5c8afb5bc0a078d993f98c5a699f4a9f95b543ef
[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 void
312 virtio_init_vring(struct virtqueue *vq)
313 {
314         int size = vq->vq_nentries;
315         struct vring *vr = &vq->vq_ring;
316         uint8_t *ring_mem = vq->vq_ring_virt_mem;
317
318         PMD_INIT_FUNC_TRACE();
319
320         /*
321          * Reinitialise since virtio port might have been stopped and restarted
322          */
323         memset(ring_mem, 0, vq->vq_ring_size);
324         vring_init(vr, size, ring_mem, VIRTIO_PCI_VRING_ALIGN);
325         vq->vq_used_cons_idx = 0;
326         vq->vq_desc_head_idx = 0;
327         vq->vq_avail_idx = 0;
328         vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
329         vq->vq_free_cnt = vq->vq_nentries;
330         memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
331
332         vring_desc_init(vr->desc, size);
333
334         /*
335          * Disable device(host) interrupting guest
336          */
337         virtqueue_disable_intr(vq);
338 }
339
340 static int
341 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
342 {
343         char vq_name[VIRTQUEUE_MAX_NAME_SZ];
344         char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
345         const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
346         unsigned int vq_size, size;
347         struct virtio_hw *hw = dev->data->dev_private;
348         struct virtnet_rx *rxvq = NULL;
349         struct virtnet_tx *txvq = NULL;
350         struct virtnet_ctl *cvq = NULL;
351         struct virtqueue *vq;
352         size_t sz_hdr_mz = 0;
353         void *sw_ring = NULL;
354         int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
355         int ret;
356
357         PMD_INIT_LOG(DEBUG, "setting up queue: %u", vtpci_queue_idx);
358
359         /*
360          * Read the virtqueue size from the Queue Size field
361          * Always power of 2 and if 0 virtqueue does not exist
362          */
363         vq_size = hw->vtpci_ops->get_queue_num(hw, vtpci_queue_idx);
364         PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
365         if (vq_size == 0) {
366                 PMD_INIT_LOG(ERR, "virtqueue does not exist");
367                 return -EINVAL;
368         }
369
370         if (!rte_is_power_of_2(vq_size)) {
371                 PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
372                 return -EINVAL;
373         }
374
375         snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
376                  dev->data->port_id, vtpci_queue_idx);
377
378         size = RTE_ALIGN_CEIL(sizeof(*vq) +
379                                 vq_size * sizeof(struct vq_desc_extra),
380                                 RTE_CACHE_LINE_SIZE);
381         if (queue_type == VTNET_TQ) {
382                 /*
383                  * For each xmit packet, allocate a virtio_net_hdr
384                  * and indirect ring elements
385                  */
386                 sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
387         } else if (queue_type == VTNET_CQ) {
388                 /* Allocate a page for control vq command, data and status */
389                 sz_hdr_mz = PAGE_SIZE;
390         }
391
392         vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
393                                 SOCKET_ID_ANY);
394         if (vq == NULL) {
395                 PMD_INIT_LOG(ERR, "can not allocate vq");
396                 return -ENOMEM;
397         }
398         hw->vqs[vtpci_queue_idx] = vq;
399
400         vq->hw = hw;
401         vq->vq_queue_index = vtpci_queue_idx;
402         vq->vq_nentries = vq_size;
403
404         /*
405          * Reserve a memzone for vring elements
406          */
407         size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
408         vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
409         PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
410                      size, vq->vq_ring_size);
411
412         mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
413                                          SOCKET_ID_ANY,
414                                          0, VIRTIO_PCI_VRING_ALIGN);
415         if (mz == NULL) {
416                 if (rte_errno == EEXIST)
417                         mz = rte_memzone_lookup(vq_name);
418                 if (mz == NULL) {
419                         ret = -ENOMEM;
420                         goto fail_q_alloc;
421                 }
422         }
423
424         memset(mz->addr, 0, sizeof(mz->len));
425
426         vq->vq_ring_mem = mz->phys_addr;
427         vq->vq_ring_virt_mem = mz->addr;
428         PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%" PRIx64,
429                      (uint64_t)mz->phys_addr);
430         PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
431                      (uint64_t)(uintptr_t)mz->addr);
432
433         virtio_init_vring(vq);
434
435         if (sz_hdr_mz) {
436                 snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
437                          dev->data->port_id, vtpci_queue_idx);
438                 hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
439                                                      SOCKET_ID_ANY, 0,
440                                                      RTE_CACHE_LINE_SIZE);
441                 if (hdr_mz == NULL) {
442                         if (rte_errno == EEXIST)
443                                 hdr_mz = rte_memzone_lookup(vq_hdr_name);
444                         if (hdr_mz == NULL) {
445                                 ret = -ENOMEM;
446                                 goto fail_q_alloc;
447                         }
448                 }
449         }
450
451         if (queue_type == VTNET_RQ) {
452                 size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
453                                sizeof(vq->sw_ring[0]);
454
455                 sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
456                                 RTE_CACHE_LINE_SIZE, SOCKET_ID_ANY);
457                 if (!sw_ring) {
458                         PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
459                         ret = -ENOMEM;
460                         goto fail_q_alloc;
461                 }
462
463                 vq->sw_ring = sw_ring;
464                 rxvq = &vq->rxq;
465                 rxvq->vq = vq;
466                 rxvq->port_id = dev->data->port_id;
467                 rxvq->mz = mz;
468         } else if (queue_type == VTNET_TQ) {
469                 txvq = &vq->txq;
470                 txvq->vq = vq;
471                 txvq->port_id = dev->data->port_id;
472                 txvq->mz = mz;
473                 txvq->virtio_net_hdr_mz = hdr_mz;
474                 txvq->virtio_net_hdr_mem = hdr_mz->phys_addr;
475         } else if (queue_type == VTNET_CQ) {
476                 cvq = &vq->cq;
477                 cvq->vq = vq;
478                 cvq->mz = mz;
479                 cvq->virtio_net_hdr_mz = hdr_mz;
480                 cvq->virtio_net_hdr_mem = hdr_mz->phys_addr;
481                 memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
482
483                 hw->cvq = cvq;
484         }
485
486         /* For virtio_user case (that is when dev->pci_dev is NULL), we use
487          * virtual address. And we need properly set _offset_, please see
488          * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
489          */
490         if (dev->pci_dev)
491                 vq->offset = offsetof(struct rte_mbuf, buf_physaddr);
492         else {
493                 vq->vq_ring_mem = (uintptr_t)mz->addr;
494                 vq->offset = offsetof(struct rte_mbuf, buf_addr);
495                 if (queue_type == VTNET_TQ)
496                         txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
497                 else if (queue_type == VTNET_CQ)
498                         cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
499         }
500
501         if (queue_type == VTNET_TQ) {
502                 struct virtio_tx_region *txr;
503                 unsigned int i;
504
505                 txr = hdr_mz->addr;
506                 memset(txr, 0, vq_size * sizeof(*txr));
507                 for (i = 0; i < vq_size; i++) {
508                         struct vring_desc *start_dp = txr[i].tx_indir;
509
510                         vring_desc_init(start_dp, RTE_DIM(txr[i].tx_indir));
511
512                         /* first indirect descriptor is always the tx header */
513                         start_dp->addr = txvq->virtio_net_hdr_mem
514                                 + i * sizeof(*txr)
515                                 + offsetof(struct virtio_tx_region, tx_hdr);
516
517                         start_dp->len = hw->vtnet_hdr_size;
518                         start_dp->flags = VRING_DESC_F_NEXT;
519                 }
520         }
521
522         if (hw->vtpci_ops->setup_queue(hw, vq) < 0) {
523                 PMD_INIT_LOG(ERR, "setup_queue failed");
524                 return -EINVAL;
525         }
526
527         return 0;
528
529 fail_q_alloc:
530         rte_free(sw_ring);
531         rte_memzone_free(hdr_mz);
532         rte_memzone_free(mz);
533         rte_free(vq);
534
535         return ret;
536 }
537
538 static void
539 virtio_free_queues(struct virtio_hw *hw)
540 {
541         uint16_t nr_vq = virtio_get_nr_vq(hw);
542         struct virtqueue *vq;
543         int queue_type;
544         uint16_t i;
545
546         for (i = 0; i < nr_vq; i++) {
547                 vq = hw->vqs[i];
548                 if (!vq)
549                         continue;
550
551                 queue_type = virtio_get_queue_type(hw, i);
552                 if (queue_type == VTNET_RQ) {
553                         rte_free(vq->sw_ring);
554                         rte_memzone_free(vq->rxq.mz);
555                 } else if (queue_type == VTNET_TQ) {
556                         rte_memzone_free(vq->txq.mz);
557                         rte_memzone_free(vq->txq.virtio_net_hdr_mz);
558                 } else {
559                         rte_memzone_free(vq->cq.mz);
560                         rte_memzone_free(vq->cq.virtio_net_hdr_mz);
561                 }
562
563                 rte_free(vq);
564         }
565
566         rte_free(hw->vqs);
567 }
568
569 static int
570 virtio_alloc_queues(struct rte_eth_dev *dev)
571 {
572         struct virtio_hw *hw = dev->data->dev_private;
573         uint16_t nr_vq = virtio_get_nr_vq(hw);
574         uint16_t i;
575         int ret;
576
577         hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
578         if (!hw->vqs) {
579                 PMD_INIT_LOG(ERR, "failed to allocate vqs");
580                 return -ENOMEM;
581         }
582
583         for (i = 0; i < nr_vq; i++) {
584                 ret = virtio_init_queue(dev, i);
585                 if (ret < 0) {
586                         virtio_free_queues(hw);
587                         return ret;
588                 }
589         }
590
591         return 0;
592 }
593
594 static void
595 virtio_dev_close(struct rte_eth_dev *dev)
596 {
597         struct virtio_hw *hw = dev->data->dev_private;
598
599         PMD_INIT_LOG(DEBUG, "virtio_dev_close");
600
601         /* reset the NIC */
602         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
603                 vtpci_irq_config(hw, VIRTIO_MSI_NO_VECTOR);
604         vtpci_reset(hw);
605         virtio_dev_free_mbufs(dev);
606         virtio_free_queues(hw);
607 }
608
609 static void
610 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
611 {
612         struct virtio_hw *hw = dev->data->dev_private;
613         struct virtio_pmd_ctrl ctrl;
614         int dlen[1];
615         int ret;
616
617         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
618                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
619                 return;
620         }
621
622         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
623         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
624         ctrl.data[0] = 1;
625         dlen[0] = 1;
626
627         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
628         if (ret)
629                 PMD_INIT_LOG(ERR, "Failed to enable promisc");
630 }
631
632 static void
633 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
634 {
635         struct virtio_hw *hw = dev->data->dev_private;
636         struct virtio_pmd_ctrl ctrl;
637         int dlen[1];
638         int ret;
639
640         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
641                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
642                 return;
643         }
644
645         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
646         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
647         ctrl.data[0] = 0;
648         dlen[0] = 1;
649
650         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
651         if (ret)
652                 PMD_INIT_LOG(ERR, "Failed to disable promisc");
653 }
654
655 static void
656 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
657 {
658         struct virtio_hw *hw = dev->data->dev_private;
659         struct virtio_pmd_ctrl ctrl;
660         int dlen[1];
661         int ret;
662
663         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
664                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
665                 return;
666         }
667
668         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
669         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
670         ctrl.data[0] = 1;
671         dlen[0] = 1;
672
673         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
674         if (ret)
675                 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
676 }
677
678 static void
679 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
680 {
681         struct virtio_hw *hw = dev->data->dev_private;
682         struct virtio_pmd_ctrl ctrl;
683         int dlen[1];
684         int ret;
685
686         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
687                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
688                 return;
689         }
690
691         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
692         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
693         ctrl.data[0] = 0;
694         dlen[0] = 1;
695
696         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
697         if (ret)
698                 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
699 }
700
701 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
702 static int
703 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
704 {
705         struct virtio_hw *hw = dev->data->dev_private;
706         uint32_t ether_hdr_len = ETHER_HDR_LEN + VLAN_TAG_LEN +
707                                  hw->vtnet_hdr_size;
708         uint32_t frame_size = mtu + ether_hdr_len;
709
710         if (mtu < ETHER_MIN_MTU || frame_size > VIRTIO_MAX_RX_PKTLEN) {
711                 PMD_INIT_LOG(ERR, "MTU should be between %d and %d\n",
712                         ETHER_MIN_MTU, VIRTIO_MAX_RX_PKTLEN - ether_hdr_len);
713                 return -EINVAL;
714         }
715         return 0;
716 }
717
718 /*
719  * dev_ops for virtio, bare necessities for basic operation
720  */
721 static const struct eth_dev_ops virtio_eth_dev_ops = {
722         .dev_configure           = virtio_dev_configure,
723         .dev_start               = virtio_dev_start,
724         .dev_stop                = virtio_dev_stop,
725         .dev_close               = virtio_dev_close,
726         .promiscuous_enable      = virtio_dev_promiscuous_enable,
727         .promiscuous_disable     = virtio_dev_promiscuous_disable,
728         .allmulticast_enable     = virtio_dev_allmulticast_enable,
729         .allmulticast_disable    = virtio_dev_allmulticast_disable,
730         .mtu_set                 = virtio_mtu_set,
731         .dev_infos_get           = virtio_dev_info_get,
732         .stats_get               = virtio_dev_stats_get,
733         .xstats_get              = virtio_dev_xstats_get,
734         .xstats_get_names        = virtio_dev_xstats_get_names,
735         .stats_reset             = virtio_dev_stats_reset,
736         .xstats_reset            = virtio_dev_stats_reset,
737         .link_update             = virtio_dev_link_update,
738         .rx_queue_setup          = virtio_dev_rx_queue_setup,
739         .rx_queue_release        = virtio_dev_queue_release,
740         .tx_queue_setup          = virtio_dev_tx_queue_setup,
741         .tx_queue_release        = virtio_dev_queue_release,
742         /* collect stats per queue */
743         .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
744         .vlan_filter_set         = virtio_vlan_filter_set,
745         .mac_addr_add            = virtio_mac_addr_add,
746         .mac_addr_remove         = virtio_mac_addr_remove,
747         .mac_addr_set            = virtio_mac_addr_set,
748 };
749
750 static inline int
751 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
752                                 struct rte_eth_link *link)
753 {
754         struct rte_eth_link *dst = link;
755         struct rte_eth_link *src = &(dev->data->dev_link);
756
757         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
758                         *(uint64_t *)src) == 0)
759                 return -1;
760
761         return 0;
762 }
763
764 /**
765  * Atomically writes the link status information into global
766  * structure rte_eth_dev.
767  *
768  * @param dev
769  *   - Pointer to the structure rte_eth_dev to read from.
770  *   - Pointer to the buffer to be saved with the link status.
771  *
772  * @return
773  *   - On success, zero.
774  *   - On failure, negative value.
775  */
776 static inline int
777 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
778                 struct rte_eth_link *link)
779 {
780         struct rte_eth_link *dst = &(dev->data->dev_link);
781         struct rte_eth_link *src = link;
782
783         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
784                                         *(uint64_t *)src) == 0)
785                 return -1;
786
787         return 0;
788 }
789
790 static void
791 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
792 {
793         unsigned i;
794
795         for (i = 0; i < dev->data->nb_tx_queues; i++) {
796                 const struct virtnet_tx *txvq = dev->data->tx_queues[i];
797                 if (txvq == NULL)
798                         continue;
799
800                 stats->opackets += txvq->stats.packets;
801                 stats->obytes += txvq->stats.bytes;
802                 stats->oerrors += txvq->stats.errors;
803
804                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
805                         stats->q_opackets[i] = txvq->stats.packets;
806                         stats->q_obytes[i] = txvq->stats.bytes;
807                 }
808         }
809
810         for (i = 0; i < dev->data->nb_rx_queues; i++) {
811                 const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
812                 if (rxvq == NULL)
813                         continue;
814
815                 stats->ipackets += rxvq->stats.packets;
816                 stats->ibytes += rxvq->stats.bytes;
817                 stats->ierrors += rxvq->stats.errors;
818
819                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
820                         stats->q_ipackets[i] = rxvq->stats.packets;
821                         stats->q_ibytes[i] = rxvq->stats.bytes;
822                 }
823         }
824
825         stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
826 }
827
828 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
829                                        struct rte_eth_xstat_name *xstats_names,
830                                        __rte_unused unsigned limit)
831 {
832         unsigned i;
833         unsigned count = 0;
834         unsigned t;
835
836         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
837                 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
838
839         if (xstats_names != NULL) {
840                 /* Note: limit checked in rte_eth_xstats_names() */
841
842                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
843                         struct virtqueue *rxvq = dev->data->rx_queues[i];
844                         if (rxvq == NULL)
845                                 continue;
846                         for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
847                                 snprintf(xstats_names[count].name,
848                                         sizeof(xstats_names[count].name),
849                                         "rx_q%u_%s", i,
850                                         rte_virtio_rxq_stat_strings[t].name);
851                                 count++;
852                         }
853                 }
854
855                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
856                         struct virtqueue *txvq = dev->data->tx_queues[i];
857                         if (txvq == NULL)
858                                 continue;
859                         for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
860                                 snprintf(xstats_names[count].name,
861                                         sizeof(xstats_names[count].name),
862                                         "tx_q%u_%s", i,
863                                         rte_virtio_txq_stat_strings[t].name);
864                                 count++;
865                         }
866                 }
867                 return count;
868         }
869         return nstats;
870 }
871
872 static int
873 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
874                       unsigned n)
875 {
876         unsigned i;
877         unsigned count = 0;
878
879         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
880                 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
881
882         if (n < nstats)
883                 return nstats;
884
885         for (i = 0; i < dev->data->nb_rx_queues; i++) {
886                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
887
888                 if (rxvq == NULL)
889                         continue;
890
891                 unsigned t;
892
893                 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
894                         xstats[count].value = *(uint64_t *)(((char *)rxvq) +
895                                 rte_virtio_rxq_stat_strings[t].offset);
896                         count++;
897                 }
898         }
899
900         for (i = 0; i < dev->data->nb_tx_queues; i++) {
901                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
902
903                 if (txvq == NULL)
904                         continue;
905
906                 unsigned t;
907
908                 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
909                         xstats[count].value = *(uint64_t *)(((char *)txvq) +
910                                 rte_virtio_txq_stat_strings[t].offset);
911                         count++;
912                 }
913         }
914
915         return count;
916 }
917
918 static void
919 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
920 {
921         virtio_update_stats(dev, stats);
922 }
923
924 static void
925 virtio_dev_stats_reset(struct rte_eth_dev *dev)
926 {
927         unsigned int i;
928
929         for (i = 0; i < dev->data->nb_tx_queues; i++) {
930                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
931                 if (txvq == NULL)
932                         continue;
933
934                 txvq->stats.packets = 0;
935                 txvq->stats.bytes = 0;
936                 txvq->stats.errors = 0;
937                 txvq->stats.multicast = 0;
938                 txvq->stats.broadcast = 0;
939                 memset(txvq->stats.size_bins, 0,
940                        sizeof(txvq->stats.size_bins[0]) * 8);
941         }
942
943         for (i = 0; i < dev->data->nb_rx_queues; i++) {
944                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
945                 if (rxvq == NULL)
946                         continue;
947
948                 rxvq->stats.packets = 0;
949                 rxvq->stats.bytes = 0;
950                 rxvq->stats.errors = 0;
951                 rxvq->stats.multicast = 0;
952                 rxvq->stats.broadcast = 0;
953                 memset(rxvq->stats.size_bins, 0,
954                        sizeof(rxvq->stats.size_bins[0]) * 8);
955         }
956 }
957
958 static void
959 virtio_set_hwaddr(struct virtio_hw *hw)
960 {
961         vtpci_write_dev_config(hw,
962                         offsetof(struct virtio_net_config, mac),
963                         &hw->mac_addr, ETHER_ADDR_LEN);
964 }
965
966 static void
967 virtio_get_hwaddr(struct virtio_hw *hw)
968 {
969         if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
970                 vtpci_read_dev_config(hw,
971                         offsetof(struct virtio_net_config, mac),
972                         &hw->mac_addr, ETHER_ADDR_LEN);
973         } else {
974                 eth_random_addr(&hw->mac_addr[0]);
975                 virtio_set_hwaddr(hw);
976         }
977 }
978
979 static void
980 virtio_mac_table_set(struct virtio_hw *hw,
981                      const struct virtio_net_ctrl_mac *uc,
982                      const struct virtio_net_ctrl_mac *mc)
983 {
984         struct virtio_pmd_ctrl ctrl;
985         int err, len[2];
986
987         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
988                 PMD_DRV_LOG(INFO, "host does not support mac table");
989                 return;
990         }
991
992         ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
993         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
994
995         len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
996         memcpy(ctrl.data, uc, len[0]);
997
998         len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
999         memcpy(ctrl.data + len[0], mc, len[1]);
1000
1001         err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1002         if (err != 0)
1003                 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1004 }
1005
1006 static void
1007 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
1008                     uint32_t index, uint32_t vmdq __rte_unused)
1009 {
1010         struct virtio_hw *hw = dev->data->dev_private;
1011         const struct ether_addr *addrs = dev->data->mac_addrs;
1012         unsigned int i;
1013         struct virtio_net_ctrl_mac *uc, *mc;
1014
1015         if (index >= VIRTIO_MAX_MAC_ADDRS) {
1016                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1017                 return;
1018         }
1019
1020         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1021         uc->entries = 0;
1022         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1023         mc->entries = 0;
1024
1025         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1026                 const struct ether_addr *addr
1027                         = (i == index) ? mac_addr : addrs + i;
1028                 struct virtio_net_ctrl_mac *tbl
1029                         = is_multicast_ether_addr(addr) ? mc : uc;
1030
1031                 memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
1032         }
1033
1034         virtio_mac_table_set(hw, uc, mc);
1035 }
1036
1037 static void
1038 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1039 {
1040         struct virtio_hw *hw = dev->data->dev_private;
1041         struct ether_addr *addrs = dev->data->mac_addrs;
1042         struct virtio_net_ctrl_mac *uc, *mc;
1043         unsigned int i;
1044
1045         if (index >= VIRTIO_MAX_MAC_ADDRS) {
1046                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1047                 return;
1048         }
1049
1050         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1051         uc->entries = 0;
1052         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1053         mc->entries = 0;
1054
1055         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1056                 struct virtio_net_ctrl_mac *tbl;
1057
1058                 if (i == index || is_zero_ether_addr(addrs + i))
1059                         continue;
1060
1061                 tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
1062                 memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
1063         }
1064
1065         virtio_mac_table_set(hw, uc, mc);
1066 }
1067
1068 static void
1069 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1070 {
1071         struct virtio_hw *hw = dev->data->dev_private;
1072
1073         memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
1074
1075         /* Use atomic update if available */
1076         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1077                 struct virtio_pmd_ctrl ctrl;
1078                 int len = ETHER_ADDR_LEN;
1079
1080                 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1081                 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1082
1083                 memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
1084                 virtio_send_command(hw->cvq, &ctrl, &len, 1);
1085         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1086                 virtio_set_hwaddr(hw);
1087 }
1088
1089 static int
1090 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1091 {
1092         struct virtio_hw *hw = dev->data->dev_private;
1093         struct virtio_pmd_ctrl ctrl;
1094         int len;
1095
1096         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1097                 return -ENOTSUP;
1098
1099         ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1100         ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1101         memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1102         len = sizeof(vlan_id);
1103
1104         return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1105 }
1106
1107 static int
1108 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1109 {
1110         uint64_t host_features;
1111
1112         /* Prepare guest_features: feature that driver wants to support */
1113         PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1114                 req_features);
1115
1116         /* Read device(host) feature bits */
1117         host_features = hw->vtpci_ops->get_features(hw);
1118         PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1119                 host_features);
1120
1121         /*
1122          * Negotiate features: Subset of device feature bits are written back
1123          * guest feature bits.
1124          */
1125         hw->guest_features = req_features;
1126         hw->guest_features = vtpci_negotiate_features(hw, host_features);
1127         PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1128                 hw->guest_features);
1129
1130         if (hw->modern) {
1131                 if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1132                         PMD_INIT_LOG(ERR,
1133                                 "VIRTIO_F_VERSION_1 features is not enabled.");
1134                         return -1;
1135                 }
1136                 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1137                 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1138                         PMD_INIT_LOG(ERR,
1139                                 "failed to set FEATURES_OK status!");
1140                         return -1;
1141                 }
1142         }
1143
1144         hw->req_guest_features = req_features;
1145
1146         return 0;
1147 }
1148
1149 /*
1150  * Process Virtio Config changed interrupt and call the callback
1151  * if link state changed.
1152  */
1153 static void
1154 virtio_interrupt_handler(__rte_unused struct rte_intr_handle *handle,
1155                          void *param)
1156 {
1157         struct rte_eth_dev *dev = param;
1158         struct virtio_hw *hw = dev->data->dev_private;
1159         uint8_t isr;
1160
1161         /* Read interrupt status which clears interrupt */
1162         isr = vtpci_isr(hw);
1163         PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1164
1165         if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0)
1166                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1167
1168         if (isr & VIRTIO_PCI_ISR_CONFIG) {
1169                 if (virtio_dev_link_update(dev, 0) == 0)
1170                         _rte_eth_dev_callback_process(dev,
1171                                                       RTE_ETH_EVENT_INTR_LSC, NULL);
1172         }
1173
1174 }
1175
1176 static void
1177 rx_func_get(struct rte_eth_dev *eth_dev)
1178 {
1179         struct virtio_hw *hw = eth_dev->data->dev_private;
1180         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF))
1181                 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1182         else
1183                 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1184 }
1185
1186 /* reset device and renegotiate features if needed */
1187 static int
1188 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1189 {
1190         struct virtio_hw *hw = eth_dev->data->dev_private;
1191         struct virtio_net_config *config;
1192         struct virtio_net_config local_config;
1193         struct rte_pci_device *pci_dev = eth_dev->pci_dev;
1194         int ret;
1195
1196         /* Reset the device although not necessary at startup */
1197         vtpci_reset(hw);
1198
1199         /* Tell the host we've noticed this device. */
1200         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1201
1202         /* Tell the host we've known how to drive the device. */
1203         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1204         if (virtio_negotiate_features(hw, req_features) < 0)
1205                 return -1;
1206
1207         /* If host does not support status then disable LSC */
1208         if (!vtpci_with_feature(hw, VIRTIO_NET_F_STATUS))
1209                 eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1210         else
1211                 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1212
1213         rte_eth_copy_pci_info(eth_dev, pci_dev);
1214
1215         rx_func_get(eth_dev);
1216
1217         /* Setting up rx_header size for the device */
1218         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1219             vtpci_with_feature(hw, VIRTIO_F_VERSION_1))
1220                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1221         else
1222                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1223
1224         /* Copy the permanent MAC address to: virtio_hw */
1225         virtio_get_hwaddr(hw);
1226         ether_addr_copy((struct ether_addr *) hw->mac_addr,
1227                         &eth_dev->data->mac_addrs[0]);
1228         PMD_INIT_LOG(DEBUG,
1229                      "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1230                      hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1231                      hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1232
1233         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1234                 config = &local_config;
1235
1236                 vtpci_read_dev_config(hw,
1237                         offsetof(struct virtio_net_config, mac),
1238                         &config->mac, sizeof(config->mac));
1239
1240                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1241                         vtpci_read_dev_config(hw,
1242                                 offsetof(struct virtio_net_config, status),
1243                                 &config->status, sizeof(config->status));
1244                 } else {
1245                         PMD_INIT_LOG(DEBUG,
1246                                      "VIRTIO_NET_F_STATUS is not supported");
1247                         config->status = 0;
1248                 }
1249
1250                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1251                         vtpci_read_dev_config(hw,
1252                                 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1253                                 &config->max_virtqueue_pairs,
1254                                 sizeof(config->max_virtqueue_pairs));
1255                 } else {
1256                         PMD_INIT_LOG(DEBUG,
1257                                      "VIRTIO_NET_F_MQ is not supported");
1258                         config->max_virtqueue_pairs = 1;
1259                 }
1260
1261                 hw->max_queue_pairs = config->max_virtqueue_pairs;
1262
1263                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1264                                 config->max_virtqueue_pairs);
1265                 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1266                 PMD_INIT_LOG(DEBUG,
1267                                 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1268                                 config->mac[0], config->mac[1],
1269                                 config->mac[2], config->mac[3],
1270                                 config->mac[4], config->mac[5]);
1271         } else {
1272                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1273                 hw->max_queue_pairs = 1;
1274         }
1275
1276         ret = virtio_alloc_queues(eth_dev);
1277         if (ret < 0)
1278                 return ret;
1279         vtpci_reinit_complete(hw);
1280
1281         if (pci_dev)
1282                 PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1283                         eth_dev->data->port_id, pci_dev->id.vendor_id,
1284                         pci_dev->id.device_id);
1285
1286         return 0;
1287 }
1288
1289 /*
1290  * This function is based on probe() function in virtio_pci.c
1291  * It returns 0 on success.
1292  */
1293 int
1294 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1295 {
1296         struct virtio_hw *hw = eth_dev->data->dev_private;
1297         struct rte_pci_device *pci_dev;
1298         uint32_t dev_flags = RTE_ETH_DEV_DETACHABLE;
1299         int ret;
1300
1301         RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1302
1303         eth_dev->dev_ops = &virtio_eth_dev_ops;
1304         eth_dev->tx_pkt_burst = &virtio_xmit_pkts;
1305
1306         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1307                 rx_func_get(eth_dev);
1308                 return 0;
1309         }
1310
1311         /* Allocate memory for storing MAC addresses */
1312         eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
1313         if (eth_dev->data->mac_addrs == NULL) {
1314                 PMD_INIT_LOG(ERR,
1315                         "Failed to allocate %d bytes needed to store MAC addresses",
1316                         VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
1317                 return -ENOMEM;
1318         }
1319
1320         pci_dev = eth_dev->pci_dev;
1321
1322         if (pci_dev) {
1323                 ret = vtpci_init(pci_dev, hw, &dev_flags);
1324                 if (ret)
1325                         return ret;
1326         }
1327
1328         eth_dev->data->dev_flags = dev_flags;
1329
1330         /* reset device and negotiate default features */
1331         ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1332         if (ret < 0)
1333                 return ret;
1334
1335         /* Setup interrupt callback  */
1336         if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1337                 rte_intr_callback_register(&pci_dev->intr_handle,
1338                         virtio_interrupt_handler, eth_dev);
1339
1340         return 0;
1341 }
1342
1343 static int
1344 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1345 {
1346         struct rte_pci_device *pci_dev;
1347
1348         PMD_INIT_FUNC_TRACE();
1349
1350         if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1351                 return -EPERM;
1352
1353         virtio_dev_stop(eth_dev);
1354         virtio_dev_close(eth_dev);
1355         pci_dev = eth_dev->pci_dev;
1356
1357         eth_dev->dev_ops = NULL;
1358         eth_dev->tx_pkt_burst = NULL;
1359         eth_dev->rx_pkt_burst = NULL;
1360
1361         rte_free(eth_dev->data->mac_addrs);
1362         eth_dev->data->mac_addrs = NULL;
1363
1364         /* reset interrupt callback  */
1365         if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1366                 rte_intr_callback_unregister(&pci_dev->intr_handle,
1367                                                 virtio_interrupt_handler,
1368                                                 eth_dev);
1369         rte_eal_pci_unmap_device(pci_dev);
1370
1371         PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1372
1373         return 0;
1374 }
1375
1376 static struct eth_driver rte_virtio_pmd = {
1377         .pci_drv = {
1378                 .driver = {
1379                         .name = "net_virtio",
1380                 },
1381                 .id_table = pci_id_virtio_map,
1382                 .drv_flags = RTE_PCI_DRV_DETACHABLE,
1383                 .probe = rte_eth_dev_pci_probe,
1384                 .remove = rte_eth_dev_pci_remove,
1385         },
1386         .eth_dev_init = eth_virtio_dev_init,
1387         .eth_dev_uninit = eth_virtio_dev_uninit,
1388         .dev_private_size = sizeof(struct virtio_hw),
1389 };
1390
1391 RTE_INIT(rte_virtio_pmd_init);
1392 static void
1393 rte_virtio_pmd_init(void)
1394 {
1395         if (rte_eal_iopl_init() != 0) {
1396                 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1397                 return;
1398         }
1399
1400         rte_eal_pci_register(&rte_virtio_pmd.pci_drv);
1401 }
1402
1403 /*
1404  * Configure virtio device
1405  * It returns 0 on success.
1406  */
1407 static int
1408 virtio_dev_configure(struct rte_eth_dev *dev)
1409 {
1410         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1411         struct virtio_hw *hw = dev->data->dev_private;
1412         uint64_t req_features;
1413         int ret;
1414
1415         PMD_INIT_LOG(DEBUG, "configure");
1416         req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
1417         if (rxmode->hw_ip_checksum)
1418                 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
1419         if (rxmode->enable_lro)
1420                 req_features |=
1421                         (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
1422                         (1ULL << VIRTIO_NET_F_GUEST_TSO6);
1423
1424         /* if request features changed, reinit the device */
1425         if (req_features != hw->req_guest_features) {
1426                 ret = virtio_init_device(dev, req_features);
1427                 if (ret < 0)
1428                         return ret;
1429         }
1430
1431         if (rxmode->hw_ip_checksum &&
1432                 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
1433                 PMD_DRV_LOG(NOTICE,
1434                         "rx ip checksum not available on this host");
1435                 return -ENOTSUP;
1436         }
1437
1438         if (rxmode->enable_lro &&
1439                 (!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
1440                         !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4))) {
1441                 PMD_DRV_LOG(NOTICE,
1442                         "lro not available on this host");
1443                 return -ENOTSUP;
1444         }
1445
1446         /* start control queue */
1447         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
1448                 virtio_dev_cq_start(dev);
1449
1450         hw->vlan_strip = rxmode->hw_vlan_strip;
1451
1452         if (rxmode->hw_vlan_filter
1453             && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1454                 PMD_DRV_LOG(NOTICE,
1455                             "vlan filtering not available on this host");
1456                 return -ENOTSUP;
1457         }
1458
1459         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1460                 if (vtpci_irq_config(hw, 0) == VIRTIO_MSI_NO_VECTOR) {
1461                         PMD_DRV_LOG(ERR, "failed to set config vector");
1462                         return -EBUSY;
1463                 }
1464
1465         return 0;
1466 }
1467
1468
1469 static int
1470 virtio_dev_start(struct rte_eth_dev *dev)
1471 {
1472         uint16_t nb_queues, i;
1473         struct virtnet_rx *rxvq;
1474         struct virtnet_tx *txvq __rte_unused;
1475         struct virtio_hw *hw = dev->data->dev_private;
1476
1477         /* check if lsc interrupt feature is enabled */
1478         if (dev->data->dev_conf.intr_conf.lsc) {
1479                 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1480                         PMD_DRV_LOG(ERR, "link status not supported by host");
1481                         return -ENOTSUP;
1482                 }
1483
1484                 if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0) {
1485                         PMD_DRV_LOG(ERR, "interrupt enable failed");
1486                         return -EIO;
1487                 }
1488         }
1489
1490         /* Initialize Link state */
1491         virtio_dev_link_update(dev, 0);
1492
1493         /*Notify the backend
1494          *Otherwise the tap backend might already stop its queue due to fullness.
1495          *vhost backend will have no chance to be waked up
1496          */
1497         nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
1498         if (hw->max_queue_pairs > 1) {
1499                 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1500                         return -EINVAL;
1501         }
1502
1503         PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1504
1505         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1506                 rxvq = dev->data->rx_queues[i];
1507                 virtqueue_notify(rxvq->vq);
1508         }
1509
1510         PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1511
1512         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1513                 rxvq = dev->data->rx_queues[i];
1514                 VIRTQUEUE_DUMP(rxvq->vq);
1515         }
1516
1517         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1518                 txvq = dev->data->tx_queues[i];
1519                 VIRTQUEUE_DUMP(txvq->vq);
1520         }
1521
1522         return 0;
1523 }
1524
1525 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1526 {
1527         struct rte_mbuf *buf;
1528         int i, mbuf_num = 0;
1529
1530         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1531                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1532
1533                 PMD_INIT_LOG(DEBUG,
1534                              "Before freeing rxq[%d] used and unused buf", i);
1535                 VIRTQUEUE_DUMP(rxvq->vq);
1536
1537                 PMD_INIT_LOG(DEBUG, "rx_queues[%d]=%p", i, rxvq);
1538                 while ((buf = virtqueue_detatch_unused(rxvq->vq)) != NULL) {
1539                         rte_pktmbuf_free(buf);
1540                         mbuf_num++;
1541                 }
1542
1543                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1544                 PMD_INIT_LOG(DEBUG,
1545                              "After freeing rxq[%d] used and unused buf", i);
1546                 VIRTQUEUE_DUMP(rxvq->vq);
1547         }
1548
1549         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1550                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1551
1552                 PMD_INIT_LOG(DEBUG,
1553                              "Before freeing txq[%d] used and unused bufs",
1554                              i);
1555                 VIRTQUEUE_DUMP(txvq->vq);
1556
1557                 mbuf_num = 0;
1558                 while ((buf = virtqueue_detatch_unused(txvq->vq)) != NULL) {
1559                         rte_pktmbuf_free(buf);
1560                         mbuf_num++;
1561                 }
1562
1563                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1564                 PMD_INIT_LOG(DEBUG,
1565                              "After freeing txq[%d] used and unused buf", i);
1566                 VIRTQUEUE_DUMP(txvq->vq);
1567         }
1568 }
1569
1570 /*
1571  * Stop device: disable interrupt and mark link down
1572  */
1573 static void
1574 virtio_dev_stop(struct rte_eth_dev *dev)
1575 {
1576         struct rte_eth_link link;
1577
1578         PMD_INIT_LOG(DEBUG, "stop");
1579
1580         if (dev->data->dev_conf.intr_conf.lsc)
1581                 rte_intr_disable(&dev->pci_dev->intr_handle);
1582
1583         memset(&link, 0, sizeof(link));
1584         virtio_dev_atomic_write_link_status(dev, &link);
1585 }
1586
1587 static int
1588 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1589 {
1590         struct rte_eth_link link, old;
1591         uint16_t status;
1592         struct virtio_hw *hw = dev->data->dev_private;
1593         memset(&link, 0, sizeof(link));
1594         virtio_dev_atomic_read_link_status(dev, &link);
1595         old = link;
1596         link.link_duplex = ETH_LINK_FULL_DUPLEX;
1597         link.link_speed  = SPEED_10G;
1598
1599         if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1600                 PMD_INIT_LOG(DEBUG, "Get link status from hw");
1601                 vtpci_read_dev_config(hw,
1602                                 offsetof(struct virtio_net_config, status),
1603                                 &status, sizeof(status));
1604                 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1605                         link.link_status = ETH_LINK_DOWN;
1606                         PMD_INIT_LOG(DEBUG, "Port %d is down",
1607                                      dev->data->port_id);
1608                 } else {
1609                         link.link_status = ETH_LINK_UP;
1610                         PMD_INIT_LOG(DEBUG, "Port %d is up",
1611                                      dev->data->port_id);
1612                 }
1613         } else {
1614                 link.link_status = ETH_LINK_UP;
1615         }
1616         virtio_dev_atomic_write_link_status(dev, &link);
1617
1618         return (old.link_status == link.link_status) ? -1 : 0;
1619 }
1620
1621 static void
1622 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1623 {
1624         uint64_t tso_mask;
1625         struct virtio_hw *hw = dev->data->dev_private;
1626
1627         dev_info->max_rx_queues =
1628                 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
1629         dev_info->max_tx_queues =
1630                 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
1631         dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1632         dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1633         dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1634         dev_info->default_txconf = (struct rte_eth_txconf) {
1635                 .txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
1636         };
1637         dev_info->rx_offload_capa =
1638                 DEV_RX_OFFLOAD_TCP_CKSUM |
1639                 DEV_RX_OFFLOAD_UDP_CKSUM |
1640                 DEV_RX_OFFLOAD_TCP_LRO;
1641         dev_info->tx_offload_capa = 0;
1642
1643         if (hw->guest_features & (1ULL << VIRTIO_NET_F_CSUM)) {
1644                 dev_info->tx_offload_capa |=
1645                         DEV_TX_OFFLOAD_UDP_CKSUM |
1646                         DEV_TX_OFFLOAD_TCP_CKSUM;
1647         }
1648
1649         tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
1650                 (1ULL << VIRTIO_NET_F_HOST_TSO6);
1651         if ((hw->guest_features & tso_mask) == tso_mask)
1652                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
1653 }
1654
1655 /*
1656  * It enables testpmd to collect per queue stats.
1657  */
1658 static int
1659 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
1660 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
1661 __rte_unused uint8_t is_rx)
1662 {
1663         return 0;
1664 }
1665
1666 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
1667 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
1668 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio");