]>
Commit | Line | Data |
---|---|---|
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/atomic.h" | |
25 | #include "qemu/option.h" | |
26 | #include "qemu/config-file.h" | |
27 | #include "sysemu/sysemu.h" | |
28 | #include "hw/hw.h" | |
29 | #include "hw/pci/msi.h" | |
30 | #include "exec/gdbstub.h" | |
31 | #include "sysemu/kvm.h" | |
32 | #include "qemu/bswap.h" | |
33 | #include "exec/memory.h" | |
34 | #include "exec/address-spaces.h" | |
35 | #include "qemu/event_notifier.h" | |
36 | #include "trace.h" | |
37 | ||
38 | /* This check must be after config-host.h is included */ | |
39 | #ifdef CONFIG_EVENTFD | |
40 | #include <sys/eventfd.h> | |
41 | #endif | |
42 | ||
43 | #ifdef CONFIG_VALGRIND_H | |
44 | #include <valgrind/memcheck.h> | |
45 | #endif | |
46 | ||
47 | /* KVM uses PAGE_SIZE in its definition of COALESCED_MMIO_MAX */ | |
48 | #define PAGE_SIZE TARGET_PAGE_SIZE | |
49 | ||
50 | //#define DEBUG_KVM | |
51 | ||
52 | #ifdef DEBUG_KVM | |
53 | #define DPRINTF(fmt, ...) \ | |
54 | do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0) | |
55 | #else | |
56 | #define DPRINTF(fmt, ...) \ | |
57 | do { } while (0) | |
58 | #endif | |
59 | ||
60 | #define KVM_MSI_HASHTAB_SIZE 256 | |
61 | ||
62 | typedef struct KVMSlot | |
63 | { | |
64 | hwaddr start_addr; | |
65 | ram_addr_t memory_size; | |
66 | void *ram; | |
67 | int slot; | |
68 | int flags; | |
69 | } KVMSlot; | |
70 | ||
71 | typedef struct kvm_dirty_log KVMDirtyLog; | |
72 | ||
73 | struct KVMState | |
74 | { | |
75 | KVMSlot slots[32]; | |
76 | int fd; | |
77 | int vmfd; | |
78 | int coalesced_mmio; | |
79 | struct kvm_coalesced_mmio_ring *coalesced_mmio_ring; | |
80 | bool coalesced_flush_in_progress; | |
81 | int broken_set_mem_region; | |
82 | int migration_log; | |
83 | int vcpu_events; | |
84 | int robust_singlestep; | |
85 | int debugregs; | |
86 | #ifdef KVM_CAP_SET_GUEST_DEBUG | |
87 | struct kvm_sw_breakpoint_head kvm_sw_breakpoints; | |
88 | #endif | |
89 | int pit_state2; | |
90 | int xsave, xcrs; | |
91 | int many_ioeventfds; | |
92 | int intx_set_mask; | |
93 | /* The man page (and posix) say ioctl numbers are signed int, but | |
94 | * they're not. Linux, glibc and *BSD all treat ioctl numbers as | |
95 | * unsigned, and treating them as signed here can break things */ | |
96 | unsigned irq_set_ioctl; | |
97 | #ifdef KVM_CAP_IRQ_ROUTING | |
98 | struct kvm_irq_routing *irq_routes; | |
99 | int nr_allocated_irq_routes; | |
100 | uint32_t *used_gsi_bitmap; | |
101 | unsigned int gsi_count; | |
102 | QTAILQ_HEAD(msi_hashtab, KVMMSIRoute) msi_hashtab[KVM_MSI_HASHTAB_SIZE]; | |
103 | bool direct_msi; | |
104 | #endif | |
105 | }; | |
106 | ||
107 | KVMState *kvm_state; | |
108 | bool kvm_kernel_irqchip; | |
109 | bool kvm_async_interrupts_allowed; | |
110 | bool kvm_irqfds_allowed; | |
111 | bool kvm_msi_via_irqfd_allowed; | |
112 | bool kvm_gsi_routing_allowed; | |
113 | bool kvm_allowed; | |
114 | ||
115 | static const KVMCapabilityInfo kvm_required_capabilites[] = { | |
116 | KVM_CAP_INFO(USER_MEMORY), | |
117 | KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS), | |
118 | KVM_CAP_LAST_INFO | |
119 | }; | |
120 | ||
121 | static KVMSlot *kvm_alloc_slot(KVMState *s) | |
122 | { | |
123 | int i; | |
124 | ||
125 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
126 | if (s->slots[i].memory_size == 0) { | |
127 | return &s->slots[i]; | |
128 | } | |
129 | } | |
130 | ||
131 | fprintf(stderr, "%s: no free slot available\n", __func__); | |
132 | abort(); | |
133 | } | |
134 | ||
135 | static KVMSlot *kvm_lookup_matching_slot(KVMState *s, | |
136 | hwaddr start_addr, | |
137 | hwaddr end_addr) | |
138 | { | |
139 | int i; | |
140 | ||
141 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
142 | KVMSlot *mem = &s->slots[i]; | |
143 | ||
144 | if (start_addr == mem->start_addr && | |
145 | end_addr == mem->start_addr + mem->memory_size) { | |
146 | return mem; | |
147 | } | |
148 | } | |
149 | ||
150 | return NULL; | |
151 | } | |
152 | ||
153 | /* | |
154 | * Find overlapping slot with lowest start address | |
155 | */ | |
156 | static KVMSlot *kvm_lookup_overlapping_slot(KVMState *s, | |
157 | hwaddr start_addr, | |
158 | hwaddr end_addr) | |
159 | { | |
160 | KVMSlot *found = NULL; | |
161 | int i; | |
162 | ||
163 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
164 | KVMSlot *mem = &s->slots[i]; | |
165 | ||
166 | if (mem->memory_size == 0 || | |
167 | (found && found->start_addr < mem->start_addr)) { | |
168 | continue; | |
169 | } | |
170 | ||
171 | if (end_addr > mem->start_addr && | |
172 | start_addr < mem->start_addr + mem->memory_size) { | |
173 | found = mem; | |
174 | } | |
175 | } | |
176 | ||
177 | return found; | |
178 | } | |
179 | ||
180 | int kvm_physical_memory_addr_from_host(KVMState *s, void *ram, | |
181 | hwaddr *phys_addr) | |
182 | { | |
183 | int i; | |
184 | ||
185 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
186 | KVMSlot *mem = &s->slots[i]; | |
187 | ||
188 | if (ram >= mem->ram && ram < mem->ram + mem->memory_size) { | |
189 | *phys_addr = mem->start_addr + (ram - mem->ram); | |
190 | return 1; | |
191 | } | |
192 | } | |
193 | ||
194 | return 0; | |
195 | } | |
196 | ||
197 | static int kvm_set_user_memory_region(KVMState *s, KVMSlot *slot) | |
198 | { | |
199 | struct kvm_userspace_memory_region mem; | |
200 | ||
201 | mem.slot = slot->slot; | |
202 | mem.guest_phys_addr = slot->start_addr; | |
203 | mem.memory_size = slot->memory_size; | |
204 | mem.userspace_addr = (unsigned long)slot->ram; | |
205 | mem.flags = slot->flags; | |
206 | if (s->migration_log) { | |
207 | mem.flags |= KVM_MEM_LOG_DIRTY_PAGES; | |
208 | } | |
209 | return kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem); | |
210 | } | |
211 | ||
212 | static void kvm_reset_vcpu(void *opaque) | |
213 | { | |
214 | CPUState *cpu = opaque; | |
215 | ||
216 | kvm_arch_reset_vcpu(cpu); | |
217 | } | |
218 | ||
219 | int kvm_init_vcpu(CPUState *cpu) | |
220 | { | |
221 | KVMState *s = kvm_state; | |
222 | long mmap_size; | |
223 | int ret; | |
224 | ||
225 | DPRINTF("kvm_init_vcpu\n"); | |
226 | ||
227 | ret = kvm_vm_ioctl(s, KVM_CREATE_VCPU, (void *)kvm_arch_vcpu_id(cpu)); | |
228 | if (ret < 0) { | |
229 | DPRINTF("kvm_create_vcpu failed\n"); | |
230 | goto err; | |
231 | } | |
232 | ||
233 | cpu->kvm_fd = ret; | |
234 | cpu->kvm_state = s; | |
235 | cpu->kvm_vcpu_dirty = true; | |
236 | ||
237 | mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0); | |
238 | if (mmap_size < 0) { | |
239 | ret = mmap_size; | |
240 | DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n"); | |
241 | goto err; | |
242 | } | |
243 | ||
244 | cpu->kvm_run = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED, | |
245 | cpu->kvm_fd, 0); | |
246 | if (cpu->kvm_run == MAP_FAILED) { | |
247 | ret = -errno; | |
248 | DPRINTF("mmap'ing vcpu state failed\n"); | |
249 | goto err; | |
250 | } | |
251 | ||
252 | if (s->coalesced_mmio && !s->coalesced_mmio_ring) { | |
253 | s->coalesced_mmio_ring = | |
254 | (void *)cpu->kvm_run + s->coalesced_mmio * PAGE_SIZE; | |
255 | } | |
256 | ||
257 | ret = kvm_arch_init_vcpu(cpu); | |
258 | if (ret == 0) { | |
259 | qemu_register_reset(kvm_reset_vcpu, cpu); | |
260 | kvm_arch_reset_vcpu(cpu); | |
261 | } | |
262 | err: | |
263 | return ret; | |
264 | } | |
265 | ||
266 | /* | |
267 | * dirty pages logging control | |
268 | */ | |
269 | ||
270 | static int kvm_mem_flags(KVMState *s, bool log_dirty) | |
271 | { | |
272 | return log_dirty ? KVM_MEM_LOG_DIRTY_PAGES : 0; | |
273 | } | |
274 | ||
275 | static int kvm_slot_dirty_pages_log_change(KVMSlot *mem, bool log_dirty) | |
276 | { | |
277 | KVMState *s = kvm_state; | |
278 | int flags, mask = KVM_MEM_LOG_DIRTY_PAGES; | |
279 | int old_flags; | |
280 | ||
281 | old_flags = mem->flags; | |
282 | ||
283 | flags = (mem->flags & ~mask) | kvm_mem_flags(s, log_dirty); | |
284 | mem->flags = flags; | |
285 | ||
286 | /* If nothing changed effectively, no need to issue ioctl */ | |
287 | if (s->migration_log) { | |
288 | flags |= KVM_MEM_LOG_DIRTY_PAGES; | |
289 | } | |
290 | ||
291 | if (flags == old_flags) { | |
292 | return 0; | |
293 | } | |
294 | ||
295 | return kvm_set_user_memory_region(s, mem); | |
296 | } | |
297 | ||
298 | static int kvm_dirty_pages_log_change(hwaddr phys_addr, | |
299 | ram_addr_t size, bool log_dirty) | |
300 | { | |
301 | KVMState *s = kvm_state; | |
302 | KVMSlot *mem = kvm_lookup_matching_slot(s, phys_addr, phys_addr + size); | |
303 | ||
304 | if (mem == NULL) { | |
305 | fprintf(stderr, "BUG: %s: invalid parameters " TARGET_FMT_plx "-" | |
306 | TARGET_FMT_plx "\n", __func__, phys_addr, | |
307 | (hwaddr)(phys_addr + size - 1)); | |
308 | return -EINVAL; | |
309 | } | |
310 | return kvm_slot_dirty_pages_log_change(mem, log_dirty); | |
311 | } | |
312 | ||
313 | static void kvm_log_start(MemoryListener *listener, | |
314 | MemoryRegionSection *section) | |
315 | { | |
316 | int r; | |
317 | ||
318 | r = kvm_dirty_pages_log_change(section->offset_within_address_space, | |
319 | section->size, true); | |
320 | if (r < 0) { | |
321 | abort(); | |
322 | } | |
323 | } | |
324 | ||
325 | static void kvm_log_stop(MemoryListener *listener, | |
326 | MemoryRegionSection *section) | |
327 | { | |
328 | int r; | |
329 | ||
330 | r = kvm_dirty_pages_log_change(section->offset_within_address_space, | |
331 | section->size, false); | |
332 | if (r < 0) { | |
333 | abort(); | |
334 | } | |
335 | } | |
336 | ||
337 | static int kvm_set_migration_log(int enable) | |
338 | { | |
339 | KVMState *s = kvm_state; | |
340 | KVMSlot *mem; | |
341 | int i, err; | |
342 | ||
343 | s->migration_log = enable; | |
344 | ||
345 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
346 | mem = &s->slots[i]; | |
347 | ||
348 | if (!mem->memory_size) { | |
349 | continue; | |
350 | } | |
351 | if (!!(mem->flags & KVM_MEM_LOG_DIRTY_PAGES) == enable) { | |
352 | continue; | |
353 | } | |
354 | err = kvm_set_user_memory_region(s, mem); | |
355 | if (err) { | |
356 | return err; | |
357 | } | |
358 | } | |
359 | return 0; | |
360 | } | |
361 | ||
362 | /* get kvm's dirty pages bitmap and update qemu's */ | |
363 | static int kvm_get_dirty_pages_log_range(MemoryRegionSection *section, | |
364 | unsigned long *bitmap) | |
365 | { | |
366 | unsigned int i, j; | |
367 | unsigned long page_number, c; | |
368 | hwaddr addr, addr1; | |
369 | unsigned int len = ((section->size / getpagesize()) + HOST_LONG_BITS - 1) / HOST_LONG_BITS; | |
370 | unsigned long hpratio = getpagesize() / TARGET_PAGE_SIZE; | |
371 | ||
372 | /* | |
373 | * bitmap-traveling is faster than memory-traveling (for addr...) | |
374 | * especially when most of the memory is not dirty. | |
375 | */ | |
376 | for (i = 0; i < len; i++) { | |
377 | if (bitmap[i] != 0) { | |
378 | c = leul_to_cpu(bitmap[i]); | |
379 | do { | |
380 | j = ffsl(c) - 1; | |
381 | c &= ~(1ul << j); | |
382 | page_number = (i * HOST_LONG_BITS + j) * hpratio; | |
383 | addr1 = page_number * TARGET_PAGE_SIZE; | |
384 | addr = section->offset_within_region + addr1; | |
385 | memory_region_set_dirty(section->mr, addr, | |
386 | TARGET_PAGE_SIZE * hpratio); | |
387 | } while (c != 0); | |
388 | } | |
389 | } | |
390 | return 0; | |
391 | } | |
392 | ||
393 | #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1)) | |
394 | ||
395 | /** | |
396 | * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space | |
397 | * This function updates qemu's dirty bitmap using | |
398 | * memory_region_set_dirty(). This means all bits are set | |
399 | * to dirty. | |
400 | * | |
401 | * @start_add: start of logged region. | |
402 | * @end_addr: end of logged region. | |
403 | */ | |
404 | static int kvm_physical_sync_dirty_bitmap(MemoryRegionSection *section) | |
405 | { | |
406 | KVMState *s = kvm_state; | |
407 | unsigned long size, allocated_size = 0; | |
408 | KVMDirtyLog d; | |
409 | KVMSlot *mem; | |
410 | int ret = 0; | |
411 | hwaddr start_addr = section->offset_within_address_space; | |
412 | hwaddr end_addr = start_addr + section->size; | |
413 | ||
414 | d.dirty_bitmap = NULL; | |
415 | while (start_addr < end_addr) { | |
416 | mem = kvm_lookup_overlapping_slot(s, start_addr, end_addr); | |
417 | if (mem == NULL) { | |
418 | break; | |
419 | } | |
420 | ||
421 | /* XXX bad kernel interface alert | |
422 | * For dirty bitmap, kernel allocates array of size aligned to | |
423 | * bits-per-long. But for case when the kernel is 64bits and | |
424 | * the userspace is 32bits, userspace can't align to the same | |
425 | * bits-per-long, since sizeof(long) is different between kernel | |
426 | * and user space. This way, userspace will provide buffer which | |
427 | * may be 4 bytes less than the kernel will use, resulting in | |
428 | * userspace memory corruption (which is not detectable by valgrind | |
429 | * too, in most cases). | |
430 | * So for now, let's align to 64 instead of HOST_LONG_BITS here, in | |
431 | * a hope that sizeof(long) wont become >8 any time soon. | |
432 | */ | |
433 | size = ALIGN(((mem->memory_size) >> TARGET_PAGE_BITS), | |
434 | /*HOST_LONG_BITS*/ 64) / 8; | |
435 | if (!d.dirty_bitmap) { | |
436 | d.dirty_bitmap = g_malloc(size); | |
437 | } else if (size > allocated_size) { | |
438 | d.dirty_bitmap = g_realloc(d.dirty_bitmap, size); | |
439 | } | |
440 | allocated_size = size; | |
441 | memset(d.dirty_bitmap, 0, allocated_size); | |
442 | ||
443 | d.slot = mem->slot; | |
444 | ||
445 | if (kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d) == -1) { | |
446 | DPRINTF("ioctl failed %d\n", errno); | |
447 | ret = -1; | |
448 | break; | |
449 | } | |
450 | ||
451 | kvm_get_dirty_pages_log_range(section, d.dirty_bitmap); | |
452 | start_addr = mem->start_addr + mem->memory_size; | |
453 | } | |
454 | g_free(d.dirty_bitmap); | |
455 | ||
456 | return ret; | |
457 | } | |
458 | ||
459 | static void kvm_coalesce_mmio_region(MemoryListener *listener, | |
460 | MemoryRegionSection *secion, | |
461 | hwaddr start, hwaddr size) | |
462 | { | |
463 | KVMState *s = kvm_state; | |
464 | ||
465 | if (s->coalesced_mmio) { | |
466 | struct kvm_coalesced_mmio_zone zone; | |
467 | ||
468 | zone.addr = start; | |
469 | zone.size = size; | |
470 | zone.pad = 0; | |
471 | ||
472 | (void)kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone); | |
473 | } | |
474 | } | |
475 | ||
476 | static void kvm_uncoalesce_mmio_region(MemoryListener *listener, | |
477 | MemoryRegionSection *secion, | |
478 | hwaddr start, hwaddr size) | |
479 | { | |
480 | KVMState *s = kvm_state; | |
481 | ||
482 | if (s->coalesced_mmio) { | |
483 | struct kvm_coalesced_mmio_zone zone; | |
484 | ||
485 | zone.addr = start; | |
486 | zone.size = size; | |
487 | zone.pad = 0; | |
488 | ||
489 | (void)kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone); | |
490 | } | |
491 | } | |
492 | ||
493 | int kvm_check_extension(KVMState *s, unsigned int extension) | |
494 | { | |
495 | int ret; | |
496 | ||
497 | ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension); | |
498 | if (ret < 0) { | |
499 | ret = 0; | |
500 | } | |
501 | ||
502 | return ret; | |
503 | } | |
504 | ||
505 | static int kvm_set_ioeventfd_mmio(int fd, uint32_t addr, uint32_t val, | |
506 | bool assign, uint32_t size, bool datamatch) | |
507 | { | |
508 | int ret; | |
509 | struct kvm_ioeventfd iofd; | |
510 | ||
511 | iofd.datamatch = datamatch ? val : 0; | |
512 | iofd.addr = addr; | |
513 | iofd.len = size; | |
514 | iofd.flags = 0; | |
515 | iofd.fd = fd; | |
516 | ||
517 | if (!kvm_enabled()) { | |
518 | return -ENOSYS; | |
519 | } | |
520 | ||
521 | if (datamatch) { | |
522 | iofd.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH; | |
523 | } | |
524 | if (!assign) { | |
525 | iofd.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN; | |
526 | } | |
527 | ||
528 | ret = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &iofd); | |
529 | ||
530 | if (ret < 0) { | |
531 | return -errno; | |
532 | } | |
533 | ||
534 | return 0; | |
535 | } | |
536 | ||
537 | static int kvm_set_ioeventfd_pio(int fd, uint16_t addr, uint16_t val, | |
538 | bool assign, uint32_t size, bool datamatch) | |
539 | { | |
540 | struct kvm_ioeventfd kick = { | |
541 | .datamatch = datamatch ? val : 0, | |
542 | .addr = addr, | |
543 | .flags = KVM_IOEVENTFD_FLAG_PIO, | |
544 | .len = size, | |
545 | .fd = fd, | |
546 | }; | |
547 | int r; | |
548 | if (!kvm_enabled()) { | |
549 | return -ENOSYS; | |
550 | } | |
551 | if (datamatch) { | |
552 | kick.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH; | |
553 | } | |
554 | if (!assign) { | |
555 | kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN; | |
556 | } | |
557 | r = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick); | |
558 | if (r < 0) { | |
559 | return r; | |
560 | } | |
561 | return 0; | |
562 | } | |
563 | ||
564 | ||
565 | static int kvm_check_many_ioeventfds(void) | |
566 | { | |
567 | /* Userspace can use ioeventfd for io notification. This requires a host | |
568 | * that supports eventfd(2) and an I/O thread; since eventfd does not | |
569 | * support SIGIO it cannot interrupt the vcpu. | |
570 | * | |
571 | * Older kernels have a 6 device limit on the KVM io bus. Find out so we | |
572 | * can avoid creating too many ioeventfds. | |
573 | */ | |
574 | #if defined(CONFIG_EVENTFD) | |
575 | int ioeventfds[7]; | |
576 | int i, ret = 0; | |
577 | for (i = 0; i < ARRAY_SIZE(ioeventfds); i++) { | |
578 | ioeventfds[i] = eventfd(0, EFD_CLOEXEC); | |
579 | if (ioeventfds[i] < 0) { | |
580 | break; | |
581 | } | |
582 | ret = kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, true, 2, true); | |
583 | if (ret < 0) { | |
584 | close(ioeventfds[i]); | |
585 | break; | |
586 | } | |
587 | } | |
588 | ||
589 | /* Decide whether many devices are supported or not */ | |
590 | ret = i == ARRAY_SIZE(ioeventfds); | |
591 | ||
592 | while (i-- > 0) { | |
593 | kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, false, 2, true); | |
594 | close(ioeventfds[i]); | |
595 | } | |
596 | return ret; | |
597 | #else | |
598 | return 0; | |
599 | #endif | |
600 | } | |
601 | ||
602 | static const KVMCapabilityInfo * | |
603 | kvm_check_extension_list(KVMState *s, const KVMCapabilityInfo *list) | |
604 | { | |
605 | while (list->name) { | |
606 | if (!kvm_check_extension(s, list->value)) { | |
607 | return list; | |
608 | } | |
609 | list++; | |
610 | } | |
611 | return NULL; | |
612 | } | |
613 | ||
614 | static void kvm_set_phys_mem(MemoryRegionSection *section, bool add) | |
615 | { | |
616 | KVMState *s = kvm_state; | |
617 | KVMSlot *mem, old; | |
618 | int err; | |
619 | MemoryRegion *mr = section->mr; | |
620 | bool log_dirty = memory_region_is_logging(mr); | |
621 | hwaddr start_addr = section->offset_within_address_space; | |
622 | ram_addr_t size = section->size; | |
623 | void *ram = NULL; | |
624 | unsigned delta; | |
625 | ||
626 | /* kvm works in page size chunks, but the function may be called | |
627 | with sub-page size and unaligned start address. */ | |
628 | delta = TARGET_PAGE_ALIGN(size) - size; | |
629 | if (delta > size) { | |
630 | return; | |
631 | } | |
632 | start_addr += delta; | |
633 | size -= delta; | |
634 | size &= TARGET_PAGE_MASK; | |
635 | if (!size || (start_addr & ~TARGET_PAGE_MASK)) { | |
636 | return; | |
637 | } | |
638 | ||
639 | if (!memory_region_is_ram(mr)) { | |
640 | return; | |
641 | } | |
642 | ||
643 | ram = memory_region_get_ram_ptr(mr) + section->offset_within_region + delta; | |
644 | ||
645 | while (1) { | |
646 | mem = kvm_lookup_overlapping_slot(s, start_addr, start_addr + size); | |
647 | if (!mem) { | |
648 | break; | |
649 | } | |
650 | ||
651 | if (add && start_addr >= mem->start_addr && | |
652 | (start_addr + size <= mem->start_addr + mem->memory_size) && | |
653 | (ram - start_addr == mem->ram - mem->start_addr)) { | |
654 | /* The new slot fits into the existing one and comes with | |
655 | * identical parameters - update flags and done. */ | |
656 | kvm_slot_dirty_pages_log_change(mem, log_dirty); | |
657 | return; | |
658 | } | |
659 | ||
660 | old = *mem; | |
661 | ||
662 | if (mem->flags & KVM_MEM_LOG_DIRTY_PAGES) { | |
663 | kvm_physical_sync_dirty_bitmap(section); | |
664 | } | |
665 | ||
666 | /* unregister the overlapping slot */ | |
667 | mem->memory_size = 0; | |
668 | err = kvm_set_user_memory_region(s, mem); | |
669 | if (err) { | |
670 | fprintf(stderr, "%s: error unregistering overlapping slot: %s\n", | |
671 | __func__, strerror(-err)); | |
672 | abort(); | |
673 | } | |
674 | ||
675 | /* Workaround for older KVM versions: we can't join slots, even not by | |
676 | * unregistering the previous ones and then registering the larger | |
677 | * slot. We have to maintain the existing fragmentation. Sigh. | |
678 | * | |
679 | * This workaround assumes that the new slot starts at the same | |
680 | * address as the first existing one. If not or if some overlapping | |
681 | * slot comes around later, we will fail (not seen in practice so far) | |
682 | * - and actually require a recent KVM version. */ | |
683 | if (s->broken_set_mem_region && | |
684 | old.start_addr == start_addr && old.memory_size < size && add) { | |
685 | mem = kvm_alloc_slot(s); | |
686 | mem->memory_size = old.memory_size; | |
687 | mem->start_addr = old.start_addr; | |
688 | mem->ram = old.ram; | |
689 | mem->flags = kvm_mem_flags(s, log_dirty); | |
690 | ||
691 | err = kvm_set_user_memory_region(s, mem); | |
692 | if (err) { | |
693 | fprintf(stderr, "%s: error updating slot: %s\n", __func__, | |
694 | strerror(-err)); | |
695 | abort(); | |
696 | } | |
697 | ||
698 | start_addr += old.memory_size; | |
699 | ram += old.memory_size; | |
700 | size -= old.memory_size; | |
701 | continue; | |
702 | } | |
703 | ||
704 | /* register prefix slot */ | |
705 | if (old.start_addr < start_addr) { | |
706 | mem = kvm_alloc_slot(s); | |
707 | mem->memory_size = start_addr - old.start_addr; | |
708 | mem->start_addr = old.start_addr; | |
709 | mem->ram = old.ram; | |
710 | mem->flags = kvm_mem_flags(s, log_dirty); | |
711 | ||
712 | err = kvm_set_user_memory_region(s, mem); | |
713 | if (err) { | |
714 | fprintf(stderr, "%s: error registering prefix slot: %s\n", | |
715 | __func__, strerror(-err)); | |
716 | #ifdef TARGET_PPC | |
717 | fprintf(stderr, "%s: This is probably because your kernel's " \ | |
718 | "PAGE_SIZE is too big. Please try to use 4k " \ | |
719 | "PAGE_SIZE!\n", __func__); | |
720 | #endif | |
721 | abort(); | |
722 | } | |
723 | } | |
724 | ||
725 | /* register suffix slot */ | |
726 | if (old.start_addr + old.memory_size > start_addr + size) { | |
727 | ram_addr_t size_delta; | |
728 | ||
729 | mem = kvm_alloc_slot(s); | |
730 | mem->start_addr = start_addr + size; | |
731 | size_delta = mem->start_addr - old.start_addr; | |
732 | mem->memory_size = old.memory_size - size_delta; | |
733 | mem->ram = old.ram + size_delta; | |
734 | mem->flags = kvm_mem_flags(s, log_dirty); | |
735 | ||
736 | err = kvm_set_user_memory_region(s, mem); | |
737 | if (err) { | |
738 | fprintf(stderr, "%s: error registering suffix slot: %s\n", | |
739 | __func__, strerror(-err)); | |
740 | abort(); | |
741 | } | |
742 | } | |
743 | } | |
744 | ||
745 | /* in case the KVM bug workaround already "consumed" the new slot */ | |
746 | if (!size) { | |
747 | return; | |
748 | } | |
749 | if (!add) { | |
750 | return; | |
751 | } | |
752 | mem = kvm_alloc_slot(s); | |
753 | mem->memory_size = size; | |
754 | mem->start_addr = start_addr; | |
755 | mem->ram = ram; | |
756 | mem->flags = kvm_mem_flags(s, log_dirty); | |
757 | ||
758 | err = kvm_set_user_memory_region(s, mem); | |
759 | if (err) { | |
760 | fprintf(stderr, "%s: error registering slot: %s\n", __func__, | |
761 | strerror(-err)); | |
762 | abort(); | |
763 | } | |
764 | } | |
765 | ||
766 | static void kvm_region_add(MemoryListener *listener, | |
767 | MemoryRegionSection *section) | |
768 | { | |
769 | kvm_set_phys_mem(section, true); | |
770 | } | |
771 | ||
772 | static void kvm_region_del(MemoryListener *listener, | |
773 | MemoryRegionSection *section) | |
774 | { | |
775 | kvm_set_phys_mem(section, false); | |
776 | } | |
777 | ||
778 | static void kvm_log_sync(MemoryListener *listener, | |
779 | MemoryRegionSection *section) | |
780 | { | |
781 | int r; | |
782 | ||
783 | r = kvm_physical_sync_dirty_bitmap(section); | |
784 | if (r < 0) { | |
785 | abort(); | |
786 | } | |
787 | } | |
788 | ||
789 | static void kvm_log_global_start(struct MemoryListener *listener) | |
790 | { | |
791 | int r; | |
792 | ||
793 | r = kvm_set_migration_log(1); | |
794 | assert(r >= 0); | |
795 | } | |
796 | ||
797 | static void kvm_log_global_stop(struct MemoryListener *listener) | |
798 | { | |
799 | int r; | |
800 | ||
801 | r = kvm_set_migration_log(0); | |
802 | assert(r >= 0); | |
803 | } | |
804 | ||
805 | static void kvm_mem_ioeventfd_add(MemoryListener *listener, | |
806 | MemoryRegionSection *section, | |
807 | bool match_data, uint64_t data, | |
808 | EventNotifier *e) | |
809 | { | |
810 | int fd = event_notifier_get_fd(e); | |
811 | int r; | |
812 | ||
813 | r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space, | |
814 | data, true, section->size, match_data); | |
815 | if (r < 0) { | |
816 | abort(); | |
817 | } | |
818 | } | |
819 | ||
820 | static void kvm_mem_ioeventfd_del(MemoryListener *listener, | |
821 | MemoryRegionSection *section, | |
822 | bool match_data, uint64_t data, | |
823 | EventNotifier *e) | |
824 | { | |
825 | int fd = event_notifier_get_fd(e); | |
826 | int r; | |
827 | ||
828 | r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space, | |
829 | data, false, section->size, match_data); | |
830 | if (r < 0) { | |
831 | abort(); | |
832 | } | |
833 | } | |
834 | ||
835 | static void kvm_io_ioeventfd_add(MemoryListener *listener, | |
836 | MemoryRegionSection *section, | |
837 | bool match_data, uint64_t data, | |
838 | EventNotifier *e) | |
839 | { | |
840 | int fd = event_notifier_get_fd(e); | |
841 | int r; | |
842 | ||
843 | r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space, | |
844 | data, true, section->size, match_data); | |
845 | if (r < 0) { | |
846 | abort(); | |
847 | } | |
848 | } | |
849 | ||
850 | static void kvm_io_ioeventfd_del(MemoryListener *listener, | |
851 | MemoryRegionSection *section, | |
852 | bool match_data, uint64_t data, | |
853 | EventNotifier *e) | |
854 | ||
855 | { | |
856 | int fd = event_notifier_get_fd(e); | |
857 | int r; | |
858 | ||
859 | r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space, | |
860 | data, false, section->size, match_data); | |
861 | if (r < 0) { | |
862 | abort(); | |
863 | } | |
864 | } | |
865 | ||
866 | static MemoryListener kvm_memory_listener = { | |
867 | .region_add = kvm_region_add, | |
868 | .region_del = kvm_region_del, | |
869 | .log_start = kvm_log_start, | |
870 | .log_stop = kvm_log_stop, | |
871 | .log_sync = kvm_log_sync, | |
872 | .log_global_start = kvm_log_global_start, | |
873 | .log_global_stop = kvm_log_global_stop, | |
874 | .eventfd_add = kvm_mem_ioeventfd_add, | |
875 | .eventfd_del = kvm_mem_ioeventfd_del, | |
876 | .coalesced_mmio_add = kvm_coalesce_mmio_region, | |
877 | .coalesced_mmio_del = kvm_uncoalesce_mmio_region, | |
878 | .priority = 10, | |
879 | }; | |
880 | ||
881 | static MemoryListener kvm_io_listener = { | |
882 | .eventfd_add = kvm_io_ioeventfd_add, | |
883 | .eventfd_del = kvm_io_ioeventfd_del, | |
884 | .priority = 10, | |
885 | }; | |
886 | ||
887 | static void kvm_handle_interrupt(CPUState *cpu, int mask) | |
888 | { | |
889 | cpu->interrupt_request |= mask; | |
890 | ||
891 | if (!qemu_cpu_is_self(cpu)) { | |
892 | qemu_cpu_kick(cpu); | |
893 | } | |
894 | } | |
895 | ||
896 | int kvm_set_irq(KVMState *s, int irq, int level) | |
897 | { | |
898 | struct kvm_irq_level event; | |
899 | int ret; | |
900 | ||
901 | assert(kvm_async_interrupts_enabled()); | |
902 | ||
903 | event.level = level; | |
904 | event.irq = irq; | |
905 | ret = kvm_vm_ioctl(s, s->irq_set_ioctl, &event); | |
906 | if (ret < 0) { | |
907 | perror("kvm_set_irq"); | |
908 | abort(); | |
909 | } | |
910 | ||
911 | return (s->irq_set_ioctl == KVM_IRQ_LINE) ? 1 : event.status; | |
912 | } | |
913 | ||
914 | #ifdef KVM_CAP_IRQ_ROUTING | |
915 | typedef struct KVMMSIRoute { | |
916 | struct kvm_irq_routing_entry kroute; | |
917 | QTAILQ_ENTRY(KVMMSIRoute) entry; | |
918 | } KVMMSIRoute; | |
919 | ||
920 | static void set_gsi(KVMState *s, unsigned int gsi) | |
921 | { | |
922 | s->used_gsi_bitmap[gsi / 32] |= 1U << (gsi % 32); | |
923 | } | |
924 | ||
925 | static void clear_gsi(KVMState *s, unsigned int gsi) | |
926 | { | |
927 | s->used_gsi_bitmap[gsi / 32] &= ~(1U << (gsi % 32)); | |
928 | } | |
929 | ||
930 | static void kvm_init_irq_routing(KVMState *s) | |
931 | { | |
932 | int gsi_count, i; | |
933 | ||
934 | gsi_count = kvm_check_extension(s, KVM_CAP_IRQ_ROUTING); | |
935 | if (gsi_count > 0) { | |
936 | unsigned int gsi_bits, i; | |
937 | ||
938 | /* Round up so we can search ints using ffs */ | |
939 | gsi_bits = ALIGN(gsi_count, 32); | |
940 | s->used_gsi_bitmap = g_malloc0(gsi_bits / 8); | |
941 | s->gsi_count = gsi_count; | |
942 | ||
943 | /* Mark any over-allocated bits as already in use */ | |
944 | for (i = gsi_count; i < gsi_bits; i++) { | |
945 | set_gsi(s, i); | |
946 | } | |
947 | } | |
948 | ||
949 | s->irq_routes = g_malloc0(sizeof(*s->irq_routes)); | |
950 | s->nr_allocated_irq_routes = 0; | |
951 | ||
952 | if (!s->direct_msi) { | |
953 | for (i = 0; i < KVM_MSI_HASHTAB_SIZE; i++) { | |
954 | QTAILQ_INIT(&s->msi_hashtab[i]); | |
955 | } | |
956 | } | |
957 | ||
958 | kvm_arch_init_irq_routing(s); | |
959 | } | |
960 | ||
961 | static void kvm_irqchip_commit_routes(KVMState *s) | |
962 | { | |
963 | int ret; | |
964 | ||
965 | s->irq_routes->flags = 0; | |
966 | ret = kvm_vm_ioctl(s, KVM_SET_GSI_ROUTING, s->irq_routes); | |
967 | assert(ret == 0); | |
968 | } | |
969 | ||
970 | static void kvm_add_routing_entry(KVMState *s, | |
971 | struct kvm_irq_routing_entry *entry) | |
972 | { | |
973 | struct kvm_irq_routing_entry *new; | |
974 | int n, size; | |
975 | ||
976 | if (s->irq_routes->nr == s->nr_allocated_irq_routes) { | |
977 | n = s->nr_allocated_irq_routes * 2; | |
978 | if (n < 64) { | |
979 | n = 64; | |
980 | } | |
981 | size = sizeof(struct kvm_irq_routing); | |
982 | size += n * sizeof(*new); | |
983 | s->irq_routes = g_realloc(s->irq_routes, size); | |
984 | s->nr_allocated_irq_routes = n; | |
985 | } | |
986 | n = s->irq_routes->nr++; | |
987 | new = &s->irq_routes->entries[n]; | |
988 | memset(new, 0, sizeof(*new)); | |
989 | new->gsi = entry->gsi; | |
990 | new->type = entry->type; | |
991 | new->flags = entry->flags; | |
992 | new->u = entry->u; | |
993 | ||
994 | set_gsi(s, entry->gsi); | |
995 | ||
996 | kvm_irqchip_commit_routes(s); | |
997 | } | |
998 | ||
999 | static int kvm_update_routing_entry(KVMState *s, | |
1000 | struct kvm_irq_routing_entry *new_entry) | |
1001 | { | |
1002 | struct kvm_irq_routing_entry *entry; | |
1003 | int n; | |
1004 | ||
1005 | for (n = 0; n < s->irq_routes->nr; n++) { | |
1006 | entry = &s->irq_routes->entries[n]; | |
1007 | if (entry->gsi != new_entry->gsi) { | |
1008 | continue; | |
1009 | } | |
1010 | ||
1011 | entry->type = new_entry->type; | |
1012 | entry->flags = new_entry->flags; | |
1013 | entry->u = new_entry->u; | |
1014 | ||
1015 | kvm_irqchip_commit_routes(s); | |
1016 | ||
1017 | return 0; | |
1018 | } | |
1019 | ||
1020 | return -ESRCH; | |
1021 | } | |
1022 | ||
1023 | void kvm_irqchip_add_irq_route(KVMState *s, int irq, int irqchip, int pin) | |
1024 | { | |
1025 | struct kvm_irq_routing_entry e; | |
1026 | ||
1027 | assert(pin < s->gsi_count); | |
1028 | ||
1029 | e.gsi = irq; | |
1030 | e.type = KVM_IRQ_ROUTING_IRQCHIP; | |
1031 | e.flags = 0; | |
1032 | e.u.irqchip.irqchip = irqchip; | |
1033 | e.u.irqchip.pin = pin; | |
1034 | kvm_add_routing_entry(s, &e); | |
1035 | } | |
1036 | ||
1037 | void kvm_irqchip_release_virq(KVMState *s, int virq) | |
1038 | { | |
1039 | struct kvm_irq_routing_entry *e; | |
1040 | int i; | |
1041 | ||
1042 | for (i = 0; i < s->irq_routes->nr; i++) { | |
1043 | e = &s->irq_routes->entries[i]; | |
1044 | if (e->gsi == virq) { | |
1045 | s->irq_routes->nr--; | |
1046 | *e = s->irq_routes->entries[s->irq_routes->nr]; | |
1047 | } | |
1048 | } | |
1049 | clear_gsi(s, virq); | |
1050 | } | |
1051 | ||
1052 | static unsigned int kvm_hash_msi(uint32_t data) | |
1053 | { | |
1054 | /* This is optimized for IA32 MSI layout. However, no other arch shall | |
1055 | * repeat the mistake of not providing a direct MSI injection API. */ | |
1056 | return data & 0xff; | |
1057 | } | |
1058 | ||
1059 | static void kvm_flush_dynamic_msi_routes(KVMState *s) | |
1060 | { | |
1061 | KVMMSIRoute *route, *next; | |
1062 | unsigned int hash; | |
1063 | ||
1064 | for (hash = 0; hash < KVM_MSI_HASHTAB_SIZE; hash++) { | |
1065 | QTAILQ_FOREACH_SAFE(route, &s->msi_hashtab[hash], entry, next) { | |
1066 | kvm_irqchip_release_virq(s, route->kroute.gsi); | |
1067 | QTAILQ_REMOVE(&s->msi_hashtab[hash], route, entry); | |
1068 | g_free(route); | |
1069 | } | |
1070 | } | |
1071 | } | |
1072 | ||
1073 | static int kvm_irqchip_get_virq(KVMState *s) | |
1074 | { | |
1075 | uint32_t *word = s->used_gsi_bitmap; | |
1076 | int max_words = ALIGN(s->gsi_count, 32) / 32; | |
1077 | int i, bit; | |
1078 | bool retry = true; | |
1079 | ||
1080 | again: | |
1081 | /* Return the lowest unused GSI in the bitmap */ | |
1082 | for (i = 0; i < max_words; i++) { | |
1083 | bit = ffs(~word[i]); | |
1084 | if (!bit) { | |
1085 | continue; | |
1086 | } | |
1087 | ||
1088 | return bit - 1 + i * 32; | |
1089 | } | |
1090 | if (!s->direct_msi && retry) { | |
1091 | retry = false; | |
1092 | kvm_flush_dynamic_msi_routes(s); | |
1093 | goto again; | |
1094 | } | |
1095 | return -ENOSPC; | |
1096 | ||
1097 | } | |
1098 | ||
1099 | static KVMMSIRoute *kvm_lookup_msi_route(KVMState *s, MSIMessage msg) | |
1100 | { | |
1101 | unsigned int hash = kvm_hash_msi(msg.data); | |
1102 | KVMMSIRoute *route; | |
1103 | ||
1104 | QTAILQ_FOREACH(route, &s->msi_hashtab[hash], entry) { | |
1105 | if (route->kroute.u.msi.address_lo == (uint32_t)msg.address && | |
1106 | route->kroute.u.msi.address_hi == (msg.address >> 32) && | |
1107 | route->kroute.u.msi.data == msg.data) { | |
1108 | return route; | |
1109 | } | |
1110 | } | |
1111 | return NULL; | |
1112 | } | |
1113 | ||
1114 | int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg) | |
1115 | { | |
1116 | struct kvm_msi msi; | |
1117 | KVMMSIRoute *route; | |
1118 | ||
1119 | if (s->direct_msi) { | |
1120 | msi.address_lo = (uint32_t)msg.address; | |
1121 | msi.address_hi = msg.address >> 32; | |
1122 | msi.data = msg.data; | |
1123 | msi.flags = 0; | |
1124 | memset(msi.pad, 0, sizeof(msi.pad)); | |
1125 | ||
1126 | return kvm_vm_ioctl(s, KVM_SIGNAL_MSI, &msi); | |
1127 | } | |
1128 | ||
1129 | route = kvm_lookup_msi_route(s, msg); | |
1130 | if (!route) { | |
1131 | int virq; | |
1132 | ||
1133 | virq = kvm_irqchip_get_virq(s); | |
1134 | if (virq < 0) { | |
1135 | return virq; | |
1136 | } | |
1137 | ||
1138 | route = g_malloc(sizeof(KVMMSIRoute)); | |
1139 | route->kroute.gsi = virq; | |
1140 | route->kroute.type = KVM_IRQ_ROUTING_MSI; | |
1141 | route->kroute.flags = 0; | |
1142 | route->kroute.u.msi.address_lo = (uint32_t)msg.address; | |
1143 | route->kroute.u.msi.address_hi = msg.address >> 32; | |
1144 | route->kroute.u.msi.data = msg.data; | |
1145 | ||
1146 | kvm_add_routing_entry(s, &route->kroute); | |
1147 | ||
1148 | QTAILQ_INSERT_TAIL(&s->msi_hashtab[kvm_hash_msi(msg.data)], route, | |
1149 | entry); | |
1150 | } | |
1151 | ||
1152 | assert(route->kroute.type == KVM_IRQ_ROUTING_MSI); | |
1153 | ||
1154 | return kvm_set_irq(s, route->kroute.gsi, 1); | |
1155 | } | |
1156 | ||
1157 | int kvm_irqchip_add_msi_route(KVMState *s, MSIMessage msg) | |
1158 | { | |
1159 | struct kvm_irq_routing_entry kroute; | |
1160 | int virq; | |
1161 | ||
1162 | if (!kvm_gsi_routing_enabled()) { | |
1163 | return -ENOSYS; | |
1164 | } | |
1165 | ||
1166 | virq = kvm_irqchip_get_virq(s); | |
1167 | if (virq < 0) { | |
1168 | return virq; | |
1169 | } | |
1170 | ||
1171 | kroute.gsi = virq; | |
1172 | kroute.type = KVM_IRQ_ROUTING_MSI; | |
1173 | kroute.flags = 0; | |
1174 | kroute.u.msi.address_lo = (uint32_t)msg.address; | |
1175 | kroute.u.msi.address_hi = msg.address >> 32; | |
1176 | kroute.u.msi.data = msg.data; | |
1177 | ||
1178 | kvm_add_routing_entry(s, &kroute); | |
1179 | ||
1180 | return virq; | |
1181 | } | |
1182 | ||
1183 | int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg) | |
1184 | { | |
1185 | struct kvm_irq_routing_entry kroute; | |
1186 | ||
1187 | if (!kvm_irqchip_in_kernel()) { | |
1188 | return -ENOSYS; | |
1189 | } | |
1190 | ||
1191 | kroute.gsi = virq; | |
1192 | kroute.type = KVM_IRQ_ROUTING_MSI; | |
1193 | kroute.flags = 0; | |
1194 | kroute.u.msi.address_lo = (uint32_t)msg.address; | |
1195 | kroute.u.msi.address_hi = msg.address >> 32; | |
1196 | kroute.u.msi.data = msg.data; | |
1197 | ||
1198 | return kvm_update_routing_entry(s, &kroute); | |
1199 | } | |
1200 | ||
1201 | static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int virq, bool assign) | |
1202 | { | |
1203 | struct kvm_irqfd irqfd = { | |
1204 | .fd = fd, | |
1205 | .gsi = virq, | |
1206 | .flags = assign ? 0 : KVM_IRQFD_FLAG_DEASSIGN, | |
1207 | }; | |
1208 | ||
1209 | if (!kvm_irqfds_enabled()) { | |
1210 | return -ENOSYS; | |
1211 | } | |
1212 | ||
1213 | return kvm_vm_ioctl(s, KVM_IRQFD, &irqfd); | |
1214 | } | |
1215 | ||
1216 | #else /* !KVM_CAP_IRQ_ROUTING */ | |
1217 | ||
1218 | static void kvm_init_irq_routing(KVMState *s) | |
1219 | { | |
1220 | } | |
1221 | ||
1222 | void kvm_irqchip_release_virq(KVMState *s, int virq) | |
1223 | { | |
1224 | } | |
1225 | ||
1226 | int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg) | |
1227 | { | |
1228 | abort(); | |
1229 | } | |
1230 | ||
1231 | int kvm_irqchip_add_msi_route(KVMState *s, MSIMessage msg) | |
1232 | { | |
1233 | return -ENOSYS; | |
1234 | } | |
1235 | ||
1236 | static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int virq, bool assign) | |
1237 | { | |
1238 | abort(); | |
1239 | } | |
1240 | ||
1241 | int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg) | |
1242 | { | |
1243 | return -ENOSYS; | |
1244 | } | |
1245 | #endif /* !KVM_CAP_IRQ_ROUTING */ | |
1246 | ||
1247 | int kvm_irqchip_add_irqfd_notifier(KVMState *s, EventNotifier *n, int virq) | |
1248 | { | |
1249 | return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n), virq, true); | |
1250 | } | |
1251 | ||
1252 | int kvm_irqchip_remove_irqfd_notifier(KVMState *s, EventNotifier *n, int virq) | |
1253 | { | |
1254 | return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n), virq, false); | |
1255 | } | |
1256 | ||
1257 | static int kvm_irqchip_create(KVMState *s) | |
1258 | { | |
1259 | QemuOptsList *list = qemu_find_opts("machine"); | |
1260 | int ret; | |
1261 | ||
1262 | if (QTAILQ_EMPTY(&list->head) || | |
1263 | !qemu_opt_get_bool(QTAILQ_FIRST(&list->head), | |
1264 | "kernel_irqchip", true) || | |
1265 | !kvm_check_extension(s, KVM_CAP_IRQCHIP)) { | |
1266 | return 0; | |
1267 | } | |
1268 | ||
1269 | ret = kvm_vm_ioctl(s, KVM_CREATE_IRQCHIP); | |
1270 | if (ret < 0) { | |
1271 | fprintf(stderr, "Create kernel irqchip failed\n"); | |
1272 | return ret; | |
1273 | } | |
1274 | ||
1275 | kvm_kernel_irqchip = true; | |
1276 | /* If we have an in-kernel IRQ chip then we must have asynchronous | |
1277 | * interrupt delivery (though the reverse is not necessarily true) | |
1278 | */ | |
1279 | kvm_async_interrupts_allowed = true; | |
1280 | ||
1281 | kvm_init_irq_routing(s); | |
1282 | ||
1283 | return 0; | |
1284 | } | |
1285 | ||
1286 | static int kvm_max_vcpus(KVMState *s) | |
1287 | { | |
1288 | int ret; | |
1289 | ||
1290 | /* Find number of supported CPUs using the recommended | |
1291 | * procedure from the kernel API documentation to cope with | |
1292 | * older kernels that may be missing capabilities. | |
1293 | */ | |
1294 | ret = kvm_check_extension(s, KVM_CAP_MAX_VCPUS); | |
1295 | if (ret) { | |
1296 | return ret; | |
1297 | } | |
1298 | ret = kvm_check_extension(s, KVM_CAP_NR_VCPUS); | |
1299 | if (ret) { | |
1300 | return ret; | |
1301 | } | |
1302 | ||
1303 | return 4; | |
1304 | } | |
1305 | ||
1306 | int kvm_init(void) | |
1307 | { | |
1308 | static const char upgrade_note[] = | |
1309 | "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n" | |
1310 | "(see http://sourceforge.net/projects/kvm).\n"; | |
1311 | KVMState *s; | |
1312 | const KVMCapabilityInfo *missing_cap; | |
1313 | int ret; | |
1314 | int i; | |
1315 | int max_vcpus; | |
1316 | ||
1317 | s = g_malloc0(sizeof(KVMState)); | |
1318 | ||
1319 | /* | |
1320 | * On systems where the kernel can support different base page | |
1321 | * sizes, host page size may be different from TARGET_PAGE_SIZE, | |
1322 | * even with KVM. TARGET_PAGE_SIZE is assumed to be the minimum | |
1323 | * page size for the system though. | |
1324 | */ | |
1325 | assert(TARGET_PAGE_SIZE <= getpagesize()); | |
1326 | ||
1327 | #ifdef KVM_CAP_SET_GUEST_DEBUG | |
1328 | QTAILQ_INIT(&s->kvm_sw_breakpoints); | |
1329 | #endif | |
1330 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
1331 | s->slots[i].slot = i; | |
1332 | } | |
1333 | s->vmfd = -1; | |
1334 | s->fd = qemu_open("/dev/kvm", O_RDWR); | |
1335 | if (s->fd == -1) { | |
1336 | fprintf(stderr, "Could not access KVM kernel module: %m\n"); | |
1337 | ret = -errno; | |
1338 | goto err; | |
1339 | } | |
1340 | ||
1341 | ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0); | |
1342 | if (ret < KVM_API_VERSION) { | |
1343 | if (ret > 0) { | |
1344 | ret = -EINVAL; | |
1345 | } | |
1346 | fprintf(stderr, "kvm version too old\n"); | |
1347 | goto err; | |
1348 | } | |
1349 | ||
1350 | if (ret > KVM_API_VERSION) { | |
1351 | ret = -EINVAL; | |
1352 | fprintf(stderr, "kvm version not supported\n"); | |
1353 | goto err; | |
1354 | } | |
1355 | ||
1356 | max_vcpus = kvm_max_vcpus(s); | |
1357 | if (smp_cpus > max_vcpus) { | |
1358 | ret = -EINVAL; | |
1359 | fprintf(stderr, "Number of SMP cpus requested (%d) exceeds max cpus " | |
1360 | "supported by KVM (%d)\n", smp_cpus, max_vcpus); | |
1361 | goto err; | |
1362 | } | |
1363 | ||
1364 | s->vmfd = kvm_ioctl(s, KVM_CREATE_VM, 0); | |
1365 | if (s->vmfd < 0) { | |
1366 | #ifdef TARGET_S390X | |
1367 | fprintf(stderr, "Please add the 'switch_amode' kernel parameter to " | |
1368 | "your host kernel command line\n"); | |
1369 | #endif | |
1370 | ret = s->vmfd; | |
1371 | goto err; | |
1372 | } | |
1373 | ||
1374 | missing_cap = kvm_check_extension_list(s, kvm_required_capabilites); | |
1375 | if (!missing_cap) { | |
1376 | missing_cap = | |
1377 | kvm_check_extension_list(s, kvm_arch_required_capabilities); | |
1378 | } | |
1379 | if (missing_cap) { | |
1380 | ret = -EINVAL; | |
1381 | fprintf(stderr, "kvm does not support %s\n%s", | |
1382 | missing_cap->name, upgrade_note); | |
1383 | goto err; | |
1384 | } | |
1385 | ||
1386 | s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO); | |
1387 | ||
1388 | s->broken_set_mem_region = 1; | |
1389 | ret = kvm_check_extension(s, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS); | |
1390 | if (ret > 0) { | |
1391 | s->broken_set_mem_region = 0; | |
1392 | } | |
1393 | ||
1394 | #ifdef KVM_CAP_VCPU_EVENTS | |
1395 | s->vcpu_events = kvm_check_extension(s, KVM_CAP_VCPU_EVENTS); | |
1396 | #endif | |
1397 | ||
1398 | s->robust_singlestep = | |
1399 | kvm_check_extension(s, KVM_CAP_X86_ROBUST_SINGLESTEP); | |
1400 | ||
1401 | #ifdef KVM_CAP_DEBUGREGS | |
1402 | s->debugregs = kvm_check_extension(s, KVM_CAP_DEBUGREGS); | |
1403 | #endif | |
1404 | ||
1405 | #ifdef KVM_CAP_XSAVE | |
1406 | s->xsave = kvm_check_extension(s, KVM_CAP_XSAVE); | |
1407 | #endif | |
1408 | ||
1409 | #ifdef KVM_CAP_XCRS | |
1410 | s->xcrs = kvm_check_extension(s, KVM_CAP_XCRS); | |
1411 | #endif | |
1412 | ||
1413 | #ifdef KVM_CAP_PIT_STATE2 | |
1414 | s->pit_state2 = kvm_check_extension(s, KVM_CAP_PIT_STATE2); | |
1415 | #endif | |
1416 | ||
1417 | #ifdef KVM_CAP_IRQ_ROUTING | |
1418 | s->direct_msi = (kvm_check_extension(s, KVM_CAP_SIGNAL_MSI) > 0); | |
1419 | #endif | |
1420 | ||
1421 | s->intx_set_mask = kvm_check_extension(s, KVM_CAP_PCI_2_3); | |
1422 | ||
1423 | s->irq_set_ioctl = KVM_IRQ_LINE; | |
1424 | if (kvm_check_extension(s, KVM_CAP_IRQ_INJECT_STATUS)) { | |
1425 | s->irq_set_ioctl = KVM_IRQ_LINE_STATUS; | |
1426 | } | |
1427 | ||
1428 | ret = kvm_arch_init(s); | |
1429 | if (ret < 0) { | |
1430 | goto err; | |
1431 | } | |
1432 | ||
1433 | ret = kvm_irqchip_create(s); | |
1434 | if (ret < 0) { | |
1435 | goto err; | |
1436 | } | |
1437 | ||
1438 | kvm_state = s; | |
1439 | memory_listener_register(&kvm_memory_listener, &address_space_memory); | |
1440 | memory_listener_register(&kvm_io_listener, &address_space_io); | |
1441 | ||
1442 | s->many_ioeventfds = kvm_check_many_ioeventfds(); | |
1443 | ||
1444 | cpu_interrupt_handler = kvm_handle_interrupt; | |
1445 | ||
1446 | return 0; | |
1447 | ||
1448 | err: | |
1449 | if (s->vmfd >= 0) { | |
1450 | close(s->vmfd); | |
1451 | } | |
1452 | if (s->fd != -1) { | |
1453 | close(s->fd); | |
1454 | } | |
1455 | g_free(s); | |
1456 | ||
1457 | return ret; | |
1458 | } | |
1459 | ||
1460 | static void kvm_handle_io(uint16_t port, void *data, int direction, int size, | |
1461 | uint32_t count) | |
1462 | { | |
1463 | int i; | |
1464 | uint8_t *ptr = data; | |
1465 | ||
1466 | for (i = 0; i < count; i++) { | |
1467 | if (direction == KVM_EXIT_IO_IN) { | |
1468 | switch (size) { | |
1469 | case 1: | |
1470 | stb_p(ptr, cpu_inb(port)); | |
1471 | break; | |
1472 | case 2: | |
1473 | stw_p(ptr, cpu_inw(port)); | |
1474 | break; | |
1475 | case 4: | |
1476 | stl_p(ptr, cpu_inl(port)); | |
1477 | break; | |
1478 | } | |
1479 | } else { | |
1480 | switch (size) { | |
1481 | case 1: | |
1482 | cpu_outb(port, ldub_p(ptr)); | |
1483 | break; | |
1484 | case 2: | |
1485 | cpu_outw(port, lduw_p(ptr)); | |
1486 | break; | |
1487 | case 4: | |
1488 | cpu_outl(port, ldl_p(ptr)); | |
1489 | break; | |
1490 | } | |
1491 | } | |
1492 | ||
1493 | ptr += size; | |
1494 | } | |
1495 | } | |
1496 | ||
1497 | static int kvm_handle_internal_error(CPUArchState *env, struct kvm_run *run) | |
1498 | { | |
1499 | CPUState *cpu = ENV_GET_CPU(env); | |
1500 | ||
1501 | fprintf(stderr, "KVM internal error."); | |
1502 | if (kvm_check_extension(kvm_state, KVM_CAP_INTERNAL_ERROR_DATA)) { | |
1503 | int i; | |
1504 | ||
1505 | fprintf(stderr, " Suberror: %d\n", run->internal.suberror); | |
1506 | for (i = 0; i < run->internal.ndata; ++i) { | |
1507 | fprintf(stderr, "extra data[%d]: %"PRIx64"\n", | |
1508 | i, (uint64_t)run->internal.data[i]); | |
1509 | } | |
1510 | } else { | |
1511 | fprintf(stderr, "\n"); | |
1512 | } | |
1513 | if (run->internal.suberror == KVM_INTERNAL_ERROR_EMULATION) { | |
1514 | fprintf(stderr, "emulation failure\n"); | |
1515 | if (!kvm_arch_stop_on_emulation_error(cpu)) { | |
1516 | cpu_dump_state(env, stderr, fprintf, CPU_DUMP_CODE); | |
1517 | return EXCP_INTERRUPT; | |
1518 | } | |
1519 | } | |
1520 | /* FIXME: Should trigger a qmp message to let management know | |
1521 | * something went wrong. | |
1522 | */ | |
1523 | return -1; | |
1524 | } | |
1525 | ||
1526 | void kvm_flush_coalesced_mmio_buffer(void) | |
1527 | { | |
1528 | KVMState *s = kvm_state; | |
1529 | ||
1530 | if (s->coalesced_flush_in_progress) { | |
1531 | return; | |
1532 | } | |
1533 | ||
1534 | s->coalesced_flush_in_progress = true; | |
1535 | ||
1536 | if (s->coalesced_mmio_ring) { | |
1537 | struct kvm_coalesced_mmio_ring *ring = s->coalesced_mmio_ring; | |
1538 | while (ring->first != ring->last) { | |
1539 | struct kvm_coalesced_mmio *ent; | |
1540 | ||
1541 | ent = &ring->coalesced_mmio[ring->first]; | |
1542 | ||
1543 | cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len); | |
1544 | smp_wmb(); | |
1545 | ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX; | |
1546 | } | |
1547 | } | |
1548 | ||
1549 | s->coalesced_flush_in_progress = false; | |
1550 | } | |
1551 | ||
1552 | static void do_kvm_cpu_synchronize_state(void *arg) | |
1553 | { | |
1554 | CPUState *cpu = arg; | |
1555 | ||
1556 | if (!cpu->kvm_vcpu_dirty) { | |
1557 | kvm_arch_get_registers(cpu); | |
1558 | cpu->kvm_vcpu_dirty = true; | |
1559 | } | |
1560 | } | |
1561 | ||
1562 | void kvm_cpu_synchronize_state(CPUArchState *env) | |
1563 | { | |
1564 | CPUState *cpu = ENV_GET_CPU(env); | |
1565 | ||
1566 | if (!cpu->kvm_vcpu_dirty) { | |
1567 | run_on_cpu(cpu, do_kvm_cpu_synchronize_state, cpu); | |
1568 | } | |
1569 | } | |
1570 | ||
1571 | void kvm_cpu_synchronize_post_reset(CPUState *cpu) | |
1572 | { | |
1573 | kvm_arch_put_registers(cpu, KVM_PUT_RESET_STATE); | |
1574 | cpu->kvm_vcpu_dirty = false; | |
1575 | } | |
1576 | ||
1577 | void kvm_cpu_synchronize_post_init(CPUState *cpu) | |
1578 | { | |
1579 | kvm_arch_put_registers(cpu, KVM_PUT_FULL_STATE); | |
1580 | cpu->kvm_vcpu_dirty = false; | |
1581 | } | |
1582 | ||
1583 | int kvm_cpu_exec(CPUArchState *env) | |
1584 | { | |
1585 | CPUState *cpu = ENV_GET_CPU(env); | |
1586 | struct kvm_run *run = cpu->kvm_run; | |
1587 | int ret, run_ret; | |
1588 | ||
1589 | DPRINTF("kvm_cpu_exec()\n"); | |
1590 | ||
1591 | if (kvm_arch_process_async_events(cpu)) { | |
1592 | cpu->exit_request = 0; | |
1593 | return EXCP_HLT; | |
1594 | } | |
1595 | ||
1596 | do { | |
1597 | if (cpu->kvm_vcpu_dirty) { | |
1598 | kvm_arch_put_registers(cpu, KVM_PUT_RUNTIME_STATE); | |
1599 | cpu->kvm_vcpu_dirty = false; | |
1600 | } | |
1601 | ||
1602 | kvm_arch_pre_run(cpu, run); | |
1603 | if (cpu->exit_request) { | |
1604 | DPRINTF("interrupt exit requested\n"); | |
1605 | /* | |
1606 | * KVM requires us to reenter the kernel after IO exits to complete | |
1607 | * instruction emulation. This self-signal will ensure that we | |
1608 | * leave ASAP again. | |
1609 | */ | |
1610 | qemu_cpu_kick_self(); | |
1611 | } | |
1612 | qemu_mutex_unlock_iothread(); | |
1613 | ||
1614 | run_ret = kvm_vcpu_ioctl(cpu, KVM_RUN, 0); | |
1615 | ||
1616 | qemu_mutex_lock_iothread(); | |
1617 | kvm_arch_post_run(cpu, run); | |
1618 | ||
1619 | if (run_ret < 0) { | |
1620 | if (run_ret == -EINTR || run_ret == -EAGAIN) { | |
1621 | DPRINTF("io window exit\n"); | |
1622 | ret = EXCP_INTERRUPT; | |
1623 | break; | |
1624 | } | |
1625 | fprintf(stderr, "error: kvm run failed %s\n", | |
1626 | strerror(-run_ret)); | |
1627 | abort(); | |
1628 | } | |
1629 | ||
1630 | trace_kvm_run_exit(cpu->cpu_index, run->exit_reason); | |
1631 | switch (run->exit_reason) { | |
1632 | case KVM_EXIT_IO: | |
1633 | DPRINTF("handle_io\n"); | |
1634 | kvm_handle_io(run->io.port, | |
1635 | (uint8_t *)run + run->io.data_offset, | |
1636 | run->io.direction, | |
1637 | run->io.size, | |
1638 | run->io.count); | |
1639 | ret = 0; | |
1640 | break; | |
1641 | case KVM_EXIT_MMIO: | |
1642 | DPRINTF("handle_mmio\n"); | |
1643 | cpu_physical_memory_rw(run->mmio.phys_addr, | |
1644 | run->mmio.data, | |
1645 | run->mmio.len, | |
1646 | run->mmio.is_write); | |
1647 | ret = 0; | |
1648 | break; | |
1649 | case KVM_EXIT_IRQ_WINDOW_OPEN: | |
1650 | DPRINTF("irq_window_open\n"); | |
1651 | ret = EXCP_INTERRUPT; | |
1652 | break; | |
1653 | case KVM_EXIT_SHUTDOWN: | |
1654 | DPRINTF("shutdown\n"); | |
1655 | qemu_system_reset_request(); | |
1656 | ret = EXCP_INTERRUPT; | |
1657 | break; | |
1658 | case KVM_EXIT_UNKNOWN: | |
1659 | fprintf(stderr, "KVM: unknown exit, hardware reason %" PRIx64 "\n", | |
1660 | (uint64_t)run->hw.hardware_exit_reason); | |
1661 | ret = -1; | |
1662 | break; | |
1663 | case KVM_EXIT_INTERNAL_ERROR: | |
1664 | ret = kvm_handle_internal_error(env, run); | |
1665 | break; | |
1666 | default: | |
1667 | DPRINTF("kvm_arch_handle_exit\n"); | |
1668 | ret = kvm_arch_handle_exit(cpu, run); | |
1669 | break; | |
1670 | } | |
1671 | } while (ret == 0); | |
1672 | ||
1673 | if (ret < 0) { | |
1674 | cpu_dump_state(env, stderr, fprintf, CPU_DUMP_CODE); | |
1675 | vm_stop(RUN_STATE_INTERNAL_ERROR); | |
1676 | } | |
1677 | ||
1678 | cpu->exit_request = 0; | |
1679 | return ret; | |
1680 | } | |
1681 | ||
1682 | int kvm_ioctl(KVMState *s, int type, ...) | |
1683 | { | |
1684 | int ret; | |
1685 | void *arg; | |
1686 | va_list ap; | |
1687 | ||
1688 | va_start(ap, type); | |
1689 | arg = va_arg(ap, void *); | |
1690 | va_end(ap); | |
1691 | ||
1692 | trace_kvm_ioctl(type, arg); | |
1693 | ret = ioctl(s->fd, type, arg); | |
1694 | if (ret == -1) { | |
1695 | ret = -errno; | |
1696 | } | |
1697 | return ret; | |
1698 | } | |
1699 | ||
1700 | int kvm_vm_ioctl(KVMState *s, int type, ...) | |
1701 | { | |
1702 | int ret; | |
1703 | void *arg; | |
1704 | va_list ap; | |
1705 | ||
1706 | va_start(ap, type); | |
1707 | arg = va_arg(ap, void *); | |
1708 | va_end(ap); | |
1709 | ||
1710 | trace_kvm_vm_ioctl(type, arg); | |
1711 | ret = ioctl(s->vmfd, type, arg); | |
1712 | if (ret == -1) { | |
1713 | ret = -errno; | |
1714 | } | |
1715 | return ret; | |
1716 | } | |
1717 | ||
1718 | int kvm_vcpu_ioctl(CPUState *cpu, int type, ...) | |
1719 | { | |
1720 | int ret; | |
1721 | void *arg; | |
1722 | va_list ap; | |
1723 | ||
1724 | va_start(ap, type); | |
1725 | arg = va_arg(ap, void *); | |
1726 | va_end(ap); | |
1727 | ||
1728 | trace_kvm_vcpu_ioctl(cpu->cpu_index, type, arg); | |
1729 | ret = ioctl(cpu->kvm_fd, type, arg); | |
1730 | if (ret == -1) { | |
1731 | ret = -errno; | |
1732 | } | |
1733 | return ret; | |
1734 | } | |
1735 | ||
1736 | int kvm_has_sync_mmu(void) | |
1737 | { | |
1738 | return kvm_check_extension(kvm_state, KVM_CAP_SYNC_MMU); | |
1739 | } | |
1740 | ||
1741 | int kvm_has_vcpu_events(void) | |
1742 | { | |
1743 | return kvm_state->vcpu_events; | |
1744 | } | |
1745 | ||
1746 | int kvm_has_robust_singlestep(void) | |
1747 | { | |
1748 | return kvm_state->robust_singlestep; | |
1749 | } | |
1750 | ||
1751 | int kvm_has_debugregs(void) | |
1752 | { | |
1753 | return kvm_state->debugregs; | |
1754 | } | |
1755 | ||
1756 | int kvm_has_xsave(void) | |
1757 | { | |
1758 | return kvm_state->xsave; | |
1759 | } | |
1760 | ||
1761 | int kvm_has_xcrs(void) | |
1762 | { | |
1763 | return kvm_state->xcrs; | |
1764 | } | |
1765 | ||
1766 | int kvm_has_pit_state2(void) | |
1767 | { | |
1768 | return kvm_state->pit_state2; | |
1769 | } | |
1770 | ||
1771 | int kvm_has_many_ioeventfds(void) | |
1772 | { | |
1773 | if (!kvm_enabled()) { | |
1774 | return 0; | |
1775 | } | |
1776 | return kvm_state->many_ioeventfds; | |
1777 | } | |
1778 | ||
1779 | int kvm_has_gsi_routing(void) | |
1780 | { | |
1781 | #ifdef KVM_CAP_IRQ_ROUTING | |
1782 | return kvm_check_extension(kvm_state, KVM_CAP_IRQ_ROUTING); | |
1783 | #else | |
1784 | return false; | |
1785 | #endif | |
1786 | } | |
1787 | ||
1788 | int kvm_has_intx_set_mask(void) | |
1789 | { | |
1790 | return kvm_state->intx_set_mask; | |
1791 | } | |
1792 | ||
1793 | void *kvm_vmalloc(ram_addr_t size) | |
1794 | { | |
1795 | #ifdef TARGET_S390X | |
1796 | void *mem; | |
1797 | ||
1798 | mem = kvm_arch_vmalloc(size); | |
1799 | if (mem) { | |
1800 | return mem; | |
1801 | } | |
1802 | #endif | |
1803 | return qemu_vmalloc(size); | |
1804 | } | |
1805 | ||
1806 | void kvm_setup_guest_memory(void *start, size_t size) | |
1807 | { | |
1808 | #ifdef CONFIG_VALGRIND_H | |
1809 | VALGRIND_MAKE_MEM_DEFINED(start, size); | |
1810 | #endif | |
1811 | if (!kvm_has_sync_mmu()) { | |
1812 | int ret = qemu_madvise(start, size, QEMU_MADV_DONTFORK); | |
1813 | ||
1814 | if (ret) { | |
1815 | perror("qemu_madvise"); | |
1816 | fprintf(stderr, | |
1817 | "Need MADV_DONTFORK in absence of synchronous KVM MMU\n"); | |
1818 | exit(1); | |
1819 | } | |
1820 | } | |
1821 | } | |
1822 | ||
1823 | #ifdef KVM_CAP_SET_GUEST_DEBUG | |
1824 | struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *cpu, | |
1825 | target_ulong pc) | |
1826 | { | |
1827 | struct kvm_sw_breakpoint *bp; | |
1828 | ||
1829 | QTAILQ_FOREACH(bp, &cpu->kvm_state->kvm_sw_breakpoints, entry) { | |
1830 | if (bp->pc == pc) { | |
1831 | return bp; | |
1832 | } | |
1833 | } | |
1834 | return NULL; | |
1835 | } | |
1836 | ||
1837 | int kvm_sw_breakpoints_active(CPUState *cpu) | |
1838 | { | |
1839 | return !QTAILQ_EMPTY(&cpu->kvm_state->kvm_sw_breakpoints); | |
1840 | } | |
1841 | ||
1842 | struct kvm_set_guest_debug_data { | |
1843 | struct kvm_guest_debug dbg; | |
1844 | CPUState *cpu; | |
1845 | int err; | |
1846 | }; | |
1847 | ||
1848 | static void kvm_invoke_set_guest_debug(void *data) | |
1849 | { | |
1850 | struct kvm_set_guest_debug_data *dbg_data = data; | |
1851 | ||
1852 | dbg_data->err = kvm_vcpu_ioctl(dbg_data->cpu, KVM_SET_GUEST_DEBUG, | |
1853 | &dbg_data->dbg); | |
1854 | } | |
1855 | ||
1856 | int kvm_update_guest_debug(CPUArchState *env, unsigned long reinject_trap) | |
1857 | { | |
1858 | CPUState *cpu = ENV_GET_CPU(env); | |
1859 | struct kvm_set_guest_debug_data data; | |
1860 | ||
1861 | data.dbg.control = reinject_trap; | |
1862 | ||
1863 | if (env->singlestep_enabled) { | |
1864 | data.dbg.control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP; | |
1865 | } | |
1866 | kvm_arch_update_guest_debug(cpu, &data.dbg); | |
1867 | data.cpu = cpu; | |
1868 | ||
1869 | run_on_cpu(cpu, kvm_invoke_set_guest_debug, &data); | |
1870 | return data.err; | |
1871 | } | |
1872 | ||
1873 | int kvm_insert_breakpoint(CPUArchState *current_env, target_ulong addr, | |
1874 | target_ulong len, int type) | |
1875 | { | |
1876 | CPUState *current_cpu = ENV_GET_CPU(current_env); | |
1877 | struct kvm_sw_breakpoint *bp; | |
1878 | CPUArchState *env; | |
1879 | int err; | |
1880 | ||
1881 | if (type == GDB_BREAKPOINT_SW) { | |
1882 | bp = kvm_find_sw_breakpoint(current_cpu, addr); | |
1883 | if (bp) { | |
1884 | bp->use_count++; | |
1885 | return 0; | |
1886 | } | |
1887 | ||
1888 | bp = g_malloc(sizeof(struct kvm_sw_breakpoint)); | |
1889 | if (!bp) { | |
1890 | return -ENOMEM; | |
1891 | } | |
1892 | ||
1893 | bp->pc = addr; | |
1894 | bp->use_count = 1; | |
1895 | err = kvm_arch_insert_sw_breakpoint(current_cpu, bp); | |
1896 | if (err) { | |
1897 | g_free(bp); | |
1898 | return err; | |
1899 | } | |
1900 | ||
1901 | QTAILQ_INSERT_HEAD(¤t_cpu->kvm_state->kvm_sw_breakpoints, | |
1902 | bp, entry); | |
1903 | } else { | |
1904 | err = kvm_arch_insert_hw_breakpoint(addr, len, type); | |
1905 | if (err) { | |
1906 | return err; | |
1907 | } | |
1908 | } | |
1909 | ||
1910 | for (env = first_cpu; env != NULL; env = env->next_cpu) { | |
1911 | err = kvm_update_guest_debug(env, 0); | |
1912 | if (err) { | |
1913 | return err; | |
1914 | } | |
1915 | } | |
1916 | return 0; | |
1917 | } | |
1918 | ||
1919 | int kvm_remove_breakpoint(CPUArchState *current_env, target_ulong addr, | |
1920 | target_ulong len, int type) | |
1921 | { | |
1922 | CPUState *current_cpu = ENV_GET_CPU(current_env); | |
1923 | struct kvm_sw_breakpoint *bp; | |
1924 | CPUArchState *env; | |
1925 | int err; | |
1926 | ||
1927 | if (type == GDB_BREAKPOINT_SW) { | |
1928 | bp = kvm_find_sw_breakpoint(current_cpu, addr); | |
1929 | if (!bp) { | |
1930 | return -ENOENT; | |
1931 | } | |
1932 | ||
1933 | if (bp->use_count > 1) { | |
1934 | bp->use_count--; | |
1935 | return 0; | |
1936 | } | |
1937 | ||
1938 | err = kvm_arch_remove_sw_breakpoint(current_cpu, bp); | |
1939 | if (err) { | |
1940 | return err; | |
1941 | } | |
1942 | ||
1943 | QTAILQ_REMOVE(¤t_cpu->kvm_state->kvm_sw_breakpoints, bp, entry); | |
1944 | g_free(bp); | |
1945 | } else { | |
1946 | err = kvm_arch_remove_hw_breakpoint(addr, len, type); | |
1947 | if (err) { | |
1948 | return err; | |
1949 | } | |
1950 | } | |
1951 | ||
1952 | for (env = first_cpu; env != NULL; env = env->next_cpu) { | |
1953 | err = kvm_update_guest_debug(env, 0); | |
1954 | if (err) { | |
1955 | return err; | |
1956 | } | |
1957 | } | |
1958 | return 0; | |
1959 | } | |
1960 | ||
1961 | void kvm_remove_all_breakpoints(CPUArchState *current_env) | |
1962 | { | |
1963 | CPUState *current_cpu = ENV_GET_CPU(current_env); | |
1964 | struct kvm_sw_breakpoint *bp, *next; | |
1965 | KVMState *s = current_cpu->kvm_state; | |
1966 | CPUArchState *env; | |
1967 | CPUState *cpu; | |
1968 | ||
1969 | QTAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) { | |
1970 | if (kvm_arch_remove_sw_breakpoint(current_cpu, bp) != 0) { | |
1971 | /* Try harder to find a CPU that currently sees the breakpoint. */ | |
1972 | for (env = first_cpu; env != NULL; env = env->next_cpu) { | |
1973 | cpu = ENV_GET_CPU(env); | |
1974 | if (kvm_arch_remove_sw_breakpoint(cpu, bp) == 0) { | |
1975 | break; | |
1976 | } | |
1977 | } | |
1978 | } | |
1979 | QTAILQ_REMOVE(&s->kvm_sw_breakpoints, bp, entry); | |
1980 | g_free(bp); | |
1981 | } | |
1982 | kvm_arch_remove_all_hw_breakpoints(); | |
1983 | ||
1984 | for (env = first_cpu; env != NULL; env = env->next_cpu) { | |
1985 | kvm_update_guest_debug(env, 0); | |
1986 | } | |
1987 | } | |
1988 | ||
1989 | #else /* !KVM_CAP_SET_GUEST_DEBUG */ | |
1990 | ||
1991 | int kvm_update_guest_debug(CPUArchState *env, unsigned long reinject_trap) | |
1992 | { | |
1993 | return -EINVAL; | |
1994 | } | |
1995 | ||
1996 | int kvm_insert_breakpoint(CPUArchState *current_env, target_ulong addr, | |
1997 | target_ulong len, int type) | |
1998 | { | |
1999 | return -EINVAL; | |
2000 | } | |
2001 | ||
2002 | int kvm_remove_breakpoint(CPUArchState *current_env, target_ulong addr, | |
2003 | target_ulong len, int type) | |
2004 | { | |
2005 | return -EINVAL; | |
2006 | } | |
2007 | ||
2008 | void kvm_remove_all_breakpoints(CPUArchState *current_env) | |
2009 | { | |
2010 | } | |
2011 | #endif /* !KVM_CAP_SET_GUEST_DEBUG */ | |
2012 | ||
2013 | int kvm_set_signal_mask(CPUArchState *env, const sigset_t *sigset) | |
2014 | { | |
2015 | CPUState *cpu = ENV_GET_CPU(env); | |
2016 | struct kvm_signal_mask *sigmask; | |
2017 | int r; | |
2018 | ||
2019 | if (!sigset) { | |
2020 | return kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, NULL); | |
2021 | } | |
2022 | ||
2023 | sigmask = g_malloc(sizeof(*sigmask) + sizeof(*sigset)); | |
2024 | ||
2025 | sigmask->len = 8; | |
2026 | memcpy(sigmask->sigset, sigset, sizeof(*sigset)); | |
2027 | r = kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, sigmask); | |
2028 | g_free(sigmask); | |
2029 | ||
2030 | return r; | |
2031 | } | |
2032 | int kvm_on_sigbus_vcpu(CPUState *cpu, int code, void *addr) | |
2033 | { | |
2034 | return kvm_arch_on_sigbus_vcpu(cpu, code, addr); | |
2035 | } | |
2036 | ||
2037 | int kvm_on_sigbus(int code, void *addr) | |
2038 | { | |
2039 | return kvm_arch_on_sigbus(code, addr); | |
2040 | } |