]>
Commit | Line | Data |
---|---|---|
05330448 AL |
1 | /* |
2 | * QEMU KVM support | |
3 | * | |
4 | * Copyright IBM, Corp. 2008 | |
5832d1f2 | 5 | * Red Hat, Inc. 2008 |
05330448 AL |
6 | * |
7 | * Authors: | |
8 | * Anthony Liguori <aliguori@us.ibm.com> | |
5832d1f2 | 9 | * Glauber Costa <gcosta@redhat.com> |
05330448 AL |
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> | |
984b5181 | 19 | #include <stdarg.h> |
05330448 AL |
20 | |
21 | #include <linux/kvm.h> | |
22 | ||
23 | #include "qemu-common.h" | |
85199474 | 24 | #include "qemu-barrier.h" |
05330448 | 25 | #include "sysemu.h" |
d33a1810 | 26 | #include "hw/hw.h" |
e22a25c9 | 27 | #include "gdbstub.h" |
05330448 | 28 | #include "kvm.h" |
8369e01c | 29 | #include "bswap.h" |
05330448 | 30 | |
d2f2b8a7 SH |
31 | /* This check must be after config-host.h is included */ |
32 | #ifdef CONFIG_EVENTFD | |
33 | #include <sys/eventfd.h> | |
34 | #endif | |
35 | ||
f65ed4c1 AL |
36 | /* KVM uses PAGE_SIZE in it's definition of COALESCED_MMIO_MAX */ |
37 | #define PAGE_SIZE TARGET_PAGE_SIZE | |
38 | ||
05330448 AL |
39 | //#define DEBUG_KVM |
40 | ||
41 | #ifdef DEBUG_KVM | |
8c0d577e | 42 | #define DPRINTF(fmt, ...) \ |
05330448 AL |
43 | do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0) |
44 | #else | |
8c0d577e | 45 | #define DPRINTF(fmt, ...) \ |
05330448 AL |
46 | do { } while (0) |
47 | #endif | |
48 | ||
34fc643f AL |
49 | typedef struct KVMSlot |
50 | { | |
c227f099 AL |
51 | target_phys_addr_t start_addr; |
52 | ram_addr_t memory_size; | |
53 | ram_addr_t phys_offset; | |
34fc643f AL |
54 | int slot; |
55 | int flags; | |
56 | } KVMSlot; | |
05330448 | 57 | |
5832d1f2 AL |
58 | typedef struct kvm_dirty_log KVMDirtyLog; |
59 | ||
05330448 AL |
60 | struct KVMState |
61 | { | |
62 | KVMSlot slots[32]; | |
63 | int fd; | |
64 | int vmfd; | |
f65ed4c1 | 65 | int coalesced_mmio; |
62a2744c | 66 | struct kvm_coalesced_mmio_ring *coalesced_mmio_ring; |
e69917e2 | 67 | int broken_set_mem_region; |
4495d6a7 | 68 | int migration_log; |
a0fb002c | 69 | int vcpu_events; |
b0b1d690 | 70 | int robust_singlestep; |
ff44f1a3 | 71 | int debugregs; |
e22a25c9 AL |
72 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
73 | struct kvm_sw_breakpoint_head kvm_sw_breakpoints; | |
74 | #endif | |
6f725c13 GC |
75 | int irqchip_in_kernel; |
76 | int pit_in_kernel; | |
f1665b21 | 77 | int xsave, xcrs; |
d2f2b8a7 | 78 | int many_ioeventfds; |
05330448 AL |
79 | }; |
80 | ||
6a7af8cb | 81 | KVMState *kvm_state; |
05330448 | 82 | |
94a8d39a JK |
83 | static const KVMCapabilityInfo kvm_required_capabilites[] = { |
84 | KVM_CAP_INFO(USER_MEMORY), | |
85 | KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS), | |
86 | KVM_CAP_LAST_INFO | |
87 | }; | |
88 | ||
05330448 AL |
89 | static KVMSlot *kvm_alloc_slot(KVMState *s) |
90 | { | |
91 | int i; | |
92 | ||
93 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
a426e122 | 94 | if (s->slots[i].memory_size == 0) { |
05330448 | 95 | return &s->slots[i]; |
a426e122 | 96 | } |
05330448 AL |
97 | } |
98 | ||
d3f8d37f AL |
99 | fprintf(stderr, "%s: no free slot available\n", __func__); |
100 | abort(); | |
101 | } | |
102 | ||
103 | static KVMSlot *kvm_lookup_matching_slot(KVMState *s, | |
c227f099 AL |
104 | target_phys_addr_t start_addr, |
105 | target_phys_addr_t end_addr) | |
d3f8d37f AL |
106 | { |
107 | int i; | |
108 | ||
109 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
110 | KVMSlot *mem = &s->slots[i]; | |
111 | ||
112 | if (start_addr == mem->start_addr && | |
113 | end_addr == mem->start_addr + mem->memory_size) { | |
114 | return mem; | |
115 | } | |
116 | } | |
117 | ||
05330448 AL |
118 | return NULL; |
119 | } | |
120 | ||
6152e2ae AL |
121 | /* |
122 | * Find overlapping slot with lowest start address | |
123 | */ | |
124 | static KVMSlot *kvm_lookup_overlapping_slot(KVMState *s, | |
c227f099 AL |
125 | target_phys_addr_t start_addr, |
126 | target_phys_addr_t end_addr) | |
05330448 | 127 | { |
6152e2ae | 128 | KVMSlot *found = NULL; |
05330448 AL |
129 | int i; |
130 | ||
131 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
132 | KVMSlot *mem = &s->slots[i]; | |
133 | ||
6152e2ae AL |
134 | if (mem->memory_size == 0 || |
135 | (found && found->start_addr < mem->start_addr)) { | |
136 | continue; | |
137 | } | |
138 | ||
139 | if (end_addr > mem->start_addr && | |
140 | start_addr < mem->start_addr + mem->memory_size) { | |
141 | found = mem; | |
142 | } | |
05330448 AL |
143 | } |
144 | ||
6152e2ae | 145 | return found; |
05330448 AL |
146 | } |
147 | ||
983dfc3b HY |
148 | int kvm_physical_memory_addr_from_ram(KVMState *s, ram_addr_t ram_addr, |
149 | target_phys_addr_t *phys_addr) | |
150 | { | |
151 | int i; | |
152 | ||
153 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
154 | KVMSlot *mem = &s->slots[i]; | |
155 | ||
156 | if (ram_addr >= mem->phys_offset && | |
157 | ram_addr < mem->phys_offset + mem->memory_size) { | |
158 | *phys_addr = mem->start_addr + (ram_addr - mem->phys_offset); | |
159 | return 1; | |
160 | } | |
161 | } | |
162 | ||
163 | return 0; | |
164 | } | |
165 | ||
5832d1f2 AL |
166 | static int kvm_set_user_memory_region(KVMState *s, KVMSlot *slot) |
167 | { | |
168 | struct kvm_userspace_memory_region mem; | |
169 | ||
170 | mem.slot = slot->slot; | |
171 | mem.guest_phys_addr = slot->start_addr; | |
172 | mem.memory_size = slot->memory_size; | |
b2e0a138 | 173 | mem.userspace_addr = (unsigned long)qemu_safe_ram_ptr(slot->phys_offset); |
5832d1f2 | 174 | mem.flags = slot->flags; |
4495d6a7 JK |
175 | if (s->migration_log) { |
176 | mem.flags |= KVM_MEM_LOG_DIRTY_PAGES; | |
177 | } | |
5832d1f2 AL |
178 | return kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem); |
179 | } | |
180 | ||
8d2ba1fb JK |
181 | static void kvm_reset_vcpu(void *opaque) |
182 | { | |
183 | CPUState *env = opaque; | |
184 | ||
caa5af0f | 185 | kvm_arch_reset_vcpu(env); |
8d2ba1fb | 186 | } |
5832d1f2 | 187 | |
6f725c13 GC |
188 | int kvm_irqchip_in_kernel(void) |
189 | { | |
190 | return kvm_state->irqchip_in_kernel; | |
191 | } | |
192 | ||
193 | int kvm_pit_in_kernel(void) | |
194 | { | |
195 | return kvm_state->pit_in_kernel; | |
196 | } | |
197 | ||
05330448 AL |
198 | int kvm_init_vcpu(CPUState *env) |
199 | { | |
200 | KVMState *s = kvm_state; | |
201 | long mmap_size; | |
202 | int ret; | |
203 | ||
8c0d577e | 204 | DPRINTF("kvm_init_vcpu\n"); |
05330448 | 205 | |
984b5181 | 206 | ret = kvm_vm_ioctl(s, KVM_CREATE_VCPU, env->cpu_index); |
05330448 | 207 | if (ret < 0) { |
8c0d577e | 208 | DPRINTF("kvm_create_vcpu failed\n"); |
05330448 AL |
209 | goto err; |
210 | } | |
211 | ||
212 | env->kvm_fd = ret; | |
213 | env->kvm_state = s; | |
214 | ||
215 | mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0); | |
216 | if (mmap_size < 0) { | |
748a680b | 217 | ret = mmap_size; |
8c0d577e | 218 | DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n"); |
05330448 AL |
219 | goto err; |
220 | } | |
221 | ||
222 | env->kvm_run = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED, | |
223 | env->kvm_fd, 0); | |
224 | if (env->kvm_run == MAP_FAILED) { | |
225 | ret = -errno; | |
8c0d577e | 226 | DPRINTF("mmap'ing vcpu state failed\n"); |
05330448 AL |
227 | goto err; |
228 | } | |
229 | ||
a426e122 JK |
230 | if (s->coalesced_mmio && !s->coalesced_mmio_ring) { |
231 | s->coalesced_mmio_ring = | |
232 | (void *)env->kvm_run + s->coalesced_mmio * PAGE_SIZE; | |
233 | } | |
62a2744c | 234 | |
05330448 | 235 | ret = kvm_arch_init_vcpu(env); |
8d2ba1fb | 236 | if (ret == 0) { |
a08d4367 | 237 | qemu_register_reset(kvm_reset_vcpu, env); |
caa5af0f | 238 | kvm_arch_reset_vcpu(env); |
8d2ba1fb | 239 | } |
05330448 AL |
240 | err: |
241 | return ret; | |
242 | } | |
243 | ||
5832d1f2 AL |
244 | /* |
245 | * dirty pages logging control | |
246 | */ | |
c227f099 AL |
247 | static int kvm_dirty_pages_log_change(target_phys_addr_t phys_addr, |
248 | ram_addr_t size, int flags, int mask) | |
5832d1f2 AL |
249 | { |
250 | KVMState *s = kvm_state; | |
d3f8d37f | 251 | KVMSlot *mem = kvm_lookup_matching_slot(s, phys_addr, phys_addr + size); |
4495d6a7 JK |
252 | int old_flags; |
253 | ||
5832d1f2 | 254 | if (mem == NULL) { |
d3f8d37f AL |
255 | fprintf(stderr, "BUG: %s: invalid parameters " TARGET_FMT_plx "-" |
256 | TARGET_FMT_plx "\n", __func__, phys_addr, | |
c227f099 | 257 | (target_phys_addr_t)(phys_addr + size - 1)); |
5832d1f2 AL |
258 | return -EINVAL; |
259 | } | |
260 | ||
4495d6a7 | 261 | old_flags = mem->flags; |
5832d1f2 | 262 | |
4495d6a7 | 263 | flags = (mem->flags & ~mask) | flags; |
5832d1f2 AL |
264 | mem->flags = flags; |
265 | ||
4495d6a7 JK |
266 | /* If nothing changed effectively, no need to issue ioctl */ |
267 | if (s->migration_log) { | |
268 | flags |= KVM_MEM_LOG_DIRTY_PAGES; | |
269 | } | |
270 | if (flags == old_flags) { | |
271 | return 0; | |
272 | } | |
273 | ||
5832d1f2 AL |
274 | return kvm_set_user_memory_region(s, mem); |
275 | } | |
276 | ||
e5896b12 AP |
277 | static int kvm_log_start(CPUPhysMemoryClient *client, |
278 | target_phys_addr_t phys_addr, ram_addr_t size) | |
5832d1f2 | 279 | { |
a426e122 JK |
280 | return kvm_dirty_pages_log_change(phys_addr, size, KVM_MEM_LOG_DIRTY_PAGES, |
281 | KVM_MEM_LOG_DIRTY_PAGES); | |
5832d1f2 AL |
282 | } |
283 | ||
e5896b12 AP |
284 | static int kvm_log_stop(CPUPhysMemoryClient *client, |
285 | target_phys_addr_t phys_addr, ram_addr_t size) | |
5832d1f2 | 286 | { |
a426e122 JK |
287 | return kvm_dirty_pages_log_change(phys_addr, size, 0, |
288 | KVM_MEM_LOG_DIRTY_PAGES); | |
5832d1f2 AL |
289 | } |
290 | ||
7b8f3b78 | 291 | static int kvm_set_migration_log(int enable) |
4495d6a7 JK |
292 | { |
293 | KVMState *s = kvm_state; | |
294 | KVMSlot *mem; | |
295 | int i, err; | |
296 | ||
297 | s->migration_log = enable; | |
298 | ||
299 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
300 | mem = &s->slots[i]; | |
301 | ||
70fedd76 AW |
302 | if (!mem->memory_size) { |
303 | continue; | |
304 | } | |
4495d6a7 JK |
305 | if (!!(mem->flags & KVM_MEM_LOG_DIRTY_PAGES) == enable) { |
306 | continue; | |
307 | } | |
308 | err = kvm_set_user_memory_region(s, mem); | |
309 | if (err) { | |
310 | return err; | |
311 | } | |
312 | } | |
313 | return 0; | |
314 | } | |
315 | ||
8369e01c MT |
316 | /* get kvm's dirty pages bitmap and update qemu's */ |
317 | static int kvm_get_dirty_pages_log_range(unsigned long start_addr, | |
318 | unsigned long *bitmap, | |
319 | unsigned long offset, | |
320 | unsigned long mem_size) | |
96c1606b | 321 | { |
8369e01c MT |
322 | unsigned int i, j; |
323 | unsigned long page_number, addr, addr1, c; | |
324 | ram_addr_t ram_addr; | |
325 | unsigned int len = ((mem_size / TARGET_PAGE_SIZE) + HOST_LONG_BITS - 1) / | |
326 | HOST_LONG_BITS; | |
327 | ||
328 | /* | |
329 | * bitmap-traveling is faster than memory-traveling (for addr...) | |
330 | * especially when most of the memory is not dirty. | |
331 | */ | |
332 | for (i = 0; i < len; i++) { | |
333 | if (bitmap[i] != 0) { | |
334 | c = leul_to_cpu(bitmap[i]); | |
335 | do { | |
336 | j = ffsl(c) - 1; | |
337 | c &= ~(1ul << j); | |
338 | page_number = i * HOST_LONG_BITS + j; | |
339 | addr1 = page_number * TARGET_PAGE_SIZE; | |
340 | addr = offset + addr1; | |
341 | ram_addr = cpu_get_physical_page_desc(addr); | |
342 | cpu_physical_memory_set_dirty(ram_addr); | |
343 | } while (c != 0); | |
344 | } | |
345 | } | |
346 | return 0; | |
96c1606b AG |
347 | } |
348 | ||
8369e01c MT |
349 | #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1)) |
350 | ||
5832d1f2 AL |
351 | /** |
352 | * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space | |
353 | * This function updates qemu's dirty bitmap using cpu_physical_memory_set_dirty(). | |
354 | * This means all bits are set to dirty. | |
355 | * | |
d3f8d37f | 356 | * @start_add: start of logged region. |
5832d1f2 AL |
357 | * @end_addr: end of logged region. |
358 | */ | |
7b8f3b78 | 359 | static int kvm_physical_sync_dirty_bitmap(target_phys_addr_t start_addr, |
a426e122 | 360 | target_phys_addr_t end_addr) |
5832d1f2 AL |
361 | { |
362 | KVMState *s = kvm_state; | |
151f7749 | 363 | unsigned long size, allocated_size = 0; |
151f7749 JK |
364 | KVMDirtyLog d; |
365 | KVMSlot *mem; | |
366 | int ret = 0; | |
5832d1f2 | 367 | |
151f7749 JK |
368 | d.dirty_bitmap = NULL; |
369 | while (start_addr < end_addr) { | |
370 | mem = kvm_lookup_overlapping_slot(s, start_addr, end_addr); | |
371 | if (mem == NULL) { | |
372 | break; | |
373 | } | |
5832d1f2 | 374 | |
8369e01c | 375 | size = ALIGN(((mem->memory_size) >> TARGET_PAGE_BITS), HOST_LONG_BITS) / 8; |
151f7749 JK |
376 | if (!d.dirty_bitmap) { |
377 | d.dirty_bitmap = qemu_malloc(size); | |
378 | } else if (size > allocated_size) { | |
379 | d.dirty_bitmap = qemu_realloc(d.dirty_bitmap, size); | |
380 | } | |
381 | allocated_size = size; | |
382 | memset(d.dirty_bitmap, 0, allocated_size); | |
5832d1f2 | 383 | |
151f7749 | 384 | d.slot = mem->slot; |
5832d1f2 | 385 | |
6e489f3f | 386 | if (kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d) == -1) { |
8c0d577e | 387 | DPRINTF("ioctl failed %d\n", errno); |
151f7749 JK |
388 | ret = -1; |
389 | break; | |
390 | } | |
5832d1f2 | 391 | |
8369e01c MT |
392 | kvm_get_dirty_pages_log_range(mem->start_addr, d.dirty_bitmap, |
393 | mem->start_addr, mem->memory_size); | |
394 | start_addr = mem->start_addr + mem->memory_size; | |
5832d1f2 | 395 | } |
5832d1f2 | 396 | qemu_free(d.dirty_bitmap); |
151f7749 JK |
397 | |
398 | return ret; | |
5832d1f2 AL |
399 | } |
400 | ||
c227f099 | 401 | int kvm_coalesce_mmio_region(target_phys_addr_t start, ram_addr_t size) |
f65ed4c1 AL |
402 | { |
403 | int ret = -ENOSYS; | |
f65ed4c1 AL |
404 | KVMState *s = kvm_state; |
405 | ||
406 | if (s->coalesced_mmio) { | |
407 | struct kvm_coalesced_mmio_zone zone; | |
408 | ||
409 | zone.addr = start; | |
410 | zone.size = size; | |
411 | ||
412 | ret = kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone); | |
413 | } | |
f65ed4c1 AL |
414 | |
415 | return ret; | |
416 | } | |
417 | ||
c227f099 | 418 | int kvm_uncoalesce_mmio_region(target_phys_addr_t start, ram_addr_t size) |
f65ed4c1 AL |
419 | { |
420 | int ret = -ENOSYS; | |
f65ed4c1 AL |
421 | KVMState *s = kvm_state; |
422 | ||
423 | if (s->coalesced_mmio) { | |
424 | struct kvm_coalesced_mmio_zone zone; | |
425 | ||
426 | zone.addr = start; | |
427 | zone.size = size; | |
428 | ||
429 | ret = kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone); | |
430 | } | |
f65ed4c1 AL |
431 | |
432 | return ret; | |
433 | } | |
434 | ||
ad7b8b33 AL |
435 | int kvm_check_extension(KVMState *s, unsigned int extension) |
436 | { | |
437 | int ret; | |
438 | ||
439 | ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension); | |
440 | if (ret < 0) { | |
441 | ret = 0; | |
442 | } | |
443 | ||
444 | return ret; | |
445 | } | |
446 | ||
d2f2b8a7 SH |
447 | static int kvm_check_many_ioeventfds(void) |
448 | { | |
d0dcac83 SH |
449 | /* Userspace can use ioeventfd for io notification. This requires a host |
450 | * that supports eventfd(2) and an I/O thread; since eventfd does not | |
451 | * support SIGIO it cannot interrupt the vcpu. | |
452 | * | |
453 | * Older kernels have a 6 device limit on the KVM io bus. Find out so we | |
d2f2b8a7 SH |
454 | * can avoid creating too many ioeventfds. |
455 | */ | |
d0dcac83 | 456 | #if defined(CONFIG_EVENTFD) && defined(CONFIG_IOTHREAD) |
d2f2b8a7 SH |
457 | int ioeventfds[7]; |
458 | int i, ret = 0; | |
459 | for (i = 0; i < ARRAY_SIZE(ioeventfds); i++) { | |
460 | ioeventfds[i] = eventfd(0, EFD_CLOEXEC); | |
461 | if (ioeventfds[i] < 0) { | |
462 | break; | |
463 | } | |
464 | ret = kvm_set_ioeventfd_pio_word(ioeventfds[i], 0, i, true); | |
465 | if (ret < 0) { | |
466 | close(ioeventfds[i]); | |
467 | break; | |
468 | } | |
469 | } | |
470 | ||
471 | /* Decide whether many devices are supported or not */ | |
472 | ret = i == ARRAY_SIZE(ioeventfds); | |
473 | ||
474 | while (i-- > 0) { | |
475 | kvm_set_ioeventfd_pio_word(ioeventfds[i], 0, i, false); | |
476 | close(ioeventfds[i]); | |
477 | } | |
478 | return ret; | |
479 | #else | |
480 | return 0; | |
481 | #endif | |
482 | } | |
483 | ||
94a8d39a JK |
484 | static const KVMCapabilityInfo * |
485 | kvm_check_extension_list(KVMState *s, const KVMCapabilityInfo *list) | |
486 | { | |
487 | while (list->name) { | |
488 | if (!kvm_check_extension(s, list->value)) { | |
489 | return list; | |
490 | } | |
491 | list++; | |
492 | } | |
493 | return NULL; | |
494 | } | |
495 | ||
a426e122 JK |
496 | static void kvm_set_phys_mem(target_phys_addr_t start_addr, ram_addr_t size, |
497 | ram_addr_t phys_offset) | |
46dbef6a MT |
498 | { |
499 | KVMState *s = kvm_state; | |
500 | ram_addr_t flags = phys_offset & ~TARGET_PAGE_MASK; | |
501 | KVMSlot *mem, old; | |
502 | int err; | |
503 | ||
14542fea GN |
504 | /* kvm works in page size chunks, but the function may be called |
505 | with sub-page size and unaligned start address. */ | |
506 | size = TARGET_PAGE_ALIGN(size); | |
507 | start_addr = TARGET_PAGE_ALIGN(start_addr); | |
46dbef6a MT |
508 | |
509 | /* KVM does not support read-only slots */ | |
510 | phys_offset &= ~IO_MEM_ROM; | |
511 | ||
512 | while (1) { | |
513 | mem = kvm_lookup_overlapping_slot(s, start_addr, start_addr + size); | |
514 | if (!mem) { | |
515 | break; | |
516 | } | |
517 | ||
518 | if (flags < IO_MEM_UNASSIGNED && start_addr >= mem->start_addr && | |
519 | (start_addr + size <= mem->start_addr + mem->memory_size) && | |
520 | (phys_offset - start_addr == mem->phys_offset - mem->start_addr)) { | |
521 | /* The new slot fits into the existing one and comes with | |
522 | * identical parameters - nothing to be done. */ | |
523 | return; | |
524 | } | |
525 | ||
526 | old = *mem; | |
527 | ||
528 | /* unregister the overlapping slot */ | |
529 | mem->memory_size = 0; | |
530 | err = kvm_set_user_memory_region(s, mem); | |
531 | if (err) { | |
532 | fprintf(stderr, "%s: error unregistering overlapping slot: %s\n", | |
533 | __func__, strerror(-err)); | |
534 | abort(); | |
535 | } | |
536 | ||
537 | /* Workaround for older KVM versions: we can't join slots, even not by | |
538 | * unregistering the previous ones and then registering the larger | |
539 | * slot. We have to maintain the existing fragmentation. Sigh. | |
540 | * | |
541 | * This workaround assumes that the new slot starts at the same | |
542 | * address as the first existing one. If not or if some overlapping | |
543 | * slot comes around later, we will fail (not seen in practice so far) | |
544 | * - and actually require a recent KVM version. */ | |
545 | if (s->broken_set_mem_region && | |
546 | old.start_addr == start_addr && old.memory_size < size && | |
547 | flags < IO_MEM_UNASSIGNED) { | |
548 | mem = kvm_alloc_slot(s); | |
549 | mem->memory_size = old.memory_size; | |
550 | mem->start_addr = old.start_addr; | |
551 | mem->phys_offset = old.phys_offset; | |
552 | mem->flags = 0; | |
553 | ||
554 | err = kvm_set_user_memory_region(s, mem); | |
555 | if (err) { | |
556 | fprintf(stderr, "%s: error updating slot: %s\n", __func__, | |
557 | strerror(-err)); | |
558 | abort(); | |
559 | } | |
560 | ||
561 | start_addr += old.memory_size; | |
562 | phys_offset += old.memory_size; | |
563 | size -= old.memory_size; | |
564 | continue; | |
565 | } | |
566 | ||
567 | /* register prefix slot */ | |
568 | if (old.start_addr < start_addr) { | |
569 | mem = kvm_alloc_slot(s); | |
570 | mem->memory_size = start_addr - old.start_addr; | |
571 | mem->start_addr = old.start_addr; | |
572 | mem->phys_offset = old.phys_offset; | |
573 | mem->flags = 0; | |
574 | ||
575 | err = kvm_set_user_memory_region(s, mem); | |
576 | if (err) { | |
577 | fprintf(stderr, "%s: error registering prefix slot: %s\n", | |
578 | __func__, strerror(-err)); | |
579 | abort(); | |
580 | } | |
581 | } | |
582 | ||
583 | /* register suffix slot */ | |
584 | if (old.start_addr + old.memory_size > start_addr + size) { | |
585 | ram_addr_t size_delta; | |
586 | ||
587 | mem = kvm_alloc_slot(s); | |
588 | mem->start_addr = start_addr + size; | |
589 | size_delta = mem->start_addr - old.start_addr; | |
590 | mem->memory_size = old.memory_size - size_delta; | |
591 | mem->phys_offset = old.phys_offset + size_delta; | |
592 | mem->flags = 0; | |
593 | ||
594 | err = kvm_set_user_memory_region(s, mem); | |
595 | if (err) { | |
596 | fprintf(stderr, "%s: error registering suffix slot: %s\n", | |
597 | __func__, strerror(-err)); | |
598 | abort(); | |
599 | } | |
600 | } | |
601 | } | |
602 | ||
603 | /* in case the KVM bug workaround already "consumed" the new slot */ | |
a426e122 | 604 | if (!size) { |
46dbef6a | 605 | return; |
a426e122 | 606 | } |
46dbef6a | 607 | /* KVM does not need to know about this memory */ |
a426e122 | 608 | if (flags >= IO_MEM_UNASSIGNED) { |
46dbef6a | 609 | return; |
a426e122 | 610 | } |
46dbef6a MT |
611 | mem = kvm_alloc_slot(s); |
612 | mem->memory_size = size; | |
613 | mem->start_addr = start_addr; | |
614 | mem->phys_offset = phys_offset; | |
615 | mem->flags = 0; | |
616 | ||
617 | err = kvm_set_user_memory_region(s, mem); | |
618 | if (err) { | |
619 | fprintf(stderr, "%s: error registering slot: %s\n", __func__, | |
620 | strerror(-err)); | |
621 | abort(); | |
622 | } | |
623 | } | |
624 | ||
7b8f3b78 | 625 | static void kvm_client_set_memory(struct CPUPhysMemoryClient *client, |
a426e122 JK |
626 | target_phys_addr_t start_addr, |
627 | ram_addr_t size, ram_addr_t phys_offset) | |
7b8f3b78 | 628 | { |
a426e122 | 629 | kvm_set_phys_mem(start_addr, size, phys_offset); |
7b8f3b78 MT |
630 | } |
631 | ||
632 | static int kvm_client_sync_dirty_bitmap(struct CPUPhysMemoryClient *client, | |
a426e122 JK |
633 | target_phys_addr_t start_addr, |
634 | target_phys_addr_t end_addr) | |
7b8f3b78 | 635 | { |
a426e122 | 636 | return kvm_physical_sync_dirty_bitmap(start_addr, end_addr); |
7b8f3b78 MT |
637 | } |
638 | ||
639 | static int kvm_client_migration_log(struct CPUPhysMemoryClient *client, | |
a426e122 | 640 | int enable) |
7b8f3b78 | 641 | { |
a426e122 | 642 | return kvm_set_migration_log(enable); |
7b8f3b78 MT |
643 | } |
644 | ||
645 | static CPUPhysMemoryClient kvm_cpu_phys_memory_client = { | |
a426e122 JK |
646 | .set_memory = kvm_client_set_memory, |
647 | .sync_dirty_bitmap = kvm_client_sync_dirty_bitmap, | |
648 | .migration_log = kvm_client_migration_log, | |
e5896b12 AP |
649 | .log_start = kvm_log_start, |
650 | .log_stop = kvm_log_stop, | |
7b8f3b78 MT |
651 | }; |
652 | ||
cad1e282 | 653 | int kvm_init(void) |
05330448 | 654 | { |
168ccc11 JK |
655 | static const char upgrade_note[] = |
656 | "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n" | |
657 | "(see http://sourceforge.net/projects/kvm).\n"; | |
05330448 | 658 | KVMState *s; |
94a8d39a | 659 | const KVMCapabilityInfo *missing_cap; |
05330448 AL |
660 | int ret; |
661 | int i; | |
662 | ||
05330448 | 663 | s = qemu_mallocz(sizeof(KVMState)); |
05330448 | 664 | |
e22a25c9 | 665 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
72cf2d4f | 666 | QTAILQ_INIT(&s->kvm_sw_breakpoints); |
e22a25c9 | 667 | #endif |
a426e122 | 668 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { |
05330448 | 669 | s->slots[i].slot = i; |
a426e122 | 670 | } |
05330448 | 671 | s->vmfd = -1; |
40ff6d7e | 672 | s->fd = qemu_open("/dev/kvm", O_RDWR); |
05330448 AL |
673 | if (s->fd == -1) { |
674 | fprintf(stderr, "Could not access KVM kernel module: %m\n"); | |
675 | ret = -errno; | |
676 | goto err; | |
677 | } | |
678 | ||
679 | ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0); | |
680 | if (ret < KVM_API_VERSION) { | |
a426e122 | 681 | if (ret > 0) { |
05330448 | 682 | ret = -EINVAL; |
a426e122 | 683 | } |
05330448 AL |
684 | fprintf(stderr, "kvm version too old\n"); |
685 | goto err; | |
686 | } | |
687 | ||
688 | if (ret > KVM_API_VERSION) { | |
689 | ret = -EINVAL; | |
690 | fprintf(stderr, "kvm version not supported\n"); | |
691 | goto err; | |
692 | } | |
693 | ||
694 | s->vmfd = kvm_ioctl(s, KVM_CREATE_VM, 0); | |
0104dcac AG |
695 | if (s->vmfd < 0) { |
696 | #ifdef TARGET_S390X | |
697 | fprintf(stderr, "Please add the 'switch_amode' kernel parameter to " | |
698 | "your host kernel command line\n"); | |
699 | #endif | |
05330448 | 700 | goto err; |
0104dcac | 701 | } |
05330448 | 702 | |
94a8d39a JK |
703 | missing_cap = kvm_check_extension_list(s, kvm_required_capabilites); |
704 | if (!missing_cap) { | |
705 | missing_cap = | |
706 | kvm_check_extension_list(s, kvm_arch_required_capabilities); | |
05330448 | 707 | } |
94a8d39a | 708 | if (missing_cap) { |
ad7b8b33 | 709 | ret = -EINVAL; |
94a8d39a JK |
710 | fprintf(stderr, "kvm does not support %s\n%s", |
711 | missing_cap->name, upgrade_note); | |
d85dc283 AL |
712 | goto err; |
713 | } | |
714 | ||
ad7b8b33 | 715 | s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO); |
f65ed4c1 | 716 | |
e69917e2 JK |
717 | s->broken_set_mem_region = 1; |
718 | #ifdef KVM_CAP_JOIN_MEMORY_REGIONS_WORKS | |
14a09518 | 719 | ret = kvm_check_extension(s, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS); |
e69917e2 JK |
720 | if (ret > 0) { |
721 | s->broken_set_mem_region = 0; | |
722 | } | |
723 | #endif | |
724 | ||
a0fb002c JK |
725 | s->vcpu_events = 0; |
726 | #ifdef KVM_CAP_VCPU_EVENTS | |
727 | s->vcpu_events = kvm_check_extension(s, KVM_CAP_VCPU_EVENTS); | |
728 | #endif | |
729 | ||
b0b1d690 JK |
730 | s->robust_singlestep = 0; |
731 | #ifdef KVM_CAP_X86_ROBUST_SINGLESTEP | |
732 | s->robust_singlestep = | |
733 | kvm_check_extension(s, KVM_CAP_X86_ROBUST_SINGLESTEP); | |
734 | #endif | |
735 | ||
ff44f1a3 JK |
736 | s->debugregs = 0; |
737 | #ifdef KVM_CAP_DEBUGREGS | |
738 | s->debugregs = kvm_check_extension(s, KVM_CAP_DEBUGREGS); | |
739 | #endif | |
740 | ||
f1665b21 SY |
741 | s->xsave = 0; |
742 | #ifdef KVM_CAP_XSAVE | |
743 | s->xsave = kvm_check_extension(s, KVM_CAP_XSAVE); | |
744 | #endif | |
745 | ||
746 | s->xcrs = 0; | |
747 | #ifdef KVM_CAP_XCRS | |
748 | s->xcrs = kvm_check_extension(s, KVM_CAP_XCRS); | |
749 | #endif | |
750 | ||
cad1e282 | 751 | ret = kvm_arch_init(s); |
a426e122 | 752 | if (ret < 0) { |
05330448 | 753 | goto err; |
a426e122 | 754 | } |
05330448 AL |
755 | |
756 | kvm_state = s; | |
7b8f3b78 | 757 | cpu_register_phys_memory_client(&kvm_cpu_phys_memory_client); |
05330448 | 758 | |
d2f2b8a7 SH |
759 | s->many_ioeventfds = kvm_check_many_ioeventfds(); |
760 | ||
05330448 AL |
761 | return 0; |
762 | ||
763 | err: | |
764 | if (s) { | |
a426e122 | 765 | if (s->vmfd != -1) { |
05330448 | 766 | close(s->vmfd); |
a426e122 JK |
767 | } |
768 | if (s->fd != -1) { | |
05330448 | 769 | close(s->fd); |
a426e122 | 770 | } |
05330448 AL |
771 | } |
772 | qemu_free(s); | |
773 | ||
774 | return ret; | |
775 | } | |
776 | ||
b30e93e9 JK |
777 | static void kvm_handle_io(uint16_t port, void *data, int direction, int size, |
778 | uint32_t count) | |
05330448 AL |
779 | { |
780 | int i; | |
781 | uint8_t *ptr = data; | |
782 | ||
783 | for (i = 0; i < count; i++) { | |
784 | if (direction == KVM_EXIT_IO_IN) { | |
785 | switch (size) { | |
786 | case 1: | |
afcea8cb | 787 | stb_p(ptr, cpu_inb(port)); |
05330448 AL |
788 | break; |
789 | case 2: | |
afcea8cb | 790 | stw_p(ptr, cpu_inw(port)); |
05330448 AL |
791 | break; |
792 | case 4: | |
afcea8cb | 793 | stl_p(ptr, cpu_inl(port)); |
05330448 AL |
794 | break; |
795 | } | |
796 | } else { | |
797 | switch (size) { | |
798 | case 1: | |
afcea8cb | 799 | cpu_outb(port, ldub_p(ptr)); |
05330448 AL |
800 | break; |
801 | case 2: | |
afcea8cb | 802 | cpu_outw(port, lduw_p(ptr)); |
05330448 AL |
803 | break; |
804 | case 4: | |
afcea8cb | 805 | cpu_outl(port, ldl_p(ptr)); |
05330448 AL |
806 | break; |
807 | } | |
808 | } | |
809 | ||
810 | ptr += size; | |
811 | } | |
05330448 AL |
812 | } |
813 | ||
7c80eef8 | 814 | #ifdef KVM_CAP_INTERNAL_ERROR_DATA |
73aaec4a | 815 | static int kvm_handle_internal_error(CPUState *env, struct kvm_run *run) |
7c80eef8 | 816 | { |
bb44e0d1 | 817 | fprintf(stderr, "KVM internal error."); |
7c80eef8 MT |
818 | if (kvm_check_extension(kvm_state, KVM_CAP_INTERNAL_ERROR_DATA)) { |
819 | int i; | |
820 | ||
bb44e0d1 | 821 | fprintf(stderr, " Suberror: %d\n", run->internal.suberror); |
7c80eef8 MT |
822 | for (i = 0; i < run->internal.ndata; ++i) { |
823 | fprintf(stderr, "extra data[%d]: %"PRIx64"\n", | |
824 | i, (uint64_t)run->internal.data[i]); | |
825 | } | |
bb44e0d1 JK |
826 | } else { |
827 | fprintf(stderr, "\n"); | |
7c80eef8 | 828 | } |
7c80eef8 MT |
829 | if (run->internal.suberror == KVM_INTERNAL_ERROR_EMULATION) { |
830 | fprintf(stderr, "emulation failure\n"); | |
a426e122 | 831 | if (!kvm_arch_stop_on_emulation_error(env)) { |
f5c848ee | 832 | cpu_dump_state(env, stderr, fprintf, CPU_DUMP_CODE); |
73aaec4a | 833 | return 0; |
a426e122 | 834 | } |
7c80eef8 MT |
835 | } |
836 | /* FIXME: Should trigger a qmp message to let management know | |
837 | * something went wrong. | |
838 | */ | |
73aaec4a | 839 | return -1; |
7c80eef8 MT |
840 | } |
841 | #endif | |
842 | ||
62a2744c | 843 | void kvm_flush_coalesced_mmio_buffer(void) |
f65ed4c1 | 844 | { |
f65ed4c1 | 845 | KVMState *s = kvm_state; |
62a2744c SY |
846 | if (s->coalesced_mmio_ring) { |
847 | struct kvm_coalesced_mmio_ring *ring = s->coalesced_mmio_ring; | |
f65ed4c1 AL |
848 | while (ring->first != ring->last) { |
849 | struct kvm_coalesced_mmio *ent; | |
850 | ||
851 | ent = &ring->coalesced_mmio[ring->first]; | |
852 | ||
853 | cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len); | |
85199474 | 854 | smp_wmb(); |
f65ed4c1 AL |
855 | ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX; |
856 | } | |
857 | } | |
f65ed4c1 AL |
858 | } |
859 | ||
2705d56a | 860 | static void do_kvm_cpu_synchronize_state(void *_env) |
4c0960c0 | 861 | { |
2705d56a JK |
862 | CPUState *env = _env; |
863 | ||
9ded2744 | 864 | if (!env->kvm_vcpu_dirty) { |
4c0960c0 | 865 | kvm_arch_get_registers(env); |
9ded2744 | 866 | env->kvm_vcpu_dirty = 1; |
4c0960c0 AK |
867 | } |
868 | } | |
869 | ||
2705d56a JK |
870 | void kvm_cpu_synchronize_state(CPUState *env) |
871 | { | |
a426e122 | 872 | if (!env->kvm_vcpu_dirty) { |
2705d56a | 873 | run_on_cpu(env, do_kvm_cpu_synchronize_state, env); |
a426e122 | 874 | } |
2705d56a JK |
875 | } |
876 | ||
ea375f9a JK |
877 | void kvm_cpu_synchronize_post_reset(CPUState *env) |
878 | { | |
879 | kvm_arch_put_registers(env, KVM_PUT_RESET_STATE); | |
880 | env->kvm_vcpu_dirty = 0; | |
881 | } | |
882 | ||
883 | void kvm_cpu_synchronize_post_init(CPUState *env) | |
884 | { | |
885 | kvm_arch_put_registers(env, KVM_PUT_FULL_STATE); | |
886 | env->kvm_vcpu_dirty = 0; | |
887 | } | |
888 | ||
05330448 AL |
889 | int kvm_cpu_exec(CPUState *env) |
890 | { | |
891 | struct kvm_run *run = env->kvm_run; | |
892 | int ret; | |
893 | ||
8c0d577e | 894 | DPRINTF("kvm_cpu_exec()\n"); |
05330448 | 895 | |
9ccfac9e JK |
896 | if (kvm_arch_process_irqchip_events(env)) { |
897 | env->exit_request = 0; | |
6792a57b | 898 | return EXCP_HLT; |
9ccfac9e | 899 | } |
0af691d7 | 900 | |
6792a57b JK |
901 | cpu_single_env = env; |
902 | ||
9ccfac9e | 903 | do { |
9ded2744 | 904 | if (env->kvm_vcpu_dirty) { |
ea375f9a | 905 | kvm_arch_put_registers(env, KVM_PUT_RUNTIME_STATE); |
9ded2744 | 906 | env->kvm_vcpu_dirty = 0; |
4c0960c0 AK |
907 | } |
908 | ||
8c14c173 | 909 | kvm_arch_pre_run(env, run); |
9ccfac9e JK |
910 | if (env->exit_request) { |
911 | DPRINTF("interrupt exit requested\n"); | |
912 | /* | |
913 | * KVM requires us to reenter the kernel after IO exits to complete | |
914 | * instruction emulation. This self-signal will ensure that we | |
915 | * leave ASAP again. | |
916 | */ | |
917 | qemu_cpu_kick_self(); | |
918 | } | |
273faf1b | 919 | cpu_single_env = NULL; |
d549db5a | 920 | qemu_mutex_unlock_iothread(); |
9ccfac9e | 921 | |
05330448 | 922 | ret = kvm_vcpu_ioctl(env, KVM_RUN, 0); |
9ccfac9e | 923 | |
d549db5a | 924 | qemu_mutex_lock_iothread(); |
273faf1b | 925 | cpu_single_env = env; |
05330448 AL |
926 | kvm_arch_post_run(env, run); |
927 | ||
b0c883b5 JK |
928 | kvm_flush_coalesced_mmio_buffer(); |
929 | ||
05330448 | 930 | if (ret == -EINTR || ret == -EAGAIN) { |
8c0d577e | 931 | DPRINTF("io window exit\n"); |
05330448 AL |
932 | ret = 0; |
933 | break; | |
934 | } | |
935 | ||
936 | if (ret < 0) { | |
8c0d577e | 937 | DPRINTF("kvm run failed %s\n", strerror(-ret)); |
05330448 AL |
938 | abort(); |
939 | } | |
940 | ||
941 | ret = 0; /* exit loop */ | |
942 | switch (run->exit_reason) { | |
943 | case KVM_EXIT_IO: | |
8c0d577e | 944 | DPRINTF("handle_io\n"); |
b30e93e9 JK |
945 | kvm_handle_io(run->io.port, |
946 | (uint8_t *)run + run->io.data_offset, | |
947 | run->io.direction, | |
948 | run->io.size, | |
949 | run->io.count); | |
950 | ret = 1; | |
05330448 AL |
951 | break; |
952 | case KVM_EXIT_MMIO: | |
8c0d577e | 953 | DPRINTF("handle_mmio\n"); |
05330448 AL |
954 | cpu_physical_memory_rw(run->mmio.phys_addr, |
955 | run->mmio.data, | |
956 | run->mmio.len, | |
957 | run->mmio.is_write); | |
958 | ret = 1; | |
959 | break; | |
960 | case KVM_EXIT_IRQ_WINDOW_OPEN: | |
8c0d577e | 961 | DPRINTF("irq_window_open\n"); |
05330448 AL |
962 | break; |
963 | case KVM_EXIT_SHUTDOWN: | |
8c0d577e | 964 | DPRINTF("shutdown\n"); |
05330448 | 965 | qemu_system_reset_request(); |
05330448 AL |
966 | break; |
967 | case KVM_EXIT_UNKNOWN: | |
bb44e0d1 JK |
968 | fprintf(stderr, "KVM: unknown exit, hardware reason %" PRIx64 "\n", |
969 | (uint64_t)run->hw.hardware_exit_reason); | |
73aaec4a | 970 | ret = -1; |
05330448 | 971 | break; |
7c80eef8 MT |
972 | #ifdef KVM_CAP_INTERNAL_ERROR_DATA |
973 | case KVM_EXIT_INTERNAL_ERROR: | |
73aaec4a | 974 | ret = kvm_handle_internal_error(env, run); |
7c80eef8 MT |
975 | break; |
976 | #endif | |
05330448 | 977 | case KVM_EXIT_DEBUG: |
8c0d577e | 978 | DPRINTF("kvm_exit_debug\n"); |
e22a25c9 AL |
979 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
980 | if (kvm_arch_debug(&run->debug.arch)) { | |
6792a57b JK |
981 | ret = EXCP_DEBUG; |
982 | goto out; | |
e22a25c9 AL |
983 | } |
984 | /* re-enter, this exception was guest-internal */ | |
985 | ret = 1; | |
986 | #endif /* KVM_CAP_SET_GUEST_DEBUG */ | |
05330448 AL |
987 | break; |
988 | default: | |
8c0d577e | 989 | DPRINTF("kvm_arch_handle_exit\n"); |
05330448 AL |
990 | ret = kvm_arch_handle_exit(env, run); |
991 | break; | |
992 | } | |
993 | } while (ret > 0); | |
994 | ||
73aaec4a | 995 | if (ret < 0) { |
f5c848ee | 996 | cpu_dump_state(env, stderr, fprintf, CPU_DUMP_CODE); |
e07bbac5 | 997 | vm_stop(VMSTOP_PANIC); |
becfc390 | 998 | } |
6792a57b | 999 | ret = EXCP_INTERRUPT; |
becfc390 | 1000 | |
6792a57b JK |
1001 | out: |
1002 | env->exit_request = 0; | |
1003 | cpu_single_env = NULL; | |
05330448 AL |
1004 | return ret; |
1005 | } | |
1006 | ||
984b5181 | 1007 | int kvm_ioctl(KVMState *s, int type, ...) |
05330448 AL |
1008 | { |
1009 | int ret; | |
984b5181 AL |
1010 | void *arg; |
1011 | va_list ap; | |
05330448 | 1012 | |
984b5181 AL |
1013 | va_start(ap, type); |
1014 | arg = va_arg(ap, void *); | |
1015 | va_end(ap); | |
1016 | ||
1017 | ret = ioctl(s->fd, type, arg); | |
a426e122 | 1018 | if (ret == -1) { |
05330448 | 1019 | ret = -errno; |
a426e122 | 1020 | } |
05330448 AL |
1021 | return ret; |
1022 | } | |
1023 | ||
984b5181 | 1024 | int kvm_vm_ioctl(KVMState *s, int type, ...) |
05330448 AL |
1025 | { |
1026 | int ret; | |
984b5181 AL |
1027 | void *arg; |
1028 | va_list ap; | |
1029 | ||
1030 | va_start(ap, type); | |
1031 | arg = va_arg(ap, void *); | |
1032 | va_end(ap); | |
05330448 | 1033 | |
984b5181 | 1034 | ret = ioctl(s->vmfd, type, arg); |
a426e122 | 1035 | if (ret == -1) { |
05330448 | 1036 | ret = -errno; |
a426e122 | 1037 | } |
05330448 AL |
1038 | return ret; |
1039 | } | |
1040 | ||
984b5181 | 1041 | int kvm_vcpu_ioctl(CPUState *env, int type, ...) |
05330448 AL |
1042 | { |
1043 | int ret; | |
984b5181 AL |
1044 | void *arg; |
1045 | va_list ap; | |
1046 | ||
1047 | va_start(ap, type); | |
1048 | arg = va_arg(ap, void *); | |
1049 | va_end(ap); | |
05330448 | 1050 | |
984b5181 | 1051 | ret = ioctl(env->kvm_fd, type, arg); |
a426e122 | 1052 | if (ret == -1) { |
05330448 | 1053 | ret = -errno; |
a426e122 | 1054 | } |
05330448 AL |
1055 | return ret; |
1056 | } | |
bd322087 AL |
1057 | |
1058 | int kvm_has_sync_mmu(void) | |
1059 | { | |
94a8d39a | 1060 | return kvm_check_extension(kvm_state, KVM_CAP_SYNC_MMU); |
bd322087 | 1061 | } |
e22a25c9 | 1062 | |
a0fb002c JK |
1063 | int kvm_has_vcpu_events(void) |
1064 | { | |
1065 | return kvm_state->vcpu_events; | |
1066 | } | |
1067 | ||
b0b1d690 JK |
1068 | int kvm_has_robust_singlestep(void) |
1069 | { | |
1070 | return kvm_state->robust_singlestep; | |
1071 | } | |
1072 | ||
ff44f1a3 JK |
1073 | int kvm_has_debugregs(void) |
1074 | { | |
1075 | return kvm_state->debugregs; | |
1076 | } | |
1077 | ||
f1665b21 SY |
1078 | int kvm_has_xsave(void) |
1079 | { | |
1080 | return kvm_state->xsave; | |
1081 | } | |
1082 | ||
1083 | int kvm_has_xcrs(void) | |
1084 | { | |
1085 | return kvm_state->xcrs; | |
1086 | } | |
1087 | ||
d2f2b8a7 SH |
1088 | int kvm_has_many_ioeventfds(void) |
1089 | { | |
1090 | if (!kvm_enabled()) { | |
1091 | return 0; | |
1092 | } | |
1093 | return kvm_state->many_ioeventfds; | |
1094 | } | |
1095 | ||
6f0437e8 JK |
1096 | void kvm_setup_guest_memory(void *start, size_t size) |
1097 | { | |
1098 | if (!kvm_has_sync_mmu()) { | |
e78815a5 | 1099 | int ret = qemu_madvise(start, size, QEMU_MADV_DONTFORK); |
6f0437e8 JK |
1100 | |
1101 | if (ret) { | |
e78815a5 AF |
1102 | perror("qemu_madvise"); |
1103 | fprintf(stderr, | |
1104 | "Need MADV_DONTFORK in absence of synchronous KVM MMU\n"); | |
6f0437e8 JK |
1105 | exit(1); |
1106 | } | |
6f0437e8 JK |
1107 | } |
1108 | } | |
1109 | ||
e22a25c9 AL |
1110 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
1111 | struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *env, | |
1112 | target_ulong pc) | |
1113 | { | |
1114 | struct kvm_sw_breakpoint *bp; | |
1115 | ||
72cf2d4f | 1116 | QTAILQ_FOREACH(bp, &env->kvm_state->kvm_sw_breakpoints, entry) { |
a426e122 | 1117 | if (bp->pc == pc) { |
e22a25c9 | 1118 | return bp; |
a426e122 | 1119 | } |
e22a25c9 AL |
1120 | } |
1121 | return NULL; | |
1122 | } | |
1123 | ||
1124 | int kvm_sw_breakpoints_active(CPUState *env) | |
1125 | { | |
72cf2d4f | 1126 | return !QTAILQ_EMPTY(&env->kvm_state->kvm_sw_breakpoints); |
e22a25c9 AL |
1127 | } |
1128 | ||
452e4751 GC |
1129 | struct kvm_set_guest_debug_data { |
1130 | struct kvm_guest_debug dbg; | |
1131 | CPUState *env; | |
1132 | int err; | |
1133 | }; | |
1134 | ||
1135 | static void kvm_invoke_set_guest_debug(void *data) | |
1136 | { | |
1137 | struct kvm_set_guest_debug_data *dbg_data = data; | |
b3807725 JK |
1138 | CPUState *env = dbg_data->env; |
1139 | ||
b3807725 | 1140 | dbg_data->err = kvm_vcpu_ioctl(env, KVM_SET_GUEST_DEBUG, &dbg_data->dbg); |
452e4751 GC |
1141 | } |
1142 | ||
e22a25c9 AL |
1143 | int kvm_update_guest_debug(CPUState *env, unsigned long reinject_trap) |
1144 | { | |
452e4751 | 1145 | struct kvm_set_guest_debug_data data; |
e22a25c9 | 1146 | |
b0b1d690 | 1147 | data.dbg.control = reinject_trap; |
e22a25c9 | 1148 | |
b0b1d690 JK |
1149 | if (env->singlestep_enabled) { |
1150 | data.dbg.control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP; | |
1151 | } | |
452e4751 | 1152 | kvm_arch_update_guest_debug(env, &data.dbg); |
452e4751 | 1153 | data.env = env; |
e22a25c9 | 1154 | |
be41cbe0 | 1155 | run_on_cpu(env, kvm_invoke_set_guest_debug, &data); |
452e4751 | 1156 | return data.err; |
e22a25c9 AL |
1157 | } |
1158 | ||
1159 | int kvm_insert_breakpoint(CPUState *current_env, target_ulong addr, | |
1160 | target_ulong len, int type) | |
1161 | { | |
1162 | struct kvm_sw_breakpoint *bp; | |
1163 | CPUState *env; | |
1164 | int err; | |
1165 | ||
1166 | if (type == GDB_BREAKPOINT_SW) { | |
1167 | bp = kvm_find_sw_breakpoint(current_env, addr); | |
1168 | if (bp) { | |
1169 | bp->use_count++; | |
1170 | return 0; | |
1171 | } | |
1172 | ||
1173 | bp = qemu_malloc(sizeof(struct kvm_sw_breakpoint)); | |
a426e122 | 1174 | if (!bp) { |
e22a25c9 | 1175 | return -ENOMEM; |
a426e122 | 1176 | } |
e22a25c9 AL |
1177 | |
1178 | bp->pc = addr; | |
1179 | bp->use_count = 1; | |
1180 | err = kvm_arch_insert_sw_breakpoint(current_env, bp); | |
1181 | if (err) { | |
1182 | free(bp); | |
1183 | return err; | |
1184 | } | |
1185 | ||
72cf2d4f | 1186 | QTAILQ_INSERT_HEAD(¤t_env->kvm_state->kvm_sw_breakpoints, |
e22a25c9 AL |
1187 | bp, entry); |
1188 | } else { | |
1189 | err = kvm_arch_insert_hw_breakpoint(addr, len, type); | |
a426e122 | 1190 | if (err) { |
e22a25c9 | 1191 | return err; |
a426e122 | 1192 | } |
e22a25c9 AL |
1193 | } |
1194 | ||
1195 | for (env = first_cpu; env != NULL; env = env->next_cpu) { | |
1196 | err = kvm_update_guest_debug(env, 0); | |
a426e122 | 1197 | if (err) { |
e22a25c9 | 1198 | return err; |
a426e122 | 1199 | } |
e22a25c9 AL |
1200 | } |
1201 | return 0; | |
1202 | } | |
1203 | ||
1204 | int kvm_remove_breakpoint(CPUState *current_env, target_ulong addr, | |
1205 | target_ulong len, int type) | |
1206 | { | |
1207 | struct kvm_sw_breakpoint *bp; | |
1208 | CPUState *env; | |
1209 | int err; | |
1210 | ||
1211 | if (type == GDB_BREAKPOINT_SW) { | |
1212 | bp = kvm_find_sw_breakpoint(current_env, addr); | |
a426e122 | 1213 | if (!bp) { |
e22a25c9 | 1214 | return -ENOENT; |
a426e122 | 1215 | } |
e22a25c9 AL |
1216 | |
1217 | if (bp->use_count > 1) { | |
1218 | bp->use_count--; | |
1219 | return 0; | |
1220 | } | |
1221 | ||
1222 | err = kvm_arch_remove_sw_breakpoint(current_env, bp); | |
a426e122 | 1223 | if (err) { |
e22a25c9 | 1224 | return err; |
a426e122 | 1225 | } |
e22a25c9 | 1226 | |
72cf2d4f | 1227 | QTAILQ_REMOVE(¤t_env->kvm_state->kvm_sw_breakpoints, bp, entry); |
e22a25c9 AL |
1228 | qemu_free(bp); |
1229 | } else { | |
1230 | err = kvm_arch_remove_hw_breakpoint(addr, len, type); | |
a426e122 | 1231 | if (err) { |
e22a25c9 | 1232 | return err; |
a426e122 | 1233 | } |
e22a25c9 AL |
1234 | } |
1235 | ||
1236 | for (env = first_cpu; env != NULL; env = env->next_cpu) { | |
1237 | err = kvm_update_guest_debug(env, 0); | |
a426e122 | 1238 | if (err) { |
e22a25c9 | 1239 | return err; |
a426e122 | 1240 | } |
e22a25c9 AL |
1241 | } |
1242 | return 0; | |
1243 | } | |
1244 | ||
1245 | void kvm_remove_all_breakpoints(CPUState *current_env) | |
1246 | { | |
1247 | struct kvm_sw_breakpoint *bp, *next; | |
1248 | KVMState *s = current_env->kvm_state; | |
1249 | CPUState *env; | |
1250 | ||
72cf2d4f | 1251 | QTAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) { |
e22a25c9 AL |
1252 | if (kvm_arch_remove_sw_breakpoint(current_env, bp) != 0) { |
1253 | /* Try harder to find a CPU that currently sees the breakpoint. */ | |
1254 | for (env = first_cpu; env != NULL; env = env->next_cpu) { | |
a426e122 | 1255 | if (kvm_arch_remove_sw_breakpoint(env, bp) == 0) { |
e22a25c9 | 1256 | break; |
a426e122 | 1257 | } |
e22a25c9 AL |
1258 | } |
1259 | } | |
1260 | } | |
1261 | kvm_arch_remove_all_hw_breakpoints(); | |
1262 | ||
a426e122 | 1263 | for (env = first_cpu; env != NULL; env = env->next_cpu) { |
e22a25c9 | 1264 | kvm_update_guest_debug(env, 0); |
a426e122 | 1265 | } |
e22a25c9 AL |
1266 | } |
1267 | ||
1268 | #else /* !KVM_CAP_SET_GUEST_DEBUG */ | |
1269 | ||
1270 | int kvm_update_guest_debug(CPUState *env, unsigned long reinject_trap) | |
1271 | { | |
1272 | return -EINVAL; | |
1273 | } | |
1274 | ||
1275 | int kvm_insert_breakpoint(CPUState *current_env, target_ulong addr, | |
1276 | target_ulong len, int type) | |
1277 | { | |
1278 | return -EINVAL; | |
1279 | } | |
1280 | ||
1281 | int kvm_remove_breakpoint(CPUState *current_env, target_ulong addr, | |
1282 | target_ulong len, int type) | |
1283 | { | |
1284 | return -EINVAL; | |
1285 | } | |
1286 | ||
1287 | void kvm_remove_all_breakpoints(CPUState *current_env) | |
1288 | { | |
1289 | } | |
1290 | #endif /* !KVM_CAP_SET_GUEST_DEBUG */ | |
cc84de95 MT |
1291 | |
1292 | int kvm_set_signal_mask(CPUState *env, const sigset_t *sigset) | |
1293 | { | |
1294 | struct kvm_signal_mask *sigmask; | |
1295 | int r; | |
1296 | ||
a426e122 | 1297 | if (!sigset) { |
cc84de95 | 1298 | return kvm_vcpu_ioctl(env, KVM_SET_SIGNAL_MASK, NULL); |
a426e122 | 1299 | } |
cc84de95 MT |
1300 | |
1301 | sigmask = qemu_malloc(sizeof(*sigmask) + sizeof(*sigset)); | |
1302 | ||
1303 | sigmask->len = 8; | |
1304 | memcpy(sigmask->sigset, sigset, sizeof(*sigset)); | |
1305 | r = kvm_vcpu_ioctl(env, KVM_SET_SIGNAL_MASK, sigmask); | |
1306 | free(sigmask); | |
1307 | ||
1308 | return r; | |
1309 | } | |
ca821806 | 1310 | |
44f1a3d8 CM |
1311 | int kvm_set_ioeventfd_mmio_long(int fd, uint32_t addr, uint32_t val, bool assign) |
1312 | { | |
1313 | #ifdef KVM_IOEVENTFD | |
1314 | int ret; | |
1315 | struct kvm_ioeventfd iofd; | |
1316 | ||
1317 | iofd.datamatch = val; | |
1318 | iofd.addr = addr; | |
1319 | iofd.len = 4; | |
1320 | iofd.flags = KVM_IOEVENTFD_FLAG_DATAMATCH; | |
1321 | iofd.fd = fd; | |
1322 | ||
1323 | if (!kvm_enabled()) { | |
1324 | return -ENOSYS; | |
1325 | } | |
1326 | ||
1327 | if (!assign) { | |
1328 | iofd.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN; | |
1329 | } | |
1330 | ||
1331 | ret = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &iofd); | |
1332 | ||
1333 | if (ret < 0) { | |
1334 | return -errno; | |
1335 | } | |
1336 | ||
1337 | return 0; | |
1338 | #else | |
1339 | return -ENOSYS; | |
1340 | #endif | |
1341 | } | |
1342 | ||
ca821806 MT |
1343 | int kvm_set_ioeventfd_pio_word(int fd, uint16_t addr, uint16_t val, bool assign) |
1344 | { | |
98c8573e | 1345 | #ifdef KVM_IOEVENTFD |
ca821806 MT |
1346 | struct kvm_ioeventfd kick = { |
1347 | .datamatch = val, | |
1348 | .addr = addr, | |
1349 | .len = 2, | |
1350 | .flags = KVM_IOEVENTFD_FLAG_DATAMATCH | KVM_IOEVENTFD_FLAG_PIO, | |
1351 | .fd = fd, | |
1352 | }; | |
1353 | int r; | |
a426e122 | 1354 | if (!kvm_enabled()) { |
ca821806 | 1355 | return -ENOSYS; |
a426e122 JK |
1356 | } |
1357 | if (!assign) { | |
ca821806 | 1358 | kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN; |
a426e122 | 1359 | } |
ca821806 | 1360 | r = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick); |
a426e122 | 1361 | if (r < 0) { |
ca821806 | 1362 | return r; |
a426e122 | 1363 | } |
ca821806 | 1364 | return 0; |
98c8573e PB |
1365 | #else |
1366 | return -ENOSYS; | |
ca821806 | 1367 | #endif |
98c8573e | 1368 | } |
a1b87fe0 JK |
1369 | |
1370 | int kvm_on_sigbus_vcpu(CPUState *env, int code, void *addr) | |
1371 | { | |
1372 | return kvm_arch_on_sigbus_vcpu(env, code, addr); | |
1373 | } | |
1374 | ||
1375 | int kvm_on_sigbus(int code, void *addr) | |
1376 | { | |
1377 | return kvm_arch_on_sigbus(code, addr); | |
1378 | } |