]>
Commit | Line | Data |
---|---|---|
d5970055 MT |
1 | /* |
2 | * vhost support | |
3 | * | |
4 | * Copyright Red Hat, Inc. 2010 | |
5 | * | |
6 | * Authors: | |
7 | * Michael S. Tsirkin <[email protected]> | |
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 | ||
d5970055 | 16 | #include <sys/ioctl.h> |
d5970055 MT |
17 | #include "vhost.h" |
18 | #include "hw/hw.h" | |
bf1b0071 | 19 | #include "range.h" |
11078ae3 | 20 | #include <linux/vhost.h> |
c49450b9 | 21 | #include "exec-memory.h" |
d5970055 MT |
22 | |
23 | static void vhost_dev_sync_region(struct vhost_dev *dev, | |
2817b260 | 24 | MemoryRegionSection *section, |
d5970055 MT |
25 | uint64_t mfirst, uint64_t mlast, |
26 | uint64_t rfirst, uint64_t rlast) | |
27 | { | |
28 | uint64_t start = MAX(mfirst, rfirst); | |
29 | uint64_t end = MIN(mlast, rlast); | |
30 | vhost_log_chunk_t *from = dev->log + start / VHOST_LOG_CHUNK; | |
31 | vhost_log_chunk_t *to = dev->log + end / VHOST_LOG_CHUNK + 1; | |
32 | uint64_t addr = (start / VHOST_LOG_CHUNK) * VHOST_LOG_CHUNK; | |
33 | ||
d5970055 MT |
34 | if (end < start) { |
35 | return; | |
36 | } | |
e314672a | 37 | assert(end / VHOST_LOG_CHUNK < dev->log_size); |
fbbaf9ae | 38 | assert(start / VHOST_LOG_CHUNK < dev->log_size); |
e314672a | 39 | |
d5970055 MT |
40 | for (;from < to; ++from) { |
41 | vhost_log_chunk_t log; | |
42 | int bit; | |
43 | /* We first check with non-atomic: much cheaper, | |
44 | * and we expect non-dirty to be the common case. */ | |
45 | if (!*from) { | |
0c600ce2 | 46 | addr += VHOST_LOG_CHUNK; |
d5970055 MT |
47 | continue; |
48 | } | |
49 | /* Data must be read atomically. We don't really | |
50 | * need the barrier semantics of __sync | |
51 | * builtins, but it's easier to use them than | |
52 | * roll our own. */ | |
53 | log = __sync_fetch_and_and(from, 0); | |
54 | while ((bit = sizeof(log) > sizeof(int) ? | |
55 | ffsll(log) : ffs(log))) { | |
7b67b18a | 56 | ram_addr_t ram_addr; |
d5970055 | 57 | bit -= 1; |
2817b260 | 58 | ram_addr = section->offset_within_region + bit * VHOST_LOG_PAGE; |
fd4aa979 | 59 | memory_region_set_dirty(section->mr, ram_addr, VHOST_LOG_PAGE); |
d5970055 MT |
60 | log &= ~(0x1ull << bit); |
61 | } | |
62 | addr += VHOST_LOG_CHUNK; | |
63 | } | |
64 | } | |
65 | ||
04097f7c | 66 | static int vhost_sync_dirty_bitmap(struct vhost_dev *dev, |
2817b260 | 67 | MemoryRegionSection *section, |
a8170e5e AK |
68 | hwaddr start_addr, |
69 | hwaddr end_addr) | |
d5970055 | 70 | { |
d5970055 | 71 | int i; |
04097f7c | 72 | |
d5970055 MT |
73 | if (!dev->log_enabled || !dev->started) { |
74 | return 0; | |
75 | } | |
76 | for (i = 0; i < dev->mem->nregions; ++i) { | |
77 | struct vhost_memory_region *reg = dev->mem->regions + i; | |
2817b260 | 78 | vhost_dev_sync_region(dev, section, start_addr, end_addr, |
d5970055 MT |
79 | reg->guest_phys_addr, |
80 | range_get_last(reg->guest_phys_addr, | |
81 | reg->memory_size)); | |
82 | } | |
83 | for (i = 0; i < dev->nvqs; ++i) { | |
84 | struct vhost_virtqueue *vq = dev->vqs + i; | |
2817b260 | 85 | vhost_dev_sync_region(dev, section, start_addr, end_addr, vq->used_phys, |
d5970055 MT |
86 | range_get_last(vq->used_phys, vq->used_size)); |
87 | } | |
88 | return 0; | |
89 | } | |
90 | ||
04097f7c AK |
91 | static void vhost_log_sync(MemoryListener *listener, |
92 | MemoryRegionSection *section) | |
93 | { | |
94 | struct vhost_dev *dev = container_of(listener, struct vhost_dev, | |
95 | memory_listener); | |
a8170e5e AK |
96 | hwaddr start_addr = section->offset_within_address_space; |
97 | hwaddr end_addr = start_addr + section->size; | |
04097f7c | 98 | |
2817b260 | 99 | vhost_sync_dirty_bitmap(dev, section, start_addr, end_addr); |
04097f7c AK |
100 | } |
101 | ||
d5970055 MT |
102 | /* Assign/unassign. Keep an unsorted array of non-overlapping |
103 | * memory regions in dev->mem. */ | |
104 | static void vhost_dev_unassign_memory(struct vhost_dev *dev, | |
105 | uint64_t start_addr, | |
106 | uint64_t size) | |
107 | { | |
108 | int from, to, n = dev->mem->nregions; | |
109 | /* Track overlapping/split regions for sanity checking. */ | |
110 | int overlap_start = 0, overlap_end = 0, overlap_middle = 0, split = 0; | |
111 | ||
112 | for (from = 0, to = 0; from < n; ++from, ++to) { | |
113 | struct vhost_memory_region *reg = dev->mem->regions + to; | |
114 | uint64_t reglast; | |
115 | uint64_t memlast; | |
116 | uint64_t change; | |
117 | ||
118 | /* clone old region */ | |
119 | if (to != from) { | |
120 | memcpy(reg, dev->mem->regions + from, sizeof *reg); | |
121 | } | |
122 | ||
123 | /* No overlap is simple */ | |
124 | if (!ranges_overlap(reg->guest_phys_addr, reg->memory_size, | |
125 | start_addr, size)) { | |
126 | continue; | |
127 | } | |
128 | ||
129 | /* Split only happens if supplied region | |
130 | * is in the middle of an existing one. Thus it can not | |
131 | * overlap with any other existing region. */ | |
132 | assert(!split); | |
133 | ||
134 | reglast = range_get_last(reg->guest_phys_addr, reg->memory_size); | |
135 | memlast = range_get_last(start_addr, size); | |
136 | ||
137 | /* Remove whole region */ | |
138 | if (start_addr <= reg->guest_phys_addr && memlast >= reglast) { | |
139 | --dev->mem->nregions; | |
140 | --to; | |
d5970055 MT |
141 | ++overlap_middle; |
142 | continue; | |
143 | } | |
144 | ||
145 | /* Shrink region */ | |
146 | if (memlast >= reglast) { | |
147 | reg->memory_size = start_addr - reg->guest_phys_addr; | |
148 | assert(reg->memory_size); | |
149 | assert(!overlap_end); | |
150 | ++overlap_end; | |
151 | continue; | |
152 | } | |
153 | ||
154 | /* Shift region */ | |
155 | if (start_addr <= reg->guest_phys_addr) { | |
156 | change = memlast + 1 - reg->guest_phys_addr; | |
157 | reg->memory_size -= change; | |
158 | reg->guest_phys_addr += change; | |
159 | reg->userspace_addr += change; | |
160 | assert(reg->memory_size); | |
161 | assert(!overlap_start); | |
162 | ++overlap_start; | |
163 | continue; | |
164 | } | |
165 | ||
166 | /* This only happens if supplied region | |
167 | * is in the middle of an existing one. Thus it can not | |
168 | * overlap with any other existing region. */ | |
169 | assert(!overlap_start); | |
170 | assert(!overlap_end); | |
171 | assert(!overlap_middle); | |
172 | /* Split region: shrink first part, shift second part. */ | |
173 | memcpy(dev->mem->regions + n, reg, sizeof *reg); | |
174 | reg->memory_size = start_addr - reg->guest_phys_addr; | |
175 | assert(reg->memory_size); | |
176 | change = memlast + 1 - reg->guest_phys_addr; | |
177 | reg = dev->mem->regions + n; | |
178 | reg->memory_size -= change; | |
179 | assert(reg->memory_size); | |
180 | reg->guest_phys_addr += change; | |
181 | reg->userspace_addr += change; | |
182 | /* Never add more than 1 region */ | |
183 | assert(dev->mem->nregions == n); | |
184 | ++dev->mem->nregions; | |
185 | ++split; | |
186 | } | |
187 | } | |
188 | ||
189 | /* Called after unassign, so no regions overlap the given range. */ | |
190 | static void vhost_dev_assign_memory(struct vhost_dev *dev, | |
191 | uint64_t start_addr, | |
192 | uint64_t size, | |
193 | uint64_t uaddr) | |
194 | { | |
195 | int from, to; | |
196 | struct vhost_memory_region *merged = NULL; | |
197 | for (from = 0, to = 0; from < dev->mem->nregions; ++from, ++to) { | |
198 | struct vhost_memory_region *reg = dev->mem->regions + to; | |
199 | uint64_t prlast, urlast; | |
200 | uint64_t pmlast, umlast; | |
201 | uint64_t s, e, u; | |
202 | ||
203 | /* clone old region */ | |
204 | if (to != from) { | |
205 | memcpy(reg, dev->mem->regions + from, sizeof *reg); | |
206 | } | |
207 | prlast = range_get_last(reg->guest_phys_addr, reg->memory_size); | |
208 | pmlast = range_get_last(start_addr, size); | |
209 | urlast = range_get_last(reg->userspace_addr, reg->memory_size); | |
210 | umlast = range_get_last(uaddr, size); | |
211 | ||
212 | /* check for overlapping regions: should never happen. */ | |
213 | assert(prlast < start_addr || pmlast < reg->guest_phys_addr); | |
214 | /* Not an adjacent or overlapping region - do not merge. */ | |
215 | if ((prlast + 1 != start_addr || urlast + 1 != uaddr) && | |
216 | (pmlast + 1 != reg->guest_phys_addr || | |
217 | umlast + 1 != reg->userspace_addr)) { | |
218 | continue; | |
219 | } | |
220 | ||
221 | if (merged) { | |
222 | --to; | |
223 | assert(to >= 0); | |
224 | } else { | |
225 | merged = reg; | |
226 | } | |
227 | u = MIN(uaddr, reg->userspace_addr); | |
228 | s = MIN(start_addr, reg->guest_phys_addr); | |
229 | e = MAX(pmlast, prlast); | |
230 | uaddr = merged->userspace_addr = u; | |
231 | start_addr = merged->guest_phys_addr = s; | |
232 | size = merged->memory_size = e - s + 1; | |
233 | assert(merged->memory_size); | |
234 | } | |
235 | ||
236 | if (!merged) { | |
237 | struct vhost_memory_region *reg = dev->mem->regions + to; | |
238 | memset(reg, 0, sizeof *reg); | |
239 | reg->memory_size = size; | |
240 | assert(reg->memory_size); | |
241 | reg->guest_phys_addr = start_addr; | |
242 | reg->userspace_addr = uaddr; | |
243 | ++to; | |
244 | } | |
245 | assert(to <= dev->mem->nregions + 1); | |
246 | dev->mem->nregions = to; | |
247 | } | |
248 | ||
249 | static uint64_t vhost_get_log_size(struct vhost_dev *dev) | |
250 | { | |
251 | uint64_t log_size = 0; | |
252 | int i; | |
253 | for (i = 0; i < dev->mem->nregions; ++i) { | |
254 | struct vhost_memory_region *reg = dev->mem->regions + i; | |
255 | uint64_t last = range_get_last(reg->guest_phys_addr, | |
256 | reg->memory_size); | |
257 | log_size = MAX(log_size, last / VHOST_LOG_CHUNK + 1); | |
258 | } | |
259 | for (i = 0; i < dev->nvqs; ++i) { | |
260 | struct vhost_virtqueue *vq = dev->vqs + i; | |
261 | uint64_t last = vq->used_phys + vq->used_size - 1; | |
262 | log_size = MAX(log_size, last / VHOST_LOG_CHUNK + 1); | |
263 | } | |
264 | return log_size; | |
265 | } | |
266 | ||
267 | static inline void vhost_dev_log_resize(struct vhost_dev* dev, uint64_t size) | |
268 | { | |
269 | vhost_log_chunk_t *log; | |
270 | uint64_t log_base; | |
2817b260 | 271 | int r, i; |
d5970055 | 272 | if (size) { |
7267c094 | 273 | log = g_malloc0(size * sizeof *log); |
d5970055 MT |
274 | } else { |
275 | log = NULL; | |
276 | } | |
277 | log_base = (uint64_t)(unsigned long)log; | |
278 | r = ioctl(dev->control, VHOST_SET_LOG_BASE, &log_base); | |
279 | assert(r >= 0); | |
2817b260 | 280 | for (i = 0; i < dev->n_mem_sections; ++i) { |
e314672a AW |
281 | /* Sync only the range covered by the old log */ |
282 | vhost_sync_dirty_bitmap(dev, &dev->mem_sections[i], 0, | |
283 | dev->log_size * VHOST_LOG_CHUNK - 1); | |
2817b260 | 284 | } |
d5970055 | 285 | if (dev->log) { |
7267c094 | 286 | g_free(dev->log); |
d5970055 MT |
287 | } |
288 | dev->log = log; | |
289 | dev->log_size = size; | |
290 | } | |
291 | ||
292 | static int vhost_verify_ring_mappings(struct vhost_dev *dev, | |
293 | uint64_t start_addr, | |
294 | uint64_t size) | |
295 | { | |
296 | int i; | |
297 | for (i = 0; i < dev->nvqs; ++i) { | |
298 | struct vhost_virtqueue *vq = dev->vqs + i; | |
a8170e5e | 299 | hwaddr l; |
d5970055 MT |
300 | void *p; |
301 | ||
302 | if (!ranges_overlap(start_addr, size, vq->ring_phys, vq->ring_size)) { | |
303 | continue; | |
304 | } | |
305 | l = vq->ring_size; | |
306 | p = cpu_physical_memory_map(vq->ring_phys, &l, 1); | |
307 | if (!p || l != vq->ring_size) { | |
308 | fprintf(stderr, "Unable to map ring buffer for ring %d\n", i); | |
309 | return -ENOMEM; | |
310 | } | |
311 | if (p != vq->ring) { | |
312 | fprintf(stderr, "Ring buffer relocated for ring %d\n", i); | |
313 | return -EBUSY; | |
314 | } | |
315 | cpu_physical_memory_unmap(p, l, 0, 0); | |
316 | } | |
317 | return 0; | |
318 | } | |
319 | ||
4e789564 MT |
320 | static struct vhost_memory_region *vhost_dev_find_reg(struct vhost_dev *dev, |
321 | uint64_t start_addr, | |
322 | uint64_t size) | |
323 | { | |
324 | int i, n = dev->mem->nregions; | |
325 | for (i = 0; i < n; ++i) { | |
326 | struct vhost_memory_region *reg = dev->mem->regions + i; | |
327 | if (ranges_overlap(reg->guest_phys_addr, reg->memory_size, | |
328 | start_addr, size)) { | |
329 | return reg; | |
330 | } | |
331 | } | |
332 | return NULL; | |
333 | } | |
334 | ||
335 | static bool vhost_dev_cmp_memory(struct vhost_dev *dev, | |
336 | uint64_t start_addr, | |
337 | uint64_t size, | |
338 | uint64_t uaddr) | |
339 | { | |
340 | struct vhost_memory_region *reg = vhost_dev_find_reg(dev, start_addr, size); | |
341 | uint64_t reglast; | |
342 | uint64_t memlast; | |
343 | ||
344 | if (!reg) { | |
345 | return true; | |
346 | } | |
347 | ||
348 | reglast = range_get_last(reg->guest_phys_addr, reg->memory_size); | |
349 | memlast = range_get_last(start_addr, size); | |
350 | ||
351 | /* Need to extend region? */ | |
352 | if (start_addr < reg->guest_phys_addr || memlast > reglast) { | |
353 | return true; | |
354 | } | |
355 | /* userspace_addr changed? */ | |
356 | return uaddr != reg->userspace_addr + start_addr - reg->guest_phys_addr; | |
357 | } | |
358 | ||
04097f7c AK |
359 | static void vhost_set_memory(MemoryListener *listener, |
360 | MemoryRegionSection *section, | |
361 | bool add) | |
d5970055 | 362 | { |
04097f7c AK |
363 | struct vhost_dev *dev = container_of(listener, struct vhost_dev, |
364 | memory_listener); | |
a8170e5e | 365 | hwaddr start_addr = section->offset_within_address_space; |
04097f7c AK |
366 | ram_addr_t size = section->size; |
367 | bool log_dirty = memory_region_is_logging(section->mr); | |
d5970055 MT |
368 | int s = offsetof(struct vhost_memory, regions) + |
369 | (dev->mem->nregions + 1) * sizeof dev->mem->regions[0]; | |
370 | uint64_t log_size; | |
371 | int r; | |
04097f7c AK |
372 | void *ram; |
373 | ||
7267c094 | 374 | dev->mem = g_realloc(dev->mem, s); |
d5970055 | 375 | |
f5a4e64f | 376 | if (log_dirty) { |
04097f7c | 377 | add = false; |
f5a4e64f MT |
378 | } |
379 | ||
d5970055 MT |
380 | assert(size); |
381 | ||
4e789564 | 382 | /* Optimize no-change case. At least cirrus_vga does this a lot at this time. */ |
d743c382 | 383 | ram = memory_region_get_ram_ptr(section->mr) + section->offset_within_region; |
04097f7c AK |
384 | if (add) { |
385 | if (!vhost_dev_cmp_memory(dev, start_addr, size, (uintptr_t)ram)) { | |
4e789564 MT |
386 | /* Region exists with same address. Nothing to do. */ |
387 | return; | |
388 | } | |
389 | } else { | |
390 | if (!vhost_dev_find_reg(dev, start_addr, size)) { | |
391 | /* Removing region that we don't access. Nothing to do. */ | |
392 | return; | |
393 | } | |
394 | } | |
395 | ||
d5970055 | 396 | vhost_dev_unassign_memory(dev, start_addr, size); |
04097f7c | 397 | if (add) { |
d5970055 | 398 | /* Add given mapping, merging adjacent regions if any */ |
04097f7c | 399 | vhost_dev_assign_memory(dev, start_addr, size, (uintptr_t)ram); |
d5970055 MT |
400 | } else { |
401 | /* Remove old mapping for this memory, if any. */ | |
402 | vhost_dev_unassign_memory(dev, start_addr, size); | |
403 | } | |
404 | ||
405 | if (!dev->started) { | |
406 | return; | |
407 | } | |
408 | ||
409 | if (dev->started) { | |
410 | r = vhost_verify_ring_mappings(dev, start_addr, size); | |
411 | assert(r >= 0); | |
412 | } | |
413 | ||
414 | if (!dev->log_enabled) { | |
415 | r = ioctl(dev->control, VHOST_SET_MEM_TABLE, dev->mem); | |
416 | assert(r >= 0); | |
417 | return; | |
418 | } | |
419 | log_size = vhost_get_log_size(dev); | |
420 | /* We allocate an extra 4K bytes to log, | |
421 | * to reduce the * number of reallocations. */ | |
422 | #define VHOST_LOG_BUFFER (0x1000 / sizeof *dev->log) | |
423 | /* To log more, must increase log size before table update. */ | |
424 | if (dev->log_size < log_size) { | |
425 | vhost_dev_log_resize(dev, log_size + VHOST_LOG_BUFFER); | |
426 | } | |
427 | r = ioctl(dev->control, VHOST_SET_MEM_TABLE, dev->mem); | |
428 | assert(r >= 0); | |
429 | /* To log less, can only decrease log size after table update. */ | |
430 | if (dev->log_size > log_size + VHOST_LOG_BUFFER) { | |
431 | vhost_dev_log_resize(dev, log_size); | |
432 | } | |
433 | } | |
434 | ||
c49450b9 AK |
435 | static bool vhost_section(MemoryRegionSection *section) |
436 | { | |
0fada674 | 437 | return memory_region_is_ram(section->mr); |
c49450b9 AK |
438 | } |
439 | ||
50c1e149 AK |
440 | static void vhost_begin(MemoryListener *listener) |
441 | { | |
442 | } | |
443 | ||
444 | static void vhost_commit(MemoryListener *listener) | |
445 | { | |
446 | } | |
447 | ||
04097f7c AK |
448 | static void vhost_region_add(MemoryListener *listener, |
449 | MemoryRegionSection *section) | |
450 | { | |
2817b260 AK |
451 | struct vhost_dev *dev = container_of(listener, struct vhost_dev, |
452 | memory_listener); | |
453 | ||
c49450b9 AK |
454 | if (!vhost_section(section)) { |
455 | return; | |
456 | } | |
457 | ||
2817b260 AK |
458 | ++dev->n_mem_sections; |
459 | dev->mem_sections = g_renew(MemoryRegionSection, dev->mem_sections, | |
460 | dev->n_mem_sections); | |
461 | dev->mem_sections[dev->n_mem_sections - 1] = *section; | |
04097f7c AK |
462 | vhost_set_memory(listener, section, true); |
463 | } | |
464 | ||
465 | static void vhost_region_del(MemoryListener *listener, | |
466 | MemoryRegionSection *section) | |
467 | { | |
2817b260 AK |
468 | struct vhost_dev *dev = container_of(listener, struct vhost_dev, |
469 | memory_listener); | |
470 | int i; | |
471 | ||
c49450b9 AK |
472 | if (!vhost_section(section)) { |
473 | return; | |
474 | } | |
475 | ||
04097f7c | 476 | vhost_set_memory(listener, section, false); |
2817b260 AK |
477 | for (i = 0; i < dev->n_mem_sections; ++i) { |
478 | if (dev->mem_sections[i].offset_within_address_space | |
479 | == section->offset_within_address_space) { | |
480 | --dev->n_mem_sections; | |
481 | memmove(&dev->mem_sections[i], &dev->mem_sections[i+1], | |
637f7a6a | 482 | (dev->n_mem_sections - i) * sizeof(*dev->mem_sections)); |
2817b260 AK |
483 | break; |
484 | } | |
485 | } | |
04097f7c AK |
486 | } |
487 | ||
50c1e149 AK |
488 | static void vhost_region_nop(MemoryListener *listener, |
489 | MemoryRegionSection *section) | |
490 | { | |
491 | } | |
492 | ||
d5970055 MT |
493 | static int vhost_virtqueue_set_addr(struct vhost_dev *dev, |
494 | struct vhost_virtqueue *vq, | |
495 | unsigned idx, bool enable_log) | |
496 | { | |
497 | struct vhost_vring_addr addr = { | |
498 | .index = idx, | |
2b3af999 SW |
499 | .desc_user_addr = (uint64_t)(unsigned long)vq->desc, |
500 | .avail_user_addr = (uint64_t)(unsigned long)vq->avail, | |
501 | .used_user_addr = (uint64_t)(unsigned long)vq->used, | |
d5970055 MT |
502 | .log_guest_addr = vq->used_phys, |
503 | .flags = enable_log ? (1 << VHOST_VRING_F_LOG) : 0, | |
504 | }; | |
505 | int r = ioctl(dev->control, VHOST_SET_VRING_ADDR, &addr); | |
506 | if (r < 0) { | |
507 | return -errno; | |
508 | } | |
509 | return 0; | |
510 | } | |
511 | ||
512 | static int vhost_dev_set_features(struct vhost_dev *dev, bool enable_log) | |
513 | { | |
514 | uint64_t features = dev->acked_features; | |
515 | int r; | |
516 | if (enable_log) { | |
517 | features |= 0x1 << VHOST_F_LOG_ALL; | |
518 | } | |
519 | r = ioctl(dev->control, VHOST_SET_FEATURES, &features); | |
520 | return r < 0 ? -errno : 0; | |
521 | } | |
522 | ||
523 | static int vhost_dev_set_log(struct vhost_dev *dev, bool enable_log) | |
524 | { | |
525 | int r, t, i; | |
526 | r = vhost_dev_set_features(dev, enable_log); | |
527 | if (r < 0) { | |
528 | goto err_features; | |
529 | } | |
530 | for (i = 0; i < dev->nvqs; ++i) { | |
531 | r = vhost_virtqueue_set_addr(dev, dev->vqs + i, i, | |
532 | enable_log); | |
533 | if (r < 0) { | |
534 | goto err_vq; | |
535 | } | |
536 | } | |
537 | return 0; | |
538 | err_vq: | |
539 | for (; i >= 0; --i) { | |
540 | t = vhost_virtqueue_set_addr(dev, dev->vqs + i, i, | |
541 | dev->log_enabled); | |
542 | assert(t >= 0); | |
543 | } | |
544 | t = vhost_dev_set_features(dev, dev->log_enabled); | |
545 | assert(t >= 0); | |
546 | err_features: | |
547 | return r; | |
548 | } | |
549 | ||
04097f7c | 550 | static int vhost_migration_log(MemoryListener *listener, int enable) |
d5970055 | 551 | { |
04097f7c AK |
552 | struct vhost_dev *dev = container_of(listener, struct vhost_dev, |
553 | memory_listener); | |
d5970055 MT |
554 | int r; |
555 | if (!!enable == dev->log_enabled) { | |
556 | return 0; | |
557 | } | |
558 | if (!dev->started) { | |
559 | dev->log_enabled = enable; | |
560 | return 0; | |
561 | } | |
562 | if (!enable) { | |
563 | r = vhost_dev_set_log(dev, false); | |
564 | if (r < 0) { | |
565 | return r; | |
566 | } | |
567 | if (dev->log) { | |
7267c094 | 568 | g_free(dev->log); |
d5970055 MT |
569 | } |
570 | dev->log = NULL; | |
571 | dev->log_size = 0; | |
572 | } else { | |
573 | vhost_dev_log_resize(dev, vhost_get_log_size(dev)); | |
574 | r = vhost_dev_set_log(dev, true); | |
575 | if (r < 0) { | |
576 | return r; | |
577 | } | |
578 | } | |
579 | dev->log_enabled = enable; | |
580 | return 0; | |
581 | } | |
582 | ||
04097f7c AK |
583 | static void vhost_log_global_start(MemoryListener *listener) |
584 | { | |
585 | int r; | |
586 | ||
587 | r = vhost_migration_log(listener, true); | |
588 | if (r < 0) { | |
589 | abort(); | |
590 | } | |
591 | } | |
592 | ||
593 | static void vhost_log_global_stop(MemoryListener *listener) | |
594 | { | |
595 | int r; | |
596 | ||
597 | r = vhost_migration_log(listener, false); | |
598 | if (r < 0) { | |
599 | abort(); | |
600 | } | |
601 | } | |
602 | ||
603 | static void vhost_log_start(MemoryListener *listener, | |
604 | MemoryRegionSection *section) | |
605 | { | |
606 | /* FIXME: implement */ | |
607 | } | |
608 | ||
609 | static void vhost_log_stop(MemoryListener *listener, | |
610 | MemoryRegionSection *section) | |
611 | { | |
612 | /* FIXME: implement */ | |
613 | } | |
614 | ||
d5970055 MT |
615 | static int vhost_virtqueue_init(struct vhost_dev *dev, |
616 | struct VirtIODevice *vdev, | |
617 | struct vhost_virtqueue *vq, | |
618 | unsigned idx) | |
619 | { | |
a8170e5e | 620 | hwaddr s, l, a; |
d5970055 MT |
621 | int r; |
622 | struct vhost_vring_file file = { | |
623 | .index = idx, | |
624 | }; | |
625 | struct vhost_vring_state state = { | |
626 | .index = idx, | |
627 | }; | |
628 | struct VirtQueue *vvq = virtio_get_queue(vdev, idx); | |
629 | ||
d5970055 MT |
630 | vq->num = state.num = virtio_queue_get_num(vdev, idx); |
631 | r = ioctl(dev->control, VHOST_SET_VRING_NUM, &state); | |
632 | if (r) { | |
633 | return -errno; | |
634 | } | |
635 | ||
636 | state.num = virtio_queue_get_last_avail_idx(vdev, idx); | |
637 | r = ioctl(dev->control, VHOST_SET_VRING_BASE, &state); | |
638 | if (r) { | |
639 | return -errno; | |
640 | } | |
641 | ||
642 | s = l = virtio_queue_get_desc_size(vdev, idx); | |
643 | a = virtio_queue_get_desc_addr(vdev, idx); | |
644 | vq->desc = cpu_physical_memory_map(a, &l, 0); | |
645 | if (!vq->desc || l != s) { | |
646 | r = -ENOMEM; | |
647 | goto fail_alloc_desc; | |
648 | } | |
649 | s = l = virtio_queue_get_avail_size(vdev, idx); | |
650 | a = virtio_queue_get_avail_addr(vdev, idx); | |
651 | vq->avail = cpu_physical_memory_map(a, &l, 0); | |
652 | if (!vq->avail || l != s) { | |
653 | r = -ENOMEM; | |
654 | goto fail_alloc_avail; | |
655 | } | |
656 | vq->used_size = s = l = virtio_queue_get_used_size(vdev, idx); | |
657 | vq->used_phys = a = virtio_queue_get_used_addr(vdev, idx); | |
658 | vq->used = cpu_physical_memory_map(a, &l, 1); | |
659 | if (!vq->used || l != s) { | |
660 | r = -ENOMEM; | |
661 | goto fail_alloc_used; | |
662 | } | |
663 | ||
664 | vq->ring_size = s = l = virtio_queue_get_ring_size(vdev, idx); | |
665 | vq->ring_phys = a = virtio_queue_get_ring_addr(vdev, idx); | |
666 | vq->ring = cpu_physical_memory_map(a, &l, 1); | |
667 | if (!vq->ring || l != s) { | |
668 | r = -ENOMEM; | |
669 | goto fail_alloc_ring; | |
670 | } | |
671 | ||
672 | r = vhost_virtqueue_set_addr(dev, vq, idx, dev->log_enabled); | |
673 | if (r < 0) { | |
674 | r = -errno; | |
675 | goto fail_alloc; | |
676 | } | |
d5970055 MT |
677 | file.fd = event_notifier_get_fd(virtio_queue_get_host_notifier(vvq)); |
678 | r = ioctl(dev->control, VHOST_SET_VRING_KICK, &file); | |
679 | if (r) { | |
c8852121 | 680 | r = -errno; |
d5970055 MT |
681 | goto fail_kick; |
682 | } | |
683 | ||
684 | file.fd = event_notifier_get_fd(virtio_queue_get_guest_notifier(vvq)); | |
685 | r = ioctl(dev->control, VHOST_SET_VRING_CALL, &file); | |
686 | if (r) { | |
c8852121 | 687 | r = -errno; |
d5970055 MT |
688 | goto fail_call; |
689 | } | |
690 | ||
691 | return 0; | |
692 | ||
693 | fail_call: | |
694 | fail_kick: | |
d5970055 MT |
695 | fail_alloc: |
696 | cpu_physical_memory_unmap(vq->ring, virtio_queue_get_ring_size(vdev, idx), | |
697 | 0, 0); | |
698 | fail_alloc_ring: | |
699 | cpu_physical_memory_unmap(vq->used, virtio_queue_get_used_size(vdev, idx), | |
700 | 0, 0); | |
701 | fail_alloc_used: | |
702 | cpu_physical_memory_unmap(vq->avail, virtio_queue_get_avail_size(vdev, idx), | |
703 | 0, 0); | |
704 | fail_alloc_avail: | |
705 | cpu_physical_memory_unmap(vq->desc, virtio_queue_get_desc_size(vdev, idx), | |
706 | 0, 0); | |
707 | fail_alloc_desc: | |
708 | return r; | |
709 | } | |
710 | ||
711 | static void vhost_virtqueue_cleanup(struct vhost_dev *dev, | |
712 | struct VirtIODevice *vdev, | |
713 | struct vhost_virtqueue *vq, | |
714 | unsigned idx) | |
715 | { | |
716 | struct vhost_vring_state state = { | |
717 | .index = idx, | |
718 | }; | |
719 | int r; | |
d5970055 MT |
720 | r = ioctl(dev->control, VHOST_GET_VRING_BASE, &state); |
721 | if (r < 0) { | |
722 | fprintf(stderr, "vhost VQ %d ring restore failed: %d\n", idx, r); | |
723 | fflush(stderr); | |
724 | } | |
725 | virtio_queue_set_last_avail_idx(vdev, idx, state.num); | |
726 | assert (r >= 0); | |
727 | cpu_physical_memory_unmap(vq->ring, virtio_queue_get_ring_size(vdev, idx), | |
728 | 0, virtio_queue_get_ring_size(vdev, idx)); | |
729 | cpu_physical_memory_unmap(vq->used, virtio_queue_get_used_size(vdev, idx), | |
730 | 1, virtio_queue_get_used_size(vdev, idx)); | |
731 | cpu_physical_memory_unmap(vq->avail, virtio_queue_get_avail_size(vdev, idx), | |
732 | 0, virtio_queue_get_avail_size(vdev, idx)); | |
733 | cpu_physical_memory_unmap(vq->desc, virtio_queue_get_desc_size(vdev, idx), | |
734 | 0, virtio_queue_get_desc_size(vdev, idx)); | |
735 | } | |
736 | ||
80a1ea37 AK |
737 | static void vhost_eventfd_add(MemoryListener *listener, |
738 | MemoryRegionSection *section, | |
753d5e14 | 739 | bool match_data, uint64_t data, EventNotifier *e) |
80a1ea37 AK |
740 | { |
741 | } | |
742 | ||
743 | static void vhost_eventfd_del(MemoryListener *listener, | |
744 | MemoryRegionSection *section, | |
753d5e14 | 745 | bool match_data, uint64_t data, EventNotifier *e) |
80a1ea37 AK |
746 | { |
747 | } | |
748 | ||
1241ed94 SH |
749 | int vhost_dev_init(struct vhost_dev *hdev, int devfd, const char *devpath, |
750 | bool force) | |
d5970055 MT |
751 | { |
752 | uint64_t features; | |
753 | int r; | |
754 | if (devfd >= 0) { | |
755 | hdev->control = devfd; | |
756 | } else { | |
1241ed94 | 757 | hdev->control = open(devpath, O_RDWR); |
d5970055 MT |
758 | if (hdev->control < 0) { |
759 | return -errno; | |
760 | } | |
761 | } | |
762 | r = ioctl(hdev->control, VHOST_SET_OWNER, NULL); | |
763 | if (r < 0) { | |
764 | goto fail; | |
765 | } | |
766 | ||
767 | r = ioctl(hdev->control, VHOST_GET_FEATURES, &features); | |
768 | if (r < 0) { | |
769 | goto fail; | |
770 | } | |
771 | hdev->features = features; | |
772 | ||
04097f7c | 773 | hdev->memory_listener = (MemoryListener) { |
50c1e149 AK |
774 | .begin = vhost_begin, |
775 | .commit = vhost_commit, | |
04097f7c AK |
776 | .region_add = vhost_region_add, |
777 | .region_del = vhost_region_del, | |
50c1e149 | 778 | .region_nop = vhost_region_nop, |
04097f7c AK |
779 | .log_start = vhost_log_start, |
780 | .log_stop = vhost_log_stop, | |
781 | .log_sync = vhost_log_sync, | |
782 | .log_global_start = vhost_log_global_start, | |
783 | .log_global_stop = vhost_log_global_stop, | |
80a1ea37 AK |
784 | .eventfd_add = vhost_eventfd_add, |
785 | .eventfd_del = vhost_eventfd_del, | |
72e22d2f | 786 | .priority = 10 |
04097f7c | 787 | }; |
7267c094 | 788 | hdev->mem = g_malloc0(offsetof(struct vhost_memory, regions)); |
2817b260 AK |
789 | hdev->n_mem_sections = 0; |
790 | hdev->mem_sections = NULL; | |
d5970055 MT |
791 | hdev->log = NULL; |
792 | hdev->log_size = 0; | |
793 | hdev->log_enabled = false; | |
794 | hdev->started = false; | |
f6790af6 | 795 | memory_listener_register(&hdev->memory_listener, &address_space_memory); |
5430a28f | 796 | hdev->force = force; |
d5970055 MT |
797 | return 0; |
798 | fail: | |
799 | r = -errno; | |
800 | close(hdev->control); | |
801 | return r; | |
802 | } | |
803 | ||
804 | void vhost_dev_cleanup(struct vhost_dev *hdev) | |
805 | { | |
04097f7c | 806 | memory_listener_unregister(&hdev->memory_listener); |
7267c094 | 807 | g_free(hdev->mem); |
2817b260 | 808 | g_free(hdev->mem_sections); |
d5970055 MT |
809 | close(hdev->control); |
810 | } | |
811 | ||
5430a28f MT |
812 | bool vhost_dev_query(struct vhost_dev *hdev, VirtIODevice *vdev) |
813 | { | |
814 | return !vdev->binding->query_guest_notifiers || | |
815 | vdev->binding->query_guest_notifiers(vdev->binding_opaque) || | |
816 | hdev->force; | |
817 | } | |
818 | ||
b0b3db79 MT |
819 | /* Stop processing guest IO notifications in qemu. |
820 | * Start processing them in vhost in kernel. | |
821 | */ | |
822 | int vhost_dev_enable_notifiers(struct vhost_dev *hdev, VirtIODevice *vdev) | |
823 | { | |
824 | int i, r; | |
825 | if (!vdev->binding->set_host_notifier) { | |
826 | fprintf(stderr, "binding does not support host notifiers\n"); | |
827 | r = -ENOSYS; | |
828 | goto fail; | |
829 | } | |
830 | ||
831 | for (i = 0; i < hdev->nvqs; ++i) { | |
832 | r = vdev->binding->set_host_notifier(vdev->binding_opaque, i, true); | |
833 | if (r < 0) { | |
834 | fprintf(stderr, "vhost VQ %d notifier binding failed: %d\n", i, -r); | |
835 | goto fail_vq; | |
836 | } | |
837 | } | |
838 | ||
839 | return 0; | |
840 | fail_vq: | |
841 | while (--i >= 0) { | |
842 | r = vdev->binding->set_host_notifier(vdev->binding_opaque, i, false); | |
843 | if (r < 0) { | |
844 | fprintf(stderr, "vhost VQ %d notifier cleanup error: %d\n", i, -r); | |
845 | fflush(stderr); | |
846 | } | |
847 | assert (r >= 0); | |
848 | } | |
849 | fail: | |
850 | return r; | |
851 | } | |
852 | ||
853 | /* Stop processing guest IO notifications in vhost. | |
854 | * Start processing them in qemu. | |
855 | * This might actually run the qemu handlers right away, | |
856 | * so virtio in qemu must be completely setup when this is called. | |
857 | */ | |
858 | void vhost_dev_disable_notifiers(struct vhost_dev *hdev, VirtIODevice *vdev) | |
859 | { | |
860 | int i, r; | |
861 | ||
862 | for (i = 0; i < hdev->nvqs; ++i) { | |
863 | r = vdev->binding->set_host_notifier(vdev->binding_opaque, i, false); | |
864 | if (r < 0) { | |
865 | fprintf(stderr, "vhost VQ %d notifier cleanup failed: %d\n", i, -r); | |
866 | fflush(stderr); | |
867 | } | |
868 | assert (r >= 0); | |
869 | } | |
870 | } | |
871 | ||
872 | /* Host notifiers must be enabled at this point. */ | |
d5970055 MT |
873 | int vhost_dev_start(struct vhost_dev *hdev, VirtIODevice *vdev) |
874 | { | |
875 | int i, r; | |
54dd9321 MT |
876 | if (!vdev->binding->set_guest_notifiers) { |
877 | fprintf(stderr, "binding does not support guest notifiers\n"); | |
878 | r = -ENOSYS; | |
879 | goto fail; | |
880 | } | |
881 | ||
882 | r = vdev->binding->set_guest_notifiers(vdev->binding_opaque, true); | |
883 | if (r < 0) { | |
884 | fprintf(stderr, "Error binding guest notifier: %d\n", -r); | |
885 | goto fail_notifiers; | |
886 | } | |
d5970055 MT |
887 | |
888 | r = vhost_dev_set_features(hdev, hdev->log_enabled); | |
889 | if (r < 0) { | |
54dd9321 | 890 | goto fail_features; |
d5970055 MT |
891 | } |
892 | r = ioctl(hdev->control, VHOST_SET_MEM_TABLE, hdev->mem); | |
893 | if (r < 0) { | |
894 | r = -errno; | |
54dd9321 | 895 | goto fail_mem; |
d5970055 | 896 | } |
d154e0ba MT |
897 | for (i = 0; i < hdev->nvqs; ++i) { |
898 | r = vhost_virtqueue_init(hdev, | |
899 | vdev, | |
900 | hdev->vqs + i, | |
901 | i); | |
902 | if (r < 0) { | |
903 | goto fail_vq; | |
904 | } | |
905 | } | |
906 | ||
d5970055 MT |
907 | if (hdev->log_enabled) { |
908 | hdev->log_size = vhost_get_log_size(hdev); | |
909 | hdev->log = hdev->log_size ? | |
7267c094 | 910 | g_malloc0(hdev->log_size * sizeof *hdev->log) : NULL; |
d5970055 MT |
911 | r = ioctl(hdev->control, VHOST_SET_LOG_BASE, |
912 | (uint64_t)(unsigned long)hdev->log); | |
913 | if (r < 0) { | |
914 | r = -errno; | |
54dd9321 | 915 | goto fail_log; |
d5970055 MT |
916 | } |
917 | } | |
d154e0ba | 918 | |
d5970055 MT |
919 | hdev->started = true; |
920 | ||
921 | return 0; | |
54dd9321 | 922 | fail_log: |
d5970055 MT |
923 | fail_vq: |
924 | while (--i >= 0) { | |
925 | vhost_virtqueue_cleanup(hdev, | |
926 | vdev, | |
927 | hdev->vqs + i, | |
928 | i); | |
929 | } | |
54dd9321 MT |
930 | fail_mem: |
931 | fail_features: | |
932 | vdev->binding->set_guest_notifiers(vdev->binding_opaque, false); | |
933 | fail_notifiers: | |
d5970055 MT |
934 | fail: |
935 | return r; | |
936 | } | |
937 | ||
b0b3db79 | 938 | /* Host notifiers must be enabled at this point. */ |
d5970055 MT |
939 | void vhost_dev_stop(struct vhost_dev *hdev, VirtIODevice *vdev) |
940 | { | |
54dd9321 MT |
941 | int i, r; |
942 | ||
d5970055 MT |
943 | for (i = 0; i < hdev->nvqs; ++i) { |
944 | vhost_virtqueue_cleanup(hdev, | |
945 | vdev, | |
946 | hdev->vqs + i, | |
947 | i); | |
948 | } | |
2817b260 AK |
949 | for (i = 0; i < hdev->n_mem_sections; ++i) { |
950 | vhost_sync_dirty_bitmap(hdev, &hdev->mem_sections[i], | |
a8170e5e | 951 | 0, (hwaddr)~0x0ull); |
2817b260 | 952 | } |
54dd9321 MT |
953 | r = vdev->binding->set_guest_notifiers(vdev->binding_opaque, false); |
954 | if (r < 0) { | |
955 | fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", r); | |
956 | fflush(stderr); | |
957 | } | |
958 | assert (r >= 0); | |
959 | ||
d5970055 | 960 | hdev->started = false; |
7267c094 | 961 | g_free(hdev->log); |
c1be973a | 962 | hdev->log = NULL; |
d5970055 MT |
963 | hdev->log_size = 0; |
964 | } |