]>
Commit | Line | Data |
---|---|---|
1e0a84ea CL |
1 | /* |
2 | * vhost-vdpa.c | |
3 | * | |
4 | * Copyright(c) 2017-2018 Intel Corporation. | |
5 | * Copyright(c) 2020 Red Hat, Inc. | |
6 | * | |
7 | * This work is licensed under the terms of the GNU GPL, version 2 or later. | |
8 | * See the COPYING file in the top-level directory. | |
9 | * | |
10 | */ | |
11 | ||
12 | #include "qemu/osdep.h" | |
13 | #include "clients.h" | |
bd907ae4 | 14 | #include "hw/virtio/virtio-net.h" |
1e0a84ea CL |
15 | #include "net/vhost_net.h" |
16 | #include "net/vhost-vdpa.h" | |
17 | #include "hw/virtio/vhost-vdpa.h" | |
18 | #include "qemu/config-file.h" | |
19 | #include "qemu/error-report.h" | |
bd907ae4 EP |
20 | #include "qemu/log.h" |
21 | #include "qemu/memalign.h" | |
1e0a84ea CL |
22 | #include "qemu/option.h" |
23 | #include "qapi/error.h" | |
40237840 | 24 | #include <linux/vhost.h> |
1e0a84ea CL |
25 | #include <sys/ioctl.h> |
26 | #include <err.h> | |
27 | #include "standard-headers/linux/virtio_net.h" | |
28 | #include "monitor/monitor.h" | |
69498430 EP |
29 | #include "migration/migration.h" |
30 | #include "migration/misc.h" | |
1e0a84ea CL |
31 | #include "hw/virtio/vhost.h" |
32 | ||
33 | /* Todo:need to add the multiqueue support here */ | |
34 | typedef struct VhostVDPAState { | |
35 | NetClientState nc; | |
36 | struct vhost_vdpa vhost_vdpa; | |
69498430 | 37 | Notifier migration_state; |
1e0a84ea | 38 | VHostNetState *vhost_net; |
2df4dd31 EP |
39 | |
40 | /* Control commands shadow buffers */ | |
17fb889f EP |
41 | void *cvq_cmd_out_buffer; |
42 | virtio_net_ctrl_ack *status; | |
43 | ||
7f211a28 EP |
44 | /* The device always have SVQ enabled */ |
45 | bool always_svq; | |
152128d6 EP |
46 | |
47 | /* The device can isolate CVQ in its own ASID */ | |
48 | bool cvq_isolated; | |
49 | ||
1e0a84ea CL |
50 | bool started; |
51 | } VhostVDPAState; | |
52 | ||
53 | const int vdpa_feature_bits[] = { | |
54 | VIRTIO_F_NOTIFY_ON_EMPTY, | |
55 | VIRTIO_RING_F_INDIRECT_DESC, | |
56 | VIRTIO_RING_F_EVENT_IDX, | |
57 | VIRTIO_F_ANY_LAYOUT, | |
58 | VIRTIO_F_VERSION_1, | |
59 | VIRTIO_NET_F_CSUM, | |
60 | VIRTIO_NET_F_GUEST_CSUM, | |
61 | VIRTIO_NET_F_GSO, | |
62 | VIRTIO_NET_F_GUEST_TSO4, | |
63 | VIRTIO_NET_F_GUEST_TSO6, | |
64 | VIRTIO_NET_F_GUEST_ECN, | |
65 | VIRTIO_NET_F_GUEST_UFO, | |
66 | VIRTIO_NET_F_HOST_TSO4, | |
67 | VIRTIO_NET_F_HOST_TSO6, | |
68 | VIRTIO_NET_F_HOST_ECN, | |
69 | VIRTIO_NET_F_HOST_UFO, | |
70 | VIRTIO_NET_F_MRG_RXBUF, | |
71 | VIRTIO_NET_F_MTU, | |
40237840 JW |
72 | VIRTIO_NET_F_CTRL_RX, |
73 | VIRTIO_NET_F_CTRL_RX_EXTRA, | |
74 | VIRTIO_NET_F_CTRL_VLAN, | |
40237840 JW |
75 | VIRTIO_NET_F_CTRL_MAC_ADDR, |
76 | VIRTIO_NET_F_RSS, | |
77 | VIRTIO_NET_F_MQ, | |
78 | VIRTIO_NET_F_CTRL_VQ, | |
1e0a84ea CL |
79 | VIRTIO_F_IOMMU_PLATFORM, |
80 | VIRTIO_F_RING_PACKED, | |
562a7d23 | 81 | VIRTIO_F_RING_RESET, |
0145c393 AM |
82 | VIRTIO_NET_F_RSS, |
83 | VIRTIO_NET_F_HASH_REPORT, | |
9aa47edd | 84 | VIRTIO_NET_F_STATUS, |
1e0a84ea CL |
85 | VHOST_INVALID_FEATURE_BIT |
86 | }; | |
87 | ||
1576dbb5 EP |
88 | /** Supported device specific feature bits with SVQ */ |
89 | static const uint64_t vdpa_svq_device_features = | |
90 | BIT_ULL(VIRTIO_NET_F_CSUM) | | |
91 | BIT_ULL(VIRTIO_NET_F_GUEST_CSUM) | | |
92 | BIT_ULL(VIRTIO_NET_F_MTU) | | |
93 | BIT_ULL(VIRTIO_NET_F_MAC) | | |
94 | BIT_ULL(VIRTIO_NET_F_GUEST_TSO4) | | |
95 | BIT_ULL(VIRTIO_NET_F_GUEST_TSO6) | | |
96 | BIT_ULL(VIRTIO_NET_F_GUEST_ECN) | | |
97 | BIT_ULL(VIRTIO_NET_F_GUEST_UFO) | | |
98 | BIT_ULL(VIRTIO_NET_F_HOST_TSO4) | | |
99 | BIT_ULL(VIRTIO_NET_F_HOST_TSO6) | | |
100 | BIT_ULL(VIRTIO_NET_F_HOST_ECN) | | |
101 | BIT_ULL(VIRTIO_NET_F_HOST_UFO) | | |
102 | BIT_ULL(VIRTIO_NET_F_MRG_RXBUF) | | |
103 | BIT_ULL(VIRTIO_NET_F_STATUS) | | |
104 | BIT_ULL(VIRTIO_NET_F_CTRL_VQ) | | |
72b99a87 | 105 | BIT_ULL(VIRTIO_NET_F_MQ) | |
1576dbb5 EP |
106 | BIT_ULL(VIRTIO_F_ANY_LAYOUT) | |
107 | BIT_ULL(VIRTIO_NET_F_CTRL_MAC_ADDR) | | |
609ab4c3 EP |
108 | /* VHOST_F_LOG_ALL is exposed by SVQ */ |
109 | BIT_ULL(VHOST_F_LOG_ALL) | | |
1576dbb5 | 110 | BIT_ULL(VIRTIO_NET_F_RSC_EXT) | |
0d74e2b7 EP |
111 | BIT_ULL(VIRTIO_NET_F_STANDBY) | |
112 | BIT_ULL(VIRTIO_NET_F_SPEED_DUPLEX); | |
1576dbb5 | 113 | |
c1a10086 EP |
114 | #define VHOST_VDPA_NET_CVQ_ASID 1 |
115 | ||
1e0a84ea CL |
116 | VHostNetState *vhost_vdpa_get_vhost_net(NetClientState *nc) |
117 | { | |
118 | VhostVDPAState *s = DO_UPCAST(VhostVDPAState, nc, nc); | |
119 | assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA); | |
120 | return s->vhost_net; | |
121 | } | |
122 | ||
915bf6cc EP |
123 | static size_t vhost_vdpa_net_cvq_cmd_len(void) |
124 | { | |
125 | /* | |
126 | * MAC_TABLE_SET is the ctrl command that produces the longer out buffer. | |
127 | * In buffer is always 1 byte, so it should fit here | |
128 | */ | |
129 | return sizeof(struct virtio_net_ctrl_hdr) + | |
130 | 2 * sizeof(struct virtio_net_ctrl_mac) + | |
131 | MAC_TABLE_ENTRIES * ETH_ALEN; | |
132 | } | |
133 | ||
134 | static size_t vhost_vdpa_net_cvq_cmd_page_len(void) | |
135 | { | |
136 | return ROUND_UP(vhost_vdpa_net_cvq_cmd_len(), qemu_real_host_page_size()); | |
137 | } | |
138 | ||
36e46472 EP |
139 | static bool vhost_vdpa_net_valid_svq_features(uint64_t features, Error **errp) |
140 | { | |
141 | uint64_t invalid_dev_features = | |
142 | features & ~vdpa_svq_device_features & | |
143 | /* Transport are all accepted at this point */ | |
144 | ~MAKE_64BIT_MASK(VIRTIO_TRANSPORT_F_START, | |
145 | VIRTIO_TRANSPORT_F_END - VIRTIO_TRANSPORT_F_START); | |
146 | ||
147 | if (invalid_dev_features) { | |
148 | error_setg(errp, "vdpa svq does not work with features 0x%" PRIx64, | |
149 | invalid_dev_features); | |
258a0394 | 150 | return false; |
36e46472 EP |
151 | } |
152 | ||
258a0394 | 153 | return vhost_svq_valid_features(features, errp); |
36e46472 EP |
154 | } |
155 | ||
1e0a84ea CL |
156 | static int vhost_vdpa_net_check_device_id(struct vhost_net *net) |
157 | { | |
158 | uint32_t device_id; | |
159 | int ret; | |
160 | struct vhost_dev *hdev; | |
161 | ||
162 | hdev = (struct vhost_dev *)&net->dev; | |
163 | ret = hdev->vhost_ops->vhost_get_device_id(hdev, &device_id); | |
164 | if (device_id != VIRTIO_ID_NET) { | |
165 | return -ENOTSUP; | |
166 | } | |
167 | return ret; | |
168 | } | |
169 | ||
40237840 JW |
170 | static int vhost_vdpa_add(NetClientState *ncs, void *be, |
171 | int queue_pair_index, int nvqs) | |
1e0a84ea CL |
172 | { |
173 | VhostNetOptions options; | |
174 | struct vhost_net *net = NULL; | |
175 | VhostVDPAState *s; | |
176 | int ret; | |
177 | ||
178 | options.backend_type = VHOST_BACKEND_TYPE_VDPA; | |
179 | assert(ncs->info->type == NET_CLIENT_DRIVER_VHOST_VDPA); | |
180 | s = DO_UPCAST(VhostVDPAState, nc, ncs); | |
181 | options.net_backend = ncs; | |
182 | options.opaque = be; | |
183 | options.busyloop_timeout = 0; | |
40237840 | 184 | options.nvqs = nvqs; |
1e0a84ea CL |
185 | |
186 | net = vhost_net_init(&options); | |
187 | if (!net) { | |
188 | error_report("failed to init vhost_net for queue"); | |
a97ef87a | 189 | goto err_init; |
1e0a84ea | 190 | } |
1e0a84ea CL |
191 | s->vhost_net = net; |
192 | ret = vhost_vdpa_net_check_device_id(net); | |
193 | if (ret) { | |
a97ef87a | 194 | goto err_check; |
1e0a84ea CL |
195 | } |
196 | return 0; | |
a97ef87a JW |
197 | err_check: |
198 | vhost_net_cleanup(net); | |
199 | g_free(net); | |
200 | err_init: | |
1e0a84ea CL |
201 | return -1; |
202 | } | |
203 | ||
204 | static void vhost_vdpa_cleanup(NetClientState *nc) | |
205 | { | |
206 | VhostVDPAState *s = DO_UPCAST(VhostVDPAState, nc, nc); | |
207 | ||
babf8b87 EP |
208 | munmap(s->cvq_cmd_out_buffer, vhost_vdpa_net_cvq_cmd_page_len()); |
209 | munmap(s->status, vhost_vdpa_net_cvq_cmd_page_len()); | |
1e0a84ea CL |
210 | if (s->vhost_net) { |
211 | vhost_net_cleanup(s->vhost_net); | |
212 | g_free(s->vhost_net); | |
213 | s->vhost_net = NULL; | |
214 | } | |
57b3a7d8 CL |
215 | if (s->vhost_vdpa.device_fd >= 0) { |
216 | qemu_close(s->vhost_vdpa.device_fd); | |
217 | s->vhost_vdpa.device_fd = -1; | |
218 | } | |
1e0a84ea CL |
219 | } |
220 | ||
221 | static bool vhost_vdpa_has_vnet_hdr(NetClientState *nc) | |
222 | { | |
223 | assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA); | |
224 | ||
225 | return true; | |
226 | } | |
227 | ||
228 | static bool vhost_vdpa_has_ufo(NetClientState *nc) | |
229 | { | |
230 | assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA); | |
231 | VhostVDPAState *s = DO_UPCAST(VhostVDPAState, nc, nc); | |
232 | uint64_t features = 0; | |
233 | features |= (1ULL << VIRTIO_NET_F_HOST_UFO); | |
234 | features = vhost_net_get_features(s->vhost_net, features); | |
235 | return !!(features & (1ULL << VIRTIO_NET_F_HOST_UFO)); | |
236 | ||
237 | } | |
238 | ||
ee8a1c63 KW |
239 | static bool vhost_vdpa_check_peer_type(NetClientState *nc, ObjectClass *oc, |
240 | Error **errp) | |
241 | { | |
242 | const char *driver = object_class_get_name(oc); | |
243 | ||
244 | if (!g_str_has_prefix(driver, "virtio-net-")) { | |
245 | error_setg(errp, "vhost-vdpa requires frontend driver virtio-net-*"); | |
246 | return false; | |
247 | } | |
248 | ||
249 | return true; | |
250 | } | |
251 | ||
846a1e85 EP |
252 | /** Dummy receive in case qemu falls back to userland tap networking */ |
253 | static ssize_t vhost_vdpa_receive(NetClientState *nc, const uint8_t *buf, | |
254 | size_t size) | |
255 | { | |
bc5add1d | 256 | return size; |
846a1e85 EP |
257 | } |
258 | ||
00ef422e EP |
259 | /** From any vdpa net client, get the netclient of the first queue pair */ |
260 | static VhostVDPAState *vhost_vdpa_net_first_nc_vdpa(VhostVDPAState *s) | |
261 | { | |
262 | NICState *nic = qemu_get_nic(s->nc.peer); | |
263 | NetClientState *nc0 = qemu_get_peer(nic->ncs, 0); | |
264 | ||
265 | return DO_UPCAST(VhostVDPAState, nc, nc0); | |
266 | } | |
267 | ||
69498430 EP |
268 | static void vhost_vdpa_net_log_global_enable(VhostVDPAState *s, bool enable) |
269 | { | |
270 | struct vhost_vdpa *v = &s->vhost_vdpa; | |
271 | VirtIONet *n; | |
272 | VirtIODevice *vdev; | |
273 | int data_queue_pairs, cvq, r; | |
274 | ||
275 | /* We are only called on the first data vqs and only if x-svq is not set */ | |
276 | if (s->vhost_vdpa.shadow_vqs_enabled == enable) { | |
277 | return; | |
278 | } | |
279 | ||
280 | vdev = v->dev->vdev; | |
281 | n = VIRTIO_NET(vdev); | |
282 | if (!n->vhost_started) { | |
283 | return; | |
284 | } | |
285 | ||
286 | data_queue_pairs = n->multiqueue ? n->max_queue_pairs : 1; | |
287 | cvq = virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) ? | |
288 | n->max_ncs - n->max_queue_pairs : 0; | |
289 | /* | |
290 | * TODO: vhost_net_stop does suspend, get_base and reset. We can be smarter | |
291 | * in the future and resume the device if read-only operations between | |
292 | * suspend and reset goes wrong. | |
293 | */ | |
294 | vhost_net_stop(vdev, n->nic->ncs, data_queue_pairs, cvq); | |
295 | ||
296 | /* Start will check migration setup_or_active to configure or not SVQ */ | |
297 | r = vhost_net_start(vdev, n->nic->ncs, data_queue_pairs, cvq); | |
298 | if (unlikely(r < 0)) { | |
299 | error_report("unable to start vhost net: %s(%d)", g_strerror(-r), -r); | |
300 | } | |
301 | } | |
302 | ||
303 | static void vdpa_net_migration_state_notifier(Notifier *notifier, void *data) | |
304 | { | |
305 | MigrationState *migration = data; | |
306 | VhostVDPAState *s = container_of(notifier, VhostVDPAState, | |
307 | migration_state); | |
308 | ||
309 | if (migration_in_setup(migration)) { | |
310 | vhost_vdpa_net_log_global_enable(s, true); | |
311 | } else if (migration_has_failed(migration)) { | |
312 | vhost_vdpa_net_log_global_enable(s, false); | |
313 | } | |
314 | } | |
315 | ||
00ef422e EP |
316 | static void vhost_vdpa_net_data_start_first(VhostVDPAState *s) |
317 | { | |
318 | struct vhost_vdpa *v = &s->vhost_vdpa; | |
319 | ||
69498430 | 320 | add_migration_state_change_notifier(&s->migration_state); |
00ef422e EP |
321 | if (v->shadow_vqs_enabled) { |
322 | v->iova_tree = vhost_iova_tree_new(v->iova_range.first, | |
323 | v->iova_range.last); | |
324 | } | |
325 | } | |
326 | ||
327 | static int vhost_vdpa_net_data_start(NetClientState *nc) | |
328 | { | |
329 | VhostVDPAState *s = DO_UPCAST(VhostVDPAState, nc, nc); | |
330 | struct vhost_vdpa *v = &s->vhost_vdpa; | |
331 | ||
332 | assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA); | |
333 | ||
69498430 EP |
334 | if (s->always_svq || |
335 | migration_is_setup_or_active(migrate_get_current()->state)) { | |
336 | v->shadow_vqs_enabled = true; | |
337 | v->shadow_data = true; | |
338 | } else { | |
339 | v->shadow_vqs_enabled = false; | |
340 | v->shadow_data = false; | |
341 | } | |
342 | ||
00ef422e EP |
343 | if (v->index == 0) { |
344 | vhost_vdpa_net_data_start_first(s); | |
345 | return 0; | |
346 | } | |
347 | ||
348 | if (v->shadow_vqs_enabled) { | |
349 | VhostVDPAState *s0 = vhost_vdpa_net_first_nc_vdpa(s); | |
350 | v->iova_tree = s0->vhost_vdpa.iova_tree; | |
351 | } | |
352 | ||
353 | return 0; | |
354 | } | |
355 | ||
356 | static void vhost_vdpa_net_client_stop(NetClientState *nc) | |
357 | { | |
358 | VhostVDPAState *s = DO_UPCAST(VhostVDPAState, nc, nc); | |
359 | struct vhost_dev *dev; | |
360 | ||
361 | assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA); | |
362 | ||
69498430 EP |
363 | if (s->vhost_vdpa.index == 0) { |
364 | remove_migration_state_change_notifier(&s->migration_state); | |
365 | } | |
366 | ||
00ef422e EP |
367 | dev = s->vhost_vdpa.dev; |
368 | if (dev->vq_index + dev->nvqs == dev->vq_index_end) { | |
369 | g_clear_pointer(&s->vhost_vdpa.iova_tree, vhost_iova_tree_delete); | |
370 | } | |
371 | } | |
372 | ||
1e0a84ea CL |
373 | static NetClientInfo net_vhost_vdpa_info = { |
374 | .type = NET_CLIENT_DRIVER_VHOST_VDPA, | |
375 | .size = sizeof(VhostVDPAState), | |
846a1e85 | 376 | .receive = vhost_vdpa_receive, |
00ef422e EP |
377 | .start = vhost_vdpa_net_data_start, |
378 | .stop = vhost_vdpa_net_client_stop, | |
1e0a84ea CL |
379 | .cleanup = vhost_vdpa_cleanup, |
380 | .has_vnet_hdr = vhost_vdpa_has_vnet_hdr, | |
381 | .has_ufo = vhost_vdpa_has_ufo, | |
ee8a1c63 | 382 | .check_peer_type = vhost_vdpa_check_peer_type, |
1e0a84ea CL |
383 | }; |
384 | ||
152128d6 EP |
385 | static int64_t vhost_vdpa_get_vring_group(int device_fd, unsigned vq_index, |
386 | Error **errp) | |
c1a10086 EP |
387 | { |
388 | struct vhost_vring_state state = { | |
389 | .index = vq_index, | |
390 | }; | |
391 | int r = ioctl(device_fd, VHOST_VDPA_GET_VRING_GROUP, &state); | |
392 | ||
393 | if (unlikely(r < 0)) { | |
0f2bb0bf | 394 | r = -errno; |
152128d6 | 395 | error_setg_errno(errp, errno, "Cannot get VQ %u group", vq_index); |
c1a10086 EP |
396 | return r; |
397 | } | |
398 | ||
399 | return state.num; | |
400 | } | |
401 | ||
402 | static int vhost_vdpa_set_address_space_id(struct vhost_vdpa *v, | |
403 | unsigned vq_group, | |
404 | unsigned asid_num) | |
405 | { | |
406 | struct vhost_vring_state asid = { | |
407 | .index = vq_group, | |
408 | .num = asid_num, | |
409 | }; | |
410 | int r; | |
411 | ||
412 | r = ioctl(v->device_fd, VHOST_VDPA_SET_GROUP_ASID, &asid); | |
413 | if (unlikely(r < 0)) { | |
414 | error_report("Can't set vq group %u asid %u, errno=%d (%s)", | |
415 | asid.index, asid.num, errno, g_strerror(errno)); | |
416 | } | |
417 | return r; | |
418 | } | |
419 | ||
2df4dd31 EP |
420 | static void vhost_vdpa_cvq_unmap_buf(struct vhost_vdpa *v, void *addr) |
421 | { | |
422 | VhostIOVATree *tree = v->iova_tree; | |
423 | DMAMap needle = { | |
424 | /* | |
425 | * No need to specify size or to look for more translations since | |
426 | * this contiguous chunk was allocated by us. | |
427 | */ | |
428 | .translated_addr = (hwaddr)(uintptr_t)addr, | |
429 | }; | |
430 | const DMAMap *map = vhost_iova_tree_find_iova(tree, &needle); | |
431 | int r; | |
432 | ||
433 | if (unlikely(!map)) { | |
434 | error_report("Cannot locate expected map"); | |
435 | return; | |
436 | } | |
437 | ||
cd831ed5 | 438 | r = vhost_vdpa_dma_unmap(v, v->address_space_id, map->iova, map->size + 1); |
2df4dd31 EP |
439 | if (unlikely(r != 0)) { |
440 | error_report("Device cannot unmap: %s(%d)", g_strerror(r), r); | |
441 | } | |
442 | ||
69292a8e | 443 | vhost_iova_tree_remove(tree, *map); |
2df4dd31 EP |
444 | } |
445 | ||
7a7f87e9 EP |
446 | /** Map CVQ buffer. */ |
447 | static int vhost_vdpa_cvq_map_buf(struct vhost_vdpa *v, void *buf, size_t size, | |
448 | bool write) | |
2df4dd31 EP |
449 | { |
450 | DMAMap map = {}; | |
451 | int r; | |
452 | ||
2df4dd31 | 453 | map.translated_addr = (hwaddr)(uintptr_t)buf; |
7a7f87e9 | 454 | map.size = size - 1; |
2df4dd31 EP |
455 | map.perm = write ? IOMMU_RW : IOMMU_RO, |
456 | r = vhost_iova_tree_map_alloc(v->iova_tree, &map); | |
457 | if (unlikely(r != IOVA_OK)) { | |
458 | error_report("Cannot map injected element"); | |
7a7f87e9 | 459 | return r; |
2df4dd31 EP |
460 | } |
461 | ||
cd831ed5 EP |
462 | r = vhost_vdpa_dma_map(v, v->address_space_id, map.iova, |
463 | vhost_vdpa_net_cvq_cmd_page_len(), buf, !write); | |
2df4dd31 EP |
464 | if (unlikely(r < 0)) { |
465 | goto dma_map_err; | |
466 | } | |
467 | ||
7a7f87e9 | 468 | return 0; |
2df4dd31 EP |
469 | |
470 | dma_map_err: | |
69292a8e | 471 | vhost_iova_tree_remove(v->iova_tree, map); |
7a7f87e9 | 472 | return r; |
2df4dd31 EP |
473 | } |
474 | ||
7a7f87e9 | 475 | static int vhost_vdpa_net_cvq_start(NetClientState *nc) |
2df4dd31 | 476 | { |
00ef422e | 477 | VhostVDPAState *s, *s0; |
c1a10086 | 478 | struct vhost_vdpa *v; |
c1a10086 | 479 | int64_t cvq_group; |
152128d6 EP |
480 | int r; |
481 | Error *err = NULL; | |
2df4dd31 | 482 | |
7a7f87e9 EP |
483 | assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA); |
484 | ||
485 | s = DO_UPCAST(VhostVDPAState, nc, nc); | |
c1a10086 EP |
486 | v = &s->vhost_vdpa; |
487 | ||
69498430 EP |
488 | s0 = vhost_vdpa_net_first_nc_vdpa(s); |
489 | v->shadow_data = s0->vhost_vdpa.shadow_vqs_enabled; | |
c1a10086 EP |
490 | v->shadow_vqs_enabled = s->always_svq; |
491 | s->vhost_vdpa.address_space_id = VHOST_VDPA_GUEST_PA_ASID; | |
492 | ||
69498430 | 493 | if (s->vhost_vdpa.shadow_data) { |
c1a10086 EP |
494 | /* SVQ is already configured for all virtqueues */ |
495 | goto out; | |
496 | } | |
497 | ||
498 | /* | |
499 | * If we early return in these cases SVQ will not be enabled. The migration | |
500 | * will be blocked as long as vhost-vdpa backends will not offer _F_LOG. | |
c1a10086 | 501 | */ |
152128d6 EP |
502 | if (!vhost_vdpa_net_valid_svq_features(v->dev->features, NULL)) { |
503 | return 0; | |
c1a10086 | 504 | } |
152128d6 EP |
505 | |
506 | if (!s->cvq_isolated) { | |
c1a10086 EP |
507 | return 0; |
508 | } | |
509 | ||
152128d6 EP |
510 | cvq_group = vhost_vdpa_get_vring_group(v->device_fd, |
511 | v->dev->vq_index_end - 1, | |
512 | &err); | |
c1a10086 | 513 | if (unlikely(cvq_group < 0)) { |
152128d6 | 514 | error_report_err(err); |
c1a10086 EP |
515 | return cvq_group; |
516 | } | |
c1a10086 EP |
517 | |
518 | r = vhost_vdpa_set_address_space_id(v, cvq_group, VHOST_VDPA_NET_CVQ_ASID); | |
519 | if (unlikely(r < 0)) { | |
520 | return r; | |
521 | } | |
522 | ||
c1a10086 EP |
523 | v->shadow_vqs_enabled = true; |
524 | s->vhost_vdpa.address_space_id = VHOST_VDPA_NET_CVQ_ASID; | |
525 | ||
526 | out: | |
7a7f87e9 EP |
527 | if (!s->vhost_vdpa.shadow_vqs_enabled) { |
528 | return 0; | |
2df4dd31 EP |
529 | } |
530 | ||
00ef422e EP |
531 | if (s0->vhost_vdpa.iova_tree) { |
532 | /* | |
533 | * SVQ is already configured for all virtqueues. Reuse IOVA tree for | |
534 | * simplicity, whether CVQ shares ASID with guest or not, because: | |
535 | * - Memory listener need access to guest's memory addresses allocated | |
536 | * in the IOVA tree. | |
537 | * - There should be plenty of IOVA address space for both ASID not to | |
538 | * worry about collisions between them. Guest's translations are | |
539 | * still validated with virtio virtqueue_pop so there is no risk for | |
540 | * the guest to access memory that it shouldn't. | |
541 | * | |
542 | * To allocate a iova tree per ASID is doable but it complicates the | |
543 | * code and it is not worth it for the moment. | |
544 | */ | |
545 | v->iova_tree = s0->vhost_vdpa.iova_tree; | |
546 | } else { | |
547 | v->iova_tree = vhost_iova_tree_new(v->iova_range.first, | |
548 | v->iova_range.last); | |
549 | } | |
550 | ||
7a7f87e9 EP |
551 | r = vhost_vdpa_cvq_map_buf(&s->vhost_vdpa, s->cvq_cmd_out_buffer, |
552 | vhost_vdpa_net_cvq_cmd_page_len(), false); | |
553 | if (unlikely(r < 0)) { | |
554 | return r; | |
555 | } | |
556 | ||
17fb889f | 557 | r = vhost_vdpa_cvq_map_buf(&s->vhost_vdpa, s->status, |
7a7f87e9 EP |
558 | vhost_vdpa_net_cvq_cmd_page_len(), true); |
559 | if (unlikely(r < 0)) { | |
2df4dd31 | 560 | vhost_vdpa_cvq_unmap_buf(&s->vhost_vdpa, s->cvq_cmd_out_buffer); |
2df4dd31 EP |
561 | } |
562 | ||
7a7f87e9 EP |
563 | return r; |
564 | } | |
565 | ||
566 | static void vhost_vdpa_net_cvq_stop(NetClientState *nc) | |
567 | { | |
568 | VhostVDPAState *s = DO_UPCAST(VhostVDPAState, nc, nc); | |
569 | ||
570 | assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA); | |
571 | ||
572 | if (s->vhost_vdpa.shadow_vqs_enabled) { | |
573 | vhost_vdpa_cvq_unmap_buf(&s->vhost_vdpa, s->cvq_cmd_out_buffer); | |
17fb889f | 574 | vhost_vdpa_cvq_unmap_buf(&s->vhost_vdpa, s->status); |
7a7f87e9 | 575 | } |
00ef422e EP |
576 | |
577 | vhost_vdpa_net_client_stop(nc); | |
2df4dd31 EP |
578 | } |
579 | ||
be4278b6 EP |
580 | static ssize_t vhost_vdpa_net_cvq_add(VhostVDPAState *s, size_t out_len, |
581 | size_t in_len) | |
582 | { | |
583 | /* Buffers for the device */ | |
584 | const struct iovec out = { | |
585 | .iov_base = s->cvq_cmd_out_buffer, | |
586 | .iov_len = out_len, | |
587 | }; | |
588 | const struct iovec in = { | |
17fb889f | 589 | .iov_base = s->status, |
be4278b6 EP |
590 | .iov_len = sizeof(virtio_net_ctrl_ack), |
591 | }; | |
592 | VhostShadowVirtqueue *svq = g_ptr_array_index(s->vhost_vdpa.shadow_vqs, 0); | |
593 | int r; | |
594 | ||
595 | r = vhost_svq_add(svq, &out, 1, &in, 1, NULL); | |
596 | if (unlikely(r != 0)) { | |
597 | if (unlikely(r == -ENOSPC)) { | |
598 | qemu_log_mask(LOG_GUEST_ERROR, "%s: No space on device queue\n", | |
599 | __func__); | |
600 | } | |
601 | return r; | |
602 | } | |
603 | ||
604 | /* | |
605 | * We can poll here since we've had BQL from the time we sent the | |
606 | * descriptor. Also, we need to take the answer before SVQ pulls by itself, | |
607 | * when BQL is released | |
608 | */ | |
609 | return vhost_svq_poll(svq); | |
610 | } | |
611 | ||
f73c0c43 EP |
612 | static ssize_t vhost_vdpa_net_load_cmd(VhostVDPAState *s, uint8_t class, |
613 | uint8_t cmd, const void *data, | |
614 | size_t data_size) | |
615 | { | |
616 | const struct virtio_net_ctrl_hdr ctrl = { | |
617 | .class = class, | |
618 | .cmd = cmd, | |
619 | }; | |
620 | ||
621 | assert(data_size < vhost_vdpa_net_cvq_cmd_page_len() - sizeof(ctrl)); | |
622 | ||
623 | memcpy(s->cvq_cmd_out_buffer, &ctrl, sizeof(ctrl)); | |
624 | memcpy(s->cvq_cmd_out_buffer + sizeof(ctrl), data, data_size); | |
625 | ||
626 | return vhost_vdpa_net_cvq_add(s, sizeof(ctrl) + data_size, | |
627 | sizeof(virtio_net_ctrl_ack)); | |
628 | } | |
629 | ||
630 | static int vhost_vdpa_net_load_mac(VhostVDPAState *s, const VirtIONet *n) | |
631 | { | |
02d3bf09 | 632 | if (virtio_vdev_has_feature(&n->parent_obj, VIRTIO_NET_F_CTRL_MAC_ADDR)) { |
f73c0c43 EP |
633 | ssize_t dev_written = vhost_vdpa_net_load_cmd(s, VIRTIO_NET_CTRL_MAC, |
634 | VIRTIO_NET_CTRL_MAC_ADDR_SET, | |
635 | n->mac, sizeof(n->mac)); | |
636 | if (unlikely(dev_written < 0)) { | |
637 | return dev_written; | |
638 | } | |
639 | ||
640 | return *s->status != VIRTIO_NET_OK; | |
641 | } | |
642 | ||
643 | return 0; | |
644 | } | |
645 | ||
f64c7cda EP |
646 | static int vhost_vdpa_net_load_mq(VhostVDPAState *s, |
647 | const VirtIONet *n) | |
648 | { | |
649 | struct virtio_net_ctrl_mq mq; | |
f64c7cda EP |
650 | ssize_t dev_written; |
651 | ||
02d3bf09 | 652 | if (!virtio_vdev_has_feature(&n->parent_obj, VIRTIO_NET_F_MQ)) { |
f64c7cda EP |
653 | return 0; |
654 | } | |
655 | ||
656 | mq.virtqueue_pairs = cpu_to_le16(n->curr_queue_pairs); | |
657 | dev_written = vhost_vdpa_net_load_cmd(s, VIRTIO_NET_CTRL_MQ, | |
658 | VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &mq, | |
659 | sizeof(mq)); | |
660 | if (unlikely(dev_written < 0)) { | |
661 | return dev_written; | |
662 | } | |
663 | ||
664 | return *s->status != VIRTIO_NET_OK; | |
665 | } | |
666 | ||
dd036d8d EP |
667 | static int vhost_vdpa_net_load(NetClientState *nc) |
668 | { | |
669 | VhostVDPAState *s = DO_UPCAST(VhostVDPAState, nc, nc); | |
f73c0c43 | 670 | struct vhost_vdpa *v = &s->vhost_vdpa; |
dd036d8d | 671 | const VirtIONet *n; |
f73c0c43 | 672 | int r; |
dd036d8d EP |
673 | |
674 | assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA); | |
675 | ||
676 | if (!v->shadow_vqs_enabled) { | |
677 | return 0; | |
678 | } | |
679 | ||
680 | n = VIRTIO_NET(v->dev->vdev); | |
f73c0c43 EP |
681 | r = vhost_vdpa_net_load_mac(s, n); |
682 | if (unlikely(r < 0)) { | |
683 | return r; | |
dd036d8d | 684 | } |
f64c7cda EP |
685 | r = vhost_vdpa_net_load_mq(s, n); |
686 | if (unlikely(r)) { | |
687 | return r; | |
688 | } | |
dd036d8d EP |
689 | |
690 | return 0; | |
691 | } | |
692 | ||
f8972b56 EP |
693 | static NetClientInfo net_vhost_vdpa_cvq_info = { |
694 | .type = NET_CLIENT_DRIVER_VHOST_VDPA, | |
695 | .size = sizeof(VhostVDPAState), | |
696 | .receive = vhost_vdpa_receive, | |
7a7f87e9 | 697 | .start = vhost_vdpa_net_cvq_start, |
dd036d8d | 698 | .load = vhost_vdpa_net_load, |
7a7f87e9 | 699 | .stop = vhost_vdpa_net_cvq_stop, |
f8972b56 EP |
700 | .cleanup = vhost_vdpa_cleanup, |
701 | .has_vnet_hdr = vhost_vdpa_has_vnet_hdr, | |
702 | .has_ufo = vhost_vdpa_has_ufo, | |
703 | .check_peer_type = vhost_vdpa_check_peer_type, | |
704 | }; | |
705 | ||
2df4dd31 EP |
706 | /** |
707 | * Validate and copy control virtqueue commands. | |
708 | * | |
709 | * Following QEMU guidelines, we offer a copy of the buffers to the device to | |
710 | * prevent TOCTOU bugs. | |
bd907ae4 EP |
711 | */ |
712 | static int vhost_vdpa_net_handle_ctrl_avail(VhostShadowVirtqueue *svq, | |
713 | VirtQueueElement *elem, | |
714 | void *opaque) | |
715 | { | |
2df4dd31 | 716 | VhostVDPAState *s = opaque; |
be4278b6 | 717 | size_t in_len; |
bd907ae4 | 718 | virtio_net_ctrl_ack status = VIRTIO_NET_ERR; |
7a7f87e9 EP |
719 | /* Out buffer sent to both the vdpa device and the device model */ |
720 | struct iovec out = { | |
721 | .iov_base = s->cvq_cmd_out_buffer, | |
722 | }; | |
2df4dd31 EP |
723 | /* in buffer used for device model */ |
724 | const struct iovec in = { | |
725 | .iov_base = &status, | |
726 | .iov_len = sizeof(status), | |
727 | }; | |
be4278b6 | 728 | ssize_t dev_written = -EINVAL; |
2df4dd31 | 729 | |
7a7f87e9 EP |
730 | out.iov_len = iov_to_buf(elem->out_sg, elem->out_num, 0, |
731 | s->cvq_cmd_out_buffer, | |
732 | vhost_vdpa_net_cvq_cmd_len()); | |
3f9a3eeb EP |
733 | if (*(uint8_t *)s->cvq_cmd_out_buffer == VIRTIO_NET_CTRL_ANNOUNCE) { |
734 | /* | |
735 | * Guest announce capability is emulated by qemu, so don't forward to | |
736 | * the device. | |
737 | */ | |
738 | dev_written = sizeof(status); | |
739 | *s->status = VIRTIO_NET_OK; | |
740 | } else { | |
741 | dev_written = vhost_vdpa_net_cvq_add(s, out.iov_len, sizeof(status)); | |
742 | if (unlikely(dev_written < 0)) { | |
743 | goto out; | |
744 | } | |
bd907ae4 EP |
745 | } |
746 | ||
bd907ae4 EP |
747 | if (unlikely(dev_written < sizeof(status))) { |
748 | error_report("Insufficient written data (%zu)", dev_written); | |
2df4dd31 EP |
749 | goto out; |
750 | } | |
751 | ||
17fb889f | 752 | if (*s->status != VIRTIO_NET_OK) { |
be4278b6 | 753 | return VIRTIO_NET_ERR; |
2df4dd31 EP |
754 | } |
755 | ||
756 | status = VIRTIO_NET_ERR; | |
7a7f87e9 | 757 | virtio_net_handle_ctrl_iov(svq->vdev, &in, 1, &out, 1); |
2df4dd31 EP |
758 | if (status != VIRTIO_NET_OK) { |
759 | error_report("Bad CVQ processing in model"); | |
bd907ae4 EP |
760 | } |
761 | ||
762 | out: | |
763 | in_len = iov_from_buf(elem->in_sg, elem->in_num, 0, &status, | |
764 | sizeof(status)); | |
765 | if (unlikely(in_len < sizeof(status))) { | |
766 | error_report("Bad device CVQ written length"); | |
767 | } | |
768 | vhost_svq_push_elem(svq, elem, MIN(in_len, sizeof(status))); | |
769 | g_free(elem); | |
be4278b6 | 770 | return dev_written < 0 ? dev_written : 0; |
bd907ae4 EP |
771 | } |
772 | ||
773 | static const VhostShadowVirtqueueOps vhost_vdpa_net_svq_ops = { | |
774 | .avail_handler = vhost_vdpa_net_handle_ctrl_avail, | |
775 | }; | |
776 | ||
152128d6 EP |
777 | /** |
778 | * Probe if CVQ is isolated | |
779 | * | |
780 | * @device_fd The vdpa device fd | |
781 | * @features Features offered by the device. | |
782 | * @cvq_index The control vq pair index | |
783 | * | |
784 | * Returns <0 in case of failure, 0 if false and 1 if true. | |
785 | */ | |
786 | static int vhost_vdpa_probe_cvq_isolation(int device_fd, uint64_t features, | |
787 | int cvq_index, Error **errp) | |
788 | { | |
789 | uint64_t backend_features; | |
790 | int64_t cvq_group; | |
791 | uint8_t status = VIRTIO_CONFIG_S_ACKNOWLEDGE | | |
792 | VIRTIO_CONFIG_S_DRIVER | | |
793 | VIRTIO_CONFIG_S_FEATURES_OK; | |
794 | int r; | |
795 | ||
796 | ERRP_GUARD(); | |
797 | ||
798 | r = ioctl(device_fd, VHOST_GET_BACKEND_FEATURES, &backend_features); | |
799 | if (unlikely(r < 0)) { | |
800 | error_setg_errno(errp, errno, "Cannot get vdpa backend_features"); | |
801 | return r; | |
802 | } | |
803 | ||
804 | if (!(backend_features & BIT_ULL(VHOST_BACKEND_F_IOTLB_ASID))) { | |
805 | return 0; | |
806 | } | |
807 | ||
808 | r = ioctl(device_fd, VHOST_SET_FEATURES, &features); | |
809 | if (unlikely(r)) { | |
810 | error_setg_errno(errp, errno, "Cannot set features"); | |
811 | } | |
812 | ||
813 | r = ioctl(device_fd, VHOST_VDPA_SET_STATUS, &status); | |
814 | if (unlikely(r)) { | |
815 | error_setg_errno(errp, -r, "Cannot set device features"); | |
816 | goto out; | |
817 | } | |
818 | ||
819 | cvq_group = vhost_vdpa_get_vring_group(device_fd, cvq_index, errp); | |
820 | if (unlikely(cvq_group < 0)) { | |
821 | if (cvq_group != -ENOTSUP) { | |
822 | r = cvq_group; | |
823 | goto out; | |
824 | } | |
825 | ||
826 | /* | |
827 | * The kernel report VHOST_BACKEND_F_IOTLB_ASID if the vdpa frontend | |
828 | * support ASID even if the parent driver does not. The CVQ cannot be | |
829 | * isolated in this case. | |
830 | */ | |
831 | error_free(*errp); | |
832 | *errp = NULL; | |
833 | r = 0; | |
834 | goto out; | |
835 | } | |
836 | ||
837 | for (int i = 0; i < cvq_index; ++i) { | |
838 | int64_t group = vhost_vdpa_get_vring_group(device_fd, i, errp); | |
839 | if (unlikely(group < 0)) { | |
840 | r = group; | |
841 | goto out; | |
842 | } | |
843 | ||
844 | if (group == (int64_t)cvq_group) { | |
845 | r = 0; | |
846 | goto out; | |
847 | } | |
848 | } | |
849 | ||
850 | r = 1; | |
851 | ||
852 | out: | |
853 | status = 0; | |
854 | ioctl(device_fd, VHOST_VDPA_SET_STATUS, &status); | |
855 | return r; | |
856 | } | |
857 | ||
654790b6 | 858 | static NetClientState *net_vhost_vdpa_init(NetClientState *peer, |
a585fad2 EP |
859 | const char *device, |
860 | const char *name, | |
861 | int vdpa_device_fd, | |
862 | int queue_pair_index, | |
863 | int nvqs, | |
864 | bool is_datapath, | |
865 | bool svq, | |
5c1ebd4c | 866 | struct vhost_vdpa_iova_range iova_range, |
152128d6 EP |
867 | uint64_t features, |
868 | Error **errp) | |
1e0a84ea CL |
869 | { |
870 | NetClientState *nc = NULL; | |
871 | VhostVDPAState *s; | |
1e0a84ea CL |
872 | int ret = 0; |
873 | assert(name); | |
152128d6 EP |
874 | int cvq_isolated; |
875 | ||
40237840 JW |
876 | if (is_datapath) { |
877 | nc = qemu_new_net_client(&net_vhost_vdpa_info, peer, device, | |
878 | name); | |
879 | } else { | |
152128d6 EP |
880 | cvq_isolated = vhost_vdpa_probe_cvq_isolation(vdpa_device_fd, features, |
881 | queue_pair_index * 2, | |
882 | errp); | |
883 | if (unlikely(cvq_isolated < 0)) { | |
884 | return NULL; | |
885 | } | |
886 | ||
f8972b56 | 887 | nc = qemu_new_net_control_client(&net_vhost_vdpa_cvq_info, peer, |
40237840 JW |
888 | device, name); |
889 | } | |
53b85d95 | 890 | qemu_set_info_str(nc, TYPE_VHOST_VDPA); |
1e0a84ea | 891 | s = DO_UPCAST(VhostVDPAState, nc, nc); |
7327813d | 892 | |
1e0a84ea | 893 | s->vhost_vdpa.device_fd = vdpa_device_fd; |
40237840 | 894 | s->vhost_vdpa.index = queue_pair_index; |
7f211a28 | 895 | s->always_svq = svq; |
69498430 | 896 | s->migration_state.notify = vdpa_net_migration_state_notifier; |
1576dbb5 | 897 | s->vhost_vdpa.shadow_vqs_enabled = svq; |
a585fad2 | 898 | s->vhost_vdpa.iova_range = iova_range; |
6188d78a | 899 | s->vhost_vdpa.shadow_data = svq; |
5c1ebd4c EP |
900 | if (queue_pair_index == 0) { |
901 | vhost_vdpa_net_valid_svq_features(features, | |
902 | &s->vhost_vdpa.migration_blocker); | |
903 | } else if (!is_datapath) { | |
babf8b87 EP |
904 | s->cvq_cmd_out_buffer = mmap(NULL, vhost_vdpa_net_cvq_cmd_page_len(), |
905 | PROT_READ | PROT_WRITE, | |
906 | MAP_SHARED | MAP_ANONYMOUS, -1, 0); | |
907 | s->status = mmap(NULL, vhost_vdpa_net_cvq_cmd_page_len(), | |
908 | PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, | |
909 | -1, 0); | |
2df4dd31 | 910 | |
bd907ae4 EP |
911 | s->vhost_vdpa.shadow_vq_ops = &vhost_vdpa_net_svq_ops; |
912 | s->vhost_vdpa.shadow_vq_ops_opaque = s; | |
152128d6 | 913 | s->cvq_isolated = cvq_isolated; |
9c363cf6 EP |
914 | |
915 | /* | |
8bc0049e EP |
916 | * TODO: We cannot migrate devices with CVQ and no x-svq enabled as |
917 | * there is no way to set the device state (MAC, MQ, etc) before | |
918 | * starting the datapath. | |
9c363cf6 EP |
919 | * |
920 | * Migration blocker ownership now belongs to s->vhost_vdpa. | |
921 | */ | |
8bc0049e EP |
922 | if (!svq) { |
923 | error_setg(&s->vhost_vdpa.migration_blocker, | |
924 | "net vdpa cannot migrate with CVQ feature"); | |
925 | } | |
bd907ae4 | 926 | } |
40237840 | 927 | ret = vhost_vdpa_add(nc, (void *)&s->vhost_vdpa, queue_pair_index, nvqs); |
74af5eec | 928 | if (ret) { |
74af5eec | 929 | qemu_del_net_client(nc); |
654790b6 | 930 | return NULL; |
74af5eec | 931 | } |
654790b6 | 932 | return nc; |
1e0a84ea CL |
933 | } |
934 | ||
8170ab3f EP |
935 | static int vhost_vdpa_get_features(int fd, uint64_t *features, Error **errp) |
936 | { | |
937 | int ret = ioctl(fd, VHOST_GET_FEATURES, features); | |
938 | if (unlikely(ret < 0)) { | |
939 | error_setg_errno(errp, errno, | |
940 | "Fail to query features from vhost-vDPA device"); | |
941 | } | |
942 | return ret; | |
943 | } | |
944 | ||
945 | static int vhost_vdpa_get_max_queue_pairs(int fd, uint64_t features, | |
946 | int *has_cvq, Error **errp) | |
40237840 JW |
947 | { |
948 | unsigned long config_size = offsetof(struct vhost_vdpa_config, buf); | |
cd523a41 | 949 | g_autofree struct vhost_vdpa_config *config = NULL; |
40237840 | 950 | __virtio16 *max_queue_pairs; |
40237840 JW |
951 | int ret; |
952 | ||
40237840 JW |
953 | if (features & (1 << VIRTIO_NET_F_CTRL_VQ)) { |
954 | *has_cvq = 1; | |
955 | } else { | |
956 | *has_cvq = 0; | |
957 | } | |
958 | ||
959 | if (features & (1 << VIRTIO_NET_F_MQ)) { | |
960 | config = g_malloc0(config_size + sizeof(*max_queue_pairs)); | |
961 | config->off = offsetof(struct virtio_net_config, max_virtqueue_pairs); | |
962 | config->len = sizeof(*max_queue_pairs); | |
963 | ||
964 | ret = ioctl(fd, VHOST_VDPA_GET_CONFIG, config); | |
965 | if (ret) { | |
966 | error_setg(errp, "Fail to get config from vhost-vDPA device"); | |
967 | return -ret; | |
968 | } | |
969 | ||
970 | max_queue_pairs = (__virtio16 *)&config->buf; | |
971 | ||
972 | return lduw_le_p(max_queue_pairs); | |
973 | } | |
974 | ||
975 | return 1; | |
976 | } | |
977 | ||
1e0a84ea CL |
978 | int net_init_vhost_vdpa(const Netdev *netdev, const char *name, |
979 | NetClientState *peer, Error **errp) | |
980 | { | |
981 | const NetdevVhostVDPAOptions *opts; | |
8170ab3f | 982 | uint64_t features; |
654790b6 | 983 | int vdpa_device_fd; |
eb3cb751 | 984 | g_autofree NetClientState **ncs = NULL; |
a585fad2 | 985 | struct vhost_vdpa_iova_range iova_range; |
eb3cb751 | 986 | NetClientState *nc; |
aed5da45 | 987 | int queue_pairs, r, i = 0, has_cvq = 0; |
1e0a84ea CL |
988 | |
989 | assert(netdev->type == NET_CLIENT_DRIVER_VHOST_VDPA); | |
990 | opts = &netdev->u.vhost_vdpa; | |
7480874a | 991 | if (!opts->vhostdev && !opts->vhostfd) { |
8801ccd0 SWL |
992 | error_setg(errp, |
993 | "vhost-vdpa: neither vhostdev= nor vhostfd= was specified"); | |
c8295404 EP |
994 | return -1; |
995 | } | |
7327813d | 996 | |
7480874a | 997 | if (opts->vhostdev && opts->vhostfd) { |
8801ccd0 SWL |
998 | error_setg(errp, |
999 | "vhost-vdpa: vhostdev= and vhostfd= are mutually exclusive"); | |
1000 | return -1; | |
1001 | } | |
1002 | ||
7480874a | 1003 | if (opts->vhostdev) { |
8801ccd0 SWL |
1004 | vdpa_device_fd = qemu_open(opts->vhostdev, O_RDWR, errp); |
1005 | if (vdpa_device_fd == -1) { | |
1006 | return -errno; | |
1007 | } | |
5107fd3e PM |
1008 | } else { |
1009 | /* has_vhostfd */ | |
8801ccd0 SWL |
1010 | vdpa_device_fd = monitor_fd_param(monitor_cur(), opts->vhostfd, errp); |
1011 | if (vdpa_device_fd == -1) { | |
1012 | error_prepend(errp, "vhost-vdpa: unable to parse vhostfd: "); | |
1013 | return -1; | |
1014 | } | |
7327813d JW |
1015 | } |
1016 | ||
8170ab3f EP |
1017 | r = vhost_vdpa_get_features(vdpa_device_fd, &features, errp); |
1018 | if (unlikely(r < 0)) { | |
aed5da45 | 1019 | goto err; |
8170ab3f EP |
1020 | } |
1021 | ||
1022 | queue_pairs = vhost_vdpa_get_max_queue_pairs(vdpa_device_fd, features, | |
40237840 JW |
1023 | &has_cvq, errp); |
1024 | if (queue_pairs < 0) { | |
7327813d | 1025 | qemu_close(vdpa_device_fd); |
40237840 JW |
1026 | return queue_pairs; |
1027 | } | |
1028 | ||
bf7a2ad8 LM |
1029 | r = vhost_vdpa_get_iova_range(vdpa_device_fd, &iova_range); |
1030 | if (unlikely(r < 0)) { | |
1031 | error_setg(errp, "vhost-vdpa: get iova range failed: %s", | |
1032 | strerror(-r)); | |
1033 | goto err; | |
1034 | } | |
1035 | ||
00ef422e EP |
1036 | if (opts->x_svq && !vhost_vdpa_net_valid_svq_features(features, errp)) { |
1037 | goto err; | |
1576dbb5 EP |
1038 | } |
1039 | ||
40237840 JW |
1040 | ncs = g_malloc0(sizeof(*ncs) * queue_pairs); |
1041 | ||
1042 | for (i = 0; i < queue_pairs; i++) { | |
1043 | ncs[i] = net_vhost_vdpa_init(peer, TYPE_VHOST_VDPA, name, | |
1576dbb5 | 1044 | vdpa_device_fd, i, 2, true, opts->x_svq, |
152128d6 | 1045 | iova_range, features, errp); |
40237840 JW |
1046 | if (!ncs[i]) |
1047 | goto err; | |
7327813d JW |
1048 | } |
1049 | ||
40237840 JW |
1050 | if (has_cvq) { |
1051 | nc = net_vhost_vdpa_init(peer, TYPE_VHOST_VDPA, name, | |
1576dbb5 | 1052 | vdpa_device_fd, i, 1, false, |
152128d6 | 1053 | opts->x_svq, iova_range, features, errp); |
40237840 JW |
1054 | if (!nc) |
1055 | goto err; | |
1056 | } | |
1057 | ||
654790b6 | 1058 | return 0; |
40237840 JW |
1059 | |
1060 | err: | |
1061 | if (i) { | |
9bd05507 SWL |
1062 | for (i--; i >= 0; i--) { |
1063 | qemu_del_net_client(ncs[i]); | |
1064 | } | |
40237840 | 1065 | } |
1576dbb5 | 1066 | |
40237840 | 1067 | qemu_close(vdpa_device_fd); |
40237840 JW |
1068 | |
1069 | return -1; | |
1e0a84ea | 1070 | } |