PMD_INIT_LOG(DEBUG,
1880                              "Before freeing txq[%d] used and unused bufs",
1881                              i);
1882                 VIRTQUEUE_DUMP(txvq->vq);
1883
1884                 mbuf_num = 0;
1885                 while ((buf = virtqueue_detatch_unused(txvq->vq)) != NULL) {
1886                         rte_pktmbuf_free(buf);
1887                         mbuf_num++;
1888                 }
1889
1890                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1891                 PMD_INIT_LOG(DEBUG,
1892                              "After freeing txq[%d] used and unused buf", i);
1893                 VIRTQUEUE_DUMP(txvq->vq);
1894         }
1895 }
1896
1897 /*
1898  * Stop device: disable interrupt and mark link down
1899  */
1900 static void
1901 virtio_dev_stop(struct rte_eth_dev *dev)
1902 {
1903         struct virtio_hw *hw = dev->data->dev_private;
1904         struct rte_eth_link link;
1905         struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
1906
1907         PMD_INIT_LOG(DEBUG, "stop");
1908
1909         if (intr_conf->lsc || intr_conf->rxq)
1910                 rte_intr_disable(dev->intr_handle);
1911
1912         hw->started = 0;
1913         memset(&link, 0, sizeof(link));
1914         virtio_dev_atomic_write_link_status(dev, &link);
1915 }
1916
1917 static int
1918 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1919 {
1920         struct rte_eth_link link, old;
1921         uint16_t status;
1922         struct virtio_hw *hw = dev->data->dev_private;
1923         memset(&link, 0, sizeof(link));
1924         virtio_dev_atomic_read_link_status(dev, &link);
1925         old = link;
1926         link.link_duplex = ETH_LINK_FULL_DUPLEX;
1927         link.link_speed  = SPEED_10G;
1928
1929         if (hw->started == 0) {
1930                 link.link_status = ETH_LINK_DOWN;
1931         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1932                 PMD_INIT_LOG(DEBUG, "Get link status from hw");
1933                 vtpci_read_dev_config(hw,
1934                                 offsetof(struct virtio_net_config, status),
1935                                 &status, sizeof(status));
1936                 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1937                         link.link_status = ETH_LINK_DOWN;
1938                         PMD_INIT_LOG(DEBUG, "Port %d is down",
1939                                      dev->data->port_id);
1940                 } else {
1941                         link.link_status = ETH_LINK_UP;
1942                         PMD_INIT_LOG(DEBUG, "Port %d is up",
1943                                      dev->data->port_id);
1944                 }
1945         } else {
1946                 link.link_status = ETH_LINK_UP;
1947         }
1948         virtio_dev_atomic_write_link_status(dev, &link);
1949
1950         return (old.link_status == link.link_status) ? -1 : 0;
1951 }
1952
1953 static void
1954 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1955 {
1956         uint64_t tso_mask, host_features;
1957         struct virtio_hw *hw = dev->data->dev_private;
1958
1959         dev_info->speed_capa = ETH_LINK_SPEED_10G; /* fake value */
1960
1961         dev_info->pci_dev = dev->device ? RTE_ETH_DEV_TO_PCI(dev) : NULL;
1962         dev_info->max_rx_queues =
1963                 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
1964         dev_info->max_tx_queues =
1965                 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
1966         dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1967         dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1968         dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1969         dev_info->default_txconf = (struct rte_eth_txconf) {
1970                 .txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
1971         };
1972
1973         host_features = VTPCI_OPS(hw)->get_features(hw);
1974         dev_info->rx_offload_capa = 0;
1975         if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
1976                 dev_info->rx_offload_capa |=
1977                         DEV_RX_OFFLOAD_TCP_CKSUM |
1978                         DEV_RX_OFFLOAD_UDP_CKSUM;
1979         }
1980         tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
1981                 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
1982         if ((host_features & tso_mask) == tso_mask)
1983                 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
1984
1985         dev_info->tx_offload_capa = 0;
1986         if (hw->guest_features & (1ULL << VIRTIO_NET_F_CSUM)) {
1987                 dev_info->tx_offload_capa |=
1988                         DEV_TX_OFFLOAD_UDP_CKSUM |
1989                         DEV_TX_OFFLOAD_TCP_CKSUM;
1990         }
1991         tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
1992                 (1ULL << VIRTIO_NET_F_HOST_TSO6);
1993         if ((hw->guest_features & tso_mask) == tso_mask)
1994                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
1995 }
1996
1997 /*
1998  * It enables testpmd to collect per queue stats.
1999  */
2000 static int
2001 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2002 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2003 __rte_unused uint8_t is_rx)
2004 {
2005         return 0;
2006 }
2007
2008 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
2009 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
2010 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");