]>
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; |
1cae88b9 | 67 | bool coalesced_flush_in_progress; |
e69917e2 | 68 | int broken_set_mem_region; |
4495d6a7 | 69 | int migration_log; |
a0fb002c | 70 | int vcpu_events; |
b0b1d690 | 71 | int robust_singlestep; |
ff44f1a3 | 72 | int debugregs; |
e22a25c9 AL |
73 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
74 | struct kvm_sw_breakpoint_head kvm_sw_breakpoints; | |
75 | #endif | |
6f725c13 GC |
76 | int irqchip_in_kernel; |
77 | int pit_in_kernel; | |
f1665b21 | 78 | int xsave, xcrs; |
d2f2b8a7 | 79 | int many_ioeventfds; |
05330448 AL |
80 | }; |
81 | ||
6a7af8cb | 82 | KVMState *kvm_state; |
05330448 | 83 | |
94a8d39a JK |
84 | static const KVMCapabilityInfo kvm_required_capabilites[] = { |
85 | KVM_CAP_INFO(USER_MEMORY), | |
86 | KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS), | |
87 | KVM_CAP_LAST_INFO | |
88 | }; | |
89 | ||
05330448 AL |
90 | static KVMSlot *kvm_alloc_slot(KVMState *s) |
91 | { | |
92 | int i; | |
93 | ||
94 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
a426e122 | 95 | if (s->slots[i].memory_size == 0) { |
05330448 | 96 | return &s->slots[i]; |
a426e122 | 97 | } |
05330448 AL |
98 | } |
99 | ||
d3f8d37f AL |
100 | fprintf(stderr, "%s: no free slot available\n", __func__); |
101 | abort(); | |
102 | } | |
103 | ||
104 | static KVMSlot *kvm_lookup_matching_slot(KVMState *s, | |
c227f099 AL |
105 | target_phys_addr_t start_addr, |
106 | target_phys_addr_t end_addr) | |
d3f8d37f AL |
107 | { |
108 | int i; | |
109 | ||
110 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
111 | KVMSlot *mem = &s->slots[i]; | |
112 | ||
113 | if (start_addr == mem->start_addr && | |
114 | end_addr == mem->start_addr + mem->memory_size) { | |
115 | return mem; | |
116 | } | |
117 | } | |
118 | ||
05330448 AL |
119 | return NULL; |
120 | } | |
121 | ||
6152e2ae AL |
122 | /* |
123 | * Find overlapping slot with lowest start address | |
124 | */ | |
125 | static KVMSlot *kvm_lookup_overlapping_slot(KVMState *s, | |
c227f099 AL |
126 | target_phys_addr_t start_addr, |
127 | target_phys_addr_t end_addr) | |
05330448 | 128 | { |
6152e2ae | 129 | KVMSlot *found = NULL; |
05330448 AL |
130 | int i; |
131 | ||
132 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
133 | KVMSlot *mem = &s->slots[i]; | |
134 | ||
6152e2ae AL |
135 | if (mem->memory_size == 0 || |
136 | (found && found->start_addr < mem->start_addr)) { | |
137 | continue; | |
138 | } | |
139 | ||
140 | if (end_addr > mem->start_addr && | |
141 | start_addr < mem->start_addr + mem->memory_size) { | |
142 | found = mem; | |
143 | } | |
05330448 AL |
144 | } |
145 | ||
6152e2ae | 146 | return found; |
05330448 AL |
147 | } |
148 | ||
983dfc3b HY |
149 | int kvm_physical_memory_addr_from_ram(KVMState *s, ram_addr_t ram_addr, |
150 | target_phys_addr_t *phys_addr) | |
151 | { | |
152 | int i; | |
153 | ||
154 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
155 | KVMSlot *mem = &s->slots[i]; | |
156 | ||
157 | if (ram_addr >= mem->phys_offset && | |
158 | ram_addr < mem->phys_offset + mem->memory_size) { | |
159 | *phys_addr = mem->start_addr + (ram_addr - mem->phys_offset); | |
160 | return 1; | |
161 | } | |
162 | } | |
163 | ||
164 | return 0; | |
165 | } | |
166 | ||
5832d1f2 AL |
167 | static int kvm_set_user_memory_region(KVMState *s, KVMSlot *slot) |
168 | { | |
169 | struct kvm_userspace_memory_region mem; | |
170 | ||
171 | mem.slot = slot->slot; | |
172 | mem.guest_phys_addr = slot->start_addr; | |
173 | mem.memory_size = slot->memory_size; | |
b2e0a138 | 174 | mem.userspace_addr = (unsigned long)qemu_safe_ram_ptr(slot->phys_offset); |
5832d1f2 | 175 | mem.flags = slot->flags; |
4495d6a7 JK |
176 | if (s->migration_log) { |
177 | mem.flags |= KVM_MEM_LOG_DIRTY_PAGES; | |
178 | } | |
5832d1f2 AL |
179 | return kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem); |
180 | } | |
181 | ||
8d2ba1fb JK |
182 | static void kvm_reset_vcpu(void *opaque) |
183 | { | |
184 | CPUState *env = opaque; | |
185 | ||
caa5af0f | 186 | kvm_arch_reset_vcpu(env); |
8d2ba1fb | 187 | } |
5832d1f2 | 188 | |
6f725c13 GC |
189 | int kvm_irqchip_in_kernel(void) |
190 | { | |
191 | return kvm_state->irqchip_in_kernel; | |
192 | } | |
193 | ||
194 | int kvm_pit_in_kernel(void) | |
195 | { | |
196 | return kvm_state->pit_in_kernel; | |
197 | } | |
198 | ||
05330448 AL |
199 | int kvm_init_vcpu(CPUState *env) |
200 | { | |
201 | KVMState *s = kvm_state; | |
202 | long mmap_size; | |
203 | int ret; | |
204 | ||
8c0d577e | 205 | DPRINTF("kvm_init_vcpu\n"); |
05330448 | 206 | |
984b5181 | 207 | ret = kvm_vm_ioctl(s, KVM_CREATE_VCPU, env->cpu_index); |
05330448 | 208 | if (ret < 0) { |
8c0d577e | 209 | DPRINTF("kvm_create_vcpu failed\n"); |
05330448 AL |
210 | goto err; |
211 | } | |
212 | ||
213 | env->kvm_fd = ret; | |
214 | env->kvm_state = s; | |
d841b6c4 | 215 | env->kvm_vcpu_dirty = 1; |
05330448 AL |
216 | |
217 | mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0); | |
218 | if (mmap_size < 0) { | |
748a680b | 219 | ret = mmap_size; |
8c0d577e | 220 | DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n"); |
05330448 AL |
221 | goto err; |
222 | } | |
223 | ||
224 | env->kvm_run = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED, | |
225 | env->kvm_fd, 0); | |
226 | if (env->kvm_run == MAP_FAILED) { | |
227 | ret = -errno; | |
8c0d577e | 228 | DPRINTF("mmap'ing vcpu state failed\n"); |
05330448 AL |
229 | goto err; |
230 | } | |
231 | ||
a426e122 JK |
232 | if (s->coalesced_mmio && !s->coalesced_mmio_ring) { |
233 | s->coalesced_mmio_ring = | |
234 | (void *)env->kvm_run + s->coalesced_mmio * PAGE_SIZE; | |
235 | } | |
62a2744c | 236 | |
05330448 | 237 | ret = kvm_arch_init_vcpu(env); |
8d2ba1fb | 238 | if (ret == 0) { |
a08d4367 | 239 | qemu_register_reset(kvm_reset_vcpu, env); |
caa5af0f | 240 | kvm_arch_reset_vcpu(env); |
8d2ba1fb | 241 | } |
05330448 AL |
242 | err: |
243 | return ret; | |
244 | } | |
245 | ||
5832d1f2 AL |
246 | /* |
247 | * dirty pages logging control | |
248 | */ | |
25254bbc MT |
249 | |
250 | static int kvm_mem_flags(KVMState *s, bool log_dirty) | |
251 | { | |
252 | return log_dirty ? KVM_MEM_LOG_DIRTY_PAGES : 0; | |
253 | } | |
254 | ||
255 | static int kvm_slot_dirty_pages_log_change(KVMSlot *mem, bool log_dirty) | |
5832d1f2 AL |
256 | { |
257 | KVMState *s = kvm_state; | |
25254bbc | 258 | int flags, mask = KVM_MEM_LOG_DIRTY_PAGES; |
4495d6a7 JK |
259 | int old_flags; |
260 | ||
4495d6a7 | 261 | old_flags = mem->flags; |
5832d1f2 | 262 | |
25254bbc | 263 | flags = (mem->flags & ~mask) | kvm_mem_flags(s, log_dirty); |
5832d1f2 AL |
264 | mem->flags = flags; |
265 | ||
4495d6a7 JK |
266 | /* If nothing changed effectively, no need to issue ioctl */ |
267 | if (s->migration_log) { | |
268 | flags |= KVM_MEM_LOG_DIRTY_PAGES; | |
269 | } | |
25254bbc | 270 | |
4495d6a7 | 271 | if (flags == old_flags) { |
25254bbc | 272 | return 0; |
4495d6a7 JK |
273 | } |
274 | ||
5832d1f2 AL |
275 | return kvm_set_user_memory_region(s, mem); |
276 | } | |
277 | ||
25254bbc MT |
278 | static int kvm_dirty_pages_log_change(target_phys_addr_t phys_addr, |
279 | ram_addr_t size, bool log_dirty) | |
280 | { | |
281 | KVMState *s = kvm_state; | |
282 | KVMSlot *mem = kvm_lookup_matching_slot(s, phys_addr, phys_addr + size); | |
283 | ||
284 | if (mem == NULL) { | |
285 | fprintf(stderr, "BUG: %s: invalid parameters " TARGET_FMT_plx "-" | |
286 | TARGET_FMT_plx "\n", __func__, phys_addr, | |
287 | (target_phys_addr_t)(phys_addr + size - 1)); | |
288 | return -EINVAL; | |
289 | } | |
290 | return kvm_slot_dirty_pages_log_change(mem, log_dirty); | |
291 | } | |
292 | ||
e5896b12 AP |
293 | static int kvm_log_start(CPUPhysMemoryClient *client, |
294 | target_phys_addr_t phys_addr, ram_addr_t size) | |
5832d1f2 | 295 | { |
25254bbc | 296 | return kvm_dirty_pages_log_change(phys_addr, size, true); |
5832d1f2 AL |
297 | } |
298 | ||
e5896b12 AP |
299 | static int kvm_log_stop(CPUPhysMemoryClient *client, |
300 | target_phys_addr_t phys_addr, ram_addr_t size) | |
5832d1f2 | 301 | { |
25254bbc | 302 | return kvm_dirty_pages_log_change(phys_addr, size, false); |
5832d1f2 AL |
303 | } |
304 | ||
7b8f3b78 | 305 | static int kvm_set_migration_log(int enable) |
4495d6a7 JK |
306 | { |
307 | KVMState *s = kvm_state; | |
308 | KVMSlot *mem; | |
309 | int i, err; | |
310 | ||
311 | s->migration_log = enable; | |
312 | ||
313 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
314 | mem = &s->slots[i]; | |
315 | ||
70fedd76 AW |
316 | if (!mem->memory_size) { |
317 | continue; | |
318 | } | |
4495d6a7 JK |
319 | if (!!(mem->flags & KVM_MEM_LOG_DIRTY_PAGES) == enable) { |
320 | continue; | |
321 | } | |
322 | err = kvm_set_user_memory_region(s, mem); | |
323 | if (err) { | |
324 | return err; | |
325 | } | |
326 | } | |
327 | return 0; | |
328 | } | |
329 | ||
8369e01c MT |
330 | /* get kvm's dirty pages bitmap and update qemu's */ |
331 | static int kvm_get_dirty_pages_log_range(unsigned long start_addr, | |
332 | unsigned long *bitmap, | |
333 | unsigned long offset, | |
334 | unsigned long mem_size) | |
96c1606b | 335 | { |
8369e01c MT |
336 | unsigned int i, j; |
337 | unsigned long page_number, addr, addr1, c; | |
338 | ram_addr_t ram_addr; | |
339 | unsigned int len = ((mem_size / TARGET_PAGE_SIZE) + HOST_LONG_BITS - 1) / | |
340 | HOST_LONG_BITS; | |
341 | ||
342 | /* | |
343 | * bitmap-traveling is faster than memory-traveling (for addr...) | |
344 | * especially when most of the memory is not dirty. | |
345 | */ | |
346 | for (i = 0; i < len; i++) { | |
347 | if (bitmap[i] != 0) { | |
348 | c = leul_to_cpu(bitmap[i]); | |
349 | do { | |
350 | j = ffsl(c) - 1; | |
351 | c &= ~(1ul << j); | |
352 | page_number = i * HOST_LONG_BITS + j; | |
353 | addr1 = page_number * TARGET_PAGE_SIZE; | |
354 | addr = offset + addr1; | |
355 | ram_addr = cpu_get_physical_page_desc(addr); | |
356 | cpu_physical_memory_set_dirty(ram_addr); | |
357 | } while (c != 0); | |
358 | } | |
359 | } | |
360 | return 0; | |
96c1606b AG |
361 | } |
362 | ||
8369e01c MT |
363 | #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1)) |
364 | ||
5832d1f2 AL |
365 | /** |
366 | * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space | |
367 | * This function updates qemu's dirty bitmap using cpu_physical_memory_set_dirty(). | |
368 | * This means all bits are set to dirty. | |
369 | * | |
d3f8d37f | 370 | * @start_add: start of logged region. |
5832d1f2 AL |
371 | * @end_addr: end of logged region. |
372 | */ | |
7b8f3b78 | 373 | static int kvm_physical_sync_dirty_bitmap(target_phys_addr_t start_addr, |
a426e122 | 374 | target_phys_addr_t end_addr) |
5832d1f2 AL |
375 | { |
376 | KVMState *s = kvm_state; | |
151f7749 | 377 | unsigned long size, allocated_size = 0; |
151f7749 JK |
378 | KVMDirtyLog d; |
379 | KVMSlot *mem; | |
380 | int ret = 0; | |
5832d1f2 | 381 | |
151f7749 JK |
382 | d.dirty_bitmap = NULL; |
383 | while (start_addr < end_addr) { | |
384 | mem = kvm_lookup_overlapping_slot(s, start_addr, end_addr); | |
385 | if (mem == NULL) { | |
386 | break; | |
387 | } | |
5832d1f2 | 388 | |
51b0c606 MT |
389 | /* XXX bad kernel interface alert |
390 | * For dirty bitmap, kernel allocates array of size aligned to | |
391 | * bits-per-long. But for case when the kernel is 64bits and | |
392 | * the userspace is 32bits, userspace can't align to the same | |
393 | * bits-per-long, since sizeof(long) is different between kernel | |
394 | * and user space. This way, userspace will provide buffer which | |
395 | * may be 4 bytes less than the kernel will use, resulting in | |
396 | * userspace memory corruption (which is not detectable by valgrind | |
397 | * too, in most cases). | |
398 | * So for now, let's align to 64 instead of HOST_LONG_BITS here, in | |
399 | * a hope that sizeof(long) wont become >8 any time soon. | |
400 | */ | |
401 | size = ALIGN(((mem->memory_size) >> TARGET_PAGE_BITS), | |
402 | /*HOST_LONG_BITS*/ 64) / 8; | |
151f7749 | 403 | if (!d.dirty_bitmap) { |
7267c094 | 404 | d.dirty_bitmap = g_malloc(size); |
151f7749 | 405 | } else if (size > allocated_size) { |
7267c094 | 406 | d.dirty_bitmap = g_realloc(d.dirty_bitmap, size); |
151f7749 JK |
407 | } |
408 | allocated_size = size; | |
409 | memset(d.dirty_bitmap, 0, allocated_size); | |
5832d1f2 | 410 | |
151f7749 | 411 | d.slot = mem->slot; |
5832d1f2 | 412 | |
6e489f3f | 413 | if (kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d) == -1) { |
8c0d577e | 414 | DPRINTF("ioctl failed %d\n", errno); |
151f7749 JK |
415 | ret = -1; |
416 | break; | |
417 | } | |
5832d1f2 | 418 | |
8369e01c MT |
419 | kvm_get_dirty_pages_log_range(mem->start_addr, d.dirty_bitmap, |
420 | mem->start_addr, mem->memory_size); | |
421 | start_addr = mem->start_addr + mem->memory_size; | |
5832d1f2 | 422 | } |
7267c094 | 423 | g_free(d.dirty_bitmap); |
151f7749 JK |
424 | |
425 | return ret; | |
5832d1f2 AL |
426 | } |
427 | ||
c227f099 | 428 | int kvm_coalesce_mmio_region(target_phys_addr_t start, ram_addr_t size) |
f65ed4c1 AL |
429 | { |
430 | int ret = -ENOSYS; | |
f65ed4c1 AL |
431 | KVMState *s = kvm_state; |
432 | ||
433 | if (s->coalesced_mmio) { | |
434 | struct kvm_coalesced_mmio_zone zone; | |
435 | ||
436 | zone.addr = start; | |
437 | zone.size = size; | |
438 | ||
439 | ret = kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone); | |
440 | } | |
f65ed4c1 AL |
441 | |
442 | return ret; | |
443 | } | |
444 | ||
c227f099 | 445 | int kvm_uncoalesce_mmio_region(target_phys_addr_t start, ram_addr_t size) |
f65ed4c1 AL |
446 | { |
447 | int ret = -ENOSYS; | |
f65ed4c1 AL |
448 | KVMState *s = kvm_state; |
449 | ||
450 | if (s->coalesced_mmio) { | |
451 | struct kvm_coalesced_mmio_zone zone; | |
452 | ||
453 | zone.addr = start; | |
454 | zone.size = size; | |
455 | ||
456 | ret = kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone); | |
457 | } | |
f65ed4c1 AL |
458 | |
459 | return ret; | |
460 | } | |
461 | ||
ad7b8b33 AL |
462 | int kvm_check_extension(KVMState *s, unsigned int extension) |
463 | { | |
464 | int ret; | |
465 | ||
466 | ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension); | |
467 | if (ret < 0) { | |
468 | ret = 0; | |
469 | } | |
470 | ||
471 | return ret; | |
472 | } | |
473 | ||
d2f2b8a7 SH |
474 | static int kvm_check_many_ioeventfds(void) |
475 | { | |
d0dcac83 SH |
476 | /* Userspace can use ioeventfd for io notification. This requires a host |
477 | * that supports eventfd(2) and an I/O thread; since eventfd does not | |
478 | * support SIGIO it cannot interrupt the vcpu. | |
479 | * | |
480 | * Older kernels have a 6 device limit on the KVM io bus. Find out so we | |
d2f2b8a7 SH |
481 | * can avoid creating too many ioeventfds. |
482 | */ | |
12d4536f | 483 | #if defined(CONFIG_EVENTFD) |
d2f2b8a7 SH |
484 | int ioeventfds[7]; |
485 | int i, ret = 0; | |
486 | for (i = 0; i < ARRAY_SIZE(ioeventfds); i++) { | |
487 | ioeventfds[i] = eventfd(0, EFD_CLOEXEC); | |
488 | if (ioeventfds[i] < 0) { | |
489 | break; | |
490 | } | |
491 | ret = kvm_set_ioeventfd_pio_word(ioeventfds[i], 0, i, true); | |
492 | if (ret < 0) { | |
493 | close(ioeventfds[i]); | |
494 | break; | |
495 | } | |
496 | } | |
497 | ||
498 | /* Decide whether many devices are supported or not */ | |
499 | ret = i == ARRAY_SIZE(ioeventfds); | |
500 | ||
501 | while (i-- > 0) { | |
502 | kvm_set_ioeventfd_pio_word(ioeventfds[i], 0, i, false); | |
503 | close(ioeventfds[i]); | |
504 | } | |
505 | return ret; | |
506 | #else | |
507 | return 0; | |
508 | #endif | |
509 | } | |
510 | ||
94a8d39a JK |
511 | static const KVMCapabilityInfo * |
512 | kvm_check_extension_list(KVMState *s, const KVMCapabilityInfo *list) | |
513 | { | |
514 | while (list->name) { | |
515 | if (!kvm_check_extension(s, list->value)) { | |
516 | return list; | |
517 | } | |
518 | list++; | |
519 | } | |
520 | return NULL; | |
521 | } | |
522 | ||
a426e122 | 523 | static void kvm_set_phys_mem(target_phys_addr_t start_addr, ram_addr_t size, |
25254bbc | 524 | ram_addr_t phys_offset, bool log_dirty) |
46dbef6a MT |
525 | { |
526 | KVMState *s = kvm_state; | |
527 | ram_addr_t flags = phys_offset & ~TARGET_PAGE_MASK; | |
528 | KVMSlot *mem, old; | |
529 | int err; | |
530 | ||
14542fea GN |
531 | /* kvm works in page size chunks, but the function may be called |
532 | with sub-page size and unaligned start address. */ | |
533 | size = TARGET_PAGE_ALIGN(size); | |
534 | start_addr = TARGET_PAGE_ALIGN(start_addr); | |
46dbef6a MT |
535 | |
536 | /* KVM does not support read-only slots */ | |
537 | phys_offset &= ~IO_MEM_ROM; | |
538 | ||
539 | while (1) { | |
540 | mem = kvm_lookup_overlapping_slot(s, start_addr, start_addr + size); | |
541 | if (!mem) { | |
542 | break; | |
543 | } | |
544 | ||
545 | if (flags < IO_MEM_UNASSIGNED && start_addr >= mem->start_addr && | |
546 | (start_addr + size <= mem->start_addr + mem->memory_size) && | |
547 | (phys_offset - start_addr == mem->phys_offset - mem->start_addr)) { | |
548 | /* The new slot fits into the existing one and comes with | |
25254bbc MT |
549 | * identical parameters - update flags and done. */ |
550 | kvm_slot_dirty_pages_log_change(mem, log_dirty); | |
46dbef6a MT |
551 | return; |
552 | } | |
553 | ||
554 | old = *mem; | |
555 | ||
556 | /* unregister the overlapping slot */ | |
557 | mem->memory_size = 0; | |
558 | err = kvm_set_user_memory_region(s, mem); | |
559 | if (err) { | |
560 | fprintf(stderr, "%s: error unregistering overlapping slot: %s\n", | |
561 | __func__, strerror(-err)); | |
562 | abort(); | |
563 | } | |
564 | ||
565 | /* Workaround for older KVM versions: we can't join slots, even not by | |
566 | * unregistering the previous ones and then registering the larger | |
567 | * slot. We have to maintain the existing fragmentation. Sigh. | |
568 | * | |
569 | * This workaround assumes that the new slot starts at the same | |
570 | * address as the first existing one. If not or if some overlapping | |
571 | * slot comes around later, we will fail (not seen in practice so far) | |
572 | * - and actually require a recent KVM version. */ | |
573 | if (s->broken_set_mem_region && | |
574 | old.start_addr == start_addr && old.memory_size < size && | |
575 | flags < IO_MEM_UNASSIGNED) { | |
576 | mem = kvm_alloc_slot(s); | |
577 | mem->memory_size = old.memory_size; | |
578 | mem->start_addr = old.start_addr; | |
579 | mem->phys_offset = old.phys_offset; | |
25254bbc | 580 | mem->flags = kvm_mem_flags(s, log_dirty); |
46dbef6a MT |
581 | |
582 | err = kvm_set_user_memory_region(s, mem); | |
583 | if (err) { | |
584 | fprintf(stderr, "%s: error updating slot: %s\n", __func__, | |
585 | strerror(-err)); | |
586 | abort(); | |
587 | } | |
588 | ||
589 | start_addr += old.memory_size; | |
590 | phys_offset += old.memory_size; | |
591 | size -= old.memory_size; | |
592 | continue; | |
593 | } | |
594 | ||
595 | /* register prefix slot */ | |
596 | if (old.start_addr < start_addr) { | |
597 | mem = kvm_alloc_slot(s); | |
598 | mem->memory_size = start_addr - old.start_addr; | |
599 | mem->start_addr = old.start_addr; | |
600 | mem->phys_offset = old.phys_offset; | |
25254bbc | 601 | mem->flags = kvm_mem_flags(s, log_dirty); |
46dbef6a MT |
602 | |
603 | err = kvm_set_user_memory_region(s, mem); | |
604 | if (err) { | |
605 | fprintf(stderr, "%s: error registering prefix slot: %s\n", | |
606 | __func__, strerror(-err)); | |
d4d6868f AG |
607 | #ifdef TARGET_PPC |
608 | fprintf(stderr, "%s: This is probably because your kernel's " \ | |
609 | "PAGE_SIZE is too big. Please try to use 4k " \ | |
610 | "PAGE_SIZE!\n", __func__); | |
611 | #endif | |
46dbef6a MT |
612 | abort(); |
613 | } | |
614 | } | |
615 | ||
616 | /* register suffix slot */ | |
617 | if (old.start_addr + old.memory_size > start_addr + size) { | |
618 | ram_addr_t size_delta; | |
619 | ||
620 | mem = kvm_alloc_slot(s); | |
621 | mem->start_addr = start_addr + size; | |
622 | size_delta = mem->start_addr - old.start_addr; | |
623 | mem->memory_size = old.memory_size - size_delta; | |
624 | mem->phys_offset = old.phys_offset + size_delta; | |
25254bbc | 625 | mem->flags = kvm_mem_flags(s, log_dirty); |
46dbef6a MT |
626 | |
627 | err = kvm_set_user_memory_region(s, mem); | |
628 | if (err) { | |
629 | fprintf(stderr, "%s: error registering suffix slot: %s\n", | |
630 | __func__, strerror(-err)); | |
631 | abort(); | |
632 | } | |
633 | } | |
634 | } | |
635 | ||
636 | /* in case the KVM bug workaround already "consumed" the new slot */ | |
a426e122 | 637 | if (!size) { |
46dbef6a | 638 | return; |
a426e122 | 639 | } |
46dbef6a | 640 | /* KVM does not need to know about this memory */ |
a426e122 | 641 | if (flags >= IO_MEM_UNASSIGNED) { |
46dbef6a | 642 | return; |
a426e122 | 643 | } |
46dbef6a MT |
644 | mem = kvm_alloc_slot(s); |
645 | mem->memory_size = size; | |
646 | mem->start_addr = start_addr; | |
647 | mem->phys_offset = phys_offset; | |
25254bbc | 648 | mem->flags = kvm_mem_flags(s, log_dirty); |
46dbef6a MT |
649 | |
650 | err = kvm_set_user_memory_region(s, mem); | |
651 | if (err) { | |
652 | fprintf(stderr, "%s: error registering slot: %s\n", __func__, | |
653 | strerror(-err)); | |
654 | abort(); | |
655 | } | |
656 | } | |
657 | ||
7b8f3b78 | 658 | static void kvm_client_set_memory(struct CPUPhysMemoryClient *client, |
a426e122 | 659 | target_phys_addr_t start_addr, |
0fd542fb MT |
660 | ram_addr_t size, ram_addr_t phys_offset, |
661 | bool log_dirty) | |
7b8f3b78 | 662 | { |
25254bbc | 663 | kvm_set_phys_mem(start_addr, size, phys_offset, log_dirty); |
7b8f3b78 MT |
664 | } |
665 | ||
666 | static int kvm_client_sync_dirty_bitmap(struct CPUPhysMemoryClient *client, | |
a426e122 JK |
667 | target_phys_addr_t start_addr, |
668 | target_phys_addr_t end_addr) | |
7b8f3b78 | 669 | { |
a426e122 | 670 | return kvm_physical_sync_dirty_bitmap(start_addr, end_addr); |
7b8f3b78 MT |
671 | } |
672 | ||
673 | static int kvm_client_migration_log(struct CPUPhysMemoryClient *client, | |
a426e122 | 674 | int enable) |
7b8f3b78 | 675 | { |
a426e122 | 676 | return kvm_set_migration_log(enable); |
7b8f3b78 MT |
677 | } |
678 | ||
679 | static CPUPhysMemoryClient kvm_cpu_phys_memory_client = { | |
a426e122 JK |
680 | .set_memory = kvm_client_set_memory, |
681 | .sync_dirty_bitmap = kvm_client_sync_dirty_bitmap, | |
682 | .migration_log = kvm_client_migration_log, | |
e5896b12 AP |
683 | .log_start = kvm_log_start, |
684 | .log_stop = kvm_log_stop, | |
7b8f3b78 MT |
685 | }; |
686 | ||
aa7f74d1 JK |
687 | static void kvm_handle_interrupt(CPUState *env, int mask) |
688 | { | |
689 | env->interrupt_request |= mask; | |
690 | ||
691 | if (!qemu_cpu_is_self(env)) { | |
692 | qemu_cpu_kick(env); | |
693 | } | |
694 | } | |
695 | ||
cad1e282 | 696 | int kvm_init(void) |
05330448 | 697 | { |
168ccc11 JK |
698 | static const char upgrade_note[] = |
699 | "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n" | |
700 | "(see http://sourceforge.net/projects/kvm).\n"; | |
05330448 | 701 | KVMState *s; |
94a8d39a | 702 | const KVMCapabilityInfo *missing_cap; |
05330448 AL |
703 | int ret; |
704 | int i; | |
705 | ||
7267c094 | 706 | s = g_malloc0(sizeof(KVMState)); |
05330448 | 707 | |
e22a25c9 | 708 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
72cf2d4f | 709 | QTAILQ_INIT(&s->kvm_sw_breakpoints); |
e22a25c9 | 710 | #endif |
a426e122 | 711 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { |
05330448 | 712 | s->slots[i].slot = i; |
a426e122 | 713 | } |
05330448 | 714 | s->vmfd = -1; |
40ff6d7e | 715 | s->fd = qemu_open("/dev/kvm", O_RDWR); |
05330448 AL |
716 | if (s->fd == -1) { |
717 | fprintf(stderr, "Could not access KVM kernel module: %m\n"); | |
718 | ret = -errno; | |
719 | goto err; | |
720 | } | |
721 | ||
722 | ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0); | |
723 | if (ret < KVM_API_VERSION) { | |
a426e122 | 724 | if (ret > 0) { |
05330448 | 725 | ret = -EINVAL; |
a426e122 | 726 | } |
05330448 AL |
727 | fprintf(stderr, "kvm version too old\n"); |
728 | goto err; | |
729 | } | |
730 | ||
731 | if (ret > KVM_API_VERSION) { | |
732 | ret = -EINVAL; | |
733 | fprintf(stderr, "kvm version not supported\n"); | |
734 | goto err; | |
735 | } | |
736 | ||
737 | s->vmfd = kvm_ioctl(s, KVM_CREATE_VM, 0); | |
0104dcac AG |
738 | if (s->vmfd < 0) { |
739 | #ifdef TARGET_S390X | |
740 | fprintf(stderr, "Please add the 'switch_amode' kernel parameter to " | |
741 | "your host kernel command line\n"); | |
742 | #endif | |
05330448 | 743 | goto err; |
0104dcac | 744 | } |
05330448 | 745 | |
94a8d39a JK |
746 | missing_cap = kvm_check_extension_list(s, kvm_required_capabilites); |
747 | if (!missing_cap) { | |
748 | missing_cap = | |
749 | kvm_check_extension_list(s, kvm_arch_required_capabilities); | |
05330448 | 750 | } |
94a8d39a | 751 | if (missing_cap) { |
ad7b8b33 | 752 | ret = -EINVAL; |
94a8d39a JK |
753 | fprintf(stderr, "kvm does not support %s\n%s", |
754 | missing_cap->name, upgrade_note); | |
d85dc283 AL |
755 | goto err; |
756 | } | |
757 | ||
ad7b8b33 | 758 | s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO); |
f65ed4c1 | 759 | |
e69917e2 | 760 | s->broken_set_mem_region = 1; |
14a09518 | 761 | ret = kvm_check_extension(s, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS); |
e69917e2 JK |
762 | if (ret > 0) { |
763 | s->broken_set_mem_region = 0; | |
764 | } | |
e69917e2 | 765 | |
a0fb002c JK |
766 | #ifdef KVM_CAP_VCPU_EVENTS |
767 | s->vcpu_events = kvm_check_extension(s, KVM_CAP_VCPU_EVENTS); | |
768 | #endif | |
769 | ||
b0b1d690 JK |
770 | s->robust_singlestep = |
771 | kvm_check_extension(s, KVM_CAP_X86_ROBUST_SINGLESTEP); | |
b0b1d690 | 772 | |
ff44f1a3 JK |
773 | #ifdef KVM_CAP_DEBUGREGS |
774 | s->debugregs = kvm_check_extension(s, KVM_CAP_DEBUGREGS); | |
775 | #endif | |
776 | ||
f1665b21 SY |
777 | #ifdef KVM_CAP_XSAVE |
778 | s->xsave = kvm_check_extension(s, KVM_CAP_XSAVE); | |
779 | #endif | |
780 | ||
f1665b21 SY |
781 | #ifdef KVM_CAP_XCRS |
782 | s->xcrs = kvm_check_extension(s, KVM_CAP_XCRS); | |
783 | #endif | |
784 | ||
cad1e282 | 785 | ret = kvm_arch_init(s); |
a426e122 | 786 | if (ret < 0) { |
05330448 | 787 | goto err; |
a426e122 | 788 | } |
05330448 AL |
789 | |
790 | kvm_state = s; | |
7b8f3b78 | 791 | cpu_register_phys_memory_client(&kvm_cpu_phys_memory_client); |
05330448 | 792 | |
d2f2b8a7 SH |
793 | s->many_ioeventfds = kvm_check_many_ioeventfds(); |
794 | ||
aa7f74d1 JK |
795 | cpu_interrupt_handler = kvm_handle_interrupt; |
796 | ||
05330448 AL |
797 | return 0; |
798 | ||
799 | err: | |
800 | if (s) { | |
a426e122 | 801 | if (s->vmfd != -1) { |
05330448 | 802 | close(s->vmfd); |
a426e122 JK |
803 | } |
804 | if (s->fd != -1) { | |
05330448 | 805 | close(s->fd); |
a426e122 | 806 | } |
05330448 | 807 | } |
7267c094 | 808 | g_free(s); |
05330448 AL |
809 | |
810 | return ret; | |
811 | } | |
812 | ||
b30e93e9 JK |
813 | static void kvm_handle_io(uint16_t port, void *data, int direction, int size, |
814 | uint32_t count) | |
05330448 AL |
815 | { |
816 | int i; | |
817 | uint8_t *ptr = data; | |
818 | ||
819 | for (i = 0; i < count; i++) { | |
820 | if (direction == KVM_EXIT_IO_IN) { | |
821 | switch (size) { | |
822 | case 1: | |
afcea8cb | 823 | stb_p(ptr, cpu_inb(port)); |
05330448 AL |
824 | break; |
825 | case 2: | |
afcea8cb | 826 | stw_p(ptr, cpu_inw(port)); |
05330448 AL |
827 | break; |
828 | case 4: | |
afcea8cb | 829 | stl_p(ptr, cpu_inl(port)); |
05330448 AL |
830 | break; |
831 | } | |
832 | } else { | |
833 | switch (size) { | |
834 | case 1: | |
afcea8cb | 835 | cpu_outb(port, ldub_p(ptr)); |
05330448 AL |
836 | break; |
837 | case 2: | |
afcea8cb | 838 | cpu_outw(port, lduw_p(ptr)); |
05330448 AL |
839 | break; |
840 | case 4: | |
afcea8cb | 841 | cpu_outl(port, ldl_p(ptr)); |
05330448 AL |
842 | break; |
843 | } | |
844 | } | |
845 | ||
846 | ptr += size; | |
847 | } | |
05330448 AL |
848 | } |
849 | ||
73aaec4a | 850 | static int kvm_handle_internal_error(CPUState *env, struct kvm_run *run) |
7c80eef8 | 851 | { |
bb44e0d1 | 852 | fprintf(stderr, "KVM internal error."); |
7c80eef8 MT |
853 | if (kvm_check_extension(kvm_state, KVM_CAP_INTERNAL_ERROR_DATA)) { |
854 | int i; | |
855 | ||
bb44e0d1 | 856 | fprintf(stderr, " Suberror: %d\n", run->internal.suberror); |
7c80eef8 MT |
857 | for (i = 0; i < run->internal.ndata; ++i) { |
858 | fprintf(stderr, "extra data[%d]: %"PRIx64"\n", | |
859 | i, (uint64_t)run->internal.data[i]); | |
860 | } | |
bb44e0d1 JK |
861 | } else { |
862 | fprintf(stderr, "\n"); | |
7c80eef8 | 863 | } |
7c80eef8 MT |
864 | if (run->internal.suberror == KVM_INTERNAL_ERROR_EMULATION) { |
865 | fprintf(stderr, "emulation failure\n"); | |
a426e122 | 866 | if (!kvm_arch_stop_on_emulation_error(env)) { |
f5c848ee | 867 | cpu_dump_state(env, stderr, fprintf, CPU_DUMP_CODE); |
d73cd8f4 | 868 | return EXCP_INTERRUPT; |
a426e122 | 869 | } |
7c80eef8 MT |
870 | } |
871 | /* FIXME: Should trigger a qmp message to let management know | |
872 | * something went wrong. | |
873 | */ | |
73aaec4a | 874 | return -1; |
7c80eef8 | 875 | } |
7c80eef8 | 876 | |
62a2744c | 877 | void kvm_flush_coalesced_mmio_buffer(void) |
f65ed4c1 | 878 | { |
f65ed4c1 | 879 | KVMState *s = kvm_state; |
1cae88b9 AK |
880 | |
881 | if (s->coalesced_flush_in_progress) { | |
882 | return; | |
883 | } | |
884 | ||
885 | s->coalesced_flush_in_progress = true; | |
886 | ||
62a2744c SY |
887 | if (s->coalesced_mmio_ring) { |
888 | struct kvm_coalesced_mmio_ring *ring = s->coalesced_mmio_ring; | |
f65ed4c1 AL |
889 | while (ring->first != ring->last) { |
890 | struct kvm_coalesced_mmio *ent; | |
891 | ||
892 | ent = &ring->coalesced_mmio[ring->first]; | |
893 | ||
894 | cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len); | |
85199474 | 895 | smp_wmb(); |
f65ed4c1 AL |
896 | ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX; |
897 | } | |
898 | } | |
1cae88b9 AK |
899 | |
900 | s->coalesced_flush_in_progress = false; | |
f65ed4c1 AL |
901 | } |
902 | ||
2705d56a | 903 | static void do_kvm_cpu_synchronize_state(void *_env) |
4c0960c0 | 904 | { |
2705d56a JK |
905 | CPUState *env = _env; |
906 | ||
9ded2744 | 907 | if (!env->kvm_vcpu_dirty) { |
4c0960c0 | 908 | kvm_arch_get_registers(env); |
9ded2744 | 909 | env->kvm_vcpu_dirty = 1; |
4c0960c0 AK |
910 | } |
911 | } | |
912 | ||
2705d56a JK |
913 | void kvm_cpu_synchronize_state(CPUState *env) |
914 | { | |
a426e122 | 915 | if (!env->kvm_vcpu_dirty) { |
2705d56a | 916 | run_on_cpu(env, do_kvm_cpu_synchronize_state, env); |
a426e122 | 917 | } |
2705d56a JK |
918 | } |
919 | ||
ea375f9a JK |
920 | void kvm_cpu_synchronize_post_reset(CPUState *env) |
921 | { | |
922 | kvm_arch_put_registers(env, KVM_PUT_RESET_STATE); | |
923 | env->kvm_vcpu_dirty = 0; | |
924 | } | |
925 | ||
926 | void kvm_cpu_synchronize_post_init(CPUState *env) | |
927 | { | |
928 | kvm_arch_put_registers(env, KVM_PUT_FULL_STATE); | |
929 | env->kvm_vcpu_dirty = 0; | |
930 | } | |
931 | ||
05330448 AL |
932 | int kvm_cpu_exec(CPUState *env) |
933 | { | |
934 | struct kvm_run *run = env->kvm_run; | |
7cbb533f | 935 | int ret, run_ret; |
05330448 | 936 | |
8c0d577e | 937 | DPRINTF("kvm_cpu_exec()\n"); |
05330448 | 938 | |
99036865 | 939 | if (kvm_arch_process_async_events(env)) { |
9ccfac9e | 940 | env->exit_request = 0; |
6792a57b | 941 | return EXCP_HLT; |
9ccfac9e | 942 | } |
0af691d7 | 943 | |
6792a57b JK |
944 | cpu_single_env = env; |
945 | ||
9ccfac9e | 946 | do { |
9ded2744 | 947 | if (env->kvm_vcpu_dirty) { |
ea375f9a | 948 | kvm_arch_put_registers(env, KVM_PUT_RUNTIME_STATE); |
9ded2744 | 949 | env->kvm_vcpu_dirty = 0; |
4c0960c0 AK |
950 | } |
951 | ||
8c14c173 | 952 | kvm_arch_pre_run(env, run); |
9ccfac9e JK |
953 | if (env->exit_request) { |
954 | DPRINTF("interrupt exit requested\n"); | |
955 | /* | |
956 | * KVM requires us to reenter the kernel after IO exits to complete | |
957 | * instruction emulation. This self-signal will ensure that we | |
958 | * leave ASAP again. | |
959 | */ | |
960 | qemu_cpu_kick_self(); | |
961 | } | |
273faf1b | 962 | cpu_single_env = NULL; |
d549db5a | 963 | qemu_mutex_unlock_iothread(); |
9ccfac9e | 964 | |
7cbb533f | 965 | run_ret = kvm_vcpu_ioctl(env, KVM_RUN, 0); |
9ccfac9e | 966 | |
d549db5a | 967 | qemu_mutex_lock_iothread(); |
273faf1b | 968 | cpu_single_env = env; |
05330448 AL |
969 | kvm_arch_post_run(env, run); |
970 | ||
b0c883b5 JK |
971 | kvm_flush_coalesced_mmio_buffer(); |
972 | ||
7cbb533f | 973 | if (run_ret < 0) { |
dc77d341 JK |
974 | if (run_ret == -EINTR || run_ret == -EAGAIN) { |
975 | DPRINTF("io window exit\n"); | |
d73cd8f4 | 976 | ret = EXCP_INTERRUPT; |
dc77d341 JK |
977 | break; |
978 | } | |
7cbb533f | 979 | DPRINTF("kvm run failed %s\n", strerror(-run_ret)); |
05330448 AL |
980 | abort(); |
981 | } | |
982 | ||
05330448 AL |
983 | switch (run->exit_reason) { |
984 | case KVM_EXIT_IO: | |
8c0d577e | 985 | DPRINTF("handle_io\n"); |
b30e93e9 JK |
986 | kvm_handle_io(run->io.port, |
987 | (uint8_t *)run + run->io.data_offset, | |
988 | run->io.direction, | |
989 | run->io.size, | |
990 | run->io.count); | |
d73cd8f4 | 991 | ret = 0; |
05330448 AL |
992 | break; |
993 | case KVM_EXIT_MMIO: | |
8c0d577e | 994 | DPRINTF("handle_mmio\n"); |
05330448 AL |
995 | cpu_physical_memory_rw(run->mmio.phys_addr, |
996 | run->mmio.data, | |
997 | run->mmio.len, | |
998 | run->mmio.is_write); | |
d73cd8f4 | 999 | ret = 0; |
05330448 AL |
1000 | break; |
1001 | case KVM_EXIT_IRQ_WINDOW_OPEN: | |
8c0d577e | 1002 | DPRINTF("irq_window_open\n"); |
d73cd8f4 | 1003 | ret = EXCP_INTERRUPT; |
05330448 AL |
1004 | break; |
1005 | case KVM_EXIT_SHUTDOWN: | |
8c0d577e | 1006 | DPRINTF("shutdown\n"); |
05330448 | 1007 | qemu_system_reset_request(); |
d73cd8f4 | 1008 | ret = EXCP_INTERRUPT; |
05330448 AL |
1009 | break; |
1010 | case KVM_EXIT_UNKNOWN: | |
bb44e0d1 JK |
1011 | fprintf(stderr, "KVM: unknown exit, hardware reason %" PRIx64 "\n", |
1012 | (uint64_t)run->hw.hardware_exit_reason); | |
73aaec4a | 1013 | ret = -1; |
05330448 | 1014 | break; |
7c80eef8 | 1015 | case KVM_EXIT_INTERNAL_ERROR: |
73aaec4a | 1016 | ret = kvm_handle_internal_error(env, run); |
7c80eef8 | 1017 | break; |
05330448 | 1018 | default: |
8c0d577e | 1019 | DPRINTF("kvm_arch_handle_exit\n"); |
05330448 AL |
1020 | ret = kvm_arch_handle_exit(env, run); |
1021 | break; | |
1022 | } | |
d73cd8f4 | 1023 | } while (ret == 0); |
05330448 | 1024 | |
73aaec4a | 1025 | if (ret < 0) { |
f5c848ee | 1026 | cpu_dump_state(env, stderr, fprintf, CPU_DUMP_CODE); |
0461d5a6 | 1027 | vm_stop(RUN_STATE_INTERNAL_ERROR); |
becfc390 AL |
1028 | } |
1029 | ||
6792a57b JK |
1030 | env->exit_request = 0; |
1031 | cpu_single_env = NULL; | |
05330448 AL |
1032 | return ret; |
1033 | } | |
1034 | ||
984b5181 | 1035 | int kvm_ioctl(KVMState *s, int type, ...) |
05330448 AL |
1036 | { |
1037 | int ret; | |
984b5181 AL |
1038 | void *arg; |
1039 | va_list ap; | |
05330448 | 1040 | |
984b5181 AL |
1041 | va_start(ap, type); |
1042 | arg = va_arg(ap, void *); | |
1043 | va_end(ap); | |
1044 | ||
1045 | ret = ioctl(s->fd, type, arg); | |
a426e122 | 1046 | if (ret == -1) { |
05330448 | 1047 | ret = -errno; |
a426e122 | 1048 | } |
05330448 AL |
1049 | return ret; |
1050 | } | |
1051 | ||
984b5181 | 1052 | int kvm_vm_ioctl(KVMState *s, int type, ...) |
05330448 AL |
1053 | { |
1054 | int ret; | |
984b5181 AL |
1055 | void *arg; |
1056 | va_list ap; | |
1057 | ||
1058 | va_start(ap, type); | |
1059 | arg = va_arg(ap, void *); | |
1060 | va_end(ap); | |
05330448 | 1061 | |
984b5181 | 1062 | ret = ioctl(s->vmfd, type, arg); |
a426e122 | 1063 | if (ret == -1) { |
05330448 | 1064 | ret = -errno; |
a426e122 | 1065 | } |
05330448 AL |
1066 | return ret; |
1067 | } | |
1068 | ||
984b5181 | 1069 | int kvm_vcpu_ioctl(CPUState *env, int type, ...) |
05330448 AL |
1070 | { |
1071 | int ret; | |
984b5181 AL |
1072 | void *arg; |
1073 | va_list ap; | |
1074 | ||
1075 | va_start(ap, type); | |
1076 | arg = va_arg(ap, void *); | |
1077 | va_end(ap); | |
05330448 | 1078 | |
984b5181 | 1079 | ret = ioctl(env->kvm_fd, type, arg); |
a426e122 | 1080 | if (ret == -1) { |
05330448 | 1081 | ret = -errno; |
a426e122 | 1082 | } |
05330448 AL |
1083 | return ret; |
1084 | } | |
bd322087 AL |
1085 | |
1086 | int kvm_has_sync_mmu(void) | |
1087 | { | |
94a8d39a | 1088 | return kvm_check_extension(kvm_state, KVM_CAP_SYNC_MMU); |
bd322087 | 1089 | } |
e22a25c9 | 1090 | |
a0fb002c JK |
1091 | int kvm_has_vcpu_events(void) |
1092 | { | |
1093 | return kvm_state->vcpu_events; | |
1094 | } | |
1095 | ||
b0b1d690 JK |
1096 | int kvm_has_robust_singlestep(void) |
1097 | { | |
1098 | return kvm_state->robust_singlestep; | |
1099 | } | |
1100 | ||
ff44f1a3 JK |
1101 | int kvm_has_debugregs(void) |
1102 | { | |
1103 | return kvm_state->debugregs; | |
1104 | } | |
1105 | ||
f1665b21 SY |
1106 | int kvm_has_xsave(void) |
1107 | { | |
1108 | return kvm_state->xsave; | |
1109 | } | |
1110 | ||
1111 | int kvm_has_xcrs(void) | |
1112 | { | |
1113 | return kvm_state->xcrs; | |
1114 | } | |
1115 | ||
d2f2b8a7 SH |
1116 | int kvm_has_many_ioeventfds(void) |
1117 | { | |
1118 | if (!kvm_enabled()) { | |
1119 | return 0; | |
1120 | } | |
1121 | return kvm_state->many_ioeventfds; | |
1122 | } | |
1123 | ||
6f0437e8 JK |
1124 | void kvm_setup_guest_memory(void *start, size_t size) |
1125 | { | |
1126 | if (!kvm_has_sync_mmu()) { | |
e78815a5 | 1127 | int ret = qemu_madvise(start, size, QEMU_MADV_DONTFORK); |
6f0437e8 JK |
1128 | |
1129 | if (ret) { | |
e78815a5 AF |
1130 | perror("qemu_madvise"); |
1131 | fprintf(stderr, | |
1132 | "Need MADV_DONTFORK in absence of synchronous KVM MMU\n"); | |
6f0437e8 JK |
1133 | exit(1); |
1134 | } | |
6f0437e8 JK |
1135 | } |
1136 | } | |
1137 | ||
e22a25c9 AL |
1138 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
1139 | struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *env, | |
1140 | target_ulong pc) | |
1141 | { | |
1142 | struct kvm_sw_breakpoint *bp; | |
1143 | ||
72cf2d4f | 1144 | QTAILQ_FOREACH(bp, &env->kvm_state->kvm_sw_breakpoints, entry) { |
a426e122 | 1145 | if (bp->pc == pc) { |
e22a25c9 | 1146 | return bp; |
a426e122 | 1147 | } |
e22a25c9 AL |
1148 | } |
1149 | return NULL; | |
1150 | } | |
1151 | ||
1152 | int kvm_sw_breakpoints_active(CPUState *env) | |
1153 | { | |
72cf2d4f | 1154 | return !QTAILQ_EMPTY(&env->kvm_state->kvm_sw_breakpoints); |
e22a25c9 AL |
1155 | } |
1156 | ||
452e4751 GC |
1157 | struct kvm_set_guest_debug_data { |
1158 | struct kvm_guest_debug dbg; | |
1159 | CPUState *env; | |
1160 | int err; | |
1161 | }; | |
1162 | ||
1163 | static void kvm_invoke_set_guest_debug(void *data) | |
1164 | { | |
1165 | struct kvm_set_guest_debug_data *dbg_data = data; | |
b3807725 JK |
1166 | CPUState *env = dbg_data->env; |
1167 | ||
b3807725 | 1168 | dbg_data->err = kvm_vcpu_ioctl(env, KVM_SET_GUEST_DEBUG, &dbg_data->dbg); |
452e4751 GC |
1169 | } |
1170 | ||
e22a25c9 AL |
1171 | int kvm_update_guest_debug(CPUState *env, unsigned long reinject_trap) |
1172 | { | |
452e4751 | 1173 | struct kvm_set_guest_debug_data data; |
e22a25c9 | 1174 | |
b0b1d690 | 1175 | data.dbg.control = reinject_trap; |
e22a25c9 | 1176 | |
b0b1d690 JK |
1177 | if (env->singlestep_enabled) { |
1178 | data.dbg.control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP; | |
1179 | } | |
452e4751 | 1180 | kvm_arch_update_guest_debug(env, &data.dbg); |
452e4751 | 1181 | data.env = env; |
e22a25c9 | 1182 | |
be41cbe0 | 1183 | run_on_cpu(env, kvm_invoke_set_guest_debug, &data); |
452e4751 | 1184 | return data.err; |
e22a25c9 AL |
1185 | } |
1186 | ||
1187 | int kvm_insert_breakpoint(CPUState *current_env, target_ulong addr, | |
1188 | target_ulong len, int type) | |
1189 | { | |
1190 | struct kvm_sw_breakpoint *bp; | |
1191 | CPUState *env; | |
1192 | int err; | |
1193 | ||
1194 | if (type == GDB_BREAKPOINT_SW) { | |
1195 | bp = kvm_find_sw_breakpoint(current_env, addr); | |
1196 | if (bp) { | |
1197 | bp->use_count++; | |
1198 | return 0; | |
1199 | } | |
1200 | ||
7267c094 | 1201 | bp = g_malloc(sizeof(struct kvm_sw_breakpoint)); |
a426e122 | 1202 | if (!bp) { |
e22a25c9 | 1203 | return -ENOMEM; |
a426e122 | 1204 | } |
e22a25c9 AL |
1205 | |
1206 | bp->pc = addr; | |
1207 | bp->use_count = 1; | |
1208 | err = kvm_arch_insert_sw_breakpoint(current_env, bp); | |
1209 | if (err) { | |
7267c094 | 1210 | g_free(bp); |
e22a25c9 AL |
1211 | return err; |
1212 | } | |
1213 | ||
72cf2d4f | 1214 | QTAILQ_INSERT_HEAD(¤t_env->kvm_state->kvm_sw_breakpoints, |
e22a25c9 AL |
1215 | bp, entry); |
1216 | } else { | |
1217 | err = kvm_arch_insert_hw_breakpoint(addr, len, type); | |
a426e122 | 1218 | if (err) { |
e22a25c9 | 1219 | return err; |
a426e122 | 1220 | } |
e22a25c9 AL |
1221 | } |
1222 | ||
1223 | for (env = first_cpu; env != NULL; env = env->next_cpu) { | |
1224 | err = kvm_update_guest_debug(env, 0); | |
a426e122 | 1225 | if (err) { |
e22a25c9 | 1226 | return err; |
a426e122 | 1227 | } |
e22a25c9 AL |
1228 | } |
1229 | return 0; | |
1230 | } | |
1231 | ||
1232 | int kvm_remove_breakpoint(CPUState *current_env, target_ulong addr, | |
1233 | target_ulong len, int type) | |
1234 | { | |
1235 | struct kvm_sw_breakpoint *bp; | |
1236 | CPUState *env; | |
1237 | int err; | |
1238 | ||
1239 | if (type == GDB_BREAKPOINT_SW) { | |
1240 | bp = kvm_find_sw_breakpoint(current_env, addr); | |
a426e122 | 1241 | if (!bp) { |
e22a25c9 | 1242 | return -ENOENT; |
a426e122 | 1243 | } |
e22a25c9 AL |
1244 | |
1245 | if (bp->use_count > 1) { | |
1246 | bp->use_count--; | |
1247 | return 0; | |
1248 | } | |
1249 | ||
1250 | err = kvm_arch_remove_sw_breakpoint(current_env, bp); | |
a426e122 | 1251 | if (err) { |
e22a25c9 | 1252 | return err; |
a426e122 | 1253 | } |
e22a25c9 | 1254 | |
72cf2d4f | 1255 | QTAILQ_REMOVE(¤t_env->kvm_state->kvm_sw_breakpoints, bp, entry); |
7267c094 | 1256 | g_free(bp); |
e22a25c9 AL |
1257 | } else { |
1258 | err = kvm_arch_remove_hw_breakpoint(addr, len, type); | |
a426e122 | 1259 | if (err) { |
e22a25c9 | 1260 | return err; |
a426e122 | 1261 | } |
e22a25c9 AL |
1262 | } |
1263 | ||
1264 | for (env = first_cpu; env != NULL; env = env->next_cpu) { | |
1265 | err = kvm_update_guest_debug(env, 0); | |
a426e122 | 1266 | if (err) { |
e22a25c9 | 1267 | return err; |
a426e122 | 1268 | } |
e22a25c9 AL |
1269 | } |
1270 | return 0; | |
1271 | } | |
1272 | ||
1273 | void kvm_remove_all_breakpoints(CPUState *current_env) | |
1274 | { | |
1275 | struct kvm_sw_breakpoint *bp, *next; | |
1276 | KVMState *s = current_env->kvm_state; | |
1277 | CPUState *env; | |
1278 | ||
72cf2d4f | 1279 | QTAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) { |
e22a25c9 AL |
1280 | if (kvm_arch_remove_sw_breakpoint(current_env, bp) != 0) { |
1281 | /* Try harder to find a CPU that currently sees the breakpoint. */ | |
1282 | for (env = first_cpu; env != NULL; env = env->next_cpu) { | |
a426e122 | 1283 | if (kvm_arch_remove_sw_breakpoint(env, bp) == 0) { |
e22a25c9 | 1284 | break; |
a426e122 | 1285 | } |
e22a25c9 AL |
1286 | } |
1287 | } | |
1288 | } | |
1289 | kvm_arch_remove_all_hw_breakpoints(); | |
1290 | ||
a426e122 | 1291 | for (env = first_cpu; env != NULL; env = env->next_cpu) { |
e22a25c9 | 1292 | kvm_update_guest_debug(env, 0); |
a426e122 | 1293 | } |
e22a25c9 AL |
1294 | } |
1295 | ||
1296 | #else /* !KVM_CAP_SET_GUEST_DEBUG */ | |
1297 | ||
1298 | int kvm_update_guest_debug(CPUState *env, unsigned long reinject_trap) | |
1299 | { | |
1300 | return -EINVAL; | |
1301 | } | |
1302 | ||
1303 | int kvm_insert_breakpoint(CPUState *current_env, target_ulong addr, | |
1304 | target_ulong len, int type) | |
1305 | { | |
1306 | return -EINVAL; | |
1307 | } | |
1308 | ||
1309 | int kvm_remove_breakpoint(CPUState *current_env, target_ulong addr, | |
1310 | target_ulong len, int type) | |
1311 | { | |
1312 | return -EINVAL; | |
1313 | } | |
1314 | ||
1315 | void kvm_remove_all_breakpoints(CPUState *current_env) | |
1316 | { | |
1317 | } | |
1318 | #endif /* !KVM_CAP_SET_GUEST_DEBUG */ | |
cc84de95 MT |
1319 | |
1320 | int kvm_set_signal_mask(CPUState *env, const sigset_t *sigset) | |
1321 | { | |
1322 | struct kvm_signal_mask *sigmask; | |
1323 | int r; | |
1324 | ||
a426e122 | 1325 | if (!sigset) { |
cc84de95 | 1326 | return kvm_vcpu_ioctl(env, KVM_SET_SIGNAL_MASK, NULL); |
a426e122 | 1327 | } |
cc84de95 | 1328 | |
7267c094 | 1329 | sigmask = g_malloc(sizeof(*sigmask) + sizeof(*sigset)); |
cc84de95 MT |
1330 | |
1331 | sigmask->len = 8; | |
1332 | memcpy(sigmask->sigset, sigset, sizeof(*sigset)); | |
1333 | r = kvm_vcpu_ioctl(env, KVM_SET_SIGNAL_MASK, sigmask); | |
7267c094 | 1334 | g_free(sigmask); |
cc84de95 MT |
1335 | |
1336 | return r; | |
1337 | } | |
ca821806 | 1338 | |
44f1a3d8 CM |
1339 | int kvm_set_ioeventfd_mmio_long(int fd, uint32_t addr, uint32_t val, bool assign) |
1340 | { | |
44f1a3d8 CM |
1341 | int ret; |
1342 | struct kvm_ioeventfd iofd; | |
1343 | ||
1344 | iofd.datamatch = val; | |
1345 | iofd.addr = addr; | |
1346 | iofd.len = 4; | |
1347 | iofd.flags = KVM_IOEVENTFD_FLAG_DATAMATCH; | |
1348 | iofd.fd = fd; | |
1349 | ||
1350 | if (!kvm_enabled()) { | |
1351 | return -ENOSYS; | |
1352 | } | |
1353 | ||
1354 | if (!assign) { | |
1355 | iofd.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN; | |
1356 | } | |
1357 | ||
1358 | ret = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &iofd); | |
1359 | ||
1360 | if (ret < 0) { | |
1361 | return -errno; | |
1362 | } | |
1363 | ||
1364 | return 0; | |
44f1a3d8 CM |
1365 | } |
1366 | ||
ca821806 MT |
1367 | int kvm_set_ioeventfd_pio_word(int fd, uint16_t addr, uint16_t val, bool assign) |
1368 | { | |
1369 | struct kvm_ioeventfd kick = { | |
1370 | .datamatch = val, | |
1371 | .addr = addr, | |
1372 | .len = 2, | |
1373 | .flags = KVM_IOEVENTFD_FLAG_DATAMATCH | KVM_IOEVENTFD_FLAG_PIO, | |
1374 | .fd = fd, | |
1375 | }; | |
1376 | int r; | |
a426e122 | 1377 | if (!kvm_enabled()) { |
ca821806 | 1378 | return -ENOSYS; |
a426e122 JK |
1379 | } |
1380 | if (!assign) { | |
ca821806 | 1381 | kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN; |
a426e122 | 1382 | } |
ca821806 | 1383 | r = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick); |
a426e122 | 1384 | if (r < 0) { |
ca821806 | 1385 | return r; |
a426e122 | 1386 | } |
ca821806 | 1387 | return 0; |
98c8573e | 1388 | } |
a1b87fe0 JK |
1389 | |
1390 | int kvm_on_sigbus_vcpu(CPUState *env, int code, void *addr) | |
1391 | { | |
1392 | return kvm_arch_on_sigbus_vcpu(env, code, addr); | |
1393 | } | |
1394 | ||
1395 | int kvm_on_sigbus(int code, void *addr) | |
1396 | { | |
1397 | return kvm_arch_on_sigbus(code, addr); | |
1398 | } |