]>
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" |
a01672d3 | 30 | #include "memory.h" |
80a1ea37 | 31 | #include "exec-memory.h" |
05330448 | 32 | |
d2f2b8a7 SH |
33 | /* This check must be after config-host.h is included */ |
34 | #ifdef CONFIG_EVENTFD | |
35 | #include <sys/eventfd.h> | |
36 | #endif | |
37 | ||
93148aa5 | 38 | /* KVM uses PAGE_SIZE in its definition of COALESCED_MMIO_MAX */ |
f65ed4c1 AL |
39 | #define PAGE_SIZE TARGET_PAGE_SIZE |
40 | ||
05330448 AL |
41 | //#define DEBUG_KVM |
42 | ||
43 | #ifdef DEBUG_KVM | |
8c0d577e | 44 | #define DPRINTF(fmt, ...) \ |
05330448 AL |
45 | do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0) |
46 | #else | |
8c0d577e | 47 | #define DPRINTF(fmt, ...) \ |
05330448 AL |
48 | do { } while (0) |
49 | #endif | |
50 | ||
34fc643f AL |
51 | typedef struct KVMSlot |
52 | { | |
c227f099 AL |
53 | target_phys_addr_t start_addr; |
54 | ram_addr_t memory_size; | |
9f213ed9 | 55 | void *ram; |
34fc643f AL |
56 | int slot; |
57 | int flags; | |
58 | } KVMSlot; | |
05330448 | 59 | |
5832d1f2 AL |
60 | typedef struct kvm_dirty_log KVMDirtyLog; |
61 | ||
05330448 AL |
62 | struct KVMState |
63 | { | |
64 | KVMSlot slots[32]; | |
65 | int fd; | |
66 | int vmfd; | |
f65ed4c1 | 67 | int coalesced_mmio; |
62a2744c | 68 | struct kvm_coalesced_mmio_ring *coalesced_mmio_ring; |
1cae88b9 | 69 | bool coalesced_flush_in_progress; |
e69917e2 | 70 | int broken_set_mem_region; |
4495d6a7 | 71 | int migration_log; |
a0fb002c | 72 | int vcpu_events; |
b0b1d690 | 73 | int robust_singlestep; |
ff44f1a3 | 74 | int debugregs; |
e22a25c9 AL |
75 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
76 | struct kvm_sw_breakpoint_head kvm_sw_breakpoints; | |
77 | #endif | |
8a7c7393 | 78 | int pit_state2; |
f1665b21 | 79 | int xsave, xcrs; |
d2f2b8a7 | 80 | int many_ioeventfds; |
92e4b519 DG |
81 | /* The man page (and posix) say ioctl numbers are signed int, but |
82 | * they're not. Linux, glibc and *BSD all treat ioctl numbers as | |
83 | * unsigned, and treating them as signed here can break things */ | |
84 | unsigned irqchip_inject_ioctl; | |
84b058d7 JK |
85 | #ifdef KVM_CAP_IRQ_ROUTING |
86 | struct kvm_irq_routing *irq_routes; | |
87 | int nr_allocated_irq_routes; | |
88 | uint32_t *used_gsi_bitmap; | |
89 | unsigned int max_gsi; | |
90 | #endif | |
05330448 AL |
91 | }; |
92 | ||
6a7af8cb | 93 | KVMState *kvm_state; |
3d4b2649 | 94 | bool kvm_kernel_irqchip; |
05330448 | 95 | |
94a8d39a JK |
96 | static const KVMCapabilityInfo kvm_required_capabilites[] = { |
97 | KVM_CAP_INFO(USER_MEMORY), | |
98 | KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS), | |
99 | KVM_CAP_LAST_INFO | |
100 | }; | |
101 | ||
05330448 AL |
102 | static KVMSlot *kvm_alloc_slot(KVMState *s) |
103 | { | |
104 | int i; | |
105 | ||
106 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
a426e122 | 107 | if (s->slots[i].memory_size == 0) { |
05330448 | 108 | return &s->slots[i]; |
a426e122 | 109 | } |
05330448 AL |
110 | } |
111 | ||
d3f8d37f AL |
112 | fprintf(stderr, "%s: no free slot available\n", __func__); |
113 | abort(); | |
114 | } | |
115 | ||
116 | static KVMSlot *kvm_lookup_matching_slot(KVMState *s, | |
c227f099 AL |
117 | target_phys_addr_t start_addr, |
118 | target_phys_addr_t end_addr) | |
d3f8d37f AL |
119 | { |
120 | int i; | |
121 | ||
122 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
123 | KVMSlot *mem = &s->slots[i]; | |
124 | ||
125 | if (start_addr == mem->start_addr && | |
126 | end_addr == mem->start_addr + mem->memory_size) { | |
127 | return mem; | |
128 | } | |
129 | } | |
130 | ||
05330448 AL |
131 | return NULL; |
132 | } | |
133 | ||
6152e2ae AL |
134 | /* |
135 | * Find overlapping slot with lowest start address | |
136 | */ | |
137 | static KVMSlot *kvm_lookup_overlapping_slot(KVMState *s, | |
c227f099 AL |
138 | target_phys_addr_t start_addr, |
139 | target_phys_addr_t end_addr) | |
05330448 | 140 | { |
6152e2ae | 141 | KVMSlot *found = NULL; |
05330448 AL |
142 | int i; |
143 | ||
144 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
145 | KVMSlot *mem = &s->slots[i]; | |
146 | ||
6152e2ae AL |
147 | if (mem->memory_size == 0 || |
148 | (found && found->start_addr < mem->start_addr)) { | |
149 | continue; | |
150 | } | |
151 | ||
152 | if (end_addr > mem->start_addr && | |
153 | start_addr < mem->start_addr + mem->memory_size) { | |
154 | found = mem; | |
155 | } | |
05330448 AL |
156 | } |
157 | ||
6152e2ae | 158 | return found; |
05330448 AL |
159 | } |
160 | ||
9f213ed9 AK |
161 | int kvm_physical_memory_addr_from_host(KVMState *s, void *ram, |
162 | target_phys_addr_t *phys_addr) | |
983dfc3b HY |
163 | { |
164 | int i; | |
165 | ||
166 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
167 | KVMSlot *mem = &s->slots[i]; | |
168 | ||
9f213ed9 AK |
169 | if (ram >= mem->ram && ram < mem->ram + mem->memory_size) { |
170 | *phys_addr = mem->start_addr + (ram - mem->ram); | |
983dfc3b HY |
171 | return 1; |
172 | } | |
173 | } | |
174 | ||
175 | return 0; | |
176 | } | |
177 | ||
5832d1f2 AL |
178 | static int kvm_set_user_memory_region(KVMState *s, KVMSlot *slot) |
179 | { | |
180 | struct kvm_userspace_memory_region mem; | |
181 | ||
182 | mem.slot = slot->slot; | |
183 | mem.guest_phys_addr = slot->start_addr; | |
184 | mem.memory_size = slot->memory_size; | |
9f213ed9 | 185 | mem.userspace_addr = (unsigned long)slot->ram; |
5832d1f2 | 186 | mem.flags = slot->flags; |
4495d6a7 JK |
187 | if (s->migration_log) { |
188 | mem.flags |= KVM_MEM_LOG_DIRTY_PAGES; | |
189 | } | |
5832d1f2 AL |
190 | return kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem); |
191 | } | |
192 | ||
8d2ba1fb JK |
193 | static void kvm_reset_vcpu(void *opaque) |
194 | { | |
9349b4f9 | 195 | CPUArchState *env = opaque; |
8d2ba1fb | 196 | |
caa5af0f | 197 | kvm_arch_reset_vcpu(env); |
8d2ba1fb | 198 | } |
5832d1f2 | 199 | |
9349b4f9 | 200 | int kvm_init_vcpu(CPUArchState *env) |
05330448 AL |
201 | { |
202 | KVMState *s = kvm_state; | |
203 | long mmap_size; | |
204 | int ret; | |
205 | ||
8c0d577e | 206 | DPRINTF("kvm_init_vcpu\n"); |
05330448 | 207 | |
984b5181 | 208 | ret = kvm_vm_ioctl(s, KVM_CREATE_VCPU, env->cpu_index); |
05330448 | 209 | if (ret < 0) { |
8c0d577e | 210 | DPRINTF("kvm_create_vcpu failed\n"); |
05330448 AL |
211 | goto err; |
212 | } | |
213 | ||
214 | env->kvm_fd = ret; | |
215 | env->kvm_state = s; | |
d841b6c4 | 216 | env->kvm_vcpu_dirty = 1; |
05330448 AL |
217 | |
218 | mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0); | |
219 | if (mmap_size < 0) { | |
748a680b | 220 | ret = mmap_size; |
8c0d577e | 221 | DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n"); |
05330448 AL |
222 | goto err; |
223 | } | |
224 | ||
225 | env->kvm_run = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED, | |
226 | env->kvm_fd, 0); | |
227 | if (env->kvm_run == MAP_FAILED) { | |
228 | ret = -errno; | |
8c0d577e | 229 | DPRINTF("mmap'ing vcpu state failed\n"); |
05330448 AL |
230 | goto err; |
231 | } | |
232 | ||
a426e122 JK |
233 | if (s->coalesced_mmio && !s->coalesced_mmio_ring) { |
234 | s->coalesced_mmio_ring = | |
235 | (void *)env->kvm_run + s->coalesced_mmio * PAGE_SIZE; | |
236 | } | |
62a2744c | 237 | |
05330448 | 238 | ret = kvm_arch_init_vcpu(env); |
8d2ba1fb | 239 | if (ret == 0) { |
a08d4367 | 240 | qemu_register_reset(kvm_reset_vcpu, env); |
caa5af0f | 241 | kvm_arch_reset_vcpu(env); |
8d2ba1fb | 242 | } |
05330448 AL |
243 | err: |
244 | return ret; | |
245 | } | |
246 | ||
5832d1f2 AL |
247 | /* |
248 | * dirty pages logging control | |
249 | */ | |
25254bbc MT |
250 | |
251 | static int kvm_mem_flags(KVMState *s, bool log_dirty) | |
252 | { | |
253 | return log_dirty ? KVM_MEM_LOG_DIRTY_PAGES : 0; | |
254 | } | |
255 | ||
256 | static int kvm_slot_dirty_pages_log_change(KVMSlot *mem, bool log_dirty) | |
5832d1f2 AL |
257 | { |
258 | KVMState *s = kvm_state; | |
25254bbc | 259 | int flags, mask = KVM_MEM_LOG_DIRTY_PAGES; |
4495d6a7 JK |
260 | int old_flags; |
261 | ||
4495d6a7 | 262 | old_flags = mem->flags; |
5832d1f2 | 263 | |
25254bbc | 264 | flags = (mem->flags & ~mask) | kvm_mem_flags(s, log_dirty); |
5832d1f2 AL |
265 | mem->flags = flags; |
266 | ||
4495d6a7 JK |
267 | /* If nothing changed effectively, no need to issue ioctl */ |
268 | if (s->migration_log) { | |
269 | flags |= KVM_MEM_LOG_DIRTY_PAGES; | |
270 | } | |
25254bbc | 271 | |
4495d6a7 | 272 | if (flags == old_flags) { |
25254bbc | 273 | return 0; |
4495d6a7 JK |
274 | } |
275 | ||
5832d1f2 AL |
276 | return kvm_set_user_memory_region(s, mem); |
277 | } | |
278 | ||
25254bbc MT |
279 | static int kvm_dirty_pages_log_change(target_phys_addr_t phys_addr, |
280 | ram_addr_t size, bool log_dirty) | |
281 | { | |
282 | KVMState *s = kvm_state; | |
283 | KVMSlot *mem = kvm_lookup_matching_slot(s, phys_addr, phys_addr + size); | |
284 | ||
285 | if (mem == NULL) { | |
286 | fprintf(stderr, "BUG: %s: invalid parameters " TARGET_FMT_plx "-" | |
287 | TARGET_FMT_plx "\n", __func__, phys_addr, | |
288 | (target_phys_addr_t)(phys_addr + size - 1)); | |
289 | return -EINVAL; | |
290 | } | |
291 | return kvm_slot_dirty_pages_log_change(mem, log_dirty); | |
292 | } | |
293 | ||
a01672d3 AK |
294 | static void kvm_log_start(MemoryListener *listener, |
295 | MemoryRegionSection *section) | |
5832d1f2 | 296 | { |
a01672d3 AK |
297 | int r; |
298 | ||
299 | r = kvm_dirty_pages_log_change(section->offset_within_address_space, | |
300 | section->size, true); | |
301 | if (r < 0) { | |
302 | abort(); | |
303 | } | |
5832d1f2 AL |
304 | } |
305 | ||
a01672d3 AK |
306 | static void kvm_log_stop(MemoryListener *listener, |
307 | MemoryRegionSection *section) | |
5832d1f2 | 308 | { |
a01672d3 AK |
309 | int r; |
310 | ||
311 | r = kvm_dirty_pages_log_change(section->offset_within_address_space, | |
312 | section->size, false); | |
313 | if (r < 0) { | |
314 | abort(); | |
315 | } | |
5832d1f2 AL |
316 | } |
317 | ||
7b8f3b78 | 318 | static int kvm_set_migration_log(int enable) |
4495d6a7 JK |
319 | { |
320 | KVMState *s = kvm_state; | |
321 | KVMSlot *mem; | |
322 | int i, err; | |
323 | ||
324 | s->migration_log = enable; | |
325 | ||
326 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
327 | mem = &s->slots[i]; | |
328 | ||
70fedd76 AW |
329 | if (!mem->memory_size) { |
330 | continue; | |
331 | } | |
4495d6a7 JK |
332 | if (!!(mem->flags & KVM_MEM_LOG_DIRTY_PAGES) == enable) { |
333 | continue; | |
334 | } | |
335 | err = kvm_set_user_memory_region(s, mem); | |
336 | if (err) { | |
337 | return err; | |
338 | } | |
339 | } | |
340 | return 0; | |
341 | } | |
342 | ||
8369e01c | 343 | /* get kvm's dirty pages bitmap and update qemu's */ |
ffcde12f AK |
344 | static int kvm_get_dirty_pages_log_range(MemoryRegionSection *section, |
345 | unsigned long *bitmap) | |
96c1606b | 346 | { |
8369e01c | 347 | unsigned int i, j; |
aa90fec7 BH |
348 | unsigned long page_number, c; |
349 | target_phys_addr_t addr, addr1; | |
ffcde12f | 350 | unsigned int len = ((section->size / TARGET_PAGE_SIZE) + HOST_LONG_BITS - 1) / HOST_LONG_BITS; |
3145fcb6 | 351 | unsigned long hpratio = getpagesize() / TARGET_PAGE_SIZE; |
8369e01c MT |
352 | |
353 | /* | |
354 | * bitmap-traveling is faster than memory-traveling (for addr...) | |
355 | * especially when most of the memory is not dirty. | |
356 | */ | |
357 | for (i = 0; i < len; i++) { | |
358 | if (bitmap[i] != 0) { | |
359 | c = leul_to_cpu(bitmap[i]); | |
360 | do { | |
361 | j = ffsl(c) - 1; | |
362 | c &= ~(1ul << j); | |
3145fcb6 | 363 | page_number = (i * HOST_LONG_BITS + j) * hpratio; |
8369e01c | 364 | addr1 = page_number * TARGET_PAGE_SIZE; |
ffcde12f | 365 | addr = section->offset_within_region + addr1; |
3145fcb6 DG |
366 | memory_region_set_dirty(section->mr, addr, |
367 | TARGET_PAGE_SIZE * hpratio); | |
8369e01c MT |
368 | } while (c != 0); |
369 | } | |
370 | } | |
371 | return 0; | |
96c1606b AG |
372 | } |
373 | ||
8369e01c MT |
374 | #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1)) |
375 | ||
5832d1f2 AL |
376 | /** |
377 | * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space | |
fd4aa979 BS |
378 | * This function updates qemu's dirty bitmap using |
379 | * memory_region_set_dirty(). This means all bits are set | |
380 | * to dirty. | |
5832d1f2 | 381 | * |
d3f8d37f | 382 | * @start_add: start of logged region. |
5832d1f2 AL |
383 | * @end_addr: end of logged region. |
384 | */ | |
ffcde12f | 385 | static int kvm_physical_sync_dirty_bitmap(MemoryRegionSection *section) |
5832d1f2 AL |
386 | { |
387 | KVMState *s = kvm_state; | |
151f7749 | 388 | unsigned long size, allocated_size = 0; |
151f7749 JK |
389 | KVMDirtyLog d; |
390 | KVMSlot *mem; | |
391 | int ret = 0; | |
ffcde12f AK |
392 | target_phys_addr_t start_addr = section->offset_within_address_space; |
393 | target_phys_addr_t end_addr = start_addr + section->size; | |
5832d1f2 | 394 | |
151f7749 JK |
395 | d.dirty_bitmap = NULL; |
396 | while (start_addr < end_addr) { | |
397 | mem = kvm_lookup_overlapping_slot(s, start_addr, end_addr); | |
398 | if (mem == NULL) { | |
399 | break; | |
400 | } | |
5832d1f2 | 401 | |
51b0c606 MT |
402 | /* XXX bad kernel interface alert |
403 | * For dirty bitmap, kernel allocates array of size aligned to | |
404 | * bits-per-long. But for case when the kernel is 64bits and | |
405 | * the userspace is 32bits, userspace can't align to the same | |
406 | * bits-per-long, since sizeof(long) is different between kernel | |
407 | * and user space. This way, userspace will provide buffer which | |
408 | * may be 4 bytes less than the kernel will use, resulting in | |
409 | * userspace memory corruption (which is not detectable by valgrind | |
410 | * too, in most cases). | |
411 | * So for now, let's align to 64 instead of HOST_LONG_BITS here, in | |
412 | * a hope that sizeof(long) wont become >8 any time soon. | |
413 | */ | |
414 | size = ALIGN(((mem->memory_size) >> TARGET_PAGE_BITS), | |
415 | /*HOST_LONG_BITS*/ 64) / 8; | |
151f7749 | 416 | if (!d.dirty_bitmap) { |
7267c094 | 417 | d.dirty_bitmap = g_malloc(size); |
151f7749 | 418 | } else if (size > allocated_size) { |
7267c094 | 419 | d.dirty_bitmap = g_realloc(d.dirty_bitmap, size); |
151f7749 JK |
420 | } |
421 | allocated_size = size; | |
422 | memset(d.dirty_bitmap, 0, allocated_size); | |
5832d1f2 | 423 | |
151f7749 | 424 | d.slot = mem->slot; |
5832d1f2 | 425 | |
6e489f3f | 426 | if (kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d) == -1) { |
8c0d577e | 427 | DPRINTF("ioctl failed %d\n", errno); |
151f7749 JK |
428 | ret = -1; |
429 | break; | |
430 | } | |
5832d1f2 | 431 | |
ffcde12f | 432 | kvm_get_dirty_pages_log_range(section, d.dirty_bitmap); |
8369e01c | 433 | start_addr = mem->start_addr + mem->memory_size; |
5832d1f2 | 434 | } |
7267c094 | 435 | g_free(d.dirty_bitmap); |
151f7749 JK |
436 | |
437 | return ret; | |
5832d1f2 AL |
438 | } |
439 | ||
c227f099 | 440 | int kvm_coalesce_mmio_region(target_phys_addr_t start, ram_addr_t size) |
f65ed4c1 AL |
441 | { |
442 | int ret = -ENOSYS; | |
f65ed4c1 AL |
443 | KVMState *s = kvm_state; |
444 | ||
445 | if (s->coalesced_mmio) { | |
446 | struct kvm_coalesced_mmio_zone zone; | |
447 | ||
448 | zone.addr = start; | |
449 | zone.size = size; | |
7e680753 | 450 | zone.pad = 0; |
f65ed4c1 AL |
451 | |
452 | ret = kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone); | |
453 | } | |
f65ed4c1 AL |
454 | |
455 | return ret; | |
456 | } | |
457 | ||
c227f099 | 458 | int kvm_uncoalesce_mmio_region(target_phys_addr_t start, ram_addr_t size) |
f65ed4c1 AL |
459 | { |
460 | int ret = -ENOSYS; | |
f65ed4c1 AL |
461 | KVMState *s = kvm_state; |
462 | ||
463 | if (s->coalesced_mmio) { | |
464 | struct kvm_coalesced_mmio_zone zone; | |
465 | ||
466 | zone.addr = start; | |
467 | zone.size = size; | |
7e680753 | 468 | zone.pad = 0; |
f65ed4c1 AL |
469 | |
470 | ret = kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone); | |
471 | } | |
f65ed4c1 AL |
472 | |
473 | return ret; | |
474 | } | |
475 | ||
ad7b8b33 AL |
476 | int kvm_check_extension(KVMState *s, unsigned int extension) |
477 | { | |
478 | int ret; | |
479 | ||
480 | ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension); | |
481 | if (ret < 0) { | |
482 | ret = 0; | |
483 | } | |
484 | ||
485 | return ret; | |
486 | } | |
487 | ||
d2f2b8a7 SH |
488 | static int kvm_check_many_ioeventfds(void) |
489 | { | |
d0dcac83 SH |
490 | /* Userspace can use ioeventfd for io notification. This requires a host |
491 | * that supports eventfd(2) and an I/O thread; since eventfd does not | |
492 | * support SIGIO it cannot interrupt the vcpu. | |
493 | * | |
494 | * Older kernels have a 6 device limit on the KVM io bus. Find out so we | |
d2f2b8a7 SH |
495 | * can avoid creating too many ioeventfds. |
496 | */ | |
12d4536f | 497 | #if defined(CONFIG_EVENTFD) |
d2f2b8a7 SH |
498 | int ioeventfds[7]; |
499 | int i, ret = 0; | |
500 | for (i = 0; i < ARRAY_SIZE(ioeventfds); i++) { | |
501 | ioeventfds[i] = eventfd(0, EFD_CLOEXEC); | |
502 | if (ioeventfds[i] < 0) { | |
503 | break; | |
504 | } | |
505 | ret = kvm_set_ioeventfd_pio_word(ioeventfds[i], 0, i, true); | |
506 | if (ret < 0) { | |
507 | close(ioeventfds[i]); | |
508 | break; | |
509 | } | |
510 | } | |
511 | ||
512 | /* Decide whether many devices are supported or not */ | |
513 | ret = i == ARRAY_SIZE(ioeventfds); | |
514 | ||
515 | while (i-- > 0) { | |
516 | kvm_set_ioeventfd_pio_word(ioeventfds[i], 0, i, false); | |
517 | close(ioeventfds[i]); | |
518 | } | |
519 | return ret; | |
520 | #else | |
521 | return 0; | |
522 | #endif | |
523 | } | |
524 | ||
94a8d39a JK |
525 | static const KVMCapabilityInfo * |
526 | kvm_check_extension_list(KVMState *s, const KVMCapabilityInfo *list) | |
527 | { | |
528 | while (list->name) { | |
529 | if (!kvm_check_extension(s, list->value)) { | |
530 | return list; | |
531 | } | |
532 | list++; | |
533 | } | |
534 | return NULL; | |
535 | } | |
536 | ||
a01672d3 | 537 | static void kvm_set_phys_mem(MemoryRegionSection *section, bool add) |
46dbef6a MT |
538 | { |
539 | KVMState *s = kvm_state; | |
46dbef6a MT |
540 | KVMSlot *mem, old; |
541 | int err; | |
a01672d3 AK |
542 | MemoryRegion *mr = section->mr; |
543 | bool log_dirty = memory_region_is_logging(mr); | |
544 | target_phys_addr_t start_addr = section->offset_within_address_space; | |
545 | ram_addr_t size = section->size; | |
9f213ed9 | 546 | void *ram = NULL; |
8f6f962b | 547 | unsigned delta; |
46dbef6a | 548 | |
14542fea GN |
549 | /* kvm works in page size chunks, but the function may be called |
550 | with sub-page size and unaligned start address. */ | |
8f6f962b AK |
551 | delta = TARGET_PAGE_ALIGN(size) - size; |
552 | if (delta > size) { | |
553 | return; | |
554 | } | |
555 | start_addr += delta; | |
556 | size -= delta; | |
557 | size &= TARGET_PAGE_MASK; | |
558 | if (!size || (start_addr & ~TARGET_PAGE_MASK)) { | |
559 | return; | |
560 | } | |
46dbef6a | 561 | |
a01672d3 AK |
562 | if (!memory_region_is_ram(mr)) { |
563 | return; | |
9f213ed9 AK |
564 | } |
565 | ||
8f6f962b | 566 | ram = memory_region_get_ram_ptr(mr) + section->offset_within_region + delta; |
a01672d3 | 567 | |
46dbef6a MT |
568 | while (1) { |
569 | mem = kvm_lookup_overlapping_slot(s, start_addr, start_addr + size); | |
570 | if (!mem) { | |
571 | break; | |
572 | } | |
573 | ||
a01672d3 | 574 | if (add && start_addr >= mem->start_addr && |
46dbef6a | 575 | (start_addr + size <= mem->start_addr + mem->memory_size) && |
9f213ed9 | 576 | (ram - start_addr == mem->ram - mem->start_addr)) { |
46dbef6a | 577 | /* The new slot fits into the existing one and comes with |
25254bbc MT |
578 | * identical parameters - update flags and done. */ |
579 | kvm_slot_dirty_pages_log_change(mem, log_dirty); | |
46dbef6a MT |
580 | return; |
581 | } | |
582 | ||
583 | old = *mem; | |
584 | ||
3fbffb62 AK |
585 | if (mem->flags & KVM_MEM_LOG_DIRTY_PAGES) { |
586 | kvm_physical_sync_dirty_bitmap(section); | |
587 | } | |
588 | ||
46dbef6a MT |
589 | /* unregister the overlapping slot */ |
590 | mem->memory_size = 0; | |
591 | err = kvm_set_user_memory_region(s, mem); | |
592 | if (err) { | |
593 | fprintf(stderr, "%s: error unregistering overlapping slot: %s\n", | |
594 | __func__, strerror(-err)); | |
595 | abort(); | |
596 | } | |
597 | ||
598 | /* Workaround for older KVM versions: we can't join slots, even not by | |
599 | * unregistering the previous ones and then registering the larger | |
600 | * slot. We have to maintain the existing fragmentation. Sigh. | |
601 | * | |
602 | * This workaround assumes that the new slot starts at the same | |
603 | * address as the first existing one. If not or if some overlapping | |
604 | * slot comes around later, we will fail (not seen in practice so far) | |
605 | * - and actually require a recent KVM version. */ | |
606 | if (s->broken_set_mem_region && | |
a01672d3 | 607 | old.start_addr == start_addr && old.memory_size < size && add) { |
46dbef6a MT |
608 | mem = kvm_alloc_slot(s); |
609 | mem->memory_size = old.memory_size; | |
610 | mem->start_addr = old.start_addr; | |
9f213ed9 | 611 | mem->ram = old.ram; |
25254bbc | 612 | mem->flags = kvm_mem_flags(s, log_dirty); |
46dbef6a MT |
613 | |
614 | err = kvm_set_user_memory_region(s, mem); | |
615 | if (err) { | |
616 | fprintf(stderr, "%s: error updating slot: %s\n", __func__, | |
617 | strerror(-err)); | |
618 | abort(); | |
619 | } | |
620 | ||
621 | start_addr += old.memory_size; | |
9f213ed9 | 622 | ram += old.memory_size; |
46dbef6a MT |
623 | size -= old.memory_size; |
624 | continue; | |
625 | } | |
626 | ||
627 | /* register prefix slot */ | |
628 | if (old.start_addr < start_addr) { | |
629 | mem = kvm_alloc_slot(s); | |
630 | mem->memory_size = start_addr - old.start_addr; | |
631 | mem->start_addr = old.start_addr; | |
9f213ed9 | 632 | mem->ram = old.ram; |
25254bbc | 633 | mem->flags = kvm_mem_flags(s, log_dirty); |
46dbef6a MT |
634 | |
635 | err = kvm_set_user_memory_region(s, mem); | |
636 | if (err) { | |
637 | fprintf(stderr, "%s: error registering prefix slot: %s\n", | |
638 | __func__, strerror(-err)); | |
d4d6868f AG |
639 | #ifdef TARGET_PPC |
640 | fprintf(stderr, "%s: This is probably because your kernel's " \ | |
641 | "PAGE_SIZE is too big. Please try to use 4k " \ | |
642 | "PAGE_SIZE!\n", __func__); | |
643 | #endif | |
46dbef6a MT |
644 | abort(); |
645 | } | |
646 | } | |
647 | ||
648 | /* register suffix slot */ | |
649 | if (old.start_addr + old.memory_size > start_addr + size) { | |
650 | ram_addr_t size_delta; | |
651 | ||
652 | mem = kvm_alloc_slot(s); | |
653 | mem->start_addr = start_addr + size; | |
654 | size_delta = mem->start_addr - old.start_addr; | |
655 | mem->memory_size = old.memory_size - size_delta; | |
9f213ed9 | 656 | mem->ram = old.ram + size_delta; |
25254bbc | 657 | mem->flags = kvm_mem_flags(s, log_dirty); |
46dbef6a MT |
658 | |
659 | err = kvm_set_user_memory_region(s, mem); | |
660 | if (err) { | |
661 | fprintf(stderr, "%s: error registering suffix slot: %s\n", | |
662 | __func__, strerror(-err)); | |
663 | abort(); | |
664 | } | |
665 | } | |
666 | } | |
667 | ||
668 | /* in case the KVM bug workaround already "consumed" the new slot */ | |
a426e122 | 669 | if (!size) { |
46dbef6a | 670 | return; |
a426e122 | 671 | } |
a01672d3 | 672 | if (!add) { |
46dbef6a | 673 | return; |
a426e122 | 674 | } |
46dbef6a MT |
675 | mem = kvm_alloc_slot(s); |
676 | mem->memory_size = size; | |
677 | mem->start_addr = start_addr; | |
9f213ed9 | 678 | mem->ram = ram; |
25254bbc | 679 | mem->flags = kvm_mem_flags(s, log_dirty); |
46dbef6a MT |
680 | |
681 | err = kvm_set_user_memory_region(s, mem); | |
682 | if (err) { | |
683 | fprintf(stderr, "%s: error registering slot: %s\n", __func__, | |
684 | strerror(-err)); | |
685 | abort(); | |
686 | } | |
687 | } | |
688 | ||
50c1e149 AK |
689 | static void kvm_begin(MemoryListener *listener) |
690 | { | |
691 | } | |
692 | ||
693 | static void kvm_commit(MemoryListener *listener) | |
694 | { | |
695 | } | |
696 | ||
a01672d3 AK |
697 | static void kvm_region_add(MemoryListener *listener, |
698 | MemoryRegionSection *section) | |
699 | { | |
700 | kvm_set_phys_mem(section, true); | |
701 | } | |
702 | ||
703 | static void kvm_region_del(MemoryListener *listener, | |
704 | MemoryRegionSection *section) | |
705 | { | |
706 | kvm_set_phys_mem(section, false); | |
707 | } | |
708 | ||
50c1e149 AK |
709 | static void kvm_region_nop(MemoryListener *listener, |
710 | MemoryRegionSection *section) | |
711 | { | |
712 | } | |
713 | ||
a01672d3 AK |
714 | static void kvm_log_sync(MemoryListener *listener, |
715 | MemoryRegionSection *section) | |
7b8f3b78 | 716 | { |
a01672d3 AK |
717 | int r; |
718 | ||
ffcde12f | 719 | r = kvm_physical_sync_dirty_bitmap(section); |
a01672d3 AK |
720 | if (r < 0) { |
721 | abort(); | |
722 | } | |
7b8f3b78 MT |
723 | } |
724 | ||
a01672d3 | 725 | static void kvm_log_global_start(struct MemoryListener *listener) |
7b8f3b78 | 726 | { |
a01672d3 AK |
727 | int r; |
728 | ||
729 | r = kvm_set_migration_log(1); | |
730 | assert(r >= 0); | |
7b8f3b78 MT |
731 | } |
732 | ||
a01672d3 | 733 | static void kvm_log_global_stop(struct MemoryListener *listener) |
7b8f3b78 | 734 | { |
a01672d3 AK |
735 | int r; |
736 | ||
737 | r = kvm_set_migration_log(0); | |
738 | assert(r >= 0); | |
7b8f3b78 MT |
739 | } |
740 | ||
80a1ea37 AK |
741 | static void kvm_mem_ioeventfd_add(MemoryRegionSection *section, |
742 | bool match_data, uint64_t data, int fd) | |
743 | { | |
744 | int r; | |
745 | ||
4b8f1c88 | 746 | assert(match_data && section->size <= 8); |
80a1ea37 | 747 | |
4b8f1c88 MT |
748 | r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space, |
749 | data, true, section->size); | |
80a1ea37 AK |
750 | if (r < 0) { |
751 | abort(); | |
752 | } | |
753 | } | |
754 | ||
755 | static void kvm_mem_ioeventfd_del(MemoryRegionSection *section, | |
756 | bool match_data, uint64_t data, int fd) | |
757 | { | |
758 | int r; | |
759 | ||
4b8f1c88 MT |
760 | r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space, |
761 | data, false, section->size); | |
80a1ea37 AK |
762 | if (r < 0) { |
763 | abort(); | |
764 | } | |
765 | } | |
766 | ||
767 | static void kvm_io_ioeventfd_add(MemoryRegionSection *section, | |
768 | bool match_data, uint64_t data, int fd) | |
769 | { | |
770 | int r; | |
771 | ||
772 | assert(match_data && section->size == 2); | |
773 | ||
774 | r = kvm_set_ioeventfd_pio_word(fd, section->offset_within_address_space, | |
775 | data, true); | |
776 | if (r < 0) { | |
777 | abort(); | |
778 | } | |
779 | } | |
780 | ||
781 | static void kvm_io_ioeventfd_del(MemoryRegionSection *section, | |
782 | bool match_data, uint64_t data, int fd) | |
783 | ||
784 | { | |
785 | int r; | |
786 | ||
787 | r = kvm_set_ioeventfd_pio_word(fd, section->offset_within_address_space, | |
788 | data, false); | |
789 | if (r < 0) { | |
790 | abort(); | |
791 | } | |
792 | } | |
793 | ||
794 | static void kvm_eventfd_add(MemoryListener *listener, | |
795 | MemoryRegionSection *section, | |
796 | bool match_data, uint64_t data, int fd) | |
797 | { | |
798 | if (section->address_space == get_system_memory()) { | |
799 | kvm_mem_ioeventfd_add(section, match_data, data, fd); | |
800 | } else { | |
801 | kvm_io_ioeventfd_add(section, match_data, data, fd); | |
802 | } | |
803 | } | |
804 | ||
805 | static void kvm_eventfd_del(MemoryListener *listener, | |
806 | MemoryRegionSection *section, | |
807 | bool match_data, uint64_t data, int fd) | |
808 | { | |
809 | if (section->address_space == get_system_memory()) { | |
810 | kvm_mem_ioeventfd_del(section, match_data, data, fd); | |
811 | } else { | |
812 | kvm_io_ioeventfd_del(section, match_data, data, fd); | |
813 | } | |
814 | } | |
815 | ||
a01672d3 | 816 | static MemoryListener kvm_memory_listener = { |
50c1e149 AK |
817 | .begin = kvm_begin, |
818 | .commit = kvm_commit, | |
a01672d3 AK |
819 | .region_add = kvm_region_add, |
820 | .region_del = kvm_region_del, | |
50c1e149 | 821 | .region_nop = kvm_region_nop, |
e5896b12 AP |
822 | .log_start = kvm_log_start, |
823 | .log_stop = kvm_log_stop, | |
a01672d3 AK |
824 | .log_sync = kvm_log_sync, |
825 | .log_global_start = kvm_log_global_start, | |
826 | .log_global_stop = kvm_log_global_stop, | |
80a1ea37 AK |
827 | .eventfd_add = kvm_eventfd_add, |
828 | .eventfd_del = kvm_eventfd_del, | |
72e22d2f | 829 | .priority = 10, |
7b8f3b78 MT |
830 | }; |
831 | ||
9349b4f9 | 832 | static void kvm_handle_interrupt(CPUArchState *env, int mask) |
aa7f74d1 JK |
833 | { |
834 | env->interrupt_request |= mask; | |
835 | ||
836 | if (!qemu_cpu_is_self(env)) { | |
837 | qemu_cpu_kick(env); | |
838 | } | |
839 | } | |
840 | ||
84b058d7 JK |
841 | int kvm_irqchip_set_irq(KVMState *s, int irq, int level) |
842 | { | |
843 | struct kvm_irq_level event; | |
844 | int ret; | |
845 | ||
3d4b2649 | 846 | assert(kvm_irqchip_in_kernel()); |
84b058d7 JK |
847 | |
848 | event.level = level; | |
849 | event.irq = irq; | |
850 | ret = kvm_vm_ioctl(s, s->irqchip_inject_ioctl, &event); | |
851 | if (ret < 0) { | |
852 | perror("kvm_set_irqchip_line"); | |
853 | abort(); | |
854 | } | |
855 | ||
856 | return (s->irqchip_inject_ioctl == KVM_IRQ_LINE) ? 1 : event.status; | |
857 | } | |
858 | ||
859 | #ifdef KVM_CAP_IRQ_ROUTING | |
860 | static void set_gsi(KVMState *s, unsigned int gsi) | |
861 | { | |
862 | assert(gsi < s->max_gsi); | |
863 | ||
864 | s->used_gsi_bitmap[gsi / 32] |= 1U << (gsi % 32); | |
865 | } | |
866 | ||
867 | static void kvm_init_irq_routing(KVMState *s) | |
868 | { | |
869 | int gsi_count; | |
870 | ||
871 | gsi_count = kvm_check_extension(s, KVM_CAP_IRQ_ROUTING); | |
872 | if (gsi_count > 0) { | |
873 | unsigned int gsi_bits, i; | |
874 | ||
875 | /* Round up so we can search ints using ffs */ | |
bc8c6788 | 876 | gsi_bits = ALIGN(gsi_count, 32); |
84b058d7 JK |
877 | s->used_gsi_bitmap = g_malloc0(gsi_bits / 8); |
878 | s->max_gsi = gsi_bits; | |
879 | ||
880 | /* Mark any over-allocated bits as already in use */ | |
881 | for (i = gsi_count; i < gsi_bits; i++) { | |
882 | set_gsi(s, i); | |
883 | } | |
884 | } | |
885 | ||
886 | s->irq_routes = g_malloc0(sizeof(*s->irq_routes)); | |
887 | s->nr_allocated_irq_routes = 0; | |
888 | ||
889 | kvm_arch_init_irq_routing(s); | |
890 | } | |
891 | ||
892 | static void kvm_add_routing_entry(KVMState *s, | |
893 | struct kvm_irq_routing_entry *entry) | |
894 | { | |
895 | struct kvm_irq_routing_entry *new; | |
896 | int n, size; | |
897 | ||
898 | if (s->irq_routes->nr == s->nr_allocated_irq_routes) { | |
899 | n = s->nr_allocated_irq_routes * 2; | |
900 | if (n < 64) { | |
901 | n = 64; | |
902 | } | |
903 | size = sizeof(struct kvm_irq_routing); | |
904 | size += n * sizeof(*new); | |
905 | s->irq_routes = g_realloc(s->irq_routes, size); | |
906 | s->nr_allocated_irq_routes = n; | |
907 | } | |
908 | n = s->irq_routes->nr++; | |
909 | new = &s->irq_routes->entries[n]; | |
910 | memset(new, 0, sizeof(*new)); | |
911 | new->gsi = entry->gsi; | |
912 | new->type = entry->type; | |
913 | new->flags = entry->flags; | |
914 | new->u = entry->u; | |
915 | ||
916 | set_gsi(s, entry->gsi); | |
917 | } | |
918 | ||
919 | void kvm_irqchip_add_route(KVMState *s, int irq, int irqchip, int pin) | |
920 | { | |
921 | struct kvm_irq_routing_entry e; | |
922 | ||
923 | e.gsi = irq; | |
924 | e.type = KVM_IRQ_ROUTING_IRQCHIP; | |
925 | e.flags = 0; | |
926 | e.u.irqchip.irqchip = irqchip; | |
927 | e.u.irqchip.pin = pin; | |
928 | kvm_add_routing_entry(s, &e); | |
929 | } | |
930 | ||
931 | int kvm_irqchip_commit_routes(KVMState *s) | |
932 | { | |
933 | s->irq_routes->flags = 0; | |
934 | return kvm_vm_ioctl(s, KVM_SET_GSI_ROUTING, s->irq_routes); | |
935 | } | |
936 | ||
937 | #else /* !KVM_CAP_IRQ_ROUTING */ | |
938 | ||
939 | static void kvm_init_irq_routing(KVMState *s) | |
940 | { | |
941 | } | |
942 | #endif /* !KVM_CAP_IRQ_ROUTING */ | |
943 | ||
944 | static int kvm_irqchip_create(KVMState *s) | |
945 | { | |
946 | QemuOptsList *list = qemu_find_opts("machine"); | |
947 | int ret; | |
948 | ||
949 | if (QTAILQ_EMPTY(&list->head) || | |
950 | !qemu_opt_get_bool(QTAILQ_FIRST(&list->head), | |
951 | "kernel_irqchip", false) || | |
952 | !kvm_check_extension(s, KVM_CAP_IRQCHIP)) { | |
953 | return 0; | |
954 | } | |
955 | ||
956 | ret = kvm_vm_ioctl(s, KVM_CREATE_IRQCHIP); | |
957 | if (ret < 0) { | |
958 | fprintf(stderr, "Create kernel irqchip failed\n"); | |
959 | return ret; | |
960 | } | |
961 | ||
962 | s->irqchip_inject_ioctl = KVM_IRQ_LINE; | |
963 | if (kvm_check_extension(s, KVM_CAP_IRQ_INJECT_STATUS)) { | |
964 | s->irqchip_inject_ioctl = KVM_IRQ_LINE_STATUS; | |
965 | } | |
3d4b2649 | 966 | kvm_kernel_irqchip = true; |
84b058d7 JK |
967 | |
968 | kvm_init_irq_routing(s); | |
969 | ||
970 | return 0; | |
971 | } | |
972 | ||
cad1e282 | 973 | int kvm_init(void) |
05330448 | 974 | { |
168ccc11 JK |
975 | static const char upgrade_note[] = |
976 | "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n" | |
977 | "(see http://sourceforge.net/projects/kvm).\n"; | |
05330448 | 978 | KVMState *s; |
94a8d39a | 979 | const KVMCapabilityInfo *missing_cap; |
05330448 AL |
980 | int ret; |
981 | int i; | |
982 | ||
7267c094 | 983 | s = g_malloc0(sizeof(KVMState)); |
05330448 | 984 | |
3145fcb6 DG |
985 | /* |
986 | * On systems where the kernel can support different base page | |
987 | * sizes, host page size may be different from TARGET_PAGE_SIZE, | |
988 | * even with KVM. TARGET_PAGE_SIZE is assumed to be the minimum | |
989 | * page size for the system though. | |
990 | */ | |
991 | assert(TARGET_PAGE_SIZE <= getpagesize()); | |
992 | ||
e22a25c9 | 993 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
72cf2d4f | 994 | QTAILQ_INIT(&s->kvm_sw_breakpoints); |
e22a25c9 | 995 | #endif |
a426e122 | 996 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { |
05330448 | 997 | s->slots[i].slot = i; |
a426e122 | 998 | } |
05330448 | 999 | s->vmfd = -1; |
40ff6d7e | 1000 | s->fd = qemu_open("/dev/kvm", O_RDWR); |
05330448 AL |
1001 | if (s->fd == -1) { |
1002 | fprintf(stderr, "Could not access KVM kernel module: %m\n"); | |
1003 | ret = -errno; | |
1004 | goto err; | |
1005 | } | |
1006 | ||
1007 | ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0); | |
1008 | if (ret < KVM_API_VERSION) { | |
a426e122 | 1009 | if (ret > 0) { |
05330448 | 1010 | ret = -EINVAL; |
a426e122 | 1011 | } |
05330448 AL |
1012 | fprintf(stderr, "kvm version too old\n"); |
1013 | goto err; | |
1014 | } | |
1015 | ||
1016 | if (ret > KVM_API_VERSION) { | |
1017 | ret = -EINVAL; | |
1018 | fprintf(stderr, "kvm version not supported\n"); | |
1019 | goto err; | |
1020 | } | |
1021 | ||
1022 | s->vmfd = kvm_ioctl(s, KVM_CREATE_VM, 0); | |
0104dcac AG |
1023 | if (s->vmfd < 0) { |
1024 | #ifdef TARGET_S390X | |
1025 | fprintf(stderr, "Please add the 'switch_amode' kernel parameter to " | |
1026 | "your host kernel command line\n"); | |
1027 | #endif | |
db9eae1c | 1028 | ret = s->vmfd; |
05330448 | 1029 | goto err; |
0104dcac | 1030 | } |
05330448 | 1031 | |
94a8d39a JK |
1032 | missing_cap = kvm_check_extension_list(s, kvm_required_capabilites); |
1033 | if (!missing_cap) { | |
1034 | missing_cap = | |
1035 | kvm_check_extension_list(s, kvm_arch_required_capabilities); | |
05330448 | 1036 | } |
94a8d39a | 1037 | if (missing_cap) { |
ad7b8b33 | 1038 | ret = -EINVAL; |
94a8d39a JK |
1039 | fprintf(stderr, "kvm does not support %s\n%s", |
1040 | missing_cap->name, upgrade_note); | |
d85dc283 AL |
1041 | goto err; |
1042 | } | |
1043 | ||
ad7b8b33 | 1044 | s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO); |
f65ed4c1 | 1045 | |
e69917e2 | 1046 | s->broken_set_mem_region = 1; |
14a09518 | 1047 | ret = kvm_check_extension(s, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS); |
e69917e2 JK |
1048 | if (ret > 0) { |
1049 | s->broken_set_mem_region = 0; | |
1050 | } | |
e69917e2 | 1051 | |
a0fb002c JK |
1052 | #ifdef KVM_CAP_VCPU_EVENTS |
1053 | s->vcpu_events = kvm_check_extension(s, KVM_CAP_VCPU_EVENTS); | |
1054 | #endif | |
1055 | ||
b0b1d690 JK |
1056 | s->robust_singlestep = |
1057 | kvm_check_extension(s, KVM_CAP_X86_ROBUST_SINGLESTEP); | |
b0b1d690 | 1058 | |
ff44f1a3 JK |
1059 | #ifdef KVM_CAP_DEBUGREGS |
1060 | s->debugregs = kvm_check_extension(s, KVM_CAP_DEBUGREGS); | |
1061 | #endif | |
1062 | ||
f1665b21 SY |
1063 | #ifdef KVM_CAP_XSAVE |
1064 | s->xsave = kvm_check_extension(s, KVM_CAP_XSAVE); | |
1065 | #endif | |
1066 | ||
f1665b21 SY |
1067 | #ifdef KVM_CAP_XCRS |
1068 | s->xcrs = kvm_check_extension(s, KVM_CAP_XCRS); | |
1069 | #endif | |
1070 | ||
8a7c7393 JK |
1071 | #ifdef KVM_CAP_PIT_STATE2 |
1072 | s->pit_state2 = kvm_check_extension(s, KVM_CAP_PIT_STATE2); | |
1073 | #endif | |
1074 | ||
cad1e282 | 1075 | ret = kvm_arch_init(s); |
a426e122 | 1076 | if (ret < 0) { |
05330448 | 1077 | goto err; |
a426e122 | 1078 | } |
05330448 | 1079 | |
84b058d7 JK |
1080 | ret = kvm_irqchip_create(s); |
1081 | if (ret < 0) { | |
1082 | goto err; | |
1083 | } | |
1084 | ||
05330448 | 1085 | kvm_state = s; |
7376e582 | 1086 | memory_listener_register(&kvm_memory_listener, NULL); |
05330448 | 1087 | |
d2f2b8a7 SH |
1088 | s->many_ioeventfds = kvm_check_many_ioeventfds(); |
1089 | ||
aa7f74d1 JK |
1090 | cpu_interrupt_handler = kvm_handle_interrupt; |
1091 | ||
05330448 AL |
1092 | return 0; |
1093 | ||
1094 | err: | |
1095 | if (s) { | |
db9eae1c | 1096 | if (s->vmfd >= 0) { |
05330448 | 1097 | close(s->vmfd); |
a426e122 JK |
1098 | } |
1099 | if (s->fd != -1) { | |
05330448 | 1100 | close(s->fd); |
a426e122 | 1101 | } |
05330448 | 1102 | } |
7267c094 | 1103 | g_free(s); |
05330448 AL |
1104 | |
1105 | return ret; | |
1106 | } | |
1107 | ||
b30e93e9 JK |
1108 | static void kvm_handle_io(uint16_t port, void *data, int direction, int size, |
1109 | uint32_t count) | |
05330448 AL |
1110 | { |
1111 | int i; | |
1112 | uint8_t *ptr = data; | |
1113 | ||
1114 | for (i = 0; i < count; i++) { | |
1115 | if (direction == KVM_EXIT_IO_IN) { | |
1116 | switch (size) { | |
1117 | case 1: | |
afcea8cb | 1118 | stb_p(ptr, cpu_inb(port)); |
05330448 AL |
1119 | break; |
1120 | case 2: | |
afcea8cb | 1121 | stw_p(ptr, cpu_inw(port)); |
05330448 AL |
1122 | break; |
1123 | case 4: | |
afcea8cb | 1124 | stl_p(ptr, cpu_inl(port)); |
05330448 AL |
1125 | break; |
1126 | } | |
1127 | } else { | |
1128 | switch (size) { | |
1129 | case 1: | |
afcea8cb | 1130 | cpu_outb(port, ldub_p(ptr)); |
05330448 AL |
1131 | break; |
1132 | case 2: | |
afcea8cb | 1133 | cpu_outw(port, lduw_p(ptr)); |
05330448 AL |
1134 | break; |
1135 | case 4: | |
afcea8cb | 1136 | cpu_outl(port, ldl_p(ptr)); |
05330448 AL |
1137 | break; |
1138 | } | |
1139 | } | |
1140 | ||
1141 | ptr += size; | |
1142 | } | |
05330448 AL |
1143 | } |
1144 | ||
9349b4f9 | 1145 | static int kvm_handle_internal_error(CPUArchState *env, struct kvm_run *run) |
7c80eef8 | 1146 | { |
bb44e0d1 | 1147 | fprintf(stderr, "KVM internal error."); |
7c80eef8 MT |
1148 | if (kvm_check_extension(kvm_state, KVM_CAP_INTERNAL_ERROR_DATA)) { |
1149 | int i; | |
1150 | ||
bb44e0d1 | 1151 | fprintf(stderr, " Suberror: %d\n", run->internal.suberror); |
7c80eef8 MT |
1152 | for (i = 0; i < run->internal.ndata; ++i) { |
1153 | fprintf(stderr, "extra data[%d]: %"PRIx64"\n", | |
1154 | i, (uint64_t)run->internal.data[i]); | |
1155 | } | |
bb44e0d1 JK |
1156 | } else { |
1157 | fprintf(stderr, "\n"); | |
7c80eef8 | 1158 | } |
7c80eef8 MT |
1159 | if (run->internal.suberror == KVM_INTERNAL_ERROR_EMULATION) { |
1160 | fprintf(stderr, "emulation failure\n"); | |
a426e122 | 1161 | if (!kvm_arch_stop_on_emulation_error(env)) { |
f5c848ee | 1162 | cpu_dump_state(env, stderr, fprintf, CPU_DUMP_CODE); |
d73cd8f4 | 1163 | return EXCP_INTERRUPT; |
a426e122 | 1164 | } |
7c80eef8 MT |
1165 | } |
1166 | /* FIXME: Should trigger a qmp message to let management know | |
1167 | * something went wrong. | |
1168 | */ | |
73aaec4a | 1169 | return -1; |
7c80eef8 | 1170 | } |
7c80eef8 | 1171 | |
62a2744c | 1172 | void kvm_flush_coalesced_mmio_buffer(void) |
f65ed4c1 | 1173 | { |
f65ed4c1 | 1174 | KVMState *s = kvm_state; |
1cae88b9 AK |
1175 | |
1176 | if (s->coalesced_flush_in_progress) { | |
1177 | return; | |
1178 | } | |
1179 | ||
1180 | s->coalesced_flush_in_progress = true; | |
1181 | ||
62a2744c SY |
1182 | if (s->coalesced_mmio_ring) { |
1183 | struct kvm_coalesced_mmio_ring *ring = s->coalesced_mmio_ring; | |
f65ed4c1 AL |
1184 | while (ring->first != ring->last) { |
1185 | struct kvm_coalesced_mmio *ent; | |
1186 | ||
1187 | ent = &ring->coalesced_mmio[ring->first]; | |
1188 | ||
1189 | cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len); | |
85199474 | 1190 | smp_wmb(); |
f65ed4c1 AL |
1191 | ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX; |
1192 | } | |
1193 | } | |
1cae88b9 AK |
1194 | |
1195 | s->coalesced_flush_in_progress = false; | |
f65ed4c1 AL |
1196 | } |
1197 | ||
2705d56a | 1198 | static void do_kvm_cpu_synchronize_state(void *_env) |
4c0960c0 | 1199 | { |
9349b4f9 | 1200 | CPUArchState *env = _env; |
2705d56a | 1201 | |
9ded2744 | 1202 | if (!env->kvm_vcpu_dirty) { |
4c0960c0 | 1203 | kvm_arch_get_registers(env); |
9ded2744 | 1204 | env->kvm_vcpu_dirty = 1; |
4c0960c0 AK |
1205 | } |
1206 | } | |
1207 | ||
9349b4f9 | 1208 | void kvm_cpu_synchronize_state(CPUArchState *env) |
2705d56a | 1209 | { |
a426e122 | 1210 | if (!env->kvm_vcpu_dirty) { |
2705d56a | 1211 | run_on_cpu(env, do_kvm_cpu_synchronize_state, env); |
a426e122 | 1212 | } |
2705d56a JK |
1213 | } |
1214 | ||
9349b4f9 | 1215 | void kvm_cpu_synchronize_post_reset(CPUArchState *env) |
ea375f9a JK |
1216 | { |
1217 | kvm_arch_put_registers(env, KVM_PUT_RESET_STATE); | |
1218 | env->kvm_vcpu_dirty = 0; | |
1219 | } | |
1220 | ||
9349b4f9 | 1221 | void kvm_cpu_synchronize_post_init(CPUArchState *env) |
ea375f9a JK |
1222 | { |
1223 | kvm_arch_put_registers(env, KVM_PUT_FULL_STATE); | |
1224 | env->kvm_vcpu_dirty = 0; | |
1225 | } | |
1226 | ||
9349b4f9 | 1227 | int kvm_cpu_exec(CPUArchState *env) |
05330448 AL |
1228 | { |
1229 | struct kvm_run *run = env->kvm_run; | |
7cbb533f | 1230 | int ret, run_ret; |
05330448 | 1231 | |
8c0d577e | 1232 | DPRINTF("kvm_cpu_exec()\n"); |
05330448 | 1233 | |
99036865 | 1234 | if (kvm_arch_process_async_events(env)) { |
9ccfac9e | 1235 | env->exit_request = 0; |
6792a57b | 1236 | return EXCP_HLT; |
9ccfac9e | 1237 | } |
0af691d7 | 1238 | |
9ccfac9e | 1239 | do { |
9ded2744 | 1240 | if (env->kvm_vcpu_dirty) { |
ea375f9a | 1241 | kvm_arch_put_registers(env, KVM_PUT_RUNTIME_STATE); |
9ded2744 | 1242 | env->kvm_vcpu_dirty = 0; |
4c0960c0 AK |
1243 | } |
1244 | ||
8c14c173 | 1245 | kvm_arch_pre_run(env, run); |
9ccfac9e JK |
1246 | if (env->exit_request) { |
1247 | DPRINTF("interrupt exit requested\n"); | |
1248 | /* | |
1249 | * KVM requires us to reenter the kernel after IO exits to complete | |
1250 | * instruction emulation. This self-signal will ensure that we | |
1251 | * leave ASAP again. | |
1252 | */ | |
1253 | qemu_cpu_kick_self(); | |
1254 | } | |
d549db5a | 1255 | qemu_mutex_unlock_iothread(); |
9ccfac9e | 1256 | |
7cbb533f | 1257 | run_ret = kvm_vcpu_ioctl(env, KVM_RUN, 0); |
9ccfac9e | 1258 | |
d549db5a | 1259 | qemu_mutex_lock_iothread(); |
05330448 AL |
1260 | kvm_arch_post_run(env, run); |
1261 | ||
b0c883b5 JK |
1262 | kvm_flush_coalesced_mmio_buffer(); |
1263 | ||
7cbb533f | 1264 | if (run_ret < 0) { |
dc77d341 JK |
1265 | if (run_ret == -EINTR || run_ret == -EAGAIN) { |
1266 | DPRINTF("io window exit\n"); | |
d73cd8f4 | 1267 | ret = EXCP_INTERRUPT; |
dc77d341 JK |
1268 | break; |
1269 | } | |
7b011fbc ME |
1270 | fprintf(stderr, "error: kvm run failed %s\n", |
1271 | strerror(-run_ret)); | |
05330448 AL |
1272 | abort(); |
1273 | } | |
1274 | ||
05330448 AL |
1275 | switch (run->exit_reason) { |
1276 | case KVM_EXIT_IO: | |
8c0d577e | 1277 | DPRINTF("handle_io\n"); |
b30e93e9 JK |
1278 | kvm_handle_io(run->io.port, |
1279 | (uint8_t *)run + run->io.data_offset, | |
1280 | run->io.direction, | |
1281 | run->io.size, | |
1282 | run->io.count); | |
d73cd8f4 | 1283 | ret = 0; |
05330448 AL |
1284 | break; |
1285 | case KVM_EXIT_MMIO: | |
8c0d577e | 1286 | DPRINTF("handle_mmio\n"); |
05330448 AL |
1287 | cpu_physical_memory_rw(run->mmio.phys_addr, |
1288 | run->mmio.data, | |
1289 | run->mmio.len, | |
1290 | run->mmio.is_write); | |
d73cd8f4 | 1291 | ret = 0; |
05330448 AL |
1292 | break; |
1293 | case KVM_EXIT_IRQ_WINDOW_OPEN: | |
8c0d577e | 1294 | DPRINTF("irq_window_open\n"); |
d73cd8f4 | 1295 | ret = EXCP_INTERRUPT; |
05330448 AL |
1296 | break; |
1297 | case KVM_EXIT_SHUTDOWN: | |
8c0d577e | 1298 | DPRINTF("shutdown\n"); |
05330448 | 1299 | qemu_system_reset_request(); |
d73cd8f4 | 1300 | ret = EXCP_INTERRUPT; |
05330448 AL |
1301 | break; |
1302 | case KVM_EXIT_UNKNOWN: | |
bb44e0d1 JK |
1303 | fprintf(stderr, "KVM: unknown exit, hardware reason %" PRIx64 "\n", |
1304 | (uint64_t)run->hw.hardware_exit_reason); | |
73aaec4a | 1305 | ret = -1; |
05330448 | 1306 | break; |
7c80eef8 | 1307 | case KVM_EXIT_INTERNAL_ERROR: |
73aaec4a | 1308 | ret = kvm_handle_internal_error(env, run); |
7c80eef8 | 1309 | break; |
05330448 | 1310 | default: |
8c0d577e | 1311 | DPRINTF("kvm_arch_handle_exit\n"); |
05330448 AL |
1312 | ret = kvm_arch_handle_exit(env, run); |
1313 | break; | |
1314 | } | |
d73cd8f4 | 1315 | } while (ret == 0); |
05330448 | 1316 | |
73aaec4a | 1317 | if (ret < 0) { |
f5c848ee | 1318 | cpu_dump_state(env, stderr, fprintf, CPU_DUMP_CODE); |
0461d5a6 | 1319 | vm_stop(RUN_STATE_INTERNAL_ERROR); |
becfc390 AL |
1320 | } |
1321 | ||
6792a57b | 1322 | env->exit_request = 0; |
05330448 AL |
1323 | return ret; |
1324 | } | |
1325 | ||
984b5181 | 1326 | int kvm_ioctl(KVMState *s, int type, ...) |
05330448 AL |
1327 | { |
1328 | int ret; | |
984b5181 AL |
1329 | void *arg; |
1330 | va_list ap; | |
05330448 | 1331 | |
984b5181 AL |
1332 | va_start(ap, type); |
1333 | arg = va_arg(ap, void *); | |
1334 | va_end(ap); | |
1335 | ||
1336 | ret = ioctl(s->fd, type, arg); | |
a426e122 | 1337 | if (ret == -1) { |
05330448 | 1338 | ret = -errno; |
a426e122 | 1339 | } |
05330448 AL |
1340 | return ret; |
1341 | } | |
1342 | ||
984b5181 | 1343 | int kvm_vm_ioctl(KVMState *s, int type, ...) |
05330448 AL |
1344 | { |
1345 | int ret; | |
984b5181 AL |
1346 | void *arg; |
1347 | va_list ap; | |
1348 | ||
1349 | va_start(ap, type); | |
1350 | arg = va_arg(ap, void *); | |
1351 | va_end(ap); | |
05330448 | 1352 | |
984b5181 | 1353 | ret = ioctl(s->vmfd, type, arg); |
a426e122 | 1354 | if (ret == -1) { |
05330448 | 1355 | ret = -errno; |
a426e122 | 1356 | } |
05330448 AL |
1357 | return ret; |
1358 | } | |
1359 | ||
9349b4f9 | 1360 | int kvm_vcpu_ioctl(CPUArchState *env, int type, ...) |
05330448 AL |
1361 | { |
1362 | int ret; | |
984b5181 AL |
1363 | void *arg; |
1364 | va_list ap; | |
1365 | ||
1366 | va_start(ap, type); | |
1367 | arg = va_arg(ap, void *); | |
1368 | va_end(ap); | |
05330448 | 1369 | |
984b5181 | 1370 | ret = ioctl(env->kvm_fd, type, arg); |
a426e122 | 1371 | if (ret == -1) { |
05330448 | 1372 | ret = -errno; |
a426e122 | 1373 | } |
05330448 AL |
1374 | return ret; |
1375 | } | |
bd322087 AL |
1376 | |
1377 | int kvm_has_sync_mmu(void) | |
1378 | { | |
94a8d39a | 1379 | return kvm_check_extension(kvm_state, KVM_CAP_SYNC_MMU); |
bd322087 | 1380 | } |
e22a25c9 | 1381 | |
a0fb002c JK |
1382 | int kvm_has_vcpu_events(void) |
1383 | { | |
1384 | return kvm_state->vcpu_events; | |
1385 | } | |
1386 | ||
b0b1d690 JK |
1387 | int kvm_has_robust_singlestep(void) |
1388 | { | |
1389 | return kvm_state->robust_singlestep; | |
1390 | } | |
1391 | ||
ff44f1a3 JK |
1392 | int kvm_has_debugregs(void) |
1393 | { | |
1394 | return kvm_state->debugregs; | |
1395 | } | |
1396 | ||
f1665b21 SY |
1397 | int kvm_has_xsave(void) |
1398 | { | |
1399 | return kvm_state->xsave; | |
1400 | } | |
1401 | ||
1402 | int kvm_has_xcrs(void) | |
1403 | { | |
1404 | return kvm_state->xcrs; | |
1405 | } | |
1406 | ||
8a7c7393 JK |
1407 | int kvm_has_pit_state2(void) |
1408 | { | |
1409 | return kvm_state->pit_state2; | |
1410 | } | |
1411 | ||
d2f2b8a7 SH |
1412 | int kvm_has_many_ioeventfds(void) |
1413 | { | |
1414 | if (!kvm_enabled()) { | |
1415 | return 0; | |
1416 | } | |
1417 | return kvm_state->many_ioeventfds; | |
1418 | } | |
1419 | ||
84b058d7 JK |
1420 | int kvm_has_gsi_routing(void) |
1421 | { | |
a9c5eb0d | 1422 | #ifdef KVM_CAP_IRQ_ROUTING |
84b058d7 | 1423 | return kvm_check_extension(kvm_state, KVM_CAP_IRQ_ROUTING); |
a9c5eb0d AG |
1424 | #else |
1425 | return false; | |
1426 | #endif | |
84b058d7 JK |
1427 | } |
1428 | ||
9b5b76d4 JK |
1429 | int kvm_allows_irq0_override(void) |
1430 | { | |
3d4b2649 | 1431 | return !kvm_irqchip_in_kernel() || kvm_has_gsi_routing(); |
9b5b76d4 JK |
1432 | } |
1433 | ||
6f0437e8 JK |
1434 | void kvm_setup_guest_memory(void *start, size_t size) |
1435 | { | |
1436 | if (!kvm_has_sync_mmu()) { | |
e78815a5 | 1437 | int ret = qemu_madvise(start, size, QEMU_MADV_DONTFORK); |
6f0437e8 JK |
1438 | |
1439 | if (ret) { | |
e78815a5 AF |
1440 | perror("qemu_madvise"); |
1441 | fprintf(stderr, | |
1442 | "Need MADV_DONTFORK in absence of synchronous KVM MMU\n"); | |
6f0437e8 JK |
1443 | exit(1); |
1444 | } | |
6f0437e8 JK |
1445 | } |
1446 | } | |
1447 | ||
e22a25c9 | 1448 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
9349b4f9 | 1449 | struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUArchState *env, |
e22a25c9 AL |
1450 | target_ulong pc) |
1451 | { | |
1452 | struct kvm_sw_breakpoint *bp; | |
1453 | ||
72cf2d4f | 1454 | QTAILQ_FOREACH(bp, &env->kvm_state->kvm_sw_breakpoints, entry) { |
a426e122 | 1455 | if (bp->pc == pc) { |
e22a25c9 | 1456 | return bp; |
a426e122 | 1457 | } |
e22a25c9 AL |
1458 | } |
1459 | return NULL; | |
1460 | } | |
1461 | ||
9349b4f9 | 1462 | int kvm_sw_breakpoints_active(CPUArchState *env) |
e22a25c9 | 1463 | { |
72cf2d4f | 1464 | return !QTAILQ_EMPTY(&env->kvm_state->kvm_sw_breakpoints); |
e22a25c9 AL |
1465 | } |
1466 | ||
452e4751 GC |
1467 | struct kvm_set_guest_debug_data { |
1468 | struct kvm_guest_debug dbg; | |
9349b4f9 | 1469 | CPUArchState *env; |
452e4751 GC |
1470 | int err; |
1471 | }; | |
1472 | ||
1473 | static void kvm_invoke_set_guest_debug(void *data) | |
1474 | { | |
1475 | struct kvm_set_guest_debug_data *dbg_data = data; | |
9349b4f9 | 1476 | CPUArchState *env = dbg_data->env; |
b3807725 | 1477 | |
b3807725 | 1478 | dbg_data->err = kvm_vcpu_ioctl(env, KVM_SET_GUEST_DEBUG, &dbg_data->dbg); |
452e4751 GC |
1479 | } |
1480 | ||
9349b4f9 | 1481 | int kvm_update_guest_debug(CPUArchState *env, unsigned long reinject_trap) |
e22a25c9 | 1482 | { |
452e4751 | 1483 | struct kvm_set_guest_debug_data data; |
e22a25c9 | 1484 | |
b0b1d690 | 1485 | data.dbg.control = reinject_trap; |
e22a25c9 | 1486 | |
b0b1d690 JK |
1487 | if (env->singlestep_enabled) { |
1488 | data.dbg.control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP; | |
1489 | } | |
452e4751 | 1490 | kvm_arch_update_guest_debug(env, &data.dbg); |
452e4751 | 1491 | data.env = env; |
e22a25c9 | 1492 | |
be41cbe0 | 1493 | run_on_cpu(env, kvm_invoke_set_guest_debug, &data); |
452e4751 | 1494 | return data.err; |
e22a25c9 AL |
1495 | } |
1496 | ||
9349b4f9 | 1497 | int kvm_insert_breakpoint(CPUArchState *current_env, target_ulong addr, |
e22a25c9 AL |
1498 | target_ulong len, int type) |
1499 | { | |
1500 | struct kvm_sw_breakpoint *bp; | |
9349b4f9 | 1501 | CPUArchState *env; |
e22a25c9 AL |
1502 | int err; |
1503 | ||
1504 | if (type == GDB_BREAKPOINT_SW) { | |
1505 | bp = kvm_find_sw_breakpoint(current_env, addr); | |
1506 | if (bp) { | |
1507 | bp->use_count++; | |
1508 | return 0; | |
1509 | } | |
1510 | ||
7267c094 | 1511 | bp = g_malloc(sizeof(struct kvm_sw_breakpoint)); |
a426e122 | 1512 | if (!bp) { |
e22a25c9 | 1513 | return -ENOMEM; |
a426e122 | 1514 | } |
e22a25c9 AL |
1515 | |
1516 | bp->pc = addr; | |
1517 | bp->use_count = 1; | |
1518 | err = kvm_arch_insert_sw_breakpoint(current_env, bp); | |
1519 | if (err) { | |
7267c094 | 1520 | g_free(bp); |
e22a25c9 AL |
1521 | return err; |
1522 | } | |
1523 | ||
72cf2d4f | 1524 | QTAILQ_INSERT_HEAD(¤t_env->kvm_state->kvm_sw_breakpoints, |
e22a25c9 AL |
1525 | bp, entry); |
1526 | } else { | |
1527 | err = kvm_arch_insert_hw_breakpoint(addr, len, type); | |
a426e122 | 1528 | if (err) { |
e22a25c9 | 1529 | return err; |
a426e122 | 1530 | } |
e22a25c9 AL |
1531 | } |
1532 | ||
1533 | for (env = first_cpu; env != NULL; env = env->next_cpu) { | |
1534 | err = kvm_update_guest_debug(env, 0); | |
a426e122 | 1535 | if (err) { |
e22a25c9 | 1536 | return err; |
a426e122 | 1537 | } |
e22a25c9 AL |
1538 | } |
1539 | return 0; | |
1540 | } | |
1541 | ||
9349b4f9 | 1542 | int kvm_remove_breakpoint(CPUArchState *current_env, target_ulong addr, |
e22a25c9 AL |
1543 | target_ulong len, int type) |
1544 | { | |
1545 | struct kvm_sw_breakpoint *bp; | |
9349b4f9 | 1546 | CPUArchState *env; |
e22a25c9 AL |
1547 | int err; |
1548 | ||
1549 | if (type == GDB_BREAKPOINT_SW) { | |
1550 | bp = kvm_find_sw_breakpoint(current_env, addr); | |
a426e122 | 1551 | if (!bp) { |
e22a25c9 | 1552 | return -ENOENT; |
a426e122 | 1553 | } |
e22a25c9 AL |
1554 | |
1555 | if (bp->use_count > 1) { | |
1556 | bp->use_count--; | |
1557 | return 0; | |
1558 | } | |
1559 | ||
1560 | err = kvm_arch_remove_sw_breakpoint(current_env, bp); | |
a426e122 | 1561 | if (err) { |
e22a25c9 | 1562 | return err; |
a426e122 | 1563 | } |
e22a25c9 | 1564 | |
72cf2d4f | 1565 | QTAILQ_REMOVE(¤t_env->kvm_state->kvm_sw_breakpoints, bp, entry); |
7267c094 | 1566 | g_free(bp); |
e22a25c9 AL |
1567 | } else { |
1568 | err = kvm_arch_remove_hw_breakpoint(addr, len, type); | |
a426e122 | 1569 | if (err) { |
e22a25c9 | 1570 | return err; |
a426e122 | 1571 | } |
e22a25c9 AL |
1572 | } |
1573 | ||
1574 | for (env = first_cpu; env != NULL; env = env->next_cpu) { | |
1575 | err = kvm_update_guest_debug(env, 0); | |
a426e122 | 1576 | if (err) { |
e22a25c9 | 1577 | return err; |
a426e122 | 1578 | } |
e22a25c9 AL |
1579 | } |
1580 | return 0; | |
1581 | } | |
1582 | ||
9349b4f9 | 1583 | void kvm_remove_all_breakpoints(CPUArchState *current_env) |
e22a25c9 AL |
1584 | { |
1585 | struct kvm_sw_breakpoint *bp, *next; | |
1586 | KVMState *s = current_env->kvm_state; | |
9349b4f9 | 1587 | CPUArchState *env; |
e22a25c9 | 1588 | |
72cf2d4f | 1589 | QTAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) { |
e22a25c9 AL |
1590 | if (kvm_arch_remove_sw_breakpoint(current_env, bp) != 0) { |
1591 | /* Try harder to find a CPU that currently sees the breakpoint. */ | |
1592 | for (env = first_cpu; env != NULL; env = env->next_cpu) { | |
a426e122 | 1593 | if (kvm_arch_remove_sw_breakpoint(env, bp) == 0) { |
e22a25c9 | 1594 | break; |
a426e122 | 1595 | } |
e22a25c9 AL |
1596 | } |
1597 | } | |
1598 | } | |
1599 | kvm_arch_remove_all_hw_breakpoints(); | |
1600 | ||
a426e122 | 1601 | for (env = first_cpu; env != NULL; env = env->next_cpu) { |
e22a25c9 | 1602 | kvm_update_guest_debug(env, 0); |
a426e122 | 1603 | } |
e22a25c9 AL |
1604 | } |
1605 | ||
1606 | #else /* !KVM_CAP_SET_GUEST_DEBUG */ | |
1607 | ||
9349b4f9 | 1608 | int kvm_update_guest_debug(CPUArchState *env, unsigned long reinject_trap) |
e22a25c9 AL |
1609 | { |
1610 | return -EINVAL; | |
1611 | } | |
1612 | ||
9349b4f9 | 1613 | int kvm_insert_breakpoint(CPUArchState *current_env, target_ulong addr, |
e22a25c9 AL |
1614 | target_ulong len, int type) |
1615 | { | |
1616 | return -EINVAL; | |
1617 | } | |
1618 | ||
9349b4f9 | 1619 | int kvm_remove_breakpoint(CPUArchState *current_env, target_ulong addr, |
e22a25c9 AL |
1620 | target_ulong len, int type) |
1621 | { | |
1622 | return -EINVAL; | |
1623 | } | |
1624 | ||
9349b4f9 | 1625 | void kvm_remove_all_breakpoints(CPUArchState *current_env) |
e22a25c9 AL |
1626 | { |
1627 | } | |
1628 | #endif /* !KVM_CAP_SET_GUEST_DEBUG */ | |
cc84de95 | 1629 | |
9349b4f9 | 1630 | int kvm_set_signal_mask(CPUArchState *env, const sigset_t *sigset) |
cc84de95 MT |
1631 | { |
1632 | struct kvm_signal_mask *sigmask; | |
1633 | int r; | |
1634 | ||
a426e122 | 1635 | if (!sigset) { |
cc84de95 | 1636 | return kvm_vcpu_ioctl(env, KVM_SET_SIGNAL_MASK, NULL); |
a426e122 | 1637 | } |
cc84de95 | 1638 | |
7267c094 | 1639 | sigmask = g_malloc(sizeof(*sigmask) + sizeof(*sigset)); |
cc84de95 MT |
1640 | |
1641 | sigmask->len = 8; | |
1642 | memcpy(sigmask->sigset, sigset, sizeof(*sigset)); | |
1643 | r = kvm_vcpu_ioctl(env, KVM_SET_SIGNAL_MASK, sigmask); | |
7267c094 | 1644 | g_free(sigmask); |
cc84de95 MT |
1645 | |
1646 | return r; | |
1647 | } | |
ca821806 | 1648 | |
4b8f1c88 MT |
1649 | int kvm_set_ioeventfd_mmio(int fd, uint32_t addr, uint32_t val, bool assign, |
1650 | uint32_t size) | |
44f1a3d8 | 1651 | { |
44f1a3d8 CM |
1652 | int ret; |
1653 | struct kvm_ioeventfd iofd; | |
1654 | ||
1655 | iofd.datamatch = val; | |
1656 | iofd.addr = addr; | |
4b8f1c88 | 1657 | iofd.len = size; |
44f1a3d8 CM |
1658 | iofd.flags = KVM_IOEVENTFD_FLAG_DATAMATCH; |
1659 | iofd.fd = fd; | |
1660 | ||
1661 | if (!kvm_enabled()) { | |
1662 | return -ENOSYS; | |
1663 | } | |
1664 | ||
1665 | if (!assign) { | |
1666 | iofd.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN; | |
1667 | } | |
1668 | ||
1669 | ret = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &iofd); | |
1670 | ||
1671 | if (ret < 0) { | |
1672 | return -errno; | |
1673 | } | |
1674 | ||
1675 | return 0; | |
44f1a3d8 CM |
1676 | } |
1677 | ||
ca821806 MT |
1678 | int kvm_set_ioeventfd_pio_word(int fd, uint16_t addr, uint16_t val, bool assign) |
1679 | { | |
1680 | struct kvm_ioeventfd kick = { | |
1681 | .datamatch = val, | |
1682 | .addr = addr, | |
1683 | .len = 2, | |
1684 | .flags = KVM_IOEVENTFD_FLAG_DATAMATCH | KVM_IOEVENTFD_FLAG_PIO, | |
1685 | .fd = fd, | |
1686 | }; | |
1687 | int r; | |
a426e122 | 1688 | if (!kvm_enabled()) { |
ca821806 | 1689 | return -ENOSYS; |
a426e122 JK |
1690 | } |
1691 | if (!assign) { | |
ca821806 | 1692 | kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN; |
a426e122 | 1693 | } |
ca821806 | 1694 | r = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick); |
a426e122 | 1695 | if (r < 0) { |
ca821806 | 1696 | return r; |
a426e122 | 1697 | } |
ca821806 | 1698 | return 0; |
98c8573e | 1699 | } |
a1b87fe0 | 1700 | |
9349b4f9 | 1701 | int kvm_on_sigbus_vcpu(CPUArchState *env, int code, void *addr) |
a1b87fe0 JK |
1702 | { |
1703 | return kvm_arch_on_sigbus_vcpu(env, code, addr); | |
1704 | } | |
1705 | ||
1706 | int kvm_on_sigbus(int code, void *addr) | |
1707 | { | |
1708 | return kvm_arch_on_sigbus(code, addr); | |
1709 | } |