root / kvm-all.c @ c05c7a73
History | View | Annotate | Download (29 kB)
1 |
/*
|
---|---|
2 |
* QEMU KVM support
|
3 |
*
|
4 |
* Copyright IBM, Corp. 2008
|
5 |
* Red Hat, Inc. 2008
|
6 |
*
|
7 |
* Authors:
|
8 |
* Anthony Liguori <aliguori@us.ibm.com>
|
9 |
* Glauber Costa <gcosta@redhat.com>
|
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> |
19 |
#include <stdarg.h> |
20 |
|
21 |
#include <linux/kvm.h> |
22 |
|
23 |
#include "qemu-common.h" |
24 |
#include "qemu-barrier.h" |
25 |
#include "sysemu.h" |
26 |
#include "hw/hw.h" |
27 |
#include "gdbstub.h" |
28 |
#include "kvm.h" |
29 |
|
30 |
/* KVM uses PAGE_SIZE in it's definition of COALESCED_MMIO_MAX */
|
31 |
#define PAGE_SIZE TARGET_PAGE_SIZE
|
32 |
|
33 |
//#define DEBUG_KVM
|
34 |
|
35 |
#ifdef DEBUG_KVM
|
36 |
#define dprintf(fmt, ...) \
|
37 |
do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0) |
38 |
#else
|
39 |
#define dprintf(fmt, ...) \
|
40 |
do { } while (0) |
41 |
#endif
|
42 |
|
43 |
typedef struct KVMSlot |
44 |
{ |
45 |
target_phys_addr_t start_addr; |
46 |
ram_addr_t memory_size; |
47 |
ram_addr_t phys_offset; |
48 |
int slot;
|
49 |
int flags;
|
50 |
} KVMSlot; |
51 |
|
52 |
typedef struct kvm_dirty_log KVMDirtyLog; |
53 |
|
54 |
int kvm_allowed = 0; |
55 |
|
56 |
struct KVMState
|
57 |
{ |
58 |
KVMSlot slots[32];
|
59 |
int fd;
|
60 |
int vmfd;
|
61 |
int coalesced_mmio;
|
62 |
#ifdef KVM_CAP_COALESCED_MMIO
|
63 |
struct kvm_coalesced_mmio_ring *coalesced_mmio_ring;
|
64 |
#endif
|
65 |
int broken_set_mem_region;
|
66 |
int migration_log;
|
67 |
int vcpu_events;
|
68 |
int robust_singlestep;
|
69 |
#ifdef KVM_CAP_SET_GUEST_DEBUG
|
70 |
struct kvm_sw_breakpoint_head kvm_sw_breakpoints;
|
71 |
#endif
|
72 |
int irqchip_in_kernel;
|
73 |
int pit_in_kernel;
|
74 |
}; |
75 |
|
76 |
static KVMState *kvm_state;
|
77 |
|
78 |
static KVMSlot *kvm_alloc_slot(KVMState *s)
|
79 |
{ |
80 |
int i;
|
81 |
|
82 |
for (i = 0; i < ARRAY_SIZE(s->slots); i++) { |
83 |
/* KVM private memory slots */
|
84 |
if (i >= 8 && i < 12) |
85 |
continue;
|
86 |
if (s->slots[i].memory_size == 0) |
87 |
return &s->slots[i];
|
88 |
} |
89 |
|
90 |
fprintf(stderr, "%s: no free slot available\n", __func__);
|
91 |
abort(); |
92 |
} |
93 |
|
94 |
static KVMSlot *kvm_lookup_matching_slot(KVMState *s,
|
95 |
target_phys_addr_t start_addr, |
96 |
target_phys_addr_t end_addr) |
97 |
{ |
98 |
int i;
|
99 |
|
100 |
for (i = 0; i < ARRAY_SIZE(s->slots); i++) { |
101 |
KVMSlot *mem = &s->slots[i]; |
102 |
|
103 |
if (start_addr == mem->start_addr &&
|
104 |
end_addr == mem->start_addr + mem->memory_size) { |
105 |
return mem;
|
106 |
} |
107 |
} |
108 |
|
109 |
return NULL; |
110 |
} |
111 |
|
112 |
/*
|
113 |
* Find overlapping slot with lowest start address
|
114 |
*/
|
115 |
static KVMSlot *kvm_lookup_overlapping_slot(KVMState *s,
|
116 |
target_phys_addr_t start_addr, |
117 |
target_phys_addr_t end_addr) |
118 |
{ |
119 |
KVMSlot *found = NULL;
|
120 |
int i;
|
121 |
|
122 |
for (i = 0; i < ARRAY_SIZE(s->slots); i++) { |
123 |
KVMSlot *mem = &s->slots[i]; |
124 |
|
125 |
if (mem->memory_size == 0 || |
126 |
(found && found->start_addr < mem->start_addr)) { |
127 |
continue;
|
128 |
} |
129 |
|
130 |
if (end_addr > mem->start_addr &&
|
131 |
start_addr < mem->start_addr + mem->memory_size) { |
132 |
found = mem; |
133 |
} |
134 |
} |
135 |
|
136 |
return found;
|
137 |
} |
138 |
|
139 |
static int kvm_set_user_memory_region(KVMState *s, KVMSlot *slot) |
140 |
{ |
141 |
struct kvm_userspace_memory_region mem;
|
142 |
|
143 |
mem.slot = slot->slot; |
144 |
mem.guest_phys_addr = slot->start_addr; |
145 |
mem.memory_size = slot->memory_size; |
146 |
mem.userspace_addr = (unsigned long)qemu_get_ram_ptr(slot->phys_offset); |
147 |
mem.flags = slot->flags; |
148 |
if (s->migration_log) {
|
149 |
mem.flags |= KVM_MEM_LOG_DIRTY_PAGES; |
150 |
} |
151 |
return kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem);
|
152 |
} |
153 |
|
154 |
static void kvm_reset_vcpu(void *opaque) |
155 |
{ |
156 |
CPUState *env = opaque; |
157 |
|
158 |
kvm_arch_reset_vcpu(env); |
159 |
} |
160 |
|
161 |
int kvm_irqchip_in_kernel(void) |
162 |
{ |
163 |
return kvm_state->irqchip_in_kernel;
|
164 |
} |
165 |
|
166 |
int kvm_pit_in_kernel(void) |
167 |
{ |
168 |
return kvm_state->pit_in_kernel;
|
169 |
} |
170 |
|
171 |
|
172 |
int kvm_init_vcpu(CPUState *env)
|
173 |
{ |
174 |
KVMState *s = kvm_state; |
175 |
long mmap_size;
|
176 |
int ret;
|
177 |
|
178 |
dprintf("kvm_init_vcpu\n");
|
179 |
|
180 |
ret = kvm_vm_ioctl(s, KVM_CREATE_VCPU, env->cpu_index); |
181 |
if (ret < 0) { |
182 |
dprintf("kvm_create_vcpu failed\n");
|
183 |
goto err;
|
184 |
} |
185 |
|
186 |
env->kvm_fd = ret; |
187 |
env->kvm_state = s; |
188 |
|
189 |
mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0);
|
190 |
if (mmap_size < 0) { |
191 |
dprintf("KVM_GET_VCPU_MMAP_SIZE failed\n");
|
192 |
goto err;
|
193 |
} |
194 |
|
195 |
env->kvm_run = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
|
196 |
env->kvm_fd, 0);
|
197 |
if (env->kvm_run == MAP_FAILED) {
|
198 |
ret = -errno; |
199 |
dprintf("mmap'ing vcpu state failed\n");
|
200 |
goto err;
|
201 |
} |
202 |
|
203 |
#ifdef KVM_CAP_COALESCED_MMIO
|
204 |
if (s->coalesced_mmio && !s->coalesced_mmio_ring)
|
205 |
s->coalesced_mmio_ring = (void *) env->kvm_run +
|
206 |
s->coalesced_mmio * PAGE_SIZE; |
207 |
#endif
|
208 |
|
209 |
ret = kvm_arch_init_vcpu(env); |
210 |
if (ret == 0) { |
211 |
qemu_register_reset(kvm_reset_vcpu, env); |
212 |
kvm_arch_reset_vcpu(env); |
213 |
} |
214 |
err:
|
215 |
return ret;
|
216 |
} |
217 |
|
218 |
/*
|
219 |
* dirty pages logging control
|
220 |
*/
|
221 |
static int kvm_dirty_pages_log_change(target_phys_addr_t phys_addr, |
222 |
ram_addr_t size, int flags, int mask) |
223 |
{ |
224 |
KVMState *s = kvm_state; |
225 |
KVMSlot *mem = kvm_lookup_matching_slot(s, phys_addr, phys_addr + size); |
226 |
int old_flags;
|
227 |
|
228 |
if (mem == NULL) { |
229 |
fprintf(stderr, "BUG: %s: invalid parameters " TARGET_FMT_plx "-" |
230 |
TARGET_FMT_plx "\n", __func__, phys_addr,
|
231 |
(target_phys_addr_t)(phys_addr + size - 1));
|
232 |
return -EINVAL;
|
233 |
} |
234 |
|
235 |
old_flags = mem->flags; |
236 |
|
237 |
flags = (mem->flags & ~mask) | flags; |
238 |
mem->flags = flags; |
239 |
|
240 |
/* If nothing changed effectively, no need to issue ioctl */
|
241 |
if (s->migration_log) {
|
242 |
flags |= KVM_MEM_LOG_DIRTY_PAGES; |
243 |
} |
244 |
if (flags == old_flags) {
|
245 |
return 0; |
246 |
} |
247 |
|
248 |
return kvm_set_user_memory_region(s, mem);
|
249 |
} |
250 |
|
251 |
int kvm_log_start(target_phys_addr_t phys_addr, ram_addr_t size)
|
252 |
{ |
253 |
return kvm_dirty_pages_log_change(phys_addr, size,
|
254 |
KVM_MEM_LOG_DIRTY_PAGES, |
255 |
KVM_MEM_LOG_DIRTY_PAGES); |
256 |
} |
257 |
|
258 |
int kvm_log_stop(target_phys_addr_t phys_addr, ram_addr_t size)
|
259 |
{ |
260 |
return kvm_dirty_pages_log_change(phys_addr, size,
|
261 |
0,
|
262 |
KVM_MEM_LOG_DIRTY_PAGES); |
263 |
} |
264 |
|
265 |
static int kvm_set_migration_log(int enable) |
266 |
{ |
267 |
KVMState *s = kvm_state; |
268 |
KVMSlot *mem; |
269 |
int i, err;
|
270 |
|
271 |
s->migration_log = enable; |
272 |
|
273 |
for (i = 0; i < ARRAY_SIZE(s->slots); i++) { |
274 |
mem = &s->slots[i]; |
275 |
|
276 |
if (!!(mem->flags & KVM_MEM_LOG_DIRTY_PAGES) == enable) {
|
277 |
continue;
|
278 |
} |
279 |
err = kvm_set_user_memory_region(s, mem); |
280 |
if (err) {
|
281 |
return err;
|
282 |
} |
283 |
} |
284 |
return 0; |
285 |
} |
286 |
|
287 |
static int test_le_bit(unsigned long nr, unsigned char *addr) |
288 |
{ |
289 |
return (addr[nr >> 3] >> (nr & 7)) & 1; |
290 |
} |
291 |
|
292 |
/**
|
293 |
* kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space
|
294 |
* This function updates qemu's dirty bitmap using cpu_physical_memory_set_dirty().
|
295 |
* This means all bits are set to dirty.
|
296 |
*
|
297 |
* @start_add: start of logged region.
|
298 |
* @end_addr: end of logged region.
|
299 |
*/
|
300 |
static int kvm_physical_sync_dirty_bitmap(target_phys_addr_t start_addr, |
301 |
target_phys_addr_t end_addr) |
302 |
{ |
303 |
KVMState *s = kvm_state; |
304 |
unsigned long size, allocated_size = 0; |
305 |
target_phys_addr_t phys_addr; |
306 |
ram_addr_t addr; |
307 |
KVMDirtyLog d; |
308 |
KVMSlot *mem; |
309 |
int ret = 0; |
310 |
|
311 |
d.dirty_bitmap = NULL;
|
312 |
while (start_addr < end_addr) {
|
313 |
mem = kvm_lookup_overlapping_slot(s, start_addr, end_addr); |
314 |
if (mem == NULL) { |
315 |
break;
|
316 |
} |
317 |
|
318 |
size = ((mem->memory_size >> TARGET_PAGE_BITS) + 7) / 8; |
319 |
if (!d.dirty_bitmap) {
|
320 |
d.dirty_bitmap = qemu_malloc(size); |
321 |
} else if (size > allocated_size) { |
322 |
d.dirty_bitmap = qemu_realloc(d.dirty_bitmap, size); |
323 |
} |
324 |
allocated_size = size; |
325 |
memset(d.dirty_bitmap, 0, allocated_size);
|
326 |
|
327 |
d.slot = mem->slot; |
328 |
|
329 |
if (kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d) == -1) { |
330 |
dprintf("ioctl failed %d\n", errno);
|
331 |
ret = -1;
|
332 |
break;
|
333 |
} |
334 |
|
335 |
for (phys_addr = mem->start_addr, addr = mem->phys_offset;
|
336 |
phys_addr < mem->start_addr + mem->memory_size; |
337 |
phys_addr += TARGET_PAGE_SIZE, addr += TARGET_PAGE_SIZE) { |
338 |
unsigned char *bitmap = (unsigned char *)d.dirty_bitmap; |
339 |
unsigned nr = (phys_addr - mem->start_addr) >> TARGET_PAGE_BITS;
|
340 |
|
341 |
if (test_le_bit(nr, bitmap)) {
|
342 |
cpu_physical_memory_set_dirty(addr); |
343 |
} |
344 |
} |
345 |
start_addr = phys_addr; |
346 |
} |
347 |
qemu_free(d.dirty_bitmap); |
348 |
|
349 |
return ret;
|
350 |
} |
351 |
|
352 |
int kvm_coalesce_mmio_region(target_phys_addr_t start, ram_addr_t size)
|
353 |
{ |
354 |
int ret = -ENOSYS;
|
355 |
#ifdef KVM_CAP_COALESCED_MMIO
|
356 |
KVMState *s = kvm_state; |
357 |
|
358 |
if (s->coalesced_mmio) {
|
359 |
struct kvm_coalesced_mmio_zone zone;
|
360 |
|
361 |
zone.addr = start; |
362 |
zone.size = size; |
363 |
|
364 |
ret = kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone); |
365 |
} |
366 |
#endif
|
367 |
|
368 |
return ret;
|
369 |
} |
370 |
|
371 |
int kvm_uncoalesce_mmio_region(target_phys_addr_t start, ram_addr_t size)
|
372 |
{ |
373 |
int ret = -ENOSYS;
|
374 |
#ifdef KVM_CAP_COALESCED_MMIO
|
375 |
KVMState *s = kvm_state; |
376 |
|
377 |
if (s->coalesced_mmio) {
|
378 |
struct kvm_coalesced_mmio_zone zone;
|
379 |
|
380 |
zone.addr = start; |
381 |
zone.size = size; |
382 |
|
383 |
ret = kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone); |
384 |
} |
385 |
#endif
|
386 |
|
387 |
return ret;
|
388 |
} |
389 |
|
390 |
int kvm_check_extension(KVMState *s, unsigned int extension) |
391 |
{ |
392 |
int ret;
|
393 |
|
394 |
ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension); |
395 |
if (ret < 0) { |
396 |
ret = 0;
|
397 |
} |
398 |
|
399 |
return ret;
|
400 |
} |
401 |
|
402 |
static void kvm_set_phys_mem(target_phys_addr_t start_addr, |
403 |
ram_addr_t size, |
404 |
ram_addr_t phys_offset) |
405 |
{ |
406 |
KVMState *s = kvm_state; |
407 |
ram_addr_t flags = phys_offset & ~TARGET_PAGE_MASK; |
408 |
KVMSlot *mem, old; |
409 |
int err;
|
410 |
|
411 |
if (start_addr & ~TARGET_PAGE_MASK) {
|
412 |
if (flags >= IO_MEM_UNASSIGNED) {
|
413 |
if (!kvm_lookup_overlapping_slot(s, start_addr,
|
414 |
start_addr + size)) { |
415 |
return;
|
416 |
} |
417 |
fprintf(stderr, "Unaligned split of a KVM memory slot\n");
|
418 |
} else {
|
419 |
fprintf(stderr, "Only page-aligned memory slots supported\n");
|
420 |
} |
421 |
abort(); |
422 |
} |
423 |
|
424 |
/* KVM does not support read-only slots */
|
425 |
phys_offset &= ~IO_MEM_ROM; |
426 |
|
427 |
while (1) { |
428 |
mem = kvm_lookup_overlapping_slot(s, start_addr, start_addr + size); |
429 |
if (!mem) {
|
430 |
break;
|
431 |
} |
432 |
|
433 |
if (flags < IO_MEM_UNASSIGNED && start_addr >= mem->start_addr &&
|
434 |
(start_addr + size <= mem->start_addr + mem->memory_size) && |
435 |
(phys_offset - start_addr == mem->phys_offset - mem->start_addr)) { |
436 |
/* The new slot fits into the existing one and comes with
|
437 |
* identical parameters - nothing to be done. */
|
438 |
return;
|
439 |
} |
440 |
|
441 |
old = *mem; |
442 |
|
443 |
/* unregister the overlapping slot */
|
444 |
mem->memory_size = 0;
|
445 |
err = kvm_set_user_memory_region(s, mem); |
446 |
if (err) {
|
447 |
fprintf(stderr, "%s: error unregistering overlapping slot: %s\n",
|
448 |
__func__, strerror(-err)); |
449 |
abort(); |
450 |
} |
451 |
|
452 |
/* Workaround for older KVM versions: we can't join slots, even not by
|
453 |
* unregistering the previous ones and then registering the larger
|
454 |
* slot. We have to maintain the existing fragmentation. Sigh.
|
455 |
*
|
456 |
* This workaround assumes that the new slot starts at the same
|
457 |
* address as the first existing one. If not or if some overlapping
|
458 |
* slot comes around later, we will fail (not seen in practice so far)
|
459 |
* - and actually require a recent KVM version. */
|
460 |
if (s->broken_set_mem_region &&
|
461 |
old.start_addr == start_addr && old.memory_size < size && |
462 |
flags < IO_MEM_UNASSIGNED) { |
463 |
mem = kvm_alloc_slot(s); |
464 |
mem->memory_size = old.memory_size; |
465 |
mem->start_addr = old.start_addr; |
466 |
mem->phys_offset = old.phys_offset; |
467 |
mem->flags = 0;
|
468 |
|
469 |
err = kvm_set_user_memory_region(s, mem); |
470 |
if (err) {
|
471 |
fprintf(stderr, "%s: error updating slot: %s\n", __func__,
|
472 |
strerror(-err)); |
473 |
abort(); |
474 |
} |
475 |
|
476 |
start_addr += old.memory_size; |
477 |
phys_offset += old.memory_size; |
478 |
size -= old.memory_size; |
479 |
continue;
|
480 |
} |
481 |
|
482 |
/* register prefix slot */
|
483 |
if (old.start_addr < start_addr) {
|
484 |
mem = kvm_alloc_slot(s); |
485 |
mem->memory_size = start_addr - old.start_addr; |
486 |
mem->start_addr = old.start_addr; |
487 |
mem->phys_offset = old.phys_offset; |
488 |
mem->flags = 0;
|
489 |
|
490 |
err = kvm_set_user_memory_region(s, mem); |
491 |
if (err) {
|
492 |
fprintf(stderr, "%s: error registering prefix slot: %s\n",
|
493 |
__func__, strerror(-err)); |
494 |
abort(); |
495 |
} |
496 |
} |
497 |
|
498 |
/* register suffix slot */
|
499 |
if (old.start_addr + old.memory_size > start_addr + size) {
|
500 |
ram_addr_t size_delta; |
501 |
|
502 |
mem = kvm_alloc_slot(s); |
503 |
mem->start_addr = start_addr + size; |
504 |
size_delta = mem->start_addr - old.start_addr; |
505 |
mem->memory_size = old.memory_size - size_delta; |
506 |
mem->phys_offset = old.phys_offset + size_delta; |
507 |
mem->flags = 0;
|
508 |
|
509 |
err = kvm_set_user_memory_region(s, mem); |
510 |
if (err) {
|
511 |
fprintf(stderr, "%s: error registering suffix slot: %s\n",
|
512 |
__func__, strerror(-err)); |
513 |
abort(); |
514 |
} |
515 |
} |
516 |
} |
517 |
|
518 |
/* in case the KVM bug workaround already "consumed" the new slot */
|
519 |
if (!size)
|
520 |
return;
|
521 |
|
522 |
/* KVM does not need to know about this memory */
|
523 |
if (flags >= IO_MEM_UNASSIGNED)
|
524 |
return;
|
525 |
|
526 |
mem = kvm_alloc_slot(s); |
527 |
mem->memory_size = size; |
528 |
mem->start_addr = start_addr; |
529 |
mem->phys_offset = phys_offset; |
530 |
mem->flags = 0;
|
531 |
|
532 |
err = kvm_set_user_memory_region(s, mem); |
533 |
if (err) {
|
534 |
fprintf(stderr, "%s: error registering slot: %s\n", __func__,
|
535 |
strerror(-err)); |
536 |
abort(); |
537 |
} |
538 |
} |
539 |
|
540 |
static void kvm_client_set_memory(struct CPUPhysMemoryClient *client, |
541 |
target_phys_addr_t start_addr, |
542 |
ram_addr_t size, |
543 |
ram_addr_t phys_offset) |
544 |
{ |
545 |
kvm_set_phys_mem(start_addr, size, phys_offset); |
546 |
} |
547 |
|
548 |
static int kvm_client_sync_dirty_bitmap(struct CPUPhysMemoryClient *client, |
549 |
target_phys_addr_t start_addr, |
550 |
target_phys_addr_t end_addr) |
551 |
{ |
552 |
return kvm_physical_sync_dirty_bitmap(start_addr, end_addr);
|
553 |
} |
554 |
|
555 |
static int kvm_client_migration_log(struct CPUPhysMemoryClient *client, |
556 |
int enable)
|
557 |
{ |
558 |
return kvm_set_migration_log(enable);
|
559 |
} |
560 |
|
561 |
static CPUPhysMemoryClient kvm_cpu_phys_memory_client = {
|
562 |
.set_memory = kvm_client_set_memory, |
563 |
.sync_dirty_bitmap = kvm_client_sync_dirty_bitmap, |
564 |
.migration_log = kvm_client_migration_log, |
565 |
}; |
566 |
|
567 |
int kvm_init(int smp_cpus) |
568 |
{ |
569 |
static const char upgrade_note[] = |
570 |
"Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
|
571 |
"(see http://sourceforge.net/projects/kvm).\n";
|
572 |
KVMState *s; |
573 |
int ret;
|
574 |
int i;
|
575 |
|
576 |
if (smp_cpus > 1) { |
577 |
fprintf(stderr, "No SMP KVM support, use '-smp 1'\n");
|
578 |
return -EINVAL;
|
579 |
} |
580 |
|
581 |
s = qemu_mallocz(sizeof(KVMState));
|
582 |
|
583 |
#ifdef KVM_CAP_SET_GUEST_DEBUG
|
584 |
QTAILQ_INIT(&s->kvm_sw_breakpoints); |
585 |
#endif
|
586 |
for (i = 0; i < ARRAY_SIZE(s->slots); i++) |
587 |
s->slots[i].slot = i; |
588 |
|
589 |
s->vmfd = -1;
|
590 |
s->fd = qemu_open("/dev/kvm", O_RDWR);
|
591 |
if (s->fd == -1) { |
592 |
fprintf(stderr, "Could not access KVM kernel module: %m\n");
|
593 |
ret = -errno; |
594 |
goto err;
|
595 |
} |
596 |
|
597 |
ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0);
|
598 |
if (ret < KVM_API_VERSION) {
|
599 |
if (ret > 0) |
600 |
ret = -EINVAL; |
601 |
fprintf(stderr, "kvm version too old\n");
|
602 |
goto err;
|
603 |
} |
604 |
|
605 |
if (ret > KVM_API_VERSION) {
|
606 |
ret = -EINVAL; |
607 |
fprintf(stderr, "kvm version not supported\n");
|
608 |
goto err;
|
609 |
} |
610 |
|
611 |
s->vmfd = kvm_ioctl(s, KVM_CREATE_VM, 0);
|
612 |
if (s->vmfd < 0) |
613 |
goto err;
|
614 |
|
615 |
/* initially, KVM allocated its own memory and we had to jump through
|
616 |
* hooks to make phys_ram_base point to this. Modern versions of KVM
|
617 |
* just use a user allocated buffer so we can use regular pages
|
618 |
* unmodified. Make sure we have a sufficiently modern version of KVM.
|
619 |
*/
|
620 |
if (!kvm_check_extension(s, KVM_CAP_USER_MEMORY)) {
|
621 |
ret = -EINVAL; |
622 |
fprintf(stderr, "kvm does not support KVM_CAP_USER_MEMORY\n%s",
|
623 |
upgrade_note); |
624 |
goto err;
|
625 |
} |
626 |
|
627 |
/* There was a nasty bug in < kvm-80 that prevents memory slots from being
|
628 |
* destroyed properly. Since we rely on this capability, refuse to work
|
629 |
* with any kernel without this capability. */
|
630 |
if (!kvm_check_extension(s, KVM_CAP_DESTROY_MEMORY_REGION_WORKS)) {
|
631 |
ret = -EINVAL; |
632 |
|
633 |
fprintf(stderr, |
634 |
"KVM kernel module broken (DESTROY_MEMORY_REGION).\n%s",
|
635 |
upgrade_note); |
636 |
goto err;
|
637 |
} |
638 |
|
639 |
s->coalesced_mmio = 0;
|
640 |
#ifdef KVM_CAP_COALESCED_MMIO
|
641 |
s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO); |
642 |
s->coalesced_mmio_ring = NULL;
|
643 |
#endif
|
644 |
|
645 |
s->broken_set_mem_region = 1;
|
646 |
#ifdef KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
|
647 |
ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS); |
648 |
if (ret > 0) { |
649 |
s->broken_set_mem_region = 0;
|
650 |
} |
651 |
#endif
|
652 |
|
653 |
s->vcpu_events = 0;
|
654 |
#ifdef KVM_CAP_VCPU_EVENTS
|
655 |
s->vcpu_events = kvm_check_extension(s, KVM_CAP_VCPU_EVENTS); |
656 |
#endif
|
657 |
|
658 |
s->robust_singlestep = 0;
|
659 |
#ifdef KVM_CAP_X86_ROBUST_SINGLESTEP
|
660 |
s->robust_singlestep = |
661 |
kvm_check_extension(s, KVM_CAP_X86_ROBUST_SINGLESTEP); |
662 |
#endif
|
663 |
|
664 |
ret = kvm_arch_init(s, smp_cpus); |
665 |
if (ret < 0) |
666 |
goto err;
|
667 |
|
668 |
kvm_state = s; |
669 |
cpu_register_phys_memory_client(&kvm_cpu_phys_memory_client); |
670 |
|
671 |
return 0; |
672 |
|
673 |
err:
|
674 |
if (s) {
|
675 |
if (s->vmfd != -1) |
676 |
close(s->vmfd); |
677 |
if (s->fd != -1) |
678 |
close(s->fd); |
679 |
} |
680 |
qemu_free(s); |
681 |
|
682 |
return ret;
|
683 |
} |
684 |
|
685 |
static int kvm_handle_io(uint16_t port, void *data, int direction, int size, |
686 |
uint32_t count) |
687 |
{ |
688 |
int i;
|
689 |
uint8_t *ptr = data; |
690 |
|
691 |
for (i = 0; i < count; i++) { |
692 |
if (direction == KVM_EXIT_IO_IN) {
|
693 |
switch (size) {
|
694 |
case 1: |
695 |
stb_p(ptr, cpu_inb(port)); |
696 |
break;
|
697 |
case 2: |
698 |
stw_p(ptr, cpu_inw(port)); |
699 |
break;
|
700 |
case 4: |
701 |
stl_p(ptr, cpu_inl(port)); |
702 |
break;
|
703 |
} |
704 |
} else {
|
705 |
switch (size) {
|
706 |
case 1: |
707 |
cpu_outb(port, ldub_p(ptr)); |
708 |
break;
|
709 |
case 2: |
710 |
cpu_outw(port, lduw_p(ptr)); |
711 |
break;
|
712 |
case 4: |
713 |
cpu_outl(port, ldl_p(ptr)); |
714 |
break;
|
715 |
} |
716 |
} |
717 |
|
718 |
ptr += size; |
719 |
} |
720 |
|
721 |
return 1; |
722 |
} |
723 |
|
724 |
void kvm_flush_coalesced_mmio_buffer(void) |
725 |
{ |
726 |
#ifdef KVM_CAP_COALESCED_MMIO
|
727 |
KVMState *s = kvm_state; |
728 |
if (s->coalesced_mmio_ring) {
|
729 |
struct kvm_coalesced_mmio_ring *ring = s->coalesced_mmio_ring;
|
730 |
while (ring->first != ring->last) {
|
731 |
struct kvm_coalesced_mmio *ent;
|
732 |
|
733 |
ent = &ring->coalesced_mmio[ring->first]; |
734 |
|
735 |
cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len); |
736 |
smp_wmb(); |
737 |
ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX;
|
738 |
} |
739 |
} |
740 |
#endif
|
741 |
} |
742 |
|
743 |
void kvm_cpu_synchronize_state(CPUState *env)
|
744 |
{ |
745 |
if (!env->kvm_vcpu_dirty) {
|
746 |
kvm_arch_get_registers(env); |
747 |
env->kvm_vcpu_dirty = 1;
|
748 |
} |
749 |
} |
750 |
|
751 |
void kvm_cpu_synchronize_post_reset(CPUState *env)
|
752 |
{ |
753 |
kvm_arch_put_registers(env, KVM_PUT_RESET_STATE); |
754 |
env->kvm_vcpu_dirty = 0;
|
755 |
} |
756 |
|
757 |
void kvm_cpu_synchronize_post_init(CPUState *env)
|
758 |
{ |
759 |
kvm_arch_put_registers(env, KVM_PUT_FULL_STATE); |
760 |
env->kvm_vcpu_dirty = 0;
|
761 |
} |
762 |
|
763 |
int kvm_cpu_exec(CPUState *env)
|
764 |
{ |
765 |
struct kvm_run *run = env->kvm_run;
|
766 |
int ret;
|
767 |
|
768 |
dprintf("kvm_cpu_exec()\n");
|
769 |
|
770 |
do {
|
771 |
#ifndef CONFIG_IOTHREAD
|
772 |
if (env->exit_request) {
|
773 |
dprintf("interrupt exit requested\n");
|
774 |
ret = 0;
|
775 |
break;
|
776 |
} |
777 |
#endif
|
778 |
|
779 |
if (env->kvm_vcpu_dirty) {
|
780 |
kvm_arch_put_registers(env, KVM_PUT_RUNTIME_STATE); |
781 |
env->kvm_vcpu_dirty = 0;
|
782 |
} |
783 |
|
784 |
kvm_arch_pre_run(env, run); |
785 |
qemu_mutex_unlock_iothread(); |
786 |
ret = kvm_vcpu_ioctl(env, KVM_RUN, 0);
|
787 |
qemu_mutex_lock_iothread(); |
788 |
kvm_arch_post_run(env, run); |
789 |
|
790 |
if (ret == -EINTR || ret == -EAGAIN) {
|
791 |
cpu_exit(env); |
792 |
dprintf("io window exit\n");
|
793 |
ret = 0;
|
794 |
break;
|
795 |
} |
796 |
|
797 |
if (ret < 0) { |
798 |
dprintf("kvm run failed %s\n", strerror(-ret));
|
799 |
abort(); |
800 |
} |
801 |
|
802 |
kvm_flush_coalesced_mmio_buffer(); |
803 |
|
804 |
ret = 0; /* exit loop */ |
805 |
switch (run->exit_reason) {
|
806 |
case KVM_EXIT_IO:
|
807 |
dprintf("handle_io\n");
|
808 |
ret = kvm_handle_io(run->io.port, |
809 |
(uint8_t *)run + run->io.data_offset, |
810 |
run->io.direction, |
811 |
run->io.size, |
812 |
run->io.count); |
813 |
break;
|
814 |
case KVM_EXIT_MMIO:
|
815 |
dprintf("handle_mmio\n");
|
816 |
cpu_physical_memory_rw(run->mmio.phys_addr, |
817 |
run->mmio.data, |
818 |
run->mmio.len, |
819 |
run->mmio.is_write); |
820 |
ret = 1;
|
821 |
break;
|
822 |
case KVM_EXIT_IRQ_WINDOW_OPEN:
|
823 |
dprintf("irq_window_open\n");
|
824 |
break;
|
825 |
case KVM_EXIT_SHUTDOWN:
|
826 |
dprintf("shutdown\n");
|
827 |
qemu_system_reset_request(); |
828 |
ret = 1;
|
829 |
break;
|
830 |
case KVM_EXIT_UNKNOWN:
|
831 |
dprintf("kvm_exit_unknown\n");
|
832 |
break;
|
833 |
case KVM_EXIT_FAIL_ENTRY:
|
834 |
dprintf("kvm_exit_fail_entry\n");
|
835 |
break;
|
836 |
case KVM_EXIT_EXCEPTION:
|
837 |
dprintf("kvm_exit_exception\n");
|
838 |
break;
|
839 |
case KVM_EXIT_DEBUG:
|
840 |
dprintf("kvm_exit_debug\n");
|
841 |
#ifdef KVM_CAP_SET_GUEST_DEBUG
|
842 |
if (kvm_arch_debug(&run->debug.arch)) {
|
843 |
gdb_set_stop_cpu(env); |
844 |
vm_stop(EXCP_DEBUG); |
845 |
env->exception_index = EXCP_DEBUG; |
846 |
return 0; |
847 |
} |
848 |
/* re-enter, this exception was guest-internal */
|
849 |
ret = 1;
|
850 |
#endif /* KVM_CAP_SET_GUEST_DEBUG */ |
851 |
break;
|
852 |
default:
|
853 |
dprintf("kvm_arch_handle_exit\n");
|
854 |
ret = kvm_arch_handle_exit(env, run); |
855 |
break;
|
856 |
} |
857 |
} while (ret > 0); |
858 |
|
859 |
if (env->exit_request) {
|
860 |
env->exit_request = 0;
|
861 |
env->exception_index = EXCP_INTERRUPT; |
862 |
} |
863 |
|
864 |
return ret;
|
865 |
} |
866 |
|
867 |
int kvm_ioctl(KVMState *s, int type, ...) |
868 |
{ |
869 |
int ret;
|
870 |
void *arg;
|
871 |
va_list ap; |
872 |
|
873 |
va_start(ap, type); |
874 |
arg = va_arg(ap, void *);
|
875 |
va_end(ap); |
876 |
|
877 |
ret = ioctl(s->fd, type, arg); |
878 |
if (ret == -1) |
879 |
ret = -errno; |
880 |
|
881 |
return ret;
|
882 |
} |
883 |
|
884 |
int kvm_vm_ioctl(KVMState *s, int type, ...) |
885 |
{ |
886 |
int ret;
|
887 |
void *arg;
|
888 |
va_list ap; |
889 |
|
890 |
va_start(ap, type); |
891 |
arg = va_arg(ap, void *);
|
892 |
va_end(ap); |
893 |
|
894 |
ret = ioctl(s->vmfd, type, arg); |
895 |
if (ret == -1) |
896 |
ret = -errno; |
897 |
|
898 |
return ret;
|
899 |
} |
900 |
|
901 |
int kvm_vcpu_ioctl(CPUState *env, int type, ...) |
902 |
{ |
903 |
int ret;
|
904 |
void *arg;
|
905 |
va_list ap; |
906 |
|
907 |
va_start(ap, type); |
908 |
arg = va_arg(ap, void *);
|
909 |
va_end(ap); |
910 |
|
911 |
ret = ioctl(env->kvm_fd, type, arg); |
912 |
if (ret == -1) |
913 |
ret = -errno; |
914 |
|
915 |
return ret;
|
916 |
} |
917 |
|
918 |
int kvm_has_sync_mmu(void) |
919 |
{ |
920 |
#ifdef KVM_CAP_SYNC_MMU
|
921 |
KVMState *s = kvm_state; |
922 |
|
923 |
return kvm_check_extension(s, KVM_CAP_SYNC_MMU);
|
924 |
#else
|
925 |
return 0; |
926 |
#endif
|
927 |
} |
928 |
|
929 |
int kvm_has_vcpu_events(void) |
930 |
{ |
931 |
return kvm_state->vcpu_events;
|
932 |
} |
933 |
|
934 |
int kvm_has_robust_singlestep(void) |
935 |
{ |
936 |
return kvm_state->robust_singlestep;
|
937 |
} |
938 |
|
939 |
void kvm_setup_guest_memory(void *start, size_t size) |
940 |
{ |
941 |
if (!kvm_has_sync_mmu()) {
|
942 |
#ifdef MADV_DONTFORK
|
943 |
int ret = madvise(start, size, MADV_DONTFORK);
|
944 |
|
945 |
if (ret) {
|
946 |
perror("madvice");
|
947 |
exit(1);
|
948 |
} |
949 |
#else
|
950 |
fprintf(stderr, |
951 |
"Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
|
952 |
exit(1);
|
953 |
#endif
|
954 |
} |
955 |
} |
956 |
|
957 |
#ifdef KVM_CAP_SET_GUEST_DEBUG
|
958 |
static void on_vcpu(CPUState *env, void (*func)(void *data), void *data) |
959 |
{ |
960 |
#ifdef CONFIG_IOTHREAD
|
961 |
if (env != cpu_single_env) {
|
962 |
abort(); |
963 |
} |
964 |
#endif
|
965 |
func(data); |
966 |
} |
967 |
|
968 |
struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *env,
|
969 |
target_ulong pc) |
970 |
{ |
971 |
struct kvm_sw_breakpoint *bp;
|
972 |
|
973 |
QTAILQ_FOREACH(bp, &env->kvm_state->kvm_sw_breakpoints, entry) { |
974 |
if (bp->pc == pc)
|
975 |
return bp;
|
976 |
} |
977 |
return NULL; |
978 |
} |
979 |
|
980 |
int kvm_sw_breakpoints_active(CPUState *env)
|
981 |
{ |
982 |
return !QTAILQ_EMPTY(&env->kvm_state->kvm_sw_breakpoints);
|
983 |
} |
984 |
|
985 |
struct kvm_set_guest_debug_data {
|
986 |
struct kvm_guest_debug dbg;
|
987 |
CPUState *env; |
988 |
int err;
|
989 |
}; |
990 |
|
991 |
static void kvm_invoke_set_guest_debug(void *data) |
992 |
{ |
993 |
struct kvm_set_guest_debug_data *dbg_data = data;
|
994 |
CPUState *env = dbg_data->env; |
995 |
|
996 |
dbg_data->err = kvm_vcpu_ioctl(env, KVM_SET_GUEST_DEBUG, &dbg_data->dbg); |
997 |
} |
998 |
|
999 |
int kvm_update_guest_debug(CPUState *env, unsigned long reinject_trap) |
1000 |
{ |
1001 |
struct kvm_set_guest_debug_data data;
|
1002 |
|
1003 |
data.dbg.control = reinject_trap; |
1004 |
|
1005 |
if (env->singlestep_enabled) {
|
1006 |
data.dbg.control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP; |
1007 |
} |
1008 |
kvm_arch_update_guest_debug(env, &data.dbg); |
1009 |
data.env = env; |
1010 |
|
1011 |
on_vcpu(env, kvm_invoke_set_guest_debug, &data); |
1012 |
return data.err;
|
1013 |
} |
1014 |
|
1015 |
int kvm_insert_breakpoint(CPUState *current_env, target_ulong addr,
|
1016 |
target_ulong len, int type)
|
1017 |
{ |
1018 |
struct kvm_sw_breakpoint *bp;
|
1019 |
CPUState *env; |
1020 |
int err;
|
1021 |
|
1022 |
if (type == GDB_BREAKPOINT_SW) {
|
1023 |
bp = kvm_find_sw_breakpoint(current_env, addr); |
1024 |
if (bp) {
|
1025 |
bp->use_count++; |
1026 |
return 0; |
1027 |
} |
1028 |
|
1029 |
bp = qemu_malloc(sizeof(struct kvm_sw_breakpoint)); |
1030 |
if (!bp)
|
1031 |
return -ENOMEM;
|
1032 |
|
1033 |
bp->pc = addr; |
1034 |
bp->use_count = 1;
|
1035 |
err = kvm_arch_insert_sw_breakpoint(current_env, bp); |
1036 |
if (err) {
|
1037 |
free(bp); |
1038 |
return err;
|
1039 |
} |
1040 |
|
1041 |
QTAILQ_INSERT_HEAD(¤t_env->kvm_state->kvm_sw_breakpoints, |
1042 |
bp, entry); |
1043 |
} else {
|
1044 |
err = kvm_arch_insert_hw_breakpoint(addr, len, type); |
1045 |
if (err)
|
1046 |
return err;
|
1047 |
} |
1048 |
|
1049 |
for (env = first_cpu; env != NULL; env = env->next_cpu) { |
1050 |
err = kvm_update_guest_debug(env, 0);
|
1051 |
if (err)
|
1052 |
return err;
|
1053 |
} |
1054 |
return 0; |
1055 |
} |
1056 |
|
1057 |
int kvm_remove_breakpoint(CPUState *current_env, target_ulong addr,
|
1058 |
target_ulong len, int type)
|
1059 |
{ |
1060 |
struct kvm_sw_breakpoint *bp;
|
1061 |
CPUState *env; |
1062 |
int err;
|
1063 |
|
1064 |
if (type == GDB_BREAKPOINT_SW) {
|
1065 |
bp = kvm_find_sw_breakpoint(current_env, addr); |
1066 |
if (!bp)
|
1067 |
return -ENOENT;
|
1068 |
|
1069 |
if (bp->use_count > 1) { |
1070 |
bp->use_count--; |
1071 |
return 0; |
1072 |
} |
1073 |
|
1074 |
err = kvm_arch_remove_sw_breakpoint(current_env, bp); |
1075 |
if (err)
|
1076 |
return err;
|
1077 |
|
1078 |
QTAILQ_REMOVE(¤t_env->kvm_state->kvm_sw_breakpoints, bp, entry); |
1079 |
qemu_free(bp); |
1080 |
} else {
|
1081 |
err = kvm_arch_remove_hw_breakpoint(addr, len, type); |
1082 |
if (err)
|
1083 |
return err;
|
1084 |
} |
1085 |
|
1086 |
for (env = first_cpu; env != NULL; env = env->next_cpu) { |
1087 |
err = kvm_update_guest_debug(env, 0);
|
1088 |
if (err)
|
1089 |
return err;
|
1090 |
} |
1091 |
return 0; |
1092 |
} |
1093 |
|
1094 |
void kvm_remove_all_breakpoints(CPUState *current_env)
|
1095 |
{ |
1096 |
struct kvm_sw_breakpoint *bp, *next;
|
1097 |
KVMState *s = current_env->kvm_state; |
1098 |
CPUState *env; |
1099 |
|
1100 |
QTAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) { |
1101 |
if (kvm_arch_remove_sw_breakpoint(current_env, bp) != 0) { |
1102 |
/* Try harder to find a CPU that currently sees the breakpoint. */
|
1103 |
for (env = first_cpu; env != NULL; env = env->next_cpu) { |
1104 |
if (kvm_arch_remove_sw_breakpoint(env, bp) == 0) |
1105 |
break;
|
1106 |
} |
1107 |
} |
1108 |
} |
1109 |
kvm_arch_remove_all_hw_breakpoints(); |
1110 |
|
1111 |
for (env = first_cpu; env != NULL; env = env->next_cpu) |
1112 |
kvm_update_guest_debug(env, 0);
|
1113 |
} |
1114 |
|
1115 |
#else /* !KVM_CAP_SET_GUEST_DEBUG */ |
1116 |
|
1117 |
int kvm_update_guest_debug(CPUState *env, unsigned long reinject_trap) |
1118 |
{ |
1119 |
return -EINVAL;
|
1120 |
} |
1121 |
|
1122 |
int kvm_insert_breakpoint(CPUState *current_env, target_ulong addr,
|
1123 |
target_ulong len, int type)
|
1124 |
{ |
1125 |
return -EINVAL;
|
1126 |
} |
1127 |
|
1128 |
int kvm_remove_breakpoint(CPUState *current_env, target_ulong addr,
|
1129 |
target_ulong len, int type)
|
1130 |
{ |
1131 |
return -EINVAL;
|
1132 |
} |
1133 |
|
1134 |
void kvm_remove_all_breakpoints(CPUState *current_env)
|
1135 |
{ |
1136 |
} |
1137 |
#endif /* !KVM_CAP_SET_GUEST_DEBUG */ |
1138 |
|
1139 |
int kvm_set_signal_mask(CPUState *env, const sigset_t *sigset) |
1140 |
{ |
1141 |
struct kvm_signal_mask *sigmask;
|
1142 |
int r;
|
1143 |
|
1144 |
if (!sigset)
|
1145 |
return kvm_vcpu_ioctl(env, KVM_SET_SIGNAL_MASK, NULL); |
1146 |
|
1147 |
sigmask = qemu_malloc(sizeof(*sigmask) + sizeof(*sigset)); |
1148 |
|
1149 |
sigmask->len = 8;
|
1150 |
memcpy(sigmask->sigset, sigset, sizeof(*sigset));
|
1151 |
r = kvm_vcpu_ioctl(env, KVM_SET_SIGNAL_MASK, sigmask); |
1152 |
free(sigmask); |
1153 |
|
1154 |
return r;
|
1155 |
} |