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