]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * QEMU KVM support | |
3 | * | |
4 | * Copyright IBM, Corp. 2008 | |
5 | * Red Hat, Inc. 2008 | |
6 | * | |
7 | * Authors: | |
8 | * Anthony Liguori <aliguori@us.ibm.com> | |
9 | * Glauber Costa <gcosta@redhat.com> | |
10 | * | |
11 | * This work is licensed under the terms of the GNU GPL, version 2 or later. | |
12 | * See the COPYING file in the top-level directory. | |
13 | * | |
14 | */ | |
15 | ||
16 | #include <sys/types.h> | |
17 | #include <sys/ioctl.h> | |
18 | #include <sys/mman.h> | |
19 | #include <stdarg.h> | |
20 | ||
21 | #include <linux/kvm.h> | |
22 | ||
23 | #include "qemu-common.h" | |
24 | #include "qemu-barrier.h" | |
25 | #include "sysemu.h" | |
26 | #include "hw/hw.h" | |
27 | #include "gdbstub.h" | |
28 | #include "kvm.h" | |
29 | #include "bswap.h" | |
30 | #include "memory.h" | |
31 | #include "exec-memory.h" | |
32 | ||
33 | /* This check must be after config-host.h is included */ | |
34 | #ifdef CONFIG_EVENTFD | |
35 | #include <sys/eventfd.h> | |
36 | #endif | |
37 | ||
38 | /* KVM uses PAGE_SIZE in its definition of COALESCED_MMIO_MAX */ | |
39 | #define PAGE_SIZE TARGET_PAGE_SIZE | |
40 | ||
41 | //#define DEBUG_KVM | |
42 | ||
43 | #ifdef DEBUG_KVM | |
44 | #define DPRINTF(fmt, ...) \ | |
45 | do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0) | |
46 | #else | |
47 | #define DPRINTF(fmt, ...) \ | |
48 | do { } while (0) | |
49 | #endif | |
50 | ||
51 | typedef struct KVMSlot | |
52 | { | |
53 | target_phys_addr_t start_addr; | |
54 | ram_addr_t memory_size; | |
55 | void *ram; | |
56 | int slot; | |
57 | int flags; | |
58 | } KVMSlot; | |
59 | ||
60 | typedef struct kvm_dirty_log KVMDirtyLog; | |
61 | ||
62 | struct KVMState | |
63 | { | |
64 | KVMSlot slots[32]; | |
65 | int fd; | |
66 | int vmfd; | |
67 | int coalesced_mmio; | |
68 | struct kvm_coalesced_mmio_ring *coalesced_mmio_ring; | |
69 | bool coalesced_flush_in_progress; | |
70 | int broken_set_mem_region; | |
71 | int migration_log; | |
72 | int vcpu_events; | |
73 | int robust_singlestep; | |
74 | int debugregs; | |
75 | #ifdef KVM_CAP_SET_GUEST_DEBUG | |
76 | struct kvm_sw_breakpoint_head kvm_sw_breakpoints; | |
77 | #endif | |
78 | int pit_state2; | |
79 | int xsave, xcrs; | |
80 | int many_ioeventfds; | |
81 | /* The man page (and posix) say ioctl numbers are signed int, but | |
82 | * they're not. Linux, glibc and *BSD all treat ioctl numbers as | |
83 | * unsigned, and treating them as signed here can break things */ | |
84 | unsigned irqchip_inject_ioctl; | |
85 | #ifdef KVM_CAP_IRQ_ROUTING | |
86 | struct kvm_irq_routing *irq_routes; | |
87 | int nr_allocated_irq_routes; | |
88 | uint32_t *used_gsi_bitmap; | |
89 | unsigned int max_gsi; | |
90 | #endif | |
91 | }; | |
92 | ||
93 | KVMState *kvm_state; | |
94 | bool kvm_kernel_irqchip; | |
95 | ||
96 | static const KVMCapabilityInfo kvm_required_capabilites[] = { | |
97 | KVM_CAP_INFO(USER_MEMORY), | |
98 | KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS), | |
99 | KVM_CAP_LAST_INFO | |
100 | }; | |
101 | ||
102 | static KVMSlot *kvm_alloc_slot(KVMState *s) | |
103 | { | |
104 | int i; | |
105 | ||
106 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
107 | if (s->slots[i].memory_size == 0) { | |
108 | return &s->slots[i]; | |
109 | } | |
110 | } | |
111 | ||
112 | fprintf(stderr, "%s: no free slot available\n", __func__); | |
113 | abort(); | |
114 | } | |
115 | ||
116 | static KVMSlot *kvm_lookup_matching_slot(KVMState *s, | |
117 | target_phys_addr_t start_addr, | |
118 | target_phys_addr_t end_addr) | |
119 | { | |
120 | int i; | |
121 | ||
122 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
123 | KVMSlot *mem = &s->slots[i]; | |
124 | ||
125 | if (start_addr == mem->start_addr && | |
126 | end_addr == mem->start_addr + mem->memory_size) { | |
127 | return mem; | |
128 | } | |
129 | } | |
130 | ||
131 | return NULL; | |
132 | } | |
133 | ||
134 | /* | |
135 | * Find overlapping slot with lowest start address | |
136 | */ | |
137 | static KVMSlot *kvm_lookup_overlapping_slot(KVMState *s, | |
138 | target_phys_addr_t start_addr, | |
139 | target_phys_addr_t end_addr) | |
140 | { | |
141 | KVMSlot *found = NULL; | |
142 | int i; | |
143 | ||
144 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
145 | KVMSlot *mem = &s->slots[i]; | |
146 | ||
147 | if (mem->memory_size == 0 || | |
148 | (found && found->start_addr < mem->start_addr)) { | |
149 | continue; | |
150 | } | |
151 | ||
152 | if (end_addr > mem->start_addr && | |
153 | start_addr < mem->start_addr + mem->memory_size) { | |
154 | found = mem; | |
155 | } | |
156 | } | |
157 | ||
158 | return found; | |
159 | } | |
160 | ||
161 | int kvm_physical_memory_addr_from_host(KVMState *s, void *ram, | |
162 | target_phys_addr_t *phys_addr) | |
163 | { | |
164 | int i; | |
165 | ||
166 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
167 | KVMSlot *mem = &s->slots[i]; | |
168 | ||
169 | if (ram >= mem->ram && ram < mem->ram + mem->memory_size) { | |
170 | *phys_addr = mem->start_addr + (ram - mem->ram); | |
171 | return 1; | |
172 | } | |
173 | } | |
174 | ||
175 | return 0; | |
176 | } | |
177 | ||
178 | static int kvm_set_user_memory_region(KVMState *s, KVMSlot *slot) | |
179 | { | |
180 | struct kvm_userspace_memory_region mem; | |
181 | ||
182 | mem.slot = slot->slot; | |
183 | mem.guest_phys_addr = slot->start_addr; | |
184 | mem.memory_size = slot->memory_size; | |
185 | mem.userspace_addr = (unsigned long)slot->ram; | |
186 | mem.flags = slot->flags; | |
187 | if (s->migration_log) { | |
188 | mem.flags |= KVM_MEM_LOG_DIRTY_PAGES; | |
189 | } | |
190 | return kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem); | |
191 | } | |
192 | ||
193 | static void kvm_reset_vcpu(void *opaque) | |
194 | { | |
195 | CPUArchState *env = opaque; | |
196 | ||
197 | kvm_arch_reset_vcpu(env); | |
198 | } | |
199 | ||
200 | int kvm_init_vcpu(CPUArchState *env) | |
201 | { | |
202 | KVMState *s = kvm_state; | |
203 | long mmap_size; | |
204 | int ret; | |
205 | ||
206 | DPRINTF("kvm_init_vcpu\n"); | |
207 | ||
208 | ret = kvm_vm_ioctl(s, KVM_CREATE_VCPU, env->cpu_index); | |
209 | if (ret < 0) { | |
210 | DPRINTF("kvm_create_vcpu failed\n"); | |
211 | goto err; | |
212 | } | |
213 | ||
214 | env->kvm_fd = ret; | |
215 | env->kvm_state = s; | |
216 | env->kvm_vcpu_dirty = 1; | |
217 | ||
218 | mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0); | |
219 | if (mmap_size < 0) { | |
220 | ret = mmap_size; | |
221 | DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n"); | |
222 | goto err; | |
223 | } | |
224 | ||
225 | env->kvm_run = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED, | |
226 | env->kvm_fd, 0); | |
227 | if (env->kvm_run == MAP_FAILED) { | |
228 | ret = -errno; | |
229 | DPRINTF("mmap'ing vcpu state failed\n"); | |
230 | goto err; | |
231 | } | |
232 | ||
233 | if (s->coalesced_mmio && !s->coalesced_mmio_ring) { | |
234 | s->coalesced_mmio_ring = | |
235 | (void *)env->kvm_run + s->coalesced_mmio * PAGE_SIZE; | |
236 | } | |
237 | ||
238 | ret = kvm_arch_init_vcpu(env); | |
239 | if (ret == 0) { | |
240 | qemu_register_reset(kvm_reset_vcpu, env); | |
241 | kvm_arch_reset_vcpu(env); | |
242 | } | |
243 | err: | |
244 | return ret; | |
245 | } | |
246 | ||
247 | /* | |
248 | * dirty pages logging control | |
249 | */ | |
250 | ||
251 | static int kvm_mem_flags(KVMState *s, bool log_dirty) | |
252 | { | |
253 | return log_dirty ? KVM_MEM_LOG_DIRTY_PAGES : 0; | |
254 | } | |
255 | ||
256 | static int kvm_slot_dirty_pages_log_change(KVMSlot *mem, bool log_dirty) | |
257 | { | |
258 | KVMState *s = kvm_state; | |
259 | int flags, mask = KVM_MEM_LOG_DIRTY_PAGES; | |
260 | int old_flags; | |
261 | ||
262 | old_flags = mem->flags; | |
263 | ||
264 | flags = (mem->flags & ~mask) | kvm_mem_flags(s, log_dirty); | |
265 | mem->flags = flags; | |
266 | ||
267 | /* If nothing changed effectively, no need to issue ioctl */ | |
268 | if (s->migration_log) { | |
269 | flags |= KVM_MEM_LOG_DIRTY_PAGES; | |
270 | } | |
271 | ||
272 | if (flags == old_flags) { | |
273 | return 0; | |
274 | } | |
275 | ||
276 | return kvm_set_user_memory_region(s, mem); | |
277 | } | |
278 | ||
279 | static int kvm_dirty_pages_log_change(target_phys_addr_t phys_addr, | |
280 | ram_addr_t size, bool log_dirty) | |
281 | { | |
282 | KVMState *s = kvm_state; | |
283 | KVMSlot *mem = kvm_lookup_matching_slot(s, phys_addr, phys_addr + size); | |
284 | ||
285 | if (mem == NULL) { | |
286 | fprintf(stderr, "BUG: %s: invalid parameters " TARGET_FMT_plx "-" | |
287 | TARGET_FMT_plx "\n", __func__, phys_addr, | |
288 | (target_phys_addr_t)(phys_addr + size - 1)); | |
289 | return -EINVAL; | |
290 | } | |
291 | return kvm_slot_dirty_pages_log_change(mem, log_dirty); | |
292 | } | |
293 | ||
294 | static void kvm_log_start(MemoryListener *listener, | |
295 | MemoryRegionSection *section) | |
296 | { | |
297 | int r; | |
298 | ||
299 | r = kvm_dirty_pages_log_change(section->offset_within_address_space, | |
300 | section->size, true); | |
301 | if (r < 0) { | |
302 | abort(); | |
303 | } | |
304 | } | |
305 | ||
306 | static void kvm_log_stop(MemoryListener *listener, | |
307 | MemoryRegionSection *section) | |
308 | { | |
309 | int r; | |
310 | ||
311 | r = kvm_dirty_pages_log_change(section->offset_within_address_space, | |
312 | section->size, false); | |
313 | if (r < 0) { | |
314 | abort(); | |
315 | } | |
316 | } | |
317 | ||
318 | static int kvm_set_migration_log(int enable) | |
319 | { | |
320 | KVMState *s = kvm_state; | |
321 | KVMSlot *mem; | |
322 | int i, err; | |
323 | ||
324 | s->migration_log = enable; | |
325 | ||
326 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
327 | mem = &s->slots[i]; | |
328 | ||
329 | if (!mem->memory_size) { | |
330 | continue; | |
331 | } | |
332 | if (!!(mem->flags & KVM_MEM_LOG_DIRTY_PAGES) == enable) { | |
333 | continue; | |
334 | } | |
335 | err = kvm_set_user_memory_region(s, mem); | |
336 | if (err) { | |
337 | return err; | |
338 | } | |
339 | } | |
340 | return 0; | |
341 | } | |
342 | ||
343 | /* get kvm's dirty pages bitmap and update qemu's */ | |
344 | static int kvm_get_dirty_pages_log_range(MemoryRegionSection *section, | |
345 | unsigned long *bitmap) | |
346 | { | |
347 | unsigned int i, j; | |
348 | unsigned long page_number, c; | |
349 | target_phys_addr_t addr, addr1; | |
350 | unsigned int len = ((section->size / TARGET_PAGE_SIZE) + HOST_LONG_BITS - 1) / HOST_LONG_BITS; | |
351 | ||
352 | /* | |
353 | * bitmap-traveling is faster than memory-traveling (for addr...) | |
354 | * especially when most of the memory is not dirty. | |
355 | */ | |
356 | for (i = 0; i < len; i++) { | |
357 | if (bitmap[i] != 0) { | |
358 | c = leul_to_cpu(bitmap[i]); | |
359 | do { | |
360 | j = ffsl(c) - 1; | |
361 | c &= ~(1ul << j); | |
362 | page_number = i * HOST_LONG_BITS + j; | |
363 | addr1 = page_number * TARGET_PAGE_SIZE; | |
364 | addr = section->offset_within_region + addr1; | |
365 | memory_region_set_dirty(section->mr, addr, TARGET_PAGE_SIZE); | |
366 | } while (c != 0); | |
367 | } | |
368 | } | |
369 | return 0; | |
370 | } | |
371 | ||
372 | #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1)) | |
373 | ||
374 | /** | |
375 | * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space | |
376 | * This function updates qemu's dirty bitmap using | |
377 | * memory_region_set_dirty(). This means all bits are set | |
378 | * to dirty. | |
379 | * | |
380 | * @start_add: start of logged region. | |
381 | * @end_addr: end of logged region. | |
382 | */ | |
383 | static int kvm_physical_sync_dirty_bitmap(MemoryRegionSection *section) | |
384 | { | |
385 | KVMState *s = kvm_state; | |
386 | unsigned long size, allocated_size = 0; | |
387 | KVMDirtyLog d; | |
388 | KVMSlot *mem; | |
389 | int ret = 0; | |
390 | target_phys_addr_t start_addr = section->offset_within_address_space; | |
391 | target_phys_addr_t end_addr = start_addr + section->size; | |
392 | ||
393 | d.dirty_bitmap = NULL; | |
394 | while (start_addr < end_addr) { | |
395 | mem = kvm_lookup_overlapping_slot(s, start_addr, end_addr); | |
396 | if (mem == NULL) { | |
397 | break; | |
398 | } | |
399 | ||
400 | /* XXX bad kernel interface alert | |
401 | * For dirty bitmap, kernel allocates array of size aligned to | |
402 | * bits-per-long. But for case when the kernel is 64bits and | |
403 | * the userspace is 32bits, userspace can't align to the same | |
404 | * bits-per-long, since sizeof(long) is different between kernel | |
405 | * and user space. This way, userspace will provide buffer which | |
406 | * may be 4 bytes less than the kernel will use, resulting in | |
407 | * userspace memory corruption (which is not detectable by valgrind | |
408 | * too, in most cases). | |
409 | * So for now, let's align to 64 instead of HOST_LONG_BITS here, in | |
410 | * a hope that sizeof(long) wont become >8 any time soon. | |
411 | */ | |
412 | size = ALIGN(((mem->memory_size) >> TARGET_PAGE_BITS), | |
413 | /*HOST_LONG_BITS*/ 64) / 8; | |
414 | if (!d.dirty_bitmap) { | |
415 | d.dirty_bitmap = g_malloc(size); | |
416 | } else if (size > allocated_size) { | |
417 | d.dirty_bitmap = g_realloc(d.dirty_bitmap, size); | |
418 | } | |
419 | allocated_size = size; | |
420 | memset(d.dirty_bitmap, 0, allocated_size); | |
421 | ||
422 | d.slot = mem->slot; | |
423 | ||
424 | if (kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d) == -1) { | |
425 | DPRINTF("ioctl failed %d\n", errno); | |
426 | ret = -1; | |
427 | break; | |
428 | } | |
429 | ||
430 | kvm_get_dirty_pages_log_range(section, d.dirty_bitmap); | |
431 | start_addr = mem->start_addr + mem->memory_size; | |
432 | } | |
433 | g_free(d.dirty_bitmap); | |
434 | ||
435 | return ret; | |
436 | } | |
437 | ||
438 | int kvm_coalesce_mmio_region(target_phys_addr_t start, ram_addr_t size) | |
439 | { | |
440 | int ret = -ENOSYS; | |
441 | KVMState *s = kvm_state; | |
442 | ||
443 | if (s->coalesced_mmio) { | |
444 | struct kvm_coalesced_mmio_zone zone; | |
445 | ||
446 | zone.addr = start; | |
447 | zone.size = size; | |
448 | zone.pad = 0; | |
449 | ||
450 | ret = kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone); | |
451 | } | |
452 | ||
453 | return ret; | |
454 | } | |
455 | ||
456 | int kvm_uncoalesce_mmio_region(target_phys_addr_t start, ram_addr_t size) | |
457 | { | |
458 | int ret = -ENOSYS; | |
459 | KVMState *s = kvm_state; | |
460 | ||
461 | if (s->coalesced_mmio) { | |
462 | struct kvm_coalesced_mmio_zone zone; | |
463 | ||
464 | zone.addr = start; | |
465 | zone.size = size; | |
466 | zone.pad = 0; | |
467 | ||
468 | ret = kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone); | |
469 | } | |
470 | ||
471 | return ret; | |
472 | } | |
473 | ||
474 | int kvm_check_extension(KVMState *s, unsigned int extension) | |
475 | { | |
476 | int ret; | |
477 | ||
478 | ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension); | |
479 | if (ret < 0) { | |
480 | ret = 0; | |
481 | } | |
482 | ||
483 | return ret; | |
484 | } | |
485 | ||
486 | static int kvm_check_many_ioeventfds(void) | |
487 | { | |
488 | /* Userspace can use ioeventfd for io notification. This requires a host | |
489 | * that supports eventfd(2) and an I/O thread; since eventfd does not | |
490 | * support SIGIO it cannot interrupt the vcpu. | |
491 | * | |
492 | * Older kernels have a 6 device limit on the KVM io bus. Find out so we | |
493 | * can avoid creating too many ioeventfds. | |
494 | */ | |
495 | #if defined(CONFIG_EVENTFD) | |
496 | int ioeventfds[7]; | |
497 | int i, ret = 0; | |
498 | for (i = 0; i < ARRAY_SIZE(ioeventfds); i++) { | |
499 | ioeventfds[i] = eventfd(0, EFD_CLOEXEC); | |
500 | if (ioeventfds[i] < 0) { | |
501 | break; | |
502 | } | |
503 | ret = kvm_set_ioeventfd_pio_word(ioeventfds[i], 0, i, true); | |
504 | if (ret < 0) { | |
505 | close(ioeventfds[i]); | |
506 | break; | |
507 | } | |
508 | } | |
509 | ||
510 | /* Decide whether many devices are supported or not */ | |
511 | ret = i == ARRAY_SIZE(ioeventfds); | |
512 | ||
513 | while (i-- > 0) { | |
514 | kvm_set_ioeventfd_pio_word(ioeventfds[i], 0, i, false); | |
515 | close(ioeventfds[i]); | |
516 | } | |
517 | return ret; | |
518 | #else | |
519 | return 0; | |
520 | #endif | |
521 | } | |
522 | ||
523 | static const KVMCapabilityInfo * | |
524 | kvm_check_extension_list(KVMState *s, const KVMCapabilityInfo *list) | |
525 | { | |
526 | while (list->name) { | |
527 | if (!kvm_check_extension(s, list->value)) { | |
528 | return list; | |
529 | } | |
530 | list++; | |
531 | } | |
532 | return NULL; | |
533 | } | |
534 | ||
535 | static void kvm_set_phys_mem(MemoryRegionSection *section, bool add) | |
536 | { | |
537 | KVMState *s = kvm_state; | |
538 | KVMSlot *mem, old; | |
539 | int err; | |
540 | MemoryRegion *mr = section->mr; | |
541 | bool log_dirty = memory_region_is_logging(mr); | |
542 | target_phys_addr_t start_addr = section->offset_within_address_space; | |
543 | ram_addr_t size = section->size; | |
544 | void *ram = NULL; | |
545 | unsigned delta; | |
546 | ||
547 | /* kvm works in page size chunks, but the function may be called | |
548 | with sub-page size and unaligned start address. */ | |
549 | delta = TARGET_PAGE_ALIGN(size) - size; | |
550 | if (delta > size) { | |
551 | return; | |
552 | } | |
553 | start_addr += delta; | |
554 | size -= delta; | |
555 | size &= TARGET_PAGE_MASK; | |
556 | if (!size || (start_addr & ~TARGET_PAGE_MASK)) { | |
557 | return; | |
558 | } | |
559 | ||
560 | if (!memory_region_is_ram(mr)) { | |
561 | return; | |
562 | } | |
563 | ||
564 | ram = memory_region_get_ram_ptr(mr) + section->offset_within_region + delta; | |
565 | ||
566 | while (1) { | |
567 | mem = kvm_lookup_overlapping_slot(s, start_addr, start_addr + size); | |
568 | if (!mem) { | |
569 | break; | |
570 | } | |
571 | ||
572 | if (add && start_addr >= mem->start_addr && | |
573 | (start_addr + size <= mem->start_addr + mem->memory_size) && | |
574 | (ram - start_addr == mem->ram - mem->start_addr)) { | |
575 | /* The new slot fits into the existing one and comes with | |
576 | * identical parameters - update flags and done. */ | |
577 | kvm_slot_dirty_pages_log_change(mem, log_dirty); | |
578 | return; | |
579 | } | |
580 | ||
581 | old = *mem; | |
582 | ||
583 | if (mem->flags & KVM_MEM_LOG_DIRTY_PAGES) { | |
584 | kvm_physical_sync_dirty_bitmap(section); | |
585 | } | |
586 | ||
587 | /* unregister the overlapping slot */ | |
588 | mem->memory_size = 0; | |
589 | err = kvm_set_user_memory_region(s, mem); | |
590 | if (err) { | |
591 | fprintf(stderr, "%s: error unregistering overlapping slot: %s\n", | |
592 | __func__, strerror(-err)); | |
593 | abort(); | |
594 | } | |
595 | ||
596 | /* Workaround for older KVM versions: we can't join slots, even not by | |
597 | * unregistering the previous ones and then registering the larger | |
598 | * slot. We have to maintain the existing fragmentation. Sigh. | |
599 | * | |
600 | * This workaround assumes that the new slot starts at the same | |
601 | * address as the first existing one. If not or if some overlapping | |
602 | * slot comes around later, we will fail (not seen in practice so far) | |
603 | * - and actually require a recent KVM version. */ | |
604 | if (s->broken_set_mem_region && | |
605 | old.start_addr == start_addr && old.memory_size < size && add) { | |
606 | mem = kvm_alloc_slot(s); | |
607 | mem->memory_size = old.memory_size; | |
608 | mem->start_addr = old.start_addr; | |
609 | mem->ram = old.ram; | |
610 | mem->flags = kvm_mem_flags(s, log_dirty); | |
611 | ||
612 | err = kvm_set_user_memory_region(s, mem); | |
613 | if (err) { | |
614 | fprintf(stderr, "%s: error updating slot: %s\n", __func__, | |
615 | strerror(-err)); | |
616 | abort(); | |
617 | } | |
618 | ||
619 | start_addr += old.memory_size; | |
620 | ram += old.memory_size; | |
621 | size -= old.memory_size; | |
622 | continue; | |
623 | } | |
624 | ||
625 | /* register prefix slot */ | |
626 | if (old.start_addr < start_addr) { | |
627 | mem = kvm_alloc_slot(s); | |
628 | mem->memory_size = start_addr - old.start_addr; | |
629 | mem->start_addr = old.start_addr; | |
630 | mem->ram = old.ram; | |
631 | mem->flags = kvm_mem_flags(s, log_dirty); | |
632 | ||
633 | err = kvm_set_user_memory_region(s, mem); | |
634 | if (err) { | |
635 | fprintf(stderr, "%s: error registering prefix slot: %s\n", | |
636 | __func__, strerror(-err)); | |
637 | #ifdef TARGET_PPC | |
638 | fprintf(stderr, "%s: This is probably because your kernel's " \ | |
639 | "PAGE_SIZE is too big. Please try to use 4k " \ | |
640 | "PAGE_SIZE!\n", __func__); | |
641 | #endif | |
642 | abort(); | |
643 | } | |
644 | } | |
645 | ||
646 | /* register suffix slot */ | |
647 | if (old.start_addr + old.memory_size > start_addr + size) { | |
648 | ram_addr_t size_delta; | |
649 | ||
650 | mem = kvm_alloc_slot(s); | |
651 | mem->start_addr = start_addr + size; | |
652 | size_delta = mem->start_addr - old.start_addr; | |
653 | mem->memory_size = old.memory_size - size_delta; | |
654 | mem->ram = old.ram + size_delta; | |
655 | mem->flags = kvm_mem_flags(s, log_dirty); | |
656 | ||
657 | err = kvm_set_user_memory_region(s, mem); | |
658 | if (err) { | |
659 | fprintf(stderr, "%s: error registering suffix slot: %s\n", | |
660 | __func__, strerror(-err)); | |
661 | abort(); | |
662 | } | |
663 | } | |
664 | } | |
665 | ||
666 | /* in case the KVM bug workaround already "consumed" the new slot */ | |
667 | if (!size) { | |
668 | return; | |
669 | } | |
670 | if (!add) { | |
671 | return; | |
672 | } | |
673 | mem = kvm_alloc_slot(s); | |
674 | mem->memory_size = size; | |
675 | mem->start_addr = start_addr; | |
676 | mem->ram = ram; | |
677 | mem->flags = kvm_mem_flags(s, log_dirty); | |
678 | ||
679 | err = kvm_set_user_memory_region(s, mem); | |
680 | if (err) { | |
681 | fprintf(stderr, "%s: error registering slot: %s\n", __func__, | |
682 | strerror(-err)); | |
683 | abort(); | |
684 | } | |
685 | } | |
686 | ||
687 | static void kvm_begin(MemoryListener *listener) | |
688 | { | |
689 | } | |
690 | ||
691 | static void kvm_commit(MemoryListener *listener) | |
692 | { | |
693 | } | |
694 | ||
695 | static void kvm_region_add(MemoryListener *listener, | |
696 | MemoryRegionSection *section) | |
697 | { | |
698 | kvm_set_phys_mem(section, true); | |
699 | } | |
700 | ||
701 | static void kvm_region_del(MemoryListener *listener, | |
702 | MemoryRegionSection *section) | |
703 | { | |
704 | kvm_set_phys_mem(section, false); | |
705 | } | |
706 | ||
707 | static void kvm_region_nop(MemoryListener *listener, | |
708 | MemoryRegionSection *section) | |
709 | { | |
710 | } | |
711 | ||
712 | static void kvm_log_sync(MemoryListener *listener, | |
713 | MemoryRegionSection *section) | |
714 | { | |
715 | int r; | |
716 | ||
717 | r = kvm_physical_sync_dirty_bitmap(section); | |
718 | if (r < 0) { | |
719 | abort(); | |
720 | } | |
721 | } | |
722 | ||
723 | static void kvm_log_global_start(struct MemoryListener *listener) | |
724 | { | |
725 | int r; | |
726 | ||
727 | r = kvm_set_migration_log(1); | |
728 | assert(r >= 0); | |
729 | } | |
730 | ||
731 | static void kvm_log_global_stop(struct MemoryListener *listener) | |
732 | { | |
733 | int r; | |
734 | ||
735 | r = kvm_set_migration_log(0); | |
736 | assert(r >= 0); | |
737 | } | |
738 | ||
739 | static void kvm_mem_ioeventfd_add(MemoryRegionSection *section, | |
740 | bool match_data, uint64_t data, int fd) | |
741 | { | |
742 | int r; | |
743 | ||
744 | assert(match_data && section->size <= 8); | |
745 | ||
746 | r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space, | |
747 | data, true, section->size); | |
748 | if (r < 0) { | |
749 | abort(); | |
750 | } | |
751 | } | |
752 | ||
753 | static void kvm_mem_ioeventfd_del(MemoryRegionSection *section, | |
754 | bool match_data, uint64_t data, int fd) | |
755 | { | |
756 | int r; | |
757 | ||
758 | r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space, | |
759 | data, false, section->size); | |
760 | if (r < 0) { | |
761 | abort(); | |
762 | } | |
763 | } | |
764 | ||
765 | static void kvm_io_ioeventfd_add(MemoryRegionSection *section, | |
766 | bool match_data, uint64_t data, int fd) | |
767 | { | |
768 | int r; | |
769 | ||
770 | assert(match_data && section->size == 2); | |
771 | ||
772 | r = kvm_set_ioeventfd_pio_word(fd, section->offset_within_address_space, | |
773 | data, true); | |
774 | if (r < 0) { | |
775 | abort(); | |
776 | } | |
777 | } | |
778 | ||
779 | static void kvm_io_ioeventfd_del(MemoryRegionSection *section, | |
780 | bool match_data, uint64_t data, int fd) | |
781 | ||
782 | { | |
783 | int r; | |
784 | ||
785 | r = kvm_set_ioeventfd_pio_word(fd, section->offset_within_address_space, | |
786 | data, false); | |
787 | if (r < 0) { | |
788 | abort(); | |
789 | } | |
790 | } | |
791 | ||
792 | static void kvm_eventfd_add(MemoryListener *listener, | |
793 | MemoryRegionSection *section, | |
794 | bool match_data, uint64_t data, int fd) | |
795 | { | |
796 | if (section->address_space == get_system_memory()) { | |
797 | kvm_mem_ioeventfd_add(section, match_data, data, fd); | |
798 | } else { | |
799 | kvm_io_ioeventfd_add(section, match_data, data, fd); | |
800 | } | |
801 | } | |
802 | ||
803 | static void kvm_eventfd_del(MemoryListener *listener, | |
804 | MemoryRegionSection *section, | |
805 | bool match_data, uint64_t data, int fd) | |
806 | { | |
807 | if (section->address_space == get_system_memory()) { | |
808 | kvm_mem_ioeventfd_del(section, match_data, data, fd); | |
809 | } else { | |
810 | kvm_io_ioeventfd_del(section, match_data, data, fd); | |
811 | } | |
812 | } | |
813 | ||
814 | static MemoryListener kvm_memory_listener = { | |
815 | .begin = kvm_begin, | |
816 | .commit = kvm_commit, | |
817 | .region_add = kvm_region_add, | |
818 | .region_del = kvm_region_del, | |
819 | .region_nop = kvm_region_nop, | |
820 | .log_start = kvm_log_start, | |
821 | .log_stop = kvm_log_stop, | |
822 | .log_sync = kvm_log_sync, | |
823 | .log_global_start = kvm_log_global_start, | |
824 | .log_global_stop = kvm_log_global_stop, | |
825 | .eventfd_add = kvm_eventfd_add, | |
826 | .eventfd_del = kvm_eventfd_del, | |
827 | .priority = 10, | |
828 | }; | |
829 | ||
830 | static void kvm_handle_interrupt(CPUArchState *env, int mask) | |
831 | { | |
832 | env->interrupt_request |= mask; | |
833 | ||
834 | if (!qemu_cpu_is_self(env)) { | |
835 | qemu_cpu_kick(env); | |
836 | } | |
837 | } | |
838 | ||
839 | int kvm_irqchip_set_irq(KVMState *s, int irq, int level) | |
840 | { | |
841 | struct kvm_irq_level event; | |
842 | int ret; | |
843 | ||
844 | assert(kvm_irqchip_in_kernel()); | |
845 | ||
846 | event.level = level; | |
847 | event.irq = irq; | |
848 | ret = kvm_vm_ioctl(s, s->irqchip_inject_ioctl, &event); | |
849 | if (ret < 0) { | |
850 | perror("kvm_set_irqchip_line"); | |
851 | abort(); | |
852 | } | |
853 | ||
854 | return (s->irqchip_inject_ioctl == KVM_IRQ_LINE) ? 1 : event.status; | |
855 | } | |
856 | ||
857 | #ifdef KVM_CAP_IRQ_ROUTING | |
858 | static void set_gsi(KVMState *s, unsigned int gsi) | |
859 | { | |
860 | assert(gsi < s->max_gsi); | |
861 | ||
862 | s->used_gsi_bitmap[gsi / 32] |= 1U << (gsi % 32); | |
863 | } | |
864 | ||
865 | static void kvm_init_irq_routing(KVMState *s) | |
866 | { | |
867 | int gsi_count; | |
868 | ||
869 | gsi_count = kvm_check_extension(s, KVM_CAP_IRQ_ROUTING); | |
870 | if (gsi_count > 0) { | |
871 | unsigned int gsi_bits, i; | |
872 | ||
873 | /* Round up so we can search ints using ffs */ | |
874 | gsi_bits = ALIGN(gsi_count, 32); | |
875 | s->used_gsi_bitmap = g_malloc0(gsi_bits / 8); | |
876 | s->max_gsi = gsi_bits; | |
877 | ||
878 | /* Mark any over-allocated bits as already in use */ | |
879 | for (i = gsi_count; i < gsi_bits; i++) { | |
880 | set_gsi(s, i); | |
881 | } | |
882 | } | |
883 | ||
884 | s->irq_routes = g_malloc0(sizeof(*s->irq_routes)); | |
885 | s->nr_allocated_irq_routes = 0; | |
886 | ||
887 | kvm_arch_init_irq_routing(s); | |
888 | } | |
889 | ||
890 | static void kvm_add_routing_entry(KVMState *s, | |
891 | struct kvm_irq_routing_entry *entry) | |
892 | { | |
893 | struct kvm_irq_routing_entry *new; | |
894 | int n, size; | |
895 | ||
896 | if (s->irq_routes->nr == s->nr_allocated_irq_routes) { | |
897 | n = s->nr_allocated_irq_routes * 2; | |
898 | if (n < 64) { | |
899 | n = 64; | |
900 | } | |
901 | size = sizeof(struct kvm_irq_routing); | |
902 | size += n * sizeof(*new); | |
903 | s->irq_routes = g_realloc(s->irq_routes, size); | |
904 | s->nr_allocated_irq_routes = n; | |
905 | } | |
906 | n = s->irq_routes->nr++; | |
907 | new = &s->irq_routes->entries[n]; | |
908 | memset(new, 0, sizeof(*new)); | |
909 | new->gsi = entry->gsi; | |
910 | new->type = entry->type; | |
911 | new->flags = entry->flags; | |
912 | new->u = entry->u; | |
913 | ||
914 | set_gsi(s, entry->gsi); | |
915 | } | |
916 | ||
917 | void kvm_irqchip_add_route(KVMState *s, int irq, int irqchip, int pin) | |
918 | { | |
919 | struct kvm_irq_routing_entry e; | |
920 | ||
921 | e.gsi = irq; | |
922 | e.type = KVM_IRQ_ROUTING_IRQCHIP; | |
923 | e.flags = 0; | |
924 | e.u.irqchip.irqchip = irqchip; | |
925 | e.u.irqchip.pin = pin; | |
926 | kvm_add_routing_entry(s, &e); | |
927 | } | |
928 | ||
929 | int kvm_irqchip_commit_routes(KVMState *s) | |
930 | { | |
931 | s->irq_routes->flags = 0; | |
932 | return kvm_vm_ioctl(s, KVM_SET_GSI_ROUTING, s->irq_routes); | |
933 | } | |
934 | ||
935 | #else /* !KVM_CAP_IRQ_ROUTING */ | |
936 | ||
937 | static void kvm_init_irq_routing(KVMState *s) | |
938 | { | |
939 | } | |
940 | #endif /* !KVM_CAP_IRQ_ROUTING */ | |
941 | ||
942 | static int kvm_irqchip_create(KVMState *s) | |
943 | { | |
944 | QemuOptsList *list = qemu_find_opts("machine"); | |
945 | int ret; | |
946 | ||
947 | if (QTAILQ_EMPTY(&list->head) || | |
948 | !qemu_opt_get_bool(QTAILQ_FIRST(&list->head), | |
949 | "kernel_irqchip", false) || | |
950 | !kvm_check_extension(s, KVM_CAP_IRQCHIP)) { | |
951 | return 0; | |
952 | } | |
953 | ||
954 | ret = kvm_vm_ioctl(s, KVM_CREATE_IRQCHIP); | |
955 | if (ret < 0) { | |
956 | fprintf(stderr, "Create kernel irqchip failed\n"); | |
957 | return ret; | |
958 | } | |
959 | ||
960 | s->irqchip_inject_ioctl = KVM_IRQ_LINE; | |
961 | if (kvm_check_extension(s, KVM_CAP_IRQ_INJECT_STATUS)) { | |
962 | s->irqchip_inject_ioctl = KVM_IRQ_LINE_STATUS; | |
963 | } | |
964 | kvm_kernel_irqchip = true; | |
965 | ||
966 | kvm_init_irq_routing(s); | |
967 | ||
968 | return 0; | |
969 | } | |
970 | ||
971 | int kvm_init(void) | |
972 | { | |
973 | static const char upgrade_note[] = | |
974 | "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n" | |
975 | "(see http://sourceforge.net/projects/kvm).\n"; | |
976 | KVMState *s; | |
977 | const KVMCapabilityInfo *missing_cap; | |
978 | int ret; | |
979 | int i; | |
980 | ||
981 | s = g_malloc0(sizeof(KVMState)); | |
982 | ||
983 | #ifdef KVM_CAP_SET_GUEST_DEBUG | |
984 | QTAILQ_INIT(&s->kvm_sw_breakpoints); | |
985 | #endif | |
986 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
987 | s->slots[i].slot = i; | |
988 | } | |
989 | s->vmfd = -1; | |
990 | s->fd = qemu_open("/dev/kvm", O_RDWR); | |
991 | if (s->fd == -1) { | |
992 | fprintf(stderr, "Could not access KVM kernel module: %m\n"); | |
993 | ret = -errno; | |
994 | goto err; | |
995 | } | |
996 | ||
997 | ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0); | |
998 | if (ret < KVM_API_VERSION) { | |
999 | if (ret > 0) { | |
1000 | ret = -EINVAL; | |
1001 | } | |
1002 | fprintf(stderr, "kvm version too old\n"); | |
1003 | goto err; | |
1004 | } | |
1005 | ||
1006 | if (ret > KVM_API_VERSION) { | |
1007 | ret = -EINVAL; | |
1008 | fprintf(stderr, "kvm version not supported\n"); | |
1009 | goto err; | |
1010 | } | |
1011 | ||
1012 | s->vmfd = kvm_ioctl(s, KVM_CREATE_VM, 0); | |
1013 | if (s->vmfd < 0) { | |
1014 | #ifdef TARGET_S390X | |
1015 | fprintf(stderr, "Please add the 'switch_amode' kernel parameter to " | |
1016 | "your host kernel command line\n"); | |
1017 | #endif | |
1018 | ret = s->vmfd; | |
1019 | goto err; | |
1020 | } | |
1021 | ||
1022 | missing_cap = kvm_check_extension_list(s, kvm_required_capabilites); | |
1023 | if (!missing_cap) { | |
1024 | missing_cap = | |
1025 | kvm_check_extension_list(s, kvm_arch_required_capabilities); | |
1026 | } | |
1027 | if (missing_cap) { | |
1028 | ret = -EINVAL; | |
1029 | fprintf(stderr, "kvm does not support %s\n%s", | |
1030 | missing_cap->name, upgrade_note); | |
1031 | goto err; | |
1032 | } | |
1033 | ||
1034 | s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO); | |
1035 | ||
1036 | s->broken_set_mem_region = 1; | |
1037 | ret = kvm_check_extension(s, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS); | |
1038 | if (ret > 0) { | |
1039 | s->broken_set_mem_region = 0; | |
1040 | } | |
1041 | ||
1042 | #ifdef KVM_CAP_VCPU_EVENTS | |
1043 | s->vcpu_events = kvm_check_extension(s, KVM_CAP_VCPU_EVENTS); | |
1044 | #endif | |
1045 | ||
1046 | s->robust_singlestep = | |
1047 | kvm_check_extension(s, KVM_CAP_X86_ROBUST_SINGLESTEP); | |
1048 | ||
1049 | #ifdef KVM_CAP_DEBUGREGS | |
1050 | s->debugregs = kvm_check_extension(s, KVM_CAP_DEBUGREGS); | |
1051 | #endif | |
1052 | ||
1053 | #ifdef KVM_CAP_XSAVE | |
1054 | s->xsave = kvm_check_extension(s, KVM_CAP_XSAVE); | |
1055 | #endif | |
1056 | ||
1057 | #ifdef KVM_CAP_XCRS | |
1058 | s->xcrs = kvm_check_extension(s, KVM_CAP_XCRS); | |
1059 | #endif | |
1060 | ||
1061 | #ifdef KVM_CAP_PIT_STATE2 | |
1062 | s->pit_state2 = kvm_check_extension(s, KVM_CAP_PIT_STATE2); | |
1063 | #endif | |
1064 | ||
1065 | ret = kvm_arch_init(s); | |
1066 | if (ret < 0) { | |
1067 | goto err; | |
1068 | } | |
1069 | ||
1070 | ret = kvm_irqchip_create(s); | |
1071 | if (ret < 0) { | |
1072 | goto err; | |
1073 | } | |
1074 | ||
1075 | kvm_state = s; | |
1076 | memory_listener_register(&kvm_memory_listener, NULL); | |
1077 | ||
1078 | s->many_ioeventfds = kvm_check_many_ioeventfds(); | |
1079 | ||
1080 | cpu_interrupt_handler = kvm_handle_interrupt; | |
1081 | ||
1082 | return 0; | |
1083 | ||
1084 | err: | |
1085 | if (s) { | |
1086 | if (s->vmfd >= 0) { | |
1087 | close(s->vmfd); | |
1088 | } | |
1089 | if (s->fd != -1) { | |
1090 | close(s->fd); | |
1091 | } | |
1092 | } | |
1093 | g_free(s); | |
1094 | ||
1095 | return ret; | |
1096 | } | |
1097 | ||
1098 | static void kvm_handle_io(uint16_t port, void *data, int direction, int size, | |
1099 | uint32_t count) | |
1100 | { | |
1101 | int i; | |
1102 | uint8_t *ptr = data; | |
1103 | ||
1104 | for (i = 0; i < count; i++) { | |
1105 | if (direction == KVM_EXIT_IO_IN) { | |
1106 | switch (size) { | |
1107 | case 1: | |
1108 | stb_p(ptr, cpu_inb(port)); | |
1109 | break; | |
1110 | case 2: | |
1111 | stw_p(ptr, cpu_inw(port)); | |
1112 | break; | |
1113 | case 4: | |
1114 | stl_p(ptr, cpu_inl(port)); | |
1115 | break; | |
1116 | } | |
1117 | } else { | |
1118 | switch (size) { | |
1119 | case 1: | |
1120 | cpu_outb(port, ldub_p(ptr)); | |
1121 | break; | |
1122 | case 2: | |
1123 | cpu_outw(port, lduw_p(ptr)); | |
1124 | break; | |
1125 | case 4: | |
1126 | cpu_outl(port, ldl_p(ptr)); | |
1127 | break; | |
1128 | } | |
1129 | } | |
1130 | ||
1131 | ptr += size; | |
1132 | } | |
1133 | } | |
1134 | ||
1135 | static int kvm_handle_internal_error(CPUArchState *env, struct kvm_run *run) | |
1136 | { | |
1137 | fprintf(stderr, "KVM internal error."); | |
1138 | if (kvm_check_extension(kvm_state, KVM_CAP_INTERNAL_ERROR_DATA)) { | |
1139 | int i; | |
1140 | ||
1141 | fprintf(stderr, " Suberror: %d\n", run->internal.suberror); | |
1142 | for (i = 0; i < run->internal.ndata; ++i) { | |
1143 | fprintf(stderr, "extra data[%d]: %"PRIx64"\n", | |
1144 | i, (uint64_t)run->internal.data[i]); | |
1145 | } | |
1146 | } else { | |
1147 | fprintf(stderr, "\n"); | |
1148 | } | |
1149 | if (run->internal.suberror == KVM_INTERNAL_ERROR_EMULATION) { | |
1150 | fprintf(stderr, "emulation failure\n"); | |
1151 | if (!kvm_arch_stop_on_emulation_error(env)) { | |
1152 | cpu_dump_state(env, stderr, fprintf, CPU_DUMP_CODE); | |
1153 | return EXCP_INTERRUPT; | |
1154 | } | |
1155 | } | |
1156 | /* FIXME: Should trigger a qmp message to let management know | |
1157 | * something went wrong. | |
1158 | */ | |
1159 | return -1; | |
1160 | } | |
1161 | ||
1162 | void kvm_flush_coalesced_mmio_buffer(void) | |
1163 | { | |
1164 | KVMState *s = kvm_state; | |
1165 | ||
1166 | if (s->coalesced_flush_in_progress) { | |
1167 | return; | |
1168 | } | |
1169 | ||
1170 | s->coalesced_flush_in_progress = true; | |
1171 | ||
1172 | if (s->coalesced_mmio_ring) { | |
1173 | struct kvm_coalesced_mmio_ring *ring = s->coalesced_mmio_ring; | |
1174 | while (ring->first != ring->last) { | |
1175 | struct kvm_coalesced_mmio *ent; | |
1176 | ||
1177 | ent = &ring->coalesced_mmio[ring->first]; | |
1178 | ||
1179 | cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len); | |
1180 | smp_wmb(); | |
1181 | ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX; | |
1182 | } | |
1183 | } | |
1184 | ||
1185 | s->coalesced_flush_in_progress = false; | |
1186 | } | |
1187 | ||
1188 | static void do_kvm_cpu_synchronize_state(void *_env) | |
1189 | { | |
1190 | CPUArchState *env = _env; | |
1191 | ||
1192 | if (!env->kvm_vcpu_dirty) { | |
1193 | kvm_arch_get_registers(env); | |
1194 | env->kvm_vcpu_dirty = 1; | |
1195 | } | |
1196 | } | |
1197 | ||
1198 | void kvm_cpu_synchronize_state(CPUArchState *env) | |
1199 | { | |
1200 | if (!env->kvm_vcpu_dirty) { | |
1201 | run_on_cpu(env, do_kvm_cpu_synchronize_state, env); | |
1202 | } | |
1203 | } | |
1204 | ||
1205 | void kvm_cpu_synchronize_post_reset(CPUArchState *env) | |
1206 | { | |
1207 | kvm_arch_put_registers(env, KVM_PUT_RESET_STATE); | |
1208 | env->kvm_vcpu_dirty = 0; | |
1209 | } | |
1210 | ||
1211 | void kvm_cpu_synchronize_post_init(CPUArchState *env) | |
1212 | { | |
1213 | kvm_arch_put_registers(env, KVM_PUT_FULL_STATE); | |
1214 | env->kvm_vcpu_dirty = 0; | |
1215 | } | |
1216 | ||
1217 | int kvm_cpu_exec(CPUArchState *env) | |
1218 | { | |
1219 | struct kvm_run *run = env->kvm_run; | |
1220 | int ret, run_ret; | |
1221 | ||
1222 | DPRINTF("kvm_cpu_exec()\n"); | |
1223 | ||
1224 | if (kvm_arch_process_async_events(env)) { | |
1225 | env->exit_request = 0; | |
1226 | return EXCP_HLT; | |
1227 | } | |
1228 | ||
1229 | do { | |
1230 | if (env->kvm_vcpu_dirty) { | |
1231 | kvm_arch_put_registers(env, KVM_PUT_RUNTIME_STATE); | |
1232 | env->kvm_vcpu_dirty = 0; | |
1233 | } | |
1234 | ||
1235 | kvm_arch_pre_run(env, run); | |
1236 | if (env->exit_request) { | |
1237 | DPRINTF("interrupt exit requested\n"); | |
1238 | /* | |
1239 | * KVM requires us to reenter the kernel after IO exits to complete | |
1240 | * instruction emulation. This self-signal will ensure that we | |
1241 | * leave ASAP again. | |
1242 | */ | |
1243 | qemu_cpu_kick_self(); | |
1244 | } | |
1245 | qemu_mutex_unlock_iothread(); | |
1246 | ||
1247 | run_ret = kvm_vcpu_ioctl(env, KVM_RUN, 0); | |
1248 | ||
1249 | qemu_mutex_lock_iothread(); | |
1250 | kvm_arch_post_run(env, run); | |
1251 | ||
1252 | kvm_flush_coalesced_mmio_buffer(); | |
1253 | ||
1254 | if (run_ret < 0) { | |
1255 | if (run_ret == -EINTR || run_ret == -EAGAIN) { | |
1256 | DPRINTF("io window exit\n"); | |
1257 | ret = EXCP_INTERRUPT; | |
1258 | break; | |
1259 | } | |
1260 | fprintf(stderr, "error: kvm run failed %s\n", | |
1261 | strerror(-run_ret)); | |
1262 | abort(); | |
1263 | } | |
1264 | ||
1265 | switch (run->exit_reason) { | |
1266 | case KVM_EXIT_IO: | |
1267 | DPRINTF("handle_io\n"); | |
1268 | kvm_handle_io(run->io.port, | |
1269 | (uint8_t *)run + run->io.data_offset, | |
1270 | run->io.direction, | |
1271 | run->io.size, | |
1272 | run->io.count); | |
1273 | ret = 0; | |
1274 | break; | |
1275 | case KVM_EXIT_MMIO: | |
1276 | DPRINTF("handle_mmio\n"); | |
1277 | cpu_physical_memory_rw(run->mmio.phys_addr, | |
1278 | run->mmio.data, | |
1279 | run->mmio.len, | |
1280 | run->mmio.is_write); | |
1281 | ret = 0; | |
1282 | break; | |
1283 | case KVM_EXIT_IRQ_WINDOW_OPEN: | |
1284 | DPRINTF("irq_window_open\n"); | |
1285 | ret = EXCP_INTERRUPT; | |
1286 | break; | |
1287 | case KVM_EXIT_SHUTDOWN: | |
1288 | DPRINTF("shutdown\n"); | |
1289 | qemu_system_reset_request(); | |
1290 | ret = EXCP_INTERRUPT; | |
1291 | break; | |
1292 | case KVM_EXIT_UNKNOWN: | |
1293 | fprintf(stderr, "KVM: unknown exit, hardware reason %" PRIx64 "\n", | |
1294 | (uint64_t)run->hw.hardware_exit_reason); | |
1295 | ret = -1; | |
1296 | break; | |
1297 | case KVM_EXIT_INTERNAL_ERROR: | |
1298 | ret = kvm_handle_internal_error(env, run); | |
1299 | break; | |
1300 | default: | |
1301 | DPRINTF("kvm_arch_handle_exit\n"); | |
1302 | ret = kvm_arch_handle_exit(env, run); | |
1303 | break; | |
1304 | } | |
1305 | } while (ret == 0); | |
1306 | ||
1307 | if (ret < 0) { | |
1308 | cpu_dump_state(env, stderr, fprintf, CPU_DUMP_CODE); | |
1309 | vm_stop(RUN_STATE_INTERNAL_ERROR); | |
1310 | } | |
1311 | ||
1312 | env->exit_request = 0; | |
1313 | return ret; | |
1314 | } | |
1315 | ||
1316 | int kvm_ioctl(KVMState *s, int type, ...) | |
1317 | { | |
1318 | int ret; | |
1319 | void *arg; | |
1320 | va_list ap; | |
1321 | ||
1322 | va_start(ap, type); | |
1323 | arg = va_arg(ap, void *); | |
1324 | va_end(ap); | |
1325 | ||
1326 | ret = ioctl(s->fd, type, arg); | |
1327 | if (ret == -1) { | |
1328 | ret = -errno; | |
1329 | } | |
1330 | return ret; | |
1331 | } | |
1332 | ||
1333 | int kvm_vm_ioctl(KVMState *s, int type, ...) | |
1334 | { | |
1335 | int ret; | |
1336 | void *arg; | |
1337 | va_list ap; | |
1338 | ||
1339 | va_start(ap, type); | |
1340 | arg = va_arg(ap, void *); | |
1341 | va_end(ap); | |
1342 | ||
1343 | ret = ioctl(s->vmfd, type, arg); | |
1344 | if (ret == -1) { | |
1345 | ret = -errno; | |
1346 | } | |
1347 | return ret; | |
1348 | } | |
1349 | ||
1350 | int kvm_vcpu_ioctl(CPUArchState *env, int type, ...) | |
1351 | { | |
1352 | int ret; | |
1353 | void *arg; | |
1354 | va_list ap; | |
1355 | ||
1356 | va_start(ap, type); | |
1357 | arg = va_arg(ap, void *); | |
1358 | va_end(ap); | |
1359 | ||
1360 | ret = ioctl(env->kvm_fd, type, arg); | |
1361 | if (ret == -1) { | |
1362 | ret = -errno; | |
1363 | } | |
1364 | return ret; | |
1365 | } | |
1366 | ||
1367 | int kvm_has_sync_mmu(void) | |
1368 | { | |
1369 | return kvm_check_extension(kvm_state, KVM_CAP_SYNC_MMU); | |
1370 | } | |
1371 | ||
1372 | int kvm_has_vcpu_events(void) | |
1373 | { | |
1374 | return kvm_state->vcpu_events; | |
1375 | } | |
1376 | ||
1377 | int kvm_has_robust_singlestep(void) | |
1378 | { | |
1379 | return kvm_state->robust_singlestep; | |
1380 | } | |
1381 | ||
1382 | int kvm_has_debugregs(void) | |
1383 | { | |
1384 | return kvm_state->debugregs; | |
1385 | } | |
1386 | ||
1387 | int kvm_has_xsave(void) | |
1388 | { | |
1389 | return kvm_state->xsave; | |
1390 | } | |
1391 | ||
1392 | int kvm_has_xcrs(void) | |
1393 | { | |
1394 | return kvm_state->xcrs; | |
1395 | } | |
1396 | ||
1397 | int kvm_has_pit_state2(void) | |
1398 | { | |
1399 | return kvm_state->pit_state2; | |
1400 | } | |
1401 | ||
1402 | int kvm_has_many_ioeventfds(void) | |
1403 | { | |
1404 | if (!kvm_enabled()) { | |
1405 | return 0; | |
1406 | } | |
1407 | return kvm_state->many_ioeventfds; | |
1408 | } | |
1409 | ||
1410 | int kvm_has_gsi_routing(void) | |
1411 | { | |
1412 | #ifdef KVM_CAP_IRQ_ROUTING | |
1413 | return kvm_check_extension(kvm_state, KVM_CAP_IRQ_ROUTING); | |
1414 | #else | |
1415 | return false; | |
1416 | #endif | |
1417 | } | |
1418 | ||
1419 | int kvm_allows_irq0_override(void) | |
1420 | { | |
1421 | return !kvm_irqchip_in_kernel() || kvm_has_gsi_routing(); | |
1422 | } | |
1423 | ||
1424 | void kvm_setup_guest_memory(void *start, size_t size) | |
1425 | { | |
1426 | if (!kvm_has_sync_mmu()) { | |
1427 | int ret = qemu_madvise(start, size, QEMU_MADV_DONTFORK); | |
1428 | ||
1429 | if (ret) { | |
1430 | perror("qemu_madvise"); | |
1431 | fprintf(stderr, | |
1432 | "Need MADV_DONTFORK in absence of synchronous KVM MMU\n"); | |
1433 | exit(1); | |
1434 | } | |
1435 | } | |
1436 | } | |
1437 | ||
1438 | #ifdef KVM_CAP_SET_GUEST_DEBUG | |
1439 | struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUArchState *env, | |
1440 | target_ulong pc) | |
1441 | { | |
1442 | struct kvm_sw_breakpoint *bp; | |
1443 | ||
1444 | QTAILQ_FOREACH(bp, &env->kvm_state->kvm_sw_breakpoints, entry) { | |
1445 | if (bp->pc == pc) { | |
1446 | return bp; | |
1447 | } | |
1448 | } | |
1449 | return NULL; | |
1450 | } | |
1451 | ||
1452 | int kvm_sw_breakpoints_active(CPUArchState *env) | |
1453 | { | |
1454 | return !QTAILQ_EMPTY(&env->kvm_state->kvm_sw_breakpoints); | |
1455 | } | |
1456 | ||
1457 | struct kvm_set_guest_debug_data { | |
1458 | struct kvm_guest_debug dbg; | |
1459 | CPUArchState *env; | |
1460 | int err; | |
1461 | }; | |
1462 | ||
1463 | static void kvm_invoke_set_guest_debug(void *data) | |
1464 | { | |
1465 | struct kvm_set_guest_debug_data *dbg_data = data; | |
1466 | CPUArchState *env = dbg_data->env; | |
1467 | ||
1468 | dbg_data->err = kvm_vcpu_ioctl(env, KVM_SET_GUEST_DEBUG, &dbg_data->dbg); | |
1469 | } | |
1470 | ||
1471 | int kvm_update_guest_debug(CPUArchState *env, unsigned long reinject_trap) | |
1472 | { | |
1473 | struct kvm_set_guest_debug_data data; | |
1474 | ||
1475 | data.dbg.control = reinject_trap; | |
1476 | ||
1477 | if (env->singlestep_enabled) { | |
1478 | data.dbg.control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP; | |
1479 | } | |
1480 | kvm_arch_update_guest_debug(env, &data.dbg); | |
1481 | data.env = env; | |
1482 | ||
1483 | run_on_cpu(env, kvm_invoke_set_guest_debug, &data); | |
1484 | return data.err; | |
1485 | } | |
1486 | ||
1487 | int kvm_insert_breakpoint(CPUArchState *current_env, target_ulong addr, | |
1488 | target_ulong len, int type) | |
1489 | { | |
1490 | struct kvm_sw_breakpoint *bp; | |
1491 | CPUArchState *env; | |
1492 | int err; | |
1493 | ||
1494 | if (type == GDB_BREAKPOINT_SW) { | |
1495 | bp = kvm_find_sw_breakpoint(current_env, addr); | |
1496 | if (bp) { | |
1497 | bp->use_count++; | |
1498 | return 0; | |
1499 | } | |
1500 | ||
1501 | bp = g_malloc(sizeof(struct kvm_sw_breakpoint)); | |
1502 | if (!bp) { | |
1503 | return -ENOMEM; | |
1504 | } | |
1505 | ||
1506 | bp->pc = addr; | |
1507 | bp->use_count = 1; | |
1508 | err = kvm_arch_insert_sw_breakpoint(current_env, bp); | |
1509 | if (err) { | |
1510 | g_free(bp); | |
1511 | return err; | |
1512 | } | |
1513 | ||
1514 | QTAILQ_INSERT_HEAD(¤t_env->kvm_state->kvm_sw_breakpoints, | |
1515 | bp, entry); | |
1516 | } else { | |
1517 | err = kvm_arch_insert_hw_breakpoint(addr, len, type); | |
1518 | if (err) { | |
1519 | return err; | |
1520 | } | |
1521 | } | |
1522 | ||
1523 | for (env = first_cpu; env != NULL; env = env->next_cpu) { | |
1524 | err = kvm_update_guest_debug(env, 0); | |
1525 | if (err) { | |
1526 | return err; | |
1527 | } | |
1528 | } | |
1529 | return 0; | |
1530 | } | |
1531 | ||
1532 | int kvm_remove_breakpoint(CPUArchState *current_env, target_ulong addr, | |
1533 | target_ulong len, int type) | |
1534 | { | |
1535 | struct kvm_sw_breakpoint *bp; | |
1536 | CPUArchState *env; | |
1537 | int err; | |
1538 | ||
1539 | if (type == GDB_BREAKPOINT_SW) { | |
1540 | bp = kvm_find_sw_breakpoint(current_env, addr); | |
1541 | if (!bp) { | |
1542 | return -ENOENT; | |
1543 | } | |
1544 | ||
1545 | if (bp->use_count > 1) { | |
1546 | bp->use_count--; | |
1547 | return 0; | |
1548 | } | |
1549 | ||
1550 | err = kvm_arch_remove_sw_breakpoint(current_env, bp); | |
1551 | if (err) { | |
1552 | return err; | |
1553 | } | |
1554 | ||
1555 | QTAILQ_REMOVE(¤t_env->kvm_state->kvm_sw_breakpoints, bp, entry); | |
1556 | g_free(bp); | |
1557 | } else { | |
1558 | err = kvm_arch_remove_hw_breakpoint(addr, len, type); | |
1559 | if (err) { | |
1560 | return err; | |
1561 | } | |
1562 | } | |
1563 | ||
1564 | for (env = first_cpu; env != NULL; env = env->next_cpu) { | |
1565 | err = kvm_update_guest_debug(env, 0); | |
1566 | if (err) { | |
1567 | return err; | |
1568 | } | |
1569 | } | |
1570 | return 0; | |
1571 | } | |
1572 | ||
1573 | void kvm_remove_all_breakpoints(CPUArchState *current_env) | |
1574 | { | |
1575 | struct kvm_sw_breakpoint *bp, *next; | |
1576 | KVMState *s = current_env->kvm_state; | |
1577 | CPUArchState *env; | |
1578 | ||
1579 | QTAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) { | |
1580 | if (kvm_arch_remove_sw_breakpoint(current_env, bp) != 0) { | |
1581 | /* Try harder to find a CPU that currently sees the breakpoint. */ | |
1582 | for (env = first_cpu; env != NULL; env = env->next_cpu) { | |
1583 | if (kvm_arch_remove_sw_breakpoint(env, bp) == 0) { | |
1584 | break; | |
1585 | } | |
1586 | } | |
1587 | } | |
1588 | } | |
1589 | kvm_arch_remove_all_hw_breakpoints(); | |
1590 | ||
1591 | for (env = first_cpu; env != NULL; env = env->next_cpu) { | |
1592 | kvm_update_guest_debug(env, 0); | |
1593 | } | |
1594 | } | |
1595 | ||
1596 | #else /* !KVM_CAP_SET_GUEST_DEBUG */ | |
1597 | ||
1598 | int kvm_update_guest_debug(CPUArchState *env, unsigned long reinject_trap) | |
1599 | { | |
1600 | return -EINVAL; | |
1601 | } | |
1602 | ||
1603 | int kvm_insert_breakpoint(CPUArchState *current_env, target_ulong addr, | |
1604 | target_ulong len, int type) | |
1605 | { | |
1606 | return -EINVAL; | |
1607 | } | |
1608 | ||
1609 | int kvm_remove_breakpoint(CPUArchState *current_env, target_ulong addr, | |
1610 | target_ulong len, int type) | |
1611 | { | |
1612 | return -EINVAL; | |
1613 | } | |
1614 | ||
1615 | void kvm_remove_all_breakpoints(CPUArchState *current_env) | |
1616 | { | |
1617 | } | |
1618 | #endif /* !KVM_CAP_SET_GUEST_DEBUG */ | |
1619 | ||
1620 | int kvm_set_signal_mask(CPUArchState *env, const sigset_t *sigset) | |
1621 | { | |
1622 | struct kvm_signal_mask *sigmask; | |
1623 | int r; | |
1624 | ||
1625 | if (!sigset) { | |
1626 | return kvm_vcpu_ioctl(env, KVM_SET_SIGNAL_MASK, NULL); | |
1627 | } | |
1628 | ||
1629 | sigmask = g_malloc(sizeof(*sigmask) + sizeof(*sigset)); | |
1630 | ||
1631 | sigmask->len = 8; | |
1632 | memcpy(sigmask->sigset, sigset, sizeof(*sigset)); | |
1633 | r = kvm_vcpu_ioctl(env, KVM_SET_SIGNAL_MASK, sigmask); | |
1634 | g_free(sigmask); | |
1635 | ||
1636 | return r; | |
1637 | } | |
1638 | ||
1639 | int kvm_set_ioeventfd_mmio(int fd, uint32_t addr, uint32_t val, bool assign, | |
1640 | uint32_t size) | |
1641 | { | |
1642 | int ret; | |
1643 | struct kvm_ioeventfd iofd; | |
1644 | ||
1645 | iofd.datamatch = val; | |
1646 | iofd.addr = addr; | |
1647 | iofd.len = size; | |
1648 | iofd.flags = KVM_IOEVENTFD_FLAG_DATAMATCH; | |
1649 | iofd.fd = fd; | |
1650 | ||
1651 | if (!kvm_enabled()) { | |
1652 | return -ENOSYS; | |
1653 | } | |
1654 | ||
1655 | if (!assign) { | |
1656 | iofd.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN; | |
1657 | } | |
1658 | ||
1659 | ret = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &iofd); | |
1660 | ||
1661 | if (ret < 0) { | |
1662 | return -errno; | |
1663 | } | |
1664 | ||
1665 | return 0; | |
1666 | } | |
1667 | ||
1668 | int kvm_set_ioeventfd_pio_word(int fd, uint16_t addr, uint16_t val, bool assign) | |
1669 | { | |
1670 | struct kvm_ioeventfd kick = { | |
1671 | .datamatch = val, | |
1672 | .addr = addr, | |
1673 | .len = 2, | |
1674 | .flags = KVM_IOEVENTFD_FLAG_DATAMATCH | KVM_IOEVENTFD_FLAG_PIO, | |
1675 | .fd = fd, | |
1676 | }; | |
1677 | int r; | |
1678 | if (!kvm_enabled()) { | |
1679 | return -ENOSYS; | |
1680 | } | |
1681 | if (!assign) { | |
1682 | kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN; | |
1683 | } | |
1684 | r = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick); | |
1685 | if (r < 0) { | |
1686 | return r; | |
1687 | } | |
1688 | return 0; | |
1689 | } | |
1690 | ||
1691 | int kvm_on_sigbus_vcpu(CPUArchState *env, int code, void *addr) | |
1692 | { | |
1693 | return kvm_arch_on_sigbus_vcpu(env, code, addr); | |
1694 | } | |
1695 | ||
1696 | int kvm_on_sigbus(int code, void *addr) | |
1697 | { | |
1698 | return kvm_arch_on_sigbus(code, addr); | |
1699 | } |