]>
Commit | Line | Data |
---|---|---|
d5970055 MT |
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. | |
6b620ca3 PB |
11 | * |
12 | * Contributions after 2012-01-13 are licensed under the terms of the | |
13 | * GNU GPL, version 2 or (at your option) any later version. | |
d5970055 MT |
14 | */ |
15 | ||
9b8bfe21 | 16 | #include "qemu/osdep.h" |
da34e65c | 17 | #include "qapi/error.h" |
0d09e41a | 18 | #include "hw/virtio/vhost.h" |
d5970055 | 19 | #include "hw/hw.h" |
5444e768 | 20 | #include "qemu/atomic.h" |
1de7afc9 | 21 | #include "qemu/range.h" |
04b7a152 | 22 | #include "qemu/error-report.h" |
15324404 | 23 | #include "qemu/memfd.h" |
11078ae3 | 24 | #include <linux/vhost.h> |
022c62cb | 25 | #include "exec/address-spaces.h" |
1c819449 | 26 | #include "hw/virtio/virtio-bus.h" |
04b7a152 | 27 | #include "hw/virtio/virtio-access.h" |
795c40b8 | 28 | #include "migration/blocker.h" |
c471ad0e | 29 | #include "sysemu/dma.h" |
aa3c40f6 | 30 | #include "trace.h" |
d5970055 | 31 | |
162bba7f MAL |
32 | /* enabled until disconnected backend stabilizes */ |
33 | #define _VHOST_DEBUG 1 | |
34 | ||
35 | #ifdef _VHOST_DEBUG | |
36 | #define VHOST_OPS_DEBUG(fmt, ...) \ | |
37 | do { error_report(fmt ": %s (%d)", ## __VA_ARGS__, \ | |
38 | strerror(errno), errno); } while (0) | |
39 | #else | |
40 | #define VHOST_OPS_DEBUG(fmt, ...) \ | |
41 | do { } while (0) | |
42 | #endif | |
43 | ||
309750fa | 44 | static struct vhost_log *vhost_log; |
15324404 | 45 | static struct vhost_log *vhost_log_shm; |
309750fa | 46 | |
2ce68e4c IM |
47 | static unsigned int used_memslots; |
48 | static QLIST_HEAD(, vhost_dev) vhost_devices = | |
49 | QLIST_HEAD_INITIALIZER(vhost_devices); | |
50 | ||
51 | bool vhost_has_free_slot(void) | |
52 | { | |
53 | unsigned int slots_limit = ~0U; | |
54 | struct vhost_dev *hdev; | |
55 | ||
56 | QLIST_FOREACH(hdev, &vhost_devices, entry) { | |
57 | unsigned int r = hdev->vhost_ops->vhost_backend_memslots_limit(hdev); | |
58 | slots_limit = MIN(slots_limit, r); | |
59 | } | |
60 | return slots_limit > used_memslots; | |
61 | } | |
62 | ||
d5970055 | 63 | static void vhost_dev_sync_region(struct vhost_dev *dev, |
2817b260 | 64 | MemoryRegionSection *section, |
d5970055 MT |
65 | uint64_t mfirst, uint64_t mlast, |
66 | uint64_t rfirst, uint64_t rlast) | |
67 | { | |
309750fa JW |
68 | vhost_log_chunk_t *log = dev->log->log; |
69 | ||
d5970055 MT |
70 | uint64_t start = MAX(mfirst, rfirst); |
71 | uint64_t end = MIN(mlast, rlast); | |
309750fa JW |
72 | vhost_log_chunk_t *from = log + start / VHOST_LOG_CHUNK; |
73 | vhost_log_chunk_t *to = log + end / VHOST_LOG_CHUNK + 1; | |
33c5793b | 74 | uint64_t addr = QEMU_ALIGN_DOWN(start, VHOST_LOG_CHUNK); |
d5970055 | 75 | |
d5970055 MT |
76 | if (end < start) { |
77 | return; | |
78 | } | |
e314672a | 79 | assert(end / VHOST_LOG_CHUNK < dev->log_size); |
fbbaf9ae | 80 | assert(start / VHOST_LOG_CHUNK < dev->log_size); |
e314672a | 81 | |
d5970055 MT |
82 | for (;from < to; ++from) { |
83 | vhost_log_chunk_t log; | |
d5970055 MT |
84 | /* We first check with non-atomic: much cheaper, |
85 | * and we expect non-dirty to be the common case. */ | |
86 | if (!*from) { | |
0c600ce2 | 87 | addr += VHOST_LOG_CHUNK; |
d5970055 MT |
88 | continue; |
89 | } | |
5444e768 PB |
90 | /* Data must be read atomically. We don't really need barrier semantics |
91 | * but it's easier to use atomic_* than roll our own. */ | |
92 | log = atomic_xchg(from, 0); | |
747eb78b NC |
93 | while (log) { |
94 | int bit = ctzl(log); | |
6b37a23d MT |
95 | hwaddr page_addr; |
96 | hwaddr section_offset; | |
97 | hwaddr mr_offset; | |
6b37a23d MT |
98 | page_addr = addr + bit * VHOST_LOG_PAGE; |
99 | section_offset = page_addr - section->offset_within_address_space; | |
100 | mr_offset = section_offset + section->offset_within_region; | |
101 | memory_region_set_dirty(section->mr, mr_offset, VHOST_LOG_PAGE); | |
d5970055 MT |
102 | log &= ~(0x1ull << bit); |
103 | } | |
104 | addr += VHOST_LOG_CHUNK; | |
105 | } | |
106 | } | |
107 | ||
04097f7c | 108 | static int vhost_sync_dirty_bitmap(struct vhost_dev *dev, |
2817b260 | 109 | MemoryRegionSection *section, |
6b37a23d MT |
110 | hwaddr first, |
111 | hwaddr last) | |
d5970055 | 112 | { |
d5970055 | 113 | int i; |
6b37a23d MT |
114 | hwaddr start_addr; |
115 | hwaddr end_addr; | |
04097f7c | 116 | |
d5970055 MT |
117 | if (!dev->log_enabled || !dev->started) { |
118 | return 0; | |
119 | } | |
6b37a23d | 120 | start_addr = section->offset_within_address_space; |
052e87b0 | 121 | end_addr = range_get_last(start_addr, int128_get64(section->size)); |
6b37a23d MT |
122 | start_addr = MAX(first, start_addr); |
123 | end_addr = MIN(last, end_addr); | |
124 | ||
d5970055 MT |
125 | for (i = 0; i < dev->mem->nregions; ++i) { |
126 | struct vhost_memory_region *reg = dev->mem->regions + i; | |
2817b260 | 127 | vhost_dev_sync_region(dev, section, start_addr, end_addr, |
d5970055 MT |
128 | reg->guest_phys_addr, |
129 | range_get_last(reg->guest_phys_addr, | |
130 | reg->memory_size)); | |
131 | } | |
132 | for (i = 0; i < dev->nvqs; ++i) { | |
133 | struct vhost_virtqueue *vq = dev->vqs + i; | |
2817b260 | 134 | vhost_dev_sync_region(dev, section, start_addr, end_addr, vq->used_phys, |
d5970055 MT |
135 | range_get_last(vq->used_phys, vq->used_size)); |
136 | } | |
137 | return 0; | |
138 | } | |
139 | ||
04097f7c AK |
140 | static void vhost_log_sync(MemoryListener *listener, |
141 | MemoryRegionSection *section) | |
142 | { | |
143 | struct vhost_dev *dev = container_of(listener, struct vhost_dev, | |
144 | memory_listener); | |
6b37a23d MT |
145 | vhost_sync_dirty_bitmap(dev, section, 0x0, ~0x0ULL); |
146 | } | |
04097f7c | 147 | |
6b37a23d MT |
148 | static void vhost_log_sync_range(struct vhost_dev *dev, |
149 | hwaddr first, hwaddr last) | |
150 | { | |
151 | int i; | |
152 | /* FIXME: this is N^2 in number of sections */ | |
153 | for (i = 0; i < dev->n_mem_sections; ++i) { | |
154 | MemoryRegionSection *section = &dev->mem_sections[i]; | |
155 | vhost_sync_dirty_bitmap(dev, section, first, last); | |
156 | } | |
04097f7c AK |
157 | } |
158 | ||
d5970055 MT |
159 | static uint64_t vhost_get_log_size(struct vhost_dev *dev) |
160 | { | |
161 | uint64_t log_size = 0; | |
162 | int i; | |
163 | for (i = 0; i < dev->mem->nregions; ++i) { | |
164 | struct vhost_memory_region *reg = dev->mem->regions + i; | |
165 | uint64_t last = range_get_last(reg->guest_phys_addr, | |
166 | reg->memory_size); | |
167 | log_size = MAX(log_size, last / VHOST_LOG_CHUNK + 1); | |
168 | } | |
169 | for (i = 0; i < dev->nvqs; ++i) { | |
170 | struct vhost_virtqueue *vq = dev->vqs + i; | |
171 | uint64_t last = vq->used_phys + vq->used_size - 1; | |
172 | log_size = MAX(log_size, last / VHOST_LOG_CHUNK + 1); | |
173 | } | |
174 | return log_size; | |
175 | } | |
15324404 MAL |
176 | |
177 | static struct vhost_log *vhost_log_alloc(uint64_t size, bool share) | |
309750fa | 178 | { |
0f2956f9 | 179 | Error *err = NULL; |
15324404 MAL |
180 | struct vhost_log *log; |
181 | uint64_t logsize = size * sizeof(*(log->log)); | |
182 | int fd = -1; | |
183 | ||
184 | log = g_new0(struct vhost_log, 1); | |
185 | if (share) { | |
186 | log->log = qemu_memfd_alloc("vhost-log", logsize, | |
187 | F_SEAL_GROW | F_SEAL_SHRINK | F_SEAL_SEAL, | |
0f2956f9 MAL |
188 | &fd, &err); |
189 | if (err) { | |
190 | error_report_err(err); | |
191 | g_free(log); | |
192 | return NULL; | |
193 | } | |
15324404 MAL |
194 | memset(log->log, 0, logsize); |
195 | } else { | |
196 | log->log = g_malloc0(logsize); | |
197 | } | |
309750fa JW |
198 | |
199 | log->size = size; | |
200 | log->refcnt = 1; | |
15324404 | 201 | log->fd = fd; |
309750fa JW |
202 | |
203 | return log; | |
204 | } | |
205 | ||
15324404 | 206 | static struct vhost_log *vhost_log_get(uint64_t size, bool share) |
309750fa | 207 | { |
15324404 MAL |
208 | struct vhost_log *log = share ? vhost_log_shm : vhost_log; |
209 | ||
210 | if (!log || log->size != size) { | |
211 | log = vhost_log_alloc(size, share); | |
212 | if (share) { | |
213 | vhost_log_shm = log; | |
214 | } else { | |
215 | vhost_log = log; | |
216 | } | |
309750fa | 217 | } else { |
15324404 | 218 | ++log->refcnt; |
309750fa JW |
219 | } |
220 | ||
15324404 | 221 | return log; |
309750fa JW |
222 | } |
223 | ||
224 | static void vhost_log_put(struct vhost_dev *dev, bool sync) | |
225 | { | |
226 | struct vhost_log *log = dev->log; | |
227 | ||
228 | if (!log) { | |
229 | return; | |
230 | } | |
231 | ||
232 | --log->refcnt; | |
233 | if (log->refcnt == 0) { | |
234 | /* Sync only the range covered by the old log */ | |
235 | if (dev->log_size && sync) { | |
236 | vhost_log_sync_range(dev, 0, dev->log_size * VHOST_LOG_CHUNK - 1); | |
237 | } | |
15324404 | 238 | |
309750fa | 239 | if (vhost_log == log) { |
15324404 | 240 | g_free(log->log); |
309750fa | 241 | vhost_log = NULL; |
15324404 MAL |
242 | } else if (vhost_log_shm == log) { |
243 | qemu_memfd_free(log->log, log->size * sizeof(*(log->log)), | |
244 | log->fd); | |
245 | vhost_log_shm = NULL; | |
309750fa | 246 | } |
15324404 | 247 | |
309750fa JW |
248 | g_free(log); |
249 | } | |
5c0ba1be FF |
250 | |
251 | dev->log = NULL; | |
252 | dev->log_size = 0; | |
309750fa | 253 | } |
d5970055 | 254 | |
15324404 MAL |
255 | static bool vhost_dev_log_is_shared(struct vhost_dev *dev) |
256 | { | |
257 | return dev->vhost_ops->vhost_requires_shm_log && | |
258 | dev->vhost_ops->vhost_requires_shm_log(dev); | |
259 | } | |
260 | ||
261 | static inline void vhost_dev_log_resize(struct vhost_dev *dev, uint64_t size) | |
d5970055 | 262 | { |
15324404 | 263 | struct vhost_log *log = vhost_log_get(size, vhost_dev_log_is_shared(dev)); |
309750fa | 264 | uint64_t log_base = (uintptr_t)log->log; |
6b37a23d | 265 | int r; |
6528499f | 266 | |
636f4ddd MAL |
267 | /* inform backend of log switching, this must be done before |
268 | releasing the current log, to ensure no logging is lost */ | |
9a78a5dd | 269 | r = dev->vhost_ops->vhost_set_log_base(dev, log_base, log); |
162bba7f MAL |
270 | if (r < 0) { |
271 | VHOST_OPS_DEBUG("vhost_set_log_base failed"); | |
272 | } | |
273 | ||
309750fa | 274 | vhost_log_put(dev, true); |
d5970055 MT |
275 | dev->log = log; |
276 | dev->log_size = size; | |
277 | } | |
278 | ||
c471ad0e JW |
279 | static int vhost_dev_has_iommu(struct vhost_dev *dev) |
280 | { | |
281 | VirtIODevice *vdev = dev->vdev; | |
c471ad0e | 282 | |
375f74f4 | 283 | return virtio_host_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM); |
c471ad0e JW |
284 | } |
285 | ||
286 | static void *vhost_memory_map(struct vhost_dev *dev, hwaddr addr, | |
287 | hwaddr *plen, int is_write) | |
288 | { | |
289 | if (!vhost_dev_has_iommu(dev)) { | |
290 | return cpu_physical_memory_map(addr, plen, is_write); | |
291 | } else { | |
292 | return (void *)(uintptr_t)addr; | |
293 | } | |
294 | } | |
295 | ||
296 | static void vhost_memory_unmap(struct vhost_dev *dev, void *buffer, | |
297 | hwaddr len, int is_write, | |
298 | hwaddr access_len) | |
299 | { | |
300 | if (!vhost_dev_has_iommu(dev)) { | |
301 | cpu_physical_memory_unmap(buffer, len, is_write, access_len); | |
302 | } | |
303 | } | |
f1f9e6c5 | 304 | |
0ca1fd2d DDAG |
305 | static int vhost_verify_ring_part_mapping(void *ring_hva, |
306 | uint64_t ring_gpa, | |
307 | uint64_t ring_size, | |
308 | void *reg_hva, | |
309 | uint64_t reg_gpa, | |
310 | uint64_t reg_size) | |
f1f9e6c5 | 311 | { |
0ca1fd2d DDAG |
312 | uint64_t hva_ring_offset; |
313 | uint64_t ring_last = range_get_last(ring_gpa, ring_size); | |
314 | uint64_t reg_last = range_get_last(reg_gpa, reg_size); | |
f1f9e6c5 | 315 | |
0ca1fd2d | 316 | if (ring_last < reg_gpa || ring_gpa > reg_last) { |
f1f9e6c5 GK |
317 | return 0; |
318 | } | |
0ca1fd2d DDAG |
319 | /* check that whole ring's is mapped */ |
320 | if (ring_last > reg_last) { | |
321 | return -ENOMEM; | |
f1f9e6c5 | 322 | } |
0ca1fd2d DDAG |
323 | /* check that ring's MemoryRegion wasn't replaced */ |
324 | hva_ring_offset = ring_gpa - reg_gpa; | |
325 | if (ring_hva != reg_hva + hva_ring_offset) { | |
326 | return -EBUSY; | |
f1f9e6c5 | 327 | } |
0ca1fd2d DDAG |
328 | |
329 | return 0; | |
f1f9e6c5 GK |
330 | } |
331 | ||
d5970055 | 332 | static int vhost_verify_ring_mappings(struct vhost_dev *dev, |
0ca1fd2d DDAG |
333 | void *reg_hva, |
334 | uint64_t reg_gpa, | |
335 | uint64_t reg_size) | |
d5970055 | 336 | { |
f1f9e6c5 | 337 | int i, j; |
8617343f | 338 | int r = 0; |
f1f9e6c5 GK |
339 | const char *part_name[] = { |
340 | "descriptor table", | |
341 | "available ring", | |
342 | "used ring" | |
343 | }; | |
8617343f | 344 | |
f1f9e6c5 | 345 | for (i = 0; i < dev->nvqs; ++i) { |
d5970055 | 346 | struct vhost_virtqueue *vq = dev->vqs + i; |
d5970055 | 347 | |
fb20fbb7 JH |
348 | if (vq->desc_phys == 0) { |
349 | continue; | |
350 | } | |
351 | ||
f1f9e6c5 | 352 | j = 0; |
0ca1fd2d DDAG |
353 | r = vhost_verify_ring_part_mapping( |
354 | vq->desc, vq->desc_phys, vq->desc_size, | |
355 | reg_hva, reg_gpa, reg_size); | |
2fe45ec3 | 356 | if (r) { |
f1f9e6c5 | 357 | break; |
d5970055 | 358 | } |
f1f9e6c5 GK |
359 | |
360 | j++; | |
0ca1fd2d | 361 | r = vhost_verify_ring_part_mapping( |
9fac50c8 | 362 | vq->avail, vq->avail_phys, vq->avail_size, |
0ca1fd2d | 363 | reg_hva, reg_gpa, reg_size); |
2fe45ec3 | 364 | if (r) { |
f1f9e6c5 | 365 | break; |
d5970055 | 366 | } |
f1f9e6c5 GK |
367 | |
368 | j++; | |
0ca1fd2d | 369 | r = vhost_verify_ring_part_mapping( |
9fac50c8 | 370 | vq->used, vq->used_phys, vq->used_size, |
0ca1fd2d | 371 | reg_hva, reg_gpa, reg_size); |
2fe45ec3 | 372 | if (r) { |
f1f9e6c5 | 373 | break; |
d5970055 | 374 | } |
f1f9e6c5 GK |
375 | } |
376 | ||
377 | if (r == -ENOMEM) { | |
378 | error_report("Unable to map %s for ring %d", part_name[j], i); | |
379 | } else if (r == -EBUSY) { | |
380 | error_report("%s relocated for ring %d", part_name[j], i); | |
d5970055 | 381 | } |
8617343f | 382 | return r; |
d5970055 MT |
383 | } |
384 | ||
af603142 NB |
385 | static bool vhost_section(MemoryRegionSection *section) |
386 | { | |
aa3c40f6 DDAG |
387 | bool result; |
388 | bool log_dirty = memory_region_get_dirty_log_mask(section->mr) & | |
389 | ~(1 << DIRTY_MEMORY_MIGRATION); | |
390 | result = memory_region_is_ram(section->mr) && | |
d56ec1e9 | 391 | !memory_region_is_rom(section->mr); |
aa3c40f6 DDAG |
392 | |
393 | /* Vhost doesn't handle any block which is doing dirty-tracking other | |
394 | * than migration; this typically fires on VGA areas. | |
395 | */ | |
396 | result &= !log_dirty; | |
397 | ||
398 | trace_vhost_section(section->mr->name, result); | |
399 | return result; | |
af603142 NB |
400 | } |
401 | ||
402 | static void vhost_begin(MemoryListener *listener) | |
403 | { | |
404 | struct vhost_dev *dev = container_of(listener, struct vhost_dev, | |
405 | memory_listener); | |
c44317ef DDAG |
406 | dev->tmp_sections = NULL; |
407 | dev->n_tmp_sections = 0; | |
af603142 | 408 | } |
d5970055 | 409 | |
af603142 NB |
410 | static void vhost_commit(MemoryListener *listener) |
411 | { | |
412 | struct vhost_dev *dev = container_of(listener, struct vhost_dev, | |
413 | memory_listener); | |
c44317ef DDAG |
414 | MemoryRegionSection *old_sections; |
415 | int n_old_sections; | |
af603142 | 416 | uint64_t log_size; |
ade6d081 | 417 | size_t regions_size; |
af603142 | 418 | int r; |
0ca1fd2d | 419 | int i; |
ade6d081 | 420 | bool changed = false; |
af603142 | 421 | |
ade6d081 DDAG |
422 | /* Note we can be called before the device is started, but then |
423 | * starting the device calls set_mem_table, so we need to have | |
424 | * built the data structures. | |
425 | */ | |
c44317ef DDAG |
426 | old_sections = dev->mem_sections; |
427 | n_old_sections = dev->n_mem_sections; | |
428 | dev->mem_sections = dev->tmp_sections; | |
429 | dev->n_mem_sections = dev->n_tmp_sections; | |
430 | ||
ade6d081 DDAG |
431 | if (dev->n_mem_sections != n_old_sections) { |
432 | changed = true; | |
433 | } else { | |
434 | /* Same size, lets check the contents */ | |
435 | changed = n_old_sections && memcmp(dev->mem_sections, old_sections, | |
436 | n_old_sections * sizeof(old_sections[0])) != 0; | |
af603142 | 437 | } |
ade6d081 DDAG |
438 | |
439 | trace_vhost_commit(dev->started, changed); | |
440 | if (!changed) { | |
c44317ef | 441 | goto out; |
d5970055 | 442 | } |
ade6d081 DDAG |
443 | |
444 | /* Rebuild the regions list from the new sections list */ | |
445 | regions_size = offsetof(struct vhost_memory, regions) + | |
446 | dev->n_mem_sections * sizeof dev->mem->regions[0]; | |
447 | dev->mem = g_realloc(dev->mem, regions_size); | |
448 | dev->mem->nregions = dev->n_mem_sections; | |
449 | used_memslots = dev->mem->nregions; | |
450 | for (i = 0; i < dev->n_mem_sections; i++) { | |
451 | struct vhost_memory_region *cur_vmr = dev->mem->regions + i; | |
452 | struct MemoryRegionSection *mrs = dev->mem_sections + i; | |
453 | ||
454 | cur_vmr->guest_phys_addr = mrs->offset_within_address_space; | |
455 | cur_vmr->memory_size = int128_get64(mrs->size); | |
456 | cur_vmr->userspace_addr = | |
457 | (uintptr_t)memory_region_get_ram_ptr(mrs->mr) + | |
458 | mrs->offset_within_region; | |
459 | cur_vmr->flags_padding = 0; | |
460 | } | |
461 | ||
462 | if (!dev->started) { | |
c44317ef | 463 | goto out; |
af603142 | 464 | } |
d5970055 | 465 | |
0ca1fd2d DDAG |
466 | for (i = 0; i < dev->mem->nregions; i++) { |
467 | if (vhost_verify_ring_mappings(dev, | |
468 | (void *)(uintptr_t)dev->mem->regions[i].userspace_addr, | |
469 | dev->mem->regions[i].guest_phys_addr, | |
470 | dev->mem->regions[i].memory_size)) { | |
471 | error_report("Verify ring failure on region %d", i); | |
472 | abort(); | |
473 | } | |
d5970055 MT |
474 | } |
475 | ||
476 | if (!dev->log_enabled) { | |
21e70425 | 477 | r = dev->vhost_ops->vhost_set_mem_table(dev, dev->mem); |
162bba7f MAL |
478 | if (r < 0) { |
479 | VHOST_OPS_DEBUG("vhost_set_mem_table failed"); | |
480 | } | |
c44317ef | 481 | goto out; |
d5970055 MT |
482 | } |
483 | log_size = vhost_get_log_size(dev); | |
484 | /* We allocate an extra 4K bytes to log, | |
485 | * to reduce the * number of reallocations. */ | |
486 | #define VHOST_LOG_BUFFER (0x1000 / sizeof *dev->log) | |
487 | /* To log more, must increase log size before table update. */ | |
488 | if (dev->log_size < log_size) { | |
489 | vhost_dev_log_resize(dev, log_size + VHOST_LOG_BUFFER); | |
490 | } | |
21e70425 | 491 | r = dev->vhost_ops->vhost_set_mem_table(dev, dev->mem); |
162bba7f MAL |
492 | if (r < 0) { |
493 | VHOST_OPS_DEBUG("vhost_set_mem_table failed"); | |
494 | } | |
d5970055 MT |
495 | /* To log less, can only decrease log size after table update. */ |
496 | if (dev->log_size > log_size + VHOST_LOG_BUFFER) { | |
497 | vhost_dev_log_resize(dev, log_size); | |
498 | } | |
c44317ef DDAG |
499 | |
500 | out: | |
501 | /* Deref the old list of sections, this must happen _after_ the | |
502 | * vhost_set_mem_table to ensure the client isn't still using the | |
503 | * section we're about to unref. | |
504 | */ | |
505 | while (n_old_sections--) { | |
506 | memory_region_unref(old_sections[n_old_sections].mr); | |
507 | } | |
508 | g_free(old_sections); | |
509 | return; | |
510 | } | |
511 | ||
48d7c975 DDAG |
512 | /* Adds the section data to the tmp_section structure. |
513 | * It relies on the listener calling us in memory address order | |
514 | * and for each region (via the _add and _nop methods) to | |
515 | * join neighbours. | |
516 | */ | |
517 | static void vhost_region_add_section(struct vhost_dev *dev, | |
518 | MemoryRegionSection *section) | |
c44317ef | 519 | { |
48d7c975 DDAG |
520 | bool need_add = true; |
521 | uint64_t mrs_size = int128_get64(section->size); | |
522 | uint64_t mrs_gpa = section->offset_within_address_space; | |
523 | uintptr_t mrs_host = (uintptr_t)memory_region_get_ram_ptr(section->mr) + | |
524 | section->offset_within_region; | |
c1ece84e DDAG |
525 | RAMBlock *mrs_rb = section->mr->ram_block; |
526 | size_t mrs_page = qemu_ram_pagesize(mrs_rb); | |
48d7c975 DDAG |
527 | |
528 | trace_vhost_region_add_section(section->mr->name, mrs_gpa, mrs_size, | |
529 | mrs_host); | |
530 | ||
c1ece84e DDAG |
531 | /* Round the section to it's page size */ |
532 | /* First align the start down to a page boundary */ | |
533 | uint64_t alignage = mrs_host & (mrs_page - 1); | |
534 | if (alignage) { | |
535 | mrs_host -= alignage; | |
536 | mrs_size += alignage; | |
537 | mrs_gpa -= alignage; | |
538 | } | |
539 | /* Now align the size up to a page boundary */ | |
540 | alignage = mrs_size & (mrs_page - 1); | |
541 | if (alignage) { | |
542 | mrs_size += mrs_page - alignage; | |
543 | } | |
544 | trace_vhost_region_add_section_aligned(section->mr->name, mrs_gpa, mrs_size, | |
545 | mrs_host); | |
546 | ||
48d7c975 DDAG |
547 | if (dev->n_tmp_sections) { |
548 | /* Since we already have at least one section, lets see if | |
549 | * this extends it; since we're scanning in order, we only | |
550 | * have to look at the last one, and the FlatView that calls | |
551 | * us shouldn't have overlaps. | |
552 | */ | |
553 | MemoryRegionSection *prev_sec = dev->tmp_sections + | |
554 | (dev->n_tmp_sections - 1); | |
555 | uint64_t prev_gpa_start = prev_sec->offset_within_address_space; | |
556 | uint64_t prev_size = int128_get64(prev_sec->size); | |
557 | uint64_t prev_gpa_end = range_get_last(prev_gpa_start, prev_size); | |
558 | uint64_t prev_host_start = | |
559 | (uintptr_t)memory_region_get_ram_ptr(prev_sec->mr) + | |
560 | prev_sec->offset_within_region; | |
561 | uint64_t prev_host_end = range_get_last(prev_host_start, prev_size); | |
562 | ||
c1ece84e DDAG |
563 | if (mrs_gpa <= (prev_gpa_end + 1)) { |
564 | /* OK, looks like overlapping/intersecting - it's possible that | |
565 | * the rounding to page sizes has made them overlap, but they should | |
566 | * match up in the same RAMBlock if they do. | |
567 | */ | |
568 | if (mrs_gpa < prev_gpa_start) { | |
569 | error_report("%s:Section rounded to %"PRIx64 | |
570 | " prior to previous %"PRIx64, | |
571 | __func__, mrs_gpa, prev_gpa_start); | |
572 | /* A way to cleanly fail here would be better */ | |
573 | return; | |
574 | } | |
575 | /* Offset from the start of the previous GPA to this GPA */ | |
576 | size_t offset = mrs_gpa - prev_gpa_start; | |
577 | ||
578 | if (prev_host_start + offset == mrs_host && | |
579 | section->mr == prev_sec->mr && | |
580 | (!dev->vhost_ops->vhost_backend_can_merge || | |
581 | dev->vhost_ops->vhost_backend_can_merge(dev, | |
48d7c975 DDAG |
582 | mrs_host, mrs_size, |
583 | prev_host_start, prev_size))) { | |
c1ece84e DDAG |
584 | uint64_t max_end = MAX(prev_host_end, mrs_host + mrs_size); |
585 | need_add = false; | |
586 | prev_sec->offset_within_address_space = | |
587 | MIN(prev_gpa_start, mrs_gpa); | |
588 | prev_sec->offset_within_region = | |
589 | MIN(prev_host_start, mrs_host) - | |
590 | (uintptr_t)memory_region_get_ram_ptr(prev_sec->mr); | |
591 | prev_sec->size = int128_make64(max_end - MIN(prev_host_start, | |
592 | mrs_host)); | |
593 | trace_vhost_region_add_section_merge(section->mr->name, | |
594 | int128_get64(prev_sec->size), | |
595 | prev_sec->offset_within_address_space, | |
596 | prev_sec->offset_within_region); | |
597 | } else { | |
598 | error_report("%s: Overlapping but not coherent sections " | |
599 | "at %"PRIx64, | |
600 | __func__, mrs_gpa); | |
601 | return; | |
602 | } | |
48d7c975 DDAG |
603 | } |
604 | } | |
605 | ||
606 | if (need_add) { | |
607 | ++dev->n_tmp_sections; | |
608 | dev->tmp_sections = g_renew(MemoryRegionSection, dev->tmp_sections, | |
609 | dev->n_tmp_sections); | |
610 | dev->tmp_sections[dev->n_tmp_sections - 1] = *section; | |
611 | /* The flatview isn't stable and we don't use it, making it NULL | |
612 | * means we can memcmp the list. | |
613 | */ | |
614 | dev->tmp_sections[dev->n_tmp_sections - 1].fv = NULL; | |
615 | memory_region_ref(section->mr); | |
616 | } | |
50c1e149 AK |
617 | } |
618 | ||
938eeb64 DDAG |
619 | /* Used for both add and nop callbacks */ |
620 | static void vhost_region_addnop(MemoryListener *listener, | |
621 | MemoryRegionSection *section) | |
04097f7c | 622 | { |
2817b260 AK |
623 | struct vhost_dev *dev = container_of(listener, struct vhost_dev, |
624 | memory_listener); | |
625 | ||
c49450b9 AK |
626 | if (!vhost_section(section)) { |
627 | return; | |
628 | } | |
48d7c975 | 629 | vhost_region_add_section(dev, section); |
04097f7c AK |
630 | } |
631 | ||
375f74f4 JW |
632 | static void vhost_iommu_unmap_notify(IOMMUNotifier *n, IOMMUTLBEntry *iotlb) |
633 | { | |
634 | struct vhost_iommu *iommu = container_of(n, struct vhost_iommu, n); | |
635 | struct vhost_dev *hdev = iommu->hdev; | |
636 | hwaddr iova = iotlb->iova + iommu->iommu_offset; | |
637 | ||
020e571b MC |
638 | if (vhost_backend_invalidate_device_iotlb(hdev, iova, |
639 | iotlb->addr_mask + 1)) { | |
375f74f4 JW |
640 | error_report("Fail to invalidate device iotlb"); |
641 | } | |
642 | } | |
643 | ||
644 | static void vhost_iommu_region_add(MemoryListener *listener, | |
645 | MemoryRegionSection *section) | |
646 | { | |
647 | struct vhost_dev *dev = container_of(listener, struct vhost_dev, | |
648 | iommu_listener); | |
649 | struct vhost_iommu *iommu; | |
698feb5e | 650 | Int128 end; |
375f74f4 JW |
651 | |
652 | if (!memory_region_is_iommu(section->mr)) { | |
653 | return; | |
654 | } | |
655 | ||
656 | iommu = g_malloc0(sizeof(*iommu)); | |
698feb5e PX |
657 | end = int128_add(int128_make64(section->offset_within_region), |
658 | section->size); | |
659 | end = int128_sub(end, int128_one()); | |
660 | iommu_notifier_init(&iommu->n, vhost_iommu_unmap_notify, | |
661 | IOMMU_NOTIFIER_UNMAP, | |
662 | section->offset_within_region, | |
663 | int128_get64(end)); | |
375f74f4 JW |
664 | iommu->mr = section->mr; |
665 | iommu->iommu_offset = section->offset_within_address_space - | |
666 | section->offset_within_region; | |
667 | iommu->hdev = dev; | |
668 | memory_region_register_iommu_notifier(section->mr, &iommu->n); | |
669 | QLIST_INSERT_HEAD(&dev->iommu_list, iommu, iommu_next); | |
670 | /* TODO: can replay help performance here? */ | |
671 | } | |
672 | ||
673 | static void vhost_iommu_region_del(MemoryListener *listener, | |
674 | MemoryRegionSection *section) | |
675 | { | |
676 | struct vhost_dev *dev = container_of(listener, struct vhost_dev, | |
677 | iommu_listener); | |
678 | struct vhost_iommu *iommu; | |
679 | ||
680 | if (!memory_region_is_iommu(section->mr)) { | |
681 | return; | |
682 | } | |
683 | ||
684 | QLIST_FOREACH(iommu, &dev->iommu_list, iommu_next) { | |
698feb5e PX |
685 | if (iommu->mr == section->mr && |
686 | iommu->n.start == section->offset_within_region) { | |
375f74f4 JW |
687 | memory_region_unregister_iommu_notifier(iommu->mr, |
688 | &iommu->n); | |
689 | QLIST_REMOVE(iommu, iommu_next); | |
690 | g_free(iommu); | |
691 | break; | |
692 | } | |
693 | } | |
694 | } | |
695 | ||
d5970055 MT |
696 | static int vhost_virtqueue_set_addr(struct vhost_dev *dev, |
697 | struct vhost_virtqueue *vq, | |
698 | unsigned idx, bool enable_log) | |
699 | { | |
700 | struct vhost_vring_addr addr = { | |
701 | .index = idx, | |
2b3af999 SW |
702 | .desc_user_addr = (uint64_t)(unsigned long)vq->desc, |
703 | .avail_user_addr = (uint64_t)(unsigned long)vq->avail, | |
704 | .used_user_addr = (uint64_t)(unsigned long)vq->used, | |
d5970055 MT |
705 | .log_guest_addr = vq->used_phys, |
706 | .flags = enable_log ? (1 << VHOST_VRING_F_LOG) : 0, | |
707 | }; | |
21e70425 | 708 | int r = dev->vhost_ops->vhost_set_vring_addr(dev, &addr); |
d5970055 | 709 | if (r < 0) { |
c6409692 | 710 | VHOST_OPS_DEBUG("vhost_set_vring_addr failed"); |
d5970055 MT |
711 | return -errno; |
712 | } | |
713 | return 0; | |
714 | } | |
715 | ||
c471ad0e JW |
716 | static int vhost_dev_set_features(struct vhost_dev *dev, |
717 | bool enable_log) | |
d5970055 MT |
718 | { |
719 | uint64_t features = dev->acked_features; | |
720 | int r; | |
721 | if (enable_log) { | |
9a2ba823 | 722 | features |= 0x1ULL << VHOST_F_LOG_ALL; |
d5970055 | 723 | } |
21e70425 | 724 | r = dev->vhost_ops->vhost_set_features(dev, features); |
c6409692 MAL |
725 | if (r < 0) { |
726 | VHOST_OPS_DEBUG("vhost_set_features failed"); | |
727 | } | |
d5970055 MT |
728 | return r < 0 ? -errno : 0; |
729 | } | |
730 | ||
731 | static int vhost_dev_set_log(struct vhost_dev *dev, bool enable_log) | |
732 | { | |
162bba7f | 733 | int r, i, idx; |
d5970055 MT |
734 | r = vhost_dev_set_features(dev, enable_log); |
735 | if (r < 0) { | |
736 | goto err_features; | |
737 | } | |
738 | for (i = 0; i < dev->nvqs; ++i) { | |
25a2a920 TC |
739 | idx = dev->vhost_ops->vhost_get_vq_index(dev, dev->vq_index + i); |
740 | r = vhost_virtqueue_set_addr(dev, dev->vqs + i, idx, | |
d5970055 MT |
741 | enable_log); |
742 | if (r < 0) { | |
743 | goto err_vq; | |
744 | } | |
745 | } | |
746 | return 0; | |
747 | err_vq: | |
748 | for (; i >= 0; --i) { | |
25a2a920 | 749 | idx = dev->vhost_ops->vhost_get_vq_index(dev, dev->vq_index + i); |
162bba7f MAL |
750 | vhost_virtqueue_set_addr(dev, dev->vqs + i, idx, |
751 | dev->log_enabled); | |
d5970055 | 752 | } |
162bba7f | 753 | vhost_dev_set_features(dev, dev->log_enabled); |
d5970055 MT |
754 | err_features: |
755 | return r; | |
756 | } | |
757 | ||
04097f7c | 758 | static int vhost_migration_log(MemoryListener *listener, int enable) |
d5970055 | 759 | { |
04097f7c AK |
760 | struct vhost_dev *dev = container_of(listener, struct vhost_dev, |
761 | memory_listener); | |
d5970055 MT |
762 | int r; |
763 | if (!!enable == dev->log_enabled) { | |
764 | return 0; | |
765 | } | |
766 | if (!dev->started) { | |
767 | dev->log_enabled = enable; | |
768 | return 0; | |
769 | } | |
770 | if (!enable) { | |
771 | r = vhost_dev_set_log(dev, false); | |
772 | if (r < 0) { | |
773 | return r; | |
774 | } | |
309750fa | 775 | vhost_log_put(dev, false); |
d5970055 MT |
776 | } else { |
777 | vhost_dev_log_resize(dev, vhost_get_log_size(dev)); | |
778 | r = vhost_dev_set_log(dev, true); | |
779 | if (r < 0) { | |
780 | return r; | |
781 | } | |
782 | } | |
783 | dev->log_enabled = enable; | |
784 | return 0; | |
785 | } | |
786 | ||
04097f7c AK |
787 | static void vhost_log_global_start(MemoryListener *listener) |
788 | { | |
789 | int r; | |
790 | ||
791 | r = vhost_migration_log(listener, true); | |
792 | if (r < 0) { | |
793 | abort(); | |
794 | } | |
795 | } | |
796 | ||
797 | static void vhost_log_global_stop(MemoryListener *listener) | |
798 | { | |
799 | int r; | |
800 | ||
801 | r = vhost_migration_log(listener, false); | |
802 | if (r < 0) { | |
803 | abort(); | |
804 | } | |
805 | } | |
806 | ||
807 | static void vhost_log_start(MemoryListener *listener, | |
b2dfd71c PB |
808 | MemoryRegionSection *section, |
809 | int old, int new) | |
04097f7c AK |
810 | { |
811 | /* FIXME: implement */ | |
812 | } | |
813 | ||
814 | static void vhost_log_stop(MemoryListener *listener, | |
b2dfd71c PB |
815 | MemoryRegionSection *section, |
816 | int old, int new) | |
04097f7c AK |
817 | { |
818 | /* FIXME: implement */ | |
819 | } | |
820 | ||
46f70ff1 GK |
821 | /* The vhost driver natively knows how to handle the vrings of non |
822 | * cross-endian legacy devices and modern devices. Only legacy devices | |
823 | * exposed to a bi-endian guest may require the vhost driver to use a | |
824 | * specific endianness. | |
825 | */ | |
a122ab24 GK |
826 | static inline bool vhost_needs_vring_endian(VirtIODevice *vdev) |
827 | { | |
e5848123 GK |
828 | if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) { |
829 | return false; | |
830 | } | |
a122ab24 | 831 | #ifdef HOST_WORDS_BIGENDIAN |
46f70ff1 | 832 | return vdev->device_endian == VIRTIO_DEVICE_ENDIAN_LITTLE; |
a122ab24 | 833 | #else |
46f70ff1 | 834 | return vdev->device_endian == VIRTIO_DEVICE_ENDIAN_BIG; |
a122ab24 | 835 | #endif |
a122ab24 GK |
836 | } |
837 | ||
04b7a152 GK |
838 | static int vhost_virtqueue_set_vring_endian_legacy(struct vhost_dev *dev, |
839 | bool is_big_endian, | |
840 | int vhost_vq_index) | |
841 | { | |
842 | struct vhost_vring_state s = { | |
843 | .index = vhost_vq_index, | |
844 | .num = is_big_endian | |
845 | }; | |
846 | ||
21e70425 | 847 | if (!dev->vhost_ops->vhost_set_vring_endian(dev, &s)) { |
04b7a152 GK |
848 | return 0; |
849 | } | |
850 | ||
c6409692 | 851 | VHOST_OPS_DEBUG("vhost_set_vring_endian failed"); |
04b7a152 GK |
852 | if (errno == ENOTTY) { |
853 | error_report("vhost does not support cross-endian"); | |
854 | return -ENOSYS; | |
855 | } | |
856 | ||
857 | return -errno; | |
858 | } | |
859 | ||
c471ad0e JW |
860 | static int vhost_memory_region_lookup(struct vhost_dev *hdev, |
861 | uint64_t gpa, uint64_t *uaddr, | |
862 | uint64_t *len) | |
863 | { | |
864 | int i; | |
865 | ||
866 | for (i = 0; i < hdev->mem->nregions; i++) { | |
867 | struct vhost_memory_region *reg = hdev->mem->regions + i; | |
868 | ||
869 | if (gpa >= reg->guest_phys_addr && | |
870 | reg->guest_phys_addr + reg->memory_size > gpa) { | |
871 | *uaddr = reg->userspace_addr + gpa - reg->guest_phys_addr; | |
872 | *len = reg->guest_phys_addr + reg->memory_size - gpa; | |
873 | return 0; | |
874 | } | |
875 | } | |
876 | ||
877 | return -EFAULT; | |
878 | } | |
879 | ||
fc58bd0d | 880 | int vhost_device_iotlb_miss(struct vhost_dev *dev, uint64_t iova, int write) |
c471ad0e JW |
881 | { |
882 | IOMMUTLBEntry iotlb; | |
883 | uint64_t uaddr, len; | |
fc58bd0d | 884 | int ret = -EFAULT; |
c471ad0e JW |
885 | |
886 | rcu_read_lock(); | |
887 | ||
888 | iotlb = address_space_get_iotlb_entry(dev->vdev->dma_as, | |
889 | iova, write); | |
890 | if (iotlb.target_as != NULL) { | |
fc58bd0d MC |
891 | ret = vhost_memory_region_lookup(dev, iotlb.translated_addr, |
892 | &uaddr, &len); | |
893 | if (ret) { | |
c471ad0e JW |
894 | error_report("Fail to lookup the translated address " |
895 | "%"PRIx64, iotlb.translated_addr); | |
896 | goto out; | |
897 | } | |
898 | ||
899 | len = MIN(iotlb.addr_mask + 1, len); | |
900 | iova = iova & ~iotlb.addr_mask; | |
901 | ||
020e571b MC |
902 | ret = vhost_backend_update_device_iotlb(dev, iova, uaddr, |
903 | len, iotlb.perm); | |
fc58bd0d | 904 | if (ret) { |
c471ad0e JW |
905 | error_report("Fail to update device iotlb"); |
906 | goto out; | |
907 | } | |
908 | } | |
909 | out: | |
910 | rcu_read_unlock(); | |
fc58bd0d MC |
911 | |
912 | return ret; | |
c471ad0e JW |
913 | } |
914 | ||
f56a1247 | 915 | static int vhost_virtqueue_start(struct vhost_dev *dev, |
d5970055 MT |
916 | struct VirtIODevice *vdev, |
917 | struct vhost_virtqueue *vq, | |
918 | unsigned idx) | |
919 | { | |
96a3d98d JW |
920 | BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(vdev))); |
921 | VirtioBusState *vbus = VIRTIO_BUS(qbus); | |
922 | VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus); | |
a8170e5e | 923 | hwaddr s, l, a; |
d5970055 | 924 | int r; |
21e70425 | 925 | int vhost_vq_index = dev->vhost_ops->vhost_get_vq_index(dev, idx); |
d5970055 | 926 | struct vhost_vring_file file = { |
a9f98bb5 | 927 | .index = vhost_vq_index |
d5970055 MT |
928 | }; |
929 | struct vhost_vring_state state = { | |
a9f98bb5 | 930 | .index = vhost_vq_index |
d5970055 MT |
931 | }; |
932 | struct VirtQueue *vvq = virtio_get_queue(vdev, idx); | |
933 | ||
fb20fbb7 JH |
934 | a = virtio_queue_get_desc_addr(vdev, idx); |
935 | if (a == 0) { | |
936 | /* Queue might not be ready for start */ | |
937 | return 0; | |
938 | } | |
a9f98bb5 | 939 | |
d5970055 | 940 | vq->num = state.num = virtio_queue_get_num(vdev, idx); |
21e70425 | 941 | r = dev->vhost_ops->vhost_set_vring_num(dev, &state); |
d5970055 | 942 | if (r) { |
c6409692 | 943 | VHOST_OPS_DEBUG("vhost_set_vring_num failed"); |
d5970055 MT |
944 | return -errno; |
945 | } | |
946 | ||
947 | state.num = virtio_queue_get_last_avail_idx(vdev, idx); | |
21e70425 | 948 | r = dev->vhost_ops->vhost_set_vring_base(dev, &state); |
d5970055 | 949 | if (r) { |
c6409692 | 950 | VHOST_OPS_DEBUG("vhost_set_vring_base failed"); |
d5970055 MT |
951 | return -errno; |
952 | } | |
953 | ||
e5848123 | 954 | if (vhost_needs_vring_endian(vdev)) { |
04b7a152 GK |
955 | r = vhost_virtqueue_set_vring_endian_legacy(dev, |
956 | virtio_is_big_endian(vdev), | |
957 | vhost_vq_index); | |
958 | if (r) { | |
959 | return -errno; | |
960 | } | |
961 | } | |
962 | ||
f1f9e6c5 | 963 | vq->desc_size = s = l = virtio_queue_get_desc_size(vdev, idx); |
fb20fbb7 | 964 | vq->desc_phys = a; |
c471ad0e | 965 | vq->desc = vhost_memory_map(dev, a, &l, 0); |
d5970055 MT |
966 | if (!vq->desc || l != s) { |
967 | r = -ENOMEM; | |
968 | goto fail_alloc_desc; | |
969 | } | |
f1f9e6c5 GK |
970 | vq->avail_size = s = l = virtio_queue_get_avail_size(vdev, idx); |
971 | vq->avail_phys = a = virtio_queue_get_avail_addr(vdev, idx); | |
c471ad0e | 972 | vq->avail = vhost_memory_map(dev, a, &l, 0); |
d5970055 MT |
973 | if (!vq->avail || l != s) { |
974 | r = -ENOMEM; | |
975 | goto fail_alloc_avail; | |
976 | } | |
977 | vq->used_size = s = l = virtio_queue_get_used_size(vdev, idx); | |
978 | vq->used_phys = a = virtio_queue_get_used_addr(vdev, idx); | |
c471ad0e | 979 | vq->used = vhost_memory_map(dev, a, &l, 1); |
d5970055 MT |
980 | if (!vq->used || l != s) { |
981 | r = -ENOMEM; | |
982 | goto fail_alloc_used; | |
983 | } | |
984 | ||
a9f98bb5 | 985 | r = vhost_virtqueue_set_addr(dev, vq, vhost_vq_index, dev->log_enabled); |
d5970055 MT |
986 | if (r < 0) { |
987 | r = -errno; | |
988 | goto fail_alloc; | |
989 | } | |
a9f98bb5 | 990 | |
d5970055 | 991 | file.fd = event_notifier_get_fd(virtio_queue_get_host_notifier(vvq)); |
21e70425 | 992 | r = dev->vhost_ops->vhost_set_vring_kick(dev, &file); |
d5970055 | 993 | if (r) { |
c6409692 | 994 | VHOST_OPS_DEBUG("vhost_set_vring_kick failed"); |
c8852121 | 995 | r = -errno; |
d5970055 MT |
996 | goto fail_kick; |
997 | } | |
998 | ||
f56a1247 MT |
999 | /* Clear and discard previous events if any. */ |
1000 | event_notifier_test_and_clear(&vq->masked_notifier); | |
d5970055 | 1001 | |
5669655a VK |
1002 | /* Init vring in unmasked state, unless guest_notifier_mask |
1003 | * will do it later. | |
1004 | */ | |
1005 | if (!vdev->use_guest_notifier_mask) { | |
1006 | /* TODO: check and handle errors. */ | |
1007 | vhost_virtqueue_mask(dev, vdev, idx, false); | |
1008 | } | |
1009 | ||
96a3d98d JW |
1010 | if (k->query_guest_notifiers && |
1011 | k->query_guest_notifiers(qbus->parent) && | |
1012 | virtio_queue_vector(vdev, idx) == VIRTIO_NO_VECTOR) { | |
1013 | file.fd = -1; | |
1014 | r = dev->vhost_ops->vhost_set_vring_call(dev, &file); | |
1015 | if (r) { | |
1016 | goto fail_vector; | |
1017 | } | |
1018 | } | |
1019 | ||
d5970055 MT |
1020 | return 0; |
1021 | ||
96a3d98d | 1022 | fail_vector: |
d5970055 | 1023 | fail_kick: |
d5970055 | 1024 | fail_alloc: |
c471ad0e JW |
1025 | vhost_memory_unmap(dev, vq->used, virtio_queue_get_used_size(vdev, idx), |
1026 | 0, 0); | |
d5970055 | 1027 | fail_alloc_used: |
c471ad0e JW |
1028 | vhost_memory_unmap(dev, vq->avail, virtio_queue_get_avail_size(vdev, idx), |
1029 | 0, 0); | |
d5970055 | 1030 | fail_alloc_avail: |
c471ad0e JW |
1031 | vhost_memory_unmap(dev, vq->desc, virtio_queue_get_desc_size(vdev, idx), |
1032 | 0, 0); | |
d5970055 MT |
1033 | fail_alloc_desc: |
1034 | return r; | |
1035 | } | |
1036 | ||
f56a1247 | 1037 | static void vhost_virtqueue_stop(struct vhost_dev *dev, |
d5970055 MT |
1038 | struct VirtIODevice *vdev, |
1039 | struct vhost_virtqueue *vq, | |
1040 | unsigned idx) | |
1041 | { | |
21e70425 | 1042 | int vhost_vq_index = dev->vhost_ops->vhost_get_vq_index(dev, idx); |
d5970055 | 1043 | struct vhost_vring_state state = { |
04b7a152 | 1044 | .index = vhost_vq_index, |
d5970055 MT |
1045 | }; |
1046 | int r; | |
fb20fbb7 JH |
1047 | int a; |
1048 | ||
1049 | a = virtio_queue_get_desc_addr(vdev, idx); | |
1050 | if (a == 0) { | |
1051 | /* Don't stop the virtqueue which might have not been started */ | |
1052 | return; | |
1053 | } | |
fc57fd99 | 1054 | |
21e70425 | 1055 | r = dev->vhost_ops->vhost_get_vring_base(dev, &state); |
d5970055 | 1056 | if (r < 0) { |
c6409692 | 1057 | VHOST_OPS_DEBUG("vhost VQ %d ring restore failed: %d", idx, r); |
2ae39a11 MC |
1058 | /* Connection to the backend is broken, so let's sync internal |
1059 | * last avail idx to the device used idx. | |
1060 | */ | |
1061 | virtio_queue_restore_last_avail_idx(vdev, idx); | |
499c5579 MAL |
1062 | } else { |
1063 | virtio_queue_set_last_avail_idx(vdev, idx, state.num); | |
d5970055 | 1064 | } |
3561ba14 | 1065 | virtio_queue_invalidate_signalled_used(vdev, idx); |
aa94d521 | 1066 | virtio_queue_update_used_idx(vdev, idx); |
04b7a152 GK |
1067 | |
1068 | /* In the cross-endian case, we need to reset the vring endianness to | |
1069 | * native as legacy devices expect so by default. | |
1070 | */ | |
e5848123 | 1071 | if (vhost_needs_vring_endian(vdev)) { |
162bba7f MAL |
1072 | vhost_virtqueue_set_vring_endian_legacy(dev, |
1073 | !virtio_is_big_endian(vdev), | |
1074 | vhost_vq_index); | |
04b7a152 GK |
1075 | } |
1076 | ||
c471ad0e JW |
1077 | vhost_memory_unmap(dev, vq->used, virtio_queue_get_used_size(vdev, idx), |
1078 | 1, virtio_queue_get_used_size(vdev, idx)); | |
1079 | vhost_memory_unmap(dev, vq->avail, virtio_queue_get_avail_size(vdev, idx), | |
1080 | 0, virtio_queue_get_avail_size(vdev, idx)); | |
1081 | vhost_memory_unmap(dev, vq->desc, virtio_queue_get_desc_size(vdev, idx), | |
1082 | 0, virtio_queue_get_desc_size(vdev, idx)); | |
d5970055 MT |
1083 | } |
1084 | ||
80a1ea37 AK |
1085 | static void vhost_eventfd_add(MemoryListener *listener, |
1086 | MemoryRegionSection *section, | |
753d5e14 | 1087 | bool match_data, uint64_t data, EventNotifier *e) |
80a1ea37 AK |
1088 | { |
1089 | } | |
1090 | ||
1091 | static void vhost_eventfd_del(MemoryListener *listener, | |
1092 | MemoryRegionSection *section, | |
753d5e14 | 1093 | bool match_data, uint64_t data, EventNotifier *e) |
80a1ea37 AK |
1094 | { |
1095 | } | |
1096 | ||
69e87b32 JW |
1097 | static int vhost_virtqueue_set_busyloop_timeout(struct vhost_dev *dev, |
1098 | int n, uint32_t timeout) | |
1099 | { | |
1100 | int vhost_vq_index = dev->vhost_ops->vhost_get_vq_index(dev, n); | |
1101 | struct vhost_vring_state state = { | |
1102 | .index = vhost_vq_index, | |
1103 | .num = timeout, | |
1104 | }; | |
1105 | int r; | |
1106 | ||
1107 | if (!dev->vhost_ops->vhost_set_vring_busyloop_timeout) { | |
1108 | return -EINVAL; | |
1109 | } | |
1110 | ||
1111 | r = dev->vhost_ops->vhost_set_vring_busyloop_timeout(dev, &state); | |
1112 | if (r) { | |
c6409692 | 1113 | VHOST_OPS_DEBUG("vhost_set_vring_busyloop_timeout failed"); |
69e87b32 JW |
1114 | return r; |
1115 | } | |
1116 | ||
1117 | return 0; | |
1118 | } | |
1119 | ||
f56a1247 MT |
1120 | static int vhost_virtqueue_init(struct vhost_dev *dev, |
1121 | struct vhost_virtqueue *vq, int n) | |
1122 | { | |
21e70425 | 1123 | int vhost_vq_index = dev->vhost_ops->vhost_get_vq_index(dev, n); |
f56a1247 | 1124 | struct vhost_vring_file file = { |
b931bfbf | 1125 | .index = vhost_vq_index, |
f56a1247 MT |
1126 | }; |
1127 | int r = event_notifier_init(&vq->masked_notifier, 0); | |
1128 | if (r < 0) { | |
1129 | return r; | |
1130 | } | |
1131 | ||
1132 | file.fd = event_notifier_get_fd(&vq->masked_notifier); | |
21e70425 | 1133 | r = dev->vhost_ops->vhost_set_vring_call(dev, &file); |
f56a1247 | 1134 | if (r) { |
c6409692 | 1135 | VHOST_OPS_DEBUG("vhost_set_vring_call failed"); |
f56a1247 MT |
1136 | r = -errno; |
1137 | goto fail_call; | |
1138 | } | |
c471ad0e JW |
1139 | |
1140 | vq->dev = dev; | |
1141 | ||
f56a1247 MT |
1142 | return 0; |
1143 | fail_call: | |
1144 | event_notifier_cleanup(&vq->masked_notifier); | |
1145 | return r; | |
1146 | } | |
1147 | ||
1148 | static void vhost_virtqueue_cleanup(struct vhost_virtqueue *vq) | |
1149 | { | |
1150 | event_notifier_cleanup(&vq->masked_notifier); | |
1151 | } | |
1152 | ||
81647a65 | 1153 | int vhost_dev_init(struct vhost_dev *hdev, void *opaque, |
69e87b32 | 1154 | VhostBackendType backend_type, uint32_t busyloop_timeout) |
d5970055 MT |
1155 | { |
1156 | uint64_t features; | |
a06db3ec | 1157 | int i, r, n_initialized_vqs = 0; |
fe44dc91 | 1158 | Error *local_err = NULL; |
81647a65 | 1159 | |
c471ad0e | 1160 | hdev->vdev = NULL; |
d2fc4402 MAL |
1161 | hdev->migration_blocker = NULL; |
1162 | ||
7cb8a9b9 MAL |
1163 | r = vhost_set_backend_type(hdev, backend_type); |
1164 | assert(r >= 0); | |
1a1bfac9 | 1165 | |
7cb8a9b9 MAL |
1166 | r = hdev->vhost_ops->vhost_backend_init(hdev, opaque); |
1167 | if (r < 0) { | |
1168 | goto fail; | |
24d1eb33 NN |
1169 | } |
1170 | ||
21e70425 | 1171 | r = hdev->vhost_ops->vhost_set_owner(hdev); |
d5970055 | 1172 | if (r < 0) { |
c6409692 | 1173 | VHOST_OPS_DEBUG("vhost_set_owner failed"); |
d5970055 MT |
1174 | goto fail; |
1175 | } | |
1176 | ||
21e70425 | 1177 | r = hdev->vhost_ops->vhost_get_features(hdev, &features); |
d5970055 | 1178 | if (r < 0) { |
c6409692 | 1179 | VHOST_OPS_DEBUG("vhost_get_features failed"); |
d5970055 MT |
1180 | goto fail; |
1181 | } | |
f56a1247 | 1182 | |
a06db3ec | 1183 | for (i = 0; i < hdev->nvqs; ++i, ++n_initialized_vqs) { |
b931bfbf | 1184 | r = vhost_virtqueue_init(hdev, hdev->vqs + i, hdev->vq_index + i); |
f56a1247 | 1185 | if (r < 0) { |
a06db3ec | 1186 | goto fail; |
f56a1247 MT |
1187 | } |
1188 | } | |
69e87b32 JW |
1189 | |
1190 | if (busyloop_timeout) { | |
1191 | for (i = 0; i < hdev->nvqs; ++i) { | |
1192 | r = vhost_virtqueue_set_busyloop_timeout(hdev, hdev->vq_index + i, | |
1193 | busyloop_timeout); | |
1194 | if (r < 0) { | |
1195 | goto fail_busyloop; | |
1196 | } | |
1197 | } | |
1198 | } | |
1199 | ||
d5970055 MT |
1200 | hdev->features = features; |
1201 | ||
04097f7c | 1202 | hdev->memory_listener = (MemoryListener) { |
50c1e149 AK |
1203 | .begin = vhost_begin, |
1204 | .commit = vhost_commit, | |
938eeb64 DDAG |
1205 | .region_add = vhost_region_addnop, |
1206 | .region_nop = vhost_region_addnop, | |
04097f7c AK |
1207 | .log_start = vhost_log_start, |
1208 | .log_stop = vhost_log_stop, | |
1209 | .log_sync = vhost_log_sync, | |
1210 | .log_global_start = vhost_log_global_start, | |
1211 | .log_global_stop = vhost_log_global_stop, | |
80a1ea37 AK |
1212 | .eventfd_add = vhost_eventfd_add, |
1213 | .eventfd_del = vhost_eventfd_del, | |
72e22d2f | 1214 | .priority = 10 |
04097f7c | 1215 | }; |
d2fc4402 | 1216 | |
375f74f4 JW |
1217 | hdev->iommu_listener = (MemoryListener) { |
1218 | .region_add = vhost_iommu_region_add, | |
1219 | .region_del = vhost_iommu_region_del, | |
1220 | }; | |
c471ad0e | 1221 | |
d2fc4402 MAL |
1222 | if (hdev->migration_blocker == NULL) { |
1223 | if (!(hdev->features & (0x1ULL << VHOST_F_LOG_ALL))) { | |
1224 | error_setg(&hdev->migration_blocker, | |
1225 | "Migration disabled: vhost lacks VHOST_F_LOG_ALL feature."); | |
0d34fbab | 1226 | } else if (vhost_dev_log_is_shared(hdev) && !qemu_memfd_check()) { |
31190ed7 MAL |
1227 | error_setg(&hdev->migration_blocker, |
1228 | "Migration disabled: failed to allocate shared memory"); | |
d2fc4402 MAL |
1229 | } |
1230 | } | |
1231 | ||
1232 | if (hdev->migration_blocker != NULL) { | |
fe44dc91 AA |
1233 | r = migrate_add_blocker(hdev->migration_blocker, &local_err); |
1234 | if (local_err) { | |
1235 | error_report_err(local_err); | |
1236 | error_free(hdev->migration_blocker); | |
1237 | goto fail_busyloop; | |
1238 | } | |
7145872e | 1239 | } |
d2fc4402 | 1240 | |
7267c094 | 1241 | hdev->mem = g_malloc0(offsetof(struct vhost_memory, regions)); |
2817b260 AK |
1242 | hdev->n_mem_sections = 0; |
1243 | hdev->mem_sections = NULL; | |
d5970055 MT |
1244 | hdev->log = NULL; |
1245 | hdev->log_size = 0; | |
1246 | hdev->log_enabled = false; | |
1247 | hdev->started = false; | |
f6790af6 | 1248 | memory_listener_register(&hdev->memory_listener, &address_space_memory); |
5be5f9be | 1249 | QLIST_INSERT_HEAD(&vhost_devices, hdev, entry); |
9e2a2a3e JZ |
1250 | |
1251 | if (used_memslots > hdev->vhost_ops->vhost_backend_memslots_limit(hdev)) { | |
1252 | error_report("vhost backend memory slots limit is less" | |
1253 | " than current number of present memory slots"); | |
1254 | r = -1; | |
1255 | if (busyloop_timeout) { | |
1256 | goto fail_busyloop; | |
1257 | } else { | |
1258 | goto fail; | |
1259 | } | |
1260 | } | |
1261 | ||
d5970055 | 1262 | return 0; |
a06db3ec | 1263 | |
69e87b32 JW |
1264 | fail_busyloop: |
1265 | while (--i >= 0) { | |
1266 | vhost_virtqueue_set_busyloop_timeout(hdev, hdev->vq_index + i, 0); | |
1267 | } | |
d5970055 | 1268 | fail: |
a06db3ec MAL |
1269 | hdev->nvqs = n_initialized_vqs; |
1270 | vhost_dev_cleanup(hdev); | |
d5970055 MT |
1271 | return r; |
1272 | } | |
1273 | ||
1274 | void vhost_dev_cleanup(struct vhost_dev *hdev) | |
1275 | { | |
f56a1247 | 1276 | int i; |
e0547b59 | 1277 | |
f56a1247 MT |
1278 | for (i = 0; i < hdev->nvqs; ++i) { |
1279 | vhost_virtqueue_cleanup(hdev->vqs + i); | |
1280 | } | |
5be5f9be MAL |
1281 | if (hdev->mem) { |
1282 | /* those are only safe after successful init */ | |
1283 | memory_listener_unregister(&hdev->memory_listener); | |
1284 | QLIST_REMOVE(hdev, entry); | |
1285 | } | |
7145872e MT |
1286 | if (hdev->migration_blocker) { |
1287 | migrate_del_blocker(hdev->migration_blocker); | |
1288 | error_free(hdev->migration_blocker); | |
1289 | } | |
7267c094 | 1290 | g_free(hdev->mem); |
2817b260 | 1291 | g_free(hdev->mem_sections); |
e0547b59 MAL |
1292 | if (hdev->vhost_ops) { |
1293 | hdev->vhost_ops->vhost_backend_cleanup(hdev); | |
1294 | } | |
7b527247 | 1295 | assert(!hdev->log); |
e0547b59 MAL |
1296 | |
1297 | memset(hdev, 0, sizeof(struct vhost_dev)); | |
d5970055 MT |
1298 | } |
1299 | ||
b0b3db79 MT |
1300 | /* Stop processing guest IO notifications in qemu. |
1301 | * Start processing them in vhost in kernel. | |
1302 | */ | |
1303 | int vhost_dev_enable_notifiers(struct vhost_dev *hdev, VirtIODevice *vdev) | |
1304 | { | |
1c819449 | 1305 | BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(vdev))); |
16617e36 | 1306 | int i, r, e; |
4afba631 | 1307 | |
310837de PB |
1308 | /* We will pass the notifiers to the kernel, make sure that QEMU |
1309 | * doesn't interfere. | |
1310 | */ | |
1311 | r = virtio_device_grab_ioeventfd(vdev); | |
1312 | if (r < 0) { | |
4afba631 | 1313 | error_report("binding does not support host notifiers"); |
b0b3db79 MT |
1314 | goto fail; |
1315 | } | |
1316 | ||
1317 | for (i = 0; i < hdev->nvqs; ++i) { | |
b1f0a33d CH |
1318 | r = virtio_bus_set_host_notifier(VIRTIO_BUS(qbus), hdev->vq_index + i, |
1319 | true); | |
b0b3db79 | 1320 | if (r < 0) { |
4afba631 | 1321 | error_report("vhost VQ %d notifier binding failed: %d", i, -r); |
b0b3db79 MT |
1322 | goto fail_vq; |
1323 | } | |
1324 | } | |
1325 | ||
1326 | return 0; | |
1327 | fail_vq: | |
1328 | while (--i >= 0) { | |
b1f0a33d CH |
1329 | e = virtio_bus_set_host_notifier(VIRTIO_BUS(qbus), hdev->vq_index + i, |
1330 | false); | |
16617e36 | 1331 | if (e < 0) { |
4afba631 | 1332 | error_report("vhost VQ %d notifier cleanup error: %d", i, -r); |
b0b3db79 | 1333 | } |
16617e36 | 1334 | assert (e >= 0); |
76143618 | 1335 | virtio_bus_cleanup_host_notifier(VIRTIO_BUS(qbus), hdev->vq_index + i); |
b0b3db79 | 1336 | } |
310837de | 1337 | virtio_device_release_ioeventfd(vdev); |
b0b3db79 MT |
1338 | fail: |
1339 | return r; | |
1340 | } | |
1341 | ||
1342 | /* Stop processing guest IO notifications in vhost. | |
1343 | * Start processing them in qemu. | |
1344 | * This might actually run the qemu handlers right away, | |
1345 | * so virtio in qemu must be completely setup when this is called. | |
1346 | */ | |
1347 | void vhost_dev_disable_notifiers(struct vhost_dev *hdev, VirtIODevice *vdev) | |
1348 | { | |
1c819449 | 1349 | BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(vdev))); |
b0b3db79 MT |
1350 | int i, r; |
1351 | ||
1352 | for (i = 0; i < hdev->nvqs; ++i) { | |
b1f0a33d CH |
1353 | r = virtio_bus_set_host_notifier(VIRTIO_BUS(qbus), hdev->vq_index + i, |
1354 | false); | |
b0b3db79 | 1355 | if (r < 0) { |
4afba631 | 1356 | error_report("vhost VQ %d notifier cleanup failed: %d", i, -r); |
b0b3db79 MT |
1357 | } |
1358 | assert (r >= 0); | |
76143618 | 1359 | virtio_bus_cleanup_host_notifier(VIRTIO_BUS(qbus), hdev->vq_index + i); |
b0b3db79 | 1360 | } |
310837de | 1361 | virtio_device_release_ioeventfd(vdev); |
b0b3db79 MT |
1362 | } |
1363 | ||
f56a1247 MT |
1364 | /* Test and clear event pending status. |
1365 | * Should be called after unmask to avoid losing events. | |
1366 | */ | |
1367 | bool vhost_virtqueue_pending(struct vhost_dev *hdev, int n) | |
1368 | { | |
a9f98bb5 | 1369 | struct vhost_virtqueue *vq = hdev->vqs + n - hdev->vq_index; |
a9f98bb5 | 1370 | assert(n >= hdev->vq_index && n < hdev->vq_index + hdev->nvqs); |
f56a1247 MT |
1371 | return event_notifier_test_and_clear(&vq->masked_notifier); |
1372 | } | |
1373 | ||
1374 | /* Mask/unmask events from this vq. */ | |
1375 | void vhost_virtqueue_mask(struct vhost_dev *hdev, VirtIODevice *vdev, int n, | |
1376 | bool mask) | |
1377 | { | |
1378 | struct VirtQueue *vvq = virtio_get_queue(vdev, n); | |
a9f98bb5 | 1379 | int r, index = n - hdev->vq_index; |
fc57fd99 | 1380 | struct vhost_vring_file file; |
f56a1247 | 1381 | |
8695de0f MAL |
1382 | /* should only be called after backend is connected */ |
1383 | assert(hdev->vhost_ops); | |
1384 | ||
f56a1247 | 1385 | if (mask) { |
5669655a | 1386 | assert(vdev->use_guest_notifier_mask); |
a9f98bb5 | 1387 | file.fd = event_notifier_get_fd(&hdev->vqs[index].masked_notifier); |
f56a1247 MT |
1388 | } else { |
1389 | file.fd = event_notifier_get_fd(virtio_queue_get_guest_notifier(vvq)); | |
1390 | } | |
fc57fd99 | 1391 | |
21e70425 MAL |
1392 | file.index = hdev->vhost_ops->vhost_get_vq_index(hdev, n); |
1393 | r = hdev->vhost_ops->vhost_set_vring_call(hdev, &file); | |
162bba7f MAL |
1394 | if (r < 0) { |
1395 | VHOST_OPS_DEBUG("vhost_set_vring_call failed"); | |
1396 | } | |
f56a1247 MT |
1397 | } |
1398 | ||
9a2ba823 CH |
1399 | uint64_t vhost_get_features(struct vhost_dev *hdev, const int *feature_bits, |
1400 | uint64_t features) | |
2e6d46d7 NN |
1401 | { |
1402 | const int *bit = feature_bits; | |
1403 | while (*bit != VHOST_INVALID_FEATURE_BIT) { | |
9a2ba823 | 1404 | uint64_t bit_mask = (1ULL << *bit); |
2e6d46d7 NN |
1405 | if (!(hdev->features & bit_mask)) { |
1406 | features &= ~bit_mask; | |
1407 | } | |
1408 | bit++; | |
1409 | } | |
1410 | return features; | |
1411 | } | |
1412 | ||
1413 | void vhost_ack_features(struct vhost_dev *hdev, const int *feature_bits, | |
9a2ba823 | 1414 | uint64_t features) |
2e6d46d7 NN |
1415 | { |
1416 | const int *bit = feature_bits; | |
1417 | while (*bit != VHOST_INVALID_FEATURE_BIT) { | |
9a2ba823 | 1418 | uint64_t bit_mask = (1ULL << *bit); |
2e6d46d7 NN |
1419 | if (features & bit_mask) { |
1420 | hdev->acked_features |= bit_mask; | |
1421 | } | |
1422 | bit++; | |
1423 | } | |
1424 | } | |
1425 | ||
4c3e257b CL |
1426 | int vhost_dev_get_config(struct vhost_dev *hdev, uint8_t *config, |
1427 | uint32_t config_len) | |
1428 | { | |
1429 | assert(hdev->vhost_ops); | |
1430 | ||
1431 | if (hdev->vhost_ops->vhost_get_config) { | |
1432 | return hdev->vhost_ops->vhost_get_config(hdev, config, config_len); | |
1433 | } | |
1434 | ||
1435 | return -1; | |
1436 | } | |
1437 | ||
1438 | int vhost_dev_set_config(struct vhost_dev *hdev, const uint8_t *data, | |
1439 | uint32_t offset, uint32_t size, uint32_t flags) | |
1440 | { | |
1441 | assert(hdev->vhost_ops); | |
1442 | ||
1443 | if (hdev->vhost_ops->vhost_set_config) { | |
1444 | return hdev->vhost_ops->vhost_set_config(hdev, data, offset, | |
1445 | size, flags); | |
1446 | } | |
1447 | ||
1448 | return -1; | |
1449 | } | |
1450 | ||
1451 | void vhost_dev_set_config_notifier(struct vhost_dev *hdev, | |
1452 | const VhostDevConfigOps *ops) | |
1453 | { | |
4c3e257b CL |
1454 | hdev->config_ops = ops; |
1455 | } | |
1456 | ||
b0b3db79 | 1457 | /* Host notifiers must be enabled at this point. */ |
d5970055 MT |
1458 | int vhost_dev_start(struct vhost_dev *hdev, VirtIODevice *vdev) |
1459 | { | |
1460 | int i, r; | |
24f4fe34 | 1461 | |
8695de0f MAL |
1462 | /* should only be called after backend is connected */ |
1463 | assert(hdev->vhost_ops); | |
1464 | ||
24f4fe34 | 1465 | hdev->started = true; |
c471ad0e | 1466 | hdev->vdev = vdev; |
24f4fe34 | 1467 | |
d5970055 MT |
1468 | r = vhost_dev_set_features(hdev, hdev->log_enabled); |
1469 | if (r < 0) { | |
54dd9321 | 1470 | goto fail_features; |
d5970055 | 1471 | } |
c471ad0e JW |
1472 | |
1473 | if (vhost_dev_has_iommu(hdev)) { | |
375f74f4 | 1474 | memory_listener_register(&hdev->iommu_listener, vdev->dma_as); |
c471ad0e JW |
1475 | } |
1476 | ||
21e70425 | 1477 | r = hdev->vhost_ops->vhost_set_mem_table(hdev, hdev->mem); |
d5970055 | 1478 | if (r < 0) { |
c6409692 | 1479 | VHOST_OPS_DEBUG("vhost_set_mem_table failed"); |
d5970055 | 1480 | r = -errno; |
54dd9321 | 1481 | goto fail_mem; |
d5970055 | 1482 | } |
d154e0ba | 1483 | for (i = 0; i < hdev->nvqs; ++i) { |
f56a1247 | 1484 | r = vhost_virtqueue_start(hdev, |
a9f98bb5 JW |
1485 | vdev, |
1486 | hdev->vqs + i, | |
1487 | hdev->vq_index + i); | |
d154e0ba MT |
1488 | if (r < 0) { |
1489 | goto fail_vq; | |
1490 | } | |
1491 | } | |
1492 | ||
d5970055 | 1493 | if (hdev->log_enabled) { |
e05ca820 MT |
1494 | uint64_t log_base; |
1495 | ||
d5970055 | 1496 | hdev->log_size = vhost_get_log_size(hdev); |
15324404 MAL |
1497 | hdev->log = vhost_log_get(hdev->log_size, |
1498 | vhost_dev_log_is_shared(hdev)); | |
309750fa | 1499 | log_base = (uintptr_t)hdev->log->log; |
c2bea314 | 1500 | r = hdev->vhost_ops->vhost_set_log_base(hdev, |
9a78a5dd MAL |
1501 | hdev->log_size ? log_base : 0, |
1502 | hdev->log); | |
d5970055 | 1503 | if (r < 0) { |
c6409692 | 1504 | VHOST_OPS_DEBUG("vhost_set_log_base failed"); |
d5970055 | 1505 | r = -errno; |
54dd9321 | 1506 | goto fail_log; |
d5970055 MT |
1507 | } |
1508 | } | |
d154e0ba | 1509 | |
c471ad0e JW |
1510 | if (vhost_dev_has_iommu(hdev)) { |
1511 | hdev->vhost_ops->vhost_set_iotlb_callback(hdev, true); | |
1512 | ||
1513 | /* Update used ring information for IOTLB to work correctly, | |
1514 | * vhost-kernel code requires for this.*/ | |
1515 | for (i = 0; i < hdev->nvqs; ++i) { | |
1516 | struct vhost_virtqueue *vq = hdev->vqs + i; | |
1517 | vhost_device_iotlb_miss(hdev, vq->used_phys, true); | |
1518 | } | |
1519 | } | |
d5970055 | 1520 | return 0; |
54dd9321 | 1521 | fail_log: |
24bfa207 | 1522 | vhost_log_put(hdev, false); |
d5970055 MT |
1523 | fail_vq: |
1524 | while (--i >= 0) { | |
f56a1247 | 1525 | vhost_virtqueue_stop(hdev, |
a9f98bb5 JW |
1526 | vdev, |
1527 | hdev->vqs + i, | |
1528 | hdev->vq_index + i); | |
d5970055 | 1529 | } |
a9f98bb5 | 1530 | i = hdev->nvqs; |
c471ad0e | 1531 | |
54dd9321 MT |
1532 | fail_mem: |
1533 | fail_features: | |
24f4fe34 MT |
1534 | |
1535 | hdev->started = false; | |
d5970055 MT |
1536 | return r; |
1537 | } | |
1538 | ||
b0b3db79 | 1539 | /* Host notifiers must be enabled at this point. */ |
d5970055 MT |
1540 | void vhost_dev_stop(struct vhost_dev *hdev, VirtIODevice *vdev) |
1541 | { | |
a9f98bb5 | 1542 | int i; |
54dd9321 | 1543 | |
8695de0f MAL |
1544 | /* should only be called after backend is connected */ |
1545 | assert(hdev->vhost_ops); | |
1546 | ||
d5970055 | 1547 | for (i = 0; i < hdev->nvqs; ++i) { |
f56a1247 | 1548 | vhost_virtqueue_stop(hdev, |
a9f98bb5 JW |
1549 | vdev, |
1550 | hdev->vqs + i, | |
1551 | hdev->vq_index + i); | |
d5970055 | 1552 | } |
54dd9321 | 1553 | |
c471ad0e JW |
1554 | if (vhost_dev_has_iommu(hdev)) { |
1555 | hdev->vhost_ops->vhost_set_iotlb_callback(hdev, false); | |
375f74f4 | 1556 | memory_listener_unregister(&hdev->iommu_listener); |
c471ad0e | 1557 | } |
309750fa | 1558 | vhost_log_put(hdev, true); |
d5970055 | 1559 | hdev->started = false; |
c471ad0e | 1560 | hdev->vdev = NULL; |
d5970055 | 1561 | } |
950d94ba MAL |
1562 | |
1563 | int vhost_net_set_backend(struct vhost_dev *hdev, | |
1564 | struct vhost_vring_file *file) | |
1565 | { | |
1566 | if (hdev->vhost_ops->vhost_net_set_backend) { | |
1567 | return hdev->vhost_ops->vhost_net_set_backend(hdev, file); | |
1568 | } | |
1569 | ||
1570 | return -1; | |
1571 | } |