root / hw / vhost.c @ e5896b12
History | View | Annotate | Download (22.6 kB)
1 |
/*
|
---|---|
2 |
* vhost support
|
3 |
*
|
4 |
* Copyright Red Hat, Inc. 2010
|
5 |
*
|
6 |
* Authors:
|
7 |
* Michael S. Tsirkin <mst@redhat.com>
|
8 |
*
|
9 |
* This work is licensed under the terms of the GNU GPL, version 2. See
|
10 |
* the COPYING file in the top-level directory.
|
11 |
*/
|
12 |
|
13 |
#include <sys/ioctl.h> |
14 |
#include "vhost.h" |
15 |
#include "hw/hw.h" |
16 |
#include "range.h" |
17 |
#include <linux/vhost.h> |
18 |
|
19 |
static void vhost_dev_sync_region(struct vhost_dev *dev, |
20 |
uint64_t mfirst, uint64_t mlast, |
21 |
uint64_t rfirst, uint64_t rlast) |
22 |
{ |
23 |
uint64_t start = MAX(mfirst, rfirst); |
24 |
uint64_t end = MIN(mlast, rlast); |
25 |
vhost_log_chunk_t *from = dev->log + start / VHOST_LOG_CHUNK; |
26 |
vhost_log_chunk_t *to = dev->log + end / VHOST_LOG_CHUNK + 1;
|
27 |
uint64_t addr = (start / VHOST_LOG_CHUNK) * VHOST_LOG_CHUNK; |
28 |
|
29 |
assert(end / VHOST_LOG_CHUNK < dev->log_size); |
30 |
assert(start / VHOST_LOG_CHUNK < dev->log_size); |
31 |
if (end < start) {
|
32 |
return;
|
33 |
} |
34 |
for (;from < to; ++from) {
|
35 |
vhost_log_chunk_t log; |
36 |
int bit;
|
37 |
/* We first check with non-atomic: much cheaper,
|
38 |
* and we expect non-dirty to be the common case. */
|
39 |
if (!*from) {
|
40 |
addr += VHOST_LOG_CHUNK; |
41 |
continue;
|
42 |
} |
43 |
/* Data must be read atomically. We don't really
|
44 |
* need the barrier semantics of __sync
|
45 |
* builtins, but it's easier to use them than
|
46 |
* roll our own. */
|
47 |
log = __sync_fetch_and_and(from, 0);
|
48 |
while ((bit = sizeof(log) > sizeof(int) ? |
49 |
ffsll(log) : ffs(log))) { |
50 |
bit -= 1;
|
51 |
cpu_physical_memory_set_dirty(addr + bit * VHOST_LOG_PAGE); |
52 |
log &= ~(0x1ull << bit);
|
53 |
} |
54 |
addr += VHOST_LOG_CHUNK; |
55 |
} |
56 |
} |
57 |
|
58 |
static int vhost_client_sync_dirty_bitmap(CPUPhysMemoryClient *client, |
59 |
target_phys_addr_t start_addr, |
60 |
target_phys_addr_t end_addr) |
61 |
{ |
62 |
struct vhost_dev *dev = container_of(client, struct vhost_dev, client); |
63 |
int i;
|
64 |
if (!dev->log_enabled || !dev->started) {
|
65 |
return 0; |
66 |
} |
67 |
for (i = 0; i < dev->mem->nregions; ++i) { |
68 |
struct vhost_memory_region *reg = dev->mem->regions + i;
|
69 |
vhost_dev_sync_region(dev, start_addr, end_addr, |
70 |
reg->guest_phys_addr, |
71 |
range_get_last(reg->guest_phys_addr, |
72 |
reg->memory_size)); |
73 |
} |
74 |
for (i = 0; i < dev->nvqs; ++i) { |
75 |
struct vhost_virtqueue *vq = dev->vqs + i;
|
76 |
vhost_dev_sync_region(dev, start_addr, end_addr, vq->used_phys, |
77 |
range_get_last(vq->used_phys, vq->used_size)); |
78 |
} |
79 |
return 0; |
80 |
} |
81 |
|
82 |
/* Assign/unassign. Keep an unsorted array of non-overlapping
|
83 |
* memory regions in dev->mem. */
|
84 |
static void vhost_dev_unassign_memory(struct vhost_dev *dev, |
85 |
uint64_t start_addr, |
86 |
uint64_t size) |
87 |
{ |
88 |
int from, to, n = dev->mem->nregions;
|
89 |
/* Track overlapping/split regions for sanity checking. */
|
90 |
int overlap_start = 0, overlap_end = 0, overlap_middle = 0, split = 0; |
91 |
|
92 |
for (from = 0, to = 0; from < n; ++from, ++to) { |
93 |
struct vhost_memory_region *reg = dev->mem->regions + to;
|
94 |
uint64_t reglast; |
95 |
uint64_t memlast; |
96 |
uint64_t change; |
97 |
|
98 |
/* clone old region */
|
99 |
if (to != from) {
|
100 |
memcpy(reg, dev->mem->regions + from, sizeof *reg);
|
101 |
} |
102 |
|
103 |
/* No overlap is simple */
|
104 |
if (!ranges_overlap(reg->guest_phys_addr, reg->memory_size,
|
105 |
start_addr, size)) { |
106 |
continue;
|
107 |
} |
108 |
|
109 |
/* Split only happens if supplied region
|
110 |
* is in the middle of an existing one. Thus it can not
|
111 |
* overlap with any other existing region. */
|
112 |
assert(!split); |
113 |
|
114 |
reglast = range_get_last(reg->guest_phys_addr, reg->memory_size); |
115 |
memlast = range_get_last(start_addr, size); |
116 |
|
117 |
/* Remove whole region */
|
118 |
if (start_addr <= reg->guest_phys_addr && memlast >= reglast) {
|
119 |
--dev->mem->nregions; |
120 |
--to; |
121 |
assert(to >= 0);
|
122 |
++overlap_middle; |
123 |
continue;
|
124 |
} |
125 |
|
126 |
/* Shrink region */
|
127 |
if (memlast >= reglast) {
|
128 |
reg->memory_size = start_addr - reg->guest_phys_addr; |
129 |
assert(reg->memory_size); |
130 |
assert(!overlap_end); |
131 |
++overlap_end; |
132 |
continue;
|
133 |
} |
134 |
|
135 |
/* Shift region */
|
136 |
if (start_addr <= reg->guest_phys_addr) {
|
137 |
change = memlast + 1 - reg->guest_phys_addr;
|
138 |
reg->memory_size -= change; |
139 |
reg->guest_phys_addr += change; |
140 |
reg->userspace_addr += change; |
141 |
assert(reg->memory_size); |
142 |
assert(!overlap_start); |
143 |
++overlap_start; |
144 |
continue;
|
145 |
} |
146 |
|
147 |
/* This only happens if supplied region
|
148 |
* is in the middle of an existing one. Thus it can not
|
149 |
* overlap with any other existing region. */
|
150 |
assert(!overlap_start); |
151 |
assert(!overlap_end); |
152 |
assert(!overlap_middle); |
153 |
/* Split region: shrink first part, shift second part. */
|
154 |
memcpy(dev->mem->regions + n, reg, sizeof *reg);
|
155 |
reg->memory_size = start_addr - reg->guest_phys_addr; |
156 |
assert(reg->memory_size); |
157 |
change = memlast + 1 - reg->guest_phys_addr;
|
158 |
reg = dev->mem->regions + n; |
159 |
reg->memory_size -= change; |
160 |
assert(reg->memory_size); |
161 |
reg->guest_phys_addr += change; |
162 |
reg->userspace_addr += change; |
163 |
/* Never add more than 1 region */
|
164 |
assert(dev->mem->nregions == n); |
165 |
++dev->mem->nregions; |
166 |
++split; |
167 |
} |
168 |
} |
169 |
|
170 |
/* Called after unassign, so no regions overlap the given range. */
|
171 |
static void vhost_dev_assign_memory(struct vhost_dev *dev, |
172 |
uint64_t start_addr, |
173 |
uint64_t size, |
174 |
uint64_t uaddr) |
175 |
{ |
176 |
int from, to;
|
177 |
struct vhost_memory_region *merged = NULL; |
178 |
for (from = 0, to = 0; from < dev->mem->nregions; ++from, ++to) { |
179 |
struct vhost_memory_region *reg = dev->mem->regions + to;
|
180 |
uint64_t prlast, urlast; |
181 |
uint64_t pmlast, umlast; |
182 |
uint64_t s, e, u; |
183 |
|
184 |
/* clone old region */
|
185 |
if (to != from) {
|
186 |
memcpy(reg, dev->mem->regions + from, sizeof *reg);
|
187 |
} |
188 |
prlast = range_get_last(reg->guest_phys_addr, reg->memory_size); |
189 |
pmlast = range_get_last(start_addr, size); |
190 |
urlast = range_get_last(reg->userspace_addr, reg->memory_size); |
191 |
umlast = range_get_last(uaddr, size); |
192 |
|
193 |
/* check for overlapping regions: should never happen. */
|
194 |
assert(prlast < start_addr || pmlast < reg->guest_phys_addr); |
195 |
/* Not an adjacent or overlapping region - do not merge. */
|
196 |
if ((prlast + 1 != start_addr || urlast + 1 != uaddr) && |
197 |
(pmlast + 1 != reg->guest_phys_addr ||
|
198 |
umlast + 1 != reg->userspace_addr)) {
|
199 |
continue;
|
200 |
} |
201 |
|
202 |
if (merged) {
|
203 |
--to; |
204 |
assert(to >= 0);
|
205 |
} else {
|
206 |
merged = reg; |
207 |
} |
208 |
u = MIN(uaddr, reg->userspace_addr); |
209 |
s = MIN(start_addr, reg->guest_phys_addr); |
210 |
e = MAX(pmlast, prlast); |
211 |
uaddr = merged->userspace_addr = u; |
212 |
start_addr = merged->guest_phys_addr = s; |
213 |
size = merged->memory_size = e - s + 1;
|
214 |
assert(merged->memory_size); |
215 |
} |
216 |
|
217 |
if (!merged) {
|
218 |
struct vhost_memory_region *reg = dev->mem->regions + to;
|
219 |
memset(reg, 0, sizeof *reg); |
220 |
reg->memory_size = size; |
221 |
assert(reg->memory_size); |
222 |
reg->guest_phys_addr = start_addr; |
223 |
reg->userspace_addr = uaddr; |
224 |
++to; |
225 |
} |
226 |
assert(to <= dev->mem->nregions + 1);
|
227 |
dev->mem->nregions = to; |
228 |
} |
229 |
|
230 |
static uint64_t vhost_get_log_size(struct vhost_dev *dev) |
231 |
{ |
232 |
uint64_t log_size = 0;
|
233 |
int i;
|
234 |
for (i = 0; i < dev->mem->nregions; ++i) { |
235 |
struct vhost_memory_region *reg = dev->mem->regions + i;
|
236 |
uint64_t last = range_get_last(reg->guest_phys_addr, |
237 |
reg->memory_size); |
238 |
log_size = MAX(log_size, last / VHOST_LOG_CHUNK + 1);
|
239 |
} |
240 |
for (i = 0; i < dev->nvqs; ++i) { |
241 |
struct vhost_virtqueue *vq = dev->vqs + i;
|
242 |
uint64_t last = vq->used_phys + vq->used_size - 1;
|
243 |
log_size = MAX(log_size, last / VHOST_LOG_CHUNK + 1);
|
244 |
} |
245 |
return log_size;
|
246 |
} |
247 |
|
248 |
static inline void vhost_dev_log_resize(struct vhost_dev* dev, uint64_t size) |
249 |
{ |
250 |
vhost_log_chunk_t *log; |
251 |
uint64_t log_base; |
252 |
int r;
|
253 |
if (size) {
|
254 |
log = qemu_mallocz(size * sizeof *log);
|
255 |
} else {
|
256 |
log = NULL;
|
257 |
} |
258 |
log_base = (uint64_t)(unsigned long)log; |
259 |
r = ioctl(dev->control, VHOST_SET_LOG_BASE, &log_base); |
260 |
assert(r >= 0);
|
261 |
vhost_client_sync_dirty_bitmap(&dev->client, 0,
|
262 |
(target_phys_addr_t)~0x0ull);
|
263 |
if (dev->log) {
|
264 |
qemu_free(dev->log); |
265 |
} |
266 |
dev->log = log; |
267 |
dev->log_size = size; |
268 |
} |
269 |
|
270 |
static int vhost_verify_ring_mappings(struct vhost_dev *dev, |
271 |
uint64_t start_addr, |
272 |
uint64_t size) |
273 |
{ |
274 |
int i;
|
275 |
for (i = 0; i < dev->nvqs; ++i) { |
276 |
struct vhost_virtqueue *vq = dev->vqs + i;
|
277 |
target_phys_addr_t l; |
278 |
void *p;
|
279 |
|
280 |
if (!ranges_overlap(start_addr, size, vq->ring_phys, vq->ring_size)) {
|
281 |
continue;
|
282 |
} |
283 |
l = vq->ring_size; |
284 |
p = cpu_physical_memory_map(vq->ring_phys, &l, 1);
|
285 |
if (!p || l != vq->ring_size) {
|
286 |
fprintf(stderr, "Unable to map ring buffer for ring %d\n", i);
|
287 |
return -ENOMEM;
|
288 |
} |
289 |
if (p != vq->ring) {
|
290 |
fprintf(stderr, "Ring buffer relocated for ring %d\n", i);
|
291 |
return -EBUSY;
|
292 |
} |
293 |
cpu_physical_memory_unmap(p, l, 0, 0); |
294 |
} |
295 |
return 0; |
296 |
} |
297 |
|
298 |
static void vhost_client_set_memory(CPUPhysMemoryClient *client, |
299 |
target_phys_addr_t start_addr, |
300 |
ram_addr_t size, |
301 |
ram_addr_t phys_offset) |
302 |
{ |
303 |
struct vhost_dev *dev = container_of(client, struct vhost_dev, client); |
304 |
ram_addr_t flags = phys_offset & ~TARGET_PAGE_MASK; |
305 |
int s = offsetof(struct vhost_memory, regions) + |
306 |
(dev->mem->nregions + 1) * sizeof dev->mem->regions[0]; |
307 |
uint64_t log_size; |
308 |
int r;
|
309 |
dev->mem = qemu_realloc(dev->mem, s); |
310 |
|
311 |
assert(size); |
312 |
|
313 |
vhost_dev_unassign_memory(dev, start_addr, size); |
314 |
if (flags == IO_MEM_RAM) {
|
315 |
/* Add given mapping, merging adjacent regions if any */
|
316 |
vhost_dev_assign_memory(dev, start_addr, size, |
317 |
(uintptr_t)qemu_get_ram_ptr(phys_offset)); |
318 |
} else {
|
319 |
/* Remove old mapping for this memory, if any. */
|
320 |
vhost_dev_unassign_memory(dev, start_addr, size); |
321 |
} |
322 |
|
323 |
if (!dev->started) {
|
324 |
return;
|
325 |
} |
326 |
|
327 |
if (dev->started) {
|
328 |
r = vhost_verify_ring_mappings(dev, start_addr, size); |
329 |
assert(r >= 0);
|
330 |
} |
331 |
|
332 |
if (!dev->log_enabled) {
|
333 |
r = ioctl(dev->control, VHOST_SET_MEM_TABLE, dev->mem); |
334 |
assert(r >= 0);
|
335 |
return;
|
336 |
} |
337 |
log_size = vhost_get_log_size(dev); |
338 |
/* We allocate an extra 4K bytes to log,
|
339 |
* to reduce the * number of reallocations. */
|
340 |
#define VHOST_LOG_BUFFER (0x1000 / sizeof *dev->log) |
341 |
/* To log more, must increase log size before table update. */
|
342 |
if (dev->log_size < log_size) {
|
343 |
vhost_dev_log_resize(dev, log_size + VHOST_LOG_BUFFER); |
344 |
} |
345 |
r = ioctl(dev->control, VHOST_SET_MEM_TABLE, dev->mem); |
346 |
assert(r >= 0);
|
347 |
/* To log less, can only decrease log size after table update. */
|
348 |
if (dev->log_size > log_size + VHOST_LOG_BUFFER) {
|
349 |
vhost_dev_log_resize(dev, log_size); |
350 |
} |
351 |
} |
352 |
|
353 |
static int vhost_virtqueue_set_addr(struct vhost_dev *dev, |
354 |
struct vhost_virtqueue *vq,
|
355 |
unsigned idx, bool enable_log) |
356 |
{ |
357 |
struct vhost_vring_addr addr = {
|
358 |
.index = idx, |
359 |
.desc_user_addr = (uint64_t)(unsigned long)vq->desc, |
360 |
.avail_user_addr = (uint64_t)(unsigned long)vq->avail, |
361 |
.used_user_addr = (uint64_t)(unsigned long)vq->used, |
362 |
.log_guest_addr = vq->used_phys, |
363 |
.flags = enable_log ? (1 << VHOST_VRING_F_LOG) : 0, |
364 |
}; |
365 |
int r = ioctl(dev->control, VHOST_SET_VRING_ADDR, &addr);
|
366 |
if (r < 0) { |
367 |
return -errno;
|
368 |
} |
369 |
return 0; |
370 |
} |
371 |
|
372 |
static int vhost_dev_set_features(struct vhost_dev *dev, bool enable_log) |
373 |
{ |
374 |
uint64_t features = dev->acked_features; |
375 |
int r;
|
376 |
if (enable_log) {
|
377 |
features |= 0x1 << VHOST_F_LOG_ALL;
|
378 |
} |
379 |
r = ioctl(dev->control, VHOST_SET_FEATURES, &features); |
380 |
return r < 0 ? -errno : 0; |
381 |
} |
382 |
|
383 |
static int vhost_dev_set_log(struct vhost_dev *dev, bool enable_log) |
384 |
{ |
385 |
int r, t, i;
|
386 |
r = vhost_dev_set_features(dev, enable_log); |
387 |
if (r < 0) { |
388 |
goto err_features;
|
389 |
} |
390 |
for (i = 0; i < dev->nvqs; ++i) { |
391 |
r = vhost_virtqueue_set_addr(dev, dev->vqs + i, i, |
392 |
enable_log); |
393 |
if (r < 0) { |
394 |
goto err_vq;
|
395 |
} |
396 |
} |
397 |
return 0; |
398 |
err_vq:
|
399 |
for (; i >= 0; --i) { |
400 |
t = vhost_virtqueue_set_addr(dev, dev->vqs + i, i, |
401 |
dev->log_enabled); |
402 |
assert(t >= 0);
|
403 |
} |
404 |
t = vhost_dev_set_features(dev, dev->log_enabled); |
405 |
assert(t >= 0);
|
406 |
err_features:
|
407 |
return r;
|
408 |
} |
409 |
|
410 |
static int vhost_client_migration_log(CPUPhysMemoryClient *client, |
411 |
int enable)
|
412 |
{ |
413 |
struct vhost_dev *dev = container_of(client, struct vhost_dev, client); |
414 |
int r;
|
415 |
if (!!enable == dev->log_enabled) {
|
416 |
return 0; |
417 |
} |
418 |
if (!dev->started) {
|
419 |
dev->log_enabled = enable; |
420 |
return 0; |
421 |
} |
422 |
if (!enable) {
|
423 |
r = vhost_dev_set_log(dev, false);
|
424 |
if (r < 0) { |
425 |
return r;
|
426 |
} |
427 |
if (dev->log) {
|
428 |
qemu_free(dev->log); |
429 |
} |
430 |
dev->log = NULL;
|
431 |
dev->log_size = 0;
|
432 |
} else {
|
433 |
vhost_dev_log_resize(dev, vhost_get_log_size(dev)); |
434 |
r = vhost_dev_set_log(dev, true);
|
435 |
if (r < 0) { |
436 |
return r;
|
437 |
} |
438 |
} |
439 |
dev->log_enabled = enable; |
440 |
return 0; |
441 |
} |
442 |
|
443 |
static int vhost_virtqueue_init(struct vhost_dev *dev, |
444 |
struct VirtIODevice *vdev,
|
445 |
struct vhost_virtqueue *vq,
|
446 |
unsigned idx)
|
447 |
{ |
448 |
target_phys_addr_t s, l, a; |
449 |
int r;
|
450 |
struct vhost_vring_file file = {
|
451 |
.index = idx, |
452 |
}; |
453 |
struct vhost_vring_state state = {
|
454 |
.index = idx, |
455 |
}; |
456 |
struct VirtQueue *vvq = virtio_get_queue(vdev, idx);
|
457 |
|
458 |
if (!vdev->binding->set_host_notifier) {
|
459 |
fprintf(stderr, "binding does not support host notifiers\n");
|
460 |
return -ENOSYS;
|
461 |
} |
462 |
|
463 |
vq->num = state.num = virtio_queue_get_num(vdev, idx); |
464 |
r = ioctl(dev->control, VHOST_SET_VRING_NUM, &state); |
465 |
if (r) {
|
466 |
return -errno;
|
467 |
} |
468 |
|
469 |
state.num = virtio_queue_get_last_avail_idx(vdev, idx); |
470 |
r = ioctl(dev->control, VHOST_SET_VRING_BASE, &state); |
471 |
if (r) {
|
472 |
return -errno;
|
473 |
} |
474 |
|
475 |
s = l = virtio_queue_get_desc_size(vdev, idx); |
476 |
a = virtio_queue_get_desc_addr(vdev, idx); |
477 |
vq->desc = cpu_physical_memory_map(a, &l, 0);
|
478 |
if (!vq->desc || l != s) {
|
479 |
r = -ENOMEM; |
480 |
goto fail_alloc_desc;
|
481 |
} |
482 |
s = l = virtio_queue_get_avail_size(vdev, idx); |
483 |
a = virtio_queue_get_avail_addr(vdev, idx); |
484 |
vq->avail = cpu_physical_memory_map(a, &l, 0);
|
485 |
if (!vq->avail || l != s) {
|
486 |
r = -ENOMEM; |
487 |
goto fail_alloc_avail;
|
488 |
} |
489 |
vq->used_size = s = l = virtio_queue_get_used_size(vdev, idx); |
490 |
vq->used_phys = a = virtio_queue_get_used_addr(vdev, idx); |
491 |
vq->used = cpu_physical_memory_map(a, &l, 1);
|
492 |
if (!vq->used || l != s) {
|
493 |
r = -ENOMEM; |
494 |
goto fail_alloc_used;
|
495 |
} |
496 |
|
497 |
vq->ring_size = s = l = virtio_queue_get_ring_size(vdev, idx); |
498 |
vq->ring_phys = a = virtio_queue_get_ring_addr(vdev, idx); |
499 |
vq->ring = cpu_physical_memory_map(a, &l, 1);
|
500 |
if (!vq->ring || l != s) {
|
501 |
r = -ENOMEM; |
502 |
goto fail_alloc_ring;
|
503 |
} |
504 |
|
505 |
r = vhost_virtqueue_set_addr(dev, vq, idx, dev->log_enabled); |
506 |
if (r < 0) { |
507 |
r = -errno; |
508 |
goto fail_alloc;
|
509 |
} |
510 |
r = vdev->binding->set_host_notifier(vdev->binding_opaque, idx, true);
|
511 |
if (r < 0) { |
512 |
fprintf(stderr, "Error binding host notifier: %d\n", -r);
|
513 |
goto fail_host_notifier;
|
514 |
} |
515 |
|
516 |
file.fd = event_notifier_get_fd(virtio_queue_get_host_notifier(vvq)); |
517 |
r = ioctl(dev->control, VHOST_SET_VRING_KICK, &file); |
518 |
if (r) {
|
519 |
r = -errno; |
520 |
goto fail_kick;
|
521 |
} |
522 |
|
523 |
file.fd = event_notifier_get_fd(virtio_queue_get_guest_notifier(vvq)); |
524 |
r = ioctl(dev->control, VHOST_SET_VRING_CALL, &file); |
525 |
if (r) {
|
526 |
r = -errno; |
527 |
goto fail_call;
|
528 |
} |
529 |
|
530 |
return 0; |
531 |
|
532 |
fail_call:
|
533 |
fail_kick:
|
534 |
vdev->binding->set_host_notifier(vdev->binding_opaque, idx, false);
|
535 |
fail_host_notifier:
|
536 |
fail_alloc:
|
537 |
cpu_physical_memory_unmap(vq->ring, virtio_queue_get_ring_size(vdev, idx), |
538 |
0, 0); |
539 |
fail_alloc_ring:
|
540 |
cpu_physical_memory_unmap(vq->used, virtio_queue_get_used_size(vdev, idx), |
541 |
0, 0); |
542 |
fail_alloc_used:
|
543 |
cpu_physical_memory_unmap(vq->avail, virtio_queue_get_avail_size(vdev, idx), |
544 |
0, 0); |
545 |
fail_alloc_avail:
|
546 |
cpu_physical_memory_unmap(vq->desc, virtio_queue_get_desc_size(vdev, idx), |
547 |
0, 0); |
548 |
fail_alloc_desc:
|
549 |
return r;
|
550 |
} |
551 |
|
552 |
static void vhost_virtqueue_cleanup(struct vhost_dev *dev, |
553 |
struct VirtIODevice *vdev,
|
554 |
struct vhost_virtqueue *vq,
|
555 |
unsigned idx)
|
556 |
{ |
557 |
struct vhost_vring_state state = {
|
558 |
.index = idx, |
559 |
}; |
560 |
int r;
|
561 |
r = vdev->binding->set_host_notifier(vdev->binding_opaque, idx, false);
|
562 |
if (r < 0) { |
563 |
fprintf(stderr, "vhost VQ %d host cleanup failed: %d\n", idx, r);
|
564 |
fflush(stderr); |
565 |
} |
566 |
assert (r >= 0);
|
567 |
r = ioctl(dev->control, VHOST_GET_VRING_BASE, &state); |
568 |
if (r < 0) { |
569 |
fprintf(stderr, "vhost VQ %d ring restore failed: %d\n", idx, r);
|
570 |
fflush(stderr); |
571 |
} |
572 |
virtio_queue_set_last_avail_idx(vdev, idx, state.num); |
573 |
assert (r >= 0);
|
574 |
cpu_physical_memory_unmap(vq->ring, virtio_queue_get_ring_size(vdev, idx), |
575 |
0, virtio_queue_get_ring_size(vdev, idx));
|
576 |
cpu_physical_memory_unmap(vq->used, virtio_queue_get_used_size(vdev, idx), |
577 |
1, virtio_queue_get_used_size(vdev, idx));
|
578 |
cpu_physical_memory_unmap(vq->avail, virtio_queue_get_avail_size(vdev, idx), |
579 |
0, virtio_queue_get_avail_size(vdev, idx));
|
580 |
cpu_physical_memory_unmap(vq->desc, virtio_queue_get_desc_size(vdev, idx), |
581 |
0, virtio_queue_get_desc_size(vdev, idx));
|
582 |
} |
583 |
|
584 |
int vhost_dev_init(struct vhost_dev *hdev, int devfd, bool force) |
585 |
{ |
586 |
uint64_t features; |
587 |
int r;
|
588 |
if (devfd >= 0) { |
589 |
hdev->control = devfd; |
590 |
} else {
|
591 |
hdev->control = open("/dev/vhost-net", O_RDWR);
|
592 |
if (hdev->control < 0) { |
593 |
return -errno;
|
594 |
} |
595 |
} |
596 |
r = ioctl(hdev->control, VHOST_SET_OWNER, NULL);
|
597 |
if (r < 0) { |
598 |
goto fail;
|
599 |
} |
600 |
|
601 |
r = ioctl(hdev->control, VHOST_GET_FEATURES, &features); |
602 |
if (r < 0) { |
603 |
goto fail;
|
604 |
} |
605 |
hdev->features = features; |
606 |
|
607 |
hdev->client.set_memory = vhost_client_set_memory; |
608 |
hdev->client.sync_dirty_bitmap = vhost_client_sync_dirty_bitmap; |
609 |
hdev->client.migration_log = vhost_client_migration_log; |
610 |
hdev->client.log_start = NULL;
|
611 |
hdev->client.log_stop = NULL;
|
612 |
hdev->mem = qemu_mallocz(offsetof(struct vhost_memory, regions));
|
613 |
hdev->log = NULL;
|
614 |
hdev->log_size = 0;
|
615 |
hdev->log_enabled = false;
|
616 |
hdev->started = false;
|
617 |
cpu_register_phys_memory_client(&hdev->client); |
618 |
hdev->force = force; |
619 |
return 0; |
620 |
fail:
|
621 |
r = -errno; |
622 |
close(hdev->control); |
623 |
return r;
|
624 |
} |
625 |
|
626 |
void vhost_dev_cleanup(struct vhost_dev *hdev) |
627 |
{ |
628 |
cpu_unregister_phys_memory_client(&hdev->client); |
629 |
qemu_free(hdev->mem); |
630 |
close(hdev->control); |
631 |
} |
632 |
|
633 |
bool vhost_dev_query(struct vhost_dev *hdev, VirtIODevice *vdev) |
634 |
{ |
635 |
return !vdev->binding->query_guest_notifiers ||
|
636 |
vdev->binding->query_guest_notifiers(vdev->binding_opaque) || |
637 |
hdev->force; |
638 |
} |
639 |
|
640 |
int vhost_dev_start(struct vhost_dev *hdev, VirtIODevice *vdev) |
641 |
{ |
642 |
int i, r;
|
643 |
if (!vdev->binding->set_guest_notifiers) {
|
644 |
fprintf(stderr, "binding does not support guest notifiers\n");
|
645 |
r = -ENOSYS; |
646 |
goto fail;
|
647 |
} |
648 |
|
649 |
r = vdev->binding->set_guest_notifiers(vdev->binding_opaque, true);
|
650 |
if (r < 0) { |
651 |
fprintf(stderr, "Error binding guest notifier: %d\n", -r);
|
652 |
goto fail_notifiers;
|
653 |
} |
654 |
|
655 |
r = vhost_dev_set_features(hdev, hdev->log_enabled); |
656 |
if (r < 0) { |
657 |
goto fail_features;
|
658 |
} |
659 |
r = ioctl(hdev->control, VHOST_SET_MEM_TABLE, hdev->mem); |
660 |
if (r < 0) { |
661 |
r = -errno; |
662 |
goto fail_mem;
|
663 |
} |
664 |
for (i = 0; i < hdev->nvqs; ++i) { |
665 |
r = vhost_virtqueue_init(hdev, |
666 |
vdev, |
667 |
hdev->vqs + i, |
668 |
i); |
669 |
if (r < 0) { |
670 |
goto fail_vq;
|
671 |
} |
672 |
} |
673 |
|
674 |
if (hdev->log_enabled) {
|
675 |
hdev->log_size = vhost_get_log_size(hdev); |
676 |
hdev->log = hdev->log_size ? |
677 |
qemu_mallocz(hdev->log_size * sizeof *hdev->log) : NULL; |
678 |
r = ioctl(hdev->control, VHOST_SET_LOG_BASE, |
679 |
(uint64_t)(unsigned long)hdev->log); |
680 |
if (r < 0) { |
681 |
r = -errno; |
682 |
goto fail_log;
|
683 |
} |
684 |
} |
685 |
|
686 |
hdev->started = true;
|
687 |
|
688 |
return 0; |
689 |
fail_log:
|
690 |
fail_vq:
|
691 |
while (--i >= 0) { |
692 |
vhost_virtqueue_cleanup(hdev, |
693 |
vdev, |
694 |
hdev->vqs + i, |
695 |
i); |
696 |
} |
697 |
fail_mem:
|
698 |
fail_features:
|
699 |
vdev->binding->set_guest_notifiers(vdev->binding_opaque, false);
|
700 |
fail_notifiers:
|
701 |
fail:
|
702 |
return r;
|
703 |
} |
704 |
|
705 |
void vhost_dev_stop(struct vhost_dev *hdev, VirtIODevice *vdev) |
706 |
{ |
707 |
int i, r;
|
708 |
|
709 |
for (i = 0; i < hdev->nvqs; ++i) { |
710 |
vhost_virtqueue_cleanup(hdev, |
711 |
vdev, |
712 |
hdev->vqs + i, |
713 |
i); |
714 |
} |
715 |
vhost_client_sync_dirty_bitmap(&hdev->client, 0,
|
716 |
(target_phys_addr_t)~0x0ull);
|
717 |
r = vdev->binding->set_guest_notifiers(vdev->binding_opaque, false);
|
718 |
if (r < 0) { |
719 |
fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", r);
|
720 |
fflush(stderr); |
721 |
} |
722 |
assert (r >= 0);
|
723 |
|
724 |
hdev->started = false;
|
725 |
qemu_free(hdev->log); |
726 |
hdev->log_size = 0;
|
727 |
} |