]>
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; | |
d841b6c4 | 214 | env->kvm_vcpu_dirty = 1; |
05330448 AL |
215 | |
216 | mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0); | |
217 | if (mmap_size < 0) { | |
748a680b | 218 | ret = mmap_size; |
8c0d577e | 219 | DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n"); |
05330448 AL |
220 | goto err; |
221 | } | |
222 | ||
223 | env->kvm_run = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED, | |
224 | env->kvm_fd, 0); | |
225 | if (env->kvm_run == MAP_FAILED) { | |
226 | ret = -errno; | |
8c0d577e | 227 | DPRINTF("mmap'ing vcpu state failed\n"); |
05330448 AL |
228 | goto err; |
229 | } | |
230 | ||
a426e122 JK |
231 | if (s->coalesced_mmio && !s->coalesced_mmio_ring) { |
232 | s->coalesced_mmio_ring = | |
233 | (void *)env->kvm_run + s->coalesced_mmio * PAGE_SIZE; | |
234 | } | |
62a2744c | 235 | |
05330448 | 236 | ret = kvm_arch_init_vcpu(env); |
8d2ba1fb | 237 | if (ret == 0) { |
a08d4367 | 238 | qemu_register_reset(kvm_reset_vcpu, env); |
caa5af0f | 239 | kvm_arch_reset_vcpu(env); |
8d2ba1fb | 240 | } |
05330448 AL |
241 | err: |
242 | return ret; | |
243 | } | |
244 | ||
5832d1f2 AL |
245 | /* |
246 | * dirty pages logging control | |
247 | */ | |
c227f099 AL |
248 | static int kvm_dirty_pages_log_change(target_phys_addr_t phys_addr, |
249 | ram_addr_t size, int flags, int mask) | |
5832d1f2 AL |
250 | { |
251 | KVMState *s = kvm_state; | |
d3f8d37f | 252 | KVMSlot *mem = kvm_lookup_matching_slot(s, phys_addr, phys_addr + size); |
4495d6a7 JK |
253 | int old_flags; |
254 | ||
5832d1f2 | 255 | if (mem == NULL) { |
d3f8d37f AL |
256 | fprintf(stderr, "BUG: %s: invalid parameters " TARGET_FMT_plx "-" |
257 | TARGET_FMT_plx "\n", __func__, phys_addr, | |
c227f099 | 258 | (target_phys_addr_t)(phys_addr + size - 1)); |
5832d1f2 AL |
259 | return -EINVAL; |
260 | } | |
261 | ||
4495d6a7 | 262 | old_flags = mem->flags; |
5832d1f2 | 263 | |
4495d6a7 | 264 | flags = (mem->flags & ~mask) | flags; |
5832d1f2 AL |
265 | mem->flags = flags; |
266 | ||
4495d6a7 JK |
267 | /* If nothing changed effectively, no need to issue ioctl */ |
268 | if (s->migration_log) { | |
269 | flags |= KVM_MEM_LOG_DIRTY_PAGES; | |
270 | } | |
271 | if (flags == old_flags) { | |
272 | return 0; | |
273 | } | |
274 | ||
5832d1f2 AL |
275 | return kvm_set_user_memory_region(s, mem); |
276 | } | |
277 | ||
e5896b12 AP |
278 | static int kvm_log_start(CPUPhysMemoryClient *client, |
279 | target_phys_addr_t phys_addr, ram_addr_t size) | |
5832d1f2 | 280 | { |
a426e122 JK |
281 | return kvm_dirty_pages_log_change(phys_addr, size, KVM_MEM_LOG_DIRTY_PAGES, |
282 | KVM_MEM_LOG_DIRTY_PAGES); | |
5832d1f2 AL |
283 | } |
284 | ||
e5896b12 AP |
285 | static int kvm_log_stop(CPUPhysMemoryClient *client, |
286 | target_phys_addr_t phys_addr, ram_addr_t size) | |
5832d1f2 | 287 | { |
a426e122 JK |
288 | return kvm_dirty_pages_log_change(phys_addr, size, 0, |
289 | KVM_MEM_LOG_DIRTY_PAGES); | |
5832d1f2 AL |
290 | } |
291 | ||
7b8f3b78 | 292 | static int kvm_set_migration_log(int enable) |
4495d6a7 JK |
293 | { |
294 | KVMState *s = kvm_state; | |
295 | KVMSlot *mem; | |
296 | int i, err; | |
297 | ||
298 | s->migration_log = enable; | |
299 | ||
300 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
301 | mem = &s->slots[i]; | |
302 | ||
70fedd76 AW |
303 | if (!mem->memory_size) { |
304 | continue; | |
305 | } | |
4495d6a7 JK |
306 | if (!!(mem->flags & KVM_MEM_LOG_DIRTY_PAGES) == enable) { |
307 | continue; | |
308 | } | |
309 | err = kvm_set_user_memory_region(s, mem); | |
310 | if (err) { | |
311 | return err; | |
312 | } | |
313 | } | |
314 | return 0; | |
315 | } | |
316 | ||
8369e01c MT |
317 | /* get kvm's dirty pages bitmap and update qemu's */ |
318 | static int kvm_get_dirty_pages_log_range(unsigned long start_addr, | |
319 | unsigned long *bitmap, | |
320 | unsigned long offset, | |
321 | unsigned long mem_size) | |
96c1606b | 322 | { |
8369e01c MT |
323 | unsigned int i, j; |
324 | unsigned long page_number, addr, addr1, c; | |
325 | ram_addr_t ram_addr; | |
326 | unsigned int len = ((mem_size / TARGET_PAGE_SIZE) + HOST_LONG_BITS - 1) / | |
327 | HOST_LONG_BITS; | |
328 | ||
329 | /* | |
330 | * bitmap-traveling is faster than memory-traveling (for addr...) | |
331 | * especially when most of the memory is not dirty. | |
332 | */ | |
333 | for (i = 0; i < len; i++) { | |
334 | if (bitmap[i] != 0) { | |
335 | c = leul_to_cpu(bitmap[i]); | |
336 | do { | |
337 | j = ffsl(c) - 1; | |
338 | c &= ~(1ul << j); | |
339 | page_number = i * HOST_LONG_BITS + j; | |
340 | addr1 = page_number * TARGET_PAGE_SIZE; | |
341 | addr = offset + addr1; | |
342 | ram_addr = cpu_get_physical_page_desc(addr); | |
343 | cpu_physical_memory_set_dirty(ram_addr); | |
344 | } while (c != 0); | |
345 | } | |
346 | } | |
347 | return 0; | |
96c1606b AG |
348 | } |
349 | ||
8369e01c MT |
350 | #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1)) |
351 | ||
5832d1f2 AL |
352 | /** |
353 | * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space | |
354 | * This function updates qemu's dirty bitmap using cpu_physical_memory_set_dirty(). | |
355 | * This means all bits are set to dirty. | |
356 | * | |
d3f8d37f | 357 | * @start_add: start of logged region. |
5832d1f2 AL |
358 | * @end_addr: end of logged region. |
359 | */ | |
7b8f3b78 | 360 | static int kvm_physical_sync_dirty_bitmap(target_phys_addr_t start_addr, |
a426e122 | 361 | target_phys_addr_t end_addr) |
5832d1f2 AL |
362 | { |
363 | KVMState *s = kvm_state; | |
151f7749 | 364 | unsigned long size, allocated_size = 0; |
151f7749 JK |
365 | KVMDirtyLog d; |
366 | KVMSlot *mem; | |
367 | int ret = 0; | |
5832d1f2 | 368 | |
151f7749 JK |
369 | d.dirty_bitmap = NULL; |
370 | while (start_addr < end_addr) { | |
371 | mem = kvm_lookup_overlapping_slot(s, start_addr, end_addr); | |
372 | if (mem == NULL) { | |
373 | break; | |
374 | } | |
5832d1f2 | 375 | |
51b0c606 MT |
376 | /* XXX bad kernel interface alert |
377 | * For dirty bitmap, kernel allocates array of size aligned to | |
378 | * bits-per-long. But for case when the kernel is 64bits and | |
379 | * the userspace is 32bits, userspace can't align to the same | |
380 | * bits-per-long, since sizeof(long) is different between kernel | |
381 | * and user space. This way, userspace will provide buffer which | |
382 | * may be 4 bytes less than the kernel will use, resulting in | |
383 | * userspace memory corruption (which is not detectable by valgrind | |
384 | * too, in most cases). | |
385 | * So for now, let's align to 64 instead of HOST_LONG_BITS here, in | |
386 | * a hope that sizeof(long) wont become >8 any time soon. | |
387 | */ | |
388 | size = ALIGN(((mem->memory_size) >> TARGET_PAGE_BITS), | |
389 | /*HOST_LONG_BITS*/ 64) / 8; | |
151f7749 JK |
390 | if (!d.dirty_bitmap) { |
391 | d.dirty_bitmap = qemu_malloc(size); | |
392 | } else if (size > allocated_size) { | |
393 | d.dirty_bitmap = qemu_realloc(d.dirty_bitmap, size); | |
394 | } | |
395 | allocated_size = size; | |
396 | memset(d.dirty_bitmap, 0, allocated_size); | |
5832d1f2 | 397 | |
151f7749 | 398 | d.slot = mem->slot; |
5832d1f2 | 399 | |
6e489f3f | 400 | if (kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d) == -1) { |
8c0d577e | 401 | DPRINTF("ioctl failed %d\n", errno); |
151f7749 JK |
402 | ret = -1; |
403 | break; | |
404 | } | |
5832d1f2 | 405 | |
8369e01c MT |
406 | kvm_get_dirty_pages_log_range(mem->start_addr, d.dirty_bitmap, |
407 | mem->start_addr, mem->memory_size); | |
408 | start_addr = mem->start_addr + mem->memory_size; | |
5832d1f2 | 409 | } |
5832d1f2 | 410 | qemu_free(d.dirty_bitmap); |
151f7749 JK |
411 | |
412 | return ret; | |
5832d1f2 AL |
413 | } |
414 | ||
c227f099 | 415 | int kvm_coalesce_mmio_region(target_phys_addr_t start, ram_addr_t size) |
f65ed4c1 AL |
416 | { |
417 | int ret = -ENOSYS; | |
f65ed4c1 AL |
418 | KVMState *s = kvm_state; |
419 | ||
420 | if (s->coalesced_mmio) { | |
421 | struct kvm_coalesced_mmio_zone zone; | |
422 | ||
423 | zone.addr = start; | |
424 | zone.size = size; | |
425 | ||
426 | ret = kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone); | |
427 | } | |
f65ed4c1 AL |
428 | |
429 | return ret; | |
430 | } | |
431 | ||
c227f099 | 432 | int kvm_uncoalesce_mmio_region(target_phys_addr_t start, ram_addr_t size) |
f65ed4c1 AL |
433 | { |
434 | int ret = -ENOSYS; | |
f65ed4c1 AL |
435 | KVMState *s = kvm_state; |
436 | ||
437 | if (s->coalesced_mmio) { | |
438 | struct kvm_coalesced_mmio_zone zone; | |
439 | ||
440 | zone.addr = start; | |
441 | zone.size = size; | |
442 | ||
443 | ret = kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone); | |
444 | } | |
f65ed4c1 AL |
445 | |
446 | return ret; | |
447 | } | |
448 | ||
ad7b8b33 AL |
449 | int kvm_check_extension(KVMState *s, unsigned int extension) |
450 | { | |
451 | int ret; | |
452 | ||
453 | ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension); | |
454 | if (ret < 0) { | |
455 | ret = 0; | |
456 | } | |
457 | ||
458 | return ret; | |
459 | } | |
460 | ||
d2f2b8a7 SH |
461 | static int kvm_check_many_ioeventfds(void) |
462 | { | |
d0dcac83 SH |
463 | /* Userspace can use ioeventfd for io notification. This requires a host |
464 | * that supports eventfd(2) and an I/O thread; since eventfd does not | |
465 | * support SIGIO it cannot interrupt the vcpu. | |
466 | * | |
467 | * Older kernels have a 6 device limit on the KVM io bus. Find out so we | |
d2f2b8a7 SH |
468 | * can avoid creating too many ioeventfds. |
469 | */ | |
d0dcac83 | 470 | #if defined(CONFIG_EVENTFD) && defined(CONFIG_IOTHREAD) |
d2f2b8a7 SH |
471 | int ioeventfds[7]; |
472 | int i, ret = 0; | |
473 | for (i = 0; i < ARRAY_SIZE(ioeventfds); i++) { | |
474 | ioeventfds[i] = eventfd(0, EFD_CLOEXEC); | |
475 | if (ioeventfds[i] < 0) { | |
476 | break; | |
477 | } | |
478 | ret = kvm_set_ioeventfd_pio_word(ioeventfds[i], 0, i, true); | |
479 | if (ret < 0) { | |
480 | close(ioeventfds[i]); | |
481 | break; | |
482 | } | |
483 | } | |
484 | ||
485 | /* Decide whether many devices are supported or not */ | |
486 | ret = i == ARRAY_SIZE(ioeventfds); | |
487 | ||
488 | while (i-- > 0) { | |
489 | kvm_set_ioeventfd_pio_word(ioeventfds[i], 0, i, false); | |
490 | close(ioeventfds[i]); | |
491 | } | |
492 | return ret; | |
493 | #else | |
494 | return 0; | |
495 | #endif | |
496 | } | |
497 | ||
94a8d39a JK |
498 | static const KVMCapabilityInfo * |
499 | kvm_check_extension_list(KVMState *s, const KVMCapabilityInfo *list) | |
500 | { | |
501 | while (list->name) { | |
502 | if (!kvm_check_extension(s, list->value)) { | |
503 | return list; | |
504 | } | |
505 | list++; | |
506 | } | |
507 | return NULL; | |
508 | } | |
509 | ||
a426e122 JK |
510 | static void kvm_set_phys_mem(target_phys_addr_t start_addr, ram_addr_t size, |
511 | ram_addr_t phys_offset) | |
46dbef6a MT |
512 | { |
513 | KVMState *s = kvm_state; | |
514 | ram_addr_t flags = phys_offset & ~TARGET_PAGE_MASK; | |
515 | KVMSlot *mem, old; | |
516 | int err; | |
517 | ||
14542fea GN |
518 | /* kvm works in page size chunks, but the function may be called |
519 | with sub-page size and unaligned start address. */ | |
520 | size = TARGET_PAGE_ALIGN(size); | |
521 | start_addr = TARGET_PAGE_ALIGN(start_addr); | |
46dbef6a MT |
522 | |
523 | /* KVM does not support read-only slots */ | |
524 | phys_offset &= ~IO_MEM_ROM; | |
525 | ||
526 | while (1) { | |
527 | mem = kvm_lookup_overlapping_slot(s, start_addr, start_addr + size); | |
528 | if (!mem) { | |
529 | break; | |
530 | } | |
531 | ||
532 | if (flags < IO_MEM_UNASSIGNED && start_addr >= mem->start_addr && | |
533 | (start_addr + size <= mem->start_addr + mem->memory_size) && | |
534 | (phys_offset - start_addr == mem->phys_offset - mem->start_addr)) { | |
535 | /* The new slot fits into the existing one and comes with | |
536 | * identical parameters - nothing to be done. */ | |
537 | return; | |
538 | } | |
539 | ||
540 | old = *mem; | |
541 | ||
542 | /* unregister the overlapping slot */ | |
543 | mem->memory_size = 0; | |
544 | err = kvm_set_user_memory_region(s, mem); | |
545 | if (err) { | |
546 | fprintf(stderr, "%s: error unregistering overlapping slot: %s\n", | |
547 | __func__, strerror(-err)); | |
548 | abort(); | |
549 | } | |
550 | ||
551 | /* Workaround for older KVM versions: we can't join slots, even not by | |
552 | * unregistering the previous ones and then registering the larger | |
553 | * slot. We have to maintain the existing fragmentation. Sigh. | |
554 | * | |
555 | * This workaround assumes that the new slot starts at the same | |
556 | * address as the first existing one. If not or if some overlapping | |
557 | * slot comes around later, we will fail (not seen in practice so far) | |
558 | * - and actually require a recent KVM version. */ | |
559 | if (s->broken_set_mem_region && | |
560 | old.start_addr == start_addr && old.memory_size < size && | |
561 | flags < IO_MEM_UNASSIGNED) { | |
562 | mem = kvm_alloc_slot(s); | |
563 | mem->memory_size = old.memory_size; | |
564 | mem->start_addr = old.start_addr; | |
565 | mem->phys_offset = old.phys_offset; | |
566 | mem->flags = 0; | |
567 | ||
568 | err = kvm_set_user_memory_region(s, mem); | |
569 | if (err) { | |
570 | fprintf(stderr, "%s: error updating slot: %s\n", __func__, | |
571 | strerror(-err)); | |
572 | abort(); | |
573 | } | |
574 | ||
575 | start_addr += old.memory_size; | |
576 | phys_offset += old.memory_size; | |
577 | size -= old.memory_size; | |
578 | continue; | |
579 | } | |
580 | ||
581 | /* register prefix slot */ | |
582 | if (old.start_addr < start_addr) { | |
583 | mem = kvm_alloc_slot(s); | |
584 | mem->memory_size = start_addr - old.start_addr; | |
585 | mem->start_addr = old.start_addr; | |
586 | mem->phys_offset = old.phys_offset; | |
587 | mem->flags = 0; | |
588 | ||
589 | err = kvm_set_user_memory_region(s, mem); | |
590 | if (err) { | |
591 | fprintf(stderr, "%s: error registering prefix slot: %s\n", | |
592 | __func__, strerror(-err)); | |
593 | abort(); | |
594 | } | |
595 | } | |
596 | ||
597 | /* register suffix slot */ | |
598 | if (old.start_addr + old.memory_size > start_addr + size) { | |
599 | ram_addr_t size_delta; | |
600 | ||
601 | mem = kvm_alloc_slot(s); | |
602 | mem->start_addr = start_addr + size; | |
603 | size_delta = mem->start_addr - old.start_addr; | |
604 | mem->memory_size = old.memory_size - size_delta; | |
605 | mem->phys_offset = old.phys_offset + size_delta; | |
606 | mem->flags = 0; | |
607 | ||
608 | err = kvm_set_user_memory_region(s, mem); | |
609 | if (err) { | |
610 | fprintf(stderr, "%s: error registering suffix slot: %s\n", | |
611 | __func__, strerror(-err)); | |
612 | abort(); | |
613 | } | |
614 | } | |
615 | } | |
616 | ||
617 | /* in case the KVM bug workaround already "consumed" the new slot */ | |
a426e122 | 618 | if (!size) { |
46dbef6a | 619 | return; |
a426e122 | 620 | } |
46dbef6a | 621 | /* KVM does not need to know about this memory */ |
a426e122 | 622 | if (flags >= IO_MEM_UNASSIGNED) { |
46dbef6a | 623 | return; |
a426e122 | 624 | } |
46dbef6a MT |
625 | mem = kvm_alloc_slot(s); |
626 | mem->memory_size = size; | |
627 | mem->start_addr = start_addr; | |
628 | mem->phys_offset = phys_offset; | |
629 | mem->flags = 0; | |
630 | ||
631 | err = kvm_set_user_memory_region(s, mem); | |
632 | if (err) { | |
633 | fprintf(stderr, "%s: error registering slot: %s\n", __func__, | |
634 | strerror(-err)); | |
635 | abort(); | |
636 | } | |
637 | } | |
638 | ||
7b8f3b78 | 639 | static void kvm_client_set_memory(struct CPUPhysMemoryClient *client, |
a426e122 JK |
640 | target_phys_addr_t start_addr, |
641 | ram_addr_t size, ram_addr_t phys_offset) | |
7b8f3b78 | 642 | { |
a426e122 | 643 | kvm_set_phys_mem(start_addr, size, phys_offset); |
7b8f3b78 MT |
644 | } |
645 | ||
646 | static int kvm_client_sync_dirty_bitmap(struct CPUPhysMemoryClient *client, | |
a426e122 JK |
647 | target_phys_addr_t start_addr, |
648 | target_phys_addr_t end_addr) | |
7b8f3b78 | 649 | { |
a426e122 | 650 | return kvm_physical_sync_dirty_bitmap(start_addr, end_addr); |
7b8f3b78 MT |
651 | } |
652 | ||
653 | static int kvm_client_migration_log(struct CPUPhysMemoryClient *client, | |
a426e122 | 654 | int enable) |
7b8f3b78 | 655 | { |
a426e122 | 656 | return kvm_set_migration_log(enable); |
7b8f3b78 MT |
657 | } |
658 | ||
659 | static CPUPhysMemoryClient kvm_cpu_phys_memory_client = { | |
a426e122 JK |
660 | .set_memory = kvm_client_set_memory, |
661 | .sync_dirty_bitmap = kvm_client_sync_dirty_bitmap, | |
662 | .migration_log = kvm_client_migration_log, | |
e5896b12 AP |
663 | .log_start = kvm_log_start, |
664 | .log_stop = kvm_log_stop, | |
7b8f3b78 MT |
665 | }; |
666 | ||
aa7f74d1 JK |
667 | static void kvm_handle_interrupt(CPUState *env, int mask) |
668 | { | |
669 | env->interrupt_request |= mask; | |
670 | ||
671 | if (!qemu_cpu_is_self(env)) { | |
672 | qemu_cpu_kick(env); | |
673 | } | |
674 | } | |
675 | ||
cad1e282 | 676 | int kvm_init(void) |
05330448 | 677 | { |
168ccc11 JK |
678 | static const char upgrade_note[] = |
679 | "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n" | |
680 | "(see http://sourceforge.net/projects/kvm).\n"; | |
05330448 | 681 | KVMState *s; |
94a8d39a | 682 | const KVMCapabilityInfo *missing_cap; |
05330448 AL |
683 | int ret; |
684 | int i; | |
685 | ||
05330448 | 686 | s = qemu_mallocz(sizeof(KVMState)); |
05330448 | 687 | |
e22a25c9 | 688 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
72cf2d4f | 689 | QTAILQ_INIT(&s->kvm_sw_breakpoints); |
e22a25c9 | 690 | #endif |
a426e122 | 691 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { |
05330448 | 692 | s->slots[i].slot = i; |
a426e122 | 693 | } |
05330448 | 694 | s->vmfd = -1; |
40ff6d7e | 695 | s->fd = qemu_open("/dev/kvm", O_RDWR); |
05330448 AL |
696 | if (s->fd == -1) { |
697 | fprintf(stderr, "Could not access KVM kernel module: %m\n"); | |
698 | ret = -errno; | |
699 | goto err; | |
700 | } | |
701 | ||
702 | ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0); | |
703 | if (ret < KVM_API_VERSION) { | |
a426e122 | 704 | if (ret > 0) { |
05330448 | 705 | ret = -EINVAL; |
a426e122 | 706 | } |
05330448 AL |
707 | fprintf(stderr, "kvm version too old\n"); |
708 | goto err; | |
709 | } | |
710 | ||
711 | if (ret > KVM_API_VERSION) { | |
712 | ret = -EINVAL; | |
713 | fprintf(stderr, "kvm version not supported\n"); | |
714 | goto err; | |
715 | } | |
716 | ||
717 | s->vmfd = kvm_ioctl(s, KVM_CREATE_VM, 0); | |
0104dcac AG |
718 | if (s->vmfd < 0) { |
719 | #ifdef TARGET_S390X | |
720 | fprintf(stderr, "Please add the 'switch_amode' kernel parameter to " | |
721 | "your host kernel command line\n"); | |
722 | #endif | |
05330448 | 723 | goto err; |
0104dcac | 724 | } |
05330448 | 725 | |
94a8d39a JK |
726 | missing_cap = kvm_check_extension_list(s, kvm_required_capabilites); |
727 | if (!missing_cap) { | |
728 | missing_cap = | |
729 | kvm_check_extension_list(s, kvm_arch_required_capabilities); | |
05330448 | 730 | } |
94a8d39a | 731 | if (missing_cap) { |
ad7b8b33 | 732 | ret = -EINVAL; |
94a8d39a JK |
733 | fprintf(stderr, "kvm does not support %s\n%s", |
734 | missing_cap->name, upgrade_note); | |
d85dc283 AL |
735 | goto err; |
736 | } | |
737 | ||
ad7b8b33 | 738 | s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO); |
f65ed4c1 | 739 | |
e69917e2 JK |
740 | s->broken_set_mem_region = 1; |
741 | #ifdef KVM_CAP_JOIN_MEMORY_REGIONS_WORKS | |
14a09518 | 742 | ret = kvm_check_extension(s, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS); |
e69917e2 JK |
743 | if (ret > 0) { |
744 | s->broken_set_mem_region = 0; | |
745 | } | |
746 | #endif | |
747 | ||
a0fb002c JK |
748 | s->vcpu_events = 0; |
749 | #ifdef KVM_CAP_VCPU_EVENTS | |
750 | s->vcpu_events = kvm_check_extension(s, KVM_CAP_VCPU_EVENTS); | |
751 | #endif | |
752 | ||
b0b1d690 JK |
753 | s->robust_singlestep = 0; |
754 | #ifdef KVM_CAP_X86_ROBUST_SINGLESTEP | |
755 | s->robust_singlestep = | |
756 | kvm_check_extension(s, KVM_CAP_X86_ROBUST_SINGLESTEP); | |
757 | #endif | |
758 | ||
ff44f1a3 JK |
759 | s->debugregs = 0; |
760 | #ifdef KVM_CAP_DEBUGREGS | |
761 | s->debugregs = kvm_check_extension(s, KVM_CAP_DEBUGREGS); | |
762 | #endif | |
763 | ||
f1665b21 SY |
764 | s->xsave = 0; |
765 | #ifdef KVM_CAP_XSAVE | |
766 | s->xsave = kvm_check_extension(s, KVM_CAP_XSAVE); | |
767 | #endif | |
768 | ||
769 | s->xcrs = 0; | |
770 | #ifdef KVM_CAP_XCRS | |
771 | s->xcrs = kvm_check_extension(s, KVM_CAP_XCRS); | |
772 | #endif | |
773 | ||
cad1e282 | 774 | ret = kvm_arch_init(s); |
a426e122 | 775 | if (ret < 0) { |
05330448 | 776 | goto err; |
a426e122 | 777 | } |
05330448 AL |
778 | |
779 | kvm_state = s; | |
7b8f3b78 | 780 | cpu_register_phys_memory_client(&kvm_cpu_phys_memory_client); |
05330448 | 781 | |
d2f2b8a7 SH |
782 | s->many_ioeventfds = kvm_check_many_ioeventfds(); |
783 | ||
aa7f74d1 JK |
784 | cpu_interrupt_handler = kvm_handle_interrupt; |
785 | ||
05330448 AL |
786 | return 0; |
787 | ||
788 | err: | |
789 | if (s) { | |
a426e122 | 790 | if (s->vmfd != -1) { |
05330448 | 791 | close(s->vmfd); |
a426e122 JK |
792 | } |
793 | if (s->fd != -1) { | |
05330448 | 794 | close(s->fd); |
a426e122 | 795 | } |
05330448 AL |
796 | } |
797 | qemu_free(s); | |
798 | ||
799 | return ret; | |
800 | } | |
801 | ||
b30e93e9 JK |
802 | static void kvm_handle_io(uint16_t port, void *data, int direction, int size, |
803 | uint32_t count) | |
05330448 AL |
804 | { |
805 | int i; | |
806 | uint8_t *ptr = data; | |
807 | ||
808 | for (i = 0; i < count; i++) { | |
809 | if (direction == KVM_EXIT_IO_IN) { | |
810 | switch (size) { | |
811 | case 1: | |
afcea8cb | 812 | stb_p(ptr, cpu_inb(port)); |
05330448 AL |
813 | break; |
814 | case 2: | |
afcea8cb | 815 | stw_p(ptr, cpu_inw(port)); |
05330448 AL |
816 | break; |
817 | case 4: | |
afcea8cb | 818 | stl_p(ptr, cpu_inl(port)); |
05330448 AL |
819 | break; |
820 | } | |
821 | } else { | |
822 | switch (size) { | |
823 | case 1: | |
afcea8cb | 824 | cpu_outb(port, ldub_p(ptr)); |
05330448 AL |
825 | break; |
826 | case 2: | |
afcea8cb | 827 | cpu_outw(port, lduw_p(ptr)); |
05330448 AL |
828 | break; |
829 | case 4: | |
afcea8cb | 830 | cpu_outl(port, ldl_p(ptr)); |
05330448 AL |
831 | break; |
832 | } | |
833 | } | |
834 | ||
835 | ptr += size; | |
836 | } | |
05330448 AL |
837 | } |
838 | ||
7c80eef8 | 839 | #ifdef KVM_CAP_INTERNAL_ERROR_DATA |
73aaec4a | 840 | static int kvm_handle_internal_error(CPUState *env, struct kvm_run *run) |
7c80eef8 | 841 | { |
bb44e0d1 | 842 | fprintf(stderr, "KVM internal error."); |
7c80eef8 MT |
843 | if (kvm_check_extension(kvm_state, KVM_CAP_INTERNAL_ERROR_DATA)) { |
844 | int i; | |
845 | ||
bb44e0d1 | 846 | fprintf(stderr, " Suberror: %d\n", run->internal.suberror); |
7c80eef8 MT |
847 | for (i = 0; i < run->internal.ndata; ++i) { |
848 | fprintf(stderr, "extra data[%d]: %"PRIx64"\n", | |
849 | i, (uint64_t)run->internal.data[i]); | |
850 | } | |
bb44e0d1 JK |
851 | } else { |
852 | fprintf(stderr, "\n"); | |
7c80eef8 | 853 | } |
7c80eef8 MT |
854 | if (run->internal.suberror == KVM_INTERNAL_ERROR_EMULATION) { |
855 | fprintf(stderr, "emulation failure\n"); | |
a426e122 | 856 | if (!kvm_arch_stop_on_emulation_error(env)) { |
f5c848ee | 857 | cpu_dump_state(env, stderr, fprintf, CPU_DUMP_CODE); |
d73cd8f4 | 858 | return EXCP_INTERRUPT; |
a426e122 | 859 | } |
7c80eef8 MT |
860 | } |
861 | /* FIXME: Should trigger a qmp message to let management know | |
862 | * something went wrong. | |
863 | */ | |
73aaec4a | 864 | return -1; |
7c80eef8 MT |
865 | } |
866 | #endif | |
867 | ||
62a2744c | 868 | void kvm_flush_coalesced_mmio_buffer(void) |
f65ed4c1 | 869 | { |
f65ed4c1 | 870 | KVMState *s = kvm_state; |
62a2744c SY |
871 | if (s->coalesced_mmio_ring) { |
872 | struct kvm_coalesced_mmio_ring *ring = s->coalesced_mmio_ring; | |
f65ed4c1 AL |
873 | while (ring->first != ring->last) { |
874 | struct kvm_coalesced_mmio *ent; | |
875 | ||
876 | ent = &ring->coalesced_mmio[ring->first]; | |
877 | ||
878 | cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len); | |
85199474 | 879 | smp_wmb(); |
f65ed4c1 AL |
880 | ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX; |
881 | } | |
882 | } | |
f65ed4c1 AL |
883 | } |
884 | ||
2705d56a | 885 | static void do_kvm_cpu_synchronize_state(void *_env) |
4c0960c0 | 886 | { |
2705d56a JK |
887 | CPUState *env = _env; |
888 | ||
9ded2744 | 889 | if (!env->kvm_vcpu_dirty) { |
4c0960c0 | 890 | kvm_arch_get_registers(env); |
9ded2744 | 891 | env->kvm_vcpu_dirty = 1; |
4c0960c0 AK |
892 | } |
893 | } | |
894 | ||
2705d56a JK |
895 | void kvm_cpu_synchronize_state(CPUState *env) |
896 | { | |
a426e122 | 897 | if (!env->kvm_vcpu_dirty) { |
2705d56a | 898 | run_on_cpu(env, do_kvm_cpu_synchronize_state, env); |
a426e122 | 899 | } |
2705d56a JK |
900 | } |
901 | ||
ea375f9a JK |
902 | void kvm_cpu_synchronize_post_reset(CPUState *env) |
903 | { | |
904 | kvm_arch_put_registers(env, KVM_PUT_RESET_STATE); | |
905 | env->kvm_vcpu_dirty = 0; | |
906 | } | |
907 | ||
908 | void kvm_cpu_synchronize_post_init(CPUState *env) | |
909 | { | |
910 | kvm_arch_put_registers(env, KVM_PUT_FULL_STATE); | |
911 | env->kvm_vcpu_dirty = 0; | |
912 | } | |
913 | ||
05330448 AL |
914 | int kvm_cpu_exec(CPUState *env) |
915 | { | |
916 | struct kvm_run *run = env->kvm_run; | |
7cbb533f | 917 | int ret, run_ret; |
05330448 | 918 | |
8c0d577e | 919 | DPRINTF("kvm_cpu_exec()\n"); |
05330448 | 920 | |
99036865 | 921 | if (kvm_arch_process_async_events(env)) { |
9ccfac9e | 922 | env->exit_request = 0; |
6792a57b | 923 | return EXCP_HLT; |
9ccfac9e | 924 | } |
0af691d7 | 925 | |
6792a57b JK |
926 | cpu_single_env = env; |
927 | ||
9ccfac9e | 928 | do { |
9ded2744 | 929 | if (env->kvm_vcpu_dirty) { |
ea375f9a | 930 | kvm_arch_put_registers(env, KVM_PUT_RUNTIME_STATE); |
9ded2744 | 931 | env->kvm_vcpu_dirty = 0; |
4c0960c0 AK |
932 | } |
933 | ||
8c14c173 | 934 | kvm_arch_pre_run(env, run); |
9ccfac9e JK |
935 | if (env->exit_request) { |
936 | DPRINTF("interrupt exit requested\n"); | |
937 | /* | |
938 | * KVM requires us to reenter the kernel after IO exits to complete | |
939 | * instruction emulation. This self-signal will ensure that we | |
940 | * leave ASAP again. | |
941 | */ | |
942 | qemu_cpu_kick_self(); | |
943 | } | |
273faf1b | 944 | cpu_single_env = NULL; |
d549db5a | 945 | qemu_mutex_unlock_iothread(); |
9ccfac9e | 946 | |
7cbb533f | 947 | run_ret = kvm_vcpu_ioctl(env, KVM_RUN, 0); |
9ccfac9e | 948 | |
d549db5a | 949 | qemu_mutex_lock_iothread(); |
273faf1b | 950 | cpu_single_env = env; |
05330448 AL |
951 | kvm_arch_post_run(env, run); |
952 | ||
b0c883b5 JK |
953 | kvm_flush_coalesced_mmio_buffer(); |
954 | ||
7cbb533f | 955 | if (run_ret < 0) { |
dc77d341 JK |
956 | if (run_ret == -EINTR || run_ret == -EAGAIN) { |
957 | DPRINTF("io window exit\n"); | |
d73cd8f4 | 958 | ret = EXCP_INTERRUPT; |
dc77d341 JK |
959 | break; |
960 | } | |
7cbb533f | 961 | DPRINTF("kvm run failed %s\n", strerror(-run_ret)); |
05330448 AL |
962 | abort(); |
963 | } | |
964 | ||
05330448 AL |
965 | switch (run->exit_reason) { |
966 | case KVM_EXIT_IO: | |
8c0d577e | 967 | DPRINTF("handle_io\n"); |
b30e93e9 JK |
968 | kvm_handle_io(run->io.port, |
969 | (uint8_t *)run + run->io.data_offset, | |
970 | run->io.direction, | |
971 | run->io.size, | |
972 | run->io.count); | |
d73cd8f4 | 973 | ret = 0; |
05330448 AL |
974 | break; |
975 | case KVM_EXIT_MMIO: | |
8c0d577e | 976 | DPRINTF("handle_mmio\n"); |
05330448 AL |
977 | cpu_physical_memory_rw(run->mmio.phys_addr, |
978 | run->mmio.data, | |
979 | run->mmio.len, | |
980 | run->mmio.is_write); | |
d73cd8f4 | 981 | ret = 0; |
05330448 AL |
982 | break; |
983 | case KVM_EXIT_IRQ_WINDOW_OPEN: | |
8c0d577e | 984 | DPRINTF("irq_window_open\n"); |
d73cd8f4 | 985 | ret = EXCP_INTERRUPT; |
05330448 AL |
986 | break; |
987 | case KVM_EXIT_SHUTDOWN: | |
8c0d577e | 988 | DPRINTF("shutdown\n"); |
05330448 | 989 | qemu_system_reset_request(); |
d73cd8f4 | 990 | ret = EXCP_INTERRUPT; |
05330448 AL |
991 | break; |
992 | case KVM_EXIT_UNKNOWN: | |
bb44e0d1 JK |
993 | fprintf(stderr, "KVM: unknown exit, hardware reason %" PRIx64 "\n", |
994 | (uint64_t)run->hw.hardware_exit_reason); | |
73aaec4a | 995 | ret = -1; |
05330448 | 996 | break; |
7c80eef8 MT |
997 | #ifdef KVM_CAP_INTERNAL_ERROR_DATA |
998 | case KVM_EXIT_INTERNAL_ERROR: | |
73aaec4a | 999 | ret = kvm_handle_internal_error(env, run); |
7c80eef8 MT |
1000 | break; |
1001 | #endif | |
05330448 | 1002 | default: |
8c0d577e | 1003 | DPRINTF("kvm_arch_handle_exit\n"); |
05330448 AL |
1004 | ret = kvm_arch_handle_exit(env, run); |
1005 | break; | |
1006 | } | |
d73cd8f4 | 1007 | } while (ret == 0); |
05330448 | 1008 | |
73aaec4a | 1009 | if (ret < 0) { |
f5c848ee | 1010 | cpu_dump_state(env, stderr, fprintf, CPU_DUMP_CODE); |
e07bbac5 | 1011 | vm_stop(VMSTOP_PANIC); |
becfc390 AL |
1012 | } |
1013 | ||
6792a57b JK |
1014 | env->exit_request = 0; |
1015 | cpu_single_env = NULL; | |
05330448 AL |
1016 | return ret; |
1017 | } | |
1018 | ||
984b5181 | 1019 | int kvm_ioctl(KVMState *s, int type, ...) |
05330448 AL |
1020 | { |
1021 | int ret; | |
984b5181 AL |
1022 | void *arg; |
1023 | va_list ap; | |
05330448 | 1024 | |
984b5181 AL |
1025 | va_start(ap, type); |
1026 | arg = va_arg(ap, void *); | |
1027 | va_end(ap); | |
1028 | ||
1029 | ret = ioctl(s->fd, type, arg); | |
a426e122 | 1030 | if (ret == -1) { |
05330448 | 1031 | ret = -errno; |
a426e122 | 1032 | } |
05330448 AL |
1033 | return ret; |
1034 | } | |
1035 | ||
984b5181 | 1036 | int kvm_vm_ioctl(KVMState *s, int type, ...) |
05330448 AL |
1037 | { |
1038 | int ret; | |
984b5181 AL |
1039 | void *arg; |
1040 | va_list ap; | |
1041 | ||
1042 | va_start(ap, type); | |
1043 | arg = va_arg(ap, void *); | |
1044 | va_end(ap); | |
05330448 | 1045 | |
984b5181 | 1046 | ret = ioctl(s->vmfd, type, arg); |
a426e122 | 1047 | if (ret == -1) { |
05330448 | 1048 | ret = -errno; |
a426e122 | 1049 | } |
05330448 AL |
1050 | return ret; |
1051 | } | |
1052 | ||
984b5181 | 1053 | int kvm_vcpu_ioctl(CPUState *env, int type, ...) |
05330448 AL |
1054 | { |
1055 | int ret; | |
984b5181 AL |
1056 | void *arg; |
1057 | va_list ap; | |
1058 | ||
1059 | va_start(ap, type); | |
1060 | arg = va_arg(ap, void *); | |
1061 | va_end(ap); | |
05330448 | 1062 | |
984b5181 | 1063 | ret = ioctl(env->kvm_fd, type, arg); |
a426e122 | 1064 | if (ret == -1) { |
05330448 | 1065 | ret = -errno; |
a426e122 | 1066 | } |
05330448 AL |
1067 | return ret; |
1068 | } | |
bd322087 AL |
1069 | |
1070 | int kvm_has_sync_mmu(void) | |
1071 | { | |
94a8d39a | 1072 | return kvm_check_extension(kvm_state, KVM_CAP_SYNC_MMU); |
bd322087 | 1073 | } |
e22a25c9 | 1074 | |
a0fb002c JK |
1075 | int kvm_has_vcpu_events(void) |
1076 | { | |
1077 | return kvm_state->vcpu_events; | |
1078 | } | |
1079 | ||
b0b1d690 JK |
1080 | int kvm_has_robust_singlestep(void) |
1081 | { | |
1082 | return kvm_state->robust_singlestep; | |
1083 | } | |
1084 | ||
ff44f1a3 JK |
1085 | int kvm_has_debugregs(void) |
1086 | { | |
1087 | return kvm_state->debugregs; | |
1088 | } | |
1089 | ||
f1665b21 SY |
1090 | int kvm_has_xsave(void) |
1091 | { | |
1092 | return kvm_state->xsave; | |
1093 | } | |
1094 | ||
1095 | int kvm_has_xcrs(void) | |
1096 | { | |
1097 | return kvm_state->xcrs; | |
1098 | } | |
1099 | ||
d2f2b8a7 SH |
1100 | int kvm_has_many_ioeventfds(void) |
1101 | { | |
1102 | if (!kvm_enabled()) { | |
1103 | return 0; | |
1104 | } | |
1105 | return kvm_state->many_ioeventfds; | |
1106 | } | |
1107 | ||
6f0437e8 JK |
1108 | void kvm_setup_guest_memory(void *start, size_t size) |
1109 | { | |
1110 | if (!kvm_has_sync_mmu()) { | |
e78815a5 | 1111 | int ret = qemu_madvise(start, size, QEMU_MADV_DONTFORK); |
6f0437e8 JK |
1112 | |
1113 | if (ret) { | |
e78815a5 AF |
1114 | perror("qemu_madvise"); |
1115 | fprintf(stderr, | |
1116 | "Need MADV_DONTFORK in absence of synchronous KVM MMU\n"); | |
6f0437e8 JK |
1117 | exit(1); |
1118 | } | |
6f0437e8 JK |
1119 | } |
1120 | } | |
1121 | ||
e22a25c9 AL |
1122 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
1123 | struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *env, | |
1124 | target_ulong pc) | |
1125 | { | |
1126 | struct kvm_sw_breakpoint *bp; | |
1127 | ||
72cf2d4f | 1128 | QTAILQ_FOREACH(bp, &env->kvm_state->kvm_sw_breakpoints, entry) { |
a426e122 | 1129 | if (bp->pc == pc) { |
e22a25c9 | 1130 | return bp; |
a426e122 | 1131 | } |
e22a25c9 AL |
1132 | } |
1133 | return NULL; | |
1134 | } | |
1135 | ||
1136 | int kvm_sw_breakpoints_active(CPUState *env) | |
1137 | { | |
72cf2d4f | 1138 | return !QTAILQ_EMPTY(&env->kvm_state->kvm_sw_breakpoints); |
e22a25c9 AL |
1139 | } |
1140 | ||
452e4751 GC |
1141 | struct kvm_set_guest_debug_data { |
1142 | struct kvm_guest_debug dbg; | |
1143 | CPUState *env; | |
1144 | int err; | |
1145 | }; | |
1146 | ||
1147 | static void kvm_invoke_set_guest_debug(void *data) | |
1148 | { | |
1149 | struct kvm_set_guest_debug_data *dbg_data = data; | |
b3807725 JK |
1150 | CPUState *env = dbg_data->env; |
1151 | ||
b3807725 | 1152 | dbg_data->err = kvm_vcpu_ioctl(env, KVM_SET_GUEST_DEBUG, &dbg_data->dbg); |
452e4751 GC |
1153 | } |
1154 | ||
e22a25c9 AL |
1155 | int kvm_update_guest_debug(CPUState *env, unsigned long reinject_trap) |
1156 | { | |
452e4751 | 1157 | struct kvm_set_guest_debug_data data; |
e22a25c9 | 1158 | |
b0b1d690 | 1159 | data.dbg.control = reinject_trap; |
e22a25c9 | 1160 | |
b0b1d690 JK |
1161 | if (env->singlestep_enabled) { |
1162 | data.dbg.control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP; | |
1163 | } | |
452e4751 | 1164 | kvm_arch_update_guest_debug(env, &data.dbg); |
452e4751 | 1165 | data.env = env; |
e22a25c9 | 1166 | |
be41cbe0 | 1167 | run_on_cpu(env, kvm_invoke_set_guest_debug, &data); |
452e4751 | 1168 | return data.err; |
e22a25c9 AL |
1169 | } |
1170 | ||
1171 | int kvm_insert_breakpoint(CPUState *current_env, target_ulong addr, | |
1172 | target_ulong len, int type) | |
1173 | { | |
1174 | struct kvm_sw_breakpoint *bp; | |
1175 | CPUState *env; | |
1176 | int err; | |
1177 | ||
1178 | if (type == GDB_BREAKPOINT_SW) { | |
1179 | bp = kvm_find_sw_breakpoint(current_env, addr); | |
1180 | if (bp) { | |
1181 | bp->use_count++; | |
1182 | return 0; | |
1183 | } | |
1184 | ||
1185 | bp = qemu_malloc(sizeof(struct kvm_sw_breakpoint)); | |
a426e122 | 1186 | if (!bp) { |
e22a25c9 | 1187 | return -ENOMEM; |
a426e122 | 1188 | } |
e22a25c9 AL |
1189 | |
1190 | bp->pc = addr; | |
1191 | bp->use_count = 1; | |
1192 | err = kvm_arch_insert_sw_breakpoint(current_env, bp); | |
1193 | if (err) { | |
4a043713 | 1194 | qemu_free(bp); |
e22a25c9 AL |
1195 | return err; |
1196 | } | |
1197 | ||
72cf2d4f | 1198 | QTAILQ_INSERT_HEAD(¤t_env->kvm_state->kvm_sw_breakpoints, |
e22a25c9 AL |
1199 | bp, entry); |
1200 | } else { | |
1201 | err = kvm_arch_insert_hw_breakpoint(addr, len, type); | |
a426e122 | 1202 | if (err) { |
e22a25c9 | 1203 | return err; |
a426e122 | 1204 | } |
e22a25c9 AL |
1205 | } |
1206 | ||
1207 | for (env = first_cpu; env != NULL; env = env->next_cpu) { | |
1208 | err = kvm_update_guest_debug(env, 0); | |
a426e122 | 1209 | if (err) { |
e22a25c9 | 1210 | return err; |
a426e122 | 1211 | } |
e22a25c9 AL |
1212 | } |
1213 | return 0; | |
1214 | } | |
1215 | ||
1216 | int kvm_remove_breakpoint(CPUState *current_env, target_ulong addr, | |
1217 | target_ulong len, int type) | |
1218 | { | |
1219 | struct kvm_sw_breakpoint *bp; | |
1220 | CPUState *env; | |
1221 | int err; | |
1222 | ||
1223 | if (type == GDB_BREAKPOINT_SW) { | |
1224 | bp = kvm_find_sw_breakpoint(current_env, addr); | |
a426e122 | 1225 | if (!bp) { |
e22a25c9 | 1226 | return -ENOENT; |
a426e122 | 1227 | } |
e22a25c9 AL |
1228 | |
1229 | if (bp->use_count > 1) { | |
1230 | bp->use_count--; | |
1231 | return 0; | |
1232 | } | |
1233 | ||
1234 | err = kvm_arch_remove_sw_breakpoint(current_env, bp); | |
a426e122 | 1235 | if (err) { |
e22a25c9 | 1236 | return err; |
a426e122 | 1237 | } |
e22a25c9 | 1238 | |
72cf2d4f | 1239 | QTAILQ_REMOVE(¤t_env->kvm_state->kvm_sw_breakpoints, bp, entry); |
e22a25c9 AL |
1240 | qemu_free(bp); |
1241 | } else { | |
1242 | err = kvm_arch_remove_hw_breakpoint(addr, len, type); | |
a426e122 | 1243 | if (err) { |
e22a25c9 | 1244 | return err; |
a426e122 | 1245 | } |
e22a25c9 AL |
1246 | } |
1247 | ||
1248 | for (env = first_cpu; env != NULL; env = env->next_cpu) { | |
1249 | err = kvm_update_guest_debug(env, 0); | |
a426e122 | 1250 | if (err) { |
e22a25c9 | 1251 | return err; |
a426e122 | 1252 | } |
e22a25c9 AL |
1253 | } |
1254 | return 0; | |
1255 | } | |
1256 | ||
1257 | void kvm_remove_all_breakpoints(CPUState *current_env) | |
1258 | { | |
1259 | struct kvm_sw_breakpoint *bp, *next; | |
1260 | KVMState *s = current_env->kvm_state; | |
1261 | CPUState *env; | |
1262 | ||
72cf2d4f | 1263 | QTAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) { |
e22a25c9 AL |
1264 | if (kvm_arch_remove_sw_breakpoint(current_env, bp) != 0) { |
1265 | /* Try harder to find a CPU that currently sees the breakpoint. */ | |
1266 | for (env = first_cpu; env != NULL; env = env->next_cpu) { | |
a426e122 | 1267 | if (kvm_arch_remove_sw_breakpoint(env, bp) == 0) { |
e22a25c9 | 1268 | break; |
a426e122 | 1269 | } |
e22a25c9 AL |
1270 | } |
1271 | } | |
1272 | } | |
1273 | kvm_arch_remove_all_hw_breakpoints(); | |
1274 | ||
a426e122 | 1275 | for (env = first_cpu; env != NULL; env = env->next_cpu) { |
e22a25c9 | 1276 | kvm_update_guest_debug(env, 0); |
a426e122 | 1277 | } |
e22a25c9 AL |
1278 | } |
1279 | ||
1280 | #else /* !KVM_CAP_SET_GUEST_DEBUG */ | |
1281 | ||
1282 | int kvm_update_guest_debug(CPUState *env, unsigned long reinject_trap) | |
1283 | { | |
1284 | return -EINVAL; | |
1285 | } | |
1286 | ||
1287 | int kvm_insert_breakpoint(CPUState *current_env, target_ulong addr, | |
1288 | target_ulong len, int type) | |
1289 | { | |
1290 | return -EINVAL; | |
1291 | } | |
1292 | ||
1293 | int kvm_remove_breakpoint(CPUState *current_env, target_ulong addr, | |
1294 | target_ulong len, int type) | |
1295 | { | |
1296 | return -EINVAL; | |
1297 | } | |
1298 | ||
1299 | void kvm_remove_all_breakpoints(CPUState *current_env) | |
1300 | { | |
1301 | } | |
1302 | #endif /* !KVM_CAP_SET_GUEST_DEBUG */ | |
cc84de95 MT |
1303 | |
1304 | int kvm_set_signal_mask(CPUState *env, const sigset_t *sigset) | |
1305 | { | |
1306 | struct kvm_signal_mask *sigmask; | |
1307 | int r; | |
1308 | ||
a426e122 | 1309 | if (!sigset) { |
cc84de95 | 1310 | return kvm_vcpu_ioctl(env, KVM_SET_SIGNAL_MASK, NULL); |
a426e122 | 1311 | } |
cc84de95 MT |
1312 | |
1313 | sigmask = qemu_malloc(sizeof(*sigmask) + sizeof(*sigset)); | |
1314 | ||
1315 | sigmask->len = 8; | |
1316 | memcpy(sigmask->sigset, sigset, sizeof(*sigset)); | |
1317 | r = kvm_vcpu_ioctl(env, KVM_SET_SIGNAL_MASK, sigmask); | |
4a043713 | 1318 | qemu_free(sigmask); |
cc84de95 MT |
1319 | |
1320 | return r; | |
1321 | } | |
ca821806 | 1322 | |
44f1a3d8 CM |
1323 | int kvm_set_ioeventfd_mmio_long(int fd, uint32_t addr, uint32_t val, bool assign) |
1324 | { | |
1325 | #ifdef KVM_IOEVENTFD | |
1326 | int ret; | |
1327 | struct kvm_ioeventfd iofd; | |
1328 | ||
1329 | iofd.datamatch = val; | |
1330 | iofd.addr = addr; | |
1331 | iofd.len = 4; | |
1332 | iofd.flags = KVM_IOEVENTFD_FLAG_DATAMATCH; | |
1333 | iofd.fd = fd; | |
1334 | ||
1335 | if (!kvm_enabled()) { | |
1336 | return -ENOSYS; | |
1337 | } | |
1338 | ||
1339 | if (!assign) { | |
1340 | iofd.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN; | |
1341 | } | |
1342 | ||
1343 | ret = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &iofd); | |
1344 | ||
1345 | if (ret < 0) { | |
1346 | return -errno; | |
1347 | } | |
1348 | ||
1349 | return 0; | |
1350 | #else | |
1351 | return -ENOSYS; | |
1352 | #endif | |
1353 | } | |
1354 | ||
ca821806 MT |
1355 | int kvm_set_ioeventfd_pio_word(int fd, uint16_t addr, uint16_t val, bool assign) |
1356 | { | |
98c8573e | 1357 | #ifdef KVM_IOEVENTFD |
ca821806 MT |
1358 | struct kvm_ioeventfd kick = { |
1359 | .datamatch = val, | |
1360 | .addr = addr, | |
1361 | .len = 2, | |
1362 | .flags = KVM_IOEVENTFD_FLAG_DATAMATCH | KVM_IOEVENTFD_FLAG_PIO, | |
1363 | .fd = fd, | |
1364 | }; | |
1365 | int r; | |
a426e122 | 1366 | if (!kvm_enabled()) { |
ca821806 | 1367 | return -ENOSYS; |
a426e122 JK |
1368 | } |
1369 | if (!assign) { | |
ca821806 | 1370 | kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN; |
a426e122 | 1371 | } |
ca821806 | 1372 | r = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick); |
a426e122 | 1373 | if (r < 0) { |
ca821806 | 1374 | return r; |
a426e122 | 1375 | } |
ca821806 | 1376 | return 0; |
98c8573e PB |
1377 | #else |
1378 | return -ENOSYS; | |
ca821806 | 1379 | #endif |
98c8573e | 1380 | } |
a1b87fe0 JK |
1381 | |
1382 | int kvm_on_sigbus_vcpu(CPUState *env, int code, void *addr) | |
1383 | { | |
1384 | return kvm_arch_on_sigbus_vcpu(env, code, addr); | |
1385 | } | |
1386 | ||
1387 | int kvm_on_sigbus(int code, void *addr) | |
1388 | { | |
1389 | return kvm_arch_on_sigbus(code, addr); | |
1390 | } |