]>
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 | */ | |
25254bbc MT |
248 | |
249 | static int kvm_mem_flags(KVMState *s, bool log_dirty) | |
250 | { | |
251 | return log_dirty ? KVM_MEM_LOG_DIRTY_PAGES : 0; | |
252 | } | |
253 | ||
254 | static int kvm_slot_dirty_pages_log_change(KVMSlot *mem, bool log_dirty) | |
5832d1f2 AL |
255 | { |
256 | KVMState *s = kvm_state; | |
25254bbc | 257 | int flags, mask = KVM_MEM_LOG_DIRTY_PAGES; |
4495d6a7 JK |
258 | int old_flags; |
259 | ||
4495d6a7 | 260 | old_flags = mem->flags; |
5832d1f2 | 261 | |
25254bbc | 262 | flags = (mem->flags & ~mask) | kvm_mem_flags(s, log_dirty); |
5832d1f2 AL |
263 | mem->flags = flags; |
264 | ||
4495d6a7 JK |
265 | /* If nothing changed effectively, no need to issue ioctl */ |
266 | if (s->migration_log) { | |
267 | flags |= KVM_MEM_LOG_DIRTY_PAGES; | |
268 | } | |
25254bbc | 269 | |
4495d6a7 | 270 | if (flags == old_flags) { |
25254bbc | 271 | return 0; |
4495d6a7 JK |
272 | } |
273 | ||
5832d1f2 AL |
274 | return kvm_set_user_memory_region(s, mem); |
275 | } | |
276 | ||
25254bbc MT |
277 | static int kvm_dirty_pages_log_change(target_phys_addr_t phys_addr, |
278 | ram_addr_t size, bool log_dirty) | |
279 | { | |
280 | KVMState *s = kvm_state; | |
281 | KVMSlot *mem = kvm_lookup_matching_slot(s, phys_addr, phys_addr + size); | |
282 | ||
283 | if (mem == NULL) { | |
284 | fprintf(stderr, "BUG: %s: invalid parameters " TARGET_FMT_plx "-" | |
285 | TARGET_FMT_plx "\n", __func__, phys_addr, | |
286 | (target_phys_addr_t)(phys_addr + size - 1)); | |
287 | return -EINVAL; | |
288 | } | |
289 | return kvm_slot_dirty_pages_log_change(mem, log_dirty); | |
290 | } | |
291 | ||
e5896b12 AP |
292 | static int kvm_log_start(CPUPhysMemoryClient *client, |
293 | target_phys_addr_t phys_addr, ram_addr_t size) | |
5832d1f2 | 294 | { |
25254bbc | 295 | return kvm_dirty_pages_log_change(phys_addr, size, true); |
5832d1f2 AL |
296 | } |
297 | ||
e5896b12 AP |
298 | static int kvm_log_stop(CPUPhysMemoryClient *client, |
299 | target_phys_addr_t phys_addr, ram_addr_t size) | |
5832d1f2 | 300 | { |
25254bbc | 301 | return kvm_dirty_pages_log_change(phys_addr, size, false); |
5832d1f2 AL |
302 | } |
303 | ||
7b8f3b78 | 304 | static int kvm_set_migration_log(int enable) |
4495d6a7 JK |
305 | { |
306 | KVMState *s = kvm_state; | |
307 | KVMSlot *mem; | |
308 | int i, err; | |
309 | ||
310 | s->migration_log = enable; | |
311 | ||
312 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
313 | mem = &s->slots[i]; | |
314 | ||
70fedd76 AW |
315 | if (!mem->memory_size) { |
316 | continue; | |
317 | } | |
4495d6a7 JK |
318 | if (!!(mem->flags & KVM_MEM_LOG_DIRTY_PAGES) == enable) { |
319 | continue; | |
320 | } | |
321 | err = kvm_set_user_memory_region(s, mem); | |
322 | if (err) { | |
323 | return err; | |
324 | } | |
325 | } | |
326 | return 0; | |
327 | } | |
328 | ||
8369e01c MT |
329 | /* get kvm's dirty pages bitmap and update qemu's */ |
330 | static int kvm_get_dirty_pages_log_range(unsigned long start_addr, | |
331 | unsigned long *bitmap, | |
332 | unsigned long offset, | |
333 | unsigned long mem_size) | |
96c1606b | 334 | { |
8369e01c MT |
335 | unsigned int i, j; |
336 | unsigned long page_number, addr, addr1, c; | |
337 | ram_addr_t ram_addr; | |
338 | unsigned int len = ((mem_size / TARGET_PAGE_SIZE) + HOST_LONG_BITS - 1) / | |
339 | HOST_LONG_BITS; | |
340 | ||
341 | /* | |
342 | * bitmap-traveling is faster than memory-traveling (for addr...) | |
343 | * especially when most of the memory is not dirty. | |
344 | */ | |
345 | for (i = 0; i < len; i++) { | |
346 | if (bitmap[i] != 0) { | |
347 | c = leul_to_cpu(bitmap[i]); | |
348 | do { | |
349 | j = ffsl(c) - 1; | |
350 | c &= ~(1ul << j); | |
351 | page_number = i * HOST_LONG_BITS + j; | |
352 | addr1 = page_number * TARGET_PAGE_SIZE; | |
353 | addr = offset + addr1; | |
354 | ram_addr = cpu_get_physical_page_desc(addr); | |
355 | cpu_physical_memory_set_dirty(ram_addr); | |
356 | } while (c != 0); | |
357 | } | |
358 | } | |
359 | return 0; | |
96c1606b AG |
360 | } |
361 | ||
8369e01c MT |
362 | #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1)) |
363 | ||
5832d1f2 AL |
364 | /** |
365 | * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space | |
366 | * This function updates qemu's dirty bitmap using cpu_physical_memory_set_dirty(). | |
367 | * This means all bits are set to dirty. | |
368 | * | |
d3f8d37f | 369 | * @start_add: start of logged region. |
5832d1f2 AL |
370 | * @end_addr: end of logged region. |
371 | */ | |
7b8f3b78 | 372 | static int kvm_physical_sync_dirty_bitmap(target_phys_addr_t start_addr, |
a426e122 | 373 | target_phys_addr_t end_addr) |
5832d1f2 AL |
374 | { |
375 | KVMState *s = kvm_state; | |
151f7749 | 376 | unsigned long size, allocated_size = 0; |
151f7749 JK |
377 | KVMDirtyLog d; |
378 | KVMSlot *mem; | |
379 | int ret = 0; | |
5832d1f2 | 380 | |
151f7749 JK |
381 | d.dirty_bitmap = NULL; |
382 | while (start_addr < end_addr) { | |
383 | mem = kvm_lookup_overlapping_slot(s, start_addr, end_addr); | |
384 | if (mem == NULL) { | |
385 | break; | |
386 | } | |
5832d1f2 | 387 | |
51b0c606 MT |
388 | /* XXX bad kernel interface alert |
389 | * For dirty bitmap, kernel allocates array of size aligned to | |
390 | * bits-per-long. But for case when the kernel is 64bits and | |
391 | * the userspace is 32bits, userspace can't align to the same | |
392 | * bits-per-long, since sizeof(long) is different between kernel | |
393 | * and user space. This way, userspace will provide buffer which | |
394 | * may be 4 bytes less than the kernel will use, resulting in | |
395 | * userspace memory corruption (which is not detectable by valgrind | |
396 | * too, in most cases). | |
397 | * So for now, let's align to 64 instead of HOST_LONG_BITS here, in | |
398 | * a hope that sizeof(long) wont become >8 any time soon. | |
399 | */ | |
400 | size = ALIGN(((mem->memory_size) >> TARGET_PAGE_BITS), | |
401 | /*HOST_LONG_BITS*/ 64) / 8; | |
151f7749 JK |
402 | if (!d.dirty_bitmap) { |
403 | d.dirty_bitmap = qemu_malloc(size); | |
404 | } else if (size > allocated_size) { | |
405 | d.dirty_bitmap = qemu_realloc(d.dirty_bitmap, size); | |
406 | } | |
407 | allocated_size = size; | |
408 | memset(d.dirty_bitmap, 0, allocated_size); | |
5832d1f2 | 409 | |
151f7749 | 410 | d.slot = mem->slot; |
5832d1f2 | 411 | |
6e489f3f | 412 | if (kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d) == -1) { |
8c0d577e | 413 | DPRINTF("ioctl failed %d\n", errno); |
151f7749 JK |
414 | ret = -1; |
415 | break; | |
416 | } | |
5832d1f2 | 417 | |
8369e01c MT |
418 | kvm_get_dirty_pages_log_range(mem->start_addr, d.dirty_bitmap, |
419 | mem->start_addr, mem->memory_size); | |
420 | start_addr = mem->start_addr + mem->memory_size; | |
5832d1f2 | 421 | } |
5832d1f2 | 422 | qemu_free(d.dirty_bitmap); |
151f7749 JK |
423 | |
424 | return ret; | |
5832d1f2 AL |
425 | } |
426 | ||
c227f099 | 427 | int kvm_coalesce_mmio_region(target_phys_addr_t start, ram_addr_t size) |
f65ed4c1 AL |
428 | { |
429 | int ret = -ENOSYS; | |
f65ed4c1 AL |
430 | KVMState *s = kvm_state; |
431 | ||
432 | if (s->coalesced_mmio) { | |
433 | struct kvm_coalesced_mmio_zone zone; | |
434 | ||
435 | zone.addr = start; | |
436 | zone.size = size; | |
437 | ||
438 | ret = kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone); | |
439 | } | |
f65ed4c1 AL |
440 | |
441 | return ret; | |
442 | } | |
443 | ||
c227f099 | 444 | int kvm_uncoalesce_mmio_region(target_phys_addr_t start, ram_addr_t size) |
f65ed4c1 AL |
445 | { |
446 | int ret = -ENOSYS; | |
f65ed4c1 AL |
447 | KVMState *s = kvm_state; |
448 | ||
449 | if (s->coalesced_mmio) { | |
450 | struct kvm_coalesced_mmio_zone zone; | |
451 | ||
452 | zone.addr = start; | |
453 | zone.size = size; | |
454 | ||
455 | ret = kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone); | |
456 | } | |
f65ed4c1 AL |
457 | |
458 | return ret; | |
459 | } | |
460 | ||
ad7b8b33 AL |
461 | int kvm_check_extension(KVMState *s, unsigned int extension) |
462 | { | |
463 | int ret; | |
464 | ||
465 | ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension); | |
466 | if (ret < 0) { | |
467 | ret = 0; | |
468 | } | |
469 | ||
470 | return ret; | |
471 | } | |
472 | ||
d2f2b8a7 SH |
473 | static int kvm_check_many_ioeventfds(void) |
474 | { | |
d0dcac83 SH |
475 | /* Userspace can use ioeventfd for io notification. This requires a host |
476 | * that supports eventfd(2) and an I/O thread; since eventfd does not | |
477 | * support SIGIO it cannot interrupt the vcpu. | |
478 | * | |
479 | * Older kernels have a 6 device limit on the KVM io bus. Find out so we | |
d2f2b8a7 SH |
480 | * can avoid creating too many ioeventfds. |
481 | */ | |
d0dcac83 | 482 | #if defined(CONFIG_EVENTFD) && defined(CONFIG_IOTHREAD) |
d2f2b8a7 SH |
483 | int ioeventfds[7]; |
484 | int i, ret = 0; | |
485 | for (i = 0; i < ARRAY_SIZE(ioeventfds); i++) { | |
486 | ioeventfds[i] = eventfd(0, EFD_CLOEXEC); | |
487 | if (ioeventfds[i] < 0) { | |
488 | break; | |
489 | } | |
490 | ret = kvm_set_ioeventfd_pio_word(ioeventfds[i], 0, i, true); | |
491 | if (ret < 0) { | |
492 | close(ioeventfds[i]); | |
493 | break; | |
494 | } | |
495 | } | |
496 | ||
497 | /* Decide whether many devices are supported or not */ | |
498 | ret = i == ARRAY_SIZE(ioeventfds); | |
499 | ||
500 | while (i-- > 0) { | |
501 | kvm_set_ioeventfd_pio_word(ioeventfds[i], 0, i, false); | |
502 | close(ioeventfds[i]); | |
503 | } | |
504 | return ret; | |
505 | #else | |
506 | return 0; | |
507 | #endif | |
508 | } | |
509 | ||
94a8d39a JK |
510 | static const KVMCapabilityInfo * |
511 | kvm_check_extension_list(KVMState *s, const KVMCapabilityInfo *list) | |
512 | { | |
513 | while (list->name) { | |
514 | if (!kvm_check_extension(s, list->value)) { | |
515 | return list; | |
516 | } | |
517 | list++; | |
518 | } | |
519 | return NULL; | |
520 | } | |
521 | ||
a426e122 | 522 | static void kvm_set_phys_mem(target_phys_addr_t start_addr, ram_addr_t size, |
25254bbc | 523 | ram_addr_t phys_offset, bool log_dirty) |
46dbef6a MT |
524 | { |
525 | KVMState *s = kvm_state; | |
526 | ram_addr_t flags = phys_offset & ~TARGET_PAGE_MASK; | |
527 | KVMSlot *mem, old; | |
528 | int err; | |
529 | ||
14542fea GN |
530 | /* kvm works in page size chunks, but the function may be called |
531 | with sub-page size and unaligned start address. */ | |
532 | size = TARGET_PAGE_ALIGN(size); | |
533 | start_addr = TARGET_PAGE_ALIGN(start_addr); | |
46dbef6a MT |
534 | |
535 | /* KVM does not support read-only slots */ | |
536 | phys_offset &= ~IO_MEM_ROM; | |
537 | ||
538 | while (1) { | |
539 | mem = kvm_lookup_overlapping_slot(s, start_addr, start_addr + size); | |
540 | if (!mem) { | |
541 | break; | |
542 | } | |
543 | ||
544 | if (flags < IO_MEM_UNASSIGNED && start_addr >= mem->start_addr && | |
545 | (start_addr + size <= mem->start_addr + mem->memory_size) && | |
546 | (phys_offset - start_addr == mem->phys_offset - mem->start_addr)) { | |
547 | /* The new slot fits into the existing one and comes with | |
25254bbc MT |
548 | * identical parameters - update flags and done. */ |
549 | kvm_slot_dirty_pages_log_change(mem, log_dirty); | |
46dbef6a MT |
550 | return; |
551 | } | |
552 | ||
553 | old = *mem; | |
554 | ||
555 | /* unregister the overlapping slot */ | |
556 | mem->memory_size = 0; | |
557 | err = kvm_set_user_memory_region(s, mem); | |
558 | if (err) { | |
559 | fprintf(stderr, "%s: error unregistering overlapping slot: %s\n", | |
560 | __func__, strerror(-err)); | |
561 | abort(); | |
562 | } | |
563 | ||
564 | /* Workaround for older KVM versions: we can't join slots, even not by | |
565 | * unregistering the previous ones and then registering the larger | |
566 | * slot. We have to maintain the existing fragmentation. Sigh. | |
567 | * | |
568 | * This workaround assumes that the new slot starts at the same | |
569 | * address as the first existing one. If not or if some overlapping | |
570 | * slot comes around later, we will fail (not seen in practice so far) | |
571 | * - and actually require a recent KVM version. */ | |
572 | if (s->broken_set_mem_region && | |
573 | old.start_addr == start_addr && old.memory_size < size && | |
574 | flags < IO_MEM_UNASSIGNED) { | |
575 | mem = kvm_alloc_slot(s); | |
576 | mem->memory_size = old.memory_size; | |
577 | mem->start_addr = old.start_addr; | |
578 | mem->phys_offset = old.phys_offset; | |
25254bbc | 579 | mem->flags = kvm_mem_flags(s, log_dirty); |
46dbef6a MT |
580 | |
581 | err = kvm_set_user_memory_region(s, mem); | |
582 | if (err) { | |
583 | fprintf(stderr, "%s: error updating slot: %s\n", __func__, | |
584 | strerror(-err)); | |
585 | abort(); | |
586 | } | |
587 | ||
588 | start_addr += old.memory_size; | |
589 | phys_offset += old.memory_size; | |
590 | size -= old.memory_size; | |
591 | continue; | |
592 | } | |
593 | ||
594 | /* register prefix slot */ | |
595 | if (old.start_addr < start_addr) { | |
596 | mem = kvm_alloc_slot(s); | |
597 | mem->memory_size = start_addr - old.start_addr; | |
598 | mem->start_addr = old.start_addr; | |
599 | mem->phys_offset = old.phys_offset; | |
25254bbc | 600 | mem->flags = kvm_mem_flags(s, log_dirty); |
46dbef6a MT |
601 | |
602 | err = kvm_set_user_memory_region(s, mem); | |
603 | if (err) { | |
604 | fprintf(stderr, "%s: error registering prefix slot: %s\n", | |
605 | __func__, strerror(-err)); | |
606 | abort(); | |
607 | } | |
608 | } | |
609 | ||
610 | /* register suffix slot */ | |
611 | if (old.start_addr + old.memory_size > start_addr + size) { | |
612 | ram_addr_t size_delta; | |
613 | ||
614 | mem = kvm_alloc_slot(s); | |
615 | mem->start_addr = start_addr + size; | |
616 | size_delta = mem->start_addr - old.start_addr; | |
617 | mem->memory_size = old.memory_size - size_delta; | |
618 | mem->phys_offset = old.phys_offset + size_delta; | |
25254bbc | 619 | mem->flags = kvm_mem_flags(s, log_dirty); |
46dbef6a MT |
620 | |
621 | err = kvm_set_user_memory_region(s, mem); | |
622 | if (err) { | |
623 | fprintf(stderr, "%s: error registering suffix slot: %s\n", | |
624 | __func__, strerror(-err)); | |
625 | abort(); | |
626 | } | |
627 | } | |
628 | } | |
629 | ||
630 | /* in case the KVM bug workaround already "consumed" the new slot */ | |
a426e122 | 631 | if (!size) { |
46dbef6a | 632 | return; |
a426e122 | 633 | } |
46dbef6a | 634 | /* KVM does not need to know about this memory */ |
a426e122 | 635 | if (flags >= IO_MEM_UNASSIGNED) { |
46dbef6a | 636 | return; |
a426e122 | 637 | } |
46dbef6a MT |
638 | mem = kvm_alloc_slot(s); |
639 | mem->memory_size = size; | |
640 | mem->start_addr = start_addr; | |
641 | mem->phys_offset = phys_offset; | |
25254bbc | 642 | mem->flags = kvm_mem_flags(s, log_dirty); |
46dbef6a MT |
643 | |
644 | err = kvm_set_user_memory_region(s, mem); | |
645 | if (err) { | |
646 | fprintf(stderr, "%s: error registering slot: %s\n", __func__, | |
647 | strerror(-err)); | |
648 | abort(); | |
649 | } | |
650 | } | |
651 | ||
7b8f3b78 | 652 | static void kvm_client_set_memory(struct CPUPhysMemoryClient *client, |
a426e122 | 653 | target_phys_addr_t start_addr, |
0fd542fb MT |
654 | ram_addr_t size, ram_addr_t phys_offset, |
655 | bool log_dirty) | |
7b8f3b78 | 656 | { |
25254bbc | 657 | kvm_set_phys_mem(start_addr, size, phys_offset, log_dirty); |
7b8f3b78 MT |
658 | } |
659 | ||
660 | static int kvm_client_sync_dirty_bitmap(struct CPUPhysMemoryClient *client, | |
a426e122 JK |
661 | target_phys_addr_t start_addr, |
662 | target_phys_addr_t end_addr) | |
7b8f3b78 | 663 | { |
a426e122 | 664 | return kvm_physical_sync_dirty_bitmap(start_addr, end_addr); |
7b8f3b78 MT |
665 | } |
666 | ||
667 | static int kvm_client_migration_log(struct CPUPhysMemoryClient *client, | |
a426e122 | 668 | int enable) |
7b8f3b78 | 669 | { |
a426e122 | 670 | return kvm_set_migration_log(enable); |
7b8f3b78 MT |
671 | } |
672 | ||
673 | static CPUPhysMemoryClient kvm_cpu_phys_memory_client = { | |
a426e122 JK |
674 | .set_memory = kvm_client_set_memory, |
675 | .sync_dirty_bitmap = kvm_client_sync_dirty_bitmap, | |
676 | .migration_log = kvm_client_migration_log, | |
e5896b12 AP |
677 | .log_start = kvm_log_start, |
678 | .log_stop = kvm_log_stop, | |
7b8f3b78 MT |
679 | }; |
680 | ||
aa7f74d1 JK |
681 | static void kvm_handle_interrupt(CPUState *env, int mask) |
682 | { | |
683 | env->interrupt_request |= mask; | |
684 | ||
685 | if (!qemu_cpu_is_self(env)) { | |
686 | qemu_cpu_kick(env); | |
687 | } | |
688 | } | |
689 | ||
cad1e282 | 690 | int kvm_init(void) |
05330448 | 691 | { |
168ccc11 JK |
692 | static const char upgrade_note[] = |
693 | "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n" | |
694 | "(see http://sourceforge.net/projects/kvm).\n"; | |
05330448 | 695 | KVMState *s; |
94a8d39a | 696 | const KVMCapabilityInfo *missing_cap; |
05330448 AL |
697 | int ret; |
698 | int i; | |
699 | ||
05330448 | 700 | s = qemu_mallocz(sizeof(KVMState)); |
05330448 | 701 | |
e22a25c9 | 702 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
72cf2d4f | 703 | QTAILQ_INIT(&s->kvm_sw_breakpoints); |
e22a25c9 | 704 | #endif |
a426e122 | 705 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { |
05330448 | 706 | s->slots[i].slot = i; |
a426e122 | 707 | } |
05330448 | 708 | s->vmfd = -1; |
40ff6d7e | 709 | s->fd = qemu_open("/dev/kvm", O_RDWR); |
05330448 AL |
710 | if (s->fd == -1) { |
711 | fprintf(stderr, "Could not access KVM kernel module: %m\n"); | |
712 | ret = -errno; | |
713 | goto err; | |
714 | } | |
715 | ||
716 | ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0); | |
717 | if (ret < KVM_API_VERSION) { | |
a426e122 | 718 | if (ret > 0) { |
05330448 | 719 | ret = -EINVAL; |
a426e122 | 720 | } |
05330448 AL |
721 | fprintf(stderr, "kvm version too old\n"); |
722 | goto err; | |
723 | } | |
724 | ||
725 | if (ret > KVM_API_VERSION) { | |
726 | ret = -EINVAL; | |
727 | fprintf(stderr, "kvm version not supported\n"); | |
728 | goto err; | |
729 | } | |
730 | ||
731 | s->vmfd = kvm_ioctl(s, KVM_CREATE_VM, 0); | |
0104dcac AG |
732 | if (s->vmfd < 0) { |
733 | #ifdef TARGET_S390X | |
734 | fprintf(stderr, "Please add the 'switch_amode' kernel parameter to " | |
735 | "your host kernel command line\n"); | |
736 | #endif | |
05330448 | 737 | goto err; |
0104dcac | 738 | } |
05330448 | 739 | |
94a8d39a JK |
740 | missing_cap = kvm_check_extension_list(s, kvm_required_capabilites); |
741 | if (!missing_cap) { | |
742 | missing_cap = | |
743 | kvm_check_extension_list(s, kvm_arch_required_capabilities); | |
05330448 | 744 | } |
94a8d39a | 745 | if (missing_cap) { |
ad7b8b33 | 746 | ret = -EINVAL; |
94a8d39a JK |
747 | fprintf(stderr, "kvm does not support %s\n%s", |
748 | missing_cap->name, upgrade_note); | |
d85dc283 AL |
749 | goto err; |
750 | } | |
751 | ||
ad7b8b33 | 752 | s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO); |
f65ed4c1 | 753 | |
e69917e2 JK |
754 | s->broken_set_mem_region = 1; |
755 | #ifdef KVM_CAP_JOIN_MEMORY_REGIONS_WORKS | |
14a09518 | 756 | ret = kvm_check_extension(s, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS); |
e69917e2 JK |
757 | if (ret > 0) { |
758 | s->broken_set_mem_region = 0; | |
759 | } | |
760 | #endif | |
761 | ||
a0fb002c JK |
762 | s->vcpu_events = 0; |
763 | #ifdef KVM_CAP_VCPU_EVENTS | |
764 | s->vcpu_events = kvm_check_extension(s, KVM_CAP_VCPU_EVENTS); | |
765 | #endif | |
766 | ||
b0b1d690 JK |
767 | s->robust_singlestep = 0; |
768 | #ifdef KVM_CAP_X86_ROBUST_SINGLESTEP | |
769 | s->robust_singlestep = | |
770 | kvm_check_extension(s, KVM_CAP_X86_ROBUST_SINGLESTEP); | |
771 | #endif | |
772 | ||
ff44f1a3 JK |
773 | s->debugregs = 0; |
774 | #ifdef KVM_CAP_DEBUGREGS | |
775 | s->debugregs = kvm_check_extension(s, KVM_CAP_DEBUGREGS); | |
776 | #endif | |
777 | ||
f1665b21 SY |
778 | s->xsave = 0; |
779 | #ifdef KVM_CAP_XSAVE | |
780 | s->xsave = kvm_check_extension(s, KVM_CAP_XSAVE); | |
781 | #endif | |
782 | ||
783 | s->xcrs = 0; | |
784 | #ifdef KVM_CAP_XCRS | |
785 | s->xcrs = kvm_check_extension(s, KVM_CAP_XCRS); | |
786 | #endif | |
787 | ||
cad1e282 | 788 | ret = kvm_arch_init(s); |
a426e122 | 789 | if (ret < 0) { |
05330448 | 790 | goto err; |
a426e122 | 791 | } |
05330448 AL |
792 | |
793 | kvm_state = s; | |
7b8f3b78 | 794 | cpu_register_phys_memory_client(&kvm_cpu_phys_memory_client); |
05330448 | 795 | |
d2f2b8a7 SH |
796 | s->many_ioeventfds = kvm_check_many_ioeventfds(); |
797 | ||
aa7f74d1 JK |
798 | cpu_interrupt_handler = kvm_handle_interrupt; |
799 | ||
05330448 AL |
800 | return 0; |
801 | ||
802 | err: | |
803 | if (s) { | |
a426e122 | 804 | if (s->vmfd != -1) { |
05330448 | 805 | close(s->vmfd); |
a426e122 JK |
806 | } |
807 | if (s->fd != -1) { | |
05330448 | 808 | close(s->fd); |
a426e122 | 809 | } |
05330448 AL |
810 | } |
811 | qemu_free(s); | |
812 | ||
813 | return ret; | |
814 | } | |
815 | ||
b30e93e9 JK |
816 | static void kvm_handle_io(uint16_t port, void *data, int direction, int size, |
817 | uint32_t count) | |
05330448 AL |
818 | { |
819 | int i; | |
820 | uint8_t *ptr = data; | |
821 | ||
822 | for (i = 0; i < count; i++) { | |
823 | if (direction == KVM_EXIT_IO_IN) { | |
824 | switch (size) { | |
825 | case 1: | |
afcea8cb | 826 | stb_p(ptr, cpu_inb(port)); |
05330448 AL |
827 | break; |
828 | case 2: | |
afcea8cb | 829 | stw_p(ptr, cpu_inw(port)); |
05330448 AL |
830 | break; |
831 | case 4: | |
afcea8cb | 832 | stl_p(ptr, cpu_inl(port)); |
05330448 AL |
833 | break; |
834 | } | |
835 | } else { | |
836 | switch (size) { | |
837 | case 1: | |
afcea8cb | 838 | cpu_outb(port, ldub_p(ptr)); |
05330448 AL |
839 | break; |
840 | case 2: | |
afcea8cb | 841 | cpu_outw(port, lduw_p(ptr)); |
05330448 AL |
842 | break; |
843 | case 4: | |
afcea8cb | 844 | cpu_outl(port, ldl_p(ptr)); |
05330448 AL |
845 | break; |
846 | } | |
847 | } | |
848 | ||
849 | ptr += size; | |
850 | } | |
05330448 AL |
851 | } |
852 | ||
7c80eef8 | 853 | #ifdef KVM_CAP_INTERNAL_ERROR_DATA |
73aaec4a | 854 | static int kvm_handle_internal_error(CPUState *env, struct kvm_run *run) |
7c80eef8 | 855 | { |
bb44e0d1 | 856 | fprintf(stderr, "KVM internal error."); |
7c80eef8 MT |
857 | if (kvm_check_extension(kvm_state, KVM_CAP_INTERNAL_ERROR_DATA)) { |
858 | int i; | |
859 | ||
bb44e0d1 | 860 | fprintf(stderr, " Suberror: %d\n", run->internal.suberror); |
7c80eef8 MT |
861 | for (i = 0; i < run->internal.ndata; ++i) { |
862 | fprintf(stderr, "extra data[%d]: %"PRIx64"\n", | |
863 | i, (uint64_t)run->internal.data[i]); | |
864 | } | |
bb44e0d1 JK |
865 | } else { |
866 | fprintf(stderr, "\n"); | |
7c80eef8 | 867 | } |
7c80eef8 MT |
868 | if (run->internal.suberror == KVM_INTERNAL_ERROR_EMULATION) { |
869 | fprintf(stderr, "emulation failure\n"); | |
a426e122 | 870 | if (!kvm_arch_stop_on_emulation_error(env)) { |
f5c848ee | 871 | cpu_dump_state(env, stderr, fprintf, CPU_DUMP_CODE); |
d73cd8f4 | 872 | return EXCP_INTERRUPT; |
a426e122 | 873 | } |
7c80eef8 MT |
874 | } |
875 | /* FIXME: Should trigger a qmp message to let management know | |
876 | * something went wrong. | |
877 | */ | |
73aaec4a | 878 | return -1; |
7c80eef8 MT |
879 | } |
880 | #endif | |
881 | ||
62a2744c | 882 | void kvm_flush_coalesced_mmio_buffer(void) |
f65ed4c1 | 883 | { |
f65ed4c1 | 884 | KVMState *s = kvm_state; |
62a2744c SY |
885 | if (s->coalesced_mmio_ring) { |
886 | struct kvm_coalesced_mmio_ring *ring = s->coalesced_mmio_ring; | |
f65ed4c1 AL |
887 | while (ring->first != ring->last) { |
888 | struct kvm_coalesced_mmio *ent; | |
889 | ||
890 | ent = &ring->coalesced_mmio[ring->first]; | |
891 | ||
892 | cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len); | |
85199474 | 893 | smp_wmb(); |
f65ed4c1 AL |
894 | ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX; |
895 | } | |
896 | } | |
f65ed4c1 AL |
897 | } |
898 | ||
2705d56a | 899 | static void do_kvm_cpu_synchronize_state(void *_env) |
4c0960c0 | 900 | { |
2705d56a JK |
901 | CPUState *env = _env; |
902 | ||
9ded2744 | 903 | if (!env->kvm_vcpu_dirty) { |
4c0960c0 | 904 | kvm_arch_get_registers(env); |
9ded2744 | 905 | env->kvm_vcpu_dirty = 1; |
4c0960c0 AK |
906 | } |
907 | } | |
908 | ||
2705d56a JK |
909 | void kvm_cpu_synchronize_state(CPUState *env) |
910 | { | |
a426e122 | 911 | if (!env->kvm_vcpu_dirty) { |
2705d56a | 912 | run_on_cpu(env, do_kvm_cpu_synchronize_state, env); |
a426e122 | 913 | } |
2705d56a JK |
914 | } |
915 | ||
ea375f9a JK |
916 | void kvm_cpu_synchronize_post_reset(CPUState *env) |
917 | { | |
918 | kvm_arch_put_registers(env, KVM_PUT_RESET_STATE); | |
919 | env->kvm_vcpu_dirty = 0; | |
920 | } | |
921 | ||
922 | void kvm_cpu_synchronize_post_init(CPUState *env) | |
923 | { | |
924 | kvm_arch_put_registers(env, KVM_PUT_FULL_STATE); | |
925 | env->kvm_vcpu_dirty = 0; | |
926 | } | |
927 | ||
05330448 AL |
928 | int kvm_cpu_exec(CPUState *env) |
929 | { | |
930 | struct kvm_run *run = env->kvm_run; | |
7cbb533f | 931 | int ret, run_ret; |
05330448 | 932 | |
8c0d577e | 933 | DPRINTF("kvm_cpu_exec()\n"); |
05330448 | 934 | |
99036865 | 935 | if (kvm_arch_process_async_events(env)) { |
9ccfac9e | 936 | env->exit_request = 0; |
6792a57b | 937 | return EXCP_HLT; |
9ccfac9e | 938 | } |
0af691d7 | 939 | |
6792a57b JK |
940 | cpu_single_env = env; |
941 | ||
9ccfac9e | 942 | do { |
9ded2744 | 943 | if (env->kvm_vcpu_dirty) { |
ea375f9a | 944 | kvm_arch_put_registers(env, KVM_PUT_RUNTIME_STATE); |
9ded2744 | 945 | env->kvm_vcpu_dirty = 0; |
4c0960c0 AK |
946 | } |
947 | ||
8c14c173 | 948 | kvm_arch_pre_run(env, run); |
9ccfac9e JK |
949 | if (env->exit_request) { |
950 | DPRINTF("interrupt exit requested\n"); | |
951 | /* | |
952 | * KVM requires us to reenter the kernel after IO exits to complete | |
953 | * instruction emulation. This self-signal will ensure that we | |
954 | * leave ASAP again. | |
955 | */ | |
956 | qemu_cpu_kick_self(); | |
957 | } | |
273faf1b | 958 | cpu_single_env = NULL; |
d549db5a | 959 | qemu_mutex_unlock_iothread(); |
9ccfac9e | 960 | |
7cbb533f | 961 | run_ret = kvm_vcpu_ioctl(env, KVM_RUN, 0); |
9ccfac9e | 962 | |
d549db5a | 963 | qemu_mutex_lock_iothread(); |
273faf1b | 964 | cpu_single_env = env; |
05330448 AL |
965 | kvm_arch_post_run(env, run); |
966 | ||
b0c883b5 JK |
967 | kvm_flush_coalesced_mmio_buffer(); |
968 | ||
7cbb533f | 969 | if (run_ret < 0) { |
dc77d341 JK |
970 | if (run_ret == -EINTR || run_ret == -EAGAIN) { |
971 | DPRINTF("io window exit\n"); | |
d73cd8f4 | 972 | ret = EXCP_INTERRUPT; |
dc77d341 JK |
973 | break; |
974 | } | |
7cbb533f | 975 | DPRINTF("kvm run failed %s\n", strerror(-run_ret)); |
05330448 AL |
976 | abort(); |
977 | } | |
978 | ||
05330448 AL |
979 | switch (run->exit_reason) { |
980 | case KVM_EXIT_IO: | |
8c0d577e | 981 | DPRINTF("handle_io\n"); |
b30e93e9 JK |
982 | kvm_handle_io(run->io.port, |
983 | (uint8_t *)run + run->io.data_offset, | |
984 | run->io.direction, | |
985 | run->io.size, | |
986 | run->io.count); | |
d73cd8f4 | 987 | ret = 0; |
05330448 AL |
988 | break; |
989 | case KVM_EXIT_MMIO: | |
8c0d577e | 990 | DPRINTF("handle_mmio\n"); |
05330448 AL |
991 | cpu_physical_memory_rw(run->mmio.phys_addr, |
992 | run->mmio.data, | |
993 | run->mmio.len, | |
994 | run->mmio.is_write); | |
d73cd8f4 | 995 | ret = 0; |
05330448 AL |
996 | break; |
997 | case KVM_EXIT_IRQ_WINDOW_OPEN: | |
8c0d577e | 998 | DPRINTF("irq_window_open\n"); |
d73cd8f4 | 999 | ret = EXCP_INTERRUPT; |
05330448 AL |
1000 | break; |
1001 | case KVM_EXIT_SHUTDOWN: | |
8c0d577e | 1002 | DPRINTF("shutdown\n"); |
05330448 | 1003 | qemu_system_reset_request(); |
d73cd8f4 | 1004 | ret = EXCP_INTERRUPT; |
05330448 AL |
1005 | break; |
1006 | case KVM_EXIT_UNKNOWN: | |
bb44e0d1 JK |
1007 | fprintf(stderr, "KVM: unknown exit, hardware reason %" PRIx64 "\n", |
1008 | (uint64_t)run->hw.hardware_exit_reason); | |
73aaec4a | 1009 | ret = -1; |
05330448 | 1010 | break; |
7c80eef8 MT |
1011 | #ifdef KVM_CAP_INTERNAL_ERROR_DATA |
1012 | case KVM_EXIT_INTERNAL_ERROR: | |
73aaec4a | 1013 | ret = kvm_handle_internal_error(env, run); |
7c80eef8 MT |
1014 | break; |
1015 | #endif | |
05330448 | 1016 | default: |
8c0d577e | 1017 | DPRINTF("kvm_arch_handle_exit\n"); |
05330448 AL |
1018 | ret = kvm_arch_handle_exit(env, run); |
1019 | break; | |
1020 | } | |
d73cd8f4 | 1021 | } while (ret == 0); |
05330448 | 1022 | |
73aaec4a | 1023 | if (ret < 0) { |
f5c848ee | 1024 | cpu_dump_state(env, stderr, fprintf, CPU_DUMP_CODE); |
e07bbac5 | 1025 | vm_stop(VMSTOP_PANIC); |
becfc390 AL |
1026 | } |
1027 | ||
6792a57b JK |
1028 | env->exit_request = 0; |
1029 | cpu_single_env = NULL; | |
05330448 AL |
1030 | return ret; |
1031 | } | |
1032 | ||
984b5181 | 1033 | int kvm_ioctl(KVMState *s, int type, ...) |
05330448 AL |
1034 | { |
1035 | int ret; | |
984b5181 AL |
1036 | void *arg; |
1037 | va_list ap; | |
05330448 | 1038 | |
984b5181 AL |
1039 | va_start(ap, type); |
1040 | arg = va_arg(ap, void *); | |
1041 | va_end(ap); | |
1042 | ||
1043 | ret = ioctl(s->fd, type, arg); | |
a426e122 | 1044 | if (ret == -1) { |
05330448 | 1045 | ret = -errno; |
a426e122 | 1046 | } |
05330448 AL |
1047 | return ret; |
1048 | } | |
1049 | ||
984b5181 | 1050 | int kvm_vm_ioctl(KVMState *s, int type, ...) |
05330448 AL |
1051 | { |
1052 | int ret; | |
984b5181 AL |
1053 | void *arg; |
1054 | va_list ap; | |
1055 | ||
1056 | va_start(ap, type); | |
1057 | arg = va_arg(ap, void *); | |
1058 | va_end(ap); | |
05330448 | 1059 | |
984b5181 | 1060 | ret = ioctl(s->vmfd, type, arg); |
a426e122 | 1061 | if (ret == -1) { |
05330448 | 1062 | ret = -errno; |
a426e122 | 1063 | } |
05330448 AL |
1064 | return ret; |
1065 | } | |
1066 | ||
984b5181 | 1067 | int kvm_vcpu_ioctl(CPUState *env, int type, ...) |
05330448 AL |
1068 | { |
1069 | int ret; | |
984b5181 AL |
1070 | void *arg; |
1071 | va_list ap; | |
1072 | ||
1073 | va_start(ap, type); | |
1074 | arg = va_arg(ap, void *); | |
1075 | va_end(ap); | |
05330448 | 1076 | |
984b5181 | 1077 | ret = ioctl(env->kvm_fd, type, arg); |
a426e122 | 1078 | if (ret == -1) { |
05330448 | 1079 | ret = -errno; |
a426e122 | 1080 | } |
05330448 AL |
1081 | return ret; |
1082 | } | |
bd322087 AL |
1083 | |
1084 | int kvm_has_sync_mmu(void) | |
1085 | { | |
94a8d39a | 1086 | return kvm_check_extension(kvm_state, KVM_CAP_SYNC_MMU); |
bd322087 | 1087 | } |
e22a25c9 | 1088 | |
a0fb002c JK |
1089 | int kvm_has_vcpu_events(void) |
1090 | { | |
1091 | return kvm_state->vcpu_events; | |
1092 | } | |
1093 | ||
b0b1d690 JK |
1094 | int kvm_has_robust_singlestep(void) |
1095 | { | |
1096 | return kvm_state->robust_singlestep; | |
1097 | } | |
1098 | ||
ff44f1a3 JK |
1099 | int kvm_has_debugregs(void) |
1100 | { | |
1101 | return kvm_state->debugregs; | |
1102 | } | |
1103 | ||
f1665b21 SY |
1104 | int kvm_has_xsave(void) |
1105 | { | |
1106 | return kvm_state->xsave; | |
1107 | } | |
1108 | ||
1109 | int kvm_has_xcrs(void) | |
1110 | { | |
1111 | return kvm_state->xcrs; | |
1112 | } | |
1113 | ||
d2f2b8a7 SH |
1114 | int kvm_has_many_ioeventfds(void) |
1115 | { | |
1116 | if (!kvm_enabled()) { | |
1117 | return 0; | |
1118 | } | |
1119 | return kvm_state->many_ioeventfds; | |
1120 | } | |
1121 | ||
6f0437e8 JK |
1122 | void kvm_setup_guest_memory(void *start, size_t size) |
1123 | { | |
1124 | if (!kvm_has_sync_mmu()) { | |
e78815a5 | 1125 | int ret = qemu_madvise(start, size, QEMU_MADV_DONTFORK); |
6f0437e8 JK |
1126 | |
1127 | if (ret) { | |
e78815a5 AF |
1128 | perror("qemu_madvise"); |
1129 | fprintf(stderr, | |
1130 | "Need MADV_DONTFORK in absence of synchronous KVM MMU\n"); | |
6f0437e8 JK |
1131 | exit(1); |
1132 | } | |
6f0437e8 JK |
1133 | } |
1134 | } | |
1135 | ||
e22a25c9 AL |
1136 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
1137 | struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *env, | |
1138 | target_ulong pc) | |
1139 | { | |
1140 | struct kvm_sw_breakpoint *bp; | |
1141 | ||
72cf2d4f | 1142 | QTAILQ_FOREACH(bp, &env->kvm_state->kvm_sw_breakpoints, entry) { |
a426e122 | 1143 | if (bp->pc == pc) { |
e22a25c9 | 1144 | return bp; |
a426e122 | 1145 | } |
e22a25c9 AL |
1146 | } |
1147 | return NULL; | |
1148 | } | |
1149 | ||
1150 | int kvm_sw_breakpoints_active(CPUState *env) | |
1151 | { | |
72cf2d4f | 1152 | return !QTAILQ_EMPTY(&env->kvm_state->kvm_sw_breakpoints); |
e22a25c9 AL |
1153 | } |
1154 | ||
452e4751 GC |
1155 | struct kvm_set_guest_debug_data { |
1156 | struct kvm_guest_debug dbg; | |
1157 | CPUState *env; | |
1158 | int err; | |
1159 | }; | |
1160 | ||
1161 | static void kvm_invoke_set_guest_debug(void *data) | |
1162 | { | |
1163 | struct kvm_set_guest_debug_data *dbg_data = data; | |
b3807725 JK |
1164 | CPUState *env = dbg_data->env; |
1165 | ||
b3807725 | 1166 | dbg_data->err = kvm_vcpu_ioctl(env, KVM_SET_GUEST_DEBUG, &dbg_data->dbg); |
452e4751 GC |
1167 | } |
1168 | ||
e22a25c9 AL |
1169 | int kvm_update_guest_debug(CPUState *env, unsigned long reinject_trap) |
1170 | { | |
452e4751 | 1171 | struct kvm_set_guest_debug_data data; |
e22a25c9 | 1172 | |
b0b1d690 | 1173 | data.dbg.control = reinject_trap; |
e22a25c9 | 1174 | |
b0b1d690 JK |
1175 | if (env->singlestep_enabled) { |
1176 | data.dbg.control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP; | |
1177 | } | |
452e4751 | 1178 | kvm_arch_update_guest_debug(env, &data.dbg); |
452e4751 | 1179 | data.env = env; |
e22a25c9 | 1180 | |
be41cbe0 | 1181 | run_on_cpu(env, kvm_invoke_set_guest_debug, &data); |
452e4751 | 1182 | return data.err; |
e22a25c9 AL |
1183 | } |
1184 | ||
1185 | int kvm_insert_breakpoint(CPUState *current_env, target_ulong addr, | |
1186 | target_ulong len, int type) | |
1187 | { | |
1188 | struct kvm_sw_breakpoint *bp; | |
1189 | CPUState *env; | |
1190 | int err; | |
1191 | ||
1192 | if (type == GDB_BREAKPOINT_SW) { | |
1193 | bp = kvm_find_sw_breakpoint(current_env, addr); | |
1194 | if (bp) { | |
1195 | bp->use_count++; | |
1196 | return 0; | |
1197 | } | |
1198 | ||
1199 | bp = qemu_malloc(sizeof(struct kvm_sw_breakpoint)); | |
a426e122 | 1200 | if (!bp) { |
e22a25c9 | 1201 | return -ENOMEM; |
a426e122 | 1202 | } |
e22a25c9 AL |
1203 | |
1204 | bp->pc = addr; | |
1205 | bp->use_count = 1; | |
1206 | err = kvm_arch_insert_sw_breakpoint(current_env, bp); | |
1207 | if (err) { | |
4a043713 | 1208 | qemu_free(bp); |
e22a25c9 AL |
1209 | return err; |
1210 | } | |
1211 | ||
72cf2d4f | 1212 | QTAILQ_INSERT_HEAD(¤t_env->kvm_state->kvm_sw_breakpoints, |
e22a25c9 AL |
1213 | bp, entry); |
1214 | } else { | |
1215 | err = kvm_arch_insert_hw_breakpoint(addr, len, type); | |
a426e122 | 1216 | if (err) { |
e22a25c9 | 1217 | return err; |
a426e122 | 1218 | } |
e22a25c9 AL |
1219 | } |
1220 | ||
1221 | for (env = first_cpu; env != NULL; env = env->next_cpu) { | |
1222 | err = kvm_update_guest_debug(env, 0); | |
a426e122 | 1223 | if (err) { |
e22a25c9 | 1224 | return err; |
a426e122 | 1225 | } |
e22a25c9 AL |
1226 | } |
1227 | return 0; | |
1228 | } | |
1229 | ||
1230 | int kvm_remove_breakpoint(CPUState *current_env, target_ulong addr, | |
1231 | target_ulong len, int type) | |
1232 | { | |
1233 | struct kvm_sw_breakpoint *bp; | |
1234 | CPUState *env; | |
1235 | int err; | |
1236 | ||
1237 | if (type == GDB_BREAKPOINT_SW) { | |
1238 | bp = kvm_find_sw_breakpoint(current_env, addr); | |
a426e122 | 1239 | if (!bp) { |
e22a25c9 | 1240 | return -ENOENT; |
a426e122 | 1241 | } |
e22a25c9 AL |
1242 | |
1243 | if (bp->use_count > 1) { | |
1244 | bp->use_count--; | |
1245 | return 0; | |
1246 | } | |
1247 | ||
1248 | err = kvm_arch_remove_sw_breakpoint(current_env, bp); | |
a426e122 | 1249 | if (err) { |
e22a25c9 | 1250 | return err; |
a426e122 | 1251 | } |
e22a25c9 | 1252 | |
72cf2d4f | 1253 | QTAILQ_REMOVE(¤t_env->kvm_state->kvm_sw_breakpoints, bp, entry); |
e22a25c9 AL |
1254 | qemu_free(bp); |
1255 | } else { | |
1256 | err = kvm_arch_remove_hw_breakpoint(addr, len, type); | |
a426e122 | 1257 | if (err) { |
e22a25c9 | 1258 | return err; |
a426e122 | 1259 | } |
e22a25c9 AL |
1260 | } |
1261 | ||
1262 | for (env = first_cpu; env != NULL; env = env->next_cpu) { | |
1263 | err = kvm_update_guest_debug(env, 0); | |
a426e122 | 1264 | if (err) { |
e22a25c9 | 1265 | return err; |
a426e122 | 1266 | } |
e22a25c9 AL |
1267 | } |
1268 | return 0; | |
1269 | } | |
1270 | ||
1271 | void kvm_remove_all_breakpoints(CPUState *current_env) | |
1272 | { | |
1273 | struct kvm_sw_breakpoint *bp, *next; | |
1274 | KVMState *s = current_env->kvm_state; | |
1275 | CPUState *env; | |
1276 | ||
72cf2d4f | 1277 | QTAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) { |
e22a25c9 AL |
1278 | if (kvm_arch_remove_sw_breakpoint(current_env, bp) != 0) { |
1279 | /* Try harder to find a CPU that currently sees the breakpoint. */ | |
1280 | for (env = first_cpu; env != NULL; env = env->next_cpu) { | |
a426e122 | 1281 | if (kvm_arch_remove_sw_breakpoint(env, bp) == 0) { |
e22a25c9 | 1282 | break; |
a426e122 | 1283 | } |
e22a25c9 AL |
1284 | } |
1285 | } | |
1286 | } | |
1287 | kvm_arch_remove_all_hw_breakpoints(); | |
1288 | ||
a426e122 | 1289 | for (env = first_cpu; env != NULL; env = env->next_cpu) { |
e22a25c9 | 1290 | kvm_update_guest_debug(env, 0); |
a426e122 | 1291 | } |
e22a25c9 AL |
1292 | } |
1293 | ||
1294 | #else /* !KVM_CAP_SET_GUEST_DEBUG */ | |
1295 | ||
1296 | int kvm_update_guest_debug(CPUState *env, unsigned long reinject_trap) | |
1297 | { | |
1298 | return -EINVAL; | |
1299 | } | |
1300 | ||
1301 | int kvm_insert_breakpoint(CPUState *current_env, target_ulong addr, | |
1302 | target_ulong len, int type) | |
1303 | { | |
1304 | return -EINVAL; | |
1305 | } | |
1306 | ||
1307 | int kvm_remove_breakpoint(CPUState *current_env, target_ulong addr, | |
1308 | target_ulong len, int type) | |
1309 | { | |
1310 | return -EINVAL; | |
1311 | } | |
1312 | ||
1313 | void kvm_remove_all_breakpoints(CPUState *current_env) | |
1314 | { | |
1315 | } | |
1316 | #endif /* !KVM_CAP_SET_GUEST_DEBUG */ | |
cc84de95 MT |
1317 | |
1318 | int kvm_set_signal_mask(CPUState *env, const sigset_t *sigset) | |
1319 | { | |
1320 | struct kvm_signal_mask *sigmask; | |
1321 | int r; | |
1322 | ||
a426e122 | 1323 | if (!sigset) { |
cc84de95 | 1324 | return kvm_vcpu_ioctl(env, KVM_SET_SIGNAL_MASK, NULL); |
a426e122 | 1325 | } |
cc84de95 MT |
1326 | |
1327 | sigmask = qemu_malloc(sizeof(*sigmask) + sizeof(*sigset)); | |
1328 | ||
1329 | sigmask->len = 8; | |
1330 | memcpy(sigmask->sigset, sigset, sizeof(*sigset)); | |
1331 | r = kvm_vcpu_ioctl(env, KVM_SET_SIGNAL_MASK, sigmask); | |
4a043713 | 1332 | qemu_free(sigmask); |
cc84de95 MT |
1333 | |
1334 | return r; | |
1335 | } | |
ca821806 | 1336 | |
44f1a3d8 CM |
1337 | int kvm_set_ioeventfd_mmio_long(int fd, uint32_t addr, uint32_t val, bool assign) |
1338 | { | |
1339 | #ifdef KVM_IOEVENTFD | |
1340 | int ret; | |
1341 | struct kvm_ioeventfd iofd; | |
1342 | ||
1343 | iofd.datamatch = val; | |
1344 | iofd.addr = addr; | |
1345 | iofd.len = 4; | |
1346 | iofd.flags = KVM_IOEVENTFD_FLAG_DATAMATCH; | |
1347 | iofd.fd = fd; | |
1348 | ||
1349 | if (!kvm_enabled()) { | |
1350 | return -ENOSYS; | |
1351 | } | |
1352 | ||
1353 | if (!assign) { | |
1354 | iofd.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN; | |
1355 | } | |
1356 | ||
1357 | ret = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &iofd); | |
1358 | ||
1359 | if (ret < 0) { | |
1360 | return -errno; | |
1361 | } | |
1362 | ||
1363 | return 0; | |
1364 | #else | |
1365 | return -ENOSYS; | |
1366 | #endif | |
1367 | } | |
1368 | ||
ca821806 MT |
1369 | int kvm_set_ioeventfd_pio_word(int fd, uint16_t addr, uint16_t val, bool assign) |
1370 | { | |
98c8573e | 1371 | #ifdef KVM_IOEVENTFD |
ca821806 MT |
1372 | struct kvm_ioeventfd kick = { |
1373 | .datamatch = val, | |
1374 | .addr = addr, | |
1375 | .len = 2, | |
1376 | .flags = KVM_IOEVENTFD_FLAG_DATAMATCH | KVM_IOEVENTFD_FLAG_PIO, | |
1377 | .fd = fd, | |
1378 | }; | |
1379 | int r; | |
a426e122 | 1380 | if (!kvm_enabled()) { |
ca821806 | 1381 | return -ENOSYS; |
a426e122 JK |
1382 | } |
1383 | if (!assign) { | |
ca821806 | 1384 | kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN; |
a426e122 | 1385 | } |
ca821806 | 1386 | r = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick); |
a426e122 | 1387 | if (r < 0) { |
ca821806 | 1388 | return r; |
a426e122 | 1389 | } |
ca821806 | 1390 | return 0; |
98c8573e PB |
1391 | #else |
1392 | return -ENOSYS; | |
ca821806 | 1393 | #endif |
98c8573e | 1394 | } |
a1b87fe0 JK |
1395 | |
1396 | int kvm_on_sigbus_vcpu(CPUState *env, int code, void *addr) | |
1397 | { | |
1398 | return kvm_arch_on_sigbus_vcpu(env, code, addr); | |
1399 | } | |
1400 | ||
1401 | int kvm_on_sigbus(int code, void *addr) | |
1402 | { | |
1403 | return kvm_arch_on_sigbus(code, addr); | |
1404 | } |