net/virtio: allocate queue at init stage
[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_uninit(struct rte_eth_dev *eth_dev);
63 static int  virtio_dev_configure(struct rte_eth_dev *dev);
64 static int  virtio_dev_start(struct rte_eth_dev *dev);
65 static void virtio_dev_stop(struct rte_eth_dev *dev);
66 static void virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
67 static void virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
68 static void virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
69 static void virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
70 static void virtio_dev_info_get(struct rte_eth_dev *dev,
71                                 struct rte_eth_dev_info *dev_info);
72 static int virtio_dev_link_update(struct rte_eth_dev *dev,
73         __rte_unused int wait_to_complete);
74
75 static void virtio_set_hwaddr(struct virtio_hw *hw);
76 static void virtio_get_hwaddr(struct virtio_hw *hw);
77
78 static void virtio_dev_stats_get(struct rte_eth_dev *dev,
79                                  struct rte_eth_stats *stats);
80 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
81                                  struct rte_eth_xstat *xstats, unsigned n);
82 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
83                                        struct rte_eth_xstat_name *xstats_names,
84                                        unsigned limit);
85 static void virtio_dev_stats_reset(struct rte_eth_dev *dev);
86 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
87 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
88                                 uint16_t vlan_id, int on);
89 static void virtio_mac_addr_add(struct rte_eth_dev *dev,
90                                 struct ether_addr *mac_addr,
91                                 uint32_t index, uint32_t vmdq __rte_unused);
92 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
93 static void virtio_mac_addr_set(struct rte_eth_dev *dev,
94                                 struct ether_addr *mac_addr);
95
96 static int virtio_dev_queue_stats_mapping_set(
97         __rte_unused struct rte_eth_dev *eth_dev,
98         __rte_unused uint16_t queue_id,
99         __rte_unused uint8_t stat_idx,
100         __rte_unused uint8_t is_rx);
101
102 /*
103  * The set of PCI devices this driver supports
104  */
105 static const struct rte_pci_id pci_id_virtio_map[] = {
106         { RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_LEGACY_DEVICEID_NET) },
107         { RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_MODERN_DEVICEID_NET) },
108         { .vendor_id = 0, /* sentinel */ },
109 };
110
111 struct rte_virtio_xstats_name_off {
112         char name[RTE_ETH_XSTATS_NAME_SIZE];
113         unsigned offset;
114 };
115
116 /* [rt]x_qX_ is prepended to the name string here */
117 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
118         {"good_packets",           offsetof(struct virtnet_rx, stats.packets)},
119         {"good_bytes",             offsetof(struct virtnet_rx, stats.bytes)},
120         {"errors",                 offsetof(struct virtnet_rx, stats.errors)},
121         {"multicast_packets",      offsetof(struct virtnet_rx, stats.multicast)},
122         {"broadcast_packets",      offsetof(struct virtnet_rx, stats.broadcast)},
123         {"undersize_packets",      offsetof(struct virtnet_rx, stats.size_bins[0])},
124         {"size_64_packets",        offsetof(struct virtnet_rx, stats.size_bins[1])},
125         {"size_65_127_packets",    offsetof(struct virtnet_rx, stats.size_bins[2])},
126         {"size_128_255_packets",   offsetof(struct virtnet_rx, stats.size_bins[3])},
127         {"size_256_511_packets",   offsetof(struct virtnet_rx, stats.size_bins[4])},
128         {"size_512_1023_packets",  offsetof(struct virtnet_rx, stats.size_bins[5])},
129         {"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
130         {"size_1519_max_packets",  offsetof(struct virtnet_rx, stats.size_bins[7])},
131 };
132
133 /* [rt]x_qX_ is prepended to the name string here */
134 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
135         {"good_packets",           offsetof(struct virtnet_tx, stats.packets)},
136         {"good_bytes",             offsetof(struct virtnet_tx, stats.bytes)},
137         {"errors",                 offsetof(struct virtnet_tx, stats.errors)},
138         {"multicast_packets",      offsetof(struct virtnet_tx, stats.multicast)},
139         {"broadcast_packets",      offsetof(struct virtnet_tx, stats.broadcast)},
140         {"undersize_packets",      offsetof(struct virtnet_tx, stats.size_bins[0])},
141         {"size_64_packets",        offsetof(struct virtnet_tx, stats.size_bins[1])},
142         {"size_65_127_packets",    offsetof(struct virtnet_tx, stats.size_bins[2])},
143         {"size_128_255_packets",   offsetof(struct virtnet_tx, stats.size_bins[3])},
144         {"size_256_511_packets",   offsetof(struct virtnet_tx, stats.size_bins[4])},
145         {"size_512_1023_packets",  offsetof(struct virtnet_tx, stats.size_bins[5])},
146         {"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
147         {"size_1519_max_packets",  offsetof(struct virtnet_tx, stats.size_bins[7])},
148 };
149
150 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
151                             sizeof(rte_virtio_rxq_stat_strings[0]))
152 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
153                             sizeof(rte_virtio_txq_stat_strings[0]))
154
155 static int
156 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
157                 int *dlen, int pkt_num)
158 {
159         uint32_t head, i;
160         int k, sum = 0;
161         virtio_net_ctrl_ack status = ~0;
162         struct virtio_pmd_ctrl result;
163         struct virtqueue *vq;
164
165         ctrl->status = status;
166
167         if (!cvq || !cvq->vq) {
168                 PMD_INIT_LOG(ERR, "Control queue is not supported.");
169                 return -1;
170         }
171         vq = cvq->vq;
172         head = vq->vq_desc_head_idx;
173
174         PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
175                 "vq->hw->cvq = %p vq = %p",
176                 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
177
178         if ((vq->vq_free_cnt < ((uint32_t)pkt_num + 2)) || (pkt_num < 1))
179                 return -1;
180
181         memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
182                 sizeof(struct virtio_pmd_ctrl));
183
184         /*
185          * Format is enforced in qemu code:
186          * One TX packet for header;
187          * At least one TX packet per argument;
188          * One RX packet for ACK.
189          */
190         vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
191         vq->vq_ring.desc[head].addr = cvq->virtio_net_hdr_mem;
192         vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
193         vq->vq_free_cnt--;
194         i = vq->vq_ring.desc[head].next;
195
196         for (k = 0; k < pkt_num; k++) {
197                 vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
198                 vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
199                         + sizeof(struct virtio_net_ctrl_hdr)
200                         + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
201                 vq->vq_ring.desc[i].len = dlen[k];
202                 sum += dlen[k];
203                 vq->vq_free_cnt--;
204                 i = vq->vq_ring.desc[i].next;
205         }
206
207         vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
208         vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
209                         + sizeof(struct virtio_net_ctrl_hdr);
210         vq->vq_ring.desc[i].len = sizeof(ctrl->status);
211         vq->vq_free_cnt--;
212
213         vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
214
215         vq_update_avail_ring(vq, head);
216         vq_update_avail_idx(vq);
217
218         PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
219
220         virtqueue_notify(vq);
221
222         rte_rmb();
223         while (VIRTQUEUE_NUSED(vq) == 0) {
224                 rte_rmb();
225                 usleep(100);
226         }
227
228         while (VIRTQUEUE_NUSED(vq)) {
229                 uint32_t idx, desc_idx, used_idx;
230                 struct vring_used_elem *uep;
231
232                 used_idx = (uint32_t)(vq->vq_used_cons_idx
233                                 & (vq->vq_nentries - 1));
234                 uep = &vq->vq_ring.used->ring[used_idx];
235                 idx = (uint32_t) uep->id;
236                 desc_idx = idx;
237
238                 while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
239                         desc_idx = vq->vq_ring.desc[desc_idx].next;
240                         vq->vq_free_cnt++;
241                 }
242
243                 vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
244                 vq->vq_desc_head_idx = idx;
245
246                 vq->vq_used_cons_idx++;
247                 vq->vq_free_cnt++;
248         }
249
250         PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
251                         vq->vq_free_cnt, vq->vq_desc_head_idx);
252
253         memcpy(&result, cvq->virtio_net_hdr_mz->addr,
254                         sizeof(struct virtio_pmd_ctrl));
255
256         return result.status;
257 }
258
259 static int
260 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
261 {
262         struct virtio_hw *hw = dev->data->dev_private;
263         struct virtio_pmd_ctrl ctrl;
264         int dlen[1];
265         int ret;
266
267         ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
268         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
269         memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
270
271         dlen[0] = sizeof(uint16_t);
272
273         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
274         if (ret) {
275                 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
276                           "failed, this is too late now...");
277                 return -EINVAL;
278         }
279
280         return 0;
281 }
282
283 static void
284 virtio_dev_queue_release(void *queue __rte_unused)
285 {
286         /* do nothing */
287 }
288
289 static int
290 virtio_get_queue_type(struct virtio_hw *hw, uint16_t vtpci_queue_idx)
291 {
292         if (vtpci_queue_idx == hw->max_queue_pairs * 2)
293                 return VTNET_CQ;
294         else if (vtpci_queue_idx % 2 == 0)
295                 return VTNET_RQ;
296         else
297                 return VTNET_TQ;
298 }
299
300 static uint16_t
301 virtio_get_nr_vq(struct virtio_hw *hw)
302 {
303         uint16_t nr_vq = hw->max_queue_pairs * 2;
304
305         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
306                 nr_vq += 1;
307
308         return nr_vq;
309 }
310
311 static int
312 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
313 {
314         char vq_name[VIRTQUEUE_MAX_NAME_SZ];
315         char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
316         const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
317         unsigned int vq_size, size;
318         struct virtio_hw *hw = dev->data->dev_private;
319         struct virtnet_rx *rxvq = NULL;
320         struct virtnet_tx *txvq = NULL;
321         struct virtnet_ctl *cvq = NULL;
322         struct virtqueue *vq;
323         size_t sz_hdr_mz = 0;
324         void *sw_ring = NULL;
325         int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
326         int ret;
327
328         PMD_INIT_LOG(DEBUG, "setting up queue: %u", vtpci_queue_idx);
329
330         /*
331          * Read the virtqueue size from the Queue Size field
332          * Always power of 2 and if 0 virtqueue does not exist
333          */
334         vq_size = hw->vtpci_ops->get_queue_num(hw, vtpci_queue_idx);
335         PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
336         if (vq_size == 0) {
337                 PMD_INIT_LOG(ERR, "virtqueue does not exist");
338                 return -EINVAL;
339         }
340
341         if (!rte_is_power_of_2(vq_size)) {
342                 PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
343                 return -EINVAL;
344         }
345
346         snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
347                  dev->data->port_id, vtpci_queue_idx);
348
349         size = RTE_ALIGN_CEIL(sizeof(*vq) +
350                                 vq_size * sizeof(struct vq_desc_extra),
351                                 RTE_CACHE_LINE_SIZE);
352         if (queue_type == VTNET_TQ) {
353                 /*
354                  * For each xmit packet, allocate a virtio_net_hdr
355                  * and indirect ring elements
356                  */
357                 sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
358         } else if (queue_type == VTNET_CQ) {
359                 /* Allocate a page for control vq command, data and status */
360                 sz_hdr_mz = PAGE_SIZE;
361         }
362
363         vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
364                                 SOCKET_ID_ANY);
365         if (vq == NULL) {
366                 PMD_INIT_LOG(ERR, "can not allocate vq");
367                 return -ENOMEM;
368         }
369         hw->vqs[vtpci_queue_idx] = vq;
370
371         vq->hw = hw;
372         vq->vq_queue_index = vtpci_queue_idx;
373         vq->vq_nentries = vq_size;
374         vq->vq_free_cnt = vq_size;
375
376         /*
377          * Reserve a memzone for vring elements
378          */
379         size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
380         vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
381         PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
382                      size, vq->vq_ring_size);
383
384         mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
385                                          SOCKET_ID_ANY,
386                                          0, VIRTIO_PCI_VRING_ALIGN);
387         if (mz == NULL) {
388                 if (rte_errno == EEXIST)
389                         mz = rte_memzone_lookup(vq_name);
390                 if (mz == NULL) {
391                         ret = -ENOMEM;
392                         goto fail_q_alloc;
393                 }
394         }
395
396         memset(mz->addr, 0, sizeof(mz->len));
397
398         vq->vq_ring_mem = mz->phys_addr;
399         vq->vq_ring_virt_mem = mz->addr;
400         PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%" PRIx64,
401                      (uint64_t)mz->phys_addr);
402         PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
403                      (uint64_t)(uintptr_t)mz->addr);
404
405         if (sz_hdr_mz) {
406                 snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
407                          dev->data->port_id, vtpci_queue_idx);
408                 hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
409                                                      SOCKET_ID_ANY, 0,
410                                                      RTE_CACHE_LINE_SIZE);
411                 if (hdr_mz == NULL) {
412                         if (rte_errno == EEXIST)
413                                 hdr_mz = rte_memzone_lookup(vq_hdr_name);
414                         if (hdr_mz == NULL) {
415                                 ret = -ENOMEM;
416                                 goto fail_q_alloc;
417                         }
418                 }
419         }
420
421         if (queue_type == VTNET_RQ) {
422                 size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
423                                sizeof(vq->sw_ring[0]);
424
425                 sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
426                                 RTE_CACHE_LINE_SIZE, SOCKET_ID_ANY);
427                 if (!sw_ring) {
428                         PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
429                         ret = -ENOMEM;
430                         goto fail_q_alloc;
431                 }
432
433                 vq->sw_ring = sw_ring;
434                 rxvq = &vq->rxq;
435                 rxvq->vq = vq;
436                 rxvq->port_id = dev->data->port_id;
437                 rxvq->mz = mz;
438         } else if (queue_type == VTNET_TQ) {
439                 txvq = &vq->txq;
440                 txvq->vq = vq;
441                 txvq->port_id = dev->data->port_id;
442                 txvq->mz = mz;
443                 txvq->virtio_net_hdr_mz = hdr_mz;
444                 txvq->virtio_net_hdr_mem = hdr_mz->phys_addr;
445         } else if (queue_type == VTNET_CQ) {
446                 cvq = &vq->cq;
447                 cvq->vq = vq;
448                 cvq->mz = mz;
449                 cvq->virtio_net_hdr_mz = hdr_mz;
450                 cvq->virtio_net_hdr_mem = hdr_mz->phys_addr;
451                 memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
452
453                 hw->cvq = cvq;
454         }
455
456         /* For virtio_user case (that is when dev->pci_dev is NULL), we use
457          * virtual address. And we need properly set _offset_, please see
458          * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
459          */
460         if (dev->pci_dev)
461                 vq->offset = offsetof(struct rte_mbuf, buf_physaddr);
462         else {
463                 vq->vq_ring_mem = (uintptr_t)mz->addr;
464                 vq->offset = offsetof(struct rte_mbuf, buf_addr);
465                 if (queue_type == VTNET_TQ)
466                         txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
467                 else if (queue_type == VTNET_CQ)
468                         cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
469         }
470
471         if (queue_type == VTNET_TQ) {
472                 struct virtio_tx_region *txr;
473                 unsigned int i;
474
475                 txr = hdr_mz->addr;
476                 memset(txr, 0, vq_size * sizeof(*txr));
477                 for (i = 0; i < vq_size; i++) {
478                         struct vring_desc *start_dp = txr[i].tx_indir;
479
480                         vring_desc_init(start_dp, RTE_DIM(txr[i].tx_indir));
481
482                         /* first indirect descriptor is always the tx header */
483                         start_dp->addr = txvq->virtio_net_hdr_mem
484                                 + i * sizeof(*txr)
485                                 + offsetof(struct virtio_tx_region, tx_hdr);
486
487                         start_dp->len = hw->vtnet_hdr_size;
488                         start_dp->flags = VRING_DESC_F_NEXT;
489                 }
490         }
491
492         if (hw->vtpci_ops->setup_queue(hw, vq) < 0) {
493                 PMD_INIT_LOG(ERR, "setup_queue failed");
494                 return -EINVAL;
495         }
496
497         return 0;
498
499 fail_q_alloc:
500         rte_free(sw_ring);
501         rte_memzone_free(hdr_mz);
502         rte_memzone_free(mz);
503         rte_free(vq);
504
505         return ret;
506 }
507
508 static void
509 virtio_free_queues(struct virtio_hw *hw)
510 {
511         uint16_t nr_vq = virtio_get_nr_vq(hw);
512         struct virtqueue *vq;
513         int queue_type;
514         uint16_t i;
515
516         for (i = 0; i < nr_vq; i++) {
517                 vq = hw->vqs[i];
518                 if (!vq)
519                         continue;
520
521                 queue_type = virtio_get_queue_type(hw, i);
522                 if (queue_type == VTNET_RQ) {
523                         rte_free(vq->sw_ring);
524                         rte_memzone_free(vq->rxq.mz);
525                 } else if (queue_type == VTNET_TQ) {
526                         rte_memzone_free(vq->txq.mz);
527                         rte_memzone_free(vq->txq.virtio_net_hdr_mz);
528                 } else {
529                         rte_memzone_free(vq->cq.mz);
530                         rte_memzone_free(vq->cq.virtio_net_hdr_mz);
531                 }
532
533                 rte_free(vq);
534         }
535
536         rte_free(hw->vqs);
537 }
538
539 static int
540 virtio_alloc_queues(struct rte_eth_dev *dev)
541 {
542         struct virtio_hw *hw = dev->data->dev_private;
543         uint16_t nr_vq = virtio_get_nr_vq(hw);
544         uint16_t i;
545         int ret;
546
547         hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
548         if (!hw->vqs) {
549                 PMD_INIT_LOG(ERR, "failed to allocate vqs");
550                 return -ENOMEM;
551         }
552
553         for (i = 0; i < nr_vq; i++) {
554                 ret = virtio_init_queue(dev, i);
555                 if (ret < 0) {
556                         virtio_free_queues(hw);
557                         return ret;
558                 }
559         }
560
561         return 0;
562 }
563
564 static void
565 virtio_dev_close(struct rte_eth_dev *dev)
566 {
567         struct virtio_hw *hw = dev->data->dev_private;
568
569         PMD_INIT_LOG(DEBUG, "virtio_dev_close");
570
571         /* reset the NIC */
572         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
573                 vtpci_irq_config(hw, VIRTIO_MSI_NO_VECTOR);
574         vtpci_reset(hw);
575         hw->started = 0;
576         virtio_dev_free_mbufs(dev);
577         virtio_free_queues(hw);
578 }
579
580 static void
581 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
582 {
583         struct virtio_hw *hw = dev->data->dev_private;
584         struct virtio_pmd_ctrl ctrl;
585         int dlen[1];
586         int ret;
587
588         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
589                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
590                 return;
591         }
592
593         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
594         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
595         ctrl.data[0] = 1;
596         dlen[0] = 1;
597
598         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
599         if (ret)
600                 PMD_INIT_LOG(ERR, "Failed to enable promisc");
601 }
602
603 static void
604 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
605 {
606         struct virtio_hw *hw = dev->data->dev_private;
607         struct virtio_pmd_ctrl ctrl;
608         int dlen[1];
609         int ret;
610
611         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
612                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
613                 return;
614         }
615
616         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
617         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
618         ctrl.data[0] = 0;
619         dlen[0] = 1;
620
621         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
622         if (ret)
623                 PMD_INIT_LOG(ERR, "Failed to disable promisc");
624 }
625
626 static void
627 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
628 {
629         struct virtio_hw *hw = dev->data->dev_private;
630         struct virtio_pmd_ctrl ctrl;
631         int dlen[1];
632         int ret;
633
634         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
635                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
636                 return;
637         }
638
639         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
640         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
641         ctrl.data[0] = 1;
642         dlen[0] = 1;
643
644         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
645         if (ret)
646                 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
647 }
648
649 static void
650 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
651 {
652         struct virtio_hw *hw = dev->data->dev_private;
653         struct virtio_pmd_ctrl ctrl;
654         int dlen[1];
655         int ret;
656
657         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
658                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
659                 return;
660         }
661
662         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
663         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
664         ctrl.data[0] = 0;
665         dlen[0] = 1;
666
667         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
668         if (ret)
669                 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
670 }
671
672 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
673 static int
674 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
675 {
676         struct virtio_hw *hw = dev->data->dev_private;
677         uint32_t ether_hdr_len = ETHER_HDR_LEN + VLAN_TAG_LEN +
678                                  hw->vtnet_hdr_size;
679         uint32_t frame_size = mtu + ether_hdr_len;
680
681         if (mtu < ETHER_MIN_MTU || frame_size > VIRTIO_MAX_RX_PKTLEN) {
682                 PMD_INIT_LOG(ERR, "MTU should be between %d and %d\n",
683                         ETHER_MIN_MTU, VIRTIO_MAX_RX_PKTLEN - ether_hdr_len);
684                 return -EINVAL;
685         }
686         return 0;
687 }
688
689 /*
690  * dev_ops for virtio, bare necessities for basic operation
691  */
692 static const struct eth_dev_ops virtio_eth_dev_ops = {
693         .dev_configure           = virtio_dev_configure,
694         .dev_start               = virtio_dev_start,
695         .dev_stop                = virtio_dev_stop,
696         .dev_close               = virtio_dev_close,
697         .promiscuous_enable      = virtio_dev_promiscuous_enable,
698         .promiscuous_disable     = virtio_dev_promiscuous_disable,
699         .allmulticast_enable     = virtio_dev_allmulticast_enable,
700         .allmulticast_disable    = virtio_dev_allmulticast_disable,
701         .mtu_set                 = virtio_mtu_set,
702         .dev_infos_get           = virtio_dev_info_get,
703         .stats_get               = virtio_dev_stats_get,
704         .xstats_get              = virtio_dev_xstats_get,
705         .xstats_get_names        = virtio_dev_xstats_get_names,
706         .stats_reset             = virtio_dev_stats_reset,
707         .xstats_reset            = virtio_dev_stats_reset,
708         .link_update             = virtio_dev_link_update,
709         .rx_queue_setup          = virtio_dev_rx_queue_setup,
710         .rx_queue_release        = virtio_dev_queue_release,
711         .tx_queue_setup          = virtio_dev_tx_queue_setup,
712         .tx_queue_release        = virtio_dev_queue_release,
713         /* collect stats per queue */
714         .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
715         .vlan_filter_set         = virtio_vlan_filter_set,
716         .mac_addr_add            = virtio_mac_addr_add,
717         .mac_addr_remove         = virtio_mac_addr_remove,
718         .mac_addr_set            = virtio_mac_addr_set,
719 };
720
721 static inline int
722 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
723                                 struct rte_eth_link *link)
724 {
725         struct rte_eth_link *dst = link;
726         struct rte_eth_link *src = &(dev->data->dev_link);
727
728         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
729                         *(uint64_t *)src) == 0)
730                 return -1;
731
732         return 0;
733 }
734
735 /**
736  * Atomically writes the link status information into global
737  * structure rte_eth_dev.
738  *
739  * @param dev
740  *   - Pointer to the structure rte_eth_dev to read from.
741  *   - Pointer to the buffer to be saved with the link status.
742  *
743  * @return
744  *   - On success, zero.
745  *   - On failure, negative value.
746  */
747 static inline int
748 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
749                 struct rte_eth_link *link)
750 {
751         struct rte_eth_link *dst = &(dev->data->dev_link);
752         struct rte_eth_link *src = link;
753
754         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
755                                         *(uint64_t *)src) == 0)
756                 return -1;
757
758         return 0;
759 }
760
761 static void
762 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
763 {
764         unsigned i;
765
766         for (i = 0; i < dev->data->nb_tx_queues; i++) {
767                 const struct virtnet_tx *txvq = dev->data->tx_queues[i];
768                 if (txvq == NULL)
769                         continue;
770
771                 stats->opackets += txvq->stats.packets;
772                 stats->obytes += txvq->stats.bytes;
773                 stats->oerrors += txvq->stats.errors;
774
775                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
776                         stats->q_opackets[i] = txvq->stats.packets;
777                         stats->q_obytes[i] = txvq->stats.bytes;
778                 }
779         }
780
781         for (i = 0; i < dev->data->nb_rx_queues; i++) {
782                 const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
783                 if (rxvq == NULL)
784                         continue;
785
786                 stats->ipackets += rxvq->stats.packets;
787                 stats->ibytes += rxvq->stats.bytes;
788                 stats->ierrors += rxvq->stats.errors;
789
790                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
791                         stats->q_ipackets[i] = rxvq->stats.packets;
792                         stats->q_ibytes[i] = rxvq->stats.bytes;
793                 }
794         }
795
796         stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
797 }
798
799 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
800                                        struct rte_eth_xstat_name *xstats_names,
801                                        __rte_unused unsigned limit)
802 {
803         unsigned i;
804         unsigned count = 0;
805         unsigned t;
806
807         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
808                 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
809
810         if (xstats_names != NULL) {
811                 /* Note: limit checked in rte_eth_xstats_names() */
812
813                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
814                         struct virtqueue *rxvq = dev->data->rx_queues[i];
815                         if (rxvq == NULL)
816                                 continue;
817                         for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
818                                 snprintf(xstats_names[count].name,
819                                         sizeof(xstats_names[count].name),
820                                         "rx_q%u_%s", i,
821                                         rte_virtio_rxq_stat_strings[t].name);
822                                 count++;
823                         }
824                 }
825
826                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
827                         struct virtqueue *txvq = dev->data->tx_queues[i];
828                         if (txvq == NULL)
829                                 continue;
830                         for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
831                                 snprintf(xstats_names[count].name,
832                                         sizeof(xstats_names[count].name),
833                                         "tx_q%u_%s", i,
834                                         rte_virtio_txq_stat_strings[t].name);
835                                 count++;
836                         }
837                 }
838                 return count;
839         }
840         return nstats;
841 }
842
843 static int
844 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
845                       unsigned n)
846 {
847         unsigned i;
848         unsigned count = 0;
849
850         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
851                 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
852
853         if (n < nstats)
854                 return nstats;
855
856         for (i = 0; i < dev->data->nb_rx_queues; i++) {
857                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
858
859                 if (rxvq == NULL)
860                         continue;
861
862                 unsigned t;
863
864                 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
865                         xstats[count].value = *(uint64_t *)(((char *)rxvq) +
866                                 rte_virtio_rxq_stat_strings[t].offset);
867                         count++;
868                 }
869         }
870
871         for (i = 0; i < dev->data->nb_tx_queues; i++) {
872                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
873
874                 if (txvq == NULL)
875                         continue;
876
877                 unsigned t;
878
879                 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
880                         xstats[count].value = *(uint64_t *)(((char *)txvq) +
881                                 rte_virtio_txq_stat_strings[t].offset);
882                         count++;
883                 }
884         }
885
886         return count;
887 }
888
889 static void
890 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
891 {
892         virtio_update_stats(dev, stats);
893 }
894
895 static void
896 virtio_dev_stats_reset(struct rte_eth_dev *dev)
897 {
898         unsigned int i;
899
900         for (i = 0; i < dev->data->nb_tx_queues; i++) {
901                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
902                 if (txvq == NULL)
903                         continue;
904
905                 txvq->stats.packets = 0;
906                 txvq->stats.bytes = 0;
907                 txvq->stats.errors = 0;
908                 txvq->stats.multicast = 0;
909                 txvq->stats.broadcast = 0;
910                 memset(txvq->stats.size_bins, 0,
911                        sizeof(txvq->stats.size_bins[0]) * 8);
912         }
913
914         for (i = 0; i < dev->data->nb_rx_queues; i++) {
915                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
916                 if (rxvq == NULL)
917                         continue;
918
919                 rxvq->stats.packets = 0;
920                 rxvq->stats.bytes = 0;
921                 rxvq->stats.errors = 0;
922                 rxvq->stats.multicast = 0;
923                 rxvq->stats.broadcast = 0;
924                 memset(rxvq->stats.size_bins, 0,
925                        sizeof(rxvq->stats.size_bins[0]) * 8);
926         }
927 }
928
929 static void
930 virtio_set_hwaddr(struct virtio_hw *hw)
931 {
932         vtpci_write_dev_config(hw,
933                         offsetof(struct virtio_net_config, mac),
934                         &hw->mac_addr, ETHER_ADDR_LEN);
935 }
936
937 static void
938 virtio_get_hwaddr(struct virtio_hw *hw)
939 {
940         if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
941                 vtpci_read_dev_config(hw,
942                         offsetof(struct virtio_net_config, mac),
943                         &hw->mac_addr, ETHER_ADDR_LEN);
944         } else {
945                 eth_random_addr(&hw->mac_addr[0]);
946                 virtio_set_hwaddr(hw);
947         }
948 }
949
950 static void
951 virtio_mac_table_set(struct virtio_hw *hw,
952                      const struct virtio_net_ctrl_mac *uc,
953                      const struct virtio_net_ctrl_mac *mc)
954 {
955         struct virtio_pmd_ctrl ctrl;
956         int err, len[2];
957
958         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
959                 PMD_DRV_LOG(INFO, "host does not support mac table");
960                 return;
961         }
962
963         ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
964         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
965
966         len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
967         memcpy(ctrl.data, uc, len[0]);
968
969         len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
970         memcpy(ctrl.data + len[0], mc, len[1]);
971
972         err = virtio_send_command(hw->cvq, &ctrl, len, 2);
973         if (err != 0)
974                 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
975 }
976
977 static void
978 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
979                     uint32_t index, uint32_t vmdq __rte_unused)
980 {
981         struct virtio_hw *hw = dev->data->dev_private;
982         const struct ether_addr *addrs = dev->data->mac_addrs;
983         unsigned int i;
984         struct virtio_net_ctrl_mac *uc, *mc;
985
986         if (index >= VIRTIO_MAX_MAC_ADDRS) {
987                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
988                 return;
989         }
990
991         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
992         uc->entries = 0;
993         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
994         mc->entries = 0;
995
996         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
997                 const struct ether_addr *addr
998                         = (i == index) ? mac_addr : addrs + i;
999                 struct virtio_net_ctrl_mac *tbl
1000                         = is_multicast_ether_addr(addr) ? mc : uc;
1001
1002                 memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
1003         }
1004
1005         virtio_mac_table_set(hw, uc, mc);
1006 }
1007
1008 static void
1009 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1010 {
1011         struct virtio_hw *hw = dev->data->dev_private;
1012         struct ether_addr *addrs = dev->data->mac_addrs;
1013         struct virtio_net_ctrl_mac *uc, *mc;
1014         unsigned int i;
1015
1016         if (index >= VIRTIO_MAX_MAC_ADDRS) {
1017                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1018                 return;
1019         }
1020
1021         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1022         uc->entries = 0;
1023         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1024         mc->entries = 0;
1025
1026         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1027                 struct virtio_net_ctrl_mac *tbl;
1028
1029                 if (i == index || is_zero_ether_addr(addrs + i))
1030                         continue;
1031
1032                 tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
1033                 memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
1034         }
1035
1036         virtio_mac_table_set(hw, uc, mc);
1037 }
1038
1039 static void
1040 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1041 {
1042         struct virtio_hw *hw = dev->data->dev_private;
1043
1044         memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
1045
1046         /* Use atomic update if available */
1047         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1048                 struct virtio_pmd_ctrl ctrl;
1049                 int len = ETHER_ADDR_LEN;
1050
1051                 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1052                 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1053
1054                 memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
1055                 virtio_send_command(hw->cvq, &ctrl, &len, 1);
1056         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1057                 virtio_set_hwaddr(hw);
1058 }
1059
1060 static int
1061 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1062 {
1063         struct virtio_hw *hw = dev->data->dev_private;
1064         struct virtio_pmd_ctrl ctrl;
1065         int len;
1066
1067         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1068                 return -ENOTSUP;
1069
1070         ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1071         ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1072         memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1073         len = sizeof(vlan_id);
1074
1075         return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1076 }
1077
1078 static int
1079 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1080 {
1081         uint64_t host_features;
1082
1083         /* Prepare guest_features: feature that driver wants to support */
1084         PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1085                 req_features);
1086
1087         /* Read device(host) feature bits */
1088         host_features = hw->vtpci_ops->get_features(hw);
1089         PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1090                 host_features);
1091
1092         /*
1093          * Negotiate features: Subset of device feature bits are written back
1094          * guest feature bits.
1095          */
1096         hw->guest_features = req_features;
1097         hw->guest_features = vtpci_negotiate_features(hw, host_features);
1098         PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1099                 hw->guest_features);
1100
1101         if (hw->modern) {
1102                 if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1103                         PMD_INIT_LOG(ERR,
1104                                 "VIRTIO_F_VERSION_1 features is not enabled.");
1105                         return -1;
1106                 }
1107                 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1108                 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1109                         PMD_INIT_LOG(ERR,
1110                                 "failed to set FEATURES_OK status!");
1111                         return -1;
1112                 }
1113         }
1114
1115         hw->req_guest_features = req_features;
1116
1117         return 0;
1118 }
1119
1120 /*
1121  * Process Virtio Config changed interrupt and call the callback
1122  * if link state changed.
1123  */
1124 static void
1125 virtio_interrupt_handler(__rte_unused struct rte_intr_handle *handle,
1126                          void *param)
1127 {
1128         struct rte_eth_dev *dev = param;
1129         struct virtio_hw *hw = dev->data->dev_private;
1130         uint8_t isr;
1131
1132         /* Read interrupt status which clears interrupt */
1133         isr = vtpci_isr(hw);
1134         PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1135
1136         if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0)
1137                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1138
1139         if (isr & VIRTIO_PCI_ISR_CONFIG) {
1140                 if (virtio_dev_link_update(dev, 0) == 0)
1141                         _rte_eth_dev_callback_process(dev,
1142                                                       RTE_ETH_EVENT_INTR_LSC, NULL);
1143         }
1144
1145 }
1146
1147 static void
1148 rx_func_get(struct rte_eth_dev *eth_dev)
1149 {
1150         struct virtio_hw *hw = eth_dev->data->dev_private;
1151         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF))
1152                 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1153         else
1154                 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1155 }
1156
1157 /* reset device and renegotiate features if needed */
1158 static int
1159 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1160 {
1161         struct virtio_hw *hw = eth_dev->data->dev_private;
1162         struct virtio_net_config *config;
1163         struct virtio_net_config local_config;
1164         struct rte_pci_device *pci_dev = eth_dev->pci_dev;
1165         int ret;
1166
1167         /* Reset the device although not necessary at startup */
1168         vtpci_reset(hw);
1169
1170         /* Tell the host we've noticed this device. */
1171         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1172
1173         /* Tell the host we've known how to drive the device. */
1174         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1175         if (virtio_negotiate_features(hw, req_features) < 0)
1176                 return -1;
1177
1178         /* If host does not support status then disable LSC */
1179         if (!vtpci_with_feature(hw, VIRTIO_NET_F_STATUS))
1180                 eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1181         else
1182                 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1183
1184         rte_eth_copy_pci_info(eth_dev, pci_dev);
1185
1186         rx_func_get(eth_dev);
1187
1188         /* Setting up rx_header size for the device */
1189         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1190             vtpci_with_feature(hw, VIRTIO_F_VERSION_1))
1191                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1192         else
1193                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1194
1195         /* Copy the permanent MAC address to: virtio_hw */
1196         virtio_get_hwaddr(hw);
1197         ether_addr_copy((struct ether_addr *) hw->mac_addr,
1198                         &eth_dev->data->mac_addrs[0]);
1199         PMD_INIT_LOG(DEBUG,
1200                      "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1201                      hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1202                      hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1203
1204         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1205                 config = &local_config;
1206
1207                 vtpci_read_dev_config(hw,
1208                         offsetof(struct virtio_net_config, mac),
1209                         &config->mac, sizeof(config->mac));
1210
1211                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1212                         vtpci_read_dev_config(hw,
1213                                 offsetof(struct virtio_net_config, status),
1214                                 &config->status, sizeof(config->status));
1215                 } else {
1216                         PMD_INIT_LOG(DEBUG,
1217                                      "VIRTIO_NET_F_STATUS is not supported");
1218                         config->status = 0;
1219                 }
1220
1221                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1222                         vtpci_read_dev_config(hw,
1223                                 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1224                                 &config->max_virtqueue_pairs,
1225                                 sizeof(config->max_virtqueue_pairs));
1226                 } else {
1227                         PMD_INIT_LOG(DEBUG,
1228                                      "VIRTIO_NET_F_MQ is not supported");
1229                         config->max_virtqueue_pairs = 1;
1230                 }
1231
1232                 hw->max_queue_pairs = config->max_virtqueue_pairs;
1233
1234                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1235                                 config->max_virtqueue_pairs);
1236                 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1237                 PMD_INIT_LOG(DEBUG,
1238                                 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1239                                 config->mac[0], config->mac[1],
1240                                 config->mac[2], config->mac[3],
1241                                 config->mac[4], config->mac[5]);
1242         } else {
1243                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1244                 hw->max_queue_pairs = 1;
1245         }
1246
1247         ret = virtio_alloc_queues(eth_dev);
1248         if (ret < 0)
1249                 return ret;
1250
1251         if (pci_dev)
1252                 PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1253                         eth_dev->data->port_id, pci_dev->id.vendor_id,
1254                         pci_dev->id.device_id);
1255
1256         return 0;
1257 }
1258
1259 /*
1260  * This function is based on probe() function in virtio_pci.c
1261  * It returns 0 on success.
1262  */
1263 int
1264 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1265 {
1266         struct virtio_hw *hw = eth_dev->data->dev_private;
1267         struct rte_pci_device *pci_dev;
1268         uint32_t dev_flags = RTE_ETH_DEV_DETACHABLE;
1269         int ret;
1270
1271         RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1272
1273         eth_dev->dev_ops = &virtio_eth_dev_ops;
1274         eth_dev->tx_pkt_burst = &virtio_xmit_pkts;
1275
1276         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1277                 rx_func_get(eth_dev);
1278                 return 0;
1279         }
1280
1281         /* Allocate memory for storing MAC addresses */
1282         eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
1283         if (eth_dev->data->mac_addrs == NULL) {
1284                 PMD_INIT_LOG(ERR,
1285                         "Failed to allocate %d bytes needed to store MAC addresses",
1286                         VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
1287                 return -ENOMEM;
1288         }
1289
1290         pci_dev = eth_dev->pci_dev;
1291
1292         if (pci_dev) {
1293                 ret = vtpci_init(pci_dev, hw, &dev_flags);
1294                 if (ret)
1295                         return ret;
1296         }
1297
1298         eth_dev->data->dev_flags = dev_flags;
1299
1300         /* reset device and negotiate default features */
1301         ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1302         if (ret < 0)
1303                 return ret;
1304
1305         /* Setup interrupt callback  */
1306         if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1307                 rte_intr_callback_register(&pci_dev->intr_handle,
1308                         virtio_interrupt_handler, eth_dev);
1309
1310         return 0;
1311 }
1312
1313 static int
1314 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1315 {
1316         struct rte_pci_device *pci_dev;
1317         struct virtio_hw *hw = eth_dev->data->dev_private;
1318
1319         PMD_INIT_FUNC_TRACE();
1320
1321         if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1322                 return -EPERM;
1323
1324         if (hw->started == 1) {
1325                 virtio_dev_stop(eth_dev);
1326                 virtio_dev_close(eth_dev);
1327         }
1328         pci_dev = eth_dev->pci_dev;
1329
1330         eth_dev->dev_ops = NULL;
1331         eth_dev->tx_pkt_burst = NULL;
1332         eth_dev->rx_pkt_burst = NULL;
1333
1334         rte_free(eth_dev->data->mac_addrs);
1335         eth_dev->data->mac_addrs = NULL;
1336
1337         /* reset interrupt callback  */
1338         if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1339                 rte_intr_callback_unregister(&pci_dev->intr_handle,
1340                                                 virtio_interrupt_handler,
1341                                                 eth_dev);
1342         rte_eal_pci_unmap_device(pci_dev);
1343
1344         PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1345
1346         return 0;
1347 }
1348
1349 static struct eth_driver rte_virtio_pmd = {
1350         .pci_drv = {
1351                 .driver = {
1352                         .name = "net_virtio",
1353                 },
1354                 .id_table = pci_id_virtio_map,
1355                 .drv_flags = RTE_PCI_DRV_DETACHABLE,
1356                 .probe = rte_eth_dev_pci_probe,
1357                 .remove = rte_eth_dev_pci_remove,
1358         },
1359         .eth_dev_init = eth_virtio_dev_init,
1360         .eth_dev_uninit = eth_virtio_dev_uninit,
1361         .dev_private_size = sizeof(struct virtio_hw),
1362 };
1363
1364 RTE_INIT(rte_virtio_pmd_init);
1365 static void
1366 rte_virtio_pmd_init(void)
1367 {
1368         if (rte_eal_iopl_init() != 0) {
1369                 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1370                 return;
1371         }
1372
1373         rte_eal_pci_register(&rte_virtio_pmd.pci_drv);
1374 }
1375
1376 /*
1377  * Configure virtio device
1378  * It returns 0 on success.
1379  */
1380 static int
1381 virtio_dev_configure(struct rte_eth_dev *dev)
1382 {
1383         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1384         struct virtio_hw *hw = dev->data->dev_private;
1385         uint64_t req_features;
1386         int ret;
1387
1388         PMD_INIT_LOG(DEBUG, "configure");
1389         req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
1390         if (rxmode->hw_ip_checksum)
1391                 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
1392         if (rxmode->enable_lro)
1393                 req_features |=
1394                         (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
1395                         (1ULL << VIRTIO_NET_F_GUEST_TSO6);
1396
1397         /* if request features changed, reinit the device */
1398         if (req_features != hw->req_guest_features) {
1399                 ret = virtio_init_device(dev, req_features);
1400                 if (ret < 0)
1401                         return ret;
1402         }
1403
1404         if (rxmode->hw_ip_checksum &&
1405                 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
1406                 PMD_DRV_LOG(NOTICE,
1407                         "rx ip checksum not available on this host");
1408                 return -ENOTSUP;
1409         }
1410
1411         if (rxmode->enable_lro &&
1412                 (!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
1413                         !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4))) {
1414                 PMD_DRV_LOG(NOTICE,
1415                         "lro not available on this host");
1416                 return -ENOTSUP;
1417         }
1418
1419         /* start control queue */
1420         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
1421                 virtio_dev_cq_start(dev);
1422
1423         hw->vlan_strip = rxmode->hw_vlan_strip;
1424
1425         if (rxmode->hw_vlan_filter
1426             && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1427                 PMD_DRV_LOG(NOTICE,
1428                             "vlan filtering not available on this host");
1429                 return -ENOTSUP;
1430         }
1431
1432         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1433                 if (vtpci_irq_config(hw, 0) == VIRTIO_MSI_NO_VECTOR) {
1434                         PMD_DRV_LOG(ERR, "failed to set config vector");
1435                         return -EBUSY;
1436                 }
1437
1438         return 0;
1439 }
1440
1441
1442 static int
1443 virtio_dev_start(struct rte_eth_dev *dev)
1444 {
1445         uint16_t nb_queues, i;
1446         struct virtio_hw *hw = dev->data->dev_private;
1447         struct virtnet_rx *rxvq;
1448         struct virtnet_tx *txvq __rte_unused;
1449
1450         /* check if lsc interrupt feature is enabled */
1451         if (dev->data->dev_conf.intr_conf.lsc) {
1452                 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1453                         PMD_DRV_LOG(ERR, "link status not supported by host");
1454                         return -ENOTSUP;
1455                 }
1456
1457                 if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0) {
1458                         PMD_DRV_LOG(ERR, "interrupt enable failed");
1459                         return -EIO;
1460                 }
1461         }
1462
1463         /* Initialize Link state */
1464         virtio_dev_link_update(dev, 0);
1465
1466         /* On restart after stop do not touch queues */
1467         if (hw->started)
1468                 return 0;
1469
1470         /* Do final configuration before rx/tx engine starts */
1471         virtio_dev_rxtx_start(dev);
1472         vtpci_reinit_complete(hw);
1473
1474         hw->started = 1;
1475
1476         /*Notify the backend
1477          *Otherwise the tap backend might already stop its queue due to fullness.
1478          *vhost backend will have no chance to be waked up
1479          */
1480         nb_queues = dev->data->nb_rx_queues;
1481         if (nb_queues > 1) {
1482                 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1483                         return -EINVAL;
1484         }
1485
1486         PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1487
1488         for (i = 0; i < nb_queues; i++) {
1489                 rxvq = dev->data->rx_queues[i];
1490                 virtqueue_notify(rxvq->vq);
1491         }
1492
1493         PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1494
1495         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1496                 rxvq = dev->data->rx_queues[i];
1497                 VIRTQUEUE_DUMP(rxvq->vq);
1498         }
1499
1500         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1501                 txvq = dev->data->tx_queues[i];
1502                 VIRTQUEUE_DUMP(txvq->vq);
1503         }
1504
1505         return 0;
1506 }
1507
1508 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1509 {
1510         struct rte_mbuf *buf;
1511         int i, mbuf_num = 0;
1512
1513         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1514                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1515
1516                 PMD_INIT_LOG(DEBUG,
1517                              "Before freeing rxq[%d] used and unused buf", i);
1518                 VIRTQUEUE_DUMP(rxvq->vq);
1519
1520                 PMD_INIT_LOG(DEBUG, "rx_queues[%d]=%p", i, rxvq);
1521                 while ((buf = virtqueue_detatch_unused(rxvq->vq)) != NULL) {
1522                         rte_pktmbuf_free(buf);
1523                         mbuf_num++;
1524                 }
1525
1526                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1527                 PMD_INIT_LOG(DEBUG,
1528                              "After freeing rxq[%d] used and unused buf", i);
1529                 VIRTQUEUE_DUMP(rxvq->vq);
1530         }
1531
1532         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1533                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1534
1535                 PMD_INIT_LOG(DEBUG,
1536                              "Before freeing txq[%d] used and unused bufs",
1537                              i);
1538                 VIRTQUEUE_DUMP(txvq->vq);
1539
1540                 mbuf_num = 0;
1541                 while ((buf = virtqueue_detatch_unused(txvq->vq)) != NULL) {
1542                         rte_pktmbuf_free(buf);
1543                         mbuf_num++;
1544                 }
1545
1546                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1547                 PMD_INIT_LOG(DEBUG,
1548                              "After freeing txq[%d] used and unused buf", i);
1549                 VIRTQUEUE_DUMP(txvq->vq);
1550         }
1551 }
1552
1553 /*
1554  * Stop device: disable interrupt and mark link down
1555  */
1556 static void
1557 virtio_dev_stop(struct rte_eth_dev *dev)
1558 {
1559         struct rte_eth_link link;
1560
1561         PMD_INIT_LOG(DEBUG, "stop");
1562
1563         if (dev->data->dev_conf.intr_conf.lsc)
1564                 rte_intr_disable(&dev->pci_dev->intr_handle);
1565
1566         memset(&link, 0, sizeof(link));
1567         virtio_dev_atomic_write_link_status(dev, &link);
1568 }
1569
1570 static int
1571 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1572 {
1573         struct rte_eth_link link, old;
1574         uint16_t status;
1575         struct virtio_hw *hw = dev->data->dev_private;
1576         memset(&link, 0, sizeof(link));
1577         virtio_dev_atomic_read_link_status(dev, &link);
1578         old = link;
1579         link.link_duplex = ETH_LINK_FULL_DUPLEX;
1580         link.link_speed  = SPEED_10G;
1581
1582         if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1583                 PMD_INIT_LOG(DEBUG, "Get link status from hw");
1584                 vtpci_read_dev_config(hw,
1585                                 offsetof(struct virtio_net_config, status),
1586                                 &status, sizeof(status));
1587                 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1588                         link.link_status = ETH_LINK_DOWN;
1589                         PMD_INIT_LOG(DEBUG, "Port %d is down",
1590                                      dev->data->port_id);
1591                 } else {
1592                         link.link_status = ETH_LINK_UP;
1593                         PMD_INIT_LOG(DEBUG, "Port %d is up",
1594                                      dev->data->port_id);
1595                 }
1596         } else {
1597                 link.link_status = ETH_LINK_UP;
1598         }
1599         virtio_dev_atomic_write_link_status(dev, &link);
1600
1601         return (old.link_status == link.link_status) ? -1 : 0;
1602 }
1603
1604 static void
1605 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1606 {
1607         uint64_t tso_mask;
1608         struct virtio_hw *hw = dev->data->dev_private;
1609
1610         if (dev->pci_dev)
1611                 dev_info->driver_name = dev->driver->pci_drv.driver.name;
1612         else
1613                 dev_info->driver_name = "virtio_user PMD";
1614         dev_info->max_rx_queues =
1615                 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
1616         dev_info->max_tx_queues =
1617                 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
1618         dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1619         dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1620         dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1621         dev_info->default_txconf = (struct rte_eth_txconf) {
1622                 .txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
1623         };
1624         dev_info->rx_offload_capa =
1625                 DEV_RX_OFFLOAD_TCP_CKSUM |
1626                 DEV_RX_OFFLOAD_UDP_CKSUM |
1627                 DEV_RX_OFFLOAD_TCP_LRO;
1628         dev_info->tx_offload_capa = 0;
1629
1630         if (hw->guest_features & (1ULL << VIRTIO_NET_F_CSUM)) {
1631                 dev_info->tx_offload_capa |=
1632                         DEV_TX_OFFLOAD_UDP_CKSUM |
1633                         DEV_TX_OFFLOAD_TCP_CKSUM;
1634         }
1635
1636         tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
1637                 (1ULL << VIRTIO_NET_F_HOST_TSO6);
1638         if ((hw->guest_features & tso_mask) == tso_mask)
1639                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
1640 }
1641
1642 /*
1643  * It enables testpmd to collect per queue stats.
1644  */
1645 static int
1646 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
1647 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
1648 __rte_unused uint8_t is_rx)
1649 {
1650         return 0;
1651 }
1652
1653 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
1654 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);