4cb27b6df81cfc01a8b2d44b41f6a6b031e41738
[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_memcpy.h>
42 #include <rte_string_fns.h>
43 #include <rte_memzone.h>
44 #include <rte_malloc.h>
45 #include <rte_atomic.h>
46 #include <rte_branch_prediction.h>
47 #include <rte_pci.h>
48 #include <rte_ether.h>
49 #include <rte_common.h>
50 #include <rte_errno.h>
51
52 #include <rte_memory.h>
53 #include <rte_eal.h>
54 #include <rte_dev.h>
55
56 #include "virtio_ethdev.h"
57 #include "virtio_pci.h"
58 #include "virtio_logs.h"
59 #include "virtqueue.h"
60 #include "virtio_rxtx.h"
61
62 static int eth_virtio_dev_init(struct rte_eth_dev *eth_dev);
63 static int eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev);
64 static int  virtio_dev_configure(struct rte_eth_dev *dev);
65 static int  virtio_dev_start(struct rte_eth_dev *dev);
66 static void virtio_dev_stop(struct rte_eth_dev *dev);
67 static void virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
68 static void virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
69 static void virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
70 static void virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
71 static void virtio_dev_info_get(struct rte_eth_dev *dev,
72                                 struct rte_eth_dev_info *dev_info);
73 static int virtio_dev_link_update(struct rte_eth_dev *dev,
74         __rte_unused int wait_to_complete);
75
76 static void virtio_set_hwaddr(struct virtio_hw *hw);
77 static void virtio_get_hwaddr(struct virtio_hw *hw);
78
79 static void virtio_dev_stats_get(struct rte_eth_dev *dev,
80                                  struct rte_eth_stats *stats);
81 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
82                                  struct rte_eth_xstat *xstats, unsigned n);
83 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
84                                        struct rte_eth_xstat_name *xstats_names,
85                                        unsigned limit);
86 static void virtio_dev_stats_reset(struct rte_eth_dev *dev);
87 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
88 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
89                                 uint16_t vlan_id, int on);
90 static void virtio_mac_addr_add(struct rte_eth_dev *dev,
91                                 struct ether_addr *mac_addr,
92                                 uint32_t index, uint32_t vmdq __rte_unused);
93 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
94 static void virtio_mac_addr_set(struct rte_eth_dev *dev,
95                                 struct ether_addr *mac_addr);
96
97 static int virtio_dev_queue_stats_mapping_set(
98         __rte_unused struct rte_eth_dev *eth_dev,
99         __rte_unused uint16_t queue_id,
100         __rte_unused uint8_t stat_idx,
101         __rte_unused uint8_t is_rx);
102
103 /*
104  * The set of PCI devices this driver supports
105  */
106 static const struct rte_pci_id pci_id_virtio_map[] = {
107
108 #define RTE_PCI_DEV_ID_DECL_VIRTIO(vend, dev) {RTE_PCI_DEVICE(vend, dev)},
109 #include "rte_pci_dev_ids.h"
110
111 { .vendor_id = 0, /* sentinel */ },
112 };
113
114 struct rte_virtio_xstats_name_off {
115         char name[RTE_ETH_XSTATS_NAME_SIZE];
116         unsigned offset;
117 };
118
119 /* [rt]x_qX_ is prepended to the name string here */
120 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
121         {"good_packets",           offsetof(struct virtnet_rx, stats.packets)},
122         {"good_bytes",             offsetof(struct virtnet_rx, stats.bytes)},
123         {"errors",                 offsetof(struct virtnet_rx, stats.errors)},
124         {"multicast_packets",      offsetof(struct virtnet_rx, stats.multicast)},
125         {"broadcast_packets",      offsetof(struct virtnet_rx, stats.broadcast)},
126         {"undersize_packets",      offsetof(struct virtnet_rx, stats.size_bins[0])},
127         {"size_64_packets",        offsetof(struct virtnet_rx, stats.size_bins[1])},
128         {"size_65_127_packets",    offsetof(struct virtnet_rx, stats.size_bins[2])},
129         {"size_128_255_packets",   offsetof(struct virtnet_rx, stats.size_bins[3])},
130         {"size_256_511_packets",   offsetof(struct virtnet_rx, stats.size_bins[4])},
131         {"size_512_1023_packets",  offsetof(struct virtnet_rx, stats.size_bins[5])},
132         {"size_1024_1517_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
133         {"size_1518_max_packets",  offsetof(struct virtnet_rx, stats.size_bins[7])},
134 };
135
136 /* [rt]x_qX_ is prepended to the name string here */
137 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
138         {"good_packets",           offsetof(struct virtnet_tx, stats.packets)},
139         {"good_bytes",             offsetof(struct virtnet_tx, stats.bytes)},
140         {"errors",                 offsetof(struct virtnet_tx, stats.errors)},
141         {"multicast_packets",      offsetof(struct virtnet_tx, stats.multicast)},
142         {"broadcast_packets",      offsetof(struct virtnet_tx, stats.broadcast)},
143         {"undersize_packets",      offsetof(struct virtnet_tx, stats.size_bins[0])},
144         {"size_64_packets",        offsetof(struct virtnet_tx, stats.size_bins[1])},
145         {"size_65_127_packets",    offsetof(struct virtnet_tx, stats.size_bins[2])},
146         {"size_128_255_packets",   offsetof(struct virtnet_tx, stats.size_bins[3])},
147         {"size_256_511_packets",   offsetof(struct virtnet_tx, stats.size_bins[4])},
148         {"size_512_1023_packets",  offsetof(struct virtnet_tx, stats.size_bins[5])},
149         {"size_1024_1517_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
150         {"size_1518_max_packets",  offsetof(struct virtnet_tx, stats.size_bins[7])},
151 };
152
153 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
154                             sizeof(rte_virtio_rxq_stat_strings[0]))
155 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
156                             sizeof(rte_virtio_txq_stat_strings[0]))
157
158 static int
159 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
160                 int *dlen, int pkt_num)
161 {
162         uint32_t head, i;
163         int k, sum = 0;
164         virtio_net_ctrl_ack status = ~0;
165         struct virtio_pmd_ctrl result;
166         struct virtqueue *vq;
167
168         ctrl->status = status;
169
170         if (!cvq && !cvq->vq) {
171                 PMD_INIT_LOG(ERR, "Control queue is not supported.");
172                 return -1;
173         }
174         vq = cvq->vq;
175         head = vq->vq_desc_head_idx;
176
177         PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
178                 "vq->hw->cvq = %p vq = %p",
179                 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
180
181         if ((vq->vq_free_cnt < ((uint32_t)pkt_num + 2)) || (pkt_num < 1))
182                 return -1;
183
184         memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
185                 sizeof(struct virtio_pmd_ctrl));
186
187         /*
188          * Format is enforced in qemu code:
189          * One TX packet for header;
190          * At least one TX packet per argument;
191          * One RX packet for ACK.
192          */
193         vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
194         vq->vq_ring.desc[head].addr = cvq->virtio_net_hdr_mz->phys_addr;
195         vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
196         vq->vq_free_cnt--;
197         i = vq->vq_ring.desc[head].next;
198
199         for (k = 0; k < pkt_num; k++) {
200                 vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
201                 vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mz->phys_addr
202                         + sizeof(struct virtio_net_ctrl_hdr)
203                         + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
204                 vq->vq_ring.desc[i].len = dlen[k];
205                 sum += dlen[k];
206                 vq->vq_free_cnt--;
207                 i = vq->vq_ring.desc[i].next;
208         }
209
210         vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
211         vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mz->phys_addr
212                         + sizeof(struct virtio_net_ctrl_hdr);
213         vq->vq_ring.desc[i].len = sizeof(ctrl->status);
214         vq->vq_free_cnt--;
215
216         vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
217
218         vq_update_avail_ring(vq, head);
219         vq_update_avail_idx(vq);
220
221         PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
222
223         virtqueue_notify(vq);
224
225         rte_rmb();
226         while (vq->vq_used_cons_idx == vq->vq_ring.used->idx) {
227                 rte_rmb();
228                 usleep(100);
229         }
230
231         while (vq->vq_used_cons_idx != vq->vq_ring.used->idx) {
232                 uint32_t idx, desc_idx, used_idx;
233                 struct vring_used_elem *uep;
234
235                 used_idx = (uint32_t)(vq->vq_used_cons_idx
236                                 & (vq->vq_nentries - 1));
237                 uep = &vq->vq_ring.used->ring[used_idx];
238                 idx = (uint32_t) uep->id;
239                 desc_idx = idx;
240
241                 while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
242                         desc_idx = vq->vq_ring.desc[desc_idx].next;
243                         vq->vq_free_cnt++;
244                 }
245
246                 vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
247                 vq->vq_desc_head_idx = idx;
248
249                 vq->vq_used_cons_idx++;
250                 vq->vq_free_cnt++;
251         }
252
253         PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
254                         vq->vq_free_cnt, vq->vq_desc_head_idx);
255
256         memcpy(&result, cvq->virtio_net_hdr_mz->addr,
257                         sizeof(struct virtio_pmd_ctrl));
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 void
287 virtio_dev_queue_release(struct virtqueue *vq)
288 {
289         struct virtio_hw *hw;
290
291         if (vq) {
292                 hw = vq->hw;
293                 if (vq->configured)
294                         hw->vtpci_ops->del_queue(hw, vq);
295
296                 rte_free(vq->sw_ring);
297                 rte_free(vq);
298         }
299 }
300
301 int virtio_dev_queue_setup(struct rte_eth_dev *dev,
302                         int queue_type,
303                         uint16_t queue_idx,
304                         uint16_t vtpci_queue_idx,
305                         uint16_t nb_desc,
306                         unsigned int socket_id,
307                         void **pvq)
308 {
309         char vq_name[VIRTQUEUE_MAX_NAME_SZ];
310         char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
311         const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
312         unsigned int vq_size, size;
313         struct virtio_hw *hw = dev->data->dev_private;
314         struct virtnet_rx *rxvq;
315         struct virtnet_tx *txvq;
316         struct virtnet_ctl *cvq;
317         struct virtqueue *vq;
318         const char *queue_names[] = {"rvq", "txq", "cvq"};
319         size_t sz_vq, sz_q = 0, sz_hdr_mz = 0;
320         void *sw_ring = NULL;
321         int ret;
322
323         PMD_INIT_LOG(DEBUG, "setting up queue: %u", vtpci_queue_idx);
324
325         /*
326          * Read the virtqueue size from the Queue Size field
327          * Always power of 2 and if 0 virtqueue does not exist
328          */
329         vq_size = hw->vtpci_ops->get_queue_num(hw, vtpci_queue_idx);
330         PMD_INIT_LOG(DEBUG, "vq_size: %u nb_desc:%u", vq_size, nb_desc);
331         if (vq_size == 0) {
332                 PMD_INIT_LOG(ERR, "virtqueue does not exist");
333                 return -EINVAL;
334         }
335
336         if (!rte_is_power_of_2(vq_size)) {
337                 PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
338                 return -EINVAL;
339         }
340
341         snprintf(vq_name, sizeof(vq_name), "port%d_%s%d",
342                  dev->data->port_id, queue_names[queue_type], queue_idx);
343
344         sz_vq = RTE_ALIGN_CEIL(sizeof(*vq) +
345                                 vq_size * sizeof(struct vq_desc_extra),
346                                 RTE_CACHE_LINE_SIZE);
347         if (queue_type == VTNET_RQ) {
348                 sz_q = sz_vq + sizeof(*rxvq);
349         } else if (queue_type == VTNET_TQ) {
350                 sz_q = sz_vq + sizeof(*txvq);
351                 /*
352                  * For each xmit packet, allocate a virtio_net_hdr
353                  * and indirect ring elements
354                  */
355                 sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
356         } else if (queue_type == VTNET_CQ) {
357                 sz_q = sz_vq + sizeof(*cvq);
358                 /* Allocate a page for control vq command, data and status */
359                 sz_hdr_mz = PAGE_SIZE;
360         }
361
362         vq = rte_zmalloc_socket(vq_name, sz_q, RTE_CACHE_LINE_SIZE, socket_id);
363         if (vq == NULL) {
364                 PMD_INIT_LOG(ERR, "can not allocate vq");
365                 return -ENOMEM;
366         }
367         vq->hw = hw;
368         vq->vq_queue_index = vtpci_queue_idx;
369         vq->vq_nentries = vq_size;
370
371         if (nb_desc == 0 || nb_desc > vq_size)
372                 nb_desc = vq_size;
373         vq->vq_free_cnt = nb_desc;
374
375         /*
376          * Reserve a memzone for vring elements
377          */
378         size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
379         vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
380         PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
381                      size, vq->vq_ring_size);
382
383         mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size, socket_id,
384                                          0, VIRTIO_PCI_VRING_ALIGN);
385         if (mz == NULL) {
386                 if (rte_errno == EEXIST)
387                         mz = rte_memzone_lookup(vq_name);
388                 if (mz == NULL) {
389                         ret = -ENOMEM;
390                         goto fail_q_alloc;
391                 }
392         }
393
394         /*
395          * Virtio PCI device VIRTIO_PCI_QUEUE_PF register is 32bit,
396          * and only accepts 32 bit page frame number.
397          * Check if the allocated physical memory exceeds 16TB.
398          */
399         if ((mz->phys_addr + vq->vq_ring_size - 1) >> (VIRTIO_PCI_QUEUE_ADDR_SHIFT + 32)) {
400                 PMD_INIT_LOG(ERR, "vring address shouldn't be above 16TB!");
401                 ret = -ENOMEM;
402                 goto fail_q_alloc;
403         }
404         memset(mz->addr, 0, sizeof(mz->len));
405
406         vq->vq_ring_mem = mz->phys_addr;
407         vq->vq_ring_virt_mem = mz->addr;
408         PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%" PRIx64,
409                      (uint64_t)mz->phys_addr);
410         PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
411                      (uint64_t)(uintptr_t)mz->addr);
412
413         if (sz_hdr_mz) {
414                 snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_%s%d_hdr",
415                          dev->data->port_id, queue_names[queue_type],
416                          queue_idx);
417                 hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
418                                                      socket_id, 0,
419                                                      RTE_CACHE_LINE_SIZE);
420                 if (hdr_mz == NULL) {
421                         if (rte_errno == EEXIST)
422                                 hdr_mz = rte_memzone_lookup(vq_hdr_name);
423                         if (hdr_mz == NULL) {
424                                 ret = -ENOMEM;
425                                 goto fail_q_alloc;
426                         }
427                 }
428         }
429
430         if (queue_type == VTNET_RQ) {
431                 size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
432                                sizeof(vq->sw_ring[0]);
433
434                 sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
435                                              RTE_CACHE_LINE_SIZE, socket_id);
436                 if (!sw_ring) {
437                         PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
438                         ret = -ENOMEM;
439                         goto fail_q_alloc;
440                 }
441
442                 vq->sw_ring = sw_ring;
443                 rxvq = (struct virtnet_rx *)RTE_PTR_ADD(vq, sz_vq);
444                 rxvq->vq = vq;
445                 rxvq->port_id = dev->data->port_id;
446                 rxvq->queue_id = queue_idx;
447                 rxvq->mz = mz;
448                 *pvq = rxvq;
449         } else if (queue_type == VTNET_TQ) {
450                 struct virtio_tx_region *txr;
451                 unsigned int i;
452
453                 txvq = (struct virtnet_tx *)RTE_PTR_ADD(vq, sz_vq);
454                 txvq->vq = vq;
455                 txvq->port_id = dev->data->port_id;
456                 txvq->queue_id = queue_idx;
457                 txvq->mz = mz;
458                 txvq->virtio_net_hdr_mz = hdr_mz;
459                 txvq->virtio_net_hdr_mem = hdr_mz->phys_addr;
460
461                 txr = hdr_mz->addr;
462                 memset(txr, 0, vq_size * sizeof(*txr));
463                 for (i = 0; i < vq_size; i++) {
464                         struct vring_desc *start_dp = txr[i].tx_indir;
465
466                         vring_desc_init(start_dp, RTE_DIM(txr[i].tx_indir));
467
468                         /* first indirect descriptor is always the tx header */
469                         start_dp->addr = txvq->virtio_net_hdr_mem
470                                 + i * sizeof(*txr)
471                                 + offsetof(struct virtio_tx_region, tx_hdr);
472
473                         start_dp->len = hw->vtnet_hdr_size;
474                         start_dp->flags = VRING_DESC_F_NEXT;
475                 }
476
477                 *pvq = txvq;
478         } else if (queue_type == VTNET_CQ) {
479                 cvq = (struct virtnet_ctl *)RTE_PTR_ADD(vq, sz_vq);
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                 *pvq = cvq;
486         }
487
488         hw->vtpci_ops->setup_queue(hw, vq);
489         vq->configured = 1;
490         return 0;
491
492 fail_q_alloc:
493         rte_free(sw_ring);
494         rte_memzone_free(hdr_mz);
495         rte_memzone_free(mz);
496         rte_free(vq);
497
498         return ret;
499 }
500
501 static int
502 virtio_dev_cq_queue_setup(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx,
503                 uint32_t socket_id)
504 {
505         struct virtnet_ctl *cvq;
506         int ret;
507         struct virtio_hw *hw = dev->data->dev_private;
508
509         PMD_INIT_FUNC_TRACE();
510         ret = virtio_dev_queue_setup(dev, VTNET_CQ, VTNET_SQ_CQ_QUEUE_IDX,
511                         vtpci_queue_idx, 0, socket_id, (void **)&cvq);
512         if (ret < 0) {
513                 PMD_INIT_LOG(ERR, "control vq initialization failed");
514                 return ret;
515         }
516
517         hw->cvq = cvq;
518         return 0;
519 }
520
521 static void
522 virtio_free_queues(struct rte_eth_dev *dev)
523 {
524         unsigned int i;
525
526         for (i = 0; i < dev->data->nb_rx_queues; i++)
527                 virtio_dev_rx_queue_release(dev->data->rx_queues[i]);
528
529         dev->data->nb_rx_queues = 0;
530
531         for (i = 0; i < dev->data->nb_tx_queues; i++)
532                 virtio_dev_tx_queue_release(dev->data->tx_queues[i]);
533
534         dev->data->nb_tx_queues = 0;
535 }
536
537 static void
538 virtio_dev_close(struct rte_eth_dev *dev)
539 {
540         struct virtio_hw *hw = dev->data->dev_private;
541
542         PMD_INIT_LOG(DEBUG, "virtio_dev_close");
543
544         if (hw->started == 1)
545                 virtio_dev_stop(dev);
546
547         /* reset the NIC */
548         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
549                 vtpci_irq_config(hw, VIRTIO_MSI_NO_VECTOR);
550         vtpci_reset(hw);
551         virtio_dev_free_mbufs(dev);
552         virtio_free_queues(dev);
553 }
554
555 static void
556 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
557 {
558         struct virtio_hw *hw = dev->data->dev_private;
559         struct virtio_pmd_ctrl ctrl;
560         int dlen[1];
561         int ret;
562
563         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
564                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
565                 return;
566         }
567
568         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
569         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
570         ctrl.data[0] = 1;
571         dlen[0] = 1;
572
573         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
574         if (ret)
575                 PMD_INIT_LOG(ERR, "Failed to enable promisc");
576 }
577
578 static void
579 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
580 {
581         struct virtio_hw *hw = dev->data->dev_private;
582         struct virtio_pmd_ctrl ctrl;
583         int dlen[1];
584         int ret;
585
586         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
587                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
588                 return;
589         }
590
591         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
592         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
593         ctrl.data[0] = 0;
594         dlen[0] = 1;
595
596         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
597         if (ret)
598                 PMD_INIT_LOG(ERR, "Failed to disable promisc");
599 }
600
601 static void
602 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
603 {
604         struct virtio_hw *hw = dev->data->dev_private;
605         struct virtio_pmd_ctrl ctrl;
606         int dlen[1];
607         int ret;
608
609         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
610                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
611                 return;
612         }
613
614         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
615         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
616         ctrl.data[0] = 1;
617         dlen[0] = 1;
618
619         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
620         if (ret)
621                 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
622 }
623
624 static void
625 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
626 {
627         struct virtio_hw *hw = dev->data->dev_private;
628         struct virtio_pmd_ctrl ctrl;
629         int dlen[1];
630         int ret;
631
632         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
633                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
634                 return;
635         }
636
637         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
638         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
639         ctrl.data[0] = 0;
640         dlen[0] = 1;
641
642         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
643         if (ret)
644                 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
645 }
646
647 /*
648  * dev_ops for virtio, bare necessities for basic operation
649  */
650 static const struct eth_dev_ops virtio_eth_dev_ops = {
651         .dev_configure           = virtio_dev_configure,
652         .dev_start               = virtio_dev_start,
653         .dev_stop                = virtio_dev_stop,
654         .dev_close               = virtio_dev_close,
655         .promiscuous_enable      = virtio_dev_promiscuous_enable,
656         .promiscuous_disable     = virtio_dev_promiscuous_disable,
657         .allmulticast_enable     = virtio_dev_allmulticast_enable,
658         .allmulticast_disable    = virtio_dev_allmulticast_disable,
659
660         .dev_infos_get           = virtio_dev_info_get,
661         .stats_get               = virtio_dev_stats_get,
662         .xstats_get              = virtio_dev_xstats_get,
663         .xstats_get_names        = virtio_dev_xstats_get_names,
664         .stats_reset             = virtio_dev_stats_reset,
665         .xstats_reset            = virtio_dev_stats_reset,
666         .link_update             = virtio_dev_link_update,
667         .rx_queue_setup          = virtio_dev_rx_queue_setup,
668         .rx_queue_release        = virtio_dev_rx_queue_release,
669         .tx_queue_setup          = virtio_dev_tx_queue_setup,
670         .tx_queue_release        = virtio_dev_tx_queue_release,
671         /* collect stats per queue */
672         .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
673         .vlan_filter_set         = virtio_vlan_filter_set,
674         .mac_addr_add            = virtio_mac_addr_add,
675         .mac_addr_remove         = virtio_mac_addr_remove,
676         .mac_addr_set            = virtio_mac_addr_set,
677 };
678
679 static inline int
680 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
681                                 struct rte_eth_link *link)
682 {
683         struct rte_eth_link *dst = link;
684         struct rte_eth_link *src = &(dev->data->dev_link);
685
686         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
687                         *(uint64_t *)src) == 0)
688                 return -1;
689
690         return 0;
691 }
692
693 /**
694  * Atomically writes the link status information into global
695  * structure rte_eth_dev.
696  *
697  * @param dev
698  *   - Pointer to the structure rte_eth_dev to read from.
699  *   - Pointer to the buffer to be saved with the link status.
700  *
701  * @return
702  *   - On success, zero.
703  *   - On failure, negative value.
704  */
705 static inline int
706 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
707                 struct rte_eth_link *link)
708 {
709         struct rte_eth_link *dst = &(dev->data->dev_link);
710         struct rte_eth_link *src = link;
711
712         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
713                                         *(uint64_t *)src) == 0)
714                 return -1;
715
716         return 0;
717 }
718
719 static void
720 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
721 {
722         unsigned i;
723
724         for (i = 0; i < dev->data->nb_tx_queues; i++) {
725                 const struct virtnet_tx *txvq = dev->data->tx_queues[i];
726                 if (txvq == NULL)
727                         continue;
728
729                 stats->opackets += txvq->stats.packets;
730                 stats->obytes += txvq->stats.bytes;
731                 stats->oerrors += txvq->stats.errors;
732
733                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
734                         stats->q_opackets[i] = txvq->stats.packets;
735                         stats->q_obytes[i] = txvq->stats.bytes;
736                 }
737         }
738
739         for (i = 0; i < dev->data->nb_rx_queues; i++) {
740                 const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
741                 if (rxvq == NULL)
742                         continue;
743
744                 stats->ipackets += rxvq->stats.packets;
745                 stats->ibytes += rxvq->stats.bytes;
746                 stats->ierrors += rxvq->stats.errors;
747
748                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
749                         stats->q_ipackets[i] = rxvq->stats.packets;
750                         stats->q_ibytes[i] = rxvq->stats.bytes;
751                 }
752         }
753
754         stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
755 }
756
757 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
758                                        struct rte_eth_xstat_name *xstats_names,
759                                        __rte_unused unsigned limit)
760 {
761         unsigned i;
762         unsigned count = 0;
763         unsigned t;
764
765         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
766                 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
767
768         if (xstats_names == NULL) {
769                 /* Note: limit checked in rte_eth_xstats_names() */
770
771                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
772                         struct virtqueue *rxvq = dev->data->rx_queues[i];
773                         if (rxvq == NULL)
774                                 continue;
775                         for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
776                                 snprintf(xstats_names[count].name,
777                                         sizeof(xstats_names[count].name),
778                                         "rx_q%u_%s", i,
779                                         rte_virtio_rxq_stat_strings[t].name);
780                                 xstats_names[count].id = count;
781                                 count++;
782                         }
783                 }
784
785                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
786                         struct virtqueue *txvq = dev->data->tx_queues[i];
787                         if (txvq == NULL)
788                                 continue;
789                         for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
790                                 snprintf(xstats_names[count].name,
791                                         sizeof(xstats_names[count].name),
792                                         "tx_q%u_%s", i,
793                                         rte_virtio_txq_stat_strings[t].name);
794                                 xstats_names[count].id = count;
795                                 count++;
796                         }
797                 }
798                 return count;
799         }
800         return nstats;
801 }
802
803 static int
804 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
805                       unsigned n)
806 {
807         unsigned i;
808         unsigned count = 0;
809
810         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
811                 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
812
813         if (n < nstats)
814                 return nstats;
815
816         for (i = 0; i < dev->data->nb_rx_queues; i++) {
817                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
818
819                 if (rxvq == NULL)
820                         continue;
821
822                 unsigned t;
823
824                 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
825                         xstats[count].id = count;
826                         xstats[count].value = *(uint64_t *)(((char *)rxvq) +
827                                 rte_virtio_rxq_stat_strings[t].offset);
828                         count++;
829                 }
830         }
831
832         for (i = 0; i < dev->data->nb_tx_queues; i++) {
833                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
834
835                 if (txvq == NULL)
836                         continue;
837
838                 unsigned t;
839
840                 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
841                         xstats[count].id = count;
842                         xstats[count].value = *(uint64_t *)(((char *)txvq) +
843                                 rte_virtio_txq_stat_strings[t].offset);
844                         count++;
845                 }
846         }
847
848         return count;
849 }
850
851 static void
852 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
853 {
854         virtio_update_stats(dev, stats);
855 }
856
857 static void
858 virtio_dev_stats_reset(struct rte_eth_dev *dev)
859 {
860         unsigned int i;
861
862         for (i = 0; i < dev->data->nb_tx_queues; i++) {
863                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
864                 if (txvq == NULL)
865                         continue;
866
867                 txvq->stats.packets = 0;
868                 txvq->stats.bytes = 0;
869                 txvq->stats.errors = 0;
870                 txvq->stats.multicast = 0;
871                 txvq->stats.broadcast = 0;
872                 memset(txvq->stats.size_bins, 0,
873                        sizeof(txvq->stats.size_bins[0]) * 8);
874         }
875
876         for (i = 0; i < dev->data->nb_rx_queues; i++) {
877                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
878                 if (rxvq == NULL)
879                         continue;
880
881                 rxvq->stats.packets = 0;
882                 rxvq->stats.bytes = 0;
883                 rxvq->stats.errors = 0;
884                 rxvq->stats.multicast = 0;
885                 rxvq->stats.broadcast = 0;
886                 memset(rxvq->stats.size_bins, 0,
887                        sizeof(rxvq->stats.size_bins[0]) * 8);
888         }
889 }
890
891 static void
892 virtio_set_hwaddr(struct virtio_hw *hw)
893 {
894         vtpci_write_dev_config(hw,
895                         offsetof(struct virtio_net_config, mac),
896                         &hw->mac_addr, ETHER_ADDR_LEN);
897 }
898
899 static void
900 virtio_get_hwaddr(struct virtio_hw *hw)
901 {
902         if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
903                 vtpci_read_dev_config(hw,
904                         offsetof(struct virtio_net_config, mac),
905                         &hw->mac_addr, ETHER_ADDR_LEN);
906         } else {
907                 eth_random_addr(&hw->mac_addr[0]);
908                 virtio_set_hwaddr(hw);
909         }
910 }
911
912 static void
913 virtio_mac_table_set(struct virtio_hw *hw,
914                      const struct virtio_net_ctrl_mac *uc,
915                      const struct virtio_net_ctrl_mac *mc)
916 {
917         struct virtio_pmd_ctrl ctrl;
918         int err, len[2];
919
920         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
921                 PMD_DRV_LOG(INFO, "host does not support mac table");
922                 return;
923         }
924
925         ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
926         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
927
928         len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
929         memcpy(ctrl.data, uc, len[0]);
930
931         len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
932         memcpy(ctrl.data + len[0], mc, len[1]);
933
934         err = virtio_send_command(hw->cvq, &ctrl, len, 2);
935         if (err != 0)
936                 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
937 }
938
939 static void
940 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
941                     uint32_t index, uint32_t vmdq __rte_unused)
942 {
943         struct virtio_hw *hw = dev->data->dev_private;
944         const struct ether_addr *addrs = dev->data->mac_addrs;
945         unsigned int i;
946         struct virtio_net_ctrl_mac *uc, *mc;
947
948         if (index >= VIRTIO_MAX_MAC_ADDRS) {
949                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
950                 return;
951         }
952
953         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
954         uc->entries = 0;
955         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
956         mc->entries = 0;
957
958         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
959                 const struct ether_addr *addr
960                         = (i == index) ? mac_addr : addrs + i;
961                 struct virtio_net_ctrl_mac *tbl
962                         = is_multicast_ether_addr(addr) ? mc : uc;
963
964                 memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
965         }
966
967         virtio_mac_table_set(hw, uc, mc);
968 }
969
970 static void
971 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
972 {
973         struct virtio_hw *hw = dev->data->dev_private;
974         struct ether_addr *addrs = dev->data->mac_addrs;
975         struct virtio_net_ctrl_mac *uc, *mc;
976         unsigned int i;
977
978         if (index >= VIRTIO_MAX_MAC_ADDRS) {
979                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
980                 return;
981         }
982
983         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
984         uc->entries = 0;
985         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
986         mc->entries = 0;
987
988         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
989                 struct virtio_net_ctrl_mac *tbl;
990
991                 if (i == index || is_zero_ether_addr(addrs + i))
992                         continue;
993
994                 tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
995                 memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
996         }
997
998         virtio_mac_table_set(hw, uc, mc);
999 }
1000
1001 static void
1002 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1003 {
1004         struct virtio_hw *hw = dev->data->dev_private;
1005
1006         memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
1007
1008         /* Use atomic update if available */
1009         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1010                 struct virtio_pmd_ctrl ctrl;
1011                 int len = ETHER_ADDR_LEN;
1012
1013                 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1014                 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1015
1016                 memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
1017                 virtio_send_command(hw->cvq, &ctrl, &len, 1);
1018         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1019                 virtio_set_hwaddr(hw);
1020 }
1021
1022 static int
1023 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1024 {
1025         struct virtio_hw *hw = dev->data->dev_private;
1026         struct virtio_pmd_ctrl ctrl;
1027         int len;
1028
1029         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1030                 return -ENOTSUP;
1031
1032         ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1033         ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1034         memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1035         len = sizeof(vlan_id);
1036
1037         return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1038 }
1039
1040 static int
1041 virtio_negotiate_features(struct virtio_hw *hw)
1042 {
1043         uint64_t host_features;
1044
1045         /* Prepare guest_features: feature that driver wants to support */
1046         hw->guest_features = VIRTIO_PMD_GUEST_FEATURES;
1047         PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1048                 hw->guest_features);
1049
1050         /* Read device(host) feature bits */
1051         host_features = hw->vtpci_ops->get_features(hw);
1052         PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1053                 host_features);
1054
1055         /*
1056          * Negotiate features: Subset of device feature bits are written back
1057          * guest feature bits.
1058          */
1059         hw->guest_features = vtpci_negotiate_features(hw, host_features);
1060         PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1061                 hw->guest_features);
1062
1063         if (hw->modern) {
1064                 if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1065                         PMD_INIT_LOG(ERR,
1066                                 "VIRTIO_F_VERSION_1 features is not enabled.");
1067                         return -1;
1068                 }
1069                 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1070                 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1071                         PMD_INIT_LOG(ERR,
1072                                 "failed to set FEATURES_OK status!");
1073                         return -1;
1074                 }
1075         }
1076
1077         return 0;
1078 }
1079
1080 /*
1081  * Process Virtio Config changed interrupt and call the callback
1082  * if link state changed.
1083  */
1084 static void
1085 virtio_interrupt_handler(__rte_unused struct rte_intr_handle *handle,
1086                          void *param)
1087 {
1088         struct rte_eth_dev *dev = param;
1089         struct virtio_hw *hw = dev->data->dev_private;
1090         uint8_t isr;
1091
1092         /* Read interrupt status which clears interrupt */
1093         isr = vtpci_isr(hw);
1094         PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1095
1096         if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0)
1097                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1098
1099         if (isr & VIRTIO_PCI_ISR_CONFIG) {
1100                 if (virtio_dev_link_update(dev, 0) == 0)
1101                         _rte_eth_dev_callback_process(dev,
1102                                                       RTE_ETH_EVENT_INTR_LSC);
1103         }
1104
1105 }
1106
1107 static void
1108 rx_func_get(struct rte_eth_dev *eth_dev)
1109 {
1110         struct virtio_hw *hw = eth_dev->data->dev_private;
1111         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF))
1112                 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1113         else
1114                 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1115 }
1116
1117 /*
1118  * This function is based on probe() function in virtio_pci.c
1119  * It returns 0 on success.
1120  */
1121 static int
1122 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1123 {
1124         struct virtio_hw *hw = eth_dev->data->dev_private;
1125         struct virtio_net_config *config;
1126         struct virtio_net_config local_config;
1127         struct rte_pci_device *pci_dev;
1128         uint32_t dev_flags = RTE_ETH_DEV_DETACHABLE;
1129         int ret;
1130
1131         RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1132
1133         eth_dev->dev_ops = &virtio_eth_dev_ops;
1134         eth_dev->tx_pkt_burst = &virtio_xmit_pkts;
1135
1136         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1137                 rx_func_get(eth_dev);
1138                 return 0;
1139         }
1140
1141         /* Allocate memory for storing MAC addresses */
1142         eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
1143         if (eth_dev->data->mac_addrs == NULL) {
1144                 PMD_INIT_LOG(ERR,
1145                         "Failed to allocate %d bytes needed to store MAC addresses",
1146                         VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
1147                 return -ENOMEM;
1148         }
1149
1150         pci_dev = eth_dev->pci_dev;
1151
1152         ret = vtpci_init(pci_dev, hw, &dev_flags);
1153         if (ret)
1154                 return ret;
1155
1156         /* Reset the device although not necessary at startup */
1157         vtpci_reset(hw);
1158
1159         /* Tell the host we've noticed this device. */
1160         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1161
1162         /* Tell the host we've known how to drive the device. */
1163         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1164         if (virtio_negotiate_features(hw) < 0)
1165                 return -1;
1166
1167         /* If host does not support status then disable LSC */
1168         if (!vtpci_with_feature(hw, VIRTIO_NET_F_STATUS))
1169                 dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1170
1171         rte_eth_copy_pci_info(eth_dev, pci_dev);
1172         eth_dev->data->dev_flags = dev_flags;
1173
1174         rx_func_get(eth_dev);
1175
1176         /* Setting up rx_header size for the device */
1177         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1178             vtpci_with_feature(hw, VIRTIO_F_VERSION_1))
1179                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1180         else
1181                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1182
1183         /* Copy the permanent MAC address to: virtio_hw */
1184         virtio_get_hwaddr(hw);
1185         ether_addr_copy((struct ether_addr *) hw->mac_addr,
1186                         &eth_dev->data->mac_addrs[0]);
1187         PMD_INIT_LOG(DEBUG,
1188                      "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1189                      hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1190                      hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1191
1192         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1193                 config = &local_config;
1194
1195                 vtpci_read_dev_config(hw,
1196                         offsetof(struct virtio_net_config, mac),
1197                         &config->mac, sizeof(config->mac));
1198
1199                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1200                         vtpci_read_dev_config(hw,
1201                                 offsetof(struct virtio_net_config, status),
1202                                 &config->status, sizeof(config->status));
1203                 } else {
1204                         PMD_INIT_LOG(DEBUG,
1205                                      "VIRTIO_NET_F_STATUS is not supported");
1206                         config->status = 0;
1207                 }
1208
1209                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1210                         vtpci_read_dev_config(hw,
1211                                 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1212                                 &config->max_virtqueue_pairs,
1213                                 sizeof(config->max_virtqueue_pairs));
1214                 } else {
1215                         PMD_INIT_LOG(DEBUG,
1216                                      "VIRTIO_NET_F_MQ is not supported");
1217                         config->max_virtqueue_pairs = 1;
1218                 }
1219
1220                 hw->max_rx_queues =
1221                         (VIRTIO_MAX_RX_QUEUES < config->max_virtqueue_pairs) ?
1222                         VIRTIO_MAX_RX_QUEUES : config->max_virtqueue_pairs;
1223                 hw->max_tx_queues =
1224                         (VIRTIO_MAX_TX_QUEUES < config->max_virtqueue_pairs) ?
1225                         VIRTIO_MAX_TX_QUEUES : config->max_virtqueue_pairs;
1226
1227                 virtio_dev_cq_queue_setup(eth_dev,
1228                                         config->max_virtqueue_pairs * 2,
1229                                         SOCKET_ID_ANY);
1230
1231                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1232                                 config->max_virtqueue_pairs);
1233                 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1234                 PMD_INIT_LOG(DEBUG,
1235                                 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1236                                 config->mac[0], config->mac[1],
1237                                 config->mac[2], config->mac[3],
1238                                 config->mac[4], config->mac[5]);
1239         } else {
1240                 hw->max_rx_queues = 1;
1241                 hw->max_tx_queues = 1;
1242         }
1243
1244         PMD_INIT_LOG(DEBUG, "hw->max_rx_queues=%d   hw->max_tx_queues=%d",
1245                         hw->max_rx_queues, hw->max_tx_queues);
1246         PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1247                         eth_dev->data->port_id, pci_dev->id.vendor_id,
1248                         pci_dev->id.device_id);
1249
1250         /* Setup interrupt callback  */
1251         if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1252                 rte_intr_callback_register(&pci_dev->intr_handle,
1253                                    virtio_interrupt_handler, eth_dev);
1254
1255         virtio_dev_cq_start(eth_dev);
1256
1257         return 0;
1258 }
1259
1260 static int
1261 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1262 {
1263         struct rte_pci_device *pci_dev;
1264         struct virtio_hw *hw = eth_dev->data->dev_private;
1265
1266         PMD_INIT_FUNC_TRACE();
1267
1268         if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1269                 return -EPERM;
1270
1271         /* Close it anyway since there's no way to know if closed */
1272         virtio_dev_close(eth_dev);
1273
1274         pci_dev = eth_dev->pci_dev;
1275
1276         eth_dev->dev_ops = NULL;
1277         eth_dev->tx_pkt_burst = NULL;
1278         eth_dev->rx_pkt_burst = NULL;
1279
1280         if (hw->cvq)
1281                 virtio_dev_queue_release(hw->cvq->vq);
1282
1283         rte_free(eth_dev->data->mac_addrs);
1284         eth_dev->data->mac_addrs = NULL;
1285
1286         /* reset interrupt callback  */
1287         if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1288                 rte_intr_callback_unregister(&pci_dev->intr_handle,
1289                                                 virtio_interrupt_handler,
1290                                                 eth_dev);
1291         rte_eal_pci_unmap_device(pci_dev);
1292
1293         PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1294
1295         return 0;
1296 }
1297
1298 static struct eth_driver rte_virtio_pmd = {
1299         .pci_drv = {
1300                 .name = "rte_virtio_pmd",
1301                 .id_table = pci_id_virtio_map,
1302                 .drv_flags = RTE_PCI_DRV_DETACHABLE,
1303         },
1304         .eth_dev_init = eth_virtio_dev_init,
1305         .eth_dev_uninit = eth_virtio_dev_uninit,
1306         .dev_private_size = sizeof(struct virtio_hw),
1307 };
1308
1309 /*
1310  * Driver initialization routine.
1311  * Invoked once at EAL init time.
1312  * Register itself as the [Poll Mode] Driver of PCI virtio devices.
1313  * Returns 0 on success.
1314  */
1315 static int
1316 rte_virtio_pmd_init(const char *name __rte_unused,
1317                     const char *param __rte_unused)
1318 {
1319         if (rte_eal_iopl_init() != 0) {
1320                 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1321                 return -1;
1322         }
1323
1324         rte_eth_driver_register(&rte_virtio_pmd);
1325         return 0;
1326 }
1327
1328 /*
1329  * Configure virtio device
1330  * It returns 0 on success.
1331  */
1332 static int
1333 virtio_dev_configure(struct rte_eth_dev *dev)
1334 {
1335         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1336         struct virtio_hw *hw = dev->data->dev_private;
1337
1338         PMD_INIT_LOG(DEBUG, "configure");
1339
1340         if (rxmode->hw_ip_checksum) {
1341                 PMD_DRV_LOG(ERR, "HW IP checksum not supported");
1342                 return -EINVAL;
1343         }
1344
1345         hw->vlan_strip = rxmode->hw_vlan_strip;
1346
1347         if (rxmode->hw_vlan_filter
1348             && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1349                 PMD_DRV_LOG(NOTICE,
1350                             "vlan filtering not available on this host");
1351                 return -ENOTSUP;
1352         }
1353
1354         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1355                 if (vtpci_irq_config(hw, 0) == VIRTIO_MSI_NO_VECTOR) {
1356                         PMD_DRV_LOG(ERR, "failed to set config vector");
1357                         return -EBUSY;
1358                 }
1359
1360         return 0;
1361 }
1362
1363
1364 static int
1365 virtio_dev_start(struct rte_eth_dev *dev)
1366 {
1367         uint16_t nb_queues, i;
1368         struct virtio_hw *hw = dev->data->dev_private;
1369         struct virtnet_rx *rxvq;
1370         struct virtnet_tx *txvq __rte_unused;
1371
1372         /* check if lsc interrupt feature is enabled */
1373         if (dev->data->dev_conf.intr_conf.lsc) {
1374                 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1375                         PMD_DRV_LOG(ERR, "link status not supported by host");
1376                         return -ENOTSUP;
1377                 }
1378
1379                 if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0) {
1380                         PMD_DRV_LOG(ERR, "interrupt enable failed");
1381                         return -EIO;
1382                 }
1383         }
1384
1385         /* Initialize Link state */
1386         virtio_dev_link_update(dev, 0);
1387
1388         /* On restart after stop do not touch queues */
1389         if (hw->started)
1390                 return 0;
1391
1392         /* Do final configuration before rx/tx engine starts */
1393         virtio_dev_rxtx_start(dev);
1394         vtpci_reinit_complete(hw);
1395
1396         hw->started = 1;
1397
1398         /*Notify the backend
1399          *Otherwise the tap backend might already stop its queue due to fullness.
1400          *vhost backend will have no chance to be waked up
1401          */
1402         nb_queues = dev->data->nb_rx_queues;
1403         if (nb_queues > 1) {
1404                 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1405                         return -EINVAL;
1406         }
1407
1408         PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1409
1410         for (i = 0; i < nb_queues; i++) {
1411                 rxvq = dev->data->rx_queues[i];
1412                 virtqueue_notify(rxvq->vq);
1413         }
1414
1415         PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1416
1417         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1418                 rxvq = dev->data->rx_queues[i];
1419                 VIRTQUEUE_DUMP(rxvq->vq);
1420         }
1421
1422         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1423                 txvq = dev->data->tx_queues[i];
1424                 VIRTQUEUE_DUMP(txvq->vq);
1425         }
1426
1427         return 0;
1428 }
1429
1430 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1431 {
1432         struct rte_mbuf *buf;
1433         int i, mbuf_num = 0;
1434
1435         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1436                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1437
1438                 PMD_INIT_LOG(DEBUG,
1439                              "Before freeing rxq[%d] used and unused buf", i);
1440                 VIRTQUEUE_DUMP(rxvq->vq);
1441
1442                 PMD_INIT_LOG(DEBUG, "rx_queues[%d]=%p", i, rxvq);
1443                 while ((buf = virtqueue_detatch_unused(rxvq->vq)) != NULL) {
1444                         rte_pktmbuf_free(buf);
1445                         mbuf_num++;
1446                 }
1447
1448                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1449                 PMD_INIT_LOG(DEBUG,
1450                              "After freeing rxq[%d] used and unused buf", i);
1451                 VIRTQUEUE_DUMP(rxvq->vq);
1452         }
1453
1454         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1455                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1456
1457                 PMD_INIT_LOG(DEBUG,
1458                              "Before freeing txq[%d] used and unused bufs",
1459                              i);
1460                 VIRTQUEUE_DUMP(txvq->vq);
1461
1462                 mbuf_num = 0;
1463                 while ((buf = virtqueue_detatch_unused(txvq->vq)) != NULL) {
1464                         rte_pktmbuf_free(buf);
1465                         mbuf_num++;
1466                 }
1467
1468                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1469                 PMD_INIT_LOG(DEBUG,
1470                              "After freeing txq[%d] used and unused buf", i);
1471                 VIRTQUEUE_DUMP(txvq->vq);
1472         }
1473 }
1474
1475 /*
1476  * Stop device: disable interrupt and mark link down
1477  */
1478 static void
1479 virtio_dev_stop(struct rte_eth_dev *dev)
1480 {
1481         struct rte_eth_link link;
1482         struct virtio_hw *hw = dev->data->dev_private;
1483
1484         PMD_INIT_LOG(DEBUG, "stop");
1485
1486         hw->started = 0;
1487
1488         if (dev->data->dev_conf.intr_conf.lsc)
1489                 rte_intr_disable(&dev->pci_dev->intr_handle);
1490
1491         memset(&link, 0, sizeof(link));
1492         virtio_dev_atomic_write_link_status(dev, &link);
1493 }
1494
1495 static int
1496 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1497 {
1498         struct rte_eth_link link, old;
1499         uint16_t status;
1500         struct virtio_hw *hw = dev->data->dev_private;
1501         memset(&link, 0, sizeof(link));
1502         virtio_dev_atomic_read_link_status(dev, &link);
1503         old = link;
1504         link.link_duplex = ETH_LINK_FULL_DUPLEX;
1505         link.link_speed  = SPEED_10G;
1506
1507         if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1508                 PMD_INIT_LOG(DEBUG, "Get link status from hw");
1509                 vtpci_read_dev_config(hw,
1510                                 offsetof(struct virtio_net_config, status),
1511                                 &status, sizeof(status));
1512                 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1513                         link.link_status = ETH_LINK_DOWN;
1514                         PMD_INIT_LOG(DEBUG, "Port %d is down",
1515                                      dev->data->port_id);
1516                 } else {
1517                         link.link_status = ETH_LINK_UP;
1518                         PMD_INIT_LOG(DEBUG, "Port %d is up",
1519                                      dev->data->port_id);
1520                 }
1521         } else {
1522                 link.link_status = ETH_LINK_UP;
1523         }
1524         virtio_dev_atomic_write_link_status(dev, &link);
1525
1526         return (old.link_status == link.link_status) ? -1 : 0;
1527 }
1528
1529 static void
1530 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1531 {
1532         struct virtio_hw *hw = dev->data->dev_private;
1533
1534         dev_info->driver_name = dev->driver->pci_drv.name;
1535         dev_info->max_rx_queues = (uint16_t)hw->max_rx_queues;
1536         dev_info->max_tx_queues = (uint16_t)hw->max_tx_queues;
1537         dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1538         dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1539         dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1540         dev_info->default_txconf = (struct rte_eth_txconf) {
1541                 .txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
1542         };
1543 }
1544
1545 /*
1546  * It enables testpmd to collect per queue stats.
1547  */
1548 static int
1549 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
1550 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
1551 __rte_unused uint8_t is_rx)
1552 {
1553         return 0;
1554 }
1555
1556 static struct rte_driver rte_virtio_driver = {
1557         .type = PMD_PDEV,
1558         .init = rte_virtio_pmd_init,
1559 };
1560
1561 PMD_REGISTER_DRIVER(rte_virtio_driver);