]>
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 | ||
f65ed4c1 AL |
38 | /* KVM uses PAGE_SIZE in it's definition of COALESCED_MMIO_MAX */ |
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 | |
6f725c13 GC |
78 | int irqchip_in_kernel; |
79 | int pit_in_kernel; | |
f1665b21 | 80 | int xsave, xcrs; |
d2f2b8a7 | 81 | int many_ioeventfds; |
84b058d7 JK |
82 | int irqchip_inject_ioctl; |
83 | #ifdef KVM_CAP_IRQ_ROUTING | |
84 | struct kvm_irq_routing *irq_routes; | |
85 | int nr_allocated_irq_routes; | |
86 | uint32_t *used_gsi_bitmap; | |
87 | unsigned int max_gsi; | |
88 | #endif | |
05330448 AL |
89 | }; |
90 | ||
6a7af8cb | 91 | KVMState *kvm_state; |
05330448 | 92 | |
94a8d39a JK |
93 | static const KVMCapabilityInfo kvm_required_capabilites[] = { |
94 | KVM_CAP_INFO(USER_MEMORY), | |
95 | KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS), | |
96 | KVM_CAP_LAST_INFO | |
97 | }; | |
98 | ||
05330448 AL |
99 | static KVMSlot *kvm_alloc_slot(KVMState *s) |
100 | { | |
101 | int i; | |
102 | ||
103 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
a426e122 | 104 | if (s->slots[i].memory_size == 0) { |
05330448 | 105 | return &s->slots[i]; |
a426e122 | 106 | } |
05330448 AL |
107 | } |
108 | ||
d3f8d37f AL |
109 | fprintf(stderr, "%s: no free slot available\n", __func__); |
110 | abort(); | |
111 | } | |
112 | ||
113 | static KVMSlot *kvm_lookup_matching_slot(KVMState *s, | |
c227f099 AL |
114 | target_phys_addr_t start_addr, |
115 | target_phys_addr_t end_addr) | |
d3f8d37f AL |
116 | { |
117 | int i; | |
118 | ||
119 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
120 | KVMSlot *mem = &s->slots[i]; | |
121 | ||
122 | if (start_addr == mem->start_addr && | |
123 | end_addr == mem->start_addr + mem->memory_size) { | |
124 | return mem; | |
125 | } | |
126 | } | |
127 | ||
05330448 AL |
128 | return NULL; |
129 | } | |
130 | ||
6152e2ae AL |
131 | /* |
132 | * Find overlapping slot with lowest start address | |
133 | */ | |
134 | static KVMSlot *kvm_lookup_overlapping_slot(KVMState *s, | |
c227f099 AL |
135 | target_phys_addr_t start_addr, |
136 | target_phys_addr_t end_addr) | |
05330448 | 137 | { |
6152e2ae | 138 | KVMSlot *found = NULL; |
05330448 AL |
139 | int i; |
140 | ||
141 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
142 | KVMSlot *mem = &s->slots[i]; | |
143 | ||
6152e2ae AL |
144 | if (mem->memory_size == 0 || |
145 | (found && found->start_addr < mem->start_addr)) { | |
146 | continue; | |
147 | } | |
148 | ||
149 | if (end_addr > mem->start_addr && | |
150 | start_addr < mem->start_addr + mem->memory_size) { | |
151 | found = mem; | |
152 | } | |
05330448 AL |
153 | } |
154 | ||
6152e2ae | 155 | return found; |
05330448 AL |
156 | } |
157 | ||
9f213ed9 AK |
158 | int kvm_physical_memory_addr_from_host(KVMState *s, void *ram, |
159 | target_phys_addr_t *phys_addr) | |
983dfc3b HY |
160 | { |
161 | int i; | |
162 | ||
163 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
164 | KVMSlot *mem = &s->slots[i]; | |
165 | ||
9f213ed9 AK |
166 | if (ram >= mem->ram && ram < mem->ram + mem->memory_size) { |
167 | *phys_addr = mem->start_addr + (ram - mem->ram); | |
983dfc3b HY |
168 | return 1; |
169 | } | |
170 | } | |
171 | ||
172 | return 0; | |
173 | } | |
174 | ||
5832d1f2 AL |
175 | static int kvm_set_user_memory_region(KVMState *s, KVMSlot *slot) |
176 | { | |
177 | struct kvm_userspace_memory_region mem; | |
178 | ||
179 | mem.slot = slot->slot; | |
180 | mem.guest_phys_addr = slot->start_addr; | |
181 | mem.memory_size = slot->memory_size; | |
9f213ed9 | 182 | mem.userspace_addr = (unsigned long)slot->ram; |
5832d1f2 | 183 | mem.flags = slot->flags; |
4495d6a7 JK |
184 | if (s->migration_log) { |
185 | mem.flags |= KVM_MEM_LOG_DIRTY_PAGES; | |
186 | } | |
5832d1f2 AL |
187 | return kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem); |
188 | } | |
189 | ||
8d2ba1fb JK |
190 | static void kvm_reset_vcpu(void *opaque) |
191 | { | |
192 | CPUState *env = opaque; | |
193 | ||
caa5af0f | 194 | kvm_arch_reset_vcpu(env); |
8d2ba1fb | 195 | } |
5832d1f2 | 196 | |
6f725c13 GC |
197 | int kvm_irqchip_in_kernel(void) |
198 | { | |
199 | return kvm_state->irqchip_in_kernel; | |
200 | } | |
201 | ||
202 | int kvm_pit_in_kernel(void) | |
203 | { | |
204 | return kvm_state->pit_in_kernel; | |
205 | } | |
206 | ||
05330448 AL |
207 | int kvm_init_vcpu(CPUState *env) |
208 | { | |
209 | KVMState *s = kvm_state; | |
210 | long mmap_size; | |
211 | int ret; | |
212 | ||
8c0d577e | 213 | DPRINTF("kvm_init_vcpu\n"); |
05330448 | 214 | |
984b5181 | 215 | ret = kvm_vm_ioctl(s, KVM_CREATE_VCPU, env->cpu_index); |
05330448 | 216 | if (ret < 0) { |
8c0d577e | 217 | DPRINTF("kvm_create_vcpu failed\n"); |
05330448 AL |
218 | goto err; |
219 | } | |
220 | ||
221 | env->kvm_fd = ret; | |
222 | env->kvm_state = s; | |
d841b6c4 | 223 | env->kvm_vcpu_dirty = 1; |
05330448 AL |
224 | |
225 | mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0); | |
226 | if (mmap_size < 0) { | |
748a680b | 227 | ret = mmap_size; |
8c0d577e | 228 | DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n"); |
05330448 AL |
229 | goto err; |
230 | } | |
231 | ||
232 | env->kvm_run = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED, | |
233 | env->kvm_fd, 0); | |
234 | if (env->kvm_run == MAP_FAILED) { | |
235 | ret = -errno; | |
8c0d577e | 236 | DPRINTF("mmap'ing vcpu state failed\n"); |
05330448 AL |
237 | goto err; |
238 | } | |
239 | ||
a426e122 JK |
240 | if (s->coalesced_mmio && !s->coalesced_mmio_ring) { |
241 | s->coalesced_mmio_ring = | |
242 | (void *)env->kvm_run + s->coalesced_mmio * PAGE_SIZE; | |
243 | } | |
62a2744c | 244 | |
05330448 | 245 | ret = kvm_arch_init_vcpu(env); |
8d2ba1fb | 246 | if (ret == 0) { |
a08d4367 | 247 | qemu_register_reset(kvm_reset_vcpu, env); |
caa5af0f | 248 | kvm_arch_reset_vcpu(env); |
8d2ba1fb | 249 | } |
05330448 AL |
250 | err: |
251 | return ret; | |
252 | } | |
253 | ||
5832d1f2 AL |
254 | /* |
255 | * dirty pages logging control | |
256 | */ | |
25254bbc MT |
257 | |
258 | static int kvm_mem_flags(KVMState *s, bool log_dirty) | |
259 | { | |
260 | return log_dirty ? KVM_MEM_LOG_DIRTY_PAGES : 0; | |
261 | } | |
262 | ||
263 | static int kvm_slot_dirty_pages_log_change(KVMSlot *mem, bool log_dirty) | |
5832d1f2 AL |
264 | { |
265 | KVMState *s = kvm_state; | |
25254bbc | 266 | int flags, mask = KVM_MEM_LOG_DIRTY_PAGES; |
4495d6a7 JK |
267 | int old_flags; |
268 | ||
4495d6a7 | 269 | old_flags = mem->flags; |
5832d1f2 | 270 | |
25254bbc | 271 | flags = (mem->flags & ~mask) | kvm_mem_flags(s, log_dirty); |
5832d1f2 AL |
272 | mem->flags = flags; |
273 | ||
4495d6a7 JK |
274 | /* If nothing changed effectively, no need to issue ioctl */ |
275 | if (s->migration_log) { | |
276 | flags |= KVM_MEM_LOG_DIRTY_PAGES; | |
277 | } | |
25254bbc | 278 | |
4495d6a7 | 279 | if (flags == old_flags) { |
25254bbc | 280 | return 0; |
4495d6a7 JK |
281 | } |
282 | ||
5832d1f2 AL |
283 | return kvm_set_user_memory_region(s, mem); |
284 | } | |
285 | ||
25254bbc MT |
286 | static int kvm_dirty_pages_log_change(target_phys_addr_t phys_addr, |
287 | ram_addr_t size, bool log_dirty) | |
288 | { | |
289 | KVMState *s = kvm_state; | |
290 | KVMSlot *mem = kvm_lookup_matching_slot(s, phys_addr, phys_addr + size); | |
291 | ||
292 | if (mem == NULL) { | |
293 | fprintf(stderr, "BUG: %s: invalid parameters " TARGET_FMT_plx "-" | |
294 | TARGET_FMT_plx "\n", __func__, phys_addr, | |
295 | (target_phys_addr_t)(phys_addr + size - 1)); | |
296 | return -EINVAL; | |
297 | } | |
298 | return kvm_slot_dirty_pages_log_change(mem, log_dirty); | |
299 | } | |
300 | ||
a01672d3 AK |
301 | static void kvm_log_start(MemoryListener *listener, |
302 | MemoryRegionSection *section) | |
5832d1f2 | 303 | { |
a01672d3 AK |
304 | int r; |
305 | ||
306 | r = kvm_dirty_pages_log_change(section->offset_within_address_space, | |
307 | section->size, true); | |
308 | if (r < 0) { | |
309 | abort(); | |
310 | } | |
5832d1f2 AL |
311 | } |
312 | ||
a01672d3 AK |
313 | static void kvm_log_stop(MemoryListener *listener, |
314 | MemoryRegionSection *section) | |
5832d1f2 | 315 | { |
a01672d3 AK |
316 | int r; |
317 | ||
318 | r = kvm_dirty_pages_log_change(section->offset_within_address_space, | |
319 | section->size, false); | |
320 | if (r < 0) { | |
321 | abort(); | |
322 | } | |
5832d1f2 AL |
323 | } |
324 | ||
7b8f3b78 | 325 | static int kvm_set_migration_log(int enable) |
4495d6a7 JK |
326 | { |
327 | KVMState *s = kvm_state; | |
328 | KVMSlot *mem; | |
329 | int i, err; | |
330 | ||
331 | s->migration_log = enable; | |
332 | ||
333 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
334 | mem = &s->slots[i]; | |
335 | ||
70fedd76 AW |
336 | if (!mem->memory_size) { |
337 | continue; | |
338 | } | |
4495d6a7 JK |
339 | if (!!(mem->flags & KVM_MEM_LOG_DIRTY_PAGES) == enable) { |
340 | continue; | |
341 | } | |
342 | err = kvm_set_user_memory_region(s, mem); | |
343 | if (err) { | |
344 | return err; | |
345 | } | |
346 | } | |
347 | return 0; | |
348 | } | |
349 | ||
8369e01c | 350 | /* get kvm's dirty pages bitmap and update qemu's */ |
ffcde12f AK |
351 | static int kvm_get_dirty_pages_log_range(MemoryRegionSection *section, |
352 | unsigned long *bitmap) | |
96c1606b | 353 | { |
8369e01c | 354 | unsigned int i, j; |
aa90fec7 BH |
355 | unsigned long page_number, c; |
356 | target_phys_addr_t addr, addr1; | |
ffcde12f | 357 | unsigned int len = ((section->size / TARGET_PAGE_SIZE) + HOST_LONG_BITS - 1) / HOST_LONG_BITS; |
8369e01c MT |
358 | |
359 | /* | |
360 | * bitmap-traveling is faster than memory-traveling (for addr...) | |
361 | * especially when most of the memory is not dirty. | |
362 | */ | |
363 | for (i = 0; i < len; i++) { | |
364 | if (bitmap[i] != 0) { | |
365 | c = leul_to_cpu(bitmap[i]); | |
366 | do { | |
367 | j = ffsl(c) - 1; | |
368 | c &= ~(1ul << j); | |
369 | page_number = i * HOST_LONG_BITS + j; | |
370 | addr1 = page_number * TARGET_PAGE_SIZE; | |
ffcde12f | 371 | addr = section->offset_within_region + addr1; |
fd4aa979 | 372 | memory_region_set_dirty(section->mr, addr, TARGET_PAGE_SIZE); |
8369e01c MT |
373 | } while (c != 0); |
374 | } | |
375 | } | |
376 | return 0; | |
96c1606b AG |
377 | } |
378 | ||
8369e01c MT |
379 | #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1)) |
380 | ||
5832d1f2 AL |
381 | /** |
382 | * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space | |
fd4aa979 BS |
383 | * This function updates qemu's dirty bitmap using |
384 | * memory_region_set_dirty(). This means all bits are set | |
385 | * to dirty. | |
5832d1f2 | 386 | * |
d3f8d37f | 387 | * @start_add: start of logged region. |
5832d1f2 AL |
388 | * @end_addr: end of logged region. |
389 | */ | |
ffcde12f | 390 | static int kvm_physical_sync_dirty_bitmap(MemoryRegionSection *section) |
5832d1f2 AL |
391 | { |
392 | KVMState *s = kvm_state; | |
151f7749 | 393 | unsigned long size, allocated_size = 0; |
151f7749 JK |
394 | KVMDirtyLog d; |
395 | KVMSlot *mem; | |
396 | int ret = 0; | |
ffcde12f AK |
397 | target_phys_addr_t start_addr = section->offset_within_address_space; |
398 | target_phys_addr_t end_addr = start_addr + section->size; | |
5832d1f2 | 399 | |
151f7749 JK |
400 | d.dirty_bitmap = NULL; |
401 | while (start_addr < end_addr) { | |
402 | mem = kvm_lookup_overlapping_slot(s, start_addr, end_addr); | |
403 | if (mem == NULL) { | |
404 | break; | |
405 | } | |
5832d1f2 | 406 | |
51b0c606 MT |
407 | /* XXX bad kernel interface alert |
408 | * For dirty bitmap, kernel allocates array of size aligned to | |
409 | * bits-per-long. But for case when the kernel is 64bits and | |
410 | * the userspace is 32bits, userspace can't align to the same | |
411 | * bits-per-long, since sizeof(long) is different between kernel | |
412 | * and user space. This way, userspace will provide buffer which | |
413 | * may be 4 bytes less than the kernel will use, resulting in | |
414 | * userspace memory corruption (which is not detectable by valgrind | |
415 | * too, in most cases). | |
416 | * So for now, let's align to 64 instead of HOST_LONG_BITS here, in | |
417 | * a hope that sizeof(long) wont become >8 any time soon. | |
418 | */ | |
419 | size = ALIGN(((mem->memory_size) >> TARGET_PAGE_BITS), | |
420 | /*HOST_LONG_BITS*/ 64) / 8; | |
151f7749 | 421 | if (!d.dirty_bitmap) { |
7267c094 | 422 | d.dirty_bitmap = g_malloc(size); |
151f7749 | 423 | } else if (size > allocated_size) { |
7267c094 | 424 | d.dirty_bitmap = g_realloc(d.dirty_bitmap, size); |
151f7749 JK |
425 | } |
426 | allocated_size = size; | |
427 | memset(d.dirty_bitmap, 0, allocated_size); | |
5832d1f2 | 428 | |
151f7749 | 429 | d.slot = mem->slot; |
5832d1f2 | 430 | |
6e489f3f | 431 | if (kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d) == -1) { |
8c0d577e | 432 | DPRINTF("ioctl failed %d\n", errno); |
151f7749 JK |
433 | ret = -1; |
434 | break; | |
435 | } | |
5832d1f2 | 436 | |
ffcde12f | 437 | kvm_get_dirty_pages_log_range(section, d.dirty_bitmap); |
8369e01c | 438 | start_addr = mem->start_addr + mem->memory_size; |
5832d1f2 | 439 | } |
7267c094 | 440 | g_free(d.dirty_bitmap); |
151f7749 JK |
441 | |
442 | return ret; | |
5832d1f2 AL |
443 | } |
444 | ||
c227f099 | 445 | int kvm_coalesce_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_REGISTER_COALESCED_MMIO, &zone); | |
457 | } | |
f65ed4c1 AL |
458 | |
459 | return ret; | |
460 | } | |
461 | ||
c227f099 | 462 | int kvm_uncoalesce_mmio_region(target_phys_addr_t start, ram_addr_t size) |
f65ed4c1 AL |
463 | { |
464 | int ret = -ENOSYS; | |
f65ed4c1 AL |
465 | KVMState *s = kvm_state; |
466 | ||
467 | if (s->coalesced_mmio) { | |
468 | struct kvm_coalesced_mmio_zone zone; | |
469 | ||
470 | zone.addr = start; | |
471 | zone.size = size; | |
472 | ||
473 | ret = kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone); | |
474 | } | |
f65ed4c1 AL |
475 | |
476 | return ret; | |
477 | } | |
478 | ||
ad7b8b33 AL |
479 | int kvm_check_extension(KVMState *s, unsigned int extension) |
480 | { | |
481 | int ret; | |
482 | ||
483 | ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension); | |
484 | if (ret < 0) { | |
485 | ret = 0; | |
486 | } | |
487 | ||
488 | return ret; | |
489 | } | |
490 | ||
d2f2b8a7 SH |
491 | static int kvm_check_many_ioeventfds(void) |
492 | { | |
d0dcac83 SH |
493 | /* Userspace can use ioeventfd for io notification. This requires a host |
494 | * that supports eventfd(2) and an I/O thread; since eventfd does not | |
495 | * support SIGIO it cannot interrupt the vcpu. | |
496 | * | |
497 | * Older kernels have a 6 device limit on the KVM io bus. Find out so we | |
d2f2b8a7 SH |
498 | * can avoid creating too many ioeventfds. |
499 | */ | |
12d4536f | 500 | #if defined(CONFIG_EVENTFD) |
d2f2b8a7 SH |
501 | int ioeventfds[7]; |
502 | int i, ret = 0; | |
503 | for (i = 0; i < ARRAY_SIZE(ioeventfds); i++) { | |
504 | ioeventfds[i] = eventfd(0, EFD_CLOEXEC); | |
505 | if (ioeventfds[i] < 0) { | |
506 | break; | |
507 | } | |
508 | ret = kvm_set_ioeventfd_pio_word(ioeventfds[i], 0, i, true); | |
509 | if (ret < 0) { | |
510 | close(ioeventfds[i]); | |
511 | break; | |
512 | } | |
513 | } | |
514 | ||
515 | /* Decide whether many devices are supported or not */ | |
516 | ret = i == ARRAY_SIZE(ioeventfds); | |
517 | ||
518 | while (i-- > 0) { | |
519 | kvm_set_ioeventfd_pio_word(ioeventfds[i], 0, i, false); | |
520 | close(ioeventfds[i]); | |
521 | } | |
522 | return ret; | |
523 | #else | |
524 | return 0; | |
525 | #endif | |
526 | } | |
527 | ||
94a8d39a JK |
528 | static const KVMCapabilityInfo * |
529 | kvm_check_extension_list(KVMState *s, const KVMCapabilityInfo *list) | |
530 | { | |
531 | while (list->name) { | |
532 | if (!kvm_check_extension(s, list->value)) { | |
533 | return list; | |
534 | } | |
535 | list++; | |
536 | } | |
537 | return NULL; | |
538 | } | |
539 | ||
a01672d3 | 540 | static void kvm_set_phys_mem(MemoryRegionSection *section, bool add) |
46dbef6a MT |
541 | { |
542 | KVMState *s = kvm_state; | |
46dbef6a MT |
543 | KVMSlot *mem, old; |
544 | int err; | |
a01672d3 AK |
545 | MemoryRegion *mr = section->mr; |
546 | bool log_dirty = memory_region_is_logging(mr); | |
547 | target_phys_addr_t start_addr = section->offset_within_address_space; | |
548 | ram_addr_t size = section->size; | |
9f213ed9 | 549 | void *ram = NULL; |
46dbef6a | 550 | |
14542fea GN |
551 | /* kvm works in page size chunks, but the function may be called |
552 | with sub-page size and unaligned start address. */ | |
553 | size = TARGET_PAGE_ALIGN(size); | |
554 | start_addr = TARGET_PAGE_ALIGN(start_addr); | |
46dbef6a | 555 | |
a01672d3 AK |
556 | if (!memory_region_is_ram(mr)) { |
557 | return; | |
9f213ed9 AK |
558 | } |
559 | ||
a01672d3 AK |
560 | ram = memory_region_get_ram_ptr(mr) + section->offset_within_region; |
561 | ||
46dbef6a MT |
562 | while (1) { |
563 | mem = kvm_lookup_overlapping_slot(s, start_addr, start_addr + size); | |
564 | if (!mem) { | |
565 | break; | |
566 | } | |
567 | ||
a01672d3 | 568 | if (add && start_addr >= mem->start_addr && |
46dbef6a | 569 | (start_addr + size <= mem->start_addr + mem->memory_size) && |
9f213ed9 | 570 | (ram - start_addr == mem->ram - mem->start_addr)) { |
46dbef6a | 571 | /* The new slot fits into the existing one and comes with |
25254bbc MT |
572 | * identical parameters - update flags and done. */ |
573 | kvm_slot_dirty_pages_log_change(mem, log_dirty); | |
46dbef6a MT |
574 | return; |
575 | } | |
576 | ||
577 | old = *mem; | |
578 | ||
3fbffb62 AK |
579 | if (mem->flags & KVM_MEM_LOG_DIRTY_PAGES) { |
580 | kvm_physical_sync_dirty_bitmap(section); | |
581 | } | |
582 | ||
46dbef6a MT |
583 | /* unregister the overlapping slot */ |
584 | mem->memory_size = 0; | |
585 | err = kvm_set_user_memory_region(s, mem); | |
586 | if (err) { | |
587 | fprintf(stderr, "%s: error unregistering overlapping slot: %s\n", | |
588 | __func__, strerror(-err)); | |
589 | abort(); | |
590 | } | |
591 | ||
592 | /* Workaround for older KVM versions: we can't join slots, even not by | |
593 | * unregistering the previous ones and then registering the larger | |
594 | * slot. We have to maintain the existing fragmentation. Sigh. | |
595 | * | |
596 | * This workaround assumes that the new slot starts at the same | |
597 | * address as the first existing one. If not or if some overlapping | |
598 | * slot comes around later, we will fail (not seen in practice so far) | |
599 | * - and actually require a recent KVM version. */ | |
600 | if (s->broken_set_mem_region && | |
a01672d3 | 601 | old.start_addr == start_addr && old.memory_size < size && add) { |
46dbef6a MT |
602 | mem = kvm_alloc_slot(s); |
603 | mem->memory_size = old.memory_size; | |
604 | mem->start_addr = old.start_addr; | |
9f213ed9 | 605 | mem->ram = old.ram; |
25254bbc | 606 | mem->flags = kvm_mem_flags(s, log_dirty); |
46dbef6a MT |
607 | |
608 | err = kvm_set_user_memory_region(s, mem); | |
609 | if (err) { | |
610 | fprintf(stderr, "%s: error updating slot: %s\n", __func__, | |
611 | strerror(-err)); | |
612 | abort(); | |
613 | } | |
614 | ||
615 | start_addr += old.memory_size; | |
9f213ed9 | 616 | ram += old.memory_size; |
46dbef6a MT |
617 | size -= old.memory_size; |
618 | continue; | |
619 | } | |
620 | ||
621 | /* register prefix slot */ | |
622 | if (old.start_addr < start_addr) { | |
623 | mem = kvm_alloc_slot(s); | |
624 | mem->memory_size = start_addr - old.start_addr; | |
625 | mem->start_addr = old.start_addr; | |
9f213ed9 | 626 | mem->ram = old.ram; |
25254bbc | 627 | mem->flags = kvm_mem_flags(s, log_dirty); |
46dbef6a MT |
628 | |
629 | err = kvm_set_user_memory_region(s, mem); | |
630 | if (err) { | |
631 | fprintf(stderr, "%s: error registering prefix slot: %s\n", | |
632 | __func__, strerror(-err)); | |
d4d6868f AG |
633 | #ifdef TARGET_PPC |
634 | fprintf(stderr, "%s: This is probably because your kernel's " \ | |
635 | "PAGE_SIZE is too big. Please try to use 4k " \ | |
636 | "PAGE_SIZE!\n", __func__); | |
637 | #endif | |
46dbef6a MT |
638 | abort(); |
639 | } | |
640 | } | |
641 | ||
642 | /* register suffix slot */ | |
643 | if (old.start_addr + old.memory_size > start_addr + size) { | |
644 | ram_addr_t size_delta; | |
645 | ||
646 | mem = kvm_alloc_slot(s); | |
647 | mem->start_addr = start_addr + size; | |
648 | size_delta = mem->start_addr - old.start_addr; | |
649 | mem->memory_size = old.memory_size - size_delta; | |
9f213ed9 | 650 | mem->ram = old.ram + size_delta; |
25254bbc | 651 | mem->flags = kvm_mem_flags(s, log_dirty); |
46dbef6a MT |
652 | |
653 | err = kvm_set_user_memory_region(s, mem); | |
654 | if (err) { | |
655 | fprintf(stderr, "%s: error registering suffix slot: %s\n", | |
656 | __func__, strerror(-err)); | |
657 | abort(); | |
658 | } | |
659 | } | |
660 | } | |
661 | ||
662 | /* in case the KVM bug workaround already "consumed" the new slot */ | |
a426e122 | 663 | if (!size) { |
46dbef6a | 664 | return; |
a426e122 | 665 | } |
a01672d3 | 666 | if (!add) { |
46dbef6a | 667 | return; |
a426e122 | 668 | } |
46dbef6a MT |
669 | mem = kvm_alloc_slot(s); |
670 | mem->memory_size = size; | |
671 | mem->start_addr = start_addr; | |
9f213ed9 | 672 | mem->ram = ram; |
25254bbc | 673 | mem->flags = kvm_mem_flags(s, log_dirty); |
46dbef6a MT |
674 | |
675 | err = kvm_set_user_memory_region(s, mem); | |
676 | if (err) { | |
677 | fprintf(stderr, "%s: error registering slot: %s\n", __func__, | |
678 | strerror(-err)); | |
679 | abort(); | |
680 | } | |
681 | } | |
682 | ||
50c1e149 AK |
683 | static void kvm_begin(MemoryListener *listener) |
684 | { | |
685 | } | |
686 | ||
687 | static void kvm_commit(MemoryListener *listener) | |
688 | { | |
689 | } | |
690 | ||
a01672d3 AK |
691 | static void kvm_region_add(MemoryListener *listener, |
692 | MemoryRegionSection *section) | |
693 | { | |
694 | kvm_set_phys_mem(section, true); | |
695 | } | |
696 | ||
697 | static void kvm_region_del(MemoryListener *listener, | |
698 | MemoryRegionSection *section) | |
699 | { | |
700 | kvm_set_phys_mem(section, false); | |
701 | } | |
702 | ||
50c1e149 AK |
703 | static void kvm_region_nop(MemoryListener *listener, |
704 | MemoryRegionSection *section) | |
705 | { | |
706 | } | |
707 | ||
a01672d3 AK |
708 | static void kvm_log_sync(MemoryListener *listener, |
709 | MemoryRegionSection *section) | |
7b8f3b78 | 710 | { |
a01672d3 AK |
711 | int r; |
712 | ||
ffcde12f | 713 | r = kvm_physical_sync_dirty_bitmap(section); |
a01672d3 AK |
714 | if (r < 0) { |
715 | abort(); | |
716 | } | |
7b8f3b78 MT |
717 | } |
718 | ||
a01672d3 | 719 | static void kvm_log_global_start(struct MemoryListener *listener) |
7b8f3b78 | 720 | { |
a01672d3 AK |
721 | int r; |
722 | ||
723 | r = kvm_set_migration_log(1); | |
724 | assert(r >= 0); | |
7b8f3b78 MT |
725 | } |
726 | ||
a01672d3 | 727 | static void kvm_log_global_stop(struct MemoryListener *listener) |
7b8f3b78 | 728 | { |
a01672d3 AK |
729 | int r; |
730 | ||
731 | r = kvm_set_migration_log(0); | |
732 | assert(r >= 0); | |
7b8f3b78 MT |
733 | } |
734 | ||
80a1ea37 AK |
735 | static void kvm_mem_ioeventfd_add(MemoryRegionSection *section, |
736 | bool match_data, uint64_t data, int fd) | |
737 | { | |
738 | int r; | |
739 | ||
740 | assert(match_data && section->size == 4); | |
741 | ||
742 | r = kvm_set_ioeventfd_mmio_long(fd, section->offset_within_address_space, | |
743 | data, true); | |
744 | if (r < 0) { | |
745 | abort(); | |
746 | } | |
747 | } | |
748 | ||
749 | static void kvm_mem_ioeventfd_del(MemoryRegionSection *section, | |
750 | bool match_data, uint64_t data, int fd) | |
751 | { | |
752 | int r; | |
753 | ||
754 | r = kvm_set_ioeventfd_mmio_long(fd, section->offset_within_address_space, | |
755 | data, false); | |
756 | if (r < 0) { | |
757 | abort(); | |
758 | } | |
759 | } | |
760 | ||
761 | static void kvm_io_ioeventfd_add(MemoryRegionSection *section, | |
762 | bool match_data, uint64_t data, int fd) | |
763 | { | |
764 | int r; | |
765 | ||
766 | assert(match_data && section->size == 2); | |
767 | ||
768 | r = kvm_set_ioeventfd_pio_word(fd, section->offset_within_address_space, | |
769 | data, true); | |
770 | if (r < 0) { | |
771 | abort(); | |
772 | } | |
773 | } | |
774 | ||
775 | static void kvm_io_ioeventfd_del(MemoryRegionSection *section, | |
776 | bool match_data, uint64_t data, int fd) | |
777 | ||
778 | { | |
779 | int r; | |
780 | ||
781 | r = kvm_set_ioeventfd_pio_word(fd, section->offset_within_address_space, | |
782 | data, false); | |
783 | if (r < 0) { | |
784 | abort(); | |
785 | } | |
786 | } | |
787 | ||
788 | static void kvm_eventfd_add(MemoryListener *listener, | |
789 | MemoryRegionSection *section, | |
790 | bool match_data, uint64_t data, int fd) | |
791 | { | |
792 | if (section->address_space == get_system_memory()) { | |
793 | kvm_mem_ioeventfd_add(section, match_data, data, fd); | |
794 | } else { | |
795 | kvm_io_ioeventfd_add(section, match_data, data, fd); | |
796 | } | |
797 | } | |
798 | ||
799 | static void kvm_eventfd_del(MemoryListener *listener, | |
800 | MemoryRegionSection *section, | |
801 | bool match_data, uint64_t data, int fd) | |
802 | { | |
803 | if (section->address_space == get_system_memory()) { | |
804 | kvm_mem_ioeventfd_del(section, match_data, data, fd); | |
805 | } else { | |
806 | kvm_io_ioeventfd_del(section, match_data, data, fd); | |
807 | } | |
808 | } | |
809 | ||
a01672d3 | 810 | static MemoryListener kvm_memory_listener = { |
50c1e149 AK |
811 | .begin = kvm_begin, |
812 | .commit = kvm_commit, | |
a01672d3 AK |
813 | .region_add = kvm_region_add, |
814 | .region_del = kvm_region_del, | |
50c1e149 | 815 | .region_nop = kvm_region_nop, |
e5896b12 AP |
816 | .log_start = kvm_log_start, |
817 | .log_stop = kvm_log_stop, | |
a01672d3 AK |
818 | .log_sync = kvm_log_sync, |
819 | .log_global_start = kvm_log_global_start, | |
820 | .log_global_stop = kvm_log_global_stop, | |
80a1ea37 AK |
821 | .eventfd_add = kvm_eventfd_add, |
822 | .eventfd_del = kvm_eventfd_del, | |
72e22d2f | 823 | .priority = 10, |
7b8f3b78 MT |
824 | }; |
825 | ||
aa7f74d1 JK |
826 | static void kvm_handle_interrupt(CPUState *env, int mask) |
827 | { | |
828 | env->interrupt_request |= mask; | |
829 | ||
830 | if (!qemu_cpu_is_self(env)) { | |
831 | qemu_cpu_kick(env); | |
832 | } | |
833 | } | |
834 | ||
84b058d7 JK |
835 | int kvm_irqchip_set_irq(KVMState *s, int irq, int level) |
836 | { | |
837 | struct kvm_irq_level event; | |
838 | int ret; | |
839 | ||
840 | assert(s->irqchip_in_kernel); | |
841 | ||
842 | event.level = level; | |
843 | event.irq = irq; | |
844 | ret = kvm_vm_ioctl(s, s->irqchip_inject_ioctl, &event); | |
845 | if (ret < 0) { | |
846 | perror("kvm_set_irqchip_line"); | |
847 | abort(); | |
848 | } | |
849 | ||
850 | return (s->irqchip_inject_ioctl == KVM_IRQ_LINE) ? 1 : event.status; | |
851 | } | |
852 | ||
853 | #ifdef KVM_CAP_IRQ_ROUTING | |
854 | static void set_gsi(KVMState *s, unsigned int gsi) | |
855 | { | |
856 | assert(gsi < s->max_gsi); | |
857 | ||
858 | s->used_gsi_bitmap[gsi / 32] |= 1U << (gsi % 32); | |
859 | } | |
860 | ||
861 | static void kvm_init_irq_routing(KVMState *s) | |
862 | { | |
863 | int gsi_count; | |
864 | ||
865 | gsi_count = kvm_check_extension(s, KVM_CAP_IRQ_ROUTING); | |
866 | if (gsi_count > 0) { | |
867 | unsigned int gsi_bits, i; | |
868 | ||
869 | /* Round up so we can search ints using ffs */ | |
870 | gsi_bits = (gsi_count + 31) / 32; | |
871 | s->used_gsi_bitmap = g_malloc0(gsi_bits / 8); | |
872 | s->max_gsi = gsi_bits; | |
873 | ||
874 | /* Mark any over-allocated bits as already in use */ | |
875 | for (i = gsi_count; i < gsi_bits; i++) { | |
876 | set_gsi(s, i); | |
877 | } | |
878 | } | |
879 | ||
880 | s->irq_routes = g_malloc0(sizeof(*s->irq_routes)); | |
881 | s->nr_allocated_irq_routes = 0; | |
882 | ||
883 | kvm_arch_init_irq_routing(s); | |
884 | } | |
885 | ||
886 | static void kvm_add_routing_entry(KVMState *s, | |
887 | struct kvm_irq_routing_entry *entry) | |
888 | { | |
889 | struct kvm_irq_routing_entry *new; | |
890 | int n, size; | |
891 | ||
892 | if (s->irq_routes->nr == s->nr_allocated_irq_routes) { | |
893 | n = s->nr_allocated_irq_routes * 2; | |
894 | if (n < 64) { | |
895 | n = 64; | |
896 | } | |
897 | size = sizeof(struct kvm_irq_routing); | |
898 | size += n * sizeof(*new); | |
899 | s->irq_routes = g_realloc(s->irq_routes, size); | |
900 | s->nr_allocated_irq_routes = n; | |
901 | } | |
902 | n = s->irq_routes->nr++; | |
903 | new = &s->irq_routes->entries[n]; | |
904 | memset(new, 0, sizeof(*new)); | |
905 | new->gsi = entry->gsi; | |
906 | new->type = entry->type; | |
907 | new->flags = entry->flags; | |
908 | new->u = entry->u; | |
909 | ||
910 | set_gsi(s, entry->gsi); | |
911 | } | |
912 | ||
913 | void kvm_irqchip_add_route(KVMState *s, int irq, int irqchip, int pin) | |
914 | { | |
915 | struct kvm_irq_routing_entry e; | |
916 | ||
917 | e.gsi = irq; | |
918 | e.type = KVM_IRQ_ROUTING_IRQCHIP; | |
919 | e.flags = 0; | |
920 | e.u.irqchip.irqchip = irqchip; | |
921 | e.u.irqchip.pin = pin; | |
922 | kvm_add_routing_entry(s, &e); | |
923 | } | |
924 | ||
925 | int kvm_irqchip_commit_routes(KVMState *s) | |
926 | { | |
927 | s->irq_routes->flags = 0; | |
928 | return kvm_vm_ioctl(s, KVM_SET_GSI_ROUTING, s->irq_routes); | |
929 | } | |
930 | ||
931 | #else /* !KVM_CAP_IRQ_ROUTING */ | |
932 | ||
933 | static void kvm_init_irq_routing(KVMState *s) | |
934 | { | |
935 | } | |
936 | #endif /* !KVM_CAP_IRQ_ROUTING */ | |
937 | ||
938 | static int kvm_irqchip_create(KVMState *s) | |
939 | { | |
940 | QemuOptsList *list = qemu_find_opts("machine"); | |
941 | int ret; | |
942 | ||
943 | if (QTAILQ_EMPTY(&list->head) || | |
944 | !qemu_opt_get_bool(QTAILQ_FIRST(&list->head), | |
945 | "kernel_irqchip", false) || | |
946 | !kvm_check_extension(s, KVM_CAP_IRQCHIP)) { | |
947 | return 0; | |
948 | } | |
949 | ||
950 | ret = kvm_vm_ioctl(s, KVM_CREATE_IRQCHIP); | |
951 | if (ret < 0) { | |
952 | fprintf(stderr, "Create kernel irqchip failed\n"); | |
953 | return ret; | |
954 | } | |
955 | ||
956 | s->irqchip_inject_ioctl = KVM_IRQ_LINE; | |
957 | if (kvm_check_extension(s, KVM_CAP_IRQ_INJECT_STATUS)) { | |
958 | s->irqchip_inject_ioctl = KVM_IRQ_LINE_STATUS; | |
959 | } | |
960 | s->irqchip_in_kernel = 1; | |
961 | ||
962 | kvm_init_irq_routing(s); | |
963 | ||
964 | return 0; | |
965 | } | |
966 | ||
cad1e282 | 967 | int kvm_init(void) |
05330448 | 968 | { |
168ccc11 JK |
969 | static const char upgrade_note[] = |
970 | "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n" | |
971 | "(see http://sourceforge.net/projects/kvm).\n"; | |
05330448 | 972 | KVMState *s; |
94a8d39a | 973 | const KVMCapabilityInfo *missing_cap; |
05330448 AL |
974 | int ret; |
975 | int i; | |
976 | ||
7267c094 | 977 | s = g_malloc0(sizeof(KVMState)); |
05330448 | 978 | |
e22a25c9 | 979 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
72cf2d4f | 980 | QTAILQ_INIT(&s->kvm_sw_breakpoints); |
e22a25c9 | 981 | #endif |
a426e122 | 982 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { |
05330448 | 983 | s->slots[i].slot = i; |
a426e122 | 984 | } |
05330448 | 985 | s->vmfd = -1; |
40ff6d7e | 986 | s->fd = qemu_open("/dev/kvm", O_RDWR); |
05330448 AL |
987 | if (s->fd == -1) { |
988 | fprintf(stderr, "Could not access KVM kernel module: %m\n"); | |
989 | ret = -errno; | |
990 | goto err; | |
991 | } | |
992 | ||
993 | ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0); | |
994 | if (ret < KVM_API_VERSION) { | |
a426e122 | 995 | if (ret > 0) { |
05330448 | 996 | ret = -EINVAL; |
a426e122 | 997 | } |
05330448 AL |
998 | fprintf(stderr, "kvm version too old\n"); |
999 | goto err; | |
1000 | } | |
1001 | ||
1002 | if (ret > KVM_API_VERSION) { | |
1003 | ret = -EINVAL; | |
1004 | fprintf(stderr, "kvm version not supported\n"); | |
1005 | goto err; | |
1006 | } | |
1007 | ||
1008 | s->vmfd = kvm_ioctl(s, KVM_CREATE_VM, 0); | |
0104dcac AG |
1009 | if (s->vmfd < 0) { |
1010 | #ifdef TARGET_S390X | |
1011 | fprintf(stderr, "Please add the 'switch_amode' kernel parameter to " | |
1012 | "your host kernel command line\n"); | |
1013 | #endif | |
db9eae1c | 1014 | ret = s->vmfd; |
05330448 | 1015 | goto err; |
0104dcac | 1016 | } |
05330448 | 1017 | |
94a8d39a JK |
1018 | missing_cap = kvm_check_extension_list(s, kvm_required_capabilites); |
1019 | if (!missing_cap) { | |
1020 | missing_cap = | |
1021 | kvm_check_extension_list(s, kvm_arch_required_capabilities); | |
05330448 | 1022 | } |
94a8d39a | 1023 | if (missing_cap) { |
ad7b8b33 | 1024 | ret = -EINVAL; |
94a8d39a JK |
1025 | fprintf(stderr, "kvm does not support %s\n%s", |
1026 | missing_cap->name, upgrade_note); | |
d85dc283 AL |
1027 | goto err; |
1028 | } | |
1029 | ||
ad7b8b33 | 1030 | s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO); |
f65ed4c1 | 1031 | |
e69917e2 | 1032 | s->broken_set_mem_region = 1; |
14a09518 | 1033 | ret = kvm_check_extension(s, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS); |
e69917e2 JK |
1034 | if (ret > 0) { |
1035 | s->broken_set_mem_region = 0; | |
1036 | } | |
e69917e2 | 1037 | |
a0fb002c JK |
1038 | #ifdef KVM_CAP_VCPU_EVENTS |
1039 | s->vcpu_events = kvm_check_extension(s, KVM_CAP_VCPU_EVENTS); | |
1040 | #endif | |
1041 | ||
b0b1d690 JK |
1042 | s->robust_singlestep = |
1043 | kvm_check_extension(s, KVM_CAP_X86_ROBUST_SINGLESTEP); | |
b0b1d690 | 1044 | |
ff44f1a3 JK |
1045 | #ifdef KVM_CAP_DEBUGREGS |
1046 | s->debugregs = kvm_check_extension(s, KVM_CAP_DEBUGREGS); | |
1047 | #endif | |
1048 | ||
f1665b21 SY |
1049 | #ifdef KVM_CAP_XSAVE |
1050 | s->xsave = kvm_check_extension(s, KVM_CAP_XSAVE); | |
1051 | #endif | |
1052 | ||
f1665b21 SY |
1053 | #ifdef KVM_CAP_XCRS |
1054 | s->xcrs = kvm_check_extension(s, KVM_CAP_XCRS); | |
1055 | #endif | |
1056 | ||
cad1e282 | 1057 | ret = kvm_arch_init(s); |
a426e122 | 1058 | if (ret < 0) { |
05330448 | 1059 | goto err; |
a426e122 | 1060 | } |
05330448 | 1061 | |
84b058d7 JK |
1062 | ret = kvm_irqchip_create(s); |
1063 | if (ret < 0) { | |
1064 | goto err; | |
1065 | } | |
1066 | ||
05330448 | 1067 | kvm_state = s; |
7376e582 | 1068 | memory_listener_register(&kvm_memory_listener, NULL); |
05330448 | 1069 | |
d2f2b8a7 SH |
1070 | s->many_ioeventfds = kvm_check_many_ioeventfds(); |
1071 | ||
aa7f74d1 JK |
1072 | cpu_interrupt_handler = kvm_handle_interrupt; |
1073 | ||
05330448 AL |
1074 | return 0; |
1075 | ||
1076 | err: | |
1077 | if (s) { | |
db9eae1c | 1078 | if (s->vmfd >= 0) { |
05330448 | 1079 | close(s->vmfd); |
a426e122 JK |
1080 | } |
1081 | if (s->fd != -1) { | |
05330448 | 1082 | close(s->fd); |
a426e122 | 1083 | } |
05330448 | 1084 | } |
7267c094 | 1085 | g_free(s); |
05330448 AL |
1086 | |
1087 | return ret; | |
1088 | } | |
1089 | ||
b30e93e9 JK |
1090 | static void kvm_handle_io(uint16_t port, void *data, int direction, int size, |
1091 | uint32_t count) | |
05330448 AL |
1092 | { |
1093 | int i; | |
1094 | uint8_t *ptr = data; | |
1095 | ||
1096 | for (i = 0; i < count; i++) { | |
1097 | if (direction == KVM_EXIT_IO_IN) { | |
1098 | switch (size) { | |
1099 | case 1: | |
afcea8cb | 1100 | stb_p(ptr, cpu_inb(port)); |
05330448 AL |
1101 | break; |
1102 | case 2: | |
afcea8cb | 1103 | stw_p(ptr, cpu_inw(port)); |
05330448 AL |
1104 | break; |
1105 | case 4: | |
afcea8cb | 1106 | stl_p(ptr, cpu_inl(port)); |
05330448 AL |
1107 | break; |
1108 | } | |
1109 | } else { | |
1110 | switch (size) { | |
1111 | case 1: | |
afcea8cb | 1112 | cpu_outb(port, ldub_p(ptr)); |
05330448 AL |
1113 | break; |
1114 | case 2: | |
afcea8cb | 1115 | cpu_outw(port, lduw_p(ptr)); |
05330448 AL |
1116 | break; |
1117 | case 4: | |
afcea8cb | 1118 | cpu_outl(port, ldl_p(ptr)); |
05330448 AL |
1119 | break; |
1120 | } | |
1121 | } | |
1122 | ||
1123 | ptr += size; | |
1124 | } | |
05330448 AL |
1125 | } |
1126 | ||
73aaec4a | 1127 | static int kvm_handle_internal_error(CPUState *env, struct kvm_run *run) |
7c80eef8 | 1128 | { |
bb44e0d1 | 1129 | fprintf(stderr, "KVM internal error."); |
7c80eef8 MT |
1130 | if (kvm_check_extension(kvm_state, KVM_CAP_INTERNAL_ERROR_DATA)) { |
1131 | int i; | |
1132 | ||
bb44e0d1 | 1133 | fprintf(stderr, " Suberror: %d\n", run->internal.suberror); |
7c80eef8 MT |
1134 | for (i = 0; i < run->internal.ndata; ++i) { |
1135 | fprintf(stderr, "extra data[%d]: %"PRIx64"\n", | |
1136 | i, (uint64_t)run->internal.data[i]); | |
1137 | } | |
bb44e0d1 JK |
1138 | } else { |
1139 | fprintf(stderr, "\n"); | |
7c80eef8 | 1140 | } |
7c80eef8 MT |
1141 | if (run->internal.suberror == KVM_INTERNAL_ERROR_EMULATION) { |
1142 | fprintf(stderr, "emulation failure\n"); | |
a426e122 | 1143 | if (!kvm_arch_stop_on_emulation_error(env)) { |
f5c848ee | 1144 | cpu_dump_state(env, stderr, fprintf, CPU_DUMP_CODE); |
d73cd8f4 | 1145 | return EXCP_INTERRUPT; |
a426e122 | 1146 | } |
7c80eef8 MT |
1147 | } |
1148 | /* FIXME: Should trigger a qmp message to let management know | |
1149 | * something went wrong. | |
1150 | */ | |
73aaec4a | 1151 | return -1; |
7c80eef8 | 1152 | } |
7c80eef8 | 1153 | |
62a2744c | 1154 | void kvm_flush_coalesced_mmio_buffer(void) |
f65ed4c1 | 1155 | { |
f65ed4c1 | 1156 | KVMState *s = kvm_state; |
1cae88b9 AK |
1157 | |
1158 | if (s->coalesced_flush_in_progress) { | |
1159 | return; | |
1160 | } | |
1161 | ||
1162 | s->coalesced_flush_in_progress = true; | |
1163 | ||
62a2744c SY |
1164 | if (s->coalesced_mmio_ring) { |
1165 | struct kvm_coalesced_mmio_ring *ring = s->coalesced_mmio_ring; | |
f65ed4c1 AL |
1166 | while (ring->first != ring->last) { |
1167 | struct kvm_coalesced_mmio *ent; | |
1168 | ||
1169 | ent = &ring->coalesced_mmio[ring->first]; | |
1170 | ||
1171 | cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len); | |
85199474 | 1172 | smp_wmb(); |
f65ed4c1 AL |
1173 | ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX; |
1174 | } | |
1175 | } | |
1cae88b9 AK |
1176 | |
1177 | s->coalesced_flush_in_progress = false; | |
f65ed4c1 AL |
1178 | } |
1179 | ||
2705d56a | 1180 | static void do_kvm_cpu_synchronize_state(void *_env) |
4c0960c0 | 1181 | { |
2705d56a JK |
1182 | CPUState *env = _env; |
1183 | ||
9ded2744 | 1184 | if (!env->kvm_vcpu_dirty) { |
4c0960c0 | 1185 | kvm_arch_get_registers(env); |
9ded2744 | 1186 | env->kvm_vcpu_dirty = 1; |
4c0960c0 AK |
1187 | } |
1188 | } | |
1189 | ||
2705d56a JK |
1190 | void kvm_cpu_synchronize_state(CPUState *env) |
1191 | { | |
a426e122 | 1192 | if (!env->kvm_vcpu_dirty) { |
2705d56a | 1193 | run_on_cpu(env, do_kvm_cpu_synchronize_state, env); |
a426e122 | 1194 | } |
2705d56a JK |
1195 | } |
1196 | ||
ea375f9a JK |
1197 | void kvm_cpu_synchronize_post_reset(CPUState *env) |
1198 | { | |
1199 | kvm_arch_put_registers(env, KVM_PUT_RESET_STATE); | |
1200 | env->kvm_vcpu_dirty = 0; | |
1201 | } | |
1202 | ||
1203 | void kvm_cpu_synchronize_post_init(CPUState *env) | |
1204 | { | |
1205 | kvm_arch_put_registers(env, KVM_PUT_FULL_STATE); | |
1206 | env->kvm_vcpu_dirty = 0; | |
1207 | } | |
1208 | ||
05330448 AL |
1209 | int kvm_cpu_exec(CPUState *env) |
1210 | { | |
1211 | struct kvm_run *run = env->kvm_run; | |
7cbb533f | 1212 | int ret, run_ret; |
05330448 | 1213 | |
8c0d577e | 1214 | DPRINTF("kvm_cpu_exec()\n"); |
05330448 | 1215 | |
99036865 | 1216 | if (kvm_arch_process_async_events(env)) { |
9ccfac9e | 1217 | env->exit_request = 0; |
6792a57b | 1218 | return EXCP_HLT; |
9ccfac9e | 1219 | } |
0af691d7 | 1220 | |
6792a57b JK |
1221 | cpu_single_env = env; |
1222 | ||
9ccfac9e | 1223 | do { |
9ded2744 | 1224 | if (env->kvm_vcpu_dirty) { |
ea375f9a | 1225 | kvm_arch_put_registers(env, KVM_PUT_RUNTIME_STATE); |
9ded2744 | 1226 | env->kvm_vcpu_dirty = 0; |
4c0960c0 AK |
1227 | } |
1228 | ||
8c14c173 | 1229 | kvm_arch_pre_run(env, run); |
9ccfac9e JK |
1230 | if (env->exit_request) { |
1231 | DPRINTF("interrupt exit requested\n"); | |
1232 | /* | |
1233 | * KVM requires us to reenter the kernel after IO exits to complete | |
1234 | * instruction emulation. This self-signal will ensure that we | |
1235 | * leave ASAP again. | |
1236 | */ | |
1237 | qemu_cpu_kick_self(); | |
1238 | } | |
273faf1b | 1239 | cpu_single_env = NULL; |
d549db5a | 1240 | qemu_mutex_unlock_iothread(); |
9ccfac9e | 1241 | |
7cbb533f | 1242 | run_ret = kvm_vcpu_ioctl(env, KVM_RUN, 0); |
9ccfac9e | 1243 | |
d549db5a | 1244 | qemu_mutex_lock_iothread(); |
273faf1b | 1245 | cpu_single_env = env; |
05330448 AL |
1246 | kvm_arch_post_run(env, run); |
1247 | ||
b0c883b5 JK |
1248 | kvm_flush_coalesced_mmio_buffer(); |
1249 | ||
7cbb533f | 1250 | if (run_ret < 0) { |
dc77d341 JK |
1251 | if (run_ret == -EINTR || run_ret == -EAGAIN) { |
1252 | DPRINTF("io window exit\n"); | |
d73cd8f4 | 1253 | ret = EXCP_INTERRUPT; |
dc77d341 JK |
1254 | break; |
1255 | } | |
7b011fbc ME |
1256 | fprintf(stderr, "error: kvm run failed %s\n", |
1257 | strerror(-run_ret)); | |
05330448 AL |
1258 | abort(); |
1259 | } | |
1260 | ||
05330448 AL |
1261 | switch (run->exit_reason) { |
1262 | case KVM_EXIT_IO: | |
8c0d577e | 1263 | DPRINTF("handle_io\n"); |
b30e93e9 JK |
1264 | kvm_handle_io(run->io.port, |
1265 | (uint8_t *)run + run->io.data_offset, | |
1266 | run->io.direction, | |
1267 | run->io.size, | |
1268 | run->io.count); | |
d73cd8f4 | 1269 | ret = 0; |
05330448 AL |
1270 | break; |
1271 | case KVM_EXIT_MMIO: | |
8c0d577e | 1272 | DPRINTF("handle_mmio\n"); |
05330448 AL |
1273 | cpu_physical_memory_rw(run->mmio.phys_addr, |
1274 | run->mmio.data, | |
1275 | run->mmio.len, | |
1276 | run->mmio.is_write); | |
d73cd8f4 | 1277 | ret = 0; |
05330448 AL |
1278 | break; |
1279 | case KVM_EXIT_IRQ_WINDOW_OPEN: | |
8c0d577e | 1280 | DPRINTF("irq_window_open\n"); |
d73cd8f4 | 1281 | ret = EXCP_INTERRUPT; |
05330448 AL |
1282 | break; |
1283 | case KVM_EXIT_SHUTDOWN: | |
8c0d577e | 1284 | DPRINTF("shutdown\n"); |
05330448 | 1285 | qemu_system_reset_request(); |
d73cd8f4 | 1286 | ret = EXCP_INTERRUPT; |
05330448 AL |
1287 | break; |
1288 | case KVM_EXIT_UNKNOWN: | |
bb44e0d1 JK |
1289 | fprintf(stderr, "KVM: unknown exit, hardware reason %" PRIx64 "\n", |
1290 | (uint64_t)run->hw.hardware_exit_reason); | |
73aaec4a | 1291 | ret = -1; |
05330448 | 1292 | break; |
7c80eef8 | 1293 | case KVM_EXIT_INTERNAL_ERROR: |
73aaec4a | 1294 | ret = kvm_handle_internal_error(env, run); |
7c80eef8 | 1295 | break; |
05330448 | 1296 | default: |
8c0d577e | 1297 | DPRINTF("kvm_arch_handle_exit\n"); |
05330448 AL |
1298 | ret = kvm_arch_handle_exit(env, run); |
1299 | break; | |
1300 | } | |
d73cd8f4 | 1301 | } while (ret == 0); |
05330448 | 1302 | |
73aaec4a | 1303 | if (ret < 0) { |
f5c848ee | 1304 | cpu_dump_state(env, stderr, fprintf, CPU_DUMP_CODE); |
0461d5a6 | 1305 | vm_stop(RUN_STATE_INTERNAL_ERROR); |
becfc390 AL |
1306 | } |
1307 | ||
6792a57b JK |
1308 | env->exit_request = 0; |
1309 | cpu_single_env = NULL; | |
05330448 AL |
1310 | return ret; |
1311 | } | |
1312 | ||
984b5181 | 1313 | int kvm_ioctl(KVMState *s, int type, ...) |
05330448 AL |
1314 | { |
1315 | int ret; | |
984b5181 AL |
1316 | void *arg; |
1317 | va_list ap; | |
05330448 | 1318 | |
984b5181 AL |
1319 | va_start(ap, type); |
1320 | arg = va_arg(ap, void *); | |
1321 | va_end(ap); | |
1322 | ||
1323 | ret = ioctl(s->fd, type, arg); | |
a426e122 | 1324 | if (ret == -1) { |
05330448 | 1325 | ret = -errno; |
a426e122 | 1326 | } |
05330448 AL |
1327 | return ret; |
1328 | } | |
1329 | ||
984b5181 | 1330 | int kvm_vm_ioctl(KVMState *s, int type, ...) |
05330448 AL |
1331 | { |
1332 | int ret; | |
984b5181 AL |
1333 | void *arg; |
1334 | va_list ap; | |
1335 | ||
1336 | va_start(ap, type); | |
1337 | arg = va_arg(ap, void *); | |
1338 | va_end(ap); | |
05330448 | 1339 | |
984b5181 | 1340 | ret = ioctl(s->vmfd, type, arg); |
a426e122 | 1341 | if (ret == -1) { |
05330448 | 1342 | ret = -errno; |
a426e122 | 1343 | } |
05330448 AL |
1344 | return ret; |
1345 | } | |
1346 | ||
984b5181 | 1347 | int kvm_vcpu_ioctl(CPUState *env, int type, ...) |
05330448 AL |
1348 | { |
1349 | int ret; | |
984b5181 AL |
1350 | void *arg; |
1351 | va_list ap; | |
1352 | ||
1353 | va_start(ap, type); | |
1354 | arg = va_arg(ap, void *); | |
1355 | va_end(ap); | |
05330448 | 1356 | |
984b5181 | 1357 | ret = ioctl(env->kvm_fd, type, arg); |
a426e122 | 1358 | if (ret == -1) { |
05330448 | 1359 | ret = -errno; |
a426e122 | 1360 | } |
05330448 AL |
1361 | return ret; |
1362 | } | |
bd322087 AL |
1363 | |
1364 | int kvm_has_sync_mmu(void) | |
1365 | { | |
94a8d39a | 1366 | return kvm_check_extension(kvm_state, KVM_CAP_SYNC_MMU); |
bd322087 | 1367 | } |
e22a25c9 | 1368 | |
a0fb002c JK |
1369 | int kvm_has_vcpu_events(void) |
1370 | { | |
1371 | return kvm_state->vcpu_events; | |
1372 | } | |
1373 | ||
b0b1d690 JK |
1374 | int kvm_has_robust_singlestep(void) |
1375 | { | |
1376 | return kvm_state->robust_singlestep; | |
1377 | } | |
1378 | ||
ff44f1a3 JK |
1379 | int kvm_has_debugregs(void) |
1380 | { | |
1381 | return kvm_state->debugregs; | |
1382 | } | |
1383 | ||
f1665b21 SY |
1384 | int kvm_has_xsave(void) |
1385 | { | |
1386 | return kvm_state->xsave; | |
1387 | } | |
1388 | ||
1389 | int kvm_has_xcrs(void) | |
1390 | { | |
1391 | return kvm_state->xcrs; | |
1392 | } | |
1393 | ||
d2f2b8a7 SH |
1394 | int kvm_has_many_ioeventfds(void) |
1395 | { | |
1396 | if (!kvm_enabled()) { | |
1397 | return 0; | |
1398 | } | |
1399 | return kvm_state->many_ioeventfds; | |
1400 | } | |
1401 | ||
84b058d7 JK |
1402 | int kvm_has_gsi_routing(void) |
1403 | { | |
a9c5eb0d | 1404 | #ifdef KVM_CAP_IRQ_ROUTING |
84b058d7 | 1405 | return kvm_check_extension(kvm_state, KVM_CAP_IRQ_ROUTING); |
a9c5eb0d AG |
1406 | #else |
1407 | return false; | |
1408 | #endif | |
84b058d7 JK |
1409 | } |
1410 | ||
9b5b76d4 JK |
1411 | int kvm_allows_irq0_override(void) |
1412 | { | |
1413 | return !kvm_enabled() || !kvm_irqchip_in_kernel() || kvm_has_gsi_routing(); | |
1414 | } | |
1415 | ||
6f0437e8 JK |
1416 | void kvm_setup_guest_memory(void *start, size_t size) |
1417 | { | |
1418 | if (!kvm_has_sync_mmu()) { | |
e78815a5 | 1419 | int ret = qemu_madvise(start, size, QEMU_MADV_DONTFORK); |
6f0437e8 JK |
1420 | |
1421 | if (ret) { | |
e78815a5 AF |
1422 | perror("qemu_madvise"); |
1423 | fprintf(stderr, | |
1424 | "Need MADV_DONTFORK in absence of synchronous KVM MMU\n"); | |
6f0437e8 JK |
1425 | exit(1); |
1426 | } | |
6f0437e8 JK |
1427 | } |
1428 | } | |
1429 | ||
e22a25c9 AL |
1430 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
1431 | struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *env, | |
1432 | target_ulong pc) | |
1433 | { | |
1434 | struct kvm_sw_breakpoint *bp; | |
1435 | ||
72cf2d4f | 1436 | QTAILQ_FOREACH(bp, &env->kvm_state->kvm_sw_breakpoints, entry) { |
a426e122 | 1437 | if (bp->pc == pc) { |
e22a25c9 | 1438 | return bp; |
a426e122 | 1439 | } |
e22a25c9 AL |
1440 | } |
1441 | return NULL; | |
1442 | } | |
1443 | ||
1444 | int kvm_sw_breakpoints_active(CPUState *env) | |
1445 | { | |
72cf2d4f | 1446 | return !QTAILQ_EMPTY(&env->kvm_state->kvm_sw_breakpoints); |
e22a25c9 AL |
1447 | } |
1448 | ||
452e4751 GC |
1449 | struct kvm_set_guest_debug_data { |
1450 | struct kvm_guest_debug dbg; | |
1451 | CPUState *env; | |
1452 | int err; | |
1453 | }; | |
1454 | ||
1455 | static void kvm_invoke_set_guest_debug(void *data) | |
1456 | { | |
1457 | struct kvm_set_guest_debug_data *dbg_data = data; | |
b3807725 JK |
1458 | CPUState *env = dbg_data->env; |
1459 | ||
b3807725 | 1460 | dbg_data->err = kvm_vcpu_ioctl(env, KVM_SET_GUEST_DEBUG, &dbg_data->dbg); |
452e4751 GC |
1461 | } |
1462 | ||
e22a25c9 AL |
1463 | int kvm_update_guest_debug(CPUState *env, unsigned long reinject_trap) |
1464 | { | |
452e4751 | 1465 | struct kvm_set_guest_debug_data data; |
e22a25c9 | 1466 | |
b0b1d690 | 1467 | data.dbg.control = reinject_trap; |
e22a25c9 | 1468 | |
b0b1d690 JK |
1469 | if (env->singlestep_enabled) { |
1470 | data.dbg.control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP; | |
1471 | } | |
452e4751 | 1472 | kvm_arch_update_guest_debug(env, &data.dbg); |
452e4751 | 1473 | data.env = env; |
e22a25c9 | 1474 | |
be41cbe0 | 1475 | run_on_cpu(env, kvm_invoke_set_guest_debug, &data); |
452e4751 | 1476 | return data.err; |
e22a25c9 AL |
1477 | } |
1478 | ||
1479 | int kvm_insert_breakpoint(CPUState *current_env, target_ulong addr, | |
1480 | target_ulong len, int type) | |
1481 | { | |
1482 | struct kvm_sw_breakpoint *bp; | |
1483 | CPUState *env; | |
1484 | int err; | |
1485 | ||
1486 | if (type == GDB_BREAKPOINT_SW) { | |
1487 | bp = kvm_find_sw_breakpoint(current_env, addr); | |
1488 | if (bp) { | |
1489 | bp->use_count++; | |
1490 | return 0; | |
1491 | } | |
1492 | ||
7267c094 | 1493 | bp = g_malloc(sizeof(struct kvm_sw_breakpoint)); |
a426e122 | 1494 | if (!bp) { |
e22a25c9 | 1495 | return -ENOMEM; |
a426e122 | 1496 | } |
e22a25c9 AL |
1497 | |
1498 | bp->pc = addr; | |
1499 | bp->use_count = 1; | |
1500 | err = kvm_arch_insert_sw_breakpoint(current_env, bp); | |
1501 | if (err) { | |
7267c094 | 1502 | g_free(bp); |
e22a25c9 AL |
1503 | return err; |
1504 | } | |
1505 | ||
72cf2d4f | 1506 | QTAILQ_INSERT_HEAD(¤t_env->kvm_state->kvm_sw_breakpoints, |
e22a25c9 AL |
1507 | bp, entry); |
1508 | } else { | |
1509 | err = kvm_arch_insert_hw_breakpoint(addr, len, type); | |
a426e122 | 1510 | if (err) { |
e22a25c9 | 1511 | return err; |
a426e122 | 1512 | } |
e22a25c9 AL |
1513 | } |
1514 | ||
1515 | for (env = first_cpu; env != NULL; env = env->next_cpu) { | |
1516 | err = kvm_update_guest_debug(env, 0); | |
a426e122 | 1517 | if (err) { |
e22a25c9 | 1518 | return err; |
a426e122 | 1519 | } |
e22a25c9 AL |
1520 | } |
1521 | return 0; | |
1522 | } | |
1523 | ||
1524 | int kvm_remove_breakpoint(CPUState *current_env, target_ulong addr, | |
1525 | target_ulong len, int type) | |
1526 | { | |
1527 | struct kvm_sw_breakpoint *bp; | |
1528 | CPUState *env; | |
1529 | int err; | |
1530 | ||
1531 | if (type == GDB_BREAKPOINT_SW) { | |
1532 | bp = kvm_find_sw_breakpoint(current_env, addr); | |
a426e122 | 1533 | if (!bp) { |
e22a25c9 | 1534 | return -ENOENT; |
a426e122 | 1535 | } |
e22a25c9 AL |
1536 | |
1537 | if (bp->use_count > 1) { | |
1538 | bp->use_count--; | |
1539 | return 0; | |
1540 | } | |
1541 | ||
1542 | err = kvm_arch_remove_sw_breakpoint(current_env, bp); | |
a426e122 | 1543 | if (err) { |
e22a25c9 | 1544 | return err; |
a426e122 | 1545 | } |
e22a25c9 | 1546 | |
72cf2d4f | 1547 | QTAILQ_REMOVE(¤t_env->kvm_state->kvm_sw_breakpoints, bp, entry); |
7267c094 | 1548 | g_free(bp); |
e22a25c9 AL |
1549 | } else { |
1550 | err = kvm_arch_remove_hw_breakpoint(addr, len, type); | |
a426e122 | 1551 | if (err) { |
e22a25c9 | 1552 | return err; |
a426e122 | 1553 | } |
e22a25c9 AL |
1554 | } |
1555 | ||
1556 | for (env = first_cpu; env != NULL; env = env->next_cpu) { | |
1557 | err = kvm_update_guest_debug(env, 0); | |
a426e122 | 1558 | if (err) { |
e22a25c9 | 1559 | return err; |
a426e122 | 1560 | } |
e22a25c9 AL |
1561 | } |
1562 | return 0; | |
1563 | } | |
1564 | ||
1565 | void kvm_remove_all_breakpoints(CPUState *current_env) | |
1566 | { | |
1567 | struct kvm_sw_breakpoint *bp, *next; | |
1568 | KVMState *s = current_env->kvm_state; | |
1569 | CPUState *env; | |
1570 | ||
72cf2d4f | 1571 | QTAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) { |
e22a25c9 AL |
1572 | if (kvm_arch_remove_sw_breakpoint(current_env, bp) != 0) { |
1573 | /* Try harder to find a CPU that currently sees the breakpoint. */ | |
1574 | for (env = first_cpu; env != NULL; env = env->next_cpu) { | |
a426e122 | 1575 | if (kvm_arch_remove_sw_breakpoint(env, bp) == 0) { |
e22a25c9 | 1576 | break; |
a426e122 | 1577 | } |
e22a25c9 AL |
1578 | } |
1579 | } | |
1580 | } | |
1581 | kvm_arch_remove_all_hw_breakpoints(); | |
1582 | ||
a426e122 | 1583 | for (env = first_cpu; env != NULL; env = env->next_cpu) { |
e22a25c9 | 1584 | kvm_update_guest_debug(env, 0); |
a426e122 | 1585 | } |
e22a25c9 AL |
1586 | } |
1587 | ||
1588 | #else /* !KVM_CAP_SET_GUEST_DEBUG */ | |
1589 | ||
1590 | int kvm_update_guest_debug(CPUState *env, unsigned long reinject_trap) | |
1591 | { | |
1592 | return -EINVAL; | |
1593 | } | |
1594 | ||
1595 | int kvm_insert_breakpoint(CPUState *current_env, target_ulong addr, | |
1596 | target_ulong len, int type) | |
1597 | { | |
1598 | return -EINVAL; | |
1599 | } | |
1600 | ||
1601 | int kvm_remove_breakpoint(CPUState *current_env, target_ulong addr, | |
1602 | target_ulong len, int type) | |
1603 | { | |
1604 | return -EINVAL; | |
1605 | } | |
1606 | ||
1607 | void kvm_remove_all_breakpoints(CPUState *current_env) | |
1608 | { | |
1609 | } | |
1610 | #endif /* !KVM_CAP_SET_GUEST_DEBUG */ | |
cc84de95 MT |
1611 | |
1612 | int kvm_set_signal_mask(CPUState *env, const sigset_t *sigset) | |
1613 | { | |
1614 | struct kvm_signal_mask *sigmask; | |
1615 | int r; | |
1616 | ||
a426e122 | 1617 | if (!sigset) { |
cc84de95 | 1618 | return kvm_vcpu_ioctl(env, KVM_SET_SIGNAL_MASK, NULL); |
a426e122 | 1619 | } |
cc84de95 | 1620 | |
7267c094 | 1621 | sigmask = g_malloc(sizeof(*sigmask) + sizeof(*sigset)); |
cc84de95 MT |
1622 | |
1623 | sigmask->len = 8; | |
1624 | memcpy(sigmask->sigset, sigset, sizeof(*sigset)); | |
1625 | r = kvm_vcpu_ioctl(env, KVM_SET_SIGNAL_MASK, sigmask); | |
7267c094 | 1626 | g_free(sigmask); |
cc84de95 MT |
1627 | |
1628 | return r; | |
1629 | } | |
ca821806 | 1630 | |
44f1a3d8 CM |
1631 | int kvm_set_ioeventfd_mmio_long(int fd, uint32_t addr, uint32_t val, bool assign) |
1632 | { | |
44f1a3d8 CM |
1633 | int ret; |
1634 | struct kvm_ioeventfd iofd; | |
1635 | ||
1636 | iofd.datamatch = val; | |
1637 | iofd.addr = addr; | |
1638 | iofd.len = 4; | |
1639 | iofd.flags = KVM_IOEVENTFD_FLAG_DATAMATCH; | |
1640 | iofd.fd = fd; | |
1641 | ||
1642 | if (!kvm_enabled()) { | |
1643 | return -ENOSYS; | |
1644 | } | |
1645 | ||
1646 | if (!assign) { | |
1647 | iofd.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN; | |
1648 | } | |
1649 | ||
1650 | ret = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &iofd); | |
1651 | ||
1652 | if (ret < 0) { | |
1653 | return -errno; | |
1654 | } | |
1655 | ||
1656 | return 0; | |
44f1a3d8 CM |
1657 | } |
1658 | ||
ca821806 MT |
1659 | int kvm_set_ioeventfd_pio_word(int fd, uint16_t addr, uint16_t val, bool assign) |
1660 | { | |
1661 | struct kvm_ioeventfd kick = { | |
1662 | .datamatch = val, | |
1663 | .addr = addr, | |
1664 | .len = 2, | |
1665 | .flags = KVM_IOEVENTFD_FLAG_DATAMATCH | KVM_IOEVENTFD_FLAG_PIO, | |
1666 | .fd = fd, | |
1667 | }; | |
1668 | int r; | |
a426e122 | 1669 | if (!kvm_enabled()) { |
ca821806 | 1670 | return -ENOSYS; |
a426e122 JK |
1671 | } |
1672 | if (!assign) { | |
ca821806 | 1673 | kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN; |
a426e122 | 1674 | } |
ca821806 | 1675 | r = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick); |
a426e122 | 1676 | if (r < 0) { |
ca821806 | 1677 | return r; |
a426e122 | 1678 | } |
ca821806 | 1679 | return 0; |
98c8573e | 1680 | } |
a1b87fe0 JK |
1681 | |
1682 | int kvm_on_sigbus_vcpu(CPUState *env, int code, void *addr) | |
1683 | { | |
1684 | return kvm_arch_on_sigbus_vcpu(env, code, addr); | |
1685 | } | |
1686 | ||
1687 | int kvm_on_sigbus(int code, void *addr) | |
1688 | { | |
1689 | return kvm_arch_on_sigbus(code, addr); | |
1690 | } |