1
2
3
4
5
6
7
8
9
10
11#include "qemu/osdep.h"
12#include "hw/virtio/vhost.h"
13#include "hw/virtio/vhost-backend.h"
14#include "qemu/error-report.h"
15#include "qemu/main-loop.h"
16#include "standard-headers/linux/vhost_types.h"
17
18#include "hw/virtio/vhost-vdpa.h"
19#ifdef CONFIG_VHOST_KERNEL
20#include <linux/vhost.h>
21#include <sys/ioctl.h>
22
23static int vhost_kernel_call(struct vhost_dev *dev, unsigned long int request,
24 void *arg)
25{
26 int fd = (uintptr_t) dev->opaque;
27 int ret;
28
29 assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_KERNEL);
30
31 ret = ioctl(fd, request, arg);
32 return ret < 0 ? -errno : ret;
33}
34
35static int vhost_kernel_init(struct vhost_dev *dev, void *opaque, Error **errp)
36{
37 assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_KERNEL);
38
39 dev->opaque = opaque;
40
41 return 0;
42}
43
44static int vhost_kernel_cleanup(struct vhost_dev *dev)
45{
46 int fd = (uintptr_t) dev->opaque;
47
48 assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_KERNEL);
49
50 return close(fd) < 0 ? -errno : 0;
51}
52
53static int vhost_kernel_memslots_limit(struct vhost_dev *dev)
54{
55 int limit = 64;
56 char *s;
57
58 if (g_file_get_contents("/sys/module/vhost/parameters/max_mem_regions",
59 &s, NULL, NULL)) {
60 uint64_t val = g_ascii_strtoull(s, NULL, 10);
61 if (val < INT_MAX && val > 0) {
62 g_free(s);
63 return val;
64 }
65 error_report("ignoring invalid max_mem_regions value in vhost module:"
66 " %s", s);
67 }
68 g_free(s);
69 return limit;
70}
71
72static int vhost_kernel_net_set_backend(struct vhost_dev *dev,
73 struct vhost_vring_file *file)
74{
75 return vhost_kernel_call(dev, VHOST_NET_SET_BACKEND, file);
76}
77
78static int vhost_kernel_scsi_set_endpoint(struct vhost_dev *dev,
79 struct vhost_scsi_target *target)
80{
81 return vhost_kernel_call(dev, VHOST_SCSI_SET_ENDPOINT, target);
82}
83
84static int vhost_kernel_scsi_clear_endpoint(struct vhost_dev *dev,
85 struct vhost_scsi_target *target)
86{
87 return vhost_kernel_call(dev, VHOST_SCSI_CLEAR_ENDPOINT, target);
88}
89
90static int vhost_kernel_scsi_get_abi_version(struct vhost_dev *dev, int *version)
91{
92 return vhost_kernel_call(dev, VHOST_SCSI_GET_ABI_VERSION, version);
93}
94
95static int vhost_kernel_set_log_base(struct vhost_dev *dev, uint64_t base,
96 struct vhost_log *log)
97{
98 return vhost_kernel_call(dev, VHOST_SET_LOG_BASE, &base);
99}
100
101static int vhost_kernel_set_mem_table(struct vhost_dev *dev,
102 struct vhost_memory *mem)
103{
104 return vhost_kernel_call(dev, VHOST_SET_MEM_TABLE, mem);
105}
106
107static int vhost_kernel_set_vring_addr(struct vhost_dev *dev,
108 struct vhost_vring_addr *addr)
109{
110 return vhost_kernel_call(dev, VHOST_SET_VRING_ADDR, addr);
111}
112
113static int vhost_kernel_set_vring_endian(struct vhost_dev *dev,
114 struct vhost_vring_state *ring)
115{
116 return vhost_kernel_call(dev, VHOST_SET_VRING_ENDIAN, ring);
117}
118
119static int vhost_kernel_set_vring_num(struct vhost_dev *dev,
120 struct vhost_vring_state *ring)
121{
122 return vhost_kernel_call(dev, VHOST_SET_VRING_NUM, ring);
123}
124
125static int vhost_kernel_set_vring_base(struct vhost_dev *dev,
126 struct vhost_vring_state *ring)
127{
128 return vhost_kernel_call(dev, VHOST_SET_VRING_BASE, ring);
129}
130
131static int vhost_kernel_get_vring_base(struct vhost_dev *dev,
132 struct vhost_vring_state *ring)
133{
134 return vhost_kernel_call(dev, VHOST_GET_VRING_BASE, ring);
135}
136
137static int vhost_kernel_set_vring_kick(struct vhost_dev *dev,
138 struct vhost_vring_file *file)
139{
140 return vhost_kernel_call(dev, VHOST_SET_VRING_KICK, file);
141}
142
143static int vhost_kernel_set_vring_call(struct vhost_dev *dev,
144 struct vhost_vring_file *file)
145{
146 return vhost_kernel_call(dev, VHOST_SET_VRING_CALL, file);
147}
148
149static int vhost_kernel_set_vring_err(struct vhost_dev *dev,
150 struct vhost_vring_file *file)
151{
152 return vhost_kernel_call(dev, VHOST_SET_VRING_ERR, file);
153}
154
155static int vhost_kernel_set_vring_busyloop_timeout(struct vhost_dev *dev,
156 struct vhost_vring_state *s)
157{
158 return vhost_kernel_call(dev, VHOST_SET_VRING_BUSYLOOP_TIMEOUT, s);
159}
160
161static int vhost_kernel_set_features(struct vhost_dev *dev,
162 uint64_t features)
163{
164 return vhost_kernel_call(dev, VHOST_SET_FEATURES, &features);
165}
166
167static int vhost_kernel_set_backend_cap(struct vhost_dev *dev)
168{
169 uint64_t features;
170 uint64_t f = 0x1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2;
171 int r;
172
173 if (vhost_kernel_call(dev, VHOST_GET_BACKEND_FEATURES, &features)) {
174 return 0;
175 }
176
177 features &= f;
178 r = vhost_kernel_call(dev, VHOST_SET_BACKEND_FEATURES,
179 &features);
180 if (r) {
181 return 0;
182 }
183
184 dev->backend_cap = features;
185
186 return 0;
187}
188
189static int vhost_kernel_get_features(struct vhost_dev *dev,
190 uint64_t *features)
191{
192 return vhost_kernel_call(dev, VHOST_GET_FEATURES, features);
193}
194
195static int vhost_kernel_set_owner(struct vhost_dev *dev)
196{
197 return vhost_kernel_call(dev, VHOST_SET_OWNER, NULL);
198}
199
200static int vhost_kernel_reset_device(struct vhost_dev *dev)
201{
202 return vhost_kernel_call(dev, VHOST_RESET_OWNER, NULL);
203}
204
205static int vhost_kernel_get_vq_index(struct vhost_dev *dev, int idx)
206{
207 assert(idx >= dev->vq_index && idx < dev->vq_index + dev->nvqs);
208
209 return idx - dev->vq_index;
210}
211
212static int vhost_kernel_vsock_set_guest_cid(struct vhost_dev *dev,
213 uint64_t guest_cid)
214{
215 return vhost_kernel_call(dev, VHOST_VSOCK_SET_GUEST_CID, &guest_cid);
216}
217
218static int vhost_kernel_vsock_set_running(struct vhost_dev *dev, int start)
219{
220 return vhost_kernel_call(dev, VHOST_VSOCK_SET_RUNNING, &start);
221}
222
223static void vhost_kernel_iotlb_read(void *opaque)
224{
225 struct vhost_dev *dev = opaque;
226 ssize_t len;
227
228 if (dev->backend_cap &
229 (0x1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2)) {
230 struct vhost_msg_v2 msg;
231
232 while ((len = read((uintptr_t)dev->opaque, &msg, sizeof msg)) > 0) {
233 if (len < sizeof msg) {
234 error_report("Wrong vhost message len: %d", (int)len);
235 break;
236 }
237 if (msg.type != VHOST_IOTLB_MSG_V2) {
238 error_report("Unknown vhost iotlb message type");
239 break;
240 }
241
242 vhost_backend_handle_iotlb_msg(dev, &msg.iotlb);
243 }
244 } else {
245 struct vhost_msg msg;
246
247 while ((len = read((uintptr_t)dev->opaque, &msg, sizeof msg)) > 0) {
248 if (len < sizeof msg) {
249 error_report("Wrong vhost message len: %d", (int)len);
250 break;
251 }
252 if (msg.type != VHOST_IOTLB_MSG) {
253 error_report("Unknown vhost iotlb message type");
254 break;
255 }
256
257 vhost_backend_handle_iotlb_msg(dev, &msg.iotlb);
258 }
259 }
260}
261
262static int vhost_kernel_send_device_iotlb_msg(struct vhost_dev *dev,
263 struct vhost_iotlb_msg *imsg)
264{
265 if (dev->backend_cap & (1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2)) {
266 struct vhost_msg_v2 msg = {};
267
268 msg.type = VHOST_IOTLB_MSG_V2;
269 msg.iotlb = *imsg;
270
271 if (write((uintptr_t)dev->opaque, &msg, sizeof msg) != sizeof msg) {
272 error_report("Fail to update device iotlb");
273 return -EFAULT;
274 }
275 } else {
276 struct vhost_msg msg = {};
277
278 msg.type = VHOST_IOTLB_MSG;
279 msg.iotlb = *imsg;
280
281 if (write((uintptr_t)dev->opaque, &msg, sizeof msg) != sizeof msg) {
282 error_report("Fail to update device iotlb");
283 return -EFAULT;
284 }
285 }
286
287 return 0;
288}
289
290static void vhost_kernel_set_iotlb_callback(struct vhost_dev *dev,
291 int enabled)
292{
293 if (enabled)
294 qemu_set_fd_handler((uintptr_t)dev->opaque,
295 vhost_kernel_iotlb_read, NULL, dev);
296 else
297 qemu_set_fd_handler((uintptr_t)dev->opaque, NULL, NULL, NULL);
298}
299
300const VhostOps kernel_ops = {
301 .backend_type = VHOST_BACKEND_TYPE_KERNEL,
302 .vhost_backend_init = vhost_kernel_init,
303 .vhost_backend_cleanup = vhost_kernel_cleanup,
304 .vhost_backend_memslots_limit = vhost_kernel_memslots_limit,
305 .vhost_net_set_backend = vhost_kernel_net_set_backend,
306 .vhost_scsi_set_endpoint = vhost_kernel_scsi_set_endpoint,
307 .vhost_scsi_clear_endpoint = vhost_kernel_scsi_clear_endpoint,
308 .vhost_scsi_get_abi_version = vhost_kernel_scsi_get_abi_version,
309 .vhost_set_log_base = vhost_kernel_set_log_base,
310 .vhost_set_mem_table = vhost_kernel_set_mem_table,
311 .vhost_set_vring_addr = vhost_kernel_set_vring_addr,
312 .vhost_set_vring_endian = vhost_kernel_set_vring_endian,
313 .vhost_set_vring_num = vhost_kernel_set_vring_num,
314 .vhost_set_vring_base = vhost_kernel_set_vring_base,
315 .vhost_get_vring_base = vhost_kernel_get_vring_base,
316 .vhost_set_vring_kick = vhost_kernel_set_vring_kick,
317 .vhost_set_vring_call = vhost_kernel_set_vring_call,
318 .vhost_set_vring_err = vhost_kernel_set_vring_err,
319 .vhost_set_vring_busyloop_timeout =
320 vhost_kernel_set_vring_busyloop_timeout,
321 .vhost_set_features = vhost_kernel_set_features,
322 .vhost_get_features = vhost_kernel_get_features,
323 .vhost_set_backend_cap = vhost_kernel_set_backend_cap,
324 .vhost_set_owner = vhost_kernel_set_owner,
325 .vhost_reset_device = vhost_kernel_reset_device,
326 .vhost_get_vq_index = vhost_kernel_get_vq_index,
327 .vhost_vsock_set_guest_cid = vhost_kernel_vsock_set_guest_cid,
328 .vhost_vsock_set_running = vhost_kernel_vsock_set_running,
329 .vhost_set_iotlb_callback = vhost_kernel_set_iotlb_callback,
330 .vhost_send_device_iotlb_msg = vhost_kernel_send_device_iotlb_msg,
331};
332#endif
333
334int vhost_backend_update_device_iotlb(struct vhost_dev *dev,
335 uint64_t iova, uint64_t uaddr,
336 uint64_t len,
337 IOMMUAccessFlags perm)
338{
339 struct vhost_iotlb_msg imsg;
340
341 imsg.iova = iova;
342 imsg.uaddr = uaddr;
343 imsg.size = len;
344 imsg.type = VHOST_IOTLB_UPDATE;
345
346 switch (perm) {
347 case IOMMU_RO:
348 imsg.perm = VHOST_ACCESS_RO;
349 break;
350 case IOMMU_WO:
351 imsg.perm = VHOST_ACCESS_WO;
352 break;
353 case IOMMU_RW:
354 imsg.perm = VHOST_ACCESS_RW;
355 break;
356 default:
357 return -EINVAL;
358 }
359
360 if (dev->vhost_ops && dev->vhost_ops->vhost_send_device_iotlb_msg)
361 return dev->vhost_ops->vhost_send_device_iotlb_msg(dev, &imsg);
362
363 return -ENODEV;
364}
365
366int vhost_backend_invalidate_device_iotlb(struct vhost_dev *dev,
367 uint64_t iova, uint64_t len)
368{
369 struct vhost_iotlb_msg imsg;
370
371 imsg.iova = iova;
372 imsg.size = len;
373 imsg.type = VHOST_IOTLB_INVALIDATE;
374
375 if (dev->vhost_ops && dev->vhost_ops->vhost_send_device_iotlb_msg)
376 return dev->vhost_ops->vhost_send_device_iotlb_msg(dev, &imsg);
377
378 return -ENODEV;
379}
380
381int vhost_backend_handle_iotlb_msg(struct vhost_dev *dev,
382 struct vhost_iotlb_msg *imsg)
383{
384 int ret = 0;
385
386 if (unlikely(!dev->vdev)) {
387 error_report("Unexpected IOTLB message when virtio device is stopped");
388 return -EINVAL;
389 }
390
391 switch (imsg->type) {
392 case VHOST_IOTLB_MISS:
393 ret = vhost_device_iotlb_miss(dev, imsg->iova,
394 imsg->perm != VHOST_ACCESS_RO);
395 break;
396 case VHOST_IOTLB_ACCESS_FAIL:
397
398 error_report("Access failure IOTLB message type not supported");
399 ret = -ENOTSUP;
400 break;
401 case VHOST_IOTLB_UPDATE:
402 case VHOST_IOTLB_INVALIDATE:
403 default:
404 error_report("Unexpected IOTLB message type");
405 ret = -EINVAL;
406 break;
407 }
408
409 return ret;
410}
411