]>
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 | |
6f725c13 | 78 | int pit_in_kernel; |
8a7c7393 | 79 | int pit_state2; |
f1665b21 | 80 | int xsave, xcrs; |
d2f2b8a7 | 81 | int many_ioeventfds; |
92e4b519 DG |
82 | /* The man page (and posix) say ioctl numbers are signed int, but |
83 | * they're not. Linux, glibc and *BSD all treat ioctl numbers as | |
84 | * unsigned, and treating them as signed here can break things */ | |
85 | unsigned irqchip_inject_ioctl; | |
84b058d7 JK |
86 | #ifdef KVM_CAP_IRQ_ROUTING |
87 | struct kvm_irq_routing *irq_routes; | |
88 | int nr_allocated_irq_routes; | |
89 | uint32_t *used_gsi_bitmap; | |
90 | unsigned int max_gsi; | |
91 | #endif | |
05330448 AL |
92 | }; |
93 | ||
6a7af8cb | 94 | KVMState *kvm_state; |
3d4b2649 | 95 | bool kvm_kernel_irqchip; |
05330448 | 96 | |
94a8d39a JK |
97 | static const KVMCapabilityInfo kvm_required_capabilites[] = { |
98 | KVM_CAP_INFO(USER_MEMORY), | |
99 | KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS), | |
100 | KVM_CAP_LAST_INFO | |
101 | }; | |
102 | ||
05330448 AL |
103 | static KVMSlot *kvm_alloc_slot(KVMState *s) |
104 | { | |
105 | int i; | |
106 | ||
107 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
a426e122 | 108 | if (s->slots[i].memory_size == 0) { |
05330448 | 109 | return &s->slots[i]; |
a426e122 | 110 | } |
05330448 AL |
111 | } |
112 | ||
d3f8d37f AL |
113 | fprintf(stderr, "%s: no free slot available\n", __func__); |
114 | abort(); | |
115 | } | |
116 | ||
117 | static KVMSlot *kvm_lookup_matching_slot(KVMState *s, | |
c227f099 AL |
118 | target_phys_addr_t start_addr, |
119 | target_phys_addr_t end_addr) | |
d3f8d37f AL |
120 | { |
121 | int i; | |
122 | ||
123 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
124 | KVMSlot *mem = &s->slots[i]; | |
125 | ||
126 | if (start_addr == mem->start_addr && | |
127 | end_addr == mem->start_addr + mem->memory_size) { | |
128 | return mem; | |
129 | } | |
130 | } | |
131 | ||
05330448 AL |
132 | return NULL; |
133 | } | |
134 | ||
6152e2ae AL |
135 | /* |
136 | * Find overlapping slot with lowest start address | |
137 | */ | |
138 | static KVMSlot *kvm_lookup_overlapping_slot(KVMState *s, | |
c227f099 AL |
139 | target_phys_addr_t start_addr, |
140 | target_phys_addr_t end_addr) | |
05330448 | 141 | { |
6152e2ae | 142 | KVMSlot *found = NULL; |
05330448 AL |
143 | int i; |
144 | ||
145 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
146 | KVMSlot *mem = &s->slots[i]; | |
147 | ||
6152e2ae AL |
148 | if (mem->memory_size == 0 || |
149 | (found && found->start_addr < mem->start_addr)) { | |
150 | continue; | |
151 | } | |
152 | ||
153 | if (end_addr > mem->start_addr && | |
154 | start_addr < mem->start_addr + mem->memory_size) { | |
155 | found = mem; | |
156 | } | |
05330448 AL |
157 | } |
158 | ||
6152e2ae | 159 | return found; |
05330448 AL |
160 | } |
161 | ||
9f213ed9 AK |
162 | int kvm_physical_memory_addr_from_host(KVMState *s, void *ram, |
163 | target_phys_addr_t *phys_addr) | |
983dfc3b HY |
164 | { |
165 | int i; | |
166 | ||
167 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
168 | KVMSlot *mem = &s->slots[i]; | |
169 | ||
9f213ed9 AK |
170 | if (ram >= mem->ram && ram < mem->ram + mem->memory_size) { |
171 | *phys_addr = mem->start_addr + (ram - mem->ram); | |
983dfc3b HY |
172 | return 1; |
173 | } | |
174 | } | |
175 | ||
176 | return 0; | |
177 | } | |
178 | ||
5832d1f2 AL |
179 | static int kvm_set_user_memory_region(KVMState *s, KVMSlot *slot) |
180 | { | |
181 | struct kvm_userspace_memory_region mem; | |
182 | ||
183 | mem.slot = slot->slot; | |
184 | mem.guest_phys_addr = slot->start_addr; | |
185 | mem.memory_size = slot->memory_size; | |
9f213ed9 | 186 | mem.userspace_addr = (unsigned long)slot->ram; |
5832d1f2 | 187 | mem.flags = slot->flags; |
4495d6a7 JK |
188 | if (s->migration_log) { |
189 | mem.flags |= KVM_MEM_LOG_DIRTY_PAGES; | |
190 | } | |
5832d1f2 AL |
191 | return kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem); |
192 | } | |
193 | ||
8d2ba1fb JK |
194 | static void kvm_reset_vcpu(void *opaque) |
195 | { | |
9349b4f9 | 196 | CPUArchState *env = opaque; |
8d2ba1fb | 197 | |
caa5af0f | 198 | kvm_arch_reset_vcpu(env); |
8d2ba1fb | 199 | } |
5832d1f2 | 200 | |
6f725c13 GC |
201 | int kvm_pit_in_kernel(void) |
202 | { | |
203 | return kvm_state->pit_in_kernel; | |
204 | } | |
205 | ||
9349b4f9 | 206 | int kvm_init_vcpu(CPUArchState *env) |
05330448 AL |
207 | { |
208 | KVMState *s = kvm_state; | |
209 | long mmap_size; | |
210 | int ret; | |
211 | ||
8c0d577e | 212 | DPRINTF("kvm_init_vcpu\n"); |
05330448 | 213 | |
984b5181 | 214 | ret = kvm_vm_ioctl(s, KVM_CREATE_VCPU, env->cpu_index); |
05330448 | 215 | if (ret < 0) { |
8c0d577e | 216 | DPRINTF("kvm_create_vcpu failed\n"); |
05330448 AL |
217 | goto err; |
218 | } | |
219 | ||
220 | env->kvm_fd = ret; | |
221 | env->kvm_state = s; | |
d841b6c4 | 222 | env->kvm_vcpu_dirty = 1; |
05330448 AL |
223 | |
224 | mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0); | |
225 | if (mmap_size < 0) { | |
748a680b | 226 | ret = mmap_size; |
8c0d577e | 227 | DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n"); |
05330448 AL |
228 | goto err; |
229 | } | |
230 | ||
231 | env->kvm_run = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED, | |
232 | env->kvm_fd, 0); | |
233 | if (env->kvm_run == MAP_FAILED) { | |
234 | ret = -errno; | |
8c0d577e | 235 | DPRINTF("mmap'ing vcpu state failed\n"); |
05330448 AL |
236 | goto err; |
237 | } | |
238 | ||
a426e122 JK |
239 | if (s->coalesced_mmio && !s->coalesced_mmio_ring) { |
240 | s->coalesced_mmio_ring = | |
241 | (void *)env->kvm_run + s->coalesced_mmio * PAGE_SIZE; | |
242 | } | |
62a2744c | 243 | |
05330448 | 244 | ret = kvm_arch_init_vcpu(env); |
8d2ba1fb | 245 | if (ret == 0) { |
a08d4367 | 246 | qemu_register_reset(kvm_reset_vcpu, env); |
caa5af0f | 247 | kvm_arch_reset_vcpu(env); |
8d2ba1fb | 248 | } |
05330448 AL |
249 | err: |
250 | return ret; | |
251 | } | |
252 | ||
5832d1f2 AL |
253 | /* |
254 | * dirty pages logging control | |
255 | */ | |
25254bbc MT |
256 | |
257 | static int kvm_mem_flags(KVMState *s, bool log_dirty) | |
258 | { | |
259 | return log_dirty ? KVM_MEM_LOG_DIRTY_PAGES : 0; | |
260 | } | |
261 | ||
262 | static int kvm_slot_dirty_pages_log_change(KVMSlot *mem, bool log_dirty) | |
5832d1f2 AL |
263 | { |
264 | KVMState *s = kvm_state; | |
25254bbc | 265 | int flags, mask = KVM_MEM_LOG_DIRTY_PAGES; |
4495d6a7 JK |
266 | int old_flags; |
267 | ||
4495d6a7 | 268 | old_flags = mem->flags; |
5832d1f2 | 269 | |
25254bbc | 270 | flags = (mem->flags & ~mask) | kvm_mem_flags(s, log_dirty); |
5832d1f2 AL |
271 | mem->flags = flags; |
272 | ||
4495d6a7 JK |
273 | /* If nothing changed effectively, no need to issue ioctl */ |
274 | if (s->migration_log) { | |
275 | flags |= KVM_MEM_LOG_DIRTY_PAGES; | |
276 | } | |
25254bbc | 277 | |
4495d6a7 | 278 | if (flags == old_flags) { |
25254bbc | 279 | return 0; |
4495d6a7 JK |
280 | } |
281 | ||
5832d1f2 AL |
282 | return kvm_set_user_memory_region(s, mem); |
283 | } | |
284 | ||
25254bbc MT |
285 | static int kvm_dirty_pages_log_change(target_phys_addr_t phys_addr, |
286 | ram_addr_t size, bool log_dirty) | |
287 | { | |
288 | KVMState *s = kvm_state; | |
289 | KVMSlot *mem = kvm_lookup_matching_slot(s, phys_addr, phys_addr + size); | |
290 | ||
291 | if (mem == NULL) { | |
292 | fprintf(stderr, "BUG: %s: invalid parameters " TARGET_FMT_plx "-" | |
293 | TARGET_FMT_plx "\n", __func__, phys_addr, | |
294 | (target_phys_addr_t)(phys_addr + size - 1)); | |
295 | return -EINVAL; | |
296 | } | |
297 | return kvm_slot_dirty_pages_log_change(mem, log_dirty); | |
298 | } | |
299 | ||
a01672d3 AK |
300 | static void kvm_log_start(MemoryListener *listener, |
301 | MemoryRegionSection *section) | |
5832d1f2 | 302 | { |
a01672d3 AK |
303 | int r; |
304 | ||
305 | r = kvm_dirty_pages_log_change(section->offset_within_address_space, | |
306 | section->size, true); | |
307 | if (r < 0) { | |
308 | abort(); | |
309 | } | |
5832d1f2 AL |
310 | } |
311 | ||
a01672d3 AK |
312 | static void kvm_log_stop(MemoryListener *listener, |
313 | MemoryRegionSection *section) | |
5832d1f2 | 314 | { |
a01672d3 AK |
315 | int r; |
316 | ||
317 | r = kvm_dirty_pages_log_change(section->offset_within_address_space, | |
318 | section->size, false); | |
319 | if (r < 0) { | |
320 | abort(); | |
321 | } | |
5832d1f2 AL |
322 | } |
323 | ||
7b8f3b78 | 324 | static int kvm_set_migration_log(int enable) |
4495d6a7 JK |
325 | { |
326 | KVMState *s = kvm_state; | |
327 | KVMSlot *mem; | |
328 | int i, err; | |
329 | ||
330 | s->migration_log = enable; | |
331 | ||
332 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
333 | mem = &s->slots[i]; | |
334 | ||
70fedd76 AW |
335 | if (!mem->memory_size) { |
336 | continue; | |
337 | } | |
4495d6a7 JK |
338 | if (!!(mem->flags & KVM_MEM_LOG_DIRTY_PAGES) == enable) { |
339 | continue; | |
340 | } | |
341 | err = kvm_set_user_memory_region(s, mem); | |
342 | if (err) { | |
343 | return err; | |
344 | } | |
345 | } | |
346 | return 0; | |
347 | } | |
348 | ||
8369e01c | 349 | /* get kvm's dirty pages bitmap and update qemu's */ |
ffcde12f AK |
350 | static int kvm_get_dirty_pages_log_range(MemoryRegionSection *section, |
351 | unsigned long *bitmap) | |
96c1606b | 352 | { |
8369e01c | 353 | unsigned int i, j; |
aa90fec7 BH |
354 | unsigned long page_number, c; |
355 | target_phys_addr_t addr, addr1; | |
ffcde12f | 356 | unsigned int len = ((section->size / TARGET_PAGE_SIZE) + HOST_LONG_BITS - 1) / HOST_LONG_BITS; |
8369e01c MT |
357 | |
358 | /* | |
359 | * bitmap-traveling is faster than memory-traveling (for addr...) | |
360 | * especially when most of the memory is not dirty. | |
361 | */ | |
362 | for (i = 0; i < len; i++) { | |
363 | if (bitmap[i] != 0) { | |
364 | c = leul_to_cpu(bitmap[i]); | |
365 | do { | |
366 | j = ffsl(c) - 1; | |
367 | c &= ~(1ul << j); | |
368 | page_number = i * HOST_LONG_BITS + j; | |
369 | addr1 = page_number * TARGET_PAGE_SIZE; | |
ffcde12f | 370 | addr = section->offset_within_region + addr1; |
fd4aa979 | 371 | memory_region_set_dirty(section->mr, addr, TARGET_PAGE_SIZE); |
8369e01c MT |
372 | } while (c != 0); |
373 | } | |
374 | } | |
375 | return 0; | |
96c1606b AG |
376 | } |
377 | ||
8369e01c MT |
378 | #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1)) |
379 | ||
5832d1f2 AL |
380 | /** |
381 | * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space | |
fd4aa979 BS |
382 | * This function updates qemu's dirty bitmap using |
383 | * memory_region_set_dirty(). This means all bits are set | |
384 | * to dirty. | |
5832d1f2 | 385 | * |
d3f8d37f | 386 | * @start_add: start of logged region. |
5832d1f2 AL |
387 | * @end_addr: end of logged region. |
388 | */ | |
ffcde12f | 389 | static int kvm_physical_sync_dirty_bitmap(MemoryRegionSection *section) |
5832d1f2 AL |
390 | { |
391 | KVMState *s = kvm_state; | |
151f7749 | 392 | unsigned long size, allocated_size = 0; |
151f7749 JK |
393 | KVMDirtyLog d; |
394 | KVMSlot *mem; | |
395 | int ret = 0; | |
ffcde12f AK |
396 | target_phys_addr_t start_addr = section->offset_within_address_space; |
397 | target_phys_addr_t end_addr = start_addr + section->size; | |
5832d1f2 | 398 | |
151f7749 JK |
399 | d.dirty_bitmap = NULL; |
400 | while (start_addr < end_addr) { | |
401 | mem = kvm_lookup_overlapping_slot(s, start_addr, end_addr); | |
402 | if (mem == NULL) { | |
403 | break; | |
404 | } | |
5832d1f2 | 405 | |
51b0c606 MT |
406 | /* XXX bad kernel interface alert |
407 | * For dirty bitmap, kernel allocates array of size aligned to | |
408 | * bits-per-long. But for case when the kernel is 64bits and | |
409 | * the userspace is 32bits, userspace can't align to the same | |
410 | * bits-per-long, since sizeof(long) is different between kernel | |
411 | * and user space. This way, userspace will provide buffer which | |
412 | * may be 4 bytes less than the kernel will use, resulting in | |
413 | * userspace memory corruption (which is not detectable by valgrind | |
414 | * too, in most cases). | |
415 | * So for now, let's align to 64 instead of HOST_LONG_BITS here, in | |
416 | * a hope that sizeof(long) wont become >8 any time soon. | |
417 | */ | |
418 | size = ALIGN(((mem->memory_size) >> TARGET_PAGE_BITS), | |
419 | /*HOST_LONG_BITS*/ 64) / 8; | |
151f7749 | 420 | if (!d.dirty_bitmap) { |
7267c094 | 421 | d.dirty_bitmap = g_malloc(size); |
151f7749 | 422 | } else if (size > allocated_size) { |
7267c094 | 423 | d.dirty_bitmap = g_realloc(d.dirty_bitmap, size); |
151f7749 JK |
424 | } |
425 | allocated_size = size; | |
426 | memset(d.dirty_bitmap, 0, allocated_size); | |
5832d1f2 | 427 | |
151f7749 | 428 | d.slot = mem->slot; |
5832d1f2 | 429 | |
6e489f3f | 430 | if (kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d) == -1) { |
8c0d577e | 431 | DPRINTF("ioctl failed %d\n", errno); |
151f7749 JK |
432 | ret = -1; |
433 | break; | |
434 | } | |
5832d1f2 | 435 | |
ffcde12f | 436 | kvm_get_dirty_pages_log_range(section, d.dirty_bitmap); |
8369e01c | 437 | start_addr = mem->start_addr + mem->memory_size; |
5832d1f2 | 438 | } |
7267c094 | 439 | g_free(d.dirty_bitmap); |
151f7749 JK |
440 | |
441 | return ret; | |
5832d1f2 AL |
442 | } |
443 | ||
c227f099 | 444 | int kvm_coalesce_mmio_region(target_phys_addr_t start, ram_addr_t size) |
f65ed4c1 AL |
445 | { |
446 | int ret = -ENOSYS; | |
f65ed4c1 AL |
447 | KVMState *s = kvm_state; |
448 | ||
449 | if (s->coalesced_mmio) { | |
450 | struct kvm_coalesced_mmio_zone zone; | |
451 | ||
452 | zone.addr = start; | |
453 | zone.size = size; | |
7e680753 | 454 | zone.pad = 0; |
f65ed4c1 AL |
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; | |
7e680753 | 472 | zone.pad = 0; |
f65ed4c1 AL |
473 | |
474 | ret = kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone); | |
475 | } | |
f65ed4c1 AL |
476 | |
477 | return ret; | |
478 | } | |
479 | ||
ad7b8b33 AL |
480 | int kvm_check_extension(KVMState *s, unsigned int extension) |
481 | { | |
482 | int ret; | |
483 | ||
484 | ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension); | |
485 | if (ret < 0) { | |
486 | ret = 0; | |
487 | } | |
488 | ||
489 | return ret; | |
490 | } | |
491 | ||
d2f2b8a7 SH |
492 | static int kvm_check_many_ioeventfds(void) |
493 | { | |
d0dcac83 SH |
494 | /* Userspace can use ioeventfd for io notification. This requires a host |
495 | * that supports eventfd(2) and an I/O thread; since eventfd does not | |
496 | * support SIGIO it cannot interrupt the vcpu. | |
497 | * | |
498 | * Older kernels have a 6 device limit on the KVM io bus. Find out so we | |
d2f2b8a7 SH |
499 | * can avoid creating too many ioeventfds. |
500 | */ | |
12d4536f | 501 | #if defined(CONFIG_EVENTFD) |
d2f2b8a7 SH |
502 | int ioeventfds[7]; |
503 | int i, ret = 0; | |
504 | for (i = 0; i < ARRAY_SIZE(ioeventfds); i++) { | |
505 | ioeventfds[i] = eventfd(0, EFD_CLOEXEC); | |
506 | if (ioeventfds[i] < 0) { | |
507 | break; | |
508 | } | |
509 | ret = kvm_set_ioeventfd_pio_word(ioeventfds[i], 0, i, true); | |
510 | if (ret < 0) { | |
511 | close(ioeventfds[i]); | |
512 | break; | |
513 | } | |
514 | } | |
515 | ||
516 | /* Decide whether many devices are supported or not */ | |
517 | ret = i == ARRAY_SIZE(ioeventfds); | |
518 | ||
519 | while (i-- > 0) { | |
520 | kvm_set_ioeventfd_pio_word(ioeventfds[i], 0, i, false); | |
521 | close(ioeventfds[i]); | |
522 | } | |
523 | return ret; | |
524 | #else | |
525 | return 0; | |
526 | #endif | |
527 | } | |
528 | ||
94a8d39a JK |
529 | static const KVMCapabilityInfo * |
530 | kvm_check_extension_list(KVMState *s, const KVMCapabilityInfo *list) | |
531 | { | |
532 | while (list->name) { | |
533 | if (!kvm_check_extension(s, list->value)) { | |
534 | return list; | |
535 | } | |
536 | list++; | |
537 | } | |
538 | return NULL; | |
539 | } | |
540 | ||
a01672d3 | 541 | static void kvm_set_phys_mem(MemoryRegionSection *section, bool add) |
46dbef6a MT |
542 | { |
543 | KVMState *s = kvm_state; | |
46dbef6a MT |
544 | KVMSlot *mem, old; |
545 | int err; | |
a01672d3 AK |
546 | MemoryRegion *mr = section->mr; |
547 | bool log_dirty = memory_region_is_logging(mr); | |
548 | target_phys_addr_t start_addr = section->offset_within_address_space; | |
549 | ram_addr_t size = section->size; | |
9f213ed9 | 550 | void *ram = NULL; |
8f6f962b | 551 | unsigned delta; |
46dbef6a | 552 | |
14542fea GN |
553 | /* kvm works in page size chunks, but the function may be called |
554 | with sub-page size and unaligned start address. */ | |
8f6f962b AK |
555 | delta = TARGET_PAGE_ALIGN(size) - size; |
556 | if (delta > size) { | |
557 | return; | |
558 | } | |
559 | start_addr += delta; | |
560 | size -= delta; | |
561 | size &= TARGET_PAGE_MASK; | |
562 | if (!size || (start_addr & ~TARGET_PAGE_MASK)) { | |
563 | return; | |
564 | } | |
46dbef6a | 565 | |
a01672d3 AK |
566 | if (!memory_region_is_ram(mr)) { |
567 | return; | |
9f213ed9 AK |
568 | } |
569 | ||
8f6f962b | 570 | ram = memory_region_get_ram_ptr(mr) + section->offset_within_region + delta; |
a01672d3 | 571 | |
46dbef6a MT |
572 | while (1) { |
573 | mem = kvm_lookup_overlapping_slot(s, start_addr, start_addr + size); | |
574 | if (!mem) { | |
575 | break; | |
576 | } | |
577 | ||
a01672d3 | 578 | if (add && start_addr >= mem->start_addr && |
46dbef6a | 579 | (start_addr + size <= mem->start_addr + mem->memory_size) && |
9f213ed9 | 580 | (ram - start_addr == mem->ram - mem->start_addr)) { |
46dbef6a | 581 | /* The new slot fits into the existing one and comes with |
25254bbc MT |
582 | * identical parameters - update flags and done. */ |
583 | kvm_slot_dirty_pages_log_change(mem, log_dirty); | |
46dbef6a MT |
584 | return; |
585 | } | |
586 | ||
587 | old = *mem; | |
588 | ||
3fbffb62 AK |
589 | if (mem->flags & KVM_MEM_LOG_DIRTY_PAGES) { |
590 | kvm_physical_sync_dirty_bitmap(section); | |
591 | } | |
592 | ||
46dbef6a MT |
593 | /* unregister the overlapping slot */ |
594 | mem->memory_size = 0; | |
595 | err = kvm_set_user_memory_region(s, mem); | |
596 | if (err) { | |
597 | fprintf(stderr, "%s: error unregistering overlapping slot: %s\n", | |
598 | __func__, strerror(-err)); | |
599 | abort(); | |
600 | } | |
601 | ||
602 | /* Workaround for older KVM versions: we can't join slots, even not by | |
603 | * unregistering the previous ones and then registering the larger | |
604 | * slot. We have to maintain the existing fragmentation. Sigh. | |
605 | * | |
606 | * This workaround assumes that the new slot starts at the same | |
607 | * address as the first existing one. If not or if some overlapping | |
608 | * slot comes around later, we will fail (not seen in practice so far) | |
609 | * - and actually require a recent KVM version. */ | |
610 | if (s->broken_set_mem_region && | |
a01672d3 | 611 | old.start_addr == start_addr && old.memory_size < size && add) { |
46dbef6a MT |
612 | mem = kvm_alloc_slot(s); |
613 | mem->memory_size = old.memory_size; | |
614 | mem->start_addr = old.start_addr; | |
9f213ed9 | 615 | mem->ram = old.ram; |
25254bbc | 616 | mem->flags = kvm_mem_flags(s, log_dirty); |
46dbef6a MT |
617 | |
618 | err = kvm_set_user_memory_region(s, mem); | |
619 | if (err) { | |
620 | fprintf(stderr, "%s: error updating slot: %s\n", __func__, | |
621 | strerror(-err)); | |
622 | abort(); | |
623 | } | |
624 | ||
625 | start_addr += old.memory_size; | |
9f213ed9 | 626 | ram += old.memory_size; |
46dbef6a MT |
627 | size -= old.memory_size; |
628 | continue; | |
629 | } | |
630 | ||
631 | /* register prefix slot */ | |
632 | if (old.start_addr < start_addr) { | |
633 | mem = kvm_alloc_slot(s); | |
634 | mem->memory_size = start_addr - old.start_addr; | |
635 | mem->start_addr = old.start_addr; | |
9f213ed9 | 636 | mem->ram = old.ram; |
25254bbc | 637 | mem->flags = kvm_mem_flags(s, log_dirty); |
46dbef6a MT |
638 | |
639 | err = kvm_set_user_memory_region(s, mem); | |
640 | if (err) { | |
641 | fprintf(stderr, "%s: error registering prefix slot: %s\n", | |
642 | __func__, strerror(-err)); | |
d4d6868f AG |
643 | #ifdef TARGET_PPC |
644 | fprintf(stderr, "%s: This is probably because your kernel's " \ | |
645 | "PAGE_SIZE is too big. Please try to use 4k " \ | |
646 | "PAGE_SIZE!\n", __func__); | |
647 | #endif | |
46dbef6a MT |
648 | abort(); |
649 | } | |
650 | } | |
651 | ||
652 | /* register suffix slot */ | |
653 | if (old.start_addr + old.memory_size > start_addr + size) { | |
654 | ram_addr_t size_delta; | |
655 | ||
656 | mem = kvm_alloc_slot(s); | |
657 | mem->start_addr = start_addr + size; | |
658 | size_delta = mem->start_addr - old.start_addr; | |
659 | mem->memory_size = old.memory_size - size_delta; | |
9f213ed9 | 660 | mem->ram = old.ram + size_delta; |
25254bbc | 661 | mem->flags = kvm_mem_flags(s, log_dirty); |
46dbef6a MT |
662 | |
663 | err = kvm_set_user_memory_region(s, mem); | |
664 | if (err) { | |
665 | fprintf(stderr, "%s: error registering suffix slot: %s\n", | |
666 | __func__, strerror(-err)); | |
667 | abort(); | |
668 | } | |
669 | } | |
670 | } | |
671 | ||
672 | /* in case the KVM bug workaround already "consumed" the new slot */ | |
a426e122 | 673 | if (!size) { |
46dbef6a | 674 | return; |
a426e122 | 675 | } |
a01672d3 | 676 | if (!add) { |
46dbef6a | 677 | return; |
a426e122 | 678 | } |
46dbef6a MT |
679 | mem = kvm_alloc_slot(s); |
680 | mem->memory_size = size; | |
681 | mem->start_addr = start_addr; | |
9f213ed9 | 682 | mem->ram = ram; |
25254bbc | 683 | mem->flags = kvm_mem_flags(s, log_dirty); |
46dbef6a MT |
684 | |
685 | err = kvm_set_user_memory_region(s, mem); | |
686 | if (err) { | |
687 | fprintf(stderr, "%s: error registering slot: %s\n", __func__, | |
688 | strerror(-err)); | |
689 | abort(); | |
690 | } | |
691 | } | |
692 | ||
50c1e149 AK |
693 | static void kvm_begin(MemoryListener *listener) |
694 | { | |
695 | } | |
696 | ||
697 | static void kvm_commit(MemoryListener *listener) | |
698 | { | |
699 | } | |
700 | ||
a01672d3 AK |
701 | static void kvm_region_add(MemoryListener *listener, |
702 | MemoryRegionSection *section) | |
703 | { | |
704 | kvm_set_phys_mem(section, true); | |
705 | } | |
706 | ||
707 | static void kvm_region_del(MemoryListener *listener, | |
708 | MemoryRegionSection *section) | |
709 | { | |
710 | kvm_set_phys_mem(section, false); | |
711 | } | |
712 | ||
50c1e149 AK |
713 | static void kvm_region_nop(MemoryListener *listener, |
714 | MemoryRegionSection *section) | |
715 | { | |
716 | } | |
717 | ||
a01672d3 AK |
718 | static void kvm_log_sync(MemoryListener *listener, |
719 | MemoryRegionSection *section) | |
7b8f3b78 | 720 | { |
a01672d3 AK |
721 | int r; |
722 | ||
ffcde12f | 723 | r = kvm_physical_sync_dirty_bitmap(section); |
a01672d3 AK |
724 | if (r < 0) { |
725 | abort(); | |
726 | } | |
7b8f3b78 MT |
727 | } |
728 | ||
a01672d3 | 729 | static void kvm_log_global_start(struct MemoryListener *listener) |
7b8f3b78 | 730 | { |
a01672d3 AK |
731 | int r; |
732 | ||
733 | r = kvm_set_migration_log(1); | |
734 | assert(r >= 0); | |
7b8f3b78 MT |
735 | } |
736 | ||
a01672d3 | 737 | static void kvm_log_global_stop(struct MemoryListener *listener) |
7b8f3b78 | 738 | { |
a01672d3 AK |
739 | int r; |
740 | ||
741 | r = kvm_set_migration_log(0); | |
742 | assert(r >= 0); | |
7b8f3b78 MT |
743 | } |
744 | ||
80a1ea37 AK |
745 | static void kvm_mem_ioeventfd_add(MemoryRegionSection *section, |
746 | bool match_data, uint64_t data, int fd) | |
747 | { | |
748 | int r; | |
749 | ||
750 | assert(match_data && section->size == 4); | |
751 | ||
752 | r = kvm_set_ioeventfd_mmio_long(fd, section->offset_within_address_space, | |
753 | data, true); | |
754 | if (r < 0) { | |
755 | abort(); | |
756 | } | |
757 | } | |
758 | ||
759 | static void kvm_mem_ioeventfd_del(MemoryRegionSection *section, | |
760 | bool match_data, uint64_t data, int fd) | |
761 | { | |
762 | int r; | |
763 | ||
764 | r = kvm_set_ioeventfd_mmio_long(fd, section->offset_within_address_space, | |
765 | data, false); | |
766 | if (r < 0) { | |
767 | abort(); | |
768 | } | |
769 | } | |
770 | ||
771 | static void kvm_io_ioeventfd_add(MemoryRegionSection *section, | |
772 | bool match_data, uint64_t data, int fd) | |
773 | { | |
774 | int r; | |
775 | ||
776 | assert(match_data && section->size == 2); | |
777 | ||
778 | r = kvm_set_ioeventfd_pio_word(fd, section->offset_within_address_space, | |
779 | data, true); | |
780 | if (r < 0) { | |
781 | abort(); | |
782 | } | |
783 | } | |
784 | ||
785 | static void kvm_io_ioeventfd_del(MemoryRegionSection *section, | |
786 | bool match_data, uint64_t data, int fd) | |
787 | ||
788 | { | |
789 | int r; | |
790 | ||
791 | r = kvm_set_ioeventfd_pio_word(fd, section->offset_within_address_space, | |
792 | data, false); | |
793 | if (r < 0) { | |
794 | abort(); | |
795 | } | |
796 | } | |
797 | ||
798 | static void kvm_eventfd_add(MemoryListener *listener, | |
799 | MemoryRegionSection *section, | |
800 | bool match_data, uint64_t data, int fd) | |
801 | { | |
802 | if (section->address_space == get_system_memory()) { | |
803 | kvm_mem_ioeventfd_add(section, match_data, data, fd); | |
804 | } else { | |
805 | kvm_io_ioeventfd_add(section, match_data, data, fd); | |
806 | } | |
807 | } | |
808 | ||
809 | static void kvm_eventfd_del(MemoryListener *listener, | |
810 | MemoryRegionSection *section, | |
811 | bool match_data, uint64_t data, int fd) | |
812 | { | |
813 | if (section->address_space == get_system_memory()) { | |
814 | kvm_mem_ioeventfd_del(section, match_data, data, fd); | |
815 | } else { | |
816 | kvm_io_ioeventfd_del(section, match_data, data, fd); | |
817 | } | |
818 | } | |
819 | ||
a01672d3 | 820 | static MemoryListener kvm_memory_listener = { |
50c1e149 AK |
821 | .begin = kvm_begin, |
822 | .commit = kvm_commit, | |
a01672d3 AK |
823 | .region_add = kvm_region_add, |
824 | .region_del = kvm_region_del, | |
50c1e149 | 825 | .region_nop = kvm_region_nop, |
e5896b12 AP |
826 | .log_start = kvm_log_start, |
827 | .log_stop = kvm_log_stop, | |
a01672d3 AK |
828 | .log_sync = kvm_log_sync, |
829 | .log_global_start = kvm_log_global_start, | |
830 | .log_global_stop = kvm_log_global_stop, | |
80a1ea37 AK |
831 | .eventfd_add = kvm_eventfd_add, |
832 | .eventfd_del = kvm_eventfd_del, | |
72e22d2f | 833 | .priority = 10, |
7b8f3b78 MT |
834 | }; |
835 | ||
9349b4f9 | 836 | static void kvm_handle_interrupt(CPUArchState *env, int mask) |
aa7f74d1 JK |
837 | { |
838 | env->interrupt_request |= mask; | |
839 | ||
840 | if (!qemu_cpu_is_self(env)) { | |
841 | qemu_cpu_kick(env); | |
842 | } | |
843 | } | |
844 | ||
84b058d7 JK |
845 | int kvm_irqchip_set_irq(KVMState *s, int irq, int level) |
846 | { | |
847 | struct kvm_irq_level event; | |
848 | int ret; | |
849 | ||
3d4b2649 | 850 | assert(kvm_irqchip_in_kernel()); |
84b058d7 JK |
851 | |
852 | event.level = level; | |
853 | event.irq = irq; | |
854 | ret = kvm_vm_ioctl(s, s->irqchip_inject_ioctl, &event); | |
855 | if (ret < 0) { | |
856 | perror("kvm_set_irqchip_line"); | |
857 | abort(); | |
858 | } | |
859 | ||
860 | return (s->irqchip_inject_ioctl == KVM_IRQ_LINE) ? 1 : event.status; | |
861 | } | |
862 | ||
863 | #ifdef KVM_CAP_IRQ_ROUTING | |
864 | static void set_gsi(KVMState *s, unsigned int gsi) | |
865 | { | |
866 | assert(gsi < s->max_gsi); | |
867 | ||
868 | s->used_gsi_bitmap[gsi / 32] |= 1U << (gsi % 32); | |
869 | } | |
870 | ||
871 | static void kvm_init_irq_routing(KVMState *s) | |
872 | { | |
873 | int gsi_count; | |
874 | ||
875 | gsi_count = kvm_check_extension(s, KVM_CAP_IRQ_ROUTING); | |
876 | if (gsi_count > 0) { | |
877 | unsigned int gsi_bits, i; | |
878 | ||
879 | /* Round up so we can search ints using ffs */ | |
880 | gsi_bits = (gsi_count + 31) / 32; | |
881 | s->used_gsi_bitmap = g_malloc0(gsi_bits / 8); | |
882 | s->max_gsi = gsi_bits; | |
883 | ||
884 | /* Mark any over-allocated bits as already in use */ | |
885 | for (i = gsi_count; i < gsi_bits; i++) { | |
886 | set_gsi(s, i); | |
887 | } | |
888 | } | |
889 | ||
890 | s->irq_routes = g_malloc0(sizeof(*s->irq_routes)); | |
891 | s->nr_allocated_irq_routes = 0; | |
892 | ||
893 | kvm_arch_init_irq_routing(s); | |
894 | } | |
895 | ||
896 | static void kvm_add_routing_entry(KVMState *s, | |
897 | struct kvm_irq_routing_entry *entry) | |
898 | { | |
899 | struct kvm_irq_routing_entry *new; | |
900 | int n, size; | |
901 | ||
902 | if (s->irq_routes->nr == s->nr_allocated_irq_routes) { | |
903 | n = s->nr_allocated_irq_routes * 2; | |
904 | if (n < 64) { | |
905 | n = 64; | |
906 | } | |
907 | size = sizeof(struct kvm_irq_routing); | |
908 | size += n * sizeof(*new); | |
909 | s->irq_routes = g_realloc(s->irq_routes, size); | |
910 | s->nr_allocated_irq_routes = n; | |
911 | } | |
912 | n = s->irq_routes->nr++; | |
913 | new = &s->irq_routes->entries[n]; | |
914 | memset(new, 0, sizeof(*new)); | |
915 | new->gsi = entry->gsi; | |
916 | new->type = entry->type; | |
917 | new->flags = entry->flags; | |
918 | new->u = entry->u; | |
919 | ||
920 | set_gsi(s, entry->gsi); | |
921 | } | |
922 | ||
923 | void kvm_irqchip_add_route(KVMState *s, int irq, int irqchip, int pin) | |
924 | { | |
925 | struct kvm_irq_routing_entry e; | |
926 | ||
927 | e.gsi = irq; | |
928 | e.type = KVM_IRQ_ROUTING_IRQCHIP; | |
929 | e.flags = 0; | |
930 | e.u.irqchip.irqchip = irqchip; | |
931 | e.u.irqchip.pin = pin; | |
932 | kvm_add_routing_entry(s, &e); | |
933 | } | |
934 | ||
935 | int kvm_irqchip_commit_routes(KVMState *s) | |
936 | { | |
937 | s->irq_routes->flags = 0; | |
938 | return kvm_vm_ioctl(s, KVM_SET_GSI_ROUTING, s->irq_routes); | |
939 | } | |
940 | ||
941 | #else /* !KVM_CAP_IRQ_ROUTING */ | |
942 | ||
943 | static void kvm_init_irq_routing(KVMState *s) | |
944 | { | |
945 | } | |
946 | #endif /* !KVM_CAP_IRQ_ROUTING */ | |
947 | ||
948 | static int kvm_irqchip_create(KVMState *s) | |
949 | { | |
950 | QemuOptsList *list = qemu_find_opts("machine"); | |
951 | int ret; | |
952 | ||
953 | if (QTAILQ_EMPTY(&list->head) || | |
954 | !qemu_opt_get_bool(QTAILQ_FIRST(&list->head), | |
955 | "kernel_irqchip", false) || | |
956 | !kvm_check_extension(s, KVM_CAP_IRQCHIP)) { | |
957 | return 0; | |
958 | } | |
959 | ||
960 | ret = kvm_vm_ioctl(s, KVM_CREATE_IRQCHIP); | |
961 | if (ret < 0) { | |
962 | fprintf(stderr, "Create kernel irqchip failed\n"); | |
963 | return ret; | |
964 | } | |
965 | ||
966 | s->irqchip_inject_ioctl = KVM_IRQ_LINE; | |
967 | if (kvm_check_extension(s, KVM_CAP_IRQ_INJECT_STATUS)) { | |
968 | s->irqchip_inject_ioctl = KVM_IRQ_LINE_STATUS; | |
969 | } | |
3d4b2649 | 970 | kvm_kernel_irqchip = true; |
84b058d7 JK |
971 | |
972 | kvm_init_irq_routing(s); | |
973 | ||
974 | return 0; | |
975 | } | |
976 | ||
cad1e282 | 977 | int kvm_init(void) |
05330448 | 978 | { |
168ccc11 JK |
979 | static const char upgrade_note[] = |
980 | "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n" | |
981 | "(see http://sourceforge.net/projects/kvm).\n"; | |
05330448 | 982 | KVMState *s; |
94a8d39a | 983 | const KVMCapabilityInfo *missing_cap; |
05330448 AL |
984 | int ret; |
985 | int i; | |
986 | ||
7267c094 | 987 | s = g_malloc0(sizeof(KVMState)); |
05330448 | 988 | |
e22a25c9 | 989 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
72cf2d4f | 990 | QTAILQ_INIT(&s->kvm_sw_breakpoints); |
e22a25c9 | 991 | #endif |
a426e122 | 992 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { |
05330448 | 993 | s->slots[i].slot = i; |
a426e122 | 994 | } |
05330448 | 995 | s->vmfd = -1; |
40ff6d7e | 996 | s->fd = qemu_open("/dev/kvm", O_RDWR); |
05330448 AL |
997 | if (s->fd == -1) { |
998 | fprintf(stderr, "Could not access KVM kernel module: %m\n"); | |
999 | ret = -errno; | |
1000 | goto err; | |
1001 | } | |
1002 | ||
1003 | ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0); | |
1004 | if (ret < KVM_API_VERSION) { | |
a426e122 | 1005 | if (ret > 0) { |
05330448 | 1006 | ret = -EINVAL; |
a426e122 | 1007 | } |
05330448 AL |
1008 | fprintf(stderr, "kvm version too old\n"); |
1009 | goto err; | |
1010 | } | |
1011 | ||
1012 | if (ret > KVM_API_VERSION) { | |
1013 | ret = -EINVAL; | |
1014 | fprintf(stderr, "kvm version not supported\n"); | |
1015 | goto err; | |
1016 | } | |
1017 | ||
1018 | s->vmfd = kvm_ioctl(s, KVM_CREATE_VM, 0); | |
0104dcac AG |
1019 | if (s->vmfd < 0) { |
1020 | #ifdef TARGET_S390X | |
1021 | fprintf(stderr, "Please add the 'switch_amode' kernel parameter to " | |
1022 | "your host kernel command line\n"); | |
1023 | #endif | |
db9eae1c | 1024 | ret = s->vmfd; |
05330448 | 1025 | goto err; |
0104dcac | 1026 | } |
05330448 | 1027 | |
94a8d39a JK |
1028 | missing_cap = kvm_check_extension_list(s, kvm_required_capabilites); |
1029 | if (!missing_cap) { | |
1030 | missing_cap = | |
1031 | kvm_check_extension_list(s, kvm_arch_required_capabilities); | |
05330448 | 1032 | } |
94a8d39a | 1033 | if (missing_cap) { |
ad7b8b33 | 1034 | ret = -EINVAL; |
94a8d39a JK |
1035 | fprintf(stderr, "kvm does not support %s\n%s", |
1036 | missing_cap->name, upgrade_note); | |
d85dc283 AL |
1037 | goto err; |
1038 | } | |
1039 | ||
ad7b8b33 | 1040 | s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO); |
f65ed4c1 | 1041 | |
e69917e2 | 1042 | s->broken_set_mem_region = 1; |
14a09518 | 1043 | ret = kvm_check_extension(s, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS); |
e69917e2 JK |
1044 | if (ret > 0) { |
1045 | s->broken_set_mem_region = 0; | |
1046 | } | |
e69917e2 | 1047 | |
a0fb002c JK |
1048 | #ifdef KVM_CAP_VCPU_EVENTS |
1049 | s->vcpu_events = kvm_check_extension(s, KVM_CAP_VCPU_EVENTS); | |
1050 | #endif | |
1051 | ||
b0b1d690 JK |
1052 | s->robust_singlestep = |
1053 | kvm_check_extension(s, KVM_CAP_X86_ROBUST_SINGLESTEP); | |
b0b1d690 | 1054 | |
ff44f1a3 JK |
1055 | #ifdef KVM_CAP_DEBUGREGS |
1056 | s->debugregs = kvm_check_extension(s, KVM_CAP_DEBUGREGS); | |
1057 | #endif | |
1058 | ||
f1665b21 SY |
1059 | #ifdef KVM_CAP_XSAVE |
1060 | s->xsave = kvm_check_extension(s, KVM_CAP_XSAVE); | |
1061 | #endif | |
1062 | ||
f1665b21 SY |
1063 | #ifdef KVM_CAP_XCRS |
1064 | s->xcrs = kvm_check_extension(s, KVM_CAP_XCRS); | |
1065 | #endif | |
1066 | ||
8a7c7393 JK |
1067 | #ifdef KVM_CAP_PIT_STATE2 |
1068 | s->pit_state2 = kvm_check_extension(s, KVM_CAP_PIT_STATE2); | |
1069 | #endif | |
1070 | ||
cad1e282 | 1071 | ret = kvm_arch_init(s); |
a426e122 | 1072 | if (ret < 0) { |
05330448 | 1073 | goto err; |
a426e122 | 1074 | } |
05330448 | 1075 | |
84b058d7 JK |
1076 | ret = kvm_irqchip_create(s); |
1077 | if (ret < 0) { | |
1078 | goto err; | |
1079 | } | |
1080 | ||
05330448 | 1081 | kvm_state = s; |
7376e582 | 1082 | memory_listener_register(&kvm_memory_listener, NULL); |
05330448 | 1083 | |
d2f2b8a7 SH |
1084 | s->many_ioeventfds = kvm_check_many_ioeventfds(); |
1085 | ||
aa7f74d1 JK |
1086 | cpu_interrupt_handler = kvm_handle_interrupt; |
1087 | ||
05330448 AL |
1088 | return 0; |
1089 | ||
1090 | err: | |
1091 | if (s) { | |
db9eae1c | 1092 | if (s->vmfd >= 0) { |
05330448 | 1093 | close(s->vmfd); |
a426e122 JK |
1094 | } |
1095 | if (s->fd != -1) { | |
05330448 | 1096 | close(s->fd); |
a426e122 | 1097 | } |
05330448 | 1098 | } |
7267c094 | 1099 | g_free(s); |
05330448 AL |
1100 | |
1101 | return ret; | |
1102 | } | |
1103 | ||
b30e93e9 JK |
1104 | static void kvm_handle_io(uint16_t port, void *data, int direction, int size, |
1105 | uint32_t count) | |
05330448 AL |
1106 | { |
1107 | int i; | |
1108 | uint8_t *ptr = data; | |
1109 | ||
1110 | for (i = 0; i < count; i++) { | |
1111 | if (direction == KVM_EXIT_IO_IN) { | |
1112 | switch (size) { | |
1113 | case 1: | |
afcea8cb | 1114 | stb_p(ptr, cpu_inb(port)); |
05330448 AL |
1115 | break; |
1116 | case 2: | |
afcea8cb | 1117 | stw_p(ptr, cpu_inw(port)); |
05330448 AL |
1118 | break; |
1119 | case 4: | |
afcea8cb | 1120 | stl_p(ptr, cpu_inl(port)); |
05330448 AL |
1121 | break; |
1122 | } | |
1123 | } else { | |
1124 | switch (size) { | |
1125 | case 1: | |
afcea8cb | 1126 | cpu_outb(port, ldub_p(ptr)); |
05330448 AL |
1127 | break; |
1128 | case 2: | |
afcea8cb | 1129 | cpu_outw(port, lduw_p(ptr)); |
05330448 AL |
1130 | break; |
1131 | case 4: | |
afcea8cb | 1132 | cpu_outl(port, ldl_p(ptr)); |
05330448 AL |
1133 | break; |
1134 | } | |
1135 | } | |
1136 | ||
1137 | ptr += size; | |
1138 | } | |
05330448 AL |
1139 | } |
1140 | ||
9349b4f9 | 1141 | static int kvm_handle_internal_error(CPUArchState *env, struct kvm_run *run) |
7c80eef8 | 1142 | { |
bb44e0d1 | 1143 | fprintf(stderr, "KVM internal error."); |
7c80eef8 MT |
1144 | if (kvm_check_extension(kvm_state, KVM_CAP_INTERNAL_ERROR_DATA)) { |
1145 | int i; | |
1146 | ||
bb44e0d1 | 1147 | fprintf(stderr, " Suberror: %d\n", run->internal.suberror); |
7c80eef8 MT |
1148 | for (i = 0; i < run->internal.ndata; ++i) { |
1149 | fprintf(stderr, "extra data[%d]: %"PRIx64"\n", | |
1150 | i, (uint64_t)run->internal.data[i]); | |
1151 | } | |
bb44e0d1 JK |
1152 | } else { |
1153 | fprintf(stderr, "\n"); | |
7c80eef8 | 1154 | } |
7c80eef8 MT |
1155 | if (run->internal.suberror == KVM_INTERNAL_ERROR_EMULATION) { |
1156 | fprintf(stderr, "emulation failure\n"); | |
a426e122 | 1157 | if (!kvm_arch_stop_on_emulation_error(env)) { |
f5c848ee | 1158 | cpu_dump_state(env, stderr, fprintf, CPU_DUMP_CODE); |
d73cd8f4 | 1159 | return EXCP_INTERRUPT; |
a426e122 | 1160 | } |
7c80eef8 MT |
1161 | } |
1162 | /* FIXME: Should trigger a qmp message to let management know | |
1163 | * something went wrong. | |
1164 | */ | |
73aaec4a | 1165 | return -1; |
7c80eef8 | 1166 | } |
7c80eef8 | 1167 | |
62a2744c | 1168 | void kvm_flush_coalesced_mmio_buffer(void) |
f65ed4c1 | 1169 | { |
f65ed4c1 | 1170 | KVMState *s = kvm_state; |
1cae88b9 AK |
1171 | |
1172 | if (s->coalesced_flush_in_progress) { | |
1173 | return; | |
1174 | } | |
1175 | ||
1176 | s->coalesced_flush_in_progress = true; | |
1177 | ||
62a2744c SY |
1178 | if (s->coalesced_mmio_ring) { |
1179 | struct kvm_coalesced_mmio_ring *ring = s->coalesced_mmio_ring; | |
f65ed4c1 AL |
1180 | while (ring->first != ring->last) { |
1181 | struct kvm_coalesced_mmio *ent; | |
1182 | ||
1183 | ent = &ring->coalesced_mmio[ring->first]; | |
1184 | ||
1185 | cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len); | |
85199474 | 1186 | smp_wmb(); |
f65ed4c1 AL |
1187 | ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX; |
1188 | } | |
1189 | } | |
1cae88b9 AK |
1190 | |
1191 | s->coalesced_flush_in_progress = false; | |
f65ed4c1 AL |
1192 | } |
1193 | ||
2705d56a | 1194 | static void do_kvm_cpu_synchronize_state(void *_env) |
4c0960c0 | 1195 | { |
9349b4f9 | 1196 | CPUArchState *env = _env; |
2705d56a | 1197 | |
9ded2744 | 1198 | if (!env->kvm_vcpu_dirty) { |
4c0960c0 | 1199 | kvm_arch_get_registers(env); |
9ded2744 | 1200 | env->kvm_vcpu_dirty = 1; |
4c0960c0 AK |
1201 | } |
1202 | } | |
1203 | ||
9349b4f9 | 1204 | void kvm_cpu_synchronize_state(CPUArchState *env) |
2705d56a | 1205 | { |
a426e122 | 1206 | if (!env->kvm_vcpu_dirty) { |
2705d56a | 1207 | run_on_cpu(env, do_kvm_cpu_synchronize_state, env); |
a426e122 | 1208 | } |
2705d56a JK |
1209 | } |
1210 | ||
9349b4f9 | 1211 | void kvm_cpu_synchronize_post_reset(CPUArchState *env) |
ea375f9a JK |
1212 | { |
1213 | kvm_arch_put_registers(env, KVM_PUT_RESET_STATE); | |
1214 | env->kvm_vcpu_dirty = 0; | |
1215 | } | |
1216 | ||
9349b4f9 | 1217 | void kvm_cpu_synchronize_post_init(CPUArchState *env) |
ea375f9a JK |
1218 | { |
1219 | kvm_arch_put_registers(env, KVM_PUT_FULL_STATE); | |
1220 | env->kvm_vcpu_dirty = 0; | |
1221 | } | |
1222 | ||
9349b4f9 | 1223 | int kvm_cpu_exec(CPUArchState *env) |
05330448 AL |
1224 | { |
1225 | struct kvm_run *run = env->kvm_run; | |
7cbb533f | 1226 | int ret, run_ret; |
05330448 | 1227 | |
8c0d577e | 1228 | DPRINTF("kvm_cpu_exec()\n"); |
05330448 | 1229 | |
99036865 | 1230 | if (kvm_arch_process_async_events(env)) { |
9ccfac9e | 1231 | env->exit_request = 0; |
6792a57b | 1232 | return EXCP_HLT; |
9ccfac9e | 1233 | } |
0af691d7 | 1234 | |
9ccfac9e | 1235 | do { |
9ded2744 | 1236 | if (env->kvm_vcpu_dirty) { |
ea375f9a | 1237 | kvm_arch_put_registers(env, KVM_PUT_RUNTIME_STATE); |
9ded2744 | 1238 | env->kvm_vcpu_dirty = 0; |
4c0960c0 AK |
1239 | } |
1240 | ||
8c14c173 | 1241 | kvm_arch_pre_run(env, run); |
9ccfac9e JK |
1242 | if (env->exit_request) { |
1243 | DPRINTF("interrupt exit requested\n"); | |
1244 | /* | |
1245 | * KVM requires us to reenter the kernel after IO exits to complete | |
1246 | * instruction emulation. This self-signal will ensure that we | |
1247 | * leave ASAP again. | |
1248 | */ | |
1249 | qemu_cpu_kick_self(); | |
1250 | } | |
d549db5a | 1251 | qemu_mutex_unlock_iothread(); |
9ccfac9e | 1252 | |
7cbb533f | 1253 | run_ret = kvm_vcpu_ioctl(env, KVM_RUN, 0); |
9ccfac9e | 1254 | |
d549db5a | 1255 | qemu_mutex_lock_iothread(); |
05330448 AL |
1256 | kvm_arch_post_run(env, run); |
1257 | ||
b0c883b5 JK |
1258 | kvm_flush_coalesced_mmio_buffer(); |
1259 | ||
7cbb533f | 1260 | if (run_ret < 0) { |
dc77d341 JK |
1261 | if (run_ret == -EINTR || run_ret == -EAGAIN) { |
1262 | DPRINTF("io window exit\n"); | |
d73cd8f4 | 1263 | ret = EXCP_INTERRUPT; |
dc77d341 JK |
1264 | break; |
1265 | } | |
7b011fbc ME |
1266 | fprintf(stderr, "error: kvm run failed %s\n", |
1267 | strerror(-run_ret)); | |
05330448 AL |
1268 | abort(); |
1269 | } | |
1270 | ||
05330448 AL |
1271 | switch (run->exit_reason) { |
1272 | case KVM_EXIT_IO: | |
8c0d577e | 1273 | DPRINTF("handle_io\n"); |
b30e93e9 JK |
1274 | kvm_handle_io(run->io.port, |
1275 | (uint8_t *)run + run->io.data_offset, | |
1276 | run->io.direction, | |
1277 | run->io.size, | |
1278 | run->io.count); | |
d73cd8f4 | 1279 | ret = 0; |
05330448 AL |
1280 | break; |
1281 | case KVM_EXIT_MMIO: | |
8c0d577e | 1282 | DPRINTF("handle_mmio\n"); |
05330448 AL |
1283 | cpu_physical_memory_rw(run->mmio.phys_addr, |
1284 | run->mmio.data, | |
1285 | run->mmio.len, | |
1286 | run->mmio.is_write); | |
d73cd8f4 | 1287 | ret = 0; |
05330448 AL |
1288 | break; |
1289 | case KVM_EXIT_IRQ_WINDOW_OPEN: | |
8c0d577e | 1290 | DPRINTF("irq_window_open\n"); |
d73cd8f4 | 1291 | ret = EXCP_INTERRUPT; |
05330448 AL |
1292 | break; |
1293 | case KVM_EXIT_SHUTDOWN: | |
8c0d577e | 1294 | DPRINTF("shutdown\n"); |
05330448 | 1295 | qemu_system_reset_request(); |
d73cd8f4 | 1296 | ret = EXCP_INTERRUPT; |
05330448 AL |
1297 | break; |
1298 | case KVM_EXIT_UNKNOWN: | |
bb44e0d1 JK |
1299 | fprintf(stderr, "KVM: unknown exit, hardware reason %" PRIx64 "\n", |
1300 | (uint64_t)run->hw.hardware_exit_reason); | |
73aaec4a | 1301 | ret = -1; |
05330448 | 1302 | break; |
7c80eef8 | 1303 | case KVM_EXIT_INTERNAL_ERROR: |
73aaec4a | 1304 | ret = kvm_handle_internal_error(env, run); |
7c80eef8 | 1305 | break; |
05330448 | 1306 | default: |
8c0d577e | 1307 | DPRINTF("kvm_arch_handle_exit\n"); |
05330448 AL |
1308 | ret = kvm_arch_handle_exit(env, run); |
1309 | break; | |
1310 | } | |
d73cd8f4 | 1311 | } while (ret == 0); |
05330448 | 1312 | |
73aaec4a | 1313 | if (ret < 0) { |
f5c848ee | 1314 | cpu_dump_state(env, stderr, fprintf, CPU_DUMP_CODE); |
0461d5a6 | 1315 | vm_stop(RUN_STATE_INTERNAL_ERROR); |
becfc390 AL |
1316 | } |
1317 | ||
6792a57b | 1318 | env->exit_request = 0; |
05330448 AL |
1319 | return ret; |
1320 | } | |
1321 | ||
984b5181 | 1322 | int kvm_ioctl(KVMState *s, int type, ...) |
05330448 AL |
1323 | { |
1324 | int ret; | |
984b5181 AL |
1325 | void *arg; |
1326 | va_list ap; | |
05330448 | 1327 | |
984b5181 AL |
1328 | va_start(ap, type); |
1329 | arg = va_arg(ap, void *); | |
1330 | va_end(ap); | |
1331 | ||
1332 | ret = ioctl(s->fd, type, arg); | |
a426e122 | 1333 | if (ret == -1) { |
05330448 | 1334 | ret = -errno; |
a426e122 | 1335 | } |
05330448 AL |
1336 | return ret; |
1337 | } | |
1338 | ||
984b5181 | 1339 | int kvm_vm_ioctl(KVMState *s, int type, ...) |
05330448 AL |
1340 | { |
1341 | int ret; | |
984b5181 AL |
1342 | void *arg; |
1343 | va_list ap; | |
1344 | ||
1345 | va_start(ap, type); | |
1346 | arg = va_arg(ap, void *); | |
1347 | va_end(ap); | |
05330448 | 1348 | |
984b5181 | 1349 | ret = ioctl(s->vmfd, type, arg); |
a426e122 | 1350 | if (ret == -1) { |
05330448 | 1351 | ret = -errno; |
a426e122 | 1352 | } |
05330448 AL |
1353 | return ret; |
1354 | } | |
1355 | ||
9349b4f9 | 1356 | int kvm_vcpu_ioctl(CPUArchState *env, int type, ...) |
05330448 AL |
1357 | { |
1358 | int ret; | |
984b5181 AL |
1359 | void *arg; |
1360 | va_list ap; | |
1361 | ||
1362 | va_start(ap, type); | |
1363 | arg = va_arg(ap, void *); | |
1364 | va_end(ap); | |
05330448 | 1365 | |
984b5181 | 1366 | ret = ioctl(env->kvm_fd, type, arg); |
a426e122 | 1367 | if (ret == -1) { |
05330448 | 1368 | ret = -errno; |
a426e122 | 1369 | } |
05330448 AL |
1370 | return ret; |
1371 | } | |
bd322087 AL |
1372 | |
1373 | int kvm_has_sync_mmu(void) | |
1374 | { | |
94a8d39a | 1375 | return kvm_check_extension(kvm_state, KVM_CAP_SYNC_MMU); |
bd322087 | 1376 | } |
e22a25c9 | 1377 | |
a0fb002c JK |
1378 | int kvm_has_vcpu_events(void) |
1379 | { | |
1380 | return kvm_state->vcpu_events; | |
1381 | } | |
1382 | ||
b0b1d690 JK |
1383 | int kvm_has_robust_singlestep(void) |
1384 | { | |
1385 | return kvm_state->robust_singlestep; | |
1386 | } | |
1387 | ||
ff44f1a3 JK |
1388 | int kvm_has_debugregs(void) |
1389 | { | |
1390 | return kvm_state->debugregs; | |
1391 | } | |
1392 | ||
f1665b21 SY |
1393 | int kvm_has_xsave(void) |
1394 | { | |
1395 | return kvm_state->xsave; | |
1396 | } | |
1397 | ||
1398 | int kvm_has_xcrs(void) | |
1399 | { | |
1400 | return kvm_state->xcrs; | |
1401 | } | |
1402 | ||
8a7c7393 JK |
1403 | int kvm_has_pit_state2(void) |
1404 | { | |
1405 | return kvm_state->pit_state2; | |
1406 | } | |
1407 | ||
d2f2b8a7 SH |
1408 | int kvm_has_many_ioeventfds(void) |
1409 | { | |
1410 | if (!kvm_enabled()) { | |
1411 | return 0; | |
1412 | } | |
1413 | return kvm_state->many_ioeventfds; | |
1414 | } | |
1415 | ||
84b058d7 JK |
1416 | int kvm_has_gsi_routing(void) |
1417 | { | |
a9c5eb0d | 1418 | #ifdef KVM_CAP_IRQ_ROUTING |
84b058d7 | 1419 | return kvm_check_extension(kvm_state, KVM_CAP_IRQ_ROUTING); |
a9c5eb0d AG |
1420 | #else |
1421 | return false; | |
1422 | #endif | |
84b058d7 JK |
1423 | } |
1424 | ||
9b5b76d4 JK |
1425 | int kvm_allows_irq0_override(void) |
1426 | { | |
3d4b2649 | 1427 | return !kvm_irqchip_in_kernel() || kvm_has_gsi_routing(); |
9b5b76d4 JK |
1428 | } |
1429 | ||
6f0437e8 JK |
1430 | void kvm_setup_guest_memory(void *start, size_t size) |
1431 | { | |
1432 | if (!kvm_has_sync_mmu()) { | |
e78815a5 | 1433 | int ret = qemu_madvise(start, size, QEMU_MADV_DONTFORK); |
6f0437e8 JK |
1434 | |
1435 | if (ret) { | |
e78815a5 AF |
1436 | perror("qemu_madvise"); |
1437 | fprintf(stderr, | |
1438 | "Need MADV_DONTFORK in absence of synchronous KVM MMU\n"); | |
6f0437e8 JK |
1439 | exit(1); |
1440 | } | |
6f0437e8 JK |
1441 | } |
1442 | } | |
1443 | ||
e22a25c9 | 1444 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
9349b4f9 | 1445 | struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUArchState *env, |
e22a25c9 AL |
1446 | target_ulong pc) |
1447 | { | |
1448 | struct kvm_sw_breakpoint *bp; | |
1449 | ||
72cf2d4f | 1450 | QTAILQ_FOREACH(bp, &env->kvm_state->kvm_sw_breakpoints, entry) { |
a426e122 | 1451 | if (bp->pc == pc) { |
e22a25c9 | 1452 | return bp; |
a426e122 | 1453 | } |
e22a25c9 AL |
1454 | } |
1455 | return NULL; | |
1456 | } | |
1457 | ||
9349b4f9 | 1458 | int kvm_sw_breakpoints_active(CPUArchState *env) |
e22a25c9 | 1459 | { |
72cf2d4f | 1460 | return !QTAILQ_EMPTY(&env->kvm_state->kvm_sw_breakpoints); |
e22a25c9 AL |
1461 | } |
1462 | ||
452e4751 GC |
1463 | struct kvm_set_guest_debug_data { |
1464 | struct kvm_guest_debug dbg; | |
9349b4f9 | 1465 | CPUArchState *env; |
452e4751 GC |
1466 | int err; |
1467 | }; | |
1468 | ||
1469 | static void kvm_invoke_set_guest_debug(void *data) | |
1470 | { | |
1471 | struct kvm_set_guest_debug_data *dbg_data = data; | |
9349b4f9 | 1472 | CPUArchState *env = dbg_data->env; |
b3807725 | 1473 | |
b3807725 | 1474 | dbg_data->err = kvm_vcpu_ioctl(env, KVM_SET_GUEST_DEBUG, &dbg_data->dbg); |
452e4751 GC |
1475 | } |
1476 | ||
9349b4f9 | 1477 | int kvm_update_guest_debug(CPUArchState *env, unsigned long reinject_trap) |
e22a25c9 | 1478 | { |
452e4751 | 1479 | struct kvm_set_guest_debug_data data; |
e22a25c9 | 1480 | |
b0b1d690 | 1481 | data.dbg.control = reinject_trap; |
e22a25c9 | 1482 | |
b0b1d690 JK |
1483 | if (env->singlestep_enabled) { |
1484 | data.dbg.control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP; | |
1485 | } | |
452e4751 | 1486 | kvm_arch_update_guest_debug(env, &data.dbg); |
452e4751 | 1487 | data.env = env; |
e22a25c9 | 1488 | |
be41cbe0 | 1489 | run_on_cpu(env, kvm_invoke_set_guest_debug, &data); |
452e4751 | 1490 | return data.err; |
e22a25c9 AL |
1491 | } |
1492 | ||
9349b4f9 | 1493 | int kvm_insert_breakpoint(CPUArchState *current_env, target_ulong addr, |
e22a25c9 AL |
1494 | target_ulong len, int type) |
1495 | { | |
1496 | struct kvm_sw_breakpoint *bp; | |
9349b4f9 | 1497 | CPUArchState *env; |
e22a25c9 AL |
1498 | int err; |
1499 | ||
1500 | if (type == GDB_BREAKPOINT_SW) { | |
1501 | bp = kvm_find_sw_breakpoint(current_env, addr); | |
1502 | if (bp) { | |
1503 | bp->use_count++; | |
1504 | return 0; | |
1505 | } | |
1506 | ||
7267c094 | 1507 | bp = g_malloc(sizeof(struct kvm_sw_breakpoint)); |
a426e122 | 1508 | if (!bp) { |
e22a25c9 | 1509 | return -ENOMEM; |
a426e122 | 1510 | } |
e22a25c9 AL |
1511 | |
1512 | bp->pc = addr; | |
1513 | bp->use_count = 1; | |
1514 | err = kvm_arch_insert_sw_breakpoint(current_env, bp); | |
1515 | if (err) { | |
7267c094 | 1516 | g_free(bp); |
e22a25c9 AL |
1517 | return err; |
1518 | } | |
1519 | ||
72cf2d4f | 1520 | QTAILQ_INSERT_HEAD(¤t_env->kvm_state->kvm_sw_breakpoints, |
e22a25c9 AL |
1521 | bp, entry); |
1522 | } else { | |
1523 | err = kvm_arch_insert_hw_breakpoint(addr, len, type); | |
a426e122 | 1524 | if (err) { |
e22a25c9 | 1525 | return err; |
a426e122 | 1526 | } |
e22a25c9 AL |
1527 | } |
1528 | ||
1529 | for (env = first_cpu; env != NULL; env = env->next_cpu) { | |
1530 | err = kvm_update_guest_debug(env, 0); | |
a426e122 | 1531 | if (err) { |
e22a25c9 | 1532 | return err; |
a426e122 | 1533 | } |
e22a25c9 AL |
1534 | } |
1535 | return 0; | |
1536 | } | |
1537 | ||
9349b4f9 | 1538 | int kvm_remove_breakpoint(CPUArchState *current_env, target_ulong addr, |
e22a25c9 AL |
1539 | target_ulong len, int type) |
1540 | { | |
1541 | struct kvm_sw_breakpoint *bp; | |
9349b4f9 | 1542 | CPUArchState *env; |
e22a25c9 AL |
1543 | int err; |
1544 | ||
1545 | if (type == GDB_BREAKPOINT_SW) { | |
1546 | bp = kvm_find_sw_breakpoint(current_env, addr); | |
a426e122 | 1547 | if (!bp) { |
e22a25c9 | 1548 | return -ENOENT; |
a426e122 | 1549 | } |
e22a25c9 AL |
1550 | |
1551 | if (bp->use_count > 1) { | |
1552 | bp->use_count--; | |
1553 | return 0; | |
1554 | } | |
1555 | ||
1556 | err = kvm_arch_remove_sw_breakpoint(current_env, bp); | |
a426e122 | 1557 | if (err) { |
e22a25c9 | 1558 | return err; |
a426e122 | 1559 | } |
e22a25c9 | 1560 | |
72cf2d4f | 1561 | QTAILQ_REMOVE(¤t_env->kvm_state->kvm_sw_breakpoints, bp, entry); |
7267c094 | 1562 | g_free(bp); |
e22a25c9 AL |
1563 | } else { |
1564 | err = kvm_arch_remove_hw_breakpoint(addr, len, type); | |
a426e122 | 1565 | if (err) { |
e22a25c9 | 1566 | return err; |
a426e122 | 1567 | } |
e22a25c9 AL |
1568 | } |
1569 | ||
1570 | for (env = first_cpu; env != NULL; env = env->next_cpu) { | |
1571 | err = kvm_update_guest_debug(env, 0); | |
a426e122 | 1572 | if (err) { |
e22a25c9 | 1573 | return err; |
a426e122 | 1574 | } |
e22a25c9 AL |
1575 | } |
1576 | return 0; | |
1577 | } | |
1578 | ||
9349b4f9 | 1579 | void kvm_remove_all_breakpoints(CPUArchState *current_env) |
e22a25c9 AL |
1580 | { |
1581 | struct kvm_sw_breakpoint *bp, *next; | |
1582 | KVMState *s = current_env->kvm_state; | |
9349b4f9 | 1583 | CPUArchState *env; |
e22a25c9 | 1584 | |
72cf2d4f | 1585 | QTAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) { |
e22a25c9 AL |
1586 | if (kvm_arch_remove_sw_breakpoint(current_env, bp) != 0) { |
1587 | /* Try harder to find a CPU that currently sees the breakpoint. */ | |
1588 | for (env = first_cpu; env != NULL; env = env->next_cpu) { | |
a426e122 | 1589 | if (kvm_arch_remove_sw_breakpoint(env, bp) == 0) { |
e22a25c9 | 1590 | break; |
a426e122 | 1591 | } |
e22a25c9 AL |
1592 | } |
1593 | } | |
1594 | } | |
1595 | kvm_arch_remove_all_hw_breakpoints(); | |
1596 | ||
a426e122 | 1597 | for (env = first_cpu; env != NULL; env = env->next_cpu) { |
e22a25c9 | 1598 | kvm_update_guest_debug(env, 0); |
a426e122 | 1599 | } |
e22a25c9 AL |
1600 | } |
1601 | ||
1602 | #else /* !KVM_CAP_SET_GUEST_DEBUG */ | |
1603 | ||
9349b4f9 | 1604 | int kvm_update_guest_debug(CPUArchState *env, unsigned long reinject_trap) |
e22a25c9 AL |
1605 | { |
1606 | return -EINVAL; | |
1607 | } | |
1608 | ||
9349b4f9 | 1609 | int kvm_insert_breakpoint(CPUArchState *current_env, target_ulong addr, |
e22a25c9 AL |
1610 | target_ulong len, int type) |
1611 | { | |
1612 | return -EINVAL; | |
1613 | } | |
1614 | ||
9349b4f9 | 1615 | int kvm_remove_breakpoint(CPUArchState *current_env, target_ulong addr, |
e22a25c9 AL |
1616 | target_ulong len, int type) |
1617 | { | |
1618 | return -EINVAL; | |
1619 | } | |
1620 | ||
9349b4f9 | 1621 | void kvm_remove_all_breakpoints(CPUArchState *current_env) |
e22a25c9 AL |
1622 | { |
1623 | } | |
1624 | #endif /* !KVM_CAP_SET_GUEST_DEBUG */ | |
cc84de95 | 1625 | |
9349b4f9 | 1626 | int kvm_set_signal_mask(CPUArchState *env, const sigset_t *sigset) |
cc84de95 MT |
1627 | { |
1628 | struct kvm_signal_mask *sigmask; | |
1629 | int r; | |
1630 | ||
a426e122 | 1631 | if (!sigset) { |
cc84de95 | 1632 | return kvm_vcpu_ioctl(env, KVM_SET_SIGNAL_MASK, NULL); |
a426e122 | 1633 | } |
cc84de95 | 1634 | |
7267c094 | 1635 | sigmask = g_malloc(sizeof(*sigmask) + sizeof(*sigset)); |
cc84de95 MT |
1636 | |
1637 | sigmask->len = 8; | |
1638 | memcpy(sigmask->sigset, sigset, sizeof(*sigset)); | |
1639 | r = kvm_vcpu_ioctl(env, KVM_SET_SIGNAL_MASK, sigmask); | |
7267c094 | 1640 | g_free(sigmask); |
cc84de95 MT |
1641 | |
1642 | return r; | |
1643 | } | |
ca821806 | 1644 | |
44f1a3d8 CM |
1645 | int kvm_set_ioeventfd_mmio_long(int fd, uint32_t addr, uint32_t val, bool assign) |
1646 | { | |
44f1a3d8 CM |
1647 | int ret; |
1648 | struct kvm_ioeventfd iofd; | |
1649 | ||
1650 | iofd.datamatch = val; | |
1651 | iofd.addr = addr; | |
1652 | iofd.len = 4; | |
1653 | iofd.flags = KVM_IOEVENTFD_FLAG_DATAMATCH; | |
1654 | iofd.fd = fd; | |
1655 | ||
1656 | if (!kvm_enabled()) { | |
1657 | return -ENOSYS; | |
1658 | } | |
1659 | ||
1660 | if (!assign) { | |
1661 | iofd.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN; | |
1662 | } | |
1663 | ||
1664 | ret = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &iofd); | |
1665 | ||
1666 | if (ret < 0) { | |
1667 | return -errno; | |
1668 | } | |
1669 | ||
1670 | return 0; | |
44f1a3d8 CM |
1671 | } |
1672 | ||
ca821806 MT |
1673 | int kvm_set_ioeventfd_pio_word(int fd, uint16_t addr, uint16_t val, bool assign) |
1674 | { | |
1675 | struct kvm_ioeventfd kick = { | |
1676 | .datamatch = val, | |
1677 | .addr = addr, | |
1678 | .len = 2, | |
1679 | .flags = KVM_IOEVENTFD_FLAG_DATAMATCH | KVM_IOEVENTFD_FLAG_PIO, | |
1680 | .fd = fd, | |
1681 | }; | |
1682 | int r; | |
a426e122 | 1683 | if (!kvm_enabled()) { |
ca821806 | 1684 | return -ENOSYS; |
a426e122 JK |
1685 | } |
1686 | if (!assign) { | |
ca821806 | 1687 | kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN; |
a426e122 | 1688 | } |
ca821806 | 1689 | r = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick); |
a426e122 | 1690 | if (r < 0) { |
ca821806 | 1691 | return r; |
a426e122 | 1692 | } |
ca821806 | 1693 | return 0; |
98c8573e | 1694 | } |
a1b87fe0 | 1695 | |
9349b4f9 | 1696 | int kvm_on_sigbus_vcpu(CPUArchState *env, int code, void *addr) |
a1b87fe0 JK |
1697 | { |
1698 | return kvm_arch_on_sigbus_vcpu(env, code, addr); | |
1699 | } | |
1700 | ||
1701 | int kvm_on_sigbus(int code, void *addr) | |
1702 | { | |
1703 | return kvm_arch_on_sigbus(code, addr); | |
1704 | } |