0b0967defbafd3d05a0c77db5898c2e849ee6e2e
[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 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
156
157 static int
158 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
159                 int *dlen, int pkt_num)
160 {
161         uint32_t head, i;
162         int k, sum = 0;
163         virtio_net_ctrl_ack status = ~0;
164         struct virtio_pmd_ctrl result;
165         struct virtqueue *vq;
166
167         ctrl->status = status;
168
169         if (!cvq || !cvq->vq) {
170                 PMD_INIT_LOG(ERR, "Control queue is not supported.");
171                 return -1;
172         }
173         vq = cvq->vq;
174         head = vq->vq_desc_head_idx;
175
176         PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
177                 "vq->hw->cvq = %p vq = %p",
178                 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
179
180         if ((vq->vq_free_cnt < ((uint32_t)pkt_num + 2)) || (pkt_num < 1))
181                 return -1;
182
183         memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
184                 sizeof(struct virtio_pmd_ctrl));
185
186         /*
187          * Format is enforced in qemu code:
188          * One TX packet for header;
189          * At least one TX packet per argument;
190          * One RX packet for ACK.
191          */
192         vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
193         vq->vq_ring.desc[head].addr = cvq->virtio_net_hdr_mem;
194         vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
195         vq->vq_free_cnt--;
196         i = vq->vq_ring.desc[head].next;
197
198         for (k = 0; k < pkt_num; k++) {
199                 vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
200                 vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
201                         + sizeof(struct virtio_net_ctrl_hdr)
202                         + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
203                 vq->vq_ring.desc[i].len = dlen[k];
204                 sum += dlen[k];
205                 vq->vq_free_cnt--;
206                 i = vq->vq_ring.desc[i].next;
207         }
208
209         vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
210         vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
211                         + sizeof(struct virtio_net_ctrl_hdr);
212         vq->vq_ring.desc[i].len = sizeof(ctrl->status);
213         vq->vq_free_cnt--;
214
215         vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
216
217         vq_update_avail_ring(vq, head);
218         vq_update_avail_idx(vq);
219
220         PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
221
222         virtqueue_notify(vq);
223
224         rte_rmb();
225         while (VIRTQUEUE_NUSED(vq) == 0) {
226                 rte_rmb();
227                 usleep(100);
228         }
229
230         while (VIRTQUEUE_NUSED(vq)) {
231                 uint32_t idx, desc_idx, used_idx;
232                 struct vring_used_elem *uep;
233
234                 used_idx = (uint32_t)(vq->vq_used_cons_idx
235                                 & (vq->vq_nentries - 1));
236                 uep = &vq->vq_ring.used->ring[used_idx];
237                 idx = (uint32_t) uep->id;
238                 desc_idx = idx;
239
240                 while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
241                         desc_idx = vq->vq_ring.desc[desc_idx].next;
242                         vq->vq_free_cnt++;
243                 }
244
245                 vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
246                 vq->vq_desc_head_idx = idx;
247
248                 vq->vq_used_cons_idx++;
249                 vq->vq_free_cnt++;
250         }
251
252         PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
253                         vq->vq_free_cnt, vq->vq_desc_head_idx);
254
255         memcpy(&result, cvq->virtio_net_hdr_mz->addr,
256                         sizeof(struct virtio_pmd_ctrl));
257
258         return result.status;
259 }
260
261 static int
262 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
263 {
264         struct virtio_hw *hw = dev->data->dev_private;
265         struct virtio_pmd_ctrl ctrl;
266         int dlen[1];
267         int ret;
268
269         ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
270         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
271         memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
272
273         dlen[0] = sizeof(uint16_t);
274
275         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
276         if (ret) {
277                 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
278                           "failed, this is too late now...");
279                 return -EINVAL;
280         }
281
282         return 0;
283 }
284
285 static void
286 virtio_dev_queue_release(void *queue __rte_unused)
287 {
288         /* do nothing */
289 }
290
291 static int
292 virtio_get_queue_type(struct virtio_hw *hw, uint16_t vtpci_queue_idx)
293 {
294         if (vtpci_queue_idx == hw->max_queue_pairs * 2)
295                 return VTNET_CQ;
296         else if (vtpci_queue_idx % 2 == 0)
297                 return VTNET_RQ;
298         else
299                 return VTNET_TQ;
300 }
301
302 static uint16_t
303 virtio_get_nr_vq(struct virtio_hw *hw)
304 {
305         uint16_t nr_vq = hw->max_queue_pairs * 2;
306
307         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
308                 nr_vq += 1;
309
310         return nr_vq;
311 }
312
313 static void
314 virtio_init_vring(struct virtqueue *vq)
315 {
316         int size = vq->vq_nentries;
317         struct vring *vr = &vq->vq_ring;
318         uint8_t *ring_mem = vq->vq_ring_virt_mem;
319
320         PMD_INIT_FUNC_TRACE();
321
322         /*
323          * Reinitialise since virtio port might have been stopped and restarted
324          */
325         memset(ring_mem, 0, vq->vq_ring_size);
326         vring_init(vr, size, ring_mem, VIRTIO_PCI_VRING_ALIGN);
327         vq->vq_used_cons_idx = 0;
328         vq->vq_desc_head_idx = 0;
329         vq->vq_avail_idx = 0;
330         vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
331         vq->vq_free_cnt = vq->vq_nentries;
332         memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
333
334         vring_desc_init(vr->desc, size);
335
336         /*
337          * Disable device(host) interrupting guest
338          */
339         virtqueue_disable_intr(vq);
340 }
341
342 static int
343 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
344 {
345         char vq_name[VIRTQUEUE_MAX_NAME_SZ];
346         char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
347         const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
348         unsigned int vq_size, size;
349         struct virtio_hw *hw = dev->data->dev_private;
350         struct virtnet_rx *rxvq = NULL;
351         struct virtnet_tx *txvq = NULL;
352         struct virtnet_ctl *cvq = NULL;
353         struct virtqueue *vq;
354         size_t sz_hdr_mz = 0;
355         void *sw_ring = NULL;
356         int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
357         int ret;
358
359         PMD_INIT_LOG(DEBUG, "setting up queue: %u", vtpci_queue_idx);
360
361         /*
362          * Read the virtqueue size from the Queue Size field
363          * Always power of 2 and if 0 virtqueue does not exist
364          */
365         vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
366         PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
367         if (vq_size == 0) {
368                 PMD_INIT_LOG(ERR, "virtqueue does not exist");
369                 return -EINVAL;
370         }
371
372         if (!rte_is_power_of_2(vq_size)) {
373                 PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
374                 return -EINVAL;
375         }
376
377         snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
378                  dev->data->port_id, vtpci_queue_idx);
379
380         size = RTE_ALIGN_CEIL(sizeof(*vq) +
381                                 vq_size * sizeof(struct vq_desc_extra),
382                                 RTE_CACHE_LINE_SIZE);
383         if (queue_type == VTNET_TQ) {
384                 /*
385                  * For each xmit packet, allocate a virtio_net_hdr
386                  * and indirect ring elements
387                  */
388                 sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
389         } else if (queue_type == VTNET_CQ) {
390                 /* Allocate a page for control vq command, data and status */
391                 sz_hdr_mz = PAGE_SIZE;
392         }
393
394         vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
395                                 SOCKET_ID_ANY);
396         if (vq == NULL) {
397                 PMD_INIT_LOG(ERR, "can not allocate vq");
398                 return -ENOMEM;
399         }
400         hw->vqs[vtpci_queue_idx] = vq;
401
402         vq->hw = hw;
403         vq->vq_queue_index = vtpci_queue_idx;
404         vq->vq_nentries = vq_size;
405
406         /*
407          * Reserve a memzone for vring elements
408          */
409         size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
410         vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
411         PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
412                      size, vq->vq_ring_size);
413
414         mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
415                                          SOCKET_ID_ANY,
416                                          0, VIRTIO_PCI_VRING_ALIGN);
417         if (mz == NULL) {
418                 if (rte_errno == EEXIST)
419                         mz = rte_memzone_lookup(vq_name);
420                 if (mz == NULL) {
421                         ret = -ENOMEM;
422                         goto fail_q_alloc;
423                 }
424         }
425
426         memset(mz->addr, 0, sizeof(mz->len));
427
428         vq->vq_ring_mem = mz->phys_addr;
429         vq->vq_ring_virt_mem = mz->addr;
430         PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%" PRIx64,
431                      (uint64_t)mz->phys_addr);
432         PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
433                      (uint64_t)(uintptr_t)mz->addr);
434
435         virtio_init_vring(vq);
436
437         if (sz_hdr_mz) {
438                 snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
439                          dev->data->port_id, vtpci_queue_idx);
440                 hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
441                                                      SOCKET_ID_ANY, 0,
442                                                      RTE_CACHE_LINE_SIZE);
443                 if (hdr_mz == NULL) {
444                         if (rte_errno == EEXIST)
445                                 hdr_mz = rte_memzone_lookup(vq_hdr_name);
446                         if (hdr_mz == NULL) {
447                                 ret = -ENOMEM;
448                                 goto fail_q_alloc;
449                         }
450                 }
451         }
452
453         if (queue_type == VTNET_RQ) {
454                 size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
455                                sizeof(vq->sw_ring[0]);
456
457                 sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
458                                 RTE_CACHE_LINE_SIZE, SOCKET_ID_ANY);
459                 if (!sw_ring) {
460                         PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
461                         ret = -ENOMEM;
462                         goto fail_q_alloc;
463                 }
464
465                 vq->sw_ring = sw_ring;
466                 rxvq = &vq->rxq;
467                 rxvq->vq = vq;
468                 rxvq->port_id = dev->data->port_id;
469                 rxvq->mz = mz;
470         } else if (queue_type == VTNET_TQ) {
471                 txvq = &vq->txq;
472                 txvq->vq = vq;
473                 txvq->port_id = dev->data->port_id;
474                 txvq->mz = mz;
475                 txvq->virtio_net_hdr_mz = hdr_mz;
476                 txvq->virtio_net_hdr_mem = hdr_mz->phys_addr;
477         } else if (queue_type == VTNET_CQ) {
478                 cvq = &vq->cq;
479                 cvq->vq = vq;
480                 cvq->mz = mz;
481                 cvq->virtio_net_hdr_mz = hdr_mz;
482                 cvq->virtio_net_hdr_mem = hdr_mz->phys_addr;
483                 memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
484
485                 hw->cvq = cvq;
486         }
487
488         /* For virtio_user case (that is when hw->dev is NULL), we use
489          * virtual address. And we need properly set _offset_, please see
490          * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
491          */
492         if (hw->dev)
493                 vq->offset = offsetof(struct rte_mbuf, buf_physaddr);
494         else {
495                 vq->vq_ring_mem = (uintptr_t)mz->addr;
496                 vq->offset = offsetof(struct rte_mbuf, buf_addr);
497                 if (queue_type == VTNET_TQ)
498                         txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
499                 else if (queue_type == VTNET_CQ)
500                         cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
501         }
502
503         if (queue_type == VTNET_TQ) {
504                 struct virtio_tx_region *txr;
505                 unsigned int i;
506
507                 txr = hdr_mz->addr;
508                 memset(txr, 0, vq_size * sizeof(*txr));
509                 for (i = 0; i < vq_size; i++) {
510                         struct vring_desc *start_dp = txr[i].tx_indir;
511
512                         vring_desc_init(start_dp, RTE_DIM(txr[i].tx_indir));
513
514                         /* first indirect descriptor is always the tx header */
515                         start_dp->addr = txvq->virtio_net_hdr_mem
516                                 + i * sizeof(*txr)
517                                 + offsetof(struct virtio_tx_region, tx_hdr);
518
519                         start_dp->len = hw->vtnet_hdr_size;
520                         start_dp->flags = VRING_DESC_F_NEXT;
521                 }
522         }
523
524         if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
525                 PMD_INIT_LOG(ERR, "setup_queue failed");
526                 return -EINVAL;
527         }
528
529         return 0;
530
531 fail_q_alloc:
532         rte_free(sw_ring);
533         rte_memzone_free(hdr_mz);
534         rte_memzone_free(mz);
535         rte_free(vq);
536
537         return ret;
538 }
539
540 static void
541 virtio_free_queues(struct virtio_hw *hw)
542 {
543         uint16_t nr_vq = virtio_get_nr_vq(hw);
544         struct virtqueue *vq;
545         int queue_type;
546         uint16_t i;
547
548         for (i = 0; i < nr_vq; i++) {
549                 vq = hw->vqs[i];
550                 if (!vq)
551                         continue;
552
553                 queue_type = virtio_get_queue_type(hw, i);
554                 if (queue_type == VTNET_RQ) {
555                         rte_free(vq->sw_ring);
556                         rte_memzone_free(vq->rxq.mz);
557                 } else if (queue_type == VTNET_TQ) {
558                         rte_memzone_free(vq->txq.mz);
559                         rte_memzone_free(vq->txq.virtio_net_hdr_mz);
560                 } else {
561                         rte_memzone_free(vq->cq.mz);
562                         rte_memzone_free(vq->cq.virtio_net_hdr_mz);
563                 }
564
565                 rte_free(vq);
566         }
567
568         rte_free(hw->vqs);
569 }
570
571 static int
572 virtio_alloc_queues(struct rte_eth_dev *dev)
573 {
574         struct virtio_hw *hw = dev->data->dev_private;
575         uint16_t nr_vq = virtio_get_nr_vq(hw);
576         uint16_t i;
577         int ret;
578
579         hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
580         if (!hw->vqs) {
581                 PMD_INIT_LOG(ERR, "failed to allocate vqs");
582                 return -ENOMEM;
583         }
584
585         for (i = 0; i < nr_vq; i++) {
586                 ret = virtio_init_queue(dev, i);
587                 if (ret < 0) {
588                         virtio_free_queues(hw);
589                         return ret;
590                 }
591         }
592
593         return 0;
594 }
595
596 static void
597 virtio_dev_close(struct rte_eth_dev *dev)
598 {
599         struct virtio_hw *hw = dev->data->dev_private;
600
601         PMD_INIT_LOG(DEBUG, "virtio_dev_close");
602
603         /* reset the NIC */
604         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
605                 vtpci_irq_config(hw, VIRTIO_MSI_NO_VECTOR);
606         vtpci_reset(hw);
607         virtio_dev_free_mbufs(dev);
608         virtio_free_queues(hw);
609 }
610
611 static void
612 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
613 {
614         struct virtio_hw *hw = dev->data->dev_private;
615         struct virtio_pmd_ctrl ctrl;
616         int dlen[1];
617         int ret;
618
619         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
620                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
621                 return;
622         }
623
624         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
625         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
626         ctrl.data[0] = 1;
627         dlen[0] = 1;
628
629         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
630         if (ret)
631                 PMD_INIT_LOG(ERR, "Failed to enable promisc");
632 }
633
634 static void
635 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
636 {
637         struct virtio_hw *hw = dev->data->dev_private;
638         struct virtio_pmd_ctrl ctrl;
639         int dlen[1];
640         int ret;
641
642         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
643                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
644                 return;
645         }
646
647         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
648         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
649         ctrl.data[0] = 0;
650         dlen[0] = 1;
651
652         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
653         if (ret)
654                 PMD_INIT_LOG(ERR, "Failed to disable promisc");
655 }
656
657 static void
658 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
659 {
660         struct virtio_hw *hw = dev->data->dev_private;
661         struct virtio_pmd_ctrl ctrl;
662         int dlen[1];
663         int ret;
664
665         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
666                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
667                 return;
668         }
669
670         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
671         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
672         ctrl.data[0] = 1;
673         dlen[0] = 1;
674
675         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
676         if (ret)
677                 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
678 }
679
680 static void
681 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
682 {
683         struct virtio_hw *hw = dev->data->dev_private;
684         struct virtio_pmd_ctrl ctrl;
685         int dlen[1];
686         int ret;
687
688         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
689                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
690                 return;
691         }
692
693         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
694         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
695         ctrl.data[0] = 0;
696         dlen[0] = 1;
697
698         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
699         if (ret)
700                 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
701 }
702
703 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
704 static int
705 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
706 {
707         struct virtio_hw *hw = dev->data->dev_private;
708         uint32_t ether_hdr_len = ETHER_HDR_LEN + VLAN_TAG_LEN +
709                                  hw->vtnet_hdr_size;
710         uint32_t frame_size = mtu + ether_hdr_len;
711
712         if (mtu < ETHER_MIN_MTU || frame_size > VIRTIO_MAX_RX_PKTLEN) {
713                 PMD_INIT_LOG(ERR, "MTU should be between %d and %d\n",
714                         ETHER_MIN_MTU, VIRTIO_MAX_RX_PKTLEN - ether_hdr_len);
715                 return -EINVAL;
716         }
717         return 0;
718 }
719
720 /*
721  * dev_ops for virtio, bare necessities for basic operation
722  */
723 static const struct eth_dev_ops virtio_eth_dev_ops = {
724         .dev_configure           = virtio_dev_configure,
725         .dev_start               = virtio_dev_start,
726         .dev_stop                = virtio_dev_stop,
727         .dev_close               = virtio_dev_close,
728         .promiscuous_enable      = virtio_dev_promiscuous_enable,
729         .promiscuous_disable     = virtio_dev_promiscuous_disable,
730         .allmulticast_enable     = virtio_dev_allmulticast_enable,
731         .allmulticast_disable    = virtio_dev_allmulticast_disable,
732         .mtu_set                 = virtio_mtu_set,
733         .dev_infos_get           = virtio_dev_info_get,
734         .stats_get               = virtio_dev_stats_get,
735         .xstats_get              = virtio_dev_xstats_get,
736         .xstats_get_names        = virtio_dev_xstats_get_names,
737         .stats_reset             = virtio_dev_stats_reset,
738         .xstats_reset            = virtio_dev_stats_reset,
739         .link_update             = virtio_dev_link_update,
740         .rx_queue_setup          = virtio_dev_rx_queue_setup,
741         .rx_queue_release        = virtio_dev_queue_release,
742         .tx_queue_setup          = virtio_dev_tx_queue_setup,
743         .tx_queue_release        = virtio_dev_queue_release,
744         /* collect stats per queue */
745         .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
746         .vlan_filter_set         = virtio_vlan_filter_set,
747         .mac_addr_add            = virtio_mac_addr_add,
748         .mac_addr_remove         = virtio_mac_addr_remove,
749         .mac_addr_set            = virtio_mac_addr_set,
750 };
751
752 static inline int
753 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
754                                 struct rte_eth_link *link)
755 {
756         struct rte_eth_link *dst = link;
757         struct rte_eth_link *src = &(dev->data->dev_link);
758
759         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
760                         *(uint64_t *)src) == 0)
761                 return -1;
762
763         return 0;
764 }
765
766 /**
767  * Atomically writes the link status information into global
768  * structure rte_eth_dev.
769  *
770  * @param dev
771  *   - Pointer to the structure rte_eth_dev to read from.
772  *   - Pointer to the buffer to be saved with the link status.
773  *
774  * @return
775  *   - On success, zero.
776  *   - On failure, negative value.
777  */
778 static inline int
779 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
780                 struct rte_eth_link *link)
781 {
782         struct rte_eth_link *dst = &(dev->data->dev_link);
783         struct rte_eth_link *src = link;
784
785         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
786                                         *(uint64_t *)src) == 0)
787                 return -1;
788
789         return 0;
790 }
791
792 static void
793 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
794 {
795         unsigned i;
796
797         for (i = 0; i < dev->data->nb_tx_queues; i++) {
798                 const struct virtnet_tx *txvq = dev->data->tx_queues[i];
799                 if (txvq == NULL)
800                         continue;
801
802                 stats->opackets += txvq->stats.packets;
803                 stats->obytes += txvq->stats.bytes;
804                 stats->oerrors += txvq->stats.errors;
805
806                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
807                         stats->q_opackets[i] = txvq->stats.packets;
808                         stats->q_obytes[i] = txvq->stats.bytes;
809                 }
810         }
811
812         for (i = 0; i < dev->data->nb_rx_queues; i++) {
813                 const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
814                 if (rxvq == NULL)
815                         continue;
816
817                 stats->ipackets += rxvq->stats.packets;
818                 stats->ibytes += rxvq->stats.bytes;
819                 stats->ierrors += rxvq->stats.errors;
820
821                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
822                         stats->q_ipackets[i] = rxvq->stats.packets;
823                         stats->q_ibytes[i] = rxvq->stats.bytes;
824                 }
825         }
826
827         stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
828 }
829
830 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
831                                        struct rte_eth_xstat_name *xstats_names,
832                                        __rte_unused unsigned limit)
833 {
834         unsigned i;
835         unsigned count = 0;
836         unsigned t;
837
838         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
839                 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
840
841         if (xstats_names != NULL) {
842                 /* Note: limit checked in rte_eth_xstats_names() */
843
844                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
845                         struct virtqueue *rxvq = dev->data->rx_queues[i];
846                         if (rxvq == NULL)
847                                 continue;
848                         for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
849                                 snprintf(xstats_names[count].name,
850                                         sizeof(xstats_names[count].name),
851                                         "rx_q%u_%s", i,
852                                         rte_virtio_rxq_stat_strings[t].name);
853                                 count++;
854                         }
855                 }
856
857                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
858                         struct virtqueue *txvq = dev->data->tx_queues[i];
859                         if (txvq == NULL)
860                                 continue;
861                         for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
862                                 snprintf(xstats_names[count].name,
863                                         sizeof(xstats_names[count].name),
864                                         "tx_q%u_%s", i,
865                                         rte_virtio_txq_stat_strings[t].name);
866                                 count++;
867                         }
868                 }
869                 return count;
870         }
871         return nstats;
872 }
873
874 static int
875 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
876                       unsigned n)
877 {
878         unsigned i;
879         unsigned count = 0;
880
881         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
882                 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
883
884         if (n < nstats)
885                 return nstats;
886
887         for (i = 0; i < dev->data->nb_rx_queues; i++) {
888                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
889
890                 if (rxvq == NULL)
891                         continue;
892
893                 unsigned t;
894
895                 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
896                         xstats[count].value = *(uint64_t *)(((char *)rxvq) +
897                                 rte_virtio_rxq_stat_strings[t].offset);
898                         xstats[count].id = count;
899                         count++;
900                 }
901         }
902
903         for (i = 0; i < dev->data->nb_tx_queues; i++) {
904                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
905
906                 if (txvq == NULL)
907                         continue;
908
909                 unsigned t;
910
911                 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
912                         xstats[count].value = *(uint64_t *)(((char *)txvq) +
913                                 rte_virtio_txq_stat_strings[t].offset);
914                         xstats[count].id = count;
915                         count++;
916                 }
917         }
918
919         return count;
920 }
921
922 static void
923 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
924 {
925         virtio_update_stats(dev, stats);
926 }
927
928 static void
929 virtio_dev_stats_reset(struct rte_eth_dev *dev)
930 {
931         unsigned int i;
932
933         for (i = 0; i < dev->data->nb_tx_queues; i++) {
934                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
935                 if (txvq == NULL)
936                         continue;
937
938                 txvq->stats.packets = 0;
939                 txvq->stats.bytes = 0;
940                 txvq->stats.errors = 0;
941                 txvq->stats.multicast = 0;
942                 txvq->stats.broadcast = 0;
943                 memset(txvq->stats.size_bins, 0,
944                        sizeof(txvq->stats.size_bins[0]) * 8);
945         }
946
947         for (i = 0; i < dev->data->nb_rx_queues; i++) {
948                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
949                 if (rxvq == NULL)
950                         continue;
951
952                 rxvq->stats.packets = 0;
953                 rxvq->stats.bytes = 0;
954                 rxvq->stats.errors = 0;
955                 rxvq->stats.multicast = 0;
956                 rxvq->stats.broadcast = 0;
957                 memset(rxvq->stats.size_bins, 0,
958                        sizeof(rxvq->stats.size_bins[0]) * 8);
959         }
960 }
961
962 static void
963 virtio_set_hwaddr(struct virtio_hw *hw)
964 {
965         vtpci_write_dev_config(hw,
966                         offsetof(struct virtio_net_config, mac),
967                         &hw->mac_addr, ETHER_ADDR_LEN);
968 }
969
970 static void
971 virtio_get_hwaddr(struct virtio_hw *hw)
972 {
973         if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
974                 vtpci_read_dev_config(hw,
975                         offsetof(struct virtio_net_config, mac),
976                         &hw->mac_addr, ETHER_ADDR_LEN);
977         } else {
978                 eth_random_addr(&hw->mac_addr[0]);
979                 virtio_set_hwaddr(hw);
980         }
981 }
982
983 static void
984 virtio_mac_table_set(struct virtio_hw *hw,
985                      const struct virtio_net_ctrl_mac *uc,
986                      const struct virtio_net_ctrl_mac *mc)
987 {
988         struct virtio_pmd_ctrl ctrl;
989         int err, len[2];
990
991         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
992                 PMD_DRV_LOG(INFO, "host does not support mac table");
993                 return;
994         }
995
996         ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
997         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
998
999         len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
1000         memcpy(ctrl.data, uc, len[0]);
1001
1002         len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
1003         memcpy(ctrl.data + len[0], mc, len[1]);
1004
1005         err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1006         if (err != 0)
1007                 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1008 }
1009
1010 static void
1011 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
1012                     uint32_t index, uint32_t vmdq __rte_unused)
1013 {
1014         struct virtio_hw *hw = dev->data->dev_private;
1015         const struct ether_addr *addrs = dev->data->mac_addrs;
1016         unsigned int i;
1017         struct virtio_net_ctrl_mac *uc, *mc;
1018
1019         if (index >= VIRTIO_MAX_MAC_ADDRS) {
1020                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1021                 return;
1022         }
1023
1024         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1025         uc->entries = 0;
1026         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1027         mc->entries = 0;
1028
1029         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1030                 const struct ether_addr *addr
1031                         = (i == index) ? mac_addr : addrs + i;
1032                 struct virtio_net_ctrl_mac *tbl
1033                         = is_multicast_ether_addr(addr) ? mc : uc;
1034
1035                 memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
1036         }
1037
1038         virtio_mac_table_set(hw, uc, mc);
1039 }
1040
1041 static void
1042 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1043 {
1044         struct virtio_hw *hw = dev->data->dev_private;
1045         struct ether_addr *addrs = dev->data->mac_addrs;
1046         struct virtio_net_ctrl_mac *uc, *mc;
1047         unsigned int i;
1048
1049         if (index >= VIRTIO_MAX_MAC_ADDRS) {
1050                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1051                 return;
1052         }
1053
1054         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1055         uc->entries = 0;
1056         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1057         mc->entries = 0;
1058
1059         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1060                 struct virtio_net_ctrl_mac *tbl;
1061
1062                 if (i == index || is_zero_ether_addr(addrs + i))
1063                         continue;
1064
1065                 tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
1066                 memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
1067         }
1068
1069         virtio_mac_table_set(hw, uc, mc);
1070 }
1071
1072 static void
1073 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1074 {
1075         struct virtio_hw *hw = dev->data->dev_private;
1076
1077         memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
1078
1079         /* Use atomic update if available */
1080         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1081                 struct virtio_pmd_ctrl ctrl;
1082                 int len = ETHER_ADDR_LEN;
1083
1084                 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1085                 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1086
1087                 memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
1088                 virtio_send_command(hw->cvq, &ctrl, &len, 1);
1089         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1090                 virtio_set_hwaddr(hw);
1091 }
1092
1093 static int
1094 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1095 {
1096         struct virtio_hw *hw = dev->data->dev_private;
1097         struct virtio_pmd_ctrl ctrl;
1098         int len;
1099
1100         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1101                 return -ENOTSUP;
1102
1103         ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1104         ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1105         memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1106         len = sizeof(vlan_id);
1107
1108         return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1109 }
1110
1111 static int
1112 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1113 {
1114         uint64_t host_features;
1115
1116         /* Prepare guest_features: feature that driver wants to support */
1117         PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1118                 req_features);
1119
1120         /* Read device(host) feature bits */
1121         host_features = VTPCI_OPS(hw)->get_features(hw);
1122         PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1123                 host_features);
1124
1125         /*
1126          * Negotiate features: Subset of device feature bits are written back
1127          * guest feature bits.
1128          */
1129         hw->guest_features = req_features;
1130         hw->guest_features = vtpci_negotiate_features(hw, host_features);
1131         PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1132                 hw->guest_features);
1133
1134         if (hw->modern) {
1135                 if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1136                         PMD_INIT_LOG(ERR,
1137                                 "VIRTIO_F_VERSION_1 features is not enabled.");
1138                         return -1;
1139                 }
1140                 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1141                 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1142                         PMD_INIT_LOG(ERR,
1143                                 "failed to set FEATURES_OK status!");
1144                         return -1;
1145                 }
1146         }
1147
1148         hw->req_guest_features = req_features;
1149
1150         return 0;
1151 }
1152
1153 /*
1154  * Process Virtio Config changed interrupt and call the callback
1155  * if link state changed.
1156  */
1157 static void
1158 virtio_interrupt_handler(struct rte_intr_handle *handle,
1159                          void *param)
1160 {
1161         struct rte_eth_dev *dev = param;
1162         struct virtio_hw *hw = dev->data->dev_private;
1163         uint8_t isr;
1164
1165         /* Read interrupt status which clears interrupt */
1166         isr = vtpci_isr(hw);
1167         PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1168
1169         if (rte_intr_enable(handle) < 0)
1170                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1171
1172         if (isr & VIRTIO_PCI_ISR_CONFIG) {
1173                 if (virtio_dev_link_update(dev, 0) == 0)
1174                         _rte_eth_dev_callback_process(dev,
1175                                                       RTE_ETH_EVENT_INTR_LSC, NULL);
1176         }
1177
1178 }
1179
1180 static void
1181 rx_func_get(struct rte_eth_dev *eth_dev)
1182 {
1183         struct virtio_hw *hw = eth_dev->data->dev_private;
1184         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF))
1185                 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1186         else
1187                 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1188 }
1189
1190 /* reset device and renegotiate features if needed */
1191 static int
1192 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1193 {
1194         struct virtio_hw *hw = eth_dev->data->dev_private;
1195         struct virtio_net_config *config;
1196         struct virtio_net_config local_config;
1197         struct rte_pci_device *pci_dev = hw->dev;
1198         int ret;
1199
1200         /* Reset the device although not necessary at startup */
1201         vtpci_reset(hw);
1202
1203         /* Tell the host we've noticed this device. */
1204         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1205
1206         /* Tell the host we've known how to drive the device. */
1207         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1208         if (virtio_negotiate_features(hw, req_features) < 0)
1209                 return -1;
1210
1211         /* If host does not support status then disable LSC */
1212         if (!vtpci_with_feature(hw, VIRTIO_NET_F_STATUS))
1213                 eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1214         else
1215                 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1216
1217         if (pci_dev) {
1218                 rte_eth_copy_pci_info(eth_dev, pci_dev);
1219                 eth_dev->data->dev_flags = RTE_ETH_DEV_DETACHABLE;
1220         }
1221
1222         rx_func_get(eth_dev);
1223
1224         /* Setting up rx_header size for the device */
1225         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1226             vtpci_with_feature(hw, VIRTIO_F_VERSION_1))
1227                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1228         else
1229                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1230
1231         /* Copy the permanent MAC address to: virtio_hw */
1232         virtio_get_hwaddr(hw);
1233         ether_addr_copy((struct ether_addr *) hw->mac_addr,
1234                         &eth_dev->data->mac_addrs[0]);
1235         PMD_INIT_LOG(DEBUG,
1236                      "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1237                      hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1238                      hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1239
1240         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1241                 config = &local_config;
1242
1243                 vtpci_read_dev_config(hw,
1244                         offsetof(struct virtio_net_config, mac),
1245                         &config->mac, sizeof(config->mac));
1246
1247                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1248                         vtpci_read_dev_config(hw,
1249                                 offsetof(struct virtio_net_config, status),
1250                                 &config->status, sizeof(config->status));
1251                 } else {
1252                         PMD_INIT_LOG(DEBUG,
1253                                      "VIRTIO_NET_F_STATUS is not supported");
1254                         config->status = 0;
1255                 }
1256
1257                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1258                         vtpci_read_dev_config(hw,
1259                                 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1260                                 &config->max_virtqueue_pairs,
1261                                 sizeof(config->max_virtqueue_pairs));
1262                 } else {
1263                         PMD_INIT_LOG(DEBUG,
1264                                      "VIRTIO_NET_F_MQ is not supported");
1265                         config->max_virtqueue_pairs = 1;
1266                 }
1267
1268                 hw->max_queue_pairs = config->max_virtqueue_pairs;
1269
1270                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1271                                 config->max_virtqueue_pairs);
1272                 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1273                 PMD_INIT_LOG(DEBUG,
1274                                 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1275                                 config->mac[0], config->mac[1],
1276                                 config->mac[2], config->mac[3],
1277                                 config->mac[4], config->mac[5]);
1278         } else {
1279                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1280                 hw->max_queue_pairs = 1;
1281         }
1282
1283         ret = virtio_alloc_queues(eth_dev);
1284         if (ret < 0)
1285                 return ret;
1286         vtpci_reinit_complete(hw);
1287
1288         if (pci_dev)
1289                 PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1290                         eth_dev->data->port_id, pci_dev->id.vendor_id,
1291                         pci_dev->id.device_id);
1292
1293         return 0;
1294 }
1295
1296 /*
1297  * This function is based on probe() function in virtio_pci.c
1298  * It returns 0 on success.
1299  */
1300 int
1301 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1302 {
1303         struct virtio_hw *hw = eth_dev->data->dev_private;
1304         uint32_t dev_flags = RTE_ETH_DEV_DETACHABLE;
1305         int ret;
1306
1307         RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1308
1309         eth_dev->dev_ops = &virtio_eth_dev_ops;
1310         eth_dev->tx_pkt_burst = &virtio_xmit_pkts;
1311
1312         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1313                 if (hw->use_simple_rxtx) {
1314                         eth_dev->tx_pkt_burst = virtio_xmit_pkts_simple;
1315                         eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1316                 } else {
1317                         rx_func_get(eth_dev);
1318                 }
1319                 return 0;
1320         }
1321
1322         /* Allocate memory for storing MAC addresses */
1323         eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
1324         if (eth_dev->data->mac_addrs == NULL) {
1325                 PMD_INIT_LOG(ERR,
1326                         "Failed to allocate %d bytes needed to store MAC addresses",
1327                         VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
1328                 return -ENOMEM;
1329         }
1330
1331         /* For virtio_user case the hw->virtio_user_dev is populated by
1332          * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1333          */
1334         if (!hw->virtio_user_dev) {
1335                 ret = vtpci_init(RTE_DEV_TO_PCI(eth_dev->device), hw,
1336                                  &dev_flags);
1337                 if (ret)
1338                         return ret;
1339         }
1340
1341         hw->port_id = eth_dev->data->port_id;
1342         eth_dev->data->dev_flags = dev_flags;
1343
1344         /* reset device and negotiate default features */
1345         ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1346         if (ret < 0)
1347                 return ret;
1348
1349         /* Setup interrupt callback  */
1350         if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1351                 rte_intr_callback_register(vtpci_intr_handle(hw),
1352                         virtio_interrupt_handler, eth_dev);
1353
1354         return 0;
1355 }
1356
1357 static int
1358 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1359 {
1360         struct virtio_hw *hw = eth_dev->data->dev_private;
1361
1362         PMD_INIT_FUNC_TRACE();
1363
1364         if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1365                 return -EPERM;
1366
1367         virtio_dev_stop(eth_dev);
1368         virtio_dev_close(eth_dev);
1369
1370         eth_dev->dev_ops = NULL;
1371         eth_dev->tx_pkt_burst = NULL;
1372         eth_dev->rx_pkt_burst = NULL;
1373
1374         rte_free(eth_dev->data->mac_addrs);
1375         eth_dev->data->mac_addrs = NULL;
1376
1377         /* reset interrupt callback  */
1378         if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1379                 rte_intr_callback_unregister(vtpci_intr_handle(hw),
1380                                                 virtio_interrupt_handler,
1381                                                 eth_dev);
1382         if (hw->dev)
1383                 rte_eal_pci_unmap_device(hw->dev);
1384
1385         PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1386
1387         return 0;
1388 }
1389
1390 static struct eth_driver rte_virtio_pmd = {
1391         .pci_drv = {
1392                 .driver = {
1393                         .name = "net_virtio",
1394                 },
1395                 .id_table = pci_id_virtio_map,
1396                 .drv_flags = 0,
1397                 .probe = rte_eth_dev_pci_probe,
1398                 .remove = rte_eth_dev_pci_remove,
1399         },
1400         .eth_dev_init = eth_virtio_dev_init,
1401         .eth_dev_uninit = eth_virtio_dev_uninit,
1402         .dev_private_size = sizeof(struct virtio_hw),
1403 };
1404
1405 RTE_INIT(rte_virtio_pmd_init);
1406 static void
1407 rte_virtio_pmd_init(void)
1408 {
1409         if (rte_eal_iopl_init() != 0) {
1410                 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1411                 return;
1412         }
1413
1414         rte_eal_pci_register(&rte_virtio_pmd.pci_drv);
1415 }
1416
1417 /*
1418  * Configure virtio device
1419  * It returns 0 on success.
1420  */
1421 static int
1422 virtio_dev_configure(struct rte_eth_dev *dev)
1423 {
1424         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1425         struct virtio_hw *hw = dev->data->dev_private;
1426         uint64_t req_features;
1427         int ret;
1428
1429         PMD_INIT_LOG(DEBUG, "configure");
1430         req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
1431         if (rxmode->hw_ip_checksum)
1432                 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
1433         if (rxmode->enable_lro)
1434                 req_features |=
1435                         (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
1436                         (1ULL << VIRTIO_NET_F_GUEST_TSO6);
1437
1438         /* if request features changed, reinit the device */
1439         if (req_features != hw->req_guest_features) {
1440                 ret = virtio_init_device(dev, req_features);
1441                 if (ret < 0)
1442                         return ret;
1443         }
1444
1445         if (rxmode->hw_ip_checksum &&
1446                 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
1447                 PMD_DRV_LOG(NOTICE,
1448                         "rx ip checksum not available on this host");
1449                 return -ENOTSUP;
1450         }
1451
1452         if (rxmode->enable_lro &&
1453                 (!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
1454                         !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4))) {
1455                 PMD_DRV_LOG(NOTICE,
1456                         "lro not available on this host");
1457                 return -ENOTSUP;
1458         }
1459
1460         /* start control queue */
1461         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
1462                 virtio_dev_cq_start(dev);
1463
1464         hw->vlan_strip = rxmode->hw_vlan_strip;
1465
1466         if (rxmode->hw_vlan_filter
1467             && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1468                 PMD_DRV_LOG(NOTICE,
1469                             "vlan filtering not available on this host");
1470                 return -ENOTSUP;
1471         }
1472
1473         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1474                 if (vtpci_irq_config(hw, 0) == VIRTIO_MSI_NO_VECTOR) {
1475                         PMD_DRV_LOG(ERR, "failed to set config vector");
1476                         return -EBUSY;
1477                 }
1478
1479         return 0;
1480 }
1481
1482
1483 static int
1484 virtio_dev_start(struct rte_eth_dev *dev)
1485 {
1486         uint16_t nb_queues, i;
1487         struct virtnet_rx *rxvq;
1488         struct virtnet_tx *txvq __rte_unused;
1489         struct virtio_hw *hw = dev->data->dev_private;
1490
1491         /* check if lsc interrupt feature is enabled */
1492         if (dev->data->dev_conf.intr_conf.lsc) {
1493                 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1494                         PMD_DRV_LOG(ERR, "link status not supported by host");
1495                         return -ENOTSUP;
1496                 }
1497
1498                 if (rte_intr_enable(vtpci_intr_handle(hw)) < 0) {
1499                         PMD_DRV_LOG(ERR, "interrupt enable failed");
1500                         return -EIO;
1501                 }
1502         }
1503
1504         /* Initialize Link state */
1505         virtio_dev_link_update(dev, 0);
1506
1507         /*Notify the backend
1508          *Otherwise the tap backend might already stop its queue due to fullness.
1509          *vhost backend will have no chance to be waked up
1510          */
1511         nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
1512         if (hw->max_queue_pairs > 1) {
1513                 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1514                         return -EINVAL;
1515         }
1516
1517         PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1518
1519         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1520                 rxvq = dev->data->rx_queues[i];
1521                 virtqueue_notify(rxvq->vq);
1522         }
1523
1524         PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1525
1526         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1527                 rxvq = dev->data->rx_queues[i];
1528                 VIRTQUEUE_DUMP(rxvq->vq);
1529         }
1530
1531         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1532                 txvq = dev->data->tx_queues[i];
1533                 VIRTQUEUE_DUMP(txvq->vq);
1534         }
1535
1536         return 0;
1537 }
1538
1539 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1540 {
1541         struct rte_mbuf *buf;
1542         int i, mbuf_num = 0;
1543
1544         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1545                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1546
1547                 PMD_INIT_LOG(DEBUG,
1548                              "Before freeing rxq[%d] used and unused buf", i);
1549                 VIRTQUEUE_DUMP(rxvq->vq);
1550
1551                 PMD_INIT_LOG(DEBUG, "rx_queues[%d]=%p", i, rxvq);
1552                 while ((buf = virtqueue_detatch_unused(rxvq->vq)) != NULL) {
1553                         rte_pktmbuf_free(buf);
1554                         mbuf_num++;
1555                 }
1556
1557                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1558                 PMD_INIT_LOG(DEBUG,
1559                              "After freeing rxq[%d] used and unused buf", i);
1560                 VIRTQUEUE_DUMP(rxvq->vq);
1561         }
1562
1563         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1564                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1565
1566                 PMD_INIT_LOG(DEBUG,
1567                              "Before freeing txq[%d] used and unused bufs",
1568                              i);
1569                 VIRTQUEUE_DUMP(txvq->vq);
1570
1571                 mbuf_num = 0;
1572                 while ((buf = virtqueue_detatch_unused(txvq->vq)) != NULL) {
1573                         rte_pktmbuf_free(buf);
1574                         mbuf_num++;
1575                 }
1576
1577                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1578                 PMD_INIT_LOG(DEBUG,
1579                              "After freeing txq[%d] used and unused buf", i);
1580                 VIRTQUEUE_DUMP(txvq->vq);
1581         }
1582 }
1583
1584 /*
1585  * Stop device: disable interrupt and mark link down
1586  */
1587 static void
1588 virtio_dev_stop(struct rte_eth_dev *dev)
1589 {
1590         struct virtio_hw *hw = dev->data->dev_private;
1591         struct rte_eth_link link;
1592
1593         PMD_INIT_LOG(DEBUG, "stop");
1594
1595         if (dev->data->dev_conf.intr_conf.lsc)
1596                 rte_intr_disable(vtpci_intr_handle(hw));
1597
1598         memset(&link, 0, sizeof(link));
1599         virtio_dev_atomic_write_link_status(dev, &link);
1600 }
1601
1602 static int
1603 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1604 {
1605         struct rte_eth_link link, old;
1606         uint16_t status;
1607         struct virtio_hw *hw = dev->data->dev_private;
1608         memset(&link, 0, sizeof(link));
1609         virtio_dev_atomic_read_link_status(dev, &link);
1610         old = link;
1611         link.link_duplex = ETH_LINK_FULL_DUPLEX;
1612         link.link_speed  = SPEED_10G;
1613
1614         if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1615                 PMD_INIT_LOG(DEBUG, "Get link status from hw");
1616                 vtpci_read_dev_config(hw,
1617                                 offsetof(struct virtio_net_config, status),
1618                                 &status, sizeof(status));
1619                 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1620                         link.link_status = ETH_LINK_DOWN;
1621                         PMD_INIT_LOG(DEBUG, "Port %d is down",
1622                                      dev->data->port_id);
1623                 } else {
1624                         link.link_status = ETH_LINK_UP;
1625                         PMD_INIT_LOG(DEBUG, "Port %d is up",
1626                                      dev->data->port_id);
1627                 }
1628         } else {
1629                 link.link_status = ETH_LINK_UP;
1630         }
1631         virtio_dev_atomic_write_link_status(dev, &link);
1632
1633         return (old.link_status == link.link_status) ? -1 : 0;
1634 }
1635
1636 static void
1637 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1638 {
1639         uint64_t tso_mask;
1640         struct virtio_hw *hw = dev->data->dev_private;
1641
1642         dev_info->pci_dev = hw->dev;
1643         dev_info->max_rx_queues =
1644                 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
1645         dev_info->max_tx_queues =
1646                 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
1647         dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1648         dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1649         dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1650         dev_info->default_txconf = (struct rte_eth_txconf) {
1651                 .txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
1652         };
1653         dev_info->rx_offload_capa =
1654                 DEV_RX_OFFLOAD_TCP_CKSUM |
1655                 DEV_RX_OFFLOAD_UDP_CKSUM |
1656                 DEV_RX_OFFLOAD_TCP_LRO;
1657         dev_info->tx_offload_capa = 0;
1658
1659         if (hw->guest_features & (1ULL << VIRTIO_NET_F_CSUM)) {
1660                 dev_info->tx_offload_capa |=
1661                         DEV_TX_OFFLOAD_UDP_CKSUM |
1662                         DEV_TX_OFFLOAD_TCP_CKSUM;
1663         }
1664
1665         tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
1666                 (1ULL << VIRTIO_NET_F_HOST_TSO6);
1667         if ((hw->guest_features & tso_mask) == tso_mask)
1668                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
1669 }
1670
1671 /*
1672  * It enables testpmd to collect per queue stats.
1673  */
1674 static int
1675 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
1676 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
1677 __rte_unused uint8_t is_rx)
1678 {
1679         return 0;
1680 }
1681
1682 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
1683 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
1684 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio");