42c2836713f9c88174d3b4d632199430bf840f71
[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         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 void
968 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
969 {
970         virtio_update_stats(dev, stats);
971 }
972
973 static void
974 virtio_dev_stats_reset(struct rte_eth_dev *dev)
975 {
976         unsigned int i;
977
978         for (i = 0; i < dev->data->nb_tx_queues; i++) {
979                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
980                 if (txvq == NULL)
981                         continue;
982
983                 txvq->stats.packets = 0;
984                 txvq->stats.bytes = 0;
985                 txvq->stats.errors = 0;
986                 txvq->stats.multicast = 0;
987                 txvq->stats.broadcast = 0;
988                 memset(txvq->stats.size_bins, 0,
989                        sizeof(txvq->stats.size_bins[0]) * 8);
990         }
991
992         for (i = 0; i < dev->data->nb_rx_queues; i++) {
993                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
994                 if (rxvq == NULL)
995                         continue;
996
997                 rxvq->stats.packets = 0;
998                 rxvq->stats.bytes = 0;
999                 rxvq->stats.errors = 0;
1000                 rxvq->stats.multicast = 0;
1001                 rxvq->stats.broadcast = 0;
1002                 memset(rxvq->stats.size_bins, 0,
1003                        sizeof(rxvq->stats.size_bins[0]) * 8);
1004         }
1005 }
1006
1007 static void
1008 virtio_set_hwaddr(struct virtio_hw *hw)
1009 {
1010         vtpci_write_dev_config(hw,
1011                         offsetof(struct virtio_net_config, mac),
1012                         &hw->mac_addr, ETHER_ADDR_LEN);
1013 }
1014
1015 static void
1016 virtio_get_hwaddr(struct virtio_hw *hw)
1017 {
1018         if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1019                 vtpci_read_dev_config(hw,
1020                         offsetof(struct virtio_net_config, mac),
1021                         &hw->mac_addr, ETHER_ADDR_LEN);
1022         } else {
1023                 eth_random_addr(&hw->mac_addr[0]);
1024                 virtio_set_hwaddr(hw);
1025         }
1026 }
1027
1028 static int
1029 virtio_mac_table_set(struct virtio_hw *hw,
1030                      const struct virtio_net_ctrl_mac *uc,
1031                      const struct virtio_net_ctrl_mac *mc)
1032 {
1033         struct virtio_pmd_ctrl ctrl;
1034         int err, len[2];
1035
1036         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1037                 PMD_DRV_LOG(INFO, "host does not support mac table");
1038                 return -1;
1039         }
1040
1041         ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1042         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1043
1044         len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
1045         memcpy(ctrl.data, uc, len[0]);
1046
1047         len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
1048         memcpy(ctrl.data + len[0], mc, len[1]);
1049
1050         err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1051         if (err != 0)
1052                 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1053         return err;
1054 }
1055
1056 static int
1057 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
1058                     uint32_t index, uint32_t vmdq __rte_unused)
1059 {
1060         struct virtio_hw *hw = dev->data->dev_private;
1061         const struct ether_addr *addrs = dev->data->mac_addrs;
1062         unsigned int i;
1063         struct virtio_net_ctrl_mac *uc, *mc;
1064
1065         if (index >= VIRTIO_MAX_MAC_ADDRS) {
1066                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1067                 return -EINVAL;
1068         }
1069
1070         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1071         uc->entries = 0;
1072         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1073         mc->entries = 0;
1074
1075         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1076                 const struct ether_addr *addr
1077                         = (i == index) ? mac_addr : addrs + i;
1078                 struct virtio_net_ctrl_mac *tbl
1079                         = is_multicast_ether_addr(addr) ? mc : uc;
1080
1081                 memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
1082         }
1083
1084         return virtio_mac_table_set(hw, uc, mc);
1085 }
1086
1087 static void
1088 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1089 {
1090         struct virtio_hw *hw = dev->data->dev_private;
1091         struct ether_addr *addrs = dev->data->mac_addrs;
1092         struct virtio_net_ctrl_mac *uc, *mc;
1093         unsigned int i;
1094
1095         if (index >= VIRTIO_MAX_MAC_ADDRS) {
1096                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1097                 return;
1098         }
1099
1100         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1101         uc->entries = 0;
1102         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1103         mc->entries = 0;
1104
1105         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1106                 struct virtio_net_ctrl_mac *tbl;
1107
1108                 if (i == index || is_zero_ether_addr(addrs + i))
1109                         continue;
1110
1111                 tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
1112                 memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
1113         }
1114
1115         virtio_mac_table_set(hw, uc, mc);
1116 }
1117
1118 static void
1119 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1120 {
1121         struct virtio_hw *hw = dev->data->dev_private;
1122
1123         memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
1124
1125         /* Use atomic update if available */
1126         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1127                 struct virtio_pmd_ctrl ctrl;
1128                 int len = ETHER_ADDR_LEN;
1129
1130                 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1131                 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1132
1133                 memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
1134                 virtio_send_command(hw->cvq, &ctrl, &len, 1);
1135         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1136                 virtio_set_hwaddr(hw);
1137 }
1138
1139 static int
1140 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1141 {
1142         struct virtio_hw *hw = dev->data->dev_private;
1143         struct virtio_pmd_ctrl ctrl;
1144         int len;
1145
1146         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1147                 return -ENOTSUP;
1148
1149         ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1150         ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1151         memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1152         len = sizeof(vlan_id);
1153
1154         return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1155 }
1156
1157 static int
1158 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1159 {
1160         uint64_t host_features;
1161
1162         /* Prepare guest_features: feature that driver wants to support */
1163         PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1164                 req_features);
1165
1166         /* Read device(host) feature bits */
1167         host_features = VTPCI_OPS(hw)->get_features(hw);
1168         PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1169                 host_features);
1170
1171         /* If supported, ensure MTU value is valid before acknowledging it. */
1172         if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1173                 struct virtio_net_config config;
1174
1175                 vtpci_read_dev_config(hw,
1176                         offsetof(struct virtio_net_config, mtu),
1177                         &config.mtu, sizeof(config.mtu));
1178
1179                 if (config.mtu < ETHER_MIN_MTU)
1180                         req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1181         }
1182
1183         /*
1184          * Negotiate features: Subset of device feature bits are written back
1185          * guest feature bits.
1186          */
1187         hw->guest_features = req_features;
1188         hw->guest_features = vtpci_negotiate_features(hw, host_features);
1189         PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1190                 hw->guest_features);
1191
1192         if (hw->modern) {
1193                 if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1194                         PMD_INIT_LOG(ERR,
1195                                 "VIRTIO_F_VERSION_1 features is not enabled.");
1196                         return -1;
1197                 }
1198                 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1199                 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1200                         PMD_INIT_LOG(ERR,
1201                                 "failed to set FEATURES_OK status!");
1202                         return -1;
1203                 }
1204         }
1205
1206         hw->req_guest_features = req_features;
1207
1208         return 0;
1209 }
1210
1211 /*
1212  * Process Virtio Config changed interrupt and call the callback
1213  * if link state changed.
1214  */
1215 void
1216 virtio_interrupt_handler(void *param)
1217 {
1218         struct rte_eth_dev *dev = param;
1219         struct virtio_hw *hw = dev->data->dev_private;
1220         uint8_t isr;
1221
1222         /* Read interrupt status which clears interrupt */
1223         isr = vtpci_isr(hw);
1224         PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1225
1226         if (rte_intr_enable(dev->intr_handle) < 0)
1227                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1228
1229         if (isr & VIRTIO_PCI_ISR_CONFIG) {
1230                 if (virtio_dev_link_update(dev, 0) == 0)
1231                         _rte_eth_dev_callback_process(dev,
1232                                                       RTE_ETH_EVENT_INTR_LSC,
1233                                                       NULL, NULL);
1234         }
1235
1236 }
1237
1238 /* set rx and tx handlers according to what is supported */
1239 static void
1240 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1241 {
1242         struct virtio_hw *hw = eth_dev->data->dev_private;
1243
1244         if (hw->use_simple_rx) {
1245                 PMD_INIT_LOG(INFO, "virtio: using simple Rx path on port %u",
1246                         eth_dev->data->port_id);
1247                 eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1248         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1249                 PMD_INIT_LOG(INFO,
1250                         "virtio: using mergeable buffer Rx path on port %u",
1251                         eth_dev->data->port_id);
1252                 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1253         } else {
1254                 PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1255                         eth_dev->data->port_id);
1256                 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1257         }
1258
1259         if (hw->use_simple_tx) {
1260                 PMD_INIT_LOG(INFO, "virtio: using simple Tx path on port %u",
1261                         eth_dev->data->port_id);
1262                 eth_dev->tx_pkt_burst = virtio_xmit_pkts_simple;
1263         } else {
1264                 PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1265                         eth_dev->data->port_id);
1266                 eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1267         }
1268 }
1269
1270 /* Only support 1:1 queue/interrupt mapping so far.
1271  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1272  * interrupt vectors (<N+1).
1273  */
1274 static int
1275 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1276 {
1277         uint32_t i;
1278         struct virtio_hw *hw = dev->data->dev_private;
1279
1280         PMD_INIT_LOG(INFO, "queue/interrupt binding");
1281         for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1282                 dev->intr_handle->intr_vec[i] = i + 1;
1283                 if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1284                                                  VIRTIO_MSI_NO_VECTOR) {
1285                         PMD_DRV_LOG(ERR, "failed to set queue vector");
1286                         return -EBUSY;
1287                 }
1288         }
1289
1290         return 0;
1291 }
1292
1293 static void
1294 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1295 {
1296         uint32_t i;
1297         struct virtio_hw *hw = dev->data->dev_private;
1298
1299         PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1300         for (i = 0; i < dev->data->nb_rx_queues; ++i)
1301                 VTPCI_OPS(hw)->set_queue_irq(hw,
1302                                              hw->vqs[i * VTNET_CQ],
1303                                              VIRTIO_MSI_NO_VECTOR);
1304 }
1305
1306 static int
1307 virtio_configure_intr(struct rte_eth_dev *dev)
1308 {
1309         struct virtio_hw *hw = dev->data->dev_private;
1310
1311         if (!rte_intr_cap_multiple(dev->intr_handle)) {
1312                 PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1313                 return -ENOTSUP;
1314         }
1315
1316         if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1317                 PMD_INIT_LOG(ERR, "Fail to create eventfd");
1318                 return -1;
1319         }
1320
1321         if (!dev->intr_handle->intr_vec) {
1322                 dev->intr_handle->intr_vec =
1323                         rte_zmalloc("intr_vec",
1324                                     hw->max_queue_pairs * sizeof(int), 0);
1325                 if (!dev->intr_handle->intr_vec) {
1326                         PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1327                                      hw->max_queue_pairs);
1328                         return -ENOMEM;
1329                 }
1330         }
1331
1332         /* Re-register callback to update max_intr */
1333         rte_intr_callback_unregister(dev->intr_handle,
1334                                      virtio_interrupt_handler,
1335                                      dev);
1336         rte_intr_callback_register(dev->intr_handle,
1337                                    virtio_interrupt_handler,
1338                                    dev);
1339
1340         /* DO NOT try to remove this! This function will enable msix, or QEMU
1341          * will encounter SIGSEGV when DRIVER_OK is sent.
1342          * And for legacy devices, this should be done before queue/vec binding
1343          * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1344          * (22) will be ignored.
1345          */
1346         if (rte_intr_enable(dev->intr_handle) < 0) {
1347                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1348                 return -1;
1349         }
1350
1351         if (virtio_queues_bind_intr(dev) < 0) {
1352                 PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1353                 return -1;
1354         }
1355
1356         return 0;
1357 }
1358
1359 /* reset device and renegotiate features if needed */
1360 static int
1361 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1362 {
1363         struct virtio_hw *hw = eth_dev->data->dev_private;
1364         struct virtio_net_config *config;
1365         struct virtio_net_config local_config;
1366         struct rte_pci_device *pci_dev = NULL;
1367         int ret;
1368
1369         /* Reset the device although not necessary at startup */
1370         vtpci_reset(hw);
1371
1372         /* Tell the host we've noticed this device. */
1373         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1374
1375         /* Tell the host we've known how to drive the device. */
1376         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1377         if (virtio_negotiate_features(hw, req_features) < 0)
1378                 return -1;
1379
1380         if (!hw->virtio_user_dev) {
1381                 pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1382                 rte_eth_copy_pci_info(eth_dev, pci_dev);
1383         }
1384
1385         eth_dev->data->dev_flags = RTE_ETH_DEV_DETACHABLE;
1386         /* If host does not support both status and MSI-X then disable LSC */
1387         if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) && hw->use_msix)
1388                 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1389         else
1390                 eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1391
1392         /* Setting up rx_header size for the device */
1393         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1394             vtpci_with_feature(hw, VIRTIO_F_VERSION_1))
1395                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1396         else
1397                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1398
1399         /* Copy the permanent MAC address to: virtio_hw */
1400         virtio_get_hwaddr(hw);
1401         ether_addr_copy((struct ether_addr *) hw->mac_addr,
1402                         &eth_dev->data->mac_addrs[0]);
1403         PMD_INIT_LOG(DEBUG,
1404                      "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1405                      hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1406                      hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1407
1408         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1409                 config = &local_config;
1410
1411                 vtpci_read_dev_config(hw,
1412                         offsetof(struct virtio_net_config, mac),
1413                         &config->mac, sizeof(config->mac));
1414
1415                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1416                         vtpci_read_dev_config(hw,
1417                                 offsetof(struct virtio_net_config, status),
1418                                 &config->status, sizeof(config->status));
1419                 } else {
1420                         PMD_INIT_LOG(DEBUG,
1421                                      "VIRTIO_NET_F_STATUS is not supported");
1422                         config->status = 0;
1423                 }
1424
1425                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1426                         vtpci_read_dev_config(hw,
1427                                 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1428                                 &config->max_virtqueue_pairs,
1429                                 sizeof(config->max_virtqueue_pairs));
1430                 } else {
1431                         PMD_INIT_LOG(DEBUG,
1432                                      "VIRTIO_NET_F_MQ is not supported");
1433                         config->max_virtqueue_pairs = 1;
1434                 }
1435
1436                 hw->max_queue_pairs = config->max_virtqueue_pairs;
1437
1438                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1439                         vtpci_read_dev_config(hw,
1440                                 offsetof(struct virtio_net_config, mtu),
1441                                 &config->mtu,
1442                                 sizeof(config->mtu));
1443
1444                         /*
1445                          * MTU value has already been checked at negotiation
1446                          * time, but check again in case it has changed since
1447                          * then, which should not happen.
1448                          */
1449                         if (config->mtu < ETHER_MIN_MTU) {
1450                                 PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1451                                                 config->mtu);
1452                                 return -1;
1453                         }
1454
1455                         hw->max_mtu = config->mtu;
1456                         /* Set initial MTU to maximum one supported by vhost */
1457                         eth_dev->data->mtu = config->mtu;
1458
1459                 } else {
1460                         hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - ETHER_HDR_LEN -
1461                                 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1462                 }
1463
1464                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1465                                 config->max_virtqueue_pairs);
1466                 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1467                 PMD_INIT_LOG(DEBUG,
1468                                 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1469                                 config->mac[0], config->mac[1],
1470                                 config->mac[2], config->mac[3],
1471                                 config->mac[4], config->mac[5]);
1472         } else {
1473                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1474                 hw->max_queue_pairs = 1;
1475         }
1476
1477         ret = virtio_alloc_queues(eth_dev);
1478         if (ret < 0)
1479                 return ret;
1480
1481         if (eth_dev->data->dev_conf.intr_conf.rxq) {
1482                 if (virtio_configure_intr(eth_dev) < 0) {
1483                         PMD_INIT_LOG(ERR, "failed to configure interrupt");
1484                         return -1;
1485                 }
1486         }
1487
1488         vtpci_reinit_complete(hw);
1489
1490         if (pci_dev)
1491                 PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1492                         eth_dev->data->port_id, pci_dev->id.vendor_id,
1493                         pci_dev->id.device_id);
1494
1495         return 0;
1496 }
1497
1498 /*
1499  * Remap the PCI device again (IO port map for legacy device and
1500  * memory map for modern device), so that the secondary process
1501  * could have the PCI initiated correctly.
1502  */
1503 static int
1504 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1505 {
1506         if (hw->modern) {
1507                 /*
1508                  * We don't have to re-parse the PCI config space, since
1509                  * rte_pci_map_device() makes sure the mapped address
1510                  * in secondary process would equal to the one mapped in
1511                  * the primary process: error will be returned if that
1512                  * requirement is not met.
1513                  *
1514                  * That said, we could simply reuse all cap pointers
1515                  * (such as dev_cfg, common_cfg, etc.) parsed from the
1516                  * primary process, which is stored in shared memory.
1517                  */
1518                 if (rte_pci_map_device(pci_dev)) {
1519                         PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1520                         return -1;
1521                 }
1522         } else {
1523                 if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1524                         return -1;
1525         }
1526
1527         return 0;
1528 }
1529
1530 static void
1531 virtio_set_vtpci_ops(struct virtio_hw *hw)
1532 {
1533 #ifdef RTE_VIRTIO_USER
1534         if (hw->virtio_user_dev)
1535                 VTPCI_OPS(hw) = &virtio_user_ops;
1536         else
1537 #endif
1538         if (hw->modern)
1539                 VTPCI_OPS(hw) = &modern_ops;
1540         else
1541                 VTPCI_OPS(hw) = &legacy_ops;
1542 }
1543
1544 /*
1545  * This function is based on probe() function in virtio_pci.c
1546  * It returns 0 on success.
1547  */
1548 int
1549 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1550 {
1551         struct virtio_hw *hw = eth_dev->data->dev_private;
1552         int ret;
1553
1554         RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1555
1556         eth_dev->dev_ops = &virtio_eth_dev_ops;
1557
1558         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1559                 if (!hw->virtio_user_dev) {
1560                         ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1561                         if (ret)
1562                                 return ret;
1563                 }
1564
1565                 virtio_set_vtpci_ops(hw);
1566                 set_rxtx_funcs(eth_dev);
1567
1568                 return 0;
1569         }
1570
1571         /* Allocate memory for storing MAC addresses */
1572         eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
1573         if (eth_dev->data->mac_addrs == NULL) {
1574                 PMD_INIT_LOG(ERR,
1575                         "Failed to allocate %d bytes needed to store MAC addresses",
1576                         VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
1577                 return -ENOMEM;
1578         }
1579
1580         hw->port_id = eth_dev->data->port_id;
1581         /* For virtio_user case the hw->virtio_user_dev is populated by
1582          * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1583          */
1584         if (!hw->virtio_user_dev) {
1585                 ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1586                 if (ret)
1587                         return ret;
1588         }
1589
1590         /* reset device and negotiate default features */
1591         ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1592         if (ret < 0)
1593                 return ret;
1594
1595         /* Setup interrupt callback  */
1596         if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1597                 rte_intr_callback_register(eth_dev->intr_handle,
1598                         virtio_interrupt_handler, eth_dev);
1599
1600         return 0;
1601 }
1602
1603 static int
1604 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1605 {
1606         PMD_INIT_FUNC_TRACE();
1607
1608         if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1609                 return -EPERM;
1610
1611         virtio_dev_stop(eth_dev);
1612         virtio_dev_close(eth_dev);
1613
1614         eth_dev->dev_ops = NULL;
1615         eth_dev->tx_pkt_burst = NULL;
1616         eth_dev->rx_pkt_burst = NULL;
1617
1618         rte_free(eth_dev->data->mac_addrs);
1619         eth_dev->data->mac_addrs = NULL;
1620
1621         /* reset interrupt callback  */
1622         if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1623                 rte_intr_callback_unregister(eth_dev->intr_handle,
1624                                                 virtio_interrupt_handler,
1625                                                 eth_dev);
1626         if (eth_dev->device)
1627                 rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1628
1629         PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1630
1631         return 0;
1632 }
1633
1634 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1635         struct rte_pci_device *pci_dev)
1636 {
1637         return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
1638                 eth_virtio_dev_init);
1639 }
1640
1641 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
1642 {
1643         return rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
1644 }
1645
1646 static struct rte_pci_driver rte_virtio_pmd = {
1647         .driver = {
1648                 .name = "net_virtio",
1649         },
1650         .id_table = pci_id_virtio_map,
1651         .drv_flags = 0,
1652         .probe = eth_virtio_pci_probe,
1653         .remove = eth_virtio_pci_remove,
1654 };
1655
1656 RTE_INIT(rte_virtio_pmd_init);
1657 static void
1658 rte_virtio_pmd_init(void)
1659 {
1660         if (rte_eal_iopl_init() != 0) {
1661                 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1662                 return;
1663         }
1664
1665         rte_pci_register(&rte_virtio_pmd);
1666 }
1667
1668 /*
1669  * Configure virtio device
1670  * It returns 0 on success.
1671  */
1672 static int
1673 virtio_dev_configure(struct rte_eth_dev *dev)
1674 {
1675         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1676         struct virtio_hw *hw = dev->data->dev_private;
1677         uint64_t req_features;
1678         int ret;
1679
1680         PMD_INIT_LOG(DEBUG, "configure");
1681         req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
1682
1683         if (dev->data->dev_conf.intr_conf.rxq) {
1684                 ret = virtio_init_device(dev, hw->req_guest_features);
1685                 if (ret < 0)
1686                         return ret;
1687         }
1688
1689         /* The name hw_ip_checksum is a bit confusing since it can be
1690          * set by the application to request L3 and/or L4 checksums. In
1691          * case of virtio, only L4 checksum is supported.
1692          */
1693         if (rxmode->hw_ip_checksum)
1694                 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
1695
1696         if (rxmode->enable_lro)
1697                 req_features |=
1698                         (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
1699                         (1ULL << VIRTIO_NET_F_GUEST_TSO6);
1700
1701         /* if request features changed, reinit the device */
1702         if (req_features != hw->req_guest_features) {
1703                 ret = virtio_init_device(dev, req_features);
1704                 if (ret < 0)
1705                         return ret;
1706         }
1707
1708         if (rxmode->hw_ip_checksum &&
1709                 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
1710                 PMD_DRV_LOG(ERR,
1711                         "rx checksum not available on this host");
1712                 return -ENOTSUP;
1713         }
1714
1715         if (rxmode->enable_lro &&
1716                 (!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
1717                         !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4))) {
1718                 PMD_DRV_LOG(ERR,
1719                         "Large Receive Offload not available on this host");
1720                 return -ENOTSUP;
1721         }
1722
1723         /* start control queue */
1724         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
1725                 virtio_dev_cq_start(dev);
1726
1727         hw->vlan_strip = rxmode->hw_vlan_strip;
1728
1729         if (rxmode->hw_vlan_filter
1730             && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1731                 PMD_DRV_LOG(ERR,
1732                             "vlan filtering not available on this host");
1733                 return -ENOTSUP;
1734         }
1735
1736         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1737                 /* Enable vector (0) for Link State Intrerrupt */
1738                 if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
1739                                 VIRTIO_MSI_NO_VECTOR) {
1740                         PMD_DRV_LOG(ERR, "failed to set config vector");
1741                         return -EBUSY;
1742                 }
1743
1744         hw->use_simple_rx = 1;
1745         hw->use_simple_tx = 1;
1746
1747 #if defined RTE_ARCH_ARM64 || defined CONFIG_RTE_ARCH_ARM
1748         if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
1749                 hw->use_simple_rx = 0;
1750                 hw->use_simple_tx = 0;
1751         }
1752 #endif
1753         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1754                 hw->use_simple_rx = 0;
1755                 hw->use_simple_tx = 0;
1756         }
1757
1758         if (rxmode->hw_ip_checksum)
1759                 hw->use_simple_rx = 0;
1760
1761         return 0;
1762 }
1763
1764
1765 static int
1766 virtio_dev_start(struct rte_eth_dev *dev)
1767 {
1768         uint16_t nb_queues, i;
1769         struct virtnet_rx *rxvq;
1770         struct virtnet_tx *txvq __rte_unused;
1771         struct virtio_hw *hw = dev->data->dev_private;
1772         int ret;
1773
1774         /* Finish the initialization of the queues */
1775         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1776                 ret = virtio_dev_rx_queue_setup_finish(dev, i);
1777                 if (ret < 0)
1778                         return ret;
1779         }
1780         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1781                 ret = virtio_dev_tx_queue_setup_finish(dev, i);
1782                 if (ret < 0)
1783                         return ret;
1784         }
1785
1786         /* check if lsc interrupt feature is enabled */
1787         if (dev->data->dev_conf.intr_conf.lsc) {
1788                 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1789                         PMD_DRV_LOG(ERR, "link status not supported by host");
1790                         return -ENOTSUP;
1791                 }
1792         }
1793
1794         /* Enable uio/vfio intr/eventfd mapping: althrough we already did that
1795          * in device configure, but it could be unmapped  when device is
1796          * stopped.
1797          */
1798         if (dev->data->dev_conf.intr_conf.lsc ||
1799             dev->data->dev_conf.intr_conf.rxq) {
1800                 rte_intr_disable(dev->intr_handle);
1801
1802                 if (rte_intr_enable(dev->intr_handle) < 0) {
1803                         PMD_DRV_LOG(ERR, "interrupt enable failed");
1804                         return -EIO;
1805                 }
1806         }
1807
1808         /*Notify the backend
1809          *Otherwise the tap backend might already stop its queue due to fullness.
1810          *vhost backend will have no chance to be waked up
1811          */
1812         nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
1813         if (hw->max_queue_pairs > 1) {
1814                 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1815                         return -EINVAL;
1816         }
1817
1818         PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1819
1820         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1821                 rxvq = dev->data->rx_queues[i];
1822                 virtqueue_notify(rxvq->vq);
1823         }
1824
1825         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1826                 txvq = dev->data->tx_queues[i];
1827                 virtqueue_notify(txvq->vq);
1828         }
1829
1830         PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1831
1832         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1833                 rxvq = dev->data->rx_queues[i];
1834                 VIRTQUEUE_DUMP(rxvq->vq);
1835         }
1836
1837         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1838                 txvq = dev->data->tx_queues[i];
1839                 VIRTQUEUE_DUMP(txvq->vq);
1840         }
1841
1842         set_rxtx_funcs(dev);
1843         hw->started = 1;
1844
1845         /* Initialize Link state */
1846         virtio_dev_link_update(dev, 0);
1847
1848         return 0;
1849 }
1850
1851 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1852 {
1853         struct rte_mbuf *buf;
1854         int i, mbuf_num = 0;
1855
1856         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1857                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1858
1859                 PMD_INIT_LOG(DEBUG,
1860                              "Before freeing rxq[%d] used and unused buf", i);
1861                 VIRTQUEUE_DUMP(rxvq->vq);
1862
1863                 PMD_INIT_LOG(DEBUG, "rx_queues[%d]=%p", i, rxvq);
1864                 while ((buf = virtqueue_detatch_unused(rxvq->vq)) != NULL) {
1865                         rte_pktmbuf_free(buf);
1866                         mbuf_num++;
1867                 }
1868
1869                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1870                 PMD_INIT_LOG(DEBUG,
1871                              "After freeing rxq[%d] used and unused buf", i);
1872                 VIRTQUEUE_DUMP(rxvq->vq);
1873         }
1874
1875         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1876                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1877
1878                 PMD_INIT_LOG(DEBUG,
1879                              "Before freeing txq[%d] used and unused bufs",
1880                              i);
1881                 VIRTQUEUE_DUMP(txvq->vq);
1882
1883                 mbuf_num = 0;
1884                 while ((buf = virtqueue_detatch_unused(txvq->vq)) != NULL) {
1885                         rte_pktmbuf_free(buf);
1886                         mbuf_num++;
1887                 }
1888
1889                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1890                 PMD_INIT_LOG(DEBUG,
1891                              "After freeing txq[%d] used and unused buf", i);
1892                 VIRTQUEUE_DUMP(txvq->vq);
1893         }
1894 }
1895
1896 /*
1897  * Stop device: disable interrupt and mark link down
1898  */
1899 static void
1900 virtio_dev_stop(struct rte_eth_dev *dev)
1901 {
1902         struct virtio_hw *hw = dev->data->dev_private;
1903         struct rte_eth_link link;
1904         struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
1905
1906         PMD_INIT_LOG(DEBUG, "stop");
1907
1908         if (intr_conf->lsc || intr_conf->rxq)
1909                 rte_intr_disable(dev->intr_handle);
1910
1911         hw->started = 0;
1912         memset(&link, 0, sizeof(link));
1913         virtio_dev_atomic_write_link_status(dev, &link);
1914 }
1915
1916 static int
1917 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1918 {
1919         struct rte_eth_link link, old;
1920         uint16_t status;
1921         struct virtio_hw *hw = dev->data->dev_private;
1922         memset(&link, 0, sizeof(link));
1923         virtio_dev_atomic_read_link_status(dev, &link);
1924         old = link;
1925         link.link_duplex = ETH_LINK_FULL_DUPLEX;
1926         link.link_speed  = SPEED_10G;
1927
1928         if (hw->started == 0) {
1929                 link.link_status = ETH_LINK_DOWN;
1930         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1931                 PMD_INIT_LOG(DEBUG, "Get link status from hw");
1932                 vtpci_read_dev_config(hw,
1933                                 offsetof(struct virtio_net_config, status),
1934                                 &status, sizeof(status));
1935                 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1936                         link.link_status = ETH_LINK_DOWN;
1937                         PMD_INIT_LOG(DEBUG, "Port %d is down",
1938                                      dev->data->port_id);
1939                 } else {
1940                         link.link_status = ETH_LINK_UP;
1941                         PMD_INIT_LOG(DEBUG, "Port %d is up",
1942                                      dev->data->port_id);
1943                 }
1944         } else {
1945                 link.link_status = ETH_LINK_UP;
1946         }
1947         virtio_dev_atomic_write_link_status(dev, &link);
1948
1949         return (old.link_status == link.link_status) ? -1 : 0;
1950 }
1951
1952 static void
1953 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1954 {
1955         uint64_t tso_mask, host_features;
1956         struct virtio_hw *hw = dev->data->dev_private;
1957
1958         dev_info->speed_capa = ETH_LINK_SPEED_10G; /* fake value */
1959
1960         dev_info->pci_dev = dev->device ? RTE_ETH_DEV_TO_PCI(dev) : NULL;
1961         dev_info->max_rx_queues =
1962                 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
1963         dev_info->max_tx_queues =
1964                 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
1965         dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1966         dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1967         dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1968         dev_info->default_txconf = (struct rte_eth_txconf) {
1969                 .txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
1970         };
1971
1972         host_features = VTPCI_OPS(hw)->get_features(hw);
1973         dev_info->rx_offload_capa = 0;
1974         if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
1975                 dev_info->rx_offload_capa |=
1976                         DEV_RX_OFFLOAD_TCP_CKSUM |
1977                         DEV_RX_OFFLOAD_UDP_CKSUM;
1978         }
1979         tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
1980                 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
1981         if ((host_features & tso_mask) == tso_mask)
1982                 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
1983
1984         dev_info->tx_offload_capa = 0;
1985         if (hw->guest_features & (1ULL << VIRTIO_NET_F_CSUM)) {
1986                 dev_info->tx_offload_capa |=
1987                         DEV_TX_OFFLOAD_UDP_CKSUM |
1988                         DEV_TX_OFFLOAD_TCP_CKSUM;
1989         }
1990         tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
1991                 (1ULL << VIRTIO_NET_F_HOST_TSO6);
1992         if ((hw->guest_features & tso_mask) == tso_mask)
1993                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
1994 }
1995
1996 /*
1997  * It enables testpmd to collect per queue stats.
1998  */
1999 static int
2000 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2001 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2002 __rte_unused uint8_t is_rx)
2003 {
2004         return 0;
2005 }
2006
2007 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
2008 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
2009 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");